Commit | Line | Data |
---|---|---|
b60503ba MW |
1 | /* |
2 | * NVM Express device driver | |
6eb0d698 | 3 | * Copyright (c) 2011-2014, Intel Corporation. |
b60503ba MW |
4 | * |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
b60503ba MW |
13 | */ |
14 | ||
15 | #include <linux/nvme.h> | |
8de05535 | 16 | #include <linux/bitops.h> |
b60503ba | 17 | #include <linux/blkdev.h> |
a4aea562 | 18 | #include <linux/blk-mq.h> |
42f61420 | 19 | #include <linux/cpu.h> |
fd63e9ce | 20 | #include <linux/delay.h> |
b60503ba MW |
21 | #include <linux/errno.h> |
22 | #include <linux/fs.h> | |
23 | #include <linux/genhd.h> | |
4cc09e2d | 24 | #include <linux/hdreg.h> |
5aff9382 | 25 | #include <linux/idr.h> |
b60503ba MW |
26 | #include <linux/init.h> |
27 | #include <linux/interrupt.h> | |
28 | #include <linux/io.h> | |
29 | #include <linux/kdev_t.h> | |
1fa6aead | 30 | #include <linux/kthread.h> |
b60503ba | 31 | #include <linux/kernel.h> |
a5768aa8 | 32 | #include <linux/list_sort.h> |
b60503ba MW |
33 | #include <linux/mm.h> |
34 | #include <linux/module.h> | |
35 | #include <linux/moduleparam.h> | |
36 | #include <linux/pci.h> | |
be7b6275 | 37 | #include <linux/poison.h> |
c3bfe717 | 38 | #include <linux/ptrace.h> |
b60503ba MW |
39 | #include <linux/sched.h> |
40 | #include <linux/slab.h> | |
e1e5e564 | 41 | #include <linux/t10-pi.h> |
b60503ba | 42 | #include <linux/types.h> |
5d0f6131 | 43 | #include <scsi/sg.h> |
797a796a HM |
44 | #include <asm-generic/io-64-nonatomic-lo-hi.h> |
45 | ||
b3fffdef | 46 | #define NVME_MINORS (1U << MINORBITS) |
9d43cf64 | 47 | #define NVME_Q_DEPTH 1024 |
d31af0a3 | 48 | #define NVME_AQ_DEPTH 256 |
b60503ba MW |
49 | #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) |
50 | #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) | |
9d43cf64 | 51 | #define ADMIN_TIMEOUT (admin_timeout * HZ) |
2484f407 | 52 | #define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ) |
9d43cf64 KB |
53 | |
54 | static unsigned char admin_timeout = 60; | |
55 | module_param(admin_timeout, byte, 0644); | |
56 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); | |
b60503ba | 57 | |
bd67608a MW |
58 | unsigned char nvme_io_timeout = 30; |
59 | module_param_named(io_timeout, nvme_io_timeout, byte, 0644); | |
b355084a | 60 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); |
b60503ba | 61 | |
2484f407 DM |
62 | static unsigned char shutdown_timeout = 5; |
63 | module_param(shutdown_timeout, byte, 0644); | |
64 | MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); | |
65 | ||
b60503ba MW |
66 | static int nvme_major; |
67 | module_param(nvme_major, int, 0); | |
68 | ||
b3fffdef KB |
69 | static int nvme_char_major; |
70 | module_param(nvme_char_major, int, 0); | |
71 | ||
58ffacb5 MW |
72 | static int use_threaded_interrupts; |
73 | module_param(use_threaded_interrupts, int, 0); | |
74 | ||
8ffaadf7 JD |
75 | static bool use_cmb_sqes = true; |
76 | module_param(use_cmb_sqes, bool, 0644); | |
77 | MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes"); | |
78 | ||
1fa6aead MW |
79 | static DEFINE_SPINLOCK(dev_list_lock); |
80 | static LIST_HEAD(dev_list); | |
81 | static struct task_struct *nvme_thread; | |
9a6b9458 | 82 | static struct workqueue_struct *nvme_workq; |
b9afca3e | 83 | static wait_queue_head_t nvme_kthread_wait; |
1fa6aead | 84 | |
b3fffdef KB |
85 | static struct class *nvme_class; |
86 | ||
d4b4ff8e | 87 | static void nvme_reset_failed_dev(struct work_struct *ws); |
4cc06521 | 88 | static int nvme_reset(struct nvme_dev *dev); |
a4aea562 | 89 | static int nvme_process_cq(struct nvme_queue *nvmeq); |
d4b4ff8e | 90 | |
4d115420 KB |
91 | struct async_cmd_info { |
92 | struct kthread_work work; | |
93 | struct kthread_worker *worker; | |
a4aea562 | 94 | struct request *req; |
4d115420 KB |
95 | u32 result; |
96 | int status; | |
97 | void *ctx; | |
98 | }; | |
1fa6aead | 99 | |
b60503ba MW |
100 | /* |
101 | * An NVM Express queue. Each device has at least two (one for admin | |
102 | * commands and one for I/O commands). | |
103 | */ | |
104 | struct nvme_queue { | |
105 | struct device *q_dmadev; | |
091b6092 | 106 | struct nvme_dev *dev; |
3193f07b | 107 | char irqname[24]; /* nvme4294967295-65535\0 */ |
b60503ba MW |
108 | spinlock_t q_lock; |
109 | struct nvme_command *sq_cmds; | |
8ffaadf7 | 110 | struct nvme_command __iomem *sq_cmds_io; |
b60503ba | 111 | volatile struct nvme_completion *cqes; |
42483228 | 112 | struct blk_mq_tags **tags; |
b60503ba MW |
113 | dma_addr_t sq_dma_addr; |
114 | dma_addr_t cq_dma_addr; | |
b60503ba MW |
115 | u32 __iomem *q_db; |
116 | u16 q_depth; | |
6222d172 | 117 | s16 cq_vector; |
b60503ba MW |
118 | u16 sq_head; |
119 | u16 sq_tail; | |
120 | u16 cq_head; | |
c30341dc | 121 | u16 qid; |
e9539f47 MW |
122 | u8 cq_phase; |
123 | u8 cqe_seen; | |
4d115420 | 124 | struct async_cmd_info cmdinfo; |
b60503ba MW |
125 | }; |
126 | ||
127 | /* | |
128 | * Check we didin't inadvertently grow the command struct | |
129 | */ | |
130 | static inline void _nvme_check_size(void) | |
131 | { | |
132 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
133 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); | |
134 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
135 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
136 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
f8ebf840 | 137 | BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); |
c30341dc | 138 | BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); |
b60503ba MW |
139 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); |
140 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); | |
141 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); | |
142 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
6ecec745 | 143 | BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); |
b60503ba MW |
144 | } |
145 | ||
edd10d33 | 146 | typedef void (*nvme_completion_fn)(struct nvme_queue *, void *, |
c2f5b650 MW |
147 | struct nvme_completion *); |
148 | ||
e85248e5 | 149 | struct nvme_cmd_info { |
c2f5b650 MW |
150 | nvme_completion_fn fn; |
151 | void *ctx; | |
c30341dc | 152 | int aborted; |
a4aea562 | 153 | struct nvme_queue *nvmeq; |
ac3dd5bd | 154 | struct nvme_iod iod[0]; |
e85248e5 MW |
155 | }; |
156 | ||
ac3dd5bd JA |
157 | /* |
158 | * Max size of iod being embedded in the request payload | |
159 | */ | |
160 | #define NVME_INT_PAGES 2 | |
161 | #define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->page_size) | |
fda631ff | 162 | #define NVME_INT_MASK 0x01 |
ac3dd5bd JA |
163 | |
164 | /* | |
165 | * Will slightly overestimate the number of pages needed. This is OK | |
166 | * as it only leads to a small amount of wasted memory for the lifetime of | |
167 | * the I/O. | |
168 | */ | |
169 | static int nvme_npages(unsigned size, struct nvme_dev *dev) | |
170 | { | |
171 | unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size); | |
172 | return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); | |
173 | } | |
174 | ||
175 | static unsigned int nvme_cmd_size(struct nvme_dev *dev) | |
176 | { | |
177 | unsigned int ret = sizeof(struct nvme_cmd_info); | |
178 | ||
179 | ret += sizeof(struct nvme_iod); | |
180 | ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev); | |
181 | ret += sizeof(struct scatterlist) * NVME_INT_PAGES; | |
182 | ||
183 | return ret; | |
184 | } | |
185 | ||
a4aea562 MB |
186 | static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
187 | unsigned int hctx_idx) | |
e85248e5 | 188 | { |
a4aea562 MB |
189 | struct nvme_dev *dev = data; |
190 | struct nvme_queue *nvmeq = dev->queues[0]; | |
191 | ||
42483228 KB |
192 | WARN_ON(hctx_idx != 0); |
193 | WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); | |
194 | WARN_ON(nvmeq->tags); | |
195 | ||
a4aea562 | 196 | hctx->driver_data = nvmeq; |
42483228 | 197 | nvmeq->tags = &dev->admin_tagset.tags[0]; |
a4aea562 | 198 | return 0; |
e85248e5 MW |
199 | } |
200 | ||
4af0e21c KB |
201 | static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) |
202 | { | |
203 | struct nvme_queue *nvmeq = hctx->driver_data; | |
204 | ||
205 | nvmeq->tags = NULL; | |
206 | } | |
207 | ||
a4aea562 MB |
208 | static int nvme_admin_init_request(void *data, struct request *req, |
209 | unsigned int hctx_idx, unsigned int rq_idx, | |
210 | unsigned int numa_node) | |
22404274 | 211 | { |
a4aea562 MB |
212 | struct nvme_dev *dev = data; |
213 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
214 | struct nvme_queue *nvmeq = dev->queues[0]; | |
215 | ||
216 | BUG_ON(!nvmeq); | |
217 | cmd->nvmeq = nvmeq; | |
218 | return 0; | |
22404274 KB |
219 | } |
220 | ||
a4aea562 MB |
221 | static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
222 | unsigned int hctx_idx) | |
b60503ba | 223 | { |
a4aea562 | 224 | struct nvme_dev *dev = data; |
42483228 | 225 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; |
a4aea562 | 226 | |
42483228 KB |
227 | if (!nvmeq->tags) |
228 | nvmeq->tags = &dev->tagset.tags[hctx_idx]; | |
b60503ba | 229 | |
42483228 | 230 | WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); |
a4aea562 MB |
231 | hctx->driver_data = nvmeq; |
232 | return 0; | |
b60503ba MW |
233 | } |
234 | ||
a4aea562 MB |
235 | static int nvme_init_request(void *data, struct request *req, |
236 | unsigned int hctx_idx, unsigned int rq_idx, | |
237 | unsigned int numa_node) | |
b60503ba | 238 | { |
a4aea562 MB |
239 | struct nvme_dev *dev = data; |
240 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
241 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; | |
242 | ||
243 | BUG_ON(!nvmeq); | |
244 | cmd->nvmeq = nvmeq; | |
245 | return 0; | |
246 | } | |
247 | ||
248 | static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx, | |
249 | nvme_completion_fn handler) | |
250 | { | |
251 | cmd->fn = handler; | |
252 | cmd->ctx = ctx; | |
253 | cmd->aborted = 0; | |
c917dfe5 | 254 | blk_mq_start_request(blk_mq_rq_from_pdu(cmd)); |
b60503ba MW |
255 | } |
256 | ||
ac3dd5bd JA |
257 | static void *iod_get_private(struct nvme_iod *iod) |
258 | { | |
259 | return (void *) (iod->private & ~0x1UL); | |
260 | } | |
261 | ||
262 | /* | |
263 | * If bit 0 is set, the iod is embedded in the request payload. | |
264 | */ | |
265 | static bool iod_should_kfree(struct nvme_iod *iod) | |
266 | { | |
fda631ff | 267 | return (iod->private & NVME_INT_MASK) == 0; |
ac3dd5bd JA |
268 | } |
269 | ||
c2f5b650 MW |
270 | /* Special values must be less than 0x1000 */ |
271 | #define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA) | |
d2d87034 MW |
272 | #define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE) |
273 | #define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE) | |
274 | #define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE) | |
be7b6275 | 275 | |
edd10d33 | 276 | static void special_completion(struct nvme_queue *nvmeq, void *ctx, |
c2f5b650 MW |
277 | struct nvme_completion *cqe) |
278 | { | |
279 | if (ctx == CMD_CTX_CANCELLED) | |
280 | return; | |
c2f5b650 | 281 | if (ctx == CMD_CTX_COMPLETED) { |
edd10d33 | 282 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
283 | "completed id %d twice on queue %d\n", |
284 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
285 | return; | |
286 | } | |
287 | if (ctx == CMD_CTX_INVALID) { | |
edd10d33 | 288 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
289 | "invalid id %d completed on queue %d\n", |
290 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
291 | return; | |
292 | } | |
edd10d33 | 293 | dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx); |
c2f5b650 MW |
294 | } |
295 | ||
a4aea562 | 296 | static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn) |
b60503ba | 297 | { |
c2f5b650 | 298 | void *ctx; |
b60503ba | 299 | |
859361a2 | 300 | if (fn) |
a4aea562 MB |
301 | *fn = cmd->fn; |
302 | ctx = cmd->ctx; | |
303 | cmd->fn = special_completion; | |
304 | cmd->ctx = CMD_CTX_CANCELLED; | |
c2f5b650 | 305 | return ctx; |
b60503ba MW |
306 | } |
307 | ||
a4aea562 MB |
308 | static void async_req_completion(struct nvme_queue *nvmeq, void *ctx, |
309 | struct nvme_completion *cqe) | |
3c0cf138 | 310 | { |
a4aea562 MB |
311 | u32 result = le32_to_cpup(&cqe->result); |
312 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
313 | ||
314 | if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) | |
315 | ++nvmeq->dev->event_limit; | |
a5768aa8 KB |
316 | if (status != NVME_SC_SUCCESS) |
317 | return; | |
318 | ||
319 | switch (result & 0xff07) { | |
320 | case NVME_AER_NOTICE_NS_CHANGED: | |
321 | dev_info(nvmeq->q_dmadev, "rescanning\n"); | |
322 | schedule_work(&nvmeq->dev->scan_work); | |
323 | default: | |
324 | dev_warn(nvmeq->q_dmadev, "async event result %08x\n", result); | |
325 | } | |
b60503ba MW |
326 | } |
327 | ||
a4aea562 MB |
328 | static void abort_completion(struct nvme_queue *nvmeq, void *ctx, |
329 | struct nvme_completion *cqe) | |
5a92e700 | 330 | { |
a4aea562 MB |
331 | struct request *req = ctx; |
332 | ||
333 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
334 | u32 result = le32_to_cpup(&cqe->result); | |
a51afb54 | 335 | |
42483228 | 336 | blk_mq_free_request(req); |
a51afb54 | 337 | |
a4aea562 MB |
338 | dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result); |
339 | ++nvmeq->dev->abort_limit; | |
5a92e700 KB |
340 | } |
341 | ||
a4aea562 MB |
342 | static void async_completion(struct nvme_queue *nvmeq, void *ctx, |
343 | struct nvme_completion *cqe) | |
b60503ba | 344 | { |
a4aea562 MB |
345 | struct async_cmd_info *cmdinfo = ctx; |
346 | cmdinfo->result = le32_to_cpup(&cqe->result); | |
347 | cmdinfo->status = le16_to_cpup(&cqe->status) >> 1; | |
348 | queue_kthread_work(cmdinfo->worker, &cmdinfo->work); | |
42483228 | 349 | blk_mq_free_request(cmdinfo->req); |
b60503ba MW |
350 | } |
351 | ||
a4aea562 MB |
352 | static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq, |
353 | unsigned int tag) | |
b60503ba | 354 | { |
42483228 | 355 | struct request *req = blk_mq_tag_to_rq(*nvmeq->tags, tag); |
a51afb54 | 356 | |
a4aea562 | 357 | return blk_mq_rq_to_pdu(req); |
4f5099af KB |
358 | } |
359 | ||
a4aea562 MB |
360 | /* |
361 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
362 | */ | |
363 | static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag, | |
364 | nvme_completion_fn *fn) | |
4f5099af | 365 | { |
a4aea562 MB |
366 | struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag); |
367 | void *ctx; | |
368 | if (tag >= nvmeq->q_depth) { | |
369 | *fn = special_completion; | |
370 | return CMD_CTX_INVALID; | |
371 | } | |
372 | if (fn) | |
373 | *fn = cmd->fn; | |
374 | ctx = cmd->ctx; | |
375 | cmd->fn = special_completion; | |
376 | cmd->ctx = CMD_CTX_COMPLETED; | |
377 | return ctx; | |
b60503ba MW |
378 | } |
379 | ||
380 | /** | |
714a7a22 | 381 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
382 | * @nvmeq: The queue to use |
383 | * @cmd: The command to send | |
384 | * | |
385 | * Safe to use from interrupt context | |
386 | */ | |
e3f879bf SB |
387 | static void __nvme_submit_cmd(struct nvme_queue *nvmeq, |
388 | struct nvme_command *cmd) | |
b60503ba | 389 | { |
a4aea562 MB |
390 | u16 tail = nvmeq->sq_tail; |
391 | ||
8ffaadf7 JD |
392 | if (nvmeq->sq_cmds_io) |
393 | memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd)); | |
394 | else | |
395 | memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); | |
396 | ||
b60503ba MW |
397 | if (++tail == nvmeq->q_depth) |
398 | tail = 0; | |
7547881d | 399 | writel(tail, nvmeq->q_db); |
b60503ba | 400 | nvmeq->sq_tail = tail; |
b60503ba MW |
401 | } |
402 | ||
e3f879bf | 403 | static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd) |
a4aea562 MB |
404 | { |
405 | unsigned long flags; | |
a4aea562 | 406 | spin_lock_irqsave(&nvmeq->q_lock, flags); |
e3f879bf | 407 | __nvme_submit_cmd(nvmeq, cmd); |
a4aea562 | 408 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); |
a4aea562 MB |
409 | } |
410 | ||
eca18b23 | 411 | static __le64 **iod_list(struct nvme_iod *iod) |
e025344c | 412 | { |
eca18b23 | 413 | return ((void *)iod) + iod->offset; |
e025344c SMM |
414 | } |
415 | ||
ac3dd5bd JA |
416 | static inline void iod_init(struct nvme_iod *iod, unsigned nbytes, |
417 | unsigned nseg, unsigned long private) | |
eca18b23 | 418 | { |
ac3dd5bd JA |
419 | iod->private = private; |
420 | iod->offset = offsetof(struct nvme_iod, sg[nseg]); | |
421 | iod->npages = -1; | |
422 | iod->length = nbytes; | |
423 | iod->nents = 0; | |
eca18b23 | 424 | } |
b60503ba | 425 | |
eca18b23 | 426 | static struct nvme_iod * |
ac3dd5bd JA |
427 | __nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev, |
428 | unsigned long priv, gfp_t gfp) | |
b60503ba | 429 | { |
eca18b23 | 430 | struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) + |
ac3dd5bd | 431 | sizeof(__le64 *) * nvme_npages(bytes, dev) + |
eca18b23 MW |
432 | sizeof(struct scatterlist) * nseg, gfp); |
433 | ||
ac3dd5bd JA |
434 | if (iod) |
435 | iod_init(iod, bytes, nseg, priv); | |
eca18b23 MW |
436 | |
437 | return iod; | |
b60503ba MW |
438 | } |
439 | ||
ac3dd5bd JA |
440 | static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev, |
441 | gfp_t gfp) | |
442 | { | |
443 | unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) : | |
444 | sizeof(struct nvme_dsm_range); | |
ac3dd5bd JA |
445 | struct nvme_iod *iod; |
446 | ||
447 | if (rq->nr_phys_segments <= NVME_INT_PAGES && | |
448 | size <= NVME_INT_BYTES(dev)) { | |
449 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq); | |
450 | ||
451 | iod = cmd->iod; | |
ac3dd5bd | 452 | iod_init(iod, size, rq->nr_phys_segments, |
fda631ff | 453 | (unsigned long) rq | NVME_INT_MASK); |
ac3dd5bd JA |
454 | return iod; |
455 | } | |
456 | ||
457 | return __nvme_alloc_iod(rq->nr_phys_segments, size, dev, | |
458 | (unsigned long) rq, gfp); | |
459 | } | |
460 | ||
d29ec824 | 461 | static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod) |
b60503ba | 462 | { |
1d090624 | 463 | const int last_prp = dev->page_size / 8 - 1; |
eca18b23 MW |
464 | int i; |
465 | __le64 **list = iod_list(iod); | |
466 | dma_addr_t prp_dma = iod->first_dma; | |
467 | ||
468 | if (iod->npages == 0) | |
469 | dma_pool_free(dev->prp_small_pool, list[0], prp_dma); | |
470 | for (i = 0; i < iod->npages; i++) { | |
471 | __le64 *prp_list = list[i]; | |
472 | dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); | |
473 | dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); | |
474 | prp_dma = next_prp_dma; | |
475 | } | |
ac3dd5bd JA |
476 | |
477 | if (iod_should_kfree(iod)) | |
478 | kfree(iod); | |
b60503ba MW |
479 | } |
480 | ||
b4ff9c8d KB |
481 | static int nvme_error_status(u16 status) |
482 | { | |
483 | switch (status & 0x7ff) { | |
484 | case NVME_SC_SUCCESS: | |
485 | return 0; | |
486 | case NVME_SC_CAP_EXCEEDED: | |
487 | return -ENOSPC; | |
488 | default: | |
489 | return -EIO; | |
490 | } | |
491 | } | |
492 | ||
52b68d7e | 493 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
e1e5e564 KB |
494 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) |
495 | { | |
496 | if (be32_to_cpu(pi->ref_tag) == v) | |
497 | pi->ref_tag = cpu_to_be32(p); | |
498 | } | |
499 | ||
500 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
501 | { | |
502 | if (be32_to_cpu(pi->ref_tag) == p) | |
503 | pi->ref_tag = cpu_to_be32(v); | |
504 | } | |
505 | ||
506 | /** | |
507 | * nvme_dif_remap - remaps ref tags to bip seed and physical lba | |
508 | * | |
509 | * The virtual start sector is the one that was originally submitted by the | |
510 | * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical | |
511 | * start sector may be different. Remap protection information to match the | |
512 | * physical LBA on writes, and back to the original seed on reads. | |
513 | * | |
514 | * Type 0 and 3 do not have a ref tag, so no remapping required. | |
515 | */ | |
516 | static void nvme_dif_remap(struct request *req, | |
517 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
518 | { | |
519 | struct nvme_ns *ns = req->rq_disk->private_data; | |
520 | struct bio_integrity_payload *bip; | |
521 | struct t10_pi_tuple *pi; | |
522 | void *p, *pmap; | |
523 | u32 i, nlb, ts, phys, virt; | |
524 | ||
525 | if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3) | |
526 | return; | |
527 | ||
528 | bip = bio_integrity(req->bio); | |
529 | if (!bip) | |
530 | return; | |
531 | ||
532 | pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset; | |
e1e5e564 KB |
533 | |
534 | p = pmap; | |
535 | virt = bip_get_seed(bip); | |
536 | phys = nvme_block_nr(ns, blk_rq_pos(req)); | |
537 | nlb = (blk_rq_bytes(req) >> ns->lba_shift); | |
538 | ts = ns->disk->integrity->tuple_size; | |
539 | ||
540 | for (i = 0; i < nlb; i++, virt++, phys++) { | |
541 | pi = (struct t10_pi_tuple *)p; | |
542 | dif_swap(phys, virt, pi); | |
543 | p += ts; | |
544 | } | |
545 | kunmap_atomic(pmap); | |
546 | } | |
547 | ||
52b68d7e KB |
548 | static int nvme_noop_verify(struct blk_integrity_iter *iter) |
549 | { | |
550 | return 0; | |
551 | } | |
552 | ||
553 | static int nvme_noop_generate(struct blk_integrity_iter *iter) | |
554 | { | |
555 | return 0; | |
556 | } | |
557 | ||
558 | struct blk_integrity nvme_meta_noop = { | |
559 | .name = "NVME_META_NOOP", | |
560 | .generate_fn = nvme_noop_generate, | |
561 | .verify_fn = nvme_noop_verify, | |
562 | }; | |
563 | ||
564 | static void nvme_init_integrity(struct nvme_ns *ns) | |
565 | { | |
566 | struct blk_integrity integrity; | |
567 | ||
568 | switch (ns->pi_type) { | |
569 | case NVME_NS_DPS_PI_TYPE3: | |
570 | integrity = t10_pi_type3_crc; | |
571 | break; | |
572 | case NVME_NS_DPS_PI_TYPE1: | |
573 | case NVME_NS_DPS_PI_TYPE2: | |
574 | integrity = t10_pi_type1_crc; | |
575 | break; | |
576 | default: | |
577 | integrity = nvme_meta_noop; | |
578 | break; | |
579 | } | |
580 | integrity.tuple_size = ns->ms; | |
581 | blk_integrity_register(ns->disk, &integrity); | |
582 | blk_queue_max_integrity_segments(ns->queue, 1); | |
583 | } | |
584 | #else /* CONFIG_BLK_DEV_INTEGRITY */ | |
585 | static void nvme_dif_remap(struct request *req, | |
586 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
587 | { | |
588 | } | |
589 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) | |
590 | { | |
591 | } | |
592 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
593 | { | |
594 | } | |
595 | static void nvme_init_integrity(struct nvme_ns *ns) | |
596 | { | |
597 | } | |
598 | #endif | |
599 | ||
a4aea562 | 600 | static void req_completion(struct nvme_queue *nvmeq, void *ctx, |
b60503ba MW |
601 | struct nvme_completion *cqe) |
602 | { | |
eca18b23 | 603 | struct nvme_iod *iod = ctx; |
ac3dd5bd | 604 | struct request *req = iod_get_private(iod); |
a4aea562 MB |
605 | struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req); |
606 | ||
b60503ba MW |
607 | u16 status = le16_to_cpup(&cqe->status) >> 1; |
608 | ||
edd10d33 | 609 | if (unlikely(status)) { |
a4aea562 MB |
610 | if (!(status & NVME_SC_DNR || blk_noretry_request(req)) |
611 | && (jiffies - req->start_time) < req->timeout) { | |
c9d3bf88 KB |
612 | unsigned long flags; |
613 | ||
a4aea562 | 614 | blk_mq_requeue_request(req); |
c9d3bf88 KB |
615 | spin_lock_irqsave(req->q->queue_lock, flags); |
616 | if (!blk_queue_stopped(req->q)) | |
617 | blk_mq_kick_requeue_list(req->q); | |
618 | spin_unlock_irqrestore(req->q->queue_lock, flags); | |
edd10d33 KB |
619 | return; |
620 | } | |
d29ec824 | 621 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) { |
17188bb4 KB |
622 | if (cmd_rq->ctx == CMD_CTX_CANCELLED) |
623 | req->errors = -EINTR; | |
624 | else | |
625 | req->errors = status; | |
d29ec824 CH |
626 | } else { |
627 | req->errors = nvme_error_status(status); | |
628 | } | |
a4aea562 MB |
629 | } else |
630 | req->errors = 0; | |
a0a931d6 KB |
631 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) { |
632 | u32 result = le32_to_cpup(&cqe->result); | |
633 | req->special = (void *)(uintptr_t)result; | |
634 | } | |
a4aea562 MB |
635 | |
636 | if (cmd_rq->aborted) | |
e75ec752 | 637 | dev_warn(nvmeq->dev->dev, |
a4aea562 MB |
638 | "completing aborted command with status:%04x\n", |
639 | status); | |
640 | ||
e1e5e564 | 641 | if (iod->nents) { |
e75ec752 | 642 | dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents, |
a4aea562 | 643 | rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
e1e5e564 KB |
644 | if (blk_integrity_rq(req)) { |
645 | if (!rq_data_dir(req)) | |
646 | nvme_dif_remap(req, nvme_dif_complete); | |
e75ec752 | 647 | dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1, |
e1e5e564 KB |
648 | rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
649 | } | |
650 | } | |
edd10d33 | 651 | nvme_free_iod(nvmeq->dev, iod); |
3291fa57 | 652 | |
a4aea562 | 653 | blk_mq_complete_request(req); |
b60503ba MW |
654 | } |
655 | ||
184d2944 | 656 | /* length is in bytes. gfp flags indicates whether we may sleep. */ |
d29ec824 CH |
657 | static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod, |
658 | int total_len, gfp_t gfp) | |
ff22b54f | 659 | { |
99802a7a | 660 | struct dma_pool *pool; |
eca18b23 MW |
661 | int length = total_len; |
662 | struct scatterlist *sg = iod->sg; | |
ff22b54f MW |
663 | int dma_len = sg_dma_len(sg); |
664 | u64 dma_addr = sg_dma_address(sg); | |
f137e0f1 MI |
665 | u32 page_size = dev->page_size; |
666 | int offset = dma_addr & (page_size - 1); | |
e025344c | 667 | __le64 *prp_list; |
eca18b23 | 668 | __le64 **list = iod_list(iod); |
e025344c | 669 | dma_addr_t prp_dma; |
eca18b23 | 670 | int nprps, i; |
ff22b54f | 671 | |
1d090624 | 672 | length -= (page_size - offset); |
ff22b54f | 673 | if (length <= 0) |
eca18b23 | 674 | return total_len; |
ff22b54f | 675 | |
1d090624 | 676 | dma_len -= (page_size - offset); |
ff22b54f | 677 | if (dma_len) { |
1d090624 | 678 | dma_addr += (page_size - offset); |
ff22b54f MW |
679 | } else { |
680 | sg = sg_next(sg); | |
681 | dma_addr = sg_dma_address(sg); | |
682 | dma_len = sg_dma_len(sg); | |
683 | } | |
684 | ||
1d090624 | 685 | if (length <= page_size) { |
edd10d33 | 686 | iod->first_dma = dma_addr; |
eca18b23 | 687 | return total_len; |
e025344c SMM |
688 | } |
689 | ||
1d090624 | 690 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
691 | if (nprps <= (256 / 8)) { |
692 | pool = dev->prp_small_pool; | |
eca18b23 | 693 | iod->npages = 0; |
99802a7a MW |
694 | } else { |
695 | pool = dev->prp_page_pool; | |
eca18b23 | 696 | iod->npages = 1; |
99802a7a MW |
697 | } |
698 | ||
b77954cb MW |
699 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
700 | if (!prp_list) { | |
edd10d33 | 701 | iod->first_dma = dma_addr; |
eca18b23 | 702 | iod->npages = -1; |
1d090624 | 703 | return (total_len - length) + page_size; |
b77954cb | 704 | } |
eca18b23 MW |
705 | list[0] = prp_list; |
706 | iod->first_dma = prp_dma; | |
e025344c SMM |
707 | i = 0; |
708 | for (;;) { | |
1d090624 | 709 | if (i == page_size >> 3) { |
e025344c | 710 | __le64 *old_prp_list = prp_list; |
b77954cb | 711 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
eca18b23 MW |
712 | if (!prp_list) |
713 | return total_len - length; | |
714 | list[iod->npages++] = prp_list; | |
7523d834 MW |
715 | prp_list[0] = old_prp_list[i - 1]; |
716 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
717 | i = 1; | |
e025344c SMM |
718 | } |
719 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
720 | dma_len -= page_size; |
721 | dma_addr += page_size; | |
722 | length -= page_size; | |
e025344c SMM |
723 | if (length <= 0) |
724 | break; | |
725 | if (dma_len > 0) | |
726 | continue; | |
727 | BUG_ON(dma_len < 0); | |
728 | sg = sg_next(sg); | |
729 | dma_addr = sg_dma_address(sg); | |
730 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
731 | } |
732 | ||
eca18b23 | 733 | return total_len; |
ff22b54f MW |
734 | } |
735 | ||
d29ec824 CH |
736 | static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req, |
737 | struct nvme_iod *iod) | |
738 | { | |
498c4394 | 739 | struct nvme_command cmnd; |
d29ec824 | 740 | |
498c4394 JD |
741 | memcpy(&cmnd, req->cmd, sizeof(cmnd)); |
742 | cmnd.rw.command_id = req->tag; | |
d29ec824 | 743 | if (req->nr_phys_segments) { |
498c4394 JD |
744 | cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); |
745 | cmnd.rw.prp2 = cpu_to_le64(iod->first_dma); | |
d29ec824 CH |
746 | } |
747 | ||
498c4394 | 748 | __nvme_submit_cmd(nvmeq, &cmnd); |
d29ec824 CH |
749 | } |
750 | ||
a4aea562 MB |
751 | /* |
752 | * We reuse the small pool to allocate the 16-byte range here as it is not | |
753 | * worth having a special pool for these or additional cases to handle freeing | |
754 | * the iod. | |
755 | */ | |
756 | static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns, | |
757 | struct request *req, struct nvme_iod *iod) | |
0e5e4f0e | 758 | { |
edd10d33 KB |
759 | struct nvme_dsm_range *range = |
760 | (struct nvme_dsm_range *)iod_list(iod)[0]; | |
498c4394 | 761 | struct nvme_command cmnd; |
0e5e4f0e | 762 | |
0e5e4f0e | 763 | range->cattr = cpu_to_le32(0); |
a4aea562 MB |
764 | range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift); |
765 | range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
0e5e4f0e | 766 | |
498c4394 JD |
767 | memset(&cmnd, 0, sizeof(cmnd)); |
768 | cmnd.dsm.opcode = nvme_cmd_dsm; | |
769 | cmnd.dsm.command_id = req->tag; | |
770 | cmnd.dsm.nsid = cpu_to_le32(ns->ns_id); | |
771 | cmnd.dsm.prp1 = cpu_to_le64(iod->first_dma); | |
772 | cmnd.dsm.nr = 0; | |
773 | cmnd.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); | |
0e5e4f0e | 774 | |
498c4394 | 775 | __nvme_submit_cmd(nvmeq, &cmnd); |
0e5e4f0e KB |
776 | } |
777 | ||
a4aea562 | 778 | static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
00df5cb4 MW |
779 | int cmdid) |
780 | { | |
498c4394 | 781 | struct nvme_command cmnd; |
00df5cb4 | 782 | |
498c4394 JD |
783 | memset(&cmnd, 0, sizeof(cmnd)); |
784 | cmnd.common.opcode = nvme_cmd_flush; | |
785 | cmnd.common.command_id = cmdid; | |
786 | cmnd.common.nsid = cpu_to_le32(ns->ns_id); | |
00df5cb4 | 787 | |
498c4394 | 788 | __nvme_submit_cmd(nvmeq, &cmnd); |
00df5cb4 MW |
789 | } |
790 | ||
a4aea562 MB |
791 | static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod, |
792 | struct nvme_ns *ns) | |
b60503ba | 793 | { |
ac3dd5bd | 794 | struct request *req = iod_get_private(iod); |
498c4394 | 795 | struct nvme_command cmnd; |
a4aea562 MB |
796 | u16 control = 0; |
797 | u32 dsmgmt = 0; | |
00df5cb4 | 798 | |
a4aea562 | 799 | if (req->cmd_flags & REQ_FUA) |
b60503ba | 800 | control |= NVME_RW_FUA; |
a4aea562 | 801 | if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) |
b60503ba MW |
802 | control |= NVME_RW_LR; |
803 | ||
a4aea562 | 804 | if (req->cmd_flags & REQ_RAHEAD) |
b60503ba MW |
805 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; |
806 | ||
498c4394 JD |
807 | memset(&cmnd, 0, sizeof(cmnd)); |
808 | cmnd.rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); | |
809 | cmnd.rw.command_id = req->tag; | |
810 | cmnd.rw.nsid = cpu_to_le32(ns->ns_id); | |
811 | cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
812 | cmnd.rw.prp2 = cpu_to_le64(iod->first_dma); | |
813 | cmnd.rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
814 | cmnd.rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); | |
e1e5e564 KB |
815 | |
816 | if (blk_integrity_rq(req)) { | |
498c4394 | 817 | cmnd.rw.metadata = cpu_to_le64(sg_dma_address(iod->meta_sg)); |
e1e5e564 KB |
818 | switch (ns->pi_type) { |
819 | case NVME_NS_DPS_PI_TYPE3: | |
820 | control |= NVME_RW_PRINFO_PRCHK_GUARD; | |
821 | break; | |
822 | case NVME_NS_DPS_PI_TYPE1: | |
823 | case NVME_NS_DPS_PI_TYPE2: | |
824 | control |= NVME_RW_PRINFO_PRCHK_GUARD | | |
825 | NVME_RW_PRINFO_PRCHK_REF; | |
498c4394 | 826 | cmnd.rw.reftag = cpu_to_le32( |
e1e5e564 KB |
827 | nvme_block_nr(ns, blk_rq_pos(req))); |
828 | break; | |
829 | } | |
830 | } else if (ns->ms) | |
831 | control |= NVME_RW_PRINFO_PRACT; | |
832 | ||
498c4394 JD |
833 | cmnd.rw.control = cpu_to_le16(control); |
834 | cmnd.rw.dsmgmt = cpu_to_le32(dsmgmt); | |
b60503ba | 835 | |
498c4394 | 836 | __nvme_submit_cmd(nvmeq, &cmnd); |
b60503ba | 837 | |
1974b1ae | 838 | return 0; |
edd10d33 KB |
839 | } |
840 | ||
d29ec824 CH |
841 | /* |
842 | * NOTE: ns is NULL when called on the admin queue. | |
843 | */ | |
a4aea562 MB |
844 | static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx, |
845 | const struct blk_mq_queue_data *bd) | |
edd10d33 | 846 | { |
a4aea562 MB |
847 | struct nvme_ns *ns = hctx->queue->queuedata; |
848 | struct nvme_queue *nvmeq = hctx->driver_data; | |
d29ec824 | 849 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 MB |
850 | struct request *req = bd->rq; |
851 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
edd10d33 | 852 | struct nvme_iod *iod; |
a4aea562 | 853 | enum dma_data_direction dma_dir; |
edd10d33 | 854 | |
e1e5e564 KB |
855 | /* |
856 | * If formated with metadata, require the block layer provide a buffer | |
857 | * unless this namespace is formated such that the metadata can be | |
858 | * stripped/generated by the controller with PRACT=1. | |
859 | */ | |
d29ec824 | 860 | if (ns && ns->ms && !blk_integrity_rq(req)) { |
71feb364 KB |
861 | if (!(ns->pi_type && ns->ms == 8) && |
862 | req->cmd_type != REQ_TYPE_DRV_PRIV) { | |
e1e5e564 KB |
863 | req->errors = -EFAULT; |
864 | blk_mq_complete_request(req); | |
865 | return BLK_MQ_RQ_QUEUE_OK; | |
866 | } | |
867 | } | |
868 | ||
d29ec824 | 869 | iod = nvme_alloc_iod(req, dev, GFP_ATOMIC); |
edd10d33 | 870 | if (!iod) |
fe54303e | 871 | return BLK_MQ_RQ_QUEUE_BUSY; |
a4aea562 | 872 | |
a4aea562 | 873 | if (req->cmd_flags & REQ_DISCARD) { |
edd10d33 KB |
874 | void *range; |
875 | /* | |
876 | * We reuse the small pool to allocate the 16-byte range here | |
877 | * as it is not worth having a special pool for these or | |
878 | * additional cases to handle freeing the iod. | |
879 | */ | |
d29ec824 | 880 | range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC, |
edd10d33 | 881 | &iod->first_dma); |
a4aea562 | 882 | if (!range) |
fe54303e | 883 | goto retry_cmd; |
edd10d33 KB |
884 | iod_list(iod)[0] = (__le64 *)range; |
885 | iod->npages = 0; | |
ac3dd5bd | 886 | } else if (req->nr_phys_segments) { |
a4aea562 MB |
887 | dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE; |
888 | ||
ac3dd5bd | 889 | sg_init_table(iod->sg, req->nr_phys_segments); |
a4aea562 | 890 | iod->nents = blk_rq_map_sg(req->q, req, iod->sg); |
fe54303e JA |
891 | if (!iod->nents) |
892 | goto error_cmd; | |
a4aea562 MB |
893 | |
894 | if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir)) | |
fe54303e | 895 | goto retry_cmd; |
a4aea562 | 896 | |
fe54303e | 897 | if (blk_rq_bytes(req) != |
d29ec824 CH |
898 | nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) { |
899 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); | |
fe54303e JA |
900 | goto retry_cmd; |
901 | } | |
e1e5e564 KB |
902 | if (blk_integrity_rq(req)) { |
903 | if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) | |
904 | goto error_cmd; | |
905 | ||
906 | sg_init_table(iod->meta_sg, 1); | |
907 | if (blk_rq_map_integrity_sg( | |
908 | req->q, req->bio, iod->meta_sg) != 1) | |
909 | goto error_cmd; | |
910 | ||
911 | if (rq_data_dir(req)) | |
912 | nvme_dif_remap(req, nvme_dif_prep); | |
913 | ||
914 | if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir)) | |
915 | goto error_cmd; | |
916 | } | |
edd10d33 | 917 | } |
1974b1ae | 918 | |
9af8785a | 919 | nvme_set_info(cmd, iod, req_completion); |
a4aea562 | 920 | spin_lock_irq(&nvmeq->q_lock); |
d29ec824 CH |
921 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) |
922 | nvme_submit_priv(nvmeq, req, iod); | |
923 | else if (req->cmd_flags & REQ_DISCARD) | |
a4aea562 MB |
924 | nvme_submit_discard(nvmeq, ns, req, iod); |
925 | else if (req->cmd_flags & REQ_FLUSH) | |
926 | nvme_submit_flush(nvmeq, ns, req->tag); | |
927 | else | |
928 | nvme_submit_iod(nvmeq, iod, ns); | |
929 | ||
930 | nvme_process_cq(nvmeq); | |
931 | spin_unlock_irq(&nvmeq->q_lock); | |
932 | return BLK_MQ_RQ_QUEUE_OK; | |
933 | ||
fe54303e | 934 | error_cmd: |
d29ec824 | 935 | nvme_free_iod(dev, iod); |
fe54303e JA |
936 | return BLK_MQ_RQ_QUEUE_ERROR; |
937 | retry_cmd: | |
d29ec824 | 938 | nvme_free_iod(dev, iod); |
fe54303e | 939 | return BLK_MQ_RQ_QUEUE_BUSY; |
b60503ba MW |
940 | } |
941 | ||
e9539f47 | 942 | static int nvme_process_cq(struct nvme_queue *nvmeq) |
b60503ba | 943 | { |
82123460 | 944 | u16 head, phase; |
b60503ba | 945 | |
b60503ba | 946 | head = nvmeq->cq_head; |
82123460 | 947 | phase = nvmeq->cq_phase; |
b60503ba MW |
948 | |
949 | for (;;) { | |
c2f5b650 MW |
950 | void *ctx; |
951 | nvme_completion_fn fn; | |
b60503ba | 952 | struct nvme_completion cqe = nvmeq->cqes[head]; |
82123460 | 953 | if ((le16_to_cpu(cqe.status) & 1) != phase) |
b60503ba MW |
954 | break; |
955 | nvmeq->sq_head = le16_to_cpu(cqe.sq_head); | |
956 | if (++head == nvmeq->q_depth) { | |
957 | head = 0; | |
82123460 | 958 | phase = !phase; |
b60503ba | 959 | } |
a4aea562 | 960 | ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn); |
edd10d33 | 961 | fn(nvmeq, ctx, &cqe); |
b60503ba MW |
962 | } |
963 | ||
964 | /* If the controller ignores the cq head doorbell and continuously | |
965 | * writes to the queue, it is theoretically possible to wrap around | |
966 | * the queue twice and mistakenly return IRQ_NONE. Linux only | |
967 | * requires that 0.1% of your interrupts are handled, so this isn't | |
968 | * a big problem. | |
969 | */ | |
82123460 | 970 | if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) |
e9539f47 | 971 | return 0; |
b60503ba | 972 | |
b80d5ccc | 973 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); |
b60503ba | 974 | nvmeq->cq_head = head; |
82123460 | 975 | nvmeq->cq_phase = phase; |
b60503ba | 976 | |
e9539f47 MW |
977 | nvmeq->cqe_seen = 1; |
978 | return 1; | |
b60503ba MW |
979 | } |
980 | ||
981 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 MW |
982 | { |
983 | irqreturn_t result; | |
984 | struct nvme_queue *nvmeq = data; | |
985 | spin_lock(&nvmeq->q_lock); | |
e9539f47 MW |
986 | nvme_process_cq(nvmeq); |
987 | result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; | |
988 | nvmeq->cqe_seen = 0; | |
58ffacb5 MW |
989 | spin_unlock(&nvmeq->q_lock); |
990 | return result; | |
991 | } | |
992 | ||
993 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
994 | { | |
995 | struct nvme_queue *nvmeq = data; | |
996 | struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head]; | |
997 | if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase) | |
998 | return IRQ_NONE; | |
999 | return IRQ_WAKE_THREAD; | |
1000 | } | |
1001 | ||
b60503ba MW |
1002 | /* |
1003 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
1004 | * if the result is positive, it's an NVM Express status code | |
1005 | */ | |
d29ec824 CH |
1006 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
1007 | void *buffer, void __user *ubuffer, unsigned bufflen, | |
1008 | u32 *result, unsigned timeout) | |
b60503ba | 1009 | { |
d29ec824 CH |
1010 | bool write = cmd->common.opcode & 1; |
1011 | struct bio *bio = NULL; | |
f705f837 | 1012 | struct request *req; |
d29ec824 | 1013 | int ret; |
b60503ba | 1014 | |
d29ec824 | 1015 | req = blk_mq_alloc_request(q, write, GFP_KERNEL, false); |
f705f837 CH |
1016 | if (IS_ERR(req)) |
1017 | return PTR_ERR(req); | |
b60503ba | 1018 | |
d29ec824 | 1019 | req->cmd_type = REQ_TYPE_DRV_PRIV; |
e112af0d | 1020 | req->cmd_flags |= REQ_FAILFAST_DRIVER; |
d29ec824 CH |
1021 | req->__data_len = 0; |
1022 | req->__sector = (sector_t) -1; | |
1023 | req->bio = req->biotail = NULL; | |
b60503ba | 1024 | |
f4ff414a | 1025 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; |
a4aea562 | 1026 | |
d29ec824 CH |
1027 | req->cmd = (unsigned char *)cmd; |
1028 | req->cmd_len = sizeof(struct nvme_command); | |
a0a931d6 | 1029 | req->special = (void *)0; |
b60503ba | 1030 | |
d29ec824 CH |
1031 | if (buffer && bufflen) { |
1032 | ret = blk_rq_map_kern(q, req, buffer, bufflen, __GFP_WAIT); | |
1033 | if (ret) | |
1034 | goto out; | |
1035 | } else if (ubuffer && bufflen) { | |
1036 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, __GFP_WAIT); | |
1037 | if (ret) | |
1038 | goto out; | |
1039 | bio = req->bio; | |
1040 | } | |
3c0cf138 | 1041 | |
d29ec824 CH |
1042 | blk_execute_rq(req->q, NULL, req, 0); |
1043 | if (bio) | |
1044 | blk_rq_unmap_user(bio); | |
b60503ba | 1045 | if (result) |
a0a931d6 | 1046 | *result = (u32)(uintptr_t)req->special; |
d29ec824 CH |
1047 | ret = req->errors; |
1048 | out: | |
f705f837 | 1049 | blk_mq_free_request(req); |
d29ec824 | 1050 | return ret; |
f705f837 CH |
1051 | } |
1052 | ||
d29ec824 CH |
1053 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
1054 | void *buffer, unsigned bufflen) | |
f705f837 | 1055 | { |
d29ec824 | 1056 | return __nvme_submit_sync_cmd(q, cmd, buffer, NULL, bufflen, NULL, 0); |
b60503ba MW |
1057 | } |
1058 | ||
a4aea562 MB |
1059 | static int nvme_submit_async_admin_req(struct nvme_dev *dev) |
1060 | { | |
1061 | struct nvme_queue *nvmeq = dev->queues[0]; | |
1062 | struct nvme_command c; | |
1063 | struct nvme_cmd_info *cmd_info; | |
1064 | struct request *req; | |
1065 | ||
1efccc9d | 1066 | req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, true); |
9f173b33 DC |
1067 | if (IS_ERR(req)) |
1068 | return PTR_ERR(req); | |
a4aea562 | 1069 | |
c917dfe5 | 1070 | req->cmd_flags |= REQ_NO_TIMEOUT; |
a4aea562 | 1071 | cmd_info = blk_mq_rq_to_pdu(req); |
1efccc9d | 1072 | nvme_set_info(cmd_info, NULL, async_req_completion); |
a4aea562 MB |
1073 | |
1074 | memset(&c, 0, sizeof(c)); | |
1075 | c.common.opcode = nvme_admin_async_event; | |
1076 | c.common.command_id = req->tag; | |
1077 | ||
42483228 | 1078 | blk_mq_free_request(req); |
e3f879bf SB |
1079 | __nvme_submit_cmd(nvmeq, &c); |
1080 | return 0; | |
a4aea562 MB |
1081 | } |
1082 | ||
1083 | static int nvme_submit_admin_async_cmd(struct nvme_dev *dev, | |
4d115420 KB |
1084 | struct nvme_command *cmd, |
1085 | struct async_cmd_info *cmdinfo, unsigned timeout) | |
1086 | { | |
a4aea562 MB |
1087 | struct nvme_queue *nvmeq = dev->queues[0]; |
1088 | struct request *req; | |
1089 | struct nvme_cmd_info *cmd_rq; | |
4d115420 | 1090 | |
a4aea562 | 1091 | req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_KERNEL, false); |
9f173b33 DC |
1092 | if (IS_ERR(req)) |
1093 | return PTR_ERR(req); | |
a4aea562 MB |
1094 | |
1095 | req->timeout = timeout; | |
1096 | cmd_rq = blk_mq_rq_to_pdu(req); | |
1097 | cmdinfo->req = req; | |
1098 | nvme_set_info(cmd_rq, cmdinfo, async_completion); | |
4d115420 | 1099 | cmdinfo->status = -EINTR; |
a4aea562 MB |
1100 | |
1101 | cmd->common.command_id = req->tag; | |
1102 | ||
e3f879bf SB |
1103 | nvme_submit_cmd(nvmeq, cmd); |
1104 | return 0; | |
4d115420 KB |
1105 | } |
1106 | ||
b60503ba MW |
1107 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
1108 | { | |
b60503ba MW |
1109 | struct nvme_command c; |
1110 | ||
1111 | memset(&c, 0, sizeof(c)); | |
1112 | c.delete_queue.opcode = opcode; | |
1113 | c.delete_queue.qid = cpu_to_le16(id); | |
1114 | ||
d29ec824 | 1115 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1116 | } |
1117 | ||
1118 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, | |
1119 | struct nvme_queue *nvmeq) | |
1120 | { | |
b60503ba MW |
1121 | struct nvme_command c; |
1122 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; | |
1123 | ||
d29ec824 CH |
1124 | /* |
1125 | * Note: we (ab)use the fact the the prp fields survive if no data | |
1126 | * is attached to the request. | |
1127 | */ | |
b60503ba MW |
1128 | memset(&c, 0, sizeof(c)); |
1129 | c.create_cq.opcode = nvme_admin_create_cq; | |
1130 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
1131 | c.create_cq.cqid = cpu_to_le16(qid); | |
1132 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1133 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
1134 | c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); | |
1135 | ||
d29ec824 | 1136 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1137 | } |
1138 | ||
1139 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
1140 | struct nvme_queue *nvmeq) | |
1141 | { | |
b60503ba MW |
1142 | struct nvme_command c; |
1143 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; | |
1144 | ||
d29ec824 CH |
1145 | /* |
1146 | * Note: we (ab)use the fact the the prp fields survive if no data | |
1147 | * is attached to the request. | |
1148 | */ | |
b60503ba MW |
1149 | memset(&c, 0, sizeof(c)); |
1150 | c.create_sq.opcode = nvme_admin_create_sq; | |
1151 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
1152 | c.create_sq.sqid = cpu_to_le16(qid); | |
1153 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1154 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
1155 | c.create_sq.cqid = cpu_to_le16(qid); | |
1156 | ||
d29ec824 | 1157 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1158 | } |
1159 | ||
1160 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
1161 | { | |
1162 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
1163 | } | |
1164 | ||
1165 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
1166 | { | |
1167 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
1168 | } | |
1169 | ||
d29ec824 | 1170 | int nvme_identify_ctrl(struct nvme_dev *dev, struct nvme_id_ctrl **id) |
bc5fc7e4 | 1171 | { |
e44ac588 | 1172 | struct nvme_command c = { }; |
d29ec824 | 1173 | int error; |
bc5fc7e4 | 1174 | |
e44ac588 AM |
1175 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ |
1176 | c.identify.opcode = nvme_admin_identify; | |
1177 | c.identify.cns = cpu_to_le32(1); | |
1178 | ||
d29ec824 CH |
1179 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); |
1180 | if (!*id) | |
1181 | return -ENOMEM; | |
bc5fc7e4 | 1182 | |
d29ec824 CH |
1183 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, |
1184 | sizeof(struct nvme_id_ctrl)); | |
1185 | if (error) | |
1186 | kfree(*id); | |
1187 | return error; | |
1188 | } | |
1189 | ||
1190 | int nvme_identify_ns(struct nvme_dev *dev, unsigned nsid, | |
1191 | struct nvme_id_ns **id) | |
1192 | { | |
e44ac588 | 1193 | struct nvme_command c = { }; |
d29ec824 | 1194 | int error; |
bc5fc7e4 | 1195 | |
e44ac588 AM |
1196 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ |
1197 | c.identify.opcode = nvme_admin_identify, | |
1198 | c.identify.nsid = cpu_to_le32(nsid), | |
1199 | ||
d29ec824 CH |
1200 | *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL); |
1201 | if (!*id) | |
1202 | return -ENOMEM; | |
1203 | ||
1204 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
1205 | sizeof(struct nvme_id_ns)); | |
1206 | if (error) | |
1207 | kfree(*id); | |
1208 | return error; | |
bc5fc7e4 MW |
1209 | } |
1210 | ||
5d0f6131 | 1211 | int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid, |
08df1e05 | 1212 | dma_addr_t dma_addr, u32 *result) |
bc5fc7e4 MW |
1213 | { |
1214 | struct nvme_command c; | |
1215 | ||
1216 | memset(&c, 0, sizeof(c)); | |
1217 | c.features.opcode = nvme_admin_get_features; | |
a42cecce | 1218 | c.features.nsid = cpu_to_le32(nsid); |
bc5fc7e4 MW |
1219 | c.features.prp1 = cpu_to_le64(dma_addr); |
1220 | c.features.fid = cpu_to_le32(fid); | |
bc5fc7e4 | 1221 | |
d29ec824 CH |
1222 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0, |
1223 | result, 0); | |
df348139 MW |
1224 | } |
1225 | ||
5d0f6131 VV |
1226 | int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11, |
1227 | dma_addr_t dma_addr, u32 *result) | |
df348139 MW |
1228 | { |
1229 | struct nvme_command c; | |
1230 | ||
1231 | memset(&c, 0, sizeof(c)); | |
1232 | c.features.opcode = nvme_admin_set_features; | |
1233 | c.features.prp1 = cpu_to_le64(dma_addr); | |
1234 | c.features.fid = cpu_to_le32(fid); | |
1235 | c.features.dword11 = cpu_to_le32(dword11); | |
1236 | ||
d29ec824 CH |
1237 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0, |
1238 | result, 0); | |
1239 | } | |
1240 | ||
1241 | int nvme_get_log_page(struct nvme_dev *dev, struct nvme_smart_log **log) | |
1242 | { | |
e44ac588 AM |
1243 | struct nvme_command c = { }; |
1244 | int error; | |
1245 | ||
1246 | c.common.opcode = nvme_admin_get_log_page, | |
1247 | c.common.nsid = cpu_to_le32(0xFFFFFFFF), | |
1248 | c.common.cdw10[0] = cpu_to_le32( | |
d29ec824 CH |
1249 | (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) | |
1250 | NVME_LOG_SMART), | |
d29ec824 CH |
1251 | |
1252 | *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL); | |
1253 | if (!*log) | |
1254 | return -ENOMEM; | |
1255 | ||
1256 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *log, | |
1257 | sizeof(struct nvme_smart_log)); | |
1258 | if (error) | |
1259 | kfree(*log); | |
1260 | return error; | |
bc5fc7e4 MW |
1261 | } |
1262 | ||
c30341dc | 1263 | /** |
a4aea562 | 1264 | * nvme_abort_req - Attempt aborting a request |
c30341dc KB |
1265 | * |
1266 | * Schedule controller reset if the command was already aborted once before and | |
1267 | * still hasn't been returned to the driver, or if this is the admin queue. | |
1268 | */ | |
a4aea562 | 1269 | static void nvme_abort_req(struct request *req) |
c30341dc | 1270 | { |
a4aea562 MB |
1271 | struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req); |
1272 | struct nvme_queue *nvmeq = cmd_rq->nvmeq; | |
c30341dc | 1273 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 MB |
1274 | struct request *abort_req; |
1275 | struct nvme_cmd_info *abort_cmd; | |
1276 | struct nvme_command cmd; | |
c30341dc | 1277 | |
a4aea562 | 1278 | if (!nvmeq->qid || cmd_rq->aborted) { |
7a509a6b KB |
1279 | unsigned long flags; |
1280 | ||
1281 | spin_lock_irqsave(&dev_list_lock, flags); | |
c30341dc | 1282 | if (work_busy(&dev->reset_work)) |
7a509a6b | 1283 | goto out; |
c30341dc | 1284 | list_del_init(&dev->node); |
e75ec752 | 1285 | dev_warn(dev->dev, "I/O %d QID %d timeout, reset controller\n", |
a4aea562 | 1286 | req->tag, nvmeq->qid); |
9ca97374 | 1287 | dev->reset_workfn = nvme_reset_failed_dev; |
c30341dc | 1288 | queue_work(nvme_workq, &dev->reset_work); |
7a509a6b KB |
1289 | out: |
1290 | spin_unlock_irqrestore(&dev_list_lock, flags); | |
c30341dc KB |
1291 | return; |
1292 | } | |
1293 | ||
1294 | if (!dev->abort_limit) | |
1295 | return; | |
1296 | ||
a4aea562 MB |
1297 | abort_req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, |
1298 | false); | |
9f173b33 | 1299 | if (IS_ERR(abort_req)) |
c30341dc KB |
1300 | return; |
1301 | ||
a4aea562 MB |
1302 | abort_cmd = blk_mq_rq_to_pdu(abort_req); |
1303 | nvme_set_info(abort_cmd, abort_req, abort_completion); | |
1304 | ||
c30341dc KB |
1305 | memset(&cmd, 0, sizeof(cmd)); |
1306 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
a4aea562 | 1307 | cmd.abort.cid = req->tag; |
c30341dc | 1308 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); |
a4aea562 | 1309 | cmd.abort.command_id = abort_req->tag; |
c30341dc KB |
1310 | |
1311 | --dev->abort_limit; | |
a4aea562 | 1312 | cmd_rq->aborted = 1; |
c30341dc | 1313 | |
a4aea562 | 1314 | dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag, |
c30341dc | 1315 | nvmeq->qid); |
e3f879bf | 1316 | nvme_submit_cmd(dev->queues[0], &cmd); |
c30341dc KB |
1317 | } |
1318 | ||
42483228 | 1319 | static void nvme_cancel_queue_ios(struct request *req, void *data, bool reserved) |
a09115b2 | 1320 | { |
a4aea562 MB |
1321 | struct nvme_queue *nvmeq = data; |
1322 | void *ctx; | |
1323 | nvme_completion_fn fn; | |
1324 | struct nvme_cmd_info *cmd; | |
cef6a948 KB |
1325 | struct nvme_completion cqe; |
1326 | ||
1327 | if (!blk_mq_request_started(req)) | |
1328 | return; | |
a09115b2 | 1329 | |
a4aea562 | 1330 | cmd = blk_mq_rq_to_pdu(req); |
a09115b2 | 1331 | |
a4aea562 MB |
1332 | if (cmd->ctx == CMD_CTX_CANCELLED) |
1333 | return; | |
1334 | ||
cef6a948 KB |
1335 | if (blk_queue_dying(req->q)) |
1336 | cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1); | |
1337 | else | |
1338 | cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1); | |
1339 | ||
1340 | ||
a4aea562 MB |
1341 | dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n", |
1342 | req->tag, nvmeq->qid); | |
1343 | ctx = cancel_cmd_info(cmd, &fn); | |
1344 | fn(nvmeq, ctx, &cqe); | |
a09115b2 MW |
1345 | } |
1346 | ||
a4aea562 | 1347 | static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) |
9e866774 | 1348 | { |
a4aea562 MB |
1349 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); |
1350 | struct nvme_queue *nvmeq = cmd->nvmeq; | |
1351 | ||
1352 | dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag, | |
1353 | nvmeq->qid); | |
7a509a6b | 1354 | spin_lock_irq(&nvmeq->q_lock); |
07836e65 | 1355 | nvme_abort_req(req); |
7a509a6b | 1356 | spin_unlock_irq(&nvmeq->q_lock); |
a4aea562 | 1357 | |
07836e65 KB |
1358 | /* |
1359 | * The aborted req will be completed on receiving the abort req. | |
1360 | * We enable the timer again. If hit twice, it'll cause a device reset, | |
1361 | * as the device then is in a faulty state. | |
1362 | */ | |
1363 | return BLK_EH_RESET_TIMER; | |
a4aea562 | 1364 | } |
22404274 | 1365 | |
a4aea562 MB |
1366 | static void nvme_free_queue(struct nvme_queue *nvmeq) |
1367 | { | |
9e866774 MW |
1368 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), |
1369 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
8ffaadf7 JD |
1370 | if (nvmeq->sq_cmds) |
1371 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
9e866774 MW |
1372 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); |
1373 | kfree(nvmeq); | |
1374 | } | |
1375 | ||
a1a5ef99 | 1376 | static void nvme_free_queues(struct nvme_dev *dev, int lowest) |
22404274 KB |
1377 | { |
1378 | int i; | |
1379 | ||
a1a5ef99 | 1380 | for (i = dev->queue_count - 1; i >= lowest; i--) { |
a4aea562 | 1381 | struct nvme_queue *nvmeq = dev->queues[i]; |
22404274 | 1382 | dev->queue_count--; |
a4aea562 | 1383 | dev->queues[i] = NULL; |
f435c282 | 1384 | nvme_free_queue(nvmeq); |
121c7ad4 | 1385 | } |
22404274 KB |
1386 | } |
1387 | ||
4d115420 KB |
1388 | /** |
1389 | * nvme_suspend_queue - put queue into suspended state | |
1390 | * @nvmeq - queue to suspend | |
4d115420 KB |
1391 | */ |
1392 | static int nvme_suspend_queue(struct nvme_queue *nvmeq) | |
b60503ba | 1393 | { |
2b25d981 | 1394 | int vector; |
b60503ba | 1395 | |
a09115b2 | 1396 | spin_lock_irq(&nvmeq->q_lock); |
2b25d981 KB |
1397 | if (nvmeq->cq_vector == -1) { |
1398 | spin_unlock_irq(&nvmeq->q_lock); | |
1399 | return 1; | |
1400 | } | |
1401 | vector = nvmeq->dev->entry[nvmeq->cq_vector].vector; | |
42f61420 | 1402 | nvmeq->dev->online_queues--; |
2b25d981 | 1403 | nvmeq->cq_vector = -1; |
a09115b2 MW |
1404 | spin_unlock_irq(&nvmeq->q_lock); |
1405 | ||
6df3dbc8 KB |
1406 | if (!nvmeq->qid && nvmeq->dev->admin_q) |
1407 | blk_mq_freeze_queue_start(nvmeq->dev->admin_q); | |
1408 | ||
aba2080f MW |
1409 | irq_set_affinity_hint(vector, NULL); |
1410 | free_irq(vector, nvmeq); | |
b60503ba | 1411 | |
4d115420 KB |
1412 | return 0; |
1413 | } | |
b60503ba | 1414 | |
4d115420 KB |
1415 | static void nvme_clear_queue(struct nvme_queue *nvmeq) |
1416 | { | |
22404274 | 1417 | spin_lock_irq(&nvmeq->q_lock); |
42483228 KB |
1418 | if (nvmeq->tags && *nvmeq->tags) |
1419 | blk_mq_all_tag_busy_iter(*nvmeq->tags, nvme_cancel_queue_ios, nvmeq); | |
22404274 | 1420 | spin_unlock_irq(&nvmeq->q_lock); |
b60503ba MW |
1421 | } |
1422 | ||
4d115420 KB |
1423 | static void nvme_disable_queue(struct nvme_dev *dev, int qid) |
1424 | { | |
a4aea562 | 1425 | struct nvme_queue *nvmeq = dev->queues[qid]; |
4d115420 KB |
1426 | |
1427 | if (!nvmeq) | |
1428 | return; | |
1429 | if (nvme_suspend_queue(nvmeq)) | |
1430 | return; | |
1431 | ||
0e53d180 KB |
1432 | /* Don't tell the adapter to delete the admin queue. |
1433 | * Don't tell a removed adapter to delete IO queues. */ | |
1434 | if (qid && readl(&dev->bar->csts) != -1) { | |
b60503ba MW |
1435 | adapter_delete_sq(dev, qid); |
1436 | adapter_delete_cq(dev, qid); | |
1437 | } | |
07836e65 KB |
1438 | |
1439 | spin_lock_irq(&nvmeq->q_lock); | |
1440 | nvme_process_cq(nvmeq); | |
1441 | spin_unlock_irq(&nvmeq->q_lock); | |
b60503ba MW |
1442 | } |
1443 | ||
8ffaadf7 JD |
1444 | static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, |
1445 | int entry_size) | |
1446 | { | |
1447 | int q_depth = dev->q_depth; | |
1448 | unsigned q_size_aligned = roundup(q_depth * entry_size, dev->page_size); | |
1449 | ||
1450 | if (q_size_aligned * nr_io_queues > dev->cmb_size) { | |
c45f5c99 JD |
1451 | u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); |
1452 | mem_per_q = round_down(mem_per_q, dev->page_size); | |
1453 | q_depth = div_u64(mem_per_q, entry_size); | |
8ffaadf7 JD |
1454 | |
1455 | /* | |
1456 | * Ensure the reduced q_depth is above some threshold where it | |
1457 | * would be better to map queues in system memory with the | |
1458 | * original depth | |
1459 | */ | |
1460 | if (q_depth < 64) | |
1461 | return -ENOMEM; | |
1462 | } | |
1463 | ||
1464 | return q_depth; | |
1465 | } | |
1466 | ||
1467 | static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, | |
1468 | int qid, int depth) | |
1469 | { | |
1470 | if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) { | |
1471 | unsigned offset = (qid - 1) * | |
1472 | roundup(SQ_SIZE(depth), dev->page_size); | |
1473 | nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset; | |
1474 | nvmeq->sq_cmds_io = dev->cmb + offset; | |
1475 | } else { | |
1476 | nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth), | |
1477 | &nvmeq->sq_dma_addr, GFP_KERNEL); | |
1478 | if (!nvmeq->sq_cmds) | |
1479 | return -ENOMEM; | |
1480 | } | |
1481 | ||
1482 | return 0; | |
1483 | } | |
1484 | ||
b60503ba | 1485 | static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, |
2b25d981 | 1486 | int depth) |
b60503ba | 1487 | { |
a4aea562 | 1488 | struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL); |
b60503ba MW |
1489 | if (!nvmeq) |
1490 | return NULL; | |
1491 | ||
e75ec752 | 1492 | nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth), |
4d51abf9 | 1493 | &nvmeq->cq_dma_addr, GFP_KERNEL); |
b60503ba MW |
1494 | if (!nvmeq->cqes) |
1495 | goto free_nvmeq; | |
b60503ba | 1496 | |
8ffaadf7 | 1497 | if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth)) |
b60503ba MW |
1498 | goto free_cqdma; |
1499 | ||
e75ec752 | 1500 | nvmeq->q_dmadev = dev->dev; |
091b6092 | 1501 | nvmeq->dev = dev; |
3193f07b MW |
1502 | snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d", |
1503 | dev->instance, qid); | |
b60503ba MW |
1504 | spin_lock_init(&nvmeq->q_lock); |
1505 | nvmeq->cq_head = 0; | |
82123460 | 1506 | nvmeq->cq_phase = 1; |
b80d5ccc | 1507 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
b60503ba | 1508 | nvmeq->q_depth = depth; |
c30341dc | 1509 | nvmeq->qid = qid; |
758dd7fd | 1510 | nvmeq->cq_vector = -1; |
a4aea562 | 1511 | dev->queues[qid] = nvmeq; |
b60503ba | 1512 | |
36a7e993 JD |
1513 | /* make sure queue descriptor is set before queue count, for kthread */ |
1514 | mb(); | |
1515 | dev->queue_count++; | |
1516 | ||
b60503ba MW |
1517 | return nvmeq; |
1518 | ||
1519 | free_cqdma: | |
e75ec752 | 1520 | dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes, |
b60503ba MW |
1521 | nvmeq->cq_dma_addr); |
1522 | free_nvmeq: | |
1523 | kfree(nvmeq); | |
1524 | return NULL; | |
1525 | } | |
1526 | ||
3001082c MW |
1527 | static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq, |
1528 | const char *name) | |
1529 | { | |
58ffacb5 MW |
1530 | if (use_threaded_interrupts) |
1531 | return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector, | |
481e5bad | 1532 | nvme_irq_check, nvme_irq, IRQF_SHARED, |
58ffacb5 | 1533 | name, nvmeq); |
3001082c | 1534 | return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq, |
481e5bad | 1535 | IRQF_SHARED, name, nvmeq); |
3001082c MW |
1536 | } |
1537 | ||
22404274 | 1538 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1539 | { |
22404274 | 1540 | struct nvme_dev *dev = nvmeq->dev; |
b60503ba | 1541 | |
7be50e93 | 1542 | spin_lock_irq(&nvmeq->q_lock); |
22404274 KB |
1543 | nvmeq->sq_tail = 0; |
1544 | nvmeq->cq_head = 0; | |
1545 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1546 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
22404274 | 1547 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); |
42f61420 | 1548 | dev->online_queues++; |
7be50e93 | 1549 | spin_unlock_irq(&nvmeq->q_lock); |
22404274 KB |
1550 | } |
1551 | ||
1552 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) | |
1553 | { | |
1554 | struct nvme_dev *dev = nvmeq->dev; | |
1555 | int result; | |
3f85d50b | 1556 | |
2b25d981 | 1557 | nvmeq->cq_vector = qid - 1; |
b60503ba MW |
1558 | result = adapter_alloc_cq(dev, qid, nvmeq); |
1559 | if (result < 0) | |
22404274 | 1560 | return result; |
b60503ba MW |
1561 | |
1562 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1563 | if (result < 0) | |
1564 | goto release_cq; | |
1565 | ||
3193f07b | 1566 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
b60503ba MW |
1567 | if (result < 0) |
1568 | goto release_sq; | |
1569 | ||
22404274 | 1570 | nvme_init_queue(nvmeq, qid); |
22404274 | 1571 | return result; |
b60503ba MW |
1572 | |
1573 | release_sq: | |
1574 | adapter_delete_sq(dev, qid); | |
1575 | release_cq: | |
1576 | adapter_delete_cq(dev, qid); | |
22404274 | 1577 | return result; |
b60503ba MW |
1578 | } |
1579 | ||
ba47e386 MW |
1580 | static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled) |
1581 | { | |
1582 | unsigned long timeout; | |
1583 | u32 bit = enabled ? NVME_CSTS_RDY : 0; | |
1584 | ||
1585 | timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
1586 | ||
1587 | while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) { | |
1588 | msleep(100); | |
1589 | if (fatal_signal_pending(current)) | |
1590 | return -EINTR; | |
1591 | if (time_after(jiffies, timeout)) { | |
e75ec752 | 1592 | dev_err(dev->dev, |
27e8166c MW |
1593 | "Device not ready; aborting %s\n", enabled ? |
1594 | "initialisation" : "reset"); | |
ba47e386 MW |
1595 | return -ENODEV; |
1596 | } | |
1597 | } | |
1598 | ||
1599 | return 0; | |
1600 | } | |
1601 | ||
1602 | /* | |
1603 | * If the device has been passed off to us in an enabled state, just clear | |
1604 | * the enabled bit. The spec says we should set the 'shutdown notification | |
1605 | * bits', but doing so may cause the device to complete commands to the | |
1606 | * admin queue ... and we don't know what memory that might be pointing at! | |
1607 | */ | |
1608 | static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap) | |
1609 | { | |
01079522 DM |
1610 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1611 | dev->ctrl_config &= ~NVME_CC_ENABLE; | |
1612 | writel(dev->ctrl_config, &dev->bar->cc); | |
44af146a | 1613 | |
ba47e386 MW |
1614 | return nvme_wait_ready(dev, cap, false); |
1615 | } | |
1616 | ||
1617 | static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap) | |
1618 | { | |
01079522 DM |
1619 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1620 | dev->ctrl_config |= NVME_CC_ENABLE; | |
1621 | writel(dev->ctrl_config, &dev->bar->cc); | |
1622 | ||
ba47e386 MW |
1623 | return nvme_wait_ready(dev, cap, true); |
1624 | } | |
1625 | ||
1894d8f1 KB |
1626 | static int nvme_shutdown_ctrl(struct nvme_dev *dev) |
1627 | { | |
1628 | unsigned long timeout; | |
1894d8f1 | 1629 | |
01079522 DM |
1630 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1631 | dev->ctrl_config |= NVME_CC_SHN_NORMAL; | |
1632 | ||
1633 | writel(dev->ctrl_config, &dev->bar->cc); | |
1894d8f1 | 1634 | |
2484f407 | 1635 | timeout = SHUTDOWN_TIMEOUT + jiffies; |
1894d8f1 KB |
1636 | while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) != |
1637 | NVME_CSTS_SHST_CMPLT) { | |
1638 | msleep(100); | |
1639 | if (fatal_signal_pending(current)) | |
1640 | return -EINTR; | |
1641 | if (time_after(jiffies, timeout)) { | |
e75ec752 | 1642 | dev_err(dev->dev, |
1894d8f1 KB |
1643 | "Device shutdown incomplete; abort shutdown\n"); |
1644 | return -ENODEV; | |
1645 | } | |
1646 | } | |
1647 | ||
1648 | return 0; | |
1649 | } | |
1650 | ||
a4aea562 | 1651 | static struct blk_mq_ops nvme_mq_admin_ops = { |
d29ec824 | 1652 | .queue_rq = nvme_queue_rq, |
a4aea562 MB |
1653 | .map_queue = blk_mq_map_queue, |
1654 | .init_hctx = nvme_admin_init_hctx, | |
4af0e21c | 1655 | .exit_hctx = nvme_admin_exit_hctx, |
a4aea562 MB |
1656 | .init_request = nvme_admin_init_request, |
1657 | .timeout = nvme_timeout, | |
1658 | }; | |
1659 | ||
1660 | static struct blk_mq_ops nvme_mq_ops = { | |
1661 | .queue_rq = nvme_queue_rq, | |
1662 | .map_queue = blk_mq_map_queue, | |
1663 | .init_hctx = nvme_init_hctx, | |
1664 | .init_request = nvme_init_request, | |
1665 | .timeout = nvme_timeout, | |
1666 | }; | |
1667 | ||
ea191d2f KB |
1668 | static void nvme_dev_remove_admin(struct nvme_dev *dev) |
1669 | { | |
1670 | if (dev->admin_q && !blk_queue_dying(dev->admin_q)) { | |
1671 | blk_cleanup_queue(dev->admin_q); | |
1672 | blk_mq_free_tag_set(&dev->admin_tagset); | |
1673 | } | |
1674 | } | |
1675 | ||
a4aea562 MB |
1676 | static int nvme_alloc_admin_tags(struct nvme_dev *dev) |
1677 | { | |
1678 | if (!dev->admin_q) { | |
1679 | dev->admin_tagset.ops = &nvme_mq_admin_ops; | |
1680 | dev->admin_tagset.nr_hw_queues = 1; | |
1681 | dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1; | |
1efccc9d | 1682 | dev->admin_tagset.reserved_tags = 1; |
a4aea562 | 1683 | dev->admin_tagset.timeout = ADMIN_TIMEOUT; |
e75ec752 | 1684 | dev->admin_tagset.numa_node = dev_to_node(dev->dev); |
ac3dd5bd | 1685 | dev->admin_tagset.cmd_size = nvme_cmd_size(dev); |
a4aea562 MB |
1686 | dev->admin_tagset.driver_data = dev; |
1687 | ||
1688 | if (blk_mq_alloc_tag_set(&dev->admin_tagset)) | |
1689 | return -ENOMEM; | |
1690 | ||
1691 | dev->admin_q = blk_mq_init_queue(&dev->admin_tagset); | |
35b489d3 | 1692 | if (IS_ERR(dev->admin_q)) { |
a4aea562 MB |
1693 | blk_mq_free_tag_set(&dev->admin_tagset); |
1694 | return -ENOMEM; | |
1695 | } | |
ea191d2f KB |
1696 | if (!blk_get_queue(dev->admin_q)) { |
1697 | nvme_dev_remove_admin(dev); | |
4af0e21c | 1698 | dev->admin_q = NULL; |
ea191d2f KB |
1699 | return -ENODEV; |
1700 | } | |
0fb59cbc KB |
1701 | } else |
1702 | blk_mq_unfreeze_queue(dev->admin_q); | |
a4aea562 MB |
1703 | |
1704 | return 0; | |
1705 | } | |
1706 | ||
8d85fce7 | 1707 | static int nvme_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1708 | { |
ba47e386 | 1709 | int result; |
b60503ba | 1710 | u32 aqa; |
ba47e386 | 1711 | u64 cap = readq(&dev->bar->cap); |
b60503ba | 1712 | struct nvme_queue *nvmeq; |
1d090624 KB |
1713 | unsigned page_shift = PAGE_SHIFT; |
1714 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12; | |
1715 | unsigned dev_page_max = NVME_CAP_MPSMAX(cap) + 12; | |
1716 | ||
1717 | if (page_shift < dev_page_min) { | |
e75ec752 | 1718 | dev_err(dev->dev, |
1d090624 KB |
1719 | "Minimum device page size (%u) too large for " |
1720 | "host (%u)\n", 1 << dev_page_min, | |
1721 | 1 << page_shift); | |
1722 | return -ENODEV; | |
1723 | } | |
1724 | if (page_shift > dev_page_max) { | |
e75ec752 | 1725 | dev_info(dev->dev, |
1d090624 KB |
1726 | "Device maximum page size (%u) smaller than " |
1727 | "host (%u); enabling work-around\n", | |
1728 | 1 << dev_page_max, 1 << page_shift); | |
1729 | page_shift = dev_page_max; | |
1730 | } | |
b60503ba | 1731 | |
dfbac8c7 KB |
1732 | dev->subsystem = readl(&dev->bar->vs) >= NVME_VS(1, 1) ? |
1733 | NVME_CAP_NSSRC(cap) : 0; | |
1734 | ||
1735 | if (dev->subsystem && (readl(&dev->bar->csts) & NVME_CSTS_NSSRO)) | |
1736 | writel(NVME_CSTS_NSSRO, &dev->bar->csts); | |
1737 | ||
ba47e386 MW |
1738 | result = nvme_disable_ctrl(dev, cap); |
1739 | if (result < 0) | |
1740 | return result; | |
b60503ba | 1741 | |
a4aea562 | 1742 | nvmeq = dev->queues[0]; |
cd638946 | 1743 | if (!nvmeq) { |
2b25d981 | 1744 | nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); |
cd638946 KB |
1745 | if (!nvmeq) |
1746 | return -ENOMEM; | |
cd638946 | 1747 | } |
b60503ba MW |
1748 | |
1749 | aqa = nvmeq->q_depth - 1; | |
1750 | aqa |= aqa << 16; | |
1751 | ||
1d090624 KB |
1752 | dev->page_size = 1 << page_shift; |
1753 | ||
01079522 | 1754 | dev->ctrl_config = NVME_CC_CSS_NVM; |
1d090624 | 1755 | dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; |
b60503ba | 1756 | dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; |
7f53f9d2 | 1757 | dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
b60503ba MW |
1758 | |
1759 | writel(aqa, &dev->bar->aqa); | |
1760 | writeq(nvmeq->sq_dma_addr, &dev->bar->asq); | |
1761 | writeq(nvmeq->cq_dma_addr, &dev->bar->acq); | |
b60503ba | 1762 | |
ba47e386 | 1763 | result = nvme_enable_ctrl(dev, cap); |
025c557a | 1764 | if (result) |
a4aea562 MB |
1765 | goto free_nvmeq; |
1766 | ||
2b25d981 | 1767 | nvmeq->cq_vector = 0; |
3193f07b | 1768 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
758dd7fd JD |
1769 | if (result) { |
1770 | nvmeq->cq_vector = -1; | |
0fb59cbc | 1771 | goto free_nvmeq; |
758dd7fd | 1772 | } |
025c557a | 1773 | |
b60503ba | 1774 | return result; |
a4aea562 | 1775 | |
a4aea562 MB |
1776 | free_nvmeq: |
1777 | nvme_free_queues(dev, 0); | |
1778 | return result; | |
b60503ba MW |
1779 | } |
1780 | ||
a53295b6 MW |
1781 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
1782 | { | |
1783 | struct nvme_dev *dev = ns->dev; | |
a53295b6 MW |
1784 | struct nvme_user_io io; |
1785 | struct nvme_command c; | |
d29ec824 | 1786 | unsigned length, meta_len; |
a67a9513 | 1787 | int status, write; |
a67a9513 KB |
1788 | dma_addr_t meta_dma = 0; |
1789 | void *meta = NULL; | |
fec558b5 | 1790 | void __user *metadata; |
a53295b6 MW |
1791 | |
1792 | if (copy_from_user(&io, uio, sizeof(io))) | |
1793 | return -EFAULT; | |
6c7d4945 MW |
1794 | |
1795 | switch (io.opcode) { | |
1796 | case nvme_cmd_write: | |
1797 | case nvme_cmd_read: | |
6bbf1acd | 1798 | case nvme_cmd_compare: |
6413214c | 1799 | break; |
6c7d4945 | 1800 | default: |
6bbf1acd | 1801 | return -EINVAL; |
6c7d4945 MW |
1802 | } |
1803 | ||
d29ec824 CH |
1804 | length = (io.nblocks + 1) << ns->lba_shift; |
1805 | meta_len = (io.nblocks + 1) * ns->ms; | |
6a398a3e | 1806 | metadata = (void __user *)(unsigned long)io.metadata; |
d29ec824 | 1807 | write = io.opcode & 1; |
a53295b6 | 1808 | |
71feb364 KB |
1809 | if (ns->ext) { |
1810 | length += meta_len; | |
1811 | meta_len = 0; | |
a67a9513 KB |
1812 | } |
1813 | if (meta_len) { | |
d29ec824 CH |
1814 | if (((io.metadata & 3) || !io.metadata) && !ns->ext) |
1815 | return -EINVAL; | |
1816 | ||
e75ec752 | 1817 | meta = dma_alloc_coherent(dev->dev, meta_len, |
a67a9513 | 1818 | &meta_dma, GFP_KERNEL); |
fec558b5 | 1819 | |
a67a9513 KB |
1820 | if (!meta) { |
1821 | status = -ENOMEM; | |
1822 | goto unmap; | |
1823 | } | |
1824 | if (write) { | |
fec558b5 | 1825 | if (copy_from_user(meta, metadata, meta_len)) { |
a67a9513 KB |
1826 | status = -EFAULT; |
1827 | goto unmap; | |
1828 | } | |
1829 | } | |
1830 | } | |
1831 | ||
a53295b6 MW |
1832 | memset(&c, 0, sizeof(c)); |
1833 | c.rw.opcode = io.opcode; | |
1834 | c.rw.flags = io.flags; | |
6c7d4945 | 1835 | c.rw.nsid = cpu_to_le32(ns->ns_id); |
a53295b6 | 1836 | c.rw.slba = cpu_to_le64(io.slba); |
6c7d4945 | 1837 | c.rw.length = cpu_to_le16(io.nblocks); |
a53295b6 | 1838 | c.rw.control = cpu_to_le16(io.control); |
1c9b5265 MW |
1839 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); |
1840 | c.rw.reftag = cpu_to_le32(io.reftag); | |
1841 | c.rw.apptag = cpu_to_le16(io.apptag); | |
1842 | c.rw.appmask = cpu_to_le16(io.appmask); | |
a67a9513 | 1843 | c.rw.metadata = cpu_to_le64(meta_dma); |
d29ec824 CH |
1844 | |
1845 | status = __nvme_submit_sync_cmd(ns->queue, &c, NULL, | |
1846 | (void __user *)io.addr, length, NULL, 0); | |
f410c680 | 1847 | unmap: |
a67a9513 KB |
1848 | if (meta) { |
1849 | if (status == NVME_SC_SUCCESS && !write) { | |
fec558b5 | 1850 | if (copy_to_user(metadata, meta, meta_len)) |
a67a9513 KB |
1851 | status = -EFAULT; |
1852 | } | |
e75ec752 | 1853 | dma_free_coherent(dev->dev, meta_len, meta, meta_dma); |
f410c680 | 1854 | } |
a53295b6 MW |
1855 | return status; |
1856 | } | |
1857 | ||
a4aea562 MB |
1858 | static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns, |
1859 | struct nvme_passthru_cmd __user *ucmd) | |
6ee44cdc | 1860 | { |
7963e521 | 1861 | struct nvme_passthru_cmd cmd; |
6ee44cdc | 1862 | struct nvme_command c; |
d29ec824 CH |
1863 | unsigned timeout = 0; |
1864 | int status; | |
6ee44cdc | 1865 | |
6bbf1acd MW |
1866 | if (!capable(CAP_SYS_ADMIN)) |
1867 | return -EACCES; | |
1868 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
6ee44cdc | 1869 | return -EFAULT; |
6ee44cdc MW |
1870 | |
1871 | memset(&c, 0, sizeof(c)); | |
6bbf1acd MW |
1872 | c.common.opcode = cmd.opcode; |
1873 | c.common.flags = cmd.flags; | |
1874 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1875 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1876 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1877 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
1878 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
1879 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
1880 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
1881 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
1882 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
1883 | ||
d29ec824 CH |
1884 | if (cmd.timeout_ms) |
1885 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
eca18b23 | 1886 | |
f705f837 | 1887 | status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c, |
d29ec824 CH |
1888 | NULL, (void __user *)cmd.addr, cmd.data_len, |
1889 | &cmd.result, timeout); | |
1890 | if (status >= 0) { | |
1891 | if (put_user(cmd.result, &ucmd->result)) | |
1892 | return -EFAULT; | |
6bbf1acd | 1893 | } |
f4f117f6 | 1894 | |
6ee44cdc MW |
1895 | return status; |
1896 | } | |
1897 | ||
81f03fed JD |
1898 | static int nvme_subsys_reset(struct nvme_dev *dev) |
1899 | { | |
1900 | if (!dev->subsystem) | |
1901 | return -ENOTTY; | |
1902 | ||
1903 | writel(0x4E564D65, &dev->bar->nssr); /* "NVMe" */ | |
1904 | return 0; | |
1905 | } | |
1906 | ||
b60503ba MW |
1907 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, |
1908 | unsigned long arg) | |
1909 | { | |
1910 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1911 | ||
1912 | switch (cmd) { | |
6bbf1acd | 1913 | case NVME_IOCTL_ID: |
c3bfe717 | 1914 | force_successful_syscall_return(); |
6bbf1acd MW |
1915 | return ns->ns_id; |
1916 | case NVME_IOCTL_ADMIN_CMD: | |
a4aea562 | 1917 | return nvme_user_cmd(ns->dev, NULL, (void __user *)arg); |
7963e521 | 1918 | case NVME_IOCTL_IO_CMD: |
a4aea562 | 1919 | return nvme_user_cmd(ns->dev, ns, (void __user *)arg); |
a53295b6 MW |
1920 | case NVME_IOCTL_SUBMIT_IO: |
1921 | return nvme_submit_io(ns, (void __user *)arg); | |
5d0f6131 VV |
1922 | case SG_GET_VERSION_NUM: |
1923 | return nvme_sg_get_version_num((void __user *)arg); | |
1924 | case SG_IO: | |
1925 | return nvme_sg_io(ns, (void __user *)arg); | |
b60503ba MW |
1926 | default: |
1927 | return -ENOTTY; | |
1928 | } | |
1929 | } | |
1930 | ||
320a3827 KB |
1931 | #ifdef CONFIG_COMPAT |
1932 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1933 | unsigned int cmd, unsigned long arg) | |
1934 | { | |
320a3827 KB |
1935 | switch (cmd) { |
1936 | case SG_IO: | |
e179729a | 1937 | return -ENOIOCTLCMD; |
320a3827 KB |
1938 | } |
1939 | return nvme_ioctl(bdev, mode, cmd, arg); | |
1940 | } | |
1941 | #else | |
1942 | #define nvme_compat_ioctl NULL | |
1943 | #endif | |
1944 | ||
9ac27090 KB |
1945 | static int nvme_open(struct block_device *bdev, fmode_t mode) |
1946 | { | |
9e60352c KB |
1947 | int ret = 0; |
1948 | struct nvme_ns *ns; | |
9ac27090 | 1949 | |
9e60352c KB |
1950 | spin_lock(&dev_list_lock); |
1951 | ns = bdev->bd_disk->private_data; | |
1952 | if (!ns) | |
1953 | ret = -ENXIO; | |
1954 | else if (!kref_get_unless_zero(&ns->dev->kref)) | |
1955 | ret = -ENXIO; | |
1956 | spin_unlock(&dev_list_lock); | |
1957 | ||
1958 | return ret; | |
9ac27090 KB |
1959 | } |
1960 | ||
1961 | static void nvme_free_dev(struct kref *kref); | |
1962 | ||
1963 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
1964 | { | |
1965 | struct nvme_ns *ns = disk->private_data; | |
1966 | struct nvme_dev *dev = ns->dev; | |
1967 | ||
1968 | kref_put(&dev->kref, nvme_free_dev); | |
1969 | } | |
1970 | ||
4cc09e2d KB |
1971 | static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo) |
1972 | { | |
1973 | /* some standard values */ | |
1974 | geo->heads = 1 << 6; | |
1975 | geo->sectors = 1 << 5; | |
1976 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
1977 | return 0; | |
1978 | } | |
1979 | ||
e1e5e564 KB |
1980 | static void nvme_config_discard(struct nvme_ns *ns) |
1981 | { | |
1982 | u32 logical_block_size = queue_logical_block_size(ns->queue); | |
1983 | ns->queue->limits.discard_zeroes_data = 0; | |
1984 | ns->queue->limits.discard_alignment = logical_block_size; | |
1985 | ns->queue->limits.discard_granularity = logical_block_size; | |
2bb4cd5c | 1986 | blk_queue_max_discard_sectors(ns->queue, 0xffffffff); |
e1e5e564 KB |
1987 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); |
1988 | } | |
1989 | ||
1b9dbf7f KB |
1990 | static int nvme_revalidate_disk(struct gendisk *disk) |
1991 | { | |
1992 | struct nvme_ns *ns = disk->private_data; | |
1993 | struct nvme_dev *dev = ns->dev; | |
1994 | struct nvme_id_ns *id; | |
a67a9513 KB |
1995 | u8 lbaf, pi_type; |
1996 | u16 old_ms; | |
e1e5e564 | 1997 | unsigned short bs; |
1b9dbf7f | 1998 | |
d29ec824 | 1999 | if (nvme_identify_ns(dev, ns->ns_id, &id)) { |
a5768aa8 KB |
2000 | dev_warn(dev->dev, "%s: Identify failure nvme%dn%d\n", __func__, |
2001 | dev->instance, ns->ns_id); | |
2002 | return -ENODEV; | |
1b9dbf7f | 2003 | } |
a5768aa8 KB |
2004 | if (id->ncap == 0) { |
2005 | kfree(id); | |
2006 | return -ENODEV; | |
e1e5e564 | 2007 | } |
1b9dbf7f | 2008 | |
e1e5e564 KB |
2009 | old_ms = ns->ms; |
2010 | lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK; | |
1b9dbf7f | 2011 | ns->lba_shift = id->lbaf[lbaf].ds; |
e1e5e564 | 2012 | ns->ms = le16_to_cpu(id->lbaf[lbaf].ms); |
a67a9513 | 2013 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); |
e1e5e564 KB |
2014 | |
2015 | /* | |
2016 | * If identify namespace failed, use default 512 byte block size so | |
2017 | * block layer can use before failing read/write for 0 capacity. | |
2018 | */ | |
2019 | if (ns->lba_shift == 0) | |
2020 | ns->lba_shift = 9; | |
2021 | bs = 1 << ns->lba_shift; | |
2022 | ||
2023 | /* XXX: PI implementation requires metadata equal t10 pi tuple size */ | |
2024 | pi_type = ns->ms == sizeof(struct t10_pi_tuple) ? | |
2025 | id->dps & NVME_NS_DPS_PI_MASK : 0; | |
2026 | ||
52b68d7e KB |
2027 | if (blk_get_integrity(disk) && (ns->pi_type != pi_type || |
2028 | ns->ms != old_ms || | |
e1e5e564 | 2029 | bs != queue_logical_block_size(disk->queue) || |
a67a9513 | 2030 | (ns->ms && ns->ext))) |
e1e5e564 KB |
2031 | blk_integrity_unregister(disk); |
2032 | ||
2033 | ns->pi_type = pi_type; | |
2034 | blk_queue_logical_block_size(ns->queue, bs); | |
2035 | ||
52b68d7e | 2036 | if (ns->ms && !blk_get_integrity(disk) && (disk->flags & GENHD_FL_UP) && |
a67a9513 | 2037 | !ns->ext) |
e1e5e564 KB |
2038 | nvme_init_integrity(ns); |
2039 | ||
a5768aa8 | 2040 | if (ns->ms && !blk_get_integrity(disk)) |
e1e5e564 KB |
2041 | set_capacity(disk, 0); |
2042 | else | |
2043 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); | |
2044 | ||
2045 | if (dev->oncs & NVME_CTRL_ONCS_DSM) | |
2046 | nvme_config_discard(ns); | |
1b9dbf7f | 2047 | |
d29ec824 | 2048 | kfree(id); |
1b9dbf7f KB |
2049 | return 0; |
2050 | } | |
2051 | ||
b60503ba MW |
2052 | static const struct block_device_operations nvme_fops = { |
2053 | .owner = THIS_MODULE, | |
2054 | .ioctl = nvme_ioctl, | |
320a3827 | 2055 | .compat_ioctl = nvme_compat_ioctl, |
9ac27090 KB |
2056 | .open = nvme_open, |
2057 | .release = nvme_release, | |
4cc09e2d | 2058 | .getgeo = nvme_getgeo, |
1b9dbf7f | 2059 | .revalidate_disk= nvme_revalidate_disk, |
b60503ba MW |
2060 | }; |
2061 | ||
1fa6aead MW |
2062 | static int nvme_kthread(void *data) |
2063 | { | |
d4b4ff8e | 2064 | struct nvme_dev *dev, *next; |
1fa6aead MW |
2065 | |
2066 | while (!kthread_should_stop()) { | |
564a232c | 2067 | set_current_state(TASK_INTERRUPTIBLE); |
1fa6aead | 2068 | spin_lock(&dev_list_lock); |
d4b4ff8e | 2069 | list_for_each_entry_safe(dev, next, &dev_list, node) { |
1fa6aead | 2070 | int i; |
dfbac8c7 KB |
2071 | u32 csts = readl(&dev->bar->csts); |
2072 | ||
2073 | if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) || | |
2074 | csts & NVME_CSTS_CFS) { | |
d4b4ff8e KB |
2075 | if (work_busy(&dev->reset_work)) |
2076 | continue; | |
2077 | list_del_init(&dev->node); | |
e75ec752 | 2078 | dev_warn(dev->dev, |
a4aea562 MB |
2079 | "Failed status: %x, reset controller\n", |
2080 | readl(&dev->bar->csts)); | |
9ca97374 | 2081 | dev->reset_workfn = nvme_reset_failed_dev; |
d4b4ff8e KB |
2082 | queue_work(nvme_workq, &dev->reset_work); |
2083 | continue; | |
2084 | } | |
1fa6aead | 2085 | for (i = 0; i < dev->queue_count; i++) { |
a4aea562 | 2086 | struct nvme_queue *nvmeq = dev->queues[i]; |
740216fc MW |
2087 | if (!nvmeq) |
2088 | continue; | |
1fa6aead | 2089 | spin_lock_irq(&nvmeq->q_lock); |
bc57a0f7 | 2090 | nvme_process_cq(nvmeq); |
6fccf938 KB |
2091 | |
2092 | while ((i == 0) && (dev->event_limit > 0)) { | |
a4aea562 | 2093 | if (nvme_submit_async_admin_req(dev)) |
6fccf938 KB |
2094 | break; |
2095 | dev->event_limit--; | |
2096 | } | |
1fa6aead MW |
2097 | spin_unlock_irq(&nvmeq->q_lock); |
2098 | } | |
2099 | } | |
2100 | spin_unlock(&dev_list_lock); | |
acb7aa0d | 2101 | schedule_timeout(round_jiffies_relative(HZ)); |
1fa6aead MW |
2102 | } |
2103 | return 0; | |
2104 | } | |
2105 | ||
e1e5e564 | 2106 | static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid) |
b60503ba MW |
2107 | { |
2108 | struct nvme_ns *ns; | |
2109 | struct gendisk *disk; | |
e75ec752 | 2110 | int node = dev_to_node(dev->dev); |
b60503ba | 2111 | |
a4aea562 | 2112 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); |
b60503ba | 2113 | if (!ns) |
e1e5e564 KB |
2114 | return; |
2115 | ||
a4aea562 | 2116 | ns->queue = blk_mq_init_queue(&dev->tagset); |
9f173b33 | 2117 | if (IS_ERR(ns->queue)) |
b60503ba | 2118 | goto out_free_ns; |
4eeb9215 MW |
2119 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue); |
2120 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); | |
a4aea562 | 2121 | queue_flag_set_unlocked(QUEUE_FLAG_SG_GAPS, ns->queue); |
b60503ba MW |
2122 | ns->dev = dev; |
2123 | ns->queue->queuedata = ns; | |
2124 | ||
a4aea562 | 2125 | disk = alloc_disk_node(0, node); |
b60503ba MW |
2126 | if (!disk) |
2127 | goto out_free_queue; | |
a4aea562 | 2128 | |
5aff9382 | 2129 | ns->ns_id = nsid; |
b60503ba | 2130 | ns->disk = disk; |
e1e5e564 KB |
2131 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ |
2132 | list_add_tail(&ns->list, &dev->namespaces); | |
2133 | ||
e9ef4636 | 2134 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); |
e824410f | 2135 | if (dev->max_hw_sectors) { |
8fc23e03 | 2136 | blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors); |
e824410f KB |
2137 | blk_queue_max_segments(ns->queue, |
2138 | ((dev->max_hw_sectors << 9) / dev->page_size) + 1); | |
2139 | } | |
a4aea562 MB |
2140 | if (dev->stripe_size) |
2141 | blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9); | |
a7d2ce28 KB |
2142 | if (dev->vwc & NVME_CTRL_VWC_PRESENT) |
2143 | blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA); | |
b60503ba MW |
2144 | |
2145 | disk->major = nvme_major; | |
469071a3 | 2146 | disk->first_minor = 0; |
b60503ba MW |
2147 | disk->fops = &nvme_fops; |
2148 | disk->private_data = ns; | |
2149 | disk->queue = ns->queue; | |
b3fffdef | 2150 | disk->driverfs_dev = dev->device; |
469071a3 | 2151 | disk->flags = GENHD_FL_EXT_DEVT; |
5aff9382 | 2152 | sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid); |
b60503ba | 2153 | |
e1e5e564 KB |
2154 | /* |
2155 | * Initialize capacity to 0 until we establish the namespace format and | |
2156 | * setup integrity extentions if necessary. The revalidate_disk after | |
2157 | * add_disk allows the driver to register with integrity if the format | |
2158 | * requires it. | |
2159 | */ | |
2160 | set_capacity(disk, 0); | |
a5768aa8 KB |
2161 | if (nvme_revalidate_disk(ns->disk)) |
2162 | goto out_free_disk; | |
2163 | ||
e1e5e564 | 2164 | add_disk(ns->disk); |
7bee6074 KB |
2165 | if (ns->ms) { |
2166 | struct block_device *bd = bdget_disk(ns->disk, 0); | |
2167 | if (!bd) | |
2168 | return; | |
2169 | if (blkdev_get(bd, FMODE_READ, NULL)) { | |
2170 | bdput(bd); | |
2171 | return; | |
2172 | } | |
2173 | blkdev_reread_part(bd); | |
2174 | blkdev_put(bd, FMODE_READ); | |
2175 | } | |
e1e5e564 | 2176 | return; |
a5768aa8 KB |
2177 | out_free_disk: |
2178 | kfree(disk); | |
2179 | list_del(&ns->list); | |
b60503ba MW |
2180 | out_free_queue: |
2181 | blk_cleanup_queue(ns->queue); | |
2182 | out_free_ns: | |
2183 | kfree(ns); | |
b60503ba MW |
2184 | } |
2185 | ||
42f61420 KB |
2186 | static void nvme_create_io_queues(struct nvme_dev *dev) |
2187 | { | |
a4aea562 | 2188 | unsigned i; |
42f61420 | 2189 | |
a4aea562 | 2190 | for (i = dev->queue_count; i <= dev->max_qid; i++) |
2b25d981 | 2191 | if (!nvme_alloc_queue(dev, i, dev->q_depth)) |
42f61420 KB |
2192 | break; |
2193 | ||
a4aea562 MB |
2194 | for (i = dev->online_queues; i <= dev->queue_count - 1; i++) |
2195 | if (nvme_create_queue(dev->queues[i], i)) | |
42f61420 KB |
2196 | break; |
2197 | } | |
2198 | ||
b3b06812 | 2199 | static int set_queue_count(struct nvme_dev *dev, int count) |
b60503ba MW |
2200 | { |
2201 | int status; | |
2202 | u32 result; | |
b3b06812 | 2203 | u32 q_count = (count - 1) | ((count - 1) << 16); |
b60503ba | 2204 | |
df348139 | 2205 | status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0, |
bc5fc7e4 | 2206 | &result); |
27e8166c MW |
2207 | if (status < 0) |
2208 | return status; | |
2209 | if (status > 0) { | |
e75ec752 | 2210 | dev_err(dev->dev, "Could not set queue count (%d)\n", status); |
badc34d4 | 2211 | return 0; |
27e8166c | 2212 | } |
b60503ba MW |
2213 | return min(result & 0xffff, result >> 16) + 1; |
2214 | } | |
2215 | ||
8ffaadf7 JD |
2216 | static void __iomem *nvme_map_cmb(struct nvme_dev *dev) |
2217 | { | |
2218 | u64 szu, size, offset; | |
2219 | u32 cmbloc; | |
2220 | resource_size_t bar_size; | |
2221 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
2222 | void __iomem *cmb; | |
2223 | dma_addr_t dma_addr; | |
2224 | ||
2225 | if (!use_cmb_sqes) | |
2226 | return NULL; | |
2227 | ||
2228 | dev->cmbsz = readl(&dev->bar->cmbsz); | |
2229 | if (!(NVME_CMB_SZ(dev->cmbsz))) | |
2230 | return NULL; | |
2231 | ||
2232 | cmbloc = readl(&dev->bar->cmbloc); | |
2233 | ||
2234 | szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz)); | |
2235 | size = szu * NVME_CMB_SZ(dev->cmbsz); | |
2236 | offset = szu * NVME_CMB_OFST(cmbloc); | |
2237 | bar_size = pci_resource_len(pdev, NVME_CMB_BIR(cmbloc)); | |
2238 | ||
2239 | if (offset > bar_size) | |
2240 | return NULL; | |
2241 | ||
2242 | /* | |
2243 | * Controllers may support a CMB size larger than their BAR, | |
2244 | * for example, due to being behind a bridge. Reduce the CMB to | |
2245 | * the reported size of the BAR | |
2246 | */ | |
2247 | if (size > bar_size - offset) | |
2248 | size = bar_size - offset; | |
2249 | ||
2250 | dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(cmbloc)) + offset; | |
2251 | cmb = ioremap_wc(dma_addr, size); | |
2252 | if (!cmb) | |
2253 | return NULL; | |
2254 | ||
2255 | dev->cmb_dma_addr = dma_addr; | |
2256 | dev->cmb_size = size; | |
2257 | return cmb; | |
2258 | } | |
2259 | ||
2260 | static inline void nvme_release_cmb(struct nvme_dev *dev) | |
2261 | { | |
2262 | if (dev->cmb) { | |
2263 | iounmap(dev->cmb); | |
2264 | dev->cmb = NULL; | |
2265 | } | |
2266 | } | |
2267 | ||
9d713c2b KB |
2268 | static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
2269 | { | |
b80d5ccc | 2270 | return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride); |
9d713c2b KB |
2271 | } |
2272 | ||
8d85fce7 | 2273 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 2274 | { |
a4aea562 | 2275 | struct nvme_queue *adminq = dev->queues[0]; |
e75ec752 | 2276 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
42f61420 | 2277 | int result, i, vecs, nr_io_queues, size; |
b60503ba | 2278 | |
42f61420 | 2279 | nr_io_queues = num_possible_cpus(); |
b348b7d5 | 2280 | result = set_queue_count(dev, nr_io_queues); |
badc34d4 | 2281 | if (result <= 0) |
1b23484b | 2282 | return result; |
b348b7d5 MW |
2283 | if (result < nr_io_queues) |
2284 | nr_io_queues = result; | |
b60503ba | 2285 | |
8ffaadf7 JD |
2286 | if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) { |
2287 | result = nvme_cmb_qdepth(dev, nr_io_queues, | |
2288 | sizeof(struct nvme_command)); | |
2289 | if (result > 0) | |
2290 | dev->q_depth = result; | |
2291 | else | |
2292 | nvme_release_cmb(dev); | |
2293 | } | |
2294 | ||
9d713c2b KB |
2295 | size = db_bar_size(dev, nr_io_queues); |
2296 | if (size > 8192) { | |
f1938f6e | 2297 | iounmap(dev->bar); |
9d713c2b KB |
2298 | do { |
2299 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
2300 | if (dev->bar) | |
2301 | break; | |
2302 | if (!--nr_io_queues) | |
2303 | return -ENOMEM; | |
2304 | size = db_bar_size(dev, nr_io_queues); | |
2305 | } while (1); | |
f1938f6e | 2306 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
5a92e700 | 2307 | adminq->q_db = dev->dbs; |
f1938f6e MW |
2308 | } |
2309 | ||
9d713c2b | 2310 | /* Deregister the admin queue's interrupt */ |
3193f07b | 2311 | free_irq(dev->entry[0].vector, adminq); |
9d713c2b | 2312 | |
e32efbfc JA |
2313 | /* |
2314 | * If we enable msix early due to not intx, disable it again before | |
2315 | * setting up the full range we need. | |
2316 | */ | |
2317 | if (!pdev->irq) | |
2318 | pci_disable_msix(pdev); | |
2319 | ||
be577fab | 2320 | for (i = 0; i < nr_io_queues; i++) |
1b23484b | 2321 | dev->entry[i].entry = i; |
be577fab AG |
2322 | vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues); |
2323 | if (vecs < 0) { | |
2324 | vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32)); | |
2325 | if (vecs < 0) { | |
2326 | vecs = 1; | |
2327 | } else { | |
2328 | for (i = 0; i < vecs; i++) | |
2329 | dev->entry[i].vector = i + pdev->irq; | |
fa08a396 RRG |
2330 | } |
2331 | } | |
2332 | ||
063a8096 MW |
2333 | /* |
2334 | * Should investigate if there's a performance win from allocating | |
2335 | * more queues than interrupt vectors; it might allow the submission | |
2336 | * path to scale better, even if the receive path is limited by the | |
2337 | * number of interrupts. | |
2338 | */ | |
2339 | nr_io_queues = vecs; | |
42f61420 | 2340 | dev->max_qid = nr_io_queues; |
063a8096 | 2341 | |
3193f07b | 2342 | result = queue_request_irq(dev, adminq, adminq->irqname); |
758dd7fd JD |
2343 | if (result) { |
2344 | adminq->cq_vector = -1; | |
22404274 | 2345 | goto free_queues; |
758dd7fd | 2346 | } |
1b23484b | 2347 | |
cd638946 | 2348 | /* Free previously allocated queues that are no longer usable */ |
42f61420 | 2349 | nvme_free_queues(dev, nr_io_queues + 1); |
a4aea562 | 2350 | nvme_create_io_queues(dev); |
9ecdc946 | 2351 | |
22404274 | 2352 | return 0; |
b60503ba | 2353 | |
22404274 | 2354 | free_queues: |
a1a5ef99 | 2355 | nvme_free_queues(dev, 1); |
22404274 | 2356 | return result; |
b60503ba MW |
2357 | } |
2358 | ||
a5768aa8 KB |
2359 | static void nvme_free_namespace(struct nvme_ns *ns) |
2360 | { | |
2361 | list_del(&ns->list); | |
2362 | ||
2363 | spin_lock(&dev_list_lock); | |
2364 | ns->disk->private_data = NULL; | |
2365 | spin_unlock(&dev_list_lock); | |
2366 | ||
2367 | put_disk(ns->disk); | |
2368 | kfree(ns); | |
2369 | } | |
2370 | ||
2371 | static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2372 | { | |
2373 | struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); | |
2374 | struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); | |
2375 | ||
2376 | return nsa->ns_id - nsb->ns_id; | |
2377 | } | |
2378 | ||
2379 | static struct nvme_ns *nvme_find_ns(struct nvme_dev *dev, unsigned nsid) | |
2380 | { | |
2381 | struct nvme_ns *ns; | |
2382 | ||
2383 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2384 | if (ns->ns_id == nsid) | |
2385 | return ns; | |
2386 | if (ns->ns_id > nsid) | |
2387 | break; | |
2388 | } | |
2389 | return NULL; | |
2390 | } | |
2391 | ||
2392 | static inline bool nvme_io_incapable(struct nvme_dev *dev) | |
2393 | { | |
2394 | return (!dev->bar || readl(&dev->bar->csts) & NVME_CSTS_CFS || | |
2395 | dev->online_queues < 2); | |
2396 | } | |
2397 | ||
2398 | static void nvme_ns_remove(struct nvme_ns *ns) | |
2399 | { | |
2400 | bool kill = nvme_io_incapable(ns->dev) && !blk_queue_dying(ns->queue); | |
2401 | ||
2402 | if (kill) | |
2403 | blk_set_queue_dying(ns->queue); | |
2404 | if (ns->disk->flags & GENHD_FL_UP) { | |
2405 | if (blk_get_integrity(ns->disk)) | |
2406 | blk_integrity_unregister(ns->disk); | |
2407 | del_gendisk(ns->disk); | |
2408 | } | |
2409 | if (kill || !blk_queue_dying(ns->queue)) { | |
2410 | blk_mq_abort_requeue_list(ns->queue); | |
2411 | blk_cleanup_queue(ns->queue); | |
2412 | } | |
2413 | } | |
2414 | ||
2415 | static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn) | |
2416 | { | |
2417 | struct nvme_ns *ns, *next; | |
2418 | unsigned i; | |
2419 | ||
2420 | for (i = 1; i <= nn; i++) { | |
2421 | ns = nvme_find_ns(dev, i); | |
2422 | if (ns) { | |
2423 | if (revalidate_disk(ns->disk)) { | |
2424 | nvme_ns_remove(ns); | |
2425 | nvme_free_namespace(ns); | |
2426 | } | |
2427 | } else | |
2428 | nvme_alloc_ns(dev, i); | |
2429 | } | |
2430 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) { | |
2431 | if (ns->ns_id > nn) { | |
2432 | nvme_ns_remove(ns); | |
2433 | nvme_free_namespace(ns); | |
2434 | } | |
2435 | } | |
2436 | list_sort(NULL, &dev->namespaces, ns_cmp); | |
2437 | } | |
2438 | ||
2439 | static void nvme_dev_scan(struct work_struct *work) | |
2440 | { | |
2441 | struct nvme_dev *dev = container_of(work, struct nvme_dev, scan_work); | |
2442 | struct nvme_id_ctrl *ctrl; | |
2443 | ||
2444 | if (!dev->tagset.tags) | |
2445 | return; | |
2446 | if (nvme_identify_ctrl(dev, &ctrl)) | |
2447 | return; | |
2448 | nvme_scan_namespaces(dev, le32_to_cpup(&ctrl->nn)); | |
2449 | kfree(ctrl); | |
2450 | } | |
2451 | ||
422ef0c7 MW |
2452 | /* |
2453 | * Return: error value if an error occurred setting up the queues or calling | |
2454 | * Identify Device. 0 if these succeeded, even if adding some of the | |
2455 | * namespaces failed. At the moment, these failures are silent. TBD which | |
2456 | * failures should be reported. | |
2457 | */ | |
8d85fce7 | 2458 | static int nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 2459 | { |
e75ec752 | 2460 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
c3bfe717 | 2461 | int res; |
51814232 | 2462 | struct nvme_id_ctrl *ctrl; |
159b67d7 | 2463 | int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12; |
b60503ba | 2464 | |
d29ec824 | 2465 | res = nvme_identify_ctrl(dev, &ctrl); |
b60503ba | 2466 | if (res) { |
e75ec752 | 2467 | dev_err(dev->dev, "Identify Controller failed (%d)\n", res); |
e1e5e564 | 2468 | return -EIO; |
b60503ba MW |
2469 | } |
2470 | ||
0e5e4f0e | 2471 | dev->oncs = le16_to_cpup(&ctrl->oncs); |
c30341dc | 2472 | dev->abort_limit = ctrl->acl + 1; |
a7d2ce28 | 2473 | dev->vwc = ctrl->vwc; |
51814232 MW |
2474 | memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn)); |
2475 | memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn)); | |
2476 | memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr)); | |
159b67d7 | 2477 | if (ctrl->mdts) |
8fc23e03 | 2478 | dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9); |
68608c26 | 2479 | if ((pdev->vendor == PCI_VENDOR_ID_INTEL) && |
a4aea562 MB |
2480 | (pdev->device == 0x0953) && ctrl->vs[3]) { |
2481 | unsigned int max_hw_sectors; | |
2482 | ||
159b67d7 | 2483 | dev->stripe_size = 1 << (ctrl->vs[3] + shift); |
a4aea562 MB |
2484 | max_hw_sectors = dev->stripe_size >> (shift - 9); |
2485 | if (dev->max_hw_sectors) { | |
2486 | dev->max_hw_sectors = min(max_hw_sectors, | |
2487 | dev->max_hw_sectors); | |
2488 | } else | |
2489 | dev->max_hw_sectors = max_hw_sectors; | |
2490 | } | |
d29ec824 | 2491 | kfree(ctrl); |
a4aea562 | 2492 | |
ffe7704d KB |
2493 | if (!dev->tagset.tags) { |
2494 | dev->tagset.ops = &nvme_mq_ops; | |
2495 | dev->tagset.nr_hw_queues = dev->online_queues - 1; | |
2496 | dev->tagset.timeout = NVME_IO_TIMEOUT; | |
2497 | dev->tagset.numa_node = dev_to_node(dev->dev); | |
2498 | dev->tagset.queue_depth = | |
a4aea562 | 2499 | min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; |
ffe7704d KB |
2500 | dev->tagset.cmd_size = nvme_cmd_size(dev); |
2501 | dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; | |
2502 | dev->tagset.driver_data = dev; | |
b60503ba | 2503 | |
ffe7704d KB |
2504 | if (blk_mq_alloc_tag_set(&dev->tagset)) |
2505 | return 0; | |
2506 | } | |
a5768aa8 | 2507 | schedule_work(&dev->scan_work); |
e1e5e564 | 2508 | return 0; |
b60503ba MW |
2509 | } |
2510 | ||
0877cb0d KB |
2511 | static int nvme_dev_map(struct nvme_dev *dev) |
2512 | { | |
42f61420 | 2513 | u64 cap; |
0877cb0d | 2514 | int bars, result = -ENOMEM; |
e75ec752 | 2515 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
0877cb0d KB |
2516 | |
2517 | if (pci_enable_device_mem(pdev)) | |
2518 | return result; | |
2519 | ||
2520 | dev->entry[0].vector = pdev->irq; | |
2521 | pci_set_master(pdev); | |
2522 | bars = pci_select_bars(pdev, IORESOURCE_MEM); | |
be7837e8 JA |
2523 | if (!bars) |
2524 | goto disable_pci; | |
2525 | ||
0877cb0d KB |
2526 | if (pci_request_selected_regions(pdev, bars, "nvme")) |
2527 | goto disable_pci; | |
2528 | ||
e75ec752 CH |
2529 | if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) && |
2530 | dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32))) | |
052d0efa | 2531 | goto disable; |
0877cb0d | 2532 | |
0877cb0d KB |
2533 | dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); |
2534 | if (!dev->bar) | |
2535 | goto disable; | |
e32efbfc | 2536 | |
0e53d180 KB |
2537 | if (readl(&dev->bar->csts) == -1) { |
2538 | result = -ENODEV; | |
2539 | goto unmap; | |
2540 | } | |
e32efbfc JA |
2541 | |
2542 | /* | |
2543 | * Some devices don't advertse INTx interrupts, pre-enable a single | |
2544 | * MSIX vec for setup. We'll adjust this later. | |
2545 | */ | |
2546 | if (!pdev->irq) { | |
2547 | result = pci_enable_msix(pdev, dev->entry, 1); | |
2548 | if (result < 0) | |
2549 | goto unmap; | |
2550 | } | |
2551 | ||
42f61420 KB |
2552 | cap = readq(&dev->bar->cap); |
2553 | dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); | |
2554 | dev->db_stride = 1 << NVME_CAP_STRIDE(cap); | |
0877cb0d | 2555 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
8ffaadf7 JD |
2556 | if (readl(&dev->bar->vs) >= NVME_VS(1, 2)) |
2557 | dev->cmb = nvme_map_cmb(dev); | |
0877cb0d KB |
2558 | |
2559 | return 0; | |
2560 | ||
0e53d180 KB |
2561 | unmap: |
2562 | iounmap(dev->bar); | |
2563 | dev->bar = NULL; | |
0877cb0d KB |
2564 | disable: |
2565 | pci_release_regions(pdev); | |
2566 | disable_pci: | |
2567 | pci_disable_device(pdev); | |
2568 | return result; | |
2569 | } | |
2570 | ||
2571 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
2572 | { | |
e75ec752 CH |
2573 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2574 | ||
2575 | if (pdev->msi_enabled) | |
2576 | pci_disable_msi(pdev); | |
2577 | else if (pdev->msix_enabled) | |
2578 | pci_disable_msix(pdev); | |
0877cb0d KB |
2579 | |
2580 | if (dev->bar) { | |
2581 | iounmap(dev->bar); | |
2582 | dev->bar = NULL; | |
e75ec752 | 2583 | pci_release_regions(pdev); |
0877cb0d KB |
2584 | } |
2585 | ||
e75ec752 CH |
2586 | if (pci_is_enabled(pdev)) |
2587 | pci_disable_device(pdev); | |
0877cb0d KB |
2588 | } |
2589 | ||
4d115420 KB |
2590 | struct nvme_delq_ctx { |
2591 | struct task_struct *waiter; | |
2592 | struct kthread_worker *worker; | |
2593 | atomic_t refcount; | |
2594 | }; | |
2595 | ||
2596 | static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev) | |
2597 | { | |
2598 | dq->waiter = current; | |
2599 | mb(); | |
2600 | ||
2601 | for (;;) { | |
2602 | set_current_state(TASK_KILLABLE); | |
2603 | if (!atomic_read(&dq->refcount)) | |
2604 | break; | |
2605 | if (!schedule_timeout(ADMIN_TIMEOUT) || | |
2606 | fatal_signal_pending(current)) { | |
0fb59cbc KB |
2607 | /* |
2608 | * Disable the controller first since we can't trust it | |
2609 | * at this point, but leave the admin queue enabled | |
2610 | * until all queue deletion requests are flushed. | |
2611 | * FIXME: This may take a while if there are more h/w | |
2612 | * queues than admin tags. | |
2613 | */ | |
4d115420 | 2614 | set_current_state(TASK_RUNNING); |
4d115420 | 2615 | nvme_disable_ctrl(dev, readq(&dev->bar->cap)); |
0fb59cbc | 2616 | nvme_clear_queue(dev->queues[0]); |
4d115420 | 2617 | flush_kthread_worker(dq->worker); |
0fb59cbc | 2618 | nvme_disable_queue(dev, 0); |
4d115420 KB |
2619 | return; |
2620 | } | |
2621 | } | |
2622 | set_current_state(TASK_RUNNING); | |
2623 | } | |
2624 | ||
2625 | static void nvme_put_dq(struct nvme_delq_ctx *dq) | |
2626 | { | |
2627 | atomic_dec(&dq->refcount); | |
2628 | if (dq->waiter) | |
2629 | wake_up_process(dq->waiter); | |
2630 | } | |
2631 | ||
2632 | static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq) | |
2633 | { | |
2634 | atomic_inc(&dq->refcount); | |
2635 | return dq; | |
2636 | } | |
2637 | ||
2638 | static void nvme_del_queue_end(struct nvme_queue *nvmeq) | |
2639 | { | |
2640 | struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx; | |
4d115420 KB |
2641 | nvme_put_dq(dq); |
2642 | } | |
2643 | ||
2644 | static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode, | |
2645 | kthread_work_func_t fn) | |
2646 | { | |
2647 | struct nvme_command c; | |
2648 | ||
2649 | memset(&c, 0, sizeof(c)); | |
2650 | c.delete_queue.opcode = opcode; | |
2651 | c.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
2652 | ||
2653 | init_kthread_work(&nvmeq->cmdinfo.work, fn); | |
a4aea562 MB |
2654 | return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo, |
2655 | ADMIN_TIMEOUT); | |
4d115420 KB |
2656 | } |
2657 | ||
2658 | static void nvme_del_cq_work_handler(struct kthread_work *work) | |
2659 | { | |
2660 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2661 | cmdinfo.work); | |
2662 | nvme_del_queue_end(nvmeq); | |
2663 | } | |
2664 | ||
2665 | static int nvme_delete_cq(struct nvme_queue *nvmeq) | |
2666 | { | |
2667 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq, | |
2668 | nvme_del_cq_work_handler); | |
2669 | } | |
2670 | ||
2671 | static void nvme_del_sq_work_handler(struct kthread_work *work) | |
2672 | { | |
2673 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2674 | cmdinfo.work); | |
2675 | int status = nvmeq->cmdinfo.status; | |
2676 | ||
2677 | if (!status) | |
2678 | status = nvme_delete_cq(nvmeq); | |
2679 | if (status) | |
2680 | nvme_del_queue_end(nvmeq); | |
2681 | } | |
2682 | ||
2683 | static int nvme_delete_sq(struct nvme_queue *nvmeq) | |
2684 | { | |
2685 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq, | |
2686 | nvme_del_sq_work_handler); | |
2687 | } | |
2688 | ||
2689 | static void nvme_del_queue_start(struct kthread_work *work) | |
2690 | { | |
2691 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2692 | cmdinfo.work); | |
4d115420 KB |
2693 | if (nvme_delete_sq(nvmeq)) |
2694 | nvme_del_queue_end(nvmeq); | |
2695 | } | |
2696 | ||
2697 | static void nvme_disable_io_queues(struct nvme_dev *dev) | |
2698 | { | |
2699 | int i; | |
2700 | DEFINE_KTHREAD_WORKER_ONSTACK(worker); | |
2701 | struct nvme_delq_ctx dq; | |
2702 | struct task_struct *kworker_task = kthread_run(kthread_worker_fn, | |
2703 | &worker, "nvme%d", dev->instance); | |
2704 | ||
2705 | if (IS_ERR(kworker_task)) { | |
e75ec752 | 2706 | dev_err(dev->dev, |
4d115420 KB |
2707 | "Failed to create queue del task\n"); |
2708 | for (i = dev->queue_count - 1; i > 0; i--) | |
2709 | nvme_disable_queue(dev, i); | |
2710 | return; | |
2711 | } | |
2712 | ||
2713 | dq.waiter = NULL; | |
2714 | atomic_set(&dq.refcount, 0); | |
2715 | dq.worker = &worker; | |
2716 | for (i = dev->queue_count - 1; i > 0; i--) { | |
a4aea562 | 2717 | struct nvme_queue *nvmeq = dev->queues[i]; |
4d115420 KB |
2718 | |
2719 | if (nvme_suspend_queue(nvmeq)) | |
2720 | continue; | |
2721 | nvmeq->cmdinfo.ctx = nvme_get_dq(&dq); | |
2722 | nvmeq->cmdinfo.worker = dq.worker; | |
2723 | init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start); | |
2724 | queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work); | |
2725 | } | |
2726 | nvme_wait_dq(&dq, dev); | |
2727 | kthread_stop(kworker_task); | |
2728 | } | |
2729 | ||
b9afca3e DM |
2730 | /* |
2731 | * Remove the node from the device list and check | |
2732 | * for whether or not we need to stop the nvme_thread. | |
2733 | */ | |
2734 | static void nvme_dev_list_remove(struct nvme_dev *dev) | |
2735 | { | |
2736 | struct task_struct *tmp = NULL; | |
2737 | ||
2738 | spin_lock(&dev_list_lock); | |
2739 | list_del_init(&dev->node); | |
2740 | if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) { | |
2741 | tmp = nvme_thread; | |
2742 | nvme_thread = NULL; | |
2743 | } | |
2744 | spin_unlock(&dev_list_lock); | |
2745 | ||
2746 | if (tmp) | |
2747 | kthread_stop(tmp); | |
2748 | } | |
2749 | ||
c9d3bf88 KB |
2750 | static void nvme_freeze_queues(struct nvme_dev *dev) |
2751 | { | |
2752 | struct nvme_ns *ns; | |
2753 | ||
2754 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2755 | blk_mq_freeze_queue_start(ns->queue); | |
2756 | ||
cddcd72b | 2757 | spin_lock_irq(ns->queue->queue_lock); |
c9d3bf88 | 2758 | queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue); |
cddcd72b | 2759 | spin_unlock_irq(ns->queue->queue_lock); |
c9d3bf88 KB |
2760 | |
2761 | blk_mq_cancel_requeue_work(ns->queue); | |
2762 | blk_mq_stop_hw_queues(ns->queue); | |
2763 | } | |
2764 | } | |
2765 | ||
2766 | static void nvme_unfreeze_queues(struct nvme_dev *dev) | |
2767 | { | |
2768 | struct nvme_ns *ns; | |
2769 | ||
2770 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2771 | queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue); | |
2772 | blk_mq_unfreeze_queue(ns->queue); | |
2773 | blk_mq_start_stopped_hw_queues(ns->queue, true); | |
2774 | blk_mq_kick_requeue_list(ns->queue); | |
2775 | } | |
2776 | } | |
2777 | ||
f0b50732 | 2778 | static void nvme_dev_shutdown(struct nvme_dev *dev) |
b60503ba | 2779 | { |
22404274 | 2780 | int i; |
7c1b2450 | 2781 | u32 csts = -1; |
22404274 | 2782 | |
b9afca3e | 2783 | nvme_dev_list_remove(dev); |
1fa6aead | 2784 | |
c9d3bf88 KB |
2785 | if (dev->bar) { |
2786 | nvme_freeze_queues(dev); | |
7c1b2450 | 2787 | csts = readl(&dev->bar->csts); |
c9d3bf88 | 2788 | } |
7c1b2450 | 2789 | if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) { |
4d115420 | 2790 | for (i = dev->queue_count - 1; i >= 0; i--) { |
a4aea562 | 2791 | struct nvme_queue *nvmeq = dev->queues[i]; |
4d115420 | 2792 | nvme_suspend_queue(nvmeq); |
4d115420 KB |
2793 | } |
2794 | } else { | |
2795 | nvme_disable_io_queues(dev); | |
1894d8f1 | 2796 | nvme_shutdown_ctrl(dev); |
4d115420 KB |
2797 | nvme_disable_queue(dev, 0); |
2798 | } | |
f0b50732 | 2799 | nvme_dev_unmap(dev); |
07836e65 KB |
2800 | |
2801 | for (i = dev->queue_count - 1; i >= 0; i--) | |
2802 | nvme_clear_queue(dev->queues[i]); | |
f0b50732 KB |
2803 | } |
2804 | ||
2805 | static void nvme_dev_remove(struct nvme_dev *dev) | |
2806 | { | |
9ac27090 | 2807 | struct nvme_ns *ns; |
f0b50732 | 2808 | |
a5768aa8 KB |
2809 | list_for_each_entry(ns, &dev->namespaces, list) |
2810 | nvme_ns_remove(ns); | |
b60503ba MW |
2811 | } |
2812 | ||
091b6092 MW |
2813 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
2814 | { | |
e75ec752 | 2815 | dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, |
091b6092 MW |
2816 | PAGE_SIZE, PAGE_SIZE, 0); |
2817 | if (!dev->prp_page_pool) | |
2818 | return -ENOMEM; | |
2819 | ||
99802a7a | 2820 | /* Optimisation for I/Os between 4k and 128k */ |
e75ec752 | 2821 | dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, |
99802a7a MW |
2822 | 256, 256, 0); |
2823 | if (!dev->prp_small_pool) { | |
2824 | dma_pool_destroy(dev->prp_page_pool); | |
2825 | return -ENOMEM; | |
2826 | } | |
091b6092 MW |
2827 | return 0; |
2828 | } | |
2829 | ||
2830 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
2831 | { | |
2832 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 2833 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
2834 | } |
2835 | ||
cd58ad7d QSA |
2836 | static DEFINE_IDA(nvme_instance_ida); |
2837 | ||
2838 | static int nvme_set_instance(struct nvme_dev *dev) | |
b60503ba | 2839 | { |
cd58ad7d QSA |
2840 | int instance, error; |
2841 | ||
2842 | do { | |
2843 | if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL)) | |
2844 | return -ENODEV; | |
2845 | ||
2846 | spin_lock(&dev_list_lock); | |
2847 | error = ida_get_new(&nvme_instance_ida, &instance); | |
2848 | spin_unlock(&dev_list_lock); | |
2849 | } while (error == -EAGAIN); | |
2850 | ||
2851 | if (error) | |
2852 | return -ENODEV; | |
2853 | ||
2854 | dev->instance = instance; | |
2855 | return 0; | |
b60503ba MW |
2856 | } |
2857 | ||
2858 | static void nvme_release_instance(struct nvme_dev *dev) | |
2859 | { | |
cd58ad7d QSA |
2860 | spin_lock(&dev_list_lock); |
2861 | ida_remove(&nvme_instance_ida, dev->instance); | |
2862 | spin_unlock(&dev_list_lock); | |
b60503ba MW |
2863 | } |
2864 | ||
9ac27090 KB |
2865 | static void nvme_free_namespaces(struct nvme_dev *dev) |
2866 | { | |
2867 | struct nvme_ns *ns, *next; | |
2868 | ||
a5768aa8 KB |
2869 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) |
2870 | nvme_free_namespace(ns); | |
9ac27090 KB |
2871 | } |
2872 | ||
5e82e952 KB |
2873 | static void nvme_free_dev(struct kref *kref) |
2874 | { | |
2875 | struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref); | |
9ac27090 | 2876 | |
e75ec752 | 2877 | put_device(dev->dev); |
b3fffdef | 2878 | put_device(dev->device); |
9ac27090 | 2879 | nvme_free_namespaces(dev); |
285dffc9 | 2880 | nvme_release_instance(dev); |
4af0e21c KB |
2881 | if (dev->tagset.tags) |
2882 | blk_mq_free_tag_set(&dev->tagset); | |
2883 | if (dev->admin_q) | |
2884 | blk_put_queue(dev->admin_q); | |
5e82e952 KB |
2885 | kfree(dev->queues); |
2886 | kfree(dev->entry); | |
2887 | kfree(dev); | |
2888 | } | |
2889 | ||
2890 | static int nvme_dev_open(struct inode *inode, struct file *f) | |
2891 | { | |
b3fffdef KB |
2892 | struct nvme_dev *dev; |
2893 | int instance = iminor(inode); | |
2894 | int ret = -ENODEV; | |
2895 | ||
2896 | spin_lock(&dev_list_lock); | |
2897 | list_for_each_entry(dev, &dev_list, node) { | |
2898 | if (dev->instance == instance) { | |
2e1d8448 KB |
2899 | if (!dev->admin_q) { |
2900 | ret = -EWOULDBLOCK; | |
2901 | break; | |
2902 | } | |
b3fffdef KB |
2903 | if (!kref_get_unless_zero(&dev->kref)) |
2904 | break; | |
2905 | f->private_data = dev; | |
2906 | ret = 0; | |
2907 | break; | |
2908 | } | |
2909 | } | |
2910 | spin_unlock(&dev_list_lock); | |
2911 | ||
2912 | return ret; | |
5e82e952 KB |
2913 | } |
2914 | ||
2915 | static int nvme_dev_release(struct inode *inode, struct file *f) | |
2916 | { | |
2917 | struct nvme_dev *dev = f->private_data; | |
2918 | kref_put(&dev->kref, nvme_free_dev); | |
2919 | return 0; | |
2920 | } | |
2921 | ||
2922 | static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg) | |
2923 | { | |
2924 | struct nvme_dev *dev = f->private_data; | |
a4aea562 MB |
2925 | struct nvme_ns *ns; |
2926 | ||
5e82e952 KB |
2927 | switch (cmd) { |
2928 | case NVME_IOCTL_ADMIN_CMD: | |
a4aea562 | 2929 | return nvme_user_cmd(dev, NULL, (void __user *)arg); |
7963e521 | 2930 | case NVME_IOCTL_IO_CMD: |
a4aea562 MB |
2931 | if (list_empty(&dev->namespaces)) |
2932 | return -ENOTTY; | |
2933 | ns = list_first_entry(&dev->namespaces, struct nvme_ns, list); | |
2934 | return nvme_user_cmd(dev, ns, (void __user *)arg); | |
4cc06521 KB |
2935 | case NVME_IOCTL_RESET: |
2936 | dev_warn(dev->dev, "resetting controller\n"); | |
2937 | return nvme_reset(dev); | |
81f03fed JD |
2938 | case NVME_IOCTL_SUBSYS_RESET: |
2939 | return nvme_subsys_reset(dev); | |
5e82e952 KB |
2940 | default: |
2941 | return -ENOTTY; | |
2942 | } | |
2943 | } | |
2944 | ||
2945 | static const struct file_operations nvme_dev_fops = { | |
2946 | .owner = THIS_MODULE, | |
2947 | .open = nvme_dev_open, | |
2948 | .release = nvme_dev_release, | |
2949 | .unlocked_ioctl = nvme_dev_ioctl, | |
2950 | .compat_ioctl = nvme_dev_ioctl, | |
2951 | }; | |
2952 | ||
a4aea562 MB |
2953 | static void nvme_set_irq_hints(struct nvme_dev *dev) |
2954 | { | |
2955 | struct nvme_queue *nvmeq; | |
2956 | int i; | |
2957 | ||
2958 | for (i = 0; i < dev->online_queues; i++) { | |
2959 | nvmeq = dev->queues[i]; | |
2960 | ||
42483228 | 2961 | if (!nvmeq->tags || !(*nvmeq->tags)) |
a4aea562 MB |
2962 | continue; |
2963 | ||
2964 | irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector, | |
42483228 | 2965 | blk_mq_tags_cpumask(*nvmeq->tags)); |
a4aea562 MB |
2966 | } |
2967 | } | |
2968 | ||
f0b50732 KB |
2969 | static int nvme_dev_start(struct nvme_dev *dev) |
2970 | { | |
2971 | int result; | |
b9afca3e | 2972 | bool start_thread = false; |
f0b50732 KB |
2973 | |
2974 | result = nvme_dev_map(dev); | |
2975 | if (result) | |
2976 | return result; | |
2977 | ||
2978 | result = nvme_configure_admin_queue(dev); | |
2979 | if (result) | |
2980 | goto unmap; | |
2981 | ||
2982 | spin_lock(&dev_list_lock); | |
b9afca3e DM |
2983 | if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) { |
2984 | start_thread = true; | |
2985 | nvme_thread = NULL; | |
2986 | } | |
f0b50732 KB |
2987 | list_add(&dev->node, &dev_list); |
2988 | spin_unlock(&dev_list_lock); | |
2989 | ||
b9afca3e DM |
2990 | if (start_thread) { |
2991 | nvme_thread = kthread_run(nvme_kthread, NULL, "nvme"); | |
387caa5a | 2992 | wake_up_all(&nvme_kthread_wait); |
b9afca3e DM |
2993 | } else |
2994 | wait_event_killable(nvme_kthread_wait, nvme_thread); | |
2995 | ||
2996 | if (IS_ERR_OR_NULL(nvme_thread)) { | |
2997 | result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR; | |
2998 | goto disable; | |
2999 | } | |
a4aea562 MB |
3000 | |
3001 | nvme_init_queue(dev->queues[0], 0); | |
0fb59cbc KB |
3002 | result = nvme_alloc_admin_tags(dev); |
3003 | if (result) | |
3004 | goto disable; | |
b9afca3e | 3005 | |
f0b50732 | 3006 | result = nvme_setup_io_queues(dev); |
badc34d4 | 3007 | if (result) |
0fb59cbc | 3008 | goto free_tags; |
f0b50732 | 3009 | |
a4aea562 MB |
3010 | nvme_set_irq_hints(dev); |
3011 | ||
1efccc9d | 3012 | dev->event_limit = 1; |
d82e8bfd | 3013 | return result; |
f0b50732 | 3014 | |
0fb59cbc KB |
3015 | free_tags: |
3016 | nvme_dev_remove_admin(dev); | |
4af0e21c KB |
3017 | blk_put_queue(dev->admin_q); |
3018 | dev->admin_q = NULL; | |
3019 | dev->queues[0]->tags = NULL; | |
f0b50732 | 3020 | disable: |
a1a5ef99 | 3021 | nvme_disable_queue(dev, 0); |
b9afca3e | 3022 | nvme_dev_list_remove(dev); |
f0b50732 KB |
3023 | unmap: |
3024 | nvme_dev_unmap(dev); | |
3025 | return result; | |
3026 | } | |
3027 | ||
9a6b9458 KB |
3028 | static int nvme_remove_dead_ctrl(void *arg) |
3029 | { | |
3030 | struct nvme_dev *dev = (struct nvme_dev *)arg; | |
e75ec752 | 3031 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
9a6b9458 KB |
3032 | |
3033 | if (pci_get_drvdata(pdev)) | |
c81f4975 | 3034 | pci_stop_and_remove_bus_device_locked(pdev); |
9a6b9458 KB |
3035 | kref_put(&dev->kref, nvme_free_dev); |
3036 | return 0; | |
3037 | } | |
3038 | ||
3039 | static void nvme_remove_disks(struct work_struct *ws) | |
3040 | { | |
9a6b9458 KB |
3041 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); |
3042 | ||
5a92e700 | 3043 | nvme_free_queues(dev, 1); |
302c6727 | 3044 | nvme_dev_remove(dev); |
9a6b9458 KB |
3045 | } |
3046 | ||
3047 | static int nvme_dev_resume(struct nvme_dev *dev) | |
3048 | { | |
3049 | int ret; | |
3050 | ||
3051 | ret = nvme_dev_start(dev); | |
badc34d4 | 3052 | if (ret) |
9a6b9458 | 3053 | return ret; |
badc34d4 | 3054 | if (dev->online_queues < 2) { |
9a6b9458 | 3055 | spin_lock(&dev_list_lock); |
9ca97374 | 3056 | dev->reset_workfn = nvme_remove_disks; |
9a6b9458 KB |
3057 | queue_work(nvme_workq, &dev->reset_work); |
3058 | spin_unlock(&dev_list_lock); | |
c9d3bf88 KB |
3059 | } else { |
3060 | nvme_unfreeze_queues(dev); | |
ffe7704d | 3061 | nvme_dev_add(dev); |
c9d3bf88 | 3062 | nvme_set_irq_hints(dev); |
9a6b9458 KB |
3063 | } |
3064 | return 0; | |
3065 | } | |
3066 | ||
de3eff2b KB |
3067 | static void nvme_dead_ctrl(struct nvme_dev *dev) |
3068 | { | |
3069 | dev_warn(dev->dev, "Device failed to resume\n"); | |
3070 | kref_get(&dev->kref); | |
3071 | if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d", | |
3072 | dev->instance))) { | |
3073 | dev_err(dev->dev, | |
3074 | "Failed to start controller remove task\n"); | |
3075 | kref_put(&dev->kref, nvme_free_dev); | |
3076 | } | |
3077 | } | |
3078 | ||
9a6b9458 KB |
3079 | static void nvme_dev_reset(struct nvme_dev *dev) |
3080 | { | |
ffe7704d KB |
3081 | bool in_probe = work_busy(&dev->probe_work); |
3082 | ||
9a6b9458 | 3083 | nvme_dev_shutdown(dev); |
ffe7704d KB |
3084 | |
3085 | /* Synchronize with device probe so that work will see failure status | |
3086 | * and exit gracefully without trying to schedule another reset */ | |
3087 | flush_work(&dev->probe_work); | |
3088 | ||
3089 | /* Fail this device if reset occured during probe to avoid | |
3090 | * infinite initialization loops. */ | |
3091 | if (in_probe) { | |
de3eff2b | 3092 | nvme_dead_ctrl(dev); |
ffe7704d | 3093 | return; |
9a6b9458 | 3094 | } |
ffe7704d KB |
3095 | /* Schedule device resume asynchronously so the reset work is available |
3096 | * to cleanup errors that may occur during reinitialization */ | |
3097 | schedule_work(&dev->probe_work); | |
9a6b9458 KB |
3098 | } |
3099 | ||
3100 | static void nvme_reset_failed_dev(struct work_struct *ws) | |
3101 | { | |
3102 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); | |
3103 | nvme_dev_reset(dev); | |
3104 | } | |
3105 | ||
9ca97374 TH |
3106 | static void nvme_reset_workfn(struct work_struct *work) |
3107 | { | |
3108 | struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work); | |
3109 | dev->reset_workfn(work); | |
3110 | } | |
3111 | ||
4cc06521 KB |
3112 | static int nvme_reset(struct nvme_dev *dev) |
3113 | { | |
3114 | int ret = -EBUSY; | |
3115 | ||
3116 | if (!dev->admin_q || blk_queue_dying(dev->admin_q)) | |
3117 | return -ENODEV; | |
3118 | ||
3119 | spin_lock(&dev_list_lock); | |
3120 | if (!work_pending(&dev->reset_work)) { | |
3121 | dev->reset_workfn = nvme_reset_failed_dev; | |
3122 | queue_work(nvme_workq, &dev->reset_work); | |
3123 | ret = 0; | |
3124 | } | |
3125 | spin_unlock(&dev_list_lock); | |
3126 | ||
3127 | if (!ret) { | |
3128 | flush_work(&dev->reset_work); | |
ffe7704d | 3129 | flush_work(&dev->probe_work); |
4cc06521 KB |
3130 | return 0; |
3131 | } | |
3132 | ||
3133 | return ret; | |
3134 | } | |
3135 | ||
3136 | static ssize_t nvme_sysfs_reset(struct device *dev, | |
3137 | struct device_attribute *attr, const char *buf, | |
3138 | size_t count) | |
3139 | { | |
3140 | struct nvme_dev *ndev = dev_get_drvdata(dev); | |
3141 | int ret; | |
3142 | ||
3143 | ret = nvme_reset(ndev); | |
3144 | if (ret < 0) | |
3145 | return ret; | |
3146 | ||
3147 | return count; | |
3148 | } | |
3149 | static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset); | |
3150 | ||
2e1d8448 | 3151 | static void nvme_async_probe(struct work_struct *work); |
8d85fce7 | 3152 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 3153 | { |
a4aea562 | 3154 | int node, result = -ENOMEM; |
b60503ba MW |
3155 | struct nvme_dev *dev; |
3156 | ||
a4aea562 MB |
3157 | node = dev_to_node(&pdev->dev); |
3158 | if (node == NUMA_NO_NODE) | |
3159 | set_dev_node(&pdev->dev, 0); | |
3160 | ||
3161 | dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); | |
b60503ba MW |
3162 | if (!dev) |
3163 | return -ENOMEM; | |
a4aea562 MB |
3164 | dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry), |
3165 | GFP_KERNEL, node); | |
b60503ba MW |
3166 | if (!dev->entry) |
3167 | goto free; | |
a4aea562 MB |
3168 | dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *), |
3169 | GFP_KERNEL, node); | |
b60503ba MW |
3170 | if (!dev->queues) |
3171 | goto free; | |
3172 | ||
3173 | INIT_LIST_HEAD(&dev->namespaces); | |
9ca97374 TH |
3174 | dev->reset_workfn = nvme_reset_failed_dev; |
3175 | INIT_WORK(&dev->reset_work, nvme_reset_workfn); | |
e75ec752 | 3176 | dev->dev = get_device(&pdev->dev); |
9a6b9458 | 3177 | pci_set_drvdata(pdev, dev); |
cd58ad7d QSA |
3178 | result = nvme_set_instance(dev); |
3179 | if (result) | |
a96d4f5c | 3180 | goto put_pci; |
b60503ba | 3181 | |
091b6092 MW |
3182 | result = nvme_setup_prp_pools(dev); |
3183 | if (result) | |
0877cb0d | 3184 | goto release; |
091b6092 | 3185 | |
fb35e914 | 3186 | kref_init(&dev->kref); |
b3fffdef KB |
3187 | dev->device = device_create(nvme_class, &pdev->dev, |
3188 | MKDEV(nvme_char_major, dev->instance), | |
3189 | dev, "nvme%d", dev->instance); | |
3190 | if (IS_ERR(dev->device)) { | |
3191 | result = PTR_ERR(dev->device); | |
2e1d8448 | 3192 | goto release_pools; |
b3fffdef KB |
3193 | } |
3194 | get_device(dev->device); | |
4cc06521 KB |
3195 | dev_set_drvdata(dev->device, dev); |
3196 | ||
3197 | result = device_create_file(dev->device, &dev_attr_reset_controller); | |
3198 | if (result) | |
3199 | goto put_dev; | |
740216fc | 3200 | |
e6e96d73 | 3201 | INIT_LIST_HEAD(&dev->node); |
a5768aa8 | 3202 | INIT_WORK(&dev->scan_work, nvme_dev_scan); |
2e1d8448 KB |
3203 | INIT_WORK(&dev->probe_work, nvme_async_probe); |
3204 | schedule_work(&dev->probe_work); | |
b60503ba MW |
3205 | return 0; |
3206 | ||
4cc06521 KB |
3207 | put_dev: |
3208 | device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance)); | |
3209 | put_device(dev->device); | |
0877cb0d | 3210 | release_pools: |
091b6092 | 3211 | nvme_release_prp_pools(dev); |
0877cb0d KB |
3212 | release: |
3213 | nvme_release_instance(dev); | |
a96d4f5c | 3214 | put_pci: |
e75ec752 | 3215 | put_device(dev->dev); |
b60503ba MW |
3216 | free: |
3217 | kfree(dev->queues); | |
3218 | kfree(dev->entry); | |
3219 | kfree(dev); | |
3220 | return result; | |
3221 | } | |
3222 | ||
2e1d8448 KB |
3223 | static void nvme_async_probe(struct work_struct *work) |
3224 | { | |
3225 | struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work); | |
2e1d8448 | 3226 | |
de3eff2b KB |
3227 | if (nvme_dev_resume(dev) && !work_busy(&dev->reset_work)) |
3228 | nvme_dead_ctrl(dev); | |
2e1d8448 KB |
3229 | } |
3230 | ||
f0d54a54 KB |
3231 | static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) |
3232 | { | |
a6739479 | 3233 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
f0d54a54 | 3234 | |
a6739479 KB |
3235 | if (prepare) |
3236 | nvme_dev_shutdown(dev); | |
3237 | else | |
3238 | nvme_dev_resume(dev); | |
f0d54a54 KB |
3239 | } |
3240 | ||
09ece142 KB |
3241 | static void nvme_shutdown(struct pci_dev *pdev) |
3242 | { | |
3243 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
3244 | nvme_dev_shutdown(dev); | |
3245 | } | |
3246 | ||
8d85fce7 | 3247 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
3248 | { |
3249 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 KB |
3250 | |
3251 | spin_lock(&dev_list_lock); | |
3252 | list_del_init(&dev->node); | |
3253 | spin_unlock(&dev_list_lock); | |
3254 | ||
3255 | pci_set_drvdata(pdev, NULL); | |
2e1d8448 | 3256 | flush_work(&dev->probe_work); |
9a6b9458 | 3257 | flush_work(&dev->reset_work); |
a5768aa8 | 3258 | flush_work(&dev->scan_work); |
4cc06521 | 3259 | device_remove_file(dev->device, &dev_attr_reset_controller); |
c9d3bf88 | 3260 | nvme_dev_remove(dev); |
3399a3f7 | 3261 | nvme_dev_shutdown(dev); |
a4aea562 | 3262 | nvme_dev_remove_admin(dev); |
b3fffdef | 3263 | device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance)); |
a1a5ef99 | 3264 | nvme_free_queues(dev, 0); |
8ffaadf7 | 3265 | nvme_release_cmb(dev); |
9a6b9458 | 3266 | nvme_release_prp_pools(dev); |
5e82e952 | 3267 | kref_put(&dev->kref, nvme_free_dev); |
b60503ba MW |
3268 | } |
3269 | ||
3270 | /* These functions are yet to be implemented */ | |
3271 | #define nvme_error_detected NULL | |
3272 | #define nvme_dump_registers NULL | |
3273 | #define nvme_link_reset NULL | |
3274 | #define nvme_slot_reset NULL | |
3275 | #define nvme_error_resume NULL | |
cd638946 | 3276 | |
671a6018 | 3277 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
3278 | static int nvme_suspend(struct device *dev) |
3279 | { | |
3280 | struct pci_dev *pdev = to_pci_dev(dev); | |
3281 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
3282 | ||
3283 | nvme_dev_shutdown(ndev); | |
3284 | return 0; | |
3285 | } | |
3286 | ||
3287 | static int nvme_resume(struct device *dev) | |
3288 | { | |
3289 | struct pci_dev *pdev = to_pci_dev(dev); | |
3290 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 3291 | |
9a6b9458 | 3292 | if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) { |
9ca97374 | 3293 | ndev->reset_workfn = nvme_reset_failed_dev; |
9a6b9458 KB |
3294 | queue_work(nvme_workq, &ndev->reset_work); |
3295 | } | |
3296 | return 0; | |
cd638946 | 3297 | } |
671a6018 | 3298 | #endif |
cd638946 KB |
3299 | |
3300 | static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); | |
b60503ba | 3301 | |
1d352035 | 3302 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba MW |
3303 | .error_detected = nvme_error_detected, |
3304 | .mmio_enabled = nvme_dump_registers, | |
3305 | .link_reset = nvme_link_reset, | |
3306 | .slot_reset = nvme_slot_reset, | |
3307 | .resume = nvme_error_resume, | |
f0d54a54 | 3308 | .reset_notify = nvme_reset_notify, |
b60503ba MW |
3309 | }; |
3310 | ||
3311 | /* Move to pci_ids.h later */ | |
3312 | #define PCI_CLASS_STORAGE_EXPRESS 0x010802 | |
3313 | ||
6eb0d698 | 3314 | static const struct pci_device_id nvme_id_table[] = { |
b60503ba MW |
3315 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
3316 | { 0, } | |
3317 | }; | |
3318 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
3319 | ||
3320 | static struct pci_driver nvme_driver = { | |
3321 | .name = "nvme", | |
3322 | .id_table = nvme_id_table, | |
3323 | .probe = nvme_probe, | |
8d85fce7 | 3324 | .remove = nvme_remove, |
09ece142 | 3325 | .shutdown = nvme_shutdown, |
cd638946 KB |
3326 | .driver = { |
3327 | .pm = &nvme_dev_pm_ops, | |
3328 | }, | |
b60503ba MW |
3329 | .err_handler = &nvme_err_handler, |
3330 | }; | |
3331 | ||
3332 | static int __init nvme_init(void) | |
3333 | { | |
0ac13140 | 3334 | int result; |
1fa6aead | 3335 | |
b9afca3e | 3336 | init_waitqueue_head(&nvme_kthread_wait); |
b60503ba | 3337 | |
9a6b9458 KB |
3338 | nvme_workq = create_singlethread_workqueue("nvme"); |
3339 | if (!nvme_workq) | |
b9afca3e | 3340 | return -ENOMEM; |
9a6b9458 | 3341 | |
5c42ea16 KB |
3342 | result = register_blkdev(nvme_major, "nvme"); |
3343 | if (result < 0) | |
9a6b9458 | 3344 | goto kill_workq; |
5c42ea16 | 3345 | else if (result > 0) |
0ac13140 | 3346 | nvme_major = result; |
b60503ba | 3347 | |
b3fffdef KB |
3348 | result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme", |
3349 | &nvme_dev_fops); | |
3350 | if (result < 0) | |
3351 | goto unregister_blkdev; | |
3352 | else if (result > 0) | |
3353 | nvme_char_major = result; | |
3354 | ||
3355 | nvme_class = class_create(THIS_MODULE, "nvme"); | |
c727040b AK |
3356 | if (IS_ERR(nvme_class)) { |
3357 | result = PTR_ERR(nvme_class); | |
b3fffdef | 3358 | goto unregister_chrdev; |
c727040b | 3359 | } |
b3fffdef | 3360 | |
f3db22fe KB |
3361 | result = pci_register_driver(&nvme_driver); |
3362 | if (result) | |
b3fffdef | 3363 | goto destroy_class; |
1fa6aead | 3364 | return 0; |
b60503ba | 3365 | |
b3fffdef KB |
3366 | destroy_class: |
3367 | class_destroy(nvme_class); | |
3368 | unregister_chrdev: | |
3369 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
1fa6aead | 3370 | unregister_blkdev: |
b60503ba | 3371 | unregister_blkdev(nvme_major, "nvme"); |
9a6b9458 KB |
3372 | kill_workq: |
3373 | destroy_workqueue(nvme_workq); | |
b60503ba MW |
3374 | return result; |
3375 | } | |
3376 | ||
3377 | static void __exit nvme_exit(void) | |
3378 | { | |
3379 | pci_unregister_driver(&nvme_driver); | |
3380 | unregister_blkdev(nvme_major, "nvme"); | |
9a6b9458 | 3381 | destroy_workqueue(nvme_workq); |
b3fffdef KB |
3382 | class_destroy(nvme_class); |
3383 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
b9afca3e | 3384 | BUG_ON(nvme_thread && !IS_ERR(nvme_thread)); |
21bd78bc | 3385 | _nvme_check_size(); |
b60503ba MW |
3386 | } |
3387 | ||
3388 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
3389 | MODULE_LICENSE("GPL"); | |
c78b4713 | 3390 | MODULE_VERSION("1.0"); |
b60503ba MW |
3391 | module_init(nvme_init); |
3392 | module_exit(nvme_exit); |