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 MW |
31 | #include <linux/kernel.h> |
32 | #include <linux/mm.h> | |
33 | #include <linux/module.h> | |
34 | #include <linux/moduleparam.h> | |
35 | #include <linux/pci.h> | |
be7b6275 | 36 | #include <linux/poison.h> |
c3bfe717 | 37 | #include <linux/ptrace.h> |
b60503ba MW |
38 | #include <linux/sched.h> |
39 | #include <linux/slab.h> | |
e1e5e564 | 40 | #include <linux/t10-pi.h> |
b60503ba | 41 | #include <linux/types.h> |
5d0f6131 | 42 | #include <scsi/sg.h> |
797a796a HM |
43 | #include <asm-generic/io-64-nonatomic-lo-hi.h> |
44 | ||
b3fffdef | 45 | #define NVME_MINORS (1U << MINORBITS) |
9d43cf64 | 46 | #define NVME_Q_DEPTH 1024 |
d31af0a3 | 47 | #define NVME_AQ_DEPTH 256 |
b60503ba MW |
48 | #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) |
49 | #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) | |
9d43cf64 | 50 | #define ADMIN_TIMEOUT (admin_timeout * HZ) |
2484f407 | 51 | #define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ) |
9d43cf64 KB |
52 | |
53 | static unsigned char admin_timeout = 60; | |
54 | module_param(admin_timeout, byte, 0644); | |
55 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); | |
b60503ba | 56 | |
bd67608a MW |
57 | unsigned char nvme_io_timeout = 30; |
58 | module_param_named(io_timeout, nvme_io_timeout, byte, 0644); | |
b355084a | 59 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); |
b60503ba | 60 | |
2484f407 DM |
61 | static unsigned char shutdown_timeout = 5; |
62 | module_param(shutdown_timeout, byte, 0644); | |
63 | MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); | |
64 | ||
b60503ba MW |
65 | static int nvme_major; |
66 | module_param(nvme_major, int, 0); | |
67 | ||
b3fffdef KB |
68 | static int nvme_char_major; |
69 | module_param(nvme_char_major, int, 0); | |
70 | ||
58ffacb5 MW |
71 | static int use_threaded_interrupts; |
72 | module_param(use_threaded_interrupts, int, 0); | |
73 | ||
1fa6aead MW |
74 | static DEFINE_SPINLOCK(dev_list_lock); |
75 | static LIST_HEAD(dev_list); | |
76 | static struct task_struct *nvme_thread; | |
9a6b9458 | 77 | static struct workqueue_struct *nvme_workq; |
b9afca3e | 78 | static wait_queue_head_t nvme_kthread_wait; |
1fa6aead | 79 | |
b3fffdef KB |
80 | static struct class *nvme_class; |
81 | ||
d4b4ff8e | 82 | static void nvme_reset_failed_dev(struct work_struct *ws); |
a4aea562 | 83 | static int nvme_process_cq(struct nvme_queue *nvmeq); |
d4b4ff8e | 84 | |
4d115420 KB |
85 | struct async_cmd_info { |
86 | struct kthread_work work; | |
87 | struct kthread_worker *worker; | |
a4aea562 | 88 | struct request *req; |
4d115420 KB |
89 | u32 result; |
90 | int status; | |
91 | void *ctx; | |
92 | }; | |
1fa6aead | 93 | |
b60503ba MW |
94 | /* |
95 | * An NVM Express queue. Each device has at least two (one for admin | |
96 | * commands and one for I/O commands). | |
97 | */ | |
98 | struct nvme_queue { | |
99 | struct device *q_dmadev; | |
091b6092 | 100 | struct nvme_dev *dev; |
3193f07b | 101 | char irqname[24]; /* nvme4294967295-65535\0 */ |
b60503ba MW |
102 | spinlock_t q_lock; |
103 | struct nvme_command *sq_cmds; | |
104 | volatile struct nvme_completion *cqes; | |
105 | dma_addr_t sq_dma_addr; | |
106 | dma_addr_t cq_dma_addr; | |
b60503ba MW |
107 | u32 __iomem *q_db; |
108 | u16 q_depth; | |
6222d172 | 109 | s16 cq_vector; |
b60503ba MW |
110 | u16 sq_head; |
111 | u16 sq_tail; | |
112 | u16 cq_head; | |
c30341dc | 113 | u16 qid; |
e9539f47 MW |
114 | u8 cq_phase; |
115 | u8 cqe_seen; | |
4d115420 | 116 | struct async_cmd_info cmdinfo; |
a4aea562 | 117 | struct blk_mq_hw_ctx *hctx; |
b60503ba MW |
118 | }; |
119 | ||
120 | /* | |
121 | * Check we didin't inadvertently grow the command struct | |
122 | */ | |
123 | static inline void _nvme_check_size(void) | |
124 | { | |
125 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
126 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); | |
127 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
128 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
129 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
f8ebf840 | 130 | BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); |
c30341dc | 131 | BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); |
b60503ba MW |
132 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); |
133 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); | |
134 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); | |
135 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
6ecec745 | 136 | BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); |
b60503ba MW |
137 | } |
138 | ||
edd10d33 | 139 | typedef void (*nvme_completion_fn)(struct nvme_queue *, void *, |
c2f5b650 MW |
140 | struct nvme_completion *); |
141 | ||
e85248e5 | 142 | struct nvme_cmd_info { |
c2f5b650 MW |
143 | nvme_completion_fn fn; |
144 | void *ctx; | |
c30341dc | 145 | int aborted; |
a4aea562 | 146 | struct nvme_queue *nvmeq; |
ac3dd5bd | 147 | struct nvme_iod iod[0]; |
e85248e5 MW |
148 | }; |
149 | ||
ac3dd5bd JA |
150 | /* |
151 | * Max size of iod being embedded in the request payload | |
152 | */ | |
153 | #define NVME_INT_PAGES 2 | |
154 | #define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->page_size) | |
fda631ff | 155 | #define NVME_INT_MASK 0x01 |
ac3dd5bd JA |
156 | |
157 | /* | |
158 | * Will slightly overestimate the number of pages needed. This is OK | |
159 | * as it only leads to a small amount of wasted memory for the lifetime of | |
160 | * the I/O. | |
161 | */ | |
162 | static int nvme_npages(unsigned size, struct nvme_dev *dev) | |
163 | { | |
164 | unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size); | |
165 | return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); | |
166 | } | |
167 | ||
168 | static unsigned int nvme_cmd_size(struct nvme_dev *dev) | |
169 | { | |
170 | unsigned int ret = sizeof(struct nvme_cmd_info); | |
171 | ||
172 | ret += sizeof(struct nvme_iod); | |
173 | ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev); | |
174 | ret += sizeof(struct scatterlist) * NVME_INT_PAGES; | |
175 | ||
176 | return ret; | |
177 | } | |
178 | ||
a4aea562 MB |
179 | static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
180 | unsigned int hctx_idx) | |
e85248e5 | 181 | { |
a4aea562 MB |
182 | struct nvme_dev *dev = data; |
183 | struct nvme_queue *nvmeq = dev->queues[0]; | |
184 | ||
185 | WARN_ON(nvmeq->hctx); | |
186 | nvmeq->hctx = hctx; | |
187 | hctx->driver_data = nvmeq; | |
188 | return 0; | |
e85248e5 MW |
189 | } |
190 | ||
a4aea562 MB |
191 | static int nvme_admin_init_request(void *data, struct request *req, |
192 | unsigned int hctx_idx, unsigned int rq_idx, | |
193 | unsigned int numa_node) | |
22404274 | 194 | { |
a4aea562 MB |
195 | struct nvme_dev *dev = data; |
196 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
197 | struct nvme_queue *nvmeq = dev->queues[0]; | |
198 | ||
199 | BUG_ON(!nvmeq); | |
200 | cmd->nvmeq = nvmeq; | |
201 | return 0; | |
22404274 KB |
202 | } |
203 | ||
2c30540b JA |
204 | static void nvme_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) |
205 | { | |
206 | struct nvme_queue *nvmeq = hctx->driver_data; | |
207 | ||
208 | nvmeq->hctx = NULL; | |
209 | } | |
210 | ||
a4aea562 MB |
211 | static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
212 | unsigned int hctx_idx) | |
b60503ba | 213 | { |
a4aea562 MB |
214 | struct nvme_dev *dev = data; |
215 | struct nvme_queue *nvmeq = dev->queues[ | |
216 | (hctx_idx % dev->queue_count) + 1]; | |
b60503ba | 217 | |
a4aea562 MB |
218 | if (!nvmeq->hctx) |
219 | nvmeq->hctx = hctx; | |
220 | ||
221 | /* nvmeq queues are shared between namespaces. We assume here that | |
222 | * blk-mq map the tags so they match up with the nvme queue tags. */ | |
223 | WARN_ON(nvmeq->hctx->tags != hctx->tags); | |
b60503ba | 224 | |
a4aea562 MB |
225 | hctx->driver_data = nvmeq; |
226 | return 0; | |
b60503ba MW |
227 | } |
228 | ||
a4aea562 MB |
229 | static int nvme_init_request(void *data, struct request *req, |
230 | unsigned int hctx_idx, unsigned int rq_idx, | |
231 | unsigned int numa_node) | |
b60503ba | 232 | { |
a4aea562 MB |
233 | struct nvme_dev *dev = data; |
234 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
235 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; | |
236 | ||
237 | BUG_ON(!nvmeq); | |
238 | cmd->nvmeq = nvmeq; | |
239 | return 0; | |
240 | } | |
241 | ||
242 | static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx, | |
243 | nvme_completion_fn handler) | |
244 | { | |
245 | cmd->fn = handler; | |
246 | cmd->ctx = ctx; | |
247 | cmd->aborted = 0; | |
c917dfe5 | 248 | blk_mq_start_request(blk_mq_rq_from_pdu(cmd)); |
b60503ba MW |
249 | } |
250 | ||
ac3dd5bd JA |
251 | static void *iod_get_private(struct nvme_iod *iod) |
252 | { | |
253 | return (void *) (iod->private & ~0x1UL); | |
254 | } | |
255 | ||
256 | /* | |
257 | * If bit 0 is set, the iod is embedded in the request payload. | |
258 | */ | |
259 | static bool iod_should_kfree(struct nvme_iod *iod) | |
260 | { | |
fda631ff | 261 | return (iod->private & NVME_INT_MASK) == 0; |
ac3dd5bd JA |
262 | } |
263 | ||
c2f5b650 MW |
264 | /* Special values must be less than 0x1000 */ |
265 | #define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA) | |
d2d87034 MW |
266 | #define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE) |
267 | #define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE) | |
268 | #define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE) | |
be7b6275 | 269 | |
edd10d33 | 270 | static void special_completion(struct nvme_queue *nvmeq, void *ctx, |
c2f5b650 MW |
271 | struct nvme_completion *cqe) |
272 | { | |
273 | if (ctx == CMD_CTX_CANCELLED) | |
274 | return; | |
c2f5b650 | 275 | if (ctx == CMD_CTX_COMPLETED) { |
edd10d33 | 276 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
277 | "completed id %d twice on queue %d\n", |
278 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
279 | return; | |
280 | } | |
281 | if (ctx == CMD_CTX_INVALID) { | |
edd10d33 | 282 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
283 | "invalid id %d completed on queue %d\n", |
284 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
285 | return; | |
286 | } | |
edd10d33 | 287 | dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx); |
c2f5b650 MW |
288 | } |
289 | ||
a4aea562 | 290 | static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn) |
b60503ba | 291 | { |
c2f5b650 | 292 | void *ctx; |
b60503ba | 293 | |
859361a2 | 294 | if (fn) |
a4aea562 MB |
295 | *fn = cmd->fn; |
296 | ctx = cmd->ctx; | |
297 | cmd->fn = special_completion; | |
298 | cmd->ctx = CMD_CTX_CANCELLED; | |
c2f5b650 | 299 | return ctx; |
b60503ba MW |
300 | } |
301 | ||
a4aea562 MB |
302 | static void async_req_completion(struct nvme_queue *nvmeq, void *ctx, |
303 | struct nvme_completion *cqe) | |
3c0cf138 | 304 | { |
a4aea562 MB |
305 | u32 result = le32_to_cpup(&cqe->result); |
306 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
307 | ||
308 | if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) | |
309 | ++nvmeq->dev->event_limit; | |
310 | if (status == NVME_SC_SUCCESS) | |
311 | dev_warn(nvmeq->q_dmadev, | |
312 | "async event result %08x\n", result); | |
b60503ba MW |
313 | } |
314 | ||
a4aea562 MB |
315 | static void abort_completion(struct nvme_queue *nvmeq, void *ctx, |
316 | struct nvme_completion *cqe) | |
5a92e700 | 317 | { |
a4aea562 MB |
318 | struct request *req = ctx; |
319 | ||
320 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
321 | u32 result = le32_to_cpup(&cqe->result); | |
a51afb54 | 322 | |
9d135bb8 | 323 | blk_mq_free_hctx_request(nvmeq->hctx, req); |
a51afb54 | 324 | |
a4aea562 MB |
325 | dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result); |
326 | ++nvmeq->dev->abort_limit; | |
5a92e700 KB |
327 | } |
328 | ||
a4aea562 MB |
329 | static void async_completion(struct nvme_queue *nvmeq, void *ctx, |
330 | struct nvme_completion *cqe) | |
b60503ba | 331 | { |
a4aea562 MB |
332 | struct async_cmd_info *cmdinfo = ctx; |
333 | cmdinfo->result = le32_to_cpup(&cqe->result); | |
334 | cmdinfo->status = le16_to_cpup(&cqe->status) >> 1; | |
335 | queue_kthread_work(cmdinfo->worker, &cmdinfo->work); | |
9d135bb8 | 336 | blk_mq_free_hctx_request(nvmeq->hctx, cmdinfo->req); |
b60503ba MW |
337 | } |
338 | ||
a4aea562 MB |
339 | static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq, |
340 | unsigned int tag) | |
b60503ba | 341 | { |
a4aea562 MB |
342 | struct blk_mq_hw_ctx *hctx = nvmeq->hctx; |
343 | struct request *req = blk_mq_tag_to_rq(hctx->tags, tag); | |
a51afb54 | 344 | |
a4aea562 | 345 | return blk_mq_rq_to_pdu(req); |
4f5099af KB |
346 | } |
347 | ||
a4aea562 MB |
348 | /* |
349 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
350 | */ | |
351 | static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag, | |
352 | nvme_completion_fn *fn) | |
4f5099af | 353 | { |
a4aea562 MB |
354 | struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag); |
355 | void *ctx; | |
356 | if (tag >= nvmeq->q_depth) { | |
357 | *fn = special_completion; | |
358 | return CMD_CTX_INVALID; | |
359 | } | |
360 | if (fn) | |
361 | *fn = cmd->fn; | |
362 | ctx = cmd->ctx; | |
363 | cmd->fn = special_completion; | |
364 | cmd->ctx = CMD_CTX_COMPLETED; | |
365 | return ctx; | |
b60503ba MW |
366 | } |
367 | ||
368 | /** | |
714a7a22 | 369 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
370 | * @nvmeq: The queue to use |
371 | * @cmd: The command to send | |
372 | * | |
373 | * Safe to use from interrupt context | |
374 | */ | |
a4aea562 | 375 | static int __nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd) |
b60503ba | 376 | { |
a4aea562 MB |
377 | u16 tail = nvmeq->sq_tail; |
378 | ||
b60503ba | 379 | memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); |
b60503ba MW |
380 | if (++tail == nvmeq->q_depth) |
381 | tail = 0; | |
7547881d | 382 | writel(tail, nvmeq->q_db); |
b60503ba | 383 | nvmeq->sq_tail = tail; |
b60503ba MW |
384 | |
385 | return 0; | |
386 | } | |
387 | ||
a4aea562 MB |
388 | static int nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd) |
389 | { | |
390 | unsigned long flags; | |
391 | int ret; | |
392 | spin_lock_irqsave(&nvmeq->q_lock, flags); | |
393 | ret = __nvme_submit_cmd(nvmeq, cmd); | |
394 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); | |
395 | return ret; | |
396 | } | |
397 | ||
eca18b23 | 398 | static __le64 **iod_list(struct nvme_iod *iod) |
e025344c | 399 | { |
eca18b23 | 400 | return ((void *)iod) + iod->offset; |
e025344c SMM |
401 | } |
402 | ||
ac3dd5bd JA |
403 | static inline void iod_init(struct nvme_iod *iod, unsigned nbytes, |
404 | unsigned nseg, unsigned long private) | |
eca18b23 | 405 | { |
ac3dd5bd JA |
406 | iod->private = private; |
407 | iod->offset = offsetof(struct nvme_iod, sg[nseg]); | |
408 | iod->npages = -1; | |
409 | iod->length = nbytes; | |
410 | iod->nents = 0; | |
eca18b23 | 411 | } |
b60503ba | 412 | |
eca18b23 | 413 | static struct nvme_iod * |
ac3dd5bd JA |
414 | __nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev, |
415 | unsigned long priv, gfp_t gfp) | |
b60503ba | 416 | { |
eca18b23 | 417 | struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) + |
ac3dd5bd | 418 | sizeof(__le64 *) * nvme_npages(bytes, dev) + |
eca18b23 MW |
419 | sizeof(struct scatterlist) * nseg, gfp); |
420 | ||
ac3dd5bd JA |
421 | if (iod) |
422 | iod_init(iod, bytes, nseg, priv); | |
eca18b23 MW |
423 | |
424 | return iod; | |
b60503ba MW |
425 | } |
426 | ||
ac3dd5bd JA |
427 | static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev, |
428 | gfp_t gfp) | |
429 | { | |
430 | unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) : | |
431 | sizeof(struct nvme_dsm_range); | |
ac3dd5bd JA |
432 | struct nvme_iod *iod; |
433 | ||
434 | if (rq->nr_phys_segments <= NVME_INT_PAGES && | |
435 | size <= NVME_INT_BYTES(dev)) { | |
436 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq); | |
437 | ||
438 | iod = cmd->iod; | |
ac3dd5bd | 439 | iod_init(iod, size, rq->nr_phys_segments, |
fda631ff | 440 | (unsigned long) rq | NVME_INT_MASK); |
ac3dd5bd JA |
441 | return iod; |
442 | } | |
443 | ||
444 | return __nvme_alloc_iod(rq->nr_phys_segments, size, dev, | |
445 | (unsigned long) rq, gfp); | |
446 | } | |
447 | ||
d29ec824 | 448 | static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod) |
b60503ba | 449 | { |
1d090624 | 450 | const int last_prp = dev->page_size / 8 - 1; |
eca18b23 MW |
451 | int i; |
452 | __le64 **list = iod_list(iod); | |
453 | dma_addr_t prp_dma = iod->first_dma; | |
454 | ||
455 | if (iod->npages == 0) | |
456 | dma_pool_free(dev->prp_small_pool, list[0], prp_dma); | |
457 | for (i = 0; i < iod->npages; i++) { | |
458 | __le64 *prp_list = list[i]; | |
459 | dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); | |
460 | dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); | |
461 | prp_dma = next_prp_dma; | |
462 | } | |
ac3dd5bd JA |
463 | |
464 | if (iod_should_kfree(iod)) | |
465 | kfree(iod); | |
b60503ba MW |
466 | } |
467 | ||
b4ff9c8d KB |
468 | static int nvme_error_status(u16 status) |
469 | { | |
470 | switch (status & 0x7ff) { | |
471 | case NVME_SC_SUCCESS: | |
472 | return 0; | |
473 | case NVME_SC_CAP_EXCEEDED: | |
474 | return -ENOSPC; | |
475 | default: | |
476 | return -EIO; | |
477 | } | |
478 | } | |
479 | ||
52b68d7e | 480 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
e1e5e564 KB |
481 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) |
482 | { | |
483 | if (be32_to_cpu(pi->ref_tag) == v) | |
484 | pi->ref_tag = cpu_to_be32(p); | |
485 | } | |
486 | ||
487 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
488 | { | |
489 | if (be32_to_cpu(pi->ref_tag) == p) | |
490 | pi->ref_tag = cpu_to_be32(v); | |
491 | } | |
492 | ||
493 | /** | |
494 | * nvme_dif_remap - remaps ref tags to bip seed and physical lba | |
495 | * | |
496 | * The virtual start sector is the one that was originally submitted by the | |
497 | * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical | |
498 | * start sector may be different. Remap protection information to match the | |
499 | * physical LBA on writes, and back to the original seed on reads. | |
500 | * | |
501 | * Type 0 and 3 do not have a ref tag, so no remapping required. | |
502 | */ | |
503 | static void nvme_dif_remap(struct request *req, | |
504 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
505 | { | |
506 | struct nvme_ns *ns = req->rq_disk->private_data; | |
507 | struct bio_integrity_payload *bip; | |
508 | struct t10_pi_tuple *pi; | |
509 | void *p, *pmap; | |
510 | u32 i, nlb, ts, phys, virt; | |
511 | ||
512 | if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3) | |
513 | return; | |
514 | ||
515 | bip = bio_integrity(req->bio); | |
516 | if (!bip) | |
517 | return; | |
518 | ||
519 | pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset; | |
e1e5e564 KB |
520 | |
521 | p = pmap; | |
522 | virt = bip_get_seed(bip); | |
523 | phys = nvme_block_nr(ns, blk_rq_pos(req)); | |
524 | nlb = (blk_rq_bytes(req) >> ns->lba_shift); | |
525 | ts = ns->disk->integrity->tuple_size; | |
526 | ||
527 | for (i = 0; i < nlb; i++, virt++, phys++) { | |
528 | pi = (struct t10_pi_tuple *)p; | |
529 | dif_swap(phys, virt, pi); | |
530 | p += ts; | |
531 | } | |
532 | kunmap_atomic(pmap); | |
533 | } | |
534 | ||
52b68d7e KB |
535 | static int nvme_noop_verify(struct blk_integrity_iter *iter) |
536 | { | |
537 | return 0; | |
538 | } | |
539 | ||
540 | static int nvme_noop_generate(struct blk_integrity_iter *iter) | |
541 | { | |
542 | return 0; | |
543 | } | |
544 | ||
545 | struct blk_integrity nvme_meta_noop = { | |
546 | .name = "NVME_META_NOOP", | |
547 | .generate_fn = nvme_noop_generate, | |
548 | .verify_fn = nvme_noop_verify, | |
549 | }; | |
550 | ||
551 | static void nvme_init_integrity(struct nvme_ns *ns) | |
552 | { | |
553 | struct blk_integrity integrity; | |
554 | ||
555 | switch (ns->pi_type) { | |
556 | case NVME_NS_DPS_PI_TYPE3: | |
557 | integrity = t10_pi_type3_crc; | |
558 | break; | |
559 | case NVME_NS_DPS_PI_TYPE1: | |
560 | case NVME_NS_DPS_PI_TYPE2: | |
561 | integrity = t10_pi_type1_crc; | |
562 | break; | |
563 | default: | |
564 | integrity = nvme_meta_noop; | |
565 | break; | |
566 | } | |
567 | integrity.tuple_size = ns->ms; | |
568 | blk_integrity_register(ns->disk, &integrity); | |
569 | blk_queue_max_integrity_segments(ns->queue, 1); | |
570 | } | |
571 | #else /* CONFIG_BLK_DEV_INTEGRITY */ | |
572 | static void nvme_dif_remap(struct request *req, | |
573 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
574 | { | |
575 | } | |
576 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) | |
577 | { | |
578 | } | |
579 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
580 | { | |
581 | } | |
582 | static void nvme_init_integrity(struct nvme_ns *ns) | |
583 | { | |
584 | } | |
585 | #endif | |
586 | ||
a4aea562 | 587 | static void req_completion(struct nvme_queue *nvmeq, void *ctx, |
b60503ba MW |
588 | struct nvme_completion *cqe) |
589 | { | |
eca18b23 | 590 | struct nvme_iod *iod = ctx; |
ac3dd5bd | 591 | struct request *req = iod_get_private(iod); |
a4aea562 MB |
592 | struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req); |
593 | ||
b60503ba MW |
594 | u16 status = le16_to_cpup(&cqe->status) >> 1; |
595 | ||
edd10d33 | 596 | if (unlikely(status)) { |
a4aea562 MB |
597 | if (!(status & NVME_SC_DNR || blk_noretry_request(req)) |
598 | && (jiffies - req->start_time) < req->timeout) { | |
c9d3bf88 KB |
599 | unsigned long flags; |
600 | ||
a4aea562 | 601 | blk_mq_requeue_request(req); |
c9d3bf88 KB |
602 | spin_lock_irqsave(req->q->queue_lock, flags); |
603 | if (!blk_queue_stopped(req->q)) | |
604 | blk_mq_kick_requeue_list(req->q); | |
605 | spin_unlock_irqrestore(req->q->queue_lock, flags); | |
edd10d33 KB |
606 | return; |
607 | } | |
d29ec824 | 608 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) { |
d29ec824 CH |
609 | req->errors = status; |
610 | } else { | |
611 | req->errors = nvme_error_status(status); | |
612 | } | |
a4aea562 MB |
613 | } else |
614 | req->errors = 0; | |
a0a931d6 KB |
615 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) { |
616 | u32 result = le32_to_cpup(&cqe->result); | |
617 | req->special = (void *)(uintptr_t)result; | |
618 | } | |
a4aea562 MB |
619 | |
620 | if (cmd_rq->aborted) | |
e75ec752 | 621 | dev_warn(nvmeq->dev->dev, |
a4aea562 MB |
622 | "completing aborted command with status:%04x\n", |
623 | status); | |
624 | ||
e1e5e564 | 625 | if (iod->nents) { |
e75ec752 | 626 | dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents, |
a4aea562 | 627 | rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
e1e5e564 KB |
628 | if (blk_integrity_rq(req)) { |
629 | if (!rq_data_dir(req)) | |
630 | nvme_dif_remap(req, nvme_dif_complete); | |
e75ec752 | 631 | dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1, |
e1e5e564 KB |
632 | rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
633 | } | |
634 | } | |
edd10d33 | 635 | nvme_free_iod(nvmeq->dev, iod); |
3291fa57 | 636 | |
a4aea562 | 637 | blk_mq_complete_request(req); |
b60503ba MW |
638 | } |
639 | ||
184d2944 | 640 | /* length is in bytes. gfp flags indicates whether we may sleep. */ |
d29ec824 CH |
641 | static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod, |
642 | int total_len, gfp_t gfp) | |
ff22b54f | 643 | { |
99802a7a | 644 | struct dma_pool *pool; |
eca18b23 MW |
645 | int length = total_len; |
646 | struct scatterlist *sg = iod->sg; | |
ff22b54f MW |
647 | int dma_len = sg_dma_len(sg); |
648 | u64 dma_addr = sg_dma_address(sg); | |
f137e0f1 MI |
649 | u32 page_size = dev->page_size; |
650 | int offset = dma_addr & (page_size - 1); | |
e025344c | 651 | __le64 *prp_list; |
eca18b23 | 652 | __le64 **list = iod_list(iod); |
e025344c | 653 | dma_addr_t prp_dma; |
eca18b23 | 654 | int nprps, i; |
ff22b54f | 655 | |
1d090624 | 656 | length -= (page_size - offset); |
ff22b54f | 657 | if (length <= 0) |
eca18b23 | 658 | return total_len; |
ff22b54f | 659 | |
1d090624 | 660 | dma_len -= (page_size - offset); |
ff22b54f | 661 | if (dma_len) { |
1d090624 | 662 | dma_addr += (page_size - offset); |
ff22b54f MW |
663 | } else { |
664 | sg = sg_next(sg); | |
665 | dma_addr = sg_dma_address(sg); | |
666 | dma_len = sg_dma_len(sg); | |
667 | } | |
668 | ||
1d090624 | 669 | if (length <= page_size) { |
edd10d33 | 670 | iod->first_dma = dma_addr; |
eca18b23 | 671 | return total_len; |
e025344c SMM |
672 | } |
673 | ||
1d090624 | 674 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
675 | if (nprps <= (256 / 8)) { |
676 | pool = dev->prp_small_pool; | |
eca18b23 | 677 | iod->npages = 0; |
99802a7a MW |
678 | } else { |
679 | pool = dev->prp_page_pool; | |
eca18b23 | 680 | iod->npages = 1; |
99802a7a MW |
681 | } |
682 | ||
b77954cb MW |
683 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
684 | if (!prp_list) { | |
edd10d33 | 685 | iod->first_dma = dma_addr; |
eca18b23 | 686 | iod->npages = -1; |
1d090624 | 687 | return (total_len - length) + page_size; |
b77954cb | 688 | } |
eca18b23 MW |
689 | list[0] = prp_list; |
690 | iod->first_dma = prp_dma; | |
e025344c SMM |
691 | i = 0; |
692 | for (;;) { | |
1d090624 | 693 | if (i == page_size >> 3) { |
e025344c | 694 | __le64 *old_prp_list = prp_list; |
b77954cb | 695 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
eca18b23 MW |
696 | if (!prp_list) |
697 | return total_len - length; | |
698 | list[iod->npages++] = prp_list; | |
7523d834 MW |
699 | prp_list[0] = old_prp_list[i - 1]; |
700 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
701 | i = 1; | |
e025344c SMM |
702 | } |
703 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
704 | dma_len -= page_size; |
705 | dma_addr += page_size; | |
706 | length -= page_size; | |
e025344c SMM |
707 | if (length <= 0) |
708 | break; | |
709 | if (dma_len > 0) | |
710 | continue; | |
711 | BUG_ON(dma_len < 0); | |
712 | sg = sg_next(sg); | |
713 | dma_addr = sg_dma_address(sg); | |
714 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
715 | } |
716 | ||
eca18b23 | 717 | return total_len; |
ff22b54f MW |
718 | } |
719 | ||
d29ec824 CH |
720 | static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req, |
721 | struct nvme_iod *iod) | |
722 | { | |
723 | struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; | |
724 | ||
725 | memcpy(cmnd, req->cmd, sizeof(struct nvme_command)); | |
726 | cmnd->rw.command_id = req->tag; | |
727 | if (req->nr_phys_segments) { | |
728 | cmnd->rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
729 | cmnd->rw.prp2 = cpu_to_le64(iod->first_dma); | |
730 | } | |
731 | ||
732 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
733 | nvmeq->sq_tail = 0; | |
734 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
735 | } | |
736 | ||
a4aea562 MB |
737 | /* |
738 | * We reuse the small pool to allocate the 16-byte range here as it is not | |
739 | * worth having a special pool for these or additional cases to handle freeing | |
740 | * the iod. | |
741 | */ | |
742 | static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns, | |
743 | struct request *req, struct nvme_iod *iod) | |
0e5e4f0e | 744 | { |
edd10d33 KB |
745 | struct nvme_dsm_range *range = |
746 | (struct nvme_dsm_range *)iod_list(iod)[0]; | |
0e5e4f0e KB |
747 | struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; |
748 | ||
0e5e4f0e | 749 | range->cattr = cpu_to_le32(0); |
a4aea562 MB |
750 | range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift); |
751 | range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
0e5e4f0e KB |
752 | |
753 | memset(cmnd, 0, sizeof(*cmnd)); | |
754 | cmnd->dsm.opcode = nvme_cmd_dsm; | |
a4aea562 | 755 | cmnd->dsm.command_id = req->tag; |
0e5e4f0e KB |
756 | cmnd->dsm.nsid = cpu_to_le32(ns->ns_id); |
757 | cmnd->dsm.prp1 = cpu_to_le64(iod->first_dma); | |
758 | cmnd->dsm.nr = 0; | |
759 | cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); | |
760 | ||
761 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
762 | nvmeq->sq_tail = 0; | |
763 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
0e5e4f0e KB |
764 | } |
765 | ||
a4aea562 | 766 | static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
00df5cb4 MW |
767 | int cmdid) |
768 | { | |
769 | struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; | |
770 | ||
771 | memset(cmnd, 0, sizeof(*cmnd)); | |
772 | cmnd->common.opcode = nvme_cmd_flush; | |
773 | cmnd->common.command_id = cmdid; | |
774 | cmnd->common.nsid = cpu_to_le32(ns->ns_id); | |
775 | ||
776 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
777 | nvmeq->sq_tail = 0; | |
778 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
00df5cb4 MW |
779 | } |
780 | ||
a4aea562 MB |
781 | static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod, |
782 | struct nvme_ns *ns) | |
b60503ba | 783 | { |
ac3dd5bd | 784 | struct request *req = iod_get_private(iod); |
ff22b54f | 785 | struct nvme_command *cmnd; |
a4aea562 MB |
786 | u16 control = 0; |
787 | u32 dsmgmt = 0; | |
00df5cb4 | 788 | |
a4aea562 | 789 | if (req->cmd_flags & REQ_FUA) |
b60503ba | 790 | control |= NVME_RW_FUA; |
a4aea562 | 791 | if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) |
b60503ba MW |
792 | control |= NVME_RW_LR; |
793 | ||
a4aea562 | 794 | if (req->cmd_flags & REQ_RAHEAD) |
b60503ba MW |
795 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; |
796 | ||
ff22b54f | 797 | cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; |
b8deb62c | 798 | memset(cmnd, 0, sizeof(*cmnd)); |
b60503ba | 799 | |
a4aea562 MB |
800 | cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); |
801 | cmnd->rw.command_id = req->tag; | |
ff22b54f | 802 | cmnd->rw.nsid = cpu_to_le32(ns->ns_id); |
edd10d33 KB |
803 | cmnd->rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); |
804 | cmnd->rw.prp2 = cpu_to_le64(iod->first_dma); | |
a4aea562 MB |
805 | cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); |
806 | cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); | |
e1e5e564 KB |
807 | |
808 | if (blk_integrity_rq(req)) { | |
809 | cmnd->rw.metadata = cpu_to_le64(sg_dma_address(iod->meta_sg)); | |
810 | switch (ns->pi_type) { | |
811 | case NVME_NS_DPS_PI_TYPE3: | |
812 | control |= NVME_RW_PRINFO_PRCHK_GUARD; | |
813 | break; | |
814 | case NVME_NS_DPS_PI_TYPE1: | |
815 | case NVME_NS_DPS_PI_TYPE2: | |
816 | control |= NVME_RW_PRINFO_PRCHK_GUARD | | |
817 | NVME_RW_PRINFO_PRCHK_REF; | |
818 | cmnd->rw.reftag = cpu_to_le32( | |
819 | nvme_block_nr(ns, blk_rq_pos(req))); | |
820 | break; | |
821 | } | |
822 | } else if (ns->ms) | |
823 | control |= NVME_RW_PRINFO_PRACT; | |
824 | ||
ff22b54f MW |
825 | cmnd->rw.control = cpu_to_le16(control); |
826 | cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); | |
b60503ba | 827 | |
b60503ba MW |
828 | if (++nvmeq->sq_tail == nvmeq->q_depth) |
829 | nvmeq->sq_tail = 0; | |
7547881d | 830 | writel(nvmeq->sq_tail, nvmeq->q_db); |
b60503ba | 831 | |
1974b1ae | 832 | return 0; |
edd10d33 KB |
833 | } |
834 | ||
d29ec824 CH |
835 | /* |
836 | * NOTE: ns is NULL when called on the admin queue. | |
837 | */ | |
a4aea562 MB |
838 | static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx, |
839 | const struct blk_mq_queue_data *bd) | |
edd10d33 | 840 | { |
a4aea562 MB |
841 | struct nvme_ns *ns = hctx->queue->queuedata; |
842 | struct nvme_queue *nvmeq = hctx->driver_data; | |
d29ec824 | 843 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 MB |
844 | struct request *req = bd->rq; |
845 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
edd10d33 | 846 | struct nvme_iod *iod; |
a4aea562 | 847 | enum dma_data_direction dma_dir; |
edd10d33 | 848 | |
e1e5e564 KB |
849 | /* |
850 | * If formated with metadata, require the block layer provide a buffer | |
851 | * unless this namespace is formated such that the metadata can be | |
852 | * stripped/generated by the controller with PRACT=1. | |
853 | */ | |
d29ec824 | 854 | if (ns && ns->ms && !blk_integrity_rq(req)) { |
e1e5e564 KB |
855 | if (!(ns->pi_type && ns->ms == 8)) { |
856 | req->errors = -EFAULT; | |
857 | blk_mq_complete_request(req); | |
858 | return BLK_MQ_RQ_QUEUE_OK; | |
859 | } | |
860 | } | |
861 | ||
d29ec824 | 862 | iod = nvme_alloc_iod(req, dev, GFP_ATOMIC); |
edd10d33 | 863 | if (!iod) |
fe54303e | 864 | return BLK_MQ_RQ_QUEUE_BUSY; |
a4aea562 | 865 | |
a4aea562 | 866 | if (req->cmd_flags & REQ_DISCARD) { |
edd10d33 KB |
867 | void *range; |
868 | /* | |
869 | * We reuse the small pool to allocate the 16-byte range here | |
870 | * as it is not worth having a special pool for these or | |
871 | * additional cases to handle freeing the iod. | |
872 | */ | |
d29ec824 | 873 | range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC, |
edd10d33 | 874 | &iod->first_dma); |
a4aea562 | 875 | if (!range) |
fe54303e | 876 | goto retry_cmd; |
edd10d33 KB |
877 | iod_list(iod)[0] = (__le64 *)range; |
878 | iod->npages = 0; | |
ac3dd5bd | 879 | } else if (req->nr_phys_segments) { |
a4aea562 MB |
880 | dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE; |
881 | ||
ac3dd5bd | 882 | sg_init_table(iod->sg, req->nr_phys_segments); |
a4aea562 | 883 | iod->nents = blk_rq_map_sg(req->q, req, iod->sg); |
fe54303e JA |
884 | if (!iod->nents) |
885 | goto error_cmd; | |
a4aea562 MB |
886 | |
887 | if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir)) | |
fe54303e | 888 | goto retry_cmd; |
a4aea562 | 889 | |
fe54303e | 890 | if (blk_rq_bytes(req) != |
d29ec824 CH |
891 | nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) { |
892 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); | |
fe54303e JA |
893 | goto retry_cmd; |
894 | } | |
e1e5e564 KB |
895 | if (blk_integrity_rq(req)) { |
896 | if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) | |
897 | goto error_cmd; | |
898 | ||
899 | sg_init_table(iod->meta_sg, 1); | |
900 | if (blk_rq_map_integrity_sg( | |
901 | req->q, req->bio, iod->meta_sg) != 1) | |
902 | goto error_cmd; | |
903 | ||
904 | if (rq_data_dir(req)) | |
905 | nvme_dif_remap(req, nvme_dif_prep); | |
906 | ||
907 | if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir)) | |
908 | goto error_cmd; | |
909 | } | |
edd10d33 | 910 | } |
1974b1ae | 911 | |
9af8785a | 912 | nvme_set_info(cmd, iod, req_completion); |
a4aea562 | 913 | spin_lock_irq(&nvmeq->q_lock); |
d29ec824 CH |
914 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) |
915 | nvme_submit_priv(nvmeq, req, iod); | |
916 | else if (req->cmd_flags & REQ_DISCARD) | |
a4aea562 MB |
917 | nvme_submit_discard(nvmeq, ns, req, iod); |
918 | else if (req->cmd_flags & REQ_FLUSH) | |
919 | nvme_submit_flush(nvmeq, ns, req->tag); | |
920 | else | |
921 | nvme_submit_iod(nvmeq, iod, ns); | |
922 | ||
923 | nvme_process_cq(nvmeq); | |
924 | spin_unlock_irq(&nvmeq->q_lock); | |
925 | return BLK_MQ_RQ_QUEUE_OK; | |
926 | ||
fe54303e | 927 | error_cmd: |
d29ec824 | 928 | nvme_free_iod(dev, iod); |
fe54303e JA |
929 | return BLK_MQ_RQ_QUEUE_ERROR; |
930 | retry_cmd: | |
d29ec824 | 931 | nvme_free_iod(dev, iod); |
fe54303e | 932 | return BLK_MQ_RQ_QUEUE_BUSY; |
b60503ba MW |
933 | } |
934 | ||
e9539f47 | 935 | static int nvme_process_cq(struct nvme_queue *nvmeq) |
b60503ba | 936 | { |
82123460 | 937 | u16 head, phase; |
b60503ba | 938 | |
b60503ba | 939 | head = nvmeq->cq_head; |
82123460 | 940 | phase = nvmeq->cq_phase; |
b60503ba MW |
941 | |
942 | for (;;) { | |
c2f5b650 MW |
943 | void *ctx; |
944 | nvme_completion_fn fn; | |
b60503ba | 945 | struct nvme_completion cqe = nvmeq->cqes[head]; |
82123460 | 946 | if ((le16_to_cpu(cqe.status) & 1) != phase) |
b60503ba MW |
947 | break; |
948 | nvmeq->sq_head = le16_to_cpu(cqe.sq_head); | |
949 | if (++head == nvmeq->q_depth) { | |
950 | head = 0; | |
82123460 | 951 | phase = !phase; |
b60503ba | 952 | } |
a4aea562 | 953 | ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn); |
edd10d33 | 954 | fn(nvmeq, ctx, &cqe); |
b60503ba MW |
955 | } |
956 | ||
957 | /* If the controller ignores the cq head doorbell and continuously | |
958 | * writes to the queue, it is theoretically possible to wrap around | |
959 | * the queue twice and mistakenly return IRQ_NONE. Linux only | |
960 | * requires that 0.1% of your interrupts are handled, so this isn't | |
961 | * a big problem. | |
962 | */ | |
82123460 | 963 | if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) |
e9539f47 | 964 | return 0; |
b60503ba | 965 | |
b80d5ccc | 966 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); |
b60503ba | 967 | nvmeq->cq_head = head; |
82123460 | 968 | nvmeq->cq_phase = phase; |
b60503ba | 969 | |
e9539f47 MW |
970 | nvmeq->cqe_seen = 1; |
971 | return 1; | |
b60503ba MW |
972 | } |
973 | ||
974 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 MW |
975 | { |
976 | irqreturn_t result; | |
977 | struct nvme_queue *nvmeq = data; | |
978 | spin_lock(&nvmeq->q_lock); | |
e9539f47 MW |
979 | nvme_process_cq(nvmeq); |
980 | result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; | |
981 | nvmeq->cqe_seen = 0; | |
58ffacb5 MW |
982 | spin_unlock(&nvmeq->q_lock); |
983 | return result; | |
984 | } | |
985 | ||
986 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
987 | { | |
988 | struct nvme_queue *nvmeq = data; | |
989 | struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head]; | |
990 | if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase) | |
991 | return IRQ_NONE; | |
992 | return IRQ_WAKE_THREAD; | |
993 | } | |
994 | ||
b60503ba MW |
995 | /* |
996 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
997 | * if the result is positive, it's an NVM Express status code | |
998 | */ | |
d29ec824 CH |
999 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
1000 | void *buffer, void __user *ubuffer, unsigned bufflen, | |
1001 | u32 *result, unsigned timeout) | |
b60503ba | 1002 | { |
d29ec824 CH |
1003 | bool write = cmd->common.opcode & 1; |
1004 | struct bio *bio = NULL; | |
f705f837 | 1005 | struct request *req; |
d29ec824 | 1006 | int ret; |
f705f837 | 1007 | |
d29ec824 | 1008 | req = blk_mq_alloc_request(q, write, GFP_KERNEL, false); |
f705f837 CH |
1009 | if (IS_ERR(req)) |
1010 | return PTR_ERR(req); | |
b60503ba | 1011 | |
d29ec824 CH |
1012 | req->cmd_type = REQ_TYPE_DRV_PRIV; |
1013 | req->__data_len = 0; | |
1014 | req->__sector = (sector_t) -1; | |
1015 | req->bio = req->biotail = NULL; | |
b60503ba | 1016 | |
f4ff414a | 1017 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; |
a4aea562 | 1018 | |
d29ec824 CH |
1019 | req->cmd = (unsigned char *)cmd; |
1020 | req->cmd_len = sizeof(struct nvme_command); | |
a0a931d6 | 1021 | req->special = (void *)0; |
b60503ba | 1022 | |
d29ec824 CH |
1023 | if (buffer && bufflen) { |
1024 | ret = blk_rq_map_kern(q, req, buffer, bufflen, __GFP_WAIT); | |
1025 | if (ret) | |
1026 | goto out; | |
1027 | } else if (ubuffer && bufflen) { | |
1028 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, __GFP_WAIT); | |
1029 | if (ret) | |
1030 | goto out; | |
1031 | bio = req->bio; | |
1032 | } | |
3c0cf138 | 1033 | |
d29ec824 CH |
1034 | blk_execute_rq(req->q, NULL, req, 0); |
1035 | if (bio) | |
1036 | blk_rq_unmap_user(bio); | |
b60503ba | 1037 | if (result) |
a0a931d6 | 1038 | *result = (u32)(uintptr_t)req->special; |
d29ec824 CH |
1039 | ret = req->errors; |
1040 | out: | |
f705f837 | 1041 | blk_mq_free_request(req); |
d29ec824 | 1042 | return ret; |
f705f837 CH |
1043 | } |
1044 | ||
d29ec824 CH |
1045 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
1046 | void *buffer, unsigned bufflen) | |
f705f837 | 1047 | { |
d29ec824 | 1048 | return __nvme_submit_sync_cmd(q, cmd, buffer, NULL, bufflen, NULL, 0); |
b60503ba MW |
1049 | } |
1050 | ||
a4aea562 MB |
1051 | static int nvme_submit_async_admin_req(struct nvme_dev *dev) |
1052 | { | |
1053 | struct nvme_queue *nvmeq = dev->queues[0]; | |
1054 | struct nvme_command c; | |
1055 | struct nvme_cmd_info *cmd_info; | |
1056 | struct request *req; | |
1057 | ||
1efccc9d | 1058 | req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, true); |
9f173b33 DC |
1059 | if (IS_ERR(req)) |
1060 | return PTR_ERR(req); | |
a4aea562 | 1061 | |
c917dfe5 | 1062 | req->cmd_flags |= REQ_NO_TIMEOUT; |
a4aea562 | 1063 | cmd_info = blk_mq_rq_to_pdu(req); |
1efccc9d | 1064 | nvme_set_info(cmd_info, NULL, async_req_completion); |
a4aea562 MB |
1065 | |
1066 | memset(&c, 0, sizeof(c)); | |
1067 | c.common.opcode = nvme_admin_async_event; | |
1068 | c.common.command_id = req->tag; | |
1069 | ||
1efccc9d | 1070 | blk_mq_free_hctx_request(nvmeq->hctx, req); |
a4aea562 MB |
1071 | return __nvme_submit_cmd(nvmeq, &c); |
1072 | } | |
1073 | ||
1074 | static int nvme_submit_admin_async_cmd(struct nvme_dev *dev, | |
4d115420 KB |
1075 | struct nvme_command *cmd, |
1076 | struct async_cmd_info *cmdinfo, unsigned timeout) | |
1077 | { | |
a4aea562 MB |
1078 | struct nvme_queue *nvmeq = dev->queues[0]; |
1079 | struct request *req; | |
1080 | struct nvme_cmd_info *cmd_rq; | |
4d115420 | 1081 | |
a4aea562 | 1082 | req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_KERNEL, false); |
9f173b33 DC |
1083 | if (IS_ERR(req)) |
1084 | return PTR_ERR(req); | |
a4aea562 MB |
1085 | |
1086 | req->timeout = timeout; | |
1087 | cmd_rq = blk_mq_rq_to_pdu(req); | |
1088 | cmdinfo->req = req; | |
1089 | nvme_set_info(cmd_rq, cmdinfo, async_completion); | |
4d115420 | 1090 | cmdinfo->status = -EINTR; |
a4aea562 MB |
1091 | |
1092 | cmd->common.command_id = req->tag; | |
1093 | ||
4f5099af | 1094 | return nvme_submit_cmd(nvmeq, cmd); |
4d115420 KB |
1095 | } |
1096 | ||
b60503ba MW |
1097 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
1098 | { | |
b60503ba MW |
1099 | struct nvme_command c; |
1100 | ||
1101 | memset(&c, 0, sizeof(c)); | |
1102 | c.delete_queue.opcode = opcode; | |
1103 | c.delete_queue.qid = cpu_to_le16(id); | |
1104 | ||
d29ec824 | 1105 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1106 | } |
1107 | ||
1108 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, | |
1109 | struct nvme_queue *nvmeq) | |
1110 | { | |
b60503ba MW |
1111 | struct nvme_command c; |
1112 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; | |
1113 | ||
d29ec824 CH |
1114 | /* |
1115 | * Note: we (ab)use the fact the the prp fields survive if no data | |
1116 | * is attached to the request. | |
1117 | */ | |
b60503ba MW |
1118 | memset(&c, 0, sizeof(c)); |
1119 | c.create_cq.opcode = nvme_admin_create_cq; | |
1120 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
1121 | c.create_cq.cqid = cpu_to_le16(qid); | |
1122 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1123 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
1124 | c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); | |
1125 | ||
d29ec824 | 1126 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1127 | } |
1128 | ||
1129 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
1130 | struct nvme_queue *nvmeq) | |
1131 | { | |
b60503ba MW |
1132 | struct nvme_command c; |
1133 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; | |
1134 | ||
d29ec824 CH |
1135 | /* |
1136 | * Note: we (ab)use the fact the the prp fields survive if no data | |
1137 | * is attached to the request. | |
1138 | */ | |
b60503ba MW |
1139 | memset(&c, 0, sizeof(c)); |
1140 | c.create_sq.opcode = nvme_admin_create_sq; | |
1141 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
1142 | c.create_sq.sqid = cpu_to_le16(qid); | |
1143 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1144 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
1145 | c.create_sq.cqid = cpu_to_le16(qid); | |
1146 | ||
d29ec824 | 1147 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1148 | } |
1149 | ||
1150 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
1151 | { | |
1152 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
1153 | } | |
1154 | ||
1155 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
1156 | { | |
1157 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
1158 | } | |
1159 | ||
d29ec824 | 1160 | int nvme_identify_ctrl(struct nvme_dev *dev, struct nvme_id_ctrl **id) |
bc5fc7e4 | 1161 | { |
d29ec824 CH |
1162 | struct nvme_command c = { |
1163 | .identify.opcode = nvme_admin_identify, | |
1164 | .identify.cns = cpu_to_le32(1), | |
1165 | }; | |
1166 | int error; | |
bc5fc7e4 | 1167 | |
d29ec824 CH |
1168 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); |
1169 | if (!*id) | |
1170 | return -ENOMEM; | |
1171 | ||
1172 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
1173 | sizeof(struct nvme_id_ctrl)); | |
1174 | if (error) | |
1175 | kfree(*id); | |
1176 | return error; | |
1177 | } | |
1178 | ||
1179 | int nvme_identify_ns(struct nvme_dev *dev, unsigned nsid, | |
1180 | struct nvme_id_ns **id) | |
1181 | { | |
1182 | struct nvme_command c = { | |
1183 | .identify.opcode = nvme_admin_identify, | |
1184 | .identify.nsid = cpu_to_le32(nsid), | |
1185 | }; | |
1186 | int error; | |
bc5fc7e4 | 1187 | |
d29ec824 CH |
1188 | *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL); |
1189 | if (!*id) | |
1190 | return -ENOMEM; | |
1191 | ||
1192 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
1193 | sizeof(struct nvme_id_ns)); | |
1194 | if (error) | |
1195 | kfree(*id); | |
1196 | return error; | |
bc5fc7e4 MW |
1197 | } |
1198 | ||
5d0f6131 | 1199 | int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid, |
08df1e05 | 1200 | dma_addr_t dma_addr, u32 *result) |
bc5fc7e4 MW |
1201 | { |
1202 | struct nvme_command c; | |
1203 | ||
1204 | memset(&c, 0, sizeof(c)); | |
1205 | c.features.opcode = nvme_admin_get_features; | |
a42cecce | 1206 | c.features.nsid = cpu_to_le32(nsid); |
bc5fc7e4 MW |
1207 | c.features.prp1 = cpu_to_le64(dma_addr); |
1208 | c.features.fid = cpu_to_le32(fid); | |
bc5fc7e4 | 1209 | |
d29ec824 CH |
1210 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0, |
1211 | result, 0); | |
df348139 MW |
1212 | } |
1213 | ||
5d0f6131 VV |
1214 | int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11, |
1215 | dma_addr_t dma_addr, u32 *result) | |
df348139 MW |
1216 | { |
1217 | struct nvme_command c; | |
1218 | ||
1219 | memset(&c, 0, sizeof(c)); | |
1220 | c.features.opcode = nvme_admin_set_features; | |
1221 | c.features.prp1 = cpu_to_le64(dma_addr); | |
1222 | c.features.fid = cpu_to_le32(fid); | |
1223 | c.features.dword11 = cpu_to_le32(dword11); | |
1224 | ||
d29ec824 CH |
1225 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0, |
1226 | result, 0); | |
1227 | } | |
1228 | ||
1229 | int nvme_get_log_page(struct nvme_dev *dev, struct nvme_smart_log **log) | |
1230 | { | |
1231 | struct nvme_command c = { | |
1232 | .common.opcode = nvme_admin_get_log_page, | |
1233 | .common.nsid = cpu_to_le32(0xFFFFFFFF), | |
1234 | .common.cdw10[0] = cpu_to_le32( | |
1235 | (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) | | |
1236 | NVME_LOG_SMART), | |
1237 | }; | |
1238 | int error; | |
1239 | ||
1240 | *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL); | |
1241 | if (!*log) | |
1242 | return -ENOMEM; | |
1243 | ||
1244 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *log, | |
1245 | sizeof(struct nvme_smart_log)); | |
1246 | if (error) | |
1247 | kfree(*log); | |
1248 | return error; | |
bc5fc7e4 MW |
1249 | } |
1250 | ||
c30341dc | 1251 | /** |
a4aea562 | 1252 | * nvme_abort_req - Attempt aborting a request |
c30341dc KB |
1253 | * |
1254 | * Schedule controller reset if the command was already aborted once before and | |
1255 | * still hasn't been returned to the driver, or if this is the admin queue. | |
1256 | */ | |
a4aea562 | 1257 | static void nvme_abort_req(struct request *req) |
c30341dc | 1258 | { |
a4aea562 MB |
1259 | struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req); |
1260 | struct nvme_queue *nvmeq = cmd_rq->nvmeq; | |
c30341dc | 1261 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 MB |
1262 | struct request *abort_req; |
1263 | struct nvme_cmd_info *abort_cmd; | |
1264 | struct nvme_command cmd; | |
c30341dc | 1265 | |
a4aea562 | 1266 | if (!nvmeq->qid || cmd_rq->aborted) { |
7a509a6b KB |
1267 | unsigned long flags; |
1268 | ||
1269 | spin_lock_irqsave(&dev_list_lock, flags); | |
c30341dc | 1270 | if (work_busy(&dev->reset_work)) |
7a509a6b | 1271 | goto out; |
c30341dc | 1272 | list_del_init(&dev->node); |
e75ec752 | 1273 | dev_warn(dev->dev, "I/O %d QID %d timeout, reset controller\n", |
a4aea562 | 1274 | req->tag, nvmeq->qid); |
9ca97374 | 1275 | dev->reset_workfn = nvme_reset_failed_dev; |
c30341dc | 1276 | queue_work(nvme_workq, &dev->reset_work); |
7a509a6b KB |
1277 | out: |
1278 | spin_unlock_irqrestore(&dev_list_lock, flags); | |
c30341dc KB |
1279 | return; |
1280 | } | |
1281 | ||
1282 | if (!dev->abort_limit) | |
1283 | return; | |
1284 | ||
a4aea562 MB |
1285 | abort_req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, |
1286 | false); | |
9f173b33 | 1287 | if (IS_ERR(abort_req)) |
c30341dc KB |
1288 | return; |
1289 | ||
a4aea562 MB |
1290 | abort_cmd = blk_mq_rq_to_pdu(abort_req); |
1291 | nvme_set_info(abort_cmd, abort_req, abort_completion); | |
1292 | ||
c30341dc KB |
1293 | memset(&cmd, 0, sizeof(cmd)); |
1294 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
a4aea562 | 1295 | cmd.abort.cid = req->tag; |
c30341dc | 1296 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); |
a4aea562 | 1297 | cmd.abort.command_id = abort_req->tag; |
c30341dc KB |
1298 | |
1299 | --dev->abort_limit; | |
a4aea562 | 1300 | cmd_rq->aborted = 1; |
c30341dc | 1301 | |
a4aea562 | 1302 | dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag, |
c30341dc | 1303 | nvmeq->qid); |
a4aea562 MB |
1304 | if (nvme_submit_cmd(dev->queues[0], &cmd) < 0) { |
1305 | dev_warn(nvmeq->q_dmadev, | |
1306 | "Could not abort I/O %d QID %d", | |
1307 | req->tag, nvmeq->qid); | |
c87fd540 | 1308 | blk_mq_free_request(abort_req); |
a4aea562 | 1309 | } |
c30341dc KB |
1310 | } |
1311 | ||
a4aea562 MB |
1312 | static void nvme_cancel_queue_ios(struct blk_mq_hw_ctx *hctx, |
1313 | struct request *req, void *data, bool reserved) | |
a09115b2 | 1314 | { |
a4aea562 MB |
1315 | struct nvme_queue *nvmeq = data; |
1316 | void *ctx; | |
1317 | nvme_completion_fn fn; | |
1318 | struct nvme_cmd_info *cmd; | |
cef6a948 KB |
1319 | struct nvme_completion cqe; |
1320 | ||
1321 | if (!blk_mq_request_started(req)) | |
1322 | return; | |
a09115b2 | 1323 | |
a4aea562 | 1324 | cmd = blk_mq_rq_to_pdu(req); |
a09115b2 | 1325 | |
a4aea562 MB |
1326 | if (cmd->ctx == CMD_CTX_CANCELLED) |
1327 | return; | |
1328 | ||
cef6a948 KB |
1329 | if (blk_queue_dying(req->q)) |
1330 | cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1); | |
1331 | else | |
1332 | cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1); | |
1333 | ||
1334 | ||
a4aea562 MB |
1335 | dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n", |
1336 | req->tag, nvmeq->qid); | |
1337 | ctx = cancel_cmd_info(cmd, &fn); | |
1338 | fn(nvmeq, ctx, &cqe); | |
a09115b2 MW |
1339 | } |
1340 | ||
a4aea562 | 1341 | static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) |
9e866774 | 1342 | { |
a4aea562 MB |
1343 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); |
1344 | struct nvme_queue *nvmeq = cmd->nvmeq; | |
1345 | ||
1346 | dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag, | |
1347 | nvmeq->qid); | |
7a509a6b | 1348 | spin_lock_irq(&nvmeq->q_lock); |
07836e65 | 1349 | nvme_abort_req(req); |
7a509a6b | 1350 | spin_unlock_irq(&nvmeq->q_lock); |
a4aea562 | 1351 | |
07836e65 KB |
1352 | /* |
1353 | * The aborted req will be completed on receiving the abort req. | |
1354 | * We enable the timer again. If hit twice, it'll cause a device reset, | |
1355 | * as the device then is in a faulty state. | |
1356 | */ | |
1357 | return BLK_EH_RESET_TIMER; | |
a4aea562 | 1358 | } |
22404274 | 1359 | |
a4aea562 MB |
1360 | static void nvme_free_queue(struct nvme_queue *nvmeq) |
1361 | { | |
9e866774 MW |
1362 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), |
1363 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
1364 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
1365 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); | |
1366 | kfree(nvmeq); | |
1367 | } | |
1368 | ||
a1a5ef99 | 1369 | static void nvme_free_queues(struct nvme_dev *dev, int lowest) |
22404274 KB |
1370 | { |
1371 | int i; | |
1372 | ||
a1a5ef99 | 1373 | for (i = dev->queue_count - 1; i >= lowest; i--) { |
a4aea562 | 1374 | struct nvme_queue *nvmeq = dev->queues[i]; |
22404274 | 1375 | dev->queue_count--; |
a4aea562 | 1376 | dev->queues[i] = NULL; |
f435c282 | 1377 | nvme_free_queue(nvmeq); |
121c7ad4 | 1378 | } |
22404274 KB |
1379 | } |
1380 | ||
4d115420 KB |
1381 | /** |
1382 | * nvme_suspend_queue - put queue into suspended state | |
1383 | * @nvmeq - queue to suspend | |
4d115420 KB |
1384 | */ |
1385 | static int nvme_suspend_queue(struct nvme_queue *nvmeq) | |
b60503ba | 1386 | { |
2b25d981 | 1387 | int vector; |
b60503ba | 1388 | |
a09115b2 | 1389 | spin_lock_irq(&nvmeq->q_lock); |
2b25d981 KB |
1390 | if (nvmeq->cq_vector == -1) { |
1391 | spin_unlock_irq(&nvmeq->q_lock); | |
1392 | return 1; | |
1393 | } | |
1394 | vector = nvmeq->dev->entry[nvmeq->cq_vector].vector; | |
42f61420 | 1395 | nvmeq->dev->online_queues--; |
2b25d981 | 1396 | nvmeq->cq_vector = -1; |
a09115b2 MW |
1397 | spin_unlock_irq(&nvmeq->q_lock); |
1398 | ||
6df3dbc8 KB |
1399 | if (!nvmeq->qid && nvmeq->dev->admin_q) |
1400 | blk_mq_freeze_queue_start(nvmeq->dev->admin_q); | |
1401 | ||
aba2080f MW |
1402 | irq_set_affinity_hint(vector, NULL); |
1403 | free_irq(vector, nvmeq); | |
b60503ba | 1404 | |
4d115420 KB |
1405 | return 0; |
1406 | } | |
b60503ba | 1407 | |
4d115420 KB |
1408 | static void nvme_clear_queue(struct nvme_queue *nvmeq) |
1409 | { | |
a4aea562 MB |
1410 | struct blk_mq_hw_ctx *hctx = nvmeq->hctx; |
1411 | ||
22404274 | 1412 | spin_lock_irq(&nvmeq->q_lock); |
a4aea562 MB |
1413 | if (hctx && hctx->tags) |
1414 | blk_mq_tag_busy_iter(hctx, nvme_cancel_queue_ios, nvmeq); | |
22404274 | 1415 | spin_unlock_irq(&nvmeq->q_lock); |
b60503ba MW |
1416 | } |
1417 | ||
4d115420 KB |
1418 | static void nvme_disable_queue(struct nvme_dev *dev, int qid) |
1419 | { | |
a4aea562 | 1420 | struct nvme_queue *nvmeq = dev->queues[qid]; |
4d115420 KB |
1421 | |
1422 | if (!nvmeq) | |
1423 | return; | |
1424 | if (nvme_suspend_queue(nvmeq)) | |
1425 | return; | |
1426 | ||
0e53d180 KB |
1427 | /* Don't tell the adapter to delete the admin queue. |
1428 | * Don't tell a removed adapter to delete IO queues. */ | |
1429 | if (qid && readl(&dev->bar->csts) != -1) { | |
b60503ba MW |
1430 | adapter_delete_sq(dev, qid); |
1431 | adapter_delete_cq(dev, qid); | |
1432 | } | |
07836e65 KB |
1433 | |
1434 | spin_lock_irq(&nvmeq->q_lock); | |
1435 | nvme_process_cq(nvmeq); | |
1436 | spin_unlock_irq(&nvmeq->q_lock); | |
b60503ba MW |
1437 | } |
1438 | ||
1439 | static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, | |
2b25d981 | 1440 | int depth) |
b60503ba | 1441 | { |
a4aea562 | 1442 | struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL); |
b60503ba MW |
1443 | if (!nvmeq) |
1444 | return NULL; | |
1445 | ||
e75ec752 | 1446 | nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth), |
4d51abf9 | 1447 | &nvmeq->cq_dma_addr, GFP_KERNEL); |
b60503ba MW |
1448 | if (!nvmeq->cqes) |
1449 | goto free_nvmeq; | |
b60503ba | 1450 | |
e75ec752 | 1451 | nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth), |
b60503ba MW |
1452 | &nvmeq->sq_dma_addr, GFP_KERNEL); |
1453 | if (!nvmeq->sq_cmds) | |
1454 | goto free_cqdma; | |
1455 | ||
e75ec752 | 1456 | nvmeq->q_dmadev = dev->dev; |
091b6092 | 1457 | nvmeq->dev = dev; |
3193f07b MW |
1458 | snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d", |
1459 | dev->instance, qid); | |
b60503ba MW |
1460 | spin_lock_init(&nvmeq->q_lock); |
1461 | nvmeq->cq_head = 0; | |
82123460 | 1462 | nvmeq->cq_phase = 1; |
b80d5ccc | 1463 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
b60503ba | 1464 | nvmeq->q_depth = depth; |
c30341dc | 1465 | nvmeq->qid = qid; |
22404274 | 1466 | dev->queue_count++; |
a4aea562 | 1467 | dev->queues[qid] = nvmeq; |
b60503ba MW |
1468 | |
1469 | return nvmeq; | |
1470 | ||
1471 | free_cqdma: | |
e75ec752 | 1472 | dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes, |
b60503ba MW |
1473 | nvmeq->cq_dma_addr); |
1474 | free_nvmeq: | |
1475 | kfree(nvmeq); | |
1476 | return NULL; | |
1477 | } | |
1478 | ||
3001082c MW |
1479 | static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq, |
1480 | const char *name) | |
1481 | { | |
58ffacb5 MW |
1482 | if (use_threaded_interrupts) |
1483 | return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector, | |
481e5bad | 1484 | nvme_irq_check, nvme_irq, IRQF_SHARED, |
58ffacb5 | 1485 | name, nvmeq); |
3001082c | 1486 | return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq, |
481e5bad | 1487 | IRQF_SHARED, name, nvmeq); |
3001082c MW |
1488 | } |
1489 | ||
22404274 | 1490 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1491 | { |
22404274 | 1492 | struct nvme_dev *dev = nvmeq->dev; |
b60503ba | 1493 | |
7be50e93 | 1494 | spin_lock_irq(&nvmeq->q_lock); |
22404274 KB |
1495 | nvmeq->sq_tail = 0; |
1496 | nvmeq->cq_head = 0; | |
1497 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1498 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
22404274 | 1499 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); |
42f61420 | 1500 | dev->online_queues++; |
7be50e93 | 1501 | spin_unlock_irq(&nvmeq->q_lock); |
22404274 KB |
1502 | } |
1503 | ||
1504 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) | |
1505 | { | |
1506 | struct nvme_dev *dev = nvmeq->dev; | |
1507 | int result; | |
3f85d50b | 1508 | |
2b25d981 | 1509 | nvmeq->cq_vector = qid - 1; |
b60503ba MW |
1510 | result = adapter_alloc_cq(dev, qid, nvmeq); |
1511 | if (result < 0) | |
22404274 | 1512 | return result; |
b60503ba MW |
1513 | |
1514 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1515 | if (result < 0) | |
1516 | goto release_cq; | |
1517 | ||
3193f07b | 1518 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
b60503ba MW |
1519 | if (result < 0) |
1520 | goto release_sq; | |
1521 | ||
22404274 | 1522 | nvme_init_queue(nvmeq, qid); |
22404274 | 1523 | return result; |
b60503ba MW |
1524 | |
1525 | release_sq: | |
1526 | adapter_delete_sq(dev, qid); | |
1527 | release_cq: | |
1528 | adapter_delete_cq(dev, qid); | |
22404274 | 1529 | return result; |
b60503ba MW |
1530 | } |
1531 | ||
ba47e386 MW |
1532 | static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled) |
1533 | { | |
1534 | unsigned long timeout; | |
1535 | u32 bit = enabled ? NVME_CSTS_RDY : 0; | |
1536 | ||
1537 | timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
1538 | ||
1539 | while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) { | |
1540 | msleep(100); | |
1541 | if (fatal_signal_pending(current)) | |
1542 | return -EINTR; | |
1543 | if (time_after(jiffies, timeout)) { | |
e75ec752 | 1544 | dev_err(dev->dev, |
27e8166c MW |
1545 | "Device not ready; aborting %s\n", enabled ? |
1546 | "initialisation" : "reset"); | |
ba47e386 MW |
1547 | return -ENODEV; |
1548 | } | |
1549 | } | |
1550 | ||
1551 | return 0; | |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * If the device has been passed off to us in an enabled state, just clear | |
1556 | * the enabled bit. The spec says we should set the 'shutdown notification | |
1557 | * bits', but doing so may cause the device to complete commands to the | |
1558 | * admin queue ... and we don't know what memory that might be pointing at! | |
1559 | */ | |
1560 | static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap) | |
1561 | { | |
01079522 DM |
1562 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1563 | dev->ctrl_config &= ~NVME_CC_ENABLE; | |
1564 | writel(dev->ctrl_config, &dev->bar->cc); | |
44af146a | 1565 | |
ba47e386 MW |
1566 | return nvme_wait_ready(dev, cap, false); |
1567 | } | |
1568 | ||
1569 | static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap) | |
1570 | { | |
01079522 DM |
1571 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1572 | dev->ctrl_config |= NVME_CC_ENABLE; | |
1573 | writel(dev->ctrl_config, &dev->bar->cc); | |
1574 | ||
ba47e386 MW |
1575 | return nvme_wait_ready(dev, cap, true); |
1576 | } | |
1577 | ||
1894d8f1 KB |
1578 | static int nvme_shutdown_ctrl(struct nvme_dev *dev) |
1579 | { | |
1580 | unsigned long timeout; | |
1894d8f1 | 1581 | |
01079522 DM |
1582 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1583 | dev->ctrl_config |= NVME_CC_SHN_NORMAL; | |
1584 | ||
1585 | writel(dev->ctrl_config, &dev->bar->cc); | |
1894d8f1 | 1586 | |
2484f407 | 1587 | timeout = SHUTDOWN_TIMEOUT + jiffies; |
1894d8f1 KB |
1588 | while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) != |
1589 | NVME_CSTS_SHST_CMPLT) { | |
1590 | msleep(100); | |
1591 | if (fatal_signal_pending(current)) | |
1592 | return -EINTR; | |
1593 | if (time_after(jiffies, timeout)) { | |
e75ec752 | 1594 | dev_err(dev->dev, |
1894d8f1 KB |
1595 | "Device shutdown incomplete; abort shutdown\n"); |
1596 | return -ENODEV; | |
1597 | } | |
1598 | } | |
1599 | ||
1600 | return 0; | |
1601 | } | |
1602 | ||
a4aea562 | 1603 | static struct blk_mq_ops nvme_mq_admin_ops = { |
d29ec824 | 1604 | .queue_rq = nvme_queue_rq, |
a4aea562 MB |
1605 | .map_queue = blk_mq_map_queue, |
1606 | .init_hctx = nvme_admin_init_hctx, | |
2c30540b | 1607 | .exit_hctx = nvme_exit_hctx, |
a4aea562 MB |
1608 | .init_request = nvme_admin_init_request, |
1609 | .timeout = nvme_timeout, | |
1610 | }; | |
1611 | ||
1612 | static struct blk_mq_ops nvme_mq_ops = { | |
1613 | .queue_rq = nvme_queue_rq, | |
1614 | .map_queue = blk_mq_map_queue, | |
1615 | .init_hctx = nvme_init_hctx, | |
2c30540b | 1616 | .exit_hctx = nvme_exit_hctx, |
a4aea562 MB |
1617 | .init_request = nvme_init_request, |
1618 | .timeout = nvme_timeout, | |
1619 | }; | |
1620 | ||
ea191d2f KB |
1621 | static void nvme_dev_remove_admin(struct nvme_dev *dev) |
1622 | { | |
1623 | if (dev->admin_q && !blk_queue_dying(dev->admin_q)) { | |
1624 | blk_cleanup_queue(dev->admin_q); | |
1625 | blk_mq_free_tag_set(&dev->admin_tagset); | |
1626 | } | |
1627 | } | |
1628 | ||
a4aea562 MB |
1629 | static int nvme_alloc_admin_tags(struct nvme_dev *dev) |
1630 | { | |
1631 | if (!dev->admin_q) { | |
1632 | dev->admin_tagset.ops = &nvme_mq_admin_ops; | |
1633 | dev->admin_tagset.nr_hw_queues = 1; | |
1634 | dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1; | |
1efccc9d | 1635 | dev->admin_tagset.reserved_tags = 1; |
a4aea562 | 1636 | dev->admin_tagset.timeout = ADMIN_TIMEOUT; |
e75ec752 | 1637 | dev->admin_tagset.numa_node = dev_to_node(dev->dev); |
ac3dd5bd | 1638 | dev->admin_tagset.cmd_size = nvme_cmd_size(dev); |
a4aea562 MB |
1639 | dev->admin_tagset.driver_data = dev; |
1640 | ||
1641 | if (blk_mq_alloc_tag_set(&dev->admin_tagset)) | |
1642 | return -ENOMEM; | |
1643 | ||
1644 | dev->admin_q = blk_mq_init_queue(&dev->admin_tagset); | |
35b489d3 | 1645 | if (IS_ERR(dev->admin_q)) { |
a4aea562 MB |
1646 | blk_mq_free_tag_set(&dev->admin_tagset); |
1647 | return -ENOMEM; | |
1648 | } | |
ea191d2f KB |
1649 | if (!blk_get_queue(dev->admin_q)) { |
1650 | nvme_dev_remove_admin(dev); | |
1651 | return -ENODEV; | |
1652 | } | |
0fb59cbc KB |
1653 | } else |
1654 | blk_mq_unfreeze_queue(dev->admin_q); | |
a4aea562 MB |
1655 | |
1656 | return 0; | |
1657 | } | |
1658 | ||
8d85fce7 | 1659 | static int nvme_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1660 | { |
ba47e386 | 1661 | int result; |
b60503ba | 1662 | u32 aqa; |
ba47e386 | 1663 | u64 cap = readq(&dev->bar->cap); |
b60503ba | 1664 | struct nvme_queue *nvmeq; |
1d090624 KB |
1665 | unsigned page_shift = PAGE_SHIFT; |
1666 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12; | |
1667 | unsigned dev_page_max = NVME_CAP_MPSMAX(cap) + 12; | |
1668 | ||
1669 | if (page_shift < dev_page_min) { | |
e75ec752 | 1670 | dev_err(dev->dev, |
1d090624 KB |
1671 | "Minimum device page size (%u) too large for " |
1672 | "host (%u)\n", 1 << dev_page_min, | |
1673 | 1 << page_shift); | |
1674 | return -ENODEV; | |
1675 | } | |
1676 | if (page_shift > dev_page_max) { | |
e75ec752 | 1677 | dev_info(dev->dev, |
1d090624 KB |
1678 | "Device maximum page size (%u) smaller than " |
1679 | "host (%u); enabling work-around\n", | |
1680 | 1 << dev_page_max, 1 << page_shift); | |
1681 | page_shift = dev_page_max; | |
1682 | } | |
b60503ba | 1683 | |
ba47e386 MW |
1684 | result = nvme_disable_ctrl(dev, cap); |
1685 | if (result < 0) | |
1686 | return result; | |
b60503ba | 1687 | |
a4aea562 | 1688 | nvmeq = dev->queues[0]; |
cd638946 | 1689 | if (!nvmeq) { |
2b25d981 | 1690 | nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); |
cd638946 KB |
1691 | if (!nvmeq) |
1692 | return -ENOMEM; | |
cd638946 | 1693 | } |
b60503ba MW |
1694 | |
1695 | aqa = nvmeq->q_depth - 1; | |
1696 | aqa |= aqa << 16; | |
1697 | ||
1d090624 KB |
1698 | dev->page_size = 1 << page_shift; |
1699 | ||
01079522 | 1700 | dev->ctrl_config = NVME_CC_CSS_NVM; |
1d090624 | 1701 | dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; |
b60503ba | 1702 | dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; |
7f53f9d2 | 1703 | dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
b60503ba MW |
1704 | |
1705 | writel(aqa, &dev->bar->aqa); | |
1706 | writeq(nvmeq->sq_dma_addr, &dev->bar->asq); | |
1707 | writeq(nvmeq->cq_dma_addr, &dev->bar->acq); | |
b60503ba | 1708 | |
ba47e386 | 1709 | result = nvme_enable_ctrl(dev, cap); |
025c557a | 1710 | if (result) |
a4aea562 MB |
1711 | goto free_nvmeq; |
1712 | ||
2b25d981 | 1713 | nvmeq->cq_vector = 0; |
3193f07b | 1714 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
025c557a | 1715 | if (result) |
0fb59cbc | 1716 | goto free_nvmeq; |
025c557a | 1717 | |
b60503ba | 1718 | return result; |
a4aea562 | 1719 | |
a4aea562 MB |
1720 | free_nvmeq: |
1721 | nvme_free_queues(dev, 0); | |
1722 | return result; | |
b60503ba MW |
1723 | } |
1724 | ||
a53295b6 MW |
1725 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
1726 | { | |
1727 | struct nvme_dev *dev = ns->dev; | |
a53295b6 MW |
1728 | struct nvme_user_io io; |
1729 | struct nvme_command c; | |
d29ec824 | 1730 | unsigned length, meta_len; |
a67a9513 | 1731 | int status, write; |
a67a9513 KB |
1732 | dma_addr_t meta_dma = 0; |
1733 | void *meta = NULL; | |
a53295b6 MW |
1734 | |
1735 | if (copy_from_user(&io, uio, sizeof(io))) | |
1736 | return -EFAULT; | |
6c7d4945 MW |
1737 | |
1738 | switch (io.opcode) { | |
1739 | case nvme_cmd_write: | |
1740 | case nvme_cmd_read: | |
6bbf1acd | 1741 | case nvme_cmd_compare: |
6413214c | 1742 | break; |
6c7d4945 | 1743 | default: |
6bbf1acd | 1744 | return -EINVAL; |
6c7d4945 MW |
1745 | } |
1746 | ||
d29ec824 CH |
1747 | length = (io.nblocks + 1) << ns->lba_shift; |
1748 | meta_len = (io.nblocks + 1) * ns->ms; | |
1749 | write = io.opcode & 1; | |
a53295b6 | 1750 | |
a67a9513 | 1751 | if (meta_len) { |
d29ec824 CH |
1752 | if (((io.metadata & 3) || !io.metadata) && !ns->ext) |
1753 | return -EINVAL; | |
1754 | ||
1755 | if (ns->ext) { | |
1756 | length += meta_len; | |
1757 | meta_len = 0; | |
1758 | } | |
1759 | ||
e75ec752 | 1760 | meta = dma_alloc_coherent(dev->dev, meta_len, |
a67a9513 KB |
1761 | &meta_dma, GFP_KERNEL); |
1762 | if (!meta) { | |
1763 | status = -ENOMEM; | |
1764 | goto unmap; | |
1765 | } | |
1766 | if (write) { | |
1767 | if (copy_from_user(meta, (void __user *)io.metadata, | |
1768 | meta_len)) { | |
1769 | status = -EFAULT; | |
1770 | goto unmap; | |
1771 | } | |
1772 | } | |
1773 | } | |
1774 | ||
a53295b6 MW |
1775 | memset(&c, 0, sizeof(c)); |
1776 | c.rw.opcode = io.opcode; | |
1777 | c.rw.flags = io.flags; | |
6c7d4945 | 1778 | c.rw.nsid = cpu_to_le32(ns->ns_id); |
a53295b6 | 1779 | c.rw.slba = cpu_to_le64(io.slba); |
6c7d4945 | 1780 | c.rw.length = cpu_to_le16(io.nblocks); |
a53295b6 | 1781 | c.rw.control = cpu_to_le16(io.control); |
1c9b5265 MW |
1782 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); |
1783 | c.rw.reftag = cpu_to_le32(io.reftag); | |
1784 | c.rw.apptag = cpu_to_le16(io.apptag); | |
1785 | c.rw.appmask = cpu_to_le16(io.appmask); | |
a67a9513 | 1786 | c.rw.metadata = cpu_to_le64(meta_dma); |
d29ec824 CH |
1787 | |
1788 | status = __nvme_submit_sync_cmd(ns->queue, &c, NULL, | |
1789 | (void __user *)io.addr, length, NULL, 0); | |
f410c680 | 1790 | unmap: |
a67a9513 KB |
1791 | if (meta) { |
1792 | if (status == NVME_SC_SUCCESS && !write) { | |
1793 | if (copy_to_user((void __user *)io.metadata, meta, | |
1794 | meta_len)) | |
1795 | status = -EFAULT; | |
1796 | } | |
e75ec752 | 1797 | dma_free_coherent(dev->dev, meta_len, meta, meta_dma); |
f410c680 | 1798 | } |
a53295b6 MW |
1799 | return status; |
1800 | } | |
1801 | ||
a4aea562 MB |
1802 | static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns, |
1803 | struct nvme_passthru_cmd __user *ucmd) | |
6ee44cdc | 1804 | { |
7963e521 | 1805 | struct nvme_passthru_cmd cmd; |
6ee44cdc | 1806 | struct nvme_command c; |
d29ec824 CH |
1807 | unsigned timeout = 0; |
1808 | int status; | |
6ee44cdc | 1809 | |
6bbf1acd MW |
1810 | if (!capable(CAP_SYS_ADMIN)) |
1811 | return -EACCES; | |
1812 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
6ee44cdc | 1813 | return -EFAULT; |
6ee44cdc MW |
1814 | |
1815 | memset(&c, 0, sizeof(c)); | |
6bbf1acd MW |
1816 | c.common.opcode = cmd.opcode; |
1817 | c.common.flags = cmd.flags; | |
1818 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1819 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1820 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1821 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
1822 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
1823 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
1824 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
1825 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
1826 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
1827 | ||
d29ec824 CH |
1828 | if (cmd.timeout_ms) |
1829 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
f705f837 CH |
1830 | |
1831 | status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c, | |
d29ec824 CH |
1832 | NULL, (void __user *)cmd.addr, cmd.data_len, |
1833 | &cmd.result, timeout); | |
1834 | if (status >= 0) { | |
1835 | if (put_user(cmd.result, &ucmd->result)) | |
1836 | return -EFAULT; | |
6bbf1acd | 1837 | } |
f4f117f6 | 1838 | |
6ee44cdc MW |
1839 | return status; |
1840 | } | |
1841 | ||
b60503ba MW |
1842 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, |
1843 | unsigned long arg) | |
1844 | { | |
1845 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1846 | ||
1847 | switch (cmd) { | |
6bbf1acd | 1848 | case NVME_IOCTL_ID: |
c3bfe717 | 1849 | force_successful_syscall_return(); |
6bbf1acd MW |
1850 | return ns->ns_id; |
1851 | case NVME_IOCTL_ADMIN_CMD: | |
a4aea562 | 1852 | return nvme_user_cmd(ns->dev, NULL, (void __user *)arg); |
7963e521 | 1853 | case NVME_IOCTL_IO_CMD: |
a4aea562 | 1854 | return nvme_user_cmd(ns->dev, ns, (void __user *)arg); |
a53295b6 MW |
1855 | case NVME_IOCTL_SUBMIT_IO: |
1856 | return nvme_submit_io(ns, (void __user *)arg); | |
5d0f6131 VV |
1857 | case SG_GET_VERSION_NUM: |
1858 | return nvme_sg_get_version_num((void __user *)arg); | |
1859 | case SG_IO: | |
1860 | return nvme_sg_io(ns, (void __user *)arg); | |
b60503ba MW |
1861 | default: |
1862 | return -ENOTTY; | |
1863 | } | |
1864 | } | |
1865 | ||
320a3827 KB |
1866 | #ifdef CONFIG_COMPAT |
1867 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1868 | unsigned int cmd, unsigned long arg) | |
1869 | { | |
320a3827 KB |
1870 | switch (cmd) { |
1871 | case SG_IO: | |
e179729a | 1872 | return -ENOIOCTLCMD; |
320a3827 KB |
1873 | } |
1874 | return nvme_ioctl(bdev, mode, cmd, arg); | |
1875 | } | |
1876 | #else | |
1877 | #define nvme_compat_ioctl NULL | |
1878 | #endif | |
1879 | ||
9ac27090 KB |
1880 | static int nvme_open(struct block_device *bdev, fmode_t mode) |
1881 | { | |
9e60352c KB |
1882 | int ret = 0; |
1883 | struct nvme_ns *ns; | |
9ac27090 | 1884 | |
9e60352c KB |
1885 | spin_lock(&dev_list_lock); |
1886 | ns = bdev->bd_disk->private_data; | |
1887 | if (!ns) | |
1888 | ret = -ENXIO; | |
1889 | else if (!kref_get_unless_zero(&ns->dev->kref)) | |
1890 | ret = -ENXIO; | |
1891 | spin_unlock(&dev_list_lock); | |
1892 | ||
1893 | return ret; | |
9ac27090 KB |
1894 | } |
1895 | ||
1896 | static void nvme_free_dev(struct kref *kref); | |
1897 | ||
1898 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
1899 | { | |
1900 | struct nvme_ns *ns = disk->private_data; | |
1901 | struct nvme_dev *dev = ns->dev; | |
1902 | ||
1903 | kref_put(&dev->kref, nvme_free_dev); | |
1904 | } | |
1905 | ||
4cc09e2d KB |
1906 | static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo) |
1907 | { | |
1908 | /* some standard values */ | |
1909 | geo->heads = 1 << 6; | |
1910 | geo->sectors = 1 << 5; | |
1911 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
1912 | return 0; | |
1913 | } | |
1914 | ||
e1e5e564 KB |
1915 | static void nvme_config_discard(struct nvme_ns *ns) |
1916 | { | |
1917 | u32 logical_block_size = queue_logical_block_size(ns->queue); | |
1918 | ns->queue->limits.discard_zeroes_data = 0; | |
1919 | ns->queue->limits.discard_alignment = logical_block_size; | |
1920 | ns->queue->limits.discard_granularity = logical_block_size; | |
1921 | ns->queue->limits.max_discard_sectors = 0xffffffff; | |
1922 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); | |
1923 | } | |
1924 | ||
1b9dbf7f KB |
1925 | static int nvme_revalidate_disk(struct gendisk *disk) |
1926 | { | |
1927 | struct nvme_ns *ns = disk->private_data; | |
1928 | struct nvme_dev *dev = ns->dev; | |
1929 | struct nvme_id_ns *id; | |
a67a9513 KB |
1930 | u8 lbaf, pi_type; |
1931 | u16 old_ms; | |
e1e5e564 | 1932 | unsigned short bs; |
1b9dbf7f | 1933 | |
d29ec824 CH |
1934 | if (nvme_identify_ns(dev, ns->ns_id, &id)) { |
1935 | dev_warn(dev->dev, "%s: Identify failure\n", __func__); | |
1b9dbf7f KB |
1936 | return 0; |
1937 | } | |
1938 | ||
e1e5e564 KB |
1939 | old_ms = ns->ms; |
1940 | lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK; | |
1b9dbf7f | 1941 | ns->lba_shift = id->lbaf[lbaf].ds; |
e1e5e564 | 1942 | ns->ms = le16_to_cpu(id->lbaf[lbaf].ms); |
a67a9513 | 1943 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); |
e1e5e564 KB |
1944 | |
1945 | /* | |
1946 | * If identify namespace failed, use default 512 byte block size so | |
1947 | * block layer can use before failing read/write for 0 capacity. | |
1948 | */ | |
1949 | if (ns->lba_shift == 0) | |
1950 | ns->lba_shift = 9; | |
1951 | bs = 1 << ns->lba_shift; | |
1952 | ||
1953 | /* XXX: PI implementation requires metadata equal t10 pi tuple size */ | |
1954 | pi_type = ns->ms == sizeof(struct t10_pi_tuple) ? | |
1955 | id->dps & NVME_NS_DPS_PI_MASK : 0; | |
1956 | ||
52b68d7e KB |
1957 | if (blk_get_integrity(disk) && (ns->pi_type != pi_type || |
1958 | ns->ms != old_ms || | |
e1e5e564 | 1959 | bs != queue_logical_block_size(disk->queue) || |
a67a9513 | 1960 | (ns->ms && ns->ext))) |
e1e5e564 KB |
1961 | blk_integrity_unregister(disk); |
1962 | ||
1963 | ns->pi_type = pi_type; | |
1964 | blk_queue_logical_block_size(ns->queue, bs); | |
1965 | ||
52b68d7e | 1966 | if (ns->ms && !blk_get_integrity(disk) && (disk->flags & GENHD_FL_UP) && |
a67a9513 | 1967 | !ns->ext) |
e1e5e564 KB |
1968 | nvme_init_integrity(ns); |
1969 | ||
52b68d7e | 1970 | if (id->ncap == 0 || (ns->ms && !blk_get_integrity(disk))) |
e1e5e564 KB |
1971 | set_capacity(disk, 0); |
1972 | else | |
1973 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); | |
1974 | ||
1975 | if (dev->oncs & NVME_CTRL_ONCS_DSM) | |
1976 | nvme_config_discard(ns); | |
1b9dbf7f | 1977 | |
d29ec824 | 1978 | kfree(id); |
1b9dbf7f KB |
1979 | return 0; |
1980 | } | |
1981 | ||
b60503ba MW |
1982 | static const struct block_device_operations nvme_fops = { |
1983 | .owner = THIS_MODULE, | |
1984 | .ioctl = nvme_ioctl, | |
320a3827 | 1985 | .compat_ioctl = nvme_compat_ioctl, |
9ac27090 KB |
1986 | .open = nvme_open, |
1987 | .release = nvme_release, | |
4cc09e2d | 1988 | .getgeo = nvme_getgeo, |
1b9dbf7f | 1989 | .revalidate_disk= nvme_revalidate_disk, |
b60503ba MW |
1990 | }; |
1991 | ||
1fa6aead MW |
1992 | static int nvme_kthread(void *data) |
1993 | { | |
d4b4ff8e | 1994 | struct nvme_dev *dev, *next; |
1fa6aead MW |
1995 | |
1996 | while (!kthread_should_stop()) { | |
564a232c | 1997 | set_current_state(TASK_INTERRUPTIBLE); |
1fa6aead | 1998 | spin_lock(&dev_list_lock); |
d4b4ff8e | 1999 | list_for_each_entry_safe(dev, next, &dev_list, node) { |
1fa6aead | 2000 | int i; |
07836e65 | 2001 | if (readl(&dev->bar->csts) & NVME_CSTS_CFS) { |
d4b4ff8e KB |
2002 | if (work_busy(&dev->reset_work)) |
2003 | continue; | |
2004 | list_del_init(&dev->node); | |
e75ec752 | 2005 | dev_warn(dev->dev, |
a4aea562 MB |
2006 | "Failed status: %x, reset controller\n", |
2007 | readl(&dev->bar->csts)); | |
9ca97374 | 2008 | dev->reset_workfn = nvme_reset_failed_dev; |
d4b4ff8e KB |
2009 | queue_work(nvme_workq, &dev->reset_work); |
2010 | continue; | |
2011 | } | |
1fa6aead | 2012 | for (i = 0; i < dev->queue_count; i++) { |
a4aea562 | 2013 | struct nvme_queue *nvmeq = dev->queues[i]; |
740216fc MW |
2014 | if (!nvmeq) |
2015 | continue; | |
1fa6aead | 2016 | spin_lock_irq(&nvmeq->q_lock); |
bc57a0f7 | 2017 | nvme_process_cq(nvmeq); |
6fccf938 KB |
2018 | |
2019 | while ((i == 0) && (dev->event_limit > 0)) { | |
a4aea562 | 2020 | if (nvme_submit_async_admin_req(dev)) |
6fccf938 KB |
2021 | break; |
2022 | dev->event_limit--; | |
2023 | } | |
1fa6aead MW |
2024 | spin_unlock_irq(&nvmeq->q_lock); |
2025 | } | |
2026 | } | |
2027 | spin_unlock(&dev_list_lock); | |
acb7aa0d | 2028 | schedule_timeout(round_jiffies_relative(HZ)); |
1fa6aead MW |
2029 | } |
2030 | return 0; | |
2031 | } | |
2032 | ||
e1e5e564 | 2033 | static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid) |
b60503ba MW |
2034 | { |
2035 | struct nvme_ns *ns; | |
2036 | struct gendisk *disk; | |
e75ec752 | 2037 | int node = dev_to_node(dev->dev); |
b60503ba | 2038 | |
a4aea562 | 2039 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); |
b60503ba | 2040 | if (!ns) |
e1e5e564 KB |
2041 | return; |
2042 | ||
a4aea562 | 2043 | ns->queue = blk_mq_init_queue(&dev->tagset); |
9f173b33 | 2044 | if (IS_ERR(ns->queue)) |
b60503ba | 2045 | goto out_free_ns; |
4eeb9215 MW |
2046 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue); |
2047 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); | |
a4aea562 | 2048 | queue_flag_set_unlocked(QUEUE_FLAG_SG_GAPS, ns->queue); |
b60503ba MW |
2049 | ns->dev = dev; |
2050 | ns->queue->queuedata = ns; | |
2051 | ||
a4aea562 | 2052 | disk = alloc_disk_node(0, node); |
b60503ba MW |
2053 | if (!disk) |
2054 | goto out_free_queue; | |
a4aea562 | 2055 | |
5aff9382 | 2056 | ns->ns_id = nsid; |
b60503ba | 2057 | ns->disk = disk; |
e1e5e564 KB |
2058 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ |
2059 | list_add_tail(&ns->list, &dev->namespaces); | |
2060 | ||
e9ef4636 | 2061 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); |
8fc23e03 KB |
2062 | if (dev->max_hw_sectors) |
2063 | blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors); | |
a4aea562 MB |
2064 | if (dev->stripe_size) |
2065 | blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9); | |
a7d2ce28 KB |
2066 | if (dev->vwc & NVME_CTRL_VWC_PRESENT) |
2067 | blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA); | |
b60503ba MW |
2068 | |
2069 | disk->major = nvme_major; | |
469071a3 | 2070 | disk->first_minor = 0; |
b60503ba MW |
2071 | disk->fops = &nvme_fops; |
2072 | disk->private_data = ns; | |
2073 | disk->queue = ns->queue; | |
b3fffdef | 2074 | disk->driverfs_dev = dev->device; |
469071a3 | 2075 | disk->flags = GENHD_FL_EXT_DEVT; |
5aff9382 | 2076 | sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid); |
b60503ba | 2077 | |
e1e5e564 KB |
2078 | /* |
2079 | * Initialize capacity to 0 until we establish the namespace format and | |
2080 | * setup integrity extentions if necessary. The revalidate_disk after | |
2081 | * add_disk allows the driver to register with integrity if the format | |
2082 | * requires it. | |
2083 | */ | |
2084 | set_capacity(disk, 0); | |
2085 | nvme_revalidate_disk(ns->disk); | |
2086 | add_disk(ns->disk); | |
2087 | if (ns->ms) | |
2088 | revalidate_disk(ns->disk); | |
2089 | return; | |
b60503ba MW |
2090 | out_free_queue: |
2091 | blk_cleanup_queue(ns->queue); | |
2092 | out_free_ns: | |
2093 | kfree(ns); | |
b60503ba MW |
2094 | } |
2095 | ||
42f61420 KB |
2096 | static void nvme_create_io_queues(struct nvme_dev *dev) |
2097 | { | |
a4aea562 | 2098 | unsigned i; |
42f61420 | 2099 | |
a4aea562 | 2100 | for (i = dev->queue_count; i <= dev->max_qid; i++) |
2b25d981 | 2101 | if (!nvme_alloc_queue(dev, i, dev->q_depth)) |
42f61420 KB |
2102 | break; |
2103 | ||
a4aea562 MB |
2104 | for (i = dev->online_queues; i <= dev->queue_count - 1; i++) |
2105 | if (nvme_create_queue(dev->queues[i], i)) | |
42f61420 KB |
2106 | break; |
2107 | } | |
2108 | ||
b3b06812 | 2109 | static int set_queue_count(struct nvme_dev *dev, int count) |
b60503ba MW |
2110 | { |
2111 | int status; | |
2112 | u32 result; | |
b3b06812 | 2113 | u32 q_count = (count - 1) | ((count - 1) << 16); |
b60503ba | 2114 | |
df348139 | 2115 | status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0, |
bc5fc7e4 | 2116 | &result); |
27e8166c MW |
2117 | if (status < 0) |
2118 | return status; | |
2119 | if (status > 0) { | |
e75ec752 | 2120 | dev_err(dev->dev, "Could not set queue count (%d)\n", status); |
badc34d4 | 2121 | return 0; |
27e8166c | 2122 | } |
b60503ba MW |
2123 | return min(result & 0xffff, result >> 16) + 1; |
2124 | } | |
2125 | ||
9d713c2b KB |
2126 | static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
2127 | { | |
b80d5ccc | 2128 | return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride); |
9d713c2b KB |
2129 | } |
2130 | ||
8d85fce7 | 2131 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 2132 | { |
a4aea562 | 2133 | struct nvme_queue *adminq = dev->queues[0]; |
e75ec752 | 2134 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
42f61420 | 2135 | int result, i, vecs, nr_io_queues, size; |
b60503ba | 2136 | |
42f61420 | 2137 | nr_io_queues = num_possible_cpus(); |
b348b7d5 | 2138 | result = set_queue_count(dev, nr_io_queues); |
badc34d4 | 2139 | if (result <= 0) |
1b23484b | 2140 | return result; |
b348b7d5 MW |
2141 | if (result < nr_io_queues) |
2142 | nr_io_queues = result; | |
b60503ba | 2143 | |
9d713c2b KB |
2144 | size = db_bar_size(dev, nr_io_queues); |
2145 | if (size > 8192) { | |
f1938f6e | 2146 | iounmap(dev->bar); |
9d713c2b KB |
2147 | do { |
2148 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
2149 | if (dev->bar) | |
2150 | break; | |
2151 | if (!--nr_io_queues) | |
2152 | return -ENOMEM; | |
2153 | size = db_bar_size(dev, nr_io_queues); | |
2154 | } while (1); | |
f1938f6e | 2155 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
5a92e700 | 2156 | adminq->q_db = dev->dbs; |
f1938f6e MW |
2157 | } |
2158 | ||
9d713c2b | 2159 | /* Deregister the admin queue's interrupt */ |
3193f07b | 2160 | free_irq(dev->entry[0].vector, adminq); |
9d713c2b | 2161 | |
e32efbfc JA |
2162 | /* |
2163 | * If we enable msix early due to not intx, disable it again before | |
2164 | * setting up the full range we need. | |
2165 | */ | |
2166 | if (!pdev->irq) | |
2167 | pci_disable_msix(pdev); | |
2168 | ||
be577fab | 2169 | for (i = 0; i < nr_io_queues; i++) |
1b23484b | 2170 | dev->entry[i].entry = i; |
be577fab AG |
2171 | vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues); |
2172 | if (vecs < 0) { | |
2173 | vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32)); | |
2174 | if (vecs < 0) { | |
2175 | vecs = 1; | |
2176 | } else { | |
2177 | for (i = 0; i < vecs; i++) | |
2178 | dev->entry[i].vector = i + pdev->irq; | |
fa08a396 RRG |
2179 | } |
2180 | } | |
2181 | ||
063a8096 MW |
2182 | /* |
2183 | * Should investigate if there's a performance win from allocating | |
2184 | * more queues than interrupt vectors; it might allow the submission | |
2185 | * path to scale better, even if the receive path is limited by the | |
2186 | * number of interrupts. | |
2187 | */ | |
2188 | nr_io_queues = vecs; | |
42f61420 | 2189 | dev->max_qid = nr_io_queues; |
063a8096 | 2190 | |
3193f07b | 2191 | result = queue_request_irq(dev, adminq, adminq->irqname); |
a4aea562 | 2192 | if (result) |
22404274 | 2193 | goto free_queues; |
1b23484b | 2194 | |
cd638946 | 2195 | /* Free previously allocated queues that are no longer usable */ |
42f61420 | 2196 | nvme_free_queues(dev, nr_io_queues + 1); |
a4aea562 | 2197 | nvme_create_io_queues(dev); |
9ecdc946 | 2198 | |
22404274 | 2199 | return 0; |
b60503ba | 2200 | |
22404274 | 2201 | free_queues: |
a1a5ef99 | 2202 | nvme_free_queues(dev, 1); |
22404274 | 2203 | return result; |
b60503ba MW |
2204 | } |
2205 | ||
422ef0c7 MW |
2206 | /* |
2207 | * Return: error value if an error occurred setting up the queues or calling | |
2208 | * Identify Device. 0 if these succeeded, even if adding some of the | |
2209 | * namespaces failed. At the moment, these failures are silent. TBD which | |
2210 | * failures should be reported. | |
2211 | */ | |
8d85fce7 | 2212 | static int nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 2213 | { |
e75ec752 | 2214 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
c3bfe717 MW |
2215 | int res; |
2216 | unsigned nn, i; | |
51814232 | 2217 | struct nvme_id_ctrl *ctrl; |
159b67d7 | 2218 | int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12; |
b60503ba | 2219 | |
d29ec824 | 2220 | res = nvme_identify_ctrl(dev, &ctrl); |
b60503ba | 2221 | if (res) { |
e75ec752 | 2222 | dev_err(dev->dev, "Identify Controller failed (%d)\n", res); |
e1e5e564 | 2223 | return -EIO; |
b60503ba MW |
2224 | } |
2225 | ||
51814232 | 2226 | nn = le32_to_cpup(&ctrl->nn); |
0e5e4f0e | 2227 | dev->oncs = le16_to_cpup(&ctrl->oncs); |
c30341dc | 2228 | dev->abort_limit = ctrl->acl + 1; |
a7d2ce28 | 2229 | dev->vwc = ctrl->vwc; |
51814232 MW |
2230 | memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn)); |
2231 | memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn)); | |
2232 | memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr)); | |
159b67d7 | 2233 | if (ctrl->mdts) |
8fc23e03 | 2234 | dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9); |
68608c26 | 2235 | if ((pdev->vendor == PCI_VENDOR_ID_INTEL) && |
a4aea562 MB |
2236 | (pdev->device == 0x0953) && ctrl->vs[3]) { |
2237 | unsigned int max_hw_sectors; | |
2238 | ||
159b67d7 | 2239 | dev->stripe_size = 1 << (ctrl->vs[3] + shift); |
a4aea562 MB |
2240 | max_hw_sectors = dev->stripe_size >> (shift - 9); |
2241 | if (dev->max_hw_sectors) { | |
2242 | dev->max_hw_sectors = min(max_hw_sectors, | |
2243 | dev->max_hw_sectors); | |
2244 | } else | |
2245 | dev->max_hw_sectors = max_hw_sectors; | |
2246 | } | |
d29ec824 | 2247 | kfree(ctrl); |
a4aea562 MB |
2248 | |
2249 | dev->tagset.ops = &nvme_mq_ops; | |
2250 | dev->tagset.nr_hw_queues = dev->online_queues - 1; | |
2251 | dev->tagset.timeout = NVME_IO_TIMEOUT; | |
e75ec752 | 2252 | dev->tagset.numa_node = dev_to_node(dev->dev); |
a4aea562 MB |
2253 | dev->tagset.queue_depth = |
2254 | min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; | |
ac3dd5bd | 2255 | dev->tagset.cmd_size = nvme_cmd_size(dev); |
a4aea562 MB |
2256 | dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; |
2257 | dev->tagset.driver_data = dev; | |
2258 | ||
2259 | if (blk_mq_alloc_tag_set(&dev->tagset)) | |
e1e5e564 | 2260 | return 0; |
b60503ba | 2261 | |
e1e5e564 KB |
2262 | for (i = 1; i <= nn; i++) |
2263 | nvme_alloc_ns(dev, i); | |
b60503ba | 2264 | |
e1e5e564 | 2265 | return 0; |
b60503ba MW |
2266 | } |
2267 | ||
0877cb0d KB |
2268 | static int nvme_dev_map(struct nvme_dev *dev) |
2269 | { | |
42f61420 | 2270 | u64 cap; |
0877cb0d | 2271 | int bars, result = -ENOMEM; |
e75ec752 | 2272 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
0877cb0d KB |
2273 | |
2274 | if (pci_enable_device_mem(pdev)) | |
2275 | return result; | |
2276 | ||
2277 | dev->entry[0].vector = pdev->irq; | |
2278 | pci_set_master(pdev); | |
2279 | bars = pci_select_bars(pdev, IORESOURCE_MEM); | |
be7837e8 JA |
2280 | if (!bars) |
2281 | goto disable_pci; | |
2282 | ||
0877cb0d KB |
2283 | if (pci_request_selected_regions(pdev, bars, "nvme")) |
2284 | goto disable_pci; | |
2285 | ||
e75ec752 CH |
2286 | if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) && |
2287 | dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32))) | |
052d0efa | 2288 | goto disable; |
0877cb0d | 2289 | |
0877cb0d KB |
2290 | dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); |
2291 | if (!dev->bar) | |
2292 | goto disable; | |
e32efbfc | 2293 | |
0e53d180 KB |
2294 | if (readl(&dev->bar->csts) == -1) { |
2295 | result = -ENODEV; | |
2296 | goto unmap; | |
2297 | } | |
e32efbfc JA |
2298 | |
2299 | /* | |
2300 | * Some devices don't advertse INTx interrupts, pre-enable a single | |
2301 | * MSIX vec for setup. We'll adjust this later. | |
2302 | */ | |
2303 | if (!pdev->irq) { | |
2304 | result = pci_enable_msix(pdev, dev->entry, 1); | |
2305 | if (result < 0) | |
2306 | goto unmap; | |
2307 | } | |
2308 | ||
42f61420 KB |
2309 | cap = readq(&dev->bar->cap); |
2310 | dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); | |
2311 | dev->db_stride = 1 << NVME_CAP_STRIDE(cap); | |
0877cb0d KB |
2312 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
2313 | ||
2314 | return 0; | |
2315 | ||
0e53d180 KB |
2316 | unmap: |
2317 | iounmap(dev->bar); | |
2318 | dev->bar = NULL; | |
0877cb0d KB |
2319 | disable: |
2320 | pci_release_regions(pdev); | |
2321 | disable_pci: | |
2322 | pci_disable_device(pdev); | |
2323 | return result; | |
2324 | } | |
2325 | ||
2326 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
2327 | { | |
e75ec752 CH |
2328 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2329 | ||
2330 | if (pdev->msi_enabled) | |
2331 | pci_disable_msi(pdev); | |
2332 | else if (pdev->msix_enabled) | |
2333 | pci_disable_msix(pdev); | |
0877cb0d KB |
2334 | |
2335 | if (dev->bar) { | |
2336 | iounmap(dev->bar); | |
2337 | dev->bar = NULL; | |
e75ec752 | 2338 | pci_release_regions(pdev); |
0877cb0d KB |
2339 | } |
2340 | ||
e75ec752 CH |
2341 | if (pci_is_enabled(pdev)) |
2342 | pci_disable_device(pdev); | |
0877cb0d KB |
2343 | } |
2344 | ||
4d115420 KB |
2345 | struct nvme_delq_ctx { |
2346 | struct task_struct *waiter; | |
2347 | struct kthread_worker *worker; | |
2348 | atomic_t refcount; | |
2349 | }; | |
2350 | ||
2351 | static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev) | |
2352 | { | |
2353 | dq->waiter = current; | |
2354 | mb(); | |
2355 | ||
2356 | for (;;) { | |
2357 | set_current_state(TASK_KILLABLE); | |
2358 | if (!atomic_read(&dq->refcount)) | |
2359 | break; | |
2360 | if (!schedule_timeout(ADMIN_TIMEOUT) || | |
2361 | fatal_signal_pending(current)) { | |
0fb59cbc KB |
2362 | /* |
2363 | * Disable the controller first since we can't trust it | |
2364 | * at this point, but leave the admin queue enabled | |
2365 | * until all queue deletion requests are flushed. | |
2366 | * FIXME: This may take a while if there are more h/w | |
2367 | * queues than admin tags. | |
2368 | */ | |
4d115420 | 2369 | set_current_state(TASK_RUNNING); |
4d115420 | 2370 | nvme_disable_ctrl(dev, readq(&dev->bar->cap)); |
0fb59cbc | 2371 | nvme_clear_queue(dev->queues[0]); |
4d115420 | 2372 | flush_kthread_worker(dq->worker); |
0fb59cbc | 2373 | nvme_disable_queue(dev, 0); |
4d115420 KB |
2374 | return; |
2375 | } | |
2376 | } | |
2377 | set_current_state(TASK_RUNNING); | |
2378 | } | |
2379 | ||
2380 | static void nvme_put_dq(struct nvme_delq_ctx *dq) | |
2381 | { | |
2382 | atomic_dec(&dq->refcount); | |
2383 | if (dq->waiter) | |
2384 | wake_up_process(dq->waiter); | |
2385 | } | |
2386 | ||
2387 | static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq) | |
2388 | { | |
2389 | atomic_inc(&dq->refcount); | |
2390 | return dq; | |
2391 | } | |
2392 | ||
2393 | static void nvme_del_queue_end(struct nvme_queue *nvmeq) | |
2394 | { | |
2395 | struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx; | |
4d115420 KB |
2396 | nvme_put_dq(dq); |
2397 | } | |
2398 | ||
2399 | static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode, | |
2400 | kthread_work_func_t fn) | |
2401 | { | |
2402 | struct nvme_command c; | |
2403 | ||
2404 | memset(&c, 0, sizeof(c)); | |
2405 | c.delete_queue.opcode = opcode; | |
2406 | c.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
2407 | ||
2408 | init_kthread_work(&nvmeq->cmdinfo.work, fn); | |
a4aea562 MB |
2409 | return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo, |
2410 | ADMIN_TIMEOUT); | |
4d115420 KB |
2411 | } |
2412 | ||
2413 | static void nvme_del_cq_work_handler(struct kthread_work *work) | |
2414 | { | |
2415 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2416 | cmdinfo.work); | |
2417 | nvme_del_queue_end(nvmeq); | |
2418 | } | |
2419 | ||
2420 | static int nvme_delete_cq(struct nvme_queue *nvmeq) | |
2421 | { | |
2422 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq, | |
2423 | nvme_del_cq_work_handler); | |
2424 | } | |
2425 | ||
2426 | static void nvme_del_sq_work_handler(struct kthread_work *work) | |
2427 | { | |
2428 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2429 | cmdinfo.work); | |
2430 | int status = nvmeq->cmdinfo.status; | |
2431 | ||
2432 | if (!status) | |
2433 | status = nvme_delete_cq(nvmeq); | |
2434 | if (status) | |
2435 | nvme_del_queue_end(nvmeq); | |
2436 | } | |
2437 | ||
2438 | static int nvme_delete_sq(struct nvme_queue *nvmeq) | |
2439 | { | |
2440 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq, | |
2441 | nvme_del_sq_work_handler); | |
2442 | } | |
2443 | ||
2444 | static void nvme_del_queue_start(struct kthread_work *work) | |
2445 | { | |
2446 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2447 | cmdinfo.work); | |
4d115420 KB |
2448 | if (nvme_delete_sq(nvmeq)) |
2449 | nvme_del_queue_end(nvmeq); | |
2450 | } | |
2451 | ||
2452 | static void nvme_disable_io_queues(struct nvme_dev *dev) | |
2453 | { | |
2454 | int i; | |
2455 | DEFINE_KTHREAD_WORKER_ONSTACK(worker); | |
2456 | struct nvme_delq_ctx dq; | |
2457 | struct task_struct *kworker_task = kthread_run(kthread_worker_fn, | |
2458 | &worker, "nvme%d", dev->instance); | |
2459 | ||
2460 | if (IS_ERR(kworker_task)) { | |
e75ec752 | 2461 | dev_err(dev->dev, |
4d115420 KB |
2462 | "Failed to create queue del task\n"); |
2463 | for (i = dev->queue_count - 1; i > 0; i--) | |
2464 | nvme_disable_queue(dev, i); | |
2465 | return; | |
2466 | } | |
2467 | ||
2468 | dq.waiter = NULL; | |
2469 | atomic_set(&dq.refcount, 0); | |
2470 | dq.worker = &worker; | |
2471 | for (i = dev->queue_count - 1; i > 0; i--) { | |
a4aea562 | 2472 | struct nvme_queue *nvmeq = dev->queues[i]; |
4d115420 KB |
2473 | |
2474 | if (nvme_suspend_queue(nvmeq)) | |
2475 | continue; | |
2476 | nvmeq->cmdinfo.ctx = nvme_get_dq(&dq); | |
2477 | nvmeq->cmdinfo.worker = dq.worker; | |
2478 | init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start); | |
2479 | queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work); | |
2480 | } | |
2481 | nvme_wait_dq(&dq, dev); | |
2482 | kthread_stop(kworker_task); | |
2483 | } | |
2484 | ||
b9afca3e DM |
2485 | /* |
2486 | * Remove the node from the device list and check | |
2487 | * for whether or not we need to stop the nvme_thread. | |
2488 | */ | |
2489 | static void nvme_dev_list_remove(struct nvme_dev *dev) | |
2490 | { | |
2491 | struct task_struct *tmp = NULL; | |
2492 | ||
2493 | spin_lock(&dev_list_lock); | |
2494 | list_del_init(&dev->node); | |
2495 | if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) { | |
2496 | tmp = nvme_thread; | |
2497 | nvme_thread = NULL; | |
2498 | } | |
2499 | spin_unlock(&dev_list_lock); | |
2500 | ||
2501 | if (tmp) | |
2502 | kthread_stop(tmp); | |
2503 | } | |
2504 | ||
c9d3bf88 KB |
2505 | static void nvme_freeze_queues(struct nvme_dev *dev) |
2506 | { | |
2507 | struct nvme_ns *ns; | |
2508 | ||
2509 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2510 | blk_mq_freeze_queue_start(ns->queue); | |
2511 | ||
cddcd72b | 2512 | spin_lock_irq(ns->queue->queue_lock); |
c9d3bf88 | 2513 | queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue); |
cddcd72b | 2514 | spin_unlock_irq(ns->queue->queue_lock); |
c9d3bf88 KB |
2515 | |
2516 | blk_mq_cancel_requeue_work(ns->queue); | |
2517 | blk_mq_stop_hw_queues(ns->queue); | |
2518 | } | |
2519 | } | |
2520 | ||
2521 | static void nvme_unfreeze_queues(struct nvme_dev *dev) | |
2522 | { | |
2523 | struct nvme_ns *ns; | |
2524 | ||
2525 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2526 | queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue); | |
2527 | blk_mq_unfreeze_queue(ns->queue); | |
2528 | blk_mq_start_stopped_hw_queues(ns->queue, true); | |
2529 | blk_mq_kick_requeue_list(ns->queue); | |
2530 | } | |
2531 | } | |
2532 | ||
f0b50732 | 2533 | static void nvme_dev_shutdown(struct nvme_dev *dev) |
b60503ba | 2534 | { |
22404274 | 2535 | int i; |
7c1b2450 | 2536 | u32 csts = -1; |
22404274 | 2537 | |
b9afca3e | 2538 | nvme_dev_list_remove(dev); |
1fa6aead | 2539 | |
c9d3bf88 KB |
2540 | if (dev->bar) { |
2541 | nvme_freeze_queues(dev); | |
7c1b2450 | 2542 | csts = readl(&dev->bar->csts); |
c9d3bf88 | 2543 | } |
7c1b2450 | 2544 | if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) { |
4d115420 | 2545 | for (i = dev->queue_count - 1; i >= 0; i--) { |
a4aea562 | 2546 | struct nvme_queue *nvmeq = dev->queues[i]; |
4d115420 | 2547 | nvme_suspend_queue(nvmeq); |
4d115420 KB |
2548 | } |
2549 | } else { | |
2550 | nvme_disable_io_queues(dev); | |
1894d8f1 | 2551 | nvme_shutdown_ctrl(dev); |
4d115420 KB |
2552 | nvme_disable_queue(dev, 0); |
2553 | } | |
f0b50732 | 2554 | nvme_dev_unmap(dev); |
07836e65 KB |
2555 | |
2556 | for (i = dev->queue_count - 1; i >= 0; i--) | |
2557 | nvme_clear_queue(dev->queues[i]); | |
f0b50732 KB |
2558 | } |
2559 | ||
2560 | static void nvme_dev_remove(struct nvme_dev *dev) | |
2561 | { | |
9ac27090 | 2562 | struct nvme_ns *ns; |
f0b50732 | 2563 | |
9ac27090 | 2564 | list_for_each_entry(ns, &dev->namespaces, list) { |
e1e5e564 | 2565 | if (ns->disk->flags & GENHD_FL_UP) { |
52b68d7e | 2566 | if (blk_get_integrity(ns->disk)) |
e1e5e564 | 2567 | blk_integrity_unregister(ns->disk); |
9ac27090 | 2568 | del_gendisk(ns->disk); |
e1e5e564 | 2569 | } |
cef6a948 KB |
2570 | if (!blk_queue_dying(ns->queue)) { |
2571 | blk_mq_abort_requeue_list(ns->queue); | |
9ac27090 | 2572 | blk_cleanup_queue(ns->queue); |
cef6a948 | 2573 | } |
b60503ba | 2574 | } |
b60503ba MW |
2575 | } |
2576 | ||
091b6092 MW |
2577 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
2578 | { | |
e75ec752 | 2579 | dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, |
091b6092 MW |
2580 | PAGE_SIZE, PAGE_SIZE, 0); |
2581 | if (!dev->prp_page_pool) | |
2582 | return -ENOMEM; | |
2583 | ||
99802a7a | 2584 | /* Optimisation for I/Os between 4k and 128k */ |
e75ec752 | 2585 | dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, |
99802a7a MW |
2586 | 256, 256, 0); |
2587 | if (!dev->prp_small_pool) { | |
2588 | dma_pool_destroy(dev->prp_page_pool); | |
2589 | return -ENOMEM; | |
2590 | } | |
091b6092 MW |
2591 | return 0; |
2592 | } | |
2593 | ||
2594 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
2595 | { | |
2596 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 2597 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
2598 | } |
2599 | ||
cd58ad7d QSA |
2600 | static DEFINE_IDA(nvme_instance_ida); |
2601 | ||
2602 | static int nvme_set_instance(struct nvme_dev *dev) | |
b60503ba | 2603 | { |
cd58ad7d QSA |
2604 | int instance, error; |
2605 | ||
2606 | do { | |
2607 | if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL)) | |
2608 | return -ENODEV; | |
2609 | ||
2610 | spin_lock(&dev_list_lock); | |
2611 | error = ida_get_new(&nvme_instance_ida, &instance); | |
2612 | spin_unlock(&dev_list_lock); | |
2613 | } while (error == -EAGAIN); | |
2614 | ||
2615 | if (error) | |
2616 | return -ENODEV; | |
2617 | ||
2618 | dev->instance = instance; | |
2619 | return 0; | |
b60503ba MW |
2620 | } |
2621 | ||
2622 | static void nvme_release_instance(struct nvme_dev *dev) | |
2623 | { | |
cd58ad7d QSA |
2624 | spin_lock(&dev_list_lock); |
2625 | ida_remove(&nvme_instance_ida, dev->instance); | |
2626 | spin_unlock(&dev_list_lock); | |
b60503ba MW |
2627 | } |
2628 | ||
9ac27090 KB |
2629 | static void nvme_free_namespaces(struct nvme_dev *dev) |
2630 | { | |
2631 | struct nvme_ns *ns, *next; | |
2632 | ||
2633 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) { | |
2634 | list_del(&ns->list); | |
9e60352c KB |
2635 | |
2636 | spin_lock(&dev_list_lock); | |
2637 | ns->disk->private_data = NULL; | |
2638 | spin_unlock(&dev_list_lock); | |
2639 | ||
9ac27090 KB |
2640 | put_disk(ns->disk); |
2641 | kfree(ns); | |
2642 | } | |
2643 | } | |
2644 | ||
5e82e952 KB |
2645 | static void nvme_free_dev(struct kref *kref) |
2646 | { | |
2647 | struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref); | |
9ac27090 | 2648 | |
e75ec752 | 2649 | put_device(dev->dev); |
b3fffdef | 2650 | put_device(dev->device); |
9ac27090 | 2651 | nvme_free_namespaces(dev); |
285dffc9 | 2652 | nvme_release_instance(dev); |
a4aea562 | 2653 | blk_mq_free_tag_set(&dev->tagset); |
ea191d2f | 2654 | blk_put_queue(dev->admin_q); |
5e82e952 KB |
2655 | kfree(dev->queues); |
2656 | kfree(dev->entry); | |
2657 | kfree(dev); | |
2658 | } | |
2659 | ||
2660 | static int nvme_dev_open(struct inode *inode, struct file *f) | |
2661 | { | |
b3fffdef KB |
2662 | struct nvme_dev *dev; |
2663 | int instance = iminor(inode); | |
2664 | int ret = -ENODEV; | |
2665 | ||
2666 | spin_lock(&dev_list_lock); | |
2667 | list_for_each_entry(dev, &dev_list, node) { | |
2668 | if (dev->instance == instance) { | |
2e1d8448 KB |
2669 | if (!dev->admin_q) { |
2670 | ret = -EWOULDBLOCK; | |
2671 | break; | |
2672 | } | |
b3fffdef KB |
2673 | if (!kref_get_unless_zero(&dev->kref)) |
2674 | break; | |
2675 | f->private_data = dev; | |
2676 | ret = 0; | |
2677 | break; | |
2678 | } | |
2679 | } | |
2680 | spin_unlock(&dev_list_lock); | |
2681 | ||
2682 | return ret; | |
5e82e952 KB |
2683 | } |
2684 | ||
2685 | static int nvme_dev_release(struct inode *inode, struct file *f) | |
2686 | { | |
2687 | struct nvme_dev *dev = f->private_data; | |
2688 | kref_put(&dev->kref, nvme_free_dev); | |
2689 | return 0; | |
2690 | } | |
2691 | ||
2692 | static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg) | |
2693 | { | |
2694 | struct nvme_dev *dev = f->private_data; | |
a4aea562 MB |
2695 | struct nvme_ns *ns; |
2696 | ||
5e82e952 KB |
2697 | switch (cmd) { |
2698 | case NVME_IOCTL_ADMIN_CMD: | |
a4aea562 | 2699 | return nvme_user_cmd(dev, NULL, (void __user *)arg); |
7963e521 | 2700 | case NVME_IOCTL_IO_CMD: |
a4aea562 MB |
2701 | if (list_empty(&dev->namespaces)) |
2702 | return -ENOTTY; | |
2703 | ns = list_first_entry(&dev->namespaces, struct nvme_ns, list); | |
2704 | return nvme_user_cmd(dev, ns, (void __user *)arg); | |
5e82e952 KB |
2705 | default: |
2706 | return -ENOTTY; | |
2707 | } | |
2708 | } | |
2709 | ||
2710 | static const struct file_operations nvme_dev_fops = { | |
2711 | .owner = THIS_MODULE, | |
2712 | .open = nvme_dev_open, | |
2713 | .release = nvme_dev_release, | |
2714 | .unlocked_ioctl = nvme_dev_ioctl, | |
2715 | .compat_ioctl = nvme_dev_ioctl, | |
2716 | }; | |
2717 | ||
a4aea562 MB |
2718 | static void nvme_set_irq_hints(struct nvme_dev *dev) |
2719 | { | |
2720 | struct nvme_queue *nvmeq; | |
2721 | int i; | |
2722 | ||
2723 | for (i = 0; i < dev->online_queues; i++) { | |
2724 | nvmeq = dev->queues[i]; | |
2725 | ||
2726 | if (!nvmeq->hctx) | |
2727 | continue; | |
2728 | ||
2729 | irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector, | |
2730 | nvmeq->hctx->cpumask); | |
2731 | } | |
2732 | } | |
2733 | ||
f0b50732 KB |
2734 | static int nvme_dev_start(struct nvme_dev *dev) |
2735 | { | |
2736 | int result; | |
b9afca3e | 2737 | bool start_thread = false; |
f0b50732 KB |
2738 | |
2739 | result = nvme_dev_map(dev); | |
2740 | if (result) | |
2741 | return result; | |
2742 | ||
2743 | result = nvme_configure_admin_queue(dev); | |
2744 | if (result) | |
2745 | goto unmap; | |
2746 | ||
2747 | spin_lock(&dev_list_lock); | |
b9afca3e DM |
2748 | if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) { |
2749 | start_thread = true; | |
2750 | nvme_thread = NULL; | |
2751 | } | |
f0b50732 KB |
2752 | list_add(&dev->node, &dev_list); |
2753 | spin_unlock(&dev_list_lock); | |
2754 | ||
b9afca3e DM |
2755 | if (start_thread) { |
2756 | nvme_thread = kthread_run(nvme_kthread, NULL, "nvme"); | |
387caa5a | 2757 | wake_up_all(&nvme_kthread_wait); |
b9afca3e DM |
2758 | } else |
2759 | wait_event_killable(nvme_kthread_wait, nvme_thread); | |
2760 | ||
2761 | if (IS_ERR_OR_NULL(nvme_thread)) { | |
2762 | result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR; | |
2763 | goto disable; | |
2764 | } | |
a4aea562 MB |
2765 | |
2766 | nvme_init_queue(dev->queues[0], 0); | |
0fb59cbc KB |
2767 | result = nvme_alloc_admin_tags(dev); |
2768 | if (result) | |
2769 | goto disable; | |
b9afca3e | 2770 | |
f0b50732 | 2771 | result = nvme_setup_io_queues(dev); |
badc34d4 | 2772 | if (result) |
0fb59cbc | 2773 | goto free_tags; |
f0b50732 | 2774 | |
a4aea562 MB |
2775 | nvme_set_irq_hints(dev); |
2776 | ||
1efccc9d | 2777 | dev->event_limit = 1; |
d82e8bfd | 2778 | return result; |
f0b50732 | 2779 | |
0fb59cbc KB |
2780 | free_tags: |
2781 | nvme_dev_remove_admin(dev); | |
f0b50732 | 2782 | disable: |
a1a5ef99 | 2783 | nvme_disable_queue(dev, 0); |
b9afca3e | 2784 | nvme_dev_list_remove(dev); |
f0b50732 KB |
2785 | unmap: |
2786 | nvme_dev_unmap(dev); | |
2787 | return result; | |
2788 | } | |
2789 | ||
9a6b9458 KB |
2790 | static int nvme_remove_dead_ctrl(void *arg) |
2791 | { | |
2792 | struct nvme_dev *dev = (struct nvme_dev *)arg; | |
e75ec752 | 2793 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
9a6b9458 KB |
2794 | |
2795 | if (pci_get_drvdata(pdev)) | |
c81f4975 | 2796 | pci_stop_and_remove_bus_device_locked(pdev); |
9a6b9458 KB |
2797 | kref_put(&dev->kref, nvme_free_dev); |
2798 | return 0; | |
2799 | } | |
2800 | ||
2801 | static void nvme_remove_disks(struct work_struct *ws) | |
2802 | { | |
9a6b9458 KB |
2803 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); |
2804 | ||
5a92e700 | 2805 | nvme_free_queues(dev, 1); |
302c6727 | 2806 | nvme_dev_remove(dev); |
9a6b9458 KB |
2807 | } |
2808 | ||
2809 | static int nvme_dev_resume(struct nvme_dev *dev) | |
2810 | { | |
2811 | int ret; | |
2812 | ||
2813 | ret = nvme_dev_start(dev); | |
badc34d4 | 2814 | if (ret) |
9a6b9458 | 2815 | return ret; |
badc34d4 | 2816 | if (dev->online_queues < 2) { |
9a6b9458 | 2817 | spin_lock(&dev_list_lock); |
9ca97374 | 2818 | dev->reset_workfn = nvme_remove_disks; |
9a6b9458 KB |
2819 | queue_work(nvme_workq, &dev->reset_work); |
2820 | spin_unlock(&dev_list_lock); | |
c9d3bf88 KB |
2821 | } else { |
2822 | nvme_unfreeze_queues(dev); | |
2823 | nvme_set_irq_hints(dev); | |
9a6b9458 KB |
2824 | } |
2825 | return 0; | |
2826 | } | |
2827 | ||
2828 | static void nvme_dev_reset(struct nvme_dev *dev) | |
2829 | { | |
2830 | nvme_dev_shutdown(dev); | |
2831 | if (nvme_dev_resume(dev)) { | |
e75ec752 | 2832 | dev_warn(dev->dev, "Device failed to resume\n"); |
9a6b9458 KB |
2833 | kref_get(&dev->kref); |
2834 | if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d", | |
2835 | dev->instance))) { | |
e75ec752 | 2836 | dev_err(dev->dev, |
9a6b9458 KB |
2837 | "Failed to start controller remove task\n"); |
2838 | kref_put(&dev->kref, nvme_free_dev); | |
2839 | } | |
2840 | } | |
2841 | } | |
2842 | ||
2843 | static void nvme_reset_failed_dev(struct work_struct *ws) | |
2844 | { | |
2845 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); | |
2846 | nvme_dev_reset(dev); | |
2847 | } | |
2848 | ||
9ca97374 TH |
2849 | static void nvme_reset_workfn(struct work_struct *work) |
2850 | { | |
2851 | struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work); | |
2852 | dev->reset_workfn(work); | |
2853 | } | |
2854 | ||
2e1d8448 | 2855 | static void nvme_async_probe(struct work_struct *work); |
8d85fce7 | 2856 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 2857 | { |
a4aea562 | 2858 | int node, result = -ENOMEM; |
b60503ba MW |
2859 | struct nvme_dev *dev; |
2860 | ||
a4aea562 MB |
2861 | node = dev_to_node(&pdev->dev); |
2862 | if (node == NUMA_NO_NODE) | |
2863 | set_dev_node(&pdev->dev, 0); | |
2864 | ||
2865 | dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); | |
b60503ba MW |
2866 | if (!dev) |
2867 | return -ENOMEM; | |
a4aea562 MB |
2868 | dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry), |
2869 | GFP_KERNEL, node); | |
b60503ba MW |
2870 | if (!dev->entry) |
2871 | goto free; | |
a4aea562 MB |
2872 | dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *), |
2873 | GFP_KERNEL, node); | |
b60503ba MW |
2874 | if (!dev->queues) |
2875 | goto free; | |
2876 | ||
2877 | INIT_LIST_HEAD(&dev->namespaces); | |
9ca97374 TH |
2878 | dev->reset_workfn = nvme_reset_failed_dev; |
2879 | INIT_WORK(&dev->reset_work, nvme_reset_workfn); | |
e75ec752 | 2880 | dev->dev = get_device(&pdev->dev); |
9a6b9458 | 2881 | pci_set_drvdata(pdev, dev); |
cd58ad7d QSA |
2882 | result = nvme_set_instance(dev); |
2883 | if (result) | |
a96d4f5c | 2884 | goto put_pci; |
b60503ba | 2885 | |
091b6092 MW |
2886 | result = nvme_setup_prp_pools(dev); |
2887 | if (result) | |
0877cb0d | 2888 | goto release; |
091b6092 | 2889 | |
fb35e914 | 2890 | kref_init(&dev->kref); |
b3fffdef KB |
2891 | dev->device = device_create(nvme_class, &pdev->dev, |
2892 | MKDEV(nvme_char_major, dev->instance), | |
2893 | dev, "nvme%d", dev->instance); | |
2894 | if (IS_ERR(dev->device)) { | |
2895 | result = PTR_ERR(dev->device); | |
2e1d8448 | 2896 | goto release_pools; |
b3fffdef KB |
2897 | } |
2898 | get_device(dev->device); | |
740216fc | 2899 | |
e6e96d73 | 2900 | INIT_LIST_HEAD(&dev->node); |
2e1d8448 KB |
2901 | INIT_WORK(&dev->probe_work, nvme_async_probe); |
2902 | schedule_work(&dev->probe_work); | |
b60503ba MW |
2903 | return 0; |
2904 | ||
0877cb0d | 2905 | release_pools: |
091b6092 | 2906 | nvme_release_prp_pools(dev); |
0877cb0d KB |
2907 | release: |
2908 | nvme_release_instance(dev); | |
a96d4f5c | 2909 | put_pci: |
e75ec752 | 2910 | put_device(dev->dev); |
b60503ba MW |
2911 | free: |
2912 | kfree(dev->queues); | |
2913 | kfree(dev->entry); | |
2914 | kfree(dev); | |
2915 | return result; | |
2916 | } | |
2917 | ||
2e1d8448 KB |
2918 | static void nvme_async_probe(struct work_struct *work) |
2919 | { | |
2920 | struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work); | |
2921 | int result; | |
2922 | ||
2923 | result = nvme_dev_start(dev); | |
2924 | if (result) | |
2925 | goto reset; | |
2926 | ||
2927 | if (dev->online_queues > 1) | |
2928 | result = nvme_dev_add(dev); | |
2929 | if (result) | |
2930 | goto reset; | |
2931 | ||
2932 | nvme_set_irq_hints(dev); | |
2e1d8448 KB |
2933 | return; |
2934 | reset: | |
07836e65 KB |
2935 | if (!work_busy(&dev->reset_work)) { |
2936 | dev->reset_workfn = nvme_reset_failed_dev; | |
2937 | queue_work(nvme_workq, &dev->reset_work); | |
2938 | } | |
2e1d8448 KB |
2939 | } |
2940 | ||
f0d54a54 KB |
2941 | static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) |
2942 | { | |
a6739479 | 2943 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
f0d54a54 | 2944 | |
a6739479 KB |
2945 | if (prepare) |
2946 | nvme_dev_shutdown(dev); | |
2947 | else | |
2948 | nvme_dev_resume(dev); | |
f0d54a54 KB |
2949 | } |
2950 | ||
09ece142 KB |
2951 | static void nvme_shutdown(struct pci_dev *pdev) |
2952 | { | |
2953 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
2954 | nvme_dev_shutdown(dev); | |
2955 | } | |
2956 | ||
8d85fce7 | 2957 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
2958 | { |
2959 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 KB |
2960 | |
2961 | spin_lock(&dev_list_lock); | |
2962 | list_del_init(&dev->node); | |
2963 | spin_unlock(&dev_list_lock); | |
2964 | ||
2965 | pci_set_drvdata(pdev, NULL); | |
2e1d8448 | 2966 | flush_work(&dev->probe_work); |
9a6b9458 | 2967 | flush_work(&dev->reset_work); |
9a6b9458 | 2968 | nvme_dev_shutdown(dev); |
c9d3bf88 | 2969 | nvme_dev_remove(dev); |
a4aea562 | 2970 | nvme_dev_remove_admin(dev); |
b3fffdef | 2971 | device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance)); |
a1a5ef99 | 2972 | nvme_free_queues(dev, 0); |
9a6b9458 | 2973 | nvme_release_prp_pools(dev); |
5e82e952 | 2974 | kref_put(&dev->kref, nvme_free_dev); |
b60503ba MW |
2975 | } |
2976 | ||
2977 | /* These functions are yet to be implemented */ | |
2978 | #define nvme_error_detected NULL | |
2979 | #define nvme_dump_registers NULL | |
2980 | #define nvme_link_reset NULL | |
2981 | #define nvme_slot_reset NULL | |
2982 | #define nvme_error_resume NULL | |
cd638946 | 2983 | |
671a6018 | 2984 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
2985 | static int nvme_suspend(struct device *dev) |
2986 | { | |
2987 | struct pci_dev *pdev = to_pci_dev(dev); | |
2988 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
2989 | ||
2990 | nvme_dev_shutdown(ndev); | |
2991 | return 0; | |
2992 | } | |
2993 | ||
2994 | static int nvme_resume(struct device *dev) | |
2995 | { | |
2996 | struct pci_dev *pdev = to_pci_dev(dev); | |
2997 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 2998 | |
9a6b9458 | 2999 | if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) { |
9ca97374 | 3000 | ndev->reset_workfn = nvme_reset_failed_dev; |
9a6b9458 KB |
3001 | queue_work(nvme_workq, &ndev->reset_work); |
3002 | } | |
3003 | return 0; | |
cd638946 | 3004 | } |
671a6018 | 3005 | #endif |
cd638946 KB |
3006 | |
3007 | static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); | |
b60503ba | 3008 | |
1d352035 | 3009 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba MW |
3010 | .error_detected = nvme_error_detected, |
3011 | .mmio_enabled = nvme_dump_registers, | |
3012 | .link_reset = nvme_link_reset, | |
3013 | .slot_reset = nvme_slot_reset, | |
3014 | .resume = nvme_error_resume, | |
f0d54a54 | 3015 | .reset_notify = nvme_reset_notify, |
b60503ba MW |
3016 | }; |
3017 | ||
3018 | /* Move to pci_ids.h later */ | |
3019 | #define PCI_CLASS_STORAGE_EXPRESS 0x010802 | |
3020 | ||
6eb0d698 | 3021 | static const struct pci_device_id nvme_id_table[] = { |
b60503ba MW |
3022 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
3023 | { 0, } | |
3024 | }; | |
3025 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
3026 | ||
3027 | static struct pci_driver nvme_driver = { | |
3028 | .name = "nvme", | |
3029 | .id_table = nvme_id_table, | |
3030 | .probe = nvme_probe, | |
8d85fce7 | 3031 | .remove = nvme_remove, |
09ece142 | 3032 | .shutdown = nvme_shutdown, |
cd638946 KB |
3033 | .driver = { |
3034 | .pm = &nvme_dev_pm_ops, | |
3035 | }, | |
b60503ba MW |
3036 | .err_handler = &nvme_err_handler, |
3037 | }; | |
3038 | ||
3039 | static int __init nvme_init(void) | |
3040 | { | |
0ac13140 | 3041 | int result; |
1fa6aead | 3042 | |
b9afca3e | 3043 | init_waitqueue_head(&nvme_kthread_wait); |
b60503ba | 3044 | |
9a6b9458 KB |
3045 | nvme_workq = create_singlethread_workqueue("nvme"); |
3046 | if (!nvme_workq) | |
b9afca3e | 3047 | return -ENOMEM; |
9a6b9458 | 3048 | |
5c42ea16 KB |
3049 | result = register_blkdev(nvme_major, "nvme"); |
3050 | if (result < 0) | |
9a6b9458 | 3051 | goto kill_workq; |
5c42ea16 | 3052 | else if (result > 0) |
0ac13140 | 3053 | nvme_major = result; |
b60503ba | 3054 | |
b3fffdef KB |
3055 | result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme", |
3056 | &nvme_dev_fops); | |
3057 | if (result < 0) | |
3058 | goto unregister_blkdev; | |
3059 | else if (result > 0) | |
3060 | nvme_char_major = result; | |
3061 | ||
3062 | nvme_class = class_create(THIS_MODULE, "nvme"); | |
c727040b AK |
3063 | if (IS_ERR(nvme_class)) { |
3064 | result = PTR_ERR(nvme_class); | |
b3fffdef | 3065 | goto unregister_chrdev; |
c727040b | 3066 | } |
b3fffdef | 3067 | |
f3db22fe KB |
3068 | result = pci_register_driver(&nvme_driver); |
3069 | if (result) | |
b3fffdef | 3070 | goto destroy_class; |
1fa6aead | 3071 | return 0; |
b60503ba | 3072 | |
b3fffdef KB |
3073 | destroy_class: |
3074 | class_destroy(nvme_class); | |
3075 | unregister_chrdev: | |
3076 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
1fa6aead | 3077 | unregister_blkdev: |
b60503ba | 3078 | unregister_blkdev(nvme_major, "nvme"); |
9a6b9458 KB |
3079 | kill_workq: |
3080 | destroy_workqueue(nvme_workq); | |
b60503ba MW |
3081 | return result; |
3082 | } | |
3083 | ||
3084 | static void __exit nvme_exit(void) | |
3085 | { | |
3086 | pci_unregister_driver(&nvme_driver); | |
3087 | unregister_blkdev(nvme_major, "nvme"); | |
9a6b9458 | 3088 | destroy_workqueue(nvme_workq); |
b3fffdef KB |
3089 | class_destroy(nvme_class); |
3090 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
b9afca3e | 3091 | BUG_ON(nvme_thread && !IS_ERR(nvme_thread)); |
21bd78bc | 3092 | _nvme_check_size(); |
b60503ba MW |
3093 | } |
3094 | ||
3095 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
3096 | MODULE_LICENSE("GPL"); | |
c78b4713 | 3097 | MODULE_VERSION("1.0"); |
b60503ba MW |
3098 | module_init(nvme_init); |
3099 | module_exit(nvme_exit); |