projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NVMe: Add nvme subsystem reset IOCTL
[linux-2.6-block.git] / drivers / block / nvme-core.c
CommitLineData
b60503ba
MW		 1/*
2 * NVM Express device driver
6eb0d698 3 * Copyright (c) 2011-2014, Intel Corporation.
b60503ba
MW		 4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
b60503ba
MW		 13 */
14
15#include <linux/nvme.h>
8de05535 16#include <linux/bitops.h>
b60503ba 17#include <linux/blkdev.h>
a4aea562 18#include <linux/blk-mq.h>
42f61420 19#include <linux/cpu.h>
fd63e9ce 20#include <linux/delay.h>
b60503ba
MW		 21#include <linux/errno.h>
22#include <linux/fs.h>
23#include <linux/genhd.h>
4cc09e2d 24#include <linux/hdreg.h>
5aff9382 25#include <linux/idr.h>
b60503ba
MW		 26#include <linux/init.h>
27#include <linux/interrupt.h>
28#include <linux/io.h>
29#include <linux/kdev_t.h>
1fa6aead 30#include <linux/kthread.h>
b60503ba 31#include <linux/kernel.h>
a5768aa8 32#include <linux/list_sort.h>
b60503ba
MW		 33#include <linux/mm.h>
34#include <linux/module.h>
35#include <linux/moduleparam.h>
36#include <linux/pci.h>
be7b6275 37#include <linux/poison.h>
c3bfe717 38#include <linux/ptrace.h>
b60503ba
MW		 39#include <linux/sched.h>
40#include <linux/slab.h>
e1e5e564 41#include <linux/t10-pi.h>
b60503ba 42#include <linux/types.h>
5d0f6131 43#include <scsi/sg.h>
797a796a
HM		 44#include <asm-generic/io-64-nonatomic-lo-hi.h>
45
b3fffdef 46#define NVME_MINORS (1U << MINORBITS)
9d43cf64 47#define NVME_Q_DEPTH 1024
d31af0a3 48#define NVME_AQ_DEPTH 256
b60503ba
MW		 49#define SQ_SIZE(depth) (depth * sizeof(struct nvme_command))
50#define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion))
9d43cf64 51#define ADMIN_TIMEOUT (admin_timeout * HZ)
2484f407 52#define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ)
9d43cf64
KB		 53
54static unsigned char admin_timeout = 60;
55module_param(admin_timeout, byte, 0644);
56MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
b60503ba 57
bd67608a
MW		 58unsigned char nvme_io_timeout = 30;
59module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
b355084a 60MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
b60503ba 61
2484f407
DM		 62static unsigned char shutdown_timeout = 5;
63module_param(shutdown_timeout, byte, 0644);
64MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
65
b60503ba
MW		 66static int nvme_major;
67module_param(nvme_major, int, 0);
68
b3fffdef
KB		 69static int nvme_char_major;
70module_param(nvme_char_major, int, 0);
71
58ffacb5
MW		 72static int use_threaded_interrupts;
73module_param(use_threaded_interrupts, int, 0);
74
8ffaadf7
JD		 75static bool use_cmb_sqes = true;
76module_param(use_cmb_sqes, bool, 0644);
77MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");
78
1fa6aead
MW		 79static DEFINE_SPINLOCK(dev_list_lock);
80static LIST_HEAD(dev_list);
81static struct task_struct *nvme_thread;
9a6b9458 82static struct workqueue_struct *nvme_workq;
b9afca3e 83static wait_queue_head_t nvme_kthread_wait;
1fa6aead 84
b3fffdef
KB		 85static struct class *nvme_class;
86
d4b4ff8e 87static void nvme_reset_failed_dev(struct work_struct *ws);
4cc06521 88static int nvme_reset(struct nvme_dev *dev);
a4aea562 89static int nvme_process_cq(struct nvme_queue *nvmeq);
d4b4ff8e 90
4d115420
KB		 91struct async_cmd_info {
92 struct kthread_work work;
93 struct kthread_worker *worker;
a4aea562 94 struct request *req;
4d115420
KB		 95 u32 result;
96 int status;
97 void *ctx;
98};
1fa6aead 99
b60503ba
MW		 100/*
101 * An NVM Express queue. Each device has at least two (one for admin
102 * commands and one for I/O commands).
103 */
104struct nvme_queue {
105 struct device *q_dmadev;
091b6092 106 struct nvme_dev *dev;
3193f07b 107 char irqname[24]; /* nvme4294967295-65535\0 */
b60503ba
MW		 108 spinlock_t q_lock;
109 struct nvme_command *sq_cmds;
8ffaadf7 110 struct nvme_command __iomem *sq_cmds_io;
b60503ba 111 volatile struct nvme_completion *cqes;
42483228 112 struct blk_mq_tags **tags;
b60503ba
MW		 113 dma_addr_t sq_dma_addr;
114 dma_addr_t cq_dma_addr;
b60503ba
MW		 115 u32 __iomem *q_db;
116 u16 q_depth;
6222d172 117 s16 cq_vector;
b60503ba
MW		 118 u16 sq_head;
119 u16 sq_tail;
120 u16 cq_head;
c30341dc 121 u16 qid;
e9539f47
MW		 122 u8 cq_phase;
123 u8 cqe_seen;
4d115420 124 struct async_cmd_info cmdinfo;
b60503ba
MW		 125};
126
127/*
128 * Check we didin't inadvertently grow the command struct
129 */
130static inline void _nvme_check_size(void)
131{
132 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
133 BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64);
134 BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64);
135 BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
136 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
f8ebf840 137 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
c30341dc 138 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
b60503ba
MW		 139 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
140 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
141 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
142 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
6ecec745 143 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
b60503ba
MW		 144}
145
edd10d33 146typedef void (*nvme_completion_fn)(struct nvme_queue *, void *,
c2f5b650
MW		 147 struct nvme_completion *);
148
e85248e5 149struct nvme_cmd_info {
c2f5b650
MW		 150 nvme_completion_fn fn;
151 void *ctx;
c30341dc 152 int aborted;
a4aea562 153 struct nvme_queue *nvmeq;
ac3dd5bd 154 struct nvme_iod iod[0];
e85248e5
MW		 155};
156
ac3dd5bd
JA		 157/*
158 * Max size of iod being embedded in the request payload
159 */
160#define NVME_INT_PAGES 2
161#define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->page_size)
fda631ff 162#define NVME_INT_MASK 0x01
ac3dd5bd
JA		 163
164/*
165 * Will slightly overestimate the number of pages needed. This is OK
166 * as it only leads to a small amount of wasted memory for the lifetime of
167 * the I/O.
168 */
169static int nvme_npages(unsigned size, struct nvme_dev *dev)
170{
171 unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size);
172 return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
173}
174
175static unsigned int nvme_cmd_size(struct nvme_dev *dev)
176{
177 unsigned int ret = sizeof(struct nvme_cmd_info);
178
179 ret += sizeof(struct nvme_iod);
180 ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev);
181 ret += sizeof(struct scatterlist) * NVME_INT_PAGES;
182
183 return ret;
184}
185
a4aea562
MB		 186static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
187 unsigned int hctx_idx)
e85248e5 188{
a4aea562
MB		 189 struct nvme_dev *dev = data;
190 struct nvme_queue *nvmeq = dev->queues[0];
191
42483228
KB		 192 WARN_ON(hctx_idx != 0);
193 WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
194 WARN_ON(nvmeq->tags);
195
a4aea562 196 hctx->driver_data = nvmeq;
42483228 197 nvmeq->tags = &dev->admin_tagset.tags[0];
a4aea562 198 return 0;
e85248e5
MW		 199}
200
4af0e21c
KB		 201static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
202{
203 struct nvme_queue *nvmeq = hctx->driver_data;
204
205 nvmeq->tags = NULL;
206}
207
a4aea562
MB		 208static int nvme_admin_init_request(void *data, struct request *req,
209 unsigned int hctx_idx, unsigned int rq_idx,
210 unsigned int numa_node)
22404274 211{
a4aea562
MB		 212 struct nvme_dev *dev = data;
213 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
214 struct nvme_queue *nvmeq = dev->queues[0];
215
216 BUG_ON(!nvmeq);
217 cmd->nvmeq = nvmeq;
218 return 0;
22404274
KB		 219}
220
a4aea562
MB		 221static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
222 unsigned int hctx_idx)
b60503ba 223{
a4aea562 224 struct nvme_dev *dev = data;
42483228 225 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
a4aea562 226
42483228
KB		 227 if (!nvmeq->tags)
228 nvmeq->tags = &dev->tagset.tags[hctx_idx];
b60503ba 229
42483228 230 WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
a4aea562
MB		 231 hctx->driver_data = nvmeq;
232 return 0;
b60503ba
MW		 233}
234
a4aea562
MB		 235static int nvme_init_request(void *data, struct request *req,
236 unsigned int hctx_idx, unsigned int rq_idx,
237 unsigned int numa_node)
b60503ba 238{
a4aea562
MB		 239 struct nvme_dev *dev = data;
240 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
241 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
242
243 BUG_ON(!nvmeq);
244 cmd->nvmeq = nvmeq;
245 return 0;
246}
247
248static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx,
249 nvme_completion_fn handler)
250{
251 cmd->fn = handler;
252 cmd->ctx = ctx;
253 cmd->aborted = 0;
c917dfe5 254 blk_mq_start_request(blk_mq_rq_from_pdu(cmd));
b60503ba
MW		 255}
256
ac3dd5bd
JA		 257static void *iod_get_private(struct nvme_iod *iod)
258{
259 return (void *) (iod->private & ~0x1UL);
260}
261
262/*
263 * If bit 0 is set, the iod is embedded in the request payload.
264 */
265static bool iod_should_kfree(struct nvme_iod *iod)
266{
fda631ff 267 return (iod->private & NVME_INT_MASK) == 0;
ac3dd5bd
JA		 268}
269
c2f5b650
MW		 270/* Special values must be less than 0x1000 */
271#define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA)
d2d87034
MW		 272#define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE)
273#define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE)
274#define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE)
be7b6275 275
edd10d33 276static void special_completion(struct nvme_queue *nvmeq, void *ctx,
c2f5b650
MW		 277 struct nvme_completion *cqe)
278{
279 if (ctx == CMD_CTX_CANCELLED)
280 return;
c2f5b650 281 if (ctx == CMD_CTX_COMPLETED) {
edd10d33 282 dev_warn(nvmeq->q_dmadev,
c2f5b650
MW		 283 "completed id %d twice on queue %d\n",
284 cqe->command_id, le16_to_cpup(&cqe->sq_id));
285 return;
286 }
287 if (ctx == CMD_CTX_INVALID) {
edd10d33 288 dev_warn(nvmeq->q_dmadev,
c2f5b650
MW		 289 "invalid id %d completed on queue %d\n",
290 cqe->command_id, le16_to_cpup(&cqe->sq_id));
291 return;
292 }
edd10d33 293 dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx);
c2f5b650
MW		 294}
295
a4aea562 296static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn)
b60503ba 297{
c2f5b650 298 void *ctx;
b60503ba 299
859361a2 300 if (fn)
a4aea562
MB		 301 *fn = cmd->fn;
302 ctx = cmd->ctx;
303 cmd->fn = special_completion;
304 cmd->ctx = CMD_CTX_CANCELLED;
c2f5b650 305 return ctx;
b60503ba
MW		 306}
307
a4aea562
MB		 308static void async_req_completion(struct nvme_queue *nvmeq, void *ctx,
309 struct nvme_completion *cqe)
3c0cf138 310{
a4aea562
MB		 311 u32 result = le32_to_cpup(&cqe->result);
312 u16 status = le16_to_cpup(&cqe->status) >> 1;
313
314 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ)
315 ++nvmeq->dev->event_limit;
a5768aa8
KB		 316 if (status != NVME_SC_SUCCESS)
317 return;
318
319 switch (result & 0xff07) {
320 case NVME_AER_NOTICE_NS_CHANGED:
321 dev_info(nvmeq->q_dmadev, "rescanning\n");
322 schedule_work(&nvmeq->dev->scan_work);
323 default:
324 dev_warn(nvmeq->q_dmadev, "async event result %08x\n", result);
325 }
b60503ba
MW		 326}
327
a4aea562
MB		 328static void abort_completion(struct nvme_queue *nvmeq, void *ctx,
329 struct nvme_completion *cqe)
5a92e700 330{
a4aea562
MB		 331 struct request *req = ctx;
332
333 u16 status = le16_to_cpup(&cqe->status) >> 1;
334 u32 result = le32_to_cpup(&cqe->result);
a51afb54 335
42483228 336 blk_mq_free_request(req);
a51afb54 337
a4aea562
MB		 338 dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result);
339 ++nvmeq->dev->abort_limit;
5a92e700
KB		 340}
341
a4aea562
MB		 342static void async_completion(struct nvme_queue *nvmeq, void *ctx,
343 struct nvme_completion *cqe)
b60503ba 344{
a4aea562
MB		 345 struct async_cmd_info *cmdinfo = ctx;
346 cmdinfo->result = le32_to_cpup(&cqe->result);
347 cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
348 queue_kthread_work(cmdinfo->worker, &cmdinfo->work);
42483228 349 blk_mq_free_request(cmdinfo->req);
b60503ba
MW		 350}
351
a4aea562
MB		 352static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq,
353 unsigned int tag)
b60503ba 354{
42483228 355 struct request *req = blk_mq_tag_to_rq(*nvmeq->tags, tag);
a51afb54 356
a4aea562 357 return blk_mq_rq_to_pdu(req);
4f5099af
KB		 358}
359
a4aea562
MB		 360/*
361 * Called with local interrupts disabled and the q_lock held. May not sleep.
362 */
363static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag,
364 nvme_completion_fn *fn)
4f5099af 365{
a4aea562
MB		 366 struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag);
367 void *ctx;
368 if (tag >= nvmeq->q_depth) {
369 *fn = special_completion;
370 return CMD_CTX_INVALID;
371 }
372 if (fn)
373 *fn = cmd->fn;
374 ctx = cmd->ctx;
375 cmd->fn = special_completion;
376 cmd->ctx = CMD_CTX_COMPLETED;
377 return ctx;
b60503ba
MW		 378}
379
380/**
714a7a22 381 * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
b60503ba
MW		 382 * @nvmeq: The queue to use
383 * @cmd: The command to send
384 *
385 * Safe to use from interrupt context
386 */
a4aea562 387static int __nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
b60503ba 388{
a4aea562
MB		 389 u16 tail = nvmeq->sq_tail;
390
8ffaadf7
JD		 391 if (nvmeq->sq_cmds_io)
392 memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd));
393 else
394 memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));
395
b60503ba
MW		 396 if (++tail == nvmeq->q_depth)
397 tail = 0;
7547881d 398 writel(tail, nvmeq->q_db);
b60503ba 399 nvmeq->sq_tail = tail;
b60503ba
MW		 400
401 return 0;
402}
403
a4aea562
MB		 404static int nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
405{
406 unsigned long flags;
407 int ret;
408 spin_lock_irqsave(&nvmeq->q_lock, flags);
409 ret = __nvme_submit_cmd(nvmeq, cmd);
410 spin_unlock_irqrestore(&nvmeq->q_lock, flags);
411 return ret;
412}
413
eca18b23 414static __le64 **iod_list(struct nvme_iod *iod)
e025344c 415{
eca18b23 416 return ((void *)iod) + iod->offset;
e025344c
SMM		 417}
418
ac3dd5bd
JA		 419static inline void iod_init(struct nvme_iod *iod, unsigned nbytes,
420 unsigned nseg, unsigned long private)
eca18b23 421{
ac3dd5bd
JA		 422 iod->private = private;
423 iod->offset = offsetof(struct nvme_iod, sg[nseg]);
424 iod->npages = -1;
425 iod->length = nbytes;
426 iod->nents = 0;
eca18b23 427}
b60503ba 428
eca18b23 429static struct nvme_iod *
ac3dd5bd
JA		 430__nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev,
431 unsigned long priv, gfp_t gfp)
b60503ba 432{
eca18b23 433 struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) +
ac3dd5bd 434 sizeof(__le64 *) * nvme_npages(bytes, dev) +
eca18b23
MW		 435 sizeof(struct scatterlist) * nseg, gfp);
436
ac3dd5bd
JA		 437 if (iod)
438 iod_init(iod, bytes, nseg, priv);
eca18b23
MW		 439
440 return iod;
b60503ba
MW		 441}
442
ac3dd5bd
JA		 443static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev,
444 gfp_t gfp)
445{
446 unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) :
447 sizeof(struct nvme_dsm_range);
ac3dd5bd
JA		 448 struct nvme_iod *iod;
449
450 if (rq->nr_phys_segments <= NVME_INT_PAGES &&
451 size <= NVME_INT_BYTES(dev)) {
452 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq);
453
454 iod = cmd->iod;
ac3dd5bd 455 iod_init(iod, size, rq->nr_phys_segments,
fda631ff 456 (unsigned long) rq | NVME_INT_MASK);
ac3dd5bd
JA		 457 return iod;
458 }
459
460 return __nvme_alloc_iod(rq->nr_phys_segments, size, dev,
461 (unsigned long) rq, gfp);
462}
463
d29ec824 464static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod)
b60503ba 465{
1d090624 466 const int last_prp = dev->page_size / 8 - 1;
eca18b23
MW		 467 int i;
468 __le64 **list = iod_list(iod);
469 dma_addr_t prp_dma = iod->first_dma;
470
471 if (iod->npages == 0)
472 dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
473 for (i = 0; i < iod->npages; i++) {
474 __le64 *prp_list = list[i];
475 dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
476 dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
477 prp_dma = next_prp_dma;
478 }
ac3dd5bd
JA		 479
480 if (iod_should_kfree(iod))
481 kfree(iod);
b60503ba
MW		 482}
483
b4ff9c8d
KB		 484static int nvme_error_status(u16 status)
485{
486 switch (status & 0x7ff) {
487 case NVME_SC_SUCCESS:
488 return 0;
489 case NVME_SC_CAP_EXCEEDED:
490 return -ENOSPC;
491 default:
492 return -EIO;
493 }
494}
495
52b68d7e 496#ifdef CONFIG_BLK_DEV_INTEGRITY
e1e5e564
KB		 497static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
498{
499 if (be32_to_cpu(pi->ref_tag) == v)
500 pi->ref_tag = cpu_to_be32(p);
501}
502
503static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
504{
505 if (be32_to_cpu(pi->ref_tag) == p)
506 pi->ref_tag = cpu_to_be32(v);
507}
508
509/**
510 * nvme_dif_remap - remaps ref tags to bip seed and physical lba
511 *
512 * The virtual start sector is the one that was originally submitted by the
513 * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical
514 * start sector may be different. Remap protection information to match the
515 * physical LBA on writes, and back to the original seed on reads.
516 *
517 * Type 0 and 3 do not have a ref tag, so no remapping required.
518 */
519static void nvme_dif_remap(struct request *req,
520 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
521{
522 struct nvme_ns *ns = req->rq_disk->private_data;
523 struct bio_integrity_payload *bip;
524 struct t10_pi_tuple *pi;
525 void *p, *pmap;
526 u32 i, nlb, ts, phys, virt;
527
528 if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3)
529 return;
530
531 bip = bio_integrity(req->bio);
532 if (!bip)
533 return;
534
535 pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset;
e1e5e564
KB		 536
537 p = pmap;
538 virt = bip_get_seed(bip);
539 phys = nvme_block_nr(ns, blk_rq_pos(req));
540 nlb = (blk_rq_bytes(req) >> ns->lba_shift);
541 ts = ns->disk->integrity->tuple_size;
542
543 for (i = 0; i < nlb; i++, virt++, phys++) {
544 pi = (struct t10_pi_tuple *)p;
545 dif_swap(phys, virt, pi);
546 p += ts;
547 }
548 kunmap_atomic(pmap);
549}
550
52b68d7e
KB		 551static int nvme_noop_verify(struct blk_integrity_iter *iter)
552{
553 return 0;
554}
555
556static int nvme_noop_generate(struct blk_integrity_iter *iter)
557{
558 return 0;
559}
560
561struct blk_integrity nvme_meta_noop = {
562 .name = "NVME_META_NOOP",
563 .generate_fn = nvme_noop_generate,
564 .verify_fn = nvme_noop_verify,
565};
566
567static void nvme_init_integrity(struct nvme_ns *ns)
568{
569 struct blk_integrity integrity;
570
571 switch (ns->pi_type) {
572 case NVME_NS_DPS_PI_TYPE3:
573 integrity = t10_pi_type3_crc;
574 break;
575 case NVME_NS_DPS_PI_TYPE1:
576 case NVME_NS_DPS_PI_TYPE2:
577 integrity = t10_pi_type1_crc;
578 break;
579 default:
580 integrity = nvme_meta_noop;
581 break;
582 }
583 integrity.tuple_size = ns->ms;
584 blk_integrity_register(ns->disk, &integrity);
585 blk_queue_max_integrity_segments(ns->queue, 1);
586}
587#else /* CONFIG_BLK_DEV_INTEGRITY */
588static void nvme_dif_remap(struct request *req,
589 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
590{
591}
592static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
593{
594}
595static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
596{
597}
598static void nvme_init_integrity(struct nvme_ns *ns)
599{
600}
601#endif
602
a4aea562 603static void req_completion(struct nvme_queue *nvmeq, void *ctx,
b60503ba
MW		 604 struct nvme_completion *cqe)
605{
eca18b23 606 struct nvme_iod *iod = ctx;
ac3dd5bd 607 struct request *req = iod_get_private(iod);
a4aea562
MB		 608 struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
609
b60503ba
MW		 610 u16 status = le16_to_cpup(&cqe->status) >> 1;
611
edd10d33 612 if (unlikely(status)) {
a4aea562
MB		 613 if (!(status & NVME_SC_DNR || blk_noretry_request(req))
614 && (jiffies - req->start_time) < req->timeout) {
c9d3bf88
KB		 615 unsigned long flags;
616
a4aea562 617 blk_mq_requeue_request(req);
c9d3bf88
KB		 618 spin_lock_irqsave(req->q->queue_lock, flags);
619 if (!blk_queue_stopped(req->q))
620 blk_mq_kick_requeue_list(req->q);
621 spin_unlock_irqrestore(req->q->queue_lock, flags);
edd10d33
KB		 622 return;
623 }
d29ec824 624 if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
17188bb4
KB		 625 if (cmd_rq->ctx == CMD_CTX_CANCELLED)
626 req->errors = -EINTR;
627 else
628 req->errors = status;
d29ec824
CH		 629 } else {
630 req->errors = nvme_error_status(status);
631 }
a4aea562
MB		 632 } else
633 req->errors = 0;
a0a931d6
KB		 634 if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
635 u32 result = le32_to_cpup(&cqe->result);
636 req->special = (void *)(uintptr_t)result;
637 }
a4aea562
MB		 638
639 if (cmd_rq->aborted)
e75ec752 640 dev_warn(nvmeq->dev->dev,
a4aea562
MB		 641 "completing aborted command with status:%04x\n",
642 status);
643
e1e5e564 644 if (iod->nents) {
e75ec752 645 dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents,
a4aea562 646 rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
e1e5e564
KB		 647 if (blk_integrity_rq(req)) {
648 if (!rq_data_dir(req))
649 nvme_dif_remap(req, nvme_dif_complete);
e75ec752 650 dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1,
e1e5e564
KB		 651 rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
652 }
653 }
edd10d33 654 nvme_free_iod(nvmeq->dev, iod);
3291fa57 655
a4aea562 656 blk_mq_complete_request(req);
b60503ba
MW		 657}
658
184d2944 659/* length is in bytes. gfp flags indicates whether we may sleep. */
d29ec824
CH		 660static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod,
661 int total_len, gfp_t gfp)
ff22b54f 662{
99802a7a 663 struct dma_pool *pool;
eca18b23
MW		 664 int length = total_len;
665 struct scatterlist *sg = iod->sg;
ff22b54f
MW		 666 int dma_len = sg_dma_len(sg);
667 u64 dma_addr = sg_dma_address(sg);
f137e0f1
MI		 668 u32 page_size = dev->page_size;
669 int offset = dma_addr & (page_size - 1);
e025344c 670 __le64 *prp_list;
eca18b23 671 __le64 **list = iod_list(iod);
e025344c 672 dma_addr_t prp_dma;
eca18b23 673 int nprps, i;
ff22b54f 674
1d090624 675 length -= (page_size - offset);
ff22b54f 676 if (length <= 0)
eca18b23 677 return total_len;
ff22b54f 678
1d090624 679 dma_len -= (page_size - offset);
ff22b54f 680 if (dma_len) {
1d090624 681 dma_addr += (page_size - offset);
ff22b54f
MW		 682 } else {
683 sg = sg_next(sg);
684 dma_addr = sg_dma_address(sg);
685 dma_len = sg_dma_len(sg);
686 }
687
1d090624 688 if (length <= page_size) {
edd10d33 689 iod->first_dma = dma_addr;
eca18b23 690 return total_len;
e025344c
SMM		 691 }
692
1d090624 693 nprps = DIV_ROUND_UP(length, page_size);
99802a7a
MW		 694 if (nprps <= (256 / 8)) {
695 pool = dev->prp_small_pool;
eca18b23 696 iod->npages = 0;
99802a7a
MW		 697 } else {
698 pool = dev->prp_page_pool;
eca18b23 699 iod->npages = 1;
99802a7a
MW		 700 }
701
b77954cb
MW		 702 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
703 if (!prp_list) {
edd10d33 704 iod->first_dma = dma_addr;
eca18b23 705 iod->npages = -1;
1d090624 706 return (total_len - length) + page_size;
b77954cb 707 }
eca18b23
MW		 708 list[0] = prp_list;
709 iod->first_dma = prp_dma;
e025344c
SMM		 710 i = 0;
711 for (;;) {
1d090624 712 if (i == page_size >> 3) {
e025344c 713 __le64 *old_prp_list = prp_list;
b77954cb 714 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
eca18b23
MW		 715 if (!prp_list)
716 return total_len - length;
717 list[iod->npages++] = prp_list;
7523d834
MW		 718 prp_list[0] = old_prp_list[i - 1];
719 old_prp_list[i - 1] = cpu_to_le64(prp_dma);
720 i = 1;
e025344c
SMM		 721 }
722 prp_list[i++] = cpu_to_le64(dma_addr);
1d090624
KB		 723 dma_len -= page_size;
724 dma_addr += page_size;
725 length -= page_size;
e025344c
SMM		 726 if (length <= 0)
727 break;
728 if (dma_len > 0)
729 continue;
730 BUG_ON(dma_len < 0);
731 sg = sg_next(sg);
732 dma_addr = sg_dma_address(sg);
733 dma_len = sg_dma_len(sg);
ff22b54f
MW		 734 }
735
eca18b23 736 return total_len;
ff22b54f
MW		 737}
738
d29ec824
CH		 739static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req,
740 struct nvme_iod *iod)
741{
498c4394 742 struct nvme_command cmnd;
d29ec824 743
498c4394
JD		 744 memcpy(&cmnd, req->cmd, sizeof(cmnd));
745 cmnd.rw.command_id = req->tag;
d29ec824 746 if (req->nr_phys_segments) {
498c4394
JD		 747 cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
748 cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
d29ec824
CH		 749 }
750
498c4394 751 __nvme_submit_cmd(nvmeq, &cmnd);
d29ec824
CH		 752}
753
a4aea562
MB		 754/*
755 * We reuse the small pool to allocate the 16-byte range here as it is not
756 * worth having a special pool for these or additional cases to handle freeing
757 * the iod.
758 */
759static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns,
760 struct request *req, struct nvme_iod *iod)
0e5e4f0e 761{
edd10d33
KB		 762 struct nvme_dsm_range *range =
763 (struct nvme_dsm_range *)iod_list(iod)[0];
498c4394 764 struct nvme_command cmnd;
0e5e4f0e 765
0e5e4f0e 766 range->cattr = cpu_to_le32(0);
a4aea562
MB		 767 range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift);
768 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
0e5e4f0e 769
498c4394
JD		 770 memset(&cmnd, 0, sizeof(cmnd));
771 cmnd.dsm.opcode = nvme_cmd_dsm;
772 cmnd.dsm.command_id = req->tag;
773 cmnd.dsm.nsid = cpu_to_le32(ns->ns_id);
774 cmnd.dsm.prp1 = cpu_to_le64(iod->first_dma);
775 cmnd.dsm.nr = 0;
776 cmnd.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
0e5e4f0e 777
498c4394 778 __nvme_submit_cmd(nvmeq, &cmnd);
0e5e4f0e
KB		 779}
780
a4aea562 781static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns,
00df5cb4
MW		 782 int cmdid)
783{
498c4394 784 struct nvme_command cmnd;
00df5cb4 785
498c4394
JD		 786 memset(&cmnd, 0, sizeof(cmnd));
787 cmnd.common.opcode = nvme_cmd_flush;
788 cmnd.common.command_id = cmdid;
789 cmnd.common.nsid = cpu_to_le32(ns->ns_id);
00df5cb4 790
498c4394 791 __nvme_submit_cmd(nvmeq, &cmnd);
00df5cb4
MW		 792}
793
a4aea562
MB		 794static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod,
795 struct nvme_ns *ns)
b60503ba 796{
ac3dd5bd 797 struct request *req = iod_get_private(iod);
498c4394 798 struct nvme_command cmnd;
a4aea562
MB		 799 u16 control = 0;
800 u32 dsmgmt = 0;
00df5cb4 801
a4aea562 802 if (req->cmd_flags & REQ_FUA)
b60503ba 803 control |= NVME_RW_FUA;
a4aea562 804 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
b60503ba
MW		 805 control |= NVME_RW_LR;
806
a4aea562 807 if (req->cmd_flags & REQ_RAHEAD)
b60503ba
MW		 808 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
809
498c4394
JD		 810 memset(&cmnd, 0, sizeof(cmnd));
811 cmnd.rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
812 cmnd.rw.command_id = req->tag;
813 cmnd.rw.nsid = cpu_to_le32(ns->ns_id);
814 cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
815 cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
816 cmnd.rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
817 cmnd.rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
e1e5e564
KB		 818
819 if (blk_integrity_rq(req)) {
498c4394 820 cmnd.rw.metadata = cpu_to_le64(sg_dma_address(iod->meta_sg));
e1e5e564
KB		 821 switch (ns->pi_type) {
822 case NVME_NS_DPS_PI_TYPE3:
823 control |= NVME_RW_PRINFO_PRCHK_GUARD;
824 break;
825 case NVME_NS_DPS_PI_TYPE1:
826 case NVME_NS_DPS_PI_TYPE2:
827 control |= NVME_RW_PRINFO_PRCHK_GUARD |
828 NVME_RW_PRINFO_PRCHK_REF;
498c4394 829 cmnd.rw.reftag = cpu_to_le32(
e1e5e564
KB		 830 nvme_block_nr(ns, blk_rq_pos(req)));
831 break;
832 }
833 } else if (ns->ms)
834 control |= NVME_RW_PRINFO_PRACT;
835
498c4394
JD		 836 cmnd.rw.control = cpu_to_le16(control);
837 cmnd.rw.dsmgmt = cpu_to_le32(dsmgmt);
b60503ba 838
498c4394 839 __nvme_submit_cmd(nvmeq, &cmnd);
b60503ba 840
1974b1ae 841 return 0;
edd10d33
KB		 842}
843
d29ec824
CH		 844/*
845 * NOTE: ns is NULL when called on the admin queue.
846 */
a4aea562
MB		 847static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
848 const struct blk_mq_queue_data *bd)
edd10d33 849{
a4aea562
MB		 850 struct nvme_ns *ns = hctx->queue->queuedata;
851 struct nvme_queue *nvmeq = hctx->driver_data;
d29ec824 852 struct nvme_dev *dev = nvmeq->dev;
a4aea562
MB		 853 struct request *req = bd->rq;
854 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
edd10d33 855 struct nvme_iod *iod;
a4aea562 856 enum dma_data_direction dma_dir;
edd10d33 857
e1e5e564
KB		 858 /*
859 * If formated with metadata, require the block layer provide a buffer
860 * unless this namespace is formated such that the metadata can be
861 * stripped/generated by the controller with PRACT=1.
862 */
d29ec824 863 if (ns && ns->ms && !blk_integrity_rq(req)) {
71feb364
KB		 864 if (!(ns->pi_type && ns->ms == 8) &&
865 req->cmd_type != REQ_TYPE_DRV_PRIV) {
e1e5e564
KB		 866 req->errors = -EFAULT;
867 blk_mq_complete_request(req);
868 return BLK_MQ_RQ_QUEUE_OK;
869 }
870 }
871
d29ec824 872 iod = nvme_alloc_iod(req, dev, GFP_ATOMIC);
edd10d33 873 if (!iod)
fe54303e 874 return BLK_MQ_RQ_QUEUE_BUSY;
a4aea562 875
a4aea562 876 if (req->cmd_flags & REQ_DISCARD) {
edd10d33
KB		 877 void *range;
878 /*
879 * We reuse the small pool to allocate the 16-byte range here
880 * as it is not worth having a special pool for these or
881 * additional cases to handle freeing the iod.
882 */
d29ec824 883 range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC,
edd10d33 884 &iod->first_dma);
a4aea562 885 if (!range)
fe54303e 886 goto retry_cmd;
edd10d33
KB		 887 iod_list(iod)[0] = (__le64 *)range;
888 iod->npages = 0;
ac3dd5bd 889 } else if (req->nr_phys_segments) {
a4aea562
MB		 890 dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
891
ac3dd5bd 892 sg_init_table(iod->sg, req->nr_phys_segments);
a4aea562 893 iod->nents = blk_rq_map_sg(req->q, req, iod->sg);
fe54303e
JA		 894 if (!iod->nents)
895 goto error_cmd;
a4aea562
MB		 896
897 if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir))
fe54303e 898 goto retry_cmd;
a4aea562 899
fe54303e 900 if (blk_rq_bytes(req) !=
d29ec824
CH		 901 nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) {
902 dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
fe54303e
JA		 903 goto retry_cmd;
904 }
e1e5e564
KB		 905 if (blk_integrity_rq(req)) {
906 if (blk_rq_count_integrity_sg(req->q, req->bio) != 1)
907 goto error_cmd;
908
909 sg_init_table(iod->meta_sg, 1);
910 if (blk_rq_map_integrity_sg(
911 req->q, req->bio, iod->meta_sg) != 1)
912 goto error_cmd;
913
914 if (rq_data_dir(req))
915 nvme_dif_remap(req, nvme_dif_prep);
916
917 if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir))
918 goto error_cmd;
919 }
edd10d33 920 }
1974b1ae 921
9af8785a 922 nvme_set_info(cmd, iod, req_completion);
a4aea562 923 spin_lock_irq(&nvmeq->q_lock);
d29ec824
CH		 924 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
925 nvme_submit_priv(nvmeq, req, iod);
926 else if (req->cmd_flags & REQ_DISCARD)
a4aea562
MB		 927 nvme_submit_discard(nvmeq, ns, req, iod);
928 else if (req->cmd_flags & REQ_FLUSH)
929 nvme_submit_flush(nvmeq, ns, req->tag);
930 else
931 nvme_submit_iod(nvmeq, iod, ns);
932
933 nvme_process_cq(nvmeq);
934 spin_unlock_irq(&nvmeq->q_lock);
935 return BLK_MQ_RQ_QUEUE_OK;
936
fe54303e 937 error_cmd:
d29ec824 938 nvme_free_iod(dev, iod);
fe54303e
JA		 939 return BLK_MQ_RQ_QUEUE_ERROR;
940 retry_cmd:
d29ec824 941 nvme_free_iod(dev, iod);
fe54303e 942 return BLK_MQ_RQ_QUEUE_BUSY;
b60503ba
MW		 943}
944
e9539f47 945static int nvme_process_cq(struct nvme_queue *nvmeq)
b60503ba 946{
82123460 947 u16 head, phase;
b60503ba 948
b60503ba 949 head = nvmeq->cq_head;
82123460 950 phase = nvmeq->cq_phase;
b60503ba
MW		 951
952 for (;;) {
c2f5b650
MW		 953 void *ctx;
954 nvme_completion_fn fn;
b60503ba 955 struct nvme_completion cqe = nvmeq->cqes[head];
82123460 956 if ((le16_to_cpu(cqe.status) & 1) != phase)
b60503ba
MW		 957 break;
958 nvmeq->sq_head = le16_to_cpu(cqe.sq_head);
959 if (++head == nvmeq->q_depth) {
960 head = 0;
82123460 961 phase = !phase;
b60503ba 962 }
a4aea562 963 ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn);
edd10d33 964 fn(nvmeq, ctx, &cqe);
b60503ba
MW		 965 }
966
967 /* If the controller ignores the cq head doorbell and continuously
968 * writes to the queue, it is theoretically possible to wrap around
969 * the queue twice and mistakenly return IRQ_NONE. Linux only
970 * requires that 0.1% of your interrupts are handled, so this isn't
971 * a big problem.
972 */
82123460 973 if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
e9539f47 974 return 0;
b60503ba 975
b80d5ccc 976 writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
b60503ba 977 nvmeq->cq_head = head;
82123460 978 nvmeq->cq_phase = phase;
b60503ba 979
e9539f47
MW		 980 nvmeq->cqe_seen = 1;
981 return 1;
b60503ba
MW		 982}
983
984static irqreturn_t nvme_irq(int irq, void *data)
58ffacb5
MW		 985{
986 irqreturn_t result;
987 struct nvme_queue *nvmeq = data;
988 spin_lock(&nvmeq->q_lock);
e9539f47
MW		 989 nvme_process_cq(nvmeq);
990 result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
991 nvmeq->cqe_seen = 0;
58ffacb5
MW		 992 spin_unlock(&nvmeq->q_lock);
993 return result;
994}
995
996static irqreturn_t nvme_irq_check(int irq, void *data)
997{
998 struct nvme_queue *nvmeq = data;
999 struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head];
1000 if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase)
1001 return IRQ_NONE;
1002 return IRQ_WAKE_THREAD;
1003}
1004
b60503ba
MW		 1005/*
1006 * Returns 0 on success. If the result is negative, it's a Linux error code;
1007 * if the result is positive, it's an NVM Express status code
1008 */
d29ec824
CH		 1009int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1010 void *buffer, void __user *ubuffer, unsigned bufflen,
1011 u32 *result, unsigned timeout)
b60503ba 1012{
d29ec824
CH		 1013 bool write = cmd->common.opcode & 1;
1014 struct bio *bio = NULL;
f705f837 1015 struct request *req;
d29ec824 1016 int ret;
b60503ba 1017
d29ec824 1018 req = blk_mq_alloc_request(q, write, GFP_KERNEL, false);
f705f837
CH		 1019 if (IS_ERR(req))
1020 return PTR_ERR(req);
b60503ba 1021
d29ec824 1022 req->cmd_type = REQ_TYPE_DRV_PRIV;
e112af0d 1023 req->cmd_flags |= REQ_FAILFAST_DRIVER;
d29ec824
CH		 1024 req->__data_len = 0;
1025 req->__sector = (sector_t) -1;
1026 req->bio = req->biotail = NULL;
b60503ba 1027
f4ff414a 1028 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
a4aea562 1029
d29ec824
CH		 1030 req->cmd = (unsigned char *)cmd;
1031 req->cmd_len = sizeof(struct nvme_command);
a0a931d6 1032 req->special = (void *)0;
b60503ba 1033
d29ec824
CH		 1034 if (buffer && bufflen) {
1035 ret = blk_rq_map_kern(q, req, buffer, bufflen, __GFP_WAIT);
1036 if (ret)
1037 goto out;
1038 } else if (ubuffer && bufflen) {
1039 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, __GFP_WAIT);
1040 if (ret)
1041 goto out;
1042 bio = req->bio;
1043 }
3c0cf138 1044
d29ec824
CH		 1045 blk_execute_rq(req->q, NULL, req, 0);
1046 if (bio)
1047 blk_rq_unmap_user(bio);
b60503ba 1048 if (result)
a0a931d6 1049 *result = (u32)(uintptr_t)req->special;
d29ec824
CH		 1050 ret = req->errors;
1051 out:
f705f837 1052 blk_mq_free_request(req);
d29ec824 1053 return ret;
f705f837
CH		 1054}
1055
d29ec824
CH		 1056int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1057 void *buffer, unsigned bufflen)
f705f837 1058{
d29ec824 1059 return __nvme_submit_sync_cmd(q, cmd, buffer, NULL, bufflen, NULL, 0);
b60503ba
MW		 1060}
1061
a4aea562
MB		 1062static int nvme_submit_async_admin_req(struct nvme_dev *dev)
1063{
1064 struct nvme_queue *nvmeq = dev->queues[0];
1065 struct nvme_command c;
1066 struct nvme_cmd_info *cmd_info;
1067 struct request *req;
1068
1efccc9d 1069 req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, true);
9f173b33
DC		 1070 if (IS_ERR(req))
1071 return PTR_ERR(req);
a4aea562 1072
c917dfe5 1073 req->cmd_flags |= REQ_NO_TIMEOUT;
a4aea562 1074 cmd_info = blk_mq_rq_to_pdu(req);
1efccc9d 1075 nvme_set_info(cmd_info, NULL, async_req_completion);
a4aea562
MB		 1076
1077 memset(&c, 0, sizeof(c));
1078 c.common.opcode = nvme_admin_async_event;
1079 c.common.command_id = req->tag;
1080
42483228 1081 blk_mq_free_request(req);
a4aea562
MB		 1082 return __nvme_submit_cmd(nvmeq, &c);
1083}
1084
1085static int nvme_submit_admin_async_cmd(struct nvme_dev *dev,
4d115420
KB		 1086 struct nvme_command *cmd,
1087 struct async_cmd_info *cmdinfo, unsigned timeout)
1088{
a4aea562
MB		 1089 struct nvme_queue *nvmeq = dev->queues[0];
1090 struct request *req;
1091 struct nvme_cmd_info *cmd_rq;
4d115420 1092
a4aea562 1093 req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_KERNEL, false);
9f173b33
DC		 1094 if (IS_ERR(req))
1095 return PTR_ERR(req);
a4aea562
MB		 1096
1097 req->timeout = timeout;
1098 cmd_rq = blk_mq_rq_to_pdu(req);
1099 cmdinfo->req = req;
1100 nvme_set_info(cmd_rq, cmdinfo, async_completion);
4d115420 1101 cmdinfo->status = -EINTR;
a4aea562
MB		 1102
1103 cmd->common.command_id = req->tag;
1104
4f5099af 1105 return nvme_submit_cmd(nvmeq, cmd);
4d115420
KB		 1106}
1107
b60503ba
MW		 1108static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
1109{
b60503ba
MW		 1110 struct nvme_command c;
1111
1112 memset(&c, 0, sizeof(c));
1113 c.delete_queue.opcode = opcode;
1114 c.delete_queue.qid = cpu_to_le16(id);
1115
d29ec824 1116 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1117}
1118
1119static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
1120 struct nvme_queue *nvmeq)
1121{
b60503ba
MW		 1122 struct nvme_command c;
1123 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;
1124
d29ec824
CH		 1125 /*
1126 * Note: we (ab)use the fact the the prp fields survive if no data
1127 * is attached to the request.
1128 */
b60503ba
MW		 1129 memset(&c, 0, sizeof(c));
1130 c.create_cq.opcode = nvme_admin_create_cq;
1131 c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr);
1132 c.create_cq.cqid = cpu_to_le16(qid);
1133 c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
1134 c.create_cq.cq_flags = cpu_to_le16(flags);
1135 c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector);
1136
d29ec824 1137 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1138}
1139
1140static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
1141 struct nvme_queue *nvmeq)
1142{
b60503ba
MW		 1143 struct nvme_command c;
1144 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM;
1145
d29ec824
CH		 1146 /*
1147 * Note: we (ab)use the fact the the prp fields survive if no data
1148 * is attached to the request.
1149 */
b60503ba
MW		 1150 memset(&c, 0, sizeof(c));
1151 c.create_sq.opcode = nvme_admin_create_sq;
1152 c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr);
1153 c.create_sq.sqid = cpu_to_le16(qid);
1154 c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
1155 c.create_sq.sq_flags = cpu_to_le16(flags);
1156 c.create_sq.cqid = cpu_to_le16(qid);
1157
d29ec824 1158 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1159}
1160
1161static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid)
1162{
1163 return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid);
1164}
1165
1166static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid)
1167{
1168 return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid);
1169}
1170
d29ec824 1171int nvme_identify_ctrl(struct nvme_dev *dev, struct nvme_id_ctrl **id)
bc5fc7e4 1172{
e44ac588 1173 struct nvme_command c = { };
d29ec824 1174 int error;
bc5fc7e4 1175
e44ac588
AM		 1176 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1177 c.identify.opcode = nvme_admin_identify;
1178 c.identify.cns = cpu_to_le32(1);
1179
d29ec824
CH		 1180 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1181 if (!*id)
1182 return -ENOMEM;
bc5fc7e4 1183
d29ec824
CH		 1184 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1185 sizeof(struct nvme_id_ctrl));
1186 if (error)
1187 kfree(*id);
1188 return error;
1189}
1190
1191int nvme_identify_ns(struct nvme_dev *dev, unsigned nsid,
1192 struct nvme_id_ns **id)
1193{
e44ac588 1194 struct nvme_command c = { };
d29ec824 1195 int error;
bc5fc7e4 1196
e44ac588
AM		 1197 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1198 c.identify.opcode = nvme_admin_identify,
1199 c.identify.nsid = cpu_to_le32(nsid),
1200
d29ec824
CH		 1201 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
1202 if (!*id)
1203 return -ENOMEM;
1204
1205 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1206 sizeof(struct nvme_id_ns));
1207 if (error)
1208 kfree(*id);
1209 return error;
bc5fc7e4
MW		 1210}
1211
5d0f6131 1212int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid,
08df1e05 1213 dma_addr_t dma_addr, u32 *result)
bc5fc7e4
MW		 1214{
1215 struct nvme_command c;
1216
1217 memset(&c, 0, sizeof(c));
1218 c.features.opcode = nvme_admin_get_features;
a42cecce 1219 c.features.nsid = cpu_to_le32(nsid);
bc5fc7e4
MW		 1220 c.features.prp1 = cpu_to_le64(dma_addr);
1221 c.features.fid = cpu_to_le32(fid);
bc5fc7e4 1222
d29ec824
CH		 1223 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0,
1224 result, 0);
df348139
MW		 1225}
1226
5d0f6131
VV		 1227int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11,
1228 dma_addr_t dma_addr, u32 *result)
df348139
MW		 1229{
1230 struct nvme_command c;
1231
1232 memset(&c, 0, sizeof(c));
1233 c.features.opcode = nvme_admin_set_features;
1234 c.features.prp1 = cpu_to_le64(dma_addr);
1235 c.features.fid = cpu_to_le32(fid);
1236 c.features.dword11 = cpu_to_le32(dword11);
1237
d29ec824
CH		 1238 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0,
1239 result, 0);
1240}
1241
1242int nvme_get_log_page(struct nvme_dev *dev, struct nvme_smart_log **log)
1243{
e44ac588
AM		 1244 struct nvme_command c = { };
1245 int error;
1246
1247 c.common.opcode = nvme_admin_get_log_page,
1248 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
1249 c.common.cdw10[0] = cpu_to_le32(
d29ec824
CH		 1250 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
1251 NVME_LOG_SMART),
d29ec824
CH		 1252
1253 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
1254 if (!*log)
1255 return -ENOMEM;
1256
1257 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
1258 sizeof(struct nvme_smart_log));
1259 if (error)
1260 kfree(*log);
1261 return error;
bc5fc7e4
MW		 1262}
1263
c30341dc 1264/**
a4aea562 1265 * nvme_abort_req - Attempt aborting a request
c30341dc
KB		 1266 *
1267 * Schedule controller reset if the command was already aborted once before and
1268 * still hasn't been returned to the driver, or if this is the admin queue.
1269 */
a4aea562 1270static void nvme_abort_req(struct request *req)
c30341dc 1271{
a4aea562
MB		 1272 struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
1273 struct nvme_queue *nvmeq = cmd_rq->nvmeq;
c30341dc 1274 struct nvme_dev *dev = nvmeq->dev;
a4aea562
MB		 1275 struct request *abort_req;
1276 struct nvme_cmd_info *abort_cmd;
1277 struct nvme_command cmd;
c30341dc 1278
a4aea562 1279 if (!nvmeq->qid || cmd_rq->aborted) {
7a509a6b
KB		 1280 unsigned long flags;
1281
1282 spin_lock_irqsave(&dev_list_lock, flags);
c30341dc 1283 if (work_busy(&dev->reset_work))
7a509a6b 1284 goto out;
c30341dc 1285 list_del_init(&dev->node);
e75ec752 1286 dev_warn(dev->dev, "I/O %d QID %d timeout, reset controller\n",
a4aea562 1287 req->tag, nvmeq->qid);
9ca97374 1288 dev->reset_workfn = nvme_reset_failed_dev;
c30341dc 1289 queue_work(nvme_workq, &dev->reset_work);
7a509a6b
KB		 1290 out:
1291 spin_unlock_irqrestore(&dev_list_lock, flags);
c30341dc
KB		 1292 return;
1293 }
1294
1295 if (!dev->abort_limit)
1296 return;
1297
a4aea562
MB		 1298 abort_req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC,
1299 false);
9f173b33 1300 if (IS_ERR(abort_req))
c30341dc
KB		 1301 return;
1302
a4aea562
MB		 1303 abort_cmd = blk_mq_rq_to_pdu(abort_req);
1304 nvme_set_info(abort_cmd, abort_req, abort_completion);
1305
c30341dc
KB		 1306 memset(&cmd, 0, sizeof(cmd));
1307 cmd.abort.opcode = nvme_admin_abort_cmd;
a4aea562 1308 cmd.abort.cid = req->tag;
c30341dc 1309 cmd.abort.sqid = cpu_to_le16(nvmeq->qid);
a4aea562 1310 cmd.abort.command_id = abort_req->tag;
c30341dc
KB		 1311
1312 --dev->abort_limit;
a4aea562 1313 cmd_rq->aborted = 1;
c30341dc 1314
a4aea562 1315 dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag,
c30341dc 1316 nvmeq->qid);
a4aea562
MB		 1317 if (nvme_submit_cmd(dev->queues[0], &cmd) < 0) {
1318 dev_warn(nvmeq->q_dmadev,
1319 "Could not abort I/O %d QID %d",
1320 req->tag, nvmeq->qid);
c87fd540 1321 blk_mq_free_request(abort_req);
a4aea562 1322 }
c30341dc
KB		 1323}
1324
42483228 1325static void nvme_cancel_queue_ios(struct request *req, void *data, bool reserved)
a09115b2 1326{
a4aea562
MB		 1327 struct nvme_queue *nvmeq = data;
1328 void *ctx;
1329 nvme_completion_fn fn;
1330 struct nvme_cmd_info *cmd;
cef6a948
KB		 1331 struct nvme_completion cqe;
1332
1333 if (!blk_mq_request_started(req))
1334 return;
a09115b2 1335
a4aea562 1336 cmd = blk_mq_rq_to_pdu(req);
a09115b2 1337
a4aea562
MB		 1338 if (cmd->ctx == CMD_CTX_CANCELLED)
1339 return;
1340
cef6a948
KB		 1341 if (blk_queue_dying(req->q))
1342 cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1);
1343 else
1344 cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);
1345
1346
a4aea562
MB		 1347 dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n",
1348 req->tag, nvmeq->qid);
1349 ctx = cancel_cmd_info(cmd, &fn);
1350 fn(nvmeq, ctx, &cqe);
a09115b2
MW		 1351}
1352
a4aea562 1353static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
9e866774 1354{
a4aea562
MB		 1355 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
1356 struct nvme_queue *nvmeq = cmd->nvmeq;
1357
1358 dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag,
1359 nvmeq->qid);
7a509a6b 1360 spin_lock_irq(&nvmeq->q_lock);
07836e65 1361 nvme_abort_req(req);
7a509a6b 1362 spin_unlock_irq(&nvmeq->q_lock);
a4aea562 1363
07836e65
KB		 1364 /*
1365 * The aborted req will be completed on receiving the abort req.
1366 * We enable the timer again. If hit twice, it'll cause a device reset,
1367 * as the device then is in a faulty state.
1368 */
1369 return BLK_EH_RESET_TIMER;
a4aea562 1370}
22404274 1371
a4aea562
MB		 1372static void nvme_free_queue(struct nvme_queue *nvmeq)
1373{
9e866774
MW		 1374 dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
1375 (void *)nvmeq->cqes, nvmeq->cq_dma_addr);
8ffaadf7
JD		 1376 if (nvmeq->sq_cmds)
1377 dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
9e866774
MW		 1378 nvmeq->sq_cmds, nvmeq->sq_dma_addr);
1379 kfree(nvmeq);
1380}
1381
a1a5ef99 1382static void nvme_free_queues(struct nvme_dev *dev, int lowest)
22404274
KB		 1383{
1384 int i;
1385
a1a5ef99 1386 for (i = dev->queue_count - 1; i >= lowest; i--) {
a4aea562 1387 struct nvme_queue *nvmeq = dev->queues[i];
22404274 1388 dev->queue_count--;
a4aea562 1389 dev->queues[i] = NULL;
f435c282 1390 nvme_free_queue(nvmeq);
121c7ad4 1391 }
22404274
KB		 1392}
1393
4d115420
KB		 1394/**
1395 * nvme_suspend_queue - put queue into suspended state
1396 * @nvmeq - queue to suspend
4d115420
KB		 1397 */
1398static int nvme_suspend_queue(struct nvme_queue *nvmeq)
b60503ba 1399{
2b25d981 1400 int vector;
b60503ba 1401
a09115b2 1402 spin_lock_irq(&nvmeq->q_lock);
2b25d981
KB		 1403 if (nvmeq->cq_vector == -1) {
1404 spin_unlock_irq(&nvmeq->q_lock);
1405 return 1;
1406 }
1407 vector = nvmeq->dev->entry[nvmeq->cq_vector].vector;
42f61420 1408 nvmeq->dev->online_queues--;
2b25d981 1409 nvmeq->cq_vector = -1;
a09115b2
MW		 1410 spin_unlock_irq(&nvmeq->q_lock);
1411
6df3dbc8
KB		 1412 if (!nvmeq->qid && nvmeq->dev->admin_q)
1413 blk_mq_freeze_queue_start(nvmeq->dev->admin_q);
1414
aba2080f
MW		 1415 irq_set_affinity_hint(vector, NULL);
1416 free_irq(vector, nvmeq);
b60503ba 1417
4d115420
KB		 1418 return 0;
1419}
b60503ba 1420
4d115420
KB		 1421static void nvme_clear_queue(struct nvme_queue *nvmeq)
1422{
22404274 1423 spin_lock_irq(&nvmeq->q_lock);
42483228
KB		 1424 if (nvmeq->tags && *nvmeq->tags)
1425 blk_mq_all_tag_busy_iter(*nvmeq->tags, nvme_cancel_queue_ios, nvmeq);
22404274 1426 spin_unlock_irq(&nvmeq->q_lock);
b60503ba
MW		 1427}
1428
4d115420
KB		 1429static void nvme_disable_queue(struct nvme_dev *dev, int qid)
1430{
a4aea562 1431 struct nvme_queue *nvmeq = dev->queues[qid];
4d115420
KB		 1432
1433 if (!nvmeq)
1434 return;
1435 if (nvme_suspend_queue(nvmeq))
1436 return;
1437
0e53d180
KB		 1438 /* Don't tell the adapter to delete the admin queue.
1439 * Don't tell a removed adapter to delete IO queues. */
1440 if (qid && readl(&dev->bar->csts) != -1) {
b60503ba
MW		 1441 adapter_delete_sq(dev, qid);
1442 adapter_delete_cq(dev, qid);
1443 }
07836e65
KB		 1444
1445 spin_lock_irq(&nvmeq->q_lock);
1446 nvme_process_cq(nvmeq);
1447 spin_unlock_irq(&nvmeq->q_lock);
b60503ba
MW		 1448}
1449
8ffaadf7
JD		 1450static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
1451 int entry_size)
1452{
1453 int q_depth = dev->q_depth;
1454 unsigned q_size_aligned = roundup(q_depth * entry_size, dev->page_size);
1455
1456 if (q_size_aligned * nr_io_queues > dev->cmb_size) {
c45f5c99
JD		 1457 u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1458 mem_per_q = round_down(mem_per_q, dev->page_size);
1459 q_depth = div_u64(mem_per_q, entry_size);
8ffaadf7
JD		 1460
1461 /*
1462 * Ensure the reduced q_depth is above some threshold where it
1463 * would be better to map queues in system memory with the
1464 * original depth
1465 */
1466 if (q_depth < 64)
1467 return -ENOMEM;
1468 }
1469
1470 return q_depth;
1471}
1472
1473static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1474 int qid, int depth)
1475{
1476 if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) {
1477 unsigned offset = (qid - 1) *
1478 roundup(SQ_SIZE(depth), dev->page_size);
1479 nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset;
1480 nvmeq->sq_cmds_io = dev->cmb + offset;
1481 } else {
1482 nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth),
1483 &nvmeq->sq_dma_addr, GFP_KERNEL);
1484 if (!nvmeq->sq_cmds)
1485 return -ENOMEM;
1486 }
1487
1488 return 0;
1489}
1490
b60503ba 1491static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
2b25d981 1492 int depth)
b60503ba 1493{
a4aea562 1494 struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
b60503ba
MW		 1495 if (!nvmeq)
1496 return NULL;
1497
e75ec752 1498 nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
4d51abf9 1499 &nvmeq->cq_dma_addr, GFP_KERNEL);
b60503ba
MW		 1500 if (!nvmeq->cqes)
1501 goto free_nvmeq;
b60503ba 1502
8ffaadf7 1503 if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
b60503ba
MW		 1504 goto free_cqdma;
1505
e75ec752 1506 nvmeq->q_dmadev = dev->dev;
091b6092 1507 nvmeq->dev = dev;
3193f07b
MW		 1508 snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
1509 dev->instance, qid);
b60503ba
MW		 1510 spin_lock_init(&nvmeq->q_lock);
1511 nvmeq->cq_head = 0;
82123460 1512 nvmeq->cq_phase = 1;
b80d5ccc 1513 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
b60503ba 1514 nvmeq->q_depth = depth;
c30341dc 1515 nvmeq->qid = qid;
758dd7fd 1516 nvmeq->cq_vector = -1;
a4aea562 1517 dev->queues[qid] = nvmeq;
b60503ba 1518
36a7e993
JD		 1519 /* make sure queue descriptor is set before queue count, for kthread */
1520 mb();
1521 dev->queue_count++;
1522
b60503ba
MW		 1523 return nvmeq;
1524
1525 free_cqdma:
e75ec752 1526 dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
b60503ba
MW		 1527 nvmeq->cq_dma_addr);
1528 free_nvmeq:
1529 kfree(nvmeq);
1530 return NULL;
1531}
1532
3001082c
MW		 1533static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1534 const char *name)
1535{
58ffacb5
MW		 1536 if (use_threaded_interrupts)
1537 return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector,
481e5bad 1538 nvme_irq_check, nvme_irq, IRQF_SHARED,
58ffacb5 1539 name, nvmeq);
3001082c 1540 return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq,
481e5bad 1541 IRQF_SHARED, name, nvmeq);
3001082c
MW		 1542}
1543
22404274 1544static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
b60503ba 1545{
22404274 1546 struct nvme_dev *dev = nvmeq->dev;
b60503ba 1547
7be50e93 1548 spin_lock_irq(&nvmeq->q_lock);
22404274
KB		 1549 nvmeq->sq_tail = 0;
1550 nvmeq->cq_head = 0;
1551 nvmeq->cq_phase = 1;
b80d5ccc 1552 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
22404274 1553 memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
42f61420 1554 dev->online_queues++;
7be50e93 1555 spin_unlock_irq(&nvmeq->q_lock);
22404274
KB		 1556}
1557
1558static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
1559{
1560 struct nvme_dev *dev = nvmeq->dev;
1561 int result;
3f85d50b 1562
2b25d981 1563 nvmeq->cq_vector = qid - 1;
b60503ba
MW		 1564 result = adapter_alloc_cq(dev, qid, nvmeq);
1565 if (result < 0)
22404274 1566 return result;
b60503ba
MW		 1567
1568 result = adapter_alloc_sq(dev, qid, nvmeq);
1569 if (result < 0)
1570 goto release_cq;
1571
3193f07b 1572 result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
b60503ba
MW		 1573 if (result < 0)
1574 goto release_sq;
1575
22404274 1576 nvme_init_queue(nvmeq, qid);
22404274 1577 return result;
b60503ba
MW		 1578
1579 release_sq:
1580 adapter_delete_sq(dev, qid);
1581 release_cq:
1582 adapter_delete_cq(dev, qid);
22404274 1583 return result;
b60503ba
MW		 1584}
1585
ba47e386
MW		 1586static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled)
1587{
1588 unsigned long timeout;
1589 u32 bit = enabled ? NVME_CSTS_RDY : 0;
1590
1591 timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1592
1593 while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) {
1594 msleep(100);
1595 if (fatal_signal_pending(current))
1596 return -EINTR;
1597 if (time_after(jiffies, timeout)) {
e75ec752 1598 dev_err(dev->dev,
27e8166c
MW		 1599 "Device not ready; aborting %s\n", enabled ?
1600 "initialisation" : "reset");
ba47e386
MW		 1601 return -ENODEV;
1602 }
1603 }
1604
1605 return 0;
1606}
1607
1608/*
1609 * If the device has been passed off to us in an enabled state, just clear
1610 * the enabled bit. The spec says we should set the 'shutdown notification
1611 * bits', but doing so may cause the device to complete commands to the
1612 * admin queue ... and we don't know what memory that might be pointing at!
1613 */
1614static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap)
1615{
01079522
DM		 1616 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1617 dev->ctrl_config &= ~NVME_CC_ENABLE;
1618 writel(dev->ctrl_config, &dev->bar->cc);
44af146a 1619
ba47e386
MW		 1620 return nvme_wait_ready(dev, cap, false);
1621}
1622
1623static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap)
1624{
01079522
DM		 1625 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1626 dev->ctrl_config |= NVME_CC_ENABLE;
1627 writel(dev->ctrl_config, &dev->bar->cc);
1628
ba47e386
MW		 1629 return nvme_wait_ready(dev, cap, true);
1630}
1631
1894d8f1
KB		 1632static int nvme_shutdown_ctrl(struct nvme_dev *dev)
1633{
1634 unsigned long timeout;
1894d8f1 1635
01079522
DM		 1636 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1637 dev->ctrl_config |= NVME_CC_SHN_NORMAL;
1638
1639 writel(dev->ctrl_config, &dev->bar->cc);
1894d8f1 1640
2484f407 1641 timeout = SHUTDOWN_TIMEOUT + jiffies;
1894d8f1
KB		 1642 while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) !=
1643 NVME_CSTS_SHST_CMPLT) {
1644 msleep(100);
1645 if (fatal_signal_pending(current))
1646 return -EINTR;
1647 if (time_after(jiffies, timeout)) {
e75ec752 1648 dev_err(dev->dev,
1894d8f1
KB		 1649 "Device shutdown incomplete; abort shutdown\n");
1650 return -ENODEV;
1651 }
1652 }
1653
1654 return 0;
1655}
1656
a4aea562 1657static struct blk_mq_ops nvme_mq_admin_ops = {
d29ec824 1658 .queue_rq = nvme_queue_rq,
a4aea562
MB		 1659 .map_queue = blk_mq_map_queue,
1660 .init_hctx = nvme_admin_init_hctx,
4af0e21c 1661 .exit_hctx = nvme_admin_exit_hctx,
a4aea562
MB		 1662 .init_request = nvme_admin_init_request,
1663 .timeout = nvme_timeout,
1664};
1665
1666static struct blk_mq_ops nvme_mq_ops = {
1667 .queue_rq = nvme_queue_rq,
1668 .map_queue = blk_mq_map_queue,
1669 .init_hctx = nvme_init_hctx,
1670 .init_request = nvme_init_request,
1671 .timeout = nvme_timeout,
1672};
1673
ea191d2f
KB		 1674static void nvme_dev_remove_admin(struct nvme_dev *dev)
1675{
1676 if (dev->admin_q && !blk_queue_dying(dev->admin_q)) {
1677 blk_cleanup_queue(dev->admin_q);
1678 blk_mq_free_tag_set(&dev->admin_tagset);
1679 }
1680}
1681
a4aea562
MB		 1682static int nvme_alloc_admin_tags(struct nvme_dev *dev)
1683{
1684 if (!dev->admin_q) {
1685 dev->admin_tagset.ops = &nvme_mq_admin_ops;
1686 dev->admin_tagset.nr_hw_queues = 1;
1687 dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1;
1efccc9d 1688 dev->admin_tagset.reserved_tags = 1;
a4aea562 1689 dev->admin_tagset.timeout = ADMIN_TIMEOUT;
e75ec752 1690 dev->admin_tagset.numa_node = dev_to_node(dev->dev);
ac3dd5bd 1691 dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
a4aea562
MB		 1692 dev->admin_tagset.driver_data = dev;
1693
1694 if (blk_mq_alloc_tag_set(&dev->admin_tagset))
1695 return -ENOMEM;
1696
1697 dev->admin_q = blk_mq_init_queue(&dev->admin_tagset);
35b489d3 1698 if (IS_ERR(dev->admin_q)) {
a4aea562
MB		 1699 blk_mq_free_tag_set(&dev->admin_tagset);
1700 return -ENOMEM;
1701 }
ea191d2f
KB		 1702 if (!blk_get_queue(dev->admin_q)) {
1703 nvme_dev_remove_admin(dev);
4af0e21c 1704 dev->admin_q = NULL;
ea191d2f
KB		 1705 return -ENODEV;
1706 }
0fb59cbc
KB		 1707 } else
1708 blk_mq_unfreeze_queue(dev->admin_q);
a4aea562
MB		 1709
1710 return 0;
1711}
1712
8d85fce7 1713static int nvme_configure_admin_queue(struct nvme_dev *dev)
b60503ba 1714{
ba47e386 1715 int result;
b60503ba 1716 u32 aqa;
ba47e386 1717 u64 cap = readq(&dev->bar->cap);
b60503ba 1718 struct nvme_queue *nvmeq;
1d090624
KB		 1719 unsigned page_shift = PAGE_SHIFT;
1720 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12;
1721 unsigned dev_page_max = NVME_CAP_MPSMAX(cap) + 12;
1722
1723 if (page_shift < dev_page_min) {
e75ec752 1724 dev_err(dev->dev,
1d090624
KB		 1725 "Minimum device page size (%u) too large for "
1726 "host (%u)\n", 1 << dev_page_min,
1727 1 << page_shift);
1728 return -ENODEV;
1729 }
1730 if (page_shift > dev_page_max) {
e75ec752 1731 dev_info(dev->dev,
1d090624
KB		 1732 "Device maximum page size (%u) smaller than "
1733 "host (%u); enabling work-around\n",
1734 1 << dev_page_max, 1 << page_shift);
1735 page_shift = dev_page_max;
1736 }
b60503ba 1737
dfbac8c7
KB		 1738 dev->subsystem = readl(&dev->bar->vs) >= NVME_VS(1, 1) ?
1739 NVME_CAP_NSSRC(cap) : 0;
1740
1741 if (dev->subsystem && (readl(&dev->bar->csts) & NVME_CSTS_NSSRO))
1742 writel(NVME_CSTS_NSSRO, &dev->bar->csts);
1743
ba47e386
MW		 1744 result = nvme_disable_ctrl(dev, cap);
1745 if (result < 0)
1746 return result;
b60503ba 1747
a4aea562 1748 nvmeq = dev->queues[0];
cd638946 1749 if (!nvmeq) {
2b25d981 1750 nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
cd638946
KB		 1751 if (!nvmeq)
1752 return -ENOMEM;
cd638946 1753 }
b60503ba
MW		 1754
1755 aqa = nvmeq->q_depth - 1;
1756 aqa |= aqa << 16;
1757
1d090624
KB		 1758 dev->page_size = 1 << page_shift;
1759
01079522 1760 dev->ctrl_config = NVME_CC_CSS_NVM;
1d090624 1761 dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
b60503ba 1762 dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
7f53f9d2 1763 dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
b60503ba
MW		 1764
1765 writel(aqa, &dev->bar->aqa);
1766 writeq(nvmeq->sq_dma_addr, &dev->bar->asq);
1767 writeq(nvmeq->cq_dma_addr, &dev->bar->acq);
b60503ba 1768
ba47e386 1769 result = nvme_enable_ctrl(dev, cap);
025c557a 1770 if (result)
a4aea562
MB		 1771 goto free_nvmeq;
1772
2b25d981 1773 nvmeq->cq_vector = 0;
3193f07b 1774 result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
758dd7fd
JD		 1775 if (result) {
1776 nvmeq->cq_vector = -1;
0fb59cbc 1777 goto free_nvmeq;
758dd7fd 1778 }
025c557a 1779
b60503ba 1780 return result;
a4aea562 1781
a4aea562
MB		 1782 free_nvmeq:
1783 nvme_free_queues(dev, 0);
1784 return result;
b60503ba
MW		 1785}
1786
a53295b6
MW		 1787static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1788{
1789 struct nvme_dev *dev = ns->dev;
a53295b6
MW		 1790 struct nvme_user_io io;
1791 struct nvme_command c;
d29ec824 1792 unsigned length, meta_len;
a67a9513 1793 int status, write;
a67a9513
KB		 1794 dma_addr_t meta_dma = 0;
1795 void *meta = NULL;
fec558b5 1796 void __user *metadata;
a53295b6
MW		 1797
1798 if (copy_from_user(&io, uio, sizeof(io)))
1799 return -EFAULT;
6c7d4945
MW		 1800
1801 switch (io.opcode) {
1802 case nvme_cmd_write:
1803 case nvme_cmd_read:
6bbf1acd 1804 case nvme_cmd_compare:
6413214c 1805 break;
6c7d4945 1806 default:
6bbf1acd 1807 return -EINVAL;
6c7d4945
MW		 1808 }
1809
d29ec824
CH		 1810 length = (io.nblocks + 1) << ns->lba_shift;
1811 meta_len = (io.nblocks + 1) * ns->ms;
6a398a3e 1812 metadata = (void __user *)(unsigned long)io.metadata;
d29ec824 1813 write = io.opcode & 1;
a53295b6 1814
71feb364
KB		 1815 if (ns->ext) {
1816 length += meta_len;
1817 meta_len = 0;
a67a9513
KB		 1818 }
1819 if (meta_len) {
d29ec824
CH		 1820 if (((io.metadata & 3) || !io.metadata) && !ns->ext)
1821 return -EINVAL;
1822
e75ec752 1823 meta = dma_alloc_coherent(dev->dev, meta_len,
a67a9513 1824 &meta_dma, GFP_KERNEL);
fec558b5 1825
a67a9513
KB		 1826 if (!meta) {
1827 status = -ENOMEM;
1828 goto unmap;
1829 }
1830 if (write) {
fec558b5 1831 if (copy_from_user(meta, metadata, meta_len)) {
a67a9513
KB		 1832 status = -EFAULT;
1833 goto unmap;
1834 }
1835 }
1836 }
1837
a53295b6
MW		 1838 memset(&c, 0, sizeof(c));
1839 c.rw.opcode = io.opcode;
1840 c.rw.flags = io.flags;
6c7d4945 1841 c.rw.nsid = cpu_to_le32(ns->ns_id);
a53295b6 1842 c.rw.slba = cpu_to_le64(io.slba);
6c7d4945 1843 c.rw.length = cpu_to_le16(io.nblocks);
a53295b6 1844 c.rw.control = cpu_to_le16(io.control);
1c9b5265
MW		 1845 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1846 c.rw.reftag = cpu_to_le32(io.reftag);
1847 c.rw.apptag = cpu_to_le16(io.apptag);
1848 c.rw.appmask = cpu_to_le16(io.appmask);
a67a9513 1849 c.rw.metadata = cpu_to_le64(meta_dma);
d29ec824
CH		 1850
1851 status = __nvme_submit_sync_cmd(ns->queue, &c, NULL,
1852 (void __user *)io.addr, length, NULL, 0);
f410c680 1853 unmap:
a67a9513
KB		 1854 if (meta) {
1855 if (status == NVME_SC_SUCCESS && !write) {
fec558b5 1856 if (copy_to_user(metadata, meta, meta_len))
a67a9513
KB		 1857 status = -EFAULT;
1858 }
e75ec752 1859 dma_free_coherent(dev->dev, meta_len, meta, meta_dma);
f410c680 1860 }
a53295b6
MW		 1861 return status;
1862}
1863
a4aea562
MB		 1864static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns,
1865 struct nvme_passthru_cmd __user *ucmd)
6ee44cdc 1866{
7963e521 1867 struct nvme_passthru_cmd cmd;
6ee44cdc 1868 struct nvme_command c;
d29ec824
CH		 1869 unsigned timeout = 0;
1870 int status;
6ee44cdc 1871
6bbf1acd
MW		 1872 if (!capable(CAP_SYS_ADMIN))
1873 return -EACCES;
1874 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
6ee44cdc 1875 return -EFAULT;
6ee44cdc
MW		 1876
1877 memset(&c, 0, sizeof(c));
6bbf1acd
MW		 1878 c.common.opcode = cmd.opcode;
1879 c.common.flags = cmd.flags;
1880 c.common.nsid = cpu_to_le32(cmd.nsid);
1881 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1882 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1883 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1884 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1885 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1886 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1887 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1888 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1889
d29ec824
CH		 1890 if (cmd.timeout_ms)
1891 timeout = msecs_to_jiffies(cmd.timeout_ms);
eca18b23 1892
f705f837 1893 status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c,
d29ec824
CH		 1894 NULL, (void __user *)cmd.addr, cmd.data_len,
1895 &cmd.result, timeout);
1896 if (status >= 0) {
1897 if (put_user(cmd.result, &ucmd->result))
1898 return -EFAULT;
6bbf1acd 1899 }
f4f117f6 1900
6ee44cdc
MW		 1901 return status;
1902}
1903
81f03fed
JD		 1904static int nvme_subsys_reset(struct nvme_dev *dev)
1905{
1906 if (!dev->subsystem)
1907 return -ENOTTY;
1908
1909 writel(0x4E564D65, &dev->bar->nssr); /* "NVMe" */
1910 return 0;
1911}
1912
b60503ba
MW		 1913static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd,
1914 unsigned long arg)
1915{
1916 struct nvme_ns *ns = bdev->bd_disk->private_data;
1917
1918 switch (cmd) {
6bbf1acd 1919 case NVME_IOCTL_ID:
c3bfe717 1920 force_successful_syscall_return();
6bbf1acd
MW		 1921 return ns->ns_id;
1922 case NVME_IOCTL_ADMIN_CMD:
a4aea562 1923 return nvme_user_cmd(ns->dev, NULL, (void __user *)arg);
7963e521 1924 case NVME_IOCTL_IO_CMD:
a4aea562 1925 return nvme_user_cmd(ns->dev, ns, (void __user *)arg);
a53295b6
MW		 1926 case NVME_IOCTL_SUBMIT_IO:
1927 return nvme_submit_io(ns, (void __user *)arg);
5d0f6131
VV		 1928 case SG_GET_VERSION_NUM:
1929 return nvme_sg_get_version_num((void __user *)arg);
1930 case SG_IO:
1931 return nvme_sg_io(ns, (void __user *)arg);
b60503ba
MW		 1932 default:
1933 return -ENOTTY;
1934 }
1935}
1936
320a3827
KB		 1937#ifdef CONFIG_COMPAT
1938static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1939 unsigned int cmd, unsigned long arg)
1940{
320a3827
KB		 1941 switch (cmd) {
1942 case SG_IO:
e179729a 1943 return -ENOIOCTLCMD;
320a3827
KB		 1944 }
1945 return nvme_ioctl(bdev, mode, cmd, arg);
1946}
1947#else
1948#define nvme_compat_ioctl NULL
1949#endif
1950
9ac27090
KB		 1951static int nvme_open(struct block_device *bdev, fmode_t mode)
1952{
9e60352c
KB		 1953 int ret = 0;
1954 struct nvme_ns *ns;
9ac27090 1955
9e60352c
KB		 1956 spin_lock(&dev_list_lock);
1957 ns = bdev->bd_disk->private_data;
1958 if (!ns)
1959 ret = -ENXIO;
1960 else if (!kref_get_unless_zero(&ns->dev->kref))
1961 ret = -ENXIO;
1962 spin_unlock(&dev_list_lock);
1963
1964 return ret;
9ac27090
KB		 1965}
1966
1967static void nvme_free_dev(struct kref *kref);
1968
1969static void nvme_release(struct gendisk *disk, fmode_t mode)
1970{
1971 struct nvme_ns *ns = disk->private_data;
1972 struct nvme_dev *dev = ns->dev;
1973
1974 kref_put(&dev->kref, nvme_free_dev);
1975}
1976
4cc09e2d
KB		 1977static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo)
1978{
1979 /* some standard values */
1980 geo->heads = 1 << 6;
1981 geo->sectors = 1 << 5;
1982 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1983 return 0;
1984}
1985
e1e5e564
KB		 1986static void nvme_config_discard(struct nvme_ns *ns)
1987{
1988 u32 logical_block_size = queue_logical_block_size(ns->queue);
1989 ns->queue->limits.discard_zeroes_data = 0;
1990 ns->queue->limits.discard_alignment = logical_block_size;
1991 ns->queue->limits.discard_granularity = logical_block_size;
2bb4cd5c 1992 blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
e1e5e564
KB		 1993 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1994}
1995
1b9dbf7f
KB		 1996static int nvme_revalidate_disk(struct gendisk *disk)
1997{
1998 struct nvme_ns *ns = disk->private_data;
1999 struct nvme_dev *dev = ns->dev;
2000 struct nvme_id_ns *id;
a67a9513
KB		 2001 u8 lbaf, pi_type;
2002 u16 old_ms;
e1e5e564 2003 unsigned short bs;
1b9dbf7f 2004
d29ec824 2005 if (nvme_identify_ns(dev, ns->ns_id, &id)) {
a5768aa8
KB		 2006 dev_warn(dev->dev, "%s: Identify failure nvme%dn%d\n", __func__,
2007 dev->instance, ns->ns_id);
2008 return -ENODEV;
1b9dbf7f 2009 }
a5768aa8
KB		 2010 if (id->ncap == 0) {
2011 kfree(id);
2012 return -ENODEV;
e1e5e564 2013 }
1b9dbf7f 2014
e1e5e564
KB		 2015 old_ms = ns->ms;
2016 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1b9dbf7f 2017 ns->lba_shift = id->lbaf[lbaf].ds;
e1e5e564 2018 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
a67a9513 2019 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
e1e5e564
KB		 2020
2021 /*
2022 * If identify namespace failed, use default 512 byte block size so
2023 * block layer can use before failing read/write for 0 capacity.
2024 */
2025 if (ns->lba_shift == 0)
2026 ns->lba_shift = 9;
2027 bs = 1 << ns->lba_shift;
2028
2029 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
2030 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
2031 id->dps & NVME_NS_DPS_PI_MASK : 0;
2032
52b68d7e
KB		 2033 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
2034 ns->ms != old_ms ||
e1e5e564 2035 bs != queue_logical_block_size(disk->queue) ||
a67a9513 2036 (ns->ms && ns->ext)))
e1e5e564
KB		 2037 blk_integrity_unregister(disk);
2038
2039 ns->pi_type = pi_type;
2040 blk_queue_logical_block_size(ns->queue, bs);
2041
52b68d7e 2042 if (ns->ms && !blk_get_integrity(disk) && (disk->flags & GENHD_FL_UP) &&
a67a9513 2043 !ns->ext)
e1e5e564
KB		 2044 nvme_init_integrity(ns);
2045
a5768aa8 2046 if (ns->ms && !blk_get_integrity(disk))
e1e5e564
KB		 2047 set_capacity(disk, 0);
2048 else
2049 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
2050
2051 if (dev->oncs & NVME_CTRL_ONCS_DSM)
2052 nvme_config_discard(ns);
1b9dbf7f 2053
d29ec824 2054 kfree(id);
1b9dbf7f
KB		 2055 return 0;
2056}
2057
b60503ba
MW		 2058static const struct block_device_operations nvme_fops = {
2059 .owner = THIS_MODULE,
2060 .ioctl = nvme_ioctl,
320a3827 2061 .compat_ioctl = nvme_compat_ioctl,
9ac27090
KB		 2062 .open = nvme_open,
2063 .release = nvme_release,
4cc09e2d 2064 .getgeo = nvme_getgeo,
1b9dbf7f 2065 .revalidate_disk= nvme_revalidate_disk,
b60503ba
MW		 2066};
2067
1fa6aead
MW		 2068static int nvme_kthread(void *data)
2069{
d4b4ff8e 2070 struct nvme_dev *dev, *next;
1fa6aead
MW		 2071
2072 while (!kthread_should_stop()) {
564a232c 2073 set_current_state(TASK_INTERRUPTIBLE);
1fa6aead 2074 spin_lock(&dev_list_lock);
d4b4ff8e 2075 list_for_each_entry_safe(dev, next, &dev_list, node) {
1fa6aead 2076 int i;
dfbac8c7
KB		 2077 u32 csts = readl(&dev->bar->csts);
2078
2079 if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
2080 csts & NVME_CSTS_CFS) {
d4b4ff8e
KB		 2081 if (work_busy(&dev->reset_work))
2082 continue;
2083 list_del_init(&dev->node);
e75ec752 2084 dev_warn(dev->dev,
a4aea562
MB		 2085 "Failed status: %x, reset controller\n",
2086 readl(&dev->bar->csts));
9ca97374 2087 dev->reset_workfn = nvme_reset_failed_dev;
d4b4ff8e
KB		 2088 queue_work(nvme_workq, &dev->reset_work);
2089 continue;
2090 }
1fa6aead 2091 for (i = 0; i < dev->queue_count; i++) {
a4aea562 2092 struct nvme_queue *nvmeq = dev->queues[i];
740216fc
MW		 2093 if (!nvmeq)
2094 continue;
1fa6aead 2095 spin_lock_irq(&nvmeq->q_lock);
bc57a0f7 2096 nvme_process_cq(nvmeq);
6fccf938
KB		 2097
2098 while ((i == 0) && (dev->event_limit > 0)) {
a4aea562 2099 if (nvme_submit_async_admin_req(dev))
6fccf938
KB		 2100 break;
2101 dev->event_limit--;
2102 }
1fa6aead
MW		 2103 spin_unlock_irq(&nvmeq->q_lock);
2104 }
2105 }
2106 spin_unlock(&dev_list_lock);
acb7aa0d 2107 schedule_timeout(round_jiffies_relative(HZ));
1fa6aead
MW		 2108 }
2109 return 0;
2110}
2111
e1e5e564 2112static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
b60503ba
MW		 2113{
2114 struct nvme_ns *ns;
2115 struct gendisk *disk;
e75ec752 2116 int node = dev_to_node(dev->dev);
b60503ba 2117
a4aea562 2118 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
b60503ba 2119 if (!ns)
e1e5e564
KB		 2120 return;
2121
a4aea562 2122 ns->queue = blk_mq_init_queue(&dev->tagset);
9f173b33 2123 if (IS_ERR(ns->queue))
b60503ba 2124 goto out_free_ns;
4eeb9215
MW		 2125 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
2126 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
a4aea562 2127 queue_flag_set_unlocked(QUEUE_FLAG_SG_GAPS, ns->queue);
b60503ba
MW		 2128 ns->dev = dev;
2129 ns->queue->queuedata = ns;
2130
a4aea562 2131 disk = alloc_disk_node(0, node);
b60503ba
MW		 2132 if (!disk)
2133 goto out_free_queue;
a4aea562 2134
5aff9382 2135 ns->ns_id = nsid;
b60503ba 2136 ns->disk = disk;
e1e5e564
KB		 2137 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2138 list_add_tail(&ns->list, &dev->namespaces);
2139
e9ef4636 2140 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
e824410f 2141 if (dev->max_hw_sectors) {
8fc23e03 2142 blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors);
e824410f
KB		 2143 blk_queue_max_segments(ns->queue,
2144 ((dev->max_hw_sectors << 9) / dev->page_size) + 1);
2145 }
a4aea562
MB		 2146 if (dev->stripe_size)
2147 blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9);
a7d2ce28
KB		 2148 if (dev->vwc & NVME_CTRL_VWC_PRESENT)
2149 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
b60503ba
MW		 2150
2151 disk->major = nvme_major;
469071a3 2152 disk->first_minor = 0;
b60503ba
MW		 2153 disk->fops = &nvme_fops;
2154 disk->private_data = ns;
2155 disk->queue = ns->queue;
b3fffdef 2156 disk->driverfs_dev = dev->device;
469071a3 2157 disk->flags = GENHD_FL_EXT_DEVT;
5aff9382 2158 sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid);
b60503ba 2159
e1e5e564
KB		 2160 /*
2161 * Initialize capacity to 0 until we establish the namespace format and
2162 * setup integrity extentions if necessary. The revalidate_disk after
2163 * add_disk allows the driver to register with integrity if the format
2164 * requires it.
2165 */
2166 set_capacity(disk, 0);
a5768aa8
KB		 2167 if (nvme_revalidate_disk(ns->disk))
2168 goto out_free_disk;
2169
e1e5e564 2170 add_disk(ns->disk);
7bee6074
KB		 2171 if (ns->ms) {
2172 struct block_device *bd = bdget_disk(ns->disk, 0);
2173 if (!bd)
2174 return;
2175 if (blkdev_get(bd, FMODE_READ, NULL)) {
2176 bdput(bd);
2177 return;
2178 }
2179 blkdev_reread_part(bd);
2180 blkdev_put(bd, FMODE_READ);
2181 }
e1e5e564 2182 return;
a5768aa8
KB		 2183 out_free_disk:
2184 kfree(disk);
2185 list_del(&ns->list);
b60503ba
MW		 2186 out_free_queue:
2187 blk_cleanup_queue(ns->queue);
2188 out_free_ns:
2189 kfree(ns);
b60503ba
MW		 2190}
2191
42f61420
KB		 2192static void nvme_create_io_queues(struct nvme_dev *dev)
2193{
a4aea562 2194 unsigned i;
42f61420 2195
a4aea562 2196 for (i = dev->queue_count; i <= dev->max_qid; i++)
2b25d981 2197 if (!nvme_alloc_queue(dev, i, dev->q_depth))
42f61420
KB		 2198 break;
2199
a4aea562
MB		 2200 for (i = dev->online_queues; i <= dev->queue_count - 1; i++)
2201 if (nvme_create_queue(dev->queues[i], i))
42f61420
KB		 2202 break;
2203}
2204
b3b06812 2205static int set_queue_count(struct nvme_dev *dev, int count)
b60503ba
MW		 2206{
2207 int status;
2208 u32 result;
b3b06812 2209 u32 q_count = (count - 1) | ((count - 1) << 16);
b60503ba 2210
df348139 2211 status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0,
bc5fc7e4 2212 &result);
27e8166c
MW		 2213 if (status < 0)
2214 return status;
2215 if (status > 0) {
e75ec752 2216 dev_err(dev->dev, "Could not set queue count (%d)\n", status);
badc34d4 2217 return 0;
27e8166c 2218 }
b60503ba
MW		 2219 return min(result & 0xffff, result >> 16) + 1;
2220}
2221
8ffaadf7
JD		 2222static void __iomem *nvme_map_cmb(struct nvme_dev *dev)
2223{
2224 u64 szu, size, offset;
2225 u32 cmbloc;
2226 resource_size_t bar_size;
2227 struct pci_dev *pdev = to_pci_dev(dev->dev);
2228 void __iomem *cmb;
2229 dma_addr_t dma_addr;
2230
2231 if (!use_cmb_sqes)
2232 return NULL;
2233
2234 dev->cmbsz = readl(&dev->bar->cmbsz);
2235 if (!(NVME_CMB_SZ(dev->cmbsz)))
2236 return NULL;
2237
2238 cmbloc = readl(&dev->bar->cmbloc);
2239
2240 szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
2241 size = szu * NVME_CMB_SZ(dev->cmbsz);
2242 offset = szu * NVME_CMB_OFST(cmbloc);
2243 bar_size = pci_resource_len(pdev, NVME_CMB_BIR(cmbloc));
2244
2245 if (offset > bar_size)
2246 return NULL;
2247
2248 /*
2249 * Controllers may support a CMB size larger than their BAR,
2250 * for example, due to being behind a bridge. Reduce the CMB to
2251 * the reported size of the BAR
2252 */
2253 if (size > bar_size - offset)
2254 size = bar_size - offset;
2255
2256 dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(cmbloc)) + offset;
2257 cmb = ioremap_wc(dma_addr, size);
2258 if (!cmb)
2259 return NULL;
2260
2261 dev->cmb_dma_addr = dma_addr;
2262 dev->cmb_size = size;
2263 return cmb;
2264}
2265
2266static inline void nvme_release_cmb(struct nvme_dev *dev)
2267{
2268 if (dev->cmb) {
2269 iounmap(dev->cmb);
2270 dev->cmb = NULL;
2271 }
2272}
2273
9d713c2b
KB		 2274static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
2275{
b80d5ccc 2276 return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
9d713c2b
KB		 2277}
2278
8d85fce7 2279static int nvme_setup_io_queues(struct nvme_dev *dev)
b60503ba 2280{
a4aea562 2281 struct nvme_queue *adminq = dev->queues[0];
e75ec752 2282 struct pci_dev *pdev = to_pci_dev(dev->dev);
42f61420 2283 int result, i, vecs, nr_io_queues, size;
b60503ba 2284
42f61420 2285 nr_io_queues = num_possible_cpus();
b348b7d5 2286 result = set_queue_count(dev, nr_io_queues);
badc34d4 2287 if (result <= 0)
1b23484b 2288 return result;
b348b7d5
MW		 2289 if (result < nr_io_queues)
2290 nr_io_queues = result;
b60503ba 2291
8ffaadf7
JD		 2292 if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) {
2293 result = nvme_cmb_qdepth(dev, nr_io_queues,
2294 sizeof(struct nvme_command));
2295 if (result > 0)
2296 dev->q_depth = result;
2297 else
2298 nvme_release_cmb(dev);
2299 }
2300
9d713c2b
KB		 2301 size = db_bar_size(dev, nr_io_queues);
2302 if (size > 8192) {
f1938f6e 2303 iounmap(dev->bar);
9d713c2b
KB		 2304 do {
2305 dev->bar = ioremap(pci_resource_start(pdev, 0), size);
2306 if (dev->bar)
2307 break;
2308 if (!--nr_io_queues)
2309 return -ENOMEM;
2310 size = db_bar_size(dev, nr_io_queues);
2311 } while (1);
f1938f6e 2312 dev->dbs = ((void __iomem *)dev->bar) + 4096;
5a92e700 2313 adminq->q_db = dev->dbs;
f1938f6e
MW		 2314 }
2315
9d713c2b 2316 /* Deregister the admin queue's interrupt */
3193f07b 2317 free_irq(dev->entry[0].vector, adminq);
9d713c2b 2318
e32efbfc
JA		 2319 /*
2320 * If we enable msix early due to not intx, disable it again before
2321 * setting up the full range we need.
2322 */
2323 if (!pdev->irq)
2324 pci_disable_msix(pdev);
2325
be577fab 2326 for (i = 0; i < nr_io_queues; i++)
1b23484b 2327 dev->entry[i].entry = i;
be577fab
AG		 2328 vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues);
2329 if (vecs < 0) {
2330 vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32));
2331 if (vecs < 0) {
2332 vecs = 1;
2333 } else {
2334 for (i = 0; i < vecs; i++)
2335 dev->entry[i].vector = i + pdev->irq;
fa08a396
RRG		 2336 }
2337 }
2338
063a8096
MW		 2339 /*
2340 * Should investigate if there's a performance win from allocating
2341 * more queues than interrupt vectors; it might allow the submission
2342 * path to scale better, even if the receive path is limited by the
2343 * number of interrupts.
2344 */
2345 nr_io_queues = vecs;
42f61420 2346 dev->max_qid = nr_io_queues;
063a8096 2347
3193f07b 2348 result = queue_request_irq(dev, adminq, adminq->irqname);
758dd7fd
JD		 2349 if (result) {
2350 adminq->cq_vector = -1;
22404274 2351 goto free_queues;
758dd7fd 2352 }
1b23484b 2353
cd638946 2354 /* Free previously allocated queues that are no longer usable */
42f61420 2355 nvme_free_queues(dev, nr_io_queues + 1);
a4aea562 2356 nvme_create_io_queues(dev);
9ecdc946 2357
22404274 2358 return 0;
b60503ba 2359
22404274 2360 free_queues:
a1a5ef99 2361 nvme_free_queues(dev, 1);
22404274 2362 return result;
b60503ba
MW		 2363}
2364
a5768aa8
KB		 2365static void nvme_free_namespace(struct nvme_ns *ns)
2366{
2367 list_del(&ns->list);
2368
2369 spin_lock(&dev_list_lock);
2370 ns->disk->private_data = NULL;
2371 spin_unlock(&dev_list_lock);
2372
2373 put_disk(ns->disk);
2374 kfree(ns);
2375}
2376
2377static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2378{
2379 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2380 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2381
2382 return nsa->ns_id - nsb->ns_id;
2383}
2384
2385static struct nvme_ns *nvme_find_ns(struct nvme_dev *dev, unsigned nsid)
2386{
2387 struct nvme_ns *ns;
2388
2389 list_for_each_entry(ns, &dev->namespaces, list) {
2390 if (ns->ns_id == nsid)
2391 return ns;
2392 if (ns->ns_id > nsid)
2393 break;
2394 }
2395 return NULL;
2396}
2397
2398static inline bool nvme_io_incapable(struct nvme_dev *dev)
2399{
2400 return (!dev->bar || readl(&dev->bar->csts) & NVME_CSTS_CFS ||
2401 dev->online_queues < 2);
2402}
2403
2404static void nvme_ns_remove(struct nvme_ns *ns)
2405{
2406 bool kill = nvme_io_incapable(ns->dev) && !blk_queue_dying(ns->queue);
2407
2408 if (kill)
2409 blk_set_queue_dying(ns->queue);
2410 if (ns->disk->flags & GENHD_FL_UP) {
2411 if (blk_get_integrity(ns->disk))
2412 blk_integrity_unregister(ns->disk);
2413 del_gendisk(ns->disk);
2414 }
2415 if (kill || !blk_queue_dying(ns->queue)) {
2416 blk_mq_abort_requeue_list(ns->queue);
2417 blk_cleanup_queue(ns->queue);
2418 }
2419}
2420
2421static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn)
2422{
2423 struct nvme_ns *ns, *next;
2424 unsigned i;
2425
2426 for (i = 1; i <= nn; i++) {
2427 ns = nvme_find_ns(dev, i);
2428 if (ns) {
2429 if (revalidate_disk(ns->disk)) {
2430 nvme_ns_remove(ns);
2431 nvme_free_namespace(ns);
2432 }
2433 } else
2434 nvme_alloc_ns(dev, i);
2435 }
2436 list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
2437 if (ns->ns_id > nn) {
2438 nvme_ns_remove(ns);
2439 nvme_free_namespace(ns);
2440 }
2441 }
2442 list_sort(NULL, &dev->namespaces, ns_cmp);
2443}
2444
2445static void nvme_dev_scan(struct work_struct *work)
2446{
2447 struct nvme_dev *dev = container_of(work, struct nvme_dev, scan_work);
2448 struct nvme_id_ctrl *ctrl;
2449
2450 if (!dev->tagset.tags)
2451 return;
2452 if (nvme_identify_ctrl(dev, &ctrl))
2453 return;
2454 nvme_scan_namespaces(dev, le32_to_cpup(&ctrl->nn));
2455 kfree(ctrl);
2456}
2457
422ef0c7
MW		 2458/*
2459 * Return: error value if an error occurred setting up the queues or calling
2460 * Identify Device. 0 if these succeeded, even if adding some of the
2461 * namespaces failed. At the moment, these failures are silent. TBD which
2462 * failures should be reported.
2463 */
8d85fce7 2464static int nvme_dev_add(struct nvme_dev *dev)
b60503ba 2465{
e75ec752 2466 struct pci_dev *pdev = to_pci_dev(dev->dev);
c3bfe717 2467 int res;
51814232 2468 struct nvme_id_ctrl *ctrl;
159b67d7 2469 int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12;
b60503ba 2470
d29ec824 2471 res = nvme_identify_ctrl(dev, &ctrl);
b60503ba 2472 if (res) {
e75ec752 2473 dev_err(dev->dev, "Identify Controller failed (%d)\n", res);
e1e5e564 2474 return -EIO;
b60503ba
MW		 2475 }
2476
0e5e4f0e 2477 dev->oncs = le16_to_cpup(&ctrl->oncs);
c30341dc 2478 dev->abort_limit = ctrl->acl + 1;
a7d2ce28 2479 dev->vwc = ctrl->vwc;
51814232
MW		 2480 memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn));
2481 memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn));
2482 memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
159b67d7 2483 if (ctrl->mdts)
8fc23e03 2484 dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9);
68608c26 2485 if ((pdev->vendor == PCI_VENDOR_ID_INTEL) &&
a4aea562
MB		 2486 (pdev->device == 0x0953) && ctrl->vs[3]) {
2487 unsigned int max_hw_sectors;
2488
159b67d7 2489 dev->stripe_size = 1 << (ctrl->vs[3] + shift);
a4aea562
MB		 2490 max_hw_sectors = dev->stripe_size >> (shift - 9);
2491 if (dev->max_hw_sectors) {
2492 dev->max_hw_sectors = min(max_hw_sectors,
2493 dev->max_hw_sectors);
2494 } else
2495 dev->max_hw_sectors = max_hw_sectors;
2496 }
d29ec824 2497 kfree(ctrl);
a4aea562 2498
ffe7704d
KB		 2499 if (!dev->tagset.tags) {
2500 dev->tagset.ops = &nvme_mq_ops;
2501 dev->tagset.nr_hw_queues = dev->online_queues - 1;
2502 dev->tagset.timeout = NVME_IO_TIMEOUT;
2503 dev->tagset.numa_node = dev_to_node(dev->dev);
2504 dev->tagset.queue_depth =
a4aea562 2505 min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
ffe7704d
KB		 2506 dev->tagset.cmd_size = nvme_cmd_size(dev);
2507 dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
2508 dev->tagset.driver_data = dev;
b60503ba 2509
ffe7704d
KB		 2510 if (blk_mq_alloc_tag_set(&dev->tagset))
2511 return 0;
2512 }
a5768aa8 2513 schedule_work(&dev->scan_work);
e1e5e564 2514 return 0;
b60503ba
MW		 2515}
2516
0877cb0d
KB		 2517static int nvme_dev_map(struct nvme_dev *dev)
2518{
42f61420 2519 u64 cap;
0877cb0d 2520 int bars, result = -ENOMEM;
e75ec752 2521 struct pci_dev *pdev = to_pci_dev(dev->dev);
0877cb0d
KB		 2522
2523 if (pci_enable_device_mem(pdev))
2524 return result;
2525
2526 dev->entry[0].vector = pdev->irq;
2527 pci_set_master(pdev);
2528 bars = pci_select_bars(pdev, IORESOURCE_MEM);
be7837e8
JA		 2529 if (!bars)
2530 goto disable_pci;
2531
0877cb0d
KB		 2532 if (pci_request_selected_regions(pdev, bars, "nvme"))
2533 goto disable_pci;
2534
e75ec752
CH		 2535 if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
2536 dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
052d0efa 2537 goto disable;
0877cb0d 2538
0877cb0d
KB		 2539 dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
2540 if (!dev->bar)
2541 goto disable;
e32efbfc 2542
0e53d180
KB		 2543 if (readl(&dev->bar->csts) == -1) {
2544 result = -ENODEV;
2545 goto unmap;
2546 }
e32efbfc
JA		 2547
2548 /*
2549 * Some devices don't advertse INTx interrupts, pre-enable a single
2550 * MSIX vec for setup. We'll adjust this later.
2551 */
2552 if (!pdev->irq) {
2553 result = pci_enable_msix(pdev, dev->entry, 1);
2554 if (result < 0)
2555 goto unmap;
2556 }
2557
42f61420
KB		 2558 cap = readq(&dev->bar->cap);
2559 dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
2560 dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
0877cb0d 2561 dev->dbs = ((void __iomem *)dev->bar) + 4096;
8ffaadf7
JD		 2562 if (readl(&dev->bar->vs) >= NVME_VS(1, 2))
2563 dev->cmb = nvme_map_cmb(dev);
0877cb0d
KB		 2564
2565 return 0;
2566
0e53d180
KB		 2567 unmap:
2568 iounmap(dev->bar);
2569 dev->bar = NULL;
0877cb0d
KB		 2570 disable:
2571 pci_release_regions(pdev);
2572 disable_pci:
2573 pci_disable_device(pdev);
2574 return result;
2575}
2576
2577static void nvme_dev_unmap(struct nvme_dev *dev)
2578{
e75ec752
CH		 2579 struct pci_dev *pdev = to_pci_dev(dev->dev);
2580
2581 if (pdev->msi_enabled)
2582 pci_disable_msi(pdev);
2583 else if (pdev->msix_enabled)
2584 pci_disable_msix(pdev);
0877cb0d
KB		 2585
2586 if (dev->bar) {
2587 iounmap(dev->bar);
2588 dev->bar = NULL;
e75ec752 2589 pci_release_regions(pdev);
0877cb0d
KB		 2590 }
2591
e75ec752
CH		 2592 if (pci_is_enabled(pdev))
2593 pci_disable_device(pdev);
0877cb0d
KB		 2594}
2595
4d115420
KB		 2596struct nvme_delq_ctx {
2597 struct task_struct *waiter;
2598 struct kthread_worker *worker;
2599 atomic_t refcount;
2600};
2601
2602static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev)
2603{
2604 dq->waiter = current;
2605 mb();
2606
2607 for (;;) {
2608 set_current_state(TASK_KILLABLE);
2609 if (!atomic_read(&dq->refcount))
2610 break;
2611 if (!schedule_timeout(ADMIN_TIMEOUT) ||
2612 fatal_signal_pending(current)) {
0fb59cbc
KB		 2613 /*
2614 * Disable the controller first since we can't trust it
2615 * at this point, but leave the admin queue enabled
2616 * until all queue deletion requests are flushed.
2617 * FIXME: This may take a while if there are more h/w
2618 * queues than admin tags.
2619 */
4d115420 2620 set_current_state(TASK_RUNNING);
4d115420 2621 nvme_disable_ctrl(dev, readq(&dev->bar->cap));
0fb59cbc 2622 nvme_clear_queue(dev->queues[0]);
4d115420 2623 flush_kthread_worker(dq->worker);
0fb59cbc 2624 nvme_disable_queue(dev, 0);
4d115420
KB		 2625 return;
2626 }
2627 }
2628 set_current_state(TASK_RUNNING);
2629}
2630
2631static void nvme_put_dq(struct nvme_delq_ctx *dq)
2632{
2633 atomic_dec(&dq->refcount);
2634 if (dq->waiter)
2635 wake_up_process(dq->waiter);
2636}
2637
2638static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq)
2639{
2640 atomic_inc(&dq->refcount);
2641 return dq;
2642}
2643
2644static void nvme_del_queue_end(struct nvme_queue *nvmeq)
2645{
2646 struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx;
4d115420
KB		 2647 nvme_put_dq(dq);
2648}
2649
2650static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode,
2651 kthread_work_func_t fn)
2652{
2653 struct nvme_command c;
2654
2655 memset(&c, 0, sizeof(c));
2656 c.delete_queue.opcode = opcode;
2657 c.delete_queue.qid = cpu_to_le16(nvmeq->qid);
2658
2659 init_kthread_work(&nvmeq->cmdinfo.work, fn);
a4aea562
MB		 2660 return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo,
2661 ADMIN_TIMEOUT);
4d115420
KB		 2662}
2663
2664static void nvme_del_cq_work_handler(struct kthread_work *work)
2665{
2666 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2667 cmdinfo.work);
2668 nvme_del_queue_end(nvmeq);
2669}
2670
2671static int nvme_delete_cq(struct nvme_queue *nvmeq)
2672{
2673 return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq,
2674 nvme_del_cq_work_handler);
2675}
2676
2677static void nvme_del_sq_work_handler(struct kthread_work *work)
2678{
2679 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2680 cmdinfo.work);
2681 int status = nvmeq->cmdinfo.status;
2682
2683 if (!status)
2684 status = nvme_delete_cq(nvmeq);
2685 if (status)
2686 nvme_del_queue_end(nvmeq);
2687}
2688
2689static int nvme_delete_sq(struct nvme_queue *nvmeq)
2690{
2691 return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq,
2692 nvme_del_sq_work_handler);
2693}
2694
2695static void nvme_del_queue_start(struct kthread_work *work)
2696{
2697 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2698 cmdinfo.work);
4d115420
KB		 2699 if (nvme_delete_sq(nvmeq))
2700 nvme_del_queue_end(nvmeq);
2701}
2702
2703static void nvme_disable_io_queues(struct nvme_dev *dev)
2704{
2705 int i;
2706 DEFINE_KTHREAD_WORKER_ONSTACK(worker);
2707 struct nvme_delq_ctx dq;
2708 struct task_struct *kworker_task = kthread_run(kthread_worker_fn,
2709 &worker, "nvme%d", dev->instance);
2710
2711 if (IS_ERR(kworker_task)) {
e75ec752 2712 dev_err(dev->dev,
4d115420
KB		 2713 "Failed to create queue del task\n");
2714 for (i = dev->queue_count - 1; i > 0; i--)
2715 nvme_disable_queue(dev, i);
2716 return;
2717 }
2718
2719 dq.waiter = NULL;
2720 atomic_set(&dq.refcount, 0);
2721 dq.worker = &worker;
2722 for (i = dev->queue_count - 1; i > 0; i--) {
a4aea562 2723 struct nvme_queue *nvmeq = dev->queues[i];
4d115420
KB		 2724
2725 if (nvme_suspend_queue(nvmeq))
2726 continue;
2727 nvmeq->cmdinfo.ctx = nvme_get_dq(&dq);
2728 nvmeq->cmdinfo.worker = dq.worker;
2729 init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start);
2730 queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work);
2731 }
2732 nvme_wait_dq(&dq, dev);
2733 kthread_stop(kworker_task);
2734}
2735
b9afca3e
DM		 2736/*
2737* Remove the node from the device list and check
2738* for whether or not we need to stop the nvme_thread.
2739*/
2740static void nvme_dev_list_remove(struct nvme_dev *dev)
2741{
2742 struct task_struct *tmp = NULL;
2743
2744 spin_lock(&dev_list_lock);
2745 list_del_init(&dev->node);
2746 if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) {
2747 tmp = nvme_thread;
2748 nvme_thread = NULL;
2749 }
2750 spin_unlock(&dev_list_lock);
2751
2752 if (tmp)
2753 kthread_stop(tmp);
2754}
2755
c9d3bf88
KB		 2756static void nvme_freeze_queues(struct nvme_dev *dev)
2757{
2758 struct nvme_ns *ns;
2759
2760 list_for_each_entry(ns, &dev->namespaces, list) {
2761 blk_mq_freeze_queue_start(ns->queue);
2762
cddcd72b 2763 spin_lock_irq(ns->queue->queue_lock);
c9d3bf88 2764 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
cddcd72b 2765 spin_unlock_irq(ns->queue->queue_lock);
c9d3bf88
KB		 2766
2767 blk_mq_cancel_requeue_work(ns->queue);
2768 blk_mq_stop_hw_queues(ns->queue);
2769 }
2770}
2771
2772static void nvme_unfreeze_queues(struct nvme_dev *dev)
2773{
2774 struct nvme_ns *ns;
2775
2776 list_for_each_entry(ns, &dev->namespaces, list) {
2777 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2778 blk_mq_unfreeze_queue(ns->queue);
2779 blk_mq_start_stopped_hw_queues(ns->queue, true);
2780 blk_mq_kick_requeue_list(ns->queue);
2781 }
2782}
2783
f0b50732 2784static void nvme_dev_shutdown(struct nvme_dev *dev)
b60503ba 2785{
22404274 2786 int i;
7c1b2450 2787 u32 csts = -1;
22404274 2788
b9afca3e 2789 nvme_dev_list_remove(dev);
1fa6aead 2790
c9d3bf88
KB		 2791 if (dev->bar) {
2792 nvme_freeze_queues(dev);
7c1b2450 2793 csts = readl(&dev->bar->csts);
c9d3bf88 2794 }
7c1b2450 2795 if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) {
4d115420 2796 for (i = dev->queue_count - 1; i >= 0; i--) {
a4aea562 2797 struct nvme_queue *nvmeq = dev->queues[i];
4d115420 2798 nvme_suspend_queue(nvmeq);
4d115420
KB		 2799 }
2800 } else {
2801 nvme_disable_io_queues(dev);
1894d8f1 2802 nvme_shutdown_ctrl(dev);
4d115420
KB		 2803 nvme_disable_queue(dev, 0);
2804 }
f0b50732 2805 nvme_dev_unmap(dev);
07836e65
KB		 2806
2807 for (i = dev->queue_count - 1; i >= 0; i--)
2808 nvme_clear_queue(dev->queues[i]);
f0b50732
KB		 2809}
2810
2811static void nvme_dev_remove(struct nvme_dev *dev)
2812{
9ac27090 2813 struct nvme_ns *ns;
f0b50732 2814
a5768aa8
KB		 2815 list_for_each_entry(ns, &dev->namespaces, list)
2816 nvme_ns_remove(ns);
b60503ba
MW		 2817}
2818
091b6092
MW		 2819static int nvme_setup_prp_pools(struct nvme_dev *dev)
2820{
e75ec752 2821 dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
091b6092
MW		 2822 PAGE_SIZE, PAGE_SIZE, 0);
2823 if (!dev->prp_page_pool)
2824 return -ENOMEM;
2825
99802a7a 2826 /* Optimisation for I/Os between 4k and 128k */
e75ec752 2827 dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
99802a7a
MW		 2828 256, 256, 0);
2829 if (!dev->prp_small_pool) {
2830 dma_pool_destroy(dev->prp_page_pool);
2831 return -ENOMEM;
2832 }
091b6092
MW		 2833 return 0;
2834}
2835
2836static void nvme_release_prp_pools(struct nvme_dev *dev)
2837{
2838 dma_pool_destroy(dev->prp_page_pool);
99802a7a 2839 dma_pool_destroy(dev->prp_small_pool);
091b6092
MW		 2840}
2841
cd58ad7d
QSA		 2842static DEFINE_IDA(nvme_instance_ida);
2843
2844static int nvme_set_instance(struct nvme_dev *dev)
b60503ba 2845{
cd58ad7d
QSA		 2846 int instance, error;
2847
2848 do {
2849 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2850 return -ENODEV;
2851
2852 spin_lock(&dev_list_lock);
2853 error = ida_get_new(&nvme_instance_ida, &instance);
2854 spin_unlock(&dev_list_lock);
2855 } while (error == -EAGAIN);
2856
2857 if (error)
2858 return -ENODEV;
2859
2860 dev->instance = instance;
2861 return 0;
b60503ba
MW		 2862}
2863
2864static void nvme_release_instance(struct nvme_dev *dev)
2865{
cd58ad7d
QSA		 2866 spin_lock(&dev_list_lock);
2867 ida_remove(&nvme_instance_ida, dev->instance);
2868 spin_unlock(&dev_list_lock);
b60503ba
MW		 2869}
2870
9ac27090
KB		 2871static void nvme_free_namespaces(struct nvme_dev *dev)
2872{
2873 struct nvme_ns *ns, *next;
2874
a5768aa8
KB		 2875 list_for_each_entry_safe(ns, next, &dev->namespaces, list)
2876 nvme_free_namespace(ns);
9ac27090
KB		 2877}
2878
5e82e952
KB		 2879static void nvme_free_dev(struct kref *kref)
2880{
2881 struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
9ac27090 2882
e75ec752 2883 put_device(dev->dev);
b3fffdef 2884 put_device(dev->device);
9ac27090 2885 nvme_free_namespaces(dev);
285dffc9 2886 nvme_release_instance(dev);
4af0e21c
KB		 2887 if (dev->tagset.tags)
2888 blk_mq_free_tag_set(&dev->tagset);
2889 if (dev->admin_q)
2890 blk_put_queue(dev->admin_q);
5e82e952
KB		 2891 kfree(dev->queues);
2892 kfree(dev->entry);
2893 kfree(dev);
2894}
2895
2896static int nvme_dev_open(struct inode *inode, struct file *f)
2897{
b3fffdef
KB		 2898 struct nvme_dev *dev;
2899 int instance = iminor(inode);
2900 int ret = -ENODEV;
2901
2902 spin_lock(&dev_list_lock);
2903 list_for_each_entry(dev, &dev_list, node) {
2904 if (dev->instance == instance) {
2e1d8448
KB		 2905 if (!dev->admin_q) {
2906 ret = -EWOULDBLOCK;
2907 break;
2908 }
b3fffdef
KB		 2909 if (!kref_get_unless_zero(&dev->kref))
2910 break;
2911 f->private_data = dev;
2912 ret = 0;
2913 break;
2914 }
2915 }
2916 spin_unlock(&dev_list_lock);
2917
2918 return ret;
5e82e952
KB		 2919}
2920
2921static int nvme_dev_release(struct inode *inode, struct file *f)
2922{
2923 struct nvme_dev *dev = f->private_data;
2924 kref_put(&dev->kref, nvme_free_dev);
2925 return 0;
2926}
2927
2928static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
2929{
2930 struct nvme_dev *dev = f->private_data;
a4aea562
MB		 2931 struct nvme_ns *ns;
2932
5e82e952
KB		 2933 switch (cmd) {
2934 case NVME_IOCTL_ADMIN_CMD:
a4aea562 2935 return nvme_user_cmd(dev, NULL, (void __user *)arg);
7963e521 2936 case NVME_IOCTL_IO_CMD:
a4aea562
MB		 2937 if (list_empty(&dev->namespaces))
2938 return -ENOTTY;
2939 ns = list_first_entry(&dev->namespaces, struct nvme_ns, list);
2940 return nvme_user_cmd(dev, ns, (void __user *)arg);
4cc06521
KB		 2941 case NVME_IOCTL_RESET:
2942 dev_warn(dev->dev, "resetting controller\n");
2943 return nvme_reset(dev);
81f03fed
JD		 2944 case NVME_IOCTL_SUBSYS_RESET:
2945 return nvme_subsys_reset(dev);
5e82e952
KB		 2946 default:
2947 return -ENOTTY;
2948 }
2949}
2950
2951static const struct file_operations nvme_dev_fops = {
2952 .owner = THIS_MODULE,
2953 .open = nvme_dev_open,
2954 .release = nvme_dev_release,
2955 .unlocked_ioctl = nvme_dev_ioctl,
2956 .compat_ioctl = nvme_dev_ioctl,
2957};
2958
a4aea562
MB		 2959static void nvme_set_irq_hints(struct nvme_dev *dev)
2960{
2961 struct nvme_queue *nvmeq;
2962 int i;
2963
2964 for (i = 0; i < dev->online_queues; i++) {
2965 nvmeq = dev->queues[i];
2966
42483228 2967 if (!nvmeq->tags || !(*nvmeq->tags))
a4aea562
MB		 2968 continue;
2969
2970 irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector,
42483228 2971 blk_mq_tags_cpumask(*nvmeq->tags));
a4aea562
MB		 2972 }
2973}
2974
f0b50732
KB		 2975static int nvme_dev_start(struct nvme_dev *dev)
2976{
2977 int result;
b9afca3e 2978 bool start_thread = false;
f0b50732
KB		 2979
2980 result = nvme_dev_map(dev);
2981 if (result)
2982 return result;
2983
2984 result = nvme_configure_admin_queue(dev);
2985 if (result)
2986 goto unmap;
2987
2988 spin_lock(&dev_list_lock);
b9afca3e
DM		 2989 if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) {
2990 start_thread = true;
2991 nvme_thread = NULL;
2992 }
f0b50732
KB		 2993 list_add(&dev->node, &dev_list);
2994 spin_unlock(&dev_list_lock);
2995
b9afca3e
DM		 2996 if (start_thread) {
2997 nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
387caa5a 2998 wake_up_all(&nvme_kthread_wait);
b9afca3e
DM		 2999 } else
3000 wait_event_killable(nvme_kthread_wait, nvme_thread);
3001
3002 if (IS_ERR_OR_NULL(nvme_thread)) {
3003 result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR;
3004 goto disable;
3005 }
a4aea562
MB		 3006
3007 nvme_init_queue(dev->queues[0], 0);
0fb59cbc
KB		 3008 result = nvme_alloc_admin_tags(dev);
3009 if (result)
3010 goto disable;
b9afca3e 3011
f0b50732 3012 result = nvme_setup_io_queues(dev);
badc34d4 3013 if (result)
0fb59cbc 3014 goto free_tags;
f0b50732 3015
a4aea562
MB		 3016 nvme_set_irq_hints(dev);
3017
1efccc9d 3018 dev->event_limit = 1;
d82e8bfd 3019 return result;
f0b50732 3020
0fb59cbc
KB		 3021 free_tags:
3022 nvme_dev_remove_admin(dev);
4af0e21c
KB		 3023 blk_put_queue(dev->admin_q);
3024 dev->admin_q = NULL;
3025 dev->queues[0]->tags = NULL;
f0b50732 3026 disable:
a1a5ef99 3027 nvme_disable_queue(dev, 0);
b9afca3e 3028 nvme_dev_list_remove(dev);
f0b50732
KB		 3029 unmap:
3030 nvme_dev_unmap(dev);
3031 return result;
3032}
3033
9a6b9458
KB		 3034static int nvme_remove_dead_ctrl(void *arg)
3035{
3036 struct nvme_dev *dev = (struct nvme_dev *)arg;
e75ec752 3037 struct pci_dev *pdev = to_pci_dev(dev->dev);
9a6b9458
KB		 3038
3039 if (pci_get_drvdata(pdev))
c81f4975 3040 pci_stop_and_remove_bus_device_locked(pdev);
9a6b9458
KB		 3041 kref_put(&dev->kref, nvme_free_dev);
3042 return 0;
3043}
3044
3045static void nvme_remove_disks(struct work_struct *ws)
3046{
9a6b9458
KB		 3047 struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
3048
5a92e700 3049 nvme_free_queues(dev, 1);
302c6727 3050 nvme_dev_remove(dev);
9a6b9458
KB		 3051}
3052
3053static int nvme_dev_resume(struct nvme_dev *dev)
3054{
3055 int ret;
3056
3057 ret = nvme_dev_start(dev);
badc34d4 3058 if (ret)
9a6b9458 3059 return ret;
badc34d4 3060 if (dev->online_queues < 2) {
9a6b9458 3061 spin_lock(&dev_list_lock);
9ca97374 3062 dev->reset_workfn = nvme_remove_disks;
9a6b9458
KB		 3063 queue_work(nvme_workq, &dev->reset_work);
3064 spin_unlock(&dev_list_lock);
c9d3bf88
KB		 3065 } else {
3066 nvme_unfreeze_queues(dev);
ffe7704d 3067 nvme_dev_add(dev);
c9d3bf88 3068 nvme_set_irq_hints(dev);
9a6b9458
KB		 3069 }
3070 return 0;
3071}
3072
de3eff2b
KB		 3073static void nvme_dead_ctrl(struct nvme_dev *dev)
3074{
3075 dev_warn(dev->dev, "Device failed to resume\n");
3076 kref_get(&dev->kref);
3077 if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d",
3078 dev->instance))) {
3079 dev_err(dev->dev,
3080 "Failed to start controller remove task\n");
3081 kref_put(&dev->kref, nvme_free_dev);
3082 }
3083}
3084
9a6b9458
KB		 3085static void nvme_dev_reset(struct nvme_dev *dev)
3086{
ffe7704d
KB		 3087 bool in_probe = work_busy(&dev->probe_work);
3088
9a6b9458 3089 nvme_dev_shutdown(dev);
ffe7704d
KB		 3090
3091 /* Synchronize with device probe so that work will see failure status
3092 * and exit gracefully without trying to schedule another reset */
3093 flush_work(&dev->probe_work);
3094
3095 /* Fail this device if reset occured during probe to avoid
3096 * infinite initialization loops. */
3097 if (in_probe) {
de3eff2b 3098 nvme_dead_ctrl(dev);
ffe7704d 3099 return;
9a6b9458 3100 }
ffe7704d
KB		 3101 /* Schedule device resume asynchronously so the reset work is available
3102 * to cleanup errors that may occur during reinitialization */
3103 schedule_work(&dev->probe_work);
9a6b9458
KB		 3104}
3105
3106static void nvme_reset_failed_dev(struct work_struct *ws)
3107{
3108 struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
3109 nvme_dev_reset(dev);
3110}
3111
9ca97374
TH		 3112static void nvme_reset_workfn(struct work_struct *work)
3113{
3114 struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work);
3115 dev->reset_workfn(work);
3116}
3117
4cc06521
KB		 3118static int nvme_reset(struct nvme_dev *dev)
3119{
3120 int ret = -EBUSY;
3121
3122 if (!dev->admin_q || blk_queue_dying(dev->admin_q))
3123 return -ENODEV;
3124
3125 spin_lock(&dev_list_lock);
3126 if (!work_pending(&dev->reset_work)) {
3127 dev->reset_workfn = nvme_reset_failed_dev;
3128 queue_work(nvme_workq, &dev->reset_work);
3129 ret = 0;
3130 }
3131 spin_unlock(&dev_list_lock);
3132
3133 if (!ret) {
3134 flush_work(&dev->reset_work);
ffe7704d 3135 flush_work(&dev->probe_work);
4cc06521
KB		 3136 return 0;
3137 }
3138
3139 return ret;
3140}
3141
3142static ssize_t nvme_sysfs_reset(struct device *dev,
3143 struct device_attribute *attr, const char *buf,
3144 size_t count)
3145{
3146 struct nvme_dev *ndev = dev_get_drvdata(dev);
3147 int ret;
3148
3149 ret = nvme_reset(ndev);
3150 if (ret < 0)
3151 return ret;
3152
3153 return count;
3154}
3155static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3156
2e1d8448 3157static void nvme_async_probe(struct work_struct *work);
8d85fce7 3158static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
b60503ba 3159{
a4aea562 3160 int node, result = -ENOMEM;
b60503ba
MW		 3161 struct nvme_dev *dev;
3162
a4aea562
MB		 3163 node = dev_to_node(&pdev->dev);
3164 if (node == NUMA_NO_NODE)
3165 set_dev_node(&pdev->dev, 0);
3166
3167 dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
b60503ba
MW		 3168 if (!dev)
3169 return -ENOMEM;
a4aea562
MB		 3170 dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry),
3171 GFP_KERNEL, node);
b60503ba
MW		 3172 if (!dev->entry)
3173 goto free;
a4aea562
MB		 3174 dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
3175 GFP_KERNEL, node);
b60503ba
MW		 3176 if (!dev->queues)
3177 goto free;
3178
3179 INIT_LIST_HEAD(&dev->namespaces);
9ca97374
TH		 3180 dev->reset_workfn = nvme_reset_failed_dev;
3181 INIT_WORK(&dev->reset_work, nvme_reset_workfn);
e75ec752 3182 dev->dev = get_device(&pdev->dev);
9a6b9458 3183 pci_set_drvdata(pdev, dev);
cd58ad7d
QSA		 3184 result = nvme_set_instance(dev);
3185 if (result)
a96d4f5c 3186 goto put_pci;
b60503ba 3187
091b6092
MW		 3188 result = nvme_setup_prp_pools(dev);
3189 if (result)
0877cb0d 3190 goto release;
091b6092 3191
fb35e914 3192 kref_init(&dev->kref);
b3fffdef
KB		 3193 dev->device = device_create(nvme_class, &pdev->dev,
3194 MKDEV(nvme_char_major, dev->instance),
3195 dev, "nvme%d", dev->instance);
3196 if (IS_ERR(dev->device)) {
3197 result = PTR_ERR(dev->device);
2e1d8448 3198 goto release_pools;
b3fffdef
KB		 3199 }
3200 get_device(dev->device);
4cc06521
KB		 3201 dev_set_drvdata(dev->device, dev);
3202
3203 result = device_create_file(dev->device, &dev_attr_reset_controller);
3204 if (result)
3205 goto put_dev;
740216fc 3206
e6e96d73 3207 INIT_LIST_HEAD(&dev->node);
a5768aa8 3208 INIT_WORK(&dev->scan_work, nvme_dev_scan);
2e1d8448
KB		 3209 INIT_WORK(&dev->probe_work, nvme_async_probe);
3210 schedule_work(&dev->probe_work);
b60503ba
MW		 3211 return 0;
3212
4cc06521
KB		 3213 put_dev:
3214 device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
3215 put_device(dev->device);
0877cb0d 3216 release_pools:
091b6092 3217 nvme_release_prp_pools(dev);
0877cb0d
KB		 3218 release:
3219 nvme_release_instance(dev);
a96d4f5c 3220 put_pci:
e75ec752 3221 put_device(dev->dev);
b60503ba
MW		 3222 free:
3223 kfree(dev->queues);
3224 kfree(dev->entry);
3225 kfree(dev);
3226 return result;
3227}
3228
2e1d8448
KB		 3229static void nvme_async_probe(struct work_struct *work)
3230{
3231 struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work);
2e1d8448 3232
de3eff2b
KB		 3233 if (nvme_dev_resume(dev) && !work_busy(&dev->reset_work))
3234 nvme_dead_ctrl(dev);
2e1d8448
KB		 3235}
3236
f0d54a54
KB		 3237static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
3238{
a6739479 3239 struct nvme_dev *dev = pci_get_drvdata(pdev);
f0d54a54 3240
a6739479
KB		 3241 if (prepare)
3242 nvme_dev_shutdown(dev);
3243 else
3244 nvme_dev_resume(dev);
f0d54a54
KB		 3245}
3246
09ece142
KB		 3247static void nvme_shutdown(struct pci_dev *pdev)
3248{
3249 struct nvme_dev *dev = pci_get_drvdata(pdev);
3250 nvme_dev_shutdown(dev);
3251}
3252
8d85fce7 3253static void nvme_remove(struct pci_dev *pdev)
b60503ba
MW		 3254{
3255 struct nvme_dev *dev = pci_get_drvdata(pdev);
9a6b9458
KB		 3256
3257 spin_lock(&dev_list_lock);
3258 list_del_init(&dev->node);
3259 spin_unlock(&dev_list_lock);
3260
3261 pci_set_drvdata(pdev, NULL);
2e1d8448 3262 flush_work(&dev->probe_work);
9a6b9458 3263 flush_work(&dev->reset_work);
a5768aa8 3264 flush_work(&dev->scan_work);
4cc06521 3265 device_remove_file(dev->device, &dev_attr_reset_controller);
c9d3bf88 3266 nvme_dev_remove(dev);
3399a3f7 3267 nvme_dev_shutdown(dev);
a4aea562 3268 nvme_dev_remove_admin(dev);
b3fffdef 3269 device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
a1a5ef99 3270 nvme_free_queues(dev, 0);
8ffaadf7 3271 nvme_release_cmb(dev);
9a6b9458 3272 nvme_release_prp_pools(dev);
5e82e952 3273 kref_put(&dev->kref, nvme_free_dev);
b60503ba
MW		 3274}
3275
3276/* These functions are yet to be implemented */
3277#define nvme_error_detected NULL
3278#define nvme_dump_registers NULL
3279#define nvme_link_reset NULL
3280#define nvme_slot_reset NULL
3281#define nvme_error_resume NULL
cd638946 3282
671a6018 3283#ifdef CONFIG_PM_SLEEP
cd638946
KB		 3284static int nvme_suspend(struct device *dev)
3285{
3286 struct pci_dev *pdev = to_pci_dev(dev);
3287 struct nvme_dev *ndev = pci_get_drvdata(pdev);
3288
3289 nvme_dev_shutdown(ndev);
3290 return 0;
3291}
3292
3293static int nvme_resume(struct device *dev)
3294{
3295 struct pci_dev *pdev = to_pci_dev(dev);
3296 struct nvme_dev *ndev = pci_get_drvdata(pdev);
cd638946 3297
9a6b9458 3298 if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) {
9ca97374 3299 ndev->reset_workfn = nvme_reset_failed_dev;
9a6b9458
KB		 3300 queue_work(nvme_workq, &ndev->reset_work);
3301 }
3302 return 0;
cd638946 3303}
671a6018 3304#endif
cd638946
KB		 3305
3306static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
b60503ba 3307
1d352035 3308static const struct pci_error_handlers nvme_err_handler = {
b60503ba
MW		 3309 .error_detected = nvme_error_detected,
3310 .mmio_enabled = nvme_dump_registers,
3311 .link_reset = nvme_link_reset,
3312 .slot_reset = nvme_slot_reset,
3313 .resume = nvme_error_resume,
f0d54a54 3314 .reset_notify = nvme_reset_notify,
b60503ba
MW		 3315};
3316
3317/* Move to pci_ids.h later */
3318#define PCI_CLASS_STORAGE_EXPRESS 0x010802
3319
6eb0d698 3320static const struct pci_device_id nvme_id_table[] = {
b60503ba
MW		 3321 { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
3322 { 0, }
3323};
3324MODULE_DEVICE_TABLE(pci, nvme_id_table);
3325
3326static struct pci_driver nvme_driver = {
3327 .name = "nvme",
3328 .id_table = nvme_id_table,
3329 .probe = nvme_probe,
8d85fce7 3330 .remove = nvme_remove,
09ece142 3331 .shutdown = nvme_shutdown,
cd638946
KB		 3332 .driver = {
3333 .pm = &nvme_dev_pm_ops,
3334 },
b60503ba
MW		 3335 .err_handler = &nvme_err_handler,
3336};
3337
3338static int __init nvme_init(void)
3339{
0ac13140 3340 int result;
1fa6aead 3341
b9afca3e 3342 init_waitqueue_head(&nvme_kthread_wait);
b60503ba 3343
9a6b9458
KB		 3344 nvme_workq = create_singlethread_workqueue("nvme");
3345 if (!nvme_workq)
b9afca3e 3346 return -ENOMEM;
9a6b9458 3347
5c42ea16
KB		 3348 result = register_blkdev(nvme_major, "nvme");
3349 if (result < 0)
9a6b9458 3350 goto kill_workq;
5c42ea16 3351 else if (result > 0)
0ac13140 3352 nvme_major = result;
b60503ba 3353
b3fffdef
KB		 3354 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
3355 &nvme_dev_fops);
3356 if (result < 0)
3357 goto unregister_blkdev;
3358 else if (result > 0)
3359 nvme_char_major = result;
3360
3361 nvme_class = class_create(THIS_MODULE, "nvme");
c727040b
AK		 3362 if (IS_ERR(nvme_class)) {
3363 result = PTR_ERR(nvme_class);
b3fffdef 3364 goto unregister_chrdev;
c727040b 3365 }
b3fffdef 3366
f3db22fe
KB		 3367 result = pci_register_driver(&nvme_driver);
3368 if (result)
b3fffdef 3369 goto destroy_class;
1fa6aead 3370 return 0;
b60503ba 3371
b3fffdef
KB		 3372 destroy_class:
3373 class_destroy(nvme_class);
3374 unregister_chrdev:
3375 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1fa6aead 3376 unregister_blkdev:
b60503ba 3377 unregister_blkdev(nvme_major, "nvme");
9a6b9458
KB		 3378 kill_workq:
3379 destroy_workqueue(nvme_workq);
b60503ba
MW		 3380 return result;
3381}
3382
3383static void __exit nvme_exit(void)
3384{
3385 pci_unregister_driver(&nvme_driver);
3386 unregister_blkdev(nvme_major, "nvme");
9a6b9458 3387 destroy_workqueue(nvme_workq);
b3fffdef
KB		 3388 class_destroy(nvme_class);
3389 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
b9afca3e 3390 BUG_ON(nvme_thread && !IS_ERR(nvme_thread));
21bd78bc 3391 _nvme_check_size();
b60503ba
MW		 3392}
3393
3394MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
3395MODULE_LICENSE("GPL");
c78b4713 3396MODULE_VERSION("1.0");
b60503ba
MW		 3397module_init(nvme_init);
3398module_exit(nvme_exit);
Linux 6.x block layer and io_uring tree(s)