]>
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 | 30 | #include <asm/unaligned.h> |
21d34711 CH |
31 | |
32 | #include "nvme.h" | |
038bd4cb | 33 | #include "fabrics.h" |
21d34711 | 34 | |
f3ca80fc CH |
35 | #define NVME_MINORS (1U << MINORBITS) |
36 | ||
8ae4e447 MO |
37 | unsigned int admin_timeout = 60; |
38 | module_param(admin_timeout, uint, 0644); | |
ba0ba7d3 | 39 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); |
576d55d6 | 40 | EXPORT_SYMBOL_GPL(admin_timeout); |
ba0ba7d3 | 41 | |
8ae4e447 MO |
42 | unsigned int nvme_io_timeout = 30; |
43 | module_param_named(io_timeout, nvme_io_timeout, uint, 0644); | |
ba0ba7d3 | 44 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); |
576d55d6 | 45 | EXPORT_SYMBOL_GPL(nvme_io_timeout); |
ba0ba7d3 | 46 | |
b3b1b0b0 | 47 | static unsigned char shutdown_timeout = 5; |
ba0ba7d3 ML |
48 | module_param(shutdown_timeout, byte, 0644); |
49 | MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); | |
50 | ||
44e44b29 CH |
51 | static u8 nvme_max_retries = 5; |
52 | module_param_named(max_retries, nvme_max_retries, byte, 0644); | |
f80ec966 | 53 | MODULE_PARM_DESC(max_retries, "max number of retries a command may have"); |
5bae7f73 | 54 | |
9947d6a0 | 55 | static unsigned long default_ps_max_latency_us = 100000; |
c5552fde AL |
56 | module_param(default_ps_max_latency_us, ulong, 0644); |
57 | MODULE_PARM_DESC(default_ps_max_latency_us, | |
58 | "max power saving latency for new devices; use PM QOS to change per device"); | |
59 | ||
c35e30b4 AL |
60 | static bool force_apst; |
61 | module_param(force_apst, bool, 0644); | |
62 | MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off"); | |
63 | ||
f5d11840 JA |
64 | static bool streams; |
65 | module_param(streams, bool, 0644); | |
66 | MODULE_PARM_DESC(streams, "turn on support for Streams write directives"); | |
67 | ||
9a6327d2 SG |
68 | struct workqueue_struct *nvme_wq; |
69 | EXPORT_SYMBOL_GPL(nvme_wq); | |
70 | ||
9843f685 | 71 | static DEFINE_IDA(nvme_instance_ida); |
a6a5149b | 72 | static dev_t nvme_chr_devt; |
f3ca80fc CH |
73 | static struct class *nvme_class; |
74 | ||
84fef62d KB |
75 | static void nvme_ns_remove(struct nvme_ns *ns); |
76 | static int nvme_revalidate_disk(struct gendisk *disk); | |
77 | ||
b6dccf7f AD |
78 | static __le32 nvme_get_log_dw10(u8 lid, size_t size) |
79 | { | |
80 | return cpu_to_le32((((size / 4) - 1) << 16) | lid); | |
81 | } | |
82 | ||
d86c4d8e CH |
83 | int nvme_reset_ctrl(struct nvme_ctrl *ctrl) |
84 | { | |
85 | if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) | |
86 | return -EBUSY; | |
87 | if (!queue_work(nvme_wq, &ctrl->reset_work)) | |
88 | return -EBUSY; | |
89 | return 0; | |
90 | } | |
91 | EXPORT_SYMBOL_GPL(nvme_reset_ctrl); | |
92 | ||
93 | static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl) | |
94 | { | |
95 | int ret; | |
96 | ||
97 | ret = nvme_reset_ctrl(ctrl); | |
98 | if (!ret) | |
99 | flush_work(&ctrl->reset_work); | |
100 | return ret; | |
101 | } | |
102 | ||
c5017e85 CH |
103 | static void nvme_delete_ctrl_work(struct work_struct *work) |
104 | { | |
105 | struct nvme_ctrl *ctrl = | |
106 | container_of(work, struct nvme_ctrl, delete_work); | |
107 | ||
4054637c | 108 | flush_work(&ctrl->reset_work); |
6cd53d14 CH |
109 | nvme_stop_ctrl(ctrl); |
110 | nvme_remove_namespaces(ctrl); | |
c5017e85 | 111 | ctrl->ops->delete_ctrl(ctrl); |
6cd53d14 CH |
112 | nvme_uninit_ctrl(ctrl); |
113 | nvme_put_ctrl(ctrl); | |
c5017e85 CH |
114 | } |
115 | ||
116 | int nvme_delete_ctrl(struct nvme_ctrl *ctrl) | |
117 | { | |
118 | if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING)) | |
119 | return -EBUSY; | |
120 | if (!queue_work(nvme_wq, &ctrl->delete_work)) | |
121 | return -EBUSY; | |
122 | return 0; | |
123 | } | |
124 | EXPORT_SYMBOL_GPL(nvme_delete_ctrl); | |
125 | ||
126 | int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl) | |
127 | { | |
128 | int ret = 0; | |
129 | ||
130 | /* | |
131 | * Keep a reference until the work is flushed since ->delete_ctrl | |
132 | * can free the controller. | |
133 | */ | |
134 | nvme_get_ctrl(ctrl); | |
135 | ret = nvme_delete_ctrl(ctrl); | |
136 | if (!ret) | |
137 | flush_work(&ctrl->delete_work); | |
138 | nvme_put_ctrl(ctrl); | |
139 | return ret; | |
140 | } | |
141 | EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync); | |
142 | ||
715ea9e0 CH |
143 | static inline bool nvme_ns_has_pi(struct nvme_ns *ns) |
144 | { | |
145 | return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple); | |
146 | } | |
147 | ||
2a842aca | 148 | static blk_status_t nvme_error_status(struct request *req) |
27fa9bc5 CH |
149 | { |
150 | switch (nvme_req(req)->status & 0x7ff) { | |
151 | case NVME_SC_SUCCESS: | |
2a842aca | 152 | return BLK_STS_OK; |
27fa9bc5 | 153 | case NVME_SC_CAP_EXCEEDED: |
2a842aca | 154 | return BLK_STS_NOSPC; |
e02ab023 | 155 | case NVME_SC_ONCS_NOT_SUPPORTED: |
2a842aca | 156 | return BLK_STS_NOTSUPP; |
e02ab023 JG |
157 | case NVME_SC_WRITE_FAULT: |
158 | case NVME_SC_READ_ERROR: | |
159 | case NVME_SC_UNWRITTEN_BLOCK: | |
a751da33 CH |
160 | case NVME_SC_ACCESS_DENIED: |
161 | case NVME_SC_READ_ONLY: | |
2a842aca | 162 | return BLK_STS_MEDIUM; |
a751da33 CH |
163 | case NVME_SC_GUARD_CHECK: |
164 | case NVME_SC_APPTAG_CHECK: | |
165 | case NVME_SC_REFTAG_CHECK: | |
166 | case NVME_SC_INVALID_PI: | |
167 | return BLK_STS_PROTECTION; | |
168 | case NVME_SC_RESERVATION_CONFLICT: | |
169 | return BLK_STS_NEXUS; | |
2a842aca CH |
170 | default: |
171 | return BLK_STS_IOERR; | |
27fa9bc5 CH |
172 | } |
173 | } | |
27fa9bc5 | 174 | |
f6324b1b | 175 | static inline bool nvme_req_needs_retry(struct request *req) |
77f02a7a | 176 | { |
f6324b1b CH |
177 | if (blk_noretry_request(req)) |
178 | return false; | |
27fa9bc5 | 179 | if (nvme_req(req)->status & NVME_SC_DNR) |
f6324b1b | 180 | return false; |
44e44b29 | 181 | if (nvme_req(req)->retries >= nvme_max_retries) |
f6324b1b | 182 | return false; |
e54b064c CH |
183 | if (blk_queue_dying(req->q)) |
184 | return false; | |
f6324b1b | 185 | return true; |
77f02a7a CH |
186 | } |
187 | ||
188 | void nvme_complete_rq(struct request *req) | |
189 | { | |
27fa9bc5 CH |
190 | if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) { |
191 | nvme_req(req)->retries++; | |
8d7b8faf | 192 | blk_mq_requeue_request(req, true); |
27fa9bc5 | 193 | return; |
77f02a7a CH |
194 | } |
195 | ||
27fa9bc5 | 196 | blk_mq_end_request(req, nvme_error_status(req)); |
77f02a7a CH |
197 | } |
198 | EXPORT_SYMBOL_GPL(nvme_complete_rq); | |
199 | ||
c55a2fd4 ML |
200 | void nvme_cancel_request(struct request *req, void *data, bool reserved) |
201 | { | |
c55a2fd4 ML |
202 | if (!blk_mq_request_started(req)) |
203 | return; | |
204 | ||
205 | dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device, | |
206 | "Cancelling I/O %d", req->tag); | |
207 | ||
e54b064c | 208 | nvme_req(req)->status = NVME_SC_ABORT_REQ; |
08e0029a | 209 | blk_mq_complete_request(req); |
27fa9bc5 | 210 | |
c55a2fd4 ML |
211 | } |
212 | EXPORT_SYMBOL_GPL(nvme_cancel_request); | |
213 | ||
bb8d261e CH |
214 | bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, |
215 | enum nvme_ctrl_state new_state) | |
216 | { | |
f6b6a28e | 217 | enum nvme_ctrl_state old_state; |
0a72bbba | 218 | unsigned long flags; |
bb8d261e CH |
219 | bool changed = false; |
220 | ||
0a72bbba | 221 | spin_lock_irqsave(&ctrl->lock, flags); |
f6b6a28e GKB |
222 | |
223 | old_state = ctrl->state; | |
bb8d261e CH |
224 | switch (new_state) { |
225 | case NVME_CTRL_LIVE: | |
226 | switch (old_state) { | |
7d2e8008 | 227 | case NVME_CTRL_NEW: |
bb8d261e | 228 | case NVME_CTRL_RESETTING: |
def61eca | 229 | case NVME_CTRL_RECONNECTING: |
bb8d261e CH |
230 | changed = true; |
231 | /* FALLTHRU */ | |
232 | default: | |
233 | break; | |
234 | } | |
235 | break; | |
236 | case NVME_CTRL_RESETTING: | |
237 | switch (old_state) { | |
238 | case NVME_CTRL_NEW: | |
def61eca | 239 | case NVME_CTRL_LIVE: |
def61eca CH |
240 | changed = true; |
241 | /* FALLTHRU */ | |
242 | default: | |
243 | break; | |
244 | } | |
245 | break; | |
246 | case NVME_CTRL_RECONNECTING: | |
247 | switch (old_state) { | |
bb8d261e | 248 | case NVME_CTRL_LIVE: |
3cec7f9d | 249 | case NVME_CTRL_RESETTING: |
bb8d261e CH |
250 | changed = true; |
251 | /* FALLTHRU */ | |
252 | default: | |
253 | break; | |
254 | } | |
255 | break; | |
256 | case NVME_CTRL_DELETING: | |
257 | switch (old_state) { | |
258 | case NVME_CTRL_LIVE: | |
259 | case NVME_CTRL_RESETTING: | |
def61eca | 260 | case NVME_CTRL_RECONNECTING: |
bb8d261e CH |
261 | changed = true; |
262 | /* FALLTHRU */ | |
263 | default: | |
264 | break; | |
265 | } | |
266 | break; | |
0ff9d4e1 KB |
267 | case NVME_CTRL_DEAD: |
268 | switch (old_state) { | |
269 | case NVME_CTRL_DELETING: | |
270 | changed = true; | |
271 | /* FALLTHRU */ | |
272 | default: | |
273 | break; | |
274 | } | |
275 | break; | |
bb8d261e CH |
276 | default: |
277 | break; | |
278 | } | |
bb8d261e CH |
279 | |
280 | if (changed) | |
281 | ctrl->state = new_state; | |
282 | ||
0a72bbba | 283 | spin_unlock_irqrestore(&ctrl->lock, flags); |
f6b6a28e | 284 | |
bb8d261e CH |
285 | return changed; |
286 | } | |
287 | EXPORT_SYMBOL_GPL(nvme_change_ctrl_state); | |
288 | ||
1673f1f0 CH |
289 | static void nvme_free_ns(struct kref *kref) |
290 | { | |
291 | struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref); | |
292 | ||
b0b4e09c MB |
293 | if (ns->ndev) |
294 | nvme_nvm_unregister(ns); | |
1673f1f0 | 295 | |
1673f1f0 | 296 | put_disk(ns->disk); |
075790eb KB |
297 | ida_simple_remove(&ns->ctrl->ns_ida, ns->instance); |
298 | nvme_put_ctrl(ns->ctrl); | |
1673f1f0 CH |
299 | kfree(ns); |
300 | } | |
301 | ||
5bae7f73 | 302 | static void nvme_put_ns(struct nvme_ns *ns) |
1673f1f0 CH |
303 | { |
304 | kref_put(&ns->kref, nvme_free_ns); | |
305 | } | |
306 | ||
4160982e | 307 | struct request *nvme_alloc_request(struct request_queue *q, |
eb71f435 | 308 | struct nvme_command *cmd, unsigned int flags, int qid) |
21d34711 | 309 | { |
aebf526b | 310 | unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; |
21d34711 | 311 | struct request *req; |
21d34711 | 312 | |
eb71f435 | 313 | if (qid == NVME_QID_ANY) { |
aebf526b | 314 | req = blk_mq_alloc_request(q, op, flags); |
eb71f435 | 315 | } else { |
aebf526b | 316 | req = blk_mq_alloc_request_hctx(q, op, flags, |
eb71f435 CH |
317 | qid ? qid - 1 : 0); |
318 | } | |
21d34711 | 319 | if (IS_ERR(req)) |
4160982e | 320 | return req; |
21d34711 | 321 | |
21d34711 | 322 | req->cmd_flags |= REQ_FAILFAST_DRIVER; |
d49187e9 | 323 | nvme_req(req)->cmd = cmd; |
21d34711 | 324 | |
4160982e CH |
325 | return req; |
326 | } | |
576d55d6 | 327 | EXPORT_SYMBOL_GPL(nvme_alloc_request); |
4160982e | 328 | |
f5d11840 JA |
329 | static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable) |
330 | { | |
331 | struct nvme_command c; | |
332 | ||
333 | memset(&c, 0, sizeof(c)); | |
334 | ||
335 | c.directive.opcode = nvme_admin_directive_send; | |
62346eae | 336 | c.directive.nsid = cpu_to_le32(NVME_NSID_ALL); |
f5d11840 JA |
337 | c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE; |
338 | c.directive.dtype = NVME_DIR_IDENTIFY; | |
339 | c.directive.tdtype = NVME_DIR_STREAMS; | |
340 | c.directive.endir = enable ? NVME_DIR_ENDIR : 0; | |
341 | ||
342 | return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0); | |
343 | } | |
344 | ||
345 | static int nvme_disable_streams(struct nvme_ctrl *ctrl) | |
346 | { | |
347 | return nvme_toggle_streams(ctrl, false); | |
348 | } | |
349 | ||
350 | static int nvme_enable_streams(struct nvme_ctrl *ctrl) | |
351 | { | |
352 | return nvme_toggle_streams(ctrl, true); | |
353 | } | |
354 | ||
355 | static int nvme_get_stream_params(struct nvme_ctrl *ctrl, | |
356 | struct streams_directive_params *s, u32 nsid) | |
357 | { | |
358 | struct nvme_command c; | |
359 | ||
360 | memset(&c, 0, sizeof(c)); | |
361 | memset(s, 0, sizeof(*s)); | |
362 | ||
363 | c.directive.opcode = nvme_admin_directive_recv; | |
364 | c.directive.nsid = cpu_to_le32(nsid); | |
a082b426 | 365 | c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1); |
f5d11840 JA |
366 | c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM; |
367 | c.directive.dtype = NVME_DIR_STREAMS; | |
368 | ||
369 | return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s)); | |
370 | } | |
371 | ||
372 | static int nvme_configure_directives(struct nvme_ctrl *ctrl) | |
373 | { | |
374 | struct streams_directive_params s; | |
375 | int ret; | |
376 | ||
377 | if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES)) | |
378 | return 0; | |
379 | if (!streams) | |
380 | return 0; | |
381 | ||
382 | ret = nvme_enable_streams(ctrl); | |
383 | if (ret) | |
384 | return ret; | |
385 | ||
62346eae | 386 | ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL); |
f5d11840 JA |
387 | if (ret) |
388 | return ret; | |
389 | ||
390 | ctrl->nssa = le16_to_cpu(s.nssa); | |
391 | if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) { | |
392 | dev_info(ctrl->device, "too few streams (%u) available\n", | |
393 | ctrl->nssa); | |
394 | nvme_disable_streams(ctrl); | |
395 | return 0; | |
396 | } | |
397 | ||
398 | ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1); | |
399 | dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams); | |
400 | return 0; | |
401 | } | |
402 | ||
403 | /* | |
404 | * Check if 'req' has a write hint associated with it. If it does, assign | |
405 | * a valid namespace stream to the write. | |
406 | */ | |
407 | static void nvme_assign_write_stream(struct nvme_ctrl *ctrl, | |
408 | struct request *req, u16 *control, | |
409 | u32 *dsmgmt) | |
410 | { | |
411 | enum rw_hint streamid = req->write_hint; | |
412 | ||
413 | if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE) | |
414 | streamid = 0; | |
415 | else { | |
416 | streamid--; | |
417 | if (WARN_ON_ONCE(streamid > ctrl->nr_streams)) | |
418 | return; | |
419 | ||
420 | *control |= NVME_RW_DTYPE_STREAMS; | |
421 | *dsmgmt |= streamid << 16; | |
422 | } | |
423 | ||
424 | if (streamid < ARRAY_SIZE(req->q->write_hints)) | |
425 | req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9; | |
426 | } | |
427 | ||
8093f7ca ML |
428 | static inline void nvme_setup_flush(struct nvme_ns *ns, |
429 | struct nvme_command *cmnd) | |
430 | { | |
431 | memset(cmnd, 0, sizeof(*cmnd)); | |
432 | cmnd->common.opcode = nvme_cmd_flush; | |
433 | cmnd->common.nsid = cpu_to_le32(ns->ns_id); | |
434 | } | |
435 | ||
fc17b653 | 436 | static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req, |
8093f7ca ML |
437 | struct nvme_command *cmnd) |
438 | { | |
b35ba01e | 439 | unsigned short segments = blk_rq_nr_discard_segments(req), n = 0; |
8093f7ca | 440 | struct nvme_dsm_range *range; |
b35ba01e | 441 | struct bio *bio; |
8093f7ca | 442 | |
b35ba01e | 443 | range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC); |
8093f7ca | 444 | if (!range) |
fc17b653 | 445 | return BLK_STS_RESOURCE; |
8093f7ca | 446 | |
b35ba01e CH |
447 | __rq_for_each_bio(bio, req) { |
448 | u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector); | |
449 | u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift; | |
450 | ||
451 | range[n].cattr = cpu_to_le32(0); | |
452 | range[n].nlb = cpu_to_le32(nlb); | |
453 | range[n].slba = cpu_to_le64(slba); | |
454 | n++; | |
455 | } | |
456 | ||
457 | if (WARN_ON_ONCE(n != segments)) { | |
458 | kfree(range); | |
fc17b653 | 459 | return BLK_STS_IOERR; |
b35ba01e | 460 | } |
8093f7ca ML |
461 | |
462 | memset(cmnd, 0, sizeof(*cmnd)); | |
463 | cmnd->dsm.opcode = nvme_cmd_dsm; | |
464 | cmnd->dsm.nsid = cpu_to_le32(ns->ns_id); | |
f1dd03a8 | 465 | cmnd->dsm.nr = cpu_to_le32(segments - 1); |
8093f7ca ML |
466 | cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); |
467 | ||
f9d03f96 CH |
468 | req->special_vec.bv_page = virt_to_page(range); |
469 | req->special_vec.bv_offset = offset_in_page(range); | |
b35ba01e | 470 | req->special_vec.bv_len = sizeof(*range) * segments; |
f9d03f96 | 471 | req->rq_flags |= RQF_SPECIAL_PAYLOAD; |
8093f7ca | 472 | |
fc17b653 | 473 | return BLK_STS_OK; |
8093f7ca | 474 | } |
8093f7ca | 475 | |
ebe6d874 CH |
476 | static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns, |
477 | struct request *req, struct nvme_command *cmnd) | |
8093f7ca | 478 | { |
f5d11840 | 479 | struct nvme_ctrl *ctrl = ns->ctrl; |
8093f7ca ML |
480 | u16 control = 0; |
481 | u32 dsmgmt = 0; | |
482 | ||
483 | if (req->cmd_flags & REQ_FUA) | |
484 | control |= NVME_RW_FUA; | |
485 | if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) | |
486 | control |= NVME_RW_LR; | |
487 | ||
488 | if (req->cmd_flags & REQ_RAHEAD) | |
489 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; | |
490 | ||
491 | memset(cmnd, 0, sizeof(*cmnd)); | |
492 | cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); | |
8093f7ca ML |
493 | cmnd->rw.nsid = cpu_to_le32(ns->ns_id); |
494 | cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
495 | cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); | |
496 | ||
f5d11840 JA |
497 | if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams) |
498 | nvme_assign_write_stream(ctrl, req, &control, &dsmgmt); | |
499 | ||
8093f7ca | 500 | if (ns->ms) { |
715ea9e0 CH |
501 | /* |
502 | * If formated with metadata, the block layer always provides a | |
503 | * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else | |
504 | * we enable the PRACT bit for protection information or set the | |
505 | * namespace capacity to zero to prevent any I/O. | |
506 | */ | |
507 | if (!blk_integrity_rq(req)) { | |
508 | if (WARN_ON_ONCE(!nvme_ns_has_pi(ns))) | |
509 | return BLK_STS_NOTSUPP; | |
510 | control |= NVME_RW_PRINFO_PRACT; | |
511 | } | |
512 | ||
8093f7ca ML |
513 | switch (ns->pi_type) { |
514 | case NVME_NS_DPS_PI_TYPE3: | |
515 | control |= NVME_RW_PRINFO_PRCHK_GUARD; | |
516 | break; | |
517 | case NVME_NS_DPS_PI_TYPE1: | |
518 | case NVME_NS_DPS_PI_TYPE2: | |
519 | control |= NVME_RW_PRINFO_PRCHK_GUARD | | |
520 | NVME_RW_PRINFO_PRCHK_REF; | |
521 | cmnd->rw.reftag = cpu_to_le32( | |
522 | nvme_block_nr(ns, blk_rq_pos(req))); | |
523 | break; | |
524 | } | |
8093f7ca ML |
525 | } |
526 | ||
527 | cmnd->rw.control = cpu_to_le16(control); | |
528 | cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); | |
ebe6d874 | 529 | return 0; |
8093f7ca ML |
530 | } |
531 | ||
fc17b653 | 532 | blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req, |
8093f7ca ML |
533 | struct nvme_command *cmd) |
534 | { | |
fc17b653 | 535 | blk_status_t ret = BLK_STS_OK; |
8093f7ca | 536 | |
987f699a | 537 | if (!(req->rq_flags & RQF_DONTPREP)) { |
44e44b29 | 538 | nvme_req(req)->retries = 0; |
27fa9bc5 | 539 | nvme_req(req)->flags = 0; |
987f699a CH |
540 | req->rq_flags |= RQF_DONTPREP; |
541 | } | |
542 | ||
aebf526b CH |
543 | switch (req_op(req)) { |
544 | case REQ_OP_DRV_IN: | |
545 | case REQ_OP_DRV_OUT: | |
d49187e9 | 546 | memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd)); |
aebf526b CH |
547 | break; |
548 | case REQ_OP_FLUSH: | |
8093f7ca | 549 | nvme_setup_flush(ns, cmd); |
aebf526b | 550 | break; |
e850fd16 CH |
551 | case REQ_OP_WRITE_ZEROES: |
552 | /* currently only aliased to deallocate for a few ctrls: */ | |
aebf526b | 553 | case REQ_OP_DISCARD: |
8093f7ca | 554 | ret = nvme_setup_discard(ns, req, cmd); |
aebf526b CH |
555 | break; |
556 | case REQ_OP_READ: | |
557 | case REQ_OP_WRITE: | |
ebe6d874 | 558 | ret = nvme_setup_rw(ns, req, cmd); |
aebf526b CH |
559 | break; |
560 | default: | |
561 | WARN_ON_ONCE(1); | |
fc17b653 | 562 | return BLK_STS_IOERR; |
aebf526b | 563 | } |
8093f7ca | 564 | |
721b3917 | 565 | cmd->common.command_id = req->tag; |
8093f7ca ML |
566 | return ret; |
567 | } | |
568 | EXPORT_SYMBOL_GPL(nvme_setup_cmd); | |
569 | ||
4160982e CH |
570 | /* |
571 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
572 | * if the result is positive, it's an NVM Express status code | |
573 | */ | |
574 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
d49187e9 | 575 | union nvme_result *result, void *buffer, unsigned bufflen, |
eb71f435 | 576 | unsigned timeout, int qid, int at_head, int flags) |
4160982e CH |
577 | { |
578 | struct request *req; | |
579 | int ret; | |
580 | ||
eb71f435 | 581 | req = nvme_alloc_request(q, cmd, flags, qid); |
4160982e CH |
582 | if (IS_ERR(req)) |
583 | return PTR_ERR(req); | |
584 | ||
585 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
586 | ||
21d34711 CH |
587 | if (buffer && bufflen) { |
588 | ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL); | |
589 | if (ret) | |
590 | goto out; | |
4160982e CH |
591 | } |
592 | ||
eb71f435 | 593 | blk_execute_rq(req->q, NULL, req, at_head); |
d49187e9 CH |
594 | if (result) |
595 | *result = nvme_req(req)->result; | |
27fa9bc5 CH |
596 | if (nvme_req(req)->flags & NVME_REQ_CANCELLED) |
597 | ret = -EINTR; | |
598 | else | |
599 | ret = nvme_req(req)->status; | |
4160982e CH |
600 | out: |
601 | blk_mq_free_request(req); | |
602 | return ret; | |
603 | } | |
eb71f435 | 604 | EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd); |
4160982e CH |
605 | |
606 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
607 | void *buffer, unsigned bufflen) | |
608 | { | |
eb71f435 CH |
609 | return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0, |
610 | NVME_QID_ANY, 0, 0); | |
4160982e | 611 | } |
576d55d6 | 612 | EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd); |
4160982e | 613 | |
1cad6562 CH |
614 | static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf, |
615 | unsigned len, u32 seed, bool write) | |
616 | { | |
617 | struct bio_integrity_payload *bip; | |
618 | int ret = -ENOMEM; | |
619 | void *buf; | |
620 | ||
621 | buf = kmalloc(len, GFP_KERNEL); | |
622 | if (!buf) | |
623 | goto out; | |
624 | ||
625 | ret = -EFAULT; | |
626 | if (write && copy_from_user(buf, ubuf, len)) | |
627 | goto out_free_meta; | |
628 | ||
629 | bip = bio_integrity_alloc(bio, GFP_KERNEL, 1); | |
630 | if (IS_ERR(bip)) { | |
631 | ret = PTR_ERR(bip); | |
632 | goto out_free_meta; | |
633 | } | |
634 | ||
635 | bip->bip_iter.bi_size = len; | |
636 | bip->bip_iter.bi_sector = seed; | |
637 | ret = bio_integrity_add_page(bio, virt_to_page(buf), len, | |
638 | offset_in_page(buf)); | |
639 | if (ret == len) | |
640 | return buf; | |
641 | ret = -ENOMEM; | |
642 | out_free_meta: | |
643 | kfree(buf); | |
644 | out: | |
645 | return ERR_PTR(ret); | |
646 | } | |
647 | ||
63263d60 | 648 | static int nvme_submit_user_cmd(struct request_queue *q, |
485783ca KB |
649 | struct nvme_command *cmd, void __user *ubuffer, |
650 | unsigned bufflen, void __user *meta_buffer, unsigned meta_len, | |
651 | u32 meta_seed, u32 *result, unsigned timeout) | |
4160982e | 652 | { |
7a5abb4b | 653 | bool write = nvme_is_write(cmd); |
0b7f1f26 KB |
654 | struct nvme_ns *ns = q->queuedata; |
655 | struct gendisk *disk = ns ? ns->disk : NULL; | |
4160982e | 656 | struct request *req; |
0b7f1f26 KB |
657 | struct bio *bio = NULL; |
658 | void *meta = NULL; | |
4160982e CH |
659 | int ret; |
660 | ||
eb71f435 | 661 | req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY); |
4160982e CH |
662 | if (IS_ERR(req)) |
663 | return PTR_ERR(req); | |
664 | ||
665 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
666 | ||
667 | if (ubuffer && bufflen) { | |
21d34711 CH |
668 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, |
669 | GFP_KERNEL); | |
670 | if (ret) | |
671 | goto out; | |
672 | bio = req->bio; | |
74d46992 | 673 | bio->bi_disk = disk; |
1cad6562 CH |
674 | if (disk && meta_buffer && meta_len) { |
675 | meta = nvme_add_user_metadata(bio, meta_buffer, meta_len, | |
676 | meta_seed, write); | |
677 | if (IS_ERR(meta)) { | |
678 | ret = PTR_ERR(meta); | |
0b7f1f26 KB |
679 | goto out_unmap; |
680 | } | |
0b7f1f26 KB |
681 | } |
682 | } | |
1cad6562 | 683 | |
0b7f1f26 | 684 | blk_execute_rq(req->q, disk, req, 0); |
27fa9bc5 CH |
685 | if (nvme_req(req)->flags & NVME_REQ_CANCELLED) |
686 | ret = -EINTR; | |
687 | else | |
688 | ret = nvme_req(req)->status; | |
21d34711 | 689 | if (result) |
d49187e9 | 690 | *result = le32_to_cpu(nvme_req(req)->result.u32); |
0b7f1f26 KB |
691 | if (meta && !ret && !write) { |
692 | if (copy_to_user(meta_buffer, meta, meta_len)) | |
693 | ret = -EFAULT; | |
694 | } | |
0b7f1f26 KB |
695 | kfree(meta); |
696 | out_unmap: | |
74d46992 | 697 | if (bio) |
0b7f1f26 | 698 | blk_rq_unmap_user(bio); |
21d34711 CH |
699 | out: |
700 | blk_mq_free_request(req); | |
701 | return ret; | |
702 | } | |
703 | ||
2a842aca | 704 | static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status) |
038bd4cb SG |
705 | { |
706 | struct nvme_ctrl *ctrl = rq->end_io_data; | |
707 | ||
708 | blk_mq_free_request(rq); | |
709 | ||
2a842aca | 710 | if (status) { |
038bd4cb | 711 | dev_err(ctrl->device, |
2a842aca CH |
712 | "failed nvme_keep_alive_end_io error=%d\n", |
713 | status); | |
038bd4cb SG |
714 | return; |
715 | } | |
716 | ||
717 | schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); | |
718 | } | |
719 | ||
720 | static int nvme_keep_alive(struct nvme_ctrl *ctrl) | |
721 | { | |
722 | struct nvme_command c; | |
723 | struct request *rq; | |
724 | ||
725 | memset(&c, 0, sizeof(c)); | |
726 | c.common.opcode = nvme_admin_keep_alive; | |
727 | ||
728 | rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED, | |
729 | NVME_QID_ANY); | |
730 | if (IS_ERR(rq)) | |
731 | return PTR_ERR(rq); | |
732 | ||
733 | rq->timeout = ctrl->kato * HZ; | |
734 | rq->end_io_data = ctrl; | |
735 | ||
736 | blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io); | |
737 | ||
738 | return 0; | |
739 | } | |
740 | ||
741 | static void nvme_keep_alive_work(struct work_struct *work) | |
742 | { | |
743 | struct nvme_ctrl *ctrl = container_of(to_delayed_work(work), | |
744 | struct nvme_ctrl, ka_work); | |
745 | ||
746 | if (nvme_keep_alive(ctrl)) { | |
747 | /* allocation failure, reset the controller */ | |
748 | dev_err(ctrl->device, "keep-alive failed\n"); | |
39bdc590 | 749 | nvme_reset_ctrl(ctrl); |
038bd4cb SG |
750 | return; |
751 | } | |
752 | } | |
753 | ||
754 | void nvme_start_keep_alive(struct nvme_ctrl *ctrl) | |
755 | { | |
756 | if (unlikely(ctrl->kato == 0)) | |
757 | return; | |
758 | ||
759 | INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work); | |
760 | schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); | |
761 | } | |
762 | EXPORT_SYMBOL_GPL(nvme_start_keep_alive); | |
763 | ||
764 | void nvme_stop_keep_alive(struct nvme_ctrl *ctrl) | |
765 | { | |
766 | if (unlikely(ctrl->kato == 0)) | |
767 | return; | |
768 | ||
769 | cancel_delayed_work_sync(&ctrl->ka_work); | |
770 | } | |
771 | EXPORT_SYMBOL_GPL(nvme_stop_keep_alive); | |
772 | ||
3f7f25a9 | 773 | static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id) |
21d34711 CH |
774 | { |
775 | struct nvme_command c = { }; | |
776 | int error; | |
777 | ||
778 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
779 | c.identify.opcode = nvme_admin_identify; | |
986994a2 | 780 | c.identify.cns = NVME_ID_CNS_CTRL; |
21d34711 CH |
781 | |
782 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); | |
783 | if (!*id) | |
784 | return -ENOMEM; | |
785 | ||
786 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
787 | sizeof(struct nvme_id_ctrl)); | |
788 | if (error) | |
789 | kfree(*id); | |
790 | return error; | |
791 | } | |
792 | ||
cdbff4f2 CH |
793 | static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid, |
794 | u8 *eui64, u8 *nguid, uuid_t *uuid) | |
3b22ba26 JT |
795 | { |
796 | struct nvme_command c = { }; | |
797 | int status; | |
798 | void *data; | |
799 | int pos; | |
800 | int len; | |
801 | ||
802 | c.identify.opcode = nvme_admin_identify; | |
803 | c.identify.nsid = cpu_to_le32(nsid); | |
804 | c.identify.cns = NVME_ID_CNS_NS_DESC_LIST; | |
805 | ||
806 | data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL); | |
807 | if (!data) | |
808 | return -ENOMEM; | |
809 | ||
cdbff4f2 | 810 | status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data, |
3b22ba26 JT |
811 | NVME_IDENTIFY_DATA_SIZE); |
812 | if (status) | |
813 | goto free_data; | |
814 | ||
815 | for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) { | |
816 | struct nvme_ns_id_desc *cur = data + pos; | |
817 | ||
818 | if (cur->nidl == 0) | |
819 | break; | |
820 | ||
821 | switch (cur->nidt) { | |
822 | case NVME_NIDT_EUI64: | |
823 | if (cur->nidl != NVME_NIDT_EUI64_LEN) { | |
cdbff4f2 | 824 | dev_warn(ctrl->device, |
3b22ba26 JT |
825 | "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n", |
826 | cur->nidl); | |
827 | goto free_data; | |
828 | } | |
829 | len = NVME_NIDT_EUI64_LEN; | |
cdbff4f2 | 830 | memcpy(eui64, data + pos + sizeof(*cur), len); |
3b22ba26 JT |
831 | break; |
832 | case NVME_NIDT_NGUID: | |
833 | if (cur->nidl != NVME_NIDT_NGUID_LEN) { | |
cdbff4f2 | 834 | dev_warn(ctrl->device, |
3b22ba26 JT |
835 | "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n", |
836 | cur->nidl); | |
837 | goto free_data; | |
838 | } | |
839 | len = NVME_NIDT_NGUID_LEN; | |
cdbff4f2 | 840 | memcpy(nguid, data + pos + sizeof(*cur), len); |
3b22ba26 JT |
841 | break; |
842 | case NVME_NIDT_UUID: | |
843 | if (cur->nidl != NVME_NIDT_UUID_LEN) { | |
cdbff4f2 | 844 | dev_warn(ctrl->device, |
3b22ba26 JT |
845 | "ctrl returned bogus length: %d for NVME_NIDT_UUID\n", |
846 | cur->nidl); | |
847 | goto free_data; | |
848 | } | |
849 | len = NVME_NIDT_UUID_LEN; | |
cdbff4f2 | 850 | uuid_copy(uuid, data + pos + sizeof(*cur)); |
3b22ba26 JT |
851 | break; |
852 | default: | |
853 | /* Skip unnkown types */ | |
854 | len = cur->nidl; | |
855 | break; | |
856 | } | |
857 | ||
858 | len += sizeof(*cur); | |
859 | } | |
860 | free_data: | |
861 | kfree(data); | |
862 | return status; | |
863 | } | |
864 | ||
540c801c KB |
865 | static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list) |
866 | { | |
867 | struct nvme_command c = { }; | |
868 | ||
869 | c.identify.opcode = nvme_admin_identify; | |
986994a2 | 870 | c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST; |
540c801c KB |
871 | c.identify.nsid = cpu_to_le32(nsid); |
872 | return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000); | |
873 | } | |
874 | ||
cdbff4f2 CH |
875 | static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl, |
876 | unsigned nsid) | |
21d34711 | 877 | { |
cdbff4f2 | 878 | struct nvme_id_ns *id; |
21d34711 CH |
879 | struct nvme_command c = { }; |
880 | int error; | |
881 | ||
882 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
778f067c MG |
883 | c.identify.opcode = nvme_admin_identify; |
884 | c.identify.nsid = cpu_to_le32(nsid); | |
986994a2 | 885 | c.identify.cns = NVME_ID_CNS_NS; |
21d34711 | 886 | |
cdbff4f2 CH |
887 | id = kmalloc(sizeof(*id), GFP_KERNEL); |
888 | if (!id) | |
889 | return NULL; | |
21d34711 | 890 | |
cdbff4f2 CH |
891 | error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id)); |
892 | if (error) { | |
893 | dev_warn(ctrl->device, "Identify namespace failed\n"); | |
894 | kfree(id); | |
895 | return NULL; | |
896 | } | |
897 | ||
898 | return id; | |
21d34711 CH |
899 | } |
900 | ||
3f7f25a9 | 901 | static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11, |
1a6fe74d | 902 | void *buffer, size_t buflen, u32 *result) |
21d34711 CH |
903 | { |
904 | struct nvme_command c; | |
d49187e9 | 905 | union nvme_result res; |
1cb3cce5 | 906 | int ret; |
21d34711 CH |
907 | |
908 | memset(&c, 0, sizeof(c)); | |
909 | c.features.opcode = nvme_admin_set_features; | |
21d34711 CH |
910 | c.features.fid = cpu_to_le32(fid); |
911 | c.features.dword11 = cpu_to_le32(dword11); | |
912 | ||
d49187e9 | 913 | ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, |
1a6fe74d | 914 | buffer, buflen, 0, NVME_QID_ANY, 0, 0); |
9b47f77a | 915 | if (ret >= 0 && result) |
d49187e9 | 916 | *result = le32_to_cpu(res.u32); |
1cb3cce5 | 917 | return ret; |
21d34711 CH |
918 | } |
919 | ||
9a0be7ab CH |
920 | int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count) |
921 | { | |
922 | u32 q_count = (*count - 1) | ((*count - 1) << 16); | |
923 | u32 result; | |
924 | int status, nr_io_queues; | |
925 | ||
1a6fe74d | 926 | status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0, |
9a0be7ab | 927 | &result); |
f5fa90dc | 928 | if (status < 0) |
9a0be7ab CH |
929 | return status; |
930 | ||
f5fa90dc CH |
931 | /* |
932 | * Degraded controllers might return an error when setting the queue | |
933 | * count. We still want to be able to bring them online and offer | |
934 | * access to the admin queue, as that might be only way to fix them up. | |
935 | */ | |
936 | if (status > 0) { | |
f0425db0 | 937 | dev_err(ctrl->device, "Could not set queue count (%d)\n", status); |
f5fa90dc CH |
938 | *count = 0; |
939 | } else { | |
940 | nr_io_queues = min(result & 0xffff, result >> 16) + 1; | |
941 | *count = min(*count, nr_io_queues); | |
942 | } | |
943 | ||
9a0be7ab CH |
944 | return 0; |
945 | } | |
576d55d6 | 946 | EXPORT_SYMBOL_GPL(nvme_set_queue_count); |
9a0be7ab | 947 | |
1673f1f0 CH |
948 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
949 | { | |
950 | struct nvme_user_io io; | |
951 | struct nvme_command c; | |
952 | unsigned length, meta_len; | |
953 | void __user *metadata; | |
954 | ||
955 | if (copy_from_user(&io, uio, sizeof(io))) | |
956 | return -EFAULT; | |
63088ec7 KB |
957 | if (io.flags) |
958 | return -EINVAL; | |
1673f1f0 CH |
959 | |
960 | switch (io.opcode) { | |
961 | case nvme_cmd_write: | |
962 | case nvme_cmd_read: | |
963 | case nvme_cmd_compare: | |
964 | break; | |
965 | default: | |
966 | return -EINVAL; | |
967 | } | |
968 | ||
969 | length = (io.nblocks + 1) << ns->lba_shift; | |
970 | meta_len = (io.nblocks + 1) * ns->ms; | |
971 | metadata = (void __user *)(uintptr_t)io.metadata; | |
972 | ||
973 | if (ns->ext) { | |
974 | length += meta_len; | |
975 | meta_len = 0; | |
976 | } else if (meta_len) { | |
977 | if ((io.metadata & 3) || !io.metadata) | |
978 | return -EINVAL; | |
979 | } | |
980 | ||
981 | memset(&c, 0, sizeof(c)); | |
982 | c.rw.opcode = io.opcode; | |
983 | c.rw.flags = io.flags; | |
984 | c.rw.nsid = cpu_to_le32(ns->ns_id); | |
985 | c.rw.slba = cpu_to_le64(io.slba); | |
986 | c.rw.length = cpu_to_le16(io.nblocks); | |
987 | c.rw.control = cpu_to_le16(io.control); | |
988 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); | |
989 | c.rw.reftag = cpu_to_le32(io.reftag); | |
990 | c.rw.apptag = cpu_to_le16(io.apptag); | |
991 | c.rw.appmask = cpu_to_le16(io.appmask); | |
992 | ||
63263d60 | 993 | return nvme_submit_user_cmd(ns->queue, &c, |
1673f1f0 CH |
994 | (void __user *)(uintptr_t)io.addr, length, |
995 | metadata, meta_len, io.slba, NULL, 0); | |
996 | } | |
997 | ||
84fef62d KB |
998 | static u32 nvme_known_admin_effects(u8 opcode) |
999 | { | |
1000 | switch (opcode) { | |
1001 | case nvme_admin_format_nvm: | |
1002 | return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC | | |
1003 | NVME_CMD_EFFECTS_CSE_MASK; | |
1004 | case nvme_admin_sanitize_nvm: | |
1005 | return NVME_CMD_EFFECTS_CSE_MASK; | |
1006 | default: | |
1007 | break; | |
1008 | } | |
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, | |
1013 | u8 opcode) | |
1014 | { | |
1015 | u32 effects = 0; | |
1016 | ||
1017 | if (ns) { | |
1018 | if (ctrl->effects) | |
1019 | effects = le32_to_cpu(ctrl->effects->iocs[opcode]); | |
1020 | if (effects & ~NVME_CMD_EFFECTS_CSUPP) | |
1021 | dev_warn(ctrl->device, | |
1022 | "IO command:%02x has unhandled effects:%08x\n", | |
1023 | opcode, effects); | |
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | if (ctrl->effects) | |
1028 | effects = le32_to_cpu(ctrl->effects->iocs[opcode]); | |
1029 | else | |
1030 | effects = nvme_known_admin_effects(opcode); | |
1031 | ||
1032 | /* | |
1033 | * For simplicity, IO to all namespaces is quiesced even if the command | |
1034 | * effects say only one namespace is affected. | |
1035 | */ | |
1036 | if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) { | |
1037 | nvme_start_freeze(ctrl); | |
1038 | nvme_wait_freeze(ctrl); | |
1039 | } | |
1040 | return effects; | |
1041 | } | |
1042 | ||
1043 | static void nvme_update_formats(struct nvme_ctrl *ctrl) | |
1044 | { | |
1045 | struct nvme_ns *ns; | |
1046 | ||
1047 | mutex_lock(&ctrl->namespaces_mutex); | |
1048 | list_for_each_entry(ns, &ctrl->namespaces, list) { | |
1049 | if (ns->disk && nvme_revalidate_disk(ns->disk)) | |
1050 | nvme_ns_remove(ns); | |
1051 | } | |
1052 | mutex_unlock(&ctrl->namespaces_mutex); | |
1053 | } | |
1054 | ||
1055 | static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects) | |
1056 | { | |
1057 | /* | |
1058 | * Revalidate LBA changes prior to unfreezing. This is necessary to | |
1059 | * prevent memory corruption if a logical block size was changed by | |
1060 | * this command. | |
1061 | */ | |
1062 | if (effects & NVME_CMD_EFFECTS_LBCC) | |
1063 | nvme_update_formats(ctrl); | |
1064 | if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) | |
1065 | nvme_unfreeze(ctrl); | |
1066 | if (effects & NVME_CMD_EFFECTS_CCC) | |
1067 | nvme_init_identify(ctrl); | |
1068 | if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) | |
1069 | nvme_queue_scan(ctrl); | |
1070 | } | |
1071 | ||
f3ca80fc | 1072 | static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns, |
1673f1f0 CH |
1073 | struct nvme_passthru_cmd __user *ucmd) |
1074 | { | |
1075 | struct nvme_passthru_cmd cmd; | |
1076 | struct nvme_command c; | |
1077 | unsigned timeout = 0; | |
84fef62d | 1078 | u32 effects; |
1673f1f0 CH |
1079 | int status; |
1080 | ||
1081 | if (!capable(CAP_SYS_ADMIN)) | |
1082 | return -EACCES; | |
1083 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
1084 | return -EFAULT; | |
63088ec7 KB |
1085 | if (cmd.flags) |
1086 | return -EINVAL; | |
1673f1f0 CH |
1087 | |
1088 | memset(&c, 0, sizeof(c)); | |
1089 | c.common.opcode = cmd.opcode; | |
1090 | c.common.flags = cmd.flags; | |
1091 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1092 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1093 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1094 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
1095 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
1096 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
1097 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
1098 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
1099 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
1100 | ||
1101 | if (cmd.timeout_ms) | |
1102 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
1103 | ||
84fef62d | 1104 | effects = nvme_passthru_start(ctrl, ns, cmd.opcode); |
1673f1f0 | 1105 | status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c, |
d1ea7be5 | 1106 | (void __user *)(uintptr_t)cmd.addr, cmd.data_len, |
63263d60 KB |
1107 | (void __user *)(uintptr_t)cmd.metadata, cmd.metadata, |
1108 | 0, &cmd.result, timeout); | |
84fef62d KB |
1109 | nvme_passthru_end(ctrl, effects); |
1110 | ||
1673f1f0 CH |
1111 | if (status >= 0) { |
1112 | if (put_user(cmd.result, &ucmd->result)) | |
1113 | return -EFAULT; | |
1114 | } | |
1115 | ||
1116 | return status; | |
1117 | } | |
1118 | ||
1119 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, | |
1120 | unsigned int cmd, unsigned long arg) | |
1121 | { | |
1122 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1123 | ||
1124 | switch (cmd) { | |
1125 | case NVME_IOCTL_ID: | |
1126 | force_successful_syscall_return(); | |
1127 | return ns->ns_id; | |
1128 | case NVME_IOCTL_ADMIN_CMD: | |
1129 | return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg); | |
1130 | case NVME_IOCTL_IO_CMD: | |
1131 | return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg); | |
1132 | case NVME_IOCTL_SUBMIT_IO: | |
1133 | return nvme_submit_io(ns, (void __user *)arg); | |
1673f1f0 | 1134 | default: |
84d4add7 MB |
1135 | #ifdef CONFIG_NVM |
1136 | if (ns->ndev) | |
1137 | return nvme_nvm_ioctl(ns, cmd, arg); | |
1138 | #endif | |
a98e58e5 | 1139 | if (is_sed_ioctl(cmd)) |
4f1244c8 | 1140 | return sed_ioctl(ns->ctrl->opal_dev, cmd, |
e225c20e | 1141 | (void __user *) arg); |
1673f1f0 CH |
1142 | return -ENOTTY; |
1143 | } | |
1144 | } | |
1145 | ||
1673f1f0 CH |
1146 | static int nvme_open(struct block_device *bdev, fmode_t mode) |
1147 | { | |
c6424a90 CH |
1148 | struct nvme_ns *ns = bdev->bd_disk->private_data; |
1149 | ||
1150 | if (!kref_get_unless_zero(&ns->kref)) | |
1151 | return -ENXIO; | |
c6424a90 | 1152 | return 0; |
1673f1f0 CH |
1153 | } |
1154 | ||
1155 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
1156 | { | |
a6a5149b | 1157 | nvme_put_ns(disk->private_data); |
1673f1f0 CH |
1158 | } |
1159 | ||
1160 | static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
1161 | { | |
1162 | /* some standard values */ | |
1163 | geo->heads = 1 << 6; | |
1164 | geo->sectors = 1 << 5; | |
1165 | geo->cylinders = get_capacity(bdev->bd_disk) >> 11; | |
1166 | return 0; | |
1167 | } | |
1168 | ||
1169 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
39b7baa4 | 1170 | static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) |
1673f1f0 CH |
1171 | { |
1172 | struct blk_integrity integrity; | |
1173 | ||
fa9a89fc | 1174 | memset(&integrity, 0, sizeof(integrity)); |
39b7baa4 | 1175 | switch (pi_type) { |
1673f1f0 CH |
1176 | case NVME_NS_DPS_PI_TYPE3: |
1177 | integrity.profile = &t10_pi_type3_crc; | |
ba36c21b NB |
1178 | integrity.tag_size = sizeof(u16) + sizeof(u32); |
1179 | integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; | |
1673f1f0 CH |
1180 | break; |
1181 | case NVME_NS_DPS_PI_TYPE1: | |
1182 | case NVME_NS_DPS_PI_TYPE2: | |
1183 | integrity.profile = &t10_pi_type1_crc; | |
ba36c21b NB |
1184 | integrity.tag_size = sizeof(u16); |
1185 | integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; | |
1673f1f0 CH |
1186 | break; |
1187 | default: | |
1188 | integrity.profile = NULL; | |
1189 | break; | |
1190 | } | |
39b7baa4 CH |
1191 | integrity.tuple_size = ms; |
1192 | blk_integrity_register(disk, &integrity); | |
1193 | blk_queue_max_integrity_segments(disk->queue, 1); | |
1673f1f0 CH |
1194 | } |
1195 | #else | |
39b7baa4 | 1196 | static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) |
1673f1f0 CH |
1197 | { |
1198 | } | |
1199 | #endif /* CONFIG_BLK_DEV_INTEGRITY */ | |
1200 | ||
6b8190d6 SB |
1201 | static void nvme_set_chunk_size(struct nvme_ns *ns) |
1202 | { | |
1203 | u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9)); | |
1204 | blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size)); | |
1205 | } | |
1206 | ||
30e5e929 CH |
1207 | static void nvme_config_discard(struct nvme_ctrl *ctrl, |
1208 | unsigned stream_alignment, struct request_queue *queue) | |
1673f1f0 | 1209 | { |
30e5e929 CH |
1210 | u32 size = queue_logical_block_size(queue); |
1211 | ||
1212 | if (stream_alignment) | |
1213 | size *= stream_alignment; | |
08095e70 | 1214 | |
b35ba01e CH |
1215 | BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) < |
1216 | NVME_DSM_MAX_RANGES); | |
1217 | ||
30e5e929 CH |
1218 | queue->limits.discard_alignment = size; |
1219 | queue->limits.discard_granularity = size; | |
f5d11840 | 1220 | |
30e5e929 CH |
1221 | blk_queue_max_discard_sectors(queue, UINT_MAX); |
1222 | blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES); | |
1223 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue); | |
e850fd16 CH |
1224 | |
1225 | if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) | |
30e5e929 | 1226 | blk_queue_max_write_zeroes_sectors(queue, UINT_MAX); |
1673f1f0 CH |
1227 | } |
1228 | ||
cdbff4f2 CH |
1229 | static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid, |
1230 | struct nvme_id_ns *id, u8 *eui64, u8 *nguid, uuid_t *uuid) | |
1673f1f0 | 1231 | { |
cdbff4f2 CH |
1232 | if (ctrl->vs >= NVME_VS(1, 1, 0)) |
1233 | memcpy(eui64, id->eui64, sizeof(id->eui64)); | |
1234 | if (ctrl->vs >= NVME_VS(1, 2, 0)) | |
1235 | memcpy(nguid, id->nguid, sizeof(id->nguid)); | |
1236 | if (ctrl->vs >= NVME_VS(1, 3, 0)) { | |
3b22ba26 JT |
1237 | /* Don't treat error as fatal we potentially |
1238 | * already have a NGUID or EUI-64 | |
1239 | */ | |
cdbff4f2 CH |
1240 | if (nvme_identify_ns_descs(ctrl, nsid, eui64, nguid, uuid)) |
1241 | dev_warn(ctrl->device, | |
3b22ba26 JT |
1242 | "%s: Identify Descriptors failed\n", __func__); |
1243 | } | |
ac81bfa9 MB |
1244 | } |
1245 | ||
24b0b58c CH |
1246 | static void nvme_update_disk_info(struct gendisk *disk, |
1247 | struct nvme_ns *ns, struct nvme_id_ns *id) | |
1248 | { | |
1249 | sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9); | |
1250 | unsigned stream_alignment = 0; | |
1251 | ||
1252 | if (ns->ctrl->nr_streams && ns->sws && ns->sgs) | |
1253 | stream_alignment = ns->sws * ns->sgs; | |
1254 | ||
1255 | blk_mq_freeze_queue(disk->queue); | |
1256 | blk_integrity_unregister(disk); | |
1257 | ||
1258 | blk_queue_logical_block_size(disk->queue, 1 << ns->lba_shift); | |
1259 | if (ns->ms && !ns->ext && | |
1260 | (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)) | |
1261 | nvme_init_integrity(disk, ns->ms, ns->pi_type); | |
715ea9e0 | 1262 | if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) |
24b0b58c CH |
1263 | capacity = 0; |
1264 | set_capacity(disk, capacity); | |
1265 | ||
1266 | if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM) | |
1267 | nvme_config_discard(ns->ctrl, stream_alignment, disk->queue); | |
1268 | blk_mq_unfreeze_queue(disk->queue); | |
1269 | } | |
1270 | ||
ac81bfa9 MB |
1271 | static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id) |
1272 | { | |
1273 | struct nvme_ns *ns = disk->private_data; | |
1673f1f0 CH |
1274 | |
1275 | /* | |
1276 | * If identify namespace failed, use default 512 byte block size so | |
1277 | * block layer can use before failing read/write for 0 capacity. | |
1278 | */ | |
c81bfba9 | 1279 | ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds; |
1673f1f0 CH |
1280 | if (ns->lba_shift == 0) |
1281 | ns->lba_shift = 9; | |
6b8190d6 | 1282 | ns->noiob = le16_to_cpu(id->noiob); |
b5be3b39 CH |
1283 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); |
1284 | ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms); | |
1285 | /* the PI implementation requires metadata equal t10 pi tuple size */ | |
1286 | if (ns->ms == sizeof(struct t10_pi_tuple)) | |
1287 | ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK; | |
1288 | else | |
1289 | ns->pi_type = 0; | |
1673f1f0 | 1290 | |
6e78f21a CH |
1291 | if (ns->noiob) |
1292 | nvme_set_chunk_size(ns); | |
24b0b58c | 1293 | nvme_update_disk_info(disk, ns, id); |
ac81bfa9 | 1294 | } |
1673f1f0 | 1295 | |
ac81bfa9 MB |
1296 | static int nvme_revalidate_disk(struct gendisk *disk) |
1297 | { | |
1298 | struct nvme_ns *ns = disk->private_data; | |
cdbff4f2 CH |
1299 | struct nvme_ctrl *ctrl = ns->ctrl; |
1300 | struct nvme_id_ns *id; | |
1d5df6af CH |
1301 | u8 eui64[8] = { 0 }, nguid[16] = { 0 }; |
1302 | uuid_t uuid = uuid_null; | |
cdbff4f2 | 1303 | int ret = 0; |
ac81bfa9 MB |
1304 | |
1305 | if (test_bit(NVME_NS_DEAD, &ns->flags)) { | |
1306 | set_capacity(disk, 0); | |
1307 | return -ENODEV; | |
1308 | } | |
1309 | ||
cdbff4f2 CH |
1310 | id = nvme_identify_ns(ctrl, ns->ns_id); |
1311 | if (!id) | |
1312 | return -ENODEV; | |
ac81bfa9 | 1313 | |
cdbff4f2 CH |
1314 | if (id->ncap == 0) { |
1315 | ret = -ENODEV; | |
1316 | goto out; | |
1317 | } | |
ac81bfa9 | 1318 | |
1d5df6af CH |
1319 | nvme_report_ns_ids(ctrl, ns->ns_id, id, eui64, nguid, &uuid); |
1320 | if (!uuid_equal(&ns->uuid, &uuid) || | |
1321 | memcmp(&ns->nguid, &nguid, sizeof(ns->nguid)) || | |
1322 | memcmp(&ns->eui, &eui64, sizeof(ns->eui))) { | |
1323 | dev_err(ctrl->device, | |
1324 | "identifiers changed for nsid %d\n", ns->ns_id); | |
1325 | ret = -ENODEV; | |
1326 | } | |
1327 | ||
cdbff4f2 CH |
1328 | out: |
1329 | kfree(id); | |
1330 | return ret; | |
1673f1f0 CH |
1331 | } |
1332 | ||
1333 | static char nvme_pr_type(enum pr_type type) | |
1334 | { | |
1335 | switch (type) { | |
1336 | case PR_WRITE_EXCLUSIVE: | |
1337 | return 1; | |
1338 | case PR_EXCLUSIVE_ACCESS: | |
1339 | return 2; | |
1340 | case PR_WRITE_EXCLUSIVE_REG_ONLY: | |
1341 | return 3; | |
1342 | case PR_EXCLUSIVE_ACCESS_REG_ONLY: | |
1343 | return 4; | |
1344 | case PR_WRITE_EXCLUSIVE_ALL_REGS: | |
1345 | return 5; | |
1346 | case PR_EXCLUSIVE_ACCESS_ALL_REGS: | |
1347 | return 6; | |
1348 | default: | |
1349 | return 0; | |
1350 | } | |
1351 | }; | |
1352 | ||
1353 | static int nvme_pr_command(struct block_device *bdev, u32 cdw10, | |
1354 | u64 key, u64 sa_key, u8 op) | |
1355 | { | |
1356 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1357 | struct nvme_command c; | |
1358 | u8 data[16] = { 0, }; | |
1359 | ||
1360 | put_unaligned_le64(key, &data[0]); | |
1361 | put_unaligned_le64(sa_key, &data[8]); | |
1362 | ||
1363 | memset(&c, 0, sizeof(c)); | |
1364 | c.common.opcode = op; | |
1365 | c.common.nsid = cpu_to_le32(ns->ns_id); | |
1366 | c.common.cdw10[0] = cpu_to_le32(cdw10); | |
1367 | ||
1368 | return nvme_submit_sync_cmd(ns->queue, &c, data, 16); | |
1369 | } | |
1370 | ||
1371 | static int nvme_pr_register(struct block_device *bdev, u64 old, | |
1372 | u64 new, unsigned flags) | |
1373 | { | |
1374 | u32 cdw10; | |
1375 | ||
1376 | if (flags & ~PR_FL_IGNORE_KEY) | |
1377 | return -EOPNOTSUPP; | |
1378 | ||
1379 | cdw10 = old ? 2 : 0; | |
1380 | cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; | |
1381 | cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ | |
1382 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register); | |
1383 | } | |
1384 | ||
1385 | static int nvme_pr_reserve(struct block_device *bdev, u64 key, | |
1386 | enum pr_type type, unsigned flags) | |
1387 | { | |
1388 | u32 cdw10; | |
1389 | ||
1390 | if (flags & ~PR_FL_IGNORE_KEY) | |
1391 | return -EOPNOTSUPP; | |
1392 | ||
1393 | cdw10 = nvme_pr_type(type) << 8; | |
1394 | cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); | |
1395 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire); | |
1396 | } | |
1397 | ||
1398 | static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, | |
1399 | enum pr_type type, bool abort) | |
1400 | { | |
1401 | u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1; | |
1402 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire); | |
1403 | } | |
1404 | ||
1405 | static int nvme_pr_clear(struct block_device *bdev, u64 key) | |
1406 | { | |
8c0b3915 | 1407 | u32 cdw10 = 1 | (key ? 1 << 3 : 0); |
1673f1f0 CH |
1408 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register); |
1409 | } | |
1410 | ||
1411 | static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
1412 | { | |
1413 | u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0; | |
1414 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release); | |
1415 | } | |
1416 | ||
1417 | static const struct pr_ops nvme_pr_ops = { | |
1418 | .pr_register = nvme_pr_register, | |
1419 | .pr_reserve = nvme_pr_reserve, | |
1420 | .pr_release = nvme_pr_release, | |
1421 | .pr_preempt = nvme_pr_preempt, | |
1422 | .pr_clear = nvme_pr_clear, | |
1423 | }; | |
1424 | ||
a98e58e5 | 1425 | #ifdef CONFIG_BLK_SED_OPAL |
4f1244c8 CH |
1426 | int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len, |
1427 | bool send) | |
a98e58e5 | 1428 | { |
4f1244c8 | 1429 | struct nvme_ctrl *ctrl = data; |
a98e58e5 | 1430 | struct nvme_command cmd; |
a98e58e5 SB |
1431 | |
1432 | memset(&cmd, 0, sizeof(cmd)); | |
1433 | if (send) | |
1434 | cmd.common.opcode = nvme_admin_security_send; | |
1435 | else | |
1436 | cmd.common.opcode = nvme_admin_security_recv; | |
a98e58e5 SB |
1437 | cmd.common.nsid = 0; |
1438 | cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8); | |
1439 | cmd.common.cdw10[1] = cpu_to_le32(len); | |
1440 | ||
1441 | return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, | |
1442 | ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0); | |
1443 | } | |
1444 | EXPORT_SYMBOL_GPL(nvme_sec_submit); | |
1445 | #endif /* CONFIG_BLK_SED_OPAL */ | |
1446 | ||
5bae7f73 | 1447 | static const struct block_device_operations nvme_fops = { |
1673f1f0 CH |
1448 | .owner = THIS_MODULE, |
1449 | .ioctl = nvme_ioctl, | |
761f2e1e | 1450 | .compat_ioctl = nvme_ioctl, |
1673f1f0 CH |
1451 | .open = nvme_open, |
1452 | .release = nvme_release, | |
1453 | .getgeo = nvme_getgeo, | |
1454 | .revalidate_disk= nvme_revalidate_disk, | |
1455 | .pr_ops = &nvme_pr_ops, | |
1456 | }; | |
1457 | ||
5fd4ce1b CH |
1458 | static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled) |
1459 | { | |
1460 | unsigned long timeout = | |
1461 | ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
1462 | u32 csts, bit = enabled ? NVME_CSTS_RDY : 0; | |
1463 | int ret; | |
1464 | ||
1465 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
0df1e4f5 KB |
1466 | if (csts == ~0) |
1467 | return -ENODEV; | |
5fd4ce1b CH |
1468 | if ((csts & NVME_CSTS_RDY) == bit) |
1469 | break; | |
1470 | ||
1471 | msleep(100); | |
1472 | if (fatal_signal_pending(current)) | |
1473 | return -EINTR; | |
1474 | if (time_after(jiffies, timeout)) { | |
1b3c47c1 | 1475 | dev_err(ctrl->device, |
5fd4ce1b CH |
1476 | "Device not ready; aborting %s\n", enabled ? |
1477 | "initialisation" : "reset"); | |
1478 | return -ENODEV; | |
1479 | } | |
1480 | } | |
1481 | ||
1482 | return ret; | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * If the device has been passed off to us in an enabled state, just clear | |
1487 | * the enabled bit. The spec says we should set the 'shutdown notification | |
1488 | * bits', but doing so may cause the device to complete commands to the | |
1489 | * admin queue ... and we don't know what memory that might be pointing at! | |
1490 | */ | |
1491 | int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap) | |
1492 | { | |
1493 | int ret; | |
1494 | ||
1495 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
1496 | ctrl->ctrl_config &= ~NVME_CC_ENABLE; | |
1497 | ||
1498 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
1499 | if (ret) | |
1500 | return ret; | |
54adc010 | 1501 | |
b5a10c5f | 1502 | if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) |
54adc010 GP |
1503 | msleep(NVME_QUIRK_DELAY_AMOUNT); |
1504 | ||
5fd4ce1b CH |
1505 | return nvme_wait_ready(ctrl, cap, false); |
1506 | } | |
576d55d6 | 1507 | EXPORT_SYMBOL_GPL(nvme_disable_ctrl); |
5fd4ce1b CH |
1508 | |
1509 | int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap) | |
1510 | { | |
1511 | /* | |
1512 | * Default to a 4K page size, with the intention to update this | |
1513 | * path in the future to accomodate architectures with differing | |
1514 | * kernel and IO page sizes. | |
1515 | */ | |
1516 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12; | |
1517 | int ret; | |
1518 | ||
1519 | if (page_shift < dev_page_min) { | |
1b3c47c1 | 1520 | dev_err(ctrl->device, |
5fd4ce1b CH |
1521 | "Minimum device page size %u too large for host (%u)\n", |
1522 | 1 << dev_page_min, 1 << page_shift); | |
1523 | return -ENODEV; | |
1524 | } | |
1525 | ||
1526 | ctrl->page_size = 1 << page_shift; | |
1527 | ||
1528 | ctrl->ctrl_config = NVME_CC_CSS_NVM; | |
1529 | ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; | |
60b43f62 | 1530 | ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE; |
5fd4ce1b CH |
1531 | ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
1532 | ctrl->ctrl_config |= NVME_CC_ENABLE; | |
1533 | ||
1534 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
1535 | if (ret) | |
1536 | return ret; | |
1537 | return nvme_wait_ready(ctrl, cap, true); | |
1538 | } | |
576d55d6 | 1539 | EXPORT_SYMBOL_GPL(nvme_enable_ctrl); |
5fd4ce1b CH |
1540 | |
1541 | int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl) | |
1542 | { | |
07fbd32a | 1543 | unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ); |
5fd4ce1b CH |
1544 | u32 csts; |
1545 | int ret; | |
1546 | ||
1547 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
1548 | ctrl->ctrl_config |= NVME_CC_SHN_NORMAL; | |
1549 | ||
1550 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
1551 | if (ret) | |
1552 | return ret; | |
1553 | ||
1554 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
1555 | if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT) | |
1556 | break; | |
1557 | ||
1558 | msleep(100); | |
1559 | if (fatal_signal_pending(current)) | |
1560 | return -EINTR; | |
1561 | if (time_after(jiffies, timeout)) { | |
1b3c47c1 | 1562 | dev_err(ctrl->device, |
5fd4ce1b CH |
1563 | "Device shutdown incomplete; abort shutdown\n"); |
1564 | return -ENODEV; | |
1565 | } | |
1566 | } | |
1567 | ||
1568 | return ret; | |
1569 | } | |
576d55d6 | 1570 | EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl); |
5fd4ce1b | 1571 | |
da35825d CH |
1572 | static void nvme_set_queue_limits(struct nvme_ctrl *ctrl, |
1573 | struct request_queue *q) | |
1574 | { | |
7c88cb00 JA |
1575 | bool vwc = false; |
1576 | ||
da35825d | 1577 | if (ctrl->max_hw_sectors) { |
45686b61 CH |
1578 | u32 max_segments = |
1579 | (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1; | |
1580 | ||
da35825d | 1581 | blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors); |
45686b61 | 1582 | blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX)); |
da35825d | 1583 | } |
e6282aef KB |
1584 | if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) |
1585 | blk_queue_chunk_sectors(q, ctrl->max_hw_sectors); | |
da35825d | 1586 | blk_queue_virt_boundary(q, ctrl->page_size - 1); |
7c88cb00 JA |
1587 | if (ctrl->vwc & NVME_CTRL_VWC_PRESENT) |
1588 | vwc = true; | |
1589 | blk_queue_write_cache(q, vwc, vwc); | |
da35825d CH |
1590 | } |
1591 | ||
dbf86b39 JD |
1592 | static int nvme_configure_timestamp(struct nvme_ctrl *ctrl) |
1593 | { | |
1594 | __le64 ts; | |
1595 | int ret; | |
1596 | ||
1597 | if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP)) | |
1598 | return 0; | |
1599 | ||
1600 | ts = cpu_to_le64(ktime_to_ms(ktime_get_real())); | |
1601 | ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts), | |
1602 | NULL); | |
1603 | if (ret) | |
1604 | dev_warn_once(ctrl->device, | |
1605 | "could not set timestamp (%d)\n", ret); | |
1606 | return ret; | |
1607 | } | |
1608 | ||
634b8325 | 1609 | static int nvme_configure_apst(struct nvme_ctrl *ctrl) |
c5552fde AL |
1610 | { |
1611 | /* | |
1612 | * APST (Autonomous Power State Transition) lets us program a | |
1613 | * table of power state transitions that the controller will | |
1614 | * perform automatically. We configure it with a simple | |
1615 | * heuristic: we are willing to spend at most 2% of the time | |
1616 | * transitioning between power states. Therefore, when running | |
1617 | * in any given state, we will enter the next lower-power | |
76e4ad09 | 1618 | * non-operational state after waiting 50 * (enlat + exlat) |
da87591b | 1619 | * microseconds, as long as that state's exit latency is under |
c5552fde AL |
1620 | * the requested maximum latency. |
1621 | * | |
1622 | * We will not autonomously enter any non-operational state for | |
1623 | * which the total latency exceeds ps_max_latency_us. Users | |
1624 | * can set ps_max_latency_us to zero to turn off APST. | |
1625 | */ | |
1626 | ||
1627 | unsigned apste; | |
1628 | struct nvme_feat_auto_pst *table; | |
fb0dc399 AL |
1629 | u64 max_lat_us = 0; |
1630 | int max_ps = -1; | |
c5552fde AL |
1631 | int ret; |
1632 | ||
1633 | /* | |
1634 | * If APST isn't supported or if we haven't been initialized yet, | |
1635 | * then don't do anything. | |
1636 | */ | |
1637 | if (!ctrl->apsta) | |
634b8325 | 1638 | return 0; |
c5552fde AL |
1639 | |
1640 | if (ctrl->npss > 31) { | |
1641 | dev_warn(ctrl->device, "NPSS is invalid; not using APST\n"); | |
634b8325 | 1642 | return 0; |
c5552fde AL |
1643 | } |
1644 | ||
1645 | table = kzalloc(sizeof(*table), GFP_KERNEL); | |
1646 | if (!table) | |
634b8325 | 1647 | return 0; |
c5552fde | 1648 | |
76a5af84 | 1649 | if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) { |
c5552fde AL |
1650 | /* Turn off APST. */ |
1651 | apste = 0; | |
fb0dc399 | 1652 | dev_dbg(ctrl->device, "APST disabled\n"); |
c5552fde AL |
1653 | } else { |
1654 | __le64 target = cpu_to_le64(0); | |
1655 | int state; | |
1656 | ||
1657 | /* | |
1658 | * Walk through all states from lowest- to highest-power. | |
1659 | * According to the spec, lower-numbered states use more | |
1660 | * power. NPSS, despite the name, is the index of the | |
1661 | * lowest-power state, not the number of states. | |
1662 | */ | |
1663 | for (state = (int)ctrl->npss; state >= 0; state--) { | |
da87591b | 1664 | u64 total_latency_us, exit_latency_us, transition_ms; |
c5552fde AL |
1665 | |
1666 | if (target) | |
1667 | table->entries[state] = target; | |
1668 | ||
ff5350a8 AL |
1669 | /* |
1670 | * Don't allow transitions to the deepest state | |
1671 | * if it's quirked off. | |
1672 | */ | |
1673 | if (state == ctrl->npss && | |
1674 | (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) | |
1675 | continue; | |
1676 | ||
c5552fde AL |
1677 | /* |
1678 | * Is this state a useful non-operational state for | |
1679 | * higher-power states to autonomously transition to? | |
1680 | */ | |
1681 | if (!(ctrl->psd[state].flags & | |
1682 | NVME_PS_FLAGS_NON_OP_STATE)) | |
1683 | continue; | |
1684 | ||
da87591b KHF |
1685 | exit_latency_us = |
1686 | (u64)le32_to_cpu(ctrl->psd[state].exit_lat); | |
1687 | if (exit_latency_us > ctrl->ps_max_latency_us) | |
c5552fde AL |
1688 | continue; |
1689 | ||
da87591b KHF |
1690 | total_latency_us = |
1691 | exit_latency_us + | |
1692 | le32_to_cpu(ctrl->psd[state].entry_lat); | |
1693 | ||
c5552fde AL |
1694 | /* |
1695 | * This state is good. Use it as the APST idle | |
1696 | * target for higher power states. | |
1697 | */ | |
1698 | transition_ms = total_latency_us + 19; | |
1699 | do_div(transition_ms, 20); | |
1700 | if (transition_ms > (1 << 24) - 1) | |
1701 | transition_ms = (1 << 24) - 1; | |
1702 | ||
1703 | target = cpu_to_le64((state << 3) | | |
1704 | (transition_ms << 8)); | |
fb0dc399 AL |
1705 | |
1706 | if (max_ps == -1) | |
1707 | max_ps = state; | |
1708 | ||
1709 | if (total_latency_us > max_lat_us) | |
1710 | max_lat_us = total_latency_us; | |
c5552fde AL |
1711 | } |
1712 | ||
1713 | apste = 1; | |
fb0dc399 AL |
1714 | |
1715 | if (max_ps == -1) { | |
1716 | dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n"); | |
1717 | } else { | |
1718 | dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n", | |
1719 | max_ps, max_lat_us, (int)sizeof(*table), table); | |
1720 | } | |
c5552fde AL |
1721 | } |
1722 | ||
1723 | ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste, | |
1724 | table, sizeof(*table), NULL); | |
1725 | if (ret) | |
1726 | dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret); | |
1727 | ||
1728 | kfree(table); | |
634b8325 | 1729 | return ret; |
c5552fde AL |
1730 | } |
1731 | ||
1732 | static void nvme_set_latency_tolerance(struct device *dev, s32 val) | |
1733 | { | |
1734 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
1735 | u64 latency; | |
1736 | ||
1737 | switch (val) { | |
1738 | case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT: | |
1739 | case PM_QOS_LATENCY_ANY: | |
1740 | latency = U64_MAX; | |
1741 | break; | |
1742 | ||
1743 | default: | |
1744 | latency = val; | |
1745 | } | |
1746 | ||
1747 | if (ctrl->ps_max_latency_us != latency) { | |
1748 | ctrl->ps_max_latency_us = latency; | |
1749 | nvme_configure_apst(ctrl); | |
1750 | } | |
1751 | } | |
1752 | ||
bd4da3ab AL |
1753 | struct nvme_core_quirk_entry { |
1754 | /* | |
1755 | * NVMe model and firmware strings are padded with spaces. For | |
1756 | * simplicity, strings in the quirk table are padded with NULLs | |
1757 | * instead. | |
1758 | */ | |
1759 | u16 vid; | |
1760 | const char *mn; | |
1761 | const char *fr; | |
1762 | unsigned long quirks; | |
1763 | }; | |
1764 | ||
1765 | static const struct nvme_core_quirk_entry core_quirks[] = { | |
c5552fde | 1766 | { |
be56945c AL |
1767 | /* |
1768 | * This Toshiba device seems to die using any APST states. See: | |
1769 | * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11 | |
1770 | */ | |
1771 | .vid = 0x1179, | |
1772 | .mn = "THNSF5256GPUK TOSHIBA", | |
c5552fde | 1773 | .quirks = NVME_QUIRK_NO_APST, |
be56945c | 1774 | } |
bd4da3ab AL |
1775 | }; |
1776 | ||
1777 | /* match is null-terminated but idstr is space-padded. */ | |
1778 | static bool string_matches(const char *idstr, const char *match, size_t len) | |
1779 | { | |
1780 | size_t matchlen; | |
1781 | ||
1782 | if (!match) | |
1783 | return true; | |
1784 | ||
1785 | matchlen = strlen(match); | |
1786 | WARN_ON_ONCE(matchlen > len); | |
1787 | ||
1788 | if (memcmp(idstr, match, matchlen)) | |
1789 | return false; | |
1790 | ||
1791 | for (; matchlen < len; matchlen++) | |
1792 | if (idstr[matchlen] != ' ') | |
1793 | return false; | |
1794 | ||
1795 | return true; | |
1796 | } | |
1797 | ||
1798 | static bool quirk_matches(const struct nvme_id_ctrl *id, | |
1799 | const struct nvme_core_quirk_entry *q) | |
1800 | { | |
1801 | return q->vid == le16_to_cpu(id->vid) && | |
1802 | string_matches(id->mn, q->mn, sizeof(id->mn)) && | |
1803 | string_matches(id->fr, q->fr, sizeof(id->fr)); | |
1804 | } | |
1805 | ||
180de007 CH |
1806 | static void nvme_init_subnqn(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id) |
1807 | { | |
1808 | size_t nqnlen; | |
1809 | int off; | |
1810 | ||
1811 | nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE); | |
1812 | if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) { | |
a47619b5 | 1813 | strncpy(ctrl->subnqn, id->subnqn, NVMF_NQN_SIZE); |
180de007 CH |
1814 | return; |
1815 | } | |
1816 | ||
1817 | if (ctrl->vs >= NVME_VS(1, 2, 1)) | |
1818 | dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n"); | |
1819 | ||
1820 | /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */ | |
1821 | off = snprintf(ctrl->subnqn, NVMF_NQN_SIZE, | |
1822 | "nqn.2014.08.org.nvmexpress:%4x%4x", | |
1823 | le16_to_cpu(id->vid), le16_to_cpu(id->ssvid)); | |
1824 | memcpy(ctrl->subnqn + off, id->sn, sizeof(id->sn)); | |
1825 | off += sizeof(id->sn); | |
1826 | memcpy(ctrl->subnqn + off, id->mn, sizeof(id->mn)); | |
1827 | off += sizeof(id->mn); | |
1828 | memset(ctrl->subnqn + off, 0, sizeof(ctrl->subnqn) - off); | |
1829 | } | |
1830 | ||
c627c487 KB |
1831 | static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log, |
1832 | size_t size) | |
1833 | { | |
1834 | struct nvme_command c = { }; | |
1835 | ||
1836 | c.common.opcode = nvme_admin_get_log_page; | |
1837 | c.common.nsid = cpu_to_le32(NVME_NSID_ALL); | |
1838 | c.common.cdw10[0] = nvme_get_log_dw10(log_page, size); | |
1839 | ||
1840 | return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size); | |
1841 | } | |
1842 | ||
84fef62d KB |
1843 | static int nvme_get_effects_log(struct nvme_ctrl *ctrl) |
1844 | { | |
1845 | int ret; | |
1846 | ||
1847 | if (!ctrl->effects) | |
1848 | ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL); | |
1849 | ||
1850 | if (!ctrl->effects) | |
1851 | return 0; | |
1852 | ||
1853 | ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects, | |
1854 | sizeof(*ctrl->effects)); | |
1855 | if (ret) { | |
1856 | kfree(ctrl->effects); | |
1857 | ctrl->effects = NULL; | |
1858 | } | |
1859 | return ret; | |
1860 | } | |
1861 | ||
7fd8930f CH |
1862 | /* |
1863 | * Initialize the cached copies of the Identify data and various controller | |
1864 | * register in our nvme_ctrl structure. This should be called as soon as | |
1865 | * the admin queue is fully up and running. | |
1866 | */ | |
1867 | int nvme_init_identify(struct nvme_ctrl *ctrl) | |
1868 | { | |
1869 | struct nvme_id_ctrl *id; | |
1870 | u64 cap; | |
1871 | int ret, page_shift; | |
a229dbf6 | 1872 | u32 max_hw_sectors; |
76a5af84 | 1873 | bool prev_apst_enabled; |
7fd8930f | 1874 | |
f3ca80fc CH |
1875 | ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs); |
1876 | if (ret) { | |
1b3c47c1 | 1877 | dev_err(ctrl->device, "Reading VS failed (%d)\n", ret); |
f3ca80fc CH |
1878 | return ret; |
1879 | } | |
1880 | ||
7fd8930f CH |
1881 | ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap); |
1882 | if (ret) { | |
1b3c47c1 | 1883 | dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret); |
7fd8930f CH |
1884 | return ret; |
1885 | } | |
1886 | page_shift = NVME_CAP_MPSMIN(cap) + 12; | |
1887 | ||
8ef2074d | 1888 | if (ctrl->vs >= NVME_VS(1, 1, 0)) |
f3ca80fc CH |
1889 | ctrl->subsystem = NVME_CAP_NSSRC(cap); |
1890 | ||
7fd8930f CH |
1891 | ret = nvme_identify_ctrl(ctrl, &id); |
1892 | if (ret) { | |
1b3c47c1 | 1893 | dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret); |
7fd8930f CH |
1894 | return -EIO; |
1895 | } | |
1896 | ||
84fef62d KB |
1897 | if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) { |
1898 | ret = nvme_get_effects_log(ctrl); | |
1899 | if (ret < 0) | |
1900 | return ret; | |
1901 | } | |
1902 | ||
180de007 CH |
1903 | nvme_init_subnqn(ctrl, id); |
1904 | ||
bd4da3ab AL |
1905 | if (!ctrl->identified) { |
1906 | /* | |
1907 | * Check for quirks. Quirk can depend on firmware version, | |
1908 | * so, in principle, the set of quirks present can change | |
1909 | * across a reset. As a possible future enhancement, we | |
1910 | * could re-scan for quirks every time we reinitialize | |
1911 | * the device, but we'd have to make sure that the driver | |
1912 | * behaves intelligently if the quirks change. | |
1913 | */ | |
1914 | ||
1915 | int i; | |
1916 | ||
1917 | for (i = 0; i < ARRAY_SIZE(core_quirks); i++) { | |
1918 | if (quirk_matches(id, &core_quirks[i])) | |
1919 | ctrl->quirks |= core_quirks[i].quirks; | |
1920 | } | |
1921 | } | |
1922 | ||
c35e30b4 | 1923 | if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) { |
f0425db0 | 1924 | dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n"); |
c35e30b4 AL |
1925 | ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS; |
1926 | } | |
1927 | ||
8a9ae523 | 1928 | ctrl->oacs = le16_to_cpu(id->oacs); |
118472ab | 1929 | ctrl->vid = le16_to_cpu(id->vid); |
7fd8930f | 1930 | ctrl->oncs = le16_to_cpup(&id->oncs); |
6bf25d16 | 1931 | atomic_set(&ctrl->abort_limit, id->acl + 1); |
7fd8930f | 1932 | ctrl->vwc = id->vwc; |
931e1c22 | 1933 | ctrl->cntlid = le16_to_cpup(&id->cntlid); |
7fd8930f CH |
1934 | memcpy(ctrl->serial, id->sn, sizeof(id->sn)); |
1935 | memcpy(ctrl->model, id->mn, sizeof(id->mn)); | |
1936 | memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr)); | |
1937 | if (id->mdts) | |
a229dbf6 | 1938 | max_hw_sectors = 1 << (id->mdts + page_shift - 9); |
7fd8930f | 1939 | else |
a229dbf6 CH |
1940 | max_hw_sectors = UINT_MAX; |
1941 | ctrl->max_hw_sectors = | |
1942 | min_not_zero(ctrl->max_hw_sectors, max_hw_sectors); | |
7fd8930f | 1943 | |
da35825d | 1944 | nvme_set_queue_limits(ctrl, ctrl->admin_q); |
07bfcd09 | 1945 | ctrl->sgls = le32_to_cpu(id->sgls); |
038bd4cb | 1946 | ctrl->kas = le16_to_cpu(id->kas); |
07bfcd09 | 1947 | |
07fbd32a MP |
1948 | if (id->rtd3e) { |
1949 | /* us -> s */ | |
1950 | u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000; | |
1951 | ||
1952 | ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time, | |
1953 | shutdown_timeout, 60); | |
1954 | ||
1955 | if (ctrl->shutdown_timeout != shutdown_timeout) | |
1956 | dev_warn(ctrl->device, | |
1957 | "Shutdown timeout set to %u seconds\n", | |
1958 | ctrl->shutdown_timeout); | |
1959 | } else | |
1960 | ctrl->shutdown_timeout = shutdown_timeout; | |
1961 | ||
c5552fde | 1962 | ctrl->npss = id->npss; |
76a5af84 KHF |
1963 | ctrl->apsta = id->apsta; |
1964 | prev_apst_enabled = ctrl->apst_enabled; | |
c35e30b4 AL |
1965 | if (ctrl->quirks & NVME_QUIRK_NO_APST) { |
1966 | if (force_apst && id->apsta) { | |
f0425db0 | 1967 | dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n"); |
76a5af84 | 1968 | ctrl->apst_enabled = true; |
c35e30b4 | 1969 | } else { |
76a5af84 | 1970 | ctrl->apst_enabled = false; |
c35e30b4 AL |
1971 | } |
1972 | } else { | |
76a5af84 | 1973 | ctrl->apst_enabled = id->apsta; |
c35e30b4 | 1974 | } |
c5552fde AL |
1975 | memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd)); |
1976 | ||
d3d5b87d | 1977 | if (ctrl->ops->flags & NVME_F_FABRICS) { |
07bfcd09 CH |
1978 | ctrl->icdoff = le16_to_cpu(id->icdoff); |
1979 | ctrl->ioccsz = le32_to_cpu(id->ioccsz); | |
1980 | ctrl->iorcsz = le32_to_cpu(id->iorcsz); | |
1981 | ctrl->maxcmd = le16_to_cpu(id->maxcmd); | |
1982 | ||
1983 | /* | |
1984 | * In fabrics we need to verify the cntlid matches the | |
1985 | * admin connect | |
1986 | */ | |
634b8325 | 1987 | if (ctrl->cntlid != le16_to_cpu(id->cntlid)) { |
07bfcd09 | 1988 | ret = -EINVAL; |
634b8325 KB |
1989 | goto out_free; |
1990 | } | |
038bd4cb SG |
1991 | |
1992 | if (!ctrl->opts->discovery_nqn && !ctrl->kas) { | |
f0425db0 | 1993 | dev_err(ctrl->device, |
038bd4cb SG |
1994 | "keep-alive support is mandatory for fabrics\n"); |
1995 | ret = -EINVAL; | |
634b8325 | 1996 | goto out_free; |
038bd4cb | 1997 | } |
07bfcd09 CH |
1998 | } else { |
1999 | ctrl->cntlid = le16_to_cpu(id->cntlid); | |
fe6d53c9 CH |
2000 | ctrl->hmpre = le32_to_cpu(id->hmpre); |
2001 | ctrl->hmmin = le32_to_cpu(id->hmmin); | |
044a9df1 CH |
2002 | ctrl->hmminds = le32_to_cpu(id->hmminds); |
2003 | ctrl->hmmaxd = le16_to_cpu(id->hmmaxd); | |
07bfcd09 | 2004 | } |
da35825d | 2005 | |
7fd8930f | 2006 | kfree(id); |
bd4da3ab | 2007 | |
76a5af84 | 2008 | if (ctrl->apst_enabled && !prev_apst_enabled) |
c5552fde | 2009 | dev_pm_qos_expose_latency_tolerance(ctrl->device); |
76a5af84 | 2010 | else if (!ctrl->apst_enabled && prev_apst_enabled) |
c5552fde AL |
2011 | dev_pm_qos_hide_latency_tolerance(ctrl->device); |
2012 | ||
634b8325 KB |
2013 | ret = nvme_configure_apst(ctrl); |
2014 | if (ret < 0) | |
2015 | return ret; | |
dbf86b39 JD |
2016 | |
2017 | ret = nvme_configure_timestamp(ctrl); | |
2018 | if (ret < 0) | |
2019 | return ret; | |
634b8325 KB |
2020 | |
2021 | ret = nvme_configure_directives(ctrl); | |
2022 | if (ret < 0) | |
2023 | return ret; | |
c5552fde | 2024 | |
bd4da3ab | 2025 | ctrl->identified = true; |
c5552fde | 2026 | |
634b8325 KB |
2027 | return 0; |
2028 | ||
2029 | out_free: | |
2030 | kfree(id); | |
07bfcd09 | 2031 | return ret; |
7fd8930f | 2032 | } |
576d55d6 | 2033 | EXPORT_SYMBOL_GPL(nvme_init_identify); |
7fd8930f | 2034 | |
f3ca80fc | 2035 | static int nvme_dev_open(struct inode *inode, struct file *file) |
1673f1f0 | 2036 | { |
a6a5149b CH |
2037 | struct nvme_ctrl *ctrl = |
2038 | container_of(inode->i_cdev, struct nvme_ctrl, cdev); | |
1673f1f0 | 2039 | |
999ada28 | 2040 | if (ctrl->state != NVME_CTRL_LIVE) |
a6a5149b CH |
2041 | return -EWOULDBLOCK; |
2042 | file->private_data = ctrl; | |
f3ca80fc CH |
2043 | return 0; |
2044 | } | |
2045 | ||
bfd89471 CH |
2046 | static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp) |
2047 | { | |
2048 | struct nvme_ns *ns; | |
2049 | int ret; | |
2050 | ||
2051 | mutex_lock(&ctrl->namespaces_mutex); | |
2052 | if (list_empty(&ctrl->namespaces)) { | |
2053 | ret = -ENOTTY; | |
2054 | goto out_unlock; | |
2055 | } | |
2056 | ||
2057 | ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list); | |
2058 | if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) { | |
1b3c47c1 | 2059 | dev_warn(ctrl->device, |
bfd89471 CH |
2060 | "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n"); |
2061 | ret = -EINVAL; | |
2062 | goto out_unlock; | |
2063 | } | |
2064 | ||
1b3c47c1 | 2065 | dev_warn(ctrl->device, |
bfd89471 CH |
2066 | "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n"); |
2067 | kref_get(&ns->kref); | |
2068 | mutex_unlock(&ctrl->namespaces_mutex); | |
2069 | ||
2070 | ret = nvme_user_cmd(ctrl, ns, argp); | |
2071 | nvme_put_ns(ns); | |
2072 | return ret; | |
2073 | ||
2074 | out_unlock: | |
2075 | mutex_unlock(&ctrl->namespaces_mutex); | |
2076 | return ret; | |
2077 | } | |
2078 | ||
f3ca80fc CH |
2079 | static long nvme_dev_ioctl(struct file *file, unsigned int cmd, |
2080 | unsigned long arg) | |
2081 | { | |
2082 | struct nvme_ctrl *ctrl = file->private_data; | |
2083 | void __user *argp = (void __user *)arg; | |
f3ca80fc CH |
2084 | |
2085 | switch (cmd) { | |
2086 | case NVME_IOCTL_ADMIN_CMD: | |
2087 | return nvme_user_cmd(ctrl, NULL, argp); | |
2088 | case NVME_IOCTL_IO_CMD: | |
bfd89471 | 2089 | return nvme_dev_user_cmd(ctrl, argp); |
f3ca80fc | 2090 | case NVME_IOCTL_RESET: |
1b3c47c1 | 2091 | dev_warn(ctrl->device, "resetting controller\n"); |
d86c4d8e | 2092 | return nvme_reset_ctrl_sync(ctrl); |
f3ca80fc CH |
2093 | case NVME_IOCTL_SUBSYS_RESET: |
2094 | return nvme_reset_subsystem(ctrl); | |
9ec3bb2f KB |
2095 | case NVME_IOCTL_RESCAN: |
2096 | nvme_queue_scan(ctrl); | |
2097 | return 0; | |
f3ca80fc CH |
2098 | default: |
2099 | return -ENOTTY; | |
2100 | } | |
2101 | } | |
2102 | ||
2103 | static const struct file_operations nvme_dev_fops = { | |
2104 | .owner = THIS_MODULE, | |
2105 | .open = nvme_dev_open, | |
f3ca80fc CH |
2106 | .unlocked_ioctl = nvme_dev_ioctl, |
2107 | .compat_ioctl = nvme_dev_ioctl, | |
2108 | }; | |
2109 | ||
2110 | static ssize_t nvme_sysfs_reset(struct device *dev, | |
2111 | struct device_attribute *attr, const char *buf, | |
2112 | size_t count) | |
2113 | { | |
2114 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2115 | int ret; | |
2116 | ||
d86c4d8e | 2117 | ret = nvme_reset_ctrl_sync(ctrl); |
f3ca80fc CH |
2118 | if (ret < 0) |
2119 | return ret; | |
2120 | return count; | |
1673f1f0 | 2121 | } |
f3ca80fc | 2122 | static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset); |
1673f1f0 | 2123 | |
9ec3bb2f KB |
2124 | static ssize_t nvme_sysfs_rescan(struct device *dev, |
2125 | struct device_attribute *attr, const char *buf, | |
2126 | size_t count) | |
2127 | { | |
2128 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2129 | ||
2130 | nvme_queue_scan(ctrl); | |
2131 | return count; | |
2132 | } | |
2133 | static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan); | |
2134 | ||
118472ab KB |
2135 | static ssize_t wwid_show(struct device *dev, struct device_attribute *attr, |
2136 | char *buf) | |
2137 | { | |
40267efd | 2138 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
118472ab KB |
2139 | struct nvme_ctrl *ctrl = ns->ctrl; |
2140 | int serial_len = sizeof(ctrl->serial); | |
2141 | int model_len = sizeof(ctrl->model); | |
2142 | ||
6484f5d1 JT |
2143 | if (!uuid_is_null(&ns->uuid)) |
2144 | return sprintf(buf, "uuid.%pU\n", &ns->uuid); | |
2145 | ||
90985b84 JT |
2146 | if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) |
2147 | return sprintf(buf, "eui.%16phN\n", ns->nguid); | |
118472ab KB |
2148 | |
2149 | if (memchr_inv(ns->eui, 0, sizeof(ns->eui))) | |
2150 | return sprintf(buf, "eui.%8phN\n", ns->eui); | |
2151 | ||
758f3735 MW |
2152 | while (serial_len > 0 && (ctrl->serial[serial_len - 1] == ' ' || |
2153 | ctrl->serial[serial_len - 1] == '\0')) | |
118472ab | 2154 | serial_len--; |
758f3735 MW |
2155 | while (model_len > 0 && (ctrl->model[model_len - 1] == ' ' || |
2156 | ctrl->model[model_len - 1] == '\0')) | |
118472ab KB |
2157 | model_len--; |
2158 | ||
2159 | return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid, | |
2160 | serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id); | |
2161 | } | |
2162 | static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL); | |
2163 | ||
d934f984 JT |
2164 | static ssize_t nguid_show(struct device *dev, struct device_attribute *attr, |
2165 | char *buf) | |
2166 | { | |
2167 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); | |
2168 | return sprintf(buf, "%pU\n", ns->nguid); | |
2169 | } | |
2170 | static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL); | |
2171 | ||
2b9b6e86 KB |
2172 | static ssize_t uuid_show(struct device *dev, struct device_attribute *attr, |
2173 | char *buf) | |
2174 | { | |
40267efd | 2175 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
d934f984 JT |
2176 | |
2177 | /* For backward compatibility expose the NGUID to userspace if | |
2178 | * we have no UUID set | |
2179 | */ | |
2180 | if (uuid_is_null(&ns->uuid)) { | |
2181 | printk_ratelimited(KERN_WARNING | |
2182 | "No UUID available providing old NGUID\n"); | |
2183 | return sprintf(buf, "%pU\n", ns->nguid); | |
2184 | } | |
2185 | return sprintf(buf, "%pU\n", &ns->uuid); | |
2b9b6e86 KB |
2186 | } |
2187 | static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL); | |
2188 | ||
2189 | static ssize_t eui_show(struct device *dev, struct device_attribute *attr, | |
2190 | char *buf) | |
2191 | { | |
40267efd | 2192 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
ab083b11 | 2193 | return sprintf(buf, "%8ph\n", ns->eui); |
2b9b6e86 KB |
2194 | } |
2195 | static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL); | |
2196 | ||
2197 | static ssize_t nsid_show(struct device *dev, struct device_attribute *attr, | |
2198 | char *buf) | |
2199 | { | |
40267efd | 2200 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
2b9b6e86 KB |
2201 | return sprintf(buf, "%d\n", ns->ns_id); |
2202 | } | |
2203 | static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL); | |
2204 | ||
2205 | static struct attribute *nvme_ns_attrs[] = { | |
118472ab | 2206 | &dev_attr_wwid.attr, |
2b9b6e86 | 2207 | &dev_attr_uuid.attr, |
d934f984 | 2208 | &dev_attr_nguid.attr, |
2b9b6e86 KB |
2209 | &dev_attr_eui.attr, |
2210 | &dev_attr_nsid.attr, | |
2211 | NULL, | |
2212 | }; | |
2213 | ||
1a353d85 | 2214 | static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj, |
2b9b6e86 KB |
2215 | struct attribute *a, int n) |
2216 | { | |
2217 | struct device *dev = container_of(kobj, struct device, kobj); | |
40267efd | 2218 | struct nvme_ns *ns = nvme_get_ns_from_dev(dev); |
2b9b6e86 KB |
2219 | |
2220 | if (a == &dev_attr_uuid.attr) { | |
d934f984 JT |
2221 | if (uuid_is_null(&ns->uuid) || |
2222 | !memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) | |
2223 | return 0; | |
2224 | } | |
2225 | if (a == &dev_attr_nguid.attr) { | |
90985b84 | 2226 | if (!memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) |
2b9b6e86 KB |
2227 | return 0; |
2228 | } | |
2229 | if (a == &dev_attr_eui.attr) { | |
2230 | if (!memchr_inv(ns->eui, 0, sizeof(ns->eui))) | |
2231 | return 0; | |
2232 | } | |
2233 | return a->mode; | |
2234 | } | |
2235 | ||
2236 | static const struct attribute_group nvme_ns_attr_group = { | |
2237 | .attrs = nvme_ns_attrs, | |
1a353d85 | 2238 | .is_visible = nvme_ns_attrs_are_visible, |
2b9b6e86 KB |
2239 | }; |
2240 | ||
931e1c22 | 2241 | #define nvme_show_str_function(field) \ |
779ff756 KB |
2242 | static ssize_t field##_show(struct device *dev, \ |
2243 | struct device_attribute *attr, char *buf) \ | |
2244 | { \ | |
2245 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ | |
2246 | return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \ | |
2247 | } \ | |
2248 | static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); | |
2249 | ||
931e1c22 ML |
2250 | #define nvme_show_int_function(field) \ |
2251 | static ssize_t field##_show(struct device *dev, \ | |
2252 | struct device_attribute *attr, char *buf) \ | |
2253 | { \ | |
2254 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ | |
2255 | return sprintf(buf, "%d\n", ctrl->field); \ | |
2256 | } \ | |
2257 | static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); | |
2258 | ||
2259 | nvme_show_str_function(model); | |
2260 | nvme_show_str_function(serial); | |
2261 | nvme_show_str_function(firmware_rev); | |
2262 | nvme_show_int_function(cntlid); | |
779ff756 | 2263 | |
1a353d85 ML |
2264 | static ssize_t nvme_sysfs_delete(struct device *dev, |
2265 | struct device_attribute *attr, const char *buf, | |
2266 | size_t count) | |
2267 | { | |
2268 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2269 | ||
2270 | if (device_remove_file_self(dev, attr)) | |
c5017e85 | 2271 | nvme_delete_ctrl_sync(ctrl); |
1a353d85 ML |
2272 | return count; |
2273 | } | |
2274 | static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete); | |
2275 | ||
2276 | static ssize_t nvme_sysfs_show_transport(struct device *dev, | |
2277 | struct device_attribute *attr, | |
2278 | char *buf) | |
2279 | { | |
2280 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2281 | ||
2282 | return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name); | |
2283 | } | |
2284 | static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL); | |
2285 | ||
8432bdb2 SG |
2286 | static ssize_t nvme_sysfs_show_state(struct device *dev, |
2287 | struct device_attribute *attr, | |
2288 | char *buf) | |
2289 | { | |
2290 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2291 | static const char *const state_name[] = { | |
2292 | [NVME_CTRL_NEW] = "new", | |
2293 | [NVME_CTRL_LIVE] = "live", | |
2294 | [NVME_CTRL_RESETTING] = "resetting", | |
2295 | [NVME_CTRL_RECONNECTING]= "reconnecting", | |
2296 | [NVME_CTRL_DELETING] = "deleting", | |
2297 | [NVME_CTRL_DEAD] = "dead", | |
2298 | }; | |
2299 | ||
2300 | if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) && | |
2301 | state_name[ctrl->state]) | |
2302 | return sprintf(buf, "%s\n", state_name[ctrl->state]); | |
2303 | ||
2304 | return sprintf(buf, "unknown state\n"); | |
2305 | } | |
2306 | ||
2307 | static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL); | |
2308 | ||
1a353d85 ML |
2309 | static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev, |
2310 | struct device_attribute *attr, | |
2311 | char *buf) | |
2312 | { | |
2313 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2314 | ||
180de007 | 2315 | return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subnqn); |
1a353d85 ML |
2316 | } |
2317 | static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL); | |
2318 | ||
2319 | static ssize_t nvme_sysfs_show_address(struct device *dev, | |
2320 | struct device_attribute *attr, | |
2321 | char *buf) | |
2322 | { | |
2323 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2324 | ||
2325 | return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE); | |
2326 | } | |
2327 | static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL); | |
2328 | ||
779ff756 KB |
2329 | static struct attribute *nvme_dev_attrs[] = { |
2330 | &dev_attr_reset_controller.attr, | |
9ec3bb2f | 2331 | &dev_attr_rescan_controller.attr, |
779ff756 KB |
2332 | &dev_attr_model.attr, |
2333 | &dev_attr_serial.attr, | |
2334 | &dev_attr_firmware_rev.attr, | |
931e1c22 | 2335 | &dev_attr_cntlid.attr, |
1a353d85 ML |
2336 | &dev_attr_delete_controller.attr, |
2337 | &dev_attr_transport.attr, | |
2338 | &dev_attr_subsysnqn.attr, | |
2339 | &dev_attr_address.attr, | |
8432bdb2 | 2340 | &dev_attr_state.attr, |
779ff756 KB |
2341 | NULL |
2342 | }; | |
2343 | ||
1a353d85 ML |
2344 | static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj, |
2345 | struct attribute *a, int n) | |
2346 | { | |
2347 | struct device *dev = container_of(kobj, struct device, kobj); | |
2348 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2349 | ||
49d3d50b CH |
2350 | if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl) |
2351 | return 0; | |
2352 | if (a == &dev_attr_address.attr && !ctrl->ops->get_address) | |
2353 | return 0; | |
1a353d85 ML |
2354 | |
2355 | return a->mode; | |
2356 | } | |
2357 | ||
779ff756 | 2358 | static struct attribute_group nvme_dev_attrs_group = { |
1a353d85 ML |
2359 | .attrs = nvme_dev_attrs, |
2360 | .is_visible = nvme_dev_attrs_are_visible, | |
779ff756 KB |
2361 | }; |
2362 | ||
2363 | static const struct attribute_group *nvme_dev_attr_groups[] = { | |
2364 | &nvme_dev_attrs_group, | |
2365 | NULL, | |
2366 | }; | |
2367 | ||
5bae7f73 CH |
2368 | static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) |
2369 | { | |
2370 | struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); | |
2371 | struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); | |
2372 | ||
2373 | return nsa->ns_id - nsb->ns_id; | |
2374 | } | |
2375 | ||
32f0c4af | 2376 | static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
5bae7f73 | 2377 | { |
32f0c4af | 2378 | struct nvme_ns *ns, *ret = NULL; |
69d3b8ac | 2379 | |
32f0c4af | 2380 | mutex_lock(&ctrl->namespaces_mutex); |
5bae7f73 | 2381 | list_for_each_entry(ns, &ctrl->namespaces, list) { |
32f0c4af | 2382 | if (ns->ns_id == nsid) { |
2dd41228 CH |
2383 | if (!kref_get_unless_zero(&ns->kref)) |
2384 | continue; | |
32f0c4af KB |
2385 | ret = ns; |
2386 | break; | |
2387 | } | |
5bae7f73 CH |
2388 | if (ns->ns_id > nsid) |
2389 | break; | |
2390 | } | |
32f0c4af KB |
2391 | mutex_unlock(&ctrl->namespaces_mutex); |
2392 | return ret; | |
5bae7f73 CH |
2393 | } |
2394 | ||
f5d11840 JA |
2395 | static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns) |
2396 | { | |
2397 | struct streams_directive_params s; | |
2398 | int ret; | |
2399 | ||
2400 | if (!ctrl->nr_streams) | |
2401 | return 0; | |
2402 | ||
2403 | ret = nvme_get_stream_params(ctrl, &s, ns->ns_id); | |
2404 | if (ret) | |
2405 | return ret; | |
2406 | ||
2407 | ns->sws = le32_to_cpu(s.sws); | |
2408 | ns->sgs = le16_to_cpu(s.sgs); | |
2409 | ||
2410 | if (ns->sws) { | |
2411 | unsigned int bs = 1 << ns->lba_shift; | |
2412 | ||
2413 | blk_queue_io_min(ns->queue, bs * ns->sws); | |
2414 | if (ns->sgs) | |
2415 | blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs); | |
2416 | } | |
2417 | ||
2418 | return 0; | |
2419 | } | |
2420 | ||
5bae7f73 CH |
2421 | static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
2422 | { | |
2423 | struct nvme_ns *ns; | |
2424 | struct gendisk *disk; | |
ac81bfa9 MB |
2425 | struct nvme_id_ns *id; |
2426 | char disk_name[DISK_NAME_LEN]; | |
5bae7f73 CH |
2427 | int node = dev_to_node(ctrl->dev); |
2428 | ||
2429 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); | |
2430 | if (!ns) | |
2431 | return; | |
2432 | ||
075790eb KB |
2433 | ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL); |
2434 | if (ns->instance < 0) | |
2435 | goto out_free_ns; | |
2436 | ||
5bae7f73 CH |
2437 | ns->queue = blk_mq_init_queue(ctrl->tagset); |
2438 | if (IS_ERR(ns->queue)) | |
075790eb | 2439 | goto out_release_instance; |
5bae7f73 CH |
2440 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); |
2441 | ns->queue->queuedata = ns; | |
2442 | ns->ctrl = ctrl; | |
2443 | ||
5bae7f73 CH |
2444 | kref_init(&ns->kref); |
2445 | ns->ns_id = nsid; | |
5bae7f73 | 2446 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ |
5bae7f73 CH |
2447 | |
2448 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); | |
da35825d | 2449 | nvme_set_queue_limits(ctrl, ns->queue); |
f5d11840 | 2450 | nvme_setup_streams_ns(ctrl, ns); |
5bae7f73 | 2451 | |
ac81bfa9 | 2452 | sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance); |
5bae7f73 | 2453 | |
cdbff4f2 CH |
2454 | id = nvme_identify_ns(ctrl, nsid); |
2455 | if (!id) | |
ac81bfa9 MB |
2456 | goto out_free_queue; |
2457 | ||
cdbff4f2 CH |
2458 | if (id->ncap == 0) |
2459 | goto out_free_id; | |
2460 | ||
2461 | nvme_report_ns_ids(ctrl, ns->ns_id, id, ns->eui, ns->nguid, &ns->uuid); | |
2462 | ||
608cc4b1 CH |
2463 | if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) { |
2464 | if (nvme_nvm_register(ns, disk_name, node)) { | |
2465 | dev_warn(ctrl->device, "LightNVM init failure\n"); | |
2466 | goto out_free_id; | |
2467 | } | |
3dc87dd0 | 2468 | } |
ac81bfa9 | 2469 | |
3dc87dd0 MB |
2470 | disk = alloc_disk_node(0, node); |
2471 | if (!disk) | |
2472 | goto out_free_id; | |
ac81bfa9 | 2473 | |
3dc87dd0 MB |
2474 | disk->fops = &nvme_fops; |
2475 | disk->private_data = ns; | |
2476 | disk->queue = ns->queue; | |
2477 | disk->flags = GENHD_FL_EXT_DEVT; | |
2478 | memcpy(disk->disk_name, disk_name, DISK_NAME_LEN); | |
2479 | ns->disk = disk; | |
2480 | ||
2481 | __nvme_revalidate_disk(disk, id); | |
5bae7f73 | 2482 | |
32f0c4af KB |
2483 | mutex_lock(&ctrl->namespaces_mutex); |
2484 | list_add_tail(&ns->list, &ctrl->namespaces); | |
2485 | mutex_unlock(&ctrl->namespaces_mutex); | |
2486 | ||
d22524a4 | 2487 | nvme_get_ctrl(ctrl); |
ac81bfa9 MB |
2488 | |
2489 | kfree(id); | |
2490 | ||
0d52c756 | 2491 | device_add_disk(ctrl->device, ns->disk); |
2b9b6e86 KB |
2492 | if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj, |
2493 | &nvme_ns_attr_group)) | |
2494 | pr_warn("%s: failed to create sysfs group for identification\n", | |
2495 | ns->disk->disk_name); | |
3dc87dd0 MB |
2496 | if (ns->ndev && nvme_nvm_register_sysfs(ns)) |
2497 | pr_warn("%s: failed to register lightnvm sysfs group for identification\n", | |
2498 | ns->disk->disk_name); | |
5bae7f73 | 2499 | return; |
ac81bfa9 MB |
2500 | out_free_id: |
2501 | kfree(id); | |
5bae7f73 CH |
2502 | out_free_queue: |
2503 | blk_cleanup_queue(ns->queue); | |
075790eb KB |
2504 | out_release_instance: |
2505 | ida_simple_remove(&ctrl->ns_ida, ns->instance); | |
5bae7f73 CH |
2506 | out_free_ns: |
2507 | kfree(ns); | |
2508 | } | |
2509 | ||
2510 | static void nvme_ns_remove(struct nvme_ns *ns) | |
2511 | { | |
646017a6 KB |
2512 | if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags)) |
2513 | return; | |
69d3b8ac | 2514 | |
b0b4e09c | 2515 | if (ns->disk && ns->disk->flags & GENHD_FL_UP) { |
5bae7f73 CH |
2516 | if (blk_get_integrity(ns->disk)) |
2517 | blk_integrity_unregister(ns->disk); | |
2b9b6e86 KB |
2518 | sysfs_remove_group(&disk_to_dev(ns->disk)->kobj, |
2519 | &nvme_ns_attr_group); | |
3dc87dd0 MB |
2520 | if (ns->ndev) |
2521 | nvme_nvm_unregister_sysfs(ns); | |
5bae7f73 | 2522 | del_gendisk(ns->disk); |
5bae7f73 CH |
2523 | blk_cleanup_queue(ns->queue); |
2524 | } | |
32f0c4af KB |
2525 | |
2526 | mutex_lock(&ns->ctrl->namespaces_mutex); | |
5bae7f73 | 2527 | list_del_init(&ns->list); |
32f0c4af KB |
2528 | mutex_unlock(&ns->ctrl->namespaces_mutex); |
2529 | ||
5bae7f73 CH |
2530 | nvme_put_ns(ns); |
2531 | } | |
2532 | ||
540c801c KB |
2533 | static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
2534 | { | |
2535 | struct nvme_ns *ns; | |
2536 | ||
32f0c4af | 2537 | ns = nvme_find_get_ns(ctrl, nsid); |
540c801c | 2538 | if (ns) { |
b0b4e09c | 2539 | if (ns->disk && revalidate_disk(ns->disk)) |
540c801c | 2540 | nvme_ns_remove(ns); |
32f0c4af | 2541 | nvme_put_ns(ns); |
540c801c KB |
2542 | } else |
2543 | nvme_alloc_ns(ctrl, nsid); | |
2544 | } | |
2545 | ||
47b0e50a SB |
2546 | static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl, |
2547 | unsigned nsid) | |
2548 | { | |
2549 | struct nvme_ns *ns, *next; | |
2550 | ||
2551 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) { | |
2552 | if (ns->ns_id > nsid) | |
2553 | nvme_ns_remove(ns); | |
2554 | } | |
2555 | } | |
2556 | ||
540c801c KB |
2557 | static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn) |
2558 | { | |
2559 | struct nvme_ns *ns; | |
2560 | __le32 *ns_list; | |
2561 | unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024); | |
2562 | int ret = 0; | |
2563 | ||
2564 | ns_list = kzalloc(0x1000, GFP_KERNEL); | |
2565 | if (!ns_list) | |
2566 | return -ENOMEM; | |
2567 | ||
2568 | for (i = 0; i < num_lists; i++) { | |
2569 | ret = nvme_identify_ns_list(ctrl, prev, ns_list); | |
2570 | if (ret) | |
47b0e50a | 2571 | goto free; |
540c801c KB |
2572 | |
2573 | for (j = 0; j < min(nn, 1024U); j++) { | |
2574 | nsid = le32_to_cpu(ns_list[j]); | |
2575 | if (!nsid) | |
2576 | goto out; | |
2577 | ||
2578 | nvme_validate_ns(ctrl, nsid); | |
2579 | ||
2580 | while (++prev < nsid) { | |
32f0c4af KB |
2581 | ns = nvme_find_get_ns(ctrl, prev); |
2582 | if (ns) { | |
540c801c | 2583 | nvme_ns_remove(ns); |
32f0c4af KB |
2584 | nvme_put_ns(ns); |
2585 | } | |
540c801c KB |
2586 | } |
2587 | } | |
2588 | nn -= j; | |
2589 | } | |
2590 | out: | |
47b0e50a SB |
2591 | nvme_remove_invalid_namespaces(ctrl, prev); |
2592 | free: | |
540c801c KB |
2593 | kfree(ns_list); |
2594 | return ret; | |
2595 | } | |
2596 | ||
5955be21 | 2597 | static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn) |
5bae7f73 | 2598 | { |
5bae7f73 CH |
2599 | unsigned i; |
2600 | ||
540c801c KB |
2601 | for (i = 1; i <= nn; i++) |
2602 | nvme_validate_ns(ctrl, i); | |
2603 | ||
47b0e50a | 2604 | nvme_remove_invalid_namespaces(ctrl, nn); |
5bae7f73 CH |
2605 | } |
2606 | ||
5955be21 | 2607 | static void nvme_scan_work(struct work_struct *work) |
5bae7f73 | 2608 | { |
5955be21 CH |
2609 | struct nvme_ctrl *ctrl = |
2610 | container_of(work, struct nvme_ctrl, scan_work); | |
5bae7f73 | 2611 | struct nvme_id_ctrl *id; |
540c801c | 2612 | unsigned nn; |
5bae7f73 | 2613 | |
5955be21 CH |
2614 | if (ctrl->state != NVME_CTRL_LIVE) |
2615 | return; | |
2616 | ||
5bae7f73 CH |
2617 | if (nvme_identify_ctrl(ctrl, &id)) |
2618 | return; | |
540c801c KB |
2619 | |
2620 | nn = le32_to_cpu(id->nn); | |
8ef2074d | 2621 | if (ctrl->vs >= NVME_VS(1, 1, 0) && |
540c801c KB |
2622 | !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) { |
2623 | if (!nvme_scan_ns_list(ctrl, nn)) | |
2624 | goto done; | |
2625 | } | |
5955be21 | 2626 | nvme_scan_ns_sequential(ctrl, nn); |
540c801c | 2627 | done: |
32f0c4af | 2628 | mutex_lock(&ctrl->namespaces_mutex); |
540c801c | 2629 | list_sort(NULL, &ctrl->namespaces, ns_cmp); |
69d3b8ac | 2630 | mutex_unlock(&ctrl->namespaces_mutex); |
5bae7f73 CH |
2631 | kfree(id); |
2632 | } | |
5955be21 CH |
2633 | |
2634 | void nvme_queue_scan(struct nvme_ctrl *ctrl) | |
2635 | { | |
2636 | /* | |
2637 | * Do not queue new scan work when a controller is reset during | |
2638 | * removal. | |
2639 | */ | |
2640 | if (ctrl->state == NVME_CTRL_LIVE) | |
c669ccdc | 2641 | queue_work(nvme_wq, &ctrl->scan_work); |
5955be21 CH |
2642 | } |
2643 | EXPORT_SYMBOL_GPL(nvme_queue_scan); | |
5bae7f73 | 2644 | |
32f0c4af KB |
2645 | /* |
2646 | * This function iterates the namespace list unlocked to allow recovery from | |
2647 | * controller failure. It is up to the caller to ensure the namespace list is | |
2648 | * not modified by scan work while this function is executing. | |
2649 | */ | |
5bae7f73 CH |
2650 | void nvme_remove_namespaces(struct nvme_ctrl *ctrl) |
2651 | { | |
2652 | struct nvme_ns *ns, *next; | |
2653 | ||
0ff9d4e1 KB |
2654 | /* |
2655 | * The dead states indicates the controller was not gracefully | |
2656 | * disconnected. In that case, we won't be able to flush any data while | |
2657 | * removing the namespaces' disks; fail all the queues now to avoid | |
2658 | * potentially having to clean up the failed sync later. | |
2659 | */ | |
2660 | if (ctrl->state == NVME_CTRL_DEAD) | |
2661 | nvme_kill_queues(ctrl); | |
2662 | ||
5bae7f73 CH |
2663 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) |
2664 | nvme_ns_remove(ns); | |
2665 | } | |
576d55d6 | 2666 | EXPORT_SYMBOL_GPL(nvme_remove_namespaces); |
5bae7f73 | 2667 | |
e3d7874d KB |
2668 | static void nvme_aen_uevent(struct nvme_ctrl *ctrl) |
2669 | { | |
2670 | char *envp[2] = { NULL, NULL }; | |
2671 | u32 aen_result = ctrl->aen_result; | |
2672 | ||
2673 | ctrl->aen_result = 0; | |
2674 | if (!aen_result) | |
2675 | return; | |
2676 | ||
2677 | envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result); | |
2678 | if (!envp[0]) | |
2679 | return; | |
2680 | kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp); | |
2681 | kfree(envp[0]); | |
2682 | } | |
2683 | ||
f866fc42 CH |
2684 | static void nvme_async_event_work(struct work_struct *work) |
2685 | { | |
2686 | struct nvme_ctrl *ctrl = | |
2687 | container_of(work, struct nvme_ctrl, async_event_work); | |
2688 | ||
e3d7874d | 2689 | nvme_aen_uevent(ctrl); |
ad22c355 | 2690 | ctrl->ops->submit_async_event(ctrl); |
f866fc42 CH |
2691 | } |
2692 | ||
b6dccf7f AD |
2693 | static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl) |
2694 | { | |
2695 | ||
2696 | u32 csts; | |
2697 | ||
2698 | if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) | |
2699 | return false; | |
2700 | ||
2701 | if (csts == ~0) | |
2702 | return false; | |
2703 | ||
2704 | return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP)); | |
2705 | } | |
2706 | ||
2707 | static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl) | |
2708 | { | |
b6dccf7f AD |
2709 | struct nvme_fw_slot_info_log *log; |
2710 | ||
2711 | log = kmalloc(sizeof(*log), GFP_KERNEL); | |
2712 | if (!log) | |
2713 | return; | |
2714 | ||
c627c487 | 2715 | if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log))) |
b6dccf7f AD |
2716 | dev_warn(ctrl->device, |
2717 | "Get FW SLOT INFO log error\n"); | |
2718 | kfree(log); | |
2719 | } | |
2720 | ||
2721 | static void nvme_fw_act_work(struct work_struct *work) | |
2722 | { | |
2723 | struct nvme_ctrl *ctrl = container_of(work, | |
2724 | struct nvme_ctrl, fw_act_work); | |
2725 | unsigned long fw_act_timeout; | |
2726 | ||
2727 | if (ctrl->mtfa) | |
2728 | fw_act_timeout = jiffies + | |
2729 | msecs_to_jiffies(ctrl->mtfa * 100); | |
2730 | else | |
2731 | fw_act_timeout = jiffies + | |
2732 | msecs_to_jiffies(admin_timeout * 1000); | |
2733 | ||
2734 | nvme_stop_queues(ctrl); | |
2735 | while (nvme_ctrl_pp_status(ctrl)) { | |
2736 | if (time_after(jiffies, fw_act_timeout)) { | |
2737 | dev_warn(ctrl->device, | |
2738 | "Fw activation timeout, reset controller\n"); | |
2739 | nvme_reset_ctrl(ctrl); | |
2740 | break; | |
2741 | } | |
2742 | msleep(100); | |
2743 | } | |
2744 | ||
2745 | if (ctrl->state != NVME_CTRL_LIVE) | |
2746 | return; | |
2747 | ||
2748 | nvme_start_queues(ctrl); | |
a806c6c8 | 2749 | /* read FW slot information to clear the AER */ |
b6dccf7f AD |
2750 | nvme_get_fw_slot_info(ctrl); |
2751 | } | |
2752 | ||
7bf58533 CH |
2753 | void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, |
2754 | union nvme_result *res) | |
f866fc42 | 2755 | { |
7bf58533 | 2756 | u32 result = le32_to_cpu(res->u32); |
f866fc42 | 2757 | |
ad22c355 | 2758 | if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS) |
f866fc42 CH |
2759 | return; |
2760 | ||
e3d7874d KB |
2761 | switch (result & 0x7) { |
2762 | case NVME_AER_ERROR: | |
2763 | case NVME_AER_SMART: | |
2764 | case NVME_AER_CSS: | |
2765 | case NVME_AER_VS: | |
2766 | ctrl->aen_result = result; | |
2767 | break; | |
2768 | default: | |
2769 | break; | |
2770 | } | |
2771 | ||
f866fc42 CH |
2772 | switch (result & 0xff07) { |
2773 | case NVME_AER_NOTICE_NS_CHANGED: | |
2774 | dev_info(ctrl->device, "rescanning\n"); | |
2775 | nvme_queue_scan(ctrl); | |
2776 | break; | |
b6dccf7f | 2777 | case NVME_AER_NOTICE_FW_ACT_STARTING: |
1a40d972 | 2778 | queue_work(nvme_wq, &ctrl->fw_act_work); |
b6dccf7f | 2779 | break; |
f866fc42 CH |
2780 | default: |
2781 | dev_warn(ctrl->device, "async event result %08x\n", result); | |
2782 | } | |
ad22c355 | 2783 | queue_work(nvme_wq, &ctrl->async_event_work); |
f866fc42 CH |
2784 | } |
2785 | EXPORT_SYMBOL_GPL(nvme_complete_async_event); | |
2786 | ||
d09f2b45 | 2787 | void nvme_stop_ctrl(struct nvme_ctrl *ctrl) |
576d55d6 | 2788 | { |
d09f2b45 | 2789 | nvme_stop_keep_alive(ctrl); |
f866fc42 | 2790 | flush_work(&ctrl->async_event_work); |
5955be21 | 2791 | flush_work(&ctrl->scan_work); |
b6dccf7f | 2792 | cancel_work_sync(&ctrl->fw_act_work); |
d09f2b45 SG |
2793 | } |
2794 | EXPORT_SYMBOL_GPL(nvme_stop_ctrl); | |
2795 | ||
2796 | void nvme_start_ctrl(struct nvme_ctrl *ctrl) | |
2797 | { | |
2798 | if (ctrl->kato) | |
2799 | nvme_start_keep_alive(ctrl); | |
2800 | ||
2801 | if (ctrl->queue_count > 1) { | |
2802 | nvme_queue_scan(ctrl); | |
d99ca609 | 2803 | queue_work(nvme_wq, &ctrl->async_event_work); |
d09f2b45 SG |
2804 | nvme_start_queues(ctrl); |
2805 | } | |
2806 | } | |
2807 | EXPORT_SYMBOL_GPL(nvme_start_ctrl); | |
5955be21 | 2808 | |
d09f2b45 SG |
2809 | void nvme_uninit_ctrl(struct nvme_ctrl *ctrl) |
2810 | { | |
a6a5149b | 2811 | cdev_device_del(&ctrl->cdev, ctrl->device); |
53029b04 | 2812 | } |
576d55d6 | 2813 | EXPORT_SYMBOL_GPL(nvme_uninit_ctrl); |
53029b04 | 2814 | |
d22524a4 | 2815 | static void nvme_free_ctrl(struct device *dev) |
53029b04 | 2816 | { |
d22524a4 CH |
2817 | struct nvme_ctrl *ctrl = |
2818 | container_of(dev, struct nvme_ctrl, ctrl_device); | |
f3ca80fc | 2819 | |
9843f685 | 2820 | ida_simple_remove(&nvme_instance_ida, ctrl->instance); |
075790eb | 2821 | ida_destroy(&ctrl->ns_ida); |
84fef62d | 2822 | kfree(ctrl->effects); |
f3ca80fc CH |
2823 | |
2824 | ctrl->ops->free_ctrl(ctrl); | |
2825 | } | |
2826 | ||
f3ca80fc CH |
2827 | /* |
2828 | * Initialize a NVMe controller structures. This needs to be called during | |
2829 | * earliest initialization so that we have the initialized structured around | |
2830 | * during probing. | |
2831 | */ | |
2832 | int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, | |
2833 | const struct nvme_ctrl_ops *ops, unsigned long quirks) | |
2834 | { | |
2835 | int ret; | |
2836 | ||
bb8d261e CH |
2837 | ctrl->state = NVME_CTRL_NEW; |
2838 | spin_lock_init(&ctrl->lock); | |
f3ca80fc | 2839 | INIT_LIST_HEAD(&ctrl->namespaces); |
69d3b8ac | 2840 | mutex_init(&ctrl->namespaces_mutex); |
f3ca80fc CH |
2841 | ctrl->dev = dev; |
2842 | ctrl->ops = ops; | |
2843 | ctrl->quirks = quirks; | |
5955be21 | 2844 | INIT_WORK(&ctrl->scan_work, nvme_scan_work); |
f866fc42 | 2845 | INIT_WORK(&ctrl->async_event_work, nvme_async_event_work); |
b6dccf7f | 2846 | INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work); |
c5017e85 | 2847 | INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work); |
f3ca80fc | 2848 | |
9843f685 CH |
2849 | ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL); |
2850 | if (ret < 0) | |
f3ca80fc | 2851 | goto out; |
9843f685 | 2852 | ctrl->instance = ret; |
f3ca80fc | 2853 | |
d22524a4 CH |
2854 | device_initialize(&ctrl->ctrl_device); |
2855 | ctrl->device = &ctrl->ctrl_device; | |
a6a5149b | 2856 | ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance); |
d22524a4 CH |
2857 | ctrl->device->class = nvme_class; |
2858 | ctrl->device->parent = ctrl->dev; | |
2859 | ctrl->device->groups = nvme_dev_attr_groups; | |
2860 | ctrl->device->release = nvme_free_ctrl; | |
2861 | dev_set_drvdata(ctrl->device, ctrl); | |
2862 | ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance); | |
2863 | if (ret) | |
f3ca80fc | 2864 | goto out_release_instance; |
a6a5149b CH |
2865 | |
2866 | cdev_init(&ctrl->cdev, &nvme_dev_fops); | |
2867 | ctrl->cdev.owner = ops->module; | |
2868 | ret = cdev_device_add(&ctrl->cdev, ctrl->device); | |
d22524a4 CH |
2869 | if (ret) |
2870 | goto out_free_name; | |
2871 | ||
075790eb | 2872 | ida_init(&ctrl->ns_ida); |
f3ca80fc | 2873 | |
c5552fde AL |
2874 | /* |
2875 | * Initialize latency tolerance controls. The sysfs files won't | |
2876 | * be visible to userspace unless the device actually supports APST. | |
2877 | */ | |
2878 | ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance; | |
2879 | dev_pm_qos_update_user_latency_tolerance(ctrl->device, | |
2880 | min(default_ps_max_latency_us, (unsigned long)S32_MAX)); | |
2881 | ||
f3ca80fc | 2882 | return 0; |
d22524a4 CH |
2883 | out_free_name: |
2884 | kfree_const(dev->kobj.name); | |
f3ca80fc | 2885 | out_release_instance: |
9843f685 | 2886 | ida_simple_remove(&nvme_instance_ida, ctrl->instance); |
f3ca80fc CH |
2887 | out: |
2888 | return ret; | |
2889 | } | |
576d55d6 | 2890 | EXPORT_SYMBOL_GPL(nvme_init_ctrl); |
f3ca80fc | 2891 | |
69d9a99c KB |
2892 | /** |
2893 | * nvme_kill_queues(): Ends all namespace queues | |
2894 | * @ctrl: the dead controller that needs to end | |
2895 | * | |
2896 | * Call this function when the driver determines it is unable to get the | |
2897 | * controller in a state capable of servicing IO. | |
2898 | */ | |
2899 | void nvme_kill_queues(struct nvme_ctrl *ctrl) | |
2900 | { | |
2901 | struct nvme_ns *ns; | |
2902 | ||
32f0c4af | 2903 | mutex_lock(&ctrl->namespaces_mutex); |
82654b6b | 2904 | |
443bd90f | 2905 | /* Forcibly unquiesce queues to avoid blocking dispatch */ |
7dd1ab16 SB |
2906 | if (ctrl->admin_q) |
2907 | blk_mq_unquiesce_queue(ctrl->admin_q); | |
443bd90f | 2908 | |
32f0c4af | 2909 | list_for_each_entry(ns, &ctrl->namespaces, list) { |
69d9a99c KB |
2910 | /* |
2911 | * Revalidating a dead namespace sets capacity to 0. This will | |
2912 | * end buffered writers dirtying pages that can't be synced. | |
2913 | */ | |
f33447b9 KB |
2914 | if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags)) |
2915 | continue; | |
2916 | revalidate_disk(ns->disk); | |
69d9a99c | 2917 | blk_set_queue_dying(ns->queue); |
806f026f | 2918 | |
443bd90f ML |
2919 | /* Forcibly unquiesce queues to avoid blocking dispatch */ |
2920 | blk_mq_unquiesce_queue(ns->queue); | |
69d9a99c | 2921 | } |
32f0c4af | 2922 | mutex_unlock(&ctrl->namespaces_mutex); |
69d9a99c | 2923 | } |
237045fc | 2924 | EXPORT_SYMBOL_GPL(nvme_kill_queues); |
69d9a99c | 2925 | |
302ad8cc KB |
2926 | void nvme_unfreeze(struct nvme_ctrl *ctrl) |
2927 | { | |
2928 | struct nvme_ns *ns; | |
2929 | ||
2930 | mutex_lock(&ctrl->namespaces_mutex); | |
2931 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
2932 | blk_mq_unfreeze_queue(ns->queue); | |
2933 | mutex_unlock(&ctrl->namespaces_mutex); | |
2934 | } | |
2935 | EXPORT_SYMBOL_GPL(nvme_unfreeze); | |
2936 | ||
2937 | void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout) | |
2938 | { | |
2939 | struct nvme_ns *ns; | |
2940 | ||
2941 | mutex_lock(&ctrl->namespaces_mutex); | |
2942 | list_for_each_entry(ns, &ctrl->namespaces, list) { | |
2943 | timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout); | |
2944 | if (timeout <= 0) | |
2945 | break; | |
2946 | } | |
2947 | mutex_unlock(&ctrl->namespaces_mutex); | |
2948 | } | |
2949 | EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout); | |
2950 | ||
2951 | void nvme_wait_freeze(struct nvme_ctrl *ctrl) | |
2952 | { | |
2953 | struct nvme_ns *ns; | |
2954 | ||
2955 | mutex_lock(&ctrl->namespaces_mutex); | |
2956 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
2957 | blk_mq_freeze_queue_wait(ns->queue); | |
2958 | mutex_unlock(&ctrl->namespaces_mutex); | |
2959 | } | |
2960 | EXPORT_SYMBOL_GPL(nvme_wait_freeze); | |
2961 | ||
2962 | void nvme_start_freeze(struct nvme_ctrl *ctrl) | |
2963 | { | |
2964 | struct nvme_ns *ns; | |
2965 | ||
2966 | mutex_lock(&ctrl->namespaces_mutex); | |
2967 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
1671d522 | 2968 | blk_freeze_queue_start(ns->queue); |
302ad8cc KB |
2969 | mutex_unlock(&ctrl->namespaces_mutex); |
2970 | } | |
2971 | EXPORT_SYMBOL_GPL(nvme_start_freeze); | |
2972 | ||
25646264 | 2973 | void nvme_stop_queues(struct nvme_ctrl *ctrl) |
363c9aac SG |
2974 | { |
2975 | struct nvme_ns *ns; | |
2976 | ||
32f0c4af | 2977 | mutex_lock(&ctrl->namespaces_mutex); |
a6eaa884 | 2978 | list_for_each_entry(ns, &ctrl->namespaces, list) |
3174dd33 | 2979 | blk_mq_quiesce_queue(ns->queue); |
32f0c4af | 2980 | mutex_unlock(&ctrl->namespaces_mutex); |
363c9aac | 2981 | } |
576d55d6 | 2982 | EXPORT_SYMBOL_GPL(nvme_stop_queues); |
363c9aac | 2983 | |
25646264 | 2984 | void nvme_start_queues(struct nvme_ctrl *ctrl) |
363c9aac SG |
2985 | { |
2986 | struct nvme_ns *ns; | |
2987 | ||
32f0c4af | 2988 | mutex_lock(&ctrl->namespaces_mutex); |
8d7b8faf | 2989 | list_for_each_entry(ns, &ctrl->namespaces, list) |
f660174e | 2990 | blk_mq_unquiesce_queue(ns->queue); |
32f0c4af | 2991 | mutex_unlock(&ctrl->namespaces_mutex); |
363c9aac | 2992 | } |
576d55d6 | 2993 | EXPORT_SYMBOL_GPL(nvme_start_queues); |
363c9aac | 2994 | |
31b84460 SG |
2995 | int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set) |
2996 | { | |
2997 | if (!ctrl->ops->reinit_request) | |
2998 | return 0; | |
2999 | ||
3000 | return blk_mq_tagset_iter(set, set->driver_data, | |
3001 | ctrl->ops->reinit_request); | |
3002 | } | |
3003 | EXPORT_SYMBOL_GPL(nvme_reinit_tagset); | |
3004 | ||
5bae7f73 CH |
3005 | int __init nvme_core_init(void) |
3006 | { | |
3007 | int result; | |
3008 | ||
9a6327d2 SG |
3009 | nvme_wq = alloc_workqueue("nvme-wq", |
3010 | WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); | |
3011 | if (!nvme_wq) | |
3012 | return -ENOMEM; | |
3013 | ||
a6a5149b | 3014 | result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme"); |
f3ca80fc | 3015 | if (result < 0) |
9a6327d2 | 3016 | goto destroy_wq; |
f3ca80fc CH |
3017 | |
3018 | nvme_class = class_create(THIS_MODULE, "nvme"); | |
3019 | if (IS_ERR(nvme_class)) { | |
3020 | result = PTR_ERR(nvme_class); | |
3021 | goto unregister_chrdev; | |
3022 | } | |
3023 | ||
5bae7f73 | 3024 | return 0; |
f3ca80fc | 3025 | |
9a6327d2 | 3026 | unregister_chrdev: |
a6a5149b | 3027 | unregister_chrdev_region(nvme_chr_devt, NVME_MINORS); |
9a6327d2 SG |
3028 | destroy_wq: |
3029 | destroy_workqueue(nvme_wq); | |
f3ca80fc | 3030 | return result; |
5bae7f73 CH |
3031 | } |
3032 | ||
3033 | void nvme_core_exit(void) | |
3034 | { | |
f3ca80fc | 3035 | class_destroy(nvme_class); |
a6a5149b | 3036 | unregister_chrdev_region(nvme_chr_devt, NVME_MINORS); |
9a6327d2 | 3037 | destroy_workqueue(nvme_wq); |
5bae7f73 | 3038 | } |
576d55d6 ML |
3039 | |
3040 | MODULE_LICENSE("GPL"); | |
3041 | MODULE_VERSION("1.0"); | |
3042 | module_init(nvme_core_init); | |
3043 | module_exit(nvme_core_exit); |