diff options
Diffstat (limited to 'drivers/s390/block')
-rw-r--r-- | drivers/s390/block/Kconfig | 18 | ||||
-rw-r--r-- | drivers/s390/block/Makefile | 6 | ||||
-rw-r--r-- | drivers/s390/block/dasd.c | 17 | ||||
-rw-r--r-- | drivers/s390/block/dasd_alias.c | 27 | ||||
-rw-r--r-- | drivers/s390/block/dasd_eckd.c | 53 | ||||
-rw-r--r-- | drivers/s390/block/dasd_ioctl.c | 4 | ||||
-rw-r--r-- | drivers/s390/block/scm_blk.c | 445 | ||||
-rw-r--r-- | drivers/s390/block/scm_blk.h | 117 | ||||
-rw-r--r-- | drivers/s390/block/scm_blk_cluster.c | 228 | ||||
-rw-r--r-- | drivers/s390/block/scm_drv.c | 81 |
10 files changed, 977 insertions, 19 deletions
diff --git a/drivers/s390/block/Kconfig b/drivers/s390/block/Kconfig index 8e477bb1f3f..4a3b6232618 100644 --- a/drivers/s390/block/Kconfig +++ b/drivers/s390/block/Kconfig @@ -70,3 +70,21 @@ config DASD_EER This driver provides a character device interface to the DASD extended error reporting. This is only needed if you want to use applications written for the EER facility. + +config SCM_BLOCK + def_tristate m + prompt "Support for Storage Class Memory" + depends on S390 && BLOCK && EADM_SCH && SCM_BUS + help + Block device driver for Storage Class Memory (SCM). This driver + provides a block device interface for each available SCM increment. + + To compile this driver as a module, choose M here: the + module will be called scm_block. + +config SCM_BLOCK_CLUSTER_WRITE + def_bool y + prompt "SCM force cluster writes" + depends on SCM_BLOCK + help + Force writes to Storage Class Memory (SCM) to be in done in clusters. diff --git a/drivers/s390/block/Makefile b/drivers/s390/block/Makefile index 0a89e080b38..c2f4e673e03 100644 --- a/drivers/s390/block/Makefile +++ b/drivers/s390/block/Makefile @@ -17,3 +17,9 @@ obj-$(CONFIG_DASD_ECKD) += dasd_eckd_mod.o obj-$(CONFIG_DASD_FBA) += dasd_fba_mod.o obj-$(CONFIG_BLK_DEV_XPRAM) += xpram.o obj-$(CONFIG_DCSSBLK) += dcssblk.o + +scm_block-objs := scm_drv.o scm_blk.o +ifdef CONFIG_SCM_BLOCK_CLUSTER_WRITE +scm_block-objs += scm_blk_cluster.o +endif +obj-$(CONFIG_SCM_BLOCK) += scm_block.o diff --git a/drivers/s390/block/dasd.c b/drivers/s390/block/dasd.c index 15370a2c5ff..0595c763daf 100644 --- a/drivers/s390/block/dasd.c +++ b/drivers/s390/block/dasd.c @@ -534,11 +534,11 @@ static void dasd_change_state(struct dasd_device *device) if (rc) device->target = device->state; - if (device->state == device->target) - wake_up(&dasd_init_waitq); - /* let user-space know that the device status changed */ kobject_uevent(&device->cdev->dev.kobj, KOBJ_CHANGE); + + if (device->state == device->target) + wake_up(&dasd_init_waitq); } /* @@ -2157,6 +2157,7 @@ static int _dasd_sleep_on(struct dasd_ccw_req *maincqr, int interruptible) test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) && (!dasd_eer_enabled(device))) { cqr->status = DASD_CQR_FAILED; + cqr->intrc = -EAGAIN; continue; } /* Don't try to start requests if device is stopped */ @@ -3270,6 +3271,16 @@ void dasd_generic_path_event(struct ccw_device *cdev, int *path_event) dasd_schedule_device_bh(device); } if (path_event[chp] & PE_PATHGROUP_ESTABLISHED) { + if (!(device->path_data.opm & eventlpm) && + !(device->path_data.tbvpm & eventlpm)) { + /* + * we can not establish a pathgroup on an + * unavailable path, so trigger a path + * verification first + */ + device->path_data.tbvpm |= eventlpm; + dasd_schedule_device_bh(device); + } DBF_DEV_EVENT(DBF_WARNING, device, "%s", "Pathgroup re-established\n"); if (device->discipline->kick_validate) diff --git a/drivers/s390/block/dasd_alias.c b/drivers/s390/block/dasd_alias.c index 157defe5e06..6b556995bb3 100644 --- a/drivers/s390/block/dasd_alias.c +++ b/drivers/s390/block/dasd_alias.c @@ -384,6 +384,29 @@ static void _remove_device_from_lcu(struct alias_lcu *lcu, group->next = NULL; }; +static int +suborder_not_supported(struct dasd_ccw_req *cqr) +{ + char *sense; + char reason; + char msg_format; + char msg_no; + + sense = dasd_get_sense(&cqr->irb); + if (!sense) + return 0; + + reason = sense[0]; + msg_format = (sense[7] & 0xF0); + msg_no = (sense[7] & 0x0F); + + /* command reject, Format 0 MSG 4 - invalid parameter */ + if ((reason == 0x80) && (msg_format == 0x00) && (msg_no == 0x04)) + return 1; + + return 0; +} + static int read_unit_address_configuration(struct dasd_device *device, struct alias_lcu *lcu) { @@ -435,6 +458,8 @@ static int read_unit_address_configuration(struct dasd_device *device, do { rc = dasd_sleep_on(cqr); + if (rc && suborder_not_supported(cqr)) + return -EOPNOTSUPP; } while (rc && (cqr->retries > 0)); if (rc) { spin_lock_irqsave(&lcu->lock, flags); @@ -521,7 +546,7 @@ static void lcu_update_work(struct work_struct *work) * processing the data */ spin_lock_irqsave(&lcu->lock, flags); - if (rc || (lcu->flags & NEED_UAC_UPDATE)) { + if ((rc && (rc != -EOPNOTSUPP)) || (lcu->flags & NEED_UAC_UPDATE)) { DBF_DEV_EVENT(DBF_WARNING, device, "could not update" " alias data in lcu (rc = %d), retry later", rc); schedule_delayed_work(&lcu->ruac_data.dwork, 30*HZ); diff --git a/drivers/s390/block/dasd_eckd.c b/drivers/s390/block/dasd_eckd.c index 2fb2b9ea97e..108332b44d9 100644 --- a/drivers/s390/block/dasd_eckd.c +++ b/drivers/s390/block/dasd_eckd.c @@ -20,6 +20,7 @@ #include <linux/compat.h> #include <linux/init.h> +#include <asm/css_chars.h> #include <asm/debug.h> #include <asm/idals.h> #include <asm/ebcdic.h> @@ -31,8 +32,6 @@ #include "dasd_int.h" #include "dasd_eckd.h" -#include "../cio/chsc.h" - #ifdef PRINTK_HEADER #undef PRINTK_HEADER @@ -140,6 +139,10 @@ dasd_eckd_set_online(struct ccw_device *cdev) static const int sizes_trk0[] = { 28, 148, 84 }; #define LABEL_SIZE 140 +/* head and record addresses of count_area read in analysis ccw */ +static const int count_area_head[] = { 0, 0, 0, 0, 2 }; +static const int count_area_rec[] = { 1, 2, 3, 4, 1 }; + static inline unsigned int round_up_multiple(unsigned int no, unsigned int mult) { @@ -212,7 +215,7 @@ check_XRC (struct ccw1 *de_ccw, rc = get_sync_clock(&data->ep_sys_time); /* Ignore return code if sync clock is switched off. */ - if (rc == -ENOSYS || rc == -EACCES) + if (rc == -EOPNOTSUPP || rc == -EACCES) rc = 0; de_ccw->count = sizeof(struct DE_eckd_data); @@ -323,7 +326,7 @@ static int check_XRC_on_prefix(struct PFX_eckd_data *pfxdata, rc = get_sync_clock(&pfxdata->define_extent.ep_sys_time); /* Ignore return code if sync clock is switched off. */ - if (rc == -ENOSYS || rc == -EACCES) + if (rc == -EOPNOTSUPP || rc == -EACCES) rc = 0; return rc; } @@ -1507,7 +1510,8 @@ static struct dasd_ccw_req *dasd_eckd_build_psf_ssc(struct dasd_device *device, * call might change behaviour of DASD devices. */ static int -dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav) +dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav, + unsigned long flags) { struct dasd_ccw_req *cqr; int rc; @@ -1516,10 +1520,19 @@ dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav) if (IS_ERR(cqr)) return PTR_ERR(cqr); + /* + * set flags e.g. turn on failfast, to prevent blocking + * the calling function should handle failed requests + */ + cqr->flags |= flags; + rc = dasd_sleep_on(cqr); if (!rc) /* trigger CIO to reprobe devices */ css_schedule_reprobe(); + else if (cqr->intrc == -EAGAIN) + rc = -EAGAIN; + dasd_sfree_request(cqr, cqr->memdev); return rc; } @@ -1527,7 +1540,8 @@ dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav) /* * Valide storage server of current device. */ -static void dasd_eckd_validate_server(struct dasd_device *device) +static int dasd_eckd_validate_server(struct dasd_device *device, + unsigned long flags) { int rc; struct dasd_eckd_private *private; @@ -1536,17 +1550,18 @@ static void dasd_eckd_validate_server(struct dasd_device *device) private = (struct dasd_eckd_private *) device->private; if (private->uid.type == UA_BASE_PAV_ALIAS || private->uid.type == UA_HYPER_PAV_ALIAS) - return; + return 0; if (dasd_nopav || MACHINE_IS_VM) enable_pav = 0; else enable_pav = 1; - rc = dasd_eckd_psf_ssc(device, enable_pav); + rc = dasd_eckd_psf_ssc(device, enable_pav, flags); /* may be requested feature is not available on server, * therefore just report error and go ahead */ DBF_EVENT_DEVID(DBF_WARNING, device->cdev, "PSF-SSC for SSID %04x " "returned rc=%d", private->uid.ssid, rc); + return rc; } /* @@ -1556,7 +1571,13 @@ static void dasd_eckd_do_validate_server(struct work_struct *work) { struct dasd_device *device = container_of(work, struct dasd_device, kick_validate); - dasd_eckd_validate_server(device); + if (dasd_eckd_validate_server(device, DASD_CQR_FLAGS_FAILFAST) + == -EAGAIN) { + /* schedule worker again if failed */ + schedule_work(&device->kick_validate); + return; + } + dasd_put_device(device); } @@ -1685,7 +1706,7 @@ dasd_eckd_check_characteristics(struct dasd_device *device) if (rc) goto out_err2; - dasd_eckd_validate_server(device); + dasd_eckd_validate_server(device, 0); /* device may report different configuration data after LCU setup */ rc = dasd_eckd_read_conf(device); @@ -1922,7 +1943,10 @@ static int dasd_eckd_end_analysis(struct dasd_block *block) count_area = NULL; for (i = 0; i < 3; i++) { if (private->count_area[i].kl != 4 || - private->count_area[i].dl != dasd_eckd_cdl_reclen(i) - 4) { + private->count_area[i].dl != dasd_eckd_cdl_reclen(i) - 4 || + private->count_area[i].cyl != 0 || + private->count_area[i].head != count_area_head[i] || + private->count_area[i].record != count_area_rec[i]) { private->uses_cdl = 0; break; } @@ -1934,7 +1958,10 @@ static int dasd_eckd_end_analysis(struct dasd_block *block) for (i = 0; i < 5; i++) { if ((private->count_area[i].kl != 0) || (private->count_area[i].dl != - private->count_area[0].dl)) + private->count_area[0].dl) || + private->count_area[i].cyl != 0 || + private->count_area[i].head != count_area_head[i] || + private->count_area[i].record != count_area_rec[i]) break; } if (i == 5) @@ -4153,7 +4180,7 @@ static int dasd_eckd_restore_device(struct dasd_device *device) rc = dasd_alias_make_device_known_to_lcu(device); if (rc) return rc; - dasd_eckd_validate_server(device); + dasd_eckd_validate_server(device, DASD_CQR_FLAGS_FAILFAST); /* RE-Read Configuration Data */ rc = dasd_eckd_read_conf(device); diff --git a/drivers/s390/block/dasd_ioctl.c b/drivers/s390/block/dasd_ioctl.c index 654c6921a6d..8252f37d04e 100644 --- a/drivers/s390/block/dasd_ioctl.c +++ b/drivers/s390/block/dasd_ioctl.c @@ -292,12 +292,12 @@ out: #else static int dasd_ioctl_reset_profile(struct dasd_block *block) { - return -ENOSYS; + return -ENOTTY; } static int dasd_ioctl_read_profile(struct dasd_block *block, void __user *argp) { - return -ENOSYS; + return -ENOTTY; } #endif diff --git a/drivers/s390/block/scm_blk.c b/drivers/s390/block/scm_blk.c new file mode 100644 index 00000000000..9978ad4433c --- /dev/null +++ b/drivers/s390/block/scm_blk.c @@ -0,0 +1,445 @@ +/* + * Block driver for s390 storage class memory. + * + * Copyright IBM Corp. 2012 + * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com> + */ + +#define KMSG_COMPONENT "scm_block" +#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt + +#include <linux/interrupt.h> +#include <linux/spinlock.h> +#include <linux/module.h> +#include <linux/blkdev.h> +#include <linux/genhd.h> +#include <linux/slab.h> +#include <linux/list.h> +#include <asm/eadm.h> +#include "scm_blk.h" + +debug_info_t *scm_debug; +static int scm_major; +static DEFINE_SPINLOCK(list_lock); +static LIST_HEAD(inactive_requests); +static unsigned int nr_requests = 64; +static atomic_t nr_devices = ATOMIC_INIT(0); +module_param(nr_requests, uint, S_IRUGO); +MODULE_PARM_DESC(nr_requests, "Number of parallel requests."); + +MODULE_DESCRIPTION("Block driver for s390 storage class memory."); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("scm:scmdev*"); + +static void __scm_free_rq(struct scm_request *scmrq) +{ + struct aob_rq_header *aobrq = to_aobrq(scmrq); + + free_page((unsigned long) scmrq->aob); + free_page((unsigned long) scmrq->aidaw); + __scm_free_rq_cluster(scmrq); + kfree(aobrq); +} + +static void scm_free_rqs(void) +{ + struct list_head *iter, *safe; + struct scm_request *scmrq; + + spin_lock_irq(&list_lock); + list_for_each_safe(iter, safe, &inactive_requests) { + scmrq = list_entry(iter, struct scm_request, list); + list_del(&scmrq->list); + __scm_free_rq(scmrq); + } + spin_unlock_irq(&list_lock); +} + +static int __scm_alloc_rq(void) +{ + struct aob_rq_header *aobrq; + struct scm_request *scmrq; + + aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL); + if (!aobrq) + return -ENOMEM; + + scmrq = (void *) aobrq->data; + scmrq->aidaw = (void *) get_zeroed_page(GFP_DMA); + scmrq->aob = (void *) get_zeroed_page(GFP_DMA); + if (!scmrq->aob || !scmrq->aidaw) { + __scm_free_rq(scmrq); + return -ENOMEM; + } + + if (__scm_alloc_rq_cluster(scmrq)) { + __scm_free_rq(scmrq); + return -ENOMEM; + } + + INIT_LIST_HEAD(&scmrq->list); + spin_lock_irq(&list_lock); + list_add(&scmrq->list, &inactive_requests); + spin_unlock_irq(&list_lock); + + return 0; +} + +static int scm_alloc_rqs(unsigned int nrqs) +{ + int ret = 0; + + while (nrqs-- && !ret) + ret = __scm_alloc_rq(); + + return ret; +} + +static struct scm_request *scm_request_fetch(void) +{ + struct scm_request *scmrq = NULL; + + spin_lock(&list_lock); + if (list_empty(&inactive_requests)) + goto out; + scmrq = list_first_entry(&inactive_requests, struct scm_request, list); + list_del(&scmrq->list); +out: + spin_unlock(&list_lock); + return scmrq; +} + +static void scm_request_done(struct scm_request *scmrq) +{ + unsigned long flags; + + spin_lock_irqsave(&list_lock, flags); + list_add(&scmrq->list, &inactive_requests); + spin_unlock_irqrestore(&list_lock, flags); +} + +static int scm_open(struct block_device *blkdev, fmode_t mode) +{ + return scm_get_ref(); +} + +static int scm_release(struct gendisk *gendisk, fmode_t mode) +{ + scm_put_ref(); + return 0; +} + +static const struct block_device_operations scm_blk_devops = { + .owner = THIS_MODULE, + .open = scm_open, + .release = scm_release, +}; + +static void scm_request_prepare(struct scm_request *scmrq) +{ + struct scm_blk_dev *bdev = scmrq->bdev; + struct scm_device *scmdev = bdev->gendisk->private_data; + struct aidaw *aidaw = scmrq->aidaw; + struct msb *msb = &scmrq->aob->msb[0]; + struct req_iterator iter; + struct bio_vec *bv; + + msb->bs = MSB_BS_4K; + scmrq->aob->request.msb_count = 1; + msb->scm_addr = scmdev->address + + ((u64) blk_rq_pos(scmrq->request) << 9); + msb->oc = (rq_data_dir(scmrq->request) == READ) ? + MSB_OC_READ : MSB_OC_WRITE; + msb->flags |= MSB_FLAG_IDA; + msb->data_addr = (u64) aidaw; + + rq_for_each_segment(bv, scmrq->request, iter) { + WARN_ON(bv->bv_offset); + msb->blk_count += bv->bv_len >> 12; + aidaw->data_addr = (u64) page_address(bv->bv_page); + aidaw++; + } +} + +static inline void scm_request_init(struct scm_blk_dev *bdev, + struct scm_request *scmrq, + struct request *req) +{ + struct aob_rq_header *aobrq = to_aobrq(scmrq); + struct aob *aob = scmrq->aob; + + memset(aob, 0, sizeof(*aob)); + memset(scmrq->aidaw, 0, PAGE_SIZE); + aobrq->scmdev = bdev->scmdev; + aob->request.cmd_code = ARQB_CMD_MOVE; + aob->request.data = (u64) aobrq; + scmrq->request = req; + scmrq->bdev = bdev; + scmrq->retries = 4; + scmrq->error = 0; + scm_request_cluster_init(scmrq); +} + +static void scm_ensure_queue_restart(struct scm_blk_dev *bdev) +{ + if (atomic_read(&bdev->queued_reqs)) { + /* Queue restart is triggered by the next interrupt. */ + return; + } + blk_delay_queue(bdev->rq, SCM_QUEUE_DELAY); +} + +void scm_request_requeue(struct scm_request *scmrq) +{ + struct scm_blk_dev *bdev = scmrq->bdev; + + scm_release_cluster(scmrq); + blk_requeue_request(bdev->rq, scmrq->request); + scm_request_done(scmrq); + scm_ensure_queue_restart(bdev); +} + +void scm_request_finish(struct scm_request *scmrq) +{ + scm_release_cluster(scmrq); + blk_end_request_all(scmrq->request, scmrq->error); + scm_request_done(scmrq); +} + +static void scm_blk_request(struct request_queue *rq) +{ + struct scm_device *scmdev = rq->queuedata; + struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev); + struct scm_request *scmrq; + struct request *req; + int ret; + + while ((req = blk_peek_request(rq))) { + if (req->cmd_type != REQ_TYPE_FS) + continue; + + scmrq = scm_request_fetch(); + if (!scmrq) { + SCM_LOG(5, "no request"); + scm_ensure_queue_restart(bdev); + return; + } + scm_request_init(bdev, scmrq, req); + if (!scm_reserve_cluster(scmrq)) { + SCM_LOG(5, "cluster busy"); + scm_request_done(scmrq); + return; + } + if (scm_need_cluster_request(scmrq)) { + blk_start_request(req); + scm_initiate_cluster_request(scmrq); + return; + } + scm_request_prepare(scmrq); + blk_start_request(req); + + ret = scm_start_aob(scmrq->aob); + if (ret) { + SCM_LOG(5, "no subchannel"); + scm_request_requeue(scmrq); + return; + } + atomic_inc(&bdev->queued_reqs); + } +} + +static void __scmrq_log_error(struct scm_request *scmrq) +{ + struct aob *aob = scmrq->aob; + + if (scmrq->error == -ETIMEDOUT) + SCM_LOG(1, "Request timeout"); + else { + SCM_LOG(1, "Request error"); + SCM_LOG_HEX(1, &aob->response, sizeof(aob->response)); + } + if (scmrq->retries) + SCM_LOG(1, "Retry request"); + else + pr_err("An I/O operation to SCM failed with rc=%d\n", + scmrq->error); +} + +void scm_blk_irq(struct scm_device *scmdev, void *data, int error) +{ + struct scm_request *scmrq = data; + struct scm_blk_dev *bdev = scmrq->bdev; + + scmrq->error = error; + if (error) + __scmrq_log_error(scmrq); + + spin_lock(&bdev->lock); + list_add_tail(&scmrq->list, &bdev->finished_requests); + spin_unlock(&bdev->lock); + tasklet_hi_schedule(&bdev->tasklet); +} + +static void scm_blk_tasklet(struct scm_blk_dev *bdev) +{ + struct scm_request *scmrq; + unsigned long flags; + + spin_lock_irqsave(&bdev->lock, flags); + while (!list_empty(&bdev->finished_requests)) { + scmrq = list_first_entry(&bdev->finished_requests, + struct scm_request, list); + list_del(&scmrq->list); + spin_unlock_irqrestore(&bdev->lock, flags); + + if (scmrq->error && scmrq->retries-- > 0) { + if (scm_start_aob(scmrq->aob)) { + spin_lock_irqsave(&bdev->rq_lock, flags); + scm_request_requeue(scmrq); + spin_unlock_irqrestore(&bdev->rq_lock, flags); + } + /* Request restarted or requeued, handle next. */ + spin_lock_irqsave(&bdev->lock, flags); + continue; + } + + if (scm_test_cluster_request(scmrq)) { + scm_cluster_request_irq(scmrq); + spin_lock_irqsave(&bdev->lock, flags); + continue; + } + + scm_request_finish(scmrq); + atomic_dec(&bdev->queued_reqs); + spin_lock_irqsave(&bdev->lock, flags); + } + spin_unlock_irqrestore(&bdev->lock, flags); + /* Look out for more requests. */ + blk_run_queue(bdev->rq); +} + +int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev) +{ + struct request_queue *rq; + int len, ret = -ENOMEM; + unsigned int devindex, nr_max_blk; + + devindex = atomic_inc_return(&nr_devices) - 1; + /* scma..scmz + scmaa..scmzz */ + if (devindex > 701) { + ret = -ENODEV; + goto out; + } + + bdev->scmdev = scmdev; + spin_lock_init(&bdev->rq_lock); + spin_lock_init(&bdev->lock); + INIT_LIST_HEAD(&bdev->finished_requests); + atomic_set(&bdev->queued_reqs, 0); + tasklet_init(&bdev->tasklet, + (void (*)(unsigned long)) scm_blk_tasklet, + (unsigned long) bdev); + + rq = blk_init_queue(scm_blk_request, &bdev->rq_lock); + if (!rq) + goto out; + + bdev->rq = rq; + nr_max_blk = min(scmdev->nr_max_block, + (unsigned int) (PAGE_SIZE / sizeof(struct aidaw))); + + blk_queue_logical_block_size(rq, 1 << 12); + blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */ + blk_queue_max_segments(rq, nr_max_blk); + queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq); + scm_blk_dev_cluster_setup(bdev); + + bdev->gendisk = alloc_disk(SCM_NR_PARTS); + if (!bdev->gendisk) + goto out_queue; + + rq->queuedata = scmdev; + bdev->gendisk->driverfs_dev = &scmdev->dev; + bdev->gendisk->private_data = scmdev; + bdev->gendisk->fops = &scm_blk_devops; + bdev->gendisk->queue = rq; + bdev->gendisk->major = scm_major; + bdev->gendisk->first_minor = devindex * SCM_NR_PARTS; + + len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm"); + if (devindex > 25) { + len += snprintf(bdev->gendisk->disk_name + len, + DISK_NAME_LEN - len, "%c", + 'a' + (devindex / 26) - 1); + devindex = devindex % 26; + } + snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c", + 'a' + devindex); + + /* 512 byte sectors */ + set_capacity(bdev->gendisk, scmdev->size >> 9); + add_disk(bdev->gendisk); + return 0; + +out_queue: + blk_cleanup_queue(rq); +out: + atomic_dec(&nr_devices); + return ret; +} + +void scm_blk_dev_cleanup(struct scm_blk_dev *bdev) +{ + tasklet_kill(&bdev->tasklet); + del_gendisk(bdev->gendisk); + blk_cleanup_queue(bdev->gendisk->queue); + put_disk(bdev->gendisk); +} + +static int __init scm_blk_init(void) +{ + int ret = -EINVAL; + + if (!scm_cluster_size_valid()) + goto out; + + ret = register_blkdev(0, "scm"); + if (ret < 0) + goto out; + + scm_major = ret; + if (scm_alloc_rqs(nr_requests)) + goto out_unreg; + + scm_debug = debug_register("scm_log", 16, 1, 16); + if (!scm_debug) + goto out_free; + + debug_register_view(scm_debug, &debug_hex_ascii_view); + debug_set_level(scm_debug, 2); + + ret = scm_drv_init(); + if (ret) + goto out_dbf; + + return ret; + +out_dbf: + debug_unregister(scm_debug); +out_free: + scm_free_rqs(); +out_unreg: + unregister_blkdev(scm_major, "scm"); +out: + return ret; +} +module_init(scm_blk_init); + +static void __exit scm_blk_cleanup(void) +{ + scm_drv_cleanup(); + debug_unregister(scm_debug); + scm_free_rqs(); + unregister_blkdev(scm_major, "scm"); +} +module_exit(scm_blk_cleanup); diff --git a/drivers/s390/block/scm_blk.h b/drivers/s390/block/scm_blk.h new file mode 100644 index 00000000000..7ac6bad919e --- /dev/null +++ b/drivers/s390/block/scm_blk.h @@ -0,0 +1,117 @@ +#ifndef SCM_BLK_H +#define SCM_BLK_H + +#include <linux/interrupt.h> +#include <linux/spinlock.h> +#include <linux/blkdev.h> +#include <linux/genhd.h> +#include <linux/list.h> + +#include <asm/debug.h> +#include <asm/eadm.h> + +#define SCM_NR_PARTS 8 +#define SCM_QUEUE_DELAY 5 + +struct scm_blk_dev { + struct tasklet_struct tasklet; + struct request_queue *rq; + struct gendisk *gendisk; + struct scm_device *scmdev; + spinlock_t rq_lock; /* guard the request queue */ + spinlock_t lock; /* guard the rest of the blockdev */ + atomic_t queued_reqs; + struct list_head finished_requests; +#ifdef CONFIG_SCM_BLOCK_CLUSTER_WRITE + struct list_head cluster_list; +#endif +}; + +struct scm_request { + struct scm_blk_dev *bdev; + struct request *request; + struct aidaw *aidaw; + struct aob *aob; + struct list_head list; + u8 retries; + int error; +#ifdef CONFIG_SCM_BLOCK_CLUSTER_WRITE + struct { + enum {CLUSTER_NONE, CLUSTER_READ, CLUSTER_WRITE} state; + struct list_head list; + void **buf; + } cluster; +#endif +}; + +#define to_aobrq(rq) container_of((void *) rq, struct aob_rq_header, data) + +int scm_blk_dev_setup(struct scm_blk_dev *, struct scm_device *); +void scm_blk_dev_cleanup(struct scm_blk_dev *); +void scm_blk_irq(struct scm_device *, void *, int); + +void scm_request_finish(struct scm_request *); +void scm_request_requeue(struct scm_request *); + +int scm_drv_init(void); +void scm_drv_cleanup(void); + +#ifdef CONFIG_SCM_BLOCK_CLUSTER_WRITE +void __scm_free_rq_cluster(struct scm_request *); +int __scm_alloc_rq_cluster(struct scm_request *); +void scm_request_cluster_init(struct scm_request *); +bool scm_reserve_cluster(struct scm_request *); +void scm_release_cluster(struct scm_request *); +void scm_blk_dev_cluster_setup(struct scm_blk_dev *); +bool scm_need_cluster_request(struct scm_request *); +void scm_initiate_cluster_request(struct scm_request *); +void scm_cluster_request_irq(struct scm_request *); +bool scm_test_cluster_request(struct scm_request *); +bool scm_cluster_size_valid(void); +#else +#define __scm_free_rq_cluster(scmrq) {} +#define __scm_alloc_rq_cluster(scmrq) 0 +#define scm_request_cluster_init(scmrq) {} +#define scm_reserve_cluster(scmrq) true +#define scm_release_cluster(scmrq) {} +#define scm_blk_dev_cluster_setup(bdev) {} +#define scm_need_cluster_request(scmrq) false +#define scm_initiate_cluster_request(scmrq) {} +#define scm_cluster_request_irq(scmrq) {} +#define scm_test_cluster_request(scmrq) false +#define scm_cluster_size_valid() true +#endif + +extern debug_info_t *scm_debug; + +#define SCM_LOG(imp, txt) do { \ + debug_text_event(scm_debug, imp, txt); \ + } while (0) + +static inline void SCM_LOG_HEX(int level, void *data, int length) +{ + if (level > scm_debug->level) + return; + while (length > 0) { + debug_event(scm_debug, level, data, length); + length -= scm_debug->buf_size; + data += scm_debug->buf_size; + } +} + +static inline void SCM_LOG_STATE(int level, struct scm_device *scmdev) +{ + struct { + u64 address; + u8 oper_state; + u8 rank; + } __packed data = { + .address = scmdev->address, + .oper_state = scmdev->attrs.oper_state, + .rank = scmdev->attrs.rank, + }; + + SCM_LOG_HEX(level, &data, sizeof(data)); +} + +#endif /* SCM_BLK_H */ diff --git a/drivers/s390/block/scm_blk_cluster.c b/drivers/s390/block/scm_blk_cluster.c new file mode 100644 index 00000000000..f4bb61b0cea --- /dev/null +++ b/drivers/s390/block/scm_blk_cluster.c @@ -0,0 +1,228 @@ +/* + * Block driver for s390 storage class memory. + * + * Copyright IBM Corp. 2012 + * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com> + */ + +#include <linux/spinlock.h> +#include <linux/module.h> +#include <linux/blkdev.h> +#include <linux/genhd.h> +#include <linux/slab.h> +#include <linux/list.h> +#include <asm/eadm.h> +#include "scm_blk.h" + +static unsigned int write_cluster_size = 64; +module_param(write_cluster_size, uint, S_IRUGO); +MODULE_PARM_DESC(write_cluster_size, + "Number of pages used for contiguous writes."); + +#define CLUSTER_SIZE (write_cluster_size * PAGE_SIZE) + +void __scm_free_rq_cluster(struct scm_request *scmrq) +{ + int i; + + if (!scmrq->cluster.buf) + return; + + for (i = 0; i < 2 * write_cluster_size; i++) + free_page((unsigned long) scmrq->cluster.buf[i]); + + kfree(scmrq->cluster.buf); +} + +int __scm_alloc_rq_cluster(struct scm_request *scmrq) +{ + int i; + + scmrq->cluster.buf = kzalloc(sizeof(void *) * 2 * write_cluster_size, + GFP_KERNEL); + if (!scmrq->cluster.buf) + return -ENOMEM; + + for (i = 0; i < 2 * write_cluster_size; i++) { + scmrq->cluster.buf[i] = (void *) get_zeroed_page(GFP_DMA); + if (!scmrq->cluster.buf[i]) + return -ENOMEM; + } + INIT_LIST_HEAD(&scmrq->cluster.list); + return 0; +} + +void scm_request_cluster_init(struct scm_request *scmrq) +{ + scmrq->cluster.state = CLUSTER_NONE; +} + +static bool clusters_intersect(struct scm_request *A, struct scm_request *B) +{ + unsigned long firstA, lastA, firstB, lastB; + + firstA = ((u64) blk_rq_pos(A->request) << 9) / CLUSTER_SIZE; + lastA = (((u64) blk_rq_pos(A->request) << 9) + + blk_rq_bytes(A->request) - 1) / CLUSTER_SIZE; + + firstB = ((u64) blk_rq_pos(B->request) << 9) / CLUSTER_SIZE; + lastB = (((u64) blk_rq_pos(B->request) << 9) + + blk_rq_bytes(B->request) - 1) / CLUSTER_SIZE; + + return (firstB <= lastA && firstA <= lastB); +} + +bool scm_reserve_cluster(struct scm_request *scmrq) +{ + struct scm_blk_dev *bdev = scmrq->bdev; + struct scm_request *iter; + + if (write_cluster_size == 0) + return true; + + spin_lock(&bdev->lock); + list_for_each_entry(iter, &bdev->cluster_list, cluster.list) { + if (clusters_intersect(scmrq, iter) && + (rq_data_dir(scmrq->request) == WRITE || + rq_data_dir(iter->request) == WRITE)) { + spin_unlock(&bdev->lock); + return false; + } + } + list_add(&scmrq->cluster.list, &bdev->cluster_list); + spin_unlock(&bdev->lock); + + return true; +} + +void scm_release_cluster(struct scm_request *scmrq) +{ + struct scm_blk_dev *bdev = scmrq->bdev; + unsigned long flags; + + if (write_cluster_size == 0) + return; + + spin_lock_irqsave(&bdev->lock, flags); + list_del(&scmrq->cluster.list); + spin_unlock_irqrestore(&bdev->lock, flags); +} + +void scm_blk_dev_cluster_setup(struct scm_blk_dev *bdev) +{ + INIT_LIST_HEAD(&bdev->cluster_list); + blk_queue_io_opt(bdev->rq, CLUSTER_SIZE); +} + +static void scm_prepare_cluster_request(struct scm_request *scmrq) +{ + struct scm_blk_dev *bdev = scmrq->bdev; + struct scm_device *scmdev = bdev->gendisk->private_data; + struct request *req = scmrq->request; + struct aidaw *aidaw = scmrq->aidaw; + struct msb *msb = &scmrq->aob->msb[0]; + struct req_iterator iter; + struct bio_vec *bv; + int i = 0; + u64 addr; + + switch (scmrq->cluster.state) { + case CLUSTER_NONE: + scmrq->cluster.state = CLUSTER_READ; + /* fall through */ + case CLUSTER_READ: + scmrq->aob->request.msb_count = 1; + msb->bs = MSB_BS_4K; + msb->oc = MSB_OC_READ; + msb->flags = MSB_FLAG_IDA; + msb->data_addr = (u64) aidaw; + msb->blk_count = write_cluster_size; + + addr = scmdev->address + ((u64) blk_rq_pos(req) << 9); + msb->scm_addr = round_down(addr, CLUSTER_SIZE); + + if (msb->scm_addr != + round_down(addr + (u64) blk_rq_bytes(req) - 1, + CLUSTER_SIZE)) + msb->blk_count = 2 * write_cluster_size; + + for (i = 0; i < msb->blk_count; i++) { + aidaw->data_addr = (u64) scmrq->cluster.buf[i]; + aidaw++; + } + + break; + case CLUSTER_WRITE: + msb->oc = MSB_OC_WRITE; + + for (addr = msb->scm_addr; + addr < scmdev->address + ((u64) blk_rq_pos(req) << 9); + addr += PAGE_SIZE) { + aidaw->data_addr = (u64) scmrq->cluster.buf[i]; + aidaw++; + i++; + } + rq_for_each_segment(bv, req, iter) { + aidaw->data_addr = (u64) page_address(bv->bv_page); + aidaw++; + i++; + } + for (; i < msb->blk_count; i++) { + aidaw->data_addr = (u64) scmrq->cluster.buf[i]; + aidaw++; + } + break; + } +} + +bool scm_need_cluster_request(struct scm_request *scmrq) +{ + if (rq_data_dir(scmrq->request) == READ) + return false; + + return blk_rq_bytes(scmrq->request) < CLUSTER_SIZE; +} + +/* Called with queue lock held. */ +void scm_initiate_cluster_request(struct scm_request *scmrq) +{ + scm_prepare_cluster_request(scmrq); + if (scm_start_aob(scmrq->aob)) + scm_request_requeue(scmrq); +} + +bool scm_test_cluster_request(struct scm_request *scmrq) +{ + return scmrq->cluster.state != CLUSTER_NONE; +} + +void scm_cluster_request_irq(struct scm_request *scmrq) +{ + struct scm_blk_dev *bdev = scmrq->bdev; + unsigned long flags; + + switch (scmrq->cluster.state) { + case CLUSTER_NONE: + BUG(); + break; + case CLUSTER_READ: + if (scmrq->error) { + scm_request_finish(scmrq); + break; + } + scmrq->cluster.state = CLUSTER_WRITE; + spin_lock_irqsave(&bdev->rq_lock, flags); + scm_initiate_cluster_request(scmrq); + spin_unlock_irqrestore(&bdev->rq_lock, flags); + break; + case CLUSTER_WRITE: + scm_request_finish(scmrq); + break; + } +} + +bool scm_cluster_size_valid(void) +{ + return write_cluster_size == 0 || write_cluster_size == 32 || + write_cluster_size == 64 || write_cluster_size == 128; +} diff --git a/drivers/s390/block/scm_drv.c b/drivers/s390/block/scm_drv.c new file mode 100644 index 00000000000..9fa0a908607 --- /dev/null +++ b/drivers/s390/block/scm_drv.c @@ -0,0 +1,81 @@ +/* + * Device driver for s390 storage class memory. + * + * Copyright IBM Corp. 2012 + * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com> + */ + +#define KMSG_COMPONENT "scm_block" +#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt + +#include <linux/module.h> +#include <linux/slab.h> +#include <asm/eadm.h> +#include "scm_blk.h" + +static void notify(struct scm_device *scmdev) +{ + pr_info("%lu: The capabilities of the SCM increment changed\n", + (unsigned long) scmdev->address); + SCM_LOG(2, "State changed"); + SCM_LOG_STATE(2, scmdev); +} + +static int scm_probe(struct scm_device *scmdev) +{ + struct scm_blk_dev *bdev; + int ret; + + SCM_LOG(2, "probe"); + SCM_LOG_STATE(2, scmdev); + + if (scmdev->attrs.oper_state != OP_STATE_GOOD) + return -EINVAL; + + bdev = kzalloc(sizeof(*bdev), GFP_KERNEL); + if (!bdev) + return -ENOMEM; + + dev_set_drvdata(&scmdev->dev, bdev); + ret = scm_blk_dev_setup(bdev, scmdev); + if (ret) { + dev_set_drvdata(&scmdev->dev, NULL); + kfree(bdev); + goto out; + } + +out: + return ret; +} + +static int scm_remove(struct scm_device *scmdev) +{ + struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev); + + scm_blk_dev_cleanup(bdev); + dev_set_drvdata(&scmdev->dev, NULL); + kfree(bdev); + + return 0; +} + +static struct scm_driver scm_drv = { + .drv = { + .name = "scm_block", + .owner = THIS_MODULE, + }, + .notify = notify, + .probe = scm_probe, + .remove = scm_remove, + .handler = scm_blk_irq, +}; + +int __init scm_drv_init(void) +{ + return scm_driver_register(&scm_drv); +} + +void scm_drv_cleanup(void) +{ + scm_driver_unregister(&scm_drv); +} |