2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
29 #include <linux/pm_qos.h>
30 #include <asm/unaligned.h>
35 #define NVME_MINORS (1U << MINORBITS)
37 unsigned char admin_timeout = 60;
38 module_param(admin_timeout, byte, 0644);
39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout);
42 unsigned char nvme_io_timeout = 30;
43 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
44 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
45 EXPORT_SYMBOL_GPL(nvme_io_timeout);
47 static unsigned char shutdown_timeout = 5;
48 module_param(shutdown_timeout, byte, 0644);
49 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
51 static u8 nvme_max_retries = 5;
52 module_param_named(max_retries, nvme_max_retries, byte, 0644);
53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
55 static int nvme_char_major;
56 module_param(nvme_char_major, int, 0);
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61 "max power saving latency for new devices; use PM QOS to change per device");
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
71 struct workqueue_struct *nvme_wq;
72 EXPORT_SYMBOL_GPL(nvme_wq);
74 static LIST_HEAD(nvme_ctrl_list);
75 static DEFINE_SPINLOCK(dev_list_lock);
77 static struct class *nvme_class;
79 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
81 return cpu_to_le32((((size / 4) - 1) << 16) | lid);
84 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
86 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
88 if (!queue_work(nvme_wq, &ctrl->reset_work))
92 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
94 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
98 ret = nvme_reset_ctrl(ctrl);
100 flush_work(&ctrl->reset_work);
104 static blk_status_t nvme_error_status(struct request *req)
106 switch (nvme_req(req)->status & 0x7ff) {
107 case NVME_SC_SUCCESS:
109 case NVME_SC_CAP_EXCEEDED:
110 return BLK_STS_NOSPC;
111 case NVME_SC_ONCS_NOT_SUPPORTED:
112 return BLK_STS_NOTSUPP;
113 case NVME_SC_WRITE_FAULT:
114 case NVME_SC_READ_ERROR:
115 case NVME_SC_UNWRITTEN_BLOCK:
116 case NVME_SC_ACCESS_DENIED:
117 case NVME_SC_READ_ONLY:
118 return BLK_STS_MEDIUM;
119 case NVME_SC_GUARD_CHECK:
120 case NVME_SC_APPTAG_CHECK:
121 case NVME_SC_REFTAG_CHECK:
122 case NVME_SC_INVALID_PI:
123 return BLK_STS_PROTECTION;
124 case NVME_SC_RESERVATION_CONFLICT:
125 return BLK_STS_NEXUS;
127 return BLK_STS_IOERR;
131 static inline bool nvme_req_needs_retry(struct request *req)
133 if (blk_noretry_request(req))
135 if (nvme_req(req)->status & NVME_SC_DNR)
137 if (jiffies - req->start_time >= req->timeout)
139 if (nvme_req(req)->retries >= nvme_max_retries)
144 void nvme_complete_rq(struct request *req)
146 if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
147 nvme_req(req)->retries++;
148 blk_mq_requeue_request(req, true);
152 blk_mq_end_request(req, nvme_error_status(req));
154 EXPORT_SYMBOL_GPL(nvme_complete_rq);
156 void nvme_cancel_request(struct request *req, void *data, bool reserved)
160 if (!blk_mq_request_started(req))
163 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
164 "Cancelling I/O %d", req->tag);
166 status = NVME_SC_ABORT_REQ;
167 if (blk_queue_dying(req->q))
168 status |= NVME_SC_DNR;
169 nvme_req(req)->status = status;
170 blk_mq_complete_request(req);
173 EXPORT_SYMBOL_GPL(nvme_cancel_request);
175 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
176 enum nvme_ctrl_state new_state)
178 enum nvme_ctrl_state old_state;
180 bool changed = false;
182 spin_lock_irqsave(&ctrl->lock, flags);
184 old_state = ctrl->state;
189 case NVME_CTRL_RESETTING:
190 case NVME_CTRL_RECONNECTING:
197 case NVME_CTRL_RESETTING:
207 case NVME_CTRL_RECONNECTING:
216 case NVME_CTRL_DELETING:
219 case NVME_CTRL_RESETTING:
220 case NVME_CTRL_RECONNECTING:
229 case NVME_CTRL_DELETING:
241 ctrl->state = new_state;
243 spin_unlock_irqrestore(&ctrl->lock, flags);
247 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
249 static void nvme_free_ns(struct kref *kref)
251 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
254 nvme_nvm_unregister(ns);
257 spin_lock(&dev_list_lock);
258 ns->disk->private_data = NULL;
259 spin_unlock(&dev_list_lock);
263 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
264 nvme_put_ctrl(ns->ctrl);
268 static void nvme_put_ns(struct nvme_ns *ns)
270 kref_put(&ns->kref, nvme_free_ns);
273 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
277 spin_lock(&dev_list_lock);
278 ns = disk->private_data;
280 if (!kref_get_unless_zero(&ns->kref))
282 if (!try_module_get(ns->ctrl->ops->module))
285 spin_unlock(&dev_list_lock);
290 kref_put(&ns->kref, nvme_free_ns);
292 spin_unlock(&dev_list_lock);
296 struct request *nvme_alloc_request(struct request_queue *q,
297 struct nvme_command *cmd, unsigned int flags, int qid)
299 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
302 if (qid == NVME_QID_ANY) {
303 req = blk_mq_alloc_request(q, op, flags);
305 req = blk_mq_alloc_request_hctx(q, op, flags,
311 req->cmd_flags |= REQ_FAILFAST_DRIVER;
312 nvme_req(req)->cmd = cmd;
316 EXPORT_SYMBOL_GPL(nvme_alloc_request);
318 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
320 struct nvme_command c;
322 memset(&c, 0, sizeof(c));
324 c.directive.opcode = nvme_admin_directive_send;
325 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
326 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
327 c.directive.dtype = NVME_DIR_IDENTIFY;
328 c.directive.tdtype = NVME_DIR_STREAMS;
329 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
331 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
334 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
336 return nvme_toggle_streams(ctrl, false);
339 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
341 return nvme_toggle_streams(ctrl, true);
344 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
345 struct streams_directive_params *s, u32 nsid)
347 struct nvme_command c;
349 memset(&c, 0, sizeof(c));
350 memset(s, 0, sizeof(*s));
352 c.directive.opcode = nvme_admin_directive_recv;
353 c.directive.nsid = cpu_to_le32(nsid);
354 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
355 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
356 c.directive.dtype = NVME_DIR_STREAMS;
358 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
361 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
363 struct streams_directive_params s;
366 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
371 ret = nvme_enable_streams(ctrl);
375 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
379 ctrl->nssa = le16_to_cpu(s.nssa);
380 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
381 dev_info(ctrl->device, "too few streams (%u) available\n",
383 nvme_disable_streams(ctrl);
387 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
388 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
393 * Check if 'req' has a write hint associated with it. If it does, assign
394 * a valid namespace stream to the write.
396 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
397 struct request *req, u16 *control,
400 enum rw_hint streamid = req->write_hint;
402 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
406 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
409 *control |= NVME_RW_DTYPE_STREAMS;
410 *dsmgmt |= streamid << 16;
413 if (streamid < ARRAY_SIZE(req->q->write_hints))
414 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
417 static inline void nvme_setup_flush(struct nvme_ns *ns,
418 struct nvme_command *cmnd)
420 memset(cmnd, 0, sizeof(*cmnd));
421 cmnd->common.opcode = nvme_cmd_flush;
422 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
425 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
426 struct nvme_command *cmnd)
428 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
429 struct nvme_dsm_range *range;
432 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
434 return BLK_STS_RESOURCE;
436 __rq_for_each_bio(bio, req) {
437 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
438 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
440 range[n].cattr = cpu_to_le32(0);
441 range[n].nlb = cpu_to_le32(nlb);
442 range[n].slba = cpu_to_le64(slba);
446 if (WARN_ON_ONCE(n != segments)) {
448 return BLK_STS_IOERR;
451 memset(cmnd, 0, sizeof(*cmnd));
452 cmnd->dsm.opcode = nvme_cmd_dsm;
453 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
454 cmnd->dsm.nr = cpu_to_le32(segments - 1);
455 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
457 req->special_vec.bv_page = virt_to_page(range);
458 req->special_vec.bv_offset = offset_in_page(range);
459 req->special_vec.bv_len = sizeof(*range) * segments;
460 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
465 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
466 struct request *req, struct nvme_command *cmnd)
468 struct nvme_ctrl *ctrl = ns->ctrl;
473 * If formated with metadata, require the block layer provide a buffer
474 * unless this namespace is formated such that the metadata can be
475 * stripped/generated by the controller with PRACT=1.
478 (!ns->pi_type || ns->ms != sizeof(struct t10_pi_tuple)) &&
479 !blk_integrity_rq(req) && !blk_rq_is_passthrough(req))
480 return BLK_STS_NOTSUPP;
482 if (req->cmd_flags & REQ_FUA)
483 control |= NVME_RW_FUA;
484 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
485 control |= NVME_RW_LR;
487 if (req->cmd_flags & REQ_RAHEAD)
488 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
490 memset(cmnd, 0, sizeof(*cmnd));
491 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
492 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
493 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
494 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
496 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
497 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
500 switch (ns->pi_type) {
501 case NVME_NS_DPS_PI_TYPE3:
502 control |= NVME_RW_PRINFO_PRCHK_GUARD;
504 case NVME_NS_DPS_PI_TYPE1:
505 case NVME_NS_DPS_PI_TYPE2:
506 control |= NVME_RW_PRINFO_PRCHK_GUARD |
507 NVME_RW_PRINFO_PRCHK_REF;
508 cmnd->rw.reftag = cpu_to_le32(
509 nvme_block_nr(ns, blk_rq_pos(req)));
512 if (!blk_integrity_rq(req))
513 control |= NVME_RW_PRINFO_PRACT;
516 cmnd->rw.control = cpu_to_le16(control);
517 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
521 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
522 struct nvme_command *cmd)
524 blk_status_t ret = BLK_STS_OK;
526 if (!(req->rq_flags & RQF_DONTPREP)) {
527 nvme_req(req)->retries = 0;
528 nvme_req(req)->flags = 0;
529 req->rq_flags |= RQF_DONTPREP;
532 switch (req_op(req)) {
535 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
538 nvme_setup_flush(ns, cmd);
540 case REQ_OP_WRITE_ZEROES:
541 /* currently only aliased to deallocate for a few ctrls: */
543 ret = nvme_setup_discard(ns, req, cmd);
547 ret = nvme_setup_rw(ns, req, cmd);
551 return BLK_STS_IOERR;
554 cmd->common.command_id = req->tag;
557 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
560 * Returns 0 on success. If the result is negative, it's a Linux error code;
561 * if the result is positive, it's an NVM Express status code
563 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
564 union nvme_result *result, void *buffer, unsigned bufflen,
565 unsigned timeout, int qid, int at_head, int flags)
570 req = nvme_alloc_request(q, cmd, flags, qid);
574 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
576 if (buffer && bufflen) {
577 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
582 blk_execute_rq(req->q, NULL, req, at_head);
584 *result = nvme_req(req)->result;
585 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
588 ret = nvme_req(req)->status;
590 blk_mq_free_request(req);
593 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
595 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
596 void *buffer, unsigned bufflen)
598 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
601 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
603 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
604 unsigned len, u32 seed, bool write)
606 struct bio_integrity_payload *bip;
610 buf = kmalloc(len, GFP_KERNEL);
615 if (write && copy_from_user(buf, ubuf, len))
618 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
624 bip->bip_iter.bi_size = len;
625 bip->bip_iter.bi_sector = seed;
626 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
627 offset_in_page(buf));
637 static int nvme_submit_user_cmd(struct request_queue *q,
638 struct nvme_command *cmd, void __user *ubuffer,
639 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
640 u32 meta_seed, u32 *result, unsigned timeout)
642 bool write = nvme_is_write(cmd);
643 struct nvme_ns *ns = q->queuedata;
644 struct gendisk *disk = ns ? ns->disk : NULL;
646 struct bio *bio = NULL;
650 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
654 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
656 if (ubuffer && bufflen) {
657 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
663 if (disk && meta_buffer && meta_len) {
664 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
673 blk_execute_rq(req->q, disk, req, 0);
674 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
677 ret = nvme_req(req)->status;
679 *result = le32_to_cpu(nvme_req(req)->result.u32);
680 if (meta && !ret && !write) {
681 if (copy_to_user(meta_buffer, meta, meta_len))
687 blk_rq_unmap_user(bio);
689 blk_mq_free_request(req);
693 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
695 struct nvme_ctrl *ctrl = rq->end_io_data;
697 blk_mq_free_request(rq);
700 dev_err(ctrl->device,
701 "failed nvme_keep_alive_end_io error=%d\n",
706 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
709 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
711 struct nvme_command c;
714 memset(&c, 0, sizeof(c));
715 c.common.opcode = nvme_admin_keep_alive;
717 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
722 rq->timeout = ctrl->kato * HZ;
723 rq->end_io_data = ctrl;
725 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
730 static void nvme_keep_alive_work(struct work_struct *work)
732 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
733 struct nvme_ctrl, ka_work);
735 if (nvme_keep_alive(ctrl)) {
736 /* allocation failure, reset the controller */
737 dev_err(ctrl->device, "keep-alive failed\n");
738 nvme_reset_ctrl(ctrl);
743 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
745 if (unlikely(ctrl->kato == 0))
748 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
749 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
751 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
753 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
755 if (unlikely(ctrl->kato == 0))
758 cancel_delayed_work_sync(&ctrl->ka_work);
760 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
762 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
764 struct nvme_command c = { };
767 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
768 c.identify.opcode = nvme_admin_identify;
769 c.identify.cns = NVME_ID_CNS_CTRL;
771 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
775 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
776 sizeof(struct nvme_id_ctrl));
782 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
783 u8 *eui64, u8 *nguid, uuid_t *uuid)
785 struct nvme_command c = { };
791 c.identify.opcode = nvme_admin_identify;
792 c.identify.nsid = cpu_to_le32(nsid);
793 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
795 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
799 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
800 NVME_IDENTIFY_DATA_SIZE);
804 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
805 struct nvme_ns_id_desc *cur = data + pos;
811 case NVME_NIDT_EUI64:
812 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
813 dev_warn(ctrl->device,
814 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
818 len = NVME_NIDT_EUI64_LEN;
819 memcpy(eui64, data + pos + sizeof(*cur), len);
821 case NVME_NIDT_NGUID:
822 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
823 dev_warn(ctrl->device,
824 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
828 len = NVME_NIDT_NGUID_LEN;
829 memcpy(nguid, data + pos + sizeof(*cur), len);
832 if (cur->nidl != NVME_NIDT_UUID_LEN) {
833 dev_warn(ctrl->device,
834 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
838 len = NVME_NIDT_UUID_LEN;
839 uuid_copy(uuid, data + pos + sizeof(*cur));
842 /* Skip unnkown types */
854 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
856 struct nvme_command c = { };
858 c.identify.opcode = nvme_admin_identify;
859 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
860 c.identify.nsid = cpu_to_le32(nsid);
861 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
864 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
867 struct nvme_id_ns *id;
868 struct nvme_command c = { };
871 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
872 c.identify.opcode = nvme_admin_identify;
873 c.identify.nsid = cpu_to_le32(nsid);
874 c.identify.cns = NVME_ID_CNS_NS;
876 id = kmalloc(sizeof(*id), GFP_KERNEL);
880 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
882 dev_warn(ctrl->device, "Identify namespace failed\n");
890 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
891 void *buffer, size_t buflen, u32 *result)
893 struct nvme_command c;
894 union nvme_result res;
897 memset(&c, 0, sizeof(c));
898 c.features.opcode = nvme_admin_set_features;
899 c.features.fid = cpu_to_le32(fid);
900 c.features.dword11 = cpu_to_le32(dword11);
902 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
903 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
904 if (ret >= 0 && result)
905 *result = le32_to_cpu(res.u32);
909 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
911 u32 q_count = (*count - 1) | ((*count - 1) << 16);
913 int status, nr_io_queues;
915 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
921 * Degraded controllers might return an error when setting the queue
922 * count. We still want to be able to bring them online and offer
923 * access to the admin queue, as that might be only way to fix them up.
926 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
929 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
930 *count = min(*count, nr_io_queues);
935 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
937 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
939 struct nvme_user_io io;
940 struct nvme_command c;
941 unsigned length, meta_len;
942 void __user *metadata;
944 if (copy_from_user(&io, uio, sizeof(io)))
952 case nvme_cmd_compare:
958 length = (io.nblocks + 1) << ns->lba_shift;
959 meta_len = (io.nblocks + 1) * ns->ms;
960 metadata = (void __user *)(uintptr_t)io.metadata;
965 } else if (meta_len) {
966 if ((io.metadata & 3) || !io.metadata)
970 memset(&c, 0, sizeof(c));
971 c.rw.opcode = io.opcode;
972 c.rw.flags = io.flags;
973 c.rw.nsid = cpu_to_le32(ns->ns_id);
974 c.rw.slba = cpu_to_le64(io.slba);
975 c.rw.length = cpu_to_le16(io.nblocks);
976 c.rw.control = cpu_to_le16(io.control);
977 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
978 c.rw.reftag = cpu_to_le32(io.reftag);
979 c.rw.apptag = cpu_to_le16(io.apptag);
980 c.rw.appmask = cpu_to_le16(io.appmask);
982 return nvme_submit_user_cmd(ns->queue, &c,
983 (void __user *)(uintptr_t)io.addr, length,
984 metadata, meta_len, io.slba, NULL, 0);
987 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
988 struct nvme_passthru_cmd __user *ucmd)
990 struct nvme_passthru_cmd cmd;
991 struct nvme_command c;
992 unsigned timeout = 0;
995 if (!capable(CAP_SYS_ADMIN))
997 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1002 memset(&c, 0, sizeof(c));
1003 c.common.opcode = cmd.opcode;
1004 c.common.flags = cmd.flags;
1005 c.common.nsid = cpu_to_le32(cmd.nsid);
1006 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1007 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1008 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1009 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1010 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1011 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1012 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1013 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1016 timeout = msecs_to_jiffies(cmd.timeout_ms);
1018 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1019 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1020 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1021 0, &cmd.result, timeout);
1023 if (put_user(cmd.result, &ucmd->result))
1030 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1031 unsigned int cmd, unsigned long arg)
1033 struct nvme_ns *ns = bdev->bd_disk->private_data;
1037 force_successful_syscall_return();
1039 case NVME_IOCTL_ADMIN_CMD:
1040 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1041 case NVME_IOCTL_IO_CMD:
1042 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1043 case NVME_IOCTL_SUBMIT_IO:
1044 return nvme_submit_io(ns, (void __user *)arg);
1048 return nvme_nvm_ioctl(ns, cmd, arg);
1050 if (is_sed_ioctl(cmd))
1051 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1052 (void __user *) arg);
1057 #ifdef CONFIG_COMPAT
1058 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1059 unsigned int cmd, unsigned long arg)
1061 return nvme_ioctl(bdev, mode, cmd, arg);
1064 #define nvme_compat_ioctl NULL
1067 static int nvme_open(struct block_device *bdev, fmode_t mode)
1069 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
1072 static void nvme_release(struct gendisk *disk, fmode_t mode)
1074 struct nvme_ns *ns = disk->private_data;
1076 module_put(ns->ctrl->ops->module);
1080 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1082 /* some standard values */
1083 geo->heads = 1 << 6;
1084 geo->sectors = 1 << 5;
1085 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1089 #ifdef CONFIG_BLK_DEV_INTEGRITY
1090 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1093 struct nvme_ns *ns = disk->private_data;
1094 u16 old_ms = ns->ms;
1097 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1098 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1100 /* PI implementation requires metadata equal t10 pi tuple size */
1101 if (ns->ms == sizeof(struct t10_pi_tuple))
1102 pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1104 if (blk_get_integrity(disk) &&
1105 (ns->pi_type != pi_type || ns->ms != old_ms ||
1106 bs != queue_logical_block_size(disk->queue) ||
1107 (ns->ms && ns->ext)))
1108 blk_integrity_unregister(disk);
1110 ns->pi_type = pi_type;
1113 static void nvme_init_integrity(struct nvme_ns *ns)
1115 struct blk_integrity integrity;
1117 memset(&integrity, 0, sizeof(integrity));
1118 switch (ns->pi_type) {
1119 case NVME_NS_DPS_PI_TYPE3:
1120 integrity.profile = &t10_pi_type3_crc;
1121 integrity.tag_size = sizeof(u16) + sizeof(u32);
1122 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1124 case NVME_NS_DPS_PI_TYPE1:
1125 case NVME_NS_DPS_PI_TYPE2:
1126 integrity.profile = &t10_pi_type1_crc;
1127 integrity.tag_size = sizeof(u16);
1128 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1131 integrity.profile = NULL;
1134 integrity.tuple_size = ns->ms;
1135 blk_integrity_register(ns->disk, &integrity);
1136 blk_queue_max_integrity_segments(ns->queue, 1);
1139 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
1143 static void nvme_init_integrity(struct nvme_ns *ns)
1146 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1148 static void nvme_set_chunk_size(struct nvme_ns *ns)
1150 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1151 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1154 static void nvme_config_discard(struct nvme_ns *ns)
1156 struct nvme_ctrl *ctrl = ns->ctrl;
1157 u32 logical_block_size = queue_logical_block_size(ns->queue);
1159 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1160 NVME_DSM_MAX_RANGES);
1162 if (ctrl->nr_streams && ns->sws && ns->sgs) {
1163 unsigned int sz = logical_block_size * ns->sws * ns->sgs;
1165 ns->queue->limits.discard_alignment = sz;
1166 ns->queue->limits.discard_granularity = sz;
1168 ns->queue->limits.discard_alignment = logical_block_size;
1169 ns->queue->limits.discard_granularity = logical_block_size;
1171 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
1172 blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
1173 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1175 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1176 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
1179 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1180 struct nvme_id_ns *id, u8 *eui64, u8 *nguid, uuid_t *uuid)
1182 if (ctrl->vs >= NVME_VS(1, 1, 0))
1183 memcpy(eui64, id->eui64, sizeof(id->eui64));
1184 if (ctrl->vs >= NVME_VS(1, 2, 0))
1185 memcpy(nguid, id->nguid, sizeof(id->nguid));
1186 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1187 /* Don't treat error as fatal we potentially
1188 * already have a NGUID or EUI-64
1190 if (nvme_identify_ns_descs(ctrl, nsid, eui64, nguid, uuid))
1191 dev_warn(ctrl->device,
1192 "%s: Identify Descriptors failed\n", __func__);
1196 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1198 struct nvme_ns *ns = disk->private_data;
1199 struct nvme_ctrl *ctrl = ns->ctrl;
1203 * If identify namespace failed, use default 512 byte block size so
1204 * block layer can use before failing read/write for 0 capacity.
1206 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1207 if (ns->lba_shift == 0)
1209 bs = 1 << ns->lba_shift;
1210 ns->noiob = le16_to_cpu(id->noiob);
1212 blk_mq_freeze_queue(disk->queue);
1214 if (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1215 nvme_prep_integrity(disk, id, bs);
1216 blk_queue_logical_block_size(ns->queue, bs);
1218 nvme_set_chunk_size(ns);
1219 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1220 nvme_init_integrity(ns);
1221 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1222 set_capacity(disk, 0);
1224 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1226 if (ctrl->oncs & NVME_CTRL_ONCS_DSM)
1227 nvme_config_discard(ns);
1228 blk_mq_unfreeze_queue(disk->queue);
1231 static int nvme_revalidate_disk(struct gendisk *disk)
1233 struct nvme_ns *ns = disk->private_data;
1234 struct nvme_ctrl *ctrl = ns->ctrl;
1235 struct nvme_id_ns *id;
1236 u8 eui64[8] = { 0 }, nguid[16] = { 0 };
1237 uuid_t uuid = uuid_null;
1240 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1241 set_capacity(disk, 0);
1245 id = nvme_identify_ns(ctrl, ns->ns_id);
1249 if (id->ncap == 0) {
1254 nvme_report_ns_ids(ctrl, ns->ns_id, id, eui64, nguid, &uuid);
1255 if (!uuid_equal(&ns->uuid, &uuid) ||
1256 memcmp(&ns->nguid, &nguid, sizeof(ns->nguid)) ||
1257 memcmp(&ns->eui, &eui64, sizeof(ns->eui))) {
1258 dev_err(ctrl->device,
1259 "identifiers changed for nsid %d\n", ns->ns_id);
1268 static char nvme_pr_type(enum pr_type type)
1271 case PR_WRITE_EXCLUSIVE:
1273 case PR_EXCLUSIVE_ACCESS:
1275 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1277 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1279 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1281 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1288 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1289 u64 key, u64 sa_key, u8 op)
1291 struct nvme_ns *ns = bdev->bd_disk->private_data;
1292 struct nvme_command c;
1293 u8 data[16] = { 0, };
1295 put_unaligned_le64(key, &data[0]);
1296 put_unaligned_le64(sa_key, &data[8]);
1298 memset(&c, 0, sizeof(c));
1299 c.common.opcode = op;
1300 c.common.nsid = cpu_to_le32(ns->ns_id);
1301 c.common.cdw10[0] = cpu_to_le32(cdw10);
1303 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1306 static int nvme_pr_register(struct block_device *bdev, u64 old,
1307 u64 new, unsigned flags)
1311 if (flags & ~PR_FL_IGNORE_KEY)
1314 cdw10 = old ? 2 : 0;
1315 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1316 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1317 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1320 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1321 enum pr_type type, unsigned flags)
1325 if (flags & ~PR_FL_IGNORE_KEY)
1328 cdw10 = nvme_pr_type(type) << 8;
1329 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1330 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1333 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1334 enum pr_type type, bool abort)
1336 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1337 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1340 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1342 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1343 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1346 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1348 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1349 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1352 static const struct pr_ops nvme_pr_ops = {
1353 .pr_register = nvme_pr_register,
1354 .pr_reserve = nvme_pr_reserve,
1355 .pr_release = nvme_pr_release,
1356 .pr_preempt = nvme_pr_preempt,
1357 .pr_clear = nvme_pr_clear,
1360 #ifdef CONFIG_BLK_SED_OPAL
1361 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1364 struct nvme_ctrl *ctrl = data;
1365 struct nvme_command cmd;
1367 memset(&cmd, 0, sizeof(cmd));
1369 cmd.common.opcode = nvme_admin_security_send;
1371 cmd.common.opcode = nvme_admin_security_recv;
1372 cmd.common.nsid = 0;
1373 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1374 cmd.common.cdw10[1] = cpu_to_le32(len);
1376 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1377 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1379 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1380 #endif /* CONFIG_BLK_SED_OPAL */
1382 static const struct block_device_operations nvme_fops = {
1383 .owner = THIS_MODULE,
1384 .ioctl = nvme_ioctl,
1385 .compat_ioctl = nvme_compat_ioctl,
1387 .release = nvme_release,
1388 .getgeo = nvme_getgeo,
1389 .revalidate_disk= nvme_revalidate_disk,
1390 .pr_ops = &nvme_pr_ops,
1393 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1395 unsigned long timeout =
1396 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1397 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1400 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1403 if ((csts & NVME_CSTS_RDY) == bit)
1407 if (fatal_signal_pending(current))
1409 if (time_after(jiffies, timeout)) {
1410 dev_err(ctrl->device,
1411 "Device not ready; aborting %s\n", enabled ?
1412 "initialisation" : "reset");
1421 * If the device has been passed off to us in an enabled state, just clear
1422 * the enabled bit. The spec says we should set the 'shutdown notification
1423 * bits', but doing so may cause the device to complete commands to the
1424 * admin queue ... and we don't know what memory that might be pointing at!
1426 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1430 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1431 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1433 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1437 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1438 msleep(NVME_QUIRK_DELAY_AMOUNT);
1440 return nvme_wait_ready(ctrl, cap, false);
1442 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1444 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1447 * Default to a 4K page size, with the intention to update this
1448 * path in the future to accomodate architectures with differing
1449 * kernel and IO page sizes.
1451 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1454 if (page_shift < dev_page_min) {
1455 dev_err(ctrl->device,
1456 "Minimum device page size %u too large for host (%u)\n",
1457 1 << dev_page_min, 1 << page_shift);
1461 ctrl->page_size = 1 << page_shift;
1463 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1464 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1465 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1466 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1467 ctrl->ctrl_config |= NVME_CC_ENABLE;
1469 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1472 return nvme_wait_ready(ctrl, cap, true);
1474 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1476 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1478 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1482 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1483 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1485 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1489 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1490 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1494 if (fatal_signal_pending(current))
1496 if (time_after(jiffies, timeout)) {
1497 dev_err(ctrl->device,
1498 "Device shutdown incomplete; abort shutdown\n");
1505 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1507 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1508 struct request_queue *q)
1512 if (ctrl->max_hw_sectors) {
1514 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1516 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1517 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1519 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1520 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1521 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1522 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1524 blk_queue_write_cache(q, vwc, vwc);
1527 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1532 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1535 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1536 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1539 dev_warn_once(ctrl->device,
1540 "could not set timestamp (%d)\n", ret);
1544 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1547 * APST (Autonomous Power State Transition) lets us program a
1548 * table of power state transitions that the controller will
1549 * perform automatically. We configure it with a simple
1550 * heuristic: we are willing to spend at most 2% of the time
1551 * transitioning between power states. Therefore, when running
1552 * in any given state, we will enter the next lower-power
1553 * non-operational state after waiting 50 * (enlat + exlat)
1554 * microseconds, as long as that state's exit latency is under
1555 * the requested maximum latency.
1557 * We will not autonomously enter any non-operational state for
1558 * which the total latency exceeds ps_max_latency_us. Users
1559 * can set ps_max_latency_us to zero to turn off APST.
1563 struct nvme_feat_auto_pst *table;
1569 * If APST isn't supported or if we haven't been initialized yet,
1570 * then don't do anything.
1575 if (ctrl->npss > 31) {
1576 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1580 table = kzalloc(sizeof(*table), GFP_KERNEL);
1584 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1585 /* Turn off APST. */
1587 dev_dbg(ctrl->device, "APST disabled\n");
1589 __le64 target = cpu_to_le64(0);
1593 * Walk through all states from lowest- to highest-power.
1594 * According to the spec, lower-numbered states use more
1595 * power. NPSS, despite the name, is the index of the
1596 * lowest-power state, not the number of states.
1598 for (state = (int)ctrl->npss; state >= 0; state--) {
1599 u64 total_latency_us, exit_latency_us, transition_ms;
1602 table->entries[state] = target;
1605 * Don't allow transitions to the deepest state
1606 * if it's quirked off.
1608 if (state == ctrl->npss &&
1609 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1613 * Is this state a useful non-operational state for
1614 * higher-power states to autonomously transition to?
1616 if (!(ctrl->psd[state].flags &
1617 NVME_PS_FLAGS_NON_OP_STATE))
1621 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1622 if (exit_latency_us > ctrl->ps_max_latency_us)
1627 le32_to_cpu(ctrl->psd[state].entry_lat);
1630 * This state is good. Use it as the APST idle
1631 * target for higher power states.
1633 transition_ms = total_latency_us + 19;
1634 do_div(transition_ms, 20);
1635 if (transition_ms > (1 << 24) - 1)
1636 transition_ms = (1 << 24) - 1;
1638 target = cpu_to_le64((state << 3) |
1639 (transition_ms << 8));
1644 if (total_latency_us > max_lat_us)
1645 max_lat_us = total_latency_us;
1651 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1653 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1654 max_ps, max_lat_us, (int)sizeof(*table), table);
1658 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1659 table, sizeof(*table), NULL);
1661 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1667 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1669 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1673 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1674 case PM_QOS_LATENCY_ANY:
1682 if (ctrl->ps_max_latency_us != latency) {
1683 ctrl->ps_max_latency_us = latency;
1684 nvme_configure_apst(ctrl);
1688 struct nvme_core_quirk_entry {
1690 * NVMe model and firmware strings are padded with spaces. For
1691 * simplicity, strings in the quirk table are padded with NULLs
1697 unsigned long quirks;
1700 static const struct nvme_core_quirk_entry core_quirks[] = {
1703 * This Toshiba device seems to die using any APST states. See:
1704 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1707 .mn = "THNSF5256GPUK TOSHIBA",
1708 .quirks = NVME_QUIRK_NO_APST,
1712 /* match is null-terminated but idstr is space-padded. */
1713 static bool string_matches(const char *idstr, const char *match, size_t len)
1720 matchlen = strlen(match);
1721 WARN_ON_ONCE(matchlen > len);
1723 if (memcmp(idstr, match, matchlen))
1726 for (; matchlen < len; matchlen++)
1727 if (idstr[matchlen] != ' ')
1733 static bool quirk_matches(const struct nvme_id_ctrl *id,
1734 const struct nvme_core_quirk_entry *q)
1736 return q->vid == le16_to_cpu(id->vid) &&
1737 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1738 string_matches(id->fr, q->fr, sizeof(id->fr));
1741 static void nvme_init_subnqn(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
1746 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1747 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
1748 strcpy(ctrl->subnqn, id->subnqn);
1752 if (ctrl->vs >= NVME_VS(1, 2, 1))
1753 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
1755 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
1756 off = snprintf(ctrl->subnqn, NVMF_NQN_SIZE,
1757 "nqn.2014.08.org.nvmexpress:%4x%4x",
1758 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
1759 memcpy(ctrl->subnqn + off, id->sn, sizeof(id->sn));
1760 off += sizeof(id->sn);
1761 memcpy(ctrl->subnqn + off, id->mn, sizeof(id->mn));
1762 off += sizeof(id->mn);
1763 memset(ctrl->subnqn + off, 0, sizeof(ctrl->subnqn) - off);
1767 * Initialize the cached copies of the Identify data and various controller
1768 * register in our nvme_ctrl structure. This should be called as soon as
1769 * the admin queue is fully up and running.
1771 int nvme_init_identify(struct nvme_ctrl *ctrl)
1773 struct nvme_id_ctrl *id;
1775 int ret, page_shift;
1777 bool prev_apst_enabled;
1779 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1781 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1785 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1787 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1790 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1792 if (ctrl->vs >= NVME_VS(1, 1, 0))
1793 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1795 ret = nvme_identify_ctrl(ctrl, &id);
1797 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1801 nvme_init_subnqn(ctrl, id);
1803 if (!ctrl->identified) {
1805 * Check for quirks. Quirk can depend on firmware version,
1806 * so, in principle, the set of quirks present can change
1807 * across a reset. As a possible future enhancement, we
1808 * could re-scan for quirks every time we reinitialize
1809 * the device, but we'd have to make sure that the driver
1810 * behaves intelligently if the quirks change.
1815 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1816 if (quirk_matches(id, &core_quirks[i]))
1817 ctrl->quirks |= core_quirks[i].quirks;
1821 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1822 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1823 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1826 ctrl->oacs = le16_to_cpu(id->oacs);
1827 ctrl->vid = le16_to_cpu(id->vid);
1828 ctrl->oncs = le16_to_cpup(&id->oncs);
1829 atomic_set(&ctrl->abort_limit, id->acl + 1);
1830 ctrl->vwc = id->vwc;
1831 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1832 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1833 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1834 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1836 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1838 max_hw_sectors = UINT_MAX;
1839 ctrl->max_hw_sectors =
1840 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1842 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1843 ctrl->sgls = le32_to_cpu(id->sgls);
1844 ctrl->kas = le16_to_cpu(id->kas);
1848 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
1850 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
1851 shutdown_timeout, 60);
1853 if (ctrl->shutdown_timeout != shutdown_timeout)
1854 dev_warn(ctrl->device,
1855 "Shutdown timeout set to %u seconds\n",
1856 ctrl->shutdown_timeout);
1858 ctrl->shutdown_timeout = shutdown_timeout;
1860 ctrl->npss = id->npss;
1861 ctrl->apsta = id->apsta;
1862 prev_apst_enabled = ctrl->apst_enabled;
1863 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1864 if (force_apst && id->apsta) {
1865 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1866 ctrl->apst_enabled = true;
1868 ctrl->apst_enabled = false;
1871 ctrl->apst_enabled = id->apsta;
1873 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1875 if (ctrl->ops->flags & NVME_F_FABRICS) {
1876 ctrl->icdoff = le16_to_cpu(id->icdoff);
1877 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1878 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1879 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1882 * In fabrics we need to verify the cntlid matches the
1885 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
1890 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1891 dev_err(ctrl->device,
1892 "keep-alive support is mandatory for fabrics\n");
1897 ctrl->cntlid = le16_to_cpu(id->cntlid);
1898 ctrl->hmpre = le32_to_cpu(id->hmpre);
1899 ctrl->hmmin = le32_to_cpu(id->hmmin);
1904 if (ctrl->apst_enabled && !prev_apst_enabled)
1905 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1906 else if (!ctrl->apst_enabled && prev_apst_enabled)
1907 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1909 ret = nvme_configure_apst(ctrl);
1913 ret = nvme_configure_timestamp(ctrl);
1917 ret = nvme_configure_directives(ctrl);
1921 ctrl->identified = true;
1929 EXPORT_SYMBOL_GPL(nvme_init_identify);
1931 static int nvme_dev_open(struct inode *inode, struct file *file)
1933 struct nvme_ctrl *ctrl;
1934 int instance = iminor(inode);
1937 spin_lock(&dev_list_lock);
1938 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1939 if (ctrl->instance != instance)
1942 if (!ctrl->admin_q) {
1946 if (!kref_get_unless_zero(&ctrl->kref))
1948 file->private_data = ctrl;
1952 spin_unlock(&dev_list_lock);
1957 static int nvme_dev_release(struct inode *inode, struct file *file)
1959 nvme_put_ctrl(file->private_data);
1963 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1968 mutex_lock(&ctrl->namespaces_mutex);
1969 if (list_empty(&ctrl->namespaces)) {
1974 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1975 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1976 dev_warn(ctrl->device,
1977 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1982 dev_warn(ctrl->device,
1983 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1984 kref_get(&ns->kref);
1985 mutex_unlock(&ctrl->namespaces_mutex);
1987 ret = nvme_user_cmd(ctrl, ns, argp);
1992 mutex_unlock(&ctrl->namespaces_mutex);
1996 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1999 struct nvme_ctrl *ctrl = file->private_data;
2000 void __user *argp = (void __user *)arg;
2003 case NVME_IOCTL_ADMIN_CMD:
2004 return nvme_user_cmd(ctrl, NULL, argp);
2005 case NVME_IOCTL_IO_CMD:
2006 return nvme_dev_user_cmd(ctrl, argp);
2007 case NVME_IOCTL_RESET:
2008 dev_warn(ctrl->device, "resetting controller\n");
2009 return nvme_reset_ctrl_sync(ctrl);
2010 case NVME_IOCTL_SUBSYS_RESET:
2011 return nvme_reset_subsystem(ctrl);
2012 case NVME_IOCTL_RESCAN:
2013 nvme_queue_scan(ctrl);
2020 static const struct file_operations nvme_dev_fops = {
2021 .owner = THIS_MODULE,
2022 .open = nvme_dev_open,
2023 .release = nvme_dev_release,
2024 .unlocked_ioctl = nvme_dev_ioctl,
2025 .compat_ioctl = nvme_dev_ioctl,
2028 static ssize_t nvme_sysfs_reset(struct device *dev,
2029 struct device_attribute *attr, const char *buf,
2032 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2035 ret = nvme_reset_ctrl_sync(ctrl);
2040 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2042 static ssize_t nvme_sysfs_rescan(struct device *dev,
2043 struct device_attribute *attr, const char *buf,
2046 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2048 nvme_queue_scan(ctrl);
2051 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2053 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2056 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2057 struct nvme_ctrl *ctrl = ns->ctrl;
2058 int serial_len = sizeof(ctrl->serial);
2059 int model_len = sizeof(ctrl->model);
2061 if (!uuid_is_null(&ns->uuid))
2062 return sprintf(buf, "uuid.%pU\n", &ns->uuid);
2064 if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2065 return sprintf(buf, "eui.%16phN\n", ns->nguid);
2067 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2068 return sprintf(buf, "eui.%8phN\n", ns->eui);
2070 while (serial_len > 0 && (ctrl->serial[serial_len - 1] == ' ' ||
2071 ctrl->serial[serial_len - 1] == '\0'))
2073 while (model_len > 0 && (ctrl->model[model_len - 1] == ' ' ||
2074 ctrl->model[model_len - 1] == '\0'))
2077 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
2078 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
2080 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2082 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2085 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2086 return sprintf(buf, "%pU\n", ns->nguid);
2088 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2090 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2093 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2095 /* For backward compatibility expose the NGUID to userspace if
2096 * we have no UUID set
2098 if (uuid_is_null(&ns->uuid)) {
2099 printk_ratelimited(KERN_WARNING
2100 "No UUID available providing old NGUID\n");
2101 return sprintf(buf, "%pU\n", ns->nguid);
2103 return sprintf(buf, "%pU\n", &ns->uuid);
2105 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2107 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2110 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2111 return sprintf(buf, "%8phd\n", ns->eui);
2113 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2115 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2118 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2119 return sprintf(buf, "%d\n", ns->ns_id);
2121 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2123 static struct attribute *nvme_ns_attrs[] = {
2124 &dev_attr_wwid.attr,
2125 &dev_attr_uuid.attr,
2126 &dev_attr_nguid.attr,
2128 &dev_attr_nsid.attr,
2132 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
2133 struct attribute *a, int n)
2135 struct device *dev = container_of(kobj, struct device, kobj);
2136 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
2138 if (a == &dev_attr_uuid.attr) {
2139 if (uuid_is_null(&ns->uuid) ||
2140 !memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2143 if (a == &dev_attr_nguid.attr) {
2144 if (!memchr_inv(ns->nguid, 0, sizeof(ns->nguid)))
2147 if (a == &dev_attr_eui.attr) {
2148 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
2154 static const struct attribute_group nvme_ns_attr_group = {
2155 .attrs = nvme_ns_attrs,
2156 .is_visible = nvme_ns_attrs_are_visible,
2159 #define nvme_show_str_function(field) \
2160 static ssize_t field##_show(struct device *dev, \
2161 struct device_attribute *attr, char *buf) \
2163 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2164 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
2166 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2168 #define nvme_show_int_function(field) \
2169 static ssize_t field##_show(struct device *dev, \
2170 struct device_attribute *attr, char *buf) \
2172 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2173 return sprintf(buf, "%d\n", ctrl->field); \
2175 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2177 nvme_show_str_function(model);
2178 nvme_show_str_function(serial);
2179 nvme_show_str_function(firmware_rev);
2180 nvme_show_int_function(cntlid);
2182 static ssize_t nvme_sysfs_delete(struct device *dev,
2183 struct device_attribute *attr, const char *buf,
2186 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2188 if (device_remove_file_self(dev, attr))
2189 ctrl->ops->delete_ctrl(ctrl);
2192 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2194 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2195 struct device_attribute *attr,
2198 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2200 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2202 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2204 static ssize_t nvme_sysfs_show_state(struct device *dev,
2205 struct device_attribute *attr,
2208 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2209 static const char *const state_name[] = {
2210 [NVME_CTRL_NEW] = "new",
2211 [NVME_CTRL_LIVE] = "live",
2212 [NVME_CTRL_RESETTING] = "resetting",
2213 [NVME_CTRL_RECONNECTING]= "reconnecting",
2214 [NVME_CTRL_DELETING] = "deleting",
2215 [NVME_CTRL_DEAD] = "dead",
2218 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2219 state_name[ctrl->state])
2220 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2222 return sprintf(buf, "unknown state\n");
2225 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2227 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2228 struct device_attribute *attr,
2231 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2233 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subnqn);
2235 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2237 static ssize_t nvme_sysfs_show_address(struct device *dev,
2238 struct device_attribute *attr,
2241 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2243 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2245 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2247 static struct attribute *nvme_dev_attrs[] = {
2248 &dev_attr_reset_controller.attr,
2249 &dev_attr_rescan_controller.attr,
2250 &dev_attr_model.attr,
2251 &dev_attr_serial.attr,
2252 &dev_attr_firmware_rev.attr,
2253 &dev_attr_cntlid.attr,
2254 &dev_attr_delete_controller.attr,
2255 &dev_attr_transport.attr,
2256 &dev_attr_subsysnqn.attr,
2257 &dev_attr_address.attr,
2258 &dev_attr_state.attr,
2262 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2263 struct attribute *a, int n)
2265 struct device *dev = container_of(kobj, struct device, kobj);
2266 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2268 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2270 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2276 static struct attribute_group nvme_dev_attrs_group = {
2277 .attrs = nvme_dev_attrs,
2278 .is_visible = nvme_dev_attrs_are_visible,
2281 static const struct attribute_group *nvme_dev_attr_groups[] = {
2282 &nvme_dev_attrs_group,
2286 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2288 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2289 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2291 return nsa->ns_id - nsb->ns_id;
2294 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2296 struct nvme_ns *ns, *ret = NULL;
2298 mutex_lock(&ctrl->namespaces_mutex);
2299 list_for_each_entry(ns, &ctrl->namespaces, list) {
2300 if (ns->ns_id == nsid) {
2301 kref_get(&ns->kref);
2305 if (ns->ns_id > nsid)
2308 mutex_unlock(&ctrl->namespaces_mutex);
2312 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2314 struct streams_directive_params s;
2317 if (!ctrl->nr_streams)
2320 ret = nvme_get_stream_params(ctrl, &s, ns->ns_id);
2324 ns->sws = le32_to_cpu(s.sws);
2325 ns->sgs = le16_to_cpu(s.sgs);
2328 unsigned int bs = 1 << ns->lba_shift;
2330 blk_queue_io_min(ns->queue, bs * ns->sws);
2332 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2338 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2341 struct gendisk *disk;
2342 struct nvme_id_ns *id;
2343 char disk_name[DISK_NAME_LEN];
2344 int node = dev_to_node(ctrl->dev);
2346 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2350 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2351 if (ns->instance < 0)
2354 ns->queue = blk_mq_init_queue(ctrl->tagset);
2355 if (IS_ERR(ns->queue))
2356 goto out_release_instance;
2357 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2358 ns->queue->queuedata = ns;
2361 kref_init(&ns->kref);
2363 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2365 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2366 nvme_set_queue_limits(ctrl, ns->queue);
2367 nvme_setup_streams_ns(ctrl, ns);
2369 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2371 id = nvme_identify_ns(ctrl, nsid);
2373 goto out_free_queue;
2378 nvme_report_ns_ids(ctrl, ns->ns_id, id, ns->eui, ns->nguid, &ns->uuid);
2380 if (nvme_nvm_ns_supported(ns, id) &&
2381 nvme_nvm_register(ns, disk_name, node)) {
2382 dev_warn(ctrl->device, "%s: LightNVM init failure\n", __func__);
2386 disk = alloc_disk_node(0, node);
2390 disk->fops = &nvme_fops;
2391 disk->private_data = ns;
2392 disk->queue = ns->queue;
2393 disk->flags = GENHD_FL_EXT_DEVT;
2394 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2397 __nvme_revalidate_disk(disk, id);
2399 mutex_lock(&ctrl->namespaces_mutex);
2400 list_add_tail(&ns->list, &ctrl->namespaces);
2401 mutex_unlock(&ctrl->namespaces_mutex);
2403 kref_get(&ctrl->kref);
2407 device_add_disk(ctrl->device, ns->disk);
2408 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2409 &nvme_ns_attr_group))
2410 pr_warn("%s: failed to create sysfs group for identification\n",
2411 ns->disk->disk_name);
2412 if (ns->ndev && nvme_nvm_register_sysfs(ns))
2413 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2414 ns->disk->disk_name);
2419 blk_cleanup_queue(ns->queue);
2420 out_release_instance:
2421 ida_simple_remove(&ctrl->ns_ida, ns->instance);
2426 static void nvme_ns_remove(struct nvme_ns *ns)
2428 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2431 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2432 if (blk_get_integrity(ns->disk))
2433 blk_integrity_unregister(ns->disk);
2434 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2435 &nvme_ns_attr_group);
2437 nvme_nvm_unregister_sysfs(ns);
2438 del_gendisk(ns->disk);
2439 blk_cleanup_queue(ns->queue);
2442 mutex_lock(&ns->ctrl->namespaces_mutex);
2443 list_del_init(&ns->list);
2444 mutex_unlock(&ns->ctrl->namespaces_mutex);
2449 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2453 ns = nvme_find_get_ns(ctrl, nsid);
2455 if (ns->disk && revalidate_disk(ns->disk))
2459 nvme_alloc_ns(ctrl, nsid);
2462 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2465 struct nvme_ns *ns, *next;
2467 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2468 if (ns->ns_id > nsid)
2473 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2477 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2480 ns_list = kzalloc(0x1000, GFP_KERNEL);
2484 for (i = 0; i < num_lists; i++) {
2485 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2489 for (j = 0; j < min(nn, 1024U); j++) {
2490 nsid = le32_to_cpu(ns_list[j]);
2494 nvme_validate_ns(ctrl, nsid);
2496 while (++prev < nsid) {
2497 ns = nvme_find_get_ns(ctrl, prev);
2507 nvme_remove_invalid_namespaces(ctrl, prev);
2513 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2517 for (i = 1; i <= nn; i++)
2518 nvme_validate_ns(ctrl, i);
2520 nvme_remove_invalid_namespaces(ctrl, nn);
2523 static void nvme_scan_work(struct work_struct *work)
2525 struct nvme_ctrl *ctrl =
2526 container_of(work, struct nvme_ctrl, scan_work);
2527 struct nvme_id_ctrl *id;
2530 if (ctrl->state != NVME_CTRL_LIVE)
2533 if (nvme_identify_ctrl(ctrl, &id))
2536 nn = le32_to_cpu(id->nn);
2537 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2538 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2539 if (!nvme_scan_ns_list(ctrl, nn))
2542 nvme_scan_ns_sequential(ctrl, nn);
2544 mutex_lock(&ctrl->namespaces_mutex);
2545 list_sort(NULL, &ctrl->namespaces, ns_cmp);
2546 mutex_unlock(&ctrl->namespaces_mutex);
2550 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2553 * Do not queue new scan work when a controller is reset during
2556 if (ctrl->state == NVME_CTRL_LIVE)
2557 queue_work(nvme_wq, &ctrl->scan_work);
2559 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2562 * This function iterates the namespace list unlocked to allow recovery from
2563 * controller failure. It is up to the caller to ensure the namespace list is
2564 * not modified by scan work while this function is executing.
2566 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2568 struct nvme_ns *ns, *next;
2571 * The dead states indicates the controller was not gracefully
2572 * disconnected. In that case, we won't be able to flush any data while
2573 * removing the namespaces' disks; fail all the queues now to avoid
2574 * potentially having to clean up the failed sync later.
2576 if (ctrl->state == NVME_CTRL_DEAD)
2577 nvme_kill_queues(ctrl);
2579 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2582 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2584 static void nvme_async_event_work(struct work_struct *work)
2586 struct nvme_ctrl *ctrl =
2587 container_of(work, struct nvme_ctrl, async_event_work);
2589 spin_lock_irq(&ctrl->lock);
2590 while (ctrl->event_limit > 0) {
2591 int aer_idx = --ctrl->event_limit;
2593 spin_unlock_irq(&ctrl->lock);
2594 ctrl->ops->submit_async_event(ctrl, aer_idx);
2595 spin_lock_irq(&ctrl->lock);
2597 spin_unlock_irq(&ctrl->lock);
2600 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
2605 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
2611 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
2614 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
2616 struct nvme_command c = { };
2617 struct nvme_fw_slot_info_log *log;
2619 log = kmalloc(sizeof(*log), GFP_KERNEL);
2623 c.common.opcode = nvme_admin_get_log_page;
2624 c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2625 c.common.cdw10[0] = nvme_get_log_dw10(NVME_LOG_FW_SLOT, sizeof(*log));
2627 if (!nvme_submit_sync_cmd(ctrl->admin_q, &c, log, sizeof(*log)))
2628 dev_warn(ctrl->device,
2629 "Get FW SLOT INFO log error\n");
2633 static void nvme_fw_act_work(struct work_struct *work)
2635 struct nvme_ctrl *ctrl = container_of(work,
2636 struct nvme_ctrl, fw_act_work);
2637 unsigned long fw_act_timeout;
2640 fw_act_timeout = jiffies +
2641 msecs_to_jiffies(ctrl->mtfa * 100);
2643 fw_act_timeout = jiffies +
2644 msecs_to_jiffies(admin_timeout * 1000);
2646 nvme_stop_queues(ctrl);
2647 while (nvme_ctrl_pp_status(ctrl)) {
2648 if (time_after(jiffies, fw_act_timeout)) {
2649 dev_warn(ctrl->device,
2650 "Fw activation timeout, reset controller\n");
2651 nvme_reset_ctrl(ctrl);
2657 if (ctrl->state != NVME_CTRL_LIVE)
2660 nvme_start_queues(ctrl);
2661 /* read FW slot informationi to clear the AER*/
2662 nvme_get_fw_slot_info(ctrl);
2665 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2666 union nvme_result *res)
2668 u32 result = le32_to_cpu(res->u32);
2671 switch (le16_to_cpu(status) >> 1) {
2672 case NVME_SC_SUCCESS:
2675 case NVME_SC_ABORT_REQ:
2676 ++ctrl->event_limit;
2677 queue_work(nvme_wq, &ctrl->async_event_work);
2686 switch (result & 0xff07) {
2687 case NVME_AER_NOTICE_NS_CHANGED:
2688 dev_info(ctrl->device, "rescanning\n");
2689 nvme_queue_scan(ctrl);
2691 case NVME_AER_NOTICE_FW_ACT_STARTING:
2692 schedule_work(&ctrl->fw_act_work);
2695 dev_warn(ctrl->device, "async event result %08x\n", result);
2698 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2700 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2702 ctrl->event_limit = NVME_NR_AERS;
2703 queue_work(nvme_wq, &ctrl->async_event_work);
2705 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2707 static DEFINE_IDA(nvme_instance_ida);
2709 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2711 int instance, error;
2714 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2717 spin_lock(&dev_list_lock);
2718 error = ida_get_new(&nvme_instance_ida, &instance);
2719 spin_unlock(&dev_list_lock);
2720 } while (error == -EAGAIN);
2725 ctrl->instance = instance;
2729 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2731 spin_lock(&dev_list_lock);
2732 ida_remove(&nvme_instance_ida, ctrl->instance);
2733 spin_unlock(&dev_list_lock);
2736 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
2738 nvme_stop_keep_alive(ctrl);
2739 flush_work(&ctrl->async_event_work);
2740 flush_work(&ctrl->scan_work);
2741 cancel_work_sync(&ctrl->fw_act_work);
2743 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
2745 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
2748 nvme_start_keep_alive(ctrl);
2750 if (ctrl->queue_count > 1) {
2751 nvme_queue_scan(ctrl);
2752 nvme_queue_async_events(ctrl);
2753 nvme_start_queues(ctrl);
2756 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
2758 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2760 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2762 spin_lock(&dev_list_lock);
2763 list_del(&ctrl->node);
2764 spin_unlock(&dev_list_lock);
2766 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2768 static void nvme_free_ctrl(struct kref *kref)
2770 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2772 put_device(ctrl->device);
2773 nvme_release_instance(ctrl);
2774 ida_destroy(&ctrl->ns_ida);
2776 ctrl->ops->free_ctrl(ctrl);
2779 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2781 kref_put(&ctrl->kref, nvme_free_ctrl);
2783 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2786 * Initialize a NVMe controller structures. This needs to be called during
2787 * earliest initialization so that we have the initialized structured around
2790 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2791 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2795 ctrl->state = NVME_CTRL_NEW;
2796 spin_lock_init(&ctrl->lock);
2797 INIT_LIST_HEAD(&ctrl->namespaces);
2798 mutex_init(&ctrl->namespaces_mutex);
2799 kref_init(&ctrl->kref);
2802 ctrl->quirks = quirks;
2803 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2804 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2805 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
2807 ret = nvme_set_instance(ctrl);
2811 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2812 MKDEV(nvme_char_major, ctrl->instance),
2813 ctrl, nvme_dev_attr_groups,
2814 "nvme%d", ctrl->instance);
2815 if (IS_ERR(ctrl->device)) {
2816 ret = PTR_ERR(ctrl->device);
2817 goto out_release_instance;
2819 get_device(ctrl->device);
2820 ida_init(&ctrl->ns_ida);
2822 spin_lock(&dev_list_lock);
2823 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2824 spin_unlock(&dev_list_lock);
2827 * Initialize latency tolerance controls. The sysfs files won't
2828 * be visible to userspace unless the device actually supports APST.
2830 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2831 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2832 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2835 out_release_instance:
2836 nvme_release_instance(ctrl);
2840 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2843 * nvme_kill_queues(): Ends all namespace queues
2844 * @ctrl: the dead controller that needs to end
2846 * Call this function when the driver determines it is unable to get the
2847 * controller in a state capable of servicing IO.
2849 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2853 mutex_lock(&ctrl->namespaces_mutex);
2855 /* Forcibly unquiesce queues to avoid blocking dispatch */
2857 blk_mq_unquiesce_queue(ctrl->admin_q);
2859 list_for_each_entry(ns, &ctrl->namespaces, list) {
2861 * Revalidating a dead namespace sets capacity to 0. This will
2862 * end buffered writers dirtying pages that can't be synced.
2864 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2866 revalidate_disk(ns->disk);
2867 blk_set_queue_dying(ns->queue);
2869 /* Forcibly unquiesce queues to avoid blocking dispatch */
2870 blk_mq_unquiesce_queue(ns->queue);
2872 mutex_unlock(&ctrl->namespaces_mutex);
2874 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2876 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2880 mutex_lock(&ctrl->namespaces_mutex);
2881 list_for_each_entry(ns, &ctrl->namespaces, list)
2882 blk_mq_unfreeze_queue(ns->queue);
2883 mutex_unlock(&ctrl->namespaces_mutex);
2885 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2887 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2891 mutex_lock(&ctrl->namespaces_mutex);
2892 list_for_each_entry(ns, &ctrl->namespaces, list) {
2893 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2897 mutex_unlock(&ctrl->namespaces_mutex);
2899 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2901 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2905 mutex_lock(&ctrl->namespaces_mutex);
2906 list_for_each_entry(ns, &ctrl->namespaces, list)
2907 blk_mq_freeze_queue_wait(ns->queue);
2908 mutex_unlock(&ctrl->namespaces_mutex);
2910 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2912 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2916 mutex_lock(&ctrl->namespaces_mutex);
2917 list_for_each_entry(ns, &ctrl->namespaces, list)
2918 blk_freeze_queue_start(ns->queue);
2919 mutex_unlock(&ctrl->namespaces_mutex);
2921 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2923 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2927 mutex_lock(&ctrl->namespaces_mutex);
2928 list_for_each_entry(ns, &ctrl->namespaces, list)
2929 blk_mq_quiesce_queue(ns->queue);
2930 mutex_unlock(&ctrl->namespaces_mutex);
2932 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2934 void nvme_start_queues(struct nvme_ctrl *ctrl)
2938 mutex_lock(&ctrl->namespaces_mutex);
2939 list_for_each_entry(ns, &ctrl->namespaces, list)
2940 blk_mq_unquiesce_queue(ns->queue);
2941 mutex_unlock(&ctrl->namespaces_mutex);
2943 EXPORT_SYMBOL_GPL(nvme_start_queues);
2945 int __init nvme_core_init(void)
2949 nvme_wq = alloc_workqueue("nvme-wq",
2950 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
2954 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2958 else if (result > 0)
2959 nvme_char_major = result;
2961 nvme_class = class_create(THIS_MODULE, "nvme");
2962 if (IS_ERR(nvme_class)) {
2963 result = PTR_ERR(nvme_class);
2964 goto unregister_chrdev;
2970 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2972 destroy_workqueue(nvme_wq);
2976 void nvme_core_exit(void)
2978 class_destroy(nvme_class);
2979 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2980 destroy_workqueue(nvme_wq);
2983 MODULE_LICENSE("GPL");
2984 MODULE_VERSION("1.0");
2985 module_init(nvme_core_init);
2986 module_exit(nvme_core_exit);