]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/block/null_blk.c
Merge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm
[mirror_ubuntu-bionic-kernel.git] / drivers / block / null_blk.c
CommitLineData
f2298c04 1#include <linux/module.h>
fc1bc354 2
f2298c04
JA
3#include <linux/moduleparam.h>
4#include <linux/sched.h>
5#include <linux/fs.h>
6#include <linux/blkdev.h>
7#include <linux/init.h>
8#include <linux/slab.h>
9#include <linux/blk-mq.h>
10#include <linux/hrtimer.h>
b2b7e001 11#include <linux/lightnvm.h>
f2298c04
JA
12
13struct nullb_cmd {
14 struct list_head list;
15 struct llist_node ll_list;
16 struct call_single_data csd;
17 struct request *rq;
18 struct bio *bio;
19 unsigned int tag;
20 struct nullb_queue *nq;
3c395a96 21 struct hrtimer timer;
f2298c04
JA
22};
23
24struct nullb_queue {
25 unsigned long *tag_map;
26 wait_queue_head_t wait;
27 unsigned int queue_depth;
28
29 struct nullb_cmd *cmds;
30};
31
32struct nullb {
33 struct list_head list;
34 unsigned int index;
35 struct request_queue *q;
36 struct gendisk *disk;
b0b4e09c 37 struct nvm_dev *ndev;
24d2f903 38 struct blk_mq_tag_set tag_set;
f2298c04
JA
39 struct hrtimer timer;
40 unsigned int queue_depth;
41 spinlock_t lock;
42
43 struct nullb_queue *queues;
44 unsigned int nr_queues;
b2b7e001 45 char disk_name[DISK_NAME_LEN];
f2298c04
JA
46};
47
48static LIST_HEAD(nullb_list);
49static struct mutex lock;
50static int null_major;
51static int nullb_indexes;
6bb9535b 52static struct kmem_cache *ppa_cache;
f2298c04 53
f2298c04
JA
54enum {
55 NULL_IRQ_NONE = 0,
56 NULL_IRQ_SOFTIRQ = 1,
57 NULL_IRQ_TIMER = 2,
ce2c350b 58};
f2298c04 59
ce2c350b 60enum {
f2298c04
JA
61 NULL_Q_BIO = 0,
62 NULL_Q_RQ = 1,
63 NULL_Q_MQ = 2,
64};
65
2d263a78 66static int submit_queues;
f2298c04
JA
67module_param(submit_queues, int, S_IRUGO);
68MODULE_PARM_DESC(submit_queues, "Number of submission queues");
69
70static int home_node = NUMA_NO_NODE;
71module_param(home_node, int, S_IRUGO);
72MODULE_PARM_DESC(home_node, "Home node for the device");
73
74static int queue_mode = NULL_Q_MQ;
709c8667
MB
75
76static int null_param_store_val(const char *str, int *val, int min, int max)
77{
78 int ret, new_val;
79
80 ret = kstrtoint(str, 10, &new_val);
81 if (ret)
82 return -EINVAL;
83
84 if (new_val < min || new_val > max)
85 return -EINVAL;
86
87 *val = new_val;
88 return 0;
89}
90
91static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
92{
93 return null_param_store_val(str, &queue_mode, NULL_Q_BIO, NULL_Q_MQ);
94}
95
9c27847d 96static const struct kernel_param_ops null_queue_mode_param_ops = {
709c8667
MB
97 .set = null_set_queue_mode,
98 .get = param_get_int,
99};
100
101device_param_cb(queue_mode, &null_queue_mode_param_ops, &queue_mode, S_IRUGO);
54ae81cd 102MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
f2298c04
JA
103
104static int gb = 250;
105module_param(gb, int, S_IRUGO);
106MODULE_PARM_DESC(gb, "Size in GB");
107
108static int bs = 512;
109module_param(bs, int, S_IRUGO);
110MODULE_PARM_DESC(bs, "Block size (in bytes)");
111
112static int nr_devices = 2;
113module_param(nr_devices, int, S_IRUGO);
114MODULE_PARM_DESC(nr_devices, "Number of devices to register");
115
b2b7e001
MB
116static bool use_lightnvm;
117module_param(use_lightnvm, bool, S_IRUGO);
118MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
119
f2298c04 120static int irqmode = NULL_IRQ_SOFTIRQ;
709c8667
MB
121
122static int null_set_irqmode(const char *str, const struct kernel_param *kp)
123{
124 return null_param_store_val(str, &irqmode, NULL_IRQ_NONE,
125 NULL_IRQ_TIMER);
126}
127
9c27847d 128static const struct kernel_param_ops null_irqmode_param_ops = {
709c8667
MB
129 .set = null_set_irqmode,
130 .get = param_get_int,
131};
132
133device_param_cb(irqmode, &null_irqmode_param_ops, &irqmode, S_IRUGO);
f2298c04
JA
134MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
135
dbac1175
AA
136static unsigned long completion_nsec = 10000;
137module_param(completion_nsec, ulong, S_IRUGO);
f2298c04
JA
138MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
139
140static int hw_queue_depth = 64;
141module_param(hw_queue_depth, int, S_IRUGO);
142MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
143
20005244 144static bool use_per_node_hctx = false;
f2298c04 145module_param(use_per_node_hctx, bool, S_IRUGO);
20005244 146MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
f2298c04
JA
147
148static void put_tag(struct nullb_queue *nq, unsigned int tag)
149{
150 clear_bit_unlock(tag, nq->tag_map);
151
152 if (waitqueue_active(&nq->wait))
153 wake_up(&nq->wait);
154}
155
156static unsigned int get_tag(struct nullb_queue *nq)
157{
158 unsigned int tag;
159
160 do {
161 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
162 if (tag >= nq->queue_depth)
163 return -1U;
164 } while (test_and_set_bit_lock(tag, nq->tag_map));
165
166 return tag;
167}
168
169static void free_cmd(struct nullb_cmd *cmd)
170{
171 put_tag(cmd->nq, cmd->tag);
172}
173
3c395a96
PV
174static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
175
f2298c04
JA
176static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
177{
178 struct nullb_cmd *cmd;
179 unsigned int tag;
180
181 tag = get_tag(nq);
182 if (tag != -1U) {
183 cmd = &nq->cmds[tag];
184 cmd->tag = tag;
185 cmd->nq = nq;
3c395a96
PV
186 if (irqmode == NULL_IRQ_TIMER) {
187 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
188 HRTIMER_MODE_REL);
189 cmd->timer.function = null_cmd_timer_expired;
190 }
f2298c04
JA
191 return cmd;
192 }
193
194 return NULL;
195}
196
197static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
198{
199 struct nullb_cmd *cmd;
200 DEFINE_WAIT(wait);
201
202 cmd = __alloc_cmd(nq);
203 if (cmd || !can_wait)
204 return cmd;
205
206 do {
207 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
208 cmd = __alloc_cmd(nq);
209 if (cmd)
210 break;
211
212 io_schedule();
213 } while (1);
214
215 finish_wait(&nq->wait, &wait);
216 return cmd;
217}
218
219static void end_cmd(struct nullb_cmd *cmd)
220{
cf8ecc5a
AA
221 struct request_queue *q = NULL;
222
e8271201
MK
223 if (cmd->rq)
224 q = cmd->rq->q;
225
ce2c350b
CH
226 switch (queue_mode) {
227 case NULL_Q_MQ:
c8a446ad 228 blk_mq_end_request(cmd->rq, 0);
ce2c350b
CH
229 return;
230 case NULL_Q_RQ:
231 INIT_LIST_HEAD(&cmd->rq->queuelist);
232 blk_end_request_all(cmd->rq, 0);
233 break;
234 case NULL_Q_BIO:
4246a0b6 235 bio_endio(cmd->bio);
48cc661e 236 break;
ce2c350b 237 }
f2298c04 238
48cc661e
JA
239 free_cmd(cmd);
240
cf8ecc5a 241 /* Restart queue if needed, as we are freeing a tag */
48cc661e 242 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
cf8ecc5a
AA
243 unsigned long flags;
244
245 spin_lock_irqsave(q->queue_lock, flags);
48cc661e 246 blk_start_queue_async(q);
cf8ecc5a 247 spin_unlock_irqrestore(q->queue_lock, flags);
f2298c04 248 }
cf8ecc5a
AA
249}
250
251static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
252{
253 end_cmd(container_of(timer, struct nullb_cmd, timer));
f2298c04
JA
254
255 return HRTIMER_NORESTART;
256}
257
258static void null_cmd_end_timer(struct nullb_cmd *cmd)
259{
8b0e1953 260 ktime_t kt = completion_nsec;
f2298c04 261
3c395a96 262 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
f2298c04
JA
263}
264
265static void null_softirq_done_fn(struct request *rq)
266{
d891fa70
JA
267 if (queue_mode == NULL_Q_MQ)
268 end_cmd(blk_mq_rq_to_pdu(rq));
269 else
270 end_cmd(rq->special);
f2298c04
JA
271}
272
f2298c04
JA
273static inline void null_handle_cmd(struct nullb_cmd *cmd)
274{
275 /* Complete IO by inline, softirq or timer */
276 switch (irqmode) {
f2298c04 277 case NULL_IRQ_SOFTIRQ:
ce2c350b
CH
278 switch (queue_mode) {
279 case NULL_Q_MQ:
f4829a9b 280 blk_mq_complete_request(cmd->rq, cmd->rq->errors);
ce2c350b
CH
281 break;
282 case NULL_Q_RQ:
283 blk_complete_request(cmd->rq);
284 break;
285 case NULL_Q_BIO:
286 /*
287 * XXX: no proper submitting cpu information available.
288 */
289 end_cmd(cmd);
290 break;
291 }
292 break;
293 case NULL_IRQ_NONE:
f2298c04 294 end_cmd(cmd);
f2298c04
JA
295 break;
296 case NULL_IRQ_TIMER:
297 null_cmd_end_timer(cmd);
298 break;
299 }
300}
301
302static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
303{
304 int index = 0;
305
306 if (nullb->nr_queues != 1)
307 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
308
309 return &nullb->queues[index];
310}
311
dece1635 312static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
f2298c04
JA
313{
314 struct nullb *nullb = q->queuedata;
315 struct nullb_queue *nq = nullb_to_queue(nullb);
316 struct nullb_cmd *cmd;
317
318 cmd = alloc_cmd(nq, 1);
319 cmd->bio = bio;
320
321 null_handle_cmd(cmd);
dece1635 322 return BLK_QC_T_NONE;
f2298c04
JA
323}
324
325static int null_rq_prep_fn(struct request_queue *q, struct request *req)
326{
327 struct nullb *nullb = q->queuedata;
328 struct nullb_queue *nq = nullb_to_queue(nullb);
329 struct nullb_cmd *cmd;
330
331 cmd = alloc_cmd(nq, 0);
332 if (cmd) {
333 cmd->rq = req;
334 req->special = cmd;
335 return BLKPREP_OK;
336 }
8b70f45e 337 blk_stop_queue(q);
f2298c04
JA
338
339 return BLKPREP_DEFER;
340}
341
342static void null_request_fn(struct request_queue *q)
343{
344 struct request *rq;
345
346 while ((rq = blk_fetch_request(q)) != NULL) {
347 struct nullb_cmd *cmd = rq->special;
348
349 spin_unlock_irq(q->queue_lock);
350 null_handle_cmd(cmd);
351 spin_lock_irq(q->queue_lock);
352 }
353}
354
74c45052
JA
355static int null_queue_rq(struct blk_mq_hw_ctx *hctx,
356 const struct blk_mq_queue_data *bd)
f2298c04 357{
74c45052 358 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
f2298c04 359
3c395a96
PV
360 if (irqmode == NULL_IRQ_TIMER) {
361 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
362 cmd->timer.function = null_cmd_timer_expired;
363 }
74c45052 364 cmd->rq = bd->rq;
f2298c04
JA
365 cmd->nq = hctx->driver_data;
366
74c45052 367 blk_mq_start_request(bd->rq);
e2490073 368
f2298c04
JA
369 null_handle_cmd(cmd);
370 return BLK_MQ_RQ_QUEUE_OK;
371}
372
2d263a78
MB
373static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
374{
375 BUG_ON(!nullb);
376 BUG_ON(!nq);
377
378 init_waitqueue_head(&nq->wait);
379 nq->queue_depth = nullb->queue_depth;
380}
381
f2298c04
JA
382static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
383 unsigned int index)
384{
385 struct nullb *nullb = data;
386 struct nullb_queue *nq = &nullb->queues[index];
387
f2298c04 388 hctx->driver_data = nq;
2d263a78
MB
389 null_init_queue(nullb, nq);
390 nullb->nr_queues++;
f2298c04
JA
391
392 return 0;
393}
394
395static struct blk_mq_ops null_mq_ops = {
396 .queue_rq = null_queue_rq,
f2298c04 397 .init_hctx = null_init_hctx,
ce2c350b 398 .complete = null_softirq_done_fn,
f2298c04
JA
399};
400
de65d2d2
MB
401static void cleanup_queue(struct nullb_queue *nq)
402{
403 kfree(nq->tag_map);
404 kfree(nq->cmds);
405}
406
407static void cleanup_queues(struct nullb *nullb)
408{
409 int i;
410
411 for (i = 0; i < nullb->nr_queues; i++)
412 cleanup_queue(&nullb->queues[i]);
413
414 kfree(nullb->queues);
415}
416
b2b7e001
MB
417#ifdef CONFIG_NVM
418
419static void null_lnvm_end_io(struct request *rq, int error)
420{
421 struct nvm_rq *rqd = rq->end_io_data;
b2b7e001 422
06894efe
MB
423 rqd->error = error;
424 nvm_end_io(rqd);
b2b7e001
MB
425
426 blk_put_request(rq);
427}
428
16f26c3a 429static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
b2b7e001 430{
16f26c3a 431 struct request_queue *q = dev->q;
b2b7e001
MB
432 struct request *rq;
433 struct bio *bio = rqd->bio;
434
aebf526b
CH
435 rq = blk_mq_alloc_request(q,
436 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
b2b7e001
MB
437 if (IS_ERR(rq))
438 return -ENOMEM;
439
b2b7e001
MB
440 rq->__sector = bio->bi_iter.bi_sector;
441 rq->ioprio = bio_prio(bio);
442
443 if (bio_has_data(bio))
444 rq->nr_phys_segments = bio_phys_segments(q, bio);
445
446 rq->__data_len = bio->bi_iter.bi_size;
447 rq->bio = rq->biotail = bio;
448
449 rq->end_io_data = rqd;
450
451 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
452
453 return 0;
454}
455
16f26c3a 456static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
b2b7e001
MB
457{
458 sector_t size = gb * 1024 * 1024 * 1024ULL;
5b40db99 459 sector_t blksize;
b2b7e001
MB
460 struct nvm_id_group *grp;
461
462 id->ver_id = 0x1;
463 id->vmnt = 0;
bf643185 464 id->cap = 0x2;
b2b7e001 465 id->dom = 0x1;
5b40db99
MB
466
467 id->ppaf.blk_offset = 0;
468 id->ppaf.blk_len = 16;
469 id->ppaf.pg_offset = 16;
470 id->ppaf.pg_len = 16;
471 id->ppaf.sect_offset = 32;
472 id->ppaf.sect_len = 8;
473 id->ppaf.pln_offset = 40;
474 id->ppaf.pln_len = 8;
475 id->ppaf.lun_offset = 48;
476 id->ppaf.lun_len = 8;
477 id->ppaf.ch_offset = 56;
478 id->ppaf.ch_len = 8;
b2b7e001 479
e93d12ae
AB
480 sector_div(size, bs); /* convert size to pages */
481 size >>= 8; /* concert size to pgs pr blk */
19bd6fe7 482 grp = &id->grp;
b2b7e001 483 grp->mtype = 0;
5b40db99 484 grp->fmtype = 0;
b2b7e001 485 grp->num_ch = 1;
b2b7e001 486 grp->num_pg = 256;
5b40db99 487 blksize = size;
e93d12ae 488 size >>= 16;
5b40db99 489 grp->num_lun = size + 1;
e93d12ae 490 sector_div(blksize, grp->num_lun);
5b40db99
MB
491 grp->num_blk = blksize;
492 grp->num_pln = 1;
493
b2b7e001
MB
494 grp->fpg_sz = bs;
495 grp->csecs = bs;
496 grp->trdt = 25000;
497 grp->trdm = 25000;
498 grp->tprt = 500000;
499 grp->tprm = 500000;
500 grp->tbet = 1500000;
501 grp->tbem = 1500000;
502 grp->mpos = 0x010101; /* single plane rwe */
503 grp->cpar = hw_queue_depth;
504
505 return 0;
506}
507
16f26c3a 508static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
b2b7e001
MB
509{
510 mempool_t *virtmem_pool;
511
6bb9535b 512 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
b2b7e001
MB
513 if (!virtmem_pool) {
514 pr_err("null_blk: Unable to create virtual memory pool\n");
515 return NULL;
516 }
517
518 return virtmem_pool;
519}
520
521static void null_lnvm_destroy_dma_pool(void *pool)
522{
523 mempool_destroy(pool);
524}
525
16f26c3a 526static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
b2b7e001
MB
527 gfp_t mem_flags, dma_addr_t *dma_handler)
528{
529 return mempool_alloc(pool, mem_flags);
530}
531
532static void null_lnvm_dev_dma_free(void *pool, void *entry,
533 dma_addr_t dma_handler)
534{
535 mempool_free(entry, pool);
536}
537
538static struct nvm_dev_ops null_lnvm_dev_ops = {
539 .identity = null_lnvm_id,
540 .submit_io = null_lnvm_submit_io,
541
542 .create_dma_pool = null_lnvm_create_dma_pool,
543 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
544 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
545 .dev_dma_free = null_lnvm_dev_dma_free,
546
547 /* Simulate nvme protocol restriction */
548 .max_phys_sect = 64,
549};
9ae2d0aa
MB
550
551static int null_nvm_register(struct nullb *nullb)
552{
b0b4e09c
MB
553 struct nvm_dev *dev;
554 int rv;
555
556 dev = nvm_alloc_dev(0);
557 if (!dev)
558 return -ENOMEM;
559
560 dev->q = nullb->q;
561 memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
562 dev->ops = &null_lnvm_dev_ops;
563
564 rv = nvm_register(dev);
565 if (rv) {
566 kfree(dev);
567 return rv;
568 }
569 nullb->ndev = dev;
570 return 0;
9ae2d0aa
MB
571}
572
573static void null_nvm_unregister(struct nullb *nullb)
574{
b0b4e09c 575 nvm_unregister(nullb->ndev);
9ae2d0aa 576}
b2b7e001 577#else
9ae2d0aa
MB
578static int null_nvm_register(struct nullb *nullb)
579{
92153d30 580 pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
9ae2d0aa
MB
581 return -EINVAL;
582}
583static void null_nvm_unregister(struct nullb *nullb) {}
b2b7e001
MB
584#endif /* CONFIG_NVM */
585
9ae2d0aa
MB
586static void null_del_dev(struct nullb *nullb)
587{
588 list_del_init(&nullb->list);
589
590 if (use_lightnvm)
591 null_nvm_unregister(nullb);
592 else
593 del_gendisk(nullb->disk);
594 blk_cleanup_queue(nullb->q);
595 if (queue_mode == NULL_Q_MQ)
596 blk_mq_free_tag_set(&nullb->tag_set);
597 if (!use_lightnvm)
598 put_disk(nullb->disk);
599 cleanup_queues(nullb);
600 kfree(nullb);
601}
602
f2298c04
JA
603static int null_open(struct block_device *bdev, fmode_t mode)
604{
605 return 0;
606}
607
608static void null_release(struct gendisk *disk, fmode_t mode)
609{
610}
611
612static const struct block_device_operations null_fops = {
613 .owner = THIS_MODULE,
614 .open = null_open,
615 .release = null_release,
616};
617
618static int setup_commands(struct nullb_queue *nq)
619{
620 struct nullb_cmd *cmd;
621 int i, tag_size;
622
623 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
624 if (!nq->cmds)
2d263a78 625 return -ENOMEM;
f2298c04
JA
626
627 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
628 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
629 if (!nq->tag_map) {
630 kfree(nq->cmds);
2d263a78 631 return -ENOMEM;
f2298c04
JA
632 }
633
634 for (i = 0; i < nq->queue_depth; i++) {
635 cmd = &nq->cmds[i];
636 INIT_LIST_HEAD(&cmd->list);
637 cmd->ll_list.next = NULL;
638 cmd->tag = -1U;
639 }
640
641 return 0;
642}
643
f2298c04
JA
644static int setup_queues(struct nullb *nullb)
645{
2d263a78
MB
646 nullb->queues = kzalloc(submit_queues * sizeof(struct nullb_queue),
647 GFP_KERNEL);
f2298c04 648 if (!nullb->queues)
2d263a78 649 return -ENOMEM;
f2298c04
JA
650
651 nullb->nr_queues = 0;
652 nullb->queue_depth = hw_queue_depth;
653
2d263a78
MB
654 return 0;
655}
656
657static int init_driver_queues(struct nullb *nullb)
658{
659 struct nullb_queue *nq;
660 int i, ret = 0;
f2298c04
JA
661
662 for (i = 0; i < submit_queues; i++) {
663 nq = &nullb->queues[i];
2d263a78
MB
664
665 null_init_queue(nullb, nq);
666
667 ret = setup_commands(nq);
668 if (ret)
31f9690e 669 return ret;
f2298c04
JA
670 nullb->nr_queues++;
671 }
2d263a78 672 return 0;
f2298c04
JA
673}
674
9ae2d0aa 675static int null_gendisk_register(struct nullb *nullb)
f2298c04
JA
676{
677 struct gendisk *disk;
f2298c04 678 sector_t size;
9ae2d0aa
MB
679
680 disk = nullb->disk = alloc_disk_node(1, home_node);
681 if (!disk)
682 return -ENOMEM;
683 size = gb * 1024 * 1024 * 1024ULL;
684 set_capacity(disk, size >> 9);
685
686 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
687 disk->major = null_major;
688 disk->first_minor = nullb->index;
689 disk->fops = &null_fops;
690 disk->private_data = nullb;
691 disk->queue = nullb->q;
692 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
693
694 add_disk(disk);
695 return 0;
696}
697
698static int null_add_dev(void)
699{
700 struct nullb *nullb;
dc501dc0 701 int rv;
f2298c04
JA
702
703 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
dc501dc0
RE
704 if (!nullb) {
705 rv = -ENOMEM;
24d2f903 706 goto out;
dc501dc0 707 }
f2298c04
JA
708
709 spin_lock_init(&nullb->lock);
710
57053d8c
MB
711 if (queue_mode == NULL_Q_MQ && use_per_node_hctx)
712 submit_queues = nr_online_nodes;
713
dc501dc0
RE
714 rv = setup_queues(nullb);
715 if (rv)
24d2f903 716 goto out_free_nullb;
f2298c04
JA
717
718 if (queue_mode == NULL_Q_MQ) {
cdef54dd 719 nullb->tag_set.ops = &null_mq_ops;
24d2f903
CH
720 nullb->tag_set.nr_hw_queues = submit_queues;
721 nullb->tag_set.queue_depth = hw_queue_depth;
722 nullb->tag_set.numa_node = home_node;
723 nullb->tag_set.cmd_size = sizeof(struct nullb_cmd);
724 nullb->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
725 nullb->tag_set.driver_data = nullb;
726
dc501dc0
RE
727 rv = blk_mq_alloc_tag_set(&nullb->tag_set);
728 if (rv)
24d2f903
CH
729 goto out_cleanup_queues;
730
731 nullb->q = blk_mq_init_queue(&nullb->tag_set);
35b489d3 732 if (IS_ERR(nullb->q)) {
dc501dc0 733 rv = -ENOMEM;
24d2f903 734 goto out_cleanup_tags;
dc501dc0 735 }
f2298c04
JA
736 } else if (queue_mode == NULL_Q_BIO) {
737 nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
dc501dc0
RE
738 if (!nullb->q) {
739 rv = -ENOMEM;
24d2f903 740 goto out_cleanup_queues;
dc501dc0 741 }
f2298c04 742 blk_queue_make_request(nullb->q, null_queue_bio);
31f9690e
JK
743 rv = init_driver_queues(nullb);
744 if (rv)
745 goto out_cleanup_blk_queue;
f2298c04
JA
746 } else {
747 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
dc501dc0
RE
748 if (!nullb->q) {
749 rv = -ENOMEM;
24d2f903 750 goto out_cleanup_queues;
dc501dc0 751 }
f2298c04 752 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
24d2f903 753 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
31f9690e
JK
754 rv = init_driver_queues(nullb);
755 if (rv)
756 goto out_cleanup_blk_queue;
f2298c04
JA
757 }
758
f2298c04
JA
759 nullb->q->queuedata = nullb;
760 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
b277da0a 761 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
f2298c04 762
f2298c04 763 mutex_lock(&lock);
f2298c04
JA
764 nullb->index = nullb_indexes++;
765 mutex_unlock(&lock);
766
767 blk_queue_logical_block_size(nullb->q, bs);
768 blk_queue_physical_block_size(nullb->q, bs);
769
b2b7e001
MB
770 sprintf(nullb->disk_name, "nullb%d", nullb->index);
771
9ae2d0aa
MB
772 if (use_lightnvm)
773 rv = null_nvm_register(nullb);
774 else
775 rv = null_gendisk_register(nullb);
b2b7e001 776
9ae2d0aa
MB
777 if (rv)
778 goto out_cleanup_blk_queue;
a514379b
MB
779
780 mutex_lock(&lock);
781 list_add_tail(&nullb->list, &nullb_list);
782 mutex_unlock(&lock);
3681c85d 783
f2298c04 784 return 0;
24d2f903
CH
785out_cleanup_blk_queue:
786 blk_cleanup_queue(nullb->q);
787out_cleanup_tags:
788 if (queue_mode == NULL_Q_MQ)
789 blk_mq_free_tag_set(&nullb->tag_set);
790out_cleanup_queues:
791 cleanup_queues(nullb);
792out_free_nullb:
793 kfree(nullb);
794out:
dc501dc0 795 return rv;
f2298c04
JA
796}
797
798static int __init null_init(void)
799{
af096e22 800 int ret = 0;
f2298c04 801 unsigned int i;
af096e22 802 struct nullb *nullb;
f2298c04 803
9967d8ac
R
804 if (bs > PAGE_SIZE) {
805 pr_warn("null_blk: invalid block size\n");
806 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
807 bs = PAGE_SIZE;
808 }
f2298c04 809
6bb9535b
MB
810 if (use_lightnvm && bs != 4096) {
811 pr_warn("null_blk: LightNVM only supports 4k block size\n");
812 pr_warn("null_blk: defaults block size to 4k\n");
813 bs = 4096;
814 }
815
b2b7e001
MB
816 if (use_lightnvm && queue_mode != NULL_Q_MQ) {
817 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
818 pr_warn("null_blk: defaults queue mode to blk-mq\n");
819 queue_mode = NULL_Q_MQ;
820 }
821
d15ee6b1 822 if (queue_mode == NULL_Q_MQ && use_per_node_hctx) {
fc1bc354 823 if (submit_queues < nr_online_nodes) {
d15ee6b1
MB
824 pr_warn("null_blk: submit_queues param is set to %u.",
825 nr_online_nodes);
fc1bc354
MB
826 submit_queues = nr_online_nodes;
827 }
d15ee6b1 828 } else if (submit_queues > nr_cpu_ids)
f2298c04
JA
829 submit_queues = nr_cpu_ids;
830 else if (!submit_queues)
831 submit_queues = 1;
832
833 mutex_init(&lock);
834
f2298c04
JA
835 null_major = register_blkdev(0, "nullb");
836 if (null_major < 0)
837 return null_major;
838
6bb9535b
MB
839 if (use_lightnvm) {
840 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
841 0, 0, NULL);
842 if (!ppa_cache) {
843 pr_err("null_blk: unable to create ppa cache\n");
af096e22
MH
844 ret = -ENOMEM;
845 goto err_ppa;
6bb9535b
MB
846 }
847 }
848
f2298c04 849 for (i = 0; i < nr_devices; i++) {
af096e22
MH
850 ret = null_add_dev();
851 if (ret)
852 goto err_dev;
f2298c04
JA
853 }
854
855 pr_info("null: module loaded\n");
856 return 0;
af096e22
MH
857
858err_dev:
859 while (!list_empty(&nullb_list)) {
860 nullb = list_entry(nullb_list.next, struct nullb, list);
861 null_del_dev(nullb);
862 }
6bb9535b 863 kmem_cache_destroy(ppa_cache);
af096e22
MH
864err_ppa:
865 unregister_blkdev(null_major, "nullb");
866 return ret;
f2298c04
JA
867}
868
869static void __exit null_exit(void)
870{
871 struct nullb *nullb;
872
873 unregister_blkdev(null_major, "nullb");
874
875 mutex_lock(&lock);
876 while (!list_empty(&nullb_list)) {
877 nullb = list_entry(nullb_list.next, struct nullb, list);
878 null_del_dev(nullb);
879 }
880 mutex_unlock(&lock);
6bb9535b
MB
881
882 kmem_cache_destroy(ppa_cache);
f2298c04
JA
883}
884
885module_init(null_init);
886module_exit(null_exit);
887
888MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
889MODULE_LICENSE("GPL");