2 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
3 * Shaohua Li <shli@fb.com>
5 #include <linux/module.h>
7 #include <linux/moduleparam.h>
8 #include <linux/sched.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>
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)
27 #define TICKS_PER_SEC 50ULL
28 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
30 static inline u64 mb_per_tick(int mbps)
32 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
36 struct list_head list;
37 struct llist_node ll_list;
38 call_single_data_t csd;
42 struct nullb_queue *nq;
48 unsigned long *tag_map;
49 wait_queue_head_t wait;
50 unsigned int queue_depth;
51 struct nullb_device *dev;
53 struct nullb_cmd *cmds;
57 * Status flags for nullb_device.
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.
64 enum nullb_device_flags {
65 NULLB_DEV_FL_CONFIGURED = 0,
67 NULLB_DEV_FL_THROTTLED = 2,
68 NULLB_DEV_FL_CACHE = 3,
72 * nullb_page is a page in memory for nullb devices.
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
87 #define NULLB_PAGE_LOCK (sizeof(unsigned long) * 8 - 1)
88 #define NULLB_PAGE_FREE (sizeof(unsigned long) * 8 - 2)
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;
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 */
119 struct nullb_device *dev;
120 struct list_head list;
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;
133 struct nullb_queue *queues;
134 unsigned int nr_queues;
135 char disk_name[DISK_NAME_LEN];
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;
147 NULL_IRQ_SOFTIRQ = 1,
157 static int g_submit_queues = 1;
158 module_param_named(submit_queues, g_submit_queues, int, S_IRUGO);
159 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
161 static int g_home_node = NUMA_NO_NODE;
162 module_param_named(home_node, g_home_node, int, S_IRUGO);
163 MODULE_PARM_DESC(home_node, "Home node for the device");
165 static int g_queue_mode = NULL_Q_MQ;
167 static int null_param_store_val(const char *str, int *val, int min, int max)
171 ret = kstrtoint(str, 10, &new_val);
175 if (new_val < min || new_val > max)
182 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
184 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
187 static const struct kernel_param_ops null_queue_mode_param_ops = {
188 .set = null_set_queue_mode,
189 .get = param_get_int,
192 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, S_IRUGO);
193 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
195 static int g_gb = 250;
196 module_param_named(gb, g_gb, int, S_IRUGO);
197 MODULE_PARM_DESC(gb, "Size in GB");
199 static int g_bs = 512;
200 module_param_named(bs, g_bs, int, S_IRUGO);
201 MODULE_PARM_DESC(bs, "Block size (in bytes)");
203 static int nr_devices = 1;
204 module_param(nr_devices, int, S_IRUGO);
205 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
207 static bool g_use_lightnvm;
208 module_param_named(use_lightnvm, g_use_lightnvm, bool, S_IRUGO);
209 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
211 static bool g_blocking;
212 module_param_named(blocking, g_blocking, bool, S_IRUGO);
213 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
215 static bool shared_tags;
216 module_param(shared_tags, bool, S_IRUGO);
217 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
219 static int g_irqmode = NULL_IRQ_SOFTIRQ;
221 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
223 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
227 static const struct kernel_param_ops null_irqmode_param_ops = {
228 .set = null_set_irqmode,
229 .get = param_get_int,
232 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, S_IRUGO);
233 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
235 static unsigned long g_completion_nsec = 10000;
236 module_param_named(completion_nsec, g_completion_nsec, ulong, S_IRUGO);
237 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
239 static int g_hw_queue_depth = 64;
240 module_param_named(hw_queue_depth, g_hw_queue_depth, int, S_IRUGO);
241 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
243 static bool g_use_per_node_hctx;
244 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, S_IRUGO);
245 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
247 static struct nullb_device *null_alloc_dev(void);
248 static void null_free_dev(struct nullb_device *dev);
249 static void null_del_dev(struct nullb *nullb);
250 static int null_add_dev(struct nullb_device *dev);
251 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
253 static inline struct nullb_device *to_nullb_device(struct config_item *item)
255 return item ? container_of(item, struct nullb_device, item) : NULL;
258 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
260 return snprintf(page, PAGE_SIZE, "%u\n", val);
263 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
266 return snprintf(page, PAGE_SIZE, "%lu\n", val);
269 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
271 return snprintf(page, PAGE_SIZE, "%u\n", val);
274 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
275 const char *page, size_t count)
280 result = kstrtouint(page, 0, &tmp);
288 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
289 const char *page, size_t count)
294 result = kstrtoul(page, 0, &tmp);
302 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
308 result = kstrtobool(page, &tmp);
316 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
317 #define NULLB_DEVICE_ATTR(NAME, TYPE) \
319 nullb_device_##NAME##_show(struct config_item *item, char *page) \
321 return nullb_device_##TYPE##_attr_show( \
322 to_nullb_device(item)->NAME, page); \
325 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
328 if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags)) \
330 return nullb_device_##TYPE##_attr_store( \
331 &to_nullb_device(item)->NAME, page, count); \
333 CONFIGFS_ATTR(nullb_device_, NAME);
335 NULLB_DEVICE_ATTR(size, ulong);
336 NULLB_DEVICE_ATTR(completion_nsec, ulong);
337 NULLB_DEVICE_ATTR(submit_queues, uint);
338 NULLB_DEVICE_ATTR(home_node, uint);
339 NULLB_DEVICE_ATTR(queue_mode, uint);
340 NULLB_DEVICE_ATTR(blocksize, uint);
341 NULLB_DEVICE_ATTR(irqmode, uint);
342 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
343 NULLB_DEVICE_ATTR(index, uint);
344 NULLB_DEVICE_ATTR(use_lightnvm, bool);
345 NULLB_DEVICE_ATTR(blocking, bool);
346 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
347 NULLB_DEVICE_ATTR(memory_backed, bool);
348 NULLB_DEVICE_ATTR(discard, bool);
349 NULLB_DEVICE_ATTR(mbps, uint);
350 NULLB_DEVICE_ATTR(cache_size, ulong);
352 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
354 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
357 static ssize_t nullb_device_power_store(struct config_item *item,
358 const char *page, size_t count)
360 struct nullb_device *dev = to_nullb_device(item);
364 ret = nullb_device_bool_attr_store(&newp, page, count);
368 if (!dev->power && newp) {
369 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
371 if (null_add_dev(dev)) {
372 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
376 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
378 } else if (dev->power && !newp) {
381 null_del_dev(dev->nullb);
383 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
389 CONFIGFS_ATTR(nullb_device_, power);
391 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
393 struct nullb_device *t_dev = to_nullb_device(item);
395 return badblocks_show(&t_dev->badblocks, page, 0);
398 static ssize_t nullb_device_badblocks_store(struct config_item *item,
399 const char *page, size_t count)
401 struct nullb_device *t_dev = to_nullb_device(item);
402 char *orig, *buf, *tmp;
406 orig = kstrndup(page, count, GFP_KERNEL);
410 buf = strstrip(orig);
413 if (buf[0] != '+' && buf[0] != '-')
415 tmp = strchr(&buf[1], '-');
419 ret = kstrtoull(buf + 1, 0, &start);
422 ret = kstrtoull(tmp + 1, 0, &end);
428 /* enable badblocks */
429 cmpxchg(&t_dev->badblocks.shift, -1, 0);
431 ret = badblocks_set(&t_dev->badblocks, start,
434 ret = badblocks_clear(&t_dev->badblocks, start,
442 CONFIGFS_ATTR(nullb_device_, badblocks);
444 static struct configfs_attribute *nullb_device_attrs[] = {
445 &nullb_device_attr_size,
446 &nullb_device_attr_completion_nsec,
447 &nullb_device_attr_submit_queues,
448 &nullb_device_attr_home_node,
449 &nullb_device_attr_queue_mode,
450 &nullb_device_attr_blocksize,
451 &nullb_device_attr_irqmode,
452 &nullb_device_attr_hw_queue_depth,
453 &nullb_device_attr_index,
454 &nullb_device_attr_use_lightnvm,
455 &nullb_device_attr_blocking,
456 &nullb_device_attr_use_per_node_hctx,
457 &nullb_device_attr_power,
458 &nullb_device_attr_memory_backed,
459 &nullb_device_attr_discard,
460 &nullb_device_attr_mbps,
461 &nullb_device_attr_cache_size,
462 &nullb_device_attr_badblocks,
466 static void nullb_device_release(struct config_item *item)
468 struct nullb_device *dev = to_nullb_device(item);
470 badblocks_exit(&dev->badblocks);
471 null_free_device_storage(dev, false);
475 static struct configfs_item_operations nullb_device_ops = {
476 .release = nullb_device_release,
479 static struct config_item_type nullb_device_type = {
480 .ct_item_ops = &nullb_device_ops,
481 .ct_attrs = nullb_device_attrs,
482 .ct_owner = THIS_MODULE,
486 config_item *nullb_group_make_item(struct config_group *group, const char *name)
488 struct nullb_device *dev;
490 dev = null_alloc_dev();
492 return ERR_PTR(-ENOMEM);
494 config_item_init_type_name(&dev->item, name, &nullb_device_type);
500 nullb_group_drop_item(struct config_group *group, struct config_item *item)
502 struct nullb_device *dev = to_nullb_device(item);
504 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
507 null_del_dev(dev->nullb);
511 config_item_put(item);
514 static ssize_t memb_group_features_show(struct config_item *item, char *page)
516 return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks\n");
519 CONFIGFS_ATTR_RO(memb_group_, features);
521 static struct configfs_attribute *nullb_group_attrs[] = {
522 &memb_group_attr_features,
526 static struct configfs_group_operations nullb_group_ops = {
527 .make_item = nullb_group_make_item,
528 .drop_item = nullb_group_drop_item,
531 static struct config_item_type nullb_group_type = {
532 .ct_group_ops = &nullb_group_ops,
533 .ct_attrs = nullb_group_attrs,
534 .ct_owner = THIS_MODULE,
537 static struct configfs_subsystem nullb_subsys = {
540 .ci_namebuf = "nullb",
541 .ci_type = &nullb_group_type,
546 static inline int null_cache_active(struct nullb *nullb)
548 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
551 static struct nullb_device *null_alloc_dev(void)
553 struct nullb_device *dev;
555 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
558 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
559 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
560 if (badblocks_init(&dev->badblocks, 0)) {
565 dev->size = g_gb * 1024;
566 dev->completion_nsec = g_completion_nsec;
567 dev->submit_queues = g_submit_queues;
568 dev->home_node = g_home_node;
569 dev->queue_mode = g_queue_mode;
570 dev->blocksize = g_bs;
571 dev->irqmode = g_irqmode;
572 dev->hw_queue_depth = g_hw_queue_depth;
573 dev->use_lightnvm = g_use_lightnvm;
574 dev->blocking = g_blocking;
575 dev->use_per_node_hctx = g_use_per_node_hctx;
579 static void null_free_dev(struct nullb_device *dev)
584 static void put_tag(struct nullb_queue *nq, unsigned int tag)
586 clear_bit_unlock(tag, nq->tag_map);
588 if (waitqueue_active(&nq->wait))
592 static unsigned int get_tag(struct nullb_queue *nq)
597 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
598 if (tag >= nq->queue_depth)
600 } while (test_and_set_bit_lock(tag, nq->tag_map));
605 static void free_cmd(struct nullb_cmd *cmd)
607 put_tag(cmd->nq, cmd->tag);
610 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
612 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
614 struct nullb_cmd *cmd;
619 cmd = &nq->cmds[tag];
622 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
623 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
625 cmd->timer.function = null_cmd_timer_expired;
633 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
635 struct nullb_cmd *cmd;
638 cmd = __alloc_cmd(nq);
639 if (cmd || !can_wait)
643 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
644 cmd = __alloc_cmd(nq);
651 finish_wait(&nq->wait, &wait);
655 static void end_cmd(struct nullb_cmd *cmd)
657 struct request_queue *q = NULL;
658 int queue_mode = cmd->nq->dev->queue_mode;
663 switch (queue_mode) {
665 blk_mq_end_request(cmd->rq, cmd->error);
668 INIT_LIST_HEAD(&cmd->rq->queuelist);
669 blk_end_request_all(cmd->rq, cmd->error);
672 cmd->bio->bi_status = cmd->error;
679 /* Restart queue if needed, as we are freeing a tag */
680 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
683 spin_lock_irqsave(q->queue_lock, flags);
684 blk_start_queue_async(q);
685 spin_unlock_irqrestore(q->queue_lock, flags);
689 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
691 end_cmd(container_of(timer, struct nullb_cmd, timer));
693 return HRTIMER_NORESTART;
696 static void null_cmd_end_timer(struct nullb_cmd *cmd)
698 ktime_t kt = cmd->nq->dev->completion_nsec;
700 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
703 static void null_softirq_done_fn(struct request *rq)
705 struct nullb *nullb = rq->q->queuedata;
707 if (nullb->dev->queue_mode == NULL_Q_MQ)
708 end_cmd(blk_mq_rq_to_pdu(rq));
710 end_cmd(rq->special);
713 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
715 struct nullb_page *t_page;
717 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
721 t_page->page = alloc_pages(gfp_flags, 0);
733 static void null_free_page(struct nullb_page *t_page)
735 __set_bit(NULLB_PAGE_FREE, &t_page->bitmap);
736 if (test_bit(NULLB_PAGE_LOCK, &t_page->bitmap))
738 __free_page(t_page->page);
742 static void null_free_sector(struct nullb *nullb, sector_t sector,
745 unsigned int sector_bit;
747 struct nullb_page *t_page, *ret;
748 struct radix_tree_root *root;
750 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
751 idx = sector >> PAGE_SECTORS_SHIFT;
752 sector_bit = (sector & SECTOR_MASK);
754 t_page = radix_tree_lookup(root, idx);
756 __clear_bit(sector_bit, &t_page->bitmap);
758 if (!t_page->bitmap) {
759 ret = radix_tree_delete_item(root, idx, t_page);
760 WARN_ON(ret != t_page);
763 nullb->dev->curr_cache -= PAGE_SIZE;
768 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
769 struct nullb_page *t_page, bool is_cache)
771 struct radix_tree_root *root;
773 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
775 if (radix_tree_insert(root, idx, t_page)) {
776 null_free_page(t_page);
777 t_page = radix_tree_lookup(root, idx);
778 WARN_ON(!t_page || t_page->page->index != idx);
780 nullb->dev->curr_cache += PAGE_SIZE;
785 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
787 unsigned long pos = 0;
789 struct nullb_page *ret, *t_pages[FREE_BATCH];
790 struct radix_tree_root *root;
792 root = is_cache ? &dev->cache : &dev->data;
797 nr_pages = radix_tree_gang_lookup(root,
798 (void **)t_pages, pos, FREE_BATCH);
800 for (i = 0; i < nr_pages; i++) {
801 pos = t_pages[i]->page->index;
802 ret = radix_tree_delete_item(root, pos, t_pages[i]);
803 WARN_ON(ret != t_pages[i]);
808 } while (nr_pages == FREE_BATCH);
814 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
815 sector_t sector, bool for_write, bool is_cache)
817 unsigned int sector_bit;
819 struct nullb_page *t_page;
820 struct radix_tree_root *root;
822 idx = sector >> PAGE_SECTORS_SHIFT;
823 sector_bit = (sector & SECTOR_MASK);
825 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
826 t_page = radix_tree_lookup(root, idx);
827 WARN_ON(t_page && t_page->page->index != idx);
829 if (t_page && (for_write || test_bit(sector_bit, &t_page->bitmap)))
835 static struct nullb_page *null_lookup_page(struct nullb *nullb,
836 sector_t sector, bool for_write, bool ignore_cache)
838 struct nullb_page *page = NULL;
841 page = __null_lookup_page(nullb, sector, for_write, true);
844 return __null_lookup_page(nullb, sector, for_write, false);
847 static struct nullb_page *null_insert_page(struct nullb *nullb,
848 sector_t sector, bool ignore_cache)
851 struct nullb_page *t_page;
853 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
857 spin_unlock_irq(&nullb->lock);
859 t_page = null_alloc_page(GFP_NOIO);
863 if (radix_tree_preload(GFP_NOIO))
866 spin_lock_irq(&nullb->lock);
867 idx = sector >> PAGE_SECTORS_SHIFT;
868 t_page->page->index = idx;
869 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
870 radix_tree_preload_end();
874 null_free_page(t_page);
876 spin_lock_irq(&nullb->lock);
877 return null_lookup_page(nullb, sector, true, ignore_cache);
880 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
885 struct nullb_page *t_page, *ret;
888 idx = c_page->page->index;
890 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
892 __clear_bit(NULLB_PAGE_LOCK, &c_page->bitmap);
893 if (test_bit(NULLB_PAGE_FREE, &c_page->bitmap)) {
894 null_free_page(c_page);
895 if (t_page && t_page->bitmap == 0) {
896 ret = radix_tree_delete_item(&nullb->dev->data,
898 null_free_page(t_page);
906 src = kmap_atomic(c_page->page);
907 dst = kmap_atomic(t_page->page);
909 for (i = 0; i < PAGE_SECTORS;
910 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
911 if (test_bit(i, &c_page->bitmap)) {
912 offset = (i << SECTOR_SHIFT);
913 memcpy(dst + offset, src + offset,
914 nullb->dev->blocksize);
915 __set_bit(i, &t_page->bitmap);
922 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
924 nullb->dev->curr_cache -= PAGE_SIZE;
929 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
931 int i, err, nr_pages;
932 struct nullb_page *c_pages[FREE_BATCH];
933 unsigned long flushed = 0, one_round;
936 if ((nullb->dev->cache_size * 1024 * 1024) >
937 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
940 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
941 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
943 * nullb_flush_cache_page could unlock before using the c_pages. To
944 * avoid race, we don't allow page free
946 for (i = 0; i < nr_pages; i++) {
947 nullb->cache_flush_pos = c_pages[i]->page->index;
949 * We found the page which is being flushed to disk by other
952 if (test_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap))
955 __set_bit(NULLB_PAGE_LOCK, &c_pages[i]->bitmap);
959 for (i = 0; i < nr_pages; i++) {
960 if (c_pages[i] == NULL)
962 err = null_flush_cache_page(nullb, c_pages[i]);
967 flushed += one_round << PAGE_SHIFT;
971 nullb->cache_flush_pos = 0;
972 if (one_round == 0) {
973 /* give other threads a chance */
974 spin_unlock_irq(&nullb->lock);
975 spin_lock_irq(&nullb->lock);
982 static int copy_to_nullb(struct nullb *nullb, struct page *source,
983 unsigned int off, sector_t sector, size_t n, bool is_fua)
985 size_t temp, count = 0;
987 struct nullb_page *t_page;
991 temp = min_t(size_t, nullb->dev->blocksize, n - count);
993 if (null_cache_active(nullb) && !is_fua)
994 null_make_cache_space(nullb, PAGE_SIZE);
996 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
997 t_page = null_insert_page(nullb, sector,
998 !null_cache_active(nullb) || is_fua);
1002 src = kmap_atomic(source);
1003 dst = kmap_atomic(t_page->page);
1004 memcpy(dst + offset, src + off + count, temp);
1008 __set_bit(sector & SECTOR_MASK, &t_page->bitmap);
1011 null_free_sector(nullb, sector, true);
1014 sector += temp >> SECTOR_SHIFT;
1019 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1020 unsigned int off, sector_t sector, size_t n)
1022 size_t temp, count = 0;
1023 unsigned int offset;
1024 struct nullb_page *t_page;
1028 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1030 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1031 t_page = null_lookup_page(nullb, sector, false,
1032 !null_cache_active(nullb));
1034 dst = kmap_atomic(dest);
1036 memset(dst + off + count, 0, temp);
1039 src = kmap_atomic(t_page->page);
1040 memcpy(dst + off + count, src + offset, temp);
1046 sector += temp >> SECTOR_SHIFT;
1051 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1055 spin_lock_irq(&nullb->lock);
1057 temp = min_t(size_t, n, nullb->dev->blocksize);
1058 null_free_sector(nullb, sector, false);
1059 if (null_cache_active(nullb))
1060 null_free_sector(nullb, sector, true);
1061 sector += temp >> SECTOR_SHIFT;
1064 spin_unlock_irq(&nullb->lock);
1067 static int null_handle_flush(struct nullb *nullb)
1071 if (!null_cache_active(nullb))
1074 spin_lock_irq(&nullb->lock);
1076 err = null_make_cache_space(nullb,
1077 nullb->dev->cache_size * 1024 * 1024);
1078 if (err || nullb->dev->curr_cache == 0)
1082 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1083 spin_unlock_irq(&nullb->lock);
1087 static int null_transfer(struct nullb *nullb, struct page *page,
1088 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1094 err = copy_from_nullb(nullb, page, off, sector, len);
1095 flush_dcache_page(page);
1097 flush_dcache_page(page);
1098 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1104 static int null_handle_rq(struct nullb_cmd *cmd)
1106 struct request *rq = cmd->rq;
1107 struct nullb *nullb = cmd->nq->dev->nullb;
1111 struct req_iterator iter;
1112 struct bio_vec bvec;
1114 sector = blk_rq_pos(rq);
1116 if (req_op(rq) == REQ_OP_DISCARD) {
1117 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1121 spin_lock_irq(&nullb->lock);
1122 rq_for_each_segment(bvec, rq, iter) {
1124 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1125 op_is_write(req_op(rq)), sector,
1126 req_op(rq) & REQ_FUA);
1128 spin_unlock_irq(&nullb->lock);
1131 sector += len >> SECTOR_SHIFT;
1133 spin_unlock_irq(&nullb->lock);
1138 static int null_handle_bio(struct nullb_cmd *cmd)
1140 struct bio *bio = cmd->bio;
1141 struct nullb *nullb = cmd->nq->dev->nullb;
1145 struct bio_vec bvec;
1146 struct bvec_iter iter;
1148 sector = bio->bi_iter.bi_sector;
1150 if (bio_op(bio) == REQ_OP_DISCARD) {
1151 null_handle_discard(nullb, sector,
1152 bio_sectors(bio) << SECTOR_SHIFT);
1156 spin_lock_irq(&nullb->lock);
1157 bio_for_each_segment(bvec, bio, iter) {
1159 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1160 op_is_write(bio_op(bio)), sector,
1161 bio_op(bio) & REQ_FUA);
1163 spin_unlock_irq(&nullb->lock);
1166 sector += len >> SECTOR_SHIFT;
1168 spin_unlock_irq(&nullb->lock);
1172 static void null_stop_queue(struct nullb *nullb)
1174 struct request_queue *q = nullb->q;
1176 if (nullb->dev->queue_mode == NULL_Q_MQ)
1177 blk_mq_stop_hw_queues(q);
1179 spin_lock_irq(q->queue_lock);
1181 spin_unlock_irq(q->queue_lock);
1185 static void null_restart_queue_async(struct nullb *nullb)
1187 struct request_queue *q = nullb->q;
1188 unsigned long flags;
1190 if (nullb->dev->queue_mode == NULL_Q_MQ)
1191 blk_mq_start_stopped_hw_queues(q, true);
1193 spin_lock_irqsave(q->queue_lock, flags);
1194 blk_start_queue_async(q);
1195 spin_unlock_irqrestore(q->queue_lock, flags);
1199 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd)
1201 struct nullb_device *dev = cmd->nq->dev;
1202 struct nullb *nullb = dev->nullb;
1205 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1206 struct request *rq = cmd->rq;
1208 if (!hrtimer_active(&nullb->bw_timer))
1209 hrtimer_restart(&nullb->bw_timer);
1211 if (atomic_long_sub_return(blk_rq_bytes(rq),
1212 &nullb->cur_bytes) < 0) {
1213 null_stop_queue(nullb);
1214 /* race with timer */
1215 if (atomic_long_read(&nullb->cur_bytes) > 0)
1216 null_restart_queue_async(nullb);
1217 if (dev->queue_mode == NULL_Q_RQ) {
1218 struct request_queue *q = nullb->q;
1220 spin_lock_irq(q->queue_lock);
1221 rq->rq_flags |= RQF_DONTPREP;
1222 blk_requeue_request(q, rq);
1223 spin_unlock_irq(q->queue_lock);
1226 /* requeue request */
1227 return BLK_STS_RESOURCE;
1231 if (nullb->dev->badblocks.shift != -1) {
1233 sector_t sector, size, first_bad;
1234 bool is_flush = true;
1236 if (dev->queue_mode == NULL_Q_BIO &&
1237 bio_op(cmd->bio) != REQ_OP_FLUSH) {
1239 sector = cmd->bio->bi_iter.bi_sector;
1240 size = bio_sectors(cmd->bio);
1242 if (dev->queue_mode != NULL_Q_BIO &&
1243 req_op(cmd->rq) != REQ_OP_FLUSH) {
1245 sector = blk_rq_pos(cmd->rq);
1246 size = blk_rq_sectors(cmd->rq);
1248 if (!is_flush && badblocks_check(&nullb->dev->badblocks, sector,
1249 size, &first_bad, &bad_sectors)) {
1250 cmd->error = BLK_STS_IOERR;
1255 if (dev->memory_backed) {
1256 if (dev->queue_mode == NULL_Q_BIO) {
1257 if (bio_op(cmd->bio) == REQ_OP_FLUSH)
1258 err = null_handle_flush(nullb);
1260 err = null_handle_bio(cmd);
1262 if (req_op(cmd->rq) == REQ_OP_FLUSH)
1263 err = null_handle_flush(nullb);
1265 err = null_handle_rq(cmd);
1268 cmd->error = errno_to_blk_status(err);
1270 /* Complete IO by inline, softirq or timer */
1271 switch (dev->irqmode) {
1272 case NULL_IRQ_SOFTIRQ:
1273 switch (dev->queue_mode) {
1275 blk_mq_complete_request(cmd->rq);
1278 blk_complete_request(cmd->rq);
1282 * XXX: no proper submitting cpu information available.
1291 case NULL_IRQ_TIMER:
1292 null_cmd_end_timer(cmd);
1298 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1300 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1301 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1302 unsigned int mbps = nullb->dev->mbps;
1304 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1305 return HRTIMER_NORESTART;
1307 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1308 null_restart_queue_async(nullb);
1310 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1312 return HRTIMER_RESTART;
1315 static void nullb_setup_bwtimer(struct nullb *nullb)
1317 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1319 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1320 nullb->bw_timer.function = nullb_bwtimer_fn;
1321 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1322 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1325 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1329 if (nullb->nr_queues != 1)
1330 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1332 return &nullb->queues[index];
1335 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1337 struct nullb *nullb = q->queuedata;
1338 struct nullb_queue *nq = nullb_to_queue(nullb);
1339 struct nullb_cmd *cmd;
1341 cmd = alloc_cmd(nq, 1);
1344 null_handle_cmd(cmd);
1345 return BLK_QC_T_NONE;
1348 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
1350 struct nullb *nullb = q->queuedata;
1351 struct nullb_queue *nq = nullb_to_queue(nullb);
1352 struct nullb_cmd *cmd;
1354 cmd = alloc_cmd(nq, 0);
1362 return BLKPREP_DEFER;
1365 static void null_request_fn(struct request_queue *q)
1369 while ((rq = blk_fetch_request(q)) != NULL) {
1370 struct nullb_cmd *cmd = rq->special;
1372 spin_unlock_irq(q->queue_lock);
1373 null_handle_cmd(cmd);
1374 spin_lock_irq(q->queue_lock);
1378 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1379 const struct blk_mq_queue_data *bd)
1381 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1382 struct nullb_queue *nq = hctx->driver_data;
1384 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1386 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1387 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1388 cmd->timer.function = null_cmd_timer_expired;
1393 blk_mq_start_request(bd->rq);
1395 return null_handle_cmd(cmd);
1398 static const struct blk_mq_ops null_mq_ops = {
1399 .queue_rq = null_queue_rq,
1400 .complete = null_softirq_done_fn,
1403 static void cleanup_queue(struct nullb_queue *nq)
1409 static void cleanup_queues(struct nullb *nullb)
1413 for (i = 0; i < nullb->nr_queues; i++)
1414 cleanup_queue(&nullb->queues[i]);
1416 kfree(nullb->queues);
1421 static void null_lnvm_end_io(struct request *rq, blk_status_t status)
1423 struct nvm_rq *rqd = rq->end_io_data;
1425 /* XXX: lighnvm core seems to expect NVM_RSP_* values here.. */
1426 rqd->error = status ? -EIO : 0;
1429 blk_put_request(rq);
1432 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
1434 struct request_queue *q = dev->q;
1436 struct bio *bio = rqd->bio;
1438 rq = blk_mq_alloc_request(q,
1439 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1443 blk_init_request_from_bio(rq, bio);
1445 rq->end_io_data = rqd;
1447 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
1452 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
1454 struct nullb *nullb = dev->q->queuedata;
1455 sector_t size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1457 struct nvm_id_group *grp;
1464 id->ppaf.blk_offset = 0;
1465 id->ppaf.blk_len = 16;
1466 id->ppaf.pg_offset = 16;
1467 id->ppaf.pg_len = 16;
1468 id->ppaf.sect_offset = 32;
1469 id->ppaf.sect_len = 8;
1470 id->ppaf.pln_offset = 40;
1471 id->ppaf.pln_len = 8;
1472 id->ppaf.lun_offset = 48;
1473 id->ppaf.lun_len = 8;
1474 id->ppaf.ch_offset = 56;
1475 id->ppaf.ch_len = 8;
1477 sector_div(size, nullb->dev->blocksize); /* convert size to pages */
1478 size >>= 8; /* concert size to pgs pr blk */
1486 grp->num_lun = size + 1;
1487 sector_div(blksize, grp->num_lun);
1488 grp->num_blk = blksize;
1491 grp->fpg_sz = nullb->dev->blocksize;
1492 grp->csecs = nullb->dev->blocksize;
1497 grp->tbet = 1500000;
1498 grp->tbem = 1500000;
1499 grp->mpos = 0x010101; /* single plane rwe */
1500 grp->cpar = nullb->dev->hw_queue_depth;
1505 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
1507 mempool_t *virtmem_pool;
1509 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
1510 if (!virtmem_pool) {
1511 pr_err("null_blk: Unable to create virtual memory pool\n");
1515 return virtmem_pool;
1518 static void null_lnvm_destroy_dma_pool(void *pool)
1520 mempool_destroy(pool);
1523 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
1524 gfp_t mem_flags, dma_addr_t *dma_handler)
1526 return mempool_alloc(pool, mem_flags);
1529 static void null_lnvm_dev_dma_free(void *pool, void *entry,
1530 dma_addr_t dma_handler)
1532 mempool_free(entry, pool);
1535 static struct nvm_dev_ops null_lnvm_dev_ops = {
1536 .identity = null_lnvm_id,
1537 .submit_io = null_lnvm_submit_io,
1539 .create_dma_pool = null_lnvm_create_dma_pool,
1540 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
1541 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
1542 .dev_dma_free = null_lnvm_dev_dma_free,
1544 /* Simulate nvme protocol restriction */
1545 .max_phys_sect = 64,
1548 static int null_nvm_register(struct nullb *nullb)
1550 struct nvm_dev *dev;
1553 dev = nvm_alloc_dev(0);
1558 memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
1559 dev->ops = &null_lnvm_dev_ops;
1561 rv = nvm_register(dev);
1570 static void null_nvm_unregister(struct nullb *nullb)
1572 nvm_unregister(nullb->ndev);
1575 static int null_nvm_register(struct nullb *nullb)
1577 pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
1580 static void null_nvm_unregister(struct nullb *nullb) {}
1581 #endif /* CONFIG_NVM */
1583 static void null_del_dev(struct nullb *nullb)
1585 struct nullb_device *dev = nullb->dev;
1587 ida_simple_remove(&nullb_indexes, nullb->index);
1589 list_del_init(&nullb->list);
1591 if (dev->use_lightnvm)
1592 null_nvm_unregister(nullb);
1594 del_gendisk(nullb->disk);
1596 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1597 hrtimer_cancel(&nullb->bw_timer);
1598 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1599 null_restart_queue_async(nullb);
1602 blk_cleanup_queue(nullb->q);
1603 if (dev->queue_mode == NULL_Q_MQ &&
1604 nullb->tag_set == &nullb->__tag_set)
1605 blk_mq_free_tag_set(nullb->tag_set);
1606 if (!dev->use_lightnvm)
1607 put_disk(nullb->disk);
1608 cleanup_queues(nullb);
1609 if (null_cache_active(nullb))
1610 null_free_device_storage(nullb->dev, true);
1615 static void null_config_discard(struct nullb *nullb)
1617 if (nullb->dev->discard == false)
1619 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1620 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1621 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1622 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nullb->q);
1625 static int null_open(struct block_device *bdev, fmode_t mode)
1630 static void null_release(struct gendisk *disk, fmode_t mode)
1634 static const struct block_device_operations null_fops = {
1635 .owner = THIS_MODULE,
1637 .release = null_release,
1640 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1645 init_waitqueue_head(&nq->wait);
1646 nq->queue_depth = nullb->queue_depth;
1647 nq->dev = nullb->dev;
1650 static void null_init_queues(struct nullb *nullb)
1652 struct request_queue *q = nullb->q;
1653 struct blk_mq_hw_ctx *hctx;
1654 struct nullb_queue *nq;
1657 queue_for_each_hw_ctx(q, hctx, i) {
1658 if (!hctx->nr_ctx || !hctx->tags)
1660 nq = &nullb->queues[i];
1661 hctx->driver_data = nq;
1662 null_init_queue(nullb, nq);
1667 static int setup_commands(struct nullb_queue *nq)
1669 struct nullb_cmd *cmd;
1672 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
1676 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1677 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
1683 for (i = 0; i < nq->queue_depth; i++) {
1685 INIT_LIST_HEAD(&cmd->list);
1686 cmd->ll_list.next = NULL;
1693 static int setup_queues(struct nullb *nullb)
1695 nullb->queues = kzalloc(nullb->dev->submit_queues *
1696 sizeof(struct nullb_queue), GFP_KERNEL);
1700 nullb->nr_queues = 0;
1701 nullb->queue_depth = nullb->dev->hw_queue_depth;
1706 static int init_driver_queues(struct nullb *nullb)
1708 struct nullb_queue *nq;
1711 for (i = 0; i < nullb->dev->submit_queues; i++) {
1712 nq = &nullb->queues[i];
1714 null_init_queue(nullb, nq);
1716 ret = setup_commands(nq);
1724 static int null_gendisk_register(struct nullb *nullb)
1726 struct gendisk *disk;
1729 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1732 size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1733 set_capacity(disk, size >> 9);
1735 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1736 disk->major = null_major;
1737 disk->first_minor = nullb->index;
1738 disk->fops = &null_fops;
1739 disk->private_data = nullb;
1740 disk->queue = nullb->q;
1741 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1747 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1749 set->ops = &null_mq_ops;
1750 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1752 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1754 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1755 set->cmd_size = sizeof(struct nullb_cmd);
1756 set->flags = BLK_MQ_F_SHOULD_MERGE;
1757 set->driver_data = NULL;
1759 if ((nullb && nullb->dev->blocking) || g_blocking)
1760 set->flags |= BLK_MQ_F_BLOCKING;
1762 return blk_mq_alloc_tag_set(set);
1765 static void null_validate_conf(struct nullb_device *dev)
1767 dev->blocksize = round_down(dev->blocksize, 512);
1768 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1769 if (dev->use_lightnvm && dev->blocksize != 4096)
1770 dev->blocksize = 4096;
1772 if (dev->use_lightnvm && dev->queue_mode != NULL_Q_MQ)
1773 dev->queue_mode = NULL_Q_MQ;
1775 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1776 if (dev->submit_queues != nr_online_nodes)
1777 dev->submit_queues = nr_online_nodes;
1778 } else if (dev->submit_queues > nr_cpu_ids)
1779 dev->submit_queues = nr_cpu_ids;
1780 else if (dev->submit_queues == 0)
1781 dev->submit_queues = 1;
1783 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1784 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1786 /* Do memory allocation, so set blocking */
1787 if (dev->memory_backed)
1788 dev->blocking = true;
1789 else /* cache is meaningless */
1790 dev->cache_size = 0;
1791 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1793 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1794 /* can not stop a queue */
1795 if (dev->queue_mode == NULL_Q_BIO)
1799 static int null_add_dev(struct nullb_device *dev)
1801 struct nullb *nullb;
1804 null_validate_conf(dev);
1806 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1814 spin_lock_init(&nullb->lock);
1816 rv = setup_queues(nullb);
1818 goto out_free_nullb;
1820 if (dev->queue_mode == NULL_Q_MQ) {
1822 nullb->tag_set = &tag_set;
1825 nullb->tag_set = &nullb->__tag_set;
1826 rv = null_init_tag_set(nullb, nullb->tag_set);
1830 goto out_cleanup_queues;
1832 nullb->q = blk_mq_init_queue(nullb->tag_set);
1833 if (IS_ERR(nullb->q)) {
1835 goto out_cleanup_tags;
1837 null_init_queues(nullb);
1838 } else if (dev->queue_mode == NULL_Q_BIO) {
1839 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node);
1842 goto out_cleanup_queues;
1844 blk_queue_make_request(nullb->q, null_queue_bio);
1845 rv = init_driver_queues(nullb);
1847 goto out_cleanup_blk_queue;
1849 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock,
1853 goto out_cleanup_queues;
1855 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
1856 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
1857 rv = init_driver_queues(nullb);
1859 goto out_cleanup_blk_queue;
1863 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1864 nullb_setup_bwtimer(nullb);
1867 if (dev->cache_size > 0) {
1868 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1869 blk_queue_write_cache(nullb->q, true, true);
1870 blk_queue_flush_queueable(nullb->q, true);
1873 nullb->q->queuedata = nullb;
1874 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
1875 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1878 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1879 dev->index = nullb->index;
1880 mutex_unlock(&lock);
1882 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1883 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1885 null_config_discard(nullb);
1887 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1889 if (dev->use_lightnvm)
1890 rv = null_nvm_register(nullb);
1892 rv = null_gendisk_register(nullb);
1895 goto out_cleanup_blk_queue;
1898 list_add_tail(&nullb->list, &nullb_list);
1899 mutex_unlock(&lock);
1902 out_cleanup_blk_queue:
1903 blk_cleanup_queue(nullb->q);
1905 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1906 blk_mq_free_tag_set(nullb->tag_set);
1908 cleanup_queues(nullb);
1915 static int __init null_init(void)
1919 struct nullb *nullb;
1920 struct nullb_device *dev;
1922 /* check for nullb_page.bitmap */
1923 if (sizeof(unsigned long) * 8 - 2 < (PAGE_SIZE >> SECTOR_SHIFT))
1926 if (g_bs > PAGE_SIZE) {
1927 pr_warn("null_blk: invalid block size\n");
1928 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
1932 if (g_use_lightnvm && g_bs != 4096) {
1933 pr_warn("null_blk: LightNVM only supports 4k block size\n");
1934 pr_warn("null_blk: defaults block size to 4k\n");
1938 if (g_use_lightnvm && g_queue_mode != NULL_Q_MQ) {
1939 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
1940 pr_warn("null_blk: defaults queue mode to blk-mq\n");
1941 g_queue_mode = NULL_Q_MQ;
1944 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1945 if (g_submit_queues != nr_online_nodes) {
1946 pr_warn("null_blk: submit_queues param is set to %u.\n",
1948 g_submit_queues = nr_online_nodes;
1950 } else if (g_submit_queues > nr_cpu_ids)
1951 g_submit_queues = nr_cpu_ids;
1952 else if (g_submit_queues <= 0)
1953 g_submit_queues = 1;
1955 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1956 ret = null_init_tag_set(NULL, &tag_set);
1961 config_group_init(&nullb_subsys.su_group);
1962 mutex_init(&nullb_subsys.su_mutex);
1964 ret = configfs_register_subsystem(&nullb_subsys);
1970 null_major = register_blkdev(0, "nullb");
1971 if (null_major < 0) {
1976 if (g_use_lightnvm) {
1977 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
1980 pr_err("null_blk: unable to create ppa cache\n");
1986 for (i = 0; i < nr_devices; i++) {
1987 dev = null_alloc_dev();
1990 ret = null_add_dev(dev);
1997 pr_info("null: module loaded\n");
2001 while (!list_empty(&nullb_list)) {
2002 nullb = list_entry(nullb_list.next, struct nullb, list);
2004 null_del_dev(nullb);
2007 kmem_cache_destroy(ppa_cache);
2009 unregister_blkdev(null_major, "nullb");
2011 configfs_unregister_subsystem(&nullb_subsys);
2013 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2014 blk_mq_free_tag_set(&tag_set);
2018 static void __exit null_exit(void)
2020 struct nullb *nullb;
2022 configfs_unregister_subsystem(&nullb_subsys);
2024 unregister_blkdev(null_major, "nullb");
2027 while (!list_empty(&nullb_list)) {
2028 struct nullb_device *dev;
2030 nullb = list_entry(nullb_list.next, struct nullb, list);
2032 null_del_dev(nullb);
2035 mutex_unlock(&lock);
2037 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2038 blk_mq_free_tag_set(&tag_set);
2040 kmem_cache_destroy(ppa_cache);
2043 module_init(null_init);
2044 module_exit(null_exit);
2046 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2047 MODULE_LICENSE("GPL");