b303cab76a7f3787149837fdd3223b6d04c7d7bb
[linux-3.10.git] / drivers / s390 / block / scm_blk.c
1 /*
2  * Block driver for s390 storage class memory.
3  *
4  * Copyright IBM Corp. 2012
5  * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
6  */
7
8 #define KMSG_COMPONENT "scm_block"
9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
10
11 #include <linux/interrupt.h>
12 #include <linux/spinlock.h>
13 #include <linux/module.h>
14 #include <linux/blkdev.h>
15 #include <linux/genhd.h>
16 #include <linux/slab.h>
17 #include <linux/list.h>
18 #include <asm/eadm.h>
19 #include "scm_blk.h"
20
21 debug_info_t *scm_debug;
22 static int scm_major;
23 static DEFINE_SPINLOCK(list_lock);
24 static LIST_HEAD(inactive_requests);
25 static unsigned int nr_requests = 64;
26 static atomic_t nr_devices = ATOMIC_INIT(0);
27 module_param(nr_requests, uint, S_IRUGO);
28 MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");
29
30 MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
31 MODULE_LICENSE("GPL");
32 MODULE_ALIAS("scm:scmdev*");
33
34 static void __scm_free_rq(struct scm_request *scmrq)
35 {
36         struct aob_rq_header *aobrq = to_aobrq(scmrq);
37
38         free_page((unsigned long) scmrq->aob);
39         free_page((unsigned long) scmrq->aidaw);
40         __scm_free_rq_cluster(scmrq);
41         kfree(aobrq);
42 }
43
44 static void scm_free_rqs(void)
45 {
46         struct list_head *iter, *safe;
47         struct scm_request *scmrq;
48
49         spin_lock_irq(&list_lock);
50         list_for_each_safe(iter, safe, &inactive_requests) {
51                 scmrq = list_entry(iter, struct scm_request, list);
52                 list_del(&scmrq->list);
53                 __scm_free_rq(scmrq);
54         }
55         spin_unlock_irq(&list_lock);
56 }
57
58 static int __scm_alloc_rq(void)
59 {
60         struct aob_rq_header *aobrq;
61         struct scm_request *scmrq;
62
63         aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
64         if (!aobrq)
65                 return -ENOMEM;
66
67         scmrq = (void *) aobrq->data;
68         scmrq->aidaw = (void *) get_zeroed_page(GFP_DMA);
69         scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
70         if (!scmrq->aob || !scmrq->aidaw) {
71                 __scm_free_rq(scmrq);
72                 return -ENOMEM;
73         }
74
75         if (__scm_alloc_rq_cluster(scmrq)) {
76                 __scm_free_rq(scmrq);
77                 return -ENOMEM;
78         }
79
80         INIT_LIST_HEAD(&scmrq->list);
81         spin_lock_irq(&list_lock);
82         list_add(&scmrq->list, &inactive_requests);
83         spin_unlock_irq(&list_lock);
84
85         return 0;
86 }
87
88 static int scm_alloc_rqs(unsigned int nrqs)
89 {
90         int ret = 0;
91
92         while (nrqs-- && !ret)
93                 ret = __scm_alloc_rq();
94
95         return ret;
96 }
97
98 static struct scm_request *scm_request_fetch(void)
99 {
100         struct scm_request *scmrq = NULL;
101
102         spin_lock(&list_lock);
103         if (list_empty(&inactive_requests))
104                 goto out;
105         scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
106         list_del(&scmrq->list);
107 out:
108         spin_unlock(&list_lock);
109         return scmrq;
110 }
111
112 static void scm_request_done(struct scm_request *scmrq)
113 {
114         unsigned long flags;
115
116         spin_lock_irqsave(&list_lock, flags);
117         list_add(&scmrq->list, &inactive_requests);
118         spin_unlock_irqrestore(&list_lock, flags);
119 }
120
121 static int scm_open(struct block_device *blkdev, fmode_t mode)
122 {
123         return scm_get_ref();
124 }
125
126 static int scm_release(struct gendisk *gendisk, fmode_t mode)
127 {
128         scm_put_ref();
129         return 0;
130 }
131
132 static const struct block_device_operations scm_blk_devops = {
133         .owner = THIS_MODULE,
134         .open = scm_open,
135         .release = scm_release,
136 };
137
138 static bool scm_permit_request(struct scm_blk_dev *bdev, struct request *req)
139 {
140         return rq_data_dir(req) != WRITE || bdev->state != SCM_WR_PROHIBIT;
141 }
142
143 static void scm_request_prepare(struct scm_request *scmrq)
144 {
145         struct scm_blk_dev *bdev = scmrq->bdev;
146         struct scm_device *scmdev = bdev->gendisk->private_data;
147         struct aidaw *aidaw = scmrq->aidaw;
148         struct msb *msb = &scmrq->aob->msb[0];
149         struct req_iterator iter;
150         struct bio_vec *bv;
151
152         msb->bs = MSB_BS_4K;
153         scmrq->aob->request.msb_count = 1;
154         msb->scm_addr = scmdev->address +
155                 ((u64) blk_rq_pos(scmrq->request) << 9);
156         msb->oc = (rq_data_dir(scmrq->request) == READ) ?
157                 MSB_OC_READ : MSB_OC_WRITE;
158         msb->flags |= MSB_FLAG_IDA;
159         msb->data_addr = (u64) aidaw;
160
161         rq_for_each_segment(bv, scmrq->request, iter) {
162                 WARN_ON(bv->bv_offset);
163                 msb->blk_count += bv->bv_len >> 12;
164                 aidaw->data_addr = (u64) page_address(bv->bv_page);
165                 aidaw++;
166         }
167 }
168
169 static inline void scm_request_init(struct scm_blk_dev *bdev,
170                                     struct scm_request *scmrq,
171                                     struct request *req)
172 {
173         struct aob_rq_header *aobrq = to_aobrq(scmrq);
174         struct aob *aob = scmrq->aob;
175
176         memset(aob, 0, sizeof(*aob));
177         memset(scmrq->aidaw, 0, PAGE_SIZE);
178         aobrq->scmdev = bdev->scmdev;
179         aob->request.cmd_code = ARQB_CMD_MOVE;
180         aob->request.data = (u64) aobrq;
181         scmrq->request = req;
182         scmrq->bdev = bdev;
183         scmrq->retries = 4;
184         scmrq->error = 0;
185         scm_request_cluster_init(scmrq);
186 }
187
188 static void scm_ensure_queue_restart(struct scm_blk_dev *bdev)
189 {
190         if (atomic_read(&bdev->queued_reqs)) {
191                 /* Queue restart is triggered by the next interrupt. */
192                 return;
193         }
194         blk_delay_queue(bdev->rq, SCM_QUEUE_DELAY);
195 }
196
197 void scm_request_requeue(struct scm_request *scmrq)
198 {
199         struct scm_blk_dev *bdev = scmrq->bdev;
200
201         scm_release_cluster(scmrq);
202         blk_requeue_request(bdev->rq, scmrq->request);
203         atomic_dec(&bdev->queued_reqs);
204         scm_request_done(scmrq);
205         scm_ensure_queue_restart(bdev);
206 }
207
208 void scm_request_finish(struct scm_request *scmrq)
209 {
210         struct scm_blk_dev *bdev = scmrq->bdev;
211
212         scm_release_cluster(scmrq);
213         blk_end_request_all(scmrq->request, scmrq->error);
214         atomic_dec(&bdev->queued_reqs);
215         scm_request_done(scmrq);
216 }
217
218 static void scm_blk_request(struct request_queue *rq)
219 {
220         struct scm_device *scmdev = rq->queuedata;
221         struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
222         struct scm_request *scmrq;
223         struct request *req;
224         int ret;
225
226         while ((req = blk_peek_request(rq))) {
227                 if (req->cmd_type != REQ_TYPE_FS)
228                         continue;
229
230                 if (!scm_permit_request(bdev, req)) {
231                         scm_ensure_queue_restart(bdev);
232                         return;
233                 }
234                 scmrq = scm_request_fetch();
235                 if (!scmrq) {
236                         SCM_LOG(5, "no request");
237                         scm_ensure_queue_restart(bdev);
238                         return;
239                 }
240                 scm_request_init(bdev, scmrq, req);
241                 if (!scm_reserve_cluster(scmrq)) {
242                         SCM_LOG(5, "cluster busy");
243                         scm_request_done(scmrq);
244                         return;
245                 }
246                 if (scm_need_cluster_request(scmrq)) {
247                         atomic_inc(&bdev->queued_reqs);
248                         blk_start_request(req);
249                         scm_initiate_cluster_request(scmrq);
250                         return;
251                 }
252                 scm_request_prepare(scmrq);
253                 atomic_inc(&bdev->queued_reqs);
254                 blk_start_request(req);
255
256                 ret = scm_start_aob(scmrq->aob);
257                 if (ret) {
258                         SCM_LOG(5, "no subchannel");
259                         scm_request_requeue(scmrq);
260                         return;
261                 }
262         }
263 }
264
265 static void __scmrq_log_error(struct scm_request *scmrq)
266 {
267         struct aob *aob = scmrq->aob;
268
269         if (scmrq->error == -ETIMEDOUT)
270                 SCM_LOG(1, "Request timeout");
271         else {
272                 SCM_LOG(1, "Request error");
273                 SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
274         }
275         if (scmrq->retries)
276                 SCM_LOG(1, "Retry request");
277         else
278                 pr_err("An I/O operation to SCM failed with rc=%d\n",
279                        scmrq->error);
280 }
281
282 void scm_blk_irq(struct scm_device *scmdev, void *data, int error)
283 {
284         struct scm_request *scmrq = data;
285         struct scm_blk_dev *bdev = scmrq->bdev;
286
287         scmrq->error = error;
288         if (error)
289                 __scmrq_log_error(scmrq);
290
291         spin_lock(&bdev->lock);
292         list_add_tail(&scmrq->list, &bdev->finished_requests);
293         spin_unlock(&bdev->lock);
294         tasklet_hi_schedule(&bdev->tasklet);
295 }
296
297 static void scm_blk_handle_error(struct scm_request *scmrq)
298 {
299         struct scm_blk_dev *bdev = scmrq->bdev;
300         unsigned long flags;
301
302         if (scmrq->error != -EIO)
303                 goto restart;
304
305         /* For -EIO the response block is valid. */
306         switch (scmrq->aob->response.eqc) {
307         case EQC_WR_PROHIBIT:
308                 spin_lock_irqsave(&bdev->lock, flags);
309                 if (bdev->state != SCM_WR_PROHIBIT)
310                         pr_info("%lx: Write access to the SCM increment is suspended\n",
311                                 (unsigned long) bdev->scmdev->address);
312                 bdev->state = SCM_WR_PROHIBIT;
313                 spin_unlock_irqrestore(&bdev->lock, flags);
314                 goto requeue;
315         default:
316                 break;
317         }
318
319 restart:
320         if (!scm_start_aob(scmrq->aob))
321                 return;
322
323 requeue:
324         spin_lock_irqsave(&bdev->rq_lock, flags);
325         scm_request_requeue(scmrq);
326         spin_unlock_irqrestore(&bdev->rq_lock, flags);
327 }
328
329 static void scm_blk_tasklet(struct scm_blk_dev *bdev)
330 {
331         struct scm_request *scmrq;
332         unsigned long flags;
333
334         spin_lock_irqsave(&bdev->lock, flags);
335         while (!list_empty(&bdev->finished_requests)) {
336                 scmrq = list_first_entry(&bdev->finished_requests,
337                                          struct scm_request, list);
338                 list_del(&scmrq->list);
339                 spin_unlock_irqrestore(&bdev->lock, flags);
340
341                 if (scmrq->error && scmrq->retries-- > 0) {
342                         scm_blk_handle_error(scmrq);
343
344                         /* Request restarted or requeued, handle next. */
345                         spin_lock_irqsave(&bdev->lock, flags);
346                         continue;
347                 }
348
349                 if (scm_test_cluster_request(scmrq)) {
350                         scm_cluster_request_irq(scmrq);
351                         spin_lock_irqsave(&bdev->lock, flags);
352                         continue;
353                 }
354
355                 scm_request_finish(scmrq);
356                 spin_lock_irqsave(&bdev->lock, flags);
357         }
358         spin_unlock_irqrestore(&bdev->lock, flags);
359         /* Look out for more requests. */
360         blk_run_queue(bdev->rq);
361 }
362
363 int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
364 {
365         struct request_queue *rq;
366         int len, ret = -ENOMEM;
367         unsigned int devindex, nr_max_blk;
368
369         devindex = atomic_inc_return(&nr_devices) - 1;
370         /* scma..scmz + scmaa..scmzz */
371         if (devindex > 701) {
372                 ret = -ENODEV;
373                 goto out;
374         }
375
376         bdev->scmdev = scmdev;
377         bdev->state = SCM_OPER;
378         spin_lock_init(&bdev->rq_lock);
379         spin_lock_init(&bdev->lock);
380         INIT_LIST_HEAD(&bdev->finished_requests);
381         atomic_set(&bdev->queued_reqs, 0);
382         tasklet_init(&bdev->tasklet,
383                      (void (*)(unsigned long)) scm_blk_tasklet,
384                      (unsigned long) bdev);
385
386         rq = blk_init_queue(scm_blk_request, &bdev->rq_lock);
387         if (!rq)
388                 goto out;
389
390         bdev->rq = rq;
391         nr_max_blk = min(scmdev->nr_max_block,
392                          (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));
393
394         blk_queue_logical_block_size(rq, 1 << 12);
395         blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */
396         blk_queue_max_segments(rq, nr_max_blk);
397         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq);
398         scm_blk_dev_cluster_setup(bdev);
399
400         bdev->gendisk = alloc_disk(SCM_NR_PARTS);
401         if (!bdev->gendisk)
402                 goto out_queue;
403
404         rq->queuedata = scmdev;
405         bdev->gendisk->driverfs_dev = &scmdev->dev;
406         bdev->gendisk->private_data = scmdev;
407         bdev->gendisk->fops = &scm_blk_devops;
408         bdev->gendisk->queue = rq;
409         bdev->gendisk->major = scm_major;
410         bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;
411
412         len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
413         if (devindex > 25) {
414                 len += snprintf(bdev->gendisk->disk_name + len,
415                                 DISK_NAME_LEN - len, "%c",
416                                 'a' + (devindex / 26) - 1);
417                 devindex = devindex % 26;
418         }
419         snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
420                  'a' + devindex);
421
422         /* 512 byte sectors */
423         set_capacity(bdev->gendisk, scmdev->size >> 9);
424         add_disk(bdev->gendisk);
425         return 0;
426
427 out_queue:
428         blk_cleanup_queue(rq);
429 out:
430         atomic_dec(&nr_devices);
431         return ret;
432 }
433
434 void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
435 {
436         tasklet_kill(&bdev->tasklet);
437         del_gendisk(bdev->gendisk);
438         blk_cleanup_queue(bdev->gendisk->queue);
439         put_disk(bdev->gendisk);
440 }
441
442 void scm_blk_set_available(struct scm_blk_dev *bdev)
443 {
444         unsigned long flags;
445
446         spin_lock_irqsave(&bdev->lock, flags);
447         if (bdev->state == SCM_WR_PROHIBIT)
448                 pr_info("%lx: Write access to the SCM increment is restored\n",
449                         (unsigned long) bdev->scmdev->address);
450         bdev->state = SCM_OPER;
451         spin_unlock_irqrestore(&bdev->lock, flags);
452 }
453
454 static int __init scm_blk_init(void)
455 {
456         int ret = -EINVAL;
457
458         if (!scm_cluster_size_valid())
459                 goto out;
460
461         ret = register_blkdev(0, "scm");
462         if (ret < 0)
463                 goto out;
464
465         scm_major = ret;
466         ret = scm_alloc_rqs(nr_requests);
467         if (ret)
468                 goto out_free;
469
470         scm_debug = debug_register("scm_log", 16, 1, 16);
471         if (!scm_debug) {
472                 ret = -ENOMEM;
473                 goto out_free;
474         }
475
476         debug_register_view(scm_debug, &debug_hex_ascii_view);
477         debug_set_level(scm_debug, 2);
478
479         ret = scm_drv_init();
480         if (ret)
481                 goto out_dbf;
482
483         return ret;
484
485 out_dbf:
486         debug_unregister(scm_debug);
487 out_free:
488         scm_free_rqs();
489         unregister_blkdev(scm_major, "scm");
490 out:
491         return ret;
492 }
493 module_init(scm_blk_init);
494
495 static void __exit scm_blk_cleanup(void)
496 {
497         scm_drv_cleanup();
498         debug_unregister(scm_debug);
499         scm_free_rqs();
500         unregister_blkdev(scm_major, "scm");
501 }
502 module_exit(scm_blk_cleanup);