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
;
37 struct blk_mq_tag_set tag_set
;
39 unsigned int queue_depth
;
42 struct nullb_queue
*queues
;
43 unsigned int nr_queues
;
44 char disk_name
[DISK_NAME_LEN
];
47 static LIST_HEAD(nullb_list
);
48 static struct mutex lock
;
49 static int null_major
;
50 static int nullb_indexes
;
51 static struct kmem_cache
*ppa_cache
;
65 static int submit_queues
;
66 module_param(submit_queues
, int, S_IRUGO
);
67 MODULE_PARM_DESC(submit_queues
, "Number of submission queues");
69 static int home_node
= NUMA_NO_NODE
;
70 module_param(home_node
, int, S_IRUGO
);
71 MODULE_PARM_DESC(home_node
, "Home node for the device");
73 static int queue_mode
= NULL_Q_MQ
;
75 static int null_param_store_val(const char *str
, int *val
, int min
, int max
)
79 ret
= kstrtoint(str
, 10, &new_val
);
83 if (new_val
< min
|| new_val
> max
)
90 static int null_set_queue_mode(const char *str
, const struct kernel_param
*kp
)
92 return null_param_store_val(str
, &queue_mode
, NULL_Q_BIO
, NULL_Q_MQ
);
95 static const struct kernel_param_ops null_queue_mode_param_ops
= {
96 .set
= null_set_queue_mode
,
100 device_param_cb(queue_mode
, &null_queue_mode_param_ops
, &queue_mode
, S_IRUGO
);
101 MODULE_PARM_DESC(queue_mode
, "Block interface to use (0=bio,1=rq,2=multiqueue)");
104 module_param(gb
, int, S_IRUGO
);
105 MODULE_PARM_DESC(gb
, "Size in GB");
108 module_param(bs
, int, S_IRUGO
);
109 MODULE_PARM_DESC(bs
, "Block size (in bytes)");
111 static int nr_devices
= 2;
112 module_param(nr_devices
, int, S_IRUGO
);
113 MODULE_PARM_DESC(nr_devices
, "Number of devices to register");
115 static bool use_lightnvm
;
116 module_param(use_lightnvm
, bool, S_IRUGO
);
117 MODULE_PARM_DESC(use_lightnvm
, "Register as a LightNVM device");
119 static int irqmode
= NULL_IRQ_SOFTIRQ
;
121 static int null_set_irqmode(const char *str
, const struct kernel_param
*kp
)
123 return null_param_store_val(str
, &irqmode
, NULL_IRQ_NONE
,
127 static const struct kernel_param_ops null_irqmode_param_ops
= {
128 .set
= null_set_irqmode
,
129 .get
= param_get_int
,
132 device_param_cb(irqmode
, &null_irqmode_param_ops
, &irqmode
, S_IRUGO
);
133 MODULE_PARM_DESC(irqmode
, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
135 static unsigned long completion_nsec
= 10000;
136 module_param(completion_nsec
, ulong
, S_IRUGO
);
137 MODULE_PARM_DESC(completion_nsec
, "Time in ns to complete a request in hardware. Default: 10,000ns");
139 static int hw_queue_depth
= 64;
140 module_param(hw_queue_depth
, int, S_IRUGO
);
141 MODULE_PARM_DESC(hw_queue_depth
, "Queue depth for each hardware queue. Default: 64");
143 static bool use_per_node_hctx
= false;
144 module_param(use_per_node_hctx
, bool, S_IRUGO
);
145 MODULE_PARM_DESC(use_per_node_hctx
, "Use per-node allocation for hardware context queues. Default: false");
147 static void put_tag(struct nullb_queue
*nq
, unsigned int tag
)
149 clear_bit_unlock(tag
, nq
->tag_map
);
151 if (waitqueue_active(&nq
->wait
))
155 static unsigned int get_tag(struct nullb_queue
*nq
)
160 tag
= find_first_zero_bit(nq
->tag_map
, nq
->queue_depth
);
161 if (tag
>= nq
->queue_depth
)
163 } while (test_and_set_bit_lock(tag
, nq
->tag_map
));
168 static void free_cmd(struct nullb_cmd
*cmd
)
170 put_tag(cmd
->nq
, cmd
->tag
);
173 static enum hrtimer_restart
null_cmd_timer_expired(struct hrtimer
*timer
);
175 static struct nullb_cmd
*__alloc_cmd(struct nullb_queue
*nq
)
177 struct nullb_cmd
*cmd
;
182 cmd
= &nq
->cmds
[tag
];
185 if (irqmode
== NULL_IRQ_TIMER
) {
186 hrtimer_init(&cmd
->timer
, CLOCK_MONOTONIC
,
188 cmd
->timer
.function
= null_cmd_timer_expired
;
196 static struct nullb_cmd
*alloc_cmd(struct nullb_queue
*nq
, int can_wait
)
198 struct nullb_cmd
*cmd
;
201 cmd
= __alloc_cmd(nq
);
202 if (cmd
|| !can_wait
)
206 prepare_to_wait(&nq
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
207 cmd
= __alloc_cmd(nq
);
214 finish_wait(&nq
->wait
, &wait
);
218 static void end_cmd(struct nullb_cmd
*cmd
)
220 struct request_queue
*q
= NULL
;
225 switch (queue_mode
) {
227 blk_mq_end_request(cmd
->rq
, 0);
230 INIT_LIST_HEAD(&cmd
->rq
->queuelist
);
231 blk_end_request_all(cmd
->rq
, 0);
240 /* Restart queue if needed, as we are freeing a tag */
241 if (queue_mode
== NULL_Q_RQ
&& blk_queue_stopped(q
)) {
244 spin_lock_irqsave(q
->queue_lock
, flags
);
245 blk_start_queue_async(q
);
246 spin_unlock_irqrestore(q
->queue_lock
, flags
);
250 static enum hrtimer_restart
null_cmd_timer_expired(struct hrtimer
*timer
)
252 end_cmd(container_of(timer
, struct nullb_cmd
, timer
));
254 return HRTIMER_NORESTART
;
257 static void null_cmd_end_timer(struct nullb_cmd
*cmd
)
259 ktime_t kt
= ktime_set(0, completion_nsec
);
261 hrtimer_start(&cmd
->timer
, kt
, HRTIMER_MODE_REL
);
264 static void null_softirq_done_fn(struct request
*rq
)
266 if (queue_mode
== NULL_Q_MQ
)
267 end_cmd(blk_mq_rq_to_pdu(rq
));
269 end_cmd(rq
->special
);
272 static inline void null_handle_cmd(struct nullb_cmd
*cmd
)
274 /* Complete IO by inline, softirq or timer */
276 case NULL_IRQ_SOFTIRQ
:
277 switch (queue_mode
) {
279 blk_mq_complete_request(cmd
->rq
, cmd
->rq
->errors
);
282 blk_complete_request(cmd
->rq
);
286 * XXX: no proper submitting cpu information available.
296 null_cmd_end_timer(cmd
);
301 static struct nullb_queue
*nullb_to_queue(struct nullb
*nullb
)
305 if (nullb
->nr_queues
!= 1)
306 index
= raw_smp_processor_id() / ((nr_cpu_ids
+ nullb
->nr_queues
- 1) / nullb
->nr_queues
);
308 return &nullb
->queues
[index
];
311 static blk_qc_t
null_queue_bio(struct request_queue
*q
, struct bio
*bio
)
313 struct nullb
*nullb
= q
->queuedata
;
314 struct nullb_queue
*nq
= nullb_to_queue(nullb
);
315 struct nullb_cmd
*cmd
;
317 cmd
= alloc_cmd(nq
, 1);
320 null_handle_cmd(cmd
);
321 return BLK_QC_T_NONE
;
324 static int null_rq_prep_fn(struct request_queue
*q
, struct request
*req
)
326 struct nullb
*nullb
= q
->queuedata
;
327 struct nullb_queue
*nq
= nullb_to_queue(nullb
);
328 struct nullb_cmd
*cmd
;
330 cmd
= alloc_cmd(nq
, 0);
338 return BLKPREP_DEFER
;
341 static void null_request_fn(struct request_queue
*q
)
345 while ((rq
= blk_fetch_request(q
)) != NULL
) {
346 struct nullb_cmd
*cmd
= rq
->special
;
348 spin_unlock_irq(q
->queue_lock
);
349 null_handle_cmd(cmd
);
350 spin_lock_irq(q
->queue_lock
);
354 static int null_queue_rq(struct blk_mq_hw_ctx
*hctx
,
355 const struct blk_mq_queue_data
*bd
)
357 struct nullb_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
359 if (irqmode
== NULL_IRQ_TIMER
) {
360 hrtimer_init(&cmd
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
361 cmd
->timer
.function
= null_cmd_timer_expired
;
364 cmd
->nq
= hctx
->driver_data
;
366 blk_mq_start_request(bd
->rq
);
368 null_handle_cmd(cmd
);
369 return BLK_MQ_RQ_QUEUE_OK
;
372 static void null_init_queue(struct nullb
*nullb
, struct nullb_queue
*nq
)
377 init_waitqueue_head(&nq
->wait
);
378 nq
->queue_depth
= nullb
->queue_depth
;
381 static int null_init_hctx(struct blk_mq_hw_ctx
*hctx
, void *data
,
384 struct nullb
*nullb
= data
;
385 struct nullb_queue
*nq
= &nullb
->queues
[index
];
387 hctx
->driver_data
= nq
;
388 null_init_queue(nullb
, nq
);
394 static struct blk_mq_ops null_mq_ops
= {
395 .queue_rq
= null_queue_rq
,
396 .map_queue
= blk_mq_map_queue
,
397 .init_hctx
= null_init_hctx
,
398 .complete
= null_softirq_done_fn
,
401 static void cleanup_queue(struct nullb_queue
*nq
)
407 static void cleanup_queues(struct nullb
*nullb
)
411 for (i
= 0; i
< nullb
->nr_queues
; i
++)
412 cleanup_queue(&nullb
->queues
[i
]);
414 kfree(nullb
->queues
);
417 static void null_del_dev(struct nullb
*nullb
)
419 list_del_init(&nullb
->list
);
422 nvm_unregister(nullb
->disk_name
);
424 del_gendisk(nullb
->disk
);
425 blk_cleanup_queue(nullb
->q
);
426 if (queue_mode
== NULL_Q_MQ
)
427 blk_mq_free_tag_set(&nullb
->tag_set
);
429 put_disk(nullb
->disk
);
430 cleanup_queues(nullb
);
436 static void null_lnvm_end_io(struct request
*rq
, int error
)
438 struct nvm_rq
*rqd
= rq
->end_io_data
;
440 nvm_end_io(rqd
, error
);
445 static int null_lnvm_submit_io(struct nvm_dev
*dev
, struct nvm_rq
*rqd
)
447 struct request_queue
*q
= dev
->q
;
449 struct bio
*bio
= rqd
->bio
;
451 rq
= blk_mq_alloc_request(q
, bio_data_dir(bio
), 0);
455 rq
->cmd_type
= REQ_TYPE_DRV_PRIV
;
456 rq
->__sector
= bio
->bi_iter
.bi_sector
;
457 rq
->ioprio
= bio_prio(bio
);
459 if (bio_has_data(bio
))
460 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
462 rq
->__data_len
= bio
->bi_iter
.bi_size
;
463 rq
->bio
= rq
->biotail
= bio
;
465 rq
->end_io_data
= rqd
;
467 blk_execute_rq_nowait(q
, NULL
, rq
, 0, null_lnvm_end_io
);
472 static int null_lnvm_id(struct nvm_dev
*dev
, struct nvm_id
*id
)
474 sector_t size
= gb
* 1024 * 1024 * 1024ULL;
476 struct nvm_id_group
*grp
;
484 id
->ppaf
.blk_offset
= 0;
485 id
->ppaf
.blk_len
= 16;
486 id
->ppaf
.pg_offset
= 16;
487 id
->ppaf
.pg_len
= 16;
488 id
->ppaf
.sect_offset
= 32;
489 id
->ppaf
.sect_len
= 8;
490 id
->ppaf
.pln_offset
= 40;
491 id
->ppaf
.pln_len
= 8;
492 id
->ppaf
.lun_offset
= 48;
493 id
->ppaf
.lun_len
= 8;
494 id
->ppaf
.ch_offset
= 56;
497 sector_div(size
, bs
); /* convert size to pages */
498 size
>>= 8; /* concert size to pgs pr blk */
499 grp
= &id
->groups
[0];
506 grp
->num_lun
= size
+ 1;
507 sector_div(blksize
, grp
->num_lun
);
508 grp
->num_blk
= blksize
;
519 grp
->mpos
= 0x010101; /* single plane rwe */
520 grp
->cpar
= hw_queue_depth
;
525 static void *null_lnvm_create_dma_pool(struct nvm_dev
*dev
, char *name
)
527 mempool_t
*virtmem_pool
;
529 virtmem_pool
= mempool_create_slab_pool(64, ppa_cache
);
531 pr_err("null_blk: Unable to create virtual memory pool\n");
538 static void null_lnvm_destroy_dma_pool(void *pool
)
540 mempool_destroy(pool
);
543 static void *null_lnvm_dev_dma_alloc(struct nvm_dev
*dev
, void *pool
,
544 gfp_t mem_flags
, dma_addr_t
*dma_handler
)
546 return mempool_alloc(pool
, mem_flags
);
549 static void null_lnvm_dev_dma_free(void *pool
, void *entry
,
550 dma_addr_t dma_handler
)
552 mempool_free(entry
, pool
);
555 static struct nvm_dev_ops null_lnvm_dev_ops
= {
556 .identity
= null_lnvm_id
,
557 .submit_io
= null_lnvm_submit_io
,
559 .create_dma_pool
= null_lnvm_create_dma_pool
,
560 .destroy_dma_pool
= null_lnvm_destroy_dma_pool
,
561 .dev_dma_alloc
= null_lnvm_dev_dma_alloc
,
562 .dev_dma_free
= null_lnvm_dev_dma_free
,
564 /* Simulate nvme protocol restriction */
568 static struct nvm_dev_ops null_lnvm_dev_ops
;
569 #endif /* CONFIG_NVM */
571 static int null_open(struct block_device
*bdev
, fmode_t mode
)
576 static void null_release(struct gendisk
*disk
, fmode_t mode
)
580 static const struct block_device_operations null_fops
= {
581 .owner
= THIS_MODULE
,
583 .release
= null_release
,
586 static int setup_commands(struct nullb_queue
*nq
)
588 struct nullb_cmd
*cmd
;
591 nq
->cmds
= kzalloc(nq
->queue_depth
* sizeof(*cmd
), GFP_KERNEL
);
595 tag_size
= ALIGN(nq
->queue_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
596 nq
->tag_map
= kzalloc(tag_size
* sizeof(unsigned long), GFP_KERNEL
);
602 for (i
= 0; i
< nq
->queue_depth
; i
++) {
604 INIT_LIST_HEAD(&cmd
->list
);
605 cmd
->ll_list
.next
= NULL
;
612 static int setup_queues(struct nullb
*nullb
)
614 nullb
->queues
= kzalloc(submit_queues
* sizeof(struct nullb_queue
),
619 nullb
->nr_queues
= 0;
620 nullb
->queue_depth
= hw_queue_depth
;
625 static int init_driver_queues(struct nullb
*nullb
)
627 struct nullb_queue
*nq
;
630 for (i
= 0; i
< submit_queues
; i
++) {
631 nq
= &nullb
->queues
[i
];
633 null_init_queue(nullb
, nq
);
635 ret
= setup_commands(nq
);
643 static int null_add_dev(void)
645 struct gendisk
*disk
;
650 nullb
= kzalloc_node(sizeof(*nullb
), GFP_KERNEL
, home_node
);
656 spin_lock_init(&nullb
->lock
);
658 if (queue_mode
== NULL_Q_MQ
&& use_per_node_hctx
)
659 submit_queues
= nr_online_nodes
;
661 rv
= setup_queues(nullb
);
665 if (queue_mode
== NULL_Q_MQ
) {
666 nullb
->tag_set
.ops
= &null_mq_ops
;
667 nullb
->tag_set
.nr_hw_queues
= submit_queues
;
668 nullb
->tag_set
.queue_depth
= hw_queue_depth
;
669 nullb
->tag_set
.numa_node
= home_node
;
670 nullb
->tag_set
.cmd_size
= sizeof(struct nullb_cmd
);
671 nullb
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
;
672 nullb
->tag_set
.driver_data
= nullb
;
674 rv
= blk_mq_alloc_tag_set(&nullb
->tag_set
);
676 goto out_cleanup_queues
;
678 nullb
->q
= blk_mq_init_queue(&nullb
->tag_set
);
679 if (IS_ERR(nullb
->q
)) {
681 goto out_cleanup_tags
;
683 } else if (queue_mode
== NULL_Q_BIO
) {
684 nullb
->q
= blk_alloc_queue_node(GFP_KERNEL
, home_node
);
687 goto out_cleanup_queues
;
689 blk_queue_make_request(nullb
->q
, null_queue_bio
);
690 rv
= init_driver_queues(nullb
);
692 goto out_cleanup_blk_queue
;
694 nullb
->q
= blk_init_queue_node(null_request_fn
, &nullb
->lock
, home_node
);
697 goto out_cleanup_queues
;
699 blk_queue_prep_rq(nullb
->q
, null_rq_prep_fn
);
700 blk_queue_softirq_done(nullb
->q
, null_softirq_done_fn
);
701 rv
= init_driver_queues(nullb
);
703 goto out_cleanup_blk_queue
;
706 nullb
->q
->queuedata
= nullb
;
707 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, nullb
->q
);
708 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, nullb
->q
);
711 nullb
->index
= nullb_indexes
++;
714 blk_queue_logical_block_size(nullb
->q
, bs
);
715 blk_queue_physical_block_size(nullb
->q
, bs
);
717 sprintf(nullb
->disk_name
, "nullb%d", nullb
->index
);
720 rv
= nvm_register(nullb
->q
, nullb
->disk_name
,
723 goto out_cleanup_blk_queue
;
727 disk
= nullb
->disk
= alloc_disk_node(1, home_node
);
730 goto out_cleanup_lightnvm
;
732 size
= gb
* 1024 * 1024 * 1024ULL;
733 set_capacity(disk
, size
>> 9);
735 disk
->flags
|= GENHD_FL_EXT_DEVT
| GENHD_FL_SUPPRESS_PARTITION_INFO
;
736 disk
->major
= null_major
;
737 disk
->first_minor
= nullb
->index
;
738 disk
->fops
= &null_fops
;
739 disk
->private_data
= nullb
;
740 disk
->queue
= nullb
->q
;
741 strncpy(disk
->disk_name
, nullb
->disk_name
, DISK_NAME_LEN
);
747 list_add_tail(&nullb
->list
, &nullb_list
);
752 out_cleanup_lightnvm
:
754 nvm_unregister(nullb
->disk_name
);
755 out_cleanup_blk_queue
:
756 blk_cleanup_queue(nullb
->q
);
758 if (queue_mode
== NULL_Q_MQ
)
759 blk_mq_free_tag_set(&nullb
->tag_set
);
761 cleanup_queues(nullb
);
768 static int __init
null_init(void)
774 if (bs
> PAGE_SIZE
) {
775 pr_warn("null_blk: invalid block size\n");
776 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE
);
780 if (use_lightnvm
&& bs
!= 4096) {
781 pr_warn("null_blk: LightNVM only supports 4k block size\n");
782 pr_warn("null_blk: defaults block size to 4k\n");
786 if (use_lightnvm
&& queue_mode
!= NULL_Q_MQ
) {
787 pr_warn("null_blk: LightNVM only supported for blk-mq\n");
788 pr_warn("null_blk: defaults queue mode to blk-mq\n");
789 queue_mode
= NULL_Q_MQ
;
792 if (queue_mode
== NULL_Q_MQ
&& use_per_node_hctx
) {
793 if (submit_queues
< nr_online_nodes
) {
794 pr_warn("null_blk: submit_queues param is set to %u.",
796 submit_queues
= nr_online_nodes
;
798 } else if (submit_queues
> nr_cpu_ids
)
799 submit_queues
= nr_cpu_ids
;
800 else if (!submit_queues
)
805 null_major
= register_blkdev(0, "nullb");
810 ppa_cache
= kmem_cache_create("ppa_cache", 64 * sizeof(u64
),
813 pr_err("null_blk: unable to create ppa cache\n");
819 for (i
= 0; i
< nr_devices
; i
++) {
820 ret
= null_add_dev();
825 pr_info("null: module loaded\n");
829 while (!list_empty(&nullb_list
)) {
830 nullb
= list_entry(nullb_list
.next
, struct nullb
, list
);
833 kmem_cache_destroy(ppa_cache
);
835 unregister_blkdev(null_major
, "nullb");
839 static void __exit
null_exit(void)
843 unregister_blkdev(null_major
, "nullb");
846 while (!list_empty(&nullb_list
)) {
847 nullb
= list_entry(nullb_list
.next
, struct nullb
, list
);
852 kmem_cache_destroy(ppa_cache
);
855 module_init(null_init
);
856 module_exit(null_exit
);
858 MODULE_AUTHOR("Jens Axboe <jaxboe@fusionio.com>");
859 MODULE_LICENSE("GPL");