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 int admin_timeout = 60;
38 module_param(admin_timeout, uint, 0644);
39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout);
42 unsigned int nvme_io_timeout = 30;
43 module_param_named(io_timeout, nvme_io_timeout, uint, 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 unsigned long default_ps_max_latency_us = 100000;
56 module_param(default_ps_max_latency_us, ulong, 0644);
57 MODULE_PARM_DESC(default_ps_max_latency_us,
58 "max power saving latency for new devices; use PM QOS to change per device");
60 static bool force_apst;
61 module_param(force_apst, bool, 0644);
62 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
65 module_param(streams, bool, 0644);
66 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
68 struct workqueue_struct *nvme_wq;
69 EXPORT_SYMBOL_GPL(nvme_wq);
71 static DEFINE_IDA(nvme_subsystems_ida);
72 static LIST_HEAD(nvme_subsystems);
73 static DEFINE_MUTEX(nvme_subsystems_lock);
75 static DEFINE_IDA(nvme_instance_ida);
76 static dev_t nvme_chr_devt;
77 static struct class *nvme_class;
78 static struct class *nvme_subsys_class;
80 static void nvme_ns_remove(struct nvme_ns *ns);
81 static int nvme_revalidate_disk(struct gendisk *disk);
83 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
85 return cpu_to_le32((((size / 4) - 1) << 16) | lid);
88 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
90 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
92 if (!queue_work(nvme_wq, &ctrl->reset_work))
96 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
98 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
102 ret = nvme_reset_ctrl(ctrl);
104 flush_work(&ctrl->reset_work);
108 static void nvme_delete_ctrl_work(struct work_struct *work)
110 struct nvme_ctrl *ctrl =
111 container_of(work, struct nvme_ctrl, delete_work);
113 flush_work(&ctrl->reset_work);
114 nvme_stop_ctrl(ctrl);
115 nvme_remove_namespaces(ctrl);
116 ctrl->ops->delete_ctrl(ctrl);
117 nvme_uninit_ctrl(ctrl);
121 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
123 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
125 if (!queue_work(nvme_wq, &ctrl->delete_work))
129 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
131 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
136 * Keep a reference until the work is flushed since ->delete_ctrl
137 * can free the controller.
140 ret = nvme_delete_ctrl(ctrl);
142 flush_work(&ctrl->delete_work);
146 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
148 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
150 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
153 static blk_status_t nvme_error_status(struct request *req)
155 switch (nvme_req(req)->status & 0x7ff) {
156 case NVME_SC_SUCCESS:
158 case NVME_SC_CAP_EXCEEDED:
159 return BLK_STS_NOSPC;
160 case NVME_SC_ONCS_NOT_SUPPORTED:
161 return BLK_STS_NOTSUPP;
162 case NVME_SC_WRITE_FAULT:
163 case NVME_SC_READ_ERROR:
164 case NVME_SC_UNWRITTEN_BLOCK:
165 case NVME_SC_ACCESS_DENIED:
166 case NVME_SC_READ_ONLY:
167 return BLK_STS_MEDIUM;
168 case NVME_SC_GUARD_CHECK:
169 case NVME_SC_APPTAG_CHECK:
170 case NVME_SC_REFTAG_CHECK:
171 case NVME_SC_INVALID_PI:
172 return BLK_STS_PROTECTION;
173 case NVME_SC_RESERVATION_CONFLICT:
174 return BLK_STS_NEXUS;
176 return BLK_STS_IOERR;
180 static inline bool nvme_req_needs_retry(struct request *req)
182 if (blk_noretry_request(req))
184 if (nvme_req(req)->status & NVME_SC_DNR)
186 if (nvme_req(req)->retries >= nvme_max_retries)
191 void nvme_complete_rq(struct request *req)
193 if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
194 if (nvme_req_needs_failover(req)) {
195 nvme_failover_req(req);
199 if (!blk_queue_dying(req->q)) {
200 nvme_req(req)->retries++;
201 blk_mq_requeue_request(req, true);
206 blk_mq_end_request(req, nvme_error_status(req));
208 EXPORT_SYMBOL_GPL(nvme_complete_rq);
210 void nvme_cancel_request(struct request *req, void *data, bool reserved)
212 if (!blk_mq_request_started(req))
215 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
216 "Cancelling I/O %d", req->tag);
218 nvme_req(req)->status = NVME_SC_ABORT_REQ;
219 blk_mq_complete_request(req);
222 EXPORT_SYMBOL_GPL(nvme_cancel_request);
224 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
225 enum nvme_ctrl_state new_state)
227 enum nvme_ctrl_state old_state;
229 bool changed = false;
231 spin_lock_irqsave(&ctrl->lock, flags);
233 old_state = ctrl->state;
238 case NVME_CTRL_RESETTING:
239 case NVME_CTRL_RECONNECTING:
246 case NVME_CTRL_RESETTING:
256 case NVME_CTRL_RECONNECTING:
259 case NVME_CTRL_RESETTING:
266 case NVME_CTRL_DELETING:
269 case NVME_CTRL_RESETTING:
270 case NVME_CTRL_RECONNECTING:
279 case NVME_CTRL_DELETING:
291 ctrl->state = new_state;
293 spin_unlock_irqrestore(&ctrl->lock, flags);
294 if (changed && ctrl->state == NVME_CTRL_LIVE)
295 nvme_kick_requeue_lists(ctrl);
298 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
300 static void nvme_free_ns_head(struct kref *ref)
302 struct nvme_ns_head *head =
303 container_of(ref, struct nvme_ns_head, ref);
305 nvme_mpath_remove_disk(head);
306 ida_simple_remove(&head->subsys->ns_ida, head->instance);
307 list_del_init(&head->entry);
308 cleanup_srcu_struct(&head->srcu);
312 static void nvme_put_ns_head(struct nvme_ns_head *head)
314 kref_put(&head->ref, nvme_free_ns_head);
317 static void nvme_free_ns(struct kref *kref)
319 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
322 nvme_nvm_unregister(ns);
325 nvme_put_ns_head(ns->head);
326 nvme_put_ctrl(ns->ctrl);
330 static void nvme_put_ns(struct nvme_ns *ns)
332 kref_put(&ns->kref, nvme_free_ns);
335 struct request *nvme_alloc_request(struct request_queue *q,
336 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
338 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
341 if (qid == NVME_QID_ANY) {
342 req = blk_mq_alloc_request(q, op, flags);
344 req = blk_mq_alloc_request_hctx(q, op, flags,
350 req->cmd_flags |= REQ_FAILFAST_DRIVER;
351 nvme_req(req)->cmd = cmd;
355 EXPORT_SYMBOL_GPL(nvme_alloc_request);
357 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
359 struct nvme_command c;
361 memset(&c, 0, sizeof(c));
363 c.directive.opcode = nvme_admin_directive_send;
364 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
365 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
366 c.directive.dtype = NVME_DIR_IDENTIFY;
367 c.directive.tdtype = NVME_DIR_STREAMS;
368 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
370 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
373 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
375 return nvme_toggle_streams(ctrl, false);
378 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
380 return nvme_toggle_streams(ctrl, true);
383 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
384 struct streams_directive_params *s, u32 nsid)
386 struct nvme_command c;
388 memset(&c, 0, sizeof(c));
389 memset(s, 0, sizeof(*s));
391 c.directive.opcode = nvme_admin_directive_recv;
392 c.directive.nsid = cpu_to_le32(nsid);
393 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
394 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
395 c.directive.dtype = NVME_DIR_STREAMS;
397 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
400 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
402 struct streams_directive_params s;
405 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
410 ret = nvme_enable_streams(ctrl);
414 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
418 ctrl->nssa = le16_to_cpu(s.nssa);
419 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
420 dev_info(ctrl->device, "too few streams (%u) available\n",
422 nvme_disable_streams(ctrl);
426 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
427 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
432 * Check if 'req' has a write hint associated with it. If it does, assign
433 * a valid namespace stream to the write.
435 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
436 struct request *req, u16 *control,
439 enum rw_hint streamid = req->write_hint;
441 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
445 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
448 *control |= NVME_RW_DTYPE_STREAMS;
449 *dsmgmt |= streamid << 16;
452 if (streamid < ARRAY_SIZE(req->q->write_hints))
453 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
456 static inline void nvme_setup_flush(struct nvme_ns *ns,
457 struct nvme_command *cmnd)
459 memset(cmnd, 0, sizeof(*cmnd));
460 cmnd->common.opcode = nvme_cmd_flush;
461 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
464 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
465 struct nvme_command *cmnd)
467 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
468 struct nvme_dsm_range *range;
471 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
473 return BLK_STS_RESOURCE;
475 __rq_for_each_bio(bio, req) {
476 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
477 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
479 range[n].cattr = cpu_to_le32(0);
480 range[n].nlb = cpu_to_le32(nlb);
481 range[n].slba = cpu_to_le64(slba);
485 if (WARN_ON_ONCE(n != segments)) {
487 return BLK_STS_IOERR;
490 memset(cmnd, 0, sizeof(*cmnd));
491 cmnd->dsm.opcode = nvme_cmd_dsm;
492 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
493 cmnd->dsm.nr = cpu_to_le32(segments - 1);
494 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
496 req->special_vec.bv_page = virt_to_page(range);
497 req->special_vec.bv_offset = offset_in_page(range);
498 req->special_vec.bv_len = sizeof(*range) * segments;
499 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
504 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
505 struct request *req, struct nvme_command *cmnd)
507 struct nvme_ctrl *ctrl = ns->ctrl;
511 if (req->cmd_flags & REQ_FUA)
512 control |= NVME_RW_FUA;
513 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
514 control |= NVME_RW_LR;
516 if (req->cmd_flags & REQ_RAHEAD)
517 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
519 memset(cmnd, 0, sizeof(*cmnd));
520 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
521 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
522 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
523 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
525 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
526 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
530 * If formated with metadata, the block layer always provides a
531 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
532 * we enable the PRACT bit for protection information or set the
533 * namespace capacity to zero to prevent any I/O.
535 if (!blk_integrity_rq(req)) {
536 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
537 return BLK_STS_NOTSUPP;
538 control |= NVME_RW_PRINFO_PRACT;
541 switch (ns->pi_type) {
542 case NVME_NS_DPS_PI_TYPE3:
543 control |= NVME_RW_PRINFO_PRCHK_GUARD;
545 case NVME_NS_DPS_PI_TYPE1:
546 case NVME_NS_DPS_PI_TYPE2:
547 control |= NVME_RW_PRINFO_PRCHK_GUARD |
548 NVME_RW_PRINFO_PRCHK_REF;
549 cmnd->rw.reftag = cpu_to_le32(
550 nvme_block_nr(ns, blk_rq_pos(req)));
555 cmnd->rw.control = cpu_to_le16(control);
556 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
560 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
561 struct nvme_command *cmd)
563 blk_status_t ret = BLK_STS_OK;
565 if (!(req->rq_flags & RQF_DONTPREP)) {
566 nvme_req(req)->retries = 0;
567 nvme_req(req)->flags = 0;
568 req->rq_flags |= RQF_DONTPREP;
571 switch (req_op(req)) {
574 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
577 nvme_setup_flush(ns, cmd);
579 case REQ_OP_WRITE_ZEROES:
580 /* currently only aliased to deallocate for a few ctrls: */
582 ret = nvme_setup_discard(ns, req, cmd);
586 ret = nvme_setup_rw(ns, req, cmd);
590 return BLK_STS_IOERR;
593 cmd->common.command_id = req->tag;
596 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
599 * Returns 0 on success. If the result is negative, it's a Linux error code;
600 * if the result is positive, it's an NVM Express status code
602 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
603 union nvme_result *result, void *buffer, unsigned bufflen,
604 unsigned timeout, int qid, int at_head,
605 blk_mq_req_flags_t flags)
610 req = nvme_alloc_request(q, cmd, flags, qid);
614 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
616 if (buffer && bufflen) {
617 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
622 blk_execute_rq(req->q, NULL, req, at_head);
624 *result = nvme_req(req)->result;
625 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
628 ret = nvme_req(req)->status;
630 blk_mq_free_request(req);
633 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
635 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
636 void *buffer, unsigned bufflen)
638 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
641 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
643 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
644 unsigned len, u32 seed, bool write)
646 struct bio_integrity_payload *bip;
650 buf = kmalloc(len, GFP_KERNEL);
655 if (write && copy_from_user(buf, ubuf, len))
658 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
664 bip->bip_iter.bi_size = len;
665 bip->bip_iter.bi_sector = seed;
666 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
667 offset_in_page(buf));
677 static int nvme_submit_user_cmd(struct request_queue *q,
678 struct nvme_command *cmd, void __user *ubuffer,
679 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
680 u32 meta_seed, u32 *result, unsigned timeout)
682 bool write = nvme_is_write(cmd);
683 struct nvme_ns *ns = q->queuedata;
684 struct gendisk *disk = ns ? ns->disk : NULL;
686 struct bio *bio = NULL;
690 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
694 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
696 if (ubuffer && bufflen) {
697 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
703 if (disk && meta_buffer && meta_len) {
704 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
713 blk_execute_rq(req->q, disk, req, 0);
714 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
717 ret = nvme_req(req)->status;
719 *result = le32_to_cpu(nvme_req(req)->result.u32);
720 if (meta && !ret && !write) {
721 if (copy_to_user(meta_buffer, meta, meta_len))
727 blk_rq_unmap_user(bio);
729 blk_mq_free_request(req);
733 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
735 struct nvme_ctrl *ctrl = rq->end_io_data;
737 blk_mq_free_request(rq);
740 dev_err(ctrl->device,
741 "failed nvme_keep_alive_end_io error=%d\n",
746 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
749 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
751 struct nvme_command c;
754 memset(&c, 0, sizeof(c));
755 c.common.opcode = nvme_admin_keep_alive;
757 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
762 rq->timeout = ctrl->kato * HZ;
763 rq->end_io_data = ctrl;
765 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
770 static void nvme_keep_alive_work(struct work_struct *work)
772 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
773 struct nvme_ctrl, ka_work);
775 if (nvme_keep_alive(ctrl)) {
776 /* allocation failure, reset the controller */
777 dev_err(ctrl->device, "keep-alive failed\n");
778 nvme_reset_ctrl(ctrl);
783 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
785 if (unlikely(ctrl->kato == 0))
788 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
789 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
791 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
793 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
795 if (unlikely(ctrl->kato == 0))
798 cancel_delayed_work_sync(&ctrl->ka_work);
800 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
802 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
804 struct nvme_command c = { };
807 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
808 c.identify.opcode = nvme_admin_identify;
809 c.identify.cns = NVME_ID_CNS_CTRL;
811 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
815 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
816 sizeof(struct nvme_id_ctrl));
822 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
823 struct nvme_ns_ids *ids)
825 struct nvme_command c = { };
831 c.identify.opcode = nvme_admin_identify;
832 c.identify.nsid = cpu_to_le32(nsid);
833 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
835 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
839 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
840 NVME_IDENTIFY_DATA_SIZE);
844 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
845 struct nvme_ns_id_desc *cur = data + pos;
851 case NVME_NIDT_EUI64:
852 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
853 dev_warn(ctrl->device,
854 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
858 len = NVME_NIDT_EUI64_LEN;
859 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
861 case NVME_NIDT_NGUID:
862 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
863 dev_warn(ctrl->device,
864 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
868 len = NVME_NIDT_NGUID_LEN;
869 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
872 if (cur->nidl != NVME_NIDT_UUID_LEN) {
873 dev_warn(ctrl->device,
874 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
878 len = NVME_NIDT_UUID_LEN;
879 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
882 /* Skip unnkown types */
894 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
896 struct nvme_command c = { };
898 c.identify.opcode = nvme_admin_identify;
899 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
900 c.identify.nsid = cpu_to_le32(nsid);
901 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
904 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
907 struct nvme_id_ns *id;
908 struct nvme_command c = { };
911 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
912 c.identify.opcode = nvme_admin_identify;
913 c.identify.nsid = cpu_to_le32(nsid);
914 c.identify.cns = NVME_ID_CNS_NS;
916 id = kmalloc(sizeof(*id), GFP_KERNEL);
920 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
922 dev_warn(ctrl->device, "Identify namespace failed\n");
930 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
931 void *buffer, size_t buflen, u32 *result)
933 struct nvme_command c;
934 union nvme_result res;
937 memset(&c, 0, sizeof(c));
938 c.features.opcode = nvme_admin_set_features;
939 c.features.fid = cpu_to_le32(fid);
940 c.features.dword11 = cpu_to_le32(dword11);
942 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
943 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
944 if (ret >= 0 && result)
945 *result = le32_to_cpu(res.u32);
949 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
951 u32 q_count = (*count - 1) | ((*count - 1) << 16);
953 int status, nr_io_queues;
955 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
961 * Degraded controllers might return an error when setting the queue
962 * count. We still want to be able to bring them online and offer
963 * access to the admin queue, as that might be only way to fix them up.
966 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
969 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
970 *count = min(*count, nr_io_queues);
975 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
977 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
979 struct nvme_user_io io;
980 struct nvme_command c;
981 unsigned length, meta_len;
982 void __user *metadata;
984 if (copy_from_user(&io, uio, sizeof(io)))
992 case nvme_cmd_compare:
998 length = (io.nblocks + 1) << ns->lba_shift;
999 meta_len = (io.nblocks + 1) * ns->ms;
1000 metadata = (void __user *)(uintptr_t)io.metadata;
1005 } else if (meta_len) {
1006 if ((io.metadata & 3) || !io.metadata)
1010 memset(&c, 0, sizeof(c));
1011 c.rw.opcode = io.opcode;
1012 c.rw.flags = io.flags;
1013 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1014 c.rw.slba = cpu_to_le64(io.slba);
1015 c.rw.length = cpu_to_le16(io.nblocks);
1016 c.rw.control = cpu_to_le16(io.control);
1017 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1018 c.rw.reftag = cpu_to_le32(io.reftag);
1019 c.rw.apptag = cpu_to_le16(io.apptag);
1020 c.rw.appmask = cpu_to_le16(io.appmask);
1022 return nvme_submit_user_cmd(ns->queue, &c,
1023 (void __user *)(uintptr_t)io.addr, length,
1024 metadata, meta_len, io.slba, NULL, 0);
1027 static u32 nvme_known_admin_effects(u8 opcode)
1030 case nvme_admin_format_nvm:
1031 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1032 NVME_CMD_EFFECTS_CSE_MASK;
1033 case nvme_admin_sanitize_nvm:
1034 return NVME_CMD_EFFECTS_CSE_MASK;
1041 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1048 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1049 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1050 dev_warn(ctrl->device,
1051 "IO command:%02x has unhandled effects:%08x\n",
1057 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1059 effects = nvme_known_admin_effects(opcode);
1062 * For simplicity, IO to all namespaces is quiesced even if the command
1063 * effects say only one namespace is affected.
1065 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1066 nvme_start_freeze(ctrl);
1067 nvme_wait_freeze(ctrl);
1072 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1076 mutex_lock(&ctrl->namespaces_mutex);
1077 list_for_each_entry(ns, &ctrl->namespaces, list) {
1078 if (ns->disk && nvme_revalidate_disk(ns->disk))
1081 mutex_unlock(&ctrl->namespaces_mutex);
1084 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1087 * Revalidate LBA changes prior to unfreezing. This is necessary to
1088 * prevent memory corruption if a logical block size was changed by
1091 if (effects & NVME_CMD_EFFECTS_LBCC)
1092 nvme_update_formats(ctrl);
1093 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1094 nvme_unfreeze(ctrl);
1095 if (effects & NVME_CMD_EFFECTS_CCC)
1096 nvme_init_identify(ctrl);
1097 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1098 nvme_queue_scan(ctrl);
1101 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1102 struct nvme_passthru_cmd __user *ucmd)
1104 struct nvme_passthru_cmd cmd;
1105 struct nvme_command c;
1106 unsigned timeout = 0;
1110 if (!capable(CAP_SYS_ADMIN))
1112 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1117 memset(&c, 0, sizeof(c));
1118 c.common.opcode = cmd.opcode;
1119 c.common.flags = cmd.flags;
1120 c.common.nsid = cpu_to_le32(cmd.nsid);
1121 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1122 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1123 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1124 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1125 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1126 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1127 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1128 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1131 timeout = msecs_to_jiffies(cmd.timeout_ms);
1133 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1134 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1135 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1136 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1137 0, &cmd.result, timeout);
1138 nvme_passthru_end(ctrl, effects);
1141 if (put_user(cmd.result, &ucmd->result))
1149 * Issue ioctl requests on the first available path. Note that unlike normal
1150 * block layer requests we will not retry failed request on another controller.
1152 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1153 struct nvme_ns_head **head, int *srcu_idx)
1155 #ifdef CONFIG_NVME_MULTIPATH
1156 if (disk->fops == &nvme_ns_head_ops) {
1157 *head = disk->private_data;
1158 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1159 return nvme_find_path(*head);
1164 return disk->private_data;
1167 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1170 srcu_read_unlock(&head->srcu, idx);
1173 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1177 force_successful_syscall_return();
1178 return ns->head->ns_id;
1179 case NVME_IOCTL_ADMIN_CMD:
1180 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1181 case NVME_IOCTL_IO_CMD:
1182 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1183 case NVME_IOCTL_SUBMIT_IO:
1184 return nvme_submit_io(ns, (void __user *)arg);
1188 return nvme_nvm_ioctl(ns, cmd, arg);
1190 if (is_sed_ioctl(cmd))
1191 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1192 (void __user *) arg);
1197 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1198 unsigned int cmd, unsigned long arg)
1200 struct nvme_ns_head *head = NULL;
1204 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1208 ret = nvme_ns_ioctl(ns, cmd, arg);
1209 nvme_put_ns_from_disk(head, srcu_idx);
1213 static int nvme_open(struct block_device *bdev, fmode_t mode)
1215 struct nvme_ns *ns = bdev->bd_disk->private_data;
1217 #ifdef CONFIG_NVME_MULTIPATH
1218 /* should never be called due to GENHD_FL_HIDDEN */
1219 if (WARN_ON_ONCE(ns->head->disk))
1222 if (!kref_get_unless_zero(&ns->kref))
1227 static void nvme_release(struct gendisk *disk, fmode_t mode)
1229 nvme_put_ns(disk->private_data);
1232 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1234 /* some standard values */
1235 geo->heads = 1 << 6;
1236 geo->sectors = 1 << 5;
1237 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1241 #ifdef CONFIG_BLK_DEV_INTEGRITY
1242 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1244 struct blk_integrity integrity;
1246 memset(&integrity, 0, sizeof(integrity));
1248 case NVME_NS_DPS_PI_TYPE3:
1249 integrity.profile = &t10_pi_type3_crc;
1250 integrity.tag_size = sizeof(u16) + sizeof(u32);
1251 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1253 case NVME_NS_DPS_PI_TYPE1:
1254 case NVME_NS_DPS_PI_TYPE2:
1255 integrity.profile = &t10_pi_type1_crc;
1256 integrity.tag_size = sizeof(u16);
1257 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1260 integrity.profile = NULL;
1263 integrity.tuple_size = ms;
1264 blk_integrity_register(disk, &integrity);
1265 blk_queue_max_integrity_segments(disk->queue, 1);
1268 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1271 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1273 static void nvme_set_chunk_size(struct nvme_ns *ns)
1275 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1276 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1279 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1280 unsigned stream_alignment, struct request_queue *queue)
1282 u32 size = queue_logical_block_size(queue);
1284 if (stream_alignment)
1285 size *= stream_alignment;
1287 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1288 NVME_DSM_MAX_RANGES);
1290 queue->limits.discard_alignment = 0;
1291 queue->limits.discard_granularity = size;
1293 blk_queue_max_discard_sectors(queue, UINT_MAX);
1294 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1295 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1297 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1298 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1301 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1302 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1304 memset(ids, 0, sizeof(*ids));
1306 if (ctrl->vs >= NVME_VS(1, 1, 0))
1307 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1308 if (ctrl->vs >= NVME_VS(1, 2, 0))
1309 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1310 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1311 /* Don't treat error as fatal we potentially
1312 * already have a NGUID or EUI-64
1314 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1315 dev_warn(ctrl->device,
1316 "%s: Identify Descriptors failed\n", __func__);
1320 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1322 return !uuid_is_null(&ids->uuid) ||
1323 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1324 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1327 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1329 return uuid_equal(&a->uuid, &b->uuid) &&
1330 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1331 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1334 static void nvme_update_disk_info(struct gendisk *disk,
1335 struct nvme_ns *ns, struct nvme_id_ns *id)
1337 sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1338 unsigned short bs = 1 << ns->lba_shift;
1339 unsigned stream_alignment = 0;
1341 if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1342 stream_alignment = ns->sws * ns->sgs;
1344 blk_mq_freeze_queue(disk->queue);
1345 blk_integrity_unregister(disk);
1347 blk_queue_logical_block_size(disk->queue, bs);
1348 blk_queue_physical_block_size(disk->queue, bs);
1349 blk_queue_io_min(disk->queue, bs);
1351 if (ns->ms && !ns->ext &&
1352 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1353 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1354 if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1356 set_capacity(disk, capacity);
1358 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1359 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1360 blk_mq_unfreeze_queue(disk->queue);
1363 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1365 struct nvme_ns *ns = disk->private_data;
1368 * If identify namespace failed, use default 512 byte block size so
1369 * block layer can use before failing read/write for 0 capacity.
1371 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1372 if (ns->lba_shift == 0)
1374 ns->noiob = le16_to_cpu(id->noiob);
1375 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1376 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1377 /* the PI implementation requires metadata equal t10 pi tuple size */
1378 if (ns->ms == sizeof(struct t10_pi_tuple))
1379 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1384 nvme_set_chunk_size(ns);
1385 nvme_update_disk_info(disk, ns, id);
1386 #ifdef CONFIG_NVME_MULTIPATH
1388 nvme_update_disk_info(ns->head->disk, ns, id);
1392 static int nvme_revalidate_disk(struct gendisk *disk)
1394 struct nvme_ns *ns = disk->private_data;
1395 struct nvme_ctrl *ctrl = ns->ctrl;
1396 struct nvme_id_ns *id;
1397 struct nvme_ns_ids ids;
1400 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1401 set_capacity(disk, 0);
1405 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1409 if (id->ncap == 0) {
1414 __nvme_revalidate_disk(disk, id);
1415 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1416 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1417 dev_err(ctrl->device,
1418 "identifiers changed for nsid %d\n", ns->head->ns_id);
1427 static char nvme_pr_type(enum pr_type type)
1430 case PR_WRITE_EXCLUSIVE:
1432 case PR_EXCLUSIVE_ACCESS:
1434 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1436 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1438 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1440 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1447 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1448 u64 key, u64 sa_key, u8 op)
1450 struct nvme_ns_head *head = NULL;
1452 struct nvme_command c;
1454 u8 data[16] = { 0, };
1456 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1458 return -EWOULDBLOCK;
1460 put_unaligned_le64(key, &data[0]);
1461 put_unaligned_le64(sa_key, &data[8]);
1463 memset(&c, 0, sizeof(c));
1464 c.common.opcode = op;
1465 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1466 c.common.cdw10[0] = cpu_to_le32(cdw10);
1468 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1469 nvme_put_ns_from_disk(head, srcu_idx);
1473 static int nvme_pr_register(struct block_device *bdev, u64 old,
1474 u64 new, unsigned flags)
1478 if (flags & ~PR_FL_IGNORE_KEY)
1481 cdw10 = old ? 2 : 0;
1482 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1483 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1484 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1487 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1488 enum pr_type type, unsigned flags)
1492 if (flags & ~PR_FL_IGNORE_KEY)
1495 cdw10 = nvme_pr_type(type) << 8;
1496 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1497 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1500 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1501 enum pr_type type, bool abort)
1503 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1504 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1507 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1509 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1510 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1513 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1515 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1516 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1519 static const struct pr_ops nvme_pr_ops = {
1520 .pr_register = nvme_pr_register,
1521 .pr_reserve = nvme_pr_reserve,
1522 .pr_release = nvme_pr_release,
1523 .pr_preempt = nvme_pr_preempt,
1524 .pr_clear = nvme_pr_clear,
1527 #ifdef CONFIG_BLK_SED_OPAL
1528 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1531 struct nvme_ctrl *ctrl = data;
1532 struct nvme_command cmd;
1534 memset(&cmd, 0, sizeof(cmd));
1536 cmd.common.opcode = nvme_admin_security_send;
1538 cmd.common.opcode = nvme_admin_security_recv;
1539 cmd.common.nsid = 0;
1540 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1541 cmd.common.cdw10[1] = cpu_to_le32(len);
1543 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1544 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1546 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1547 #endif /* CONFIG_BLK_SED_OPAL */
1549 static const struct block_device_operations nvme_fops = {
1550 .owner = THIS_MODULE,
1551 .ioctl = nvme_ioctl,
1552 .compat_ioctl = nvme_ioctl,
1554 .release = nvme_release,
1555 .getgeo = nvme_getgeo,
1556 .revalidate_disk= nvme_revalidate_disk,
1557 .pr_ops = &nvme_pr_ops,
1560 #ifdef CONFIG_NVME_MULTIPATH
1561 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1563 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1565 if (!kref_get_unless_zero(&head->ref))
1570 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1572 nvme_put_ns_head(disk->private_data);
1575 const struct block_device_operations nvme_ns_head_ops = {
1576 .owner = THIS_MODULE,
1577 .open = nvme_ns_head_open,
1578 .release = nvme_ns_head_release,
1579 .ioctl = nvme_ioctl,
1580 .compat_ioctl = nvme_ioctl,
1581 .getgeo = nvme_getgeo,
1582 .pr_ops = &nvme_pr_ops,
1584 #endif /* CONFIG_NVME_MULTIPATH */
1586 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1588 unsigned long timeout =
1589 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1590 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1593 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1596 if ((csts & NVME_CSTS_RDY) == bit)
1600 if (fatal_signal_pending(current))
1602 if (time_after(jiffies, timeout)) {
1603 dev_err(ctrl->device,
1604 "Device not ready; aborting %s\n", enabled ?
1605 "initialisation" : "reset");
1614 * If the device has been passed off to us in an enabled state, just clear
1615 * the enabled bit. The spec says we should set the 'shutdown notification
1616 * bits', but doing so may cause the device to complete commands to the
1617 * admin queue ... and we don't know what memory that might be pointing at!
1619 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1623 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1624 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1626 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1630 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1631 msleep(NVME_QUIRK_DELAY_AMOUNT);
1633 return nvme_wait_ready(ctrl, cap, false);
1635 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1637 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1640 * Default to a 4K page size, with the intention to update this
1641 * path in the future to accomodate architectures with differing
1642 * kernel and IO page sizes.
1644 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1647 if (page_shift < dev_page_min) {
1648 dev_err(ctrl->device,
1649 "Minimum device page size %u too large for host (%u)\n",
1650 1 << dev_page_min, 1 << page_shift);
1654 ctrl->page_size = 1 << page_shift;
1656 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1657 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1658 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1659 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1660 ctrl->ctrl_config |= NVME_CC_ENABLE;
1662 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1665 return nvme_wait_ready(ctrl, cap, true);
1667 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1669 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1671 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1675 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1676 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1678 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1682 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1683 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1687 if (fatal_signal_pending(current))
1689 if (time_after(jiffies, timeout)) {
1690 dev_err(ctrl->device,
1691 "Device shutdown incomplete; abort shutdown\n");
1698 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1700 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1701 struct request_queue *q)
1705 if (ctrl->max_hw_sectors) {
1707 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1709 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1710 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1712 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1713 is_power_of_2(ctrl->max_hw_sectors))
1714 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1715 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1716 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1718 blk_queue_write_cache(q, vwc, vwc);
1721 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1726 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1729 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1730 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1733 dev_warn_once(ctrl->device,
1734 "could not set timestamp (%d)\n", ret);
1738 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1741 * APST (Autonomous Power State Transition) lets us program a
1742 * table of power state transitions that the controller will
1743 * perform automatically. We configure it with a simple
1744 * heuristic: we are willing to spend at most 2% of the time
1745 * transitioning between power states. Therefore, when running
1746 * in any given state, we will enter the next lower-power
1747 * non-operational state after waiting 50 * (enlat + exlat)
1748 * microseconds, as long as that state's exit latency is under
1749 * the requested maximum latency.
1751 * We will not autonomously enter any non-operational state for
1752 * which the total latency exceeds ps_max_latency_us. Users
1753 * can set ps_max_latency_us to zero to turn off APST.
1757 struct nvme_feat_auto_pst *table;
1763 * If APST isn't supported or if we haven't been initialized yet,
1764 * then don't do anything.
1769 if (ctrl->npss > 31) {
1770 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1774 table = kzalloc(sizeof(*table), GFP_KERNEL);
1778 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1779 /* Turn off APST. */
1781 dev_dbg(ctrl->device, "APST disabled\n");
1783 __le64 target = cpu_to_le64(0);
1787 * Walk through all states from lowest- to highest-power.
1788 * According to the spec, lower-numbered states use more
1789 * power. NPSS, despite the name, is the index of the
1790 * lowest-power state, not the number of states.
1792 for (state = (int)ctrl->npss; state >= 0; state--) {
1793 u64 total_latency_us, exit_latency_us, transition_ms;
1796 table->entries[state] = target;
1799 * Don't allow transitions to the deepest state
1800 * if it's quirked off.
1802 if (state == ctrl->npss &&
1803 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1807 * Is this state a useful non-operational state for
1808 * higher-power states to autonomously transition to?
1810 if (!(ctrl->psd[state].flags &
1811 NVME_PS_FLAGS_NON_OP_STATE))
1815 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1816 if (exit_latency_us > ctrl->ps_max_latency_us)
1821 le32_to_cpu(ctrl->psd[state].entry_lat);
1824 * This state is good. Use it as the APST idle
1825 * target for higher power states.
1827 transition_ms = total_latency_us + 19;
1828 do_div(transition_ms, 20);
1829 if (transition_ms > (1 << 24) - 1)
1830 transition_ms = (1 << 24) - 1;
1832 target = cpu_to_le64((state << 3) |
1833 (transition_ms << 8));
1838 if (total_latency_us > max_lat_us)
1839 max_lat_us = total_latency_us;
1845 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1847 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1848 max_ps, max_lat_us, (int)sizeof(*table), table);
1852 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1853 table, sizeof(*table), NULL);
1855 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1861 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1863 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1867 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1868 case PM_QOS_LATENCY_ANY:
1876 if (ctrl->ps_max_latency_us != latency) {
1877 ctrl->ps_max_latency_us = latency;
1878 nvme_configure_apst(ctrl);
1882 struct nvme_core_quirk_entry {
1884 * NVMe model and firmware strings are padded with spaces. For
1885 * simplicity, strings in the quirk table are padded with NULLs
1891 unsigned long quirks;
1894 static const struct nvme_core_quirk_entry core_quirks[] = {
1897 * This Toshiba device seems to die using any APST states. See:
1898 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1901 .mn = "THNSF5256GPUK TOSHIBA",
1902 .quirks = NVME_QUIRK_NO_APST,
1906 /* match is null-terminated but idstr is space-padded. */
1907 static bool string_matches(const char *idstr, const char *match, size_t len)
1914 matchlen = strlen(match);
1915 WARN_ON_ONCE(matchlen > len);
1917 if (memcmp(idstr, match, matchlen))
1920 for (; matchlen < len; matchlen++)
1921 if (idstr[matchlen] != ' ')
1927 static bool quirk_matches(const struct nvme_id_ctrl *id,
1928 const struct nvme_core_quirk_entry *q)
1930 return q->vid == le16_to_cpu(id->vid) &&
1931 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1932 string_matches(id->fr, q->fr, sizeof(id->fr));
1935 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
1936 struct nvme_id_ctrl *id)
1941 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1942 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1943 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
1947 if (ctrl->vs >= NVME_VS(1, 2, 1))
1948 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1950 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1951 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
1952 "nqn.2014.08.org.nvmexpress:%4x%4x",
1953 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1954 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
1955 off += sizeof(id->sn);
1956 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
1957 off += sizeof(id->mn);
1958 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
1961 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
1963 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
1967 static void nvme_release_subsystem(struct device *dev)
1969 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
1972 static void nvme_destroy_subsystem(struct kref *ref)
1974 struct nvme_subsystem *subsys =
1975 container_of(ref, struct nvme_subsystem, ref);
1977 mutex_lock(&nvme_subsystems_lock);
1978 list_del(&subsys->entry);
1979 mutex_unlock(&nvme_subsystems_lock);
1981 ida_destroy(&subsys->ns_ida);
1982 device_del(&subsys->dev);
1983 put_device(&subsys->dev);
1986 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
1988 kref_put(&subsys->ref, nvme_destroy_subsystem);
1991 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
1993 struct nvme_subsystem *subsys;
1995 lockdep_assert_held(&nvme_subsystems_lock);
1997 list_for_each_entry(subsys, &nvme_subsystems, entry) {
1998 if (strcmp(subsys->subnqn, subsysnqn))
2000 if (!kref_get_unless_zero(&subsys->ref))
2008 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2009 struct device_attribute subsys_attr_##_name = \
2010 __ATTR(_name, _mode, _show, NULL)
2012 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2013 struct device_attribute *attr,
2016 struct nvme_subsystem *subsys =
2017 container_of(dev, struct nvme_subsystem, dev);
2019 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2021 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2023 #define nvme_subsys_show_str_function(field) \
2024 static ssize_t subsys_##field##_show(struct device *dev, \
2025 struct device_attribute *attr, char *buf) \
2027 struct nvme_subsystem *subsys = \
2028 container_of(dev, struct nvme_subsystem, dev); \
2029 return sprintf(buf, "%.*s\n", \
2030 (int)sizeof(subsys->field), subsys->field); \
2032 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2034 nvme_subsys_show_str_function(model);
2035 nvme_subsys_show_str_function(serial);
2036 nvme_subsys_show_str_function(firmware_rev);
2038 static struct attribute *nvme_subsys_attrs[] = {
2039 &subsys_attr_model.attr,
2040 &subsys_attr_serial.attr,
2041 &subsys_attr_firmware_rev.attr,
2042 &subsys_attr_subsysnqn.attr,
2046 static struct attribute_group nvme_subsys_attrs_group = {
2047 .attrs = nvme_subsys_attrs,
2050 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2051 &nvme_subsys_attrs_group,
2055 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2057 struct nvme_subsystem *subsys, *found;
2060 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2063 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2068 subsys->instance = ret;
2069 mutex_init(&subsys->lock);
2070 kref_init(&subsys->ref);
2071 INIT_LIST_HEAD(&subsys->ctrls);
2072 INIT_LIST_HEAD(&subsys->nsheads);
2073 nvme_init_subnqn(subsys, ctrl, id);
2074 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2075 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2076 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2077 subsys->vendor_id = le16_to_cpu(id->vid);
2078 subsys->cmic = id->cmic;
2080 subsys->dev.class = nvme_subsys_class;
2081 subsys->dev.release = nvme_release_subsystem;
2082 subsys->dev.groups = nvme_subsys_attrs_groups;
2083 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2084 device_initialize(&subsys->dev);
2086 mutex_lock(&nvme_subsystems_lock);
2087 found = __nvme_find_get_subsystem(subsys->subnqn);
2090 * Verify that the subsystem actually supports multiple
2091 * controllers, else bail out.
2093 if (!(id->cmic & (1 << 1))) {
2094 dev_err(ctrl->device,
2095 "ignoring ctrl due to duplicate subnqn (%s).\n",
2097 nvme_put_subsystem(found);
2102 __nvme_release_subsystem(subsys);
2105 ret = device_add(&subsys->dev);
2107 dev_err(ctrl->device,
2108 "failed to register subsystem device.\n");
2111 ida_init(&subsys->ns_ida);
2112 list_add_tail(&subsys->entry, &nvme_subsystems);
2115 ctrl->subsys = subsys;
2116 mutex_unlock(&nvme_subsystems_lock);
2118 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2119 dev_name(ctrl->device))) {
2120 dev_err(ctrl->device,
2121 "failed to create sysfs link from subsystem.\n");
2122 /* the transport driver will eventually put the subsystem */
2126 mutex_lock(&subsys->lock);
2127 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2128 mutex_unlock(&subsys->lock);
2133 mutex_unlock(&nvme_subsystems_lock);
2134 put_device(&subsys->dev);
2138 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2141 struct nvme_command c = { };
2143 c.common.opcode = nvme_admin_get_log_page;
2144 c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2145 c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2147 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2150 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2155 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2160 ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2161 sizeof(*ctrl->effects));
2163 kfree(ctrl->effects);
2164 ctrl->effects = NULL;
2170 * Initialize the cached copies of the Identify data and various controller
2171 * register in our nvme_ctrl structure. This should be called as soon as
2172 * the admin queue is fully up and running.
2174 int nvme_init_identify(struct nvme_ctrl *ctrl)
2176 struct nvme_id_ctrl *id;
2178 int ret, page_shift;
2180 bool prev_apst_enabled;
2182 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2184 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2188 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2190 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2193 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2195 if (ctrl->vs >= NVME_VS(1, 1, 0))
2196 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2198 ret = nvme_identify_ctrl(ctrl, &id);
2200 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2204 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2205 ret = nvme_get_effects_log(ctrl);
2210 if (!ctrl->identified) {
2213 ret = nvme_init_subsystem(ctrl, id);
2218 * Check for quirks. Quirk can depend on firmware version,
2219 * so, in principle, the set of quirks present can change
2220 * across a reset. As a possible future enhancement, we
2221 * could re-scan for quirks every time we reinitialize
2222 * the device, but we'd have to make sure that the driver
2223 * behaves intelligently if the quirks change.
2225 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2226 if (quirk_matches(id, &core_quirks[i]))
2227 ctrl->quirks |= core_quirks[i].quirks;
2231 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2232 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2233 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2236 ctrl->oacs = le16_to_cpu(id->oacs);
2237 ctrl->oncs = le16_to_cpup(&id->oncs);
2238 atomic_set(&ctrl->abort_limit, id->acl + 1);
2239 ctrl->vwc = id->vwc;
2240 ctrl->cntlid = le16_to_cpup(&id->cntlid);
2242 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2244 max_hw_sectors = UINT_MAX;
2245 ctrl->max_hw_sectors =
2246 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2248 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2249 ctrl->sgls = le32_to_cpu(id->sgls);
2250 ctrl->kas = le16_to_cpu(id->kas);
2254 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2256 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2257 shutdown_timeout, 60);
2259 if (ctrl->shutdown_timeout != shutdown_timeout)
2260 dev_warn(ctrl->device,
2261 "Shutdown timeout set to %u seconds\n",
2262 ctrl->shutdown_timeout);
2264 ctrl->shutdown_timeout = shutdown_timeout;
2266 ctrl->npss = id->npss;
2267 ctrl->apsta = id->apsta;
2268 prev_apst_enabled = ctrl->apst_enabled;
2269 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2270 if (force_apst && id->apsta) {
2271 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2272 ctrl->apst_enabled = true;
2274 ctrl->apst_enabled = false;
2277 ctrl->apst_enabled = id->apsta;
2279 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2281 if (ctrl->ops->flags & NVME_F_FABRICS) {
2282 ctrl->icdoff = le16_to_cpu(id->icdoff);
2283 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2284 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2285 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2288 * In fabrics we need to verify the cntlid matches the
2291 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2296 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2297 dev_err(ctrl->device,
2298 "keep-alive support is mandatory for fabrics\n");
2303 ctrl->cntlid = le16_to_cpu(id->cntlid);
2304 ctrl->hmpre = le32_to_cpu(id->hmpre);
2305 ctrl->hmmin = le32_to_cpu(id->hmmin);
2306 ctrl->hmminds = le32_to_cpu(id->hmminds);
2307 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2312 if (ctrl->apst_enabled && !prev_apst_enabled)
2313 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2314 else if (!ctrl->apst_enabled && prev_apst_enabled)
2315 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2317 ret = nvme_configure_apst(ctrl);
2321 ret = nvme_configure_timestamp(ctrl);
2325 ret = nvme_configure_directives(ctrl);
2329 ctrl->identified = true;
2337 EXPORT_SYMBOL_GPL(nvme_init_identify);
2339 static int nvme_dev_open(struct inode *inode, struct file *file)
2341 struct nvme_ctrl *ctrl =
2342 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2344 if (ctrl->state != NVME_CTRL_LIVE)
2345 return -EWOULDBLOCK;
2346 file->private_data = ctrl;
2350 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2355 mutex_lock(&ctrl->namespaces_mutex);
2356 if (list_empty(&ctrl->namespaces)) {
2361 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2362 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2363 dev_warn(ctrl->device,
2364 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2369 dev_warn(ctrl->device,
2370 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2371 kref_get(&ns->kref);
2372 mutex_unlock(&ctrl->namespaces_mutex);
2374 ret = nvme_user_cmd(ctrl, ns, argp);
2379 mutex_unlock(&ctrl->namespaces_mutex);
2383 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2386 struct nvme_ctrl *ctrl = file->private_data;
2387 void __user *argp = (void __user *)arg;
2390 case NVME_IOCTL_ADMIN_CMD:
2391 return nvme_user_cmd(ctrl, NULL, argp);
2392 case NVME_IOCTL_IO_CMD:
2393 return nvme_dev_user_cmd(ctrl, argp);
2394 case NVME_IOCTL_RESET:
2395 dev_warn(ctrl->device, "resetting controller\n");
2396 return nvme_reset_ctrl_sync(ctrl);
2397 case NVME_IOCTL_SUBSYS_RESET:
2398 return nvme_reset_subsystem(ctrl);
2399 case NVME_IOCTL_RESCAN:
2400 nvme_queue_scan(ctrl);
2407 static const struct file_operations nvme_dev_fops = {
2408 .owner = THIS_MODULE,
2409 .open = nvme_dev_open,
2410 .unlocked_ioctl = nvme_dev_ioctl,
2411 .compat_ioctl = nvme_dev_ioctl,
2414 static ssize_t nvme_sysfs_reset(struct device *dev,
2415 struct device_attribute *attr, const char *buf,
2418 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2421 ret = nvme_reset_ctrl_sync(ctrl);
2426 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2428 static ssize_t nvme_sysfs_rescan(struct device *dev,
2429 struct device_attribute *attr, const char *buf,
2432 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2434 nvme_queue_scan(ctrl);
2437 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2439 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2441 struct gendisk *disk = dev_to_disk(dev);
2443 if (disk->fops == &nvme_fops)
2444 return nvme_get_ns_from_dev(dev)->head;
2446 return disk->private_data;
2449 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2452 struct nvme_ns_head *head = dev_to_ns_head(dev);
2453 struct nvme_ns_ids *ids = &head->ids;
2454 struct nvme_subsystem *subsys = head->subsys;
2455 int serial_len = sizeof(subsys->serial);
2456 int model_len = sizeof(subsys->model);
2458 if (!uuid_is_null(&ids->uuid))
2459 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2461 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2462 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2464 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2465 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2467 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2468 subsys->serial[serial_len - 1] == '\0'))
2470 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2471 subsys->model[model_len - 1] == '\0'))
2474 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2475 serial_len, subsys->serial, model_len, subsys->model,
2478 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2480 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2483 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2485 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2487 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2490 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2492 /* For backward compatibility expose the NGUID to userspace if
2493 * we have no UUID set
2495 if (uuid_is_null(&ids->uuid)) {
2496 printk_ratelimited(KERN_WARNING
2497 "No UUID available providing old NGUID\n");
2498 return sprintf(buf, "%pU\n", ids->nguid);
2500 return sprintf(buf, "%pU\n", &ids->uuid);
2502 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2504 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2507 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2509 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2511 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2514 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2516 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2518 static struct attribute *nvme_ns_id_attrs[] = {
2519 &dev_attr_wwid.attr,
2520 &dev_attr_uuid.attr,
2521 &dev_attr_nguid.attr,
2523 &dev_attr_nsid.attr,
2527 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2528 struct attribute *a, int n)
2530 struct device *dev = container_of(kobj, struct device, kobj);
2531 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2533 if (a == &dev_attr_uuid.attr) {
2534 if (uuid_is_null(&ids->uuid) &&
2535 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2538 if (a == &dev_attr_nguid.attr) {
2539 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2542 if (a == &dev_attr_eui.attr) {
2543 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2549 const struct attribute_group nvme_ns_id_attr_group = {
2550 .attrs = nvme_ns_id_attrs,
2551 .is_visible = nvme_ns_id_attrs_are_visible,
2554 #define nvme_show_str_function(field) \
2555 static ssize_t field##_show(struct device *dev, \
2556 struct device_attribute *attr, char *buf) \
2558 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2559 return sprintf(buf, "%.*s\n", \
2560 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2562 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2564 nvme_show_str_function(model);
2565 nvme_show_str_function(serial);
2566 nvme_show_str_function(firmware_rev);
2568 #define nvme_show_int_function(field) \
2569 static ssize_t field##_show(struct device *dev, \
2570 struct device_attribute *attr, char *buf) \
2572 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2573 return sprintf(buf, "%d\n", ctrl->field); \
2575 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2577 nvme_show_int_function(cntlid);
2579 static ssize_t nvme_sysfs_delete(struct device *dev,
2580 struct device_attribute *attr, const char *buf,
2583 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2585 if (device_remove_file_self(dev, attr))
2586 nvme_delete_ctrl_sync(ctrl);
2589 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2591 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2592 struct device_attribute *attr,
2595 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2597 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2599 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2601 static ssize_t nvme_sysfs_show_state(struct device *dev,
2602 struct device_attribute *attr,
2605 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2606 static const char *const state_name[] = {
2607 [NVME_CTRL_NEW] = "new",
2608 [NVME_CTRL_LIVE] = "live",
2609 [NVME_CTRL_RESETTING] = "resetting",
2610 [NVME_CTRL_RECONNECTING]= "reconnecting",
2611 [NVME_CTRL_DELETING] = "deleting",
2612 [NVME_CTRL_DEAD] = "dead",
2615 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2616 state_name[ctrl->state])
2617 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2619 return sprintf(buf, "unknown state\n");
2622 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2624 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2625 struct device_attribute *attr,
2628 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2630 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2632 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2634 static ssize_t nvme_sysfs_show_address(struct device *dev,
2635 struct device_attribute *attr,
2638 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2640 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2642 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2644 static struct attribute *nvme_dev_attrs[] = {
2645 &dev_attr_reset_controller.attr,
2646 &dev_attr_rescan_controller.attr,
2647 &dev_attr_model.attr,
2648 &dev_attr_serial.attr,
2649 &dev_attr_firmware_rev.attr,
2650 &dev_attr_cntlid.attr,
2651 &dev_attr_delete_controller.attr,
2652 &dev_attr_transport.attr,
2653 &dev_attr_subsysnqn.attr,
2654 &dev_attr_address.attr,
2655 &dev_attr_state.attr,
2659 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2660 struct attribute *a, int n)
2662 struct device *dev = container_of(kobj, struct device, kobj);
2663 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2665 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2667 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2673 static struct attribute_group nvme_dev_attrs_group = {
2674 .attrs = nvme_dev_attrs,
2675 .is_visible = nvme_dev_attrs_are_visible,
2678 static const struct attribute_group *nvme_dev_attr_groups[] = {
2679 &nvme_dev_attrs_group,
2683 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2686 struct nvme_ns_head *h;
2688 lockdep_assert_held(&subsys->lock);
2690 list_for_each_entry(h, &subsys->nsheads, entry) {
2691 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2698 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2699 struct nvme_ns_head *new)
2701 struct nvme_ns_head *h;
2703 lockdep_assert_held(&subsys->lock);
2705 list_for_each_entry(h, &subsys->nsheads, entry) {
2706 if (nvme_ns_ids_valid(&new->ids) &&
2707 nvme_ns_ids_equal(&new->ids, &h->ids))
2714 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2715 unsigned nsid, struct nvme_id_ns *id)
2717 struct nvme_ns_head *head;
2720 head = kzalloc(sizeof(*head), GFP_KERNEL);
2723 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2726 head->instance = ret;
2727 INIT_LIST_HEAD(&head->list);
2728 init_srcu_struct(&head->srcu);
2729 head->subsys = ctrl->subsys;
2731 kref_init(&head->ref);
2733 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2735 ret = __nvme_check_ids(ctrl->subsys, head);
2737 dev_err(ctrl->device,
2738 "duplicate IDs for nsid %d\n", nsid);
2739 goto out_cleanup_srcu;
2742 ret = nvme_mpath_alloc_disk(ctrl, head);
2744 goto out_cleanup_srcu;
2746 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2749 cleanup_srcu_struct(&head->srcu);
2750 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2754 return ERR_PTR(ret);
2757 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2758 struct nvme_id_ns *id, bool *new)
2760 struct nvme_ctrl *ctrl = ns->ctrl;
2761 bool is_shared = id->nmic & (1 << 0);
2762 struct nvme_ns_head *head = NULL;
2765 mutex_lock(&ctrl->subsys->lock);
2767 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2769 head = nvme_alloc_ns_head(ctrl, nsid, id);
2771 ret = PTR_ERR(head);
2777 struct nvme_ns_ids ids;
2779 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2780 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2781 dev_err(ctrl->device,
2782 "IDs don't match for shared namespace %d\n",
2791 list_add_tail(&ns->siblings, &head->list);
2795 mutex_unlock(&ctrl->subsys->lock);
2799 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2801 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2802 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2804 return nsa->head->ns_id - nsb->head->ns_id;
2807 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2809 struct nvme_ns *ns, *ret = NULL;
2811 mutex_lock(&ctrl->namespaces_mutex);
2812 list_for_each_entry(ns, &ctrl->namespaces, list) {
2813 if (ns->head->ns_id == nsid) {
2814 if (!kref_get_unless_zero(&ns->kref))
2819 if (ns->head->ns_id > nsid)
2822 mutex_unlock(&ctrl->namespaces_mutex);
2826 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2828 struct streams_directive_params s;
2831 if (!ctrl->nr_streams)
2834 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2838 ns->sws = le32_to_cpu(s.sws);
2839 ns->sgs = le16_to_cpu(s.sgs);
2842 unsigned int bs = 1 << ns->lba_shift;
2844 blk_queue_io_min(ns->queue, bs * ns->sws);
2846 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2852 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2855 struct gendisk *disk;
2856 struct nvme_id_ns *id;
2857 char disk_name[DISK_NAME_LEN];
2858 int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2861 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2865 ns->queue = blk_mq_init_queue(ctrl->tagset);
2866 if (IS_ERR(ns->queue))
2868 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2869 ns->queue->queuedata = ns;
2872 kref_init(&ns->kref);
2873 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2875 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2876 nvme_set_queue_limits(ctrl, ns->queue);
2878 id = nvme_identify_ns(ctrl, nsid);
2880 goto out_free_queue;
2885 if (nvme_init_ns_head(ns, nsid, id, &new))
2887 nvme_setup_streams_ns(ctrl, ns);
2889 #ifdef CONFIG_NVME_MULTIPATH
2891 * If multipathing is enabled we need to always use the subsystem
2892 * instance number for numbering our devices to avoid conflicts
2893 * between subsystems that have multiple controllers and thus use
2894 * the multipath-aware subsystem node and those that have a single
2895 * controller and use the controller node directly.
2897 if (ns->head->disk) {
2898 sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2899 ctrl->cntlid, ns->head->instance);
2900 flags = GENHD_FL_HIDDEN;
2902 sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2903 ns->head->instance);
2907 * But without the multipath code enabled, multiple controller per
2908 * subsystems are visible as devices and thus we cannot use the
2909 * subsystem instance.
2911 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2914 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
2915 if (nvme_nvm_register(ns, disk_name, node)) {
2916 dev_warn(ctrl->device, "LightNVM init failure\n");
2921 disk = alloc_disk_node(0, node);
2925 disk->fops = &nvme_fops;
2926 disk->private_data = ns;
2927 disk->queue = ns->queue;
2928 disk->flags = flags;
2929 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2932 __nvme_revalidate_disk(disk, id);
2934 mutex_lock(&ctrl->namespaces_mutex);
2935 list_add_tail(&ns->list, &ctrl->namespaces);
2936 mutex_unlock(&ctrl->namespaces_mutex);
2938 nvme_get_ctrl(ctrl);
2942 device_add_disk(ctrl->device, ns->disk);
2943 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2944 &nvme_ns_id_attr_group))
2945 pr_warn("%s: failed to create sysfs group for identification\n",
2946 ns->disk->disk_name);
2947 if (ns->ndev && nvme_nvm_register_sysfs(ns))
2948 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2949 ns->disk->disk_name);
2952 nvme_mpath_add_disk(ns->head);
2953 nvme_mpath_add_disk_links(ns);
2956 mutex_lock(&ctrl->subsys->lock);
2957 list_del_rcu(&ns->siblings);
2958 mutex_unlock(&ctrl->subsys->lock);
2962 blk_cleanup_queue(ns->queue);
2967 static void nvme_ns_remove(struct nvme_ns *ns)
2969 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2972 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2973 nvme_mpath_remove_disk_links(ns);
2974 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2975 &nvme_ns_id_attr_group);
2977 nvme_nvm_unregister_sysfs(ns);
2978 del_gendisk(ns->disk);
2979 blk_cleanup_queue(ns->queue);
2980 if (blk_get_integrity(ns->disk))
2981 blk_integrity_unregister(ns->disk);
2984 mutex_lock(&ns->ctrl->subsys->lock);
2985 nvme_mpath_clear_current_path(ns);
2986 list_del_rcu(&ns->siblings);
2987 mutex_unlock(&ns->ctrl->subsys->lock);
2989 mutex_lock(&ns->ctrl->namespaces_mutex);
2990 list_del_init(&ns->list);
2991 mutex_unlock(&ns->ctrl->namespaces_mutex);
2993 synchronize_srcu(&ns->head->srcu);
2994 nvme_mpath_check_last_path(ns);
2998 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3002 ns = nvme_find_get_ns(ctrl, nsid);
3004 if (ns->disk && revalidate_disk(ns->disk))
3008 nvme_alloc_ns(ctrl, nsid);
3011 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3014 struct nvme_ns *ns, *next;
3016 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3017 if (ns->head->ns_id > nsid)
3022 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3026 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3029 ns_list = kzalloc(0x1000, GFP_KERNEL);
3033 for (i = 0; i < num_lists; i++) {
3034 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3038 for (j = 0; j < min(nn, 1024U); j++) {
3039 nsid = le32_to_cpu(ns_list[j]);
3043 nvme_validate_ns(ctrl, nsid);
3045 while (++prev < nsid) {
3046 ns = nvme_find_get_ns(ctrl, prev);
3056 nvme_remove_invalid_namespaces(ctrl, prev);
3062 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3066 for (i = 1; i <= nn; i++)
3067 nvme_validate_ns(ctrl, i);
3069 nvme_remove_invalid_namespaces(ctrl, nn);
3072 static void nvme_scan_work(struct work_struct *work)
3074 struct nvme_ctrl *ctrl =
3075 container_of(work, struct nvme_ctrl, scan_work);
3076 struct nvme_id_ctrl *id;
3079 if (ctrl->state != NVME_CTRL_LIVE)
3082 if (nvme_identify_ctrl(ctrl, &id))
3085 nn = le32_to_cpu(id->nn);
3086 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3087 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3088 if (!nvme_scan_ns_list(ctrl, nn))
3091 nvme_scan_ns_sequential(ctrl, nn);
3093 mutex_lock(&ctrl->namespaces_mutex);
3094 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3095 mutex_unlock(&ctrl->namespaces_mutex);
3099 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3102 * Do not queue new scan work when a controller is reset during
3105 if (ctrl->state == NVME_CTRL_LIVE)
3106 queue_work(nvme_wq, &ctrl->scan_work);
3108 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3111 * This function iterates the namespace list unlocked to allow recovery from
3112 * controller failure. It is up to the caller to ensure the namespace list is
3113 * not modified by scan work while this function is executing.
3115 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3117 struct nvme_ns *ns, *next;
3120 * The dead states indicates the controller was not gracefully
3121 * disconnected. In that case, we won't be able to flush any data while
3122 * removing the namespaces' disks; fail all the queues now to avoid
3123 * potentially having to clean up the failed sync later.
3125 if (ctrl->state == NVME_CTRL_DEAD)
3126 nvme_kill_queues(ctrl);
3128 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3131 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3133 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3135 char *envp[2] = { NULL, NULL };
3136 u32 aen_result = ctrl->aen_result;
3138 ctrl->aen_result = 0;
3142 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3145 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3149 static void nvme_async_event_work(struct work_struct *work)
3151 struct nvme_ctrl *ctrl =
3152 container_of(work, struct nvme_ctrl, async_event_work);
3154 nvme_aen_uevent(ctrl);
3155 ctrl->ops->submit_async_event(ctrl);
3158 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3163 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3169 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3172 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3174 struct nvme_fw_slot_info_log *log;
3176 log = kmalloc(sizeof(*log), GFP_KERNEL);
3180 if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3181 dev_warn(ctrl->device,
3182 "Get FW SLOT INFO log error\n");
3186 static void nvme_fw_act_work(struct work_struct *work)
3188 struct nvme_ctrl *ctrl = container_of(work,
3189 struct nvme_ctrl, fw_act_work);
3190 unsigned long fw_act_timeout;
3193 fw_act_timeout = jiffies +
3194 msecs_to_jiffies(ctrl->mtfa * 100);
3196 fw_act_timeout = jiffies +
3197 msecs_to_jiffies(admin_timeout * 1000);
3199 nvme_stop_queues(ctrl);
3200 while (nvme_ctrl_pp_status(ctrl)) {
3201 if (time_after(jiffies, fw_act_timeout)) {
3202 dev_warn(ctrl->device,
3203 "Fw activation timeout, reset controller\n");
3204 nvme_reset_ctrl(ctrl);
3210 if (ctrl->state != NVME_CTRL_LIVE)
3213 nvme_start_queues(ctrl);
3214 /* read FW slot information to clear the AER */
3215 nvme_get_fw_slot_info(ctrl);
3218 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3219 union nvme_result *res)
3221 u32 result = le32_to_cpu(res->u32);
3223 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3226 switch (result & 0x7) {
3227 case NVME_AER_ERROR:
3228 case NVME_AER_SMART:
3231 ctrl->aen_result = result;
3237 switch (result & 0xff07) {
3238 case NVME_AER_NOTICE_NS_CHANGED:
3239 dev_info(ctrl->device, "rescanning\n");
3240 nvme_queue_scan(ctrl);
3242 case NVME_AER_NOTICE_FW_ACT_STARTING:
3243 queue_work(nvme_wq, &ctrl->fw_act_work);
3246 dev_warn(ctrl->device, "async event result %08x\n", result);
3248 queue_work(nvme_wq, &ctrl->async_event_work);
3250 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3252 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3254 nvme_stop_keep_alive(ctrl);
3255 flush_work(&ctrl->async_event_work);
3256 flush_work(&ctrl->scan_work);
3257 cancel_work_sync(&ctrl->fw_act_work);
3259 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3261 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3264 nvme_start_keep_alive(ctrl);
3266 if (ctrl->queue_count > 1) {
3267 nvme_queue_scan(ctrl);
3268 queue_work(nvme_wq, &ctrl->async_event_work);
3269 nvme_start_queues(ctrl);
3272 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3274 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3276 cdev_device_del(&ctrl->cdev, ctrl->device);
3278 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3280 static void nvme_free_ctrl(struct device *dev)
3282 struct nvme_ctrl *ctrl =
3283 container_of(dev, struct nvme_ctrl, ctrl_device);
3284 struct nvme_subsystem *subsys = ctrl->subsys;
3286 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3287 kfree(ctrl->effects);
3290 mutex_lock(&subsys->lock);
3291 list_del(&ctrl->subsys_entry);
3292 mutex_unlock(&subsys->lock);
3293 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3296 ctrl->ops->free_ctrl(ctrl);
3299 nvme_put_subsystem(subsys);
3303 * Initialize a NVMe controller structures. This needs to be called during
3304 * earliest initialization so that we have the initialized structured around
3307 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3308 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3312 ctrl->state = NVME_CTRL_NEW;
3313 spin_lock_init(&ctrl->lock);
3314 INIT_LIST_HEAD(&ctrl->namespaces);
3315 mutex_init(&ctrl->namespaces_mutex);
3318 ctrl->quirks = quirks;
3319 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3320 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3321 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3322 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3324 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3327 ctrl->instance = ret;
3329 device_initialize(&ctrl->ctrl_device);
3330 ctrl->device = &ctrl->ctrl_device;
3331 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3332 ctrl->device->class = nvme_class;
3333 ctrl->device->parent = ctrl->dev;
3334 ctrl->device->groups = nvme_dev_attr_groups;
3335 ctrl->device->release = nvme_free_ctrl;
3336 dev_set_drvdata(ctrl->device, ctrl);
3337 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3339 goto out_release_instance;
3341 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3342 ctrl->cdev.owner = ops->module;
3343 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3348 * Initialize latency tolerance controls. The sysfs files won't
3349 * be visible to userspace unless the device actually supports APST.
3351 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3352 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3353 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3357 kfree_const(dev->kobj.name);
3358 out_release_instance:
3359 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3363 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3366 * nvme_kill_queues(): Ends all namespace queues
3367 * @ctrl: the dead controller that needs to end
3369 * Call this function when the driver determines it is unable to get the
3370 * controller in a state capable of servicing IO.
3372 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3376 mutex_lock(&ctrl->namespaces_mutex);
3378 /* Forcibly unquiesce queues to avoid blocking dispatch */
3380 blk_mq_unquiesce_queue(ctrl->admin_q);
3382 list_for_each_entry(ns, &ctrl->namespaces, list) {
3384 * Revalidating a dead namespace sets capacity to 0. This will
3385 * end buffered writers dirtying pages that can't be synced.
3387 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3389 revalidate_disk(ns->disk);
3390 blk_set_queue_dying(ns->queue);
3392 /* Forcibly unquiesce queues to avoid blocking dispatch */
3393 blk_mq_unquiesce_queue(ns->queue);
3395 mutex_unlock(&ctrl->namespaces_mutex);
3397 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3399 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3403 mutex_lock(&ctrl->namespaces_mutex);
3404 list_for_each_entry(ns, &ctrl->namespaces, list)
3405 blk_mq_unfreeze_queue(ns->queue);
3406 mutex_unlock(&ctrl->namespaces_mutex);
3408 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3410 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3414 mutex_lock(&ctrl->namespaces_mutex);
3415 list_for_each_entry(ns, &ctrl->namespaces, list) {
3416 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3420 mutex_unlock(&ctrl->namespaces_mutex);
3422 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3424 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3428 mutex_lock(&ctrl->namespaces_mutex);
3429 list_for_each_entry(ns, &ctrl->namespaces, list)
3430 blk_mq_freeze_queue_wait(ns->queue);
3431 mutex_unlock(&ctrl->namespaces_mutex);
3433 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3435 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3439 mutex_lock(&ctrl->namespaces_mutex);
3440 list_for_each_entry(ns, &ctrl->namespaces, list)
3441 blk_freeze_queue_start(ns->queue);
3442 mutex_unlock(&ctrl->namespaces_mutex);
3444 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3446 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3450 mutex_lock(&ctrl->namespaces_mutex);
3451 list_for_each_entry(ns, &ctrl->namespaces, list)
3452 blk_mq_quiesce_queue(ns->queue);
3453 mutex_unlock(&ctrl->namespaces_mutex);
3455 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3457 void nvme_start_queues(struct nvme_ctrl *ctrl)
3461 mutex_lock(&ctrl->namespaces_mutex);
3462 list_for_each_entry(ns, &ctrl->namespaces, list)
3463 blk_mq_unquiesce_queue(ns->queue);
3464 mutex_unlock(&ctrl->namespaces_mutex);
3466 EXPORT_SYMBOL_GPL(nvme_start_queues);
3468 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3470 if (!ctrl->ops->reinit_request)
3473 return blk_mq_tagset_iter(set, set->driver_data,
3474 ctrl->ops->reinit_request);
3476 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3478 int __init nvme_core_init(void)
3482 nvme_wq = alloc_workqueue("nvme-wq",
3483 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3487 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3491 nvme_class = class_create(THIS_MODULE, "nvme");
3492 if (IS_ERR(nvme_class)) {
3493 result = PTR_ERR(nvme_class);
3494 goto unregister_chrdev;
3497 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3498 if (IS_ERR(nvme_subsys_class)) {
3499 result = PTR_ERR(nvme_subsys_class);
3505 class_destroy(nvme_class);
3507 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3509 destroy_workqueue(nvme_wq);
3513 void nvme_core_exit(void)
3515 ida_destroy(&nvme_subsystems_ida);
3516 class_destroy(nvme_subsys_class);
3517 class_destroy(nvme_class);
3518 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3519 destroy_workqueue(nvme_wq);
3522 MODULE_LICENSE("GPL");
3523 MODULE_VERSION("1.0");
3524 module_init(nvme_core_init);
3525 module_exit(nvme_core_exit);