2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
29 #include <linux/pm_qos.h>
30 #include <asm/unaligned.h>
35 #define NVME_MINORS (1U << MINORBITS)
37 unsigned char admin_timeout = 60;
38 module_param(admin_timeout, byte, 0644);
39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout);
42 unsigned char nvme_io_timeout = 30;
43 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
44 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
45 EXPORT_SYMBOL_GPL(nvme_io_timeout);
47 static unsigned char shutdown_timeout = 5;
48 module_param(shutdown_timeout, byte, 0644);
49 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
51 static u8 nvme_max_retries = 5;
52 module_param_named(max_retries, nvme_max_retries, byte, 0644);
53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
55 static int nvme_char_major;
56 module_param(nvme_char_major, int, 0);
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61 "max power saving latency for new devices; use PM QOS to change per device");
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
71 struct workqueue_struct *nvme_wq;
72 EXPORT_SYMBOL_GPL(nvme_wq);
74 static LIST_HEAD(nvme_ctrl_list);
75 static DEFINE_SPINLOCK(dev_list_lock);
77 static struct class *nvme_class;
79 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
81 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
83 if (!queue_work(nvme_wq, &ctrl->reset_work))
87 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
89 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
93 ret = nvme_reset_ctrl(ctrl);
95 flush_work(&ctrl->reset_work);
99 static blk_status_t nvme_error_status(struct request *req)
101 switch (nvme_req(req)->status & 0x7ff) {
102 case NVME_SC_SUCCESS:
104 case NVME_SC_CAP_EXCEEDED:
105 return BLK_STS_NOSPC;
106 case NVME_SC_ONCS_NOT_SUPPORTED:
107 return BLK_STS_NOTSUPP;
108 case NVME_SC_WRITE_FAULT:
109 case NVME_SC_READ_ERROR:
110 case NVME_SC_UNWRITTEN_BLOCK:
111 return BLK_STS_MEDIUM;
113 return BLK_STS_IOERR;
117 static inline bool nvme_req_needs_retry(struct request *req)
119 if (blk_noretry_request(req))
121 if (nvme_req(req)->status & NVME_SC_DNR)
123 if (jiffies - req->start_time >= req->timeout)
125 if (nvme_req(req)->retries >= nvme_max_retries)
130 void nvme_complete_rq(struct request *req)
132 if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
133 nvme_req(req)->retries++;
134 blk_mq_requeue_request(req, true);
138 blk_mq_end_request(req, nvme_error_status(req));
140 EXPORT_SYMBOL_GPL(nvme_complete_rq);
142 void nvme_cancel_request(struct request *req, void *data, bool reserved)
146 if (!blk_mq_request_started(req))
149 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
150 "Cancelling I/O %d", req->tag);
152 status = NVME_SC_ABORT_REQ;
153 if (blk_queue_dying(req->q))
154 status |= NVME_SC_DNR;
155 nvme_req(req)->status = status;
156 blk_mq_complete_request(req);
159 EXPORT_SYMBOL_GPL(nvme_cancel_request);
161 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
162 enum nvme_ctrl_state new_state)
164 enum nvme_ctrl_state old_state;
165 bool changed = false;
167 spin_lock_irq(&ctrl->lock);
169 old_state = ctrl->state;
174 case NVME_CTRL_RESETTING:
175 case NVME_CTRL_RECONNECTING:
182 case NVME_CTRL_RESETTING:
192 case NVME_CTRL_RECONNECTING:
201 case NVME_CTRL_DELETING:
204 case NVME_CTRL_RESETTING:
205 case NVME_CTRL_RECONNECTING:
214 case NVME_CTRL_DELETING:
226 ctrl->state = new_state;
228 spin_unlock_irq(&ctrl->lock);
232 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
234 static void nvme_free_ns(struct kref *kref)
236 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
239 nvme_nvm_unregister(ns);
242 spin_lock(&dev_list_lock);
243 ns->disk->private_data = NULL;
244 spin_unlock(&dev_list_lock);
248 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
249 nvme_put_ctrl(ns->ctrl);
253 static void nvme_put_ns(struct nvme_ns *ns)
255 kref_put(&ns->kref, nvme_free_ns);
258 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
262 spin_lock(&dev_list_lock);
263 ns = disk->private_data;
265 if (!kref_get_unless_zero(&ns->kref))
267 if (!try_module_get(ns->ctrl->ops->module))
270 spin_unlock(&dev_list_lock);
275 kref_put(&ns->kref, nvme_free_ns);
277 spin_unlock(&dev_list_lock);
281 struct request *nvme_alloc_request(struct request_queue *q,
282 struct nvme_command *cmd, unsigned int flags, int qid)
284 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
287 if (qid == NVME_QID_ANY) {
288 req = blk_mq_alloc_request(q, op, flags);
290 req = blk_mq_alloc_request_hctx(q, op, flags,
296 req->cmd_flags |= REQ_FAILFAST_DRIVER;
297 nvme_req(req)->cmd = cmd;
301 EXPORT_SYMBOL_GPL(nvme_alloc_request);
303 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
305 struct nvme_command c;
307 memset(&c, 0, sizeof(c));
309 c.directive.opcode = nvme_admin_directive_send;
310 c.directive.nsid = cpu_to_le32(0xffffffff);
311 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
312 c.directive.dtype = NVME_DIR_IDENTIFY;
313 c.directive.tdtype = NVME_DIR_STREAMS;
314 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
316 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
319 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
321 return nvme_toggle_streams(ctrl, false);
324 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
326 return nvme_toggle_streams(ctrl, true);
329 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
330 struct streams_directive_params *s, u32 nsid)
332 struct nvme_command c;
334 memset(&c, 0, sizeof(c));
335 memset(s, 0, sizeof(*s));
337 c.directive.opcode = nvme_admin_directive_recv;
338 c.directive.nsid = cpu_to_le32(nsid);
339 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
340 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
341 c.directive.dtype = NVME_DIR_STREAMS;
343 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
346 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
348 struct streams_directive_params s;
351 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
356 ret = nvme_enable_streams(ctrl);
360 ret = nvme_get_stream_params(ctrl, &s, 0xffffffff);
364 ctrl->nssa = le16_to_cpu(s.nssa);
365 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
366 dev_info(ctrl->device, "too few streams (%u) available\n",
368 nvme_disable_streams(ctrl);
372 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
373 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
378 * Check if 'req' has a write hint associated with it. If it does, assign
379 * a valid namespace stream to the write.
381 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
382 struct request *req, u16 *control,
385 enum rw_hint streamid = req->write_hint;
387 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
391 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
394 *control |= NVME_RW_DTYPE_STREAMS;
395 *dsmgmt |= streamid << 16;
398 if (streamid < ARRAY_SIZE(req->q->write_hints))
399 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
402 static inline void nvme_setup_flush(struct nvme_ns *ns,
403 struct nvme_command *cmnd)
405 memset(cmnd, 0, sizeof(*cmnd));
406 cmnd->common.opcode = nvme_cmd_flush;
407 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
410 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
411 struct nvme_command *cmnd)
413 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
414 struct nvme_dsm_range *range;
417 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
419 return BLK_STS_RESOURCE;
421 __rq_for_each_bio(bio, req) {
422 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
423 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
425 range[n].cattr = cpu_to_le32(0);
426 range[n].nlb = cpu_to_le32(nlb);
427 range[n].slba = cpu_to_le64(slba);
431 if (WARN_ON_ONCE(n != segments)) {
433 return BLK_STS_IOERR;
436 memset(cmnd, 0, sizeof(*cmnd));
437 cmnd->dsm.opcode = nvme_cmd_dsm;
438 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
439 cmnd->dsm.nr = cpu_to_le32(segments - 1);
440 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
442 req->special_vec.bv_page = virt_to_page(range);
443 req->special_vec.bv_offset = offset_in_page(range);
444 req->special_vec.bv_len = sizeof(*range) * segments;
445 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
450 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
451 struct request *req, struct nvme_command *cmnd)
453 struct nvme_ctrl *ctrl = ns->ctrl;
458 * If formated with metadata, require the block layer provide a buffer
459 * unless this namespace is formated such that the metadata can be
460 * stripped/generated by the controller with PRACT=1.
463 (!ns->pi_type || ns->ms != sizeof(struct t10_pi_tuple)) &&
464 !blk_integrity_rq(req) && !blk_rq_is_passthrough(req))
465 return BLK_STS_NOTSUPP;
467 if (req->cmd_flags & REQ_FUA)
468 control |= NVME_RW_FUA;
469 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
470 control |= NVME_RW_LR;
472 if (req->cmd_flags & REQ_RAHEAD)
473 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
475 memset(cmnd, 0, sizeof(*cmnd));
476 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
477 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
478 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
479 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
481 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
482 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
485 switch (ns->pi_type) {
486 case NVME_NS_DPS_PI_TYPE3:
487 control |= NVME_RW_PRINFO_PRCHK_GUARD;
489 case NVME_NS_DPS_PI_TYPE1:
490 case NVME_NS_DPS_PI_TYPE2:
491 control |= NVME_RW_PRINFO_PRCHK_GUARD |
492 NVME_RW_PRINFO_PRCHK_REF;
493 cmnd->rw.reftag = cpu_to_le32(
494 nvme_block_nr(ns, blk_rq_pos(req)));
497 if (!blk_integrity_rq(req))
498 control |= NVME_RW_PRINFO_PRACT;
501 cmnd->rw.control = cpu_to_le16(control);
502 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
506 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
507 struct nvme_command *cmd)
509 blk_status_t ret = BLK_STS_OK;
511 if (!(req->rq_flags & RQF_DONTPREP)) {
512 nvme_req(req)->retries = 0;
513 nvme_req(req)->flags = 0;
514 req->rq_flags |= RQF_DONTPREP;
517 switch (req_op(req)) {
520 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
523 nvme_setup_flush(ns, cmd);
525 case REQ_OP_WRITE_ZEROES:
526 /* currently only aliased to deallocate for a few ctrls: */
528 ret = nvme_setup_discard(ns, req, cmd);
532 ret = nvme_setup_rw(ns, req, cmd);
536 return BLK_STS_IOERR;
539 cmd->common.command_id = req->tag;
542 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
545 * Returns 0 on success. If the result is negative, it's a Linux error code;
546 * if the result is positive, it's an NVM Express status code
548 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
549 union nvme_result *result, void *buffer, unsigned bufflen,
550 unsigned timeout, int qid, int at_head, int flags)
555 req = nvme_alloc_request(q, cmd, flags, qid);
559 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
561 if (buffer && bufflen) {
562 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
567 blk_execute_rq(req->q, NULL, req, at_head);
569 *result = nvme_req(req)->result;
570 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
573 ret = nvme_req(req)->status;
575 blk_mq_free_request(req);
578 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
580 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
581 void *buffer, unsigned bufflen)
583 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
586 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
588 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
589 void __user *ubuffer, unsigned bufflen,
590 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
591 u32 *result, unsigned timeout)
593 bool write = nvme_is_write(cmd);
594 struct nvme_ns *ns = q->queuedata;
595 struct gendisk *disk = ns ? ns->disk : NULL;
597 struct bio *bio = NULL;
601 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
605 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
607 if (ubuffer && bufflen) {
608 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
616 bio->bi_bdev = bdget_disk(disk, 0);
622 if (meta_buffer && meta_len) {
623 struct bio_integrity_payload *bip;
625 meta = kmalloc(meta_len, GFP_KERNEL);
632 if (copy_from_user(meta, meta_buffer,
639 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
645 bip->bip_iter.bi_size = meta_len;
646 bip->bip_iter.bi_sector = meta_seed;
648 ret = bio_integrity_add_page(bio, virt_to_page(meta),
649 meta_len, offset_in_page(meta));
650 if (ret != meta_len) {
657 blk_execute_rq(req->q, disk, req, 0);
658 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
661 ret = nvme_req(req)->status;
663 *result = le32_to_cpu(nvme_req(req)->result.u32);
664 if (meta && !ret && !write) {
665 if (copy_to_user(meta_buffer, meta, meta_len))
672 if (disk && bio->bi_bdev)
674 blk_rq_unmap_user(bio);
677 blk_mq_free_request(req);
681 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
682 void __user *ubuffer, unsigned bufflen, u32 *result,
685 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
689 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
691 struct nvme_ctrl *ctrl = rq->end_io_data;
693 blk_mq_free_request(rq);
696 dev_err(ctrl->device,
697 "failed nvme_keep_alive_end_io error=%d\n",
702 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
705 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
707 struct nvme_command c;
710 memset(&c, 0, sizeof(c));
711 c.common.opcode = nvme_admin_keep_alive;
713 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
718 rq->timeout = ctrl->kato * HZ;
719 rq->end_io_data = ctrl;
721 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
726 static void nvme_keep_alive_work(struct work_struct *work)
728 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
729 struct nvme_ctrl, ka_work);
731 if (nvme_keep_alive(ctrl)) {
732 /* allocation failure, reset the controller */
733 dev_err(ctrl->device, "keep-alive failed\n");
734 nvme_reset_ctrl(ctrl);
739 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
741 if (unlikely(ctrl->kato == 0))
744 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
745 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
747 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
749 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
751 if (unlikely(ctrl->kato == 0))
754 cancel_delayed_work_sync(&ctrl->ka_work);
756 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
758 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
760 struct nvme_command c = { };
763 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
764 c.identify.opcode = nvme_admin_identify;
765 c.identify.cns = NVME_ID_CNS_CTRL;
767 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
771 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
772 sizeof(struct nvme_id_ctrl));
778 static int nvme_identify_ns_descs(struct nvme_ns *ns, unsigned nsid)
780 struct nvme_command c = { };
786 c.identify.opcode = nvme_admin_identify;
787 c.identify.nsid = cpu_to_le32(nsid);
788 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
790 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
794 status = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, data,
795 NVME_IDENTIFY_DATA_SIZE);
799 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
800 struct nvme_ns_id_desc *cur = data + pos;
806 case NVME_NIDT_EUI64:
807 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
808 dev_warn(ns->ctrl->device,
809 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
813 len = NVME_NIDT_EUI64_LEN;
814 memcpy(ns->eui, data + pos + sizeof(*cur), len);
816 case NVME_NIDT_NGUID:
817 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
818 dev_warn(ns->ctrl->device,
819 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
823 len = NVME_NIDT_NGUID_LEN;
824 memcpy(ns->nguid, data + pos + sizeof(*cur), len);
827 if (cur->nidl != NVME_NIDT_UUID_LEN) {
828 dev_warn(ns->ctrl->device,
829 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
833 len = NVME_NIDT_UUID_LEN;
834 uuid_copy(&ns->uuid, data + pos + sizeof(*cur));
837 /* Skip unnkown types */
849 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
851 struct nvme_command c = { };
853 c.identify.opcode = nvme_admin_identify;
854 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
855 c.identify.nsid = cpu_to_le32(nsid);
856 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
859 static int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
860 struct nvme_id_ns **id)
862 struct nvme_command c = { };
865 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
866 c.identify.opcode = nvme_admin_identify;
867 c.identify.nsid = cpu_to_le32(nsid);
868 c.identify.cns = NVME_ID_CNS_NS;
870 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
874 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
875 sizeof(struct nvme_id_ns));
881 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
882 void *buffer, size_t buflen, u32 *result)
884 struct nvme_command c;
885 union nvme_result res;
888 memset(&c, 0, sizeof(c));
889 c.features.opcode = nvme_admin_set_features;
890 c.features.fid = cpu_to_le32(fid);
891 c.features.dword11 = cpu_to_le32(dword11);
893 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
894 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
895 if (ret >= 0 && result)
896 *result = le32_to_cpu(res.u32);
900 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
902 u32 q_count = (*count - 1) | ((*count - 1) << 16);
904 int status, nr_io_queues;
906 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
912 * Degraded controllers might return an error when setting the queue
913 * count. We still want to be able to bring them online and offer
914 * access to the admin queue, as that might be only way to fix them up.
917 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
920 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
921 *count = min(*count, nr_io_queues);
926 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
928 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
930 struct nvme_user_io io;
931 struct nvme_command c;
932 unsigned length, meta_len;
933 void __user *metadata;
935 if (copy_from_user(&io, uio, sizeof(io)))
943 case nvme_cmd_compare:
949 length = (io.nblocks + 1) << ns->lba_shift;
950 meta_len = (io.nblocks + 1) * ns->ms;
951 metadata = (void __user *)(uintptr_t)io.metadata;
956 } else if (meta_len) {
957 if ((io.metadata & 3) || !io.metadata)
961 memset(&c, 0, sizeof(c));
962 c.rw.opcode = io.opcode;
963 c.rw.flags = io.flags;
964 c.rw.nsid = cpu_to_le32(ns->ns_id);
965 c.rw.slba = cpu_to_le64(io.slba);
966 c.rw.length = cpu_to_le16(io.nblocks);
967 c.rw.control = cpu_to_le16(io.control);
968 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
969 c.rw.reftag = cpu_to_le32(io.reftag);
970 c.rw.apptag = cpu_to_le16(io.apptag);
971 c.rw.appmask = cpu_to_le16(io.appmask);
973 return __nvme_submit_user_cmd(ns->queue, &c,
974 (void __user *)(uintptr_t)io.addr, length,
975 metadata, meta_len, io.slba, NULL, 0);
978 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
979 struct nvme_passthru_cmd __user *ucmd)
981 struct nvme_passthru_cmd cmd;
982 struct nvme_command c;
983 unsigned timeout = 0;
986 if (!capable(CAP_SYS_ADMIN))
988 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
993 memset(&c, 0, sizeof(c));
994 c.common.opcode = cmd.opcode;
995 c.common.flags = cmd.flags;
996 c.common.nsid = cpu_to_le32(cmd.nsid);
997 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
998 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
999 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1000 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1001 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1002 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1003 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1004 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1007 timeout = msecs_to_jiffies(cmd.timeout_ms);
1009 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1010 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1011 &cmd.result, timeout);
1013 if (put_user(cmd.result, &ucmd->result))
1020 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1021 unsigned int cmd, unsigned long arg)
1023 struct nvme_ns *ns = bdev->bd_disk->private_data;
1027 force_successful_syscall_return();
1029 case NVME_IOCTL_ADMIN_CMD:
1030 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1031 case NVME_IOCTL_IO_CMD:
1032 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1033 case NVME_IOCTL_SUBMIT_IO:
1034 return nvme_submit_io(ns, (void __user *)arg);
1038 return nvme_nvm_ioctl(ns, cmd, arg);
1040 if (is_sed_ioctl(cmd))
1041 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1042 (void __user *) arg);
1047 #ifdef CONFIG_COMPAT
1048 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1049 unsigned int cmd, unsigned long arg)
1051 return nvme_ioctl(bdev, mode, cmd, arg);
1054 #define nvme_compat_ioctl NULL
1057 static int nvme_open(struct block_device *bdev, fmode_t mode)
1059 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
1062 static void nvme_release(struct gendisk *disk, fmode_t mode)
1064 struct nvme_ns *ns = disk->private_data;
1066 module_put(ns->ctrl->ops->module);
1070 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1072 /* some standard values */
1073 geo->heads = 1 << 6;
1074 geo->sectors = 1 << 5;
1075 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1079 #ifdef CONFIG_BLK_DEV_INTEGRITY
1080 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1083 struct nvme_ns *ns = disk->private_data;
1084 u16 old_ms = ns->ms;
1087 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1088 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1090 /* PI implementation requires metadata equal t10 pi tuple size */
1091 if (ns->ms == sizeof(struct t10_pi_tuple))
1092 pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1094 if (blk_get_integrity(disk) &&
1095 (ns->pi_type != pi_type || ns->ms != old_ms ||
1096 bs != queue_logical_block_size(disk->queue) ||
1097 (ns->ms && ns->ext)))
1098 blk_integrity_unregister(disk);
1100 ns->pi_type = pi_type;
1103 static void nvme_init_integrity(struct nvme_ns *ns)
1105 struct blk_integrity integrity;
1107 memset(&integrity, 0, sizeof(integrity));
1108 switch (ns->pi_type) {
1109 case NVME_NS_DPS_PI_TYPE3:
1110 integrity.profile = &t10_pi_type3_crc;
1111 integrity.tag_size = sizeof(u16) + sizeof(u32);
1112 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1114 case NVME_NS_DPS_PI_TYPE1:
1115 case NVME_NS_DPS_PI_TYPE2:
1116 integrity.profile = &t10_pi_type1_crc;
1117 integrity.tag_size = sizeof(u16);
1118 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1121 integrity.profile = NULL;
1124 integrity.tuple_size = ns->ms;
1125 blk_integrity_register(ns->disk, &integrity);
1126 blk_queue_max_integrity_segments(ns->queue, 1);
1129 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1133 static void nvme_init_integrity(struct nvme_ns *ns)
1136 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1138 static void nvme_set_chunk_size(struct nvme_ns *ns)
1140 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1141 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1144 static void nvme_config_discard(struct nvme_ns *ns)
1146 struct nvme_ctrl *ctrl = ns->ctrl;
1147 u32 logical_block_size = queue_logical_block_size(ns->queue);
1149 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1150 NVME_DSM_MAX_RANGES);
1152 if (ctrl->nr_streams && ns->sws && ns->sgs) {
1153 unsigned int sz = logical_block_size * ns->sws * ns->sgs;
1155 ns->queue->limits.discard_alignment = sz;
1156 ns->queue->limits.discard_granularity = sz;
1158 ns->queue->limits.discard_alignment = logical_block_size;
1159 ns->queue->limits.discard_granularity = logical_block_size;
1161 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
1162 blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
1163 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1165 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1166 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
1169 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
1171 if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
1172 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
1176 if ((*id)->ncap == 0) {
1181 if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
1182 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
1183 if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
1184 memcpy(ns->nguid, (*id)->nguid, sizeof(ns->nguid));
1185 if (ns->ctrl->vs >= NVME_VS(1, 3, 0)) {
1186 /* Don't treat error as fatal we potentially
1187 * already have a NGUID or EUI-64
1189 if (nvme_identify_ns_descs(ns, ns->ns_id))
1190 dev_warn(ns->ctrl->device,
1191 "%s: Identify Descriptors failed\n", __func__);
1197 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1199 struct nvme_ns *ns = disk->private_data;
1200 struct nvme_ctrl *ctrl = ns->ctrl;
1204 * If identify namespace failed, use default 512 byte block size so
1205 * block layer can use before failing read/write for 0 capacity.
1207 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1208 if (ns->lba_shift == 0)
1210 bs = 1 << ns->lba_shift;
1211 ns->noiob = le16_to_cpu(id->noiob);
1213 blk_mq_freeze_queue(disk->queue);
1215 if (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1216 nvme_prep_integrity(disk, id, bs);
1217 blk_queue_logical_block_size(ns->queue, bs);
1219 nvme_set_chunk_size(ns);
1220 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1221 nvme_init_integrity(ns);
1222 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1223 set_capacity(disk, 0);
1225 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1227 if (ctrl->oncs & NVME_CTRL_ONCS_DSM)
1228 nvme_config_discard(ns);
1229 blk_mq_unfreeze_queue(disk->queue);
1232 static int nvme_revalidate_disk(struct gendisk *disk)
1234 struct nvme_ns *ns = disk->private_data;
1235 struct nvme_id_ns *id = NULL;
1238 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1239 set_capacity(disk, 0);
1243 ret = nvme_revalidate_ns(ns, &id);
1247 __nvme_revalidate_disk(disk, id);
1253 static char nvme_pr_type(enum pr_type type)
1256 case PR_WRITE_EXCLUSIVE:
1258 case PR_EXCLUSIVE_ACCESS:
1260 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1262 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1264 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1266 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1273 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1274 u64 key, u64 sa_key, u8 op)
1276 struct nvme_ns *ns = bdev->bd_disk->private_data;
1277 struct nvme_command c;
1278 u8 data[16] = { 0, };
1280 put_unaligned_le64(key, &data[0]);
1281 put_unaligned_le64(sa_key, &data[8]);
1283 memset(&c, 0, sizeof(c));
1284 c.common.opcode = op;
1285 c.common.nsid = cpu_to_le32(ns->ns_id);
1286 c.common.cdw10[0] = cpu_to_le32(cdw10);
1288 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1291 static int nvme_pr_register(struct block_device *bdev, u64 old,
1292 u64 new, unsigned flags)
1296 if (flags & ~PR_FL_IGNORE_KEY)
1299 cdw10 = old ? 2 : 0;
1300 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1301 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1302 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1305 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1306 enum pr_type type, unsigned flags)
1310 if (flags & ~PR_FL_IGNORE_KEY)
1313 cdw10 = nvme_pr_type(type) << 8;
1314 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1315 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1318 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1319 enum pr_type type, bool abort)
1321 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1322 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1325 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1327 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1328 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1331 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1333 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1334 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1337 static const struct pr_ops nvme_pr_ops = {
1338 .pr_register = nvme_pr_register,
1339 .pr_reserve = nvme_pr_reserve,
1340 .pr_release = nvme_pr_release,
1341 .pr_preempt = nvme_pr_preempt,
1342 .pr_clear = nvme_pr_clear,
1345 #ifdef CONFIG_BLK_SED_OPAL
1346 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1349 struct nvme_ctrl *ctrl = data;
1350 struct nvme_command cmd;
1352 memset(&cmd, 0, sizeof(cmd));
1354 cmd.common.opcode = nvme_admin_security_send;
1356 cmd.common.opcode = nvme_admin_security_recv;
1357 cmd.common.nsid = 0;
1358 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1359 cmd.common.cdw10[1] = cpu_to_le32(len);
1361 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1362 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1364 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1365 #endif /* CONFIG_BLK_SED_OPAL */
1367 static const struct block_device_operations nvme_fops = {
1368 .owner = THIS_MODULE,
1369 .ioctl = nvme_ioctl,
1370 .compat_ioctl = nvme_compat_ioctl,
1372 .release = nvme_release,
1373 .getgeo = nvme_getgeo,
1374 .revalidate_disk= nvme_revalidate_disk,
1375 .pr_ops = &nvme_pr_ops,
1378 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1380 unsigned long timeout =
1381 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1382 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1385 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1388 if ((csts & NVME_CSTS_RDY) == bit)
1392 if (fatal_signal_pending(current))
1394 if (time_after(jiffies, timeout)) {
1395 dev_err(ctrl->device,
1396 "Device not ready; aborting %s\n", enabled ?
1397 "initialisation" : "reset");
1406 * If the device has been passed off to us in an enabled state, just clear
1407 * the enabled bit. The spec says we should set the 'shutdown notification
1408 * bits', but doing so may cause the device to complete commands to the
1409 * admin queue ... and we don't know what memory that might be pointing at!
1411 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1415 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1416 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1418 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1422 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1423 msleep(NVME_QUIRK_DELAY_AMOUNT);
1425 return nvme_wait_ready(ctrl, cap, false);
1427 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1429 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1432 * Default to a 4K page size, with the intention to update this
1433 * path in the future to accomodate architectures with differing
1434 * kernel and IO page sizes.
1436 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1439 if (page_shift < dev_page_min) {
1440 dev_err(ctrl->device,
1441 "Minimum device page size %u too large for host (%u)\n",
1442 1 << dev_page_min, 1 << page_shift);
1446 ctrl->page_size = 1 << page_shift;
1448 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1449 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1450 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1451 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1452 ctrl->ctrl_config |= NVME_CC_ENABLE;
1454 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1457 return nvme_wait_ready(ctrl, cap, true);
1459 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1461 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1463 unsigned long timeout = jiffies + (shutdown_timeout * HZ);
1467 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1468 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1470 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1474 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1475 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1479 if (fatal_signal_pending(current))
1481 if (time_after(jiffies, timeout)) {
1482 dev_err(ctrl->device,
1483 "Device shutdown incomplete; abort shutdown\n");
1490 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1492 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1493 struct request_queue *q)
1497 if (ctrl->max_hw_sectors) {
1499 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1501 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1502 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1504 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1505 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1506 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1507 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1509 blk_queue_write_cache(q, vwc, vwc);
1512 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1515 * APST (Autonomous Power State Transition) lets us program a
1516 * table of power state transitions that the controller will
1517 * perform automatically. We configure it with a simple
1518 * heuristic: we are willing to spend at most 2% of the time
1519 * transitioning between power states. Therefore, when running
1520 * in any given state, we will enter the next lower-power
1521 * non-operational state after waiting 50 * (enlat + exlat)
1522 * microseconds, as long as that state's exit latency is under
1523 * the requested maximum latency.
1525 * We will not autonomously enter any non-operational state for
1526 * which the total latency exceeds ps_max_latency_us. Users
1527 * can set ps_max_latency_us to zero to turn off APST.
1531 struct nvme_feat_auto_pst *table;
1537 * If APST isn't supported or if we haven't been initialized yet,
1538 * then don't do anything.
1543 if (ctrl->npss > 31) {
1544 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1548 table = kzalloc(sizeof(*table), GFP_KERNEL);
1552 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1553 /* Turn off APST. */
1555 dev_dbg(ctrl->device, "APST disabled\n");
1557 __le64 target = cpu_to_le64(0);
1561 * Walk through all states from lowest- to highest-power.
1562 * According to the spec, lower-numbered states use more
1563 * power. NPSS, despite the name, is the index of the
1564 * lowest-power state, not the number of states.
1566 for (state = (int)ctrl->npss; state >= 0; state--) {
1567 u64 total_latency_us, exit_latency_us, transition_ms;
1570 table->entries[state] = target;
1573 * Don't allow transitions to the deepest state
1574 * if it's quirked off.
1576 if (state == ctrl->npss &&
1577 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1581 * Is this state a useful non-operational state for
1582 * higher-power states to autonomously transition to?
1584 if (!(ctrl->psd[state].flags &
1585 NVME_PS_FLAGS_NON_OP_STATE))
1589 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1590 if (exit_latency_us > ctrl->ps_max_latency_us)
1595 le32_to_cpu(ctrl->psd[state].entry_lat);
1598 * This state is good. Use it as the APST idle
1599 * target for higher power states.
1601 transition_ms = total_latency_us + 19;
1602 do_div(transition_ms, 20);
1603 if (transition_ms > (1 << 24) - 1)
1604 transition_ms = (1 << 24) - 1;
1606 target = cpu_to_le64((state << 3) |
1607 (transition_ms << 8));
1612 if (total_latency_us > max_lat_us)
1613 max_lat_us = total_latency_us;
1619 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1621 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1622 max_ps, max_lat_us, (int)sizeof(*table), table);
1626 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1627 table, sizeof(*table), NULL);
1629 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1635 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1637 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1641 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1642 case PM_QOS_LATENCY_ANY:
1650 if (ctrl->ps_max_latency_us != latency) {
1651 ctrl->ps_max_latency_us = latency;
1652 nvme_configure_apst(ctrl);
1656 struct nvme_core_quirk_entry {
1658 * NVMe model and firmware strings are padded with spaces. For
1659 * simplicity, strings in the quirk table are padded with NULLs
1665 unsigned long quirks;
1668 static const struct nvme_core_quirk_entry core_quirks[] = {
1671 * This Toshiba device seems to die using any APST states. See:
1672 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1675 .mn = "THNSF5256GPUK TOSHIBA",
1676 .quirks = NVME_QUIRK_NO_APST,
1680 /* match is null-terminated but idstr is space-padded. */
1681 static bool string_matches(const char *idstr, const char *match, size_t len)
1688 matchlen = strlen(match);
1689 WARN_ON_ONCE(matchlen > len);
1691 if (memcmp(idstr, match, matchlen))
1694 for (; matchlen < len; matchlen++)
1695 if (idstr[matchlen] != ' ')
1701 static bool quirk_matches(const struct nvme_id_ctrl *id,
1702 const struct nvme_core_quirk_entry *q)
1704 return q->vid == le16_to_cpu(id->vid) &&
1705 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1706 string_matches(id->fr, q->fr, sizeof(id->fr));
1709 static void nvme_init_subnqn(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
1714 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1715 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1716 strcpy(ctrl->subnqn, id->subnqn);
1720 if (ctrl->vs >= NVME_VS(1, 2, 1))
1721 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1723 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1724 off = snprintf(ctrl->subnqn, NVMF_NQN_SIZE,
1725 "nqn.2014.08.org.nvmexpress:%4x%4x",
1726 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1727 memcpy(ctrl->subnqn + off, id->sn, sizeof(id->sn));
1728 off += sizeof(id->sn);
1729 memcpy(ctrl->subnqn + off, id->mn, sizeof(id->mn));
1730 off += sizeof(id->mn);
1731 memset(ctrl->subnqn + off, 0, sizeof(ctrl->subnqn) - off);
1735 * Initialize the cached copies of the Identify data and various controller
1736 * register in our nvme_ctrl structure. This should be called as soon as
1737 * the admin queue is fully up and running.
1739 int nvme_init_identify(struct nvme_ctrl *ctrl)
1741 struct nvme_id_ctrl *id;
1743 int ret, page_shift;
1745 bool prev_apst_enabled;
1747 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1749 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1753 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1755 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1758 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1760 if (ctrl->vs >= NVME_VS(1, 1, 0))
1761 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1763 ret = nvme_identify_ctrl(ctrl, &id);
1765 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1769 nvme_init_subnqn(ctrl, id);
1771 if (!ctrl->identified) {
1773 * Check for quirks. Quirk can depend on firmware version,
1774 * so, in principle, the set of quirks present can change
1775 * across a reset. As a possible future enhancement, we
1776 * could re-scan for quirks every time we reinitialize
1777 * the device, but we'd have to make sure that the driver
1778 * behaves intelligently if the quirks change.
1783 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1784 if (quirk_matches(id, &core_quirks[i]))
1785 ctrl->quirks |= core_quirks[i].quirks;
1789 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1790 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1791 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1794 ctrl->oacs = le16_to_cpu(id->oacs);
1795 ctrl->vid = le16_to_cpu(id->vid);
1796 ctrl->oncs = le16_to_cpup(&id->oncs);
1797 atomic_set(&ctrl->abort_limit, id->acl + 1);
1798 ctrl->vwc = id->vwc;
1799 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1800 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1801 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1802 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1804 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1806 max_hw_sectors = UINT_MAX;
1807 ctrl->max_hw_sectors =
1808 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1810 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1811 ctrl->sgls = le32_to_cpu(id->sgls);
1812 ctrl->kas = le16_to_cpu(id->kas);
1814 ctrl->npss = id->npss;
1815 ctrl->apsta = id->apsta;
1816 prev_apst_enabled = ctrl->apst_enabled;
1817 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1818 if (force_apst && id->apsta) {
1819 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1820 ctrl->apst_enabled = true;
1822 ctrl->apst_enabled = false;
1825 ctrl->apst_enabled = id->apsta;
1827 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1829 if (ctrl->ops->flags & NVME_F_FABRICS) {
1830 ctrl->icdoff = le16_to_cpu(id->icdoff);
1831 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1832 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1833 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1836 * In fabrics we need to verify the cntlid matches the
1839 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
1844 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1845 dev_err(ctrl->device,
1846 "keep-alive support is mandatory for fabrics\n");
1851 ctrl->cntlid = le16_to_cpu(id->cntlid);
1852 ctrl->hmpre = le32_to_cpu(id->hmpre);
1853 ctrl->hmmin = le32_to_cpu(id->hmmin);
1858 if (ctrl->apst_enabled && !prev_apst_enabled)
1859 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1860 else if (!ctrl->apst_enabled && prev_apst_enabled)
1861 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1863 ret = nvme_configure_apst(ctrl);
1867 ret = nvme_configure_directives(ctrl);
1871 ctrl->identified = true;
1879 EXPORT_SYMBOL_GPL(nvme_init_identify);
1881 static int nvme_dev_open(struct inode *inode, struct file *file)
1883 struct nvme_ctrl *ctrl;
1884 int instance = iminor(inode);
1887 spin_lock(&dev_list_lock);
1888 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1889 if (ctrl->instance != instance)
1892 if (!ctrl->admin_q) {
1896 if (!kref_get_unless_zero(&ctrl->kref))
1898 file->private_data = ctrl;
1902 spin_unlock(&dev_list_lock);
1907 static int nvme_dev_release(struct inode *inode, struct file *file)
1909 nvme_put_ctrl(file->private_data);
1913 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1918 mutex_lock(&ctrl->namespaces_mutex);
1919 if (list_empty(&ctrl->namespaces)) {
1924 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1925 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1926 dev_warn(ctrl->device,
1927 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1932 dev_warn(ctrl->device,
1933 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1934 kref_get(&ns->kref);
1935 mutex_unlock(&ctrl->namespaces_mutex);
1937 ret = nvme_user_cmd(ctrl, ns, argp);
1942 mutex_unlock(&ctrl->namespaces_mutex);
1946 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1949 struct nvme_ctrl *ctrl = file->private_data;
1950 void __user *argp = (void __user *)arg;
1953 case NVME_IOCTL_ADMIN_CMD:
1954 return nvme_user_cmd(ctrl, NULL, argp);
1955 case NVME_IOCTL_IO_CMD:
1956 return nvme_dev_user_cmd(ctrl, argp);
1957 case NVME_IOCTL_RESET:
1958 dev_warn(ctrl->device, "resetting controller\n");
1959 return nvme_reset_ctrl_sync(ctrl);
1960 case NVME_IOCTL_SUBSYS_RESET:
1961 return nvme_reset_subsystem(ctrl);
1962 case NVME_IOCTL_RESCAN:
1963 nvme_queue_scan(ctrl);
1970 static const struct file_operations nvme_dev_fops = {
1971 .owner = THIS_MODULE,
1972 .open = nvme_dev_open,
1973 .release = nvme_dev_release,
1974 .unlocked_ioctl = nvme_dev_ioctl,
1975 .compat_ioctl = nvme_dev_ioctl,
1978 static ssize_t nvme_sysfs_reset(struct device *dev,
1979 struct device_attribute *attr, const char *buf,
1982 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1985 ret = nvme_reset_ctrl_sync(ctrl);
1990 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1992 static ssize_t nvme_sysfs_rescan(struct device *dev,
1993 struct device_attribute *attr, const char *buf,
1996 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1998 nvme_queue_scan(ctrl);
2001 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2003 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2006 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2007 struct nvme_ctrl *ctrl = ns->ctrl;
2008 int serial_len = sizeof(ctrl->serial);
2009 int model_len = sizeof(ctrl->model);
2011 if (!uuid_is_null(&ns->uuid))
2012 return sprintf(buf, "uuid.%pU\n", &ns->uuid);
2014 if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2015 return sprintf(buf, "eui.%16phN\n", ns->nguid);
2017 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2018 return sprintf(buf, "eui.%8phN\n", ns->eui);
2020 while (serial_len > 0 && (ctrl->serial[serial_len - 1] == ' ' ||
2021 ctrl->serial[serial_len - 1] == '\0'))
2023 while (model_len > 0 && (ctrl->model[model_len - 1] == ' ' ||
2024 ctrl->model[model_len - 1] == '\0'))
2027 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
2028 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
2030 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2032 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2035 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2036 return sprintf(buf, "%pU\n", ns->nguid);
2038 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2040 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2043 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2045 /* For backward compatibility expose the NGUID to userspace if
2046 * we have no UUID set
2048 if (uuid_is_null(&ns->uuid)) {
2049 printk_ratelimited(KERN_WARNING
2050 "No UUID available providing old NGUID\n");
2051 return sprintf(buf, "%pU\n", ns->nguid);
2053 return sprintf(buf, "%pU\n", &ns->uuid);
2055 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2057 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2060 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2061 return sprintf(buf, "%8phd\n", ns->eui);
2063 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2065 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2068 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2069 return sprintf(buf, "%d\n", ns->ns_id);
2071 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2073 static struct attribute *nvme_ns_attrs[] = {
2074 &dev_attr_wwid.attr,
2075 &dev_attr_uuid.attr,
2076 &dev_attr_nguid.attr,
2078 &dev_attr_nsid.attr,
2082 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
2083 struct attribute *a, int n)
2085 struct device *dev = container_of(kobj, struct device, kobj);
2086 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2088 if (a == &dev_attr_uuid.attr) {
2089 if (uuid_is_null(&ns->uuid) ||
2090 !memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2093 if (a == &dev_attr_nguid.attr) {
2094 if (!memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2097 if (a == &dev_attr_eui.attr) {
2098 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2104 static const struct attribute_group nvme_ns_attr_group = {
2105 .attrs = nvme_ns_attrs,
2106 .is_visible = nvme_ns_attrs_are_visible,
2109 #define nvme_show_str_function(field) \
2110 static ssize_t field##_show(struct device *dev, \
2111 struct device_attribute *attr, char *buf) \
2113 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2114 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
2116 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2118 #define nvme_show_int_function(field) \
2119 static ssize_t field##_show(struct device *dev, \
2120 struct device_attribute *attr, char *buf) \
2122 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2123 return sprintf(buf, "%d\n", ctrl->field); \
2125 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2127 nvme_show_str_function(model);
2128 nvme_show_str_function(serial);
2129 nvme_show_str_function(firmware_rev);
2130 nvme_show_int_function(cntlid);
2132 static ssize_t nvme_sysfs_delete(struct device *dev,
2133 struct device_attribute *attr, const char *buf,
2136 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2138 if (device_remove_file_self(dev, attr))
2139 ctrl->ops->delete_ctrl(ctrl);
2142 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2144 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2145 struct device_attribute *attr,
2148 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2150 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2152 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2154 static ssize_t nvme_sysfs_show_state(struct device *dev,
2155 struct device_attribute *attr,
2158 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2159 static const char *const state_name[] = {
2160 [NVME_CTRL_NEW] = "new",
2161 [NVME_CTRL_LIVE] = "live",
2162 [NVME_CTRL_RESETTING] = "resetting",
2163 [NVME_CTRL_RECONNECTING]= "reconnecting",
2164 [NVME_CTRL_DELETING] = "deleting",
2165 [NVME_CTRL_DEAD] = "dead",
2168 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2169 state_name[ctrl->state])
2170 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2172 return sprintf(buf, "unknown state\n");
2175 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2177 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2178 struct device_attribute *attr,
2181 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2183 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subnqn);
2185 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2187 static ssize_t nvme_sysfs_show_address(struct device *dev,
2188 struct device_attribute *attr,
2191 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2193 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2195 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2197 static struct attribute *nvme_dev_attrs[] = {
2198 &dev_attr_reset_controller.attr,
2199 &dev_attr_rescan_controller.attr,
2200 &dev_attr_model.attr,
2201 &dev_attr_serial.attr,
2202 &dev_attr_firmware_rev.attr,
2203 &dev_attr_cntlid.attr,
2204 &dev_attr_delete_controller.attr,
2205 &dev_attr_transport.attr,
2206 &dev_attr_subsysnqn.attr,
2207 &dev_attr_address.attr,
2208 &dev_attr_state.attr,
2212 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2213 struct attribute *a, int n)
2215 struct device *dev = container_of(kobj, struct device, kobj);
2216 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2218 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2220 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2226 static struct attribute_group nvme_dev_attrs_group = {
2227 .attrs = nvme_dev_attrs,
2228 .is_visible = nvme_dev_attrs_are_visible,
2231 static const struct attribute_group *nvme_dev_attr_groups[] = {
2232 &nvme_dev_attrs_group,
2236 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2238 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2239 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2241 return nsa->ns_id - nsb->ns_id;
2244 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2246 struct nvme_ns *ns, *ret = NULL;
2248 mutex_lock(&ctrl->namespaces_mutex);
2249 list_for_each_entry(ns, &ctrl->namespaces, list) {
2250 if (ns->ns_id == nsid) {
2251 kref_get(&ns->kref);
2255 if (ns->ns_id > nsid)
2258 mutex_unlock(&ctrl->namespaces_mutex);
2262 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2264 struct streams_directive_params s;
2267 if (!ctrl->nr_streams)
2270 ret = nvme_get_stream_params(ctrl, &s, ns->ns_id);
2274 ns->sws = le32_to_cpu(s.sws);
2275 ns->sgs = le16_to_cpu(s.sgs);
2278 unsigned int bs = 1 << ns->lba_shift;
2280 blk_queue_io_min(ns->queue, bs * ns->sws);
2282 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2288 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2291 struct gendisk *disk;
2292 struct nvme_id_ns *id;
2293 char disk_name[DISK_NAME_LEN];
2294 int node = dev_to_node(ctrl->dev);
2296 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2300 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2301 if (ns->instance < 0)
2304 ns->queue = blk_mq_init_queue(ctrl->tagset);
2305 if (IS_ERR(ns->queue))
2306 goto out_release_instance;
2307 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2308 ns->queue->queuedata = ns;
2311 kref_init(&ns->kref);
2313 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2315 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2316 nvme_set_queue_limits(ctrl, ns->queue);
2317 nvme_setup_streams_ns(ctrl, ns);
2319 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2321 if (nvme_revalidate_ns(ns, &id))
2322 goto out_free_queue;
2324 if (nvme_nvm_ns_supported(ns, id) &&
2325 nvme_nvm_register(ns, disk_name, node)) {
2326 dev_warn(ctrl->device, "%s: LightNVM init failure\n", __func__);
2330 disk = alloc_disk_node(0, node);
2334 disk->fops = &nvme_fops;
2335 disk->private_data = ns;
2336 disk->queue = ns->queue;
2337 disk->flags = GENHD_FL_EXT_DEVT;
2338 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2341 __nvme_revalidate_disk(disk, id);
2343 mutex_lock(&ctrl->namespaces_mutex);
2344 list_add_tail(&ns->list, &ctrl->namespaces);
2345 mutex_unlock(&ctrl->namespaces_mutex);
2347 kref_get(&ctrl->kref);
2351 device_add_disk(ctrl->device, ns->disk);
2352 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2353 &nvme_ns_attr_group))
2354 pr_warn("%s: failed to create sysfs group for identification\n",
2355 ns->disk->disk_name);
2356 if (ns->ndev && nvme_nvm_register_sysfs(ns))
2357 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2358 ns->disk->disk_name);
2363 blk_cleanup_queue(ns->queue);
2364 out_release_instance:
2365 ida_simple_remove(&ctrl->ns_ida, ns->instance);
2370 static void nvme_ns_remove(struct nvme_ns *ns)
2372 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2375 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2376 if (blk_get_integrity(ns->disk))
2377 blk_integrity_unregister(ns->disk);
2378 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2379 &nvme_ns_attr_group);
2381 nvme_nvm_unregister_sysfs(ns);
2382 del_gendisk(ns->disk);
2383 blk_cleanup_queue(ns->queue);
2386 mutex_lock(&ns->ctrl->namespaces_mutex);
2387 list_del_init(&ns->list);
2388 mutex_unlock(&ns->ctrl->namespaces_mutex);
2393 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2397 ns = nvme_find_get_ns(ctrl, nsid);
2399 if (ns->disk && revalidate_disk(ns->disk))
2403 nvme_alloc_ns(ctrl, nsid);
2406 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2409 struct nvme_ns *ns, *next;
2411 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2412 if (ns->ns_id > nsid)
2417 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2421 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2424 ns_list = kzalloc(0x1000, GFP_KERNEL);
2428 for (i = 0; i < num_lists; i++) {
2429 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2433 for (j = 0; j < min(nn, 1024U); j++) {
2434 nsid = le32_to_cpu(ns_list[j]);
2438 nvme_validate_ns(ctrl, nsid);
2440 while (++prev < nsid) {
2441 ns = nvme_find_get_ns(ctrl, prev);
2451 nvme_remove_invalid_namespaces(ctrl, prev);
2457 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2461 for (i = 1; i <= nn; i++)
2462 nvme_validate_ns(ctrl, i);
2464 nvme_remove_invalid_namespaces(ctrl, nn);
2467 static void nvme_scan_work(struct work_struct *work)
2469 struct nvme_ctrl *ctrl =
2470 container_of(work, struct nvme_ctrl, scan_work);
2471 struct nvme_id_ctrl *id;
2474 if (ctrl->state != NVME_CTRL_LIVE)
2477 if (nvme_identify_ctrl(ctrl, &id))
2480 nn = le32_to_cpu(id->nn);
2481 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2482 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2483 if (!nvme_scan_ns_list(ctrl, nn))
2486 nvme_scan_ns_sequential(ctrl, nn);
2488 mutex_lock(&ctrl->namespaces_mutex);
2489 list_sort(NULL, &ctrl->namespaces, ns_cmp);
2490 mutex_unlock(&ctrl->namespaces_mutex);
2494 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2497 * Do not queue new scan work when a controller is reset during
2500 if (ctrl->state == NVME_CTRL_LIVE)
2501 queue_work(nvme_wq, &ctrl->scan_work);
2503 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2506 * This function iterates the namespace list unlocked to allow recovery from
2507 * controller failure. It is up to the caller to ensure the namespace list is
2508 * not modified by scan work while this function is executing.
2510 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2512 struct nvme_ns *ns, *next;
2515 * The dead states indicates the controller was not gracefully
2516 * disconnected. In that case, we won't be able to flush any data while
2517 * removing the namespaces' disks; fail all the queues now to avoid
2518 * potentially having to clean up the failed sync later.
2520 if (ctrl->state == NVME_CTRL_DEAD)
2521 nvme_kill_queues(ctrl);
2523 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2526 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2528 static void nvme_async_event_work(struct work_struct *work)
2530 struct nvme_ctrl *ctrl =
2531 container_of(work, struct nvme_ctrl, async_event_work);
2533 spin_lock_irq(&ctrl->lock);
2534 while (ctrl->event_limit > 0) {
2535 int aer_idx = --ctrl->event_limit;
2537 spin_unlock_irq(&ctrl->lock);
2538 ctrl->ops->submit_async_event(ctrl, aer_idx);
2539 spin_lock_irq(&ctrl->lock);
2541 spin_unlock_irq(&ctrl->lock);
2544 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2545 union nvme_result *res)
2547 u32 result = le32_to_cpu(res->u32);
2550 switch (le16_to_cpu(status) >> 1) {
2551 case NVME_SC_SUCCESS:
2554 case NVME_SC_ABORT_REQ:
2555 ++ctrl->event_limit;
2556 queue_work(nvme_wq, &ctrl->async_event_work);
2565 switch (result & 0xff07) {
2566 case NVME_AER_NOTICE_NS_CHANGED:
2567 dev_info(ctrl->device, "rescanning\n");
2568 nvme_queue_scan(ctrl);
2571 dev_warn(ctrl->device, "async event result %08x\n", result);
2574 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2576 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2578 ctrl->event_limit = NVME_NR_AERS;
2579 queue_work(nvme_wq, &ctrl->async_event_work);
2581 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2583 static DEFINE_IDA(nvme_instance_ida);
2585 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2587 int instance, error;
2590 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2593 spin_lock(&dev_list_lock);
2594 error = ida_get_new(&nvme_instance_ida, &instance);
2595 spin_unlock(&dev_list_lock);
2596 } while (error == -EAGAIN);
2601 ctrl->instance = instance;
2605 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2607 spin_lock(&dev_list_lock);
2608 ida_remove(&nvme_instance_ida, ctrl->instance);
2609 spin_unlock(&dev_list_lock);
2612 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
2614 nvme_stop_keep_alive(ctrl);
2615 flush_work(&ctrl->async_event_work);
2616 flush_work(&ctrl->scan_work);
2618 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
2620 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
2623 nvme_start_keep_alive(ctrl);
2625 if (ctrl->queue_count > 1) {
2626 nvme_queue_scan(ctrl);
2627 nvme_queue_async_events(ctrl);
2628 nvme_start_queues(ctrl);
2631 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
2633 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2635 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2637 spin_lock(&dev_list_lock);
2638 list_del(&ctrl->node);
2639 spin_unlock(&dev_list_lock);
2641 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2643 static void nvme_free_ctrl(struct kref *kref)
2645 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2647 put_device(ctrl->device);
2648 nvme_release_instance(ctrl);
2649 ida_destroy(&ctrl->ns_ida);
2651 ctrl->ops->free_ctrl(ctrl);
2654 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2656 kref_put(&ctrl->kref, nvme_free_ctrl);
2658 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2661 * Initialize a NVMe controller structures. This needs to be called during
2662 * earliest initialization so that we have the initialized structured around
2665 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2666 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2670 ctrl->state = NVME_CTRL_NEW;
2671 spin_lock_init(&ctrl->lock);
2672 INIT_LIST_HEAD(&ctrl->namespaces);
2673 mutex_init(&ctrl->namespaces_mutex);
2674 kref_init(&ctrl->kref);
2677 ctrl->quirks = quirks;
2678 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2679 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2681 ret = nvme_set_instance(ctrl);
2685 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2686 MKDEV(nvme_char_major, ctrl->instance),
2687 ctrl, nvme_dev_attr_groups,
2688 "nvme%d", ctrl->instance);
2689 if (IS_ERR(ctrl->device)) {
2690 ret = PTR_ERR(ctrl->device);
2691 goto out_release_instance;
2693 get_device(ctrl->device);
2694 ida_init(&ctrl->ns_ida);
2696 spin_lock(&dev_list_lock);
2697 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2698 spin_unlock(&dev_list_lock);
2701 * Initialize latency tolerance controls. The sysfs files won't
2702 * be visible to userspace unless the device actually supports APST.
2704 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2705 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2706 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2709 out_release_instance:
2710 nvme_release_instance(ctrl);
2714 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2717 * nvme_kill_queues(): Ends all namespace queues
2718 * @ctrl: the dead controller that needs to end
2720 * Call this function when the driver determines it is unable to get the
2721 * controller in a state capable of servicing IO.
2723 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2727 mutex_lock(&ctrl->namespaces_mutex);
2729 /* Forcibly unquiesce queues to avoid blocking dispatch */
2731 blk_mq_unquiesce_queue(ctrl->admin_q);
2733 list_for_each_entry(ns, &ctrl->namespaces, list) {
2735 * Revalidating a dead namespace sets capacity to 0. This will
2736 * end buffered writers dirtying pages that can't be synced.
2738 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2740 revalidate_disk(ns->disk);
2741 blk_set_queue_dying(ns->queue);
2743 /* Forcibly unquiesce queues to avoid blocking dispatch */
2744 blk_mq_unquiesce_queue(ns->queue);
2746 mutex_unlock(&ctrl->namespaces_mutex);
2748 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2750 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2754 mutex_lock(&ctrl->namespaces_mutex);
2755 list_for_each_entry(ns, &ctrl->namespaces, list)
2756 blk_mq_unfreeze_queue(ns->queue);
2757 mutex_unlock(&ctrl->namespaces_mutex);
2759 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2761 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2765 mutex_lock(&ctrl->namespaces_mutex);
2766 list_for_each_entry(ns, &ctrl->namespaces, list) {
2767 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2771 mutex_unlock(&ctrl->namespaces_mutex);
2773 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2775 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2779 mutex_lock(&ctrl->namespaces_mutex);
2780 list_for_each_entry(ns, &ctrl->namespaces, list)
2781 blk_mq_freeze_queue_wait(ns->queue);
2782 mutex_unlock(&ctrl->namespaces_mutex);
2784 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2786 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2790 mutex_lock(&ctrl->namespaces_mutex);
2791 list_for_each_entry(ns, &ctrl->namespaces, list)
2792 blk_freeze_queue_start(ns->queue);
2793 mutex_unlock(&ctrl->namespaces_mutex);
2795 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2797 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2801 mutex_lock(&ctrl->namespaces_mutex);
2802 list_for_each_entry(ns, &ctrl->namespaces, list)
2803 blk_mq_quiesce_queue(ns->queue);
2804 mutex_unlock(&ctrl->namespaces_mutex);
2806 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2808 void nvme_start_queues(struct nvme_ctrl *ctrl)
2812 mutex_lock(&ctrl->namespaces_mutex);
2813 list_for_each_entry(ns, &ctrl->namespaces, list)
2814 blk_mq_unquiesce_queue(ns->queue);
2815 mutex_unlock(&ctrl->namespaces_mutex);
2817 EXPORT_SYMBOL_GPL(nvme_start_queues);
2819 int __init nvme_core_init(void)
2823 nvme_wq = alloc_workqueue("nvme-wq",
2824 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
2828 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2832 else if (result > 0)
2833 nvme_char_major = result;
2835 nvme_class = class_create(THIS_MODULE, "nvme");
2836 if (IS_ERR(nvme_class)) {
2837 result = PTR_ERR(nvme_class);
2838 goto unregister_chrdev;
2844 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2846 destroy_workqueue(nvme_wq);
2850 void nvme_core_exit(void)
2852 class_destroy(nvme_class);
2853 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2854 destroy_workqueue(nvme_wq);
2857 MODULE_LICENSE("GPL");
2858 MODULE_VERSION("1.0");
2859 module_init(nvme_core_init);
2860 module_exit(nvme_core_exit);