1 #include <linux/module.h>
3 #include <linux/moduleparam.h>
4 #include <linux/sched.h>
6 #include <linux/blkdev.h>
7 #include <linux/init.h>
8 #include <linux/slab.h>
9 #include <linux/blk-mq.h>
10 #include <linux/hrtimer.h>
11 #include <linux/lightnvm.h>
14 struct list_head list;
15 struct llist_node ll_list;
16 struct call_single_data csd;
20 struct nullb_queue *nq;
25 unsigned long *tag_map;
26 wait_queue_head_t wait;
27 unsigned int queue_depth;
29 struct nullb_cmd *cmds;
33 struct list_head list;
35 struct request_queue *q;
38 struct blk_mq_tag_set *tag_set;
39 struct blk_mq_tag_set __tag_set;
41 unsigned int queue_depth;
44 struct nullb_queue *queues;
45 unsigned int nr_queues;
46 char disk_name[DISK_NAME_LEN];
49 static LIST_HEAD(nullb_list);
50 static struct mutex lock;
51 static int null_major;
52 static int nullb_indexes;
53 static struct kmem_cache *ppa_cache;
54 static struct blk_mq_tag_set tag_set;
68 static int submit_queues;
69 module_param(submit_queues, int, S_IRUGO);
70 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
72 static int home_node = NUMA_NO_NODE;
73 module_param(home_node, int, S_IRUGO);
74 MODULE_PARM_DESC(home_node, "Home node for the device");
76 static int queue_mode = NULL_Q_MQ;
78 static int null_param_store_val(const char *str, int *val, int min, int max)
82 ret = kstrtoint(str, 10, &new_val);
86 if (new_val < min || new_val > max)
93 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
95 return null_param_store_val(str, &queue_mode, NULL_Q_BIO, NULL_Q_MQ);
98 static const struct kernel_param_ops null_queue_mode_param_ops = {
99 .set = null_set_queue_mode,
100 .get = param_get_int,
103 device_param_cb(queue_mode, &null_queue_mode_param_ops, &queue_mode, S_IRUGO);
104 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
107 module_param(gb, int, S_IRUGO);
108 MODULE_PARM_DESC(gb, "Size in GB");
111 module_param(bs, int, S_IRUGO);
112 MODULE_PARM_DESC(bs, "Block size (in bytes)");
114 static int nr_devices = 1;
115 module_param(nr_devices, int, S_IRUGO);
116 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
118 static bool use_lightnvm;
119 module_param(use_lightnvm, bool, S_IRUGO);
120 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
122 static bool blocking;
123 module_param(blocking, bool, S_IRUGO);
124 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
126 static bool shared_tags;
127 module_param(shared_tags, bool, S_IRUGO);
128 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
130 static int irqmode = NULL_IRQ_SOFTIRQ;
132 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
134 return null_param_store_val(str, &irqmode, NULL_IRQ_NONE,
138 static const struct kernel_param_ops null_irqmode_param_ops = {
139 .set = null_set_irqmode,
140 .get = param_get_int,
143 device_param_cb(irqmode, &null_irqmode_param_ops, &irqmode, S_IRUGO);
144 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
146 static unsigned long completion_nsec = 10000;
147 module_param(completion_nsec, ulong, S_IRUGO);
148 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
150 static int hw_queue_depth = 64;
151 module_param(hw_queue_depth, int, S_IRUGO);
152 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
154 static bool use_per_node_hctx = false;
155 module_param(use_per_node_hctx, bool, S_IRUGO);
156 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
158 static void put_tag(struct nullb_queue *nq, unsigned int tag)
160 clear_bit_unlock(tag, nq->tag_map);
162 if (waitqueue_active(&nq->wait))
166 static unsigned int get_tag(struct nullb_queue *nq)
171 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
172 if (tag >= nq->queue_depth)
174 } while (test_and_set_bit_lock(tag, nq->tag_map));
179 static void free_cmd(struct nullb_cmd *cmd)
181 put_tag(cmd->nq, cmd->tag);
184 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
186 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
188 struct nullb_cmd *cmd;
193 cmd = &nq->cmds[tag];
196 if (irqmode == NULL_IRQ_TIMER) {
197 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
199 cmd->timer.function = null_cmd_timer_expired;
207 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
209 struct nullb_cmd *cmd;
212 cmd = __alloc_cmd(nq);
213 if (cmd || !can_wait)
217 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
218 cmd = __alloc_cmd(nq);
225 finish_wait(&nq->wait, &wait);
229 static void end_cmd(struct nullb_cmd *cmd)
231 struct request_queue *q = NULL;
236 switch (queue_mode) {
238 blk_mq_end_request(cmd->rq, BLK_STS_OK);
241 INIT_LIST_HEAD(&cmd->rq->queuelist);
242 blk_end_request_all(cmd->rq, BLK_STS_OK);
251 /* Restart queue if needed, as we are freeing a tag */
252 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
255 spin_lock_irqsave(q->queue_lock, flags);
256 blk_start_queue_async(q);
257 spin_unlock_irqrestore(q->queue_lock, flags);
261 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
263 end_cmd(container_of(timer, struct nullb_cmd, timer));
265 return HRTIMER_NORESTART;
268 static void null_cmd_end_timer(struct nullb_cmd *cmd)
270 ktime_t kt = completion_nsec;
272 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
275 static void null_softirq_done_fn(struct request *rq)
277 if (queue_mode == NULL_Q_MQ)
278 end_cmd(blk_mq_rq_to_pdu(rq));
280 end_cmd(rq->special);
283 static inline void null_handle_cmd(struct nullb_cmd *cmd)
285 /* Complete IO by inline, softirq or timer */
287 case NULL_IRQ_SOFTIRQ:
288 switch (queue_mode) {
290 blk_mq_complete_request(cmd->rq);
293 blk_complete_request(cmd->rq);
297 * XXX: no proper submitting cpu information available.
307 null_cmd_end_timer(cmd);
312 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
316 if (nullb->nr_queues != 1)
317 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
319 return &nullb->queues[index];
322 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
324 struct nullb *nullb = q->queuedata;
325 struct nullb_queue *nq = nullb_to_queue(nullb);
326 struct nullb_cmd *cmd;
328 cmd = alloc_cmd(nq, 1);
331 null_handle_cmd(cmd);
332 return BLK_QC_T_NONE;
335 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
337 struct nullb *nullb = q->queuedata;
338 struct nullb_queue *nq = nullb_to_queue(nullb);
339 struct nullb_cmd *cmd;
341 cmd = alloc_cmd(nq, 0);
349 return BLKPREP_DEFER;
352 static void null_request_fn(struct request_queue *q)
356 while ((rq = blk_fetch_request(q)) != NULL) {
357 struct nullb_cmd *cmd = rq->special;
359 spin_unlock_irq(q->queue_lock);
360 null_handle_cmd(cmd);
361 spin_lock_irq(q->queue_lock);
365 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
366 const struct blk_mq_queue_data *bd)
368 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
370 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
372 if (irqmode == NULL_IRQ_TIMER) {
373 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
374 cmd->timer.function = null_cmd_timer_expired;
377 cmd->nq = hctx->driver_data;
379 blk_mq_start_request(bd->rq);
381 null_handle_cmd(cmd);
385 static const struct blk_mq_ops null_mq_ops = {
386 .queue_rq = null_queue_rq,
387 .complete = null_softirq_done_fn,
390 static void cleanup_queue(struct nullb_queue *nq)
396 static void cleanup_queues(struct nullb *nullb)
400 for (i = 0; i < nullb->nr_queues; i++)
401 cleanup_queue(&nullb->queues[i]);
403 kfree(nullb->queues);
408 static void null_lnvm_end_io(struct request *rq, blk_status_t status)
410 struct nvm_rq *rqd = rq->end_io_data;
412 /* XXX: lighnvm core seems to expect NVM_RSP_* values here.. */
413 rqd->error = status ? -EIO : 0;
419 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
421 struct request_queue *q = dev->q;
423 struct bio *bio = rqd->bio;
425 rq = blk_mq_alloc_request(q,
426 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
430 blk_init_request_from_bio(rq, bio);
432 rq->end_io_data = rqd;
434 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
439 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
441 sector_t size = gb * 1024 * 1024 * 1024ULL;
443 struct nvm_id_group *grp;
450 id->ppaf.blk_offset = 0;
451 id->ppaf.blk_len = 16;
452 id->ppaf.pg_offset = 16;
453 id->ppaf.pg_len = 16;
454 id->ppaf.sect_offset = 32;
455 id->ppaf.sect_len = 8;
456 id->ppaf.pln_offset = 40;
457 id->ppaf.pln_len = 8;
458 id->ppaf.lun_offset = 48;
459 id->ppaf.lun_len = 8;
460 id->ppaf.ch_offset = 56;
463 sector_div(size, bs); /* convert size to pages */
464 size >>= 8; /* concert size to pgs pr blk */
472 grp->num_lun = size + 1;
473 sector_div(blksize, grp->num_lun);
474 grp->num_blk = blksize;
485 grp->mpos = 0x010101; /* single plane rwe */
486 grp->cpar = hw_queue_depth;
491 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
493 mempool_t *virtmem_pool;
495 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
497 pr_err("null_blk: Unable to create virtual memory pool\n");
504 static void null_lnvm_destroy_dma_pool(void *pool)
506 mempool_destroy(pool);
509 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
510 gfp_t mem_flags, dma_addr_t *dma_handler)
512 return mempool_alloc(pool, mem_flags);
515 static void null_lnvm_dev_dma_free(void *pool, void *entry,
516 dma_addr_t dma_handler)
518 mempool_free(entry, pool);
521 static struct nvm_dev_ops null_lnvm_dev_ops = {
522 .identity = null_lnvm_id,
523 .submit_io = null_lnvm_submit_io,
525 .create_dma_pool = null_lnvm_create_dma_pool,
526 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
527 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
528 .dev_dma_free = null_lnvm_dev_dma_free,
530 /* Simulate nvme protocol restriction */
534 static int null_nvm_register(struct nullb *nullb)
539 dev = nvm_alloc_dev(0);
544 memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
545 dev->ops = &null_lnvm_dev_ops;
547 rv = nvm_register(dev);
556 static void null_nvm_unregister(struct nullb *nullb)
558 nvm_unregister(nullb->ndev);
561 static int null_nvm_register(struct nullb *nullb)
563 pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
566 static void null_nvm_unregister(struct nullb *nullb) {}
567 #endif /* CONFIG_NVM */
569 static void null_del_dev(struct nullb *nullb)
571 list_del_init(&nullb->list);
574 null_nvm_unregister(nullb);
576 del_gendisk(nullb->disk);
577 blk_cleanup_queue(nullb->q);
578 if (queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
579 blk_mq_free_tag_set(nullb->tag_set);
581 put_disk(nullb->disk);
582 cleanup_queues(nullb);
586 static int null_open(struct block_device *bdev, fmode_t mode)
591 static void null_release(struct gendisk *disk, fmode_t mode)
595 static const struct block_device_operations null_fops = {
596 .owner = THIS_MODULE,
598 .release = null_release,
601 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
606 init_waitqueue_head(&nq->wait);
607 nq->queue_depth = nullb->queue_depth;
610 static void null_init_queues(struct nullb *nullb)
612 struct request_queue *q = nullb->q;
613 struct blk_mq_hw_ctx *hctx;
614 struct nullb_queue *nq;
617 queue_for_each_hw_ctx(q, hctx, i) {
618 if (!hctx->nr_ctx || !hctx->tags)
620 nq = &nullb->queues[i];
621 hctx->driver_data = nq;
622 null_init_queue(nullb, nq);
627 static int setup_commands(struct nullb_queue *nq)
629 struct nullb_cmd *cmd;
632 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
636 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
637 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
643 for (i = 0; i < nq->queue_depth; i++) {
645 INIT_LIST_HEAD(&cmd->list);
646 cmd->ll_list.next = NULL;
653 static int setup_queues(struct nullb *nullb)
655 nullb->queues = kzalloc(submit_queues * sizeof(struct nullb_queue),
660 nullb->nr_queues = 0;
661 nullb->queue_depth = hw_queue_depth;
666 static int init_driver_queues(struct nullb *nullb)
668 struct nullb_queue *nq;
671 for (i = 0; i < submit_queues; i++) {
672 nq = &nullb->queues[i];
674 null_init_queue(nullb, nq);
676 ret = setup_commands(nq);
684 static int null_gendisk_register(struct nullb *nullb)
686 struct gendisk *disk;
689 disk = nullb->disk = alloc_disk_node(1, home_node);
692 size = gb * 1024 * 1024 * 1024ULL;
693 set_capacity(disk, size >> 9);
695 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
696 disk->major = null_major;
697 disk->first_minor = nullb->index;
698 disk->fops = &null_fops;
699 disk->private_data = nullb;
700 disk->queue = nullb->q;
701 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
707 static int null_init_tag_set(struct blk_mq_tag_set *set)
709 set->ops = &null_mq_ops;
710 set->nr_hw_queues = submit_queues;
711 set->queue_depth = hw_queue_depth;
712 set->numa_node = home_node;
713 set->cmd_size = sizeof(struct nullb_cmd);
714 set->flags = BLK_MQ_F_SHOULD_MERGE;
715 set->driver_data = NULL;
718 set->flags |= BLK_MQ_F_BLOCKING;
720 return blk_mq_alloc_tag_set(set);
723 static int null_add_dev(void)
728 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
734 spin_lock_init(&nullb->lock);
736 if (queue_mode == NULL_Q_MQ && use_per_node_hctx)
737 submit_queues = nr_online_nodes;
739 rv = setup_queues(nullb);
743 if (queue_mode == NULL_Q_MQ) {
745 nullb->tag_set = &tag_set;
748 nullb->tag_set = &nullb->__tag_set;
749 rv = null_init_tag_set(nullb->tag_set);
753 goto out_cleanup_queues;
755 nullb->q = blk_mq_init_queue(nullb->tag_set);
756 if (IS_ERR(nullb->q)) {
758 goto out_cleanup_tags;
760 null_init_queues(nullb);
761 } else if (queue_mode == NULL_Q_BIO) {
762 nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
765 goto out_cleanup_queues;
767 blk_queue_make_request(nullb->q, null_queue_bio);
768 rv = init_driver_queues(nullb);
770 goto out_cleanup_blk_queue;
772 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
775 goto out_cleanup_queues;
777 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
778 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
779 rv = init_driver_queues(nullb);
781 goto out_cleanup_blk_queue;
784 nullb->q->queuedata = nullb;
785 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
786 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
789 nullb->index = nullb_indexes++;
792 blk_queue_logical_block_size(nullb->q, bs);
793 blk_queue_physical_block_size(nullb->q, bs);
795 sprintf(nullb->disk_name, "nullb%d", nullb->index);
798 rv = null_nvm_register(nullb);
800 rv = null_gendisk_register(nullb);
803 goto out_cleanup_blk_queue;
806 list_add_tail(&nullb->list, &nullb_list);
810 out_cleanup_blk_queue:
811 blk_cleanup_queue(nullb->q);
813 if (queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
814 blk_mq_free_tag_set(nullb->tag_set);
816 cleanup_queues(nullb);
823 static int __init null_init(void)
829 if (bs > PAGE_SIZE) {
830 pr_warn("null_blk: invalid block size\n");
831 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
835 if (use_lightnvm && bs != 4096) {
836 pr_warn("null_blk: LightNVM only supports 4k block size\n");
837 pr_warn("null_blk: defaults block size to 4k\n");
841 if (use_lightnvm && queue_mode != NULL_Q_MQ) {
842 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
843 pr_warn("null_blk: defaults queue mode to blk-mq\n");
844 queue_mode = NULL_Q_MQ;
847 if (queue_mode == NULL_Q_MQ && shared_tags)
848 null_init_tag_set(&tag_set);
850 if (queue_mode == NULL_Q_MQ && use_per_node_hctx) {
851 if (submit_queues < nr_online_nodes) {
852 pr_warn("null_blk: submit_queues param is set to %u.",
854 submit_queues = nr_online_nodes;
856 } else if (submit_queues > nr_cpu_ids)
857 submit_queues = nr_cpu_ids;
858 else if (!submit_queues)
863 null_major = register_blkdev(0, "nullb");
868 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
871 pr_err("null_blk: unable to create ppa cache\n");
877 for (i = 0; i < nr_devices; i++) {
878 ret = null_add_dev();
883 pr_info("null: module loaded\n");
887 while (!list_empty(&nullb_list)) {
888 nullb = list_entry(nullb_list.next, struct nullb, list);
891 kmem_cache_destroy(ppa_cache);
893 unregister_blkdev(null_major, "nullb");
897 static void __exit null_exit(void)
901 unregister_blkdev(null_major, "nullb");
904 while (!list_empty(&nullb_list)) {
905 nullb = list_entry(nullb_list.next, struct nullb, list);
910 if (queue_mode == NULL_Q_MQ && shared_tags)
911 blk_mq_free_tag_set(&tag_set);
913 kmem_cache_destroy(ppa_cache);
916 module_init(null_init);
917 module_exit(null_exit);
919 MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
920 MODULE_LICENSE("GPL");