Commit | Line | Data |
---|---|---|
21d34711 CH |
1 | /* |
2 | * NVM Express device driver | |
3 | * Copyright (c) 2011-2014, Intel Corporation. | |
4 | * | |
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. | |
8 | * | |
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 | |
12 | * more details. | |
13 | */ | |
14 | ||
15 | #include <linux/blkdev.h> | |
16 | #include <linux/blk-mq.h> | |
5fd4ce1b | 17 | #include <linux/delay.h> |
21d34711 | 18 | #include <linux/errno.h> |
1673f1f0 | 19 | #include <linux/hdreg.h> |
21d34711 | 20 | #include <linux/kernel.h> |
5bae7f73 CH |
21 | #include <linux/module.h> |
22 | #include <linux/list_sort.h> | |
21d34711 CH |
23 | #include <linux/slab.h> |
24 | #include <linux/types.h> | |
1673f1f0 CH |
25 | #include <linux/pr.h> |
26 | #include <linux/ptrace.h> | |
27 | #include <linux/nvme_ioctl.h> | |
28 | #include <linux/t10-pi.h> | |
c5552fde | 29 | #include <linux/pm_qos.h> |
1673f1f0 CH |
30 | #include <scsi/sg.h> |
31 | #include <asm/unaligned.h> | |
21d34711 CH |
32 | |
33 | #include "nvme.h" | |
038bd4cb | 34 | #include "fabrics.h" |
21d34711 | 35 | |
f3ca80fc CH |
36 | #define NVME_MINORS (1U << MINORBITS) |
37 | ||
ba0ba7d3 ML |
38 | unsigned char admin_timeout = 60; |
39 | module_param(admin_timeout, byte, 0644); | |
40 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); | |
576d55d6 | 41 | EXPORT_SYMBOL_GPL(admin_timeout); |
ba0ba7d3 ML |
42 | |
43 | unsigned char nvme_io_timeout = 30; | |
44 | module_param_named(io_timeout, nvme_io_timeout, byte, 0644); | |
45 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); | |
576d55d6 | 46 | EXPORT_SYMBOL_GPL(nvme_io_timeout); |
ba0ba7d3 ML |
47 | |
48 | unsigned char shutdown_timeout = 5; | |
49 | module_param(shutdown_timeout, byte, 0644); | |
50 | MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); | |
51 | ||
44e44b29 CH |
52 | static u8 nvme_max_retries = 5; |
53 | module_param_named(max_retries, nvme_max_retries, byte, 0644); | |
f80ec966 | 54 | MODULE_PARM_DESC(max_retries, "max number of retries a command may have"); |
5bae7f73 | 55 | |
f3ca80fc CH |
56 | static int nvme_char_major; |
57 | module_param(nvme_char_major, int, 0); | |
58 | ||
c5552fde AL |
59 | static unsigned long default_ps_max_latency_us = 25000; |
60 | module_param(default_ps_max_latency_us, ulong, 0644); | |
61 | MODULE_PARM_DESC(default_ps_max_latency_us, | |
62 | "max power saving latency for new devices; use PM QOS to change per device"); | |
63 | ||
c35e30b4 AL |
64 | static bool force_apst; |
65 | module_param(force_apst, bool, 0644); | |
66 | MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off"); | |
67 | ||
f3ca80fc | 68 | static LIST_HEAD(nvme_ctrl_list); |
9f2482b9 | 69 | static DEFINE_SPINLOCK(dev_list_lock); |
1673f1f0 | 70 | |
f3ca80fc CH |
71 | static struct class *nvme_class; |
72 | ||
65ba6b54 | 73 | static int nvme_error_status(struct request *req) |
27fa9bc5 CH |
74 | { |
75 | switch (nvme_req(req)->status & 0x7ff) { | |
76 | case NVME_SC_SUCCESS: | |
77 | return 0; | |
78 | case NVME_SC_CAP_EXCEEDED: | |
79 | return -ENOSPC; | |
80 | default: | |
81 | return -EIO; | |
e02ab023 JG |
82 | |
83 | /* | |
84 | * XXX: these errors are a nasty side-band protocol to | |
85 | * drivers/md/dm-mpath.c:noretry_error() that aren't documented | |
86 | * anywhere.. | |
87 | */ | |
88 | case NVME_SC_CMD_SEQ_ERROR: | |
89 | return -EILSEQ; | |
90 | case NVME_SC_ONCS_NOT_SUPPORTED: | |
91 | return -EOPNOTSUPP; | |
92 | case NVME_SC_WRITE_FAULT: | |
93 | case NVME_SC_READ_ERROR: | |
94 | case NVME_SC_UNWRITTEN_BLOCK: | |
95 | return -ENODATA; | |
27fa9bc5 CH |
96 | } |
97 | } | |
27fa9bc5 | 98 | |
f6324b1b | 99 | static inline bool nvme_req_needs_retry(struct request *req) |
77f02a7a | 100 | { |
f6324b1b CH |
101 | if (blk_noretry_request(req)) |
102 | return false; | |
27fa9bc5 | 103 | if (nvme_req(req)->status & NVME_SC_DNR) |
f6324b1b CH |
104 | return false; |
105 | if (jiffies - req->start_time >= req->timeout) | |
106 | return false; | |
44e44b29 | 107 | if (nvme_req(req)->retries >= nvme_max_retries) |
f6324b1b CH |
108 | return false; |
109 | return true; | |
77f02a7a CH |
110 | } |
111 | ||
112 | void nvme_complete_rq(struct request *req) | |
113 | { | |
27fa9bc5 CH |
114 | if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) { |
115 | nvme_req(req)->retries++; | |
116 | blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q)); | |
117 | return; | |
77f02a7a CH |
118 | } |
119 | ||
27fa9bc5 | 120 | blk_mq_end_request(req, nvme_error_status(req)); |
77f02a7a CH |
121 | } |
122 | EXPORT_SYMBOL_GPL(nvme_complete_rq); | |
123 | ||
c55a2fd4 ML |
124 | void nvme_cancel_request(struct request *req, void *data, bool reserved) |
125 | { | |
126 | int status; | |
127 | ||
128 | if (!blk_mq_request_started(req)) | |
129 | return; | |
130 | ||
131 | dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device, | |
132 | "Cancelling I/O %d", req->tag); | |
133 | ||
134 | status = NVME_SC_ABORT_REQ; | |
135 | if (blk_queue_dying(req->q)) | |
136 | status |= NVME_SC_DNR; | |
27fa9bc5 | 137 | nvme_req(req)->status = status; |
08e0029a | 138 | blk_mq_complete_request(req); |
27fa9bc5 | 139 | |
c55a2fd4 ML |
140 | } |
141 | EXPORT_SYMBOL_GPL(nvme_cancel_request); | |
142 | ||
bb8d261e CH |
143 | bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, |
144 | enum nvme_ctrl_state new_state) | |
145 | { | |
f6b6a28e | 146 | enum nvme_ctrl_state old_state; |
bb8d261e CH |
147 | bool changed = false; |
148 | ||
149 | spin_lock_irq(&ctrl->lock); | |
f6b6a28e GKB |
150 | |
151 | old_state = ctrl->state; | |
bb8d261e CH |
152 | switch (new_state) { |
153 | case NVME_CTRL_LIVE: | |
154 | switch (old_state) { | |
7d2e8008 | 155 | case NVME_CTRL_NEW: |
bb8d261e | 156 | case NVME_CTRL_RESETTING: |
def61eca | 157 | case NVME_CTRL_RECONNECTING: |
bb8d261e CH |
158 | changed = true; |
159 | /* FALLTHRU */ | |
160 | default: | |
161 | break; | |
162 | } | |
163 | break; | |
164 | case NVME_CTRL_RESETTING: | |
165 | switch (old_state) { | |
166 | case NVME_CTRL_NEW: | |
def61eca CH |
167 | case NVME_CTRL_LIVE: |
168 | case NVME_CTRL_RECONNECTING: | |
169 | changed = true; | |
170 | /* FALLTHRU */ | |
171 | default: | |
172 | break; | |
173 | } | |
174 | break; | |
175 | case NVME_CTRL_RECONNECTING: | |
176 | switch (old_state) { | |
bb8d261e CH |
177 | case NVME_CTRL_LIVE: |
178 | changed = true; | |
179 | /* FALLTHRU */ | |
180 | default: | |
181 | break; | |
182 | } | |
183 | break; | |
184 | case NVME_CTRL_DELETING: | |
185 | switch (old_state) { | |
186 | case NVME_CTRL_LIVE: | |
187 | case NVME_CTRL_RESETTING: | |
def61eca | 188 | case NVME_CTRL_RECONNECTING: |
bb8d261e CH |
189 | changed = true; |
190 | /* FALLTHRU */ | |
191 | default: | |
192 | break; | |
193 | } | |
194 | break; | |
0ff9d4e1 KB |
195 | case NVME_CTRL_DEAD: |
196 | switch (old_state) { | |
197 | case NVME_CTRL_DELETING: | |
198 | changed = true; | |
199 | /* FALLTHRU */ | |
200 | default: | |
201 | break; | |
202 | } | |
203 | break; | |
bb8d261e CH |
204 | default: |
205 | break; | |
206 | } | |
bb8d261e CH |
207 | |
208 | if (changed) | |
209 | ctrl->state = new_state; | |
210 | ||
f6b6a28e GKB |
211 | spin_unlock_irq(&ctrl->lock); |
212 | ||
bb8d261e CH |
213 | return changed; |
214 | } | |
215 | EXPORT_SYMBOL_GPL(nvme_change_ctrl_state); | |
216 | ||
1673f1f0 CH |
217 | static void nvme_free_ns(struct kref *kref) |
218 | { | |
219 | struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref); | |
220 | ||
b0b4e09c MB |
221 | if (ns->ndev) |
222 | nvme_nvm_unregister(ns); | |
1673f1f0 | 223 | |
b0b4e09c MB |
224 | if (ns->disk) { |
225 | spin_lock(&dev_list_lock); | |
226 | ns->disk->private_data = NULL; | |
227 | spin_unlock(&dev_list_lock); | |
228 | } | |
1673f1f0 | 229 | |
1673f1f0 | 230 | put_disk(ns->disk); |
075790eb KB |
231 | ida_simple_remove(&ns->ctrl->ns_ida, ns->instance); |
232 | nvme_put_ctrl(ns->ctrl); | |
1673f1f0 CH |
233 | kfree(ns); |
234 | } | |
235 | ||
5bae7f73 | 236 | static void nvme_put_ns(struct nvme_ns *ns) |
1673f1f0 CH |
237 | { |
238 | kref_put(&ns->kref, nvme_free_ns); | |
239 | } | |
240 | ||
241 | static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk) | |
242 | { | |
243 | struct nvme_ns *ns; | |
244 | ||
245 | spin_lock(&dev_list_lock); | |
246 | ns = disk->private_data; | |
e439bb12 SG |
247 | if (ns) { |
248 | if (!kref_get_unless_zero(&ns->kref)) | |
249 | goto fail; | |
250 | if (!try_module_get(ns->ctrl->ops->module)) | |
251 | goto fail_put_ns; | |
252 | } | |
1673f1f0 CH |
253 | spin_unlock(&dev_list_lock); |
254 | ||
255 | return ns; | |
e439bb12 SG |
256 | |
257 | fail_put_ns: | |
258 | kref_put(&ns->kref, nvme_free_ns); | |
259 | fail: | |
260 | spin_unlock(&dev_list_lock); | |
261 | return NULL; | |
1673f1f0 CH |
262 | } |
263 | ||
4160982e | 264 | struct request *nvme_alloc_request(struct request_queue *q, |
eb71f435 | 265 | struct nvme_command *cmd, unsigned int flags, int qid) |
21d34711 | 266 | { |
aebf526b | 267 | unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; |
21d34711 | 268 | struct request *req; |
21d34711 | 269 | |
eb71f435 | 270 | if (qid == NVME_QID_ANY) { |
aebf526b | 271 | req = blk_mq_alloc_request(q, op, flags); |
eb71f435 | 272 | } else { |
aebf526b | 273 | req = blk_mq_alloc_request_hctx(q, op, flags, |
eb71f435 CH |
274 | qid ? qid - 1 : 0); |
275 | } | |
21d34711 | 276 | if (IS_ERR(req)) |
4160982e | 277 | return req; |
21d34711 | 278 | |
21d34711 | 279 | req->cmd_flags |= REQ_FAILFAST_DRIVER; |
d49187e9 | 280 | nvme_req(req)->cmd = cmd; |
21d34711 | 281 | |
4160982e CH |
282 | return req; |
283 | } | |
576d55d6 | 284 | EXPORT_SYMBOL_GPL(nvme_alloc_request); |
4160982e | 285 | |
8093f7ca ML |
286 | static inline void nvme_setup_flush(struct nvme_ns *ns, |
287 | struct nvme_command *cmnd) | |
288 | { | |
289 | memset(cmnd, 0, sizeof(*cmnd)); | |
290 | cmnd->common.opcode = nvme_cmd_flush; | |
291 | cmnd->common.nsid = cpu_to_le32(ns->ns_id); | |
292 | } | |
293 | ||
294 | static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req, | |
295 | struct nvme_command *cmnd) | |
296 | { | |
b35ba01e | 297 | unsigned short segments = blk_rq_nr_discard_segments(req), n = 0; |
8093f7ca | 298 | struct nvme_dsm_range *range; |
b35ba01e | 299 | struct bio *bio; |
8093f7ca | 300 | |
b35ba01e | 301 | range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC); |
8093f7ca ML |
302 | if (!range) |
303 | return BLK_MQ_RQ_QUEUE_BUSY; | |
304 | ||
b35ba01e CH |
305 | __rq_for_each_bio(bio, req) { |
306 | u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector); | |
307 | u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift; | |
308 | ||
309 | range[n].cattr = cpu_to_le32(0); | |
310 | range[n].nlb = cpu_to_le32(nlb); | |
311 | range[n].slba = cpu_to_le64(slba); | |
312 | n++; | |
313 | } | |
314 | ||
315 | if (WARN_ON_ONCE(n != segments)) { | |
316 | kfree(range); | |
317 | return BLK_MQ_RQ_QUEUE_ERROR; | |
318 | } | |
8093f7ca ML |
319 | |
320 | memset(cmnd, 0, sizeof(*cmnd)); | |
321 | cmnd->dsm.opcode = nvme_cmd_dsm; | |
322 | cmnd->dsm.nsid = cpu_to_le32(ns->ns_id); | |
f1dd03a8 | 323 | cmnd->dsm.nr = cpu_to_le32(segments - 1); |
8093f7ca ML |
324 | cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); |
325 | ||
f9d03f96 CH |
326 | req->special_vec.bv_page = virt_to_page(range); |
327 | req->special_vec.bv_offset = offset_in_page(range); | |
b35ba01e | 328 | req->special_vec.bv_len = sizeof(*range) * segments; |
f9d03f96 | 329 | req->rq_flags |= RQF_SPECIAL_PAYLOAD; |
8093f7ca | 330 | |
bac0000a | 331 | return BLK_MQ_RQ_QUEUE_OK; |
8093f7ca | 332 | } |
8093f7ca | 333 | |
8093f7ca ML |
334 | static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req, |
335 | struct nvme_command *cmnd) | |
336 | { | |
337 | u16 control = 0; | |
338 | u32 dsmgmt = 0; | |
339 | ||
340 | if (req->cmd_flags & REQ_FUA) | |
341 | control |= NVME_RW_FUA; | |
342 | if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) | |
343 | control |= NVME_RW_LR; | |
344 | ||
345 | if (req->cmd_flags & REQ_RAHEAD) | |
346 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; | |
347 | ||
348 | memset(cmnd, 0, sizeof(*cmnd)); | |
349 | cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); | |
8093f7ca ML |
350 | cmnd->rw.nsid = cpu_to_le32(ns->ns_id); |
351 | cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
352 | cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); | |
353 | ||
354 | if (ns->ms) { | |
355 | switch (ns->pi_type) { | |
356 | case NVME_NS_DPS_PI_TYPE3: | |
357 | control |= NVME_RW_PRINFO_PRCHK_GUARD; | |
358 | break; | |
359 | case NVME_NS_DPS_PI_TYPE1: | |
360 | case NVME_NS_DPS_PI_TYPE2: | |
361 | control |= NVME_RW_PRINFO_PRCHK_GUARD | | |
362 | NVME_RW_PRINFO_PRCHK_REF; | |
363 | cmnd->rw.reftag = cpu_to_le32( | |
364 | nvme_block_nr(ns, blk_rq_pos(req))); | |
365 | break; | |
366 | } | |
367 | if (!blk_integrity_rq(req)) | |
368 | control |= NVME_RW_PRINFO_PRACT; | |
369 | } | |
370 | ||
371 | cmnd->rw.control = cpu_to_le16(control); | |
372 | cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); | |
373 | } | |
374 | ||
375 | int nvme_setup_cmd(struct nvme_ns *ns, struct request *req, | |
376 | struct nvme_command *cmd) | |
377 | { | |
bac0000a | 378 | int ret = BLK_MQ_RQ_QUEUE_OK; |
8093f7ca | 379 | |
987f699a | 380 | if (!(req->rq_flags & RQF_DONTPREP)) { |
44e44b29 | 381 | nvme_req(req)->retries = 0; |
27fa9bc5 | 382 | nvme_req(req)->flags = 0; |
987f699a CH |
383 | req->rq_flags |= RQF_DONTPREP; |
384 | } | |
385 | ||
aebf526b CH |
386 | switch (req_op(req)) { |
387 | case REQ_OP_DRV_IN: | |
388 | case REQ_OP_DRV_OUT: | |
d49187e9 | 389 | memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd)); |
aebf526b CH |
390 | break; |
391 | case REQ_OP_FLUSH: | |
8093f7ca | 392 | nvme_setup_flush(ns, cmd); |
aebf526b | 393 | break; |
e850fd16 CH |
394 | case REQ_OP_WRITE_ZEROES: |
395 | /* currently only aliased to deallocate for a few ctrls: */ | |
aebf526b | 396 | case REQ_OP_DISCARD: |
8093f7ca | 397 | ret = nvme_setup_discard(ns, req, cmd); |
aebf526b CH |
398 | break; |
399 | case REQ_OP_READ: | |
400 | case REQ_OP_WRITE: | |
8093f7ca | 401 | nvme_setup_rw(ns, req, cmd); |
aebf526b CH |
402 | break; |
403 | default: | |
404 | WARN_ON_ONCE(1); | |
405 | return BLK_MQ_RQ_QUEUE_ERROR; | |
406 | } | |
8093f7ca | 407 | |
721b3917 | 408 | cmd->common.command_id = req->tag; |
8093f7ca ML |
409 | return ret; |
410 | } | |
411 | EXPORT_SYMBOL_GPL(nvme_setup_cmd); | |
412 | ||
4160982e CH |
413 | /* |
414 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
415 | * if the result is positive, it's an NVM Express status code | |
416 | */ | |
417 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
d49187e9 | 418 | union nvme_result *result, void *buffer, unsigned bufflen, |
eb71f435 | 419 | unsigned timeout, int qid, int at_head, int flags) |
4160982e CH |
420 | { |
421 | struct request *req; | |
422 | int ret; | |
423 | ||
eb71f435 | 424 | req = nvme_alloc_request(q, cmd, flags, qid); |
4160982e CH |
425 | if (IS_ERR(req)) |
426 | return PTR_ERR(req); | |
427 | ||
428 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
429 | ||
21d34711 CH |
430 | if (buffer && bufflen) { |
431 | ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL); | |
432 | if (ret) | |
433 | goto out; | |
4160982e CH |
434 | } |
435 | ||
eb71f435 | 436 | blk_execute_rq(req->q, NULL, req, at_head); |
d49187e9 CH |
437 | if (result) |
438 | *result = nvme_req(req)->result; | |
27fa9bc5 CH |
439 | if (nvme_req(req)->flags & NVME_REQ_CANCELLED) |
440 | ret = -EINTR; | |
441 | else | |
442 | ret = nvme_req(req)->status; | |
4160982e CH |
443 | out: |
444 | blk_mq_free_request(req); | |
445 | return ret; | |
446 | } | |
eb71f435 | 447 | EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd); |
4160982e CH |
448 | |
449 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
450 | void *buffer, unsigned bufflen) | |
451 | { | |
eb71f435 CH |
452 | return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0, |
453 | NVME_QID_ANY, 0, 0); | |
4160982e | 454 | } |
576d55d6 | 455 | EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd); |
4160982e | 456 | |
0b7f1f26 KB |
457 | int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd, |
458 | void __user *ubuffer, unsigned bufflen, | |
459 | void __user *meta_buffer, unsigned meta_len, u32 meta_seed, | |
460 | u32 *result, unsigned timeout) | |
4160982e | 461 | { |
7a5abb4b | 462 | bool write = nvme_is_write(cmd); |
0b7f1f26 KB |
463 | struct nvme_ns *ns = q->queuedata; |
464 | struct gendisk *disk = ns ? ns->disk : NULL; | |
4160982e | 465 | struct request *req; |
0b7f1f26 KB |
466 | struct bio *bio = NULL; |
467 | void *meta = NULL; | |
4160982e CH |
468 | int ret; |
469 | ||
eb71f435 | 470 | req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY); |
4160982e CH |
471 | if (IS_ERR(req)) |
472 | return PTR_ERR(req); | |
473 | ||
474 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
475 | ||
476 | if (ubuffer && bufflen) { | |
21d34711 CH |
477 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, |
478 | GFP_KERNEL); | |
479 | if (ret) | |
480 | goto out; | |
481 | bio = req->bio; | |
21d34711 | 482 | |
0b7f1f26 KB |
483 | if (!disk) |
484 | goto submit; | |
485 | bio->bi_bdev = bdget_disk(disk, 0); | |
486 | if (!bio->bi_bdev) { | |
487 | ret = -ENODEV; | |
488 | goto out_unmap; | |
489 | } | |
490 | ||
e9fc63d6 | 491 | if (meta_buffer && meta_len) { |
0b7f1f26 KB |
492 | struct bio_integrity_payload *bip; |
493 | ||
494 | meta = kmalloc(meta_len, GFP_KERNEL); | |
495 | if (!meta) { | |
496 | ret = -ENOMEM; | |
497 | goto out_unmap; | |
498 | } | |
499 | ||
500 | if (write) { | |
501 | if (copy_from_user(meta, meta_buffer, | |
502 | meta_len)) { | |
503 | ret = -EFAULT; | |
504 | goto out_free_meta; | |
505 | } | |
506 | } | |
507 | ||
508 | bip = bio_integrity_alloc(bio, GFP_KERNEL, 1); | |
06c1e390 KB |
509 | if (IS_ERR(bip)) { |
510 | ret = PTR_ERR(bip); | |
0b7f1f26 KB |
511 | goto out_free_meta; |
512 | } | |
513 | ||
514 | bip->bip_iter.bi_size = meta_len; | |
515 | bip->bip_iter.bi_sector = meta_seed; | |
516 | ||
517 | ret = bio_integrity_add_page(bio, virt_to_page(meta), | |
518 | meta_len, offset_in_page(meta)); | |
519 | if (ret != meta_len) { | |
520 | ret = -ENOMEM; | |
521 | goto out_free_meta; | |
522 | } | |
523 | } | |
524 | } | |
525 | submit: | |
526 | blk_execute_rq(req->q, disk, req, 0); | |
27fa9bc5 CH |
527 | if (nvme_req(req)->flags & NVME_REQ_CANCELLED) |
528 | ret = -EINTR; | |
529 | else | |
530 | ret = nvme_req(req)->status; | |
21d34711 | 531 | if (result) |
d49187e9 | 532 | *result = le32_to_cpu(nvme_req(req)->result.u32); |
0b7f1f26 KB |
533 | if (meta && !ret && !write) { |
534 | if (copy_to_user(meta_buffer, meta, meta_len)) | |
535 | ret = -EFAULT; | |
536 | } | |
537 | out_free_meta: | |
538 | kfree(meta); | |
539 | out_unmap: | |
540 | if (bio) { | |
541 | if (disk && bio->bi_bdev) | |
542 | bdput(bio->bi_bdev); | |
543 | blk_rq_unmap_user(bio); | |
544 | } | |
21d34711 CH |
545 | out: |
546 | blk_mq_free_request(req); | |
547 | return ret; | |
548 | } | |
549 | ||
0b7f1f26 KB |
550 | int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd, |
551 | void __user *ubuffer, unsigned bufflen, u32 *result, | |
552 | unsigned timeout) | |
553 | { | |
554 | return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0, | |
555 | result, timeout); | |
556 | } | |
557 | ||
038bd4cb SG |
558 | static void nvme_keep_alive_end_io(struct request *rq, int error) |
559 | { | |
560 | struct nvme_ctrl *ctrl = rq->end_io_data; | |
561 | ||
562 | blk_mq_free_request(rq); | |
563 | ||
564 | if (error) { | |
565 | dev_err(ctrl->device, | |
566 | "failed nvme_keep_alive_end_io error=%d\n", error); | |
567 | return; | |
568 | } | |
569 | ||
570 | schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); | |
571 | } | |
572 | ||
573 | static int nvme_keep_alive(struct nvme_ctrl *ctrl) | |
574 | { | |
575 | struct nvme_command c; | |
576 | struct request *rq; | |
577 | ||
578 | memset(&c, 0, sizeof(c)); | |
579 | c.common.opcode = nvme_admin_keep_alive; | |
580 | ||
581 | rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED, | |
582 | NVME_QID_ANY); | |
583 | if (IS_ERR(rq)) | |
584 | return PTR_ERR(rq); | |
585 | ||
586 | rq->timeout = ctrl->kato * HZ; | |
587 | rq->end_io_data = ctrl; | |
588 | ||
589 | blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io); | |
590 | ||
591 | return 0; | |
592 | } | |
593 | ||
594 | static void nvme_keep_alive_work(struct work_struct *work) | |
595 | { | |
596 | struct nvme_ctrl *ctrl = container_of(to_delayed_work(work), | |
597 | struct nvme_ctrl, ka_work); | |
598 | ||
599 | if (nvme_keep_alive(ctrl)) { | |
600 | /* allocation failure, reset the controller */ | |
601 | dev_err(ctrl->device, "keep-alive failed\n"); | |
602 | ctrl->ops->reset_ctrl(ctrl); | |
603 | return; | |
604 | } | |
605 | } | |
606 | ||
607 | void nvme_start_keep_alive(struct nvme_ctrl *ctrl) | |
608 | { | |
609 | if (unlikely(ctrl->kato == 0)) | |
610 | return; | |
611 | ||
612 | INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work); | |
613 | schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(nvme_start_keep_alive); | |
616 | ||
617 | void nvme_stop_keep_alive(struct nvme_ctrl *ctrl) | |
618 | { | |
619 | if (unlikely(ctrl->kato == 0)) | |
620 | return; | |
621 | ||
622 | cancel_delayed_work_sync(&ctrl->ka_work); | |
623 | } | |
624 | EXPORT_SYMBOL_GPL(nvme_stop_keep_alive); | |
625 | ||
1c63dc66 | 626 | int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id) |
21d34711 CH |
627 | { |
628 | struct nvme_command c = { }; | |
629 | int error; | |
630 | ||
631 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
632 | c.identify.opcode = nvme_admin_identify; | |
986994a2 | 633 | c.identify.cns = NVME_ID_CNS_CTRL; |
21d34711 CH |
634 | |
635 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); | |
636 | if (!*id) | |
637 | return -ENOMEM; | |
638 | ||
639 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
640 | sizeof(struct nvme_id_ctrl)); | |
641 | if (error) | |
642 | kfree(*id); | |
643 | return error; | |
644 | } | |
645 | ||
540c801c KB |
646 | static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list) |
647 | { | |
648 | struct nvme_command c = { }; | |
649 | ||
650 | c.identify.opcode = nvme_admin_identify; | |
986994a2 | 651 | c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST; |
540c801c KB |
652 | c.identify.nsid = cpu_to_le32(nsid); |
653 | return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000); | |
654 | } | |
655 | ||
1c63dc66 | 656 | int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid, |
21d34711 CH |
657 | struct nvme_id_ns **id) |
658 | { | |
659 | struct nvme_command c = { }; | |
660 | int error; | |
661 | ||
662 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
778f067c MG |
663 | c.identify.opcode = nvme_admin_identify; |
664 | c.identify.nsid = cpu_to_le32(nsid); | |
986994a2 | 665 | c.identify.cns = NVME_ID_CNS_NS; |
21d34711 CH |
666 | |
667 | *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL); | |
668 | if (!*id) | |
669 | return -ENOMEM; | |
670 | ||
671 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
672 | sizeof(struct nvme_id_ns)); | |
673 | if (error) | |
674 | kfree(*id); | |
675 | return error; | |
676 | } | |
677 | ||
1c63dc66 | 678 | int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid, |
1a6fe74d | 679 | void *buffer, size_t buflen, u32 *result) |
21d34711 CH |
680 | { |
681 | struct nvme_command c; | |
d49187e9 | 682 | union nvme_result res; |
1cb3cce5 | 683 | int ret; |
21d34711 CH |
684 | |
685 | memset(&c, 0, sizeof(c)); | |
686 | c.features.opcode = nvme_admin_get_features; | |
687 | c.features.nsid = cpu_to_le32(nsid); | |
21d34711 CH |
688 | c.features.fid = cpu_to_le32(fid); |
689 | ||
d49187e9 | 690 | ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0, |
eb71f435 | 691 | NVME_QID_ANY, 0, 0); |
9b47f77a | 692 | if (ret >= 0 && result) |
d49187e9 | 693 | *result = le32_to_cpu(res.u32); |
1cb3cce5 | 694 | return ret; |
21d34711 CH |
695 | } |
696 | ||
1c63dc66 | 697 | int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11, |
1a6fe74d | 698 | void *buffer, size_t buflen, u32 *result) |
21d34711 CH |
699 | { |
700 | struct nvme_command c; | |
d49187e9 | 701 | union nvme_result res; |
1cb3cce5 | 702 | int ret; |
21d34711 CH |
703 | |
704 | memset(&c, 0, sizeof(c)); | |
705 | c.features.opcode = nvme_admin_set_features; | |
21d34711 CH |
706 | c.features.fid = cpu_to_le32(fid); |
707 | c.features.dword11 = cpu_to_le32(dword11); | |
708 | ||
d49187e9 | 709 | ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, |
1a6fe74d | 710 | buffer, buflen, 0, NVME_QID_ANY, 0, 0); |
9b47f77a | 711 | if (ret >= 0 && result) |
d49187e9 | 712 | *result = le32_to_cpu(res.u32); |
1cb3cce5 | 713 | return ret; |
21d34711 CH |
714 | } |
715 | ||
1c63dc66 | 716 | int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log) |
21d34711 CH |
717 | { |
718 | struct nvme_command c = { }; | |
719 | int error; | |
720 | ||
721 | c.common.opcode = nvme_admin_get_log_page, | |
722 | c.common.nsid = cpu_to_le32(0xFFFFFFFF), | |
723 | c.common.cdw10[0] = cpu_to_le32( | |
724 | (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) | | |
725 | NVME_LOG_SMART), | |
726 | ||
727 | *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL); | |
728 | if (!*log) | |
729 | return -ENOMEM; | |
730 | ||
731 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *log, | |
732 | sizeof(struct nvme_smart_log)); | |
733 | if (error) | |
734 | kfree(*log); | |
735 | return error; | |
736 | } | |
1673f1f0 | 737 | |
9a0be7ab CH |
738 | int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count) |
739 | { | |
740 | u32 q_count = (*count - 1) | ((*count - 1) << 16); | |
741 | u32 result; | |
742 | int status, nr_io_queues; | |
743 | ||
1a6fe74d | 744 | status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0, |
9a0be7ab | 745 | &result); |
f5fa90dc | 746 | if (status < 0) |
9a0be7ab CH |
747 | return status; |
748 | ||
f5fa90dc CH |
749 | /* |
750 | * Degraded controllers might return an error when setting the queue | |
751 | * count. We still want to be able to bring them online and offer | |
752 | * access to the admin queue, as that might be only way to fix them up. | |
753 | */ | |
754 | if (status > 0) { | |
755 | dev_err(ctrl->dev, "Could not set queue count (%d)\n", status); | |
756 | *count = 0; | |
757 | } else { | |
758 | nr_io_queues = min(result & 0xffff, result >> 16) + 1; | |
759 | *count = min(*count, nr_io_queues); | |
760 | } | |
761 | ||
9a0be7ab CH |
762 | return 0; |
763 | } | |
576d55d6 | 764 | EXPORT_SYMBOL_GPL(nvme_set_queue_count); |
9a0be7ab | 765 | |
1673f1f0 CH |
766 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
767 | { | |
768 | struct nvme_user_io io; | |
769 | struct nvme_command c; | |
770 | unsigned length, meta_len; | |
771 | void __user *metadata; | |
772 | ||
773 | if (copy_from_user(&io, uio, sizeof(io))) | |
774 | return -EFAULT; | |
63088ec7 KB |
775 | if (io.flags) |
776 | return -EINVAL; | |
1673f1f0 CH |
777 | |
778 | switch (io.opcode) { | |
779 | case nvme_cmd_write: | |
780 | case nvme_cmd_read: | |
781 | case nvme_cmd_compare: | |
782 | break; | |
783 | default: | |
784 | return -EINVAL; | |
785 | } | |
786 | ||
787 | length = (io.nblocks + 1) << ns->lba_shift; | |
788 | meta_len = (io.nblocks + 1) * ns->ms; | |
789 | metadata = (void __user *)(uintptr_t)io.metadata; | |
790 | ||
791 | if (ns->ext) { | |
792 | length += meta_len; | |
793 | meta_len = 0; | |
794 | } else if (meta_len) { | |
795 | if ((io.metadata & 3) || !io.metadata) | |
796 | return -EINVAL; | |
797 | } | |
798 | ||
799 | memset(&c, 0, sizeof(c)); | |
800 | c.rw.opcode = io.opcode; | |
801 | c.rw.flags = io.flags; | |
802 | c.rw.nsid = cpu_to_le32(ns->ns_id); | |
803 | c.rw.slba = cpu_to_le64(io.slba); | |
804 | c.rw.length = cpu_to_le16(io.nblocks); | |
805 | c.rw.control = cpu_to_le16(io.control); | |
806 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); | |
807 | c.rw.reftag = cpu_to_le32(io.reftag); | |
808 | c.rw.apptag = cpu_to_le16(io.apptag); | |
809 | c.rw.appmask = cpu_to_le16(io.appmask); | |
810 | ||
811 | return __nvme_submit_user_cmd(ns->queue, &c, | |
812 | (void __user *)(uintptr_t)io.addr, length, | |
813 | metadata, meta_len, io.slba, NULL, 0); | |
814 | } | |
815 | ||
f3ca80fc | 816 | static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns, |
1673f1f0 CH |
817 | struct nvme_passthru_cmd __user *ucmd) |
818 | { | |
819 | struct nvme_passthru_cmd cmd; | |
820 | struct nvme_command c; | |
821 | unsigned timeout = 0; | |
822 | int status; | |
823 | ||
824 | if (!capable(CAP_SYS_ADMIN)) | |
825 | return -EACCES; | |
826 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
827 | return -EFAULT; | |
63088ec7 KB |
828 | if (cmd.flags) |
829 | return -EINVAL; | |
1673f1f0 CH |
830 | |
831 | memset(&c, 0, sizeof(c)); | |
832 | c.common.opcode = cmd.opcode; | |
833 | c.common.flags = cmd.flags; | |
834 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
835 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
836 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
837 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
838 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
839 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
840 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
841 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
842 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
843 | ||
844 | if (cmd.timeout_ms) | |
845 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
846 | ||
847 | status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c, | |
d1ea7be5 | 848 | (void __user *)(uintptr_t)cmd.addr, cmd.data_len, |
1673f1f0 CH |
849 | &cmd.result, timeout); |
850 | if (status >= 0) { | |
851 | if (put_user(cmd.result, &ucmd->result)) | |
852 | return -EFAULT; | |
853 | } | |
854 | ||
855 | return status; | |
856 | } | |
857 | ||
858 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, | |
859 | unsigned int cmd, unsigned long arg) | |
860 | { | |
861 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
862 | ||
863 | switch (cmd) { | |
864 | case NVME_IOCTL_ID: | |
865 | force_successful_syscall_return(); | |
866 | return ns->ns_id; | |
867 | case NVME_IOCTL_ADMIN_CMD: | |
868 | return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg); | |
869 | case NVME_IOCTL_IO_CMD: | |
870 | return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg); | |
871 | case NVME_IOCTL_SUBMIT_IO: | |
872 | return nvme_submit_io(ns, (void __user *)arg); | |
44907332 | 873 | #ifdef CONFIG_BLK_DEV_NVME_SCSI |
1673f1f0 CH |
874 | case SG_GET_VERSION_NUM: |
875 | return nvme_sg_get_version_num((void __user *)arg); | |
876 | case SG_IO: | |
877 | return nvme_sg_io(ns, (void __user *)arg); | |
44907332 | 878 | #endif |
1673f1f0 | 879 | default: |
84d4add7 MB |
880 | #ifdef CONFIG_NVM |
881 | if (ns->ndev) | |
882 | return nvme_nvm_ioctl(ns, cmd, arg); | |
883 | #endif | |
a98e58e5 | 884 | if (is_sed_ioctl(cmd)) |
4f1244c8 | 885 | return sed_ioctl(ns->ctrl->opal_dev, cmd, |
e225c20e | 886 | (void __user *) arg); |
1673f1f0 CH |
887 | return -ENOTTY; |
888 | } | |
889 | } | |
890 | ||
891 | #ifdef CONFIG_COMPAT | |
892 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
893 | unsigned int cmd, unsigned long arg) | |
894 | { | |
895 | switch (cmd) { | |
896 | case SG_IO: | |
897 | return -ENOIOCTLCMD; | |
898 | } | |
899 | return nvme_ioctl(bdev, mode, cmd, arg); | |
900 | } | |
901 | #else | |
902 | #define nvme_compat_ioctl NULL | |
903 | #endif | |
904 | ||
905 | static int nvme_open(struct block_device *bdev, fmode_t mode) | |
906 | { | |
907 | return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO; | |
908 | } | |
909 | ||
910 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
911 | { | |
e439bb12 SG |
912 | struct nvme_ns *ns = disk->private_data; |
913 | ||
914 | module_put(ns->ctrl->ops->module); | |
915 | nvme_put_ns(ns); | |
1673f1f0 CH |
916 | } |
917 | ||
918 | static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
919 | { | |
920 | /* some standard values */ | |
921 | geo->heads = 1 << 6; | |
922 | geo->sectors = 1 << 5; | |
923 | geo->cylinders = get_capacity(bdev->bd_disk) >> 11; | |
924 | return 0; | |
925 | } | |
926 | ||
927 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
c81bfba9 CH |
928 | static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id, |
929 | u16 bs) | |
930 | { | |
931 | struct nvme_ns *ns = disk->private_data; | |
932 | u16 old_ms = ns->ms; | |
933 | u8 pi_type = 0; | |
934 | ||
935 | ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms); | |
936 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); | |
937 | ||
938 | /* PI implementation requires metadata equal t10 pi tuple size */ | |
939 | if (ns->ms == sizeof(struct t10_pi_tuple)) | |
940 | pi_type = id->dps & NVME_NS_DPS_PI_MASK; | |
941 | ||
942 | if (blk_get_integrity(disk) && | |
943 | (ns->pi_type != pi_type || ns->ms != old_ms || | |
944 | bs != queue_logical_block_size(disk->queue) || | |
945 | (ns->ms && ns->ext))) | |
946 | blk_integrity_unregister(disk); | |
947 | ||
948 | ns->pi_type = pi_type; | |
949 | } | |
950 | ||
1673f1f0 CH |
951 | static void nvme_init_integrity(struct nvme_ns *ns) |
952 | { | |
953 | struct blk_integrity integrity; | |
954 | ||
fa9a89fc | 955 | memset(&integrity, 0, sizeof(integrity)); |
1673f1f0 CH |
956 | switch (ns->pi_type) { |
957 | case NVME_NS_DPS_PI_TYPE3: | |
958 | integrity.profile = &t10_pi_type3_crc; | |
ba36c21b NB |
959 | integrity.tag_size = sizeof(u16) + sizeof(u32); |
960 | integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; | |
1673f1f0 CH |
961 | break; |
962 | case NVME_NS_DPS_PI_TYPE1: | |
963 | case NVME_NS_DPS_PI_TYPE2: | |
964 | integrity.profile = &t10_pi_type1_crc; | |
ba36c21b NB |
965 | integrity.tag_size = sizeof(u16); |
966 | integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; | |
1673f1f0 CH |
967 | break; |
968 | default: | |
969 | integrity.profile = NULL; | |
970 | break; | |
971 | } | |
972 | integrity.tuple_size = ns->ms; | |
973 | blk_integrity_register(ns->disk, &integrity); | |
974 | blk_queue_max_integrity_segments(ns->queue, 1); | |
975 | } | |
976 | #else | |
c81bfba9 CH |
977 | static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id, |
978 | u16 bs) | |
979 | { | |
980 | } | |
1673f1f0 CH |
981 | static void nvme_init_integrity(struct nvme_ns *ns) |
982 | { | |
983 | } | |
984 | #endif /* CONFIG_BLK_DEV_INTEGRITY */ | |
985 | ||
986 | static void nvme_config_discard(struct nvme_ns *ns) | |
987 | { | |
08095e70 | 988 | struct nvme_ctrl *ctrl = ns->ctrl; |
1673f1f0 | 989 | u32 logical_block_size = queue_logical_block_size(ns->queue); |
08095e70 | 990 | |
b35ba01e CH |
991 | BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) < |
992 | NVME_DSM_MAX_RANGES); | |
993 | ||
1673f1f0 CH |
994 | ns->queue->limits.discard_alignment = logical_block_size; |
995 | ns->queue->limits.discard_granularity = logical_block_size; | |
bd0fc288 | 996 | blk_queue_max_discard_sectors(ns->queue, UINT_MAX); |
b35ba01e | 997 | blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES); |
1673f1f0 | 998 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); |
e850fd16 CH |
999 | |
1000 | if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) | |
1001 | blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX); | |
1673f1f0 CH |
1002 | } |
1003 | ||
ac81bfa9 | 1004 | static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id) |
1673f1f0 | 1005 | { |
ac81bfa9 | 1006 | if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) { |
b0b4e09c | 1007 | dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__); |
1673f1f0 CH |
1008 | return -ENODEV; |
1009 | } | |
1673f1f0 | 1010 | |
ac81bfa9 MB |
1011 | if ((*id)->ncap == 0) { |
1012 | kfree(*id); | |
1013 | return -ENODEV; | |
1673f1f0 CH |
1014 | } |
1015 | ||
8ef2074d | 1016 | if (ns->ctrl->vs >= NVME_VS(1, 1, 0)) |
ac81bfa9 | 1017 | memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui)); |
8ef2074d | 1018 | if (ns->ctrl->vs >= NVME_VS(1, 2, 0)) |
ac81bfa9 MB |
1019 | memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid)); |
1020 | ||
1021 | return 0; | |
1022 | } | |
1023 | ||
1024 | static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id) | |
1025 | { | |
1026 | struct nvme_ns *ns = disk->private_data; | |
c81bfba9 | 1027 | u16 bs; |
1673f1f0 CH |
1028 | |
1029 | /* | |
1030 | * If identify namespace failed, use default 512 byte block size so | |
1031 | * block layer can use before failing read/write for 0 capacity. | |
1032 | */ | |
c81bfba9 | 1033 | ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds; |
1673f1f0 CH |
1034 | if (ns->lba_shift == 0) |
1035 | ns->lba_shift = 9; | |
1036 | bs = 1 << ns->lba_shift; | |
1673f1f0 CH |
1037 | |
1038 | blk_mq_freeze_queue(disk->queue); | |
1673f1f0 | 1039 | |
c81bfba9 CH |
1040 | if (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED) |
1041 | nvme_prep_integrity(disk, id, bs); | |
1673f1f0 | 1042 | blk_queue_logical_block_size(ns->queue, bs); |
4b9d5b15 | 1043 | if (ns->ms && !blk_get_integrity(disk) && !ns->ext) |
1673f1f0 | 1044 | nvme_init_integrity(ns); |
1673f1f0 CH |
1045 | if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk)) |
1046 | set_capacity(disk, 0); | |
1047 | else | |
1048 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); | |
1049 | ||
1050 | if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM) | |
1051 | nvme_config_discard(ns); | |
1052 | blk_mq_unfreeze_queue(disk->queue); | |
ac81bfa9 | 1053 | } |
1673f1f0 | 1054 | |
ac81bfa9 MB |
1055 | static int nvme_revalidate_disk(struct gendisk *disk) |
1056 | { | |
1057 | struct nvme_ns *ns = disk->private_data; | |
1058 | struct nvme_id_ns *id = NULL; | |
1059 | int ret; | |
1060 | ||
1061 | if (test_bit(NVME_NS_DEAD, &ns->flags)) { | |
1062 | set_capacity(disk, 0); | |
1063 | return -ENODEV; | |
1064 | } | |
1065 | ||
1066 | ret = nvme_revalidate_ns(ns, &id); | |
1067 | if (ret) | |
1068 | return ret; | |
1069 | ||
1070 | __nvme_revalidate_disk(disk, id); | |
1673f1f0 | 1071 | kfree(id); |
ac81bfa9 | 1072 | |
1673f1f0 CH |
1073 | return 0; |
1074 | } | |
1075 | ||
1076 | static char nvme_pr_type(enum pr_type type) | |
1077 | { | |
1078 | switch (type) { | |
1079 | case PR_WRITE_EXCLUSIVE: | |
1080 | return 1; | |
1081 | case PR_EXCLUSIVE_ACCESS: | |
1082 | return 2; | |
1083 | case PR_WRITE_EXCLUSIVE_REG_ONLY: | |
1084 | return 3; | |
1085 | case PR_EXCLUSIVE_ACCESS_REG_ONLY: | |
1086 | return 4; | |
1087 | case PR_WRITE_EXCLUSIVE_ALL_REGS: | |
1088 | return 5; | |
1089 | case PR_EXCLUSIVE_ACCESS_ALL_REGS: | |
1090 | return 6; | |
1091 | default: | |
1092 | return 0; | |
1093 | } | |
1094 | }; | |
1095 | ||
1096 | static int nvme_pr_command(struct block_device *bdev, u32 cdw10, | |
1097 | u64 key, u64 sa_key, u8 op) | |
1098 | { | |
1099 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1100 | struct nvme_command c; | |
1101 | u8 data[16] = { 0, }; | |
1102 | ||
1103 | put_unaligned_le64(key, &data[0]); | |
1104 | put_unaligned_le64(sa_key, &data[8]); | |
1105 | ||
1106 | memset(&c, 0, sizeof(c)); | |
1107 | c.common.opcode = op; | |
1108 | c.common.nsid = cpu_to_le32(ns->ns_id); | |
1109 | c.common.cdw10[0] = cpu_to_le32(cdw10); | |
1110 | ||
1111 | return nvme_submit_sync_cmd(ns->queue, &c, data, 16); | |
1112 | } | |
1113 | ||
1114 | static int nvme_pr_register(struct block_device *bdev, u64 old, | |
1115 | u64 new, unsigned flags) | |
1116 | { | |
1117 | u32 cdw10; | |
1118 | ||
1119 | if (flags & ~PR_FL_IGNORE_KEY) | |
1120 | return -EOPNOTSUPP; | |
1121 | ||
1122 | cdw10 = old ? 2 : 0; | |
1123 | cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; | |
1124 | cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ | |
1125 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register); | |
1126 | } | |
1127 | ||
1128 | static int nvme_pr_reserve(struct block_device *bdev, u64 key, | |
1129 | enum pr_type type, unsigned flags) | |
1130 | { | |
1131 | u32 cdw10; | |
1132 | ||
1133 | if (flags & ~PR_FL_IGNORE_KEY) | |
1134 | return -EOPNOTSUPP; | |
1135 | ||
1136 | cdw10 = nvme_pr_type(type) << 8; | |
1137 | cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); | |
1138 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire); | |
1139 | } | |
1140 | ||
1141 | static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, | |
1142 | enum pr_type type, bool abort) | |
1143 | { | |
1144 | u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1; | |
1145 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire); | |
1146 | } | |
1147 | ||
1148 | static int nvme_pr_clear(struct block_device *bdev, u64 key) | |
1149 | { | |
8c0b3915 | 1150 | u32 cdw10 = 1 | (key ? 1 << 3 : 0); |
1673f1f0 CH |
1151 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register); |
1152 | } | |
1153 | ||
1154 | static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
1155 | { | |
1156 | u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0; | |
1157 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release); | |
1158 | } | |
1159 | ||
1160 | static const struct pr_ops nvme_pr_ops = { | |
1161 | .pr_register = nvme_pr_register, | |
1162 | .pr_reserve = nvme_pr_reserve, | |
1163 | .pr_release = nvme_pr_release, | |
1164 | .pr_preempt = nvme_pr_preempt, | |
1165 | .pr_clear = nvme_pr_clear, | |
1166 | }; | |
1167 | ||
a98e58e5 | 1168 | #ifdef CONFIG_BLK_SED_OPAL |
4f1244c8 CH |
1169 | int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len, |
1170 | bool send) | |
a98e58e5 | 1171 | { |
4f1244c8 | 1172 | struct nvme_ctrl *ctrl = data; |
a98e58e5 | 1173 | struct nvme_command cmd; |
a98e58e5 SB |
1174 | |
1175 | memset(&cmd, 0, sizeof(cmd)); | |
1176 | if (send) | |
1177 | cmd.common.opcode = nvme_admin_security_send; | |
1178 | else | |
1179 | cmd.common.opcode = nvme_admin_security_recv; | |
a98e58e5 SB |
1180 | cmd.common.nsid = 0; |
1181 | cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8); | |
1182 | cmd.common.cdw10[1] = cpu_to_le32(len); | |
1183 | ||
1184 | return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, | |
1185 | ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0); | |
1186 | } | |
1187 | EXPORT_SYMBOL_GPL(nvme_sec_submit); | |
1188 | #endif /* CONFIG_BLK_SED_OPAL */ | |
1189 | ||
5bae7f73 | 1190 | static const struct block_device_operations nvme_fops = { |
1673f1f0 CH |
1191 | .owner = THIS_MODULE, |
1192 | .ioctl = nvme_ioctl, | |
1193 | .compat_ioctl = nvme_compat_ioctl, | |
1194 | .open = nvme_open, | |
1195 | .release = nvme_release, | |
1196 | .getgeo = nvme_getgeo, | |
1197 | .revalidate_disk= nvme_revalidate_disk, | |
1198 | .pr_ops = &nvme_pr_ops, | |
1199 | }; | |
1200 | ||
5fd4ce1b CH |
1201 | static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled) |
1202 | { | |
1203 | unsigned long timeout = | |
1204 | ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
1205 | u32 csts, bit = enabled ? NVME_CSTS_RDY : 0; | |
1206 | int ret; | |
1207 | ||
1208 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
0df1e4f5 KB |
1209 | if (csts == ~0) |
1210 | return -ENODEV; | |
5fd4ce1b CH |
1211 | if ((csts & NVME_CSTS_RDY) == bit) |
1212 | break; | |
1213 | ||
1214 | msleep(100); | |
1215 | if (fatal_signal_pending(current)) | |
1216 | return -EINTR; | |
1217 | if (time_after(jiffies, timeout)) { | |
1b3c47c1 | 1218 | dev_err(ctrl->device, |
5fd4ce1b CH |
1219 | "Device not ready; aborting %s\n", enabled ? |
1220 | "initialisation" : "reset"); | |
1221 | return -ENODEV; | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | return ret; | |
1226 | } | |
1227 | ||
1228 | /* | |
1229 | * If the device has been passed off to us in an enabled state, just clear | |
1230 | * the enabled bit. The spec says we should set the 'shutdown notification | |
1231 | * bits', but doing so may cause the device to complete commands to the | |
1232 | * admin queue ... and we don't know what memory that might be pointing at! | |
1233 | */ | |
1234 | int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap) | |
1235 | { | |
1236 | int ret; | |
1237 | ||
1238 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
1239 | ctrl->ctrl_config &= ~NVME_CC_ENABLE; | |
1240 | ||
1241 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
1242 | if (ret) | |
1243 | return ret; | |
54adc010 | 1244 | |
b5a10c5f | 1245 | if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) |
54adc010 GP |
1246 | msleep(NVME_QUIRK_DELAY_AMOUNT); |
1247 | ||
5fd4ce1b CH |
1248 | return nvme_wait_ready(ctrl, cap, false); |
1249 | } | |
576d55d6 | 1250 | EXPORT_SYMBOL_GPL(nvme_disable_ctrl); |
5fd4ce1b CH |
1251 | |
1252 | int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap) | |
1253 | { | |
1254 | /* | |
1255 | * Default to a 4K page size, with the intention to update this | |
1256 | * path in the future to accomodate architectures with differing | |
1257 | * kernel and IO page sizes. | |
1258 | */ | |
1259 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12; | |
1260 | int ret; | |
1261 | ||
1262 | if (page_shift < dev_page_min) { | |
1b3c47c1 | 1263 | dev_err(ctrl->device, |
5fd4ce1b CH |
1264 | "Minimum device page size %u too large for host (%u)\n", |
1265 | 1 << dev_page_min, 1 << page_shift); | |
1266 | return -ENODEV; | |
1267 | } | |
1268 | ||
1269 | ctrl->page_size = 1 << page_shift; | |
1270 | ||
1271 | ctrl->ctrl_config = NVME_CC_CSS_NVM; | |
1272 | ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; | |
1273 | ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; | |
1274 | ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; | |
1275 | ctrl->ctrl_config |= NVME_CC_ENABLE; | |
1276 | ||
1277 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
1278 | if (ret) | |
1279 | return ret; | |
1280 | return nvme_wait_ready(ctrl, cap, true); | |
1281 | } | |
576d55d6 | 1282 | EXPORT_SYMBOL_GPL(nvme_enable_ctrl); |
5fd4ce1b CH |
1283 | |
1284 | int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl) | |
1285 | { | |
1286 | unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies; | |
1287 | u32 csts; | |
1288 | int ret; | |
1289 | ||
1290 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
1291 | ctrl->ctrl_config |= NVME_CC_SHN_NORMAL; | |
1292 | ||
1293 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
1294 | if (ret) | |
1295 | return ret; | |
1296 | ||
1297 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
1298 | if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT) | |
1299 | break; | |
1300 | ||
1301 | msleep(100); | |
1302 | if (fatal_signal_pending(current)) | |
1303 | return -EINTR; | |
1304 | if (time_after(jiffies, timeout)) { | |
1b3c47c1 | 1305 | dev_err(ctrl->device, |
5fd4ce1b CH |
1306 | "Device shutdown incomplete; abort shutdown\n"); |
1307 | return -ENODEV; | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | return ret; | |
1312 | } | |
576d55d6 | 1313 | EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl); |
5fd4ce1b | 1314 | |
da35825d CH |
1315 | static void nvme_set_queue_limits(struct nvme_ctrl *ctrl, |
1316 | struct request_queue *q) | |
1317 | { | |
7c88cb00 JA |
1318 | bool vwc = false; |
1319 | ||
da35825d | 1320 | if (ctrl->max_hw_sectors) { |
45686b61 CH |
1321 | u32 max_segments = |
1322 | (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1; | |
1323 | ||
da35825d | 1324 | blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors); |
45686b61 | 1325 | blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX)); |
da35825d | 1326 | } |
e6282aef KB |
1327 | if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) |
1328 | blk_queue_chunk_sectors(q, ctrl->max_hw_sectors); | |
da35825d | 1329 | blk_queue_virt_boundary(q, ctrl->page_size - 1); |
7c88cb00 JA |
1330 | if (ctrl->vwc & NVME_CTRL_VWC_PRESENT) |
1331 | vwc = true; | |
1332 | blk_queue_write_cache(q, vwc, vwc); | |
da35825d CH |
1333 | } |
1334 | ||
c5552fde AL |
1335 | static void nvme_configure_apst(struct nvme_ctrl *ctrl) |
1336 | { | |
1337 | /* | |
1338 | * APST (Autonomous Power State Transition) lets us program a | |
1339 | * table of power state transitions that the controller will | |
1340 | * perform automatically. We configure it with a simple | |
1341 | * heuristic: we are willing to spend at most 2% of the time | |
1342 | * transitioning between power states. Therefore, when running | |
1343 | * in any given state, we will enter the next lower-power | |
76e4ad09 | 1344 | * non-operational state after waiting 50 * (enlat + exlat) |
c5552fde AL |
1345 | * microseconds, as long as that state's total latency is under |
1346 | * the requested maximum latency. | |
1347 | * | |
1348 | * We will not autonomously enter any non-operational state for | |
1349 | * which the total latency exceeds ps_max_latency_us. Users | |
1350 | * can set ps_max_latency_us to zero to turn off APST. | |
1351 | */ | |
1352 | ||
1353 | unsigned apste; | |
1354 | struct nvme_feat_auto_pst *table; | |
fb0dc399 AL |
1355 | u64 max_lat_us = 0; |
1356 | int max_ps = -1; | |
c5552fde AL |
1357 | int ret; |
1358 | ||
1359 | /* | |
1360 | * If APST isn't supported or if we haven't been initialized yet, | |
1361 | * then don't do anything. | |
1362 | */ | |
1363 | if (!ctrl->apsta) | |
1364 | return; | |
1365 | ||
1366 | if (ctrl->npss > 31) { | |
1367 | dev_warn(ctrl->device, "NPSS is invalid; not using APST\n"); | |
1368 | return; | |
1369 | } | |
1370 | ||
1371 | table = kzalloc(sizeof(*table), GFP_KERNEL); | |
1372 | if (!table) | |
1373 | return; | |
1374 | ||
1375 | if (ctrl->ps_max_latency_us == 0) { | |
1376 | /* Turn off APST. */ | |
1377 | apste = 0; | |
fb0dc399 | 1378 | dev_dbg(ctrl->device, "APST disabled\n"); |
c5552fde AL |
1379 | } else { |
1380 | __le64 target = cpu_to_le64(0); | |
1381 | int state; | |
1382 | ||
1383 | /* | |
1384 | * Walk through all states from lowest- to highest-power. | |
1385 | * According to the spec, lower-numbered states use more | |
1386 | * power. NPSS, despite the name, is the index of the | |
1387 | * lowest-power state, not the number of states. | |
1388 | */ | |
1389 | for (state = (int)ctrl->npss; state >= 0; state--) { | |
1390 | u64 total_latency_us, transition_ms; | |
1391 | ||
1392 | if (target) | |
1393 | table->entries[state] = target; | |
1394 | ||
ff5350a8 AL |
1395 | /* |
1396 | * Don't allow transitions to the deepest state | |
1397 | * if it's quirked off. | |
1398 | */ | |
1399 | if (state == ctrl->npss && | |
1400 | (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) | |
1401 | continue; | |
1402 | ||
c5552fde AL |
1403 | /* |
1404 | * Is this state a useful non-operational state for | |
1405 | * higher-power states to autonomously transition to? | |
1406 | */ | |
1407 | if (!(ctrl->psd[state].flags & | |
1408 | NVME_PS_FLAGS_NON_OP_STATE)) | |
1409 | continue; | |
1410 | ||
1411 | total_latency_us = | |
1412 | (u64)le32_to_cpu(ctrl->psd[state].entry_lat) + | |
1413 | + le32_to_cpu(ctrl->psd[state].exit_lat); | |
1414 | if (total_latency_us > ctrl->ps_max_latency_us) | |
1415 | continue; | |
1416 | ||
1417 | /* | |
1418 | * This state is good. Use it as the APST idle | |
1419 | * target for higher power states. | |
1420 | */ | |
1421 | transition_ms = total_latency_us + 19; | |
1422 | do_div(transition_ms, 20); | |
1423 | if (transition_ms > (1 << 24) - 1) | |
1424 | transition_ms = (1 << 24) - 1; | |
1425 | ||
1426 | target = cpu_to_le64((state << 3) | | |
1427 | (transition_ms << 8)); | |
fb0dc399 AL |
1428 | |
1429 | if (max_ps == -1) | |
1430 | max_ps = state; | |
1431 | ||
1432 | if (total_latency_us > max_lat_us) | |
1433 | max_lat_us = total_latency_us; | |
c5552fde AL |
1434 | } |
1435 | ||
1436 | apste = 1; | |
fb0dc399 AL |
1437 | |
1438 | if (max_ps == -1) { | |
1439 | dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n"); | |
1440 | } else { | |
1441 | dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n", | |
1442 | max_ps, max_lat_us, (int)sizeof(*table), table); | |
1443 | } | |
c5552fde AL |
1444 | } |
1445 | ||
1446 | ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste, | |
1447 | table, sizeof(*table), NULL); | |
1448 | if (ret) | |
1449 | dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret); | |
1450 | ||
1451 | kfree(table); | |
1452 | } | |
1453 | ||
1454 | static void nvme_set_latency_tolerance(struct device *dev, s32 val) | |
1455 | { | |
1456 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1457 | u64 latency; | |
1458 | ||
1459 | switch (val) { | |
1460 | case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT: | |
1461 | case PM_QOS_LATENCY_ANY: | |
1462 | latency = U64_MAX; | |
1463 | break; | |
1464 | ||
1465 | default: | |
1466 | latency = val; | |
1467 | } | |
1468 | ||
1469 | if (ctrl->ps_max_latency_us != latency) { | |
1470 | ctrl->ps_max_latency_us = latency; | |
1471 | nvme_configure_apst(ctrl); | |
1472 | } | |
1473 | } | |
1474 | ||
bd4da3ab AL |
1475 | struct nvme_core_quirk_entry { |
1476 | /* | |
1477 | * NVMe model and firmware strings are padded with spaces. For | |
1478 | * simplicity, strings in the quirk table are padded with NULLs | |
1479 | * instead. | |
1480 | */ | |
1481 | u16 vid; | |
1482 | const char *mn; | |
1483 | const char *fr; | |
1484 | unsigned long quirks; | |
1485 | }; | |
1486 | ||
1487 | static const struct nvme_core_quirk_entry core_quirks[] = { | |
c5552fde | 1488 | { |
be56945c AL |
1489 | /* |
1490 | * This Toshiba device seems to die using any APST states. See: | |
1491 | * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11 | |
1492 | */ | |
1493 | .vid = 0x1179, | |
1494 | .mn = "THNSF5256GPUK TOSHIBA", | |
c5552fde | 1495 | .quirks = NVME_QUIRK_NO_APST, |
be56945c | 1496 | } |
bd4da3ab AL |
1497 | }; |
1498 | ||
1499 | /* match is null-terminated but idstr is space-padded. */ | |
1500 | static bool string_matches(const char *idstr, const char *match, size_t len) | |
1501 | { | |
1502 | size_t matchlen; | |
1503 | ||
1504 | if (!match) | |
1505 | return true; | |
1506 | ||
1507 | matchlen = strlen(match); | |
1508 | WARN_ON_ONCE(matchlen > len); | |
1509 | ||
1510 | if (memcmp(idstr, match, matchlen)) | |
1511 | return false; | |
1512 | ||
1513 | for (; matchlen < len; matchlen++) | |
1514 | if (idstr[matchlen] != ' ') | |
1515 | return false; | |
1516 | ||
1517 | return true; | |
1518 | } | |
1519 | ||
1520 | static bool quirk_matches(const struct nvme_id_ctrl *id, | |
1521 | const struct nvme_core_quirk_entry *q) | |
1522 | { | |
1523 | return q->vid == le16_to_cpu(id->vid) && | |
1524 | string_matches(id->mn, q->mn, sizeof(id->mn)) && | |
1525 | string_matches(id->fr, q->fr, sizeof(id->fr)); | |
1526 | } | |
1527 | ||
7fd8930f CH |
1528 | /* |
1529 | * Initialize the cached copies of the Identify data and various controller | |
1530 | * register in our nvme_ctrl structure. This should be called as soon as | |
1531 | * the admin queue is fully up and running. | |
1532 | */ | |
1533 | int nvme_init_identify(struct nvme_ctrl *ctrl) | |
1534 | { | |
1535 | struct nvme_id_ctrl *id; | |
1536 | u64 cap; | |
1537 | int ret, page_shift; | |
a229dbf6 | 1538 | u32 max_hw_sectors; |
c5552fde | 1539 | u8 prev_apsta; |
7fd8930f | 1540 | |
f3ca80fc CH |
1541 | ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs); |
1542 | if (ret) { | |
1b3c47c1 | 1543 | dev_err(ctrl->device, "Reading VS failed (%d)\n", ret); |
f3ca80fc CH |
1544 | return ret; |
1545 | } | |
1546 | ||
7fd8930f CH |
1547 | ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap); |
1548 | if (ret) { | |
1b3c47c1 | 1549 | dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret); |
7fd8930f CH |
1550 | return ret; |
1551 | } | |
1552 | page_shift = NVME_CAP_MPSMIN(cap) + 12; | |
1553 | ||
8ef2074d | 1554 | if (ctrl->vs >= NVME_VS(1, 1, 0)) |
f3ca80fc CH |
1555 | ctrl->subsystem = NVME_CAP_NSSRC(cap); |
1556 | ||
7fd8930f CH |
1557 | ret = nvme_identify_ctrl(ctrl, &id); |
1558 | if (ret) { | |
1b3c47c1 | 1559 | dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret); |
7fd8930f CH |
1560 | return -EIO; |
1561 | } | |
1562 | ||
bd4da3ab AL |
1563 | if (!ctrl->identified) { |
1564 | /* | |
1565 | * Check for quirks. Quirk can depend on firmware version, | |
1566 | * so, in principle, the set of quirks present can change | |
1567 | * across a reset. As a possible future enhancement, we | |
1568 | * could re-scan for quirks every time we reinitialize | |
1569 | * the device, but we'd have to make sure that the driver | |
1570 | * behaves intelligently if the quirks change. | |
1571 | */ | |
1572 | ||
1573 | int i; | |
1574 | ||
1575 | for (i = 0; i < ARRAY_SIZE(core_quirks); i++) { | |
1576 | if (quirk_matches(id, &core_quirks[i])) | |
1577 | ctrl->quirks |= core_quirks[i].quirks; | |
1578 | } | |
1579 | } | |
1580 | ||
c35e30b4 AL |
1581 | if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) { |
1582 | dev_warn(ctrl->dev, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n"); | |
1583 | ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS; | |
1584 | } | |
1585 | ||
8a9ae523 | 1586 | ctrl->oacs = le16_to_cpu(id->oacs); |
118472ab | 1587 | ctrl->vid = le16_to_cpu(id->vid); |
7fd8930f | 1588 | ctrl->oncs = le16_to_cpup(&id->oncs); |
6bf25d16 | 1589 | atomic_set(&ctrl->abort_limit, id->acl + 1); |
7fd8930f | 1590 | ctrl->vwc = id->vwc; |
931e1c22 | 1591 | ctrl->cntlid = le16_to_cpup(&id->cntlid); |
7fd8930f CH |
1592 | memcpy(ctrl->serial, id->sn, sizeof(id->sn)); |
1593 | memcpy(ctrl->model, id->mn, sizeof(id->mn)); | |
1594 | memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr)); | |
1595 | if (id->mdts) | |
a229dbf6 | 1596 | max_hw_sectors = 1 << (id->mdts + page_shift - 9); |
7fd8930f | 1597 | else |
a229dbf6 CH |
1598 | max_hw_sectors = UINT_MAX; |
1599 | ctrl->max_hw_sectors = | |
1600 | min_not_zero(ctrl->max_hw_sectors, max_hw_sectors); | |
7fd8930f | 1601 | |
da35825d | 1602 | nvme_set_queue_limits(ctrl, ctrl->admin_q); |
07bfcd09 | 1603 | ctrl->sgls = le32_to_cpu(id->sgls); |
038bd4cb | 1604 | ctrl->kas = le16_to_cpu(id->kas); |
07bfcd09 | 1605 | |
c5552fde AL |
1606 | ctrl->npss = id->npss; |
1607 | prev_apsta = ctrl->apsta; | |
c35e30b4 AL |
1608 | if (ctrl->quirks & NVME_QUIRK_NO_APST) { |
1609 | if (force_apst && id->apsta) { | |
1610 | dev_warn(ctrl->dev, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n"); | |
1611 | ctrl->apsta = 1; | |
1612 | } else { | |
1613 | ctrl->apsta = 0; | |
1614 | } | |
1615 | } else { | |
1616 | ctrl->apsta = id->apsta; | |
1617 | } | |
c5552fde AL |
1618 | memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd)); |
1619 | ||
d3d5b87d | 1620 | if (ctrl->ops->flags & NVME_F_FABRICS) { |
07bfcd09 CH |
1621 | ctrl->icdoff = le16_to_cpu(id->icdoff); |
1622 | ctrl->ioccsz = le32_to_cpu(id->ioccsz); | |
1623 | ctrl->iorcsz = le32_to_cpu(id->iorcsz); | |
1624 | ctrl->maxcmd = le16_to_cpu(id->maxcmd); | |
1625 | ||
1626 | /* | |
1627 | * In fabrics we need to verify the cntlid matches the | |
1628 | * admin connect | |
1629 | */ | |
1630 | if (ctrl->cntlid != le16_to_cpu(id->cntlid)) | |
1631 | ret = -EINVAL; | |
038bd4cb SG |
1632 | |
1633 | if (!ctrl->opts->discovery_nqn && !ctrl->kas) { | |
1634 | dev_err(ctrl->dev, | |
1635 | "keep-alive support is mandatory for fabrics\n"); | |
1636 | ret = -EINVAL; | |
1637 | } | |
07bfcd09 CH |
1638 | } else { |
1639 | ctrl->cntlid = le16_to_cpu(id->cntlid); | |
1640 | } | |
da35825d | 1641 | |
7fd8930f | 1642 | kfree(id); |
bd4da3ab | 1643 | |
c5552fde AL |
1644 | if (ctrl->apsta && !prev_apsta) |
1645 | dev_pm_qos_expose_latency_tolerance(ctrl->device); | |
1646 | else if (!ctrl->apsta && prev_apsta) | |
1647 | dev_pm_qos_hide_latency_tolerance(ctrl->device); | |
1648 | ||
1649 | nvme_configure_apst(ctrl); | |
1650 | ||
bd4da3ab | 1651 | ctrl->identified = true; |
c5552fde | 1652 | |
07bfcd09 | 1653 | return ret; |
7fd8930f | 1654 | } |
576d55d6 | 1655 | EXPORT_SYMBOL_GPL(nvme_init_identify); |
7fd8930f | 1656 | |
f3ca80fc | 1657 | static int nvme_dev_open(struct inode *inode, struct file *file) |
1673f1f0 | 1658 | { |
f3ca80fc CH |
1659 | struct nvme_ctrl *ctrl; |
1660 | int instance = iminor(inode); | |
1661 | int ret = -ENODEV; | |
1673f1f0 | 1662 | |
f3ca80fc CH |
1663 | spin_lock(&dev_list_lock); |
1664 | list_for_each_entry(ctrl, &nvme_ctrl_list, node) { | |
1665 | if (ctrl->instance != instance) | |
1666 | continue; | |
1667 | ||
1668 | if (!ctrl->admin_q) { | |
1669 | ret = -EWOULDBLOCK; | |
1670 | break; | |
1671 | } | |
1672 | if (!kref_get_unless_zero(&ctrl->kref)) | |
1673 | break; | |
1674 | file->private_data = ctrl; | |
1675 | ret = 0; | |
1676 | break; | |
1677 | } | |
1678 | spin_unlock(&dev_list_lock); | |
1679 | ||
1680 | return ret; | |
1673f1f0 CH |
1681 | } |
1682 | ||
f3ca80fc | 1683 | static int nvme_dev_release(struct inode *inode, struct file *file) |
1673f1f0 | 1684 | { |
f3ca80fc CH |
1685 | nvme_put_ctrl(file->private_data); |
1686 | return 0; | |
1687 | } | |
1688 | ||
bfd89471 CH |
1689 | static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp) |
1690 | { | |
1691 | struct nvme_ns *ns; | |
1692 | int ret; | |
1693 | ||
1694 | mutex_lock(&ctrl->namespaces_mutex); | |
1695 | if (list_empty(&ctrl->namespaces)) { | |
1696 | ret = -ENOTTY; | |
1697 | goto out_unlock; | |
1698 | } | |
1699 | ||
1700 | ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list); | |
1701 | if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) { | |
1b3c47c1 | 1702 | dev_warn(ctrl->device, |
bfd89471 CH |
1703 | "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n"); |
1704 | ret = -EINVAL; | |
1705 | goto out_unlock; | |
1706 | } | |
1707 | ||
1b3c47c1 | 1708 | dev_warn(ctrl->device, |
bfd89471 CH |
1709 | "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n"); |
1710 | kref_get(&ns->kref); | |
1711 | mutex_unlock(&ctrl->namespaces_mutex); | |
1712 | ||
1713 | ret = nvme_user_cmd(ctrl, ns, argp); | |
1714 | nvme_put_ns(ns); | |
1715 | return ret; | |
1716 | ||
1717 | out_unlock: | |
1718 | mutex_unlock(&ctrl->namespaces_mutex); | |
1719 | return ret; | |
1720 | } | |
1721 | ||
f3ca80fc CH |
1722 | static long nvme_dev_ioctl(struct file *file, unsigned int cmd, |
1723 | unsigned long arg) | |
1724 | { | |
1725 | struct nvme_ctrl *ctrl = file->private_data; | |
1726 | void __user *argp = (void __user *)arg; | |
f3ca80fc CH |
1727 | |
1728 | switch (cmd) { | |
1729 | case NVME_IOCTL_ADMIN_CMD: | |
1730 | return nvme_user_cmd(ctrl, NULL, argp); | |
1731 | case NVME_IOCTL_IO_CMD: | |
bfd89471 | 1732 | return nvme_dev_user_cmd(ctrl, argp); |
f3ca80fc | 1733 | case NVME_IOCTL_RESET: |
1b3c47c1 | 1734 | dev_warn(ctrl->device, "resetting controller\n"); |
f3ca80fc CH |
1735 | return ctrl->ops->reset_ctrl(ctrl); |
1736 | case NVME_IOCTL_SUBSYS_RESET: | |
1737 | return nvme_reset_subsystem(ctrl); | |
9ec3bb2f KB |
1738 | case NVME_IOCTL_RESCAN: |
1739 | nvme_queue_scan(ctrl); | |
1740 | return 0; | |
f3ca80fc CH |
1741 | default: |
1742 | return -ENOTTY; | |
1743 | } | |
1744 | } | |
1745 | ||
1746 | static const struct file_operations nvme_dev_fops = { | |
1747 | .owner = THIS_MODULE, | |
1748 | .open = nvme_dev_open, | |
1749 | .release = nvme_dev_release, | |
1750 | .unlocked_ioctl = nvme_dev_ioctl, | |
1751 | .compat_ioctl = nvme_dev_ioctl, | |
1752 | }; | |
1753 | ||
1754 | static ssize_t nvme_sysfs_reset(struct device *dev, | |
1755 | struct device_attribute *attr, const char *buf, | |
1756 | size_t count) | |
1757 | { | |
1758 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1759 | int ret; | |
1760 | ||
1761 | ret = ctrl->ops->reset_ctrl(ctrl); | |
1762 | if (ret < 0) | |
1763 | return ret; | |
1764 | return count; | |
1673f1f0 | 1765 | } |
f3ca80fc | 1766 | static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset); |
1673f1f0 | 1767 | |
9ec3bb2f KB |
1768 | static ssize_t nvme_sysfs_rescan(struct device *dev, |
1769 | struct device_attribute *attr, const char *buf, | |
1770 | size_t count) | |
1771 | { | |
1772 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1773 | ||
1774 | nvme_queue_scan(ctrl); | |
1775 | return count; | |
1776 | } | |
1777 | static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan); | |
1778 | ||
118472ab KB |
1779 | static ssize_t wwid_show(struct device *dev, struct device_attribute *attr, |
1780 | char *buf) | |
1781 | { | |
40267efd | 1782 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
118472ab KB |
1783 | struct nvme_ctrl *ctrl = ns->ctrl; |
1784 | int serial_len = sizeof(ctrl->serial); | |
1785 | int model_len = sizeof(ctrl->model); | |
1786 | ||
1787 | if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid))) | |
1788 | return sprintf(buf, "eui.%16phN\n", ns->uuid); | |
1789 | ||
1790 | if (memchr_inv(ns->eui, 0, sizeof(ns->eui))) | |
1791 | return sprintf(buf, "eui.%8phN\n", ns->eui); | |
1792 | ||
1793 | while (ctrl->serial[serial_len - 1] == ' ') | |
1794 | serial_len--; | |
1795 | while (ctrl->model[model_len - 1] == ' ') | |
1796 | model_len--; | |
1797 | ||
1798 | return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid, | |
1799 | serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id); | |
1800 | } | |
1801 | static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL); | |
1802 | ||
2b9b6e86 KB |
1803 | static ssize_t uuid_show(struct device *dev, struct device_attribute *attr, |
1804 | char *buf) | |
1805 | { | |
40267efd | 1806 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
2b9b6e86 KB |
1807 | return sprintf(buf, "%pU\n", ns->uuid); |
1808 | } | |
1809 | static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL); | |
1810 | ||
1811 | static ssize_t eui_show(struct device *dev, struct device_attribute *attr, | |
1812 | char *buf) | |
1813 | { | |
40267efd | 1814 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
2b9b6e86 KB |
1815 | return sprintf(buf, "%8phd\n", ns->eui); |
1816 | } | |
1817 | static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL); | |
1818 | ||
1819 | static ssize_t nsid_show(struct device *dev, struct device_attribute *attr, | |
1820 | char *buf) | |
1821 | { | |
40267efd | 1822 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
2b9b6e86 KB |
1823 | return sprintf(buf, "%d\n", ns->ns_id); |
1824 | } | |
1825 | static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL); | |
1826 | ||
1827 | static struct attribute *nvme_ns_attrs[] = { | |
118472ab | 1828 | &dev_attr_wwid.attr, |
2b9b6e86 KB |
1829 | &dev_attr_uuid.attr, |
1830 | &dev_attr_eui.attr, | |
1831 | &dev_attr_nsid.attr, | |
1832 | NULL, | |
1833 | }; | |
1834 | ||
1a353d85 | 1835 | static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj, |
2b9b6e86 KB |
1836 | struct attribute *a, int n) |
1837 | { | |
1838 | struct device *dev = container_of(kobj, struct device, kobj); | |
40267efd | 1839 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
2b9b6e86 KB |
1840 | |
1841 | if (a == &dev_attr_uuid.attr) { | |
1842 | if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid))) | |
1843 | return 0; | |
1844 | } | |
1845 | if (a == &dev_attr_eui.attr) { | |
1846 | if (!memchr_inv(ns->eui, 0, sizeof(ns->eui))) | |
1847 | return 0; | |
1848 | } | |
1849 | return a->mode; | |
1850 | } | |
1851 | ||
1852 | static const struct attribute_group nvme_ns_attr_group = { | |
1853 | .attrs = nvme_ns_attrs, | |
1a353d85 | 1854 | .is_visible = nvme_ns_attrs_are_visible, |
2b9b6e86 KB |
1855 | }; |
1856 | ||
931e1c22 | 1857 | #define nvme_show_str_function(field) \ |
779ff756 KB |
1858 | static ssize_t field##_show(struct device *dev, \ |
1859 | struct device_attribute *attr, char *buf) \ | |
1860 | { \ | |
1861 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ | |
1862 | return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \ | |
1863 | } \ | |
1864 | static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); | |
1865 | ||
931e1c22 ML |
1866 | #define nvme_show_int_function(field) \ |
1867 | static ssize_t field##_show(struct device *dev, \ | |
1868 | struct device_attribute *attr, char *buf) \ | |
1869 | { \ | |
1870 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ | |
1871 | return sprintf(buf, "%d\n", ctrl->field); \ | |
1872 | } \ | |
1873 | static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); | |
1874 | ||
1875 | nvme_show_str_function(model); | |
1876 | nvme_show_str_function(serial); | |
1877 | nvme_show_str_function(firmware_rev); | |
1878 | nvme_show_int_function(cntlid); | |
779ff756 | 1879 | |
1a353d85 ML |
1880 | static ssize_t nvme_sysfs_delete(struct device *dev, |
1881 | struct device_attribute *attr, const char *buf, | |
1882 | size_t count) | |
1883 | { | |
1884 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1885 | ||
1886 | if (device_remove_file_self(dev, attr)) | |
1887 | ctrl->ops->delete_ctrl(ctrl); | |
1888 | return count; | |
1889 | } | |
1890 | static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete); | |
1891 | ||
1892 | static ssize_t nvme_sysfs_show_transport(struct device *dev, | |
1893 | struct device_attribute *attr, | |
1894 | char *buf) | |
1895 | { | |
1896 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1897 | ||
1898 | return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name); | |
1899 | } | |
1900 | static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL); | |
1901 | ||
8432bdb2 SG |
1902 | static ssize_t nvme_sysfs_show_state(struct device *dev, |
1903 | struct device_attribute *attr, | |
1904 | char *buf) | |
1905 | { | |
1906 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1907 | static const char *const state_name[] = { | |
1908 | [NVME_CTRL_NEW] = "new", | |
1909 | [NVME_CTRL_LIVE] = "live", | |
1910 | [NVME_CTRL_RESETTING] = "resetting", | |
1911 | [NVME_CTRL_RECONNECTING]= "reconnecting", | |
1912 | [NVME_CTRL_DELETING] = "deleting", | |
1913 | [NVME_CTRL_DEAD] = "dead", | |
1914 | }; | |
1915 | ||
1916 | if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) && | |
1917 | state_name[ctrl->state]) | |
1918 | return sprintf(buf, "%s\n", state_name[ctrl->state]); | |
1919 | ||
1920 | return sprintf(buf, "unknown state\n"); | |
1921 | } | |
1922 | ||
1923 | static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL); | |
1924 | ||
1a353d85 ML |
1925 | static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev, |
1926 | struct device_attribute *attr, | |
1927 | char *buf) | |
1928 | { | |
1929 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1930 | ||
1931 | return snprintf(buf, PAGE_SIZE, "%s\n", | |
1932 | ctrl->ops->get_subsysnqn(ctrl)); | |
1933 | } | |
1934 | static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL); | |
1935 | ||
1936 | static ssize_t nvme_sysfs_show_address(struct device *dev, | |
1937 | struct device_attribute *attr, | |
1938 | char *buf) | |
1939 | { | |
1940 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1941 | ||
1942 | return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE); | |
1943 | } | |
1944 | static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL); | |
1945 | ||
779ff756 KB |
1946 | static struct attribute *nvme_dev_attrs[] = { |
1947 | &dev_attr_reset_controller.attr, | |
9ec3bb2f | 1948 | &dev_attr_rescan_controller.attr, |
779ff756 KB |
1949 | &dev_attr_model.attr, |
1950 | &dev_attr_serial.attr, | |
1951 | &dev_attr_firmware_rev.attr, | |
931e1c22 | 1952 | &dev_attr_cntlid.attr, |
1a353d85 ML |
1953 | &dev_attr_delete_controller.attr, |
1954 | &dev_attr_transport.attr, | |
1955 | &dev_attr_subsysnqn.attr, | |
1956 | &dev_attr_address.attr, | |
8432bdb2 | 1957 | &dev_attr_state.attr, |
779ff756 KB |
1958 | NULL |
1959 | }; | |
1960 | ||
1a353d85 ML |
1961 | #define CHECK_ATTR(ctrl, a, name) \ |
1962 | if ((a) == &dev_attr_##name.attr && \ | |
1963 | !(ctrl)->ops->get_##name) \ | |
1964 | return 0 | |
1965 | ||
1966 | static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj, | |
1967 | struct attribute *a, int n) | |
1968 | { | |
1969 | struct device *dev = container_of(kobj, struct device, kobj); | |
1970 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1971 | ||
1972 | if (a == &dev_attr_delete_controller.attr) { | |
1973 | if (!ctrl->ops->delete_ctrl) | |
1974 | return 0; | |
1975 | } | |
1976 | ||
1977 | CHECK_ATTR(ctrl, a, subsysnqn); | |
1978 | CHECK_ATTR(ctrl, a, address); | |
1979 | ||
1980 | return a->mode; | |
1981 | } | |
1982 | ||
779ff756 | 1983 | static struct attribute_group nvme_dev_attrs_group = { |
1a353d85 ML |
1984 | .attrs = nvme_dev_attrs, |
1985 | .is_visible = nvme_dev_attrs_are_visible, | |
779ff756 KB |
1986 | }; |
1987 | ||
1988 | static const struct attribute_group *nvme_dev_attr_groups[] = { | |
1989 | &nvme_dev_attrs_group, | |
1990 | NULL, | |
1991 | }; | |
1992 | ||
5bae7f73 CH |
1993 | static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) |
1994 | { | |
1995 | struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); | |
1996 | struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); | |
1997 | ||
1998 | return nsa->ns_id - nsb->ns_id; | |
1999 | } | |
2000 | ||
32f0c4af | 2001 | static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
5bae7f73 | 2002 | { |
32f0c4af | 2003 | struct nvme_ns *ns, *ret = NULL; |
69d3b8ac | 2004 | |
32f0c4af | 2005 | mutex_lock(&ctrl->namespaces_mutex); |
5bae7f73 | 2006 | list_for_each_entry(ns, &ctrl->namespaces, list) { |
32f0c4af KB |
2007 | if (ns->ns_id == nsid) { |
2008 | kref_get(&ns->kref); | |
2009 | ret = ns; | |
2010 | break; | |
2011 | } | |
5bae7f73 CH |
2012 | if (ns->ns_id > nsid) |
2013 | break; | |
2014 | } | |
32f0c4af KB |
2015 | mutex_unlock(&ctrl->namespaces_mutex); |
2016 | return ret; | |
5bae7f73 CH |
2017 | } |
2018 | ||
2019 | static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid) | |
2020 | { | |
2021 | struct nvme_ns *ns; | |
2022 | struct gendisk *disk; | |
ac81bfa9 MB |
2023 | struct nvme_id_ns *id; |
2024 | char disk_name[DISK_NAME_LEN]; | |
5bae7f73 CH |
2025 | int node = dev_to_node(ctrl->dev); |
2026 | ||
2027 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); | |
2028 | if (!ns) | |
2029 | return; | |
2030 | ||
075790eb KB |
2031 | ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL); |
2032 | if (ns->instance < 0) | |
2033 | goto out_free_ns; | |
2034 | ||
5bae7f73 CH |
2035 | ns->queue = blk_mq_init_queue(ctrl->tagset); |
2036 | if (IS_ERR(ns->queue)) | |
075790eb | 2037 | goto out_release_instance; |
5bae7f73 CH |
2038 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); |
2039 | ns->queue->queuedata = ns; | |
2040 | ns->ctrl = ctrl; | |
2041 | ||
5bae7f73 CH |
2042 | kref_init(&ns->kref); |
2043 | ns->ns_id = nsid; | |
5bae7f73 | 2044 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ |
5bae7f73 CH |
2045 | |
2046 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); | |
da35825d | 2047 | nvme_set_queue_limits(ctrl, ns->queue); |
5bae7f73 | 2048 | |
ac81bfa9 | 2049 | sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance); |
5bae7f73 | 2050 | |
ac81bfa9 MB |
2051 | if (nvme_revalidate_ns(ns, &id)) |
2052 | goto out_free_queue; | |
2053 | ||
3dc87dd0 MB |
2054 | if (nvme_nvm_ns_supported(ns, id) && |
2055 | nvme_nvm_register(ns, disk_name, node)) { | |
2056 | dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__); | |
2057 | goto out_free_id; | |
2058 | } | |
ac81bfa9 | 2059 | |
3dc87dd0 MB |
2060 | disk = alloc_disk_node(0, node); |
2061 | if (!disk) | |
2062 | goto out_free_id; | |
ac81bfa9 | 2063 | |
3dc87dd0 MB |
2064 | disk->fops = &nvme_fops; |
2065 | disk->private_data = ns; | |
2066 | disk->queue = ns->queue; | |
2067 | disk->flags = GENHD_FL_EXT_DEVT; | |
2068 | memcpy(disk->disk_name, disk_name, DISK_NAME_LEN); | |
2069 | ns->disk = disk; | |
2070 | ||
2071 | __nvme_revalidate_disk(disk, id); | |
5bae7f73 | 2072 | |
32f0c4af KB |
2073 | mutex_lock(&ctrl->namespaces_mutex); |
2074 | list_add_tail(&ns->list, &ctrl->namespaces); | |
2075 | mutex_unlock(&ctrl->namespaces_mutex); | |
2076 | ||
5bae7f73 | 2077 | kref_get(&ctrl->kref); |
ac81bfa9 MB |
2078 | |
2079 | kfree(id); | |
2080 | ||
0d52c756 | 2081 | device_add_disk(ctrl->device, ns->disk); |
2b9b6e86 KB |
2082 | if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj, |
2083 | &nvme_ns_attr_group)) | |
2084 | pr_warn("%s: failed to create sysfs group for identification\n", | |
2085 | ns->disk->disk_name); | |
3dc87dd0 MB |
2086 | if (ns->ndev && nvme_nvm_register_sysfs(ns)) |
2087 | pr_warn("%s: failed to register lightnvm sysfs group for identification\n", | |
2088 | ns->disk->disk_name); | |
5bae7f73 | 2089 | return; |
ac81bfa9 MB |
2090 | out_free_id: |
2091 | kfree(id); | |
5bae7f73 CH |
2092 | out_free_queue: |
2093 | blk_cleanup_queue(ns->queue); | |
075790eb KB |
2094 | out_release_instance: |
2095 | ida_simple_remove(&ctrl->ns_ida, ns->instance); | |
5bae7f73 CH |
2096 | out_free_ns: |
2097 | kfree(ns); | |
2098 | } | |
2099 | ||
2100 | static void nvme_ns_remove(struct nvme_ns *ns) | |
2101 | { | |
646017a6 KB |
2102 | if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags)) |
2103 | return; | |
69d3b8ac | 2104 | |
b0b4e09c | 2105 | if (ns->disk && ns->disk->flags & GENHD_FL_UP) { |
5bae7f73 CH |
2106 | if (blk_get_integrity(ns->disk)) |
2107 | blk_integrity_unregister(ns->disk); | |
2b9b6e86 KB |
2108 | sysfs_remove_group(&disk_to_dev(ns->disk)->kobj, |
2109 | &nvme_ns_attr_group); | |
3dc87dd0 MB |
2110 | if (ns->ndev) |
2111 | nvme_nvm_unregister_sysfs(ns); | |
5bae7f73 | 2112 | del_gendisk(ns->disk); |
5bae7f73 CH |
2113 | blk_cleanup_queue(ns->queue); |
2114 | } | |
32f0c4af KB |
2115 | |
2116 | mutex_lock(&ns->ctrl->namespaces_mutex); | |
5bae7f73 | 2117 | list_del_init(&ns->list); |
32f0c4af KB |
2118 | mutex_unlock(&ns->ctrl->namespaces_mutex); |
2119 | ||
5bae7f73 CH |
2120 | nvme_put_ns(ns); |
2121 | } | |
2122 | ||
540c801c KB |
2123 | static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
2124 | { | |
2125 | struct nvme_ns *ns; | |
2126 | ||
32f0c4af | 2127 | ns = nvme_find_get_ns(ctrl, nsid); |
540c801c | 2128 | if (ns) { |
b0b4e09c | 2129 | if (ns->disk && revalidate_disk(ns->disk)) |
540c801c | 2130 | nvme_ns_remove(ns); |
32f0c4af | 2131 | nvme_put_ns(ns); |
540c801c KB |
2132 | } else |
2133 | nvme_alloc_ns(ctrl, nsid); | |
2134 | } | |
2135 | ||
47b0e50a SB |
2136 | static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl, |
2137 | unsigned nsid) | |
2138 | { | |
2139 | struct nvme_ns *ns, *next; | |
2140 | ||
2141 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) { | |
2142 | if (ns->ns_id > nsid) | |
2143 | nvme_ns_remove(ns); | |
2144 | } | |
2145 | } | |
2146 | ||
540c801c KB |
2147 | static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn) |
2148 | { | |
2149 | struct nvme_ns *ns; | |
2150 | __le32 *ns_list; | |
2151 | unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024); | |
2152 | int ret = 0; | |
2153 | ||
2154 | ns_list = kzalloc(0x1000, GFP_KERNEL); | |
2155 | if (!ns_list) | |
2156 | return -ENOMEM; | |
2157 | ||
2158 | for (i = 0; i < num_lists; i++) { | |
2159 | ret = nvme_identify_ns_list(ctrl, prev, ns_list); | |
2160 | if (ret) | |
47b0e50a | 2161 | goto free; |
540c801c KB |
2162 | |
2163 | for (j = 0; j < min(nn, 1024U); j++) { | |
2164 | nsid = le32_to_cpu(ns_list[j]); | |
2165 | if (!nsid) | |
2166 | goto out; | |
2167 | ||
2168 | nvme_validate_ns(ctrl, nsid); | |
2169 | ||
2170 | while (++prev < nsid) { | |
32f0c4af KB |
2171 | ns = nvme_find_get_ns(ctrl, prev); |
2172 | if (ns) { | |
540c801c | 2173 | nvme_ns_remove(ns); |
32f0c4af KB |
2174 | nvme_put_ns(ns); |
2175 | } | |
540c801c KB |
2176 | } |
2177 | } | |
2178 | nn -= j; | |
2179 | } | |
2180 | out: | |
47b0e50a SB |
2181 | nvme_remove_invalid_namespaces(ctrl, prev); |
2182 | free: | |
540c801c KB |
2183 | kfree(ns_list); |
2184 | return ret; | |
2185 | } | |
2186 | ||
5955be21 | 2187 | static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn) |
5bae7f73 | 2188 | { |
5bae7f73 CH |
2189 | unsigned i; |
2190 | ||
540c801c KB |
2191 | for (i = 1; i <= nn; i++) |
2192 | nvme_validate_ns(ctrl, i); | |
2193 | ||
47b0e50a | 2194 | nvme_remove_invalid_namespaces(ctrl, nn); |
5bae7f73 CH |
2195 | } |
2196 | ||
5955be21 | 2197 | static void nvme_scan_work(struct work_struct *work) |
5bae7f73 | 2198 | { |
5955be21 CH |
2199 | struct nvme_ctrl *ctrl = |
2200 | container_of(work, struct nvme_ctrl, scan_work); | |
5bae7f73 | 2201 | struct nvme_id_ctrl *id; |
540c801c | 2202 | unsigned nn; |
5bae7f73 | 2203 | |
5955be21 CH |
2204 | if (ctrl->state != NVME_CTRL_LIVE) |
2205 | return; | |
2206 | ||
5bae7f73 CH |
2207 | if (nvme_identify_ctrl(ctrl, &id)) |
2208 | return; | |
540c801c KB |
2209 | |
2210 | nn = le32_to_cpu(id->nn); | |
8ef2074d | 2211 | if (ctrl->vs >= NVME_VS(1, 1, 0) && |
540c801c KB |
2212 | !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) { |
2213 | if (!nvme_scan_ns_list(ctrl, nn)) | |
2214 | goto done; | |
2215 | } | |
5955be21 | 2216 | nvme_scan_ns_sequential(ctrl, nn); |
540c801c | 2217 | done: |
32f0c4af | 2218 | mutex_lock(&ctrl->namespaces_mutex); |
540c801c | 2219 | list_sort(NULL, &ctrl->namespaces, ns_cmp); |
69d3b8ac | 2220 | mutex_unlock(&ctrl->namespaces_mutex); |
5bae7f73 CH |
2221 | kfree(id); |
2222 | } | |
5955be21 CH |
2223 | |
2224 | void nvme_queue_scan(struct nvme_ctrl *ctrl) | |
2225 | { | |
2226 | /* | |
2227 | * Do not queue new scan work when a controller is reset during | |
2228 | * removal. | |
2229 | */ | |
2230 | if (ctrl->state == NVME_CTRL_LIVE) | |
2231 | schedule_work(&ctrl->scan_work); | |
2232 | } | |
2233 | EXPORT_SYMBOL_GPL(nvme_queue_scan); | |
5bae7f73 | 2234 | |
32f0c4af KB |
2235 | /* |
2236 | * This function iterates the namespace list unlocked to allow recovery from | |
2237 | * controller failure. It is up to the caller to ensure the namespace list is | |
2238 | * not modified by scan work while this function is executing. | |
2239 | */ | |
5bae7f73 CH |
2240 | void nvme_remove_namespaces(struct nvme_ctrl *ctrl) |
2241 | { | |
2242 | struct nvme_ns *ns, *next; | |
2243 | ||
0ff9d4e1 KB |
2244 | /* |
2245 | * The dead states indicates the controller was not gracefully | |
2246 | * disconnected. In that case, we won't be able to flush any data while | |
2247 | * removing the namespaces' disks; fail all the queues now to avoid | |
2248 | * potentially having to clean up the failed sync later. | |
2249 | */ | |
2250 | if (ctrl->state == NVME_CTRL_DEAD) | |
2251 | nvme_kill_queues(ctrl); | |
2252 | ||
5bae7f73 CH |
2253 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) |
2254 | nvme_ns_remove(ns); | |
2255 | } | |
576d55d6 | 2256 | EXPORT_SYMBOL_GPL(nvme_remove_namespaces); |
5bae7f73 | 2257 | |
f866fc42 CH |
2258 | static void nvme_async_event_work(struct work_struct *work) |
2259 | { | |
2260 | struct nvme_ctrl *ctrl = | |
2261 | container_of(work, struct nvme_ctrl, async_event_work); | |
2262 | ||
2263 | spin_lock_irq(&ctrl->lock); | |
2264 | while (ctrl->event_limit > 0) { | |
2265 | int aer_idx = --ctrl->event_limit; | |
2266 | ||
2267 | spin_unlock_irq(&ctrl->lock); | |
2268 | ctrl->ops->submit_async_event(ctrl, aer_idx); | |
2269 | spin_lock_irq(&ctrl->lock); | |
2270 | } | |
2271 | spin_unlock_irq(&ctrl->lock); | |
2272 | } | |
2273 | ||
7bf58533 CH |
2274 | void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, |
2275 | union nvme_result *res) | |
f866fc42 | 2276 | { |
7bf58533 CH |
2277 | u32 result = le32_to_cpu(res->u32); |
2278 | bool done = true; | |
f866fc42 | 2279 | |
7bf58533 CH |
2280 | switch (le16_to_cpu(status) >> 1) { |
2281 | case NVME_SC_SUCCESS: | |
2282 | done = false; | |
2283 | /*FALLTHRU*/ | |
2284 | case NVME_SC_ABORT_REQ: | |
f866fc42 CH |
2285 | ++ctrl->event_limit; |
2286 | schedule_work(&ctrl->async_event_work); | |
7bf58533 CH |
2287 | break; |
2288 | default: | |
2289 | break; | |
f866fc42 CH |
2290 | } |
2291 | ||
7bf58533 | 2292 | if (done) |
f866fc42 CH |
2293 | return; |
2294 | ||
2295 | switch (result & 0xff07) { | |
2296 | case NVME_AER_NOTICE_NS_CHANGED: | |
2297 | dev_info(ctrl->device, "rescanning\n"); | |
2298 | nvme_queue_scan(ctrl); | |
2299 | break; | |
2300 | default: | |
2301 | dev_warn(ctrl->device, "async event result %08x\n", result); | |
2302 | } | |
2303 | } | |
2304 | EXPORT_SYMBOL_GPL(nvme_complete_async_event); | |
2305 | ||
2306 | void nvme_queue_async_events(struct nvme_ctrl *ctrl) | |
2307 | { | |
2308 | ctrl->event_limit = NVME_NR_AERS; | |
2309 | schedule_work(&ctrl->async_event_work); | |
2310 | } | |
2311 | EXPORT_SYMBOL_GPL(nvme_queue_async_events); | |
2312 | ||
f3ca80fc CH |
2313 | static DEFINE_IDA(nvme_instance_ida); |
2314 | ||
2315 | static int nvme_set_instance(struct nvme_ctrl *ctrl) | |
2316 | { | |
2317 | int instance, error; | |
2318 | ||
2319 | do { | |
2320 | if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL)) | |
2321 | return -ENODEV; | |
2322 | ||
2323 | spin_lock(&dev_list_lock); | |
2324 | error = ida_get_new(&nvme_instance_ida, &instance); | |
2325 | spin_unlock(&dev_list_lock); | |
2326 | } while (error == -EAGAIN); | |
2327 | ||
2328 | if (error) | |
2329 | return -ENODEV; | |
2330 | ||
2331 | ctrl->instance = instance; | |
2332 | return 0; | |
2333 | } | |
2334 | ||
2335 | static void nvme_release_instance(struct nvme_ctrl *ctrl) | |
2336 | { | |
2337 | spin_lock(&dev_list_lock); | |
2338 | ida_remove(&nvme_instance_ida, ctrl->instance); | |
2339 | spin_unlock(&dev_list_lock); | |
2340 | } | |
2341 | ||
53029b04 | 2342 | void nvme_uninit_ctrl(struct nvme_ctrl *ctrl) |
576d55d6 | 2343 | { |
f866fc42 | 2344 | flush_work(&ctrl->async_event_work); |
5955be21 CH |
2345 | flush_work(&ctrl->scan_work); |
2346 | nvme_remove_namespaces(ctrl); | |
2347 | ||
53029b04 | 2348 | device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance)); |
f3ca80fc CH |
2349 | |
2350 | spin_lock(&dev_list_lock); | |
2351 | list_del(&ctrl->node); | |
2352 | spin_unlock(&dev_list_lock); | |
53029b04 | 2353 | } |
576d55d6 | 2354 | EXPORT_SYMBOL_GPL(nvme_uninit_ctrl); |
53029b04 KB |
2355 | |
2356 | static void nvme_free_ctrl(struct kref *kref) | |
2357 | { | |
2358 | struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref); | |
f3ca80fc CH |
2359 | |
2360 | put_device(ctrl->device); | |
2361 | nvme_release_instance(ctrl); | |
075790eb | 2362 | ida_destroy(&ctrl->ns_ida); |
f3ca80fc CH |
2363 | |
2364 | ctrl->ops->free_ctrl(ctrl); | |
2365 | } | |
2366 | ||
2367 | void nvme_put_ctrl(struct nvme_ctrl *ctrl) | |
2368 | { | |
2369 | kref_put(&ctrl->kref, nvme_free_ctrl); | |
2370 | } | |
576d55d6 | 2371 | EXPORT_SYMBOL_GPL(nvme_put_ctrl); |
f3ca80fc CH |
2372 | |
2373 | /* | |
2374 | * Initialize a NVMe controller structures. This needs to be called during | |
2375 | * earliest initialization so that we have the initialized structured around | |
2376 | * during probing. | |
2377 | */ | |
2378 | int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, | |
2379 | const struct nvme_ctrl_ops *ops, unsigned long quirks) | |
2380 | { | |
2381 | int ret; | |
2382 | ||
bb8d261e CH |
2383 | ctrl->state = NVME_CTRL_NEW; |
2384 | spin_lock_init(&ctrl->lock); | |
f3ca80fc | 2385 | INIT_LIST_HEAD(&ctrl->namespaces); |
69d3b8ac | 2386 | mutex_init(&ctrl->namespaces_mutex); |
f3ca80fc CH |
2387 | kref_init(&ctrl->kref); |
2388 | ctrl->dev = dev; | |
2389 | ctrl->ops = ops; | |
2390 | ctrl->quirks = quirks; | |
5955be21 | 2391 | INIT_WORK(&ctrl->scan_work, nvme_scan_work); |
f866fc42 | 2392 | INIT_WORK(&ctrl->async_event_work, nvme_async_event_work); |
f3ca80fc CH |
2393 | |
2394 | ret = nvme_set_instance(ctrl); | |
2395 | if (ret) | |
2396 | goto out; | |
2397 | ||
779ff756 | 2398 | ctrl->device = device_create_with_groups(nvme_class, ctrl->dev, |
f3ca80fc | 2399 | MKDEV(nvme_char_major, ctrl->instance), |
f4f0f63e | 2400 | ctrl, nvme_dev_attr_groups, |
779ff756 | 2401 | "nvme%d", ctrl->instance); |
f3ca80fc CH |
2402 | if (IS_ERR(ctrl->device)) { |
2403 | ret = PTR_ERR(ctrl->device); | |
2404 | goto out_release_instance; | |
2405 | } | |
2406 | get_device(ctrl->device); | |
075790eb | 2407 | ida_init(&ctrl->ns_ida); |
f3ca80fc | 2408 | |
f3ca80fc CH |
2409 | spin_lock(&dev_list_lock); |
2410 | list_add_tail(&ctrl->node, &nvme_ctrl_list); | |
2411 | spin_unlock(&dev_list_lock); | |
2412 | ||
c5552fde AL |
2413 | /* |
2414 | * Initialize latency tolerance controls. The sysfs files won't | |
2415 | * be visible to userspace unless the device actually supports APST. | |
2416 | */ | |
2417 | ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance; | |
2418 | dev_pm_qos_update_user_latency_tolerance(ctrl->device, | |
2419 | min(default_ps_max_latency_us, (unsigned long)S32_MAX)); | |
2420 | ||
f3ca80fc | 2421 | return 0; |
f3ca80fc CH |
2422 | out_release_instance: |
2423 | nvme_release_instance(ctrl); | |
2424 | out: | |
2425 | return ret; | |
2426 | } | |
576d55d6 | 2427 | EXPORT_SYMBOL_GPL(nvme_init_ctrl); |
f3ca80fc | 2428 | |
69d9a99c KB |
2429 | /** |
2430 | * nvme_kill_queues(): Ends all namespace queues | |
2431 | * @ctrl: the dead controller that needs to end | |
2432 | * | |
2433 | * Call this function when the driver determines it is unable to get the | |
2434 | * controller in a state capable of servicing IO. | |
2435 | */ | |
2436 | void nvme_kill_queues(struct nvme_ctrl *ctrl) | |
2437 | { | |
2438 | struct nvme_ns *ns; | |
2439 | ||
32f0c4af KB |
2440 | mutex_lock(&ctrl->namespaces_mutex); |
2441 | list_for_each_entry(ns, &ctrl->namespaces, list) { | |
69d9a99c KB |
2442 | /* |
2443 | * Revalidating a dead namespace sets capacity to 0. This will | |
2444 | * end buffered writers dirtying pages that can't be synced. | |
2445 | */ | |
f33447b9 KB |
2446 | if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags)) |
2447 | continue; | |
2448 | revalidate_disk(ns->disk); | |
69d9a99c | 2449 | blk_set_queue_dying(ns->queue); |
806f026f ML |
2450 | |
2451 | /* | |
2452 | * Forcibly start all queues to avoid having stuck requests. | |
2453 | * Note that we must ensure the queues are not stopped | |
2454 | * when the final removal happens. | |
2455 | */ | |
2456 | blk_mq_start_hw_queues(ns->queue); | |
986f75c8 ML |
2457 | |
2458 | /* draining requests in requeue list */ | |
2459 | blk_mq_kick_requeue_list(ns->queue); | |
69d9a99c | 2460 | } |
32f0c4af | 2461 | mutex_unlock(&ctrl->namespaces_mutex); |
69d9a99c | 2462 | } |
237045fc | 2463 | EXPORT_SYMBOL_GPL(nvme_kill_queues); |
69d9a99c | 2464 | |
302ad8cc KB |
2465 | void nvme_unfreeze(struct nvme_ctrl *ctrl) |
2466 | { | |
2467 | struct nvme_ns *ns; | |
2468 | ||
2469 | mutex_lock(&ctrl->namespaces_mutex); | |
2470 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
2471 | blk_mq_unfreeze_queue(ns->queue); | |
2472 | mutex_unlock(&ctrl->namespaces_mutex); | |
2473 | } | |
2474 | EXPORT_SYMBOL_GPL(nvme_unfreeze); | |
2475 | ||
2476 | void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout) | |
2477 | { | |
2478 | struct nvme_ns *ns; | |
2479 | ||
2480 | mutex_lock(&ctrl->namespaces_mutex); | |
2481 | list_for_each_entry(ns, &ctrl->namespaces, list) { | |
2482 | timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout); | |
2483 | if (timeout <= 0) | |
2484 | break; | |
2485 | } | |
2486 | mutex_unlock(&ctrl->namespaces_mutex); | |
2487 | } | |
2488 | EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout); | |
2489 | ||
2490 | void nvme_wait_freeze(struct nvme_ctrl *ctrl) | |
2491 | { | |
2492 | struct nvme_ns *ns; | |
2493 | ||
2494 | mutex_lock(&ctrl->namespaces_mutex); | |
2495 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
2496 | blk_mq_freeze_queue_wait(ns->queue); | |
2497 | mutex_unlock(&ctrl->namespaces_mutex); | |
2498 | } | |
2499 | EXPORT_SYMBOL_GPL(nvme_wait_freeze); | |
2500 | ||
2501 | void nvme_start_freeze(struct nvme_ctrl *ctrl) | |
2502 | { | |
2503 | struct nvme_ns *ns; | |
2504 | ||
2505 | mutex_lock(&ctrl->namespaces_mutex); | |
2506 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
1671d522 | 2507 | blk_freeze_queue_start(ns->queue); |
302ad8cc KB |
2508 | mutex_unlock(&ctrl->namespaces_mutex); |
2509 | } | |
2510 | EXPORT_SYMBOL_GPL(nvme_start_freeze); | |
2511 | ||
25646264 | 2512 | void nvme_stop_queues(struct nvme_ctrl *ctrl) |
363c9aac SG |
2513 | { |
2514 | struct nvme_ns *ns; | |
2515 | ||
32f0c4af | 2516 | mutex_lock(&ctrl->namespaces_mutex); |
a6eaa884 | 2517 | list_for_each_entry(ns, &ctrl->namespaces, list) |
3174dd33 | 2518 | blk_mq_quiesce_queue(ns->queue); |
32f0c4af | 2519 | mutex_unlock(&ctrl->namespaces_mutex); |
363c9aac | 2520 | } |
576d55d6 | 2521 | EXPORT_SYMBOL_GPL(nvme_stop_queues); |
363c9aac | 2522 | |
25646264 | 2523 | void nvme_start_queues(struct nvme_ctrl *ctrl) |
363c9aac SG |
2524 | { |
2525 | struct nvme_ns *ns; | |
2526 | ||
32f0c4af KB |
2527 | mutex_lock(&ctrl->namespaces_mutex); |
2528 | list_for_each_entry(ns, &ctrl->namespaces, list) { | |
363c9aac SG |
2529 | blk_mq_start_stopped_hw_queues(ns->queue, true); |
2530 | blk_mq_kick_requeue_list(ns->queue); | |
2531 | } | |
32f0c4af | 2532 | mutex_unlock(&ctrl->namespaces_mutex); |
363c9aac | 2533 | } |
576d55d6 | 2534 | EXPORT_SYMBOL_GPL(nvme_start_queues); |
363c9aac | 2535 | |
5bae7f73 CH |
2536 | int __init nvme_core_init(void) |
2537 | { | |
2538 | int result; | |
2539 | ||
f3ca80fc CH |
2540 | result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme", |
2541 | &nvme_dev_fops); | |
2542 | if (result < 0) | |
b09dcf58 | 2543 | return result; |
f3ca80fc CH |
2544 | else if (result > 0) |
2545 | nvme_char_major = result; | |
2546 | ||
2547 | nvme_class = class_create(THIS_MODULE, "nvme"); | |
2548 | if (IS_ERR(nvme_class)) { | |
2549 | result = PTR_ERR(nvme_class); | |
2550 | goto unregister_chrdev; | |
2551 | } | |
2552 | ||
5bae7f73 | 2553 | return 0; |
f3ca80fc CH |
2554 | |
2555 | unregister_chrdev: | |
2556 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
f3ca80fc | 2557 | return result; |
5bae7f73 CH |
2558 | } |
2559 | ||
2560 | void nvme_core_exit(void) | |
2561 | { | |
f3ca80fc CH |
2562 | class_destroy(nvme_class); |
2563 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
5bae7f73 | 2564 | } |
576d55d6 ML |
2565 | |
2566 | MODULE_LICENSE("GPL"); | |
2567 | MODULE_VERSION("1.0"); | |
2568 | module_init(nvme_core_init); | |
2569 | module_exit(nvme_core_exit); |