Commit | Line | Data |
---|---|---|
b60503ba MW |
1 | /* |
2 | * NVM Express device driver | |
6eb0d698 | 3 | * Copyright (c) 2011-2014, Intel Corporation. |
b60503ba MW |
4 | * |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
b60503ba MW |
13 | */ |
14 | ||
15 | #include <linux/nvme.h> | |
16 | #include <linux/bio.h> | |
8de05535 | 17 | #include <linux/bitops.h> |
b60503ba | 18 | #include <linux/blkdev.h> |
42f61420 | 19 | #include <linux/cpu.h> |
fd63e9ce | 20 | #include <linux/delay.h> |
b60503ba MW |
21 | #include <linux/errno.h> |
22 | #include <linux/fs.h> | |
23 | #include <linux/genhd.h> | |
4cc09e2d | 24 | #include <linux/hdreg.h> |
5aff9382 | 25 | #include <linux/idr.h> |
b60503ba MW |
26 | #include <linux/init.h> |
27 | #include <linux/interrupt.h> | |
28 | #include <linux/io.h> | |
29 | #include <linux/kdev_t.h> | |
1fa6aead | 30 | #include <linux/kthread.h> |
b60503ba MW |
31 | #include <linux/kernel.h> |
32 | #include <linux/mm.h> | |
33 | #include <linux/module.h> | |
34 | #include <linux/moduleparam.h> | |
35 | #include <linux/pci.h> | |
42f61420 | 36 | #include <linux/percpu.h> |
be7b6275 | 37 | #include <linux/poison.h> |
c3bfe717 | 38 | #include <linux/ptrace.h> |
b60503ba MW |
39 | #include <linux/sched.h> |
40 | #include <linux/slab.h> | |
41 | #include <linux/types.h> | |
5d0f6131 | 42 | #include <scsi/sg.h> |
797a796a HM |
43 | #include <asm-generic/io-64-nonatomic-lo-hi.h> |
44 | ||
3291fa57 KB |
45 | #include <trace/events/block.h> |
46 | ||
9d43cf64 | 47 | #define NVME_Q_DEPTH 1024 |
b60503ba MW |
48 | #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) |
49 | #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) | |
9d43cf64 KB |
50 | #define ADMIN_TIMEOUT (admin_timeout * HZ) |
51 | #define IOD_TIMEOUT (retry_time * HZ) | |
52 | ||
53 | static unsigned char admin_timeout = 60; | |
54 | module_param(admin_timeout, byte, 0644); | |
55 | MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); | |
b60503ba | 56 | |
bd67608a MW |
57 | unsigned char nvme_io_timeout = 30; |
58 | module_param_named(io_timeout, nvme_io_timeout, byte, 0644); | |
b355084a | 59 | MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); |
b60503ba | 60 | |
61e4ce08 KB |
61 | static unsigned char retry_time = 30; |
62 | module_param(retry_time, byte, 0644); | |
63 | MODULE_PARM_DESC(retry_time, "time in seconds to retry failed I/O"); | |
64 | ||
b60503ba MW |
65 | static int nvme_major; |
66 | module_param(nvme_major, int, 0); | |
67 | ||
58ffacb5 MW |
68 | static int use_threaded_interrupts; |
69 | module_param(use_threaded_interrupts, int, 0); | |
70 | ||
1fa6aead MW |
71 | static DEFINE_SPINLOCK(dev_list_lock); |
72 | static LIST_HEAD(dev_list); | |
73 | static struct task_struct *nvme_thread; | |
9a6b9458 | 74 | static struct workqueue_struct *nvme_workq; |
b9afca3e | 75 | static wait_queue_head_t nvme_kthread_wait; |
f3db22fe | 76 | static struct notifier_block nvme_nb; |
1fa6aead | 77 | |
d4b4ff8e KB |
78 | static void nvme_reset_failed_dev(struct work_struct *ws); |
79 | ||
4d115420 KB |
80 | struct async_cmd_info { |
81 | struct kthread_work work; | |
82 | struct kthread_worker *worker; | |
83 | u32 result; | |
84 | int status; | |
85 | void *ctx; | |
86 | }; | |
1fa6aead | 87 | |
b60503ba MW |
88 | /* |
89 | * An NVM Express queue. Each device has at least two (one for admin | |
90 | * commands and one for I/O commands). | |
91 | */ | |
92 | struct nvme_queue { | |
5a92e700 | 93 | struct rcu_head r_head; |
b60503ba | 94 | struct device *q_dmadev; |
091b6092 | 95 | struct nvme_dev *dev; |
3193f07b | 96 | char irqname[24]; /* nvme4294967295-65535\0 */ |
b60503ba MW |
97 | spinlock_t q_lock; |
98 | struct nvme_command *sq_cmds; | |
99 | volatile struct nvme_completion *cqes; | |
100 | dma_addr_t sq_dma_addr; | |
101 | dma_addr_t cq_dma_addr; | |
102 | wait_queue_head_t sq_full; | |
1fa6aead | 103 | wait_queue_t sq_cong_wait; |
b60503ba | 104 | struct bio_list sq_cong; |
edd10d33 | 105 | struct list_head iod_bio; |
b60503ba MW |
106 | u32 __iomem *q_db; |
107 | u16 q_depth; | |
108 | u16 cq_vector; | |
109 | u16 sq_head; | |
110 | u16 sq_tail; | |
111 | u16 cq_head; | |
c30341dc | 112 | u16 qid; |
e9539f47 MW |
113 | u8 cq_phase; |
114 | u8 cqe_seen; | |
22404274 | 115 | u8 q_suspended; |
42f61420 | 116 | cpumask_var_t cpu_mask; |
4d115420 | 117 | struct async_cmd_info cmdinfo; |
b60503ba MW |
118 | unsigned long cmdid_data[]; |
119 | }; | |
120 | ||
121 | /* | |
122 | * Check we didin't inadvertently grow the command struct | |
123 | */ | |
124 | static inline void _nvme_check_size(void) | |
125 | { | |
126 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
127 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); | |
128 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
129 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
130 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
f8ebf840 | 131 | BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); |
c30341dc | 132 | BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); |
b60503ba MW |
133 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); |
134 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); | |
135 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); | |
136 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
6ecec745 | 137 | BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); |
b60503ba MW |
138 | } |
139 | ||
edd10d33 | 140 | typedef void (*nvme_completion_fn)(struct nvme_queue *, void *, |
c2f5b650 MW |
141 | struct nvme_completion *); |
142 | ||
e85248e5 | 143 | struct nvme_cmd_info { |
c2f5b650 MW |
144 | nvme_completion_fn fn; |
145 | void *ctx; | |
e85248e5 | 146 | unsigned long timeout; |
c30341dc | 147 | int aborted; |
e85248e5 MW |
148 | }; |
149 | ||
150 | static struct nvme_cmd_info *nvme_cmd_info(struct nvme_queue *nvmeq) | |
151 | { | |
152 | return (void *)&nvmeq->cmdid_data[BITS_TO_LONGS(nvmeq->q_depth)]; | |
153 | } | |
154 | ||
22404274 KB |
155 | static unsigned nvme_queue_extra(int depth) |
156 | { | |
157 | return DIV_ROUND_UP(depth, 8) + (depth * sizeof(struct nvme_cmd_info)); | |
158 | } | |
159 | ||
b60503ba | 160 | /** |
714a7a22 MW |
161 | * alloc_cmdid() - Allocate a Command ID |
162 | * @nvmeq: The queue that will be used for this command | |
163 | * @ctx: A pointer that will be passed to the handler | |
c2f5b650 | 164 | * @handler: The function to call on completion |
b60503ba MW |
165 | * |
166 | * Allocate a Command ID for a queue. The data passed in will | |
167 | * be passed to the completion handler. This is implemented by using | |
168 | * the bottom two bits of the ctx pointer to store the handler ID. | |
169 | * Passing in a pointer that's not 4-byte aligned will cause a BUG. | |
170 | * We can change this if it becomes a problem. | |
184d2944 MW |
171 | * |
172 | * May be called with local interrupts disabled and the q_lock held, | |
173 | * or with interrupts enabled and no locks held. | |
b60503ba | 174 | */ |
c2f5b650 MW |
175 | static int alloc_cmdid(struct nvme_queue *nvmeq, void *ctx, |
176 | nvme_completion_fn handler, unsigned timeout) | |
b60503ba | 177 | { |
e6d15f79 | 178 | int depth = nvmeq->q_depth - 1; |
e85248e5 | 179 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); |
b60503ba MW |
180 | int cmdid; |
181 | ||
b60503ba MW |
182 | do { |
183 | cmdid = find_first_zero_bit(nvmeq->cmdid_data, depth); | |
184 | if (cmdid >= depth) | |
185 | return -EBUSY; | |
186 | } while (test_and_set_bit(cmdid, nvmeq->cmdid_data)); | |
187 | ||
c2f5b650 MW |
188 | info[cmdid].fn = handler; |
189 | info[cmdid].ctx = ctx; | |
e85248e5 | 190 | info[cmdid].timeout = jiffies + timeout; |
c30341dc | 191 | info[cmdid].aborted = 0; |
b60503ba MW |
192 | return cmdid; |
193 | } | |
194 | ||
195 | static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx, | |
c2f5b650 | 196 | nvme_completion_fn handler, unsigned timeout) |
b60503ba MW |
197 | { |
198 | int cmdid; | |
199 | wait_event_killable(nvmeq->sq_full, | |
e85248e5 | 200 | (cmdid = alloc_cmdid(nvmeq, ctx, handler, timeout)) >= 0); |
b60503ba MW |
201 | return (cmdid < 0) ? -EINTR : cmdid; |
202 | } | |
203 | ||
c2f5b650 MW |
204 | /* Special values must be less than 0x1000 */ |
205 | #define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA) | |
d2d87034 MW |
206 | #define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE) |
207 | #define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE) | |
208 | #define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE) | |
53562be7 | 209 | #define CMD_CTX_ABORT (0x318 + CMD_CTX_BASE) |
6fccf938 | 210 | #define CMD_CTX_ASYNC (0x31C + CMD_CTX_BASE) |
be7b6275 | 211 | |
edd10d33 | 212 | static void special_completion(struct nvme_queue *nvmeq, void *ctx, |
c2f5b650 MW |
213 | struct nvme_completion *cqe) |
214 | { | |
215 | if (ctx == CMD_CTX_CANCELLED) | |
216 | return; | |
c30341dc | 217 | if (ctx == CMD_CTX_ABORT) { |
edd10d33 | 218 | ++nvmeq->dev->abort_limit; |
c30341dc KB |
219 | return; |
220 | } | |
c2f5b650 | 221 | if (ctx == CMD_CTX_COMPLETED) { |
edd10d33 | 222 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
223 | "completed id %d twice on queue %d\n", |
224 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
225 | return; | |
226 | } | |
227 | if (ctx == CMD_CTX_INVALID) { | |
edd10d33 | 228 | dev_warn(nvmeq->q_dmadev, |
c2f5b650 MW |
229 | "invalid id %d completed on queue %d\n", |
230 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
231 | return; | |
232 | } | |
6fccf938 KB |
233 | if (ctx == CMD_CTX_ASYNC) { |
234 | u32 result = le32_to_cpup(&cqe->result); | |
235 | u16 status = le16_to_cpup(&cqe->status) >> 1; | |
236 | ||
237 | if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) | |
238 | ++nvmeq->dev->event_limit; | |
239 | if (status == NVME_SC_SUCCESS) | |
240 | dev_warn(nvmeq->q_dmadev, | |
241 | "async event result %08x\n", result); | |
242 | return; | |
243 | } | |
c2f5b650 | 244 | |
edd10d33 | 245 | dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx); |
c2f5b650 MW |
246 | } |
247 | ||
edd10d33 | 248 | static void async_completion(struct nvme_queue *nvmeq, void *ctx, |
4d115420 KB |
249 | struct nvme_completion *cqe) |
250 | { | |
251 | struct async_cmd_info *cmdinfo = ctx; | |
252 | cmdinfo->result = le32_to_cpup(&cqe->result); | |
253 | cmdinfo->status = le16_to_cpup(&cqe->status) >> 1; | |
254 | queue_kthread_work(cmdinfo->worker, &cmdinfo->work); | |
255 | } | |
256 | ||
184d2944 MW |
257 | /* |
258 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
259 | */ | |
c2f5b650 MW |
260 | static void *free_cmdid(struct nvme_queue *nvmeq, int cmdid, |
261 | nvme_completion_fn *fn) | |
b60503ba | 262 | { |
c2f5b650 | 263 | void *ctx; |
e85248e5 | 264 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); |
b60503ba | 265 | |
94bbac40 KB |
266 | if (cmdid >= nvmeq->q_depth || !info[cmdid].fn) { |
267 | if (fn) | |
268 | *fn = special_completion; | |
48e3d398 | 269 | return CMD_CTX_INVALID; |
c2f5b650 | 270 | } |
859361a2 KB |
271 | if (fn) |
272 | *fn = info[cmdid].fn; | |
c2f5b650 MW |
273 | ctx = info[cmdid].ctx; |
274 | info[cmdid].fn = special_completion; | |
e85248e5 | 275 | info[cmdid].ctx = CMD_CTX_COMPLETED; |
b60503ba MW |
276 | clear_bit(cmdid, nvmeq->cmdid_data); |
277 | wake_up(&nvmeq->sq_full); | |
c2f5b650 | 278 | return ctx; |
b60503ba MW |
279 | } |
280 | ||
c2f5b650 MW |
281 | static void *cancel_cmdid(struct nvme_queue *nvmeq, int cmdid, |
282 | nvme_completion_fn *fn) | |
3c0cf138 | 283 | { |
c2f5b650 | 284 | void *ctx; |
e85248e5 | 285 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); |
c2f5b650 MW |
286 | if (fn) |
287 | *fn = info[cmdid].fn; | |
288 | ctx = info[cmdid].ctx; | |
289 | info[cmdid].fn = special_completion; | |
e85248e5 | 290 | info[cmdid].ctx = CMD_CTX_CANCELLED; |
c2f5b650 | 291 | return ctx; |
3c0cf138 MW |
292 | } |
293 | ||
5a92e700 | 294 | static struct nvme_queue *raw_nvmeq(struct nvme_dev *dev, int qid) |
b60503ba | 295 | { |
5a92e700 | 296 | return rcu_dereference_raw(dev->queues[qid]); |
b60503ba MW |
297 | } |
298 | ||
4f5099af | 299 | static struct nvme_queue *get_nvmeq(struct nvme_dev *dev) __acquires(RCU) |
5a92e700 | 300 | { |
a51afb54 | 301 | struct nvme_queue *nvmeq; |
42f61420 | 302 | unsigned queue_id = get_cpu_var(*dev->io_queue); |
a51afb54 | 303 | |
5a92e700 | 304 | rcu_read_lock(); |
a51afb54 KB |
305 | nvmeq = rcu_dereference(dev->queues[queue_id]); |
306 | if (nvmeq) | |
307 | return nvmeq; | |
308 | ||
309 | rcu_read_unlock(); | |
310 | put_cpu_var(*dev->io_queue); | |
311 | return NULL; | |
5a92e700 KB |
312 | } |
313 | ||
4f5099af | 314 | static void put_nvmeq(struct nvme_queue *nvmeq) __releases(RCU) |
b60503ba | 315 | { |
5a92e700 | 316 | rcu_read_unlock(); |
42f61420 | 317 | put_cpu_var(nvmeq->dev->io_queue); |
b60503ba MW |
318 | } |
319 | ||
4f5099af KB |
320 | static struct nvme_queue *lock_nvmeq(struct nvme_dev *dev, int q_idx) |
321 | __acquires(RCU) | |
b60503ba | 322 | { |
a51afb54 KB |
323 | struct nvme_queue *nvmeq; |
324 | ||
4f5099af | 325 | rcu_read_lock(); |
a51afb54 KB |
326 | nvmeq = rcu_dereference(dev->queues[q_idx]); |
327 | if (nvmeq) | |
328 | return nvmeq; | |
329 | ||
330 | rcu_read_unlock(); | |
331 | return NULL; | |
4f5099af KB |
332 | } |
333 | ||
334 | static void unlock_nvmeq(struct nvme_queue *nvmeq) __releases(RCU) | |
335 | { | |
336 | rcu_read_unlock(); | |
b60503ba MW |
337 | } |
338 | ||
339 | /** | |
714a7a22 | 340 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
341 | * @nvmeq: The queue to use |
342 | * @cmd: The command to send | |
343 | * | |
344 | * Safe to use from interrupt context | |
345 | */ | |
346 | static int nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd) | |
347 | { | |
348 | unsigned long flags; | |
349 | u16 tail; | |
b60503ba | 350 | spin_lock_irqsave(&nvmeq->q_lock, flags); |
4f5099af KB |
351 | if (nvmeq->q_suspended) { |
352 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); | |
353 | return -EBUSY; | |
354 | } | |
b60503ba MW |
355 | tail = nvmeq->sq_tail; |
356 | memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); | |
b60503ba MW |
357 | if (++tail == nvmeq->q_depth) |
358 | tail = 0; | |
7547881d | 359 | writel(tail, nvmeq->q_db); |
b60503ba MW |
360 | nvmeq->sq_tail = tail; |
361 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); | |
362 | ||
363 | return 0; | |
364 | } | |
365 | ||
eca18b23 | 366 | static __le64 **iod_list(struct nvme_iod *iod) |
e025344c | 367 | { |
eca18b23 | 368 | return ((void *)iod) + iod->offset; |
e025344c SMM |
369 | } |
370 | ||
eca18b23 MW |
371 | /* |
372 | * Will slightly overestimate the number of pages needed. This is OK | |
373 | * as it only leads to a small amount of wasted memory for the lifetime of | |
374 | * the I/O. | |
375 | */ | |
1d090624 | 376 | static int nvme_npages(unsigned size, struct nvme_dev *dev) |
eca18b23 | 377 | { |
1d090624 KB |
378 | unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size); |
379 | return DIV_ROUND_UP(8 * nprps, dev->page_size - 8); | |
eca18b23 | 380 | } |
b60503ba | 381 | |
eca18b23 | 382 | static struct nvme_iod * |
1d090624 | 383 | nvme_alloc_iod(unsigned nseg, unsigned nbytes, struct nvme_dev *dev, gfp_t gfp) |
b60503ba | 384 | { |
eca18b23 | 385 | struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) + |
1d090624 | 386 | sizeof(__le64 *) * nvme_npages(nbytes, dev) + |
eca18b23 MW |
387 | sizeof(struct scatterlist) * nseg, gfp); |
388 | ||
389 | if (iod) { | |
390 | iod->offset = offsetof(struct nvme_iod, sg[nseg]); | |
391 | iod->npages = -1; | |
392 | iod->length = nbytes; | |
2b196034 | 393 | iod->nents = 0; |
edd10d33 | 394 | iod->first_dma = 0ULL; |
6198221f | 395 | iod->start_time = jiffies; |
eca18b23 MW |
396 | } |
397 | ||
398 | return iod; | |
b60503ba MW |
399 | } |
400 | ||
5d0f6131 | 401 | void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod) |
b60503ba | 402 | { |
1d090624 | 403 | const int last_prp = dev->page_size / 8 - 1; |
eca18b23 MW |
404 | int i; |
405 | __le64 **list = iod_list(iod); | |
406 | dma_addr_t prp_dma = iod->first_dma; | |
407 | ||
408 | if (iod->npages == 0) | |
409 | dma_pool_free(dev->prp_small_pool, list[0], prp_dma); | |
410 | for (i = 0; i < iod->npages; i++) { | |
411 | __le64 *prp_list = list[i]; | |
412 | dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); | |
413 | dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); | |
414 | prp_dma = next_prp_dma; | |
415 | } | |
416 | kfree(iod); | |
b60503ba MW |
417 | } |
418 | ||
6198221f KB |
419 | static void nvme_start_io_acct(struct bio *bio) |
420 | { | |
421 | struct gendisk *disk = bio->bi_bdev->bd_disk; | |
b4e75cbf SB |
422 | if (blk_queue_io_stat(disk->queue)) { |
423 | const int rw = bio_data_dir(bio); | |
424 | int cpu = part_stat_lock(); | |
425 | part_round_stats(cpu, &disk->part0); | |
426 | part_stat_inc(cpu, &disk->part0, ios[rw]); | |
427 | part_stat_add(cpu, &disk->part0, sectors[rw], | |
428 | bio_sectors(bio)); | |
429 | part_inc_in_flight(&disk->part0, rw); | |
430 | part_stat_unlock(); | |
431 | } | |
6198221f KB |
432 | } |
433 | ||
434 | static void nvme_end_io_acct(struct bio *bio, unsigned long start_time) | |
435 | { | |
436 | struct gendisk *disk = bio->bi_bdev->bd_disk; | |
b4e75cbf SB |
437 | if (blk_queue_io_stat(disk->queue)) { |
438 | const int rw = bio_data_dir(bio); | |
439 | unsigned long duration = jiffies - start_time; | |
440 | int cpu = part_stat_lock(); | |
441 | part_stat_add(cpu, &disk->part0, ticks[rw], duration); | |
442 | part_round_stats(cpu, &disk->part0); | |
443 | part_dec_in_flight(&disk->part0, rw); | |
444 | part_stat_unlock(); | |
445 | } | |
6198221f KB |
446 | } |
447 | ||
edd10d33 | 448 | static void bio_completion(struct nvme_queue *nvmeq, void *ctx, |
b60503ba MW |
449 | struct nvme_completion *cqe) |
450 | { | |
eca18b23 MW |
451 | struct nvme_iod *iod = ctx; |
452 | struct bio *bio = iod->private; | |
b60503ba | 453 | u16 status = le16_to_cpup(&cqe->status) >> 1; |
3291fa57 | 454 | int error = 0; |
b60503ba | 455 | |
edd10d33 KB |
456 | if (unlikely(status)) { |
457 | if (!(status & NVME_SC_DNR || | |
458 | bio->bi_rw & REQ_FAILFAST_MASK) && | |
459 | (jiffies - iod->start_time) < IOD_TIMEOUT) { | |
460 | if (!waitqueue_active(&nvmeq->sq_full)) | |
461 | add_wait_queue(&nvmeq->sq_full, | |
462 | &nvmeq->sq_cong_wait); | |
463 | list_add_tail(&iod->node, &nvmeq->iod_bio); | |
464 | wake_up(&nvmeq->sq_full); | |
465 | return; | |
466 | } | |
3291fa57 | 467 | error = -EIO; |
edd10d33 | 468 | } |
9e59d091 | 469 | if (iod->nents) { |
edd10d33 | 470 | dma_unmap_sg(nvmeq->q_dmadev, iod->sg, iod->nents, |
b60503ba | 471 | bio_data_dir(bio) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
9e59d091 KB |
472 | nvme_end_io_acct(bio, iod->start_time); |
473 | } | |
edd10d33 | 474 | nvme_free_iod(nvmeq->dev, iod); |
3291fa57 KB |
475 | |
476 | trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio, error); | |
477 | bio_endio(bio, error); | |
b60503ba MW |
478 | } |
479 | ||
184d2944 | 480 | /* length is in bytes. gfp flags indicates whether we may sleep. */ |
edd10d33 KB |
481 | int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod, int total_len, |
482 | gfp_t gfp) | |
ff22b54f | 483 | { |
99802a7a | 484 | struct dma_pool *pool; |
eca18b23 MW |
485 | int length = total_len; |
486 | struct scatterlist *sg = iod->sg; | |
ff22b54f MW |
487 | int dma_len = sg_dma_len(sg); |
488 | u64 dma_addr = sg_dma_address(sg); | |
489 | int offset = offset_in_page(dma_addr); | |
e025344c | 490 | __le64 *prp_list; |
eca18b23 | 491 | __le64 **list = iod_list(iod); |
e025344c | 492 | dma_addr_t prp_dma; |
eca18b23 | 493 | int nprps, i; |
1d090624 | 494 | u32 page_size = dev->page_size; |
ff22b54f | 495 | |
1d090624 | 496 | length -= (page_size - offset); |
ff22b54f | 497 | if (length <= 0) |
eca18b23 | 498 | return total_len; |
ff22b54f | 499 | |
1d090624 | 500 | dma_len -= (page_size - offset); |
ff22b54f | 501 | if (dma_len) { |
1d090624 | 502 | dma_addr += (page_size - offset); |
ff22b54f MW |
503 | } else { |
504 | sg = sg_next(sg); | |
505 | dma_addr = sg_dma_address(sg); | |
506 | dma_len = sg_dma_len(sg); | |
507 | } | |
508 | ||
1d090624 | 509 | if (length <= page_size) { |
edd10d33 | 510 | iod->first_dma = dma_addr; |
eca18b23 | 511 | return total_len; |
e025344c SMM |
512 | } |
513 | ||
1d090624 | 514 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
515 | if (nprps <= (256 / 8)) { |
516 | pool = dev->prp_small_pool; | |
eca18b23 | 517 | iod->npages = 0; |
99802a7a MW |
518 | } else { |
519 | pool = dev->prp_page_pool; | |
eca18b23 | 520 | iod->npages = 1; |
99802a7a MW |
521 | } |
522 | ||
b77954cb MW |
523 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
524 | if (!prp_list) { | |
edd10d33 | 525 | iod->first_dma = dma_addr; |
eca18b23 | 526 | iod->npages = -1; |
1d090624 | 527 | return (total_len - length) + page_size; |
b77954cb | 528 | } |
eca18b23 MW |
529 | list[0] = prp_list; |
530 | iod->first_dma = prp_dma; | |
e025344c SMM |
531 | i = 0; |
532 | for (;;) { | |
1d090624 | 533 | if (i == page_size >> 3) { |
e025344c | 534 | __le64 *old_prp_list = prp_list; |
b77954cb | 535 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
eca18b23 MW |
536 | if (!prp_list) |
537 | return total_len - length; | |
538 | list[iod->npages++] = prp_list; | |
7523d834 MW |
539 | prp_list[0] = old_prp_list[i - 1]; |
540 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
541 | i = 1; | |
e025344c SMM |
542 | } |
543 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
544 | dma_len -= page_size; |
545 | dma_addr += page_size; | |
546 | length -= page_size; | |
e025344c SMM |
547 | if (length <= 0) |
548 | break; | |
549 | if (dma_len > 0) | |
550 | continue; | |
551 | BUG_ON(dma_len < 0); | |
552 | sg = sg_next(sg); | |
553 | dma_addr = sg_dma_address(sg); | |
554 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
555 | } |
556 | ||
eca18b23 | 557 | return total_len; |
ff22b54f MW |
558 | } |
559 | ||
427e9708 | 560 | static int nvme_split_and_submit(struct bio *bio, struct nvme_queue *nvmeq, |
20d0189b | 561 | int len) |
427e9708 | 562 | { |
20d0189b KO |
563 | struct bio *split = bio_split(bio, len >> 9, GFP_ATOMIC, NULL); |
564 | if (!split) | |
427e9708 KB |
565 | return -ENOMEM; |
566 | ||
3291fa57 KB |
567 | trace_block_split(bdev_get_queue(bio->bi_bdev), bio, |
568 | split->bi_iter.bi_sector); | |
20d0189b KO |
569 | bio_chain(split, bio); |
570 | ||
edd10d33 | 571 | if (!waitqueue_active(&nvmeq->sq_full)) |
427e9708 | 572 | add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait); |
20d0189b KO |
573 | bio_list_add(&nvmeq->sq_cong, split); |
574 | bio_list_add(&nvmeq->sq_cong, bio); | |
edd10d33 | 575 | wake_up(&nvmeq->sq_full); |
427e9708 KB |
576 | |
577 | return 0; | |
578 | } | |
579 | ||
1ad2f893 MW |
580 | /* NVMe scatterlists require no holes in the virtual address */ |
581 | #define BIOVEC_NOT_VIRT_MERGEABLE(vec1, vec2) ((vec2)->bv_offset || \ | |
582 | (((vec1)->bv_offset + (vec1)->bv_len) % PAGE_SIZE)) | |
583 | ||
427e9708 | 584 | static int nvme_map_bio(struct nvme_queue *nvmeq, struct nvme_iod *iod, |
b60503ba MW |
585 | struct bio *bio, enum dma_data_direction dma_dir, int psegs) |
586 | { | |
7988613b KO |
587 | struct bio_vec bvec, bvprv; |
588 | struct bvec_iter iter; | |
76830840 | 589 | struct scatterlist *sg = NULL; |
7988613b KO |
590 | int length = 0, nsegs = 0, split_len = bio->bi_iter.bi_size; |
591 | int first = 1; | |
159b67d7 KB |
592 | |
593 | if (nvmeq->dev->stripe_size) | |
594 | split_len = nvmeq->dev->stripe_size - | |
4f024f37 KO |
595 | ((bio->bi_iter.bi_sector << 9) & |
596 | (nvmeq->dev->stripe_size - 1)); | |
b60503ba | 597 | |
eca18b23 | 598 | sg_init_table(iod->sg, psegs); |
7988613b KO |
599 | bio_for_each_segment(bvec, bio, iter) { |
600 | if (!first && BIOVEC_PHYS_MERGEABLE(&bvprv, &bvec)) { | |
601 | sg->length += bvec.bv_len; | |
76830840 | 602 | } else { |
7988613b KO |
603 | if (!first && BIOVEC_NOT_VIRT_MERGEABLE(&bvprv, &bvec)) |
604 | return nvme_split_and_submit(bio, nvmeq, | |
20d0189b | 605 | length); |
427e9708 | 606 | |
eca18b23 | 607 | sg = sg ? sg + 1 : iod->sg; |
7988613b KO |
608 | sg_set_page(sg, bvec.bv_page, |
609 | bvec.bv_len, bvec.bv_offset); | |
76830840 MW |
610 | nsegs++; |
611 | } | |
159b67d7 | 612 | |
7988613b | 613 | if (split_len - length < bvec.bv_len) |
20d0189b | 614 | return nvme_split_and_submit(bio, nvmeq, split_len); |
7988613b | 615 | length += bvec.bv_len; |
76830840 | 616 | bvprv = bvec; |
7988613b | 617 | first = 0; |
b60503ba | 618 | } |
eca18b23 | 619 | iod->nents = nsegs; |
76830840 | 620 | sg_mark_end(sg); |
427e9708 | 621 | if (dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir) == 0) |
1ad2f893 | 622 | return -ENOMEM; |
427e9708 | 623 | |
4f024f37 | 624 | BUG_ON(length != bio->bi_iter.bi_size); |
1ad2f893 | 625 | return length; |
b60503ba MW |
626 | } |
627 | ||
0e5e4f0e KB |
628 | static int nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
629 | struct bio *bio, struct nvme_iod *iod, int cmdid) | |
630 | { | |
edd10d33 KB |
631 | struct nvme_dsm_range *range = |
632 | (struct nvme_dsm_range *)iod_list(iod)[0]; | |
0e5e4f0e KB |
633 | struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; |
634 | ||
0e5e4f0e | 635 | range->cattr = cpu_to_le32(0); |
4f024f37 KO |
636 | range->nlb = cpu_to_le32(bio->bi_iter.bi_size >> ns->lba_shift); |
637 | range->slba = cpu_to_le64(nvme_block_nr(ns, bio->bi_iter.bi_sector)); | |
0e5e4f0e KB |
638 | |
639 | memset(cmnd, 0, sizeof(*cmnd)); | |
640 | cmnd->dsm.opcode = nvme_cmd_dsm; | |
641 | cmnd->dsm.command_id = cmdid; | |
642 | cmnd->dsm.nsid = cpu_to_le32(ns->ns_id); | |
643 | cmnd->dsm.prp1 = cpu_to_le64(iod->first_dma); | |
644 | cmnd->dsm.nr = 0; | |
645 | cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); | |
646 | ||
647 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
648 | nvmeq->sq_tail = 0; | |
649 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
650 | ||
651 | return 0; | |
652 | } | |
653 | ||
00df5cb4 MW |
654 | static int nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
655 | int cmdid) | |
656 | { | |
657 | struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; | |
658 | ||
659 | memset(cmnd, 0, sizeof(*cmnd)); | |
660 | cmnd->common.opcode = nvme_cmd_flush; | |
661 | cmnd->common.command_id = cmdid; | |
662 | cmnd->common.nsid = cpu_to_le32(ns->ns_id); | |
663 | ||
664 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
665 | nvmeq->sq_tail = 0; | |
666 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
667 | ||
668 | return 0; | |
669 | } | |
670 | ||
edd10d33 | 671 | static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod) |
b60503ba | 672 | { |
edd10d33 KB |
673 | struct bio *bio = iod->private; |
674 | struct nvme_ns *ns = bio->bi_bdev->bd_disk->private_data; | |
ff22b54f | 675 | struct nvme_command *cmnd; |
edd10d33 | 676 | int cmdid; |
b60503ba MW |
677 | u16 control; |
678 | u32 dsmgmt; | |
00df5cb4 | 679 | |
ff976d72 | 680 | cmdid = alloc_cmdid(nvmeq, iod, bio_completion, NVME_IO_TIMEOUT); |
b60503ba | 681 | if (unlikely(cmdid < 0)) |
edd10d33 | 682 | return cmdid; |
b60503ba | 683 | |
edd10d33 KB |
684 | if (bio->bi_rw & REQ_DISCARD) |
685 | return nvme_submit_discard(nvmeq, ns, bio, iod, cmdid); | |
53562be7 | 686 | if (bio->bi_rw & REQ_FLUSH) |
00df5cb4 MW |
687 | return nvme_submit_flush(nvmeq, ns, cmdid); |
688 | ||
b60503ba MW |
689 | control = 0; |
690 | if (bio->bi_rw & REQ_FUA) | |
691 | control |= NVME_RW_FUA; | |
692 | if (bio->bi_rw & (REQ_FAILFAST_DEV | REQ_RAHEAD)) | |
693 | control |= NVME_RW_LR; | |
694 | ||
695 | dsmgmt = 0; | |
696 | if (bio->bi_rw & REQ_RAHEAD) | |
697 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; | |
698 | ||
ff22b54f | 699 | cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; |
b8deb62c | 700 | memset(cmnd, 0, sizeof(*cmnd)); |
b60503ba | 701 | |
edd10d33 | 702 | cmnd->rw.opcode = bio_data_dir(bio) ? nvme_cmd_write : nvme_cmd_read; |
ff22b54f MW |
703 | cmnd->rw.command_id = cmdid; |
704 | cmnd->rw.nsid = cpu_to_le32(ns->ns_id); | |
edd10d33 KB |
705 | cmnd->rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); |
706 | cmnd->rw.prp2 = cpu_to_le64(iod->first_dma); | |
4f024f37 | 707 | cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, bio->bi_iter.bi_sector)); |
edd10d33 KB |
708 | cmnd->rw.length = |
709 | cpu_to_le16((bio->bi_iter.bi_size >> ns->lba_shift) - 1); | |
ff22b54f MW |
710 | cmnd->rw.control = cpu_to_le16(control); |
711 | cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); | |
b60503ba | 712 | |
b60503ba MW |
713 | if (++nvmeq->sq_tail == nvmeq->q_depth) |
714 | nvmeq->sq_tail = 0; | |
7547881d | 715 | writel(nvmeq->sq_tail, nvmeq->q_db); |
b60503ba | 716 | |
1974b1ae | 717 | return 0; |
edd10d33 KB |
718 | } |
719 | ||
53562be7 KB |
720 | static int nvme_split_flush_data(struct nvme_queue *nvmeq, struct bio *bio) |
721 | { | |
722 | struct bio *split = bio_clone(bio, GFP_ATOMIC); | |
723 | if (!split) | |
724 | return -ENOMEM; | |
725 | ||
726 | split->bi_iter.bi_size = 0; | |
727 | split->bi_phys_segments = 0; | |
728 | bio->bi_rw &= ~REQ_FLUSH; | |
729 | bio_chain(split, bio); | |
730 | ||
731 | if (!waitqueue_active(&nvmeq->sq_full)) | |
732 | add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait); | |
733 | bio_list_add(&nvmeq->sq_cong, split); | |
734 | bio_list_add(&nvmeq->sq_cong, bio); | |
735 | wake_up_process(nvme_thread); | |
736 | ||
737 | return 0; | |
738 | } | |
739 | ||
edd10d33 KB |
740 | /* |
741 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
742 | */ | |
743 | static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns, | |
744 | struct bio *bio) | |
745 | { | |
746 | struct nvme_iod *iod; | |
747 | int psegs = bio_phys_segments(ns->queue, bio); | |
748 | int result; | |
749 | ||
53562be7 KB |
750 | if ((bio->bi_rw & REQ_FLUSH) && psegs) |
751 | return nvme_split_flush_data(nvmeq, bio); | |
edd10d33 | 752 | |
1d090624 | 753 | iod = nvme_alloc_iod(psegs, bio->bi_iter.bi_size, ns->dev, GFP_ATOMIC); |
edd10d33 KB |
754 | if (!iod) |
755 | return -ENOMEM; | |
756 | ||
757 | iod->private = bio; | |
758 | if (bio->bi_rw & REQ_DISCARD) { | |
759 | void *range; | |
760 | /* | |
761 | * We reuse the small pool to allocate the 16-byte range here | |
762 | * as it is not worth having a special pool for these or | |
763 | * additional cases to handle freeing the iod. | |
764 | */ | |
765 | range = dma_pool_alloc(nvmeq->dev->prp_small_pool, | |
766 | GFP_ATOMIC, | |
767 | &iod->first_dma); | |
768 | if (!range) { | |
769 | result = -ENOMEM; | |
770 | goto free_iod; | |
771 | } | |
772 | iod_list(iod)[0] = (__le64 *)range; | |
773 | iod->npages = 0; | |
774 | } else if (psegs) { | |
775 | result = nvme_map_bio(nvmeq, iod, bio, | |
776 | bio_data_dir(bio) ? DMA_TO_DEVICE : DMA_FROM_DEVICE, | |
777 | psegs); | |
778 | if (result <= 0) | |
779 | goto free_iod; | |
780 | if (nvme_setup_prps(nvmeq->dev, iod, result, GFP_ATOMIC) != | |
781 | result) { | |
782 | result = -ENOMEM; | |
783 | goto free_iod; | |
784 | } | |
785 | nvme_start_io_acct(bio); | |
786 | } | |
787 | if (unlikely(nvme_submit_iod(nvmeq, iod))) { | |
788 | if (!waitqueue_active(&nvmeq->sq_full)) | |
789 | add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait); | |
790 | list_add_tail(&iod->node, &nvmeq->iod_bio); | |
791 | } | |
792 | return 0; | |
1974b1ae | 793 | |
eca18b23 MW |
794 | free_iod: |
795 | nvme_free_iod(nvmeq->dev, iod); | |
eeee3226 | 796 | return result; |
b60503ba MW |
797 | } |
798 | ||
e9539f47 | 799 | static int nvme_process_cq(struct nvme_queue *nvmeq) |
b60503ba | 800 | { |
82123460 | 801 | u16 head, phase; |
b60503ba | 802 | |
b60503ba | 803 | head = nvmeq->cq_head; |
82123460 | 804 | phase = nvmeq->cq_phase; |
b60503ba MW |
805 | |
806 | for (;;) { | |
c2f5b650 MW |
807 | void *ctx; |
808 | nvme_completion_fn fn; | |
b60503ba | 809 | struct nvme_completion cqe = nvmeq->cqes[head]; |
82123460 | 810 | if ((le16_to_cpu(cqe.status) & 1) != phase) |
b60503ba MW |
811 | break; |
812 | nvmeq->sq_head = le16_to_cpu(cqe.sq_head); | |
813 | if (++head == nvmeq->q_depth) { | |
814 | head = 0; | |
82123460 | 815 | phase = !phase; |
b60503ba MW |
816 | } |
817 | ||
c2f5b650 | 818 | ctx = free_cmdid(nvmeq, cqe.command_id, &fn); |
edd10d33 | 819 | fn(nvmeq, ctx, &cqe); |
b60503ba MW |
820 | } |
821 | ||
822 | /* If the controller ignores the cq head doorbell and continuously | |
823 | * writes to the queue, it is theoretically possible to wrap around | |
824 | * the queue twice and mistakenly return IRQ_NONE. Linux only | |
825 | * requires that 0.1% of your interrupts are handled, so this isn't | |
826 | * a big problem. | |
827 | */ | |
82123460 | 828 | if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) |
e9539f47 | 829 | return 0; |
b60503ba | 830 | |
b80d5ccc | 831 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); |
b60503ba | 832 | nvmeq->cq_head = head; |
82123460 | 833 | nvmeq->cq_phase = phase; |
b60503ba | 834 | |
e9539f47 MW |
835 | nvmeq->cqe_seen = 1; |
836 | return 1; | |
b60503ba MW |
837 | } |
838 | ||
7d822457 MW |
839 | static void nvme_make_request(struct request_queue *q, struct bio *bio) |
840 | { | |
841 | struct nvme_ns *ns = q->queuedata; | |
842 | struct nvme_queue *nvmeq = get_nvmeq(ns->dev); | |
843 | int result = -EBUSY; | |
844 | ||
cd638946 | 845 | if (!nvmeq) { |
cd638946 KB |
846 | bio_endio(bio, -EIO); |
847 | return; | |
848 | } | |
849 | ||
7d822457 | 850 | spin_lock_irq(&nvmeq->q_lock); |
22404274 | 851 | if (!nvmeq->q_suspended && bio_list_empty(&nvmeq->sq_cong)) |
7d822457 MW |
852 | result = nvme_submit_bio_queue(nvmeq, ns, bio); |
853 | if (unlikely(result)) { | |
edd10d33 | 854 | if (!waitqueue_active(&nvmeq->sq_full)) |
7d822457 MW |
855 | add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait); |
856 | bio_list_add(&nvmeq->sq_cong, bio); | |
857 | } | |
858 | ||
859 | nvme_process_cq(nvmeq); | |
860 | spin_unlock_irq(&nvmeq->q_lock); | |
861 | put_nvmeq(nvmeq); | |
862 | } | |
863 | ||
b60503ba | 864 | static irqreturn_t nvme_irq(int irq, void *data) |
58ffacb5 MW |
865 | { |
866 | irqreturn_t result; | |
867 | struct nvme_queue *nvmeq = data; | |
868 | spin_lock(&nvmeq->q_lock); | |
e9539f47 MW |
869 | nvme_process_cq(nvmeq); |
870 | result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; | |
871 | nvmeq->cqe_seen = 0; | |
58ffacb5 MW |
872 | spin_unlock(&nvmeq->q_lock); |
873 | return result; | |
874 | } | |
875 | ||
876 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
877 | { | |
878 | struct nvme_queue *nvmeq = data; | |
879 | struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head]; | |
880 | if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase) | |
881 | return IRQ_NONE; | |
882 | return IRQ_WAKE_THREAD; | |
883 | } | |
884 | ||
3c0cf138 MW |
885 | static void nvme_abort_command(struct nvme_queue *nvmeq, int cmdid) |
886 | { | |
887 | spin_lock_irq(&nvmeq->q_lock); | |
c2f5b650 | 888 | cancel_cmdid(nvmeq, cmdid, NULL); |
3c0cf138 MW |
889 | spin_unlock_irq(&nvmeq->q_lock); |
890 | } | |
891 | ||
c2f5b650 MW |
892 | struct sync_cmd_info { |
893 | struct task_struct *task; | |
894 | u32 result; | |
895 | int status; | |
896 | }; | |
897 | ||
edd10d33 | 898 | static void sync_completion(struct nvme_queue *nvmeq, void *ctx, |
c2f5b650 MW |
899 | struct nvme_completion *cqe) |
900 | { | |
901 | struct sync_cmd_info *cmdinfo = ctx; | |
902 | cmdinfo->result = le32_to_cpup(&cqe->result); | |
903 | cmdinfo->status = le16_to_cpup(&cqe->status) >> 1; | |
904 | wake_up_process(cmdinfo->task); | |
905 | } | |
906 | ||
b60503ba MW |
907 | /* |
908 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
909 | * if the result is positive, it's an NVM Express status code | |
910 | */ | |
4f5099af KB |
911 | static int nvme_submit_sync_cmd(struct nvme_dev *dev, int q_idx, |
912 | struct nvme_command *cmd, | |
5d0f6131 | 913 | u32 *result, unsigned timeout) |
b60503ba | 914 | { |
4f5099af | 915 | int cmdid, ret; |
b60503ba | 916 | struct sync_cmd_info cmdinfo; |
4f5099af KB |
917 | struct nvme_queue *nvmeq; |
918 | ||
919 | nvmeq = lock_nvmeq(dev, q_idx); | |
a51afb54 | 920 | if (!nvmeq) |
4f5099af | 921 | return -ENODEV; |
b60503ba MW |
922 | |
923 | cmdinfo.task = current; | |
924 | cmdinfo.status = -EINTR; | |
925 | ||
4f5099af KB |
926 | cmdid = alloc_cmdid(nvmeq, &cmdinfo, sync_completion, timeout); |
927 | if (cmdid < 0) { | |
928 | unlock_nvmeq(nvmeq); | |
b60503ba | 929 | return cmdid; |
4f5099af | 930 | } |
b60503ba MW |
931 | cmd->common.command_id = cmdid; |
932 | ||
3c0cf138 | 933 | set_current_state(TASK_KILLABLE); |
4f5099af KB |
934 | ret = nvme_submit_cmd(nvmeq, cmd); |
935 | if (ret) { | |
936 | free_cmdid(nvmeq, cmdid, NULL); | |
937 | unlock_nvmeq(nvmeq); | |
938 | set_current_state(TASK_RUNNING); | |
939 | return ret; | |
940 | } | |
941 | unlock_nvmeq(nvmeq); | |
78f8d257 | 942 | schedule_timeout(timeout); |
b60503ba | 943 | |
3c0cf138 | 944 | if (cmdinfo.status == -EINTR) { |
4f5099af | 945 | nvmeq = lock_nvmeq(dev, q_idx); |
a51afb54 | 946 | if (nvmeq) { |
4f5099af | 947 | nvme_abort_command(nvmeq, cmdid); |
a51afb54 KB |
948 | unlock_nvmeq(nvmeq); |
949 | } | |
3c0cf138 MW |
950 | return -EINTR; |
951 | } | |
952 | ||
b60503ba MW |
953 | if (result) |
954 | *result = cmdinfo.result; | |
955 | ||
956 | return cmdinfo.status; | |
957 | } | |
958 | ||
4d115420 KB |
959 | static int nvme_submit_async_cmd(struct nvme_queue *nvmeq, |
960 | struct nvme_command *cmd, | |
961 | struct async_cmd_info *cmdinfo, unsigned timeout) | |
962 | { | |
963 | int cmdid; | |
964 | ||
965 | cmdid = alloc_cmdid_killable(nvmeq, cmdinfo, async_completion, timeout); | |
966 | if (cmdid < 0) | |
967 | return cmdid; | |
968 | cmdinfo->status = -EINTR; | |
969 | cmd->common.command_id = cmdid; | |
4f5099af | 970 | return nvme_submit_cmd(nvmeq, cmd); |
4d115420 KB |
971 | } |
972 | ||
5d0f6131 | 973 | int nvme_submit_admin_cmd(struct nvme_dev *dev, struct nvme_command *cmd, |
b60503ba MW |
974 | u32 *result) |
975 | { | |
4f5099af KB |
976 | return nvme_submit_sync_cmd(dev, 0, cmd, result, ADMIN_TIMEOUT); |
977 | } | |
978 | ||
979 | int nvme_submit_io_cmd(struct nvme_dev *dev, struct nvme_command *cmd, | |
980 | u32 *result) | |
981 | { | |
982 | return nvme_submit_sync_cmd(dev, smp_processor_id() + 1, cmd, result, | |
983 | NVME_IO_TIMEOUT); | |
b60503ba MW |
984 | } |
985 | ||
4d115420 KB |
986 | static int nvme_submit_admin_cmd_async(struct nvme_dev *dev, |
987 | struct nvme_command *cmd, struct async_cmd_info *cmdinfo) | |
988 | { | |
5a92e700 | 989 | return nvme_submit_async_cmd(raw_nvmeq(dev, 0), cmd, cmdinfo, |
4d115420 KB |
990 | ADMIN_TIMEOUT); |
991 | } | |
992 | ||
b60503ba MW |
993 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
994 | { | |
995 | int status; | |
996 | struct nvme_command c; | |
997 | ||
998 | memset(&c, 0, sizeof(c)); | |
999 | c.delete_queue.opcode = opcode; | |
1000 | c.delete_queue.qid = cpu_to_le16(id); | |
1001 | ||
1002 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
1003 | if (status) | |
1004 | return -EIO; | |
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, | |
1009 | struct nvme_queue *nvmeq) | |
1010 | { | |
1011 | int status; | |
1012 | struct nvme_command c; | |
1013 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; | |
1014 | ||
1015 | memset(&c, 0, sizeof(c)); | |
1016 | c.create_cq.opcode = nvme_admin_create_cq; | |
1017 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
1018 | c.create_cq.cqid = cpu_to_le16(qid); | |
1019 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1020 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
1021 | c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); | |
1022 | ||
1023 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
1024 | if (status) | |
1025 | return -EIO; | |
1026 | return 0; | |
1027 | } | |
1028 | ||
1029 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
1030 | struct nvme_queue *nvmeq) | |
1031 | { | |
1032 | int status; | |
1033 | struct nvme_command c; | |
1034 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; | |
1035 | ||
1036 | memset(&c, 0, sizeof(c)); | |
1037 | c.create_sq.opcode = nvme_admin_create_sq; | |
1038 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
1039 | c.create_sq.sqid = cpu_to_le16(qid); | |
1040 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1041 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
1042 | c.create_sq.cqid = cpu_to_le16(qid); | |
1043 | ||
1044 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
1045 | if (status) | |
1046 | return -EIO; | |
1047 | return 0; | |
1048 | } | |
1049 | ||
1050 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
1051 | { | |
1052 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
1053 | } | |
1054 | ||
1055 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
1056 | { | |
1057 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
1058 | } | |
1059 | ||
5d0f6131 | 1060 | int nvme_identify(struct nvme_dev *dev, unsigned nsid, unsigned cns, |
bc5fc7e4 MW |
1061 | dma_addr_t dma_addr) |
1062 | { | |
1063 | struct nvme_command c; | |
1064 | ||
1065 | memset(&c, 0, sizeof(c)); | |
1066 | c.identify.opcode = nvme_admin_identify; | |
1067 | c.identify.nsid = cpu_to_le32(nsid); | |
1068 | c.identify.prp1 = cpu_to_le64(dma_addr); | |
1069 | c.identify.cns = cpu_to_le32(cns); | |
1070 | ||
1071 | return nvme_submit_admin_cmd(dev, &c, NULL); | |
1072 | } | |
1073 | ||
5d0f6131 | 1074 | int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid, |
08df1e05 | 1075 | dma_addr_t dma_addr, u32 *result) |
bc5fc7e4 MW |
1076 | { |
1077 | struct nvme_command c; | |
1078 | ||
1079 | memset(&c, 0, sizeof(c)); | |
1080 | c.features.opcode = nvme_admin_get_features; | |
a42cecce | 1081 | c.features.nsid = cpu_to_le32(nsid); |
bc5fc7e4 MW |
1082 | c.features.prp1 = cpu_to_le64(dma_addr); |
1083 | c.features.fid = cpu_to_le32(fid); | |
bc5fc7e4 | 1084 | |
08df1e05 | 1085 | return nvme_submit_admin_cmd(dev, &c, result); |
df348139 MW |
1086 | } |
1087 | ||
5d0f6131 VV |
1088 | int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11, |
1089 | dma_addr_t dma_addr, u32 *result) | |
df348139 MW |
1090 | { |
1091 | struct nvme_command c; | |
1092 | ||
1093 | memset(&c, 0, sizeof(c)); | |
1094 | c.features.opcode = nvme_admin_set_features; | |
1095 | c.features.prp1 = cpu_to_le64(dma_addr); | |
1096 | c.features.fid = cpu_to_le32(fid); | |
1097 | c.features.dword11 = cpu_to_le32(dword11); | |
1098 | ||
bc5fc7e4 MW |
1099 | return nvme_submit_admin_cmd(dev, &c, result); |
1100 | } | |
1101 | ||
c30341dc KB |
1102 | /** |
1103 | * nvme_abort_cmd - Attempt aborting a command | |
1104 | * @cmdid: Command id of a timed out IO | |
1105 | * @queue: The queue with timed out IO | |
1106 | * | |
1107 | * Schedule controller reset if the command was already aborted once before and | |
1108 | * still hasn't been returned to the driver, or if this is the admin queue. | |
1109 | */ | |
1110 | static void nvme_abort_cmd(int cmdid, struct nvme_queue *nvmeq) | |
1111 | { | |
1112 | int a_cmdid; | |
1113 | struct nvme_command cmd; | |
1114 | struct nvme_dev *dev = nvmeq->dev; | |
1115 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); | |
5a92e700 | 1116 | struct nvme_queue *adminq; |
c30341dc KB |
1117 | |
1118 | if (!nvmeq->qid || info[cmdid].aborted) { | |
1119 | if (work_busy(&dev->reset_work)) | |
1120 | return; | |
1121 | list_del_init(&dev->node); | |
1122 | dev_warn(&dev->pci_dev->dev, | |
1123 | "I/O %d QID %d timeout, reset controller\n", cmdid, | |
1124 | nvmeq->qid); | |
9ca97374 | 1125 | dev->reset_workfn = nvme_reset_failed_dev; |
c30341dc KB |
1126 | queue_work(nvme_workq, &dev->reset_work); |
1127 | return; | |
1128 | } | |
1129 | ||
1130 | if (!dev->abort_limit) | |
1131 | return; | |
1132 | ||
5a92e700 KB |
1133 | adminq = rcu_dereference(dev->queues[0]); |
1134 | a_cmdid = alloc_cmdid(adminq, CMD_CTX_ABORT, special_completion, | |
c30341dc KB |
1135 | ADMIN_TIMEOUT); |
1136 | if (a_cmdid < 0) | |
1137 | return; | |
1138 | ||
1139 | memset(&cmd, 0, sizeof(cmd)); | |
1140 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
1141 | cmd.abort.cid = cmdid; | |
1142 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); | |
1143 | cmd.abort.command_id = a_cmdid; | |
1144 | ||
1145 | --dev->abort_limit; | |
1146 | info[cmdid].aborted = 1; | |
1147 | info[cmdid].timeout = jiffies + ADMIN_TIMEOUT; | |
1148 | ||
1149 | dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", cmdid, | |
1150 | nvmeq->qid); | |
5a92e700 | 1151 | nvme_submit_cmd(adminq, &cmd); |
c30341dc KB |
1152 | } |
1153 | ||
a09115b2 MW |
1154 | /** |
1155 | * nvme_cancel_ios - Cancel outstanding I/Os | |
1156 | * @queue: The queue to cancel I/Os on | |
1157 | * @timeout: True to only cancel I/Os which have timed out | |
1158 | */ | |
1159 | static void nvme_cancel_ios(struct nvme_queue *nvmeq, bool timeout) | |
1160 | { | |
1161 | int depth = nvmeq->q_depth - 1; | |
1162 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); | |
1163 | unsigned long now = jiffies; | |
1164 | int cmdid; | |
1165 | ||
1166 | for_each_set_bit(cmdid, nvmeq->cmdid_data, depth) { | |
1167 | void *ctx; | |
1168 | nvme_completion_fn fn; | |
1169 | static struct nvme_completion cqe = { | |
af2d9ca7 | 1170 | .status = cpu_to_le16(NVME_SC_ABORT_REQ << 1), |
a09115b2 MW |
1171 | }; |
1172 | ||
1173 | if (timeout && !time_after(now, info[cmdid].timeout)) | |
1174 | continue; | |
053ab702 KB |
1175 | if (info[cmdid].ctx == CMD_CTX_CANCELLED) |
1176 | continue; | |
6fccf938 KB |
1177 | if (timeout && info[cmdid].ctx == CMD_CTX_ASYNC) |
1178 | continue; | |
c30341dc KB |
1179 | if (timeout && nvmeq->dev->initialized) { |
1180 | nvme_abort_cmd(cmdid, nvmeq); | |
1181 | continue; | |
1182 | } | |
1183 | dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n", cmdid, | |
1184 | nvmeq->qid); | |
a09115b2 | 1185 | ctx = cancel_cmdid(nvmeq, cmdid, &fn); |
edd10d33 | 1186 | fn(nvmeq, ctx, &cqe); |
a09115b2 MW |
1187 | } |
1188 | } | |
1189 | ||
5a92e700 | 1190 | static void nvme_free_queue(struct rcu_head *r) |
9e866774 | 1191 | { |
5a92e700 KB |
1192 | struct nvme_queue *nvmeq = container_of(r, struct nvme_queue, r_head); |
1193 | ||
22404274 KB |
1194 | spin_lock_irq(&nvmeq->q_lock); |
1195 | while (bio_list_peek(&nvmeq->sq_cong)) { | |
1196 | struct bio *bio = bio_list_pop(&nvmeq->sq_cong); | |
1197 | bio_endio(bio, -EIO); | |
1198 | } | |
edd10d33 KB |
1199 | while (!list_empty(&nvmeq->iod_bio)) { |
1200 | static struct nvme_completion cqe = { | |
1201 | .status = cpu_to_le16( | |
1202 | (NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1), | |
1203 | }; | |
1204 | struct nvme_iod *iod = list_first_entry(&nvmeq->iod_bio, | |
1205 | struct nvme_iod, | |
1206 | node); | |
1207 | list_del(&iod->node); | |
1208 | bio_completion(nvmeq, iod, &cqe); | |
1209 | } | |
22404274 KB |
1210 | spin_unlock_irq(&nvmeq->q_lock); |
1211 | ||
9e866774 MW |
1212 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), |
1213 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
1214 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
1215 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); | |
42f61420 KB |
1216 | if (nvmeq->qid) |
1217 | free_cpumask_var(nvmeq->cpu_mask); | |
9e866774 MW |
1218 | kfree(nvmeq); |
1219 | } | |
1220 | ||
a1a5ef99 | 1221 | static void nvme_free_queues(struct nvme_dev *dev, int lowest) |
22404274 KB |
1222 | { |
1223 | int i; | |
1224 | ||
a1a5ef99 | 1225 | for (i = dev->queue_count - 1; i >= lowest; i--) { |
5a92e700 KB |
1226 | struct nvme_queue *nvmeq = raw_nvmeq(dev, i); |
1227 | rcu_assign_pointer(dev->queues[i], NULL); | |
1228 | call_rcu(&nvmeq->r_head, nvme_free_queue); | |
22404274 | 1229 | dev->queue_count--; |
22404274 KB |
1230 | } |
1231 | } | |
1232 | ||
4d115420 KB |
1233 | /** |
1234 | * nvme_suspend_queue - put queue into suspended state | |
1235 | * @nvmeq - queue to suspend | |
1236 | * | |
1237 | * Returns 1 if already suspended, 0 otherwise. | |
1238 | */ | |
1239 | static int nvme_suspend_queue(struct nvme_queue *nvmeq) | |
b60503ba | 1240 | { |
4d115420 | 1241 | int vector = nvmeq->dev->entry[nvmeq->cq_vector].vector; |
b60503ba | 1242 | |
a09115b2 | 1243 | spin_lock_irq(&nvmeq->q_lock); |
22404274 KB |
1244 | if (nvmeq->q_suspended) { |
1245 | spin_unlock_irq(&nvmeq->q_lock); | |
4d115420 | 1246 | return 1; |
3295874b | 1247 | } |
22404274 | 1248 | nvmeq->q_suspended = 1; |
42f61420 | 1249 | nvmeq->dev->online_queues--; |
a09115b2 MW |
1250 | spin_unlock_irq(&nvmeq->q_lock); |
1251 | ||
aba2080f MW |
1252 | irq_set_affinity_hint(vector, NULL); |
1253 | free_irq(vector, nvmeq); | |
b60503ba | 1254 | |
4d115420 KB |
1255 | return 0; |
1256 | } | |
b60503ba | 1257 | |
4d115420 KB |
1258 | static void nvme_clear_queue(struct nvme_queue *nvmeq) |
1259 | { | |
22404274 KB |
1260 | spin_lock_irq(&nvmeq->q_lock); |
1261 | nvme_process_cq(nvmeq); | |
1262 | nvme_cancel_ios(nvmeq, false); | |
1263 | spin_unlock_irq(&nvmeq->q_lock); | |
b60503ba MW |
1264 | } |
1265 | ||
4d115420 KB |
1266 | static void nvme_disable_queue(struct nvme_dev *dev, int qid) |
1267 | { | |
5a92e700 | 1268 | struct nvme_queue *nvmeq = raw_nvmeq(dev, qid); |
4d115420 KB |
1269 | |
1270 | if (!nvmeq) | |
1271 | return; | |
1272 | if (nvme_suspend_queue(nvmeq)) | |
1273 | return; | |
1274 | ||
0e53d180 KB |
1275 | /* Don't tell the adapter to delete the admin queue. |
1276 | * Don't tell a removed adapter to delete IO queues. */ | |
1277 | if (qid && readl(&dev->bar->csts) != -1) { | |
b60503ba MW |
1278 | adapter_delete_sq(dev, qid); |
1279 | adapter_delete_cq(dev, qid); | |
1280 | } | |
4d115420 | 1281 | nvme_clear_queue(nvmeq); |
b60503ba MW |
1282 | } |
1283 | ||
1284 | static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, | |
1285 | int depth, int vector) | |
1286 | { | |
1287 | struct device *dmadev = &dev->pci_dev->dev; | |
22404274 | 1288 | unsigned extra = nvme_queue_extra(depth); |
b60503ba MW |
1289 | struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq) + extra, GFP_KERNEL); |
1290 | if (!nvmeq) | |
1291 | return NULL; | |
1292 | ||
4d51abf9 JP |
1293 | nvmeq->cqes = dma_zalloc_coherent(dmadev, CQ_SIZE(depth), |
1294 | &nvmeq->cq_dma_addr, GFP_KERNEL); | |
b60503ba MW |
1295 | if (!nvmeq->cqes) |
1296 | goto free_nvmeq; | |
b60503ba MW |
1297 | |
1298 | nvmeq->sq_cmds = dma_alloc_coherent(dmadev, SQ_SIZE(depth), | |
1299 | &nvmeq->sq_dma_addr, GFP_KERNEL); | |
1300 | if (!nvmeq->sq_cmds) | |
1301 | goto free_cqdma; | |
1302 | ||
42f61420 KB |
1303 | if (qid && !zalloc_cpumask_var(&nvmeq->cpu_mask, GFP_KERNEL)) |
1304 | goto free_sqdma; | |
1305 | ||
b60503ba | 1306 | nvmeq->q_dmadev = dmadev; |
091b6092 | 1307 | nvmeq->dev = dev; |
3193f07b MW |
1308 | snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d", |
1309 | dev->instance, qid); | |
b60503ba MW |
1310 | spin_lock_init(&nvmeq->q_lock); |
1311 | nvmeq->cq_head = 0; | |
82123460 | 1312 | nvmeq->cq_phase = 1; |
b60503ba | 1313 | init_waitqueue_head(&nvmeq->sq_full); |
1fa6aead | 1314 | init_waitqueue_entry(&nvmeq->sq_cong_wait, nvme_thread); |
b60503ba | 1315 | bio_list_init(&nvmeq->sq_cong); |
edd10d33 | 1316 | INIT_LIST_HEAD(&nvmeq->iod_bio); |
b80d5ccc | 1317 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
b60503ba MW |
1318 | nvmeq->q_depth = depth; |
1319 | nvmeq->cq_vector = vector; | |
c30341dc | 1320 | nvmeq->qid = qid; |
22404274 KB |
1321 | nvmeq->q_suspended = 1; |
1322 | dev->queue_count++; | |
5a92e700 | 1323 | rcu_assign_pointer(dev->queues[qid], nvmeq); |
b60503ba MW |
1324 | |
1325 | return nvmeq; | |
1326 | ||
42f61420 KB |
1327 | free_sqdma: |
1328 | dma_free_coherent(dmadev, SQ_SIZE(depth), (void *)nvmeq->sq_cmds, | |
1329 | nvmeq->sq_dma_addr); | |
b60503ba | 1330 | free_cqdma: |
68b8eca5 | 1331 | dma_free_coherent(dmadev, CQ_SIZE(depth), (void *)nvmeq->cqes, |
b60503ba MW |
1332 | nvmeq->cq_dma_addr); |
1333 | free_nvmeq: | |
1334 | kfree(nvmeq); | |
1335 | return NULL; | |
1336 | } | |
1337 | ||
3001082c MW |
1338 | static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq, |
1339 | const char *name) | |
1340 | { | |
58ffacb5 MW |
1341 | if (use_threaded_interrupts) |
1342 | return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector, | |
481e5bad | 1343 | nvme_irq_check, nvme_irq, IRQF_SHARED, |
58ffacb5 | 1344 | name, nvmeq); |
3001082c | 1345 | return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq, |
481e5bad | 1346 | IRQF_SHARED, name, nvmeq); |
3001082c MW |
1347 | } |
1348 | ||
22404274 | 1349 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1350 | { |
22404274 KB |
1351 | struct nvme_dev *dev = nvmeq->dev; |
1352 | unsigned extra = nvme_queue_extra(nvmeq->q_depth); | |
b60503ba | 1353 | |
22404274 KB |
1354 | nvmeq->sq_tail = 0; |
1355 | nvmeq->cq_head = 0; | |
1356 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1357 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
22404274 KB |
1358 | memset(nvmeq->cmdid_data, 0, extra); |
1359 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); | |
1360 | nvme_cancel_ios(nvmeq, false); | |
1361 | nvmeq->q_suspended = 0; | |
42f61420 | 1362 | dev->online_queues++; |
22404274 KB |
1363 | } |
1364 | ||
1365 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) | |
1366 | { | |
1367 | struct nvme_dev *dev = nvmeq->dev; | |
1368 | int result; | |
3f85d50b | 1369 | |
b60503ba MW |
1370 | result = adapter_alloc_cq(dev, qid, nvmeq); |
1371 | if (result < 0) | |
22404274 | 1372 | return result; |
b60503ba MW |
1373 | |
1374 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1375 | if (result < 0) | |
1376 | goto release_cq; | |
1377 | ||
3193f07b | 1378 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
b60503ba MW |
1379 | if (result < 0) |
1380 | goto release_sq; | |
1381 | ||
0a8d44cb | 1382 | spin_lock_irq(&nvmeq->q_lock); |
22404274 | 1383 | nvme_init_queue(nvmeq, qid); |
0a8d44cb | 1384 | spin_unlock_irq(&nvmeq->q_lock); |
22404274 KB |
1385 | |
1386 | return result; | |
b60503ba MW |
1387 | |
1388 | release_sq: | |
1389 | adapter_delete_sq(dev, qid); | |
1390 | release_cq: | |
1391 | adapter_delete_cq(dev, qid); | |
22404274 | 1392 | return result; |
b60503ba MW |
1393 | } |
1394 | ||
ba47e386 MW |
1395 | static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled) |
1396 | { | |
1397 | unsigned long timeout; | |
1398 | u32 bit = enabled ? NVME_CSTS_RDY : 0; | |
1399 | ||
1400 | timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
1401 | ||
1402 | while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) { | |
1403 | msleep(100); | |
1404 | if (fatal_signal_pending(current)) | |
1405 | return -EINTR; | |
1406 | if (time_after(jiffies, timeout)) { | |
1407 | dev_err(&dev->pci_dev->dev, | |
27e8166c MW |
1408 | "Device not ready; aborting %s\n", enabled ? |
1409 | "initialisation" : "reset"); | |
ba47e386 MW |
1410 | return -ENODEV; |
1411 | } | |
1412 | } | |
1413 | ||
1414 | return 0; | |
1415 | } | |
1416 | ||
1417 | /* | |
1418 | * If the device has been passed off to us in an enabled state, just clear | |
1419 | * the enabled bit. The spec says we should set the 'shutdown notification | |
1420 | * bits', but doing so may cause the device to complete commands to the | |
1421 | * admin queue ... and we don't know what memory that might be pointing at! | |
1422 | */ | |
1423 | static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap) | |
1424 | { | |
01079522 DM |
1425 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1426 | dev->ctrl_config &= ~NVME_CC_ENABLE; | |
1427 | writel(dev->ctrl_config, &dev->bar->cc); | |
44af146a | 1428 | |
ba47e386 MW |
1429 | return nvme_wait_ready(dev, cap, false); |
1430 | } | |
1431 | ||
1432 | static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap) | |
1433 | { | |
01079522 DM |
1434 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1435 | dev->ctrl_config |= NVME_CC_ENABLE; | |
1436 | writel(dev->ctrl_config, &dev->bar->cc); | |
1437 | ||
ba47e386 MW |
1438 | return nvme_wait_ready(dev, cap, true); |
1439 | } | |
1440 | ||
1894d8f1 KB |
1441 | static int nvme_shutdown_ctrl(struct nvme_dev *dev) |
1442 | { | |
1443 | unsigned long timeout; | |
1894d8f1 | 1444 | |
01079522 DM |
1445 | dev->ctrl_config &= ~NVME_CC_SHN_MASK; |
1446 | dev->ctrl_config |= NVME_CC_SHN_NORMAL; | |
1447 | ||
1448 | writel(dev->ctrl_config, &dev->bar->cc); | |
1894d8f1 KB |
1449 | |
1450 | timeout = 2 * HZ + jiffies; | |
1451 | while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) != | |
1452 | NVME_CSTS_SHST_CMPLT) { | |
1453 | msleep(100); | |
1454 | if (fatal_signal_pending(current)) | |
1455 | return -EINTR; | |
1456 | if (time_after(jiffies, timeout)) { | |
1457 | dev_err(&dev->pci_dev->dev, | |
1458 | "Device shutdown incomplete; abort shutdown\n"); | |
1459 | return -ENODEV; | |
1460 | } | |
1461 | } | |
1462 | ||
1463 | return 0; | |
1464 | } | |
1465 | ||
8d85fce7 | 1466 | static int nvme_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1467 | { |
ba47e386 | 1468 | int result; |
b60503ba | 1469 | u32 aqa; |
ba47e386 | 1470 | u64 cap = readq(&dev->bar->cap); |
b60503ba | 1471 | struct nvme_queue *nvmeq; |
1d090624 KB |
1472 | unsigned page_shift = PAGE_SHIFT; |
1473 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12; | |
1474 | unsigned dev_page_max = NVME_CAP_MPSMAX(cap) + 12; | |
1475 | ||
1476 | if (page_shift < dev_page_min) { | |
1477 | dev_err(&dev->pci_dev->dev, | |
1478 | "Minimum device page size (%u) too large for " | |
1479 | "host (%u)\n", 1 << dev_page_min, | |
1480 | 1 << page_shift); | |
1481 | return -ENODEV; | |
1482 | } | |
1483 | if (page_shift > dev_page_max) { | |
1484 | dev_info(&dev->pci_dev->dev, | |
1485 | "Device maximum page size (%u) smaller than " | |
1486 | "host (%u); enabling work-around\n", | |
1487 | 1 << dev_page_max, 1 << page_shift); | |
1488 | page_shift = dev_page_max; | |
1489 | } | |
b60503ba | 1490 | |
ba47e386 MW |
1491 | result = nvme_disable_ctrl(dev, cap); |
1492 | if (result < 0) | |
1493 | return result; | |
b60503ba | 1494 | |
5a92e700 | 1495 | nvmeq = raw_nvmeq(dev, 0); |
cd638946 KB |
1496 | if (!nvmeq) { |
1497 | nvmeq = nvme_alloc_queue(dev, 0, 64, 0); | |
1498 | if (!nvmeq) | |
1499 | return -ENOMEM; | |
cd638946 | 1500 | } |
b60503ba MW |
1501 | |
1502 | aqa = nvmeq->q_depth - 1; | |
1503 | aqa |= aqa << 16; | |
1504 | ||
1d090624 KB |
1505 | dev->page_size = 1 << page_shift; |
1506 | ||
01079522 | 1507 | dev->ctrl_config = NVME_CC_CSS_NVM; |
1d090624 | 1508 | dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; |
b60503ba | 1509 | dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; |
7f53f9d2 | 1510 | dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
b60503ba MW |
1511 | |
1512 | writel(aqa, &dev->bar->aqa); | |
1513 | writeq(nvmeq->sq_dma_addr, &dev->bar->asq); | |
1514 | writeq(nvmeq->cq_dma_addr, &dev->bar->acq); | |
b60503ba | 1515 | |
ba47e386 | 1516 | result = nvme_enable_ctrl(dev, cap); |
025c557a | 1517 | if (result) |
cd638946 | 1518 | return result; |
9e866774 | 1519 | |
3193f07b | 1520 | result = queue_request_irq(dev, nvmeq, nvmeq->irqname); |
025c557a | 1521 | if (result) |
cd638946 | 1522 | return result; |
025c557a | 1523 | |
0a8d44cb | 1524 | spin_lock_irq(&nvmeq->q_lock); |
22404274 | 1525 | nvme_init_queue(nvmeq, 0); |
0a8d44cb | 1526 | spin_unlock_irq(&nvmeq->q_lock); |
b60503ba MW |
1527 | return result; |
1528 | } | |
1529 | ||
5d0f6131 | 1530 | struct nvme_iod *nvme_map_user_pages(struct nvme_dev *dev, int write, |
eca18b23 | 1531 | unsigned long addr, unsigned length) |
b60503ba | 1532 | { |
36c14ed9 | 1533 | int i, err, count, nents, offset; |
7fc3cdab MW |
1534 | struct scatterlist *sg; |
1535 | struct page **pages; | |
eca18b23 | 1536 | struct nvme_iod *iod; |
36c14ed9 MW |
1537 | |
1538 | if (addr & 3) | |
eca18b23 | 1539 | return ERR_PTR(-EINVAL); |
5460fc03 | 1540 | if (!length || length > INT_MAX - PAGE_SIZE) |
eca18b23 | 1541 | return ERR_PTR(-EINVAL); |
7fc3cdab | 1542 | |
36c14ed9 | 1543 | offset = offset_in_page(addr); |
7fc3cdab MW |
1544 | count = DIV_ROUND_UP(offset + length, PAGE_SIZE); |
1545 | pages = kcalloc(count, sizeof(*pages), GFP_KERNEL); | |
22fff826 DC |
1546 | if (!pages) |
1547 | return ERR_PTR(-ENOMEM); | |
36c14ed9 MW |
1548 | |
1549 | err = get_user_pages_fast(addr, count, 1, pages); | |
1550 | if (err < count) { | |
1551 | count = err; | |
1552 | err = -EFAULT; | |
1553 | goto put_pages; | |
1554 | } | |
7fc3cdab | 1555 | |
6808c5fb | 1556 | err = -ENOMEM; |
1d090624 | 1557 | iod = nvme_alloc_iod(count, length, dev, GFP_KERNEL); |
6808c5fb S |
1558 | if (!iod) |
1559 | goto put_pages; | |
1560 | ||
eca18b23 | 1561 | sg = iod->sg; |
36c14ed9 | 1562 | sg_init_table(sg, count); |
d0ba1e49 MW |
1563 | for (i = 0; i < count; i++) { |
1564 | sg_set_page(&sg[i], pages[i], | |
5460fc03 DC |
1565 | min_t(unsigned, length, PAGE_SIZE - offset), |
1566 | offset); | |
d0ba1e49 MW |
1567 | length -= (PAGE_SIZE - offset); |
1568 | offset = 0; | |
7fc3cdab | 1569 | } |
fe304c43 | 1570 | sg_mark_end(&sg[i - 1]); |
1c2ad9fa | 1571 | iod->nents = count; |
7fc3cdab | 1572 | |
7fc3cdab MW |
1573 | nents = dma_map_sg(&dev->pci_dev->dev, sg, count, |
1574 | write ? DMA_TO_DEVICE : DMA_FROM_DEVICE); | |
36c14ed9 | 1575 | if (!nents) |
eca18b23 | 1576 | goto free_iod; |
b60503ba | 1577 | |
7fc3cdab | 1578 | kfree(pages); |
eca18b23 | 1579 | return iod; |
b60503ba | 1580 | |
eca18b23 MW |
1581 | free_iod: |
1582 | kfree(iod); | |
7fc3cdab MW |
1583 | put_pages: |
1584 | for (i = 0; i < count; i++) | |
1585 | put_page(pages[i]); | |
1586 | kfree(pages); | |
eca18b23 | 1587 | return ERR_PTR(err); |
7fc3cdab | 1588 | } |
b60503ba | 1589 | |
5d0f6131 | 1590 | void nvme_unmap_user_pages(struct nvme_dev *dev, int write, |
1c2ad9fa | 1591 | struct nvme_iod *iod) |
7fc3cdab | 1592 | { |
1c2ad9fa | 1593 | int i; |
b60503ba | 1594 | |
1c2ad9fa MW |
1595 | dma_unmap_sg(&dev->pci_dev->dev, iod->sg, iod->nents, |
1596 | write ? DMA_TO_DEVICE : DMA_FROM_DEVICE); | |
7fc3cdab | 1597 | |
1c2ad9fa MW |
1598 | for (i = 0; i < iod->nents; i++) |
1599 | put_page(sg_page(&iod->sg[i])); | |
7fc3cdab | 1600 | } |
b60503ba | 1601 | |
a53295b6 MW |
1602 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
1603 | { | |
1604 | struct nvme_dev *dev = ns->dev; | |
a53295b6 MW |
1605 | struct nvme_user_io io; |
1606 | struct nvme_command c; | |
f410c680 KB |
1607 | unsigned length, meta_len; |
1608 | int status, i; | |
1609 | struct nvme_iod *iod, *meta_iod = NULL; | |
1610 | dma_addr_t meta_dma_addr; | |
1611 | void *meta, *uninitialized_var(meta_mem); | |
a53295b6 MW |
1612 | |
1613 | if (copy_from_user(&io, uio, sizeof(io))) | |
1614 | return -EFAULT; | |
6c7d4945 | 1615 | length = (io.nblocks + 1) << ns->lba_shift; |
f410c680 KB |
1616 | meta_len = (io.nblocks + 1) * ns->ms; |
1617 | ||
1618 | if (meta_len && ((io.metadata & 3) || !io.metadata)) | |
1619 | return -EINVAL; | |
6c7d4945 MW |
1620 | |
1621 | switch (io.opcode) { | |
1622 | case nvme_cmd_write: | |
1623 | case nvme_cmd_read: | |
6bbf1acd | 1624 | case nvme_cmd_compare: |
eca18b23 | 1625 | iod = nvme_map_user_pages(dev, io.opcode & 1, io.addr, length); |
6413214c | 1626 | break; |
6c7d4945 | 1627 | default: |
6bbf1acd | 1628 | return -EINVAL; |
6c7d4945 MW |
1629 | } |
1630 | ||
eca18b23 MW |
1631 | if (IS_ERR(iod)) |
1632 | return PTR_ERR(iod); | |
a53295b6 MW |
1633 | |
1634 | memset(&c, 0, sizeof(c)); | |
1635 | c.rw.opcode = io.opcode; | |
1636 | c.rw.flags = io.flags; | |
6c7d4945 | 1637 | c.rw.nsid = cpu_to_le32(ns->ns_id); |
a53295b6 | 1638 | c.rw.slba = cpu_to_le64(io.slba); |
6c7d4945 | 1639 | c.rw.length = cpu_to_le16(io.nblocks); |
a53295b6 | 1640 | c.rw.control = cpu_to_le16(io.control); |
1c9b5265 MW |
1641 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); |
1642 | c.rw.reftag = cpu_to_le32(io.reftag); | |
1643 | c.rw.apptag = cpu_to_le16(io.apptag); | |
1644 | c.rw.appmask = cpu_to_le16(io.appmask); | |
f410c680 KB |
1645 | |
1646 | if (meta_len) { | |
1b56749e KB |
1647 | meta_iod = nvme_map_user_pages(dev, io.opcode & 1, io.metadata, |
1648 | meta_len); | |
f410c680 KB |
1649 | if (IS_ERR(meta_iod)) { |
1650 | status = PTR_ERR(meta_iod); | |
1651 | meta_iod = NULL; | |
1652 | goto unmap; | |
1653 | } | |
1654 | ||
1655 | meta_mem = dma_alloc_coherent(&dev->pci_dev->dev, meta_len, | |
1656 | &meta_dma_addr, GFP_KERNEL); | |
1657 | if (!meta_mem) { | |
1658 | status = -ENOMEM; | |
1659 | goto unmap; | |
1660 | } | |
1661 | ||
1662 | if (io.opcode & 1) { | |
1663 | int meta_offset = 0; | |
1664 | ||
1665 | for (i = 0; i < meta_iod->nents; i++) { | |
1666 | meta = kmap_atomic(sg_page(&meta_iod->sg[i])) + | |
1667 | meta_iod->sg[i].offset; | |
1668 | memcpy(meta_mem + meta_offset, meta, | |
1669 | meta_iod->sg[i].length); | |
1670 | kunmap_atomic(meta); | |
1671 | meta_offset += meta_iod->sg[i].length; | |
1672 | } | |
1673 | } | |
1674 | ||
1675 | c.rw.metadata = cpu_to_le64(meta_dma_addr); | |
1676 | } | |
1677 | ||
edd10d33 KB |
1678 | length = nvme_setup_prps(dev, iod, length, GFP_KERNEL); |
1679 | c.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
1680 | c.rw.prp2 = cpu_to_le64(iod->first_dma); | |
a53295b6 | 1681 | |
b77954cb MW |
1682 | if (length != (io.nblocks + 1) << ns->lba_shift) |
1683 | status = -ENOMEM; | |
1684 | else | |
4f5099af | 1685 | status = nvme_submit_io_cmd(dev, &c, NULL); |
a53295b6 | 1686 | |
f410c680 KB |
1687 | if (meta_len) { |
1688 | if (status == NVME_SC_SUCCESS && !(io.opcode & 1)) { | |
1689 | int meta_offset = 0; | |
1690 | ||
1691 | for (i = 0; i < meta_iod->nents; i++) { | |
1692 | meta = kmap_atomic(sg_page(&meta_iod->sg[i])) + | |
1693 | meta_iod->sg[i].offset; | |
1694 | memcpy(meta, meta_mem + meta_offset, | |
1695 | meta_iod->sg[i].length); | |
1696 | kunmap_atomic(meta); | |
1697 | meta_offset += meta_iod->sg[i].length; | |
1698 | } | |
1699 | } | |
1700 | ||
1701 | dma_free_coherent(&dev->pci_dev->dev, meta_len, meta_mem, | |
1702 | meta_dma_addr); | |
1703 | } | |
1704 | ||
1705 | unmap: | |
1c2ad9fa | 1706 | nvme_unmap_user_pages(dev, io.opcode & 1, iod); |
eca18b23 | 1707 | nvme_free_iod(dev, iod); |
f410c680 KB |
1708 | |
1709 | if (meta_iod) { | |
1710 | nvme_unmap_user_pages(dev, io.opcode & 1, meta_iod); | |
1711 | nvme_free_iod(dev, meta_iod); | |
1712 | } | |
1713 | ||
a53295b6 MW |
1714 | return status; |
1715 | } | |
1716 | ||
50af8bae | 1717 | static int nvme_user_admin_cmd(struct nvme_dev *dev, |
6bbf1acd | 1718 | struct nvme_admin_cmd __user *ucmd) |
6ee44cdc | 1719 | { |
6bbf1acd | 1720 | struct nvme_admin_cmd cmd; |
6ee44cdc | 1721 | struct nvme_command c; |
eca18b23 | 1722 | int status, length; |
c7d36ab8 | 1723 | struct nvme_iod *uninitialized_var(iod); |
94f370ca | 1724 | unsigned timeout; |
6ee44cdc | 1725 | |
6bbf1acd MW |
1726 | if (!capable(CAP_SYS_ADMIN)) |
1727 | return -EACCES; | |
1728 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
6ee44cdc | 1729 | return -EFAULT; |
6ee44cdc MW |
1730 | |
1731 | memset(&c, 0, sizeof(c)); | |
6bbf1acd MW |
1732 | c.common.opcode = cmd.opcode; |
1733 | c.common.flags = cmd.flags; | |
1734 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1735 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1736 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1737 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
1738 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
1739 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
1740 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
1741 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
1742 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
1743 | ||
1744 | length = cmd.data_len; | |
1745 | if (cmd.data_len) { | |
49742188 MW |
1746 | iod = nvme_map_user_pages(dev, cmd.opcode & 1, cmd.addr, |
1747 | length); | |
eca18b23 MW |
1748 | if (IS_ERR(iod)) |
1749 | return PTR_ERR(iod); | |
edd10d33 KB |
1750 | length = nvme_setup_prps(dev, iod, length, GFP_KERNEL); |
1751 | c.common.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
1752 | c.common.prp2 = cpu_to_le64(iod->first_dma); | |
6bbf1acd MW |
1753 | } |
1754 | ||
94f370ca KB |
1755 | timeout = cmd.timeout_ms ? msecs_to_jiffies(cmd.timeout_ms) : |
1756 | ADMIN_TIMEOUT; | |
6bbf1acd | 1757 | if (length != cmd.data_len) |
b77954cb MW |
1758 | status = -ENOMEM; |
1759 | else | |
4f5099af | 1760 | status = nvme_submit_sync_cmd(dev, 0, &c, &cmd.result, timeout); |
eca18b23 | 1761 | |
6bbf1acd | 1762 | if (cmd.data_len) { |
1c2ad9fa | 1763 | nvme_unmap_user_pages(dev, cmd.opcode & 1, iod); |
eca18b23 | 1764 | nvme_free_iod(dev, iod); |
6bbf1acd | 1765 | } |
f4f117f6 | 1766 | |
cf90bc48 | 1767 | if ((status >= 0) && copy_to_user(&ucmd->result, &cmd.result, |
f4f117f6 KB |
1768 | sizeof(cmd.result))) |
1769 | status = -EFAULT; | |
1770 | ||
6ee44cdc MW |
1771 | return status; |
1772 | } | |
1773 | ||
b60503ba MW |
1774 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, |
1775 | unsigned long arg) | |
1776 | { | |
1777 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1778 | ||
1779 | switch (cmd) { | |
6bbf1acd | 1780 | case NVME_IOCTL_ID: |
c3bfe717 | 1781 | force_successful_syscall_return(); |
6bbf1acd MW |
1782 | return ns->ns_id; |
1783 | case NVME_IOCTL_ADMIN_CMD: | |
50af8bae | 1784 | return nvme_user_admin_cmd(ns->dev, (void __user *)arg); |
a53295b6 MW |
1785 | case NVME_IOCTL_SUBMIT_IO: |
1786 | return nvme_submit_io(ns, (void __user *)arg); | |
5d0f6131 VV |
1787 | case SG_GET_VERSION_NUM: |
1788 | return nvme_sg_get_version_num((void __user *)arg); | |
1789 | case SG_IO: | |
1790 | return nvme_sg_io(ns, (void __user *)arg); | |
b60503ba MW |
1791 | default: |
1792 | return -ENOTTY; | |
1793 | } | |
1794 | } | |
1795 | ||
320a3827 KB |
1796 | #ifdef CONFIG_COMPAT |
1797 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
1798 | unsigned int cmd, unsigned long arg) | |
1799 | { | |
1800 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1801 | ||
1802 | switch (cmd) { | |
1803 | case SG_IO: | |
1804 | return nvme_sg_io32(ns, arg); | |
1805 | } | |
1806 | return nvme_ioctl(bdev, mode, cmd, arg); | |
1807 | } | |
1808 | #else | |
1809 | #define nvme_compat_ioctl NULL | |
1810 | #endif | |
1811 | ||
9ac27090 KB |
1812 | static int nvme_open(struct block_device *bdev, fmode_t mode) |
1813 | { | |
1814 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1815 | struct nvme_dev *dev = ns->dev; | |
1816 | ||
1817 | kref_get(&dev->kref); | |
1818 | return 0; | |
1819 | } | |
1820 | ||
1821 | static void nvme_free_dev(struct kref *kref); | |
1822 | ||
1823 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
1824 | { | |
1825 | struct nvme_ns *ns = disk->private_data; | |
1826 | struct nvme_dev *dev = ns->dev; | |
1827 | ||
1828 | kref_put(&dev->kref, nvme_free_dev); | |
1829 | } | |
1830 | ||
4cc09e2d KB |
1831 | static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo) |
1832 | { | |
1833 | /* some standard values */ | |
1834 | geo->heads = 1 << 6; | |
1835 | geo->sectors = 1 << 5; | |
1836 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
1837 | return 0; | |
1838 | } | |
1839 | ||
b60503ba MW |
1840 | static const struct block_device_operations nvme_fops = { |
1841 | .owner = THIS_MODULE, | |
1842 | .ioctl = nvme_ioctl, | |
320a3827 | 1843 | .compat_ioctl = nvme_compat_ioctl, |
9ac27090 KB |
1844 | .open = nvme_open, |
1845 | .release = nvme_release, | |
4cc09e2d | 1846 | .getgeo = nvme_getgeo, |
b60503ba MW |
1847 | }; |
1848 | ||
edd10d33 KB |
1849 | static void nvme_resubmit_iods(struct nvme_queue *nvmeq) |
1850 | { | |
1851 | struct nvme_iod *iod, *next; | |
1852 | ||
1853 | list_for_each_entry_safe(iod, next, &nvmeq->iod_bio, node) { | |
1854 | if (unlikely(nvme_submit_iod(nvmeq, iod))) | |
1855 | break; | |
1856 | list_del(&iod->node); | |
1857 | if (bio_list_empty(&nvmeq->sq_cong) && | |
1858 | list_empty(&nvmeq->iod_bio)) | |
1859 | remove_wait_queue(&nvmeq->sq_full, | |
1860 | &nvmeq->sq_cong_wait); | |
1861 | } | |
1862 | } | |
1863 | ||
1fa6aead MW |
1864 | static void nvme_resubmit_bios(struct nvme_queue *nvmeq) |
1865 | { | |
1866 | while (bio_list_peek(&nvmeq->sq_cong)) { | |
1867 | struct bio *bio = bio_list_pop(&nvmeq->sq_cong); | |
1868 | struct nvme_ns *ns = bio->bi_bdev->bd_disk->private_data; | |
427e9708 | 1869 | |
edd10d33 KB |
1870 | if (bio_list_empty(&nvmeq->sq_cong) && |
1871 | list_empty(&nvmeq->iod_bio)) | |
427e9708 KB |
1872 | remove_wait_queue(&nvmeq->sq_full, |
1873 | &nvmeq->sq_cong_wait); | |
1fa6aead | 1874 | if (nvme_submit_bio_queue(nvmeq, ns, bio)) { |
edd10d33 | 1875 | if (!waitqueue_active(&nvmeq->sq_full)) |
427e9708 KB |
1876 | add_wait_queue(&nvmeq->sq_full, |
1877 | &nvmeq->sq_cong_wait); | |
1fa6aead MW |
1878 | bio_list_add_head(&nvmeq->sq_cong, bio); |
1879 | break; | |
1880 | } | |
1881 | } | |
1882 | } | |
1883 | ||
6fccf938 KB |
1884 | static int nvme_submit_async_req(struct nvme_queue *nvmeq) |
1885 | { | |
1886 | struct nvme_command *c; | |
1887 | int cmdid; | |
1888 | ||
1889 | cmdid = alloc_cmdid(nvmeq, CMD_CTX_ASYNC, special_completion, 0); | |
1890 | if (cmdid < 0) | |
1891 | return cmdid; | |
1892 | ||
1893 | c = &nvmeq->sq_cmds[nvmeq->sq_tail]; | |
1894 | memset(c, 0, sizeof(*c)); | |
1895 | c->common.opcode = nvme_admin_async_event; | |
1896 | c->common.command_id = cmdid; | |
1897 | ||
1898 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
1899 | nvmeq->sq_tail = 0; | |
1900 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
1901 | ||
1902 | return 0; | |
1903 | } | |
1904 | ||
1fa6aead MW |
1905 | static int nvme_kthread(void *data) |
1906 | { | |
d4b4ff8e | 1907 | struct nvme_dev *dev, *next; |
1fa6aead MW |
1908 | |
1909 | while (!kthread_should_stop()) { | |
564a232c | 1910 | set_current_state(TASK_INTERRUPTIBLE); |
1fa6aead | 1911 | spin_lock(&dev_list_lock); |
d4b4ff8e | 1912 | list_for_each_entry_safe(dev, next, &dev_list, node) { |
1fa6aead | 1913 | int i; |
d4b4ff8e KB |
1914 | if (readl(&dev->bar->csts) & NVME_CSTS_CFS && |
1915 | dev->initialized) { | |
1916 | if (work_busy(&dev->reset_work)) | |
1917 | continue; | |
1918 | list_del_init(&dev->node); | |
1919 | dev_warn(&dev->pci_dev->dev, | |
1920 | "Failed status, reset controller\n"); | |
9ca97374 | 1921 | dev->reset_workfn = nvme_reset_failed_dev; |
d4b4ff8e KB |
1922 | queue_work(nvme_workq, &dev->reset_work); |
1923 | continue; | |
1924 | } | |
5a92e700 | 1925 | rcu_read_lock(); |
1fa6aead | 1926 | for (i = 0; i < dev->queue_count; i++) { |
5a92e700 KB |
1927 | struct nvme_queue *nvmeq = |
1928 | rcu_dereference(dev->queues[i]); | |
740216fc MW |
1929 | if (!nvmeq) |
1930 | continue; | |
1fa6aead | 1931 | spin_lock_irq(&nvmeq->q_lock); |
22404274 KB |
1932 | if (nvmeq->q_suspended) |
1933 | goto unlock; | |
bc57a0f7 | 1934 | nvme_process_cq(nvmeq); |
a09115b2 | 1935 | nvme_cancel_ios(nvmeq, true); |
1fa6aead | 1936 | nvme_resubmit_bios(nvmeq); |
edd10d33 | 1937 | nvme_resubmit_iods(nvmeq); |
6fccf938 KB |
1938 | |
1939 | while ((i == 0) && (dev->event_limit > 0)) { | |
1940 | if (nvme_submit_async_req(nvmeq)) | |
1941 | break; | |
1942 | dev->event_limit--; | |
1943 | } | |
22404274 | 1944 | unlock: |
1fa6aead MW |
1945 | spin_unlock_irq(&nvmeq->q_lock); |
1946 | } | |
5a92e700 | 1947 | rcu_read_unlock(); |
1fa6aead MW |
1948 | } |
1949 | spin_unlock(&dev_list_lock); | |
acb7aa0d | 1950 | schedule_timeout(round_jiffies_relative(HZ)); |
1fa6aead MW |
1951 | } |
1952 | return 0; | |
1953 | } | |
1954 | ||
0e5e4f0e KB |
1955 | static void nvme_config_discard(struct nvme_ns *ns) |
1956 | { | |
1957 | u32 logical_block_size = queue_logical_block_size(ns->queue); | |
1958 | ns->queue->limits.discard_zeroes_data = 0; | |
1959 | ns->queue->limits.discard_alignment = logical_block_size; | |
1960 | ns->queue->limits.discard_granularity = logical_block_size; | |
1961 | ns->queue->limits.max_discard_sectors = 0xffffffff; | |
1962 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); | |
1963 | } | |
1964 | ||
c3bfe717 | 1965 | static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid, |
b60503ba MW |
1966 | struct nvme_id_ns *id, struct nvme_lba_range_type *rt) |
1967 | { | |
1968 | struct nvme_ns *ns; | |
1969 | struct gendisk *disk; | |
1970 | int lbaf; | |
1971 | ||
1972 | if (rt->attributes & NVME_LBART_ATTRIB_HIDE) | |
1973 | return NULL; | |
1974 | ||
1975 | ns = kzalloc(sizeof(*ns), GFP_KERNEL); | |
1976 | if (!ns) | |
1977 | return NULL; | |
1978 | ns->queue = blk_alloc_queue(GFP_KERNEL); | |
1979 | if (!ns->queue) | |
1980 | goto out_free_ns; | |
4eeb9215 MW |
1981 | ns->queue->queue_flags = QUEUE_FLAG_DEFAULT; |
1982 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue); | |
1983 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); | |
b277da0a | 1984 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, ns->queue); |
b60503ba MW |
1985 | blk_queue_make_request(ns->queue, nvme_make_request); |
1986 | ns->dev = dev; | |
1987 | ns->queue->queuedata = ns; | |
1988 | ||
469071a3 | 1989 | disk = alloc_disk(0); |
b60503ba MW |
1990 | if (!disk) |
1991 | goto out_free_queue; | |
5aff9382 | 1992 | ns->ns_id = nsid; |
b60503ba MW |
1993 | ns->disk = disk; |
1994 | lbaf = id->flbas & 0xf; | |
1995 | ns->lba_shift = id->lbaf[lbaf].ds; | |
f410c680 | 1996 | ns->ms = le16_to_cpu(id->lbaf[lbaf].ms); |
e9ef4636 | 1997 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); |
8fc23e03 KB |
1998 | if (dev->max_hw_sectors) |
1999 | blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors); | |
a7d2ce28 KB |
2000 | if (dev->vwc & NVME_CTRL_VWC_PRESENT) |
2001 | blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA); | |
b60503ba MW |
2002 | |
2003 | disk->major = nvme_major; | |
469071a3 | 2004 | disk->first_minor = 0; |
b60503ba MW |
2005 | disk->fops = &nvme_fops; |
2006 | disk->private_data = ns; | |
2007 | disk->queue = ns->queue; | |
388f037f | 2008 | disk->driverfs_dev = &dev->pci_dev->dev; |
469071a3 | 2009 | disk->flags = GENHD_FL_EXT_DEVT; |
5aff9382 | 2010 | sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid); |
b60503ba MW |
2011 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); |
2012 | ||
0e5e4f0e KB |
2013 | if (dev->oncs & NVME_CTRL_ONCS_DSM) |
2014 | nvme_config_discard(ns); | |
2015 | ||
b60503ba MW |
2016 | return ns; |
2017 | ||
2018 | out_free_queue: | |
2019 | blk_cleanup_queue(ns->queue); | |
2020 | out_free_ns: | |
2021 | kfree(ns); | |
2022 | return NULL; | |
2023 | } | |
2024 | ||
42f61420 KB |
2025 | static int nvme_find_closest_node(int node) |
2026 | { | |
2027 | int n, val, min_val = INT_MAX, best_node = node; | |
2028 | ||
2029 | for_each_online_node(n) { | |
2030 | if (n == node) | |
2031 | continue; | |
2032 | val = node_distance(node, n); | |
2033 | if (val < min_val) { | |
2034 | min_val = val; | |
2035 | best_node = n; | |
2036 | } | |
2037 | } | |
2038 | return best_node; | |
2039 | } | |
2040 | ||
2041 | static void nvme_set_queue_cpus(cpumask_t *qmask, struct nvme_queue *nvmeq, | |
2042 | int count) | |
2043 | { | |
2044 | int cpu; | |
2045 | for_each_cpu(cpu, qmask) { | |
2046 | if (cpumask_weight(nvmeq->cpu_mask) >= count) | |
2047 | break; | |
2048 | if (!cpumask_test_and_set_cpu(cpu, nvmeq->cpu_mask)) | |
2049 | *per_cpu_ptr(nvmeq->dev->io_queue, cpu) = nvmeq->qid; | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | static void nvme_add_cpus(cpumask_t *mask, const cpumask_t *unassigned_cpus, | |
2054 | const cpumask_t *new_mask, struct nvme_queue *nvmeq, int cpus_per_queue) | |
2055 | { | |
2056 | int next_cpu; | |
2057 | for_each_cpu(next_cpu, new_mask) { | |
2058 | cpumask_or(mask, mask, get_cpu_mask(next_cpu)); | |
2059 | cpumask_or(mask, mask, topology_thread_cpumask(next_cpu)); | |
2060 | cpumask_and(mask, mask, unassigned_cpus); | |
2061 | nvme_set_queue_cpus(mask, nvmeq, cpus_per_queue); | |
2062 | } | |
2063 | } | |
2064 | ||
2065 | static void nvme_create_io_queues(struct nvme_dev *dev) | |
2066 | { | |
2067 | unsigned i, max; | |
2068 | ||
2069 | max = min(dev->max_qid, num_online_cpus()); | |
2070 | for (i = dev->queue_count; i <= max; i++) | |
2071 | if (!nvme_alloc_queue(dev, i, dev->q_depth, i - 1)) | |
2072 | break; | |
2073 | ||
2074 | max = min(dev->queue_count - 1, num_online_cpus()); | |
2075 | for (i = dev->online_queues; i <= max; i++) | |
2076 | if (nvme_create_queue(raw_nvmeq(dev, i), i)) | |
2077 | break; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * If there are fewer queues than online cpus, this will try to optimally | |
2082 | * assign a queue to multiple cpus by grouping cpus that are "close" together: | |
2083 | * thread siblings, core, socket, closest node, then whatever else is | |
2084 | * available. | |
2085 | */ | |
2086 | static void nvme_assign_io_queues(struct nvme_dev *dev) | |
2087 | { | |
2088 | unsigned cpu, cpus_per_queue, queues, remainder, i; | |
2089 | cpumask_var_t unassigned_cpus; | |
2090 | ||
2091 | nvme_create_io_queues(dev); | |
2092 | ||
2093 | queues = min(dev->online_queues - 1, num_online_cpus()); | |
2094 | if (!queues) | |
2095 | return; | |
2096 | ||
2097 | cpus_per_queue = num_online_cpus() / queues; | |
2098 | remainder = queues - (num_online_cpus() - queues * cpus_per_queue); | |
2099 | ||
2100 | if (!alloc_cpumask_var(&unassigned_cpus, GFP_KERNEL)) | |
2101 | return; | |
2102 | ||
2103 | cpumask_copy(unassigned_cpus, cpu_online_mask); | |
2104 | cpu = cpumask_first(unassigned_cpus); | |
2105 | for (i = 1; i <= queues; i++) { | |
2106 | struct nvme_queue *nvmeq = lock_nvmeq(dev, i); | |
2107 | cpumask_t mask; | |
2108 | ||
2109 | cpumask_clear(nvmeq->cpu_mask); | |
2110 | if (!cpumask_weight(unassigned_cpus)) { | |
2111 | unlock_nvmeq(nvmeq); | |
2112 | break; | |
2113 | } | |
2114 | ||
2115 | mask = *get_cpu_mask(cpu); | |
2116 | nvme_set_queue_cpus(&mask, nvmeq, cpus_per_queue); | |
2117 | if (cpus_weight(mask) < cpus_per_queue) | |
2118 | nvme_add_cpus(&mask, unassigned_cpus, | |
2119 | topology_thread_cpumask(cpu), | |
2120 | nvmeq, cpus_per_queue); | |
2121 | if (cpus_weight(mask) < cpus_per_queue) | |
2122 | nvme_add_cpus(&mask, unassigned_cpus, | |
2123 | topology_core_cpumask(cpu), | |
2124 | nvmeq, cpus_per_queue); | |
2125 | if (cpus_weight(mask) < cpus_per_queue) | |
2126 | nvme_add_cpus(&mask, unassigned_cpus, | |
2127 | cpumask_of_node(cpu_to_node(cpu)), | |
2128 | nvmeq, cpus_per_queue); | |
2129 | if (cpus_weight(mask) < cpus_per_queue) | |
2130 | nvme_add_cpus(&mask, unassigned_cpus, | |
2131 | cpumask_of_node( | |
2132 | nvme_find_closest_node( | |
2133 | cpu_to_node(cpu))), | |
2134 | nvmeq, cpus_per_queue); | |
2135 | if (cpus_weight(mask) < cpus_per_queue) | |
2136 | nvme_add_cpus(&mask, unassigned_cpus, | |
2137 | unassigned_cpus, | |
2138 | nvmeq, cpus_per_queue); | |
2139 | ||
2140 | WARN(cpumask_weight(nvmeq->cpu_mask) != cpus_per_queue, | |
2141 | "nvme%d qid:%d mis-matched queue-to-cpu assignment\n", | |
2142 | dev->instance, i); | |
2143 | ||
2144 | irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector, | |
2145 | nvmeq->cpu_mask); | |
2146 | cpumask_andnot(unassigned_cpus, unassigned_cpus, | |
2147 | nvmeq->cpu_mask); | |
2148 | cpu = cpumask_next(cpu, unassigned_cpus); | |
2149 | if (remainder && !--remainder) | |
2150 | cpus_per_queue++; | |
2151 | unlock_nvmeq(nvmeq); | |
2152 | } | |
2153 | WARN(cpumask_weight(unassigned_cpus), "nvme%d unassigned online cpus\n", | |
2154 | dev->instance); | |
2155 | i = 0; | |
2156 | cpumask_andnot(unassigned_cpus, cpu_possible_mask, cpu_online_mask); | |
2157 | for_each_cpu(cpu, unassigned_cpus) | |
2158 | *per_cpu_ptr(dev->io_queue, cpu) = (i++ % queues) + 1; | |
2159 | free_cpumask_var(unassigned_cpus); | |
2160 | } | |
2161 | ||
b3b06812 | 2162 | static int set_queue_count(struct nvme_dev *dev, int count) |
b60503ba MW |
2163 | { |
2164 | int status; | |
2165 | u32 result; | |
b3b06812 | 2166 | u32 q_count = (count - 1) | ((count - 1) << 16); |
b60503ba | 2167 | |
df348139 | 2168 | status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0, |
bc5fc7e4 | 2169 | &result); |
27e8166c MW |
2170 | if (status < 0) |
2171 | return status; | |
2172 | if (status > 0) { | |
2173 | dev_err(&dev->pci_dev->dev, "Could not set queue count (%d)\n", | |
2174 | status); | |
badc34d4 | 2175 | return 0; |
27e8166c | 2176 | } |
b60503ba MW |
2177 | return min(result & 0xffff, result >> 16) + 1; |
2178 | } | |
2179 | ||
9d713c2b KB |
2180 | static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
2181 | { | |
b80d5ccc | 2182 | return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride); |
9d713c2b KB |
2183 | } |
2184 | ||
f3db22fe KB |
2185 | static void nvme_cpu_workfn(struct work_struct *work) |
2186 | { | |
2187 | struct nvme_dev *dev = container_of(work, struct nvme_dev, cpu_work); | |
2188 | if (dev->initialized) | |
2189 | nvme_assign_io_queues(dev); | |
2190 | } | |
2191 | ||
33b1e95c KB |
2192 | static int nvme_cpu_notify(struct notifier_block *self, |
2193 | unsigned long action, void *hcpu) | |
2194 | { | |
f3db22fe KB |
2195 | struct nvme_dev *dev; |
2196 | ||
33b1e95c KB |
2197 | switch (action) { |
2198 | case CPU_ONLINE: | |
2199 | case CPU_DEAD: | |
f3db22fe KB |
2200 | spin_lock(&dev_list_lock); |
2201 | list_for_each_entry(dev, &dev_list, node) | |
2202 | schedule_work(&dev->cpu_work); | |
2203 | spin_unlock(&dev_list_lock); | |
33b1e95c KB |
2204 | break; |
2205 | } | |
2206 | return NOTIFY_OK; | |
2207 | } | |
2208 | ||
8d85fce7 | 2209 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 2210 | { |
5a92e700 | 2211 | struct nvme_queue *adminq = raw_nvmeq(dev, 0); |
fa08a396 | 2212 | struct pci_dev *pdev = dev->pci_dev; |
42f61420 | 2213 | int result, i, vecs, nr_io_queues, size; |
b60503ba | 2214 | |
42f61420 | 2215 | nr_io_queues = num_possible_cpus(); |
b348b7d5 | 2216 | result = set_queue_count(dev, nr_io_queues); |
badc34d4 | 2217 | if (result <= 0) |
1b23484b | 2218 | return result; |
b348b7d5 MW |
2219 | if (result < nr_io_queues) |
2220 | nr_io_queues = result; | |
b60503ba | 2221 | |
9d713c2b KB |
2222 | size = db_bar_size(dev, nr_io_queues); |
2223 | if (size > 8192) { | |
f1938f6e | 2224 | iounmap(dev->bar); |
9d713c2b KB |
2225 | do { |
2226 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
2227 | if (dev->bar) | |
2228 | break; | |
2229 | if (!--nr_io_queues) | |
2230 | return -ENOMEM; | |
2231 | size = db_bar_size(dev, nr_io_queues); | |
2232 | } while (1); | |
f1938f6e | 2233 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
5a92e700 | 2234 | adminq->q_db = dev->dbs; |
f1938f6e MW |
2235 | } |
2236 | ||
9d713c2b | 2237 | /* Deregister the admin queue's interrupt */ |
3193f07b | 2238 | free_irq(dev->entry[0].vector, adminq); |
9d713c2b | 2239 | |
be577fab | 2240 | for (i = 0; i < nr_io_queues; i++) |
1b23484b | 2241 | dev->entry[i].entry = i; |
be577fab AG |
2242 | vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues); |
2243 | if (vecs < 0) { | |
2244 | vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32)); | |
2245 | if (vecs < 0) { | |
2246 | vecs = 1; | |
2247 | } else { | |
2248 | for (i = 0; i < vecs; i++) | |
2249 | dev->entry[i].vector = i + pdev->irq; | |
fa08a396 RRG |
2250 | } |
2251 | } | |
2252 | ||
063a8096 MW |
2253 | /* |
2254 | * Should investigate if there's a performance win from allocating | |
2255 | * more queues than interrupt vectors; it might allow the submission | |
2256 | * path to scale better, even if the receive path is limited by the | |
2257 | * number of interrupts. | |
2258 | */ | |
2259 | nr_io_queues = vecs; | |
42f61420 | 2260 | dev->max_qid = nr_io_queues; |
063a8096 | 2261 | |
3193f07b | 2262 | result = queue_request_irq(dev, adminq, adminq->irqname); |
9d713c2b | 2263 | if (result) { |
3193f07b | 2264 | adminq->q_suspended = 1; |
22404274 | 2265 | goto free_queues; |
9d713c2b | 2266 | } |
1b23484b | 2267 | |
cd638946 | 2268 | /* Free previously allocated queues that are no longer usable */ |
42f61420 KB |
2269 | nvme_free_queues(dev, nr_io_queues + 1); |
2270 | nvme_assign_io_queues(dev); | |
9ecdc946 | 2271 | |
22404274 | 2272 | return 0; |
b60503ba | 2273 | |
22404274 | 2274 | free_queues: |
a1a5ef99 | 2275 | nvme_free_queues(dev, 1); |
22404274 | 2276 | return result; |
b60503ba MW |
2277 | } |
2278 | ||
422ef0c7 MW |
2279 | /* |
2280 | * Return: error value if an error occurred setting up the queues or calling | |
2281 | * Identify Device. 0 if these succeeded, even if adding some of the | |
2282 | * namespaces failed. At the moment, these failures are silent. TBD which | |
2283 | * failures should be reported. | |
2284 | */ | |
8d85fce7 | 2285 | static int nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 2286 | { |
68608c26 | 2287 | struct pci_dev *pdev = dev->pci_dev; |
c3bfe717 MW |
2288 | int res; |
2289 | unsigned nn, i; | |
cbb6218f | 2290 | struct nvme_ns *ns; |
51814232 | 2291 | struct nvme_id_ctrl *ctrl; |
bc5fc7e4 MW |
2292 | struct nvme_id_ns *id_ns; |
2293 | void *mem; | |
b60503ba | 2294 | dma_addr_t dma_addr; |
159b67d7 | 2295 | int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12; |
b60503ba | 2296 | |
68608c26 | 2297 | mem = dma_alloc_coherent(&pdev->dev, 8192, &dma_addr, GFP_KERNEL); |
a9ef4343 KB |
2298 | if (!mem) |
2299 | return -ENOMEM; | |
b60503ba | 2300 | |
bc5fc7e4 | 2301 | res = nvme_identify(dev, 0, 1, dma_addr); |
b60503ba | 2302 | if (res) { |
27e8166c | 2303 | dev_err(&pdev->dev, "Identify Controller failed (%d)\n", res); |
b60503ba | 2304 | res = -EIO; |
cbb6218f | 2305 | goto out; |
b60503ba MW |
2306 | } |
2307 | ||
bc5fc7e4 | 2308 | ctrl = mem; |
51814232 | 2309 | nn = le32_to_cpup(&ctrl->nn); |
0e5e4f0e | 2310 | dev->oncs = le16_to_cpup(&ctrl->oncs); |
c30341dc | 2311 | dev->abort_limit = ctrl->acl + 1; |
a7d2ce28 | 2312 | dev->vwc = ctrl->vwc; |
6fccf938 | 2313 | dev->event_limit = min(ctrl->aerl + 1, 8); |
51814232 MW |
2314 | memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn)); |
2315 | memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn)); | |
2316 | memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr)); | |
159b67d7 | 2317 | if (ctrl->mdts) |
8fc23e03 | 2318 | dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9); |
68608c26 MW |
2319 | if ((pdev->vendor == PCI_VENDOR_ID_INTEL) && |
2320 | (pdev->device == 0x0953) && ctrl->vs[3]) | |
159b67d7 | 2321 | dev->stripe_size = 1 << (ctrl->vs[3] + shift); |
b60503ba | 2322 | |
bc5fc7e4 | 2323 | id_ns = mem; |
2b2c1896 | 2324 | for (i = 1; i <= nn; i++) { |
bc5fc7e4 | 2325 | res = nvme_identify(dev, i, 0, dma_addr); |
b60503ba MW |
2326 | if (res) |
2327 | continue; | |
2328 | ||
bc5fc7e4 | 2329 | if (id_ns->ncap == 0) |
b60503ba MW |
2330 | continue; |
2331 | ||
bc5fc7e4 | 2332 | res = nvme_get_features(dev, NVME_FEAT_LBA_RANGE, i, |
08df1e05 | 2333 | dma_addr + 4096, NULL); |
b60503ba | 2334 | if (res) |
12209036 | 2335 | memset(mem + 4096, 0, 4096); |
b60503ba | 2336 | |
bc5fc7e4 | 2337 | ns = nvme_alloc_ns(dev, i, mem, mem + 4096); |
b60503ba MW |
2338 | if (ns) |
2339 | list_add_tail(&ns->list, &dev->namespaces); | |
2340 | } | |
2341 | list_for_each_entry(ns, &dev->namespaces, list) | |
2342 | add_disk(ns->disk); | |
422ef0c7 | 2343 | res = 0; |
b60503ba | 2344 | |
bc5fc7e4 | 2345 | out: |
684f5c20 | 2346 | dma_free_coherent(&dev->pci_dev->dev, 8192, mem, dma_addr); |
b60503ba MW |
2347 | return res; |
2348 | } | |
2349 | ||
0877cb0d KB |
2350 | static int nvme_dev_map(struct nvme_dev *dev) |
2351 | { | |
42f61420 | 2352 | u64 cap; |
0877cb0d KB |
2353 | int bars, result = -ENOMEM; |
2354 | struct pci_dev *pdev = dev->pci_dev; | |
2355 | ||
2356 | if (pci_enable_device_mem(pdev)) | |
2357 | return result; | |
2358 | ||
2359 | dev->entry[0].vector = pdev->irq; | |
2360 | pci_set_master(pdev); | |
2361 | bars = pci_select_bars(pdev, IORESOURCE_MEM); | |
2362 | if (pci_request_selected_regions(pdev, bars, "nvme")) | |
2363 | goto disable_pci; | |
2364 | ||
052d0efa RK |
2365 | if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) && |
2366 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32))) | |
2367 | goto disable; | |
0877cb0d | 2368 | |
0877cb0d KB |
2369 | dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); |
2370 | if (!dev->bar) | |
2371 | goto disable; | |
0e53d180 KB |
2372 | if (readl(&dev->bar->csts) == -1) { |
2373 | result = -ENODEV; | |
2374 | goto unmap; | |
2375 | } | |
42f61420 KB |
2376 | cap = readq(&dev->bar->cap); |
2377 | dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); | |
2378 | dev->db_stride = 1 << NVME_CAP_STRIDE(cap); | |
0877cb0d KB |
2379 | dev->dbs = ((void __iomem *)dev->bar) + 4096; |
2380 | ||
2381 | return 0; | |
2382 | ||
0e53d180 KB |
2383 | unmap: |
2384 | iounmap(dev->bar); | |
2385 | dev->bar = NULL; | |
0877cb0d KB |
2386 | disable: |
2387 | pci_release_regions(pdev); | |
2388 | disable_pci: | |
2389 | pci_disable_device(pdev); | |
2390 | return result; | |
2391 | } | |
2392 | ||
2393 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
2394 | { | |
2395 | if (dev->pci_dev->msi_enabled) | |
2396 | pci_disable_msi(dev->pci_dev); | |
2397 | else if (dev->pci_dev->msix_enabled) | |
2398 | pci_disable_msix(dev->pci_dev); | |
2399 | ||
2400 | if (dev->bar) { | |
2401 | iounmap(dev->bar); | |
2402 | dev->bar = NULL; | |
9a6b9458 | 2403 | pci_release_regions(dev->pci_dev); |
0877cb0d KB |
2404 | } |
2405 | ||
0877cb0d KB |
2406 | if (pci_is_enabled(dev->pci_dev)) |
2407 | pci_disable_device(dev->pci_dev); | |
2408 | } | |
2409 | ||
4d115420 KB |
2410 | struct nvme_delq_ctx { |
2411 | struct task_struct *waiter; | |
2412 | struct kthread_worker *worker; | |
2413 | atomic_t refcount; | |
2414 | }; | |
2415 | ||
2416 | static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev) | |
2417 | { | |
2418 | dq->waiter = current; | |
2419 | mb(); | |
2420 | ||
2421 | for (;;) { | |
2422 | set_current_state(TASK_KILLABLE); | |
2423 | if (!atomic_read(&dq->refcount)) | |
2424 | break; | |
2425 | if (!schedule_timeout(ADMIN_TIMEOUT) || | |
2426 | fatal_signal_pending(current)) { | |
2427 | set_current_state(TASK_RUNNING); | |
2428 | ||
2429 | nvme_disable_ctrl(dev, readq(&dev->bar->cap)); | |
2430 | nvme_disable_queue(dev, 0); | |
2431 | ||
2432 | send_sig(SIGKILL, dq->worker->task, 1); | |
2433 | flush_kthread_worker(dq->worker); | |
2434 | return; | |
2435 | } | |
2436 | } | |
2437 | set_current_state(TASK_RUNNING); | |
2438 | } | |
2439 | ||
2440 | static void nvme_put_dq(struct nvme_delq_ctx *dq) | |
2441 | { | |
2442 | atomic_dec(&dq->refcount); | |
2443 | if (dq->waiter) | |
2444 | wake_up_process(dq->waiter); | |
2445 | } | |
2446 | ||
2447 | static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq) | |
2448 | { | |
2449 | atomic_inc(&dq->refcount); | |
2450 | return dq; | |
2451 | } | |
2452 | ||
2453 | static void nvme_del_queue_end(struct nvme_queue *nvmeq) | |
2454 | { | |
2455 | struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx; | |
2456 | ||
2457 | nvme_clear_queue(nvmeq); | |
2458 | nvme_put_dq(dq); | |
2459 | } | |
2460 | ||
2461 | static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode, | |
2462 | kthread_work_func_t fn) | |
2463 | { | |
2464 | struct nvme_command c; | |
2465 | ||
2466 | memset(&c, 0, sizeof(c)); | |
2467 | c.delete_queue.opcode = opcode; | |
2468 | c.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
2469 | ||
2470 | init_kthread_work(&nvmeq->cmdinfo.work, fn); | |
2471 | return nvme_submit_admin_cmd_async(nvmeq->dev, &c, &nvmeq->cmdinfo); | |
2472 | } | |
2473 | ||
2474 | static void nvme_del_cq_work_handler(struct kthread_work *work) | |
2475 | { | |
2476 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2477 | cmdinfo.work); | |
2478 | nvme_del_queue_end(nvmeq); | |
2479 | } | |
2480 | ||
2481 | static int nvme_delete_cq(struct nvme_queue *nvmeq) | |
2482 | { | |
2483 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq, | |
2484 | nvme_del_cq_work_handler); | |
2485 | } | |
2486 | ||
2487 | static void nvme_del_sq_work_handler(struct kthread_work *work) | |
2488 | { | |
2489 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2490 | cmdinfo.work); | |
2491 | int status = nvmeq->cmdinfo.status; | |
2492 | ||
2493 | if (!status) | |
2494 | status = nvme_delete_cq(nvmeq); | |
2495 | if (status) | |
2496 | nvme_del_queue_end(nvmeq); | |
2497 | } | |
2498 | ||
2499 | static int nvme_delete_sq(struct nvme_queue *nvmeq) | |
2500 | { | |
2501 | return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq, | |
2502 | nvme_del_sq_work_handler); | |
2503 | } | |
2504 | ||
2505 | static void nvme_del_queue_start(struct kthread_work *work) | |
2506 | { | |
2507 | struct nvme_queue *nvmeq = container_of(work, struct nvme_queue, | |
2508 | cmdinfo.work); | |
2509 | allow_signal(SIGKILL); | |
2510 | if (nvme_delete_sq(nvmeq)) | |
2511 | nvme_del_queue_end(nvmeq); | |
2512 | } | |
2513 | ||
2514 | static void nvme_disable_io_queues(struct nvme_dev *dev) | |
2515 | { | |
2516 | int i; | |
2517 | DEFINE_KTHREAD_WORKER_ONSTACK(worker); | |
2518 | struct nvme_delq_ctx dq; | |
2519 | struct task_struct *kworker_task = kthread_run(kthread_worker_fn, | |
2520 | &worker, "nvme%d", dev->instance); | |
2521 | ||
2522 | if (IS_ERR(kworker_task)) { | |
2523 | dev_err(&dev->pci_dev->dev, | |
2524 | "Failed to create queue del task\n"); | |
2525 | for (i = dev->queue_count - 1; i > 0; i--) | |
2526 | nvme_disable_queue(dev, i); | |
2527 | return; | |
2528 | } | |
2529 | ||
2530 | dq.waiter = NULL; | |
2531 | atomic_set(&dq.refcount, 0); | |
2532 | dq.worker = &worker; | |
2533 | for (i = dev->queue_count - 1; i > 0; i--) { | |
5a92e700 | 2534 | struct nvme_queue *nvmeq = raw_nvmeq(dev, i); |
4d115420 KB |
2535 | |
2536 | if (nvme_suspend_queue(nvmeq)) | |
2537 | continue; | |
2538 | nvmeq->cmdinfo.ctx = nvme_get_dq(&dq); | |
2539 | nvmeq->cmdinfo.worker = dq.worker; | |
2540 | init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start); | |
2541 | queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work); | |
2542 | } | |
2543 | nvme_wait_dq(&dq, dev); | |
2544 | kthread_stop(kworker_task); | |
2545 | } | |
2546 | ||
b9afca3e DM |
2547 | /* |
2548 | * Remove the node from the device list and check | |
2549 | * for whether or not we need to stop the nvme_thread. | |
2550 | */ | |
2551 | static void nvme_dev_list_remove(struct nvme_dev *dev) | |
2552 | { | |
2553 | struct task_struct *tmp = NULL; | |
2554 | ||
2555 | spin_lock(&dev_list_lock); | |
2556 | list_del_init(&dev->node); | |
2557 | if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) { | |
2558 | tmp = nvme_thread; | |
2559 | nvme_thread = NULL; | |
2560 | } | |
2561 | spin_unlock(&dev_list_lock); | |
2562 | ||
2563 | if (tmp) | |
2564 | kthread_stop(tmp); | |
2565 | } | |
2566 | ||
f0b50732 | 2567 | static void nvme_dev_shutdown(struct nvme_dev *dev) |
b60503ba | 2568 | { |
22404274 KB |
2569 | int i; |
2570 | ||
d4b4ff8e | 2571 | dev->initialized = 0; |
b9afca3e | 2572 | nvme_dev_list_remove(dev); |
1fa6aead | 2573 | |
4d115420 KB |
2574 | if (!dev->bar || (dev->bar && readl(&dev->bar->csts) == -1)) { |
2575 | for (i = dev->queue_count - 1; i >= 0; i--) { | |
5a92e700 | 2576 | struct nvme_queue *nvmeq = raw_nvmeq(dev, i); |
4d115420 KB |
2577 | nvme_suspend_queue(nvmeq); |
2578 | nvme_clear_queue(nvmeq); | |
2579 | } | |
2580 | } else { | |
2581 | nvme_disable_io_queues(dev); | |
1894d8f1 | 2582 | nvme_shutdown_ctrl(dev); |
4d115420 KB |
2583 | nvme_disable_queue(dev, 0); |
2584 | } | |
f0b50732 KB |
2585 | nvme_dev_unmap(dev); |
2586 | } | |
2587 | ||
2588 | static void nvme_dev_remove(struct nvme_dev *dev) | |
2589 | { | |
9ac27090 | 2590 | struct nvme_ns *ns; |
f0b50732 | 2591 | |
9ac27090 KB |
2592 | list_for_each_entry(ns, &dev->namespaces, list) { |
2593 | if (ns->disk->flags & GENHD_FL_UP) | |
2594 | del_gendisk(ns->disk); | |
2595 | if (!blk_queue_dying(ns->queue)) | |
2596 | blk_cleanup_queue(ns->queue); | |
b60503ba | 2597 | } |
b60503ba MW |
2598 | } |
2599 | ||
091b6092 MW |
2600 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
2601 | { | |
2602 | struct device *dmadev = &dev->pci_dev->dev; | |
2603 | dev->prp_page_pool = dma_pool_create("prp list page", dmadev, | |
2604 | PAGE_SIZE, PAGE_SIZE, 0); | |
2605 | if (!dev->prp_page_pool) | |
2606 | return -ENOMEM; | |
2607 | ||
99802a7a MW |
2608 | /* Optimisation for I/Os between 4k and 128k */ |
2609 | dev->prp_small_pool = dma_pool_create("prp list 256", dmadev, | |
2610 | 256, 256, 0); | |
2611 | if (!dev->prp_small_pool) { | |
2612 | dma_pool_destroy(dev->prp_page_pool); | |
2613 | return -ENOMEM; | |
2614 | } | |
091b6092 MW |
2615 | return 0; |
2616 | } | |
2617 | ||
2618 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
2619 | { | |
2620 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 2621 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
2622 | } |
2623 | ||
cd58ad7d QSA |
2624 | static DEFINE_IDA(nvme_instance_ida); |
2625 | ||
2626 | static int nvme_set_instance(struct nvme_dev *dev) | |
b60503ba | 2627 | { |
cd58ad7d QSA |
2628 | int instance, error; |
2629 | ||
2630 | do { | |
2631 | if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL)) | |
2632 | return -ENODEV; | |
2633 | ||
2634 | spin_lock(&dev_list_lock); | |
2635 | error = ida_get_new(&nvme_instance_ida, &instance); | |
2636 | spin_unlock(&dev_list_lock); | |
2637 | } while (error == -EAGAIN); | |
2638 | ||
2639 | if (error) | |
2640 | return -ENODEV; | |
2641 | ||
2642 | dev->instance = instance; | |
2643 | return 0; | |
b60503ba MW |
2644 | } |
2645 | ||
2646 | static void nvme_release_instance(struct nvme_dev *dev) | |
2647 | { | |
cd58ad7d QSA |
2648 | spin_lock(&dev_list_lock); |
2649 | ida_remove(&nvme_instance_ida, dev->instance); | |
2650 | spin_unlock(&dev_list_lock); | |
b60503ba MW |
2651 | } |
2652 | ||
9ac27090 KB |
2653 | static void nvme_free_namespaces(struct nvme_dev *dev) |
2654 | { | |
2655 | struct nvme_ns *ns, *next; | |
2656 | ||
2657 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) { | |
2658 | list_del(&ns->list); | |
2659 | put_disk(ns->disk); | |
2660 | kfree(ns); | |
2661 | } | |
2662 | } | |
2663 | ||
5e82e952 KB |
2664 | static void nvme_free_dev(struct kref *kref) |
2665 | { | |
2666 | struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref); | |
9ac27090 KB |
2667 | |
2668 | nvme_free_namespaces(dev); | |
42f61420 | 2669 | free_percpu(dev->io_queue); |
5e82e952 KB |
2670 | kfree(dev->queues); |
2671 | kfree(dev->entry); | |
2672 | kfree(dev); | |
2673 | } | |
2674 | ||
2675 | static int nvme_dev_open(struct inode *inode, struct file *f) | |
2676 | { | |
2677 | struct nvme_dev *dev = container_of(f->private_data, struct nvme_dev, | |
2678 | miscdev); | |
2679 | kref_get(&dev->kref); | |
2680 | f->private_data = dev; | |
2681 | return 0; | |
2682 | } | |
2683 | ||
2684 | static int nvme_dev_release(struct inode *inode, struct file *f) | |
2685 | { | |
2686 | struct nvme_dev *dev = f->private_data; | |
2687 | kref_put(&dev->kref, nvme_free_dev); | |
2688 | return 0; | |
2689 | } | |
2690 | ||
2691 | static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg) | |
2692 | { | |
2693 | struct nvme_dev *dev = f->private_data; | |
2694 | switch (cmd) { | |
2695 | case NVME_IOCTL_ADMIN_CMD: | |
2696 | return nvme_user_admin_cmd(dev, (void __user *)arg); | |
2697 | default: | |
2698 | return -ENOTTY; | |
2699 | } | |
2700 | } | |
2701 | ||
2702 | static const struct file_operations nvme_dev_fops = { | |
2703 | .owner = THIS_MODULE, | |
2704 | .open = nvme_dev_open, | |
2705 | .release = nvme_dev_release, | |
2706 | .unlocked_ioctl = nvme_dev_ioctl, | |
2707 | .compat_ioctl = nvme_dev_ioctl, | |
2708 | }; | |
2709 | ||
f0b50732 KB |
2710 | static int nvme_dev_start(struct nvme_dev *dev) |
2711 | { | |
2712 | int result; | |
b9afca3e | 2713 | bool start_thread = false; |
f0b50732 KB |
2714 | |
2715 | result = nvme_dev_map(dev); | |
2716 | if (result) | |
2717 | return result; | |
2718 | ||
2719 | result = nvme_configure_admin_queue(dev); | |
2720 | if (result) | |
2721 | goto unmap; | |
2722 | ||
2723 | spin_lock(&dev_list_lock); | |
b9afca3e DM |
2724 | if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) { |
2725 | start_thread = true; | |
2726 | nvme_thread = NULL; | |
2727 | } | |
f0b50732 KB |
2728 | list_add(&dev->node, &dev_list); |
2729 | spin_unlock(&dev_list_lock); | |
2730 | ||
b9afca3e DM |
2731 | if (start_thread) { |
2732 | nvme_thread = kthread_run(nvme_kthread, NULL, "nvme"); | |
2733 | wake_up(&nvme_kthread_wait); | |
2734 | } else | |
2735 | wait_event_killable(nvme_kthread_wait, nvme_thread); | |
2736 | ||
2737 | if (IS_ERR_OR_NULL(nvme_thread)) { | |
2738 | result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR; | |
2739 | goto disable; | |
2740 | } | |
2741 | ||
f0b50732 | 2742 | result = nvme_setup_io_queues(dev); |
badc34d4 | 2743 | if (result) |
f0b50732 KB |
2744 | goto disable; |
2745 | ||
d82e8bfd | 2746 | return result; |
f0b50732 KB |
2747 | |
2748 | disable: | |
a1a5ef99 | 2749 | nvme_disable_queue(dev, 0); |
b9afca3e | 2750 | nvme_dev_list_remove(dev); |
f0b50732 KB |
2751 | unmap: |
2752 | nvme_dev_unmap(dev); | |
2753 | return result; | |
2754 | } | |
2755 | ||
9a6b9458 KB |
2756 | static int nvme_remove_dead_ctrl(void *arg) |
2757 | { | |
2758 | struct nvme_dev *dev = (struct nvme_dev *)arg; | |
2759 | struct pci_dev *pdev = dev->pci_dev; | |
2760 | ||
2761 | if (pci_get_drvdata(pdev)) | |
c81f4975 | 2762 | pci_stop_and_remove_bus_device_locked(pdev); |
9a6b9458 KB |
2763 | kref_put(&dev->kref, nvme_free_dev); |
2764 | return 0; | |
2765 | } | |
2766 | ||
2767 | static void nvme_remove_disks(struct work_struct *ws) | |
2768 | { | |
9a6b9458 KB |
2769 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); |
2770 | ||
2771 | nvme_dev_remove(dev); | |
5a92e700 | 2772 | nvme_free_queues(dev, 1); |
9a6b9458 KB |
2773 | } |
2774 | ||
2775 | static int nvme_dev_resume(struct nvme_dev *dev) | |
2776 | { | |
2777 | int ret; | |
2778 | ||
2779 | ret = nvme_dev_start(dev); | |
badc34d4 | 2780 | if (ret) |
9a6b9458 | 2781 | return ret; |
badc34d4 | 2782 | if (dev->online_queues < 2) { |
9a6b9458 | 2783 | spin_lock(&dev_list_lock); |
9ca97374 | 2784 | dev->reset_workfn = nvme_remove_disks; |
9a6b9458 KB |
2785 | queue_work(nvme_workq, &dev->reset_work); |
2786 | spin_unlock(&dev_list_lock); | |
2787 | } | |
d4b4ff8e | 2788 | dev->initialized = 1; |
9a6b9458 KB |
2789 | return 0; |
2790 | } | |
2791 | ||
2792 | static void nvme_dev_reset(struct nvme_dev *dev) | |
2793 | { | |
2794 | nvme_dev_shutdown(dev); | |
2795 | if (nvme_dev_resume(dev)) { | |
2796 | dev_err(&dev->pci_dev->dev, "Device failed to resume\n"); | |
2797 | kref_get(&dev->kref); | |
2798 | if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d", | |
2799 | dev->instance))) { | |
2800 | dev_err(&dev->pci_dev->dev, | |
2801 | "Failed to start controller remove task\n"); | |
2802 | kref_put(&dev->kref, nvme_free_dev); | |
2803 | } | |
2804 | } | |
2805 | } | |
2806 | ||
2807 | static void nvme_reset_failed_dev(struct work_struct *ws) | |
2808 | { | |
2809 | struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work); | |
2810 | nvme_dev_reset(dev); | |
2811 | } | |
2812 | ||
9ca97374 TH |
2813 | static void nvme_reset_workfn(struct work_struct *work) |
2814 | { | |
2815 | struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work); | |
2816 | dev->reset_workfn(work); | |
2817 | } | |
2818 | ||
8d85fce7 | 2819 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 2820 | { |
0877cb0d | 2821 | int result = -ENOMEM; |
b60503ba MW |
2822 | struct nvme_dev *dev; |
2823 | ||
2824 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2825 | if (!dev) | |
2826 | return -ENOMEM; | |
2827 | dev->entry = kcalloc(num_possible_cpus(), sizeof(*dev->entry), | |
2828 | GFP_KERNEL); | |
2829 | if (!dev->entry) | |
2830 | goto free; | |
1b23484b MW |
2831 | dev->queues = kcalloc(num_possible_cpus() + 1, sizeof(void *), |
2832 | GFP_KERNEL); | |
b60503ba MW |
2833 | if (!dev->queues) |
2834 | goto free; | |
42f61420 KB |
2835 | dev->io_queue = alloc_percpu(unsigned short); |
2836 | if (!dev->io_queue) | |
2837 | goto free; | |
b60503ba MW |
2838 | |
2839 | INIT_LIST_HEAD(&dev->namespaces); | |
9ca97374 TH |
2840 | dev->reset_workfn = nvme_reset_failed_dev; |
2841 | INIT_WORK(&dev->reset_work, nvme_reset_workfn); | |
f3db22fe | 2842 | INIT_WORK(&dev->cpu_work, nvme_cpu_workfn); |
b60503ba | 2843 | dev->pci_dev = pdev; |
9a6b9458 | 2844 | pci_set_drvdata(pdev, dev); |
cd58ad7d QSA |
2845 | result = nvme_set_instance(dev); |
2846 | if (result) | |
0877cb0d | 2847 | goto free; |
b60503ba | 2848 | |
091b6092 MW |
2849 | result = nvme_setup_prp_pools(dev); |
2850 | if (result) | |
0877cb0d | 2851 | goto release; |
091b6092 | 2852 | |
fb35e914 | 2853 | kref_init(&dev->kref); |
f0b50732 | 2854 | result = nvme_dev_start(dev); |
badc34d4 | 2855 | if (result) |
0877cb0d | 2856 | goto release_pools; |
b60503ba | 2857 | |
badc34d4 KB |
2858 | if (dev->online_queues > 1) |
2859 | result = nvme_dev_add(dev); | |
d82e8bfd | 2860 | if (result) |
f0b50732 | 2861 | goto shutdown; |
740216fc | 2862 | |
5e82e952 KB |
2863 | scnprintf(dev->name, sizeof(dev->name), "nvme%d", dev->instance); |
2864 | dev->miscdev.minor = MISC_DYNAMIC_MINOR; | |
2865 | dev->miscdev.parent = &pdev->dev; | |
2866 | dev->miscdev.name = dev->name; | |
2867 | dev->miscdev.fops = &nvme_dev_fops; | |
2868 | result = misc_register(&dev->miscdev); | |
2869 | if (result) | |
2870 | goto remove; | |
2871 | ||
d4b4ff8e | 2872 | dev->initialized = 1; |
b60503ba MW |
2873 | return 0; |
2874 | ||
5e82e952 KB |
2875 | remove: |
2876 | nvme_dev_remove(dev); | |
9ac27090 | 2877 | nvme_free_namespaces(dev); |
f0b50732 KB |
2878 | shutdown: |
2879 | nvme_dev_shutdown(dev); | |
0877cb0d | 2880 | release_pools: |
a1a5ef99 | 2881 | nvme_free_queues(dev, 0); |
091b6092 | 2882 | nvme_release_prp_pools(dev); |
0877cb0d KB |
2883 | release: |
2884 | nvme_release_instance(dev); | |
b60503ba | 2885 | free: |
42f61420 | 2886 | free_percpu(dev->io_queue); |
b60503ba MW |
2887 | kfree(dev->queues); |
2888 | kfree(dev->entry); | |
2889 | kfree(dev); | |
2890 | return result; | |
2891 | } | |
2892 | ||
f0d54a54 KB |
2893 | static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) |
2894 | { | |
2895 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
2896 | ||
2897 | if (prepare) | |
2898 | nvme_dev_shutdown(dev); | |
2899 | else | |
2900 | nvme_dev_resume(dev); | |
2901 | } | |
2902 | ||
09ece142 KB |
2903 | static void nvme_shutdown(struct pci_dev *pdev) |
2904 | { | |
2905 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
2906 | nvme_dev_shutdown(dev); | |
2907 | } | |
2908 | ||
8d85fce7 | 2909 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
2910 | { |
2911 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 KB |
2912 | |
2913 | spin_lock(&dev_list_lock); | |
2914 | list_del_init(&dev->node); | |
2915 | spin_unlock(&dev_list_lock); | |
2916 | ||
2917 | pci_set_drvdata(pdev, NULL); | |
2918 | flush_work(&dev->reset_work); | |
f3db22fe | 2919 | flush_work(&dev->cpu_work); |
5e82e952 | 2920 | misc_deregister(&dev->miscdev); |
9a6b9458 KB |
2921 | nvme_dev_remove(dev); |
2922 | nvme_dev_shutdown(dev); | |
a1a5ef99 | 2923 | nvme_free_queues(dev, 0); |
5a92e700 | 2924 | rcu_barrier(); |
9a6b9458 KB |
2925 | nvme_release_instance(dev); |
2926 | nvme_release_prp_pools(dev); | |
5e82e952 | 2927 | kref_put(&dev->kref, nvme_free_dev); |
b60503ba MW |
2928 | } |
2929 | ||
2930 | /* These functions are yet to be implemented */ | |
2931 | #define nvme_error_detected NULL | |
2932 | #define nvme_dump_registers NULL | |
2933 | #define nvme_link_reset NULL | |
2934 | #define nvme_slot_reset NULL | |
2935 | #define nvme_error_resume NULL | |
cd638946 | 2936 | |
671a6018 | 2937 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
2938 | static int nvme_suspend(struct device *dev) |
2939 | { | |
2940 | struct pci_dev *pdev = to_pci_dev(dev); | |
2941 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
2942 | ||
2943 | nvme_dev_shutdown(ndev); | |
2944 | return 0; | |
2945 | } | |
2946 | ||
2947 | static int nvme_resume(struct device *dev) | |
2948 | { | |
2949 | struct pci_dev *pdev = to_pci_dev(dev); | |
2950 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 2951 | |
9a6b9458 | 2952 | if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) { |
9ca97374 | 2953 | ndev->reset_workfn = nvme_reset_failed_dev; |
9a6b9458 KB |
2954 | queue_work(nvme_workq, &ndev->reset_work); |
2955 | } | |
2956 | return 0; | |
cd638946 | 2957 | } |
671a6018 | 2958 | #endif |
cd638946 KB |
2959 | |
2960 | static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); | |
b60503ba | 2961 | |
1d352035 | 2962 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba MW |
2963 | .error_detected = nvme_error_detected, |
2964 | .mmio_enabled = nvme_dump_registers, | |
2965 | .link_reset = nvme_link_reset, | |
2966 | .slot_reset = nvme_slot_reset, | |
2967 | .resume = nvme_error_resume, | |
f0d54a54 | 2968 | .reset_notify = nvme_reset_notify, |
b60503ba MW |
2969 | }; |
2970 | ||
2971 | /* Move to pci_ids.h later */ | |
2972 | #define PCI_CLASS_STORAGE_EXPRESS 0x010802 | |
2973 | ||
6eb0d698 | 2974 | static const struct pci_device_id nvme_id_table[] = { |
b60503ba MW |
2975 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
2976 | { 0, } | |
2977 | }; | |
2978 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
2979 | ||
2980 | static struct pci_driver nvme_driver = { | |
2981 | .name = "nvme", | |
2982 | .id_table = nvme_id_table, | |
2983 | .probe = nvme_probe, | |
8d85fce7 | 2984 | .remove = nvme_remove, |
09ece142 | 2985 | .shutdown = nvme_shutdown, |
cd638946 KB |
2986 | .driver = { |
2987 | .pm = &nvme_dev_pm_ops, | |
2988 | }, | |
b60503ba MW |
2989 | .err_handler = &nvme_err_handler, |
2990 | }; | |
2991 | ||
2992 | static int __init nvme_init(void) | |
2993 | { | |
0ac13140 | 2994 | int result; |
1fa6aead | 2995 | |
b9afca3e | 2996 | init_waitqueue_head(&nvme_kthread_wait); |
b60503ba | 2997 | |
9a6b9458 KB |
2998 | nvme_workq = create_singlethread_workqueue("nvme"); |
2999 | if (!nvme_workq) | |
b9afca3e | 3000 | return -ENOMEM; |
9a6b9458 | 3001 | |
5c42ea16 KB |
3002 | result = register_blkdev(nvme_major, "nvme"); |
3003 | if (result < 0) | |
9a6b9458 | 3004 | goto kill_workq; |
5c42ea16 | 3005 | else if (result > 0) |
0ac13140 | 3006 | nvme_major = result; |
b60503ba | 3007 | |
f3db22fe KB |
3008 | nvme_nb.notifier_call = &nvme_cpu_notify; |
3009 | result = register_hotcpu_notifier(&nvme_nb); | |
1fa6aead MW |
3010 | if (result) |
3011 | goto unregister_blkdev; | |
f3db22fe KB |
3012 | |
3013 | result = pci_register_driver(&nvme_driver); | |
3014 | if (result) | |
3015 | goto unregister_hotcpu; | |
1fa6aead | 3016 | return 0; |
b60503ba | 3017 | |
f3db22fe KB |
3018 | unregister_hotcpu: |
3019 | unregister_hotcpu_notifier(&nvme_nb); | |
1fa6aead | 3020 | unregister_blkdev: |
b60503ba | 3021 | unregister_blkdev(nvme_major, "nvme"); |
9a6b9458 KB |
3022 | kill_workq: |
3023 | destroy_workqueue(nvme_workq); | |
b60503ba MW |
3024 | return result; |
3025 | } | |
3026 | ||
3027 | static void __exit nvme_exit(void) | |
3028 | { | |
3029 | pci_unregister_driver(&nvme_driver); | |
f3db22fe | 3030 | unregister_hotcpu_notifier(&nvme_nb); |
b60503ba | 3031 | unregister_blkdev(nvme_major, "nvme"); |
9a6b9458 | 3032 | destroy_workqueue(nvme_workq); |
b9afca3e | 3033 | BUG_ON(nvme_thread && !IS_ERR(nvme_thread)); |
21bd78bc | 3034 | _nvme_check_size(); |
b60503ba MW |
3035 | } |
3036 | ||
3037 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
3038 | MODULE_LICENSE("GPL"); | |
6eb0d698 | 3039 | MODULE_VERSION("0.9"); |
b60503ba MW |
3040 | module_init(nvme_init); |
3041 | module_exit(nvme_exit); |