]>
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 | ||
8de05535 | 15 | #include <linux/bitops.h> |
b60503ba | 16 | #include <linux/blkdev.h> |
a4aea562 | 17 | #include <linux/blk-mq.h> |
42f61420 | 18 | #include <linux/cpu.h> |
fd63e9ce | 19 | #include <linux/delay.h> |
b60503ba MW |
20 | #include <linux/errno.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/genhd.h> | |
4cc09e2d | 23 | #include <linux/hdreg.h> |
5aff9382 | 24 | #include <linux/idr.h> |
b60503ba MW |
25 | #include <linux/init.h> |
26 | #include <linux/interrupt.h> | |
27 | #include <linux/io.h> | |
28 | #include <linux/kdev_t.h> | |
1fa6aead | 29 | #include <linux/kthread.h> |
b60503ba | 30 | #include <linux/kernel.h> |
a5768aa8 | 31 | #include <linux/list_sort.h> |
b60503ba MW |
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> |
1d277a63 | 42 | #include <linux/pr.h> |
5d0f6131 | 43 | #include <scsi/sg.h> |
797a796a | 44 | #include <asm-generic/io-64-nonatomic-lo-hi.h> |
1d277a63 | 45 | #include <asm/unaligned.h> |
797a796a | 46 | |
9d99a8dd | 47 | #include <uapi/linux/nvme_ioctl.h> |
f11bb3e2 CH |
48 | #include "nvme.h" |
49 | ||
b3fffdef | 50 | #define NVME_MINORS (1U << MINORBITS) |
9d43cf64 | 51 | #define NVME_Q_DEPTH 1024 |
d31af0a3 | 52 | #define NVME_AQ_DEPTH 256 |
b60503ba MW |
53 | #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) |
54 | #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) | |
9d43cf64 | 55 | #define ADMIN_TIMEOUT (admin_timeout * HZ) |
2484f407 | 56 | #define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ) |
9d43cf64 KB |
57 | |
58 | static unsigned char admin_timeout = 60; | |
59 | module_param(admin_timeout, byte, 0644); | |
60 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); | |
b60503ba | 61 | |
bd67608a MW |
62 | unsigned char nvme_io_timeout = 30; |
63 | module_param_named(io_timeout, nvme_io_timeout, byte, 0644); | |
b355084a | 64 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); |
b60503ba | 65 | |
2484f407 DM |
66 | static unsigned char shutdown_timeout = 5; |
67 | module_param(shutdown_timeout, byte, 0644); | |
68 | MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); | |
69 | ||
b60503ba MW |
70 | static int nvme_major; |
71 | module_param(nvme_major, int, 0); | |
72 | ||
b3fffdef KB |
73 | static int nvme_char_major; |
74 | module_param(nvme_char_major, int, 0); | |
75 | ||
58ffacb5 MW |
76 | static int use_threaded_interrupts; |
77 | module_param(use_threaded_interrupts, int, 0); | |
78 | ||
8ffaadf7 JD |
79 | static bool use_cmb_sqes = true; |
80 | module_param(use_cmb_sqes, bool, 0644); | |
81 | MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes"); | |
82 | ||
1fa6aead MW |
83 | static DEFINE_SPINLOCK(dev_list_lock); |
84 | static LIST_HEAD(dev_list); | |
85 | static struct task_struct *nvme_thread; | |
9a6b9458 | 86 | static struct workqueue_struct *nvme_workq; |
b9afca3e | 87 | static wait_queue_head_t nvme_kthread_wait; |
1fa6aead | 88 | |
b3fffdef KB |
89 | static struct class *nvme_class; |
90 | ||
90667892 | 91 | static int __nvme_reset(struct nvme_dev *dev); |
4cc06521 | 92 | static int nvme_reset(struct nvme_dev *dev); |
a4aea562 | 93 | static int nvme_process_cq(struct nvme_queue *nvmeq); |
3cf519b5 | 94 | static void nvme_dead_ctrl(struct nvme_dev *dev); |
d4b4ff8e | 95 | |
4d115420 KB |
96 | struct async_cmd_info { |
97 | struct kthread_work work; | |
98 | struct kthread_worker *worker; | |
a4aea562 | 99 | struct request *req; |
4d115420 KB |
100 | u32 result; |
101 | int status; | |
102 | void *ctx; | |
103 | }; | |
1fa6aead | 104 | |
b60503ba MW |
105 | /* |
106 | * An NVM Express queue. Each device has at least two (one for admin | |
107 | * commands and one for I/O commands). | |
108 | */ | |
109 | struct nvme_queue { | |
110 | struct device *q_dmadev; | |
091b6092 | 111 | struct nvme_dev *dev; |
3193f07b | 112 | char irqname[24]; /* nvme4294967295-65535\0 */ |
b60503ba MW |
113 | spinlock_t q_lock; |
114 | struct nvme_command *sq_cmds; | |
8ffaadf7 | 115 | struct nvme_command __iomem *sq_cmds_io; |
b60503ba | 116 | volatile struct nvme_completion *cqes; |
42483228 | 117 | struct blk_mq_tags **tags; |
b60503ba MW |
118 | dma_addr_t sq_dma_addr; |
119 | dma_addr_t cq_dma_addr; | |
b60503ba MW |
120 | u32 __iomem *q_db; |
121 | u16 q_depth; | |
6222d172 | 122 | s16 cq_vector; |
b60503ba MW |
123 | u16 sq_head; |
124 | u16 sq_tail; | |
125 | u16 cq_head; | |
c30341dc | 126 | u16 qid; |
e9539f47 MW |
127 | u8 cq_phase; |
128 | u8 cqe_seen; | |
4d115420 | 129 | struct async_cmd_info cmdinfo; |
b60503ba MW |
130 | }; |
131 | ||
132 | /* | |
133 | * Check we didin't inadvertently grow the command struct | |
134 | */ | |
135 | static inline void _nvme_check_size(void) | |
136 | { | |
137 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
138 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); | |
139 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
140 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
141 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
f8ebf840 | 142 | BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); |
c30341dc | 143 | BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); |
b60503ba MW |
144 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); |
145 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); | |
146 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); | |
147 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
6ecec745 | 148 | BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); |
b60503ba MW |
149 | } |
150 | ||
edd10d33 | 151 | typedef void (*nvme_completion_fn)(struct nvme_queue *, void *, |
c2f5b650 MW |
152 | struct nvme_completion *); |
153 | ||
e85248e5 | 154 | struct nvme_cmd_info { |
c2f5b650 MW |
155 | nvme_completion_fn fn; |
156 | void *ctx; | |
c30341dc | 157 | int aborted; |
a4aea562 | 158 | struct nvme_queue *nvmeq; |
ac3dd5bd | 159 | struct nvme_iod iod[0]; |
e85248e5 MW |
160 | }; |
161 | ||
ac3dd5bd JA |
162 | /* |
163 | * Max size of iod being embedded in the request payload | |
164 | */ | |
165 | #define NVME_INT_PAGES 2 | |
166 | #define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->page_size) | |
fda631ff | 167 | #define NVME_INT_MASK 0x01 |
ac3dd5bd JA |
168 | |
169 | /* | |
170 | * Will slightly overestimate the number of pages needed. This is OK | |
171 | * as it only leads to a small amount of wasted memory for the lifetime of | |
172 | * the I/O. | |
173 | */ | |
174 | static int nvme_npages(unsigned size, struct nvme_dev *dev) | |
175 | { | |
176 | unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size); | |
177 | return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); | |
178 | } | |
179 | ||
180 | static unsigned int nvme_cmd_size(struct nvme_dev *dev) | |
181 | { | |
182 | unsigned int ret = sizeof(struct nvme_cmd_info); | |
183 | ||
184 | ret += sizeof(struct nvme_iod); | |
185 | ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev); | |
186 | ret += sizeof(struct scatterlist) * NVME_INT_PAGES; | |
187 | ||
188 | return ret; | |
189 | } | |
190 | ||
a4aea562 MB |
191 | static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
192 | unsigned int hctx_idx) | |
e85248e5 | 193 | { |
a4aea562 MB |
194 | struct nvme_dev *dev = data; |
195 | struct nvme_queue *nvmeq = dev->queues[0]; | |
196 | ||
42483228 KB |
197 | WARN_ON(hctx_idx != 0); |
198 | WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); | |
199 | WARN_ON(nvmeq->tags); | |
200 | ||
a4aea562 | 201 | hctx->driver_data = nvmeq; |
42483228 | 202 | nvmeq->tags = &dev->admin_tagset.tags[0]; |
a4aea562 | 203 | return 0; |
e85248e5 MW |
204 | } |
205 | ||
4af0e21c KB |
206 | static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) |
207 | { | |
208 | struct nvme_queue *nvmeq = hctx->driver_data; | |
209 | ||
210 | nvmeq->tags = NULL; | |
211 | } | |
212 | ||
a4aea562 MB |
213 | static int nvme_admin_init_request(void *data, struct request *req, |
214 | unsigned int hctx_idx, unsigned int rq_idx, | |
215 | unsigned int numa_node) | |
22404274 | 216 | { |
a4aea562 MB |
217 | struct nvme_dev *dev = data; |
218 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
219 | struct nvme_queue *nvmeq = dev->queues[0]; | |
220 | ||
221 | BUG_ON(!nvmeq); | |
222 | cmd->nvmeq = nvmeq; | |
223 | return 0; | |
22404274 KB |
224 | } |
225 | ||
a4aea562 MB |
226 | static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
227 | unsigned int hctx_idx) | |
b60503ba | 228 | { |
a4aea562 | 229 | struct nvme_dev *dev = data; |
42483228 | 230 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; |
a4aea562 | 231 | |
42483228 KB |
232 | if (!nvmeq->tags) |
233 | nvmeq->tags = &dev->tagset.tags[hctx_idx]; | |
b60503ba | 234 | |
42483228 | 235 | WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); |
a4aea562 MB |
236 | hctx->driver_data = nvmeq; |
237 | return 0; | |
b60503ba MW |
238 | } |
239 | ||
a4aea562 MB |
240 | static int nvme_init_request(void *data, struct request *req, |
241 | unsigned int hctx_idx, unsigned int rq_idx, | |
242 | unsigned int numa_node) | |
b60503ba | 243 | { |
a4aea562 MB |
244 | struct nvme_dev *dev = data; |
245 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
246 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; | |
247 | ||
248 | BUG_ON(!nvmeq); | |
249 | cmd->nvmeq = nvmeq; | |
250 | return 0; | |
251 | } | |
252 | ||
253 | static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx, | |
254 | nvme_completion_fn handler) | |
255 | { | |
256 | cmd->fn = handler; | |
257 | cmd->ctx = ctx; | |
258 | cmd->aborted = 0; | |
c917dfe5 | 259 | blk_mq_start_request(blk_mq_rq_from_pdu(cmd)); |
b60503ba MW |
260 | } |
261 | ||
ac3dd5bd JA |
262 | static void *iod_get_private(struct nvme_iod *iod) |
263 | { | |
264 | return (void *) (iod->private & ~0x1UL); | |
265 | } | |
266 | ||
267 | /* | |
268 | * If bit 0 is set, the iod is embedded in the request payload. | |
269 | */ | |
270 | static bool iod_should_kfree(struct nvme_iod *iod) | |
271 | { | |
fda631ff | 272 | return (iod->private & NVME_INT_MASK) == 0; |
ac3dd5bd JA |
273 | } |
274 | ||
c2f5b650 MW |
275 | /* Special values must be less than 0x1000 */ |
276 | #define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA) | |
d2d87034 MW |
277 | #define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE) |
278 | #define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE) | |
279 | #define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE) | |
be7b6275 | 280 | |
edd10d33 | 281 | static void special_completion(struct nvme_queue *nvmeq, void *ctx, |
c2f5b650 MW |
282 | struct nvme_completion *cqe) |
283 | { | |
284 | if (ctx == CMD_CTX_CANCELLED) | |
285 | return; | |
c2f5b650 | 286 | if (ctx == CMD_CTX_COMPLETED) { |
edd10d33 | 287 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
288 | "completed id %d twice on queue %d\n", |
289 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
290 | return; | |
291 | } | |
292 | if (ctx == CMD_CTX_INVALID) { | |
edd10d33 | 293 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
294 | "invalid id %d completed on queue %d\n", |
295 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
296 | return; | |
297 | } | |
edd10d33 | 298 | dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx); |
c2f5b650 MW |
299 | } |
300 | ||
a4aea562 | 301 | static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn) |
b60503ba | 302 | { |
c2f5b650 | 303 | void *ctx; |
b60503ba | 304 | |
859361a2 | 305 | if (fn) |
a4aea562 MB |
306 | *fn = cmd->fn; |
307 | ctx = cmd->ctx; | |
308 | cmd->fn = special_completion; | |
309 | cmd->ctx = CMD_CTX_CANCELLED; | |
c2f5b650 | 310 | return ctx; |
b60503ba MW |
311 | } |
312 | ||
a4aea562 MB |
313 | static void async_req_completion(struct nvme_queue *nvmeq, void *ctx, |
314 | struct nvme_completion *cqe) | |
3c0cf138 | 315 | { |
a4aea562 MB |
316 | u32 result = le32_to_cpup(&cqe->result); |
317 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
318 | ||
319 | if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) | |
320 | ++nvmeq->dev->event_limit; | |
a5768aa8 KB |
321 | if (status != NVME_SC_SUCCESS) |
322 | return; | |
323 | ||
324 | switch (result & 0xff07) { | |
325 | case NVME_AER_NOTICE_NS_CHANGED: | |
326 | dev_info(nvmeq->q_dmadev, "rescanning\n"); | |
327 | schedule_work(&nvmeq->dev->scan_work); | |
328 | default: | |
329 | dev_warn(nvmeq->q_dmadev, "async event result %08x\n", result); | |
330 | } | |
b60503ba MW |
331 | } |
332 | ||
a4aea562 MB |
333 | static void abort_completion(struct nvme_queue *nvmeq, void *ctx, |
334 | struct nvme_completion *cqe) | |
5a92e700 | 335 | { |
a4aea562 MB |
336 | struct request *req = ctx; |
337 | ||
338 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
339 | u32 result = le32_to_cpup(&cqe->result); | |
a51afb54 | 340 | |
42483228 | 341 | blk_mq_free_request(req); |
a51afb54 | 342 | |
a4aea562 MB |
343 | dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result); |
344 | ++nvmeq->dev->abort_limit; | |
5a92e700 KB |
345 | } |
346 | ||
a4aea562 MB |
347 | static void async_completion(struct nvme_queue *nvmeq, void *ctx, |
348 | struct nvme_completion *cqe) | |
b60503ba | 349 | { |
a4aea562 MB |
350 | struct async_cmd_info *cmdinfo = ctx; |
351 | cmdinfo->result = le32_to_cpup(&cqe->result); | |
352 | cmdinfo->status = le16_to_cpup(&cqe->status) >> 1; | |
353 | queue_kthread_work(cmdinfo->worker, &cmdinfo->work); | |
42483228 | 354 | blk_mq_free_request(cmdinfo->req); |
b60503ba MW |
355 | } |
356 | ||
a4aea562 MB |
357 | static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq, |
358 | unsigned int tag) | |
b60503ba | 359 | { |
42483228 | 360 | struct request *req = blk_mq_tag_to_rq(*nvmeq->tags, tag); |
a51afb54 | 361 | |
a4aea562 | 362 | return blk_mq_rq_to_pdu(req); |
4f5099af KB |
363 | } |
364 | ||
a4aea562 MB |
365 | /* |
366 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
367 | */ | |
368 | static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag, | |
369 | nvme_completion_fn *fn) | |
4f5099af | 370 | { |
a4aea562 MB |
371 | struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag); |
372 | void *ctx; | |
373 | if (tag >= nvmeq->q_depth) { | |
374 | *fn = special_completion; | |
375 | return CMD_CTX_INVALID; | |
376 | } | |
377 | if (fn) | |
378 | *fn = cmd->fn; | |
379 | ctx = cmd->ctx; | |
380 | cmd->fn = special_completion; | |
381 | cmd->ctx = CMD_CTX_COMPLETED; | |
382 | return ctx; | |
b60503ba MW |
383 | } |
384 | ||
385 | /** | |
714a7a22 | 386 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
387 | * @nvmeq: The queue to use |
388 | * @cmd: The command to send | |
389 | * | |
390 | * Safe to use from interrupt context | |
391 | */ | |
e3f879bf SB |
392 | static void __nvme_submit_cmd(struct nvme_queue *nvmeq, |
393 | struct nvme_command *cmd) | |
b60503ba | 394 | { |
a4aea562 MB |
395 | u16 tail = nvmeq->sq_tail; |
396 | ||
8ffaadf7 JD |
397 | if (nvmeq->sq_cmds_io) |
398 | memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd)); | |
399 | else | |
400 | memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); | |
401 | ||
b60503ba MW |
402 | if (++tail == nvmeq->q_depth) |
403 | tail = 0; | |
7547881d | 404 | writel(tail, nvmeq->q_db); |
b60503ba | 405 | nvmeq->sq_tail = tail; |
b60503ba MW |
406 | } |
407 | ||
e3f879bf | 408 | static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd) |
a4aea562 MB |
409 | { |
410 | unsigned long flags; | |
a4aea562 | 411 | spin_lock_irqsave(&nvmeq->q_lock, flags); |
e3f879bf | 412 | __nvme_submit_cmd(nvmeq, cmd); |
a4aea562 | 413 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); |
a4aea562 MB |
414 | } |
415 | ||
eca18b23 | 416 | static __le64 **iod_list(struct nvme_iod *iod) |
e025344c | 417 | { |
eca18b23 | 418 | return ((void *)iod) + iod->offset; |
e025344c SMM |
419 | } |
420 | ||
ac3dd5bd JA |
421 | static inline void iod_init(struct nvme_iod *iod, unsigned nbytes, |
422 | unsigned nseg, unsigned long private) | |
eca18b23 | 423 | { |
ac3dd5bd JA |
424 | iod->private = private; |
425 | iod->offset = offsetof(struct nvme_iod, sg[nseg]); | |
426 | iod->npages = -1; | |
427 | iod->length = nbytes; | |
428 | iod->nents = 0; | |
eca18b23 | 429 | } |
b60503ba | 430 | |
eca18b23 | 431 | static struct nvme_iod * |
ac3dd5bd JA |
432 | __nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev, |
433 | unsigned long priv, gfp_t gfp) | |
b60503ba | 434 | { |
eca18b23 | 435 | struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) + |
ac3dd5bd | 436 | sizeof(__le64 *) * nvme_npages(bytes, dev) + |
eca18b23 MW |
437 | sizeof(struct scatterlist) * nseg, gfp); |
438 | ||
ac3dd5bd JA |
439 | if (iod) |
440 | iod_init(iod, bytes, nseg, priv); | |
eca18b23 MW |
441 | |
442 | return iod; | |
b60503ba MW |
443 | } |
444 | ||
ac3dd5bd JA |
445 | static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev, |
446 | gfp_t gfp) | |
447 | { | |
448 | unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) : | |
449 | sizeof(struct nvme_dsm_range); | |
ac3dd5bd JA |
450 | struct nvme_iod *iod; |
451 | ||
452 | if (rq->nr_phys_segments <= NVME_INT_PAGES && | |
453 | size <= NVME_INT_BYTES(dev)) { | |
454 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq); | |
455 | ||
456 | iod = cmd->iod; | |
ac3dd5bd | 457 | iod_init(iod, size, rq->nr_phys_segments, |
fda631ff | 458 | (unsigned long) rq | NVME_INT_MASK); |
ac3dd5bd JA |
459 | return iod; |
460 | } | |
461 | ||
462 | return __nvme_alloc_iod(rq->nr_phys_segments, size, dev, | |
463 | (unsigned long) rq, gfp); | |
464 | } | |
465 | ||
d29ec824 | 466 | static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod) |
b60503ba | 467 | { |
1d090624 | 468 | const int last_prp = dev->page_size / 8 - 1; |
eca18b23 MW |
469 | int i; |
470 | __le64 **list = iod_list(iod); | |
471 | dma_addr_t prp_dma = iod->first_dma; | |
472 | ||
473 | if (iod->npages == 0) | |
474 | dma_pool_free(dev->prp_small_pool, list[0], prp_dma); | |
475 | for (i = 0; i < iod->npages; i++) { | |
476 | __le64 *prp_list = list[i]; | |
477 | dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); | |
478 | dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); | |
479 | prp_dma = next_prp_dma; | |
480 | } | |
ac3dd5bd JA |
481 | |
482 | if (iod_should_kfree(iod)) | |
483 | kfree(iod); | |
b60503ba MW |
484 | } |
485 | ||
b4ff9c8d KB |
486 | static int nvme_error_status(u16 status) |
487 | { | |
488 | switch (status & 0x7ff) { | |
489 | case NVME_SC_SUCCESS: | |
490 | return 0; | |
491 | case NVME_SC_CAP_EXCEEDED: | |
492 | return -ENOSPC; | |
493 | default: | |
494 | return -EIO; | |
495 | } | |
496 | } | |
497 | ||
52b68d7e | 498 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
e1e5e564 KB |
499 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) |
500 | { | |
501 | if (be32_to_cpu(pi->ref_tag) == v) | |
502 | pi->ref_tag = cpu_to_be32(p); | |
503 | } | |
504 | ||
505 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
506 | { | |
507 | if (be32_to_cpu(pi->ref_tag) == p) | |
508 | pi->ref_tag = cpu_to_be32(v); | |
509 | } | |
510 | ||
511 | /** | |
512 | * nvme_dif_remap - remaps ref tags to bip seed and physical lba | |
513 | * | |
514 | * The virtual start sector is the one that was originally submitted by the | |
515 | * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical | |
516 | * start sector may be different. Remap protection information to match the | |
517 | * physical LBA on writes, and back to the original seed on reads. | |
518 | * | |
519 | * Type 0 and 3 do not have a ref tag, so no remapping required. | |
520 | */ | |
521 | static void nvme_dif_remap(struct request *req, | |
522 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
523 | { | |
524 | struct nvme_ns *ns = req->rq_disk->private_data; | |
525 | struct bio_integrity_payload *bip; | |
526 | struct t10_pi_tuple *pi; | |
527 | void *p, *pmap; | |
528 | u32 i, nlb, ts, phys, virt; | |
529 | ||
530 | if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3) | |
531 | return; | |
532 | ||
533 | bip = bio_integrity(req->bio); | |
534 | if (!bip) | |
535 | return; | |
536 | ||
537 | pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset; | |
e1e5e564 KB |
538 | |
539 | p = pmap; | |
540 | virt = bip_get_seed(bip); | |
541 | phys = nvme_block_nr(ns, blk_rq_pos(req)); | |
542 | nlb = (blk_rq_bytes(req) >> ns->lba_shift); | |
ac6fc48c | 543 | ts = ns->disk->queue->integrity.tuple_size; |
e1e5e564 KB |
544 | |
545 | for (i = 0; i < nlb; i++, virt++, phys++) { | |
546 | pi = (struct t10_pi_tuple *)p; | |
547 | dif_swap(phys, virt, pi); | |
548 | p += ts; | |
549 | } | |
550 | kunmap_atomic(pmap); | |
551 | } | |
552 | ||
52b68d7e KB |
553 | static void nvme_init_integrity(struct nvme_ns *ns) |
554 | { | |
555 | struct blk_integrity integrity; | |
556 | ||
557 | switch (ns->pi_type) { | |
558 | case NVME_NS_DPS_PI_TYPE3: | |
0f8087ec | 559 | integrity.profile = &t10_pi_type3_crc; |
52b68d7e KB |
560 | break; |
561 | case NVME_NS_DPS_PI_TYPE1: | |
562 | case NVME_NS_DPS_PI_TYPE2: | |
0f8087ec | 563 | integrity.profile = &t10_pi_type1_crc; |
52b68d7e KB |
564 | break; |
565 | default: | |
4125a09b | 566 | integrity.profile = NULL; |
52b68d7e KB |
567 | break; |
568 | } | |
569 | integrity.tuple_size = ns->ms; | |
570 | blk_integrity_register(ns->disk, &integrity); | |
571 | blk_queue_max_integrity_segments(ns->queue, 1); | |
572 | } | |
573 | #else /* CONFIG_BLK_DEV_INTEGRITY */ | |
574 | static void nvme_dif_remap(struct request *req, | |
575 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
576 | { | |
577 | } | |
578 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) | |
579 | { | |
580 | } | |
581 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
582 | { | |
583 | } | |
584 | static void nvme_init_integrity(struct nvme_ns *ns) | |
585 | { | |
586 | } | |
587 | #endif | |
588 | ||
a4aea562 | 589 | static void req_completion(struct nvme_queue *nvmeq, void *ctx, |
b60503ba MW |
590 | struct nvme_completion *cqe) |
591 | { | |
eca18b23 | 592 | struct nvme_iod *iod = ctx; |
ac3dd5bd | 593 | struct request *req = iod_get_private(iod); |
a4aea562 | 594 | struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req); |
b60503ba | 595 | u16 status = le16_to_cpup(&cqe->status) >> 1; |
0dfc70c3 | 596 | bool requeue = false; |
81c04b94 | 597 | int error = 0; |
b60503ba | 598 | |
edd10d33 | 599 | if (unlikely(status)) { |
a4aea562 MB |
600 | if (!(status & NVME_SC_DNR || blk_noretry_request(req)) |
601 | && (jiffies - req->start_time) < req->timeout) { | |
c9d3bf88 KB |
602 | unsigned long flags; |
603 | ||
0dfc70c3 | 604 | requeue = true; |
a4aea562 | 605 | blk_mq_requeue_request(req); |
c9d3bf88 KB |
606 | spin_lock_irqsave(req->q->queue_lock, flags); |
607 | if (!blk_queue_stopped(req->q)) | |
608 | blk_mq_kick_requeue_list(req->q); | |
609 | spin_unlock_irqrestore(req->q->queue_lock, flags); | |
0dfc70c3 | 610 | goto release_iod; |
edd10d33 | 611 | } |
f4829a9b | 612 | |
d29ec824 | 613 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) { |
17188bb4 | 614 | if (cmd_rq->ctx == CMD_CTX_CANCELLED) |
81c04b94 CH |
615 | error = -EINTR; |
616 | else | |
617 | error = status; | |
d29ec824 | 618 | } else { |
81c04b94 | 619 | error = nvme_error_status(status); |
d29ec824 | 620 | } |
f4829a9b CH |
621 | } |
622 | ||
a0a931d6 KB |
623 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) { |
624 | u32 result = le32_to_cpup(&cqe->result); | |
625 | req->special = (void *)(uintptr_t)result; | |
626 | } | |
a4aea562 MB |
627 | |
628 | if (cmd_rq->aborted) | |
e75ec752 | 629 | dev_warn(nvmeq->dev->dev, |
a4aea562 | 630 | "completing aborted command with status:%04x\n", |
81c04b94 | 631 | error); |
a4aea562 | 632 | |
0dfc70c3 | 633 | release_iod: |
e1e5e564 | 634 | if (iod->nents) { |
e75ec752 | 635 | dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents, |
a4aea562 | 636 | rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
e1e5e564 KB |
637 | if (blk_integrity_rq(req)) { |
638 | if (!rq_data_dir(req)) | |
639 | nvme_dif_remap(req, nvme_dif_complete); | |
e75ec752 | 640 | dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1, |
e1e5e564 KB |
641 | rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
642 | } | |
643 | } | |
edd10d33 | 644 | nvme_free_iod(nvmeq->dev, iod); |
3291fa57 | 645 | |
0dfc70c3 KB |
646 | if (likely(!requeue)) |
647 | blk_mq_complete_request(req, error); | |
b60503ba MW |
648 | } |
649 | ||
184d2944 | 650 | /* length is in bytes. gfp flags indicates whether we may sleep. */ |
d29ec824 CH |
651 | static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod, |
652 | int total_len, gfp_t gfp) | |
ff22b54f | 653 | { |
99802a7a | 654 | struct dma_pool *pool; |
eca18b23 MW |
655 | int length = total_len; |
656 | struct scatterlist *sg = iod->sg; | |
ff22b54f MW |
657 | int dma_len = sg_dma_len(sg); |
658 | u64 dma_addr = sg_dma_address(sg); | |
f137e0f1 MI |
659 | u32 page_size = dev->page_size; |
660 | int offset = dma_addr & (page_size - 1); | |
e025344c | 661 | __le64 *prp_list; |
eca18b23 | 662 | __le64 **list = iod_list(iod); |
e025344c | 663 | dma_addr_t prp_dma; |
eca18b23 | 664 | int nprps, i; |
ff22b54f | 665 | |
1d090624 | 666 | length -= (page_size - offset); |
ff22b54f | 667 | if (length <= 0) |
eca18b23 | 668 | return total_len; |
ff22b54f | 669 | |
1d090624 | 670 | dma_len -= (page_size - offset); |
ff22b54f | 671 | if (dma_len) { |
1d090624 | 672 | dma_addr += (page_size - offset); |
ff22b54f MW |
673 | } else { |
674 | sg = sg_next(sg); | |
675 | dma_addr = sg_dma_address(sg); | |
676 | dma_len = sg_dma_len(sg); | |
677 | } | |
678 | ||
1d090624 | 679 | if (length <= page_size) { |
edd10d33 | 680 | iod->first_dma = dma_addr; |
eca18b23 | 681 | return total_len; |
e025344c SMM |
682 | } |
683 | ||
1d090624 | 684 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
685 | if (nprps <= (256 / 8)) { |
686 | pool = dev->prp_small_pool; | |
eca18b23 | 687 | iod->npages = 0; |
99802a7a MW |
688 | } else { |
689 | pool = dev->prp_page_pool; | |
eca18b23 | 690 | iod->npages = 1; |
99802a7a MW |
691 | } |
692 | ||
b77954cb MW |
693 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
694 | if (!prp_list) { | |
edd10d33 | 695 | iod->first_dma = dma_addr; |
eca18b23 | 696 | iod->npages = -1; |
1d090624 | 697 | return (total_len - length) + page_size; |
b77954cb | 698 | } |
eca18b23 MW |
699 | list[0] = prp_list; |
700 | iod->first_dma = prp_dma; | |
e025344c SMM |
701 | i = 0; |
702 | for (;;) { | |
1d090624 | 703 | if (i == page_size >> 3) { |
e025344c | 704 | __le64 *old_prp_list = prp_list; |
b77954cb | 705 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
eca18b23 MW |
706 | if (!prp_list) |
707 | return total_len - length; | |
708 | list[iod->npages++] = prp_list; | |
7523d834 MW |
709 | prp_list[0] = old_prp_list[i - 1]; |
710 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
711 | i = 1; | |
e025344c SMM |
712 | } |
713 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
714 | dma_len -= page_size; |
715 | dma_addr += page_size; | |
716 | length -= page_size; | |
e025344c SMM |
717 | if (length <= 0) |
718 | break; | |
719 | if (dma_len > 0) | |
720 | continue; | |
721 | BUG_ON(dma_len < 0); | |
722 | sg = sg_next(sg); | |
723 | dma_addr = sg_dma_address(sg); | |
724 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
725 | } |
726 | ||
eca18b23 | 727 | return total_len; |
ff22b54f MW |
728 | } |
729 | ||
d29ec824 CH |
730 | static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req, |
731 | struct nvme_iod *iod) | |
732 | { | |
498c4394 | 733 | struct nvme_command cmnd; |
d29ec824 | 734 | |
498c4394 JD |
735 | memcpy(&cmnd, req->cmd, sizeof(cmnd)); |
736 | cmnd.rw.command_id = req->tag; | |
d29ec824 | 737 | if (req->nr_phys_segments) { |
498c4394 JD |
738 | cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); |
739 | cmnd.rw.prp2 = cpu_to_le64(iod->first_dma); | |
d29ec824 CH |
740 | } |
741 | ||
498c4394 | 742 | __nvme_submit_cmd(nvmeq, &cmnd); |
d29ec824 CH |
743 | } |
744 | ||
a4aea562 MB |
745 | /* |
746 | * We reuse the small pool to allocate the 16-byte range here as it is not | |
747 | * worth having a special pool for these or additional cases to handle freeing | |
748 | * the iod. | |
749 | */ | |
750 | static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns, | |
751 | struct request *req, struct nvme_iod *iod) | |
0e5e4f0e | 752 | { |
edd10d33 KB |
753 | struct nvme_dsm_range *range = |
754 | (struct nvme_dsm_range *)iod_list(iod)[0]; | |
498c4394 | 755 | struct nvme_command cmnd; |
0e5e4f0e | 756 | |
0e5e4f0e | 757 | range->cattr = cpu_to_le32(0); |
a4aea562 MB |
758 | range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift); |
759 | range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
0e5e4f0e | 760 | |
498c4394 JD |
761 | memset(&cmnd, 0, sizeof(cmnd)); |
762 | cmnd.dsm.opcode = nvme_cmd_dsm; | |
763 | cmnd.dsm.command_id = req->tag; | |
764 | cmnd.dsm.nsid = cpu_to_le32(ns->ns_id); | |
765 | cmnd.dsm.prp1 = cpu_to_le64(iod->first_dma); | |
766 | cmnd.dsm.nr = 0; | |
767 | cmnd.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); | |
0e5e4f0e | 768 | |
498c4394 | 769 | __nvme_submit_cmd(nvmeq, &cmnd); |
0e5e4f0e KB |
770 | } |
771 | ||
a4aea562 | 772 | static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
00df5cb4 MW |
773 | int cmdid) |
774 | { | |
498c4394 | 775 | struct nvme_command cmnd; |
00df5cb4 | 776 | |
498c4394 JD |
777 | memset(&cmnd, 0, sizeof(cmnd)); |
778 | cmnd.common.opcode = nvme_cmd_flush; | |
779 | cmnd.common.command_id = cmdid; | |
780 | cmnd.common.nsid = cpu_to_le32(ns->ns_id); | |
00df5cb4 | 781 | |
498c4394 | 782 | __nvme_submit_cmd(nvmeq, &cmnd); |
00df5cb4 MW |
783 | } |
784 | ||
a4aea562 MB |
785 | static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod, |
786 | struct nvme_ns *ns) | |
b60503ba | 787 | { |
ac3dd5bd | 788 | struct request *req = iod_get_private(iod); |
498c4394 | 789 | struct nvme_command cmnd; |
a4aea562 MB |
790 | u16 control = 0; |
791 | u32 dsmgmt = 0; | |
00df5cb4 | 792 | |
a4aea562 | 793 | if (req->cmd_flags & REQ_FUA) |
b60503ba | 794 | control |= NVME_RW_FUA; |
a4aea562 | 795 | if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) |
b60503ba MW |
796 | control |= NVME_RW_LR; |
797 | ||
a4aea562 | 798 | if (req->cmd_flags & REQ_RAHEAD) |
b60503ba MW |
799 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; |
800 | ||
498c4394 JD |
801 | memset(&cmnd, 0, sizeof(cmnd)); |
802 | cmnd.rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); | |
803 | cmnd.rw.command_id = req->tag; | |
804 | cmnd.rw.nsid = cpu_to_le32(ns->ns_id); | |
805 | cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
806 | cmnd.rw.prp2 = cpu_to_le64(iod->first_dma); | |
807 | cmnd.rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); | |
808 | cmnd.rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); | |
b60503ba | 809 | |
e19b127f | 810 | if (ns->ms) { |
e1e5e564 KB |
811 | switch (ns->pi_type) { |
812 | case NVME_NS_DPS_PI_TYPE3: | |
813 | control |= NVME_RW_PRINFO_PRCHK_GUARD; | |
814 | break; | |
815 | case NVME_NS_DPS_PI_TYPE1: | |
816 | case NVME_NS_DPS_PI_TYPE2: | |
817 | control |= NVME_RW_PRINFO_PRCHK_GUARD | | |
818 | NVME_RW_PRINFO_PRCHK_REF; | |
498c4394 | 819 | cmnd.rw.reftag = cpu_to_le32( |
e1e5e564 KB |
820 | nvme_block_nr(ns, blk_rq_pos(req))); |
821 | break; | |
822 | } | |
e19b127f AP |
823 | if (blk_integrity_rq(req)) |
824 | cmnd.rw.metadata = | |
825 | cpu_to_le64(sg_dma_address(iod->meta_sg)); | |
826 | else | |
827 | control |= NVME_RW_PRINFO_PRACT; | |
828 | } | |
e1e5e564 | 829 | |
498c4394 JD |
830 | cmnd.rw.control = cpu_to_le16(control); |
831 | cmnd.rw.dsmgmt = cpu_to_le32(dsmgmt); | |
b60503ba | 832 | |
498c4394 | 833 | __nvme_submit_cmd(nvmeq, &cmnd); |
b60503ba | 834 | |
1974b1ae | 835 | return 0; |
edd10d33 KB |
836 | } |
837 | ||
d29ec824 CH |
838 | /* |
839 | * NOTE: ns is NULL when called on the admin queue. | |
840 | */ | |
a4aea562 MB |
841 | static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx, |
842 | const struct blk_mq_queue_data *bd) | |
edd10d33 | 843 | { |
a4aea562 MB |
844 | struct nvme_ns *ns = hctx->queue->queuedata; |
845 | struct nvme_queue *nvmeq = hctx->driver_data; | |
d29ec824 | 846 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 MB |
847 | struct request *req = bd->rq; |
848 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); | |
edd10d33 | 849 | struct nvme_iod *iod; |
a4aea562 | 850 | enum dma_data_direction dma_dir; |
edd10d33 | 851 | |
e1e5e564 KB |
852 | /* |
853 | * If formated with metadata, require the block layer provide a buffer | |
854 | * unless this namespace is formated such that the metadata can be | |
855 | * stripped/generated by the controller with PRACT=1. | |
856 | */ | |
d29ec824 | 857 | if (ns && ns->ms && !blk_integrity_rq(req)) { |
71feb364 KB |
858 | if (!(ns->pi_type && ns->ms == 8) && |
859 | req->cmd_type != REQ_TYPE_DRV_PRIV) { | |
f4829a9b | 860 | blk_mq_complete_request(req, -EFAULT); |
e1e5e564 KB |
861 | return BLK_MQ_RQ_QUEUE_OK; |
862 | } | |
863 | } | |
864 | ||
d29ec824 | 865 | iod = nvme_alloc_iod(req, dev, GFP_ATOMIC); |
edd10d33 | 866 | if (!iod) |
fe54303e | 867 | return BLK_MQ_RQ_QUEUE_BUSY; |
a4aea562 | 868 | |
a4aea562 | 869 | if (req->cmd_flags & REQ_DISCARD) { |
edd10d33 KB |
870 | void *range; |
871 | /* | |
872 | * We reuse the small pool to allocate the 16-byte range here | |
873 | * as it is not worth having a special pool for these or | |
874 | * additional cases to handle freeing the iod. | |
875 | */ | |
d29ec824 | 876 | range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC, |
edd10d33 | 877 | &iod->first_dma); |
a4aea562 | 878 | if (!range) |
fe54303e | 879 | goto retry_cmd; |
edd10d33 KB |
880 | iod_list(iod)[0] = (__le64 *)range; |
881 | iod->npages = 0; | |
ac3dd5bd | 882 | } else if (req->nr_phys_segments) { |
a4aea562 MB |
883 | dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE; |
884 | ||
ac3dd5bd | 885 | sg_init_table(iod->sg, req->nr_phys_segments); |
a4aea562 | 886 | iod->nents = blk_rq_map_sg(req->q, req, iod->sg); |
fe54303e JA |
887 | if (!iod->nents) |
888 | goto error_cmd; | |
a4aea562 MB |
889 | |
890 | if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir)) | |
fe54303e | 891 | goto retry_cmd; |
a4aea562 | 892 | |
fe54303e | 893 | if (blk_rq_bytes(req) != |
d29ec824 CH |
894 | nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) { |
895 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); | |
fe54303e JA |
896 | goto retry_cmd; |
897 | } | |
e1e5e564 KB |
898 | if (blk_integrity_rq(req)) { |
899 | if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) | |
900 | goto error_cmd; | |
901 | ||
902 | sg_init_table(iod->meta_sg, 1); | |
903 | if (blk_rq_map_integrity_sg( | |
904 | req->q, req->bio, iod->meta_sg) != 1) | |
905 | goto error_cmd; | |
906 | ||
907 | if (rq_data_dir(req)) | |
908 | nvme_dif_remap(req, nvme_dif_prep); | |
909 | ||
910 | if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir)) | |
911 | goto error_cmd; | |
912 | } | |
edd10d33 | 913 | } |
1974b1ae | 914 | |
9af8785a | 915 | nvme_set_info(cmd, iod, req_completion); |
a4aea562 | 916 | spin_lock_irq(&nvmeq->q_lock); |
d29ec824 CH |
917 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) |
918 | nvme_submit_priv(nvmeq, req, iod); | |
919 | else if (req->cmd_flags & REQ_DISCARD) | |
a4aea562 MB |
920 | nvme_submit_discard(nvmeq, ns, req, iod); |
921 | else if (req->cmd_flags & REQ_FLUSH) | |
922 | nvme_submit_flush(nvmeq, ns, req->tag); | |
923 | else | |
924 | nvme_submit_iod(nvmeq, iod, ns); | |
925 | ||
926 | nvme_process_cq(nvmeq); | |
927 | spin_unlock_irq(&nvmeq->q_lock); | |
928 | return BLK_MQ_RQ_QUEUE_OK; | |
929 | ||
fe54303e | 930 | error_cmd: |
d29ec824 | 931 | nvme_free_iod(dev, iod); |
fe54303e JA |
932 | return BLK_MQ_RQ_QUEUE_ERROR; |
933 | retry_cmd: | |
d29ec824 | 934 | nvme_free_iod(dev, iod); |
fe54303e | 935 | return BLK_MQ_RQ_QUEUE_BUSY; |
b60503ba MW |
936 | } |
937 | ||
e9539f47 | 938 | static int nvme_process_cq(struct nvme_queue *nvmeq) |
b60503ba | 939 | { |
82123460 | 940 | u16 head, phase; |
b60503ba | 941 | |
b60503ba | 942 | head = nvmeq->cq_head; |
82123460 | 943 | phase = nvmeq->cq_phase; |
b60503ba MW |
944 | |
945 | for (;;) { | |
c2f5b650 MW |
946 | void *ctx; |
947 | nvme_completion_fn fn; | |
b60503ba | 948 | struct nvme_completion cqe = nvmeq->cqes[head]; |
82123460 | 949 | if ((le16_to_cpu(cqe.status) & 1) != phase) |
b60503ba MW |
950 | break; |
951 | nvmeq->sq_head = le16_to_cpu(cqe.sq_head); | |
952 | if (++head == nvmeq->q_depth) { | |
953 | head = 0; | |
82123460 | 954 | phase = !phase; |
b60503ba | 955 | } |
a4aea562 | 956 | ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn); |
edd10d33 | 957 | fn(nvmeq, ctx, &cqe); |
b60503ba MW |
958 | } |
959 | ||
960 | /* If the controller ignores the cq head doorbell and continuously | |
961 | * writes to the queue, it is theoretically possible to wrap around | |
962 | * the queue twice and mistakenly return IRQ_NONE. Linux only | |
963 | * requires that 0.1% of your interrupts are handled, so this isn't | |
964 | * a big problem. | |
965 | */ | |
82123460 | 966 | if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) |
e9539f47 | 967 | return 0; |
b60503ba | 968 | |
b80d5ccc | 969 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); |
b60503ba | 970 | nvmeq->cq_head = head; |
82123460 | 971 | nvmeq->cq_phase = phase; |
b60503ba | 972 | |
e9539f47 MW |
973 | nvmeq->cqe_seen = 1; |
974 | return 1; | |
b60503ba MW |
975 | } |
976 | ||
977 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 MW |
978 | { |
979 | irqreturn_t result; | |
980 | struct nvme_queue *nvmeq = data; | |
981 | spin_lock(&nvmeq->q_lock); | |
e9539f47 MW |
982 | nvme_process_cq(nvmeq); |
983 | result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; | |
984 | nvmeq->cqe_seen = 0; | |
58ffacb5 MW |
985 | spin_unlock(&nvmeq->q_lock); |
986 | return result; | |
987 | } | |
988 | ||
989 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
990 | { | |
991 | struct nvme_queue *nvmeq = data; | |
992 | struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head]; | |
993 | if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase) | |
994 | return IRQ_NONE; | |
995 | return IRQ_WAKE_THREAD; | |
996 | } | |
997 | ||
b60503ba MW |
998 | /* |
999 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
1000 | * if the result is positive, it's an NVM Express status code | |
1001 | */ | |
d29ec824 CH |
1002 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
1003 | void *buffer, void __user *ubuffer, unsigned bufflen, | |
1004 | u32 *result, unsigned timeout) | |
b60503ba | 1005 | { |
d29ec824 CH |
1006 | bool write = cmd->common.opcode & 1; |
1007 | struct bio *bio = NULL; | |
f705f837 | 1008 | struct request *req; |
d29ec824 | 1009 | int ret; |
b60503ba | 1010 | |
d29ec824 | 1011 | req = blk_mq_alloc_request(q, write, GFP_KERNEL, false); |
f705f837 CH |
1012 | if (IS_ERR(req)) |
1013 | return PTR_ERR(req); | |
b60503ba | 1014 | |
d29ec824 | 1015 | req->cmd_type = REQ_TYPE_DRV_PRIV; |
e112af0d | 1016 | req->cmd_flags |= REQ_FAILFAST_DRIVER; |
d29ec824 CH |
1017 | req->__data_len = 0; |
1018 | req->__sector = (sector_t) -1; | |
1019 | req->bio = req->biotail = NULL; | |
b60503ba | 1020 | |
f4ff414a | 1021 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; |
a4aea562 | 1022 | |
d29ec824 CH |
1023 | req->cmd = (unsigned char *)cmd; |
1024 | req->cmd_len = sizeof(struct nvme_command); | |
a0a931d6 | 1025 | req->special = (void *)0; |
b60503ba | 1026 | |
d29ec824 | 1027 | if (buffer && bufflen) { |
71baba4b MG |
1028 | ret = blk_rq_map_kern(q, req, buffer, bufflen, |
1029 | __GFP_DIRECT_RECLAIM); | |
d29ec824 CH |
1030 | if (ret) |
1031 | goto out; | |
1032 | } else if (ubuffer && bufflen) { | |
71baba4b MG |
1033 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, |
1034 | __GFP_DIRECT_RECLAIM); | |
d29ec824 CH |
1035 | if (ret) |
1036 | goto out; | |
1037 | bio = req->bio; | |
1038 | } | |
3c0cf138 | 1039 | |
d29ec824 CH |
1040 | blk_execute_rq(req->q, NULL, req, 0); |
1041 | if (bio) | |
1042 | blk_rq_unmap_user(bio); | |
b60503ba | 1043 | if (result) |
a0a931d6 | 1044 | *result = (u32)(uintptr_t)req->special; |
d29ec824 CH |
1045 | ret = req->errors; |
1046 | out: | |
f705f837 | 1047 | blk_mq_free_request(req); |
d29ec824 | 1048 | return ret; |
f705f837 CH |
1049 | } |
1050 | ||
d29ec824 CH |
1051 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
1052 | void *buffer, unsigned bufflen) | |
f705f837 | 1053 | { |
d29ec824 | 1054 | return __nvme_submit_sync_cmd(q, cmd, buffer, NULL, bufflen, NULL, 0); |
b60503ba MW |
1055 | } |
1056 | ||
a4aea562 MB |
1057 | static int nvme_submit_async_admin_req(struct nvme_dev *dev) |
1058 | { | |
1059 | struct nvme_queue *nvmeq = dev->queues[0]; | |
1060 | struct nvme_command c; | |
1061 | struct nvme_cmd_info *cmd_info; | |
1062 | struct request *req; | |
1063 | ||
1efccc9d | 1064 | req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, true); |
9f173b33 DC |
1065 | if (IS_ERR(req)) |
1066 | return PTR_ERR(req); | |
a4aea562 | 1067 | |
c917dfe5 | 1068 | req->cmd_flags |= REQ_NO_TIMEOUT; |
a4aea562 | 1069 | cmd_info = blk_mq_rq_to_pdu(req); |
1efccc9d | 1070 | nvme_set_info(cmd_info, NULL, async_req_completion); |
a4aea562 MB |
1071 | |
1072 | memset(&c, 0, sizeof(c)); | |
1073 | c.common.opcode = nvme_admin_async_event; | |
1074 | c.common.command_id = req->tag; | |
1075 | ||
42483228 | 1076 | blk_mq_free_request(req); |
e3f879bf SB |
1077 | __nvme_submit_cmd(nvmeq, &c); |
1078 | return 0; | |
a4aea562 MB |
1079 | } |
1080 | ||
1081 | static int nvme_submit_admin_async_cmd(struct nvme_dev *dev, | |
4d115420 KB |
1082 | struct nvme_command *cmd, |
1083 | struct async_cmd_info *cmdinfo, unsigned timeout) | |
1084 | { | |
a4aea562 MB |
1085 | struct nvme_queue *nvmeq = dev->queues[0]; |
1086 | struct request *req; | |
1087 | struct nvme_cmd_info *cmd_rq; | |
4d115420 | 1088 | |
a4aea562 | 1089 | req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_KERNEL, false); |
9f173b33 DC |
1090 | if (IS_ERR(req)) |
1091 | return PTR_ERR(req); | |
a4aea562 MB |
1092 | |
1093 | req->timeout = timeout; | |
1094 | cmd_rq = blk_mq_rq_to_pdu(req); | |
1095 | cmdinfo->req = req; | |
1096 | nvme_set_info(cmd_rq, cmdinfo, async_completion); | |
4d115420 | 1097 | cmdinfo->status = -EINTR; |
a4aea562 MB |
1098 | |
1099 | cmd->common.command_id = req->tag; | |
1100 | ||
e3f879bf SB |
1101 | nvme_submit_cmd(nvmeq, cmd); |
1102 | return 0; | |
4d115420 KB |
1103 | } |
1104 | ||
b60503ba MW |
1105 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
1106 | { | |
b60503ba MW |
1107 | struct nvme_command c; |
1108 | ||
1109 | memset(&c, 0, sizeof(c)); | |
1110 | c.delete_queue.opcode = opcode; | |
1111 | c.delete_queue.qid = cpu_to_le16(id); | |
1112 | ||
d29ec824 | 1113 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1114 | } |
1115 | ||
1116 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, | |
1117 | struct nvme_queue *nvmeq) | |
1118 | { | |
b60503ba MW |
1119 | struct nvme_command c; |
1120 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; | |
1121 | ||
d29ec824 CH |
1122 | /* |
1123 | * Note: we (ab)use the fact the the prp fields survive if no data | |
1124 | * is attached to the request. | |
1125 | */ | |
b60503ba MW |
1126 | memset(&c, 0, sizeof(c)); |
1127 | c.create_cq.opcode = nvme_admin_create_cq; | |
1128 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
1129 | c.create_cq.cqid = cpu_to_le16(qid); | |
1130 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1131 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
1132 | c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); | |
1133 | ||
d29ec824 | 1134 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1135 | } |
1136 | ||
1137 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
1138 | struct nvme_queue *nvmeq) | |
1139 | { | |
b60503ba MW |
1140 | struct nvme_command c; |
1141 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; | |
1142 | ||
d29ec824 CH |
1143 | /* |
1144 | * Note: we (ab)use the fact the the prp fields survive if no data | |
1145 | * is attached to the request. | |
1146 | */ | |
b60503ba MW |
1147 | memset(&c, 0, sizeof(c)); |
1148 | c.create_sq.opcode = nvme_admin_create_sq; | |
1149 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
1150 | c.create_sq.sqid = cpu_to_le16(qid); | |
1151 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1152 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
1153 | c.create_sq.cqid = cpu_to_le16(qid); | |
1154 | ||
d29ec824 | 1155 | return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0); |
b60503ba MW |
1156 | } |
1157 | ||
1158 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
1159 | { | |
1160 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
1161 | } | |
1162 | ||
1163 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
1164 | { | |
1165 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
1166 | } | |
1167 | ||
d29ec824 | 1168 | int nvme_identify_ctrl(struct nvme_dev *dev, struct nvme_id_ctrl **id) |
bc5fc7e4 | 1169 | { |
e44ac588 | 1170 | struct nvme_command c = { }; |
d29ec824 | 1171 | int error; |
bc5fc7e4 | 1172 | |
e44ac588 AM |
1173 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ |
1174 | c.identify.opcode = nvme_admin_identify; | |
1175 | c.identify.cns = cpu_to_le32(1); | |
1176 | ||
d29ec824 CH |
1177 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); |
1178 | if (!*id) | |
1179 | return -ENOMEM; | |
bc5fc7e4 | 1180 | |
d29ec824 CH |
1181 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, |
1182 | sizeof(struct nvme_id_ctrl)); | |
1183 | if (error) | |
1184 | kfree(*id); | |
1185 | return error; | |
1186 | } | |
1187 | ||
1188 | int nvme_identify_ns(struct nvme_dev *dev, unsigned nsid, | |
1189 | struct nvme_id_ns **id) | |
1190 | { | |
e44ac588 | 1191 | struct nvme_command c = { }; |
d29ec824 | 1192 | int error; |
bc5fc7e4 | 1193 | |
e44ac588 AM |
1194 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ |
1195 | c.identify.opcode = nvme_admin_identify, | |
1196 | c.identify.nsid = cpu_to_le32(nsid), | |
1197 | ||
d29ec824 CH |
1198 | *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL); |
1199 | if (!*id) | |
1200 | return -ENOMEM; | |
1201 | ||
1202 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
1203 | sizeof(struct nvme_id_ns)); | |
1204 | if (error) | |
1205 | kfree(*id); | |
1206 | return error; | |
bc5fc7e4 MW |
1207 | } |
1208 | ||
5d0f6131 | 1209 | int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid, |
08df1e05 | 1210 | dma_addr_t dma_addr, u32 *result) |
bc5fc7e4 MW |
1211 | { |
1212 | struct nvme_command c; | |
1213 | ||
1214 | memset(&c, 0, sizeof(c)); | |
1215 | c.features.opcode = nvme_admin_get_features; | |
a42cecce | 1216 | c.features.nsid = cpu_to_le32(nsid); |
bc5fc7e4 MW |
1217 | c.features.prp1 = cpu_to_le64(dma_addr); |
1218 | c.features.fid = cpu_to_le32(fid); | |
bc5fc7e4 | 1219 | |
d29ec824 CH |
1220 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0, |
1221 | result, 0); | |
df348139 MW |
1222 | } |
1223 | ||
5d0f6131 VV |
1224 | int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11, |
1225 | dma_addr_t dma_addr, u32 *result) | |
df348139 MW |
1226 | { |
1227 | struct nvme_command c; | |
1228 | ||
1229 | memset(&c, 0, sizeof(c)); | |
1230 | c.features.opcode = nvme_admin_set_features; | |
1231 | c.features.prp1 = cpu_to_le64(dma_addr); | |
1232 | c.features.fid = cpu_to_le32(fid); | |
1233 | c.features.dword11 = cpu_to_le32(dword11); | |
1234 | ||
d29ec824 CH |
1235 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0, |
1236 | result, 0); | |
1237 | } | |
1238 | ||
1239 | int nvme_get_log_page(struct nvme_dev *dev, struct nvme_smart_log **log) | |
1240 | { | |
e44ac588 AM |
1241 | struct nvme_command c = { }; |
1242 | int error; | |
1243 | ||
1244 | c.common.opcode = nvme_admin_get_log_page, | |
1245 | c.common.nsid = cpu_to_le32(0xFFFFFFFF), | |
1246 | c.common.cdw10[0] = cpu_to_le32( | |
d29ec824 CH |
1247 | (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) | |
1248 | NVME_LOG_SMART), | |
d29ec824 CH |
1249 | |
1250 | *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL); | |
1251 | if (!*log) | |
1252 | return -ENOMEM; | |
1253 | ||
1254 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *log, | |
1255 | sizeof(struct nvme_smart_log)); | |
1256 | if (error) | |
1257 | kfree(*log); | |
1258 | return error; | |
bc5fc7e4 MW |
1259 | } |
1260 | ||
c30341dc | 1261 | /** |
a4aea562 | 1262 | * nvme_abort_req - Attempt aborting a request |
c30341dc KB |
1263 | * |
1264 | * Schedule controller reset if the command was already aborted once before and | |
1265 | * still hasn't been returned to the driver, or if this is the admin queue. | |
1266 | */ | |
a4aea562 | 1267 | static void nvme_abort_req(struct request *req) |
c30341dc | 1268 | { |
a4aea562 MB |
1269 | struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req); |
1270 | struct nvme_queue *nvmeq = cmd_rq->nvmeq; | |
c30341dc | 1271 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 MB |
1272 | struct request *abort_req; |
1273 | struct nvme_cmd_info *abort_cmd; | |
1274 | struct nvme_command cmd; | |
c30341dc | 1275 | |
a4aea562 | 1276 | if (!nvmeq->qid || cmd_rq->aborted) { |
90667892 CH |
1277 | spin_lock(&dev_list_lock); |
1278 | if (!__nvme_reset(dev)) { | |
1279 | dev_warn(dev->dev, | |
1280 | "I/O %d QID %d timeout, reset controller\n", | |
1281 | req->tag, nvmeq->qid); | |
1282 | } | |
1283 | spin_unlock(&dev_list_lock); | |
c30341dc KB |
1284 | return; |
1285 | } | |
1286 | ||
1287 | if (!dev->abort_limit) | |
1288 | return; | |
1289 | ||
a4aea562 MB |
1290 | abort_req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, |
1291 | false); | |
9f173b33 | 1292 | if (IS_ERR(abort_req)) |
c30341dc KB |
1293 | return; |
1294 | ||
a4aea562 MB |
1295 | abort_cmd = blk_mq_rq_to_pdu(abort_req); |
1296 | nvme_set_info(abort_cmd, abort_req, abort_completion); | |
1297 | ||
c30341dc KB |
1298 | memset(&cmd, 0, sizeof(cmd)); |
1299 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
a4aea562 | 1300 | cmd.abort.cid = req->tag; |
c30341dc | 1301 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); |
a4aea562 | 1302 | cmd.abort.command_id = abort_req->tag; |
c30341dc KB |
1303 | |
1304 | --dev->abort_limit; | |
a4aea562 | 1305 | cmd_rq->aborted = 1; |
c30341dc | 1306 | |
a4aea562 | 1307 | dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag, |
c30341dc | 1308 | nvmeq->qid); |
e3f879bf | 1309 | nvme_submit_cmd(dev->queues[0], &cmd); |
c30341dc KB |
1310 | } |
1311 | ||
42483228 | 1312 | static void nvme_cancel_queue_ios(struct request *req, void *data, bool reserved) |
a09115b2 | 1313 | { |
a4aea562 MB |
1314 | struct nvme_queue *nvmeq = data; |
1315 | void *ctx; | |
1316 | nvme_completion_fn fn; | |
1317 | struct nvme_cmd_info *cmd; | |
cef6a948 KB |
1318 | struct nvme_completion cqe; |
1319 | ||
1320 | if (!blk_mq_request_started(req)) | |
1321 | return; | |
a09115b2 | 1322 | |
a4aea562 | 1323 | cmd = blk_mq_rq_to_pdu(req); |
a09115b2 | 1324 | |
a4aea562 MB |
1325 | if (cmd->ctx == CMD_CTX_CANCELLED) |
1326 | return; | |
1327 | ||
cef6a948 KB |
1328 | if (blk_queue_dying(req->q)) |
1329 | cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1); | |
1330 | else | |
1331 | cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1); | |
1332 | ||
1333 | ||
a4aea562 MB |
1334 | dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n", |
1335 | req->tag, nvmeq->qid); | |
1336 | ctx = cancel_cmd_info(cmd, &fn); | |
1337 | fn(nvmeq, ctx, &cqe); | |
a09115b2 MW |
1338 | } |
1339 | ||
a4aea562 | 1340 | static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) |
9e866774 | 1341 | { |
a4aea562 MB |
1342 | struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req); |
1343 | struct nvme_queue *nvmeq = cmd->nvmeq; | |
1344 | ||
1345 | dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag, | |
1346 | nvmeq->qid); | |
7a509a6b | 1347 | spin_lock_irq(&nvmeq->q_lock); |
07836e65 | 1348 | nvme_abort_req(req); |
7a509a6b | 1349 | spin_unlock_irq(&nvmeq->q_lock); |
a4aea562 | 1350 | |
07836e65 KB |
1351 | /* |
1352 | * The aborted req will be completed on receiving the abort req. | |
1353 | * We enable the timer again. If hit twice, it'll cause a device reset, | |
1354 | * as the device then is in a faulty state. | |
1355 | */ | |
1356 | return BLK_EH_RESET_TIMER; | |
a4aea562 | 1357 | } |
22404274 | 1358 | |
a4aea562 MB |
1359 | static void nvme_free_queue(struct nvme_queue *nvmeq) |
1360 | { | |
9e866774 MW |
1361 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), |
1362 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
8ffaadf7 JD |
1363 | if (nvmeq->sq_cmds) |
1364 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
9e866774 MW |
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 | { | |
22404274 | 1410 | spin_lock_irq(&nvmeq->q_lock); |
42483228 KB |
1411 | if (nvmeq->tags && *nvmeq->tags) |
1412 | blk_mq_all_tag_busy_iter(*nvmeq->tags, nvme_cancel_queue_ios, nvmeq); | |
22404274 | 1413 | spin_unlock_irq(&nvmeq->q_lock); |
b60503ba MW |
1414 | } |
1415 | ||
4d115420 KB |
1416 | static void nvme_disable_queue(struct nvme_dev *dev, int qid) |
1417 | { | |
a4aea562 | 1418 | struct nvme_queue *nvmeq = dev->queues[qid]; |
4d115420 KB |
1419 | |
1420 | if (!nvmeq) | |
1421 | return; | |
1422 | if (nvme_suspend_queue(nvmeq)) | |
1423 | return; | |
1424 | ||
0e53d180 KB |
1425 | /* Don't tell the adapter to delete the admin queue. |
1426 | * Don't tell a removed adapter to delete IO queues. */ | |
1427 | if (qid && readl(&dev->bar->csts) != -1) { | |
b60503ba MW |
1428 | adapter_delete_sq(dev, qid); |
1429 | adapter_delete_cq(dev, qid); | |
1430 | } | |
07836e65 KB |
1431 | |
1432 | spin_lock_irq(&nvmeq->q_lock); | |
1433 | nvme_process_cq(nvmeq); | |
1434 | spin_unlock_irq(&nvmeq->q_lock); | |
b60503ba MW |
1435 | } |
1436 | ||
8ffaadf7 JD |
1437 | static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, |
1438 | int entry_size) | |
1439 | { | |
1440 | int q_depth = dev->q_depth; | |
1441 | unsigned q_size_aligned = roundup(q_depth * entry_size, dev->page_size); | |
1442 | ||
1443 | if (q_size_aligned * nr_io_queues > dev->cmb_size) { | |
c45f5c99 JD |
1444 | u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); |
1445 | mem_per_q = round_down(mem_per_q, dev->page_size); | |
1446 | q_depth = div_u64(mem_per_q, entry_size); | |
8ffaadf7 JD |
1447 | |
1448 | /* | |
1449 | * Ensure the reduced q_depth is above some threshold where it | |
1450 | * would be better to map queues in system memory with the | |
1451 | * original depth | |
1452 | */ | |
1453 | if (q_depth < 64) | |
1454 | return -ENOMEM; | |
1455 | } | |
1456 | ||
1457 | return q_depth; | |
1458 | } | |
1459 | ||
1460 | static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, | |
1461 | int qid, int depth) | |
1462 | { | |
1463 | if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) { | |
1464 | unsigned offset = (qid - 1) * | |
1465 | roundup(SQ_SIZE(depth), dev->page_size); | |
1466 | nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset; | |
1467 | nvmeq->sq_cmds_io = dev->cmb + offset; | |
1468 | } else { | |
1469 | nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth), | |
1470 | &nvmeq->sq_dma_addr, GFP_KERNEL); | |
1471 | if (!nvmeq->sq_cmds) | |
1472 | return -ENOMEM; | |
1473 | } | |
1474 | ||
1475 | return 0; | |
1476 | } | |
1477 | ||
b60503ba | 1478 | static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, |
2b25d981 | 1479 | int depth) |
b60503ba | 1480 | { |
a4aea562 | 1481 | struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL); |
b60503ba MW |
1482 | if (!nvmeq) |
1483 | return NULL; | |
1484 | ||
e75ec752 | 1485 | nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth), |
4d51abf9 | 1486 | &nvmeq->cq_dma_addr, GFP_KERNEL); |
b60503ba MW |
1487 | if (!nvmeq->cqes) |
1488 | goto free_nvmeq; | |
b60503ba | 1489 | |
8ffaadf7 | 1490 | if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth)) |
b60503ba MW |
1491 | goto free_cqdma; |
1492 | ||
e75ec752 | 1493 | nvmeq->q_dmadev = dev->dev; |
091b6092 | 1494 | nvmeq->dev = dev; |
3193f07b MW |
1495 | snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d", |
1496 | dev->instance, qid); | |
b60503ba MW |
1497 | spin_lock_init(&nvmeq->q_lock); |
1498 | nvmeq->cq_head = 0; | |
82123460 | 1499 | nvmeq->cq_phase = 1; |
b80d5ccc | 1500 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
b60503ba | 1501 | nvmeq->q_depth = depth; |
c30341dc | 1502 | nvmeq->qid = qid; |
758dd7fd | 1503 | nvmeq->cq_vector = -1; |
a4aea562 | 1504 | dev->queues[qid] = nvmeq; |
b60503ba | 1505 | |
36a7e993 JD |
1506 | /* make sure queue descriptor is set before queue count, for kthread */ |
1507 | mb(); | |
1508 | dev->queue_count++; | |
1509 | ||
b60503ba MW |
1510 | return nvmeq; |
1511 | ||
1512 | free_cqdma: | |
e75ec752 | 1513 | dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes, |
b60503ba MW |
1514 | nvmeq->cq_dma_addr); |
1515 | free_nvmeq: | |
1516 | kfree(nvmeq); | |
1517 | return NULL; | |
1518 | } | |
1519 | ||
3001082c MW |
1520 | static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq, |
1521 | const char *name) | |
1522 | { | |
58ffacb5 MW |
1523 | if (use_threaded_interrupts) |
1524 | return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector, | |
481e5bad | 1525 | nvme_irq_check, nvme_irq, IRQF_SHARED, |
58ffacb5 | 1526 | name, nvmeq); |
3001082c | 1527 | return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq, |
481e5bad | 1528 | IRQF_SHARED, name, nvmeq); |
3001082c MW |
1529 | } |
1530 | ||
22404274 | 1531 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1532 | { |
22404274 | 1533 | struct nvme_dev *dev = nvmeq->dev; |
b60503ba | 1534 | |
7be50e93 | 1535 | spin_lock_irq(&nvmeq->q_lock); |
22404274 KB |
1536 | nvmeq->sq_tail = 0; |
1537 | nvmeq->cq_head = 0; | |
1538 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1539 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
22404274 | 1540 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); |
42f61420 | 1541 | dev->online_queues++; |
7be50e93 | 1542 | spin_unlock_irq(&nvmeq->q_lock); |
22404274 KB |
1543 | } |
1544 | ||
1545 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) | |
1546 | { | |
1547 | struct nvme_dev *dev = nvmeq->dev; | |
1548 | int result; | |
3f85d50b | 1549 | |
2b25d981 | 1550 | nvmeq->cq_vector = qid - 1; |
b60503ba MW |
1551 | result = adapter_alloc_cq(dev, qid, nvmeq); |
1552 | if (result < 0) | |
22404274 | 1553 | return result; |
b60503ba MW |
1554 | |
1555 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1556 | if (result < 0) | |
1557 | goto release_cq; | |
1558 | ||
3193f07b | 1559 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
b60503ba MW |
1560 | if (result < 0) |
1561 | goto release_sq; | |
1562 | ||
22404274 | 1563 | nvme_init_queue(nvmeq, qid); |
22404274 | 1564 | return result; |
b60503ba MW |
1565 | |
1566 | release_sq: | |
1567 | adapter_delete_sq(dev, qid); | |
1568 | release_cq: | |
1569 | adapter_delete_cq(dev, qid); | |
22404274 | 1570 | return result; |
b60503ba MW |
1571 | } |
1572 | ||
ba47e386 MW |
1573 | static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled) |
1574 | { | |
1575 | unsigned long timeout; | |
1576 | u32 bit = enabled ? NVME_CSTS_RDY : 0; | |
1577 | ||
1578 | timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
1579 | ||
1580 | while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) { | |
1581 | msleep(100); | |
1582 | if (fatal_signal_pending(current)) | |
1583 | return -EINTR; | |
1584 | if (time_after(jiffies, timeout)) { | |
e75ec752 | 1585 | dev_err(dev->dev, |
27e8166c MW |
1586 | "Device not ready; aborting %s\n", enabled ? |
1587 | "initialisation" : "reset"); | |
ba47e386 MW |
1588 | return -ENODEV; |
1589 | } | |
1590 | } | |
1591 | ||
1592 | return 0; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * If the device has been passed off to us in an enabled state, just clear | |
1597 | * the enabled bit. The spec says we should set the 'shutdown notification | |
1598 | * bits', but doing so may cause the device to complete commands to the | |
1599 | * admin queue ... and we don't know what memory that might be pointing at! | |
1600 | */ | |
1601 | static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap) | |
1602 | { | |
01079522 DM |
1603 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1604 | dev->ctrl_config &= ~NVME_CC_ENABLE; | |
1605 | writel(dev->ctrl_config, &dev->bar->cc); | |
44af146a | 1606 | |
ba47e386 MW |
1607 | return nvme_wait_ready(dev, cap, false); |
1608 | } | |
1609 | ||
1610 | static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap) | |
1611 | { | |
01079522 DM |
1612 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1613 | dev->ctrl_config |= NVME_CC_ENABLE; | |
1614 | writel(dev->ctrl_config, &dev->bar->cc); | |
1615 | ||
ba47e386 MW |
1616 | return nvme_wait_ready(dev, cap, true); |
1617 | } | |
1618 | ||
1894d8f1 KB |
1619 | static int nvme_shutdown_ctrl(struct nvme_dev *dev) |
1620 | { | |
1621 | unsigned long timeout; | |
1894d8f1 | 1622 | |
01079522 DM |
1623 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1624 | dev->ctrl_config |= NVME_CC_SHN_NORMAL; | |
1625 | ||
1626 | writel(dev->ctrl_config, &dev->bar->cc); | |
1894d8f1 | 1627 | |
2484f407 | 1628 | timeout = SHUTDOWN_TIMEOUT + jiffies; |
1894d8f1 KB |
1629 | while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) != |
1630 | NVME_CSTS_SHST_CMPLT) { | |
1631 | msleep(100); | |
1632 | if (fatal_signal_pending(current)) | |
1633 | return -EINTR; | |
1634 | if (time_after(jiffies, timeout)) { | |
e75ec752 | 1635 | dev_err(dev->dev, |
1894d8f1 KB |
1636 | "Device shutdown incomplete; abort shutdown\n"); |
1637 | return -ENODEV; | |
1638 | } | |
1639 | } | |
1640 | ||
1641 | return 0; | |
1642 | } | |
1643 | ||
a4aea562 | 1644 | static struct blk_mq_ops nvme_mq_admin_ops = { |
d29ec824 | 1645 | .queue_rq = nvme_queue_rq, |
a4aea562 MB |
1646 | .map_queue = blk_mq_map_queue, |
1647 | .init_hctx = nvme_admin_init_hctx, | |
4af0e21c | 1648 | .exit_hctx = nvme_admin_exit_hctx, |
a4aea562 MB |
1649 | .init_request = nvme_admin_init_request, |
1650 | .timeout = nvme_timeout, | |
1651 | }; | |
1652 | ||
1653 | static struct blk_mq_ops nvme_mq_ops = { | |
1654 | .queue_rq = nvme_queue_rq, | |
1655 | .map_queue = blk_mq_map_queue, | |
1656 | .init_hctx = nvme_init_hctx, | |
1657 | .init_request = nvme_init_request, | |
1658 | .timeout = nvme_timeout, | |
1659 | }; | |
1660 | ||
ea191d2f KB |
1661 | static void nvme_dev_remove_admin(struct nvme_dev *dev) |
1662 | { | |
1663 | if (dev->admin_q && !blk_queue_dying(dev->admin_q)) { | |
1664 | blk_cleanup_queue(dev->admin_q); | |
1665 | blk_mq_free_tag_set(&dev->admin_tagset); | |
1666 | } | |
1667 | } | |
1668 | ||
a4aea562 MB |
1669 | static int nvme_alloc_admin_tags(struct nvme_dev *dev) |
1670 | { | |
1671 | if (!dev->admin_q) { | |
1672 | dev->admin_tagset.ops = &nvme_mq_admin_ops; | |
1673 | dev->admin_tagset.nr_hw_queues = 1; | |
1674 | dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1; | |
1efccc9d | 1675 | dev->admin_tagset.reserved_tags = 1; |
a4aea562 | 1676 | dev->admin_tagset.timeout = ADMIN_TIMEOUT; |
e75ec752 | 1677 | dev->admin_tagset.numa_node = dev_to_node(dev->dev); |
ac3dd5bd | 1678 | dev->admin_tagset.cmd_size = nvme_cmd_size(dev); |
a4aea562 MB |
1679 | dev->admin_tagset.driver_data = dev; |
1680 | ||
1681 | if (blk_mq_alloc_tag_set(&dev->admin_tagset)) | |
1682 | return -ENOMEM; | |
1683 | ||
1684 | dev->admin_q = blk_mq_init_queue(&dev->admin_tagset); | |
35b489d3 | 1685 | if (IS_ERR(dev->admin_q)) { |
a4aea562 MB |
1686 | blk_mq_free_tag_set(&dev->admin_tagset); |
1687 | return -ENOMEM; | |
1688 | } | |
ea191d2f KB |
1689 | if (!blk_get_queue(dev->admin_q)) { |
1690 | nvme_dev_remove_admin(dev); | |
4af0e21c | 1691 | dev->admin_q = NULL; |
ea191d2f KB |
1692 | return -ENODEV; |
1693 | } | |
0fb59cbc KB |
1694 | } else |
1695 | blk_mq_unfreeze_queue(dev->admin_q); | |
a4aea562 MB |
1696 | |
1697 | return 0; | |
1698 | } | |
1699 | ||
8d85fce7 | 1700 | static int nvme_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1701 | { |
ba47e386 | 1702 | int result; |
b60503ba | 1703 | u32 aqa; |
ba47e386 | 1704 | u64 cap = readq(&dev->bar->cap); |
b60503ba | 1705 | struct nvme_queue *nvmeq; |
1d090624 KB |
1706 | unsigned page_shift = PAGE_SHIFT; |
1707 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12; | |
1708 | unsigned dev_page_max = NVME_CAP_MPSMAX(cap) + 12; | |
1709 | ||
1710 | if (page_shift < dev_page_min) { | |
e75ec752 | 1711 | dev_err(dev->dev, |
1d090624 KB |
1712 | "Minimum device page size (%u) too large for " |
1713 | "host (%u)\n", 1 << dev_page_min, | |
1714 | 1 << page_shift); | |
1715 | return -ENODEV; | |
1716 | } | |
1717 | if (page_shift > dev_page_max) { | |
e75ec752 | 1718 | dev_info(dev->dev, |
1d090624 KB |
1719 | "Device maximum page size (%u) smaller than " |
1720 | "host (%u); enabling work-around\n", | |
1721 | 1 << dev_page_max, 1 << page_shift); | |
1722 | page_shift = dev_page_max; | |
1723 | } | |
b60503ba | 1724 | |
dfbac8c7 KB |
1725 | dev->subsystem = readl(&dev->bar->vs) >= NVME_VS(1, 1) ? |
1726 | NVME_CAP_NSSRC(cap) : 0; | |
1727 | ||
1728 | if (dev->subsystem && (readl(&dev->bar->csts) & NVME_CSTS_NSSRO)) | |
1729 | writel(NVME_CSTS_NSSRO, &dev->bar->csts); | |
1730 | ||
ba47e386 MW |
1731 | result = nvme_disable_ctrl(dev, cap); |
1732 | if (result < 0) | |
1733 | return result; | |
b60503ba | 1734 | |
a4aea562 | 1735 | nvmeq = dev->queues[0]; |
cd638946 | 1736 | if (!nvmeq) { |
2b25d981 | 1737 | nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); |
cd638946 KB |
1738 | if (!nvmeq) |
1739 | return -ENOMEM; | |
cd638946 | 1740 | } |
b60503ba MW |
1741 | |
1742 | aqa = nvmeq->q_depth - 1; | |
1743 | aqa |= aqa << 16; | |
1744 | ||
1d090624 KB |
1745 | dev->page_size = 1 << page_shift; |
1746 | ||
01079522 | 1747 | dev->ctrl_config = NVME_CC_CSS_NVM; |
1d090624 | 1748 | dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; |
b60503ba | 1749 | dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; |
7f53f9d2 | 1750 | dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
b60503ba MW |
1751 | |
1752 | writel(aqa, &dev->bar->aqa); | |
1753 | writeq(nvmeq->sq_dma_addr, &dev->bar->asq); | |
1754 | writeq(nvmeq->cq_dma_addr, &dev->bar->acq); | |
b60503ba | 1755 | |
ba47e386 | 1756 | result = nvme_enable_ctrl(dev, cap); |
025c557a | 1757 | if (result) |
a4aea562 MB |
1758 | goto free_nvmeq; |
1759 | ||
2b25d981 | 1760 | nvmeq->cq_vector = 0; |
3193f07b | 1761 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
758dd7fd JD |
1762 | if (result) { |
1763 | nvmeq->cq_vector = -1; | |
0fb59cbc | 1764 | goto free_nvmeq; |
758dd7fd | 1765 | } |
025c557a | 1766 | |
b60503ba | 1767 | return result; |
a4aea562 | 1768 | |
a4aea562 MB |
1769 | free_nvmeq: |
1770 | nvme_free_queues(dev, 0); | |
1771 | return result; | |
b60503ba MW |
1772 | } |
1773 | ||
a53295b6 MW |
1774 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
1775 | { | |
1776 | struct nvme_dev *dev = ns->dev; | |
a53295b6 MW |
1777 | struct nvme_user_io io; |
1778 | struct nvme_command c; | |
d29ec824 | 1779 | unsigned length, meta_len; |
a67a9513 | 1780 | int status, write; |
a67a9513 KB |
1781 | dma_addr_t meta_dma = 0; |
1782 | void *meta = NULL; | |
fec558b5 | 1783 | void __user *metadata; |
a53295b6 MW |
1784 | |
1785 | if (copy_from_user(&io, uio, sizeof(io))) | |
1786 | return -EFAULT; | |
6c7d4945 MW |
1787 | |
1788 | switch (io.opcode) { | |
1789 | case nvme_cmd_write: | |
1790 | case nvme_cmd_read: | |
6bbf1acd | 1791 | case nvme_cmd_compare: |
6413214c | 1792 | break; |
6c7d4945 | 1793 | default: |
6bbf1acd | 1794 | return -EINVAL; |
6c7d4945 MW |
1795 | } |
1796 | ||
d29ec824 CH |
1797 | length = (io.nblocks + 1) << ns->lba_shift; |
1798 | meta_len = (io.nblocks + 1) * ns->ms; | |
835da3f9 | 1799 | metadata = (void __user *)(uintptr_t)io.metadata; |
d29ec824 | 1800 | write = io.opcode & 1; |
a53295b6 | 1801 | |
71feb364 KB |
1802 | if (ns->ext) { |
1803 | length += meta_len; | |
1804 | meta_len = 0; | |
a67a9513 KB |
1805 | } |
1806 | if (meta_len) { | |
d29ec824 CH |
1807 | if (((io.metadata & 3) || !io.metadata) && !ns->ext) |
1808 | return -EINVAL; | |
1809 | ||
e75ec752 | 1810 | meta = dma_alloc_coherent(dev->dev, meta_len, |
a67a9513 | 1811 | &meta_dma, GFP_KERNEL); |
fec558b5 | 1812 | |
a67a9513 KB |
1813 | if (!meta) { |
1814 | status = -ENOMEM; | |
1815 | goto unmap; | |
1816 | } | |
1817 | if (write) { | |
fec558b5 | 1818 | if (copy_from_user(meta, metadata, meta_len)) { |
a67a9513 KB |
1819 | status = -EFAULT; |
1820 | goto unmap; | |
1821 | } | |
1822 | } | |
1823 | } | |
1824 | ||
a53295b6 MW |
1825 | memset(&c, 0, sizeof(c)); |
1826 | c.rw.opcode = io.opcode; | |
1827 | c.rw.flags = io.flags; | |
6c7d4945 | 1828 | c.rw.nsid = cpu_to_le32(ns->ns_id); |
a53295b6 | 1829 | c.rw.slba = cpu_to_le64(io.slba); |
6c7d4945 | 1830 | c.rw.length = cpu_to_le16(io.nblocks); |
a53295b6 | 1831 | c.rw.control = cpu_to_le16(io.control); |
1c9b5265 MW |
1832 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); |
1833 | c.rw.reftag = cpu_to_le32(io.reftag); | |
1834 | c.rw.apptag = cpu_to_le16(io.apptag); | |
1835 | c.rw.appmask = cpu_to_le16(io.appmask); | |
a67a9513 | 1836 | c.rw.metadata = cpu_to_le64(meta_dma); |
d29ec824 CH |
1837 | |
1838 | status = __nvme_submit_sync_cmd(ns->queue, &c, NULL, | |
835da3f9 | 1839 | (void __user *)(uintptr_t)io.addr, length, NULL, 0); |
f410c680 | 1840 | unmap: |
a67a9513 KB |
1841 | if (meta) { |
1842 | if (status == NVME_SC_SUCCESS && !write) { | |
fec558b5 | 1843 | if (copy_to_user(metadata, meta, meta_len)) |
a67a9513 KB |
1844 | status = -EFAULT; |
1845 | } | |
e75ec752 | 1846 | dma_free_coherent(dev->dev, meta_len, meta, meta_dma); |
f410c680 | 1847 | } |
a53295b6 MW |
1848 | return status; |
1849 | } | |
1850 | ||
a4aea562 MB |
1851 | static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns, |
1852 | struct nvme_passthru_cmd __user *ucmd) | |
6ee44cdc | 1853 | { |
7963e521 | 1854 | struct nvme_passthru_cmd cmd; |
6ee44cdc | 1855 | struct nvme_command c; |
d29ec824 CH |
1856 | unsigned timeout = 0; |
1857 | int status; | |
6ee44cdc | 1858 | |
6bbf1acd MW |
1859 | if (!capable(CAP_SYS_ADMIN)) |
1860 | return -EACCES; | |
1861 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
6ee44cdc | 1862 | return -EFAULT; |
6ee44cdc MW |
1863 | |
1864 | memset(&c, 0, sizeof(c)); | |
6bbf1acd MW |
1865 | c.common.opcode = cmd.opcode; |
1866 | c.common.flags = cmd.flags; | |
1867 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1868 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1869 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1870 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
1871 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
1872 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
1873 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
1874 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
1875 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
1876 | ||
d29ec824 CH |
1877 | if (cmd.timeout_ms) |
1878 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
eca18b23 | 1879 | |
f705f837 | 1880 | status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c, |
835da3f9 | 1881 | NULL, (void __user *)(uintptr_t)cmd.addr, cmd.data_len, |
d29ec824 CH |
1882 | &cmd.result, timeout); |
1883 | if (status >= 0) { | |
1884 | if (put_user(cmd.result, &ucmd->result)) | |
1885 | return -EFAULT; | |
6bbf1acd | 1886 | } |
f4f117f6 | 1887 | |
6ee44cdc MW |
1888 | return status; |
1889 | } | |
1890 | ||
81f03fed JD |
1891 | static int nvme_subsys_reset(struct nvme_dev *dev) |
1892 | { | |
1893 | if (!dev->subsystem) | |
1894 | return -ENOTTY; | |
1895 | ||
1896 | writel(0x4E564D65, &dev->bar->nssr); /* "NVMe" */ | |
1897 | return 0; | |
1898 | } | |
1899 | ||
b60503ba MW |
1900 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, |
1901 | unsigned long arg) | |
1902 | { | |
1903 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1904 | ||
1905 | switch (cmd) { | |
6bbf1acd | 1906 | case NVME_IOCTL_ID: |
c3bfe717 | 1907 | force_successful_syscall_return(); |
6bbf1acd MW |
1908 | return ns->ns_id; |
1909 | case NVME_IOCTL_ADMIN_CMD: | |
a4aea562 | 1910 | return nvme_user_cmd(ns->dev, NULL, (void __user *)arg); |
7963e521 | 1911 | case NVME_IOCTL_IO_CMD: |
a4aea562 | 1912 | return nvme_user_cmd(ns->dev, ns, (void __user *)arg); |
a53295b6 MW |
1913 | case NVME_IOCTL_SUBMIT_IO: |
1914 | return nvme_submit_io(ns, (void __user *)arg); | |
5d0f6131 VV |
1915 | case SG_GET_VERSION_NUM: |
1916 | return nvme_sg_get_version_num((void __user *)arg); | |
1917 | case SG_IO: | |
1918 | return nvme_sg_io(ns, (void __user *)arg); | |
b60503ba MW |
1919 | default: |
1920 | return -ENOTTY; | |
1921 | } | |
1922 | } | |
1923 | ||
320a3827 KB |
1924 | #ifdef CONFIG_COMPAT |
1925 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1926 | unsigned int cmd, unsigned long arg) | |
1927 | { | |
320a3827 KB |
1928 | switch (cmd) { |
1929 | case SG_IO: | |
e179729a | 1930 | return -ENOIOCTLCMD; |
320a3827 KB |
1931 | } |
1932 | return nvme_ioctl(bdev, mode, cmd, arg); | |
1933 | } | |
1934 | #else | |
1935 | #define nvme_compat_ioctl NULL | |
1936 | #endif | |
1937 | ||
5105aa55 | 1938 | static void nvme_free_dev(struct kref *kref); |
188c3568 KB |
1939 | static void nvme_free_ns(struct kref *kref) |
1940 | { | |
1941 | struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref); | |
1942 | ||
ca064085 MB |
1943 | if (ns->type == NVME_NS_LIGHTNVM) |
1944 | nvme_nvm_unregister(ns->queue, ns->disk->disk_name); | |
1945 | ||
188c3568 KB |
1946 | spin_lock(&dev_list_lock); |
1947 | ns->disk->private_data = NULL; | |
1948 | spin_unlock(&dev_list_lock); | |
1949 | ||
5105aa55 | 1950 | kref_put(&ns->dev->kref, nvme_free_dev); |
188c3568 KB |
1951 | put_disk(ns->disk); |
1952 | kfree(ns); | |
1953 | } | |
1954 | ||
9ac27090 KB |
1955 | static int nvme_open(struct block_device *bdev, fmode_t mode) |
1956 | { | |
9e60352c KB |
1957 | int ret = 0; |
1958 | struct nvme_ns *ns; | |
9ac27090 | 1959 | |
9e60352c KB |
1960 | spin_lock(&dev_list_lock); |
1961 | ns = bdev->bd_disk->private_data; | |
1962 | if (!ns) | |
1963 | ret = -ENXIO; | |
188c3568 | 1964 | else if (!kref_get_unless_zero(&ns->kref)) |
9e60352c KB |
1965 | ret = -ENXIO; |
1966 | spin_unlock(&dev_list_lock); | |
1967 | ||
1968 | return ret; | |
9ac27090 KB |
1969 | } |
1970 | ||
9ac27090 KB |
1971 | static void nvme_release(struct gendisk *disk, fmode_t mode) |
1972 | { | |
1973 | struct nvme_ns *ns = disk->private_data; | |
188c3568 | 1974 | kref_put(&ns->kref, nvme_free_ns); |
9ac27090 KB |
1975 | } |
1976 | ||
4cc09e2d KB |
1977 | static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo) |
1978 | { | |
1979 | /* some standard values */ | |
1980 | geo->heads = 1 << 6; | |
1981 | geo->sectors = 1 << 5; | |
1982 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
1983 | return 0; | |
1984 | } | |
1985 | ||
e1e5e564 KB |
1986 | static void nvme_config_discard(struct nvme_ns *ns) |
1987 | { | |
1988 | u32 logical_block_size = queue_logical_block_size(ns->queue); | |
1989 | ns->queue->limits.discard_zeroes_data = 0; | |
1990 | ns->queue->limits.discard_alignment = logical_block_size; | |
1991 | ns->queue->limits.discard_granularity = logical_block_size; | |
2bb4cd5c | 1992 | blk_queue_max_discard_sectors(ns->queue, 0xffffffff); |
e1e5e564 KB |
1993 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); |
1994 | } | |
1995 | ||
1b9dbf7f KB |
1996 | static int nvme_revalidate_disk(struct gendisk *disk) |
1997 | { | |
1998 | struct nvme_ns *ns = disk->private_data; | |
1999 | struct nvme_dev *dev = ns->dev; | |
2000 | struct nvme_id_ns *id; | |
a67a9513 KB |
2001 | u8 lbaf, pi_type; |
2002 | u16 old_ms; | |
e1e5e564 | 2003 | unsigned short bs; |
1b9dbf7f | 2004 | |
d29ec824 | 2005 | if (nvme_identify_ns(dev, ns->ns_id, &id)) { |
a5768aa8 KB |
2006 | dev_warn(dev->dev, "%s: Identify failure nvme%dn%d\n", __func__, |
2007 | dev->instance, ns->ns_id); | |
2008 | return -ENODEV; | |
1b9dbf7f | 2009 | } |
a5768aa8 KB |
2010 | if (id->ncap == 0) { |
2011 | kfree(id); | |
2012 | return -ENODEV; | |
e1e5e564 | 2013 | } |
1b9dbf7f | 2014 | |
ca064085 MB |
2015 | if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) { |
2016 | if (nvme_nvm_register(ns->queue, disk->disk_name)) { | |
2017 | dev_warn(dev->dev, | |
2018 | "%s: LightNVM init failure\n", __func__); | |
2019 | kfree(id); | |
2020 | return -ENODEV; | |
2021 | } | |
2022 | ns->type = NVME_NS_LIGHTNVM; | |
2023 | } | |
2024 | ||
e1e5e564 KB |
2025 | old_ms = ns->ms; |
2026 | lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK; | |
1b9dbf7f | 2027 | ns->lba_shift = id->lbaf[lbaf].ds; |
e1e5e564 | 2028 | ns->ms = le16_to_cpu(id->lbaf[lbaf].ms); |
a67a9513 | 2029 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); |
e1e5e564 KB |
2030 | |
2031 | /* | |
2032 | * If identify namespace failed, use default 512 byte block size so | |
2033 | * block layer can use before failing read/write for 0 capacity. | |
2034 | */ | |
2035 | if (ns->lba_shift == 0) | |
2036 | ns->lba_shift = 9; | |
2037 | bs = 1 << ns->lba_shift; | |
2038 | ||
2039 | /* XXX: PI implementation requires metadata equal t10 pi tuple size */ | |
2040 | pi_type = ns->ms == sizeof(struct t10_pi_tuple) ? | |
2041 | id->dps & NVME_NS_DPS_PI_MASK : 0; | |
2042 | ||
4cfc766e | 2043 | blk_mq_freeze_queue(disk->queue); |
52b68d7e KB |
2044 | if (blk_get_integrity(disk) && (ns->pi_type != pi_type || |
2045 | ns->ms != old_ms || | |
e1e5e564 | 2046 | bs != queue_logical_block_size(disk->queue) || |
a67a9513 | 2047 | (ns->ms && ns->ext))) |
e1e5e564 KB |
2048 | blk_integrity_unregister(disk); |
2049 | ||
2050 | ns->pi_type = pi_type; | |
2051 | blk_queue_logical_block_size(ns->queue, bs); | |
2052 | ||
25520d55 | 2053 | if (ns->ms && !ns->ext) |
e1e5e564 KB |
2054 | nvme_init_integrity(ns); |
2055 | ||
ca064085 MB |
2056 | if ((ns->ms && !(ns->ms == 8 && ns->pi_type) && |
2057 | !blk_get_integrity(disk)) || | |
2058 | ns->type == NVME_NS_LIGHTNVM) | |
e1e5e564 KB |
2059 | set_capacity(disk, 0); |
2060 | else | |
2061 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); | |
2062 | ||
2063 | if (dev->oncs & NVME_CTRL_ONCS_DSM) | |
2064 | nvme_config_discard(ns); | |
4cfc766e | 2065 | blk_mq_unfreeze_queue(disk->queue); |
1b9dbf7f | 2066 | |
d29ec824 | 2067 | kfree(id); |
1b9dbf7f KB |
2068 | return 0; |
2069 | } | |
2070 | ||
1d277a63 KB |
2071 | static char nvme_pr_type(enum pr_type type) |
2072 | { | |
2073 | switch (type) { | |
2074 | case PR_WRITE_EXCLUSIVE: | |
2075 | return 1; | |
2076 | case PR_EXCLUSIVE_ACCESS: | |
2077 | return 2; | |
2078 | case PR_WRITE_EXCLUSIVE_REG_ONLY: | |
2079 | return 3; | |
2080 | case PR_EXCLUSIVE_ACCESS_REG_ONLY: | |
2081 | return 4; | |
2082 | case PR_WRITE_EXCLUSIVE_ALL_REGS: | |
2083 | return 5; | |
2084 | case PR_EXCLUSIVE_ACCESS_ALL_REGS: | |
2085 | return 6; | |
2086 | default: | |
2087 | return 0; | |
2088 | } | |
2089 | }; | |
2090 | ||
2091 | static int nvme_pr_command(struct block_device *bdev, u32 cdw10, | |
2092 | u64 key, u64 sa_key, u8 op) | |
2093 | { | |
2094 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
2095 | struct nvme_command c; | |
2096 | u8 data[16] = { 0, }; | |
2097 | ||
2098 | put_unaligned_le64(key, &data[0]); | |
2099 | put_unaligned_le64(sa_key, &data[8]); | |
2100 | ||
2101 | memset(&c, 0, sizeof(c)); | |
2102 | c.common.opcode = op; | |
a6dd1020 CH |
2103 | c.common.nsid = cpu_to_le32(ns->ns_id); |
2104 | c.common.cdw10[0] = cpu_to_le32(cdw10); | |
1d277a63 KB |
2105 | |
2106 | return nvme_submit_sync_cmd(ns->queue, &c, data, 16); | |
2107 | } | |
2108 | ||
2109 | static int nvme_pr_register(struct block_device *bdev, u64 old, | |
2110 | u64 new, unsigned flags) | |
2111 | { | |
2112 | u32 cdw10; | |
2113 | ||
2114 | if (flags & ~PR_FL_IGNORE_KEY) | |
2115 | return -EOPNOTSUPP; | |
2116 | ||
2117 | cdw10 = old ? 2 : 0; | |
2118 | cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; | |
2119 | cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ | |
2120 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register); | |
2121 | } | |
2122 | ||
2123 | static int nvme_pr_reserve(struct block_device *bdev, u64 key, | |
2124 | enum pr_type type, unsigned flags) | |
2125 | { | |
2126 | u32 cdw10; | |
2127 | ||
2128 | if (flags & ~PR_FL_IGNORE_KEY) | |
2129 | return -EOPNOTSUPP; | |
2130 | ||
2131 | cdw10 = nvme_pr_type(type) << 8; | |
2132 | cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); | |
2133 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire); | |
2134 | } | |
2135 | ||
2136 | static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, | |
2137 | enum pr_type type, bool abort) | |
2138 | { | |
2139 | u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1; | |
2140 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire); | |
2141 | } | |
2142 | ||
2143 | static int nvme_pr_clear(struct block_device *bdev, u64 key) | |
2144 | { | |
73fcf4e2 | 2145 | u32 cdw10 = 1 | (key ? 1 << 3 : 0); |
1d277a63 KB |
2146 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register); |
2147 | } | |
2148 | ||
2149 | static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2150 | { | |
2151 | u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0; | |
2152 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release); | |
2153 | } | |
2154 | ||
2155 | static const struct pr_ops nvme_pr_ops = { | |
2156 | .pr_register = nvme_pr_register, | |
2157 | .pr_reserve = nvme_pr_reserve, | |
2158 | .pr_release = nvme_pr_release, | |
2159 | .pr_preempt = nvme_pr_preempt, | |
2160 | .pr_clear = nvme_pr_clear, | |
2161 | }; | |
2162 | ||
b60503ba MW |
2163 | static const struct block_device_operations nvme_fops = { |
2164 | .owner = THIS_MODULE, | |
2165 | .ioctl = nvme_ioctl, | |
320a3827 | 2166 | .compat_ioctl = nvme_compat_ioctl, |
9ac27090 KB |
2167 | .open = nvme_open, |
2168 | .release = nvme_release, | |
4cc09e2d | 2169 | .getgeo = nvme_getgeo, |
1b9dbf7f | 2170 | .revalidate_disk= nvme_revalidate_disk, |
1d277a63 | 2171 | .pr_ops = &nvme_pr_ops, |
b60503ba MW |
2172 | }; |
2173 | ||
1fa6aead MW |
2174 | static int nvme_kthread(void *data) |
2175 | { | |
d4b4ff8e | 2176 | struct nvme_dev *dev, *next; |
1fa6aead MW |
2177 | |
2178 | while (!kthread_should_stop()) { | |
564a232c | 2179 | set_current_state(TASK_INTERRUPTIBLE); |
1fa6aead | 2180 | spin_lock(&dev_list_lock); |
d4b4ff8e | 2181 | list_for_each_entry_safe(dev, next, &dev_list, node) { |
1fa6aead | 2182 | int i; |
dfbac8c7 KB |
2183 | u32 csts = readl(&dev->bar->csts); |
2184 | ||
2185 | if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) || | |
2186 | csts & NVME_CSTS_CFS) { | |
90667892 CH |
2187 | if (!__nvme_reset(dev)) { |
2188 | dev_warn(dev->dev, | |
2189 | "Failed status: %x, reset controller\n", | |
2190 | readl(&dev->bar->csts)); | |
2191 | } | |
d4b4ff8e KB |
2192 | continue; |
2193 | } | |
1fa6aead | 2194 | for (i = 0; i < dev->queue_count; i++) { |
a4aea562 | 2195 | struct nvme_queue *nvmeq = dev->queues[i]; |
740216fc MW |
2196 | if (!nvmeq) |
2197 | continue; | |
1fa6aead | 2198 | spin_lock_irq(&nvmeq->q_lock); |
bc57a0f7 | 2199 | nvme_process_cq(nvmeq); |
6fccf938 KB |
2200 | |
2201 | while ((i == 0) && (dev->event_limit > 0)) { | |
a4aea562 | 2202 | if (nvme_submit_async_admin_req(dev)) |
6fccf938 KB |
2203 | break; |
2204 | dev->event_limit--; | |
2205 | } | |
1fa6aead MW |
2206 | spin_unlock_irq(&nvmeq->q_lock); |
2207 | } | |
2208 | } | |
2209 | spin_unlock(&dev_list_lock); | |
acb7aa0d | 2210 | schedule_timeout(round_jiffies_relative(HZ)); |
1fa6aead MW |
2211 | } |
2212 | return 0; | |
2213 | } | |
2214 | ||
e1e5e564 | 2215 | static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid) |
b60503ba MW |
2216 | { |
2217 | struct nvme_ns *ns; | |
2218 | struct gendisk *disk; | |
e75ec752 | 2219 | int node = dev_to_node(dev->dev); |
b60503ba | 2220 | |
a4aea562 | 2221 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); |
b60503ba | 2222 | if (!ns) |
e1e5e564 KB |
2223 | return; |
2224 | ||
a4aea562 | 2225 | ns->queue = blk_mq_init_queue(&dev->tagset); |
9f173b33 | 2226 | if (IS_ERR(ns->queue)) |
b60503ba | 2227 | goto out_free_ns; |
4eeb9215 MW |
2228 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue); |
2229 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); | |
b60503ba MW |
2230 | ns->dev = dev; |
2231 | ns->queue->queuedata = ns; | |
2232 | ||
a4aea562 | 2233 | disk = alloc_disk_node(0, node); |
b60503ba MW |
2234 | if (!disk) |
2235 | goto out_free_queue; | |
a4aea562 | 2236 | |
188c3568 | 2237 | kref_init(&ns->kref); |
5aff9382 | 2238 | ns->ns_id = nsid; |
b60503ba | 2239 | ns->disk = disk; |
e1e5e564 KB |
2240 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ |
2241 | list_add_tail(&ns->list, &dev->namespaces); | |
2242 | ||
e9ef4636 | 2243 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); |
e824410f | 2244 | if (dev->max_hw_sectors) { |
8fc23e03 | 2245 | blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors); |
e824410f KB |
2246 | blk_queue_max_segments(ns->queue, |
2247 | ((dev->max_hw_sectors << 9) / dev->page_size) + 1); | |
2248 | } | |
a4aea562 MB |
2249 | if (dev->stripe_size) |
2250 | blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9); | |
a7d2ce28 KB |
2251 | if (dev->vwc & NVME_CTRL_VWC_PRESENT) |
2252 | blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA); | |
03100aad | 2253 | blk_queue_virt_boundary(ns->queue, dev->page_size - 1); |
b60503ba MW |
2254 | |
2255 | disk->major = nvme_major; | |
469071a3 | 2256 | disk->first_minor = 0; |
b60503ba MW |
2257 | disk->fops = &nvme_fops; |
2258 | disk->private_data = ns; | |
2259 | disk->queue = ns->queue; | |
b3fffdef | 2260 | disk->driverfs_dev = dev->device; |
469071a3 | 2261 | disk->flags = GENHD_FL_EXT_DEVT; |
5aff9382 | 2262 | sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid); |
b60503ba | 2263 | |
e1e5e564 KB |
2264 | /* |
2265 | * Initialize capacity to 0 until we establish the namespace format and | |
2266 | * setup integrity extentions if necessary. The revalidate_disk after | |
2267 | * add_disk allows the driver to register with integrity if the format | |
2268 | * requires it. | |
2269 | */ | |
2270 | set_capacity(disk, 0); | |
a5768aa8 KB |
2271 | if (nvme_revalidate_disk(ns->disk)) |
2272 | goto out_free_disk; | |
2273 | ||
5105aa55 | 2274 | kref_get(&dev->kref); |
ca064085 MB |
2275 | if (ns->type != NVME_NS_LIGHTNVM) { |
2276 | add_disk(ns->disk); | |
2277 | if (ns->ms) { | |
2278 | struct block_device *bd = bdget_disk(ns->disk, 0); | |
2279 | if (!bd) | |
2280 | return; | |
2281 | if (blkdev_get(bd, FMODE_READ, NULL)) { | |
2282 | bdput(bd); | |
2283 | return; | |
2284 | } | |
2285 | blkdev_reread_part(bd); | |
2286 | blkdev_put(bd, FMODE_READ); | |
7bee6074 | 2287 | } |
7bee6074 | 2288 | } |
e1e5e564 | 2289 | return; |
a5768aa8 KB |
2290 | out_free_disk: |
2291 | kfree(disk); | |
2292 | list_del(&ns->list); | |
b60503ba MW |
2293 | out_free_queue: |
2294 | blk_cleanup_queue(ns->queue); | |
2295 | out_free_ns: | |
2296 | kfree(ns); | |
b60503ba MW |
2297 | } |
2298 | ||
2659e57b CH |
2299 | /* |
2300 | * Create I/O queues. Failing to create an I/O queue is not an issue, | |
2301 | * we can continue with less than the desired amount of queues, and | |
2302 | * even a controller without I/O queues an still be used to issue | |
2303 | * admin commands. This might be useful to upgrade a buggy firmware | |
2304 | * for example. | |
2305 | */ | |
42f61420 KB |
2306 | static void nvme_create_io_queues(struct nvme_dev *dev) |
2307 | { | |
a4aea562 | 2308 | unsigned i; |
42f61420 | 2309 | |
a4aea562 | 2310 | for (i = dev->queue_count; i <= dev->max_qid; i++) |
2b25d981 | 2311 | if (!nvme_alloc_queue(dev, i, dev->q_depth)) |
42f61420 KB |
2312 | break; |
2313 | ||
a4aea562 | 2314 | for (i = dev->online_queues; i <= dev->queue_count - 1; i++) |
2659e57b CH |
2315 | if (nvme_create_queue(dev->queues[i], i)) { |
2316 | nvme_free_queues(dev, i); | |
42f61420 | 2317 | break; |
2659e57b | 2318 | } |
42f61420 KB |
2319 | } |
2320 | ||
b3b06812 | 2321 | static int set_queue_count(struct nvme_dev *dev, int count) |
b60503ba MW |
2322 | { |
2323 | int status; | |
2324 | u32 result; | |
b3b06812 | 2325 | u32 q_count = (count - 1) | ((count - 1) << 16); |
b60503ba | 2326 | |
df348139 | 2327 | status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0, |
bc5fc7e4 | 2328 | &result); |
27e8166c MW |
2329 | if (status < 0) |
2330 | return status; | |
2331 | if (status > 0) { | |
e75ec752 | 2332 | dev_err(dev->dev, "Could not set queue count (%d)\n", status); |
badc34d4 | 2333 | return 0; |
27e8166c | 2334 | } |
b60503ba MW |
2335 | return min(result & 0xffff, result >> 16) + 1; |
2336 | } | |
2337 | ||
8ffaadf7 JD |
2338 | static void __iomem *nvme_map_cmb(struct nvme_dev *dev) |
2339 | { | |
2340 | u64 szu, size, offset; | |
2341 | u32 cmbloc; | |
2342 | resource_size_t bar_size; | |
2343 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
2344 | void __iomem *cmb; | |
2345 | dma_addr_t dma_addr; | |
2346 | ||
2347 | if (!use_cmb_sqes) | |
2348 | return NULL; | |
2349 | ||
2350 | dev->cmbsz = readl(&dev->bar->cmbsz); | |
2351 | if (!(NVME_CMB_SZ(dev->cmbsz))) | |
2352 | return NULL; | |
2353 | ||
2354 | cmbloc = readl(&dev->bar->cmbloc); | |
2355 | ||
2356 | szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz)); | |
2357 | size = szu * NVME_CMB_SZ(dev->cmbsz); | |
2358 | offset = szu * NVME_CMB_OFST(cmbloc); | |
2359 | bar_size = pci_resource_len(pdev, NVME_CMB_BIR(cmbloc)); | |
2360 | ||
2361 | if (offset > bar_size) | |
2362 | return NULL; | |
2363 | ||
2364 | /* | |
2365 | * Controllers may support a CMB size larger than their BAR, | |
2366 | * for example, due to being behind a bridge. Reduce the CMB to | |
2367 | * the reported size of the BAR | |
2368 | */ | |
2369 | if (size > bar_size - offset) | |
2370 | size = bar_size - offset; | |
2371 | ||
2372 | dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(cmbloc)) + offset; | |
2373 | cmb = ioremap_wc(dma_addr, size); | |
2374 | if (!cmb) | |
2375 | return NULL; | |
2376 | ||
2377 | dev->cmb_dma_addr = dma_addr; | |
2378 | dev->cmb_size = size; | |
2379 | return cmb; | |
2380 | } | |
2381 | ||
2382 | static inline void nvme_release_cmb(struct nvme_dev *dev) | |
2383 | { | |
2384 | if (dev->cmb) { | |
2385 | iounmap(dev->cmb); | |
2386 | dev->cmb = NULL; | |
2387 | } | |
2388 | } | |
2389 | ||
9d713c2b KB |
2390 | static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
2391 | { | |
b80d5ccc | 2392 | return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride); |
9d713c2b KB |
2393 | } |
2394 | ||
8d85fce7 | 2395 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 2396 | { |
a4aea562 | 2397 | struct nvme_queue *adminq = dev->queues[0]; |
e75ec752 | 2398 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
42f61420 | 2399 | int result, i, vecs, nr_io_queues, size; |
b60503ba | 2400 | |
42f61420 | 2401 | nr_io_queues = num_possible_cpus(); |
b348b7d5 | 2402 | result = set_queue_count(dev, nr_io_queues); |
badc34d4 | 2403 | if (result <= 0) |
1b23484b | 2404 | return result; |
b348b7d5 MW |
2405 | if (result < nr_io_queues) |
2406 | nr_io_queues = result; | |
b60503ba | 2407 | |
8ffaadf7 JD |
2408 | if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) { |
2409 | result = nvme_cmb_qdepth(dev, nr_io_queues, | |
2410 | sizeof(struct nvme_command)); | |
2411 | if (result > 0) | |
2412 | dev->q_depth = result; | |
2413 | else | |
2414 | nvme_release_cmb(dev); | |
2415 | } | |
2416 | ||
9d713c2b KB |
2417 | size = db_bar_size(dev, nr_io_queues); |
2418 | if (size > 8192) { | |
f1938f6e | 2419 | iounmap(dev->bar); |
9d713c2b KB |
2420 | do { |
2421 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
2422 | if (dev->bar) | |
2423 | break; | |
2424 | if (!--nr_io_queues) | |
2425 | return -ENOMEM; | |
2426 | size = db_bar_size(dev, nr_io_queues); | |
2427 | } while (1); | |
f1938f6e | 2428 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
5a92e700 | 2429 | adminq->q_db = dev->dbs; |
f1938f6e MW |
2430 | } |
2431 | ||
9d713c2b | 2432 | /* Deregister the admin queue's interrupt */ |
3193f07b | 2433 | free_irq(dev->entry[0].vector, adminq); |
9d713c2b | 2434 | |
e32efbfc JA |
2435 | /* |
2436 | * If we enable msix early due to not intx, disable it again before | |
2437 | * setting up the full range we need. | |
2438 | */ | |
2439 | if (!pdev->irq) | |
2440 | pci_disable_msix(pdev); | |
2441 | ||
be577fab | 2442 | for (i = 0; i < nr_io_queues; i++) |
1b23484b | 2443 | dev->entry[i].entry = i; |
be577fab AG |
2444 | vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues); |
2445 | if (vecs < 0) { | |
2446 | vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32)); | |
2447 | if (vecs < 0) { | |
2448 | vecs = 1; | |
2449 | } else { | |
2450 | for (i = 0; i < vecs; i++) | |
2451 | dev->entry[i].vector = i + pdev->irq; | |
fa08a396 RRG |
2452 | } |
2453 | } | |
2454 | ||
063a8096 MW |
2455 | /* |
2456 | * Should investigate if there's a performance win from allocating | |
2457 | * more queues than interrupt vectors; it might allow the submission | |
2458 | * path to scale better, even if the receive path is limited by the | |
2459 | * number of interrupts. | |
2460 | */ | |
2461 | nr_io_queues = vecs; | |
42f61420 | 2462 | dev->max_qid = nr_io_queues; |
063a8096 | 2463 | |
3193f07b | 2464 | result = queue_request_irq(dev, adminq, adminq->irqname); |
758dd7fd JD |
2465 | if (result) { |
2466 | adminq->cq_vector = -1; | |
22404274 | 2467 | goto free_queues; |
758dd7fd | 2468 | } |
1b23484b | 2469 | |
cd638946 | 2470 | /* Free previously allocated queues that are no longer usable */ |
42f61420 | 2471 | nvme_free_queues(dev, nr_io_queues + 1); |
a4aea562 | 2472 | nvme_create_io_queues(dev); |
9ecdc946 | 2473 | |
22404274 | 2474 | return 0; |
b60503ba | 2475 | |
22404274 | 2476 | free_queues: |
a1a5ef99 | 2477 | nvme_free_queues(dev, 1); |
22404274 | 2478 | return result; |
b60503ba MW |
2479 | } |
2480 | ||
a5768aa8 KB |
2481 | static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) |
2482 | { | |
2483 | struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); | |
2484 | struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); | |
2485 | ||
2486 | return nsa->ns_id - nsb->ns_id; | |
2487 | } | |
2488 | ||
2489 | static struct nvme_ns *nvme_find_ns(struct nvme_dev *dev, unsigned nsid) | |
2490 | { | |
2491 | struct nvme_ns *ns; | |
2492 | ||
2493 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2494 | if (ns->ns_id == nsid) | |
2495 | return ns; | |
2496 | if (ns->ns_id > nsid) | |
2497 | break; | |
2498 | } | |
2499 | return NULL; | |
2500 | } | |
2501 | ||
2502 | static inline bool nvme_io_incapable(struct nvme_dev *dev) | |
2503 | { | |
2504 | return (!dev->bar || readl(&dev->bar->csts) & NVME_CSTS_CFS || | |
2505 | dev->online_queues < 2); | |
2506 | } | |
2507 | ||
2508 | static void nvme_ns_remove(struct nvme_ns *ns) | |
2509 | { | |
2510 | bool kill = nvme_io_incapable(ns->dev) && !blk_queue_dying(ns->queue); | |
2511 | ||
2512 | if (kill) | |
2513 | blk_set_queue_dying(ns->queue); | |
9609b994 | 2514 | if (ns->disk->flags & GENHD_FL_UP) |
a5768aa8 | 2515 | del_gendisk(ns->disk); |
a5768aa8 KB |
2516 | if (kill || !blk_queue_dying(ns->queue)) { |
2517 | blk_mq_abort_requeue_list(ns->queue); | |
2518 | blk_cleanup_queue(ns->queue); | |
5105aa55 KB |
2519 | } |
2520 | list_del_init(&ns->list); | |
2521 | kref_put(&ns->kref, nvme_free_ns); | |
a5768aa8 KB |
2522 | } |
2523 | ||
2524 | static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn) | |
2525 | { | |
2526 | struct nvme_ns *ns, *next; | |
2527 | unsigned i; | |
2528 | ||
2529 | for (i = 1; i <= nn; i++) { | |
2530 | ns = nvme_find_ns(dev, i); | |
2531 | if (ns) { | |
5105aa55 | 2532 | if (revalidate_disk(ns->disk)) |
a5768aa8 | 2533 | nvme_ns_remove(ns); |
a5768aa8 KB |
2534 | } else |
2535 | nvme_alloc_ns(dev, i); | |
2536 | } | |
2537 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) { | |
5105aa55 | 2538 | if (ns->ns_id > nn) |
a5768aa8 | 2539 | nvme_ns_remove(ns); |
a5768aa8 KB |
2540 | } |
2541 | list_sort(NULL, &dev->namespaces, ns_cmp); | |
2542 | } | |
2543 | ||
bda4e0fb KB |
2544 | static void nvme_set_irq_hints(struct nvme_dev *dev) |
2545 | { | |
2546 | struct nvme_queue *nvmeq; | |
2547 | int i; | |
2548 | ||
2549 | for (i = 0; i < dev->online_queues; i++) { | |
2550 | nvmeq = dev->queues[i]; | |
2551 | ||
2552 | if (!nvmeq->tags || !(*nvmeq->tags)) | |
2553 | continue; | |
2554 | ||
2555 | irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector, | |
2556 | blk_mq_tags_cpumask(*nvmeq->tags)); | |
2557 | } | |
2558 | } | |
2559 | ||
a5768aa8 KB |
2560 | static void nvme_dev_scan(struct work_struct *work) |
2561 | { | |
2562 | struct nvme_dev *dev = container_of(work, struct nvme_dev, scan_work); | |
2563 | struct nvme_id_ctrl *ctrl; | |
2564 | ||
2565 | if (!dev->tagset.tags) | |
2566 | return; | |
2567 | if (nvme_identify_ctrl(dev, &ctrl)) | |
2568 | return; | |
2569 | nvme_scan_namespaces(dev, le32_to_cpup(&ctrl->nn)); | |
2570 | kfree(ctrl); | |
bda4e0fb | 2571 | nvme_set_irq_hints(dev); |
a5768aa8 KB |
2572 | } |
2573 | ||
422ef0c7 MW |
2574 | /* |
2575 | * Return: error value if an error occurred setting up the queues or calling | |
2576 | * Identify Device. 0 if these succeeded, even if adding some of the | |
2577 | * namespaces failed. At the moment, these failures are silent. TBD which | |
2578 | * failures should be reported. | |
2579 | */ | |
8d85fce7 | 2580 | static int nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 2581 | { |
e75ec752 | 2582 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
c3bfe717 | 2583 | int res; |
51814232 | 2584 | struct nvme_id_ctrl *ctrl; |
159b67d7 | 2585 | int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12; |
b60503ba | 2586 | |
d29ec824 | 2587 | res = nvme_identify_ctrl(dev, &ctrl); |
b60503ba | 2588 | if (res) { |
e75ec752 | 2589 | dev_err(dev->dev, "Identify Controller failed (%d)\n", res); |
e1e5e564 | 2590 | return -EIO; |
b60503ba MW |
2591 | } |
2592 | ||
0e5e4f0e | 2593 | dev->oncs = le16_to_cpup(&ctrl->oncs); |
c30341dc | 2594 | dev->abort_limit = ctrl->acl + 1; |
a7d2ce28 | 2595 | dev->vwc = ctrl->vwc; |
51814232 MW |
2596 | memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn)); |
2597 | memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn)); | |
2598 | memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr)); | |
159b67d7 | 2599 | if (ctrl->mdts) |
8fc23e03 | 2600 | dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9); |
68608c26 | 2601 | if ((pdev->vendor == PCI_VENDOR_ID_INTEL) && |
a4aea562 MB |
2602 | (pdev->device == 0x0953) && ctrl->vs[3]) { |
2603 | unsigned int max_hw_sectors; | |
2604 | ||
159b67d7 | 2605 | dev->stripe_size = 1 << (ctrl->vs[3] + shift); |
a4aea562 MB |
2606 | max_hw_sectors = dev->stripe_size >> (shift - 9); |
2607 | if (dev->max_hw_sectors) { | |
2608 | dev->max_hw_sectors = min(max_hw_sectors, | |
2609 | dev->max_hw_sectors); | |
2610 | } else | |
2611 | dev->max_hw_sectors = max_hw_sectors; | |
2612 | } | |
d29ec824 | 2613 | kfree(ctrl); |
a4aea562 | 2614 | |
ffe7704d KB |
2615 | if (!dev->tagset.tags) { |
2616 | dev->tagset.ops = &nvme_mq_ops; | |
2617 | dev->tagset.nr_hw_queues = dev->online_queues - 1; | |
2618 | dev->tagset.timeout = NVME_IO_TIMEOUT; | |
2619 | dev->tagset.numa_node = dev_to_node(dev->dev); | |
2620 | dev->tagset.queue_depth = | |
a4aea562 | 2621 | min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; |
ffe7704d KB |
2622 | dev->tagset.cmd_size = nvme_cmd_size(dev); |
2623 | dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; | |
2624 | dev->tagset.driver_data = dev; | |
b60503ba | 2625 | |
ffe7704d KB |
2626 | if (blk_mq_alloc_tag_set(&dev->tagset)) |
2627 | return 0; | |
2628 | } | |
a5768aa8 | 2629 | schedule_work(&dev->scan_work); |
e1e5e564 | 2630 | return 0; |
b60503ba MW |
2631 | } |
2632 | ||
0877cb0d KB |
2633 | static int nvme_dev_map(struct nvme_dev *dev) |
2634 | { | |
42f61420 | 2635 | u64 cap; |
0877cb0d | 2636 | int bars, result = -ENOMEM; |
e75ec752 | 2637 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
0877cb0d KB |
2638 | |
2639 | if (pci_enable_device_mem(pdev)) | |
2640 | return result; | |
2641 | ||
2642 | dev->entry[0].vector = pdev->irq; | |
2643 | pci_set_master(pdev); | |
2644 | bars = pci_select_bars(pdev, IORESOURCE_MEM); | |
be7837e8 JA |
2645 | if (!bars) |
2646 | goto disable_pci; | |
2647 | ||
0877cb0d KB |
2648 | if (pci_request_selected_regions(pdev, bars, "nvme")) |
2649 | goto disable_pci; | |
2650 | ||
e75ec752 CH |
2651 | if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) && |
2652 | dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32))) | |
052d0efa | 2653 | goto disable; |
0877cb0d | 2654 | |
0877cb0d KB |
2655 | dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); |
2656 | if (!dev->bar) | |
2657 | goto disable; | |
e32efbfc | 2658 | |
0e53d180 KB |
2659 | if (readl(&dev->bar->csts) == -1) { |
2660 | result = -ENODEV; | |
2661 | goto unmap; | |
2662 | } | |
e32efbfc JA |
2663 | |
2664 | /* | |
2665 | * Some devices don't advertse INTx interrupts, pre-enable a single | |
2666 | * MSIX vec for setup. We'll adjust this later. | |
2667 | */ | |
2668 | if (!pdev->irq) { | |
2669 | result = pci_enable_msix(pdev, dev->entry, 1); | |
2670 | if (result < 0) | |
2671 | goto unmap; | |
2672 | } | |
2673 | ||
42f61420 KB |
2674 | cap = readq(&dev->bar->cap); |
2675 | dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); | |
2676 | dev->db_stride = 1 << NVME_CAP_STRIDE(cap); | |
0877cb0d | 2677 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
8ffaadf7 JD |
2678 | if (readl(&dev->bar->vs) >= NVME_VS(1, 2)) |
2679 | dev->cmb = nvme_map_cmb(dev); | |
0877cb0d KB |
2680 | |
2681 | return 0; | |
2682 | ||
0e53d180 KB |
2683 | unmap: |
2684 | iounmap(dev->bar); | |
2685 | dev->bar = NULL; | |
0877cb0d KB |
2686 | disable: |
2687 | pci_release_regions(pdev); | |
2688 | disable_pci: | |
2689 | pci_disable_device(pdev); | |
2690 | return result; | |
2691 | } | |
2692 | ||
2693 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
2694 | { | |
e75ec752 CH |
2695 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2696 | ||
2697 | if (pdev->msi_enabled) | |
2698 | pci_disable_msi(pdev); | |
2699 | else if (pdev->msix_enabled) | |
2700 | pci_disable_msix(pdev); | |
0877cb0d KB |
2701 | |
2702 | if (dev->bar) { | |
2703 | iounmap(dev->bar); | |
2704 | dev->bar = NULL; | |
e75ec752 | 2705 | pci_release_regions(pdev); |
0877cb0d KB |
2706 | } |
2707 | ||
e75ec752 CH |
2708 | if (pci_is_enabled(pdev)) |
2709 | pci_disable_device(pdev); | |
0877cb0d KB |
2710 | } |
2711 | ||
4d115420 KB |
2712 | struct nvme_delq_ctx { |
2713 | struct task_struct *waiter; | |
2714 | struct kthread_worker *worker; | |
2715 | atomic_t refcount; | |
2716 | }; | |
2717 | ||
2718 | static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev) | |
2719 | { | |
2720 | dq->waiter = current; | |
2721 | mb(); | |
2722 | ||
2723 | for (;;) { | |
2724 | set_current_state(TASK_KILLABLE); | |
2725 | if (!atomic_read(&dq->refcount)) | |
2726 | break; | |
2727 | if (!schedule_timeout(ADMIN_TIMEOUT) || | |
2728 | fatal_signal_pending(current)) { | |
0fb59cbc KB |
2729 | /* |
2730 | * Disable the controller first since we can't trust it | |
2731 | * at this point, but leave the admin queue enabled | |
2732 | * until all queue deletion requests are flushed. | |
2733 | * FIXME: This may take a while if there are more h/w | |
2734 | * queues than admin tags. | |
2735 | */ | |
4d115420 | 2736 | set_current_state(TASK_RUNNING); |
4d115420 | 2737 | nvme_disable_ctrl(dev, readq(&dev->bar->cap)); |
0fb59cbc | 2738 | nvme_clear_queue(dev->queues[0]); |
4d115420 | 2739 | flush_kthread_worker(dq->worker); |
0fb59cbc | 2740 | nvme_disable_queue(dev, 0); |
4d115420 KB |
2741 | return; |
2742 | } | |
2743 | } | |
2744 | set_current_state(TASK_RUNNING); | |
2745 | } | |
2746 | ||
2747 | static void nvme_put_dq(struct nvme_delq_ctx *dq) | |
2748 | { | |
2749 | atomic_dec(&dq->refcount); | |
2750 | if (dq->waiter) | |
2751 | wake_up_process(dq->waiter); | |
2752 | } | |
2753 | ||
2754 | static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq) | |
2755 | { | |
2756 | atomic_inc(&dq->refcount); | |
2757 | return dq; | |
2758 | } | |
2759 | ||
2760 | static void nvme_del_queue_end(struct nvme_queue *nvmeq) | |
2761 | { | |
2762 | struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx; | |
4d115420 KB |
2763 | nvme_put_dq(dq); |
2764 | } | |
2765 | ||
2766 | static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode, | |
2767 | kthread_work_func_t fn) | |
2768 | { | |
2769 | struct nvme_command c; | |
2770 | ||
2771 | memset(&c, 0, sizeof(c)); | |
2772 | c.delete_queue.opcode = opcode; | |
2773 | c.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
2774 | ||
2775 | init_kthread_work(&nvmeq->cmdinfo.work, fn); | |
a4aea562 MB |
2776 | return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo, |
2777 | ADMIN_TIMEOUT); | |
4d115420 KB |
2778 | } |
2779 | ||
2780 | static void nvme_del_cq_work_handler(struct kthread_work *work) | |
2781 | { | |
2782 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2783 | cmdinfo.work); | |
2784 | nvme_del_queue_end(nvmeq); | |
2785 | } | |
2786 | ||
2787 | static int nvme_delete_cq(struct nvme_queue *nvmeq) | |
2788 | { | |
2789 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq, | |
2790 | nvme_del_cq_work_handler); | |
2791 | } | |
2792 | ||
2793 | static void nvme_del_sq_work_handler(struct kthread_work *work) | |
2794 | { | |
2795 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2796 | cmdinfo.work); | |
2797 | int status = nvmeq->cmdinfo.status; | |
2798 | ||
2799 | if (!status) | |
2800 | status = nvme_delete_cq(nvmeq); | |
2801 | if (status) | |
2802 | nvme_del_queue_end(nvmeq); | |
2803 | } | |
2804 | ||
2805 | static int nvme_delete_sq(struct nvme_queue *nvmeq) | |
2806 | { | |
2807 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq, | |
2808 | nvme_del_sq_work_handler); | |
2809 | } | |
2810 | ||
2811 | static void nvme_del_queue_start(struct kthread_work *work) | |
2812 | { | |
2813 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2814 | cmdinfo.work); | |
4d115420 KB |
2815 | if (nvme_delete_sq(nvmeq)) |
2816 | nvme_del_queue_end(nvmeq); | |
2817 | } | |
2818 | ||
2819 | static void nvme_disable_io_queues(struct nvme_dev *dev) | |
2820 | { | |
2821 | int i; | |
2822 | DEFINE_KTHREAD_WORKER_ONSTACK(worker); | |
2823 | struct nvme_delq_ctx dq; | |
2824 | struct task_struct *kworker_task = kthread_run(kthread_worker_fn, | |
2825 | &worker, "nvme%d", dev->instance); | |
2826 | ||
2827 | if (IS_ERR(kworker_task)) { | |
e75ec752 | 2828 | dev_err(dev->dev, |
4d115420 KB |
2829 | "Failed to create queue del task\n"); |
2830 | for (i = dev->queue_count - 1; i > 0; i--) | |
2831 | nvme_disable_queue(dev, i); | |
2832 | return; | |
2833 | } | |
2834 | ||
2835 | dq.waiter = NULL; | |
2836 | atomic_set(&dq.refcount, 0); | |
2837 | dq.worker = &worker; | |
2838 | for (i = dev->queue_count - 1; i > 0; i--) { | |
a4aea562 | 2839 | struct nvme_queue *nvmeq = dev->queues[i]; |
4d115420 KB |
2840 | |
2841 | if (nvme_suspend_queue(nvmeq)) | |
2842 | continue; | |
2843 | nvmeq->cmdinfo.ctx = nvme_get_dq(&dq); | |
2844 | nvmeq->cmdinfo.worker = dq.worker; | |
2845 | init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start); | |
2846 | queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work); | |
2847 | } | |
2848 | nvme_wait_dq(&dq, dev); | |
2849 | kthread_stop(kworker_task); | |
2850 | } | |
2851 | ||
b9afca3e DM |
2852 | /* |
2853 | * Remove the node from the device list and check | |
2854 | * for whether or not we need to stop the nvme_thread. | |
2855 | */ | |
2856 | static void nvme_dev_list_remove(struct nvme_dev *dev) | |
2857 | { | |
2858 | struct task_struct *tmp = NULL; | |
2859 | ||
2860 | spin_lock(&dev_list_lock); | |
2861 | list_del_init(&dev->node); | |
2862 | if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) { | |
2863 | tmp = nvme_thread; | |
2864 | nvme_thread = NULL; | |
2865 | } | |
2866 | spin_unlock(&dev_list_lock); | |
2867 | ||
2868 | if (tmp) | |
2869 | kthread_stop(tmp); | |
2870 | } | |
2871 | ||
c9d3bf88 KB |
2872 | static void nvme_freeze_queues(struct nvme_dev *dev) |
2873 | { | |
2874 | struct nvme_ns *ns; | |
2875 | ||
2876 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2877 | blk_mq_freeze_queue_start(ns->queue); | |
2878 | ||
cddcd72b | 2879 | spin_lock_irq(ns->queue->queue_lock); |
c9d3bf88 | 2880 | queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue); |
cddcd72b | 2881 | spin_unlock_irq(ns->queue->queue_lock); |
c9d3bf88 KB |
2882 | |
2883 | blk_mq_cancel_requeue_work(ns->queue); | |
2884 | blk_mq_stop_hw_queues(ns->queue); | |
2885 | } | |
2886 | } | |
2887 | ||
2888 | static void nvme_unfreeze_queues(struct nvme_dev *dev) | |
2889 | { | |
2890 | struct nvme_ns *ns; | |
2891 | ||
2892 | list_for_each_entry(ns, &dev->namespaces, list) { | |
2893 | queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue); | |
2894 | blk_mq_unfreeze_queue(ns->queue); | |
2895 | blk_mq_start_stopped_hw_queues(ns->queue, true); | |
2896 | blk_mq_kick_requeue_list(ns->queue); | |
2897 | } | |
2898 | } | |
2899 | ||
f0b50732 | 2900 | static void nvme_dev_shutdown(struct nvme_dev *dev) |
b60503ba | 2901 | { |
22404274 | 2902 | int i; |
7c1b2450 | 2903 | u32 csts = -1; |
22404274 | 2904 | |
b9afca3e | 2905 | nvme_dev_list_remove(dev); |
1fa6aead | 2906 | |
c9d3bf88 KB |
2907 | if (dev->bar) { |
2908 | nvme_freeze_queues(dev); | |
7c1b2450 | 2909 | csts = readl(&dev->bar->csts); |
c9d3bf88 | 2910 | } |
7c1b2450 | 2911 | if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) { |
4d115420 | 2912 | for (i = dev->queue_count - 1; i >= 0; i--) { |
a4aea562 | 2913 | struct nvme_queue *nvmeq = dev->queues[i]; |
4d115420 | 2914 | nvme_suspend_queue(nvmeq); |
4d115420 KB |
2915 | } |
2916 | } else { | |
2917 | nvme_disable_io_queues(dev); | |
1894d8f1 | 2918 | nvme_shutdown_ctrl(dev); |
4d115420 KB |
2919 | nvme_disable_queue(dev, 0); |
2920 | } | |
f0b50732 | 2921 | nvme_dev_unmap(dev); |
07836e65 KB |
2922 | |
2923 | for (i = dev->queue_count - 1; i >= 0; i--) | |
2924 | nvme_clear_queue(dev->queues[i]); | |
f0b50732 KB |
2925 | } |
2926 | ||
2927 | static void nvme_dev_remove(struct nvme_dev *dev) | |
2928 | { | |
5105aa55 | 2929 | struct nvme_ns *ns, *next; |
f0b50732 | 2930 | |
5105aa55 | 2931 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) |
a5768aa8 | 2932 | nvme_ns_remove(ns); |
b60503ba MW |
2933 | } |
2934 | ||
091b6092 MW |
2935 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
2936 | { | |
e75ec752 | 2937 | dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, |
091b6092 MW |
2938 | PAGE_SIZE, PAGE_SIZE, 0); |
2939 | if (!dev->prp_page_pool) | |
2940 | return -ENOMEM; | |
2941 | ||
99802a7a | 2942 | /* Optimisation for I/Os between 4k and 128k */ |
e75ec752 | 2943 | dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, |
99802a7a MW |
2944 | 256, 256, 0); |
2945 | if (!dev->prp_small_pool) { | |
2946 | dma_pool_destroy(dev->prp_page_pool); | |
2947 | return -ENOMEM; | |
2948 | } | |
091b6092 MW |
2949 | return 0; |
2950 | } | |
2951 | ||
2952 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
2953 | { | |
2954 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 2955 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
2956 | } |
2957 | ||
cd58ad7d QSA |
2958 | static DEFINE_IDA(nvme_instance_ida); |
2959 | ||
2960 | static int nvme_set_instance(struct nvme_dev *dev) | |
b60503ba | 2961 | { |
cd58ad7d QSA |
2962 | int instance, error; |
2963 | ||
2964 | do { | |
2965 | if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL)) | |
2966 | return -ENODEV; | |
2967 | ||
2968 | spin_lock(&dev_list_lock); | |
2969 | error = ida_get_new(&nvme_instance_ida, &instance); | |
2970 | spin_unlock(&dev_list_lock); | |
2971 | } while (error == -EAGAIN); | |
2972 | ||
2973 | if (error) | |
2974 | return -ENODEV; | |
2975 | ||
2976 | dev->instance = instance; | |
2977 | return 0; | |
b60503ba MW |
2978 | } |
2979 | ||
2980 | static void nvme_release_instance(struct nvme_dev *dev) | |
2981 | { | |
cd58ad7d QSA |
2982 | spin_lock(&dev_list_lock); |
2983 | ida_remove(&nvme_instance_ida, dev->instance); | |
2984 | spin_unlock(&dev_list_lock); | |
b60503ba MW |
2985 | } |
2986 | ||
5e82e952 KB |
2987 | static void nvme_free_dev(struct kref *kref) |
2988 | { | |
2989 | struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref); | |
9ac27090 | 2990 | |
e75ec752 | 2991 | put_device(dev->dev); |
b3fffdef | 2992 | put_device(dev->device); |
285dffc9 | 2993 | nvme_release_instance(dev); |
4af0e21c KB |
2994 | if (dev->tagset.tags) |
2995 | blk_mq_free_tag_set(&dev->tagset); | |
2996 | if (dev->admin_q) | |
2997 | blk_put_queue(dev->admin_q); | |
5e82e952 KB |
2998 | kfree(dev->queues); |
2999 | kfree(dev->entry); | |
3000 | kfree(dev); | |
3001 | } | |
3002 | ||
3003 | static int nvme_dev_open(struct inode *inode, struct file *f) | |
3004 | { | |
b3fffdef KB |
3005 | struct nvme_dev *dev; |
3006 | int instance = iminor(inode); | |
3007 | int ret = -ENODEV; | |
3008 | ||
3009 | spin_lock(&dev_list_lock); | |
3010 | list_for_each_entry(dev, &dev_list, node) { | |
3011 | if (dev->instance == instance) { | |
2e1d8448 KB |
3012 | if (!dev->admin_q) { |
3013 | ret = -EWOULDBLOCK; | |
3014 | break; | |
3015 | } | |
b3fffdef KB |
3016 | if (!kref_get_unless_zero(&dev->kref)) |
3017 | break; | |
3018 | f->private_data = dev; | |
3019 | ret = 0; | |
3020 | break; | |
3021 | } | |
3022 | } | |
3023 | spin_unlock(&dev_list_lock); | |
3024 | ||
3025 | return ret; | |
5e82e952 KB |
3026 | } |
3027 | ||
3028 | static int nvme_dev_release(struct inode *inode, struct file *f) | |
3029 | { | |
3030 | struct nvme_dev *dev = f->private_data; | |
3031 | kref_put(&dev->kref, nvme_free_dev); | |
3032 | return 0; | |
3033 | } | |
3034 | ||
3035 | static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg) | |
3036 | { | |
3037 | struct nvme_dev *dev = f->private_data; | |
a4aea562 MB |
3038 | struct nvme_ns *ns; |
3039 | ||
5e82e952 KB |
3040 | switch (cmd) { |
3041 | case NVME_IOCTL_ADMIN_CMD: | |
a4aea562 | 3042 | return nvme_user_cmd(dev, NULL, (void __user *)arg); |
7963e521 | 3043 | case NVME_IOCTL_IO_CMD: |
a4aea562 MB |
3044 | if (list_empty(&dev->namespaces)) |
3045 | return -ENOTTY; | |
3046 | ns = list_first_entry(&dev->namespaces, struct nvme_ns, list); | |
3047 | return nvme_user_cmd(dev, ns, (void __user *)arg); | |
4cc06521 KB |
3048 | case NVME_IOCTL_RESET: |
3049 | dev_warn(dev->dev, "resetting controller\n"); | |
3050 | return nvme_reset(dev); | |
81f03fed JD |
3051 | case NVME_IOCTL_SUBSYS_RESET: |
3052 | return nvme_subsys_reset(dev); | |
5e82e952 KB |
3053 | default: |
3054 | return -ENOTTY; | |
3055 | } | |
3056 | } | |
3057 | ||
3058 | static const struct file_operations nvme_dev_fops = { | |
3059 | .owner = THIS_MODULE, | |
3060 | .open = nvme_dev_open, | |
3061 | .release = nvme_dev_release, | |
3062 | .unlocked_ioctl = nvme_dev_ioctl, | |
3063 | .compat_ioctl = nvme_dev_ioctl, | |
3064 | }; | |
3065 | ||
3cf519b5 | 3066 | static void nvme_probe_work(struct work_struct *work) |
f0b50732 | 3067 | { |
3cf519b5 | 3068 | struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work); |
b9afca3e | 3069 | bool start_thread = false; |
3cf519b5 | 3070 | int result; |
f0b50732 KB |
3071 | |
3072 | result = nvme_dev_map(dev); | |
3073 | if (result) | |
3cf519b5 | 3074 | goto out; |
f0b50732 KB |
3075 | |
3076 | result = nvme_configure_admin_queue(dev); | |
3077 | if (result) | |
3078 | goto unmap; | |
3079 | ||
3080 | spin_lock(&dev_list_lock); | |
b9afca3e DM |
3081 | if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) { |
3082 | start_thread = true; | |
3083 | nvme_thread = NULL; | |
3084 | } | |
f0b50732 KB |
3085 | list_add(&dev->node, &dev_list); |
3086 | spin_unlock(&dev_list_lock); | |
3087 | ||
b9afca3e DM |
3088 | if (start_thread) { |
3089 | nvme_thread = kthread_run(nvme_kthread, NULL, "nvme"); | |
387caa5a | 3090 | wake_up_all(&nvme_kthread_wait); |
b9afca3e DM |
3091 | } else |
3092 | wait_event_killable(nvme_kthread_wait, nvme_thread); | |
3093 | ||
3094 | if (IS_ERR_OR_NULL(nvme_thread)) { | |
3095 | result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR; | |
3096 | goto disable; | |
3097 | } | |
a4aea562 MB |
3098 | |
3099 | nvme_init_queue(dev->queues[0], 0); | |
0fb59cbc KB |
3100 | result = nvme_alloc_admin_tags(dev); |
3101 | if (result) | |
3102 | goto disable; | |
b9afca3e | 3103 | |
f0b50732 | 3104 | result = nvme_setup_io_queues(dev); |
badc34d4 | 3105 | if (result) |
0fb59cbc | 3106 | goto free_tags; |
f0b50732 | 3107 | |
1efccc9d | 3108 | dev->event_limit = 1; |
3cf519b5 | 3109 | |
2659e57b CH |
3110 | /* |
3111 | * Keep the controller around but remove all namespaces if we don't have | |
3112 | * any working I/O queue. | |
3113 | */ | |
3cf519b5 CH |
3114 | if (dev->online_queues < 2) { |
3115 | dev_warn(dev->dev, "IO queues not created\n"); | |
3cf519b5 CH |
3116 | nvme_dev_remove(dev); |
3117 | } else { | |
3118 | nvme_unfreeze_queues(dev); | |
3119 | nvme_dev_add(dev); | |
3120 | } | |
3121 | ||
3122 | return; | |
f0b50732 | 3123 | |
0fb59cbc KB |
3124 | free_tags: |
3125 | nvme_dev_remove_admin(dev); | |
4af0e21c KB |
3126 | blk_put_queue(dev->admin_q); |
3127 | dev->admin_q = NULL; | |
3128 | dev->queues[0]->tags = NULL; | |
f0b50732 | 3129 | disable: |
a1a5ef99 | 3130 | nvme_disable_queue(dev, 0); |
b9afca3e | 3131 | nvme_dev_list_remove(dev); |
f0b50732 KB |
3132 | unmap: |
3133 | nvme_dev_unmap(dev); | |
3cf519b5 CH |
3134 | out: |
3135 | if (!work_busy(&dev->reset_work)) | |
3136 | nvme_dead_ctrl(dev); | |
f0b50732 KB |
3137 | } |
3138 | ||
9a6b9458 KB |
3139 | static int nvme_remove_dead_ctrl(void *arg) |
3140 | { | |
3141 | struct nvme_dev *dev = (struct nvme_dev *)arg; | |
e75ec752 | 3142 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
9a6b9458 KB |
3143 | |
3144 | if (pci_get_drvdata(pdev)) | |
c81f4975 | 3145 | pci_stop_and_remove_bus_device_locked(pdev); |
9a6b9458 KB |
3146 | kref_put(&dev->kref, nvme_free_dev); |
3147 | return 0; | |
3148 | } | |
3149 | ||
de3eff2b KB |
3150 | static void nvme_dead_ctrl(struct nvme_dev *dev) |
3151 | { | |
3152 | dev_warn(dev->dev, "Device failed to resume\n"); | |
3153 | kref_get(&dev->kref); | |
3154 | if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d", | |
3155 | dev->instance))) { | |
3156 | dev_err(dev->dev, | |
3157 | "Failed to start controller remove task\n"); | |
3158 | kref_put(&dev->kref, nvme_free_dev); | |
3159 | } | |
3160 | } | |
3161 | ||
77b50d9e | 3162 | static void nvme_reset_work(struct work_struct *ws) |
9a6b9458 | 3163 | { |
77b50d9e | 3164 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); |
ffe7704d KB |
3165 | bool in_probe = work_busy(&dev->probe_work); |
3166 | ||
9a6b9458 | 3167 | nvme_dev_shutdown(dev); |
ffe7704d KB |
3168 | |
3169 | /* Synchronize with device probe so that work will see failure status | |
3170 | * and exit gracefully without trying to schedule another reset */ | |
3171 | flush_work(&dev->probe_work); | |
3172 | ||
3173 | /* Fail this device if reset occured during probe to avoid | |
3174 | * infinite initialization loops. */ | |
3175 | if (in_probe) { | |
de3eff2b | 3176 | nvme_dead_ctrl(dev); |
ffe7704d | 3177 | return; |
9a6b9458 | 3178 | } |
ffe7704d KB |
3179 | /* Schedule device resume asynchronously so the reset work is available |
3180 | * to cleanup errors that may occur during reinitialization */ | |
3181 | schedule_work(&dev->probe_work); | |
9a6b9458 KB |
3182 | } |
3183 | ||
90667892 | 3184 | static int __nvme_reset(struct nvme_dev *dev) |
9ca97374 | 3185 | { |
90667892 CH |
3186 | if (work_pending(&dev->reset_work)) |
3187 | return -EBUSY; | |
3188 | list_del_init(&dev->node); | |
3189 | queue_work(nvme_workq, &dev->reset_work); | |
3190 | return 0; | |
9ca97374 TH |
3191 | } |
3192 | ||
4cc06521 KB |
3193 | static int nvme_reset(struct nvme_dev *dev) |
3194 | { | |
90667892 | 3195 | int ret; |
4cc06521 KB |
3196 | |
3197 | if (!dev->admin_q || blk_queue_dying(dev->admin_q)) | |
3198 | return -ENODEV; | |
3199 | ||
3200 | spin_lock(&dev_list_lock); | |
90667892 | 3201 | ret = __nvme_reset(dev); |
4cc06521 KB |
3202 | spin_unlock(&dev_list_lock); |
3203 | ||
3204 | if (!ret) { | |
3205 | flush_work(&dev->reset_work); | |
ffe7704d | 3206 | flush_work(&dev->probe_work); |
4cc06521 KB |
3207 | return 0; |
3208 | } | |
3209 | ||
3210 | return ret; | |
3211 | } | |
3212 | ||
3213 | static ssize_t nvme_sysfs_reset(struct device *dev, | |
3214 | struct device_attribute *attr, const char *buf, | |
3215 | size_t count) | |
3216 | { | |
3217 | struct nvme_dev *ndev = dev_get_drvdata(dev); | |
3218 | int ret; | |
3219 | ||
3220 | ret = nvme_reset(ndev); | |
3221 | if (ret < 0) | |
3222 | return ret; | |
3223 | ||
3224 | return count; | |
3225 | } | |
3226 | static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset); | |
3227 | ||
8d85fce7 | 3228 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 3229 | { |
a4aea562 | 3230 | int node, result = -ENOMEM; |
b60503ba MW |
3231 | struct nvme_dev *dev; |
3232 | ||
a4aea562 MB |
3233 | node = dev_to_node(&pdev->dev); |
3234 | if (node == NUMA_NO_NODE) | |
3235 | set_dev_node(&pdev->dev, 0); | |
3236 | ||
3237 | dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); | |
b60503ba MW |
3238 | if (!dev) |
3239 | return -ENOMEM; | |
a4aea562 MB |
3240 | dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry), |
3241 | GFP_KERNEL, node); | |
b60503ba MW |
3242 | if (!dev->entry) |
3243 | goto free; | |
a4aea562 MB |
3244 | dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *), |
3245 | GFP_KERNEL, node); | |
b60503ba MW |
3246 | if (!dev->queues) |
3247 | goto free; | |
3248 | ||
3249 | INIT_LIST_HEAD(&dev->namespaces); | |
77b50d9e | 3250 | INIT_WORK(&dev->reset_work, nvme_reset_work); |
e75ec752 | 3251 | dev->dev = get_device(&pdev->dev); |
9a6b9458 | 3252 | pci_set_drvdata(pdev, dev); |
cd58ad7d QSA |
3253 | result = nvme_set_instance(dev); |
3254 | if (result) | |
a96d4f5c | 3255 | goto put_pci; |
b60503ba | 3256 | |
091b6092 MW |
3257 | result = nvme_setup_prp_pools(dev); |
3258 | if (result) | |
0877cb0d | 3259 | goto release; |
091b6092 | 3260 | |
fb35e914 | 3261 | kref_init(&dev->kref); |
b3fffdef KB |
3262 | dev->device = device_create(nvme_class, &pdev->dev, |
3263 | MKDEV(nvme_char_major, dev->instance), | |
3264 | dev, "nvme%d", dev->instance); | |
3265 | if (IS_ERR(dev->device)) { | |
3266 | result = PTR_ERR(dev->device); | |
2e1d8448 | 3267 | goto release_pools; |
b3fffdef KB |
3268 | } |
3269 | get_device(dev->device); | |
4cc06521 KB |
3270 | dev_set_drvdata(dev->device, dev); |
3271 | ||
3272 | result = device_create_file(dev->device, &dev_attr_reset_controller); | |
3273 | if (result) | |
3274 | goto put_dev; | |
740216fc | 3275 | |
e6e96d73 | 3276 | INIT_LIST_HEAD(&dev->node); |
a5768aa8 | 3277 | INIT_WORK(&dev->scan_work, nvme_dev_scan); |
3cf519b5 | 3278 | INIT_WORK(&dev->probe_work, nvme_probe_work); |
2e1d8448 | 3279 | schedule_work(&dev->probe_work); |
b60503ba MW |
3280 | return 0; |
3281 | ||
4cc06521 KB |
3282 | put_dev: |
3283 | device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance)); | |
3284 | put_device(dev->device); | |
0877cb0d | 3285 | release_pools: |
091b6092 | 3286 | nvme_release_prp_pools(dev); |
0877cb0d KB |
3287 | release: |
3288 | nvme_release_instance(dev); | |
a96d4f5c | 3289 | put_pci: |
e75ec752 | 3290 | put_device(dev->dev); |
b60503ba MW |
3291 | free: |
3292 | kfree(dev->queues); | |
3293 | kfree(dev->entry); | |
3294 | kfree(dev); | |
3295 | return result; | |
3296 | } | |
3297 | ||
f0d54a54 KB |
3298 | static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) |
3299 | { | |
a6739479 | 3300 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
f0d54a54 | 3301 | |
a6739479 KB |
3302 | if (prepare) |
3303 | nvme_dev_shutdown(dev); | |
3304 | else | |
0a7385ad | 3305 | schedule_work(&dev->probe_work); |
f0d54a54 KB |
3306 | } |
3307 | ||
09ece142 KB |
3308 | static void nvme_shutdown(struct pci_dev *pdev) |
3309 | { | |
3310 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
3311 | nvme_dev_shutdown(dev); | |
3312 | } | |
3313 | ||
8d85fce7 | 3314 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
3315 | { |
3316 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 KB |
3317 | |
3318 | spin_lock(&dev_list_lock); | |
3319 | list_del_init(&dev->node); | |
3320 | spin_unlock(&dev_list_lock); | |
3321 | ||
3322 | pci_set_drvdata(pdev, NULL); | |
2e1d8448 | 3323 | flush_work(&dev->probe_work); |
9a6b9458 | 3324 | flush_work(&dev->reset_work); |
a5768aa8 | 3325 | flush_work(&dev->scan_work); |
4cc06521 | 3326 | device_remove_file(dev->device, &dev_attr_reset_controller); |
c9d3bf88 | 3327 | nvme_dev_remove(dev); |
3399a3f7 | 3328 | nvme_dev_shutdown(dev); |
a4aea562 | 3329 | nvme_dev_remove_admin(dev); |
b3fffdef | 3330 | device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance)); |
a1a5ef99 | 3331 | nvme_free_queues(dev, 0); |
8ffaadf7 | 3332 | nvme_release_cmb(dev); |
9a6b9458 | 3333 | nvme_release_prp_pools(dev); |
5e82e952 | 3334 | kref_put(&dev->kref, nvme_free_dev); |
b60503ba MW |
3335 | } |
3336 | ||
3337 | /* These functions are yet to be implemented */ | |
3338 | #define nvme_error_detected NULL | |
3339 | #define nvme_dump_registers NULL | |
3340 | #define nvme_link_reset NULL | |
3341 | #define nvme_slot_reset NULL | |
3342 | #define nvme_error_resume NULL | |
cd638946 | 3343 | |
671a6018 | 3344 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
3345 | static int nvme_suspend(struct device *dev) |
3346 | { | |
3347 | struct pci_dev *pdev = to_pci_dev(dev); | |
3348 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
3349 | ||
3350 | nvme_dev_shutdown(ndev); | |
3351 | return 0; | |
3352 | } | |
3353 | ||
3354 | static int nvme_resume(struct device *dev) | |
3355 | { | |
3356 | struct pci_dev *pdev = to_pci_dev(dev); | |
3357 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 3358 | |
0a7385ad | 3359 | schedule_work(&ndev->probe_work); |
9a6b9458 | 3360 | return 0; |
cd638946 | 3361 | } |
671a6018 | 3362 | #endif |
cd638946 KB |
3363 | |
3364 | static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); | |
b60503ba | 3365 | |
1d352035 | 3366 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba MW |
3367 | .error_detected = nvme_error_detected, |
3368 | .mmio_enabled = nvme_dump_registers, | |
3369 | .link_reset = nvme_link_reset, | |
3370 | .slot_reset = nvme_slot_reset, | |
3371 | .resume = nvme_error_resume, | |
f0d54a54 | 3372 | .reset_notify = nvme_reset_notify, |
b60503ba MW |
3373 | }; |
3374 | ||
3375 | /* Move to pci_ids.h later */ | |
3376 | #define PCI_CLASS_STORAGE_EXPRESS 0x010802 | |
3377 | ||
6eb0d698 | 3378 | static const struct pci_device_id nvme_id_table[] = { |
b60503ba MW |
3379 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
3380 | { 0, } | |
3381 | }; | |
3382 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
3383 | ||
3384 | static struct pci_driver nvme_driver = { | |
3385 | .name = "nvme", | |
3386 | .id_table = nvme_id_table, | |
3387 | .probe = nvme_probe, | |
8d85fce7 | 3388 | .remove = nvme_remove, |
09ece142 | 3389 | .shutdown = nvme_shutdown, |
cd638946 KB |
3390 | .driver = { |
3391 | .pm = &nvme_dev_pm_ops, | |
3392 | }, | |
b60503ba MW |
3393 | .err_handler = &nvme_err_handler, |
3394 | }; | |
3395 | ||
3396 | static int __init nvme_init(void) | |
3397 | { | |
0ac13140 | 3398 | int result; |
1fa6aead | 3399 | |
b9afca3e | 3400 | init_waitqueue_head(&nvme_kthread_wait); |
b60503ba | 3401 | |
9a6b9458 KB |
3402 | nvme_workq = create_singlethread_workqueue("nvme"); |
3403 | if (!nvme_workq) | |
b9afca3e | 3404 | return -ENOMEM; |
9a6b9458 | 3405 | |
5c42ea16 KB |
3406 | result = register_blkdev(nvme_major, "nvme"); |
3407 | if (result < 0) | |
9a6b9458 | 3408 | goto kill_workq; |
5c42ea16 | 3409 | else if (result > 0) |
0ac13140 | 3410 | nvme_major = result; |
b60503ba | 3411 | |
b3fffdef KB |
3412 | result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme", |
3413 | &nvme_dev_fops); | |
3414 | if (result < 0) | |
3415 | goto unregister_blkdev; | |
3416 | else if (result > 0) | |
3417 | nvme_char_major = result; | |
3418 | ||
3419 | nvme_class = class_create(THIS_MODULE, "nvme"); | |
c727040b AK |
3420 | if (IS_ERR(nvme_class)) { |
3421 | result = PTR_ERR(nvme_class); | |
b3fffdef | 3422 | goto unregister_chrdev; |
c727040b | 3423 | } |
b3fffdef | 3424 | |
f3db22fe KB |
3425 | result = pci_register_driver(&nvme_driver); |
3426 | if (result) | |
b3fffdef | 3427 | goto destroy_class; |
1fa6aead | 3428 | return 0; |
b60503ba | 3429 | |
b3fffdef KB |
3430 | destroy_class: |
3431 | class_destroy(nvme_class); | |
3432 | unregister_chrdev: | |
3433 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
1fa6aead | 3434 | unregister_blkdev: |
b60503ba | 3435 | unregister_blkdev(nvme_major, "nvme"); |
9a6b9458 KB |
3436 | kill_workq: |
3437 | destroy_workqueue(nvme_workq); | |
b60503ba MW |
3438 | return result; |
3439 | } | |
3440 | ||
3441 | static void __exit nvme_exit(void) | |
3442 | { | |
3443 | pci_unregister_driver(&nvme_driver); | |
3444 | unregister_blkdev(nvme_major, "nvme"); | |
9a6b9458 | 3445 | destroy_workqueue(nvme_workq); |
b3fffdef KB |
3446 | class_destroy(nvme_class); |
3447 | __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme"); | |
b9afca3e | 3448 | BUG_ON(nvme_thread && !IS_ERR(nvme_thread)); |
21bd78bc | 3449 | _nvme_check_size(); |
b60503ba MW |
3450 | } |
3451 | ||
3452 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
3453 | MODULE_LICENSE("GPL"); | |
c78b4713 | 3454 | MODULE_VERSION("1.0"); |
b60503ba MW |
3455 | module_init(nvme_init); |
3456 | module_exit(nvme_exit); |