diff options
Diffstat (limited to 'drivers/infiniband/hw/ipath')
31 files changed, 10034 insertions, 1154 deletions
diff --git a/drivers/infiniband/hw/ipath/Makefile b/drivers/infiniband/hw/ipath/Makefile index fe673882686..75a6c91944c 100644 --- a/drivers/infiniband/hw/ipath/Makefile +++ b/drivers/infiniband/hw/ipath/Makefile @@ -20,17 +20,20 @@ ib_ipath-y := \ ipath_qp.o \ ipath_rc.o \ ipath_ruc.o \ + ipath_sdma.o \ ipath_srq.o \ ipath_stats.o \ ipath_sysfs.o \ ipath_uc.o \ ipath_ud.o \ ipath_user_pages.o \ + ipath_user_sdma.o \ ipath_verbs_mcast.o \ ipath_verbs.o ib_ipath-$(CONFIG_HT_IRQ) += ipath_iba6110.o ib_ipath-$(CONFIG_PCI_MSI) += ipath_iba6120.o +ib_ipath-$(CONFIG_PCI_MSI) += ipath_iba7220.o ipath_sd7220.o ipath_sd7220_img.o ib_ipath-$(CONFIG_X86_64) += ipath_wc_x86_64.o ib_ipath-$(CONFIG_PPC64) += ipath_wc_ppc64.o diff --git a/drivers/infiniband/hw/ipath/ipath_7220.h b/drivers/infiniband/hw/ipath/ipath_7220.h new file mode 100644 index 00000000000..74fa5cc5131 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_7220.h @@ -0,0 +1,57 @@ +#ifndef _IPATH_7220_H +#define _IPATH_7220_H +/* + * Copyright (c) 2007 QLogic Corporation. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +/* + * This header file provides the declarations and common definitions + * for (mostly) manipulation of the SerDes blocks within the IBA7220. + * the functions declared should only be called from within other + * 7220-related files such as ipath_iba7220.c or ipath_sd7220.c. + */ +int ipath_sd7220_presets(struct ipath_devdata *dd); +int ipath_sd7220_init(struct ipath_devdata *dd, int was_reset); +int ipath_sd7220_prog_ld(struct ipath_devdata *dd, int sdnum, u8 *img, + int len, int offset); +int ipath_sd7220_prog_vfy(struct ipath_devdata *dd, int sdnum, const u8 *img, + int len, int offset); +/* + * Below used for sdnum parameter, selecting one of the two sections + * used for PCIe, or the single SerDes used for IB, which is the + * only one currently used + */ +#define IB_7220_SERDES 2 + +int ipath_sd7220_ib_load(struct ipath_devdata *dd); +int ipath_sd7220_ib_vfy(struct ipath_devdata *dd); + +#endif /* _IPATH_7220_H */ diff --git a/drivers/infiniband/hw/ipath/ipath_common.h b/drivers/infiniband/hw/ipath/ipath_common.h index 591901aab6b..28cfe97cf1e 100644 --- a/drivers/infiniband/hw/ipath/ipath_common.h +++ b/drivers/infiniband/hw/ipath/ipath_common.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -80,6 +80,8 @@ #define IPATH_IB_LINKDOWN_DISABLE 5 #define IPATH_IB_LINK_LOOPBACK 6 /* enable local loopback */ #define IPATH_IB_LINK_EXTERNAL 7 /* normal, disable local loopback */ +#define IPATH_IB_LINK_NO_HRTBT 8 /* disable Heartbeat, e.g. for loopback */ +#define IPATH_IB_LINK_HRTBT 9 /* enable heartbeat, normal, non-loopback */ /* * These 3 values (SDR and DDR may be ORed for auto-speed @@ -198,7 +200,8 @@ typedef enum _ipath_ureg { #define IPATH_RUNTIME_FORCE_WC_ORDER 0x4 #define IPATH_RUNTIME_RCVHDR_COPY 0x8 #define IPATH_RUNTIME_MASTER 0x10 -/* 0x20 and 0x40 are no longer used, but are reserved for ABI compatibility */ +#define IPATH_RUNTIME_NODMA_RTAIL 0x80 +#define IPATH_RUNTIME_SDMA 0x200 #define IPATH_RUNTIME_FORCE_PIOAVAIL 0x400 #define IPATH_RUNTIME_PIO_REGSWAPPED 0x800 @@ -444,8 +447,9 @@ struct ipath_user_info { #define IPATH_CMD_PIOAVAILUPD 27 /* force an update of PIOAvail reg */ #define IPATH_CMD_POLL_TYPE 28 /* set the kind of polling we want */ #define IPATH_CMD_ARMLAUNCH_CTRL 29 /* armlaunch detection control */ - -#define IPATH_CMD_MAX 29 +/* 30 is unused */ +#define IPATH_CMD_SDMA_INFLIGHT 31 /* sdma inflight counter request */ +#define IPATH_CMD_SDMA_COMPLETE 32 /* sdma completion counter request */ /* * Poll types @@ -483,6 +487,17 @@ struct ipath_cmd { union { struct ipath_tid_info tid_info; struct ipath_user_info user_info; + + /* + * address in userspace where we should put the sdma + * inflight counter + */ + __u64 sdma_inflight; + /* + * address in userspace where we should put the sdma + * completion counter + */ + __u64 sdma_complete; /* address in userspace of struct ipath_port_info to write result to */ __u64 port_info; @@ -537,7 +552,7 @@ struct ipath_diag_pkt { /* The second diag_pkt struct is the expanded version that allows * more control over the packet, specifically, by allowing a custom - * pbc (+ extra) qword, so that special modes and deliberate + * pbc (+ static rate) qword, so that special modes and deliberate * changes to CRCs can be used. The elements were also re-ordered * for better alignment and to avoid padding issues. */ @@ -662,8 +677,12 @@ struct infinipath_counters { #define INFINIPATH_RHF_LENGTH_SHIFT 0 #define INFINIPATH_RHF_RCVTYPE_MASK 0x7 #define INFINIPATH_RHF_RCVTYPE_SHIFT 11 -#define INFINIPATH_RHF_EGRINDEX_MASK 0x7FF +#define INFINIPATH_RHF_EGRINDEX_MASK 0xFFF #define INFINIPATH_RHF_EGRINDEX_SHIFT 16 +#define INFINIPATH_RHF_SEQ_MASK 0xF +#define INFINIPATH_RHF_SEQ_SHIFT 0 +#define INFINIPATH_RHF_HDRQ_OFFSET_MASK 0x7FF +#define INFINIPATH_RHF_HDRQ_OFFSET_SHIFT 4 #define INFINIPATH_RHF_H_ICRCERR 0x80000000 #define INFINIPATH_RHF_H_VCRCERR 0x40000000 #define INFINIPATH_RHF_H_PARITYERR 0x20000000 @@ -673,6 +692,8 @@ struct infinipath_counters { #define INFINIPATH_RHF_H_TIDERR 0x02000000 #define INFINIPATH_RHF_H_MKERR 0x01000000 #define INFINIPATH_RHF_H_IBERR 0x00800000 +#define INFINIPATH_RHF_H_ERR_MASK 0xFF800000 +#define INFINIPATH_RHF_L_USE_EGR 0x80000000 #define INFINIPATH_RHF_L_SWA 0x00008000 #define INFINIPATH_RHF_L_SWB 0x00004000 @@ -696,6 +717,7 @@ struct infinipath_counters { /* SendPIO per-buffer control */ #define INFINIPATH_SP_TEST 0x40 #define INFINIPATH_SP_TESTEBP 0x20 +#define INFINIPATH_SP_TRIGGER_SHIFT 15 /* SendPIOAvail bits */ #define INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT 1 @@ -762,6 +784,7 @@ struct ether_header { #define IPATH_MSN_MASK 0xFFFFFF #define IPATH_QPN_MASK 0xFFFFFF #define IPATH_MULTICAST_LID_BASE 0xC000 +#define IPATH_EAGER_TID_ID INFINIPATH_I_TID_MASK #define IPATH_MULTICAST_QPN 0xFFFFFF /* Receive Header Queue: receive type (from infinipath) */ @@ -781,7 +804,7 @@ struct ether_header { */ static inline __u32 ipath_hdrget_err_flags(const __le32 * rbuf) { - return __le32_to_cpu(rbuf[1]); + return __le32_to_cpu(rbuf[1]) & INFINIPATH_RHF_H_ERR_MASK; } static inline __u32 ipath_hdrget_rcv_type(const __le32 * rbuf) @@ -802,6 +825,23 @@ static inline __u32 ipath_hdrget_index(const __le32 * rbuf) & INFINIPATH_RHF_EGRINDEX_MASK; } +static inline __u32 ipath_hdrget_seq(const __le32 *rbuf) +{ + return (__le32_to_cpu(rbuf[1]) >> INFINIPATH_RHF_SEQ_SHIFT) + & INFINIPATH_RHF_SEQ_MASK; +} + +static inline __u32 ipath_hdrget_offset(const __le32 *rbuf) +{ + return (__le32_to_cpu(rbuf[1]) >> INFINIPATH_RHF_HDRQ_OFFSET_SHIFT) + & INFINIPATH_RHF_HDRQ_OFFSET_MASK; +} + +static inline __u32 ipath_hdrget_use_egr_buf(const __le32 *rbuf) +{ + return __le32_to_cpu(rbuf[0]) & INFINIPATH_RHF_L_USE_EGR; +} + static inline __u32 ipath_hdrget_ipath_ver(__le32 hdrword) { return (__le32_to_cpu(hdrword) >> INFINIPATH_I_VERS_SHIFT) diff --git a/drivers/infiniband/hw/ipath/ipath_debug.h b/drivers/infiniband/hw/ipath/ipath_debug.h index d6f69532d83..65926cd3575 100644 --- a/drivers/infiniband/hw/ipath/ipath_debug.h +++ b/drivers/infiniband/hw/ipath/ipath_debug.h @@ -66,6 +66,7 @@ #define __IPATH_IPATHERR 0x40000 /* Ethernet (IPATH) errors */ #define __IPATH_IPATHPD 0x80000 /* Ethernet (IPATH) packet dump */ #define __IPATH_IPATHTABLE 0x100000 /* Ethernet (IPATH) table dump */ +#define __IPATH_LINKVERBDBG 0x200000 /* very verbose linkchange debug */ #else /* _IPATH_DEBUGGING */ @@ -89,6 +90,7 @@ #define __IPATH_IPATHERR 0x0 /* Ethernet (IPATH) errors on */ #define __IPATH_IPATHPD 0x0 /* Ethernet (IPATH) packet dump on */ #define __IPATH_IPATHTABLE 0x0 /* Ethernet (IPATH) packet dump on */ +#define __IPATH_LINKVERBDBG 0x0 /* very verbose linkchange debug */ #endif /* _IPATH_DEBUGGING */ diff --git a/drivers/infiniband/hw/ipath/ipath_diag.c b/drivers/infiniband/hw/ipath/ipath_diag.c index 4137c7770f1..6d49d2f18a8 100644 --- a/drivers/infiniband/hw/ipath/ipath_diag.c +++ b/drivers/infiniband/hw/ipath/ipath_diag.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -330,13 +330,19 @@ static ssize_t ipath_diagpkt_write(struct file *fp, struct ipath_devdata *dd; ssize_t ret = 0; u64 val; + u32 l_state, lt_state; /* LinkState, LinkTrainingState */ - if (count != sizeof(dp)) { + if (count < sizeof(odp)) { ret = -EINVAL; goto bail; } - if (copy_from_user(&dp, data, sizeof(dp))) { + if (count == sizeof(dp)) { + if (copy_from_user(&dp, data, sizeof(dp))) { + ret = -EFAULT; + goto bail; + } + } else if (copy_from_user(&odp, data, sizeof(odp))) { ret = -EFAULT; goto bail; } @@ -396,10 +402,17 @@ static ssize_t ipath_diagpkt_write(struct file *fp, ret = -ENODEV; goto bail; } - /* Check link state, but not if we have custom PBC */ - val = dd->ipath_lastibcstat & IPATH_IBSTATE_MASK; - if (!dp.pbc_wd && val != IPATH_IBSTATE_INIT && - val != IPATH_IBSTATE_ARM && val != IPATH_IBSTATE_ACTIVE) { + /* + * Want to skip check for l_state if using custom PBC, + * because we might be trying to force an SM packet out. + * first-cut, skip _all_ state checking in that case. + */ + val = ipath_ib_state(dd, dd->ipath_lastibcstat); + lt_state = ipath_ib_linktrstate(dd, dd->ipath_lastibcstat); + l_state = ipath_ib_linkstate(dd, dd->ipath_lastibcstat); + if (!dp.pbc_wd && (lt_state != INFINIPATH_IBCS_LT_STATE_LINKUP || + (val != dd->ib_init && val != dd->ib_arm && + val != dd->ib_active))) { ipath_cdbg(VERBOSE, "unit %u not ready (state %llx)\n", dd->ipath_unit, (unsigned long long) val); ret = -EINVAL; @@ -431,15 +444,17 @@ static ssize_t ipath_diagpkt_write(struct file *fp, goto bail; } - piobuf = ipath_getpiobuf(dd, &pbufn); + plen >>= 2; /* in dwords */ + + piobuf = ipath_getpiobuf(dd, plen, &pbufn); if (!piobuf) { ipath_cdbg(VERBOSE, "No PIO buffers avail unit for %u\n", dd->ipath_unit); ret = -EBUSY; goto bail; } - - plen >>= 2; /* in dwords */ + /* disarm it just to be extra sure */ + ipath_disarm_piobufs(dd, pbufn, 1); if (ipath_debug & __IPATH_PKTDBG) ipath_cdbg(VERBOSE, "unit %u 0x%x+1w pio%d\n", diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c index ca4d0acc678..e0a64f070b9 100644 --- a/drivers/infiniband/hw/ipath/ipath_driver.c +++ b/drivers/infiniband/hw/ipath/ipath_driver.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -41,7 +41,6 @@ #include "ipath_kernel.h" #include "ipath_verbs.h" -#include "ipath_common.h" static void ipath_update_pio_bufs(struct ipath_devdata *); @@ -73,10 +72,27 @@ module_param_named(debug, ipath_debug, uint, S_IWUSR | S_IRUGO); MODULE_PARM_DESC(debug, "mask for debug prints"); EXPORT_SYMBOL_GPL(ipath_debug); +unsigned ipath_mtu4096 = 1; /* max 4KB IB mtu by default, if supported */ +module_param_named(mtu4096, ipath_mtu4096, uint, S_IRUGO); +MODULE_PARM_DESC(mtu4096, "enable MTU of 4096 bytes, if supported"); + +static unsigned ipath_hol_timeout_ms = 13000; +module_param_named(hol_timeout_ms, ipath_hol_timeout_ms, uint, S_IRUGO); +MODULE_PARM_DESC(hol_timeout_ms, + "duration of user app suspension after link failure"); + +unsigned ipath_linkrecovery = 1; +module_param_named(linkrecovery, ipath_linkrecovery, uint, S_IWUSR | S_IRUGO); +MODULE_PARM_DESC(linkrecovery, "enable workaround for link recovery issue"); + MODULE_LICENSE("GPL"); -MODULE_AUTHOR("QLogic <support@pathscale.com>"); +MODULE_AUTHOR("QLogic <support@qlogic.com>"); MODULE_DESCRIPTION("QLogic InfiniPath driver"); +/* + * Table to translate the LINKTRAININGSTATE portion of + * IBCStatus to a human-readable form. + */ const char *ipath_ibcstatus_str[] = { "Disabled", "LinkUp", @@ -91,9 +107,20 @@ const char *ipath_ibcstatus_str[] = { "CfgWaitRmt", "CfgIdle", "RecovRetrain", - "LState0xD", /* unused */ + "CfgTxRevLane", /* unused before IBA7220 */ "RecovWaitRmt", "RecovIdle", + /* below were added for IBA7220 */ + "CfgEnhanced", + "CfgTest", + "CfgWaitRmtTest", + "CfgWaitCfgEnhanced", + "SendTS_T", + "SendTstIdles", + "RcvTS_T", + "SendTst_TS1s", + "LTState18", "LTState19", "LTState1A", "LTState1B", + "LTState1C", "LTState1D", "LTState1E", "LTState1F" }; static void __devexit ipath_remove_one(struct pci_dev *); @@ -102,8 +129,10 @@ static int __devinit ipath_init_one(struct pci_dev *, /* Only needed for registration, nothing else needs this info */ #define PCI_VENDOR_ID_PATHSCALE 0x1fc1 +#define PCI_VENDOR_ID_QLOGIC 0x1077 #define PCI_DEVICE_ID_INFINIPATH_HT 0xd #define PCI_DEVICE_ID_INFINIPATH_PE800 0x10 +#define PCI_DEVICE_ID_INFINIPATH_7220 0x7220 /* Number of seconds before our card status check... */ #define STATUS_TIMEOUT 60 @@ -111,6 +140,7 @@ static int __devinit ipath_init_one(struct pci_dev *, static const struct pci_device_id ipath_pci_tbl[] = { { PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_HT) }, { PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_PE800) }, + { PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_INFINIPATH_7220) }, { 0, } }; @@ -126,19 +156,6 @@ static struct pci_driver ipath_driver = { }, }; -static void ipath_check_status(struct work_struct *work) -{ - struct ipath_devdata *dd = container_of(work, struct ipath_devdata, - status_work.work); - - /* - * If we don't have any interrupts, let the user know and - * don't bother checking again. - */ - if (dd->ipath_int_counter == 0) - dev_err(&dd->pcidev->dev, "No interrupts detected.\n"); -} - static inline void read_bars(struct ipath_devdata *dd, struct pci_dev *dev, u32 *bar0, u32 *bar1) { @@ -206,8 +223,6 @@ static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev) dd->pcidev = pdev; pci_set_drvdata(pdev, dd); - INIT_DELAYED_WORK(&dd->status_work, ipath_check_status); - list_add(&dd->ipath_list, &ipath_dev_list); bail_unlock: @@ -234,12 +249,12 @@ struct ipath_devdata *ipath_lookup(int unit) return dd; } -int ipath_count_units(int *npresentp, int *nupp, u32 *maxportsp) +int ipath_count_units(int *npresentp, int *nupp, int *maxportsp) { int nunits, npresent, nup; struct ipath_devdata *dd; unsigned long flags; - u32 maxports; + int maxports; nunits = npresent = nup = maxports = 0; @@ -304,7 +319,7 @@ static void ipath_verify_pioperf(struct ipath_devdata *dd) u32 *addr; u64 msecs, emsecs; - piobuf = ipath_getpiobuf(dd, &pbnum); + piobuf = ipath_getpiobuf(dd, 0, &pbnum); if (!piobuf) { dev_info(&dd->pcidev->dev, "No PIObufs for checking perf, skipping\n"); @@ -336,7 +351,14 @@ static void ipath_verify_pioperf(struct ipath_devdata *dd) ipath_disable_armlaunch(dd); - writeq(0, piobuf); /* length 0, no dwords actually sent */ + /* + * length 0, no dwords actually sent, and mark as VL15 + * on chips where that may matter (due to IB flowcontrol) + */ + if ((dd->ipath_flags & IPATH_HAS_PBC_CNT)) + writeq(1UL << 63, piobuf); + else + writeq(0, piobuf); ipath_flush_wc(); /* @@ -377,6 +399,7 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, struct ipath_devdata *dd; unsigned long long addr; u32 bar0 = 0, bar1 = 0; + u8 rev; dd = ipath_alloc_devdata(pdev); if (IS_ERR(dd)) { @@ -408,7 +431,7 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, } addr = pci_resource_start(pdev, 0); len = pci_resource_len(pdev, 0); - ipath_cdbg(VERBOSE, "regbase (0) %llx len %d pdev->irq %d, vend %x/%x " + ipath_cdbg(VERBOSE, "regbase (0) %llx len %d irq %d, vend %x/%x " "driver_data %lx\n", addr, len, pdev->irq, ent->vendor, ent->device, ent->driver_data); @@ -512,6 +535,13 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, "CONFIG_PCI_MSI is not enabled\n", ent->device); return -ENODEV; #endif + case PCI_DEVICE_ID_INFINIPATH_7220: +#ifndef CONFIG_PCI_MSI + ipath_dbg("CONFIG_PCI_MSI is not enabled, " + "using IntX for unit %u\n", dd->ipath_unit); +#endif + ipath_init_iba7220_funcs(dd); + break; default: ipath_dev_err(dd, "Found unknown QLogic deviceid 0x%x, " "failing\n", ent->device); @@ -533,7 +563,13 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, goto bail_regions; } - dd->ipath_pcirev = pdev->revision; + ret = pci_read_config_byte(pdev, PCI_REVISION_ID, &rev); + if (ret) { + ipath_dev_err(dd, "Failed to read PCI revision ID unit " + "%u: err %d\n", dd->ipath_unit, -ret); + goto bail_regions; /* shouldn't ever happen */ + } + dd->ipath_pcirev = rev; #if defined(__powerpc__) /* There isn't a generic way to specify writethrough mappings */ @@ -556,14 +592,6 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, ipath_cdbg(VERBOSE, "mapped io addr %llx to kregbase %p\n", addr, dd->ipath_kregbase); - /* - * clear ipath_flags here instead of in ipath_init_chip as it is set - * by ipath_setup_htconfig. - */ - dd->ipath_flags = 0; - dd->ipath_lli_counter = 0; - dd->ipath_lli_errors = 0; - if (dd->ipath_f_bus(dd, pdev)) ipath_dev_err(dd, "Failed to setup config space; " "continuing anyway\n"); @@ -608,13 +636,11 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, ipath_diag_add(dd); ipath_register_ib_device(dd); - /* Check that card status in STATUS_TIMEOUT seconds. */ - schedule_delayed_work(&dd->status_work, HZ * STATUS_TIMEOUT); - goto bail; bail_irqsetup: - if (pdev->irq) free_irq(pdev->irq, dd); + if (pdev->irq) + free_irq(pdev->irq, dd); bail_iounmap: iounmap((volatile void __iomem *) dd->ipath_kregbase); @@ -654,6 +680,10 @@ static void __devexit cleanup_device(struct ipath_devdata *dd) ipath_disable_wc(dd); } + if (dd->ipath_spectriggerhit) + dev_info(&dd->pcidev->dev, "%lu special trigger hits\n", + dd->ipath_spectriggerhit); + if (dd->ipath_pioavailregs_dma) { dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE, (void *) dd->ipath_pioavailregs_dma, @@ -706,6 +736,8 @@ static void __devexit cleanup_device(struct ipath_devdata *dd) tmpp = dd->ipath_pageshadow; dd->ipath_pageshadow = NULL; vfree(tmpp); + + dd->ipath_egrtidbase = NULL; } /* @@ -738,7 +770,6 @@ static void __devexit ipath_remove_one(struct pci_dev *pdev) */ ipath_shutdown_device(dd); - cancel_delayed_work(&dd->status_work); flush_scheduled_work(); if (dd->verbs_dev) @@ -823,20 +854,8 @@ void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first, ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); } - - /* - * Disable PIOAVAILUPD, then re-enable, reading scratch in - * between. This seems to avoid a chip timing race that causes - * pioavail updates to memory to stop. We xor as we don't - * know the state of the bit when we're called. - */ - spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl ^ INFINIPATH_S_PIOBUFAVAILUPD); - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl); - spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + /* on some older chips, update may not happen after cancel */ + ipath_force_pio_avail_update(dd); } /** @@ -873,18 +892,52 @@ int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, int msecs) (unsigned long long) ipath_read_kreg64( dd, dd->ipath_kregs->kr_ibcctrl), (unsigned long long) val, - ipath_ibcstatus_str[val & 0xf]); + ipath_ibcstatus_str[val & dd->ibcs_lts_mask]); } return (dd->ipath_flags & state) ? 0 : -ETIMEDOUT; } +static void decode_sdma_errs(struct ipath_devdata *dd, ipath_err_t err, + char *buf, size_t blen) +{ + static const struct { + ipath_err_t err; + const char *msg; + } errs[] = { + { INFINIPATH_E_SDMAGENMISMATCH, "SDmaGenMismatch" }, + { INFINIPATH_E_SDMAOUTOFBOUND, "SDmaOutOfBound" }, + { INFINIPATH_E_SDMATAILOUTOFBOUND, "SDmaTailOutOfBound" }, + { INFINIPATH_E_SDMABASE, "SDmaBase" }, + { INFINIPATH_E_SDMA1STDESC, "SDma1stDesc" }, + { INFINIPATH_E_SDMARPYTAG, "SDmaRpyTag" }, + { INFINIPATH_E_SDMADWEN, "SDmaDwEn" }, + { INFINIPATH_E_SDMAMISSINGDW, "SDmaMissingDw" }, + { INFINIPATH_E_SDMAUNEXPDATA, "SDmaUnexpData" }, + { INFINIPATH_E_SDMADESCADDRMISALIGN, "SDmaDescAddrMisalign" }, + { INFINIPATH_E_SENDBUFMISUSE, "SendBufMisuse" }, + { INFINIPATH_E_SDMADISABLED, "SDmaDisabled" }, + }; + int i; + int expected; + size_t bidx = 0; + + for (i = 0; i < ARRAY_SIZE(errs); i++) { + expected = (errs[i].err != INFINIPATH_E_SDMADISABLED) ? 0 : + test_bit(IPATH_SDMA_ABORTING, &dd->ipath_sdma_status); + if ((err & errs[i].err) && !expected) + bidx += snprintf(buf + bidx, blen - bidx, + "%s ", errs[i].msg); + } +} + /* * Decode the error status into strings, deciding whether to always * print * it or not depending on "normal packet errors" vs everything * else. Return 1 if "real" errors, otherwise 0 if only packet * errors, so caller can decide what to print with the string. */ -int ipath_decode_err(char *buf, size_t blen, ipath_err_t err) +int ipath_decode_err(struct ipath_devdata *dd, char *buf, size_t blen, + ipath_err_t err) { int iserr = 1; *buf = '\0'; @@ -922,6 +975,8 @@ int ipath_decode_err(char *buf, size_t blen, ipath_err_t err) strlcat(buf, "rbadversion ", blen); if (err & INFINIPATH_E_RHDR) strlcat(buf, "rhdr ", blen); + if (err & INFINIPATH_E_SENDSPECIALTRIGGER) + strlcat(buf, "sendspecialtrigger ", blen); if (err & INFINIPATH_E_RLONGPKTLEN) strlcat(buf, "rlongpktlen ", blen); if (err & INFINIPATH_E_RMAXPKTLEN) @@ -964,6 +1019,10 @@ int ipath_decode_err(char *buf, size_t blen, ipath_err_t err) strlcat(buf, "hardware ", blen); if (err & INFINIPATH_E_RESET) strlcat(buf, "reset ", blen); + if (err & INFINIPATH_E_SDMAERRS) + decode_sdma_errs(dd, err, buf, blen); + if (err & INFINIPATH_E_INVALIDEEPCMD) + strlcat(buf, "invalideepromcmd ", blen); done: return iserr; } @@ -1076,18 +1135,17 @@ static void ipath_rcv_hdrerr(struct ipath_devdata *dd, u32 eflags, u32 l, u32 etail, - u64 *rc) + __le32 *rhf_addr, + struct ipath_message_header *hdr) { char emsg[128]; - struct ipath_message_header *hdr; get_rhf_errstring(eflags, emsg, sizeof emsg); - hdr = (struct ipath_message_header *)&rc[1]; ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u " "tlen=%x opcode=%x egridx=%x: %s\n", eflags, l, - ipath_hdrget_rcv_type((__le32 *) rc), - ipath_hdrget_length_in_bytes((__le32 *) rc), + ipath_hdrget_rcv_type(rhf_addr), + ipath_hdrget_length_in_bytes(rhf_addr), be32_to_cpu(hdr->bth[0]) >> 24, etail, emsg); @@ -1112,55 +1170,52 @@ static void ipath_rcv_hdrerr(struct ipath_devdata *dd, */ void ipath_kreceive(struct ipath_portdata *pd) { - u64 *rc; struct ipath_devdata *dd = pd->port_dd; + __le32 *rhf_addr; void *ebuf; const u32 rsize = dd->ipath_rcvhdrentsize; /* words */ const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize; /* words */ u32 etail = -1, l, hdrqtail; struct ipath_message_header *hdr; - u32 eflags, i, etype, tlen, pkttot = 0, updegr=0, reloop=0; + u32 eflags, i, etype, tlen, pkttot = 0, updegr = 0, reloop = 0; static u64 totcalls; /* stats, may eventually remove */ - - if (!dd->ipath_hdrqtailptr) { - ipath_dev_err(dd, - "hdrqtailptr not set, can't do receives\n"); - goto bail; - } + int last; l = pd->port_head; - hdrqtail = ipath_get_rcvhdrtail(pd); - if (l == hdrqtail) - goto bail; - -reloop: - for (i = 0; l != hdrqtail; i++) { - u32 qp; - u8 *bthbytes; + rhf_addr = (__le32 *) pd->port_rcvhdrq + l + dd->ipath_rhf_offset; + if (dd->ipath_flags & IPATH_NODMA_RTAIL) { + u32 seq = ipath_hdrget_seq(rhf_addr); - rc = (u64 *) (pd->port_rcvhdrq + (l << 2)); - hdr = (struct ipath_message_header *)&rc[1]; - /* - * could make a network order version of IPATH_KD_QP, and - * do the obvious shift before masking to speed this up. - */ - qp = ntohl(hdr->bth[1]) & 0xffffff; - bthbytes = (u8 *) hdr->bth; + if (seq != pd->port_seq_cnt) + goto bail; + hdrqtail = 0; + } else { + hdrqtail = ipath_get_rcvhdrtail(pd); + if (l == hdrqtail) + goto bail; + smp_rmb(); + } - eflags = ipath_hdrget_err_flags((__le32 *) rc); - etype = ipath_hdrget_rcv_type((__le32 *) rc); +reloop: + for (last = 0, i = 1; !last; i++) { + hdr = dd->ipath_f_get_msgheader(dd, rhf_addr); + eflags = ipath_hdrget_err_flags(rhf_addr); + etype = ipath_hdrget_rcv_type(rhf_addr); /* total length */ - tlen = ipath_hdrget_length_in_bytes((__le32 *) rc); + tlen = ipath_hdrget_length_in_bytes(rhf_addr); ebuf = NULL; - if (etype != RCVHQ_RCV_TYPE_EXPECTED) { + if ((dd->ipath_flags & IPATH_NODMA_RTAIL) ? + ipath_hdrget_use_egr_buf(rhf_addr) : + (etype != RCVHQ_RCV_TYPE_EXPECTED)) { /* - * it turns out that the chips uses an eager buffer + * It turns out that the chip uses an eager buffer * for all non-expected packets, whether it "needs" * one or not. So always get the index, but don't * set ebuf (so we try to copy data) unless the * length requires it. */ - etail = ipath_hdrget_index((__le32 *) rc); + etail = ipath_hdrget_index(rhf_addr); + updegr = 1; if (tlen > sizeof(*hdr) || etype == RCVHQ_RCV_TYPE_NON_KD) ebuf = ipath_get_egrbuf(dd, etail); @@ -1171,75 +1226,91 @@ reloop: * packets; only ipathhdrerr should be set. */ - if (etype != RCVHQ_RCV_TYPE_NON_KD && etype != - RCVHQ_RCV_TYPE_ERROR && ipath_hdrget_ipath_ver( - hdr->iph.ver_port_tid_offset) != - IPS_PROTO_VERSION) { + if (etype != RCVHQ_RCV_TYPE_NON_KD && + etype != RCVHQ_RCV_TYPE_ERROR && + ipath_hdrget_ipath_ver(hdr->iph.ver_port_tid_offset) != + IPS_PROTO_VERSION) ipath_cdbg(PKT, "Bad InfiniPath protocol version " "%x\n", etype); - } if (unlikely(eflags)) - ipath_rcv_hdrerr(dd, eflags, l, etail, rc); + ipath_rcv_hdrerr(dd, eflags, l, etail, rhf_addr, hdr); else if (etype == RCVHQ_RCV_TYPE_NON_KD) { - ipath_ib_rcv(dd->verbs_dev, rc + 1, ebuf, tlen); + ipath_ib_rcv(dd->verbs_dev, (u32 *)hdr, ebuf, tlen); if (dd->ipath_lli_counter) dd->ipath_lli_counter--; + } else if (etype == RCVHQ_RCV_TYPE_EAGER) { + u8 opcode = be32_to_cpu(hdr->bth[0]) >> 24; + u32 qp = be32_to_cpu(hdr->bth[1]) & 0xffffff; ipath_cdbg(PKT, "typ %x, opcode %x (eager, " "qp=%x), len %x; ignored\n", - etype, bthbytes[0], qp, tlen); + etype, opcode, qp, tlen); } - else if (etype == RCVHQ_RCV_TYPE_EAGER) - ipath_cdbg(PKT, "typ %x, opcode %x (eager, " - "qp=%x), len %x; ignored\n", - etype, bthbytes[0], qp, tlen); else if (etype == RCVHQ_RCV_TYPE_EXPECTED) ipath_dbg("Bug: Expected TID, opcode %x; ignored\n", - be32_to_cpu(hdr->bth[0]) & 0xff); + be32_to_cpu(hdr->bth[0]) >> 24); else { /* * error packet, type of error unknown. * Probably type 3, but we don't know, so don't * even try to print the opcode, etc. + * Usually caused by a "bad packet", that has no + * BTH, when the LRH says it should. */ - ipath_dbg("Error Pkt, but no eflags! egrbuf %x, " - "len %x\nhdrq@%lx;hdrq+%x rhf: %llx; " - "hdr %llx %llx %llx %llx %llx\n", - etail, tlen, (unsigned long) rc, l, - (unsigned long long) rc[0], - (unsigned long long) rc[1], - (unsigned long long) rc[2], - (unsigned long long) rc[3], - (unsigned long long) rc[4], - (unsigned long long) rc[5]); + ipath_cdbg(ERRPKT, "Error Pkt, but no eflags! egrbuf" + " %x, len %x hdrq+%x rhf: %Lx\n", + etail, tlen, l, + le64_to_cpu(*(__le64 *) rhf_addr)); + if (ipath_debug & __IPATH_ERRPKTDBG) { + u32 j, *d, dw = rsize-2; + if (rsize > (tlen>>2)) + dw = tlen>>2; + d = (u32 *)hdr; + printk(KERN_DEBUG "EPkt rcvhdr(%x dw):\n", + dw); + for (j = 0; j < dw; j++) + printk(KERN_DEBUG "%8x%s", d[j], + (j%8) == 7 ? "\n" : " "); + printk(KERN_DEBUG ".\n"); + } } l += rsize; if (l >= maxcnt) l = 0; - if (etype != RCVHQ_RCV_TYPE_EXPECTED) - updegr = 1; + rhf_addr = (__le32 *) pd->port_rcvhdrq + + l + dd->ipath_rhf_offset; + if (dd->ipath_flags & IPATH_NODMA_RTAIL) { + u32 seq = ipath_hdrget_seq(rhf_addr); + + if (++pd->port_seq_cnt > 13) + pd->port_seq_cnt = 1; + if (seq != pd->port_seq_cnt) + last = 1; + } else if (l == hdrqtail) + last = 1; /* * update head regs on last packet, and every 16 packets. * Reduce bus traffic, while still trying to prevent * rcvhdrq overflows, for when the queue is nearly full */ - if (l == hdrqtail || (i && !(i&0xf))) { - u64 lval; - if (l == hdrqtail) - /* request IBA6120 interrupt only on last */ - lval = dd->ipath_rhdrhead_intr_off | l; - else - lval = l; - (void)ipath_write_ureg(dd, ur_rcvhdrhead, lval, 0); + if (last || !(i & 0xf)) { + u64 lval = l; + + /* request IBA6120 and 7220 interrupt only on last */ + if (last) + lval |= dd->ipath_rhdrhead_intr_off; + ipath_write_ureg(dd, ur_rcvhdrhead, lval, + pd->port_port); if (updegr) { - (void)ipath_write_ureg(dd, ur_rcvegrindexhead, - etail, 0); + ipath_write_ureg(dd, ur_rcvegrindexhead, + etail, pd->port_port); updegr = 0; } } } - if (!dd->ipath_rhdrhead_intr_off && !reloop) { + if (!dd->ipath_rhdrhead_intr_off && !reloop && + !(dd->ipath_flags & IPATH_NODMA_RTAIL)) { /* IBA6110 workaround; we can have a race clearing chip * interrupt with another interrupt about to be delivered, * and can clear it before it is delivered on the GPIO @@ -1301,7 +1372,6 @@ static void ipath_update_pio_bufs(struct ipath_devdata *dd) * happens when all buffers are in use, so only cpu overhead, not * latency or bandwidth is affected. */ -#define _IPATH_ALL_CHECKBITS 0x5555555555555555ULL if (!dd->ipath_pioavailregs_dma) { ipath_dbg("Update shadow pioavail, but regs_dma NULL!\n"); return; @@ -1346,7 +1416,7 @@ static void ipath_update_pio_bufs(struct ipath_devdata *dd) piov = le64_to_cpu(dd->ipath_pioavailregs_dma[i ^ 1]); else piov = le64_to_cpu(dd->ipath_pioavailregs_dma[i]); - pchg = _IPATH_ALL_CHECKBITS & + pchg = dd->ipath_pioavailkernel[i] & ~(dd->ipath_pioavailshadow[i] ^ piov); pchbusy = pchg << INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT; if (pchg && (pchbusy & dd->ipath_pioavailshadow[i])) { @@ -1397,27 +1467,63 @@ int ipath_setrcvhdrsize(struct ipath_devdata *dd, unsigned rhdrsize) return ret; } -/** - * ipath_getpiobuf - find an available pio buffer - * @dd: the infinipath device - * @pbufnum: the buffer number is placed here +/* + * debugging code and stats updates if no pio buffers available. + */ +static noinline void no_pio_bufs(struct ipath_devdata *dd) +{ + unsigned long *shadow = dd->ipath_pioavailshadow; + __le64 *dma = (__le64 *)dd->ipath_pioavailregs_dma; + + dd->ipath_upd_pio_shadow = 1; + + /* + * not atomic, but if we lose a stat count in a while, that's OK + */ + ipath_stats.sps_nopiobufs++; + if (!(++dd->ipath_consec_nopiobuf % 100000)) { + ipath_dbg("%u pio sends with no bufavail; dmacopy: " + "%llx %llx %llx %llx; shadow: %lx %lx %lx %lx\n", + dd->ipath_consec_nopiobuf, + (unsigned long long) le64_to_cpu(dma[0]), + (unsigned long long) le64_to_cpu(dma[1]), + (unsigned long long) le64_to_cpu(dma[2]), + (unsigned long long) le64_to_cpu(dma[3]), + shadow[0], shadow[1], shadow[2], shadow[3]); + /* + * 4 buffers per byte, 4 registers above, cover rest + * below + */ + if ((dd->ipath_piobcnt2k + dd->ipath_piobcnt4k) > + (sizeof(shadow[0]) * 4 * 4)) + ipath_dbg("2nd group: dmacopy: %llx %llx " + "%llx %llx; shadow: %lx %lx %lx %lx\n", + (unsigned long long)le64_to_cpu(dma[4]), + (unsigned long long)le64_to_cpu(dma[5]), + (unsigned long long)le64_to_cpu(dma[6]), + (unsigned long long)le64_to_cpu(dma[7]), + shadow[4], shadow[5], shadow[6], + shadow[7]); + } +} + +/* + * common code for normal driver pio buffer allocation, and reserved + * allocation. * * do appropriate marking as busy, etc. * returns buffer number if one found (>=0), negative number is error. - * Used by ipath_layer_send */ -u32 __iomem *ipath_getpiobuf(struct ipath_devdata *dd, u32 * pbufnum) +static u32 __iomem *ipath_getpiobuf_range(struct ipath_devdata *dd, + u32 *pbufnum, u32 first, u32 last, u32 firsti) { - int i, j, starti, updated = 0; - unsigned piobcnt, iter; + int i, j, updated = 0; + unsigned piobcnt; unsigned long flags; unsigned long *shadow = dd->ipath_pioavailshadow; u32 __iomem *buf; - piobcnt = (unsigned)(dd->ipath_piobcnt2k - + dd->ipath_piobcnt4k); - starti = dd->ipath_lastport_piobuf; - iter = piobcnt - starti; + piobcnt = last - first; if (dd->ipath_upd_pio_shadow) { /* * Minor optimization. If we had no buffers on last call, @@ -1425,12 +1531,10 @@ u32 __iomem *ipath_getpiobuf(struct ipath_devdata *dd, u32 * pbufnum) * if no buffers were updated, to be paranoid */ ipath_update_pio_bufs(dd); - /* we scanned here, don't do it at end of scan */ - updated = 1; - i = starti; + updated++; + i = first; } else - i = dd->ipath_lastpioindex; - + i = firsti; rescan: /* * while test_and_set_bit() is atomic, we do that and then the @@ -1438,104 +1542,141 @@ rescan: * of the remaining armlaunch errors. */ spin_lock_irqsave(&ipath_pioavail_lock, flags); - for (j = 0; j < iter; j++, i++) { - if (i >= piobcnt) - i = starti; - /* - * To avoid bus lock overhead, we first find a candidate - * buffer, then do the test and set, and continue if that - * fails. - */ - if (test_bit((2 * i) + 1, shadow) || - test_and_set_bit((2 * i) + 1, shadow)) + for (j = 0; j < piobcnt; j++, i++) { + if (i >= last) + i = first; + if (__test_and_set_bit((2 * i) + 1, shadow)) continue; /* flip generation bit */ - change_bit(2 * i, shadow); + __change_bit(2 * i, shadow); break; } spin_unlock_irqrestore(&ipath_pioavail_lock, flags); - if (j == iter) { - volatile __le64 *dma = dd->ipath_pioavailregs_dma; - - /* - * first time through; shadow exhausted, but may be real - * buffers available, so go see; if any updated, rescan - * (once) - */ + if (j == piobcnt) { if (!updated) { + /* + * first time through; shadow exhausted, but may be + * buffers available, try an update and then rescan. + */ ipath_update_pio_bufs(dd); - updated = 1; - i = starti; + updated++; + i = first; goto rescan; - } - dd->ipath_upd_pio_shadow = 1; - /* - * not atomic, but if we lose one once in a while, that's OK - */ - ipath_stats.sps_nopiobufs++; - if (!(++dd->ipath_consec_nopiobuf % 100000)) { - ipath_dbg( - "%u pio sends with no bufavail; dmacopy: " - "%llx %llx %llx %llx; shadow: " - "%lx %lx %lx %lx\n", - dd->ipath_consec_nopiobuf, - (unsigned long long) le64_to_cpu(dma[0]), - (unsigned long long) le64_to_cpu(dma[1]), - (unsigned long long) le64_to_cpu(dma[2]), - (unsigned long long) le64_to_cpu(dma[3]), - shadow[0], shadow[1], shadow[2], - shadow[3]); + } else if (updated == 1 && piobcnt <= + ((dd->ipath_sendctrl + >> INFINIPATH_S_UPDTHRESH_SHIFT) & + INFINIPATH_S_UPDTHRESH_MASK)) { /* - * 4 buffers per byte, 4 registers above, cover rest - * below + * for chips supporting and using the update + * threshold we need to force an update of the + * in-memory copy if the count is less than the + * thershold, then check one more time. */ - if ((dd->ipath_piobcnt2k + dd->ipath_piobcnt4k) > - (sizeof(shadow[0]) * 4 * 4)) - ipath_dbg("2nd group: dmacopy: %llx %llx " - "%llx %llx; shadow: %lx %lx " - "%lx %lx\n", - (unsigned long long) - le64_to_cpu(dma[4]), - (unsigned long long) - le64_to_cpu(dma[5]), - (unsigned long long) - le64_to_cpu(dma[6]), - (unsigned long long) - le64_to_cpu(dma[7]), - shadow[4], shadow[5], - shadow[6], shadow[7]); + ipath_force_pio_avail_update(dd); + ipath_update_pio_bufs(dd); + updated++; + i = first; + goto rescan; } + + no_pio_bufs(dd); buf = NULL; - goto bail; + } else { + if (i < dd->ipath_piobcnt2k) + buf = (u32 __iomem *) (dd->ipath_pio2kbase + + i * dd->ipath_palign); + else + buf = (u32 __iomem *) + (dd->ipath_pio4kbase + + (i - dd->ipath_piobcnt2k) * dd->ipath_4kalign); + if (pbufnum) + *pbufnum = i; } - /* - * set next starting place. Since it's just an optimization, - * it doesn't matter who wins on this, so no locking - */ - dd->ipath_lastpioindex = i + 1; - if (dd->ipath_upd_pio_shadow) - dd->ipath_upd_pio_shadow = 0; - if (dd->ipath_consec_nopiobuf) - dd->ipath_consec_nopiobuf = 0; - if (i < dd->ipath_piobcnt2k) - buf = (u32 __iomem *) (dd->ipath_pio2kbase + - i * dd->ipath_palign); - else - buf = (u32 __iomem *) - (dd->ipath_pio4kbase + - (i - dd->ipath_piobcnt2k) * dd->ipath_4kalign); - ipath_cdbg(VERBOSE, "Return piobuf%u %uk @ %p\n", - i, (i < dd->ipath_piobcnt2k) ? 2 : 4, buf); - if (pbufnum) - *pbufnum = i; + return buf; +} -bail: +/** + * ipath_getpiobuf - find an available pio buffer + * @dd: the infinipath device + * @plen: the size of the PIO buffer needed in 32-bit words + * @pbufnum: the buffer number is placed here + */ +u32 __iomem *ipath_getpiobuf(struct ipath_devdata *dd, u32 plen, u32 *pbufnum) +{ + u32 __iomem *buf; + u32 pnum, nbufs; + u32 first, lasti; + + if (plen + 1 >= IPATH_SMALLBUF_DWORDS) { + first = dd->ipath_piobcnt2k; + lasti = dd->ipath_lastpioindexl; + } else { + first = 0; + lasti = dd->ipath_lastpioindex; + } + nbufs = dd->ipath_piobcnt2k + dd->ipath_piobcnt4k; + buf = ipath_getpiobuf_range(dd, &pnum, first, nbufs, lasti); + + if (buf) { + /* + * Set next starting place. It's just an optimization, + * it doesn't matter who wins on this, so no locking + */ + if (plen + 1 >= IPATH_SMALLBUF_DWORDS) + dd->ipath_lastpioindexl = pnum + 1; + else + dd->ipath_lastpioindex = pnum + 1; + if (dd->ipath_upd_pio_shadow) + dd->ipath_upd_pio_shadow = 0; + if (dd->ipath_consec_nopiobuf) + dd->ipath_consec_nopiobuf = 0; + ipath_cdbg(VERBOSE, "Return piobuf%u %uk @ %p\n", + pnum, (pnum < dd->ipath_piobcnt2k) ? 2 : 4, buf); + if (pbufnum) + *pbufnum = pnum; + + } return buf; } /** + * ipath_chg_pioavailkernel - change which send buffers are available for kernel + * @dd: the infinipath device + * @start: the starting send buffer number + * @len: the number of send buffers + * @avail: true if the buffers are available for kernel use, false otherwise + */ +void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start, + unsigned len, int avail) +{ + unsigned long flags; + unsigned end; + + /* There are two bits per send buffer (busy and generation) */ + start *= 2; + len *= 2; + end = start + len; + + /* Set or clear the generation bits. */ + spin_lock_irqsave(&ipath_pioavail_lock, flags); + while (start < end) { + if (avail) { + __clear_bit(start + INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT, + dd->ipath_pioavailshadow); + __set_bit(start, dd->ipath_pioavailkernel); + } else { + __set_bit(start + INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT, + dd->ipath_pioavailshadow); + __clear_bit(start, dd->ipath_pioavailkernel); + } + start += 2; + } + spin_unlock_irqrestore(&ipath_pioavail_lock, flags); +} + +/** * ipath_create_rcvhdrq - create a receive header queue * @dd: the infinipath device * @pd: the port data @@ -1566,19 +1707,27 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd, ret = -ENOMEM; goto bail; } - pd->port_rcvhdrtail_kvaddr = dma_alloc_coherent( - &dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail, GFP_KERNEL); - if (!pd->port_rcvhdrtail_kvaddr) { - ipath_dev_err(dd, "attempt to allocate 1 page " - "for port %u rcvhdrqtailaddr failed\n", - pd->port_port); - ret = -ENOMEM; - dma_free_coherent(&dd->pcidev->dev, amt, - pd->port_rcvhdrq, pd->port_rcvhdrq_phys); - pd->port_rcvhdrq = NULL; - goto bail; + + if (!(dd->ipath_flags & IPATH_NODMA_RTAIL)) { + pd->port_rcvhdrtail_kvaddr = dma_alloc_coherent( + &dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail, + GFP_KERNEL); + if (!pd->port_rcvhdrtail_kvaddr) { + ipath_dev_err(dd, "attempt to allocate 1 page " + "for port %u rcvhdrqtailaddr " + "failed\n", pd->port_port); + ret = -ENOMEM; + dma_free_coherent(&dd->pcidev->dev, amt, + pd->port_rcvhdrq, + pd->port_rcvhdrq_phys); + pd->port_rcvhdrq = NULL; + goto bail; + } + pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail; + ipath_cdbg(VERBOSE, "port %d hdrtailaddr, %llx " + "physical\n", pd->port_port, + (unsigned long long) phys_hdrqtail); } - pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail; pd->port_rcvhdrq_size = amt; @@ -1588,10 +1737,6 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd, (unsigned long) pd->port_rcvhdrq_phys, (unsigned long) pd->port_rcvhdrq_size, pd->port_port); - - ipath_cdbg(VERBOSE, "port %d hdrtailaddr, %llx physical\n", - pd->port_port, - (unsigned long long) phys_hdrqtail); } else ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; " @@ -1615,7 +1760,6 @@ int ipath_create_rcvhdrq(struct ipath_devdata *dd, ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdraddr, pd->port_port, pd->port_rcvhdrq_phys); - ret = 0; bail: return ret; } @@ -1632,52 +1776,149 @@ bail: */ void ipath_cancel_sends(struct ipath_devdata *dd, int restore_sendctrl) { + unsigned long flags; + + if (dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) { + ipath_cdbg(VERBOSE, "Ignore while in autonegotiation\n"); + goto bail; + } + /* + * If we have SDMA, and it's not disabled, we have to kick off the + * abort state machine, provided we aren't already aborting. + * If we are in the process of aborting SDMA (!DISABLED, but ABORTING), + * we skip the rest of this routine. It is already "in progress" + */ + if (dd->ipath_flags & IPATH_HAS_SEND_DMA) { + int skip_cancel; + u64 *statp = &dd->ipath_sdma_status; + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + skip_cancel = + !test_bit(IPATH_SDMA_DISABLED, statp) && + test_and_set_bit(IPATH_SDMA_ABORTING, statp); + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + if (skip_cancel) + goto bail; + } + ipath_dbg("Cancelling all in-progress send buffers\n"); - dd->ipath_lastcancel = jiffies+HZ/2; /* skip armlaunch errs a bit */ + + /* skip armlaunch errs for a while */ + dd->ipath_lastcancel = jiffies + HZ / 2; + /* - * the abort bit is auto-clearing. We read scratch to be sure - * that cancels and the abort have taken effect in the chip. + * The abort bit is auto-clearing. We also don't want pioavail + * update happening during this, and we don't want any other + * sends going out, so turn those off for the duration. We read + * the scratch register to be sure that cancels and the abort + * have taken effect in the chip. Otherwise two parts are same + * as ipath_force_pio_avail_update() */ + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl &= ~(INFINIPATH_S_PIOBUFAVAILUPD + | INFINIPATH_S_PIOENABLE); ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - INFINIPATH_S_ABORT); + dd->ipath_sendctrl | INFINIPATH_S_ABORT); ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + + /* disarm all send buffers */ ipath_disarm_piobufs(dd, 0, - (unsigned)(dd->ipath_piobcnt2k + dd->ipath_piobcnt4k)); - if (restore_sendctrl) /* else done by caller later */ + dd->ipath_piobcnt2k + dd->ipath_piobcnt4k); + + if (restore_sendctrl) { + /* else done by caller later if needed */ + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl |= INFINIPATH_S_PIOBUFAVAILUPD | + INFINIPATH_S_PIOENABLE; ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl); + dd->ipath_sendctrl); + /* and again, be sure all have hit the chip */ + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + } - /* and again, be sure all have hit the chip */ - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + if ((dd->ipath_flags & IPATH_HAS_SEND_DMA) && + !test_bit(IPATH_SDMA_DISABLED, &dd->ipath_sdma_status) && + test_bit(IPATH_SDMA_RUNNING, &dd->ipath_sdma_status)) { + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + /* only wait so long for intr */ + dd->ipath_sdma_abort_intr_timeout = jiffies + HZ; + dd->ipath_sdma_reset_wait = 200; + __set_bit(IPATH_SDMA_DISARMED, &dd->ipath_sdma_status); + if (!test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + tasklet_hi_schedule(&dd->ipath_sdma_abort_task); + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + } +bail:; } +/* + * Force an update of in-memory copy of the pioavail registers, when + * needed for any of a variety of reasons. We read the scratch register + * to make it highly likely that the update will have happened by the + * time we return. If already off (as in cancel_sends above), this + * routine is a nop, on the assumption that the caller will "do the + * right thing". + */ +void ipath_force_pio_avail_update(struct ipath_devdata *dd) +{ + unsigned long flags; + + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + if (dd->ipath_sendctrl & INFINIPATH_S_PIOBUFAVAILUPD) { + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl & ~INFINIPATH_S_PIOBUFAVAILUPD); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + } + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); +} -static void ipath_set_ib_lstate(struct ipath_devdata *dd, int which) +static void ipath_set_ib_lstate(struct ipath_devdata *dd, int linkcmd, + int linitcmd) { + u64 mod_wd; static const char *what[4] = { [0] = "NOP", [INFINIPATH_IBCC_LINKCMD_DOWN] = "DOWN", [INFINIPATH_IBCC_LINKCMD_ARMED] = "ARMED", [INFINIPATH_IBCC_LINKCMD_ACTIVE] = "ACTIVE" }; - int linkcmd = (which >> INFINIPATH_IBCC_LINKCMD_SHIFT) & - INFINIPATH_IBCC_LINKCMD_MASK; - - ipath_cdbg(VERBOSE, "Trying to move unit %u to %s, current ltstate " - "is %s\n", dd->ipath_unit, - what[linkcmd], - ipath_ibcstatus_str[ - (ipath_read_kreg64 - (dd, dd->ipath_kregs->kr_ibcstatus) >> - INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) & - INFINIPATH_IBCS_LINKTRAININGSTATE_MASK]); - /* flush all queued sends when going to DOWN to be sure that - * they don't block MAD packets */ - if (linkcmd == INFINIPATH_IBCC_LINKCMD_DOWN) - ipath_cancel_sends(dd, 1); + + if (linitcmd == INFINIPATH_IBCC_LINKINITCMD_DISABLE) { + /* + * If we are told to disable, note that so link-recovery + * code does not attempt to bring us back up. + */ + preempt_disable(); + dd->ipath_flags |= IPATH_IB_LINK_DISABLED; + preempt_enable(); + } else if (linitcmd) { + /* + * Any other linkinitcmd will lead to LINKDOWN and then + * to INIT (if all is well), so clear flag to let + * link-recovery code attempt to bring us back up. + */ + preempt_disable(); + dd->ipath_flags &= ~IPATH_IB_LINK_DISABLED; + preempt_enable(); + } + + mod_wd = (linkcmd << dd->ibcc_lc_shift) | + (linitcmd << INFINIPATH_IBCC_LINKINITCMD_SHIFT); + ipath_cdbg(VERBOSE, + "Moving unit %u to %s (initcmd=0x%x), current ltstate is %s\n", + dd->ipath_unit, what[linkcmd], linitcmd, + ipath_ibcstatus_str[ipath_ib_linktrstate(dd, + ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus))]); ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, - dd->ipath_ibcctrl | which); + dd->ipath_ibcctrl | mod_wd); + /* read from chip so write is flushed */ + (void) ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); } int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) @@ -1687,30 +1928,28 @@ int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) switch (newstate) { case IPATH_IB_LINKDOWN_ONLY: - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN << - INFINIPATH_IBCC_LINKCMD_SHIFT); + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, 0); /* don't wait */ ret = 0; goto bail; case IPATH_IB_LINKDOWN: - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_POLL << - INFINIPATH_IBCC_LINKINITCMD_SHIFT); + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, + INFINIPATH_IBCC_LINKINITCMD_POLL); /* don't wait */ ret = 0; goto bail; case IPATH_IB_LINKDOWN_SLEEP: - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_SLEEP << - INFINIPATH_IBCC_LINKINITCMD_SHIFT); + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, + INFINIPATH_IBCC_LINKINITCMD_SLEEP); /* don't wait */ ret = 0; goto bail; case IPATH_IB_LINKDOWN_DISABLE: - ipath_set_ib_lstate(dd, - INFINIPATH_IBCC_LINKINITCMD_DISABLE << - INFINIPATH_IBCC_LINKINITCMD_SHIFT); + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, + INFINIPATH_IBCC_LINKINITCMD_DISABLE); /* don't wait */ ret = 0; goto bail; @@ -1725,8 +1964,8 @@ int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) ret = -EINVAL; goto bail; } - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED << - INFINIPATH_IBCC_LINKCMD_SHIFT); + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED, 0); + /* * Since the port can transition to ACTIVE by receiving * a non VL 15 packet, wait for either state. @@ -1743,8 +1982,7 @@ int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) ret = -EINVAL; goto bail; } - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE << - INFINIPATH_IBCC_LINKCMD_SHIFT); + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE, 0); lstate = IPATH_LINKACTIVE; break; @@ -1753,16 +1991,41 @@ int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) dd->ipath_ibcctrl |= INFINIPATH_IBCC_LOOPBACK; ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, dd->ipath_ibcctrl); + + /* turn heartbeat off, as it causes loopback to fail */ + dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, + IPATH_IB_HRTBT_OFF); + /* don't wait */ ret = 0; - goto bail; // no state change to wait for + goto bail; case IPATH_IB_LINK_EXTERNAL: - dev_info(&dd->pcidev->dev, "Disabling IB local loopback (normal)\n"); + dev_info(&dd->pcidev->dev, + "Disabling IB local loopback (normal)\n"); + dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, + IPATH_IB_HRTBT_ON); dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LOOPBACK; ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, dd->ipath_ibcctrl); + /* don't wait */ ret = 0; - goto bail; // no state change to wait for + goto bail; + + /* + * Heartbeat can be explicitly enabled by the user via + * "hrtbt_enable" "file", and if disabled, trying to enable here + * will have no effect. Implicit changes (heartbeat off when + * loopback on, and vice versa) are included to ease testing. + */ + case IPATH_IB_LINK_HRTBT: + ret = dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, + IPATH_IB_HRTBT_ON); + goto bail; + + case IPATH_IB_LINK_NO_HRTBT: + ret = dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, + IPATH_IB_HRTBT_OFF); + goto bail; default: ipath_dbg("Invalid linkstate 0x%x requested\n", newstate); @@ -1785,7 +2048,7 @@ bail: * sanity checking on this, and we don't deal with what happens to * programs that are already running when the size changes. * NOTE: changing the MTU will usually cause the IBC to go back to - * link initialize (IPATH_IBSTATE_INIT) state... + * link INIT state... */ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg) { @@ -1800,7 +2063,7 @@ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg) * piosize). We check that it's one of the valid IB sizes. */ if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 && - arg != 4096) { + (arg != 4096 || !ipath_mtu4096)) { ipath_dbg("Trying to set invalid mtu %u, failing\n", arg); ret = -EINVAL; goto bail; @@ -1816,6 +2079,8 @@ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg) if (arg >= (piosize - IPATH_PIO_MAXIBHDR)) { /* Only if it's not the initial value (or reset to it) */ if (piosize != dd->ipath_init_ibmaxlen) { + if (arg > piosize && arg <= dd->ipath_init_ibmaxlen) + piosize = dd->ipath_init_ibmaxlen; dd->ipath_ibmaxlen = piosize; changed = 1; } @@ -1829,24 +2094,17 @@ int ipath_set_mtu(struct ipath_devdata *dd, u16 arg) } if (changed) { + u64 ibc = dd->ipath_ibcctrl, ibdw; /* - * set the IBC maxpktlength to the size of our pio - * buffers in words + * update our housekeeping variables, and set IBC max + * size, same as init code; max IBC is max we allow in + * buffer, less the qword pbc, plus 1 for ICRC, in dwords */ - u64 ibc = dd->ipath_ibcctrl; + dd->ipath_ibmaxlen = piosize - 2 * sizeof(u32); + ibdw = (dd->ipath_ibmaxlen >> 2) + 1; ibc &= ~(INFINIPATH_IBCC_MAXPKTLEN_MASK << - INFINIPATH_IBCC_MAXPKTLEN_SHIFT); - - piosize = piosize - 2 * sizeof(u32); /* ignore pbc */ - dd->ipath_ibmaxlen = piosize; - piosize /= sizeof(u32); /* in words */ - /* - * for ICRC, which we only send in diag test pkt mode, and - * we don't need to worry about that for mtu - */ - piosize += 1; - - ibc |= piosize << INFINIPATH_IBCC_MAXPKTLEN_SHIFT; + dd->ibcc_mpl_shift); + ibc |= ibdw << dd->ibcc_mpl_shift; dd->ipath_ibcctrl = ibc; ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, dd->ipath_ibcctrl); @@ -1859,11 +2117,16 @@ bail: return ret; } -int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc) +int ipath_set_lid(struct ipath_devdata *dd, u32 lid, u8 lmc) { - dd->ipath_lid = arg; + dd->ipath_lid = lid; dd->ipath_lmc = lmc; + dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_LIDLMC, lid | + (~((1U << lmc) - 1)) << 16); + + dev_info(&dd->pcidev->dev, "We got a lid: 0x%x\n", lid); + return 0; } @@ -1925,10 +2188,8 @@ static void ipath_run_led_override(unsigned long opaque) * but leave that to per-chip functions. */ val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); - ltstate = (val >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) & - INFINIPATH_IBCS_LINKTRAININGSTATE_MASK; - lstate = (val >> INFINIPATH_IBCS_LINKSTATE_SHIFT) & - INFINIPATH_IBCS_LINKSTATE_MASK; + ltstate = ipath_ib_linktrstate(dd, val); + lstate = ipath_ib_linkstate(dd, val); dd->ipath_f_setextled(dd, lstate, ltstate); mod_timer(&dd->ipath_led_override_timer, jiffies + timeoff); @@ -1969,9 +2230,8 @@ void ipath_set_led_override(struct ipath_devdata *dd, unsigned int val) dd->ipath_led_override_timer.data = (unsigned long) dd; dd->ipath_led_override_timer.expires = jiffies + 1; add_timer(&dd->ipath_led_override_timer); - } else { + } else atomic_dec(&dd->ipath_led_override_timer_active); - } } /** @@ -1989,6 +2249,8 @@ void ipath_shutdown_device(struct ipath_devdata *dd) ipath_dbg("Shutting down the device\n"); + ipath_hol_up(dd); /* make sure user processes aren't suspended */ + dd->ipath_flags |= IPATH_LINKUNK; dd->ipath_flags &= ~(IPATH_INITTED | IPATH_LINKDOWN | IPATH_LINKINIT | IPATH_LINKARMED | @@ -2003,6 +2265,9 @@ void ipath_shutdown_device(struct ipath_devdata *dd) ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, dd->ipath_rcvctrl); + if (dd->ipath_flags & IPATH_HAS_SEND_DMA) + teardown_sdma(dd); + /* * gracefully stop all sends allowing any in progress to trickle out * first. @@ -2020,10 +2285,16 @@ void ipath_shutdown_device(struct ipath_devdata *dd) */ udelay(5); - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_DISABLE << - INFINIPATH_IBCC_LINKINITCMD_SHIFT); + dd->ipath_f_setextled(dd, 0, 0); /* make sure LEDs are off */ + + ipath_set_ib_lstate(dd, 0, INFINIPATH_IBCC_LINKINITCMD_DISABLE); ipath_cancel_sends(dd, 0); + /* + * we are shutting down, so tell components that care. We don't do + * this on just a link state change, much like ethernet, a cable + * unplug, etc. doesn't change driver state + */ signal_ib_event(dd, IB_EVENT_PORT_ERR); /* disable IBC */ @@ -2038,10 +2309,20 @@ void ipath_shutdown_device(struct ipath_devdata *dd) */ dd->ipath_f_quiet_serdes(dd); + /* stop all the timers that might still be running */ + del_timer_sync(&dd->ipath_hol_timer); if (dd->ipath_stats_timer_active) { del_timer_sync(&dd->ipath_stats_timer); dd->ipath_stats_timer_active = 0; } + if (dd->ipath_intrchk_timer.data) { + del_timer_sync(&dd->ipath_intrchk_timer); + dd->ipath_intrchk_timer.data = 0; + } + if (atomic_read(&dd->ipath_led_override_timer_active)) { + del_timer_sync(&dd->ipath_led_override_timer); + atomic_set(&dd->ipath_led_override_timer_active, 0); + } /* * clear all interrupts and errors, so that the next time the driver @@ -2115,13 +2396,13 @@ void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd) ipath_cdbg(VERBOSE, "free closed port %d " "ipath_port0_skbinfo @ %p\n", pd->port_port, skbinfo); - for (e = 0; e < dd->ipath_rcvegrcnt; e++) - if (skbinfo[e].skb) { - pci_unmap_single(dd->pcidev, skbinfo[e].phys, - dd->ipath_ibmaxlen, - PCI_DMA_FROMDEVICE); - dev_kfree_skb(skbinfo[e].skb); - } + for (e = 0; e < dd->ipath_p0_rcvegrcnt; e++) + if (skbinfo[e].skb) { + pci_unmap_single(dd->pcidev, skbinfo[e].phys, + dd->ipath_ibmaxlen, + PCI_DMA_FROMDEVICE); + dev_kfree_skb(skbinfo[e].skb); + } vfree(skbinfo); } kfree(pd->port_tid_pg_list); @@ -2144,6 +2425,7 @@ static int __init infinipath_init(void) */ idr_init(&unit_table); if (!idr_pre_get(&unit_table, GFP_KERNEL)) { + printk(KERN_ERR IPATH_DRV_NAME ": idr_pre_get() failed\n"); ret = -ENOMEM; goto bail; } @@ -2235,13 +2517,18 @@ int ipath_reset_device(int unit) } } + if (dd->ipath_flags & IPATH_HAS_SEND_DMA) + teardown_sdma(dd); + dd->ipath_flags &= ~IPATH_INITTED; + ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask, 0ULL); ret = dd->ipath_f_reset(dd); - if (ret != 1) - ipath_dbg("reset was not successful\n"); - ipath_dbg("Trying to reinitialize unit %u after reset attempt\n", - unit); - ret = ipath_init_chip(dd, 1); + if (ret == 1) { + ipath_dbg("Reinitializing unit %u after reset attempt\n", + unit); + ret = ipath_init_chip(dd, 1); + } else + ret = -EAGAIN; if (ret) ipath_dev_err(dd, "Reinitialize unit %u after " "reset failed with %d\n", unit, ret); @@ -2253,13 +2540,121 @@ bail: return ret; } +/* + * send a signal to all the processes that have the driver open + * through the normal interfaces (i.e., everything other than diags + * interface). Returns number of signalled processes. + */ +static int ipath_signal_procs(struct ipath_devdata *dd, int sig) +{ + int i, sub, any = 0; + pid_t pid; + + if (!dd->ipath_pd) + return 0; + for (i = 1; i < dd->ipath_cfgports; i++) { + if (!dd->ipath_pd[i] || !dd->ipath_pd[i]->port_cnt || + !dd->ipath_pd[i]->port_pid) + continue; + pid = dd->ipath_pd[i]->port_pid; + dev_info(&dd->pcidev->dev, "context %d in use " + "(PID %u), sending signal %d\n", + i, pid, sig); + kill_proc(pid, sig, 1); + any++; + for (sub = 0; sub < INFINIPATH_MAX_SUBPORT; sub++) { + pid = dd->ipath_pd[i]->port_subpid[sub]; + if (!pid) + continue; + dev_info(&dd->pcidev->dev, "sub-context " + "%d:%d in use (PID %u), sending " + "signal %d\n", i, sub, pid, sig); + kill_proc(pid, sig, 1); + any++; + } + } + return any; +} + +static void ipath_hol_signal_down(struct ipath_devdata *dd) +{ + if (ipath_signal_procs(dd, SIGSTOP)) + ipath_dbg("Stopped some processes\n"); + ipath_cancel_sends(dd, 1); +} + + +static void ipath_hol_signal_up(struct ipath_devdata *dd) +{ + if (ipath_signal_procs(dd, SIGCONT)) + ipath_dbg("Continued some processes\n"); +} + +/* + * link is down, stop any users processes, and flush pending sends + * to prevent HoL blocking, then start the HoL timer that + * periodically continues, then stop procs, so they can detect + * link down if they want, and do something about it. + * Timer may already be running, so use __mod_timer, not add_timer. + */ +void ipath_hol_down(struct ipath_devdata *dd) +{ + dd->ipath_hol_state = IPATH_HOL_DOWN; + ipath_hol_signal_down(dd); + dd->ipath_hol_next = IPATH_HOL_DOWNCONT; + dd->ipath_hol_timer.expires = jiffies + + msecs_to_jiffies(ipath_hol_timeout_ms); + __mod_timer(&dd->ipath_hol_timer, dd->ipath_hol_timer.expires); +} + +/* + * link is up, continue any user processes, and ensure timer + * is a nop, if running. Let timer keep running, if set; it + * will nop when it sees the link is up + */ +void ipath_hol_up(struct ipath_devdata *dd) +{ + ipath_hol_signal_up(dd); + dd->ipath_hol_state = IPATH_HOL_UP; +} + +/* + * toggle the running/not running state of user proceses + * to prevent HoL blocking on chip resources, but still allow + * user processes to do link down special case handling. + * Should only be called via the timer + */ +void ipath_hol_event(unsigned long opaque) +{ + struct ipath_devdata *dd = (struct ipath_devdata *)opaque; + + if (dd->ipath_hol_next == IPATH_HOL_DOWNSTOP + && dd->ipath_hol_state != IPATH_HOL_UP) { + dd->ipath_hol_next = IPATH_HOL_DOWNCONT; + ipath_dbg("Stopping processes\n"); + ipath_hol_signal_down(dd); + } else { /* may do "extra" if also in ipath_hol_up() */ + dd->ipath_hol_next = IPATH_HOL_DOWNSTOP; + ipath_dbg("Continuing processes\n"); + ipath_hol_signal_up(dd); + } + if (dd->ipath_hol_state == IPATH_HOL_UP) + ipath_dbg("link's up, don't resched timer\n"); + else { + dd->ipath_hol_timer.expires = jiffies + + msecs_to_jiffies(ipath_hol_timeout_ms); + __mod_timer(&dd->ipath_hol_timer, + dd->ipath_hol_timer.expires); + } +} + int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv) { u64 val; - if ( new_pol_inv > INFINIPATH_XGXS_RX_POL_MASK ) { + + if (new_pol_inv > INFINIPATH_XGXS_RX_POL_MASK) return -1; - } - if ( dd->ipath_rx_pol_inv != new_pol_inv ) { + if (dd->ipath_rx_pol_inv != new_pol_inv) { dd->ipath_rx_pol_inv = new_pol_inv; val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig); val &= ~(INFINIPATH_XGXS_RX_POL_MASK << diff --git a/drivers/infiniband/hw/ipath/ipath_eeprom.c b/drivers/infiniband/hw/ipath/ipath_eeprom.c index e28a42f5376..dc37277f1c8 100644 --- a/drivers/infiniband/hw/ipath/ipath_eeprom.c +++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -62,6 +62,33 @@ * accessing eeprom contents from within the kernel, only via sysfs. */ +/* Added functionality for IBA7220-based cards */ +#define IPATH_EEPROM_DEV_V1 0xA0 +#define IPATH_EEPROM_DEV_V2 0xA2 +#define IPATH_TEMP_DEV 0x98 +#define IPATH_BAD_DEV (IPATH_EEPROM_DEV_V2+2) +#define IPATH_NO_DEV (0xFF) + +/* + * The number of I2C chains is proliferating. Table below brings + * some order to the madness. The basic principle is that the + * table is scanned from the top, and a "probe" is made to the + * device probe_dev. If that succeeds, the chain is considered + * to be of that type, and dd->i2c_chain_type is set to the index+1 + * of the entry. + * The +1 is so static initialization can mean "unknown, do probe." + */ +static struct i2c_chain_desc { + u8 probe_dev; /* If seen at probe, chain is this type */ + u8 eeprom_dev; /* Dev addr (if any) for EEPROM */ + u8 temp_dev; /* Dev Addr (if any) for Temp-sense */ +} i2c_chains[] = { + { IPATH_BAD_DEV, IPATH_NO_DEV, IPATH_NO_DEV }, /* pre-iba7220 bds */ + { IPATH_EEPROM_DEV_V1, IPATH_EEPROM_DEV_V1, IPATH_TEMP_DEV}, /* V1 */ + { IPATH_EEPROM_DEV_V2, IPATH_EEPROM_DEV_V2, IPATH_TEMP_DEV}, /* V2 */ + { IPATH_NO_DEV } +}; + enum i2c_type { i2c_line_scl = 0, i2c_line_sda @@ -75,13 +102,6 @@ enum i2c_state { #define READ_CMD 1 #define WRITE_CMD 0 -static int eeprom_init; - -/* - * The gpioval manipulation really should be protected by spinlocks - * or be converted to use atomic operations. - */ - /** * i2c_gpio_set - set a GPIO line * @dd: the infinipath device @@ -241,6 +261,27 @@ static int i2c_ackrcv(struct ipath_devdata *dd) } /** + * rd_byte - read a byte, leaving ACK, STOP, etc up to caller + * @dd: the infinipath device + * + * Returns byte shifted out of device + */ +static int rd_byte(struct ipath_devdata *dd) +{ + int bit_cntr, data; + + data = 0; + + for (bit_cntr = 7; bit_cntr >= 0; --bit_cntr) { + data <<= 1; + scl_out(dd, i2c_line_high); + data |= sda_in(dd, 0); + scl_out(dd, i2c_line_low); + } + return data; +} + +/** * wr_byte - write a byte, one bit at a time * @dd: the infinipath device * @data: the byte to write @@ -331,7 +372,6 @@ static int eeprom_reset(struct ipath_devdata *dd) ipath_cdbg(VERBOSE, "Resetting i2c eeprom; initial gpioout reg " "is %llx\n", (unsigned long long) *gpioval); - eeprom_init = 1; /* * This is to get the i2c into a known state, by first going low, * then tristate sda (and then tristate scl as first thing @@ -340,12 +380,17 @@ static int eeprom_reset(struct ipath_devdata *dd) scl_out(dd, i2c_line_low); sda_out(dd, i2c_line_high); + /* Clock up to 9 cycles looking for SDA hi, then issue START and STOP */ while (clock_cycles_left--) { scl_out(dd, i2c_line_high); + /* SDA seen high, issue START by dropping it while SCL high */ if (sda_in(dd, 0)) { sda_out(dd, i2c_line_low); scl_out(dd, i2c_line_low); + /* ATMEL spec says must be followed by STOP. */ + scl_out(dd, i2c_line_high); + sda_out(dd, i2c_line_high); ret = 0; goto bail; } @@ -359,29 +404,121 @@ bail: return ret; } -/** - * ipath_eeprom_read - receives bytes from the eeprom via I2C - * @dd: the infinipath device - * @eeprom_offset: address to read from - * @buffer: where to store result - * @len: number of bytes to receive +/* + * Probe for I2C device at specified address. Returns 0 for "success" + * to match rest of this file. + * Leave bus in "reasonable" state for further commands. */ +static int i2c_probe(struct ipath_devdata *dd, int devaddr) +{ + int ret = 0; + + ret = eeprom_reset(dd); + if (ret) { + ipath_dev_err(dd, "Failed reset probing device 0x%02X\n", + devaddr); + return ret; + } + /* + * Reset no longer leaves bus in start condition, so normal + * i2c_startcmd() will do. + */ + ret = i2c_startcmd(dd, devaddr | READ_CMD); + if (ret) + ipath_cdbg(VERBOSE, "Failed startcmd for device 0x%02X\n", + devaddr); + else { + /* + * Device did respond. Complete a single-byte read, because some + * devices apparently cannot handle STOP immediately after they + * ACK the start-cmd. + */ + int data; + data = rd_byte(dd); + stop_cmd(dd); + ipath_cdbg(VERBOSE, "Response from device 0x%02X\n", devaddr); + } + return ret; +} + +/* + * Returns the "i2c type". This is a pointer to a struct that describes + * the I2C chain on this board. To minimize impact on struct ipath_devdata, + * the (small integer) index into the table is actually memoized, rather + * then the pointer. + * Memoization is because the type is determined on the first call per chip. + * An alternative would be to move type determination to early + * init code. + */ +static struct i2c_chain_desc *ipath_i2c_type(struct ipath_devdata *dd) +{ + int idx; + + /* Get memoized index, from previous successful probes */ + idx = dd->ipath_i2c_chain_type - 1; + if (idx >= 0 && idx < (ARRAY_SIZE(i2c_chains) - 1)) + goto done; + + idx = 0; + while (i2c_chains[idx].probe_dev != IPATH_NO_DEV) { + /* if probe succeeds, this is type */ + if (!i2c_probe(dd, i2c_chains[idx].probe_dev)) + break; + ++idx; + } + + /* + * Old EEPROM (first entry) may require a reset after probe, + * rather than being able to "start" after "stop" + */ + if (idx == 0) + eeprom_reset(dd); + + if (i2c_chains[idx].probe_dev == IPATH_NO_DEV) + idx = -1; + else + dd->ipath_i2c_chain_type = idx + 1; +done: + return (idx >= 0) ? i2c_chains + idx : NULL; +} static int ipath_eeprom_internal_read(struct ipath_devdata *dd, u8 eeprom_offset, void *buffer, int len) { - /* compiler complains unless initialized */ - u8 single_byte = 0; - int bit_cntr; int ret; + struct i2c_chain_desc *icd; + u8 *bp = buffer; - if (!eeprom_init) - eeprom_reset(dd); - - eeprom_offset = (eeprom_offset << 1) | READ_CMD; + ret = 1; + icd = ipath_i2c_type(dd); + if (!icd) + goto bail; - if (i2c_startcmd(dd, eeprom_offset)) { - ipath_dbg("Failed startcmd\n"); + if (icd->eeprom_dev == IPATH_NO_DEV) { + /* legacy not-really-I2C */ + ipath_cdbg(VERBOSE, "Start command only address\n"); + eeprom_offset = (eeprom_offset << 1) | READ_CMD; + ret = i2c_startcmd(dd, eeprom_offset); + } else { + /* Actual I2C */ + ipath_cdbg(VERBOSE, "Start command uses devaddr\n"); + if (i2c_startcmd(dd, icd->eeprom_dev | WRITE_CMD)) { + ipath_dbg("Failed EEPROM startcmd\n"); + stop_cmd(dd); + ret = 1; + goto bail; + } + ret = wr_byte(dd, eeprom_offset); + stop_cmd(dd); + if (ret) { + ipath_dev_err(dd, "Failed to write EEPROM address\n"); + ret = 1; + goto bail; + } + ret = i2c_startcmd(dd, icd->eeprom_dev | READ_CMD); + } + if (ret) { + ipath_dbg("Failed startcmd for dev %02X\n", icd->eeprom_dev); stop_cmd(dd); ret = 1; goto bail; @@ -392,22 +529,11 @@ static int ipath_eeprom_internal_read(struct ipath_devdata *dd, * incrementing the address. */ while (len-- > 0) { - /* get data */ - single_byte = 0; - for (bit_cntr = 8; bit_cntr; bit_cntr--) { - u8 bit; - scl_out(dd, i2c_line_high); - bit = sda_in(dd, 0); - single_byte |= bit << (bit_cntr - 1); - scl_out(dd, i2c_line_low); - } - + /* get and store data */ + *bp++ = rd_byte(dd); /* send ack if not the last byte */ if (len) send_ack(dd); - - *((u8 *) buffer) = single_byte; - buffer++; } stop_cmd(dd); @@ -418,31 +544,40 @@ bail: return ret; } - -/** - * ipath_eeprom_write - writes data to the eeprom via I2C - * @dd: the infinipath device - * @eeprom_offset: where to place data - * @buffer: data to write - * @len: number of bytes to write - */ static int ipath_eeprom_internal_write(struct ipath_devdata *dd, u8 eeprom_offset, const void *buffer, int len) { - u8 single_byte; int sub_len; const u8 *bp = buffer; int max_wait_time, i; int ret; + struct i2c_chain_desc *icd; - if (!eeprom_init) - eeprom_reset(dd); + ret = 1; + icd = ipath_i2c_type(dd); + if (!icd) + goto bail; while (len > 0) { - if (i2c_startcmd(dd, (eeprom_offset << 1) | WRITE_CMD)) { - ipath_dbg("Failed to start cmd offset %u\n", - eeprom_offset); - goto failed_write; + if (icd->eeprom_dev == IPATH_NO_DEV) { + if (i2c_startcmd(dd, + (eeprom_offset << 1) | WRITE_CMD)) { + ipath_dbg("Failed to start cmd offset %u\n", + eeprom_offset); + goto failed_write; + } + } else { + /* Real I2C */ + if (i2c_startcmd(dd, icd->eeprom_dev | WRITE_CMD)) { + ipath_dbg("Failed EEPROM startcmd\n"); + goto failed_write; + } + ret = wr_byte(dd, eeprom_offset); + if (ret) { + ipath_dev_err(dd, "Failed to write EEPROM " + "address\n"); + goto failed_write; + } } sub_len = min(len, 4); @@ -468,9 +603,11 @@ static int ipath_eeprom_internal_write(struct ipath_devdata *dd, u8 eeprom_offse * the writes have completed. We do this inline to avoid * the debug prints that are in the real read routine * if the startcmd fails. + * We also use the proper device address, so it doesn't matter + * whether we have real eeprom_dev. legacy likes any address. */ max_wait_time = 100; - while (i2c_startcmd(dd, READ_CMD)) { + while (i2c_startcmd(dd, icd->eeprom_dev | READ_CMD)) { stop_cmd(dd); if (!--max_wait_time) { ipath_dbg("Did not get successful read to " @@ -478,15 +615,8 @@ static int ipath_eeprom_internal_write(struct ipath_devdata *dd, u8 eeprom_offse goto failed_write; } } - /* now read the zero byte */ - for (i = single_byte = 0; i < 8; i++) { - u8 bit; - scl_out(dd, i2c_line_high); - bit = sda_in(dd, 0); - scl_out(dd, i2c_line_low); - single_byte <<= 1; - single_byte |= bit; - } + /* now read (and ignore) the resulting byte */ + rd_byte(dd); stop_cmd(dd); } @@ -501,9 +631,12 @@ bail: return ret; } -/* - * The public entry-points ipath_eeprom_read() and ipath_eeprom_write() - * are now just wrappers around the internal functions. +/** + * ipath_eeprom_read - receives bytes from the eeprom via I2C + * @dd: the infinipath device + * @eeprom_offset: address to read from + * @buffer: where to store result + * @len: number of bytes to receive */ int ipath_eeprom_read(struct ipath_devdata *dd, u8 eeprom_offset, void *buff, int len) @@ -519,6 +652,13 @@ int ipath_eeprom_read(struct ipath_devdata *dd, u8 eeprom_offset, return ret; } +/** + * ipath_eeprom_write - writes data to the eeprom via I2C + * @dd: the infinipath device + * @eeprom_offset: where to place data + * @buffer: data to write + * @len: number of bytes to write + */ int ipath_eeprom_write(struct ipath_devdata *dd, u8 eeprom_offset, const void *buff, int len) { @@ -820,7 +960,7 @@ int ipath_update_eeprom_log(struct ipath_devdata *dd) * if we log an hour at 31 minutes, then we would need to set * active_time to -29 to accurately count the _next_ hour. */ - if (new_time > 3600) { + if (new_time >= 3600) { new_hrs = new_time / 3600; atomic_sub((new_hrs * 3600), &dd->ipath_active_time); new_hrs += dd->ipath_eep_hrs; @@ -885,3 +1025,159 @@ void ipath_inc_eeprom_err(struct ipath_devdata *dd, u32 eidx, u32 incr) spin_unlock_irqrestore(&dd->ipath_eep_st_lock, flags); return; } + +static int ipath_tempsense_internal_read(struct ipath_devdata *dd, u8 regnum) +{ + int ret; + struct i2c_chain_desc *icd; + + ret = -ENOENT; + + icd = ipath_i2c_type(dd); + if (!icd) + goto bail; + + if (icd->temp_dev == IPATH_NO_DEV) { + /* tempsense only exists on new, real-I2C boards */ + ret = -ENXIO; + goto bail; + } + + if (i2c_startcmd(dd, icd->temp_dev | WRITE_CMD)) { + ipath_dbg("Failed tempsense startcmd\n"); + stop_cmd(dd); + ret = -ENXIO; + goto bail; + } + ret = wr_byte(dd, regnum); + stop_cmd(dd); + if (ret) { + ipath_dev_err(dd, "Failed tempsense WR command %02X\n", + regnum); + ret = -ENXIO; + goto bail; + } + if (i2c_startcmd(dd, icd->temp_dev | READ_CMD)) { + ipath_dbg("Failed tempsense RD startcmd\n"); + stop_cmd(dd); + ret = -ENXIO; + goto bail; + } + /* + * We can only clock out one byte per command, sensibly + */ + ret = rd_byte(dd); + stop_cmd(dd); + +bail: + return ret; +} + +#define VALID_TS_RD_REG_MASK 0xBF + +/** + * ipath_tempsense_read - read register of temp sensor via I2C + * @dd: the infinipath device + * @regnum: register to read from + * + * returns reg contents (0..255) or < 0 for error + */ +int ipath_tempsense_read(struct ipath_devdata *dd, u8 regnum) +{ + int ret; + + if (regnum > 7) + return -EINVAL; + + /* return a bogus value for (the one) register we do not have */ + if (!((1 << regnum) & VALID_TS_RD_REG_MASK)) + return 0; + + ret = mutex_lock_interruptible(&dd->ipath_eep_lock); + if (!ret) { + ret = ipath_tempsense_internal_read(dd, regnum); + mutex_unlock(&dd->ipath_eep_lock); + } + + /* + * There are three possibilities here: + * ret is actual value (0..255) + * ret is -ENXIO or -EINVAL from code in this file + * ret is -EINTR from mutex_lock_interruptible. + */ + return ret; +} + +static int ipath_tempsense_internal_write(struct ipath_devdata *dd, + u8 regnum, u8 data) +{ + int ret = -ENOENT; + struct i2c_chain_desc *icd; + + icd = ipath_i2c_type(dd); + if (!icd) + goto bail; + + if (icd->temp_dev == IPATH_NO_DEV) { + /* tempsense only exists on new, real-I2C boards */ + ret = -ENXIO; + goto bail; + } + if (i2c_startcmd(dd, icd->temp_dev | WRITE_CMD)) { + ipath_dbg("Failed tempsense startcmd\n"); + stop_cmd(dd); + ret = -ENXIO; + goto bail; + } + ret = wr_byte(dd, regnum); + if (ret) { + stop_cmd(dd); + ipath_dev_err(dd, "Failed to write tempsense command %02X\n", + regnum); + ret = -ENXIO; + goto bail; + } + ret = wr_byte(dd, data); + stop_cmd(dd); + ret = i2c_startcmd(dd, icd->temp_dev | READ_CMD); + if (ret) { + ipath_dev_err(dd, "Failed tempsense data wrt to %02X\n", + regnum); + ret = -ENXIO; + } + +bail: + return ret; +} + +#define VALID_TS_WR_REG_MASK ((1 << 9) | (1 << 0xB) | (1 << 0xD)) + +/** + * ipath_tempsense_write - write register of temp sensor via I2C + * @dd: the infinipath device + * @regnum: register to write + * @data: data to write + * + * returns 0 for success or < 0 for error + */ +int ipath_tempsense_write(struct ipath_devdata *dd, u8 regnum, u8 data) +{ + int ret; + + if (regnum > 15 || !((1 << regnum) & VALID_TS_WR_REG_MASK)) + return -EINVAL; + + ret = mutex_lock_interruptible(&dd->ipath_eep_lock); + if (!ret) { + ret = ipath_tempsense_internal_write(dd, regnum, data); + mutex_unlock(&dd->ipath_eep_lock); + } + + /* + * There are three possibilities here: + * ret is 0 for success + * ret is -ENXIO or -EINVAL from code in this file + * ret is -EINTR from mutex_lock_interruptible. + */ + return ret; +} diff --git a/drivers/infiniband/hw/ipath/ipath_file_ops.c b/drivers/infiniband/hw/ipath/ipath_file_ops.c index 7e025c8e01b..1e627aab18b 100644 --- a/drivers/infiniband/hw/ipath/ipath_file_ops.c +++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -36,21 +36,28 @@ #include <linux/cdev.h> #include <linux/swap.h> #include <linux/vmalloc.h> +#include <linux/highmem.h> +#include <linux/io.h> +#include <linux/jiffies.h> #include <asm/pgtable.h> #include "ipath_kernel.h" #include "ipath_common.h" +#include "ipath_user_sdma.h" static int ipath_open(struct inode *, struct file *); static int ipath_close(struct inode *, struct file *); static ssize_t ipath_write(struct file *, const char __user *, size_t, loff_t *); +static ssize_t ipath_writev(struct kiocb *, const struct iovec *, + unsigned long , loff_t); static unsigned int ipath_poll(struct file *, struct poll_table_struct *); static int ipath_mmap(struct file *, struct vm_area_struct *); static const struct file_operations ipath_file_ops = { .owner = THIS_MODULE, .write = ipath_write, + .aio_write = ipath_writev, .open = ipath_open, .release = ipath_close, .poll = ipath_poll, @@ -184,6 +191,29 @@ static int ipath_get_base_info(struct file *fp, kinfo->spi_piobufbase = (u64) pd->port_piobufs + dd->ipath_palign * kinfo->spi_piocnt * slave; } + + /* + * Set the PIO avail update threshold to no larger + * than the number of buffers per process. Note that + * we decrease it here, but won't ever increase it. + */ + if (dd->ipath_pioupd_thresh && + kinfo->spi_piocnt < dd->ipath_pioupd_thresh) { + unsigned long flags; + + dd->ipath_pioupd_thresh = kinfo->spi_piocnt; + ipath_dbg("Decreased pio update threshold to %u\n", + dd->ipath_pioupd_thresh); + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl &= ~(INFINIPATH_S_UPDTHRESH_MASK + << INFINIPATH_S_UPDTHRESH_SHIFT); + dd->ipath_sendctrl |= dd->ipath_pioupd_thresh + << INFINIPATH_S_UPDTHRESH_SHIFT; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + } + if (shared) { kinfo->spi_port_uregbase = (u64) dd->ipath_uregbase + dd->ipath_ureg_align * pd->port_port; @@ -219,7 +249,12 @@ static int ipath_get_base_info(struct file *fp, kinfo->spi_pioalign = dd->ipath_palign; kinfo->spi_qpair = IPATH_KD_QP; - kinfo->spi_piosize = dd->ipath_ibmaxlen; + /* + * user mode PIO buffers are always 2KB, even when 4KB can + * be received, and sent via the kernel; this is ibmaxlen + * for 2K MTU. + */ + kinfo->spi_piosize = dd->ipath_piosize2k - 2 * sizeof(u32); kinfo->spi_mtu = dd->ipath_ibmaxlen; /* maxlen, not ibmtu */ kinfo->spi_port = pd->port_port; kinfo->spi_subport = subport_fp(fp); @@ -1598,6 +1633,9 @@ static int try_alloc_port(struct ipath_devdata *dd, int port, port_fp(fp) = pd; pd->port_pid = current->pid; strncpy(pd->port_comm, current->comm, sizeof(pd->port_comm)); + ipath_chg_pioavailkernel(dd, + dd->ipath_pbufsport * (pd->port_port - 1), + dd->ipath_pbufsport, 0); ipath_stats.sps_ports++; ret = 0; } else @@ -1760,7 +1798,7 @@ static int find_shared_port(struct file *fp, for (ndev = 0; ndev < devmax; ndev++) { struct ipath_devdata *dd = ipath_lookup(ndev); - if (!dd) + if (!usable(dd)) continue; for (i = 1; i < dd->ipath_cfgports; i++) { struct ipath_portdata *pd = dd->ipath_pd[i]; @@ -1839,10 +1877,9 @@ static int ipath_assign_port(struct file *fp, if (ipath_compatible_subports(swmajor, swminor) && uinfo->spu_subport_cnt && (ret = find_shared_port(fp, uinfo))) { - mutex_unlock(&ipath_mutex); if (ret > 0) ret = 0; - goto done; + goto done_chk_sdma; } i_minor = iminor(fp->f_path.dentry->d_inode) - IPATH_USER_MINOR_BASE; @@ -1854,6 +1891,21 @@ static int ipath_assign_port(struct file *fp, else ret = find_best_unit(fp, uinfo); +done_chk_sdma: + if (!ret) { + struct ipath_filedata *fd = fp->private_data; + const struct ipath_portdata *pd = fd->pd; + const struct ipath_devdata *dd = pd->port_dd; + + fd->pq = ipath_user_sdma_queue_create(&dd->pcidev->dev, + dd->ipath_unit, + pd->port_port, + fd->subport); + + if (!fd->pq) + ret = -ENOMEM; + } + mutex_unlock(&ipath_mutex); done: @@ -1922,22 +1974,25 @@ static int ipath_do_user_init(struct file *fp, pd->port_hdrqfull_poll = pd->port_hdrqfull; /* - * now enable the port; the tail registers will be written to memory - * by the chip as soon as it sees the write to - * dd->ipath_kregs->kr_rcvctrl. The update only happens on - * transition from 0 to 1, so clear it first, then set it as part of - * enabling the port. This will (very briefly) affect any other - * open ports, but it shouldn't be long enough to be an issue. - * We explictly set the in-memory copy to 0 beforehand, so we don't - * have to wait to be sure the DMA update has happened. + * Now enable the port for receive. + * For chips that are set to DMA the tail register to memory + * when they change (and when the update bit transitions from + * 0 to 1. So for those chips, we turn it off and then back on. + * This will (very briefly) affect any other open ports, but the + * duration is very short, and therefore isn't an issue. We + * explictly set the in-memory tail copy to 0 beforehand, so we + * don't have to wait to be sure the DMA update has happened + * (chip resets head/tail to 0 on transition to enable). */ - if (pd->port_rcvhdrtail_kvaddr) - ipath_clear_rcvhdrtail(pd); set_bit(dd->ipath_r_portenable_shift + pd->port_port, &dd->ipath_rcvctrl); - ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, + if (!(dd->ipath_flags & IPATH_NODMA_RTAIL)) { + if (pd->port_rcvhdrtail_kvaddr) + ipath_clear_rcvhdrtail(pd); + ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, dd->ipath_rcvctrl & ~(1ULL << dd->ipath_r_tailupd_shift)); + } ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, dd->ipath_rcvctrl); /* Notify any waiting slaves */ @@ -1965,14 +2020,15 @@ static void unlock_expected_tids(struct ipath_portdata *pd) ipath_cdbg(VERBOSE, "Port %u unlocking any locked expTID pages\n", pd->port_port); for (i = port_tidbase; i < maxtid; i++) { - if (!dd->ipath_pageshadow[i]) + struct page *ps = dd->ipath_pageshadow[i]; + + if (!ps) continue; + dd->ipath_pageshadow[i] = NULL; pci_unmap_page(dd->pcidev, dd->ipath_physshadow[i], PAGE_SIZE, PCI_DMA_FROMDEVICE); - ipath_release_user_pages_on_close(&dd->ipath_pageshadow[i], - 1); - dd->ipath_pageshadow[i] = NULL; + ipath_release_user_pages_on_close(&ps, 1); cnt++; ipath_stats.sps_pageunlocks++; } @@ -2007,6 +2063,13 @@ static int ipath_close(struct inode *in, struct file *fp) mutex_unlock(&ipath_mutex); goto bail; } + + dd = pd->port_dd; + + /* drain user sdma queue */ + ipath_user_sdma_queue_drain(dd, fd->pq); + ipath_user_sdma_queue_destroy(fd->pq); + if (--pd->port_cnt) { /* * XXX If the master closes the port before the slave(s), @@ -2019,7 +2082,6 @@ static int ipath_close(struct inode *in, struct file *fp) goto bail; } port = pd->port_port; - dd = pd->port_dd; if (pd->port_hdrqfull) { ipath_cdbg(PROC, "%s[%u] had %u rcvhdrqfull errors " @@ -2039,7 +2101,7 @@ static int ipath_close(struct inode *in, struct file *fp) pd->port_rcvnowait = pd->port_pionowait = 0; } if (pd->port_flag) { - ipath_dbg("port %u port_flag still set to 0x%lx\n", + ipath_cdbg(PROC, "port %u port_flag set: 0x%lx\n", pd->port_port, pd->port_flag); pd->port_flag = 0; } @@ -2076,6 +2138,7 @@ static int ipath_close(struct inode *in, struct file *fp) i = dd->ipath_pbufsport * (port - 1); ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport); + ipath_chg_pioavailkernel(dd, i, dd->ipath_pbufsport, 1); dd->ipath_f_clear_tids(dd, pd->port_port); @@ -2140,17 +2203,31 @@ static int ipath_get_slave_info(struct ipath_portdata *pd, return ret; } -static int ipath_force_pio_avail_update(struct ipath_devdata *dd) +static int ipath_sdma_get_inflight(struct ipath_user_sdma_queue *pq, + u32 __user *inflightp) { - unsigned long flags; + const u32 val = ipath_user_sdma_inflight_counter(pq); - spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl & ~INFINIPATH_S_PIOBUFAVAILUPD); - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); - spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + if (put_user(val, inflightp)) + return -EFAULT; + + return 0; +} + +static int ipath_sdma_get_complete(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + u32 __user *completep) +{ + u32 val; + int err; + + err = ipath_user_sdma_make_progress(dd, pq); + if (err < 0) + return err; + + val = ipath_user_sdma_complete_counter(pq); + if (put_user(val, completep)) + return -EFAULT; return 0; } @@ -2229,6 +2306,16 @@ static ssize_t ipath_write(struct file *fp, const char __user *data, dest = &cmd.cmd.armlaunch_ctrl; src = &ucmd->cmd.armlaunch_ctrl; break; + case IPATH_CMD_SDMA_INFLIGHT: + copy = sizeof(cmd.cmd.sdma_inflight); + dest = &cmd.cmd.sdma_inflight; + src = &ucmd->cmd.sdma_inflight; + break; + case IPATH_CMD_SDMA_COMPLETE: + copy = sizeof(cmd.cmd.sdma_complete); + dest = &cmd.cmd.sdma_complete; + src = &ucmd->cmd.sdma_complete; + break; default: ret = -EINVAL; goto bail; @@ -2299,7 +2386,7 @@ static ssize_t ipath_write(struct file *fp, const char __user *data, cmd.cmd.slave_mask_addr); break; case IPATH_CMD_PIOAVAILUPD: - ret = ipath_force_pio_avail_update(pd->port_dd); + ipath_force_pio_avail_update(pd->port_dd); break; case IPATH_CMD_POLL_TYPE: pd->poll_type = cmd.cmd.poll_type; @@ -2310,6 +2397,17 @@ static ssize_t ipath_write(struct file *fp, const char __user *data, else ipath_disable_armlaunch(pd->port_dd); break; + case IPATH_CMD_SDMA_INFLIGHT: + ret = ipath_sdma_get_inflight(user_sdma_queue_fp(fp), + (u32 __user *) (unsigned long) + cmd.cmd.sdma_inflight); + break; + case IPATH_CMD_SDMA_COMPLETE: + ret = ipath_sdma_get_complete(pd->port_dd, + user_sdma_queue_fp(fp), + (u32 __user *) (unsigned long) + cmd.cmd.sdma_complete); + break; } if (ret >= 0) @@ -2319,6 +2417,20 @@ bail: return ret; } +static ssize_t ipath_writev(struct kiocb *iocb, const struct iovec *iov, + unsigned long dim, loff_t off) +{ + struct file *filp = iocb->ki_filp; + struct ipath_filedata *fp = filp->private_data; + struct ipath_portdata *pd = port_fp(filp); + struct ipath_user_sdma_queue *pq = fp->pq; + + if (!dim) + return -EINVAL; + + return ipath_user_sdma_writev(pd->port_dd, pq, iov, dim); +} + static struct class *ipath_class; static int init_cdev(int minor, char *name, const struct file_operations *fops, diff --git a/drivers/infiniband/hw/ipath/ipath_iba6110.c b/drivers/infiniband/hw/ipath/ipath_iba6110.c index 9e2ced3cdc5..02831ad070b 100644 --- a/drivers/infiniband/hw/ipath/ipath_iba6110.c +++ b/drivers/infiniband/hw/ipath/ipath_iba6110.c @@ -40,6 +40,7 @@ #include <linux/pci.h> #include <linux/delay.h> #include <linux/htirq.h> +#include <rdma/ib_verbs.h> #include "ipath_kernel.h" #include "ipath_registers.h" @@ -305,7 +306,9 @@ static const struct ipath_cregs ipath_ht_cregs = { /* kr_intstatus, kr_intclear, kr_intmask bits */ #define INFINIPATH_I_RCVURG_MASK ((1U<<9)-1) +#define INFINIPATH_I_RCVURG_SHIFT 0 #define INFINIPATH_I_RCVAVAIL_MASK ((1U<<9)-1) +#define INFINIPATH_I_RCVAVAIL_SHIFT 12 /* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */ #define INFINIPATH_HWE_HTCMEMPARITYERR_SHIFT 0 @@ -476,7 +479,13 @@ static const struct ipath_hwerror_msgs ipath_6110_hwerror_msgs[] = { #define RXE_EAGER_PARITY (INFINIPATH_HWE_RXEMEMPARITYERR_EAGERTID \ << INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) -static int ipath_ht_txe_recover(struct ipath_devdata *); +static void ipath_ht_txe_recover(struct ipath_devdata *dd) +{ + ++ipath_stats.sps_txeparity; + dev_info(&dd->pcidev->dev, + "Recovering from TXE PIO parity error\n"); +} + /** * ipath_ht_handle_hwerrors - display hardware errors. @@ -557,11 +566,11 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg, * occur if a processor speculative read is done to the PIO * buffer while we are sending a packet, for example. */ - if ((hwerrs & TXE_PIO_PARITY) && ipath_ht_txe_recover(dd)) + if (hwerrs & TXE_PIO_PARITY) { + ipath_ht_txe_recover(dd); hwerrs &= ~TXE_PIO_PARITY; - if (hwerrs & RXE_EAGER_PARITY) - ipath_dev_err(dd, "RXE parity, Eager TID error is not " - "recoverable\n"); + } + if (!hwerrs) { ipath_dbg("Clearing freezemode on ignored or " "recovered hardware error\n"); @@ -735,11 +744,10 @@ static int ipath_ht_boardname(struct ipath_devdata *dd, char *name, */ dd->ipath_flags |= IPATH_32BITCOUNTERS; dd->ipath_flags |= IPATH_GPIO_INTR; - if (dd->ipath_htspeed != 800) + if (dd->ipath_lbus_speed != 800) ipath_dev_err(dd, "Incorrectly configured for HT @ %uMHz\n", - dd->ipath_htspeed); - ret = 0; + dd->ipath_lbus_speed); /* * set here, not in ipath_init_*_funcs because we have to do @@ -839,7 +847,7 @@ static void slave_or_pri_blk(struct ipath_devdata *dd, struct pci_dev *pdev, /* * now write them back to clear the error. */ - pci_write_config_byte(pdev, link_off, + pci_write_config_word(pdev, link_off, linkctrl & (0xf << 8)); } } @@ -904,7 +912,7 @@ static void slave_or_pri_blk(struct ipath_devdata *dd, struct pci_dev *pdev, break; } - dd->ipath_htwidth = width; + dd->ipath_lbus_width = width; if (linkwidth != 0x11) { ipath_dev_err(dd, "Not configured for 16 bit HT " @@ -952,8 +960,13 @@ static void slave_or_pri_blk(struct ipath_devdata *dd, struct pci_dev *pdev, speed = 200; break; } - dd->ipath_htspeed = speed; + dd->ipath_lbus_speed = speed; } + + snprintf(dd->ipath_lbus_info, sizeof(dd->ipath_lbus_info), + "HyperTransport,%uMHz,x%u\n", + dd->ipath_lbus_speed, + dd->ipath_lbus_width); } static int ipath_ht_intconfig(struct ipath_devdata *dd) @@ -1653,22 +1666,6 @@ static int ipath_ht_early_init(struct ipath_devdata *dd) } -static int ipath_ht_txe_recover(struct ipath_devdata *dd) -{ - int cnt = ++ipath_stats.sps_txeparity; - if (cnt >= IPATH_MAX_PARITY_ATTEMPTS) { - if (cnt == IPATH_MAX_PARITY_ATTEMPTS) - ipath_dev_err(dd, - "Too many attempts to recover from " - "TXE parity, giving up\n"); - return 0; - } - dev_info(&dd->pcidev->dev, - "Recovering from TXE PIO parity error\n"); - return 1; -} - - /** * ipath_init_ht_get_base_info - set chip-specific flags for user code * @dd: the infinipath device diff --git a/drivers/infiniband/hw/ipath/ipath_iba6120.c b/drivers/infiniband/hw/ipath/ipath_iba6120.c index c7a2f50824c..421cc2af891 100644 --- a/drivers/infiniband/hw/ipath/ipath_iba6120.c +++ b/drivers/infiniband/hw/ipath/ipath_iba6120.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -38,7 +38,7 @@ #include <linux/interrupt.h> #include <linux/pci.h> #include <linux/delay.h> - +#include <rdma/ib_verbs.h> #include "ipath_kernel.h" #include "ipath_registers.h" @@ -311,9 +311,14 @@ static const struct ipath_cregs ipath_pe_cregs = { .cr_ibsymbolerrcnt = IPATH_CREG_OFFSET(IBSymbolErrCnt) }; +/* kr_control bits */ +#define INFINIPATH_C_RESET 1U + /* kr_intstatus, kr_intclear, kr_intmask bits */ #define INFINIPATH_I_RCVURG_MASK ((1U<<5)-1) +#define INFINIPATH_I_RCVURG_SHIFT 0 #define INFINIPATH_I_RCVAVAIL_MASK ((1U<<5)-1) +#define INFINIPATH_I_RCVAVAIL_SHIFT 12 /* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */ #define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK 0x000000000000003fULL @@ -338,6 +343,9 @@ static const struct ipath_cregs ipath_pe_cregs = { #define INFINIPATH_EXTS_MEMBIST_ENDTEST 0x0000000000004000 #define INFINIPATH_EXTS_MEMBIST_FOUND 0x0000000000008000 +/* kr_xgxsconfig bits */ +#define INFINIPATH_XGXS_RESET 0x5ULL + #define _IPATH_GPIO_SDA_NUM 1 #define _IPATH_GPIO_SCL_NUM 0 @@ -346,6 +354,16 @@ static const struct ipath_cregs ipath_pe_cregs = { #define IPATH_GPIO_SCL (1ULL << \ (_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT)) +#define INFINIPATH_RT_BUFSIZE_MASK 0xe0000000ULL +#define INFINIPATH_RT_BUFSIZE_SHIFTVAL(tid) \ + ((((tid) & INFINIPATH_RT_BUFSIZE_MASK) >> 29) + 11 - 1) +#define INFINIPATH_RT_BUFSIZE(tid) (1 << INFINIPATH_RT_BUFSIZE_SHIFTVAL(tid)) +#define INFINIPATH_RT_IS_VALID(tid) \ + (((tid) & INFINIPATH_RT_BUFSIZE_MASK) && \ + ((((tid) & INFINIPATH_RT_BUFSIZE_MASK) != INFINIPATH_RT_BUFSIZE_MASK))) +#define INFINIPATH_RT_ADDR_MASK 0x1FFFFFFFULL /* 29 bits valid */ +#define INFINIPATH_RT_ADDR_SHIFT 10 + #define INFINIPATH_R_INTRAVAIL_SHIFT 16 #define INFINIPATH_R_TAILUPD_SHIFT 31 @@ -372,6 +390,8 @@ static const struct ipath_hwerror_msgs ipath_6120_hwerror_msgs[] = { #define TXE_PIO_PARITY ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | \ INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) \ << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) +#define RXE_EAGER_PARITY (INFINIPATH_HWE_RXEMEMPARITYERR_EAGERTID \ + << INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) static void ipath_pe_put_tid_2(struct ipath_devdata *, u64 __iomem *, u32, unsigned long); @@ -450,10 +470,8 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg, * make sure we get this much out, unless told to be quiet, * or it's occurred within the last 5 seconds */ - if ((hwerrs & ~(dd->ipath_lasthwerror | - ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | - INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) - << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) || + if ((hwerrs & ~(dd->ipath_lasthwerror | TXE_PIO_PARITY | + RXE_EAGER_PARITY)) || (ipath_debug & __IPATH_VERBDBG)) dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx " "(cleared)\n", (unsigned long long) hwerrs); @@ -465,7 +483,7 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg, (hwerrs & ~dd->ipath_hwe_bitsextant)); ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control); - if (ctrl & INFINIPATH_C_FREEZEMODE) { + if ((ctrl & INFINIPATH_C_FREEZEMODE) && !ipath_diag_inuse) { /* * parity errors in send memory are recoverable, * just cancel the send (if indicated in * sendbuffererror), @@ -540,12 +558,40 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg, dd->ipath_hwerrmask); } - if (*msg) + if (hwerrs) { + /* + * if any set that we aren't ignoring; only + * make the complaint once, in case it's stuck + * or recurring, and we get here multiple + * times. + */ ipath_dev_err(dd, "%s hardware error\n", msg); - if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) { + if (dd->ipath_flags & IPATH_INITTED) { + ipath_set_linkstate(dd, IPATH_IB_LINKDOWN); + ipath_setup_pe_setextled(dd, + INFINIPATH_IBCS_L_STATE_DOWN, + INFINIPATH_IBCS_LT_STATE_DISABLED); + ipath_dev_err(dd, "Fatal Hardware Error (freeze " + "mode), no longer usable, SN %.16s\n", + dd->ipath_serial); + isfatal = 1; + } + *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY; + /* mark as having had error */ + *dd->ipath_statusp |= IPATH_STATUS_HWERROR; /* - * for /sys status file ; if no trailing } is copied, we'll - * know it was truncated. + * mark as not usable, at a minimum until driver + * is reloaded, probably until reboot, since no + * other reset is possible. + */ + dd->ipath_flags &= ~IPATH_INITTED; + } else + *msg = 0; /* recovered from all of them */ + + if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg && msg) { + /* + * for /sys status file ; if no trailing brace is copied, + * we'll know it was truncated. */ snprintf(dd->ipath_freezemsg, dd->ipath_freezelen, "{%s}", msg); @@ -610,7 +656,6 @@ static int ipath_pe_boardname(struct ipath_devdata *dd, char *name, dd->ipath_f_put_tid = ipath_pe_put_tid_2; } - /* * set here, not in ipath_init_*_funcs because we have to do * it after we can read chip registers. @@ -838,7 +883,7 @@ static void ipath_setup_pe_setextled(struct ipath_devdata *dd, u64 lst, extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON | INFINIPATH_EXTC_LED2PRIPORT_ON); - if (ltst & INFINIPATH_IBCS_LT_STATE_LINKUP) + if (ltst == INFINIPATH_IBCS_LT_STATE_LINKUP) extctl |= INFINIPATH_EXTC_LED2PRIPORT_ON; if (lst == INFINIPATH_IBCS_L_STATE_ACTIVE) extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON; @@ -863,6 +908,62 @@ static void ipath_setup_pe_cleanup(struct ipath_devdata *dd) pci_disable_msi(dd->pcidev); } +static void ipath_6120_pcie_params(struct ipath_devdata *dd) +{ + u16 linkstat, speed; + int pos; + + pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP); + if (!pos) { + ipath_dev_err(dd, "Can't find PCI Express capability!\n"); + goto bail; + } + + pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA, + &linkstat); + /* + * speed is bits 0-4, linkwidth is bits 4-8 + * no defines for them in headers + */ + speed = linkstat & 0xf; + linkstat >>= 4; + linkstat &= 0x1f; + dd->ipath_lbus_width = linkstat; + + switch (speed) { + case 1: + dd->ipath_lbus_speed = 2500; /* Gen1, 2.5GHz */ + break; + case 2: + dd->ipath_lbus_speed = 5000; /* Gen1, 5GHz */ + break; + default: /* not defined, assume gen1 */ + dd->ipath_lbus_speed = 2500; + break; + } + + if (linkstat < 8) + ipath_dev_err(dd, + "PCIe width %u (x8 HCA), performance reduced\n", + linkstat); + else + ipath_cdbg(VERBOSE, "PCIe speed %u width %u (x8 HCA)\n", + dd->ipath_lbus_speed, linkstat); + + if (speed != 1) + ipath_dev_err(dd, + "PCIe linkspeed %u is incorrect; " + "should be 1 (2500)!\n", speed); +bail: + /* fill in string, even on errors */ + snprintf(dd->ipath_lbus_info, sizeof(dd->ipath_lbus_info), + "PCIe,%uMHz,x%u\n", + dd->ipath_lbus_speed, + dd->ipath_lbus_width); + + return; +} + /** * ipath_setup_pe_config - setup PCIe config related stuff * @dd: the infinipath device @@ -920,19 +1021,8 @@ static int ipath_setup_pe_config(struct ipath_devdata *dd, } else ipath_dev_err(dd, "Can't find MSI capability, " "can't save MSI settings for reset\n"); - if ((pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP))) { - u16 linkstat; - pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA, - &linkstat); - linkstat >>= 4; - linkstat &= 0x1f; - if (linkstat != 8) - ipath_dev_err(dd, "PCIe width %u, " - "performance reduced\n", linkstat); - } - else - ipath_dev_err(dd, "Can't find PCI Express " - "capability!\n"); + + ipath_6120_pcie_params(dd); dd->ipath_link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X; dd->ipath_link_speed_supported = IPATH_IB_SDR; @@ -1065,10 +1155,7 @@ static void ipath_init_pe_variables(struct ipath_devdata *dd) INFINIPATH_HWE_RXEMEMPARITYERR_MASK << INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT; - dd->ipath_eep_st_masks[2].errs_to_log = - INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET; - - + dd->ipath_eep_st_masks[2].errs_to_log = INFINIPATH_E_RESET; dd->delay_mult = 2; /* SDR, 4X, can't change */ } @@ -1142,6 +1229,9 @@ static int ipath_setup_pe_reset(struct ipath_devdata *dd) u64 val; int i; int ret; + u16 cmdval; + + pci_read_config_word(dd->pcidev, PCI_COMMAND, &cmdval); /* Use ERROR so it shows up in logs, etc. */ ipath_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->ipath_unit); @@ -1169,10 +1259,14 @@ static int ipath_setup_pe_reset(struct ipath_devdata *dd) ipath_dev_err(dd, "rewrite of BAR1 failed: %d\n", r); /* now re-enable memory access */ + pci_write_config_word(dd->pcidev, PCI_COMMAND, cmdval); if ((r = pci_enable_device(dd->pcidev))) ipath_dev_err(dd, "pci_enable_device failed after " "reset: %d\n", r); - /* whether it worked or not, mark as present, again */ + /* + * whether it fully enabled or not, mark as present, + * again (but not INITTED) + */ dd->ipath_flags |= IPATH_PRESENT; val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision); if (val == dd->ipath_revision) { @@ -1190,6 +1284,8 @@ static int ipath_setup_pe_reset(struct ipath_devdata *dd) ret = 0; /* failed */ bail: + if (ret) + ipath_6120_pcie_params(dd); return ret; } @@ -1209,16 +1305,21 @@ static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr, { u32 __iomem *tidp32 = (u32 __iomem *)tidptr; unsigned long flags = 0; /* keep gcc quiet */ + int tidx; + spinlock_t *tidlockp; + + if (!dd->ipath_kregbase) + return; if (pa != dd->ipath_tidinvalid) { if (pa & ((1U << 11) - 1)) { dev_info(&dd->pcidev->dev, "BUG: physaddr %lx " - "not 4KB aligned!\n", pa); + "not 2KB aligned!\n", pa); return; } pa >>= 11; /* paranoia check */ - if (pa & (7<<29)) + if (pa & ~INFINIPATH_RT_ADDR_MASK) ipath_dev_err(dd, "BUG: Physical page address 0x%lx " "has bits set in 31-29\n", pa); @@ -1238,14 +1339,22 @@ static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr, * call can be done from interrupt level for the port 0 eager TIDs, * so we have to use irqsave locks. */ - spin_lock_irqsave(&dd->ipath_tid_lock, flags); + /* + * Assumes tidptr always > ipath_egrtidbase + * if type == RCVHQ_RCV_TYPE_EAGER. + */ + tidx = tidptr - dd->ipath_egrtidbase; + + tidlockp = (type == RCVHQ_RCV_TYPE_EAGER && tidx < dd->ipath_rcvegrcnt) + ? &dd->ipath_kernel_tid_lock : &dd->ipath_user_tid_lock; + spin_lock_irqsave(tidlockp, flags); ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xfeeddeaf); - if (dd->ipath_kregbase) - writel(pa, tidp32); + writel(pa, tidp32); ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xdeadbeef); mmiowb(); - spin_unlock_irqrestore(&dd->ipath_tid_lock, flags); + spin_unlock_irqrestore(tidlockp, flags); } + /** * ipath_pe_put_tid_2 - write a TID in chip, Revision 2 or higher * @dd: the infinipath device @@ -1261,6 +1370,10 @@ static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr, u32 type, unsigned long pa) { u32 __iomem *tidp32 = (u32 __iomem *)tidptr; + u32 tidx; + + if (!dd->ipath_kregbase) + return; if (pa != dd->ipath_tidinvalid) { if (pa & ((1U << 11) - 1)) { @@ -1270,7 +1383,7 @@ static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr, } pa >>= 11; /* paranoia check */ - if (pa & (7<<29)) + if (pa & ~INFINIPATH_RT_ADDR_MASK) ipath_dev_err(dd, "BUG: Physical page address 0x%lx " "has bits set in 31-29\n", pa); @@ -1280,8 +1393,8 @@ static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr, else /* for now, always full 4KB page */ pa |= 2 << 29; } - if (dd->ipath_kregbase) - writel(pa, tidp32); + tidx = tidptr - dd->ipath_egrtidbase; + writel(pa, tidp32); mmiowb(); } @@ -1379,17 +1492,13 @@ static int ipath_pe_early_init(struct ipath_devdata *dd) dd->ipath_egrtidbase = (u64 __iomem *) ((char __iomem *) dd->ipath_kregbase + dd->ipath_rcvegrbase); - /* - * To truly support a 4KB MTU (for usermode), we need to - * bump this to a larger value. For now, we use them for - * the kernel only. - */ - dd->ipath_rcvegrbufsize = 2048; + dd->ipath_rcvegrbufsize = ipath_mtu4096 ? 4096 : 2048; /* * the min() check here is currently a nop, but it may not always * be, depending on just how we do ipath_rcvegrbufsize */ - dd->ipath_ibmaxlen = min(dd->ipath_piosize2k, + dd->ipath_ibmaxlen = min(ipath_mtu4096 ? dd->ipath_piosize4k : + dd->ipath_piosize2k, dd->ipath_rcvegrbufsize + (dd->ipath_rcvhdrentsize << 2)); dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen; diff --git a/drivers/infiniband/hw/ipath/ipath_iba7220.c b/drivers/infiniband/hw/ipath/ipath_iba7220.c new file mode 100644 index 00000000000..1b2de2cfb69 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_iba7220.c @@ -0,0 +1,2571 @@ +/* + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. + * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +/* + * This file contains all of the code that is specific to the + * InfiniPath 7220 chip (except that specific to the SerDes) + */ + +#include <linux/interrupt.h> +#include <linux/pci.h> +#include <linux/delay.h> +#include <linux/io.h> +#include <rdma/ib_verbs.h> + +#include "ipath_kernel.h" +#include "ipath_registers.h" +#include "ipath_7220.h" + +static void ipath_setup_7220_setextled(struct ipath_devdata *, u64, u64); + +static unsigned ipath_compat_ddr_negotiate = 1; + +module_param_named(compat_ddr_negotiate, ipath_compat_ddr_negotiate, uint, + S_IWUSR | S_IRUGO); +MODULE_PARM_DESC(compat_ddr_negotiate, + "Attempt pre-IBTA 1.2 DDR speed negotiation"); + +static unsigned ipath_sdma_fetch_arb = 1; +module_param_named(fetch_arb, ipath_sdma_fetch_arb, uint, S_IRUGO); +MODULE_PARM_DESC(fetch_arb, "IBA7220: change SDMA descriptor arbitration"); + +/* + * This file contains almost all the chip-specific register information and + * access functions for the QLogic InfiniPath 7220 PCI-Express chip, with the + * exception of SerDes support, which in in ipath_sd7220.c. + * + * This lists the InfiniPath registers, in the actual chip layout. + * This structure should never be directly accessed. + */ +struct _infinipath_do_not_use_kernel_regs { + unsigned long long Revision; + unsigned long long Control; + unsigned long long PageAlign; + unsigned long long PortCnt; + unsigned long long DebugPortSelect; + unsigned long long DebugSigsIntSel; /* was Reserved0;*/ + unsigned long long SendRegBase; + unsigned long long UserRegBase; + unsigned long long CounterRegBase; + unsigned long long Scratch; + unsigned long long EEPROMAddrCmd; /* was Reserved1; */ + unsigned long long EEPROMData; /* was Reserved2; */ + unsigned long long IntBlocked; + unsigned long long IntMask; + unsigned long long IntStatus; + unsigned long long IntClear; + unsigned long long ErrorMask; + unsigned long long ErrorStatus; + unsigned long long ErrorClear; + unsigned long long HwErrMask; + unsigned long long HwErrStatus; + unsigned long long HwErrClear; + unsigned long long HwDiagCtrl; + unsigned long long MDIO; + unsigned long long IBCStatus; + unsigned long long IBCCtrl; + unsigned long long ExtStatus; + unsigned long long ExtCtrl; + unsigned long long GPIOOut; + unsigned long long GPIOMask; + unsigned long long GPIOStatus; + unsigned long long GPIOClear; + unsigned long long RcvCtrl; + unsigned long long RcvBTHQP; + unsigned long long RcvHdrSize; + unsigned long long RcvHdrCnt; + unsigned long long RcvHdrEntSize; + unsigned long long RcvTIDBase; + unsigned long long RcvTIDCnt; + unsigned long long RcvEgrBase; + unsigned long long RcvEgrCnt; + unsigned long long RcvBufBase; + unsigned long long RcvBufSize; + unsigned long long RxIntMemBase; + unsigned long long RxIntMemSize; + unsigned long long RcvPartitionKey; + unsigned long long RcvQPMulticastPort; + unsigned long long RcvPktLEDCnt; + unsigned long long IBCDDRCtrl; + unsigned long long HRTBT_GUID; + unsigned long long IB_SDTEST_IF_TX; + unsigned long long IB_SDTEST_IF_RX; + unsigned long long IBCDDRCtrl2; + unsigned long long IBCDDRStatus; + unsigned long long JIntReload; + unsigned long long IBNCModeCtrl; + unsigned long long SendCtrl; + unsigned long long SendBufBase; + unsigned long long SendBufSize; + unsigned long long SendBufCnt; + unsigned long long SendAvailAddr; + unsigned long long TxIntMemBase; + unsigned long long TxIntMemSize; + unsigned long long SendDmaBase; + unsigned long long SendDmaLenGen; + unsigned long long SendDmaTail; + unsigned long long SendDmaHead; + unsigned long long SendDmaHeadAddr; + unsigned long long SendDmaBufMask0; + unsigned long long SendDmaBufMask1; + unsigned long long SendDmaBufMask2; + unsigned long long SendDmaStatus; + unsigned long long SendBufferError; + unsigned long long SendBufferErrorCONT1; + unsigned long long SendBufErr2; /* was Reserved6SBE[0/6] */ + unsigned long long Reserved6L[2]; + unsigned long long AvailUpdCount; + unsigned long long RcvHdrAddr0; + unsigned long long RcvHdrAddrs[16]; /* Why enumerate? */ + unsigned long long Reserved7hdtl; /* Align next to 300 */ + unsigned long long RcvHdrTailAddr0; /* 300, like others */ + unsigned long long RcvHdrTailAddrs[16]; + unsigned long long Reserved9SW[7]; /* was [8]; we have 17 ports */ + unsigned long long IbsdEpbAccCtl; /* IB Serdes EPB access control */ + unsigned long long IbsdEpbTransReg; /* IB Serdes EPB Transaction */ + unsigned long long Reserved10sds; /* was SerdesStatus on */ + unsigned long long XGXSConfig; + unsigned long long IBSerDesCtrl; /* Was IBPLLCfg on Monty */ + unsigned long long EEPCtlStat; /* for "boot" EEPROM/FLASH */ + unsigned long long EEPAddrCmd; + unsigned long long EEPData; + unsigned long long PcieEpbAccCtl; + unsigned long long PcieEpbTransCtl; + unsigned long long EfuseCtl; /* E-Fuse control */ + unsigned long long EfuseData[4]; + unsigned long long ProcMon; + /* this chip moves following two from previous 200, 208 */ + unsigned long long PCIeRBufTestReg0; + unsigned long long PCIeRBufTestReg1; + /* added for this chip */ + unsigned long long PCIeRBufTestReg2; + unsigned long long PCIeRBufTestReg3; + /* added for this chip, debug only */ + unsigned long long SPC_JTAG_ACCESS_REG; + unsigned long long LAControlReg; + unsigned long long GPIODebugSelReg; + unsigned long long DebugPortValueReg; + /* added for this chip, DMA */ + unsigned long long SendDmaBufUsed[3]; + unsigned long long SendDmaReqTagUsed; + /* + * added for this chip, EFUSE: note that these program 64-bit + * words 2 and 3 */ + unsigned long long efuse_pgm_data[2]; + unsigned long long Reserved11LAalign[10]; /* Skip 4B0..4F8 */ + /* we have 30 regs for DDS and RXEQ in IB SERDES */ + unsigned long long SerDesDDSRXEQ[30]; + unsigned long long Reserved12LAalign[2]; /* Skip 5F0, 5F8 */ + /* added for LA debug support */ + unsigned long long LAMemory[32]; +}; + +struct _infinipath_do_not_use_counters { + __u64 LBIntCnt; + __u64 LBFlowStallCnt; + __u64 TxSDmaDescCnt; /* was Reserved1 */ + __u64 TxUnsupVLErrCnt; + __u64 TxDataPktCnt; + __u64 TxFlowPktCnt; + __u64 TxDwordCnt; + __u64 TxLenErrCnt; + __u64 TxMaxMinLenErrCnt; + __u64 TxUnderrunCnt; + __u64 TxFlowStallCnt; + __u64 TxDroppedPktCnt; + __u64 RxDroppedPktCnt; + __u64 RxDataPktCnt; + __u64 RxFlowPktCnt; + __u64 RxDwordCnt; + __u64 RxLenErrCnt; + __u64 RxMaxMinLenErrCnt; + __u64 RxICRCErrCnt; + __u64 RxVCRCErrCnt; + __u64 RxFlowCtrlErrCnt; + __u64 RxBadFormatCnt; + __u64 RxLinkProblemCnt; + __u64 RxEBPCnt; + __u64 RxLPCRCErrCnt; + __u64 RxBufOvflCnt; + __u64 RxTIDFullErrCnt; + __u64 RxTIDValidErrCnt; + __u64 RxPKeyMismatchCnt; + __u64 RxP0HdrEgrOvflCnt; + __u64 RxP1HdrEgrOvflCnt; + __u64 RxP2HdrEgrOvflCnt; + __u64 RxP3HdrEgrOvflCnt; + __u64 RxP4HdrEgrOvflCnt; + __u64 RxP5HdrEgrOvflCnt; + __u64 RxP6HdrEgrOvflCnt; + __u64 RxP7HdrEgrOvflCnt; + __u64 RxP8HdrEgrOvflCnt; + __u64 RxP9HdrEgrOvflCnt; /* was Reserved6 */ + __u64 RxP10HdrEgrOvflCnt; /* was Reserved7 */ + __u64 RxP11HdrEgrOvflCnt; /* new for IBA7220 */ + __u64 RxP12HdrEgrOvflCnt; /* new for IBA7220 */ + __u64 RxP13HdrEgrOvflCnt; /* new for IBA7220 */ + __u64 RxP14HdrEgrOvflCnt; /* new for IBA7220 */ + __u64 RxP15HdrEgrOvflCnt; /* new for IBA7220 */ + __u64 RxP16HdrEgrOvflCnt; /* new for IBA7220 */ + __u64 IBStatusChangeCnt; + __u64 IBLinkErrRecoveryCnt; + __u64 IBLinkDownedCnt; + __u64 IBSymbolErrCnt; + /* The following are new for IBA7220 */ + __u64 RxVL15DroppedPktCnt; + __u64 RxOtherLocalPhyErrCnt; + __u64 PcieRetryBufDiagQwordCnt; + __u64 ExcessBufferOvflCnt; + __u64 LocalLinkIntegrityErrCnt; + __u64 RxVlErrCnt; + __u64 RxDlidFltrCnt; + __u64 Reserved8[7]; + __u64 PSStat; + __u64 PSStart; + __u64 PSInterval; + __u64 PSRcvDataCount; + __u64 PSRcvPktsCount; + __u64 PSXmitDataCount; + __u64 PSXmitPktsCount; + __u64 PSXmitWaitCount; +}; + +#define IPATH_KREG_OFFSET(field) (offsetof( \ + struct _infinipath_do_not_use_kernel_regs, field) / sizeof(u64)) +#define IPATH_CREG_OFFSET(field) (offsetof( \ + struct _infinipath_do_not_use_counters, field) / sizeof(u64)) + +static const struct ipath_kregs ipath_7220_kregs = { + .kr_control = IPATH_KREG_OFFSET(Control), + .kr_counterregbase = IPATH_KREG_OFFSET(CounterRegBase), + .kr_debugportselect = IPATH_KREG_OFFSET(DebugPortSelect), + .kr_errorclear = IPATH_KREG_OFFSET(ErrorClear), + .kr_errormask = IPATH_KREG_OFFSET(ErrorMask), + .kr_errorstatus = IPATH_KREG_OFFSET(ErrorStatus), + .kr_extctrl = IPATH_KREG_OFFSET(ExtCtrl), + .kr_extstatus = IPATH_KREG_OFFSET(ExtStatus), + .kr_gpio_clear = IPATH_KREG_OFFSET(GPIOClear), + .kr_gpio_mask = IPATH_KREG_OFFSET(GPIOMask), + .kr_gpio_out = IPATH_KREG_OFFSET(GPIOOut), + .kr_gpio_status = IPATH_KREG_OFFSET(GPIOStatus), + .kr_hwdiagctrl = IPATH_KREG_OFFSET(HwDiagCtrl), + .kr_hwerrclear = IPATH_KREG_OFFSET(HwErrClear), + .kr_hwerrmask = IPATH_KREG_OFFSET(HwErrMask), + .kr_hwerrstatus = IPATH_KREG_OFFSET(HwErrStatus), + .kr_ibcctrl = IPATH_KREG_OFFSET(IBCCtrl), + .kr_ibcstatus = IPATH_KREG_OFFSET(IBCStatus), + .kr_intblocked = IPATH_KREG_OFFSET(IntBlocked), + .kr_intclear = IPATH_KREG_OFFSET(IntClear), + .kr_intmask = IPATH_KREG_OFFSET(IntMask), + .kr_intstatus = IPATH_KREG_OFFSET(IntStatus), + .kr_mdio = IPATH_KREG_OFFSET(MDIO), + .kr_pagealign = IPATH_KREG_OFFSET(PageAlign), + .kr_partitionkey = IPATH_KREG_OFFSET(RcvPartitionKey), + .kr_portcnt = IPATH_KREG_OFFSET(PortCnt), + .kr_rcvbthqp = IPATH_KREG_OFFSET(RcvBTHQP), + .kr_rcvbufbase = IPATH_KREG_OFFSET(RcvBufBase), + .kr_rcvbufsize = IPATH_KREG_OFFSET(RcvBufSize), + .kr_rcvctrl = IPATH_KREG_OFFSET(RcvCtrl), + .kr_rcvegrbase = IPATH_KREG_OFFSET(RcvEgrBase), + .kr_rcvegrcnt = IPATH_KREG_OFFSET(RcvEgrCnt), + .kr_rcvhdrcnt = IPATH_KREG_OFFSET(RcvHdrCnt), + .kr_rcvhdrentsize = IPATH_KREG_OFFSET(RcvHdrEntSize), + .kr_rcvhdrsize = IPATH_KREG_OFFSET(RcvHdrSize), + .kr_rcvintmembase = IPATH_KREG_OFFSET(RxIntMemBase), + .kr_rcvintmemsize = IPATH_KREG_OFFSET(RxIntMemSize), + .kr_rcvtidbase = IPATH_KREG_OFFSET(RcvTIDBase), + .kr_rcvtidcnt = IPATH_KREG_OFFSET(RcvTIDCnt), + .kr_revision = IPATH_KREG_OFFSET(Revision), + .kr_scratch = IPATH_KREG_OFFSET(Scratch), + .kr_sendbuffererror = IPATH_KREG_OFFSET(SendBufferError), + .kr_sendctrl = IPATH_KREG_OFFSET(SendCtrl), + .kr_sendpioavailaddr = IPATH_KREG_OFFSET(SendAvailAddr), + .kr_sendpiobufbase = IPATH_KREG_OFFSET(SendBufBase), + .kr_sendpiobufcnt = IPATH_KREG_OFFSET(SendBufCnt), + .kr_sendpiosize = IPATH_KREG_OFFSET(SendBufSize), + .kr_sendregbase = IPATH_KREG_OFFSET(SendRegBase), + .kr_txintmembase = IPATH_KREG_OFFSET(TxIntMemBase), + .kr_txintmemsize = IPATH_KREG_OFFSET(TxIntMemSize), + .kr_userregbase = IPATH_KREG_OFFSET(UserRegBase), + + .kr_xgxsconfig = IPATH_KREG_OFFSET(XGXSConfig), + + /* send dma related regs */ + .kr_senddmabase = IPATH_KREG_OFFSET(SendDmaBase), + .kr_senddmalengen = IPATH_KREG_OFFSET(SendDmaLenGen), + .kr_senddmatail = IPATH_KREG_OFFSET(SendDmaTail), + .kr_senddmahead = IPATH_KREG_OFFSET(SendDmaHead), + .kr_senddmaheadaddr = IPATH_KREG_OFFSET(SendDmaHeadAddr), + .kr_senddmabufmask0 = IPATH_KREG_OFFSET(SendDmaBufMask0), + .kr_senddmabufmask1 = IPATH_KREG_OFFSET(SendDmaBufMask1), + .kr_senddmabufmask2 = IPATH_KREG_OFFSET(SendDmaBufMask2), + .kr_senddmastatus = IPATH_KREG_OFFSET(SendDmaStatus), + + /* SerDes related regs */ + .kr_ibserdesctrl = IPATH_KREG_OFFSET(IBSerDesCtrl), + .kr_ib_epbacc = IPATH_KREG_OFFSET(IbsdEpbAccCtl), + .kr_ib_epbtrans = IPATH_KREG_OFFSET(IbsdEpbTransReg), + .kr_pcie_epbacc = IPATH_KREG_OFFSET(PcieEpbAccCtl), + .kr_pcie_epbtrans = IPATH_KREG_OFFSET(PcieEpbTransCtl), + .kr_ib_ddsrxeq = IPATH_KREG_OFFSET(SerDesDDSRXEQ), + + /* + * These should not be used directly via ipath_read_kreg64(), + * use them with ipath_read_kreg64_port() + */ + .kr_rcvhdraddr = IPATH_KREG_OFFSET(RcvHdrAddr0), + .kr_rcvhdrtailaddr = IPATH_KREG_OFFSET(RcvHdrTailAddr0), + + /* + * The rcvpktled register controls one of the debug port signals, so + * a packet activity LED can be connected to it. + */ + .kr_rcvpktledcnt = IPATH_KREG_OFFSET(RcvPktLEDCnt), + .kr_pcierbuftestreg0 = IPATH_KREG_OFFSET(PCIeRBufTestReg0), + .kr_pcierbuftestreg1 = IPATH_KREG_OFFSET(PCIeRBufTestReg1), + + .kr_hrtbt_guid = IPATH_KREG_OFFSET(HRTBT_GUID), + .kr_ibcddrctrl = IPATH_KREG_OFFSET(IBCDDRCtrl), + .kr_ibcddrstatus = IPATH_KREG_OFFSET(IBCDDRStatus), + .kr_jintreload = IPATH_KREG_OFFSET(JIntReload) +}; + +static const struct ipath_cregs ipath_7220_cregs = { + .cr_badformatcnt = IPATH_CREG_OFFSET(RxBadFormatCnt), + .cr_erricrccnt = IPATH_CREG_OFFSET(RxICRCErrCnt), + .cr_errlinkcnt = IPATH_CREG_OFFSET(RxLinkProblemCnt), + .cr_errlpcrccnt = IPATH_CREG_OFFSET(RxLPCRCErrCnt), + .cr_errpkey = IPATH_CREG_OFFSET(RxPKeyMismatchCnt), + .cr_errrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowCtrlErrCnt), + .cr_err_rlencnt = IPATH_CREG_OFFSET(RxLenErrCnt), + .cr_errslencnt = IPATH_CREG_OFFSET(TxLenErrCnt), + .cr_errtidfull = IPATH_CREG_OFFSET(RxTIDFullErrCnt), + .cr_errtidvalid = IPATH_CREG_OFFSET(RxTIDValidErrCnt), + .cr_errvcrccnt = IPATH_CREG_OFFSET(RxVCRCErrCnt), + .cr_ibstatuschange = IPATH_CREG_OFFSET(IBStatusChangeCnt), + .cr_intcnt = IPATH_CREG_OFFSET(LBIntCnt), + .cr_invalidrlencnt = IPATH_CREG_OFFSET(RxMaxMinLenErrCnt), + .cr_invalidslencnt = IPATH_CREG_OFFSET(TxMaxMinLenErrCnt), + .cr_lbflowstallcnt = IPATH_CREG_OFFSET(LBFlowStallCnt), + .cr_pktrcvcnt = IPATH_CREG_OFFSET(RxDataPktCnt), + .cr_pktrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowPktCnt), + .cr_pktsendcnt = IPATH_CREG_OFFSET(TxDataPktCnt), + .cr_pktsendflowcnt = IPATH_CREG_OFFSET(TxFlowPktCnt), + .cr_portovflcnt = IPATH_CREG_OFFSET(RxP0HdrEgrOvflCnt), + .cr_rcvebpcnt = IPATH_CREG_OFFSET(RxEBPCnt), + .cr_rcvovflcnt = IPATH_CREG_OFFSET(RxBufOvflCnt), + .cr_senddropped = IPATH_CREG_OFFSET(TxDroppedPktCnt), + .cr_sendstallcnt = IPATH_CREG_OFFSET(TxFlowStallCnt), + .cr_sendunderruncnt = IPATH_CREG_OFFSET(TxUnderrunCnt), + .cr_wordrcvcnt = IPATH_CREG_OFFSET(RxDwordCnt), + .cr_wordsendcnt = IPATH_CREG_OFFSET(TxDwordCnt), + .cr_unsupvlcnt = IPATH_CREG_OFFSET(TxUnsupVLErrCnt), + .cr_rxdroppktcnt = IPATH_CREG_OFFSET(RxDroppedPktCnt), + .cr_iblinkerrrecovcnt = IPATH_CREG_OFFSET(IBLinkErrRecoveryCnt), + .cr_iblinkdowncnt = IPATH_CREG_OFFSET(IBLinkDownedCnt), + .cr_ibsymbolerrcnt = IPATH_CREG_OFFSET(IBSymbolErrCnt), + .cr_vl15droppedpktcnt = IPATH_CREG_OFFSET(RxVL15DroppedPktCnt), + .cr_rxotherlocalphyerrcnt = + IPATH_CREG_OFFSET(RxOtherLocalPhyErrCnt), + .cr_excessbufferovflcnt = IPATH_CREG_OFFSET(ExcessBufferOvflCnt), + .cr_locallinkintegrityerrcnt = + IPATH_CREG_OFFSET(LocalLinkIntegrityErrCnt), + .cr_rxvlerrcnt = IPATH_CREG_OFFSET(RxVlErrCnt), + .cr_rxdlidfltrcnt = IPATH_CREG_OFFSET(RxDlidFltrCnt), + .cr_psstat = IPATH_CREG_OFFSET(PSStat), + .cr_psstart = IPATH_CREG_OFFSET(PSStart), + .cr_psinterval = IPATH_CREG_OFFSET(PSInterval), + .cr_psrcvdatacount = IPATH_CREG_OFFSET(PSRcvDataCount), + .cr_psrcvpktscount = IPATH_CREG_OFFSET(PSRcvPktsCount), + .cr_psxmitdatacount = IPATH_CREG_OFFSET(PSXmitDataCount), + .cr_psxmitpktscount = IPATH_CREG_OFFSET(PSXmitPktsCount), + .cr_psxmitwaitcount = IPATH_CREG_OFFSET(PSXmitWaitCount), +}; + +/* kr_control bits */ +#define INFINIPATH_C_RESET (1U<<7) + +/* kr_intstatus, kr_intclear, kr_intmask bits */ +#define INFINIPATH_I_RCVURG_MASK ((1ULL<<17)-1) +#define INFINIPATH_I_RCVURG_SHIFT 32 +#define INFINIPATH_I_RCVAVAIL_MASK ((1ULL<<17)-1) +#define INFINIPATH_I_RCVAVAIL_SHIFT 0 +#define INFINIPATH_I_SERDESTRIMDONE (1ULL<<27) + +/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */ +#define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK 0x00000000000000ffULL +#define INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT 0 +#define INFINIPATH_HWE_PCIEPOISONEDTLP 0x0000000010000000ULL +#define INFINIPATH_HWE_PCIECPLTIMEOUT 0x0000000020000000ULL +#define INFINIPATH_HWE_PCIEBUSPARITYXTLH 0x0000000040000000ULL +#define INFINIPATH_HWE_PCIEBUSPARITYXADM 0x0000000080000000ULL +#define INFINIPATH_HWE_PCIEBUSPARITYRADM 0x0000000100000000ULL +#define INFINIPATH_HWE_COREPLL_FBSLIP 0x0080000000000000ULL +#define INFINIPATH_HWE_COREPLL_RFSLIP 0x0100000000000000ULL +#define INFINIPATH_HWE_PCIE1PLLFAILED 0x0400000000000000ULL +#define INFINIPATH_HWE_PCIE0PLLFAILED 0x0800000000000000ULL +#define INFINIPATH_HWE_SERDESPLLFAILED 0x1000000000000000ULL +/* specific to this chip */ +#define INFINIPATH_HWE_PCIECPLDATAQUEUEERR 0x0000000000000040ULL +#define INFINIPATH_HWE_PCIECPLHDRQUEUEERR 0x0000000000000080ULL +#define INFINIPATH_HWE_SDMAMEMREADERR 0x0000000010000000ULL +#define INFINIPATH_HWE_CLK_UC_PLLNOTLOCKED 0x2000000000000000ULL +#define INFINIPATH_HWE_PCIESERDESQ0PCLKNOTDETECT 0x0100000000000000ULL +#define INFINIPATH_HWE_PCIESERDESQ1PCLKNOTDETECT 0x0200000000000000ULL +#define INFINIPATH_HWE_PCIESERDESQ2PCLKNOTDETECT 0x0400000000000000ULL +#define INFINIPATH_HWE_PCIESERDESQ3PCLKNOTDETECT 0x0800000000000000ULL +#define INFINIPATH_HWE_DDSRXEQMEMORYPARITYERR 0x0000008000000000ULL +#define INFINIPATH_HWE_IB_UC_MEMORYPARITYERR 0x0000004000000000ULL +#define INFINIPATH_HWE_PCIE_UC_OCT0MEMORYPARITYERR 0x0000001000000000ULL +#define INFINIPATH_HWE_PCIE_UC_OCT1MEMORYPARITYERR 0x0000002000000000ULL + +#define IBA7220_IBCS_LINKTRAININGSTATE_MASK 0x1F +#define IBA7220_IBCS_LINKSTATE_SHIFT 5 +#define IBA7220_IBCS_LINKSPEED_SHIFT 8 +#define IBA7220_IBCS_LINKWIDTH_SHIFT 9 + +#define IBA7220_IBCC_LINKINITCMD_MASK 0x7ULL +#define IBA7220_IBCC_LINKCMD_SHIFT 19 +#define IBA7220_IBCC_MAXPKTLEN_SHIFT 21 + +/* kr_ibcddrctrl bits */ +#define IBA7220_IBC_DLIDLMC_MASK 0xFFFFFFFFUL +#define IBA7220_IBC_DLIDLMC_SHIFT 32 +#define IBA7220_IBC_HRTBT_MASK 3 +#define IBA7220_IBC_HRTBT_SHIFT 16 +#define IBA7220_IBC_HRTBT_ENB 0x10000UL +#define IBA7220_IBC_LANE_REV_SUPPORTED (1<<8) +#define IBA7220_IBC_LREV_MASK 1 +#define IBA7220_IBC_LREV_SHIFT 8 +#define IBA7220_IBC_RXPOL_MASK 1 +#define IBA7220_IBC_RXPOL_SHIFT 7 +#define IBA7220_IBC_WIDTH_SHIFT 5 +#define IBA7220_IBC_WIDTH_MASK 0x3 +#define IBA7220_IBC_WIDTH_1X_ONLY (0<<IBA7220_IBC_WIDTH_SHIFT) +#define IBA7220_IBC_WIDTH_4X_ONLY (1<<IBA7220_IBC_WIDTH_SHIFT) +#define IBA7220_IBC_WIDTH_AUTONEG (2<<IBA7220_IBC_WIDTH_SHIFT) +#define IBA7220_IBC_SPEED_AUTONEG (1<<1) +#define IBA7220_IBC_SPEED_SDR (1<<2) +#define IBA7220_IBC_SPEED_DDR (1<<3) +#define IBA7220_IBC_SPEED_AUTONEG_MASK (0x7<<1) +#define IBA7220_IBC_IBTA_1_2_MASK (1) + +/* kr_ibcddrstatus */ +/* link latency shift is 0, don't bother defining */ +#define IBA7220_DDRSTAT_LINKLAT_MASK 0x3ffffff + +/* kr_extstatus bits */ +#define INFINIPATH_EXTS_FREQSEL 0x2 +#define INFINIPATH_EXTS_SERDESSEL 0x4 +#define INFINIPATH_EXTS_MEMBIST_ENDTEST 0x0000000000004000 +#define INFINIPATH_EXTS_MEMBIST_DISABLED 0x0000000000008000 + +/* kr_xgxsconfig bits */ +#define INFINIPATH_XGXS_RESET 0x5ULL +#define INFINIPATH_XGXS_FC_SAFE (1ULL<<63) + +/* kr_rcvpktledcnt */ +#define IBA7220_LEDBLINK_ON_SHIFT 32 /* 4ns period on after packet */ +#define IBA7220_LEDBLINK_OFF_SHIFT 0 /* 4ns period off before next on */ + +#define _IPATH_GPIO_SDA_NUM 1 +#define _IPATH_GPIO_SCL_NUM 0 + +#define IPATH_GPIO_SDA (1ULL << \ + (_IPATH_GPIO_SDA_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT)) +#define IPATH_GPIO_SCL (1ULL << \ + (_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT)) + +#define IBA7220_R_INTRAVAIL_SHIFT 17 +#define IBA7220_R_TAILUPD_SHIFT 35 +#define IBA7220_R_PORTCFG_SHIFT 36 + +#define INFINIPATH_JINT_PACKETSHIFT 16 +#define INFINIPATH_JINT_DEFAULT_IDLE_TICKS 0 +#define INFINIPATH_JINT_DEFAULT_MAX_PACKETS 0 + +#define IBA7220_HDRHEAD_PKTINT_SHIFT 32 /* interrupt cnt in upper 32 bits */ + +/* + * the size bits give us 2^N, in KB units. 0 marks as invalid, + * and 7 is reserved. We currently use only 2KB and 4KB + */ +#define IBA7220_TID_SZ_SHIFT 37 /* shift to 3bit size selector */ +#define IBA7220_TID_SZ_2K (1UL<<IBA7220_TID_SZ_SHIFT) /* 2KB */ +#define IBA7220_TID_SZ_4K (2UL<<IBA7220_TID_SZ_SHIFT) /* 4KB */ +#define IBA7220_TID_PA_SHIFT 11U /* TID addr in chip stored w/o low bits */ + +#define IPATH_AUTONEG_TRIES 5 /* sequential retries to negotiate DDR */ + +static char int_type[16] = "auto"; +module_param_string(interrupt_type, int_type, sizeof(int_type), 0444); +MODULE_PARM_DESC(int_type, " interrupt_type=auto|force_msi|force_intx\n"); + +/* packet rate matching delay; chip has support */ +static u8 rate_to_delay[2][2] = { + /* 1x, 4x */ + { 8, 2 }, /* SDR */ + { 4, 1 } /* DDR */ +}; + +/* 7220 specific hardware errors... */ +static const struct ipath_hwerror_msgs ipath_7220_hwerror_msgs[] = { + INFINIPATH_HWE_MSG(PCIEPOISONEDTLP, "PCIe Poisoned TLP"), + INFINIPATH_HWE_MSG(PCIECPLTIMEOUT, "PCIe completion timeout"), + /* + * In practice, it's unlikely wthat we'll see PCIe PLL, or bus + * parity or memory parity error failures, because most likely we + * won't be able to talk to the core of the chip. Nonetheless, we + * might see them, if they are in parts of the PCIe core that aren't + * essential. + */ + INFINIPATH_HWE_MSG(PCIE1PLLFAILED, "PCIePLL1"), + INFINIPATH_HWE_MSG(PCIE0PLLFAILED, "PCIePLL0"), + INFINIPATH_HWE_MSG(PCIEBUSPARITYXTLH, "PCIe XTLH core parity"), + INFINIPATH_HWE_MSG(PCIEBUSPARITYXADM, "PCIe ADM TX core parity"), + INFINIPATH_HWE_MSG(PCIEBUSPARITYRADM, "PCIe ADM RX core parity"), + INFINIPATH_HWE_MSG(RXDSYNCMEMPARITYERR, "Rx Dsync"), + INFINIPATH_HWE_MSG(SERDESPLLFAILED, "SerDes PLL"), + INFINIPATH_HWE_MSG(PCIECPLDATAQUEUEERR, "PCIe cpl header queue"), + INFINIPATH_HWE_MSG(PCIECPLHDRQUEUEERR, "PCIe cpl data queue"), + INFINIPATH_HWE_MSG(SDMAMEMREADERR, "Send DMA memory read"), + INFINIPATH_HWE_MSG(CLK_UC_PLLNOTLOCKED, "uC PLL clock not locked"), + INFINIPATH_HWE_MSG(PCIESERDESQ0PCLKNOTDETECT, + "PCIe serdes Q0 no clock"), + INFINIPATH_HWE_MSG(PCIESERDESQ1PCLKNOTDETECT, + "PCIe serdes Q1 no clock"), + INFINIPATH_HWE_MSG(PCIESERDESQ2PCLKNOTDETECT, + "PCIe serdes Q2 no clock"), + INFINIPATH_HWE_MSG(PCIESERDESQ3PCLKNOTDETECT, + "PCIe serdes Q3 no clock"), + INFINIPATH_HWE_MSG(DDSRXEQMEMORYPARITYERR, + "DDS RXEQ memory parity"), + INFINIPATH_HWE_MSG(IB_UC_MEMORYPARITYERR, "IB uC memory parity"), + INFINIPATH_HWE_MSG(PCIE_UC_OCT0MEMORYPARITYERR, + "PCIe uC oct0 memory parity"), + INFINIPATH_HWE_MSG(PCIE_UC_OCT1MEMORYPARITYERR, + "PCIe uC oct1 memory parity"), +}; + +static void autoneg_work(struct work_struct *); + +/* + * the offset is different for different configured port numbers, since + * port0 is fixed in size, but others can vary. Make it a function to + * make the issue more obvious. +*/ +static inline u32 port_egrtid_idx(struct ipath_devdata *dd, unsigned port) +{ + return port ? dd->ipath_p0_rcvegrcnt + + (port-1) * dd->ipath_rcvegrcnt : 0; +} + +static void ipath_7220_txe_recover(struct ipath_devdata *dd) +{ + ++ipath_stats.sps_txeparity; + + dev_info(&dd->pcidev->dev, + "Recovering from TXE PIO parity error\n"); + ipath_disarm_senderrbufs(dd, 1); +} + + +/** + * ipath_7220_handle_hwerrors - display hardware errors. + * @dd: the infinipath device + * @msg: the output buffer + * @msgl: the size of the output buffer + * + * Use same msg buffer as regular errors to avoid excessive stack + * use. Most hardware errors are catastrophic, but for right now, + * we'll print them and continue. We reuse the same message buffer as + * ipath_handle_errors() to avoid excessive stack usage. + */ +static void ipath_7220_handle_hwerrors(struct ipath_devdata *dd, char *msg, + size_t msgl) +{ + ipath_err_t hwerrs; + u32 bits, ctrl; + int isfatal = 0; + char bitsmsg[64]; + int log_idx; + + hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus); + if (!hwerrs) { + /* + * better than printing cofusing messages + * This seems to be related to clearing the crc error, or + * the pll error during init. + */ + ipath_cdbg(VERBOSE, "Called but no hardware errors set\n"); + goto bail; + } else if (hwerrs == ~0ULL) { + ipath_dev_err(dd, "Read of hardware error status failed " + "(all bits set); ignoring\n"); + goto bail; + } + ipath_stats.sps_hwerrs++; + + /* + * Always clear the error status register, except MEMBISTFAIL, + * regardless of whether we continue or stop using the chip. + * We want that set so we know it failed, even across driver reload. + * We'll still ignore it in the hwerrmask. We do this partly for + * diagnostics, but also for support. + */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear, + hwerrs&~INFINIPATH_HWE_MEMBISTFAILED); + + hwerrs &= dd->ipath_hwerrmask; + + /* We log some errors to EEPROM, check if we have any of those. */ + for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx) + if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log) + ipath_inc_eeprom_err(dd, log_idx, 1); + /* + * Make sure we get this much out, unless told to be quiet, + * or it's occurred within the last 5 seconds. + */ + if ((hwerrs & ~(dd->ipath_lasthwerror | + ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | + INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) + << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) || + (ipath_debug & __IPATH_VERBDBG)) + dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx " + "(cleared)\n", (unsigned long long) hwerrs); + dd->ipath_lasthwerror |= hwerrs; + + if (hwerrs & ~dd->ipath_hwe_bitsextant) + ipath_dev_err(dd, "hwerror interrupt with unknown errors " + "%llx set\n", (unsigned long long) + (hwerrs & ~dd->ipath_hwe_bitsextant)); + + if (hwerrs & INFINIPATH_HWE_IB_UC_MEMORYPARITYERR) + ipath_sd7220_clr_ibpar(dd); + + ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control); + if ((ctrl & INFINIPATH_C_FREEZEMODE) && !ipath_diag_inuse) { + /* + * Parity errors in send memory are recoverable, + * just cancel the send (if indicated in * sendbuffererror), + * count the occurrence, unfreeze (if no other handled + * hardware error bits are set), and continue. + */ + if (hwerrs & ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | + INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) + << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)) { + ipath_7220_txe_recover(dd); + hwerrs &= ~((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | + INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) + << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT); + if (!hwerrs) { + /* else leave in freeze mode */ + ipath_write_kreg(dd, + dd->ipath_kregs->kr_control, + dd->ipath_control); + goto bail; + } + } + if (hwerrs) { + /* + * If any set that we aren't ignoring only make the + * complaint once, in case it's stuck or recurring, + * and we get here multiple times + * Force link down, so switch knows, and + * LEDs are turned off. + */ + if (dd->ipath_flags & IPATH_INITTED) { + ipath_set_linkstate(dd, IPATH_IB_LINKDOWN); + ipath_setup_7220_setextled(dd, + INFINIPATH_IBCS_L_STATE_DOWN, + INFINIPATH_IBCS_LT_STATE_DISABLED); + ipath_dev_err(dd, "Fatal Hardware Error " + "(freeze mode), no longer" + " usable, SN %.16s\n", + dd->ipath_serial); + isfatal = 1; + } + /* + * Mark as having had an error for driver, and also + * for /sys and status word mapped to user programs. + * This marks unit as not usable, until reset. + */ + *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY; + *dd->ipath_statusp |= IPATH_STATUS_HWERROR; + dd->ipath_flags &= ~IPATH_INITTED; + } else { + ipath_dbg("Clearing freezemode on ignored hardware " + "error\n"); + ipath_clear_freeze(dd); + } + } + + *msg = '\0'; + + if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) { + strlcat(msg, "[Memory BIST test failed, " + "InfiniPath hardware unusable]", msgl); + /* ignore from now on, so disable until driver reloaded */ + *dd->ipath_statusp |= IPATH_STATUS_HWERROR; + dd->ipath_hwerrmask &= ~INFINIPATH_HWE_MEMBISTFAILED; + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, + dd->ipath_hwerrmask); + } + + ipath_format_hwerrors(hwerrs, + ipath_7220_hwerror_msgs, + ARRAY_SIZE(ipath_7220_hwerror_msgs), + msg, msgl); + + if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK + << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) { + bits = (u32) ((hwerrs >> + INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) & + INFINIPATH_HWE_PCIEMEMPARITYERR_MASK); + snprintf(bitsmsg, sizeof bitsmsg, + "[PCIe Mem Parity Errs %x] ", bits); + strlcat(msg, bitsmsg, msgl); + } + +#define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP | \ + INFINIPATH_HWE_COREPLL_RFSLIP) + + if (hwerrs & _IPATH_PLL_FAIL) { + snprintf(bitsmsg, sizeof bitsmsg, + "[PLL failed (%llx), InfiniPath hardware unusable]", + (unsigned long long) hwerrs & _IPATH_PLL_FAIL); + strlcat(msg, bitsmsg, msgl); + /* ignore from now on, so disable until driver reloaded */ + dd->ipath_hwerrmask &= ~(hwerrs & _IPATH_PLL_FAIL); + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, + dd->ipath_hwerrmask); + } + + if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED) { + /* + * If it occurs, it is left masked since the eternal + * interface is unused. + */ + dd->ipath_hwerrmask &= ~INFINIPATH_HWE_SERDESPLLFAILED; + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, + dd->ipath_hwerrmask); + } + + ipath_dev_err(dd, "%s hardware error\n", msg); + /* + * For /sys status file. if no trailing } is copied, we'll + * know it was truncated. + */ + if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg) + snprintf(dd->ipath_freezemsg, dd->ipath_freezelen, + "{%s}", msg); +bail:; +} + +/** + * ipath_7220_boardname - fill in the board name + * @dd: the infinipath device + * @name: the output buffer + * @namelen: the size of the output buffer + * + * info is based on the board revision register + */ +static int ipath_7220_boardname(struct ipath_devdata *dd, char *name, + size_t namelen) +{ + char *n = NULL; + u8 boardrev = dd->ipath_boardrev; + int ret; + + if (boardrev == 15) { + /* + * Emulator sometimes comes up all-ones, rather than zero. + */ + boardrev = 0; + dd->ipath_boardrev = boardrev; + } + switch (boardrev) { + case 0: + n = "InfiniPath_7220_Emulation"; + break; + case 1: + n = "InfiniPath_QLE7240"; + break; + case 2: + n = "InfiniPath_QLE7280"; + break; + case 3: + n = "InfiniPath_QLE7242"; + break; + case 4: + n = "InfiniPath_QEM7240"; + break; + case 5: + n = "InfiniPath_QMI7240"; + break; + case 6: + n = "InfiniPath_QMI7264"; + break; + case 7: + n = "InfiniPath_QMH7240"; + break; + case 8: + n = "InfiniPath_QME7240"; + break; + case 9: + n = "InfiniPath_QLE7250"; + break; + case 10: + n = "InfiniPath_QLE7290"; + break; + case 11: + n = "InfiniPath_QEM7250"; + break; + case 12: + n = "InfiniPath_QLE-Bringup"; + break; + default: + ipath_dev_err(dd, + "Don't yet know about board with ID %u\n", + boardrev); + snprintf(name, namelen, "Unknown_InfiniPath_PCIe_%u", + boardrev); + break; + } + if (n) + snprintf(name, namelen, "%s", n); + + if (dd->ipath_majrev != 5 || !dd->ipath_minrev || + dd->ipath_minrev > 2) { + ipath_dev_err(dd, "Unsupported InfiniPath hardware " + "revision %u.%u!\n", + dd->ipath_majrev, dd->ipath_minrev); + ret = 1; + } else if (dd->ipath_minrev == 1) { + /* Rev1 chips are prototype. Complain, but allow use */ + ipath_dev_err(dd, "Unsupported hardware " + "revision %u.%u, Contact support@qlogic.com\n", + dd->ipath_majrev, dd->ipath_minrev); + ret = 0; + } else + ret = 0; + + /* + * Set here not in ipath_init_*_funcs because we have to do + * it after we can read chip registers. + */ + dd->ipath_ureg_align = 0x10000; /* 64KB alignment */ + + return ret; +} + +/** + * ipath_7220_init_hwerrors - enable hardware errors + * @dd: the infinipath device + * + * now that we have finished initializing everything that might reasonably + * cause a hardware error, and cleared those errors bits as they occur, + * we can enable hardware errors in the mask (potentially enabling + * freeze mode), and enable hardware errors as errors (along with + * everything else) in errormask + */ +static void ipath_7220_init_hwerrors(struct ipath_devdata *dd) +{ + ipath_err_t val; + u64 extsval; + + extsval = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus); + + if (!(extsval & (INFINIPATH_EXTS_MEMBIST_ENDTEST | + INFINIPATH_EXTS_MEMBIST_DISABLED))) + ipath_dev_err(dd, "MemBIST did not complete!\n"); + if (extsval & INFINIPATH_EXTS_MEMBIST_DISABLED) + dev_info(&dd->pcidev->dev, "MemBIST is disabled.\n"); + + val = ~0ULL; /* barring bugs, all hwerrors become interrupts, */ + + if (!dd->ipath_boardrev) /* no PLL for Emulator */ + val &= ~INFINIPATH_HWE_SERDESPLLFAILED; + + if (dd->ipath_minrev == 1) + val &= ~(1ULL << 42); /* TXE LaunchFIFO Parity rev1 issue */ + + val &= ~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR; + dd->ipath_hwerrmask = val; + + /* + * special trigger "error" is for debugging purposes. It + * works around a processor/chipset problem. The error + * interrupt allows us to count occurrences, but we don't + * want to pay the overhead for normal use. Emulation only + */ + if (!dd->ipath_boardrev) + dd->ipath_maskederrs = INFINIPATH_E_SENDSPECIALTRIGGER; +} + +/* + * All detailed interaction with the SerDes has been moved to ipath_sd7220.c + * + * The portion of IBA7220-specific bringup_serdes() that actually deals with + * registers and memory within the SerDes itself is ipath_sd7220_init(). + */ + +/** + * ipath_7220_bringup_serdes - bring up the serdes + * @dd: the infinipath device + */ +static int ipath_7220_bringup_serdes(struct ipath_devdata *dd) +{ + int ret = 0; + u64 val, prev_val, guid; + int was_reset; /* Note whether uC was reset */ + + ipath_dbg("Trying to bringup serdes\n"); + + if (ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus) & + INFINIPATH_HWE_SERDESPLLFAILED) { + ipath_dbg("At start, serdes PLL failed bit set " + "in hwerrstatus, clearing and continuing\n"); + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear, + INFINIPATH_HWE_SERDESPLLFAILED); + } + + if (!dd->ipath_ibcddrctrl) { + /* not on re-init after reset */ + dd->ipath_ibcddrctrl = + ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcddrctrl); + + if (dd->ipath_link_speed_enabled == + (IPATH_IB_SDR | IPATH_IB_DDR)) + dd->ipath_ibcddrctrl |= + IBA7220_IBC_SPEED_AUTONEG_MASK | + IBA7220_IBC_IBTA_1_2_MASK; + else + dd->ipath_ibcddrctrl |= + dd->ipath_link_speed_enabled == IPATH_IB_DDR + ? IBA7220_IBC_SPEED_DDR : + IBA7220_IBC_SPEED_SDR; + if ((dd->ipath_link_width_enabled & (IB_WIDTH_1X | + IB_WIDTH_4X)) == (IB_WIDTH_1X | IB_WIDTH_4X)) + dd->ipath_ibcddrctrl |= IBA7220_IBC_WIDTH_AUTONEG; + else + dd->ipath_ibcddrctrl |= + dd->ipath_link_width_enabled == IB_WIDTH_4X + ? IBA7220_IBC_WIDTH_4X_ONLY : + IBA7220_IBC_WIDTH_1X_ONLY; + + /* always enable these on driver reload, not sticky */ + dd->ipath_ibcddrctrl |= + IBA7220_IBC_RXPOL_MASK << IBA7220_IBC_RXPOL_SHIFT; + dd->ipath_ibcddrctrl |= + IBA7220_IBC_HRTBT_MASK << IBA7220_IBC_HRTBT_SHIFT; + /* + * automatic lane reversal detection for receive + * doesn't work correctly in rev 1, so disable it + * on that rev, otherwise enable (disabling not + * sticky across reload for >rev1) + */ + if (dd->ipath_minrev == 1) + dd->ipath_ibcddrctrl &= + ~IBA7220_IBC_LANE_REV_SUPPORTED; + else + dd->ipath_ibcddrctrl |= + IBA7220_IBC_LANE_REV_SUPPORTED; + } + + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcddrctrl, + dd->ipath_ibcddrctrl); + + ipath_write_kreg(dd, IPATH_KREG_OFFSET(IBNCModeCtrl), 0Ull); + + /* IBA7220 has SERDES MPU reset in D0 of what _was_ IBPLLCfg */ + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibserdesctrl); + /* remember if uC was in Reset or not, for dactrim */ + was_reset = (val & 1); + ipath_cdbg(VERBOSE, "IBReset %s xgxsconfig %llx\n", + was_reset ? "Asserted" : "Negated", (unsigned long long) + ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig)); + + if (dd->ipath_boardrev) { + /* + * Hardware is not emulator, and may have been reset. Init it. + * Below will release reset, but needs to know if chip was + * originally in reset, to only trim DACs on first time + * after chip reset or powercycle (not driver reload) + */ + ret = ipath_sd7220_init(dd, was_reset); + } + + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig); + prev_val = val; + val |= INFINIPATH_XGXS_FC_SAFE; + if (val != prev_val) { + ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + } + if (val & INFINIPATH_XGXS_RESET) + val &= ~INFINIPATH_XGXS_RESET; + if (val != prev_val) + ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val); + + ipath_cdbg(VERBOSE, "done: xgxs=%llx from %llx\n", + (unsigned long long) + ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig), + prev_val); + + guid = be64_to_cpu(dd->ipath_guid); + + if (!guid) { + /* have to have something, so use likely unique tsc */ + guid = get_cycles(); + ipath_dbg("No GUID for heartbeat, faking %llx\n", + (unsigned long long)guid); + } else + ipath_cdbg(VERBOSE, "Wrote %llX to HRTBT_GUID\n", guid); + ipath_write_kreg(dd, dd->ipath_kregs->kr_hrtbt_guid, guid); + return ret; +} + +static void ipath_7220_config_jint(struct ipath_devdata *dd, + u16 idle_ticks, u16 max_packets) +{ + + /* + * We can request a receive interrupt for 1 or more packets + * from current offset. + */ + if (idle_ticks == 0 || max_packets == 0) + /* interrupt after one packet if no mitigation */ + dd->ipath_rhdrhead_intr_off = + 1ULL << IBA7220_HDRHEAD_PKTINT_SHIFT; + else + /* Turn off RcvHdrHead interrupts if using mitigation */ + dd->ipath_rhdrhead_intr_off = 0ULL; + + /* refresh kernel RcvHdrHead registers... */ + ipath_write_ureg(dd, ur_rcvhdrhead, + dd->ipath_rhdrhead_intr_off | + dd->ipath_pd[0]->port_head, 0); + + dd->ipath_jint_max_packets = max_packets; + dd->ipath_jint_idle_ticks = idle_ticks; + ipath_write_kreg(dd, dd->ipath_kregs->kr_jintreload, + ((u64) max_packets << INFINIPATH_JINT_PACKETSHIFT) | + idle_ticks); +} + +/** + * ipath_7220_quiet_serdes - set serdes to txidle + * @dd: the infinipath device + * Called when driver is being unloaded + */ +static void ipath_7220_quiet_serdes(struct ipath_devdata *dd) +{ + u64 val; + dd->ipath_flags &= ~IPATH_IB_AUTONEG_INPROG; + wake_up(&dd->ipath_autoneg_wait); + cancel_delayed_work(&dd->ipath_autoneg_work); + flush_scheduled_work(); + ipath_shutdown_relock_poll(dd); + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig); + val |= INFINIPATH_XGXS_RESET; + ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val); +} + +static int ipath_7220_intconfig(struct ipath_devdata *dd) +{ + ipath_7220_config_jint(dd, dd->ipath_jint_idle_ticks, + dd->ipath_jint_max_packets); + return 0; +} + +/** + * ipath_setup_7220_setextled - set the state of the two external LEDs + * @dd: the infinipath device + * @lst: the L state + * @ltst: the LT state + * + * These LEDs indicate the physical and logical state of IB link. + * For this chip (at least with recommended board pinouts), LED1 + * is Yellow (logical state) and LED2 is Green (physical state), + * + * Note: We try to match the Mellanox HCA LED behavior as best + * we can. Green indicates physical link state is OK (something is + * plugged in, and we can train). + * Amber indicates the link is logically up (ACTIVE). + * Mellanox further blinks the amber LED to indicate data packet + * activity, but we have no hardware support for that, so it would + * require waking up every 10-20 msecs and checking the counters + * on the chip, and then turning the LED off if appropriate. That's + * visible overhead, so not something we will do. + * + */ +static void ipath_setup_7220_setextled(struct ipath_devdata *dd, u64 lst, + u64 ltst) +{ + u64 extctl, ledblink = 0; + unsigned long flags = 0; + + /* the diags use the LED to indicate diag info, so we leave + * the external LED alone when the diags are running */ + if (ipath_diag_inuse) + return; + + /* Allow override of LED display for, e.g. Locating system in rack */ + if (dd->ipath_led_override) { + ltst = (dd->ipath_led_override & IPATH_LED_PHYS) + ? INFINIPATH_IBCS_LT_STATE_LINKUP + : INFINIPATH_IBCS_LT_STATE_DISABLED; + lst = (dd->ipath_led_override & IPATH_LED_LOG) + ? INFINIPATH_IBCS_L_STATE_ACTIVE + : INFINIPATH_IBCS_L_STATE_DOWN; + } + + spin_lock_irqsave(&dd->ipath_gpio_lock, flags); + extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON | + INFINIPATH_EXTC_LED2PRIPORT_ON); + if (ltst == INFINIPATH_IBCS_LT_STATE_LINKUP) { + extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON; + /* + * counts are in chip clock (4ns) periods. + * This is 1/16 sec (66.6ms) on, + * 3/16 sec (187.5 ms) off, with packets rcvd + */ + ledblink = ((66600*1000UL/4) << IBA7220_LEDBLINK_ON_SHIFT) + | ((187500*1000UL/4) << IBA7220_LEDBLINK_OFF_SHIFT); + } + if (lst == INFINIPATH_IBCS_L_STATE_ACTIVE) + extctl |= INFINIPATH_EXTC_LED2PRIPORT_ON; + dd->ipath_extctrl = extctl; + ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl); + spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags); + + if (ledblink) /* blink the LED on packet receive */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvpktledcnt, + ledblink); +} + +/* + * Similar to pci_intx(pdev, 1), except that we make sure + * msi is off... + */ +static void ipath_enable_intx(struct pci_dev *pdev) +{ + u16 cw, new; + int pos; + + /* first, turn on INTx */ + pci_read_config_word(pdev, PCI_COMMAND, &cw); + new = cw & ~PCI_COMMAND_INTX_DISABLE; + if (new != cw) + pci_write_config_word(pdev, PCI_COMMAND, new); + + /* then turn off MSI */ + pos = pci_find_capability(pdev, PCI_CAP_ID_MSI); + if (pos) { + pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw); + new = cw & ~PCI_MSI_FLAGS_ENABLE; + if (new != cw) + pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new); + } +} + +static int ipath_msi_enabled(struct pci_dev *pdev) +{ + int pos, ret = 0; + + pos = pci_find_capability(pdev, PCI_CAP_ID_MSI); + if (pos) { + u16 cw; + + pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw); + ret = !!(cw & PCI_MSI_FLAGS_ENABLE); + } + return ret; +} + +/* + * disable msi interrupt if enabled, and clear the flag. + * flag is used primarily for the fallback to IntX, but + * is also used in reinit after reset as a flag. + */ +static void ipath_7220_nomsi(struct ipath_devdata *dd) +{ + dd->ipath_msi_lo = 0; +#ifdef CONFIG_PCI_MSI + if (ipath_msi_enabled(dd->pcidev)) { + /* + * free, but don't zero; later kernels require + * it be freed before disable_msi, so the intx + * setup has to request it again. + */ + if (dd->ipath_irq) + free_irq(dd->ipath_irq, dd); + pci_disable_msi(dd->pcidev); + } +#endif +} + +/* + * ipath_setup_7220_cleanup - clean up any per-chip chip-specific stuff + * @dd: the infinipath device + * + * Nothing but msi interrupt cleanup for now. + * + * This is called during driver unload. + */ +static void ipath_setup_7220_cleanup(struct ipath_devdata *dd) +{ + ipath_7220_nomsi(dd); +} + + +static void ipath_7220_pcie_params(struct ipath_devdata *dd, u32 boardrev) +{ + u16 linkstat, minwidth, speed; + int pos; + + pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP); + if (!pos) { + ipath_dev_err(dd, "Can't find PCI Express capability!\n"); + goto bail; + } + + pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA, + &linkstat); + /* + * speed is bits 0-4, linkwidth is bits 4-8 + * no defines for them in headers + */ + speed = linkstat & 0xf; + linkstat >>= 4; + linkstat &= 0x1f; + dd->ipath_lbus_width = linkstat; + switch (boardrev) { + case 0: + case 2: + case 10: + case 12: + minwidth = 16; /* x16 capable boards */ + break; + default: + minwidth = 8; /* x8 capable boards */ + break; + } + + switch (speed) { + case 1: + dd->ipath_lbus_speed = 2500; /* Gen1, 2.5GHz */ + break; + case 2: + dd->ipath_lbus_speed = 5000; /* Gen1, 5GHz */ + break; + default: /* not defined, assume gen1 */ + dd->ipath_lbus_speed = 2500; + break; + } + + if (linkstat < minwidth) + ipath_dev_err(dd, + "PCIe width %u (x%u HCA), performance " + "reduced\n", linkstat, minwidth); + else + ipath_cdbg(VERBOSE, "PCIe speed %u width %u (x%u HCA)\n", + dd->ipath_lbus_speed, linkstat, minwidth); + + if (speed != 1) + ipath_dev_err(dd, + "PCIe linkspeed %u is incorrect; " + "should be 1 (2500)!\n", speed); + +bail: + /* fill in string, even on errors */ + snprintf(dd->ipath_lbus_info, sizeof(dd->ipath_lbus_info), + "PCIe,%uMHz,x%u\n", + dd->ipath_lbus_speed, + dd->ipath_lbus_width); + return; +} + + +/** + * ipath_setup_7220_config - setup PCIe config related stuff + * @dd: the infinipath device + * @pdev: the PCI device + * + * The pci_enable_msi() call will fail on systems with MSI quirks + * such as those with AMD8131, even if the device of interest is not + * attached to that device, (in the 2.6.13 - 2.6.15 kernels, at least, fixed + * late in 2.6.16). + * All that can be done is to edit the kernel source to remove the quirk + * check until that is fixed. + * We do not need to call enable_msi() for our HyperTransport chip, + * even though it uses MSI, and we want to avoid the quirk warning, so + * So we call enable_msi only for PCIe. If we do end up needing + * pci_enable_msi at some point in the future for HT, we'll move the + * call back into the main init_one code. + * We save the msi lo and hi values, so we can restore them after + * chip reset (the kernel PCI infrastructure doesn't yet handle that + * correctly). + */ +static int ipath_setup_7220_config(struct ipath_devdata *dd, + struct pci_dev *pdev) +{ + int pos, ret = -1; + u32 boardrev; + + dd->ipath_msi_lo = 0; /* used as a flag during reset processing */ +#ifdef CONFIG_PCI_MSI + pos = pci_find_capability(pdev, PCI_CAP_ID_MSI); + if (!strcmp(int_type, "force_msi") || !strcmp(int_type, "auto")) + ret = pci_enable_msi(pdev); + if (ret) { + if (!strcmp(int_type, "force_msi")) { + ipath_dev_err(dd, "pci_enable_msi failed: %d, " + "force_msi is on, so not continuing.\n", + ret); + return ret; + } + + ipath_enable_intx(pdev); + if (!strcmp(int_type, "auto")) + ipath_dev_err(dd, "pci_enable_msi failed: %d, " + "falling back to INTx\n", ret); + } else if (pos) { + u16 control; + pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO, + &dd->ipath_msi_lo); + pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI, + &dd->ipath_msi_hi); + pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, + &control); + /* now save the data (vector) info */ + pci_read_config_word(pdev, + pos + ((control & PCI_MSI_FLAGS_64BIT) + ? PCI_MSI_DATA_64 : + PCI_MSI_DATA_32), + &dd->ipath_msi_data); + } else + ipath_dev_err(dd, "Can't find MSI capability, " + "can't save MSI settings for reset\n"); +#else + ipath_dbg("PCI_MSI not configured, using IntX interrupts\n"); + ipath_enable_intx(pdev); +#endif + + dd->ipath_irq = pdev->irq; + + /* + * We save the cachelinesize also, although it doesn't + * really matter. + */ + pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE, + &dd->ipath_pci_cacheline); + + /* + * this function called early, ipath_boardrev not set yet. Can't + * use ipath_read_kreg64() yet, too early in init, so use readq() + */ + boardrev = (readq(&dd->ipath_kregbase[dd->ipath_kregs->kr_revision]) + >> INFINIPATH_R_BOARDID_SHIFT) & INFINIPATH_R_BOARDID_MASK; + + ipath_7220_pcie_params(dd, boardrev); + + dd->ipath_flags |= IPATH_NODMA_RTAIL | IPATH_HAS_SEND_DMA | + IPATH_HAS_PBC_CNT | IPATH_HAS_THRESH_UPDATE; + dd->ipath_pioupd_thresh = 4U; /* set default update threshold */ + return 0; +} + +static void ipath_init_7220_variables(struct ipath_devdata *dd) +{ + /* + * setup the register offsets, since they are different for each + * chip + */ + dd->ipath_kregs = &ipath_7220_kregs; + dd->ipath_cregs = &ipath_7220_cregs; + + /* + * bits for selecting i2c direction and values, + * used for I2C serial flash + */ + dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM; + dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM; + dd->ipath_gpio_sda = IPATH_GPIO_SDA; + dd->ipath_gpio_scl = IPATH_GPIO_SCL; + + /* + * Fill in data for field-values that change in IBA7220. + * We dynamically specify only the mask for LINKTRAININGSTATE + * and only the shift for LINKSTATE, as they are the only ones + * that change. Also precalculate the 3 link states of interest + * and the combined mask. + */ + dd->ibcs_ls_shift = IBA7220_IBCS_LINKSTATE_SHIFT; + dd->ibcs_lts_mask = IBA7220_IBCS_LINKTRAININGSTATE_MASK; + dd->ibcs_mask = (INFINIPATH_IBCS_LINKSTATE_MASK << + dd->ibcs_ls_shift) | dd->ibcs_lts_mask; + dd->ib_init = (INFINIPATH_IBCS_LT_STATE_LINKUP << + INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) | + (INFINIPATH_IBCS_L_STATE_INIT << dd->ibcs_ls_shift); + dd->ib_arm = (INFINIPATH_IBCS_LT_STATE_LINKUP << + INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) | + (INFINIPATH_IBCS_L_STATE_ARM << dd->ibcs_ls_shift); + dd->ib_active = (INFINIPATH_IBCS_LT_STATE_LINKUP << + INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) | + (INFINIPATH_IBCS_L_STATE_ACTIVE << dd->ibcs_ls_shift); + + /* + * Fill in data for ibcc field-values that change in IBA7220. + * We dynamically specify only the mask for LINKINITCMD + * and only the shift for LINKCMD and MAXPKTLEN, as they are + * the only ones that change. + */ + dd->ibcc_lic_mask = IBA7220_IBCC_LINKINITCMD_MASK; + dd->ibcc_lc_shift = IBA7220_IBCC_LINKCMD_SHIFT; + dd->ibcc_mpl_shift = IBA7220_IBCC_MAXPKTLEN_SHIFT; + + /* Fill in shifts for RcvCtrl. */ + dd->ipath_r_portenable_shift = INFINIPATH_R_PORTENABLE_SHIFT; + dd->ipath_r_intravail_shift = IBA7220_R_INTRAVAIL_SHIFT; + dd->ipath_r_tailupd_shift = IBA7220_R_TAILUPD_SHIFT; + dd->ipath_r_portcfg_shift = IBA7220_R_PORTCFG_SHIFT; + + /* variables for sanity checking interrupt and errors */ + dd->ipath_hwe_bitsextant = + (INFINIPATH_HWE_RXEMEMPARITYERR_MASK << + INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) | + (INFINIPATH_HWE_TXEMEMPARITYERR_MASK << + INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) | + (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK << + INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) | + INFINIPATH_HWE_PCIE1PLLFAILED | + INFINIPATH_HWE_PCIE0PLLFAILED | + INFINIPATH_HWE_PCIEPOISONEDTLP | + INFINIPATH_HWE_PCIECPLTIMEOUT | + INFINIPATH_HWE_PCIEBUSPARITYXTLH | + INFINIPATH_HWE_PCIEBUSPARITYXADM | + INFINIPATH_HWE_PCIEBUSPARITYRADM | + INFINIPATH_HWE_MEMBISTFAILED | + INFINIPATH_HWE_COREPLL_FBSLIP | + INFINIPATH_HWE_COREPLL_RFSLIP | + INFINIPATH_HWE_SERDESPLLFAILED | + INFINIPATH_HWE_IBCBUSTOSPCPARITYERR | + INFINIPATH_HWE_IBCBUSFRSPCPARITYERR | + INFINIPATH_HWE_PCIECPLDATAQUEUEERR | + INFINIPATH_HWE_PCIECPLHDRQUEUEERR | + INFINIPATH_HWE_SDMAMEMREADERR | + INFINIPATH_HWE_CLK_UC_PLLNOTLOCKED | + INFINIPATH_HWE_PCIESERDESQ0PCLKNOTDETECT | + INFINIPATH_HWE_PCIESERDESQ1PCLKNOTDETECT | + INFINIPATH_HWE_PCIESERDESQ2PCLKNOTDETECT | + INFINIPATH_HWE_PCIESERDESQ3PCLKNOTDETECT | + INFINIPATH_HWE_DDSRXEQMEMORYPARITYERR | + INFINIPATH_HWE_IB_UC_MEMORYPARITYERR | + INFINIPATH_HWE_PCIE_UC_OCT0MEMORYPARITYERR | + INFINIPATH_HWE_PCIE_UC_OCT1MEMORYPARITYERR; + dd->ipath_i_bitsextant = + INFINIPATH_I_SDMAINT | INFINIPATH_I_SDMADISABLED | + (INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) | + (INFINIPATH_I_RCVAVAIL_MASK << + INFINIPATH_I_RCVAVAIL_SHIFT) | + INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT | + INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO | + INFINIPATH_I_JINT | INFINIPATH_I_SERDESTRIMDONE; + dd->ipath_e_bitsextant = + INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC | + INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN | + INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN | + INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RUNEXPCHAR | + INFINIPATH_E_RUNSUPVL | INFINIPATH_E_REBP | + INFINIPATH_E_RIBFLOW | INFINIPATH_E_RBADVERSION | + INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL | + INFINIPATH_E_RBADTID | INFINIPATH_E_RHDRLEN | + INFINIPATH_E_RHDR | INFINIPATH_E_RIBLOSTLINK | + INFINIPATH_E_SENDSPECIALTRIGGER | + INFINIPATH_E_SDMADISABLED | INFINIPATH_E_SMINPKTLEN | + INFINIPATH_E_SMAXPKTLEN | INFINIPATH_E_SUNDERRUN | + INFINIPATH_E_SPKTLEN | INFINIPATH_E_SDROPPEDSMPPKT | + INFINIPATH_E_SDROPPEDDATAPKT | + INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM | + INFINIPATH_E_SUNSUPVL | INFINIPATH_E_SENDBUFMISUSE | + INFINIPATH_E_SDMAGENMISMATCH | INFINIPATH_E_SDMAOUTOFBOUND | + INFINIPATH_E_SDMATAILOUTOFBOUND | INFINIPATH_E_SDMABASE | + INFINIPATH_E_SDMA1STDESC | INFINIPATH_E_SDMARPYTAG | + INFINIPATH_E_SDMADWEN | INFINIPATH_E_SDMAMISSINGDW | + INFINIPATH_E_SDMAUNEXPDATA | + INFINIPATH_E_IBSTATUSCHANGED | INFINIPATH_E_INVALIDADDR | + INFINIPATH_E_RESET | INFINIPATH_E_HARDWARE | + INFINIPATH_E_SDMADESCADDRMISALIGN | + INFINIPATH_E_INVALIDEEPCMD; + + dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK; + dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK; + dd->ipath_i_rcvavail_shift = INFINIPATH_I_RCVAVAIL_SHIFT; + dd->ipath_i_rcvurg_shift = INFINIPATH_I_RCVURG_SHIFT; + dd->ipath_flags |= IPATH_INTREG_64 | IPATH_HAS_MULT_IB_SPEED + | IPATH_HAS_LINK_LATENCY; + + /* + * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity. + * 2 is Some Misc, 3 is reserved for future. + */ + dd->ipath_eep_st_masks[0].hwerrs_to_log = + INFINIPATH_HWE_TXEMEMPARITYERR_MASK << + INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT; + + dd->ipath_eep_st_masks[1].hwerrs_to_log = + INFINIPATH_HWE_RXEMEMPARITYERR_MASK << + INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT; + + dd->ipath_eep_st_masks[2].errs_to_log = INFINIPATH_E_RESET; + + ipath_linkrecovery = 0; + + init_waitqueue_head(&dd->ipath_autoneg_wait); + INIT_DELAYED_WORK(&dd->ipath_autoneg_work, autoneg_work); + + dd->ipath_link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X; + dd->ipath_link_speed_supported = IPATH_IB_SDR | IPATH_IB_DDR; + + dd->ipath_link_width_enabled = dd->ipath_link_width_supported; + dd->ipath_link_speed_enabled = dd->ipath_link_speed_supported; + /* + * set the initial values to reasonable default, will be set + * for real when link is up. + */ + dd->ipath_link_width_active = IB_WIDTH_4X; + dd->ipath_link_speed_active = IPATH_IB_SDR; + dd->delay_mult = rate_to_delay[0][1]; +} + + +/* + * Setup the MSI stuff again after a reset. I'd like to just call + * pci_enable_msi() and request_irq() again, but when I do that, + * the MSI enable bit doesn't get set in the command word, and + * we switch to to a different interrupt vector, which is confusing, + * so I instead just do it all inline. Perhaps somehow can tie this + * into the PCIe hotplug support at some point + * Note, because I'm doing it all here, I don't call pci_disable_msi() + * or free_irq() at the start of ipath_setup_7220_reset(). + */ +static int ipath_reinit_msi(struct ipath_devdata *dd) +{ + int ret = 0; +#ifdef CONFIG_PCI_MSI + int pos; + u16 control; + if (!dd->ipath_msi_lo) /* Using intX, or init problem */ + goto bail; + + pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI); + if (!pos) { + ipath_dev_err(dd, "Can't find MSI capability, " + "can't restore MSI settings\n"); + goto bail; + } + ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n", + dd->ipath_msi_lo, pos + PCI_MSI_ADDRESS_LO); + pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO, + dd->ipath_msi_lo); + ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n", + dd->ipath_msi_hi, pos + PCI_MSI_ADDRESS_HI); + pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI, + dd->ipath_msi_hi); + pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control); + if (!(control & PCI_MSI_FLAGS_ENABLE)) { + ipath_cdbg(VERBOSE, "MSI control at off %x was %x, " + "setting MSI enable (%x)\n", pos + PCI_MSI_FLAGS, + control, control | PCI_MSI_FLAGS_ENABLE); + control |= PCI_MSI_FLAGS_ENABLE; + pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, + control); + } + /* now rewrite the data (vector) info */ + pci_write_config_word(dd->pcidev, pos + + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8), + dd->ipath_msi_data); + ret = 1; +bail: +#endif + if (!ret) { + ipath_dbg("Using IntX, MSI disabled or not configured\n"); + ipath_enable_intx(dd->pcidev); + ret = 1; + } + /* + * We restore the cachelinesize also, although it doesn't really + * matter. + */ + pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, + dd->ipath_pci_cacheline); + /* and now set the pci master bit again */ + pci_set_master(dd->pcidev); + + return ret; +} + +/* + * This routine sleeps, so it can only be called from user context, not + * from interrupt context. If we need interrupt context, we can split + * it into two routines. + */ +static int ipath_setup_7220_reset(struct ipath_devdata *dd) +{ + u64 val; + int i; + int ret; + u16 cmdval; + + pci_read_config_word(dd->pcidev, PCI_COMMAND, &cmdval); + + /* Use dev_err so it shows up in logs, etc. */ + ipath_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->ipath_unit); + + /* keep chip from being accessed in a few places */ + dd->ipath_flags &= ~(IPATH_INITTED | IPATH_PRESENT); + val = dd->ipath_control | INFINIPATH_C_RESET; + ipath_write_kreg(dd, dd->ipath_kregs->kr_control, val); + mb(); + + for (i = 1; i <= 5; i++) { + int r; + + /* + * Allow MBIST, etc. to complete; longer on each retry. + * We sometimes get machine checks from bus timeout if no + * response, so for now, make it *really* long. + */ + msleep(1000 + (1 + i) * 2000); + r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0, + dd->ipath_pcibar0); + if (r) + ipath_dev_err(dd, "rewrite of BAR0 failed: %d\n", r); + r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1, + dd->ipath_pcibar1); + if (r) + ipath_dev_err(dd, "rewrite of BAR1 failed: %d\n", r); + /* now re-enable memory access */ + pci_write_config_word(dd->pcidev, PCI_COMMAND, cmdval); + r = pci_enable_device(dd->pcidev); + if (r) + ipath_dev_err(dd, "pci_enable_device failed after " + "reset: %d\n", r); + /* + * whether it fully enabled or not, mark as present, + * again (but not INITTED) + */ + dd->ipath_flags |= IPATH_PRESENT; + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision); + if (val == dd->ipath_revision) { + ipath_cdbg(VERBOSE, "Got matching revision " + "register %llx on try %d\n", + (unsigned long long) val, i); + ret = ipath_reinit_msi(dd); + goto bail; + } + /* Probably getting -1 back */ + ipath_dbg("Didn't get expected revision register, " + "got %llx, try %d\n", (unsigned long long) val, + i + 1); + } + ret = 0; /* failed */ + +bail: + if (ret) + ipath_7220_pcie_params(dd, dd->ipath_boardrev); + + return ret; +} + +/** + * ipath_7220_put_tid - write a TID to the chip + * @dd: the infinipath device + * @tidptr: pointer to the expected TID (in chip) to udpate + * @tidtype: 0 for eager, 1 for expected + * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing + * + * This exists as a separate routine to allow for selection of the + * appropriate "flavor". The static calls in cleanup just use the + * revision-agnostic form, as they are not performance critical. + */ +static void ipath_7220_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr, + u32 type, unsigned long pa) +{ + if (pa != dd->ipath_tidinvalid) { + u64 chippa = pa >> IBA7220_TID_PA_SHIFT; + + /* paranoia checks */ + if (pa != (chippa << IBA7220_TID_PA_SHIFT)) { + dev_info(&dd->pcidev->dev, "BUG: physaddr %lx " + "not 2KB aligned!\n", pa); + return; + } + if (pa >= (1UL << IBA7220_TID_SZ_SHIFT)) { + ipath_dev_err(dd, + "BUG: Physical page address 0x%lx " + "larger than supported\n", pa); + return; + } + + if (type == RCVHQ_RCV_TYPE_EAGER) + chippa |= dd->ipath_tidtemplate; + else /* for now, always full 4KB page */ + chippa |= IBA7220_TID_SZ_4K; + writeq(chippa, tidptr); + } else + writeq(pa, tidptr); + mmiowb(); +} + +/** + * ipath_7220_clear_tid - clear all TID entries for a port, expected and eager + * @dd: the infinipath device + * @port: the port + * + * clear all TID entries for a port, expected and eager. + * Used from ipath_close(). On this chip, TIDs are only 32 bits, + * not 64, but they are still on 64 bit boundaries, so tidbase + * is declared as u64 * for the pointer math, even though we write 32 bits + */ +static void ipath_7220_clear_tids(struct ipath_devdata *dd, unsigned port) +{ + u64 __iomem *tidbase; + unsigned long tidinv; + int i; + + if (!dd->ipath_kregbase) + return; + + ipath_cdbg(VERBOSE, "Invalidate TIDs for port %u\n", port); + + tidinv = dd->ipath_tidinvalid; + tidbase = (u64 __iomem *) + ((char __iomem *)(dd->ipath_kregbase) + + dd->ipath_rcvtidbase + + port * dd->ipath_rcvtidcnt * sizeof(*tidbase)); + + for (i = 0; i < dd->ipath_rcvtidcnt; i++) + ipath_7220_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED, + tidinv); + + tidbase = (u64 __iomem *) + ((char __iomem *)(dd->ipath_kregbase) + + dd->ipath_rcvegrbase + port_egrtid_idx(dd, port) + * sizeof(*tidbase)); + + for (i = port ? dd->ipath_rcvegrcnt : dd->ipath_p0_rcvegrcnt; i; i--) + ipath_7220_put_tid(dd, &tidbase[i-1], RCVHQ_RCV_TYPE_EAGER, + tidinv); +} + +/** + * ipath_7220_tidtemplate - setup constants for TID updates + * @dd: the infinipath device + * + * We setup stuff that we use a lot, to avoid calculating each time + */ +static void ipath_7220_tidtemplate(struct ipath_devdata *dd) +{ + /* For now, we always allocate 4KB buffers (at init) so we can + * receive max size packets. We may want a module parameter to + * specify 2KB or 4KB and/or make be per port instead of per device + * for those who want to reduce memory footprint. Note that the + * ipath_rcvhdrentsize size must be large enough to hold the largest + * IB header (currently 96 bytes) that we expect to handle (plus of + * course the 2 dwords of RHF). + */ + if (dd->ipath_rcvegrbufsize == 2048) + dd->ipath_tidtemplate = IBA7220_TID_SZ_2K; + else if (dd->ipath_rcvegrbufsize == 4096) + dd->ipath_tidtemplate = IBA7220_TID_SZ_4K; + else { + dev_info(&dd->pcidev->dev, "BUG: unsupported egrbufsize " + "%u, using %u\n", dd->ipath_rcvegrbufsize, + 4096); + dd->ipath_tidtemplate = IBA7220_TID_SZ_4K; + } + dd->ipath_tidinvalid = 0; +} + +static int ipath_7220_early_init(struct ipath_devdata *dd) +{ + u32 i, s; + + if (strcmp(int_type, "auto") && + strcmp(int_type, "force_msi") && + strcmp(int_type, "force_intx")) { + ipath_dev_err(dd, "Invalid interrupt_type: '%s', expecting " + "auto, force_msi or force_intx\n", int_type); + return -EINVAL; + } + + /* + * Control[4] has been added to change the arbitration within + * the SDMA engine between favoring data fetches over descriptor + * fetches. ipath_sdma_fetch_arb==0 gives data fetches priority. + */ + if (ipath_sdma_fetch_arb && (dd->ipath_minrev > 1)) + dd->ipath_control |= 1<<4; + + dd->ipath_flags |= IPATH_4BYTE_TID; + + /* + * For openfabrics, we need to be able to handle an IB header of + * 24 dwords. HT chip has arbitrary sized receive buffers, so we + * made them the same size as the PIO buffers. This chip does not + * handle arbitrary size buffers, so we need the header large enough + * to handle largest IB header, but still have room for a 2KB MTU + * standard IB packet. + */ + dd->ipath_rcvhdrentsize = 24; + dd->ipath_rcvhdrsize = IPATH_DFLT_RCVHDRSIZE; + dd->ipath_rhf_offset = + dd->ipath_rcvhdrentsize - sizeof(u64) / sizeof(u32); + + dd->ipath_rcvegrbufsize = ipath_mtu4096 ? 4096 : 2048; + /* + * the min() check here is currently a nop, but it may not always + * be, depending on just how we do ipath_rcvegrbufsize + */ + dd->ipath_ibmaxlen = min(ipath_mtu4096 ? dd->ipath_piosize4k : + dd->ipath_piosize2k, + dd->ipath_rcvegrbufsize + + (dd->ipath_rcvhdrentsize << 2)); + dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen; + + ipath_7220_config_jint(dd, INFINIPATH_JINT_DEFAULT_IDLE_TICKS, + INFINIPATH_JINT_DEFAULT_MAX_PACKETS); + + if (dd->ipath_boardrev) /* no eeprom on emulator */ + ipath_get_eeprom_info(dd); + + /* start of code to check and print procmon */ + s = ipath_read_kreg32(dd, IPATH_KREG_OFFSET(ProcMon)); + s &= ~(1U<<31); /* clear done bit */ + s |= 1U<<14; /* clear counter (write 1 to clear) */ + ipath_write_kreg(dd, IPATH_KREG_OFFSET(ProcMon), s); + /* make sure clear_counter low long enough before start */ + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + + s &= ~(1U<<14); /* allow counter to count (before starting) */ + ipath_write_kreg(dd, IPATH_KREG_OFFSET(ProcMon), s); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + s = ipath_read_kreg32(dd, IPATH_KREG_OFFSET(ProcMon)); + + s |= 1U<<15; /* start the counter */ + s &= ~(1U<<31); /* clear done bit */ + s &= ~0x7ffU; /* clear frequency bits */ + s |= 0xe29; /* set frequency bits, in case cleared */ + ipath_write_kreg(dd, IPATH_KREG_OFFSET(ProcMon), s); + + s = 0; + for (i = 500; i > 0 && !(s&(1ULL<<31)); i--) { + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + s = ipath_read_kreg32(dd, IPATH_KREG_OFFSET(ProcMon)); + } + if (!(s&(1U<<31))) + ipath_dev_err(dd, "ProcMon register not valid: 0x%x\n", s); + else + ipath_dbg("ProcMon=0x%x, count=0x%x\n", s, (s>>16)&0x1ff); + + return 0; +} + +/** + * ipath_init_7220_get_base_info - set chip-specific flags for user code + * @pd: the infinipath port + * @kbase: ipath_base_info pointer + * + * We set the PCIE flag because the lower bandwidth on PCIe vs + * HyperTransport can affect some user packet algorithims. + */ +static int ipath_7220_get_base_info(struct ipath_portdata *pd, void *kbase) +{ + struct ipath_base_info *kinfo = kbase; + + kinfo->spi_runtime_flags |= + IPATH_RUNTIME_PCIE | IPATH_RUNTIME_NODMA_RTAIL | + IPATH_RUNTIME_SDMA; + + return 0; +} + +static void ipath_7220_free_irq(struct ipath_devdata *dd) +{ + free_irq(dd->ipath_irq, dd); + dd->ipath_irq = 0; +} + +static struct ipath_message_header * +ipath_7220_get_msgheader(struct ipath_devdata *dd, __le32 *rhf_addr) +{ + u32 offset = ipath_hdrget_offset(rhf_addr); + + return (struct ipath_message_header *) + (rhf_addr - dd->ipath_rhf_offset + offset); +} + +static void ipath_7220_config_ports(struct ipath_devdata *dd, ushort cfgports) +{ + u32 nchipports; + + nchipports = ipath_read_kreg32(dd, dd->ipath_kregs->kr_portcnt); + if (!cfgports) { + int ncpus = num_online_cpus(); + + if (ncpus <= 4) + dd->ipath_portcnt = 5; + else if (ncpus <= 8) + dd->ipath_portcnt = 9; + if (dd->ipath_portcnt) + ipath_dbg("Auto-configured for %u ports, %d cpus " + "online\n", dd->ipath_portcnt, ncpus); + } else if (cfgports <= nchipports) + dd->ipath_portcnt = cfgports; + if (!dd->ipath_portcnt) /* none of the above, set to max */ + dd->ipath_portcnt = nchipports; + /* + * chip can be configured for 5, 9, or 17 ports, and choice + * affects number of eager TIDs per port (1K, 2K, 4K). + */ + if (dd->ipath_portcnt > 9) + dd->ipath_rcvctrl |= 2ULL << IBA7220_R_PORTCFG_SHIFT; + else if (dd->ipath_portcnt > 5) + dd->ipath_rcvctrl |= 1ULL << IBA7220_R_PORTCFG_SHIFT; + /* else configure for default 5 receive ports */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, + dd->ipath_rcvctrl); + dd->ipath_p0_rcvegrcnt = 2048; /* always */ + if (dd->ipath_flags & IPATH_HAS_SEND_DMA) + dd->ipath_pioreserved = 1; /* reserve a buffer */ +} + + +static int ipath_7220_get_ib_cfg(struct ipath_devdata *dd, int which) +{ + int lsb, ret = 0; + u64 maskr; /* right-justified mask */ + + switch (which) { + case IPATH_IB_CFG_HRTBT: /* Get Heartbeat off/enable/auto */ + lsb = IBA7220_IBC_HRTBT_SHIFT; + maskr = IBA7220_IBC_HRTBT_MASK; + break; + + case IPATH_IB_CFG_LWID_ENB: /* Get allowed Link-width */ + ret = dd->ipath_link_width_enabled; + goto done; + + case IPATH_IB_CFG_LWID: /* Get currently active Link-width */ + ret = dd->ipath_link_width_active; + goto done; + + case IPATH_IB_CFG_SPD_ENB: /* Get allowed Link speeds */ + ret = dd->ipath_link_speed_enabled; + goto done; + + case IPATH_IB_CFG_SPD: /* Get current Link spd */ + ret = dd->ipath_link_speed_active; + goto done; + + case IPATH_IB_CFG_RXPOL_ENB: /* Get Auto-RX-polarity enable */ + lsb = IBA7220_IBC_RXPOL_SHIFT; + maskr = IBA7220_IBC_RXPOL_MASK; + break; + + case IPATH_IB_CFG_LREV_ENB: /* Get Auto-Lane-reversal enable */ + lsb = IBA7220_IBC_LREV_SHIFT; + maskr = IBA7220_IBC_LREV_MASK; + break; + + case IPATH_IB_CFG_LINKLATENCY: + ret = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcddrstatus) + & IBA7220_DDRSTAT_LINKLAT_MASK; + goto done; + + default: + ret = -ENOTSUPP; + goto done; + } + ret = (int)((dd->ipath_ibcddrctrl >> lsb) & maskr); +done: + return ret; +} + +static int ipath_7220_set_ib_cfg(struct ipath_devdata *dd, int which, u32 val) +{ + int lsb, ret = 0, setforce = 0; + u64 maskr; /* right-justified mask */ + + switch (which) { + case IPATH_IB_CFG_LIDLMC: + /* + * Set LID and LMC. Combined to avoid possible hazard + * caller puts LMC in 16MSbits, DLID in 16LSbits of val + */ + lsb = IBA7220_IBC_DLIDLMC_SHIFT; + maskr = IBA7220_IBC_DLIDLMC_MASK; + break; + + case IPATH_IB_CFG_HRTBT: /* set Heartbeat off/enable/auto */ + if (val & IPATH_IB_HRTBT_ON && + (dd->ipath_flags & IPATH_NO_HRTBT)) + goto bail; + lsb = IBA7220_IBC_HRTBT_SHIFT; + maskr = IBA7220_IBC_HRTBT_MASK; + break; + + case IPATH_IB_CFG_LWID_ENB: /* set allowed Link-width */ + /* + * As with speed, only write the actual register if + * the link is currently down, otherwise takes effect + * on next link change. + */ + dd->ipath_link_width_enabled = val; + if ((dd->ipath_flags & (IPATH_LINKDOWN|IPATH_LINKINIT)) != + IPATH_LINKDOWN) + goto bail; + /* + * We set the IPATH_IB_FORCE_NOTIFY bit so updown + * will get called because we want update + * link_width_active, and the change may not take + * effect for some time (if we are in POLL), so this + * flag will force the updown routine to be called + * on the next ibstatuschange down interrupt, even + * if it's not an down->up transition. + */ + val--; /* convert from IB to chip */ + maskr = IBA7220_IBC_WIDTH_MASK; + lsb = IBA7220_IBC_WIDTH_SHIFT; + setforce = 1; + dd->ipath_flags |= IPATH_IB_FORCE_NOTIFY; + break; + + case IPATH_IB_CFG_SPD_ENB: /* set allowed Link speeds */ + /* + * If we turn off IB1.2, need to preset SerDes defaults, + * but not right now. Set a flag for the next time + * we command the link down. As with width, only write the + * actual register if the link is currently down, otherwise + * takes effect on next link change. Since setting is being + * explictly requested (via MAD or sysfs), clear autoneg + * failure status if speed autoneg is enabled. + */ + dd->ipath_link_speed_enabled = val; + if (dd->ipath_ibcddrctrl & IBA7220_IBC_IBTA_1_2_MASK && + !(val & (val - 1))) + dd->ipath_presets_needed = 1; + if ((dd->ipath_flags & (IPATH_LINKDOWN|IPATH_LINKINIT)) != + IPATH_LINKDOWN) + goto bail; + /* + * We set the IPATH_IB_FORCE_NOTIFY bit so updown + * will get called because we want update + * link_speed_active, and the change may not take + * effect for some time (if we are in POLL), so this + * flag will force the updown routine to be called + * on the next ibstatuschange down interrupt, even + * if it's not an down->up transition. When setting + * speed autoneg, clear AUTONEG_FAILED. + */ + if (val == (IPATH_IB_SDR | IPATH_IB_DDR)) { + val = IBA7220_IBC_SPEED_AUTONEG_MASK | + IBA7220_IBC_IBTA_1_2_MASK; + dd->ipath_flags &= ~IPATH_IB_AUTONEG_FAILED; + } else + val = val == IPATH_IB_DDR ? IBA7220_IBC_SPEED_DDR + : IBA7220_IBC_SPEED_SDR; + maskr = IBA7220_IBC_SPEED_AUTONEG_MASK | + IBA7220_IBC_IBTA_1_2_MASK; + lsb = 0; /* speed bits are low bits */ + setforce = 1; + break; + + case IPATH_IB_CFG_RXPOL_ENB: /* set Auto-RX-polarity enable */ + lsb = IBA7220_IBC_RXPOL_SHIFT; + maskr = IBA7220_IBC_RXPOL_MASK; + break; + + case IPATH_IB_CFG_LREV_ENB: /* set Auto-Lane-reversal enable */ + lsb = IBA7220_IBC_LREV_SHIFT; + maskr = IBA7220_IBC_LREV_MASK; + break; + + default: + ret = -ENOTSUPP; + goto bail; + } + dd->ipath_ibcddrctrl &= ~(maskr << lsb); + dd->ipath_ibcddrctrl |= (((u64) val & maskr) << lsb); + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcddrctrl, + dd->ipath_ibcddrctrl); + if (setforce) + dd->ipath_flags |= IPATH_IB_FORCE_NOTIFY; +bail: + return ret; +} + +static void ipath_7220_read_counters(struct ipath_devdata *dd, + struct infinipath_counters *cntrs) +{ + u64 *counters = (u64 *) cntrs; + int i; + + for (i = 0; i < sizeof(*cntrs) / sizeof(u64); i++) + counters[i] = ipath_snap_cntr(dd, i); +} + +/* if we are using MSI, try to fallback to IntX */ +static int ipath_7220_intr_fallback(struct ipath_devdata *dd) +{ + if (dd->ipath_msi_lo) { + dev_info(&dd->pcidev->dev, "MSI interrupt not detected," + " trying IntX interrupts\n"); + ipath_7220_nomsi(dd); + ipath_enable_intx(dd->pcidev); + /* + * some newer kernels require free_irq before disable_msi, + * and irq can be changed during disable and intx enable + * and we need to therefore use the pcidev->irq value, + * not our saved MSI value. + */ + dd->ipath_irq = dd->pcidev->irq; + if (request_irq(dd->ipath_irq, ipath_intr, IRQF_SHARED, + IPATH_DRV_NAME, dd)) + ipath_dev_err(dd, + "Could not re-request_irq for IntX\n"); + return 1; + } + return 0; +} + +/* + * reset the XGXS (between serdes and IBC). Slightly less intrusive + * than resetting the IBC or external link state, and useful in some + * cases to cause some retraining. To do this right, we reset IBC + * as well. + */ +static void ipath_7220_xgxs_reset(struct ipath_devdata *dd) +{ + u64 val, prev_val; + + prev_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig); + val = prev_val | INFINIPATH_XGXS_RESET; + prev_val &= ~INFINIPATH_XGXS_RESET; /* be sure */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_control, + dd->ipath_control & ~INFINIPATH_C_LINKENABLE); + ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, prev_val); + ipath_write_kreg(dd, dd->ipath_kregs->kr_control, + dd->ipath_control); +} + + +/* Still needs cleanup, too much hardwired stuff */ +static void autoneg_send(struct ipath_devdata *dd, + u32 *hdr, u32 dcnt, u32 *data) +{ + int i; + u64 cnt; + u32 __iomem *piobuf; + u32 pnum; + + i = 0; + cnt = 7 + dcnt + 1; /* 7 dword header, dword data, icrc */ + while (!(piobuf = ipath_getpiobuf(dd, cnt, &pnum))) { + if (i++ > 15) { + ipath_dbg("Couldn't get pio buffer for send\n"); + return; + } + udelay(2); + } + if (dd->ipath_flags&IPATH_HAS_PBC_CNT) + cnt |= 0x80000000UL<<32; /* mark as VL15 */ + writeq(cnt, piobuf); + ipath_flush_wc(); + __iowrite32_copy(piobuf + 2, hdr, 7); + __iowrite32_copy(piobuf + 9, data, dcnt); + ipath_flush_wc(); +} + +/* + * _start packet gets sent twice at start, _done gets sent twice at end + */ +static void ipath_autoneg_send(struct ipath_devdata *dd, int which) +{ + static u32 swapped; + u32 dw, i, hcnt, dcnt, *data; + static u32 hdr[7] = { 0xf002ffff, 0x48ffff, 0x6400abba }; + static u32 madpayload_start[0x40] = { + 0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0, + 0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0x1, 0x1388, 0x15e, 0x1, /* rest 0's */ + }; + static u32 madpayload_done[0x40] = { + 0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0, + 0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0x40000001, 0x1388, 0x15e, /* rest 0's */ + }; + dcnt = sizeof(madpayload_start)/sizeof(madpayload_start[0]); + hcnt = sizeof(hdr)/sizeof(hdr[0]); + if (!swapped) { + /* for maintainability, do it at runtime */ + for (i = 0; i < hcnt; i++) { + dw = (__force u32) cpu_to_be32(hdr[i]); + hdr[i] = dw; + } + for (i = 0; i < dcnt; i++) { + dw = (__force u32) cpu_to_be32(madpayload_start[i]); + madpayload_start[i] = dw; + dw = (__force u32) cpu_to_be32(madpayload_done[i]); + madpayload_done[i] = dw; + } + swapped = 1; + } + + data = which ? madpayload_done : madpayload_start; + ipath_cdbg(PKT, "Sending %s special MADs\n", which?"done":"start"); + + autoneg_send(dd, hdr, dcnt, data); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + udelay(2); + autoneg_send(dd, hdr, dcnt, data); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + udelay(2); +} + + + +/* + * Do the absolute minimum to cause an IB speed change, and make it + * ready, but don't actually trigger the change. The caller will + * do that when ready (if link is in Polling training state, it will + * happen immediately, otherwise when link next goes down) + * + * This routine should only be used as part of the DDR autonegotation + * code for devices that are not compliant with IB 1.2 (or code that + * fixes things up for same). + * + * When link has gone down, and autoneg enabled, or autoneg has + * failed and we give up until next time we set both speeds, and + * then we want IBTA enabled as well as "use max enabled speed. + */ +static void set_speed_fast(struct ipath_devdata *dd, u32 speed) +{ + dd->ipath_ibcddrctrl &= ~(IBA7220_IBC_SPEED_AUTONEG_MASK | + IBA7220_IBC_IBTA_1_2_MASK | + (IBA7220_IBC_WIDTH_MASK << IBA7220_IBC_WIDTH_SHIFT)); + + if (speed == (IPATH_IB_SDR | IPATH_IB_DDR)) + dd->ipath_ibcddrctrl |= IBA7220_IBC_SPEED_AUTONEG_MASK | + IBA7220_IBC_IBTA_1_2_MASK; + else + dd->ipath_ibcddrctrl |= speed == IPATH_IB_DDR ? + IBA7220_IBC_SPEED_DDR : IBA7220_IBC_SPEED_SDR; + + /* + * Convert from IB-style 1 = 1x, 2 = 4x, 3 = auto + * to chip-centric 0 = 1x, 1 = 4x, 2 = auto + */ + dd->ipath_ibcddrctrl |= (u64)(dd->ipath_link_width_enabled - 1) << + IBA7220_IBC_WIDTH_SHIFT; + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcddrctrl, + dd->ipath_ibcddrctrl); + ipath_cdbg(VERBOSE, "setup for IB speed (%x) done\n", speed); +} + + +/* + * this routine is only used when we are not talking to another + * IB 1.2-compliant device that we think can do DDR. + * (This includes all existing switch chips as of Oct 2007.) + * 1.2-compliant devices go directly to DDR prior to reaching INIT + */ +static void try_auto_neg(struct ipath_devdata *dd) +{ + /* + * required for older non-IB1.2 DDR switches. Newer + * non-IB-compliant switches don't need it, but so far, + * aren't bothered by it either. "Magic constant" + */ + ipath_write_kreg(dd, IPATH_KREG_OFFSET(IBNCModeCtrl), + 0x3b9dc07); + dd->ipath_flags |= IPATH_IB_AUTONEG_INPROG; + ipath_autoneg_send(dd, 0); + set_speed_fast(dd, IPATH_IB_DDR); + ipath_toggle_rclkrls(dd); + /* 2 msec is minimum length of a poll cycle */ + schedule_delayed_work(&dd->ipath_autoneg_work, + msecs_to_jiffies(2)); +} + + +static int ipath_7220_ib_updown(struct ipath_devdata *dd, int ibup, u64 ibcs) +{ + int ret = 0; + u32 ltstate = ipath_ib_linkstate(dd, ibcs); + + dd->ipath_link_width_active = + ((ibcs >> IBA7220_IBCS_LINKWIDTH_SHIFT) & 1) ? + IB_WIDTH_4X : IB_WIDTH_1X; + dd->ipath_link_speed_active = + ((ibcs >> IBA7220_IBCS_LINKSPEED_SHIFT) & 1) ? + IPATH_IB_DDR : IPATH_IB_SDR; + + if (!ibup) { + /* + * when link goes down we don't want aeq running, so it + * won't't interfere with IBC training, etc., and we need + * to go back to the static SerDes preset values + */ + if (dd->ipath_x1_fix_tries && + ltstate <= INFINIPATH_IBCS_LT_STATE_SLEEPQUIET && + ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP) + dd->ipath_x1_fix_tries = 0; + if (!(dd->ipath_flags & (IPATH_IB_AUTONEG_FAILED | + IPATH_IB_AUTONEG_INPROG))) + set_speed_fast(dd, dd->ipath_link_speed_enabled); + if (!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG)) { + ipath_cdbg(VERBOSE, "Setting RXEQ defaults\n"); + ipath_sd7220_presets(dd); + } + /* this might better in ipath_sd7220_presets() */ + ipath_set_relock_poll(dd, ibup); + } else { + if (ipath_compat_ddr_negotiate && + !(dd->ipath_flags & (IPATH_IB_AUTONEG_FAILED | + IPATH_IB_AUTONEG_INPROG)) && + dd->ipath_link_speed_active == IPATH_IB_SDR && + (dd->ipath_link_speed_enabled & + (IPATH_IB_DDR | IPATH_IB_SDR)) == + (IPATH_IB_DDR | IPATH_IB_SDR) && + dd->ipath_autoneg_tries < IPATH_AUTONEG_TRIES) { + /* we are SDR, and DDR auto-negotiation enabled */ + ++dd->ipath_autoneg_tries; + ipath_dbg("DDR negotiation try, %u/%u\n", + dd->ipath_autoneg_tries, + IPATH_AUTONEG_TRIES); + try_auto_neg(dd); + ret = 1; /* no other IB status change processing */ + } else if ((dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) + && dd->ipath_link_speed_active == IPATH_IB_SDR) { + ipath_autoneg_send(dd, 1); + set_speed_fast(dd, IPATH_IB_DDR); + udelay(2); + ipath_toggle_rclkrls(dd); + ret = 1; /* no other IB status change processing */ + } else { + if ((dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) && + (dd->ipath_link_speed_active & IPATH_IB_DDR)) { + ipath_dbg("Got to INIT with DDR autoneg\n"); + dd->ipath_flags &= ~(IPATH_IB_AUTONEG_INPROG + | IPATH_IB_AUTONEG_FAILED); + dd->ipath_autoneg_tries = 0; + /* re-enable SDR, for next link down */ + set_speed_fast(dd, + dd->ipath_link_speed_enabled); + wake_up(&dd->ipath_autoneg_wait); + } else if (dd->ipath_flags & IPATH_IB_AUTONEG_FAILED) { + /* + * clear autoneg failure flag, and do setup + * so we'll try next time link goes down and + * back to INIT (possibly connected to different + * device). + */ + ipath_dbg("INIT %sDR after autoneg failure\n", + (dd->ipath_link_speed_active & + IPATH_IB_DDR) ? "D" : "S"); + dd->ipath_flags &= ~IPATH_IB_AUTONEG_FAILED; + dd->ipath_ibcddrctrl |= + IBA7220_IBC_IBTA_1_2_MASK; + ipath_write_kreg(dd, + IPATH_KREG_OFFSET(IBNCModeCtrl), 0); + } + } + /* + * if we are in 1X, and are in autoneg width, it + * could be due to an xgxs problem, so if we haven't + * already tried, try twice to get to 4X; if we + * tried, and couldn't, report it, since it will + * probably not be what is desired. + */ + if ((dd->ipath_link_width_enabled & (IB_WIDTH_1X | + IB_WIDTH_4X)) == (IB_WIDTH_1X | IB_WIDTH_4X) + && dd->ipath_link_width_active == IB_WIDTH_1X + && dd->ipath_x1_fix_tries < 3) { + if (++dd->ipath_x1_fix_tries == 3) + dev_info(&dd->pcidev->dev, + "IB link is in 1X mode\n"); + else { + ipath_cdbg(VERBOSE, "IB 1X in " + "auto-width, try %u to be " + "sure it's really 1X; " + "ltstate %u\n", + dd->ipath_x1_fix_tries, + ltstate); + dd->ipath_f_xgxs_reset(dd); + ret = 1; /* skip other processing */ + } + } + + if (!ret) { + dd->delay_mult = rate_to_delay + [(ibcs >> IBA7220_IBCS_LINKSPEED_SHIFT) & 1] + [(ibcs >> IBA7220_IBCS_LINKWIDTH_SHIFT) & 1]; + + ipath_set_relock_poll(dd, ibup); + } + } + + if (!ret) + ipath_setup_7220_setextled(dd, ipath_ib_linkstate(dd, ibcs), + ltstate); + return ret; +} + + +/* + * Handle the empirically determined mechanism for auto-negotiation + * of DDR speed with switches. + */ +static void autoneg_work(struct work_struct *work) +{ + struct ipath_devdata *dd; + u64 startms; + u32 lastlts, i; + + dd = container_of(work, struct ipath_devdata, + ipath_autoneg_work.work); + + startms = jiffies_to_msecs(jiffies); + + /* + * busy wait for this first part, it should be at most a + * few hundred usec, since we scheduled ourselves for 2msec. + */ + for (i = 0; i < 25; i++) { + lastlts = ipath_ib_linktrstate(dd, dd->ipath_lastibcstat); + if (lastlts == INFINIPATH_IBCS_LT_STATE_POLLQUIET) { + ipath_set_linkstate(dd, IPATH_IB_LINKDOWN_DISABLE); + break; + } + udelay(100); + } + + if (!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG)) + goto done; /* we got there early or told to stop */ + + /* we expect this to timeout */ + if (wait_event_timeout(dd->ipath_autoneg_wait, + !(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG), + msecs_to_jiffies(90))) + goto done; + + ipath_toggle_rclkrls(dd); + + /* we expect this to timeout */ + if (wait_event_timeout(dd->ipath_autoneg_wait, + !(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG), + msecs_to_jiffies(1700))) + goto done; + + set_speed_fast(dd, IPATH_IB_SDR); + ipath_toggle_rclkrls(dd); + + /* + * wait up to 250 msec for link to train and get to INIT at DDR; + * this should terminate early + */ + wait_event_timeout(dd->ipath_autoneg_wait, + !(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG), + msecs_to_jiffies(250)); +done: + if (dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) { + ipath_dbg("Did not get to DDR INIT (%x) after %Lu msecs\n", + ipath_ib_state(dd, dd->ipath_lastibcstat), + jiffies_to_msecs(jiffies)-startms); + dd->ipath_flags &= ~IPATH_IB_AUTONEG_INPROG; + if (dd->ipath_autoneg_tries == IPATH_AUTONEG_TRIES) { + dd->ipath_flags |= IPATH_IB_AUTONEG_FAILED; + ipath_dbg("Giving up on DDR until next IB " + "link Down\n"); + dd->ipath_autoneg_tries = 0; + } + set_speed_fast(dd, dd->ipath_link_speed_enabled); + } +} + + +/** + * ipath_init_iba7220_funcs - set up the chip-specific function pointers + * @dd: the infinipath device + * + * This is global, and is called directly at init to set up the + * chip-specific function pointers for later use. + */ +void ipath_init_iba7220_funcs(struct ipath_devdata *dd) +{ + dd->ipath_f_intrsetup = ipath_7220_intconfig; + dd->ipath_f_bus = ipath_setup_7220_config; + dd->ipath_f_reset = ipath_setup_7220_reset; + dd->ipath_f_get_boardname = ipath_7220_boardname; + dd->ipath_f_init_hwerrors = ipath_7220_init_hwerrors; + dd->ipath_f_early_init = ipath_7220_early_init; + dd->ipath_f_handle_hwerrors = ipath_7220_handle_hwerrors; + dd->ipath_f_quiet_serdes = ipath_7220_quiet_serdes; + dd->ipath_f_bringup_serdes = ipath_7220_bringup_serdes; + dd->ipath_f_clear_tids = ipath_7220_clear_tids; + dd->ipath_f_put_tid = ipath_7220_put_tid; + dd->ipath_f_cleanup = ipath_setup_7220_cleanup; + dd->ipath_f_setextled = ipath_setup_7220_setextled; + dd->ipath_f_get_base_info = ipath_7220_get_base_info; + dd->ipath_f_free_irq = ipath_7220_free_irq; + dd->ipath_f_tidtemplate = ipath_7220_tidtemplate; + dd->ipath_f_intr_fallback = ipath_7220_intr_fallback; + dd->ipath_f_xgxs_reset = ipath_7220_xgxs_reset; + dd->ipath_f_get_ib_cfg = ipath_7220_get_ib_cfg; + dd->ipath_f_set_ib_cfg = ipath_7220_set_ib_cfg; + dd->ipath_f_config_jint = ipath_7220_config_jint; + dd->ipath_f_config_ports = ipath_7220_config_ports; + dd->ipath_f_read_counters = ipath_7220_read_counters; + dd->ipath_f_get_msgheader = ipath_7220_get_msgheader; + dd->ipath_f_ib_updown = ipath_7220_ib_updown; + + /* initialize chip-specific variables */ + ipath_init_7220_variables(dd); +} diff --git a/drivers/infiniband/hw/ipath/ipath_init_chip.c b/drivers/infiniband/hw/ipath/ipath_init_chip.c index 4471674975c..27dd8947666 100644 --- a/drivers/infiniband/hw/ipath/ipath_init_chip.c +++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -155,24 +155,13 @@ static int bringup_link(struct ipath_devdata *dd) dd->ipath_control); /* - * Note that prior to try 14 or 15 of IB, the credit scaling - * wasn't working, because it was swapped for writes with the - * 1 bit default linkstate field + * set initial max size pkt IBC will send, including ICRC; it's the + * PIO buffer size in dwords, less 1; also see ipath_set_mtu() */ + val = (dd->ipath_ibmaxlen >> 2) + 1; + ibc = val << dd->ibcc_mpl_shift; - /* ignore pbc and align word */ - val = dd->ipath_piosize2k - 2 * sizeof(u32); - /* - * for ICRC, which we only send in diag test pkt mode, and we - * don't need to worry about that for mtu - */ - val += 1; - /* - * Set the IBC maxpktlength to the size of our pio buffers the - * maxpktlength is in words. This is *not* the IB data MTU. - */ - ibc = (val / sizeof(u32)) << INFINIPATH_IBCC_MAXPKTLEN_SHIFT; - /* in KB */ + /* flowcontrolwatermark is in units of KBytes */ ibc |= 0x5ULL << INFINIPATH_IBCC_FLOWCTRLWATERMARK_SHIFT; /* * How often flowctrl sent. More or less in usecs; balance against @@ -191,10 +180,13 @@ static int bringup_link(struct ipath_devdata *dd) /* * Want to start out with both LINKCMD and LINKINITCMD in NOP * (0 and 0). Don't put linkinitcmd in ipath_ibcctrl, want that - * to stay a NOP + * to stay a NOP. Flag that we are disabled, for the (unlikely) + * case that some recovery path is trying to bring the link up + * before we are ready. */ ibc |= INFINIPATH_IBCC_LINKINITCMD_DISABLE << INFINIPATH_IBCC_LINKINITCMD_SHIFT; + dd->ipath_flags |= IPATH_IB_LINK_DISABLED; ipath_cdbg(VERBOSE, "Writing 0x%llx to ibcctrl\n", (unsigned long long) ibc); ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, ibc); @@ -227,17 +219,26 @@ static struct ipath_portdata *create_portdata0(struct ipath_devdata *dd) pd->port_cnt = 1; /* The port 0 pkey table is used by the layer interface. */ pd->port_pkeys[0] = IPATH_DEFAULT_P_KEY; + pd->port_seq_cnt = 1; } return pd; } -static int init_chip_first(struct ipath_devdata *dd, - struct ipath_portdata **pdp) +static int init_chip_first(struct ipath_devdata *dd) { - struct ipath_portdata *pd = NULL; + struct ipath_portdata *pd; int ret = 0; u64 val; + spin_lock_init(&dd->ipath_kernel_tid_lock); + spin_lock_init(&dd->ipath_user_tid_lock); + spin_lock_init(&dd->ipath_sendctrl_lock); + spin_lock_init(&dd->ipath_sdma_lock); + spin_lock_init(&dd->ipath_gpio_lock); + spin_lock_init(&dd->ipath_eep_st_lock); + spin_lock_init(&dd->ipath_sdepb_lock); + mutex_init(&dd->ipath_eep_lock); + /* * skip cfgports stuff because we are not allocating memory, * and we don't want problems if the portcnt changed due to @@ -250,12 +251,14 @@ static int init_chip_first(struct ipath_devdata *dd, else if (ipath_cfgports <= dd->ipath_portcnt) { dd->ipath_cfgports = ipath_cfgports; ipath_dbg("Configured to use %u ports out of %u in chip\n", - dd->ipath_cfgports, dd->ipath_portcnt); + dd->ipath_cfgports, ipath_read_kreg32(dd, + dd->ipath_kregs->kr_portcnt)); } else { dd->ipath_cfgports = dd->ipath_portcnt; ipath_dbg("Tried to configured to use %u ports; chip " "only supports %u\n", ipath_cfgports, - dd->ipath_portcnt); + ipath_read_kreg32(dd, + dd->ipath_kregs->kr_portcnt)); } /* * Allocate full portcnt array, rather than just cfgports, because @@ -295,12 +298,9 @@ static int init_chip_first(struct ipath_devdata *dd, val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendpiosize); dd->ipath_piosize2k = val & ~0U; dd->ipath_piosize4k = val >> 32; - /* - * Note: the chips support a maximum MTU of 4096, but the driver - * hasn't implemented this feature yet, so set the initial value - * to 2048. - */ - dd->ipath_ibmtu = 2048; + if (dd->ipath_piosize4k == 0 && ipath_mtu4096) + ipath_mtu4096 = 0; /* 4KB not supported by this chip */ + dd->ipath_ibmtu = ipath_mtu4096 ? 4096 : 2048; val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendpiobufcnt); dd->ipath_piobcnt2k = val & ~0U; dd->ipath_piobcnt4k = val >> 32; @@ -328,43 +328,46 @@ static int init_chip_first(struct ipath_devdata *dd, else ipath_dbg("%u 2k piobufs @ %p\n", dd->ipath_piobcnt2k, dd->ipath_pio2kbase); - spin_lock_init(&dd->ipath_tid_lock); - spin_lock_init(&dd->ipath_sendctrl_lock); - spin_lock_init(&dd->ipath_gpio_lock); - spin_lock_init(&dd->ipath_eep_st_lock); - mutex_init(&dd->ipath_eep_lock); - done: - *pdp = pd; return ret; } /** * init_chip_reset - re-initialize after a reset, or enable * @dd: the infinipath device - * @pdp: output for port data * * sanity check at least some of the values after reset, and * ensure no receive or transmit (explictly, in case reset * failed */ -static int init_chip_reset(struct ipath_devdata *dd, - struct ipath_portdata **pdp) +static int init_chip_reset(struct ipath_devdata *dd) { u32 rtmp; + int i; + unsigned long flags; + + /* + * ensure chip does no sends or receives, tail updates, or + * pioavail updates while we re-initialize + */ + dd->ipath_rcvctrl &= ~(1ULL << dd->ipath_r_tailupd_shift); + for (i = 0; i < dd->ipath_portcnt; i++) { + clear_bit(dd->ipath_r_portenable_shift + i, + &dd->ipath_rcvctrl); + clear_bit(dd->ipath_r_intravail_shift + i, + &dd->ipath_rcvctrl); + } + ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, + dd->ipath_rcvctrl); - *pdp = dd->ipath_pd[0]; - /* ensure chip does no sends or receives while we re-initialize */ - dd->ipath_control = dd->ipath_sendctrl = dd->ipath_rcvctrl = 0U; - ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, dd->ipath_rcvctrl); + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl = 0U; /* no sdma, etc */ ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); - ipath_write_kreg(dd, dd->ipath_kregs->kr_control, dd->ipath_control); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + + ipath_write_kreg(dd, dd->ipath_kregs->kr_control, 0ULL); - rtmp = ipath_read_kreg32(dd, dd->ipath_kregs->kr_portcnt); - if (dd->ipath_portcnt != rtmp) - dev_info(&dd->pcidev->dev, "portcnt was %u before " - "reset, now %u, using original\n", - dd->ipath_portcnt, rtmp); rtmp = ipath_read_kreg32(dd, dd->ipath_kregs->kr_rcvtidcnt); if (rtmp != dd->ipath_rcvtidcnt) dev_info(&dd->pcidev->dev, "tidcnt was %u before " @@ -467,10 +470,10 @@ static void init_shadow_tids(struct ipath_devdata *dd) dd->ipath_physshadow = addrs; } -static void enable_chip(struct ipath_devdata *dd, - struct ipath_portdata *pd, int reinit) +static void enable_chip(struct ipath_devdata *dd, int reinit) { u32 val; + u64 rcvmask; unsigned long flags; int i; @@ -484,17 +487,28 @@ static void enable_chip(struct ipath_devdata *dd, /* Enable PIO send, and update of PIOavail regs to memory. */ dd->ipath_sendctrl = INFINIPATH_S_PIOENABLE | INFINIPATH_S_PIOBUFAVAILUPD; + + /* + * Set the PIO avail update threshold to host memory + * on chips that support it. + */ + if (dd->ipath_pioupd_thresh) + dd->ipath_sendctrl |= dd->ipath_pioupd_thresh + << INFINIPATH_S_UPDTHRESH_SHIFT; ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); /* - * enable port 0 receive, and receive interrupt. other ports - * done as user opens and inits them. + * Enable kernel ports' receive and receive interrupt. + * Other ports done as user opens and inits them. */ - dd->ipath_rcvctrl = (1ULL << dd->ipath_r_tailupd_shift) | - (1ULL << dd->ipath_r_portenable_shift) | - (1ULL << dd->ipath_r_intravail_shift); + rcvmask = 1ULL; + dd->ipath_rcvctrl |= (rcvmask << dd->ipath_r_portenable_shift) | + (rcvmask << dd->ipath_r_intravail_shift); + if (!(dd->ipath_flags & IPATH_NODMA_RTAIL)) + dd->ipath_rcvctrl |= (1ULL << dd->ipath_r_tailupd_shift); + ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, dd->ipath_rcvctrl); @@ -505,16 +519,16 @@ static void enable_chip(struct ipath_devdata *dd, dd->ipath_flags |= IPATH_INITTED; /* - * init our shadow copies of head from tail values, and write - * head values to match. + * Init our shadow copies of head from tail values, + * and write head values to match. */ val = ipath_read_ureg32(dd, ur_rcvegrindextail, 0); - (void)ipath_write_ureg(dd, ur_rcvegrindexhead, val, 0); + ipath_write_ureg(dd, ur_rcvegrindexhead, val, 0); /* Initialize so we interrupt on next packet received */ - (void)ipath_write_ureg(dd, ur_rcvhdrhead, - dd->ipath_rhdrhead_intr_off | - dd->ipath_pd[0]->port_head, 0); + ipath_write_ureg(dd, ur_rcvhdrhead, + dd->ipath_rhdrhead_intr_off | + dd->ipath_pd[0]->port_head, 0); /* * by now pioavail updates to memory should have occurred, so @@ -523,25 +537,26 @@ static void enable_chip(struct ipath_devdata *dd, * initial values of the generation bit correct. */ for (i = 0; i < dd->ipath_pioavregs; i++) { - __le64 val; + __le64 pioavail; /* * Chip Errata bug 6641; even and odd qwords>3 are swapped. */ if (i > 3 && (dd->ipath_flags & IPATH_SWAP_PIOBUFS)) - val = dd->ipath_pioavailregs_dma[i ^ 1]; + pioavail = dd->ipath_pioavailregs_dma[i ^ 1]; else - val = dd->ipath_pioavailregs_dma[i]; - dd->ipath_pioavailshadow[i] = le64_to_cpu(val); + pioavail = dd->ipath_pioavailregs_dma[i]; + dd->ipath_pioavailshadow[i] = le64_to_cpu(pioavail) | + (~dd->ipath_pioavailkernel[i] << + INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT); } /* can get counters, stats, etc. */ dd->ipath_flags |= IPATH_PRESENT; } -static int init_housekeeping(struct ipath_devdata *dd, - struct ipath_portdata **pdp, int reinit) +static int init_housekeeping(struct ipath_devdata *dd, int reinit) { - char boardn[32]; + char boardn[40]; int ret = 0; /* @@ -600,18 +615,9 @@ static int init_housekeeping(struct ipath_devdata *dd, ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear, INFINIPATH_E_RESET); - if (reinit) - ret = init_chip_reset(dd, pdp); - else - ret = init_chip_first(dd, pdp); - - if (ret) - goto done; - - ipath_cdbg(VERBOSE, "Revision %llx (PCI %x), %u ports, %u tids, " - "%u egrtids\n", (unsigned long long) dd->ipath_revision, - dd->ipath_pcirev, dd->ipath_portcnt, dd->ipath_rcvtidcnt, - dd->ipath_rcvegrcnt); + ipath_cdbg(VERBOSE, "Revision %llx (PCI %x)\n", + (unsigned long long) dd->ipath_revision, + dd->ipath_pcirev); if (((dd->ipath_revision >> INFINIPATH_R_SOFTWARE_SHIFT) & INFINIPATH_R_SOFTWARE_MASK) != IPATH_CHIP_SWVERSION) { @@ -650,10 +656,39 @@ static int init_housekeeping(struct ipath_devdata *dd, ipath_dbg("%s", dd->ipath_boardversion); + if (ret) + goto done; + + if (reinit) + ret = init_chip_reset(dd); + else + ret = init_chip_first(dd); + done: return ret; } +static void verify_interrupt(unsigned long opaque) +{ + struct ipath_devdata *dd = (struct ipath_devdata *) opaque; + + if (!dd) + return; /* being torn down */ + + /* + * If we don't have any interrupts, let the user know and + * don't bother checking again. + */ + if (dd->ipath_int_counter == 0) { + if (!dd->ipath_f_intr_fallback(dd)) + dev_err(&dd->pcidev->dev, "No interrupts detected, " + "not usable.\n"); + else /* re-arm the timer to see if fallback works */ + mod_timer(&dd->ipath_intrchk_timer, jiffies + HZ/2); + } else + ipath_cdbg(VERBOSE, "%u interrupts at timer check\n", + dd->ipath_int_counter); +} /** * ipath_init_chip - do the actual initialization sequence on the chip @@ -676,11 +711,11 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) u32 val32, kpiobufs; u32 piobufs, uports; u64 val; - struct ipath_portdata *pd = NULL; /* keep gcc4 happy */ + struct ipath_portdata *pd; gfp_t gfp_flags = GFP_USER | __GFP_COMP; unsigned long flags; - ret = init_housekeeping(dd, &pd, reinit); + ret = init_housekeeping(dd, reinit); if (ret) goto done; @@ -700,7 +735,7 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) * we now use routines that backend onto __get_free_pages, the * rest would be wasted. */ - dd->ipath_rcvhdrcnt = dd->ipath_rcvegrcnt; + dd->ipath_rcvhdrcnt = max(dd->ipath_p0_rcvegrcnt, dd->ipath_rcvegrcnt); ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvhdrcnt, dd->ipath_rcvhdrcnt); @@ -731,8 +766,8 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) if (kpiobufs + (uports * IPATH_MIN_USER_PORT_BUFCNT) > piobufs) { int i = (int) piobufs - (int) (uports * IPATH_MIN_USER_PORT_BUFCNT); - if (i < 0) - i = 0; + if (i < 1) + i = 1; dev_info(&dd->pcidev->dev, "Allocating %d PIO bufs of " "%d for kernel leaves too few for %d user ports " "(%d each); using %u\n", kpiobufs, @@ -751,24 +786,40 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) ipath_dbg("allocating %u pbufs/port leaves %u unused, " "add to kernel\n", dd->ipath_pbufsport, val32); dd->ipath_lastport_piobuf -= val32; + kpiobufs += val32; ipath_dbg("%u pbufs/port leaves %u unused, add to kernel\n", dd->ipath_pbufsport, val32); } - dd->ipath_lastpioindex = dd->ipath_lastport_piobuf; + dd->ipath_lastpioindex = 0; + dd->ipath_lastpioindexl = dd->ipath_piobcnt2k; + ipath_chg_pioavailkernel(dd, 0, piobufs, 1); ipath_cdbg(VERBOSE, "%d PIO bufs for kernel out of %d total %u " "each for %u user ports\n", kpiobufs, piobufs, dd->ipath_pbufsport, uports); + if (dd->ipath_pioupd_thresh) { + if (dd->ipath_pbufsport < dd->ipath_pioupd_thresh) + dd->ipath_pioupd_thresh = dd->ipath_pbufsport; + if (kpiobufs < dd->ipath_pioupd_thresh) + dd->ipath_pioupd_thresh = kpiobufs; + } + + ret = dd->ipath_f_early_init(dd); + if (ret) { + ipath_dev_err(dd, "Early initialization failure\n"); + goto done; + } - dd->ipath_f_early_init(dd); /* - * cancel any possible active sends from early driver load. + * Cancel any possible active sends from early driver load. * Follows early_init because some chips have to initialize * PIO buffers in early_init to avoid false parity errors. */ ipath_cancel_sends(dd, 0); - /* early_init sets rcvhdrentsize and rcvhdrsize, so this must be - * done after early_init */ + /* + * Early_init sets rcvhdrentsize and rcvhdrsize, so this must be + * done after early_init. + */ dd->ipath_hdrqlast = dd->ipath_rcvhdrentsize * (dd->ipath_rcvhdrcnt - 1); ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvhdrentsize, @@ -783,8 +834,8 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) goto done; } - (void)ipath_write_kreg(dd, dd->ipath_kregs->kr_sendpioavailaddr, - dd->ipath_pioavailregs_phys); + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendpioavailaddr, + dd->ipath_pioavailregs_phys); /* * this is to detect s/w errors, which the h/w works around by * ignoring the low 6 bits of address, if it wasn't aligned. @@ -843,58 +894,65 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) /* enable errors that are masked, at least this first time. */ ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, ~dd->ipath_maskederrs); - dd->ipath_errormask = ipath_read_kreg64(dd, - dd->ipath_kregs->kr_errormask); + dd->ipath_maskederrs = 0; /* don't re-enable ignored in timer */ + dd->ipath_errormask = + ipath_read_kreg64(dd, dd->ipath_kregs->kr_errormask); /* clear any interrupts up to this point (ints still not enabled) */ ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, -1LL); + dd->ipath_f_tidtemplate(dd); + /* * Set up the port 0 (kernel) rcvhdr q and egr TIDs. If doing * re-init, the simplest way to handle this is to free * existing, and re-allocate. * Need to re-create rest of port 0 portdata as well. */ + pd = dd->ipath_pd[0]; if (reinit) { - /* Alloc and init new ipath_portdata for port0, + struct ipath_portdata *npd; + + /* + * Alloc and init new ipath_portdata for port0, * Then free old pd. Could lead to fragmentation, but also * makes later support for hot-swap easier. */ - struct ipath_portdata *npd; npd = create_portdata0(dd); if (npd) { ipath_free_pddata(dd, pd); - dd->ipath_pd[0] = pd = npd; + dd->ipath_pd[0] = npd; + pd = npd; } else { - ipath_dev_err(dd, "Unable to allocate portdata for" - " port 0, failing\n"); + ipath_dev_err(dd, "Unable to allocate portdata" + " for port 0, failing\n"); ret = -ENOMEM; goto done; } } - dd->ipath_f_tidtemplate(dd); ret = ipath_create_rcvhdrq(dd, pd); - if (!ret) { - dd->ipath_hdrqtailptr = - (volatile __le64 *)pd->port_rcvhdrtail_kvaddr; + if (!ret) ret = create_port0_egr(dd); - } - if (ret) - ipath_dev_err(dd, "failed to allocate port 0 (kernel) " + if (ret) { + ipath_dev_err(dd, "failed to allocate kernel port's " "rcvhdrq and/or egr bufs\n"); + goto done; + } else - enable_chip(dd, pd, reinit); - + enable_chip(dd, reinit); - if (!ret && !reinit) { - /* used when we close a port, for DMA already in flight at close */ + if (!reinit) { + /* + * Used when we close a port, for DMA already in flight + * at close. + */ dd->ipath_dummy_hdrq = dma_alloc_coherent( - &dd->pcidev->dev, pd->port_rcvhdrq_size, + &dd->pcidev->dev, dd->ipath_pd[0]->port_rcvhdrq_size, &dd->ipath_dummy_hdrq_phys, gfp_flags); - if (!dd->ipath_dummy_hdrq ) { + if (!dd->ipath_dummy_hdrq) { dev_info(&dd->pcidev->dev, "Couldn't allocate 0x%lx bytes for dummy hdrq\n", - pd->port_rcvhdrq_size); + dd->ipath_pd[0]->port_rcvhdrq_size); /* fallback to just 0'ing */ dd->ipath_dummy_hdrq_phys = 0UL; } @@ -906,7 +964,7 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) */ ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, 0ULL); - if(!dd->ipath_stats_timer_active) { + if (!dd->ipath_stats_timer_active) { /* * first init, or after an admin disable/enable * set up stats retrieval timer, even if we had errors @@ -922,6 +980,16 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit) dd->ipath_stats_timer_active = 1; } + /* Set up SendDMA if chip supports it */ + if (dd->ipath_flags & IPATH_HAS_SEND_DMA) + ret = setup_sdma(dd); + + /* Set up HoL state */ + init_timer(&dd->ipath_hol_timer); + dd->ipath_hol_timer.function = ipath_hol_event; + dd->ipath_hol_timer.data = (unsigned long)dd; + dd->ipath_hol_state = IPATH_HOL_UP; + done: if (!ret) { *dd->ipath_statusp |= IPATH_STATUS_CHIP_PRESENT; @@ -934,6 +1002,20 @@ done: 0ULL); /* chip is usable; mark it as initialized */ *dd->ipath_statusp |= IPATH_STATUS_INITTED; + + /* + * setup to verify we get an interrupt, and fallback + * to an alternate if necessary and possible + */ + if (!reinit) { + init_timer(&dd->ipath_intrchk_timer); + dd->ipath_intrchk_timer.function = + verify_interrupt; + dd->ipath_intrchk_timer.data = + (unsigned long) dd; + } + dd->ipath_intrchk_timer.expires = jiffies + HZ/2; + add_timer(&dd->ipath_intrchk_timer); } else ipath_dev_err(dd, "No interrupts enabled, couldn't " "setup interrupt address\n"); diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c index 92e58c92152..1b58f4737c7 100644 --- a/drivers/infiniband/hw/ipath/ipath_intr.c +++ b/drivers/infiniband/hw/ipath/ipath_intr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -32,6 +32,7 @@ */ #include <linux/pci.h> +#include <linux/delay.h> #include "ipath_kernel.h" #include "ipath_verbs.h" @@ -59,9 +60,11 @@ static void ipath_clrpiobuf(struct ipath_devdata *dd, u32 pnum) dev_info(&dd->pcidev->dev, "Rewrite PIO buffer %u, to recover from parity error\n", pnum); - *pbuf = dwcnt+1; /* no flush required, since already in freeze */ - while(--dwcnt) - *pbuf++ = 0; + + /* no flush required, since already in freeze */ + writel(dwcnt + 1, pbuf); + while (--dwcnt) + writel(0, pbuf++); } /* @@ -70,7 +73,7 @@ static void ipath_clrpiobuf(struct ipath_devdata *dd, u32 pnum) * If rewrite is true, and bits are set in the sendbufferror registers, * we'll write to the buffer, for error recovery on parity errors. */ -static void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite) +void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite) { u32 piobcnt; unsigned long sbuf[4]; @@ -84,12 +87,14 @@ static void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite) dd, dd->ipath_kregs->kr_sendbuffererror); sbuf[1] = ipath_read_kreg64( dd, dd->ipath_kregs->kr_sendbuffererror + 1); - if (piobcnt > 128) { + if (piobcnt > 128) sbuf[2] = ipath_read_kreg64( dd, dd->ipath_kregs->kr_sendbuffererror + 2); + if (piobcnt > 192) sbuf[3] = ipath_read_kreg64( dd, dd->ipath_kregs->kr_sendbuffererror + 3); - } + else + sbuf[3] = 0; if (sbuf[0] || sbuf[1] || (piobcnt > 128 && (sbuf[2] || sbuf[3]))) { int i; @@ -254,24 +259,20 @@ void ipath_format_hwerrors(u64 hwerrs, } /* return the strings for the most common link states */ -static char *ib_linkstate(u32 linkstate) +static char *ib_linkstate(struct ipath_devdata *dd, u64 ibcs) { char *ret; + u32 state; - switch (linkstate) { - case IPATH_IBSTATE_INIT: + state = ipath_ib_state(dd, ibcs); + if (state == dd->ib_init) ret = "Init"; - break; - case IPATH_IBSTATE_ARM: + else if (state == dd->ib_arm) ret = "Arm"; - break; - case IPATH_IBSTATE_ACTIVE: + else if (state == dd->ib_active) ret = "Active"; - break; - default: + else ret = "Down"; - } - return ret; } @@ -286,103 +287,172 @@ void signal_ib_event(struct ipath_devdata *dd, enum ib_event_type ev) } static void handle_e_ibstatuschanged(struct ipath_devdata *dd, - ipath_err_t errs, int noprint) + ipath_err_t errs) { - u64 val; - u32 ltstate, lstate; + u32 ltstate, lstate, ibstate, lastlstate; + u32 init = dd->ib_init; + u32 arm = dd->ib_arm; + u32 active = dd->ib_active; + const u64 ibcs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); + + lstate = ipath_ib_linkstate(dd, ibcs); /* linkstate */ + ibstate = ipath_ib_state(dd, ibcs); + /* linkstate at last interrupt */ + lastlstate = ipath_ib_linkstate(dd, dd->ipath_lastibcstat); + ltstate = ipath_ib_linktrstate(dd, ibcs); /* linktrainingtate */ /* - * even if diags are enabled, we want to notice LINKINIT, etc. - * We just don't want to change the LED state, or - * dd->ipath_kregs->kr_ibcctrl + * Since going into a recovery state causes the link state to go + * down and since recovery is transitory, it is better if we "miss" + * ever seeing the link training state go into recovery (i.e., + * ignore this transition for link state special handling purposes) + * without even updating ipath_lastibcstat. */ - val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); - lstate = val & IPATH_IBSTATE_MASK; + if ((ltstate == INFINIPATH_IBCS_LT_STATE_RECOVERRETRAIN) || + (ltstate == INFINIPATH_IBCS_LT_STATE_RECOVERWAITRMT) || + (ltstate == INFINIPATH_IBCS_LT_STATE_RECOVERIDLE)) + goto done; /* - * this is confusing enough when it happens that I want to always put it - * on the console and in the logs. If it was a requested state change, - * we'll have already cleared the flags, so we won't print this warning + * if linkstate transitions into INIT from any of the various down + * states, or if it transitions from any of the up (INIT or better) + * states into any of the down states (except link recovery), then + * call the chip-specific code to take appropriate actions. */ - if ((lstate != IPATH_IBSTATE_ARM && lstate != IPATH_IBSTATE_ACTIVE) - && (dd->ipath_flags & (IPATH_LINKARMED | IPATH_LINKACTIVE))) { - dev_info(&dd->pcidev->dev, "Link state changed from %s to %s\n", - (dd->ipath_flags & IPATH_LINKARMED) ? "ARM" : "ACTIVE", - ib_linkstate(lstate)); - /* - * Flush all queued sends when link went to DOWN or INIT, - * to be sure that they don't block SMA and other MAD packets - */ - ipath_cancel_sends(dd, 1); - } - else if (lstate == IPATH_IBSTATE_INIT || lstate == IPATH_IBSTATE_ARM || - lstate == IPATH_IBSTATE_ACTIVE) { - /* - * only print at SMA if there is a change, debug if not - * (sometimes we want to know that, usually not). - */ - if (lstate == ((unsigned) dd->ipath_lastibcstat - & IPATH_IBSTATE_MASK)) { - ipath_dbg("Status change intr but no change (%s)\n", - ib_linkstate(lstate)); + if (lstate >= INFINIPATH_IBCS_L_STATE_INIT && + lastlstate == INFINIPATH_IBCS_L_STATE_DOWN) { + /* transitioned to UP */ + if (dd->ipath_f_ib_updown(dd, 1, ibcs)) { + /* link came up, so we must no longer be disabled */ + dd->ipath_flags &= ~IPATH_IB_LINK_DISABLED; + ipath_cdbg(LINKVERB, "LinkUp handled, skipped\n"); + goto skip_ibchange; /* chip-code handled */ + } + } else if ((lastlstate >= INFINIPATH_IBCS_L_STATE_INIT || + (dd->ipath_flags & IPATH_IB_FORCE_NOTIFY)) && + ltstate <= INFINIPATH_IBCS_LT_STATE_CFGWAITRMT && + ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP) { + int handled; + handled = dd->ipath_f_ib_updown(dd, 0, ibcs); + dd->ipath_flags &= ~IPATH_IB_FORCE_NOTIFY; + if (handled) { + ipath_cdbg(LINKVERB, "LinkDown handled, skipped\n"); + goto skip_ibchange; /* chip-code handled */ } - else - ipath_cdbg(VERBOSE, "Unit %u link state %s, last " - "was %s\n", dd->ipath_unit, - ib_linkstate(lstate), - ib_linkstate((unsigned) - dd->ipath_lastibcstat - & IPATH_IBSTATE_MASK)); } - else { - lstate = dd->ipath_lastibcstat & IPATH_IBSTATE_MASK; - if (lstate == IPATH_IBSTATE_INIT || - lstate == IPATH_IBSTATE_ARM || - lstate == IPATH_IBSTATE_ACTIVE) - ipath_cdbg(VERBOSE, "Unit %u link state down" - " (state 0x%x), from %s\n", - dd->ipath_unit, - (u32)val & IPATH_IBSTATE_MASK, - ib_linkstate(lstate)); - else - ipath_cdbg(VERBOSE, "Unit %u link state changed " - "to 0x%x from down (%x)\n", - dd->ipath_unit, (u32) val, lstate); + + /* + * Significant enough to always print and get into logs, if it was + * unexpected. If it was a requested state change, we'll have + * already cleared the flags, so we won't print this warning + */ + if ((ibstate != arm && ibstate != active) && + (dd->ipath_flags & (IPATH_LINKARMED | IPATH_LINKACTIVE))) { + dev_info(&dd->pcidev->dev, "Link state changed from %s " + "to %s\n", (dd->ipath_flags & IPATH_LINKARMED) ? + "ARM" : "ACTIVE", ib_linkstate(dd, ibcs)); } - ltstate = (val >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) & - INFINIPATH_IBCS_LINKTRAININGSTATE_MASK; - lstate = (val >> INFINIPATH_IBCS_LINKSTATE_SHIFT) & - INFINIPATH_IBCS_LINKSTATE_MASK; if (ltstate == INFINIPATH_IBCS_LT_STATE_POLLACTIVE || ltstate == INFINIPATH_IBCS_LT_STATE_POLLQUIET) { - u32 last_ltstate; - + u32 lastlts; + lastlts = ipath_ib_linktrstate(dd, dd->ipath_lastibcstat); /* - * Ignore cycling back and forth from Polling.Active - * to Polling.Quiet while waiting for the other end of - * the link to come up. We will cycle back and forth - * between them if no cable is plugged in, - * the other device is powered off or disabled, etc. + * Ignore cycling back and forth from Polling.Active to + * Polling.Quiet while waiting for the other end of the link + * to come up, except to try and decide if we are connected + * to a live IB device or not. We will cycle back and + * forth between them if no cable is plugged in, the other + * device is powered off or disabled, etc. */ - last_ltstate = (dd->ipath_lastibcstat >> - INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) - & INFINIPATH_IBCS_LINKTRAININGSTATE_MASK; - if (last_ltstate == INFINIPATH_IBCS_LT_STATE_POLLACTIVE - || last_ltstate == - INFINIPATH_IBCS_LT_STATE_POLLQUIET) { - if (dd->ipath_ibpollcnt > 40) { + if (lastlts == INFINIPATH_IBCS_LT_STATE_POLLACTIVE || + lastlts == INFINIPATH_IBCS_LT_STATE_POLLQUIET) { + if (!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) && + (++dd->ipath_ibpollcnt == 40)) { dd->ipath_flags |= IPATH_NOCABLE; *dd->ipath_statusp |= IPATH_STATUS_IB_NOCABLE; - } else - dd->ipath_ibpollcnt++; + ipath_cdbg(LINKVERB, "Set NOCABLE\n"); + } + ipath_cdbg(LINKVERB, "POLL change to %s (%x)\n", + ipath_ibcstatus_str[ltstate], ibstate); goto skip_ibchange; } } - dd->ipath_ibpollcnt = 0; /* some state other than 2 or 3 */ + + dd->ipath_ibpollcnt = 0; /* not poll*, now */ ipath_stats.sps_iblink++; - if (ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP) { + + if (ibstate != init && dd->ipath_lastlinkrecov && ipath_linkrecovery) { + u64 linkrecov; + linkrecov = ipath_snap_cntr(dd, + dd->ipath_cregs->cr_iblinkerrrecovcnt); + if (linkrecov != dd->ipath_lastlinkrecov) { + ipath_dbg("IB linkrecov up %Lx (%s %s) recov %Lu\n", + ibcs, ib_linkstate(dd, ibcs), + ipath_ibcstatus_str[ltstate], + linkrecov); + /* and no more until active again */ + dd->ipath_lastlinkrecov = 0; + ipath_set_linkstate(dd, IPATH_IB_LINKDOWN); + goto skip_ibchange; + } + } + + if (ibstate == init || ibstate == arm || ibstate == active) { + *dd->ipath_statusp &= ~IPATH_STATUS_IB_NOCABLE; + if (ibstate == init || ibstate == arm) { + *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY; + if (dd->ipath_flags & IPATH_LINKACTIVE) + signal_ib_event(dd, IB_EVENT_PORT_ERR); + } + if (ibstate == arm) { + dd->ipath_flags |= IPATH_LINKARMED; + dd->ipath_flags &= ~(IPATH_LINKUNK | + IPATH_LINKINIT | IPATH_LINKDOWN | + IPATH_LINKACTIVE | IPATH_NOCABLE); + ipath_hol_down(dd); + } else if (ibstate == init) { + /* + * set INIT and DOWN. Down is checked by + * most of the other code, but INIT is + * useful to know in a few places. + */ + dd->ipath_flags |= IPATH_LINKINIT | + IPATH_LINKDOWN; + dd->ipath_flags &= ~(IPATH_LINKUNK | + IPATH_LINKARMED | IPATH_LINKACTIVE | + IPATH_NOCABLE); + ipath_hol_down(dd); + } else { /* active */ + dd->ipath_lastlinkrecov = ipath_snap_cntr(dd, + dd->ipath_cregs->cr_iblinkerrrecovcnt); + *dd->ipath_statusp |= + IPATH_STATUS_IB_READY | IPATH_STATUS_IB_CONF; + dd->ipath_flags |= IPATH_LINKACTIVE; + dd->ipath_flags &= ~(IPATH_LINKUNK | IPATH_LINKINIT + | IPATH_LINKDOWN | IPATH_LINKARMED | + IPATH_NOCABLE); + if (dd->ipath_flags & IPATH_HAS_SEND_DMA) + ipath_restart_sdma(dd); + signal_ib_event(dd, IB_EVENT_PORT_ACTIVE); + /* LED active not handled in chip _f_updown */ + dd->ipath_f_setextled(dd, lstate, ltstate); + ipath_hol_up(dd); + } + + /* + * print after we've already done the work, so as not to + * delay the state changes and notifications, for debugging + */ + if (lstate == lastlstate) + ipath_cdbg(LINKVERB, "Unchanged from last: %s " + "(%x)\n", ib_linkstate(dd, ibcs), ibstate); + else + ipath_cdbg(VERBOSE, "Unit %u: link up to %s %s (%x)\n", + dd->ipath_unit, ib_linkstate(dd, ibcs), + ipath_ibcstatus_str[ltstate], ibstate); + } else { /* down */ if (dd->ipath_flags & IPATH_LINKACTIVE) signal_ib_event(dd, IB_EVENT_PORT_ERR); dd->ipath_flags |= IPATH_LINKDOWN; @@ -391,69 +461,28 @@ static void handle_e_ibstatuschanged(struct ipath_devdata *dd, IPATH_LINKARMED); *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY; dd->ipath_lli_counter = 0; - if (!noprint) { - if (((dd->ipath_lastibcstat >> - INFINIPATH_IBCS_LINKSTATE_SHIFT) & - INFINIPATH_IBCS_LINKSTATE_MASK) - == INFINIPATH_IBCS_L_STATE_ACTIVE) - /* if from up to down be more vocal */ - ipath_cdbg(VERBOSE, - "Unit %u link now down (%s)\n", - dd->ipath_unit, - ipath_ibcstatus_str[ltstate]); - else - ipath_cdbg(VERBOSE, "Unit %u link is " - "down (%s)\n", dd->ipath_unit, - ipath_ibcstatus_str[ltstate]); - } - dd->ipath_f_setextled(dd, lstate, ltstate); - } else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_ACTIVE) { - dd->ipath_flags |= IPATH_LINKACTIVE; - dd->ipath_flags &= - ~(IPATH_LINKUNK | IPATH_LINKINIT | IPATH_LINKDOWN | - IPATH_LINKARMED | IPATH_NOCABLE); - *dd->ipath_statusp &= ~IPATH_STATUS_IB_NOCABLE; - *dd->ipath_statusp |= - IPATH_STATUS_IB_READY | IPATH_STATUS_IB_CONF; - dd->ipath_f_setextled(dd, lstate, ltstate); - signal_ib_event(dd, IB_EVENT_PORT_ACTIVE); - } else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_INIT) { - if (dd->ipath_flags & IPATH_LINKACTIVE) - signal_ib_event(dd, IB_EVENT_PORT_ERR); - /* - * set INIT and DOWN. Down is checked by most of the other - * code, but INIT is useful to know in a few places. - */ - dd->ipath_flags |= IPATH_LINKINIT | IPATH_LINKDOWN; - dd->ipath_flags &= - ~(IPATH_LINKUNK | IPATH_LINKACTIVE | IPATH_LINKARMED - | IPATH_NOCABLE); - *dd->ipath_statusp &= ~(IPATH_STATUS_IB_NOCABLE - | IPATH_STATUS_IB_READY); - dd->ipath_f_setextled(dd, lstate, ltstate); - } else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_ARM) { - if (dd->ipath_flags & IPATH_LINKACTIVE) - signal_ib_event(dd, IB_EVENT_PORT_ERR); - dd->ipath_flags |= IPATH_LINKARMED; - dd->ipath_flags &= - ~(IPATH_LINKUNK | IPATH_LINKDOWN | IPATH_LINKINIT | - IPATH_LINKACTIVE | IPATH_NOCABLE); - *dd->ipath_statusp &= ~(IPATH_STATUS_IB_NOCABLE - | IPATH_STATUS_IB_READY); - dd->ipath_f_setextled(dd, lstate, ltstate); - } else { - if (!noprint) - ipath_dbg("IBstatuschange unit %u: %s (%x)\n", - dd->ipath_unit, - ipath_ibcstatus_str[ltstate], ltstate); + if (lastlstate != INFINIPATH_IBCS_L_STATE_DOWN) + ipath_cdbg(VERBOSE, "Unit %u link state down " + "(state 0x%x), from %s\n", + dd->ipath_unit, lstate, + ib_linkstate(dd, dd->ipath_lastibcstat)); + else + ipath_cdbg(LINKVERB, "Unit %u link state changed " + "to %s (0x%x) from down (%x)\n", + dd->ipath_unit, + ipath_ibcstatus_str[ltstate], + ibstate, lastlstate); } + skip_ibchange: - dd->ipath_lastibcstat = val; + dd->ipath_lastibcstat = ibcs; +done: + return; } static void handle_supp_msgs(struct ipath_devdata *dd, - unsigned supp_msgs, char *msg, int msgsz) + unsigned supp_msgs, char *msg, u32 msgsz) { /* * Print the message unless it's ibc status change only, which @@ -461,12 +490,19 @@ static void handle_supp_msgs(struct ipath_devdata *dd, */ if (dd->ipath_lasterror & ~INFINIPATH_E_IBSTATUSCHANGED) { int iserr; - iserr = ipath_decode_err(msg, msgsz, + ipath_err_t mask; + iserr = ipath_decode_err(dd, msg, msgsz, dd->ipath_lasterror & ~INFINIPATH_E_IBSTATUSCHANGED); - if (dd->ipath_lasterror & - ~(INFINIPATH_E_RRCVEGRFULL | - INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS)) + + mask = INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL | + INFINIPATH_E_PKTERRS | INFINIPATH_E_SDMADISABLED; + + /* if we're in debug, then don't mask SDMADISABLED msgs */ + if (ipath_debug & __IPATH_DBG) + mask &= ~INFINIPATH_E_SDMADISABLED; + + if (dd->ipath_lasterror & ~mask) ipath_dev_err(dd, "Suppressed %u messages for " "fast-repeating errors (%s) (%llx)\n", supp_msgs, msg, @@ -493,7 +529,7 @@ static void handle_supp_msgs(struct ipath_devdata *dd, static unsigned handle_frequent_errors(struct ipath_devdata *dd, ipath_err_t errs, char *msg, - int msgsz, int *noprint) + u32 msgsz, int *noprint) { unsigned long nc; static unsigned long nextmsg_time; @@ -523,19 +559,125 @@ static unsigned handle_frequent_errors(struct ipath_devdata *dd, return supp_msgs; } +static void handle_sdma_errors(struct ipath_devdata *dd, ipath_err_t errs) +{ + unsigned long flags; + int expected; + + if (ipath_debug & __IPATH_DBG) { + char msg[128]; + ipath_decode_err(dd, msg, sizeof msg, errs & + INFINIPATH_E_SDMAERRS); + ipath_dbg("errors %lx (%s)\n", (unsigned long)errs, msg); + } + if (ipath_debug & __IPATH_VERBDBG) { + unsigned long tl, hd, status, lengen; + tl = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmatail); + hd = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmahead); + status = ipath_read_kreg64(dd + , dd->ipath_kregs->kr_senddmastatus); + lengen = ipath_read_kreg64(dd, + dd->ipath_kregs->kr_senddmalengen); + ipath_cdbg(VERBOSE, "sdma tl 0x%lx hd 0x%lx status 0x%lx " + "lengen 0x%lx\n", tl, hd, status, lengen); + } + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + __set_bit(IPATH_SDMA_DISABLED, &dd->ipath_sdma_status); + expected = test_bit(IPATH_SDMA_ABORTING, &dd->ipath_sdma_status); + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + if (!expected) + ipath_cancel_sends(dd, 1); +} + +static void handle_sdma_intr(struct ipath_devdata *dd, u64 istat) +{ + unsigned long flags; + int expected; + + if ((istat & INFINIPATH_I_SDMAINT) && + !test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + ipath_sdma_intr(dd); + + if (istat & INFINIPATH_I_SDMADISABLED) { + expected = test_bit(IPATH_SDMA_ABORTING, + &dd->ipath_sdma_status); + ipath_dbg("%s SDmaDisabled intr\n", + expected ? "expected" : "unexpected"); + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + __set_bit(IPATH_SDMA_DISABLED, &dd->ipath_sdma_status); + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + if (!expected) + ipath_cancel_sends(dd, 1); + if (!test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + tasklet_hi_schedule(&dd->ipath_sdma_abort_task); + } +} + +static int handle_hdrq_full(struct ipath_devdata *dd) +{ + int chkerrpkts = 0; + u32 hd, tl; + u32 i; + + ipath_stats.sps_hdrqfull++; + for (i = 0; i < dd->ipath_cfgports; i++) { + struct ipath_portdata *pd = dd->ipath_pd[i]; + + if (i == 0) { + /* + * For kernel receive queues, we just want to know + * if there are packets in the queue that we can + * process. + */ + if (pd->port_head != ipath_get_hdrqtail(pd)) + chkerrpkts |= 1 << i; + continue; + } + + /* Skip if user context is not open */ + if (!pd || !pd->port_cnt) + continue; + + /* Don't report the same point multiple times. */ + if (dd->ipath_flags & IPATH_NODMA_RTAIL) + tl = ipath_read_ureg32(dd, ur_rcvhdrtail, i); + else + tl = ipath_get_rcvhdrtail(pd); + if (tl == pd->port_lastrcvhdrqtail) + continue; + + hd = ipath_read_ureg32(dd, ur_rcvhdrhead, i); + if (hd == (tl + 1) || (!hd && tl == dd->ipath_hdrqlast)) { + pd->port_lastrcvhdrqtail = tl; + pd->port_hdrqfull++; + /* flush hdrqfull so that poll() sees it */ + wmb(); + wake_up_interruptible(&pd->port_wait); + } + } + + return chkerrpkts; +} + static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) { char msg[128]; u64 ignore_this_time = 0; - int i, iserr = 0; + u64 iserr = 0; int chkerrpkts = 0, noprint = 0; unsigned supp_msgs; int log_idx; - supp_msgs = handle_frequent_errors(dd, errs, msg, sizeof msg, &noprint); + /* + * don't report errors that are masked, either at init + * (not set in ipath_errormask), or temporarily (set in + * ipath_maskederrs) + */ + errs &= dd->ipath_errormask & ~dd->ipath_maskederrs; - /* don't report errors that are masked */ - errs &= ~dd->ipath_maskederrs; + supp_msgs = handle_frequent_errors(dd, errs, msg, (u32)sizeof msg, + &noprint); /* do these first, they are most important */ if (errs & INFINIPATH_E_HARDWARE) { @@ -550,6 +692,9 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) } } + if (errs & INFINIPATH_E_SDMAERRS) + handle_sdma_errors(dd, errs); + if (!noprint && (errs & ~dd->ipath_e_bitsextant)) ipath_dev_err(dd, "error interrupt with unknown errors " "%llx set\n", (unsigned long long) @@ -580,18 +725,19 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) * ones on this particular interrupt, which also isn't great */ dd->ipath_maskederrs |= dd->ipath_lasterror | errs; + dd->ipath_errormask &= ~dd->ipath_maskederrs; ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, - dd->ipath_errormask); - s_iserr = ipath_decode_err(msg, sizeof msg, - dd->ipath_maskederrs); + dd->ipath_errormask); + s_iserr = ipath_decode_err(dd, msg, sizeof msg, + dd->ipath_maskederrs); if (dd->ipath_maskederrs & - ~(INFINIPATH_E_RRCVEGRFULL | - INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS)) + ~(INFINIPATH_E_RRCVEGRFULL | + INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS)) ipath_dev_err(dd, "Temporarily disabling " "error(s) %llx reporting; too frequent (%s)\n", - (unsigned long long)dd->ipath_maskederrs, + (unsigned long long) dd->ipath_maskederrs, msg); else { /* @@ -633,26 +779,43 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) INFINIPATH_E_IBSTATUSCHANGED); } - /* likely due to cancel, so suppress */ + if (errs & INFINIPATH_E_SENDSPECIALTRIGGER) { + dd->ipath_spectriggerhit++; + ipath_dbg("%lu special trigger hits\n", + dd->ipath_spectriggerhit); + } + + /* likely due to cancel; so suppress message unless verbose */ if ((errs & (INFINIPATH_E_SPKTLEN | INFINIPATH_E_SPIOARMLAUNCH)) && dd->ipath_lastcancel > jiffies) { - ipath_dbg("Suppressed armlaunch/spktlen after error send cancel\n"); + /* armlaunch takes precedence; it often causes both. */ + ipath_cdbg(VERBOSE, + "Suppressed %s error (%llx) after sendbuf cancel\n", + (errs & INFINIPATH_E_SPIOARMLAUNCH) ? + "armlaunch" : "sendpktlen", (unsigned long long)errs); errs &= ~(INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SPKTLEN); } if (!errs) return 0; - if (!noprint) + if (!noprint) { + ipath_err_t mask; /* - * the ones we mask off are handled specially below or above + * The ones we mask off are handled specially below + * or above. Also mask SDMADISABLED by default as it + * is too chatty. */ - ipath_decode_err(msg, sizeof msg, - errs & ~(INFINIPATH_E_IBSTATUSCHANGED | - INFINIPATH_E_RRCVEGRFULL | - INFINIPATH_E_RRCVHDRFULL | - INFINIPATH_E_HARDWARE)); - else + mask = INFINIPATH_E_IBSTATUSCHANGED | + INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL | + INFINIPATH_E_HARDWARE | INFINIPATH_E_SDMADISABLED; + + /* if we're in debug, then don't mask SDMADISABLED msgs */ + if (ipath_debug & __IPATH_DBG) + mask &= ~INFINIPATH_E_SDMADISABLED; + + ipath_decode_err(dd, msg, sizeof msg, errs & ~mask); + } else /* so we don't need if (!noprint) at strlcat's below */ *msg = 0; @@ -677,40 +840,8 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) * fast_stats, no more than every 5 seconds, user ports get printed * on close */ - if (errs & INFINIPATH_E_RRCVHDRFULL) { - u32 hd, tl; - ipath_stats.sps_hdrqfull++; - for (i = 0; i < dd->ipath_cfgports; i++) { - struct ipath_portdata *pd = dd->ipath_pd[i]; - if (i == 0) { - hd = pd->port_head; - tl = (u32) le64_to_cpu( - *dd->ipath_hdrqtailptr); - } else if (pd && pd->port_cnt && - pd->port_rcvhdrtail_kvaddr) { - /* - * don't report same point multiple times, - * except kernel - */ - tl = *(u64 *) pd->port_rcvhdrtail_kvaddr; - if (tl == pd->port_lastrcvhdrqtail) - continue; - hd = ipath_read_ureg32(dd, ur_rcvhdrhead, - i); - } else - continue; - if (hd == (tl + 1) || - (!hd && tl == dd->ipath_hdrqlast)) { - if (i == 0) - chkerrpkts = 1; - pd->port_lastrcvhdrqtail = tl; - pd->port_hdrqfull++; - /* flush hdrqfull so that poll() sees it */ - wmb(); - wake_up_interruptible(&pd->port_wait); - } - } - } + if (errs & INFINIPATH_E_RRCVHDRFULL) + chkerrpkts |= handle_hdrq_full(dd); if (errs & INFINIPATH_E_RRCVEGRFULL) { struct ipath_portdata *pd = dd->ipath_pd[0]; @@ -721,9 +852,8 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) * vs user) */ ipath_stats.sps_etidfull++; - if (pd->port_head != - (u32) le64_to_cpu(*dd->ipath_hdrqtailptr)) - chkerrpkts = 1; + if (pd->port_head != ipath_get_hdrqtail(pd)) + chkerrpkts |= 1; } /* @@ -741,16 +871,13 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) dd->ipath_flags &= ~(IPATH_LINKUNK | IPATH_LINKINIT | IPATH_LINKARMED | IPATH_LINKACTIVE); *dd->ipath_statusp &= ~IPATH_STATUS_IB_READY; - if (!noprint) { - u64 st = ipath_read_kreg64( - dd, dd->ipath_kregs->kr_ibcstatus); - ipath_dbg("Lost link, link now down (%s)\n", - ipath_ibcstatus_str[st & 0xf]); - } + ipath_dbg("Lost link, link now down (%s)\n", + ipath_ibcstatus_str[ipath_read_kreg64(dd, + dd->ipath_kregs->kr_ibcstatus) & 0xf]); } if (errs & INFINIPATH_E_IBSTATUSCHANGED) - handle_e_ibstatuschanged(dd, errs, noprint); + handle_e_ibstatuschanged(dd, errs); if (errs & INFINIPATH_E_RESET) { if (!noprint) @@ -765,9 +892,6 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) if (!noprint && *msg) { if (iserr) ipath_dev_err(dd, "%s error\n", msg); - else - dev_info(&dd->pcidev->dev, "%s packet problems\n", - msg); } if (dd->ipath_state_wanted & dd->ipath_flags) { ipath_cdbg(VERBOSE, "driver wanted state %x, iflags now %x, " @@ -779,7 +903,6 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) return chkerrpkts; } - /* * try to cleanup as much as possible for anything that might have gone * wrong while in freeze mode, such as pio buffers being written by user @@ -796,8 +919,7 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs) void ipath_clear_freeze(struct ipath_devdata *dd) { int i, im; - __le64 val; - unsigned long flags; + u64 val; /* disable error interrupts, to avoid confusion */ ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, 0ULL); @@ -816,14 +938,7 @@ void ipath_clear_freeze(struct ipath_devdata *dd) dd->ipath_control); /* ensure pio avail updates continue */ - spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl & ~INFINIPATH_S_PIOBUFAVAILUPD); - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl); - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); - spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + ipath_force_pio_avail_update(dd); /* * We just enabled pioavailupdate, so dma copy is almost certainly @@ -831,10 +946,13 @@ void ipath_clear_freeze(struct ipath_devdata *dd) */ for (i = 0; i < dd->ipath_pioavregs; i++) { /* deal with 6110 chip bug */ - im = i > 3 ? i ^ 1 : i; + im = (i > 3 && (dd->ipath_flags & IPATH_SWAP_PIOBUFS)) ? + i ^ 1 : i; val = ipath_read_kreg64(dd, (0x1000 / sizeof(u64)) + im); - dd->ipath_pioavailregs_dma[i] = dd->ipath_pioavailshadow[i] - = le64_to_cpu(val); + dd->ipath_pioavailregs_dma[i] = cpu_to_le64(val); + dd->ipath_pioavailshadow[i] = val | + (~dd->ipath_pioavailkernel[i] << + INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT); } /* @@ -950,7 +1068,7 @@ set: * process was waiting for a packet to arrive, and didn't want * to poll */ -static void handle_urcv(struct ipath_devdata *dd, u32 istat) +static void handle_urcv(struct ipath_devdata *dd, u64 istat) { u64 portr; int i; @@ -966,12 +1084,13 @@ static void handle_urcv(struct ipath_devdata *dd, u32 istat) * and ipath_poll_next()... */ rmb(); - portr = ((istat >> INFINIPATH_I_RCVAVAIL_SHIFT) & - dd->ipath_i_rcvavail_mask) - | ((istat >> INFINIPATH_I_RCVURG_SHIFT) & - dd->ipath_i_rcvurg_mask); + portr = ((istat >> dd->ipath_i_rcvavail_shift) & + dd->ipath_i_rcvavail_mask) | + ((istat >> dd->ipath_i_rcvurg_shift) & + dd->ipath_i_rcvurg_mask); for (i = 1; i < dd->ipath_cfgports; i++) { struct ipath_portdata *pd = dd->ipath_pd[i]; + if (portr & (1 << i) && pd && pd->port_cnt) { if (test_and_clear_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag)) { @@ -988,7 +1107,7 @@ static void handle_urcv(struct ipath_devdata *dd, u32 istat) } if (rcvdint) { /* only want to take one interrupt, so turn off the rcv - * interrupt for all the ports that we did the wakeup on + * interrupt for all the ports that we set the rcv_waiting * (but never for kernel port) */ ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, @@ -999,12 +1118,11 @@ static void handle_urcv(struct ipath_devdata *dd, u32 istat) irqreturn_t ipath_intr(int irq, void *data) { struct ipath_devdata *dd = data; - u32 istat, chk0rcv = 0; + u64 istat, chk0rcv = 0; ipath_err_t estat = 0; irqreturn_t ret; static unsigned unexpected = 0; - static const u32 port0rbits = (1U<<INFINIPATH_I_RCVAVAIL_SHIFT) | - (1U<<INFINIPATH_I_RCVURG_SHIFT); + u64 kportrbits; ipath_stats.sps_ints++; @@ -1053,17 +1171,17 @@ irqreturn_t ipath_intr(int irq, void *data) if (unlikely(istat & ~dd->ipath_i_bitsextant)) ipath_dev_err(dd, - "interrupt with unknown interrupts %x set\n", - istat & (u32) ~ dd->ipath_i_bitsextant); - else - ipath_cdbg(VERBOSE, "intr stat=0x%x\n", istat); + "interrupt with unknown interrupts %Lx set\n", + istat & ~dd->ipath_i_bitsextant); + else if (istat & ~INFINIPATH_I_ERROR) /* errors do own printing */ + ipath_cdbg(VERBOSE, "intr stat=0x%Lx\n", istat); - if (unlikely(istat & INFINIPATH_I_ERROR)) { + if (istat & INFINIPATH_I_ERROR) { ipath_stats.sps_errints++; estat = ipath_read_kreg64(dd, dd->ipath_kregs->kr_errorstatus); if (!estat) - dev_info(&dd->pcidev->dev, "error interrupt (%x), " + dev_info(&dd->pcidev->dev, "error interrupt (%Lx), " "but no error bits set!\n", istat); else if (estat == -1LL) /* @@ -1073,9 +1191,7 @@ irqreturn_t ipath_intr(int irq, void *data) ipath_dev_err(dd, "Read of error status failed " "(all bits set); ignoring\n"); else - if (handle_errors(dd, estat)) - /* force calling ipath_kreceive() */ - chk0rcv = 1; + chk0rcv |= handle_errors(dd, estat); } if (istat & INFINIPATH_I_GPIO) { @@ -1093,8 +1209,7 @@ irqreturn_t ipath_intr(int irq, void *data) gpiostatus = ipath_read_kreg32( dd, dd->ipath_kregs->kr_gpio_status); - /* First the error-counter case. - */ + /* First the error-counter case. */ if ((gpiostatus & IPATH_GPIO_ERRINTR_MASK) && (dd->ipath_flags & IPATH_GPIO_ERRINTRS)) { /* want to clear the bits we see asserted. */ @@ -1156,7 +1271,6 @@ irqreturn_t ipath_intr(int irq, void *data) (u64) to_clear); } } - chk0rcv |= istat & port0rbits; /* * Clear the interrupt bits we found set, unless they are receive @@ -1169,22 +1283,25 @@ irqreturn_t ipath_intr(int irq, void *data) ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, istat); /* - * handle port0 receive before checking for pio buffers available, - * since receives can overflow; piobuf waiters can afford a few - * extra cycles, since they were waiting anyway, and user's waiting - * for receive are at the bottom. + * Handle kernel receive queues before checking for pio buffers + * available since receives can overflow; piobuf waiters can afford + * a few extra cycles, since they were waiting anyway, and user's + * waiting for receive are at the bottom. */ - if (chk0rcv) { + kportrbits = (1ULL << dd->ipath_i_rcvavail_shift) | + (1ULL << dd->ipath_i_rcvurg_shift); + if (chk0rcv || (istat & kportrbits)) { + istat &= ~kportrbits; ipath_kreceive(dd->ipath_pd[0]); - istat &= ~port0rbits; } - if (istat & ((dd->ipath_i_rcvavail_mask << - INFINIPATH_I_RCVAVAIL_SHIFT) - | (dd->ipath_i_rcvurg_mask << - INFINIPATH_I_RCVURG_SHIFT))) + if (istat & ((dd->ipath_i_rcvavail_mask << dd->ipath_i_rcvavail_shift) | + (dd->ipath_i_rcvurg_mask << dd->ipath_i_rcvurg_shift))) handle_urcv(dd, istat); + if (istat & (INFINIPATH_I_SDMAINT | INFINIPATH_I_SDMADISABLED)) + handle_sdma_intr(dd, istat); + if (istat & INFINIPATH_I_SPIOBUFAVAIL) { unsigned long flags; @@ -1195,7 +1312,10 @@ irqreturn_t ipath_intr(int irq, void *data) ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); - handle_layer_pioavail(dd); + if (!(dd->ipath_flags & IPATH_HAS_SEND_DMA)) + handle_layer_pioavail(dd); + else + ipath_dbg("unexpected BUFAVAIL intr\n"); } ret = IRQ_HANDLED; diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h index ecf3f7ff771..5863cbe9930 100644 --- a/drivers/infiniband/hw/ipath/ipath_kernel.h +++ b/drivers/infiniband/hw/ipath/ipath_kernel.h @@ -1,7 +1,7 @@ #ifndef _IPATH_KERNEL_H #define _IPATH_KERNEL_H /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -42,6 +42,8 @@ #include <linux/pci.h> #include <linux/dma-mapping.h> #include <linux/mutex.h> +#include <linux/list.h> +#include <linux/scatterlist.h> #include <asm/io.h> #include <rdma/ib_verbs.h> @@ -175,9 +177,13 @@ struct ipath_portdata { u16 poll_type; /* port rcvhdrq head offset */ u32 port_head; + /* receive packet sequence counter */ + u32 port_seq_cnt; }; struct sk_buff; +struct ipath_sge_state; +struct ipath_verbs_txreq; /* * control information for layered drivers @@ -191,6 +197,40 @@ struct ipath_skbinfo { dma_addr_t phys; }; +struct ipath_sdma_txreq { + int flags; + int sg_count; + union { + struct scatterlist *sg; + void *map_addr; + }; + void (*callback)(void *, int); + void *callback_cookie; + int callback_status; + u16 start_idx; /* sdma private */ + u16 next_descq_idx; /* sdma private */ + struct list_head list; /* sdma private */ +}; + +struct ipath_sdma_desc { + __le64 qw[2]; +}; + +#define IPATH_SDMA_TXREQ_F_USELARGEBUF 0x1 +#define IPATH_SDMA_TXREQ_F_HEADTOHOST 0x2 +#define IPATH_SDMA_TXREQ_F_INTREQ 0x4 +#define IPATH_SDMA_TXREQ_F_FREEBUF 0x8 +#define IPATH_SDMA_TXREQ_F_FREEDESC 0x10 +#define IPATH_SDMA_TXREQ_F_VL15 0x20 + +#define IPATH_SDMA_TXREQ_S_OK 0 +#define IPATH_SDMA_TXREQ_S_SENDERROR 1 +#define IPATH_SDMA_TXREQ_S_ABORTED 2 +#define IPATH_SDMA_TXREQ_S_SHUTDOWN 3 + +/* max dwords in small buffer packet */ +#define IPATH_SMALLBUF_DWORDS (dd->ipath_piosize2k >> 2) + /* * Possible IB config parameters for ipath_f_get/set_ib_cfg() */ @@ -221,11 +261,6 @@ struct ipath_devdata { unsigned long ipath_physaddr; /* base of memory alloced for ipath_kregbase, for free */ u64 *ipath_kregalloc; - /* - * virtual address where port0 rcvhdrqtail updated for this unit. - * only written to by the chip, not the driver. - */ - volatile __le64 *ipath_hdrqtailptr; /* ipath_cfgports pointers */ struct ipath_portdata **ipath_pd; /* sk_buffs used by port 0 eager receive queue */ @@ -283,6 +318,7 @@ struct ipath_devdata { /* per chip actions needed for IB Link up/down changes */ int (*ipath_f_ib_updown)(struct ipath_devdata *, int, u64); + unsigned ipath_lastegr_idx; struct ipath_ibdev *verbs_dev; struct timer_list verbs_timer; /* total dwords sent (summed from counter) */ @@ -309,6 +345,7 @@ struct ipath_devdata { ipath_err_t ipath_lasthwerror; /* errors masked because they occur too fast */ ipath_err_t ipath_maskederrs; + u64 ipath_lastlinkrecov; /* link recoveries at last ACTIVE */ /* time in jiffies at which to re-enable maskederrs */ unsigned long ipath_unmasktime; /* count of egrfull errors, combined for all ports */ @@ -347,6 +384,7 @@ struct ipath_devdata { u32 ipath_lastrpkts; /* pio bufs allocated per port */ u32 ipath_pbufsport; + u32 ipath_pioupd_thresh; /* update threshold, some chips */ /* * number of ports configured as max; zero is set to number chip * supports, less gives more pio bufs/port, etc. @@ -365,6 +403,7 @@ struct ipath_devdata { * get to multiple devices */ u32 ipath_lastpioindex; + u32 ipath_lastpioindexl; /* max length of freezemsg */ u32 ipath_freezelen; /* @@ -381,6 +420,15 @@ struct ipath_devdata { u32 ipath_pcibar0; /* so we can rewrite it after a chip reset */ u32 ipath_pcibar1; + u32 ipath_x1_fix_tries; + u32 ipath_autoneg_tries; + u32 serdes_first_init_done; + + struct ipath_relock { + atomic_t ipath_relock_timer_active; + struct timer_list ipath_relock_timer; + unsigned int ipath_relock_interval; /* in jiffies */ + } ipath_relock_singleton; /* interrupt number */ int ipath_irq; @@ -403,7 +451,7 @@ struct ipath_devdata { u64 __iomem *ipath_egrtidbase; /* lock to workaround chip bug 9437 and others */ spinlock_t ipath_kernel_tid_lock; - spinlock_t ipath_tid_lock; + spinlock_t ipath_user_tid_lock; spinlock_t ipath_sendctrl_lock; /* @@ -422,11 +470,48 @@ struct ipath_devdata { struct class_device *diag_class_dev; /* timer used to prevent stats overflow, error throttling, etc. */ struct timer_list ipath_stats_timer; + /* timer to verify interrupts work, and fallback if possible */ + struct timer_list ipath_intrchk_timer; void *ipath_dummy_hdrq; /* used after port close */ dma_addr_t ipath_dummy_hdrq_phys; + /* SendDMA related entries */ + spinlock_t ipath_sdma_lock; + u64 ipath_sdma_status; + unsigned long ipath_sdma_abort_jiffies; + unsigned long ipath_sdma_abort_intr_timeout; + unsigned long ipath_sdma_buf_jiffies; + struct ipath_sdma_desc *ipath_sdma_descq; + u64 ipath_sdma_descq_added; + u64 ipath_sdma_descq_removed; + int ipath_sdma_desc_nreserved; + u16 ipath_sdma_descq_cnt; + u16 ipath_sdma_descq_tail; + u16 ipath_sdma_descq_head; + u16 ipath_sdma_next_intr; + u16 ipath_sdma_reset_wait; + u8 ipath_sdma_generation; + struct tasklet_struct ipath_sdma_abort_task; + struct tasklet_struct ipath_sdma_notify_task; + struct list_head ipath_sdma_activelist; + struct list_head ipath_sdma_notifylist; + atomic_t ipath_sdma_vl15_count; + struct timer_list ipath_sdma_vl15_timer; + + dma_addr_t ipath_sdma_descq_phys; + volatile __le64 *ipath_sdma_head_dma; + dma_addr_t ipath_sdma_head_phys; + unsigned long ipath_ureg_align; /* user register alignment */ + struct delayed_work ipath_autoneg_work; + wait_queue_head_t ipath_autoneg_wait; + + /* HoL blocking / user app forward-progress state */ + unsigned ipath_hol_state; + unsigned ipath_hol_next; + struct timer_list ipath_hol_timer; + /* * Shadow copies of registers; size indicates read access size. * Most of them are readonly, but some are write-only register, @@ -447,6 +532,8 @@ struct ipath_devdata { * init time. */ unsigned long ipath_pioavailshadow[8]; + /* bitmap of send buffers available for the kernel to use with PIO. */ + unsigned long ipath_pioavailkernel[8]; /* shadow of kr_gpio_out, for rmw ops */ u64 ipath_gpio_out; /* shadow the gpio mask register */ @@ -472,6 +559,8 @@ struct ipath_devdata { u64 ipath_intconfig; /* kr_sendpiobufbase value */ u64 ipath_piobufbase; + /* kr_ibcddrctrl shadow */ + u64 ipath_ibcddrctrl; /* these are the "32 bit" regs */ @@ -488,7 +577,10 @@ struct ipath_devdata { unsigned long ipath_rcvctrl; /* shadow kr_sendctrl */ unsigned long ipath_sendctrl; - unsigned long ipath_lastcancel; /* to not count armlaunch after cancel */ + /* to not count armlaunch after cancel */ + unsigned long ipath_lastcancel; + /* count cases where special trigger was needed (double write) */ + unsigned long ipath_spectriggerhit; /* value we put in kr_rcvhdrcnt */ u32 ipath_rcvhdrcnt; @@ -510,6 +602,7 @@ struct ipath_devdata { u32 ipath_piobcnt4k; /* size in bytes of "4KB" PIO buffers */ u32 ipath_piosize4k; + u32 ipath_pioreserved; /* reserved special-inkernel; */ /* kr_rcvegrbase value */ u32 ipath_rcvegrbase; /* kr_rcvegrcnt value */ @@ -546,10 +639,10 @@ struct ipath_devdata { u32 ipath_init_ibmaxlen; /* size of each rcvegrbuffer */ u32 ipath_rcvegrbufsize; - /* width (2,4,8,16,32) from HT config reg */ - u32 ipath_htwidth; - /* HT speed (200,400,800,1000) from HT config */ - u32 ipath_htspeed; + /* localbus width (1, 2,4,8,16,32) from config space */ + u32 ipath_lbus_width; + /* localbus speed (HT: 200,400,800,1000; PCIe 2500) */ + u32 ipath_lbus_speed; /* * number of sequential ibcstatus change for polling active/quiet * (i.e., link not coming up). @@ -573,21 +666,14 @@ struct ipath_devdata { */ u8 ipath_serial[16]; /* human readable board version */ - u8 ipath_boardversion[80]; + u8 ipath_boardversion[96]; + u8 ipath_lbus_info[32]; /* human readable localbus info */ /* chip major rev, from ipath_revision */ u8 ipath_majrev; /* chip minor rev, from ipath_revision */ u8 ipath_minrev; /* board rev, from ipath_revision */ u8 ipath_boardrev; - - u8 ipath_r_portenable_shift; - u8 ipath_r_intravail_shift; - u8 ipath_r_tailupd_shift; - u8 ipath_r_portcfg_shift; - - /* unit # of this chip, if present */ - int ipath_unit; /* saved for restore after reset */ u8 ipath_pci_cacheline; /* LID mask control */ @@ -603,6 +689,14 @@ struct ipath_devdata { /* Rx Polarity inversion (compensate for ~tx on partner) */ u8 ipath_rx_pol_inv; + u8 ipath_r_portenable_shift; + u8 ipath_r_intravail_shift; + u8 ipath_r_tailupd_shift; + u8 ipath_r_portcfg_shift; + + /* unit # of this chip, if present */ + int ipath_unit; + /* local link integrity counter */ u32 ipath_lli_counter; /* local link integrity errors */ @@ -617,9 +711,6 @@ struct ipath_devdata { u32 ipath_overrun_thresh_errs; u32 ipath_lli_errs; - /* status check work */ - struct delayed_work status_work; - /* * Not all devices managed by a driver instance are the same * type, so these fields must be per-device. @@ -632,8 +723,8 @@ struct ipath_devdata { * Below should be computable from number of ports, * since they are never modified. */ - u32 ipath_i_rcvavail_mask; - u32 ipath_i_rcvurg_mask; + u64 ipath_i_rcvavail_mask; + u64 ipath_i_rcvurg_mask; u16 ipath_i_rcvurg_shift; u16 ipath_i_rcvavail_shift; @@ -641,8 +732,9 @@ struct ipath_devdata { * Register bits for selecting i2c direction and values, used for * I2C serial flash. */ - u16 ipath_gpio_sda_num; - u16 ipath_gpio_scl_num; + u8 ipath_gpio_sda_num; + u8 ipath_gpio_scl_num; + u8 ipath_i2c_chain_type; u64 ipath_gpio_sda; u64 ipath_gpio_scl; @@ -703,13 +795,51 @@ struct ipath_devdata { /* interrupt mitigation reload register info */ u16 ipath_jint_idle_ticks; /* idle clock ticks */ u16 ipath_jint_max_packets; /* max packets across all ports */ + + /* + * lock for access to SerDes, and flags to sequence preset + * versus steady-state. 7220-only at the moment. + */ + spinlock_t ipath_sdepb_lock; + u8 ipath_presets_needed; /* Set if presets to be restored next DOWN */ }; +/* ipath_hol_state values (stopping/starting user proc, send flushing) */ +#define IPATH_HOL_UP 0 +#define IPATH_HOL_DOWN 1 +/* ipath_hol_next toggle values, used when hol_state IPATH_HOL_DOWN */ +#define IPATH_HOL_DOWNSTOP 0 +#define IPATH_HOL_DOWNCONT 1 + +/* bit positions for sdma_status */ +#define IPATH_SDMA_ABORTING 0 +#define IPATH_SDMA_DISARMED 1 +#define IPATH_SDMA_DISABLED 2 +#define IPATH_SDMA_LAYERBUF 3 +#define IPATH_SDMA_RUNNING 62 +#define IPATH_SDMA_SHUTDOWN 63 + +/* bit combinations that correspond to abort states */ +#define IPATH_SDMA_ABORT_NONE 0 +#define IPATH_SDMA_ABORT_ABORTING (1UL << IPATH_SDMA_ABORTING) +#define IPATH_SDMA_ABORT_DISARMED ((1UL << IPATH_SDMA_ABORTING) | \ + (1UL << IPATH_SDMA_DISARMED)) +#define IPATH_SDMA_ABORT_DISABLED ((1UL << IPATH_SDMA_ABORTING) | \ + (1UL << IPATH_SDMA_DISABLED)) +#define IPATH_SDMA_ABORT_ABORTED ((1UL << IPATH_SDMA_ABORTING) | \ + (1UL << IPATH_SDMA_DISARMED) | (1UL << IPATH_SDMA_DISABLED)) +#define IPATH_SDMA_ABORT_MASK ((1UL<<IPATH_SDMA_ABORTING) | \ + (1UL << IPATH_SDMA_DISARMED) | (1UL << IPATH_SDMA_DISABLED)) + +#define IPATH_SDMA_BUF_NONE 0 +#define IPATH_SDMA_BUF_MASK (1UL<<IPATH_SDMA_LAYERBUF) + /* Private data for file operations */ struct ipath_filedata { struct ipath_portdata *pd; unsigned subport; unsigned tidcursor; + struct ipath_user_sdma_queue *pq; }; extern struct list_head ipath_dev_list; extern spinlock_t ipath_devs_lock; @@ -718,7 +848,7 @@ extern struct ipath_devdata *ipath_lookup(int unit); int ipath_init_chip(struct ipath_devdata *, int); int ipath_enable_wc(struct ipath_devdata *dd); void ipath_disable_wc(struct ipath_devdata *dd); -int ipath_count_units(int *npresentp, int *nupp, u32 *maxportsp); +int ipath_count_units(int *npresentp, int *nupp, int *maxportsp); void ipath_shutdown_device(struct ipath_devdata *); void ipath_clear_freeze(struct ipath_devdata *); @@ -741,7 +871,8 @@ struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd, gfp_t); extern int ipath_diag_inuse; irqreturn_t ipath_intr(int irq, void *devid); -int ipath_decode_err(char *buf, size_t blen, ipath_err_t err); +int ipath_decode_err(struct ipath_devdata *dd, char *buf, size_t blen, + ipath_err_t err); #if __IPATH_INFO || __IPATH_DBG extern const char *ipath_ibcstatus_str[]; #endif @@ -774,6 +905,13 @@ int ipath_set_lid(struct ipath_devdata *, u32, u8); int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv); void ipath_enable_armlaunch(struct ipath_devdata *); void ipath_disable_armlaunch(struct ipath_devdata *); +void ipath_hol_down(struct ipath_devdata *); +void ipath_hol_up(struct ipath_devdata *); +void ipath_hol_event(unsigned long); +void ipath_toggle_rclkrls(struct ipath_devdata *); +void ipath_sd7220_clr_ibpar(struct ipath_devdata *); +void ipath_set_relock_poll(struct ipath_devdata *, int); +void ipath_shutdown_relock_poll(struct ipath_devdata *); /* for use in system calls, where we want to know device type, etc. */ #define port_fp(fp) ((struct ipath_filedata *)(fp)->private_data)->pd @@ -781,11 +919,15 @@ void ipath_disable_armlaunch(struct ipath_devdata *); ((struct ipath_filedata *)(fp)->private_data)->subport #define tidcursor_fp(fp) \ ((struct ipath_filedata *)(fp)->private_data)->tidcursor +#define user_sdma_queue_fp(fp) \ + ((struct ipath_filedata *)(fp)->private_data)->pq /* * values for ipath_flags */ -/* The chip is up and initted */ + /* chip can report link latency (IB 1.2) */ +#define IPATH_HAS_LINK_LATENCY 0x1 + /* The chip is up and initted */ #define IPATH_INITTED 0x2 /* set if any user code has set kr_rcvhdrsize */ #define IPATH_RCVHDRSZ_SET 0x4 @@ -809,6 +951,8 @@ void ipath_disable_armlaunch(struct ipath_devdata *); #define IPATH_LINKUNK 0x400 /* Write combining flush needed for PIO */ #define IPATH_PIO_FLUSH_WC 0x1000 + /* DMA Receive tail pointer */ +#define IPATH_NODMA_RTAIL 0x2000 /* no IB cable, or no device on IB cable */ #define IPATH_NOCABLE 0x4000 /* Supports port zero per packet receive interrupts via @@ -819,16 +963,26 @@ void ipath_disable_armlaunch(struct ipath_devdata *); /* packet/word counters are 32 bit, else those 4 counters * are 64bit */ #define IPATH_32BITCOUNTERS 0x20000 - /* can miss port0 rx interrupts */ /* Interrupt register is 64 bits */ #define IPATH_INTREG_64 0x40000 + /* can miss port0 rx interrupts */ #define IPATH_DISABLED 0x80000 /* administratively disabled */ /* Use GPIO interrupts for new counters */ #define IPATH_GPIO_ERRINTRS 0x100000 #define IPATH_SWAP_PIOBUFS 0x200000 + /* Supports Send DMA */ +#define IPATH_HAS_SEND_DMA 0x400000 + /* Supports Send Count (not just word count) in PBC */ +#define IPATH_HAS_PBC_CNT 0x800000 /* Suppress heartbeat, even if turning off loopback */ #define IPATH_NO_HRTBT 0x1000000 +#define IPATH_HAS_THRESH_UPDATE 0x4000000 #define IPATH_HAS_MULT_IB_SPEED 0x8000000 +#define IPATH_IB_AUTONEG_INPROG 0x10000000 +#define IPATH_IB_AUTONEG_FAILED 0x20000000 + /* Linkdown-disable intentionally, Do not attempt to bring up */ +#define IPATH_IB_LINK_DISABLED 0x40000000 +#define IPATH_IB_FORCE_NOTIFY 0x80000000 /* force notify on next ib change */ /* Bits in GPIO for the added interrupts */ #define IPATH_GPIO_PORT0_BIT 2 @@ -847,13 +1001,18 @@ void ipath_disable_armlaunch(struct ipath_devdata *); /* free up any allocated data at closes */ void ipath_free_data(struct ipath_portdata *dd); -u32 __iomem *ipath_getpiobuf(struct ipath_devdata *, u32 *); +u32 __iomem *ipath_getpiobuf(struct ipath_devdata *, u32, u32 *); +void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start, + unsigned len, int avail); +void ipath_init_iba7220_funcs(struct ipath_devdata *); void ipath_init_iba6120_funcs(struct ipath_devdata *); void ipath_init_iba6110_funcs(struct ipath_devdata *); void ipath_get_eeprom_info(struct ipath_devdata *); int ipath_update_eeprom_log(struct ipath_devdata *dd); void ipath_inc_eeprom_err(struct ipath_devdata *dd, u32 eidx, u32 incr); u64 ipath_snap_cntr(struct ipath_devdata *, ipath_creg); +void ipath_disarm_senderrbufs(struct ipath_devdata *, int); +void ipath_force_pio_avail_update(struct ipath_devdata *); void signal_ib_event(struct ipath_devdata *dd, enum ib_event_type ev); /* @@ -865,6 +1024,34 @@ void signal_ib_event(struct ipath_devdata *dd, enum ib_event_type ev); #define IPATH_LED_LOG 2 /* Logical (link) YELLOW LED */ void ipath_set_led_override(struct ipath_devdata *dd, unsigned int val); +/* send dma routines */ +int setup_sdma(struct ipath_devdata *); +void teardown_sdma(struct ipath_devdata *); +void ipath_restart_sdma(struct ipath_devdata *); +void ipath_sdma_intr(struct ipath_devdata *); +int ipath_sdma_verbs_send(struct ipath_devdata *, struct ipath_sge_state *, + u32, struct ipath_verbs_txreq *); +/* ipath_sdma_lock should be locked before calling this. */ +int ipath_sdma_make_progress(struct ipath_devdata *dd); + +/* must be called under ipath_sdma_lock */ +static inline u16 ipath_sdma_descq_freecnt(const struct ipath_devdata *dd) +{ + return dd->ipath_sdma_descq_cnt - + (dd->ipath_sdma_descq_added - dd->ipath_sdma_descq_removed) - + 1 - dd->ipath_sdma_desc_nreserved; +} + +static inline void ipath_sdma_desc_reserve(struct ipath_devdata *dd, u16 cnt) +{ + dd->ipath_sdma_desc_nreserved += cnt; +} + +static inline void ipath_sdma_desc_unreserve(struct ipath_devdata *dd, u16 cnt) +{ + dd->ipath_sdma_desc_nreserved -= cnt; +} + /* * number of words used for protocol header if not set by ipath_userinit(); */ @@ -875,6 +1062,8 @@ void ipath_release_user_pages(struct page **, size_t); void ipath_release_user_pages_on_close(struct page **, size_t); int ipath_eeprom_read(struct ipath_devdata *, u8, void *, int); int ipath_eeprom_write(struct ipath_devdata *, u8, const void *, int); +int ipath_tempsense_read(struct ipath_devdata *, u8 regnum); +int ipath_tempsense_write(struct ipath_devdata *, u8 regnum, u8 data); /* these are used for the registers that vary with port */ void ipath_write_kreg_port(const struct ipath_devdata *, ipath_kreg, @@ -891,8 +1080,7 @@ void ipath_write_kreg_port(const struct ipath_devdata *, ipath_kreg, /* * At the moment, none of the s-registers are writable, so no - * ipath_write_sreg(), and none of the c-registers are writable, so no - * ipath_write_creg(). + * ipath_write_sreg(). */ /** @@ -1001,6 +1189,27 @@ static inline u32 ipath_get_rcvhdrtail(const struct ipath_portdata *pd) pd->port_rcvhdrtail_kvaddr)); } +static inline u32 ipath_get_hdrqtail(const struct ipath_portdata *pd) +{ + const struct ipath_devdata *dd = pd->port_dd; + u32 hdrqtail; + + if (dd->ipath_flags & IPATH_NODMA_RTAIL) { + __le32 *rhf_addr; + u32 seq; + + rhf_addr = (__le32 *) pd->port_rcvhdrq + + pd->port_head + dd->ipath_rhf_offset; + seq = ipath_hdrget_seq(rhf_addr); + hdrqtail = pd->port_head; + if (seq == pd->port_seq_cnt) + hdrqtail++; + } else + hdrqtail = ipath_get_rcvhdrtail(pd); + + return hdrqtail; +} + static inline u64 ipath_read_ireg(const struct ipath_devdata *dd, ipath_kreg r) { return (dd->ipath_flags & IPATH_INTREG_64) ? @@ -1029,6 +1238,21 @@ static inline u32 ipath_ib_linktrstate(struct ipath_devdata *dd, u64 ibcs) } /* + * from contents of IBCStatus (or a saved copy), return logical link state + * combination of link state and linktraining state (down, active, init, + * arm, etc. + */ +static inline u32 ipath_ib_state(struct ipath_devdata *dd, u64 ibcs) +{ + u32 ibs; + ibs = (u32)(ibcs >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) & + dd->ibcs_lts_mask; + ibs |= (u32)(ibcs & + (INFINIPATH_IBCS_LINKSTATE_MASK << dd->ibcs_ls_shift)); + return ibs; +} + +/* * sysfs interface. */ @@ -1053,6 +1277,7 @@ int ipathfs_remove_device(struct ipath_devdata *); dma_addr_t ipath_map_page(struct pci_dev *, struct page *, unsigned long, size_t, int); dma_addr_t ipath_map_single(struct pci_dev *, void *, size_t, int); +const char *ipath_get_unit_name(int unit); /* * Flush write combining store buffers (if present) and perform a write @@ -1065,11 +1290,8 @@ dma_addr_t ipath_map_single(struct pci_dev *, void *, size_t, int); #endif extern unsigned ipath_debug; /* debugging bit mask */ - -#define IPATH_MAX_PARITY_ATTEMPTS 10000 /* max times to try recovery */ - -const char *ipath_get_unit_name(int unit); - +extern unsigned ipath_linkrecovery; +extern unsigned ipath_mtu4096; extern struct mutex ipath_mutex; #define IPATH_DRV_NAME "ib_ipath" @@ -1096,7 +1318,7 @@ extern struct mutex ipath_mutex; # define __IPATH_DBG_WHICH(which,fmt,...) \ do { \ - if(unlikely(ipath_debug&(which))) \ + if (unlikely(ipath_debug & (which))) \ printk(KERN_DEBUG IPATH_DRV_NAME ": %s: " fmt, \ __func__,##__VA_ARGS__); \ } while(0) diff --git a/drivers/infiniband/hw/ipath/ipath_mad.c b/drivers/infiniband/hw/ipath/ipath_mad.c index b34b91d3723..1ff46ae7dd9 100644 --- a/drivers/infiniband/hw/ipath/ipath_mad.c +++ b/drivers/infiniband/hw/ipath/ipath_mad.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -146,6 +146,15 @@ static int recv_subn_get_guidinfo(struct ib_smp *smp, return reply(smp); } +static void set_link_width_enabled(struct ipath_devdata *dd, u32 w) +{ + (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_LWID_ENB, w); +} + +static void set_link_speed_enabled(struct ipath_devdata *dd, u32 s) +{ + (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_SPD_ENB, s); +} static int get_overrunthreshold(struct ipath_devdata *dd) { @@ -226,6 +235,7 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, struct ib_device *ibdev, u8 port) { struct ipath_ibdev *dev; + struct ipath_devdata *dd; struct ib_port_info *pip = (struct ib_port_info *)smp->data; u16 lid; u8 ibcstat; @@ -239,6 +249,7 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, } dev = to_idev(ibdev); + dd = dev->dd; /* Clear all fields. Only set the non-zero fields. */ memset(smp->data, 0, sizeof(smp->data)); @@ -248,25 +259,28 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, dev->mkeyprot == 0) pip->mkey = dev->mkey; pip->gid_prefix = dev->gid_prefix; - lid = dev->dd->ipath_lid; + lid = dd->ipath_lid; pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE; pip->sm_lid = cpu_to_be16(dev->sm_lid); pip->cap_mask = cpu_to_be32(dev->port_cap_flags); /* pip->diag_code; */ pip->mkey_lease_period = cpu_to_be16(dev->mkey_lease_period); pip->local_port_num = port; - pip->link_width_enabled = dev->link_width_enabled; - pip->link_width_supported = 3; /* 1x or 4x */ - pip->link_width_active = 2; /* 4x */ - pip->linkspeed_portstate = 0x10; /* 2.5Gbps */ - ibcstat = dev->dd->ipath_lastibcstat; - pip->linkspeed_portstate |= ((ibcstat >> 4) & 0x3) + 1; + pip->link_width_enabled = dd->ipath_link_width_enabled; + pip->link_width_supported = dd->ipath_link_width_supported; + pip->link_width_active = dd->ipath_link_width_active; + pip->linkspeed_portstate = dd->ipath_link_speed_supported << 4; + ibcstat = dd->ipath_lastibcstat; + /* map LinkState to IB portinfo values. */ + pip->linkspeed_portstate |= ipath_ib_linkstate(dd, ibcstat) + 1; + pip->portphysstate_linkdown = - (ipath_cvt_physportstate[ibcstat & 0xf] << 4) | - (get_linkdowndefaultstate(dev->dd) ? 1 : 2); - pip->mkeyprot_resv_lmc = (dev->mkeyprot << 6) | dev->dd->ipath_lmc; - pip->linkspeedactive_enabled = 0x11; /* 2.5Gbps, 2.5Gbps */ - switch (dev->dd->ipath_ibmtu) { + (ipath_cvt_physportstate[ibcstat & dd->ibcs_lts_mask] << 4) | + (get_linkdowndefaultstate(dd) ? 1 : 2); + pip->mkeyprot_resv_lmc = (dev->mkeyprot << 6) | dd->ipath_lmc; + pip->linkspeedactive_enabled = (dd->ipath_link_speed_active << 4) | + dd->ipath_link_speed_enabled; + switch (dd->ipath_ibmtu) { case 4096: mtu = IB_MTU_4096; break; @@ -292,19 +306,15 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, /* pip->vl_arb_high_cap; // only one VL */ /* pip->vl_arb_low_cap; // only one VL */ /* InitTypeReply = 0 */ - /* - * Note: the chips support a maximum MTU of 4096, but the driver - * hasn't implemented this feature yet, so set the maximum value - * to 2048. - */ - pip->inittypereply_mtucap = IB_MTU_2048; - // HCAs ignore VLStallCount and HOQLife + /* our mtu cap depends on whether 4K MTU enabled or not */ + pip->inittypereply_mtucap = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048; + /* HCAs ignore VLStallCount and HOQLife */ /* pip->vlstallcnt_hoqlife; */ pip->operationalvl_pei_peo_fpi_fpo = 0x10; /* OVLs = 1 */ pip->mkey_violations = cpu_to_be16(dev->mkey_violations); /* P_KeyViolations are counted by hardware. */ pip->pkey_violations = - cpu_to_be16((ipath_get_cr_errpkey(dev->dd) - + cpu_to_be16((ipath_get_cr_errpkey(dd) - dev->z_pkey_violations) & 0xFFFF); pip->qkey_violations = cpu_to_be16(dev->qkey_violations); /* Only the hardware GUID is supported for now */ @@ -313,10 +323,17 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, /* 32.768 usec. response time (guessing) */ pip->resv_resptimevalue = 3; pip->localphyerrors_overrunerrors = - (get_phyerrthreshold(dev->dd) << 4) | - get_overrunthreshold(dev->dd); + (get_phyerrthreshold(dd) << 4) | + get_overrunthreshold(dd); /* pip->max_credit_hint; */ - /* pip->link_roundtrip_latency[3]; */ + if (dev->port_cap_flags & IB_PORT_LINK_LATENCY_SUP) { + u32 v; + + v = dd->ipath_f_get_ib_cfg(dd, IPATH_IB_CFG_LINKLATENCY); + pip->link_roundtrip_latency[0] = v >> 16; + pip->link_roundtrip_latency[1] = v >> 8; + pip->link_roundtrip_latency[2] = v; + } ret = reply(smp); @@ -444,19 +461,25 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, ib_dispatch_event(&event); } - /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */ + /* Allow 1x or 4x to be set (see 14.2.6.6). */ lwe = pip->link_width_enabled; - if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE)) - goto err; - if (lwe == 0xFF) - dev->link_width_enabled = 3; /* 1x or 4x */ - else if (lwe) - dev->link_width_enabled = lwe; + if (lwe) { + if (lwe == 0xFF) + lwe = dd->ipath_link_width_supported; + else if (lwe >= 16 || (lwe & ~dd->ipath_link_width_supported)) + goto err; + set_link_width_enabled(dd, lwe); + } - /* Only 2.5 Gbs supported. */ + /* Allow 2.5 or 5.0 Gbs. */ lse = pip->linkspeedactive_enabled & 0xF; - if (lse >= 2 && lse <= 0xE) - goto err; + if (lse) { + if (lse == 15) + lse = dd->ipath_link_speed_supported; + else if (lse >= 8 || (lse & ~dd->ipath_link_speed_supported)) + goto err; + set_link_speed_enabled(dd, lse); + } /* Set link down default state. */ switch (pip->portphysstate_linkdown & 0xF) { @@ -491,6 +514,8 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, mtu = 2048; break; case IB_MTU_4096: + if (!ipath_mtu4096) + goto err; mtu = 4096; break; default: @@ -565,6 +590,10 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, else goto err; ipath_set_linkstate(dd, lstate); + if (lstate == IPATH_IB_LINKDOWN_DISABLE) { + ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED; + goto done; + } ipath_wait_linkstate(dd, IPATH_LINKINIT | IPATH_LINKARMED | IPATH_LINKACTIVE, 1000); break; @@ -948,10 +977,14 @@ static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp, * nsec. 0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec. Sample * intervals are counted in ticks. Since we use Linux timers, that * count in jiffies, we can't sample for less than 1000 ticks if HZ - * == 1000 (4000 ticks if HZ is 250). + * == 1000 (4000 ticks if HZ is 250). link_speed_active returns 2 for + * DDR, 1 for SDR, set the tick to 1 for DDR, 0 for SDR on chips that + * have hardware support for delaying packets. */ - /* XXX This is WRONG. */ - p->tick = 250; /* 1 usec. */ + if (crp->cr_psstat) + p->tick = dev->dd->ipath_link_speed_active - 1; + else + p->tick = 250; /* 1 usec. */ p->counter_width = 4; /* 32 bit counters */ p->counter_mask0_9 = COUNTER_MASK0_9; spin_lock_irqsave(&dev->pending_lock, flags); @@ -1364,7 +1397,8 @@ static int process_subn(struct ib_device *ibdev, int mad_flags, } /* Is the mkey in the process of expiring? */ - if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) { + if (dev->mkey_lease_timeout && + time_after_eq(jiffies, dev->mkey_lease_timeout)) { /* Clear timeout and mkey protection field. */ dev->mkey_lease_timeout = 0; dev->mkeyprot = 0; diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c index 087ed316647..dd5b6e9d57c 100644 --- a/drivers/infiniband/hw/ipath/ipath_qp.c +++ b/drivers/infiniband/hw/ipath/ipath_qp.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -340,6 +340,7 @@ static void ipath_reset_qp(struct ipath_qp *qp, enum ib_qp_type type) qp->s_flags &= IPATH_S_SIGNAL_REQ_WR; qp->s_hdrwords = 0; qp->s_wqe = NULL; + qp->s_pkt_delay = 0; qp->s_psn = 0; qp->r_psn = 0; qp->r_msn = 0; @@ -392,7 +393,6 @@ int ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err) qp->ibqp.qp_num, qp->remote_qpn, err); spin_lock(&dev->pending_lock); - /* XXX What if its already removed by the timeout code? */ if (!list_empty(&qp->timerwait)) list_del_init(&qp->timerwait); if (!list_empty(&qp->piowait)) @@ -516,13 +516,13 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, goto inval; /* - * Note: the chips support a maximum MTU of 4096, but the driver - * hasn't implemented this feature yet, so don't allow Path MTU - * values greater than 2048. + * don't allow invalid Path MTU values or greater than 2048 + * unless we are configured for a 4KB MTU */ - if (attr_mask & IB_QP_PATH_MTU) - if (attr->path_mtu > IB_MTU_2048) - goto inval; + if ((attr_mask & IB_QP_PATH_MTU) && + (ib_mtu_enum_to_int(attr->path_mtu) == -1 || + (attr->path_mtu > IB_MTU_2048 && !ipath_mtu4096))) + goto inval; if (attr_mask & IB_QP_PATH_MIG_STATE) if (attr->path_mig_state != IB_MIG_MIGRATED && @@ -564,8 +564,10 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, if (attr_mask & IB_QP_ACCESS_FLAGS) qp->qp_access_flags = attr->qp_access_flags; - if (attr_mask & IB_QP_AV) + if (attr_mask & IB_QP_AV) { qp->remote_ah_attr = attr->ah_attr; + qp->s_dmult = ipath_ib_rate_to_mult(attr->ah_attr.static_rate); + } if (attr_mask & IB_QP_PATH_MTU) qp->path_mtu = attr->path_mtu; @@ -748,22 +750,33 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, size_t sz; struct ib_qp *ret; - if (init_attr->cap.max_send_sge > ib_ipath_max_sges || - init_attr->cap.max_recv_sge > ib_ipath_max_sges || - init_attr->cap.max_send_wr > ib_ipath_max_qp_wrs || - init_attr->cap.max_recv_wr > ib_ipath_max_qp_wrs) { - ret = ERR_PTR(-ENOMEM); + if (init_attr->create_flags) { + ret = ERR_PTR(-EINVAL); goto bail; } - if (init_attr->cap.max_send_sge + - init_attr->cap.max_recv_sge + - init_attr->cap.max_send_wr + - init_attr->cap.max_recv_wr == 0) { + if (init_attr->cap.max_send_sge > ib_ipath_max_sges || + init_attr->cap.max_send_wr > ib_ipath_max_qp_wrs) { ret = ERR_PTR(-EINVAL); goto bail; } + /* Check receive queue parameters if no SRQ is specified. */ + if (!init_attr->srq) { + if (init_attr->cap.max_recv_sge > ib_ipath_max_sges || + init_attr->cap.max_recv_wr > ib_ipath_max_qp_wrs) { + ret = ERR_PTR(-EINVAL); + goto bail; + } + if (init_attr->cap.max_send_sge + + init_attr->cap.max_send_wr + + init_attr->cap.max_recv_sge + + init_attr->cap.max_recv_wr == 0) { + ret = ERR_PTR(-EINVAL); + goto bail; + } + } + switch (init_attr->qp_type) { case IB_QPT_UC: case IB_QPT_RC: @@ -840,6 +853,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, goto bail_qp; } qp->ip = NULL; + qp->s_tx = NULL; ipath_reset_qp(qp, init_attr->qp_type); break; @@ -945,12 +959,20 @@ int ipath_destroy_qp(struct ib_qp *ibqp) /* Stop the sending tasklet. */ tasklet_kill(&qp->s_task); + if (qp->s_tx) { + atomic_dec(&qp->refcount); + if (qp->s_tx->txreq.flags & IPATH_SDMA_TXREQ_F_FREEBUF) + kfree(qp->s_tx->txreq.map_addr); + } + /* Make sure the QP isn't on the timeout list. */ spin_lock_irqsave(&dev->pending_lock, flags); if (!list_empty(&qp->timerwait)) list_del_init(&qp->timerwait); if (!list_empty(&qp->piowait)) list_del_init(&qp->piowait); + if (qp->s_tx) + list_add(&qp->s_tx->txreq.list, &dev->txreq_free); spin_unlock_irqrestore(&dev->pending_lock, flags); /* @@ -1021,7 +1043,6 @@ void ipath_sqerror_qp(struct ipath_qp *qp, struct ib_wc *wc) qp->ibqp.qp_num, qp->remote_qpn, wc->status); spin_lock(&dev->pending_lock); - /* XXX What if its already removed by the timeout code? */ if (!list_empty(&qp->timerwait)) list_del_init(&qp->timerwait); if (!list_empty(&qp->piowait)) diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c index 40f3e37d7ad..c405dfba553 100644 --- a/drivers/infiniband/hw/ipath/ipath_rc.c +++ b/drivers/infiniband/hw/ipath/ipath_rc.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -31,6 +31,8 @@ * SOFTWARE. */ +#include <linux/io.h> + #include "ipath_verbs.h" #include "ipath_kernel.h" @@ -306,7 +308,7 @@ int ipath_make_rc_req(struct ipath_qp *qp) else { qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ - ohdr->u.imm_data = wqe->wr.imm_data; + ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; } if (wqe->wr.send_flags & IB_SEND_SOLICITED) @@ -344,7 +346,7 @@ int ipath_make_rc_req(struct ipath_qp *qp) qp->s_state = OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE); /* Immediate data comes after RETH */ - ohdr->u.rc.imm_data = wqe->wr.imm_data; + ohdr->u.rc.imm_data = wqe->wr.ex.imm_data; hwords += 1; if (wqe->wr.send_flags & IB_SEND_SOLICITED) bth0 |= 1 << 23; @@ -488,7 +490,7 @@ int ipath_make_rc_req(struct ipath_qp *qp) else { qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ - ohdr->u.imm_data = wqe->wr.imm_data; + ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; } if (wqe->wr.send_flags & IB_SEND_SOLICITED) @@ -524,7 +526,7 @@ int ipath_make_rc_req(struct ipath_qp *qp) else { qp->s_state = OP(RDMA_WRITE_LAST_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ - ohdr->u.imm_data = wqe->wr.imm_data; + ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; if (wqe->wr.send_flags & IB_SEND_SOLICITED) bth0 |= 1 << 23; @@ -585,19 +587,39 @@ bail: static void send_rc_ack(struct ipath_qp *qp) { struct ipath_ibdev *dev = to_idev(qp->ibqp.device); + struct ipath_devdata *dd; u16 lrh0; u32 bth0; u32 hwords; + u32 __iomem *piobuf; struct ipath_ib_header hdr; struct ipath_other_headers *ohdr; unsigned long flags; + spin_lock_irqsave(&qp->s_lock, flags); + /* Don't send ACK or NAK if a RDMA read or atomic is pending. */ if (qp->r_head_ack_queue != qp->s_tail_ack_queue || (qp->s_flags & IPATH_S_ACK_PENDING) || qp->s_ack_state != OP(ACKNOWLEDGE)) goto queue_ack; + spin_unlock_irqrestore(&qp->s_lock, flags); + + dd = dev->dd; + piobuf = ipath_getpiobuf(dd, 0, NULL); + if (!piobuf) { + /* + * We are out of PIO buffers at the moment. + * Pass responsibility for sending the ACK to the + * send tasklet so that when a PIO buffer becomes + * available, the ACK is sent ahead of other outgoing + * packets. + */ + spin_lock_irqsave(&qp->s_lock, flags); + goto queue_ack; + } + /* Construct the header. */ ohdr = &hdr.u.oth; lrh0 = IPATH_LRH_BTH; @@ -611,7 +633,7 @@ static void send_rc_ack(struct ipath_qp *qp) lrh0 = IPATH_LRH_GRH; } /* read pkey_index w/o lock (its atomic) */ - bth0 = ipath_get_pkey(dev->dd, qp->s_pkey_index) | + bth0 = ipath_get_pkey(dd, qp->s_pkey_index) | (OP(ACKNOWLEDGE) << 24) | (1 << 22); if (qp->r_nak_state) ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) | @@ -623,30 +645,29 @@ static void send_rc_ack(struct ipath_qp *qp) hdr.lrh[0] = cpu_to_be16(lrh0); hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid); hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC); - hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid); + hdr.lrh[3] = cpu_to_be16(dd->ipath_lid); ohdr->bth[0] = cpu_to_be32(bth0); ohdr->bth[1] = cpu_to_be32(qp->remote_qpn); ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK); - /* - * If we can send the ACK, clear the ACK state. - */ - if (ipath_verbs_send(qp, &hdr, hwords, NULL, 0) == 0) { - dev->n_unicast_xmit++; - goto done; - } + writeq(hwords + 1, piobuf); - /* - * We are out of PIO buffers at the moment. - * Pass responsibility for sending the ACK to the - * send tasklet so that when a PIO buffer becomes - * available, the ACK is sent ahead of other outgoing - * packets. - */ - dev->n_rc_qacks++; + if (dd->ipath_flags & IPATH_PIO_FLUSH_WC) { + u32 *hdrp = (u32 *) &hdr; + + ipath_flush_wc(); + __iowrite32_copy(piobuf + 2, hdrp, hwords - 1); + ipath_flush_wc(); + __raw_writel(hdrp[hwords - 1], piobuf + hwords + 1); + } else + __iowrite32_copy(piobuf + 2, (u32 *) &hdr, hwords); + + ipath_flush_wc(); + + dev->n_unicast_xmit++; + goto done; queue_ack: - spin_lock_irqsave(&qp->s_lock, flags); dev->n_rc_qacks++; qp->s_flags |= IPATH_S_ACK_PENDING; qp->s_nak_state = qp->r_nak_state; diff --git a/drivers/infiniband/hw/ipath/ipath_registers.h b/drivers/infiniband/hw/ipath/ipath_registers.h index 92ad73a7fff..8f44d0cf383 100644 --- a/drivers/infiniband/hw/ipath/ipath_registers.h +++ b/drivers/infiniband/hw/ipath/ipath_registers.h @@ -63,67 +63,92 @@ /* kr_control bits */ #define INFINIPATH_C_FREEZEMODE 0x00000002 #define INFINIPATH_C_LINKENABLE 0x00000004 -#define INFINIPATH_C_RESET 0x00000001 /* kr_sendctrl bits */ #define INFINIPATH_S_DISARMPIOBUF_SHIFT 16 +#define INFINIPATH_S_UPDTHRESH_SHIFT 24 +#define INFINIPATH_S_UPDTHRESH_MASK 0x1f #define IPATH_S_ABORT 0 #define IPATH_S_PIOINTBUFAVAIL 1 #define IPATH_S_PIOBUFAVAILUPD 2 #define IPATH_S_PIOENABLE 3 +#define IPATH_S_SDMAINTENABLE 9 +#define IPATH_S_SDMASINGLEDESCRIPTOR 10 +#define IPATH_S_SDMAENABLE 11 +#define IPATH_S_SDMAHALT 12 #define IPATH_S_DISARM 31 #define INFINIPATH_S_ABORT (1U << IPATH_S_ABORT) #define INFINIPATH_S_PIOINTBUFAVAIL (1U << IPATH_S_PIOINTBUFAVAIL) #define INFINIPATH_S_PIOBUFAVAILUPD (1U << IPATH_S_PIOBUFAVAILUPD) #define INFINIPATH_S_PIOENABLE (1U << IPATH_S_PIOENABLE) +#define INFINIPATH_S_SDMAINTENABLE (1U << IPATH_S_SDMAINTENABLE) +#define INFINIPATH_S_SDMASINGLEDESCRIPTOR \ + (1U << IPATH_S_SDMASINGLEDESCRIPTOR) +#define INFINIPATH_S_SDMAENABLE (1U << IPATH_S_SDMAENABLE) +#define INFINIPATH_S_SDMAHALT (1U << IPATH_S_SDMAHALT) #define INFINIPATH_S_DISARM (1U << IPATH_S_DISARM) -/* kr_rcvctrl bits */ +/* kr_rcvctrl bits that are the same on multiple chips */ #define INFINIPATH_R_PORTENABLE_SHIFT 0 #define INFINIPATH_R_QPMAP_ENABLE (1ULL << 38) /* kr_intstatus, kr_intclear, kr_intmask bits */ -#define INFINIPATH_I_RCVURG_SHIFT 0 -#define INFINIPATH_I_RCVAVAIL_SHIFT 12 -#define INFINIPATH_I_ERROR 0x80000000 -#define INFINIPATH_I_SPIOSENT 0x40000000 -#define INFINIPATH_I_SPIOBUFAVAIL 0x20000000 -#define INFINIPATH_I_GPIO 0x10000000 +#define INFINIPATH_I_SDMAINT 0x8000000000000000ULL +#define INFINIPATH_I_SDMADISABLED 0x4000000000000000ULL +#define INFINIPATH_I_ERROR 0x0000000080000000ULL +#define INFINIPATH_I_SPIOSENT 0x0000000040000000ULL +#define INFINIPATH_I_SPIOBUFAVAIL 0x0000000020000000ULL +#define INFINIPATH_I_GPIO 0x0000000010000000ULL +#define INFINIPATH_I_JINT 0x0000000004000000ULL /* kr_errorstatus, kr_errorclear, kr_errormask bits */ -#define INFINIPATH_E_RFORMATERR 0x0000000000000001ULL -#define INFINIPATH_E_RVCRC 0x0000000000000002ULL -#define INFINIPATH_E_RICRC 0x0000000000000004ULL -#define INFINIPATH_E_RMINPKTLEN 0x0000000000000008ULL -#define INFINIPATH_E_RMAXPKTLEN 0x0000000000000010ULL -#define INFINIPATH_E_RLONGPKTLEN 0x0000000000000020ULL -#define INFINIPATH_E_RSHORTPKTLEN 0x0000000000000040ULL -#define INFINIPATH_E_RUNEXPCHAR 0x0000000000000080ULL -#define INFINIPATH_E_RUNSUPVL 0x0000000000000100ULL -#define INFINIPATH_E_REBP 0x0000000000000200ULL -#define INFINIPATH_E_RIBFLOW 0x0000000000000400ULL -#define INFINIPATH_E_RBADVERSION 0x0000000000000800ULL -#define INFINIPATH_E_RRCVEGRFULL 0x0000000000001000ULL -#define INFINIPATH_E_RRCVHDRFULL 0x0000000000002000ULL -#define INFINIPATH_E_RBADTID 0x0000000000004000ULL -#define INFINIPATH_E_RHDRLEN 0x0000000000008000ULL -#define INFINIPATH_E_RHDR 0x0000000000010000ULL -#define INFINIPATH_E_RIBLOSTLINK 0x0000000000020000ULL -#define INFINIPATH_E_SMINPKTLEN 0x0000000020000000ULL -#define INFINIPATH_E_SMAXPKTLEN 0x0000000040000000ULL -#define INFINIPATH_E_SUNDERRUN 0x0000000080000000ULL -#define INFINIPATH_E_SPKTLEN 0x0000000100000000ULL -#define INFINIPATH_E_SDROPPEDSMPPKT 0x0000000200000000ULL -#define INFINIPATH_E_SDROPPEDDATAPKT 0x0000000400000000ULL -#define INFINIPATH_E_SPIOARMLAUNCH 0x0000000800000000ULL -#define INFINIPATH_E_SUNEXPERRPKTNUM 0x0000001000000000ULL -#define INFINIPATH_E_SUNSUPVL 0x0000002000000000ULL -#define INFINIPATH_E_IBSTATUSCHANGED 0x0001000000000000ULL -#define INFINIPATH_E_INVALIDADDR 0x0002000000000000ULL -#define INFINIPATH_E_RESET 0x0004000000000000ULL -#define INFINIPATH_E_HARDWARE 0x0008000000000000ULL +#define INFINIPATH_E_RFORMATERR 0x0000000000000001ULL +#define INFINIPATH_E_RVCRC 0x0000000000000002ULL +#define INFINIPATH_E_RICRC 0x0000000000000004ULL +#define INFINIPATH_E_RMINPKTLEN 0x0000000000000008ULL +#define INFINIPATH_E_RMAXPKTLEN 0x0000000000000010ULL +#define INFINIPATH_E_RLONGPKTLEN 0x0000000000000020ULL +#define INFINIPATH_E_RSHORTPKTLEN 0x0000000000000040ULL +#define INFINIPATH_E_RUNEXPCHAR 0x0000000000000080ULL +#define INFINIPATH_E_RUNSUPVL 0x0000000000000100ULL +#define INFINIPATH_E_REBP 0x0000000000000200ULL +#define INFINIPATH_E_RIBFLOW 0x0000000000000400ULL +#define INFINIPATH_E_RBADVERSION 0x0000000000000800ULL +#define INFINIPATH_E_RRCVEGRFULL 0x0000000000001000ULL +#define INFINIPATH_E_RRCVHDRFULL 0x0000000000002000ULL +#define INFINIPATH_E_RBADTID 0x0000000000004000ULL +#define INFINIPATH_E_RHDRLEN 0x0000000000008000ULL +#define INFINIPATH_E_RHDR 0x0000000000010000ULL +#define INFINIPATH_E_RIBLOSTLINK 0x0000000000020000ULL +#define INFINIPATH_E_SENDSPECIALTRIGGER 0x0000000008000000ULL +#define INFINIPATH_E_SDMADISABLED 0x0000000010000000ULL +#define INFINIPATH_E_SMINPKTLEN 0x0000000020000000ULL +#define INFINIPATH_E_SMAXPKTLEN 0x0000000040000000ULL +#define INFINIPATH_E_SUNDERRUN 0x0000000080000000ULL +#define INFINIPATH_E_SPKTLEN 0x0000000100000000ULL +#define INFINIPATH_E_SDROPPEDSMPPKT 0x0000000200000000ULL +#define INFINIPATH_E_SDROPPEDDATAPKT 0x0000000400000000ULL +#define INFINIPATH_E_SPIOARMLAUNCH 0x0000000800000000ULL +#define INFINIPATH_E_SUNEXPERRPKTNUM 0x0000001000000000ULL +#define INFINIPATH_E_SUNSUPVL 0x0000002000000000ULL +#define INFINIPATH_E_SENDBUFMISUSE 0x0000004000000000ULL +#define INFINIPATH_E_SDMAGENMISMATCH 0x0000008000000000ULL +#define INFINIPATH_E_SDMAOUTOFBOUND 0x0000010000000000ULL +#define INFINIPATH_E_SDMATAILOUTOFBOUND 0x0000020000000000ULL +#define INFINIPATH_E_SDMABASE 0x0000040000000000ULL +#define INFINIPATH_E_SDMA1STDESC 0x0000080000000000ULL +#define INFINIPATH_E_SDMARPYTAG 0x0000100000000000ULL +#define INFINIPATH_E_SDMADWEN 0x0000200000000000ULL +#define INFINIPATH_E_SDMAMISSINGDW 0x0000400000000000ULL +#define INFINIPATH_E_SDMAUNEXPDATA 0x0000800000000000ULL +#define INFINIPATH_E_IBSTATUSCHANGED 0x0001000000000000ULL +#define INFINIPATH_E_INVALIDADDR 0x0002000000000000ULL +#define INFINIPATH_E_RESET 0x0004000000000000ULL +#define INFINIPATH_E_HARDWARE 0x0008000000000000ULL +#define INFINIPATH_E_SDMADESCADDRMISALIGN 0x0010000000000000ULL +#define INFINIPATH_E_INVALIDEEPCMD 0x0020000000000000ULL /* * this is used to print "common" packet errors only when the @@ -134,6 +159,17 @@ | INFINIPATH_E_RICRC | INFINIPATH_E_RSHORTPKTLEN \ | INFINIPATH_E_REBP ) +/* Convenience for decoding Send DMA errors */ +#define INFINIPATH_E_SDMAERRS ( \ + INFINIPATH_E_SDMAGENMISMATCH | INFINIPATH_E_SDMAOUTOFBOUND | \ + INFINIPATH_E_SDMATAILOUTOFBOUND | INFINIPATH_E_SDMABASE | \ + INFINIPATH_E_SDMA1STDESC | INFINIPATH_E_SDMARPYTAG | \ + INFINIPATH_E_SDMADWEN | INFINIPATH_E_SDMAMISSINGDW | \ + INFINIPATH_E_SDMAUNEXPDATA | \ + INFINIPATH_E_SDMADESCADDRMISALIGN | \ + INFINIPATH_E_SDMADISABLED | \ + INFINIPATH_E_SENDBUFMISUSE) + /* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */ /* TXEMEMPARITYERR bit 0: PIObuf, 1: PIOpbc, 2: launchfifo * RXEMEMPARITYERR bit 0: rcvbuf, 1: lookupq, 2: expTID, 3: eagerTID @@ -158,7 +194,7 @@ #define INFINIPATH_HWE_RXEMEMPARITYERR_HDRINFO 0x40ULL /* waldo specific -- find the rest in ipath_6110.c */ #define INFINIPATH_HWE_RXDSYNCMEMPARITYERR 0x0000000400000000ULL -/* monty specific -- find the rest in ipath_6120.c */ +/* 6120/7220 specific -- find the rest in ipath_6120.c and ipath_7220.c */ #define INFINIPATH_HWE_MEMBISTFAILED 0x0040000000000000ULL /* kr_hwdiagctrl bits */ @@ -185,8 +221,8 @@ #define INFINIPATH_IBCC_LINKINITCMD_SLEEP 3 #define INFINIPATH_IBCC_LINKINITCMD_SHIFT 16 #define INFINIPATH_IBCC_LINKCMD_MASK 0x3ULL -#define INFINIPATH_IBCC_LINKCMD_DOWN 1 /* move to 0x11 */ -#define INFINIPATH_IBCC_LINKCMD_ARMED 2 /* move to 0x21 */ +#define INFINIPATH_IBCC_LINKCMD_DOWN 1 /* move to 0x11 */ +#define INFINIPATH_IBCC_LINKCMD_ARMED 2 /* move to 0x21 */ #define INFINIPATH_IBCC_LINKCMD_ACTIVE 3 /* move to 0x31 */ #define INFINIPATH_IBCC_LINKCMD_SHIFT 18 #define INFINIPATH_IBCC_MAXPKTLEN_MASK 0x7FFULL @@ -201,10 +237,9 @@ #define INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE 0x4000000000000000ULL /* kr_ibcstatus bits */ -#define INFINIPATH_IBCS_LINKTRAININGSTATE_MASK 0xF #define INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT 0 #define INFINIPATH_IBCS_LINKSTATE_MASK 0x7 -#define INFINIPATH_IBCS_LINKSTATE_SHIFT 4 + #define INFINIPATH_IBCS_TXREADY 0x40000000 #define INFINIPATH_IBCS_TXCREDITOK 0x80000000 /* link training states (shift by @@ -222,30 +257,13 @@ #define INFINIPATH_IBCS_LT_STATE_RECOVERRETRAIN 0x0c #define INFINIPATH_IBCS_LT_STATE_RECOVERWAITRMT 0x0e #define INFINIPATH_IBCS_LT_STATE_RECOVERIDLE 0x0f -/* link state machine states (shift by INFINIPATH_IBCS_LINKSTATE_SHIFT) */ +/* link state machine states (shift by ibcs_ls_shift) */ #define INFINIPATH_IBCS_L_STATE_DOWN 0x0 #define INFINIPATH_IBCS_L_STATE_INIT 0x1 #define INFINIPATH_IBCS_L_STATE_ARM 0x2 #define INFINIPATH_IBCS_L_STATE_ACTIVE 0x3 #define INFINIPATH_IBCS_L_STATE_ACT_DEFER 0x4 -/* combination link status states that we use with some frequency */ -#define IPATH_IBSTATE_MASK ((INFINIPATH_IBCS_LINKTRAININGSTATE_MASK \ - << INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) | \ - (INFINIPATH_IBCS_LINKSTATE_MASK \ - <<INFINIPATH_IBCS_LINKSTATE_SHIFT)) -#define IPATH_IBSTATE_INIT ((INFINIPATH_IBCS_L_STATE_INIT \ - << INFINIPATH_IBCS_LINKSTATE_SHIFT) | \ - (INFINIPATH_IBCS_LT_STATE_LINKUP \ - <<INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT)) -#define IPATH_IBSTATE_ARM ((INFINIPATH_IBCS_L_STATE_ARM \ - << INFINIPATH_IBCS_LINKSTATE_SHIFT) | \ - (INFINIPATH_IBCS_LT_STATE_LINKUP \ - <<INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT)) -#define IPATH_IBSTATE_ACTIVE ((INFINIPATH_IBCS_L_STATE_ACTIVE \ - << INFINIPATH_IBCS_LINKSTATE_SHIFT) | \ - (INFINIPATH_IBCS_LT_STATE_LINKUP \ - <<INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT)) /* kr_extstatus bits */ #define INFINIPATH_EXTS_SERDESPLLLOCK 0x1 @@ -286,8 +304,7 @@ /* L1 Power down; use with RXDETECT, Otherwise not used on IB side */ #define INFINIPATH_SERDC0_L1PWR_DN 0xF0ULL -/* kr_xgxsconfig bits */ -#define INFINIPATH_XGXS_RESET 0x7ULL +/* common kr_xgxsconfig bits (or safe in all, even if not implemented) */ #define INFINIPATH_XGXS_RX_POL_SHIFT 19 #define INFINIPATH_XGXS_RX_POL_MASK 0xfULL @@ -417,6 +434,29 @@ struct ipath_kregs { ipath_kreg kr_pcieq1serdesconfig0; ipath_kreg kr_pcieq1serdesconfig1; ipath_kreg kr_pcieq1serdesstatus; + ipath_kreg kr_hrtbt_guid; + ipath_kreg kr_ibcddrctrl; + ipath_kreg kr_ibcddrstatus; + ipath_kreg kr_jintreload; + + /* send dma related regs */ + ipath_kreg kr_senddmabase; + ipath_kreg kr_senddmalengen; + ipath_kreg kr_senddmatail; + ipath_kreg kr_senddmahead; + ipath_kreg kr_senddmaheadaddr; + ipath_kreg kr_senddmabufmask0; + ipath_kreg kr_senddmabufmask1; + ipath_kreg kr_senddmabufmask2; + ipath_kreg kr_senddmastatus; + + /* SerDes related regs (IBA7220-only) */ + ipath_kreg kr_ibserdesctrl; + ipath_kreg kr_ib_epbacc; + ipath_kreg kr_ib_epbtrans; + ipath_kreg kr_pcie_epbacc; + ipath_kreg kr_pcie_epbtrans; + ipath_kreg kr_ib_ddsrxeq; }; struct ipath_cregs { diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c index a59bdbd0ed8..8ac5c1d82cc 100644 --- a/drivers/infiniband/hw/ipath/ipath_ruc.c +++ b/drivers/infiniband/hw/ipath/ipath_ruc.c @@ -310,7 +310,7 @@ again: switch (wqe->wr.opcode) { case IB_WR_SEND_WITH_IMM: wc.wc_flags = IB_WC_WITH_IMM; - wc.imm_data = wqe->wr.imm_data; + wc.imm_data = wqe->wr.ex.imm_data; /* FALLTHROUGH */ case IB_WR_SEND: if (!ipath_get_rwqe(qp, 0)) { @@ -339,7 +339,7 @@ again: goto err; } wc.wc_flags = IB_WC_WITH_IMM; - wc.imm_data = wqe->wr.imm_data; + wc.imm_data = wqe->wr.ex.imm_data; if (!ipath_get_rwqe(qp, 1)) goto rnr_nak; /* FALLTHROUGH */ @@ -483,14 +483,16 @@ done: static void want_buffer(struct ipath_devdata *dd) { - unsigned long flags; - - spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); - dd->ipath_sendctrl |= INFINIPATH_S_PIOINTBUFAVAIL; - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl); - ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); - spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + if (!(dd->ipath_flags & IPATH_HAS_SEND_DMA)) { + unsigned long flags; + + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl |= INFINIPATH_S_PIOINTBUFAVAIL; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + } } /** diff --git a/drivers/infiniband/hw/ipath/ipath_sd7220.c b/drivers/infiniband/hw/ipath/ipath_sd7220.c new file mode 100644 index 00000000000..aa47eb54952 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_sd7220.c @@ -0,0 +1,1462 @@ +/* + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. + * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +/* + * This file contains all of the code that is specific to the SerDes + * on the InfiniPath 7220 chip. + */ + +#include <linux/pci.h> +#include <linux/delay.h> + +#include "ipath_kernel.h" +#include "ipath_registers.h" +#include "ipath_7220.h" + +/* + * The IBSerDesMappTable is a memory that holds values to be stored in + * various SerDes registers by IBC. It is not part of the normal kregs + * map and is used in exactly one place, hence the #define below. + */ +#define KR_IBSerDesMappTable (0x94000 / (sizeof(uint64_t))) + +/* + * Below used for sdnum parameter, selecting one of the two sections + * used for PCIe, or the single SerDes used for IB. + */ +#define PCIE_SERDES0 0 +#define PCIE_SERDES1 1 + +/* + * The EPB requires addressing in a particular form. EPB_LOC() is intended + * to make #definitions a little more readable. + */ +#define EPB_ADDR_SHF 8 +#define EPB_LOC(chn, elt, reg) \ + (((elt & 0xf) | ((chn & 7) << 4) | ((reg & 0x3f) << 9)) << \ + EPB_ADDR_SHF) +#define EPB_IB_QUAD0_CS_SHF (25) +#define EPB_IB_QUAD0_CS (1U << EPB_IB_QUAD0_CS_SHF) +#define EPB_IB_UC_CS_SHF (26) +#define EPB_PCIE_UC_CS_SHF (27) +#define EPB_GLOBAL_WR (1U << (EPB_ADDR_SHF + 8)) + +/* Forward declarations. */ +static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc, + u32 data, u32 mask); +static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val, + int mask); +static int ipath_sd_trimdone_poll(struct ipath_devdata *dd); +static void ipath_sd_trimdone_monitor(struct ipath_devdata *dd, + const char *where); +static int ipath_sd_setvals(struct ipath_devdata *dd); +static int ipath_sd_early(struct ipath_devdata *dd); +static int ipath_sd_dactrim(struct ipath_devdata *dd); +/* Set the registers that IBC may muck with to their default "preset" values */ +int ipath_sd7220_presets(struct ipath_devdata *dd); +static int ipath_internal_presets(struct ipath_devdata *dd); +/* Tweak the register (CMUCTRL5) that contains the TRIMSELF controls */ +static int ipath_sd_trimself(struct ipath_devdata *dd, int val); +static int epb_access(struct ipath_devdata *dd, int sdnum, int claim); + +void ipath_set_relock_poll(struct ipath_devdata *dd, int ibup); + +/* + * Below keeps track of whether the "once per power-on" initialization has + * been done, because uC code Version 1.32.17 or higher allows the uC to + * be reset at will, and Automatic Equalization may require it. So the + * state of the reset "pin", as reflected in was_reset parameter to + * ipath_sd7220_init() is no longer valid. Instead, we check for the + * actual uC code having been loaded. + */ +static int ipath_ibsd_ucode_loaded(struct ipath_devdata *dd) +{ + if (!dd->serdes_first_init_done && (ipath_sd7220_ib_vfy(dd) > 0)) + dd->serdes_first_init_done = 1; + return dd->serdes_first_init_done; +} + +/* repeat #define for local use. "Real" #define is in ipath_iba7220.c */ +#define INFINIPATH_HWE_IB_UC_MEMORYPARITYERR 0x0000004000000000ULL +#define IB_MPREG5 (EPB_LOC(6, 0, 0xE) | (1L << EPB_IB_UC_CS_SHF)) +#define IB_MPREG6 (EPB_LOC(6, 0, 0xF) | (1U << EPB_IB_UC_CS_SHF)) +#define UC_PAR_CLR_D 8 +#define UC_PAR_CLR_M 0xC +#define IB_CTRL2(chn) (EPB_LOC(chn, 7, 3) | EPB_IB_QUAD0_CS) +#define START_EQ1(chan) EPB_LOC(chan, 7, 0x27) + +void ipath_sd7220_clr_ibpar(struct ipath_devdata *dd) +{ + int ret; + + /* clear, then re-enable parity errs */ + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, + UC_PAR_CLR_D, UC_PAR_CLR_M); + if (ret < 0) { + ipath_dev_err(dd, "Failed clearing IBSerDes Parity err\n"); + goto bail; + } + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0, + UC_PAR_CLR_M); + + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + udelay(4); + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear, + INFINIPATH_HWE_IB_UC_MEMORYPARITYERR); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); +bail: + return; +} + +/* + * After a reset or other unusual event, the epb interface may need + * to be re-synchronized, between the host and the uC. + * returns <0 for failure to resync within IBSD_RESYNC_TRIES (not expected) + */ +#define IBSD_RESYNC_TRIES 3 +#define IB_PGUDP(chn) (EPB_LOC((chn), 2, 1) | EPB_IB_QUAD0_CS) +#define IB_CMUDONE(chn) (EPB_LOC((chn), 7, 0xF) | EPB_IB_QUAD0_CS) + +static int ipath_resync_ibepb(struct ipath_devdata *dd) +{ + int ret, pat, tries, chn; + u32 loc; + + ret = -1; + chn = 0; + for (tries = 0; tries < (4 * IBSD_RESYNC_TRIES); ++tries) { + loc = IB_PGUDP(chn); + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0); + if (ret < 0) { + ipath_dev_err(dd, "Failed read in resync\n"); + continue; + } + if (ret != 0xF0 && ret != 0x55 && tries == 0) + ipath_dev_err(dd, "unexpected pattern in resync\n"); + pat = ret ^ 0xA5; /* alternate F0 and 55 */ + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, pat, 0xFF); + if (ret < 0) { + ipath_dev_err(dd, "Failed write in resync\n"); + continue; + } + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0); + if (ret < 0) { + ipath_dev_err(dd, "Failed re-read in resync\n"); + continue; + } + if (ret != pat) { + ipath_dev_err(dd, "Failed compare1 in resync\n"); + continue; + } + loc = IB_CMUDONE(chn); + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0); + if (ret < 0) { + ipath_dev_err(dd, "Failed CMUDONE rd in resync\n"); + continue; + } + if ((ret & 0x70) != ((chn << 4) | 0x40)) { + ipath_dev_err(dd, "Bad CMUDONE value %02X, chn %d\n", + ret, chn); + continue; + } + if (++chn == 4) + break; /* Success */ + } + ipath_cdbg(VERBOSE, "Resync in %d tries\n", tries); + return (ret > 0) ? 0 : ret; +} + +/* + * Localize the stuff that should be done to change IB uC reset + * returns <0 for errors. + */ +static int ipath_ibsd_reset(struct ipath_devdata *dd, int assert_rst) +{ + u64 rst_val; + int ret = 0; + unsigned long flags; + + rst_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibserdesctrl); + if (assert_rst) { + /* + * Vendor recommends "interrupting" uC before reset, to + * minimize possible glitches. + */ + spin_lock_irqsave(&dd->ipath_sdepb_lock, flags); + epb_access(dd, IB_7220_SERDES, 1); + rst_val |= 1ULL; + /* Squelch possible parity error from _asserting_ reset */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, + dd->ipath_hwerrmask & + ~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR); + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibserdesctrl, rst_val); + /* flush write, delay to ensure it took effect */ + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + udelay(2); + /* once it's reset, can remove interrupt */ + epb_access(dd, IB_7220_SERDES, -1); + spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags); + } else { + /* + * Before we de-assert reset, we need to deal with + * possible glitch on the Parity-error line. + * Suppress it around the reset, both in chip-level + * hwerrmask and in IB uC control reg. uC will allow + * it again during startup. + */ + u64 val; + rst_val &= ~(1ULL); + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, + dd->ipath_hwerrmask & + ~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR); + + ret = ipath_resync_ibepb(dd); + if (ret < 0) + ipath_dev_err(dd, "unable to re-sync IB EPB\n"); + + /* set uC control regs to suppress parity errs */ + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG5, 1, 1); + if (ret < 0) + goto bail; + /* IB uC code past Version 1.32.17 allow suppression of wdog */ + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80, + 0x80); + if (ret < 0) { + ipath_dev_err(dd, "Failed to set WDOG disable\n"); + goto bail; + } + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibserdesctrl, rst_val); + /* flush write, delay for startup */ + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + udelay(1); + /* clear, then re-enable parity errs */ + ipath_sd7220_clr_ibpar(dd); + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus); + if (val & INFINIPATH_HWE_IB_UC_MEMORYPARITYERR) { + ipath_dev_err(dd, "IBUC Parity still set after RST\n"); + dd->ipath_hwerrmask &= + ~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR; + } + ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask, + dd->ipath_hwerrmask); + } + +bail: + return ret; +} + +static void ipath_sd_trimdone_monitor(struct ipath_devdata *dd, + const char *where) +{ + int ret, chn, baduns; + u64 val; + + if (!where) + where = "?"; + + /* give time for reset to settle out in EPB */ + udelay(2); + + ret = ipath_resync_ibepb(dd); + if (ret < 0) + ipath_dev_err(dd, "not able to re-sync IB EPB (%s)\n", where); + + /* Do "sacrificial read" to get EPB in sane state after reset */ + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_CTRL2(0), 0, 0); + if (ret < 0) + ipath_dev_err(dd, "Failed TRIMDONE 1st read, (%s)\n", where); + + /* Check/show "summary" Trim-done bit in IBCStatus */ + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); + if (val & (1ULL << 11)) + ipath_cdbg(VERBOSE, "IBCS TRIMDONE set (%s)\n", where); + else + ipath_dev_err(dd, "IBCS TRIMDONE clear (%s)\n", where); + + udelay(2); + + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80, 0x80); + if (ret < 0) + ipath_dev_err(dd, "Failed Dummy RMW, (%s)\n", where); + udelay(10); + + baduns = 0; + + for (chn = 3; chn >= 0; --chn) { + /* Read CTRL reg for each channel to check TRIMDONE */ + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, + IB_CTRL2(chn), 0, 0); + if (ret < 0) + ipath_dev_err(dd, "Failed checking TRIMDONE, chn %d" + " (%s)\n", chn, where); + + if (!(ret & 0x10)) { + int probe; + baduns |= (1 << chn); + ipath_dev_err(dd, "TRIMDONE cleared on chn %d (%02X)." + " (%s)\n", chn, ret, where); + probe = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, + IB_PGUDP(0), 0, 0); + ipath_dev_err(dd, "probe is %d (%02X)\n", + probe, probe); + probe = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, + IB_CTRL2(chn), 0, 0); + ipath_dev_err(dd, "re-read: %d (%02X)\n", + probe, probe); + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, + IB_CTRL2(chn), 0x10, 0x10); + if (ret < 0) + ipath_dev_err(dd, + "Err on TRIMDONE rewrite1\n"); + } + } + for (chn = 3; chn >= 0; --chn) { + /* Read CTRL reg for each channel to check TRIMDONE */ + if (baduns & (1 << chn)) { + ipath_dev_err(dd, + "Reseting TRIMDONE on chn %d (%s)\n", + chn, where); + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, + IB_CTRL2(chn), 0x10, 0x10); + if (ret < 0) + ipath_dev_err(dd, "Failed re-setting " + "TRIMDONE, chn %d (%s)\n", + chn, where); + } + } +} + +/* + * Below is portion of IBA7220-specific bringup_serdes() that actually + * deals with registers and memory within the SerDes itself. + * Post IB uC code version 1.32.17, was_reset being 1 is not really + * informative, so we double-check. + */ +int ipath_sd7220_init(struct ipath_devdata *dd, int was_reset) +{ + int ret = 1; /* default to failure */ + int first_reset; + int val_stat; + + if (!was_reset) { + /* entered with reset not asserted, we need to do it */ + ipath_ibsd_reset(dd, 1); + ipath_sd_trimdone_monitor(dd, "Driver-reload"); + } + + /* Substitute our deduced value for was_reset */ + ret = ipath_ibsd_ucode_loaded(dd); + if (ret < 0) { + ret = 1; + goto done; + } + first_reset = !ret; /* First reset if IBSD uCode not yet loaded */ + + /* + * Alter some regs per vendor latest doc, reset-defaults + * are not right for IB. + */ + ret = ipath_sd_early(dd); + if (ret < 0) { + ipath_dev_err(dd, "Failed to set IB SERDES early defaults\n"); + ret = 1; + goto done; + } + + /* + * Set DAC manual trim IB. + * We only do this once after chip has been reset (usually + * same as once per system boot). + */ + if (first_reset) { + ret = ipath_sd_dactrim(dd); + if (ret < 0) { + ipath_dev_err(dd, "Failed IB SERDES DAC trim\n"); + ret = 1; + goto done; + } + } + + /* + * Set various registers (DDS and RXEQ) that will be + * controlled by IBC (in 1.2 mode) to reasonable preset values + * Calling the "internal" version avoids the "check for needed" + * and "trimdone monitor" that might be counter-productive. + */ + ret = ipath_internal_presets(dd); + if (ret < 0) { + ipath_dev_err(dd, "Failed to set IB SERDES presets\n"); + ret = 1; + goto done; + } + ret = ipath_sd_trimself(dd, 0x80); + if (ret < 0) { + ipath_dev_err(dd, "Failed to set IB SERDES TRIMSELF\n"); + ret = 1; + goto done; + } + + /* Load image, then try to verify */ + ret = 0; /* Assume success */ + if (first_reset) { + int vfy; + int trim_done; + ipath_dbg("SerDes uC was reset, reloading PRAM\n"); + ret = ipath_sd7220_ib_load(dd); + if (ret < 0) { + ipath_dev_err(dd, "Failed to load IB SERDES image\n"); + ret = 1; + goto done; + } + + /* Loaded image, try to verify */ + vfy = ipath_sd7220_ib_vfy(dd); + if (vfy != ret) { + ipath_dev_err(dd, "SERDES PRAM VFY failed\n"); + ret = 1; + goto done; + } + /* + * Loaded and verified. Almost good... + * hold "success" in ret + */ + ret = 0; + + /* + * Prev steps all worked, continue bringup + * De-assert RESET to uC, only in first reset, to allow + * trimming. + * + * Since our default setup sets START_EQ1 to + * PRESET, we need to clear that for this very first run. + */ + ret = ibsd_mod_allchnls(dd, START_EQ1(0), 0, 0x38); + if (ret < 0) { + ipath_dev_err(dd, "Failed clearing START_EQ1\n"); + ret = 1; + goto done; + } + + ipath_ibsd_reset(dd, 0); + /* + * If this is not the first reset, trimdone should be set + * already. + */ + trim_done = ipath_sd_trimdone_poll(dd); + /* + * Whether or not trimdone succeeded, we need to put the + * uC back into reset to avoid a possible fight with the + * IBC state-machine. + */ + ipath_ibsd_reset(dd, 1); + + if (!trim_done) { + ipath_dev_err(dd, "No TRIMDONE seen\n"); + ret = 1; + goto done; + } + + ipath_sd_trimdone_monitor(dd, "First-reset"); + /* Remember so we do not re-do the load, dactrim, etc. */ + dd->serdes_first_init_done = 1; + } + /* + * Setup for channel training and load values for + * RxEq and DDS in tables used by IBC in IB1.2 mode + */ + + val_stat = ipath_sd_setvals(dd); + if (val_stat < 0) + ret = 1; +done: + /* start relock timer regardless, but start at 1 second */ + ipath_set_relock_poll(dd, -1); + return ret; +} + +#define EPB_ACC_REQ 1 +#define EPB_ACC_GNT 0x100 +#define EPB_DATA_MASK 0xFF +#define EPB_RD (1ULL << 24) +#define EPB_TRANS_RDY (1ULL << 31) +#define EPB_TRANS_ERR (1ULL << 30) +#define EPB_TRANS_TRIES 5 + +/* + * query, claim, release ownership of the EPB (External Parallel Bus) + * for a specified SERDES. + * the "claim" parameter is >0 to claim, <0 to release, 0 to query. + * Returns <0 for errors, >0 if we had ownership, else 0. + */ +static int epb_access(struct ipath_devdata *dd, int sdnum, int claim) +{ + u16 acc; + u64 accval; + int owned = 0; + u64 oct_sel = 0; + + switch (sdnum) { + case IB_7220_SERDES : + /* + * The IB SERDES "ownership" is fairly simple. A single each + * request/grant. + */ + acc = dd->ipath_kregs->kr_ib_epbacc; + break; + case PCIE_SERDES0 : + case PCIE_SERDES1 : + /* PCIe SERDES has two "octants", need to select which */ + acc = dd->ipath_kregs->kr_pcie_epbacc; + oct_sel = (2 << (sdnum - PCIE_SERDES0)); + break; + default : + return 0; + } + + /* Make sure any outstanding transaction was seen */ + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + udelay(15); + + accval = ipath_read_kreg32(dd, acc); + + owned = !!(accval & EPB_ACC_GNT); + if (claim < 0) { + /* Need to release */ + u64 pollval; + /* + * The only writeable bits are the request and CS. + * Both should be clear + */ + u64 newval = 0; + ipath_write_kreg(dd, acc, newval); + /* First read after write is not trustworthy */ + pollval = ipath_read_kreg32(dd, acc); + udelay(5); + pollval = ipath_read_kreg32(dd, acc); + if (pollval & EPB_ACC_GNT) + owned = -1; + } else if (claim > 0) { + /* Need to claim */ + u64 pollval; + u64 newval = EPB_ACC_REQ | oct_sel; + ipath_write_kreg(dd, acc, newval); + /* First read after write is not trustworthy */ + pollval = ipath_read_kreg32(dd, acc); + udelay(5); + pollval = ipath_read_kreg32(dd, acc); + if (!(pollval & EPB_ACC_GNT)) + owned = -1; + } + return owned; +} + +/* + * Lemma to deal with race condition of write..read to epb regs + */ +static int epb_trans(struct ipath_devdata *dd, u16 reg, u64 i_val, u64 *o_vp) +{ + int tries; + u64 transval; + + + ipath_write_kreg(dd, reg, i_val); + /* Throw away first read, as RDY bit may be stale */ + transval = ipath_read_kreg64(dd, reg); + + for (tries = EPB_TRANS_TRIES; tries; --tries) { + transval = ipath_read_kreg32(dd, reg); + if (transval & EPB_TRANS_RDY) + break; + udelay(5); + } + if (transval & EPB_TRANS_ERR) + return -1; + if (tries > 0 && o_vp) + *o_vp = transval; + return tries; +} + +/** + * + * ipath_sd7220_reg_mod - modify SERDES register + * @dd: the infinipath device + * @sdnum: which SERDES to access + * @loc: location - channel, element, register, as packed by EPB_LOC() macro. + * @wd: Write Data - value to set in register + * @mask: ones where data should be spliced into reg. + * + * Basic register read/modify/write, with un-needed acesses elided. That is, + * a mask of zero will prevent write, while a mask of 0xFF will prevent read. + * returns current (presumed, if a write was done) contents of selected + * register, or <0 if errors. + */ +static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc, + u32 wd, u32 mask) +{ + u16 trans; + u64 transval; + int owned; + int tries, ret; + unsigned long flags; + + switch (sdnum) { + case IB_7220_SERDES : + trans = dd->ipath_kregs->kr_ib_epbtrans; + break; + case PCIE_SERDES0 : + case PCIE_SERDES1 : + trans = dd->ipath_kregs->kr_pcie_epbtrans; + break; + default : + return -1; + } + + /* + * All access is locked in software (vs other host threads) and + * hardware (vs uC access). + */ + spin_lock_irqsave(&dd->ipath_sdepb_lock, flags); + + owned = epb_access(dd, sdnum, 1); + if (owned < 0) { + spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags); + return -1; + } + ret = 0; + for (tries = EPB_TRANS_TRIES; tries; --tries) { + transval = ipath_read_kreg32(dd, trans); + if (transval & EPB_TRANS_RDY) + break; + udelay(5); + } + + if (tries > 0) { + tries = 1; /* to make read-skip work */ + if (mask != 0xFF) { + /* + * Not a pure write, so need to read. + * loc encodes chip-select as well as address + */ + transval = loc | EPB_RD; + tries = epb_trans(dd, trans, transval, &transval); + } + if (tries > 0 && mask != 0) { + /* + * Not a pure read, so need to write. + */ + wd = (wd & mask) | (transval & ~mask); + transval = loc | (wd & EPB_DATA_MASK); + tries = epb_trans(dd, trans, transval, &transval); + } + } + /* else, failed to see ready, what error-handling? */ + + /* + * Release bus. Failure is an error. + */ + if (epb_access(dd, sdnum, -1) < 0) + ret = -1; + else + ret = transval & EPB_DATA_MASK; + + spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags); + if (tries <= 0) + ret = -1; + return ret; +} + +#define EPB_ROM_R (2) +#define EPB_ROM_W (1) +/* + * Below, all uC-related, use appropriate UC_CS, depending + * on which SerDes is used. + */ +#define EPB_UC_CTL EPB_LOC(6, 0, 0) +#define EPB_MADDRL EPB_LOC(6, 0, 2) +#define EPB_MADDRH EPB_LOC(6, 0, 3) +#define EPB_ROMDATA EPB_LOC(6, 0, 4) +#define EPB_RAMDATA EPB_LOC(6, 0, 5) + +/* Transfer date to/from uC Program RAM of IB or PCIe SerDes */ +static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc, + u8 *buf, int cnt, int rd_notwr) +{ + u16 trans; + u64 transval; + u64 csbit; + int owned; + int tries; + int sofar; + int addr; + int ret; + unsigned long flags; + const char *op; + + /* Pick appropriate transaction reg and "Chip select" for this serdes */ + switch (sdnum) { + case IB_7220_SERDES : + csbit = 1ULL << EPB_IB_UC_CS_SHF; + trans = dd->ipath_kregs->kr_ib_epbtrans; + break; + case PCIE_SERDES0 : + case PCIE_SERDES1 : + /* PCIe SERDES has uC "chip select" in different bit, too */ + csbit = 1ULL << EPB_PCIE_UC_CS_SHF; + trans = dd->ipath_kregs->kr_pcie_epbtrans; + break; + default : + return -1; + } + + op = rd_notwr ? "Rd" : "Wr"; + spin_lock_irqsave(&dd->ipath_sdepb_lock, flags); + + owned = epb_access(dd, sdnum, 1); + if (owned < 0) { + spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags); + ipath_dbg("Could not get %s access to %s EPB: %X, loc %X\n", + op, (sdnum == IB_7220_SERDES) ? "IB" : "PCIe", + owned, loc); + return -1; + } + + /* + * In future code, we may need to distinguish several address ranges, + * and select various memories based on this. For now, just trim + * "loc" (location including address and memory select) to + * "addr" (address within memory). we will only support PRAM + * The memory is 8KB. + */ + addr = loc & 0x1FFF; + for (tries = EPB_TRANS_TRIES; tries; --tries) { + transval = ipath_read_kreg32(dd, trans); + if (transval & EPB_TRANS_RDY) + break; + udelay(5); + } + + sofar = 0; + if (tries <= 0) + ipath_dbg("No initial RDY on EPB access request\n"); + else { + /* + * Every "memory" access is doubly-indirect. + * We set two bytes of address, then read/write + * one or mores bytes of data. + */ + + /* First, we set control to "Read" or "Write" */ + transval = csbit | EPB_UC_CTL | + (rd_notwr ? EPB_ROM_R : EPB_ROM_W); + tries = epb_trans(dd, trans, transval, &transval); + if (tries <= 0) + ipath_dbg("No EPB response to uC %s cmd\n", op); + while (tries > 0 && sofar < cnt) { + if (!sofar) { + /* Only set address at start of chunk */ + int addrbyte = (addr + sofar) >> 8; + transval = csbit | EPB_MADDRH | addrbyte; + tries = epb_trans(dd, trans, transval, + &transval); + if (tries <= 0) { + ipath_dbg("No EPB response ADDRH\n"); + break; + } + addrbyte = (addr + sofar) & 0xFF; + transval = csbit | EPB_MADDRL | addrbyte; + tries = epb_trans(dd, trans, transval, + &transval); + if (tries <= 0) { + ipath_dbg("No EPB response ADDRL\n"); + break; + } + } + + if (rd_notwr) + transval = csbit | EPB_ROMDATA | EPB_RD; + else + transval = csbit | EPB_ROMDATA | buf[sofar]; + tries = epb_trans(dd, trans, transval, &transval); + if (tries <= 0) { + ipath_dbg("No EPB response DATA\n"); + break; + } + if (rd_notwr) + buf[sofar] = transval & EPB_DATA_MASK; + ++sofar; + } + /* Finally, clear control-bit for Read or Write */ + transval = csbit | EPB_UC_CTL; + tries = epb_trans(dd, trans, transval, &transval); + if (tries <= 0) + ipath_dbg("No EPB response to drop of uC %s cmd\n", op); + } + + ret = sofar; + /* Release bus. Failure is an error */ + if (epb_access(dd, sdnum, -1) < 0) + ret = -1; + + spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags); + if (tries <= 0) { + ipath_dbg("SERDES PRAM %s failed after %d bytes\n", op, sofar); + ret = -1; + } + return ret; +} + +#define PROG_CHUNK 64 + +int ipath_sd7220_prog_ld(struct ipath_devdata *dd, int sdnum, + u8 *img, int len, int offset) +{ + int cnt, sofar, req; + + sofar = 0; + while (sofar < len) { + req = len - sofar; + if (req > PROG_CHUNK) + req = PROG_CHUNK; + cnt = ipath_sd7220_ram_xfer(dd, sdnum, offset + sofar, + img + sofar, req, 0); + if (cnt < req) { + sofar = -1; + break; + } + sofar += req; + } + return sofar; +} + +#define VFY_CHUNK 64 +#define SD_PRAM_ERROR_LIMIT 42 + +int ipath_sd7220_prog_vfy(struct ipath_devdata *dd, int sdnum, + const u8 *img, int len, int offset) +{ + int cnt, sofar, req, idx, errors; + unsigned char readback[VFY_CHUNK]; + + errors = 0; + sofar = 0; + while (sofar < len) { + req = len - sofar; + if (req > VFY_CHUNK) + req = VFY_CHUNK; + cnt = ipath_sd7220_ram_xfer(dd, sdnum, sofar + offset, + readback, req, 1); + if (cnt < req) { + /* failed in read itself */ + sofar = -1; + break; + } + for (idx = 0; idx < cnt; ++idx) { + if (readback[idx] != img[idx+sofar]) + ++errors; + } + sofar += cnt; + } + return errors ? -errors : sofar; +} + +/* IRQ not set up at this point in init, so we poll. */ +#define IB_SERDES_TRIM_DONE (1ULL << 11) +#define TRIM_TMO (30) + +static int ipath_sd_trimdone_poll(struct ipath_devdata *dd) +{ + int trim_tmo, ret; + uint64_t val; + + /* + * Default to failure, so IBC will not start + * without IB_SERDES_TRIM_DONE. + */ + ret = 0; + for (trim_tmo = 0; trim_tmo < TRIM_TMO; ++trim_tmo) { + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); + if (val & IB_SERDES_TRIM_DONE) { + ipath_cdbg(VERBOSE, "TRIMDONE after %d\n", trim_tmo); + ret = 1; + break; + } + msleep(10); + } + if (trim_tmo >= TRIM_TMO) { + ipath_dev_err(dd, "No TRIMDONE in %d tries\n", trim_tmo); + ret = 0; + } + return ret; +} + +#define TX_FAST_ELT (9) + +/* + * Set the "negotiation" values for SERDES. These are used by the IB1.2 + * link negotiation. Macros below are attempt to keep the values a + * little more human-editable. + * First, values related to Drive De-emphasis Settings. + */ + +#define NUM_DDS_REGS 6 +#define DDS_REG_MAP 0x76A910 /* LSB-first list of regs (in elt 9) to mod */ + +#define DDS_VAL(amp_d, main_d, ipst_d, ipre_d, amp_s, main_s, ipst_s, ipre_s) \ + { { ((amp_d & 0x1F) << 1) | 1, ((amp_s & 0x1F) << 1) | 1, \ + (main_d << 3) | 4 | (ipre_d >> 2), \ + (main_s << 3) | 4 | (ipre_s >> 2), \ + ((ipst_d & 0xF) << 1) | ((ipre_d & 3) << 6) | 0x21, \ + ((ipst_s & 0xF) << 1) | ((ipre_s & 3) << 6) | 0x21 } } + +static struct dds_init { + uint8_t reg_vals[NUM_DDS_REGS]; +} dds_init_vals[] = { + /* DDR(FDR) SDR(HDR) */ + /* Vendor recommends below for 3m cable */ +#define DDS_3M 0 + DDS_VAL(31, 19, 12, 0, 29, 22, 9, 0), + DDS_VAL(31, 12, 15, 4, 31, 15, 15, 1), + DDS_VAL(31, 13, 15, 3, 31, 16, 15, 0), + DDS_VAL(31, 14, 15, 2, 31, 17, 14, 0), + DDS_VAL(31, 15, 15, 1, 31, 18, 13, 0), + DDS_VAL(31, 16, 15, 0, 31, 19, 12, 0), + DDS_VAL(31, 17, 14, 0, 31, 20, 11, 0), + DDS_VAL(31, 18, 13, 0, 30, 21, 10, 0), + DDS_VAL(31, 20, 11, 0, 28, 23, 8, 0), + DDS_VAL(31, 21, 10, 0, 27, 24, 7, 0), + DDS_VAL(31, 22, 9, 0, 26, 25, 6, 0), + DDS_VAL(30, 23, 8, 0, 25, 26, 5, 0), + DDS_VAL(29, 24, 7, 0, 23, 27, 4, 0), + /* Vendor recommends below for 1m cable */ +#define DDS_1M 13 + DDS_VAL(28, 25, 6, 0, 21, 28, 3, 0), + DDS_VAL(27, 26, 5, 0, 19, 29, 2, 0), + DDS_VAL(25, 27, 4, 0, 17, 30, 1, 0) +}; + +/* + * Next, values related to Receive Equalization. + * In comments, FDR (Full) is IB DDR, HDR (Half) is IB SDR + */ +/* Hardware packs an element number and register address thus: */ +#define RXEQ_INIT_RDESC(elt, addr) (((elt) & 0xF) | ((addr) << 4)) +#define RXEQ_VAL(elt, adr, val0, val1, val2, val3) \ + {RXEQ_INIT_RDESC((elt), (adr)), {(val0), (val1), (val2), (val3)} } + +#define RXEQ_VAL_ALL(elt, adr, val) \ + {RXEQ_INIT_RDESC((elt), (adr)), {(val), (val), (val), (val)} } + +#define RXEQ_SDR_DFELTH 0 +#define RXEQ_SDR_TLTH 0 +#define RXEQ_SDR_G1CNT_Z1CNT 0x11 +#define RXEQ_SDR_ZCNT 23 + +static struct rxeq_init { + u16 rdesc; /* in form used in SerDesDDSRXEQ */ + u8 rdata[4]; +} rxeq_init_vals[] = { + /* Set Rcv Eq. to Preset node */ + RXEQ_VAL_ALL(7, 0x27, 0x10), + /* Set DFELTHFDR/HDR thresholds */ + RXEQ_VAL(7, 8, 0, 0, 0, 0), /* FDR */ + RXEQ_VAL(7, 0x21, 0, 0, 0, 0), /* HDR */ + /* Set TLTHFDR/HDR theshold */ + RXEQ_VAL(7, 9, 2, 2, 2, 2), /* FDR */ + RXEQ_VAL(7, 0x23, 2, 2, 2, 2), /* HDR */ + /* Set Preamp setting 2 (ZFR/ZCNT) */ + RXEQ_VAL(7, 0x1B, 12, 12, 12, 12), /* FDR */ + RXEQ_VAL(7, 0x1C, 12, 12, 12, 12), /* HDR */ + /* Set Preamp DC gain and Setting 1 (GFR/GHR) */ + RXEQ_VAL(7, 0x1E, 0x10, 0x10, 0x10, 0x10), /* FDR */ + RXEQ_VAL(7, 0x1F, 0x10, 0x10, 0x10, 0x10), /* HDR */ + /* Toggle RELOCK (in VCDL_CTRL0) to lock to data */ + RXEQ_VAL_ALL(6, 6, 0x20), /* Set D5 High */ + RXEQ_VAL_ALL(6, 6, 0), /* Set D5 Low */ +}; + +/* There are 17 values from vendor, but IBC only accesses the first 16 */ +#define DDS_ROWS (16) +#define RXEQ_ROWS ARRAY_SIZE(rxeq_init_vals) + +static int ipath_sd_setvals(struct ipath_devdata *dd) +{ + int idx, midx; + int min_idx; /* Minimum index for this portion of table */ + uint32_t dds_reg_map; + u64 __iomem *taddr, *iaddr; + uint64_t data; + uint64_t sdctl; + + taddr = dd->ipath_kregbase + KR_IBSerDesMappTable; + iaddr = dd->ipath_kregbase + dd->ipath_kregs->kr_ib_ddsrxeq; + + /* + * Init the DDS section of the table. + * Each "row" of the table provokes NUM_DDS_REG writes, to the + * registers indicated in DDS_REG_MAP. + */ + sdctl = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibserdesctrl); + sdctl = (sdctl & ~(0x1f << 8)) | (NUM_DDS_REGS << 8); + sdctl = (sdctl & ~(0x1f << 13)) | (RXEQ_ROWS << 13); + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibserdesctrl, sdctl); + + /* + * Iterate down table within loop for each register to store. + */ + dds_reg_map = DDS_REG_MAP; + for (idx = 0; idx < NUM_DDS_REGS; ++idx) { + data = ((dds_reg_map & 0xF) << 4) | TX_FAST_ELT; + writeq(data, iaddr + idx); + mmiowb(); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + dds_reg_map >>= 4; + for (midx = 0; midx < DDS_ROWS; ++midx) { + u64 __iomem *daddr = taddr + ((midx << 4) + idx); + data = dds_init_vals[midx].reg_vals[idx]; + writeq(data, daddr); + mmiowb(); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + } /* End inner for (vals for this reg, each row) */ + } /* end outer for (regs to be stored) */ + + /* + * Init the RXEQ section of the table. As explained above the table + * rxeq_init_vals[], this runs in a different order, as the pattern + * of register references is more complex, but there are only + * four "data" values per register. + */ + min_idx = idx; /* RXEQ indices pick up where DDS left off */ + taddr += 0x100; /* RXEQ data is in second half of table */ + /* Iterate through RXEQ register addresses */ + for (idx = 0; idx < RXEQ_ROWS; ++idx) { + int didx; /* "destination" */ + int vidx; + + /* didx is offset by min_idx to address RXEQ range of regs */ + didx = idx + min_idx; + /* Store the next RXEQ register address */ + writeq(rxeq_init_vals[idx].rdesc, iaddr + didx); + mmiowb(); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + /* Iterate through RXEQ values */ + for (vidx = 0; vidx < 4; vidx++) { + data = rxeq_init_vals[idx].rdata[vidx]; + writeq(data, taddr + (vidx << 6) + idx); + mmiowb(); + ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch); + } + } /* end outer for (Reg-writes for RXEQ) */ + return 0; +} + +#define CMUCTRL5 EPB_LOC(7, 0, 0x15) +#define RXHSCTRL0(chan) EPB_LOC(chan, 6, 0) +#define VCDL_DAC2(chan) EPB_LOC(chan, 6, 5) +#define VCDL_CTRL0(chan) EPB_LOC(chan, 6, 6) +#define VCDL_CTRL2(chan) EPB_LOC(chan, 6, 8) +#define START_EQ2(chan) EPB_LOC(chan, 7, 0x28) + +static int ibsd_sto_noisy(struct ipath_devdata *dd, int loc, int val, int mask) +{ + int ret = -1; + int sloc; /* shifted loc, for messages */ + + loc |= (1U << EPB_IB_QUAD0_CS_SHF); + sloc = loc >> EPB_ADDR_SHF; + + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, mask); + if (ret < 0) + ipath_dev_err(dd, "Write failed: elt %d," + " addr 0x%X, chnl %d, val 0x%02X, mask 0x%02X\n", + (sloc & 0xF), (sloc >> 9) & 0x3f, (sloc >> 4) & 7, + val & 0xFF, mask & 0xFF); + return ret; +} + +/* + * Repeat a "store" across all channels of the IB SerDes. + * Although nominally it inherits the "read value" of the last + * channel it modified, the only really useful return is <0 for + * failure, >= 0 for success. The parameter 'loc' is assumed to + * be the location for the channel-0 copy of the register to + * be modified. + */ +static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val, + int mask) +{ + int ret = -1; + int chnl; + + if (loc & EPB_GLOBAL_WR) { + /* + * Our caller has assured us that we can set all four + * channels at once. Trust that. If mask is not 0xFF, + * we will read the _specified_ channel for our starting + * value. + */ + loc |= (1U << EPB_IB_QUAD0_CS_SHF); + chnl = (loc >> (4 + EPB_ADDR_SHF)) & 7; + if (mask != 0xFF) { + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, + loc & ~EPB_GLOBAL_WR, 0, 0); + if (ret < 0) { + int sloc = loc >> EPB_ADDR_SHF; + ipath_dev_err(dd, "pre-read failed: elt %d," + " addr 0x%X, chnl %d\n", (sloc & 0xF), + (sloc >> 9) & 0x3f, chnl); + return ret; + } + val = (ret & ~mask) | (val & mask); + } + loc &= ~(7 << (4+EPB_ADDR_SHF)); + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, 0xFF); + if (ret < 0) { + int sloc = loc >> EPB_ADDR_SHF; + ipath_dev_err(dd, "Global WR failed: elt %d," + " addr 0x%X, val %02X\n", + (sloc & 0xF), (sloc >> 9) & 0x3f, val); + } + return ret; + } + /* Clear "channel" and set CS so we can simply iterate */ + loc &= ~(7 << (4+EPB_ADDR_SHF)); + loc |= (1U << EPB_IB_QUAD0_CS_SHF); + for (chnl = 0; chnl < 4; ++chnl) { + int cloc; + cloc = loc | (chnl << (4+EPB_ADDR_SHF)); + ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, cloc, val, mask); + if (ret < 0) { + int sloc = loc >> EPB_ADDR_SHF; + ipath_dev_err(dd, "Write failed: elt %d," + " addr 0x%X, chnl %d, val 0x%02X," + " mask 0x%02X\n", + (sloc & 0xF), (sloc >> 9) & 0x3f, chnl, + val & 0xFF, mask & 0xFF); + break; + } + } + return ret; +} + +/* + * Set the Tx values normally modified by IBC in IB1.2 mode to default + * values, as gotten from first row of init table. + */ +static int set_dds_vals(struct ipath_devdata *dd, struct dds_init *ddi) +{ + int ret; + int idx, reg, data; + uint32_t regmap; + + regmap = DDS_REG_MAP; + for (idx = 0; idx < NUM_DDS_REGS; ++idx) { + reg = (regmap & 0xF); + regmap >>= 4; + data = ddi->reg_vals[idx]; + /* Vendor says RMW not needed for these regs, use 0xFF mask */ + ret = ibsd_mod_allchnls(dd, EPB_LOC(0, 9, reg), data, 0xFF); + if (ret < 0) + break; + } + return ret; +} + +/* + * Set the Rx values normally modified by IBC in IB1.2 mode to default + * values, as gotten from selected column of init table. + */ +static int set_rxeq_vals(struct ipath_devdata *dd, int vsel) +{ + int ret; + int ridx; + int cnt = ARRAY_SIZE(rxeq_init_vals); + + for (ridx = 0; ridx < cnt; ++ridx) { + int elt, reg, val, loc; + elt = rxeq_init_vals[ridx].rdesc & 0xF; + reg = rxeq_init_vals[ridx].rdesc >> 4; + loc = EPB_LOC(0, elt, reg); + val = rxeq_init_vals[ridx].rdata[vsel]; + /* mask of 0xFF, because hardware does full-byte store. */ + ret = ibsd_mod_allchnls(dd, loc, val, 0xFF); + if (ret < 0) + break; + } + return ret; +} + +/* + * Set the default values (row 0) for DDR Driver Demphasis. + * we do this initially and whenever we turn off IB-1.2 + * The "default" values for Rx equalization are also stored to + * SerDes registers. Formerly (and still default), we used set 2. + * For experimenting with cables and link-partners, we allow changing + * that via a module parameter. + */ +static unsigned ipath_rxeq_set = 2; +module_param_named(rxeq_default_set, ipath_rxeq_set, uint, + S_IWUSR | S_IRUGO); +MODULE_PARM_DESC(rxeq_default_set, + "Which set [0..3] of Rx Equalization values is default"); + +static int ipath_internal_presets(struct ipath_devdata *dd) +{ + int ret = 0; + + ret = set_dds_vals(dd, dds_init_vals + DDS_3M); + + if (ret < 0) + ipath_dev_err(dd, "Failed to set default DDS values\n"); + ret = set_rxeq_vals(dd, ipath_rxeq_set & 3); + if (ret < 0) + ipath_dev_err(dd, "Failed to set default RXEQ values\n"); + return ret; +} + +int ipath_sd7220_presets(struct ipath_devdata *dd) +{ + int ret = 0; + + if (!dd->ipath_presets_needed) + return ret; + dd->ipath_presets_needed = 0; + /* Assert uC reset, so we don't clash with it. */ + ipath_ibsd_reset(dd, 1); + udelay(2); + ipath_sd_trimdone_monitor(dd, "link-down"); + + ret = ipath_internal_presets(dd); +return ret; +} + +static int ipath_sd_trimself(struct ipath_devdata *dd, int val) +{ + return ibsd_sto_noisy(dd, CMUCTRL5, val, 0xFF); +} + +static int ipath_sd_early(struct ipath_devdata *dd) +{ + int ret = -1; /* Default failed */ + int chnl; + + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, RXHSCTRL0(chnl), 0xD4, 0xFF); + if (ret < 0) + goto bail; + } + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, VCDL_DAC2(chnl), 0x2D, 0xFF); + if (ret < 0) + goto bail; + } + /* more fine-tuning of what will be default */ + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, VCDL_CTRL2(chnl), 3, 0xF); + if (ret < 0) + goto bail; + } + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, START_EQ1(chnl), 0x10, 0xFF); + if (ret < 0) + goto bail; + } + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, START_EQ2(chnl), 0x30, 0xFF); + if (ret < 0) + goto bail; + } +bail: + return ret; +} + +#define BACTRL(chnl) EPB_LOC(chnl, 6, 0x0E) +#define LDOUTCTRL1(chnl) EPB_LOC(chnl, 7, 6) +#define RXHSSTATUS(chnl) EPB_LOC(chnl, 6, 0xF) + +static int ipath_sd_dactrim(struct ipath_devdata *dd) +{ + int ret = -1; /* Default failed */ + int chnl; + + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, BACTRL(chnl), 0x40, 0xFF); + if (ret < 0) + goto bail; + } + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, LDOUTCTRL1(chnl), 0x04, 0xFF); + if (ret < 0) + goto bail; + } + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, RXHSSTATUS(chnl), 0x04, 0xFF); + if (ret < 0) + goto bail; + } + /* + * delay for max possible number of steps, with slop. + * Each step is about 4usec. + */ + udelay(415); + for (chnl = 0; chnl < 4; ++chnl) { + ret = ibsd_sto_noisy(dd, LDOUTCTRL1(chnl), 0x00, 0xFF); + if (ret < 0) + goto bail; + } +bail: + return ret; +} + +#define RELOCK_FIRST_MS 3 +#define RXLSPPM(chan) EPB_LOC(chan, 0, 2) +void ipath_toggle_rclkrls(struct ipath_devdata *dd) +{ + int loc = RXLSPPM(0) | EPB_GLOBAL_WR; + int ret; + + ret = ibsd_mod_allchnls(dd, loc, 0, 0x80); + if (ret < 0) + ipath_dev_err(dd, "RCLKRLS failed to clear D7\n"); + else { + udelay(1); + ibsd_mod_allchnls(dd, loc, 0x80, 0x80); + } + /* And again for good measure */ + udelay(1); + ret = ibsd_mod_allchnls(dd, loc, 0, 0x80); + if (ret < 0) + ipath_dev_err(dd, "RCLKRLS failed to clear D7\n"); + else { + udelay(1); + ibsd_mod_allchnls(dd, loc, 0x80, 0x80); + } + /* Now reset xgxs and IBC to complete the recovery */ + dd->ipath_f_xgxs_reset(dd); +} + +/* + * Shut down the timer that polls for relock occasions, if needed + * this is "hooked" from ipath_7220_quiet_serdes(), which is called + * just before ipath_shutdown_device() in ipath_driver.c shuts down all + * the other timers + */ +void ipath_shutdown_relock_poll(struct ipath_devdata *dd) +{ + struct ipath_relock *irp = &dd->ipath_relock_singleton; + if (atomic_read(&irp->ipath_relock_timer_active)) { + del_timer_sync(&irp->ipath_relock_timer); + atomic_set(&irp->ipath_relock_timer_active, 0); + } +} + +static unsigned ipath_relock_by_timer = 1; +module_param_named(relock_by_timer, ipath_relock_by_timer, uint, + S_IWUSR | S_IRUGO); +MODULE_PARM_DESC(relock_by_timer, "Allow relock attempt if link not up"); + +static void ipath_run_relock(unsigned long opaque) +{ + struct ipath_devdata *dd = (struct ipath_devdata *)opaque; + struct ipath_relock *irp = &dd->ipath_relock_singleton; + u64 val, ltstate; + + if (!(dd->ipath_flags & IPATH_INITTED)) { + /* Not yet up, just reenable the timer for later */ + irp->ipath_relock_interval = HZ; + mod_timer(&irp->ipath_relock_timer, jiffies + HZ); + return; + } + + /* + * Check link-training state for "stuck" state. + * if found, try relock and schedule another try at + * exponentially growing delay, maxed at one second. + * if not stuck, our work is done. + */ + val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); + ltstate = ipath_ib_linktrstate(dd, val); + + if (ltstate <= INFINIPATH_IBCS_LT_STATE_CFGWAITRMT + && ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP) { + int timeoff; + /* Not up yet. Try again, if allowed by module-param */ + if (ipath_relock_by_timer) { + if (dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) + ipath_cdbg(VERBOSE, "Skip RELOCK in AUTONEG\n"); + else if (!(dd->ipath_flags & IPATH_IB_LINK_DISABLED)) { + ipath_cdbg(VERBOSE, "RELOCK\n"); + ipath_toggle_rclkrls(dd); + } + } + /* re-set timer for next check */ + timeoff = irp->ipath_relock_interval << 1; + if (timeoff > HZ) + timeoff = HZ; + irp->ipath_relock_interval = timeoff; + + mod_timer(&irp->ipath_relock_timer, jiffies + timeoff); + } else { + /* Up, so no more need to check so often */ + mod_timer(&irp->ipath_relock_timer, jiffies + HZ); + } +} + +void ipath_set_relock_poll(struct ipath_devdata *dd, int ibup) +{ + struct ipath_relock *irp = &dd->ipath_relock_singleton; + + if (ibup > 0) { + /* we are now up, so relax timer to 1 second interval */ + if (atomic_read(&irp->ipath_relock_timer_active)) + mod_timer(&irp->ipath_relock_timer, jiffies + HZ); + } else { + /* Transition to down, (re-)set timer to short interval. */ + int timeout; + timeout = (HZ * ((ibup == -1) ? 1000 : RELOCK_FIRST_MS))/1000; + if (timeout == 0) + timeout = 1; + /* If timer has not yet been started, do so. */ + if (atomic_inc_return(&irp->ipath_relock_timer_active) == 1) { + init_timer(&irp->ipath_relock_timer); + irp->ipath_relock_timer.function = ipath_run_relock; + irp->ipath_relock_timer.data = (unsigned long) dd; + irp->ipath_relock_interval = timeout; + irp->ipath_relock_timer.expires = jiffies + timeout; + add_timer(&irp->ipath_relock_timer); + } else { + irp->ipath_relock_interval = timeout; + mod_timer(&irp->ipath_relock_timer, jiffies + timeout); + atomic_dec(&irp->ipath_relock_timer_active); + } + } +} + diff --git a/drivers/infiniband/hw/ipath/ipath_sd7220_img.c b/drivers/infiniband/hw/ipath/ipath_sd7220_img.c new file mode 100644 index 00000000000..5ef59da9270 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_sd7220_img.c @@ -0,0 +1,1082 @@ +/* + * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +/* + * This file contains the memory image from the vendor, to be copied into + * the IB SERDES of the IBA7220 during initialization. + * The file also includes the two functions which use this image. + */ +#include <linux/pci.h> +#include <linux/delay.h> + +#include "ipath_kernel.h" +#include "ipath_registers.h" +#include "ipath_7220.h" + +static unsigned char ipath_sd7220_ib_img[] = { +/*0000*/0x02, 0x0A, 0x29, 0x02, 0x0A, 0x87, 0xE5, 0xE6, + 0x30, 0xE6, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, +/*0010*/0x00, 0xE5, 0xE2, 0x30, 0xE4, 0x04, 0x7E, 0x01, + 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x5F, 0x60, 0x08, +/*0020*/0x53, 0xF9, 0xF7, 0xE4, 0xF5, 0xFE, 0x80, 0x08, + 0x7F, 0x0A, 0x12, 0x17, 0x31, 0x12, 0x0E, 0xA2, +/*0030*/0x75, 0xFC, 0x08, 0xE4, 0xF5, 0xFD, 0xE5, 0xE7, + 0x20, 0xE7, 0x03, 0x43, 0xF9, 0x08, 0x22, 0x00, +/*0040*/0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x75, + 0x51, 0x01, 0xE4, 0xF5, 0x52, 0xF5, 0x53, 0xF5, +/*0050*/0x52, 0xF5, 0x7E, 0x7F, 0x04, 0x02, 0x04, 0x38, + 0xC2, 0x36, 0x05, 0x52, 0xE5, 0x52, 0xD3, 0x94, +/*0060*/0x0C, 0x40, 0x05, 0x75, 0x52, 0x01, 0xD2, 0x36, + 0x90, 0x07, 0x0C, 0x74, 0x07, 0xF0, 0xA3, 0x74, +/*0070*/0xFF, 0xF0, 0xE4, 0xF5, 0x0C, 0xA3, 0xF0, 0x90, + 0x07, 0x14, 0xF0, 0xA3, 0xF0, 0x75, 0x0B, 0x20, +/*0080*/0xF5, 0x09, 0xE4, 0xF5, 0x08, 0xE5, 0x08, 0xD3, + 0x94, 0x30, 0x40, 0x03, 0x02, 0x04, 0x04, 0x12, +/*0090*/0x00, 0x06, 0x15, 0x0B, 0xE5, 0x08, 0x70, 0x04, + 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x09, +/*00A0*/0x70, 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, + 0xEE, 0x5F, 0x60, 0x05, 0x12, 0x18, 0x71, 0xD2, +/*00B0*/0x35, 0x53, 0xE1, 0xF7, 0xE5, 0x08, 0x45, 0x09, + 0xFF, 0xE5, 0x0B, 0x25, 0xE0, 0x25, 0xE0, 0x24, +/*00C0*/0x83, 0xF5, 0x82, 0xE4, 0x34, 0x07, 0xF5, 0x83, + 0xEF, 0xF0, 0x85, 0xE2, 0x20, 0xE5, 0x52, 0xD3, +/*00D0*/0x94, 0x01, 0x40, 0x0D, 0x12, 0x19, 0xF3, 0xE0, + 0x54, 0xA0, 0x64, 0x40, 0x70, 0x03, 0x02, 0x03, +/*00E0*/0xFB, 0x53, 0xF9, 0xF8, 0x90, 0x94, 0x70, 0xE4, + 0xF0, 0xE0, 0xF5, 0x10, 0xAF, 0x09, 0x12, 0x1E, +/*00F0*/0xB3, 0xAF, 0x08, 0xEF, 0x44, 0x08, 0xF5, 0x82, + 0x75, 0x83, 0x80, 0xE0, 0xF5, 0x29, 0xEF, 0x44, +/*0100*/0x07, 0x12, 0x1A, 0x3C, 0xF5, 0x22, 0x54, 0x40, + 0xD3, 0x94, 0x00, 0x40, 0x1E, 0xE5, 0x29, 0x54, +/*0110*/0xF0, 0x70, 0x21, 0x12, 0x19, 0xF3, 0xE0, 0x44, + 0x80, 0xF0, 0xE5, 0x22, 0x54, 0x30, 0x65, 0x08, +/*0120*/0x70, 0x09, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xBF, + 0xF0, 0x80, 0x09, 0x12, 0x19, 0xF3, 0x74, 0x40, +/*0130*/0xF0, 0x02, 0x03, 0xFB, 0x12, 0x1A, 0x12, 0x75, + 0x83, 0xAE, 0x74, 0xFF, 0xF0, 0xAF, 0x08, 0x7E, +/*0140*/0x00, 0xEF, 0x44, 0x07, 0xF5, 0x82, 0xE0, 0xFD, + 0xE5, 0x0B, 0x25, 0xE0, 0x25, 0xE0, 0x24, 0x81, +/*0150*/0xF5, 0x82, 0xE4, 0x34, 0x07, 0xF5, 0x83, 0xED, + 0xF0, 0x90, 0x07, 0x0E, 0xE0, 0x04, 0xF0, 0xEF, +/*0160*/0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0x98, 0xE0, + 0xF5, 0x28, 0x12, 0x1A, 0x23, 0x40, 0x0C, 0x12, +/*0170*/0x19, 0xF3, 0xE0, 0x44, 0x01, 0x12, 0x1A, 0x32, + 0x02, 0x03, 0xF6, 0xAF, 0x08, 0x7E, 0x00, 0x74, +/*0180*/0x80, 0xCD, 0xEF, 0xCD, 0x8D, 0x82, 0xF5, 0x83, + 0xE0, 0x30, 0xE0, 0x0A, 0x12, 0x19, 0xF3, 0xE0, +/*0190*/0x44, 0x20, 0xF0, 0x02, 0x03, 0xFB, 0x12, 0x19, + 0xF3, 0xE0, 0x54, 0xDF, 0xF0, 0xEE, 0x44, 0xAE, +/*01A0*/0x12, 0x1A, 0x43, 0x30, 0xE4, 0x03, 0x02, 0x03, + 0xFB, 0x74, 0x9E, 0x12, 0x1A, 0x05, 0x20, 0xE0, +/*01B0*/0x03, 0x02, 0x03, 0xFB, 0x8F, 0x82, 0x8E, 0x83, + 0xE0, 0x20, 0xE0, 0x03, 0x02, 0x03, 0xFB, 0x12, +/*01C0*/0x19, 0xF3, 0xE0, 0x44, 0x10, 0xF0, 0xE5, 0xE3, + 0x20, 0xE7, 0x08, 0xE5, 0x08, 0x12, 0x1A, 0x3A, +/*01D0*/0x44, 0x04, 0xF0, 0xAF, 0x08, 0x7E, 0x00, 0xEF, + 0x12, 0x1A, 0x3A, 0x20, 0xE2, 0x34, 0x12, 0x19, +/*01E0*/0xF3, 0xE0, 0x44, 0x08, 0xF0, 0xE5, 0xE4, 0x30, + 0xE6, 0x04, 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, +/*01F0*/0xE5, 0x7E, 0xC3, 0x94, 0x04, 0x50, 0x04, 0x7C, + 0x01, 0x80, 0x02, 0x7C, 0x00, 0xEC, 0x4D, 0x60, +/*0200*/0x05, 0xC2, 0x35, 0x02, 0x03, 0xFB, 0xEE, 0x44, + 0xD2, 0x12, 0x1A, 0x43, 0x44, 0x40, 0xF0, 0x02, +/*0210*/0x03, 0xFB, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xF7, + 0xF0, 0x12, 0x1A, 0x12, 0x75, 0x83, 0xD2, 0xE0, +/*0220*/0x54, 0xBF, 0xF0, 0x90, 0x07, 0x14, 0xE0, 0x04, + 0xF0, 0xE5, 0x7E, 0x70, 0x03, 0x75, 0x7E, 0x01, +/*0230*/0xAF, 0x08, 0x7E, 0x00, 0x12, 0x1A, 0x23, 0x40, + 0x12, 0x12, 0x19, 0xF3, 0xE0, 0x44, 0x01, 0x12, +/*0240*/0x19, 0xF2, 0xE0, 0x54, 0x02, 0x12, 0x1A, 0x32, + 0x02, 0x03, 0xFB, 0x12, 0x19, 0xF3, 0xE0, 0x44, +/*0250*/0x02, 0x12, 0x19, 0xF2, 0xE0, 0x54, 0xFE, 0xF0, + 0xC2, 0x35, 0xEE, 0x44, 0x8A, 0x8F, 0x82, 0xF5, +/*0260*/0x83, 0xE0, 0xF5, 0x17, 0x54, 0x8F, 0x44, 0x40, + 0xF0, 0x74, 0x90, 0xFC, 0xE5, 0x08, 0x44, 0x07, +/*0270*/0xFD, 0xF5, 0x82, 0x8C, 0x83, 0xE0, 0x54, 0x3F, + 0x90, 0x07, 0x02, 0xF0, 0xE0, 0x54, 0xC0, 0x8D, +/*0280*/0x82, 0x8C, 0x83, 0xF0, 0x74, 0x92, 0x12, 0x1A, + 0x05, 0x90, 0x07, 0x03, 0x12, 0x1A, 0x19, 0x74, +/*0290*/0x82, 0x12, 0x1A, 0x05, 0x90, 0x07, 0x04, 0x12, + 0x1A, 0x19, 0x74, 0xB4, 0x12, 0x1A, 0x05, 0x90, +/*02A0*/0x07, 0x05, 0x12, 0x1A, 0x19, 0x74, 0x94, 0xFE, + 0xE5, 0x08, 0x44, 0x06, 0x12, 0x1A, 0x0A, 0xF5, +/*02B0*/0x10, 0x30, 0xE0, 0x04, 0xD2, 0x37, 0x80, 0x02, + 0xC2, 0x37, 0xE5, 0x10, 0x54, 0x7F, 0x8F, 0x82, +/*02C0*/0x8E, 0x83, 0xF0, 0x30, 0x44, 0x30, 0x12, 0x1A, + 0x03, 0x54, 0x80, 0xD3, 0x94, 0x00, 0x40, 0x04, +/*02D0*/0xD2, 0x39, 0x80, 0x02, 0xC2, 0x39, 0x8F, 0x82, + 0x8E, 0x83, 0xE0, 0x44, 0x80, 0xF0, 0x12, 0x1A, +/*02E0*/0x03, 0x54, 0x40, 0xD3, 0x94, 0x00, 0x40, 0x04, + 0xD2, 0x3A, 0x80, 0x02, 0xC2, 0x3A, 0x8F, 0x82, +/*02F0*/0x8E, 0x83, 0xE0, 0x44, 0x40, 0xF0, 0x74, 0x92, + 0xFE, 0xE5, 0x08, 0x44, 0x06, 0x12, 0x1A, 0x0A, +/*0300*/0x30, 0xE7, 0x04, 0xD2, 0x38, 0x80, 0x02, 0xC2, + 0x38, 0x8F, 0x82, 0x8E, 0x83, 0xE0, 0x54, 0x7F, +/*0310*/0xF0, 0x12, 0x1E, 0x46, 0xE4, 0xF5, 0x0A, 0x20, + 0x03, 0x02, 0x80, 0x03, 0x30, 0x43, 0x03, 0x12, +/*0320*/0x19, 0x95, 0x20, 0x02, 0x02, 0x80, 0x03, 0x30, + 0x42, 0x03, 0x12, 0x0C, 0x8F, 0x30, 0x30, 0x06, +/*0330*/0x12, 0x19, 0x95, 0x12, 0x0C, 0x8F, 0x12, 0x0D, + 0x47, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xFB, 0xF0, +/*0340*/0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40, 0x46, 0x43, + 0xE1, 0x08, 0x12, 0x19, 0xF3, 0xE0, 0x44, 0x04, +/*0350*/0xF0, 0xE5, 0xE4, 0x20, 0xE7, 0x2A, 0x12, 0x1A, + 0x12, 0x75, 0x83, 0xD2, 0xE0, 0x54, 0x08, 0xD3, +/*0360*/0x94, 0x00, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02, + 0x7F, 0x00, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40, +/*0370*/0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF, + 0x5E, 0x60, 0x05, 0x12, 0x1D, 0xD7, 0x80, 0x17, +/*0380*/0x12, 0x1A, 0x12, 0x75, 0x83, 0xD2, 0xE0, 0x44, + 0x08, 0xF0, 0x02, 0x03, 0xFB, 0x12, 0x1A, 0x12, +/*0390*/0x75, 0x83, 0xD2, 0xE0, 0x54, 0xF7, 0xF0, 0x12, + 0x1E, 0x46, 0x7F, 0x08, 0x12, 0x17, 0x31, 0x74, +/*03A0*/0x8E, 0xFE, 0x12, 0x1A, 0x12, 0x8E, 0x83, 0xE0, + 0xF5, 0x10, 0x54, 0xFE, 0xF0, 0xE5, 0x10, 0x44, +/*03B0*/0x01, 0xFF, 0xE5, 0x08, 0xFD, 0xED, 0x44, 0x07, + 0xF5, 0x82, 0xEF, 0xF0, 0xE5, 0x10, 0x54, 0xFE, +/*03C0*/0xFF, 0xED, 0x44, 0x07, 0xF5, 0x82, 0xEF, 0x12, + 0x1A, 0x11, 0x75, 0x83, 0x86, 0xE0, 0x44, 0x10, +/*03D0*/0x12, 0x1A, 0x11, 0xE0, 0x44, 0x10, 0xF0, 0x12, + 0x19, 0xF3, 0xE0, 0x54, 0xFD, 0x44, 0x01, 0xFF, +/*03E0*/0x12, 0x19, 0xF3, 0xEF, 0x12, 0x1A, 0x32, 0x30, + 0x32, 0x0C, 0xE5, 0x08, 0x44, 0x08, 0xF5, 0x82, +/*03F0*/0x75, 0x83, 0x82, 0x74, 0x05, 0xF0, 0xAF, 0x0B, + 0x12, 0x18, 0xD7, 0x74, 0x10, 0x25, 0x08, 0xF5, +/*0400*/0x08, 0x02, 0x00, 0x85, 0x05, 0x09, 0xE5, 0x09, + 0xD3, 0x94, 0x07, 0x50, 0x03, 0x02, 0x00, 0x82, +/*0410*/0xE5, 0x7E, 0xD3, 0x94, 0x00, 0x40, 0x04, 0x7F, + 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x7E, 0xC3, +/*0420*/0x94, 0xFA, 0x50, 0x04, 0x7E, 0x01, 0x80, 0x02, + 0x7E, 0x00, 0xEE, 0x5F, 0x60, 0x02, 0x05, 0x7E, +/*0430*/0x30, 0x35, 0x0B, 0x43, 0xE1, 0x01, 0x7F, 0x09, + 0x12, 0x17, 0x31, 0x02, 0x00, 0x58, 0x53, 0xE1, +/*0440*/0xFE, 0x02, 0x00, 0x58, 0x8E, 0x6A, 0x8F, 0x6B, + 0x8C, 0x6C, 0x8D, 0x6D, 0x75, 0x6E, 0x01, 0x75, +/*0450*/0x6F, 0x01, 0x75, 0x70, 0x01, 0xE4, 0xF5, 0x73, + 0xF5, 0x74, 0xF5, 0x75, 0x90, 0x07, 0x2F, 0xF0, +/*0460*/0xF5, 0x3C, 0xF5, 0x3E, 0xF5, 0x46, 0xF5, 0x47, + 0xF5, 0x3D, 0xF5, 0x3F, 0xF5, 0x6F, 0xE5, 0x6F, +/*0470*/0x70, 0x0F, 0xE5, 0x6B, 0x45, 0x6A, 0x12, 0x07, + 0x2A, 0x75, 0x83, 0x80, 0x74, 0x3A, 0xF0, 0x80, +/*0480*/0x09, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x80, 0x74, + 0x1A, 0xF0, 0xE4, 0xF5, 0x6E, 0xC3, 0x74, 0x3F, +/*0490*/0x95, 0x6E, 0xFF, 0x12, 0x08, 0x65, 0x75, 0x83, + 0x82, 0xEF, 0xF0, 0x12, 0x1A, 0x4D, 0x12, 0x08, +/*04A0*/0xC6, 0xE5, 0x33, 0xF0, 0x12, 0x08, 0xFA, 0x12, + 0x08, 0xB1, 0x40, 0xE1, 0xE5, 0x6F, 0x70, 0x0B, +/*04B0*/0x12, 0x07, 0x2A, 0x75, 0x83, 0x80, 0x74, 0x36, + 0xF0, 0x80, 0x09, 0x12, 0x07, 0x2A, 0x75, 0x83, +/*04C0*/0x80, 0x74, 0x16, 0xF0, 0x75, 0x6E, 0x01, 0x12, + 0x07, 0x2A, 0x75, 0x83, 0xB4, 0xE5, 0x6E, 0xF0, +/*04D0*/0x12, 0x1A, 0x4D, 0x74, 0x3F, 0x25, 0x6E, 0xF5, + 0x82, 0xE4, 0x34, 0x00, 0xF5, 0x83, 0xE5, 0x33, +/*04E0*/0xF0, 0x74, 0xBF, 0x25, 0x6E, 0xF5, 0x82, 0xE4, + 0x34, 0x00, 0x12, 0x08, 0xB1, 0x40, 0xD8, 0xE4, +/*04F0*/0xF5, 0x70, 0xF5, 0x46, 0xF5, 0x47, 0xF5, 0x6E, + 0x12, 0x08, 0xFA, 0xF5, 0x83, 0xE0, 0xFE, 0x12, +/*0500*/0x08, 0xC6, 0xE0, 0x7C, 0x00, 0x24, 0x00, 0xFF, + 0xEC, 0x3E, 0xFE, 0xAD, 0x3B, 0xD3, 0xEF, 0x9D, +/*0510*/0xEE, 0x9C, 0x50, 0x04, 0x7B, 0x01, 0x80, 0x02, + 0x7B, 0x00, 0xE5, 0x70, 0x70, 0x04, 0x7A, 0x01, +/*0520*/0x80, 0x02, 0x7A, 0x00, 0xEB, 0x5A, 0x60, 0x06, + 0x85, 0x6E, 0x46, 0x75, 0x70, 0x01, 0xD3, 0xEF, +/*0530*/0x9D, 0xEE, 0x9C, 0x50, 0x04, 0x7F, 0x01, 0x80, + 0x02, 0x7F, 0x00, 0xE5, 0x70, 0xB4, 0x01, 0x04, +/*0540*/0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF, 0x5E, + 0x60, 0x03, 0x85, 0x6E, 0x47, 0x05, 0x6E, 0xE5, +/*0550*/0x6E, 0x64, 0x7F, 0x70, 0xA3, 0xE5, 0x46, 0x60, + 0x05, 0xE5, 0x47, 0xB4, 0x7E, 0x03, 0x85, 0x46, +/*0560*/0x47, 0xE5, 0x6F, 0x70, 0x08, 0x85, 0x46, 0x76, + 0x85, 0x47, 0x77, 0x80, 0x0E, 0xC3, 0x74, 0x7F, +/*0570*/0x95, 0x46, 0xF5, 0x78, 0xC3, 0x74, 0x7F, 0x95, + 0x47, 0xF5, 0x79, 0xE5, 0x6F, 0x70, 0x37, 0xE5, +/*0580*/0x46, 0x65, 0x47, 0x70, 0x0C, 0x75, 0x73, 0x01, + 0x75, 0x74, 0x01, 0xF5, 0x3C, 0xF5, 0x3D, 0x80, +/*0590*/0x35, 0xE4, 0xF5, 0x4E, 0xC3, 0xE5, 0x47, 0x95, + 0x46, 0xF5, 0x3C, 0xC3, 0x13, 0xF5, 0x71, 0x25, +/*05A0*/0x46, 0xF5, 0x72, 0xC3, 0x94, 0x3F, 0x40, 0x05, + 0xE4, 0xF5, 0x3D, 0x80, 0x40, 0xC3, 0x74, 0x3F, +/*05B0*/0x95, 0x72, 0xF5, 0x3D, 0x80, 0x37, 0xE5, 0x46, + 0x65, 0x47, 0x70, 0x0F, 0x75, 0x73, 0x01, 0x75, +/*05C0*/0x75, 0x01, 0xF5, 0x3E, 0xF5, 0x3F, 0x75, 0x4E, + 0x01, 0x80, 0x22, 0xE4, 0xF5, 0x4E, 0xC3, 0xE5, +/*05D0*/0x47, 0x95, 0x46, 0xF5, 0x3E, 0xC3, 0x13, 0xF5, + 0x71, 0x25, 0x46, 0xF5, 0x72, 0xD3, 0x94, 0x3F, +/*05E0*/0x50, 0x05, 0xE4, 0xF5, 0x3F, 0x80, 0x06, 0xE5, + 0x72, 0x24, 0xC1, 0xF5, 0x3F, 0x05, 0x6F, 0xE5, +/*05F0*/0x6F, 0xC3, 0x94, 0x02, 0x50, 0x03, 0x02, 0x04, + 0x6E, 0xE5, 0x6D, 0x45, 0x6C, 0x70, 0x02, 0x80, +/*0600*/0x04, 0xE5, 0x74, 0x45, 0x75, 0x90, 0x07, 0x2F, + 0xF0, 0x7F, 0x01, 0xE5, 0x3E, 0x60, 0x04, 0xE5, +/*0610*/0x3C, 0x70, 0x14, 0xE4, 0xF5, 0x3C, 0xF5, 0x3D, + 0xF5, 0x3E, 0xF5, 0x3F, 0x12, 0x08, 0xD2, 0x70, +/*0620*/0x04, 0xF0, 0x02, 0x06, 0xA4, 0x80, 0x7A, 0xE5, + 0x3C, 0xC3, 0x95, 0x3E, 0x40, 0x07, 0xE5, 0x3C, +/*0630*/0x95, 0x3E, 0xFF, 0x80, 0x06, 0xC3, 0xE5, 0x3E, + 0x95, 0x3C, 0xFF, 0xE5, 0x76, 0xD3, 0x95, 0x79, +/*0640*/0x40, 0x05, 0x85, 0x76, 0x7A, 0x80, 0x03, 0x85, + 0x79, 0x7A, 0xE5, 0x77, 0xC3, 0x95, 0x78, 0x50, +/*0650*/0x05, 0x85, 0x77, 0x7B, 0x80, 0x03, 0x85, 0x78, + 0x7B, 0xE5, 0x7B, 0xD3, 0x95, 0x7A, 0x40, 0x30, +/*0660*/0xE5, 0x7B, 0x95, 0x7A, 0xF5, 0x3C, 0xF5, 0x3E, + 0xC3, 0xE5, 0x7B, 0x95, 0x7A, 0x90, 0x07, 0x19, +/*0670*/0xF0, 0xE5, 0x3C, 0xC3, 0x13, 0xF5, 0x71, 0x25, + 0x7A, 0xF5, 0x72, 0xC3, 0x94, 0x3F, 0x40, 0x05, +/*0680*/0xE4, 0xF5, 0x3D, 0x80, 0x1F, 0xC3, 0x74, 0x3F, + 0x95, 0x72, 0xF5, 0x3D, 0xF5, 0x3F, 0x80, 0x14, +/*0690*/0xE4, 0xF5, 0x3C, 0xF5, 0x3E, 0x90, 0x07, 0x19, + 0xF0, 0x12, 0x08, 0xD2, 0x70, 0x03, 0xF0, 0x80, +/*06A0*/0x03, 0x74, 0x01, 0xF0, 0x12, 0x08, 0x65, 0x75, + 0x83, 0xD0, 0xE0, 0x54, 0x0F, 0xFE, 0xAD, 0x3C, +/*06B0*/0x70, 0x02, 0x7E, 0x07, 0xBE, 0x0F, 0x02, 0x7E, + 0x80, 0xEE, 0xFB, 0xEF, 0xD3, 0x9B, 0x74, 0x80, +/*06C0*/0xF8, 0x98, 0x40, 0x1F, 0xE4, 0xF5, 0x3C, 0xF5, + 0x3E, 0x12, 0x08, 0xD2, 0x70, 0x03, 0xF0, 0x80, +/*06D0*/0x12, 0x74, 0x01, 0xF0, 0xE5, 0x08, 0xFB, 0xEB, + 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xD2, 0xE0, +/*06E0*/0x44, 0x10, 0xF0, 0xE5, 0x08, 0xFB, 0xEB, 0x44, + 0x09, 0xF5, 0x82, 0x75, 0x83, 0x9E, 0xED, 0xF0, +/*06F0*/0xEB, 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xCA, + 0xED, 0xF0, 0x12, 0x08, 0x65, 0x75, 0x83, 0xCC, +/*0700*/0xEF, 0xF0, 0x22, 0xE5, 0x08, 0x44, 0x07, 0xF5, + 0x82, 0x75, 0x83, 0xBC, 0xE0, 0x54, 0xF0, 0xF0, +/*0710*/0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, + 0xBE, 0xE0, 0x54, 0xF0, 0xF0, 0xE5, 0x08, 0x44, +/*0720*/0x07, 0xF5, 0x82, 0x75, 0x83, 0xC0, 0xE0, 0x54, + 0xF0, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82, +/*0730*/0x22, 0xF0, 0x90, 0x07, 0x28, 0xE0, 0xFE, 0xA3, + 0xE0, 0xF5, 0x82, 0x8E, 0x83, 0x22, 0x85, 0x42, +/*0740*/0x42, 0x85, 0x41, 0x41, 0x85, 0x40, 0x40, 0x74, + 0xC0, 0x2F, 0xF5, 0x82, 0x74, 0x02, 0x3E, 0xF5, +/*0750*/0x83, 0xE5, 0x42, 0xF0, 0x74, 0xE0, 0x2F, 0xF5, + 0x82, 0x74, 0x02, 0x3E, 0xF5, 0x83, 0x22, 0xE5, +/*0760*/0x42, 0x29, 0xFD, 0xE4, 0x33, 0xFC, 0xE5, 0x3C, + 0xC3, 0x9D, 0xEC, 0x64, 0x80, 0xF8, 0x74, 0x80, +/*0770*/0x98, 0x22, 0xF5, 0x83, 0xE0, 0x90, 0x07, 0x22, + 0x54, 0x1F, 0xFD, 0xE0, 0xFA, 0xA3, 0xE0, 0xF5, +/*0780*/0x82, 0x8A, 0x83, 0xED, 0xF0, 0x22, 0x90, 0x07, + 0x22, 0xE0, 0xFC, 0xA3, 0xE0, 0xF5, 0x82, 0x8C, +/*0790*/0x83, 0x22, 0x90, 0x07, 0x24, 0xFF, 0xED, 0x44, + 0x07, 0xCF, 0xF0, 0xA3, 0xEF, 0xF0, 0x22, 0x85, +/*07A0*/0x38, 0x38, 0x85, 0x39, 0x39, 0x85, 0x3A, 0x3A, + 0x74, 0xC0, 0x2F, 0xF5, 0x82, 0x74, 0x02, 0x3E, +/*07B0*/0xF5, 0x83, 0x22, 0x90, 0x07, 0x26, 0xFF, 0xED, + 0x44, 0x07, 0xCF, 0xF0, 0xA3, 0xEF, 0xF0, 0x22, +/*07C0*/0xF0, 0x74, 0xA0, 0x2F, 0xF5, 0x82, 0x74, 0x02, + 0x3E, 0xF5, 0x83, 0x22, 0x74, 0xC0, 0x25, 0x11, +/*07D0*/0xF5, 0x82, 0xE4, 0x34, 0x01, 0xF5, 0x83, 0x22, + 0x74, 0x00, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34, +/*07E0*/0x02, 0xF5, 0x83, 0x22, 0x74, 0x60, 0x25, 0x11, + 0xF5, 0x82, 0xE4, 0x34, 0x03, 0xF5, 0x83, 0x22, +/*07F0*/0x74, 0x80, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34, + 0x03, 0xF5, 0x83, 0x22, 0x74, 0xE0, 0x25, 0x11, +/*0800*/0xF5, 0x82, 0xE4, 0x34, 0x03, 0xF5, 0x83, 0x22, + 0x74, 0x40, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34, +/*0810*/0x06, 0xF5, 0x83, 0x22, 0x74, 0x80, 0x2F, 0xF5, + 0x82, 0x74, 0x02, 0x3E, 0xF5, 0x83, 0x22, 0xAF, +/*0820*/0x08, 0x7E, 0x00, 0xEF, 0x44, 0x07, 0xF5, 0x82, + 0x22, 0xF5, 0x83, 0xE5, 0x82, 0x44, 0x07, 0xF5, +/*0830*/0x82, 0xE5, 0x40, 0xF0, 0x22, 0x74, 0x40, 0x25, + 0x11, 0xF5, 0x82, 0xE4, 0x34, 0x02, 0xF5, 0x83, +/*0840*/0x22, 0x74, 0xC0, 0x25, 0x11, 0xF5, 0x82, 0xE4, + 0x34, 0x03, 0xF5, 0x83, 0x22, 0x74, 0x00, 0x25, +/*0850*/0x11, 0xF5, 0x82, 0xE4, 0x34, 0x06, 0xF5, 0x83, + 0x22, 0x74, 0x20, 0x25, 0x11, 0xF5, 0x82, 0xE4, +/*0860*/0x34, 0x06, 0xF5, 0x83, 0x22, 0xE5, 0x08, 0xFD, + 0xED, 0x44, 0x07, 0xF5, 0x82, 0x22, 0xE5, 0x41, +/*0870*/0xF0, 0xE5, 0x65, 0x64, 0x01, 0x45, 0x64, 0x22, + 0x7E, 0x00, 0xFB, 0x7A, 0x00, 0xFD, 0x7C, 0x00, +/*0880*/0x22, 0x74, 0x20, 0x25, 0x11, 0xF5, 0x82, 0xE4, + 0x34, 0x02, 0x22, 0x74, 0xA0, 0x25, 0x11, 0xF5, +/*0890*/0x82, 0xE4, 0x34, 0x03, 0x22, 0x85, 0x3E, 0x42, + 0x85, 0x3F, 0x41, 0x8F, 0x40, 0x22, 0x85, 0x3C, +/*08A0*/0x42, 0x85, 0x3D, 0x41, 0x8F, 0x40, 0x22, 0x75, + 0x45, 0x3F, 0x90, 0x07, 0x20, 0xE4, 0xF0, 0xA3, +/*08B0*/0x22, 0xF5, 0x83, 0xE5, 0x32, 0xF0, 0x05, 0x6E, + 0xE5, 0x6E, 0xC3, 0x94, 0x40, 0x22, 0xF0, 0xE5, +/*08C0*/0x08, 0x44, 0x06, 0xF5, 0x82, 0x22, 0x74, 0x00, + 0x25, 0x6E, 0xF5, 0x82, 0xE4, 0x34, 0x00, 0xF5, +/*08D0*/0x83, 0x22, 0xE5, 0x6D, 0x45, 0x6C, 0x90, 0x07, + 0x2F, 0x22, 0xE4, 0xF9, 0xE5, 0x3C, 0xD3, 0x95, +/*08E0*/0x3E, 0x22, 0x74, 0x80, 0x2E, 0xF5, 0x82, 0xE4, + 0x34, 0x02, 0xF5, 0x83, 0xE0, 0x22, 0x74, 0xA0, +/*08F0*/0x2E, 0xF5, 0x82, 0xE4, 0x34, 0x02, 0xF5, 0x83, + 0xE0, 0x22, 0x74, 0x80, 0x25, 0x6E, 0xF5, 0x82, +/*0900*/0xE4, 0x34, 0x00, 0x22, 0x25, 0x42, 0xFD, 0xE4, + 0x33, 0xFC, 0x22, 0x85, 0x42, 0x42, 0x85, 0x41, +/*0910*/0x41, 0x85, 0x40, 0x40, 0x22, 0xED, 0x4C, 0x60, + 0x03, 0x02, 0x09, 0xE5, 0xEF, 0x4E, 0x70, 0x37, +/*0920*/0x90, 0x07, 0x26, 0x12, 0x07, 0x89, 0xE0, 0xFD, + 0x12, 0x07, 0xCC, 0xED, 0xF0, 0x90, 0x07, 0x28, +/*0930*/0x12, 0x07, 0x89, 0xE0, 0xFD, 0x12, 0x07, 0xD8, + 0xED, 0xF0, 0x12, 0x07, 0x86, 0xE0, 0x54, 0x1F, +/*0940*/0xFD, 0x12, 0x08, 0x81, 0xF5, 0x83, 0xED, 0xF0, + 0x90, 0x07, 0x24, 0x12, 0x07, 0x89, 0xE0, 0x54, +/*0950*/0x1F, 0xFD, 0x12, 0x08, 0x35, 0xED, 0xF0, 0xEF, + 0x64, 0x04, 0x4E, 0x70, 0x37, 0x90, 0x07, 0x26, +/*0960*/0x12, 0x07, 0x89, 0xE0, 0xFD, 0x12, 0x07, 0xE4, + 0xED, 0xF0, 0x90, 0x07, 0x28, 0x12, 0x07, 0x89, +/*0970*/0xE0, 0xFD, 0x12, 0x07, 0xF0, 0xED, 0xF0, 0x12, + 0x07, 0x86, 0xE0, 0x54, 0x1F, 0xFD, 0x12, 0x08, +/*0980*/0x8B, 0xF5, 0x83, 0xED, 0xF0, 0x90, 0x07, 0x24, + 0x12, 0x07, 0x89, 0xE0, 0x54, 0x1F, 0xFD, 0x12, +/*0990*/0x08, 0x41, 0xED, 0xF0, 0xEF, 0x64, 0x01, 0x4E, + 0x70, 0x04, 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, +/*09A0*/0xEF, 0x64, 0x02, 0x4E, 0x70, 0x04, 0x7F, 0x01, + 0x80, 0x02, 0x7F, 0x00, 0xEF, 0x4D, 0x60, 0x78, +/*09B0*/0x90, 0x07, 0x26, 0x12, 0x07, 0x35, 0xE0, 0xFF, + 0x12, 0x07, 0xFC, 0xEF, 0x12, 0x07, 0x31, 0xE0, +/*09C0*/0xFF, 0x12, 0x08, 0x08, 0xEF, 0xF0, 0x90, 0x07, + 0x22, 0x12, 0x07, 0x35, 0xE0, 0x54, 0x1F, 0xFF, +/*09D0*/0x12, 0x08, 0x4D, 0xEF, 0xF0, 0x90, 0x07, 0x24, + 0x12, 0x07, 0x35, 0xE0, 0x54, 0x1F, 0xFF, 0x12, +/*09E0*/0x08, 0x59, 0xEF, 0xF0, 0x22, 0x12, 0x07, 0xCC, + 0xE4, 0xF0, 0x12, 0x07, 0xD8, 0xE4, 0xF0, 0x12, +/*09F0*/0x08, 0x81, 0xF5, 0x83, 0xE4, 0xF0, 0x12, 0x08, + 0x35, 0x74, 0x14, 0xF0, 0x12, 0x07, 0xE4, 0xE4, +/*0A00*/0xF0, 0x12, 0x07, 0xF0, 0xE4, 0xF0, 0x12, 0x08, + 0x8B, 0xF5, 0x83, 0xE4, 0xF0, 0x12, 0x08, 0x41, +/*0A10*/0x74, 0x14, 0xF0, 0x12, 0x07, 0xFC, 0xE4, 0xF0, + 0x12, 0x08, 0x08, 0xE4, 0xF0, 0x12, 0x08, 0x4D, +/*0A20*/0xE4, 0xF0, 0x12, 0x08, 0x59, 0x74, 0x14, 0xF0, + 0x22, 0x53, 0xF9, 0xF7, 0x75, 0xFC, 0x10, 0xE4, +/*0A30*/0xF5, 0xFD, 0x75, 0xFE, 0x30, 0xF5, 0xFF, 0xE5, + 0xE7, 0x20, 0xE7, 0x03, 0x43, 0xF9, 0x08, 0xE5, +/*0A40*/0xE6, 0x20, 0xE7, 0x0B, 0x78, 0xFF, 0xE4, 0xF6, + 0xD8, 0xFD, 0x53, 0xE6, 0xFE, 0x80, 0x09, 0x78, +/*0A50*/0x08, 0xE4, 0xF6, 0xD8, 0xFD, 0x53, 0xE6, 0xFE, + 0x75, 0x81, 0x80, 0xE4, 0xF5, 0xA8, 0xD2, 0xA8, +/*0A60*/0xC2, 0xA9, 0xD2, 0xAF, 0xE5, 0xE2, 0x20, 0xE5, + 0x05, 0x20, 0xE6, 0x02, 0x80, 0x03, 0x43, 0xE1, +/*0A70*/0x02, 0xE5, 0xE2, 0x20, 0xE0, 0x0E, 0x90, 0x00, + 0x00, 0x7F, 0x00, 0x7E, 0x08, 0xE4, 0xF0, 0xA3, +/*0A80*/0xDF, 0xFC, 0xDE, 0xFA, 0x02, 0x0A, 0xDB, 0x43, + 0xFA, 0x01, 0xC0, 0xE0, 0xC0, 0xF0, 0xC0, 0x83, +/*0A90*/0xC0, 0x82, 0xC0, 0xD0, 0x12, 0x1C, 0xE7, 0xD0, + 0xD0, 0xD0, 0x82, 0xD0, 0x83, 0xD0, 0xF0, 0xD0, +/*0AA0*/0xE0, 0x53, 0xFA, 0xFE, 0x32, 0x02, 0x1B, 0x55, + 0xE4, 0x93, 0xA3, 0xF8, 0xE4, 0x93, 0xA3, 0xF6, +/*0AB0*/0x08, 0xDF, 0xF9, 0x80, 0x29, 0xE4, 0x93, 0xA3, + 0xF8, 0x54, 0x07, 0x24, 0x0C, 0xC8, 0xC3, 0x33, +/*0AC0*/0xC4, 0x54, 0x0F, 0x44, 0x20, 0xC8, 0x83, 0x40, + 0x04, 0xF4, 0x56, 0x80, 0x01, 0x46, 0xF6, 0xDF, +/*0AD0*/0xE4, 0x80, 0x0B, 0x01, 0x02, 0x04, 0x08, 0x10, + 0x20, 0x40, 0x80, 0x90, 0x00, 0x3F, 0xE4, 0x7E, +/*0AE0*/0x01, 0x93, 0x60, 0xC1, 0xA3, 0xFF, 0x54, 0x3F, + 0x30, 0xE5, 0x09, 0x54, 0x1F, 0xFE, 0xE4, 0x93, +/*0AF0*/0xA3, 0x60, 0x01, 0x0E, 0xCF, 0x54, 0xC0, 0x25, + 0xE0, 0x60, 0xAD, 0x40, 0xB8, 0x80, 0xFE, 0x8C, +/*0B00*/0x64, 0x8D, 0x65, 0x8A, 0x66, 0x8B, 0x67, 0xE4, + 0xF5, 0x69, 0xEF, 0x4E, 0x70, 0x03, 0x02, 0x1D, +/*0B10*/0x55, 0xE4, 0xF5, 0x68, 0xE5, 0x67, 0x45, 0x66, + 0x70, 0x32, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x90, +/*0B20*/0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE4, + 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0xE4, 0x12, +/*0B30*/0x08, 0x70, 0x70, 0x29, 0x12, 0x07, 0x2A, 0x75, + 0x83, 0x92, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, +/*0B40*/0xC6, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC8, + 0xE4, 0xF0, 0x80, 0x11, 0x90, 0x07, 0x26, 0x12, +/*0B50*/0x07, 0x35, 0xE4, 0x12, 0x08, 0x70, 0x70, 0x05, + 0x12, 0x07, 0x32, 0xE4, 0xF0, 0x12, 0x1D, 0x55, +/*0B60*/0x12, 0x1E, 0xBF, 0xE5, 0x67, 0x45, 0x66, 0x70, + 0x33, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x90, 0xE5, +/*0B70*/0x41, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE5, + 0x41, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0x12, +/*0B80*/0x08, 0x6E, 0x70, 0x29, 0x12, 0x07, 0x2A, 0x75, + 0x83, 0x92, 0xE5, 0x40, 0x12, 0x07, 0x29, 0x75, +/*0B90*/0x83, 0xC6, 0xE5, 0x40, 0x12, 0x07, 0x29, 0x75, + 0x83, 0xC8, 0x80, 0x0E, 0x90, 0x07, 0x26, 0x12, +/*0BA0*/0x07, 0x35, 0x12, 0x08, 0x6E, 0x70, 0x06, 0x12, + 0x07, 0x32, 0xE5, 0x40, 0xF0, 0xAF, 0x69, 0x7E, +/*0BB0*/0x00, 0xAD, 0x67, 0xAC, 0x66, 0x12, 0x04, 0x44, + 0x12, 0x07, 0x2A, 0x75, 0x83, 0xCA, 0xE0, 0xD3, +/*0BC0*/0x94, 0x00, 0x50, 0x0C, 0x05, 0x68, 0xE5, 0x68, + 0xC3, 0x94, 0x05, 0x50, 0x03, 0x02, 0x0B, 0x14, +/*0BD0*/0x22, 0x8C, 0x60, 0x8D, 0x61, 0x12, 0x08, 0xDA, + 0x74, 0x20, 0x40, 0x0D, 0x2F, 0xF5, 0x82, 0x74, +/*0BE0*/0x03, 0x3E, 0xF5, 0x83, 0xE5, 0x3E, 0xF0, 0x80, + 0x0B, 0x2F, 0xF5, 0x82, 0x74, 0x03, 0x3E, 0xF5, +/*0BF0*/0x83, 0xE5, 0x3C, 0xF0, 0xE5, 0x3C, 0xD3, 0x95, + 0x3E, 0x40, 0x3C, 0xE5, 0x61, 0x45, 0x60, 0x70, +/*0C00*/0x10, 0xE9, 0x12, 0x09, 0x04, 0xE5, 0x3E, 0x12, + 0x07, 0x68, 0x40, 0x3B, 0x12, 0x08, 0x95, 0x80, +/*0C10*/0x18, 0xE5, 0x3E, 0xC3, 0x95, 0x38, 0x40, 0x1D, + 0x85, 0x3E, 0x38, 0xE5, 0x3E, 0x60, 0x05, 0x85, +/*0C20*/0x3F, 0x39, 0x80, 0x03, 0x85, 0x39, 0x39, 0x8F, + 0x3A, 0x12, 0x08, 0x14, 0xE5, 0x3E, 0x12, 0x07, +/*0C30*/0xC0, 0xE5, 0x3F, 0xF0, 0x22, 0x80, 0x43, 0xE5, + 0x61, 0x45, 0x60, 0x70, 0x19, 0x12, 0x07, 0x5F, +/*0C40*/0x40, 0x05, 0x12, 0x08, 0x9E, 0x80, 0x27, 0x12, + 0x09, 0x0B, 0x12, 0x08, 0x14, 0xE5, 0x42, 0x12, +/*0C50*/0x07, 0xC0, 0xE5, 0x41, 0xF0, 0x22, 0xE5, 0x3C, + 0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3C, 0x38, +/*0C60*/0xE5, 0x3C, 0x60, 0x05, 0x85, 0x3D, 0x39, 0x80, + 0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, 0x08, +/*0C70*/0x14, 0xE5, 0x3C, 0x12, 0x07, 0xC0, 0xE5, 0x3D, + 0xF0, 0x22, 0x85, 0x38, 0x38, 0x85, 0x39, 0x39, +/*0C80*/0x85, 0x3A, 0x3A, 0x12, 0x08, 0x14, 0xE5, 0x38, + 0x12, 0x07, 0xC0, 0xE5, 0x39, 0xF0, 0x22, 0x7F, +/*0C90*/0x06, 0x12, 0x17, 0x31, 0x12, 0x1D, 0x23, 0x12, + 0x0E, 0x04, 0x12, 0x0E, 0x33, 0xE0, 0x44, 0x0A, +/*0CA0*/0xF0, 0x74, 0x8E, 0xFE, 0x12, 0x0E, 0x04, 0x12, + 0x0E, 0x0B, 0xEF, 0xF0, 0xE5, 0x28, 0x30, 0xE5, +/*0CB0*/0x03, 0xD3, 0x80, 0x01, 0xC3, 0x40, 0x05, 0x75, + 0x14, 0x20, 0x80, 0x03, 0x75, 0x14, 0x08, 0x12, +/*0CC0*/0x0E, 0x04, 0x75, 0x83, 0x8A, 0xE5, 0x14, 0xF0, + 0xB4, 0xFF, 0x05, 0x75, 0x12, 0x80, 0x80, 0x06, +/*0CD0*/0xE5, 0x14, 0xC3, 0x13, 0xF5, 0x12, 0xE4, 0xF5, + 0x16, 0xF5, 0x7F, 0x12, 0x19, 0x36, 0x12, 0x13, +/*0CE0*/0xA3, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x50, 0x09, + 0x05, 0x16, 0xE5, 0x16, 0xC3, 0x94, 0x14, 0x40, +/*0CF0*/0xEA, 0xE5, 0xE4, 0x20, 0xE7, 0x28, 0x12, 0x0E, + 0x04, 0x75, 0x83, 0xD2, 0xE0, 0x54, 0x08, 0xD3, +/*0D00*/0x94, 0x00, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02, + 0x7F, 0x00, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40, +/*0D10*/0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF, + 0x5E, 0x60, 0x03, 0x12, 0x1D, 0xD7, 0xE5, 0x7F, +/*0D20*/0xC3, 0x94, 0x11, 0x40, 0x14, 0x12, 0x0E, 0x04, + 0x75, 0x83, 0xD2, 0xE0, 0x44, 0x80, 0xF0, 0xE5, +/*0D30*/0xE4, 0x20, 0xE7, 0x0F, 0x12, 0x1D, 0xD7, 0x80, + 0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0xD2, 0xE0, +/*0D40*/0x54, 0x7F, 0xF0, 0x12, 0x1D, 0x23, 0x22, 0x74, + 0x8A, 0x85, 0x08, 0x82, 0xF5, 0x83, 0xE5, 0x17, +/*0D50*/0xF0, 0x12, 0x0E, 0x3A, 0xE4, 0xF0, 0x90, 0x07, + 0x02, 0xE0, 0x12, 0x0E, 0x17, 0x75, 0x83, 0x90, +/*0D60*/0xEF, 0xF0, 0x74, 0x92, 0xFE, 0xE5, 0x08, 0x44, + 0x07, 0xFF, 0xF5, 0x82, 0x8E, 0x83, 0xE0, 0x54, +/*0D70*/0xC0, 0xFD, 0x90, 0x07, 0x03, 0xE0, 0x54, 0x3F, + 0x4D, 0x8F, 0x82, 0x8E, 0x83, 0xF0, 0x90, 0x07, +/*0D80*/0x04, 0xE0, 0x12, 0x0E, 0x17, 0x75, 0x83, 0x82, + 0xEF, 0xF0, 0x90, 0x07, 0x05, 0xE0, 0xFF, 0xED, +/*0D90*/0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xB4, 0xEF, + 0x12, 0x0E, 0x03, 0x75, 0x83, 0x80, 0xE0, 0x54, +/*0DA0*/0xBF, 0xF0, 0x30, 0x37, 0x0A, 0x12, 0x0E, 0x91, + 0x75, 0x83, 0x94, 0xE0, 0x44, 0x80, 0xF0, 0x30, +/*0DB0*/0x38, 0x0A, 0x12, 0x0E, 0x91, 0x75, 0x83, 0x92, + 0xE0, 0x44, 0x80, 0xF0, 0xE5, 0x28, 0x30, 0xE4, +/*0DC0*/0x1A, 0x20, 0x39, 0x0A, 0x12, 0x0E, 0x04, 0x75, + 0x83, 0x88, 0xE0, 0x54, 0x7F, 0xF0, 0x20, 0x3A, +/*0DD0*/0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0x88, 0xE0, + 0x54, 0xBF, 0xF0, 0x74, 0x8C, 0xFE, 0x12, 0x0E, +/*0DE0*/0x04, 0x8E, 0x83, 0xE0, 0x54, 0x0F, 0x12, 0x0E, + 0x03, 0x75, 0x83, 0x86, 0xE0, 0x54, 0xBF, 0xF0, +/*0DF0*/0xE5, 0x08, 0x44, 0x06, 0x12, 0x0D, 0xFD, 0x75, + 0x83, 0x8A, 0xE4, 0xF0, 0x22, 0xF5, 0x82, 0x75, +/*0E00*/0x83, 0x82, 0xE4, 0xF0, 0xE5, 0x08, 0x44, 0x07, + 0xF5, 0x82, 0x22, 0x8E, 0x83, 0xE0, 0xF5, 0x10, +/*0E10*/0x54, 0xFE, 0xF0, 0xE5, 0x10, 0x44, 0x01, 0xFF, + 0xE5, 0x08, 0xFD, 0xED, 0x44, 0x07, 0xF5, 0x82, +/*0E20*/0x22, 0xE5, 0x15, 0xC4, 0x54, 0x07, 0xFF, 0xE5, + 0x08, 0xFD, 0xED, 0x44, 0x08, 0xF5, 0x82, 0x75, +/*0E30*/0x83, 0x82, 0x22, 0x75, 0x83, 0x80, 0xE0, 0x44, + 0x40, 0xF0, 0xE5, 0x08, 0x44, 0x08, 0xF5, 0x82, +/*0E40*/0x75, 0x83, 0x8A, 0x22, 0xE5, 0x16, 0x25, 0xE0, + 0x25, 0xE0, 0x24, 0xAF, 0xF5, 0x82, 0xE4, 0x34, +/*0E50*/0x1A, 0xF5, 0x83, 0xE4, 0x93, 0xF5, 0x0D, 0x22, + 0x43, 0xE1, 0x10, 0x43, 0xE1, 0x80, 0x53, 0xE1, +/*0E60*/0xFD, 0x85, 0xE1, 0x10, 0x22, 0xE5, 0x16, 0x25, + 0xE0, 0x25, 0xE0, 0x24, 0xB2, 0xF5, 0x82, 0xE4, +/*0E70*/0x34, 0x1A, 0xF5, 0x83, 0xE4, 0x93, 0x22, 0x85, + 0x55, 0x82, 0x85, 0x54, 0x83, 0xE5, 0x15, 0xF0, +/*0E80*/0x22, 0xE5, 0xE2, 0x54, 0x20, 0xD3, 0x94, 0x00, + 0x22, 0xE5, 0xE2, 0x54, 0x40, 0xD3, 0x94, 0x00, +/*0E90*/0x22, 0xE5, 0x08, 0x44, 0x06, 0xF5, 0x82, 0x22, + 0xFD, 0xE5, 0x08, 0xFB, 0xEB, 0x44, 0x07, 0xF5, +/*0EA0*/0x82, 0x22, 0x53, 0xF9, 0xF7, 0x75, 0xFE, 0x30, + 0x22, 0xEF, 0x4E, 0x70, 0x26, 0x12, 0x07, 0xCC, +/*0EB0*/0xE0, 0xFD, 0x90, 0x07, 0x26, 0x12, 0x07, 0x7B, + 0x12, 0x07, 0xD8, 0xE0, 0xFD, 0x90, 0x07, 0x28, +/*0EC0*/0x12, 0x07, 0x7B, 0x12, 0x08, 0x81, 0x12, 0x07, + 0x72, 0x12, 0x08, 0x35, 0xE0, 0x90, 0x07, 0x24, +/*0ED0*/0x12, 0x07, 0x78, 0xEF, 0x64, 0x04, 0x4E, 0x70, + 0x29, 0x12, 0x07, 0xE4, 0xE0, 0xFD, 0x90, 0x07, +/*0EE0*/0x26, 0x12, 0x07, 0x7B, 0x12, 0x07, 0xF0, 0xE0, + 0xFD, 0x90, 0x07, 0x28, 0x12, 0x07, 0x7B, 0x12, +/*0EF0*/0x08, 0x8B, 0x12, 0x07, 0x72, 0x12, 0x08, 0x41, + 0xE0, 0x54, 0x1F, 0xFD, 0x90, 0x07, 0x24, 0x12, +/*0F00*/0x07, 0x7B, 0xEF, 0x64, 0x01, 0x4E, 0x70, 0x04, + 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, 0xEF, 0x64, +/*0F10*/0x02, 0x4E, 0x70, 0x04, 0x7F, 0x01, 0x80, 0x02, + 0x7F, 0x00, 0xEF, 0x4D, 0x60, 0x35, 0x12, 0x07, +/*0F20*/0xFC, 0xE0, 0xFF, 0x90, 0x07, 0x26, 0x12, 0x07, + 0x89, 0xEF, 0xF0, 0x12, 0x08, 0x08, 0xE0, 0xFF, +/*0F30*/0x90, 0x07, 0x28, 0x12, 0x07, 0x89, 0xEF, 0xF0, + 0x12, 0x08, 0x4D, 0xE0, 0x54, 0x1F, 0xFF, 0x12, +/*0F40*/0x07, 0x86, 0xEF, 0xF0, 0x12, 0x08, 0x59, 0xE0, + 0x54, 0x1F, 0xFF, 0x90, 0x07, 0x24, 0x12, 0x07, +/*0F50*/0x89, 0xEF, 0xF0, 0x22, 0xE4, 0xF5, 0x53, 0x12, + 0x0E, 0x81, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02, +/*0F60*/0x7F, 0x00, 0x12, 0x0E, 0x89, 0x40, 0x04, 0x7E, + 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x70, +/*0F70*/0x03, 0x02, 0x0F, 0xF6, 0x85, 0xE1, 0x10, 0x43, + 0xE1, 0x02, 0x53, 0xE1, 0x0F, 0x85, 0xE1, 0x10, +/*0F80*/0xE4, 0xF5, 0x51, 0xE5, 0xE3, 0x54, 0x3F, 0xF5, + 0x52, 0x12, 0x0E, 0x89, 0x40, 0x1D, 0xAD, 0x52, +/*0F90*/0xAF, 0x51, 0x12, 0x11, 0x18, 0xEF, 0x60, 0x08, + 0x85, 0xE1, 0x10, 0x43, 0xE1, 0x40, 0x80, 0x0B, +/*0FA0*/0x53, 0xE1, 0xBF, 0x12, 0x0E, 0x58, 0x12, 0x00, + 0x06, 0x80, 0xFB, 0xE5, 0xE3, 0x54, 0x3F, 0xF5, +/*0FB0*/0x51, 0xE5, 0xE4, 0x54, 0x3F, 0xF5, 0x52, 0x12, + 0x0E, 0x81, 0x40, 0x1D, 0xAD, 0x52, 0xAF, 0x51, +/*0FC0*/0x12, 0x11, 0x18, 0xEF, 0x60, 0x08, 0x85, 0xE1, + 0x10, 0x43, 0xE1, 0x20, 0x80, 0x0B, 0x53, 0xE1, +/*0FD0*/0xDF, 0x12, 0x0E, 0x58, 0x12, 0x00, 0x06, 0x80, + 0xFB, 0x12, 0x0E, 0x81, 0x40, 0x04, 0x7F, 0x01, +/*0FE0*/0x80, 0x02, 0x7F, 0x00, 0x12, 0x0E, 0x89, 0x40, + 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, +/*0FF0*/0x4F, 0x60, 0x03, 0x12, 0x0E, 0x5B, 0x22, 0x12, + 0x0E, 0x21, 0xEF, 0xF0, 0x12, 0x10, 0x91, 0x22, +/*1000*/0x02, 0x11, 0x00, 0x02, 0x10, 0x40, 0x02, 0x10, + 0x90, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1010*/0x01, 0x20, 0x01, 0x20, 0xE4, 0xF5, 0x57, 0x12, + 0x16, 0xBD, 0x12, 0x16, 0x44, 0xE4, 0x12, 0x10, +/*1020*/0x56, 0x12, 0x14, 0xB7, 0x90, 0x07, 0x26, 0x12, + 0x07, 0x35, 0xE4, 0x12, 0x07, 0x31, 0xE4, 0xF0, +/*1030*/0x12, 0x10, 0x56, 0x12, 0x14, 0xB7, 0x90, 0x07, + 0x26, 0x12, 0x07, 0x35, 0xE5, 0x41, 0x12, 0x07, +/*1040*/0x31, 0xE5, 0x40, 0xF0, 0xAF, 0x57, 0x7E, 0x00, + 0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, 0xAF, +/*1050*/0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0xFF, 0x90, + 0x07, 0x20, 0xA3, 0xE0, 0xFD, 0xE4, 0xF5, 0x56, +/*1060*/0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, 0x12, + 0x11, 0x51, 0x7F, 0x0F, 0x7D, 0x18, 0xE4, 0xF5, +/*1070*/0x56, 0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, + 0x12, 0x15, 0x41, 0xAF, 0x56, 0x7E, 0x00, 0x12, +/*1080*/0x1A, 0xFF, 0xE4, 0xFF, 0xF5, 0x56, 0x7D, 0x1F, + 0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, 0x22, +/*1090*/0x22, 0xE4, 0xF5, 0x55, 0xE5, 0x08, 0xFD, 0x74, + 0xA0, 0xF5, 0x56, 0xED, 0x44, 0x07, 0xF5, 0x57, +/*10A0*/0xE5, 0x28, 0x30, 0xE5, 0x03, 0xD3, 0x80, 0x01, + 0xC3, 0x40, 0x05, 0x7F, 0x28, 0xEF, 0x80, 0x04, +/*10B0*/0x7F, 0x14, 0xEF, 0xC3, 0x13, 0xF5, 0x54, 0xE4, + 0xF9, 0x12, 0x0E, 0x18, 0x75, 0x83, 0x8E, 0xE0, +/*10C0*/0xF5, 0x10, 0xCE, 0xEF, 0xCE, 0xEE, 0xD3, 0x94, + 0x00, 0x40, 0x26, 0xE5, 0x10, 0x54, 0xFE, 0x12, +/*10D0*/0x0E, 0x98, 0x75, 0x83, 0x8E, 0xED, 0xF0, 0xE5, + 0x10, 0x44, 0x01, 0xFD, 0xEB, 0x44, 0x07, 0xF5, +/*10E0*/0x82, 0xED, 0xF0, 0x85, 0x57, 0x82, 0x85, 0x56, + 0x83, 0xE0, 0x30, 0xE3, 0x01, 0x09, 0x1E, 0x80, +/*10F0*/0xD4, 0xC2, 0x34, 0xE9, 0xC3, 0x95, 0x54, 0x40, + 0x02, 0xD2, 0x34, 0x22, 0x02, 0x00, 0x06, 0x22, +/*1100*/0x30, 0x30, 0x11, 0x90, 0x10, 0x00, 0xE4, 0x93, + 0xF5, 0x10, 0x90, 0x10, 0x10, 0xE4, 0x93, 0xF5, +/*1110*/0x10, 0x12, 0x10, 0x90, 0x12, 0x11, 0x50, 0x22, + 0xE4, 0xFC, 0xC3, 0xED, 0x9F, 0xFA, 0xEF, 0xF5, +/*1120*/0x83, 0x75, 0x82, 0x00, 0x79, 0xFF, 0xE4, 0x93, + 0xCC, 0x6C, 0xCC, 0xA3, 0xD9, 0xF8, 0xDA, 0xF6, +/*1130*/0xE5, 0xE2, 0x30, 0xE4, 0x02, 0x8C, 0xE5, 0xED, + 0x24, 0xFF, 0xFF, 0xEF, 0x75, 0x82, 0xFF, 0xF5, +/*1140*/0x83, 0xE4, 0x93, 0x6C, 0x70, 0x03, 0x7F, 0x01, + 0x22, 0x7F, 0x00, 0x22, 0x22, 0x11, 0x00, 0x00, +/*1150*/0x22, 0x8E, 0x58, 0x8F, 0x59, 0x8C, 0x5A, 0x8D, + 0x5B, 0x8A, 0x5C, 0x8B, 0x5D, 0x75, 0x5E, 0x01, +/*1160*/0xE4, 0xF5, 0x5F, 0xF5, 0x60, 0xF5, 0x62, 0x12, + 0x07, 0x2A, 0x75, 0x83, 0xD0, 0xE0, 0xFF, 0xC4, +/*1170*/0x54, 0x0F, 0xF5, 0x61, 0x12, 0x1E, 0xA5, 0x85, + 0x59, 0x5E, 0xD3, 0xE5, 0x5E, 0x95, 0x5B, 0xE5, +/*1180*/0x5A, 0x12, 0x07, 0x6B, 0x50, 0x4B, 0x12, 0x07, + 0x03, 0x75, 0x83, 0xBC, 0xE0, 0x45, 0x5E, 0x12, +/*1190*/0x07, 0x29, 0x75, 0x83, 0xBE, 0xE0, 0x45, 0x5E, + 0x12, 0x07, 0x29, 0x75, 0x83, 0xC0, 0xE0, 0x45, +/*11A0*/0x5E, 0xF0, 0xAF, 0x5F, 0xE5, 0x60, 0x12, 0x08, + 0x78, 0x12, 0x0A, 0xFF, 0xAF, 0x62, 0x7E, 0x00, +/*11B0*/0xAD, 0x5D, 0xAC, 0x5C, 0x12, 0x04, 0x44, 0xE5, + 0x61, 0xAF, 0x5E, 0x7E, 0x00, 0xB4, 0x03, 0x05, +/*11C0*/0x12, 0x1E, 0x21, 0x80, 0x07, 0xAD, 0x5D, 0xAC, + 0x5C, 0x12, 0x13, 0x17, 0x05, 0x5E, 0x02, 0x11, +/*11D0*/0x7A, 0x12, 0x07, 0x03, 0x75, 0x83, 0xBC, 0xE0, + 0x45, 0x40, 0x12, 0x07, 0x29, 0x75, 0x83, 0xBE, +/*11E0*/0xE0, 0x45, 0x40, 0x12, 0x07, 0x29, 0x75, 0x83, + 0xC0, 0xE0, 0x45, 0x40, 0xF0, 0x22, 0x8E, 0x58, +/*11F0*/0x8F, 0x59, 0x75, 0x5A, 0x01, 0x79, 0x01, 0x75, + 0x5B, 0x01, 0xE4, 0xFB, 0x12, 0x07, 0x2A, 0x75, +/*1200*/0x83, 0xAE, 0xE0, 0x54, 0x1A, 0xFF, 0x12, 0x08, + 0x65, 0xE0, 0xC4, 0x13, 0x54, 0x07, 0xFE, 0xEF, +/*1210*/0x70, 0x0C, 0xEE, 0x65, 0x35, 0x70, 0x07, 0x90, + 0x07, 0x2F, 0xE0, 0xB4, 0x01, 0x0D, 0xAF, 0x35, +/*1220*/0x7E, 0x00, 0x12, 0x0E, 0xA9, 0xCF, 0xEB, 0xCF, + 0x02, 0x1E, 0x60, 0xE5, 0x59, 0x64, 0x02, 0x45, +/*1230*/0x58, 0x70, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, + 0x00, 0xE5, 0x59, 0x45, 0x58, 0x70, 0x04, 0x7E, +/*1240*/0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x60, + 0x23, 0x85, 0x41, 0x49, 0x85, 0x40, 0x4B, 0xE5, +/*1250*/0x59, 0x45, 0x58, 0x70, 0x2C, 0xAF, 0x5A, 0xFE, + 0xCD, 0xE9, 0xCD, 0xFC, 0xAB, 0x59, 0xAA, 0x58, +/*1260*/0x12, 0x0A, 0xFF, 0xAF, 0x5B, 0x7E, 0x00, 0x12, + 0x1E, 0x60, 0x80, 0x15, 0xAF, 0x5B, 0x7E, 0x00, +/*1270*/0x12, 0x1E, 0x60, 0x90, 0x07, 0x26, 0x12, 0x07, + 0x35, 0xE5, 0x49, 0x12, 0x07, 0x31, 0xE5, 0x4B, +/*1280*/0xF0, 0xE4, 0xFD, 0xAF, 0x35, 0xFE, 0xFC, 0x12, + 0x09, 0x15, 0x22, 0x8C, 0x64, 0x8D, 0x65, 0x12, +/*1290*/0x08, 0xDA, 0x40, 0x3C, 0xE5, 0x65, 0x45, 0x64, + 0x70, 0x10, 0x12, 0x09, 0x04, 0xC3, 0xE5, 0x3E, +/*12A0*/0x12, 0x07, 0x69, 0x40, 0x3B, 0x12, 0x08, 0x95, + 0x80, 0x18, 0xE5, 0x3E, 0xC3, 0x95, 0x38, 0x40, +/*12B0*/0x1D, 0x85, 0x3E, 0x38, 0xE5, 0x3E, 0x60, 0x05, + 0x85, 0x3F, 0x39, 0x80, 0x03, 0x85, 0x39, 0x39, +/*12C0*/0x8F, 0x3A, 0x12, 0x07, 0xA8, 0xE5, 0x3E, 0x12, + 0x07, 0x53, 0xE5, 0x3F, 0xF0, 0x22, 0x80, 0x3B, +/*12D0*/0xE5, 0x65, 0x45, 0x64, 0x70, 0x11, 0x12, 0x07, + 0x5F, 0x40, 0x05, 0x12, 0x08, 0x9E, 0x80, 0x1F, +/*12E0*/0x12, 0x07, 0x3E, 0xE5, 0x41, 0xF0, 0x22, 0xE5, + 0x3C, 0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3C, +/*12F0*/0x38, 0xE5, 0x3C, 0x60, 0x05, 0x85, 0x3D, 0x39, + 0x80, 0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, +/*1300*/0x07, 0xA8, 0xE5, 0x3C, 0x12, 0x07, 0x53, 0xE5, + 0x3D, 0xF0, 0x22, 0x12, 0x07, 0x9F, 0xE5, 0x38, +/*1310*/0x12, 0x07, 0x53, 0xE5, 0x39, 0xF0, 0x22, 0x8C, + 0x63, 0x8D, 0x64, 0x12, 0x08, 0xDA, 0x40, 0x3C, +/*1320*/0xE5, 0x64, 0x45, 0x63, 0x70, 0x10, 0x12, 0x09, + 0x04, 0xC3, 0xE5, 0x3E, 0x12, 0x07, 0x69, 0x40, +/*1330*/0x3B, 0x12, 0x08, 0x95, 0x80, 0x18, 0xE5, 0x3E, + 0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3E, 0x38, +/*1340*/0xE5, 0x3E, 0x60, 0x05, 0x85, 0x3F, 0x39, 0x80, + 0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, 0x07, +/*1350*/0xA8, 0xE5, 0x3E, 0x12, 0x07, 0x53, 0xE5, 0x3F, + 0xF0, 0x22, 0x80, 0x3B, 0xE5, 0x64, 0x45, 0x63, +/*1360*/0x70, 0x11, 0x12, 0x07, 0x5F, 0x40, 0x05, 0x12, + 0x08, 0x9E, 0x80, 0x1F, 0x12, 0x07, 0x3E, 0xE5, +/*1370*/0x41, 0xF0, 0x22, 0xE5, 0x3C, 0xC3, 0x95, 0x38, + 0x40, 0x1D, 0x85, 0x3C, 0x38, 0xE5, 0x3C, 0x60, +/*1380*/0x05, 0x85, 0x3D, 0x39, 0x80, 0x03, 0x85, 0x39, + 0x39, 0x8F, 0x3A, 0x12, 0x07, 0xA8, 0xE5, 0x3C, +/*1390*/0x12, 0x07, 0x53, 0xE5, 0x3D, 0xF0, 0x22, 0x12, + 0x07, 0x9F, 0xE5, 0x38, 0x12, 0x07, 0x53, 0xE5, +/*13A0*/0x39, 0xF0, 0x22, 0xE5, 0x0D, 0xFE, 0xE5, 0x08, + 0x8E, 0x54, 0x44, 0x05, 0xF5, 0x55, 0x75, 0x15, +/*13B0*/0x0F, 0xF5, 0x82, 0x12, 0x0E, 0x7A, 0x12, 0x17, + 0xA3, 0x20, 0x31, 0x05, 0x75, 0x15, 0x03, 0x80, +/*13C0*/0x03, 0x75, 0x15, 0x0B, 0xE5, 0x0A, 0xC3, 0x94, + 0x01, 0x50, 0x38, 0x12, 0x14, 0x20, 0x20, 0x31, +/*13D0*/0x06, 0x05, 0x15, 0x05, 0x15, 0x80, 0x04, 0x15, + 0x15, 0x15, 0x15, 0xE5, 0x0A, 0xC3, 0x94, 0x01, +/*13E0*/0x50, 0x21, 0x12, 0x14, 0x20, 0x20, 0x31, 0x04, + 0x05, 0x15, 0x80, 0x02, 0x15, 0x15, 0xE5, 0x0A, +/*13F0*/0xC3, 0x94, 0x01, 0x50, 0x0E, 0x12, 0x0E, 0x77, + 0x12, 0x17, 0xA3, 0x20, 0x31, 0x05, 0x05, 0x15, +/*1400*/0x12, 0x0E, 0x77, 0xE5, 0x15, 0xB4, 0x08, 0x04, + 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x15, +/*1410*/0xB4, 0x07, 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, + 0x00, 0xEE, 0x4F, 0x60, 0x02, 0x05, 0x7F, 0x22, +/*1420*/0x85, 0x55, 0x82, 0x85, 0x54, 0x83, 0xE5, 0x15, + 0xF0, 0x12, 0x17, 0xA3, 0x22, 0x12, 0x07, 0x2A, +/*1430*/0x75, 0x83, 0xAE, 0x74, 0xFF, 0x12, 0x07, 0x29, + 0xE0, 0x54, 0x1A, 0xF5, 0x34, 0xE0, 0xC4, 0x13, +/*1440*/0x54, 0x07, 0xF5, 0x35, 0x24, 0xFE, 0x60, 0x24, + 0x24, 0xFE, 0x60, 0x3C, 0x24, 0x04, 0x70, 0x63, +/*1450*/0x75, 0x31, 0x2D, 0xE5, 0x08, 0xFD, 0x74, 0xB6, + 0x12, 0x07, 0x92, 0x74, 0xBC, 0x90, 0x07, 0x22, +/*1460*/0x12, 0x07, 0x95, 0x74, 0x90, 0x12, 0x07, 0xB3, + 0x74, 0x92, 0x80, 0x3C, 0x75, 0x31, 0x3A, 0xE5, +/*1470*/0x08, 0xFD, 0x74, 0xBA, 0x12, 0x07, 0x92, 0x74, + 0xC0, 0x90, 0x07, 0x22, 0x12, 0x07, 0xB6, 0x74, +/*1480*/0xC4, 0x12, 0x07, 0xB3, 0x74, 0xC8, 0x80, 0x20, + 0x75, 0x31, 0x35, 0xE5, 0x08, 0xFD, 0x74, 0xB8, +/*1490*/0x12, 0x07, 0x92, 0x74, 0xBE, 0xFF, 0xED, 0x44, + 0x07, 0x90, 0x07, 0x22, 0xCF, 0xF0, 0xA3, 0xEF, +/*14A0*/0xF0, 0x74, 0xC2, 0x12, 0x07, 0xB3, 0x74, 0xC6, + 0xFF, 0xED, 0x44, 0x07, 0xA3, 0xCF, 0xF0, 0xA3, +/*14B0*/0xEF, 0xF0, 0x22, 0x75, 0x34, 0x01, 0x22, 0x8E, + 0x58, 0x8F, 0x59, 0x8C, 0x5A, 0x8D, 0x5B, 0x8A, +/*14C0*/0x5C, 0x8B, 0x5D, 0x75, 0x5E, 0x01, 0xE4, 0xF5, + 0x5F, 0x12, 0x1E, 0xA5, 0x85, 0x59, 0x5E, 0xD3, +/*14D0*/0xE5, 0x5E, 0x95, 0x5B, 0xE5, 0x5A, 0x12, 0x07, + 0x6B, 0x50, 0x57, 0xE5, 0x5D, 0x45, 0x5C, 0x70, +/*14E0*/0x30, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x92, 0xE5, + 0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC6, 0xE5, +/*14F0*/0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC8, 0xE5, + 0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0x90, 0xE5, +/*1500*/0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE5, + 0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0x80, +/*1510*/0x03, 0x12, 0x07, 0x32, 0xE5, 0x5E, 0xF0, 0xAF, + 0x5F, 0x7E, 0x00, 0xAD, 0x5D, 0xAC, 0x5C, 0x12, +/*1520*/0x04, 0x44, 0xAF, 0x5E, 0x7E, 0x00, 0xAD, 0x5D, + 0xAC, 0x5C, 0x12, 0x0B, 0xD1, 0x05, 0x5E, 0x02, +/*1530*/0x14, 0xCF, 0xAB, 0x5D, 0xAA, 0x5C, 0xAD, 0x5B, + 0xAC, 0x5A, 0xAF, 0x59, 0xAE, 0x58, 0x02, 0x1B, +/*1540*/0xFB, 0x8C, 0x5C, 0x8D, 0x5D, 0x8A, 0x5E, 0x8B, + 0x5F, 0x75, 0x60, 0x01, 0xE4, 0xF5, 0x61, 0xF5, +/*1550*/0x62, 0xF5, 0x63, 0x12, 0x1E, 0xA5, 0x8F, 0x60, + 0xD3, 0xE5, 0x60, 0x95, 0x5D, 0xE5, 0x5C, 0x12, +/*1560*/0x07, 0x6B, 0x50, 0x61, 0xE5, 0x5F, 0x45, 0x5E, + 0x70, 0x27, 0x12, 0x07, 0x2A, 0x75, 0x83, 0xB6, +/*1570*/0xE5, 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0xB8, + 0xE5, 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0xBA, +/*1580*/0xE5, 0x60, 0xF0, 0xAF, 0x61, 0x7E, 0x00, 0xE5, + 0x62, 0x12, 0x08, 0x7A, 0x12, 0x0A, 0xFF, 0x80, +/*1590*/0x19, 0x90, 0x07, 0x24, 0x12, 0x07, 0x35, 0xE5, + 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0x8E, 0xE4, +/*15A0*/0x12, 0x07, 0x29, 0x74, 0x01, 0x12, 0x07, 0x29, + 0xE4, 0xF0, 0xAF, 0x63, 0x7E, 0x00, 0xAD, 0x5F, +/*15B0*/0xAC, 0x5E, 0x12, 0x04, 0x44, 0xAF, 0x60, 0x7E, + 0x00, 0xAD, 0x5F, 0xAC, 0x5E, 0x12, 0x12, 0x8B, +/*15C0*/0x05, 0x60, 0x02, 0x15, 0x58, 0x22, 0x90, 0x11, + 0x4D, 0xE4, 0x93, 0x90, 0x07, 0x2E, 0xF0, 0x12, +/*15D0*/0x08, 0x1F, 0x75, 0x83, 0xAE, 0xE0, 0x54, 0x1A, + 0xF5, 0x34, 0x70, 0x67, 0xEF, 0x44, 0x07, 0xF5, +/*15E0*/0x82, 0x75, 0x83, 0xCE, 0xE0, 0xFF, 0x13, 0x13, + 0x13, 0x54, 0x07, 0xF5, 0x36, 0x54, 0x0F, 0xD3, +/*15F0*/0x94, 0x00, 0x40, 0x06, 0x12, 0x14, 0x2D, 0x12, + 0x1B, 0xA9, 0xE5, 0x36, 0x54, 0x0F, 0x24, 0xFE, +/*1600*/0x60, 0x0C, 0x14, 0x60, 0x0C, 0x14, 0x60, 0x19, + 0x24, 0x03, 0x70, 0x37, 0x80, 0x10, 0x02, 0x1E, +/*1610*/0x91, 0x12, 0x1E, 0x91, 0x12, 0x07, 0x2A, 0x75, + 0x83, 0xCE, 0xE0, 0x54, 0xEF, 0xF0, 0x02, 0x1D, +/*1620*/0xAE, 0x12, 0x10, 0x14, 0xE4, 0xF5, 0x55, 0x12, + 0x1D, 0x85, 0x05, 0x55, 0xE5, 0x55, 0xC3, 0x94, +/*1630*/0x05, 0x40, 0xF4, 0x12, 0x07, 0x2A, 0x75, 0x83, + 0xCE, 0xE0, 0x54, 0xC7, 0x12, 0x07, 0x29, 0xE0, +/*1640*/0x44, 0x08, 0xF0, 0x22, 0xE4, 0xF5, 0x58, 0xF5, + 0x59, 0xAF, 0x08, 0xEF, 0x44, 0x07, 0xF5, 0x82, +/*1650*/0x75, 0x83, 0xD0, 0xE0, 0xFD, 0xC4, 0x54, 0x0F, + 0xF5, 0x5A, 0xEF, 0x44, 0x07, 0xF5, 0x82, 0x75, +/*1660*/0x83, 0x80, 0x74, 0x01, 0xF0, 0x12, 0x08, 0x21, + 0x75, 0x83, 0x82, 0xE5, 0x45, 0xF0, 0xEF, 0x44, +/*1670*/0x07, 0xF5, 0x82, 0x75, 0x83, 0x8A, 0x74, 0xFF, + 0xF0, 0x12, 0x1A, 0x4D, 0x12, 0x07, 0x2A, 0x75, +/*1680*/0x83, 0xBC, 0xE0, 0x54, 0xEF, 0x12, 0x07, 0x29, + 0x75, 0x83, 0xBE, 0xE0, 0x54, 0xEF, 0x12, 0x07, +/*1690*/0x29, 0x75, 0x83, 0xC0, 0xE0, 0x54, 0xEF, 0x12, + 0x07, 0x29, 0x75, 0x83, 0xBC, 0xE0, 0x44, 0x10, +/*16A0*/0x12, 0x07, 0x29, 0x75, 0x83, 0xBE, 0xE0, 0x44, + 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC0, 0xE0, +/*16B0*/0x44, 0x10, 0xF0, 0xAF, 0x58, 0xE5, 0x59, 0x12, + 0x08, 0x78, 0x02, 0x0A, 0xFF, 0xE4, 0xF5, 0x58, +/*16C0*/0x7D, 0x01, 0xF5, 0x59, 0xAF, 0x35, 0xFE, 0xFC, + 0x12, 0x09, 0x15, 0x12, 0x07, 0x2A, 0x75, 0x83, +/*16D0*/0xB6, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, + 0xB8, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, +/*16E0*/0xBA, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, + 0xBC, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, +/*16F0*/0xBE, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, + 0xC0, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83, +/*1700*/0x90, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, + 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0xE4, +/*1710*/0x12, 0x07, 0x29, 0x75, 0x83, 0x92, 0xE4, 0x12, + 0x07, 0x29, 0x75, 0x83, 0xC6, 0xE4, 0x12, 0x07, +/*1720*/0x29, 0x75, 0x83, 0xC8, 0xE4, 0xF0, 0xAF, 0x58, + 0xFE, 0xE5, 0x59, 0x12, 0x08, 0x7A, 0x02, 0x0A, +/*1730*/0xFF, 0xE5, 0xE2, 0x30, 0xE4, 0x6C, 0xE5, 0xE7, + 0x54, 0xC0, 0x64, 0x40, 0x70, 0x64, 0xE5, 0x09, +/*1740*/0xC4, 0x54, 0x30, 0xFE, 0xE5, 0x08, 0x25, 0xE0, + 0x25, 0xE0, 0x54, 0xC0, 0x4E, 0xFE, 0xEF, 0x54, +/*1750*/0x3F, 0x4E, 0xFD, 0xE5, 0x2B, 0xAE, 0x2A, 0x78, + 0x02, 0xC3, 0x33, 0xCE, 0x33, 0xCE, 0xD8, 0xF9, +/*1760*/0xF5, 0x82, 0x8E, 0x83, 0xED, 0xF0, 0xE5, 0x2B, + 0xAE, 0x2A, 0x78, 0x02, 0xC3, 0x33, 0xCE, 0x33, +/*1770*/0xCE, 0xD8, 0xF9, 0xFF, 0xF5, 0x82, 0x8E, 0x83, + 0xA3, 0xE5, 0xFE, 0xF0, 0x8F, 0x82, 0x8E, 0x83, +/*1780*/0xA3, 0xA3, 0xE5, 0xFD, 0xF0, 0x8F, 0x82, 0x8E, + 0x83, 0xA3, 0xA3, 0xA3, 0xE5, 0xFC, 0xF0, 0xC3, +/*1790*/0xE5, 0x2B, 0x94, 0xFA, 0xE5, 0x2A, 0x94, 0x00, + 0x50, 0x08, 0x05, 0x2B, 0xE5, 0x2B, 0x70, 0x02, +/*17A0*/0x05, 0x2A, 0x22, 0xE4, 0xFF, 0xE4, 0xF5, 0x58, + 0xF5, 0x56, 0xF5, 0x57, 0x74, 0x82, 0xFC, 0x12, +/*17B0*/0x0E, 0x04, 0x8C, 0x83, 0xE0, 0xF5, 0x10, 0x54, + 0x7F, 0xF0, 0xE5, 0x10, 0x44, 0x80, 0x12, 0x0E, +/*17C0*/0x98, 0xED, 0xF0, 0x7E, 0x0A, 0x12, 0x0E, 0x04, + 0x75, 0x83, 0xA0, 0xE0, 0x20, 0xE0, 0x26, 0xDE, +/*17D0*/0xF4, 0x05, 0x57, 0xE5, 0x57, 0x70, 0x02, 0x05, + 0x56, 0xE5, 0x14, 0x24, 0x01, 0xFD, 0xE4, 0x33, +/*17E0*/0xFC, 0xD3, 0xE5, 0x57, 0x9D, 0xE5, 0x56, 0x9C, + 0x40, 0xD9, 0xE5, 0x0A, 0x94, 0x20, 0x50, 0x02, +/*17F0*/0x05, 0x0A, 0x43, 0xE1, 0x08, 0xC2, 0x31, 0x12, + 0x0E, 0x04, 0x75, 0x83, 0xA6, 0xE0, 0x55, 0x12, +/*1800*/0x65, 0x12, 0x70, 0x03, 0xD2, 0x31, 0x22, 0xC2, + 0x31, 0x22, 0x90, 0x07, 0x26, 0xE0, 0xFA, 0xA3, +/*1810*/0xE0, 0xF5, 0x82, 0x8A, 0x83, 0xE0, 0xF5, 0x41, + 0xE5, 0x39, 0xC3, 0x95, 0x41, 0x40, 0x26, 0xE5, +/*1820*/0x39, 0x95, 0x41, 0xC3, 0x9F, 0xEE, 0x12, 0x07, + 0x6B, 0x40, 0x04, 0x7C, 0x01, 0x80, 0x02, 0x7C, +/*1830*/0x00, 0xE5, 0x41, 0x64, 0x3F, 0x60, 0x04, 0x7B, + 0x01, 0x80, 0x02, 0x7B, 0x00, 0xEC, 0x5B, 0x60, +/*1840*/0x29, 0x05, 0x41, 0x80, 0x28, 0xC3, 0xE5, 0x41, + 0x95, 0x39, 0xC3, 0x9F, 0xEE, 0x12, 0x07, 0x6B, +/*1850*/0x40, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, + 0xE5, 0x41, 0x60, 0x04, 0x7E, 0x01, 0x80, 0x02, +/*1860*/0x7E, 0x00, 0xEF, 0x5E, 0x60, 0x04, 0x15, 0x41, + 0x80, 0x03, 0x85, 0x39, 0x41, 0x85, 0x3A, 0x40, +/*1870*/0x22, 0xE5, 0xE2, 0x30, 0xE4, 0x60, 0xE5, 0xE1, + 0x30, 0xE2, 0x5B, 0xE5, 0x09, 0x70, 0x04, 0x7F, +/*1880*/0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x08, 0x70, + 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, +/*1890*/0x5F, 0x60, 0x43, 0x53, 0xF9, 0xF8, 0xE5, 0xE2, + 0x30, 0xE4, 0x3B, 0xE5, 0xE1, 0x30, 0xE2, 0x2E, +/*18A0*/0x43, 0xFA, 0x02, 0x53, 0xFA, 0xFB, 0xE4, 0xF5, + 0x10, 0x90, 0x94, 0x70, 0xE5, 0x10, 0xF0, 0xE5, +/*18B0*/0xE1, 0x30, 0xE2, 0xE7, 0x90, 0x94, 0x70, 0xE0, + 0x65, 0x10, 0x60, 0x03, 0x43, 0xFA, 0x04, 0x05, +/*18C0*/0x10, 0x90, 0x94, 0x70, 0xE5, 0x10, 0xF0, 0x70, + 0xE6, 0x12, 0x00, 0x06, 0x80, 0xE1, 0x53, 0xFA, +/*18D0*/0xFD, 0x53, 0xFA, 0xFB, 0x80, 0xC0, 0x22, 0x8F, + 0x54, 0x12, 0x00, 0x06, 0xE5, 0xE1, 0x30, 0xE0, +/*18E0*/0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, + 0x7E, 0xD3, 0x94, 0x05, 0x40, 0x04, 0x7E, 0x01, +/*18F0*/0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x60, 0x3D, + 0x85, 0x54, 0x11, 0xE5, 0xE2, 0x20, 0xE1, 0x32, +/*1900*/0x74, 0xCE, 0x12, 0x1A, 0x05, 0x30, 0xE7, 0x04, + 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, 0x8F, 0x82, +/*1910*/0x8E, 0x83, 0xE0, 0x30, 0xE6, 0x04, 0x7F, 0x01, + 0x80, 0x02, 0x7F, 0x00, 0xEF, 0x5D, 0x70, 0x15, +/*1920*/0x12, 0x15, 0xC6, 0x74, 0xCE, 0x12, 0x1A, 0x05, + 0x30, 0xE6, 0x07, 0xE0, 0x44, 0x80, 0xF0, 0x43, +/*1930*/0xF9, 0x80, 0x12, 0x18, 0x71, 0x22, 0x12, 0x0E, + 0x44, 0xE5, 0x16, 0x25, 0xE0, 0x25, 0xE0, 0x24, +/*1940*/0xB0, 0xF5, 0x82, 0xE4, 0x34, 0x1A, 0xF5, 0x83, + 0xE4, 0x93, 0xF5, 0x0F, 0xE5, 0x16, 0x25, 0xE0, +/*1950*/0x25, 0xE0, 0x24, 0xB1, 0xF5, 0x82, 0xE4, 0x34, + 0x1A, 0xF5, 0x83, 0xE4, 0x93, 0xF5, 0x0E, 0x12, +/*1960*/0x0E, 0x65, 0xF5, 0x10, 0xE5, 0x0F, 0x54, 0xF0, + 0x12, 0x0E, 0x17, 0x75, 0x83, 0x8C, 0xEF, 0xF0, +/*1970*/0xE5, 0x0F, 0x30, 0xE0, 0x0C, 0x12, 0x0E, 0x04, + 0x75, 0x83, 0x86, 0xE0, 0x44, 0x40, 0xF0, 0x80, +/*1980*/0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0x86, 0xE0, + 0x54, 0xBF, 0xF0, 0x12, 0x0E, 0x91, 0x75, 0x83, +/*1990*/0x82, 0xE5, 0x0E, 0xF0, 0x22, 0x7F, 0x05, 0x12, + 0x17, 0x31, 0x12, 0x0E, 0x04, 0x12, 0x0E, 0x33, +/*19A0*/0x74, 0x02, 0xF0, 0x74, 0x8E, 0xFE, 0x12, 0x0E, + 0x04, 0x12, 0x0E, 0x0B, 0xEF, 0xF0, 0x75, 0x15, +/*19B0*/0x70, 0x12, 0x0F, 0xF7, 0x20, 0x34, 0x05, 0x75, + 0x15, 0x10, 0x80, 0x03, 0x75, 0x15, 0x50, 0x12, +/*19C0*/0x0F, 0xF7, 0x20, 0x34, 0x04, 0x74, 0x10, 0x80, + 0x02, 0x74, 0xF0, 0x25, 0x15, 0xF5, 0x15, 0x12, +/*19D0*/0x0E, 0x21, 0xEF, 0xF0, 0x12, 0x10, 0x91, 0x20, + 0x34, 0x17, 0xE5, 0x15, 0x64, 0x30, 0x60, 0x0C, +/*19E0*/0x74, 0x10, 0x25, 0x15, 0xF5, 0x15, 0xB4, 0x80, + 0x03, 0xE4, 0xF5, 0x15, 0x12, 0x0E, 0x21, 0xEF, +/*19F0*/0xF0, 0x22, 0xF0, 0xE5, 0x0B, 0x25, 0xE0, 0x25, + 0xE0, 0x24, 0x82, 0xF5, 0x82, 0xE4, 0x34, 0x07, +/*1A00*/0xF5, 0x83, 0x22, 0x74, 0x88, 0xFE, 0xE5, 0x08, + 0x44, 0x07, 0xFF, 0xF5, 0x82, 0x8E, 0x83, 0xE0, +/*1A10*/0x22, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82, + 0x22, 0xF0, 0xE0, 0x54, 0xC0, 0x8F, 0x82, 0x8E, +/*1A20*/0x83, 0xF0, 0x22, 0xEF, 0x44, 0x07, 0xF5, 0x82, + 0x75, 0x83, 0x86, 0xE0, 0x54, 0x10, 0xD3, 0x94, +/*1A30*/0x00, 0x22, 0xF0, 0x90, 0x07, 0x15, 0xE0, 0x04, + 0xF0, 0x22, 0x44, 0x06, 0xF5, 0x82, 0x75, 0x83, +/*1A40*/0x9E, 0xE0, 0x22, 0xFE, 0xEF, 0x44, 0x07, 0xF5, + 0x82, 0x8E, 0x83, 0xE0, 0x22, 0xE4, 0x90, 0x07, +/*1A50*/0x2A, 0xF0, 0xA3, 0xF0, 0x12, 0x07, 0x2A, 0x75, + 0x83, 0x82, 0xE0, 0x54, 0x7F, 0x12, 0x07, 0x29, +/*1A60*/0xE0, 0x44, 0x80, 0xF0, 0x12, 0x10, 0xFC, 0x12, + 0x08, 0x1F, 0x75, 0x83, 0xA0, 0xE0, 0x20, 0xE0, +/*1A70*/0x1A, 0x90, 0x07, 0x2B, 0xE0, 0x04, 0xF0, 0x70, + 0x06, 0x90, 0x07, 0x2A, 0xE0, 0x04, 0xF0, 0x90, +/*1A80*/0x07, 0x2A, 0xE0, 0xB4, 0x10, 0xE1, 0xA3, 0xE0, + 0xB4, 0x00, 0xDC, 0xEE, 0x44, 0xA6, 0xFC, 0xEF, +/*1A90*/0x44, 0x07, 0xF5, 0x82, 0x8C, 0x83, 0xE0, 0xF5, + 0x32, 0xEE, 0x44, 0xA8, 0xFE, 0xEF, 0x44, 0x07, +/*1AA0*/0xF5, 0x82, 0x8E, 0x83, 0xE0, 0xF5, 0x33, 0x22, + 0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x90, +/*1AB0*/0x00, 0x20, 0x0F, 0x92, 0x00, 0x21, 0x0F, 0x94, + 0x00, 0x22, 0x0F, 0x96, 0x00, 0x23, 0x0F, 0x98, +/*1AC0*/0x00, 0x24, 0x0F, 0x9A, 0x00, 0x25, 0x0F, 0x9C, + 0x00, 0x26, 0x0F, 0x9E, 0x00, 0x27, 0x0F, 0xA0, +/*1AD0*/0x01, 0x20, 0x01, 0xA2, 0x01, 0x21, 0x01, 0xA4, + 0x01, 0x22, 0x01, 0xA6, 0x01, 0x23, 0x01, 0xA8, +/*1AE0*/0x01, 0x24, 0x01, 0xAA, 0x01, 0x25, 0x01, 0xAC, + 0x01, 0x26, 0x01, 0xAE, 0x01, 0x27, 0x01, 0xB0, +/*1AF0*/0x01, 0x28, 0x01, 0xB4, 0x00, 0x28, 0x0F, 0xB6, + 0x40, 0x28, 0x0F, 0xB8, 0x61, 0x28, 0x01, 0xCB, +/*1B00*/0xEF, 0xCB, 0xCA, 0xEE, 0xCA, 0x7F, 0x01, 0xE4, + 0xFD, 0xEB, 0x4A, 0x70, 0x24, 0xE5, 0x08, 0xF5, +/*1B10*/0x82, 0x74, 0xB6, 0x12, 0x08, 0x29, 0xE5, 0x08, + 0xF5, 0x82, 0x74, 0xB8, 0x12, 0x08, 0x29, 0xE5, +/*1B20*/0x08, 0xF5, 0x82, 0x74, 0xBA, 0x12, 0x08, 0x29, + 0x7E, 0x00, 0x7C, 0x00, 0x12, 0x0A, 0xFF, 0x80, +/*1B30*/0x12, 0x90, 0x07, 0x26, 0x12, 0x07, 0x35, 0xE5, + 0x41, 0xF0, 0x90, 0x07, 0x24, 0x12, 0x07, 0x35, +/*1B40*/0xE5, 0x40, 0xF0, 0x12, 0x07, 0x2A, 0x75, 0x83, + 0x8E, 0xE4, 0x12, 0x07, 0x29, 0x74, 0x01, 0x12, +/*1B50*/0x07, 0x29, 0xE4, 0xF0, 0x22, 0xE4, 0xF5, 0x26, + 0xF5, 0x27, 0x53, 0xE1, 0xFE, 0xF5, 0x2A, 0x75, +/*1B60*/0x2B, 0x01, 0xF5, 0x08, 0x7F, 0x01, 0x12, 0x17, + 0x31, 0x30, 0x30, 0x1C, 0x90, 0x1A, 0xA9, 0xE4, +/*1B70*/0x93, 0xF5, 0x10, 0x90, 0x1F, 0xF9, 0xE4, 0x93, + 0xF5, 0x10, 0x90, 0x00, 0x41, 0xE4, 0x93, 0xF5, +/*1B80*/0x10, 0x90, 0x1E, 0xCA, 0xE4, 0x93, 0xF5, 0x10, + 0x7F, 0x02, 0x12, 0x17, 0x31, 0x12, 0x0F, 0x54, +/*1B90*/0x7F, 0x03, 0x12, 0x17, 0x31, 0x12, 0x00, 0x06, + 0xE5, 0xE2, 0x30, 0xE7, 0x09, 0x12, 0x10, 0x00, +/*1BA0*/0x30, 0x30, 0x03, 0x12, 0x11, 0x00, 0x02, 0x00, + 0x47, 0x12, 0x08, 0x1F, 0x75, 0x83, 0xD0, 0xE0, +/*1BB0*/0xC4, 0x54, 0x0F, 0xFD, 0x75, 0x43, 0x01, 0x75, + 0x44, 0xFF, 0x12, 0x08, 0xAA, 0x74, 0x04, 0xF0, +/*1BC0*/0x75, 0x3B, 0x01, 0xED, 0x14, 0x60, 0x0C, 0x14, + 0x60, 0x0B, 0x14, 0x60, 0x0F, 0x24, 0x03, 0x70, +/*1BD0*/0x0B, 0x80, 0x09, 0x80, 0x00, 0x12, 0x08, 0xA7, + 0x04, 0xF0, 0x80, 0x06, 0x12, 0x08, 0xA7, 0x74, +/*1BE0*/0x04, 0xF0, 0xEE, 0x44, 0x82, 0xFE, 0xEF, 0x44, + 0x07, 0xF5, 0x82, 0x8E, 0x83, 0xE5, 0x45, 0x12, +/*1BF0*/0x08, 0xBE, 0x75, 0x83, 0x82, 0xE5, 0x31, 0xF0, + 0x02, 0x11, 0x4C, 0x8E, 0x60, 0x8F, 0x61, 0x12, +/*1C00*/0x1E, 0xA5, 0xE4, 0xFF, 0xCE, 0xED, 0xCE, 0xEE, + 0xD3, 0x95, 0x61, 0xE5, 0x60, 0x12, 0x07, 0x6B, +/*1C10*/0x40, 0x39, 0x74, 0x20, 0x2E, 0xF5, 0x82, 0xE4, + 0x34, 0x03, 0xF5, 0x83, 0xE0, 0x70, 0x03, 0xFF, +/*1C20*/0x80, 0x26, 0x12, 0x08, 0xE2, 0xFD, 0xC3, 0x9F, + 0x40, 0x1E, 0xCF, 0xED, 0xCF, 0xEB, 0x4A, 0x70, +/*1C30*/0x0B, 0x8D, 0x42, 0x12, 0x08, 0xEE, 0xF5, 0x41, + 0x8E, 0x40, 0x80, 0x0C, 0x12, 0x08, 0xE2, 0xF5, +/*1C40*/0x38, 0x12, 0x08, 0xEE, 0xF5, 0x39, 0x8E, 0x3A, + 0x1E, 0x80, 0xBC, 0x22, 0x75, 0x58, 0x01, 0xE5, +/*1C50*/0x35, 0x70, 0x0C, 0x12, 0x07, 0xCC, 0xE0, 0xF5, + 0x4A, 0x12, 0x07, 0xD8, 0xE0, 0xF5, 0x4C, 0xE5, +/*1C60*/0x35, 0xB4, 0x04, 0x0C, 0x12, 0x07, 0xE4, 0xE0, + 0xF5, 0x4A, 0x12, 0x07, 0xF0, 0xE0, 0xF5, 0x4C, +/*1C70*/0xE5, 0x35, 0xB4, 0x01, 0x04, 0x7F, 0x01, 0x80, + 0x02, 0x7F, 0x00, 0xE5, 0x35, 0xB4, 0x02, 0x04, +/*1C80*/0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, + 0x60, 0x0C, 0x12, 0x07, 0xFC, 0xE0, 0xF5, 0x4A, +/*1C90*/0x12, 0x08, 0x08, 0xE0, 0xF5, 0x4C, 0x85, 0x41, + 0x49, 0x85, 0x40, 0x4B, 0x22, 0x75, 0x5B, 0x01, +/*1CA0*/0x90, 0x07, 0x24, 0x12, 0x07, 0x35, 0xE0, 0x54, + 0x1F, 0xFF, 0xD3, 0x94, 0x02, 0x50, 0x04, 0x8F, +/*1CB0*/0x58, 0x80, 0x05, 0xEF, 0x24, 0xFE, 0xF5, 0x58, + 0xEF, 0xC3, 0x94, 0x18, 0x40, 0x05, 0x75, 0x59, +/*1CC0*/0x18, 0x80, 0x04, 0xEF, 0x04, 0xF5, 0x59, 0x85, + 0x43, 0x5A, 0xAF, 0x58, 0x7E, 0x00, 0xAD, 0x59, +/*1CD0*/0x7C, 0x00, 0xAB, 0x5B, 0x7A, 0x00, 0x12, 0x15, + 0x41, 0xAF, 0x5A, 0x7E, 0x00, 0x12, 0x18, 0x0A, +/*1CE0*/0xAF, 0x5B, 0x7E, 0x00, 0x02, 0x1A, 0xFF, 0xE5, + 0xE2, 0x30, 0xE7, 0x0E, 0x12, 0x10, 0x03, 0xC2, +/*1CF0*/0x30, 0x30, 0x30, 0x03, 0x12, 0x10, 0xFF, 0x20, + 0x33, 0x28, 0xE5, 0xE7, 0x30, 0xE7, 0x05, 0x12, +/*1D00*/0x0E, 0xA2, 0x80, 0x0D, 0xE5, 0xFE, 0xC3, 0x94, + 0x20, 0x50, 0x06, 0x12, 0x0E, 0xA2, 0x43, 0xF9, +/*1D10*/0x08, 0xE5, 0xF2, 0x30, 0xE7, 0x03, 0x53, 0xF9, + 0x7F, 0xE5, 0xF1, 0x54, 0x70, 0xD3, 0x94, 0x00, +/*1D20*/0x50, 0xD8, 0x22, 0x12, 0x0E, 0x04, 0x75, 0x83, + 0x80, 0xE4, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0x12, +/*1D30*/0x0D, 0xFD, 0x75, 0x83, 0x84, 0x12, 0x0E, 0x02, + 0x75, 0x83, 0x86, 0x12, 0x0E, 0x02, 0x75, 0x83, +/*1D40*/0x8C, 0xE0, 0x54, 0xF3, 0x12, 0x0E, 0x03, 0x75, + 0x83, 0x8E, 0x12, 0x0E, 0x02, 0x75, 0x83, 0x94, +/*1D50*/0xE0, 0x54, 0xFB, 0xF0, 0x22, 0x12, 0x07, 0x2A, + 0x75, 0x83, 0x8E, 0xE4, 0x12, 0x07, 0x29, 0x74, +/*1D60*/0x01, 0x12, 0x07, 0x29, 0xE4, 0x12, 0x08, 0xBE, + 0x75, 0x83, 0x8C, 0xE0, 0x44, 0x20, 0x12, 0x08, +/*1D70*/0xBE, 0xE0, 0x54, 0xDF, 0xF0, 0x74, 0x84, 0x85, + 0x08, 0x82, 0xF5, 0x83, 0xE0, 0x54, 0x7F, 0xF0, +/*1D80*/0xE0, 0x44, 0x80, 0xF0, 0x22, 0x75, 0x56, 0x01, + 0xE4, 0xFD, 0xF5, 0x57, 0xAF, 0x35, 0xFE, 0xFC, +/*1D90*/0x12, 0x09, 0x15, 0x12, 0x1C, 0x9D, 0x12, 0x1E, + 0x7A, 0x12, 0x1C, 0x4C, 0xAF, 0x57, 0x7E, 0x00, +/*1DA0*/0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, 0xAF, + 0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0x75, 0x56, +/*1DB0*/0x01, 0xE4, 0xFD, 0xF5, 0x57, 0xAF, 0x35, 0xFE, + 0xFC, 0x12, 0x09, 0x15, 0x12, 0x1C, 0x9D, 0x12, +/*1DC0*/0x1E, 0x7A, 0x12, 0x1C, 0x4C, 0xAF, 0x57, 0x7E, + 0x00, 0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, +/*1DD0*/0xAF, 0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0xE4, + 0xF5, 0x16, 0x12, 0x0E, 0x44, 0xFE, 0xE5, 0x08, +/*1DE0*/0x44, 0x05, 0xFF, 0x12, 0x0E, 0x65, 0x8F, 0x82, + 0x8E, 0x83, 0xF0, 0x05, 0x16, 0xE5, 0x16, 0xC3, +/*1DF0*/0x94, 0x14, 0x40, 0xE6, 0xE5, 0x08, 0x12, 0x0E, + 0x2B, 0xE4, 0xF0, 0x22, 0xE4, 0xF5, 0x58, 0xF5, +/*1E00*/0x59, 0xF5, 0x5A, 0xFF, 0xFE, 0xAD, 0x58, 0xFC, + 0x12, 0x09, 0x15, 0x7F, 0x04, 0x7E, 0x00, 0xAD, +/*1E10*/0x58, 0x7C, 0x00, 0x12, 0x09, 0x15, 0x7F, 0x02, + 0x7E, 0x00, 0xAD, 0x58, 0x7C, 0x00, 0x02, 0x09, +/*1E20*/0x15, 0xE5, 0x3C, 0x25, 0x3E, 0xFC, 0xE5, 0x42, + 0x24, 0x00, 0xFB, 0xE4, 0x33, 0xFA, 0xEC, 0xC3, +/*1E30*/0x9B, 0xEA, 0x12, 0x07, 0x6B, 0x40, 0x0B, 0x8C, + 0x42, 0xE5, 0x3D, 0x25, 0x3F, 0xF5, 0x41, 0x8F, +/*1E40*/0x40, 0x22, 0x12, 0x09, 0x0B, 0x22, 0x74, 0x84, + 0xF5, 0x18, 0x85, 0x08, 0x19, 0x85, 0x19, 0x82, +/*1E50*/0x85, 0x18, 0x83, 0xE0, 0x54, 0x7F, 0xF0, 0xE0, + 0x44, 0x80, 0xF0, 0xE0, 0x44, 0x80, 0xF0, 0x22, +/*1E60*/0xEF, 0x4E, 0x70, 0x0B, 0x12, 0x07, 0x2A, 0x75, + 0x83, 0xD2, 0xE0, 0x54, 0xDF, 0xF0, 0x22, 0x12, +/*1E70*/0x07, 0x2A, 0x75, 0x83, 0xD2, 0xE0, 0x44, 0x20, + 0xF0, 0x22, 0x75, 0x58, 0x01, 0x90, 0x07, 0x26, +/*1E80*/0x12, 0x07, 0x35, 0xE0, 0x54, 0x3F, 0xF5, 0x41, + 0x12, 0x07, 0x32, 0xE0, 0x54, 0x3F, 0xF5, 0x40, +/*1E90*/0x22, 0x75, 0x56, 0x02, 0xE4, 0xF5, 0x57, 0x12, + 0x1D, 0xFC, 0xAF, 0x57, 0x7E, 0x00, 0xAD, 0x56, +/*1EA0*/0x7C, 0x00, 0x02, 0x04, 0x44, 0xE4, 0xF5, 0x42, + 0xF5, 0x41, 0xF5, 0x40, 0xF5, 0x38, 0xF5, 0x39, +/*1EB0*/0xF5, 0x3A, 0x22, 0xEF, 0x54, 0x07, 0xFF, 0xE5, + 0xF9, 0x54, 0xF8, 0x4F, 0xF5, 0xF9, 0x22, 0x7F, +/*1EC0*/0x01, 0xE4, 0xFE, 0x0F, 0x0E, 0xBE, 0xFF, 0xFB, + 0x22, 0x01, 0x20, 0x00, 0x01, 0x04, 0x20, 0x00, +/*1ED0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1EE0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1EF0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F00*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F10*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F20*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F30*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F40*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F50*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F60*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F70*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F80*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1F90*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1FA0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1FB0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1FC0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1FD0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1FE0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, +/*1FF0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x81 +}; + +int ipath_sd7220_ib_load(struct ipath_devdata *dd) +{ + return ipath_sd7220_prog_ld(dd, IB_7220_SERDES, ipath_sd7220_ib_img, + sizeof(ipath_sd7220_ib_img), 0); +} + +int ipath_sd7220_ib_vfy(struct ipath_devdata *dd) +{ + return ipath_sd7220_prog_vfy(dd, IB_7220_SERDES, ipath_sd7220_ib_img, + sizeof(ipath_sd7220_ib_img), 0); +} diff --git a/drivers/infiniband/hw/ipath/ipath_sdma.c b/drivers/infiniband/hw/ipath/ipath_sdma.c new file mode 100644 index 00000000000..1974df7a9f7 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_sdma.c @@ -0,0 +1,790 @@ +/* + * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/spinlock.h> + +#include "ipath_kernel.h" +#include "ipath_verbs.h" +#include "ipath_common.h" + +#define SDMA_DESCQ_SZ PAGE_SIZE /* 256 entries per 4KB page */ + +static void vl15_watchdog_enq(struct ipath_devdata *dd) +{ + /* ipath_sdma_lock must already be held */ + if (atomic_inc_return(&dd->ipath_sdma_vl15_count) == 1) { + unsigned long interval = (HZ + 19) / 20; + dd->ipath_sdma_vl15_timer.expires = jiffies + interval; + add_timer(&dd->ipath_sdma_vl15_timer); + } +} + +static void vl15_watchdog_deq(struct ipath_devdata *dd) +{ + /* ipath_sdma_lock must already be held */ + if (atomic_dec_return(&dd->ipath_sdma_vl15_count) != 0) { + unsigned long interval = (HZ + 19) / 20; + mod_timer(&dd->ipath_sdma_vl15_timer, jiffies + interval); + } else { + del_timer(&dd->ipath_sdma_vl15_timer); + } +} + +static void vl15_watchdog_timeout(unsigned long opaque) +{ + struct ipath_devdata *dd = (struct ipath_devdata *)opaque; + + if (atomic_read(&dd->ipath_sdma_vl15_count) != 0) { + ipath_dbg("vl15 watchdog timeout - clearing\n"); + ipath_cancel_sends(dd, 1); + ipath_hol_down(dd); + } else { + ipath_dbg("vl15 watchdog timeout - " + "condition already cleared\n"); + } +} + +static void unmap_desc(struct ipath_devdata *dd, unsigned head) +{ + __le64 *descqp = &dd->ipath_sdma_descq[head].qw[0]; + u64 desc[2]; + dma_addr_t addr; + size_t len; + + desc[0] = le64_to_cpu(descqp[0]); + desc[1] = le64_to_cpu(descqp[1]); + + addr = (desc[1] << 32) | (desc[0] >> 32); + len = (desc[0] >> 14) & (0x7ffULL << 2); + dma_unmap_single(&dd->pcidev->dev, addr, len, DMA_TO_DEVICE); +} + +/* + * ipath_sdma_lock should be locked before calling this. + */ +int ipath_sdma_make_progress(struct ipath_devdata *dd) +{ + struct list_head *lp = NULL; + struct ipath_sdma_txreq *txp = NULL; + u16 dmahead; + u16 start_idx = 0; + int progress = 0; + + if (!list_empty(&dd->ipath_sdma_activelist)) { + lp = dd->ipath_sdma_activelist.next; + txp = list_entry(lp, struct ipath_sdma_txreq, list); + start_idx = txp->start_idx; + } + + /* + * Read the SDMA head register in order to know that the + * interrupt clear has been written to the chip. + * Otherwise, we may not get an interrupt for the last + * descriptor in the queue. + */ + dmahead = (u16)ipath_read_kreg32(dd, dd->ipath_kregs->kr_senddmahead); + /* sanity check return value for error handling (chip reset, etc.) */ + if (dmahead >= dd->ipath_sdma_descq_cnt) + goto done; + + while (dd->ipath_sdma_descq_head != dmahead) { + if (txp && txp->flags & IPATH_SDMA_TXREQ_F_FREEDESC && + dd->ipath_sdma_descq_head == start_idx) { + unmap_desc(dd, dd->ipath_sdma_descq_head); + start_idx++; + if (start_idx == dd->ipath_sdma_descq_cnt) + start_idx = 0; + } + + /* increment free count and head */ + dd->ipath_sdma_descq_removed++; + if (++dd->ipath_sdma_descq_head == dd->ipath_sdma_descq_cnt) + dd->ipath_sdma_descq_head = 0; + + if (txp && txp->next_descq_idx == dd->ipath_sdma_descq_head) { + /* move to notify list */ + if (txp->flags & IPATH_SDMA_TXREQ_F_VL15) + vl15_watchdog_deq(dd); + list_move_tail(lp, &dd->ipath_sdma_notifylist); + if (!list_empty(&dd->ipath_sdma_activelist)) { + lp = dd->ipath_sdma_activelist.next; + txp = list_entry(lp, struct ipath_sdma_txreq, + list); + start_idx = txp->start_idx; + } else { + lp = NULL; + txp = NULL; + } + } + progress = 1; + } + + if (progress) + tasklet_hi_schedule(&dd->ipath_sdma_notify_task); + +done: + return progress; +} + +static void ipath_sdma_notify(struct ipath_devdata *dd, struct list_head *list) +{ + struct ipath_sdma_txreq *txp, *txp_next; + + list_for_each_entry_safe(txp, txp_next, list, list) { + list_del_init(&txp->list); + + if (txp->callback) + (*txp->callback)(txp->callback_cookie, + txp->callback_status); + } +} + +static void sdma_notify_taskbody(struct ipath_devdata *dd) +{ + unsigned long flags; + struct list_head list; + + INIT_LIST_HEAD(&list); + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + + list_splice_init(&dd->ipath_sdma_notifylist, &list); + + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + ipath_sdma_notify(dd, &list); + + /* + * The IB verbs layer needs to see the callback before getting + * the call to ipath_ib_piobufavail() because the callback + * handles releasing resources the next send will need. + * Otherwise, we could do these calls in + * ipath_sdma_make_progress(). + */ + ipath_ib_piobufavail(dd->verbs_dev); +} + +static void sdma_notify_task(unsigned long opaque) +{ + struct ipath_devdata *dd = (struct ipath_devdata *)opaque; + + if (!test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + sdma_notify_taskbody(dd); +} + +static void dump_sdma_state(struct ipath_devdata *dd) +{ + unsigned long reg; + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmastatus); + ipath_cdbg(VERBOSE, "kr_senddmastatus: 0x%016lx\n", reg); + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendctrl); + ipath_cdbg(VERBOSE, "kr_sendctrl: 0x%016lx\n", reg); + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmabufmask0); + ipath_cdbg(VERBOSE, "kr_senddmabufmask0: 0x%016lx\n", reg); + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmabufmask1); + ipath_cdbg(VERBOSE, "kr_senddmabufmask1: 0x%016lx\n", reg); + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmabufmask2); + ipath_cdbg(VERBOSE, "kr_senddmabufmask2: 0x%016lx\n", reg); + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmatail); + ipath_cdbg(VERBOSE, "kr_senddmatail: 0x%016lx\n", reg); + + reg = ipath_read_kreg64(dd, dd->ipath_kregs->kr_senddmahead); + ipath_cdbg(VERBOSE, "kr_senddmahead: 0x%016lx\n", reg); +} + +static void sdma_abort_task(unsigned long opaque) +{ + struct ipath_devdata *dd = (struct ipath_devdata *) opaque; + u64 status; + unsigned long flags; + + if (test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + return; + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + + status = dd->ipath_sdma_status & IPATH_SDMA_ABORT_MASK; + + /* nothing to do */ + if (status == IPATH_SDMA_ABORT_NONE) + goto unlock; + + /* ipath_sdma_abort() is done, waiting for interrupt */ + if (status == IPATH_SDMA_ABORT_DISARMED) { + if (jiffies < dd->ipath_sdma_abort_intr_timeout) + goto resched_noprint; + /* give up, intr got lost somewhere */ + ipath_dbg("give up waiting for SDMADISABLED intr\n"); + __set_bit(IPATH_SDMA_DISABLED, &dd->ipath_sdma_status); + status = IPATH_SDMA_ABORT_ABORTED; + } + + /* everything is stopped, time to clean up and restart */ + if (status == IPATH_SDMA_ABORT_ABORTED) { + struct ipath_sdma_txreq *txp, *txpnext; + u64 hwstatus; + int notify = 0; + + hwstatus = ipath_read_kreg64(dd, + dd->ipath_kregs->kr_senddmastatus); + + if (/* ScoreBoardDrainInProg */ + test_bit(63, &hwstatus) || + /* AbortInProg */ + test_bit(62, &hwstatus) || + /* InternalSDmaEnable */ + test_bit(61, &hwstatus) || + /* ScbEmpty */ + !test_bit(30, &hwstatus)) { + if (dd->ipath_sdma_reset_wait > 0) { + /* not done shutting down sdma */ + --dd->ipath_sdma_reset_wait; + goto resched; + } + ipath_cdbg(VERBOSE, "gave up waiting for quiescent " + "status after SDMA reset, continuing\n"); + dump_sdma_state(dd); + } + + /* dequeue all "sent" requests */ + list_for_each_entry_safe(txp, txpnext, + &dd->ipath_sdma_activelist, list) { + txp->callback_status = IPATH_SDMA_TXREQ_S_ABORTED; + if (txp->flags & IPATH_SDMA_TXREQ_F_VL15) + vl15_watchdog_deq(dd); + list_move_tail(&txp->list, &dd->ipath_sdma_notifylist); + notify = 1; + } + if (notify) + tasklet_hi_schedule(&dd->ipath_sdma_notify_task); + + /* reset our notion of head and tail */ + dd->ipath_sdma_descq_tail = 0; + dd->ipath_sdma_descq_head = 0; + dd->ipath_sdma_head_dma[0] = 0; + dd->ipath_sdma_generation = 0; + dd->ipath_sdma_descq_removed = dd->ipath_sdma_descq_added; + + /* Reset SendDmaLenGen */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmalengen, + (u64) dd->ipath_sdma_descq_cnt | (1ULL << 18)); + + /* done with sdma state for a bit */ + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + /* + * Don't restart sdma here. Wait until link is up to ACTIVE. + * VL15 MADs used to bring the link up use PIO, and multiple + * link transitions otherwise cause the sdma engine to be + * stopped and started multiple times. + * The disable is done here, including the shadow, so the + * state is kept consistent. + * See ipath_restart_sdma() for the actual starting of sdma. + */ + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl &= ~INFINIPATH_S_SDMAENABLE; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + + /* make sure I see next message */ + dd->ipath_sdma_abort_jiffies = 0; + + goto done; + } + +resched: + /* + * for now, keep spinning + * JAG - this is bad to just have default be a loop without + * state change + */ + if (jiffies > dd->ipath_sdma_abort_jiffies) { + ipath_dbg("looping with status 0x%016llx\n", + dd->ipath_sdma_status); + dd->ipath_sdma_abort_jiffies = jiffies + 5 * HZ; + } +resched_noprint: + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + if (!test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + tasklet_hi_schedule(&dd->ipath_sdma_abort_task); + return; + +unlock: + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); +done: + return; +} + +/* + * This is called from interrupt context. + */ +void ipath_sdma_intr(struct ipath_devdata *dd) +{ + unsigned long flags; + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + + (void) ipath_sdma_make_progress(dd); + + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); +} + +static int alloc_sdma(struct ipath_devdata *dd) +{ + int ret = 0; + + /* Allocate memory for SendDMA descriptor FIFO */ + dd->ipath_sdma_descq = dma_alloc_coherent(&dd->pcidev->dev, + SDMA_DESCQ_SZ, &dd->ipath_sdma_descq_phys, GFP_KERNEL); + + if (!dd->ipath_sdma_descq) { + ipath_dev_err(dd, "failed to allocate SendDMA descriptor " + "FIFO memory\n"); + ret = -ENOMEM; + goto done; + } + + dd->ipath_sdma_descq_cnt = + SDMA_DESCQ_SZ / sizeof(struct ipath_sdma_desc); + + /* Allocate memory for DMA of head register to memory */ + dd->ipath_sdma_head_dma = dma_alloc_coherent(&dd->pcidev->dev, + PAGE_SIZE, &dd->ipath_sdma_head_phys, GFP_KERNEL); + if (!dd->ipath_sdma_head_dma) { + ipath_dev_err(dd, "failed to allocate SendDMA head memory\n"); + ret = -ENOMEM; + goto cleanup_descq; + } + dd->ipath_sdma_head_dma[0] = 0; + + init_timer(&dd->ipath_sdma_vl15_timer); + dd->ipath_sdma_vl15_timer.function = vl15_watchdog_timeout; + dd->ipath_sdma_vl15_timer.data = (unsigned long)dd; + atomic_set(&dd->ipath_sdma_vl15_count, 0); + + goto done; + +cleanup_descq: + dma_free_coherent(&dd->pcidev->dev, SDMA_DESCQ_SZ, + (void *)dd->ipath_sdma_descq, dd->ipath_sdma_descq_phys); + dd->ipath_sdma_descq = NULL; + dd->ipath_sdma_descq_phys = 0; +done: + return ret; +} + +int setup_sdma(struct ipath_devdata *dd) +{ + int ret = 0; + unsigned i, n; + u64 tmp64; + u64 senddmabufmask[3] = { 0 }; + unsigned long flags; + + ret = alloc_sdma(dd); + if (ret) + goto done; + + if (!dd->ipath_sdma_descq) { + ipath_dev_err(dd, "SendDMA memory not allocated\n"); + goto done; + } + + dd->ipath_sdma_status = 0; + dd->ipath_sdma_abort_jiffies = 0; + dd->ipath_sdma_generation = 0; + dd->ipath_sdma_descq_tail = 0; + dd->ipath_sdma_descq_head = 0; + dd->ipath_sdma_descq_removed = 0; + dd->ipath_sdma_descq_added = 0; + + /* Set SendDmaBase */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabase, + dd->ipath_sdma_descq_phys); + /* Set SendDmaLenGen */ + tmp64 = dd->ipath_sdma_descq_cnt; + tmp64 |= 1<<18; /* enable generation checking */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmalengen, tmp64); + /* Set SendDmaTail */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmatail, + dd->ipath_sdma_descq_tail); + /* Set SendDmaHeadAddr */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmaheadaddr, + dd->ipath_sdma_head_phys); + + /* Reserve all the former "kernel" piobufs */ + n = dd->ipath_piobcnt2k + dd->ipath_piobcnt4k - dd->ipath_pioreserved; + for (i = dd->ipath_lastport_piobuf; i < n; ++i) { + unsigned word = i / 64; + unsigned bit = i & 63; + BUG_ON(word >= 3); + senddmabufmask[word] |= 1ULL << bit; + } + ipath_chg_pioavailkernel(dd, dd->ipath_lastport_piobuf, + n - dd->ipath_lastport_piobuf, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabufmask0, + senddmabufmask[0]); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabufmask1, + senddmabufmask[1]); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabufmask2, + senddmabufmask[2]); + + INIT_LIST_HEAD(&dd->ipath_sdma_activelist); + INIT_LIST_HEAD(&dd->ipath_sdma_notifylist); + + tasklet_init(&dd->ipath_sdma_notify_task, sdma_notify_task, + (unsigned long) dd); + tasklet_init(&dd->ipath_sdma_abort_task, sdma_abort_task, + (unsigned long) dd); + + /* + * No use to turn on SDMA here, as link is probably not ACTIVE + * Just mark it RUNNING and enable the interrupt, and let the + * ipath_restart_sdma() on link transition to ACTIVE actually + * enable it. + */ + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl |= INFINIPATH_S_SDMAINTENABLE; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + __set_bit(IPATH_SDMA_RUNNING, &dd->ipath_sdma_status); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + +done: + return ret; +} + +void teardown_sdma(struct ipath_devdata *dd) +{ + struct ipath_sdma_txreq *txp, *txpnext; + unsigned long flags; + dma_addr_t sdma_head_phys = 0; + dma_addr_t sdma_descq_phys = 0; + void *sdma_descq = NULL; + void *sdma_head_dma = NULL; + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + __clear_bit(IPATH_SDMA_RUNNING, &dd->ipath_sdma_status); + __set_bit(IPATH_SDMA_ABORTING, &dd->ipath_sdma_status); + __set_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status); + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + tasklet_kill(&dd->ipath_sdma_abort_task); + tasklet_kill(&dd->ipath_sdma_notify_task); + + /* turn off sdma */ + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + dd->ipath_sendctrl &= ~INFINIPATH_S_SDMAENABLE; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + /* dequeue all "sent" requests */ + list_for_each_entry_safe(txp, txpnext, &dd->ipath_sdma_activelist, + list) { + txp->callback_status = IPATH_SDMA_TXREQ_S_SHUTDOWN; + if (txp->flags & IPATH_SDMA_TXREQ_F_VL15) + vl15_watchdog_deq(dd); + list_move_tail(&txp->list, &dd->ipath_sdma_notifylist); + } + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + sdma_notify_taskbody(dd); + + del_timer_sync(&dd->ipath_sdma_vl15_timer); + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + + dd->ipath_sdma_abort_jiffies = 0; + + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabase, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmalengen, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmatail, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmaheadaddr, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabufmask0, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabufmask1, 0); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmabufmask2, 0); + + if (dd->ipath_sdma_head_dma) { + sdma_head_dma = (void *) dd->ipath_sdma_head_dma; + sdma_head_phys = dd->ipath_sdma_head_phys; + dd->ipath_sdma_head_dma = NULL; + dd->ipath_sdma_head_phys = 0; + } + + if (dd->ipath_sdma_descq) { + sdma_descq = dd->ipath_sdma_descq; + sdma_descq_phys = dd->ipath_sdma_descq_phys; + dd->ipath_sdma_descq = NULL; + dd->ipath_sdma_descq_phys = 0; + } + + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + if (sdma_head_dma) + dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE, + sdma_head_dma, sdma_head_phys); + + if (sdma_descq) + dma_free_coherent(&dd->pcidev->dev, SDMA_DESCQ_SZ, + sdma_descq, sdma_descq_phys); +} + +/* + * [Re]start SDMA, if we use it, and it's not already OK. + * This is called on transition to link ACTIVE, either the first or + * subsequent times. + */ +void ipath_restart_sdma(struct ipath_devdata *dd) +{ + unsigned long flags; + int needed = 1; + + if (!(dd->ipath_flags & IPATH_HAS_SEND_DMA)) + goto bail; + + /* + * First, make sure we should, which is to say, + * check that we are "RUNNING" (not in teardown) + * and not "SHUTDOWN" + */ + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + if (!test_bit(IPATH_SDMA_RUNNING, &dd->ipath_sdma_status) + || test_bit(IPATH_SDMA_SHUTDOWN, &dd->ipath_sdma_status)) + needed = 0; + else { + __clear_bit(IPATH_SDMA_DISABLED, &dd->ipath_sdma_status); + __clear_bit(IPATH_SDMA_DISARMED, &dd->ipath_sdma_status); + __clear_bit(IPATH_SDMA_ABORTING, &dd->ipath_sdma_status); + } + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + if (!needed) { + ipath_dbg("invalid attempt to restart SDMA, status 0x%016llx\n", + dd->ipath_sdma_status); + goto bail; + } + spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); + /* + * First clear, just to be safe. Enable is only done + * in chip on 0->1 transition + */ + dd->ipath_sendctrl &= ~INFINIPATH_S_SDMAENABLE; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + dd->ipath_sendctrl |= INFINIPATH_S_SDMAENABLE; + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); + ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); + spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); + +bail: + return; +} + +static inline void make_sdma_desc(struct ipath_devdata *dd, + u64 *sdmadesc, u64 addr, u64 dwlen, u64 dwoffset) +{ + WARN_ON(addr & 3); + /* SDmaPhyAddr[47:32] */ + sdmadesc[1] = addr >> 32; + /* SDmaPhyAddr[31:0] */ + sdmadesc[0] = (addr & 0xfffffffcULL) << 32; + /* SDmaGeneration[1:0] */ + sdmadesc[0] |= (dd->ipath_sdma_generation & 3ULL) << 30; + /* SDmaDwordCount[10:0] */ + sdmadesc[0] |= (dwlen & 0x7ffULL) << 16; + /* SDmaBufOffset[12:2] */ + sdmadesc[0] |= dwoffset & 0x7ffULL; +} + +/* + * This function queues one IB packet onto the send DMA queue per call. + * The caller is responsible for checking: + * 1) The number of send DMA descriptor entries is less than the size of + * the descriptor queue. + * 2) The IB SGE addresses and lengths are 32-bit aligned + * (except possibly the last SGE's length) + * 3) The SGE addresses are suitable for passing to dma_map_single(). + */ +int ipath_sdma_verbs_send(struct ipath_devdata *dd, + struct ipath_sge_state *ss, u32 dwords, + struct ipath_verbs_txreq *tx) +{ + + unsigned long flags; + struct ipath_sge *sge; + int ret = 0; + u16 tail; + __le64 *descqp; + u64 sdmadesc[2]; + u32 dwoffset; + dma_addr_t addr; + + if ((tx->map_len + (dwords<<2)) > dd->ipath_ibmaxlen) { + ipath_dbg("packet size %X > ibmax %X, fail\n", + tx->map_len + (dwords<<2), dd->ipath_ibmaxlen); + ret = -EMSGSIZE; + goto fail; + } + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + +retry: + if (unlikely(test_bit(IPATH_SDMA_ABORTING, &dd->ipath_sdma_status))) { + ret = -EBUSY; + goto unlock; + } + + if (tx->txreq.sg_count > ipath_sdma_descq_freecnt(dd)) { + if (ipath_sdma_make_progress(dd)) + goto retry; + ret = -ENOBUFS; + goto unlock; + } + + addr = dma_map_single(&dd->pcidev->dev, tx->txreq.map_addr, + tx->map_len, DMA_TO_DEVICE); + if (dma_mapping_error(addr)) { + ret = -EIO; + goto unlock; + } + + dwoffset = tx->map_len >> 2; + make_sdma_desc(dd, sdmadesc, (u64) addr, dwoffset, 0); + + /* SDmaFirstDesc */ + sdmadesc[0] |= 1ULL << 12; + if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_USELARGEBUF) + sdmadesc[0] |= 1ULL << 14; /* SDmaUseLargeBuf */ + + /* write to the descq */ + tail = dd->ipath_sdma_descq_tail; + descqp = &dd->ipath_sdma_descq[tail].qw[0]; + *descqp++ = cpu_to_le64(sdmadesc[0]); + *descqp++ = cpu_to_le64(sdmadesc[1]); + + if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_FREEDESC) + tx->txreq.start_idx = tail; + + /* increment the tail */ + if (++tail == dd->ipath_sdma_descq_cnt) { + tail = 0; + descqp = &dd->ipath_sdma_descq[0].qw[0]; + ++dd->ipath_sdma_generation; + } + + sge = &ss->sge; + while (dwords) { + u32 dw; + u32 len; + + len = dwords << 2; + if (len > sge->length) + len = sge->length; + if (len > sge->sge_length) + len = sge->sge_length; + BUG_ON(len == 0); + dw = (len + 3) >> 2; + addr = dma_map_single(&dd->pcidev->dev, sge->vaddr, dw << 2, + DMA_TO_DEVICE); + make_sdma_desc(dd, sdmadesc, (u64) addr, dw, dwoffset); + /* SDmaUseLargeBuf has to be set in every descriptor */ + if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_USELARGEBUF) + sdmadesc[0] |= 1ULL << 14; + /* write to the descq */ + *descqp++ = cpu_to_le64(sdmadesc[0]); + *descqp++ = cpu_to_le64(sdmadesc[1]); + + /* increment the tail */ + if (++tail == dd->ipath_sdma_descq_cnt) { + tail = 0; + descqp = &dd->ipath_sdma_descq[0].qw[0]; + ++dd->ipath_sdma_generation; + } + sge->vaddr += len; + sge->length -= len; + sge->sge_length -= len; + if (sge->sge_length == 0) { + if (--ss->num_sge) + *sge = *ss->sg_list++; + } else if (sge->length == 0 && sge->mr != NULL) { + if (++sge->n >= IPATH_SEGSZ) { + if (++sge->m >= sge->mr->mapsz) + break; + sge->n = 0; + } + sge->vaddr = + sge->mr->map[sge->m]->segs[sge->n].vaddr; + sge->length = + sge->mr->map[sge->m]->segs[sge->n].length; + } + + dwoffset += dw; + dwords -= dw; + } + + if (!tail) + descqp = &dd->ipath_sdma_descq[dd->ipath_sdma_descq_cnt].qw[0]; + descqp -= 2; + /* SDmaLastDesc */ + descqp[0] |= __constant_cpu_to_le64(1ULL << 11); + if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_INTREQ) { + /* SDmaIntReq */ + descqp[0] |= __constant_cpu_to_le64(1ULL << 15); + } + + /* Commit writes to memory and advance the tail on the chip */ + wmb(); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmatail, tail); + + tx->txreq.next_descq_idx = tail; + tx->txreq.callback_status = IPATH_SDMA_TXREQ_S_OK; + dd->ipath_sdma_descq_tail = tail; + dd->ipath_sdma_descq_added += tx->txreq.sg_count; + list_add_tail(&tx->txreq.list, &dd->ipath_sdma_activelist); + if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_VL15) + vl15_watchdog_enq(dd); + +unlock: + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); +fail: + return ret; +} diff --git a/drivers/infiniband/hw/ipath/ipath_srq.c b/drivers/infiniband/hw/ipath/ipath_srq.c index f772102e471..e3d80ca84c1 100644 --- a/drivers/infiniband/hw/ipath/ipath_srq.c +++ b/drivers/infiniband/hw/ipath/ipath_srq.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -245,7 +245,8 @@ int ipath_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, sizeof(offset_addr)); if (ret) goto bail_free; - udata->outbuf = (void __user *) offset_addr; + udata->outbuf = + (void __user *) (unsigned long) offset_addr; ret = ib_copy_to_udata(udata, &offset, sizeof(offset)); if (ret) diff --git a/drivers/infiniband/hw/ipath/ipath_stats.c b/drivers/infiniband/hw/ipath/ipath_stats.c index d2725cd11bd..c8e3d65f0de 100644 --- a/drivers/infiniband/hw/ipath/ipath_stats.c +++ b/drivers/infiniband/hw/ipath/ipath_stats.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -136,6 +136,7 @@ static void ipath_qcheck(struct ipath_devdata *dd) struct ipath_portdata *pd = dd->ipath_pd[0]; size_t blen = 0; char buf[128]; + u32 hdrqtail; *buf = 0; if (pd->port_hdrqfull != dd->ipath_p0_hdrqfull) { @@ -174,17 +175,18 @@ static void ipath_qcheck(struct ipath_devdata *dd) if (blen) ipath_dbg("%s\n", buf); - if (pd->port_head != (u32) - le64_to_cpu(*dd->ipath_hdrqtailptr)) { + hdrqtail = ipath_get_hdrqtail(pd); + if (pd->port_head != hdrqtail) { if (dd->ipath_lastport0rcv_cnt == ipath_stats.sps_port0pkts) { ipath_cdbg(PKT, "missing rcv interrupts? " - "port0 hd=%llx tl=%x; port0pkts %llx\n", - (unsigned long long) - le64_to_cpu(*dd->ipath_hdrqtailptr), - pd->port_head, + "port0 hd=%x tl=%x; port0pkts %llx; write" + " hd (w/intr)\n", + pd->port_head, hdrqtail, (unsigned long long) ipath_stats.sps_port0pkts); + ipath_write_ureg(dd, ur_rcvhdrhead, hdrqtail | + dd->ipath_rhdrhead_intr_off, pd->port_port); } dd->ipath_lastport0rcv_cnt = ipath_stats.sps_port0pkts; } @@ -290,11 +292,11 @@ void ipath_get_faststats(unsigned long opaque) && time_after(jiffies, dd->ipath_unmasktime)) { char ebuf[256]; int iserr; - iserr = ipath_decode_err(ebuf, sizeof ebuf, - dd->ipath_maskederrs); + iserr = ipath_decode_err(dd, ebuf, sizeof ebuf, + dd->ipath_maskederrs); if (dd->ipath_maskederrs & - ~(INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL | - INFINIPATH_E_PKTERRS )) + ~(INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL | + INFINIPATH_E_PKTERRS)) ipath_dev_err(dd, "Re-enabling masked errors " "(%s)\n", ebuf); else { @@ -306,17 +308,18 @@ void ipath_get_faststats(unsigned long opaque) * level. */ if (iserr) - ipath_dbg("Re-enabling queue full errors (%s)\n", - ebuf); + ipath_dbg( + "Re-enabling queue full errors (%s)\n", + ebuf); else ipath_cdbg(ERRPKT, "Re-enabling packet" - " problem interrupt (%s)\n", ebuf); + " problem interrupt (%s)\n", ebuf); } /* re-enable masked errors */ dd->ipath_errormask |= dd->ipath_maskederrs; ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, - dd->ipath_errormask); + dd->ipath_errormask); dd->ipath_maskederrs = 0; } diff --git a/drivers/infiniband/hw/ipath/ipath_sysfs.c b/drivers/infiniband/hw/ipath/ipath_sysfs.c index 56dfc8a2344..a6c8efbdc0c 100644 --- a/drivers/infiniband/hw/ipath/ipath_sysfs.c +++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -34,6 +34,7 @@ #include <linux/ctype.h> #include "ipath_kernel.h" +#include "ipath_verbs.h" #include "ipath_common.h" /** @@ -163,6 +164,15 @@ static ssize_t show_boardversion(struct device *dev, return scnprintf(buf, PAGE_SIZE, "%s", dd->ipath_boardversion); } +static ssize_t show_localbus_info(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct ipath_devdata *dd = dev_get_drvdata(dev); + /* The string printed here is already newline-terminated. */ + return scnprintf(buf, PAGE_SIZE, "%s", dd->ipath_lbus_info); +} + static ssize_t show_lmc(struct device *dev, struct device_attribute *attr, char *buf) @@ -311,6 +321,8 @@ static ssize_t store_guid(struct device *dev, dd->ipath_guid = new_guid; dd->ipath_nguid = 1; + if (dd->verbs_dev) + dd->verbs_dev->ibdev.node_guid = new_guid; ret = strlen(buf); goto bail; @@ -919,21 +931,21 @@ static ssize_t store_rx_polinv_enb(struct device *dev, u16 val; ret = ipath_parse_ushort(buf, &val); - if (ret < 0 || val > 1) - goto invalid; + if (ret >= 0 && val > 1) { + ipath_dev_err(dd, + "attempt to set invalid Rx Polarity (enable)\n"); + ret = -EINVAL; + goto bail; + } r = dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_RXPOL_ENB, val); - if (r < 0) { + if (r < 0) ret = r; - goto bail; - } - goto bail; -invalid: - ipath_dev_err(dd, "attempt to set invalid Rx Polarity (enable)\n"); bail: return ret; } + /* * Get/Set RX lane-reversal enable. 0=no, 1=yes. */ @@ -988,6 +1000,75 @@ static struct attribute_group driver_attr_group = { .attrs = driver_attributes }; +static ssize_t store_tempsense(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct ipath_devdata *dd = dev_get_drvdata(dev); + int ret, stat; + u16 val; + + ret = ipath_parse_ushort(buf, &val); + if (ret <= 0) { + ipath_dev_err(dd, "attempt to set invalid tempsense config\n"); + goto bail; + } + /* If anything but the highest limit, enable T_CRIT_A "interrupt" */ + stat = ipath_tempsense_write(dd, 9, (val == 0x7f7f) ? 0x80 : 0); + if (stat) { + ipath_dev_err(dd, "Unable to set tempsense config\n"); + ret = -1; + goto bail; + } + stat = ipath_tempsense_write(dd, 0xB, (u8) (val & 0xFF)); + if (stat) { + ipath_dev_err(dd, "Unable to set local Tcrit\n"); + ret = -1; + goto bail; + } + stat = ipath_tempsense_write(dd, 0xD, (u8) (val >> 8)); + if (stat) { + ipath_dev_err(dd, "Unable to set remote Tcrit\n"); + ret = -1; + goto bail; + } + +bail: + return ret; +} + +/* + * dump tempsense regs. in decimal, to ease shell-scripts. + */ +static ssize_t show_tempsense(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct ipath_devdata *dd = dev_get_drvdata(dev); + int ret; + int idx; + u8 regvals[8]; + + ret = -ENXIO; + for (idx = 0; idx < 8; ++idx) { + if (idx == 6) + continue; + ret = ipath_tempsense_read(dd, idx); + if (ret < 0) + break; + regvals[idx] = ret; + } + if (idx == 8) + ret = scnprintf(buf, PAGE_SIZE, "%d %d %02X %02X %d %d\n", + *(signed char *)(regvals), + *(signed char *)(regvals + 1), + regvals[2], regvals[3], + *(signed char *)(regvals + 5), + *(signed char *)(regvals + 7)); + return ret; +} + struct attribute_group *ipath_driver_attr_groups[] = { &driver_attr_group, NULL, @@ -1011,10 +1092,13 @@ static DEVICE_ATTR(unit, S_IRUGO, show_unit, NULL); static DEVICE_ATTR(rx_pol_inv, S_IWUSR, NULL, store_rx_pol_inv); static DEVICE_ATTR(led_override, S_IWUSR, NULL, store_led_override); static DEVICE_ATTR(logged_errors, S_IRUGO, show_logged_errs, NULL); +static DEVICE_ATTR(localbus_info, S_IRUGO, show_localbus_info, NULL); static DEVICE_ATTR(jint_max_packets, S_IWUSR | S_IRUGO, show_jint_max_packets, store_jint_max_packets); static DEVICE_ATTR(jint_idle_ticks, S_IWUSR | S_IRUGO, show_jint_idle_ticks, store_jint_idle_ticks); +static DEVICE_ATTR(tempsense, S_IWUSR | S_IRUGO, + show_tempsense, store_tempsense); static struct attribute *dev_attributes[] = { &dev_attr_guid.attr, @@ -1034,6 +1118,8 @@ static struct attribute *dev_attributes[] = { &dev_attr_rx_pol_inv.attr, &dev_attr_led_override.attr, &dev_attr_logged_errors.attr, + &dev_attr_tempsense.attr, + &dev_attr_localbus_info.attr, NULL }; diff --git a/drivers/infiniband/hw/ipath/ipath_uc.c b/drivers/infiniband/hw/ipath/ipath_uc.c index 2dd8de20d22..bfe8926b551 100644 --- a/drivers/infiniband/hw/ipath/ipath_uc.c +++ b/drivers/infiniband/hw/ipath/ipath_uc.c @@ -94,7 +94,7 @@ int ipath_make_uc_req(struct ipath_qp *qp) qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ - ohdr->u.imm_data = wqe->wr.imm_data; + ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; } if (wqe->wr.send_flags & IB_SEND_SOLICITED) @@ -123,7 +123,7 @@ int ipath_make_uc_req(struct ipath_qp *qp) qp->s_state = OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE); /* Immediate data comes after the RETH */ - ohdr->u.rc.imm_data = wqe->wr.imm_data; + ohdr->u.rc.imm_data = wqe->wr.ex.imm_data; hwords += 1; if (wqe->wr.send_flags & IB_SEND_SOLICITED) bth0 |= 1 << 23; @@ -152,7 +152,7 @@ int ipath_make_uc_req(struct ipath_qp *qp) else { qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ - ohdr->u.imm_data = wqe->wr.imm_data; + ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; } if (wqe->wr.send_flags & IB_SEND_SOLICITED) @@ -177,7 +177,7 @@ int ipath_make_uc_req(struct ipath_qp *qp) qp->s_state = OP(RDMA_WRITE_LAST_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ - ohdr->u.imm_data = wqe->wr.imm_data; + ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; if (wqe->wr.send_flags & IB_SEND_SOLICITED) bth0 |= 1 << 23; diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c index de67eed08ed..8b6a261c89e 100644 --- a/drivers/infiniband/hw/ipath/ipath_ud.c +++ b/drivers/infiniband/hw/ipath/ipath_ud.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -95,7 +95,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe) if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) { wc.wc_flags = IB_WC_WITH_IMM; - wc.imm_data = swqe->wr.imm_data; + wc.imm_data = swqe->wr.ex.imm_data; } else { wc.wc_flags = 0; wc.imm_data = 0; @@ -303,6 +303,7 @@ int ipath_make_ud_req(struct ipath_qp *qp) qp->s_hdrwords = 7; qp->s_cur_size = wqe->length; qp->s_cur_sge = &qp->s_sge; + qp->s_dmult = ah_attr->static_rate; qp->s_wqe = wqe; qp->s_sge.sge = wqe->sg_list[0]; qp->s_sge.sg_list = wqe->sg_list + 1; @@ -326,7 +327,7 @@ int ipath_make_ud_req(struct ipath_qp *qp) } if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) { qp->s_hdrwords++; - ohdr->u.ud.imm_data = wqe->wr.imm_data; + ohdr->u.ud.imm_data = wqe->wr.ex.imm_data; bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24; } else bth0 = IB_OPCODE_UD_SEND_ONLY << 24; diff --git a/drivers/infiniband/hw/ipath/ipath_user_sdma.c b/drivers/infiniband/hw/ipath/ipath_user_sdma.c new file mode 100644 index 00000000000..86e016916cd --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_user_sdma.c @@ -0,0 +1,879 @@ +/* + * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/mm.h> +#include <linux/types.h> +#include <linux/device.h> +#include <linux/dmapool.h> +#include <linux/slab.h> +#include <linux/list.h> +#include <linux/highmem.h> +#include <linux/io.h> +#include <linux/uio.h> +#include <linux/rbtree.h> +#include <linux/spinlock.h> +#include <linux/delay.h> + +#include "ipath_kernel.h" +#include "ipath_user_sdma.h" + +/* minimum size of header */ +#define IPATH_USER_SDMA_MIN_HEADER_LENGTH 64 +/* expected size of headers (for dma_pool) */ +#define IPATH_USER_SDMA_EXP_HEADER_LENGTH 64 +/* length mask in PBC (lower 11 bits) */ +#define IPATH_PBC_LENGTH_MASK ((1 << 11) - 1) + +struct ipath_user_sdma_pkt { + u8 naddr; /* dimension of addr (1..3) ... */ + u32 counter; /* sdma pkts queued counter for this entry */ + u64 added; /* global descq number of entries */ + + struct { + u32 offset; /* offset for kvaddr, addr */ + u32 length; /* length in page */ + u8 put_page; /* should we put_page? */ + u8 dma_mapped; /* is page dma_mapped? */ + struct page *page; /* may be NULL (coherent mem) */ + void *kvaddr; /* FIXME: only for pio hack */ + dma_addr_t addr; + } addr[4]; /* max pages, any more and we coalesce */ + struct list_head list; /* list element */ +}; + +struct ipath_user_sdma_queue { + /* + * pkts sent to dma engine are queued on this + * list head. the type of the elements of this + * list are struct ipath_user_sdma_pkt... + */ + struct list_head sent; + + /* headers with expected length are allocated from here... */ + char header_cache_name[64]; + struct dma_pool *header_cache; + + /* packets are allocated from the slab cache... */ + char pkt_slab_name[64]; + struct kmem_cache *pkt_slab; + + /* as packets go on the queued queue, they are counted... */ + u32 counter; + u32 sent_counter; + + /* dma page table */ + struct rb_root dma_pages_root; + + /* protect everything above... */ + struct mutex lock; +}; + +struct ipath_user_sdma_queue * +ipath_user_sdma_queue_create(struct device *dev, int unit, int port, int sport) +{ + struct ipath_user_sdma_queue *pq = + kmalloc(sizeof(struct ipath_user_sdma_queue), GFP_KERNEL); + + if (!pq) + goto done; + + pq->counter = 0; + pq->sent_counter = 0; + INIT_LIST_HEAD(&pq->sent); + + mutex_init(&pq->lock); + + snprintf(pq->pkt_slab_name, sizeof(pq->pkt_slab_name), + "ipath-user-sdma-pkts-%u-%02u.%02u", unit, port, sport); + pq->pkt_slab = kmem_cache_create(pq->pkt_slab_name, + sizeof(struct ipath_user_sdma_pkt), + 0, 0, NULL); + + if (!pq->pkt_slab) + goto err_kfree; + + snprintf(pq->header_cache_name, sizeof(pq->header_cache_name), + "ipath-user-sdma-headers-%u-%02u.%02u", unit, port, sport); + pq->header_cache = dma_pool_create(pq->header_cache_name, + dev, + IPATH_USER_SDMA_EXP_HEADER_LENGTH, + 4, 0); + if (!pq->header_cache) + goto err_slab; + + pq->dma_pages_root = RB_ROOT; + + goto done; + +err_slab: + kmem_cache_destroy(pq->pkt_slab); +err_kfree: + kfree(pq); + pq = NULL; + +done: + return pq; +} + +static void ipath_user_sdma_init_frag(struct ipath_user_sdma_pkt *pkt, + int i, size_t offset, size_t len, + int put_page, int dma_mapped, + struct page *page, + void *kvaddr, dma_addr_t dma_addr) +{ + pkt->addr[i].offset = offset; + pkt->addr[i].length = len; + pkt->addr[i].put_page = put_page; + pkt->addr[i].dma_mapped = dma_mapped; + pkt->addr[i].page = page; + pkt->addr[i].kvaddr = kvaddr; + pkt->addr[i].addr = dma_addr; +} + +static void ipath_user_sdma_init_header(struct ipath_user_sdma_pkt *pkt, + u32 counter, size_t offset, + size_t len, int dma_mapped, + struct page *page, + void *kvaddr, dma_addr_t dma_addr) +{ + pkt->naddr = 1; + pkt->counter = counter; + ipath_user_sdma_init_frag(pkt, 0, offset, len, 0, dma_mapped, page, + kvaddr, dma_addr); +} + +/* we've too many pages in the iovec, coalesce to a single page */ +static int ipath_user_sdma_coalesce(const struct ipath_devdata *dd, + struct ipath_user_sdma_pkt *pkt, + const struct iovec *iov, + unsigned long niov) { + int ret = 0; + struct page *page = alloc_page(GFP_KERNEL); + void *mpage_save; + char *mpage; + int i; + int len = 0; + dma_addr_t dma_addr; + + if (!page) { + ret = -ENOMEM; + goto done; + } + + mpage = kmap(page); + mpage_save = mpage; + for (i = 0; i < niov; i++) { + int cfur; + + cfur = copy_from_user(mpage, + iov[i].iov_base, iov[i].iov_len); + if (cfur) { + ret = -EFAULT; + goto free_unmap; + } + + mpage += iov[i].iov_len; + len += iov[i].iov_len; + } + + dma_addr = dma_map_page(&dd->pcidev->dev, page, 0, len, + DMA_TO_DEVICE); + if (dma_mapping_error(dma_addr)) { + ret = -ENOMEM; + goto free_unmap; + } + + ipath_user_sdma_init_frag(pkt, 1, 0, len, 0, 1, page, mpage_save, + dma_addr); + pkt->naddr = 2; + + goto done; + +free_unmap: + kunmap(page); + __free_page(page); +done: + return ret; +} + +/* how many pages in this iovec element? */ +static int ipath_user_sdma_num_pages(const struct iovec *iov) +{ + const unsigned long addr = (unsigned long) iov->iov_base; + const unsigned long len = iov->iov_len; + const unsigned long spage = addr & PAGE_MASK; + const unsigned long epage = (addr + len - 1) & PAGE_MASK; + + return 1 + ((epage - spage) >> PAGE_SHIFT); +} + +/* truncate length to page boundry */ +static int ipath_user_sdma_page_length(unsigned long addr, unsigned long len) +{ + const unsigned long offset = addr & ~PAGE_MASK; + + return ((offset + len) > PAGE_SIZE) ? (PAGE_SIZE - offset) : len; +} + +static void ipath_user_sdma_free_pkt_frag(struct device *dev, + struct ipath_user_sdma_queue *pq, + struct ipath_user_sdma_pkt *pkt, + int frag) +{ + const int i = frag; + + if (pkt->addr[i].page) { + if (pkt->addr[i].dma_mapped) + dma_unmap_page(dev, + pkt->addr[i].addr, + pkt->addr[i].length, + DMA_TO_DEVICE); + + if (pkt->addr[i].kvaddr) + kunmap(pkt->addr[i].page); + + if (pkt->addr[i].put_page) + put_page(pkt->addr[i].page); + else + __free_page(pkt->addr[i].page); + } else if (pkt->addr[i].kvaddr) + /* free coherent mem from cache... */ + dma_pool_free(pq->header_cache, + pkt->addr[i].kvaddr, pkt->addr[i].addr); +} + +/* return number of pages pinned... */ +static int ipath_user_sdma_pin_pages(const struct ipath_devdata *dd, + struct ipath_user_sdma_pkt *pkt, + unsigned long addr, int tlen, int npages) +{ + struct page *pages[2]; + int j; + int ret; + + ret = get_user_pages(current, current->mm, addr, + npages, 0, 1, pages, NULL); + + if (ret != npages) { + int i; + + for (i = 0; i < ret; i++) + put_page(pages[i]); + + ret = -ENOMEM; + goto done; + } + + for (j = 0; j < npages; j++) { + /* map the pages... */ + const int flen = + ipath_user_sdma_page_length(addr, tlen); + dma_addr_t dma_addr = + dma_map_page(&dd->pcidev->dev, + pages[j], 0, flen, DMA_TO_DEVICE); + unsigned long fofs = addr & ~PAGE_MASK; + + if (dma_mapping_error(dma_addr)) { + ret = -ENOMEM; + goto done; + } + + ipath_user_sdma_init_frag(pkt, pkt->naddr, fofs, flen, 1, 1, + pages[j], kmap(pages[j]), + dma_addr); + + pkt->naddr++; + addr += flen; + tlen -= flen; + } + +done: + return ret; +} + +static int ipath_user_sdma_pin_pkt(const struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + struct ipath_user_sdma_pkt *pkt, + const struct iovec *iov, + unsigned long niov) +{ + int ret = 0; + unsigned long idx; + + for (idx = 0; idx < niov; idx++) { + const int npages = ipath_user_sdma_num_pages(iov + idx); + const unsigned long addr = (unsigned long) iov[idx].iov_base; + + ret = ipath_user_sdma_pin_pages(dd, pkt, + addr, iov[idx].iov_len, + npages); + if (ret < 0) + goto free_pkt; + } + + goto done; + +free_pkt: + for (idx = 0; idx < pkt->naddr; idx++) + ipath_user_sdma_free_pkt_frag(&dd->pcidev->dev, pq, pkt, idx); + +done: + return ret; +} + +static int ipath_user_sdma_init_payload(const struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + struct ipath_user_sdma_pkt *pkt, + const struct iovec *iov, + unsigned long niov, int npages) +{ + int ret = 0; + + if (npages >= ARRAY_SIZE(pkt->addr)) + ret = ipath_user_sdma_coalesce(dd, pkt, iov, niov); + else + ret = ipath_user_sdma_pin_pkt(dd, pq, pkt, iov, niov); + + return ret; +} + +/* free a packet list -- return counter value of last packet */ +static void ipath_user_sdma_free_pkt_list(struct device *dev, + struct ipath_user_sdma_queue *pq, + struct list_head *list) +{ + struct ipath_user_sdma_pkt *pkt, *pkt_next; + + list_for_each_entry_safe(pkt, pkt_next, list, list) { + int i; + + for (i = 0; i < pkt->naddr; i++) + ipath_user_sdma_free_pkt_frag(dev, pq, pkt, i); + + kmem_cache_free(pq->pkt_slab, pkt); + } +} + +/* + * copy headers, coalesce etc -- pq->lock must be held + * + * we queue all the packets to list, returning the + * number of bytes total. list must be empty initially, + * as, if there is an error we clean it... + */ +static int ipath_user_sdma_queue_pkts(const struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + struct list_head *list, + const struct iovec *iov, + unsigned long niov, + int maxpkts) +{ + unsigned long idx = 0; + int ret = 0; + int npkts = 0; + struct page *page = NULL; + __le32 *pbc; + dma_addr_t dma_addr; + struct ipath_user_sdma_pkt *pkt = NULL; + size_t len; + size_t nw; + u32 counter = pq->counter; + int dma_mapped = 0; + + while (idx < niov && npkts < maxpkts) { + const unsigned long addr = (unsigned long) iov[idx].iov_base; + const unsigned long idx_save = idx; + unsigned pktnw; + unsigned pktnwc; + int nfrags = 0; + int npages = 0; + int cfur; + + dma_mapped = 0; + len = iov[idx].iov_len; + nw = len >> 2; + page = NULL; + + pkt = kmem_cache_alloc(pq->pkt_slab, GFP_KERNEL); + if (!pkt) { + ret = -ENOMEM; + goto free_list; + } + + if (len < IPATH_USER_SDMA_MIN_HEADER_LENGTH || + len > PAGE_SIZE || len & 3 || addr & 3) { + ret = -EINVAL; + goto free_pkt; + } + + if (len == IPATH_USER_SDMA_EXP_HEADER_LENGTH) + pbc = dma_pool_alloc(pq->header_cache, GFP_KERNEL, + &dma_addr); + else + pbc = NULL; + + if (!pbc) { + page = alloc_page(GFP_KERNEL); + if (!page) { + ret = -ENOMEM; + goto free_pkt; + } + pbc = kmap(page); + } + + cfur = copy_from_user(pbc, iov[idx].iov_base, len); + if (cfur) { + ret = -EFAULT; + goto free_pbc; + } + + /* + * this assignment is a bit strange. it's because the + * the pbc counts the number of 32 bit words in the full + * packet _except_ the first word of the pbc itself... + */ + pktnwc = nw - 1; + + /* + * pktnw computation yields the number of 32 bit words + * that the caller has indicated in the PBC. note that + * this is one less than the total number of words that + * goes to the send DMA engine as the first 32 bit word + * of the PBC itself is not counted. Armed with this count, + * we can verify that the packet is consistent with the + * iovec lengths. + */ + pktnw = le32_to_cpu(*pbc) & IPATH_PBC_LENGTH_MASK; + if (pktnw < pktnwc || pktnw > pktnwc + (PAGE_SIZE >> 2)) { + ret = -EINVAL; + goto free_pbc; + } + + + idx++; + while (pktnwc < pktnw && idx < niov) { + const size_t slen = iov[idx].iov_len; + const unsigned long faddr = + (unsigned long) iov[idx].iov_base; + + if (slen & 3 || faddr & 3 || !slen || + slen > PAGE_SIZE) { + ret = -EINVAL; + goto free_pbc; + } + + npages++; + if ((faddr & PAGE_MASK) != + ((faddr + slen - 1) & PAGE_MASK)) + npages++; + + pktnwc += slen >> 2; + idx++; + nfrags++; + } + + if (pktnwc != pktnw) { + ret = -EINVAL; + goto free_pbc; + } + + if (page) { + dma_addr = dma_map_page(&dd->pcidev->dev, + page, 0, len, DMA_TO_DEVICE); + if (dma_mapping_error(dma_addr)) { + ret = -ENOMEM; + goto free_pbc; + } + + dma_mapped = 1; + } + + ipath_user_sdma_init_header(pkt, counter, 0, len, dma_mapped, + page, pbc, dma_addr); + + if (nfrags) { + ret = ipath_user_sdma_init_payload(dd, pq, pkt, + iov + idx_save + 1, + nfrags, npages); + if (ret < 0) + goto free_pbc_dma; + } + + counter++; + npkts++; + + list_add_tail(&pkt->list, list); + } + + ret = idx; + goto done; + +free_pbc_dma: + if (dma_mapped) + dma_unmap_page(&dd->pcidev->dev, dma_addr, len, DMA_TO_DEVICE); +free_pbc: + if (page) { + kunmap(page); + __free_page(page); + } else + dma_pool_free(pq->header_cache, pbc, dma_addr); +free_pkt: + kmem_cache_free(pq->pkt_slab, pkt); +free_list: + ipath_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, list); +done: + return ret; +} + +static void ipath_user_sdma_set_complete_counter(struct ipath_user_sdma_queue *pq, + u32 c) +{ + pq->sent_counter = c; +} + +/* try to clean out queue -- needs pq->lock */ +static int ipath_user_sdma_queue_clean(const struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq) +{ + struct list_head free_list; + struct ipath_user_sdma_pkt *pkt; + struct ipath_user_sdma_pkt *pkt_prev; + int ret = 0; + + INIT_LIST_HEAD(&free_list); + + list_for_each_entry_safe(pkt, pkt_prev, &pq->sent, list) { + s64 descd = dd->ipath_sdma_descq_removed - pkt->added; + + if (descd < 0) + break; + + list_move_tail(&pkt->list, &free_list); + + /* one more packet cleaned */ + ret++; + } + + if (!list_empty(&free_list)) { + u32 counter; + + pkt = list_entry(free_list.prev, + struct ipath_user_sdma_pkt, list); + counter = pkt->counter; + + ipath_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list); + ipath_user_sdma_set_complete_counter(pq, counter); + } + + return ret; +} + +void ipath_user_sdma_queue_destroy(struct ipath_user_sdma_queue *pq) +{ + if (!pq) + return; + + kmem_cache_destroy(pq->pkt_slab); + dma_pool_destroy(pq->header_cache); + kfree(pq); +} + +/* clean descriptor queue, returns > 0 if some elements cleaned */ +static int ipath_user_sdma_hwqueue_clean(struct ipath_devdata *dd) +{ + int ret; + unsigned long flags; + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + ret = ipath_sdma_make_progress(dd); + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + return ret; +} + +/* we're in close, drain packets so that we can cleanup successfully... */ +void ipath_user_sdma_queue_drain(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq) +{ + int i; + + if (!pq) + return; + + for (i = 0; i < 100; i++) { + mutex_lock(&pq->lock); + if (list_empty(&pq->sent)) { + mutex_unlock(&pq->lock); + break; + } + ipath_user_sdma_hwqueue_clean(dd); + ipath_user_sdma_queue_clean(dd, pq); + mutex_unlock(&pq->lock); + msleep(10); + } + + if (!list_empty(&pq->sent)) { + struct list_head free_list; + + printk(KERN_INFO "drain: lists not empty: forcing!\n"); + INIT_LIST_HEAD(&free_list); + mutex_lock(&pq->lock); + list_splice_init(&pq->sent, &free_list); + ipath_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list); + mutex_unlock(&pq->lock); + } +} + +static inline __le64 ipath_sdma_make_desc0(struct ipath_devdata *dd, + u64 addr, u64 dwlen, u64 dwoffset) +{ + return cpu_to_le64(/* SDmaPhyAddr[31:0] */ + ((addr & 0xfffffffcULL) << 32) | + /* SDmaGeneration[1:0] */ + ((dd->ipath_sdma_generation & 3ULL) << 30) | + /* SDmaDwordCount[10:0] */ + ((dwlen & 0x7ffULL) << 16) | + /* SDmaBufOffset[12:2] */ + (dwoffset & 0x7ffULL)); +} + +static inline __le64 ipath_sdma_make_first_desc0(__le64 descq) +{ + return descq | __constant_cpu_to_le64(1ULL << 12); +} + +static inline __le64 ipath_sdma_make_last_desc0(__le64 descq) +{ + /* last */ /* dma head */ + return descq | __constant_cpu_to_le64(1ULL << 11 | 1ULL << 13); +} + +static inline __le64 ipath_sdma_make_desc1(u64 addr) +{ + /* SDmaPhyAddr[47:32] */ + return cpu_to_le64(addr >> 32); +} + +static void ipath_user_sdma_send_frag(struct ipath_devdata *dd, + struct ipath_user_sdma_pkt *pkt, int idx, + unsigned ofs, u16 tail) +{ + const u64 addr = (u64) pkt->addr[idx].addr + + (u64) pkt->addr[idx].offset; + const u64 dwlen = (u64) pkt->addr[idx].length / 4; + __le64 *descqp; + __le64 descq0; + + descqp = &dd->ipath_sdma_descq[tail].qw[0]; + + descq0 = ipath_sdma_make_desc0(dd, addr, dwlen, ofs); + if (idx == 0) + descq0 = ipath_sdma_make_first_desc0(descq0); + if (idx == pkt->naddr - 1) + descq0 = ipath_sdma_make_last_desc0(descq0); + + descqp[0] = descq0; + descqp[1] = ipath_sdma_make_desc1(addr); +} + +/* pq->lock must be held, get packets on the wire... */ +static int ipath_user_sdma_push_pkts(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + struct list_head *pktlist) +{ + int ret = 0; + unsigned long flags; + u16 tail; + + if (list_empty(pktlist)) + return 0; + + if (unlikely(!(dd->ipath_flags & IPATH_LINKACTIVE))) + return -ECOMM; + + spin_lock_irqsave(&dd->ipath_sdma_lock, flags); + + if (unlikely(dd->ipath_sdma_status & IPATH_SDMA_ABORT_MASK)) { + ret = -ECOMM; + goto unlock; + } + + tail = dd->ipath_sdma_descq_tail; + while (!list_empty(pktlist)) { + struct ipath_user_sdma_pkt *pkt = + list_entry(pktlist->next, struct ipath_user_sdma_pkt, + list); + int i; + unsigned ofs = 0; + u16 dtail = tail; + + if (pkt->naddr > ipath_sdma_descq_freecnt(dd)) + goto unlock_check_tail; + + for (i = 0; i < pkt->naddr; i++) { + ipath_user_sdma_send_frag(dd, pkt, i, ofs, tail); + ofs += pkt->addr[i].length >> 2; + + if (++tail == dd->ipath_sdma_descq_cnt) { + tail = 0; + ++dd->ipath_sdma_generation; + } + } + + if ((ofs<<2) > dd->ipath_ibmaxlen) { + ipath_dbg("packet size %X > ibmax %X, fail\n", + ofs<<2, dd->ipath_ibmaxlen); + ret = -EMSGSIZE; + goto unlock; + } + + /* + * if the packet is >= 2KB mtu equivalent, we have to use + * the large buffers, and have to mark each descriptor as + * part of a large buffer packet. + */ + if (ofs >= IPATH_SMALLBUF_DWORDS) { + for (i = 0; i < pkt->naddr; i++) { + dd->ipath_sdma_descq[dtail].qw[0] |= + __constant_cpu_to_le64(1ULL << 14); + if (++dtail == dd->ipath_sdma_descq_cnt) + dtail = 0; + } + } + + dd->ipath_sdma_descq_added += pkt->naddr; + pkt->added = dd->ipath_sdma_descq_added; + list_move_tail(&pkt->list, &pq->sent); + ret++; + } + +unlock_check_tail: + /* advance the tail on the chip if necessary */ + if (dd->ipath_sdma_descq_tail != tail) { + wmb(); + ipath_write_kreg(dd, dd->ipath_kregs->kr_senddmatail, tail); + dd->ipath_sdma_descq_tail = tail; + } + +unlock: + spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags); + + return ret; +} + +int ipath_user_sdma_writev(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + const struct iovec *iov, + unsigned long dim) +{ + int ret = 0; + struct list_head list; + int npkts = 0; + + INIT_LIST_HEAD(&list); + + mutex_lock(&pq->lock); + + if (dd->ipath_sdma_descq_added != dd->ipath_sdma_descq_removed) { + ipath_user_sdma_hwqueue_clean(dd); + ipath_user_sdma_queue_clean(dd, pq); + } + + while (dim) { + const int mxp = 8; + + down_write(¤t->mm->mmap_sem); + ret = ipath_user_sdma_queue_pkts(dd, pq, &list, iov, dim, mxp); + up_write(¤t->mm->mmap_sem); + + if (ret <= 0) + goto done_unlock; + else { + dim -= ret; + iov += ret; + } + + /* force packets onto the sdma hw queue... */ + if (!list_empty(&list)) { + /* + * lazily clean hw queue. the 4 is a guess of about + * how many sdma descriptors a packet will take (it + * doesn't have to be perfect). + */ + if (ipath_sdma_descq_freecnt(dd) < ret * 4) { + ipath_user_sdma_hwqueue_clean(dd); + ipath_user_sdma_queue_clean(dd, pq); + } + + ret = ipath_user_sdma_push_pkts(dd, pq, &list); + if (ret < 0) + goto done_unlock; + else { + npkts += ret; + pq->counter += ret; + + if (!list_empty(&list)) + goto done_unlock; + } + } + } + +done_unlock: + if (!list_empty(&list)) + ipath_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &list); + mutex_unlock(&pq->lock); + + return (ret < 0) ? ret : npkts; +} + +int ipath_user_sdma_make_progress(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq) +{ + int ret = 0; + + mutex_lock(&pq->lock); + ipath_user_sdma_hwqueue_clean(dd); + ret = ipath_user_sdma_queue_clean(dd, pq); + mutex_unlock(&pq->lock); + + return ret; +} + +u32 ipath_user_sdma_complete_counter(const struct ipath_user_sdma_queue *pq) +{ + return pq->sent_counter; +} + +u32 ipath_user_sdma_inflight_counter(struct ipath_user_sdma_queue *pq) +{ + return pq->counter; +} + diff --git a/drivers/infiniband/hw/ipath/ipath_user_sdma.h b/drivers/infiniband/hw/ipath/ipath_user_sdma.h new file mode 100644 index 00000000000..e70946c1428 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_user_sdma.h @@ -0,0 +1,54 @@ +/* + * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/device.h> + +struct ipath_user_sdma_queue; + +struct ipath_user_sdma_queue * +ipath_user_sdma_queue_create(struct device *dev, int unit, int port, int sport); +void ipath_user_sdma_queue_destroy(struct ipath_user_sdma_queue *pq); + +int ipath_user_sdma_writev(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq, + const struct iovec *iov, + unsigned long dim); + +int ipath_user_sdma_make_progress(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq); + +int ipath_user_sdma_pkt_sent(const struct ipath_user_sdma_queue *pq, + u32 counter); +void ipath_user_sdma_queue_drain(struct ipath_devdata *dd, + struct ipath_user_sdma_queue *pq); + +u32 ipath_user_sdma_complete_counter(const struct ipath_user_sdma_queue *pq); +u32 ipath_user_sdma_inflight_counter(struct ipath_user_sdma_queue *pq); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index 32d8f882e56..320a6d018de 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -242,6 +242,93 @@ static void ipath_flush_wqe(struct ipath_qp *qp, struct ib_send_wr *wr) ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 1); } +/* + * Count the number of DMA descriptors needed to send length bytes of data. + * Don't modify the ipath_sge_state to get the count. + * Return zero if any of the segments is not aligned. + */ +static u32 ipath_count_sge(struct ipath_sge_state *ss, u32 length) +{ + struct ipath_sge *sg_list = ss->sg_list; + struct ipath_sge sge = ss->sge; + u8 num_sge = ss->num_sge; + u32 ndesc = 1; /* count the header */ + + while (length) { + u32 len = sge.length; + + if (len > length) + len = length; + if (len > sge.sge_length) + len = sge.sge_length; + BUG_ON(len == 0); + if (((long) sge.vaddr & (sizeof(u32) - 1)) || + (len != length && (len & (sizeof(u32) - 1)))) { + ndesc = 0; + break; + } + ndesc++; + sge.vaddr += len; + sge.length -= len; + sge.sge_length -= len; + if (sge.sge_length == 0) { + if (--num_sge) + sge = *sg_list++; + } else if (sge.length == 0 && sge.mr != NULL) { + if (++sge.n >= IPATH_SEGSZ) { + if (++sge.m >= sge.mr->mapsz) + break; + sge.n = 0; + } + sge.vaddr = + sge.mr->map[sge.m]->segs[sge.n].vaddr; + sge.length = + sge.mr->map[sge.m]->segs[sge.n].length; + } + length -= len; + } + return ndesc; +} + +/* + * Copy from the SGEs to the data buffer. + */ +static void ipath_copy_from_sge(void *data, struct ipath_sge_state *ss, + u32 length) +{ + struct ipath_sge *sge = &ss->sge; + + while (length) { + u32 len = sge->length; + + if (len > length) + len = length; + if (len > sge->sge_length) + len = sge->sge_length; + BUG_ON(len == 0); + memcpy(data, sge->vaddr, len); + sge->vaddr += len; + sge->length -= len; + sge->sge_length -= len; + if (sge->sge_length == 0) { + if (--ss->num_sge) + *sge = *ss->sg_list++; + } else if (sge->length == 0 && sge->mr != NULL) { + if (++sge->n >= IPATH_SEGSZ) { + if (++sge->m >= sge->mr->mapsz) + break; + sge->n = 0; + } + sge->vaddr = + sge->mr->map[sge->m]->segs[sge->n].vaddr; + sge->length = + sge->mr->map[sge->m]->segs[sge->n].length; + } + data += len; + length -= len; + } +} + /** * ipath_post_one_send - post one RC, UC, or UD send work request * @qp: the QP to post on @@ -866,27 +953,257 @@ static void copy_io(u32 __iomem *piobuf, struct ipath_sge_state *ss, __raw_writel(last, piobuf); } -static int ipath_verbs_send_pio(struct ipath_qp *qp, u32 *hdr, u32 hdrwords, +/* + * Convert IB rate to delay multiplier. + */ +unsigned ipath_ib_rate_to_mult(enum ib_rate rate) +{ + switch (rate) { + case IB_RATE_2_5_GBPS: return 8; + case IB_RATE_5_GBPS: return 4; + case IB_RATE_10_GBPS: return 2; + case IB_RATE_20_GBPS: return 1; + default: return 0; + } +} + +/* + * Convert delay multiplier to IB rate + */ +static enum ib_rate ipath_mult_to_ib_rate(unsigned mult) +{ + switch (mult) { + case 8: return IB_RATE_2_5_GBPS; + case 4: return IB_RATE_5_GBPS; + case 2: return IB_RATE_10_GBPS; + case 1: return IB_RATE_20_GBPS; + default: return IB_RATE_PORT_CURRENT; + } +} + +static inline struct ipath_verbs_txreq *get_txreq(struct ipath_ibdev *dev) +{ + struct ipath_verbs_txreq *tx = NULL; + unsigned long flags; + + spin_lock_irqsave(&dev->pending_lock, flags); + if (!list_empty(&dev->txreq_free)) { + struct list_head *l = dev->txreq_free.next; + + list_del(l); + tx = list_entry(l, struct ipath_verbs_txreq, txreq.list); + } + spin_unlock_irqrestore(&dev->pending_lock, flags); + return tx; +} + +static inline void put_txreq(struct ipath_ibdev *dev, + struct ipath_verbs_txreq *tx) +{ + unsigned long flags; + + spin_lock_irqsave(&dev->pending_lock, flags); + list_add(&tx->txreq.list, &dev->txreq_free); + spin_unlock_irqrestore(&dev->pending_lock, flags); +} + +static void sdma_complete(void *cookie, int status) +{ + struct ipath_verbs_txreq *tx = cookie; + struct ipath_qp *qp = tx->qp; + struct ipath_ibdev *dev = to_idev(qp->ibqp.device); + + /* Generate a completion queue entry if needed */ + if (qp->ibqp.qp_type != IB_QPT_RC && tx->wqe) { + enum ib_wc_status ibs = status == IPATH_SDMA_TXREQ_S_OK ? + IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR; + + ipath_send_complete(qp, tx->wqe, ibs); + } + + if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_FREEBUF) + kfree(tx->txreq.map_addr); + put_txreq(dev, tx); + + if (atomic_dec_and_test(&qp->refcount)) + wake_up(&qp->wait); +} + +/* + * Compute the number of clock cycles of delay before sending the next packet. + * The multipliers reflect the number of clocks for the fastest rate so + * one tick at 4xDDR is 8 ticks at 1xSDR. + * If the destination port will take longer to receive a packet than + * the outgoing link can send it, we need to delay sending the next packet + * by the difference in time it takes the receiver to receive and the sender + * to send this packet. + * Note that this delay is always correct for UC and RC but not always + * optimal for UD. For UD, the destination HCA can be different for each + * packet, in which case, we could send packets to a different destination + * while "waiting" for the delay. The overhead for doing this without + * HW support is more than just paying the cost of delaying some packets + * unnecessarily. + */ +static inline unsigned ipath_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult) +{ + return (rcv_mult > snd_mult) ? + (plen * (rcv_mult - snd_mult) + 1) >> 1 : 0; +} + +static int ipath_verbs_send_dma(struct ipath_qp *qp, + struct ipath_ib_header *hdr, u32 hdrwords, + struct ipath_sge_state *ss, u32 len, + u32 plen, u32 dwords) +{ + struct ipath_ibdev *dev = to_idev(qp->ibqp.device); + struct ipath_devdata *dd = dev->dd; + struct ipath_verbs_txreq *tx; + u32 *piobuf; + u32 control; + u32 ndesc; + int ret; + + tx = qp->s_tx; + if (tx) { + qp->s_tx = NULL; + /* resend previously constructed packet */ + ret = ipath_sdma_verbs_send(dd, tx->ss, tx->len, tx); + if (ret) + qp->s_tx = tx; + goto bail; + } + + tx = get_txreq(dev); + if (!tx) { + ret = -EBUSY; + goto bail; + } + + /* + * Get the saved delay count we computed for the previous packet + * and save the delay count for this packet to be used next time + * we get here. + */ + control = qp->s_pkt_delay; + qp->s_pkt_delay = ipath_pkt_delay(plen, dd->delay_mult, qp->s_dmult); + + tx->qp = qp; + atomic_inc(&qp->refcount); + tx->wqe = qp->s_wqe; + tx->txreq.callback = sdma_complete; + tx->txreq.callback_cookie = tx; + tx->txreq.flags = IPATH_SDMA_TXREQ_F_HEADTOHOST | + IPATH_SDMA_TXREQ_F_INTREQ | IPATH_SDMA_TXREQ_F_FREEDESC; + if (plen + 1 >= IPATH_SMALLBUF_DWORDS) + tx->txreq.flags |= IPATH_SDMA_TXREQ_F_USELARGEBUF; + + /* VL15 packets bypass credit check */ + if ((be16_to_cpu(hdr->lrh[0]) >> 12) == 15) { + control |= 1ULL << 31; + tx->txreq.flags |= IPATH_SDMA_TXREQ_F_VL15; + } + + if (len) { + /* + * Don't try to DMA if it takes more descriptors than + * the queue holds. + */ + ndesc = ipath_count_sge(ss, len); + if (ndesc >= dd->ipath_sdma_descq_cnt) + ndesc = 0; + } else + ndesc = 1; + if (ndesc) { + tx->hdr.pbc[0] = cpu_to_le32(plen); + tx->hdr.pbc[1] = cpu_to_le32(control); + memcpy(&tx->hdr.hdr, hdr, hdrwords << 2); + tx->txreq.sg_count = ndesc; + tx->map_len = (hdrwords + 2) << 2; + tx->txreq.map_addr = &tx->hdr; + ret = ipath_sdma_verbs_send(dd, ss, dwords, tx); + if (ret) { + /* save ss and length in dwords */ + tx->ss = ss; + tx->len = dwords; + qp->s_tx = tx; + } + goto bail; + } + + /* Allocate a buffer and copy the header and payload to it. */ + tx->map_len = (plen + 1) << 2; + piobuf = kmalloc(tx->map_len, GFP_ATOMIC); + if (unlikely(piobuf == NULL)) { + ret = -EBUSY; + goto err_tx; + } + tx->txreq.map_addr = piobuf; + tx->txreq.flags |= IPATH_SDMA_TXREQ_F_FREEBUF; + tx->txreq.sg_count = 1; + + *piobuf++ = (__force u32) cpu_to_le32(plen); + *piobuf++ = (__force u32) cpu_to_le32(control); + memcpy(piobuf, hdr, hdrwords << 2); + ipath_copy_from_sge(piobuf + hdrwords, ss, len); + + ret = ipath_sdma_verbs_send(dd, NULL, 0, tx); + /* + * If we couldn't queue the DMA request, save the info + * and try again later rather than destroying the + * buffer and undoing the side effects of the copy. + */ + if (ret) { + tx->ss = NULL; + tx->len = 0; + qp->s_tx = tx; + } + dev->n_unaligned++; + goto bail; + +err_tx: + if (atomic_dec_and_test(&qp->refcount)) + wake_up(&qp->wait); + put_txreq(dev, tx); +bail: + return ret; +} + +static int ipath_verbs_send_pio(struct ipath_qp *qp, + struct ipath_ib_header *ibhdr, u32 hdrwords, struct ipath_sge_state *ss, u32 len, u32 plen, u32 dwords) { struct ipath_devdata *dd = to_idev(qp->ibqp.device)->dd; + u32 *hdr = (u32 *) ibhdr; u32 __iomem *piobuf; unsigned flush_wc; + u32 control; int ret; - piobuf = ipath_getpiobuf(dd, NULL); + piobuf = ipath_getpiobuf(dd, plen, NULL); if (unlikely(piobuf == NULL)) { ret = -EBUSY; goto bail; } /* - * Write len to control qword, no flags. + * Get the saved delay count we computed for the previous packet + * and save the delay count for this packet to be used next time + * we get here. + */ + control = qp->s_pkt_delay; + qp->s_pkt_delay = ipath_pkt_delay(plen, dd->delay_mult, qp->s_dmult); + + /* VL15 packets bypass credit check */ + if ((be16_to_cpu(ibhdr->lrh[0]) >> 12) == 15) + control |= 1ULL << 31; + + /* + * Write the length to the control qword plus any needed flags. * We have to flush after the PBC for correctness on some cpus * or WC buffer can be written out of order. */ - writeq(plen, piobuf); + writeq(((u64) control << 32) | plen, piobuf); piobuf += 2; flush_wc = dd->ipath_flags & IPATH_PIO_FLUSH_WC; @@ -961,15 +1278,25 @@ int ipath_verbs_send(struct ipath_qp *qp, struct ipath_ib_header *hdr, */ plen = hdrwords + dwords + 1; - /* Drop non-VL15 packets if we are not in the active state */ - if (!(dd->ipath_flags & IPATH_LINKACTIVE) && - qp->ibqp.qp_type != IB_QPT_SMI) { + /* + * VL15 packets (IB_QPT_SMI) will always use PIO, so we + * can defer SDMA restart until link goes ACTIVE without + * worrying about just how we got there. + */ + if (qp->ibqp.qp_type == IB_QPT_SMI) + ret = ipath_verbs_send_pio(qp, hdr, hdrwords, ss, len, + plen, dwords); + /* All non-VL15 packets are dropped if link is not ACTIVE */ + else if (!(dd->ipath_flags & IPATH_LINKACTIVE)) { if (qp->s_wqe) ipath_send_complete(qp, qp->s_wqe, IB_WC_SUCCESS); ret = 0; - } else - ret = ipath_verbs_send_pio(qp, (u32 *) hdr, hdrwords, - ss, len, plen, dwords); + } else if (dd->ipath_flags & IPATH_HAS_SEND_DMA) + ret = ipath_verbs_send_dma(qp, hdr, hdrwords, ss, len, + plen, dwords); + else + ret = ipath_verbs_send_pio(qp, hdr, hdrwords, ss, len, + plen, dwords); return ret; } @@ -1038,6 +1365,12 @@ int ipath_get_counters(struct ipath_devdata *dd, ipath_snap_cntr(dd, crp->cr_errlpcrccnt) + ipath_snap_cntr(dd, crp->cr_badformatcnt) + dd->ipath_rxfc_unsupvl_errs; + if (crp->cr_rxotherlocalphyerrcnt) + cntrs->port_rcv_errors += + ipath_snap_cntr(dd, crp->cr_rxotherlocalphyerrcnt); + if (crp->cr_rxvlerrcnt) + cntrs->port_rcv_errors += + ipath_snap_cntr(dd, crp->cr_rxvlerrcnt); cntrs->port_rcv_remphys_errors = ipath_snap_cntr(dd, crp->cr_rcvebpcnt); cntrs->port_xmit_discards = ipath_snap_cntr(dd, crp->cr_unsupvlcnt); @@ -1046,9 +1379,16 @@ int ipath_get_counters(struct ipath_devdata *dd, cntrs->port_xmit_packets = ipath_snap_cntr(dd, crp->cr_pktsendcnt); cntrs->port_rcv_packets = ipath_snap_cntr(dd, crp->cr_pktrcvcnt); cntrs->local_link_integrity_errors = - (dd->ipath_flags & IPATH_GPIO_ERRINTRS) ? - dd->ipath_lli_errs : dd->ipath_lli_errors; - cntrs->excessive_buffer_overrun_errors = dd->ipath_overrun_thresh_errs; + crp->cr_locallinkintegrityerrcnt ? + ipath_snap_cntr(dd, crp->cr_locallinkintegrityerrcnt) : + ((dd->ipath_flags & IPATH_GPIO_ERRINTRS) ? + dd->ipath_lli_errs : dd->ipath_lli_errors); + cntrs->excessive_buffer_overrun_errors = + crp->cr_excessbufferovflcnt ? + ipath_snap_cntr(dd, crp->cr_excessbufferovflcnt) : + dd->ipath_overrun_thresh_errs; + cntrs->vl15_dropped = crp->cr_vl15droppedpktcnt ? + ipath_snap_cntr(dd, crp->cr_vl15droppedpktcnt) : 0; ret = 0; @@ -1183,7 +1523,9 @@ static int ipath_query_port(struct ib_device *ibdev, props->sm_lid = dev->sm_lid; props->sm_sl = dev->sm_sl; ibcstat = dd->ipath_lastibcstat; - props->state = ((ibcstat >> 4) & 0x3) + 1; + /* map LinkState to IB portinfo values. */ + props->state = ipath_ib_linkstate(dd, ibcstat) + 1; + /* See phys_state_show() */ props->phys_state = /* MEA: assumes shift == 0 */ ipath_cvt_physportstate[dd->ipath_lastibcstat & @@ -1195,18 +1537,13 @@ static int ipath_query_port(struct ib_device *ibdev, props->bad_pkey_cntr = ipath_get_cr_errpkey(dd) - dev->z_pkey_violations; props->qkey_viol_cntr = dev->qkey_violations; - props->active_width = IB_WIDTH_4X; + props->active_width = dd->ipath_link_width_active; /* See rate_show() */ - props->active_speed = 1; /* Regular 10Mbs speed. */ + props->active_speed = dd->ipath_link_speed_active; props->max_vl_num = 1; /* VLCap = VL0 */ props->init_type_reply = 0; - /* - * Note: the chip supports a maximum MTU of 4096, but the driver - * hasn't implemented this feature yet, so set the maximum value - * to 2048. - */ - props->max_mtu = IB_MTU_2048; + props->max_mtu = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048; switch (dd->ipath_ibmtu) { case 4096: mtu = IB_MTU_4096; @@ -1399,6 +1736,7 @@ static struct ib_ah *ipath_create_ah(struct ib_pd *pd, /* ib_create_ah() will initialize ah->ibah. */ ah->attr = *ah_attr; + ah->attr.static_rate = ipath_ib_rate_to_mult(ah_attr->static_rate); ret = &ah->ibah; @@ -1432,6 +1770,7 @@ static int ipath_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr) struct ipath_ah *ah = to_iah(ibah); *ah_attr = ah->attr; + ah_attr->static_rate = ipath_mult_to_ib_rate(ah->attr.static_rate); return 0; } @@ -1581,6 +1920,8 @@ int ipath_register_ib_device(struct ipath_devdata *dd) struct ipath_verbs_counters cntrs; struct ipath_ibdev *idev; struct ib_device *dev; + struct ipath_verbs_txreq *tx; + unsigned i; int ret; idev = (struct ipath_ibdev *)ib_alloc_device(sizeof *idev); @@ -1591,6 +1932,17 @@ int ipath_register_ib_device(struct ipath_devdata *dd) dev = &idev->ibdev; + if (dd->ipath_sdma_descq_cnt) { + tx = kmalloc(dd->ipath_sdma_descq_cnt * sizeof *tx, + GFP_KERNEL); + if (tx == NULL) { + ret = -ENOMEM; + goto err_tx; + } + } else + tx = NULL; + idev->txreq_bufs = tx; + /* Only need to initialize non-zero fields. */ spin_lock_init(&idev->n_pds_lock); spin_lock_init(&idev->n_ahs_lock); @@ -1631,15 +1983,17 @@ int ipath_register_ib_device(struct ipath_devdata *dd) INIT_LIST_HEAD(&idev->pending[2]); INIT_LIST_HEAD(&idev->piowait); INIT_LIST_HEAD(&idev->rnrwait); + INIT_LIST_HEAD(&idev->txreq_free); idev->pending_index = 0; idev->port_cap_flags = IB_PORT_SYS_IMAGE_GUID_SUP | IB_PORT_CLIENT_REG_SUP; + if (dd->ipath_flags & IPATH_HAS_LINK_LATENCY) + idev->port_cap_flags |= IB_PORT_LINK_LATENCY_SUP; idev->pma_counter_select[0] = IB_PMA_PORT_XMIT_DATA; idev->pma_counter_select[1] = IB_PMA_PORT_RCV_DATA; idev->pma_counter_select[2] = IB_PMA_PORT_XMIT_PKTS; idev->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS; idev->pma_counter_select[4] = IB_PMA_PORT_XMIT_WAIT; - idev->link_width_enabled = 3; /* 1x or 4x */ /* Snapshot current HW counters to "clear" them. */ ipath_get_counters(dd, &cntrs); @@ -1661,6 +2015,9 @@ int ipath_register_ib_device(struct ipath_devdata *dd) cntrs.excessive_buffer_overrun_errors; idev->z_vl15_dropped = cntrs.vl15_dropped; + for (i = 0; i < dd->ipath_sdma_descq_cnt; i++, tx++) + list_add(&tx->txreq.list, &idev->txreq_free); + /* * The system image GUID is supposed to be the same for all * IB HCAs in a single system but since there can be other @@ -1710,6 +2067,7 @@ int ipath_register_ib_device(struct ipath_devdata *dd) dev->phys_port_cnt = 1; dev->num_comp_vectors = 1; dev->dma_device = &dd->pcidev->dev; + dev->class_dev.dev = dev->dma_device; dev->query_device = ipath_query_device; dev->modify_device = ipath_modify_device; dev->query_port = ipath_query_port; @@ -1774,6 +2132,8 @@ err_reg: err_lk: kfree(idev->qp_table.table); err_qp: + kfree(idev->txreq_bufs); +err_tx: ib_dealloc_device(dev); ipath_dev_err(dd, "cannot register verbs: %d!\n", -ret); idev = NULL; @@ -1808,6 +2168,7 @@ void ipath_unregister_ib_device(struct ipath_ibdev *dev) ipath_free_all_qps(&dev->qp_table); kfree(dev->qp_table.table); kfree(dev->lk_table.table); + kfree(dev->txreq_bufs); ib_dealloc_device(ibdev); } @@ -1855,13 +2216,15 @@ static ssize_t show_stats(struct class_device *cdev, char *buf) "RC stalls %d\n" "piobuf wait %d\n" "no piobuf %d\n" + "unaligned %d\n" "PKT drops %d\n" "WQE errs %d\n", dev->n_rc_resends, dev->n_rc_qacks, dev->n_rc_acks, dev->n_seq_naks, dev->n_rdma_seq, dev->n_rnr_naks, dev->n_other_naks, dev->n_timeouts, dev->n_rdma_dup_busy, dev->n_rc_stalls, dev->n_piowait, - dev->n_no_piobuf, dev->n_pkt_drops, dev->n_wqe_errs); + dev->n_no_piobuf, dev->n_unaligned, + dev->n_pkt_drops, dev->n_wqe_errs); for (i = 0; i < ARRAY_SIZE(dev->opstats); i++) { const struct ipath_opcode_stats *si = &dev->opstats[i]; diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 3d59736b49b..6514aa8306c 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. + * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved. * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. * * This software is available to you under a choice of one of two @@ -138,6 +138,11 @@ struct ipath_ib_header { } u; } __attribute__ ((packed)); +struct ipath_pio_header { + __le32 pbc[2]; + struct ipath_ib_header hdr; +} __attribute__ ((packed)); + /* * There is one struct ipath_mcast for each multicast GID. * All attached QPs are then stored as a list of @@ -319,6 +324,7 @@ struct ipath_sge_state { struct ipath_sge *sg_list; /* next SGE to be used if any */ struct ipath_sge sge; /* progress state for the current SGE */ u8 num_sge; + u8 static_rate; }; /* @@ -356,6 +362,7 @@ struct ipath_qp { struct tasklet_struct s_task; struct ipath_mmap_info *ip; struct ipath_sge_state *s_cur_sge; + struct ipath_verbs_txreq *s_tx; struct ipath_sge_state s_sge; /* current send request data */ struct ipath_ack_entry s_ack_queue[IPATH_MAX_RDMA_ATOMIC + 1]; struct ipath_sge_state s_ack_rdma_sge; @@ -363,7 +370,8 @@ struct ipath_qp { struct ipath_sge_state r_sge; /* current receive data */ spinlock_t s_lock; unsigned long s_busy; - u32 s_hdrwords; /* size of s_hdr in 32 bit words */ + u16 s_pkt_delay; + u16 s_hdrwords; /* size of s_hdr in 32 bit words */ u32 s_cur_size; /* size of send packet in bytes */ u32 s_len; /* total length of s_sge */ u32 s_rdma_read_len; /* total length of s_rdma_read_sge */ @@ -387,7 +395,6 @@ struct ipath_qp { u8 r_nak_state; /* non-zero if NAK is pending */ u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */ u8 r_reuse_sge; /* for UC receive errors */ - u8 r_sge_inx; /* current index into sg_list */ u8 r_wrid_valid; /* r_wrid set but CQ entry not yet made */ u8 r_max_rd_atomic; /* max number of RDMA read/atomic to receive */ u8 r_head_ack_queue; /* index into s_ack_queue[] */ @@ -403,6 +410,7 @@ struct ipath_qp { u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */ u8 s_tail_ack_queue; /* index into s_ack_queue[] */ u8 s_flags; + u8 s_dmult; u8 timeout; /* Timeout for this QP */ enum ib_mtu path_mtu; u32 remote_qpn; @@ -510,6 +518,8 @@ struct ipath_ibdev { struct ipath_lkey_table lk_table; struct list_head pending[3]; /* FIFO of QPs waiting for ACKs */ struct list_head piowait; /* list for wait PIO buf */ + struct list_head txreq_free; + void *txreq_bufs; /* list of QPs waiting for RNR timer */ struct list_head rnrwait; spinlock_t pending_lock; @@ -570,6 +580,7 @@ struct ipath_ibdev { u32 n_rdma_dup_busy; u32 n_piowait; u32 n_no_piobuf; + u32 n_unaligned; u32 port_cap_flags; u32 pma_sample_start; u32 pma_sample_interval; @@ -581,7 +592,6 @@ struct ipath_ibdev { u16 pending_index; /* which pending queue is active */ u8 pma_sample_status; u8 subnet_timeout; - u8 link_width_enabled; u8 vl_high_limit; struct ipath_opcode_stats opstats[128]; }; @@ -602,6 +612,16 @@ struct ipath_verbs_counters { u32 vl15_dropped; }; +struct ipath_verbs_txreq { + struct ipath_qp *qp; + struct ipath_swqe *wqe; + u32 map_len; + u32 len; + struct ipath_sge_state *ss; + struct ipath_pio_header hdr; + struct ipath_sdma_txreq txreq; +}; + static inline struct ipath_mr *to_imr(struct ib_mr *ibmr) { return container_of(ibmr, struct ipath_mr, ibmr); @@ -694,11 +714,11 @@ void ipath_sqerror_qp(struct ipath_qp *qp, struct ib_wc *wc); void ipath_get_credit(struct ipath_qp *qp, u32 aeth); +unsigned ipath_ib_rate_to_mult(enum ib_rate rate); + int ipath_verbs_send(struct ipath_qp *qp, struct ipath_ib_header *hdr, u32 hdrwords, struct ipath_sge_state *ss, u32 len); -void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int sig); - void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length); void ipath_skip_sge(struct ipath_sge_state *ss, u32 length); |