]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/block/null_blk.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mmc: core: prepend 0x to OCR entry in sysfs
[mirror_ubuntu-bionic-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 call_single_data_t csd;
39 struct request *rq;
40 struct bio *bio;
41 unsigned int tag;
42 struct nullb_queue *nq;
43 struct hrtimer timer;
44 blk_status_t error;
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 badblocks_exit(&dev->badblocks);
475 null_free_device_storage(dev, false);
476 null_free_dev(dev);
477 }
478
479 static struct configfs_item_operations nullb_device_ops = {
480 .release = nullb_device_release,
481 };
482
483 static const struct config_item_type nullb_device_type = {
484 .ct_item_ops = &nullb_device_ops,
485 .ct_attrs = nullb_device_attrs,
486 .ct_owner = THIS_MODULE,
487 };
488
489 static struct
490 config_item *nullb_group_make_item(struct config_group *group, const char *name)
491 {
492 struct nullb_device *dev;
493
494 dev = null_alloc_dev();
495 if (!dev)
496 return ERR_PTR(-ENOMEM);
497
498 config_item_init_type_name(&dev->item, name, &nullb_device_type);
499
500 return &dev->item;
501 }
502
503 static void
504 nullb_group_drop_item(struct config_group *group, struct config_item *item)
505 {
506 struct nullb_device *dev = to_nullb_device(item);
507
508 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
509 mutex_lock(&lock);
510 dev->power = false;
511 null_del_dev(dev->nullb);
512 mutex_unlock(&lock);
513 }
514
515 config_item_put(item);
516 }
517
518 static ssize_t memb_group_features_show(struct config_item *item, char *page)
519 {
520 return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks\n");
521 }
522
523 CONFIGFS_ATTR_RO(memb_group_, features);
524
525 static struct configfs_attribute *nullb_group_attrs[] = {
526 &memb_group_attr_features,
527 NULL,
528 };
529
530 static struct configfs_group_operations nullb_group_ops = {
531 .make_item = nullb_group_make_item,
532 .drop_item = nullb_group_drop_item,
533 };
534
535 static const struct config_item_type nullb_group_type = {
536 .ct_group_ops = &nullb_group_ops,
537 .ct_attrs = nullb_group_attrs,
538 .ct_owner = THIS_MODULE,
539 };
540
541 static struct configfs_subsystem nullb_subsys = {
542 .su_group = {
543 .cg_item = {
544 .ci_namebuf = "nullb",
545 .ci_type = &nullb_group_type,
546 },
547 },
548 };
549
550 static inline int null_cache_active(struct nullb *nullb)
551 {
552 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
553 }
554
555 static struct nullb_device *null_alloc_dev(void)
556 {
557 struct nullb_device *dev;
558
559 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
560 if (!dev)
561 return NULL;
562 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
563 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
564 if (badblocks_init(&dev->badblocks, 0)) {
565 kfree(dev);
566 return NULL;
567 }
568
569 dev->size = g_gb * 1024;
570 dev->completion_nsec = g_completion_nsec;
571 dev->submit_queues = g_submit_queues;
572 dev->home_node = g_home_node;
573 dev->queue_mode = g_queue_mode;
574 dev->blocksize = g_bs;
575 dev->irqmode = g_irqmode;
576 dev->hw_queue_depth = g_hw_queue_depth;
577 dev->use_lightnvm = g_use_lightnvm;
578 dev->blocking = g_blocking;
579 dev->use_per_node_hctx = g_use_per_node_hctx;
580 return dev;
581 }
582
583 static void null_free_dev(struct nullb_device *dev)
584 {
585 kfree(dev);
586 }
587
588 static void put_tag(struct nullb_queue *nq, unsigned int tag)
589 {
590 clear_bit_unlock(tag, nq->tag_map);
591
592 if (waitqueue_active(&nq->wait))
593 wake_up(&nq->wait);
594 }
595
596 static unsigned int get_tag(struct nullb_queue *nq)
597 {
598 unsigned int tag;
599
600 do {
601 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
602 if (tag >= nq->queue_depth)
603 return -1U;
604 } while (test_and_set_bit_lock(tag, nq->tag_map));
605
606 return tag;
607 }
608
609 static void free_cmd(struct nullb_cmd *cmd)
610 {
611 put_tag(cmd->nq, cmd->tag);
612 }
613
614 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
615
616 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
617 {
618 struct nullb_cmd *cmd;
619 unsigned int tag;
620
621 tag = get_tag(nq);
622 if (tag != -1U) {
623 cmd = &nq->cmds[tag];
624 cmd->tag = tag;
625 cmd->nq = nq;
626 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
627 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
628 HRTIMER_MODE_REL);
629 cmd->timer.function = null_cmd_timer_expired;
630 }
631 return cmd;
632 }
633
634 return NULL;
635 }
636
637 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
638 {
639 struct nullb_cmd *cmd;
640 DEFINE_WAIT(wait);
641
642 cmd = __alloc_cmd(nq);
643 if (cmd || !can_wait)
644 return cmd;
645
646 do {
647 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
648 cmd = __alloc_cmd(nq);
649 if (cmd)
650 break;
651
652 io_schedule();
653 } while (1);
654
655 finish_wait(&nq->wait, &wait);
656 return cmd;
657 }
658
659 static void end_cmd(struct nullb_cmd *cmd)
660 {
661 struct request_queue *q = NULL;
662 int queue_mode = cmd->nq->dev->queue_mode;
663
664 if (cmd->rq)
665 q = cmd->rq->q;
666
667 switch (queue_mode) {
668 case NULL_Q_MQ:
669 blk_mq_end_request(cmd->rq, cmd->error);
670 return;
671 case NULL_Q_RQ:
672 INIT_LIST_HEAD(&cmd->rq->queuelist);
673 blk_end_request_all(cmd->rq, cmd->error);
674 break;
675 case NULL_Q_BIO:
676 cmd->bio->bi_status = cmd->error;
677 bio_endio(cmd->bio);
678 break;
679 }
680
681 free_cmd(cmd);
682
683 /* Restart queue if needed, as we are freeing a tag */
684 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
685 unsigned long flags;
686
687 spin_lock_irqsave(q->queue_lock, flags);
688 blk_start_queue_async(q);
689 spin_unlock_irqrestore(q->queue_lock, flags);
690 }
691 }
692
693 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
694 {
695 end_cmd(container_of(timer, struct nullb_cmd, timer));
696
697 return HRTIMER_NORESTART;
698 }
699
700 static void null_cmd_end_timer(struct nullb_cmd *cmd)
701 {
702 ktime_t kt = cmd->nq->dev->completion_nsec;
703
704 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
705 }
706
707 static void null_softirq_done_fn(struct request *rq)
708 {
709 struct nullb *nullb = rq->q->queuedata;
710
711 if (nullb->dev->queue_mode == NULL_Q_MQ)
712 end_cmd(blk_mq_rq_to_pdu(rq));
713 else
714 end_cmd(rq->special);
715 }
716
717 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
718 {
719 struct nullb_page *t_page;
720
721 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
722 if (!t_page)
723 goto out;
724
725 t_page->page = alloc_pages(gfp_flags, 0);
726 if (!t_page->page)
727 goto out_freepage;
728
729 t_page->bitmap = 0;
730 return t_page;
731 out_freepage:
732 kfree(t_page);
733 out:
734 return NULL;
735 }
736
737 static void null_free_page(struct nullb_page *t_page)
738 {
739 __set_bit(NULLB_PAGE_FREE, &t_page->bitmap);
740 if (test_bit(NULLB_PAGE_LOCK, &t_page->bitmap))
741 return;
742 __free_page(t_page->page);
743 kfree(t_page);
744 }
745
746 static void null_free_sector(struct nullb *nullb, sector_t sector,
747 bool is_cache)
748 {
749 unsigned int sector_bit;
750 u64 idx;
751 struct nullb_page *t_page, *ret;
752 struct radix_tree_root *root;
753
754 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
755 idx = sector >> PAGE_SECTORS_SHIFT;
756 sector_bit = (sector & SECTOR_MASK);
757
758 t_page = radix_tree_lookup(root, idx);
759 if (t_page) {
760 __clear_bit(sector_bit, &t_page->bitmap);
761
762 if (!t_page->bitmap) {
763 ret = radix_tree_delete_item(root, idx, t_page);
764 WARN_ON(ret != t_page);
765 null_free_page(ret);
766 if (is_cache)
767 nullb->dev->curr_cache -= PAGE_SIZE;
768 }
769 }
770 }
771
772 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
773 struct nullb_page *t_page, bool is_cache)
774 {
775 struct radix_tree_root *root;
776
777 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
778
779 if (radix_tree_insert(root, idx, t_page)) {
780 null_free_page(t_page);
781 t_page = radix_tree_lookup(root, idx);
782 WARN_ON(!t_page || t_page->page->index != idx);
783 } else if (is_cache)
784 nullb->dev->curr_cache += PAGE_SIZE;
785
786 return t_page;
787 }
788
789 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
790 {
791 unsigned long pos = 0;
792 int nr_pages;
793 struct nullb_page *ret, *t_pages[FREE_BATCH];
794 struct radix_tree_root *root;
795
796 root = is_cache ? &dev->cache : &dev->data;
797
798 do {
799 int i;
800
801 nr_pages = radix_tree_gang_lookup(root,
802 (void **)t_pages, pos, FREE_BATCH);
803
804 for (i = 0; i < nr_pages; i++) {
805 pos = t_pages[i]->page->index;
806 ret = radix_tree_delete_item(root, pos, t_pages[i]);
807 WARN_ON(ret != t_pages[i]);
808 null_free_page(ret);
809 }
810
811 pos++;
812 } while (nr_pages == FREE_BATCH);
813
814 if (is_cache)
815 dev->curr_cache = 0;
816 }
817
818 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
819 sector_t sector, bool for_write, bool is_cache)
820 {
821 unsigned int sector_bit;
822 u64 idx;
823 struct nullb_page *t_page;
824 struct radix_tree_root *root;
825
826 idx = sector >> PAGE_SECTORS_SHIFT;
827 sector_bit = (sector & SECTOR_MASK);
828
829 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
830 t_page = radix_tree_lookup(root, idx);
831 WARN_ON(t_page && t_page->page->index != idx);
832
833 if (t_page && (for_write || test_bit(sector_bit, &t_page->bitmap)))
834 return t_page;
835
836 return NULL;
837 }
838
839 static struct nullb_page *null_lookup_page(struct nullb *nullb,
840 sector_t sector, bool for_write, bool ignore_cache)
841 {
842 struct nullb_page *page = NULL;
843
844 if (!ignore_cache)
845 page = __null_lookup_page(nullb, sector, for_write, true);
846 if (page)
847 return page;
848 return __null_lookup_page(nullb, sector, for_write, false);
849 }
850
851 static struct nullb_page *null_insert_page(struct nullb *nullb,
852 sector_t sector, bool ignore_cache)
853 {
854 u64 idx;
855 struct nullb_page *t_page;
856
857 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
858 if (t_page)
859 return t_page;
860
861 spin_unlock_irq(&nullb->lock);
862
863 t_page = null_alloc_page(GFP_NOIO);
864 if (!t_page)
865 goto out_lock;
866
867 if (radix_tree_preload(GFP_NOIO))
868 goto out_freepage;
869
870 spin_lock_irq(&nullb->lock);
871 idx = sector >> PAGE_SECTORS_SHIFT;
872 t_page->page->index = idx;
873 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
874 radix_tree_preload_end();
875
876 return t_page;
877 out_freepage:
878 null_free_page(t_page);
879 out_lock:
880 spin_lock_irq(&nullb->lock);
881 return null_lookup_page(nullb, sector, true, ignore_cache);
882 }
883
884 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
885 {
886 int i;
887 unsigned int offset;
888 u64 idx;
889 struct nullb_page *t_page, *ret;
890 void *dst, *src;
891
892 idx = c_page->page->index;
893
894 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
895
896 __clear_bit(NULLB_PAGE_LOCK, &c_page->bitmap);
897 if (test_bit(NULLB_PAGE_FREE, &c_page->bitmap)) {
898 null_free_page(c_page);
899 if (t_page && t_page->bitmap == 0) {
900 ret = radix_tree_delete_item(&nullb->dev->data,
901 idx, t_page);
902 null_free_page(t_page);
903 }
904 return 0;
905 }
906
907 if (!t_page)
908 return -ENOMEM;
909
910 src = kmap_atomic(c_page->page);
911 dst = kmap_atomic(t_page->page);
912
913 for (i = 0; i < PAGE_SECTORS;
914 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
915 if (test_bit(i, &c_page->bitmap)) {
916 offset = (i << SECTOR_SHIFT);
917 memcpy(dst + offset, src + offset,
918 nullb->dev->blocksize);
919 __set_bit(i, &t_page->bitmap);
920 }
921 }
922
923 kunmap_atomic(dst);
924 kunmap_atomic(src);
925
926 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
927 null_free_page(ret);
928 nullb->dev->curr_cache -= PAGE_SIZE;
929
930 return 0;
931 }
932
933 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
934 {
935 int i, err, nr_pages;
936 struct nullb_page *c_pages[FREE_BATCH];
937 unsigned long flushed = 0, one_round;
938
939 again:
940 if ((nullb->dev->cache_size * 1024 * 1024) >
941 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
942 return 0;
943
944 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
945 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
946 /*
947 * nullb_flush_cache_page could unlock before using the c_pages. To
948 * avoid race, we don't allow page free
949 */
950 for (i = 0; i < nr_pages; i++) {
951 nullb->cache_flush_pos = c_pages[i]->page->index;
952 /*
953 * We found the page which is being flushed to disk by other
954 * threads
955 */
956 if (test_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap))
957 c_pages[i] = NULL;
958 else
959 __set_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap);
960 }
961
962 one_round = 0;
963 for (i = 0; i < nr_pages; i++) {
964 if (c_pages[i] == NULL)
965 continue;
966 err = null_flush_cache_page(nullb, c_pages[i]);
967 if (err)
968 return err;
969 one_round++;
970 }
971 flushed += one_round << PAGE_SHIFT;
972
973 if (n > flushed) {
974 if (nr_pages == 0)
975 nullb->cache_flush_pos = 0;
976 if (one_round == 0) {
977 /* give other threads a chance */
978 spin_unlock_irq(&nullb->lock);
979 spin_lock_irq(&nullb->lock);
980 }
981 goto again;
982 }
983 return 0;
984 }
985
986 static int copy_to_nullb(struct nullb *nullb, struct page *source,
987 unsigned int off, sector_t sector, size_t n, bool is_fua)
988 {
989 size_t temp, count = 0;
990 unsigned int offset;
991 struct nullb_page *t_page;
992 void *dst, *src;
993
994 while (count < n) {
995 temp = min_t(size_t, nullb->dev->blocksize, n - count);
996
997 if (null_cache_active(nullb) && !is_fua)
998 null_make_cache_space(nullb, PAGE_SIZE);
999
1000 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1001 t_page = null_insert_page(nullb, sector,
1002 !null_cache_active(nullb) || is_fua);
1003 if (!t_page)
1004 return -ENOSPC;
1005
1006 src = kmap_atomic(source);
1007 dst = kmap_atomic(t_page->page);
1008 memcpy(dst + offset, src + off + count, temp);
1009 kunmap_atomic(dst);
1010 kunmap_atomic(src);
1011
1012 __set_bit(sector & SECTOR_MASK, &t_page->bitmap);
1013
1014 if (is_fua)
1015 null_free_sector(nullb, sector, true);
1016
1017 count += temp;
1018 sector += temp >> SECTOR_SHIFT;
1019 }
1020 return 0;
1021 }
1022
1023 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1024 unsigned int off, sector_t sector, size_t n)
1025 {
1026 size_t temp, count = 0;
1027 unsigned int offset;
1028 struct nullb_page *t_page;
1029 void *dst, *src;
1030
1031 while (count < n) {
1032 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1033
1034 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1035 t_page = null_lookup_page(nullb, sector, false,
1036 !null_cache_active(nullb));
1037
1038 dst = kmap_atomic(dest);
1039 if (!t_page) {
1040 memset(dst + off + count, 0, temp);
1041 goto next;
1042 }
1043 src = kmap_atomic(t_page->page);
1044 memcpy(dst + off + count, src + offset, temp);
1045 kunmap_atomic(src);
1046 next:
1047 kunmap_atomic(dst);
1048
1049 count += temp;
1050 sector += temp >> SECTOR_SHIFT;
1051 }
1052 return 0;
1053 }
1054
1055 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1056 {
1057 size_t temp;
1058
1059 spin_lock_irq(&nullb->lock);
1060 while (n > 0) {
1061 temp = min_t(size_t, n, nullb->dev->blocksize);
1062 null_free_sector(nullb, sector, false);
1063 if (null_cache_active(nullb))
1064 null_free_sector(nullb, sector, true);
1065 sector += temp >> SECTOR_SHIFT;
1066 n -= temp;
1067 }
1068 spin_unlock_irq(&nullb->lock);
1069 }
1070
1071 static int null_handle_flush(struct nullb *nullb)
1072 {
1073 int err;
1074
1075 if (!null_cache_active(nullb))
1076 return 0;
1077
1078 spin_lock_irq(&nullb->lock);
1079 while (true) {
1080 err = null_make_cache_space(nullb,
1081 nullb->dev->cache_size * 1024 * 1024);
1082 if (err || nullb->dev->curr_cache == 0)
1083 break;
1084 }
1085
1086 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1087 spin_unlock_irq(&nullb->lock);
1088 return err;
1089 }
1090
1091 static int null_transfer(struct nullb *nullb, struct page *page,
1092 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1093 bool is_fua)
1094 {
1095 int err = 0;
1096
1097 if (!is_write) {
1098 err = copy_from_nullb(nullb, page, off, sector, len);
1099 flush_dcache_page(page);
1100 } else {
1101 flush_dcache_page(page);
1102 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1103 }
1104
1105 return err;
1106 }
1107
1108 static int null_handle_rq(struct nullb_cmd *cmd)
1109 {
1110 struct request *rq = cmd->rq;
1111 struct nullb *nullb = cmd->nq->dev->nullb;
1112 int err;
1113 unsigned int len;
1114 sector_t sector;
1115 struct req_iterator iter;
1116 struct bio_vec bvec;
1117
1118 sector = blk_rq_pos(rq);
1119
1120 if (req_op(rq) == REQ_OP_DISCARD) {
1121 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1122 return 0;
1123 }
1124
1125 spin_lock_irq(&nullb->lock);
1126 rq_for_each_segment(bvec, rq, iter) {
1127 len = bvec.bv_len;
1128 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1129 op_is_write(req_op(rq)), sector,
1130 req_op(rq) & REQ_FUA);
1131 if (err) {
1132 spin_unlock_irq(&nullb->lock);
1133 return err;
1134 }
1135 sector += len >> SECTOR_SHIFT;
1136 }
1137 spin_unlock_irq(&nullb->lock);
1138
1139 return 0;
1140 }
1141
1142 static int null_handle_bio(struct nullb_cmd *cmd)
1143 {
1144 struct bio *bio = cmd->bio;
1145 struct nullb *nullb = cmd->nq->dev->nullb;
1146 int err;
1147 unsigned int len;
1148 sector_t sector;
1149 struct bio_vec bvec;
1150 struct bvec_iter iter;
1151
1152 sector = bio->bi_iter.bi_sector;
1153
1154 if (bio_op(bio) == REQ_OP_DISCARD) {
1155 null_handle_discard(nullb, sector,
1156 bio_sectors(bio) << SECTOR_SHIFT);
1157 return 0;
1158 }
1159
1160 spin_lock_irq(&nullb->lock);
1161 bio_for_each_segment(bvec, bio, iter) {
1162 len = bvec.bv_len;
1163 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1164 op_is_write(bio_op(bio)), sector,
1165 bio_op(bio) & REQ_FUA);
1166 if (err) {
1167 spin_unlock_irq(&nullb->lock);
1168 return err;
1169 }
1170 sector += len >> SECTOR_SHIFT;
1171 }
1172 spin_unlock_irq(&nullb->lock);
1173 return 0;
1174 }
1175
1176 static void null_stop_queue(struct nullb *nullb)
1177 {
1178 struct request_queue *q = nullb->q;
1179
1180 if (nullb->dev->queue_mode == NULL_Q_MQ)
1181 blk_mq_stop_hw_queues(q);
1182 else {
1183 spin_lock_irq(q->queue_lock);
1184 blk_stop_queue(q);
1185 spin_unlock_irq(q->queue_lock);
1186 }
1187 }
1188
1189 static void null_restart_queue_async(struct nullb *nullb)
1190 {
1191 struct request_queue *q = nullb->q;
1192 unsigned long flags;
1193
1194 if (nullb->dev->queue_mode == NULL_Q_MQ)
1195 blk_mq_start_stopped_hw_queues(q, true);
1196 else {
1197 spin_lock_irqsave(q->queue_lock, flags);
1198 blk_start_queue_async(q);
1199 spin_unlock_irqrestore(q->queue_lock, flags);
1200 }
1201 }
1202
1203 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd)
1204 {
1205 struct nullb_device *dev = cmd->nq->dev;
1206 struct nullb *nullb = dev->nullb;
1207 int err = 0;
1208
1209 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1210 struct request *rq = cmd->rq;
1211
1212 if (!hrtimer_active(&nullb->bw_timer))
1213 hrtimer_restart(&nullb->bw_timer);
1214
1215 if (atomic_long_sub_return(blk_rq_bytes(rq),
1216 &nullb->cur_bytes) < 0) {
1217 null_stop_queue(nullb);
1218 /* race with timer */
1219 if (atomic_long_read(&nullb->cur_bytes) > 0)
1220 null_restart_queue_async(nullb);
1221 if (dev->queue_mode == NULL_Q_RQ) {
1222 struct request_queue *q = nullb->q;
1223
1224 spin_lock_irq(q->queue_lock);
1225 rq->rq_flags |= RQF_DONTPREP;
1226 blk_requeue_request(q, rq);
1227 spin_unlock_irq(q->queue_lock);
1228 return BLK_STS_OK;
1229 } else
1230 /* requeue request */
1231 return BLK_STS_RESOURCE;
1232 }
1233 }
1234
1235 if (nullb->dev->badblocks.shift != -1) {
1236 int bad_sectors;
1237 sector_t sector, size, first_bad;
1238 bool is_flush = true;
1239
1240 if (dev->queue_mode == NULL_Q_BIO &&
1241 bio_op(cmd->bio) != REQ_OP_FLUSH) {
1242 is_flush = false;
1243 sector = cmd->bio->bi_iter.bi_sector;
1244 size = bio_sectors(cmd->bio);
1245 }
1246 if (dev->queue_mode != NULL_Q_BIO &&
1247 req_op(cmd->rq) != REQ_OP_FLUSH) {
1248 is_flush = false;
1249 sector = blk_rq_pos(cmd->rq);
1250 size = blk_rq_sectors(cmd->rq);
1251 }
1252 if (!is_flush && badblocks_check(&nullb->dev->badblocks, sector,
1253 size, &first_bad, &bad_sectors)) {
1254 cmd->error = BLK_STS_IOERR;
1255 goto out;
1256 }
1257 }
1258
1259 if (dev->memory_backed) {
1260 if (dev->queue_mode == NULL_Q_BIO) {
1261 if (bio_op(cmd->bio) == REQ_OP_FLUSH)
1262 err = null_handle_flush(nullb);
1263 else
1264 err = null_handle_bio(cmd);
1265 } else {
1266 if (req_op(cmd->rq) == REQ_OP_FLUSH)
1267 err = null_handle_flush(nullb);
1268 else
1269 err = null_handle_rq(cmd);
1270 }
1271 }
1272 cmd->error = errno_to_blk_status(err);
1273 out:
1274 /* Complete IO by inline, softirq or timer */
1275 switch (dev->irqmode) {
1276 case NULL_IRQ_SOFTIRQ:
1277 switch (dev->queue_mode) {
1278 case NULL_Q_MQ:
1279 blk_mq_complete_request(cmd->rq);
1280 break;
1281 case NULL_Q_RQ:
1282 blk_complete_request(cmd->rq);
1283 break;
1284 case NULL_Q_BIO:
1285 /*
1286 * XXX: no proper submitting cpu information available.
1287 */
1288 end_cmd(cmd);
1289 break;
1290 }
1291 break;
1292 case NULL_IRQ_NONE:
1293 end_cmd(cmd);
1294 break;
1295 case NULL_IRQ_TIMER:
1296 null_cmd_end_timer(cmd);
1297 break;
1298 }
1299 return BLK_STS_OK;
1300 }
1301
1302 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1303 {
1304 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1305 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1306 unsigned int mbps = nullb->dev->mbps;
1307
1308 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1309 return HRTIMER_NORESTART;
1310
1311 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1312 null_restart_queue_async(nullb);
1313
1314 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1315
1316 return HRTIMER_RESTART;
1317 }
1318
1319 static void nullb_setup_bwtimer(struct nullb *nullb)
1320 {
1321 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1322
1323 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1324 nullb->bw_timer.function = nullb_bwtimer_fn;
1325 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1326 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1327 }
1328
1329 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1330 {
1331 int index = 0;
1332
1333 if (nullb->nr_queues != 1)
1334 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1335
1336 return &nullb->queues[index];
1337 }
1338
1339 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1340 {
1341 struct nullb *nullb = q->queuedata;
1342 struct nullb_queue *nq = nullb_to_queue(nullb);
1343 struct nullb_cmd *cmd;
1344
1345 cmd = alloc_cmd(nq, 1);
1346 cmd->bio = bio;
1347
1348 null_handle_cmd(cmd);
1349 return BLK_QC_T_NONE;
1350 }
1351
1352 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
1353 {
1354 struct nullb *nullb = q->queuedata;
1355 struct nullb_queue *nq = nullb_to_queue(nullb);
1356 struct nullb_cmd *cmd;
1357
1358 cmd = alloc_cmd(nq, 0);
1359 if (cmd) {
1360 cmd->rq = req;
1361 req->special = cmd;
1362 return BLKPREP_OK;
1363 }
1364 blk_stop_queue(q);
1365
1366 return BLKPREP_DEFER;
1367 }
1368
1369 static void null_request_fn(struct request_queue *q)
1370 {
1371 struct request *rq;
1372
1373 while ((rq = blk_fetch_request(q)) != NULL) {
1374 struct nullb_cmd *cmd = rq->special;
1375
1376 spin_unlock_irq(q->queue_lock);
1377 null_handle_cmd(cmd);
1378 spin_lock_irq(q->queue_lock);
1379 }
1380 }
1381
1382 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1383 const struct blk_mq_queue_data *bd)
1384 {
1385 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1386 struct nullb_queue *nq = hctx->driver_data;
1387
1388 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1389
1390 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1391 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1392 cmd->timer.function = null_cmd_timer_expired;
1393 }
1394 cmd->rq = bd->rq;
1395 cmd->nq = nq;
1396
1397 blk_mq_start_request(bd->rq);
1398
1399 return null_handle_cmd(cmd);
1400 }
1401
1402 static const struct blk_mq_ops null_mq_ops = {
1403 .queue_rq = null_queue_rq,
1404 .complete = null_softirq_done_fn,
1405 };
1406
1407 static void cleanup_queue(struct nullb_queue *nq)
1408 {
1409 kfree(nq->tag_map);
1410 kfree(nq->cmds);
1411 }
1412
1413 static void cleanup_queues(struct nullb *nullb)
1414 {
1415 int i;
1416
1417 for (i = 0; i < nullb->nr_queues; i++)
1418 cleanup_queue(&nullb->queues[i]);
1419
1420 kfree(nullb->queues);
1421 }
1422
1423 #ifdef CONFIG_NVM
1424
1425 static void null_lnvm_end_io(struct request *rq, blk_status_t status)
1426 {
1427 struct nvm_rq *rqd = rq->end_io_data;
1428
1429 /* XXX: lighnvm core seems to expect NVM_RSP_* values here.. */
1430 rqd->error = status ? -EIO : 0;
1431 nvm_end_io(rqd);
1432
1433 blk_put_request(rq);
1434 }
1435
1436 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
1437 {
1438 struct request_queue *q = dev->q;
1439 struct request *rq;
1440 struct bio *bio = rqd->bio;
1441
1442 rq = blk_mq_alloc_request(q,
1443 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1444 if (IS_ERR(rq))
1445 return -ENOMEM;
1446
1447 blk_init_request_from_bio(rq, bio);
1448
1449 rq->end_io_data = rqd;
1450
1451 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
1452
1453 return 0;
1454 }
1455
1456 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
1457 {
1458 struct nullb *nullb = dev->q->queuedata;
1459 sector_t size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1460 sector_t blksize;
1461 struct nvm_id_group *grp;
1462
1463 id->ver_id = 0x1;
1464 id->vmnt = 0;
1465 id->cap = 0x2;
1466 id->dom = 0x1;
1467
1468 id->ppaf.blk_offset = 0;
1469 id->ppaf.blk_len = 16;
1470 id->ppaf.pg_offset = 16;
1471 id->ppaf.pg_len = 16;
1472 id->ppaf.sect_offset = 32;
1473 id->ppaf.sect_len = 8;
1474 id->ppaf.pln_offset = 40;
1475 id->ppaf.pln_len = 8;
1476 id->ppaf.lun_offset = 48;
1477 id->ppaf.lun_len = 8;
1478 id->ppaf.ch_offset = 56;
1479 id->ppaf.ch_len = 8;
1480
1481 sector_div(size, nullb->dev->blocksize); /* convert size to pages */
1482 size >>= 8; /* concert size to pgs pr blk */
1483 grp = &id->grp;
1484 grp->mtype = 0;
1485 grp->fmtype = 0;
1486 grp->num_ch = 1;
1487 grp->num_pg = 256;
1488 blksize = size;
1489 size >>= 16;
1490 grp->num_lun = size + 1;
1491 sector_div(blksize, grp->num_lun);
1492 grp->num_blk = blksize;
1493 grp->num_pln = 1;
1494
1495 grp->fpg_sz = nullb->dev->blocksize;
1496 grp->csecs = nullb->dev->blocksize;
1497 grp->trdt = 25000;
1498 grp->trdm = 25000;
1499 grp->tprt = 500000;
1500 grp->tprm = 500000;
1501 grp->tbet = 1500000;
1502 grp->tbem = 1500000;
1503 grp->mpos = 0x010101; /* single plane rwe */
1504 grp->cpar = nullb->dev->hw_queue_depth;
1505
1506 return 0;
1507 }
1508
1509 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
1510 {
1511 mempool_t *virtmem_pool;
1512
1513 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
1514 if (!virtmem_pool) {
1515 pr_err("null_blk: Unable to create virtual memory pool\n");
1516 return NULL;
1517 }
1518
1519 return virtmem_pool;
1520 }
1521
1522 static void null_lnvm_destroy_dma_pool(void *pool)
1523 {
1524 mempool_destroy(pool);
1525 }
1526
1527 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
1528 gfp_t mem_flags, dma_addr_t *dma_handler)
1529 {
1530 return mempool_alloc(pool, mem_flags);
1531 }
1532
1533 static void null_lnvm_dev_dma_free(void *pool, void *entry,
1534 dma_addr_t dma_handler)
1535 {
1536 mempool_free(entry, pool);
1537 }
1538
1539 static struct nvm_dev_ops null_lnvm_dev_ops = {
1540 .identity = null_lnvm_id,
1541 .submit_io = null_lnvm_submit_io,
1542
1543 .create_dma_pool = null_lnvm_create_dma_pool,
1544 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
1545 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
1546 .dev_dma_free = null_lnvm_dev_dma_free,
1547
1548 /* Simulate nvme protocol restriction */
1549 .max_phys_sect = 64,
1550 };
1551
1552 static int null_nvm_register(struct nullb *nullb)
1553 {
1554 struct nvm_dev *dev;
1555 int rv;
1556
1557 dev = nvm_alloc_dev(0);
1558 if (!dev)
1559 return -ENOMEM;
1560
1561 dev->q = nullb->q;
1562 memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
1563 dev->ops = &null_lnvm_dev_ops;
1564
1565 rv = nvm_register(dev);
1566 if (rv) {
1567 kfree(dev);
1568 return rv;
1569 }
1570 nullb->ndev = dev;
1571 return 0;
1572 }
1573
1574 static void null_nvm_unregister(struct nullb *nullb)
1575 {
1576 nvm_unregister(nullb->ndev);
1577 }
1578 #else
1579 static int null_nvm_register(struct nullb *nullb)
1580 {
1581 pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
1582 return -EINVAL;
1583 }
1584 static void null_nvm_unregister(struct nullb *nullb) {}
1585 #endif /* CONFIG_NVM */
1586
1587 static void null_del_dev(struct nullb *nullb)
1588 {
1589 struct nullb_device *dev = nullb->dev;
1590
1591 ida_simple_remove(&nullb_indexes, nullb->index);
1592
1593 list_del_init(&nullb->list);
1594
1595 if (dev->use_lightnvm)
1596 null_nvm_unregister(nullb);
1597 else
1598 del_gendisk(nullb->disk);
1599
1600 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1601 hrtimer_cancel(&nullb->bw_timer);
1602 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1603 null_restart_queue_async(nullb);
1604 }
1605
1606 blk_cleanup_queue(nullb->q);
1607 if (dev->queue_mode == NULL_Q_MQ &&
1608 nullb->tag_set == &nullb->__tag_set)
1609 blk_mq_free_tag_set(nullb->tag_set);
1610 if (!dev->use_lightnvm)
1611 put_disk(nullb->disk);
1612 cleanup_queues(nullb);
1613 if (null_cache_active(nullb))
1614 null_free_device_storage(nullb->dev, true);
1615 kfree(nullb);
1616 dev->nullb = NULL;
1617 }
1618
1619 static void null_config_discard(struct nullb *nullb)
1620 {
1621 if (nullb->dev->discard == false)
1622 return;
1623 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1624 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1625 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1626 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nullb->q);
1627 }
1628
1629 static int null_open(struct block_device *bdev, fmode_t mode)
1630 {
1631 return 0;
1632 }
1633
1634 static void null_release(struct gendisk *disk, fmode_t mode)
1635 {
1636 }
1637
1638 static const struct block_device_operations null_fops = {
1639 .owner = THIS_MODULE,
1640 .open = null_open,
1641 .release = null_release,
1642 };
1643
1644 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1645 {
1646 BUG_ON(!nullb);
1647 BUG_ON(!nq);
1648
1649 init_waitqueue_head(&nq->wait);
1650 nq->queue_depth = nullb->queue_depth;
1651 nq->dev = nullb->dev;
1652 }
1653
1654 static void null_init_queues(struct nullb *nullb)
1655 {
1656 struct request_queue *q = nullb->q;
1657 struct blk_mq_hw_ctx *hctx;
1658 struct nullb_queue *nq;
1659 int i;
1660
1661 queue_for_each_hw_ctx(q, hctx, i) {
1662 if (!hctx->nr_ctx || !hctx->tags)
1663 continue;
1664 nq = &nullb->queues[i];
1665 hctx->driver_data = nq;
1666 null_init_queue(nullb, nq);
1667 nullb->nr_queues++;
1668 }
1669 }
1670
1671 static int setup_commands(struct nullb_queue *nq)
1672 {
1673 struct nullb_cmd *cmd;
1674 int i, tag_size;
1675
1676 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
1677 if (!nq->cmds)
1678 return -ENOMEM;
1679
1680 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1681 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
1682 if (!nq->tag_map) {
1683 kfree(nq->cmds);
1684 return -ENOMEM;
1685 }
1686
1687 for (i = 0; i < nq->queue_depth; i++) {
1688 cmd = &nq->cmds[i];
1689 INIT_LIST_HEAD(&cmd->list);
1690 cmd->ll_list.next = NULL;
1691 cmd->tag = -1U;
1692 }
1693
1694 return 0;
1695 }
1696
1697 static int setup_queues(struct nullb *nullb)
1698 {
1699 nullb->queues = kzalloc(nullb->dev->submit_queues *
1700 sizeof(struct nullb_queue), GFP_KERNEL);
1701 if (!nullb->queues)
1702 return -ENOMEM;
1703
1704 nullb->nr_queues = 0;
1705 nullb->queue_depth = nullb->dev->hw_queue_depth;
1706
1707 return 0;
1708 }
1709
1710 static int init_driver_queues(struct nullb *nullb)
1711 {
1712 struct nullb_queue *nq;
1713 int i, ret = 0;
1714
1715 for (i = 0; i < nullb->dev->submit_queues; i++) {
1716 nq = &nullb->queues[i];
1717
1718 null_init_queue(nullb, nq);
1719
1720 ret = setup_commands(nq);
1721 if (ret)
1722 return ret;
1723 nullb->nr_queues++;
1724 }
1725 return 0;
1726 }
1727
1728 static int null_gendisk_register(struct nullb *nullb)
1729 {
1730 struct gendisk *disk;
1731 sector_t size;
1732
1733 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1734 if (!disk)
1735 return -ENOMEM;
1736 size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1737 set_capacity(disk, size >> 9);
1738
1739 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1740 disk->major = null_major;
1741 disk->first_minor = nullb->index;
1742 disk->fops = &null_fops;
1743 disk->private_data = nullb;
1744 disk->queue = nullb->q;
1745 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1746
1747 add_disk(disk);
1748 return 0;
1749 }
1750
1751 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1752 {
1753 set->ops = &null_mq_ops;
1754 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1755 g_submit_queues;
1756 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1757 g_hw_queue_depth;
1758 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1759 set->cmd_size = sizeof(struct nullb_cmd);
1760 set->flags = BLK_MQ_F_SHOULD_MERGE;
1761 if (g_no_sched)
1762 set->flags |= BLK_MQ_F_NO_SCHED;
1763 set->driver_data = NULL;
1764
1765 if ((nullb && nullb->dev->blocking) || g_blocking)
1766 set->flags |= BLK_MQ_F_BLOCKING;
1767
1768 return blk_mq_alloc_tag_set(set);
1769 }
1770
1771 static void null_validate_conf(struct nullb_device *dev)
1772 {
1773 dev->blocksize = round_down(dev->blocksize, 512);
1774 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1775 if (dev->use_lightnvm && dev->blocksize != 4096)
1776 dev->blocksize = 4096;
1777
1778 if (dev->use_lightnvm && dev->queue_mode != NULL_Q_MQ)
1779 dev->queue_mode = NULL_Q_MQ;
1780
1781 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1782 if (dev->submit_queues != nr_online_nodes)
1783 dev->submit_queues = nr_online_nodes;
1784 } else if (dev->submit_queues > nr_cpu_ids)
1785 dev->submit_queues = nr_cpu_ids;
1786 else if (dev->submit_queues == 0)
1787 dev->submit_queues = 1;
1788
1789 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1790 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1791
1792 /* Do memory allocation, so set blocking */
1793 if (dev->memory_backed)
1794 dev->blocking = true;
1795 else /* cache is meaningless */
1796 dev->cache_size = 0;
1797 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1798 dev->cache_size);
1799 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1800 /* can not stop a queue */
1801 if (dev->queue_mode == NULL_Q_BIO)
1802 dev->mbps = 0;
1803 }
1804
1805 static int null_add_dev(struct nullb_device *dev)
1806 {
1807 struct nullb *nullb;
1808 int rv;
1809
1810 null_validate_conf(dev);
1811
1812 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1813 if (!nullb) {
1814 rv = -ENOMEM;
1815 goto out;
1816 }
1817 nullb->dev = dev;
1818 dev->nullb = nullb;
1819
1820 spin_lock_init(&nullb->lock);
1821
1822 rv = setup_queues(nullb);
1823 if (rv)
1824 goto out_free_nullb;
1825
1826 if (dev->queue_mode == NULL_Q_MQ) {
1827 if (shared_tags) {
1828 nullb->tag_set = &tag_set;
1829 rv = 0;
1830 } else {
1831 nullb->tag_set = &nullb->__tag_set;
1832 rv = null_init_tag_set(nullb, nullb->tag_set);
1833 }
1834
1835 if (rv)
1836 goto out_cleanup_queues;
1837
1838 nullb->q = blk_mq_init_queue(nullb->tag_set);
1839 if (IS_ERR(nullb->q)) {
1840 rv = -ENOMEM;
1841 goto out_cleanup_tags;
1842 }
1843 null_init_queues(nullb);
1844 } else if (dev->queue_mode == NULL_Q_BIO) {
1845 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node);
1846 if (!nullb->q) {
1847 rv = -ENOMEM;
1848 goto out_cleanup_queues;
1849 }
1850 blk_queue_make_request(nullb->q, null_queue_bio);
1851 rv = init_driver_queues(nullb);
1852 if (rv)
1853 goto out_cleanup_blk_queue;
1854 } else {
1855 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock,
1856 dev->home_node);
1857 if (!nullb->q) {
1858 rv = -ENOMEM;
1859 goto out_cleanup_queues;
1860 }
1861 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
1862 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
1863 rv = init_driver_queues(nullb);
1864 if (rv)
1865 goto out_cleanup_blk_queue;
1866 }
1867
1868 if (dev->mbps) {
1869 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1870 nullb_setup_bwtimer(nullb);
1871 }
1872
1873 if (dev->cache_size > 0) {
1874 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1875 blk_queue_write_cache(nullb->q, true, true);
1876 blk_queue_flush_queueable(nullb->q, true);
1877 }
1878
1879 nullb->q->queuedata = nullb;
1880 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
1881 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1882
1883 mutex_lock(&lock);
1884 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1885 dev->index = nullb->index;
1886 mutex_unlock(&lock);
1887
1888 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1889 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1890
1891 null_config_discard(nullb);
1892
1893 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1894
1895 if (dev->use_lightnvm)
1896 rv = null_nvm_register(nullb);
1897 else
1898 rv = null_gendisk_register(nullb);
1899
1900 if (rv)
1901 goto out_cleanup_blk_queue;
1902
1903 mutex_lock(&lock);
1904 list_add_tail(&nullb->list, &nullb_list);
1905 mutex_unlock(&lock);
1906
1907 return 0;
1908 out_cleanup_blk_queue:
1909 blk_cleanup_queue(nullb->q);
1910 out_cleanup_tags:
1911 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1912 blk_mq_free_tag_set(nullb->tag_set);
1913 out_cleanup_queues:
1914 cleanup_queues(nullb);
1915 out_free_nullb:
1916 kfree(nullb);
1917 out:
1918 return rv;
1919 }
1920
1921 static int __init null_init(void)
1922 {
1923 int ret = 0;
1924 unsigned int i;
1925 struct nullb *nullb;
1926 struct nullb_device *dev;
1927
1928 /* check for nullb_page.bitmap */
1929 if (sizeof(unsigned long) * 8 - 2 < (PAGE_SIZE >> SECTOR_SHIFT))
1930 return -EINVAL;
1931
1932 if (g_bs > PAGE_SIZE) {
1933 pr_warn("null_blk: invalid block size\n");
1934 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
1935 g_bs = PAGE_SIZE;
1936 }
1937
1938 if (g_use_lightnvm && g_bs != 4096) {
1939 pr_warn("null_blk: LightNVM only supports 4k block size\n");
1940 pr_warn("null_blk: defaults block size to 4k\n");
1941 g_bs = 4096;
1942 }
1943
1944 if (g_use_lightnvm && g_queue_mode != NULL_Q_MQ) {
1945 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
1946 pr_warn("null_blk: defaults queue mode to blk-mq\n");
1947 g_queue_mode = NULL_Q_MQ;
1948 }
1949
1950 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1951 if (g_submit_queues != nr_online_nodes) {
1952 pr_warn("null_blk: submit_queues param is set to %u.\n",
1953 nr_online_nodes);
1954 g_submit_queues = nr_online_nodes;
1955 }
1956 } else if (g_submit_queues > nr_cpu_ids)
1957 g_submit_queues = nr_cpu_ids;
1958 else if (g_submit_queues <= 0)
1959 g_submit_queues = 1;
1960
1961 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1962 ret = null_init_tag_set(NULL, &tag_set);
1963 if (ret)
1964 return ret;
1965 }
1966
1967 config_group_init(&nullb_subsys.su_group);
1968 mutex_init(&nullb_subsys.su_mutex);
1969
1970 ret = configfs_register_subsystem(&nullb_subsys);
1971 if (ret)
1972 goto err_tagset;
1973
1974 mutex_init(&lock);
1975
1976 null_major = register_blkdev(0, "nullb");
1977 if (null_major < 0) {
1978 ret = null_major;
1979 goto err_conf;
1980 }
1981
1982 if (g_use_lightnvm) {
1983 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
1984 0, 0, NULL);
1985 if (!ppa_cache) {
1986 pr_err("null_blk: unable to create ppa cache\n");
1987 ret = -ENOMEM;
1988 goto err_ppa;
1989 }
1990 }
1991
1992 for (i = 0; i < nr_devices; i++) {
1993 dev = null_alloc_dev();
1994 if (!dev) {
1995 ret = -ENOMEM;
1996 goto err_dev;
1997 }
1998 ret = null_add_dev(dev);
1999 if (ret) {
2000 null_free_dev(dev);
2001 goto err_dev;
2002 }
2003 }
2004
2005 pr_info("null: module loaded\n");
2006 return 0;
2007
2008 err_dev:
2009 while (!list_empty(&nullb_list)) {
2010 nullb = list_entry(nullb_list.next, struct nullb, list);
2011 dev = nullb->dev;
2012 null_del_dev(nullb);
2013 null_free_dev(dev);
2014 }
2015 kmem_cache_destroy(ppa_cache);
2016 err_ppa:
2017 unregister_blkdev(null_major, "nullb");
2018 err_conf:
2019 configfs_unregister_subsystem(&nullb_subsys);
2020 err_tagset:
2021 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2022 blk_mq_free_tag_set(&tag_set);
2023 return ret;
2024 }
2025
2026 static void __exit null_exit(void)
2027 {
2028 struct nullb *nullb;
2029
2030 configfs_unregister_subsystem(&nullb_subsys);
2031
2032 unregister_blkdev(null_major, "nullb");
2033
2034 mutex_lock(&lock);
2035 while (!list_empty(&nullb_list)) {
2036 struct nullb_device *dev;
2037
2038 nullb = list_entry(nullb_list.next, struct nullb, list);
2039 dev = nullb->dev;
2040 null_del_dev(nullb);
2041 null_free_dev(dev);
2042 }
2043 mutex_unlock(&lock);
2044
2045 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2046 blk_mq_free_tag_set(&tag_set);
2047
2048 kmem_cache_destroy(ppa_cache);
2049 }
2050
2051 module_init(null_init);
2052 module_exit(null_exit);
2053
2054 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2055 MODULE_LICENSE("GPL");
ubuntu-bionic-kernel mirror