]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/block/null_blk.c
block: fix module reference leak on put_disk() call for cgroups throttle
[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;
24d2f903 37 struct blk_mq_tag_set tag_set;
f2298c04
JA
38 struct hrtimer timer;
39 unsigned int queue_depth;
40 spinlock_t lock;
41
42 struct nullb_queue *queues;
43 unsigned int nr_queues;
b2b7e001 44 char disk_name[DISK_NAME_LEN];
f2298c04
JA
45};
46
47static LIST_HEAD(nullb_list);
48static struct mutex lock;
49static int null_major;
50static int nullb_indexes;
6bb9535b 51static struct kmem_cache *ppa_cache;
f2298c04 52
f2298c04
JA
53enum {
54 NULL_IRQ_NONE = 0,
55 NULL_IRQ_SOFTIRQ = 1,
56 NULL_IRQ_TIMER = 2,
ce2c350b 57};
f2298c04 58
ce2c350b 59enum {
f2298c04
JA
60 NULL_Q_BIO = 0,
61 NULL_Q_RQ = 1,
62 NULL_Q_MQ = 2,
63};
64
2d263a78 65static int submit_queues;
f2298c04
JA
66module_param(submit_queues, int, S_IRUGO);
67MODULE_PARM_DESC(submit_queues, "Number of submission queues");
68
69static int home_node = NUMA_NO_NODE;
70module_param(home_node, int, S_IRUGO);
71MODULE_PARM_DESC(home_node, "Home node for the device");
72
73static int queue_mode = NULL_Q_MQ;
709c8667
MB
74
75static int null_param_store_val(const char *str, int *val, int min, int max)
76{
77 int ret, new_val;
78
79 ret = kstrtoint(str, 10, &new_val);
80 if (ret)
81 return -EINVAL;
82
83 if (new_val < min || new_val > max)
84 return -EINVAL;
85
86 *val = new_val;
87 return 0;
88}
89
90static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
91{
92 return null_param_store_val(str, &queue_mode, NULL_Q_BIO, NULL_Q_MQ);
93}
94
9c27847d 95static const struct kernel_param_ops null_queue_mode_param_ops = {
709c8667
MB
96 .set = null_set_queue_mode,
97 .get = param_get_int,
98};
99
100device_param_cb(queue_mode, &null_queue_mode_param_ops, &queue_mode, S_IRUGO);
54ae81cd 101MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
f2298c04
JA
102
103static int gb = 250;
104module_param(gb, int, S_IRUGO);
105MODULE_PARM_DESC(gb, "Size in GB");
106
107static int bs = 512;
108module_param(bs, int, S_IRUGO);
109MODULE_PARM_DESC(bs, "Block size (in bytes)");
110
111static int nr_devices = 2;
112module_param(nr_devices, int, S_IRUGO);
113MODULE_PARM_DESC(nr_devices, "Number of devices to register");
114
b2b7e001
MB
115static bool use_lightnvm;
116module_param(use_lightnvm, bool, S_IRUGO);
117MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
118
f2298c04 119static int irqmode = NULL_IRQ_SOFTIRQ;
709c8667
MB
120
121static int null_set_irqmode(const char *str, const struct kernel_param *kp)
122{
123 return null_param_store_val(str, &irqmode, NULL_IRQ_NONE,
124 NULL_IRQ_TIMER);
125}
126
9c27847d 127static const struct kernel_param_ops null_irqmode_param_ops = {
709c8667
MB
128 .set = null_set_irqmode,
129 .get = param_get_int,
130};
131
132device_param_cb(irqmode, &null_irqmode_param_ops, &irqmode, S_IRUGO);
f2298c04
JA
133MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
134
dbac1175
AA
135static unsigned long completion_nsec = 10000;
136module_param(completion_nsec, ulong, S_IRUGO);
f2298c04
JA
137MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
138
139static int hw_queue_depth = 64;
140module_param(hw_queue_depth, int, S_IRUGO);
141MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
142
20005244 143static bool use_per_node_hctx = false;
f2298c04 144module_param(use_per_node_hctx, bool, S_IRUGO);
20005244 145MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
f2298c04
JA
146
147static void put_tag(struct nullb_queue *nq, unsigned int tag)
148{
149 clear_bit_unlock(tag, nq->tag_map);
150
151 if (waitqueue_active(&nq->wait))
152 wake_up(&nq->wait);
153}
154
155static unsigned int get_tag(struct nullb_queue *nq)
156{
157 unsigned int tag;
158
159 do {
160 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
161 if (tag >= nq->queue_depth)
162 return -1U;
163 } while (test_and_set_bit_lock(tag, nq->tag_map));
164
165 return tag;
166}
167
168static void free_cmd(struct nullb_cmd *cmd)
169{
170 put_tag(cmd->nq, cmd->tag);
171}
172
3c395a96
PV
173static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
174
f2298c04
JA
175static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
176{
177 struct nullb_cmd *cmd;
178 unsigned int tag;
179
180 tag = get_tag(nq);
181 if (tag != -1U) {
182 cmd = &nq->cmds[tag];
183 cmd->tag = tag;
184 cmd->nq = nq;
3c395a96
PV
185 if (irqmode == NULL_IRQ_TIMER) {
186 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
187 HRTIMER_MODE_REL);
188 cmd->timer.function = null_cmd_timer_expired;
189 }
f2298c04
JA
190 return cmd;
191 }
192
193 return NULL;
194}
195
196static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
197{
198 struct nullb_cmd *cmd;
199 DEFINE_WAIT(wait);
200
201 cmd = __alloc_cmd(nq);
202 if (cmd || !can_wait)
203 return cmd;
204
205 do {
206 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
207 cmd = __alloc_cmd(nq);
208 if (cmd)
209 break;
210
211 io_schedule();
212 } while (1);
213
214 finish_wait(&nq->wait, &wait);
215 return cmd;
216}
217
218static void end_cmd(struct nullb_cmd *cmd)
219{
cf8ecc5a
AA
220 struct request_queue *q = NULL;
221
e8271201
MK
222 if (cmd->rq)
223 q = cmd->rq->q;
224
ce2c350b
CH
225 switch (queue_mode) {
226 case NULL_Q_MQ:
c8a446ad 227 blk_mq_end_request(cmd->rq, 0);
ce2c350b
CH
228 return;
229 case NULL_Q_RQ:
230 INIT_LIST_HEAD(&cmd->rq->queuelist);
231 blk_end_request_all(cmd->rq, 0);
232 break;
233 case NULL_Q_BIO:
4246a0b6 234 bio_endio(cmd->bio);
48cc661e 235 break;
ce2c350b 236 }
f2298c04 237
48cc661e
JA
238 free_cmd(cmd);
239
cf8ecc5a 240 /* Restart queue if needed, as we are freeing a tag */
48cc661e 241 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
cf8ecc5a
AA
242 unsigned long flags;
243
244 spin_lock_irqsave(q->queue_lock, flags);
48cc661e 245 blk_start_queue_async(q);
cf8ecc5a 246 spin_unlock_irqrestore(q->queue_lock, flags);
f2298c04 247 }
cf8ecc5a
AA
248}
249
250static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
251{
252 end_cmd(container_of(timer, struct nullb_cmd, timer));
f2298c04
JA
253
254 return HRTIMER_NORESTART;
255}
256
257static void null_cmd_end_timer(struct nullb_cmd *cmd)
258{
3c395a96 259 ktime_t kt = ktime_set(0, completion_nsec);
f2298c04 260
3c395a96 261 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
f2298c04
JA
262}
263
264static void null_softirq_done_fn(struct request *rq)
265{
d891fa70
JA
266 if (queue_mode == NULL_Q_MQ)
267 end_cmd(blk_mq_rq_to_pdu(rq));
268 else
269 end_cmd(rq->special);
f2298c04
JA
270}
271
f2298c04
JA
272static inline void null_handle_cmd(struct nullb_cmd *cmd)
273{
274 /* Complete IO by inline, softirq or timer */
275 switch (irqmode) {
f2298c04 276 case NULL_IRQ_SOFTIRQ:
ce2c350b
CH
277 switch (queue_mode) {
278 case NULL_Q_MQ:
f4829a9b 279 blk_mq_complete_request(cmd->rq, cmd->rq->errors);
ce2c350b
CH
280 break;
281 case NULL_Q_RQ:
282 blk_complete_request(cmd->rq);
283 break;
284 case NULL_Q_BIO:
285 /*
286 * XXX: no proper submitting cpu information available.
287 */
288 end_cmd(cmd);
289 break;
290 }
291 break;
292 case NULL_IRQ_NONE:
f2298c04 293 end_cmd(cmd);
f2298c04
JA
294 break;
295 case NULL_IRQ_TIMER:
296 null_cmd_end_timer(cmd);
297 break;
298 }
299}
300
301static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
302{
303 int index = 0;
304
305 if (nullb->nr_queues != 1)
306 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
307
308 return &nullb->queues[index];
309}
310
dece1635 311static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
f2298c04
JA
312{
313 struct nullb *nullb = q->queuedata;
314 struct nullb_queue *nq = nullb_to_queue(nullb);
315 struct nullb_cmd *cmd;
316
317 cmd = alloc_cmd(nq, 1);
318 cmd->bio = bio;
319
320 null_handle_cmd(cmd);
dece1635 321 return BLK_QC_T_NONE;
f2298c04
JA
322}
323
324static int null_rq_prep_fn(struct request_queue *q, struct request *req)
325{
326 struct nullb *nullb = q->queuedata;
327 struct nullb_queue *nq = nullb_to_queue(nullb);
328 struct nullb_cmd *cmd;
329
330 cmd = alloc_cmd(nq, 0);
331 if (cmd) {
332 cmd->rq = req;
333 req->special = cmd;
334 return BLKPREP_OK;
335 }
8b70f45e 336 blk_stop_queue(q);
f2298c04
JA
337
338 return BLKPREP_DEFER;
339}
340
341static void null_request_fn(struct request_queue *q)
342{
343 struct request *rq;
344
345 while ((rq = blk_fetch_request(q)) != NULL) {
346 struct nullb_cmd *cmd = rq->special;
347
348 spin_unlock_irq(q->queue_lock);
349 null_handle_cmd(cmd);
350 spin_lock_irq(q->queue_lock);
351 }
352}
353
74c45052
JA
354static int null_queue_rq(struct blk_mq_hw_ctx *hctx,
355 const struct blk_mq_queue_data *bd)
f2298c04 356{
74c45052 357 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
f2298c04 358
3c395a96
PV
359 if (irqmode == NULL_IRQ_TIMER) {
360 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
361 cmd->timer.function = null_cmd_timer_expired;
362 }
74c45052 363 cmd->rq = bd->rq;
f2298c04
JA
364 cmd->nq = hctx->driver_data;
365
74c45052 366 blk_mq_start_request(bd->rq);
e2490073 367
f2298c04
JA
368 null_handle_cmd(cmd);
369 return BLK_MQ_RQ_QUEUE_OK;
370}
371
2d263a78
MB
372static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
373{
374 BUG_ON(!nullb);
375 BUG_ON(!nq);
376
377 init_waitqueue_head(&nq->wait);
378 nq->queue_depth = nullb->queue_depth;
379}
380
f2298c04
JA
381static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
382 unsigned int index)
383{
384 struct nullb *nullb = data;
385 struct nullb_queue *nq = &nullb->queues[index];
386
f2298c04 387 hctx->driver_data = nq;
2d263a78
MB
388 null_init_queue(nullb, nq);
389 nullb->nr_queues++;
f2298c04
JA
390
391 return 0;
392}
393
394static struct blk_mq_ops null_mq_ops = {
395 .queue_rq = null_queue_rq,
396 .map_queue = blk_mq_map_queue,
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
f2298c04
JA
417static void null_del_dev(struct nullb *nullb)
418{
419 list_del_init(&nullb->list);
420
b2b7e001 421 if (use_lightnvm)
54514aa4
MB
422 nvm_unregister(nullb->disk_name);
423 else
424 del_gendisk(nullb->disk);
518d00b7 425 blk_cleanup_queue(nullb->q);
24d2f903
CH
426 if (queue_mode == NULL_Q_MQ)
427 blk_mq_free_tag_set(&nullb->tag_set);
54514aa4
MB
428 if (!use_lightnvm)
429 put_disk(nullb->disk);
de65d2d2 430 cleanup_queues(nullb);
f2298c04
JA
431 kfree(nullb);
432}
433
b2b7e001
MB
434#ifdef CONFIG_NVM
435
436static void null_lnvm_end_io(struct request *rq, int error)
437{
438 struct nvm_rq *rqd = rq->end_io_data;
b2b7e001 439
91276162 440 nvm_end_io(rqd, error);
b2b7e001
MB
441
442 blk_put_request(rq);
443}
444
16f26c3a 445static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
b2b7e001 446{
16f26c3a 447 struct request_queue *q = dev->q;
b2b7e001
MB
448 struct request *rq;
449 struct bio *bio = rqd->bio;
450
6f3b0e8b 451 rq = blk_mq_alloc_request(q, bio_rw(bio), 0);
b2b7e001
MB
452 if (IS_ERR(rq))
453 return -ENOMEM;
454
455 rq->cmd_type = REQ_TYPE_DRV_PRIV;
456 rq->__sector = bio->bi_iter.bi_sector;
457 rq->ioprio = bio_prio(bio);
458
459 if (bio_has_data(bio))
460 rq->nr_phys_segments = bio_phys_segments(q, bio);
461
462 rq->__data_len = bio->bi_iter.bi_size;
463 rq->bio = rq->biotail = bio;
464
465 rq->end_io_data = rqd;
466
467 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
468
469 return 0;
470}
471
16f26c3a 472static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
b2b7e001
MB
473{
474 sector_t size = gb * 1024 * 1024 * 1024ULL;
5b40db99 475 sector_t blksize;
b2b7e001
MB
476 struct nvm_id_group *grp;
477
478 id->ver_id = 0x1;
479 id->vmnt = 0;
480 id->cgrps = 1;
bf643185 481 id->cap = 0x2;
b2b7e001 482 id->dom = 0x1;
5b40db99
MB
483
484 id->ppaf.blk_offset = 0;
485 id->ppaf.blk_len = 16;
486 id->ppaf.pg_offset = 16;
487 id->ppaf.pg_len = 16;
488 id->ppaf.sect_offset = 32;
489 id->ppaf.sect_len = 8;
490 id->ppaf.pln_offset = 40;
491 id->ppaf.pln_len = 8;
492 id->ppaf.lun_offset = 48;
493 id->ppaf.lun_len = 8;
494 id->ppaf.ch_offset = 56;
495 id->ppaf.ch_len = 8;
b2b7e001 496
e93d12ae
AB
497 sector_div(size, bs); /* convert size to pages */
498 size >>= 8; /* concert size to pgs pr blk */
b2b7e001
MB
499 grp = &id->groups[0];
500 grp->mtype = 0;
5b40db99 501 grp->fmtype = 0;
b2b7e001 502 grp->num_ch = 1;
b2b7e001 503 grp->num_pg = 256;
5b40db99 504 blksize = size;
e93d12ae 505 size >>= 16;
5b40db99 506 grp->num_lun = size + 1;
e93d12ae 507 sector_div(blksize, grp->num_lun);
5b40db99
MB
508 grp->num_blk = blksize;
509 grp->num_pln = 1;
510
b2b7e001
MB
511 grp->fpg_sz = bs;
512 grp->csecs = bs;
513 grp->trdt = 25000;
514 grp->trdm = 25000;
515 grp->tprt = 500000;
516 grp->tprm = 500000;
517 grp->tbet = 1500000;
518 grp->tbem = 1500000;
519 grp->mpos = 0x010101; /* single plane rwe */
520 grp->cpar = hw_queue_depth;
521
522 return 0;
523}
524
16f26c3a 525static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
b2b7e001
MB
526{
527 mempool_t *virtmem_pool;
528
6bb9535b 529 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
b2b7e001
MB
530 if (!virtmem_pool) {
531 pr_err("null_blk: Unable to create virtual memory pool\n");
532 return NULL;
533 }
534
535 return virtmem_pool;
536}
537
538static void null_lnvm_destroy_dma_pool(void *pool)
539{
540 mempool_destroy(pool);
541}
542
16f26c3a 543static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
b2b7e001
MB
544 gfp_t mem_flags, dma_addr_t *dma_handler)
545{
546 return mempool_alloc(pool, mem_flags);
547}
548
549static void null_lnvm_dev_dma_free(void *pool, void *entry,
550 dma_addr_t dma_handler)
551{
552 mempool_free(entry, pool);
553}
554
555static struct nvm_dev_ops null_lnvm_dev_ops = {
556 .identity = null_lnvm_id,
557 .submit_io = null_lnvm_submit_io,
558
559 .create_dma_pool = null_lnvm_create_dma_pool,
560 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
561 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
562 .dev_dma_free = null_lnvm_dev_dma_free,
563
564 /* Simulate nvme protocol restriction */
565 .max_phys_sect = 64,
566};
567#else
568static struct nvm_dev_ops null_lnvm_dev_ops;
569#endif /* CONFIG_NVM */
570
f2298c04
JA
571static int null_open(struct block_device *bdev, fmode_t mode)
572{
573 return 0;
574}
575
576static void null_release(struct gendisk *disk, fmode_t mode)
577{
578}
579
580static const struct block_device_operations null_fops = {
581 .owner = THIS_MODULE,
582 .open = null_open,
583 .release = null_release,
584};
585
586static int setup_commands(struct nullb_queue *nq)
587{
588 struct nullb_cmd *cmd;
589 int i, tag_size;
590
591 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
592 if (!nq->cmds)
2d263a78 593 return -ENOMEM;
f2298c04
JA
594
595 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
596 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
597 if (!nq->tag_map) {
598 kfree(nq->cmds);
2d263a78 599 return -ENOMEM;
f2298c04
JA
600 }
601
602 for (i = 0; i < nq->queue_depth; i++) {
603 cmd = &nq->cmds[i];
604 INIT_LIST_HEAD(&cmd->list);
605 cmd->ll_list.next = NULL;
606 cmd->tag = -1U;
607 }
608
609 return 0;
610}
611
f2298c04
JA
612static int setup_queues(struct nullb *nullb)
613{
2d263a78
MB
614 nullb->queues = kzalloc(submit_queues * sizeof(struct nullb_queue),
615 GFP_KERNEL);
f2298c04 616 if (!nullb->queues)
2d263a78 617 return -ENOMEM;
f2298c04
JA
618
619 nullb->nr_queues = 0;
620 nullb->queue_depth = hw_queue_depth;
621
2d263a78
MB
622 return 0;
623}
624
625static int init_driver_queues(struct nullb *nullb)
626{
627 struct nullb_queue *nq;
628 int i, ret = 0;
f2298c04
JA
629
630 for (i = 0; i < submit_queues; i++) {
631 nq = &nullb->queues[i];
2d263a78
MB
632
633 null_init_queue(nullb, nq);
634
635 ret = setup_commands(nq);
636 if (ret)
31f9690e 637 return ret;
f2298c04
JA
638 nullb->nr_queues++;
639 }
2d263a78 640 return 0;
f2298c04
JA
641}
642
643static int null_add_dev(void)
644{
645 struct gendisk *disk;
646 struct nullb *nullb;
647 sector_t size;
dc501dc0 648 int rv;
f2298c04
JA
649
650 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
dc501dc0
RE
651 if (!nullb) {
652 rv = -ENOMEM;
24d2f903 653 goto out;
dc501dc0 654 }
f2298c04
JA
655
656 spin_lock_init(&nullb->lock);
657
57053d8c
MB
658 if (queue_mode == NULL_Q_MQ && use_per_node_hctx)
659 submit_queues = nr_online_nodes;
660
dc501dc0
RE
661 rv = setup_queues(nullb);
662 if (rv)
24d2f903 663 goto out_free_nullb;
f2298c04
JA
664
665 if (queue_mode == NULL_Q_MQ) {
cdef54dd 666 nullb->tag_set.ops = &null_mq_ops;
24d2f903
CH
667 nullb->tag_set.nr_hw_queues = submit_queues;
668 nullb->tag_set.queue_depth = hw_queue_depth;
669 nullb->tag_set.numa_node = home_node;
670 nullb->tag_set.cmd_size = sizeof(struct nullb_cmd);
671 nullb->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
672 nullb->tag_set.driver_data = nullb;
673
dc501dc0
RE
674 rv = blk_mq_alloc_tag_set(&nullb->tag_set);
675 if (rv)
24d2f903
CH
676 goto out_cleanup_queues;
677
678 nullb->q = blk_mq_init_queue(&nullb->tag_set);
35b489d3 679 if (IS_ERR(nullb->q)) {
dc501dc0 680 rv = -ENOMEM;
24d2f903 681 goto out_cleanup_tags;
dc501dc0 682 }
f2298c04
JA
683 } else if (queue_mode == NULL_Q_BIO) {
684 nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
dc501dc0
RE
685 if (!nullb->q) {
686 rv = -ENOMEM;
24d2f903 687 goto out_cleanup_queues;
dc501dc0 688 }
f2298c04 689 blk_queue_make_request(nullb->q, null_queue_bio);
31f9690e
JK
690 rv = init_driver_queues(nullb);
691 if (rv)
692 goto out_cleanup_blk_queue;
f2298c04
JA
693 } else {
694 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
dc501dc0
RE
695 if (!nullb->q) {
696 rv = -ENOMEM;
24d2f903 697 goto out_cleanup_queues;
dc501dc0 698 }
f2298c04 699 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
24d2f903 700 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
31f9690e
JK
701 rv = init_driver_queues(nullb);
702 if (rv)
703 goto out_cleanup_blk_queue;
f2298c04
JA
704 }
705
f2298c04
JA
706 nullb->q->queuedata = nullb;
707 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
b277da0a 708 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
f2298c04 709
f2298c04
JA
710
711 mutex_lock(&lock);
712 list_add_tail(&nullb->list, &nullb_list);
713 nullb->index = nullb_indexes++;
714 mutex_unlock(&lock);
715
716 blk_queue_logical_block_size(nullb->q, bs);
717 blk_queue_physical_block_size(nullb->q, bs);
718
b2b7e001
MB
719 sprintf(nullb->disk_name, "nullb%d", nullb->index);
720
721 if (use_lightnvm) {
722 rv = nvm_register(nullb->q, nullb->disk_name,
723 &null_lnvm_dev_ops);
724 if (rv)
725 goto out_cleanup_blk_queue;
726 goto done;
727 }
728
729 disk = nullb->disk = alloc_disk_node(1, home_node);
730 if (!disk) {
731 rv = -ENOMEM;
732 goto out_cleanup_lightnvm;
733 }
f2298c04 734 size = gb * 1024 * 1024 * 1024ULL;
5fdb7e1b 735 set_capacity(disk, size >> 9);
f2298c04 736
227290b4 737 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
f2298c04
JA
738 disk->major = null_major;
739 disk->first_minor = nullb->index;
740 disk->fops = &null_fops;
741 disk->private_data = nullb;
742 disk->queue = nullb->q;
b2b7e001
MB
743 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
744
f2298c04 745 add_disk(disk);
b2b7e001 746done:
f2298c04 747 return 0;
24d2f903 748
b2b7e001
MB
749out_cleanup_lightnvm:
750 if (use_lightnvm)
751 nvm_unregister(nullb->disk_name);
24d2f903
CH
752out_cleanup_blk_queue:
753 blk_cleanup_queue(nullb->q);
754out_cleanup_tags:
755 if (queue_mode == NULL_Q_MQ)
756 blk_mq_free_tag_set(&nullb->tag_set);
757out_cleanup_queues:
758 cleanup_queues(nullb);
759out_free_nullb:
760 kfree(nullb);
761out:
dc501dc0 762 return rv;
f2298c04
JA
763}
764
765static int __init null_init(void)
766{
af096e22 767 int ret = 0;
f2298c04 768 unsigned int i;
af096e22 769 struct nullb *nullb;
f2298c04 770
9967d8ac
R
771 if (bs > PAGE_SIZE) {
772 pr_warn("null_blk: invalid block size\n");
773 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
774 bs = PAGE_SIZE;
775 }
f2298c04 776
6bb9535b
MB
777 if (use_lightnvm && bs != 4096) {
778 pr_warn("null_blk: LightNVM only supports 4k block size\n");
779 pr_warn("null_blk: defaults block size to 4k\n");
780 bs = 4096;
781 }
782
b2b7e001
MB
783 if (use_lightnvm && queue_mode != NULL_Q_MQ) {
784 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
785 pr_warn("null_blk: defaults queue mode to blk-mq\n");
786 queue_mode = NULL_Q_MQ;
787 }
788
d15ee6b1 789 if (queue_mode == NULL_Q_MQ && use_per_node_hctx) {
fc1bc354 790 if (submit_queues < nr_online_nodes) {
d15ee6b1
MB
791 pr_warn("null_blk: submit_queues param is set to %u.",
792 nr_online_nodes);
fc1bc354
MB
793 submit_queues = nr_online_nodes;
794 }
d15ee6b1 795 } else if (submit_queues > nr_cpu_ids)
f2298c04
JA
796 submit_queues = nr_cpu_ids;
797 else if (!submit_queues)
798 submit_queues = 1;
799
800 mutex_init(&lock);
801
f2298c04
JA
802 null_major = register_blkdev(0, "nullb");
803 if (null_major < 0)
804 return null_major;
805
6bb9535b
MB
806 if (use_lightnvm) {
807 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
808 0, 0, NULL);
809 if (!ppa_cache) {
810 pr_err("null_blk: unable to create ppa cache\n");
af096e22
MH
811 ret = -ENOMEM;
812 goto err_ppa;
6bb9535b
MB
813 }
814 }
815
f2298c04 816 for (i = 0; i < nr_devices; i++) {
af096e22
MH
817 ret = null_add_dev();
818 if (ret)
819 goto err_dev;
f2298c04
JA
820 }
821
822 pr_info("null: module loaded\n");
823 return 0;
af096e22
MH
824
825err_dev:
826 while (!list_empty(&nullb_list)) {
827 nullb = list_entry(nullb_list.next, struct nullb, list);
828 null_del_dev(nullb);
829 }
6bb9535b 830 kmem_cache_destroy(ppa_cache);
af096e22
MH
831err_ppa:
832 unregister_blkdev(null_major, "nullb");
833 return ret;
f2298c04
JA
834}
835
836static void __exit null_exit(void)
837{
838 struct nullb *nullb;
839
840 unregister_blkdev(null_major, "nullb");
841
842 mutex_lock(&lock);
843 while (!list_empty(&nullb_list)) {
844 nullb = list_entry(nullb_list.next, struct nullb, list);
845 null_del_dev(nullb);
846 }
847 mutex_unlock(&lock);
6bb9535b
MB
848
849 kmem_cache_destroy(ppa_cache);
f2298c04
JA
850}
851
852module_init(null_init);
853module_exit(null_exit);
854
855MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
856MODULE_LICENSE("GPL");