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;
40 unsigned int queue_depth;
43 struct nullb_queue *queues;
44 unsigned int nr_queues;
45 char disk_name[DISK_NAME_LEN];
48 static LIST_HEAD(nullb_list);
49 static struct mutex lock;
50 static int null_major;
51 static int nullb_indexes;
52 static struct kmem_cache *ppa_cache;
66 static int submit_queues;
67 module_param(submit_queues, int, S_IRUGO);
68 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
70 static int home_node = NUMA_NO_NODE;
71 module_param(home_node, int, S_IRUGO);
72 MODULE_PARM_DESC(home_node, "Home node for the device");
74 static int queue_mode = NULL_Q_MQ;
76 static int null_param_store_val(const char *str, int *val, int min, int max)
80 ret = kstrtoint(str, 10, &new_val);
84 if (new_val < min || new_val > max)
91 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
93 return null_param_store_val(str, &queue_mode, NULL_Q_BIO, NULL_Q_MQ);
96 static const struct kernel_param_ops null_queue_mode_param_ops = {
97 .set = null_set_queue_mode,
101 device_param_cb(queue_mode, &null_queue_mode_param_ops, &queue_mode, S_IRUGO);
102 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
105 module_param(gb, int, S_IRUGO);
106 MODULE_PARM_DESC(gb, "Size in GB");
109 module_param(bs, int, S_IRUGO);
110 MODULE_PARM_DESC(bs, "Block size (in bytes)");
112 static int nr_devices = 2;
113 module_param(nr_devices, int, S_IRUGO);
114 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
116 static bool use_lightnvm;
117 module_param(use_lightnvm, bool, S_IRUGO);
118 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
120 static bool blocking;
121 module_param(blocking, bool, S_IRUGO);
122 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
124 static int irqmode = NULL_IRQ_SOFTIRQ;
126 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
128 return null_param_store_val(str, &irqmode, NULL_IRQ_NONE,
132 static const struct kernel_param_ops null_irqmode_param_ops = {
133 .set = null_set_irqmode,
134 .get = param_get_int,
137 device_param_cb(irqmode, &null_irqmode_param_ops, &irqmode, S_IRUGO);
138 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
140 static unsigned long completion_nsec = 10000;
141 module_param(completion_nsec, ulong, S_IRUGO);
142 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
144 static int hw_queue_depth = 64;
145 module_param(hw_queue_depth, int, S_IRUGO);
146 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
148 static bool use_per_node_hctx = false;
149 module_param(use_per_node_hctx, bool, S_IRUGO);
150 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
152 static void put_tag(struct nullb_queue *nq, unsigned int tag)
154 clear_bit_unlock(tag, nq->tag_map);
156 if (waitqueue_active(&nq->wait))
160 static unsigned int get_tag(struct nullb_queue *nq)
165 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
166 if (tag >= nq->queue_depth)
168 } while (test_and_set_bit_lock(tag, nq->tag_map));
173 static void free_cmd(struct nullb_cmd *cmd)
175 put_tag(cmd->nq, cmd->tag);
178 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
180 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
182 struct nullb_cmd *cmd;
187 cmd = &nq->cmds[tag];
190 if (irqmode == NULL_IRQ_TIMER) {
191 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
193 cmd->timer.function = null_cmd_timer_expired;
201 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
203 struct nullb_cmd *cmd;
206 cmd = __alloc_cmd(nq);
207 if (cmd || !can_wait)
211 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
212 cmd = __alloc_cmd(nq);
219 finish_wait(&nq->wait, &wait);
223 static void end_cmd(struct nullb_cmd *cmd)
225 struct request_queue *q = NULL;
230 switch (queue_mode) {
232 blk_mq_end_request(cmd->rq, 0);
235 INIT_LIST_HEAD(&cmd->rq->queuelist);
236 blk_end_request_all(cmd->rq, 0);
245 /* Restart queue if needed, as we are freeing a tag */
246 if (queue_mode == NULL_Q_RQ && blk_queue_stopped(q)) {
249 spin_lock_irqsave(q->queue_lock, flags);
250 blk_start_queue_async(q);
251 spin_unlock_irqrestore(q->queue_lock, flags);
255 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
257 end_cmd(container_of(timer, struct nullb_cmd, timer));
259 return HRTIMER_NORESTART;
262 static void null_cmd_end_timer(struct nullb_cmd *cmd)
264 ktime_t kt = completion_nsec;
266 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
269 static void null_softirq_done_fn(struct request *rq)
271 if (queue_mode == NULL_Q_MQ)
272 end_cmd(blk_mq_rq_to_pdu(rq));
274 end_cmd(rq->special);
277 static inline void null_handle_cmd(struct nullb_cmd *cmd)
279 /* Complete IO by inline, softirq or timer */
281 case NULL_IRQ_SOFTIRQ:
282 switch (queue_mode) {
284 blk_mq_complete_request(cmd->rq);
287 blk_complete_request(cmd->rq);
291 * XXX: no proper submitting cpu information available.
301 null_cmd_end_timer(cmd);
306 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
310 if (nullb->nr_queues != 1)
311 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
313 return &nullb->queues[index];
316 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
318 struct nullb *nullb = q->queuedata;
319 struct nullb_queue *nq = nullb_to_queue(nullb);
320 struct nullb_cmd *cmd;
322 cmd = alloc_cmd(nq, 1);
325 null_handle_cmd(cmd);
326 return BLK_QC_T_NONE;
329 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
331 struct nullb *nullb = q->queuedata;
332 struct nullb_queue *nq = nullb_to_queue(nullb);
333 struct nullb_cmd *cmd;
335 cmd = alloc_cmd(nq, 0);
343 return BLKPREP_DEFER;
346 static void null_request_fn(struct request_queue *q)
350 while ((rq = blk_fetch_request(q)) != NULL) {
351 struct nullb_cmd *cmd = rq->special;
353 spin_unlock_irq(q->queue_lock);
354 null_handle_cmd(cmd);
355 spin_lock_irq(q->queue_lock);
359 static int null_queue_rq(struct blk_mq_hw_ctx *hctx,
360 const struct blk_mq_queue_data *bd)
362 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
364 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
366 if (irqmode == NULL_IRQ_TIMER) {
367 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
368 cmd->timer.function = null_cmd_timer_expired;
371 cmd->nq = hctx->driver_data;
373 blk_mq_start_request(bd->rq);
375 null_handle_cmd(cmd);
376 return BLK_MQ_RQ_QUEUE_OK;
379 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
384 init_waitqueue_head(&nq->wait);
385 nq->queue_depth = nullb->queue_depth;
388 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
391 struct nullb *nullb = data;
392 struct nullb_queue *nq = &nullb->queues[index];
394 hctx->driver_data = nq;
395 null_init_queue(nullb, nq);
401 static const struct blk_mq_ops null_mq_ops = {
402 .queue_rq = null_queue_rq,
403 .init_hctx = null_init_hctx,
404 .complete = null_softirq_done_fn,
407 static void cleanup_queue(struct nullb_queue *nq)
413 static void cleanup_queues(struct nullb *nullb)
417 for (i = 0; i < nullb->nr_queues; i++)
418 cleanup_queue(&nullb->queues[i]);
420 kfree(nullb->queues);
425 static void null_lnvm_end_io(struct request *rq, int error)
427 struct nvm_rq *rqd = rq->end_io_data;
435 static int null_lnvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
437 struct request_queue *q = dev->q;
439 struct bio *bio = rqd->bio;
441 rq = blk_mq_alloc_request(q,
442 op_is_write(bio_op(bio)) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
446 blk_init_request_from_bio(rq, bio);
448 rq->end_io_data = rqd;
450 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
455 static int null_lnvm_id(struct nvm_dev *dev, struct nvm_id *id)
457 sector_t size = gb * 1024 * 1024 * 1024ULL;
459 struct nvm_id_group *grp;
466 id->ppaf.blk_offset = 0;
467 id->ppaf.blk_len = 16;
468 id->ppaf.pg_offset = 16;
469 id->ppaf.pg_len = 16;
470 id->ppaf.sect_offset = 32;
471 id->ppaf.sect_len = 8;
472 id->ppaf.pln_offset = 40;
473 id->ppaf.pln_len = 8;
474 id->ppaf.lun_offset = 48;
475 id->ppaf.lun_len = 8;
476 id->ppaf.ch_offset = 56;
479 sector_div(size, bs); /* convert size to pages */
480 size >>= 8; /* concert size to pgs pr blk */
488 grp->num_lun = size + 1;
489 sector_div(blksize, grp->num_lun);
490 grp->num_blk = blksize;
501 grp->mpos = 0x010101; /* single plane rwe */
502 grp->cpar = hw_queue_depth;
507 static void *null_lnvm_create_dma_pool(struct nvm_dev *dev, char *name)
509 mempool_t *virtmem_pool;
511 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
513 pr_err("null_blk: Unable to create virtual memory pool\n");
520 static void null_lnvm_destroy_dma_pool(void *pool)
522 mempool_destroy(pool);
525 static void *null_lnvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
526 gfp_t mem_flags, dma_addr_t *dma_handler)
528 return mempool_alloc(pool, mem_flags);
531 static void null_lnvm_dev_dma_free(void *pool, void *entry,
532 dma_addr_t dma_handler)
534 mempool_free(entry, pool);
537 static struct nvm_dev_ops null_lnvm_dev_ops = {
538 .identity = null_lnvm_id,
539 .submit_io = null_lnvm_submit_io,
541 .create_dma_pool = null_lnvm_create_dma_pool,
542 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
543 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
544 .dev_dma_free = null_lnvm_dev_dma_free,
546 /* Simulate nvme protocol restriction */
550 static int null_nvm_register(struct nullb *nullb)
555 dev = nvm_alloc_dev(0);
560 memcpy(dev->name, nullb->disk_name, DISK_NAME_LEN);
561 dev->ops = &null_lnvm_dev_ops;
563 rv = nvm_register(dev);
572 static void null_nvm_unregister(struct nullb *nullb)
574 nvm_unregister(nullb->ndev);
577 static int null_nvm_register(struct nullb *nullb)
579 pr_err("null_blk: CONFIG_NVM needs to be enabled for LightNVM\n");
582 static void null_nvm_unregister(struct nullb *nullb) {}
583 #endif /* CONFIG_NVM */
585 static void null_del_dev(struct nullb *nullb)
587 list_del_init(&nullb->list);
590 null_nvm_unregister(nullb);
592 del_gendisk(nullb->disk);
593 blk_cleanup_queue(nullb->q);
594 if (queue_mode == NULL_Q_MQ)
595 blk_mq_free_tag_set(&nullb->tag_set);
597 put_disk(nullb->disk);
598 cleanup_queues(nullb);
602 static int null_open(struct block_device *bdev, fmode_t mode)
607 static void null_release(struct gendisk *disk, fmode_t mode)
611 static const struct block_device_operations null_fops = {
612 .owner = THIS_MODULE,
614 .release = null_release,
617 static int setup_commands(struct nullb_queue *nq)
619 struct nullb_cmd *cmd;
622 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
626 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
627 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
633 for (i = 0; i < nq->queue_depth; i++) {
635 INIT_LIST_HEAD(&cmd->list);
636 cmd->ll_list.next = NULL;
643 static int setup_queues(struct nullb *nullb)
645 nullb->queues = kzalloc(submit_queues * sizeof(struct nullb_queue),
650 nullb->nr_queues = 0;
651 nullb->queue_depth = hw_queue_depth;
656 static int init_driver_queues(struct nullb *nullb)
658 struct nullb_queue *nq;
661 for (i = 0; i < submit_queues; i++) {
662 nq = &nullb->queues[i];
664 null_init_queue(nullb, nq);
666 ret = setup_commands(nq);
674 static int null_gendisk_register(struct nullb *nullb)
676 struct gendisk *disk;
679 disk = nullb->disk = alloc_disk_node(1, home_node);
682 size = gb * 1024 * 1024 * 1024ULL;
683 set_capacity(disk, size >> 9);
685 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
686 disk->major = null_major;
687 disk->first_minor = nullb->index;
688 disk->fops = &null_fops;
689 disk->private_data = nullb;
690 disk->queue = nullb->q;
691 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
697 static int null_add_dev(void)
702 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
708 spin_lock_init(&nullb->lock);
710 if (queue_mode == NULL_Q_MQ && use_per_node_hctx)
711 submit_queues = nr_online_nodes;
713 rv = setup_queues(nullb);
717 if (queue_mode == NULL_Q_MQ) {
718 nullb->tag_set.ops = &null_mq_ops;
719 nullb->tag_set.nr_hw_queues = submit_queues;
720 nullb->tag_set.queue_depth = hw_queue_depth;
721 nullb->tag_set.numa_node = home_node;
722 nullb->tag_set.cmd_size = sizeof(struct nullb_cmd);
723 nullb->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
724 nullb->tag_set.driver_data = nullb;
727 nullb->tag_set.flags |= BLK_MQ_F_BLOCKING;
729 rv = blk_mq_alloc_tag_set(&nullb->tag_set);
731 goto out_cleanup_queues;
733 nullb->q = blk_mq_init_queue(&nullb->tag_set);
734 if (IS_ERR(nullb->q)) {
736 goto out_cleanup_tags;
738 } else if (queue_mode == NULL_Q_BIO) {
739 nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
742 goto out_cleanup_queues;
744 blk_queue_make_request(nullb->q, null_queue_bio);
745 rv = init_driver_queues(nullb);
747 goto out_cleanup_blk_queue;
749 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
752 goto out_cleanup_queues;
754 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
755 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
756 rv = init_driver_queues(nullb);
758 goto out_cleanup_blk_queue;
761 nullb->q->queuedata = nullb;
762 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
763 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
766 nullb->index = nullb_indexes++;
769 blk_queue_logical_block_size(nullb->q, bs);
770 blk_queue_physical_block_size(nullb->q, bs);
772 sprintf(nullb->disk_name, "nullb%d", nullb->index);
775 rv = null_nvm_register(nullb);
777 rv = null_gendisk_register(nullb);
780 goto out_cleanup_blk_queue;
783 list_add_tail(&nullb->list, &nullb_list);
787 out_cleanup_blk_queue:
788 blk_cleanup_queue(nullb->q);
790 if (queue_mode == NULL_Q_MQ)
791 blk_mq_free_tag_set(&nullb->tag_set);
793 cleanup_queues(nullb);
800 static int __init null_init(void)
806 if (bs > PAGE_SIZE) {
807 pr_warn("null_blk: invalid block size\n");
808 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
812 if (use_lightnvm && bs != 4096) {
813 pr_warn("null_blk: LightNVM only supports 4k block size\n");
814 pr_warn("null_blk: defaults block size to 4k\n");
818 if (use_lightnvm && queue_mode != NULL_Q_MQ) {
819 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
820 pr_warn("null_blk: defaults queue mode to blk-mq\n");
821 queue_mode = NULL_Q_MQ;
824 if (queue_mode == NULL_Q_MQ && use_per_node_hctx) {
825 if (submit_queues < nr_online_nodes) {
826 pr_warn("null_blk: submit_queues param is set to %u.",
828 submit_queues = nr_online_nodes;
830 } else if (submit_queues > nr_cpu_ids)
831 submit_queues = nr_cpu_ids;
832 else if (!submit_queues)
837 null_major = register_blkdev(0, "nullb");
842 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
845 pr_err("null_blk: unable to create ppa cache\n");
851 for (i = 0; i < nr_devices; i++) {
852 ret = null_add_dev();
857 pr_info("null: module loaded\n");
861 while (!list_empty(&nullb_list)) {
862 nullb = list_entry(nullb_list.next, struct nullb, list);
865 kmem_cache_destroy(ppa_cache);
867 unregister_blkdev(null_major, "nullb");
871 static void __exit null_exit(void)
875 unregister_blkdev(null_major, "nullb");
878 while (!list_empty(&nullb_list)) {
879 nullb = list_entry(nullb_list.next, struct nullb, list);
884 kmem_cache_destroy(ppa_cache);
887 module_init(null_init);
888 module_exit(null_exit);
890 MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
891 MODULE_LICENSE("GPL");