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;
24 unsigned long *tag_map;
25 wait_queue_head_t wait;
26 unsigned int queue_depth;
28 struct nullb_cmd *cmds;
32 struct list_head list;
34 struct request_queue *q;
36 struct blk_mq_tag_set tag_set;
38 unsigned int queue_depth;
41 struct nullb_queue *queues;
42 unsigned int nr_queues;
43 char disk_name[DISK_NAME_LEN];
46 static LIST_HEAD(nullb_list);
47 static struct mutex lock;
48 static int null_major;
49 static int nullb_indexes;
50 static struct kmem_cache *ppa_cache;
52 struct completion_queue {
53 struct llist_head list;
58 * These are per-cpu for now, they will need to be configured by the
59 * complete_queues parameter and appropriately mapped.
61 static DEFINE_PER_CPU(struct completion_queue, completion_queues);
75 static int submit_queues;
76 module_param(submit_queues, int, S_IRUGO);
77 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
79 static int home_node = NUMA_NO_NODE;
80 module_param(home_node, int, S_IRUGO);
81 MODULE_PARM_DESC(home_node, "Home node for the device");
83 static int queue_mode = NULL_Q_MQ;
85 static int null_param_store_val(const char *str, int *val, int min, int max)
89 ret = kstrtoint(str, 10, &new_val);
93 if (new_val < min || new_val > max)
100 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
102 return null_param_store_val(str, &queue_mode, NULL_Q_BIO, NULL_Q_MQ);
105 static const struct kernel_param_ops null_queue_mode_param_ops = {
106 .set = null_set_queue_mode,
107 .get = param_get_int,
110 device_param_cb(queue_mode, &null_queue_mode_param_ops, &queue_mode, S_IRUGO);
111 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
114 module_param(gb, int, S_IRUGO);
115 MODULE_PARM_DESC(gb, "Size in GB");
118 module_param(bs, int, S_IRUGO);
119 MODULE_PARM_DESC(bs, "Block size (in bytes)");
121 static int nr_devices = 2;
122 module_param(nr_devices, int, S_IRUGO);
123 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
125 static bool use_lightnvm;
126 module_param(use_lightnvm, bool, S_IRUGO);
127 MODULE_PARM_DESC(use_lightnvm, "Register as a LightNVM device");
129 static int irqmode = NULL_IRQ_SOFTIRQ;
131 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
133 return null_param_store_val(str, &irqmode, NULL_IRQ_NONE,
137 static const struct kernel_param_ops null_irqmode_param_ops = {
138 .set = null_set_irqmode,
139 .get = param_get_int,
142 device_param_cb(irqmode, &null_irqmode_param_ops, &irqmode, S_IRUGO);
143 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
145 static int completion_nsec = 10000;
146 module_param(completion_nsec, int, S_IRUGO);
147 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
149 static int hw_queue_depth = 64;
150 module_param(hw_queue_depth, int, S_IRUGO);
151 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
153 static bool use_per_node_hctx = false;
154 module_param(use_per_node_hctx, bool, S_IRUGO);
155 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
157 static void put_tag(struct nullb_queue *nq, unsigned int tag)
159 clear_bit_unlock(tag, nq->tag_map);
161 if (waitqueue_active(&nq->wait))
165 static unsigned int get_tag(struct nullb_queue *nq)
170 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
171 if (tag >= nq->queue_depth)
173 } while (test_and_set_bit_lock(tag, nq->tag_map));
178 static void free_cmd(struct nullb_cmd *cmd)
180 put_tag(cmd->nq, cmd->tag);
183 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
185 struct nullb_cmd *cmd;
190 cmd = &nq->cmds[tag];
199 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
201 struct nullb_cmd *cmd;
204 cmd = __alloc_cmd(nq);
205 if (cmd || !can_wait)
209 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
210 cmd = __alloc_cmd(nq);
217 finish_wait(&nq->wait, &wait);
221 static void end_cmd(struct nullb_cmd *cmd)
223 switch (queue_mode) {
225 blk_mq_end_request(cmd->rq, 0);
228 INIT_LIST_HEAD(&cmd->rq->queuelist);
229 blk_end_request_all(cmd->rq, 0);
239 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
241 struct completion_queue *cq;
242 struct llist_node *entry;
243 struct nullb_cmd *cmd;
245 cq = &per_cpu(completion_queues, smp_processor_id());
247 while ((entry = llist_del_all(&cq->list)) != NULL) {
248 entry = llist_reverse_order(entry);
250 struct request_queue *q = NULL;
252 cmd = container_of(entry, struct nullb_cmd, ll_list);
258 if (q && !q->mq_ops && blk_queue_stopped(q)) {
259 spin_lock(q->queue_lock);
260 if (blk_queue_stopped(q))
262 spin_unlock(q->queue_lock);
267 return HRTIMER_NORESTART;
270 static void null_cmd_end_timer(struct nullb_cmd *cmd)
272 struct completion_queue *cq = &per_cpu(completion_queues, get_cpu());
274 cmd->ll_list.next = NULL;
275 if (llist_add(&cmd->ll_list, &cq->list)) {
276 ktime_t kt = ktime_set(0, completion_nsec);
278 hrtimer_start(&cq->timer, kt, HRTIMER_MODE_REL_PINNED);
284 static void null_softirq_done_fn(struct request *rq)
286 if (queue_mode == NULL_Q_MQ)
287 end_cmd(blk_mq_rq_to_pdu(rq));
289 end_cmd(rq->special);
292 static inline void null_handle_cmd(struct nullb_cmd *cmd)
294 /* Complete IO by inline, softirq or timer */
296 case NULL_IRQ_SOFTIRQ:
297 switch (queue_mode) {
299 blk_mq_complete_request(cmd->rq, cmd->rq->errors);
302 blk_complete_request(cmd->rq);
306 * XXX: no proper submitting cpu information available.
316 null_cmd_end_timer(cmd);
321 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
325 if (nullb->nr_queues != 1)
326 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
328 return &nullb->queues[index];
331 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
333 struct nullb *nullb = q->queuedata;
334 struct nullb_queue *nq = nullb_to_queue(nullb);
335 struct nullb_cmd *cmd;
337 cmd = alloc_cmd(nq, 1);
340 null_handle_cmd(cmd);
341 return BLK_QC_T_NONE;
344 static int null_rq_prep_fn(struct request_queue *q, struct request *req)
346 struct nullb *nullb = q->queuedata;
347 struct nullb_queue *nq = nullb_to_queue(nullb);
348 struct nullb_cmd *cmd;
350 cmd = alloc_cmd(nq, 0);
358 return BLKPREP_DEFER;
361 static void null_request_fn(struct request_queue *q)
365 while ((rq = blk_fetch_request(q)) != NULL) {
366 struct nullb_cmd *cmd = rq->special;
368 spin_unlock_irq(q->queue_lock);
369 null_handle_cmd(cmd);
370 spin_lock_irq(q->queue_lock);
374 static int null_queue_rq(struct blk_mq_hw_ctx *hctx,
375 const struct blk_mq_queue_data *bd)
377 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
380 cmd->nq = hctx->driver_data;
382 blk_mq_start_request(bd->rq);
384 null_handle_cmd(cmd);
385 return BLK_MQ_RQ_QUEUE_OK;
388 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
393 init_waitqueue_head(&nq->wait);
394 nq->queue_depth = nullb->queue_depth;
397 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
400 struct nullb *nullb = data;
401 struct nullb_queue *nq = &nullb->queues[index];
403 hctx->driver_data = nq;
404 null_init_queue(nullb, nq);
410 static struct blk_mq_ops null_mq_ops = {
411 .queue_rq = null_queue_rq,
412 .map_queue = blk_mq_map_queue,
413 .init_hctx = null_init_hctx,
414 .complete = null_softirq_done_fn,
417 static void cleanup_queue(struct nullb_queue *nq)
423 static void cleanup_queues(struct nullb *nullb)
427 for (i = 0; i < nullb->nr_queues; i++)
428 cleanup_queue(&nullb->queues[i]);
430 kfree(nullb->queues);
433 static void null_del_dev(struct nullb *nullb)
435 list_del_init(&nullb->list);
438 nvm_unregister(nullb->disk_name);
440 del_gendisk(nullb->disk);
441 blk_cleanup_queue(nullb->q);
442 if (queue_mode == NULL_Q_MQ)
443 blk_mq_free_tag_set(&nullb->tag_set);
445 put_disk(nullb->disk);
446 cleanup_queues(nullb);
452 static void null_lnvm_end_io(struct request *rq, int error)
454 struct nvm_rq *rqd = rq->end_io_data;
455 struct nvm_dev *dev = rqd->dev;
457 dev->mt->end_io(rqd, error);
462 static int null_lnvm_submit_io(struct request_queue *q, struct nvm_rq *rqd)
465 struct bio *bio = rqd->bio;
467 rq = blk_mq_alloc_request(q, bio_rw(bio), GFP_KERNEL, 0);
471 rq->cmd_type = REQ_TYPE_DRV_PRIV;
472 rq->__sector = bio->bi_iter.bi_sector;
473 rq->ioprio = bio_prio(bio);
475 if (bio_has_data(bio))
476 rq->nr_phys_segments = bio_phys_segments(q, bio);
478 rq->__data_len = bio->bi_iter.bi_size;
479 rq->bio = rq->biotail = bio;
481 rq->end_io_data = rqd;
483 blk_execute_rq_nowait(q, NULL, rq, 0, null_lnvm_end_io);
488 static int null_lnvm_id(struct request_queue *q, struct nvm_id *id)
490 sector_t size = gb * 1024 * 1024 * 1024ULL;
492 struct nvm_id_group *grp;
500 id->ppaf.blk_offset = 0;
501 id->ppaf.blk_len = 16;
502 id->ppaf.pg_offset = 16;
503 id->ppaf.pg_len = 16;
504 id->ppaf.sect_offset = 32;
505 id->ppaf.sect_len = 8;
506 id->ppaf.pln_offset = 40;
507 id->ppaf.pln_len = 8;
508 id->ppaf.lun_offset = 48;
509 id->ppaf.lun_len = 8;
510 id->ppaf.ch_offset = 56;
513 do_div(size, bs); /* convert size to pages */
514 do_div(size, 256); /* concert size to pgs pr blk */
515 grp = &id->groups[0];
521 do_div(size, (1 << 16));
522 grp->num_lun = size + 1;
523 do_div(blksize, grp->num_lun);
524 grp->num_blk = blksize;
535 grp->mpos = 0x010101; /* single plane rwe */
536 grp->cpar = hw_queue_depth;
541 static void *null_lnvm_create_dma_pool(struct request_queue *q, char *name)
543 mempool_t *virtmem_pool;
545 virtmem_pool = mempool_create_slab_pool(64, ppa_cache);
547 pr_err("null_blk: Unable to create virtual memory pool\n");
554 static void null_lnvm_destroy_dma_pool(void *pool)
556 mempool_destroy(pool);
559 static void *null_lnvm_dev_dma_alloc(struct request_queue *q, void *pool,
560 gfp_t mem_flags, dma_addr_t *dma_handler)
562 return mempool_alloc(pool, mem_flags);
565 static void null_lnvm_dev_dma_free(void *pool, void *entry,
566 dma_addr_t dma_handler)
568 mempool_free(entry, pool);
571 static struct nvm_dev_ops null_lnvm_dev_ops = {
572 .identity = null_lnvm_id,
573 .submit_io = null_lnvm_submit_io,
575 .create_dma_pool = null_lnvm_create_dma_pool,
576 .destroy_dma_pool = null_lnvm_destroy_dma_pool,
577 .dev_dma_alloc = null_lnvm_dev_dma_alloc,
578 .dev_dma_free = null_lnvm_dev_dma_free,
580 /* Simulate nvme protocol restriction */
584 static struct nvm_dev_ops null_lnvm_dev_ops;
585 #endif /* CONFIG_NVM */
587 static int null_open(struct block_device *bdev, fmode_t mode)
592 static void null_release(struct gendisk *disk, fmode_t mode)
596 static const struct block_device_operations null_fops = {
597 .owner = THIS_MODULE,
599 .release = null_release,
602 static int setup_commands(struct nullb_queue *nq)
604 struct nullb_cmd *cmd;
607 nq->cmds = kzalloc(nq->queue_depth * sizeof(*cmd), GFP_KERNEL);
611 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
612 nq->tag_map = kzalloc(tag_size * sizeof(unsigned long), GFP_KERNEL);
618 for (i = 0; i < nq->queue_depth; i++) {
620 INIT_LIST_HEAD(&cmd->list);
621 cmd->ll_list.next = NULL;
628 static int setup_queues(struct nullb *nullb)
630 nullb->queues = kzalloc(submit_queues * sizeof(struct nullb_queue),
635 nullb->nr_queues = 0;
636 nullb->queue_depth = hw_queue_depth;
641 static int init_driver_queues(struct nullb *nullb)
643 struct nullb_queue *nq;
646 for (i = 0; i < submit_queues; i++) {
647 nq = &nullb->queues[i];
649 null_init_queue(nullb, nq);
651 ret = setup_commands(nq);
659 static int null_add_dev(void)
661 struct gendisk *disk;
666 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, home_node);
672 spin_lock_init(&nullb->lock);
674 if (queue_mode == NULL_Q_MQ && use_per_node_hctx)
675 submit_queues = nr_online_nodes;
677 rv = setup_queues(nullb);
681 if (queue_mode == NULL_Q_MQ) {
682 nullb->tag_set.ops = &null_mq_ops;
683 nullb->tag_set.nr_hw_queues = submit_queues;
684 nullb->tag_set.queue_depth = hw_queue_depth;
685 nullb->tag_set.numa_node = home_node;
686 nullb->tag_set.cmd_size = sizeof(struct nullb_cmd);
687 nullb->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
688 nullb->tag_set.driver_data = nullb;
690 rv = blk_mq_alloc_tag_set(&nullb->tag_set);
692 goto out_cleanup_queues;
694 nullb->q = blk_mq_init_queue(&nullb->tag_set);
695 if (IS_ERR(nullb->q)) {
697 goto out_cleanup_tags;
699 } else if (queue_mode == NULL_Q_BIO) {
700 nullb->q = blk_alloc_queue_node(GFP_KERNEL, home_node);
703 goto out_cleanup_queues;
705 blk_queue_make_request(nullb->q, null_queue_bio);
706 rv = init_driver_queues(nullb);
708 goto out_cleanup_blk_queue;
710 nullb->q = blk_init_queue_node(null_request_fn, &nullb->lock, home_node);
713 goto out_cleanup_queues;
715 blk_queue_prep_rq(nullb->q, null_rq_prep_fn);
716 blk_queue_softirq_done(nullb->q, null_softirq_done_fn);
717 rv = init_driver_queues(nullb);
719 goto out_cleanup_blk_queue;
722 nullb->q->queuedata = nullb;
723 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, nullb->q);
724 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, nullb->q);
728 list_add_tail(&nullb->list, &nullb_list);
729 nullb->index = nullb_indexes++;
732 blk_queue_logical_block_size(nullb->q, bs);
733 blk_queue_physical_block_size(nullb->q, bs);
735 sprintf(nullb->disk_name, "nullb%d", nullb->index);
738 rv = nvm_register(nullb->q, nullb->disk_name,
741 goto out_cleanup_blk_queue;
745 disk = nullb->disk = alloc_disk_node(1, home_node);
748 goto out_cleanup_lightnvm;
750 size = gb * 1024 * 1024 * 1024ULL;
751 set_capacity(disk, size >> 9);
753 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
754 disk->major = null_major;
755 disk->first_minor = nullb->index;
756 disk->fops = &null_fops;
757 disk->private_data = nullb;
758 disk->queue = nullb->q;
759 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
765 out_cleanup_lightnvm:
767 nvm_unregister(nullb->disk_name);
768 out_cleanup_blk_queue:
769 blk_cleanup_queue(nullb->q);
771 if (queue_mode == NULL_Q_MQ)
772 blk_mq_free_tag_set(&nullb->tag_set);
774 cleanup_queues(nullb);
781 static int __init null_init(void)
785 if (bs > PAGE_SIZE) {
786 pr_warn("null_blk: invalid block size\n");
787 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
791 if (use_lightnvm && bs != 4096) {
792 pr_warn("null_blk: LightNVM only supports 4k block size\n");
793 pr_warn("null_blk: defaults block size to 4k\n");
797 if (use_lightnvm && queue_mode != NULL_Q_MQ) {
798 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
799 pr_warn("null_blk: defaults queue mode to blk-mq\n");
800 queue_mode = NULL_Q_MQ;
803 if (queue_mode == NULL_Q_MQ && use_per_node_hctx) {
804 if (submit_queues < nr_online_nodes) {
805 pr_warn("null_blk: submit_queues param is set to %u.",
807 submit_queues = nr_online_nodes;
809 } else if (submit_queues > nr_cpu_ids)
810 submit_queues = nr_cpu_ids;
811 else if (!submit_queues)
816 /* Initialize a separate list for each CPU for issuing softirqs */
817 for_each_possible_cpu(i) {
818 struct completion_queue *cq = &per_cpu(completion_queues, i);
820 init_llist_head(&cq->list);
822 if (irqmode != NULL_IRQ_TIMER)
825 hrtimer_init(&cq->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
826 cq->timer.function = null_cmd_timer_expired;
829 null_major = register_blkdev(0, "nullb");
834 ppa_cache = kmem_cache_create("ppa_cache", 64 * sizeof(u64),
837 pr_err("null_blk: unable to create ppa cache\n");
842 for (i = 0; i < nr_devices; i++) {
843 if (null_add_dev()) {
844 unregister_blkdev(null_major, "nullb");
849 pr_info("null: module loaded\n");
852 kmem_cache_destroy(ppa_cache);
856 static void __exit null_exit(void)
860 unregister_blkdev(null_major, "nullb");
863 while (!list_empty(&nullb_list)) {
864 nullb = list_entry(nullb_list.next, struct nullb, list);
869 kmem_cache_destroy(ppa_cache);
872 module_init(null_init);
873 module_exit(null_exit);
875 MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
876 MODULE_LICENSE("GPL");