1
2
3
4
5
6
7
8
9#define KMSG_COMPONENT "scm_block"
10#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
11
12#include <linux/interrupt.h>
13#include <linux/spinlock.h>
14#include <linux/mempool.h>
15#include <linux/module.h>
16#include <linux/blkdev.h>
17#include <linux/blk-mq.h>
18#include <linux/genhd.h>
19#include <linux/slab.h>
20#include <linux/list.h>
21#include <asm/eadm.h>
22#include "scm_blk.h"
23
24debug_info_t *scm_debug;
25static int scm_major;
26static mempool_t *aidaw_pool;
27static DEFINE_SPINLOCK(list_lock);
28static LIST_HEAD(inactive_requests);
29static unsigned int nr_requests = 64;
30static unsigned int nr_requests_per_io = 8;
31static atomic_t nr_devices = ATOMIC_INIT(0);
32module_param(nr_requests, uint, S_IRUGO);
33MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");
34
35module_param(nr_requests_per_io, uint, S_IRUGO);
36MODULE_PARM_DESC(nr_requests_per_io, "Number of requests per IO.");
37
38MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
39MODULE_LICENSE("GPL");
40MODULE_ALIAS("scm:scmdev*");
41
42static void __scm_free_rq(struct scm_request *scmrq)
43{
44 struct aob_rq_header *aobrq = to_aobrq(scmrq);
45
46 free_page((unsigned long) scmrq->aob);
47 kfree(scmrq->request);
48 kfree(aobrq);
49}
50
51static void scm_free_rqs(void)
52{
53 struct list_head *iter, *safe;
54 struct scm_request *scmrq;
55
56 spin_lock_irq(&list_lock);
57 list_for_each_safe(iter, safe, &inactive_requests) {
58 scmrq = list_entry(iter, struct scm_request, list);
59 list_del(&scmrq->list);
60 __scm_free_rq(scmrq);
61 }
62 spin_unlock_irq(&list_lock);
63
64 mempool_destroy(aidaw_pool);
65}
66
67static int __scm_alloc_rq(void)
68{
69 struct aob_rq_header *aobrq;
70 struct scm_request *scmrq;
71
72 aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
73 if (!aobrq)
74 return -ENOMEM;
75
76 scmrq = (void *) aobrq->data;
77 scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
78 if (!scmrq->aob)
79 goto free;
80
81 scmrq->request = kcalloc(nr_requests_per_io, sizeof(scmrq->request[0]),
82 GFP_KERNEL);
83 if (!scmrq->request)
84 goto free;
85
86 INIT_LIST_HEAD(&scmrq->list);
87 spin_lock_irq(&list_lock);
88 list_add(&scmrq->list, &inactive_requests);
89 spin_unlock_irq(&list_lock);
90
91 return 0;
92free:
93 __scm_free_rq(scmrq);
94 return -ENOMEM;
95}
96
97static int scm_alloc_rqs(unsigned int nrqs)
98{
99 int ret = 0;
100
101 aidaw_pool = mempool_create_page_pool(max(nrqs/8, 1U), 0);
102 if (!aidaw_pool)
103 return -ENOMEM;
104
105 while (nrqs-- && !ret)
106 ret = __scm_alloc_rq();
107
108 return ret;
109}
110
111static struct scm_request *scm_request_fetch(void)
112{
113 struct scm_request *scmrq = NULL;
114
115 spin_lock_irq(&list_lock);
116 if (list_empty(&inactive_requests))
117 goto out;
118 scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
119 list_del(&scmrq->list);
120out:
121 spin_unlock_irq(&list_lock);
122 return scmrq;
123}
124
125static void scm_request_done(struct scm_request *scmrq)
126{
127 unsigned long flags;
128 struct msb *msb;
129 u64 aidaw;
130 int i;
131
132 for (i = 0; i < nr_requests_per_io && scmrq->request[i]; i++) {
133 msb = &scmrq->aob->msb[i];
134 aidaw = msb->data_addr;
135
136 if ((msb->flags & MSB_FLAG_IDA) && aidaw &&
137 IS_ALIGNED(aidaw, PAGE_SIZE))
138 mempool_free(virt_to_page(aidaw), aidaw_pool);
139 }
140
141 spin_lock_irqsave(&list_lock, flags);
142 list_add(&scmrq->list, &inactive_requests);
143 spin_unlock_irqrestore(&list_lock, flags);
144}
145
146static bool scm_permit_request(struct scm_blk_dev *bdev, struct request *req)
147{
148 return rq_data_dir(req) != WRITE || bdev->state != SCM_WR_PROHIBIT;
149}
150
151static inline struct aidaw *scm_aidaw_alloc(void)
152{
153 struct page *page = mempool_alloc(aidaw_pool, GFP_ATOMIC);
154
155 return page ? page_address(page) : NULL;
156}
157
158static inline unsigned long scm_aidaw_bytes(struct aidaw *aidaw)
159{
160 unsigned long _aidaw = (unsigned long) aidaw;
161 unsigned long bytes = ALIGN(_aidaw, PAGE_SIZE) - _aidaw;
162
163 return (bytes / sizeof(*aidaw)) * PAGE_SIZE;
164}
165
166struct aidaw *scm_aidaw_fetch(struct scm_request *scmrq, unsigned int bytes)
167{
168 struct aidaw *aidaw;
169
170 if (scm_aidaw_bytes(scmrq->next_aidaw) >= bytes)
171 return scmrq->next_aidaw;
172
173 aidaw = scm_aidaw_alloc();
174 if (aidaw)
175 memset(aidaw, 0, PAGE_SIZE);
176 return aidaw;
177}
178
179static int scm_request_prepare(struct scm_request *scmrq)
180{
181 struct scm_blk_dev *bdev = scmrq->bdev;
182 struct scm_device *scmdev = bdev->gendisk->private_data;
183 int pos = scmrq->aob->request.msb_count;
184 struct msb *msb = &scmrq->aob->msb[pos];
185 struct request *req = scmrq->request[pos];
186 struct req_iterator iter;
187 struct aidaw *aidaw;
188 struct bio_vec bv;
189
190 aidaw = scm_aidaw_fetch(scmrq, blk_rq_bytes(req));
191 if (!aidaw)
192 return -ENOMEM;
193
194 msb->bs = MSB_BS_4K;
195 scmrq->aob->request.msb_count++;
196 msb->scm_addr = scmdev->address + ((u64) blk_rq_pos(req) << 9);
197 msb->oc = (rq_data_dir(req) == READ) ? MSB_OC_READ : MSB_OC_WRITE;
198 msb->flags |= MSB_FLAG_IDA;
199 msb->data_addr = (u64) aidaw;
200
201 rq_for_each_segment(bv, req, iter) {
202 WARN_ON(bv.bv_offset);
203 msb->blk_count += bv.bv_len >> 12;
204 aidaw->data_addr = (u64) page_address(bv.bv_page);
205 aidaw++;
206 }
207
208 scmrq->next_aidaw = aidaw;
209 return 0;
210}
211
212static inline void scm_request_set(struct scm_request *scmrq,
213 struct request *req)
214{
215 scmrq->request[scmrq->aob->request.msb_count] = req;
216}
217
218static inline void scm_request_init(struct scm_blk_dev *bdev,
219 struct scm_request *scmrq)
220{
221 struct aob_rq_header *aobrq = to_aobrq(scmrq);
222 struct aob *aob = scmrq->aob;
223
224 memset(scmrq->request, 0,
225 nr_requests_per_io * sizeof(scmrq->request[0]));
226 memset(aob, 0, sizeof(*aob));
227 aobrq->scmdev = bdev->scmdev;
228 aob->request.cmd_code = ARQB_CMD_MOVE;
229 aob->request.data = (u64) aobrq;
230 scmrq->bdev = bdev;
231 scmrq->retries = 4;
232 scmrq->error = BLK_STS_OK;
233
234 scmrq->next_aidaw = (void *) &aob->msb[nr_requests_per_io];
235}
236
237static void scm_request_requeue(struct scm_request *scmrq)
238{
239 struct scm_blk_dev *bdev = scmrq->bdev;
240 int i;
241
242 for (i = 0; i < nr_requests_per_io && scmrq->request[i]; i++)
243 blk_mq_requeue_request(scmrq->request[i], false);
244
245 atomic_dec(&bdev->queued_reqs);
246 scm_request_done(scmrq);
247 blk_mq_kick_requeue_list(bdev->rq);
248}
249
250static void scm_request_finish(struct scm_request *scmrq)
251{
252 struct scm_blk_dev *bdev = scmrq->bdev;
253 blk_status_t *error;
254 int i;
255
256 for (i = 0; i < nr_requests_per_io && scmrq->request[i]; i++) {
257 error = blk_mq_rq_to_pdu(scmrq->request[i]);
258 *error = scmrq->error;
259 blk_mq_complete_request(scmrq->request[i]);
260 }
261
262 atomic_dec(&bdev->queued_reqs);
263 scm_request_done(scmrq);
264}
265
266static void scm_request_start(struct scm_request *scmrq)
267{
268 struct scm_blk_dev *bdev = scmrq->bdev;
269
270 atomic_inc(&bdev->queued_reqs);
271 if (eadm_start_aob(scmrq->aob)) {
272 SCM_LOG(5, "no subchannel");
273 scm_request_requeue(scmrq);
274 }
275}
276
277struct scm_queue {
278 struct scm_request *scmrq;
279 spinlock_t lock;
280};
281
282static blk_status_t scm_blk_request(struct blk_mq_hw_ctx *hctx,
283 const struct blk_mq_queue_data *qd)
284{
285 struct scm_device *scmdev = hctx->queue->queuedata;
286 struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
287 struct scm_queue *sq = hctx->driver_data;
288 struct request *req = qd->rq;
289 struct scm_request *scmrq;
290
291 spin_lock(&sq->lock);
292 if (!scm_permit_request(bdev, req)) {
293 spin_unlock(&sq->lock);
294 return BLK_STS_RESOURCE;
295 }
296
297 scmrq = sq->scmrq;
298 if (!scmrq) {
299 scmrq = scm_request_fetch();
300 if (!scmrq) {
301 SCM_LOG(5, "no request");
302 spin_unlock(&sq->lock);
303 return BLK_STS_RESOURCE;
304 }
305 scm_request_init(bdev, scmrq);
306 sq->scmrq = scmrq;
307 }
308 scm_request_set(scmrq, req);
309
310 if (scm_request_prepare(scmrq)) {
311 SCM_LOG(5, "aidaw alloc failed");
312 scm_request_set(scmrq, NULL);
313
314 if (scmrq->aob->request.msb_count)
315 scm_request_start(scmrq);
316
317 sq->scmrq = NULL;
318 spin_unlock(&sq->lock);
319 return BLK_STS_RESOURCE;
320 }
321 blk_mq_start_request(req);
322
323 if (qd->last || scmrq->aob->request.msb_count == nr_requests_per_io) {
324 scm_request_start(scmrq);
325 sq->scmrq = NULL;
326 }
327 spin_unlock(&sq->lock);
328 return BLK_STS_OK;
329}
330
331static int scm_blk_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
332 unsigned int idx)
333{
334 struct scm_queue *qd = kzalloc(sizeof(*qd), GFP_KERNEL);
335
336 if (!qd)
337 return -ENOMEM;
338
339 spin_lock_init(&qd->lock);
340 hctx->driver_data = qd;
341
342 return 0;
343}
344
345static void scm_blk_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int idx)
346{
347 struct scm_queue *qd = hctx->driver_data;
348
349 WARN_ON(qd->scmrq);
350 kfree(hctx->driver_data);
351 hctx->driver_data = NULL;
352}
353
354static void __scmrq_log_error(struct scm_request *scmrq)
355{
356 struct aob *aob = scmrq->aob;
357
358 if (scmrq->error == BLK_STS_TIMEOUT)
359 SCM_LOG(1, "Request timeout");
360 else {
361 SCM_LOG(1, "Request error");
362 SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
363 }
364 if (scmrq->retries)
365 SCM_LOG(1, "Retry request");
366 else
367 pr_err("An I/O operation to SCM failed with rc=%d\n",
368 scmrq->error);
369}
370
371static void scm_blk_handle_error(struct scm_request *scmrq)
372{
373 struct scm_blk_dev *bdev = scmrq->bdev;
374 unsigned long flags;
375
376 if (scmrq->error != BLK_STS_IOERR)
377 goto restart;
378
379
380 switch (scmrq->aob->response.eqc) {
381 case EQC_WR_PROHIBIT:
382 spin_lock_irqsave(&bdev->lock, flags);
383 if (bdev->state != SCM_WR_PROHIBIT)
384 pr_info("%lx: Write access to the SCM increment is suspended\n",
385 (unsigned long) bdev->scmdev->address);
386 bdev->state = SCM_WR_PROHIBIT;
387 spin_unlock_irqrestore(&bdev->lock, flags);
388 goto requeue;
389 default:
390 break;
391 }
392
393restart:
394 if (!eadm_start_aob(scmrq->aob))
395 return;
396
397requeue:
398 scm_request_requeue(scmrq);
399}
400
401void scm_blk_irq(struct scm_device *scmdev, void *data, blk_status_t error)
402{
403 struct scm_request *scmrq = data;
404
405 scmrq->error = error;
406 if (error) {
407 __scmrq_log_error(scmrq);
408 if (scmrq->retries-- > 0) {
409 scm_blk_handle_error(scmrq);
410 return;
411 }
412 }
413
414 scm_request_finish(scmrq);
415}
416
417static void scm_blk_request_done(struct request *req)
418{
419 blk_status_t *error = blk_mq_rq_to_pdu(req);
420
421 blk_mq_end_request(req, *error);
422}
423
424static const struct block_device_operations scm_blk_devops = {
425 .owner = THIS_MODULE,
426};
427
428static const struct blk_mq_ops scm_mq_ops = {
429 .queue_rq = scm_blk_request,
430 .complete = scm_blk_request_done,
431 .init_hctx = scm_blk_init_hctx,
432 .exit_hctx = scm_blk_exit_hctx,
433};
434
435int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
436{
437 unsigned int devindex, nr_max_blk;
438 struct request_queue *rq;
439 int len, ret;
440
441 devindex = atomic_inc_return(&nr_devices) - 1;
442
443 if (devindex > 701) {
444 ret = -ENODEV;
445 goto out;
446 }
447
448 bdev->scmdev = scmdev;
449 bdev->state = SCM_OPER;
450 spin_lock_init(&bdev->lock);
451 atomic_set(&bdev->queued_reqs, 0);
452
453 bdev->tag_set.ops = &scm_mq_ops;
454 bdev->tag_set.cmd_size = sizeof(blk_status_t);
455 bdev->tag_set.nr_hw_queues = nr_requests;
456 bdev->tag_set.queue_depth = nr_requests_per_io * nr_requests;
457 bdev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
458
459 ret = blk_mq_alloc_tag_set(&bdev->tag_set);
460 if (ret)
461 goto out;
462
463 rq = blk_mq_init_queue(&bdev->tag_set);
464 if (IS_ERR(rq)) {
465 ret = PTR_ERR(rq);
466 goto out_tag;
467 }
468 bdev->rq = rq;
469 nr_max_blk = min(scmdev->nr_max_block,
470 (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));
471
472 blk_queue_logical_block_size(rq, 1 << 12);
473 blk_queue_max_hw_sectors(rq, nr_max_blk << 3);
474 blk_queue_max_segments(rq, nr_max_blk);
475 blk_queue_flag_set(QUEUE_FLAG_NONROT, rq);
476 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, rq);
477
478 bdev->gendisk = alloc_disk(SCM_NR_PARTS);
479 if (!bdev->gendisk) {
480 ret = -ENOMEM;
481 goto out_queue;
482 }
483 rq->queuedata = scmdev;
484 bdev->gendisk->private_data = scmdev;
485 bdev->gendisk->fops = &scm_blk_devops;
486 bdev->gendisk->queue = rq;
487 bdev->gendisk->major = scm_major;
488 bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;
489
490 len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
491 if (devindex > 25) {
492 len += snprintf(bdev->gendisk->disk_name + len,
493 DISK_NAME_LEN - len, "%c",
494 'a' + (devindex / 26) - 1);
495 devindex = devindex % 26;
496 }
497 snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
498 'a' + devindex);
499
500
501 set_capacity(bdev->gendisk, scmdev->size >> 9);
502 device_add_disk(&scmdev->dev, bdev->gendisk);
503 return 0;
504
505out_queue:
506 blk_cleanup_queue(rq);
507out_tag:
508 blk_mq_free_tag_set(&bdev->tag_set);
509out:
510 atomic_dec(&nr_devices);
511 return ret;
512}
513
514void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
515{
516 del_gendisk(bdev->gendisk);
517 blk_cleanup_queue(bdev->gendisk->queue);
518 blk_mq_free_tag_set(&bdev->tag_set);
519 put_disk(bdev->gendisk);
520}
521
522void scm_blk_set_available(struct scm_blk_dev *bdev)
523{
524 unsigned long flags;
525
526 spin_lock_irqsave(&bdev->lock, flags);
527 if (bdev->state == SCM_WR_PROHIBIT)
528 pr_info("%lx: Write access to the SCM increment is restored\n",
529 (unsigned long) bdev->scmdev->address);
530 bdev->state = SCM_OPER;
531 spin_unlock_irqrestore(&bdev->lock, flags);
532}
533
534static bool __init scm_blk_params_valid(void)
535{
536 if (!nr_requests_per_io || nr_requests_per_io > 64)
537 return false;
538
539 return true;
540}
541
542static int __init scm_blk_init(void)
543{
544 int ret = -EINVAL;
545
546 if (!scm_blk_params_valid())
547 goto out;
548
549 ret = register_blkdev(0, "scm");
550 if (ret < 0)
551 goto out;
552
553 scm_major = ret;
554 ret = scm_alloc_rqs(nr_requests);
555 if (ret)
556 goto out_free;
557
558 scm_debug = debug_register("scm_log", 16, 1, 16);
559 if (!scm_debug) {
560 ret = -ENOMEM;
561 goto out_free;
562 }
563
564 debug_register_view(scm_debug, &debug_hex_ascii_view);
565 debug_set_level(scm_debug, 2);
566
567 ret = scm_drv_init();
568 if (ret)
569 goto out_dbf;
570
571 return ret;
572
573out_dbf:
574 debug_unregister(scm_debug);
575out_free:
576 scm_free_rqs();
577 unregister_blkdev(scm_major, "scm");
578out:
579 return ret;
580}
581module_init(scm_blk_init);
582
583static void __exit scm_blk_cleanup(void)
584{
585 scm_drv_cleanup();
586 debug_unregister(scm_debug);
587 scm_free_rqs();
588 unregister_blkdev(scm_major, "scm");
589}
590module_exit(scm_blk_cleanup);
591
