2 * Block driver for s390 storage class memory.
4 * Copyright IBM Corp. 2012
5 * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
8 #define KMSG_COMPONENT "scm_block"
9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
11 #include <linux/interrupt.h>
12 #include <linux/spinlock.h>
13 #include <linux/mempool.h>
14 #include <linux/module.h>
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/genhd.h>
18 #include <linux/slab.h>
19 #include <linux/list.h>
23 debug_info_t *scm_debug;
25 static mempool_t *aidaw_pool;
26 static DEFINE_SPINLOCK(list_lock);
27 static LIST_HEAD(inactive_requests);
28 static unsigned int nr_requests = 64;
29 static unsigned int nr_requests_per_io = 8;
30 static atomic_t nr_devices = ATOMIC_INIT(0);
31 module_param(nr_requests, uint, S_IRUGO);
32 MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");
34 module_param(nr_requests_per_io, uint, S_IRUGO);
35 MODULE_PARM_DESC(nr_requests_per_io, "Number of requests per IO.");
37 MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
38 MODULE_LICENSE("GPL");
39 MODULE_ALIAS("scm:scmdev*");
41 static void __scm_free_rq(struct scm_request *scmrq)
43 struct aob_rq_header *aobrq = to_aobrq(scmrq);
45 free_page((unsigned long) scmrq->aob);
46 kfree(scmrq->request);
50 static void scm_free_rqs(void)
52 struct list_head *iter, *safe;
53 struct scm_request *scmrq;
55 spin_lock_irq(&list_lock);
56 list_for_each_safe(iter, safe, &inactive_requests) {
57 scmrq = list_entry(iter, struct scm_request, list);
58 list_del(&scmrq->list);
61 spin_unlock_irq(&list_lock);
63 mempool_destroy(aidaw_pool);
66 static int __scm_alloc_rq(void)
68 struct aob_rq_header *aobrq;
69 struct scm_request *scmrq;
71 aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
75 scmrq = (void *) aobrq->data;
76 scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
80 scmrq->request = kcalloc(nr_requests_per_io, sizeof(scmrq->request[0]),
85 INIT_LIST_HEAD(&scmrq->list);
86 spin_lock_irq(&list_lock);
87 list_add(&scmrq->list, &inactive_requests);
88 spin_unlock_irq(&list_lock);
96 static int scm_alloc_rqs(unsigned int nrqs)
100 aidaw_pool = mempool_create_page_pool(max(nrqs/8, 1U), 0);
104 while (nrqs-- && !ret)
105 ret = __scm_alloc_rq();
110 static struct scm_request *scm_request_fetch(void)
112 struct scm_request *scmrq = NULL;
114 spin_lock_irq(&list_lock);
115 if (list_empty(&inactive_requests))
117 scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
118 list_del(&scmrq->list);
120 spin_unlock_irq(&list_lock);
124 static void scm_request_done(struct scm_request *scmrq)
131 for (i = 0; i < nr_requests_per_io && scmrq->request[i]; i++) {
132 msb = &scmrq->aob->msb[i];
133 aidaw = msb->data_addr;
135 if ((msb->flags & MSB_FLAG_IDA) && aidaw &&
136 IS_ALIGNED(aidaw, PAGE_SIZE))
137 mempool_free(virt_to_page(aidaw), aidaw_pool);
140 spin_lock_irqsave(&list_lock, flags);
141 list_add(&scmrq->list, &inactive_requests);
142 spin_unlock_irqrestore(&list_lock, flags);
145 static bool scm_permit_request(struct scm_blk_dev *bdev, struct request *req)
147 return rq_data_dir(req) != WRITE || bdev->state != SCM_WR_PROHIBIT;
150 static inline struct aidaw *scm_aidaw_alloc(void)
152 struct page *page = mempool_alloc(aidaw_pool, GFP_ATOMIC);
154 return page ? page_address(page) : NULL;
157 static inline unsigned long scm_aidaw_bytes(struct aidaw *aidaw)
159 unsigned long _aidaw = (unsigned long) aidaw;
160 unsigned long bytes = ALIGN(_aidaw, PAGE_SIZE) - _aidaw;
162 return (bytes / sizeof(*aidaw)) * PAGE_SIZE;
165 struct aidaw *scm_aidaw_fetch(struct scm_request *scmrq, unsigned int bytes)
169 if (scm_aidaw_bytes(scmrq->next_aidaw) >= bytes)
170 return scmrq->next_aidaw;
172 aidaw = scm_aidaw_alloc();
174 memset(aidaw, 0, PAGE_SIZE);
178 static int scm_request_prepare(struct scm_request *scmrq)
180 struct scm_blk_dev *bdev = scmrq->bdev;
181 struct scm_device *scmdev = bdev->gendisk->private_data;
182 int pos = scmrq->aob->request.msb_count;
183 struct msb *msb = &scmrq->aob->msb[pos];
184 struct request *req = scmrq->request[pos];
185 struct req_iterator iter;
189 aidaw = scm_aidaw_fetch(scmrq, blk_rq_bytes(req));
194 scmrq->aob->request.msb_count++;
195 msb->scm_addr = scmdev->address + ((u64) blk_rq_pos(req) << 9);
196 msb->oc = (rq_data_dir(req) == READ) ? MSB_OC_READ : MSB_OC_WRITE;
197 msb->flags |= MSB_FLAG_IDA;
198 msb->data_addr = (u64) aidaw;
200 rq_for_each_segment(bv, req, iter) {
201 WARN_ON(bv.bv_offset);
202 msb->blk_count += bv.bv_len >> 12;
203 aidaw->data_addr = (u64) page_address(bv.bv_page);
207 scmrq->next_aidaw = aidaw;
211 static inline void scm_request_set(struct scm_request *scmrq,
214 scmrq->request[scmrq->aob->request.msb_count] = req;
217 static inline void scm_request_init(struct scm_blk_dev *bdev,
218 struct scm_request *scmrq)
220 struct aob_rq_header *aobrq = to_aobrq(scmrq);
221 struct aob *aob = scmrq->aob;
223 memset(scmrq->request, 0,
224 nr_requests_per_io * sizeof(scmrq->request[0]));
225 memset(aob, 0, sizeof(*aob));
226 aobrq->scmdev = bdev->scmdev;
227 aob->request.cmd_code = ARQB_CMD_MOVE;
228 aob->request.data = (u64) aobrq;
231 scmrq->error = BLK_STS_OK;
232 /* We don't use all msbs - place aidaws at the end of the aob page. */
233 scmrq->next_aidaw = (void *) &aob->msb[nr_requests_per_io];
236 static void scm_request_requeue(struct scm_request *scmrq)
238 struct scm_blk_dev *bdev = scmrq->bdev;
241 for (i = 0; i < nr_requests_per_io && scmrq->request[i]; i++)
242 blk_mq_requeue_request(scmrq->request[i], false);
244 atomic_dec(&bdev->queued_reqs);
245 scm_request_done(scmrq);
246 blk_mq_kick_requeue_list(bdev->rq);
249 static void scm_request_finish(struct scm_request *scmrq)
251 struct scm_blk_dev *bdev = scmrq->bdev;
255 for (i = 0; i < nr_requests_per_io && scmrq->request[i]; i++) {
256 error = blk_mq_rq_to_pdu(scmrq->request[i]);
257 *error = scmrq->error;
258 blk_mq_complete_request(scmrq->request[i]);
261 atomic_dec(&bdev->queued_reqs);
262 scm_request_done(scmrq);
265 static void scm_request_start(struct scm_request *scmrq)
267 struct scm_blk_dev *bdev = scmrq->bdev;
269 atomic_inc(&bdev->queued_reqs);
270 if (eadm_start_aob(scmrq->aob)) {
271 SCM_LOG(5, "no subchannel");
272 scm_request_requeue(scmrq);
277 struct scm_request *scmrq;
281 static blk_status_t scm_blk_request(struct blk_mq_hw_ctx *hctx,
282 const struct blk_mq_queue_data *qd)
284 struct scm_device *scmdev = hctx->queue->queuedata;
285 struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
286 struct scm_queue *sq = hctx->driver_data;
287 struct request *req = qd->rq;
288 struct scm_request *scmrq;
290 spin_lock(&sq->lock);
291 if (!scm_permit_request(bdev, req)) {
292 spin_unlock(&sq->lock);
293 return BLK_STS_RESOURCE;
298 scmrq = scm_request_fetch();
300 SCM_LOG(5, "no request");
301 spin_unlock(&sq->lock);
302 return BLK_STS_RESOURCE;
304 scm_request_init(bdev, scmrq);
307 scm_request_set(scmrq, req);
309 if (scm_request_prepare(scmrq)) {
310 SCM_LOG(5, "aidaw alloc failed");
311 scm_request_set(scmrq, NULL);
313 if (scmrq->aob->request.msb_count)
314 scm_request_start(scmrq);
317 spin_unlock(&sq->lock);
318 return BLK_STS_RESOURCE;
320 blk_mq_start_request(req);
322 if (qd->last || scmrq->aob->request.msb_count == nr_requests_per_io) {
323 scm_request_start(scmrq);
326 spin_unlock(&sq->lock);
330 static int scm_blk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
333 struct scm_queue *qd = kzalloc(sizeof(*qd), GFP_KERNEL);
338 spin_lock_init(&qd->lock);
339 hctx->driver_data = qd;
344 static void scm_blk_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int idx)
346 struct scm_queue *qd = hctx->driver_data;
349 kfree(hctx->driver_data);
350 hctx->driver_data = NULL;
353 static void __scmrq_log_error(struct scm_request *scmrq)
355 struct aob *aob = scmrq->aob;
357 if (scmrq->error == BLK_STS_TIMEOUT)
358 SCM_LOG(1, "Request timeout");
360 SCM_LOG(1, "Request error");
361 SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
364 SCM_LOG(1, "Retry request");
366 pr_err("An I/O operation to SCM failed with rc=%d\n",
370 static void scm_blk_handle_error(struct scm_request *scmrq)
372 struct scm_blk_dev *bdev = scmrq->bdev;
375 if (scmrq->error != BLK_STS_IOERR)
378 /* For -EIO the response block is valid. */
379 switch (scmrq->aob->response.eqc) {
380 case EQC_WR_PROHIBIT:
381 spin_lock_irqsave(&bdev->lock, flags);
382 if (bdev->state != SCM_WR_PROHIBIT)
383 pr_info("%lx: Write access to the SCM increment is suspended\n",
384 (unsigned long) bdev->scmdev->address);
385 bdev->state = SCM_WR_PROHIBIT;
386 spin_unlock_irqrestore(&bdev->lock, flags);
393 if (!eadm_start_aob(scmrq->aob))
397 scm_request_requeue(scmrq);
400 void scm_blk_irq(struct scm_device *scmdev, void *data, blk_status_t error)
402 struct scm_request *scmrq = data;
404 scmrq->error = error;
406 __scmrq_log_error(scmrq);
407 if (scmrq->retries-- > 0) {
408 scm_blk_handle_error(scmrq);
413 scm_request_finish(scmrq);
416 static void scm_blk_request_done(struct request *req)
418 blk_status_t *error = blk_mq_rq_to_pdu(req);
420 blk_mq_end_request(req, *error);
423 static const struct block_device_operations scm_blk_devops = {
424 .owner = THIS_MODULE,
427 static const struct blk_mq_ops scm_mq_ops = {
428 .queue_rq = scm_blk_request,
429 .complete = scm_blk_request_done,
430 .init_hctx = scm_blk_init_hctx,
431 .exit_hctx = scm_blk_exit_hctx,
434 int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
436 unsigned int devindex, nr_max_blk;
437 struct request_queue *rq;
440 devindex = atomic_inc_return(&nr_devices) - 1;
441 /* scma..scmz + scmaa..scmzz */
442 if (devindex > 701) {
447 bdev->scmdev = scmdev;
448 bdev->state = SCM_OPER;
449 spin_lock_init(&bdev->lock);
450 atomic_set(&bdev->queued_reqs, 0);
452 bdev->tag_set.ops = &scm_mq_ops;
453 bdev->tag_set.cmd_size = sizeof(blk_status_t);
454 bdev->tag_set.nr_hw_queues = nr_requests;
455 bdev->tag_set.queue_depth = nr_requests_per_io * nr_requests;
456 bdev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
458 ret = blk_mq_alloc_tag_set(&bdev->tag_set);
462 rq = blk_mq_init_queue(&bdev->tag_set);
468 nr_max_blk = min(scmdev->nr_max_block,
469 (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));
471 blk_queue_logical_block_size(rq, 1 << 12);
472 blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */
473 blk_queue_max_segments(rq, nr_max_blk);
474 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq);
475 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, rq);
477 bdev->gendisk = alloc_disk(SCM_NR_PARTS);
478 if (!bdev->gendisk) {
482 rq->queuedata = scmdev;
483 bdev->gendisk->private_data = scmdev;
484 bdev->gendisk->fops = &scm_blk_devops;
485 bdev->gendisk->queue = rq;
486 bdev->gendisk->major = scm_major;
487 bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;
489 len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
491 len += snprintf(bdev->gendisk->disk_name + len,
492 DISK_NAME_LEN - len, "%c",
493 'a' + (devindex / 26) - 1);
494 devindex = devindex % 26;
496 snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
499 /* 512 byte sectors */
500 set_capacity(bdev->gendisk, scmdev->size >> 9);
501 device_add_disk(&scmdev->dev, bdev->gendisk);
505 blk_cleanup_queue(rq);
507 blk_mq_free_tag_set(&bdev->tag_set);
509 atomic_dec(&nr_devices);
513 void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
515 del_gendisk(bdev->gendisk);
516 blk_cleanup_queue(bdev->gendisk->queue);
517 blk_mq_free_tag_set(&bdev->tag_set);
518 put_disk(bdev->gendisk);
521 void scm_blk_set_available(struct scm_blk_dev *bdev)
525 spin_lock_irqsave(&bdev->lock, flags);
526 if (bdev->state == SCM_WR_PROHIBIT)
527 pr_info("%lx: Write access to the SCM increment is restored\n",
528 (unsigned long) bdev->scmdev->address);
529 bdev->state = SCM_OPER;
530 spin_unlock_irqrestore(&bdev->lock, flags);
533 static bool __init scm_blk_params_valid(void)
535 if (!nr_requests_per_io || nr_requests_per_io > 64)
541 static int __init scm_blk_init(void)
545 if (!scm_blk_params_valid())
548 ret = register_blkdev(0, "scm");
553 ret = scm_alloc_rqs(nr_requests);
557 scm_debug = debug_register("scm_log", 16, 1, 16);
563 debug_register_view(scm_debug, &debug_hex_ascii_view);
564 debug_set_level(scm_debug, 2);
566 ret = scm_drv_init();
573 debug_unregister(scm_debug);
576 unregister_blkdev(scm_major, "scm");
580 module_init(scm_blk_init);
582 static void __exit scm_blk_cleanup(void)
585 debug_unregister(scm_debug);
587 unregister_blkdev(scm_major, "scm");
589 module_exit(scm_blk_cleanup);