]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blob - drivers/block/null_blk.c
projects / mirror_ubuntu-focal-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'asoc/topic/pcm512x' into asoc-next
[mirror_ubuntu-focal-kernel.git] / drivers / block / null_blk.c
1 /*
2 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
3 * Shaohua Li <shli@fb.com>
4 */
5 #include <linux/module.h>
6
7 #include <linux/moduleparam.h>
8 #include <linux/sched.h>
9 #include <linux/fs.h>
10 #include <linux/blkdev.h>
11 #include <linux/init.h>
12 #include <linux/slab.h>
13 #include <linux/blk-mq.h>
14 #include <linux/hrtimer.h>
15 #include <linux/lightnvm.h>
16 #include <linux/configfs.h>
17 #include <linux/badblocks.h>
18
19 #define SECTOR_SHIFT 9
20 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
21 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
22 #define SECTOR_SIZE (1 << SECTOR_SHIFT)
23 #define SECTOR_MASK (PAGE_SECTORS - 1)
24
25 #define FREE_BATCH 16
26
27 #define TICKS_PER_SEC 50ULL
28 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
29
30 static inline u64 mb_per_tick(int mbps)
31 {
32 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
33 }
34
35 struct nullb_cmd {
36 struct list_head list;
37 struct llist_node ll_list;
38 struct __call_single_data csd;
39 struct request *rq;
40 struct bio *bio;
41 unsigned int tag;
42 blk_status_t error;
43 struct nullb_queue *nq;
44 struct hrtimer timer;
45 };
46
47 struct nullb_queue {
48 unsigned long *tag_map;
49 wait_queue_head_t wait;
50 unsigned int queue_depth;
51 struct nullb_device *dev;
52
53 struct nullb_cmd *cmds;
54 };
55
56 /*
57 * Status flags for nullb_device.
58 *
59 * CONFIGURED: Device has been configured and turned on. Cannot reconfigure.
60 * UP: Device is currently on and visible in userspace.
61 * THROTTLED: Device is being throttled.
62 * CACHE: Device is using a write-back cache.
63 */
64 enum nullb_device_flags {
65 NULLB_DEV_FL_CONFIGURED = 0,
66 NULLB_DEV_FL_UP = 1,
67 NULLB_DEV_FL_THROTTLED = 2,
68 NULLB_DEV_FL_CACHE = 3,
69 };
70
71 /*
72 * nullb_page is a page in memory for nullb devices.
73 *
74 * @page: The page holding the data.
75 * @bitmap: The bitmap represents which sector in the page has data.
76 * Each bit represents one block size. For example, sector 8
77 * will use the 7th bit
78 * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
79 * page is being flushing to storage. FREE means the cache page is freed and
80 * should be skipped from flushing to storage. Please see
81 * null_make_cache_space
82 */
83 struct nullb_page {
84 struct page *page;
85 unsigned long bitmap;
86 };
87 #define NULLB_PAGE_LOCK (sizeof(unsigned long) * 8 - 1)
88 #define NULLB_PAGE_FREE (sizeof(unsigned long) * 8 - 2)
89
90 struct nullb_device {
91 struct nullb *nullb;
92 struct config_item item;
93 struct radix_tree_root data; /* data stored in the disk */
94 struct radix_tree_root cache; /* disk cache data */
95 unsigned long flags; /* device flags */
96 unsigned int curr_cache;
97 struct badblocks badblocks;
98
99 unsigned long size; /* device size in MB */
100 unsigned long completion_nsec; /* time in ns to complete a request */
101 unsigned long cache_size; /* disk cache size in MB */
102 unsigned int submit_queues; /* number of submission queues */
103 unsigned int home_node; /* home node for the device */
104 unsigned int queue_mode; /* block interface */
105 unsigned int blocksize; /* block size */
106 unsigned int irqmode; /* IRQ completion handler */
107 unsigned int hw_queue_depth; /* queue depth */
108 unsigned int index; /* index of the disk, only valid with a disk */
109 unsigned int mbps; /* Bandwidth throttle cap (in MB/s) */
110 bool use_lightnvm; /* register as a LightNVM device */
111 bool blocking; /* blocking blk-mq device */
112 bool use_per_node_hctx; /* use per-node allocation for hardware context */
113 bool power; /* power on/off the device */
114 bool memory_backed; /* if data is stored in memory */
115 bool discard; /* if support discard */
116 };
117
118 struct nullb {
119 struct nullb_device *dev;
120 struct list_head list;
121 unsigned int index;
122 struct request_queue *q;
123 struct gendisk *disk;
124 struct nvm_dev *ndev;
125 struct blk_mq_tag_set *tag_set;
126 struct blk_mq_tag_set __tag_set;
127 unsigned int queue_depth;
128 atomic_long_t cur_bytes;
129 struct hrtimer bw_timer;
130 unsigned long cache_flush_pos;
131 spinlock_t lock;
132
133 struct nullb_queue *queues;
134 unsigned int nr_queues;
135 char disk_name[DISK_NAME_LEN];
136 };
137
138 static LIST_HEAD(nullb_list);
139 static struct mutex lock;
140 static int null_major;
141 static DEFINE_IDA(nullb_indexes);
142 static struct kmem_cache *ppa_cache;
143 static struct blk_mq_tag_set tag_set;
144
145 enum {
146 NULL_IRQ_NONE = 0,
147 NULL_IRQ_SOFTIRQ = 1,
148 NULL_IRQ_TIMER = 2,
149 };
150
151 enum {
152 NULL_Q_BIO = 0,
153 NULL_Q_RQ = 1,
154 NULL_Q_MQ = 2,
155 };
156
157 static int g_no_sched;
158 module_param_named(no_sched, g_no_sched, int, S_IRUGO);
159 MODULE_PARM_DESC(no_sched, "No io scheduler");
160
161 static int g_submit_queues = 1;
162 module_param_named(submit_queues, g_submit_queues, int, S_IRUGO);
163 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
164
165 static int g_home_node = NUMA_NO_NODE;
166 module_param_named(home_node, g_home_node, int, S_IRUGO);
167 MODULE_PARM_DESC(home_node, "Home node for the device");
168
169 static int g_queue_mode = NULL_Q_MQ;
170
171 static int null_param_store_val(const char *str, int *val, int min, int max)
172 {
173 int ret, new_val;
174
175 ret = kstrtoint(str, 10, &new_val);
176 if (ret)
177 return -EINVAL;
178
179 if (new_val < min || new_val > max)
180 return -EINVAL;
181
182 *val = new_val;
183 return 0;
184 }
185
186 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
187 {
188 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
189 }
190
191 static const struct kernel_param_ops null_queue_mode_param_ops = {
192 .set = null_set_queue_mode,
193 .get = param_get_int,
194 };
195
196 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, S_IRUGO);
197 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
198
199 static int g_gb = 250;
200 module_param_named(gb, g_gb, int, S_IRUGO);
201 MODULE_PARM_DESC(gb, "Size in GB");
202
203 static int g_bs = 512;
204 module_param_named(bs, g_bs, int, S_IRUGO);
205 MODULE_PARM_DESC(bs, "Block size (in bytes)");
206
207 static int nr_devices = 1;
208 module_param(nr_devices, int, S_IRUGO);
209 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
210
211 static bool g_use_lightnvm;
212 module_param_named(use_lightnvm, g_use_lightnvm, bool, S_IRUGO);
213 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
214
215 static bool g_blocking;
216 module_param_named(blocking, g_blocking, bool, S_IRUGO);
217 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
218
219 static bool shared_tags;
220 module_param(shared_tags, bool, S_IRUGO);
221 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
222
223 static int g_irqmode = NULL_IRQ_SOFTIRQ;
224
225 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
226 {
227 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
228 NULL_IRQ_TIMER);
229 }
230
231 static const struct kernel_param_ops null_irqmode_param_ops = {
232 .set = null_set_irqmode,
233 .get = param_get_int,
234 };
235
236 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, S_IRUGO);
237 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
238
239 static unsigned long g_completion_nsec = 10000;
240 module_param_named(completion_nsec, g_completion_nsec, ulong, S_IRUGO);
241 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
242
243 static int g_hw_queue_depth = 64;
244 module_param_named(hw_queue_depth, g_hw_queue_depth, int, S_IRUGO);
245 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
246
247 static bool g_use_per_node_hctx;
248 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, S_IRUGO);
249 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
250
251 static struct nullb_device *null_alloc_dev(void);
252 static void null_free_dev(struct nullb_device *dev);
253 static void null_del_dev(struct nullb *nullb);
254 static int null_add_dev(struct nullb_device *dev);
255 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
256
257 static inline struct nullb_device *to_nullb_device(struct config_item *item)
258 {
259 return item ? container_of(item, struct nullb_device, item) : NULL;
260 }
261
262 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
263 {
264 return snprintf(page, PAGE_SIZE, "%u\n", val);
265 }
266
267 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
268 char *page)
269 {
270 return snprintf(page, PAGE_SIZE, "%lu\n", val);
271 }
272
273 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
274 {
275 return snprintf(page, PAGE_SIZE, "%u\n", val);
276 }
277
278 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
279 const char *page, size_t count)
280 {
281 unsigned int tmp;
282 int result;
283
284 result = kstrtouint(page, 0, &tmp);
285 if (result)
286 return result;
287
288 *val = tmp;
289 return count;
290 }
291
292 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
293 const char *page, size_t count)
294 {
295 int result;
296 unsigned long tmp;
297
298 result = kstrtoul(page, 0, &tmp);
299 if (result)
300 return result;
301
302 *val = tmp;
303 return count;
304 }
305
306 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
307 size_t count)
308 {
309 bool tmp;
310 int result;
311
312 result = kstrtobool(page, &tmp);
313 if (result)
314 return result;
315
316 *val = tmp;
317 return count;
318 }
319
320 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
321 #define NULLB_DEVICE_ATTR(NAME, TYPE) \
322 static ssize_t \
323 nullb_device_##NAME##_show(struct config_item *item, char *page) \
324 { \
325 return nullb_device_##TYPE##_attr_show( \
326 to_nullb_device(item)->NAME, page); \
327 } \
328 static ssize_t \
329 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
330 size_t count) \
331 { \
332 if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags)) \
333 return -EBUSY; \
334 return nullb_device_##TYPE##_attr_store( \
335 &to_nullb_device(item)->NAME, page, count); \
336 } \
337 CONFIGFS_ATTR(nullb_device_, NAME);
338
339 NULLB_DEVICE_ATTR(size, ulong);
340 NULLB_DEVICE_ATTR(completion_nsec, ulong);
341 NULLB_DEVICE_ATTR(submit_queues, uint);
342 NULLB_DEVICE_ATTR(home_node, uint);
343 NULLB_DEVICE_ATTR(queue_mode, uint);
344 NULLB_DEVICE_ATTR(blocksize, uint);
345 NULLB_DEVICE_ATTR(irqmode, uint);
346 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
347 NULLB_DEVICE_ATTR(index, uint);
348 NULLB_DEVICE_ATTR(use_lightnvm, bool);
349 NULLB_DEVICE_ATTR(blocking, bool);
350 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
351 NULLB_DEVICE_ATTR(memory_backed, bool);
352 NULLB_DEVICE_ATTR(discard, bool);
353 NULLB_DEVICE_ATTR(mbps, uint);
354 NULLB_DEVICE_ATTR(cache_size, ulong);
355
356 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
357 {
358 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
359 }
360
361 static ssize_t nullb_device_power_store(struct config_item *item,
362 const char *page, size_t count)
363 {
364 struct nullb_device *dev = to_nullb_device(item);
365 bool newp = false;
366 ssize_t ret;
367
368 ret = nullb_device_bool_attr_store(&newp, page, count);
369 if (ret < 0)
370 return ret;
371
372 if (!dev->power && newp) {
373 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
374 return count;
375 if (null_add_dev(dev)) {
376 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
377 return -ENOMEM;
378 }
379
380 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
381 dev->power = newp;
382 } else if (dev->power && !newp) {
383 mutex_lock(&lock);
384 dev->power = newp;
385 null_del_dev(dev->nullb);
386 mutex_unlock(&lock);
387 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
388 }
389
390 return count;
391 }
392
393 CONFIGFS_ATTR(nullb_device_, power);
394
395 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
396 {
397 struct nullb_device *t_dev = to_nullb_device(item);
398
399 return badblocks_show(&t_dev->badblocks, page, 0);
400 }
401
402 static ssize_t nullb_device_badblocks_store(struct config_item *item,
403 const char *page, size_t count)
404 {
405 struct nullb_device *t_dev = to_nullb_device(item);
406 char *orig, *buf, *tmp;
407 u64 start, end;
408 int ret;
409
410 orig = kstrndup(page, count, GFP_KERNEL);
411 if (!orig)
412 return -ENOMEM;
413
414 buf = strstrip(orig);
415
416 ret = -EINVAL;
417 if (buf[0] != '+' && buf[0] != '-')
418 goto out;
419 tmp = strchr(&buf[1], '-');
420 if (!tmp)
421 goto out;
422 *tmp = '\0';
423 ret = kstrtoull(buf + 1, 0, &start);
424 if (ret)
425 goto out;
426 ret = kstrtoull(tmp + 1, 0, &end);
427 if (ret)
428 goto out;
429 ret = -EINVAL;
430 if (start > end)
431 goto out;
432 /* enable badblocks */
433 cmpxchg(&t_dev->badblocks.shift, -1, 0);
434 if (buf[0] == '+')
435 ret = badblocks_set(&t_dev->badblocks, start,
436 end - start + 1, 1);
437 else
438 ret = badblocks_clear(&t_dev->badblocks, start,
439 end - start + 1);
440 if (ret == 0)
441 ret = count;
442 out:
443 kfree(orig);
444 return ret;
445 }
446 CONFIGFS_ATTR(nullb_device_, badblocks);
447
448 static struct configfs_attribute *nullb_device_attrs[] = {
449 &nullb_device_attr_size,
450 &nullb_device_attr_completion_nsec,
451 &nullb_device_attr_submit_queues,
452 &nullb_device_attr_home_node,
453 &nullb_device_attr_queue_mode,
454 &nullb_device_attr_blocksize,
455 &nullb_device_attr_irqmode,
456 &nullb_device_attr_hw_queue_depth,
457 &nullb_device_attr_index,
458 &nullb_device_attr_use_lightnvm,
459 &nullb_device_attr_blocking,
460 &nullb_device_attr_use_per_node_hctx,
461 &nullb_device_attr_power,
462 &nullb_device_attr_memory_backed,
463 &nullb_device_attr_discard,
464 &nullb_device_attr_mbps,
465 &nullb_device_attr_cache_size,
466 &nullb_device_attr_badblocks,
467 NULL,
468 };
469
470 static void nullb_device_release(struct config_item *item)
471 {
472 struct nullb_device *dev = to_nullb_device(item);
473
474 null_free_device_storage(dev, false);
475 null_free_dev(dev);
476 }
477
478 static struct configfs_item_operations nullb_device_ops = {
479 .release = nullb_device_release,
480 };
481
482 static const struct config_item_type nullb_device_type = {
483 .ct_item_ops = &nullb_device_ops,
484 .ct_attrs = nullb_device_attrs,
485 .ct_owner = THIS_MODULE,
486 };
487
488 static struct
489 config_item *nullb_group_make_item(struct config_group *group, const char *name)
490 {
491 struct nullb_device *dev;
492
493 dev = null_alloc_dev();
494 if (!dev)
495 return ERR_PTR(-ENOMEM);
496
497 config_item_init_type_name(&dev->item, name, &nullb_device_type);
498
499 return &dev->item;
500 }
501
502 static void
503 nullb_group_drop_item(struct config_group *group, struct config_item *item)
504 {
505 struct nullb_device *dev = to_nullb_device(item);
506
507 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
508 mutex_lock(&lock);
509 dev->power = false;
510 null_del_dev(dev->nullb);
511 mutex_unlock(&lock);
512 }
513
514 config_item_put(item);
515 }
516
517 static ssize_t memb_group_features_show(struct config_item *item, char *page)
518 {
519 return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks\n");
520 }
521
522 CONFIGFS_ATTR_RO(memb_group_, features);
523
524 static struct configfs_attribute *nullb_group_attrs[] = {
525 &memb_group_attr_features,
526 NULL,
527 };
528
529 static struct configfs_group_operations nullb_group_ops = {
530 .make_item = nullb_group_make_item,
531 .drop_item = nullb_group_drop_item,
532 };
533
534 static const struct config_item_type nullb_group_type = {
535 .ct_group_ops = &nullb_group_ops,
536 .ct_attrs = nullb_group_attrs,
537 .ct_owner = THIS_MODULE,
538 };
539
540 static struct configfs_subsystem nullb_subsys = {
541 .su_group = {
542 .cg_item = {
543 .ci_namebuf = "nullb",
544 .ci_type = &nullb_group_type,
545 },
546 },
547 };
548
549 static inline int null_cache_active(struct nullb *nullb)
550 {
551 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
552 }
553
554 static struct nullb_device *null_alloc_dev(void)
555 {
556 struct nullb_device *dev;
557
558 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
559 if (!dev)
560 return NULL;
561 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
562 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
563 if (badblocks_init(&dev->badblocks, 0)) {
564 kfree(dev);
565 return NULL;
566 }
567
568 dev->size = g_gb * 1024;
569 dev->completion_nsec = g_completion_nsec;
570 dev->submit_queues = g_submit_queues;
571 dev->home_node = g_home_node;
572 dev->queue_mode = g_queue_mode;
573 dev->blocksize = g_bs;
574 dev->irqmode = g_irqmode;
575 dev->hw_queue_depth = g_hw_queue_depth;
576 dev->use_lightnvm = g_use_lightnvm;
577 dev->blocking = g_blocking;
578 dev->use_per_node_hctx = g_use_per_node_hctx;
579 return dev;
580 }
581
582 static void null_free_dev(struct nullb_device *dev)
583 {
584 if (!dev)
585 return;
586
587 badblocks_exit(&dev->badblocks);
588 kfree(dev);
589 }
590
591 static void put_tag(struct nullb_queue *nq, unsigned int tag)
592 {
593 clear_bit_unlock(tag, nq->tag_map);
594
595 if (waitqueue_active(&nq->wait))
596 wake_up(&nq->wait);
597 }
598
599 static unsigned int get_tag(struct nullb_queue *nq)
600 {
601 unsigned int tag;
602
603 do {
604 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
605 if (tag >= nq->queue_depth)
606 return -1U;
607 } while (test_and_set_bit_lock(tag, nq->tag_map));
608
609 return tag;
610 }
611
612 static void free_cmd(struct nullb_cmd *cmd)
613 {
614 put_tag(cmd->nq, cmd->tag);
615 }
616
617 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
618
619 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
620 {
621 struct nullb_cmd *cmd;
622 unsigned int tag;
623
624 tag = get_tag(nq);
625 if (tag != -1U) {
626 cmd = &nq->cmds[tag];
627 cmd->tag = tag;
628 cmd->nq = nq;
629 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
630 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
631 HRTIMER_MODE_REL);
632 cmd->timer.function = null_cmd_timer_expired;
633 }
634 return cmd;
635 }
636
637 return NULL;
638 }
639
640 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
641 {
642 struct nullb_cmd *cmd;
643 DEFINE_WAIT(wait);
644
645 cmd = __alloc_cmd(nq);
646 if (cmd || !can_wait)
647 return cmd;
648
649 do {
650 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
651 cmd = __alloc_cmd(nq);
652 if (cmd)
653 break;
654
655 io_schedule();
656 } while (1);
657
658 finish_wait(&nq->wait, &wait);
659 return cmd;
660 }
661
662 static void end_cmd(struct nullb_cmd *cmd)
663 {
664 struct request_queue *q = NULL;
665 int queue_mode = cmd->nq->dev->queue_mode;
666
667 if (cmd->rq)
668 q = cmd->rq->q;
669
670 switch (queue_mode) {
671 case NULL_Q_MQ:
672 blk_mq_end_request(cmd->rq, cmd->error);
673 return;
674 case NULL_Q_RQ:
675 INIT_LIST_HEAD(&cmd->rq->queuelist);
676 blk_end_request_all(cmd->rq, cmd->error);
677 break;
678 case NULL_Q_BIO:
679 cmd->bio->bi_status = cmd->error;
680 bio_endio(cmd->bio);
681 break;
682 }
683
684 free_cmd(cmd);
685
686 /* Restart queue if needed, as we are freeing a tag */
687 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
688 unsigned long flags;
689
690 spin_lock_irqsave(q->queue_lock, flags);
691 blk_start_queue_async(q);
692 spin_unlock_irqrestore(q->queue_lock, flags);
693 }
694 }
695
696 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
697 {
698 end_cmd(container_of(timer, struct nullb_cmd, timer));
699
700 return HRTIMER_NORESTART;
701 }
702
703 static void null_cmd_end_timer(struct nullb_cmd *cmd)
704 {
705 ktime_t kt = cmd->nq->dev->completion_nsec;
706
707 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
708 }
709
710 static void null_softirq_done_fn(struct request *rq)
711 {
712 struct nullb *nullb = rq->q->queuedata;
713
714 if (nullb->dev->queue_mode == NULL_Q_MQ)
715 end_cmd(blk_mq_rq_to_pdu(rq));
716 else
717 end_cmd(rq->special);
718 }
719
720 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
721 {
722 struct nullb_page *t_page;
723
724 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
725 if (!t_page)
726 goto out;
727
728 t_page->page = alloc_pages(gfp_flags, 0);
729 if (!t_page->page)
730 goto out_freepage;
731
732 t_page->bitmap = 0;
733 return t_page;
734 out_freepage:
735 kfree(t_page);
736 out:
737 return NULL;
738 }
739
740 static void null_free_page(struct nullb_page *t_page)
741 {
742 __set_bit(NULLB_PAGE_FREE, &t_page->bitmap);
743 if (test_bit(NULLB_PAGE_LOCK, &t_page->bitmap))
744 return;
745 __free_page(t_page->page);
746 kfree(t_page);
747 }
748
749 static void null_free_sector(struct nullb *nullb, sector_t sector,
750 bool is_cache)
751 {
752 unsigned int sector_bit;
753 u64 idx;
754 struct nullb_page *t_page, *ret;
755 struct radix_tree_root *root;
756
757 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
758 idx = sector >> PAGE_SECTORS_SHIFT;
759 sector_bit = (sector & SECTOR_MASK);
760
761 t_page = radix_tree_lookup(root, idx);
762 if (t_page) {
763 __clear_bit(sector_bit, &t_page->bitmap);
764
765 if (!t_page->bitmap) {
766 ret = radix_tree_delete_item(root, idx, t_page);
767 WARN_ON(ret != t_page);
768 null_free_page(ret);
769 if (is_cache)
770 nullb->dev->curr_cache -= PAGE_SIZE;
771 }
772 }
773 }
774
775 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
776 struct nullb_page *t_page, bool is_cache)
777 {
778 struct radix_tree_root *root;
779
780 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
781
782 if (radix_tree_insert(root, idx, t_page)) {
783 null_free_page(t_page);
784 t_page = radix_tree_lookup(root, idx);
785 WARN_ON(!t_page || t_page->page->index != idx);
786 } else if (is_cache)
787 nullb->dev->curr_cache += PAGE_SIZE;
788
789 return t_page;
790 }
791
792 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
793 {
794 unsigned long pos = 0;
795 int nr_pages;
796 struct nullb_page *ret, *t_pages[FREE_BATCH];
797 struct radix_tree_root *root;
798
799 root = is_cache ? &dev->cache : &dev->data;
800
801 do {
802 int i;
803
804 nr_pages = radix_tree_gang_lookup(root,
805 (void **)t_pages, pos, FREE_BATCH);
806
807 for (i = 0; i < nr_pages; i++) {
808 pos = t_pages[i]->page->index;
809 ret = radix_tree_delete_item(root, pos, t_pages[i]);
810 WARN_ON(ret != t_pages[i]);
811 null_free_page(ret);
812 }
813
814 pos++;
815 } while (nr_pages == FREE_BATCH);
816
817 if (is_cache)
818 dev->curr_cache = 0;
819 }
820
821 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
822 sector_t sector, bool for_write, bool is_cache)
823 {
824 unsigned int sector_bit;
825 u64 idx;
826 struct nullb_page *t_page;
827 struct radix_tree_root *root;
828
829 idx = sector >> PAGE_SECTORS_SHIFT;
830 sector_bit = (sector & SECTOR_MASK);
831
832 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
833 t_page = radix_tree_lookup(root, idx);
834 WARN_ON(t_page && t_page->page->index != idx);
835
836 if (t_page && (for_write || test_bit(sector_bit, &t_page->bitmap)))
837 return t_page;
838
839 return NULL;
840 }
841
842 static struct nullb_page *null_lookup_page(struct nullb *nullb,
843 sector_t sector, bool for_write, bool ignore_cache)
844 {
845 struct nullb_page *page = NULL;
846
847 if (!ignore_cache)
848 page = __null_lookup_page(nullb, sector, for_write, true);
849 if (page)
850 return page;
851 return __null_lookup_page(nullb, sector, for_write, false);
852 }
853
854 static struct nullb_page *null_insert_page(struct nullb *nullb,
855 sector_t sector, bool ignore_cache)
856 {
857 u64 idx;
858 struct nullb_page *t_page;
859
860 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
861 if (t_page)
862 return t_page;
863
864 spin_unlock_irq(&nullb->lock);
865
866 t_page = null_alloc_page(GFP_NOIO);
867 if (!t_page)
868 goto out_lock;
869
870 if (radix_tree_preload(GFP_NOIO))
871 goto out_freepage;
872
873 spin_lock_irq(&nullb->lock);
874 idx = sector >> PAGE_SECTORS_SHIFT;
875 t_page->page->index = idx;
876 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
877 radix_tree_preload_end();
878
879 return t_page;
880 out_freepage:
881 null_free_page(t_page);
882 out_lock:
883 spin_lock_irq(&nullb->lock);
884 return null_lookup_page(nullb, sector, true, ignore_cache);
885 }
886
887 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
888 {
889 int i;
890 unsigned int offset;
891 u64 idx;
892 struct nullb_page *t_page, *ret;
893 void *dst, *src;
894
895 idx = c_page->page->index;
896
897 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
898
899 __clear_bit(NULLB_PAGE_LOCK, &c_page->bitmap);
900 if (test_bit(NULLB_PAGE_FREE, &c_page->bitmap)) {
901 null_free_page(c_page);
902 if (t_page && t_page->bitmap == 0) {
903 ret = radix_tree_delete_item(&nullb->dev->data,
904 idx, t_page);
905 null_free_page(t_page);
906 }
907 return 0;
908 }
909
910 if (!t_page)
911 return -ENOMEM;
912
913 src = kmap_atomic(c_page->page);
914 dst = kmap_atomic(t_page->page);
915
916 for (i = 0; i < PAGE_SECTORS;
917 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
918 if (test_bit(i, &c_page->bitmap)) {
919 offset = (i << SECTOR_SHIFT);
920 memcpy(dst + offset, src + offset,
921 nullb->dev->blocksize);
922 __set_bit(i, &t_page->bitmap);
923 }
924 }
925
926 kunmap_atomic(dst);
927 kunmap_atomic(src);
928
929 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
930 null_free_page(ret);
931 nullb->dev->curr_cache -= PAGE_SIZE;
932
933 return 0;
934 }
935
936 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
937 {
938 int i, err, nr_pages;
939 struct nullb_page *c_pages[FREE_BATCH];
940 unsigned long flushed = 0, one_round;
941
942 again:
943 if ((nullb->dev->cache_size * 1024 * 1024) >
944 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
945 return 0;
946
947 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
948 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
949 /*
950 * nullb_flush_cache_page could unlock before using the c_pages. To
951 * avoid race, we don't allow page free
952 */
953 for (i = 0; i < nr_pages; i++) {
954 nullb->cache_flush_pos = c_pages[i]->page->index;
955 /*
956 * We found the page which is being flushed to disk by other
957 * threads
958 */
959 if (test_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap))
960 c_pages[i] = NULL;
961 else
962 __set_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap);
963 }
964
965 one_round = 0;
966 for (i = 0; i < nr_pages; i++) {
967 if (c_pages[i] == NULL)
968 continue;
969 err = null_flush_cache_page(nullb, c_pages[i]);
970 if (err)
971 return err;
972 one_round++;
973 }
974 flushed += one_round << PAGE_SHIFT;
975
976 if (n > flushed) {
977 if (nr_pages == 0)
978 nullb->cache_flush_pos = 0;
979 if (one_round == 0) {
980 /* give other threads a chance */
981 spin_unlock_irq(&nullb->lock);
982 spin_lock_irq(&nullb->lock);
983 }
984 goto again;
985 }
986 return 0;
987 }
988
989 static int copy_to_nullb(struct nullb *nullb, struct page *source,
990 unsigned int off, sector_t sector, size_t n, bool is_fua)
991 {
992 size_t temp, count = 0;
993 unsigned int offset;
994 struct nullb_page *t_page;
995 void *dst, *src;
996
997 while (count < n) {
998 temp = min_t(size_t, nullb->dev->blocksize, n - count);
999
1000 if (null_cache_active(nullb) && !is_fua)
1001 null_make_cache_space(nullb, PAGE_SIZE);
1002
1003 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1004 t_page = null_insert_page(nullb, sector,
1005 !null_cache_active(nullb) || is_fua);
1006 if (!t_page)
1007 return -ENOSPC;
1008
1009 src = kmap_atomic(source);
1010 dst = kmap_atomic(t_page->page);
1011 memcpy(dst + offset, src + off + count, temp);
1012 kunmap_atomic(dst);
1013 kunmap_atomic(src);
1014
1015 __set_bit(sector & SECTOR_MASK, &t_page->bitmap);
1016
1017 if (is_fua)
1018 null_free_sector(nullb, sector, true);
1019
1020 count += temp;
1021 sector += temp >> SECTOR_SHIFT;
1022 }
1023 return 0;
1024 }
1025
1026 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1027 unsigned int off, sector_t sector, size_t n)
1028 {
1029 size_t temp, count = 0;
1030 unsigned int offset;
1031 struct nullb_page *t_page;
1032 void *dst, *src;
1033
1034 while (count < n) {
1035 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1036
1037 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1038 t_page = null_lookup_page(nullb, sector, false,
1039 !null_cache_active(nullb));
1040
1041 dst = kmap_atomic(dest);
1042 if (!t_page) {
1043 memset(dst + off + count, 0, temp);
1044 goto next;
1045 }
1046 src = kmap_atomic(t_page->page);
1047 memcpy(dst + off + count, src + offset, temp);
1048 kunmap_atomic(src);
1049 next:
1050 kunmap_atomic(dst);
1051
1052 count += temp;
1053 sector += temp >> SECTOR_SHIFT;
1054 }
1055 return 0;
1056 }
1057
1058 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1059 {
1060 size_t temp;
1061
1062 spin_lock_irq(&nullb->lock);
1063 while (n > 0) {
1064 temp = min_t(size_t, n, nullb->dev->blocksize);
1065 null_free_sector(nullb, sector, false);
1066 if (null_cache_active(nullb))
1067 null_free_sector(nullb, sector, true);
1068 sector += temp >> SECTOR_SHIFT;
1069 n -= temp;
1070 }
1071 spin_unlock_irq(&nullb->lock);
1072 }
1073
1074 static int null_handle_flush(struct nullb *nullb)
1075 {
1076 int err;
1077
1078 if (!null_cache_active(nullb))
1079 return 0;
1080
1081 spin_lock_irq(&nullb->lock);
1082 while (true) {
1083 err = null_make_cache_space(nullb,
1084 nullb->dev->cache_size * 1024 * 1024);
1085 if (err || nullb->dev->curr_cache == 0)
1086 break;
1087 }
1088
1089 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1090 spin_unlock_irq(&nullb->lock);
1091 return err;
1092 }
1093
1094 static int null_transfer(struct nullb *nullb, struct page *page,
1095 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1096 bool is_fua)
1097 {
1098 int err = 0;
1099
1100 if (!is_write) {
1101 err = copy_from_nullb(nullb, page, off, sector, len);
1102 flush_dcache_page(page);
1103 } else {
1104 flush_dcache_page(page);
1105 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1106 }
1107
1108 return err;
1109 }
1110
1111 static int null_handle_rq(struct nullb_cmd *cmd)
1112 {
1113 struct request *rq = cmd->rq;
1114 struct nullb *nullb = cmd->nq->dev->nullb;
1115 int err;
1116 unsigned int len;
1117 sector_t sector;
1118 struct req_iterator iter;
1119 struct bio_vec bvec;
1120
1121 sector = blk_rq_pos(rq);
1122
1123 if (req_op(rq) == REQ_OP_DISCARD) {
1124 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1125 return 0;
1126 }
1127
1128 spin_lock_irq(&nullb->lock);
1129 rq_for_each_segment(bvec, rq, iter) {
1130 len = bvec.bv_len;
1131 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1132 op_is_write(req_op(rq)), sector,
1133 req_op(rq) & REQ_FUA);
1134 if (err) {
1135 spin_unlock_irq(&nullb->lock);
1136 return err;
1137 }
1138 sector += len >> SECTOR_SHIFT;
1139 }
1140 spin_unlock_irq(&nullb->lock);
1141
1142 return 0;
1143 }
1144
1145 static int null_handle_bio(struct nullb_cmd *cmd)
1146 {
1147 struct bio *bio = cmd->bio;
1148 struct nullb *nullb = cmd->nq->dev->nullb;
1149 int err;
1150 unsigned int len;
1151 sector_t sector;
1152 struct bio_vec bvec;
1153 struct bvec_iter iter;
1154
1155 sector = bio->bi_iter.bi_sector;
1156
1157 if (bio_op(bio) == REQ_OP_DISCARD) {
1158 null_handle_discard(nullb, sector,
1159 bio_sectors(bio) << SECTOR_SHIFT);
1160 return 0;
1161 }
1162
1163 spin_lock_irq(&nullb->lock);
1164 bio_for_each_segment(bvec, bio, iter) {
1165 len = bvec.bv_len;
1166 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1167 op_is_write(bio_op(bio)), sector,
1168 bio_op(bio) & REQ_FUA);
1169 if (err) {
1170 spin_unlock_irq(&nullb->lock);
1171 return err;
1172 }
1173 sector += len >> SECTOR_SHIFT;
1174 }
1175 spin_unlock_irq(&nullb->lock);
1176 return 0;
1177 }
1178
1179 static void null_stop_queue(struct nullb *nullb)
1180 {
1181 struct request_queue *q = nullb->q;
1182
1183 if (nullb->dev->queue_mode == NULL_Q_MQ)
1184 blk_mq_stop_hw_queues(q);
1185 else {
1186 spin_lock_irq(q->queue_lock);
1187 blk_stop_queue(q);
1188 spin_unlock_irq(q->queue_lock);
1189 }
1190 }
1191
1192 static void null_restart_queue_async(struct nullb *nullb)
1193 {
1194 struct request_queue *q = nullb->q;
1195 unsigned long flags;
1196
1197 if (nullb->dev->queue_mode == NULL_Q_MQ)
1198 blk_mq_start_stopped_hw_queues(q, true);
1199 else {
1200 spin_lock_irqsave(q->queue_lock, flags);
1201 blk_start_queue_async(q);
1202 spin_unlock_irqrestore(q->queue_lock, flags);
1203 }
1204 }
1205
1206 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd)
1207 {
1208 struct nullb_device *dev = cmd->nq->dev;
1209 struct nullb *nullb = dev->nullb;
1210 int err = 0;
1211
1212 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1213 struct request *rq = cmd->rq;
1214
1215 if (!hrtimer_active(&nullb->bw_timer))
1216 hrtimer_restart(&nullb->bw_timer);
1217
1218 if (atomic_long_sub_return(blk_rq_bytes(rq),
1219 &nullb->cur_bytes) < 0) {
1220 null_stop_queue(nullb);
1221 /* race with timer */
1222 if (atomic_long_read(&nullb->cur_bytes) > 0)
1223 null_restart_queue_async(nullb);
1224 if (dev->queue_mode == NULL_Q_RQ) {
1225 struct request_queue *q = nullb->q;
1226
1227 spin_lock_irq(q->queue_lock);
1228 rq->rq_flags |= RQF_DONTPREP;
1229 blk_requeue_request(q, rq);
1230 spin_unlock_irq(q->queue_lock);
1231 return BLK_STS_OK;
1232 } else
1233 /* requeue request */
1234 return BLK_STS_RESOURCE;
1235 }
1236 }
1237
1238 if (nullb->dev->badblocks.shift != -1) {
1239 int bad_sectors;
1240 sector_t sector, size, first_bad;
1241 bool is_flush = true;
1242
1243 if (dev->queue_mode == NULL_Q_BIO &&
1244 bio_op(cmd->bio) != REQ_OP_FLUSH) {
1245 is_flush = false;
1246 sector = cmd->bio->bi_iter.bi_sector;
1247 size = bio_sectors(cmd->bio);
1248 }
1249 if (dev->queue_mode != NULL_Q_BIO &&
1250 req_op(cmd->rq) != REQ_OP_FLUSH) {
1251 is_flush = false;
1252 sector = blk_rq_pos(cmd->rq);
1253 size = blk_rq_sectors(cmd->rq);
1254 }
1255 if (!is_flush && badblocks_check(&nullb->dev->badblocks, sector,
1256 size, &first_bad, &bad_sectors)) {
1257 cmd->error = BLK_STS_IOERR;
1258 goto out;
1259 }
1260 }
1261
1262 if (dev->memory_backed) {
1263 if (dev->queue_mode == NULL_Q_BIO) {
1264 if (bio_op(cmd->bio) == REQ_OP_FLUSH)
1265 err = null_handle_flush(nullb);
1266 else
1267 err = null_handle_bio(cmd);
1268 } else {
1269 if (req_op(cmd->rq) == REQ_OP_FLUSH)
1270 err = null_handle_flush(nullb);
1271 else
1272 err = null_handle_rq(cmd);
1273 }
1274 }
1275 cmd->error = errno_to_blk_status(err);
1276 out:
1277 /* Complete IO by inline, softirq or timer */
1278 switch (dev->irqmode) {
1279 case NULL_IRQ_SOFTIRQ:
1280 switch (dev->queue_mode) {
1281 case NULL_Q_MQ:
1282 blk_mq_complete_request(cmd->rq);
1283 break;
1284 case NULL_Q_RQ:
1285 blk_complete_request(cmd->rq);
1286 break;
1287 case NULL_Q_BIO:
1288 /*
1289 * XXX: no proper submitting cpu information available.
1290 */
1291 end_cmd(cmd);
1292 break;
1293 }
1294 break;
1295 case NULL_IRQ_NONE:
1296 end_cmd(cmd);
1297 break;
1298 case NULL_IRQ_TIMER:
1299 null_cmd_end_timer(cmd);
1300 break;
1301 }
1302 return BLK_STS_OK;
1303 }
1304
1305 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1306 {
1307 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1308 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1309 unsigned int mbps = nullb->dev->mbps;
1310
1311 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1312 return HRTIMER_NORESTART;
1313
1314 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1315 null_restart_queue_async(nullb);
1316
1317 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1318
1319 return HRTIMER_RESTART;
1320 }
1321
1322 static void nullb_setup_bwtimer(struct nullb *nullb)
1323 {
1324 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1325
1326 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1327 nullb->bw_timer.function = nullb_bwtimer_fn;
1328 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1329 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1330 }
1331
1332 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1333 {
1334 int index = 0;
1335
1336 if (nullb->nr_queues != 1)
1337 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1338
1339 return &nullb->queues[index];
1340 }
1341
1342 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1343 {
1344 struct nullb *nullb = q->queuedata;
1345 struct nullb_queue *nq = nullb_to_queue(nullb);
1346 struct nullb_cmd *cmd;
1347
1348 cmd = alloc_cmd(nq, 1);
1349 cmd->bio = bio;
1350
1351 null_handle_cmd(cmd);
1352 return BLK_QC_T_NONE;
1353 }
1354
1355 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
1356 {
1357 struct nullb *nullb = q->queuedata;
1358 struct nullb_queue *nq = nullb_to_queue(nullb);
1359 struct nullb_cmd *cmd;
1360
1361 cmd = alloc_cmd(nq, 0);
1362 if (cmd) {
1363 cmd->rq = req;
1364 req->special = cmd;
1365 return BLKPREP_OK;
1366 }
1367 blk_stop_queue(q);
1368
1369 return BLKPREP_DEFER;
1370 }
1371
1372 static void null_request_fn(struct request_queue *q)
1373 {
1374 struct request *rq;
1375
1376 while ((rq = blk_fetch_request(q)) != NULL) {
1377 struct nullb_cmd *cmd = rq->special;
1378
1379 spin_unlock_irq(q->queue_lock);
1380 null_handle_cmd(cmd);
1381 spin_lock_irq(q->queue_lock);
1382 }
1383 }
1384
1385 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1386 const struct blk_mq_queue_data *bd)
1387 {
1388 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1389 struct nullb_queue *nq = hctx->driver_data;
1390
1391 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1392
1393 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1394 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1395 cmd->timer.function = null_cmd_timer_expired;
1396 }
1397 cmd->rq = bd->rq;
1398 cmd->nq = nq;
1399
1400 blk_mq_start_request(bd->rq);
1401
1402 return null_handle_cmd(cmd);
1403 }
1404
1405 static const struct blk_mq_ops null_mq_ops = {
1406 .queue_rq = null_queue_rq,
1407 .complete = null_softirq_done_fn,
1408 };
1409
1410 static void cleanup_queue(struct nullb_queue *nq)
1411 {
1412 kfree(nq->tag_map);
1413 kfree(nq->cmds);
1414 }
1415
1416 static void cleanup_queues(struct nullb *nullb)
1417 {
1418 int i;
1419
1420 for (i = 0; i < nullb->nr_queues; i++)
1421 cleanup_queue(&nullb->queues[i]);
1422
1423 kfree(nullb->queues);
1424 }
1425
1426 #ifdef CONFIG_NVM
1427
1428 static void null_lnvm_end_io(struct request *rq, blk_status_t status)
1429 {
1430 struct nvm_rq *rqd = rq->end_io_data;
1431
1432 /* XXX: lighnvm core seems to expect NVM_RSP_* values here.. */
1433 rqd->error = status ? -EIO : 0;
1434 nvm_end_io(rqd);
1435
1436 blk_put_request(rq);
1437 }
1438
1439 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
1440 {
1441 struct request_queue *q = dev->q;
1442 struct request *rq;
1443 struct bio *bio = rqd->bio;
1444
1445 rq = blk_mq_alloc_request(q,
1446 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1447 if (IS_ERR(rq))
1448 return -ENOMEM;
1449
1450 blk_init_request_from_bio(rq, bio);
1451
1452 rq->end_io_data = rqd;
1453
1454 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
1455
1456 return 0;
1457 }
1458
1459 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
1460 {
1461 struct nullb *nullb = dev->q->queuedata;
1462 sector_t size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1463 sector_t blksize;
1464 struct nvm_id_group *grp;
1465
1466 id->ver_id = 0x1;
1467 id->vmnt = 0;
1468 id->cap = 0x2;
1469 id->dom = 0x1;
1470
1471 id->ppaf.blk_offset = 0;
1472 id->ppaf.blk_len = 16;
1473 id->ppaf.pg_offset = 16;
1474 id->ppaf.pg_len = 16;
1475 id->ppaf.sect_offset = 32;
1476 id->ppaf.sect_len = 8;
1477 id->ppaf.pln_offset = 40;
1478 id->ppaf.pln_len = 8;
1479 id->ppaf.lun_offset = 48;
1480 id->ppaf.lun_len = 8;
1481 id->ppaf.ch_offset = 56;
1482 id->ppaf.ch_len = 8;
1483
1484 sector_div(size, nullb->dev->blocksize); /* convert size to pages */
1485 size >>= 8; /* concert size to pgs pr blk */
1486 grp = &id->grp;
1487 grp->mtype = 0;
1488 grp->fmtype = 0;
1489 grp->num_ch = 1;
1490 grp->num_pg = 256;
1491 blksize = size;
1492 size >>= 16;
1493 grp->num_lun = size + 1;
1494 sector_div(blksize, grp->num_lun);
1495 grp->num_blk = blksize;
1496 grp->num_pln = 1;
1497
1498 grp->fpg_sz = nullb->dev->blocksize;
1499 grp->csecs = nullb->dev->blocksize;
1500 grp->trdt = 25000;
1501 grp->trdm = 25000;
1502 grp->tprt = 500000;
1503 grp->tprm = 500000;
1504 grp->tbet = 1500000;
1505 grp->tbem = 1500000;
1506 grp->mpos = 0x010101; /* single plane rwe */
1507 grp->cpar = nullb->dev->hw_queue_depth;
1508
1509 return 0;
1510 }
1511
1512 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
1513 {
1514 mempool_t *virtmem_pool;
1515
1516 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
1517 if (!virtmem_pool) {
1518 pr_err("null_blk: Unable to create virtual memory pool\n");
1519 return NULL;
1520 }
1521
1522 return virtmem_pool;
1523 }
1524
1525 static void null_lnvm_destroy_dma_pool(void *pool)
1526 {
1527 mempool_destroy(pool);
1528 }
1529
1530 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
1531 gfp_t mem_flags, dma_addr_t *dma_handler)
1532 {
1533 return mempool_alloc(pool, mem_flags);
1534 }
1535
1536 static void null_lnvm_dev_dma_free(void *pool, void *entry,
1537 dma_addr_t dma_handler)
1538 {
1539 mempool_free(entry, pool);
1540 }
1541
1542 static struct nvm_dev_ops null_lnvm_dev_ops = {
1543 .identity = null_lnvm_id,
1544 .submit_io = null_lnvm_submit_io,
1545
1546 .create_dma_pool = null_lnvm_create_dma_pool,
1547 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
1548 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
1549 .dev_dma_free = null_lnvm_dev_dma_free,
1550
1551 /* Simulate nvme protocol restriction */
1552 .max_phys_sect = 64,
1553 };
1554
1555 static int null_nvm_register(struct nullb *nullb)
1556 {
1557 struct nvm_dev *dev;
1558 int rv;
1559
1560 dev = nvm_alloc_dev(0);
1561 if (!dev)
1562 return -ENOMEM;
1563
1564 dev->q = nullb->q;
1565 memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
1566 dev->ops = &null_lnvm_dev_ops;
1567
1568 rv = nvm_register(dev);
1569 if (rv) {
1570 kfree(dev);
1571 return rv;
1572 }
1573 nullb->ndev = dev;
1574 return 0;
1575 }
1576
1577 static void null_nvm_unregister(struct nullb *nullb)
1578 {
1579 nvm_unregister(nullb->ndev);
1580 }
1581 #else
1582 static int null_nvm_register(struct nullb *nullb)
1583 {
1584 pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
1585 return -EINVAL;
1586 }
1587 static void null_nvm_unregister(struct nullb *nullb) {}
1588 #endif /* CONFIG_NVM */
1589
1590 static void null_del_dev(struct nullb *nullb)
1591 {
1592 struct nullb_device *dev = nullb->dev;
1593
1594 ida_simple_remove(&nullb_indexes, nullb->index);
1595
1596 list_del_init(&nullb->list);
1597
1598 if (dev->use_lightnvm)
1599 null_nvm_unregister(nullb);
1600 else
1601 del_gendisk(nullb->disk);
1602
1603 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1604 hrtimer_cancel(&nullb->bw_timer);
1605 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1606 null_restart_queue_async(nullb);
1607 }
1608
1609 blk_cleanup_queue(nullb->q);
1610 if (dev->queue_mode == NULL_Q_MQ &&
1611 nullb->tag_set == &nullb->__tag_set)
1612 blk_mq_free_tag_set(nullb->tag_set);
1613 if (!dev->use_lightnvm)
1614 put_disk(nullb->disk);
1615 cleanup_queues(nullb);
1616 if (null_cache_active(nullb))
1617 null_free_device_storage(nullb->dev, true);
1618 kfree(nullb);
1619 dev->nullb = NULL;
1620 }
1621
1622 static void null_config_discard(struct nullb *nullb)
1623 {
1624 if (nullb->dev->discard == false)
1625 return;
1626 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1627 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1628 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1629 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nullb->q);
1630 }
1631
1632 static int null_open(struct block_device *bdev, fmode_t mode)
1633 {
1634 return 0;
1635 }
1636
1637 static void null_release(struct gendisk *disk, fmode_t mode)
1638 {
1639 }
1640
1641 static const struct block_device_operations null_fops = {
1642 .owner = THIS_MODULE,
1643 .open = null_open,
1644 .release = null_release,
1645 };
1646
1647 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1648 {
1649 BUG_ON(!nullb);
1650 BUG_ON(!nq);
1651
1652 init_waitqueue_head(&nq->wait);
1653 nq->queue_depth = nullb->queue_depth;
1654 nq->dev = nullb->dev;
1655 }
1656
1657 static void null_init_queues(struct nullb *nullb)
1658 {
1659 struct request_queue *q = nullb->q;
1660 struct blk_mq_hw_ctx *hctx;
1661 struct nullb_queue *nq;
1662 int i;
1663
1664 queue_for_each_hw_ctx(q, hctx, i) {
1665 if (!hctx->nr_ctx || !hctx->tags)
1666 continue;
1667 nq = &nullb->queues[i];
1668 hctx->driver_data = nq;
1669 null_init_queue(nullb, nq);
1670 nullb->nr_queues++;
1671 }
1672 }
1673
1674 static int setup_commands(struct nullb_queue *nq)
1675 {
1676 struct nullb_cmd *cmd;
1677 int i, tag_size;
1678
1679 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
1680 if (!nq->cmds)
1681 return -ENOMEM;
1682
1683 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1684 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
1685 if (!nq->tag_map) {
1686 kfree(nq->cmds);
1687 return -ENOMEM;
1688 }
1689
1690 for (i = 0; i < nq->queue_depth; i++) {
1691 cmd = &nq->cmds[i];
1692 INIT_LIST_HEAD(&cmd->list);
1693 cmd->ll_list.next = NULL;
1694 cmd->tag = -1U;
1695 }
1696
1697 return 0;
1698 }
1699
1700 static int setup_queues(struct nullb *nullb)
1701 {
1702 nullb->queues = kzalloc(nullb->dev->submit_queues *
1703 sizeof(struct nullb_queue), GFP_KERNEL);
1704 if (!nullb->queues)
1705 return -ENOMEM;
1706
1707 nullb->nr_queues = 0;
1708 nullb->queue_depth = nullb->dev->hw_queue_depth;
1709
1710 return 0;
1711 }
1712
1713 static int init_driver_queues(struct nullb *nullb)
1714 {
1715 struct nullb_queue *nq;
1716 int i, ret = 0;
1717
1718 for (i = 0; i < nullb->dev->submit_queues; i++) {
1719 nq = &nullb->queues[i];
1720
1721 null_init_queue(nullb, nq);
1722
1723 ret = setup_commands(nq);
1724 if (ret)
1725 return ret;
1726 nullb->nr_queues++;
1727 }
1728 return 0;
1729 }
1730
1731 static int null_gendisk_register(struct nullb *nullb)
1732 {
1733 struct gendisk *disk;
1734 sector_t size;
1735
1736 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1737 if (!disk)
1738 return -ENOMEM;
1739 size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1740 set_capacity(disk, size >> 9);
1741
1742 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1743 disk->major = null_major;
1744 disk->first_minor = nullb->index;
1745 disk->fops = &null_fops;
1746 disk->private_data = nullb;
1747 disk->queue = nullb->q;
1748 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1749
1750 add_disk(disk);
1751 return 0;
1752 }
1753
1754 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1755 {
1756 set->ops = &null_mq_ops;
1757 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1758 g_submit_queues;
1759 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1760 g_hw_queue_depth;
1761 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1762 set->cmd_size = sizeof(struct nullb_cmd);
1763 set->flags = BLK_MQ_F_SHOULD_MERGE;
1764 if (g_no_sched)
1765 set->flags |= BLK_MQ_F_NO_SCHED;
1766 set->driver_data = NULL;
1767
1768 if ((nullb && nullb->dev->blocking) || g_blocking)
1769 set->flags |= BLK_MQ_F_BLOCKING;
1770
1771 return blk_mq_alloc_tag_set(set);
1772 }
1773
1774 static void null_validate_conf(struct nullb_device *dev)
1775 {
1776 dev->blocksize = round_down(dev->blocksize, 512);
1777 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1778 if (dev->use_lightnvm && dev->blocksize != 4096)
1779 dev->blocksize = 4096;
1780
1781 if (dev->use_lightnvm && dev->queue_mode != NULL_Q_MQ)
1782 dev->queue_mode = NULL_Q_MQ;
1783
1784 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1785 if (dev->submit_queues != nr_online_nodes)
1786 dev->submit_queues = nr_online_nodes;
1787 } else if (dev->submit_queues > nr_cpu_ids)
1788 dev->submit_queues = nr_cpu_ids;
1789 else if (dev->submit_queues == 0)
1790 dev->submit_queues = 1;
1791
1792 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1793 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1794
1795 /* Do memory allocation, so set blocking */
1796 if (dev->memory_backed)
1797 dev->blocking = true;
1798 else /* cache is meaningless */
1799 dev->cache_size = 0;
1800 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1801 dev->cache_size);
1802 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1803 /* can not stop a queue */
1804 if (dev->queue_mode == NULL_Q_BIO)
1805 dev->mbps = 0;
1806 }
1807
1808 static int null_add_dev(struct nullb_device *dev)
1809 {
1810 struct nullb *nullb;
1811 int rv;
1812
1813 null_validate_conf(dev);
1814
1815 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1816 if (!nullb) {
1817 rv = -ENOMEM;
1818 goto out;
1819 }
1820 nullb->dev = dev;
1821 dev->nullb = nullb;
1822
1823 spin_lock_init(&nullb->lock);
1824
1825 rv = setup_queues(nullb);
1826 if (rv)
1827 goto out_free_nullb;
1828
1829 if (dev->queue_mode == NULL_Q_MQ) {
1830 if (shared_tags) {
1831 nullb->tag_set = &tag_set;
1832 rv = 0;
1833 } else {
1834 nullb->tag_set = &nullb->__tag_set;
1835 rv = null_init_tag_set(nullb, nullb->tag_set);
1836 }
1837
1838 if (rv)
1839 goto out_cleanup_queues;
1840
1841 nullb->q = blk_mq_init_queue(nullb->tag_set);
1842 if (IS_ERR(nullb->q)) {
1843 rv = -ENOMEM;
1844 goto out_cleanup_tags;
1845 }
1846 null_init_queues(nullb);
1847 } else if (dev->queue_mode == NULL_Q_BIO) {
1848 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node);
1849 if (!nullb->q) {
1850 rv = -ENOMEM;
1851 goto out_cleanup_queues;
1852 }
1853 blk_queue_make_request(nullb->q, null_queue_bio);
1854 rv = init_driver_queues(nullb);
1855 if (rv)
1856 goto out_cleanup_blk_queue;
1857 } else {
1858 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock,
1859 dev->home_node);
1860 if (!nullb->q) {
1861 rv = -ENOMEM;
1862 goto out_cleanup_queues;
1863 }
1864 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
1865 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
1866 rv = init_driver_queues(nullb);
1867 if (rv)
1868 goto out_cleanup_blk_queue;
1869 }
1870
1871 if (dev->mbps) {
1872 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1873 nullb_setup_bwtimer(nullb);
1874 }
1875
1876 if (dev->cache_size > 0) {
1877 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1878 blk_queue_write_cache(nullb->q, true, true);
1879 blk_queue_flush_queueable(nullb->q, true);
1880 }
1881
1882 nullb->q->queuedata = nullb;
1883 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
1884 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1885
1886 mutex_lock(&lock);
1887 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1888 dev->index = nullb->index;
1889 mutex_unlock(&lock);
1890
1891 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1892 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1893
1894 null_config_discard(nullb);
1895
1896 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1897
1898 if (dev->use_lightnvm)
1899 rv = null_nvm_register(nullb);
1900 else
1901 rv = null_gendisk_register(nullb);
1902
1903 if (rv)
1904 goto out_cleanup_blk_queue;
1905
1906 mutex_lock(&lock);
1907 list_add_tail(&nullb->list, &nullb_list);
1908 mutex_unlock(&lock);
1909
1910 return 0;
1911 out_cleanup_blk_queue:
1912 blk_cleanup_queue(nullb->q);
1913 out_cleanup_tags:
1914 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1915 blk_mq_free_tag_set(nullb->tag_set);
1916 out_cleanup_queues:
1917 cleanup_queues(nullb);
1918 out_free_nullb:
1919 kfree(nullb);
1920 out:
1921 return rv;
1922 }
1923
1924 static int __init null_init(void)
1925 {
1926 int ret = 0;
1927 unsigned int i;
1928 struct nullb *nullb;
1929 struct nullb_device *dev;
1930
1931 /* check for nullb_page.bitmap */
1932 if (sizeof(unsigned long) * 8 - 2 < (PAGE_SIZE >> SECTOR_SHIFT))
1933 return -EINVAL;
1934
1935 if (g_bs > PAGE_SIZE) {
1936 pr_warn("null_blk: invalid block size\n");
1937 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
1938 g_bs = PAGE_SIZE;
1939 }
1940
1941 if (g_use_lightnvm && g_bs != 4096) {
1942 pr_warn("null_blk: LightNVM only supports 4k block size\n");
1943 pr_warn("null_blk: defaults block size to 4k\n");
1944 g_bs = 4096;
1945 }
1946
1947 if (g_use_lightnvm && g_queue_mode != NULL_Q_MQ) {
1948 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
1949 pr_warn("null_blk: defaults queue mode to blk-mq\n");
1950 g_queue_mode = NULL_Q_MQ;
1951 }
1952
1953 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1954 if (g_submit_queues != nr_online_nodes) {
1955 pr_warn("null_blk: submit_queues param is set to %u.\n",
1956 nr_online_nodes);
1957 g_submit_queues = nr_online_nodes;
1958 }
1959 } else if (g_submit_queues > nr_cpu_ids)
1960 g_submit_queues = nr_cpu_ids;
1961 else if (g_submit_queues <= 0)
1962 g_submit_queues = 1;
1963
1964 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1965 ret = null_init_tag_set(NULL, &tag_set);
1966 if (ret)
1967 return ret;
1968 }
1969
1970 config_group_init(&nullb_subsys.su_group);
1971 mutex_init(&nullb_subsys.su_mutex);
1972
1973 ret = configfs_register_subsystem(&nullb_subsys);
1974 if (ret)
1975 goto err_tagset;
1976
1977 mutex_init(&lock);
1978
1979 null_major = register_blkdev(0, "nullb");
1980 if (null_major < 0) {
1981 ret = null_major;
1982 goto err_conf;
1983 }
1984
1985 if (g_use_lightnvm) {
1986 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
1987 0, 0, NULL);
1988 if (!ppa_cache) {
1989 pr_err("null_blk: unable to create ppa cache\n");
1990 ret = -ENOMEM;
1991 goto err_ppa;
1992 }
1993 }
1994
1995 for (i = 0; i < nr_devices; i++) {
1996 dev = null_alloc_dev();
1997 if (!dev) {
1998 ret = -ENOMEM;
1999 goto err_dev;
2000 }
2001 ret = null_add_dev(dev);
2002 if (ret) {
2003 null_free_dev(dev);
2004 goto err_dev;
2005 }
2006 }
2007
2008 pr_info("null: module loaded\n");
2009 return 0;
2010
2011 err_dev:
2012 while (!list_empty(&nullb_list)) {
2013 nullb = list_entry(nullb_list.next, struct nullb, list);
2014 dev = nullb->dev;
2015 null_del_dev(nullb);
2016 null_free_dev(dev);
2017 }
2018 kmem_cache_destroy(ppa_cache);
2019 err_ppa:
2020 unregister_blkdev(null_major, "nullb");
2021 err_conf:
2022 configfs_unregister_subsystem(&nullb_subsys);
2023 err_tagset:
2024 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2025 blk_mq_free_tag_set(&tag_set);
2026 return ret;
2027 }
2028
2029 static void __exit null_exit(void)
2030 {
2031 struct nullb *nullb;
2032
2033 configfs_unregister_subsystem(&nullb_subsys);
2034
2035 unregister_blkdev(null_major, "nullb");
2036
2037 mutex_lock(&lock);
2038 while (!list_empty(&nullb_list)) {
2039 struct nullb_device *dev;
2040
2041 nullb = list_entry(nullb_list.next, struct nullb, list);
2042 dev = nullb->dev;
2043 null_del_dev(nullb);
2044 null_free_dev(dev);
2045 }
2046 mutex_unlock(&lock);
2047
2048 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2049 blk_mq_free_tag_set(&tag_set);
2050
2051 kmem_cache_destroy(ppa_cache);
2052 }
2053
2054 module_init(null_init);
2055 module_exit(null_exit);
2056
2057 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2058 MODULE_LICENSE("GPL");
Ubuntu Focal Linux kernel mirror