]>
Commit | Line | Data |
---|---|---|
bc50ad75 | 1 | // SPDX-License-Identifier: GPL-2.0 |
21d34711 CH |
2 | /* |
3 | * NVM Express device driver | |
4 | * Copyright (c) 2011-2014, Intel Corporation. | |
21d34711 CH |
5 | */ |
6 | ||
7 | #include <linux/blkdev.h> | |
8 | #include <linux/blk-mq.h> | |
c95b708d | 9 | #include <linux/compat.h> |
5fd4ce1b | 10 | #include <linux/delay.h> |
21d34711 | 11 | #include <linux/errno.h> |
1673f1f0 | 12 | #include <linux/hdreg.h> |
21d34711 | 13 | #include <linux/kernel.h> |
5bae7f73 | 14 | #include <linux/module.h> |
958f2a0f | 15 | #include <linux/backing-dev.h> |
5bae7f73 | 16 | #include <linux/list_sort.h> |
21d34711 CH |
17 | #include <linux/slab.h> |
18 | #include <linux/types.h> | |
1673f1f0 CH |
19 | #include <linux/pr.h> |
20 | #include <linux/ptrace.h> | |
21 | #include <linux/nvme_ioctl.h> | |
22 | #include <linux/t10-pi.h> | |
c5552fde | 23 | #include <linux/pm_qos.h> |
1673f1f0 | 24 | #include <asm/unaligned.h> |
21d34711 CH |
25 | |
26 | #include "nvme.h" | |
038bd4cb | 27 | #include "fabrics.h" |
21d34711 | 28 | |
35fe0d12 HR |
29 | #define CREATE_TRACE_POINTS |
30 | #include "trace.h" | |
31 | ||
f3ca80fc CH |
32 | #define NVME_MINORS (1U << MINORBITS) |
33 | ||
8ae4e447 MO |
34 | unsigned int admin_timeout = 60; |
35 | module_param(admin_timeout, uint, 0644); | |
ba0ba7d3 | 36 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); |
576d55d6 | 37 | EXPORT_SYMBOL_GPL(admin_timeout); |
ba0ba7d3 | 38 | |
8ae4e447 MO |
39 | unsigned int nvme_io_timeout = 30; |
40 | module_param_named(io_timeout, nvme_io_timeout, uint, 0644); | |
ba0ba7d3 | 41 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); |
576d55d6 | 42 | EXPORT_SYMBOL_GPL(nvme_io_timeout); |
ba0ba7d3 | 43 | |
b3b1b0b0 | 44 | static unsigned char shutdown_timeout = 5; |
ba0ba7d3 ML |
45 | module_param(shutdown_timeout, byte, 0644); |
46 | MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); | |
47 | ||
44e44b29 CH |
48 | static u8 nvme_max_retries = 5; |
49 | module_param_named(max_retries, nvme_max_retries, byte, 0644); | |
f80ec966 | 50 | MODULE_PARM_DESC(max_retries, "max number of retries a command may have"); |
5bae7f73 | 51 | |
9947d6a0 | 52 | static unsigned long default_ps_max_latency_us = 100000; |
c5552fde AL |
53 | module_param(default_ps_max_latency_us, ulong, 0644); |
54 | MODULE_PARM_DESC(default_ps_max_latency_us, | |
55 | "max power saving latency for new devices; use PM QOS to change per device"); | |
56 | ||
c35e30b4 AL |
57 | static bool force_apst; |
58 | module_param(force_apst, bool, 0644); | |
59 | MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off"); | |
60 | ||
f5d11840 JA |
61 | static bool streams; |
62 | module_param(streams, bool, 0644); | |
63 | MODULE_PARM_DESC(streams, "turn on support for Streams write directives"); | |
64 | ||
b227c59b RS |
65 | /* |
66 | * nvme_wq - hosts nvme related works that are not reset or delete | |
67 | * nvme_reset_wq - hosts nvme reset works | |
68 | * nvme_delete_wq - hosts nvme delete works | |
69 | * | |
97b2512a NK |
70 | * nvme_wq will host works such as scan, aen handling, fw activation, |
71 | * keep-alive, periodic reconnects etc. nvme_reset_wq | |
b227c59b RS |
72 | * runs reset works which also flush works hosted on nvme_wq for |
73 | * serialization purposes. nvme_delete_wq host controller deletion | |
74 | * works which flush reset works for serialization. | |
75 | */ | |
9a6327d2 SG |
76 | struct workqueue_struct *nvme_wq; |
77 | EXPORT_SYMBOL_GPL(nvme_wq); | |
78 | ||
b227c59b RS |
79 | struct workqueue_struct *nvme_reset_wq; |
80 | EXPORT_SYMBOL_GPL(nvme_reset_wq); | |
81 | ||
82 | struct workqueue_struct *nvme_delete_wq; | |
83 | EXPORT_SYMBOL_GPL(nvme_delete_wq); | |
84 | ||
ab9e00cc CH |
85 | static LIST_HEAD(nvme_subsystems); |
86 | static DEFINE_MUTEX(nvme_subsystems_lock); | |
1673f1f0 | 87 | |
9843f685 | 88 | static DEFINE_IDA(nvme_instance_ida); |
a6a5149b | 89 | static dev_t nvme_chr_devt; |
f3ca80fc | 90 | static struct class *nvme_class; |
ab9e00cc | 91 | static struct class *nvme_subsys_class; |
f3ca80fc | 92 | |
84fef62d | 93 | static int nvme_revalidate_disk(struct gendisk *disk); |
12d9f070 | 94 | static void nvme_put_subsystem(struct nvme_subsystem *subsys); |
cf39a6bc SB |
95 | static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl, |
96 | unsigned nsid); | |
97 | ||
98 | static void nvme_set_queue_dying(struct nvme_ns *ns) | |
99 | { | |
100 | /* | |
101 | * Revalidating a dead namespace sets capacity to 0. This will end | |
102 | * buffered writers dirtying pages that can't be synced. | |
103 | */ | |
104 | if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags)) | |
105 | return; | |
cf39a6bc SB |
106 | blk_set_queue_dying(ns->queue); |
107 | /* Forcibly unquiesce queues to avoid blocking dispatch */ | |
108 | blk_mq_unquiesce_queue(ns->queue); | |
b224726d BS |
109 | /* |
110 | * Revalidate after unblocking dispatchers that may be holding bd_butex | |
111 | */ | |
112 | revalidate_disk(ns->disk); | |
cf39a6bc | 113 | } |
f3ca80fc | 114 | |
50e8d8ee CH |
115 | static void nvme_queue_scan(struct nvme_ctrl *ctrl) |
116 | { | |
117 | /* | |
118 | * Only new queue scan work when admin and IO queues are both alive | |
119 | */ | |
5d02a5c1 | 120 | if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset) |
50e8d8ee CH |
121 | queue_work(nvme_wq, &ctrl->scan_work); |
122 | } | |
123 | ||
4c75f877 KB |
124 | /* |
125 | * Use this function to proceed with scheduling reset_work for a controller | |
126 | * that had previously been set to the resetting state. This is intended for | |
127 | * code paths that can't be interrupted by other reset attempts. A hot removal | |
128 | * may prevent this from succeeding. | |
129 | */ | |
c1ac9a4b | 130 | int nvme_try_sched_reset(struct nvme_ctrl *ctrl) |
4c75f877 KB |
131 | { |
132 | if (ctrl->state != NVME_CTRL_RESETTING) | |
133 | return -EBUSY; | |
134 | if (!queue_work(nvme_reset_wq, &ctrl->reset_work)) | |
135 | return -EBUSY; | |
136 | return 0; | |
137 | } | |
c1ac9a4b | 138 | EXPORT_SYMBOL_GPL(nvme_try_sched_reset); |
4c75f877 | 139 | |
d86c4d8e CH |
140 | int nvme_reset_ctrl(struct nvme_ctrl *ctrl) |
141 | { | |
142 | if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) | |
143 | return -EBUSY; | |
b227c59b | 144 | if (!queue_work(nvme_reset_wq, &ctrl->reset_work)) |
d86c4d8e CH |
145 | return -EBUSY; |
146 | return 0; | |
147 | } | |
148 | EXPORT_SYMBOL_GPL(nvme_reset_ctrl); | |
149 | ||
79c48ccf | 150 | int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl) |
d86c4d8e CH |
151 | { |
152 | int ret; | |
153 | ||
154 | ret = nvme_reset_ctrl(ctrl); | |
8000d1fd | 155 | if (!ret) { |
d86c4d8e | 156 | flush_work(&ctrl->reset_work); |
5d02a5c1 | 157 | if (ctrl->state != NVME_CTRL_LIVE) |
8000d1fd NC |
158 | ret = -ENETRESET; |
159 | } | |
160 | ||
d86c4d8e CH |
161 | return ret; |
162 | } | |
79c48ccf | 163 | EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync); |
d86c4d8e | 164 | |
a686ed75 | 165 | static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl) |
c5017e85 | 166 | { |
77d0612d MG |
167 | dev_info(ctrl->device, |
168 | "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn); | |
169 | ||
4054637c | 170 | flush_work(&ctrl->reset_work); |
6cd53d14 CH |
171 | nvme_stop_ctrl(ctrl); |
172 | nvme_remove_namespaces(ctrl); | |
c5017e85 | 173 | ctrl->ops->delete_ctrl(ctrl); |
6cd53d14 | 174 | nvme_uninit_ctrl(ctrl); |
c5017e85 CH |
175 | } |
176 | ||
a686ed75 BVA |
177 | static void nvme_delete_ctrl_work(struct work_struct *work) |
178 | { | |
179 | struct nvme_ctrl *ctrl = | |
180 | container_of(work, struct nvme_ctrl, delete_work); | |
181 | ||
182 | nvme_do_delete_ctrl(ctrl); | |
183 | } | |
184 | ||
c5017e85 CH |
185 | int nvme_delete_ctrl(struct nvme_ctrl *ctrl) |
186 | { | |
187 | if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING)) | |
188 | return -EBUSY; | |
b227c59b | 189 | if (!queue_work(nvme_delete_wq, &ctrl->delete_work)) |
c5017e85 CH |
190 | return -EBUSY; |
191 | return 0; | |
192 | } | |
193 | EXPORT_SYMBOL_GPL(nvme_delete_ctrl); | |
194 | ||
6721c18a | 195 | static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl) |
c5017e85 | 196 | { |
c5017e85 | 197 | /* |
01fc08ff YY |
198 | * Keep a reference until nvme_do_delete_ctrl() complete, |
199 | * since ->delete_ctrl can free the controller. | |
c5017e85 CH |
200 | */ |
201 | nvme_get_ctrl(ctrl); | |
6721c18a | 202 | if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING)) |
b9c77583 | 203 | nvme_do_delete_ctrl(ctrl); |
c5017e85 | 204 | nvme_put_ctrl(ctrl); |
c5017e85 | 205 | } |
c5017e85 | 206 | |
715ea9e0 CH |
207 | static inline bool nvme_ns_has_pi(struct nvme_ns *ns) |
208 | { | |
209 | return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple); | |
210 | } | |
211 | ||
2f9c1736 | 212 | static blk_status_t nvme_error_status(u16 status) |
27fa9bc5 | 213 | { |
2f9c1736 | 214 | switch (status & 0x7ff) { |
27fa9bc5 | 215 | case NVME_SC_SUCCESS: |
2a842aca | 216 | return BLK_STS_OK; |
27fa9bc5 | 217 | case NVME_SC_CAP_EXCEEDED: |
2a842aca | 218 | return BLK_STS_NOSPC; |
e96fef2c | 219 | case NVME_SC_LBA_RANGE: |
35038bff KB |
220 | case NVME_SC_CMD_INTERRUPTED: |
221 | case NVME_SC_NS_NOT_READY: | |
e96fef2c KB |
222 | return BLK_STS_TARGET; |
223 | case NVME_SC_BAD_ATTRIBUTES: | |
e02ab023 | 224 | case NVME_SC_ONCS_NOT_SUPPORTED: |
e96fef2c KB |
225 | case NVME_SC_INVALID_OPCODE: |
226 | case NVME_SC_INVALID_FIELD: | |
227 | case NVME_SC_INVALID_NS: | |
2a842aca | 228 | return BLK_STS_NOTSUPP; |
e02ab023 JG |
229 | case NVME_SC_WRITE_FAULT: |
230 | case NVME_SC_READ_ERROR: | |
231 | case NVME_SC_UNWRITTEN_BLOCK: | |
a751da33 CH |
232 | case NVME_SC_ACCESS_DENIED: |
233 | case NVME_SC_READ_ONLY: | |
e96fef2c | 234 | case NVME_SC_COMPARE_FAILED: |
2a842aca | 235 | return BLK_STS_MEDIUM; |
a751da33 CH |
236 | case NVME_SC_GUARD_CHECK: |
237 | case NVME_SC_APPTAG_CHECK: | |
238 | case NVME_SC_REFTAG_CHECK: | |
239 | case NVME_SC_INVALID_PI: | |
240 | return BLK_STS_PROTECTION; | |
241 | case NVME_SC_RESERVATION_CONFLICT: | |
242 | return BLK_STS_NEXUS; | |
1c0d12c0 SG |
243 | case NVME_SC_HOST_PATH_ERROR: |
244 | return BLK_STS_TRANSPORT; | |
2a842aca CH |
245 | default: |
246 | return BLK_STS_IOERR; | |
27fa9bc5 CH |
247 | } |
248 | } | |
27fa9bc5 | 249 | |
f6324b1b | 250 | static inline bool nvme_req_needs_retry(struct request *req) |
77f02a7a | 251 | { |
f6324b1b CH |
252 | if (blk_noretry_request(req)) |
253 | return false; | |
27fa9bc5 | 254 | if (nvme_req(req)->status & NVME_SC_DNR) |
f6324b1b | 255 | return false; |
44e44b29 | 256 | if (nvme_req(req)->retries >= nvme_max_retries) |
f6324b1b CH |
257 | return false; |
258 | return true; | |
77f02a7a CH |
259 | } |
260 | ||
49cd84b6 KB |
261 | static void nvme_retry_req(struct request *req) |
262 | { | |
263 | struct nvme_ns *ns = req->q->queuedata; | |
264 | unsigned long delay = 0; | |
265 | u16 crd; | |
266 | ||
267 | /* The mask and shift result must be <= 3 */ | |
268 | crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11; | |
269 | if (ns && crd) | |
270 | delay = ns->ctrl->crdt[crd - 1] * 100; | |
271 | ||
272 | nvme_req(req)->retries++; | |
273 | blk_mq_requeue_request(req, false); | |
274 | blk_mq_delay_kick_requeue_list(req->q, delay); | |
275 | } | |
276 | ||
77f02a7a CH |
277 | void nvme_complete_rq(struct request *req) |
278 | { | |
2f9c1736 | 279 | blk_status_t status = nvme_error_status(nvme_req(req)->status); |
908e4564 | 280 | |
ca5554a6 JT |
281 | trace_nvme_complete_rq(req); |
282 | ||
16686f3a MG |
283 | nvme_cleanup_cmd(req); |
284 | ||
6e3ca03e SG |
285 | if (nvme_req(req)->ctrl->kas) |
286 | nvme_req(req)->ctrl->comp_seen = true; | |
287 | ||
908e4564 | 288 | if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) { |
764e9332 | 289 | if ((req->cmd_flags & REQ_NVME_MPATH) && nvme_failover_req(req)) |
32acab31 | 290 | return; |
32acab31 CH |
291 | |
292 | if (!blk_queue_dying(req->q)) { | |
49cd84b6 | 293 | nvme_retry_req(req); |
32acab31 CH |
294 | return; |
295 | } | |
77f02a7a | 296 | } |
35fe0d12 HR |
297 | |
298 | nvme_trace_bio_complete(req, status); | |
908e4564 | 299 | blk_mq_end_request(req, status); |
77f02a7a CH |
300 | } |
301 | EXPORT_SYMBOL_GPL(nvme_complete_rq); | |
302 | ||
7baa8572 | 303 | bool nvme_cancel_request(struct request *req, void *data, bool reserved) |
c55a2fd4 | 304 | { |
c55a2fd4 ML |
305 | dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device, |
306 | "Cancelling I/O %d", req->tag); | |
307 | ||
78ca4072 ML |
308 | /* don't abort one completed request */ |
309 | if (blk_mq_request_completed(req)) | |
310 | return true; | |
311 | ||
2dc3947b | 312 | nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD; |
a87ccce0 | 313 | blk_mq_complete_request(req); |
7baa8572 | 314 | return true; |
c55a2fd4 ML |
315 | } |
316 | EXPORT_SYMBOL_GPL(nvme_cancel_request); | |
317 | ||
bb8d261e CH |
318 | bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, |
319 | enum nvme_ctrl_state new_state) | |
320 | { | |
f6b6a28e | 321 | enum nvme_ctrl_state old_state; |
0a72bbba | 322 | unsigned long flags; |
bb8d261e CH |
323 | bool changed = false; |
324 | ||
0a72bbba | 325 | spin_lock_irqsave(&ctrl->lock, flags); |
f6b6a28e GKB |
326 | |
327 | old_state = ctrl->state; | |
bb8d261e CH |
328 | switch (new_state) { |
329 | case NVME_CTRL_LIVE: | |
330 | switch (old_state) { | |
7d2e8008 | 331 | case NVME_CTRL_NEW: |
bb8d261e | 332 | case NVME_CTRL_RESETTING: |
ad6a0a52 | 333 | case NVME_CTRL_CONNECTING: |
bb8d261e CH |
334 | changed = true; |
335 | /* FALLTHRU */ | |
336 | default: | |
337 | break; | |
338 | } | |
339 | break; | |
340 | case NVME_CTRL_RESETTING: | |
341 | switch (old_state) { | |
342 | case NVME_CTRL_NEW: | |
def61eca | 343 | case NVME_CTRL_LIVE: |
def61eca CH |
344 | changed = true; |
345 | /* FALLTHRU */ | |
346 | default: | |
347 | break; | |
348 | } | |
349 | break; | |
ad6a0a52 | 350 | case NVME_CTRL_CONNECTING: |
def61eca | 351 | switch (old_state) { |
b754a32c | 352 | case NVME_CTRL_NEW: |
3cec7f9d | 353 | case NVME_CTRL_RESETTING: |
bb8d261e CH |
354 | changed = true; |
355 | /* FALLTHRU */ | |
356 | default: | |
357 | break; | |
358 | } | |
359 | break; | |
360 | case NVME_CTRL_DELETING: | |
361 | switch (old_state) { | |
362 | case NVME_CTRL_LIVE: | |
363 | case NVME_CTRL_RESETTING: | |
ad6a0a52 | 364 | case NVME_CTRL_CONNECTING: |
bb8d261e CH |
365 | changed = true; |
366 | /* FALLTHRU */ | |
367 | default: | |
368 | break; | |
369 | } | |
370 | break; | |
0ff9d4e1 KB |
371 | case NVME_CTRL_DEAD: |
372 | switch (old_state) { | |
373 | case NVME_CTRL_DELETING: | |
374 | changed = true; | |
375 | /* FALLTHRU */ | |
376 | default: | |
377 | break; | |
378 | } | |
379 | break; | |
bb8d261e CH |
380 | default: |
381 | break; | |
382 | } | |
bb8d261e | 383 | |
c1ac9a4b | 384 | if (changed) { |
bb8d261e | 385 | ctrl->state = new_state; |
c1ac9a4b KB |
386 | wake_up_all(&ctrl->state_wq); |
387 | } | |
bb8d261e | 388 | |
0a72bbba | 389 | spin_unlock_irqrestore(&ctrl->lock, flags); |
32acab31 CH |
390 | if (changed && ctrl->state == NVME_CTRL_LIVE) |
391 | nvme_kick_requeue_lists(ctrl); | |
bb8d261e CH |
392 | return changed; |
393 | } | |
394 | EXPORT_SYMBOL_GPL(nvme_change_ctrl_state); | |
395 | ||
c1ac9a4b KB |
396 | /* |
397 | * Returns true for sink states that can't ever transition back to live. | |
398 | */ | |
399 | static bool nvme_state_terminal(struct nvme_ctrl *ctrl) | |
400 | { | |
401 | switch (ctrl->state) { | |
402 | case NVME_CTRL_NEW: | |
403 | case NVME_CTRL_LIVE: | |
404 | case NVME_CTRL_RESETTING: | |
405 | case NVME_CTRL_CONNECTING: | |
406 | return false; | |
407 | case NVME_CTRL_DELETING: | |
408 | case NVME_CTRL_DEAD: | |
409 | return true; | |
410 | default: | |
411 | WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state); | |
412 | return true; | |
413 | } | |
414 | } | |
415 | ||
416 | /* | |
417 | * Waits for the controller state to be resetting, or returns false if it is | |
418 | * not possible to ever transition to that state. | |
419 | */ | |
420 | bool nvme_wait_reset(struct nvme_ctrl *ctrl) | |
421 | { | |
422 | wait_event(ctrl->state_wq, | |
423 | nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) || | |
424 | nvme_state_terminal(ctrl)); | |
425 | return ctrl->state == NVME_CTRL_RESETTING; | |
426 | } | |
427 | EXPORT_SYMBOL_GPL(nvme_wait_reset); | |
428 | ||
ed754e5d CH |
429 | static void nvme_free_ns_head(struct kref *ref) |
430 | { | |
431 | struct nvme_ns_head *head = | |
432 | container_of(ref, struct nvme_ns_head, ref); | |
433 | ||
32acab31 | 434 | nvme_mpath_remove_disk(head); |
ed754e5d CH |
435 | ida_simple_remove(&head->subsys->ns_ida, head->instance); |
436 | list_del_init(&head->entry); | |
f5ad3991 | 437 | cleanup_srcu_struct(&head->srcu); |
12d9f070 | 438 | nvme_put_subsystem(head->subsys); |
ed754e5d CH |
439 | kfree(head); |
440 | } | |
441 | ||
442 | static void nvme_put_ns_head(struct nvme_ns_head *head) | |
443 | { | |
444 | kref_put(&head->ref, nvme_free_ns_head); | |
445 | } | |
446 | ||
1673f1f0 CH |
447 | static void nvme_free_ns(struct kref *kref) |
448 | { | |
449 | struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref); | |
450 | ||
b0b4e09c MB |
451 | if (ns->ndev) |
452 | nvme_nvm_unregister(ns); | |
1673f1f0 | 453 | |
1673f1f0 | 454 | put_disk(ns->disk); |
ed754e5d | 455 | nvme_put_ns_head(ns->head); |
075790eb | 456 | nvme_put_ctrl(ns->ctrl); |
1673f1f0 CH |
457 | kfree(ns); |
458 | } | |
459 | ||
5bae7f73 | 460 | static void nvme_put_ns(struct nvme_ns *ns) |
1673f1f0 CH |
461 | { |
462 | kref_put(&ns->kref, nvme_free_ns); | |
463 | } | |
464 | ||
bb06ec31 JS |
465 | static inline void nvme_clear_nvme_request(struct request *req) |
466 | { | |
467 | if (!(req->rq_flags & RQF_DONTPREP)) { | |
468 | nvme_req(req)->retries = 0; | |
469 | nvme_req(req)->flags = 0; | |
470 | req->rq_flags |= RQF_DONTPREP; | |
471 | } | |
472 | } | |
473 | ||
4160982e | 474 | struct request *nvme_alloc_request(struct request_queue *q, |
9a95e4ef | 475 | struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid) |
21d34711 | 476 | { |
aebf526b | 477 | unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; |
21d34711 | 478 | struct request *req; |
21d34711 | 479 | |
eb71f435 | 480 | if (qid == NVME_QID_ANY) { |
aebf526b | 481 | req = blk_mq_alloc_request(q, op, flags); |
eb71f435 | 482 | } else { |
aebf526b | 483 | req = blk_mq_alloc_request_hctx(q, op, flags, |
eb71f435 CH |
484 | qid ? qid - 1 : 0); |
485 | } | |
21d34711 | 486 | if (IS_ERR(req)) |
4160982e | 487 | return req; |
21d34711 | 488 | |
21d34711 | 489 | req->cmd_flags |= REQ_FAILFAST_DRIVER; |
bb06ec31 | 490 | nvme_clear_nvme_request(req); |
d49187e9 | 491 | nvme_req(req)->cmd = cmd; |
21d34711 | 492 | |
4160982e CH |
493 | return req; |
494 | } | |
576d55d6 | 495 | EXPORT_SYMBOL_GPL(nvme_alloc_request); |
4160982e | 496 | |
f5d11840 JA |
497 | static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable) |
498 | { | |
499 | struct nvme_command c; | |
500 | ||
501 | memset(&c, 0, sizeof(c)); | |
502 | ||
503 | c.directive.opcode = nvme_admin_directive_send; | |
62346eae | 504 | c.directive.nsid = cpu_to_le32(NVME_NSID_ALL); |
f5d11840 JA |
505 | c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE; |
506 | c.directive.dtype = NVME_DIR_IDENTIFY; | |
507 | c.directive.tdtype = NVME_DIR_STREAMS; | |
508 | c.directive.endir = enable ? NVME_DIR_ENDIR : 0; | |
509 | ||
510 | return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0); | |
511 | } | |
512 | ||
513 | static int nvme_disable_streams(struct nvme_ctrl *ctrl) | |
514 | { | |
515 | return nvme_toggle_streams(ctrl, false); | |
516 | } | |
517 | ||
518 | static int nvme_enable_streams(struct nvme_ctrl *ctrl) | |
519 | { | |
520 | return nvme_toggle_streams(ctrl, true); | |
521 | } | |
522 | ||
523 | static int nvme_get_stream_params(struct nvme_ctrl *ctrl, | |
524 | struct streams_directive_params *s, u32 nsid) | |
525 | { | |
526 | struct nvme_command c; | |
527 | ||
528 | memset(&c, 0, sizeof(c)); | |
529 | memset(s, 0, sizeof(*s)); | |
530 | ||
531 | c.directive.opcode = nvme_admin_directive_recv; | |
532 | c.directive.nsid = cpu_to_le32(nsid); | |
a082b426 | 533 | c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1); |
f5d11840 JA |
534 | c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM; |
535 | c.directive.dtype = NVME_DIR_STREAMS; | |
536 | ||
537 | return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s)); | |
538 | } | |
539 | ||
540 | static int nvme_configure_directives(struct nvme_ctrl *ctrl) | |
541 | { | |
542 | struct streams_directive_params s; | |
543 | int ret; | |
544 | ||
545 | if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES)) | |
546 | return 0; | |
547 | if (!streams) | |
548 | return 0; | |
549 | ||
550 | ret = nvme_enable_streams(ctrl); | |
551 | if (ret) | |
552 | return ret; | |
553 | ||
62346eae | 554 | ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL); |
f5d11840 JA |
555 | if (ret) |
556 | return ret; | |
557 | ||
558 | ctrl->nssa = le16_to_cpu(s.nssa); | |
559 | if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) { | |
560 | dev_info(ctrl->device, "too few streams (%u) available\n", | |
561 | ctrl->nssa); | |
562 | nvme_disable_streams(ctrl); | |
563 | return 0; | |
564 | } | |
565 | ||
566 | ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1); | |
567 | dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams); | |
568 | return 0; | |
569 | } | |
570 | ||
571 | /* | |
572 | * Check if 'req' has a write hint associated with it. If it does, assign | |
573 | * a valid namespace stream to the write. | |
574 | */ | |
575 | static void nvme_assign_write_stream(struct nvme_ctrl *ctrl, | |
576 | struct request *req, u16 *control, | |
577 | u32 *dsmgmt) | |
578 | { | |
579 | enum rw_hint streamid = req->write_hint; | |
580 | ||
581 | if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE) | |
582 | streamid = 0; | |
583 | else { | |
584 | streamid--; | |
585 | if (WARN_ON_ONCE(streamid > ctrl->nr_streams)) | |
586 | return; | |
587 | ||
588 | *control |= NVME_RW_DTYPE_STREAMS; | |
589 | *dsmgmt |= streamid << 16; | |
590 | } | |
591 | ||
592 | if (streamid < ARRAY_SIZE(req->q->write_hints)) | |
593 | req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9; | |
594 | } | |
595 | ||
8093f7ca ML |
596 | static inline void nvme_setup_flush(struct nvme_ns *ns, |
597 | struct nvme_command *cmnd) | |
598 | { | |
8093f7ca | 599 | cmnd->common.opcode = nvme_cmd_flush; |
ed754e5d | 600 | cmnd->common.nsid = cpu_to_le32(ns->head->ns_id); |
8093f7ca ML |
601 | } |
602 | ||
fc17b653 | 603 | static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req, |
8093f7ca ML |
604 | struct nvme_command *cmnd) |
605 | { | |
b35ba01e | 606 | unsigned short segments = blk_rq_nr_discard_segments(req), n = 0; |
8093f7ca | 607 | struct nvme_dsm_range *range; |
b35ba01e | 608 | struct bio *bio; |
8093f7ca | 609 | |
530436c4 EH |
610 | /* |
611 | * Some devices do not consider the DSM 'Number of Ranges' field when | |
612 | * determining how much data to DMA. Always allocate memory for maximum | |
613 | * number of segments to prevent device reading beyond end of buffer. | |
614 | */ | |
615 | static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES; | |
616 | ||
617 | range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN); | |
cb5b7262 JA |
618 | if (!range) { |
619 | /* | |
620 | * If we fail allocation our range, fallback to the controller | |
621 | * discard page. If that's also busy, it's safe to return | |
622 | * busy, as we know we can make progress once that's freed. | |
623 | */ | |
624 | if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy)) | |
625 | return BLK_STS_RESOURCE; | |
626 | ||
627 | range = page_address(ns->ctrl->discard_page); | |
628 | } | |
8093f7ca | 629 | |
b35ba01e | 630 | __rq_for_each_bio(bio, req) { |
314d48dd | 631 | u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector); |
b35ba01e CH |
632 | u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift; |
633 | ||
8cb6af7b KB |
634 | if (n < segments) { |
635 | range[n].cattr = cpu_to_le32(0); | |
636 | range[n].nlb = cpu_to_le32(nlb); | |
637 | range[n].slba = cpu_to_le64(slba); | |
638 | } | |
b35ba01e CH |
639 | n++; |
640 | } | |
641 | ||
642 | if (WARN_ON_ONCE(n != segments)) { | |
cb5b7262 JA |
643 | if (virt_to_page(range) == ns->ctrl->discard_page) |
644 | clear_bit_unlock(0, &ns->ctrl->discard_page_busy); | |
645 | else | |
646 | kfree(range); | |
fc17b653 | 647 | return BLK_STS_IOERR; |
b35ba01e | 648 | } |
8093f7ca | 649 | |
8093f7ca | 650 | cmnd->dsm.opcode = nvme_cmd_dsm; |
ed754e5d | 651 | cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id); |
f1dd03a8 | 652 | cmnd->dsm.nr = cpu_to_le32(segments - 1); |
8093f7ca ML |
653 | cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); |
654 | ||
f9d03f96 CH |
655 | req->special_vec.bv_page = virt_to_page(range); |
656 | req->special_vec.bv_offset = offset_in_page(range); | |
530436c4 | 657 | req->special_vec.bv_len = alloc_size; |
f9d03f96 | 658 | req->rq_flags |= RQF_SPECIAL_PAYLOAD; |
8093f7ca | 659 | |
fc17b653 | 660 | return BLK_STS_OK; |
8093f7ca | 661 | } |
8093f7ca | 662 | |
6e02318e CK |
663 | static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns, |
664 | struct request *req, struct nvme_command *cmnd) | |
665 | { | |
666 | if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) | |
667 | return nvme_setup_discard(ns, req, cmnd); | |
668 | ||
669 | cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes; | |
670 | cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id); | |
671 | cmnd->write_zeroes.slba = | |
314d48dd | 672 | cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req))); |
6e02318e CK |
673 | cmnd->write_zeroes.length = |
674 | cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); | |
675 | cmnd->write_zeroes.control = 0; | |
676 | return BLK_STS_OK; | |
677 | } | |
678 | ||
ebe6d874 CH |
679 | static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns, |
680 | struct request *req, struct nvme_command *cmnd) | |
8093f7ca | 681 | { |
f5d11840 | 682 | struct nvme_ctrl *ctrl = ns->ctrl; |
8093f7ca ML |
683 | u16 control = 0; |
684 | u32 dsmgmt = 0; | |
685 | ||
686 | if (req->cmd_flags & REQ_FUA) | |
687 | control |= NVME_RW_FUA; | |
688 | if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) | |
689 | control |= NVME_RW_LR; | |
690 | ||
691 | if (req->cmd_flags & REQ_RAHEAD) | |
692 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; | |
693 | ||
8093f7ca | 694 | cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); |
ed754e5d | 695 | cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id); |
314d48dd | 696 | cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req))); |
8093f7ca ML |
697 | cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); |
698 | ||
f5d11840 JA |
699 | if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams) |
700 | nvme_assign_write_stream(ctrl, req, &control, &dsmgmt); | |
701 | ||
8093f7ca | 702 | if (ns->ms) { |
715ea9e0 CH |
703 | /* |
704 | * If formated with metadata, the block layer always provides a | |
705 | * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else | |
706 | * we enable the PRACT bit for protection information or set the | |
707 | * namespace capacity to zero to prevent any I/O. | |
708 | */ | |
709 | if (!blk_integrity_rq(req)) { | |
710 | if (WARN_ON_ONCE(!nvme_ns_has_pi(ns))) | |
711 | return BLK_STS_NOTSUPP; | |
712 | control |= NVME_RW_PRINFO_PRACT; | |
713 | } | |
714 | ||
8093f7ca ML |
715 | switch (ns->pi_type) { |
716 | case NVME_NS_DPS_PI_TYPE3: | |
717 | control |= NVME_RW_PRINFO_PRCHK_GUARD; | |
718 | break; | |
719 | case NVME_NS_DPS_PI_TYPE1: | |
720 | case NVME_NS_DPS_PI_TYPE2: | |
721 | control |= NVME_RW_PRINFO_PRCHK_GUARD | | |
722 | NVME_RW_PRINFO_PRCHK_REF; | |
ddd0bc75 | 723 | cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req)); |
8093f7ca ML |
724 | break; |
725 | } | |
8093f7ca ML |
726 | } |
727 | ||
728 | cmnd->rw.control = cpu_to_le16(control); | |
729 | cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); | |
ebe6d874 | 730 | return 0; |
8093f7ca ML |
731 | } |
732 | ||
f7f1fc36 MG |
733 | void nvme_cleanup_cmd(struct request *req) |
734 | { | |
f7f1fc36 | 735 | if (req->rq_flags & RQF_SPECIAL_PAYLOAD) { |
cb5b7262 JA |
736 | struct nvme_ns *ns = req->rq_disk->private_data; |
737 | struct page *page = req->special_vec.bv_page; | |
738 | ||
739 | if (page == ns->ctrl->discard_page) | |
740 | clear_bit_unlock(0, &ns->ctrl->discard_page_busy); | |
741 | else | |
742 | kfree(page_address(page) + req->special_vec.bv_offset); | |
f7f1fc36 MG |
743 | } |
744 | } | |
745 | EXPORT_SYMBOL_GPL(nvme_cleanup_cmd); | |
746 | ||
fc17b653 | 747 | blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req, |
8093f7ca ML |
748 | struct nvme_command *cmd) |
749 | { | |
fc17b653 | 750 | blk_status_t ret = BLK_STS_OK; |
8093f7ca | 751 | |
bb06ec31 | 752 | nvme_clear_nvme_request(req); |
987f699a | 753 | |
11902035 | 754 | memset(cmd, 0, sizeof(*cmd)); |
aebf526b CH |
755 | switch (req_op(req)) { |
756 | case REQ_OP_DRV_IN: | |
757 | case REQ_OP_DRV_OUT: | |
d49187e9 | 758 | memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd)); |
aebf526b CH |
759 | break; |
760 | case REQ_OP_FLUSH: | |
8093f7ca | 761 | nvme_setup_flush(ns, cmd); |
aebf526b | 762 | break; |
e850fd16 | 763 | case REQ_OP_WRITE_ZEROES: |
6e02318e CK |
764 | ret = nvme_setup_write_zeroes(ns, req, cmd); |
765 | break; | |
aebf526b | 766 | case REQ_OP_DISCARD: |
8093f7ca | 767 | ret = nvme_setup_discard(ns, req, cmd); |
aebf526b CH |
768 | break; |
769 | case REQ_OP_READ: | |
770 | case REQ_OP_WRITE: | |
ebe6d874 | 771 | ret = nvme_setup_rw(ns, req, cmd); |
aebf526b CH |
772 | break; |
773 | default: | |
774 | WARN_ON_ONCE(1); | |
fc17b653 | 775 | return BLK_STS_IOERR; |
aebf526b | 776 | } |
8093f7ca | 777 | |
721b3917 | 778 | cmd->common.command_id = req->tag; |
5d87eb94 | 779 | trace_nvme_setup_cmd(req, cmd); |
8093f7ca ML |
780 | return ret; |
781 | } | |
782 | EXPORT_SYMBOL_GPL(nvme_setup_cmd); | |
783 | ||
6287b51c SG |
784 | static void nvme_end_sync_rq(struct request *rq, blk_status_t error) |
785 | { | |
786 | struct completion *waiting = rq->end_io_data; | |
787 | ||
788 | rq->end_io_data = NULL; | |
789 | complete(waiting); | |
790 | } | |
791 | ||
792 | static void nvme_execute_rq_polled(struct request_queue *q, | |
793 | struct gendisk *bd_disk, struct request *rq, int at_head) | |
794 | { | |
795 | DECLARE_COMPLETION_ONSTACK(wait); | |
796 | ||
797 | WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)); | |
798 | ||
799 | rq->cmd_flags |= REQ_HIPRI; | |
800 | rq->end_io_data = &wait; | |
801 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq); | |
802 | ||
803 | while (!completion_done(&wait)) { | |
804 | blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true); | |
805 | cond_resched(); | |
806 | } | |
807 | } | |
808 | ||
4160982e CH |
809 | /* |
810 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
811 | * if the result is positive, it's an NVM Express status code | |
812 | */ | |
813 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
d49187e9 | 814 | union nvme_result *result, void *buffer, unsigned bufflen, |
9a95e4ef | 815 | unsigned timeout, int qid, int at_head, |
6287b51c | 816 | blk_mq_req_flags_t flags, bool poll) |
4160982e CH |
817 | { |
818 | struct request *req; | |
819 | int ret; | |
820 | ||
eb71f435 | 821 | req = nvme_alloc_request(q, cmd, flags, qid); |
4160982e CH |
822 | if (IS_ERR(req)) |
823 | return PTR_ERR(req); | |
824 | ||
825 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
826 | ||
21d34711 CH |
827 | if (buffer && bufflen) { |
828 | ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL); | |
829 | if (ret) | |
830 | goto out; | |
4160982e CH |
831 | } |
832 | ||
6287b51c SG |
833 | if (poll) |
834 | nvme_execute_rq_polled(req->q, NULL, req, at_head); | |
835 | else | |
836 | blk_execute_rq(req->q, NULL, req, at_head); | |
d49187e9 CH |
837 | if (result) |
838 | *result = nvme_req(req)->result; | |
27fa9bc5 CH |
839 | if (nvme_req(req)->flags & NVME_REQ_CANCELLED) |
840 | ret = -EINTR; | |
841 | else | |
842 | ret = nvme_req(req)->status; | |
4160982e CH |
843 | out: |
844 | blk_mq_free_request(req); | |
845 | return ret; | |
846 | } | |
eb71f435 | 847 | EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd); |
4160982e CH |
848 | |
849 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
850 | void *buffer, unsigned bufflen) | |
851 | { | |
eb71f435 | 852 | return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0, |
6287b51c | 853 | NVME_QID_ANY, 0, 0, false); |
4160982e | 854 | } |
576d55d6 | 855 | EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd); |
4160982e | 856 | |
1cad6562 CH |
857 | static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf, |
858 | unsigned len, u32 seed, bool write) | |
859 | { | |
860 | struct bio_integrity_payload *bip; | |
861 | int ret = -ENOMEM; | |
862 | void *buf; | |
863 | ||
864 | buf = kmalloc(len, GFP_KERNEL); | |
865 | if (!buf) | |
866 | goto out; | |
867 | ||
868 | ret = -EFAULT; | |
869 | if (write && copy_from_user(buf, ubuf, len)) | |
870 | goto out_free_meta; | |
871 | ||
872 | bip = bio_integrity_alloc(bio, GFP_KERNEL, 1); | |
873 | if (IS_ERR(bip)) { | |
874 | ret = PTR_ERR(bip); | |
875 | goto out_free_meta; | |
876 | } | |
877 | ||
878 | bip->bip_iter.bi_size = len; | |
879 | bip->bip_iter.bi_sector = seed; | |
880 | ret = bio_integrity_add_page(bio, virt_to_page(buf), len, | |
881 | offset_in_page(buf)); | |
882 | if (ret == len) | |
883 | return buf; | |
884 | ret = -ENOMEM; | |
885 | out_free_meta: | |
886 | kfree(buf); | |
887 | out: | |
888 | return ERR_PTR(ret); | |
889 | } | |
890 | ||
63263d60 | 891 | static int nvme_submit_user_cmd(struct request_queue *q, |
485783ca KB |
892 | struct nvme_command *cmd, void __user *ubuffer, |
893 | unsigned bufflen, void __user *meta_buffer, unsigned meta_len, | |
65e68edc | 894 | u32 meta_seed, u64 *result, unsigned timeout) |
4160982e | 895 | { |
7a5abb4b | 896 | bool write = nvme_is_write(cmd); |
0b7f1f26 KB |
897 | struct nvme_ns *ns = q->queuedata; |
898 | struct gendisk *disk = ns ? ns->disk : NULL; | |
4160982e | 899 | struct request *req; |
0b7f1f26 KB |
900 | struct bio *bio = NULL; |
901 | void *meta = NULL; | |
4160982e CH |
902 | int ret; |
903 | ||
eb71f435 | 904 | req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY); |
4160982e CH |
905 | if (IS_ERR(req)) |
906 | return PTR_ERR(req); | |
907 | ||
908 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
bb06ec31 | 909 | nvme_req(req)->flags |= NVME_REQ_USERCMD; |
4160982e CH |
910 | |
911 | if (ubuffer && bufflen) { | |
21d34711 CH |
912 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, |
913 | GFP_KERNEL); | |
914 | if (ret) | |
915 | goto out; | |
916 | bio = req->bio; | |
74d46992 | 917 | bio->bi_disk = disk; |
1cad6562 CH |
918 | if (disk && meta_buffer && meta_len) { |
919 | meta = nvme_add_user_metadata(bio, meta_buffer, meta_len, | |
920 | meta_seed, write); | |
921 | if (IS_ERR(meta)) { | |
922 | ret = PTR_ERR(meta); | |
0b7f1f26 KB |
923 | goto out_unmap; |
924 | } | |
f31a2110 | 925 | req->cmd_flags |= REQ_INTEGRITY; |
0b7f1f26 KB |
926 | } |
927 | } | |
1cad6562 | 928 | |
0b7f1f26 | 929 | blk_execute_rq(req->q, disk, req, 0); |
27fa9bc5 CH |
930 | if (nvme_req(req)->flags & NVME_REQ_CANCELLED) |
931 | ret = -EINTR; | |
932 | else | |
933 | ret = nvme_req(req)->status; | |
21d34711 | 934 | if (result) |
65e68edc | 935 | *result = le64_to_cpu(nvme_req(req)->result.u64); |
0b7f1f26 KB |
936 | if (meta && !ret && !write) { |
937 | if (copy_to_user(meta_buffer, meta, meta_len)) | |
938 | ret = -EFAULT; | |
939 | } | |
0b7f1f26 KB |
940 | kfree(meta); |
941 | out_unmap: | |
74d46992 | 942 | if (bio) |
0b7f1f26 | 943 | blk_rq_unmap_user(bio); |
21d34711 CH |
944 | out: |
945 | blk_mq_free_request(req); | |
946 | return ret; | |
947 | } | |
948 | ||
2a842aca | 949 | static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status) |
038bd4cb SG |
950 | { |
951 | struct nvme_ctrl *ctrl = rq->end_io_data; | |
86880d64 JS |
952 | unsigned long flags; |
953 | bool startka = false; | |
038bd4cb SG |
954 | |
955 | blk_mq_free_request(rq); | |
956 | ||
2a842aca | 957 | if (status) { |
038bd4cb | 958 | dev_err(ctrl->device, |
2a842aca CH |
959 | "failed nvme_keep_alive_end_io error=%d\n", |
960 | status); | |
038bd4cb SG |
961 | return; |
962 | } | |
963 | ||
6e3ca03e | 964 | ctrl->comp_seen = false; |
86880d64 JS |
965 | spin_lock_irqsave(&ctrl->lock, flags); |
966 | if (ctrl->state == NVME_CTRL_LIVE || | |
967 | ctrl->state == NVME_CTRL_CONNECTING) | |
968 | startka = true; | |
969 | spin_unlock_irqrestore(&ctrl->lock, flags); | |
970 | if (startka) | |
97b2512a | 971 | queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ); |
038bd4cb SG |
972 | } |
973 | ||
974 | static int nvme_keep_alive(struct nvme_ctrl *ctrl) | |
975 | { | |
038bd4cb SG |
976 | struct request *rq; |
977 | ||
0a34e466 | 978 | rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED, |
038bd4cb SG |
979 | NVME_QID_ANY); |
980 | if (IS_ERR(rq)) | |
981 | return PTR_ERR(rq); | |
982 | ||
983 | rq->timeout = ctrl->kato * HZ; | |
984 | rq->end_io_data = ctrl; | |
985 | ||
986 | blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io); | |
987 | ||
988 | return 0; | |
989 | } | |
990 | ||
991 | static void nvme_keep_alive_work(struct work_struct *work) | |
992 | { | |
993 | struct nvme_ctrl *ctrl = container_of(to_delayed_work(work), | |
994 | struct nvme_ctrl, ka_work); | |
6e3ca03e SG |
995 | bool comp_seen = ctrl->comp_seen; |
996 | ||
997 | if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) { | |
998 | dev_dbg(ctrl->device, | |
999 | "reschedule traffic based keep-alive timer\n"); | |
1000 | ctrl->comp_seen = false; | |
97b2512a | 1001 | queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ); |
6e3ca03e SG |
1002 | return; |
1003 | } | |
038bd4cb SG |
1004 | |
1005 | if (nvme_keep_alive(ctrl)) { | |
1006 | /* allocation failure, reset the controller */ | |
1007 | dev_err(ctrl->device, "keep-alive failed\n"); | |
39bdc590 | 1008 | nvme_reset_ctrl(ctrl); |
038bd4cb SG |
1009 | return; |
1010 | } | |
1011 | } | |
1012 | ||
00b683db | 1013 | static void nvme_start_keep_alive(struct nvme_ctrl *ctrl) |
038bd4cb SG |
1014 | { |
1015 | if (unlikely(ctrl->kato == 0)) | |
1016 | return; | |
1017 | ||
97b2512a | 1018 | queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ); |
038bd4cb | 1019 | } |
038bd4cb SG |
1020 | |
1021 | void nvme_stop_keep_alive(struct nvme_ctrl *ctrl) | |
1022 | { | |
1023 | if (unlikely(ctrl->kato == 0)) | |
1024 | return; | |
1025 | ||
1026 | cancel_delayed_work_sync(&ctrl->ka_work); | |
1027 | } | |
1028 | EXPORT_SYMBOL_GPL(nvme_stop_keep_alive); | |
1029 | ||
3f7f25a9 | 1030 | static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id) |
21d34711 CH |
1031 | { |
1032 | struct nvme_command c = { }; | |
1033 | int error; | |
1034 | ||
1035 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
1036 | c.identify.opcode = nvme_admin_identify; | |
986994a2 | 1037 | c.identify.cns = NVME_ID_CNS_CTRL; |
21d34711 CH |
1038 | |
1039 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); | |
1040 | if (!*id) | |
1041 | return -ENOMEM; | |
1042 | ||
1043 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
1044 | sizeof(struct nvme_id_ctrl)); | |
1045 | if (error) | |
1046 | kfree(*id); | |
1047 | return error; | |
1048 | } | |
1049 | ||
ad95a613 CK |
1050 | static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids, |
1051 | struct nvme_ns_id_desc *cur) | |
1052 | { | |
1053 | const char *warn_str = "ctrl returned bogus length:"; | |
1054 | void *data = cur; | |
1055 | ||
1056 | switch (cur->nidt) { | |
1057 | case NVME_NIDT_EUI64: | |
1058 | if (cur->nidl != NVME_NIDT_EUI64_LEN) { | |
1059 | dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n", | |
1060 | warn_str, cur->nidl); | |
1061 | return -1; | |
1062 | } | |
1063 | memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN); | |
1064 | return NVME_NIDT_EUI64_LEN; | |
1065 | case NVME_NIDT_NGUID: | |
1066 | if (cur->nidl != NVME_NIDT_NGUID_LEN) { | |
1067 | dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n", | |
1068 | warn_str, cur->nidl); | |
1069 | return -1; | |
1070 | } | |
1071 | memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN); | |
1072 | return NVME_NIDT_NGUID_LEN; | |
1073 | case NVME_NIDT_UUID: | |
1074 | if (cur->nidl != NVME_NIDT_UUID_LEN) { | |
1075 | dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n", | |
1076 | warn_str, cur->nidl); | |
1077 | return -1; | |
1078 | } | |
1079 | uuid_copy(&ids->uuid, data + sizeof(*cur)); | |
1080 | return NVME_NIDT_UUID_LEN; | |
1081 | default: | |
1082 | /* Skip unknown types */ | |
1083 | return cur->nidl; | |
1084 | } | |
1085 | } | |
1086 | ||
cdbff4f2 | 1087 | static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid, |
002fab04 | 1088 | struct nvme_ns_ids *ids) |
3b22ba26 JT |
1089 | { |
1090 | struct nvme_command c = { }; | |
1091 | int status; | |
1092 | void *data; | |
1093 | int pos; | |
1094 | int len; | |
1095 | ||
1096 | c.identify.opcode = nvme_admin_identify; | |
1097 | c.identify.nsid = cpu_to_le32(nsid); | |
1098 | c.identify.cns = NVME_ID_CNS_NS_DESC_LIST; | |
1099 | ||
1100 | data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL); | |
1101 | if (!data) | |
1102 | return -ENOMEM; | |
1103 | ||
cdbff4f2 | 1104 | status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data, |
3b22ba26 | 1105 | NVME_IDENTIFY_DATA_SIZE); |
fb314eb0 CH |
1106 | if (status) { |
1107 | dev_warn(ctrl->device, | |
1108 | "Identify Descriptors failed (%d)\n", status); | |
1109 | /* | |
1110 | * Don't treat an error as fatal, as we potentially already | |
1111 | * have a NGUID or EUI-64. | |
1112 | */ | |
59c7c3ca | 1113 | if (status > 0 && !(status & NVME_SC_DNR)) |
fb314eb0 | 1114 | status = 0; |
3b22ba26 | 1115 | goto free_data; |
fb314eb0 | 1116 | } |
3b22ba26 JT |
1117 | |
1118 | for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) { | |
1119 | struct nvme_ns_id_desc *cur = data + pos; | |
1120 | ||
1121 | if (cur->nidl == 0) | |
1122 | break; | |
1123 | ||
ad95a613 CK |
1124 | len = nvme_process_ns_desc(ctrl, ids, cur); |
1125 | if (len < 0) | |
1126 | goto free_data; | |
3b22ba26 JT |
1127 | |
1128 | len += sizeof(*cur); | |
1129 | } | |
1130 | free_data: | |
1131 | kfree(data); | |
1132 | return status; | |
1133 | } | |
1134 | ||
540c801c KB |
1135 | static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list) |
1136 | { | |
1137 | struct nvme_command c = { }; | |
1138 | ||
1139 | c.identify.opcode = nvme_admin_identify; | |
986994a2 | 1140 | c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST; |
540c801c | 1141 | c.identify.nsid = cpu_to_le32(nsid); |
42595eb7 MI |
1142 | return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, |
1143 | NVME_IDENTIFY_DATA_SIZE); | |
540c801c KB |
1144 | } |
1145 | ||
331813f6 SG |
1146 | static int nvme_identify_ns(struct nvme_ctrl *ctrl, |
1147 | unsigned nsid, struct nvme_id_ns **id) | |
21d34711 CH |
1148 | { |
1149 | struct nvme_command c = { }; | |
1150 | int error; | |
1151 | ||
1152 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
778f067c MG |
1153 | c.identify.opcode = nvme_admin_identify; |
1154 | c.identify.nsid = cpu_to_le32(nsid); | |
986994a2 | 1155 | c.identify.cns = NVME_ID_CNS_NS; |
21d34711 | 1156 | |
331813f6 SG |
1157 | *id = kmalloc(sizeof(**id), GFP_KERNEL); |
1158 | if (!*id) | |
1159 | return -ENOMEM; | |
21d34711 | 1160 | |
331813f6 | 1161 | error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id)); |
cdbff4f2 | 1162 | if (error) { |
d0de579c | 1163 | dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error); |
331813f6 | 1164 | kfree(*id); |
cdbff4f2 CH |
1165 | } |
1166 | ||
331813f6 | 1167 | return error; |
21d34711 CH |
1168 | } |
1169 | ||
1a87ee65 KB |
1170 | static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid, |
1171 | unsigned int dword11, void *buffer, size_t buflen, u32 *result) | |
21d34711 | 1172 | { |
15755854 | 1173 | union nvme_result res = { 0 }; |
21d34711 | 1174 | struct nvme_command c; |
1cb3cce5 | 1175 | int ret; |
21d34711 CH |
1176 | |
1177 | memset(&c, 0, sizeof(c)); | |
1a87ee65 | 1178 | c.features.opcode = op; |
21d34711 CH |
1179 | c.features.fid = cpu_to_le32(fid); |
1180 | c.features.dword11 = cpu_to_le32(dword11); | |
1181 | ||
d49187e9 | 1182 | ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, |
6287b51c | 1183 | buffer, buflen, 0, NVME_QID_ANY, 0, 0, false); |
9b47f77a | 1184 | if (ret >= 0 && result) |
d49187e9 | 1185 | *result = le32_to_cpu(res.u32); |
1cb3cce5 | 1186 | return ret; |
21d34711 CH |
1187 | } |
1188 | ||
1a87ee65 KB |
1189 | int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid, |
1190 | unsigned int dword11, void *buffer, size_t buflen, | |
1191 | u32 *result) | |
1192 | { | |
1193 | return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer, | |
1194 | buflen, result); | |
1195 | } | |
1196 | EXPORT_SYMBOL_GPL(nvme_set_features); | |
1197 | ||
1198 | int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid, | |
1199 | unsigned int dword11, void *buffer, size_t buflen, | |
1200 | u32 *result) | |
1201 | { | |
1202 | return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer, | |
1203 | buflen, result); | |
1204 | } | |
1205 | EXPORT_SYMBOL_GPL(nvme_get_features); | |
1206 | ||
9a0be7ab CH |
1207 | int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count) |
1208 | { | |
1209 | u32 q_count = (*count - 1) | ((*count - 1) << 16); | |
1210 | u32 result; | |
1211 | int status, nr_io_queues; | |
1212 | ||
1a6fe74d | 1213 | status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0, |
9a0be7ab | 1214 | &result); |
f5fa90dc | 1215 | if (status < 0) |
9a0be7ab CH |
1216 | return status; |
1217 | ||
f5fa90dc CH |
1218 | /* |
1219 | * Degraded controllers might return an error when setting the queue | |
1220 | * count. We still want to be able to bring them online and offer | |
1221 | * access to the admin queue, as that might be only way to fix them up. | |
1222 | */ | |
1223 | if (status > 0) { | |
f0425db0 | 1224 | dev_err(ctrl->device, "Could not set queue count (%d)\n", status); |
f5fa90dc CH |
1225 | *count = 0; |
1226 | } else { | |
1227 | nr_io_queues = min(result & 0xffff, result >> 16) + 1; | |
1228 | *count = min(*count, nr_io_queues); | |
1229 | } | |
1230 | ||
9a0be7ab CH |
1231 | return 0; |
1232 | } | |
576d55d6 | 1233 | EXPORT_SYMBOL_GPL(nvme_set_queue_count); |
9a0be7ab | 1234 | |
c0561f82 | 1235 | #define NVME_AEN_SUPPORTED \ |
85f8a435 SG |
1236 | (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \ |
1237 | NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE) | |
c0561f82 HR |
1238 | |
1239 | static void nvme_enable_aen(struct nvme_ctrl *ctrl) | |
1240 | { | |
fa441b71 | 1241 | u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED; |
c0561f82 HR |
1242 | int status; |
1243 | ||
fa441b71 WZ |
1244 | if (!supported_aens) |
1245 | return; | |
1246 | ||
1247 | status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens, | |
1248 | NULL, 0, &result); | |
c0561f82 HR |
1249 | if (status) |
1250 | dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n", | |
fa441b71 | 1251 | supported_aens); |
93da4023 SG |
1252 | |
1253 | queue_work(nvme_wq, &ctrl->async_event_work); | |
c0561f82 HR |
1254 | } |
1255 | ||
c95b708d NB |
1256 | /* |
1257 | * Convert integer values from ioctl structures to user pointers, silently | |
1258 | * ignoring the upper bits in the compat case to match behaviour of 32-bit | |
1259 | * kernels. | |
1260 | */ | |
1261 | static void __user *nvme_to_user_ptr(uintptr_t ptrval) | |
1262 | { | |
1263 | if (in_compat_syscall()) | |
1264 | ptrval = (compat_uptr_t)ptrval; | |
1265 | return (void __user *)ptrval; | |
1266 | } | |
1267 | ||
1673f1f0 CH |
1268 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
1269 | { | |
1270 | struct nvme_user_io io; | |
1271 | struct nvme_command c; | |
1272 | unsigned length, meta_len; | |
1273 | void __user *metadata; | |
1274 | ||
1275 | if (copy_from_user(&io, uio, sizeof(io))) | |
1276 | return -EFAULT; | |
63088ec7 KB |
1277 | if (io.flags) |
1278 | return -EINVAL; | |
1673f1f0 CH |
1279 | |
1280 | switch (io.opcode) { | |
1281 | case nvme_cmd_write: | |
1282 | case nvme_cmd_read: | |
1283 | case nvme_cmd_compare: | |
1284 | break; | |
1285 | default: | |
1286 | return -EINVAL; | |
1287 | } | |
1288 | ||
1289 | length = (io.nblocks + 1) << ns->lba_shift; | |
1290 | meta_len = (io.nblocks + 1) * ns->ms; | |
c95b708d | 1291 | metadata = nvme_to_user_ptr(io.metadata); |
1673f1f0 CH |
1292 | |
1293 | if (ns->ext) { | |
1294 | length += meta_len; | |
1295 | meta_len = 0; | |
1296 | } else if (meta_len) { | |
1297 | if ((io.metadata & 3) || !io.metadata) | |
1298 | return -EINVAL; | |
1299 | } | |
1300 | ||
1301 | memset(&c, 0, sizeof(c)); | |
1302 | c.rw.opcode = io.opcode; | |
1303 | c.rw.flags = io.flags; | |
ed754e5d | 1304 | c.rw.nsid = cpu_to_le32(ns->head->ns_id); |
1673f1f0 CH |
1305 | c.rw.slba = cpu_to_le64(io.slba); |
1306 | c.rw.length = cpu_to_le16(io.nblocks); | |
1307 | c.rw.control = cpu_to_le16(io.control); | |
1308 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); | |
1309 | c.rw.reftag = cpu_to_le32(io.reftag); | |
1310 | c.rw.apptag = cpu_to_le16(io.apptag); | |
1311 | c.rw.appmask = cpu_to_le16(io.appmask); | |
1312 | ||
63263d60 | 1313 | return nvme_submit_user_cmd(ns->queue, &c, |
c95b708d | 1314 | nvme_to_user_ptr(io.addr), length, |
202359c0 | 1315 | metadata, meta_len, lower_32_bits(io.slba), NULL, 0); |
1673f1f0 CH |
1316 | } |
1317 | ||
84fef62d KB |
1318 | static u32 nvme_known_admin_effects(u8 opcode) |
1319 | { | |
1320 | switch (opcode) { | |
1321 | case nvme_admin_format_nvm: | |
1322 | return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC | | |
1323 | NVME_CMD_EFFECTS_CSE_MASK; | |
1324 | case nvme_admin_sanitize_nvm: | |
1325 | return NVME_CMD_EFFECTS_CSE_MASK; | |
1326 | default: | |
1327 | break; | |
1328 | } | |
1329 | return 0; | |
1330 | } | |
1331 | ||
1332 | static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, | |
1333 | u8 opcode) | |
1334 | { | |
1335 | u32 effects = 0; | |
1336 | ||
1337 | if (ns) { | |
1338 | if (ctrl->effects) | |
1339 | effects = le32_to_cpu(ctrl->effects->iocs[opcode]); | |
415df90b | 1340 | if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC)) |
84fef62d KB |
1341 | dev_warn(ctrl->device, |
1342 | "IO command:%02x has unhandled effects:%08x\n", | |
1343 | opcode, effects); | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | if (ctrl->effects) | |
62843c2e | 1348 | effects = le32_to_cpu(ctrl->effects->acs[opcode]); |
6fa0321a | 1349 | effects |= nvme_known_admin_effects(opcode); |
84fef62d KB |
1350 | |
1351 | /* | |
1352 | * For simplicity, IO to all namespaces is quiesced even if the command | |
1353 | * effects say only one namespace is affected. | |
1354 | */ | |
1355 | if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) { | |
e7ad43c3 | 1356 | mutex_lock(&ctrl->scan_lock); |
b9156dae SG |
1357 | mutex_lock(&ctrl->subsys->lock); |
1358 | nvme_mpath_start_freeze(ctrl->subsys); | |
1359 | nvme_mpath_wait_freeze(ctrl->subsys); | |
84fef62d KB |
1360 | nvme_start_freeze(ctrl); |
1361 | nvme_wait_freeze(ctrl); | |
1362 | } | |
1363 | return effects; | |
1364 | } | |
1365 | ||
1366 | static void nvme_update_formats(struct nvme_ctrl *ctrl) | |
1367 | { | |
cf39a6bc | 1368 | struct nvme_ns *ns; |
84fef62d | 1369 | |
cf39a6bc SB |
1370 | down_read(&ctrl->namespaces_rwsem); |
1371 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
1372 | if (ns->disk && nvme_revalidate_disk(ns->disk)) | |
1373 | nvme_set_queue_dying(ns); | |
1374 | up_read(&ctrl->namespaces_rwsem); | |
84fef62d KB |
1375 | } |
1376 | ||
1377 | static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects) | |
1378 | { | |
1379 | /* | |
1380 | * Revalidate LBA changes prior to unfreezing. This is necessary to | |
1381 | * prevent memory corruption if a logical block size was changed by | |
1382 | * this command. | |
1383 | */ | |
1384 | if (effects & NVME_CMD_EFFECTS_LBCC) | |
1385 | nvme_update_formats(ctrl); | |
e7ad43c3 | 1386 | if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) { |
84fef62d | 1387 | nvme_unfreeze(ctrl); |
b9156dae SG |
1388 | nvme_mpath_unfreeze(ctrl->subsys); |
1389 | mutex_unlock(&ctrl->subsys->lock); | |
6abff1b9 | 1390 | nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL); |
e7ad43c3 KB |
1391 | mutex_unlock(&ctrl->scan_lock); |
1392 | } | |
84fef62d KB |
1393 | if (effects & NVME_CMD_EFFECTS_CCC) |
1394 | nvme_init_identify(ctrl); | |
1395 | if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) | |
1396 | nvme_queue_scan(ctrl); | |
1397 | } | |
1398 | ||
f3ca80fc | 1399 | static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns, |
1673f1f0 CH |
1400 | struct nvme_passthru_cmd __user *ucmd) |
1401 | { | |
1402 | struct nvme_passthru_cmd cmd; | |
1403 | struct nvme_command c; | |
1404 | unsigned timeout = 0; | |
84fef62d | 1405 | u32 effects; |
65e68edc MR |
1406 | u64 result; |
1407 | int status; | |
1408 | ||
1409 | if (!capable(CAP_SYS_ADMIN)) | |
1410 | return -EACCES; | |
1411 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
1412 | return -EFAULT; | |
1413 | if (cmd.flags) | |
1414 | return -EINVAL; | |
1415 | ||
1416 | memset(&c, 0, sizeof(c)); | |
1417 | c.common.opcode = cmd.opcode; | |
1418 | c.common.flags = cmd.flags; | |
1419 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1420 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1421 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1422 | c.common.cdw10 = cpu_to_le32(cmd.cdw10); | |
1423 | c.common.cdw11 = cpu_to_le32(cmd.cdw11); | |
1424 | c.common.cdw12 = cpu_to_le32(cmd.cdw12); | |
1425 | c.common.cdw13 = cpu_to_le32(cmd.cdw13); | |
1426 | c.common.cdw14 = cpu_to_le32(cmd.cdw14); | |
1427 | c.common.cdw15 = cpu_to_le32(cmd.cdw15); | |
1428 | ||
1429 | if (cmd.timeout_ms) | |
1430 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
1431 | ||
1432 | effects = nvme_passthru_start(ctrl, ns, cmd.opcode); | |
1433 | status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c, | |
c95b708d NB |
1434 | nvme_to_user_ptr(cmd.addr), cmd.data_len, |
1435 | nvme_to_user_ptr(cmd.metadata), cmd.metadata_len, | |
1436 | 0, &result, timeout); | |
65e68edc MR |
1437 | nvme_passthru_end(ctrl, effects); |
1438 | ||
1439 | if (status >= 0) { | |
1440 | if (put_user(result, &ucmd->result)) | |
1441 | return -EFAULT; | |
1442 | } | |
1443 | ||
1444 | return status; | |
1445 | } | |
1446 | ||
1447 | static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns, | |
1448 | struct nvme_passthru_cmd64 __user *ucmd) | |
1449 | { | |
1450 | struct nvme_passthru_cmd64 cmd; | |
1451 | struct nvme_command c; | |
1452 | unsigned timeout = 0; | |
1453 | u32 effects; | |
1673f1f0 CH |
1454 | int status; |
1455 | ||
1456 | if (!capable(CAP_SYS_ADMIN)) | |
1457 | return -EACCES; | |
1458 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
1459 | return -EFAULT; | |
63088ec7 KB |
1460 | if (cmd.flags) |
1461 | return -EINVAL; | |
1673f1f0 CH |
1462 | |
1463 | memset(&c, 0, sizeof(c)); | |
1464 | c.common.opcode = cmd.opcode; | |
1465 | c.common.flags = cmd.flags; | |
1466 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1467 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1468 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
b7c8f366 CK |
1469 | c.common.cdw10 = cpu_to_le32(cmd.cdw10); |
1470 | c.common.cdw11 = cpu_to_le32(cmd.cdw11); | |
1471 | c.common.cdw12 = cpu_to_le32(cmd.cdw12); | |
1472 | c.common.cdw13 = cpu_to_le32(cmd.cdw13); | |
1473 | c.common.cdw14 = cpu_to_le32(cmd.cdw14); | |
1474 | c.common.cdw15 = cpu_to_le32(cmd.cdw15); | |
1673f1f0 CH |
1475 | |
1476 | if (cmd.timeout_ms) | |
1477 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
1478 | ||
84fef62d | 1479 | effects = nvme_passthru_start(ctrl, ns, cmd.opcode); |
1673f1f0 | 1480 | status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c, |
c95b708d NB |
1481 | nvme_to_user_ptr(cmd.addr), cmd.data_len, |
1482 | nvme_to_user_ptr(cmd.metadata), cmd.metadata_len, | |
63263d60 | 1483 | 0, &cmd.result, timeout); |
84fef62d KB |
1484 | nvme_passthru_end(ctrl, effects); |
1485 | ||
1673f1f0 CH |
1486 | if (status >= 0) { |
1487 | if (put_user(cmd.result, &ucmd->result)) | |
1488 | return -EFAULT; | |
1489 | } | |
1490 | ||
1491 | return status; | |
1492 | } | |
1493 | ||
32acab31 CH |
1494 | /* |
1495 | * Issue ioctl requests on the first available path. Note that unlike normal | |
1496 | * block layer requests we will not retry failed request on another controller. | |
1497 | */ | |
1498 | static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk, | |
1499 | struct nvme_ns_head **head, int *srcu_idx) | |
1673f1f0 | 1500 | { |
32acab31 CH |
1501 | #ifdef CONFIG_NVME_MULTIPATH |
1502 | if (disk->fops == &nvme_ns_head_ops) { | |
100c815c CH |
1503 | struct nvme_ns *ns; |
1504 | ||
32acab31 CH |
1505 | *head = disk->private_data; |
1506 | *srcu_idx = srcu_read_lock(&(*head)->srcu); | |
100c815c CH |
1507 | ns = nvme_find_path(*head); |
1508 | if (!ns) | |
1509 | srcu_read_unlock(&(*head)->srcu, *srcu_idx); | |
1510 | return ns; | |
32acab31 CH |
1511 | } |
1512 | #endif | |
1513 | *head = NULL; | |
1514 | *srcu_idx = -1; | |
1515 | return disk->private_data; | |
1516 | } | |
1673f1f0 | 1517 | |
32acab31 CH |
1518 | static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx) |
1519 | { | |
1520 | if (head) | |
1521 | srcu_read_unlock(&head->srcu, idx); | |
1522 | } | |
1673f1f0 | 1523 | |
65e68edc MR |
1524 | static bool is_ctrl_ioctl(unsigned int cmd) |
1525 | { | |
1526 | if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD) | |
1527 | return true; | |
1528 | if (is_sed_ioctl(cmd)) | |
1529 | return true; | |
1530 | return false; | |
1531 | } | |
1532 | ||
1533 | static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd, | |
1534 | void __user *argp, | |
1535 | struct nvme_ns_head *head, | |
1536 | int srcu_idx) | |
1537 | { | |
1538 | struct nvme_ctrl *ctrl = ns->ctrl; | |
1539 | int ret; | |
1540 | ||
1541 | nvme_get_ctrl(ns->ctrl); | |
1542 | nvme_put_ns_from_disk(head, srcu_idx); | |
1543 | ||
1544 | switch (cmd) { | |
1545 | case NVME_IOCTL_ADMIN_CMD: | |
1546 | ret = nvme_user_cmd(ctrl, NULL, argp); | |
1547 | break; | |
1548 | case NVME_IOCTL_ADMIN64_CMD: | |
1549 | ret = nvme_user_cmd64(ctrl, NULL, argp); | |
1550 | break; | |
1551 | default: | |
1552 | ret = sed_ioctl(ctrl->opal_dev, cmd, argp); | |
1553 | break; | |
1554 | } | |
1555 | nvme_put_ctrl(ctrl); | |
1556 | return ret; | |
1557 | } | |
1558 | ||
32acab31 CH |
1559 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, |
1560 | unsigned int cmd, unsigned long arg) | |
1673f1f0 | 1561 | { |
32acab31 | 1562 | struct nvme_ns_head *head = NULL; |
90ec611a | 1563 | void __user *argp = (void __user *)arg; |
32acab31 CH |
1564 | struct nvme_ns *ns; |
1565 | int srcu_idx, ret; | |
1566 | ||
1567 | ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx); | |
1568 | if (unlikely(!ns)) | |
100c815c CH |
1569 | return -EWOULDBLOCK; |
1570 | ||
5fb4aac7 CH |
1571 | /* |
1572 | * Handle ioctls that apply to the controller instead of the namespace | |
1573 | * seperately and drop the ns SRCU reference early. This avoids a | |
1574 | * deadlock when deleting namespaces using the passthrough interface. | |
1575 | */ | |
65e68edc MR |
1576 | if (is_ctrl_ioctl(cmd)) |
1577 | return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx); | |
5fb4aac7 | 1578 | |
90ec611a CH |
1579 | switch (cmd) { |
1580 | case NVME_IOCTL_ID: | |
1581 | force_successful_syscall_return(); | |
1582 | ret = ns->head->ns_id; | |
1583 | break; | |
90ec611a CH |
1584 | case NVME_IOCTL_IO_CMD: |
1585 | ret = nvme_user_cmd(ns->ctrl, ns, argp); | |
1586 | break; | |
1587 | case NVME_IOCTL_SUBMIT_IO: | |
1588 | ret = nvme_submit_io(ns, argp); | |
1589 | break; | |
65e68edc MR |
1590 | case NVME_IOCTL_IO64_CMD: |
1591 | ret = nvme_user_cmd64(ns->ctrl, ns, argp); | |
1592 | break; | |
90ec611a CH |
1593 | default: |
1594 | if (ns->ndev) | |
1595 | ret = nvme_nvm_ioctl(ns, cmd, arg); | |
90ec611a CH |
1596 | else |
1597 | ret = -ENOTTY; | |
1598 | } | |
1599 | ||
32acab31 CH |
1600 | nvme_put_ns_from_disk(head, srcu_idx); |
1601 | return ret; | |
1673f1f0 | 1602 | } |
1673f1f0 | 1603 | |
c225b610 | 1604 | #ifdef CONFIG_COMPAT |
1605 | struct nvme_user_io32 { | |
1606 | __u8 opcode; | |
1607 | __u8 flags; | |
1608 | __u16 control; | |
1609 | __u16 nblocks; | |
1610 | __u16 rsvd; | |
1611 | __u64 metadata; | |
1612 | __u64 addr; | |
1613 | __u64 slba; | |
1614 | __u32 dsmgmt; | |
1615 | __u32 reftag; | |
1616 | __u16 apptag; | |
1617 | __u16 appmask; | |
1618 | } __attribute__((__packed__)); | |
1619 | ||
1620 | #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32) | |
1621 | ||
1622 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1623 | unsigned int cmd, unsigned long arg) | |
1624 | { | |
1625 | /* | |
1626 | * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO | |
1627 | * between 32 bit programs and 64 bit kernel. | |
1628 | * The cause is that the results of sizeof(struct nvme_user_io), | |
1629 | * which is used to define NVME_IOCTL_SUBMIT_IO, | |
1630 | * are not same between 32 bit compiler and 64 bit compiler. | |
1631 | * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling | |
1632 | * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs. | |
1633 | * Other IOCTL numbers are same between 32 bit and 64 bit. | |
1634 | * So there is nothing to do regarding to other IOCTL numbers. | |
1635 | */ | |
1636 | if (cmd == NVME_IOCTL_SUBMIT_IO32) | |
1637 | return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg); | |
1638 | ||
1639 | return nvme_ioctl(bdev, mode, cmd, arg); | |
1640 | } | |
1641 | #else | |
1642 | #define nvme_compat_ioctl NULL | |
1643 | #endif /* CONFIG_COMPAT */ | |
1644 | ||
1673f1f0 CH |
1645 | static int nvme_open(struct block_device *bdev, fmode_t mode) |
1646 | { | |
c6424a90 CH |
1647 | struct nvme_ns *ns = bdev->bd_disk->private_data; |
1648 | ||
32acab31 CH |
1649 | #ifdef CONFIG_NVME_MULTIPATH |
1650 | /* should never be called due to GENHD_FL_HIDDEN */ | |
1651 | if (WARN_ON_ONCE(ns->head->disk)) | |
85088c4a | 1652 | goto fail; |
32acab31 | 1653 | #endif |
c6424a90 | 1654 | if (!kref_get_unless_zero(&ns->kref)) |
85088c4a NC |
1655 | goto fail; |
1656 | if (!try_module_get(ns->ctrl->ops->module)) | |
1657 | goto fail_put_ns; | |
1658 | ||
c6424a90 | 1659 | return 0; |
85088c4a NC |
1660 | |
1661 | fail_put_ns: | |
1662 | nvme_put_ns(ns); | |
1663 | fail: | |
1664 | return -ENXIO; | |
1673f1f0 CH |
1665 | } |
1666 | ||
1667 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
1668 | { | |
85088c4a NC |
1669 | struct nvme_ns *ns = disk->private_data; |
1670 | ||
1671 | module_put(ns->ctrl->ops->module); | |
1672 | nvme_put_ns(ns); | |
1673f1f0 CH |
1673 | } |
1674 | ||
1675 | static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
1676 | { | |
1677 | /* some standard values */ | |
1678 | geo->heads = 1 << 6; | |
1679 | geo->sectors = 1 << 5; | |
1680 | geo->cylinders = get_capacity(bdev->bd_disk) >> 11; | |
1681 | return 0; | |
1682 | } | |
1683 | ||
1684 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
39b7baa4 | 1685 | static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) |
1673f1f0 CH |
1686 | { |
1687 | struct blk_integrity integrity; | |
1688 | ||
fa9a89fc | 1689 | memset(&integrity, 0, sizeof(integrity)); |
39b7baa4 | 1690 | switch (pi_type) { |
1673f1f0 CH |
1691 | case NVME_NS_DPS_PI_TYPE3: |
1692 | integrity.profile = &t10_pi_type3_crc; | |
ba36c21b NB |
1693 | integrity.tag_size = sizeof(u16) + sizeof(u32); |
1694 | integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; | |
1673f1f0 CH |
1695 | break; |
1696 | case NVME_NS_DPS_PI_TYPE1: | |
1697 | case NVME_NS_DPS_PI_TYPE2: | |
1698 | integrity.profile = &t10_pi_type1_crc; | |
ba36c21b NB |
1699 | integrity.tag_size = sizeof(u16); |
1700 | integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; | |
1673f1f0 CH |
1701 | break; |
1702 | default: | |
1703 | integrity.profile = NULL; | |
1704 | break; | |
1705 | } | |
39b7baa4 CH |
1706 | integrity.tuple_size = ms; |
1707 | blk_integrity_register(disk, &integrity); | |
1708 | blk_queue_max_integrity_segments(disk->queue, 1); | |
1673f1f0 CH |
1709 | } |
1710 | #else | |
39b7baa4 | 1711 | static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) |
1673f1f0 CH |
1712 | { |
1713 | } | |
1714 | #endif /* CONFIG_BLK_DEV_INTEGRITY */ | |
1715 | ||
6b8190d6 SB |
1716 | static void nvme_set_chunk_size(struct nvme_ns *ns) |
1717 | { | |
e08f2ae8 | 1718 | u32 chunk_size = nvme_lba_to_sect(ns, ns->noiob); |
6b8190d6 SB |
1719 | blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size)); |
1720 | } | |
1721 | ||
26318571 | 1722 | static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns) |
1673f1f0 | 1723 | { |
3831761e | 1724 | struct nvme_ctrl *ctrl = ns->ctrl; |
26318571 | 1725 | struct request_queue *queue = disk->queue; |
30e5e929 CH |
1726 | u32 size = queue_logical_block_size(queue); |
1727 | ||
3831761e JA |
1728 | if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) { |
1729 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue); | |
1730 | return; | |
1731 | } | |
1732 | ||
1733 | if (ctrl->nr_streams && ns->sws && ns->sgs) | |
1734 | size *= ns->sws * ns->sgs; | |
08095e70 | 1735 | |
b35ba01e CH |
1736 | BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) < |
1737 | NVME_DSM_MAX_RANGES); | |
1738 | ||
b224f613 | 1739 | queue->limits.discard_alignment = 0; |
30e5e929 | 1740 | queue->limits.discard_granularity = size; |
f5d11840 | 1741 | |
3831761e JA |
1742 | /* If discard is already enabled, don't reset queue limits */ |
1743 | if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue)) | |
1744 | return; | |
1745 | ||
30e5e929 CH |
1746 | blk_queue_max_discard_sectors(queue, UINT_MAX); |
1747 | blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES); | |
e850fd16 CH |
1748 | |
1749 | if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) | |
30e5e929 | 1750 | blk_queue_max_write_zeroes_sectors(queue, UINT_MAX); |
1673f1f0 CH |
1751 | } |
1752 | ||
9f0916ab | 1753 | static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns) |
6e02318e | 1754 | { |
e08f2ae8 | 1755 | u64 max_blocks; |
6e02318e | 1756 | |
7b210e4e CH |
1757 | if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) || |
1758 | (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES)) | |
6e02318e CK |
1759 | return; |
1760 | /* | |
1761 | * Even though NVMe spec explicitly states that MDTS is not | |
1762 | * applicable to the write-zeroes:- "The restriction does not apply to | |
1763 | * commands that do not transfer data between the host and the | |
1764 | * controller (e.g., Write Uncorrectable ro Write Zeroes command).". | |
1765 | * In order to be more cautious use controller's max_hw_sectors value | |
1766 | * to configure the maximum sectors for the write-zeroes which is | |
1767 | * configured based on the controller's MDTS field in the | |
1768 | * nvme_init_identify() if available. | |
1769 | */ | |
1770 | if (ns->ctrl->max_hw_sectors == UINT_MAX) | |
e08f2ae8 | 1771 | max_blocks = (u64)USHRT_MAX + 1; |
6e02318e | 1772 | else |
e08f2ae8 | 1773 | max_blocks = ns->ctrl->max_hw_sectors + 1; |
6e02318e | 1774 | |
e08f2ae8 DLM |
1775 | blk_queue_max_write_zeroes_sectors(disk->queue, |
1776 | nvme_lba_to_sect(ns, max_blocks)); | |
6e02318e CK |
1777 | } |
1778 | ||
538af88e | 1779 | static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid, |
002fab04 | 1780 | struct nvme_id_ns *id, struct nvme_ns_ids *ids) |
1673f1f0 | 1781 | { |
002fab04 CH |
1782 | memset(ids, 0, sizeof(*ids)); |
1783 | ||
cdbff4f2 | 1784 | if (ctrl->vs >= NVME_VS(1, 1, 0)) |
002fab04 | 1785 | memcpy(ids->eui64, id->eui64, sizeof(id->eui64)); |
cdbff4f2 | 1786 | if (ctrl->vs >= NVME_VS(1, 2, 0)) |
002fab04 | 1787 | memcpy(ids->nguid, id->nguid, sizeof(id->nguid)); |
fb314eb0 CH |
1788 | if (ctrl->vs >= NVME_VS(1, 3, 0)) |
1789 | return nvme_identify_ns_descs(ctrl, nsid, ids); | |
1790 | return 0; | |
ac81bfa9 MB |
1791 | } |
1792 | ||
ed754e5d CH |
1793 | static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids) |
1794 | { | |
1795 | return !uuid_is_null(&ids->uuid) || | |
1796 | memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) || | |
1797 | memchr_inv(ids->eui64, 0, sizeof(ids->eui64)); | |
1798 | } | |
1799 | ||
002fab04 CH |
1800 | static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b) |
1801 | { | |
1802 | return uuid_equal(&a->uuid, &b->uuid) && | |
1803 | memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 && | |
1804 | memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0; | |
1805 | } | |
1806 | ||
24b0b58c CH |
1807 | static void nvme_update_disk_info(struct gendisk *disk, |
1808 | struct nvme_ns *ns, struct nvme_id_ns *id) | |
1809 | { | |
e08f2ae8 | 1810 | sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze)); |
cee160fd | 1811 | unsigned short bs = 1 << ns->lba_shift; |
81adb863 | 1812 | u32 atomic_bs, phys_bs, io_opt; |
24b0b58c | 1813 | |
01fa0174 SG |
1814 | if (ns->lba_shift > PAGE_SHIFT) { |
1815 | /* unsupported block size, set capacity to 0 later */ | |
1816 | bs = (1 << 9); | |
1817 | } | |
24b0b58c CH |
1818 | blk_mq_freeze_queue(disk->queue); |
1819 | blk_integrity_unregister(disk); | |
1820 | ||
81adb863 BVA |
1821 | if (id->nabo == 0) { |
1822 | /* | |
1823 | * Bit 1 indicates whether NAWUPF is defined for this namespace | |
1824 | * and whether it should be used instead of AWUPF. If NAWUPF == | |
1825 | * 0 then AWUPF must be used instead. | |
1826 | */ | |
1827 | if (id->nsfeat & (1 << 1) && id->nawupf) | |
1828 | atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs; | |
1829 | else | |
1830 | atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs; | |
1831 | } else { | |
1832 | atomic_bs = bs; | |
1833 | } | |
1834 | phys_bs = bs; | |
1835 | io_opt = bs; | |
1836 | if (id->nsfeat & (1 << 4)) { | |
1837 | /* NPWG = Namespace Preferred Write Granularity */ | |
1838 | phys_bs *= 1 + le16_to_cpu(id->npwg); | |
1839 | /* NOWS = Namespace Optimal Write Size */ | |
1840 | io_opt *= 1 + le16_to_cpu(id->nows); | |
1841 | } | |
1842 | ||
cee160fd | 1843 | blk_queue_logical_block_size(disk->queue, bs); |
81adb863 BVA |
1844 | /* |
1845 | * Linux filesystems assume writing a single physical block is | |
1846 | * an atomic operation. Hence limit the physical block size to the | |
1847 | * value of the Atomic Write Unit Power Fail parameter. | |
1848 | */ | |
1849 | blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs)); | |
1850 | blk_queue_io_min(disk->queue, phys_bs); | |
1851 | blk_queue_io_opt(disk->queue, io_opt); | |
cee160fd | 1852 | |
24b0b58c CH |
1853 | if (ns->ms && !ns->ext && |
1854 | (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)) | |
1855 | nvme_init_integrity(disk, ns->ms, ns->pi_type); | |
01fa0174 SG |
1856 | if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) || |
1857 | ns->lba_shift > PAGE_SHIFT) | |
24b0b58c | 1858 | capacity = 0; |
24b0b58c | 1859 | |
cb224c3a | 1860 | set_capacity_revalidate_and_notify(disk, capacity, false); |
b1aafb35 | 1861 | |
26318571 | 1862 | nvme_config_discard(disk, ns); |
9f0916ab | 1863 | nvme_config_write_zeroes(disk, ns); |
1293477f CK |
1864 | |
1865 | if (id->nsattr & (1 << 0)) | |
1866 | set_disk_ro(disk, true); | |
1867 | else | |
1868 | set_disk_ro(disk, false); | |
1869 | ||
24b0b58c CH |
1870 | blk_mq_unfreeze_queue(disk->queue); |
1871 | } | |
1872 | ||
ac81bfa9 MB |
1873 | static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id) |
1874 | { | |
1875 | struct nvme_ns *ns = disk->private_data; | |
1673f1f0 CH |
1876 | |
1877 | /* | |
1878 | * If identify namespace failed, use default 512 byte block size so | |
1879 | * block layer can use before failing read/write for 0 capacity. | |
1880 | */ | |
c81bfba9 | 1881 | ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds; |
1673f1f0 CH |
1882 | if (ns->lba_shift == 0) |
1883 | ns->lba_shift = 9; | |
6b8190d6 | 1884 | ns->noiob = le16_to_cpu(id->noiob); |
b5be3b39 | 1885 | ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms); |
c97f414c | 1886 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); |
b5be3b39 CH |
1887 | /* the PI implementation requires metadata equal t10 pi tuple size */ |
1888 | if (ns->ms == sizeof(struct t10_pi_tuple)) | |
1889 | ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK; | |
1890 | else | |
1891 | ns->pi_type = 0; | |
1673f1f0 | 1892 | |
6b8190d6 SB |
1893 | if (ns->noiob) |
1894 | nvme_set_chunk_size(ns); | |
24b0b58c | 1895 | nvme_update_disk_info(disk, ns, id); |
32acab31 | 1896 | #ifdef CONFIG_NVME_MULTIPATH |
8f676b85 | 1897 | if (ns->head->disk) { |
32acab31 | 1898 | nvme_update_disk_info(ns->head->disk, ns, id); |
8f676b85 | 1899 | blk_queue_stack_limits(ns->head->disk->queue, ns->queue); |
74e4d20e SG |
1900 | if (bdi_cap_stable_pages_required(ns->queue->backing_dev_info)) { |
1901 | struct backing_dev_info *info = | |
1902 | ns->head->disk->queue->backing_dev_info; | |
1903 | ||
1904 | info->capabilities |= BDI_CAP_STABLE_WRITES; | |
1905 | } | |
1906 | ||
fab7772b | 1907 | revalidate_disk(ns->head->disk); |
8f676b85 | 1908 | } |
32acab31 | 1909 | #endif |
ac81bfa9 | 1910 | } |
1673f1f0 | 1911 | |
ac81bfa9 MB |
1912 | static int nvme_revalidate_disk(struct gendisk *disk) |
1913 | { | |
1914 | struct nvme_ns *ns = disk->private_data; | |
cdbff4f2 CH |
1915 | struct nvme_ctrl *ctrl = ns->ctrl; |
1916 | struct nvme_id_ns *id; | |
002fab04 | 1917 | struct nvme_ns_ids ids; |
cdbff4f2 | 1918 | int ret = 0; |
ac81bfa9 MB |
1919 | |
1920 | if (test_bit(NVME_NS_DEAD, &ns->flags)) { | |
1921 | set_capacity(disk, 0); | |
1922 | return -ENODEV; | |
1923 | } | |
1924 | ||
331813f6 SG |
1925 | ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id); |
1926 | if (ret) | |
1927 | goto out; | |
ac81bfa9 | 1928 | |
cdbff4f2 CH |
1929 | if (id->ncap == 0) { |
1930 | ret = -ENODEV; | |
331813f6 | 1931 | goto free_id; |
cdbff4f2 | 1932 | } |
ac81bfa9 | 1933 | |
5e0fab57 | 1934 | __nvme_revalidate_disk(disk, id); |
538af88e SG |
1935 | ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids); |
1936 | if (ret) | |
1937 | goto free_id; | |
1938 | ||
ed754e5d | 1939 | if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) { |
1d5df6af | 1940 | dev_err(ctrl->device, |
ed754e5d | 1941 | "identifiers changed for nsid %d\n", ns->head->ns_id); |
1d5df6af CH |
1942 | ret = -ENODEV; |
1943 | } | |
1944 | ||
331813f6 | 1945 | free_id: |
cdbff4f2 | 1946 | kfree(id); |
331813f6 | 1947 | out: |
205da243 SG |
1948 | /* |
1949 | * Only fail the function if we got a fatal error back from the | |
1950 | * device, otherwise ignore the error and just move on. | |
1951 | */ | |
1952 | if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR))) | |
1953 | ret = 0; | |
1954 | else if (ret > 0) | |
331813f6 | 1955 | ret = blk_status_to_errno(nvme_error_status(ret)); |
cdbff4f2 | 1956 | return ret; |
1673f1f0 CH |
1957 | } |
1958 | ||
1959 | static char nvme_pr_type(enum pr_type type) | |
1960 | { | |
1961 | switch (type) { | |
1962 | case PR_WRITE_EXCLUSIVE: | |
1963 | return 1; | |
1964 | case PR_EXCLUSIVE_ACCESS: | |
1965 | return 2; | |
1966 | case PR_WRITE_EXCLUSIVE_REG_ONLY: | |
1967 | return 3; | |
1968 | case PR_EXCLUSIVE_ACCESS_REG_ONLY: | |
1969 | return 4; | |
1970 | case PR_WRITE_EXCLUSIVE_ALL_REGS: | |
1971 | return 5; | |
1972 | case PR_EXCLUSIVE_ACCESS_ALL_REGS: | |
1973 | return 6; | |
1974 | default: | |
1975 | return 0; | |
1976 | } | |
1977 | }; | |
1978 | ||
1979 | static int nvme_pr_command(struct block_device *bdev, u32 cdw10, | |
1980 | u64 key, u64 sa_key, u8 op) | |
1981 | { | |
32acab31 CH |
1982 | struct nvme_ns_head *head = NULL; |
1983 | struct nvme_ns *ns; | |
1673f1f0 | 1984 | struct nvme_command c; |
32acab31 | 1985 | int srcu_idx, ret; |
1673f1f0 CH |
1986 | u8 data[16] = { 0, }; |
1987 | ||
b0d61d58 KB |
1988 | ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx); |
1989 | if (unlikely(!ns)) | |
1990 | return -EWOULDBLOCK; | |
1991 | ||
1673f1f0 CH |
1992 | put_unaligned_le64(key, &data[0]); |
1993 | put_unaligned_le64(sa_key, &data[8]); | |
1994 | ||
1995 | memset(&c, 0, sizeof(c)); | |
1996 | c.common.opcode = op; | |
b0d61d58 | 1997 | c.common.nsid = cpu_to_le32(ns->head->ns_id); |
b7c8f366 | 1998 | c.common.cdw10 = cpu_to_le32(cdw10); |
1673f1f0 | 1999 | |
b0d61d58 | 2000 | ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16); |
32acab31 CH |
2001 | nvme_put_ns_from_disk(head, srcu_idx); |
2002 | return ret; | |
1673f1f0 CH |
2003 | } |
2004 | ||
2005 | static int nvme_pr_register(struct block_device *bdev, u64 old, | |
2006 | u64 new, unsigned flags) | |
2007 | { | |
2008 | u32 cdw10; | |
2009 | ||
2010 | if (flags & ~PR_FL_IGNORE_KEY) | |
2011 | return -EOPNOTSUPP; | |
2012 | ||
2013 | cdw10 = old ? 2 : 0; | |
2014 | cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; | |
2015 | cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ | |
2016 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register); | |
2017 | } | |
2018 | ||
2019 | static int nvme_pr_reserve(struct block_device *bdev, u64 key, | |
2020 | enum pr_type type, unsigned flags) | |
2021 | { | |
2022 | u32 cdw10; | |
2023 | ||
2024 | if (flags & ~PR_FL_IGNORE_KEY) | |
2025 | return -EOPNOTSUPP; | |
2026 | ||
2027 | cdw10 = nvme_pr_type(type) << 8; | |
2028 | cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); | |
2029 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire); | |
2030 | } | |
2031 | ||
2032 | static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, | |
2033 | enum pr_type type, bool abort) | |
2034 | { | |
e9a9853c | 2035 | u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1); |
1673f1f0 CH |
2036 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire); |
2037 | } | |
2038 | ||
2039 | static int nvme_pr_clear(struct block_device *bdev, u64 key) | |
2040 | { | |
8c0b3915 | 2041 | u32 cdw10 = 1 | (key ? 1 << 3 : 0); |
1673f1f0 CH |
2042 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register); |
2043 | } | |
2044 | ||
2045 | static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2046 | { | |
e9a9853c | 2047 | u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0); |
1673f1f0 CH |
2048 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release); |
2049 | } | |
2050 | ||
2051 | static const struct pr_ops nvme_pr_ops = { | |
2052 | .pr_register = nvme_pr_register, | |
2053 | .pr_reserve = nvme_pr_reserve, | |
2054 | .pr_release = nvme_pr_release, | |
2055 | .pr_preempt = nvme_pr_preempt, | |
2056 | .pr_clear = nvme_pr_clear, | |
2057 | }; | |
2058 | ||
a98e58e5 | 2059 | #ifdef CONFIG_BLK_SED_OPAL |
4f1244c8 CH |
2060 | int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len, |
2061 | bool send) | |
a98e58e5 | 2062 | { |
4f1244c8 | 2063 | struct nvme_ctrl *ctrl = data; |
a98e58e5 | 2064 | struct nvme_command cmd; |
a98e58e5 SB |
2065 | |
2066 | memset(&cmd, 0, sizeof(cmd)); | |
2067 | if (send) | |
2068 | cmd.common.opcode = nvme_admin_security_send; | |
2069 | else | |
2070 | cmd.common.opcode = nvme_admin_security_recv; | |
a98e58e5 | 2071 | cmd.common.nsid = 0; |
b7c8f366 CK |
2072 | cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8); |
2073 | cmd.common.cdw11 = cpu_to_le32(len); | |
a98e58e5 SB |
2074 | |
2075 | return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, | |
6287b51c | 2076 | ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false); |
a98e58e5 SB |
2077 | } |
2078 | EXPORT_SYMBOL_GPL(nvme_sec_submit); | |
2079 | #endif /* CONFIG_BLK_SED_OPAL */ | |
2080 | ||
5bae7f73 | 2081 | static const struct block_device_operations nvme_fops = { |
1673f1f0 CH |
2082 | .owner = THIS_MODULE, |
2083 | .ioctl = nvme_ioctl, | |
c225b610 | 2084 | .compat_ioctl = nvme_compat_ioctl, |
1673f1f0 CH |
2085 | .open = nvme_open, |
2086 | .release = nvme_release, | |
2087 | .getgeo = nvme_getgeo, | |
2088 | .revalidate_disk= nvme_revalidate_disk, | |
2089 | .pr_ops = &nvme_pr_ops, | |
2090 | }; | |
2091 | ||
32acab31 CH |
2092 | #ifdef CONFIG_NVME_MULTIPATH |
2093 | static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode) | |
2094 | { | |
2095 | struct nvme_ns_head *head = bdev->bd_disk->private_data; | |
2096 | ||
2097 | if (!kref_get_unless_zero(&head->ref)) | |
2098 | return -ENXIO; | |
2099 | return 0; | |
2100 | } | |
2101 | ||
2102 | static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode) | |
2103 | { | |
2104 | nvme_put_ns_head(disk->private_data); | |
2105 | } | |
2106 | ||
2107 | const struct block_device_operations nvme_ns_head_ops = { | |
2108 | .owner = THIS_MODULE, | |
2109 | .open = nvme_ns_head_open, | |
2110 | .release = nvme_ns_head_release, | |
2111 | .ioctl = nvme_ioctl, | |
c225b610 | 2112 | .compat_ioctl = nvme_compat_ioctl, |
32acab31 CH |
2113 | .getgeo = nvme_getgeo, |
2114 | .pr_ops = &nvme_pr_ops, | |
2115 | }; | |
2116 | #endif /* CONFIG_NVME_MULTIPATH */ | |
2117 | ||
5fd4ce1b CH |
2118 | static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled) |
2119 | { | |
2120 | unsigned long timeout = | |
2121 | ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
2122 | u32 csts, bit = enabled ? NVME_CSTS_RDY : 0; | |
2123 | int ret; | |
2124 | ||
2125 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
0df1e4f5 KB |
2126 | if (csts == ~0) |
2127 | return -ENODEV; | |
5fd4ce1b CH |
2128 | if ((csts & NVME_CSTS_RDY) == bit) |
2129 | break; | |
2130 | ||
3e98c244 | 2131 | usleep_range(1000, 2000); |
5fd4ce1b CH |
2132 | if (fatal_signal_pending(current)) |
2133 | return -EINTR; | |
2134 | if (time_after(jiffies, timeout)) { | |
1b3c47c1 | 2135 | dev_err(ctrl->device, |
94d2e705 RG |
2136 | "Device not ready; aborting %s, CSTS=0x%x\n", |
2137 | enabled ? "initialisation" : "reset", csts); | |
5fd4ce1b CH |
2138 | return -ENODEV; |
2139 | } | |
2140 | } | |
2141 | ||
2142 | return ret; | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * If the device has been passed off to us in an enabled state, just clear | |
2147 | * the enabled bit. The spec says we should set the 'shutdown notification | |
2148 | * bits', but doing so may cause the device to complete commands to the | |
2149 | * admin queue ... and we don't know what memory that might be pointing at! | |
2150 | */ | |
b5b05048 | 2151 | int nvme_disable_ctrl(struct nvme_ctrl *ctrl) |
5fd4ce1b CH |
2152 | { |
2153 | int ret; | |
2154 | ||
2155 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
2156 | ctrl->ctrl_config &= ~NVME_CC_ENABLE; | |
2157 | ||
2158 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
2159 | if (ret) | |
2160 | return ret; | |
54adc010 | 2161 | |
b5a10c5f | 2162 | if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) |
54adc010 GP |
2163 | msleep(NVME_QUIRK_DELAY_AMOUNT); |
2164 | ||
b5b05048 | 2165 | return nvme_wait_ready(ctrl, ctrl->cap, false); |
5fd4ce1b | 2166 | } |
576d55d6 | 2167 | EXPORT_SYMBOL_GPL(nvme_disable_ctrl); |
5fd4ce1b | 2168 | |
c0f2f45b | 2169 | int nvme_enable_ctrl(struct nvme_ctrl *ctrl) |
5fd4ce1b CH |
2170 | { |
2171 | /* | |
2172 | * Default to a 4K page size, with the intention to update this | |
2173 | * path in the future to accomodate architectures with differing | |
2174 | * kernel and IO page sizes. | |
2175 | */ | |
c0f2f45b | 2176 | unsigned dev_page_min, page_shift = 12; |
5fd4ce1b CH |
2177 | int ret; |
2178 | ||
c0f2f45b SG |
2179 | ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap); |
2180 | if (ret) { | |
2181 | dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret); | |
2182 | return ret; | |
2183 | } | |
2184 | dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12; | |
2185 | ||
5fd4ce1b | 2186 | if (page_shift < dev_page_min) { |
1b3c47c1 | 2187 | dev_err(ctrl->device, |
5fd4ce1b CH |
2188 | "Minimum device page size %u too large for host (%u)\n", |
2189 | 1 << dev_page_min, 1 << page_shift); | |
2190 | return -ENODEV; | |
2191 | } | |
2192 | ||
2193 | ctrl->page_size = 1 << page_shift; | |
2194 | ||
2195 | ctrl->ctrl_config = NVME_CC_CSS_NVM; | |
2196 | ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; | |
60b43f62 | 2197 | ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE; |
5fd4ce1b CH |
2198 | ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
2199 | ctrl->ctrl_config |= NVME_CC_ENABLE; | |
2200 | ||
2201 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
2202 | if (ret) | |
2203 | return ret; | |
c0f2f45b | 2204 | return nvme_wait_ready(ctrl, ctrl->cap, true); |
5fd4ce1b | 2205 | } |
576d55d6 | 2206 | EXPORT_SYMBOL_GPL(nvme_enable_ctrl); |
5fd4ce1b CH |
2207 | |
2208 | int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl) | |
2209 | { | |
07fbd32a | 2210 | unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ); |
5fd4ce1b CH |
2211 | u32 csts; |
2212 | int ret; | |
2213 | ||
2214 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
2215 | ctrl->ctrl_config |= NVME_CC_SHN_NORMAL; | |
2216 | ||
2217 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
2218 | if (ret) | |
2219 | return ret; | |
2220 | ||
2221 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
2222 | if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT) | |
2223 | break; | |
2224 | ||
2225 | msleep(100); | |
2226 | if (fatal_signal_pending(current)) | |
2227 | return -EINTR; | |
2228 | if (time_after(jiffies, timeout)) { | |
1b3c47c1 | 2229 | dev_err(ctrl->device, |
5fd4ce1b CH |
2230 | "Device shutdown incomplete; abort shutdown\n"); |
2231 | return -ENODEV; | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | return ret; | |
2236 | } | |
576d55d6 | 2237 | EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl); |
5fd4ce1b | 2238 | |
da35825d CH |
2239 | static void nvme_set_queue_limits(struct nvme_ctrl *ctrl, |
2240 | struct request_queue *q) | |
2241 | { | |
7c88cb00 JA |
2242 | bool vwc = false; |
2243 | ||
da35825d | 2244 | if (ctrl->max_hw_sectors) { |
45686b61 CH |
2245 | u32 max_segments = |
2246 | (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1; | |
2247 | ||
943e942e | 2248 | max_segments = min_not_zero(max_segments, ctrl->max_segments); |
da35825d | 2249 | blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors); |
45686b61 | 2250 | blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX)); |
da35825d | 2251 | } |
249159c5 KB |
2252 | if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && |
2253 | is_power_of_2(ctrl->max_hw_sectors)) | |
e6282aef | 2254 | blk_queue_chunk_sectors(q, ctrl->max_hw_sectors); |
da35825d | 2255 | blk_queue_virt_boundary(q, ctrl->page_size - 1); |
7c88cb00 JA |
2256 | if (ctrl->vwc & NVME_CTRL_VWC_PRESENT) |
2257 | vwc = true; | |
2258 | blk_queue_write_cache(q, vwc, vwc); | |
da35825d CH |
2259 | } |
2260 | ||
dbf86b39 JD |
2261 | static int nvme_configure_timestamp(struct nvme_ctrl *ctrl) |
2262 | { | |
2263 | __le64 ts; | |
2264 | int ret; | |
2265 | ||
2266 | if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP)) | |
2267 | return 0; | |
2268 | ||
2269 | ts = cpu_to_le64(ktime_to_ms(ktime_get_real())); | |
2270 | ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts), | |
2271 | NULL); | |
2272 | if (ret) | |
2273 | dev_warn_once(ctrl->device, | |
2274 | "could not set timestamp (%d)\n", ret); | |
2275 | return ret; | |
2276 | } | |
2277 | ||
49cd84b6 KB |
2278 | static int nvme_configure_acre(struct nvme_ctrl *ctrl) |
2279 | { | |
2280 | struct nvme_feat_host_behavior *host; | |
2281 | int ret; | |
2282 | ||
2283 | /* Don't bother enabling the feature if retry delay is not reported */ | |
2284 | if (!ctrl->crdt[0]) | |
2285 | return 0; | |
2286 | ||
2287 | host = kzalloc(sizeof(*host), GFP_KERNEL); | |
2288 | if (!host) | |
2289 | return 0; | |
2290 | ||
2291 | host->acre = NVME_ENABLE_ACRE; | |
2292 | ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0, | |
2293 | host, sizeof(*host), NULL); | |
2294 | kfree(host); | |
2295 | return ret; | |
2296 | } | |
2297 | ||
634b8325 | 2298 | static int nvme_configure_apst(struct nvme_ctrl *ctrl) |
c5552fde AL |
2299 | { |
2300 | /* | |
2301 | * APST (Autonomous Power State Transition) lets us program a | |
2302 | * table of power state transitions that the controller will | |
2303 | * perform automatically. We configure it with a simple | |
2304 | * heuristic: we are willing to spend at most 2% of the time | |
2305 | * transitioning between power states. Therefore, when running | |
2306 | * in any given state, we will enter the next lower-power | |
76e4ad09 | 2307 | * non-operational state after waiting 50 * (enlat + exlat) |
da87591b | 2308 | * microseconds, as long as that state's exit latency is under |
c5552fde AL |
2309 | * the requested maximum latency. |
2310 | * | |
2311 | * We will not autonomously enter any non-operational state for | |
2312 | * which the total latency exceeds ps_max_latency_us. Users | |
2313 | * can set ps_max_latency_us to zero to turn off APST. | |
2314 | */ | |
2315 | ||
2316 | unsigned apste; | |
2317 | struct nvme_feat_auto_pst *table; | |
fb0dc399 AL |
2318 | u64 max_lat_us = 0; |
2319 | int max_ps = -1; | |
c5552fde AL |
2320 | int ret; |
2321 | ||
2322 | /* | |
2323 | * If APST isn't supported or if we haven't been initialized yet, | |
2324 | * then don't do anything. | |
2325 | */ | |
2326 | if (!ctrl->apsta) | |
634b8325 | 2327 | return 0; |
c5552fde AL |
2328 | |
2329 | if (ctrl->npss > 31) { | |
2330 | dev_warn(ctrl->device, "NPSS is invalid; not using APST\n"); | |
634b8325 | 2331 | return 0; |
c5552fde AL |
2332 | } |
2333 | ||
2334 | table = kzalloc(sizeof(*table), GFP_KERNEL); | |
2335 | if (!table) | |
634b8325 | 2336 | return 0; |
c5552fde | 2337 | |
76a5af84 | 2338 | if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) { |
c5552fde AL |
2339 | /* Turn off APST. */ |
2340 | apste = 0; | |
fb0dc399 | 2341 | dev_dbg(ctrl->device, "APST disabled\n"); |
c5552fde AL |
2342 | } else { |
2343 | __le64 target = cpu_to_le64(0); | |
2344 | int state; | |
2345 | ||
2346 | /* | |
2347 | * Walk through all states from lowest- to highest-power. | |
2348 | * According to the spec, lower-numbered states use more | |
2349 | * power. NPSS, despite the name, is the index of the | |
2350 | * lowest-power state, not the number of states. | |
2351 | */ | |
2352 | for (state = (int)ctrl->npss; state >= 0; state--) { | |
da87591b | 2353 | u64 total_latency_us, exit_latency_us, transition_ms; |
c5552fde AL |
2354 | |
2355 | if (target) | |
2356 | table->entries[state] = target; | |
2357 | ||
ff5350a8 AL |
2358 | /* |
2359 | * Don't allow transitions to the deepest state | |
2360 | * if it's quirked off. | |
2361 | */ | |
2362 | if (state == ctrl->npss && | |
2363 | (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) | |
2364 | continue; | |
2365 | ||
c5552fde AL |
2366 | /* |
2367 | * Is this state a useful non-operational state for | |
2368 | * higher-power states to autonomously transition to? | |
2369 | */ | |
2370 | if (!(ctrl->psd[state].flags & | |
2371 | NVME_PS_FLAGS_NON_OP_STATE)) | |
2372 | continue; | |
2373 | ||
da87591b KHF |
2374 | exit_latency_us = |
2375 | (u64)le32_to_cpu(ctrl->psd[state].exit_lat); | |
2376 | if (exit_latency_us > ctrl->ps_max_latency_us) | |
c5552fde AL |
2377 | continue; |
2378 | ||
da87591b KHF |
2379 | total_latency_us = |
2380 | exit_latency_us + | |
2381 | le32_to_cpu(ctrl->psd[state].entry_lat); | |
2382 | ||
c5552fde AL |
2383 | /* |
2384 | * This state is good. Use it as the APST idle | |
2385 | * target for higher power states. | |
2386 | */ | |
2387 | transition_ms = total_latency_us + 19; | |
2388 | do_div(transition_ms, 20); | |
2389 | if (transition_ms > (1 << 24) - 1) | |
2390 | transition_ms = (1 << 24) - 1; | |
2391 | ||
2392 | target = cpu_to_le64((state << 3) | | |
2393 | (transition_ms << 8)); | |
fb0dc399 AL |
2394 | |
2395 | if (max_ps == -1) | |
2396 | max_ps = state; | |
2397 | ||
2398 | if (total_latency_us > max_lat_us) | |
2399 | max_lat_us = total_latency_us; | |
c5552fde AL |
2400 | } |
2401 | ||
2402 | apste = 1; | |
fb0dc399 AL |
2403 | |
2404 | if (max_ps == -1) { | |
2405 | dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n"); | |
2406 | } else { | |
2407 | dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n", | |
2408 | max_ps, max_lat_us, (int)sizeof(*table), table); | |
2409 | } | |
c5552fde AL |
2410 | } |
2411 | ||
2412 | ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste, | |
2413 | table, sizeof(*table), NULL); | |
2414 | if (ret) | |
2415 | dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret); | |
2416 | ||
2417 | kfree(table); | |
634b8325 | 2418 | return ret; |
c5552fde AL |
2419 | } |
2420 | ||
2421 | static void nvme_set_latency_tolerance(struct device *dev, s32 val) | |
2422 | { | |
2423 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
2424 | u64 latency; | |
2425 | ||
2426 | switch (val) { | |
2427 | case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT: | |
2428 | case PM_QOS_LATENCY_ANY: | |
2429 | latency = U64_MAX; | |
2430 | break; | |
2431 | ||
2432 | default: | |
2433 | latency = val; | |
2434 | } | |
2435 | ||
2436 | if (ctrl->ps_max_latency_us != latency) { | |
2437 | ctrl->ps_max_latency_us = latency; | |
2438 | nvme_configure_apst(ctrl); | |
2439 | } | |
2440 | } | |
2441 | ||
bd4da3ab AL |
2442 | struct nvme_core_quirk_entry { |
2443 | /* | |
2444 | * NVMe model and firmware strings are padded with spaces. For | |
2445 | * simplicity, strings in the quirk table are padded with NULLs | |
2446 | * instead. | |
2447 | */ | |
2448 | u16 vid; | |
2449 | const char *mn; | |
2450 | const char *fr; | |
2451 | unsigned long quirks; | |
2452 | }; | |
2453 | ||
2454 | static const struct nvme_core_quirk_entry core_quirks[] = { | |
c5552fde | 2455 | { |
be56945c AL |
2456 | /* |
2457 | * This Toshiba device seems to die using any APST states. See: | |
2458 | * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11 | |
2459 | */ | |
2460 | .vid = 0x1179, | |
2461 | .mn = "THNSF5256GPUK TOSHIBA", | |
c5552fde | 2462 | .quirks = NVME_QUIRK_NO_APST, |
cb32de1b ML |
2463 | }, |
2464 | { | |
2465 | /* | |
2466 | * This LiteON CL1-3D*-Q11 firmware version has a race | |
2467 | * condition associated with actions related to suspend to idle | |
2468 | * LiteON has resolved the problem in future firmware | |
2469 | */ | |
2470 | .vid = 0x14a4, | |
2471 | .fr = "22301111", | |
2472 | .quirks = NVME_QUIRK_SIMPLE_SUSPEND, | |
be56945c | 2473 | } |
bd4da3ab AL |
2474 | }; |
2475 | ||
2476 | /* match is null-terminated but idstr is space-padded. */ | |
2477 | static bool string_matches(const char *idstr, const char *match, size_t len) | |
2478 | { | |
2479 | size_t matchlen; | |
2480 | ||
2481 | if (!match) | |
2482 | return true; | |
2483 | ||
2484 | matchlen = strlen(match); | |
2485 | WARN_ON_ONCE(matchlen > len); | |
2486 | ||
2487 | if (memcmp(idstr, match, matchlen)) | |
2488 | return false; | |
2489 | ||
2490 | for (; matchlen < len; matchlen++) | |
2491 | if (idstr[matchlen] != ' ') | |
2492 | return false; | |
2493 | ||
2494 | return true; | |
2495 | } | |
2496 | ||
2497 | static bool quirk_matches(const struct nvme_id_ctrl *id, | |
2498 | const struct nvme_core_quirk_entry *q) | |
2499 | { | |
2500 | return q->vid == le16_to_cpu(id->vid) && | |
2501 | string_matches(id->mn, q->mn, sizeof(id->mn)) && | |
2502 | string_matches(id->fr, q->fr, sizeof(id->fr)); | |
2503 | } | |
2504 | ||
ab9e00cc CH |
2505 | static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl, |
2506 | struct nvme_id_ctrl *id) | |
180de007 CH |
2507 | { |
2508 | size_t nqnlen; | |
2509 | int off; | |
2510 | ||
6299358d JD |
2511 | if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) { |
2512 | nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE); | |
2513 | if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) { | |
2514 | strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE); | |
2515 | return; | |
2516 | } | |
180de007 | 2517 | |
6299358d JD |
2518 | if (ctrl->vs >= NVME_VS(1, 2, 1)) |
2519 | dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n"); | |
2520 | } | |
180de007 CH |
2521 | |
2522 | /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */ | |
ab9e00cc | 2523 | off = snprintf(subsys->subnqn, NVMF_NQN_SIZE, |
3da584f5 | 2524 | "nqn.2014.08.org.nvmexpress:%04x%04x", |
180de007 | 2525 | le16_to_cpu(id->vid), le16_to_cpu(id->ssvid)); |
ab9e00cc | 2526 | memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn)); |
180de007 | 2527 | off += sizeof(id->sn); |
ab9e00cc | 2528 | memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn)); |
180de007 | 2529 | off += sizeof(id->mn); |
ab9e00cc CH |
2530 | memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off); |
2531 | } | |
2532 | ||
e654dfd3 | 2533 | static void nvme_release_subsystem(struct device *dev) |
ab9e00cc | 2534 | { |
e654dfd3 LG |
2535 | struct nvme_subsystem *subsys = |
2536 | container_of(dev, struct nvme_subsystem, dev); | |
2537 | ||
733e4b69 KB |
2538 | if (subsys->instance >= 0) |
2539 | ida_simple_remove(&nvme_instance_ida, subsys->instance); | |
ab9e00cc CH |
2540 | kfree(subsys); |
2541 | } | |
2542 | ||
ab9e00cc CH |
2543 | static void nvme_destroy_subsystem(struct kref *ref) |
2544 | { | |
2545 | struct nvme_subsystem *subsys = | |
2546 | container_of(ref, struct nvme_subsystem, ref); | |
2547 | ||
2548 | mutex_lock(&nvme_subsystems_lock); | |
2549 | list_del(&subsys->entry); | |
2550 | mutex_unlock(&nvme_subsystems_lock); | |
2551 | ||
ed754e5d | 2552 | ida_destroy(&subsys->ns_ida); |
ab9e00cc CH |
2553 | device_del(&subsys->dev); |
2554 | put_device(&subsys->dev); | |
2555 | } | |
2556 | ||
2557 | static void nvme_put_subsystem(struct nvme_subsystem *subsys) | |
2558 | { | |
2559 | kref_put(&subsys->ref, nvme_destroy_subsystem); | |
2560 | } | |
2561 | ||
2562 | static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn) | |
2563 | { | |
2564 | struct nvme_subsystem *subsys; | |
2565 | ||
2566 | lockdep_assert_held(&nvme_subsystems_lock); | |
2567 | ||
c26aa572 JS |
2568 | /* |
2569 | * Fail matches for discovery subsystems. This results | |
2570 | * in each discovery controller bound to a unique subsystem. | |
2571 | * This avoids issues with validating controller values | |
2572 | * that can only be true when there is a single unique subsystem. | |
2573 | * There may be multiple and completely independent entities | |
2574 | * that provide discovery controllers. | |
2575 | */ | |
2576 | if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME)) | |
2577 | return NULL; | |
2578 | ||
ab9e00cc CH |
2579 | list_for_each_entry(subsys, &nvme_subsystems, entry) { |
2580 | if (strcmp(subsys->subnqn, subsysnqn)) | |
2581 | continue; | |
2582 | if (!kref_get_unless_zero(&subsys->ref)) | |
2583 | continue; | |
2584 | return subsys; | |
2585 | } | |
2586 | ||
2587 | return NULL; | |
2588 | } | |
2589 | ||
1e496938 HR |
2590 | #define SUBSYS_ATTR_RO(_name, _mode, _show) \ |
2591 | struct device_attribute subsys_attr_##_name = \ | |
2592 | __ATTR(_name, _mode, _show, NULL) | |
2593 | ||
2594 | static ssize_t nvme_subsys_show_nqn(struct device *dev, | |
2595 | struct device_attribute *attr, | |
2596 | char *buf) | |
2597 | { | |
2598 | struct nvme_subsystem *subsys = | |
2599 | container_of(dev, struct nvme_subsystem, dev); | |
2600 | ||
2601 | return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn); | |
2602 | } | |
2603 | static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn); | |
2604 | ||
2605 | #define nvme_subsys_show_str_function(field) \ | |
2606 | static ssize_t subsys_##field##_show(struct device *dev, \ | |
2607 | struct device_attribute *attr, char *buf) \ | |
2608 | { \ | |
2609 | struct nvme_subsystem *subsys = \ | |
2610 | container_of(dev, struct nvme_subsystem, dev); \ | |
2611 | return sprintf(buf, "%.*s\n", \ | |
2612 | (int)sizeof(subsys->field), subsys->field); \ | |
2613 | } \ | |
2614 | static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show); | |
2615 | ||
2616 | nvme_subsys_show_str_function(model); | |
2617 | nvme_subsys_show_str_function(serial); | |
2618 | nvme_subsys_show_str_function(firmware_rev); | |
2619 | ||
2620 | static struct attribute *nvme_subsys_attrs[] = { | |
2621 | &subsys_attr_model.attr, | |
2622 | &subsys_attr_serial.attr, | |
2623 | &subsys_attr_firmware_rev.attr, | |
2624 | &subsys_attr_subsysnqn.attr, | |
75c10e73 HR |
2625 | #ifdef CONFIG_NVME_MULTIPATH |
2626 | &subsys_attr_iopolicy.attr, | |
2627 | #endif | |
1e496938 HR |
2628 | NULL, |
2629 | }; | |
2630 | ||
2631 | static struct attribute_group nvme_subsys_attrs_group = { | |
2632 | .attrs = nvme_subsys_attrs, | |
2633 | }; | |
2634 | ||
2635 | static const struct attribute_group *nvme_subsys_attrs_groups[] = { | |
2636 | &nvme_subsys_attrs_group, | |
2637 | NULL, | |
2638 | }; | |
2639 | ||
1b1031ca CH |
2640 | static bool nvme_validate_cntlid(struct nvme_subsystem *subsys, |
2641 | struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id) | |
b837b283 | 2642 | { |
1b1031ca | 2643 | struct nvme_ctrl *tmp; |
b837b283 | 2644 | |
32fd90c4 CH |
2645 | lockdep_assert_held(&nvme_subsystems_lock); |
2646 | ||
1b1031ca | 2647 | list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) { |
e7c43fea | 2648 | if (nvme_state_terminal(tmp)) |
1b1031ca CH |
2649 | continue; |
2650 | ||
2651 | if (tmp->cntlid == ctrl->cntlid) { | |
2652 | dev_err(ctrl->device, | |
2653 | "Duplicate cntlid %u with %s, rejecting\n", | |
2654 | ctrl->cntlid, dev_name(tmp->device)); | |
2655 | return false; | |
2656 | } | |
b837b283 | 2657 | |
1b1031ca CH |
2658 | if ((id->cmic & (1 << 1)) || |
2659 | (ctrl->opts && ctrl->opts->discovery_nqn)) | |
2660 | continue; | |
2661 | ||
2662 | dev_err(ctrl->device, | |
2663 | "Subsystem does not support multiple controllers\n"); | |
2664 | return false; | |
b837b283 | 2665 | } |
b837b283 | 2666 | |
1b1031ca | 2667 | return true; |
b837b283 IR |
2668 | } |
2669 | ||
ab9e00cc CH |
2670 | static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id) |
2671 | { | |
2672 | struct nvme_subsystem *subsys, *found; | |
2673 | int ret; | |
2674 | ||
2675 | subsys = kzalloc(sizeof(*subsys), GFP_KERNEL); | |
2676 | if (!subsys) | |
2677 | return -ENOMEM; | |
733e4b69 KB |
2678 | |
2679 | subsys->instance = -1; | |
ab9e00cc CH |
2680 | mutex_init(&subsys->lock); |
2681 | kref_init(&subsys->ref); | |
2682 | INIT_LIST_HEAD(&subsys->ctrls); | |
ed754e5d | 2683 | INIT_LIST_HEAD(&subsys->nsheads); |
ab9e00cc CH |
2684 | nvme_init_subnqn(subsys, ctrl, id); |
2685 | memcpy(subsys->serial, id->sn, sizeof(subsys->serial)); | |
2686 | memcpy(subsys->model, id->mn, sizeof(subsys->model)); | |
2687 | memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev)); | |
2688 | subsys->vendor_id = le16_to_cpu(id->vid); | |
2689 | subsys->cmic = id->cmic; | |
81adb863 | 2690 | subsys->awupf = le16_to_cpu(id->awupf); |
75c10e73 HR |
2691 | #ifdef CONFIG_NVME_MULTIPATH |
2692 | subsys->iopolicy = NVME_IOPOLICY_NUMA; | |
2693 | #endif | |
ab9e00cc CH |
2694 | |
2695 | subsys->dev.class = nvme_subsys_class; | |
2696 | subsys->dev.release = nvme_release_subsystem; | |
1e496938 | 2697 | subsys->dev.groups = nvme_subsys_attrs_groups; |
733e4b69 | 2698 | dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance); |
ab9e00cc CH |
2699 | device_initialize(&subsys->dev); |
2700 | ||
2701 | mutex_lock(&nvme_subsystems_lock); | |
2702 | found = __nvme_find_get_subsystem(subsys->subnqn); | |
2703 | if (found) { | |
e654dfd3 | 2704 | put_device(&subsys->dev); |
ab9e00cc | 2705 | subsys = found; |
32fd90c4 | 2706 | |
1b1031ca | 2707 | if (!nvme_validate_cntlid(subsys, ctrl, id)) { |
ab9e00cc | 2708 | ret = -EINVAL; |
32fd90c4 | 2709 | goto out_put_subsystem; |
ab9e00cc | 2710 | } |
ab9e00cc CH |
2711 | } else { |
2712 | ret = device_add(&subsys->dev); | |
2713 | if (ret) { | |
2714 | dev_err(ctrl->device, | |
2715 | "failed to register subsystem device.\n"); | |
8c36e66f | 2716 | put_device(&subsys->dev); |
ab9e00cc CH |
2717 | goto out_unlock; |
2718 | } | |
ed754e5d | 2719 | ida_init(&subsys->ns_ida); |
ab9e00cc CH |
2720 | list_add_tail(&subsys->entry, &nvme_subsystems); |
2721 | } | |
2722 | ||
bc4f6e06 DC |
2723 | ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj, |
2724 | dev_name(ctrl->device)); | |
2725 | if (ret) { | |
ab9e00cc CH |
2726 | dev_err(ctrl->device, |
2727 | "failed to create sysfs link from subsystem.\n"); | |
32fd90c4 | 2728 | goto out_put_subsystem; |
ab9e00cc CH |
2729 | } |
2730 | ||
733e4b69 KB |
2731 | if (!found) |
2732 | subsys->instance = ctrl->instance; | |
32fd90c4 | 2733 | ctrl->subsys = subsys; |
ab9e00cc | 2734 | list_add_tail(&ctrl->subsys_entry, &subsys->ctrls); |
32fd90c4 | 2735 | mutex_unlock(&nvme_subsystems_lock); |
ab9e00cc CH |
2736 | return 0; |
2737 | ||
32fd90c4 CH |
2738 | out_put_subsystem: |
2739 | nvme_put_subsystem(subsys); | |
ab9e00cc CH |
2740 | out_unlock: |
2741 | mutex_unlock(&nvme_subsystems_lock); | |
ab9e00cc | 2742 | return ret; |
180de007 CH |
2743 | } |
2744 | ||
0e98719b CH |
2745 | int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, |
2746 | void *log, size_t size, u64 offset) | |
c627c487 KB |
2747 | { |
2748 | struct nvme_command c = { }; | |
70da6094 MB |
2749 | unsigned long dwlen = size / 4 - 1; |
2750 | ||
2751 | c.get_log_page.opcode = nvme_admin_get_log_page; | |
0e98719b | 2752 | c.get_log_page.nsid = cpu_to_le32(nsid); |
70da6094 | 2753 | c.get_log_page.lid = log_page; |
0e98719b | 2754 | c.get_log_page.lsp = lsp; |
70da6094 MB |
2755 | c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1)); |
2756 | c.get_log_page.numdu = cpu_to_le16(dwlen >> 16); | |
7ec6074f MB |
2757 | c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset)); |
2758 | c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset)); | |
c627c487 KB |
2759 | |
2760 | return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size); | |
2761 | } | |
2762 | ||
84fef62d KB |
2763 | static int nvme_get_effects_log(struct nvme_ctrl *ctrl) |
2764 | { | |
2765 | int ret; | |
2766 | ||
2767 | if (!ctrl->effects) | |
2768 | ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL); | |
2769 | ||
2770 | if (!ctrl->effects) | |
2771 | return 0; | |
2772 | ||
0e98719b CH |
2773 | ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0, |
2774 | ctrl->effects, sizeof(*ctrl->effects), 0); | |
84fef62d KB |
2775 | if (ret) { |
2776 | kfree(ctrl->effects); | |
2777 | ctrl->effects = NULL; | |
2778 | } | |
2779 | return ret; | |
180de007 CH |
2780 | } |
2781 | ||
7fd8930f CH |
2782 | /* |
2783 | * Initialize the cached copies of the Identify data and various controller | |
2784 | * register in our nvme_ctrl structure. This should be called as soon as | |
2785 | * the admin queue is fully up and running. | |
2786 | */ | |
2787 | int nvme_init_identify(struct nvme_ctrl *ctrl) | |
2788 | { | |
2789 | struct nvme_id_ctrl *id; | |
7fd8930f | 2790 | int ret, page_shift; |
a229dbf6 | 2791 | u32 max_hw_sectors; |
76a5af84 | 2792 | bool prev_apst_enabled; |
7fd8930f | 2793 | |
f3ca80fc CH |
2794 | ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs); |
2795 | if (ret) { | |
1b3c47c1 | 2796 | dev_err(ctrl->device, "Reading VS failed (%d)\n", ret); |
f3ca80fc CH |
2797 | return ret; |
2798 | } | |
4fba4458 | 2799 | page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12; |
c0f2f45b | 2800 | ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize); |
7fd8930f | 2801 | |
8ef2074d | 2802 | if (ctrl->vs >= NVME_VS(1, 1, 0)) |
4fba4458 | 2803 | ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap); |
f3ca80fc | 2804 | |
7fd8930f CH |
2805 | ret = nvme_identify_ctrl(ctrl, &id); |
2806 | if (ret) { | |
1b3c47c1 | 2807 | dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret); |
7fd8930f CH |
2808 | return -EIO; |
2809 | } | |
2810 | ||
84fef62d KB |
2811 | if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) { |
2812 | ret = nvme_get_effects_log(ctrl); | |
2813 | if (ret < 0) | |
75c8b19a | 2814 | goto out_free; |
84fef62d | 2815 | } |
180de007 | 2816 | |
a89fcca8 GP |
2817 | if (!(ctrl->ops->flags & NVME_F_FABRICS)) |
2818 | ctrl->cntlid = le16_to_cpu(id->cntlid); | |
2819 | ||
bd4da3ab | 2820 | if (!ctrl->identified) { |
ab9e00cc CH |
2821 | int i; |
2822 | ||
2823 | ret = nvme_init_subsystem(ctrl, id); | |
2824 | if (ret) | |
2825 | goto out_free; | |
2826 | ||
bd4da3ab AL |
2827 | /* |
2828 | * Check for quirks. Quirk can depend on firmware version, | |
2829 | * so, in principle, the set of quirks present can change | |
2830 | * across a reset. As a possible future enhancement, we | |
2831 | * could re-scan for quirks every time we reinitialize | |
2832 | * the device, but we'd have to make sure that the driver | |
2833 | * behaves intelligently if the quirks change. | |
2834 | */ | |
bd4da3ab AL |
2835 | for (i = 0; i < ARRAY_SIZE(core_quirks); i++) { |
2836 | if (quirk_matches(id, &core_quirks[i])) | |
2837 | ctrl->quirks |= core_quirks[i].quirks; | |
2838 | } | |
2839 | } | |
2840 | ||
c35e30b4 | 2841 | if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) { |
f0425db0 | 2842 | dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n"); |
c35e30b4 AL |
2843 | ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS; |
2844 | } | |
2845 | ||
49cd84b6 KB |
2846 | ctrl->crdt[0] = le16_to_cpu(id->crdt1); |
2847 | ctrl->crdt[1] = le16_to_cpu(id->crdt2); | |
2848 | ctrl->crdt[2] = le16_to_cpu(id->crdt3); | |
2849 | ||
8a9ae523 | 2850 | ctrl->oacs = le16_to_cpu(id->oacs); |
43e2d08d | 2851 | ctrl->oncs = le16_to_cpu(id->oncs); |
2d466c7a | 2852 | ctrl->mtfa = le16_to_cpu(id->mtfa); |
c0561f82 | 2853 | ctrl->oaes = le32_to_cpu(id->oaes); |
400b6a7b GR |
2854 | ctrl->wctemp = le16_to_cpu(id->wctemp); |
2855 | ctrl->cctemp = le16_to_cpu(id->cctemp); | |
2856 | ||
6bf25d16 | 2857 | atomic_set(&ctrl->abort_limit, id->acl + 1); |
7fd8930f | 2858 | ctrl->vwc = id->vwc; |
7fd8930f | 2859 | if (id->mdts) |
a229dbf6 | 2860 | max_hw_sectors = 1 << (id->mdts + page_shift - 9); |
7fd8930f | 2861 | else |
a229dbf6 CH |
2862 | max_hw_sectors = UINT_MAX; |
2863 | ctrl->max_hw_sectors = | |
2864 | min_not_zero(ctrl->max_hw_sectors, max_hw_sectors); | |
7fd8930f | 2865 | |
da35825d | 2866 | nvme_set_queue_limits(ctrl, ctrl->admin_q); |
07bfcd09 | 2867 | ctrl->sgls = le32_to_cpu(id->sgls); |
038bd4cb | 2868 | ctrl->kas = le16_to_cpu(id->kas); |
0d0b660f | 2869 | ctrl->max_namespaces = le32_to_cpu(id->mnan); |
3e53ba38 | 2870 | ctrl->ctratt = le32_to_cpu(id->ctratt); |
07bfcd09 | 2871 | |
07fbd32a MP |
2872 | if (id->rtd3e) { |
2873 | /* us -> s */ | |
2874 | u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000; | |
2875 | ||
2876 | ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time, | |
2877 | shutdown_timeout, 60); | |
2878 | ||
2879 | if (ctrl->shutdown_timeout != shutdown_timeout) | |
1a3838d7 | 2880 | dev_info(ctrl->device, |
07fbd32a MP |
2881 | "Shutdown timeout set to %u seconds\n", |
2882 | ctrl->shutdown_timeout); | |
2883 | } else | |
2884 | ctrl->shutdown_timeout = shutdown_timeout; | |
2885 | ||
c5552fde | 2886 | ctrl->npss = id->npss; |
76a5af84 KHF |
2887 | ctrl->apsta = id->apsta; |
2888 | prev_apst_enabled = ctrl->apst_enabled; | |
c35e30b4 AL |
2889 | if (ctrl->quirks & NVME_QUIRK_NO_APST) { |
2890 | if (force_apst && id->apsta) { | |
f0425db0 | 2891 | dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n"); |
76a5af84 | 2892 | ctrl->apst_enabled = true; |
c35e30b4 | 2893 | } else { |
76a5af84 | 2894 | ctrl->apst_enabled = false; |
c35e30b4 AL |
2895 | } |
2896 | } else { | |
76a5af84 | 2897 | ctrl->apst_enabled = id->apsta; |
c35e30b4 | 2898 | } |
c5552fde AL |
2899 | memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd)); |
2900 | ||
d3d5b87d | 2901 | if (ctrl->ops->flags & NVME_F_FABRICS) { |
07bfcd09 CH |
2902 | ctrl->icdoff = le16_to_cpu(id->icdoff); |
2903 | ctrl->ioccsz = le32_to_cpu(id->ioccsz); | |
2904 | ctrl->iorcsz = le32_to_cpu(id->iorcsz); | |
2905 | ctrl->maxcmd = le16_to_cpu(id->maxcmd); | |
2906 | ||
2907 | /* | |
2908 | * In fabrics we need to verify the cntlid matches the | |
2909 | * admin connect | |
2910 | */ | |
634b8325 | 2911 | if (ctrl->cntlid != le16_to_cpu(id->cntlid)) { |
a8157ff3 JS |
2912 | dev_err(ctrl->device, |
2913 | "Mismatching cntlid: Connect %u vs Identify " | |
2914 | "%u, rejecting\n", | |
2915 | ctrl->cntlid, le16_to_cpu(id->cntlid)); | |
07bfcd09 | 2916 | ret = -EINVAL; |
634b8325 KB |
2917 | goto out_free; |
2918 | } | |
038bd4cb SG |
2919 | |
2920 | if (!ctrl->opts->discovery_nqn && !ctrl->kas) { | |
f0425db0 | 2921 | dev_err(ctrl->device, |
038bd4cb SG |
2922 | "keep-alive support is mandatory for fabrics\n"); |
2923 | ret = -EINVAL; | |
634b8325 | 2924 | goto out_free; |
038bd4cb | 2925 | } |
07bfcd09 | 2926 | } else { |
fe6d53c9 CH |
2927 | ctrl->hmpre = le32_to_cpu(id->hmpre); |
2928 | ctrl->hmmin = le32_to_cpu(id->hmmin); | |
044a9df1 CH |
2929 | ctrl->hmminds = le32_to_cpu(id->hmminds); |
2930 | ctrl->hmmaxd = le16_to_cpu(id->hmmaxd); | |
07bfcd09 | 2931 | } |
da35825d | 2932 | |
0d0b660f | 2933 | ret = nvme_mpath_init(ctrl, id); |
7fd8930f | 2934 | kfree(id); |
bd4da3ab | 2935 | |
0d0b660f CH |
2936 | if (ret < 0) |
2937 | return ret; | |
2938 | ||
76a5af84 | 2939 | if (ctrl->apst_enabled && !prev_apst_enabled) |
c5552fde | 2940 | dev_pm_qos_expose_latency_tolerance(ctrl->device); |
76a5af84 | 2941 | else if (!ctrl->apst_enabled && prev_apst_enabled) |
c5552fde AL |
2942 | dev_pm_qos_hide_latency_tolerance(ctrl->device); |
2943 | ||
634b8325 KB |
2944 | ret = nvme_configure_apst(ctrl); |
2945 | if (ret < 0) | |
2946 | return ret; | |
dbf86b39 JD |
2947 | |
2948 | ret = nvme_configure_timestamp(ctrl); | |
2949 | if (ret < 0) | |
2950 | return ret; | |
634b8325 KB |
2951 | |
2952 | ret = nvme_configure_directives(ctrl); | |
2953 | if (ret < 0) | |
2954 | return ret; | |
c5552fde | 2955 | |
49cd84b6 KB |
2956 | ret = nvme_configure_acre(ctrl); |
2957 | if (ret < 0) | |
2958 | return ret; | |
2959 | ||
400b6a7b GR |
2960 | if (!ctrl->identified) |
2961 | nvme_hwmon_init(ctrl); | |
2962 | ||
bd4da3ab | 2963 | ctrl->identified = true; |
c5552fde | 2964 | |
634b8325 KB |
2965 | return 0; |
2966 | ||
2967 | out_free: | |
2968 | kfree(id); | |
07bfcd09 | 2969 | return ret; |
7fd8930f | 2970 | } |
576d55d6 | 2971 | EXPORT_SYMBOL_GPL(nvme_init_identify); |
7fd8930f | 2972 | |
f3ca80fc | 2973 | static int nvme_dev_open(struct inode *inode, struct file *file) |
1673f1f0 | 2974 | { |
a6a5149b CH |
2975 | struct nvme_ctrl *ctrl = |
2976 | container_of(inode->i_cdev, struct nvme_ctrl, cdev); | |
1673f1f0 | 2977 | |
2b1b7e78 JW |
2978 | switch (ctrl->state) { |
2979 | case NVME_CTRL_LIVE: | |
2b1b7e78 JW |
2980 | break; |
2981 | default: | |
a6a5149b | 2982 | return -EWOULDBLOCK; |
2b1b7e78 JW |
2983 | } |
2984 | ||
a6a5149b | 2985 | file->private_data = ctrl; |
f3ca80fc CH |
2986 | return 0; |
2987 | } | |
2988 | ||
bfd89471 CH |
2989 | static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp) |
2990 | { | |
2991 | struct nvme_ns *ns; | |
2992 | int ret; | |
2993 | ||
765cc031 | 2994 | down_read(&ctrl->namespaces_rwsem); |
bfd89471 CH |
2995 | if (list_empty(&ctrl->namespaces)) { |
2996 | ret = -ENOTTY; | |
2997 | goto out_unlock; | |
2998 | } | |
2999 | ||
3000 | ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list); | |
3001 | if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) { | |
1b3c47c1 | 3002 | dev_warn(ctrl->device, |
bfd89471 CH |
3003 | "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n"); |
3004 | ret = -EINVAL; | |
3005 | goto out_unlock; | |
3006 | } | |
3007 | ||
1b3c47c1 | 3008 | dev_warn(ctrl->device, |
bfd89471 CH |
3009 | "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n"); |
3010 | kref_get(&ns->kref); | |
765cc031 | 3011 | up_read(&ctrl->namespaces_rwsem); |
bfd89471 CH |
3012 | |
3013 | ret = nvme_user_cmd(ctrl, ns, argp); | |
3014 | nvme_put_ns(ns); | |
3015 | return ret; | |
3016 | ||
3017 | out_unlock: | |
765cc031 | 3018 | up_read(&ctrl->namespaces_rwsem); |
bfd89471 CH |
3019 | return ret; |
3020 | } | |
3021 | ||
f3ca80fc CH |
3022 | static long nvme_dev_ioctl(struct file *file, unsigned int cmd, |
3023 | unsigned long arg) | |
3024 | { | |
3025 | struct nvme_ctrl *ctrl = file->private_data; | |
3026 | void __user *argp = (void __user *)arg; | |
f3ca80fc CH |
3027 | |
3028 | switch (cmd) { | |
3029 | case NVME_IOCTL_ADMIN_CMD: | |
3030 | return nvme_user_cmd(ctrl, NULL, argp); | |
65e68edc MR |
3031 | case NVME_IOCTL_ADMIN64_CMD: |
3032 | return nvme_user_cmd64(ctrl, NULL, argp); | |
f3ca80fc | 3033 | case NVME_IOCTL_IO_CMD: |
bfd89471 | 3034 | return nvme_dev_user_cmd(ctrl, argp); |
f3ca80fc | 3035 | case NVME_IOCTL_RESET: |
1b3c47c1 | 3036 | dev_warn(ctrl->device, "resetting controller\n"); |
d86c4d8e | 3037 | return nvme_reset_ctrl_sync(ctrl); |
f3ca80fc CH |
3038 | case NVME_IOCTL_SUBSYS_RESET: |
3039 | return nvme_reset_subsystem(ctrl); | |
9ec3bb2f KB |
3040 | case NVME_IOCTL_RESCAN: |
3041 | nvme_queue_scan(ctrl); | |
3042 | return 0; | |
f3ca80fc CH |
3043 | default: |
3044 | return -ENOTTY; | |
3045 | } | |
3046 | } | |
3047 | ||
3048 | static const struct file_operations nvme_dev_fops = { | |
3049 | .owner = THIS_MODULE, | |
3050 | .open = nvme_dev_open, | |
f3ca80fc | 3051 | .unlocked_ioctl = nvme_dev_ioctl, |
1832f2d8 | 3052 | .compat_ioctl = compat_ptr_ioctl, |
f3ca80fc CH |
3053 | }; |
3054 | ||
3055 | static ssize_t nvme_sysfs_reset(struct device *dev, | |
3056 | struct device_attribute *attr, const char *buf, | |
3057 | size_t count) | |
3058 | { | |
3059 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3060 | int ret; | |
3061 | ||
d86c4d8e | 3062 | ret = nvme_reset_ctrl_sync(ctrl); |
f3ca80fc CH |
3063 | if (ret < 0) |
3064 | return ret; | |
3065 | return count; | |
1673f1f0 | 3066 | } |
f3ca80fc | 3067 | static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset); |
1673f1f0 | 3068 | |
9ec3bb2f KB |
3069 | static ssize_t nvme_sysfs_rescan(struct device *dev, |
3070 | struct device_attribute *attr, const char *buf, | |
3071 | size_t count) | |
3072 | { | |
3073 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3074 | ||
3075 | nvme_queue_scan(ctrl); | |
3076 | return count; | |
3077 | } | |
3078 | static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan); | |
3079 | ||
5b85b826 CH |
3080 | static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev) |
3081 | { | |
3082 | struct gendisk *disk = dev_to_disk(dev); | |
3083 | ||
3084 | if (disk->fops == &nvme_fops) | |
3085 | return nvme_get_ns_from_dev(dev)->head; | |
3086 | else | |
3087 | return disk->private_data; | |
3088 | } | |
3089 | ||
118472ab | 3090 | static ssize_t wwid_show(struct device *dev, struct device_attribute *attr, |
5b85b826 | 3091 | char *buf) |
118472ab | 3092 | { |
5b85b826 CH |
3093 | struct nvme_ns_head *head = dev_to_ns_head(dev); |
3094 | struct nvme_ns_ids *ids = &head->ids; | |
3095 | struct nvme_subsystem *subsys = head->subsys; | |
ab9e00cc CH |
3096 | int serial_len = sizeof(subsys->serial); |
3097 | int model_len = sizeof(subsys->model); | |
118472ab | 3098 | |
002fab04 CH |
3099 | if (!uuid_is_null(&ids->uuid)) |
3100 | return sprintf(buf, "uuid.%pU\n", &ids->uuid); | |
6484f5d1 | 3101 | |
002fab04 CH |
3102 | if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) |
3103 | return sprintf(buf, "eui.%16phN\n", ids->nguid); | |
118472ab | 3104 | |
002fab04 CH |
3105 | if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64))) |
3106 | return sprintf(buf, "eui.%8phN\n", ids->eui64); | |
118472ab | 3107 | |
ab9e00cc CH |
3108 | while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' || |
3109 | subsys->serial[serial_len - 1] == '\0')) | |
118472ab | 3110 | serial_len--; |
ab9e00cc CH |
3111 | while (model_len > 0 && (subsys->model[model_len - 1] == ' ' || |
3112 | subsys->model[model_len - 1] == '\0')) | |
118472ab KB |
3113 | model_len--; |
3114 | ||
ab9e00cc CH |
3115 | return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id, |
3116 | serial_len, subsys->serial, model_len, subsys->model, | |
5b85b826 | 3117 | head->ns_id); |
118472ab | 3118 | } |
c828a892 | 3119 | static DEVICE_ATTR_RO(wwid); |
118472ab | 3120 | |
d934f984 | 3121 | static ssize_t nguid_show(struct device *dev, struct device_attribute *attr, |
5b85b826 | 3122 | char *buf) |
d934f984 | 3123 | { |
5b85b826 | 3124 | return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid); |
d934f984 | 3125 | } |
c828a892 | 3126 | static DEVICE_ATTR_RO(nguid); |
d934f984 | 3127 | |
2b9b6e86 | 3128 | static ssize_t uuid_show(struct device *dev, struct device_attribute *attr, |
5b85b826 | 3129 | char *buf) |
2b9b6e86 | 3130 | { |
5b85b826 | 3131 | struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids; |
d934f984 JT |
3132 | |
3133 | /* For backward compatibility expose the NGUID to userspace if | |
3134 | * we have no UUID set | |
3135 | */ | |
002fab04 | 3136 | if (uuid_is_null(&ids->uuid)) { |
d934f984 JT |
3137 | printk_ratelimited(KERN_WARNING |
3138 | "No UUID available providing old NGUID\n"); | |
002fab04 | 3139 | return sprintf(buf, "%pU\n", ids->nguid); |
d934f984 | 3140 | } |
002fab04 | 3141 | return sprintf(buf, "%pU\n", &ids->uuid); |
2b9b6e86 | 3142 | } |
c828a892 | 3143 | static DEVICE_ATTR_RO(uuid); |
2b9b6e86 KB |
3144 | |
3145 | static ssize_t eui_show(struct device *dev, struct device_attribute *attr, | |
5b85b826 | 3146 | char *buf) |
2b9b6e86 | 3147 | { |
5b85b826 | 3148 | return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64); |
2b9b6e86 | 3149 | } |
c828a892 | 3150 | static DEVICE_ATTR_RO(eui); |
2b9b6e86 KB |
3151 | |
3152 | static ssize_t nsid_show(struct device *dev, struct device_attribute *attr, | |
5b85b826 | 3153 | char *buf) |
2b9b6e86 | 3154 | { |
5b85b826 | 3155 | return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id); |
2b9b6e86 | 3156 | } |
c828a892 | 3157 | static DEVICE_ATTR_RO(nsid); |
2b9b6e86 | 3158 | |
5b85b826 | 3159 | static struct attribute *nvme_ns_id_attrs[] = { |
118472ab | 3160 | &dev_attr_wwid.attr, |
2b9b6e86 | 3161 | &dev_attr_uuid.attr, |
d934f984 | 3162 | &dev_attr_nguid.attr, |
2b9b6e86 KB |
3163 | &dev_attr_eui.attr, |
3164 | &dev_attr_nsid.attr, | |
0d0b660f CH |
3165 | #ifdef CONFIG_NVME_MULTIPATH |
3166 | &dev_attr_ana_grpid.attr, | |
3167 | &dev_attr_ana_state.attr, | |
3168 | #endif | |
2b9b6e86 KB |
3169 | NULL, |
3170 | }; | |
3171 | ||
5b85b826 | 3172 | static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj, |
2b9b6e86 KB |
3173 | struct attribute *a, int n) |
3174 | { | |
3175 | struct device *dev = container_of(kobj, struct device, kobj); | |
5b85b826 | 3176 | struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids; |
2b9b6e86 KB |
3177 | |
3178 | if (a == &dev_attr_uuid.attr) { | |
a04b5de5 | 3179 | if (uuid_is_null(&ids->uuid) && |
002fab04 | 3180 | !memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) |
d934f984 JT |
3181 | return 0; |
3182 | } | |
3183 | if (a == &dev_attr_nguid.attr) { | |
002fab04 | 3184 | if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) |
2b9b6e86 KB |
3185 | return 0; |
3186 | } | |
3187 | if (a == &dev_attr_eui.attr) { | |
002fab04 | 3188 | if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64))) |
2b9b6e86 KB |
3189 | return 0; |
3190 | } | |
0d0b660f CH |
3191 | #ifdef CONFIG_NVME_MULTIPATH |
3192 | if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) { | |
3193 | if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */ | |
3194 | return 0; | |
3195 | if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl)) | |
3196 | return 0; | |
3197 | } | |
3198 | #endif | |
2b9b6e86 KB |
3199 | return a->mode; |
3200 | } | |
3201 | ||
eb090c4c | 3202 | static const struct attribute_group nvme_ns_id_attr_group = { |
5b85b826 CH |
3203 | .attrs = nvme_ns_id_attrs, |
3204 | .is_visible = nvme_ns_id_attrs_are_visible, | |
2b9b6e86 KB |
3205 | }; |
3206 | ||
33b14f67 HR |
3207 | const struct attribute_group *nvme_ns_id_attr_groups[] = { |
3208 | &nvme_ns_id_attr_group, | |
3209 | #ifdef CONFIG_NVM | |
3210 | &nvme_nvm_attr_group, | |
3211 | #endif | |
3212 | NULL, | |
3213 | }; | |
3214 | ||
931e1c22 | 3215 | #define nvme_show_str_function(field) \ |
779ff756 KB |
3216 | static ssize_t field##_show(struct device *dev, \ |
3217 | struct device_attribute *attr, char *buf) \ | |
3218 | { \ | |
3219 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ | |
ab9e00cc CH |
3220 | return sprintf(buf, "%.*s\n", \ |
3221 | (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \ | |
779ff756 KB |
3222 | } \ |
3223 | static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); | |
3224 | ||
ab9e00cc CH |
3225 | nvme_show_str_function(model); |
3226 | nvme_show_str_function(serial); | |
3227 | nvme_show_str_function(firmware_rev); | |
3228 | ||
931e1c22 ML |
3229 | #define nvme_show_int_function(field) \ |
3230 | static ssize_t field##_show(struct device *dev, \ | |
3231 | struct device_attribute *attr, char *buf) \ | |
3232 | { \ | |
3233 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ | |
3234 | return sprintf(buf, "%d\n", ctrl->field); \ | |
3235 | } \ | |
3236 | static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); | |
3237 | ||
931e1c22 | 3238 | nvme_show_int_function(cntlid); |
103e515e | 3239 | nvme_show_int_function(numa_node); |
2b1ff255 JS |
3240 | nvme_show_int_function(queue_count); |
3241 | nvme_show_int_function(sqsize); | |
779ff756 | 3242 | |
1a353d85 ML |
3243 | static ssize_t nvme_sysfs_delete(struct device *dev, |
3244 | struct device_attribute *attr, const char *buf, | |
3245 | size_t count) | |
3246 | { | |
3247 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3248 | ||
ce151813 IR |
3249 | /* Can't delete non-created controllers */ |
3250 | if (!ctrl->created) | |
3251 | return -EBUSY; | |
3252 | ||
1a353d85 | 3253 | if (device_remove_file_self(dev, attr)) |
c5017e85 | 3254 | nvme_delete_ctrl_sync(ctrl); |
1a353d85 ML |
3255 | return count; |
3256 | } | |
3257 | static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete); | |
3258 | ||
3259 | static ssize_t nvme_sysfs_show_transport(struct device *dev, | |
3260 | struct device_attribute *attr, | |
3261 | char *buf) | |
3262 | { | |
3263 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3264 | ||
3265 | return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name); | |
3266 | } | |
3267 | static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL); | |
3268 | ||
8432bdb2 SG |
3269 | static ssize_t nvme_sysfs_show_state(struct device *dev, |
3270 | struct device_attribute *attr, | |
3271 | char *buf) | |
3272 | { | |
3273 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3274 | static const char *const state_name[] = { | |
3275 | [NVME_CTRL_NEW] = "new", | |
3276 | [NVME_CTRL_LIVE] = "live", | |
3277 | [NVME_CTRL_RESETTING] = "resetting", | |
ad6a0a52 | 3278 | [NVME_CTRL_CONNECTING] = "connecting", |
8432bdb2 SG |
3279 | [NVME_CTRL_DELETING] = "deleting", |
3280 | [NVME_CTRL_DEAD] = "dead", | |
3281 | }; | |
3282 | ||
3283 | if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) && | |
3284 | state_name[ctrl->state]) | |
3285 | return sprintf(buf, "%s\n", state_name[ctrl->state]); | |
3286 | ||
3287 | return sprintf(buf, "unknown state\n"); | |
3288 | } | |
3289 | ||
3290 | static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL); | |
3291 | ||
1a353d85 ML |
3292 | static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev, |
3293 | struct device_attribute *attr, | |
3294 | char *buf) | |
3295 | { | |
3296 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3297 | ||
ab9e00cc | 3298 | return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn); |
1a353d85 ML |
3299 | } |
3300 | static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL); | |
3301 | ||
76171c6c SG |
3302 | static ssize_t nvme_sysfs_show_hostnqn(struct device *dev, |
3303 | struct device_attribute *attr, | |
3304 | char *buf) | |
3305 | { | |
3306 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3307 | ||
3308 | return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn); | |
3309 | } | |
3310 | static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL); | |
3311 | ||
45fb19f7 SG |
3312 | static ssize_t nvme_sysfs_show_hostid(struct device *dev, |
3313 | struct device_attribute *attr, | |
3314 | char *buf) | |
3315 | { | |
3316 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3317 | ||
3318 | return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id); | |
3319 | } | |
3320 | static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL); | |
3321 | ||
1a353d85 ML |
3322 | static ssize_t nvme_sysfs_show_address(struct device *dev, |
3323 | struct device_attribute *attr, | |
3324 | char *buf) | |
3325 | { | |
3326 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3327 | ||
3328 | return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE); | |
3329 | } | |
3330 | static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL); | |
3331 | ||
779ff756 KB |
3332 | static struct attribute *nvme_dev_attrs[] = { |
3333 | &dev_attr_reset_controller.attr, | |
9ec3bb2f | 3334 | &dev_attr_rescan_controller.attr, |
779ff756 KB |
3335 | &dev_attr_model.attr, |
3336 | &dev_attr_serial.attr, | |
3337 | &dev_attr_firmware_rev.attr, | |
931e1c22 | 3338 | &dev_attr_cntlid.attr, |
1a353d85 ML |
3339 | &dev_attr_delete_controller.attr, |
3340 | &dev_attr_transport.attr, | |
3341 | &dev_attr_subsysnqn.attr, | |
3342 | &dev_attr_address.attr, | |
8432bdb2 | 3343 | &dev_attr_state.attr, |
103e515e | 3344 | &dev_attr_numa_node.attr, |
2b1ff255 JS |
3345 | &dev_attr_queue_count.attr, |
3346 | &dev_attr_sqsize.attr, | |
76171c6c | 3347 | &dev_attr_hostnqn.attr, |
45fb19f7 | 3348 | &dev_attr_hostid.attr, |
779ff756 KB |
3349 | NULL |
3350 | }; | |
3351 | ||
1a353d85 ML |
3352 | static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj, |
3353 | struct attribute *a, int n) | |
3354 | { | |
3355 | struct device *dev = container_of(kobj, struct device, kobj); | |
3356 | struct nvme_ctrl *ctrl = dev_get_drvdata(dev); | |
3357 | ||
49d3d50b CH |
3358 | if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl) |
3359 | return 0; | |
3360 | if (a == &dev_attr_address.attr && !ctrl->ops->get_address) | |
3361 | return 0; | |
76171c6c SG |
3362 | if (a == &dev_attr_hostnqn.attr && !ctrl->opts) |
3363 | return 0; | |
45fb19f7 SG |
3364 | if (a == &dev_attr_hostid.attr && !ctrl->opts) |
3365 | return 0; | |
1a353d85 ML |
3366 | |
3367 | return a->mode; | |
3368 | } | |
3369 | ||
779ff756 | 3370 | static struct attribute_group nvme_dev_attrs_group = { |
1a353d85 ML |
3371 | .attrs = nvme_dev_attrs, |
3372 | .is_visible = nvme_dev_attrs_are_visible, | |
779ff756 KB |
3373 | }; |
3374 | ||
3375 | static const struct attribute_group *nvme_dev_attr_groups[] = { | |
3376 | &nvme_dev_attrs_group, | |
3377 | NULL, | |
3378 | }; | |
3379 | ||
026d2ef7 | 3380 | static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys, |
ed754e5d CH |
3381 | unsigned nsid) |
3382 | { | |
3383 | struct nvme_ns_head *h; | |
3384 | ||
3385 | lockdep_assert_held(&subsys->lock); | |
3386 | ||
3387 | list_for_each_entry(h, &subsys->nsheads, entry) { | |
3388 | if (h->ns_id == nsid && kref_get_unless_zero(&h->ref)) | |
3389 | return h; | |
3390 | } | |
3391 | ||
3392 | return NULL; | |
3393 | } | |
3394 | ||
3395 | static int __nvme_check_ids(struct nvme_subsystem *subsys, | |
3396 | struct nvme_ns_head *new) | |
3397 | { | |
3398 | struct nvme_ns_head *h; | |
3399 | ||
3400 | lockdep_assert_held(&subsys->lock); | |
3401 | ||
3402 | list_for_each_entry(h, &subsys->nsheads, entry) { | |
3403 | if (nvme_ns_ids_valid(&new->ids) && | |
2079699c | 3404 | !list_empty(&h->list) && |
ed754e5d CH |
3405 | nvme_ns_ids_equal(&new->ids, &h->ids)) |
3406 | return -EINVAL; | |
3407 | } | |
3408 | ||
3409 | return 0; | |
3410 | } | |
3411 | ||
3412 | static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl, | |
43fcd9e1 CH |
3413 | unsigned nsid, struct nvme_id_ns *id, |
3414 | struct nvme_ns_ids *ids) | |
ed754e5d CH |
3415 | { |
3416 | struct nvme_ns_head *head; | |
f3334447 | 3417 | size_t size = sizeof(*head); |
ed754e5d CH |
3418 | int ret = -ENOMEM; |
3419 | ||
f3334447 CH |
3420 | #ifdef CONFIG_NVME_MULTIPATH |
3421 | size += num_possible_nodes() * sizeof(struct nvme_ns *); | |
3422 | #endif | |
3423 | ||
3424 | head = kzalloc(size, GFP_KERNEL); | |
ed754e5d CH |
3425 | if (!head) |
3426 | goto out; | |
3427 | ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL); | |
3428 | if (ret < 0) | |
3429 | goto out_free_head; | |
3430 | head->instance = ret; | |
3431 | INIT_LIST_HEAD(&head->list); | |
fd92c77f MG |
3432 | ret = init_srcu_struct(&head->srcu); |
3433 | if (ret) | |
3434 | goto out_ida_remove; | |
ed754e5d CH |
3435 | head->subsys = ctrl->subsys; |
3436 | head->ns_id = nsid; | |
43fcd9e1 | 3437 | head->ids = *ids; |
ed754e5d CH |
3438 | kref_init(&head->ref); |
3439 | ||
ed754e5d CH |
3440 | ret = __nvme_check_ids(ctrl->subsys, head); |
3441 | if (ret) { | |
3442 | dev_err(ctrl->device, | |
3443 | "duplicate IDs for nsid %d\n", nsid); | |
3444 | goto out_cleanup_srcu; | |
3445 | } | |
3446 | ||
32acab31 CH |
3447 | ret = nvme_mpath_alloc_disk(ctrl, head); |
3448 | if (ret) | |
3449 | goto out_cleanup_srcu; | |
3450 | ||
ed754e5d | 3451 | list_add_tail(&head->entry, &ctrl->subsys->nsheads); |
12d9f070 JW |
3452 | |
3453 | kref_get(&ctrl->subsys->ref); | |
3454 | ||
ed754e5d CH |
3455 | return head; |
3456 | out_cleanup_srcu: | |
3457 | cleanup_srcu_struct(&head->srcu); | |
fd92c77f | 3458 | out_ida_remove: |
ed754e5d CH |
3459 | ida_simple_remove(&ctrl->subsys->ns_ida, head->instance); |
3460 | out_free_head: | |
3461 | kfree(head); | |
3462 | out: | |
538af88e SG |
3463 | if (ret > 0) |
3464 | ret = blk_status_to_errno(nvme_error_status(ret)); | |
ed754e5d CH |
3465 | return ERR_PTR(ret); |
3466 | } | |
3467 | ||
3468 | static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid, | |
9bd82b1a | 3469 | struct nvme_id_ns *id) |
ed754e5d CH |
3470 | { |
3471 | struct nvme_ctrl *ctrl = ns->ctrl; | |
3472 | bool is_shared = id->nmic & (1 << 0); | |
3473 | struct nvme_ns_head *head = NULL; | |
43fcd9e1 | 3474 | struct nvme_ns_ids ids; |
ed754e5d CH |
3475 | int ret = 0; |
3476 | ||
43fcd9e1 CH |
3477 | ret = nvme_report_ns_ids(ctrl, nsid, id, &ids); |
3478 | if (ret) | |
3479 | goto out; | |
3480 | ||
ed754e5d CH |
3481 | mutex_lock(&ctrl->subsys->lock); |
3482 | if (is_shared) | |
026d2ef7 | 3483 | head = nvme_find_ns_head(ctrl->subsys, nsid); |
ed754e5d | 3484 | if (!head) { |
43fcd9e1 | 3485 | head = nvme_alloc_ns_head(ctrl, nsid, id, &ids); |
ed754e5d CH |
3486 | if (IS_ERR(head)) { |
3487 | ret = PTR_ERR(head); | |
3488 | goto out_unlock; | |
3489 | } | |
ed754e5d | 3490 | } else { |
ed754e5d CH |
3491 | if (!nvme_ns_ids_equal(&head->ids, &ids)) { |
3492 | dev_err(ctrl->device, | |
3493 | "IDs don't match for shared namespace %d\n", | |
3494 | nsid); | |
3495 | ret = -EINVAL; | |
3496 | goto out_unlock; | |
3497 | } | |
ed754e5d CH |
3498 | } |
3499 | ||
3500 | list_add_tail(&ns->siblings, &head->list); | |
3501 | ns->head = head; | |
3502 | ||
3503 | out_unlock: | |
3504 | mutex_unlock(&ctrl->subsys->lock); | |
43fcd9e1 | 3505 | out: |
538af88e SG |
3506 | if (ret > 0) |
3507 | ret = blk_status_to_errno(nvme_error_status(ret)); | |
ed754e5d CH |
3508 | return ret; |
3509 | } | |
3510 | ||
5bae7f73 CH |
3511 | static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) |
3512 | { | |
3513 | struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); | |
3514 | struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); | |
3515 | ||
ed754e5d | 3516 | return nsa->head->ns_id - nsb->head->ns_id; |
5bae7f73 CH |
3517 | } |
3518 | ||
32f0c4af | 3519 | static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
5bae7f73 | 3520 | { |
32f0c4af | 3521 | struct nvme_ns *ns, *ret = NULL; |
69d3b8ac | 3522 | |
765cc031 | 3523 | down_read(&ctrl->namespaces_rwsem); |
5bae7f73 | 3524 | list_for_each_entry(ns, &ctrl->namespaces, list) { |
ed754e5d | 3525 | if (ns->head->ns_id == nsid) { |
2dd41228 CH |
3526 | if (!kref_get_unless_zero(&ns->kref)) |
3527 | continue; | |
32f0c4af KB |
3528 | ret = ns; |
3529 | break; | |
3530 | } | |
ed754e5d | 3531 | if (ns->head->ns_id > nsid) |
5bae7f73 CH |
3532 | break; |
3533 | } | |
765cc031 | 3534 | up_read(&ctrl->namespaces_rwsem); |
32f0c4af | 3535 | return ret; |
5bae7f73 CH |
3536 | } |
3537 | ||
f5d11840 JA |
3538 | static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns) |
3539 | { | |
3540 | struct streams_directive_params s; | |
3541 | int ret; | |
3542 | ||
3543 | if (!ctrl->nr_streams) | |
3544 | return 0; | |
3545 | ||
ed754e5d | 3546 | ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id); |
f5d11840 JA |
3547 | if (ret) |
3548 | return ret; | |
3549 | ||
3550 | ns->sws = le32_to_cpu(s.sws); | |
3551 | ns->sgs = le16_to_cpu(s.sgs); | |
3552 | ||
3553 | if (ns->sws) { | |
3554 | unsigned int bs = 1 << ns->lba_shift; | |
3555 | ||
3556 | blk_queue_io_min(ns->queue, bs * ns->sws); | |
3557 | if (ns->sgs) | |
3558 | blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs); | |
3559 | } | |
3560 | ||
3561 | return 0; | |
3562 | } | |
3563 | ||
adce7e98 | 3564 | static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
5bae7f73 CH |
3565 | { |
3566 | struct nvme_ns *ns; | |
3567 | struct gendisk *disk; | |
ac81bfa9 MB |
3568 | struct nvme_id_ns *id; |
3569 | char disk_name[DISK_NAME_LEN]; | |
ab4ab09c | 3570 | int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret; |
5bae7f73 CH |
3571 | |
3572 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); | |
3573 | if (!ns) | |
adce7e98 | 3574 | return; |
5bae7f73 CH |
3575 | |
3576 | ns->queue = blk_mq_init_queue(ctrl->tagset); | |
adce7e98 | 3577 | if (IS_ERR(ns->queue)) |
ed754e5d | 3578 | goto out_free_ns; |
e0596ab2 | 3579 | |
7d30c81b | 3580 | if (ctrl->opts && ctrl->opts->data_digest) |
958f2a0f MS |
3581 | ns->queue->backing_dev_info->capabilities |
3582 | |= BDI_CAP_STABLE_WRITES; | |
3583 | ||
8b904b5b | 3584 | blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue); |
e0596ab2 LG |
3585 | if (ctrl->ops->flags & NVME_F_PCI_P2PDMA) |
3586 | blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue); | |
3587 | ||
5bae7f73 CH |
3588 | ns->queue->queuedata = ns; |
3589 | ns->ctrl = ctrl; | |
3590 | ||
5bae7f73 | 3591 | kref_init(&ns->kref); |
5bae7f73 | 3592 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ |
5bae7f73 CH |
3593 | |
3594 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); | |
da35825d | 3595 | nvme_set_queue_limits(ctrl, ns->queue); |
5bae7f73 | 3596 | |
331813f6 SG |
3597 | ret = nvme_identify_ns(ctrl, nsid, &id); |
3598 | if (ret) | |
ac81bfa9 MB |
3599 | goto out_free_queue; |
3600 | ||
adce7e98 | 3601 | if (id->ncap == 0) /* no namespace (legacy quirk) */ |
cdbff4f2 CH |
3602 | goto out_free_id; |
3603 | ||
ab4ab09c HR |
3604 | ret = nvme_init_ns_head(ns, nsid, id); |
3605 | if (ret) | |
ed754e5d | 3606 | goto out_free_id; |
654b4a4a | 3607 | nvme_setup_streams_ns(ctrl, ns); |
a785dbcc | 3608 | nvme_set_disk_name(disk_name, ns, ctrl, &flags); |
cdbff4f2 | 3609 | |
3dc87dd0 | 3610 | disk = alloc_disk_node(0, node); |
adce7e98 | 3611 | if (!disk) |
ed754e5d | 3612 | goto out_unlink_ns; |
ac81bfa9 | 3613 | |
3dc87dd0 MB |
3614 | disk->fops = &nvme_fops; |
3615 | disk->private_data = ns; | |
3616 | disk->queue = ns->queue; | |
32acab31 | 3617 | disk->flags = flags; |
3dc87dd0 MB |
3618 | memcpy(disk->disk_name, disk_name, DISK_NAME_LEN); |
3619 | ns->disk = disk; | |
3620 | ||
3621 | __nvme_revalidate_disk(disk, id); | |
5bae7f73 | 3622 | |
85136c01 | 3623 | if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) { |
ab4ab09c HR |
3624 | ret = nvme_nvm_register(ns, disk_name, node); |
3625 | if (ret) { | |
85136c01 MB |
3626 | dev_warn(ctrl->device, "LightNVM init failure\n"); |
3627 | goto out_put_disk; | |
3628 | } | |
3629 | } | |
3630 | ||
765cc031 | 3631 | down_write(&ctrl->namespaces_rwsem); |
32f0c4af | 3632 | list_add_tail(&ns->list, &ctrl->namespaces); |
765cc031 | 3633 | up_write(&ctrl->namespaces_rwsem); |
32f0c4af | 3634 | |
d22524a4 | 3635 | nvme_get_ctrl(ctrl); |
ac81bfa9 | 3636 | |
33b14f67 | 3637 | device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups); |
32acab31 | 3638 | |
0d0b660f | 3639 | nvme_mpath_add_disk(ns, id); |
a3646451 | 3640 | nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name); |
0d0b660f CH |
3641 | kfree(id); |
3642 | ||
adce7e98 | 3643 | return; |
85136c01 | 3644 | out_put_disk: |
132be623 NC |
3645 | /* prevent double queue cleanup */ |
3646 | ns->disk->queue = NULL; | |
85136c01 | 3647 | put_disk(ns->disk); |
ed754e5d CH |
3648 | out_unlink_ns: |
3649 | mutex_lock(&ctrl->subsys->lock); | |
3650 | list_del_rcu(&ns->siblings); | |
3651 | mutex_unlock(&ctrl->subsys->lock); | |
a63b8370 | 3652 | nvme_put_ns_head(ns->head); |
ac81bfa9 MB |
3653 | out_free_id: |
3654 | kfree(id); | |
5bae7f73 CH |
3655 | out_free_queue: |
3656 | blk_cleanup_queue(ns->queue); | |
3657 | out_free_ns: | |
3658 | kfree(ns); | |
3659 | } | |
3660 | ||
3661 | static void nvme_ns_remove(struct nvme_ns *ns) | |
3662 | { | |
646017a6 KB |
3663 | if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags)) |
3664 | return; | |
69d3b8ac | 3665 | |
a3646451 | 3666 | nvme_fault_inject_fini(&ns->fault_inject); |
2181e455 AE |
3667 | |
3668 | mutex_lock(&ns->ctrl->subsys->lock); | |
3669 | list_del_rcu(&ns->siblings); | |
3670 | mutex_unlock(&ns->ctrl->subsys->lock); | |
3671 | synchronize_rcu(); /* guarantee not available in head->list */ | |
3672 | nvme_mpath_clear_current_path(ns); | |
3673 | synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */ | |
3674 | ||
b0b4e09c | 3675 | if (ns->disk && ns->disk->flags & GENHD_FL_UP) { |
5bae7f73 | 3676 | del_gendisk(ns->disk); |
5bae7f73 | 3677 | blk_cleanup_queue(ns->queue); |
bd9f5d65 ML |
3678 | if (blk_get_integrity(ns->disk)) |
3679 | blk_integrity_unregister(ns->disk); | |
5bae7f73 | 3680 | } |
32f0c4af | 3681 | |
765cc031 | 3682 | down_write(&ns->ctrl->namespaces_rwsem); |
5bae7f73 | 3683 | list_del_init(&ns->list); |
765cc031 | 3684 | up_write(&ns->ctrl->namespaces_rwsem); |
32f0c4af | 3685 | |
479a322f | 3686 | nvme_mpath_check_last_path(ns); |
5bae7f73 CH |
3687 | nvme_put_ns(ns); |
3688 | } | |
3689 | ||
540c801c KB |
3690 | static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid) |
3691 | { | |
3692 | struct nvme_ns *ns; | |
3693 | ||
32f0c4af | 3694 | ns = nvme_find_get_ns(ctrl, nsid); |
540c801c | 3695 | if (ns) { |
b0b4e09c | 3696 | if (ns->disk && revalidate_disk(ns->disk)) |
540c801c | 3697 | nvme_ns_remove(ns); |
32f0c4af | 3698 | nvme_put_ns(ns); |
540c801c KB |
3699 | } else |
3700 | nvme_alloc_ns(ctrl, nsid); | |
3701 | } | |
3702 | ||
47b0e50a SB |
3703 | static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl, |
3704 | unsigned nsid) | |
3705 | { | |
3706 | struct nvme_ns *ns, *next; | |
6f8e0d78 | 3707 | LIST_HEAD(rm_list); |
47b0e50a | 3708 | |
765cc031 | 3709 | down_write(&ctrl->namespaces_rwsem); |
47b0e50a | 3710 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) { |
cf39a6bc | 3711 | if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags)) |
6f8e0d78 | 3712 | list_move_tail(&ns->list, &rm_list); |
47b0e50a | 3713 | } |
765cc031 | 3714 | up_write(&ctrl->namespaces_rwsem); |
6f8e0d78 JW |
3715 | |
3716 | list_for_each_entry_safe(ns, next, &rm_list, list) | |
3717 | nvme_ns_remove(ns); | |
3718 | ||
47b0e50a SB |
3719 | } |
3720 | ||
540c801c KB |
3721 | static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn) |
3722 | { | |
3723 | struct nvme_ns *ns; | |
3724 | __le32 *ns_list; | |
c8e8c77b JL |
3725 | unsigned i, j, nsid, prev = 0; |
3726 | unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024); | |
540c801c KB |
3727 | int ret = 0; |
3728 | ||
42595eb7 | 3729 | ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL); |
540c801c KB |
3730 | if (!ns_list) |
3731 | return -ENOMEM; | |
3732 | ||
3733 | for (i = 0; i < num_lists; i++) { | |
3734 | ret = nvme_identify_ns_list(ctrl, prev, ns_list); | |
3735 | if (ret) | |
47b0e50a | 3736 | goto free; |
540c801c KB |
3737 | |
3738 | for (j = 0; j < min(nn, 1024U); j++) { | |
3739 | nsid = le32_to_cpu(ns_list[j]); | |
3740 | if (!nsid) | |
3741 | goto out; | |
3742 | ||
3743 | nvme_validate_ns(ctrl, nsid); | |
3744 | ||
3745 | while (++prev < nsid) { | |
32f0c4af KB |
3746 | ns = nvme_find_get_ns(ctrl, prev); |
3747 | if (ns) { | |
540c801c | 3748 | nvme_ns_remove(ns); |
32f0c4af KB |
3749 | nvme_put_ns(ns); |
3750 | } | |
540c801c KB |
3751 | } |
3752 | } | |
3753 | nn -= j; | |
3754 | } | |
3755 | out: | |
47b0e50a SB |
3756 | nvme_remove_invalid_namespaces(ctrl, prev); |
3757 | free: | |
540c801c KB |
3758 | kfree(ns_list); |
3759 | return ret; | |
3760 | } | |
3761 | ||
5955be21 | 3762 | static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn) |
5bae7f73 | 3763 | { |
5bae7f73 CH |
3764 | unsigned i; |
3765 | ||
540c801c KB |
3766 | for (i = 1; i <= nn; i++) |
3767 | nvme_validate_ns(ctrl, i); | |
3768 | ||
47b0e50a | 3769 | nvme_remove_invalid_namespaces(ctrl, nn); |
5bae7f73 CH |
3770 | } |
3771 | ||
f493af37 | 3772 | static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl) |
30d90964 CH |
3773 | { |
3774 | size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32); | |
3775 | __le32 *log; | |
f493af37 | 3776 | int error; |
30d90964 CH |
3777 | |
3778 | log = kzalloc(log_size, GFP_KERNEL); | |
3779 | if (!log) | |
f493af37 | 3780 | return; |
30d90964 | 3781 | |
f493af37 CH |
3782 | /* |
3783 | * We need to read the log to clear the AEN, but we don't want to rely | |
3784 | * on it for the changed namespace information as userspace could have | |
3785 | * raced with us in reading the log page, which could cause us to miss | |
3786 | * updates. | |
3787 | */ | |
0e98719b CH |
3788 | error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log, |
3789 | log_size, 0); | |
f493af37 | 3790 | if (error) |
30d90964 CH |
3791 | dev_warn(ctrl->device, |
3792 | "reading changed ns log failed: %d\n", error); | |
30d90964 | 3793 | |
30d90964 | 3794 | kfree(log); |
30d90964 CH |
3795 | } |
3796 | ||
5955be21 | 3797 | static void nvme_scan_work(struct work_struct *work) |
5bae7f73 | 3798 | { |
5955be21 CH |
3799 | struct nvme_ctrl *ctrl = |
3800 | container_of(work, struct nvme_ctrl, scan_work); | |
5bae7f73 | 3801 | struct nvme_id_ctrl *id; |
540c801c | 3802 | unsigned nn; |
5bae7f73 | 3803 | |
5d02a5c1 KB |
3804 | /* No tagset on a live ctrl means IO queues could not created */ |
3805 | if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset) | |
5955be21 CH |
3806 | return; |
3807 | ||
77016199 | 3808 | if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) { |
30d90964 | 3809 | dev_info(ctrl->device, "rescanning namespaces.\n"); |
f493af37 | 3810 | nvme_clear_changed_ns_log(ctrl); |
30d90964 CH |
3811 | } |
3812 | ||
5bae7f73 CH |
3813 | if (nvme_identify_ctrl(ctrl, &id)) |
3814 | return; | |
540c801c | 3815 | |
e7ad43c3 | 3816 | mutex_lock(&ctrl->scan_lock); |
540c801c | 3817 | nn = le32_to_cpu(id->nn); |
8ef2074d | 3818 | if (ctrl->vs >= NVME_VS(1, 1, 0) && |
540c801c KB |
3819 | !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) { |
3820 | if (!nvme_scan_ns_list(ctrl, nn)) | |
30d90964 | 3821 | goto out_free_id; |
540c801c | 3822 | } |
5955be21 | 3823 | nvme_scan_ns_sequential(ctrl, nn); |
30d90964 | 3824 | out_free_id: |
e7ad43c3 | 3825 | mutex_unlock(&ctrl->scan_lock); |
30d90964 | 3826 | kfree(id); |
765cc031 | 3827 | down_write(&ctrl->namespaces_rwsem); |
540c801c | 3828 | list_sort(NULL, &ctrl->namespaces, ns_cmp); |
765cc031 | 3829 | up_write(&ctrl->namespaces_rwsem); |
5955be21 | 3830 | } |
5bae7f73 | 3831 | |
32f0c4af KB |
3832 | /* |
3833 | * This function iterates the namespace list unlocked to allow recovery from | |
3834 | * controller failure. It is up to the caller to ensure the namespace list is | |
3835 | * not modified by scan work while this function is executing. | |
3836 | */ | |
5bae7f73 CH |
3837 | void nvme_remove_namespaces(struct nvme_ctrl *ctrl) |
3838 | { | |
3839 | struct nvme_ns *ns, *next; | |
6f8e0d78 | 3840 | LIST_HEAD(ns_list); |
5bae7f73 | 3841 | |
0157ec8d SG |
3842 | /* |
3843 | * make sure to requeue I/O to all namespaces as these | |
3844 | * might result from the scan itself and must complete | |
3845 | * for the scan_work to make progress | |
3846 | */ | |
3847 | nvme_mpath_clear_ctrl_paths(ctrl); | |
3848 | ||
f6c8e432 SG |
3849 | /* prevent racing with ns scanning */ |
3850 | flush_work(&ctrl->scan_work); | |
3851 | ||
0ff9d4e1 KB |
3852 | /* |
3853 | * The dead states indicates the controller was not gracefully | |
3854 | * disconnected. In that case, we won't be able to flush any data while | |
3855 | * removing the namespaces' disks; fail all the queues now to avoid | |
3856 | * potentially having to clean up the failed sync later. | |
3857 | */ | |
3858 | if (ctrl->state == NVME_CTRL_DEAD) | |
3859 | nvme_kill_queues(ctrl); | |
3860 | ||
765cc031 | 3861 | down_write(&ctrl->namespaces_rwsem); |
6f8e0d78 | 3862 | list_splice_init(&ctrl->namespaces, &ns_list); |
765cc031 | 3863 | up_write(&ctrl->namespaces_rwsem); |
6f8e0d78 JW |
3864 | |
3865 | list_for_each_entry_safe(ns, next, &ns_list, list) | |
5bae7f73 CH |
3866 | nvme_ns_remove(ns); |
3867 | } | |
576d55d6 | 3868 | EXPORT_SYMBOL_GPL(nvme_remove_namespaces); |
5bae7f73 | 3869 | |
a42f42e5 SG |
3870 | static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env) |
3871 | { | |
3872 | struct nvme_ctrl *ctrl = | |
3873 | container_of(dev, struct nvme_ctrl, ctrl_device); | |
3874 | struct nvmf_ctrl_options *opts = ctrl->opts; | |
3875 | int ret; | |
3876 | ||
3877 | ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name); | |
3878 | if (ret) | |
3879 | return ret; | |
3880 | ||
3881 | if (opts) { | |
3882 | ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr); | |
3883 | if (ret) | |
3884 | return ret; | |
3885 | ||
3886 | ret = add_uevent_var(env, "NVME_TRSVCID=%s", | |
3887 | opts->trsvcid ?: "none"); | |
3888 | if (ret) | |
3889 | return ret; | |
3890 | ||
3891 | ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s", | |
3892 | opts->host_traddr ?: "none"); | |
3893 | } | |
3894 | return ret; | |
3895 | } | |
3896 | ||
e3d7874d KB |
3897 | static void nvme_aen_uevent(struct nvme_ctrl *ctrl) |
3898 | { | |
3899 | char *envp[2] = { NULL, NULL }; | |
3900 | u32 aen_result = ctrl->aen_result; | |
3901 | ||
3902 | ctrl->aen_result = 0; | |
3903 | if (!aen_result) | |
3904 | return; | |
3905 | ||
3906 | envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result); | |
3907 | if (!envp[0]) | |
3908 | return; | |
3909 | kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp); | |
3910 | kfree(envp[0]); | |
3911 | } | |
3912 | ||
f866fc42 CH |
3913 | static void nvme_async_event_work(struct work_struct *work) |
3914 | { | |
3915 | struct nvme_ctrl *ctrl = | |
3916 | container_of(work, struct nvme_ctrl, async_event_work); | |
3917 | ||
e3d7874d | 3918 | nvme_aen_uevent(ctrl); |
ad22c355 | 3919 | ctrl->ops->submit_async_event(ctrl); |
f866fc42 CH |
3920 | } |
3921 | ||
b6dccf7f AD |
3922 | static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl) |
3923 | { | |
3924 | ||
3925 | u32 csts; | |
3926 | ||
3927 | if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) | |
3928 | return false; | |
3929 | ||
3930 | if (csts == ~0) | |
3931 | return false; | |
3932 | ||
3933 | return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP)); | |
3934 | } | |
3935 | ||
3936 | static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl) | |
3937 | { | |
b6dccf7f AD |
3938 | struct nvme_fw_slot_info_log *log; |
3939 | ||
3940 | log = kmalloc(sizeof(*log), GFP_KERNEL); | |
3941 | if (!log) | |
3942 | return; | |
3943 | ||
f25372ff | 3944 | if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, log, |
0e98719b CH |
3945 | sizeof(*log), 0)) |
3946 | dev_warn(ctrl->device, "Get FW SLOT INFO log error\n"); | |
b6dccf7f AD |
3947 | kfree(log); |
3948 | } | |
3949 | ||
3950 | static void nvme_fw_act_work(struct work_struct *work) | |
3951 | { | |
3952 | struct nvme_ctrl *ctrl = container_of(work, | |
3953 | struct nvme_ctrl, fw_act_work); | |
3954 | unsigned long fw_act_timeout; | |
3955 | ||
3956 | if (ctrl->mtfa) | |
3957 | fw_act_timeout = jiffies + | |
3958 | msecs_to_jiffies(ctrl->mtfa * 100); | |
3959 | else | |
3960 | fw_act_timeout = jiffies + | |
3961 | msecs_to_jiffies(admin_timeout * 1000); | |
3962 | ||
3963 | nvme_stop_queues(ctrl); | |
3964 | while (nvme_ctrl_pp_status(ctrl)) { | |
3965 | if (time_after(jiffies, fw_act_timeout)) { | |
3966 | dev_warn(ctrl->device, | |
3967 | "Fw activation timeout, reset controller\n"); | |
4c75f877 KB |
3968 | nvme_try_sched_reset(ctrl); |
3969 | return; | |
b6dccf7f AD |
3970 | } |
3971 | msleep(100); | |
3972 | } | |
3973 | ||
4c75f877 | 3974 | if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE)) |
b6dccf7f AD |
3975 | return; |
3976 | ||
3977 | nvme_start_queues(ctrl); | |
a806c6c8 | 3978 | /* read FW slot information to clear the AER */ |
b6dccf7f AD |
3979 | nvme_get_fw_slot_info(ctrl); |
3980 | } | |
3981 | ||
868c2392 CH |
3982 | static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result) |
3983 | { | |
09bd1ff4 CK |
3984 | u32 aer_notice_type = (result & 0xff00) >> 8; |
3985 | ||
521cfb8e CK |
3986 | trace_nvme_async_event(ctrl, aer_notice_type); |
3987 | ||
09bd1ff4 | 3988 | switch (aer_notice_type) { |
868c2392 | 3989 | case NVME_AER_NOTICE_NS_CHANGED: |
77016199 | 3990 | set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events); |
868c2392 CH |
3991 | nvme_queue_scan(ctrl); |
3992 | break; | |
3993 | case NVME_AER_NOTICE_FW_ACT_STARTING: | |
4c75f877 KB |
3994 | /* |
3995 | * We are (ab)using the RESETTING state to prevent subsequent | |
3996 | * recovery actions from interfering with the controller's | |
3997 | * firmware activation. | |
3998 | */ | |
3999 | if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) | |
4000 | queue_work(nvme_wq, &ctrl->fw_act_work); | |
868c2392 | 4001 | break; |
0d0b660f CH |
4002 | #ifdef CONFIG_NVME_MULTIPATH |
4003 | case NVME_AER_NOTICE_ANA: | |
4004 | if (!ctrl->ana_log_buf) | |
4005 | break; | |
4006 | queue_work(nvme_wq, &ctrl->ana_work); | |
4007 | break; | |
4008 | #endif | |
85f8a435 SG |
4009 | case NVME_AER_NOTICE_DISC_CHANGED: |
4010 | ctrl->aen_result = result; | |
4011 | break; | |
868c2392 CH |
4012 | default: |
4013 | dev_warn(ctrl->device, "async event result %08x\n", result); | |
4014 | } | |
4015 | } | |
4016 | ||
7bf58533 | 4017 | void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, |
287a63eb | 4018 | volatile union nvme_result *res) |
f866fc42 | 4019 | { |
7bf58533 | 4020 | u32 result = le32_to_cpu(res->u32); |
09bd1ff4 | 4021 | u32 aer_type = result & 0x07; |
f866fc42 | 4022 | |
ad22c355 | 4023 | if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS) |
f866fc42 CH |
4024 | return; |
4025 | ||
09bd1ff4 | 4026 | switch (aer_type) { |
868c2392 CH |
4027 | case NVME_AER_NOTICE: |
4028 | nvme_handle_aen_notice(ctrl, result); | |
4029 | break; | |
e3d7874d KB |
4030 | case NVME_AER_ERROR: |
4031 | case NVME_AER_SMART: | |
4032 | case NVME_AER_CSS: | |
4033 | case NVME_AER_VS: | |
09bd1ff4 | 4034 | trace_nvme_async_event(ctrl, aer_type); |
e3d7874d | 4035 | ctrl->aen_result = result; |
7bf58533 CH |
4036 | break; |
4037 | default: | |
4038 | break; | |
f866fc42 | 4039 | } |
c669ccdc | 4040 | queue_work(nvme_wq, &ctrl->async_event_work); |
f866fc42 | 4041 | } |
f866fc42 | 4042 | EXPORT_SYMBOL_GPL(nvme_complete_async_event); |
f3ca80fc | 4043 | |
d09f2b45 | 4044 | void nvme_stop_ctrl(struct nvme_ctrl *ctrl) |
576d55d6 | 4045 | { |
0d0b660f | 4046 | nvme_mpath_stop(ctrl); |
d09f2b45 | 4047 | nvme_stop_keep_alive(ctrl); |
f866fc42 | 4048 | flush_work(&ctrl->async_event_work); |
b6dccf7f | 4049 | cancel_work_sync(&ctrl->fw_act_work); |
d09f2b45 SG |
4050 | } |
4051 | EXPORT_SYMBOL_GPL(nvme_stop_ctrl); | |
4052 | ||
4053 | void nvme_start_ctrl(struct nvme_ctrl *ctrl) | |
4054 | { | |
4055 | if (ctrl->kato) | |
4056 | nvme_start_keep_alive(ctrl); | |
4057 | ||
93da4023 SG |
4058 | nvme_enable_aen(ctrl); |
4059 | ||
d09f2b45 SG |
4060 | if (ctrl->queue_count > 1) { |
4061 | nvme_queue_scan(ctrl); | |
d09f2b45 SG |
4062 | nvme_start_queues(ctrl); |
4063 | } | |
ce151813 | 4064 | ctrl->created = true; |
d09f2b45 SG |
4065 | } |
4066 | EXPORT_SYMBOL_GPL(nvme_start_ctrl); | |
5955be21 | 4067 | |
d09f2b45 SG |
4068 | void nvme_uninit_ctrl(struct nvme_ctrl *ctrl) |
4069 | { | |
f79d5fda | 4070 | nvme_fault_inject_fini(&ctrl->fault_inject); |
510a405d | 4071 | dev_pm_qos_hide_latency_tolerance(ctrl->device); |
a6a5149b | 4072 | cdev_device_del(&ctrl->cdev, ctrl->device); |
726612b6 | 4073 | nvme_put_ctrl(ctrl); |
53029b04 | 4074 | } |
576d55d6 | 4075 | EXPORT_SYMBOL_GPL(nvme_uninit_ctrl); |
53029b04 | 4076 | |
d22524a4 | 4077 | static void nvme_free_ctrl(struct device *dev) |
53029b04 | 4078 | { |
d22524a4 CH |
4079 | struct nvme_ctrl *ctrl = |
4080 | container_of(dev, struct nvme_ctrl, ctrl_device); | |
ab9e00cc | 4081 | struct nvme_subsystem *subsys = ctrl->subsys; |
f3ca80fc | 4082 | |
733e4b69 KB |
4083 | if (subsys && ctrl->instance != subsys->instance) |
4084 | ida_simple_remove(&nvme_instance_ida, ctrl->instance); | |
4085 | ||
84fef62d | 4086 | kfree(ctrl->effects); |
0d0b660f | 4087 | nvme_mpath_uninit(ctrl); |
092ff052 | 4088 | __free_page(ctrl->discard_page); |
f3ca80fc | 4089 | |
ab9e00cc | 4090 | if (subsys) { |
32fd90c4 | 4091 | mutex_lock(&nvme_subsystems_lock); |
ab9e00cc | 4092 | list_del(&ctrl->subsys_entry); |
ab9e00cc | 4093 | sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device)); |
32fd90c4 | 4094 | mutex_unlock(&nvme_subsystems_lock); |
ab9e00cc | 4095 | } |
f3ca80fc CH |
4096 | |
4097 | ctrl->ops->free_ctrl(ctrl); | |
f3ca80fc | 4098 | |
ab9e00cc CH |
4099 | if (subsys) |
4100 | nvme_put_subsystem(subsys); | |
f3ca80fc CH |
4101 | } |
4102 | ||
4103 | /* | |
4104 | * Initialize a NVMe controller structures. This needs to be called during | |
4105 | * earliest initialization so that we have the initialized structured around | |
4106 | * during probing. | |
4107 | */ | |
4108 | int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, | |
4109 | const struct nvme_ctrl_ops *ops, unsigned long quirks) | |
4110 | { | |
4111 | int ret; | |
4112 | ||
bb8d261e CH |
4113 | ctrl->state = NVME_CTRL_NEW; |
4114 | spin_lock_init(&ctrl->lock); | |
e7ad43c3 | 4115 | mutex_init(&ctrl->scan_lock); |
f3ca80fc | 4116 | INIT_LIST_HEAD(&ctrl->namespaces); |
765cc031 | 4117 | init_rwsem(&ctrl->namespaces_rwsem); |
f3ca80fc CH |
4118 | ctrl->dev = dev; |
4119 | ctrl->ops = ops; | |
4120 | ctrl->quirks = quirks; | |
5955be21 | 4121 | INIT_WORK(&ctrl->scan_work, nvme_scan_work); |
f866fc42 | 4122 | INIT_WORK(&ctrl->async_event_work, nvme_async_event_work); |
b6dccf7f | 4123 | INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work); |
c5017e85 | 4124 | INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work); |
c1ac9a4b | 4125 | init_waitqueue_head(&ctrl->state_wq); |
f3ca80fc | 4126 | |
230f1f9e JS |
4127 | INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work); |
4128 | memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd)); | |
4129 | ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive; | |
4130 | ||
cb5b7262 JA |
4131 | BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) > |
4132 | PAGE_SIZE); | |
4133 | ctrl->discard_page = alloc_page(GFP_KERNEL); | |
4134 | if (!ctrl->discard_page) { | |
4135 | ret = -ENOMEM; | |
4136 | goto out; | |
4137 | } | |
4138 | ||
9843f685 CH |
4139 | ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL); |
4140 | if (ret < 0) | |
f3ca80fc | 4141 | goto out; |
9843f685 | 4142 | ctrl->instance = ret; |
f3ca80fc | 4143 | |
d22524a4 CH |
4144 | device_initialize(&ctrl->ctrl_device); |
4145 | ctrl->device = &ctrl->ctrl_device; | |
a6a5149b | 4146 | ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance); |
d22524a4 CH |
4147 | ctrl->device->class = nvme_class; |
4148 | ctrl->device->parent = ctrl->dev; | |
4149 | ctrl->device->groups = nvme_dev_attr_groups; | |
4150 | ctrl->device->release = nvme_free_ctrl; | |
4151 | dev_set_drvdata(ctrl->device, ctrl); | |
4152 | ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance); | |
4153 | if (ret) | |
f3ca80fc | 4154 | goto out_release_instance; |
f3ca80fc | 4155 | |
b780d741 | 4156 | nvme_get_ctrl(ctrl); |
a6a5149b CH |
4157 | cdev_init(&ctrl->cdev, &nvme_dev_fops); |
4158 | ctrl->cdev.owner = ops->module; | |
4159 | ret = cdev_device_add(&ctrl->cdev, ctrl->device); | |
d22524a4 CH |
4160 | if (ret) |
4161 | goto out_free_name; | |
f3ca80fc | 4162 | |
c5552fde AL |
4163 | /* |
4164 | * Initialize latency tolerance controls. The sysfs files won't | |
4165 | * be visible to userspace unless the device actually supports APST. | |
4166 | */ | |
4167 | ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance; | |
4168 | dev_pm_qos_update_user_latency_tolerance(ctrl->device, | |
4169 | min(default_ps_max_latency_us, (unsigned long)S32_MAX)); | |
4170 | ||
f79d5fda AM |
4171 | nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device)); |
4172 | ||
f3ca80fc | 4173 | return 0; |
d22524a4 | 4174 | out_free_name: |
b780d741 | 4175 | nvme_put_ctrl(ctrl); |
d6a2b953 | 4176 | kfree_const(ctrl->device->kobj.name); |
f3ca80fc | 4177 | out_release_instance: |
9843f685 | 4178 | ida_simple_remove(&nvme_instance_ida, ctrl->instance); |
f3ca80fc | 4179 | out: |
cb5b7262 JA |
4180 | if (ctrl->discard_page) |
4181 | __free_page(ctrl->discard_page); | |
f3ca80fc CH |
4182 | return ret; |
4183 | } | |
576d55d6 | 4184 | EXPORT_SYMBOL_GPL(nvme_init_ctrl); |
f3ca80fc | 4185 | |
69d9a99c KB |
4186 | /** |
4187 | * nvme_kill_queues(): Ends all namespace queues | |
4188 | * @ctrl: the dead controller that needs to end | |
4189 | * | |
4190 | * Call this function when the driver determines it is unable to get the | |
4191 | * controller in a state capable of servicing IO. | |
4192 | */ | |
4193 | void nvme_kill_queues(struct nvme_ctrl *ctrl) | |
4194 | { | |
4195 | struct nvme_ns *ns; | |
4196 | ||
765cc031 | 4197 | down_read(&ctrl->namespaces_rwsem); |
82654b6b | 4198 | |
443bd90f | 4199 | /* Forcibly unquiesce queues to avoid blocking dispatch */ |
751a0cc0 | 4200 | if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q)) |
7dd1ab16 | 4201 | blk_mq_unquiesce_queue(ctrl->admin_q); |
443bd90f | 4202 | |
cf39a6bc SB |
4203 | list_for_each_entry(ns, &ctrl->namespaces, list) |
4204 | nvme_set_queue_dying(ns); | |
806f026f | 4205 | |
765cc031 | 4206 | up_read(&ctrl->namespaces_rwsem); |
69d9a99c | 4207 | } |
237045fc | 4208 | EXPORT_SYMBOL_GPL(nvme_kill_queues); |
69d9a99c | 4209 | |
302ad8cc KB |
4210 | void nvme_unfreeze(struct nvme_ctrl *ctrl) |
4211 | { | |
4212 | struct nvme_ns *ns; | |
4213 | ||
765cc031 | 4214 | down_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4215 | list_for_each_entry(ns, &ctrl->namespaces, list) |
4216 | blk_mq_unfreeze_queue(ns->queue); | |
765cc031 | 4217 | up_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4218 | } |
4219 | EXPORT_SYMBOL_GPL(nvme_unfreeze); | |
4220 | ||
4221 | void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout) | |
4222 | { | |
4223 | struct nvme_ns *ns; | |
4224 | ||
765cc031 | 4225 | down_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4226 | list_for_each_entry(ns, &ctrl->namespaces, list) { |
4227 | timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout); | |
4228 | if (timeout <= 0) | |
4229 | break; | |
4230 | } | |
765cc031 | 4231 | up_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4232 | } |
4233 | EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout); | |
4234 | ||
4235 | void nvme_wait_freeze(struct nvme_ctrl *ctrl) | |
4236 | { | |
4237 | struct nvme_ns *ns; | |
4238 | ||
765cc031 | 4239 | down_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4240 | list_for_each_entry(ns, &ctrl->namespaces, list) |
4241 | blk_mq_freeze_queue_wait(ns->queue); | |
765cc031 | 4242 | up_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4243 | } |
4244 | EXPORT_SYMBOL_GPL(nvme_wait_freeze); | |
4245 | ||
4246 | void nvme_start_freeze(struct nvme_ctrl *ctrl) | |
4247 | { | |
4248 | struct nvme_ns *ns; | |
4249 | ||
765cc031 | 4250 | down_read(&ctrl->namespaces_rwsem); |
302ad8cc | 4251 | list_for_each_entry(ns, &ctrl->namespaces, list) |
1671d522 | 4252 | blk_freeze_queue_start(ns->queue); |
765cc031 | 4253 | up_read(&ctrl->namespaces_rwsem); |
302ad8cc KB |
4254 | } |
4255 | EXPORT_SYMBOL_GPL(nvme_start_freeze); | |
4256 | ||
25646264 | 4257 | void nvme_stop_queues(struct nvme_ctrl *ctrl) |
363c9aac SG |
4258 | { |
4259 | struct nvme_ns *ns; | |
4260 | ||
765cc031 | 4261 | down_read(&ctrl->namespaces_rwsem); |
a6eaa884 | 4262 | list_for_each_entry(ns, &ctrl->namespaces, list) |
3174dd33 | 4263 | blk_mq_quiesce_queue(ns->queue); |
765cc031 | 4264 | up_read(&ctrl->namespaces_rwsem); |
363c9aac | 4265 | } |
576d55d6 | 4266 | EXPORT_SYMBOL_GPL(nvme_stop_queues); |
363c9aac | 4267 | |
25646264 | 4268 | void nvme_start_queues(struct nvme_ctrl *ctrl) |
363c9aac SG |
4269 | { |
4270 | struct nvme_ns *ns; | |
4271 | ||
765cc031 | 4272 | down_read(&ctrl->namespaces_rwsem); |
8d7b8faf | 4273 | list_for_each_entry(ns, &ctrl->namespaces, list) |
f660174e | 4274 | blk_mq_unquiesce_queue(ns->queue); |
765cc031 | 4275 | up_read(&ctrl->namespaces_rwsem); |
363c9aac | 4276 | } |
576d55d6 | 4277 | EXPORT_SYMBOL_GPL(nvme_start_queues); |
363c9aac | 4278 | |
d6135c3a KB |
4279 | |
4280 | void nvme_sync_queues(struct nvme_ctrl *ctrl) | |
4281 | { | |
4282 | struct nvme_ns *ns; | |
4283 | ||
4284 | down_read(&ctrl->namespaces_rwsem); | |
4285 | list_for_each_entry(ns, &ctrl->namespaces, list) | |
4286 | blk_sync_queue(ns->queue); | |
4287 | up_read(&ctrl->namespaces_rwsem); | |
03894b7a EN |
4288 | |
4289 | if (ctrl->admin_q) | |
4290 | blk_sync_queue(ctrl->admin_q); | |
d6135c3a KB |
4291 | } |
4292 | EXPORT_SYMBOL_GPL(nvme_sync_queues); | |
4293 | ||
81101540 CH |
4294 | /* |
4295 | * Check we didn't inadvertently grow the command structure sizes: | |
4296 | */ | |
4297 | static inline void _nvme_check_size(void) | |
4298 | { | |
4299 | BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64); | |
4300 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
4301 | BUILD_BUG_ON(sizeof(struct nvme_identify) != 64); | |
4302 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
4303 | BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64); | |
4304 | BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); | |
4305 | BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64); | |
4306 | BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64); | |
4307 | BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); | |
4308 | BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64); | |
4309 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); | |
4310 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE); | |
4311 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE); | |
4312 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
4313 | BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); | |
4314 | BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64); | |
4315 | BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64); | |
4316 | } | |
4317 | ||
4318 | ||
893a74b7 | 4319 | static int __init nvme_core_init(void) |
5bae7f73 | 4320 | { |
b227c59b | 4321 | int result = -ENOMEM; |
5bae7f73 | 4322 | |
81101540 CH |
4323 | _nvme_check_size(); |
4324 | ||
9a6327d2 SG |
4325 | nvme_wq = alloc_workqueue("nvme-wq", |
4326 | WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); | |
4327 | if (!nvme_wq) | |
b227c59b RS |
4328 | goto out; |
4329 | ||
4330 | nvme_reset_wq = alloc_workqueue("nvme-reset-wq", | |
4331 | WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); | |
4332 | if (!nvme_reset_wq) | |
4333 | goto destroy_wq; | |
4334 | ||
4335 | nvme_delete_wq = alloc_workqueue("nvme-delete-wq", | |
4336 | WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); | |
4337 | if (!nvme_delete_wq) | |
4338 | goto destroy_reset_wq; | |
9a6327d2 | 4339 | |
a6a5149b | 4340 | result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme"); |
f3ca80fc | 4341 | if (result < 0) |
b227c59b | 4342 | goto destroy_delete_wq; |
f3ca80fc CH |
4343 | |
4344 | nvme_class = class_create(THIS_MODULE, "nvme"); | |
4345 | if (IS_ERR(nvme_class)) { | |
4346 | result = PTR_ERR(nvme_class); | |
4347 | goto unregister_chrdev; | |
4348 | } | |
a42f42e5 | 4349 | nvme_class->dev_uevent = nvme_class_uevent; |
f3ca80fc | 4350 | |
ab9e00cc CH |
4351 | nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem"); |
4352 | if (IS_ERR(nvme_subsys_class)) { | |
4353 | result = PTR_ERR(nvme_subsys_class); | |
4354 | goto destroy_class; | |
4355 | } | |
5bae7f73 | 4356 | return 0; |
f3ca80fc | 4357 | |
ab9e00cc CH |
4358 | destroy_class: |
4359 | class_destroy(nvme_class); | |
9a6327d2 | 4360 | unregister_chrdev: |
a6a5149b | 4361 | unregister_chrdev_region(nvme_chr_devt, NVME_MINORS); |
b227c59b RS |
4362 | destroy_delete_wq: |
4363 | destroy_workqueue(nvme_delete_wq); | |
4364 | destroy_reset_wq: | |
4365 | destroy_workqueue(nvme_reset_wq); | |
9a6327d2 SG |
4366 | destroy_wq: |
4367 | destroy_workqueue(nvme_wq); | |
b227c59b | 4368 | out: |
f3ca80fc | 4369 | return result; |
5bae7f73 CH |
4370 | } |
4371 | ||
893a74b7 | 4372 | static void __exit nvme_core_exit(void) |
5bae7f73 | 4373 | { |
ab9e00cc | 4374 | class_destroy(nvme_subsys_class); |
f3ca80fc | 4375 | class_destroy(nvme_class); |
a6a5149b | 4376 | unregister_chrdev_region(nvme_chr_devt, NVME_MINORS); |
b227c59b RS |
4377 | destroy_workqueue(nvme_delete_wq); |
4378 | destroy_workqueue(nvme_reset_wq); | |
9a6327d2 | 4379 | destroy_workqueue(nvme_wq); |
f41cfd5d | 4380 | ida_destroy(&nvme_instance_ida); |
5bae7f73 | 4381 | } |
576d55d6 ML |
4382 | |
4383 | MODULE_LICENSE("GPL"); | |
4384 | MODULE_VERSION("1.0"); | |
4385 | module_init(nvme_core_init); | |
4386 | module_exit(nvme_core_exit); |