]>
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 | ||
a0a3408e | 15 | #include <linux/aer.h> |
8de05535 | 16 | #include <linux/bitops.h> |
b60503ba | 17 | #include <linux/blkdev.h> |
a4aea562 | 18 | #include <linux/blk-mq.h> |
dca51e78 | 19 | #include <linux/blk-mq-pci.h> |
42f61420 | 20 | #include <linux/cpu.h> |
fd63e9ce | 21 | #include <linux/delay.h> |
b60503ba MW |
22 | #include <linux/errno.h> |
23 | #include <linux/fs.h> | |
24 | #include <linux/genhd.h> | |
4cc09e2d | 25 | #include <linux/hdreg.h> |
5aff9382 | 26 | #include <linux/idr.h> |
b60503ba MW |
27 | #include <linux/init.h> |
28 | #include <linux/interrupt.h> | |
29 | #include <linux/io.h> | |
30 | #include <linux/kdev_t.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/mm.h> | |
33 | #include <linux/module.h> | |
34 | #include <linux/moduleparam.h> | |
77bf25ea | 35 | #include <linux/mutex.h> |
b60503ba | 36 | #include <linux/pci.h> |
be7b6275 | 37 | #include <linux/poison.h> |
c3bfe717 | 38 | #include <linux/ptrace.h> |
b60503ba MW |
39 | #include <linux/sched.h> |
40 | #include <linux/slab.h> | |
e1e5e564 | 41 | #include <linux/t10-pi.h> |
2d55cd5f | 42 | #include <linux/timer.h> |
b60503ba | 43 | #include <linux/types.h> |
2f8e2c87 | 44 | #include <linux/io-64-nonatomic-lo-hi.h> |
1d277a63 | 45 | #include <asm/unaligned.h> |
a98e58e5 | 46 | #include <linux/sed-opal.h> |
797a796a | 47 | |
f11bb3e2 CH |
48 | #include "nvme.h" |
49 | ||
9d43cf64 | 50 | #define NVME_Q_DEPTH 1024 |
d31af0a3 | 51 | #define NVME_AQ_DEPTH 256 |
b60503ba MW |
52 | #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) |
53 | #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) | |
c965809c | 54 | |
adf68f21 CH |
55 | /* |
56 | * We handle AEN commands ourselves and don't even let the | |
57 | * block layer know about them. | |
58 | */ | |
f866fc42 | 59 | #define NVME_AQ_BLKMQ_DEPTH (NVME_AQ_DEPTH - NVME_NR_AERS) |
9d43cf64 | 60 | |
58ffacb5 MW |
61 | static int use_threaded_interrupts; |
62 | module_param(use_threaded_interrupts, int, 0); | |
63 | ||
8ffaadf7 JD |
64 | static bool use_cmb_sqes = true; |
65 | module_param(use_cmb_sqes, bool, 0644); | |
66 | MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes"); | |
67 | ||
9a6b9458 | 68 | static struct workqueue_struct *nvme_workq; |
1fa6aead | 69 | |
1c63dc66 CH |
70 | struct nvme_dev; |
71 | struct nvme_queue; | |
b3fffdef | 72 | |
4cc06521 | 73 | static int nvme_reset(struct nvme_dev *dev); |
a0fa9647 | 74 | static void nvme_process_cq(struct nvme_queue *nvmeq); |
a5cdb68c | 75 | static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown); |
d4b4ff8e | 76 | |
1c63dc66 CH |
77 | /* |
78 | * Represents an NVM Express device. Each nvme_dev is a PCI function. | |
79 | */ | |
80 | struct nvme_dev { | |
1c63dc66 CH |
81 | struct nvme_queue **queues; |
82 | struct blk_mq_tag_set tagset; | |
83 | struct blk_mq_tag_set admin_tagset; | |
84 | u32 __iomem *dbs; | |
85 | struct device *dev; | |
86 | struct dma_pool *prp_page_pool; | |
87 | struct dma_pool *prp_small_pool; | |
88 | unsigned queue_count; | |
89 | unsigned online_queues; | |
90 | unsigned max_qid; | |
91 | int q_depth; | |
92 | u32 db_stride; | |
1c63dc66 | 93 | void __iomem *bar; |
1c63dc66 | 94 | struct work_struct reset_work; |
5c8809e6 | 95 | struct work_struct remove_work; |
2d55cd5f | 96 | struct timer_list watchdog_timer; |
77bf25ea | 97 | struct mutex shutdown_lock; |
1c63dc66 | 98 | bool subsystem; |
1c63dc66 CH |
99 | void __iomem *cmb; |
100 | dma_addr_t cmb_dma_addr; | |
101 | u64 cmb_size; | |
102 | u32 cmbsz; | |
202021c1 | 103 | u32 cmbloc; |
1c63dc66 | 104 | struct nvme_ctrl ctrl; |
db3cbfff | 105 | struct completion ioq_wait; |
4d115420 | 106 | }; |
1fa6aead | 107 | |
1c63dc66 CH |
108 | static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl) |
109 | { | |
110 | return container_of(ctrl, struct nvme_dev, ctrl); | |
111 | } | |
112 | ||
b60503ba MW |
113 | /* |
114 | * An NVM Express queue. Each device has at least two (one for admin | |
115 | * commands and one for I/O commands). | |
116 | */ | |
117 | struct nvme_queue { | |
118 | struct device *q_dmadev; | |
091b6092 | 119 | struct nvme_dev *dev; |
3193f07b | 120 | char irqname[24]; /* nvme4294967295-65535\0 */ |
b60503ba MW |
121 | spinlock_t q_lock; |
122 | struct nvme_command *sq_cmds; | |
8ffaadf7 | 123 | struct nvme_command __iomem *sq_cmds_io; |
b60503ba | 124 | volatile struct nvme_completion *cqes; |
42483228 | 125 | struct blk_mq_tags **tags; |
b60503ba MW |
126 | dma_addr_t sq_dma_addr; |
127 | dma_addr_t cq_dma_addr; | |
b60503ba MW |
128 | u32 __iomem *q_db; |
129 | u16 q_depth; | |
6222d172 | 130 | s16 cq_vector; |
b60503ba MW |
131 | u16 sq_tail; |
132 | u16 cq_head; | |
c30341dc | 133 | u16 qid; |
e9539f47 MW |
134 | u8 cq_phase; |
135 | u8 cqe_seen; | |
b60503ba MW |
136 | }; |
137 | ||
71bd150c CH |
138 | /* |
139 | * The nvme_iod describes the data in an I/O, including the list of PRP | |
140 | * entries. You can't see it in this data structure because C doesn't let | |
f4800d6d | 141 | * me express that. Use nvme_init_iod to ensure there's enough space |
71bd150c CH |
142 | * allocated to store the PRP list. |
143 | */ | |
144 | struct nvme_iod { | |
d49187e9 | 145 | struct nvme_request req; |
f4800d6d CH |
146 | struct nvme_queue *nvmeq; |
147 | int aborted; | |
71bd150c | 148 | int npages; /* In the PRP list. 0 means small pool in use */ |
71bd150c CH |
149 | int nents; /* Used in scatterlist */ |
150 | int length; /* Of data, in bytes */ | |
151 | dma_addr_t first_dma; | |
bf684057 | 152 | struct scatterlist meta_sg; /* metadata requires single contiguous buffer */ |
f4800d6d CH |
153 | struct scatterlist *sg; |
154 | struct scatterlist inline_sg[0]; | |
b60503ba MW |
155 | }; |
156 | ||
157 | /* | |
158 | * Check we didin't inadvertently grow the command struct | |
159 | */ | |
160 | static inline void _nvme_check_size(void) | |
161 | { | |
162 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
163 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); | |
164 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
165 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
166 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
f8ebf840 | 167 | BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); |
c30341dc | 168 | BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); |
b60503ba MW |
169 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); |
170 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); | |
171 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); | |
172 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
6ecec745 | 173 | BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); |
b60503ba MW |
174 | } |
175 | ||
ac3dd5bd JA |
176 | /* |
177 | * Max size of iod being embedded in the request payload | |
178 | */ | |
179 | #define NVME_INT_PAGES 2 | |
5fd4ce1b | 180 | #define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->ctrl.page_size) |
ac3dd5bd JA |
181 | |
182 | /* | |
183 | * Will slightly overestimate the number of pages needed. This is OK | |
184 | * as it only leads to a small amount of wasted memory for the lifetime of | |
185 | * the I/O. | |
186 | */ | |
187 | static int nvme_npages(unsigned size, struct nvme_dev *dev) | |
188 | { | |
5fd4ce1b CH |
189 | unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size, |
190 | dev->ctrl.page_size); | |
ac3dd5bd JA |
191 | return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); |
192 | } | |
193 | ||
f4800d6d CH |
194 | static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev, |
195 | unsigned int size, unsigned int nseg) | |
ac3dd5bd | 196 | { |
f4800d6d CH |
197 | return sizeof(__le64 *) * nvme_npages(size, dev) + |
198 | sizeof(struct scatterlist) * nseg; | |
199 | } | |
ac3dd5bd | 200 | |
f4800d6d CH |
201 | static unsigned int nvme_cmd_size(struct nvme_dev *dev) |
202 | { | |
203 | return sizeof(struct nvme_iod) + | |
204 | nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES); | |
ac3dd5bd JA |
205 | } |
206 | ||
dca51e78 CH |
207 | static int nvmeq_irq(struct nvme_queue *nvmeq) |
208 | { | |
209 | return pci_irq_vector(to_pci_dev(nvmeq->dev->dev), nvmeq->cq_vector); | |
210 | } | |
211 | ||
a4aea562 MB |
212 | static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
213 | unsigned int hctx_idx) | |
e85248e5 | 214 | { |
a4aea562 MB |
215 | struct nvme_dev *dev = data; |
216 | struct nvme_queue *nvmeq = dev->queues[0]; | |
217 | ||
42483228 KB |
218 | WARN_ON(hctx_idx != 0); |
219 | WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); | |
220 | WARN_ON(nvmeq->tags); | |
221 | ||
a4aea562 | 222 | hctx->driver_data = nvmeq; |
42483228 | 223 | nvmeq->tags = &dev->admin_tagset.tags[0]; |
a4aea562 | 224 | return 0; |
e85248e5 MW |
225 | } |
226 | ||
4af0e21c KB |
227 | static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) |
228 | { | |
229 | struct nvme_queue *nvmeq = hctx->driver_data; | |
230 | ||
231 | nvmeq->tags = NULL; | |
232 | } | |
233 | ||
a4aea562 MB |
234 | static int nvme_admin_init_request(void *data, struct request *req, |
235 | unsigned int hctx_idx, unsigned int rq_idx, | |
236 | unsigned int numa_node) | |
22404274 | 237 | { |
a4aea562 | 238 | struct nvme_dev *dev = data; |
f4800d6d | 239 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
a4aea562 MB |
240 | struct nvme_queue *nvmeq = dev->queues[0]; |
241 | ||
242 | BUG_ON(!nvmeq); | |
f4800d6d | 243 | iod->nvmeq = nvmeq; |
a4aea562 | 244 | return 0; |
22404274 KB |
245 | } |
246 | ||
a4aea562 MB |
247 | static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
248 | unsigned int hctx_idx) | |
b60503ba | 249 | { |
a4aea562 | 250 | struct nvme_dev *dev = data; |
42483228 | 251 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; |
a4aea562 | 252 | |
42483228 KB |
253 | if (!nvmeq->tags) |
254 | nvmeq->tags = &dev->tagset.tags[hctx_idx]; | |
b60503ba | 255 | |
42483228 | 256 | WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); |
a4aea562 MB |
257 | hctx->driver_data = nvmeq; |
258 | return 0; | |
b60503ba MW |
259 | } |
260 | ||
a4aea562 MB |
261 | static int nvme_init_request(void *data, struct request *req, |
262 | unsigned int hctx_idx, unsigned int rq_idx, | |
263 | unsigned int numa_node) | |
b60503ba | 264 | { |
a4aea562 | 265 | struct nvme_dev *dev = data; |
f4800d6d | 266 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
a4aea562 MB |
267 | struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; |
268 | ||
269 | BUG_ON(!nvmeq); | |
f4800d6d | 270 | iod->nvmeq = nvmeq; |
a4aea562 MB |
271 | return 0; |
272 | } | |
273 | ||
dca51e78 CH |
274 | static int nvme_pci_map_queues(struct blk_mq_tag_set *set) |
275 | { | |
276 | struct nvme_dev *dev = set->driver_data; | |
277 | ||
278 | return blk_mq_pci_map_queues(set, to_pci_dev(dev->dev)); | |
279 | } | |
280 | ||
b60503ba | 281 | /** |
adf68f21 | 282 | * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
283 | * @nvmeq: The queue to use |
284 | * @cmd: The command to send | |
285 | * | |
286 | * Safe to use from interrupt context | |
287 | */ | |
e3f879bf SB |
288 | static void __nvme_submit_cmd(struct nvme_queue *nvmeq, |
289 | struct nvme_command *cmd) | |
b60503ba | 290 | { |
a4aea562 MB |
291 | u16 tail = nvmeq->sq_tail; |
292 | ||
8ffaadf7 JD |
293 | if (nvmeq->sq_cmds_io) |
294 | memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd)); | |
295 | else | |
296 | memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); | |
297 | ||
b60503ba MW |
298 | if (++tail == nvmeq->q_depth) |
299 | tail = 0; | |
7547881d | 300 | writel(tail, nvmeq->q_db); |
b60503ba | 301 | nvmeq->sq_tail = tail; |
b60503ba MW |
302 | } |
303 | ||
f4800d6d | 304 | static __le64 **iod_list(struct request *req) |
b60503ba | 305 | { |
f4800d6d | 306 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
f9d03f96 | 307 | return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req)); |
b60503ba MW |
308 | } |
309 | ||
58b45602 ML |
310 | static int nvme_init_iod(struct request *rq, unsigned size, |
311 | struct nvme_dev *dev) | |
ac3dd5bd | 312 | { |
f4800d6d | 313 | struct nvme_iod *iod = blk_mq_rq_to_pdu(rq); |
f9d03f96 | 314 | int nseg = blk_rq_nr_phys_segments(rq); |
ac3dd5bd | 315 | |
f4800d6d CH |
316 | if (nseg > NVME_INT_PAGES || size > NVME_INT_BYTES(dev)) { |
317 | iod->sg = kmalloc(nvme_iod_alloc_size(dev, size, nseg), GFP_ATOMIC); | |
318 | if (!iod->sg) | |
319 | return BLK_MQ_RQ_QUEUE_BUSY; | |
320 | } else { | |
321 | iod->sg = iod->inline_sg; | |
ac3dd5bd JA |
322 | } |
323 | ||
f4800d6d CH |
324 | iod->aborted = 0; |
325 | iod->npages = -1; | |
326 | iod->nents = 0; | |
327 | iod->length = size; | |
f80ec966 | 328 | |
e8064021 | 329 | if (!(rq->rq_flags & RQF_DONTPREP)) { |
f80ec966 | 330 | rq->retries = 0; |
e8064021 | 331 | rq->rq_flags |= RQF_DONTPREP; |
f80ec966 | 332 | } |
bac0000a | 333 | return BLK_MQ_RQ_QUEUE_OK; |
ac3dd5bd JA |
334 | } |
335 | ||
f4800d6d | 336 | static void nvme_free_iod(struct nvme_dev *dev, struct request *req) |
b60503ba | 337 | { |
f4800d6d | 338 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
5fd4ce1b | 339 | const int last_prp = dev->ctrl.page_size / 8 - 1; |
eca18b23 | 340 | int i; |
f4800d6d | 341 | __le64 **list = iod_list(req); |
eca18b23 MW |
342 | dma_addr_t prp_dma = iod->first_dma; |
343 | ||
344 | if (iod->npages == 0) | |
345 | dma_pool_free(dev->prp_small_pool, list[0], prp_dma); | |
346 | for (i = 0; i < iod->npages; i++) { | |
347 | __le64 *prp_list = list[i]; | |
348 | dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); | |
349 | dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); | |
350 | prp_dma = next_prp_dma; | |
351 | } | |
ac3dd5bd | 352 | |
f4800d6d CH |
353 | if (iod->sg != iod->inline_sg) |
354 | kfree(iod->sg); | |
b4ff9c8d KB |
355 | } |
356 | ||
52b68d7e | 357 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
e1e5e564 KB |
358 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) |
359 | { | |
360 | if (be32_to_cpu(pi->ref_tag) == v) | |
361 | pi->ref_tag = cpu_to_be32(p); | |
362 | } | |
363 | ||
364 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
365 | { | |
366 | if (be32_to_cpu(pi->ref_tag) == p) | |
367 | pi->ref_tag = cpu_to_be32(v); | |
368 | } | |
369 | ||
370 | /** | |
371 | * nvme_dif_remap - remaps ref tags to bip seed and physical lba | |
372 | * | |
373 | * The virtual start sector is the one that was originally submitted by the | |
374 | * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical | |
375 | * start sector may be different. Remap protection information to match the | |
376 | * physical LBA on writes, and back to the original seed on reads. | |
377 | * | |
378 | * Type 0 and 3 do not have a ref tag, so no remapping required. | |
379 | */ | |
380 | static void nvme_dif_remap(struct request *req, | |
381 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
382 | { | |
383 | struct nvme_ns *ns = req->rq_disk->private_data; | |
384 | struct bio_integrity_payload *bip; | |
385 | struct t10_pi_tuple *pi; | |
386 | void *p, *pmap; | |
387 | u32 i, nlb, ts, phys, virt; | |
388 | ||
389 | if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3) | |
390 | return; | |
391 | ||
392 | bip = bio_integrity(req->bio); | |
393 | if (!bip) | |
394 | return; | |
395 | ||
396 | pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset; | |
e1e5e564 KB |
397 | |
398 | p = pmap; | |
399 | virt = bip_get_seed(bip); | |
400 | phys = nvme_block_nr(ns, blk_rq_pos(req)); | |
401 | nlb = (blk_rq_bytes(req) >> ns->lba_shift); | |
ac6fc48c | 402 | ts = ns->disk->queue->integrity.tuple_size; |
e1e5e564 KB |
403 | |
404 | for (i = 0; i < nlb; i++, virt++, phys++) { | |
405 | pi = (struct t10_pi_tuple *)p; | |
406 | dif_swap(phys, virt, pi); | |
407 | p += ts; | |
408 | } | |
409 | kunmap_atomic(pmap); | |
410 | } | |
52b68d7e KB |
411 | #else /* CONFIG_BLK_DEV_INTEGRITY */ |
412 | static void nvme_dif_remap(struct request *req, | |
413 | void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) | |
414 | { | |
415 | } | |
416 | static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) | |
417 | { | |
418 | } | |
419 | static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) | |
420 | { | |
421 | } | |
52b68d7e KB |
422 | #endif |
423 | ||
f4800d6d | 424 | static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req, |
69d2b571 | 425 | int total_len) |
ff22b54f | 426 | { |
f4800d6d | 427 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
99802a7a | 428 | struct dma_pool *pool; |
eca18b23 MW |
429 | int length = total_len; |
430 | struct scatterlist *sg = iod->sg; | |
ff22b54f MW |
431 | int dma_len = sg_dma_len(sg); |
432 | u64 dma_addr = sg_dma_address(sg); | |
5fd4ce1b | 433 | u32 page_size = dev->ctrl.page_size; |
f137e0f1 | 434 | int offset = dma_addr & (page_size - 1); |
e025344c | 435 | __le64 *prp_list; |
f4800d6d | 436 | __le64 **list = iod_list(req); |
e025344c | 437 | dma_addr_t prp_dma; |
eca18b23 | 438 | int nprps, i; |
ff22b54f | 439 | |
1d090624 | 440 | length -= (page_size - offset); |
ff22b54f | 441 | if (length <= 0) |
69d2b571 | 442 | return true; |
ff22b54f | 443 | |
1d090624 | 444 | dma_len -= (page_size - offset); |
ff22b54f | 445 | if (dma_len) { |
1d090624 | 446 | dma_addr += (page_size - offset); |
ff22b54f MW |
447 | } else { |
448 | sg = sg_next(sg); | |
449 | dma_addr = sg_dma_address(sg); | |
450 | dma_len = sg_dma_len(sg); | |
451 | } | |
452 | ||
1d090624 | 453 | if (length <= page_size) { |
edd10d33 | 454 | iod->first_dma = dma_addr; |
69d2b571 | 455 | return true; |
e025344c SMM |
456 | } |
457 | ||
1d090624 | 458 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
459 | if (nprps <= (256 / 8)) { |
460 | pool = dev->prp_small_pool; | |
eca18b23 | 461 | iod->npages = 0; |
99802a7a MW |
462 | } else { |
463 | pool = dev->prp_page_pool; | |
eca18b23 | 464 | iod->npages = 1; |
99802a7a MW |
465 | } |
466 | ||
69d2b571 | 467 | prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); |
b77954cb | 468 | if (!prp_list) { |
edd10d33 | 469 | iod->first_dma = dma_addr; |
eca18b23 | 470 | iod->npages = -1; |
69d2b571 | 471 | return false; |
b77954cb | 472 | } |
eca18b23 MW |
473 | list[0] = prp_list; |
474 | iod->first_dma = prp_dma; | |
e025344c SMM |
475 | i = 0; |
476 | for (;;) { | |
1d090624 | 477 | if (i == page_size >> 3) { |
e025344c | 478 | __le64 *old_prp_list = prp_list; |
69d2b571 | 479 | prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); |
eca18b23 | 480 | if (!prp_list) |
69d2b571 | 481 | return false; |
eca18b23 | 482 | list[iod->npages++] = prp_list; |
7523d834 MW |
483 | prp_list[0] = old_prp_list[i - 1]; |
484 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
485 | i = 1; | |
e025344c SMM |
486 | } |
487 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
488 | dma_len -= page_size; |
489 | dma_addr += page_size; | |
490 | length -= page_size; | |
e025344c SMM |
491 | if (length <= 0) |
492 | break; | |
493 | if (dma_len > 0) | |
494 | continue; | |
495 | BUG_ON(dma_len < 0); | |
496 | sg = sg_next(sg); | |
497 | dma_addr = sg_dma_address(sg); | |
498 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
499 | } |
500 | ||
69d2b571 | 501 | return true; |
ff22b54f MW |
502 | } |
503 | ||
f4800d6d | 504 | static int nvme_map_data(struct nvme_dev *dev, struct request *req, |
03b5929e | 505 | unsigned size, struct nvme_command *cmnd) |
d29ec824 | 506 | { |
f4800d6d | 507 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
ba1ca37e CH |
508 | struct request_queue *q = req->q; |
509 | enum dma_data_direction dma_dir = rq_data_dir(req) ? | |
510 | DMA_TO_DEVICE : DMA_FROM_DEVICE; | |
511 | int ret = BLK_MQ_RQ_QUEUE_ERROR; | |
d29ec824 | 512 | |
f9d03f96 | 513 | sg_init_table(iod->sg, blk_rq_nr_phys_segments(req)); |
ba1ca37e CH |
514 | iod->nents = blk_rq_map_sg(q, req, iod->sg); |
515 | if (!iod->nents) | |
516 | goto out; | |
d29ec824 | 517 | |
ba1ca37e | 518 | ret = BLK_MQ_RQ_QUEUE_BUSY; |
2b6b535d MFO |
519 | if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir, |
520 | DMA_ATTR_NO_WARN)) | |
ba1ca37e | 521 | goto out; |
d29ec824 | 522 | |
03b5929e | 523 | if (!nvme_setup_prps(dev, req, size)) |
ba1ca37e | 524 | goto out_unmap; |
0e5e4f0e | 525 | |
ba1ca37e CH |
526 | ret = BLK_MQ_RQ_QUEUE_ERROR; |
527 | if (blk_integrity_rq(req)) { | |
528 | if (blk_rq_count_integrity_sg(q, req->bio) != 1) | |
529 | goto out_unmap; | |
0e5e4f0e | 530 | |
bf684057 CH |
531 | sg_init_table(&iod->meta_sg, 1); |
532 | if (blk_rq_map_integrity_sg(q, req->bio, &iod->meta_sg) != 1) | |
ba1ca37e | 533 | goto out_unmap; |
0e5e4f0e | 534 | |
ba1ca37e CH |
535 | if (rq_data_dir(req)) |
536 | nvme_dif_remap(req, nvme_dif_prep); | |
0e5e4f0e | 537 | |
bf684057 | 538 | if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir)) |
ba1ca37e | 539 | goto out_unmap; |
d29ec824 | 540 | } |
00df5cb4 | 541 | |
eb793e2c CH |
542 | cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); |
543 | cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma); | |
ba1ca37e | 544 | if (blk_integrity_rq(req)) |
bf684057 | 545 | cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg)); |
ba1ca37e | 546 | return BLK_MQ_RQ_QUEUE_OK; |
00df5cb4 | 547 | |
ba1ca37e CH |
548 | out_unmap: |
549 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); | |
550 | out: | |
551 | return ret; | |
00df5cb4 MW |
552 | } |
553 | ||
f4800d6d | 554 | static void nvme_unmap_data(struct nvme_dev *dev, struct request *req) |
b60503ba | 555 | { |
f4800d6d | 556 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
d4f6c3ab CH |
557 | enum dma_data_direction dma_dir = rq_data_dir(req) ? |
558 | DMA_TO_DEVICE : DMA_FROM_DEVICE; | |
559 | ||
560 | if (iod->nents) { | |
561 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); | |
562 | if (blk_integrity_rq(req)) { | |
563 | if (!rq_data_dir(req)) | |
564 | nvme_dif_remap(req, nvme_dif_complete); | |
bf684057 | 565 | dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir); |
e1e5e564 | 566 | } |
e19b127f | 567 | } |
e1e5e564 | 568 | |
f9d03f96 | 569 | nvme_cleanup_cmd(req); |
f4800d6d | 570 | nvme_free_iod(dev, req); |
d4f6c3ab | 571 | } |
b60503ba | 572 | |
d29ec824 CH |
573 | /* |
574 | * NOTE: ns is NULL when called on the admin queue. | |
575 | */ | |
a4aea562 MB |
576 | static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx, |
577 | const struct blk_mq_queue_data *bd) | |
edd10d33 | 578 | { |
a4aea562 MB |
579 | struct nvme_ns *ns = hctx->queue->queuedata; |
580 | struct nvme_queue *nvmeq = hctx->driver_data; | |
d29ec824 | 581 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 | 582 | struct request *req = bd->rq; |
ba1ca37e | 583 | struct nvme_command cmnd; |
58b45602 | 584 | unsigned map_len; |
ba1ca37e | 585 | int ret = BLK_MQ_RQ_QUEUE_OK; |
edd10d33 | 586 | |
e1e5e564 KB |
587 | /* |
588 | * If formated with metadata, require the block layer provide a buffer | |
589 | * unless this namespace is formated such that the metadata can be | |
590 | * stripped/generated by the controller with PRACT=1. | |
591 | */ | |
d29ec824 | 592 | if (ns && ns->ms && !blk_integrity_rq(req)) { |
71feb364 KB |
593 | if (!(ns->pi_type && ns->ms == 8) && |
594 | req->cmd_type != REQ_TYPE_DRV_PRIV) { | |
eee417b0 | 595 | blk_mq_end_request(req, -EFAULT); |
e1e5e564 KB |
596 | return BLK_MQ_RQ_QUEUE_OK; |
597 | } | |
598 | } | |
599 | ||
f9d03f96 | 600 | ret = nvme_setup_cmd(ns, req, &cmnd); |
bac0000a | 601 | if (ret != BLK_MQ_RQ_QUEUE_OK) |
f4800d6d | 602 | return ret; |
a4aea562 | 603 | |
f9d03f96 CH |
604 | map_len = nvme_map_len(req); |
605 | ret = nvme_init_iod(req, map_len, dev); | |
bac0000a | 606 | if (ret != BLK_MQ_RQ_QUEUE_OK) |
f9d03f96 | 607 | goto out_free_cmd; |
a4aea562 | 608 | |
f9d03f96 | 609 | if (blk_rq_nr_phys_segments(req)) |
03b5929e | 610 | ret = nvme_map_data(dev, req, map_len, &cmnd); |
a4aea562 | 611 | |
bac0000a | 612 | if (ret != BLK_MQ_RQ_QUEUE_OK) |
f9d03f96 | 613 | goto out_cleanup_iod; |
a4aea562 | 614 | |
aae239e1 | 615 | blk_mq_start_request(req); |
a4aea562 | 616 | |
ba1ca37e | 617 | spin_lock_irq(&nvmeq->q_lock); |
ae1fba20 | 618 | if (unlikely(nvmeq->cq_vector < 0)) { |
69d9a99c KB |
619 | if (ns && !test_bit(NVME_NS_DEAD, &ns->flags)) |
620 | ret = BLK_MQ_RQ_QUEUE_BUSY; | |
621 | else | |
622 | ret = BLK_MQ_RQ_QUEUE_ERROR; | |
ae1fba20 | 623 | spin_unlock_irq(&nvmeq->q_lock); |
f9d03f96 | 624 | goto out_cleanup_iod; |
ae1fba20 | 625 | } |
ba1ca37e | 626 | __nvme_submit_cmd(nvmeq, &cmnd); |
a4aea562 MB |
627 | nvme_process_cq(nvmeq); |
628 | spin_unlock_irq(&nvmeq->q_lock); | |
629 | return BLK_MQ_RQ_QUEUE_OK; | |
f9d03f96 | 630 | out_cleanup_iod: |
f4800d6d | 631 | nvme_free_iod(dev, req); |
f9d03f96 CH |
632 | out_free_cmd: |
633 | nvme_cleanup_cmd(req); | |
ba1ca37e | 634 | return ret; |
b60503ba | 635 | } |
e1e5e564 | 636 | |
eee417b0 CH |
637 | static void nvme_complete_rq(struct request *req) |
638 | { | |
f4800d6d CH |
639 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
640 | struct nvme_dev *dev = iod->nvmeq->dev; | |
eee417b0 | 641 | int error = 0; |
e1e5e564 | 642 | |
f4800d6d | 643 | nvme_unmap_data(dev, req); |
e1e5e564 | 644 | |
eee417b0 CH |
645 | if (unlikely(req->errors)) { |
646 | if (nvme_req_needs_retry(req, req->errors)) { | |
f80ec966 | 647 | req->retries++; |
eee417b0 CH |
648 | nvme_requeue_req(req); |
649 | return; | |
e1e5e564 | 650 | } |
1974b1ae | 651 | |
eee417b0 CH |
652 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) |
653 | error = req->errors; | |
654 | else | |
655 | error = nvme_error_status(req->errors); | |
656 | } | |
a4aea562 | 657 | |
f4800d6d | 658 | if (unlikely(iod->aborted)) { |
1b3c47c1 | 659 | dev_warn(dev->ctrl.device, |
eee417b0 CH |
660 | "completing aborted command with status: %04x\n", |
661 | req->errors); | |
662 | } | |
a4aea562 | 663 | |
eee417b0 | 664 | blk_mq_end_request(req, error); |
b60503ba MW |
665 | } |
666 | ||
d783e0bd MR |
667 | /* We read the CQE phase first to check if the rest of the entry is valid */ |
668 | static inline bool nvme_cqe_valid(struct nvme_queue *nvmeq, u16 head, | |
669 | u16 phase) | |
670 | { | |
671 | return (le16_to_cpu(nvmeq->cqes[head].status) & 1) == phase; | |
672 | } | |
673 | ||
a0fa9647 | 674 | static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag) |
b60503ba | 675 | { |
82123460 | 676 | u16 head, phase; |
b60503ba | 677 | |
b60503ba | 678 | head = nvmeq->cq_head; |
82123460 | 679 | phase = nvmeq->cq_phase; |
b60503ba | 680 | |
d783e0bd | 681 | while (nvme_cqe_valid(nvmeq, head, phase)) { |
b60503ba | 682 | struct nvme_completion cqe = nvmeq->cqes[head]; |
eee417b0 | 683 | struct request *req; |
adf68f21 | 684 | |
b60503ba MW |
685 | if (++head == nvmeq->q_depth) { |
686 | head = 0; | |
82123460 | 687 | phase = !phase; |
b60503ba | 688 | } |
adf68f21 | 689 | |
a0fa9647 JA |
690 | if (tag && *tag == cqe.command_id) |
691 | *tag = -1; | |
adf68f21 | 692 | |
aae239e1 | 693 | if (unlikely(cqe.command_id >= nvmeq->q_depth)) { |
1b3c47c1 | 694 | dev_warn(nvmeq->dev->ctrl.device, |
aae239e1 CH |
695 | "invalid id %d completed on queue %d\n", |
696 | cqe.command_id, le16_to_cpu(cqe.sq_id)); | |
697 | continue; | |
698 | } | |
699 | ||
adf68f21 CH |
700 | /* |
701 | * AEN requests are special as they don't time out and can | |
702 | * survive any kind of queue freeze and often don't respond to | |
703 | * aborts. We don't even bother to allocate a struct request | |
704 | * for them but rather special case them here. | |
705 | */ | |
706 | if (unlikely(nvmeq->qid == 0 && | |
707 | cqe.command_id >= NVME_AQ_BLKMQ_DEPTH)) { | |
7bf58533 CH |
708 | nvme_complete_async_event(&nvmeq->dev->ctrl, |
709 | cqe.status, &cqe.result); | |
adf68f21 CH |
710 | continue; |
711 | } | |
712 | ||
eee417b0 | 713 | req = blk_mq_tag_to_rq(*nvmeq->tags, cqe.command_id); |
d49187e9 | 714 | nvme_req(req)->result = cqe.result; |
d783e0bd | 715 | blk_mq_complete_request(req, le16_to_cpu(cqe.status) >> 1); |
b60503ba MW |
716 | } |
717 | ||
82123460 | 718 | if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) |
a0fa9647 | 719 | return; |
b60503ba | 720 | |
604e8c8d KB |
721 | if (likely(nvmeq->cq_vector >= 0)) |
722 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); | |
b60503ba | 723 | nvmeq->cq_head = head; |
82123460 | 724 | nvmeq->cq_phase = phase; |
b60503ba | 725 | |
e9539f47 | 726 | nvmeq->cqe_seen = 1; |
a0fa9647 JA |
727 | } |
728 | ||
729 | static void nvme_process_cq(struct nvme_queue *nvmeq) | |
730 | { | |
731 | __nvme_process_cq(nvmeq, NULL); | |
b60503ba MW |
732 | } |
733 | ||
734 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 MW |
735 | { |
736 | irqreturn_t result; | |
737 | struct nvme_queue *nvmeq = data; | |
738 | spin_lock(&nvmeq->q_lock); | |
e9539f47 MW |
739 | nvme_process_cq(nvmeq); |
740 | result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; | |
741 | nvmeq->cqe_seen = 0; | |
58ffacb5 MW |
742 | spin_unlock(&nvmeq->q_lock); |
743 | return result; | |
744 | } | |
745 | ||
746 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
747 | { | |
748 | struct nvme_queue *nvmeq = data; | |
d783e0bd MR |
749 | if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) |
750 | return IRQ_WAKE_THREAD; | |
751 | return IRQ_NONE; | |
58ffacb5 MW |
752 | } |
753 | ||
a0fa9647 JA |
754 | static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag) |
755 | { | |
756 | struct nvme_queue *nvmeq = hctx->driver_data; | |
757 | ||
d783e0bd | 758 | if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) { |
a0fa9647 JA |
759 | spin_lock_irq(&nvmeq->q_lock); |
760 | __nvme_process_cq(nvmeq, &tag); | |
761 | spin_unlock_irq(&nvmeq->q_lock); | |
762 | ||
763 | if (tag == -1) | |
764 | return 1; | |
765 | } | |
766 | ||
767 | return 0; | |
768 | } | |
769 | ||
f866fc42 | 770 | static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx) |
b60503ba | 771 | { |
f866fc42 | 772 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
9396dec9 | 773 | struct nvme_queue *nvmeq = dev->queues[0]; |
a4aea562 | 774 | struct nvme_command c; |
b60503ba | 775 | |
a4aea562 MB |
776 | memset(&c, 0, sizeof(c)); |
777 | c.common.opcode = nvme_admin_async_event; | |
f866fc42 | 778 | c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx; |
3c0cf138 | 779 | |
9396dec9 | 780 | spin_lock_irq(&nvmeq->q_lock); |
f866fc42 | 781 | __nvme_submit_cmd(nvmeq, &c); |
9396dec9 | 782 | spin_unlock_irq(&nvmeq->q_lock); |
f705f837 CH |
783 | } |
784 | ||
b60503ba | 785 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
f705f837 | 786 | { |
b60503ba MW |
787 | struct nvme_command c; |
788 | ||
789 | memset(&c, 0, sizeof(c)); | |
790 | c.delete_queue.opcode = opcode; | |
791 | c.delete_queue.qid = cpu_to_le16(id); | |
792 | ||
1c63dc66 | 793 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
794 | } |
795 | ||
b60503ba MW |
796 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, |
797 | struct nvme_queue *nvmeq) | |
798 | { | |
b60503ba MW |
799 | struct nvme_command c; |
800 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; | |
801 | ||
d29ec824 CH |
802 | /* |
803 | * Note: we (ab)use the fact the the prp fields survive if no data | |
804 | * is attached to the request. | |
805 | */ | |
b60503ba MW |
806 | memset(&c, 0, sizeof(c)); |
807 | c.create_cq.opcode = nvme_admin_create_cq; | |
808 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
809 | c.create_cq.cqid = cpu_to_le16(qid); | |
810 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
811 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
812 | c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); | |
813 | ||
1c63dc66 | 814 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
815 | } |
816 | ||
817 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
818 | struct nvme_queue *nvmeq) | |
819 | { | |
b60503ba MW |
820 | struct nvme_command c; |
821 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; | |
822 | ||
d29ec824 CH |
823 | /* |
824 | * Note: we (ab)use the fact the the prp fields survive if no data | |
825 | * is attached to the request. | |
826 | */ | |
b60503ba MW |
827 | memset(&c, 0, sizeof(c)); |
828 | c.create_sq.opcode = nvme_admin_create_sq; | |
829 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
830 | c.create_sq.sqid = cpu_to_le16(qid); | |
831 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
832 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
833 | c.create_sq.cqid = cpu_to_le16(qid); | |
834 | ||
1c63dc66 | 835 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
836 | } |
837 | ||
838 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
839 | { | |
840 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
841 | } | |
842 | ||
843 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
844 | { | |
845 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
846 | } | |
847 | ||
e7a2a87d | 848 | static void abort_endio(struct request *req, int error) |
bc5fc7e4 | 849 | { |
f4800d6d CH |
850 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
851 | struct nvme_queue *nvmeq = iod->nvmeq; | |
e7a2a87d | 852 | u16 status = req->errors; |
e44ac588 | 853 | |
1cb3cce5 | 854 | dev_warn(nvmeq->dev->ctrl.device, "Abort status: 0x%x", status); |
e7a2a87d | 855 | atomic_inc(&nvmeq->dev->ctrl.abort_limit); |
e7a2a87d | 856 | blk_mq_free_request(req); |
bc5fc7e4 MW |
857 | } |
858 | ||
31c7c7d2 | 859 | static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) |
c30341dc | 860 | { |
f4800d6d CH |
861 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
862 | struct nvme_queue *nvmeq = iod->nvmeq; | |
c30341dc | 863 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 | 864 | struct request *abort_req; |
a4aea562 | 865 | struct nvme_command cmd; |
c30341dc | 866 | |
31c7c7d2 | 867 | /* |
fd634f41 CH |
868 | * Shutdown immediately if controller times out while starting. The |
869 | * reset work will see the pci device disabled when it gets the forced | |
870 | * cancellation error. All outstanding requests are completed on | |
871 | * shutdown, so we return BLK_EH_HANDLED. | |
872 | */ | |
bb8d261e | 873 | if (dev->ctrl.state == NVME_CTRL_RESETTING) { |
1b3c47c1 | 874 | dev_warn(dev->ctrl.device, |
fd634f41 CH |
875 | "I/O %d QID %d timeout, disable controller\n", |
876 | req->tag, nvmeq->qid); | |
a5cdb68c | 877 | nvme_dev_disable(dev, false); |
fd634f41 CH |
878 | req->errors = NVME_SC_CANCELLED; |
879 | return BLK_EH_HANDLED; | |
c30341dc KB |
880 | } |
881 | ||
fd634f41 CH |
882 | /* |
883 | * Shutdown the controller immediately and schedule a reset if the | |
884 | * command was already aborted once before and still hasn't been | |
885 | * returned to the driver, or if this is the admin queue. | |
31c7c7d2 | 886 | */ |
f4800d6d | 887 | if (!nvmeq->qid || iod->aborted) { |
1b3c47c1 | 888 | dev_warn(dev->ctrl.device, |
e1569a16 KB |
889 | "I/O %d QID %d timeout, reset controller\n", |
890 | req->tag, nvmeq->qid); | |
a5cdb68c | 891 | nvme_dev_disable(dev, false); |
c5f6ce97 | 892 | nvme_reset(dev); |
c30341dc | 893 | |
e1569a16 KB |
894 | /* |
895 | * Mark the request as handled, since the inline shutdown | |
896 | * forces all outstanding requests to complete. | |
897 | */ | |
898 | req->errors = NVME_SC_CANCELLED; | |
899 | return BLK_EH_HANDLED; | |
c30341dc | 900 | } |
c30341dc | 901 | |
e7a2a87d | 902 | if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) { |
6bf25d16 | 903 | atomic_inc(&dev->ctrl.abort_limit); |
31c7c7d2 | 904 | return BLK_EH_RESET_TIMER; |
6bf25d16 | 905 | } |
7bf7d778 | 906 | iod->aborted = 1; |
a4aea562 | 907 | |
c30341dc KB |
908 | memset(&cmd, 0, sizeof(cmd)); |
909 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
a4aea562 | 910 | cmd.abort.cid = req->tag; |
c30341dc | 911 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); |
c30341dc | 912 | |
1b3c47c1 SG |
913 | dev_warn(nvmeq->dev->ctrl.device, |
914 | "I/O %d QID %d timeout, aborting\n", | |
915 | req->tag, nvmeq->qid); | |
e7a2a87d CH |
916 | |
917 | abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd, | |
eb71f435 | 918 | BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); |
e7a2a87d CH |
919 | if (IS_ERR(abort_req)) { |
920 | atomic_inc(&dev->ctrl.abort_limit); | |
921 | return BLK_EH_RESET_TIMER; | |
922 | } | |
923 | ||
924 | abort_req->timeout = ADMIN_TIMEOUT; | |
925 | abort_req->end_io_data = NULL; | |
926 | blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio); | |
c30341dc | 927 | |
31c7c7d2 CH |
928 | /* |
929 | * The aborted req will be completed on receiving the abort req. | |
930 | * We enable the timer again. If hit twice, it'll cause a device reset, | |
931 | * as the device then is in a faulty state. | |
932 | */ | |
933 | return BLK_EH_RESET_TIMER; | |
c30341dc KB |
934 | } |
935 | ||
a4aea562 MB |
936 | static void nvme_free_queue(struct nvme_queue *nvmeq) |
937 | { | |
9e866774 MW |
938 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), |
939 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
8ffaadf7 JD |
940 | if (nvmeq->sq_cmds) |
941 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
9e866774 MW |
942 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); |
943 | kfree(nvmeq); | |
944 | } | |
945 | ||
a1a5ef99 | 946 | static void nvme_free_queues(struct nvme_dev *dev, int lowest) |
22404274 KB |
947 | { |
948 | int i; | |
949 | ||
a1a5ef99 | 950 | for (i = dev->queue_count - 1; i >= lowest; i--) { |
a4aea562 | 951 | struct nvme_queue *nvmeq = dev->queues[i]; |
22404274 | 952 | dev->queue_count--; |
a4aea562 | 953 | dev->queues[i] = NULL; |
f435c282 | 954 | nvme_free_queue(nvmeq); |
121c7ad4 | 955 | } |
22404274 KB |
956 | } |
957 | ||
4d115420 KB |
958 | /** |
959 | * nvme_suspend_queue - put queue into suspended state | |
960 | * @nvmeq - queue to suspend | |
4d115420 KB |
961 | */ |
962 | static int nvme_suspend_queue(struct nvme_queue *nvmeq) | |
b60503ba | 963 | { |
2b25d981 | 964 | int vector; |
b60503ba | 965 | |
a09115b2 | 966 | spin_lock_irq(&nvmeq->q_lock); |
2b25d981 KB |
967 | if (nvmeq->cq_vector == -1) { |
968 | spin_unlock_irq(&nvmeq->q_lock); | |
969 | return 1; | |
970 | } | |
dca51e78 | 971 | vector = nvmeq_irq(nvmeq); |
42f61420 | 972 | nvmeq->dev->online_queues--; |
2b25d981 | 973 | nvmeq->cq_vector = -1; |
a09115b2 MW |
974 | spin_unlock_irq(&nvmeq->q_lock); |
975 | ||
1c63dc66 | 976 | if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q) |
25646264 | 977 | blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q); |
6df3dbc8 | 978 | |
aba2080f | 979 | free_irq(vector, nvmeq); |
b60503ba | 980 | |
4d115420 KB |
981 | return 0; |
982 | } | |
b60503ba | 983 | |
a5cdb68c | 984 | static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown) |
4d115420 | 985 | { |
a5cdb68c | 986 | struct nvme_queue *nvmeq = dev->queues[0]; |
4d115420 KB |
987 | |
988 | if (!nvmeq) | |
989 | return; | |
990 | if (nvme_suspend_queue(nvmeq)) | |
991 | return; | |
992 | ||
a5cdb68c KB |
993 | if (shutdown) |
994 | nvme_shutdown_ctrl(&dev->ctrl); | |
995 | else | |
996 | nvme_disable_ctrl(&dev->ctrl, lo_hi_readq( | |
997 | dev->bar + NVME_REG_CAP)); | |
07836e65 KB |
998 | |
999 | spin_lock_irq(&nvmeq->q_lock); | |
1000 | nvme_process_cq(nvmeq); | |
1001 | spin_unlock_irq(&nvmeq->q_lock); | |
b60503ba MW |
1002 | } |
1003 | ||
8ffaadf7 JD |
1004 | static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, |
1005 | int entry_size) | |
1006 | { | |
1007 | int q_depth = dev->q_depth; | |
5fd4ce1b CH |
1008 | unsigned q_size_aligned = roundup(q_depth * entry_size, |
1009 | dev->ctrl.page_size); | |
8ffaadf7 JD |
1010 | |
1011 | if (q_size_aligned * nr_io_queues > dev->cmb_size) { | |
c45f5c99 | 1012 | u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); |
5fd4ce1b | 1013 | mem_per_q = round_down(mem_per_q, dev->ctrl.page_size); |
c45f5c99 | 1014 | q_depth = div_u64(mem_per_q, entry_size); |
8ffaadf7 JD |
1015 | |
1016 | /* | |
1017 | * Ensure the reduced q_depth is above some threshold where it | |
1018 | * would be better to map queues in system memory with the | |
1019 | * original depth | |
1020 | */ | |
1021 | if (q_depth < 64) | |
1022 | return -ENOMEM; | |
1023 | } | |
1024 | ||
1025 | return q_depth; | |
1026 | } | |
1027 | ||
1028 | static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, | |
1029 | int qid, int depth) | |
1030 | { | |
1031 | if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) { | |
5fd4ce1b CH |
1032 | unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth), |
1033 | dev->ctrl.page_size); | |
8ffaadf7 JD |
1034 | nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset; |
1035 | nvmeq->sq_cmds_io = dev->cmb + offset; | |
1036 | } else { | |
1037 | nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth), | |
1038 | &nvmeq->sq_dma_addr, GFP_KERNEL); | |
1039 | if (!nvmeq->sq_cmds) | |
1040 | return -ENOMEM; | |
1041 | } | |
1042 | ||
1043 | return 0; | |
1044 | } | |
1045 | ||
b60503ba | 1046 | static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, |
2b25d981 | 1047 | int depth) |
b60503ba | 1048 | { |
a4aea562 | 1049 | struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL); |
b60503ba MW |
1050 | if (!nvmeq) |
1051 | return NULL; | |
1052 | ||
e75ec752 | 1053 | nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth), |
4d51abf9 | 1054 | &nvmeq->cq_dma_addr, GFP_KERNEL); |
b60503ba MW |
1055 | if (!nvmeq->cqes) |
1056 | goto free_nvmeq; | |
b60503ba | 1057 | |
8ffaadf7 | 1058 | if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth)) |
b60503ba MW |
1059 | goto free_cqdma; |
1060 | ||
e75ec752 | 1061 | nvmeq->q_dmadev = dev->dev; |
091b6092 | 1062 | nvmeq->dev = dev; |
3193f07b | 1063 | snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d", |
1c63dc66 | 1064 | dev->ctrl.instance, qid); |
b60503ba MW |
1065 | spin_lock_init(&nvmeq->q_lock); |
1066 | nvmeq->cq_head = 0; | |
82123460 | 1067 | nvmeq->cq_phase = 1; |
b80d5ccc | 1068 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
b60503ba | 1069 | nvmeq->q_depth = depth; |
c30341dc | 1070 | nvmeq->qid = qid; |
758dd7fd | 1071 | nvmeq->cq_vector = -1; |
a4aea562 | 1072 | dev->queues[qid] = nvmeq; |
36a7e993 JD |
1073 | dev->queue_count++; |
1074 | ||
b60503ba MW |
1075 | return nvmeq; |
1076 | ||
1077 | free_cqdma: | |
e75ec752 | 1078 | dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes, |
b60503ba MW |
1079 | nvmeq->cq_dma_addr); |
1080 | free_nvmeq: | |
1081 | kfree(nvmeq); | |
1082 | return NULL; | |
1083 | } | |
1084 | ||
dca51e78 | 1085 | static int queue_request_irq(struct nvme_queue *nvmeq) |
3001082c | 1086 | { |
58ffacb5 | 1087 | if (use_threaded_interrupts) |
dca51e78 CH |
1088 | return request_threaded_irq(nvmeq_irq(nvmeq), nvme_irq_check, |
1089 | nvme_irq, IRQF_SHARED, nvmeq->irqname, nvmeq); | |
1090 | else | |
1091 | return request_irq(nvmeq_irq(nvmeq), nvme_irq, IRQF_SHARED, | |
1092 | nvmeq->irqname, nvmeq); | |
3001082c MW |
1093 | } |
1094 | ||
22404274 | 1095 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1096 | { |
22404274 | 1097 | struct nvme_dev *dev = nvmeq->dev; |
b60503ba | 1098 | |
7be50e93 | 1099 | spin_lock_irq(&nvmeq->q_lock); |
22404274 KB |
1100 | nvmeq->sq_tail = 0; |
1101 | nvmeq->cq_head = 0; | |
1102 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1103 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
22404274 | 1104 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); |
42f61420 | 1105 | dev->online_queues++; |
7be50e93 | 1106 | spin_unlock_irq(&nvmeq->q_lock); |
22404274 KB |
1107 | } |
1108 | ||
1109 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) | |
1110 | { | |
1111 | struct nvme_dev *dev = nvmeq->dev; | |
1112 | int result; | |
3f85d50b | 1113 | |
2b25d981 | 1114 | nvmeq->cq_vector = qid - 1; |
b60503ba MW |
1115 | result = adapter_alloc_cq(dev, qid, nvmeq); |
1116 | if (result < 0) | |
22404274 | 1117 | return result; |
b60503ba MW |
1118 | |
1119 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1120 | if (result < 0) | |
1121 | goto release_cq; | |
1122 | ||
dca51e78 | 1123 | result = queue_request_irq(nvmeq); |
b60503ba MW |
1124 | if (result < 0) |
1125 | goto release_sq; | |
1126 | ||
22404274 | 1127 | nvme_init_queue(nvmeq, qid); |
22404274 | 1128 | return result; |
b60503ba MW |
1129 | |
1130 | release_sq: | |
1131 | adapter_delete_sq(dev, qid); | |
1132 | release_cq: | |
1133 | adapter_delete_cq(dev, qid); | |
22404274 | 1134 | return result; |
b60503ba MW |
1135 | } |
1136 | ||
a4aea562 | 1137 | static struct blk_mq_ops nvme_mq_admin_ops = { |
d29ec824 | 1138 | .queue_rq = nvme_queue_rq, |
eee417b0 | 1139 | .complete = nvme_complete_rq, |
a4aea562 | 1140 | .init_hctx = nvme_admin_init_hctx, |
4af0e21c | 1141 | .exit_hctx = nvme_admin_exit_hctx, |
a4aea562 MB |
1142 | .init_request = nvme_admin_init_request, |
1143 | .timeout = nvme_timeout, | |
1144 | }; | |
1145 | ||
1146 | static struct blk_mq_ops nvme_mq_ops = { | |
1147 | .queue_rq = nvme_queue_rq, | |
eee417b0 | 1148 | .complete = nvme_complete_rq, |
a4aea562 MB |
1149 | .init_hctx = nvme_init_hctx, |
1150 | .init_request = nvme_init_request, | |
dca51e78 | 1151 | .map_queues = nvme_pci_map_queues, |
a4aea562 | 1152 | .timeout = nvme_timeout, |
a0fa9647 | 1153 | .poll = nvme_poll, |
a4aea562 MB |
1154 | }; |
1155 | ||
ea191d2f KB |
1156 | static void nvme_dev_remove_admin(struct nvme_dev *dev) |
1157 | { | |
1c63dc66 | 1158 | if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) { |
69d9a99c KB |
1159 | /* |
1160 | * If the controller was reset during removal, it's possible | |
1161 | * user requests may be waiting on a stopped queue. Start the | |
1162 | * queue to flush these to completion. | |
1163 | */ | |
1164 | blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true); | |
1c63dc66 | 1165 | blk_cleanup_queue(dev->ctrl.admin_q); |
ea191d2f KB |
1166 | blk_mq_free_tag_set(&dev->admin_tagset); |
1167 | } | |
1168 | } | |
1169 | ||
a4aea562 MB |
1170 | static int nvme_alloc_admin_tags(struct nvme_dev *dev) |
1171 | { | |
1c63dc66 | 1172 | if (!dev->ctrl.admin_q) { |
a4aea562 MB |
1173 | dev->admin_tagset.ops = &nvme_mq_admin_ops; |
1174 | dev->admin_tagset.nr_hw_queues = 1; | |
e3e9d50c KB |
1175 | |
1176 | /* | |
1177 | * Subtract one to leave an empty queue entry for 'Full Queue' | |
1178 | * condition. See NVM-Express 1.2 specification, section 4.1.2. | |
1179 | */ | |
1180 | dev->admin_tagset.queue_depth = NVME_AQ_BLKMQ_DEPTH - 1; | |
a4aea562 | 1181 | dev->admin_tagset.timeout = ADMIN_TIMEOUT; |
e75ec752 | 1182 | dev->admin_tagset.numa_node = dev_to_node(dev->dev); |
ac3dd5bd | 1183 | dev->admin_tagset.cmd_size = nvme_cmd_size(dev); |
d3484991 | 1184 | dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED; |
a4aea562 MB |
1185 | dev->admin_tagset.driver_data = dev; |
1186 | ||
1187 | if (blk_mq_alloc_tag_set(&dev->admin_tagset)) | |
1188 | return -ENOMEM; | |
1189 | ||
1c63dc66 CH |
1190 | dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset); |
1191 | if (IS_ERR(dev->ctrl.admin_q)) { | |
a4aea562 MB |
1192 | blk_mq_free_tag_set(&dev->admin_tagset); |
1193 | return -ENOMEM; | |
1194 | } | |
1c63dc66 | 1195 | if (!blk_get_queue(dev->ctrl.admin_q)) { |
ea191d2f | 1196 | nvme_dev_remove_admin(dev); |
1c63dc66 | 1197 | dev->ctrl.admin_q = NULL; |
ea191d2f KB |
1198 | return -ENODEV; |
1199 | } | |
0fb59cbc | 1200 | } else |
25646264 | 1201 | blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true); |
a4aea562 MB |
1202 | |
1203 | return 0; | |
1204 | } | |
1205 | ||
8d85fce7 | 1206 | static int nvme_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1207 | { |
ba47e386 | 1208 | int result; |
b60503ba | 1209 | u32 aqa; |
7a67cbea | 1210 | u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP); |
b60503ba MW |
1211 | struct nvme_queue *nvmeq; |
1212 | ||
8ef2074d | 1213 | dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ? |
dfbac8c7 KB |
1214 | NVME_CAP_NSSRC(cap) : 0; |
1215 | ||
7a67cbea CH |
1216 | if (dev->subsystem && |
1217 | (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO)) | |
1218 | writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS); | |
dfbac8c7 | 1219 | |
5fd4ce1b | 1220 | result = nvme_disable_ctrl(&dev->ctrl, cap); |
ba47e386 MW |
1221 | if (result < 0) |
1222 | return result; | |
b60503ba | 1223 | |
a4aea562 | 1224 | nvmeq = dev->queues[0]; |
cd638946 | 1225 | if (!nvmeq) { |
2b25d981 | 1226 | nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); |
cd638946 KB |
1227 | if (!nvmeq) |
1228 | return -ENOMEM; | |
cd638946 | 1229 | } |
b60503ba MW |
1230 | |
1231 | aqa = nvmeq->q_depth - 1; | |
1232 | aqa |= aqa << 16; | |
1233 | ||
7a67cbea CH |
1234 | writel(aqa, dev->bar + NVME_REG_AQA); |
1235 | lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ); | |
1236 | lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ); | |
b60503ba | 1237 | |
5fd4ce1b | 1238 | result = nvme_enable_ctrl(&dev->ctrl, cap); |
025c557a | 1239 | if (result) |
d4875622 | 1240 | return result; |
a4aea562 | 1241 | |
2b25d981 | 1242 | nvmeq->cq_vector = 0; |
dca51e78 | 1243 | result = queue_request_irq(nvmeq); |
758dd7fd JD |
1244 | if (result) { |
1245 | nvmeq->cq_vector = -1; | |
d4875622 | 1246 | return result; |
758dd7fd | 1247 | } |
025c557a | 1248 | |
b60503ba MW |
1249 | return result; |
1250 | } | |
1251 | ||
c875a709 GP |
1252 | static bool nvme_should_reset(struct nvme_dev *dev, u32 csts) |
1253 | { | |
1254 | ||
1255 | /* If true, indicates loss of adapter communication, possibly by a | |
1256 | * NVMe Subsystem reset. | |
1257 | */ | |
1258 | bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO); | |
1259 | ||
1260 | /* If there is a reset ongoing, we shouldn't reset again. */ | |
1261 | if (work_busy(&dev->reset_work)) | |
1262 | return false; | |
1263 | ||
1264 | /* We shouldn't reset unless the controller is on fatal error state | |
1265 | * _or_ if we lost the communication with it. | |
1266 | */ | |
1267 | if (!(csts & NVME_CSTS_CFS) && !nssro) | |
1268 | return false; | |
1269 | ||
1270 | /* If PCI error recovery process is happening, we cannot reset or | |
1271 | * the recovery mechanism will surely fail. | |
1272 | */ | |
1273 | if (pci_channel_offline(to_pci_dev(dev->dev))) | |
1274 | return false; | |
1275 | ||
1276 | return true; | |
1277 | } | |
1278 | ||
d2a61918 AL |
1279 | static void nvme_warn_reset(struct nvme_dev *dev, u32 csts) |
1280 | { | |
1281 | /* Read a config register to help see what died. */ | |
1282 | u16 pci_status; | |
1283 | int result; | |
1284 | ||
1285 | result = pci_read_config_word(to_pci_dev(dev->dev), PCI_STATUS, | |
1286 | &pci_status); | |
1287 | if (result == PCIBIOS_SUCCESSFUL) | |
1288 | dev_warn(dev->dev, | |
1289 | "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n", | |
1290 | csts, pci_status); | |
1291 | else | |
1292 | dev_warn(dev->dev, | |
1293 | "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n", | |
1294 | csts, result); | |
1295 | } | |
1296 | ||
2d55cd5f | 1297 | static void nvme_watchdog_timer(unsigned long data) |
1fa6aead | 1298 | { |
2d55cd5f CH |
1299 | struct nvme_dev *dev = (struct nvme_dev *)data; |
1300 | u32 csts = readl(dev->bar + NVME_REG_CSTS); | |
1fa6aead | 1301 | |
c875a709 GP |
1302 | /* Skip controllers under certain specific conditions. */ |
1303 | if (nvme_should_reset(dev, csts)) { | |
c5f6ce97 | 1304 | if (!nvme_reset(dev)) |
d2a61918 | 1305 | nvme_warn_reset(dev, csts); |
2d55cd5f | 1306 | return; |
1fa6aead | 1307 | } |
2d55cd5f CH |
1308 | |
1309 | mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ)); | |
1fa6aead MW |
1310 | } |
1311 | ||
749941f2 | 1312 | static int nvme_create_io_queues(struct nvme_dev *dev) |
42f61420 | 1313 | { |
949928c1 | 1314 | unsigned i, max; |
749941f2 | 1315 | int ret = 0; |
42f61420 | 1316 | |
749941f2 CH |
1317 | for (i = dev->queue_count; i <= dev->max_qid; i++) { |
1318 | if (!nvme_alloc_queue(dev, i, dev->q_depth)) { | |
1319 | ret = -ENOMEM; | |
42f61420 | 1320 | break; |
749941f2 CH |
1321 | } |
1322 | } | |
42f61420 | 1323 | |
949928c1 KB |
1324 | max = min(dev->max_qid, dev->queue_count - 1); |
1325 | for (i = dev->online_queues; i <= max; i++) { | |
749941f2 | 1326 | ret = nvme_create_queue(dev->queues[i], i); |
d4875622 | 1327 | if (ret) |
42f61420 | 1328 | break; |
27e8166c | 1329 | } |
749941f2 CH |
1330 | |
1331 | /* | |
1332 | * Ignore failing Create SQ/CQ commands, we can continue with less | |
1333 | * than the desired aount of queues, and even a controller without | |
1334 | * I/O queues an still be used to issue admin commands. This might | |
1335 | * be useful to upgrade a buggy firmware for example. | |
1336 | */ | |
1337 | return ret >= 0 ? 0 : ret; | |
b60503ba MW |
1338 | } |
1339 | ||
202021c1 SB |
1340 | static ssize_t nvme_cmb_show(struct device *dev, |
1341 | struct device_attribute *attr, | |
1342 | char *buf) | |
1343 | { | |
1344 | struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev)); | |
1345 | ||
c965809c | 1346 | return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz : x%08x\n", |
202021c1 SB |
1347 | ndev->cmbloc, ndev->cmbsz); |
1348 | } | |
1349 | static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL); | |
1350 | ||
8ffaadf7 JD |
1351 | static void __iomem *nvme_map_cmb(struct nvme_dev *dev) |
1352 | { | |
1353 | u64 szu, size, offset; | |
8ffaadf7 JD |
1354 | resource_size_t bar_size; |
1355 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
1356 | void __iomem *cmb; | |
1357 | dma_addr_t dma_addr; | |
1358 | ||
7a67cbea | 1359 | dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ); |
8ffaadf7 JD |
1360 | if (!(NVME_CMB_SZ(dev->cmbsz))) |
1361 | return NULL; | |
202021c1 | 1362 | dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC); |
8ffaadf7 | 1363 | |
202021c1 SB |
1364 | if (!use_cmb_sqes) |
1365 | return NULL; | |
8ffaadf7 JD |
1366 | |
1367 | szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz)); | |
1368 | size = szu * NVME_CMB_SZ(dev->cmbsz); | |
202021c1 SB |
1369 | offset = szu * NVME_CMB_OFST(dev->cmbloc); |
1370 | bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc)); | |
8ffaadf7 JD |
1371 | |
1372 | if (offset > bar_size) | |
1373 | return NULL; | |
1374 | ||
1375 | /* | |
1376 | * Controllers may support a CMB size larger than their BAR, | |
1377 | * for example, due to being behind a bridge. Reduce the CMB to | |
1378 | * the reported size of the BAR | |
1379 | */ | |
1380 | if (size > bar_size - offset) | |
1381 | size = bar_size - offset; | |
1382 | ||
202021c1 | 1383 | dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset; |
8ffaadf7 JD |
1384 | cmb = ioremap_wc(dma_addr, size); |
1385 | if (!cmb) | |
1386 | return NULL; | |
1387 | ||
1388 | dev->cmb_dma_addr = dma_addr; | |
1389 | dev->cmb_size = size; | |
1390 | return cmb; | |
1391 | } | |
1392 | ||
1393 | static inline void nvme_release_cmb(struct nvme_dev *dev) | |
1394 | { | |
1395 | if (dev->cmb) { | |
1396 | iounmap(dev->cmb); | |
1397 | dev->cmb = NULL; | |
1398 | } | |
1399 | } | |
1400 | ||
9d713c2b KB |
1401 | static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
1402 | { | |
b80d5ccc | 1403 | return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride); |
9d713c2b KB |
1404 | } |
1405 | ||
8d85fce7 | 1406 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 1407 | { |
a4aea562 | 1408 | struct nvme_queue *adminq = dev->queues[0]; |
e75ec752 | 1409 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
dca51e78 | 1410 | int result, nr_io_queues, size; |
b60503ba | 1411 | |
2800b8e7 | 1412 | nr_io_queues = num_online_cpus(); |
9a0be7ab CH |
1413 | result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues); |
1414 | if (result < 0) | |
1b23484b | 1415 | return result; |
9a0be7ab | 1416 | |
f5fa90dc | 1417 | if (nr_io_queues == 0) |
a5229050 | 1418 | return 0; |
b60503ba | 1419 | |
8ffaadf7 JD |
1420 | if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) { |
1421 | result = nvme_cmb_qdepth(dev, nr_io_queues, | |
1422 | sizeof(struct nvme_command)); | |
1423 | if (result > 0) | |
1424 | dev->q_depth = result; | |
1425 | else | |
1426 | nvme_release_cmb(dev); | |
1427 | } | |
1428 | ||
9d713c2b KB |
1429 | size = db_bar_size(dev, nr_io_queues); |
1430 | if (size > 8192) { | |
f1938f6e | 1431 | iounmap(dev->bar); |
9d713c2b KB |
1432 | do { |
1433 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
1434 | if (dev->bar) | |
1435 | break; | |
1436 | if (!--nr_io_queues) | |
1437 | return -ENOMEM; | |
1438 | size = db_bar_size(dev, nr_io_queues); | |
1439 | } while (1); | |
7a67cbea | 1440 | dev->dbs = dev->bar + 4096; |
5a92e700 | 1441 | adminq->q_db = dev->dbs; |
f1938f6e MW |
1442 | } |
1443 | ||
9d713c2b | 1444 | /* Deregister the admin queue's interrupt */ |
dca51e78 | 1445 | free_irq(pci_irq_vector(pdev, 0), adminq); |
9d713c2b | 1446 | |
e32efbfc JA |
1447 | /* |
1448 | * If we enable msix early due to not intx, disable it again before | |
1449 | * setting up the full range we need. | |
1450 | */ | |
dca51e78 CH |
1451 | pci_free_irq_vectors(pdev); |
1452 | nr_io_queues = pci_alloc_irq_vectors(pdev, 1, nr_io_queues, | |
1453 | PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY); | |
1454 | if (nr_io_queues <= 0) | |
1455 | return -EIO; | |
1456 | dev->max_qid = nr_io_queues; | |
fa08a396 | 1457 | |
063a8096 MW |
1458 | /* |
1459 | * Should investigate if there's a performance win from allocating | |
1460 | * more queues than interrupt vectors; it might allow the submission | |
1461 | * path to scale better, even if the receive path is limited by the | |
1462 | * number of interrupts. | |
1463 | */ | |
063a8096 | 1464 | |
dca51e78 | 1465 | result = queue_request_irq(adminq); |
758dd7fd JD |
1466 | if (result) { |
1467 | adminq->cq_vector = -1; | |
d4875622 | 1468 | return result; |
758dd7fd | 1469 | } |
749941f2 | 1470 | return nvme_create_io_queues(dev); |
b60503ba MW |
1471 | } |
1472 | ||
db3cbfff | 1473 | static void nvme_del_queue_end(struct request *req, int error) |
a5768aa8 | 1474 | { |
db3cbfff | 1475 | struct nvme_queue *nvmeq = req->end_io_data; |
b5875222 | 1476 | |
db3cbfff KB |
1477 | blk_mq_free_request(req); |
1478 | complete(&nvmeq->dev->ioq_wait); | |
a5768aa8 KB |
1479 | } |
1480 | ||
db3cbfff | 1481 | static void nvme_del_cq_end(struct request *req, int error) |
a5768aa8 | 1482 | { |
db3cbfff | 1483 | struct nvme_queue *nvmeq = req->end_io_data; |
a5768aa8 | 1484 | |
db3cbfff KB |
1485 | if (!error) { |
1486 | unsigned long flags; | |
1487 | ||
2e39e0f6 ML |
1488 | /* |
1489 | * We might be called with the AQ q_lock held | |
1490 | * and the I/O queue q_lock should always | |
1491 | * nest inside the AQ one. | |
1492 | */ | |
1493 | spin_lock_irqsave_nested(&nvmeq->q_lock, flags, | |
1494 | SINGLE_DEPTH_NESTING); | |
db3cbfff KB |
1495 | nvme_process_cq(nvmeq); |
1496 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); | |
a5768aa8 | 1497 | } |
db3cbfff KB |
1498 | |
1499 | nvme_del_queue_end(req, error); | |
a5768aa8 KB |
1500 | } |
1501 | ||
db3cbfff | 1502 | static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode) |
bda4e0fb | 1503 | { |
db3cbfff KB |
1504 | struct request_queue *q = nvmeq->dev->ctrl.admin_q; |
1505 | struct request *req; | |
1506 | struct nvme_command cmd; | |
bda4e0fb | 1507 | |
db3cbfff KB |
1508 | memset(&cmd, 0, sizeof(cmd)); |
1509 | cmd.delete_queue.opcode = opcode; | |
1510 | cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
bda4e0fb | 1511 | |
eb71f435 | 1512 | req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); |
db3cbfff KB |
1513 | if (IS_ERR(req)) |
1514 | return PTR_ERR(req); | |
bda4e0fb | 1515 | |
db3cbfff KB |
1516 | req->timeout = ADMIN_TIMEOUT; |
1517 | req->end_io_data = nvmeq; | |
1518 | ||
1519 | blk_execute_rq_nowait(q, NULL, req, false, | |
1520 | opcode == nvme_admin_delete_cq ? | |
1521 | nvme_del_cq_end : nvme_del_queue_end); | |
1522 | return 0; | |
bda4e0fb KB |
1523 | } |
1524 | ||
70659060 | 1525 | static void nvme_disable_io_queues(struct nvme_dev *dev, int queues) |
a5768aa8 | 1526 | { |
70659060 | 1527 | int pass; |
db3cbfff KB |
1528 | unsigned long timeout; |
1529 | u8 opcode = nvme_admin_delete_sq; | |
a5768aa8 | 1530 | |
db3cbfff | 1531 | for (pass = 0; pass < 2; pass++) { |
014a0d60 | 1532 | int sent = 0, i = queues; |
db3cbfff KB |
1533 | |
1534 | reinit_completion(&dev->ioq_wait); | |
1535 | retry: | |
1536 | timeout = ADMIN_TIMEOUT; | |
c21377f8 GKB |
1537 | for (; i > 0; i--, sent++) |
1538 | if (nvme_delete_queue(dev->queues[i], opcode)) | |
db3cbfff | 1539 | break; |
c21377f8 | 1540 | |
db3cbfff KB |
1541 | while (sent--) { |
1542 | timeout = wait_for_completion_io_timeout(&dev->ioq_wait, timeout); | |
1543 | if (timeout == 0) | |
1544 | return; | |
1545 | if (i) | |
1546 | goto retry; | |
1547 | } | |
1548 | opcode = nvme_admin_delete_cq; | |
1549 | } | |
a5768aa8 KB |
1550 | } |
1551 | ||
422ef0c7 MW |
1552 | /* |
1553 | * Return: error value if an error occurred setting up the queues or calling | |
1554 | * Identify Device. 0 if these succeeded, even if adding some of the | |
1555 | * namespaces failed. At the moment, these failures are silent. TBD which | |
1556 | * failures should be reported. | |
1557 | */ | |
8d85fce7 | 1558 | static int nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 1559 | { |
5bae7f73 | 1560 | if (!dev->ctrl.tagset) { |
ffe7704d KB |
1561 | dev->tagset.ops = &nvme_mq_ops; |
1562 | dev->tagset.nr_hw_queues = dev->online_queues - 1; | |
1563 | dev->tagset.timeout = NVME_IO_TIMEOUT; | |
1564 | dev->tagset.numa_node = dev_to_node(dev->dev); | |
1565 | dev->tagset.queue_depth = | |
a4aea562 | 1566 | min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; |
ffe7704d KB |
1567 | dev->tagset.cmd_size = nvme_cmd_size(dev); |
1568 | dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; | |
1569 | dev->tagset.driver_data = dev; | |
b60503ba | 1570 | |
ffe7704d KB |
1571 | if (blk_mq_alloc_tag_set(&dev->tagset)) |
1572 | return 0; | |
5bae7f73 | 1573 | dev->ctrl.tagset = &dev->tagset; |
949928c1 KB |
1574 | } else { |
1575 | blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1); | |
1576 | ||
1577 | /* Free previously allocated queues that are no longer usable */ | |
1578 | nvme_free_queues(dev, dev->online_queues); | |
ffe7704d | 1579 | } |
949928c1 | 1580 | |
e1e5e564 | 1581 | return 0; |
b60503ba MW |
1582 | } |
1583 | ||
b00a726a | 1584 | static int nvme_pci_enable(struct nvme_dev *dev) |
0877cb0d | 1585 | { |
42f61420 | 1586 | u64 cap; |
b00a726a | 1587 | int result = -ENOMEM; |
e75ec752 | 1588 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
0877cb0d KB |
1589 | |
1590 | if (pci_enable_device_mem(pdev)) | |
1591 | return result; | |
1592 | ||
0877cb0d | 1593 | pci_set_master(pdev); |
0877cb0d | 1594 | |
e75ec752 CH |
1595 | if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) && |
1596 | dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32))) | |
052d0efa | 1597 | goto disable; |
0877cb0d | 1598 | |
7a67cbea | 1599 | if (readl(dev->bar + NVME_REG_CSTS) == -1) { |
0e53d180 | 1600 | result = -ENODEV; |
b00a726a | 1601 | goto disable; |
0e53d180 | 1602 | } |
e32efbfc JA |
1603 | |
1604 | /* | |
a5229050 KB |
1605 | * Some devices and/or platforms don't advertise or work with INTx |
1606 | * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll | |
1607 | * adjust this later. | |
e32efbfc | 1608 | */ |
dca51e78 CH |
1609 | result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES); |
1610 | if (result < 0) | |
1611 | return result; | |
e32efbfc | 1612 | |
7a67cbea CH |
1613 | cap = lo_hi_readq(dev->bar + NVME_REG_CAP); |
1614 | ||
42f61420 KB |
1615 | dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); |
1616 | dev->db_stride = 1 << NVME_CAP_STRIDE(cap); | |
7a67cbea | 1617 | dev->dbs = dev->bar + 4096; |
1f390c1f SG |
1618 | |
1619 | /* | |
1620 | * Temporary fix for the Apple controller found in the MacBook8,1 and | |
1621 | * some MacBook7,1 to avoid controller resets and data loss. | |
1622 | */ | |
1623 | if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) { | |
1624 | dev->q_depth = 2; | |
1625 | dev_warn(dev->dev, "detected Apple NVMe controller, set " | |
1626 | "queue depth=%u to work around controller resets\n", | |
1627 | dev->q_depth); | |
1628 | } | |
1629 | ||
202021c1 SB |
1630 | /* |
1631 | * CMBs can currently only exist on >=1.2 PCIe devices. We only | |
1632 | * populate sysfs if a CMB is implemented. Note that we add the | |
1633 | * CMB attribute to the nvme_ctrl kobj which removes the need to remove | |
1634 | * it on exit. Since nvme_dev_attrs_group has no name we can pass | |
1635 | * NULL as final argument to sysfs_add_file_to_group. | |
1636 | */ | |
1637 | ||
8ef2074d | 1638 | if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) { |
8ffaadf7 | 1639 | dev->cmb = nvme_map_cmb(dev); |
0877cb0d | 1640 | |
202021c1 SB |
1641 | if (dev->cmbsz) { |
1642 | if (sysfs_add_file_to_group(&dev->ctrl.device->kobj, | |
1643 | &dev_attr_cmb.attr, NULL)) | |
1644 | dev_warn(dev->dev, | |
1645 | "failed to add sysfs attribute for CMB\n"); | |
1646 | } | |
1647 | } | |
1648 | ||
a0a3408e KB |
1649 | pci_enable_pcie_error_reporting(pdev); |
1650 | pci_save_state(pdev); | |
0877cb0d KB |
1651 | return 0; |
1652 | ||
1653 | disable: | |
0877cb0d KB |
1654 | pci_disable_device(pdev); |
1655 | return result; | |
1656 | } | |
1657 | ||
1658 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
b00a726a KB |
1659 | { |
1660 | if (dev->bar) | |
1661 | iounmap(dev->bar); | |
a1f447b3 | 1662 | pci_release_mem_regions(to_pci_dev(dev->dev)); |
b00a726a KB |
1663 | } |
1664 | ||
1665 | static void nvme_pci_disable(struct nvme_dev *dev) | |
0877cb0d | 1666 | { |
e75ec752 CH |
1667 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
1668 | ||
dca51e78 | 1669 | pci_free_irq_vectors(pdev); |
0877cb0d | 1670 | |
a0a3408e KB |
1671 | if (pci_is_enabled(pdev)) { |
1672 | pci_disable_pcie_error_reporting(pdev); | |
e75ec752 | 1673 | pci_disable_device(pdev); |
4d115420 | 1674 | } |
4d115420 KB |
1675 | } |
1676 | ||
a5cdb68c | 1677 | static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown) |
b60503ba | 1678 | { |
70659060 | 1679 | int i, queues; |
7c1b2450 | 1680 | u32 csts = -1; |
22404274 | 1681 | |
2d55cd5f | 1682 | del_timer_sync(&dev->watchdog_timer); |
1fa6aead | 1683 | |
77bf25ea | 1684 | mutex_lock(&dev->shutdown_lock); |
b00a726a | 1685 | if (pci_is_enabled(to_pci_dev(dev->dev))) { |
25646264 | 1686 | nvme_stop_queues(&dev->ctrl); |
7a67cbea | 1687 | csts = readl(dev->bar + NVME_REG_CSTS); |
c9d3bf88 | 1688 | } |
c21377f8 | 1689 | |
70659060 | 1690 | queues = dev->online_queues - 1; |
c21377f8 GKB |
1691 | for (i = dev->queue_count - 1; i > 0; i--) |
1692 | nvme_suspend_queue(dev->queues[i]); | |
1693 | ||
7c1b2450 | 1694 | if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) { |
82469c59 GKB |
1695 | /* A device might become IO incapable very soon during |
1696 | * probe, before the admin queue is configured. Thus, | |
1697 | * queue_count can be 0 here. | |
1698 | */ | |
1699 | if (dev->queue_count) | |
1700 | nvme_suspend_queue(dev->queues[0]); | |
4d115420 | 1701 | } else { |
70659060 | 1702 | nvme_disable_io_queues(dev, queues); |
a5cdb68c | 1703 | nvme_disable_admin_queue(dev, shutdown); |
4d115420 | 1704 | } |
b00a726a | 1705 | nvme_pci_disable(dev); |
07836e65 | 1706 | |
e1958e65 ML |
1707 | blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl); |
1708 | blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl); | |
77bf25ea | 1709 | mutex_unlock(&dev->shutdown_lock); |
b60503ba MW |
1710 | } |
1711 | ||
091b6092 MW |
1712 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
1713 | { | |
e75ec752 | 1714 | dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, |
091b6092 MW |
1715 | PAGE_SIZE, PAGE_SIZE, 0); |
1716 | if (!dev->prp_page_pool) | |
1717 | return -ENOMEM; | |
1718 | ||
99802a7a | 1719 | /* Optimisation for I/Os between 4k and 128k */ |
e75ec752 | 1720 | dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, |
99802a7a MW |
1721 | 256, 256, 0); |
1722 | if (!dev->prp_small_pool) { | |
1723 | dma_pool_destroy(dev->prp_page_pool); | |
1724 | return -ENOMEM; | |
1725 | } | |
091b6092 MW |
1726 | return 0; |
1727 | } | |
1728 | ||
1729 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
1730 | { | |
1731 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 1732 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
1733 | } |
1734 | ||
1673f1f0 | 1735 | static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl) |
5e82e952 | 1736 | { |
1673f1f0 | 1737 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
9ac27090 | 1738 | |
e75ec752 | 1739 | put_device(dev->dev); |
4af0e21c KB |
1740 | if (dev->tagset.tags) |
1741 | blk_mq_free_tag_set(&dev->tagset); | |
1c63dc66 CH |
1742 | if (dev->ctrl.admin_q) |
1743 | blk_put_queue(dev->ctrl.admin_q); | |
5e82e952 | 1744 | kfree(dev->queues); |
5e82e952 KB |
1745 | kfree(dev); |
1746 | } | |
1747 | ||
f58944e2 KB |
1748 | static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status) |
1749 | { | |
237045fc | 1750 | dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status); |
f58944e2 KB |
1751 | |
1752 | kref_get(&dev->ctrl.kref); | |
69d9a99c | 1753 | nvme_dev_disable(dev, false); |
f58944e2 KB |
1754 | if (!schedule_work(&dev->remove_work)) |
1755 | nvme_put_ctrl(&dev->ctrl); | |
1756 | } | |
1757 | ||
fd634f41 | 1758 | static void nvme_reset_work(struct work_struct *work) |
5e82e952 | 1759 | { |
fd634f41 | 1760 | struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work); |
a98e58e5 | 1761 | bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL); |
f58944e2 | 1762 | int result = -ENODEV; |
5e82e952 | 1763 | |
bb8d261e | 1764 | if (WARN_ON(dev->ctrl.state == NVME_CTRL_RESETTING)) |
fd634f41 | 1765 | goto out; |
5e82e952 | 1766 | |
fd634f41 CH |
1767 | /* |
1768 | * If we're called to reset a live controller first shut it down before | |
1769 | * moving on. | |
1770 | */ | |
b00a726a | 1771 | if (dev->ctrl.ctrl_config & NVME_CC_ENABLE) |
a5cdb68c | 1772 | nvme_dev_disable(dev, false); |
5e82e952 | 1773 | |
bb8d261e | 1774 | if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING)) |
9bf2b972 KB |
1775 | goto out; |
1776 | ||
b00a726a | 1777 | result = nvme_pci_enable(dev); |
f0b50732 | 1778 | if (result) |
3cf519b5 | 1779 | goto out; |
f0b50732 KB |
1780 | |
1781 | result = nvme_configure_admin_queue(dev); | |
1782 | if (result) | |
f58944e2 | 1783 | goto out; |
f0b50732 | 1784 | |
a4aea562 | 1785 | nvme_init_queue(dev->queues[0], 0); |
0fb59cbc KB |
1786 | result = nvme_alloc_admin_tags(dev); |
1787 | if (result) | |
f58944e2 | 1788 | goto out; |
b9afca3e | 1789 | |
ce4541f4 CH |
1790 | result = nvme_init_identify(&dev->ctrl); |
1791 | if (result) | |
f58944e2 | 1792 | goto out; |
ce4541f4 | 1793 | |
a98e58e5 SB |
1794 | init_opal_dev(&dev->ctrl.opal_dev, &nvme_sec_submit); |
1795 | ||
1796 | if (was_suspend) | |
1797 | opal_unlock_from_suspend(&dev->ctrl.opal_dev); | |
1798 | ||
f0b50732 | 1799 | result = nvme_setup_io_queues(dev); |
badc34d4 | 1800 | if (result) |
f58944e2 | 1801 | goto out; |
f0b50732 | 1802 | |
21f033f7 KB |
1803 | /* |
1804 | * A controller that can not execute IO typically requires user | |
1805 | * intervention to correct. For such degraded controllers, the driver | |
1806 | * should not submit commands the user did not request, so skip | |
1807 | * registering for asynchronous event notification on this condition. | |
1808 | */ | |
f866fc42 CH |
1809 | if (dev->online_queues > 1) |
1810 | nvme_queue_async_events(&dev->ctrl); | |
3cf519b5 | 1811 | |
2d55cd5f | 1812 | mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ)); |
3cf519b5 | 1813 | |
2659e57b CH |
1814 | /* |
1815 | * Keep the controller around but remove all namespaces if we don't have | |
1816 | * any working I/O queue. | |
1817 | */ | |
3cf519b5 | 1818 | if (dev->online_queues < 2) { |
1b3c47c1 | 1819 | dev_warn(dev->ctrl.device, "IO queues not created\n"); |
3b24774e | 1820 | nvme_kill_queues(&dev->ctrl); |
5bae7f73 | 1821 | nvme_remove_namespaces(&dev->ctrl); |
3cf519b5 | 1822 | } else { |
25646264 | 1823 | nvme_start_queues(&dev->ctrl); |
3cf519b5 CH |
1824 | nvme_dev_add(dev); |
1825 | } | |
1826 | ||
bb8d261e CH |
1827 | if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) { |
1828 | dev_warn(dev->ctrl.device, "failed to mark controller live\n"); | |
1829 | goto out; | |
1830 | } | |
92911a55 CH |
1831 | |
1832 | if (dev->online_queues > 1) | |
5955be21 | 1833 | nvme_queue_scan(&dev->ctrl); |
3cf519b5 | 1834 | return; |
f0b50732 | 1835 | |
3cf519b5 | 1836 | out: |
f58944e2 | 1837 | nvme_remove_dead_ctrl(dev, result); |
f0b50732 KB |
1838 | } |
1839 | ||
5c8809e6 | 1840 | static void nvme_remove_dead_ctrl_work(struct work_struct *work) |
9a6b9458 | 1841 | { |
5c8809e6 | 1842 | struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work); |
e75ec752 | 1843 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
9a6b9458 | 1844 | |
69d9a99c | 1845 | nvme_kill_queues(&dev->ctrl); |
9a6b9458 | 1846 | if (pci_get_drvdata(pdev)) |
921920ab | 1847 | device_release_driver(&pdev->dev); |
1673f1f0 | 1848 | nvme_put_ctrl(&dev->ctrl); |
9a6b9458 KB |
1849 | } |
1850 | ||
4cc06521 | 1851 | static int nvme_reset(struct nvme_dev *dev) |
9a6b9458 | 1852 | { |
1c63dc66 | 1853 | if (!dev->ctrl.admin_q || blk_queue_dying(dev->ctrl.admin_q)) |
4cc06521 | 1854 | return -ENODEV; |
c5f6ce97 KB |
1855 | if (work_busy(&dev->reset_work)) |
1856 | return -ENODEV; | |
846cc05f CH |
1857 | if (!queue_work(nvme_workq, &dev->reset_work)) |
1858 | return -EBUSY; | |
846cc05f | 1859 | return 0; |
9a6b9458 KB |
1860 | } |
1861 | ||
1c63dc66 | 1862 | static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) |
9ca97374 | 1863 | { |
1c63dc66 | 1864 | *val = readl(to_nvme_dev(ctrl)->bar + off); |
90667892 | 1865 | return 0; |
9ca97374 TH |
1866 | } |
1867 | ||
5fd4ce1b | 1868 | static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) |
4cc06521 | 1869 | { |
5fd4ce1b CH |
1870 | writel(val, to_nvme_dev(ctrl)->bar + off); |
1871 | return 0; | |
1872 | } | |
4cc06521 | 1873 | |
7fd8930f CH |
1874 | static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) |
1875 | { | |
1876 | *val = readq(to_nvme_dev(ctrl)->bar + off); | |
1877 | return 0; | |
4cc06521 KB |
1878 | } |
1879 | ||
f3ca80fc CH |
1880 | static int nvme_pci_reset_ctrl(struct nvme_ctrl *ctrl) |
1881 | { | |
c5f6ce97 KB |
1882 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
1883 | int ret = nvme_reset(dev); | |
1884 | ||
1885 | if (!ret) | |
1886 | flush_work(&dev->reset_work); | |
1887 | return ret; | |
4cc06521 | 1888 | } |
f3ca80fc | 1889 | |
1c63dc66 | 1890 | static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = { |
1a353d85 | 1891 | .name = "pcie", |
e439bb12 | 1892 | .module = THIS_MODULE, |
1c63dc66 | 1893 | .reg_read32 = nvme_pci_reg_read32, |
5fd4ce1b | 1894 | .reg_write32 = nvme_pci_reg_write32, |
7fd8930f | 1895 | .reg_read64 = nvme_pci_reg_read64, |
f3ca80fc | 1896 | .reset_ctrl = nvme_pci_reset_ctrl, |
1673f1f0 | 1897 | .free_ctrl = nvme_pci_free_ctrl, |
f866fc42 | 1898 | .submit_async_event = nvme_pci_submit_async_event, |
1c63dc66 | 1899 | }; |
4cc06521 | 1900 | |
b00a726a KB |
1901 | static int nvme_dev_map(struct nvme_dev *dev) |
1902 | { | |
b00a726a KB |
1903 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
1904 | ||
a1f447b3 | 1905 | if (pci_request_mem_regions(pdev, "nvme")) |
b00a726a KB |
1906 | return -ENODEV; |
1907 | ||
1908 | dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); | |
1909 | if (!dev->bar) | |
1910 | goto release; | |
1911 | ||
9fa196e7 | 1912 | return 0; |
b00a726a | 1913 | release: |
9fa196e7 MG |
1914 | pci_release_mem_regions(pdev); |
1915 | return -ENODEV; | |
b00a726a KB |
1916 | } |
1917 | ||
8d85fce7 | 1918 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 1919 | { |
a4aea562 | 1920 | int node, result = -ENOMEM; |
b60503ba MW |
1921 | struct nvme_dev *dev; |
1922 | ||
a4aea562 MB |
1923 | node = dev_to_node(&pdev->dev); |
1924 | if (node == NUMA_NO_NODE) | |
2fa84351 | 1925 | set_dev_node(&pdev->dev, first_memory_node); |
a4aea562 MB |
1926 | |
1927 | dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); | |
b60503ba MW |
1928 | if (!dev) |
1929 | return -ENOMEM; | |
a4aea562 MB |
1930 | dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *), |
1931 | GFP_KERNEL, node); | |
b60503ba MW |
1932 | if (!dev->queues) |
1933 | goto free; | |
1934 | ||
e75ec752 | 1935 | dev->dev = get_device(&pdev->dev); |
9a6b9458 | 1936 | pci_set_drvdata(pdev, dev); |
1c63dc66 | 1937 | |
b00a726a KB |
1938 | result = nvme_dev_map(dev); |
1939 | if (result) | |
1940 | goto free; | |
1941 | ||
f3ca80fc | 1942 | INIT_WORK(&dev->reset_work, nvme_reset_work); |
5c8809e6 | 1943 | INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work); |
2d55cd5f CH |
1944 | setup_timer(&dev->watchdog_timer, nvme_watchdog_timer, |
1945 | (unsigned long)dev); | |
77bf25ea | 1946 | mutex_init(&dev->shutdown_lock); |
db3cbfff | 1947 | init_completion(&dev->ioq_wait); |
b60503ba | 1948 | |
091b6092 MW |
1949 | result = nvme_setup_prp_pools(dev); |
1950 | if (result) | |
a96d4f5c | 1951 | goto put_pci; |
4cc06521 | 1952 | |
f3ca80fc CH |
1953 | result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops, |
1954 | id->driver_data); | |
4cc06521 | 1955 | if (result) |
2e1d8448 | 1956 | goto release_pools; |
740216fc | 1957 | |
1b3c47c1 SG |
1958 | dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev)); |
1959 | ||
92f7a162 | 1960 | queue_work(nvme_workq, &dev->reset_work); |
b60503ba MW |
1961 | return 0; |
1962 | ||
0877cb0d | 1963 | release_pools: |
091b6092 | 1964 | nvme_release_prp_pools(dev); |
a96d4f5c | 1965 | put_pci: |
e75ec752 | 1966 | put_device(dev->dev); |
b00a726a | 1967 | nvme_dev_unmap(dev); |
b60503ba MW |
1968 | free: |
1969 | kfree(dev->queues); | |
b60503ba MW |
1970 | kfree(dev); |
1971 | return result; | |
1972 | } | |
1973 | ||
f0d54a54 KB |
1974 | static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) |
1975 | { | |
a6739479 | 1976 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
f0d54a54 | 1977 | |
a6739479 | 1978 | if (prepare) |
a5cdb68c | 1979 | nvme_dev_disable(dev, false); |
a6739479 | 1980 | else |
c5f6ce97 | 1981 | nvme_reset(dev); |
f0d54a54 KB |
1982 | } |
1983 | ||
09ece142 KB |
1984 | static void nvme_shutdown(struct pci_dev *pdev) |
1985 | { | |
1986 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
a5cdb68c | 1987 | nvme_dev_disable(dev, true); |
09ece142 KB |
1988 | } |
1989 | ||
f58944e2 KB |
1990 | /* |
1991 | * The driver's remove may be called on a device in a partially initialized | |
1992 | * state. This function must not have any dependencies on the device state in | |
1993 | * order to proceed. | |
1994 | */ | |
8d85fce7 | 1995 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
1996 | { |
1997 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 | 1998 | |
bb8d261e CH |
1999 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); |
2000 | ||
9a6b9458 | 2001 | pci_set_drvdata(pdev, NULL); |
0ff9d4e1 KB |
2002 | |
2003 | if (!pci_device_is_present(pdev)) | |
2004 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD); | |
2005 | ||
9bf2b972 | 2006 | flush_work(&dev->reset_work); |
53029b04 | 2007 | nvme_uninit_ctrl(&dev->ctrl); |
a5cdb68c | 2008 | nvme_dev_disable(dev, true); |
a4aea562 | 2009 | nvme_dev_remove_admin(dev); |
a1a5ef99 | 2010 | nvme_free_queues(dev, 0); |
8ffaadf7 | 2011 | nvme_release_cmb(dev); |
9a6b9458 | 2012 | nvme_release_prp_pools(dev); |
b00a726a | 2013 | nvme_dev_unmap(dev); |
1673f1f0 | 2014 | nvme_put_ctrl(&dev->ctrl); |
b60503ba MW |
2015 | } |
2016 | ||
13880f5b KB |
2017 | static int nvme_pci_sriov_configure(struct pci_dev *pdev, int numvfs) |
2018 | { | |
2019 | int ret = 0; | |
2020 | ||
2021 | if (numvfs == 0) { | |
2022 | if (pci_vfs_assigned(pdev)) { | |
2023 | dev_warn(&pdev->dev, | |
2024 | "Cannot disable SR-IOV VFs while assigned\n"); | |
2025 | return -EPERM; | |
2026 | } | |
2027 | pci_disable_sriov(pdev); | |
2028 | return 0; | |
2029 | } | |
2030 | ||
2031 | ret = pci_enable_sriov(pdev, numvfs); | |
2032 | return ret ? ret : numvfs; | |
2033 | } | |
2034 | ||
671a6018 | 2035 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
2036 | static int nvme_suspend(struct device *dev) |
2037 | { | |
2038 | struct pci_dev *pdev = to_pci_dev(dev); | |
2039 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
2040 | ||
a5cdb68c | 2041 | nvme_dev_disable(ndev, true); |
cd638946 KB |
2042 | return 0; |
2043 | } | |
2044 | ||
2045 | static int nvme_resume(struct device *dev) | |
2046 | { | |
2047 | struct pci_dev *pdev = to_pci_dev(dev); | |
2048 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 2049 | |
c5f6ce97 | 2050 | nvme_reset(ndev); |
9a6b9458 | 2051 | return 0; |
cd638946 | 2052 | } |
671a6018 | 2053 | #endif |
cd638946 KB |
2054 | |
2055 | static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); | |
b60503ba | 2056 | |
a0a3408e KB |
2057 | static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev, |
2058 | pci_channel_state_t state) | |
2059 | { | |
2060 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
2061 | ||
2062 | /* | |
2063 | * A frozen channel requires a reset. When detected, this method will | |
2064 | * shutdown the controller to quiesce. The controller will be restarted | |
2065 | * after the slot reset through driver's slot_reset callback. | |
2066 | */ | |
a0a3408e KB |
2067 | switch (state) { |
2068 | case pci_channel_io_normal: | |
2069 | return PCI_ERS_RESULT_CAN_RECOVER; | |
2070 | case pci_channel_io_frozen: | |
d011fb31 KB |
2071 | dev_warn(dev->ctrl.device, |
2072 | "frozen state error detected, reset controller\n"); | |
a5cdb68c | 2073 | nvme_dev_disable(dev, false); |
a0a3408e KB |
2074 | return PCI_ERS_RESULT_NEED_RESET; |
2075 | case pci_channel_io_perm_failure: | |
d011fb31 KB |
2076 | dev_warn(dev->ctrl.device, |
2077 | "failure state error detected, request disconnect\n"); | |
a0a3408e KB |
2078 | return PCI_ERS_RESULT_DISCONNECT; |
2079 | } | |
2080 | return PCI_ERS_RESULT_NEED_RESET; | |
2081 | } | |
2082 | ||
2083 | static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev) | |
2084 | { | |
2085 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
2086 | ||
1b3c47c1 | 2087 | dev_info(dev->ctrl.device, "restart after slot reset\n"); |
a0a3408e | 2088 | pci_restore_state(pdev); |
c5f6ce97 | 2089 | nvme_reset(dev); |
a0a3408e KB |
2090 | return PCI_ERS_RESULT_RECOVERED; |
2091 | } | |
2092 | ||
2093 | static void nvme_error_resume(struct pci_dev *pdev) | |
2094 | { | |
2095 | pci_cleanup_aer_uncorrect_error_status(pdev); | |
2096 | } | |
2097 | ||
1d352035 | 2098 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba | 2099 | .error_detected = nvme_error_detected, |
b60503ba MW |
2100 | .slot_reset = nvme_slot_reset, |
2101 | .resume = nvme_error_resume, | |
f0d54a54 | 2102 | .reset_notify = nvme_reset_notify, |
b60503ba MW |
2103 | }; |
2104 | ||
6eb0d698 | 2105 | static const struct pci_device_id nvme_id_table[] = { |
106198ed | 2106 | { PCI_VDEVICE(INTEL, 0x0953), |
08095e70 KB |
2107 | .driver_data = NVME_QUIRK_STRIPE_SIZE | |
2108 | NVME_QUIRK_DISCARD_ZEROES, }, | |
99466e70 KB |
2109 | { PCI_VDEVICE(INTEL, 0x0a53), |
2110 | .driver_data = NVME_QUIRK_STRIPE_SIZE | | |
2111 | NVME_QUIRK_DISCARD_ZEROES, }, | |
2112 | { PCI_VDEVICE(INTEL, 0x0a54), | |
2113 | .driver_data = NVME_QUIRK_STRIPE_SIZE | | |
2114 | NVME_QUIRK_DISCARD_ZEROES, }, | |
540c801c KB |
2115 | { PCI_VDEVICE(INTEL, 0x5845), /* Qemu emulated controller */ |
2116 | .driver_data = NVME_QUIRK_IDENTIFY_CNS, }, | |
54adc010 GP |
2117 | { PCI_DEVICE(0x1c58, 0x0003), /* HGST adapter */ |
2118 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
015282c9 WW |
2119 | { PCI_DEVICE(0x1c5f, 0x0540), /* Memblaze Pblaze4 adapter */ |
2120 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
b60503ba | 2121 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
c74dc780 | 2122 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) }, |
b60503ba MW |
2123 | { 0, } |
2124 | }; | |
2125 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
2126 | ||
2127 | static struct pci_driver nvme_driver = { | |
2128 | .name = "nvme", | |
2129 | .id_table = nvme_id_table, | |
2130 | .probe = nvme_probe, | |
8d85fce7 | 2131 | .remove = nvme_remove, |
09ece142 | 2132 | .shutdown = nvme_shutdown, |
cd638946 KB |
2133 | .driver = { |
2134 | .pm = &nvme_dev_pm_ops, | |
2135 | }, | |
13880f5b | 2136 | .sriov_configure = nvme_pci_sriov_configure, |
b60503ba MW |
2137 | .err_handler = &nvme_err_handler, |
2138 | }; | |
2139 | ||
2140 | static int __init nvme_init(void) | |
2141 | { | |
0ac13140 | 2142 | int result; |
1fa6aead | 2143 | |
92f7a162 | 2144 | nvme_workq = alloc_workqueue("nvme", WQ_UNBOUND | WQ_MEM_RECLAIM, 0); |
9a6b9458 | 2145 | if (!nvme_workq) |
b9afca3e | 2146 | return -ENOMEM; |
9a6b9458 | 2147 | |
f3db22fe KB |
2148 | result = pci_register_driver(&nvme_driver); |
2149 | if (result) | |
576d55d6 | 2150 | destroy_workqueue(nvme_workq); |
b60503ba MW |
2151 | return result; |
2152 | } | |
2153 | ||
2154 | static void __exit nvme_exit(void) | |
2155 | { | |
2156 | pci_unregister_driver(&nvme_driver); | |
9a6b9458 | 2157 | destroy_workqueue(nvme_workq); |
21bd78bc | 2158 | _nvme_check_size(); |
b60503ba MW |
2159 | } |
2160 | ||
2161 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
2162 | MODULE_LICENSE("GPL"); | |
c78b4713 | 2163 | MODULE_VERSION("1.0"); |
b60503ba MW |
2164 | module_init(nvme_init); |
2165 | module_exit(nvme_exit); |