1 // SPDX-License-Identifier: GPL-2.0-only
3 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
4 * Shaohua Li <shli@fb.com>
6 #include <linux/module.h>
8 #include <linux/moduleparam.h>
9 #include <linux/sched.h>
11 #include <linux/init.h>
14 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
15 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
16 #define SECTOR_MASK (PAGE_SECTORS - 1)
20 #define TICKS_PER_SEC 50ULL
21 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
23 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
24 static DECLARE_FAULT_ATTR(null_timeout_attr);
25 static DECLARE_FAULT_ATTR(null_requeue_attr);
28 static inline u64 mb_per_tick(int mbps)
30 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
34 * Status flags for nullb_device.
36 * CONFIGURED: Device has been configured and turned on. Cannot reconfigure.
37 * UP: Device is currently on and visible in userspace.
38 * THROTTLED: Device is being throttled.
39 * CACHE: Device is using a write-back cache.
41 enum nullb_device_flags {
42 NULLB_DEV_FL_CONFIGURED = 0,
44 NULLB_DEV_FL_THROTTLED = 2,
45 NULLB_DEV_FL_CACHE = 3,
48 #define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
50 * nullb_page is a page in memory for nullb devices.
52 * @page: The page holding the data.
53 * @bitmap: The bitmap represents which sector in the page has data.
54 * Each bit represents one block size. For example, sector 8
55 * will use the 7th bit
56 * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
57 * page is being flushing to storage. FREE means the cache page is freed and
58 * should be skipped from flushing to storage. Please see
59 * null_make_cache_space
63 DECLARE_BITMAP(bitmap, MAP_SZ);
65 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
66 #define NULLB_PAGE_FREE (MAP_SZ - 2)
68 static LIST_HEAD(nullb_list);
69 static struct mutex lock;
70 static int null_major;
71 static DEFINE_IDA(nullb_indexes);
72 static struct blk_mq_tag_set tag_set;
86 static int g_no_sched;
87 module_param_named(no_sched, g_no_sched, int, 0444);
88 MODULE_PARM_DESC(no_sched, "No io scheduler");
90 static int g_submit_queues = 1;
91 module_param_named(submit_queues, g_submit_queues, int, 0444);
92 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
94 static int g_home_node = NUMA_NO_NODE;
95 module_param_named(home_node, g_home_node, int, 0444);
96 MODULE_PARM_DESC(home_node, "Home node for the device");
98 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
99 static char g_timeout_str[80];
100 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
102 static char g_requeue_str[80];
103 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
106 static int g_queue_mode = NULL_Q_MQ;
108 static int null_param_store_val(const char *str, int *val, int min, int max)
112 ret = kstrtoint(str, 10, &new_val);
116 if (new_val < min || new_val > max)
123 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
125 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
128 static const struct kernel_param_ops null_queue_mode_param_ops = {
129 .set = null_set_queue_mode,
130 .get = param_get_int,
133 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
134 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
136 static int g_gb = 250;
137 module_param_named(gb, g_gb, int, 0444);
138 MODULE_PARM_DESC(gb, "Size in GB");
140 static int g_bs = 512;
141 module_param_named(bs, g_bs, int, 0444);
142 MODULE_PARM_DESC(bs, "Block size (in bytes)");
144 static unsigned int nr_devices = 1;
145 module_param(nr_devices, uint, 0444);
146 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
148 static bool g_blocking;
149 module_param_named(blocking, g_blocking, bool, 0444);
150 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
152 static bool shared_tags;
153 module_param(shared_tags, bool, 0444);
154 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
156 static int g_irqmode = NULL_IRQ_SOFTIRQ;
158 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
160 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
164 static const struct kernel_param_ops null_irqmode_param_ops = {
165 .set = null_set_irqmode,
166 .get = param_get_int,
169 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
170 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
172 static unsigned long g_completion_nsec = 10000;
173 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
174 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
176 static int g_hw_queue_depth = 64;
177 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
178 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
180 static bool g_use_per_node_hctx;
181 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
182 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
185 module_param_named(zoned, g_zoned, bool, S_IRUGO);
186 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
188 static unsigned long g_zone_size = 256;
189 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
190 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
192 static unsigned int g_zone_nr_conv;
193 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
194 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
196 static struct nullb_device *null_alloc_dev(void);
197 static void null_free_dev(struct nullb_device *dev);
198 static void null_del_dev(struct nullb *nullb);
199 static int null_add_dev(struct nullb_device *dev);
200 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
202 static inline struct nullb_device *to_nullb_device(struct config_item *item)
204 return item ? container_of(item, struct nullb_device, item) : NULL;
207 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
209 return snprintf(page, PAGE_SIZE, "%u\n", val);
212 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
215 return snprintf(page, PAGE_SIZE, "%lu\n", val);
218 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
220 return snprintf(page, PAGE_SIZE, "%u\n", val);
223 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
224 const char *page, size_t count)
229 result = kstrtouint(page, 0, &tmp);
237 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
238 const char *page, size_t count)
243 result = kstrtoul(page, 0, &tmp);
251 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
257 result = kstrtobool(page, &tmp);
265 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
266 #define NULLB_DEVICE_ATTR(NAME, TYPE) \
268 nullb_device_##NAME##_show(struct config_item *item, char *page) \
270 return nullb_device_##TYPE##_attr_show( \
271 to_nullb_device(item)->NAME, page); \
274 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
277 if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags)) \
279 return nullb_device_##TYPE##_attr_store( \
280 &to_nullb_device(item)->NAME, page, count); \
282 CONFIGFS_ATTR(nullb_device_, NAME);
284 NULLB_DEVICE_ATTR(size, ulong);
285 NULLB_DEVICE_ATTR(completion_nsec, ulong);
286 NULLB_DEVICE_ATTR(submit_queues, uint);
287 NULLB_DEVICE_ATTR(home_node, uint);
288 NULLB_DEVICE_ATTR(queue_mode, uint);
289 NULLB_DEVICE_ATTR(blocksize, uint);
290 NULLB_DEVICE_ATTR(irqmode, uint);
291 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
292 NULLB_DEVICE_ATTR(index, uint);
293 NULLB_DEVICE_ATTR(blocking, bool);
294 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
295 NULLB_DEVICE_ATTR(memory_backed, bool);
296 NULLB_DEVICE_ATTR(discard, bool);
297 NULLB_DEVICE_ATTR(mbps, uint);
298 NULLB_DEVICE_ATTR(cache_size, ulong);
299 NULLB_DEVICE_ATTR(zoned, bool);
300 NULLB_DEVICE_ATTR(zone_size, ulong);
301 NULLB_DEVICE_ATTR(zone_nr_conv, uint);
303 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
305 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
308 static ssize_t nullb_device_power_store(struct config_item *item,
309 const char *page, size_t count)
311 struct nullb_device *dev = to_nullb_device(item);
315 ret = nullb_device_bool_attr_store(&newp, page, count);
319 if (!dev->power && newp) {
320 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
322 if (null_add_dev(dev)) {
323 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
327 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
329 } else if (dev->power && !newp) {
330 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
333 null_del_dev(dev->nullb);
336 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
342 CONFIGFS_ATTR(nullb_device_, power);
344 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
346 struct nullb_device *t_dev = to_nullb_device(item);
348 return badblocks_show(&t_dev->badblocks, page, 0);
351 static ssize_t nullb_device_badblocks_store(struct config_item *item,
352 const char *page, size_t count)
354 struct nullb_device *t_dev = to_nullb_device(item);
355 char *orig, *buf, *tmp;
359 orig = kstrndup(page, count, GFP_KERNEL);
363 buf = strstrip(orig);
366 if (buf[0] != '+' && buf[0] != '-')
368 tmp = strchr(&buf[1], '-');
372 ret = kstrtoull(buf + 1, 0, &start);
375 ret = kstrtoull(tmp + 1, 0, &end);
381 /* enable badblocks */
382 cmpxchg(&t_dev->badblocks.shift, -1, 0);
384 ret = badblocks_set(&t_dev->badblocks, start,
387 ret = badblocks_clear(&t_dev->badblocks, start,
395 CONFIGFS_ATTR(nullb_device_, badblocks);
397 static struct configfs_attribute *nullb_device_attrs[] = {
398 &nullb_device_attr_size,
399 &nullb_device_attr_completion_nsec,
400 &nullb_device_attr_submit_queues,
401 &nullb_device_attr_home_node,
402 &nullb_device_attr_queue_mode,
403 &nullb_device_attr_blocksize,
404 &nullb_device_attr_irqmode,
405 &nullb_device_attr_hw_queue_depth,
406 &nullb_device_attr_index,
407 &nullb_device_attr_blocking,
408 &nullb_device_attr_use_per_node_hctx,
409 &nullb_device_attr_power,
410 &nullb_device_attr_memory_backed,
411 &nullb_device_attr_discard,
412 &nullb_device_attr_mbps,
413 &nullb_device_attr_cache_size,
414 &nullb_device_attr_badblocks,
415 &nullb_device_attr_zoned,
416 &nullb_device_attr_zone_size,
417 &nullb_device_attr_zone_nr_conv,
421 static void nullb_device_release(struct config_item *item)
423 struct nullb_device *dev = to_nullb_device(item);
425 null_free_device_storage(dev, false);
429 static struct configfs_item_operations nullb_device_ops = {
430 .release = nullb_device_release,
433 static const struct config_item_type nullb_device_type = {
434 .ct_item_ops = &nullb_device_ops,
435 .ct_attrs = nullb_device_attrs,
436 .ct_owner = THIS_MODULE,
440 config_item *nullb_group_make_item(struct config_group *group, const char *name)
442 struct nullb_device *dev;
444 dev = null_alloc_dev();
446 return ERR_PTR(-ENOMEM);
448 config_item_init_type_name(&dev->item, name, &nullb_device_type);
454 nullb_group_drop_item(struct config_group *group, struct config_item *item)
456 struct nullb_device *dev = to_nullb_device(item);
458 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
461 null_del_dev(dev->nullb);
465 config_item_put(item);
468 static ssize_t memb_group_features_show(struct config_item *item, char *page)
470 return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size\n");
473 CONFIGFS_ATTR_RO(memb_group_, features);
475 static struct configfs_attribute *nullb_group_attrs[] = {
476 &memb_group_attr_features,
480 static struct configfs_group_operations nullb_group_ops = {
481 .make_item = nullb_group_make_item,
482 .drop_item = nullb_group_drop_item,
485 static const struct config_item_type nullb_group_type = {
486 .ct_group_ops = &nullb_group_ops,
487 .ct_attrs = nullb_group_attrs,
488 .ct_owner = THIS_MODULE,
491 static struct configfs_subsystem nullb_subsys = {
494 .ci_namebuf = "nullb",
495 .ci_type = &nullb_group_type,
500 static inline int null_cache_active(struct nullb *nullb)
502 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
505 static struct nullb_device *null_alloc_dev(void)
507 struct nullb_device *dev;
509 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
512 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
513 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
514 if (badblocks_init(&dev->badblocks, 0)) {
519 dev->size = g_gb * 1024;
520 dev->completion_nsec = g_completion_nsec;
521 dev->submit_queues = g_submit_queues;
522 dev->home_node = g_home_node;
523 dev->queue_mode = g_queue_mode;
524 dev->blocksize = g_bs;
525 dev->irqmode = g_irqmode;
526 dev->hw_queue_depth = g_hw_queue_depth;
527 dev->blocking = g_blocking;
528 dev->use_per_node_hctx = g_use_per_node_hctx;
529 dev->zoned = g_zoned;
530 dev->zone_size = g_zone_size;
531 dev->zone_nr_conv = g_zone_nr_conv;
535 static void null_free_dev(struct nullb_device *dev)
541 badblocks_exit(&dev->badblocks);
545 static void put_tag(struct nullb_queue *nq, unsigned int tag)
547 clear_bit_unlock(tag, nq->tag_map);
549 if (waitqueue_active(&nq->wait))
553 static unsigned int get_tag(struct nullb_queue *nq)
558 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
559 if (tag >= nq->queue_depth)
561 } while (test_and_set_bit_lock(tag, nq->tag_map));
566 static void free_cmd(struct nullb_cmd *cmd)
568 put_tag(cmd->nq, cmd->tag);
571 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
573 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
575 struct nullb_cmd *cmd;
580 cmd = &nq->cmds[tag];
583 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
584 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
586 cmd->timer.function = null_cmd_timer_expired;
594 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
596 struct nullb_cmd *cmd;
599 cmd = __alloc_cmd(nq);
600 if (cmd || !can_wait)
604 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
605 cmd = __alloc_cmd(nq);
612 finish_wait(&nq->wait, &wait);
616 static void end_cmd(struct nullb_cmd *cmd)
618 int queue_mode = cmd->nq->dev->queue_mode;
620 switch (queue_mode) {
622 blk_mq_end_request(cmd->rq, cmd->error);
625 cmd->bio->bi_status = cmd->error;
633 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
635 end_cmd(container_of(timer, struct nullb_cmd, timer));
637 return HRTIMER_NORESTART;
640 static void null_cmd_end_timer(struct nullb_cmd *cmd)
642 ktime_t kt = cmd->nq->dev->completion_nsec;
644 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
647 static void null_complete_rq(struct request *rq)
649 end_cmd(blk_mq_rq_to_pdu(rq));
652 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
654 struct nullb_page *t_page;
656 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
660 t_page->page = alloc_pages(gfp_flags, 0);
664 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
672 static void null_free_page(struct nullb_page *t_page)
674 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
675 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
677 __free_page(t_page->page);
681 static bool null_page_empty(struct nullb_page *page)
683 int size = MAP_SZ - 2;
685 return find_first_bit(page->bitmap, size) == size;
688 static void null_free_sector(struct nullb *nullb, sector_t sector,
691 unsigned int sector_bit;
693 struct nullb_page *t_page, *ret;
694 struct radix_tree_root *root;
696 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
697 idx = sector >> PAGE_SECTORS_SHIFT;
698 sector_bit = (sector & SECTOR_MASK);
700 t_page = radix_tree_lookup(root, idx);
702 __clear_bit(sector_bit, t_page->bitmap);
704 if (null_page_empty(t_page)) {
705 ret = radix_tree_delete_item(root, idx, t_page);
706 WARN_ON(ret != t_page);
709 nullb->dev->curr_cache -= PAGE_SIZE;
714 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
715 struct nullb_page *t_page, bool is_cache)
717 struct radix_tree_root *root;
719 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
721 if (radix_tree_insert(root, idx, t_page)) {
722 null_free_page(t_page);
723 t_page = radix_tree_lookup(root, idx);
724 WARN_ON(!t_page || t_page->page->index != idx);
726 nullb->dev->curr_cache += PAGE_SIZE;
731 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
733 unsigned long pos = 0;
735 struct nullb_page *ret, *t_pages[FREE_BATCH];
736 struct radix_tree_root *root;
738 root = is_cache ? &dev->cache : &dev->data;
743 nr_pages = radix_tree_gang_lookup(root,
744 (void **)t_pages, pos, FREE_BATCH);
746 for (i = 0; i < nr_pages; i++) {
747 pos = t_pages[i]->page->index;
748 ret = radix_tree_delete_item(root, pos, t_pages[i]);
749 WARN_ON(ret != t_pages[i]);
754 } while (nr_pages == FREE_BATCH);
760 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
761 sector_t sector, bool for_write, bool is_cache)
763 unsigned int sector_bit;
765 struct nullb_page *t_page;
766 struct radix_tree_root *root;
768 idx = sector >> PAGE_SECTORS_SHIFT;
769 sector_bit = (sector & SECTOR_MASK);
771 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
772 t_page = radix_tree_lookup(root, idx);
773 WARN_ON(t_page && t_page->page->index != idx);
775 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
781 static struct nullb_page *null_lookup_page(struct nullb *nullb,
782 sector_t sector, bool for_write, bool ignore_cache)
784 struct nullb_page *page = NULL;
787 page = __null_lookup_page(nullb, sector, for_write, true);
790 return __null_lookup_page(nullb, sector, for_write, false);
793 static struct nullb_page *null_insert_page(struct nullb *nullb,
794 sector_t sector, bool ignore_cache)
795 __releases(&nullb->lock)
796 __acquires(&nullb->lock)
799 struct nullb_page *t_page;
801 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
805 spin_unlock_irq(&nullb->lock);
807 t_page = null_alloc_page(GFP_NOIO);
811 if (radix_tree_preload(GFP_NOIO))
814 spin_lock_irq(&nullb->lock);
815 idx = sector >> PAGE_SECTORS_SHIFT;
816 t_page->page->index = idx;
817 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
818 radix_tree_preload_end();
822 null_free_page(t_page);
824 spin_lock_irq(&nullb->lock);
825 return null_lookup_page(nullb, sector, true, ignore_cache);
828 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
833 struct nullb_page *t_page, *ret;
836 idx = c_page->page->index;
838 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
840 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
841 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
842 null_free_page(c_page);
843 if (t_page && null_page_empty(t_page)) {
844 ret = radix_tree_delete_item(&nullb->dev->data,
846 null_free_page(t_page);
854 src = kmap_atomic(c_page->page);
855 dst = kmap_atomic(t_page->page);
857 for (i = 0; i < PAGE_SECTORS;
858 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
859 if (test_bit(i, c_page->bitmap)) {
860 offset = (i << SECTOR_SHIFT);
861 memcpy(dst + offset, src + offset,
862 nullb->dev->blocksize);
863 __set_bit(i, t_page->bitmap);
870 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
872 nullb->dev->curr_cache -= PAGE_SIZE;
877 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
879 int i, err, nr_pages;
880 struct nullb_page *c_pages[FREE_BATCH];
881 unsigned long flushed = 0, one_round;
884 if ((nullb->dev->cache_size * 1024 * 1024) >
885 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
888 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
889 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
891 * nullb_flush_cache_page could unlock before using the c_pages. To
892 * avoid race, we don't allow page free
894 for (i = 0; i < nr_pages; i++) {
895 nullb->cache_flush_pos = c_pages[i]->page->index;
897 * We found the page which is being flushed to disk by other
900 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
903 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
907 for (i = 0; i < nr_pages; i++) {
908 if (c_pages[i] == NULL)
910 err = null_flush_cache_page(nullb, c_pages[i]);
915 flushed += one_round << PAGE_SHIFT;
919 nullb->cache_flush_pos = 0;
920 if (one_round == 0) {
921 /* give other threads a chance */
922 spin_unlock_irq(&nullb->lock);
923 spin_lock_irq(&nullb->lock);
930 static int copy_to_nullb(struct nullb *nullb, struct page *source,
931 unsigned int off, sector_t sector, size_t n, bool is_fua)
933 size_t temp, count = 0;
935 struct nullb_page *t_page;
939 temp = min_t(size_t, nullb->dev->blocksize, n - count);
941 if (null_cache_active(nullb) && !is_fua)
942 null_make_cache_space(nullb, PAGE_SIZE);
944 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
945 t_page = null_insert_page(nullb, sector,
946 !null_cache_active(nullb) || is_fua);
950 src = kmap_atomic(source);
951 dst = kmap_atomic(t_page->page);
952 memcpy(dst + offset, src + off + count, temp);
956 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
959 null_free_sector(nullb, sector, true);
962 sector += temp >> SECTOR_SHIFT;
967 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
968 unsigned int off, sector_t sector, size_t n)
970 size_t temp, count = 0;
972 struct nullb_page *t_page;
976 temp = min_t(size_t, nullb->dev->blocksize, n - count);
978 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
979 t_page = null_lookup_page(nullb, sector, false,
980 !null_cache_active(nullb));
982 dst = kmap_atomic(dest);
984 memset(dst + off + count, 0, temp);
987 src = kmap_atomic(t_page->page);
988 memcpy(dst + off + count, src + offset, temp);
994 sector += temp >> SECTOR_SHIFT;
999 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1003 spin_lock_irq(&nullb->lock);
1005 temp = min_t(size_t, n, nullb->dev->blocksize);
1006 null_free_sector(nullb, sector, false);
1007 if (null_cache_active(nullb))
1008 null_free_sector(nullb, sector, true);
1009 sector += temp >> SECTOR_SHIFT;
1012 spin_unlock_irq(&nullb->lock);
1015 static int null_handle_flush(struct nullb *nullb)
1019 if (!null_cache_active(nullb))
1022 spin_lock_irq(&nullb->lock);
1024 err = null_make_cache_space(nullb,
1025 nullb->dev->cache_size * 1024 * 1024);
1026 if (err || nullb->dev->curr_cache == 0)
1030 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1031 spin_unlock_irq(&nullb->lock);
1035 static int null_transfer(struct nullb *nullb, struct page *page,
1036 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1042 err = copy_from_nullb(nullb, page, off, sector, len);
1043 flush_dcache_page(page);
1045 flush_dcache_page(page);
1046 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1052 static int null_handle_rq(struct nullb_cmd *cmd)
1054 struct request *rq = cmd->rq;
1055 struct nullb *nullb = cmd->nq->dev->nullb;
1059 struct req_iterator iter;
1060 struct bio_vec bvec;
1062 sector = blk_rq_pos(rq);
1064 if (req_op(rq) == REQ_OP_DISCARD) {
1065 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1069 spin_lock_irq(&nullb->lock);
1070 rq_for_each_segment(bvec, rq, iter) {
1072 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1073 op_is_write(req_op(rq)), sector,
1074 req_op(rq) & REQ_FUA);
1076 spin_unlock_irq(&nullb->lock);
1079 sector += len >> SECTOR_SHIFT;
1081 spin_unlock_irq(&nullb->lock);
1086 static int null_handle_bio(struct nullb_cmd *cmd)
1088 struct bio *bio = cmd->bio;
1089 struct nullb *nullb = cmd->nq->dev->nullb;
1093 struct bio_vec bvec;
1094 struct bvec_iter iter;
1096 sector = bio->bi_iter.bi_sector;
1098 if (bio_op(bio) == REQ_OP_DISCARD) {
1099 null_handle_discard(nullb, sector,
1100 bio_sectors(bio) << SECTOR_SHIFT);
1104 spin_lock_irq(&nullb->lock);
1105 bio_for_each_segment(bvec, bio, iter) {
1107 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1108 op_is_write(bio_op(bio)), sector,
1109 bio->bi_opf & REQ_FUA);
1111 spin_unlock_irq(&nullb->lock);
1114 sector += len >> SECTOR_SHIFT;
1116 spin_unlock_irq(&nullb->lock);
1120 static void null_stop_queue(struct nullb *nullb)
1122 struct request_queue *q = nullb->q;
1124 if (nullb->dev->queue_mode == NULL_Q_MQ)
1125 blk_mq_stop_hw_queues(q);
1128 static void null_restart_queue_async(struct nullb *nullb)
1130 struct request_queue *q = nullb->q;
1132 if (nullb->dev->queue_mode == NULL_Q_MQ)
1133 blk_mq_start_stopped_hw_queues(q, true);
1136 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1138 struct nullb_device *dev = cmd->nq->dev;
1139 struct nullb *nullb = dev->nullb;
1140 blk_status_t sts = BLK_STS_OK;
1141 struct request *rq = cmd->rq;
1143 if (!hrtimer_active(&nullb->bw_timer))
1144 hrtimer_restart(&nullb->bw_timer);
1146 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1147 null_stop_queue(nullb);
1148 /* race with timer */
1149 if (atomic_long_read(&nullb->cur_bytes) > 0)
1150 null_restart_queue_async(nullb);
1151 /* requeue request */
1152 sts = BLK_STS_DEV_RESOURCE;
1157 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1159 sector_t nr_sectors)
1161 struct badblocks *bb = &cmd->nq->dev->badblocks;
1165 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1166 return BLK_STS_IOERR;
1171 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1174 struct nullb_device *dev = cmd->nq->dev;
1177 if (dev->queue_mode == NULL_Q_BIO)
1178 err = null_handle_bio(cmd);
1180 err = null_handle_rq(cmd);
1182 return errno_to_blk_status(err);
1185 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1187 /* Complete IO by inline, softirq or timer */
1188 switch (cmd->nq->dev->irqmode) {
1189 case NULL_IRQ_SOFTIRQ:
1190 switch (cmd->nq->dev->queue_mode) {
1192 blk_mq_complete_request(cmd->rq);
1196 * XXX: no proper submitting cpu information available.
1205 case NULL_IRQ_TIMER:
1206 null_cmd_end_timer(cmd);
1211 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1212 sector_t nr_sectors, enum req_opf op)
1214 struct nullb_device *dev = cmd->nq->dev;
1215 struct nullb *nullb = dev->nullb;
1218 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1219 sts = null_handle_throttled(cmd);
1220 if (sts != BLK_STS_OK)
1224 if (op == REQ_OP_FLUSH) {
1225 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1229 if (nullb->dev->badblocks.shift != -1) {
1230 cmd->error = null_handle_badblocks(cmd, sector, nr_sectors);
1231 if (cmd->error != BLK_STS_OK)
1235 if (dev->memory_backed)
1236 cmd->error = null_handle_memory_backed(cmd, op);
1238 if (!cmd->error && dev->zoned)
1239 cmd->error = null_handle_zoned(cmd, op, sector, nr_sectors);
1242 nullb_complete_cmd(cmd);
1246 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1248 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1249 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1250 unsigned int mbps = nullb->dev->mbps;
1252 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1253 return HRTIMER_NORESTART;
1255 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1256 null_restart_queue_async(nullb);
1258 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1260 return HRTIMER_RESTART;
1263 static void nullb_setup_bwtimer(struct nullb *nullb)
1265 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1267 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1268 nullb->bw_timer.function = nullb_bwtimer_fn;
1269 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1270 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1273 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1277 if (nullb->nr_queues != 1)
1278 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1280 return &nullb->queues[index];
1283 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1285 sector_t sector = bio->bi_iter.bi_sector;
1286 sector_t nr_sectors = bio_sectors(bio);
1287 struct nullb *nullb = q->queuedata;
1288 struct nullb_queue *nq = nullb_to_queue(nullb);
1289 struct nullb_cmd *cmd;
1291 cmd = alloc_cmd(nq, 1);
1294 null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1295 return BLK_QC_T_NONE;
1298 static bool should_timeout_request(struct request *rq)
1300 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1301 if (g_timeout_str[0])
1302 return should_fail(&null_timeout_attr, 1);
1307 static bool should_requeue_request(struct request *rq)
1309 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1310 if (g_requeue_str[0])
1311 return should_fail(&null_requeue_attr, 1);
1316 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1318 pr_info("rq %p timed out\n", rq);
1319 blk_mq_complete_request(rq);
1323 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1324 const struct blk_mq_queue_data *bd)
1326 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1327 struct nullb_queue *nq = hctx->driver_data;
1328 sector_t nr_sectors = blk_rq_sectors(bd->rq);
1329 sector_t sector = blk_rq_pos(bd->rq);
1331 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1333 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1334 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1335 cmd->timer.function = null_cmd_timer_expired;
1340 blk_mq_start_request(bd->rq);
1342 if (should_requeue_request(bd->rq)) {
1344 * Alternate between hitting the core BUSY path, and the
1345 * driver driven requeue path
1347 nq->requeue_selection++;
1348 if (nq->requeue_selection & 1)
1349 return BLK_STS_RESOURCE;
1351 blk_mq_requeue_request(bd->rq, true);
1355 if (should_timeout_request(bd->rq))
1358 return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1361 static const struct blk_mq_ops null_mq_ops = {
1362 .queue_rq = null_queue_rq,
1363 .complete = null_complete_rq,
1364 .timeout = null_timeout_rq,
1367 static void cleanup_queue(struct nullb_queue *nq)
1373 static void cleanup_queues(struct nullb *nullb)
1377 for (i = 0; i < nullb->nr_queues; i++)
1378 cleanup_queue(&nullb->queues[i]);
1380 kfree(nullb->queues);
1383 static void null_del_dev(struct nullb *nullb)
1385 struct nullb_device *dev = nullb->dev;
1387 ida_simple_remove(&nullb_indexes, nullb->index);
1389 list_del_init(&nullb->list);
1391 del_gendisk(nullb->disk);
1393 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1394 hrtimer_cancel(&nullb->bw_timer);
1395 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1396 null_restart_queue_async(nullb);
1399 blk_cleanup_queue(nullb->q);
1400 if (dev->queue_mode == NULL_Q_MQ &&
1401 nullb->tag_set == &nullb->__tag_set)
1402 blk_mq_free_tag_set(nullb->tag_set);
1403 put_disk(nullb->disk);
1404 cleanup_queues(nullb);
1405 if (null_cache_active(nullb))
1406 null_free_device_storage(nullb->dev, true);
1411 static void null_config_discard(struct nullb *nullb)
1413 if (nullb->dev->discard == false)
1415 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1416 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1417 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1418 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1421 static int null_open(struct block_device *bdev, fmode_t mode)
1426 static void null_release(struct gendisk *disk, fmode_t mode)
1430 static const struct block_device_operations null_fops = {
1431 .owner = THIS_MODULE,
1433 .release = null_release,
1434 .report_zones = null_zone_report,
1437 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1442 init_waitqueue_head(&nq->wait);
1443 nq->queue_depth = nullb->queue_depth;
1444 nq->dev = nullb->dev;
1447 static void null_init_queues(struct nullb *nullb)
1449 struct request_queue *q = nullb->q;
1450 struct blk_mq_hw_ctx *hctx;
1451 struct nullb_queue *nq;
1454 queue_for_each_hw_ctx(q, hctx, i) {
1455 if (!hctx->nr_ctx || !hctx->tags)
1457 nq = &nullb->queues[i];
1458 hctx->driver_data = nq;
1459 null_init_queue(nullb, nq);
1464 static int setup_commands(struct nullb_queue *nq)
1466 struct nullb_cmd *cmd;
1469 nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1473 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1474 nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1480 for (i = 0; i < nq->queue_depth; i++) {
1482 INIT_LIST_HEAD(&cmd->list);
1483 cmd->ll_list.next = NULL;
1490 static int setup_queues(struct nullb *nullb)
1492 nullb->queues = kcalloc(nullb->dev->submit_queues,
1493 sizeof(struct nullb_queue),
1498 nullb->queue_depth = nullb->dev->hw_queue_depth;
1503 static int init_driver_queues(struct nullb *nullb)
1505 struct nullb_queue *nq;
1508 for (i = 0; i < nullb->dev->submit_queues; i++) {
1509 nq = &nullb->queues[i];
1511 null_init_queue(nullb, nq);
1513 ret = setup_commands(nq);
1521 static int null_gendisk_register(struct nullb *nullb)
1523 struct gendisk *disk;
1526 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1529 size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1530 set_capacity(disk, size >> 9);
1532 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1533 disk->major = null_major;
1534 disk->first_minor = nullb->index;
1535 disk->fops = &null_fops;
1536 disk->private_data = nullb;
1537 disk->queue = nullb->q;
1538 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1540 if (nullb->dev->zoned) {
1541 int ret = blk_revalidate_disk_zones(disk);
1551 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1553 set->ops = &null_mq_ops;
1554 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1556 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1558 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1559 set->cmd_size = sizeof(struct nullb_cmd);
1560 set->flags = BLK_MQ_F_SHOULD_MERGE;
1562 set->flags |= BLK_MQ_F_NO_SCHED;
1563 set->driver_data = NULL;
1565 if ((nullb && nullb->dev->blocking) || g_blocking)
1566 set->flags |= BLK_MQ_F_BLOCKING;
1568 return blk_mq_alloc_tag_set(set);
1571 static void null_validate_conf(struct nullb_device *dev)
1573 dev->blocksize = round_down(dev->blocksize, 512);
1574 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1576 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1577 if (dev->submit_queues != nr_online_nodes)
1578 dev->submit_queues = nr_online_nodes;
1579 } else if (dev->submit_queues > nr_cpu_ids)
1580 dev->submit_queues = nr_cpu_ids;
1581 else if (dev->submit_queues == 0)
1582 dev->submit_queues = 1;
1584 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1585 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1587 /* Do memory allocation, so set blocking */
1588 if (dev->memory_backed)
1589 dev->blocking = true;
1590 else /* cache is meaningless */
1591 dev->cache_size = 0;
1592 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1594 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1595 /* can not stop a queue */
1596 if (dev->queue_mode == NULL_Q_BIO)
1600 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1601 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1606 if (!setup_fault_attr(attr, str))
1614 static bool null_setup_fault(void)
1616 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1617 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1619 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1625 static int null_add_dev(struct nullb_device *dev)
1627 struct nullb *nullb;
1630 null_validate_conf(dev);
1632 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1640 spin_lock_init(&nullb->lock);
1642 rv = setup_queues(nullb);
1644 goto out_free_nullb;
1646 if (dev->queue_mode == NULL_Q_MQ) {
1648 nullb->tag_set = &tag_set;
1651 nullb->tag_set = &nullb->__tag_set;
1652 rv = null_init_tag_set(nullb, nullb->tag_set);
1656 goto out_cleanup_queues;
1658 if (!null_setup_fault())
1659 goto out_cleanup_queues;
1661 nullb->tag_set->timeout = 5 * HZ;
1662 nullb->q = blk_mq_init_queue(nullb->tag_set);
1663 if (IS_ERR(nullb->q)) {
1665 goto out_cleanup_tags;
1667 null_init_queues(nullb);
1668 } else if (dev->queue_mode == NULL_Q_BIO) {
1669 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node);
1672 goto out_cleanup_queues;
1674 blk_queue_make_request(nullb->q, null_queue_bio);
1675 rv = init_driver_queues(nullb);
1677 goto out_cleanup_blk_queue;
1681 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1682 nullb_setup_bwtimer(nullb);
1685 if (dev->cache_size > 0) {
1686 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1687 blk_queue_write_cache(nullb->q, true, true);
1691 rv = null_zone_init(dev);
1693 goto out_cleanup_blk_queue;
1695 blk_queue_chunk_sectors(nullb->q, dev->zone_size_sects);
1696 nullb->q->limits.zoned = BLK_ZONED_HM;
1697 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, nullb->q);
1698 blk_queue_required_elevator_features(nullb->q,
1699 ELEVATOR_F_ZBD_SEQ_WRITE);
1702 nullb->q->queuedata = nullb;
1703 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1704 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1707 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1708 dev->index = nullb->index;
1709 mutex_unlock(&lock);
1711 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1712 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1714 null_config_discard(nullb);
1716 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1718 rv = null_gendisk_register(nullb);
1720 goto out_cleanup_zone;
1723 list_add_tail(&nullb->list, &nullb_list);
1724 mutex_unlock(&lock);
1729 null_zone_exit(dev);
1730 out_cleanup_blk_queue:
1731 blk_cleanup_queue(nullb->q);
1733 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1734 blk_mq_free_tag_set(nullb->tag_set);
1736 cleanup_queues(nullb);
1743 static int __init null_init(void)
1747 struct nullb *nullb;
1748 struct nullb_device *dev;
1750 if (g_bs > PAGE_SIZE) {
1751 pr_warn("invalid block size\n");
1752 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1756 if (!is_power_of_2(g_zone_size)) {
1757 pr_err("zone_size must be power-of-two\n");
1761 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1762 pr_err("invalid home_node value\n");
1763 g_home_node = NUMA_NO_NODE;
1766 if (g_queue_mode == NULL_Q_RQ) {
1767 pr_err("legacy IO path no longer available\n");
1770 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1771 if (g_submit_queues != nr_online_nodes) {
1772 pr_warn("submit_queues param is set to %u.\n",
1774 g_submit_queues = nr_online_nodes;
1776 } else if (g_submit_queues > nr_cpu_ids)
1777 g_submit_queues = nr_cpu_ids;
1778 else if (g_submit_queues <= 0)
1779 g_submit_queues = 1;
1781 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1782 ret = null_init_tag_set(NULL, &tag_set);
1787 config_group_init(&nullb_subsys.su_group);
1788 mutex_init(&nullb_subsys.su_mutex);
1790 ret = configfs_register_subsystem(&nullb_subsys);
1796 null_major = register_blkdev(0, "nullb");
1797 if (null_major < 0) {
1802 for (i = 0; i < nr_devices; i++) {
1803 dev = null_alloc_dev();
1808 ret = null_add_dev(dev);
1815 pr_info("module loaded\n");
1819 while (!list_empty(&nullb_list)) {
1820 nullb = list_entry(nullb_list.next, struct nullb, list);
1822 null_del_dev(nullb);
1825 unregister_blkdev(null_major, "nullb");
1827 configfs_unregister_subsystem(&nullb_subsys);
1829 if (g_queue_mode == NULL_Q_MQ && shared_tags)
1830 blk_mq_free_tag_set(&tag_set);
1834 static void __exit null_exit(void)
1836 struct nullb *nullb;
1838 configfs_unregister_subsystem(&nullb_subsys);
1840 unregister_blkdev(null_major, "nullb");
1843 while (!list_empty(&nullb_list)) {
1844 struct nullb_device *dev;
1846 nullb = list_entry(nullb_list.next, struct nullb, list);
1848 null_del_dev(nullb);
1851 mutex_unlock(&lock);
1853 if (g_queue_mode == NULL_Q_MQ && shared_tags)
1854 blk_mq_free_tag_set(&tag_set);
1857 module_init(null_init);
1858 module_exit(null_exit);
1860 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
1861 MODULE_LICENSE("GPL");