From 9bc57e2d19db4da81c1150120658cc3658a99ed4 Mon Sep 17 00:00:00 2001 From: Ralph Campbell Date: Fri, 11 Aug 2006 14:58:09 -0700 Subject: IB/uverbs: Pass userspace data to modify_srq and modify_qp methods Pass a struct ib_udata to the low-level driver's ->modify_srq() and ->modify_qp() methods, so that it can get to the device-specific data passed in by the userspace driver. Signed-off-by: Ralph Campbell Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_verbs.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 2df684727dc..698396778f0 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -579,7 +579,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, int ipath_destroy_qp(struct ib_qp *ibqp); int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, - int attr_mask); + int attr_mask, struct ib_udata *udata); int ipath_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask, struct ib_qp_init_attr *init_attr); @@ -638,7 +638,8 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd, struct ib_udata *udata); int ipath_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, - enum ib_srq_attr_mask attr_mask); + enum ib_srq_attr_mask attr_mask, + struct ib_udata *udata); int ipath_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr); -- cgit v1.2.3-70-g09d2 From 373d9915803aebbbf7fd3841efd9dac31c32e148 Mon Sep 17 00:00:00 2001 From: Ralph Campbell Date: Fri, 22 Sep 2006 15:22:26 -0700 Subject: IB/ipath: Performance improvements via mmap of queues Improve performance of userspace post receive, post SRQ receive, and poll CQ operations for ipath by allowing userspace to directly mmap() receive queues and completion queues. This eliminates the copying between userspace and the kernel in the data path. Signed-off-by: Ralph Campbell Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/Makefile | 1 + drivers/infiniband/hw/ipath/ipath_cq.c | 176 ++++++++++++++++------ drivers/infiniband/hw/ipath/ipath_mmap.c | 122 +++++++++++++++ drivers/infiniband/hw/ipath/ipath_qp.c | 156 +++++++++++++------ drivers/infiniband/hw/ipath/ipath_ruc.c | 138 +++++++++++------ drivers/infiniband/hw/ipath/ipath_srq.c | 240 +++++++++++++++++++----------- drivers/infiniband/hw/ipath/ipath_ud.c | 169 +++++++++++++-------- drivers/infiniband/hw/ipath/ipath_verbs.c | 50 ++----- drivers/infiniband/hw/ipath/ipath_verbs.h | 115 +++++++------- 9 files changed, 785 insertions(+), 382 deletions(-) create mode 100644 drivers/infiniband/hw/ipath/ipath_mmap.c (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/Makefile b/drivers/infiniband/hw/ipath/Makefile index b0bf7286413..6bb43474d10 100644 --- a/drivers/infiniband/hw/ipath/Makefile +++ b/drivers/infiniband/hw/ipath/Makefile @@ -25,6 +25,7 @@ ib_ipath-y := \ ipath_cq.o \ ipath_keys.o \ ipath_mad.o \ + ipath_mmap.o \ ipath_mr.o \ ipath_qp.o \ ipath_rc.o \ diff --git a/drivers/infiniband/hw/ipath/ipath_cq.c b/drivers/infiniband/hw/ipath/ipath_cq.c index 3efee341c9b..3c4c198a451 100644 --- a/drivers/infiniband/hw/ipath/ipath_cq.c +++ b/drivers/infiniband/hw/ipath/ipath_cq.c @@ -42,20 +42,28 @@ * @entry: work completion entry to add * @sig: true if @entry is a solicitated entry * - * This may be called with one of the qp->s_lock or qp->r_rq.lock held. + * This may be called with qp->s_lock held. */ void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited) { + struct ipath_cq_wc *wc = cq->queue; unsigned long flags; + u32 head; u32 next; spin_lock_irqsave(&cq->lock, flags); - if (cq->head == cq->ibcq.cqe) + /* + * Note that the head pointer might be writable by user processes. + * Take care to verify it is a sane value. + */ + head = wc->head; + if (head >= (unsigned) cq->ibcq.cqe) { + head = cq->ibcq.cqe; next = 0; - else - next = cq->head + 1; - if (unlikely(next == cq->tail)) { + } else + next = head + 1; + if (unlikely(next == wc->tail)) { spin_unlock_irqrestore(&cq->lock, flags); if (cq->ibcq.event_handler) { struct ib_event ev; @@ -67,8 +75,8 @@ void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited) } return; } - cq->queue[cq->head] = *entry; - cq->head = next; + wc->queue[head] = *entry; + wc->head = next; if (cq->notify == IB_CQ_NEXT_COMP || (cq->notify == IB_CQ_SOLICITED && solicited)) { @@ -101,19 +109,20 @@ void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited) int ipath_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry) { struct ipath_cq *cq = to_icq(ibcq); + struct ipath_cq_wc *wc = cq->queue; unsigned long flags; int npolled; spin_lock_irqsave(&cq->lock, flags); for (npolled = 0; npolled < num_entries; ++npolled, ++entry) { - if (cq->tail == cq->head) + if (wc->tail == wc->head) break; - *entry = cq->queue[cq->tail]; - if (cq->tail == cq->ibcq.cqe) - cq->tail = 0; + *entry = wc->queue[wc->tail]; + if (wc->tail >= cq->ibcq.cqe) + wc->tail = 0; else - cq->tail++; + wc->tail++; } spin_unlock_irqrestore(&cq->lock, flags); @@ -160,38 +169,74 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries, { struct ipath_ibdev *dev = to_idev(ibdev); struct ipath_cq *cq; - struct ib_wc *wc; + struct ipath_cq_wc *wc; struct ib_cq *ret; if (entries > ib_ipath_max_cqes) { ret = ERR_PTR(-EINVAL); - goto bail; + goto done; } if (dev->n_cqs_allocated == ib_ipath_max_cqs) { ret = ERR_PTR(-ENOMEM); - goto bail; + goto done; } - /* - * Need to use vmalloc() if we want to support large #s of - * entries. - */ + /* Allocate the completion queue structure. */ cq = kmalloc(sizeof(*cq), GFP_KERNEL); if (!cq) { ret = ERR_PTR(-ENOMEM); - goto bail; + goto done; } /* - * Need to use vmalloc() if we want to support large #s of entries. + * Allocate the completion queue entries and head/tail pointers. + * This is allocated separately so that it can be resized and + * also mapped into user space. + * We need to use vmalloc() in order to support mmap and large + * numbers of entries. */ - wc = vmalloc(sizeof(*wc) * (entries + 1)); + wc = vmalloc_user(sizeof(*wc) + sizeof(struct ib_wc) * entries); if (!wc) { - kfree(cq); ret = ERR_PTR(-ENOMEM); - goto bail; + goto bail_cq; } + + /* + * Return the address of the WC as the offset to mmap. + * See ipath_mmap() for details. + */ + if (udata && udata->outlen >= sizeof(__u64)) { + struct ipath_mmap_info *ip; + __u64 offset = (__u64) wc; + int err; + + err = ib_copy_to_udata(udata, &offset, sizeof(offset)); + if (err) { + ret = ERR_PTR(err); + goto bail_wc; + } + + /* Allocate info for ipath_mmap(). */ + ip = kmalloc(sizeof(*ip), GFP_KERNEL); + if (!ip) { + ret = ERR_PTR(-ENOMEM); + goto bail_wc; + } + cq->ip = ip; + ip->context = context; + ip->obj = wc; + kref_init(&ip->ref); + ip->mmap_cnt = 0; + ip->size = PAGE_ALIGN(sizeof(*wc) + + sizeof(struct ib_wc) * entries); + spin_lock_irq(&dev->pending_lock); + ip->next = dev->pending_mmaps; + dev->pending_mmaps = ip; + spin_unlock_irq(&dev->pending_lock); + } else + cq->ip = NULL; + /* * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe. * The number of entries should be >= the number requested or return @@ -202,15 +247,22 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries, cq->triggered = 0; spin_lock_init(&cq->lock); tasklet_init(&cq->comptask, send_complete, (unsigned long)cq); - cq->head = 0; - cq->tail = 0; + wc->head = 0; + wc->tail = 0; cq->queue = wc; ret = &cq->ibcq; dev->n_cqs_allocated++; + goto done; -bail: +bail_wc: + vfree(wc); + +bail_cq: + kfree(cq); + +done: return ret; } @@ -229,7 +281,10 @@ int ipath_destroy_cq(struct ib_cq *ibcq) tasklet_kill(&cq->comptask); dev->n_cqs_allocated--; - vfree(cq->queue); + if (cq->ip) + kref_put(&cq->ip->ref, ipath_release_mmap_info); + else + vfree(cq->queue); kfree(cq); return 0; @@ -253,7 +308,7 @@ int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) spin_lock_irqsave(&cq->lock, flags); /* * Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow - * any other transitions. + * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2). */ if (cq->notify != IB_CQ_NEXT_COMP) cq->notify = notify; @@ -264,46 +319,81 @@ int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata) { struct ipath_cq *cq = to_icq(ibcq); - struct ib_wc *wc, *old_wc; - u32 n; + struct ipath_cq_wc *old_wc = cq->queue; + struct ipath_cq_wc *wc; + u32 head, tail, n; int ret; /* * Need to use vmalloc() if we want to support large #s of entries. */ - wc = vmalloc(sizeof(*wc) * (cqe + 1)); + wc = vmalloc_user(sizeof(*wc) + sizeof(struct ib_wc) * cqe); if (!wc) { ret = -ENOMEM; goto bail; } + /* + * Return the address of the WC as the offset to mmap. + * See ipath_mmap() for details. + */ + if (udata && udata->outlen >= sizeof(__u64)) { + __u64 offset = (__u64) wc; + + ret = ib_copy_to_udata(udata, &offset, sizeof(offset)); + if (ret) + goto bail; + } + spin_lock_irq(&cq->lock); - if (cq->head < cq->tail) - n = cq->ibcq.cqe + 1 + cq->head - cq->tail; + /* + * Make sure head and tail are sane since they + * might be user writable. + */ + head = old_wc->head; + if (head > (u32) cq->ibcq.cqe) + head = (u32) cq->ibcq.cqe; + tail = old_wc->tail; + if (tail > (u32) cq->ibcq.cqe) + tail = (u32) cq->ibcq.cqe; + if (head < tail) + n = cq->ibcq.cqe + 1 + head - tail; else - n = cq->head - cq->tail; + n = head - tail; if (unlikely((u32)cqe < n)) { spin_unlock_irq(&cq->lock); vfree(wc); ret = -EOVERFLOW; goto bail; } - for (n = 0; cq->tail != cq->head; n++) { - wc[n] = cq->queue[cq->tail]; - if (cq->tail == cq->ibcq.cqe) - cq->tail = 0; + for (n = 0; tail != head; n++) { + wc->queue[n] = old_wc->queue[tail]; + if (tail == (u32) cq->ibcq.cqe) + tail = 0; else - cq->tail++; + tail++; } cq->ibcq.cqe = cqe; - cq->head = n; - cq->tail = 0; - old_wc = cq->queue; + wc->head = n; + wc->tail = 0; cq->queue = wc; spin_unlock_irq(&cq->lock); vfree(old_wc); + if (cq->ip) { + struct ipath_ibdev *dev = to_idev(ibcq->device); + struct ipath_mmap_info *ip = cq->ip; + + ip->obj = wc; + ip->size = PAGE_ALIGN(sizeof(*wc) + + sizeof(struct ib_wc) * cqe); + spin_lock_irq(&dev->pending_lock); + ip->next = dev->pending_mmaps; + dev->pending_mmaps = ip; + spin_unlock_irq(&dev->pending_lock); + } + ret = 0; bail: diff --git a/drivers/infiniband/hw/ipath/ipath_mmap.c b/drivers/infiniband/hw/ipath/ipath_mmap.c new file mode 100644 index 00000000000..11b7378ff21 --- /dev/null +++ b/drivers/infiniband/hw/ipath/ipath_mmap.c @@ -0,0 +1,122 @@ +/* + * Copyright (c) 2006 QLogic, 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. + */ + +#include +#include +#include +#include +#include +#include + +#include "ipath_verbs.h" + +/** + * ipath_release_mmap_info - free mmap info structure + * @ref: a pointer to the kref within struct ipath_mmap_info + */ +void ipath_release_mmap_info(struct kref *ref) +{ + struct ipath_mmap_info *ip = + container_of(ref, struct ipath_mmap_info, ref); + + vfree(ip->obj); + kfree(ip); +} + +/* + * open and close keep track of how many times the CQ is mapped, + * to avoid releasing it. + */ +static void ipath_vma_open(struct vm_area_struct *vma) +{ + struct ipath_mmap_info *ip = vma->vm_private_data; + + kref_get(&ip->ref); + ip->mmap_cnt++; +} + +static void ipath_vma_close(struct vm_area_struct *vma) +{ + struct ipath_mmap_info *ip = vma->vm_private_data; + + ip->mmap_cnt--; + kref_put(&ip->ref, ipath_release_mmap_info); +} + +static struct vm_operations_struct ipath_vm_ops = { + .open = ipath_vma_open, + .close = ipath_vma_close, +}; + +/** + * ipath_mmap - create a new mmap region + * @context: the IB user context of the process making the mmap() call + * @vma: the VMA to be initialized + * Return zero if the mmap is OK. Otherwise, return an errno. + */ +int ipath_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) +{ + struct ipath_ibdev *dev = to_idev(context->device); + unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; + unsigned long size = vma->vm_end - vma->vm_start; + struct ipath_mmap_info *ip, **pp; + int ret = -EINVAL; + + /* + * Search the device's list of objects waiting for a mmap call. + * Normally, this list is very short since a call to create a + * CQ, QP, or SRQ is soon followed by a call to mmap(). + */ + spin_lock_irq(&dev->pending_lock); + for (pp = &dev->pending_mmaps; (ip = *pp); pp = &ip->next) { + /* Only the creator is allowed to mmap the object */ + if (context != ip->context || (void *) offset != ip->obj) + continue; + /* Don't allow a mmap larger than the object. */ + if (size > ip->size) + break; + + *pp = ip->next; + spin_unlock_irq(&dev->pending_lock); + + ret = remap_vmalloc_range(vma, ip->obj, 0); + if (ret) + goto done; + vma->vm_ops = &ipath_vm_ops; + vma->vm_private_data = ip; + ipath_vma_open(vma); + goto done; + } + spin_unlock_irq(&dev->pending_lock); +done: + return ret; +} diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c index 44c32d2db99..1ccfc909db1 100644 --- a/drivers/infiniband/hw/ipath/ipath_qp.c +++ b/drivers/infiniband/hw/ipath/ipath_qp.c @@ -35,7 +35,7 @@ #include #include "ipath_verbs.h" -#include "ipath_common.h" +#include "ipath_kernel.h" #define BITS_PER_PAGE (PAGE_SIZE*BITS_PER_BYTE) #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1) @@ -44,19 +44,6 @@ #define find_next_offset(map, off) find_next_zero_bit((map)->page, \ BITS_PER_PAGE, off) -#define TRANS_INVALID 0 -#define TRANS_ANY2RST 1 -#define TRANS_RST2INIT 2 -#define TRANS_INIT2INIT 3 -#define TRANS_INIT2RTR 4 -#define TRANS_RTR2RTS 5 -#define TRANS_RTS2RTS 6 -#define TRANS_SQERR2RTS 7 -#define TRANS_ANY2ERR 8 -#define TRANS_RTS2SQD 9 /* XXX Wait for expected ACKs & signal event */ -#define TRANS_SQD2SQD 10 /* error if not drained & parameter change */ -#define TRANS_SQD2RTS 11 /* error if not drained */ - /* * Convert the AETH credit code into the number of credits. */ @@ -355,8 +342,10 @@ static void ipath_reset_qp(struct ipath_qp *qp) qp->s_last = 0; qp->s_ssn = 1; qp->s_lsn = 0; - qp->r_rq.head = 0; - qp->r_rq.tail = 0; + if (qp->r_rq.wq) { + qp->r_rq.wq->head = 0; + qp->r_rq.wq->tail = 0; + } qp->r_reuse_sge = 0; } @@ -410,15 +399,32 @@ void ipath_error_qp(struct ipath_qp *qp) qp->s_hdrwords = 0; qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE; - wc.opcode = IB_WC_RECV; - spin_lock(&qp->r_rq.lock); - while (qp->r_rq.tail != qp->r_rq.head) { - wc.wr_id = get_rwqe_ptr(&qp->r_rq, qp->r_rq.tail)->wr_id; - if (++qp->r_rq.tail >= qp->r_rq.size) - qp->r_rq.tail = 0; - ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1); + if (qp->r_rq.wq) { + struct ipath_rwq *wq; + u32 head; + u32 tail; + + spin_lock(&qp->r_rq.lock); + + /* sanity check pointers before trusting them */ + wq = qp->r_rq.wq; + head = wq->head; + if (head >= qp->r_rq.size) + head = 0; + tail = wq->tail; + if (tail >= qp->r_rq.size) + tail = 0; + wc.opcode = IB_WC_RECV; + while (tail != head) { + wc.wr_id = get_rwqe_ptr(&qp->r_rq, tail)->wr_id; + if (++tail >= qp->r_rq.size) + tail = 0; + ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1); + } + wq->tail = tail; + + spin_unlock(&qp->r_rq.lock); } - spin_unlock(&qp->r_rq.lock); } /** @@ -544,7 +550,7 @@ int ipath_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, attr->dest_qp_num = qp->remote_qpn; attr->qp_access_flags = qp->qp_access_flags; attr->cap.max_send_wr = qp->s_size - 1; - attr->cap.max_recv_wr = qp->r_rq.size - 1; + attr->cap.max_recv_wr = qp->ibqp.srq ? 0 : qp->r_rq.size - 1; attr->cap.max_send_sge = qp->s_max_sge; attr->cap.max_recv_sge = qp->r_rq.max_sge; attr->cap.max_inline_data = 0; @@ -597,13 +603,23 @@ __be32 ipath_compute_aeth(struct ipath_qp *qp) } else { u32 min, max, x; u32 credits; - + struct ipath_rwq *wq = qp->r_rq.wq; + u32 head; + u32 tail; + + /* sanity check pointers before trusting them */ + head = wq->head; + if (head >= qp->r_rq.size) + head = 0; + tail = wq->tail; + if (tail >= qp->r_rq.size) + tail = 0; /* * Compute the number of credits available (RWQEs). * XXX Not holding the r_rq.lock here so there is a small * chance that the pair of reads are not atomic. */ - credits = qp->r_rq.head - qp->r_rq.tail; + credits = head - tail; if ((int)credits < 0) credits += qp->r_rq.size; /* @@ -680,27 +696,37 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, case IB_QPT_UD: case IB_QPT_SMI: case IB_QPT_GSI: - qp = kmalloc(sizeof(*qp), GFP_KERNEL); + sz = sizeof(*qp); + if (init_attr->srq) { + struct ipath_srq *srq = to_isrq(init_attr->srq); + + sz += sizeof(*qp->r_sg_list) * + srq->rq.max_sge; + } else + sz += sizeof(*qp->r_sg_list) * + init_attr->cap.max_recv_sge; + qp = kmalloc(sz, GFP_KERNEL); if (!qp) { - vfree(swq); ret = ERR_PTR(-ENOMEM); - goto bail; + goto bail_swq; } if (init_attr->srq) { + sz = 0; qp->r_rq.size = 0; qp->r_rq.max_sge = 0; qp->r_rq.wq = NULL; + init_attr->cap.max_recv_wr = 0; + init_attr->cap.max_recv_sge = 0; } else { qp->r_rq.size = init_attr->cap.max_recv_wr + 1; qp->r_rq.max_sge = init_attr->cap.max_recv_sge; - sz = (sizeof(struct ipath_sge) * qp->r_rq.max_sge) + + sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) + sizeof(struct ipath_rwqe); - qp->r_rq.wq = vmalloc(qp->r_rq.size * sz); + qp->r_rq.wq = vmalloc_user(sizeof(struct ipath_rwq) + + qp->r_rq.size * sz); if (!qp->r_rq.wq) { - kfree(qp); - vfree(swq); ret = ERR_PTR(-ENOMEM); - goto bail; + goto bail_qp; } } @@ -726,12 +752,10 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, err = ipath_alloc_qpn(&dev->qp_table, qp, init_attr->qp_type); if (err) { - vfree(swq); - vfree(qp->r_rq.wq); - kfree(qp); ret = ERR_PTR(err); - goto bail; + goto bail_rwq; } + qp->ip = NULL; ipath_reset_qp(qp); /* Tell the core driver that the kernel SMA is present. */ @@ -748,8 +772,51 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, init_attr->cap.max_inline_data = 0; + /* + * Return the address of the RWQ as the offset to mmap. + * See ipath_mmap() for details. + */ + if (udata && udata->outlen >= sizeof(__u64)) { + struct ipath_mmap_info *ip; + __u64 offset = (__u64) qp->r_rq.wq; + int err; + + err = ib_copy_to_udata(udata, &offset, sizeof(offset)); + if (err) { + ret = ERR_PTR(err); + goto bail_rwq; + } + + if (qp->r_rq.wq) { + /* Allocate info for ipath_mmap(). */ + ip = kmalloc(sizeof(*ip), GFP_KERNEL); + if (!ip) { + ret = ERR_PTR(-ENOMEM); + goto bail_rwq; + } + qp->ip = ip; + ip->context = ibpd->uobject->context; + ip->obj = qp->r_rq.wq; + kref_init(&ip->ref); + ip->mmap_cnt = 0; + ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) + + qp->r_rq.size * sz); + spin_lock_irq(&dev->pending_lock); + ip->next = dev->pending_mmaps; + dev->pending_mmaps = ip; + spin_unlock_irq(&dev->pending_lock); + } + } + ret = &qp->ibqp; + goto bail; +bail_rwq: + vfree(qp->r_rq.wq); +bail_qp: + kfree(qp); +bail_swq: + vfree(swq); bail: return ret; } @@ -773,11 +840,9 @@ int ipath_destroy_qp(struct ib_qp *ibqp) if (qp->ibqp.qp_type == IB_QPT_SMI) ipath_layer_set_verbs_flags(dev->dd, 0); - spin_lock_irqsave(&qp->r_rq.lock, flags); - spin_lock(&qp->s_lock); + spin_lock_irqsave(&qp->s_lock, flags); qp->state = IB_QPS_ERR; - spin_unlock(&qp->s_lock); - spin_unlock_irqrestore(&qp->r_rq.lock, flags); + spin_unlock_irqrestore(&qp->s_lock, flags); /* Stop the sending tasklet. */ tasklet_kill(&qp->s_task); @@ -798,8 +863,11 @@ int ipath_destroy_qp(struct ib_qp *ibqp) if (atomic_read(&qp->refcount) != 0) ipath_free_qp(&dev->qp_table, qp); + if (qp->ip) + kref_put(&qp->ip->ref, ipath_release_mmap_info); + else + vfree(qp->r_rq.wq); vfree(qp->s_wq); - vfree(qp->r_rq.wq); kfree(qp); return 0; } diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c index 772bc59fb85..dd09420d677 100644 --- a/drivers/infiniband/hw/ipath/ipath_ruc.c +++ b/drivers/infiniband/hw/ipath/ipath_ruc.c @@ -32,7 +32,7 @@ */ #include "ipath_verbs.h" -#include "ipath_common.h" +#include "ipath_kernel.h" /* * Convert the AETH RNR timeout code into the number of milliseconds. @@ -106,6 +106,54 @@ void ipath_insert_rnr_queue(struct ipath_qp *qp) spin_unlock_irqrestore(&dev->pending_lock, flags); } +static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe) +{ + struct ipath_ibdev *dev = to_idev(qp->ibqp.device); + int user = to_ipd(qp->ibqp.pd)->user; + int i, j, ret; + struct ib_wc wc; + + qp->r_len = 0; + for (i = j = 0; i < wqe->num_sge; i++) { + if (wqe->sg_list[i].length == 0) + continue; + /* Check LKEY */ + if ((user && wqe->sg_list[i].lkey == 0) || + !ipath_lkey_ok(&dev->lk_table, + &qp->r_sg_list[j], &wqe->sg_list[i], + IB_ACCESS_LOCAL_WRITE)) + goto bad_lkey; + qp->r_len += wqe->sg_list[i].length; + j++; + } + qp->r_sge.sge = qp->r_sg_list[0]; + qp->r_sge.sg_list = qp->r_sg_list + 1; + qp->r_sge.num_sge = j; + ret = 1; + goto bail; + +bad_lkey: + wc.wr_id = wqe->wr_id; + wc.status = IB_WC_LOC_PROT_ERR; + wc.opcode = IB_WC_RECV; + wc.vendor_err = 0; + wc.byte_len = 0; + wc.imm_data = 0; + wc.qp_num = qp->ibqp.qp_num; + wc.src_qp = 0; + wc.wc_flags = 0; + wc.pkey_index = 0; + wc.slid = 0; + wc.sl = 0; + wc.dlid_path_bits = 0; + wc.port_num = 0; + /* Signal solicited completion event. */ + ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1); + ret = 0; +bail: + return ret; +} + /** * ipath_get_rwqe - copy the next RWQE into the QP's RWQE * @qp: the QP @@ -119,71 +167,71 @@ int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only) { unsigned long flags; struct ipath_rq *rq; + struct ipath_rwq *wq; struct ipath_srq *srq; struct ipath_rwqe *wqe; - int ret = 1; + void (*handler)(struct ib_event *, void *); + u32 tail; + int ret; - if (!qp->ibqp.srq) { + if (qp->ibqp.srq) { + srq = to_isrq(qp->ibqp.srq); + handler = srq->ibsrq.event_handler; + rq = &srq->rq; + } else { + srq = NULL; + handler = NULL; rq = &qp->r_rq; - spin_lock_irqsave(&rq->lock, flags); - - if (unlikely(rq->tail == rq->head)) { - ret = 0; - goto done; - } - wqe = get_rwqe_ptr(rq, rq->tail); - qp->r_wr_id = wqe->wr_id; - if (!wr_id_only) { - qp->r_sge.sge = wqe->sg_list[0]; - qp->r_sge.sg_list = wqe->sg_list + 1; - qp->r_sge.num_sge = wqe->num_sge; - qp->r_len = wqe->length; - } - if (++rq->tail >= rq->size) - rq->tail = 0; - goto done; } - srq = to_isrq(qp->ibqp.srq); - rq = &srq->rq; spin_lock_irqsave(&rq->lock, flags); - - if (unlikely(rq->tail == rq->head)) { - ret = 0; - goto done; - } - wqe = get_rwqe_ptr(rq, rq->tail); + wq = rq->wq; + tail = wq->tail; + /* Validate tail before using it since it is user writable. */ + if (tail >= rq->size) + tail = 0; + do { + if (unlikely(tail == wq->head)) { + spin_unlock_irqrestore(&rq->lock, flags); + ret = 0; + goto bail; + } + wqe = get_rwqe_ptr(rq, tail); + if (++tail >= rq->size) + tail = 0; + } while (!wr_id_only && !init_sge(qp, wqe)); qp->r_wr_id = wqe->wr_id; - if (!wr_id_only) { - qp->r_sge.sge = wqe->sg_list[0]; - qp->r_sge.sg_list = wqe->sg_list + 1; - qp->r_sge.num_sge = wqe->num_sge; - qp->r_len = wqe->length; - } - if (++rq->tail >= rq->size) - rq->tail = 0; - if (srq->ibsrq.event_handler) { - struct ib_event ev; + wq->tail = tail; + + ret = 1; + if (handler) { u32 n; - if (rq->head < rq->tail) - n = rq->size + rq->head - rq->tail; + /* + * validate head pointer value and compute + * the number of remaining WQEs. + */ + n = wq->head; + if (n >= rq->size) + n = 0; + if (n < tail) + n += rq->size - tail; else - n = rq->head - rq->tail; + n -= tail; if (n < srq->limit) { + struct ib_event ev; + srq->limit = 0; spin_unlock_irqrestore(&rq->lock, flags); ev.device = qp->ibqp.device; ev.element.srq = qp->ibqp.srq; ev.event = IB_EVENT_SRQ_LIMIT_REACHED; - srq->ibsrq.event_handler(&ev, - srq->ibsrq.srq_context); + handler(&ev, srq->ibsrq.srq_context); goto bail; } } - -done: spin_unlock_irqrestore(&rq->lock, flags); + bail: return ret; } diff --git a/drivers/infiniband/hw/ipath/ipath_srq.c b/drivers/infiniband/hw/ipath/ipath_srq.c index fa77da6667e..941e866d951 100644 --- a/drivers/infiniband/hw/ipath/ipath_srq.c +++ b/drivers/infiniband/hw/ipath/ipath_srq.c @@ -48,66 +48,39 @@ int ipath_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr, struct ib_recv_wr **bad_wr) { struct ipath_srq *srq = to_isrq(ibsrq); - struct ipath_ibdev *dev = to_idev(ibsrq->device); + struct ipath_rwq *wq; unsigned long flags; int ret; for (; wr; wr = wr->next) { struct ipath_rwqe *wqe; u32 next; - int i, j; + int i; - if (wr->num_sge > srq->rq.max_sge) { + if ((unsigned) wr->num_sge > srq->rq.max_sge) { *bad_wr = wr; ret = -ENOMEM; goto bail; } spin_lock_irqsave(&srq->rq.lock, flags); - next = srq->rq.head + 1; + wq = srq->rq.wq; + next = wq->head + 1; if (next >= srq->rq.size) next = 0; - if (next == srq->rq.tail) { + if (next == wq->tail) { spin_unlock_irqrestore(&srq->rq.lock, flags); *bad_wr = wr; ret = -ENOMEM; goto bail; } - wqe = get_rwqe_ptr(&srq->rq, srq->rq.head); + wqe = get_rwqe_ptr(&srq->rq, wq->head); wqe->wr_id = wr->wr_id; - wqe->sg_list[0].mr = NULL; - wqe->sg_list[0].vaddr = NULL; - wqe->sg_list[0].length = 0; - wqe->sg_list[0].sge_length = 0; - wqe->length = 0; - for (i = 0, j = 0; i < wr->num_sge; i++) { - /* Check LKEY */ - if (to_ipd(srq->ibsrq.pd)->user && - wr->sg_list[i].lkey == 0) { - spin_unlock_irqrestore(&srq->rq.lock, - flags); - *bad_wr = wr; - ret = -EINVAL; - goto bail; - } - if (wr->sg_list[i].length == 0) - continue; - if (!ipath_lkey_ok(&dev->lk_table, - &wqe->sg_list[j], - &wr->sg_list[i], - IB_ACCESS_LOCAL_WRITE)) { - spin_unlock_irqrestore(&srq->rq.lock, - flags); - *bad_wr = wr; - ret = -EINVAL; - goto bail; - } - wqe->length += wr->sg_list[i].length; - j++; - } - wqe->num_sge = j; - srq->rq.head = next; + wqe->num_sge = wr->num_sge; + for (i = 0; i < wr->num_sge; i++) + wqe->sg_list[i] = wr->sg_list[i]; + wq->head = next; spin_unlock_irqrestore(&srq->rq.lock, flags); } ret = 0; @@ -133,53 +106,95 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd, if (dev->n_srqs_allocated == ib_ipath_max_srqs) { ret = ERR_PTR(-ENOMEM); - goto bail; + goto done; } if (srq_init_attr->attr.max_wr == 0) { ret = ERR_PTR(-EINVAL); - goto bail; + goto done; } if ((srq_init_attr->attr.max_sge > ib_ipath_max_srq_sges) || (srq_init_attr->attr.max_wr > ib_ipath_max_srq_wrs)) { ret = ERR_PTR(-EINVAL); - goto bail; + goto done; } srq = kmalloc(sizeof(*srq), GFP_KERNEL); if (!srq) { ret = ERR_PTR(-ENOMEM); - goto bail; + goto done; } /* * Need to use vmalloc() if we want to support large #s of entries. */ srq->rq.size = srq_init_attr->attr.max_wr + 1; - sz = sizeof(struct ipath_sge) * srq_init_attr->attr.max_sge + + srq->rq.max_sge = srq_init_attr->attr.max_sge; + sz = sizeof(struct ib_sge) * srq->rq.max_sge + sizeof(struct ipath_rwqe); - srq->rq.wq = vmalloc(srq->rq.size * sz); + srq->rq.wq = vmalloc_user(sizeof(struct ipath_rwq) + srq->rq.size * sz); if (!srq->rq.wq) { - kfree(srq); ret = ERR_PTR(-ENOMEM); - goto bail; + goto bail_srq; } + /* + * Return the address of the RWQ as the offset to mmap. + * See ipath_mmap() for details. + */ + if (udata && udata->outlen >= sizeof(__u64)) { + struct ipath_mmap_info *ip; + __u64 offset = (__u64) srq->rq.wq; + int err; + + err = ib_copy_to_udata(udata, &offset, sizeof(offset)); + if (err) { + ret = ERR_PTR(err); + goto bail_wq; + } + + /* Allocate info for ipath_mmap(). */ + ip = kmalloc(sizeof(*ip), GFP_KERNEL); + if (!ip) { + ret = ERR_PTR(-ENOMEM); + goto bail_wq; + } + srq->ip = ip; + ip->context = ibpd->uobject->context; + ip->obj = srq->rq.wq; + kref_init(&ip->ref); + ip->mmap_cnt = 0; + ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) + + srq->rq.size * sz); + spin_lock_irq(&dev->pending_lock); + ip->next = dev->pending_mmaps; + dev->pending_mmaps = ip; + spin_unlock_irq(&dev->pending_lock); + } else + srq->ip = NULL; + /* * ib_create_srq() will initialize srq->ibsrq. */ spin_lock_init(&srq->rq.lock); - srq->rq.head = 0; - srq->rq.tail = 0; + srq->rq.wq->head = 0; + srq->rq.wq->tail = 0; srq->rq.max_sge = srq_init_attr->attr.max_sge; srq->limit = srq_init_attr->attr.srq_limit; + dev->n_srqs_allocated++; + ret = &srq->ibsrq; + goto done; - dev->n_srqs_allocated++; +bail_wq: + vfree(srq->rq.wq); -bail: +bail_srq: + kfree(srq); + +done: return ret; } @@ -195,78 +210,123 @@ int ipath_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, struct ib_udata *udata) { struct ipath_srq *srq = to_isrq(ibsrq); - unsigned long flags; - int ret; + int ret = 0; - if (attr_mask & IB_SRQ_MAX_WR) - if ((attr->max_wr > ib_ipath_max_srq_wrs) || - (attr->max_sge > srq->rq.max_sge)) { - ret = -EINVAL; - goto bail; - } + if (attr_mask & IB_SRQ_MAX_WR) { + struct ipath_rwq *owq; + struct ipath_rwq *wq; + struct ipath_rwqe *p; + u32 sz, size, n, head, tail; - if (attr_mask & IB_SRQ_LIMIT) - if (attr->srq_limit >= srq->rq.size) { + /* Check that the requested sizes are below the limits. */ + if ((attr->max_wr > ib_ipath_max_srq_wrs) || + ((attr_mask & IB_SRQ_LIMIT) ? + attr->srq_limit : srq->limit) > attr->max_wr) { ret = -EINVAL; goto bail; } - if (attr_mask & IB_SRQ_MAX_WR) { - struct ipath_rwqe *wq, *p; - u32 sz, size, n; - sz = sizeof(struct ipath_rwqe) + - attr->max_sge * sizeof(struct ipath_sge); + srq->rq.max_sge * sizeof(struct ib_sge); size = attr->max_wr + 1; - wq = vmalloc(size * sz); + wq = vmalloc_user(sizeof(struct ipath_rwq) + size * sz); if (!wq) { ret = -ENOMEM; goto bail; } - spin_lock_irqsave(&srq->rq.lock, flags); - if (srq->rq.head < srq->rq.tail) - n = srq->rq.size + srq->rq.head - srq->rq.tail; + /* + * Return the address of the RWQ as the offset to mmap. + * See ipath_mmap() for details. + */ + if (udata && udata->inlen >= sizeof(__u64)) { + __u64 offset_addr; + __u64 offset = (__u64) wq; + + ret = ib_copy_from_udata(&offset_addr, udata, + sizeof(offset_addr)); + if (ret) { + vfree(wq); + goto bail; + } + udata->outbuf = (void __user *) offset_addr; + ret = ib_copy_to_udata(udata, &offset, + sizeof(offset)); + if (ret) { + vfree(wq); + goto bail; + } + } + + spin_lock_irq(&srq->rq.lock); + /* + * validate head pointer value and compute + * the number of remaining WQEs. + */ + owq = srq->rq.wq; + head = owq->head; + if (head >= srq->rq.size) + head = 0; + tail = owq->tail; + if (tail >= srq->rq.size) + tail = 0; + n = head; + if (n < tail) + n += srq->rq.size - tail; else - n = srq->rq.head - srq->rq.tail; - if (size <= n || size <= srq->limit) { - spin_unlock_irqrestore(&srq->rq.lock, flags); + n -= tail; + if (size <= n) { + spin_unlock_irq(&srq->rq.lock); vfree(wq); ret = -EINVAL; goto bail; } n = 0; - p = wq; - while (srq->rq.tail != srq->rq.head) { + p = wq->wq; + while (tail != head) { struct ipath_rwqe *wqe; int i; - wqe = get_rwqe_ptr(&srq->rq, srq->rq.tail); + wqe = get_rwqe_ptr(&srq->rq, tail); p->wr_id = wqe->wr_id; - p->length = wqe->length; p->num_sge = wqe->num_sge; for (i = 0; i < wqe->num_sge; i++) p->sg_list[i] = wqe->sg_list[i]; n++; p = (struct ipath_rwqe *)((char *) p + sz); - if (++srq->rq.tail >= srq->rq.size) - srq->rq.tail = 0; + if (++tail >= srq->rq.size) + tail = 0; } - vfree(srq->rq.wq); srq->rq.wq = wq; srq->rq.size = size; - srq->rq.head = n; - srq->rq.tail = 0; - srq->rq.max_sge = attr->max_sge; - spin_unlock_irqrestore(&srq->rq.lock, flags); - } - - if (attr_mask & IB_SRQ_LIMIT) { - spin_lock_irqsave(&srq->rq.lock, flags); - srq->limit = attr->srq_limit; - spin_unlock_irqrestore(&srq->rq.lock, flags); + wq->head = n; + wq->tail = 0; + if (attr_mask & IB_SRQ_LIMIT) + srq->limit = attr->srq_limit; + spin_unlock_irq(&srq->rq.lock); + + vfree(owq); + + if (srq->ip) { + struct ipath_mmap_info *ip = srq->ip; + struct ipath_ibdev *dev = to_idev(srq->ibsrq.device); + + ip->obj = wq; + ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) + + size * sz); + spin_lock_irq(&dev->pending_lock); + ip->next = dev->pending_mmaps; + dev->pending_mmaps = ip; + spin_unlock_irq(&dev->pending_lock); + } + } else if (attr_mask & IB_SRQ_LIMIT) { + spin_lock_irq(&srq->rq.lock); + if (attr->srq_limit >= srq->rq.size) + ret = -EINVAL; + else + srq->limit = attr->srq_limit; + spin_unlock_irq(&srq->rq.lock); } - ret = 0; bail: return ret; diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c index 3466129af80..82439fcfc2f 100644 --- a/drivers/infiniband/hw/ipath/ipath_ud.c +++ b/drivers/infiniband/hw/ipath/ipath_ud.c @@ -34,7 +34,54 @@ #include #include "ipath_verbs.h" -#include "ipath_common.h" +#include "ipath_kernel.h" + +static int init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe, + u32 *lengthp, struct ipath_sge_state *ss) +{ + struct ipath_ibdev *dev = to_idev(qp->ibqp.device); + int user = to_ipd(qp->ibqp.pd)->user; + int i, j, ret; + struct ib_wc wc; + + *lengthp = 0; + for (i = j = 0; i < wqe->num_sge; i++) { + if (wqe->sg_list[i].length == 0) + continue; + /* Check LKEY */ + if ((user && wqe->sg_list[i].lkey == 0) || + !ipath_lkey_ok(&dev->lk_table, + j ? &ss->sg_list[j - 1] : &ss->sge, + &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE)) + goto bad_lkey; + *lengthp += wqe->sg_list[i].length; + j++; + } + ss->num_sge = j; + ret = 1; + goto bail; + +bad_lkey: + wc.wr_id = wqe->wr_id; + wc.status = IB_WC_LOC_PROT_ERR; + wc.opcode = IB_WC_RECV; + wc.vendor_err = 0; + wc.byte_len = 0; + wc.imm_data = 0; + wc.qp_num = qp->ibqp.qp_num; + wc.src_qp = 0; + wc.wc_flags = 0; + wc.pkey_index = 0; + wc.slid = 0; + wc.sl = 0; + wc.dlid_path_bits = 0; + wc.port_num = 0; + /* Signal solicited completion event. */ + ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1); + ret = 0; +bail: + return ret; +} /** * ipath_ud_loopback - handle send on loopback QPs @@ -46,6 +93,8 @@ * * This is called from ipath_post_ud_send() to forward a WQE addressed * to the same HCA. + * Note that the receive interrupt handler may be calling ipath_ud_rcv() + * while this is being called. */ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_sge_state *ss, @@ -60,7 +109,11 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_srq *srq; struct ipath_sge_state rsge; struct ipath_sge *sge; + struct ipath_rwq *wq; struct ipath_rwqe *wqe; + void (*handler)(struct ib_event *, void *); + u32 tail; + u32 rlen; qp = ipath_lookup_qpn(&dev->qp_table, wr->wr.ud.remote_qpn); if (!qp) @@ -94,6 +147,13 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, wc->imm_data = 0; } + if (wr->num_sge > 1) { + rsge.sg_list = kmalloc((wr->num_sge - 1) * + sizeof(struct ipath_sge), + GFP_ATOMIC); + } else + rsge.sg_list = NULL; + /* * Get the next work request entry to find where to put the data. * Note that it is safe to drop the lock after changing rq->tail @@ -101,37 +161,52 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, */ if (qp->ibqp.srq) { srq = to_isrq(qp->ibqp.srq); + handler = srq->ibsrq.event_handler; rq = &srq->rq; } else { srq = NULL; + handler = NULL; rq = &qp->r_rq; } + spin_lock_irqsave(&rq->lock, flags); - if (rq->tail == rq->head) { - spin_unlock_irqrestore(&rq->lock, flags); - dev->n_pkt_drops++; - goto done; + wq = rq->wq; + tail = wq->tail; + while (1) { + if (unlikely(tail == wq->head)) { + spin_unlock_irqrestore(&rq->lock, flags); + dev->n_pkt_drops++; + goto bail_sge; + } + wqe = get_rwqe_ptr(rq, tail); + if (++tail >= rq->size) + tail = 0; + if (init_sge(qp, wqe, &rlen, &rsge)) + break; + wq->tail = tail; } /* Silently drop packets which are too big. */ - wqe = get_rwqe_ptr(rq, rq->tail); - if (wc->byte_len > wqe->length) { + if (wc->byte_len > rlen) { spin_unlock_irqrestore(&rq->lock, flags); dev->n_pkt_drops++; - goto done; + goto bail_sge; } + wq->tail = tail; wc->wr_id = wqe->wr_id; - rsge.sge = wqe->sg_list[0]; - rsge.sg_list = wqe->sg_list + 1; - rsge.num_sge = wqe->num_sge; - if (++rq->tail >= rq->size) - rq->tail = 0; - if (srq && srq->ibsrq.event_handler) { + if (handler) { u32 n; - if (rq->head < rq->tail) - n = rq->size + rq->head - rq->tail; + /* + * validate head pointer value and compute + * the number of remaining WQEs. + */ + n = wq->head; + if (n >= rq->size) + n = 0; + if (n < tail) + n += rq->size - tail; else - n = rq->head - rq->tail; + n -= tail; if (n < srq->limit) { struct ib_event ev; @@ -140,12 +215,12 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, ev.device = qp->ibqp.device; ev.element.srq = qp->ibqp.srq; ev.event = IB_EVENT_SRQ_LIMIT_REACHED; - srq->ibsrq.event_handler(&ev, - srq->ibsrq.srq_context); + handler(&ev, srq->ibsrq.srq_context); } else spin_unlock_irqrestore(&rq->lock, flags); } else spin_unlock_irqrestore(&rq->lock, flags); + ah_attr = &to_iah(wr->wr.ud.ah)->attr; if (ah_attr->ah_flags & IB_AH_GRH) { ipath_copy_sge(&rsge, &ah_attr->grh, sizeof(struct ib_grh)); @@ -186,7 +261,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, wc->src_qp = sqp->ibqp.qp_num; /* XXX do we know which pkey matched? Only needed for GSI. */ wc->pkey_index = 0; - wc->slid = ipath_layer_get_lid(dev->dd) | + wc->slid = dev->dd->ipath_lid | (ah_attr->src_path_bits & ((1 << (dev->mkeyprot_resv_lmc & 7)) - 1)); wc->sl = ah_attr->sl; @@ -196,6 +271,8 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, ipath_cq_enter(to_icq(qp->ibqp.recv_cq), wc, wr->send_flags & IB_SEND_SOLICITED); +bail_sge: + kfree(rsge.sg_list); done: if (atomic_dec_and_test(&qp->refcount)) wake_up(&qp->wait); @@ -433,13 +510,9 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, int opcode; u32 hdrsize; u32 pad; - unsigned long flags; struct ib_wc wc; u32 qkey; u32 src_qp; - struct ipath_rq *rq; - struct ipath_srq *srq; - struct ipath_rwqe *wqe; u16 dlid; int header_in_data; @@ -547,19 +620,10 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, /* * Get the next work request entry to find where to put the data. - * Note that it is safe to drop the lock after changing rq->tail - * since ipath_post_receive() won't fill the empty slot. */ - if (qp->ibqp.srq) { - srq = to_isrq(qp->ibqp.srq); - rq = &srq->rq; - } else { - srq = NULL; - rq = &qp->r_rq; - } - spin_lock_irqsave(&rq->lock, flags); - if (rq->tail == rq->head) { - spin_unlock_irqrestore(&rq->lock, flags); + if (qp->r_reuse_sge) + qp->r_reuse_sge = 0; + else if (!ipath_get_rwqe(qp, 0)) { /* * Count VL15 packets dropped due to no receive buffer. * Otherwise, count them as buffer overruns since usually, @@ -573,39 +637,11 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, goto bail; } /* Silently drop packets which are too big. */ - wqe = get_rwqe_ptr(rq, rq->tail); - if (wc.byte_len > wqe->length) { - spin_unlock_irqrestore(&rq->lock, flags); + if (wc.byte_len > qp->r_len) { + qp->r_reuse_sge = 1; dev->n_pkt_drops++; goto bail; } - wc.wr_id = wqe->wr_id; - qp->r_sge.sge = wqe->sg_list[0]; - qp->r_sge.sg_list = wqe->sg_list + 1; - qp->r_sge.num_sge = wqe->num_sge; - if (++rq->tail >= rq->size) - rq->tail = 0; - if (srq && srq->ibsrq.event_handler) { - u32 n; - - if (rq->head < rq->tail) - n = rq->size + rq->head - rq->tail; - else - n = rq->head - rq->tail; - if (n < srq->limit) { - struct ib_event ev; - - srq->limit = 0; - spin_unlock_irqrestore(&rq->lock, flags); - ev.device = qp->ibqp.device; - ev.element.srq = qp->ibqp.srq; - ev.event = IB_EVENT_SRQ_LIMIT_REACHED; - srq->ibsrq.event_handler(&ev, - srq->ibsrq.srq_context); - } else - spin_unlock_irqrestore(&rq->lock, flags); - } else - spin_unlock_irqrestore(&rq->lock, flags); if (has_grh) { ipath_copy_sge(&qp->r_sge, &hdr->u.l.grh, sizeof(struct ib_grh)); @@ -614,6 +650,7 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, ipath_skip_sge(&qp->r_sge, sizeof(struct ib_grh)); ipath_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh)); + wc.wr_id = qp->r_wr_id; wc.status = IB_WC_SUCCESS; wc.opcode = IB_WC_RECV; wc.vendor_err = 0; diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index d70a9b6b523..a2b4c70192d 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -277,11 +277,12 @@ static int ipath_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, struct ib_recv_wr **bad_wr) { struct ipath_qp *qp = to_iqp(ibqp); + struct ipath_rwq *wq = qp->r_rq.wq; unsigned long flags; int ret; /* Check that state is OK to post receive. */ - if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_RECV_OK)) { + if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_RECV_OK) || !wq) { *bad_wr = wr; ret = -EINVAL; goto bail; @@ -290,59 +291,31 @@ static int ipath_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, for (; wr; wr = wr->next) { struct ipath_rwqe *wqe; u32 next; - int i, j; + int i; - if (wr->num_sge > qp->r_rq.max_sge) { + if ((unsigned) wr->num_sge > qp->r_rq.max_sge) { *bad_wr = wr; ret = -ENOMEM; goto bail; } spin_lock_irqsave(&qp->r_rq.lock, flags); - next = qp->r_rq.head + 1; + next = wq->head + 1; if (next >= qp->r_rq.size) next = 0; - if (next == qp->r_rq.tail) { + if (next == wq->tail) { spin_unlock_irqrestore(&qp->r_rq.lock, flags); *bad_wr = wr; ret = -ENOMEM; goto bail; } - wqe = get_rwqe_ptr(&qp->r_rq, qp->r_rq.head); + wqe = get_rwqe_ptr(&qp->r_rq, wq->head); wqe->wr_id = wr->wr_id; - wqe->sg_list[0].mr = NULL; - wqe->sg_list[0].vaddr = NULL; - wqe->sg_list[0].length = 0; - wqe->sg_list[0].sge_length = 0; - wqe->length = 0; - for (i = 0, j = 0; i < wr->num_sge; i++) { - /* Check LKEY */ - if (to_ipd(qp->ibqp.pd)->user && - wr->sg_list[i].lkey == 0) { - spin_unlock_irqrestore(&qp->r_rq.lock, - flags); - *bad_wr = wr; - ret = -EINVAL; - goto bail; - } - if (wr->sg_list[i].length == 0) - continue; - if (!ipath_lkey_ok( - &to_idev(qp->ibqp.device)->lk_table, - &wqe->sg_list[j], &wr->sg_list[i], - IB_ACCESS_LOCAL_WRITE)) { - spin_unlock_irqrestore(&qp->r_rq.lock, - flags); - *bad_wr = wr; - ret = -EINVAL; - goto bail; - } - wqe->length += wr->sg_list[i].length; - j++; - } - wqe->num_sge = j; - qp->r_rq.head = next; + wqe->num_sge = wr->num_sge; + for (i = 0; i < wr->num_sge; i++) + wqe->sg_list[i] = wr->sg_list[i]; + wq->head = next; spin_unlock_irqrestore(&qp->r_rq.lock, flags); } ret = 0; @@ -1137,6 +1110,7 @@ static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd) dev->attach_mcast = ipath_multicast_attach; dev->detach_mcast = ipath_multicast_detach; dev->process_mad = ipath_process_mad; + dev->mmap = ipath_mmap; snprintf(dev->node_desc, sizeof(dev->node_desc), IPATH_IDSTR " %s kernel_SMA", system_utsname.nodename); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 698396778f0..7d2ba72609f 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -38,6 +38,7 @@ #include #include #include +#include #include #include "ipath_layer.h" @@ -50,7 +51,7 @@ * Increment this value if any changes that break userspace ABI * compatibility are made. */ -#define IPATH_UVERBS_ABI_VERSION 1 +#define IPATH_UVERBS_ABI_VERSION 2 /* * Define an ib_cq_notify value that is not valid so we know when CQ @@ -178,58 +179,41 @@ struct ipath_ah { }; /* - * Quick description of our CQ/QP locking scheme: - * - * We have one global lock that protects dev->cq/qp_table. Each - * struct ipath_cq/qp also has its own lock. An individual qp lock - * may be taken inside of an individual cq lock. Both cqs attached to - * a qp may be locked, with the send cq locked first. No other - * nesting should be done. - * - * Each struct ipath_cq/qp also has an atomic_t ref count. The - * pointer from the cq/qp_table to the struct counts as one reference. - * This reference also is good for access through the consumer API, so - * modifying the CQ/QP etc doesn't need to take another reference. - * Access because of a completion being polled does need a reference. - * - * Finally, each struct ipath_cq/qp has a wait_queue_head_t for the - * destroy function to sleep on. - * - * This means that access from the consumer API requires nothing but - * taking the struct's lock. - * - * Access because of a completion event should go as follows: - * - lock cq/qp_table and look up struct - * - increment ref count in struct - * - drop cq/qp_table lock - * - lock struct, do your thing, and unlock struct - * - decrement ref count; if zero, wake up waiters - * - * To destroy a CQ/QP, we can do the following: - * - lock cq/qp_table, remove pointer, unlock cq/qp_table lock - * - decrement ref count - * - wait_event until ref count is zero - * - * It is the consumer's responsibilty to make sure that no QP - * operations (WQE posting or state modification) are pending when the - * QP is destroyed. Also, the consumer must make sure that calls to - * qp_modify are serialized. - * - * Possible optimizations (wait for profile data to see if/where we - * have locks bouncing between CPUs): - * - split cq/qp table lock into n separate (cache-aligned) locks, - * indexed (say) by the page in the table + * This structure is used by ipath_mmap() to validate an offset + * when an mmap() request is made. The vm_area_struct then uses + * this as its vm_private_data. */ +struct ipath_mmap_info { + struct ipath_mmap_info *next; + struct ib_ucontext *context; + void *obj; + struct kref ref; + unsigned size; + unsigned mmap_cnt; +}; +/* + * This structure is used to contain the head pointer, tail pointer, + * and completion queue entries as a single memory allocation so + * it can be mmap'ed into user space. + */ +struct ipath_cq_wc { + u32 head; /* index of next entry to fill */ + u32 tail; /* index of next ib_poll_cq() entry */ + struct ib_wc queue[1]; /* this is actually size ibcq.cqe + 1 */ +}; + +/* + * The completion queue structure. + */ struct ipath_cq { struct ib_cq ibcq; struct tasklet_struct comptask; spinlock_t lock; u8 notify; u8 triggered; - u32 head; /* new records added to the head */ - u32 tail; /* poll_cq() reads from here. */ - struct ib_wc *queue; /* this is actually ibcq.cqe + 1 */ + struct ipath_cq_wc *queue; + struct ipath_mmap_info *ip; }; /* @@ -248,28 +232,40 @@ struct ipath_swqe { /* * Receive work request queue entry. - * The size of the sg_list is determined when the QP is created and stored - * in qp->r_max_sge. + * The size of the sg_list is determined when the QP (or SRQ) is created + * and stored in qp->r_rq.max_sge (or srq->rq.max_sge). */ struct ipath_rwqe { u64 wr_id; - u32 length; /* total length of data in sg_list */ u8 num_sge; - struct ipath_sge sg_list[0]; + struct ib_sge sg_list[0]; }; -struct ipath_rq { - spinlock_t lock; +/* + * This structure is used to contain the head pointer, tail pointer, + * and receive work queue entries as a single memory allocation so + * it can be mmap'ed into user space. + * Note that the wq array elements are variable size so you can't + * just index into the array to get the N'th element; + * use get_rwqe_ptr() instead. + */ +struct ipath_rwq { u32 head; /* new work requests posted to the head */ u32 tail; /* receives pull requests from here. */ + struct ipath_rwqe wq[0]; +}; + +struct ipath_rq { + struct ipath_rwq *wq; + spinlock_t lock; u32 size; /* size of RWQE array */ u8 max_sge; - struct ipath_rwqe *wq; /* RWQE array */ }; struct ipath_srq { struct ib_srq ibsrq; struct ipath_rq rq; + struct ipath_mmap_info *ip; /* send signal when number of RWQEs < limit */ u32 limit; }; @@ -293,6 +289,7 @@ struct ipath_qp { atomic_t refcount; wait_queue_head_t wait; struct tasklet_struct s_task; + struct ipath_mmap_info *ip; struct ipath_sge_state *s_cur_sge; struct ipath_sge_state s_sge; /* current send request data */ /* current RDMA read send data */ @@ -345,7 +342,8 @@ struct ipath_qp { u32 s_ssn; /* SSN of tail entry */ u32 s_lsn; /* limit sequence number (credit) */ struct ipath_swqe *s_wq; /* send work queue */ - struct ipath_rq r_rq; /* receive work queue */ + struct ipath_rq r_rq; /* receive work queue */ + struct ipath_sge r_sg_list[0]; /* verified SGEs */ }; /* @@ -369,15 +367,15 @@ static inline struct ipath_swqe *get_swqe_ptr(struct ipath_qp *qp, /* * Since struct ipath_rwqe is not a fixed size, we can't simply index into - * struct ipath_rq.wq. This function does the array index computation. + * struct ipath_rwq.wq. This function does the array index computation. */ static inline struct ipath_rwqe *get_rwqe_ptr(struct ipath_rq *rq, unsigned n) { return (struct ipath_rwqe *) - ((char *) rq->wq + + ((char *) rq->wq->wq + (sizeof(struct ipath_rwqe) + - rq->max_sge * sizeof(struct ipath_sge)) * n); + rq->max_sge * sizeof(struct ib_sge)) * n); } /* @@ -417,6 +415,7 @@ struct ipath_ibdev { struct ib_device ibdev; struct list_head dev_list; struct ipath_devdata *dd; + struct ipath_mmap_info *pending_mmaps; int ib_unit; /* This is the device number */ u16 sm_lid; /* in host order */ u8 sm_sl; @@ -681,6 +680,10 @@ int ipath_unmap_fmr(struct list_head *fmr_list); int ipath_dealloc_fmr(struct ib_fmr *ibfmr); +void ipath_release_mmap_info(struct kref *ref); + +int ipath_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); + void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev); void ipath_insert_rnr_queue(struct ipath_qp *qp); -- cgit v1.2.3-70-g09d2 From c27fef26271d352b5546c33239edeb0dcb4fc0cc Mon Sep 17 00:00:00 2001 From: Bryan O'Sullivan Date: Fri, 25 Aug 2006 11:24:27 -0700 Subject: IB/ipath: lock resource limit counters correctly Signed-off-by: Bryan O'Sullivan Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_verbs.c | 34 +++++++++++++++++++------ drivers/infiniband/hw/ipath/ipath_verbs.h | 7 +++++ drivers/infiniband/hw/ipath/ipath_verbs_mcast.c | 5 ++++ 3 files changed, 38 insertions(+), 8 deletions(-) (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index a2b4c70192d..5b8ee65c6cd 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -776,18 +776,22 @@ static struct ib_pd *ipath_alloc_pd(struct ib_device *ibdev, * we allow allocations of more than we report for this value. */ - if (dev->n_pds_allocated == ib_ipath_max_pds) { + pd = kmalloc(sizeof *pd, GFP_KERNEL); + if (!pd) { ret = ERR_PTR(-ENOMEM); goto bail; } - pd = kmalloc(sizeof *pd, GFP_KERNEL); - if (!pd) { + spin_lock(&dev->n_pds_lock); + if (dev->n_pds_allocated == ib_ipath_max_pds) { + spin_unlock(&dev->n_pds_lock); + kfree(pd); ret = ERR_PTR(-ENOMEM); goto bail; } dev->n_pds_allocated++; + spin_unlock(&dev->n_pds_lock); /* ib_alloc_pd() will initialize pd->ibpd. */ pd->user = udata != NULL; @@ -803,7 +807,9 @@ static int ipath_dealloc_pd(struct ib_pd *ibpd) struct ipath_pd *pd = to_ipd(ibpd); struct ipath_ibdev *dev = to_idev(ibpd->device); + spin_lock(&dev->n_pds_lock); dev->n_pds_allocated--; + spin_unlock(&dev->n_pds_lock); kfree(pd); @@ -824,11 +830,6 @@ static struct ib_ah *ipath_create_ah(struct ib_pd *pd, struct ib_ah *ret; struct ipath_ibdev *dev = to_idev(pd->device); - if (dev->n_ahs_allocated == ib_ipath_max_ahs) { - ret = ERR_PTR(-ENOMEM); - goto bail; - } - /* A multicast address requires a GRH (see ch. 8.4.1). */ if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE && ah_attr->dlid != IPATH_PERMISSIVE_LID && @@ -854,7 +855,16 @@ static struct ib_ah *ipath_create_ah(struct ib_pd *pd, goto bail; } + spin_lock(&dev->n_ahs_lock); + if (dev->n_ahs_allocated == ib_ipath_max_ahs) { + spin_unlock(&dev->n_ahs_lock); + kfree(ah); + ret = ERR_PTR(-ENOMEM); + goto bail; + } + dev->n_ahs_allocated++; + spin_unlock(&dev->n_ahs_lock); /* ib_create_ah() will initialize ah->ibah. */ ah->attr = *ah_attr; @@ -876,7 +886,9 @@ static int ipath_destroy_ah(struct ib_ah *ibah) struct ipath_ibdev *dev = to_idev(ibah->device); struct ipath_ah *ah = to_iah(ibah); + spin_lock(&dev->n_ahs_lock); dev->n_ahs_allocated--; + spin_unlock(&dev->n_ahs_lock); kfree(ah); @@ -963,6 +975,12 @@ static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd) dev = &idev->ibdev; /* Only need to initialize non-zero fields. */ + spin_lock_init(&idev->n_pds_lock); + spin_lock_init(&idev->n_ahs_lock); + spin_lock_init(&idev->n_cqs_lock); + spin_lock_init(&idev->n_srqs_lock); + spin_lock_init(&idev->n_mcast_grps_lock); + spin_lock_init(&idev->qp_table.lock); spin_lock_init(&idev->lk_table.lock); idev->sm_lid = __constant_be16_to_cpu(IB_LID_PERMISSIVE); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 7d2ba72609f..a9baa910143 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -434,11 +434,18 @@ struct ipath_ibdev { __be64 sys_image_guid; /* in network order */ __be64 gid_prefix; /* in network order */ __be64 mkey; + u32 n_pds_allocated; /* number of PDs allocated for device */ + spinlock_t n_pds_lock; u32 n_ahs_allocated; /* number of AHs allocated for device */ + spinlock_t n_ahs_lock; u32 n_cqs_allocated; /* number of CQs allocated for device */ + spinlock_t n_cqs_lock; u32 n_srqs_allocated; /* number of SRQs allocated for device */ + spinlock_t n_srqs_lock; u32 n_mcast_grps_allocated; /* number of mcast groups allocated */ + spinlock_t n_mcast_grps_lock; + u64 ipath_sword; /* total dwords sent (sample result) */ u64 ipath_rword; /* total dwords received (sample result) */ u64 ipath_spkts; /* total packets sent (sample result) */ diff --git a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c index ee0e1d96d72..cb35679e4a1 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c @@ -207,12 +207,15 @@ static int ipath_mcast_add(struct ipath_ibdev *dev, goto bail; } + spin_lock(&dev->n_mcast_grps_lock); if (dev->n_mcast_grps_allocated == ib_ipath_max_mcast_grps) { + spin_unlock(&dev->n_mcast_grps_lock); ret = ENOMEM; goto bail; } dev->n_mcast_grps_allocated++; + spin_unlock(&dev->n_mcast_grps_lock); list_add_tail_rcu(&mqp->list, &mcast->qp_list); @@ -343,7 +346,9 @@ int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) atomic_dec(&mcast->refcount); wait_event(mcast->wait, !atomic_read(&mcast->refcount)); ipath_mcast_free(mcast); + spin_lock(&dev->n_mcast_grps_lock); dev->n_mcast_grps_allocated--; + spin_unlock(&dev->n_mcast_grps_lock); } ret = 0; -- cgit v1.2.3-70-g09d2 From b1c1b6a30eac88665a35a207cc5e6233090b9d65 Mon Sep 17 00:00:00 2001 From: Bryan O'Sullivan Date: Fri, 25 Aug 2006 11:24:31 -0700 Subject: IB/ipath: merge ipath_core and ib_ipath drivers There is little point in keeping the two drivers separate, so we are merging them. Signed-off-by: Bryan O'Sullivan Signed-off-by: Roland Dreier --- drivers/infiniband/Makefile | 2 +- drivers/infiniband/hw/ipath/Kconfig | 21 +-- drivers/infiniband/hw/ipath/Makefile | 27 ++-- drivers/infiniband/hw/ipath/ipath_driver.c | 13 +- drivers/infiniband/hw/ipath/ipath_intr.c | 3 +- drivers/infiniband/hw/ipath/ipath_kernel.h | 11 +- drivers/infiniband/hw/ipath/ipath_layer.c | 214 ++--------------------------- drivers/infiniband/hw/ipath/ipath_layer.h | 8 -- drivers/infiniband/hw/ipath/ipath_verbs.c | 45 ++---- drivers/infiniband/hw/ipath/ipath_verbs.h | 10 ++ 10 files changed, 62 insertions(+), 292 deletions(-) (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/Makefile b/drivers/infiniband/Makefile index 893bee0a50b..08cff32d900 100644 --- a/drivers/infiniband/Makefile +++ b/drivers/infiniband/Makefile @@ -1,6 +1,6 @@ obj-$(CONFIG_INFINIBAND) += core/ obj-$(CONFIG_INFINIBAND_MTHCA) += hw/mthca/ -obj-$(CONFIG_IPATH_CORE) += hw/ipath/ +obj-$(CONFIG_INFINIBAND_IPATH) += hw/ipath/ obj-$(CONFIG_INFINIBAND_EHCA) += hw/ehca/ obj-$(CONFIG_INFINIBAND_IPOIB) += ulp/ipoib/ obj-$(CONFIG_INFINIBAND_SRP) += ulp/srp/ diff --git a/drivers/infiniband/hw/ipath/Kconfig b/drivers/infiniband/hw/ipath/Kconfig index 1db9489f1e8..574a678e7fd 100644 --- a/drivers/infiniband/hw/ipath/Kconfig +++ b/drivers/infiniband/hw/ipath/Kconfig @@ -1,16 +1,9 @@ -config IPATH_CORE - tristate "QLogic InfiniPath Driver" - depends on 64BIT && PCI_MSI && NET - ---help--- - This is a low-level driver for QLogic InfiniPath host channel - adapters (HCAs) based on the HT-400 and PE-800 chips. - config INFINIBAND_IPATH - tristate "QLogic InfiniPath Verbs Driver" - depends on IPATH_CORE && INFINIBAND + tristate "QLogic InfiniPath Driver" + depends on PCI_MSI && 64BIT && INFINIBAND ---help--- - This is a driver that provides InfiniBand verbs support for - QLogic InfiniPath host channel adapters (HCAs). This - allows these devices to be used with both kernel upper level - protocols such as IP-over-InfiniBand as well as with userspace - applications (in conjunction with InfiniBand userspace access). + This is a driver for QLogic InfiniPath host channel adapters, + including InfiniBand verbs support. This driver allows these + devices to be used with both kernel upper level protocols such + as IP-over-InfiniBand as well as with userspace applications + (in conjunction with InfiniBand userspace access). diff --git a/drivers/infiniband/hw/ipath/Makefile b/drivers/infiniband/hw/ipath/Makefile index 075e313941f..690dc713e63 100644 --- a/drivers/infiniband/hw/ipath/Makefile +++ b/drivers/infiniband/hw/ipath/Makefile @@ -1,10 +1,10 @@ EXTRA_CFLAGS += -DIPATH_IDSTR='"QLogic kernel.org driver"' \ -DIPATH_KERN_TYPE=0 -obj-$(CONFIG_IPATH_CORE) += ipath_core.o obj-$(CONFIG_INFINIBAND_IPATH) += ib_ipath.o -ipath_core-y := \ +ib_ipath-y := \ + ipath_cq.o \ ipath_diag.o \ ipath_driver.o \ ipath_eeprom.o \ @@ -13,26 +13,23 @@ ipath_core-y := \ ipath_ht400.o \ ipath_init_chip.o \ ipath_intr.o \ - ipath_layer.o \ - ipath_pe800.o \ - ipath_stats.o \ - ipath_sysfs.o \ - ipath_user_pages.o - -ipath_core-$(CONFIG_X86_64) += ipath_wc_x86_64.o -ipath_core-$(CONFIG_PPC64) += ipath_wc_ppc64.o - -ib_ipath-y := \ - ipath_cq.o \ ipath_keys.o \ + ipath_layer.o \ ipath_mad.o \ ipath_mmap.o \ ipath_mr.o \ + ipath_pe800.o \ ipath_qp.o \ ipath_rc.o \ ipath_ruc.o \ ipath_srq.o \ + ipath_stats.o \ + ipath_sysfs.o \ ipath_uc.o \ ipath_ud.o \ - ipath_verbs.o \ - ipath_verbs_mcast.o + ipath_user_pages.o \ + ipath_verbs_mcast.o \ + ipath_verbs.o + +ib_ipath-$(CONFIG_X86_64) += ipath_wc_x86_64.o +ib_ipath-$(CONFIG_PPC64) += ipath_wc_ppc64.o diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c index 6ded914f9eb..9af7406d6a6 100644 --- a/drivers/infiniband/hw/ipath/ipath_driver.c +++ b/drivers/infiniband/hw/ipath/ipath_driver.c @@ -40,6 +40,7 @@ #include "ipath_kernel.h" #include "ipath_layer.h" +#include "ipath_verbs.h" #include "ipath_common.h" static void ipath_update_pio_bufs(struct ipath_devdata *); @@ -51,8 +52,6 @@ const char *ipath_get_unit_name(int unit) return iname; } -EXPORT_SYMBOL_GPL(ipath_get_unit_name); - #define DRIVER_LOAD_MSG "QLogic " IPATH_DRV_NAME " loaded: " #define PFX IPATH_DRV_NAME ": " @@ -510,6 +509,7 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, ipath_user_add(dd); ipath_diag_add(dd); ipath_layer_add(dd); + ipath_register_ib_device(dd); goto bail; @@ -538,6 +538,7 @@ static void __devexit ipath_remove_one(struct pci_dev *pdev) return; dd = pci_get_drvdata(pdev); + ipath_unregister_ib_device(dd->verbs_dev); ipath_layer_remove(dd); ipath_diag_remove(dd); ipath_user_remove(dd); @@ -978,12 +979,8 @@ reloop: if (unlikely(eflags)) ipath_rcv_hdrerr(dd, eflags, l, etail, rc); else if (etype == RCVHQ_RCV_TYPE_NON_KD) { - int ret = __ipath_verbs_rcv(dd, rc + 1, - ebuf, tlen); - if (ret == -ENODEV) - ipath_cdbg(VERBOSE, - "received IB packet, " - "not SMA (QP=%x)\n", qp); + ipath_ib_rcv(dd->verbs_dev, rc + 1, ebuf, + tlen); if (dd->ipath_lli_counter) dd->ipath_lli_counter--; diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c index 280e732660a..ed54f8f2945 100644 --- a/drivers/infiniband/hw/ipath/ipath_intr.c +++ b/drivers/infiniband/hw/ipath/ipath_intr.c @@ -35,6 +35,7 @@ #include "ipath_kernel.h" #include "ipath_layer.h" +#include "ipath_verbs.h" #include "ipath_common.h" /* These are all rcv-related errors which we want to count for stats */ @@ -712,7 +713,7 @@ static void handle_layer_pioavail(struct ipath_devdata *dd) if (ret > 0) goto set; - ret = __ipath_verbs_piobufavail(dd); + ret = ipath_ib_piobufavail(dd->verbs_dev); if (ret > 0) goto set; diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h index e9f374fb641..f1931105adb 100644 --- a/drivers/infiniband/hw/ipath/ipath_kernel.h +++ b/drivers/infiniband/hw/ipath/ipath_kernel.h @@ -132,12 +132,6 @@ struct _ipath_layer { void *l_arg; }; -/* Verbs layer interface */ -struct _verbs_layer { - void *l_arg; - struct timer_list l_timer; -}; - struct ipath_devdata { struct list_head ipath_list; @@ -198,7 +192,8 @@ struct ipath_devdata { void (*ipath_f_setextled)(struct ipath_devdata *, u64, u64); /* fill out chip-specific fields */ int (*ipath_f_get_base_info)(struct ipath_portdata *, void *); - struct _verbs_layer verbs_layer; + struct ipath_ibdev *verbs_dev; + struct timer_list verbs_timer; /* total dwords sent (summed from counter) */ u64 ipath_sword; /* total dwords rcvd (summed from counter) */ @@ -529,8 +524,6 @@ extern int ipath_layer_intr(struct ipath_devdata *, u32); extern int __ipath_layer_rcv(struct ipath_devdata *, void *, struct sk_buff *); extern int __ipath_layer_rcv_lid(struct ipath_devdata *, void *); -extern int __ipath_verbs_piobufavail(struct ipath_devdata *); -extern int __ipath_verbs_rcv(struct ipath_devdata *, void *, void *, u32); void ipath_layer_add(struct ipath_devdata *); void ipath_layer_remove(struct ipath_devdata *); diff --git a/drivers/infiniband/hw/ipath/ipath_layer.c b/drivers/infiniband/hw/ipath/ipath_layer.c index b28c6f81c73..acc32200cc0 100644 --- a/drivers/infiniband/hw/ipath/ipath_layer.c +++ b/drivers/infiniband/hw/ipath/ipath_layer.c @@ -42,26 +42,20 @@ #include "ipath_kernel.h" #include "ipath_layer.h" +#include "ipath_verbs.h" #include "ipath_common.h" /* Acquire before ipath_devs_lock. */ static DEFINE_MUTEX(ipath_layer_mutex); -static int ipath_verbs_registered; - u16 ipath_layer_rcv_opcode; static int (*layer_intr)(void *, u32); static int (*layer_rcv)(void *, void *, struct sk_buff *); static int (*layer_rcv_lid)(void *, void *); -static int (*verbs_piobufavail)(void *); -static void (*verbs_rcv)(void *, void *, void *, u32); static void *(*layer_add_one)(int, struct ipath_devdata *); static void (*layer_remove_one)(void *); -static void *(*verbs_add_one)(int, struct ipath_devdata *); -static void (*verbs_remove_one)(void *); -static void (*verbs_timer_cb)(void *); int __ipath_layer_intr(struct ipath_devdata *dd, u32 arg) { @@ -107,29 +101,6 @@ int __ipath_layer_rcv_lid(struct ipath_devdata *dd, void *hdr) return ret; } -int __ipath_verbs_piobufavail(struct ipath_devdata *dd) -{ - int ret = -ENODEV; - - if (dd->verbs_layer.l_arg && verbs_piobufavail) - ret = verbs_piobufavail(dd->verbs_layer.l_arg); - - return ret; -} - -int __ipath_verbs_rcv(struct ipath_devdata *dd, void *rc, void *ebuf, - u32 tlen) -{ - int ret = -ENODEV; - - if (dd->verbs_layer.l_arg && verbs_rcv) { - verbs_rcv(dd->verbs_layer.l_arg, rc, ebuf, tlen); - ret = 0; - } - - return ret; -} - int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 newstate) { u32 lstate; @@ -212,8 +183,6 @@ bail: return ret; } -EXPORT_SYMBOL_GPL(ipath_layer_set_linkstate); - /** * ipath_layer_set_mtu - set the MTU * @dd: the infinipath device @@ -298,8 +267,6 @@ bail: return ret; } -EXPORT_SYMBOL_GPL(ipath_layer_set_mtu); - int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc) { dd->ipath_lid = arg; @@ -315,8 +282,6 @@ int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc) return 0; } -EXPORT_SYMBOL_GPL(ipath_set_lid); - int ipath_layer_set_guid(struct ipath_devdata *dd, __be64 guid) { /* XXX - need to inform anyone who cares this just happened. */ @@ -324,85 +289,56 @@ int ipath_layer_set_guid(struct ipath_devdata *dd, __be64 guid) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_set_guid); - __be64 ipath_layer_get_guid(struct ipath_devdata *dd) { return dd->ipath_guid; } -EXPORT_SYMBOL_GPL(ipath_layer_get_guid); - -u32 ipath_layer_get_nguid(struct ipath_devdata *dd) -{ - return dd->ipath_nguid; -} - -EXPORT_SYMBOL_GPL(ipath_layer_get_nguid); - u32 ipath_layer_get_majrev(struct ipath_devdata *dd) { return dd->ipath_majrev; } -EXPORT_SYMBOL_GPL(ipath_layer_get_majrev); - u32 ipath_layer_get_minrev(struct ipath_devdata *dd) { return dd->ipath_minrev; } -EXPORT_SYMBOL_GPL(ipath_layer_get_minrev); - u32 ipath_layer_get_pcirev(struct ipath_devdata *dd) { return dd->ipath_pcirev; } -EXPORT_SYMBOL_GPL(ipath_layer_get_pcirev); - u32 ipath_layer_get_flags(struct ipath_devdata *dd) { return dd->ipath_flags; } -EXPORT_SYMBOL_GPL(ipath_layer_get_flags); - struct device *ipath_layer_get_device(struct ipath_devdata *dd) { return &dd->pcidev->dev; } -EXPORT_SYMBOL_GPL(ipath_layer_get_device); - u16 ipath_layer_get_deviceid(struct ipath_devdata *dd) { return dd->ipath_deviceid; } -EXPORT_SYMBOL_GPL(ipath_layer_get_deviceid); - u32 ipath_layer_get_vendorid(struct ipath_devdata *dd) { return dd->ipath_vendorid; } -EXPORT_SYMBOL_GPL(ipath_layer_get_vendorid); - u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd) { return dd->ipath_lastibcstat; } -EXPORT_SYMBOL_GPL(ipath_layer_get_lastibcstat); - u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd) { return dd->ipath_ibmtu; } -EXPORT_SYMBOL_GPL(ipath_layer_get_ibmtu); - void ipath_layer_add(struct ipath_devdata *dd) { mutex_lock(&ipath_layer_mutex); @@ -411,10 +347,6 @@ void ipath_layer_add(struct ipath_devdata *dd) dd->ipath_layer.l_arg = layer_add_one(dd->ipath_unit, dd); - if (verbs_add_one) - dd->verbs_layer.l_arg = - verbs_add_one(dd->ipath_unit, dd); - mutex_unlock(&ipath_layer_mutex); } @@ -427,11 +359,6 @@ void ipath_layer_remove(struct ipath_devdata *dd) dd->ipath_layer.l_arg = NULL; } - if (dd->verbs_layer.l_arg && verbs_remove_one) { - verbs_remove_one(dd->verbs_layer.l_arg); - dd->verbs_layer.l_arg = NULL; - } - mutex_unlock(&ipath_layer_mutex); } @@ -521,95 +448,10 @@ static void __ipath_verbs_timer(unsigned long arg) ipath_kreceive(dd); /* Handle verbs layer timeouts. */ - if (dd->verbs_layer.l_arg && verbs_timer_cb) - verbs_timer_cb(dd->verbs_layer.l_arg); - - mod_timer(&dd->verbs_layer.l_timer, jiffies + 1); -} - -/** - * ipath_verbs_register - verbs layer registration - * @l_piobufavail: callback for when PIO buffers become available - * @l_rcv: callback for receiving a packet - * @l_timer_cb: timer callback - * @ipath_devdata: device data structure is put here - */ -int ipath_verbs_register(void *(*l_add)(int, struct ipath_devdata *), - void (*l_remove)(void *arg), - int (*l_piobufavail) (void *arg), - void (*l_rcv) (void *arg, void *rhdr, - void *data, u32 tlen), - void (*l_timer_cb) (void *arg)) -{ - struct ipath_devdata *dd, *tmp; - unsigned long flags; - - mutex_lock(&ipath_layer_mutex); - - verbs_add_one = l_add; - verbs_remove_one = l_remove; - verbs_piobufavail = l_piobufavail; - verbs_rcv = l_rcv; - verbs_timer_cb = l_timer_cb; - - spin_lock_irqsave(&ipath_devs_lock, flags); - - list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) { - if (!(dd->ipath_flags & IPATH_INITTED)) - continue; - - if (dd->verbs_layer.l_arg) - continue; - - spin_unlock_irqrestore(&ipath_devs_lock, flags); - dd->verbs_layer.l_arg = l_add(dd->ipath_unit, dd); - spin_lock_irqsave(&ipath_devs_lock, flags); - } - - spin_unlock_irqrestore(&ipath_devs_lock, flags); - mutex_unlock(&ipath_layer_mutex); - - ipath_verbs_registered = 1; - - return 0; -} - -EXPORT_SYMBOL_GPL(ipath_verbs_register); - -void ipath_verbs_unregister(void) -{ - struct ipath_devdata *dd, *tmp; - unsigned long flags; - - mutex_lock(&ipath_layer_mutex); - spin_lock_irqsave(&ipath_devs_lock, flags); - - list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) { - *dd->ipath_statusp &= ~IPATH_STATUS_OIB_SMA; - - if (dd->verbs_layer.l_arg && verbs_remove_one) { - spin_unlock_irqrestore(&ipath_devs_lock, flags); - verbs_remove_one(dd->verbs_layer.l_arg); - spin_lock_irqsave(&ipath_devs_lock, flags); - dd->verbs_layer.l_arg = NULL; - } - } - - spin_unlock_irqrestore(&ipath_devs_lock, flags); - - verbs_add_one = NULL; - verbs_remove_one = NULL; - verbs_piobufavail = NULL; - verbs_rcv = NULL; - verbs_timer_cb = NULL; - - ipath_verbs_registered = 0; - - mutex_unlock(&ipath_layer_mutex); + ipath_ib_timer(dd->verbs_dev); + mod_timer(&dd->verbs_timer, jiffies + 1); } -EXPORT_SYMBOL_GPL(ipath_verbs_unregister); - int ipath_layer_open(struct ipath_devdata *dd, u32 * pktmax) { int ret; @@ -703,8 +545,6 @@ u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd) return ipath_read_creg32(dd, dd->ipath_cregs->cr_errpkey); } -EXPORT_SYMBOL_GPL(ipath_layer_get_cr_errpkey); - static void update_sge(struct ipath_sge_state *ss, u32 length) { struct ipath_sge *sge = &ss->sge; @@ -981,8 +821,6 @@ bail: return ret; } -EXPORT_SYMBOL_GPL(ipath_verbs_send); - int ipath_layer_snapshot_counters(struct ipath_devdata *dd, u64 *swords, u64 *rwords, u64 *spkts, u64 *rpkts, u64 *xmit_wait) @@ -1007,8 +845,6 @@ bail: return ret; } -EXPORT_SYMBOL_GPL(ipath_layer_snapshot_counters); - /** * ipath_layer_get_counters - get various chip counters * @dd: the infinipath device @@ -1069,8 +905,6 @@ bail: return ret; } -EXPORT_SYMBOL_GPL(ipath_layer_get_counters); - int ipath_layer_want_buffer(struct ipath_devdata *dd) { set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl); @@ -1080,8 +914,6 @@ int ipath_layer_want_buffer(struct ipath_devdata *dd) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_want_buffer); - int ipath_layer_send_hdr(struct ipath_devdata *dd, struct ether_header *hdr) { int ret = 0; @@ -1174,30 +1006,26 @@ int ipath_layer_enable_timer(struct ipath_devdata *dd) (u64) (1 << 2)); } - init_timer(&dd->verbs_layer.l_timer); - dd->verbs_layer.l_timer.function = __ipath_verbs_timer; - dd->verbs_layer.l_timer.data = (unsigned long)dd; - dd->verbs_layer.l_timer.expires = jiffies + 1; - add_timer(&dd->verbs_layer.l_timer); + init_timer(&dd->verbs_timer); + dd->verbs_timer.function = __ipath_verbs_timer; + dd->verbs_timer.data = (unsigned long)dd; + dd->verbs_timer.expires = jiffies + 1; + add_timer(&dd->verbs_timer); return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_enable_timer); - int ipath_layer_disable_timer(struct ipath_devdata *dd) { /* Disable GPIO bit 2 interrupt */ if (dd->ipath_flags & IPATH_GPIO_INTR) ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 0); - del_timer_sync(&dd->verbs_layer.l_timer); + del_timer_sync(&dd->verbs_timer); return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_disable_timer); - /** * ipath_layer_set_verbs_flags - set the verbs layer flags * @dd: the infinipath device @@ -1225,8 +1053,6 @@ int ipath_layer_set_verbs_flags(struct ipath_devdata *dd, unsigned flags) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_set_verbs_flags); - /** * ipath_layer_get_npkeys - return the size of the PKEY table for port 0 * @dd: the infinipath device @@ -1236,8 +1062,6 @@ unsigned ipath_layer_get_npkeys(struct ipath_devdata *dd) return ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys); } -EXPORT_SYMBOL_GPL(ipath_layer_get_npkeys); - /** * ipath_layer_get_pkey - return the indexed PKEY from the port 0 PKEY table * @dd: the infinipath device @@ -1255,8 +1079,6 @@ unsigned ipath_layer_get_pkey(struct ipath_devdata *dd, unsigned index) return ret; } -EXPORT_SYMBOL_GPL(ipath_layer_get_pkey); - /** * ipath_layer_get_pkeys - return the PKEY table for port 0 * @dd: the infinipath device @@ -1271,8 +1093,6 @@ int ipath_layer_get_pkeys(struct ipath_devdata *dd, u16 * pkeys) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_get_pkeys); - /** * rm_pkey - decrecment the reference count for the given PKEY * @dd: the infinipath device @@ -1419,8 +1239,6 @@ int ipath_layer_set_pkeys(struct ipath_devdata *dd, u16 * pkeys) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_set_pkeys); - /** * ipath_layer_get_linkdowndefaultstate - get the default linkdown state * @dd: the infinipath device @@ -1432,8 +1250,6 @@ int ipath_layer_get_linkdowndefaultstate(struct ipath_devdata *dd) return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE); } -EXPORT_SYMBOL_GPL(ipath_layer_get_linkdowndefaultstate); - /** * ipath_layer_set_linkdowndefaultstate - set the default linkdown state * @dd: the infinipath device @@ -1453,8 +1269,6 @@ int ipath_layer_set_linkdowndefaultstate(struct ipath_devdata *dd, return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_set_linkdowndefaultstate); - int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd) { return (dd->ipath_ibcctrl >> @@ -1462,8 +1276,6 @@ int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd) INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK; } -EXPORT_SYMBOL_GPL(ipath_layer_get_phyerrthreshold); - /** * ipath_layer_set_phyerrthreshold - set the physical error threshold * @dd: the infinipath device @@ -1489,8 +1301,6 @@ int ipath_layer_set_phyerrthreshold(struct ipath_devdata *dd, unsigned n) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_set_phyerrthreshold); - int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd) { return (dd->ipath_ibcctrl >> @@ -1498,8 +1308,6 @@ int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd) INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK; } -EXPORT_SYMBOL_GPL(ipath_layer_get_overrunthreshold); - /** * ipath_layer_set_overrunthreshold - set the overrun threshold * @dd: the infinipath device @@ -1525,17 +1333,13 @@ int ipath_layer_set_overrunthreshold(struct ipath_devdata *dd, unsigned n) return 0; } -EXPORT_SYMBOL_GPL(ipath_layer_set_overrunthreshold); - int ipath_layer_get_boardname(struct ipath_devdata *dd, char *name, size_t namelen) { return dd->ipath_f_get_boardname(dd, name, namelen); } -EXPORT_SYMBOL_GPL(ipath_layer_get_boardname); u32 ipath_layer_get_rcvhdrentsize(struct ipath_devdata *dd) { return dd->ipath_rcvhdrentsize; } -EXPORT_SYMBOL_GPL(ipath_layer_get_rcvhdrentsize); diff --git a/drivers/infiniband/hw/ipath/ipath_layer.h b/drivers/infiniband/hw/ipath/ipath_layer.h index 71485096fca..57c990a5715 100644 --- a/drivers/infiniband/hw/ipath/ipath_layer.h +++ b/drivers/infiniband/hw/ipath/ipath_layer.h @@ -114,14 +114,7 @@ int ipath_layer_register(void *(*l_add)(int, struct ipath_devdata *), struct sk_buff *), u16 rcv_opcode, int (*l_rcv_lid)(void *, void *)); -int ipath_verbs_register(void *(*l_add)(int, struct ipath_devdata *), - void (*l_remove)(void *arg), - int (*l_piobufavail)(void *arg), - void (*l_rcv)(void *arg, void *rhdr, - void *data, u32 tlen), - void (*l_timer_cb)(void *arg)); void ipath_layer_unregister(void); -void ipath_verbs_unregister(void); int ipath_layer_open(struct ipath_devdata *, u32 * pktmax); u16 ipath_layer_get_lid(struct ipath_devdata *dd); int ipath_layer_get_mac(struct ipath_devdata *dd, u8 *); @@ -145,7 +138,6 @@ int ipath_layer_get_counters(struct ipath_devdata *dd, int ipath_layer_want_buffer(struct ipath_devdata *dd); int ipath_layer_set_guid(struct ipath_devdata *, __be64 guid); __be64 ipath_layer_get_guid(struct ipath_devdata *); -u32 ipath_layer_get_nguid(struct ipath_devdata *); u32 ipath_layer_get_majrev(struct ipath_devdata *); u32 ipath_layer_get_minrev(struct ipath_devdata *); u32 ipath_layer_get_pcirev(struct ipath_devdata *); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index 5b8ee65c6cd..15edec9227e 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -368,7 +368,7 @@ static void ipath_qp_rcv(struct ipath_ibdev *dev, } /** - * ipath_ib_rcv - process and incoming packet + * ipath_ib_rcv - process an incoming packet * @arg: the device pointer * @rhdr: the header of the packet * @data: the packet data @@ -377,9 +377,9 @@ static void ipath_qp_rcv(struct ipath_ibdev *dev, * This is called from ipath_kreceive() to process an incoming packet at * interrupt level. Tlen is the length of the header + data + CRC in bytes. */ -static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen) +void ipath_ib_rcv(struct ipath_ibdev *dev, void *rhdr, void *data, + u32 tlen) { - struct ipath_ibdev *dev = (struct ipath_ibdev *) arg; struct ipath_ib_header *hdr = rhdr; struct ipath_other_headers *ohdr; struct ipath_qp *qp; @@ -468,9 +468,8 @@ bail:; * This is called from ipath_do_rcv_timer() at interrupt level to check for * QPs which need retransmits and to collect performance numbers. */ -static void ipath_ib_timer(void *arg) +void ipath_ib_timer(struct ipath_ibdev *dev) { - struct ipath_ibdev *dev = (struct ipath_ibdev *) arg; struct ipath_qp *resend = NULL; struct list_head *last; struct ipath_qp *qp; @@ -564,9 +563,8 @@ static void ipath_ib_timer(void *arg) * QPs waiting for buffers (for now, just do a tasklet_hi_schedule and * return zero). */ -static int ipath_ib_piobufavail(void *arg) +int ipath_ib_piobufavail(struct ipath_ibdev *dev) { - struct ipath_ibdev *dev = (struct ipath_ibdev *) arg; struct ipath_qp *qp; unsigned long flags; @@ -957,11 +955,10 @@ static int ipath_verbs_register_sysfs(struct ib_device *dev); /** * ipath_register_ib_device - register our device with the infiniband core - * @unit: the device number to register * @dd: the device data structure * Return the allocated ipath_ibdev pointer or NULL on error. */ -static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd) +int ipath_register_ib_device(struct ipath_devdata *dd) { struct ipath_layer_counters cntrs; struct ipath_ibdev *idev; @@ -969,8 +966,10 @@ static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd) int ret; idev = (struct ipath_ibdev *)ib_alloc_device(sizeof *idev); - if (idev == NULL) + if (idev == NULL) { + ret = -ENOMEM; goto bail; + } dev = &idev->ibdev; @@ -1047,7 +1046,7 @@ static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd) if (!sys_image_guid) sys_image_guid = ipath_layer_get_guid(dd); idev->sys_image_guid = sys_image_guid; - idev->ib_unit = unit; + idev->ib_unit = dd->ipath_unit; idev->dd = dd; strlcpy(dev->name, "ipath%d", IB_DEVICE_NAME_MAX); @@ -1153,16 +1152,16 @@ err_lk: err_qp: ib_dealloc_device(dev); _VERBS_ERROR("ib_ipath%d cannot register verbs (%d)!\n", - unit, -ret); + dd->ipath_unit, -ret); idev = NULL; bail: - return idev; + dd->verbs_dev = idev; + return ret; } -static void ipath_unregister_ib_device(void *arg) +void ipath_unregister_ib_device(struct ipath_ibdev *dev) { - struct ipath_ibdev *dev = (struct ipath_ibdev *) arg; struct ib_device *ibdev = &dev->ibdev; ipath_layer_disable_timer(dev->dd); @@ -1193,19 +1192,6 @@ static void ipath_unregister_ib_device(void *arg) ib_dealloc_device(ibdev); } -static int __init ipath_verbs_init(void) -{ - return ipath_verbs_register(ipath_register_ib_device, - ipath_unregister_ib_device, - ipath_ib_piobufavail, ipath_ib_rcv, - ipath_ib_timer); -} - -static void __exit ipath_verbs_cleanup(void) -{ - ipath_verbs_unregister(); -} - static ssize_t show_rev(struct class_device *cdev, char *buf) { struct ipath_ibdev *dev = @@ -1297,6 +1283,3 @@ static int ipath_verbs_register_sysfs(struct ib_device *dev) bail: return ret; } - -module_init(ipath_verbs_init); -module_exit(ipath_verbs_cleanup); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index a9baa910143..d6faa4ba606 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -711,6 +711,16 @@ int ipath_make_rc_req(struct ipath_qp *qp, struct ipath_other_headers *ohdr, int ipath_make_uc_req(struct ipath_qp *qp, struct ipath_other_headers *ohdr, u32 pmtu, u32 *bth0p, u32 *bth2p); +int ipath_register_ib_device(struct ipath_devdata *); + +void ipath_unregister_ib_device(struct ipath_ibdev *); + +void ipath_ib_rcv(struct ipath_ibdev *, void *, void *, u32); + +int ipath_ib_piobufavail(struct ipath_ibdev *); + +void ipath_ib_timer(struct ipath_ibdev *); + extern const enum ib_wc_opcode ib_ipath_wc_opcode[]; extern const u8 ipath_cvt_physportstate[]; -- cgit v1.2.3-70-g09d2 From 34b2aafea38efdf02cd8107a6e1057e2a297c447 Mon Sep 17 00:00:00 2001 From: Bryan O'Sullivan Date: Fri, 25 Aug 2006 11:24:32 -0700 Subject: IB/ipath: simplify layering code A lot of ipath layer code was only called in one place. Now that the ipath_core and ib_ipath drivers are merged, it's more sensible to simply inline the simple stuff that the layer code was doing. Signed-off-by: Bryan O'Sullivan Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_diag.c | 1 - drivers/infiniband/hw/ipath/ipath_driver.c | 275 +++++--- drivers/infiniband/hw/ipath/ipath_file_ops.c | 1 - drivers/infiniband/hw/ipath/ipath_intr.c | 7 - drivers/infiniband/hw/ipath/ipath_kernel.h | 15 +- drivers/infiniband/hw/ipath/ipath_layer.c | 978 +-------------------------- drivers/infiniband/hw/ipath/ipath_layer.h | 104 --- drivers/infiniband/hw/ipath/ipath_mad.c | 339 ++++++++-- drivers/infiniband/hw/ipath/ipath_mr.c | 12 + drivers/infiniband/hw/ipath/ipath_qp.c | 34 +- drivers/infiniband/hw/ipath/ipath_rc.c | 9 +- drivers/infiniband/hw/ipath/ipath_ruc.c | 22 +- drivers/infiniband/hw/ipath/ipath_sysfs.c | 6 +- drivers/infiniband/hw/ipath/ipath_uc.c | 5 +- drivers/infiniband/hw/ipath/ipath_ud.c | 13 +- drivers/infiniband/hw/ipath/ipath_verbs.c | 525 ++++++++++++-- drivers/infiniband/hw/ipath/ipath_verbs.h | 109 ++- 17 files changed, 1126 insertions(+), 1329 deletions(-) (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/ipath_diag.c b/drivers/infiniband/hw/ipath/ipath_diag.c index 147dd89e21c..5d77a74aa57 100644 --- a/drivers/infiniband/hw/ipath/ipath_diag.c +++ b/drivers/infiniband/hw/ipath/ipath_diag.c @@ -45,7 +45,6 @@ #include #include "ipath_kernel.h" -#include "ipath_layer.h" #include "ipath_common.h" int ipath_diag_inuse; diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c index 9af7406d6a6..958cc9b33c8 100644 --- a/drivers/infiniband/hw/ipath/ipath_driver.c +++ b/drivers/infiniband/hw/ipath/ipath_driver.c @@ -39,7 +39,6 @@ #include #include "ipath_kernel.h" -#include "ipath_layer.h" #include "ipath_verbs.h" #include "ipath_common.h" @@ -508,7 +507,6 @@ static int __devinit ipath_init_one(struct pci_dev *pdev, ipathfs_add_device(dd); ipath_user_add(dd); ipath_diag_add(dd); - ipath_layer_add(dd); ipath_register_ib_device(dd); goto bail; @@ -539,7 +537,6 @@ static void __devexit ipath_remove_one(struct pci_dev *pdev) dd = pci_get_drvdata(pdev); ipath_unregister_ib_device(dd->verbs_dev); - ipath_layer_remove(dd); ipath_diag_remove(dd); ipath_user_remove(dd); ipathfs_remove_device(dd); @@ -614,11 +611,12 @@ void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first, * * wait up to msecs milliseconds for IB link state change to occur for * now, take the easy polling route. Currently used only by - * ipath_layer_set_linkstate. Returns 0 if state reached, otherwise + * ipath_set_linkstate. Returns 0 if state reached, otherwise * -ETIMEDOUT state can have multiple states set, for any of several * transitions. */ -int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, int msecs) +static int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, + int msecs) { dd->ipath_sma_state_wanted = state; wait_event_interruptible_timeout(ipath_sma_state_wait, @@ -814,58 +812,6 @@ bail: return skb; } -/** - * ipath_rcv_layer - receive a packet for the layered (ethernet) driver - * @dd: the infinipath device - * @etail: the sk_buff number - * @tlen: the total packet length - * @hdr: the ethernet header - * - * Separate routine for better overall optimization - */ -static void ipath_rcv_layer(struct ipath_devdata *dd, u32 etail, - u32 tlen, struct ether_header *hdr) -{ - u32 elen; - u8 pad, *bthbytes; - struct sk_buff *skb, *nskb; - - if (dd->ipath_port0_skbs && - hdr->sub_opcode == IPATH_ITH4X_OPCODE_ENCAP) { - /* - * Allocate a new sk_buff to replace the one we give - * to the network stack. - */ - nskb = ipath_alloc_skb(dd, GFP_ATOMIC); - if (!nskb) { - /* count OK packets that we drop */ - ipath_stats.sps_krdrops++; - return; - } - - bthbytes = (u8 *) hdr->bth; - pad = (bthbytes[1] >> 4) & 3; - /* +CRC32 */ - elen = tlen - (sizeof(*hdr) + pad + sizeof(u32)); - - skb = dd->ipath_port0_skbs[etail]; - dd->ipath_port0_skbs[etail] = nskb; - skb_put(skb, elen); - - dd->ipath_f_put_tid(dd, etail + (u64 __iomem *) - ((char __iomem *) dd->ipath_kregbase - + dd->ipath_rcvegrbase), 0, - virt_to_phys(nskb->data)); - - __ipath_layer_rcv(dd, hdr, skb); - - /* another ether packet received */ - ipath_stats.sps_ether_rpkts++; - } - else if (hdr->sub_opcode == IPATH_ITH4X_OPCODE_LID_ARP) - __ipath_layer_rcv_lid(dd, hdr); -} - static void ipath_rcv_hdrerr(struct ipath_devdata *dd, u32 eflags, u32 l, @@ -979,22 +925,17 @@ reloop: if (unlikely(eflags)) ipath_rcv_hdrerr(dd, eflags, l, etail, rc); else if (etype == RCVHQ_RCV_TYPE_NON_KD) { - ipath_ib_rcv(dd->verbs_dev, rc + 1, ebuf, - tlen); - if (dd->ipath_lli_counter) - dd->ipath_lli_counter--; - - } else if (etype == RCVHQ_RCV_TYPE_EAGER) { - if (qp == IPATH_KD_QP && - bthbytes[0] == ipath_layer_rcv_opcode && - ebuf) - ipath_rcv_layer(dd, etail, tlen, - (struct ether_header *)hdr); - else - ipath_cdbg(PKT, "typ %x, opcode %x (eager, " - "qp=%x), len %x; ignored\n", - etype, bthbytes[0], qp, tlen); + ipath_ib_rcv(dd->verbs_dev, rc + 1, ebuf, tlen); + if (dd->ipath_lli_counter) + dd->ipath_lli_counter--; + 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_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); @@ -1320,13 +1261,6 @@ rescan: goto bail; } - if (updated) - /* - * ran out of bufs, now some (at least this one we just - * got) are now available, so tell the layered driver. - */ - __ipath_layer_intr(dd, IPATH_LAYER_INT_SEND_CONTINUE); - /* * set next starting place. Since it's just an optimization, * it doesn't matter who wins on this, so no locking @@ -1503,7 +1437,7 @@ int ipath_waitfor_mdio_cmdready(struct ipath_devdata *dd) return ret; } -void ipath_set_ib_lstate(struct ipath_devdata *dd, int which) +static void ipath_set_ib_lstate(struct ipath_devdata *dd, int which) { static const char *what[4] = { [0] = "DOWN", @@ -1537,6 +1471,180 @@ void ipath_set_ib_lstate(struct ipath_devdata *dd, int which) dd->ipath_ibcctrl | which); } +int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) +{ + u32 lstate; + int ret; + + switch (newstate) { + case IPATH_IB_LINKDOWN: + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_POLL << + INFINIPATH_IBCC_LINKINITCMD_SHIFT); + /* 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); + /* 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); + /* don't wait */ + ret = 0; + goto bail; + + case IPATH_IB_LINKINIT: + if (dd->ipath_flags & IPATH_LINKINIT) { + ret = 0; + goto bail; + } + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_INIT << + INFINIPATH_IBCC_LINKCMD_SHIFT); + lstate = IPATH_LINKINIT; + break; + + case IPATH_IB_LINKARM: + if (dd->ipath_flags & IPATH_LINKARMED) { + ret = 0; + goto bail; + } + if (!(dd->ipath_flags & + (IPATH_LINKINIT | IPATH_LINKACTIVE))) { + ret = -EINVAL; + goto bail; + } + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED << + INFINIPATH_IBCC_LINKCMD_SHIFT); + /* + * Since the port can transition to ACTIVE by receiving + * a non VL 15 packet, wait for either state. + */ + lstate = IPATH_LINKARMED | IPATH_LINKACTIVE; + break; + + case IPATH_IB_LINKACTIVE: + if (dd->ipath_flags & IPATH_LINKACTIVE) { + ret = 0; + goto bail; + } + if (!(dd->ipath_flags & IPATH_LINKARMED)) { + ret = -EINVAL; + goto bail; + } + ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE << + INFINIPATH_IBCC_LINKCMD_SHIFT); + lstate = IPATH_LINKACTIVE; + break; + + default: + ipath_dbg("Invalid linkstate 0x%x requested\n", newstate); + ret = -EINVAL; + goto bail; + } + ret = ipath_wait_linkstate(dd, lstate, 2000); + +bail: + return ret; +} + +/** + * ipath_set_mtu - set the MTU + * @dd: the infinipath device + * @arg: the new MTU + * + * we can handle "any" incoming size, the issue here is whether we + * need to restrict our outgoing size. For now, we don't do any + * 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... + */ +int ipath_set_mtu(struct ipath_devdata *dd, u16 arg) +{ + u32 piosize; + int changed = 0; + int ret; + + /* + * mtu is IB data payload max. It's the largest power of 2 less + * than piosize (or even larger, since it only really controls the + * largest we can receive; we can send the max of the mtu and + * piosize). We check that it's one of the valid IB sizes. + */ + if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 && + arg != 4096) { + ipath_dbg("Trying to set invalid mtu %u, failing\n", arg); + ret = -EINVAL; + goto bail; + } + if (dd->ipath_ibmtu == arg) { + ret = 0; /* same as current */ + goto bail; + } + + piosize = dd->ipath_ibmaxlen; + dd->ipath_ibmtu = 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) { + dd->ipath_ibmaxlen = piosize; + changed = 1; + } + } else if ((arg + IPATH_PIO_MAXIBHDR) != dd->ipath_ibmaxlen) { + piosize = arg + IPATH_PIO_MAXIBHDR; + ipath_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x " + "(mtu 0x%x)\n", dd->ipath_ibmaxlen, piosize, + arg); + dd->ipath_ibmaxlen = piosize; + changed = 1; + } + + if (changed) { + /* + * set the IBC maxpktlength to the size of our pio + * buffers in words + */ + u64 ibc = dd->ipath_ibcctrl; + 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->ipath_ibcctrl = ibc; + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, + dd->ipath_ibcctrl); + dd->ipath_f_tidtemplate(dd); + } + + ret = 0; + +bail: + return ret; +} + +int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc) +{ + dd->ipath_lid = arg; + dd->ipath_lmc = lmc; + + return 0; +} + /** * ipath_read_kreg64_port - read a device's per-port 64-bit kernel register * @dd: the infinipath device @@ -1640,13 +1748,6 @@ void ipath_shutdown_device(struct ipath_devdata *dd) ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_DISABLE << INFINIPATH_IBCC_LINKINITCMD_SHIFT); - /* - * we are shutting down, so tell the layered driver. We don't do - * this on just a link state change, much like ethernet, a cable - * unplug, etc. doesn't change driver state - */ - ipath_layer_intr(dd, IPATH_LAYER_INT_IF_DOWN); - /* disable IBC */ dd->ipath_control &= ~INFINIPATH_C_LINKENABLE; ipath_write_kreg(dd, dd->ipath_kregs->kr_control, diff --git a/drivers/infiniband/hw/ipath/ipath_file_ops.c b/drivers/infiniband/hw/ipath/ipath_file_ops.c index e999a46bef9..f865ce89b73 100644 --- a/drivers/infiniband/hw/ipath/ipath_file_ops.c +++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c @@ -39,7 +39,6 @@ #include #include "ipath_kernel.h" -#include "ipath_layer.h" #include "ipath_common.h" static int ipath_open(struct inode *, struct file *); diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c index ed54f8f2945..250e2a9f01b 100644 --- a/drivers/infiniband/hw/ipath/ipath_intr.c +++ b/drivers/infiniband/hw/ipath/ipath_intr.c @@ -34,7 +34,6 @@ #include #include "ipath_kernel.h" -#include "ipath_layer.h" #include "ipath_verbs.h" #include "ipath_common.h" @@ -290,8 +289,6 @@ static void handle_e_ibstatuschanged(struct ipath_devdata *dd, *dd->ipath_statusp |= IPATH_STATUS_IB_READY | IPATH_STATUS_IB_CONF; dd->ipath_f_setextled(dd, lstate, ltstate); - - __ipath_layer_intr(dd, IPATH_LAYER_INT_IF_UP); } else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_INIT) { /* * set INIT and DOWN. Down is checked by most of the other @@ -709,10 +706,6 @@ static void handle_layer_pioavail(struct ipath_devdata *dd) { int ret; - ret = __ipath_layer_intr(dd, IPATH_LAYER_INT_SEND_CONTINUE); - if (ret > 0) - goto set; - ret = ipath_ib_piobufavail(dd->verbs_dev); if (ret > 0) goto set; diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h index f1931105adb..999249b7f27 100644 --- a/drivers/infiniband/hw/ipath/ipath_kernel.h +++ b/drivers/infiniband/hw/ipath/ipath_kernel.h @@ -518,16 +518,6 @@ extern struct list_head ipath_dev_list; extern spinlock_t ipath_devs_lock; extern struct ipath_devdata *ipath_lookup(int unit); -extern u16 ipath_layer_rcv_opcode; -extern int __ipath_layer_intr(struct ipath_devdata *, u32); -extern int ipath_layer_intr(struct ipath_devdata *, u32); -extern int __ipath_layer_rcv(struct ipath_devdata *, void *, - struct sk_buff *); -extern int __ipath_layer_rcv_lid(struct ipath_devdata *, void *); - -void ipath_layer_add(struct ipath_devdata *); -void ipath_layer_remove(struct ipath_devdata *); - int ipath_init_chip(struct ipath_devdata *, int); int ipath_enable_wc(struct ipath_devdata *dd); void ipath_disable_wc(struct ipath_devdata *dd); @@ -575,12 +565,13 @@ void ipath_free_pddata(struct ipath_devdata *, struct ipath_portdata *); int ipath_parse_ushort(const char *str, unsigned short *valp); -int ipath_wait_linkstate(struct ipath_devdata *, u32, int); -void ipath_set_ib_lstate(struct ipath_devdata *, int); void ipath_kreceive(struct ipath_devdata *); int ipath_setrcvhdrsize(struct ipath_devdata *, unsigned); int ipath_reset_device(int); void ipath_get_faststats(unsigned long); +int ipath_set_linkstate(struct ipath_devdata *, u8); +int ipath_set_mtu(struct ipath_devdata *, u16); +int ipath_set_lid(struct ipath_devdata *, u32, u8); /* for use in system calls, where we want to know device type, etc. */ #define port_fp(fp) ((struct ipath_portdata *) (fp)->private_data) diff --git a/drivers/infiniband/hw/ipath/ipath_layer.c b/drivers/infiniband/hw/ipath/ipath_layer.c index acc32200cc0..10f578e2aed 100644 --- a/drivers/infiniband/hw/ipath/ipath_layer.c +++ b/drivers/infiniband/hw/ipath/ipath_layer.c @@ -101,242 +101,14 @@ int __ipath_layer_rcv_lid(struct ipath_devdata *dd, void *hdr) return ret; } -int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 newstate) +void ipath_layer_lid_changed(struct ipath_devdata *dd) { - u32 lstate; - int ret; - - switch (newstate) { - case IPATH_IB_LINKDOWN: - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_POLL << - INFINIPATH_IBCC_LINKINITCMD_SHIFT); - /* 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); - /* 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); - /* don't wait */ - ret = 0; - goto bail; - - case IPATH_IB_LINKINIT: - if (dd->ipath_flags & IPATH_LINKINIT) { - ret = 0; - goto bail; - } - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_INIT << - INFINIPATH_IBCC_LINKCMD_SHIFT); - lstate = IPATH_LINKINIT; - break; - - case IPATH_IB_LINKARM: - if (dd->ipath_flags & IPATH_LINKARMED) { - ret = 0; - goto bail; - } - if (!(dd->ipath_flags & - (IPATH_LINKINIT | IPATH_LINKACTIVE))) { - ret = -EINVAL; - goto bail; - } - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED << - INFINIPATH_IBCC_LINKCMD_SHIFT); - /* - * Since the port can transition to ACTIVE by receiving - * a non VL 15 packet, wait for either state. - */ - lstate = IPATH_LINKARMED | IPATH_LINKACTIVE; - break; - - case IPATH_IB_LINKACTIVE: - if (dd->ipath_flags & IPATH_LINKACTIVE) { - ret = 0; - goto bail; - } - if (!(dd->ipath_flags & IPATH_LINKARMED)) { - ret = -EINVAL; - goto bail; - } - ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE << - INFINIPATH_IBCC_LINKCMD_SHIFT); - lstate = IPATH_LINKACTIVE; - break; - - default: - ipath_dbg("Invalid linkstate 0x%x requested\n", newstate); - ret = -EINVAL; - goto bail; - } - ret = ipath_wait_linkstate(dd, lstate, 2000); - -bail: - return ret; -} - -/** - * ipath_layer_set_mtu - set the MTU - * @dd: the infinipath device - * @arg: the new MTU - * - * we can handle "any" incoming size, the issue here is whether we - * need to restrict our outgoing size. For now, we don't do any - * 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... - */ -int ipath_layer_set_mtu(struct ipath_devdata *dd, u16 arg) -{ - u32 piosize; - int changed = 0; - int ret; - - /* - * mtu is IB data payload max. It's the largest power of 2 less - * than piosize (or even larger, since it only really controls the - * largest we can receive; we can send the max of the mtu and - * piosize). We check that it's one of the valid IB sizes. - */ - if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 && - arg != 4096) { - ipath_dbg("Trying to set invalid mtu %u, failing\n", arg); - ret = -EINVAL; - goto bail; - } - if (dd->ipath_ibmtu == arg) { - ret = 0; /* same as current */ - goto bail; - } - - piosize = dd->ipath_ibmaxlen; - dd->ipath_ibmtu = 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) { - dd->ipath_ibmaxlen = piosize; - changed = 1; - } - } else if ((arg + IPATH_PIO_MAXIBHDR) != dd->ipath_ibmaxlen) { - piosize = arg + IPATH_PIO_MAXIBHDR; - ipath_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x " - "(mtu 0x%x)\n", dd->ipath_ibmaxlen, piosize, - arg); - dd->ipath_ibmaxlen = piosize; - changed = 1; - } - - if (changed) { - /* - * set the IBC maxpktlength to the size of our pio - * buffers in words - */ - u64 ibc = dd->ipath_ibcctrl; - 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->ipath_ibcctrl = ibc; - ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, - dd->ipath_ibcctrl); - dd->ipath_f_tidtemplate(dd); - } - - ret = 0; - -bail: - return ret; -} - -int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc) -{ - dd->ipath_lid = arg; - dd->ipath_lmc = lmc; - mutex_lock(&ipath_layer_mutex); if (dd->ipath_layer.l_arg && layer_intr) layer_intr(dd->ipath_layer.l_arg, IPATH_LAYER_INT_LID); mutex_unlock(&ipath_layer_mutex); - - return 0; -} - -int ipath_layer_set_guid(struct ipath_devdata *dd, __be64 guid) -{ - /* XXX - need to inform anyone who cares this just happened. */ - dd->ipath_guid = guid; - return 0; -} - -__be64 ipath_layer_get_guid(struct ipath_devdata *dd) -{ - return dd->ipath_guid; -} - -u32 ipath_layer_get_majrev(struct ipath_devdata *dd) -{ - return dd->ipath_majrev; -} - -u32 ipath_layer_get_minrev(struct ipath_devdata *dd) -{ - return dd->ipath_minrev; -} - -u32 ipath_layer_get_pcirev(struct ipath_devdata *dd) -{ - return dd->ipath_pcirev; -} - -u32 ipath_layer_get_flags(struct ipath_devdata *dd) -{ - return dd->ipath_flags; -} - -struct device *ipath_layer_get_device(struct ipath_devdata *dd) -{ - return &dd->pcidev->dev; -} - -u16 ipath_layer_get_deviceid(struct ipath_devdata *dd) -{ - return dd->ipath_deviceid; -} - -u32 ipath_layer_get_vendorid(struct ipath_devdata *dd) -{ - return dd->ipath_vendorid; -} - -u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd) -{ - return dd->ipath_lastibcstat; -} - -u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd) -{ - return dd->ipath_ibmtu; } void ipath_layer_add(struct ipath_devdata *dd) @@ -436,22 +208,6 @@ void ipath_layer_unregister(void) EXPORT_SYMBOL_GPL(ipath_layer_unregister); -static void __ipath_verbs_timer(unsigned long arg) -{ - struct ipath_devdata *dd = (struct ipath_devdata *) arg; - - /* - * If port 0 receive packet interrupts are not available, or - * can be missed, poll the receive queue - */ - if (dd->ipath_flags & IPATH_POLL_RX_INTR) - ipath_kreceive(dd); - - /* Handle verbs layer timeouts. */ - ipath_ib_timer(dd->verbs_dev); - mod_timer(&dd->verbs_timer, jiffies + 1); -} - int ipath_layer_open(struct ipath_devdata *dd, u32 * pktmax) { int ret; @@ -540,380 +296,6 @@ u16 ipath_layer_get_bcast(struct ipath_devdata *dd) EXPORT_SYMBOL_GPL(ipath_layer_get_bcast); -u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd) -{ - return ipath_read_creg32(dd, dd->ipath_cregs->cr_errpkey); -} - -static void update_sge(struct ipath_sge_state *ss, u32 length) -{ - struct ipath_sge *sge = &ss->sge; - - sge->vaddr += length; - sge->length -= length; - sge->sge_length -= length; - 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) - return; - 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; - } -} - -#ifdef __LITTLE_ENDIAN -static inline u32 get_upper_bits(u32 data, u32 shift) -{ - return data >> shift; -} - -static inline u32 set_upper_bits(u32 data, u32 shift) -{ - return data << shift; -} - -static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off) -{ - data <<= ((sizeof(u32) - n) * BITS_PER_BYTE); - data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE); - return data; -} -#else -static inline u32 get_upper_bits(u32 data, u32 shift) -{ - return data << shift; -} - -static inline u32 set_upper_bits(u32 data, u32 shift) -{ - return data >> shift; -} - -static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off) -{ - data >>= ((sizeof(u32) - n) * BITS_PER_BYTE); - data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE); - return data; -} -#endif - -static void copy_io(u32 __iomem *piobuf, struct ipath_sge_state *ss, - u32 length) -{ - u32 extra = 0; - u32 data = 0; - u32 last; - - while (1) { - u32 len = ss->sge.length; - u32 off; - - BUG_ON(len == 0); - if (len > length) - len = length; - if (len > ss->sge.sge_length) - len = ss->sge.sge_length; - /* If the source address is not aligned, try to align it. */ - off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1); - if (off) { - u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr & - ~(sizeof(u32) - 1)); - u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE); - u32 y; - - y = sizeof(u32) - off; - if (len > y) - len = y; - if (len + extra >= sizeof(u32)) { - data |= set_upper_bits(v, extra * - BITS_PER_BYTE); - len = sizeof(u32) - extra; - if (len == length) { - last = data; - break; - } - __raw_writel(data, piobuf); - piobuf++; - extra = 0; - data = 0; - } else { - /* Clear unused upper bytes */ - data |= clear_upper_bytes(v, len, extra); - if (len == length) { - last = data; - break; - } - extra += len; - } - } else if (extra) { - /* Source address is aligned. */ - u32 *addr = (u32 *) ss->sge.vaddr; - int shift = extra * BITS_PER_BYTE; - int ushift = 32 - shift; - u32 l = len; - - while (l >= sizeof(u32)) { - u32 v = *addr; - - data |= set_upper_bits(v, shift); - __raw_writel(data, piobuf); - data = get_upper_bits(v, ushift); - piobuf++; - addr++; - l -= sizeof(u32); - } - /* - * We still have 'extra' number of bytes leftover. - */ - if (l) { - u32 v = *addr; - - if (l + extra >= sizeof(u32)) { - data |= set_upper_bits(v, shift); - len -= l + extra - sizeof(u32); - if (len == length) { - last = data; - break; - } - __raw_writel(data, piobuf); - piobuf++; - extra = 0; - data = 0; - } else { - /* Clear unused upper bytes */ - data |= clear_upper_bytes(v, l, - extra); - if (len == length) { - last = data; - break; - } - extra += l; - } - } else if (len == length) { - last = data; - break; - } - } else if (len == length) { - u32 w; - - /* - * Need to round up for the last dword in the - * packet. - */ - w = (len + 3) >> 2; - __iowrite32_copy(piobuf, ss->sge.vaddr, w - 1); - piobuf += w - 1; - last = ((u32 *) ss->sge.vaddr)[w - 1]; - break; - } else { - u32 w = len >> 2; - - __iowrite32_copy(piobuf, ss->sge.vaddr, w); - piobuf += w; - - extra = len & (sizeof(u32) - 1); - if (extra) { - u32 v = ((u32 *) ss->sge.vaddr)[w]; - - /* Clear unused upper bytes */ - data = clear_upper_bytes(v, extra, 0); - } - } - update_sge(ss, len); - length -= len; - } - /* Update address before sending packet. */ - update_sge(ss, length); - /* must flush early everything before trigger word */ - ipath_flush_wc(); - __raw_writel(last, piobuf); - /* be sure trigger word is written */ - ipath_flush_wc(); -} - -/** - * ipath_verbs_send - send a packet from the verbs layer - * @dd: the infinipath device - * @hdrwords: the number of words in the header - * @hdr: the packet header - * @len: the length of the packet in bytes - * @ss: the SGE to send - * - * This is like ipath_sma_send_pkt() in that we need to be able to send - * packets after the chip is initialized (MADs) but also like - * ipath_layer_send_hdr() since its used by the verbs layer. - */ -int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords, - u32 *hdr, u32 len, struct ipath_sge_state *ss) -{ - u32 __iomem *piobuf; - u32 plen; - int ret; - - /* +1 is for the qword padding of pbc */ - plen = hdrwords + ((len + 3) >> 2) + 1; - if (unlikely((plen << 2) > dd->ipath_ibmaxlen)) { - ipath_dbg("packet len 0x%x too long, failing\n", plen); - ret = -EINVAL; - goto bail; - } - - /* Get a PIO buffer to use. */ - piobuf = ipath_getpiobuf(dd, NULL); - if (unlikely(piobuf == NULL)) { - ret = -EBUSY; - goto bail; - } - - /* - * Write len to control qword, no 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); - ipath_flush_wc(); - piobuf += 2; - if (len == 0) { - /* - * If there is just the header portion, must flush before - * writing last word of header for correctness, and after - * the last header word (trigger word). - */ - __iowrite32_copy(piobuf, hdr, hdrwords - 1); - ipath_flush_wc(); - __raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1); - ipath_flush_wc(); - ret = 0; - goto bail; - } - - __iowrite32_copy(piobuf, hdr, hdrwords); - piobuf += hdrwords; - - /* The common case is aligned and contained in one segment. */ - if (likely(ss->num_sge == 1 && len <= ss->sge.length && - !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) { - u32 w; - u32 *addr = (u32 *) ss->sge.vaddr; - - /* Update address before sending packet. */ - update_sge(ss, len); - /* Need to round up for the last dword in the packet. */ - w = (len + 3) >> 2; - __iowrite32_copy(piobuf, addr, w - 1); - /* must flush early everything before trigger word */ - ipath_flush_wc(); - __raw_writel(addr[w - 1], piobuf + w - 1); - /* be sure trigger word is written */ - ipath_flush_wc(); - ret = 0; - goto bail; - } - copy_io(piobuf, ss, len); - ret = 0; - -bail: - return ret; -} - -int ipath_layer_snapshot_counters(struct ipath_devdata *dd, u64 *swords, - u64 *rwords, u64 *spkts, u64 *rpkts, - u64 *xmit_wait) -{ - int ret; - - if (!(dd->ipath_flags & IPATH_INITTED)) { - /* no hardware, freeze, etc. */ - ipath_dbg("unit %u not usable\n", dd->ipath_unit); - ret = -EINVAL; - goto bail; - } - *swords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt); - *rwords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt); - *spkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt); - *rpkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt); - *xmit_wait = ipath_snap_cntr(dd, dd->ipath_cregs->cr_sendstallcnt); - - ret = 0; - -bail: - return ret; -} - -/** - * ipath_layer_get_counters - get various chip counters - * @dd: the infinipath device - * @cntrs: counters are placed here - * - * Return the counters needed by recv_pma_get_portcounters(). - */ -int ipath_layer_get_counters(struct ipath_devdata *dd, - struct ipath_layer_counters *cntrs) -{ - int ret; - - if (!(dd->ipath_flags & IPATH_INITTED)) { - /* no hardware, freeze, etc. */ - ipath_dbg("unit %u not usable\n", dd->ipath_unit); - ret = -EINVAL; - goto bail; - } - cntrs->symbol_error_counter = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_ibsymbolerrcnt); - cntrs->link_error_recovery_counter = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkerrrecovcnt); - /* - * The link downed counter counts when the other side downs the - * connection. We add in the number of times we downed the link - * due to local link integrity errors to compensate. - */ - cntrs->link_downed_counter = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkdowncnt); - cntrs->port_rcv_errors = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_rxdroppktcnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvovflcnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_portovflcnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_err_rlencnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_invalidrlencnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) + - ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt); - cntrs->port_rcv_remphys_errors = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt); - cntrs->port_xmit_discards = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_unsupvlcnt); - cntrs->port_xmit_data = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt); - cntrs->port_rcv_data = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt); - cntrs->port_xmit_packets = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt); - cntrs->port_rcv_packets = - ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt); - cntrs->local_link_integrity_errors = dd->ipath_lli_errors; - cntrs->excessive_buffer_overrun_errors = 0; /* XXX */ - - ret = 0; - -bail: - return ret; -} - -int ipath_layer_want_buffer(struct ipath_devdata *dd) -{ - set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl); - ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, - dd->ipath_sendctrl); - - return 0; -} - int ipath_layer_send_hdr(struct ipath_devdata *dd, struct ether_header *hdr) { int ret = 0; @@ -985,361 +367,3 @@ int ipath_layer_set_piointbufavail_int(struct ipath_devdata *dd) } EXPORT_SYMBOL_GPL(ipath_layer_set_piointbufavail_int); - -int ipath_layer_enable_timer(struct ipath_devdata *dd) -{ - /* - * HT-400 has a design flaw where the chip and kernel idea - * of the tail register don't always agree, and therefore we won't - * get an interrupt on the next packet received. - * If the board supports per packet receive interrupts, use it. - * Otherwise, the timer function periodically checks for packets - * to cover this case. - * Either way, the timer is needed for verbs layer related - * processing. - */ - if (dd->ipath_flags & IPATH_GPIO_INTR) { - ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect, - 0x2074076542310ULL); - /* Enable GPIO bit 2 interrupt */ - ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, - (u64) (1 << 2)); - } - - init_timer(&dd->verbs_timer); - dd->verbs_timer.function = __ipath_verbs_timer; - dd->verbs_timer.data = (unsigned long)dd; - dd->verbs_timer.expires = jiffies + 1; - add_timer(&dd->verbs_timer); - - return 0; -} - -int ipath_layer_disable_timer(struct ipath_devdata *dd) -{ - /* Disable GPIO bit 2 interrupt */ - if (dd->ipath_flags & IPATH_GPIO_INTR) - ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 0); - - del_timer_sync(&dd->verbs_timer); - - return 0; -} - -/** - * ipath_layer_set_verbs_flags - set the verbs layer flags - * @dd: the infinipath device - * @flags: the flags to set - */ -int ipath_layer_set_verbs_flags(struct ipath_devdata *dd, unsigned flags) -{ - struct ipath_devdata *ss; - unsigned long lflags; - - spin_lock_irqsave(&ipath_devs_lock, lflags); - - list_for_each_entry(ss, &ipath_dev_list, ipath_list) { - if (!(ss->ipath_flags & IPATH_INITTED)) - continue; - if ((flags & IPATH_VERBS_KERNEL_SMA) && - !(*ss->ipath_statusp & IPATH_STATUS_SMA)) - *ss->ipath_statusp |= IPATH_STATUS_OIB_SMA; - else - *ss->ipath_statusp &= ~IPATH_STATUS_OIB_SMA; - } - - spin_unlock_irqrestore(&ipath_devs_lock, lflags); - - return 0; -} - -/** - * ipath_layer_get_npkeys - return the size of the PKEY table for port 0 - * @dd: the infinipath device - */ -unsigned ipath_layer_get_npkeys(struct ipath_devdata *dd) -{ - return ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys); -} - -/** - * ipath_layer_get_pkey - return the indexed PKEY from the port 0 PKEY table - * @dd: the infinipath device - * @index: the PKEY index - */ -unsigned ipath_layer_get_pkey(struct ipath_devdata *dd, unsigned index) -{ - unsigned ret; - - if (index >= ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys)) - ret = 0; - else - ret = dd->ipath_pd[0]->port_pkeys[index]; - - return ret; -} - -/** - * ipath_layer_get_pkeys - return the PKEY table for port 0 - * @dd: the infinipath device - * @pkeys: the pkey table is placed here - */ -int ipath_layer_get_pkeys(struct ipath_devdata *dd, u16 * pkeys) -{ - struct ipath_portdata *pd = dd->ipath_pd[0]; - - memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys)); - - return 0; -} - -/** - * rm_pkey - decrecment the reference count for the given PKEY - * @dd: the infinipath device - * @key: the PKEY index - * - * Return true if this was the last reference and the hardware table entry - * needs to be changed. - */ -static int rm_pkey(struct ipath_devdata *dd, u16 key) -{ - int i; - int ret; - - for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) { - if (dd->ipath_pkeys[i] != key) - continue; - if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) { - dd->ipath_pkeys[i] = 0; - ret = 1; - goto bail; - } - break; - } - - ret = 0; - -bail: - return ret; -} - -/** - * add_pkey - add the given PKEY to the hardware table - * @dd: the infinipath device - * @key: the PKEY - * - * Return an error code if unable to add the entry, zero if no change, - * or 1 if the hardware PKEY register needs to be updated. - */ -static int add_pkey(struct ipath_devdata *dd, u16 key) -{ - int i; - u16 lkey = key & 0x7FFF; - int any = 0; - int ret; - - if (lkey == 0x7FFF) { - ret = 0; - goto bail; - } - - /* Look for an empty slot or a matching PKEY. */ - for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) { - if (!dd->ipath_pkeys[i]) { - any++; - continue; - } - /* If it matches exactly, try to increment the ref count */ - if (dd->ipath_pkeys[i] == key) { - if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) { - ret = 0; - goto bail; - } - /* Lost the race. Look for an empty slot below. */ - atomic_dec(&dd->ipath_pkeyrefs[i]); - any++; - } - /* - * It makes no sense to have both the limited and unlimited - * PKEY set at the same time since the unlimited one will - * disable the limited one. - */ - if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) { - ret = -EEXIST; - goto bail; - } - } - if (!any) { - ret = -EBUSY; - goto bail; - } - for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) { - if (!dd->ipath_pkeys[i] && - atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) { - /* for ipathstats, etc. */ - ipath_stats.sps_pkeys[i] = lkey; - dd->ipath_pkeys[i] = key; - ret = 1; - goto bail; - } - } - ret = -EBUSY; - -bail: - return ret; -} - -/** - * ipath_layer_set_pkeys - set the PKEY table for port 0 - * @dd: the infinipath device - * @pkeys: the PKEY table - */ -int ipath_layer_set_pkeys(struct ipath_devdata *dd, u16 * pkeys) -{ - struct ipath_portdata *pd; - int i; - int changed = 0; - - pd = dd->ipath_pd[0]; - - for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) { - u16 key = pkeys[i]; - u16 okey = pd->port_pkeys[i]; - - if (key == okey) - continue; - /* - * The value of this PKEY table entry is changing. - * Remove the old entry in the hardware's array of PKEYs. - */ - if (okey & 0x7FFF) - changed |= rm_pkey(dd, okey); - if (key & 0x7FFF) { - int ret = add_pkey(dd, key); - - if (ret < 0) - key = 0; - else - changed |= ret; - } - pd->port_pkeys[i] = key; - } - if (changed) { - u64 pkey; - - pkey = (u64) dd->ipath_pkeys[0] | - ((u64) dd->ipath_pkeys[1] << 16) | - ((u64) dd->ipath_pkeys[2] << 32) | - ((u64) dd->ipath_pkeys[3] << 48); - ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n", - (unsigned long long) pkey); - ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey, - pkey); - } - return 0; -} - -/** - * ipath_layer_get_linkdowndefaultstate - get the default linkdown state - * @dd: the infinipath device - * - * Returns zero if the default is POLL, 1 if the default is SLEEP. - */ -int ipath_layer_get_linkdowndefaultstate(struct ipath_devdata *dd) -{ - return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE); -} - -/** - * ipath_layer_set_linkdowndefaultstate - set the default linkdown state - * @dd: the infinipath device - * @sleep: the new state - * - * Note that this will only take effect when the link state changes. - */ -int ipath_layer_set_linkdowndefaultstate(struct ipath_devdata *dd, - int sleep) -{ - if (sleep) - dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE; - else - dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE; - ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, - dd->ipath_ibcctrl); - return 0; -} - -int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd) -{ - return (dd->ipath_ibcctrl >> - INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) & - INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK; -} - -/** - * ipath_layer_set_phyerrthreshold - set the physical error threshold - * @dd: the infinipath device - * @n: the new threshold - * - * Note that this will only take effect when the link state changes. - */ -int ipath_layer_set_phyerrthreshold(struct ipath_devdata *dd, unsigned n) -{ - unsigned v; - - v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) & - INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK; - if (v != n) { - dd->ipath_ibcctrl &= - ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK << - INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT); - dd->ipath_ibcctrl |= - (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT; - ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, - dd->ipath_ibcctrl); - } - return 0; -} - -int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd) -{ - return (dd->ipath_ibcctrl >> - INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) & - INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK; -} - -/** - * ipath_layer_set_overrunthreshold - set the overrun threshold - * @dd: the infinipath device - * @n: the new threshold - * - * Note that this will only take effect when the link state changes. - */ -int ipath_layer_set_overrunthreshold(struct ipath_devdata *dd, unsigned n) -{ - unsigned v; - - v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) & - INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK; - if (v != n) { - dd->ipath_ibcctrl &= - ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK << - INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT); - dd->ipath_ibcctrl |= - (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT; - ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, - dd->ipath_ibcctrl); - } - return 0; -} - -int ipath_layer_get_boardname(struct ipath_devdata *dd, char *name, - size_t namelen) -{ - return dd->ipath_f_get_boardname(dd, name, namelen); -} - -u32 ipath_layer_get_rcvhdrentsize(struct ipath_devdata *dd) -{ - return dd->ipath_rcvhdrentsize; -} diff --git a/drivers/infiniband/hw/ipath/ipath_layer.h b/drivers/infiniband/hw/ipath/ipath_layer.h index 57c990a5715..4a27ede4994 100644 --- a/drivers/infiniband/hw/ipath/ipath_layer.h +++ b/drivers/infiniband/hw/ipath/ipath_layer.h @@ -40,73 +40,9 @@ */ struct sk_buff; -struct ipath_sge_state; struct ipath_devdata; struct ether_header; -struct ipath_layer_counters { - u64 symbol_error_counter; - u64 link_error_recovery_counter; - u64 link_downed_counter; - u64 port_rcv_errors; - u64 port_rcv_remphys_errors; - u64 port_xmit_discards; - u64 port_xmit_data; - u64 port_rcv_data; - u64 port_xmit_packets; - u64 port_rcv_packets; - u32 local_link_integrity_errors; - u32 excessive_buffer_overrun_errors; -}; - -/* - * A segment is a linear region of low physical memory. - * XXX Maybe we should use phys addr here and kmap()/kunmap(). - * Used by the verbs layer. - */ -struct ipath_seg { - void *vaddr; - size_t length; -}; - -/* The number of ipath_segs that fit in a page. */ -#define IPATH_SEGSZ (PAGE_SIZE / sizeof (struct ipath_seg)) - -struct ipath_segarray { - struct ipath_seg segs[IPATH_SEGSZ]; -}; - -struct ipath_mregion { - u64 user_base; /* User's address for this region */ - u64 iova; /* IB start address of this region */ - size_t length; - u32 lkey; - u32 offset; /* offset (bytes) to start of region */ - int access_flags; - u32 max_segs; /* number of ipath_segs in all the arrays */ - u32 mapsz; /* size of the map array */ - struct ipath_segarray *map[0]; /* the segments */ -}; - -/* - * These keep track of the copy progress within a memory region. - * Used by the verbs layer. - */ -struct ipath_sge { - struct ipath_mregion *mr; - void *vaddr; /* current pointer into the segment */ - u32 sge_length; /* length of the SGE */ - u32 length; /* remaining length of the segment */ - u16 m; /* current index: mr->map[m] */ - u16 n; /* current index: mr->map[m]->segs[n] */ -}; - -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; -}; - int ipath_layer_register(void *(*l_add)(int, struct ipath_devdata *), void (*l_remove)(void *), int (*l_intr)(void *, u32), @@ -119,49 +55,9 @@ int ipath_layer_open(struct ipath_devdata *, u32 * pktmax); u16 ipath_layer_get_lid(struct ipath_devdata *dd); int ipath_layer_get_mac(struct ipath_devdata *dd, u8 *); u16 ipath_layer_get_bcast(struct ipath_devdata *dd); -u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd); -int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 state); -int ipath_layer_set_mtu(struct ipath_devdata *, u16); -int ipath_set_lid(struct ipath_devdata *, u32, u8); int ipath_layer_send_hdr(struct ipath_devdata *dd, struct ether_header *hdr); -int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords, - u32 * hdr, u32 len, struct ipath_sge_state *ss); int ipath_layer_set_piointbufavail_int(struct ipath_devdata *dd); -int ipath_layer_get_boardname(struct ipath_devdata *dd, char *name, - size_t namelen); -int ipath_layer_snapshot_counters(struct ipath_devdata *dd, u64 *swords, - u64 *rwords, u64 *spkts, u64 *rpkts, - u64 *xmit_wait); -int ipath_layer_get_counters(struct ipath_devdata *dd, - struct ipath_layer_counters *cntrs); -int ipath_layer_want_buffer(struct ipath_devdata *dd); -int ipath_layer_set_guid(struct ipath_devdata *, __be64 guid); -__be64 ipath_layer_get_guid(struct ipath_devdata *); -u32 ipath_layer_get_majrev(struct ipath_devdata *); -u32 ipath_layer_get_minrev(struct ipath_devdata *); -u32 ipath_layer_get_pcirev(struct ipath_devdata *); -u32 ipath_layer_get_flags(struct ipath_devdata *dd); -struct device *ipath_layer_get_device(struct ipath_devdata *dd); -u16 ipath_layer_get_deviceid(struct ipath_devdata *dd); -u32 ipath_layer_get_vendorid(struct ipath_devdata *); -u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd); -u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd); -int ipath_layer_enable_timer(struct ipath_devdata *dd); -int ipath_layer_disable_timer(struct ipath_devdata *dd); -int ipath_layer_set_verbs_flags(struct ipath_devdata *dd, unsigned flags); -unsigned ipath_layer_get_npkeys(struct ipath_devdata *dd); -unsigned ipath_layer_get_pkey(struct ipath_devdata *dd, unsigned index); -int ipath_layer_get_pkeys(struct ipath_devdata *dd, u16 *pkeys); -int ipath_layer_set_pkeys(struct ipath_devdata *dd, u16 *pkeys); -int ipath_layer_get_linkdowndefaultstate(struct ipath_devdata *dd); -int ipath_layer_set_linkdowndefaultstate(struct ipath_devdata *dd, - int sleep); -int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd); -int ipath_layer_set_phyerrthreshold(struct ipath_devdata *dd, unsigned n); -int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd); -int ipath_layer_set_overrunthreshold(struct ipath_devdata *dd, unsigned n); -u32 ipath_layer_get_rcvhdrentsize(struct ipath_devdata *dd); /* ipath_ether interrupt values */ #define IPATH_LAYER_INT_IF_UP 0x2 diff --git a/drivers/infiniband/hw/ipath/ipath_mad.c b/drivers/infiniband/hw/ipath/ipath_mad.c index d3402341b7d..72d1db89db8 100644 --- a/drivers/infiniband/hw/ipath/ipath_mad.c +++ b/drivers/infiniband/hw/ipath/ipath_mad.c @@ -101,15 +101,15 @@ static int recv_subn_get_nodeinfo(struct ib_smp *smp, nip->num_ports = ibdev->phys_port_cnt; /* This is already in network order */ nip->sys_guid = to_idev(ibdev)->sys_image_guid; - nip->node_guid = ipath_layer_get_guid(dd); + nip->node_guid = dd->ipath_guid; nip->port_guid = nip->sys_guid; - nip->partition_cap = cpu_to_be16(ipath_layer_get_npkeys(dd)); - nip->device_id = cpu_to_be16(ipath_layer_get_deviceid(dd)); - majrev = ipath_layer_get_majrev(dd); - minrev = ipath_layer_get_minrev(dd); + nip->partition_cap = cpu_to_be16(ipath_get_npkeys(dd)); + nip->device_id = cpu_to_be16(dd->ipath_deviceid); + majrev = dd->ipath_majrev; + minrev = dd->ipath_minrev; nip->revision = cpu_to_be32((majrev << 16) | minrev); nip->local_port_num = port; - vendor = ipath_layer_get_vendorid(dd); + vendor = dd->ipath_vendorid; nip->vendor_id[0] = 0; nip->vendor_id[1] = vendor >> 8; nip->vendor_id[2] = vendor; @@ -133,13 +133,89 @@ static int recv_subn_get_guidinfo(struct ib_smp *smp, */ if (startgx == 0) /* The first is a copy of the read-only HW GUID. */ - *p = ipath_layer_get_guid(to_idev(ibdev)->dd); + *p = to_idev(ibdev)->dd->ipath_guid; else smp->status |= IB_SMP_INVALID_FIELD; return reply(smp); } + +static int get_overrunthreshold(struct ipath_devdata *dd) +{ + return (dd->ipath_ibcctrl >> + INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) & + INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK; +} + +/** + * set_overrunthreshold - set the overrun threshold + * @dd: the infinipath device + * @n: the new threshold + * + * Note that this will only take effect when the link state changes. + */ +static int set_overrunthreshold(struct ipath_devdata *dd, unsigned n) +{ + unsigned v; + + v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) & + INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK; + if (v != n) { + dd->ipath_ibcctrl &= + ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK << + INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT); + dd->ipath_ibcctrl |= + (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT; + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, + dd->ipath_ibcctrl); + } + return 0; +} + +static int get_phyerrthreshold(struct ipath_devdata *dd) +{ + return (dd->ipath_ibcctrl >> + INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) & + INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK; +} + +/** + * set_phyerrthreshold - set the physical error threshold + * @dd: the infinipath device + * @n: the new threshold + * + * Note that this will only take effect when the link state changes. + */ +static int set_phyerrthreshold(struct ipath_devdata *dd, unsigned n) +{ + unsigned v; + + v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) & + INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK; + if (v != n) { + dd->ipath_ibcctrl &= + ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK << + INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT); + dd->ipath_ibcctrl |= + (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT; + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, + dd->ipath_ibcctrl); + } + return 0; +} + +/** + * get_linkdowndefaultstate - get the default linkdown state + * @dd: the infinipath device + * + * Returns zero if the default is POLL, 1 if the default is SLEEP. + */ +static int get_linkdowndefaultstate(struct ipath_devdata *dd) +{ + return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE); +} + static int recv_subn_get_portinfo(struct ib_smp *smp, struct ib_device *ibdev, u8 port) { @@ -166,7 +242,7 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, (dev->mkeyprot_resv_lmc >> 6) == 0) pip->mkey = dev->mkey; pip->gid_prefix = dev->gid_prefix; - lid = ipath_layer_get_lid(dev->dd); + lid = dev->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); @@ -177,14 +253,14 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, pip->link_width_supported = 3; /* 1x or 4x */ pip->link_width_active = 2; /* 4x */ pip->linkspeed_portstate = 0x10; /* 2.5Gbps */ - ibcstat = ipath_layer_get_lastibcstat(dev->dd); + ibcstat = dev->dd->ipath_lastibcstat; pip->linkspeed_portstate |= ((ibcstat >> 4) & 0x3) + 1; pip->portphysstate_linkdown = (ipath_cvt_physportstate[ibcstat & 0xf] << 4) | - (ipath_layer_get_linkdowndefaultstate(dev->dd) ? 1 : 2); + (get_linkdowndefaultstate(dev->dd) ? 1 : 2); pip->mkeyprot_resv_lmc = dev->mkeyprot_resv_lmc; pip->linkspeedactive_enabled = 0x11; /* 2.5Gbps, 2.5Gbps */ - switch (ipath_layer_get_ibmtu(dev->dd)) { + switch (dev->dd->ipath_ibmtu) { case 4096: mtu = IB_MTU_4096; break; @@ -217,7 +293,7 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, pip->mkey_violations = cpu_to_be16(dev->mkey_violations); /* P_KeyViolations are counted by hardware. */ pip->pkey_violations = - cpu_to_be16((ipath_layer_get_cr_errpkey(dev->dd) - + cpu_to_be16((ipath_get_cr_errpkey(dev->dd) - dev->z_pkey_violations) & 0xFFFF); pip->qkey_violations = cpu_to_be16(dev->qkey_violations); /* Only the hardware GUID is supported for now */ @@ -226,8 +302,8 @@ static int recv_subn_get_portinfo(struct ib_smp *smp, /* 32.768 usec. response time (guessing) */ pip->resv_resptimevalue = 3; pip->localphyerrors_overrunerrors = - (ipath_layer_get_phyerrthreshold(dev->dd) << 4) | - ipath_layer_get_overrunthreshold(dev->dd); + (get_phyerrthreshold(dev->dd) << 4) | + get_overrunthreshold(dev->dd); /* pip->max_credit_hint; */ /* pip->link_roundtrip_latency[3]; */ @@ -237,6 +313,20 @@ bail: return ret; } +/** + * get_pkeys - return the PKEY table for port 0 + * @dd: the infinipath device + * @pkeys: the pkey table is placed here + */ +static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys) +{ + struct ipath_portdata *pd = dd->ipath_pd[0]; + + memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys)); + + return 0; +} + static int recv_subn_get_pkeytable(struct ib_smp *smp, struct ib_device *ibdev) { @@ -249,9 +339,9 @@ static int recv_subn_get_pkeytable(struct ib_smp *smp, memset(smp->data, 0, sizeof(smp->data)); if (startpx == 0) { struct ipath_ibdev *dev = to_idev(ibdev); - unsigned i, n = ipath_layer_get_npkeys(dev->dd); + unsigned i, n = ipath_get_npkeys(dev->dd); - ipath_layer_get_pkeys(dev->dd, p); + get_pkeys(dev->dd, p); for (i = 0; i < n; i++) q[i] = cpu_to_be16(p[i]); @@ -268,6 +358,24 @@ static int recv_subn_set_guidinfo(struct ib_smp *smp, return recv_subn_get_guidinfo(smp, ibdev); } +/** + * set_linkdowndefaultstate - set the default linkdown state + * @dd: the infinipath device + * @sleep: the new state + * + * Note that this will only take effect when the link state changes. + */ +static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep) +{ + if (sleep) + dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE; + else + dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE; + ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, + dd->ipath_ibcctrl); + return 0; +} + /** * recv_subn_set_portinfo - set port information * @smp: the incoming SM packet @@ -290,7 +398,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, u8 state; u16 lstate; u32 mtu; - int ret; + int ret, ore; if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt) goto err; @@ -304,7 +412,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period); lid = be16_to_cpu(pip->lid); - if (lid != ipath_layer_get_lid(dev->dd)) { + if (lid != dev->dd->ipath_lid) { /* Must be a valid unicast LID address. */ if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE) goto err; @@ -342,11 +450,11 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, case 0: /* NOP */ break; case 1: /* SLEEP */ - if (ipath_layer_set_linkdowndefaultstate(dev->dd, 1)) + if (set_linkdowndefaultstate(dev->dd, 1)) goto err; break; case 2: /* POLL */ - if (ipath_layer_set_linkdowndefaultstate(dev->dd, 0)) + if (set_linkdowndefaultstate(dev->dd, 0)) goto err; break; default: @@ -376,7 +484,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, /* XXX We have already partially updated our state! */ goto err; } - ipath_layer_set_mtu(dev->dd, mtu); + ipath_set_mtu(dev->dd, mtu); dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF; @@ -392,20 +500,16 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, * later. */ if (pip->pkey_violations == 0) - dev->z_pkey_violations = - ipath_layer_get_cr_errpkey(dev->dd); + dev->z_pkey_violations = ipath_get_cr_errpkey(dev->dd); if (pip->qkey_violations == 0) dev->qkey_violations = 0; - if (ipath_layer_set_phyerrthreshold( - dev->dd, - (pip->localphyerrors_overrunerrors >> 4) & 0xF)) + ore = pip->localphyerrors_overrunerrors; + if (set_phyerrthreshold(dev->dd, (ore >> 4) & 0xF)) goto err; - if (ipath_layer_set_overrunthreshold( - dev->dd, - (pip->localphyerrors_overrunerrors & 0xF))) + if (set_overrunthreshold(dev->dd, (ore & 0xF))) goto err; dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F; @@ -423,7 +527,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, * is down or is being set to down. */ state = pip->linkspeed_portstate & 0xF; - flags = ipath_layer_get_flags(dev->dd); + flags = dev->dd->ipath_flags; lstate = (pip->portphysstate_linkdown >> 4) & 0xF; if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP)) goto err; @@ -439,7 +543,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, /* FALLTHROUGH */ case IB_PORT_DOWN: if (lstate == 0) - if (ipath_layer_get_linkdowndefaultstate(dev->dd)) + if (get_linkdowndefaultstate(dev->dd)) lstate = IPATH_IB_LINKDOWN_SLEEP; else lstate = IPATH_IB_LINKDOWN; @@ -451,7 +555,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, lstate = IPATH_IB_LINKDOWN_DISABLE; else goto err; - ipath_layer_set_linkstate(dev->dd, lstate); + ipath_set_linkstate(dev->dd, lstate); if (flags & IPATH_LINKACTIVE) { event.event = IB_EVENT_PORT_ERR; ib_dispatch_event(&event); @@ -460,7 +564,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, case IB_PORT_ARMED: if (!(flags & (IPATH_LINKINIT | IPATH_LINKACTIVE))) break; - ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKARM); + ipath_set_linkstate(dev->dd, IPATH_IB_LINKARM); if (flags & IPATH_LINKACTIVE) { event.event = IB_EVENT_PORT_ERR; ib_dispatch_event(&event); @@ -469,7 +573,7 @@ static int recv_subn_set_portinfo(struct ib_smp *smp, case IB_PORT_ACTIVE: if (!(flags & IPATH_LINKARMED)) break; - ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKACTIVE); + ipath_set_linkstate(dev->dd, IPATH_IB_LINKACTIVE); event.event = IB_EVENT_PORT_ACTIVE; ib_dispatch_event(&event); break; @@ -493,6 +597,152 @@ done: return ret; } +/** + * rm_pkey - decrecment the reference count for the given PKEY + * @dd: the infinipath device + * @key: the PKEY index + * + * Return true if this was the last reference and the hardware table entry + * needs to be changed. + */ +static int rm_pkey(struct ipath_devdata *dd, u16 key) +{ + int i; + int ret; + + for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) { + if (dd->ipath_pkeys[i] != key) + continue; + if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) { + dd->ipath_pkeys[i] = 0; + ret = 1; + goto bail; + } + break; + } + + ret = 0; + +bail: + return ret; +} + +/** + * add_pkey - add the given PKEY to the hardware table + * @dd: the infinipath device + * @key: the PKEY + * + * Return an error code if unable to add the entry, zero if no change, + * or 1 if the hardware PKEY register needs to be updated. + */ +static int add_pkey(struct ipath_devdata *dd, u16 key) +{ + int i; + u16 lkey = key & 0x7FFF; + int any = 0; + int ret; + + if (lkey == 0x7FFF) { + ret = 0; + goto bail; + } + + /* Look for an empty slot or a matching PKEY. */ + for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) { + if (!dd->ipath_pkeys[i]) { + any++; + continue; + } + /* If it matches exactly, try to increment the ref count */ + if (dd->ipath_pkeys[i] == key) { + if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) { + ret = 0; + goto bail; + } + /* Lost the race. Look for an empty slot below. */ + atomic_dec(&dd->ipath_pkeyrefs[i]); + any++; + } + /* + * It makes no sense to have both the limited and unlimited + * PKEY set at the same time since the unlimited one will + * disable the limited one. + */ + if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) { + ret = -EEXIST; + goto bail; + } + } + if (!any) { + ret = -EBUSY; + goto bail; + } + for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) { + if (!dd->ipath_pkeys[i] && + atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) { + /* for ipathstats, etc. */ + ipath_stats.sps_pkeys[i] = lkey; + dd->ipath_pkeys[i] = key; + ret = 1; + goto bail; + } + } + ret = -EBUSY; + +bail: + return ret; +} + +/** + * set_pkeys - set the PKEY table for port 0 + * @dd: the infinipath device + * @pkeys: the PKEY table + */ +static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys) +{ + struct ipath_portdata *pd; + int i; + int changed = 0; + + pd = dd->ipath_pd[0]; + + for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) { + u16 key = pkeys[i]; + u16 okey = pd->port_pkeys[i]; + + if (key == okey) + continue; + /* + * The value of this PKEY table entry is changing. + * Remove the old entry in the hardware's array of PKEYs. + */ + if (okey & 0x7FFF) + changed |= rm_pkey(dd, okey); + if (key & 0x7FFF) { + int ret = add_pkey(dd, key); + + if (ret < 0) + key = 0; + else + changed |= ret; + } + pd->port_pkeys[i] = key; + } + if (changed) { + u64 pkey; + + pkey = (u64) dd->ipath_pkeys[0] | + ((u64) dd->ipath_pkeys[1] << 16) | + ((u64) dd->ipath_pkeys[2] << 32) | + ((u64) dd->ipath_pkeys[3] << 48); + ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n", + (unsigned long long) pkey); + ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey, + pkey); + } + return 0; +} + static int recv_subn_set_pkeytable(struct ib_smp *smp, struct ib_device *ibdev) { @@ -500,13 +750,12 @@ static int recv_subn_set_pkeytable(struct ib_smp *smp, __be16 *p = (__be16 *) smp->data; u16 *q = (u16 *) smp->data; struct ipath_ibdev *dev = to_idev(ibdev); - unsigned i, n = ipath_layer_get_npkeys(dev->dd); + unsigned i, n = ipath_get_npkeys(dev->dd); for (i = 0; i < n; i++) q[i] = be16_to_cpu(p[i]); - if (startpx != 0 || - ipath_layer_set_pkeys(dev->dd, q) != 0) + if (startpx != 0 || set_pkeys(dev->dd, q) != 0) smp->status |= IB_SMP_INVALID_FIELD; return recv_subn_get_pkeytable(smp, ibdev); @@ -844,10 +1093,10 @@ static int recv_pma_get_portcounters(struct ib_perf *pmp, struct ib_pma_portcounters *p = (struct ib_pma_portcounters *) pmp->data; struct ipath_ibdev *dev = to_idev(ibdev); - struct ipath_layer_counters cntrs; + struct ipath_verbs_counters cntrs; u8 port_select = p->port_select; - ipath_layer_get_counters(dev->dd, &cntrs); + ipath_get_counters(dev->dd, &cntrs); /* Adjust counters for any resets done. */ cntrs.symbol_error_counter -= dev->z_symbol_error_counter; @@ -944,8 +1193,8 @@ static int recv_pma_get_portcounters_ext(struct ib_perf *pmp, u64 swords, rwords, spkts, rpkts, xwait; u8 port_select = p->port_select; - ipath_layer_snapshot_counters(dev->dd, &swords, &rwords, &spkts, - &rpkts, &xwait); + ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts, + &rpkts, &xwait); /* Adjust counters for any resets done. */ swords -= dev->z_port_xmit_data; @@ -978,13 +1227,13 @@ static int recv_pma_set_portcounters(struct ib_perf *pmp, struct ib_pma_portcounters *p = (struct ib_pma_portcounters *) pmp->data; struct ipath_ibdev *dev = to_idev(ibdev); - struct ipath_layer_counters cntrs; + struct ipath_verbs_counters cntrs; /* * Since the HW doesn't support clearing counters, we save the * current count and subtract it from future responses. */ - ipath_layer_get_counters(dev->dd, &cntrs); + ipath_get_counters(dev->dd, &cntrs); if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR) dev->z_symbol_error_counter = cntrs.symbol_error_counter; @@ -1041,8 +1290,8 @@ static int recv_pma_set_portcounters_ext(struct ib_perf *pmp, struct ipath_ibdev *dev = to_idev(ibdev); u64 swords, rwords, spkts, rpkts, xwait; - ipath_layer_snapshot_counters(dev->dd, &swords, &rwords, &spkts, - &rpkts, &xwait); + ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts, + &rpkts, &xwait); if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA) dev->z_port_xmit_data = swords; diff --git a/drivers/infiniband/hw/ipath/ipath_mr.c b/drivers/infiniband/hw/ipath/ipath_mr.c index 4ac31a5da33..b36f6fb3e37 100644 --- a/drivers/infiniband/hw/ipath/ipath_mr.c +++ b/drivers/infiniband/hw/ipath/ipath_mr.c @@ -36,6 +36,18 @@ #include "ipath_verbs.h" +/* Fast memory region */ +struct ipath_fmr { + struct ib_fmr ibfmr; + u8 page_shift; + struct ipath_mregion mr; /* must be last */ +}; + +static inline struct ipath_fmr *to_ifmr(struct ib_fmr *ibfmr) +{ + return container_of(ibfmr, struct ipath_fmr, ibfmr); +} + /** * ipath_get_dma_mr - get a DMA memory region * @pd: protection domain for this memory region diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c index 1ccfc909db1..9228d59b59d 100644 --- a/drivers/infiniband/hw/ipath/ipath_qp.c +++ b/drivers/infiniband/hw/ipath/ipath_qp.c @@ -461,7 +461,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, goto inval; if (attr_mask & IB_QP_PKEY_INDEX) - if (attr->pkey_index >= ipath_layer_get_npkeys(dev->dd)) + if (attr->pkey_index >= ipath_get_npkeys(dev->dd)) goto inval; if (attr_mask & IB_QP_MIN_RNR_TIMER) @@ -644,6 +644,33 @@ __be32 ipath_compute_aeth(struct ipath_qp *qp) return cpu_to_be32(aeth); } +/** + * set_verbs_flags - set the verbs layer flags + * @dd: the infinipath device + * @flags: the flags to set + */ +static int set_verbs_flags(struct ipath_devdata *dd, unsigned flags) +{ + struct ipath_devdata *ss; + unsigned long lflags; + + spin_lock_irqsave(&ipath_devs_lock, lflags); + + list_for_each_entry(ss, &ipath_dev_list, ipath_list) { + if (!(ss->ipath_flags & IPATH_INITTED)) + continue; + if ((flags & IPATH_VERBS_KERNEL_SMA) && + !(*ss->ipath_statusp & IPATH_STATUS_SMA)) + *ss->ipath_statusp |= IPATH_STATUS_OIB_SMA; + else + *ss->ipath_statusp &= ~IPATH_STATUS_OIB_SMA; + } + + spin_unlock_irqrestore(&ipath_devs_lock, lflags); + + return 0; +} + /** * ipath_create_qp - create a queue pair for a device * @ibpd: the protection domain who's device we create the queue pair for @@ -760,8 +787,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, /* Tell the core driver that the kernel SMA is present. */ if (init_attr->qp_type == IB_QPT_SMI) - ipath_layer_set_verbs_flags(dev->dd, - IPATH_VERBS_KERNEL_SMA); + set_verbs_flags(dev->dd, IPATH_VERBS_KERNEL_SMA); break; default: @@ -838,7 +864,7 @@ int ipath_destroy_qp(struct ib_qp *ibqp) /* Tell the core driver that the kernel SMA is gone. */ if (qp->ibqp.qp_type == IB_QPT_SMI) - ipath_layer_set_verbs_flags(dev->dd, 0); + set_verbs_flags(dev->dd, 0); spin_lock_irqsave(&qp->s_lock, flags); qp->state = IB_QPS_ERR; diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c index 774d1615ce2..a08654042c0 100644 --- a/drivers/infiniband/hw/ipath/ipath_rc.c +++ b/drivers/infiniband/hw/ipath/ipath_rc.c @@ -32,7 +32,7 @@ */ #include "ipath_verbs.h" -#include "ipath_common.h" +#include "ipath_kernel.h" /* cut down ridiculously long IB macro names */ #define OP(x) IB_OPCODE_RC_##x @@ -540,7 +540,7 @@ static void send_rc_ack(struct ipath_qp *qp) lrh0 = IPATH_LRH_GRH; } /* read pkey_index w/o lock (its atomic) */ - bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index); + bth0 = ipath_get_pkey(dev->dd, qp->s_pkey_index); if (qp->r_nak_state) ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) | (qp->r_nak_state << @@ -557,7 +557,7 @@ 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(ipath_layer_get_lid(dev->dd)); + hdr.lrh[3] = cpu_to_be16(dev->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); @@ -1323,8 +1323,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, * the eager header buffer size to 56 bytes so the last 4 * bytes of the BTH header (PSN) is in the data buffer. */ - header_in_data = - ipath_layer_get_rcvhdrentsize(dev->dd) == 16; + header_in_data = dev->dd->ipath_rcvhdrentsize == 16; if (header_in_data) { psn = be32_to_cpu(((__be32 *) data)[0]); data += sizeof(__be32); diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c index dd09420d677..5c1da2d25e0 100644 --- a/drivers/infiniband/hw/ipath/ipath_ruc.c +++ b/drivers/infiniband/hw/ipath/ipath_ruc.c @@ -470,6 +470,15 @@ done: wake_up(&qp->wait); } +static int want_buffer(struct ipath_devdata *dd) +{ + set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl); + ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, + dd->ipath_sendctrl); + + return 0; +} + /** * ipath_no_bufs_available - tell the layer driver we need buffers * @qp: the QP that caused the problem @@ -486,7 +495,7 @@ void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev) list_add_tail(&qp->piowait, &dev->piowait); spin_unlock_irqrestore(&dev->pending_lock, flags); /* - * Note that as soon as ipath_layer_want_buffer() is called and + * Note that as soon as want_buffer() is called and * possibly before it returns, ipath_ib_piobufavail() * could be called. If we are still in the tasklet function, * tasklet_hi_schedule() will not call us until the next time @@ -496,7 +505,7 @@ void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev) */ clear_bit(IPATH_S_BUSY, &qp->s_flags); tasklet_unlock(&qp->s_task); - ipath_layer_want_buffer(dev->dd); + want_buffer(dev->dd); dev->n_piowait++; } @@ -611,7 +620,7 @@ u32 ipath_make_grh(struct ipath_ibdev *dev, struct ib_grh *hdr, hdr->hop_limit = grh->hop_limit; /* The SGID is 32-bit aligned. */ hdr->sgid.global.subnet_prefix = dev->gid_prefix; - hdr->sgid.global.interface_id = ipath_layer_get_guid(dev->dd); + hdr->sgid.global.interface_id = dev->dd->ipath_guid; hdr->dgid = grh->dgid; /* GRH header size in 32-bit words. */ @@ -643,8 +652,7 @@ void ipath_do_ruc_send(unsigned long data) if (test_and_set_bit(IPATH_S_BUSY, &qp->s_flags)) goto bail; - if (unlikely(qp->remote_ah_attr.dlid == - ipath_layer_get_lid(dev->dd))) { + if (unlikely(qp->remote_ah_attr.dlid == dev->dd->ipath_lid)) { ipath_ruc_loopback(qp); goto clear; } @@ -711,8 +719,8 @@ again: qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid); qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC); - qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd)); - bth0 |= ipath_layer_get_pkey(dev->dd, qp->s_pkey_index); + qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid); + bth0 |= ipath_get_pkey(dev->dd, qp->s_pkey_index); bth0 |= extra_bytes << 20; ohdr->bth[0] = cpu_to_be32(bth0); ohdr->bth[1] = cpu_to_be32(qp->remote_qpn); diff --git a/drivers/infiniband/hw/ipath/ipath_sysfs.c b/drivers/infiniband/hw/ipath/ipath_sysfs.c index b98821d7801..7396a63840d 100644 --- a/drivers/infiniband/hw/ipath/ipath_sysfs.c +++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c @@ -35,7 +35,6 @@ #include #include "ipath_kernel.h" -#include "ipath_layer.h" #include "ipath_common.h" /** @@ -227,7 +226,6 @@ static ssize_t store_mlid(struct device *dev, unit = dd->ipath_unit; dd->ipath_mlid = mlid; - ipath_layer_intr(dd, IPATH_LAYER_INT_BCAST); goto bail; invalid: @@ -467,7 +465,7 @@ static ssize_t store_link_state(struct device *dev, if (ret < 0) goto invalid; - r = ipath_layer_set_linkstate(dd, state); + r = ipath_set_linkstate(dd, state); if (r < 0) { ret = r; goto bail; @@ -502,7 +500,7 @@ static ssize_t store_mtu(struct device *dev, if (ret < 0) goto invalid; - r = ipath_layer_set_mtu(dd, mtu); + r = ipath_set_mtu(dd, mtu); if (r < 0) ret = r; diff --git a/drivers/infiniband/hw/ipath/ipath_uc.c b/drivers/infiniband/hw/ipath/ipath_uc.c index c33abea2d5a..0fd3cded16b 100644 --- a/drivers/infiniband/hw/ipath/ipath_uc.c +++ b/drivers/infiniband/hw/ipath/ipath_uc.c @@ -32,7 +32,7 @@ */ #include "ipath_verbs.h" -#include "ipath_common.h" +#include "ipath_kernel.h" /* cut down ridiculously long IB macro names */ #define OP(x) IB_OPCODE_UC_##x @@ -261,8 +261,7 @@ void ipath_uc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, * size to 56 bytes so the last 4 bytes of * the BTH header (PSN) is in the data buffer. */ - header_in_data = - ipath_layer_get_rcvhdrentsize(dev->dd) == 16; + header_in_data = dev->dd->ipath_rcvhdrentsize == 16; if (header_in_data) { psn = be32_to_cpu(((__be32 *) data)[0]); data += sizeof(__be32); diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c index 82439fcfc2f..6991d1d74e3 100644 --- a/drivers/infiniband/hw/ipath/ipath_ud.c +++ b/drivers/infiniband/hw/ipath/ipath_ud.c @@ -353,7 +353,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr) ss.num_sge++; } /* Check for invalid packet size. */ - if (len > ipath_layer_get_ibmtu(dev->dd)) { + if (len > dev->dd->ipath_ibmtu) { ret = -EINVAL; goto bail; } @@ -375,7 +375,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr) dev->n_unicast_xmit++; lid = ah_attr->dlid & ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1); - if (unlikely(lid == ipath_layer_get_lid(dev->dd))) { + if (unlikely(lid == dev->dd->ipath_lid)) { /* * Pass in an uninitialized ib_wc to save stack * space. @@ -404,7 +404,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr) qp->s_hdr.u.l.grh.sgid.global.subnet_prefix = dev->gid_prefix; qp->s_hdr.u.l.grh.sgid.global.interface_id = - ipath_layer_get_guid(dev->dd); + dev->dd->ipath_guid; qp->s_hdr.u.l.grh.dgid = ah_attr->grh.dgid; /* * Don't worry about sending to locally attached multicast @@ -434,7 +434,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr) qp->s_hdr.lrh[0] = cpu_to_be16(lrh0); qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid); /* DEST LID */ qp->s_hdr.lrh[2] = cpu_to_be16(hwords + nwords + SIZE_OF_CRC); - lid = ipath_layer_get_lid(dev->dd); + lid = dev->dd->ipath_lid; if (lid) { lid |= ah_attr->src_path_bits & ((1 << (dev->mkeyprot_resv_lmc & 7)) - 1); @@ -445,7 +445,7 @@ int ipath_post_ud_send(struct ipath_qp *qp, struct ib_send_wr *wr) bth0 |= 1 << 23; bth0 |= extra_bytes << 20; bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? IPATH_DEFAULT_P_KEY : - ipath_layer_get_pkey(dev->dd, qp->s_pkey_index); + ipath_get_pkey(dev->dd, qp->s_pkey_index); ohdr->bth[0] = cpu_to_be32(bth0); /* * Use the multicast QP if the destination LID is a multicast LID. @@ -531,8 +531,7 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr, * the eager header buffer size to 56 bytes so the last 12 * bytes of the IB header is in the data buffer. */ - header_in_data = - ipath_layer_get_rcvhdrentsize(dev->dd) == 16; + header_in_data = dev->dd->ipath_rcvhdrentsize == 16; if (header_in_data) { qkey = be32_to_cpu(((__be32 *) data)[1]); src_qp = be32_to_cpu(((__be32 *) data)[2]); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index 15edec9227e..3c47620e988 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -33,15 +33,13 @@ #include #include +#include #include #include "ipath_kernel.h" #include "ipath_verbs.h" #include "ipath_common.h" -/* Not static, because we don't want the compiler removing it */ -const char ipath_verbs_version[] = "ipath_verbs " IPATH_IDSTR; - static unsigned int ib_ipath_qp_table_size = 251; module_param_named(qp_table_size, ib_ipath_qp_table_size, uint, S_IRUGO); MODULE_PARM_DESC(qp_table_size, "QP table size"); @@ -109,10 +107,6 @@ module_param_named(max_srq_wrs, ib_ipath_max_srq_wrs, uint, S_IWUSR | S_IRUGO); MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support"); -MODULE_LICENSE("GPL"); -MODULE_AUTHOR("QLogic "); -MODULE_DESCRIPTION("QLogic InfiniPath driver"); - const int ib_ipath_state_ops[IB_QPS_ERR + 1] = { [IB_QPS_RESET] = 0, [IB_QPS_INIT] = IPATH_POST_RECV_OK, @@ -125,6 +119,16 @@ const int ib_ipath_state_ops[IB_QPS_ERR + 1] = { [IB_QPS_ERR] = 0, }; +struct ipath_ucontext { + struct ib_ucontext ibucontext; +}; + +static inline struct ipath_ucontext *to_iucontext(struct ib_ucontext + *ibucontext) +{ + return container_of(ibucontext, struct ipath_ucontext, ibucontext); +} + /* * Translate ib_wr_opcode into ib_wc_opcode. */ @@ -400,7 +404,7 @@ void ipath_ib_rcv(struct ipath_ibdev *dev, void *rhdr, void *data, lid = be16_to_cpu(hdr->lrh[1]); if (lid < IPATH_MULTICAST_LID_BASE) { lid &= ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1); - if (unlikely(lid != ipath_layer_get_lid(dev->dd))) { + if (unlikely(lid != dev->dd->ipath_lid)) { dev->rcv_errors++; goto bail; } @@ -511,19 +515,19 @@ void ipath_ib_timer(struct ipath_ibdev *dev) if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_STARTED && --dev->pma_sample_start == 0) { dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING; - ipath_layer_snapshot_counters(dev->dd, &dev->ipath_sword, - &dev->ipath_rword, - &dev->ipath_spkts, - &dev->ipath_rpkts, - &dev->ipath_xmit_wait); + ipath_snapshot_counters(dev->dd, &dev->ipath_sword, + &dev->ipath_rword, + &dev->ipath_spkts, + &dev->ipath_rpkts, + &dev->ipath_xmit_wait); } if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_RUNNING) { if (dev->pma_sample_interval == 0) { u64 ta, tb, tc, td, te; dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE; - ipath_layer_snapshot_counters(dev->dd, &ta, &tb, - &tc, &td, &te); + ipath_snapshot_counters(dev->dd, &ta, &tb, + &tc, &td, &te); dev->ipath_sword = ta - dev->ipath_sword; dev->ipath_rword = tb - dev->ipath_rword; @@ -553,6 +557,362 @@ void ipath_ib_timer(struct ipath_ibdev *dev) } } +static void update_sge(struct ipath_sge_state *ss, u32 length) +{ + struct ipath_sge *sge = &ss->sge; + + sge->vaddr += length; + sge->length -= length; + sge->sge_length -= length; + 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) + return; + 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; + } +} + +#ifdef __LITTLE_ENDIAN +static inline u32 get_upper_bits(u32 data, u32 shift) +{ + return data >> shift; +} + +static inline u32 set_upper_bits(u32 data, u32 shift) +{ + return data << shift; +} + +static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off) +{ + data <<= ((sizeof(u32) - n) * BITS_PER_BYTE); + data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE); + return data; +} +#else +static inline u32 get_upper_bits(u32 data, u32 shift) +{ + return data << shift; +} + +static inline u32 set_upper_bits(u32 data, u32 shift) +{ + return data >> shift; +} + +static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off) +{ + data >>= ((sizeof(u32) - n) * BITS_PER_BYTE); + data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE); + return data; +} +#endif + +static void copy_io(u32 __iomem *piobuf, struct ipath_sge_state *ss, + u32 length) +{ + u32 extra = 0; + u32 data = 0; + u32 last; + + while (1) { + u32 len = ss->sge.length; + u32 off; + + BUG_ON(len == 0); + if (len > length) + len = length; + if (len > ss->sge.sge_length) + len = ss->sge.sge_length; + /* If the source address is not aligned, try to align it. */ + off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1); + if (off) { + u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr & + ~(sizeof(u32) - 1)); + u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE); + u32 y; + + y = sizeof(u32) - off; + if (len > y) + len = y; + if (len + extra >= sizeof(u32)) { + data |= set_upper_bits(v, extra * + BITS_PER_BYTE); + len = sizeof(u32) - extra; + if (len == length) { + last = data; + break; + } + __raw_writel(data, piobuf); + piobuf++; + extra = 0; + data = 0; + } else { + /* Clear unused upper bytes */ + data |= clear_upper_bytes(v, len, extra); + if (len == length) { + last = data; + break; + } + extra += len; + } + } else if (extra) { + /* Source address is aligned. */ + u32 *addr = (u32 *) ss->sge.vaddr; + int shift = extra * BITS_PER_BYTE; + int ushift = 32 - shift; + u32 l = len; + + while (l >= sizeof(u32)) { + u32 v = *addr; + + data |= set_upper_bits(v, shift); + __raw_writel(data, piobuf); + data = get_upper_bits(v, ushift); + piobuf++; + addr++; + l -= sizeof(u32); + } + /* + * We still have 'extra' number of bytes leftover. + */ + if (l) { + u32 v = *addr; + + if (l + extra >= sizeof(u32)) { + data |= set_upper_bits(v, shift); + len -= l + extra - sizeof(u32); + if (len == length) { + last = data; + break; + } + __raw_writel(data, piobuf); + piobuf++; + extra = 0; + data = 0; + } else { + /* Clear unused upper bytes */ + data |= clear_upper_bytes(v, l, + extra); + if (len == length) { + last = data; + break; + } + extra += l; + } + } else if (len == length) { + last = data; + break; + } + } else if (len == length) { + u32 w; + + /* + * Need to round up for the last dword in the + * packet. + */ + w = (len + 3) >> 2; + __iowrite32_copy(piobuf, ss->sge.vaddr, w - 1); + piobuf += w - 1; + last = ((u32 *) ss->sge.vaddr)[w - 1]; + break; + } else { + u32 w = len >> 2; + + __iowrite32_copy(piobuf, ss->sge.vaddr, w); + piobuf += w; + + extra = len & (sizeof(u32) - 1); + if (extra) { + u32 v = ((u32 *) ss->sge.vaddr)[w]; + + /* Clear unused upper bytes */ + data = clear_upper_bytes(v, extra, 0); + } + } + update_sge(ss, len); + length -= len; + } + /* Update address before sending packet. */ + update_sge(ss, length); + /* must flush early everything before trigger word */ + ipath_flush_wc(); + __raw_writel(last, piobuf); + /* be sure trigger word is written */ + ipath_flush_wc(); +} + +/** + * ipath_verbs_send - send a packet + * @dd: the infinipath device + * @hdrwords: the number of words in the header + * @hdr: the packet header + * @len: the length of the packet in bytes + * @ss: the SGE to send + */ +int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords, + u32 *hdr, u32 len, struct ipath_sge_state *ss) +{ + u32 __iomem *piobuf; + u32 plen; + int ret; + + /* +1 is for the qword padding of pbc */ + plen = hdrwords + ((len + 3) >> 2) + 1; + if (unlikely((plen << 2) > dd->ipath_ibmaxlen)) { + ipath_dbg("packet len 0x%x too long, failing\n", plen); + ret = -EINVAL; + goto bail; + } + + /* Get a PIO buffer to use. */ + piobuf = ipath_getpiobuf(dd, NULL); + if (unlikely(piobuf == NULL)) { + ret = -EBUSY; + goto bail; + } + + /* + * Write len to control qword, no 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); + ipath_flush_wc(); + piobuf += 2; + if (len == 0) { + /* + * If there is just the header portion, must flush before + * writing last word of header for correctness, and after + * the last header word (trigger word). + */ + __iowrite32_copy(piobuf, hdr, hdrwords - 1); + ipath_flush_wc(); + __raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1); + ipath_flush_wc(); + ret = 0; + goto bail; + } + + __iowrite32_copy(piobuf, hdr, hdrwords); + piobuf += hdrwords; + + /* The common case is aligned and contained in one segment. */ + if (likely(ss->num_sge == 1 && len <= ss->sge.length && + !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) { + u32 w; + u32 *addr = (u32 *) ss->sge.vaddr; + + /* Update address before sending packet. */ + update_sge(ss, len); + /* Need to round up for the last dword in the packet. */ + w = (len + 3) >> 2; + __iowrite32_copy(piobuf, addr, w - 1); + /* must flush early everything before trigger word */ + ipath_flush_wc(); + __raw_writel(addr[w - 1], piobuf + w - 1); + /* be sure trigger word is written */ + ipath_flush_wc(); + ret = 0; + goto bail; + } + copy_io(piobuf, ss, len); + ret = 0; + +bail: + return ret; +} + +int ipath_snapshot_counters(struct ipath_devdata *dd, u64 *swords, + u64 *rwords, u64 *spkts, u64 *rpkts, + u64 *xmit_wait) +{ + int ret; + + if (!(dd->ipath_flags & IPATH_INITTED)) { + /* no hardware, freeze, etc. */ + ipath_dbg("unit %u not usable\n", dd->ipath_unit); + ret = -EINVAL; + goto bail; + } + *swords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt); + *rwords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt); + *spkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt); + *rpkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt); + *xmit_wait = ipath_snap_cntr(dd, dd->ipath_cregs->cr_sendstallcnt); + + ret = 0; + +bail: + return ret; +} + +/** + * ipath_get_counters - get various chip counters + * @dd: the infinipath device + * @cntrs: counters are placed here + * + * Return the counters needed by recv_pma_get_portcounters(). + */ +int ipath_get_counters(struct ipath_devdata *dd, + struct ipath_verbs_counters *cntrs) +{ + int ret; + + if (!(dd->ipath_flags & IPATH_INITTED)) { + /* no hardware, freeze, etc. */ + ipath_dbg("unit %u not usable\n", dd->ipath_unit); + ret = -EINVAL; + goto bail; + } + cntrs->symbol_error_counter = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_ibsymbolerrcnt); + cntrs->link_error_recovery_counter = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkerrrecovcnt); + /* + * The link downed counter counts when the other side downs the + * connection. We add in the number of times we downed the link + * due to local link integrity errors to compensate. + */ + cntrs->link_downed_counter = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkdowncnt); + cntrs->port_rcv_errors = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_rxdroppktcnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvovflcnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_portovflcnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_err_rlencnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_invalidrlencnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) + + ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt); + cntrs->port_rcv_remphys_errors = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt); + cntrs->port_xmit_discards = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_unsupvlcnt); + cntrs->port_xmit_data = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt); + cntrs->port_rcv_data = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt); + cntrs->port_xmit_packets = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt); + cntrs->port_rcv_packets = + ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt); + cntrs->local_link_integrity_errors = dd->ipath_lli_errors; + cntrs->excessive_buffer_overrun_errors = 0; /* XXX */ + + ret = 0; + +bail: + return ret; +} + /** * ipath_ib_piobufavail - callback when a PIO buffer is available * @arg: the device pointer @@ -595,9 +955,9 @@ static int ipath_query_device(struct ib_device *ibdev, IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT | IB_DEVICE_SYS_IMAGE_GUID; props->page_size_cap = PAGE_SIZE; - props->vendor_id = ipath_layer_get_vendorid(dev->dd); - props->vendor_part_id = ipath_layer_get_deviceid(dev->dd); - props->hw_ver = ipath_layer_get_pcirev(dev->dd); + props->vendor_id = dev->dd->ipath_vendorid; + props->vendor_part_id = dev->dd->ipath_deviceid; + props->hw_ver = dev->dd->ipath_pcirev; props->sys_image_guid = dev->sys_image_guid; @@ -618,7 +978,7 @@ static int ipath_query_device(struct ib_device *ibdev, props->max_srq_sge = ib_ipath_max_srq_sges; /* props->local_ca_ack_delay */ props->atomic_cap = IB_ATOMIC_HCA; - props->max_pkeys = ipath_layer_get_npkeys(dev->dd); + props->max_pkeys = ipath_get_npkeys(dev->dd); props->max_mcast_grp = ib_ipath_max_mcast_grps; props->max_mcast_qp_attach = ib_ipath_max_mcast_qp_attached; props->max_total_mcast_qp_attach = props->max_mcast_qp_attach * @@ -643,12 +1003,17 @@ const u8 ipath_cvt_physportstate[16] = { [INFINIPATH_IBCS_LT_STATE_RECOVERIDLE] = 6, }; +u32 ipath_get_cr_errpkey(struct ipath_devdata *dd) +{ + return ipath_read_creg32(dd, dd->ipath_cregs->cr_errpkey); +} + static int ipath_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *props) { struct ipath_ibdev *dev = to_idev(ibdev); enum ib_mtu mtu; - u16 lid = ipath_layer_get_lid(dev->dd); + u16 lid = dev->dd->ipath_lid; u64 ibcstat; memset(props, 0, sizeof(*props)); @@ -656,16 +1021,16 @@ static int ipath_query_port(struct ib_device *ibdev, props->lmc = dev->mkeyprot_resv_lmc & 7; props->sm_lid = dev->sm_lid; props->sm_sl = dev->sm_sl; - ibcstat = ipath_layer_get_lastibcstat(dev->dd); + ibcstat = dev->dd->ipath_lastibcstat; props->state = ((ibcstat >> 4) & 0x3) + 1; /* See phys_state_show() */ props->phys_state = ipath_cvt_physportstate[ - ipath_layer_get_lastibcstat(dev->dd) & 0xf]; + dev->dd->ipath_lastibcstat & 0xf]; props->port_cap_flags = dev->port_cap_flags; props->gid_tbl_len = 1; props->max_msg_sz = 0x80000000; - props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd); - props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) - + props->pkey_tbl_len = ipath_get_npkeys(dev->dd); + props->bad_pkey_cntr = ipath_get_cr_errpkey(dev->dd) - dev->z_pkey_violations; props->qkey_viol_cntr = dev->qkey_violations; props->active_width = IB_WIDTH_4X; @@ -675,7 +1040,7 @@ static int ipath_query_port(struct ib_device *ibdev, props->init_type_reply = 0; props->max_mtu = IB_MTU_4096; - switch (ipath_layer_get_ibmtu(dev->dd)) { + switch (dev->dd->ipath_ibmtu) { case 4096: mtu = IB_MTU_4096; break; @@ -734,7 +1099,7 @@ static int ipath_modify_port(struct ib_device *ibdev, dev->port_cap_flags |= props->set_port_cap_mask; dev->port_cap_flags &= ~props->clr_port_cap_mask; if (port_modify_mask & IB_PORT_SHUTDOWN) - ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKDOWN); + ipath_set_linkstate(dev->dd, IPATH_IB_LINKDOWN); if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR) dev->qkey_violations = 0; return 0; @@ -751,7 +1116,7 @@ static int ipath_query_gid(struct ib_device *ibdev, u8 port, goto bail; } gid->global.subnet_prefix = dev->gid_prefix; - gid->global.interface_id = ipath_layer_get_guid(dev->dd); + gid->global.interface_id = dev->dd->ipath_guid; ret = 0; @@ -902,25 +1267,50 @@ static int ipath_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr) return 0; } +/** + * ipath_get_npkeys - return the size of the PKEY table for port 0 + * @dd: the infinipath device + */ +unsigned ipath_get_npkeys(struct ipath_devdata *dd) +{ + return ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys); +} + +/** + * ipath_get_pkey - return the indexed PKEY from the port 0 PKEY table + * @dd: the infinipath device + * @index: the PKEY index + */ +unsigned ipath_get_pkey(struct ipath_devdata *dd, unsigned index) +{ + unsigned ret; + + if (index >= ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys)) + ret = 0; + else + ret = dd->ipath_pd[0]->port_pkeys[index]; + + return ret; +} + static int ipath_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) { struct ipath_ibdev *dev = to_idev(ibdev); int ret; - if (index >= ipath_layer_get_npkeys(dev->dd)) { + if (index >= ipath_get_npkeys(dev->dd)) { ret = -EINVAL; goto bail; } - *pkey = ipath_layer_get_pkey(dev->dd, index); + *pkey = ipath_get_pkey(dev->dd, index); ret = 0; bail: return ret; } - /** * ipath_alloc_ucontext - allocate a ucontest * @ibdev: the infiniband device @@ -953,6 +1343,63 @@ static int ipath_dealloc_ucontext(struct ib_ucontext *context) static int ipath_verbs_register_sysfs(struct ib_device *dev); +static void __verbs_timer(unsigned long arg) +{ + struct ipath_devdata *dd = (struct ipath_devdata *) arg; + + /* + * If port 0 receive packet interrupts are not available, or + * can be missed, poll the receive queue + */ + if (dd->ipath_flags & IPATH_POLL_RX_INTR) + ipath_kreceive(dd); + + /* Handle verbs layer timeouts. */ + ipath_ib_timer(dd->verbs_dev); + + mod_timer(&dd->verbs_timer, jiffies + 1); +} + +static int enable_timer(struct ipath_devdata *dd) +{ + /* + * Early chips had a design flaw where the chip and kernel idea + * of the tail register don't always agree, and therefore we won't + * get an interrupt on the next packet received. + * If the board supports per packet receive interrupts, use it. + * Otherwise, the timer function periodically checks for packets + * to cover this case. + * Either way, the timer is needed for verbs layer related + * processing. + */ + if (dd->ipath_flags & IPATH_GPIO_INTR) { + ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect, + 0x2074076542310ULL); + /* Enable GPIO bit 2 interrupt */ + ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, + (u64) (1 << 2)); + } + + init_timer(&dd->verbs_timer); + dd->verbs_timer.function = __verbs_timer; + dd->verbs_timer.data = (unsigned long)dd; + dd->verbs_timer.expires = jiffies + 1; + add_timer(&dd->verbs_timer); + + return 0; +} + +static int disable_timer(struct ipath_devdata *dd) +{ + /* Disable GPIO bit 2 interrupt */ + if (dd->ipath_flags & IPATH_GPIO_INTR) + ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 0); + + del_timer_sync(&dd->verbs_timer); + + return 0; +} + /** * ipath_register_ib_device - register our device with the infiniband core * @dd: the device data structure @@ -960,7 +1407,7 @@ static int ipath_verbs_register_sysfs(struct ib_device *dev); */ int ipath_register_ib_device(struct ipath_devdata *dd) { - struct ipath_layer_counters cntrs; + struct ipath_verbs_counters cntrs; struct ipath_ibdev *idev; struct ib_device *dev; int ret; @@ -1020,7 +1467,7 @@ int ipath_register_ib_device(struct ipath_devdata *dd) idev->link_width_enabled = 3; /* 1x or 4x */ /* Snapshot current HW counters to "clear" them. */ - ipath_layer_get_counters(dd, &cntrs); + ipath_get_counters(dd, &cntrs); idev->z_symbol_error_counter = cntrs.symbol_error_counter; idev->z_link_error_recovery_counter = cntrs.link_error_recovery_counter; @@ -1044,14 +1491,14 @@ int ipath_register_ib_device(struct ipath_devdata *dd) * device types in the system, we can't be sure this is unique. */ if (!sys_image_guid) - sys_image_guid = ipath_layer_get_guid(dd); + sys_image_guid = dd->ipath_guid; idev->sys_image_guid = sys_image_guid; idev->ib_unit = dd->ipath_unit; idev->dd = dd; strlcpy(dev->name, "ipath%d", IB_DEVICE_NAME_MAX); dev->owner = THIS_MODULE; - dev->node_guid = ipath_layer_get_guid(dd); + dev->node_guid = dd->ipath_guid; dev->uverbs_abi_ver = IPATH_UVERBS_ABI_VERSION; dev->uverbs_cmd_mask = (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | @@ -1085,7 +1532,7 @@ int ipath_register_ib_device(struct ipath_devdata *dd) (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV); dev->node_type = IB_NODE_CA; dev->phys_port_cnt = 1; - dev->dma_device = ipath_layer_get_device(dd); + 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; @@ -1139,7 +1586,7 @@ int ipath_register_ib_device(struct ipath_devdata *dd) if (ipath_verbs_register_sysfs(dev)) goto err_class; - ipath_layer_enable_timer(dd); + enable_timer(dd); goto bail; @@ -1164,7 +1611,7 @@ void ipath_unregister_ib_device(struct ipath_ibdev *dev) { struct ib_device *ibdev = &dev->ibdev; - ipath_layer_disable_timer(dev->dd); + disable_timer(dev->dd); ib_unregister_device(ibdev); @@ -1197,7 +1644,7 @@ static ssize_t show_rev(struct class_device *cdev, char *buf) struct ipath_ibdev *dev = container_of(cdev, struct ipath_ibdev, ibdev.class_dev); - return sprintf(buf, "%x\n", ipath_layer_get_pcirev(dev->dd)); + return sprintf(buf, "%x\n", dev->dd->ipath_pcirev); } static ssize_t show_hca(struct class_device *cdev, char *buf) @@ -1206,7 +1653,7 @@ static ssize_t show_hca(struct class_device *cdev, char *buf) container_of(cdev, struct ipath_ibdev, ibdev.class_dev); int ret; - ret = ipath_layer_get_boardname(dev->dd, buf, 128); + ret = dev->dd->ipath_f_get_boardname(dev->dd, buf, 128); if (ret < 0) goto bail; strcat(buf, "\n"); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index d6faa4ba606..00f4cecc258 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -153,19 +153,6 @@ struct ipath_mcast { int n_attached; }; -/* Memory region */ -struct ipath_mr { - struct ib_mr ibmr; - struct ipath_mregion mr; /* must be last */ -}; - -/* Fast memory region */ -struct ipath_fmr { - struct ib_fmr ibfmr; - u8 page_shift; - struct ipath_mregion mr; /* must be last */ -}; - /* Protection domain */ struct ipath_pd { struct ib_pd ibpd; @@ -216,6 +203,54 @@ struct ipath_cq { struct ipath_mmap_info *ip; }; +/* + * A segment is a linear region of low physical memory. + * XXX Maybe we should use phys addr here and kmap()/kunmap(). + * Used by the verbs layer. + */ +struct ipath_seg { + void *vaddr; + size_t length; +}; + +/* The number of ipath_segs that fit in a page. */ +#define IPATH_SEGSZ (PAGE_SIZE / sizeof (struct ipath_seg)) + +struct ipath_segarray { + struct ipath_seg segs[IPATH_SEGSZ]; +}; + +struct ipath_mregion { + u64 user_base; /* User's address for this region */ + u64 iova; /* IB start address of this region */ + size_t length; + u32 lkey; + u32 offset; /* offset (bytes) to start of region */ + int access_flags; + u32 max_segs; /* number of ipath_segs in all the arrays */ + u32 mapsz; /* size of the map array */ + struct ipath_segarray *map[0]; /* the segments */ +}; + +/* + * These keep track of the copy progress within a memory region. + * Used by the verbs layer. + */ +struct ipath_sge { + struct ipath_mregion *mr; + void *vaddr; /* current pointer into the segment */ + u32 sge_length; /* length of the SGE */ + u32 length; /* remaining length of the segment */ + u16 m; /* current index: mr->map[m] */ + u16 n; /* current index: mr->map[m]->segs[n] */ +}; + +/* Memory region */ +struct ipath_mr { + struct ib_mr ibmr; + struct ipath_mregion mr; /* must be last */ +}; + /* * Send work request queue entry. * The size of the sg_list is determined when the QP is created and stored @@ -270,6 +305,12 @@ struct ipath_srq { u32 limit; }; +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; +}; + /* * Variables prefixed with s_ are for the requester (sender). * Variables prefixed with r_ are for the responder (receiver). @@ -500,8 +541,19 @@ struct ipath_ibdev { struct ipath_opcode_stats opstats[128]; }; -struct ipath_ucontext { - struct ib_ucontext ibucontext; +struct ipath_verbs_counters { + u64 symbol_error_counter; + u64 link_error_recovery_counter; + u64 link_downed_counter; + u64 port_rcv_errors; + u64 port_rcv_remphys_errors; + u64 port_xmit_discards; + u64 port_xmit_data; + u64 port_rcv_data; + u64 port_xmit_packets; + u64 port_rcv_packets; + u32 local_link_integrity_errors; + u32 excessive_buffer_overrun_errors; }; static inline struct ipath_mr *to_imr(struct ib_mr *ibmr) @@ -509,11 +561,6 @@ static inline struct ipath_mr *to_imr(struct ib_mr *ibmr) return container_of(ibmr, struct ipath_mr, ibmr); } -static inline struct ipath_fmr *to_ifmr(struct ib_fmr *ibfmr) -{ - return container_of(ibfmr, struct ipath_fmr, ibfmr); -} - static inline struct ipath_pd *to_ipd(struct ib_pd *ibpd) { return container_of(ibpd, struct ipath_pd, ibpd); @@ -551,12 +598,6 @@ int ipath_process_mad(struct ib_device *ibdev, struct ib_grh *in_grh, struct ib_mad *in_mad, struct ib_mad *out_mad); -static inline struct ipath_ucontext *to_iucontext(struct ib_ucontext - *ibucontext) -{ - return container_of(ibucontext, struct ipath_ucontext, ibucontext); -} - /* * Compare the lower 24 bits of the two values. * Returns an integer <, ==, or > than zero. @@ -568,6 +609,13 @@ static inline int ipath_cmp24(u32 a, u32 b) struct ipath_mcast *ipath_mcast_find(union ib_gid *mgid); +int ipath_snapshot_counters(struct ipath_devdata *dd, u64 *swords, + u64 *rwords, u64 *spkts, u64 *rpkts, + u64 *xmit_wait); + +int ipath_get_counters(struct ipath_devdata *dd, + struct ipath_verbs_counters *cntrs); + int ipath_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid); int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid); @@ -598,6 +646,9 @@ void ipath_sqerror_qp(struct ipath_qp *qp, struct ib_wc *wc); void ipath_get_credit(struct ipath_qp *qp, u32 aeth); +int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords, + u32 *hdr, u32 len, struct ipath_sge_state *ss); + void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int sig); int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss, @@ -721,6 +772,12 @@ int ipath_ib_piobufavail(struct ipath_ibdev *); void ipath_ib_timer(struct ipath_ibdev *); +unsigned ipath_get_npkeys(struct ipath_devdata *); + +u32 ipath_get_cr_errpkey(struct ipath_devdata *); + +unsigned ipath_get_pkey(struct ipath_devdata *, unsigned); + extern const enum ib_wc_opcode ib_ipath_wc_opcode[]; extern const u8 ipath_cvt_physportstate[]; -- cgit v1.2.3-70-g09d2 From b55f4f06c834a67f949a5219c5f97ffafa240989 Mon Sep 17 00:00:00 2001 From: Bryan O'Sullivan Date: Fri, 25 Aug 2006 11:24:33 -0700 Subject: IB/ipath: simplify debugging code after ipath_core and ib_ipath merger Signed-off-by: Bryan O'Sullivan Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_driver.c | 4 +- drivers/infiniband/hw/ipath/ipath_kernel.h | 4 +- drivers/infiniband/hw/ipath/ipath_keys.c | 3 +- drivers/infiniband/hw/ipath/ipath_qp.c | 10 +-- drivers/infiniband/hw/ipath/ipath_sysfs.c | 2 +- drivers/infiniband/hw/ipath/ipath_verbs.c | 19 ++--- drivers/infiniband/hw/ipath/ipath_verbs.h | 1 - drivers/infiniband/hw/ipath/verbs_debug.h | 108 ----------------------------- 8 files changed, 17 insertions(+), 134 deletions(-) delete mode 100644 drivers/infiniband/hw/ipath/verbs_debug.h (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c index 958cc9b33c8..e6261bb9a8a 100644 --- a/drivers/infiniband/hw/ipath/ipath_driver.c +++ b/drivers/infiniband/hw/ipath/ipath_driver.c @@ -58,7 +58,7 @@ const char *ipath_get_unit_name(int unit) * The size has to be longer than this string, so we can append * board/chip information to it in the init code. */ -const char ipath_core_version[] = IPATH_IDSTR "\n"; +const char ib_ipath_version[] = IPATH_IDSTR "\n"; static struct idr unit_table; DEFINE_SPINLOCK(ipath_devs_lock); @@ -1847,7 +1847,7 @@ static int __init infinipath_init(void) { int ret; - ipath_dbg(KERN_INFO DRIVER_LOAD_MSG "%s", ipath_core_version); + ipath_dbg(KERN_INFO DRIVER_LOAD_MSG "%s", ib_ipath_version); /* * These must be called before the driver is registered with diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h index 999249b7f27..af342314b36 100644 --- a/drivers/infiniband/hw/ipath/ipath_kernel.h +++ b/drivers/infiniband/hw/ipath/ipath_kernel.h @@ -785,7 +785,7 @@ static inline u32 ipath_read_creg32(const struct ipath_devdata *dd, struct device_driver; -extern const char ipath_core_version[]; +extern const char ib_ipath_version[]; int ipath_driver_create_group(struct device_driver *); void ipath_driver_remove_group(struct device_driver *); @@ -815,7 +815,7 @@ const char *ipath_get_unit_name(int unit); extern struct mutex ipath_mutex; -#define IPATH_DRV_NAME "ipath_core" +#define IPATH_DRV_NAME "ib_ipath" #define IPATH_MAJOR 233 #define IPATH_USER_MINOR_BASE 0 #define IPATH_SMA_MINOR 128 diff --git a/drivers/infiniband/hw/ipath/ipath_keys.c b/drivers/infiniband/hw/ipath/ipath_keys.c index a5ca279370a..ba1b93226ca 100644 --- a/drivers/infiniband/hw/ipath/ipath_keys.c +++ b/drivers/infiniband/hw/ipath/ipath_keys.c @@ -34,6 +34,7 @@ #include #include "ipath_verbs.h" +#include "ipath_kernel.h" /** * ipath_alloc_lkey - allocate an lkey @@ -60,7 +61,7 @@ int ipath_alloc_lkey(struct ipath_lkey_table *rkt, struct ipath_mregion *mr) r = (r + 1) & (rkt->max - 1); if (r == n) { spin_unlock_irqrestore(&rkt->lock, flags); - _VERBS_INFO("LKEY table full\n"); + ipath_dbg(KERN_INFO "LKEY table full\n"); ret = 0; goto bail; } diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c index 9228d59b59d..b86858e70f2 100644 --- a/drivers/infiniband/hw/ipath/ipath_qp.c +++ b/drivers/infiniband/hw/ipath/ipath_qp.c @@ -274,7 +274,7 @@ void ipath_free_all_qps(struct ipath_qp_table *qpt) free_qpn(qpt, qp->ibqp.qp_num); if (!atomic_dec_and_test(&qp->refcount) || !ipath_destroy_qp(&qp->ibqp)) - _VERBS_INFO("QP memory leak!\n"); + ipath_dbg(KERN_INFO "QP memory leak!\n"); qp = nqp; } } @@ -362,8 +362,8 @@ void ipath_error_qp(struct ipath_qp *qp) struct ipath_ibdev *dev = to_idev(qp->ibqp.device); struct ib_wc wc; - _VERBS_INFO("QP%d/%d in error state\n", - qp->ibqp.qp_num, qp->remote_qpn); + ipath_dbg(KERN_INFO "QP%d/%d in error state\n", + qp->ibqp.qp_num, qp->remote_qpn); spin_lock(&dev->pending_lock); /* XXX What if its already removed by the timeout code? */ @@ -945,8 +945,8 @@ void ipath_sqerror_qp(struct ipath_qp *qp, struct ib_wc *wc) struct ipath_ibdev *dev = to_idev(qp->ibqp.device); struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last); - _VERBS_INFO("Send queue error on QP%d/%d: err: %d\n", - qp->ibqp.qp_num, qp->remote_qpn, wc->status); + ipath_dbg(KERN_INFO "Send queue error on QP%d/%d: err: %d\n", + qp->ibqp.qp_num, qp->remote_qpn, wc->status); spin_lock(&dev->pending_lock); /* XXX What if its already removed by the timeout code? */ diff --git a/drivers/infiniband/hw/ipath/ipath_sysfs.c b/drivers/infiniband/hw/ipath/ipath_sysfs.c index 7396a63840d..56f12202ff4 100644 --- a/drivers/infiniband/hw/ipath/ipath_sysfs.c +++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c @@ -75,7 +75,7 @@ bail: static ssize_t show_version(struct device_driver *dev, char *buf) { /* The string printed here is already newline-terminated. */ - return scnprintf(buf, PAGE_SIZE, "%s", ipath_core_version); + return scnprintf(buf, PAGE_SIZE, "%s", ib_ipath_version); } static ssize_t show_num_units(struct device_driver *dev, char *buf) diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index 3c47620e988..1776330914e 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -50,10 +50,6 @@ module_param_named(lkey_table_size, ib_ipath_lkey_table_size, uint, MODULE_PARM_DESC(lkey_table_size, "LKEY table size in bits (2^n, 1 <= n <= 23)"); -unsigned int ib_ipath_debug; /* debug mask */ -module_param_named(debug, ib_ipath_debug, uint, S_IWUSR | S_IRUGO); -MODULE_PARM_DESC(debug, "Verbs debug mask"); - static unsigned int ib_ipath_max_pds = 0xFFFF; module_param_named(max_pds, ib_ipath_max_pds, uint, S_IWUSR | S_IRUGO); MODULE_PARM_DESC(max_pds, @@ -1598,8 +1594,7 @@ err_lk: kfree(idev->qp_table.table); err_qp: ib_dealloc_device(dev); - _VERBS_ERROR("ib_ipath%d cannot register verbs (%d)!\n", - dd->ipath_unit, -ret); + ipath_dev_err(dd, "cannot register verbs: %d!\n", -ret); idev = NULL; bail: @@ -1618,17 +1613,13 @@ void ipath_unregister_ib_device(struct ipath_ibdev *dev) if (!list_empty(&dev->pending[0]) || !list_empty(&dev->pending[1]) || !list_empty(&dev->pending[2])) - _VERBS_ERROR("ipath%d pending list not empty!\n", - dev->ib_unit); + ipath_dev_err(dev->dd, "pending list not empty!\n"); if (!list_empty(&dev->piowait)) - _VERBS_ERROR("ipath%d piowait list not empty!\n", - dev->ib_unit); + ipath_dev_err(dev->dd, "piowait list not empty!\n"); if (!list_empty(&dev->rnrwait)) - _VERBS_ERROR("ipath%d rnrwait list not empty!\n", - dev->ib_unit); + ipath_dev_err(dev->dd, "rnrwait list not empty!\n"); if (!ipath_mcast_tree_empty()) - _VERBS_ERROR("ipath%d multicast table memory leak!\n", - dev->ib_unit); + ipath_dev_err(dev->dd, "multicast table memory leak!\n"); /* * Note that ipath_unregister_ib_device() can be called before all * the QPs are destroyed! diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 00f4cecc258..9cc0dbfe860 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -42,7 +42,6 @@ #include #include "ipath_layer.h" -#include "verbs_debug.h" #define QPN_MAX (1 << 24) #define QPNMAP_ENTRIES (QPN_MAX / PAGE_SIZE / BITS_PER_BYTE) diff --git a/drivers/infiniband/hw/ipath/verbs_debug.h b/drivers/infiniband/hw/ipath/verbs_debug.h deleted file mode 100644 index 6186676f2a1..00000000000 --- a/drivers/infiniband/hw/ipath/verbs_debug.h +++ /dev/null @@ -1,108 +0,0 @@ -/* - * Copyright (c) 2006 QLogic, Inc. 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. - */ - -#ifndef _VERBS_DEBUG_H -#define _VERBS_DEBUG_H - -/* - * This file contains tracing code for the ib_ipath kernel module. - */ -#ifndef _VERBS_DEBUGGING /* tracing enabled or not */ -#define _VERBS_DEBUGGING 1 -#endif - -extern unsigned ib_ipath_debug; - -#define _VERBS_ERROR(fmt,...) \ - do { \ - printk(KERN_ERR "%s: " fmt, "ib_ipath", ##__VA_ARGS__); \ - } while(0) - -#define _VERBS_UNIT_ERROR(unit,fmt,...) \ - do { \ - printk(KERN_ERR "%s: " fmt, "ib_ipath", ##__VA_ARGS__); \ - } while(0) - -#if _VERBS_DEBUGGING - -/* - * Mask values for debugging. The scheme allows us to compile out any - * of the debug tracing stuff, and if compiled in, to enable or - * disable dynamically. - * This can be set at modprobe time also: - * modprobe ib_path ib_ipath_debug=3 - */ - -#define __VERBS_INFO 0x1 /* generic low verbosity stuff */ -#define __VERBS_DBG 0x2 /* generic debug */ -#define __VERBS_VDBG 0x4 /* verbose debug */ -#define __VERBS_SMADBG 0x8000 /* sma packet debug */ - -#define _VERBS_INFO(fmt,...) \ - do { \ - if (unlikely(ib_ipath_debug&__VERBS_INFO)) \ - printk(KERN_INFO "%s: " fmt,"ib_ipath", \ - ##__VA_ARGS__); \ - } while(0) - -#define _VERBS_DBG(fmt,...) \ - do { \ - if (unlikely(ib_ipath_debug&__VERBS_DBG)) \ - printk(KERN_DEBUG "%s: " fmt, __func__, \ - ##__VA_ARGS__); \ - } while(0) - -#define _VERBS_VDBG(fmt,...) \ - do { \ - if (unlikely(ib_ipath_debug&__VERBS_VDBG)) \ - printk(KERN_DEBUG "%s: " fmt, __func__, \ - ##__VA_ARGS__); \ - } while(0) - -#define _VERBS_SMADBG(fmt,...) \ - do { \ - if (unlikely(ib_ipath_debug&__VERBS_SMADBG)) \ - printk(KERN_DEBUG "%s: " fmt, __func__, \ - ##__VA_ARGS__); \ - } while(0) - -#else /* ! _VERBS_DEBUGGING */ - -#define _VERBS_INFO(fmt,...) -#define _VERBS_DBG(fmt,...) -#define _VERBS_VDBG(fmt,...) -#define _VERBS_SMADBG(fmt,...) - -#endif /* _VERBS_DEBUGGING */ - -#endif /* _VERBS_DEBUG_H */ -- cgit v1.2.3-70-g09d2 From fc8cf8cdfc3ce328c577b18ebcd60a5595f2a283 Mon Sep 17 00:00:00 2001 From: Bryan O'Sullivan Date: Fri, 25 Aug 2006 11:24:41 -0700 Subject: IB/ipath: be more strict about testing the modify QP verb Signed-off-by: Bryan O'Sullivan Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_qp.c | 33 +++++++++++++++++++++++++++++-- drivers/infiniband/hw/ipath/ipath_verbs.h | 1 + 2 files changed, 32 insertions(+), 2 deletions(-) (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c index c0267cf8ca8..502d555fdf3 100644 --- a/drivers/infiniband/hw/ipath/ipath_qp.c +++ b/drivers/infiniband/hw/ipath/ipath_qp.c @@ -455,11 +455,16 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, attr_mask)) goto inval; - if (attr_mask & IB_QP_AV) + if (attr_mask & IB_QP_AV) { if (attr->ah_attr.dlid == 0 || attr->ah_attr.dlid >= IPATH_MULTICAST_LID_BASE) goto inval; + if ((attr->ah_attr.ah_flags & IB_AH_GRH) && + (attr->ah_attr.grh.sgid_index > 1)) + goto inval; + } + if (attr_mask & IB_QP_PKEY_INDEX) if (attr->pkey_index >= ipath_get_npkeys(dev->dd)) goto inval; @@ -468,6 +473,27 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, if (attr->min_rnr_timer > 31) goto inval; + if (attr_mask & IB_QP_PORT) + if (attr->port_num == 0 || + attr->port_num > ibqp->device->phys_port_cnt) + goto inval; + + if (attr_mask & IB_QP_PATH_MTU) + if (attr->path_mtu > IB_MTU_4096) + goto inval; + + if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) + if (attr->max_dest_rd_atomic > 1) + goto inval; + + if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) + if (attr->max_rd_atomic > 1) + goto inval; + + if (attr_mask & IB_QP_PATH_MIG_STATE) + if (attr->path_mig_state != IB_MIG_MIGRATED) + goto inval; + switch (new_state) { case IB_QPS_RESET: ipath_reset_qp(qp); @@ -518,6 +544,9 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, if (attr_mask & IB_QP_MIN_RNR_TIMER) qp->r_min_rnr_timer = attr->min_rnr_timer; + if (attr_mask & IB_QP_TIMEOUT) + qp->timeout = attr->timeout; + if (attr_mask & IB_QP_QKEY) qp->qkey = attr->qkey; @@ -564,7 +593,7 @@ int ipath_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, attr->max_dest_rd_atomic = 1; attr->min_rnr_timer = qp->r_min_rnr_timer; attr->port_num = 1; - attr->timeout = 0; + attr->timeout = qp->timeout; attr->retry_cnt = qp->s_retry_cnt; attr->rnr_retry = qp->s_rnr_retry; attr->alt_port_num = 0; diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 9cc0dbfe860..f2956090d93 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -371,6 +371,7 @@ struct ipath_qp { u8 s_retry; /* requester retry counter */ u8 s_rnr_retry; /* requester RNR retry counter */ u8 s_pkey_index; /* PKEY index to use */ + u8 timeout; /* Timeout for this QP */ enum ib_mtu path_mtu; u32 remote_qpn; u32 qkey; /* QKEY for this QP (for UD or RD) */ -- cgit v1.2.3-70-g09d2 From 0b81e4f79af8322c7142701982f40d1431dedf19 Mon Sep 17 00:00:00 2001 From: Bryan O'Sullivan Date: Fri, 25 Aug 2006 11:24:43 -0700 Subject: IB/ipath: put a limit on the number of QPs that can be created Signed-off-by: Bryan O'Sullivan Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_qp.c | 15 +++++++++++++++ drivers/infiniband/hw/ipath/ipath_verbs.c | 7 ++++++- drivers/infiniband/hw/ipath/ipath_verbs.h | 4 ++++ 3 files changed, 25 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw/ipath/ipath_verbs.h') diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c index 77391886d2f..607ba72af2f 100644 --- a/drivers/infiniband/hw/ipath/ipath_qp.c +++ b/drivers/infiniband/hw/ipath/ipath_qp.c @@ -833,9 +833,21 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd, } } + spin_lock(&dev->n_qps_lock); + if (dev->n_qps_allocated == ib_ipath_max_qps) { + spin_unlock(&dev->n_qps_lock); + ret = ERR_PTR(-ENOMEM); + goto bail_ip; + } + + dev->n_qps_allocated++; + spin_unlock(&dev->n_qps_lock); + ret = &qp->ibqp; goto bail; +bail_ip: + kfree(qp->ip); bail_rwq: vfree(qp->r_rq.wq); bail_qp: @@ -864,6 +876,9 @@ int ipath_destroy_qp(struct ib_qp *ibqp) spin_lock_irqsave(&qp->s_lock, flags); qp->state = IB_QPS_ERR; spin_unlock_irqrestore(&qp->s_lock, flags); + spin_lock(&dev->n_qps_lock); + dev->n_qps_allocated--; + spin_unlock(&dev->n_qps_lock); /* Stop the sending tasklet. */ tasklet_kill(&qp->s_task); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c index ab0006288b4..b9be0fd2ed3 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.c +++ b/drivers/infiniband/hw/ipath/ipath_verbs.c @@ -73,6 +73,10 @@ module_param_named(max_qp_wrs, ib_ipath_max_qp_wrs, uint, S_IWUSR | S_IRUGO); MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support"); +unsigned int ib_ipath_max_qps = 16384; +module_param_named(max_qps, ib_ipath_max_qps, uint, S_IWUSR | S_IRUGO); +MODULE_PARM_DESC(max_qps, "Maximum number of QPs to support"); + unsigned int ib_ipath_max_sges = 0x60; module_param_named(max_sges, ib_ipath_max_sges, uint, S_IWUSR | S_IRUGO); MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support"); @@ -958,7 +962,7 @@ static int ipath_query_device(struct ib_device *ibdev, props->sys_image_guid = dev->sys_image_guid; props->max_mr_size = ~0ull; - props->max_qp = dev->qp_table.max; + props->max_qp = ib_ipath_max_qps; props->max_qp_wr = ib_ipath_max_qp_wrs; props->max_sge = ib_ipath_max_sges; props->max_cq = ib_ipath_max_cqs; @@ -1420,6 +1424,7 @@ int ipath_register_ib_device(struct ipath_devdata *dd) spin_lock_init(&idev->n_pds_lock); spin_lock_init(&idev->n_ahs_lock); spin_lock_init(&idev->n_cqs_lock); + spin_lock_init(&idev->n_qps_lock); spin_lock_init(&idev->n_srqs_lock); spin_lock_init(&idev->n_mcast_grps_lock); diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index f2956090d93..09bbb3f9a21 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -482,6 +482,8 @@ struct ipath_ibdev { spinlock_t n_ahs_lock; u32 n_cqs_allocated; /* number of CQs allocated for device */ spinlock_t n_cqs_lock; + u32 n_qps_allocated; /* number of QPs allocated for device */ + spinlock_t n_qps_lock; u32 n_srqs_allocated; /* number of SRQs allocated for device */ spinlock_t n_srqs_lock; u32 n_mcast_grps_allocated; /* number of mcast groups allocated */ @@ -792,6 +794,8 @@ extern unsigned int ib_ipath_max_cqs; extern unsigned int ib_ipath_max_qp_wrs; +extern unsigned int ib_ipath_max_qps; + extern unsigned int ib_ipath_max_sges; extern unsigned int ib_ipath_max_mcast_grps; -- cgit v1.2.3-70-g09d2