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_instance_ida);
72 static dev_t nvme_chr_devt;
73 static struct class *nvme_class;
75 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
77 return cpu_to_le32((((size / 4) - 1) << 16) | lid);
80 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
82 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
84 if (!queue_work(nvme_wq, &ctrl->reset_work))
88 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
90 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
94 ret = nvme_reset_ctrl(ctrl);
96 flush_work(&ctrl->reset_work);
100 static void nvme_delete_ctrl_work(struct work_struct *work)
102 struct nvme_ctrl *ctrl =
103 container_of(work, struct nvme_ctrl, delete_work);
105 flush_work(&ctrl->reset_work);
106 nvme_stop_ctrl(ctrl);
107 nvme_remove_namespaces(ctrl);
108 ctrl->ops->delete_ctrl(ctrl);
109 nvme_uninit_ctrl(ctrl);
113 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
115 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
117 if (!queue_work(nvme_wq, &ctrl->delete_work))
121 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
123 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
128 * Keep a reference until the work is flushed since ->delete_ctrl
129 * can free the controller.
132 ret = nvme_delete_ctrl(ctrl);
134 flush_work(&ctrl->delete_work);
138 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
140 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
142 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
145 static blk_status_t nvme_error_status(struct request *req)
147 switch (nvme_req(req)->status & 0x7ff) {
148 case NVME_SC_SUCCESS:
150 case NVME_SC_CAP_EXCEEDED:
151 return BLK_STS_NOSPC;
152 case NVME_SC_ONCS_NOT_SUPPORTED:
153 return BLK_STS_NOTSUPP;
154 case NVME_SC_WRITE_FAULT:
155 case NVME_SC_READ_ERROR:
156 case NVME_SC_UNWRITTEN_BLOCK:
157 case NVME_SC_ACCESS_DENIED:
158 case NVME_SC_READ_ONLY:
159 return BLK_STS_MEDIUM;
160 case NVME_SC_GUARD_CHECK:
161 case NVME_SC_APPTAG_CHECK:
162 case NVME_SC_REFTAG_CHECK:
163 case NVME_SC_INVALID_PI:
164 return BLK_STS_PROTECTION;
165 case NVME_SC_RESERVATION_CONFLICT:
166 return BLK_STS_NEXUS;
168 return BLK_STS_IOERR;
172 static inline bool nvme_req_needs_retry(struct request *req)
174 if (blk_noretry_request(req))
176 if (nvme_req(req)->status & NVME_SC_DNR)
178 if (nvme_req(req)->retries >= nvme_max_retries)
180 if (blk_queue_dying(req->q))
185 void nvme_complete_rq(struct request *req)
187 if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
188 nvme_req(req)->retries++;
189 blk_mq_requeue_request(req, true);
193 blk_mq_end_request(req, nvme_error_status(req));
195 EXPORT_SYMBOL_GPL(nvme_complete_rq);
197 void nvme_cancel_request(struct request *req, void *data, bool reserved)
199 if (!blk_mq_request_started(req))
202 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
203 "Cancelling I/O %d", req->tag);
205 nvme_req(req)->status = NVME_SC_ABORT_REQ;
206 blk_mq_complete_request(req);
209 EXPORT_SYMBOL_GPL(nvme_cancel_request);
211 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
212 enum nvme_ctrl_state new_state)
214 enum nvme_ctrl_state old_state;
216 bool changed = false;
218 spin_lock_irqsave(&ctrl->lock, flags);
220 old_state = ctrl->state;
225 case NVME_CTRL_RESETTING:
226 case NVME_CTRL_RECONNECTING:
233 case NVME_CTRL_RESETTING:
243 case NVME_CTRL_RECONNECTING:
246 case NVME_CTRL_RESETTING:
253 case NVME_CTRL_DELETING:
256 case NVME_CTRL_RESETTING:
257 case NVME_CTRL_RECONNECTING:
266 case NVME_CTRL_DELETING:
278 ctrl->state = new_state;
280 spin_unlock_irqrestore(&ctrl->lock, flags);
284 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
286 static void nvme_free_ns(struct kref *kref)
288 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
291 nvme_nvm_unregister(ns);
294 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
295 nvme_put_ctrl(ns->ctrl);
299 static void nvme_put_ns(struct nvme_ns *ns)
301 kref_put(&ns->kref, nvme_free_ns);
304 struct request *nvme_alloc_request(struct request_queue *q,
305 struct nvme_command *cmd, unsigned int flags, int qid)
307 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
310 if (qid == NVME_QID_ANY) {
311 req = blk_mq_alloc_request(q, op, flags);
313 req = blk_mq_alloc_request_hctx(q, op, flags,
319 req->cmd_flags |= REQ_FAILFAST_DRIVER;
320 nvme_req(req)->cmd = cmd;
324 EXPORT_SYMBOL_GPL(nvme_alloc_request);
326 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
328 struct nvme_command c;
330 memset(&c, 0, sizeof(c));
332 c.directive.opcode = nvme_admin_directive_send;
333 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
334 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
335 c.directive.dtype = NVME_DIR_IDENTIFY;
336 c.directive.tdtype = NVME_DIR_STREAMS;
337 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
339 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
342 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
344 return nvme_toggle_streams(ctrl, false);
347 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
349 return nvme_toggle_streams(ctrl, true);
352 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
353 struct streams_directive_params *s, u32 nsid)
355 struct nvme_command c;
357 memset(&c, 0, sizeof(c));
358 memset(s, 0, sizeof(*s));
360 c.directive.opcode = nvme_admin_directive_recv;
361 c.directive.nsid = cpu_to_le32(nsid);
362 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
363 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
364 c.directive.dtype = NVME_DIR_STREAMS;
366 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
369 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
371 struct streams_directive_params s;
374 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
379 ret = nvme_enable_streams(ctrl);
383 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
387 ctrl->nssa = le16_to_cpu(s.nssa);
388 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
389 dev_info(ctrl->device, "too few streams (%u) available\n",
391 nvme_disable_streams(ctrl);
395 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
396 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
401 * Check if 'req' has a write hint associated with it. If it does, assign
402 * a valid namespace stream to the write.
404 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
405 struct request *req, u16 *control,
408 enum rw_hint streamid = req->write_hint;
410 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
414 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
417 *control |= NVME_RW_DTYPE_STREAMS;
418 *dsmgmt |= streamid << 16;
421 if (streamid < ARRAY_SIZE(req->q->write_hints))
422 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
425 static inline void nvme_setup_flush(struct nvme_ns *ns,
426 struct nvme_command *cmnd)
428 memset(cmnd, 0, sizeof(*cmnd));
429 cmnd->common.opcode = nvme_cmd_flush;
430 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
433 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
434 struct nvme_command *cmnd)
436 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
437 struct nvme_dsm_range *range;
440 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
442 return BLK_STS_RESOURCE;
444 __rq_for_each_bio(bio, req) {
445 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
446 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
448 range[n].cattr = cpu_to_le32(0);
449 range[n].nlb = cpu_to_le32(nlb);
450 range[n].slba = cpu_to_le64(slba);
454 if (WARN_ON_ONCE(n != segments)) {
456 return BLK_STS_IOERR;
459 memset(cmnd, 0, sizeof(*cmnd));
460 cmnd->dsm.opcode = nvme_cmd_dsm;
461 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
462 cmnd->dsm.nr = cpu_to_le32(segments - 1);
463 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
465 req->special_vec.bv_page = virt_to_page(range);
466 req->special_vec.bv_offset = offset_in_page(range);
467 req->special_vec.bv_len = sizeof(*range) * segments;
468 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
473 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
474 struct request *req, struct nvme_command *cmnd)
476 struct nvme_ctrl *ctrl = ns->ctrl;
480 if (req->cmd_flags & REQ_FUA)
481 control |= NVME_RW_FUA;
482 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
483 control |= NVME_RW_LR;
485 if (req->cmd_flags & REQ_RAHEAD)
486 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
488 memset(cmnd, 0, sizeof(*cmnd));
489 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
490 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
491 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
492 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
494 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
495 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
499 * If formated with metadata, the block layer always provides a
500 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
501 * we enable the PRACT bit for protection information or set the
502 * namespace capacity to zero to prevent any I/O.
504 if (!blk_integrity_rq(req)) {
505 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
506 return BLK_STS_NOTSUPP;
507 control |= NVME_RW_PRINFO_PRACT;
510 switch (ns->pi_type) {
511 case NVME_NS_DPS_PI_TYPE3:
512 control |= NVME_RW_PRINFO_PRCHK_GUARD;
514 case NVME_NS_DPS_PI_TYPE1:
515 case NVME_NS_DPS_PI_TYPE2:
516 control |= NVME_RW_PRINFO_PRCHK_GUARD |
517 NVME_RW_PRINFO_PRCHK_REF;
518 cmnd->rw.reftag = cpu_to_le32(
519 nvme_block_nr(ns, blk_rq_pos(req)));
524 cmnd->rw.control = cpu_to_le16(control);
525 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
529 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
530 struct nvme_command *cmd)
532 blk_status_t ret = BLK_STS_OK;
534 if (!(req->rq_flags & RQF_DONTPREP)) {
535 nvme_req(req)->retries = 0;
536 nvme_req(req)->flags = 0;
537 req->rq_flags |= RQF_DONTPREP;
540 switch (req_op(req)) {
543 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
546 nvme_setup_flush(ns, cmd);
548 case REQ_OP_WRITE_ZEROES:
549 /* currently only aliased to deallocate for a few ctrls: */
551 ret = nvme_setup_discard(ns, req, cmd);
555 ret = nvme_setup_rw(ns, req, cmd);
559 return BLK_STS_IOERR;
562 cmd->common.command_id = req->tag;
565 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
568 * Returns 0 on success. If the result is negative, it's a Linux error code;
569 * if the result is positive, it's an NVM Express status code
571 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
572 union nvme_result *result, void *buffer, unsigned bufflen,
573 unsigned timeout, int qid, int at_head, int flags)
578 req = nvme_alloc_request(q, cmd, flags, qid);
582 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
584 if (buffer && bufflen) {
585 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
590 blk_execute_rq(req->q, NULL, req, at_head);
592 *result = nvme_req(req)->result;
593 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
596 ret = nvme_req(req)->status;
598 blk_mq_free_request(req);
601 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
603 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
604 void *buffer, unsigned bufflen)
606 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
609 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
611 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
612 unsigned len, u32 seed, bool write)
614 struct bio_integrity_payload *bip;
618 buf = kmalloc(len, GFP_KERNEL);
623 if (write && copy_from_user(buf, ubuf, len))
626 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
632 bip->bip_iter.bi_size = len;
633 bip->bip_iter.bi_sector = seed;
634 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
635 offset_in_page(buf));
645 static int nvme_submit_user_cmd(struct request_queue *q,
646 struct nvme_command *cmd, void __user *ubuffer,
647 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
648 u32 meta_seed, u32 *result, unsigned timeout)
650 bool write = nvme_is_write(cmd);
651 struct nvme_ns *ns = q->queuedata;
652 struct gendisk *disk = ns ? ns->disk : NULL;
654 struct bio *bio = NULL;
658 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
662 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
664 if (ubuffer && bufflen) {
665 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
671 if (disk && meta_buffer && meta_len) {
672 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
681 blk_execute_rq(req->q, disk, req, 0);
682 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
685 ret = nvme_req(req)->status;
687 *result = le32_to_cpu(nvme_req(req)->result.u32);
688 if (meta && !ret && !write) {
689 if (copy_to_user(meta_buffer, meta, meta_len))
695 blk_rq_unmap_user(bio);
697 blk_mq_free_request(req);
701 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
703 struct nvme_ctrl *ctrl = rq->end_io_data;
705 blk_mq_free_request(rq);
708 dev_err(ctrl->device,
709 "failed nvme_keep_alive_end_io error=%d\n",
714 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
717 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
719 struct nvme_command c;
722 memset(&c, 0, sizeof(c));
723 c.common.opcode = nvme_admin_keep_alive;
725 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
730 rq->timeout = ctrl->kato * HZ;
731 rq->end_io_data = ctrl;
733 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
738 static void nvme_keep_alive_work(struct work_struct *work)
740 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
741 struct nvme_ctrl, ka_work);
743 if (nvme_keep_alive(ctrl)) {
744 /* allocation failure, reset the controller */
745 dev_err(ctrl->device, "keep-alive failed\n");
746 nvme_reset_ctrl(ctrl);
751 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
753 if (unlikely(ctrl->kato == 0))
756 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
757 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
759 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
761 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
763 if (unlikely(ctrl->kato == 0))
766 cancel_delayed_work_sync(&ctrl->ka_work);
768 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
770 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
772 struct nvme_command c = { };
775 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
776 c.identify.opcode = nvme_admin_identify;
777 c.identify.cns = NVME_ID_CNS_CTRL;
779 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
783 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
784 sizeof(struct nvme_id_ctrl));
790 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
791 u8 *eui64, u8 *nguid, uuid_t *uuid)
793 struct nvme_command c = { };
799 c.identify.opcode = nvme_admin_identify;
800 c.identify.nsid = cpu_to_le32(nsid);
801 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
803 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
807 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
808 NVME_IDENTIFY_DATA_SIZE);
812 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
813 struct nvme_ns_id_desc *cur = data + pos;
819 case NVME_NIDT_EUI64:
820 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
821 dev_warn(ctrl->device,
822 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
826 len = NVME_NIDT_EUI64_LEN;
827 memcpy(eui64, data + pos + sizeof(*cur), len);
829 case NVME_NIDT_NGUID:
830 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
831 dev_warn(ctrl->device,
832 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
836 len = NVME_NIDT_NGUID_LEN;
837 memcpy(nguid, data + pos + sizeof(*cur), len);
840 if (cur->nidl != NVME_NIDT_UUID_LEN) {
841 dev_warn(ctrl->device,
842 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
846 len = NVME_NIDT_UUID_LEN;
847 uuid_copy(uuid, data + pos + sizeof(*cur));
850 /* Skip unnkown types */
862 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
864 struct nvme_command c = { };
866 c.identify.opcode = nvme_admin_identify;
867 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
868 c.identify.nsid = cpu_to_le32(nsid);
869 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
872 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
875 struct nvme_id_ns *id;
876 struct nvme_command c = { };
879 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
880 c.identify.opcode = nvme_admin_identify;
881 c.identify.nsid = cpu_to_le32(nsid);
882 c.identify.cns = NVME_ID_CNS_NS;
884 id = kmalloc(sizeof(*id), GFP_KERNEL);
888 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
890 dev_warn(ctrl->device, "Identify namespace failed\n");
898 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
899 void *buffer, size_t buflen, u32 *result)
901 struct nvme_command c;
902 union nvme_result res;
905 memset(&c, 0, sizeof(c));
906 c.features.opcode = nvme_admin_set_features;
907 c.features.fid = cpu_to_le32(fid);
908 c.features.dword11 = cpu_to_le32(dword11);
910 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
911 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
912 if (ret >= 0 && result)
913 *result = le32_to_cpu(res.u32);
917 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
919 u32 q_count = (*count - 1) | ((*count - 1) << 16);
921 int status, nr_io_queues;
923 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
929 * Degraded controllers might return an error when setting the queue
930 * count. We still want to be able to bring them online and offer
931 * access to the admin queue, as that might be only way to fix them up.
934 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
937 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
938 *count = min(*count, nr_io_queues);
943 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
945 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
947 struct nvme_user_io io;
948 struct nvme_command c;
949 unsigned length, meta_len;
950 void __user *metadata;
952 if (copy_from_user(&io, uio, sizeof(io)))
960 case nvme_cmd_compare:
966 length = (io.nblocks + 1) << ns->lba_shift;
967 meta_len = (io.nblocks + 1) * ns->ms;
968 metadata = (void __user *)(uintptr_t)io.metadata;
973 } else if (meta_len) {
974 if ((io.metadata & 3) || !io.metadata)
978 memset(&c, 0, sizeof(c));
979 c.rw.opcode = io.opcode;
980 c.rw.flags = io.flags;
981 c.rw.nsid = cpu_to_le32(ns->ns_id);
982 c.rw.slba = cpu_to_le64(io.slba);
983 c.rw.length = cpu_to_le16(io.nblocks);
984 c.rw.control = cpu_to_le16(io.control);
985 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
986 c.rw.reftag = cpu_to_le32(io.reftag);
987 c.rw.apptag = cpu_to_le16(io.apptag);
988 c.rw.appmask = cpu_to_le16(io.appmask);
990 return nvme_submit_user_cmd(ns->queue, &c,
991 (void __user *)(uintptr_t)io.addr, length,
992 metadata, meta_len, io.slba, NULL, 0);
995 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
996 struct nvme_passthru_cmd __user *ucmd)
998 struct nvme_passthru_cmd cmd;
999 struct nvme_command c;
1000 unsigned timeout = 0;
1003 if (!capable(CAP_SYS_ADMIN))
1005 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1010 memset(&c, 0, sizeof(c));
1011 c.common.opcode = cmd.opcode;
1012 c.common.flags = cmd.flags;
1013 c.common.nsid = cpu_to_le32(cmd.nsid);
1014 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1015 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1016 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1017 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1018 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1019 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1020 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1021 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1024 timeout = msecs_to_jiffies(cmd.timeout_ms);
1026 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1027 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1028 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1029 0, &cmd.result, timeout);
1031 if (put_user(cmd.result, &ucmd->result))
1038 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1039 unsigned int cmd, unsigned long arg)
1041 struct nvme_ns *ns = bdev->bd_disk->private_data;
1045 force_successful_syscall_return();
1047 case NVME_IOCTL_ADMIN_CMD:
1048 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1049 case NVME_IOCTL_IO_CMD:
1050 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1051 case NVME_IOCTL_SUBMIT_IO:
1052 return nvme_submit_io(ns, (void __user *)arg);
1056 return nvme_nvm_ioctl(ns, cmd, arg);
1058 if (is_sed_ioctl(cmd))
1059 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1060 (void __user *) arg);
1065 static int nvme_open(struct block_device *bdev, fmode_t mode)
1067 struct nvme_ns *ns = bdev->bd_disk->private_data;
1069 if (!kref_get_unless_zero(&ns->kref))
1074 static void nvme_release(struct gendisk *disk, fmode_t mode)
1076 nvme_put_ns(disk->private_data);
1079 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1081 /* some standard values */
1082 geo->heads = 1 << 6;
1083 geo->sectors = 1 << 5;
1084 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1088 #ifdef CONFIG_BLK_DEV_INTEGRITY
1089 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1091 struct blk_integrity integrity;
1093 memset(&integrity, 0, sizeof(integrity));
1095 case NVME_NS_DPS_PI_TYPE3:
1096 integrity.profile = &t10_pi_type3_crc;
1097 integrity.tag_size = sizeof(u16) + sizeof(u32);
1098 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1100 case NVME_NS_DPS_PI_TYPE1:
1101 case NVME_NS_DPS_PI_TYPE2:
1102 integrity.profile = &t10_pi_type1_crc;
1103 integrity.tag_size = sizeof(u16);
1104 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1107 integrity.profile = NULL;
1110 integrity.tuple_size = ms;
1111 blk_integrity_register(disk, &integrity);
1112 blk_queue_max_integrity_segments(disk->queue, 1);
1115 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1118 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1120 static void nvme_set_chunk_size(struct nvme_ns *ns)
1122 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1123 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1126 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1127 unsigned stream_alignment, struct request_queue *queue)
1129 u32 size = queue_logical_block_size(queue);
1131 if (stream_alignment)
1132 size *= stream_alignment;
1134 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1135 NVME_DSM_MAX_RANGES);
1137 queue->limits.discard_alignment = size;
1138 queue->limits.discard_granularity = size;
1140 blk_queue_max_discard_sectors(queue, UINT_MAX);
1141 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1142 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1144 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1145 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1148 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1149 struct nvme_id_ns *id, u8 *eui64, u8 *nguid, uuid_t *uuid)
1151 if (ctrl->vs >= NVME_VS(1, 1, 0))
1152 memcpy(eui64, id->eui64, sizeof(id->eui64));
1153 if (ctrl->vs >= NVME_VS(1, 2, 0))
1154 memcpy(nguid, id->nguid, sizeof(id->nguid));
1155 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1156 /* Don't treat error as fatal we potentially
1157 * already have a NGUID or EUI-64
1159 if (nvme_identify_ns_descs(ctrl, nsid, eui64, nguid, uuid))
1160 dev_warn(ctrl->device,
1161 "%s: Identify Descriptors failed\n", __func__);
1165 static void nvme_update_disk_info(struct gendisk *disk,
1166 struct nvme_ns *ns, struct nvme_id_ns *id)
1168 sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1169 unsigned stream_alignment = 0;
1171 if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1172 stream_alignment = ns->sws * ns->sgs;
1174 blk_mq_freeze_queue(disk->queue);
1175 blk_integrity_unregister(disk);
1177 blk_queue_logical_block_size(disk->queue, 1 << ns->lba_shift);
1178 if (ns->ms && !ns->ext &&
1179 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1180 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1181 if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1183 set_capacity(disk, capacity);
1185 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1186 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1187 blk_mq_unfreeze_queue(disk->queue);
1190 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1192 struct nvme_ns *ns = disk->private_data;
1195 * If identify namespace failed, use default 512 byte block size so
1196 * block layer can use before failing read/write for 0 capacity.
1198 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1199 if (ns->lba_shift == 0)
1201 ns->noiob = le16_to_cpu(id->noiob);
1202 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1203 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1204 /* the PI implementation requires metadata equal t10 pi tuple size */
1205 if (ns->ms == sizeof(struct t10_pi_tuple))
1206 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1211 nvme_set_chunk_size(ns);
1212 nvme_update_disk_info(disk, ns, id);
1215 static int nvme_revalidate_disk(struct gendisk *disk)
1217 struct nvme_ns *ns = disk->private_data;
1218 struct nvme_ctrl *ctrl = ns->ctrl;
1219 struct nvme_id_ns *id;
1220 u8 eui64[8] = { 0 }, nguid[16] = { 0 };
1221 uuid_t uuid = uuid_null;
1224 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1225 set_capacity(disk, 0);
1229 id = nvme_identify_ns(ctrl, ns->ns_id);
1233 if (id->ncap == 0) {
1238 nvme_report_ns_ids(ctrl, ns->ns_id, id, eui64, nguid, &uuid);
1239 if (!uuid_equal(&ns->uuid, &uuid) ||
1240 memcmp(&ns->nguid, &nguid, sizeof(ns->nguid)) ||
1241 memcmp(&ns->eui, &eui64, sizeof(ns->eui))) {
1242 dev_err(ctrl->device,
1243 "identifiers changed for nsid %d\n", ns->ns_id);
1252 static char nvme_pr_type(enum pr_type type)
1255 case PR_WRITE_EXCLUSIVE:
1257 case PR_EXCLUSIVE_ACCESS:
1259 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1261 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1263 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1265 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1272 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1273 u64 key, u64 sa_key, u8 op)
1275 struct nvme_ns *ns = bdev->bd_disk->private_data;
1276 struct nvme_command c;
1277 u8 data[16] = { 0, };
1279 put_unaligned_le64(key, &data[0]);
1280 put_unaligned_le64(sa_key, &data[8]);
1282 memset(&c, 0, sizeof(c));
1283 c.common.opcode = op;
1284 c.common.nsid = cpu_to_le32(ns->ns_id);
1285 c.common.cdw10[0] = cpu_to_le32(cdw10);
1287 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1290 static int nvme_pr_register(struct block_device *bdev, u64 old,
1291 u64 new, unsigned flags)
1295 if (flags & ~PR_FL_IGNORE_KEY)
1298 cdw10 = old ? 2 : 0;
1299 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1300 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1301 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1304 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1305 enum pr_type type, unsigned flags)
1309 if (flags & ~PR_FL_IGNORE_KEY)
1312 cdw10 = nvme_pr_type(type) << 8;
1313 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1314 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1317 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1318 enum pr_type type, bool abort)
1320 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1321 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1324 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1326 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1327 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1330 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1332 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1333 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1336 static const struct pr_ops nvme_pr_ops = {
1337 .pr_register = nvme_pr_register,
1338 .pr_reserve = nvme_pr_reserve,
1339 .pr_release = nvme_pr_release,
1340 .pr_preempt = nvme_pr_preempt,
1341 .pr_clear = nvme_pr_clear,
1344 #ifdef CONFIG_BLK_SED_OPAL
1345 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1348 struct nvme_ctrl *ctrl = data;
1349 struct nvme_command cmd;
1351 memset(&cmd, 0, sizeof(cmd));
1353 cmd.common.opcode = nvme_admin_security_send;
1355 cmd.common.opcode = nvme_admin_security_recv;
1356 cmd.common.nsid = 0;
1357 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1358 cmd.common.cdw10[1] = cpu_to_le32(len);
1360 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1361 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1363 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1364 #endif /* CONFIG_BLK_SED_OPAL */
1366 static const struct block_device_operations nvme_fops = {
1367 .owner = THIS_MODULE,
1368 .ioctl = nvme_ioctl,
1369 .compat_ioctl = nvme_ioctl,
1371 .release = nvme_release,
1372 .getgeo = nvme_getgeo,
1373 .revalidate_disk= nvme_revalidate_disk,
1374 .pr_ops = &nvme_pr_ops,
1377 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1379 unsigned long timeout =
1380 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1381 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1384 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1387 if ((csts & NVME_CSTS_RDY) == bit)
1391 if (fatal_signal_pending(current))
1393 if (time_after(jiffies, timeout)) {
1394 dev_err(ctrl->device,
1395 "Device not ready; aborting %s\n", enabled ?
1396 "initialisation" : "reset");
1405 * If the device has been passed off to us in an enabled state, just clear
1406 * the enabled bit. The spec says we should set the 'shutdown notification
1407 * bits', but doing so may cause the device to complete commands to the
1408 * admin queue ... and we don't know what memory that might be pointing at!
1410 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1414 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1415 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1417 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1421 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1422 msleep(NVME_QUIRK_DELAY_AMOUNT);
1424 return nvme_wait_ready(ctrl, cap, false);
1426 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1428 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1431 * Default to a 4K page size, with the intention to update this
1432 * path in the future to accomodate architectures with differing
1433 * kernel and IO page sizes.
1435 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1438 if (page_shift < dev_page_min) {
1439 dev_err(ctrl->device,
1440 "Minimum device page size %u too large for host (%u)\n",
1441 1 << dev_page_min, 1 << page_shift);
1445 ctrl->page_size = 1 << page_shift;
1447 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1448 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1449 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1450 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1451 ctrl->ctrl_config |= NVME_CC_ENABLE;
1453 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1456 return nvme_wait_ready(ctrl, cap, true);
1458 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1460 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1462 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1466 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1467 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1469 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1473 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1474 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1478 if (fatal_signal_pending(current))
1480 if (time_after(jiffies, timeout)) {
1481 dev_err(ctrl->device,
1482 "Device shutdown incomplete; abort shutdown\n");
1489 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1491 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1492 struct request_queue *q)
1496 if (ctrl->max_hw_sectors) {
1498 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1500 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1501 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1503 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1504 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1505 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1506 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1508 blk_queue_write_cache(q, vwc, vwc);
1511 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1516 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1519 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1520 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1523 dev_warn_once(ctrl->device,
1524 "could not set timestamp (%d)\n", ret);
1528 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1531 * APST (Autonomous Power State Transition) lets us program a
1532 * table of power state transitions that the controller will
1533 * perform automatically. We configure it with a simple
1534 * heuristic: we are willing to spend at most 2% of the time
1535 * transitioning between power states. Therefore, when running
1536 * in any given state, we will enter the next lower-power
1537 * non-operational state after waiting 50 * (enlat + exlat)
1538 * microseconds, as long as that state's exit latency is under
1539 * the requested maximum latency.
1541 * We will not autonomously enter any non-operational state for
1542 * which the total latency exceeds ps_max_latency_us. Users
1543 * can set ps_max_latency_us to zero to turn off APST.
1547 struct nvme_feat_auto_pst *table;
1553 * If APST isn't supported or if we haven't been initialized yet,
1554 * then don't do anything.
1559 if (ctrl->npss > 31) {
1560 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1564 table = kzalloc(sizeof(*table), GFP_KERNEL);
1568 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1569 /* Turn off APST. */
1571 dev_dbg(ctrl->device, "APST disabled\n");
1573 __le64 target = cpu_to_le64(0);
1577 * Walk through all states from lowest- to highest-power.
1578 * According to the spec, lower-numbered states use more
1579 * power. NPSS, despite the name, is the index of the
1580 * lowest-power state, not the number of states.
1582 for (state = (int)ctrl->npss; state >= 0; state--) {
1583 u64 total_latency_us, exit_latency_us, transition_ms;
1586 table->entries[state] = target;
1589 * Don't allow transitions to the deepest state
1590 * if it's quirked off.
1592 if (state == ctrl->npss &&
1593 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1597 * Is this state a useful non-operational state for
1598 * higher-power states to autonomously transition to?
1600 if (!(ctrl->psd[state].flags &
1601 NVME_PS_FLAGS_NON_OP_STATE))
1605 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1606 if (exit_latency_us > ctrl->ps_max_latency_us)
1611 le32_to_cpu(ctrl->psd[state].entry_lat);
1614 * This state is good. Use it as the APST idle
1615 * target for higher power states.
1617 transition_ms = total_latency_us + 19;
1618 do_div(transition_ms, 20);
1619 if (transition_ms > (1 << 24) - 1)
1620 transition_ms = (1 << 24) - 1;
1622 target = cpu_to_le64((state << 3) |
1623 (transition_ms << 8));
1628 if (total_latency_us > max_lat_us)
1629 max_lat_us = total_latency_us;
1635 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1637 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1638 max_ps, max_lat_us, (int)sizeof(*table), table);
1642 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1643 table, sizeof(*table), NULL);
1645 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1651 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1653 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1657 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1658 case PM_QOS_LATENCY_ANY:
1666 if (ctrl->ps_max_latency_us != latency) {
1667 ctrl->ps_max_latency_us = latency;
1668 nvme_configure_apst(ctrl);
1672 struct nvme_core_quirk_entry {
1674 * NVMe model and firmware strings are padded with spaces. For
1675 * simplicity, strings in the quirk table are padded with NULLs
1681 unsigned long quirks;
1684 static const struct nvme_core_quirk_entry core_quirks[] = {
1687 * This Toshiba device seems to die using any APST states. See:
1688 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1691 .mn = "THNSF5256GPUK TOSHIBA",
1692 .quirks = NVME_QUIRK_NO_APST,
1696 /* match is null-terminated but idstr is space-padded. */
1697 static bool string_matches(const char *idstr, const char *match, size_t len)
1704 matchlen = strlen(match);
1705 WARN_ON_ONCE(matchlen > len);
1707 if (memcmp(idstr, match, matchlen))
1710 for (; matchlen < len; matchlen++)
1711 if (idstr[matchlen] != ' ')
1717 static bool quirk_matches(const struct nvme_id_ctrl *id,
1718 const struct nvme_core_quirk_entry *q)
1720 return q->vid == le16_to_cpu(id->vid) &&
1721 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1722 string_matches(id->fr, q->fr, sizeof(id->fr));
1725 static void nvme_init_subnqn(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
1730 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1731 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1732 strcpy(ctrl->subnqn, id->subnqn);
1736 if (ctrl->vs >= NVME_VS(1, 2, 1))
1737 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1739 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1740 off = snprintf(ctrl->subnqn, NVMF_NQN_SIZE,
1741 "nqn.2014.08.org.nvmexpress:%4x%4x",
1742 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1743 memcpy(ctrl->subnqn + off, id->sn, sizeof(id->sn));
1744 off += sizeof(id->sn);
1745 memcpy(ctrl->subnqn + off, id->mn, sizeof(id->mn));
1746 off += sizeof(id->mn);
1747 memset(ctrl->subnqn + off, 0, sizeof(ctrl->subnqn) - off);
1750 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
1753 struct nvme_command c = { };
1755 c.common.opcode = nvme_admin_get_log_page;
1756 c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
1757 c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
1759 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
1763 * Initialize the cached copies of the Identify data and various controller
1764 * register in our nvme_ctrl structure. This should be called as soon as
1765 * the admin queue is fully up and running.
1767 int nvme_init_identify(struct nvme_ctrl *ctrl)
1769 struct nvme_id_ctrl *id;
1771 int ret, page_shift;
1773 bool prev_apst_enabled;
1775 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1777 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1781 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1783 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1786 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1788 if (ctrl->vs >= NVME_VS(1, 1, 0))
1789 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1791 ret = nvme_identify_ctrl(ctrl, &id);
1793 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1797 nvme_init_subnqn(ctrl, id);
1799 if (!ctrl->identified) {
1801 * Check for quirks. Quirk can depend on firmware version,
1802 * so, in principle, the set of quirks present can change
1803 * across a reset. As a possible future enhancement, we
1804 * could re-scan for quirks every time we reinitialize
1805 * the device, but we'd have to make sure that the driver
1806 * behaves intelligently if the quirks change.
1811 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1812 if (quirk_matches(id, &core_quirks[i]))
1813 ctrl->quirks |= core_quirks[i].quirks;
1817 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1818 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1819 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1822 ctrl->oacs = le16_to_cpu(id->oacs);
1823 ctrl->vid = le16_to_cpu(id->vid);
1824 ctrl->oncs = le16_to_cpup(&id->oncs);
1825 atomic_set(&ctrl->abort_limit, id->acl + 1);
1826 ctrl->vwc = id->vwc;
1827 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1828 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1829 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1830 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1832 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1834 max_hw_sectors = UINT_MAX;
1835 ctrl->max_hw_sectors =
1836 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1838 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1839 ctrl->sgls = le32_to_cpu(id->sgls);
1840 ctrl->kas = le16_to_cpu(id->kas);
1844 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
1846 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
1847 shutdown_timeout, 60);
1849 if (ctrl->shutdown_timeout != shutdown_timeout)
1850 dev_warn(ctrl->device,
1851 "Shutdown timeout set to %u seconds\n",
1852 ctrl->shutdown_timeout);
1854 ctrl->shutdown_timeout = shutdown_timeout;
1856 ctrl->npss = id->npss;
1857 ctrl->apsta = id->apsta;
1858 prev_apst_enabled = ctrl->apst_enabled;
1859 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1860 if (force_apst && id->apsta) {
1861 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1862 ctrl->apst_enabled = true;
1864 ctrl->apst_enabled = false;
1867 ctrl->apst_enabled = id->apsta;
1869 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1871 if (ctrl->ops->flags & NVME_F_FABRICS) {
1872 ctrl->icdoff = le16_to_cpu(id->icdoff);
1873 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1874 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1875 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1878 * In fabrics we need to verify the cntlid matches the
1881 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
1886 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1887 dev_err(ctrl->device,
1888 "keep-alive support is mandatory for fabrics\n");
1893 ctrl->cntlid = le16_to_cpu(id->cntlid);
1894 ctrl->hmpre = le32_to_cpu(id->hmpre);
1895 ctrl->hmmin = le32_to_cpu(id->hmmin);
1896 ctrl->hmminds = le32_to_cpu(id->hmminds);
1897 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
1902 if (ctrl->apst_enabled && !prev_apst_enabled)
1903 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1904 else if (!ctrl->apst_enabled && prev_apst_enabled)
1905 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1907 ret = nvme_configure_apst(ctrl);
1911 ret = nvme_configure_timestamp(ctrl);
1915 ret = nvme_configure_directives(ctrl);
1919 ctrl->identified = true;
1927 EXPORT_SYMBOL_GPL(nvme_init_identify);
1929 static int nvme_dev_open(struct inode *inode, struct file *file)
1931 struct nvme_ctrl *ctrl =
1932 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
1934 if (ctrl->state != NVME_CTRL_LIVE)
1935 return -EWOULDBLOCK;
1936 file->private_data = ctrl;
1940 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1945 mutex_lock(&ctrl->namespaces_mutex);
1946 if (list_empty(&ctrl->namespaces)) {
1951 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1952 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1953 dev_warn(ctrl->device,
1954 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1959 dev_warn(ctrl->device,
1960 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1961 kref_get(&ns->kref);
1962 mutex_unlock(&ctrl->namespaces_mutex);
1964 ret = nvme_user_cmd(ctrl, ns, argp);
1969 mutex_unlock(&ctrl->namespaces_mutex);
1973 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1976 struct nvme_ctrl *ctrl = file->private_data;
1977 void __user *argp = (void __user *)arg;
1980 case NVME_IOCTL_ADMIN_CMD:
1981 return nvme_user_cmd(ctrl, NULL, argp);
1982 case NVME_IOCTL_IO_CMD:
1983 return nvme_dev_user_cmd(ctrl, argp);
1984 case NVME_IOCTL_RESET:
1985 dev_warn(ctrl->device, "resetting controller\n");
1986 return nvme_reset_ctrl_sync(ctrl);
1987 case NVME_IOCTL_SUBSYS_RESET:
1988 return nvme_reset_subsystem(ctrl);
1989 case NVME_IOCTL_RESCAN:
1990 nvme_queue_scan(ctrl);
1997 static const struct file_operations nvme_dev_fops = {
1998 .owner = THIS_MODULE,
1999 .open = nvme_dev_open,
2000 .unlocked_ioctl = nvme_dev_ioctl,
2001 .compat_ioctl = nvme_dev_ioctl,
2004 static ssize_t nvme_sysfs_reset(struct device *dev,
2005 struct device_attribute *attr, const char *buf,
2008 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2011 ret = nvme_reset_ctrl_sync(ctrl);
2016 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2018 static ssize_t nvme_sysfs_rescan(struct device *dev,
2019 struct device_attribute *attr, const char *buf,
2022 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2024 nvme_queue_scan(ctrl);
2027 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2029 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2032 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2033 struct nvme_ctrl *ctrl = ns->ctrl;
2034 int serial_len = sizeof(ctrl->serial);
2035 int model_len = sizeof(ctrl->model);
2037 if (!uuid_is_null(&ns->uuid))
2038 return sprintf(buf, "uuid.%pU\n", &ns->uuid);
2040 if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2041 return sprintf(buf, "eui.%16phN\n", ns->nguid);
2043 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2044 return sprintf(buf, "eui.%8phN\n", ns->eui);
2046 while (serial_len > 0 && (ctrl->serial[serial_len - 1] == ' ' ||
2047 ctrl->serial[serial_len - 1] == '\0'))
2049 while (model_len > 0 && (ctrl->model[model_len - 1] == ' ' ||
2050 ctrl->model[model_len - 1] == '\0'))
2053 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
2054 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
2056 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2058 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2061 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2062 return sprintf(buf, "%pU\n", ns->nguid);
2064 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2066 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2069 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2071 /* For backward compatibility expose the NGUID to userspace if
2072 * we have no UUID set
2074 if (uuid_is_null(&ns->uuid)) {
2075 printk_ratelimited(KERN_WARNING
2076 "No UUID available providing old NGUID\n");
2077 return sprintf(buf, "%pU\n", ns->nguid);
2079 return sprintf(buf, "%pU\n", &ns->uuid);
2081 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2083 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2086 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2087 return sprintf(buf, "%8phd\n", ns->eui);
2089 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2091 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2094 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2095 return sprintf(buf, "%d\n", ns->ns_id);
2097 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2099 static struct attribute *nvme_ns_attrs[] = {
2100 &dev_attr_wwid.attr,
2101 &dev_attr_uuid.attr,
2102 &dev_attr_nguid.attr,
2104 &dev_attr_nsid.attr,
2108 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
2109 struct attribute *a, int n)
2111 struct device *dev = container_of(kobj, struct device, kobj);
2112 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2114 if (a == &dev_attr_uuid.attr) {
2115 if (uuid_is_null(&ns->uuid) ||
2116 !memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2119 if (a == &dev_attr_nguid.attr) {
2120 if (!memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2123 if (a == &dev_attr_eui.attr) {
2124 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2130 static const struct attribute_group nvme_ns_attr_group = {
2131 .attrs = nvme_ns_attrs,
2132 .is_visible = nvme_ns_attrs_are_visible,
2135 #define nvme_show_str_function(field) \
2136 static ssize_t field##_show(struct device *dev, \
2137 struct device_attribute *attr, char *buf) \
2139 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2140 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
2142 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2144 #define nvme_show_int_function(field) \
2145 static ssize_t field##_show(struct device *dev, \
2146 struct device_attribute *attr, char *buf) \
2148 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2149 return sprintf(buf, "%d\n", ctrl->field); \
2151 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2153 nvme_show_str_function(model);
2154 nvme_show_str_function(serial);
2155 nvme_show_str_function(firmware_rev);
2156 nvme_show_int_function(cntlid);
2158 static ssize_t nvme_sysfs_delete(struct device *dev,
2159 struct device_attribute *attr, const char *buf,
2162 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2164 if (device_remove_file_self(dev, attr))
2165 nvme_delete_ctrl_sync(ctrl);
2168 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2170 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2171 struct device_attribute *attr,
2174 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2176 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2178 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2180 static ssize_t nvme_sysfs_show_state(struct device *dev,
2181 struct device_attribute *attr,
2184 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2185 static const char *const state_name[] = {
2186 [NVME_CTRL_NEW] = "new",
2187 [NVME_CTRL_LIVE] = "live",
2188 [NVME_CTRL_RESETTING] = "resetting",
2189 [NVME_CTRL_RECONNECTING]= "reconnecting",
2190 [NVME_CTRL_DELETING] = "deleting",
2191 [NVME_CTRL_DEAD] = "dead",
2194 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2195 state_name[ctrl->state])
2196 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2198 return sprintf(buf, "unknown state\n");
2201 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2203 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2204 struct device_attribute *attr,
2207 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2209 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subnqn);
2211 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2213 static ssize_t nvme_sysfs_show_address(struct device *dev,
2214 struct device_attribute *attr,
2217 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2219 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2221 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2223 static struct attribute *nvme_dev_attrs[] = {
2224 &dev_attr_reset_controller.attr,
2225 &dev_attr_rescan_controller.attr,
2226 &dev_attr_model.attr,
2227 &dev_attr_serial.attr,
2228 &dev_attr_firmware_rev.attr,
2229 &dev_attr_cntlid.attr,
2230 &dev_attr_delete_controller.attr,
2231 &dev_attr_transport.attr,
2232 &dev_attr_subsysnqn.attr,
2233 &dev_attr_address.attr,
2234 &dev_attr_state.attr,
2238 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2239 struct attribute *a, int n)
2241 struct device *dev = container_of(kobj, struct device, kobj);
2242 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2244 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2246 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2252 static struct attribute_group nvme_dev_attrs_group = {
2253 .attrs = nvme_dev_attrs,
2254 .is_visible = nvme_dev_attrs_are_visible,
2257 static const struct attribute_group *nvme_dev_attr_groups[] = {
2258 &nvme_dev_attrs_group,
2262 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2264 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2265 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2267 return nsa->ns_id - nsb->ns_id;
2270 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2272 struct nvme_ns *ns, *ret = NULL;
2274 mutex_lock(&ctrl->namespaces_mutex);
2275 list_for_each_entry(ns, &ctrl->namespaces, list) {
2276 if (ns->ns_id == nsid) {
2277 if (!kref_get_unless_zero(&ns->kref))
2282 if (ns->ns_id > nsid)
2285 mutex_unlock(&ctrl->namespaces_mutex);
2289 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2291 struct streams_directive_params s;
2294 if (!ctrl->nr_streams)
2297 ret = nvme_get_stream_params(ctrl, &s, ns->ns_id);
2301 ns->sws = le32_to_cpu(s.sws);
2302 ns->sgs = le16_to_cpu(s.sgs);
2305 unsigned int bs = 1 << ns->lba_shift;
2307 blk_queue_io_min(ns->queue, bs * ns->sws);
2309 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2315 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2318 struct gendisk *disk;
2319 struct nvme_id_ns *id;
2320 char disk_name[DISK_NAME_LEN];
2321 int node = dev_to_node(ctrl->dev);
2323 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2327 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2328 if (ns->instance < 0)
2331 ns->queue = blk_mq_init_queue(ctrl->tagset);
2332 if (IS_ERR(ns->queue))
2333 goto out_release_instance;
2334 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2335 ns->queue->queuedata = ns;
2338 kref_init(&ns->kref);
2340 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2342 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2343 nvme_set_queue_limits(ctrl, ns->queue);
2344 nvme_setup_streams_ns(ctrl, ns);
2346 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2348 id = nvme_identify_ns(ctrl, nsid);
2350 goto out_free_queue;
2355 nvme_report_ns_ids(ctrl, ns->ns_id, id, ns->eui, ns->nguid, &ns->uuid);
2357 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
2358 if (nvme_nvm_register(ns, disk_name, node)) {
2359 dev_warn(ctrl->device, "LightNVM init failure\n");
2364 disk = alloc_disk_node(0, node);
2368 disk->fops = &nvme_fops;
2369 disk->private_data = ns;
2370 disk->queue = ns->queue;
2371 disk->flags = GENHD_FL_EXT_DEVT;
2372 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2375 __nvme_revalidate_disk(disk, id);
2377 mutex_lock(&ctrl->namespaces_mutex);
2378 list_add_tail(&ns->list, &ctrl->namespaces);
2379 mutex_unlock(&ctrl->namespaces_mutex);
2381 nvme_get_ctrl(ctrl);
2385 device_add_disk(ctrl->device, ns->disk);
2386 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2387 &nvme_ns_attr_group))
2388 pr_warn("%s: failed to create sysfs group for identification\n",
2389 ns->disk->disk_name);
2390 if (ns->ndev && nvme_nvm_register_sysfs(ns))
2391 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2392 ns->disk->disk_name);
2397 blk_cleanup_queue(ns->queue);
2398 out_release_instance:
2399 ida_simple_remove(&ctrl->ns_ida, ns->instance);
2404 static void nvme_ns_remove(struct nvme_ns *ns)
2406 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2409 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2410 if (blk_get_integrity(ns->disk))
2411 blk_integrity_unregister(ns->disk);
2412 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2413 &nvme_ns_attr_group);
2415 nvme_nvm_unregister_sysfs(ns);
2416 del_gendisk(ns->disk);
2417 blk_cleanup_queue(ns->queue);
2420 mutex_lock(&ns->ctrl->namespaces_mutex);
2421 list_del_init(&ns->list);
2422 mutex_unlock(&ns->ctrl->namespaces_mutex);
2427 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2431 ns = nvme_find_get_ns(ctrl, nsid);
2433 if (ns->disk && revalidate_disk(ns->disk))
2437 nvme_alloc_ns(ctrl, nsid);
2440 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2443 struct nvme_ns *ns, *next;
2445 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2446 if (ns->ns_id > nsid)
2451 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2455 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2458 ns_list = kzalloc(0x1000, GFP_KERNEL);
2462 for (i = 0; i < num_lists; i++) {
2463 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2467 for (j = 0; j < min(nn, 1024U); j++) {
2468 nsid = le32_to_cpu(ns_list[j]);
2472 nvme_validate_ns(ctrl, nsid);
2474 while (++prev < nsid) {
2475 ns = nvme_find_get_ns(ctrl, prev);
2485 nvme_remove_invalid_namespaces(ctrl, prev);
2491 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2495 for (i = 1; i <= nn; i++)
2496 nvme_validate_ns(ctrl, i);
2498 nvme_remove_invalid_namespaces(ctrl, nn);
2501 static void nvme_scan_work(struct work_struct *work)
2503 struct nvme_ctrl *ctrl =
2504 container_of(work, struct nvme_ctrl, scan_work);
2505 struct nvme_id_ctrl *id;
2508 if (ctrl->state != NVME_CTRL_LIVE)
2511 if (nvme_identify_ctrl(ctrl, &id))
2514 nn = le32_to_cpu(id->nn);
2515 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2516 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2517 if (!nvme_scan_ns_list(ctrl, nn))
2520 nvme_scan_ns_sequential(ctrl, nn);
2522 mutex_lock(&ctrl->namespaces_mutex);
2523 list_sort(NULL, &ctrl->namespaces, ns_cmp);
2524 mutex_unlock(&ctrl->namespaces_mutex);
2528 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2531 * Do not queue new scan work when a controller is reset during
2534 if (ctrl->state == NVME_CTRL_LIVE)
2535 queue_work(nvme_wq, &ctrl->scan_work);
2537 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2540 * This function iterates the namespace list unlocked to allow recovery from
2541 * controller failure. It is up to the caller to ensure the namespace list is
2542 * not modified by scan work while this function is executing.
2544 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2546 struct nvme_ns *ns, *next;
2549 * The dead states indicates the controller was not gracefully
2550 * disconnected. In that case, we won't be able to flush any data while
2551 * removing the namespaces' disks; fail all the queues now to avoid
2552 * potentially having to clean up the failed sync later.
2554 if (ctrl->state == NVME_CTRL_DEAD)
2555 nvme_kill_queues(ctrl);
2557 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2560 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2562 static void nvme_async_event_work(struct work_struct *work)
2564 struct nvme_ctrl *ctrl =
2565 container_of(work, struct nvme_ctrl, async_event_work);
2567 spin_lock_irq(&ctrl->lock);
2568 while (ctrl->state == NVME_CTRL_LIVE && ctrl->event_limit > 0) {
2569 int aer_idx = --ctrl->event_limit;
2571 spin_unlock_irq(&ctrl->lock);
2572 ctrl->ops->submit_async_event(ctrl, aer_idx);
2573 spin_lock_irq(&ctrl->lock);
2575 spin_unlock_irq(&ctrl->lock);
2578 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
2583 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
2589 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
2592 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
2594 struct nvme_fw_slot_info_log *log;
2596 log = kmalloc(sizeof(*log), GFP_KERNEL);
2600 if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
2601 dev_warn(ctrl->device,
2602 "Get FW SLOT INFO log error\n");
2606 static void nvme_fw_act_work(struct work_struct *work)
2608 struct nvme_ctrl *ctrl = container_of(work,
2609 struct nvme_ctrl, fw_act_work);
2610 unsigned long fw_act_timeout;
2613 fw_act_timeout = jiffies +
2614 msecs_to_jiffies(ctrl->mtfa * 100);
2616 fw_act_timeout = jiffies +
2617 msecs_to_jiffies(admin_timeout * 1000);
2619 nvme_stop_queues(ctrl);
2620 while (nvme_ctrl_pp_status(ctrl)) {
2621 if (time_after(jiffies, fw_act_timeout)) {
2622 dev_warn(ctrl->device,
2623 "Fw activation timeout, reset controller\n");
2624 nvme_reset_ctrl(ctrl);
2630 if (ctrl->state != NVME_CTRL_LIVE)
2633 nvme_start_queues(ctrl);
2634 /* read FW slot information to clear the AER */
2635 nvme_get_fw_slot_info(ctrl);
2638 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2639 union nvme_result *res)
2641 u32 result = le32_to_cpu(res->u32);
2644 switch (le16_to_cpu(status) >> 1) {
2645 case NVME_SC_SUCCESS:
2648 case NVME_SC_ABORT_REQ:
2649 ++ctrl->event_limit;
2650 if (ctrl->state == NVME_CTRL_LIVE)
2651 queue_work(nvme_wq, &ctrl->async_event_work);
2660 switch (result & 0xff07) {
2661 case NVME_AER_NOTICE_NS_CHANGED:
2662 dev_info(ctrl->device, "rescanning\n");
2663 nvme_queue_scan(ctrl);
2665 case NVME_AER_NOTICE_FW_ACT_STARTING:
2666 queue_work(nvme_wq, &ctrl->fw_act_work);
2669 dev_warn(ctrl->device, "async event result %08x\n", result);
2672 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2674 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2676 ctrl->event_limit = NVME_NR_AERS;
2677 queue_work(nvme_wq, &ctrl->async_event_work);
2679 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2681 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
2683 nvme_stop_keep_alive(ctrl);
2684 flush_work(&ctrl->async_event_work);
2685 flush_work(&ctrl->scan_work);
2686 cancel_work_sync(&ctrl->fw_act_work);
2688 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
2690 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
2693 nvme_start_keep_alive(ctrl);
2695 if (ctrl->queue_count > 1) {
2696 nvme_queue_scan(ctrl);
2697 nvme_queue_async_events(ctrl);
2698 nvme_start_queues(ctrl);
2701 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
2703 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2705 cdev_device_del(&ctrl->cdev, ctrl->device);
2707 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2709 static void nvme_free_ctrl(struct device *dev)
2711 struct nvme_ctrl *ctrl =
2712 container_of(dev, struct nvme_ctrl, ctrl_device);
2714 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
2715 ida_destroy(&ctrl->ns_ida);
2717 ctrl->ops->free_ctrl(ctrl);
2721 * Initialize a NVMe controller structures. This needs to be called during
2722 * earliest initialization so that we have the initialized structured around
2725 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2726 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2730 ctrl->state = NVME_CTRL_NEW;
2731 spin_lock_init(&ctrl->lock);
2732 INIT_LIST_HEAD(&ctrl->namespaces);
2733 mutex_init(&ctrl->namespaces_mutex);
2736 ctrl->quirks = quirks;
2737 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2738 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2739 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
2740 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
2742 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
2745 ctrl->instance = ret;
2747 device_initialize(&ctrl->ctrl_device);
2748 ctrl->device = &ctrl->ctrl_device;
2749 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
2750 ctrl->device->class = nvme_class;
2751 ctrl->device->parent = ctrl->dev;
2752 ctrl->device->groups = nvme_dev_attr_groups;
2753 ctrl->device->release = nvme_free_ctrl;
2754 dev_set_drvdata(ctrl->device, ctrl);
2755 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
2757 goto out_release_instance;
2759 cdev_init(&ctrl->cdev, &nvme_dev_fops);
2760 ctrl->cdev.owner = ops->module;
2761 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
2765 ida_init(&ctrl->ns_ida);
2768 * Initialize latency tolerance controls. The sysfs files won't
2769 * be visible to userspace unless the device actually supports APST.
2771 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2772 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2773 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2777 kfree_const(dev->kobj.name);
2778 out_release_instance:
2779 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
2783 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2786 * nvme_kill_queues(): Ends all namespace queues
2787 * @ctrl: the dead controller that needs to end
2789 * Call this function when the driver determines it is unable to get the
2790 * controller in a state capable of servicing IO.
2792 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2796 mutex_lock(&ctrl->namespaces_mutex);
2798 /* Forcibly unquiesce queues to avoid blocking dispatch */
2800 blk_mq_unquiesce_queue(ctrl->admin_q);
2802 list_for_each_entry(ns, &ctrl->namespaces, list) {
2804 * Revalidating a dead namespace sets capacity to 0. This will
2805 * end buffered writers dirtying pages that can't be synced.
2807 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2809 revalidate_disk(ns->disk);
2810 blk_set_queue_dying(ns->queue);
2812 /* Forcibly unquiesce queues to avoid blocking dispatch */
2813 blk_mq_unquiesce_queue(ns->queue);
2815 mutex_unlock(&ctrl->namespaces_mutex);
2817 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2819 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2823 mutex_lock(&ctrl->namespaces_mutex);
2824 list_for_each_entry(ns, &ctrl->namespaces, list)
2825 blk_mq_unfreeze_queue(ns->queue);
2826 mutex_unlock(&ctrl->namespaces_mutex);
2828 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2830 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2834 mutex_lock(&ctrl->namespaces_mutex);
2835 list_for_each_entry(ns, &ctrl->namespaces, list) {
2836 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2840 mutex_unlock(&ctrl->namespaces_mutex);
2842 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2844 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2848 mutex_lock(&ctrl->namespaces_mutex);
2849 list_for_each_entry(ns, &ctrl->namespaces, list)
2850 blk_mq_freeze_queue_wait(ns->queue);
2851 mutex_unlock(&ctrl->namespaces_mutex);
2853 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2855 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2859 mutex_lock(&ctrl->namespaces_mutex);
2860 list_for_each_entry(ns, &ctrl->namespaces, list)
2861 blk_freeze_queue_start(ns->queue);
2862 mutex_unlock(&ctrl->namespaces_mutex);
2864 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2866 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2870 mutex_lock(&ctrl->namespaces_mutex);
2871 list_for_each_entry(ns, &ctrl->namespaces, list)
2872 blk_mq_quiesce_queue(ns->queue);
2873 mutex_unlock(&ctrl->namespaces_mutex);
2875 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2877 void nvme_start_queues(struct nvme_ctrl *ctrl)
2881 mutex_lock(&ctrl->namespaces_mutex);
2882 list_for_each_entry(ns, &ctrl->namespaces, list)
2883 blk_mq_unquiesce_queue(ns->queue);
2884 mutex_unlock(&ctrl->namespaces_mutex);
2886 EXPORT_SYMBOL_GPL(nvme_start_queues);
2888 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
2890 if (!ctrl->ops->reinit_request)
2893 return blk_mq_tagset_iter(set, set->driver_data,
2894 ctrl->ops->reinit_request);
2896 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
2898 int __init nvme_core_init(void)
2902 nvme_wq = alloc_workqueue("nvme-wq",
2903 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
2907 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
2911 nvme_class = class_create(THIS_MODULE, "nvme");
2912 if (IS_ERR(nvme_class)) {
2913 result = PTR_ERR(nvme_class);
2914 goto unregister_chrdev;
2920 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
2922 destroy_workqueue(nvme_wq);
2926 void nvme_core_exit(void)
2928 class_destroy(nvme_class);
2929 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
2930 destroy_workqueue(nvme_wq);
2933 MODULE_LICENSE("GPL");
2934 MODULE_VERSION("1.0");
2935 module_init(nvme_core_init);
2936 module_exit(nvme_core_exit);