projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
nvme: split command submission helpers out of pci.c
[linux-2.6-block.git] / drivers / nvme / host / pci.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
8de05535 15#include <linux/bitops.h>
b60503ba 16#include <linux/blkdev.h>
a4aea562 17#include <linux/blk-mq.h>
42f61420 18#include <linux/cpu.h>
fd63e9ce 19#include <linux/delay.h>
b60503ba
MW		 20#include <linux/errno.h>
21#include <linux/fs.h>
22#include <linux/genhd.h>
4cc09e2d 23#include <linux/hdreg.h>
5aff9382 24#include <linux/idr.h>
b60503ba
MW		 25#include <linux/init.h>
26#include <linux/interrupt.h>
27#include <linux/io.h>
28#include <linux/kdev_t.h>
1fa6aead 29#include <linux/kthread.h>
b60503ba 30#include <linux/kernel.h>
a5768aa8 31#include <linux/list_sort.h>
b60503ba
MW		 32#include <linux/mm.h>
33#include <linux/module.h>
34#include <linux/moduleparam.h>
35#include <linux/pci.h>
be7b6275 36#include <linux/poison.h>
c3bfe717 37#include <linux/ptrace.h>
b60503ba
MW		 38#include <linux/sched.h>
39#include <linux/slab.h>
e1e5e564 40#include <linux/t10-pi.h>
b60503ba 41#include <linux/types.h>
1d277a63 42#include <linux/pr.h>
5d0f6131 43#include <scsi/sg.h>
2f8e2c87 44#include <linux/io-64-nonatomic-lo-hi.h>
1d277a63 45#include <asm/unaligned.h>
797a796a 46
9d99a8dd 47#include <uapi/linux/nvme_ioctl.h>
f11bb3e2
CH		 48#include "nvme.h"
49
b3fffdef 50#define NVME_MINORS (1U << MINORBITS)
9d43cf64 51#define NVME_Q_DEPTH 1024
d31af0a3 52#define NVME_AQ_DEPTH 256
b60503ba
MW		 53#define SQ_SIZE(depth) (depth * sizeof(struct nvme_command))
54#define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion))
2484f407 55#define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ)
9d43cf64 56
21d34711 57unsigned char admin_timeout = 60;
9d43cf64
KB		 58module_param(admin_timeout, byte, 0644);
59MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
b60503ba 60
bd67608a
MW		 61unsigned char nvme_io_timeout = 30;
62module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
b355084a 63MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
b60503ba 64
2484f407
DM		 65static unsigned char shutdown_timeout = 5;
66module_param(shutdown_timeout, byte, 0644);
67MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
68
b60503ba
MW		 69static int nvme_major;
70module_param(nvme_major, int, 0);
71
b3fffdef
KB		 72static int nvme_char_major;
73module_param(nvme_char_major, int, 0);
74
58ffacb5
MW		 75static int use_threaded_interrupts;
76module_param(use_threaded_interrupts, int, 0);
77
8ffaadf7
JD		 78static bool use_cmb_sqes = true;
79module_param(use_cmb_sqes, bool, 0644);
80MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");
81
1fa6aead
MW		 82static DEFINE_SPINLOCK(dev_list_lock);
83static LIST_HEAD(dev_list);
84static struct task_struct *nvme_thread;
9a6b9458 85static struct workqueue_struct *nvme_workq;
b9afca3e 86static wait_queue_head_t nvme_kthread_wait;
1fa6aead 87
b3fffdef
KB		 88static struct class *nvme_class;
89
90667892 90static int __nvme_reset(struct nvme_dev *dev);
4cc06521 91static int nvme_reset(struct nvme_dev *dev);
a0fa9647 92static void nvme_process_cq(struct nvme_queue *nvmeq);
3cf519b5 93static void nvme_dead_ctrl(struct nvme_dev *dev);
d4b4ff8e 94
4d115420
KB		 95struct async_cmd_info {
96 struct kthread_work work;
97 struct kthread_worker *worker;
a4aea562 98 struct request *req;
4d115420
KB		 99 u32 result;
100 int status;
101 void *ctx;
102};
1fa6aead 103
b60503ba
MW		 104/*
105 * An NVM Express queue. Each device has at least two (one for admin
106 * commands and one for I/O commands).
107 */
108struct nvme_queue {
109 struct device *q_dmadev;
091b6092 110 struct nvme_dev *dev;
3193f07b 111 char irqname[24]; /* nvme4294967295-65535\0 */
b60503ba
MW		 112 spinlock_t q_lock;
113 struct nvme_command *sq_cmds;
8ffaadf7 114 struct nvme_command __iomem *sq_cmds_io;
b60503ba 115 volatile struct nvme_completion *cqes;
42483228 116 struct blk_mq_tags **tags;
b60503ba
MW		 117 dma_addr_t sq_dma_addr;
118 dma_addr_t cq_dma_addr;
b60503ba
MW		 119 u32 __iomem *q_db;
120 u16 q_depth;
6222d172 121 s16 cq_vector;
b60503ba
MW		 122 u16 sq_head;
123 u16 sq_tail;
124 u16 cq_head;
c30341dc 125 u16 qid;
e9539f47
MW		 126 u8 cq_phase;
127 u8 cqe_seen;
4d115420 128 struct async_cmd_info cmdinfo;
b60503ba
MW		 129};
130
71bd150c
CH		 131/*
132 * The nvme_iod describes the data in an I/O, including the list of PRP
133 * entries. You can't see it in this data structure because C doesn't let
134 * me express that. Use nvme_alloc_iod to ensure there's enough space
135 * allocated to store the PRP list.
136 */
137struct nvme_iod {
138 unsigned long private; /* For the use of the submitter of the I/O */
139 int npages; /* In the PRP list. 0 means small pool in use */
140 int offset; /* Of PRP list */
141 int nents; /* Used in scatterlist */
142 int length; /* Of data, in bytes */
143 dma_addr_t first_dma;
144 struct scatterlist meta_sg[1]; /* metadata requires single contiguous buffer */
145 struct scatterlist sg[0];
146};
147
b60503ba
MW		 148/*
149 * Check we didin't inadvertently grow the command struct
150 */
151static inline void _nvme_check_size(void)
152{
153 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
154 BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64);
155 BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64);
156 BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
157 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
f8ebf840 158 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
c30341dc 159 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
b60503ba
MW		 160 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
161 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
162 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
163 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
6ecec745 164 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
b60503ba
MW		 165}
166
edd10d33 167typedef void (*nvme_completion_fn)(struct nvme_queue *, void *,
c2f5b650
MW		 168 struct nvme_completion *);
169
e85248e5 170struct nvme_cmd_info {
c2f5b650
MW		 171 nvme_completion_fn fn;
172 void *ctx;
c30341dc 173 int aborted;
a4aea562 174 struct nvme_queue *nvmeq;
ac3dd5bd 175 struct nvme_iod iod[0];
e85248e5
MW		 176};
177
ac3dd5bd
JA		 178/*
179 * Max size of iod being embedded in the request payload
180 */
181#define NVME_INT_PAGES 2
182#define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->page_size)
fda631ff 183#define NVME_INT_MASK 0x01
ac3dd5bd
JA		 184
185/*
186 * Will slightly overestimate the number of pages needed. This is OK
187 * as it only leads to a small amount of wasted memory for the lifetime of
188 * the I/O.
189 */
190static int nvme_npages(unsigned size, struct nvme_dev *dev)
191{
192 unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size);
193 return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
194}
195
196static unsigned int nvme_cmd_size(struct nvme_dev *dev)
197{
198 unsigned int ret = sizeof(struct nvme_cmd_info);
199
200 ret += sizeof(struct nvme_iod);
201 ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev);
202 ret += sizeof(struct scatterlist) * NVME_INT_PAGES;
203
204 return ret;
205}
206
a4aea562
MB		 207static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
208 unsigned int hctx_idx)
e85248e5 209{
a4aea562
MB		 210 struct nvme_dev *dev = data;
211 struct nvme_queue *nvmeq = dev->queues[0];
212
42483228
KB		 213 WARN_ON(hctx_idx != 0);
214 WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
215 WARN_ON(nvmeq->tags);
216
a4aea562 217 hctx->driver_data = nvmeq;
42483228 218 nvmeq->tags = &dev->admin_tagset.tags[0];
a4aea562 219 return 0;
e85248e5
MW		 220}
221
4af0e21c
KB		 222static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
223{
224 struct nvme_queue *nvmeq = hctx->driver_data;
225
226 nvmeq->tags = NULL;
227}
228
a4aea562
MB		 229static int nvme_admin_init_request(void *data, struct request *req,
230 unsigned int hctx_idx, unsigned int rq_idx,
231 unsigned int numa_node)
22404274 232{
a4aea562
MB		 233 struct nvme_dev *dev = data;
234 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
235 struct nvme_queue *nvmeq = dev->queues[0];
236
237 BUG_ON(!nvmeq);
238 cmd->nvmeq = nvmeq;
239 return 0;
22404274
KB		 240}
241
a4aea562
MB		 242static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
243 unsigned int hctx_idx)
b60503ba 244{
a4aea562 245 struct nvme_dev *dev = data;
42483228 246 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
a4aea562 247
42483228
KB		 248 if (!nvmeq->tags)
249 nvmeq->tags = &dev->tagset.tags[hctx_idx];
b60503ba 250
42483228 251 WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
a4aea562
MB		 252 hctx->driver_data = nvmeq;
253 return 0;
b60503ba
MW		 254}
255
a4aea562
MB		 256static int nvme_init_request(void *data, struct request *req,
257 unsigned int hctx_idx, unsigned int rq_idx,
258 unsigned int numa_node)
b60503ba 259{
a4aea562
MB		 260 struct nvme_dev *dev = data;
261 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
262 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
263
264 BUG_ON(!nvmeq);
265 cmd->nvmeq = nvmeq;
266 return 0;
267}
268
269static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx,
270 nvme_completion_fn handler)
271{
272 cmd->fn = handler;
273 cmd->ctx = ctx;
274 cmd->aborted = 0;
c917dfe5 275 blk_mq_start_request(blk_mq_rq_from_pdu(cmd));
b60503ba
MW		 276}
277
ac3dd5bd
JA		 278static void *iod_get_private(struct nvme_iod *iod)
279{
280 return (void *) (iod->private & ~0x1UL);
281}
282
283/*
284 * If bit 0 is set, the iod is embedded in the request payload.
285 */
286static bool iod_should_kfree(struct nvme_iod *iod)
287{
fda631ff 288 return (iod->private & NVME_INT_MASK) == 0;
ac3dd5bd
JA		 289}
290
c2f5b650
MW		 291/* Special values must be less than 0x1000 */
292#define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA)
d2d87034
MW		 293#define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE)
294#define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE)
295#define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE)
be7b6275 296
edd10d33 297static void special_completion(struct nvme_queue *nvmeq, void *ctx,
c2f5b650
MW		 298 struct nvme_completion *cqe)
299{
300 if (ctx == CMD_CTX_CANCELLED)
301 return;
c2f5b650 302 if (ctx == CMD_CTX_COMPLETED) {
edd10d33 303 dev_warn(nvmeq->q_dmadev,
c2f5b650
MW		 304 "completed id %d twice on queue %d\n",
305 cqe->command_id, le16_to_cpup(&cqe->sq_id));
306 return;
307 }
308 if (ctx == CMD_CTX_INVALID) {
edd10d33 309 dev_warn(nvmeq->q_dmadev,
c2f5b650
MW		 310 "invalid id %d completed on queue %d\n",
311 cqe->command_id, le16_to_cpup(&cqe->sq_id));
312 return;
313 }
edd10d33 314 dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx);
c2f5b650
MW		 315}
316
a4aea562 317static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn)
b60503ba 318{
c2f5b650 319 void *ctx;
b60503ba 320
859361a2 321 if (fn)
a4aea562
MB		 322 *fn = cmd->fn;
323 ctx = cmd->ctx;
324 cmd->fn = special_completion;
325 cmd->ctx = CMD_CTX_CANCELLED;
c2f5b650 326 return ctx;
b60503ba
MW		 327}
328
a4aea562
MB		 329static void async_req_completion(struct nvme_queue *nvmeq, void *ctx,
330 struct nvme_completion *cqe)
3c0cf138 331{
a4aea562
MB		 332 u32 result = le32_to_cpup(&cqe->result);
333 u16 status = le16_to_cpup(&cqe->status) >> 1;
334
335 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ)
336 ++nvmeq->dev->event_limit;
a5768aa8
KB		 337 if (status != NVME_SC_SUCCESS)
338 return;
339
340 switch (result & 0xff07) {
341 case NVME_AER_NOTICE_NS_CHANGED:
342 dev_info(nvmeq->q_dmadev, "rescanning\n");
343 schedule_work(&nvmeq->dev->scan_work);
344 default:
345 dev_warn(nvmeq->q_dmadev, "async event result %08x\n", result);
346 }
b60503ba
MW		 347}
348
a4aea562
MB		 349static void abort_completion(struct nvme_queue *nvmeq, void *ctx,
350 struct nvme_completion *cqe)
5a92e700 351{
a4aea562
MB		 352 struct request *req = ctx;
353
354 u16 status = le16_to_cpup(&cqe->status) >> 1;
355 u32 result = le32_to_cpup(&cqe->result);
a51afb54 356
42483228 357 blk_mq_free_request(req);
a51afb54 358
a4aea562
MB		 359 dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result);
360 ++nvmeq->dev->abort_limit;
5a92e700
KB		 361}
362
a4aea562
MB		 363static void async_completion(struct nvme_queue *nvmeq, void *ctx,
364 struct nvme_completion *cqe)
b60503ba 365{
a4aea562
MB		 366 struct async_cmd_info *cmdinfo = ctx;
367 cmdinfo->result = le32_to_cpup(&cqe->result);
368 cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
369 queue_kthread_work(cmdinfo->worker, &cmdinfo->work);
42483228 370 blk_mq_free_request(cmdinfo->req);
b60503ba
MW		 371}
372
a4aea562
MB		 373static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq,
374 unsigned int tag)
b60503ba 375{
42483228 376 struct request *req = blk_mq_tag_to_rq(*nvmeq->tags, tag);
a51afb54 377
a4aea562 378 return blk_mq_rq_to_pdu(req);
4f5099af
KB		 379}
380
a4aea562
MB		 381/*
382 * Called with local interrupts disabled and the q_lock held. May not sleep.
383 */
384static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag,
385 nvme_completion_fn *fn)
4f5099af 386{
a4aea562
MB		 387 struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag);
388 void *ctx;
389 if (tag >= nvmeq->q_depth) {
390 *fn = special_completion;
391 return CMD_CTX_INVALID;
392 }
393 if (fn)
394 *fn = cmd->fn;
395 ctx = cmd->ctx;
396 cmd->fn = special_completion;
397 cmd->ctx = CMD_CTX_COMPLETED;
398 return ctx;
b60503ba
MW		 399}
400
401/**
714a7a22 402 * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
b60503ba
MW		 403 * @nvmeq: The queue to use
404 * @cmd: The command to send
405 *
406 * Safe to use from interrupt context
407 */
e3f879bf
SB		 408static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
409 struct nvme_command *cmd)
b60503ba 410{
a4aea562
MB		 411 u16 tail = nvmeq->sq_tail;
412
8ffaadf7
JD		 413 if (nvmeq->sq_cmds_io)
414 memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd));
415 else
416 memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));
417
b60503ba
MW		 418 if (++tail == nvmeq->q_depth)
419 tail = 0;
7547881d 420 writel(tail, nvmeq->q_db);
b60503ba 421 nvmeq->sq_tail = tail;
b60503ba
MW		 422}
423
e3f879bf 424static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
a4aea562
MB		 425{
426 unsigned long flags;
a4aea562 427 spin_lock_irqsave(&nvmeq->q_lock, flags);
e3f879bf 428 __nvme_submit_cmd(nvmeq, cmd);
a4aea562 429 spin_unlock_irqrestore(&nvmeq->q_lock, flags);
a4aea562
MB		 430}
431
eca18b23 432static __le64 **iod_list(struct nvme_iod *iod)
e025344c 433{
eca18b23 434 return ((void *)iod) + iod->offset;
e025344c
SMM		 435}
436
ac3dd5bd
JA		 437static inline void iod_init(struct nvme_iod *iod, unsigned nbytes,
438 unsigned nseg, unsigned long private)
eca18b23 439{
ac3dd5bd
JA		 440 iod->private = private;
441 iod->offset = offsetof(struct nvme_iod, sg[nseg]);
442 iod->npages = -1;
443 iod->length = nbytes;
444 iod->nents = 0;
eca18b23 445}
b60503ba 446
eca18b23 447static struct nvme_iod *
ac3dd5bd
JA		 448__nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev,
449 unsigned long priv, gfp_t gfp)
b60503ba 450{
eca18b23 451 struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) +
ac3dd5bd 452 sizeof(__le64 *) * nvme_npages(bytes, dev) +
eca18b23
MW		 453 sizeof(struct scatterlist) * nseg, gfp);
454
ac3dd5bd
JA		 455 if (iod)
456 iod_init(iod, bytes, nseg, priv);
eca18b23
MW		 457
458 return iod;
b60503ba
MW		 459}
460
ac3dd5bd
JA		 461static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev,
462 gfp_t gfp)
463{
464 unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) :
465 sizeof(struct nvme_dsm_range);
ac3dd5bd
JA		 466 struct nvme_iod *iod;
467
468 if (rq->nr_phys_segments <= NVME_INT_PAGES &&
469 size <= NVME_INT_BYTES(dev)) {
470 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq);
471
472 iod = cmd->iod;
ac3dd5bd 473 iod_init(iod, size, rq->nr_phys_segments,
fda631ff 474 (unsigned long) rq | NVME_INT_MASK);
ac3dd5bd
JA		 475 return iod;
476 }
477
478 return __nvme_alloc_iod(rq->nr_phys_segments, size, dev,
479 (unsigned long) rq, gfp);
480}
481
d29ec824 482static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod)
b60503ba 483{
1d090624 484 const int last_prp = dev->page_size / 8 - 1;
eca18b23
MW		 485 int i;
486 __le64 **list = iod_list(iod);
487 dma_addr_t prp_dma = iod->first_dma;
488
489 if (iod->npages == 0)
490 dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
491 for (i = 0; i < iod->npages; i++) {
492 __le64 *prp_list = list[i];
493 dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
494 dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
495 prp_dma = next_prp_dma;
496 }
ac3dd5bd
JA		 497
498 if (iod_should_kfree(iod))
499 kfree(iod);
b60503ba
MW		 500}
501
b4ff9c8d
KB		 502static int nvme_error_status(u16 status)
503{
504 switch (status & 0x7ff) {
505 case NVME_SC_SUCCESS:
506 return 0;
507 case NVME_SC_CAP_EXCEEDED:
508 return -ENOSPC;
509 default:
510 return -EIO;
511 }
512}
513
52b68d7e 514#ifdef CONFIG_BLK_DEV_INTEGRITY
e1e5e564
KB		 515static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
516{
517 if (be32_to_cpu(pi->ref_tag) == v)
518 pi->ref_tag = cpu_to_be32(p);
519}
520
521static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
522{
523 if (be32_to_cpu(pi->ref_tag) == p)
524 pi->ref_tag = cpu_to_be32(v);
525}
526
527/**
528 * nvme_dif_remap - remaps ref tags to bip seed and physical lba
529 *
530 * The virtual start sector is the one that was originally submitted by the
531 * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical
532 * start sector may be different. Remap protection information to match the
533 * physical LBA on writes, and back to the original seed on reads.
534 *
535 * Type 0 and 3 do not have a ref tag, so no remapping required.
536 */
537static void nvme_dif_remap(struct request *req,
538 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
539{
540 struct nvme_ns *ns = req->rq_disk->private_data;
541 struct bio_integrity_payload *bip;
542 struct t10_pi_tuple *pi;
543 void *p, *pmap;
544 u32 i, nlb, ts, phys, virt;
545
546 if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3)
547 return;
548
549 bip = bio_integrity(req->bio);
550 if (!bip)
551 return;
552
553 pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset;
e1e5e564
KB		 554
555 p = pmap;
556 virt = bip_get_seed(bip);
557 phys = nvme_block_nr(ns, blk_rq_pos(req));
558 nlb = (blk_rq_bytes(req) >> ns->lba_shift);
ac6fc48c 559 ts = ns->disk->queue->integrity.tuple_size;
e1e5e564
KB		 560
561 for (i = 0; i < nlb; i++, virt++, phys++) {
562 pi = (struct t10_pi_tuple *)p;
563 dif_swap(phys, virt, pi);
564 p += ts;
565 }
566 kunmap_atomic(pmap);
567}
568
52b68d7e
KB		 569static void nvme_init_integrity(struct nvme_ns *ns)
570{
571 struct blk_integrity integrity;
572
573 switch (ns->pi_type) {
574 case NVME_NS_DPS_PI_TYPE3:
0f8087ec 575 integrity.profile = &t10_pi_type3_crc;
52b68d7e
KB		 576 break;
577 case NVME_NS_DPS_PI_TYPE1:
578 case NVME_NS_DPS_PI_TYPE2:
0f8087ec 579 integrity.profile = &t10_pi_type1_crc;
52b68d7e
KB		 580 break;
581 default:
4125a09b 582 integrity.profile = NULL;
52b68d7e
KB		 583 break;
584 }
585 integrity.tuple_size = ns->ms;
586 blk_integrity_register(ns->disk, &integrity);
587 blk_queue_max_integrity_segments(ns->queue, 1);
588}
589#else /* CONFIG_BLK_DEV_INTEGRITY */
590static void nvme_dif_remap(struct request *req,
591 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
592{
593}
594static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
595{
596}
597static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
598{
599}
600static void nvme_init_integrity(struct nvme_ns *ns)
601{
602}
603#endif
604
a4aea562 605static void req_completion(struct nvme_queue *nvmeq, void *ctx,
b60503ba
MW		 606 struct nvme_completion *cqe)
607{
eca18b23 608 struct nvme_iod *iod = ctx;
ac3dd5bd 609 struct request *req = iod_get_private(iod);
a4aea562 610 struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
b60503ba 611 u16 status = le16_to_cpup(&cqe->status) >> 1;
0dfc70c3 612 bool requeue = false;
81c04b94 613 int error = 0;
b60503ba 614
edd10d33 615 if (unlikely(status)) {
a4aea562
MB		 616 if (!(status & NVME_SC_DNR || blk_noretry_request(req))
617 && (jiffies - req->start_time) < req->timeout) {
c9d3bf88
KB		 618 unsigned long flags;
619
0dfc70c3 620 requeue = true;
a4aea562 621 blk_mq_requeue_request(req);
c9d3bf88
KB		 622 spin_lock_irqsave(req->q->queue_lock, flags);
623 if (!blk_queue_stopped(req->q))
624 blk_mq_kick_requeue_list(req->q);
625 spin_unlock_irqrestore(req->q->queue_lock, flags);
0dfc70c3 626 goto release_iod;
edd10d33 627 }
f4829a9b 628
d29ec824 629 if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
17188bb4 630 if (cmd_rq->ctx == CMD_CTX_CANCELLED)
81c04b94
CH		 631 error = -EINTR;
632 else
633 error = status;
d29ec824 634 } else {
81c04b94 635 error = nvme_error_status(status);
d29ec824 636 }
f4829a9b
CH		 637 }
638
a0a931d6
KB		 639 if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
640 u32 result = le32_to_cpup(&cqe->result);
641 req->special = (void *)(uintptr_t)result;
642 }
a4aea562
MB		 643
644 if (cmd_rq->aborted)
e75ec752 645 dev_warn(nvmeq->dev->dev,
a4aea562 646 "completing aborted command with status:%04x\n",
81c04b94 647 error);
a4aea562 648
0dfc70c3 649release_iod:
e1e5e564 650 if (iod->nents) {
e75ec752 651 dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents,
a4aea562 652 rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
e1e5e564
KB		 653 if (blk_integrity_rq(req)) {
654 if (!rq_data_dir(req))
655 nvme_dif_remap(req, nvme_dif_complete);
e75ec752 656 dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1,
e1e5e564
KB		 657 rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
658 }
659 }
edd10d33 660 nvme_free_iod(nvmeq->dev, iod);
3291fa57 661
0dfc70c3
KB		 662 if (likely(!requeue))
663 blk_mq_complete_request(req, error);
b60503ba
MW		 664}
665
184d2944 666/* length is in bytes. gfp flags indicates whether we may sleep. */
d29ec824
CH		 667static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod,
668 int total_len, gfp_t gfp)
ff22b54f 669{
99802a7a 670 struct dma_pool *pool;
eca18b23
MW		 671 int length = total_len;
672 struct scatterlist *sg = iod->sg;
ff22b54f
MW		 673 int dma_len = sg_dma_len(sg);
674 u64 dma_addr = sg_dma_address(sg);
f137e0f1
MI		 675 u32 page_size = dev->page_size;
676 int offset = dma_addr & (page_size - 1);
e025344c 677 __le64 *prp_list;
eca18b23 678 __le64 **list = iod_list(iod);
e025344c 679 dma_addr_t prp_dma;
eca18b23 680 int nprps, i;
ff22b54f 681
1d090624 682 length -= (page_size - offset);
ff22b54f 683 if (length <= 0)
eca18b23 684 return total_len;
ff22b54f 685
1d090624 686 dma_len -= (page_size - offset);
ff22b54f 687 if (dma_len) {
1d090624 688 dma_addr += (page_size - offset);
ff22b54f
MW		 689 } else {
690 sg = sg_next(sg);
691 dma_addr = sg_dma_address(sg);
692 dma_len = sg_dma_len(sg);
693 }
694
1d090624 695 if (length <= page_size) {
edd10d33 696 iod->first_dma = dma_addr;
eca18b23 697 return total_len;
e025344c
SMM		 698 }
699
1d090624 700 nprps = DIV_ROUND_UP(length, page_size);
99802a7a
MW		 701 if (nprps <= (256 / 8)) {
702 pool = dev->prp_small_pool;
eca18b23 703 iod->npages = 0;
99802a7a
MW		 704 } else {
705 pool = dev->prp_page_pool;
eca18b23 706 iod->npages = 1;
99802a7a
MW		 707 }
708
b77954cb
MW		 709 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
710 if (!prp_list) {
edd10d33 711 iod->first_dma = dma_addr;
eca18b23 712 iod->npages = -1;
1d090624 713 return (total_len - length) + page_size;
b77954cb 714 }
eca18b23
MW		 715 list[0] = prp_list;
716 iod->first_dma = prp_dma;
e025344c
SMM		 717 i = 0;
718 for (;;) {
1d090624 719 if (i == page_size >> 3) {
e025344c 720 __le64 *old_prp_list = prp_list;
b77954cb 721 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
eca18b23
MW		 722 if (!prp_list)
723 return total_len - length;
724 list[iod->npages++] = prp_list;
7523d834
MW		 725 prp_list[0] = old_prp_list[i - 1];
726 old_prp_list[i - 1] = cpu_to_le64(prp_dma);
727 i = 1;
e025344c
SMM		 728 }
729 prp_list[i++] = cpu_to_le64(dma_addr);
1d090624
KB		 730 dma_len -= page_size;
731 dma_addr += page_size;
732 length -= page_size;
e025344c
SMM		 733 if (length <= 0)
734 break;
735 if (dma_len > 0)
736 continue;
737 BUG_ON(dma_len < 0);
738 sg = sg_next(sg);
739 dma_addr = sg_dma_address(sg);
740 dma_len = sg_dma_len(sg);
ff22b54f
MW		 741 }
742
eca18b23 743 return total_len;
ff22b54f
MW		 744}
745
d29ec824
CH		 746static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req,
747 struct nvme_iod *iod)
748{
498c4394 749 struct nvme_command cmnd;
d29ec824 750
498c4394
JD		 751 memcpy(&cmnd, req->cmd, sizeof(cmnd));
752 cmnd.rw.command_id = req->tag;
d29ec824 753 if (req->nr_phys_segments) {
498c4394
JD		 754 cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
755 cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
d29ec824
CH		 756 }
757
498c4394 758 __nvme_submit_cmd(nvmeq, &cmnd);
d29ec824
CH		 759}
760
a4aea562
MB		 761/*
762 * We reuse the small pool to allocate the 16-byte range here as it is not
763 * worth having a special pool for these or additional cases to handle freeing
764 * the iod.
765 */
766static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns,
767 struct request *req, struct nvme_iod *iod)
0e5e4f0e 768{
edd10d33
KB		 769 struct nvme_dsm_range *range =
770 (struct nvme_dsm_range *)iod_list(iod)[0];
498c4394 771 struct nvme_command cmnd;
0e5e4f0e 772
0e5e4f0e 773 range->cattr = cpu_to_le32(0);
a4aea562
MB		 774 range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift);
775 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
0e5e4f0e 776
498c4394
JD		 777 memset(&cmnd, 0, sizeof(cmnd));
778 cmnd.dsm.opcode = nvme_cmd_dsm;
779 cmnd.dsm.command_id = req->tag;
780 cmnd.dsm.nsid = cpu_to_le32(ns->ns_id);
781 cmnd.dsm.prp1 = cpu_to_le64(iod->first_dma);
782 cmnd.dsm.nr = 0;
783 cmnd.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
0e5e4f0e 784
498c4394 785 __nvme_submit_cmd(nvmeq, &cmnd);
0e5e4f0e
KB		 786}
787
a4aea562 788static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns,
00df5cb4
MW		 789 int cmdid)
790{
498c4394 791 struct nvme_command cmnd;
00df5cb4 792
498c4394
JD		 793 memset(&cmnd, 0, sizeof(cmnd));
794 cmnd.common.opcode = nvme_cmd_flush;
795 cmnd.common.command_id = cmdid;
796 cmnd.common.nsid = cpu_to_le32(ns->ns_id);
00df5cb4 797
498c4394 798 __nvme_submit_cmd(nvmeq, &cmnd);
00df5cb4
MW		 799}
800
a4aea562
MB		 801static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod,
802 struct nvme_ns *ns)
b60503ba 803{
ac3dd5bd 804 struct request *req = iod_get_private(iod);
498c4394 805 struct nvme_command cmnd;
a4aea562
MB		 806 u16 control = 0;
807 u32 dsmgmt = 0;
00df5cb4 808
a4aea562 809 if (req->cmd_flags & REQ_FUA)
b60503ba 810 control |= NVME_RW_FUA;
a4aea562 811 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
b60503ba
MW		 812 control |= NVME_RW_LR;
813
a4aea562 814 if (req->cmd_flags & REQ_RAHEAD)
b60503ba
MW		 815 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
816
498c4394
JD		 817 memset(&cmnd, 0, sizeof(cmnd));
818 cmnd.rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
819 cmnd.rw.command_id = req->tag;
820 cmnd.rw.nsid = cpu_to_le32(ns->ns_id);
821 cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
822 cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
823 cmnd.rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
824 cmnd.rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
b60503ba 825
e19b127f 826 if (ns->ms) {
e1e5e564
KB		 827 switch (ns->pi_type) {
828 case NVME_NS_DPS_PI_TYPE3:
829 control |= NVME_RW_PRINFO_PRCHK_GUARD;
830 break;
831 case NVME_NS_DPS_PI_TYPE1:
832 case NVME_NS_DPS_PI_TYPE2:
833 control |= NVME_RW_PRINFO_PRCHK_GUARD |
834 NVME_RW_PRINFO_PRCHK_REF;
498c4394 835 cmnd.rw.reftag = cpu_to_le32(
e1e5e564
KB		 836 nvme_block_nr(ns, blk_rq_pos(req)));
837 break;
838 }
e19b127f
AP		 839 if (blk_integrity_rq(req))
840 cmnd.rw.metadata =
841 cpu_to_le64(sg_dma_address(iod->meta_sg));
842 else
843 control |= NVME_RW_PRINFO_PRACT;
844 }
e1e5e564 845
498c4394
JD		 846 cmnd.rw.control = cpu_to_le16(control);
847 cmnd.rw.dsmgmt = cpu_to_le32(dsmgmt);
b60503ba 848
498c4394 849 __nvme_submit_cmd(nvmeq, &cmnd);
b60503ba 850
1974b1ae 851 return 0;
edd10d33
KB		 852}
853
d29ec824
CH		 854/*
855 * NOTE: ns is NULL when called on the admin queue.
856 */
a4aea562
MB		 857static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
858 const struct blk_mq_queue_data *bd)
edd10d33 859{
a4aea562
MB		 860 struct nvme_ns *ns = hctx->queue->queuedata;
861 struct nvme_queue *nvmeq = hctx->driver_data;
d29ec824 862 struct nvme_dev *dev = nvmeq->dev;
a4aea562
MB		 863 struct request *req = bd->rq;
864 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
edd10d33 865 struct nvme_iod *iod;
a4aea562 866 enum dma_data_direction dma_dir;
edd10d33 867
e1e5e564
KB		 868 /*
869 * If formated with metadata, require the block layer provide a buffer
870 * unless this namespace is formated such that the metadata can be
871 * stripped/generated by the controller with PRACT=1.
872 */
d29ec824 873 if (ns && ns->ms && !blk_integrity_rq(req)) {
71feb364
KB		 874 if (!(ns->pi_type && ns->ms == 8) &&
875 req->cmd_type != REQ_TYPE_DRV_PRIV) {
f4829a9b 876 blk_mq_complete_request(req, -EFAULT);
e1e5e564
KB		 877 return BLK_MQ_RQ_QUEUE_OK;
878 }
879 }
880
d29ec824 881 iod = nvme_alloc_iod(req, dev, GFP_ATOMIC);
edd10d33 882 if (!iod)
fe54303e 883 return BLK_MQ_RQ_QUEUE_BUSY;
a4aea562 884
a4aea562 885 if (req->cmd_flags & REQ_DISCARD) {
edd10d33
KB		 886 void *range;
887 /*
888 * We reuse the small pool to allocate the 16-byte range here
889 * as it is not worth having a special pool for these or
890 * additional cases to handle freeing the iod.
891 */
d29ec824 892 range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC,
edd10d33 893 &iod->first_dma);
a4aea562 894 if (!range)
fe54303e 895 goto retry_cmd;
edd10d33
KB		 896 iod_list(iod)[0] = (__le64 *)range;
897 iod->npages = 0;
ac3dd5bd 898 } else if (req->nr_phys_segments) {
a4aea562
MB		 899 dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
900
ac3dd5bd 901 sg_init_table(iod->sg, req->nr_phys_segments);
a4aea562 902 iod->nents = blk_rq_map_sg(req->q, req, iod->sg);
fe54303e
JA		 903 if (!iod->nents)
904 goto error_cmd;
a4aea562
MB		 905
906 if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir))
fe54303e 907 goto retry_cmd;
a4aea562 908
fe54303e 909 if (blk_rq_bytes(req) !=
d29ec824
CH		 910 nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) {
911 dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
fe54303e
JA		 912 goto retry_cmd;
913 }
e1e5e564 914 if (blk_integrity_rq(req)) {
bf508e91
CH		 915 if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) {
916 dma_unmap_sg(dev->dev, iod->sg, iod->nents,
917 dma_dir);
e1e5e564 918 goto error_cmd;
bf508e91 919 }
e1e5e564
KB		 920
921 sg_init_table(iod->meta_sg, 1);
922 if (blk_rq_map_integrity_sg(
bf508e91
CH		 923 req->q, req->bio, iod->meta_sg) != 1) {
924 dma_unmap_sg(dev->dev, iod->sg, iod->nents,
925 dma_dir);
e1e5e564 926 goto error_cmd;
bf508e91 927 }
e1e5e564
KB		 928
929 if (rq_data_dir(req))
930 nvme_dif_remap(req, nvme_dif_prep);
931
bf508e91
CH		 932 if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir)) {
933 dma_unmap_sg(dev->dev, iod->sg, iod->nents,
934 dma_dir);
e1e5e564 935 goto error_cmd;
bf508e91 936 }
e1e5e564 937 }
edd10d33 938 }
1974b1ae 939
9af8785a 940 nvme_set_info(cmd, iod, req_completion);
a4aea562 941 spin_lock_irq(&nvmeq->q_lock);
d29ec824
CH		 942 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
943 nvme_submit_priv(nvmeq, req, iod);
944 else if (req->cmd_flags & REQ_DISCARD)
a4aea562
MB		 945 nvme_submit_discard(nvmeq, ns, req, iod);
946 else if (req->cmd_flags & REQ_FLUSH)
947 nvme_submit_flush(nvmeq, ns, req->tag);
948 else
949 nvme_submit_iod(nvmeq, iod, ns);
950
951 nvme_process_cq(nvmeq);
952 spin_unlock_irq(&nvmeq->q_lock);
953 return BLK_MQ_RQ_QUEUE_OK;
954
fe54303e 955 error_cmd:
d29ec824 956 nvme_free_iod(dev, iod);
fe54303e
JA		 957 return BLK_MQ_RQ_QUEUE_ERROR;
958 retry_cmd:
d29ec824 959 nvme_free_iod(dev, iod);
fe54303e 960 return BLK_MQ_RQ_QUEUE_BUSY;
b60503ba
MW		 961}
962
a0fa9647 963static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag)
b60503ba 964{
82123460 965 u16 head, phase;
b60503ba 966
b60503ba 967 head = nvmeq->cq_head;
82123460 968 phase = nvmeq->cq_phase;
b60503ba
MW		 969
970 for (;;) {
c2f5b650
MW		 971 void *ctx;
972 nvme_completion_fn fn;
b60503ba 973 struct nvme_completion cqe = nvmeq->cqes[head];
82123460 974 if ((le16_to_cpu(cqe.status) & 1) != phase)
b60503ba
MW		 975 break;
976 nvmeq->sq_head = le16_to_cpu(cqe.sq_head);
977 if (++head == nvmeq->q_depth) {
978 head = 0;
82123460 979 phase = !phase;
b60503ba 980 }
a0fa9647
JA		 981 if (tag && *tag == cqe.command_id)
982 *tag = -1;
a4aea562 983 ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn);
edd10d33 984 fn(nvmeq, ctx, &cqe);
b60503ba
MW		 985 }
986
987 /* If the controller ignores the cq head doorbell and continuously
988 * writes to the queue, it is theoretically possible to wrap around
989 * the queue twice and mistakenly return IRQ_NONE. Linux only
990 * requires that 0.1% of your interrupts are handled, so this isn't
991 * a big problem.
992 */
82123460 993 if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
a0fa9647 994 return;
b60503ba 995
604e8c8d
KB		 996 if (likely(nvmeq->cq_vector >= 0))
997 writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
b60503ba 998 nvmeq->cq_head = head;
82123460 999 nvmeq->cq_phase = phase;
b60503ba 1000
e9539f47 1001 nvmeq->cqe_seen = 1;
a0fa9647
JA		 1002}
1003
1004static void nvme_process_cq(struct nvme_queue *nvmeq)
1005{
1006 __nvme_process_cq(nvmeq, NULL);
b60503ba
MW		 1007}
1008
1009static irqreturn_t nvme_irq(int irq, void *data)
58ffacb5
MW		 1010{
1011 irqreturn_t result;
1012 struct nvme_queue *nvmeq = data;
1013 spin_lock(&nvmeq->q_lock);
e9539f47
MW		 1014 nvme_process_cq(nvmeq);
1015 result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
1016 nvmeq->cqe_seen = 0;
58ffacb5
MW		 1017 spin_unlock(&nvmeq->q_lock);
1018 return result;
1019}
1020
1021static irqreturn_t nvme_irq_check(int irq, void *data)
1022{
1023 struct nvme_queue *nvmeq = data;
1024 struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head];
1025 if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase)
1026 return IRQ_NONE;
1027 return IRQ_WAKE_THREAD;
1028}
1029
a0fa9647
JA		 1030static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
1031{
1032 struct nvme_queue *nvmeq = hctx->driver_data;
1033
1034 if ((le16_to_cpu(nvmeq->cqes[nvmeq->cq_head].status) & 1) ==
1035 nvmeq->cq_phase) {
1036 spin_lock_irq(&nvmeq->q_lock);
1037 __nvme_process_cq(nvmeq, &tag);
1038 spin_unlock_irq(&nvmeq->q_lock);
1039
1040 if (tag == -1)
1041 return 1;
1042 }
1043
1044 return 0;
1045}
1046
a4aea562
MB		 1047static int nvme_submit_async_admin_req(struct nvme_dev *dev)
1048{
1049 struct nvme_queue *nvmeq = dev->queues[0];
1050 struct nvme_command c;
1051 struct nvme_cmd_info *cmd_info;
1052 struct request *req;
1053
6f3b0e8b
CH		 1054 req = blk_mq_alloc_request(dev->admin_q, WRITE,
1055 BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_RESERVED);
9f173b33
DC		 1056 if (IS_ERR(req))
1057 return PTR_ERR(req);
a4aea562 1058
c917dfe5 1059 req->cmd_flags |= REQ_NO_TIMEOUT;
a4aea562 1060 cmd_info = blk_mq_rq_to_pdu(req);
1efccc9d 1061 nvme_set_info(cmd_info, NULL, async_req_completion);
a4aea562
MB		 1062
1063 memset(&c, 0, sizeof(c));
1064 c.common.opcode = nvme_admin_async_event;
1065 c.common.command_id = req->tag;
1066
42483228 1067 blk_mq_free_request(req);
e3f879bf
SB		 1068 __nvme_submit_cmd(nvmeq, &c);
1069 return 0;
a4aea562
MB		 1070}
1071
1072static int nvme_submit_admin_async_cmd(struct nvme_dev *dev,
4d115420
KB		 1073 struct nvme_command *cmd,
1074 struct async_cmd_info *cmdinfo, unsigned timeout)
1075{
a4aea562
MB		 1076 struct nvme_queue *nvmeq = dev->queues[0];
1077 struct request *req;
1078 struct nvme_cmd_info *cmd_rq;
4d115420 1079
6f3b0e8b 1080 req = blk_mq_alloc_request(dev->admin_q, WRITE, 0);
9f173b33
DC		 1081 if (IS_ERR(req))
1082 return PTR_ERR(req);
a4aea562
MB		 1083
1084 req->timeout = timeout;
1085 cmd_rq = blk_mq_rq_to_pdu(req);
1086 cmdinfo->req = req;
1087 nvme_set_info(cmd_rq, cmdinfo, async_completion);
4d115420 1088 cmdinfo->status = -EINTR;
a4aea562
MB		 1089
1090 cmd->common.command_id = req->tag;
1091
e3f879bf
SB		 1092 nvme_submit_cmd(nvmeq, cmd);
1093 return 0;
4d115420
KB		 1094}
1095
b60503ba
MW		 1096static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
1097{
b60503ba
MW		 1098 struct nvme_command c;
1099
1100 memset(&c, 0, sizeof(c));
1101 c.delete_queue.opcode = opcode;
1102 c.delete_queue.qid = cpu_to_le16(id);
1103
d29ec824 1104 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1105}
1106
1107static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
1108 struct nvme_queue *nvmeq)
1109{
b60503ba
MW		 1110 struct nvme_command c;
1111 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;
1112
d29ec824
CH		 1113 /*
1114 * Note: we (ab)use the fact the the prp fields survive if no data
1115 * is attached to the request.
1116 */
b60503ba
MW		 1117 memset(&c, 0, sizeof(c));
1118 c.create_cq.opcode = nvme_admin_create_cq;
1119 c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr);
1120 c.create_cq.cqid = cpu_to_le16(qid);
1121 c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
1122 c.create_cq.cq_flags = cpu_to_le16(flags);
1123 c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector);
1124
d29ec824 1125 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1126}
1127
1128static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
1129 struct nvme_queue *nvmeq)
1130{
b60503ba
MW		 1131 struct nvme_command c;
1132 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM;
1133
d29ec824
CH		 1134 /*
1135 * Note: we (ab)use the fact the the prp fields survive if no data
1136 * is attached to the request.
1137 */
b60503ba
MW		 1138 memset(&c, 0, sizeof(c));
1139 c.create_sq.opcode = nvme_admin_create_sq;
1140 c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr);
1141 c.create_sq.sqid = cpu_to_le16(qid);
1142 c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
1143 c.create_sq.sq_flags = cpu_to_le16(flags);
1144 c.create_sq.cqid = cpu_to_le16(qid);
1145
d29ec824 1146 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1147}
1148
1149static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid)
1150{
1151 return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid);
1152}
1153
1154static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid)
1155{
1156 return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid);
1157}
1158
c30341dc 1159/**
a4aea562 1160 * nvme_abort_req - Attempt aborting a request
c30341dc
KB		 1161 *
1162 * Schedule controller reset if the command was already aborted once before and
1163 * still hasn't been returned to the driver, or if this is the admin queue.
1164 */
a4aea562 1165static void nvme_abort_req(struct request *req)
c30341dc 1166{
a4aea562
MB		 1167 struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
1168 struct nvme_queue *nvmeq = cmd_rq->nvmeq;
c30341dc 1169 struct nvme_dev *dev = nvmeq->dev;
a4aea562
MB		 1170 struct request *abort_req;
1171 struct nvme_cmd_info *abort_cmd;
1172 struct nvme_command cmd;
c30341dc 1173
a4aea562 1174 if (!nvmeq->qid || cmd_rq->aborted) {
90667892
CH		 1175 spin_lock(&dev_list_lock);
1176 if (!__nvme_reset(dev)) {
1177 dev_warn(dev->dev,
1178 "I/O %d QID %d timeout, reset controller\n",
1179 req->tag, nvmeq->qid);
1180 }
1181 spin_unlock(&dev_list_lock);
c30341dc
KB		 1182 return;
1183 }
1184
1185 if (!dev->abort_limit)
1186 return;
1187
6f3b0e8b
CH		 1188 abort_req = blk_mq_alloc_request(dev->admin_q, WRITE,
1189 BLK_MQ_REQ_NOWAIT);
9f173b33 1190 if (IS_ERR(abort_req))
c30341dc
KB		 1191 return;
1192
a4aea562
MB		 1193 abort_cmd = blk_mq_rq_to_pdu(abort_req);
1194 nvme_set_info(abort_cmd, abort_req, abort_completion);
1195
c30341dc
KB		 1196 memset(&cmd, 0, sizeof(cmd));
1197 cmd.abort.opcode = nvme_admin_abort_cmd;
a4aea562 1198 cmd.abort.cid = req->tag;
c30341dc 1199 cmd.abort.sqid = cpu_to_le16(nvmeq->qid);
a4aea562 1200 cmd.abort.command_id = abort_req->tag;
c30341dc
KB		 1201
1202 --dev->abort_limit;
a4aea562 1203 cmd_rq->aborted = 1;
c30341dc 1204
a4aea562 1205 dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag,
c30341dc 1206 nvmeq->qid);
e3f879bf 1207 nvme_submit_cmd(dev->queues[0], &cmd);
c30341dc
KB		 1208}
1209
42483228 1210static void nvme_cancel_queue_ios(struct request *req, void *data, bool reserved)
a09115b2 1211{
a4aea562
MB		 1212 struct nvme_queue *nvmeq = data;
1213 void *ctx;
1214 nvme_completion_fn fn;
1215 struct nvme_cmd_info *cmd;
cef6a948
KB		 1216 struct nvme_completion cqe;
1217
1218 if (!blk_mq_request_started(req))
1219 return;
a09115b2 1220
a4aea562 1221 cmd = blk_mq_rq_to_pdu(req);
a09115b2 1222
a4aea562
MB		 1223 if (cmd->ctx == CMD_CTX_CANCELLED)
1224 return;
1225
cef6a948
KB		 1226 if (blk_queue_dying(req->q))
1227 cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1);
1228 else
1229 cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);
1230
1231
a4aea562
MB		 1232 dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n",
1233 req->tag, nvmeq->qid);
1234 ctx = cancel_cmd_info(cmd, &fn);
1235 fn(nvmeq, ctx, &cqe);
a09115b2
MW		 1236}
1237
a4aea562 1238static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
9e866774 1239{
a4aea562
MB		 1240 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
1241 struct nvme_queue *nvmeq = cmd->nvmeq;
1242
1243 dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag,
1244 nvmeq->qid);
7a509a6b 1245 spin_lock_irq(&nvmeq->q_lock);
07836e65 1246 nvme_abort_req(req);
7a509a6b 1247 spin_unlock_irq(&nvmeq->q_lock);
a4aea562 1248
07836e65
KB		 1249 /*
1250 * The aborted req will be completed on receiving the abort req.
1251 * We enable the timer again. If hit twice, it'll cause a device reset,
1252 * as the device then is in a faulty state.
1253 */
1254 return BLK_EH_RESET_TIMER;
a4aea562 1255}
22404274 1256
a4aea562
MB		 1257static void nvme_free_queue(struct nvme_queue *nvmeq)
1258{
9e866774
MW		 1259 dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
1260 (void *)nvmeq->cqes, nvmeq->cq_dma_addr);
8ffaadf7
JD		 1261 if (nvmeq->sq_cmds)
1262 dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
9e866774
MW		 1263 nvmeq->sq_cmds, nvmeq->sq_dma_addr);
1264 kfree(nvmeq);
1265}
1266
a1a5ef99 1267static void nvme_free_queues(struct nvme_dev *dev, int lowest)
22404274
KB		 1268{
1269 int i;
1270
a1a5ef99 1271 for (i = dev->queue_count - 1; i >= lowest; i--) {
a4aea562 1272 struct nvme_queue *nvmeq = dev->queues[i];
22404274 1273 dev->queue_count--;
a4aea562 1274 dev->queues[i] = NULL;
f435c282 1275 nvme_free_queue(nvmeq);
121c7ad4 1276 }
22404274
KB		 1277}
1278
4d115420
KB		 1279/**
1280 * nvme_suspend_queue - put queue into suspended state
1281 * @nvmeq - queue to suspend
4d115420
KB		 1282 */
1283static int nvme_suspend_queue(struct nvme_queue *nvmeq)
b60503ba 1284{
2b25d981 1285 int vector;
b60503ba 1286
a09115b2 1287 spin_lock_irq(&nvmeq->q_lock);
2b25d981
KB		 1288 if (nvmeq->cq_vector == -1) {
1289 spin_unlock_irq(&nvmeq->q_lock);
1290 return 1;
1291 }
1292 vector = nvmeq->dev->entry[nvmeq->cq_vector].vector;
42f61420 1293 nvmeq->dev->online_queues--;
2b25d981 1294 nvmeq->cq_vector = -1;
a09115b2
MW		 1295 spin_unlock_irq(&nvmeq->q_lock);
1296
6df3dbc8
KB		 1297 if (!nvmeq->qid && nvmeq->dev->admin_q)
1298 blk_mq_freeze_queue_start(nvmeq->dev->admin_q);
1299
aba2080f
MW		 1300 irq_set_affinity_hint(vector, NULL);
1301 free_irq(vector, nvmeq);
b60503ba 1302
4d115420
KB		 1303 return 0;
1304}
b60503ba 1305
4d115420
KB		 1306static void nvme_clear_queue(struct nvme_queue *nvmeq)
1307{
22404274 1308 spin_lock_irq(&nvmeq->q_lock);
42483228
KB		 1309 if (nvmeq->tags && *nvmeq->tags)
1310 blk_mq_all_tag_busy_iter(*nvmeq->tags, nvme_cancel_queue_ios, nvmeq);
22404274 1311 spin_unlock_irq(&nvmeq->q_lock);
b60503ba
MW		 1312}
1313
4d115420
KB		 1314static void nvme_disable_queue(struct nvme_dev *dev, int qid)
1315{
a4aea562 1316 struct nvme_queue *nvmeq = dev->queues[qid];
4d115420
KB		 1317
1318 if (!nvmeq)
1319 return;
1320 if (nvme_suspend_queue(nvmeq))
1321 return;
1322
0e53d180
KB		 1323 /* Don't tell the adapter to delete the admin queue.
1324 * Don't tell a removed adapter to delete IO queues. */
1325 if (qid && readl(&dev->bar->csts) != -1) {
b60503ba
MW		 1326 adapter_delete_sq(dev, qid);
1327 adapter_delete_cq(dev, qid);
1328 }
07836e65
KB		 1329
1330 spin_lock_irq(&nvmeq->q_lock);
1331 nvme_process_cq(nvmeq);
1332 spin_unlock_irq(&nvmeq->q_lock);
b60503ba
MW		 1333}
1334
8ffaadf7
JD		 1335static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
1336 int entry_size)
1337{
1338 int q_depth = dev->q_depth;
1339 unsigned q_size_aligned = roundup(q_depth * entry_size, dev->page_size);
1340
1341 if (q_size_aligned * nr_io_queues > dev->cmb_size) {
c45f5c99
JD		 1342 u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1343 mem_per_q = round_down(mem_per_q, dev->page_size);
1344 q_depth = div_u64(mem_per_q, entry_size);
8ffaadf7
JD		 1345
1346 /*
1347 * Ensure the reduced q_depth is above some threshold where it
1348 * would be better to map queues in system memory with the
1349 * original depth
1350 */
1351 if (q_depth < 64)
1352 return -ENOMEM;
1353 }
1354
1355 return q_depth;
1356}
1357
1358static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1359 int qid, int depth)
1360{
1361 if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) {
1362 unsigned offset = (qid - 1) *
1363 roundup(SQ_SIZE(depth), dev->page_size);
1364 nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset;
1365 nvmeq->sq_cmds_io = dev->cmb + offset;
1366 } else {
1367 nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth),
1368 &nvmeq->sq_dma_addr, GFP_KERNEL);
1369 if (!nvmeq->sq_cmds)
1370 return -ENOMEM;
1371 }
1372
1373 return 0;
1374}
1375
b60503ba 1376static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
2b25d981 1377 int depth)
b60503ba 1378{
a4aea562 1379 struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
b60503ba
MW		 1380 if (!nvmeq)
1381 return NULL;
1382
e75ec752 1383 nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
4d51abf9 1384 &nvmeq->cq_dma_addr, GFP_KERNEL);
b60503ba
MW		 1385 if (!nvmeq->cqes)
1386 goto free_nvmeq;
b60503ba 1387
8ffaadf7 1388 if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
b60503ba
MW		 1389 goto free_cqdma;
1390
e75ec752 1391 nvmeq->q_dmadev = dev->dev;
091b6092 1392 nvmeq->dev = dev;
3193f07b
MW		 1393 snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
1394 dev->instance, qid);
b60503ba
MW		 1395 spin_lock_init(&nvmeq->q_lock);
1396 nvmeq->cq_head = 0;
82123460 1397 nvmeq->cq_phase = 1;
b80d5ccc 1398 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
b60503ba 1399 nvmeq->q_depth = depth;
c30341dc 1400 nvmeq->qid = qid;
758dd7fd 1401 nvmeq->cq_vector = -1;
a4aea562 1402 dev->queues[qid] = nvmeq;
b60503ba 1403
36a7e993
JD		 1404 /* make sure queue descriptor is set before queue count, for kthread */
1405 mb();
1406 dev->queue_count++;
1407
b60503ba
MW		 1408 return nvmeq;
1409
1410 free_cqdma:
e75ec752 1411 dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
b60503ba
MW		 1412 nvmeq->cq_dma_addr);
1413 free_nvmeq:
1414 kfree(nvmeq);
1415 return NULL;
1416}
1417
3001082c
MW		 1418static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1419 const char *name)
1420{
58ffacb5
MW		 1421 if (use_threaded_interrupts)
1422 return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector,
481e5bad 1423 nvme_irq_check, nvme_irq, IRQF_SHARED,
58ffacb5 1424 name, nvmeq);
3001082c 1425 return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq,
481e5bad 1426 IRQF_SHARED, name, nvmeq);
3001082c
MW		 1427}
1428
22404274 1429static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
b60503ba 1430{
22404274 1431 struct nvme_dev *dev = nvmeq->dev;
b60503ba 1432
7be50e93 1433 spin_lock_irq(&nvmeq->q_lock);
22404274
KB		 1434 nvmeq->sq_tail = 0;
1435 nvmeq->cq_head = 0;
1436 nvmeq->cq_phase = 1;
b80d5ccc 1437 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
22404274 1438 memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
42f61420 1439 dev->online_queues++;
7be50e93 1440 spin_unlock_irq(&nvmeq->q_lock);
22404274
KB		 1441}
1442
1443static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
1444{
1445 struct nvme_dev *dev = nvmeq->dev;
1446 int result;
3f85d50b 1447
2b25d981 1448 nvmeq->cq_vector = qid - 1;
b60503ba
MW		 1449 result = adapter_alloc_cq(dev, qid, nvmeq);
1450 if (result < 0)
22404274 1451 return result;
b60503ba
MW		 1452
1453 result = adapter_alloc_sq(dev, qid, nvmeq);
1454 if (result < 0)
1455 goto release_cq;
1456
3193f07b 1457 result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
b60503ba
MW		 1458 if (result < 0)
1459 goto release_sq;
1460
22404274 1461 nvme_init_queue(nvmeq, qid);
22404274 1462 return result;
b60503ba
MW		 1463
1464 release_sq:
1465 adapter_delete_sq(dev, qid);
1466 release_cq:
1467 adapter_delete_cq(dev, qid);
22404274 1468 return result;
b60503ba
MW		 1469}
1470
ba47e386
MW		 1471static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled)
1472{
1473 unsigned long timeout;
1474 u32 bit = enabled ? NVME_CSTS_RDY : 0;
1475
1476 timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1477
1478 while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) {
1479 msleep(100);
1480 if (fatal_signal_pending(current))
1481 return -EINTR;
1482 if (time_after(jiffies, timeout)) {
e75ec752 1483 dev_err(dev->dev,
27e8166c
MW		 1484 "Device not ready; aborting %s\n", enabled ?
1485 "initialisation" : "reset");
ba47e386
MW		 1486 return -ENODEV;
1487 }
1488 }
1489
1490 return 0;
1491}
1492
1493/*
1494 * If the device has been passed off to us in an enabled state, just clear
1495 * the enabled bit. The spec says we should set the 'shutdown notification
1496 * bits', but doing so may cause the device to complete commands to the
1497 * admin queue ... and we don't know what memory that might be pointing at!
1498 */
1499static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap)
1500{
01079522
DM		 1501 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1502 dev->ctrl_config &= ~NVME_CC_ENABLE;
1503 writel(dev->ctrl_config, &dev->bar->cc);
44af146a 1504
ba47e386
MW		 1505 return nvme_wait_ready(dev, cap, false);
1506}
1507
1508static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap)
1509{
01079522
DM		 1510 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1511 dev->ctrl_config |= NVME_CC_ENABLE;
1512 writel(dev->ctrl_config, &dev->bar->cc);
1513
ba47e386
MW		 1514 return nvme_wait_ready(dev, cap, true);
1515}
1516
1894d8f1
KB		 1517static int nvme_shutdown_ctrl(struct nvme_dev *dev)
1518{
1519 unsigned long timeout;
1894d8f1 1520
01079522
DM		 1521 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1522 dev->ctrl_config |= NVME_CC_SHN_NORMAL;
1523
1524 writel(dev->ctrl_config, &dev->bar->cc);
1894d8f1 1525
2484f407 1526 timeout = SHUTDOWN_TIMEOUT + jiffies;
1894d8f1
KB		 1527 while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) !=
1528 NVME_CSTS_SHST_CMPLT) {
1529 msleep(100);
1530 if (fatal_signal_pending(current))
1531 return -EINTR;
1532 if (time_after(jiffies, timeout)) {
e75ec752 1533 dev_err(dev->dev,
1894d8f1
KB		 1534 "Device shutdown incomplete; abort shutdown\n");
1535 return -ENODEV;
1536 }
1537 }
1538
1539 return 0;
1540}
1541
a4aea562 1542static struct blk_mq_ops nvme_mq_admin_ops = {
d29ec824 1543 .queue_rq = nvme_queue_rq,
a4aea562
MB		 1544 .map_queue = blk_mq_map_queue,
1545 .init_hctx = nvme_admin_init_hctx,
4af0e21c 1546 .exit_hctx = nvme_admin_exit_hctx,
a4aea562
MB		 1547 .init_request = nvme_admin_init_request,
1548 .timeout = nvme_timeout,
1549};
1550
1551static struct blk_mq_ops nvme_mq_ops = {
1552 .queue_rq = nvme_queue_rq,
1553 .map_queue = blk_mq_map_queue,
1554 .init_hctx = nvme_init_hctx,
1555 .init_request = nvme_init_request,
1556 .timeout = nvme_timeout,
a0fa9647 1557 .poll = nvme_poll,
a4aea562
MB		 1558};
1559
ea191d2f
KB		 1560static void nvme_dev_remove_admin(struct nvme_dev *dev)
1561{
1562 if (dev->admin_q && !blk_queue_dying(dev->admin_q)) {
1563 blk_cleanup_queue(dev->admin_q);
1564 blk_mq_free_tag_set(&dev->admin_tagset);
1565 }
1566}
1567
a4aea562
MB		 1568static int nvme_alloc_admin_tags(struct nvme_dev *dev)
1569{
1570 if (!dev->admin_q) {
1571 dev->admin_tagset.ops = &nvme_mq_admin_ops;
1572 dev->admin_tagset.nr_hw_queues = 1;
1573 dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1;
1efccc9d 1574 dev->admin_tagset.reserved_tags = 1;
a4aea562 1575 dev->admin_tagset.timeout = ADMIN_TIMEOUT;
e75ec752 1576 dev->admin_tagset.numa_node = dev_to_node(dev->dev);
ac3dd5bd 1577 dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
a4aea562
MB		 1578 dev->admin_tagset.driver_data = dev;
1579
1580 if (blk_mq_alloc_tag_set(&dev->admin_tagset))
1581 return -ENOMEM;
1582
1583 dev->admin_q = blk_mq_init_queue(&dev->admin_tagset);
35b489d3 1584 if (IS_ERR(dev->admin_q)) {
a4aea562
MB		 1585 blk_mq_free_tag_set(&dev->admin_tagset);
1586 return -ENOMEM;
1587 }
ea191d2f
KB		 1588 if (!blk_get_queue(dev->admin_q)) {
1589 nvme_dev_remove_admin(dev);
4af0e21c 1590 dev->admin_q = NULL;
ea191d2f
KB		 1591 return -ENODEV;
1592 }
0fb59cbc
KB		 1593 } else
1594 blk_mq_unfreeze_queue(dev->admin_q);
a4aea562
MB		 1595
1596 return 0;
1597}
1598
8d85fce7 1599static int nvme_configure_admin_queue(struct nvme_dev *dev)
b60503ba 1600{
ba47e386 1601 int result;
b60503ba 1602 u32 aqa;
a310acd7 1603 u64 cap = lo_hi_readq(&dev->bar->cap);
b60503ba 1604 struct nvme_queue *nvmeq;
c5c9f25b
NA		 1605 /*
1606 * default to a 4K page size, with the intention to update this
1607 * path in the future to accomodate architectures with differing
1608 * kernel and IO page sizes.
1609 */
1610 unsigned page_shift = 12;
1d090624 1611 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12;
1d090624
KB		 1612
1613 if (page_shift < dev_page_min) {
e75ec752 1614 dev_err(dev->dev,
1d090624
KB		 1615 "Minimum device page size (%u) too large for "
1616 "host (%u)\n", 1 << dev_page_min,
1617 1 << page_shift);
1618 return -ENODEV;
1619 }
b60503ba 1620
dfbac8c7
KB		 1621 dev->subsystem = readl(&dev->bar->vs) >= NVME_VS(1, 1) ?
1622 NVME_CAP_NSSRC(cap) : 0;
1623
1624 if (dev->subsystem && (readl(&dev->bar->csts) & NVME_CSTS_NSSRO))
1625 writel(NVME_CSTS_NSSRO, &dev->bar->csts);
1626
ba47e386
MW		 1627 result = nvme_disable_ctrl(dev, cap);
1628 if (result < 0)
1629 return result;
b60503ba 1630
a4aea562 1631 nvmeq = dev->queues[0];
cd638946 1632 if (!nvmeq) {
2b25d981 1633 nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
cd638946
KB		 1634 if (!nvmeq)
1635 return -ENOMEM;
cd638946 1636 }
b60503ba
MW		 1637
1638 aqa = nvmeq->q_depth - 1;
1639 aqa |= aqa << 16;
1640
1d090624
KB		 1641 dev->page_size = 1 << page_shift;
1642
01079522 1643 dev->ctrl_config = NVME_CC_CSS_NVM;
1d090624 1644 dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
b60503ba 1645 dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
7f53f9d2 1646 dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
b60503ba
MW		 1647
1648 writel(aqa, &dev->bar->aqa);
a310acd7
SG		 1649 lo_hi_writeq(nvmeq->sq_dma_addr, &dev->bar->asq);
1650 lo_hi_writeq(nvmeq->cq_dma_addr, &dev->bar->acq);
b60503ba 1651
ba47e386 1652 result = nvme_enable_ctrl(dev, cap);
025c557a 1653 if (result)
a4aea562
MB		 1654 goto free_nvmeq;
1655
2b25d981 1656 nvmeq->cq_vector = 0;
3193f07b 1657 result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
758dd7fd
JD		 1658 if (result) {
1659 nvmeq->cq_vector = -1;
0fb59cbc 1660 goto free_nvmeq;
758dd7fd 1661 }
025c557a 1662
b60503ba 1663 return result;
a4aea562 1664
a4aea562
MB		 1665 free_nvmeq:
1666 nvme_free_queues(dev, 0);
1667 return result;
b60503ba
MW		 1668}
1669
a53295b6
MW		 1670static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1671{
1672 struct nvme_dev *dev = ns->dev;
a53295b6
MW		 1673 struct nvme_user_io io;
1674 struct nvme_command c;
d29ec824 1675 unsigned length, meta_len;
a67a9513 1676 int status, write;
a67a9513
KB		 1677 dma_addr_t meta_dma = 0;
1678 void *meta = NULL;
fec558b5 1679 void __user *metadata;
a53295b6
MW		 1680
1681 if (copy_from_user(&io, uio, sizeof(io)))
1682 return -EFAULT;
6c7d4945
MW		 1683
1684 switch (io.opcode) {
1685 case nvme_cmd_write:
1686 case nvme_cmd_read:
6bbf1acd 1687 case nvme_cmd_compare:
6413214c 1688 break;
6c7d4945 1689 default:
6bbf1acd 1690 return -EINVAL;
6c7d4945
MW		 1691 }
1692
d29ec824
CH		 1693 length = (io.nblocks + 1) << ns->lba_shift;
1694 meta_len = (io.nblocks + 1) * ns->ms;
835da3f9 1695 metadata = (void __user *)(uintptr_t)io.metadata;
d29ec824 1696 write = io.opcode & 1;
a53295b6 1697
71feb364
KB		 1698 if (ns->ext) {
1699 length += meta_len;
1700 meta_len = 0;
a67a9513
KB		 1701 }
1702 if (meta_len) {
d29ec824
CH		 1703 if (((io.metadata & 3) || !io.metadata) && !ns->ext)
1704 return -EINVAL;
1705
e75ec752 1706 meta = dma_alloc_coherent(dev->dev, meta_len,
a67a9513 1707 &meta_dma, GFP_KERNEL);
fec558b5 1708
a67a9513
KB		 1709 if (!meta) {
1710 status = -ENOMEM;
1711 goto unmap;
1712 }
1713 if (write) {
fec558b5 1714 if (copy_from_user(meta, metadata, meta_len)) {
a67a9513
KB		 1715 status = -EFAULT;
1716 goto unmap;
1717 }
1718 }
1719 }
1720
a53295b6
MW		 1721 memset(&c, 0, sizeof(c));
1722 c.rw.opcode = io.opcode;
1723 c.rw.flags = io.flags;
6c7d4945 1724 c.rw.nsid = cpu_to_le32(ns->ns_id);
a53295b6 1725 c.rw.slba = cpu_to_le64(io.slba);
6c7d4945 1726 c.rw.length = cpu_to_le16(io.nblocks);
a53295b6 1727 c.rw.control = cpu_to_le16(io.control);
1c9b5265
MW		 1728 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1729 c.rw.reftag = cpu_to_le32(io.reftag);
1730 c.rw.apptag = cpu_to_le16(io.apptag);
1731 c.rw.appmask = cpu_to_le16(io.appmask);
a67a9513 1732 c.rw.metadata = cpu_to_le64(meta_dma);
d29ec824
CH		 1733
1734 status = __nvme_submit_sync_cmd(ns->queue, &c, NULL,
835da3f9 1735 (void __user *)(uintptr_t)io.addr, length, NULL, 0);
f410c680 1736 unmap:
a67a9513
KB		 1737 if (meta) {
1738 if (status == NVME_SC_SUCCESS && !write) {
fec558b5 1739 if (copy_to_user(metadata, meta, meta_len))
a67a9513
KB		 1740 status = -EFAULT;
1741 }
e75ec752 1742 dma_free_coherent(dev->dev, meta_len, meta, meta_dma);
f410c680 1743 }
a53295b6
MW		 1744 return status;
1745}
1746
a4aea562
MB		 1747static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns,
1748 struct nvme_passthru_cmd __user *ucmd)
6ee44cdc 1749{
7963e521 1750 struct nvme_passthru_cmd cmd;
6ee44cdc 1751 struct nvme_command c;
d29ec824
CH		 1752 unsigned timeout = 0;
1753 int status;
6ee44cdc 1754
6bbf1acd
MW		 1755 if (!capable(CAP_SYS_ADMIN))
1756 return -EACCES;
1757 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
6ee44cdc 1758 return -EFAULT;
6ee44cdc
MW		 1759
1760 memset(&c, 0, sizeof(c));
6bbf1acd
MW		 1761 c.common.opcode = cmd.opcode;
1762 c.common.flags = cmd.flags;
1763 c.common.nsid = cpu_to_le32(cmd.nsid);
1764 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1765 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1766 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1767 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1768 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1769 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1770 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1771 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1772
d29ec824
CH		 1773 if (cmd.timeout_ms)
1774 timeout = msecs_to_jiffies(cmd.timeout_ms);
eca18b23 1775
f705f837 1776 status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c,
835da3f9 1777 NULL, (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
d29ec824
CH		 1778 &cmd.result, timeout);
1779 if (status >= 0) {
1780 if (put_user(cmd.result, &ucmd->result))
1781 return -EFAULT;
6bbf1acd 1782 }
f4f117f6 1783
6ee44cdc
MW		 1784 return status;
1785}
1786
81f03fed
JD		 1787static int nvme_subsys_reset(struct nvme_dev *dev)
1788{
1789 if (!dev->subsystem)
1790 return -ENOTTY;
1791
1792 writel(0x4E564D65, &dev->bar->nssr); /* "NVMe" */
1793 return 0;
1794}
1795
b60503ba
MW		 1796static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd,
1797 unsigned long arg)
1798{
1799 struct nvme_ns *ns = bdev->bd_disk->private_data;
1800
1801 switch (cmd) {
6bbf1acd 1802 case NVME_IOCTL_ID:
c3bfe717 1803 force_successful_syscall_return();
6bbf1acd
MW		 1804 return ns->ns_id;
1805 case NVME_IOCTL_ADMIN_CMD:
a4aea562 1806 return nvme_user_cmd(ns->dev, NULL, (void __user *)arg);
7963e521 1807 case NVME_IOCTL_IO_CMD:
a4aea562 1808 return nvme_user_cmd(ns->dev, ns, (void __user *)arg);
a53295b6
MW		 1809 case NVME_IOCTL_SUBMIT_IO:
1810 return nvme_submit_io(ns, (void __user *)arg);
5d0f6131
VV		 1811 case SG_GET_VERSION_NUM:
1812 return nvme_sg_get_version_num((void __user *)arg);
1813 case SG_IO:
1814 return nvme_sg_io(ns, (void __user *)arg);
b60503ba
MW		 1815 default:
1816 return -ENOTTY;
1817 }
1818}
1819
320a3827
KB		 1820#ifdef CONFIG_COMPAT
1821static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1822 unsigned int cmd, unsigned long arg)
1823{
320a3827
KB		 1824 switch (cmd) {
1825 case SG_IO:
e179729a 1826 return -ENOIOCTLCMD;
320a3827
KB		 1827 }
1828 return nvme_ioctl(bdev, mode, cmd, arg);
1829}
1830#else
1831#define nvme_compat_ioctl NULL
1832#endif
1833
5105aa55 1834static void nvme_free_dev(struct kref *kref);
188c3568
KB		 1835static void nvme_free_ns(struct kref *kref)
1836{
1837 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
1838
ca064085
MB		 1839 if (ns->type == NVME_NS_LIGHTNVM)
1840 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
1841
188c3568
KB		 1842 spin_lock(&dev_list_lock);
1843 ns->disk->private_data = NULL;
1844 spin_unlock(&dev_list_lock);
1845
5105aa55 1846 kref_put(&ns->dev->kref, nvme_free_dev);
188c3568
KB		 1847 put_disk(ns->disk);
1848 kfree(ns);
1849}
1850
9ac27090
KB		 1851static int nvme_open(struct block_device *bdev, fmode_t mode)
1852{
9e60352c
KB		 1853 int ret = 0;
1854 struct nvme_ns *ns;
9ac27090 1855
9e60352c
KB		 1856 spin_lock(&dev_list_lock);
1857 ns = bdev->bd_disk->private_data;
1858 if (!ns)
1859 ret = -ENXIO;
188c3568 1860 else if (!kref_get_unless_zero(&ns->kref))
9e60352c
KB		 1861 ret = -ENXIO;
1862 spin_unlock(&dev_list_lock);
1863
1864 return ret;
9ac27090
KB		 1865}
1866
9ac27090
KB		 1867static void nvme_release(struct gendisk *disk, fmode_t mode)
1868{
1869 struct nvme_ns *ns = disk->private_data;
188c3568 1870 kref_put(&ns->kref, nvme_free_ns);
9ac27090
KB		 1871}
1872
4cc09e2d
KB		 1873static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo)
1874{
1875 /* some standard values */
1876 geo->heads = 1 << 6;
1877 geo->sectors = 1 << 5;
1878 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1879 return 0;
1880}
1881
e1e5e564
KB		 1882static void nvme_config_discard(struct nvme_ns *ns)
1883{
1884 u32 logical_block_size = queue_logical_block_size(ns->queue);
1885 ns->queue->limits.discard_zeroes_data = 0;
1886 ns->queue->limits.discard_alignment = logical_block_size;
1887 ns->queue->limits.discard_granularity = logical_block_size;
2bb4cd5c 1888 blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
e1e5e564
KB		 1889 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1890}
1891
1b9dbf7f
KB		 1892static int nvme_revalidate_disk(struct gendisk *disk)
1893{
1894 struct nvme_ns *ns = disk->private_data;
1895 struct nvme_dev *dev = ns->dev;
1896 struct nvme_id_ns *id;
a67a9513
KB		 1897 u8 lbaf, pi_type;
1898 u16 old_ms;
e1e5e564 1899 unsigned short bs;
1b9dbf7f 1900
d29ec824 1901 if (nvme_identify_ns(dev, ns->ns_id, &id)) {
a5768aa8
KB		 1902 dev_warn(dev->dev, "%s: Identify failure nvme%dn%d\n", __func__,
1903 dev->instance, ns->ns_id);
1904 return -ENODEV;
1b9dbf7f 1905 }
a5768aa8
KB		 1906 if (id->ncap == 0) {
1907 kfree(id);
1908 return -ENODEV;
e1e5e564 1909 }
1b9dbf7f 1910
ca064085
MB		 1911 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
1912 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
1913 dev_warn(dev->dev,
1914 "%s: LightNVM init failure\n", __func__);
1915 kfree(id);
1916 return -ENODEV;
1917 }
1918 ns->type = NVME_NS_LIGHTNVM;
1919 }
1920
e1e5e564
KB		 1921 old_ms = ns->ms;
1922 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1b9dbf7f 1923 ns->lba_shift = id->lbaf[lbaf].ds;
e1e5e564 1924 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
a67a9513 1925 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
e1e5e564
KB		 1926
1927 /*
1928 * If identify namespace failed, use default 512 byte block size so
1929 * block layer can use before failing read/write for 0 capacity.
1930 */
1931 if (ns->lba_shift == 0)
1932 ns->lba_shift = 9;
1933 bs = 1 << ns->lba_shift;
1934
1935 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
1936 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
1937 id->dps & NVME_NS_DPS_PI_MASK : 0;
1938
4cfc766e 1939 blk_mq_freeze_queue(disk->queue);
52b68d7e
KB		 1940 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
1941 ns->ms != old_ms ||
e1e5e564 1942 bs != queue_logical_block_size(disk->queue) ||
a67a9513 1943 (ns->ms && ns->ext)))
e1e5e564
KB		 1944 blk_integrity_unregister(disk);
1945
1946 ns->pi_type = pi_type;
1947 blk_queue_logical_block_size(ns->queue, bs);
1948
25520d55 1949 if (ns->ms && !ns->ext)
e1e5e564
KB		 1950 nvme_init_integrity(ns);
1951
ca064085
MB		 1952 if ((ns->ms && !(ns->ms == 8 && ns->pi_type) &&
1953 !blk_get_integrity(disk)) ||
1954 ns->type == NVME_NS_LIGHTNVM)
e1e5e564
KB		 1955 set_capacity(disk, 0);
1956 else
1957 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1958
1959 if (dev->oncs & NVME_CTRL_ONCS_DSM)
1960 nvme_config_discard(ns);
4cfc766e 1961 blk_mq_unfreeze_queue(disk->queue);
1b9dbf7f 1962
d29ec824 1963 kfree(id);
1b9dbf7f
KB		 1964 return 0;
1965}
1966
1d277a63
KB		 1967static char nvme_pr_type(enum pr_type type)
1968{
1969 switch (type) {
1970 case PR_WRITE_EXCLUSIVE:
1971 return 1;
1972 case PR_EXCLUSIVE_ACCESS:
1973 return 2;
1974 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1975 return 3;
1976 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1977 return 4;
1978 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1979 return 5;
1980 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1981 return 6;
1982 default:
1983 return 0;
1984 }
1985};
1986
1987static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1988 u64 key, u64 sa_key, u8 op)
1989{
1990 struct nvme_ns *ns = bdev->bd_disk->private_data;
1991 struct nvme_command c;
1992 u8 data[16] = { 0, };
1993
1994 put_unaligned_le64(key, &data[0]);
1995 put_unaligned_le64(sa_key, &data[8]);
1996
1997 memset(&c, 0, sizeof(c));
1998 c.common.opcode = op;
a6dd1020
CH		 1999 c.common.nsid = cpu_to_le32(ns->ns_id);
2000 c.common.cdw10[0] = cpu_to_le32(cdw10);
1d277a63
KB		 2001
2002 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2003}
2004
2005static int nvme_pr_register(struct block_device *bdev, u64 old,
2006 u64 new, unsigned flags)
2007{
2008 u32 cdw10;
2009
2010 if (flags & ~PR_FL_IGNORE_KEY)
2011 return -EOPNOTSUPP;
2012
2013 cdw10 = old ? 2 : 0;
2014 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2015 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2016 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2017}
2018
2019static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2020 enum pr_type type, unsigned flags)
2021{
2022 u32 cdw10;
2023
2024 if (flags & ~PR_FL_IGNORE_KEY)
2025 return -EOPNOTSUPP;
2026
2027 cdw10 = nvme_pr_type(type) << 8;
2028 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2029 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2030}
2031
2032static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2033 enum pr_type type, bool abort)
2034{
2035 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
2036 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2037}
2038
2039static int nvme_pr_clear(struct block_device *bdev, u64 key)
2040{
73fcf4e2 2041 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1d277a63
KB		 2042 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2043}
2044
2045static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2046{
2047 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
2048 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2049}
2050
2051static const struct pr_ops nvme_pr_ops = {
2052 .pr_register = nvme_pr_register,
2053 .pr_reserve = nvme_pr_reserve,
2054 .pr_release = nvme_pr_release,
2055 .pr_preempt = nvme_pr_preempt,
2056 .pr_clear = nvme_pr_clear,
2057};
2058
b60503ba
MW		 2059static const struct block_device_operations nvme_fops = {
2060 .owner = THIS_MODULE,
2061 .ioctl = nvme_ioctl,
320a3827 2062 .compat_ioctl = nvme_compat_ioctl,
9ac27090
KB		 2063 .open = nvme_open,
2064 .release = nvme_release,
4cc09e2d 2065 .getgeo = nvme_getgeo,
1b9dbf7f 2066 .revalidate_disk= nvme_revalidate_disk,
1d277a63 2067 .pr_ops = &nvme_pr_ops,
b60503ba
MW		 2068};
2069
1fa6aead
MW		 2070static int nvme_kthread(void *data)
2071{
d4b4ff8e 2072 struct nvme_dev *dev, *next;
1fa6aead
MW		 2073
2074 while (!kthread_should_stop()) {
564a232c 2075 set_current_state(TASK_INTERRUPTIBLE);
1fa6aead 2076 spin_lock(&dev_list_lock);
d4b4ff8e 2077 list_for_each_entry_safe(dev, next, &dev_list, node) {
1fa6aead 2078 int i;
dfbac8c7
KB		 2079 u32 csts = readl(&dev->bar->csts);
2080
2081 if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
2082 csts & NVME_CSTS_CFS) {
90667892
CH		 2083 if (!__nvme_reset(dev)) {
2084 dev_warn(dev->dev,
2085 "Failed status: %x, reset controller\n",
2086 readl(&dev->bar->csts));
2087 }
d4b4ff8e
KB		 2088 continue;
2089 }
1fa6aead 2090 for (i = 0; i < dev->queue_count; i++) {
a4aea562 2091 struct nvme_queue *nvmeq = dev->queues[i];
740216fc
MW		 2092 if (!nvmeq)
2093 continue;
1fa6aead 2094 spin_lock_irq(&nvmeq->q_lock);
bc57a0f7 2095 nvme_process_cq(nvmeq);
6fccf938
KB		 2096
2097 while ((i == 0) && (dev->event_limit > 0)) {
a4aea562 2098 if (nvme_submit_async_admin_req(dev))
6fccf938
KB		 2099 break;
2100 dev->event_limit--;
2101 }
1fa6aead
MW		 2102 spin_unlock_irq(&nvmeq->q_lock);
2103 }
2104 }
2105 spin_unlock(&dev_list_lock);
acb7aa0d 2106 schedule_timeout(round_jiffies_relative(HZ));
1fa6aead
MW		 2107 }
2108 return 0;
2109}
2110
e1e5e564 2111static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
b60503ba
MW		 2112{
2113 struct nvme_ns *ns;
2114 struct gendisk *disk;
e75ec752 2115 int node = dev_to_node(dev->dev);
b60503ba 2116
a4aea562 2117 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
b60503ba 2118 if (!ns)
e1e5e564
KB		 2119 return;
2120
a4aea562 2121 ns->queue = blk_mq_init_queue(&dev->tagset);
9f173b33 2122 if (IS_ERR(ns->queue))
b60503ba 2123 goto out_free_ns;
4eeb9215
MW		 2124 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
2125 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
b60503ba
MW		 2126 ns->dev = dev;
2127 ns->queue->queuedata = ns;
2128
a4aea562 2129 disk = alloc_disk_node(0, node);
b60503ba
MW		 2130 if (!disk)
2131 goto out_free_queue;
a4aea562 2132
188c3568 2133 kref_init(&ns->kref);
5aff9382 2134 ns->ns_id = nsid;
b60503ba 2135 ns->disk = disk;
e1e5e564
KB		 2136 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2137 list_add_tail(&ns->list, &dev->namespaces);
2138
e9ef4636 2139 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
e824410f 2140 if (dev->max_hw_sectors) {
8fc23e03 2141 blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors);
e824410f 2142 blk_queue_max_segments(ns->queue,
6824c5ef 2143 (dev->max_hw_sectors / (dev->page_size >> 9)) + 1);
e824410f 2144 }
a4aea562
MB		 2145 if (dev->stripe_size)
2146 blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9);
a7d2ce28
KB		 2147 if (dev->vwc & NVME_CTRL_VWC_PRESENT)
2148 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
03100aad 2149 blk_queue_virt_boundary(ns->queue, dev->page_size - 1);
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
5105aa55 2170 kref_get(&dev->kref);
ca064085
MB		 2171 if (ns->type != NVME_NS_LIGHTNVM) {
2172 add_disk(ns->disk);
2173 if (ns->ms) {
2174 struct block_device *bd = bdget_disk(ns->disk, 0);
2175 if (!bd)
2176 return;
2177 if (blkdev_get(bd, FMODE_READ, NULL)) {
2178 bdput(bd);
2179 return;
2180 }
2181 blkdev_reread_part(bd);
2182 blkdev_put(bd, FMODE_READ);
7bee6074 2183 }
7bee6074 2184 }
e1e5e564 2185 return;
a5768aa8
KB		 2186 out_free_disk:
2187 kfree(disk);
2188 list_del(&ns->list);
b60503ba
MW		 2189 out_free_queue:
2190 blk_cleanup_queue(ns->queue);
2191 out_free_ns:
2192 kfree(ns);
b60503ba
MW		 2193}
2194
2659e57b
CH		 2195/*
2196 * Create I/O queues. Failing to create an I/O queue is not an issue,
2197 * we can continue with less than the desired amount of queues, and
2198 * even a controller without I/O queues an still be used to issue
2199 * admin commands. This might be useful to upgrade a buggy firmware
2200 * for example.
2201 */
42f61420
KB		 2202static void nvme_create_io_queues(struct nvme_dev *dev)
2203{
a4aea562 2204 unsigned i;
42f61420 2205
a4aea562 2206 for (i = dev->queue_count; i <= dev->max_qid; i++)
2b25d981 2207 if (!nvme_alloc_queue(dev, i, dev->q_depth))
42f61420
KB		 2208 break;
2209
a4aea562 2210 for (i = dev->online_queues; i <= dev->queue_count - 1; i++)
2659e57b
CH		 2211 if (nvme_create_queue(dev->queues[i], i)) {
2212 nvme_free_queues(dev, i);
42f61420 2213 break;
2659e57b 2214 }
42f61420
KB		 2215}
2216
b3b06812 2217static int set_queue_count(struct nvme_dev *dev, int count)
b60503ba
MW		 2218{
2219 int status;
2220 u32 result;
b3b06812 2221 u32 q_count = (count - 1) | ((count - 1) << 16);
b60503ba 2222
df348139 2223 status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0,
bc5fc7e4 2224 &result);
27e8166c
MW		 2225 if (status < 0)
2226 return status;
2227 if (status > 0) {
e75ec752 2228 dev_err(dev->dev, "Could not set queue count (%d)\n", status);
badc34d4 2229 return 0;
27e8166c 2230 }
b60503ba
MW		 2231 return min(result & 0xffff, result >> 16) + 1;
2232}
2233
8ffaadf7
JD		 2234static void __iomem *nvme_map_cmb(struct nvme_dev *dev)
2235{
2236 u64 szu, size, offset;
2237 u32 cmbloc;
2238 resource_size_t bar_size;
2239 struct pci_dev *pdev = to_pci_dev(dev->dev);
2240 void __iomem *cmb;
2241 dma_addr_t dma_addr;
2242
2243 if (!use_cmb_sqes)
2244 return NULL;
2245
2246 dev->cmbsz = readl(&dev->bar->cmbsz);
2247 if (!(NVME_CMB_SZ(dev->cmbsz)))
2248 return NULL;
2249
2250 cmbloc = readl(&dev->bar->cmbloc);
2251
2252 szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
2253 size = szu * NVME_CMB_SZ(dev->cmbsz);
2254 offset = szu * NVME_CMB_OFST(cmbloc);
2255 bar_size = pci_resource_len(pdev, NVME_CMB_BIR(cmbloc));
2256
2257 if (offset > bar_size)
2258 return NULL;
2259
2260 /*
2261 * Controllers may support a CMB size larger than their BAR,
2262 * for example, due to being behind a bridge. Reduce the CMB to
2263 * the reported size of the BAR
2264 */
2265 if (size > bar_size - offset)
2266 size = bar_size - offset;
2267
2268 dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(cmbloc)) + offset;
2269 cmb = ioremap_wc(dma_addr, size);
2270 if (!cmb)
2271 return NULL;
2272
2273 dev->cmb_dma_addr = dma_addr;
2274 dev->cmb_size = size;
2275 return cmb;
2276}
2277
2278static inline void nvme_release_cmb(struct nvme_dev *dev)
2279{
2280 if (dev->cmb) {
2281 iounmap(dev->cmb);
2282 dev->cmb = NULL;
2283 }
2284}
2285
9d713c2b
KB		 2286static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
2287{
b80d5ccc 2288 return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
9d713c2b
KB		 2289}
2290
8d85fce7 2291static int nvme_setup_io_queues(struct nvme_dev *dev)
b60503ba 2292{
a4aea562 2293 struct nvme_queue *adminq = dev->queues[0];
e75ec752 2294 struct pci_dev *pdev = to_pci_dev(dev->dev);
42f61420 2295 int result, i, vecs, nr_io_queues, size;
b60503ba 2296
42f61420 2297 nr_io_queues = num_possible_cpus();
b348b7d5 2298 result = set_queue_count(dev, nr_io_queues);
badc34d4 2299 if (result <= 0)
1b23484b 2300 return result;
b348b7d5
MW		 2301 if (result < nr_io_queues)
2302 nr_io_queues = result;
b60503ba 2303
8ffaadf7
JD		 2304 if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) {
2305 result = nvme_cmb_qdepth(dev, nr_io_queues,
2306 sizeof(struct nvme_command));
2307 if (result > 0)
2308 dev->q_depth = result;
2309 else
2310 nvme_release_cmb(dev);
2311 }
2312
9d713c2b
KB		 2313 size = db_bar_size(dev, nr_io_queues);
2314 if (size > 8192) {
f1938f6e 2315 iounmap(dev->bar);
9d713c2b
KB		 2316 do {
2317 dev->bar = ioremap(pci_resource_start(pdev, 0), size);
2318 if (dev->bar)
2319 break;
2320 if (!--nr_io_queues)
2321 return -ENOMEM;
2322 size = db_bar_size(dev, nr_io_queues);
2323 } while (1);
f1938f6e 2324 dev->dbs = ((void __iomem *)dev->bar) + 4096;
5a92e700 2325 adminq->q_db = dev->dbs;
f1938f6e
MW		 2326 }
2327
9d713c2b 2328 /* Deregister the admin queue's interrupt */
3193f07b 2329 free_irq(dev->entry[0].vector, adminq);
9d713c2b 2330
e32efbfc
JA		 2331 /*
2332 * If we enable msix early due to not intx, disable it again before
2333 * setting up the full range we need.
2334 */
2335 if (!pdev->irq)
2336 pci_disable_msix(pdev);
2337
be577fab 2338 for (i = 0; i < nr_io_queues; i++)
1b23484b 2339 dev->entry[i].entry = i;
be577fab
AG		 2340 vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues);
2341 if (vecs < 0) {
2342 vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32));
2343 if (vecs < 0) {
2344 vecs = 1;
2345 } else {
2346 for (i = 0; i < vecs; i++)
2347 dev->entry[i].vector = i + pdev->irq;
fa08a396
RRG		 2348 }
2349 }
2350
063a8096
MW		 2351 /*
2352 * Should investigate if there's a performance win from allocating
2353 * more queues than interrupt vectors; it might allow the submission
2354 * path to scale better, even if the receive path is limited by the
2355 * number of interrupts.
2356 */
2357 nr_io_queues = vecs;
42f61420 2358 dev->max_qid = nr_io_queues;
063a8096 2359
3193f07b 2360 result = queue_request_irq(dev, adminq, adminq->irqname);
758dd7fd
JD		 2361 if (result) {
2362 adminq->cq_vector = -1;
22404274 2363 goto free_queues;
758dd7fd 2364 }
1b23484b 2365
cd638946 2366 /* Free previously allocated queues that are no longer usable */
42f61420 2367 nvme_free_queues(dev, nr_io_queues + 1);
a4aea562 2368 nvme_create_io_queues(dev);
9ecdc946 2369
22404274 2370 return 0;
b60503ba 2371
22404274 2372 free_queues:
a1a5ef99 2373 nvme_free_queues(dev, 1);
22404274 2374 return result;
b60503ba
MW		 2375}
2376
a5768aa8
KB		 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);
9609b994 2410 if (ns->disk->flags & GENHD_FL_UP)
a5768aa8 2411 del_gendisk(ns->disk);
a5768aa8
KB		 2412 if (kill || !blk_queue_dying(ns->queue)) {
2413 blk_mq_abort_requeue_list(ns->queue);
2414 blk_cleanup_queue(ns->queue);
5105aa55
KB		 2415 }
2416 list_del_init(&ns->list);
2417 kref_put(&ns->kref, nvme_free_ns);
a5768aa8
KB		 2418}
2419
2420static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn)
2421{
2422 struct nvme_ns *ns, *next;
2423 unsigned i;
2424
2425 for (i = 1; i <= nn; i++) {
2426 ns = nvme_find_ns(dev, i);
2427 if (ns) {
5105aa55 2428 if (revalidate_disk(ns->disk))
a5768aa8 2429 nvme_ns_remove(ns);
a5768aa8
KB		 2430 } else
2431 nvme_alloc_ns(dev, i);
2432 }
2433 list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
5105aa55 2434 if (ns->ns_id > nn)
a5768aa8 2435 nvme_ns_remove(ns);
a5768aa8
KB		 2436 }
2437 list_sort(NULL, &dev->namespaces, ns_cmp);
2438}
2439
bda4e0fb
KB		 2440static void nvme_set_irq_hints(struct nvme_dev *dev)
2441{
2442 struct nvme_queue *nvmeq;
2443 int i;
2444
2445 for (i = 0; i < dev->online_queues; i++) {
2446 nvmeq = dev->queues[i];
2447
2448 if (!nvmeq->tags || !(*nvmeq->tags))
2449 continue;
2450
2451 irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector,
2452 blk_mq_tags_cpumask(*nvmeq->tags));
2453 }
2454}
2455
a5768aa8
KB		 2456static void nvme_dev_scan(struct work_struct *work)
2457{
2458 struct nvme_dev *dev = container_of(work, struct nvme_dev, scan_work);
2459 struct nvme_id_ctrl *ctrl;
2460
2461 if (!dev->tagset.tags)
2462 return;
2463 if (nvme_identify_ctrl(dev, &ctrl))
2464 return;
2465 nvme_scan_namespaces(dev, le32_to_cpup(&ctrl->nn));
2466 kfree(ctrl);
bda4e0fb 2467 nvme_set_irq_hints(dev);
a5768aa8
KB		 2468}
2469
422ef0c7
MW		 2470/*
2471 * Return: error value if an error occurred setting up the queues or calling
2472 * Identify Device. 0 if these succeeded, even if adding some of the
2473 * namespaces failed. At the moment, these failures are silent. TBD which
2474 * failures should be reported.
2475 */
8d85fce7 2476static int nvme_dev_add(struct nvme_dev *dev)
b60503ba 2477{
e75ec752 2478 struct pci_dev *pdev = to_pci_dev(dev->dev);
c3bfe717 2479 int res;
51814232 2480 struct nvme_id_ctrl *ctrl;
a310acd7 2481 int shift = NVME_CAP_MPSMIN(lo_hi_readq(&dev->bar->cap)) + 12;
b60503ba 2482
d29ec824 2483 res = nvme_identify_ctrl(dev, &ctrl);
b60503ba 2484 if (res) {
e75ec752 2485 dev_err(dev->dev, "Identify Controller failed (%d)\n", res);
e1e5e564 2486 return -EIO;
b60503ba
MW		 2487 }
2488
0e5e4f0e 2489 dev->oncs = le16_to_cpup(&ctrl->oncs);
c30341dc 2490 dev->abort_limit = ctrl->acl + 1;
a7d2ce28 2491 dev->vwc = ctrl->vwc;
51814232
MW		 2492 memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn));
2493 memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn));
2494 memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
159b67d7 2495 if (ctrl->mdts)
8fc23e03 2496 dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9);
b12363d0
S		 2497 else
2498 dev->max_hw_sectors = UINT_MAX;
68608c26 2499 if ((pdev->vendor == PCI_VENDOR_ID_INTEL) &&
a4aea562
MB		 2500 (pdev->device == 0x0953) && ctrl->vs[3]) {
2501 unsigned int max_hw_sectors;
2502
159b67d7 2503 dev->stripe_size = 1 << (ctrl->vs[3] + shift);
a4aea562
MB		 2504 max_hw_sectors = dev->stripe_size >> (shift - 9);
2505 if (dev->max_hw_sectors) {
2506 dev->max_hw_sectors = min(max_hw_sectors,
2507 dev->max_hw_sectors);
2508 } else
2509 dev->max_hw_sectors = max_hw_sectors;
2510 }
d29ec824 2511 kfree(ctrl);
a4aea562 2512
ffe7704d
KB		 2513 if (!dev->tagset.tags) {
2514 dev->tagset.ops = &nvme_mq_ops;
2515 dev->tagset.nr_hw_queues = dev->online_queues - 1;
2516 dev->tagset.timeout = NVME_IO_TIMEOUT;
2517 dev->tagset.numa_node = dev_to_node(dev->dev);
2518 dev->tagset.queue_depth =
a4aea562 2519 min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
ffe7704d
KB		 2520 dev->tagset.cmd_size = nvme_cmd_size(dev);
2521 dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
2522 dev->tagset.driver_data = dev;
b60503ba 2523
ffe7704d
KB		 2524 if (blk_mq_alloc_tag_set(&dev->tagset))
2525 return 0;
2526 }
a5768aa8 2527 schedule_work(&dev->scan_work);
e1e5e564 2528 return 0;
b60503ba
MW		 2529}
2530
0877cb0d
KB		 2531static int nvme_dev_map(struct nvme_dev *dev)
2532{
42f61420 2533 u64 cap;
0877cb0d 2534 int bars, result = -ENOMEM;
e75ec752 2535 struct pci_dev *pdev = to_pci_dev(dev->dev);
0877cb0d
KB		 2536
2537 if (pci_enable_device_mem(pdev))
2538 return result;
2539
2540 dev->entry[0].vector = pdev->irq;
2541 pci_set_master(pdev);
2542 bars = pci_select_bars(pdev, IORESOURCE_MEM);
be7837e8
JA		 2543 if (!bars)
2544 goto disable_pci;
2545
0877cb0d
KB		 2546 if (pci_request_selected_regions(pdev, bars, "nvme"))
2547 goto disable_pci;
2548
e75ec752
CH		 2549 if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
2550 dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
052d0efa 2551 goto disable;
0877cb0d 2552
0877cb0d
KB		 2553 dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
2554 if (!dev->bar)
2555 goto disable;
e32efbfc 2556
0e53d180
KB		 2557 if (readl(&dev->bar->csts) == -1) {
2558 result = -ENODEV;
2559 goto unmap;
2560 }
e32efbfc
JA		 2561
2562 /*
2563 * Some devices don't advertse INTx interrupts, pre-enable a single
2564 * MSIX vec for setup. We'll adjust this later.
2565 */
2566 if (!pdev->irq) {
2567 result = pci_enable_msix(pdev, dev->entry, 1);
2568 if (result < 0)
2569 goto unmap;
2570 }
2571
a310acd7 2572 cap = lo_hi_readq(&dev->bar->cap);
42f61420
KB		 2573 dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
2574 dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
0877cb0d 2575 dev->dbs = ((void __iomem *)dev->bar) + 4096;
8ffaadf7
JD		 2576 if (readl(&dev->bar->vs) >= NVME_VS(1, 2))
2577 dev->cmb = nvme_map_cmb(dev);
0877cb0d
KB		 2578
2579 return 0;
2580
0e53d180
KB		 2581 unmap:
2582 iounmap(dev->bar);
2583 dev->bar = NULL;
0877cb0d
KB		 2584 disable:
2585 pci_release_regions(pdev);
2586 disable_pci:
2587 pci_disable_device(pdev);
2588 return result;
2589}
2590
2591static void nvme_dev_unmap(struct nvme_dev *dev)
2592{
e75ec752
CH		 2593 struct pci_dev *pdev = to_pci_dev(dev->dev);
2594
2595 if (pdev->msi_enabled)
2596 pci_disable_msi(pdev);
2597 else if (pdev->msix_enabled)
2598 pci_disable_msix(pdev);
0877cb0d
KB		 2599
2600 if (dev->bar) {
2601 iounmap(dev->bar);
2602 dev->bar = NULL;
e75ec752 2603 pci_release_regions(pdev);
0877cb0d
KB		 2604 }
2605
e75ec752
CH		 2606 if (pci_is_enabled(pdev))
2607 pci_disable_device(pdev);
0877cb0d
KB		 2608}
2609
4d115420
KB		 2610struct nvme_delq_ctx {
2611 struct task_struct *waiter;
2612 struct kthread_worker *worker;
2613 atomic_t refcount;
2614};
2615
2616static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev)
2617{
2618 dq->waiter = current;
2619 mb();
2620
2621 for (;;) {
2622 set_current_state(TASK_KILLABLE);
2623 if (!atomic_read(&dq->refcount))
2624 break;
2625 if (!schedule_timeout(ADMIN_TIMEOUT) ||
2626 fatal_signal_pending(current)) {
0fb59cbc
KB		 2627 /*
2628 * Disable the controller first since we can't trust it
2629 * at this point, but leave the admin queue enabled
2630 * until all queue deletion requests are flushed.
2631 * FIXME: This may take a while if there are more h/w
2632 * queues than admin tags.
2633 */
4d115420 2634 set_current_state(TASK_RUNNING);
a310acd7 2635 nvme_disable_ctrl(dev, lo_hi_readq(&dev->bar->cap));
0fb59cbc 2636 nvme_clear_queue(dev->queues[0]);
4d115420 2637 flush_kthread_worker(dq->worker);
0fb59cbc 2638 nvme_disable_queue(dev, 0);
4d115420
KB		 2639 return;
2640 }
2641 }
2642 set_current_state(TASK_RUNNING);
2643}
2644
2645static void nvme_put_dq(struct nvme_delq_ctx *dq)
2646{
2647 atomic_dec(&dq->refcount);
2648 if (dq->waiter)
2649 wake_up_process(dq->waiter);
2650}
2651
2652static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq)
2653{
2654 atomic_inc(&dq->refcount);
2655 return dq;
2656}
2657
2658static void nvme_del_queue_end(struct nvme_queue *nvmeq)
2659{
2660 struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx;
4d115420 2661 nvme_put_dq(dq);
604e8c8d
KB		 2662
2663 spin_lock_irq(&nvmeq->q_lock);
2664 nvme_process_cq(nvmeq);
2665 spin_unlock_irq(&nvmeq->q_lock);
4d115420
KB		 2666}
2667
2668static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode,
2669 kthread_work_func_t fn)
2670{
2671 struct nvme_command c;
2672
2673 memset(&c, 0, sizeof(c));
2674 c.delete_queue.opcode = opcode;
2675 c.delete_queue.qid = cpu_to_le16(nvmeq->qid);
2676
2677 init_kthread_work(&nvmeq->cmdinfo.work, fn);
a4aea562
MB		 2678 return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo,
2679 ADMIN_TIMEOUT);
4d115420
KB		 2680}
2681
2682static void nvme_del_cq_work_handler(struct kthread_work *work)
2683{
2684 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2685 cmdinfo.work);
2686 nvme_del_queue_end(nvmeq);
2687}
2688
2689static int nvme_delete_cq(struct nvme_queue *nvmeq)
2690{
2691 return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq,
2692 nvme_del_cq_work_handler);
2693}
2694
2695static void nvme_del_sq_work_handler(struct kthread_work *work)
2696{
2697 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2698 cmdinfo.work);
2699 int status = nvmeq->cmdinfo.status;
2700
2701 if (!status)
2702 status = nvme_delete_cq(nvmeq);
2703 if (status)
2704 nvme_del_queue_end(nvmeq);
2705}
2706
2707static int nvme_delete_sq(struct nvme_queue *nvmeq)
2708{
2709 return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq,
2710 nvme_del_sq_work_handler);
2711}
2712
2713static void nvme_del_queue_start(struct kthread_work *work)
2714{
2715 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2716 cmdinfo.work);
4d115420
KB		 2717 if (nvme_delete_sq(nvmeq))
2718 nvme_del_queue_end(nvmeq);
2719}
2720
2721static void nvme_disable_io_queues(struct nvme_dev *dev)
2722{
2723 int i;
2724 DEFINE_KTHREAD_WORKER_ONSTACK(worker);
2725 struct nvme_delq_ctx dq;
2726 struct task_struct *kworker_task = kthread_run(kthread_worker_fn,
2727 &worker, "nvme%d", dev->instance);
2728
2729 if (IS_ERR(kworker_task)) {
e75ec752 2730 dev_err(dev->dev,
4d115420
KB		 2731 "Failed to create queue del task\n");
2732 for (i = dev->queue_count - 1; i > 0; i--)
2733 nvme_disable_queue(dev, i);
2734 return;
2735 }
2736
2737 dq.waiter = NULL;
2738 atomic_set(&dq.refcount, 0);
2739 dq.worker = &worker;
2740 for (i = dev->queue_count - 1; i > 0; i--) {
a4aea562 2741 struct nvme_queue *nvmeq = dev->queues[i];
4d115420
KB		 2742
2743 if (nvme_suspend_queue(nvmeq))
2744 continue;
2745 nvmeq->cmdinfo.ctx = nvme_get_dq(&dq);
2746 nvmeq->cmdinfo.worker = dq.worker;
2747 init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start);
2748 queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work);
2749 }
2750 nvme_wait_dq(&dq, dev);
2751 kthread_stop(kworker_task);
2752}
2753
b9afca3e
DM		 2754/*
2755* Remove the node from the device list and check
2756* for whether or not we need to stop the nvme_thread.
2757*/
2758static void nvme_dev_list_remove(struct nvme_dev *dev)
2759{
2760 struct task_struct *tmp = NULL;
2761
2762 spin_lock(&dev_list_lock);
2763 list_del_init(&dev->node);
2764 if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) {
2765 tmp = nvme_thread;
2766 nvme_thread = NULL;
2767 }
2768 spin_unlock(&dev_list_lock);
2769
2770 if (tmp)
2771 kthread_stop(tmp);
2772}
2773
c9d3bf88
KB		 2774static void nvme_freeze_queues(struct nvme_dev *dev)
2775{
2776 struct nvme_ns *ns;
2777
2778 list_for_each_entry(ns, &dev->namespaces, list) {
2779 blk_mq_freeze_queue_start(ns->queue);
2780
cddcd72b 2781 spin_lock_irq(ns->queue->queue_lock);
c9d3bf88 2782 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
cddcd72b 2783 spin_unlock_irq(ns->queue->queue_lock);
c9d3bf88
KB		 2784
2785 blk_mq_cancel_requeue_work(ns->queue);
2786 blk_mq_stop_hw_queues(ns->queue);
2787 }
2788}
2789
2790static void nvme_unfreeze_queues(struct nvme_dev *dev)
2791{
2792 struct nvme_ns *ns;
2793
2794 list_for_each_entry(ns, &dev->namespaces, list) {
2795 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2796 blk_mq_unfreeze_queue(ns->queue);
2797 blk_mq_start_stopped_hw_queues(ns->queue, true);
2798 blk_mq_kick_requeue_list(ns->queue);
2799 }
2800}
2801
f0b50732 2802static void nvme_dev_shutdown(struct nvme_dev *dev)
b60503ba 2803{
22404274 2804 int i;
7c1b2450 2805 u32 csts = -1;
22404274 2806
b9afca3e 2807 nvme_dev_list_remove(dev);
1fa6aead 2808
c9d3bf88
KB		 2809 if (dev->bar) {
2810 nvme_freeze_queues(dev);
7c1b2450 2811 csts = readl(&dev->bar->csts);
c9d3bf88 2812 }
7c1b2450 2813 if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) {
4d115420 2814 for (i = dev->queue_count - 1; i >= 0; i--) {
a4aea562 2815 struct nvme_queue *nvmeq = dev->queues[i];
4d115420 2816 nvme_suspend_queue(nvmeq);
4d115420
KB		 2817 }
2818 } else {
2819 nvme_disable_io_queues(dev);
1894d8f1 2820 nvme_shutdown_ctrl(dev);
4d115420
KB		 2821 nvme_disable_queue(dev, 0);
2822 }
f0b50732 2823 nvme_dev_unmap(dev);
07836e65
KB		 2824
2825 for (i = dev->queue_count - 1; i >= 0; i--)
2826 nvme_clear_queue(dev->queues[i]);
f0b50732
KB		 2827}
2828
2829static void nvme_dev_remove(struct nvme_dev *dev)
2830{
5105aa55 2831 struct nvme_ns *ns, *next;
f0b50732 2832
5105aa55 2833 list_for_each_entry_safe(ns, next, &dev->namespaces, list)
a5768aa8 2834 nvme_ns_remove(ns);
b60503ba
MW		 2835}
2836
091b6092
MW		 2837static int nvme_setup_prp_pools(struct nvme_dev *dev)
2838{
e75ec752 2839 dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
091b6092
MW		 2840 PAGE_SIZE, PAGE_SIZE, 0);
2841 if (!dev->prp_page_pool)
2842 return -ENOMEM;
2843
99802a7a 2844 /* Optimisation for I/Os between 4k and 128k */
e75ec752 2845 dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
99802a7a
MW		 2846 256, 256, 0);
2847 if (!dev->prp_small_pool) {
2848 dma_pool_destroy(dev->prp_page_pool);
2849 return -ENOMEM;
2850 }
091b6092
MW		 2851 return 0;
2852}
2853
2854static void nvme_release_prp_pools(struct nvme_dev *dev)
2855{
2856 dma_pool_destroy(dev->prp_page_pool);
99802a7a 2857 dma_pool_destroy(dev->prp_small_pool);
091b6092
MW		 2858}
2859
cd58ad7d
QSA		 2860static DEFINE_IDA(nvme_instance_ida);
2861
2862static int nvme_set_instance(struct nvme_dev *dev)
b60503ba 2863{
cd58ad7d
QSA		 2864 int instance, error;
2865
2866 do {
2867 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2868 return -ENODEV;
2869
2870 spin_lock(&dev_list_lock);
2871 error = ida_get_new(&nvme_instance_ida, &instance);
2872 spin_unlock(&dev_list_lock);
2873 } while (error == -EAGAIN);
2874
2875 if (error)
2876 return -ENODEV;
2877
2878 dev->instance = instance;
2879 return 0;
b60503ba
MW		 2880}
2881
2882static void nvme_release_instance(struct nvme_dev *dev)
2883{
cd58ad7d
QSA		 2884 spin_lock(&dev_list_lock);
2885 ida_remove(&nvme_instance_ida, dev->instance);
2886 spin_unlock(&dev_list_lock);
b60503ba
MW		 2887}
2888
5e82e952
KB		 2889static void nvme_free_dev(struct kref *kref)
2890{
2891 struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
9ac27090 2892
e75ec752 2893 put_device(dev->dev);
b3fffdef 2894 put_device(dev->device);
285dffc9 2895 nvme_release_instance(dev);
4af0e21c
KB		 2896 if (dev->tagset.tags)
2897 blk_mq_free_tag_set(&dev->tagset);
2898 if (dev->admin_q)
2899 blk_put_queue(dev->admin_q);
5e82e952
KB		 2900 kfree(dev->queues);
2901 kfree(dev->entry);
2902 kfree(dev);
2903}
2904
2905static int nvme_dev_open(struct inode *inode, struct file *f)
2906{
b3fffdef
KB		 2907 struct nvme_dev *dev;
2908 int instance = iminor(inode);
2909 int ret = -ENODEV;
2910
2911 spin_lock(&dev_list_lock);
2912 list_for_each_entry(dev, &dev_list, node) {
2913 if (dev->instance == instance) {
2e1d8448
KB		 2914 if (!dev->admin_q) {
2915 ret = -EWOULDBLOCK;
2916 break;
2917 }
b3fffdef
KB		 2918 if (!kref_get_unless_zero(&dev->kref))
2919 break;
2920 f->private_data = dev;
2921 ret = 0;
2922 break;
2923 }
2924 }
2925 spin_unlock(&dev_list_lock);
2926
2927 return ret;
5e82e952
KB		 2928}
2929
2930static int nvme_dev_release(struct inode *inode, struct file *f)
2931{
2932 struct nvme_dev *dev = f->private_data;
2933 kref_put(&dev->kref, nvme_free_dev);
2934 return 0;
2935}
2936
2937static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
2938{
2939 struct nvme_dev *dev = f->private_data;
a4aea562
MB		 2940 struct nvme_ns *ns;
2941
5e82e952
KB		 2942 switch (cmd) {
2943 case NVME_IOCTL_ADMIN_CMD:
a4aea562 2944 return nvme_user_cmd(dev, NULL, (void __user *)arg);
7963e521 2945 case NVME_IOCTL_IO_CMD:
a4aea562
MB		 2946 if (list_empty(&dev->namespaces))
2947 return -ENOTTY;
2948 ns = list_first_entry(&dev->namespaces, struct nvme_ns, list);
2949 return nvme_user_cmd(dev, ns, (void __user *)arg);
4cc06521
KB		 2950 case NVME_IOCTL_RESET:
2951 dev_warn(dev->dev, "resetting controller\n");
2952 return nvme_reset(dev);
81f03fed
JD		 2953 case NVME_IOCTL_SUBSYS_RESET:
2954 return nvme_subsys_reset(dev);
5e82e952
KB		 2955 default:
2956 return -ENOTTY;
2957 }
2958}
2959
2960static const struct file_operations nvme_dev_fops = {
2961 .owner = THIS_MODULE,
2962 .open = nvme_dev_open,
2963 .release = nvme_dev_release,
2964 .unlocked_ioctl = nvme_dev_ioctl,
2965 .compat_ioctl = nvme_dev_ioctl,
2966};
2967
3cf519b5 2968static void nvme_probe_work(struct work_struct *work)
f0b50732 2969{
3cf519b5 2970 struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work);
b9afca3e 2971 bool start_thread = false;
3cf519b5 2972 int result;
f0b50732
KB		 2973
2974 result = nvme_dev_map(dev);
2975 if (result)
3cf519b5 2976 goto out;
f0b50732
KB		 2977
2978 result = nvme_configure_admin_queue(dev);
2979 if (result)
2980 goto unmap;
2981
2982 spin_lock(&dev_list_lock);
b9afca3e
DM		 2983 if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) {
2984 start_thread = true;
2985 nvme_thread = NULL;
2986 }
f0b50732
KB		 2987 list_add(&dev->node, &dev_list);
2988 spin_unlock(&dev_list_lock);
2989
b9afca3e
DM		 2990 if (start_thread) {
2991 nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
387caa5a 2992 wake_up_all(&nvme_kthread_wait);
b9afca3e
DM		 2993 } else
2994 wait_event_killable(nvme_kthread_wait, nvme_thread);
2995
2996 if (IS_ERR_OR_NULL(nvme_thread)) {
2997 result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR;
2998 goto disable;
2999 }
a4aea562
MB		 3000
3001 nvme_init_queue(dev->queues[0], 0);
0fb59cbc
KB		 3002 result = nvme_alloc_admin_tags(dev);
3003 if (result)
3004 goto disable;
b9afca3e 3005
f0b50732 3006 result = nvme_setup_io_queues(dev);
badc34d4 3007 if (result)
0fb59cbc 3008 goto free_tags;
f0b50732 3009
1efccc9d 3010 dev->event_limit = 1;
3cf519b5 3011
2659e57b
CH		 3012 /*
3013 * Keep the controller around but remove all namespaces if we don't have
3014 * any working I/O queue.
3015 */
3cf519b5
CH		 3016 if (dev->online_queues < 2) {
3017 dev_warn(dev->dev, "IO queues not created\n");
3cf519b5
CH		 3018 nvme_dev_remove(dev);
3019 } else {
3020 nvme_unfreeze_queues(dev);
3021 nvme_dev_add(dev);
3022 }
3023
3024 return;
f0b50732 3025
0fb59cbc
KB		 3026 free_tags:
3027 nvme_dev_remove_admin(dev);
4af0e21c
KB		 3028 blk_put_queue(dev->admin_q);
3029 dev->admin_q = NULL;
3030 dev->queues[0]->tags = NULL;
f0b50732 3031 disable:
a1a5ef99 3032 nvme_disable_queue(dev, 0);
b9afca3e 3033 nvme_dev_list_remove(dev);
f0b50732
KB		 3034 unmap:
3035 nvme_dev_unmap(dev);
3cf519b5
CH		 3036 out:
3037 if (!work_busy(&dev->reset_work))
3038 nvme_dead_ctrl(dev);
f0b50732
KB		 3039}
3040
9a6b9458
KB		 3041static int nvme_remove_dead_ctrl(void *arg)
3042{
3043 struct nvme_dev *dev = (struct nvme_dev *)arg;
e75ec752 3044 struct pci_dev *pdev = to_pci_dev(dev->dev);
9a6b9458
KB		 3045
3046 if (pci_get_drvdata(pdev))
c81f4975 3047 pci_stop_and_remove_bus_device_locked(pdev);
9a6b9458
KB		 3048 kref_put(&dev->kref, nvme_free_dev);
3049 return 0;
3050}
3051
de3eff2b
KB		 3052static void nvme_dead_ctrl(struct nvme_dev *dev)
3053{
3054 dev_warn(dev->dev, "Device failed to resume\n");
3055 kref_get(&dev->kref);
3056 if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d",
3057 dev->instance))) {
3058 dev_err(dev->dev,
3059 "Failed to start controller remove task\n");
3060 kref_put(&dev->kref, nvme_free_dev);
3061 }
3062}
3063
77b50d9e 3064static void nvme_reset_work(struct work_struct *ws)
9a6b9458 3065{
77b50d9e 3066 struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
ffe7704d
KB		 3067 bool in_probe = work_busy(&dev->probe_work);
3068
9a6b9458 3069 nvme_dev_shutdown(dev);
ffe7704d
KB		 3070
3071 /* Synchronize with device probe so that work will see failure status
3072 * and exit gracefully without trying to schedule another reset */
3073 flush_work(&dev->probe_work);
3074
3075 /* Fail this device if reset occured during probe to avoid
3076 * infinite initialization loops. */
3077 if (in_probe) {
de3eff2b 3078 nvme_dead_ctrl(dev);
ffe7704d 3079 return;
9a6b9458 3080 }
ffe7704d
KB		 3081 /* Schedule device resume asynchronously so the reset work is available
3082 * to cleanup errors that may occur during reinitialization */
3083 schedule_work(&dev->probe_work);
9a6b9458
KB		 3084}
3085
90667892 3086static int __nvme_reset(struct nvme_dev *dev)
9ca97374 3087{
90667892
CH		 3088 if (work_pending(&dev->reset_work))
3089 return -EBUSY;
3090 list_del_init(&dev->node);
3091 queue_work(nvme_workq, &dev->reset_work);
3092 return 0;
9ca97374
TH		 3093}
3094
4cc06521
KB		 3095static int nvme_reset(struct nvme_dev *dev)
3096{
90667892 3097 int ret;
4cc06521
KB		 3098
3099 if (!dev->admin_q || blk_queue_dying(dev->admin_q))
3100 return -ENODEV;
3101
3102 spin_lock(&dev_list_lock);
90667892 3103 ret = __nvme_reset(dev);
4cc06521
KB		 3104 spin_unlock(&dev_list_lock);
3105
3106 if (!ret) {
3107 flush_work(&dev->reset_work);
ffe7704d 3108 flush_work(&dev->probe_work);
4cc06521
KB		 3109 return 0;
3110 }
3111
3112 return ret;
3113}
3114
3115static ssize_t nvme_sysfs_reset(struct device *dev,
3116 struct device_attribute *attr, const char *buf,
3117 size_t count)
3118{
3119 struct nvme_dev *ndev = dev_get_drvdata(dev);
3120 int ret;
3121
3122 ret = nvme_reset(ndev);
3123 if (ret < 0)
3124 return ret;
3125
3126 return count;
3127}
3128static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3129
8d85fce7 3130static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
b60503ba 3131{
a4aea562 3132 int node, result = -ENOMEM;
b60503ba
MW		 3133 struct nvme_dev *dev;
3134
a4aea562
MB		 3135 node = dev_to_node(&pdev->dev);
3136 if (node == NUMA_NO_NODE)
3137 set_dev_node(&pdev->dev, 0);
3138
3139 dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
b60503ba
MW		 3140 if (!dev)
3141 return -ENOMEM;
a4aea562
MB		 3142 dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry),
3143 GFP_KERNEL, node);
b60503ba
MW		 3144 if (!dev->entry)
3145 goto free;
a4aea562
MB		 3146 dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
3147 GFP_KERNEL, node);
b60503ba
MW		 3148 if (!dev->queues)
3149 goto free;
3150
3151 INIT_LIST_HEAD(&dev->namespaces);
77b50d9e 3152 INIT_WORK(&dev->reset_work, nvme_reset_work);
e75ec752 3153 dev->dev = get_device(&pdev->dev);
9a6b9458 3154 pci_set_drvdata(pdev, dev);
cd58ad7d
QSA		 3155 result = nvme_set_instance(dev);
3156 if (result)
a96d4f5c 3157 goto put_pci;
b60503ba 3158
091b6092
MW		 3159 result = nvme_setup_prp_pools(dev);
3160 if (result)
0877cb0d 3161 goto release;
091b6092 3162
fb35e914 3163 kref_init(&dev->kref);
b3fffdef
KB		 3164 dev->device = device_create(nvme_class, &pdev->dev,
3165 MKDEV(nvme_char_major, dev->instance),
3166 dev, "nvme%d", dev->instance);
3167 if (IS_ERR(dev->device)) {
3168 result = PTR_ERR(dev->device);
2e1d8448 3169 goto release_pools;
b3fffdef
KB		 3170 }
3171 get_device(dev->device);
4cc06521
KB		 3172 dev_set_drvdata(dev->device, dev);
3173
3174 result = device_create_file(dev->device, &dev_attr_reset_controller);
3175 if (result)
3176 goto put_dev;
740216fc 3177
e6e96d73 3178 INIT_LIST_HEAD(&dev->node);
a5768aa8 3179 INIT_WORK(&dev->scan_work, nvme_dev_scan);
3cf519b5 3180 INIT_WORK(&dev->probe_work, nvme_probe_work);
2e1d8448 3181 schedule_work(&dev->probe_work);
b60503ba
MW		 3182 return 0;
3183
4cc06521
KB		 3184 put_dev:
3185 device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
3186 put_device(dev->device);
0877cb0d 3187 release_pools:
091b6092 3188 nvme_release_prp_pools(dev);
0877cb0d
KB		 3189 release:
3190 nvme_release_instance(dev);
a96d4f5c 3191 put_pci:
e75ec752 3192 put_device(dev->dev);
b60503ba
MW		 3193 free:
3194 kfree(dev->queues);
3195 kfree(dev->entry);
3196 kfree(dev);
3197 return result;
3198}
3199
f0d54a54
KB		 3200static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
3201{
a6739479 3202 struct nvme_dev *dev = pci_get_drvdata(pdev);
f0d54a54 3203
a6739479
KB		 3204 if (prepare)
3205 nvme_dev_shutdown(dev);
3206 else
0a7385ad 3207 schedule_work(&dev->probe_work);
f0d54a54
KB		 3208}
3209
09ece142
KB		 3210static void nvme_shutdown(struct pci_dev *pdev)
3211{
3212 struct nvme_dev *dev = pci_get_drvdata(pdev);
3213 nvme_dev_shutdown(dev);
3214}
3215
8d85fce7 3216static void nvme_remove(struct pci_dev *pdev)
b60503ba
MW		 3217{
3218 struct nvme_dev *dev = pci_get_drvdata(pdev);
9a6b9458
KB		 3219
3220 spin_lock(&dev_list_lock);
3221 list_del_init(&dev->node);
3222 spin_unlock(&dev_list_lock);
3223
3224 pci_set_drvdata(pdev, NULL);
2e1d8448 3225 flush_work(&dev->probe_work);
9a6b9458 3226 flush_work(&dev->reset_work);
a5768aa8 3227 flush_work(&dev->scan_work);
4cc06521 3228 device_remove_file(dev->device, &dev_attr_reset_controller);
c9d3bf88 3229 nvme_dev_remove(dev);
3399a3f7 3230 nvme_dev_shutdown(dev);
a4aea562 3231 nvme_dev_remove_admin(dev);
b3fffdef 3232 device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
a1a5ef99 3233 nvme_free_queues(dev, 0);
8ffaadf7 3234 nvme_release_cmb(dev);
9a6b9458 3235 nvme_release_prp_pools(dev);
5e82e952 3236 kref_put(&dev->kref, nvme_free_dev);
b60503ba
MW		 3237}
3238
3239/* These functions are yet to be implemented */
3240#define nvme_error_detected NULL
3241#define nvme_dump_registers NULL
3242#define nvme_link_reset NULL
3243#define nvme_slot_reset NULL
3244#define nvme_error_resume NULL
cd638946 3245
671a6018 3246#ifdef CONFIG_PM_SLEEP
cd638946
KB		 3247static int nvme_suspend(struct device *dev)
3248{
3249 struct pci_dev *pdev = to_pci_dev(dev);
3250 struct nvme_dev *ndev = pci_get_drvdata(pdev);
3251
3252 nvme_dev_shutdown(ndev);
3253 return 0;
3254}
3255
3256static int nvme_resume(struct device *dev)
3257{
3258 struct pci_dev *pdev = to_pci_dev(dev);
3259 struct nvme_dev *ndev = pci_get_drvdata(pdev);
cd638946 3260
0a7385ad 3261 schedule_work(&ndev->probe_work);
9a6b9458 3262 return 0;
cd638946 3263}
671a6018 3264#endif
cd638946
KB		 3265
3266static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
b60503ba 3267
1d352035 3268static const struct pci_error_handlers nvme_err_handler = {
b60503ba
MW		 3269 .error_detected = nvme_error_detected,
3270 .mmio_enabled = nvme_dump_registers,
3271 .link_reset = nvme_link_reset,
3272 .slot_reset = nvme_slot_reset,
3273 .resume = nvme_error_resume,
f0d54a54 3274 .reset_notify = nvme_reset_notify,
b60503ba
MW		 3275};
3276
3277/* Move to pci_ids.h later */
3278#define PCI_CLASS_STORAGE_EXPRESS 0x010802
3279
6eb0d698 3280static const struct pci_device_id nvme_id_table[] = {
b60503ba 3281 { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
c74dc780 3282 { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
b60503ba
MW		 3283 { 0, }
3284};
3285MODULE_DEVICE_TABLE(pci, nvme_id_table);
3286
3287static struct pci_driver nvme_driver = {
3288 .name = "nvme",
3289 .id_table = nvme_id_table,
3290 .probe = nvme_probe,
8d85fce7 3291 .remove = nvme_remove,
09ece142 3292 .shutdown = nvme_shutdown,
cd638946
KB		 3293 .driver = {
3294 .pm = &nvme_dev_pm_ops,
3295 },
b60503ba
MW		 3296 .err_handler = &nvme_err_handler,
3297};
3298
3299static int __init nvme_init(void)
3300{
0ac13140 3301 int result;
1fa6aead 3302
b9afca3e 3303 init_waitqueue_head(&nvme_kthread_wait);
b60503ba 3304
9a6b9458
KB		 3305 nvme_workq = create_singlethread_workqueue("nvme");
3306 if (!nvme_workq)
b9afca3e 3307 return -ENOMEM;
9a6b9458 3308
5c42ea16
KB		 3309 result = register_blkdev(nvme_major, "nvme");
3310 if (result < 0)
9a6b9458 3311 goto kill_workq;
5c42ea16 3312 else if (result > 0)
0ac13140 3313 nvme_major = result;
b60503ba 3314
b3fffdef
KB		 3315 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
3316 &nvme_dev_fops);
3317 if (result < 0)
3318 goto unregister_blkdev;
3319 else if (result > 0)
3320 nvme_char_major = result;
3321
3322 nvme_class = class_create(THIS_MODULE, "nvme");
c727040b
AK		 3323 if (IS_ERR(nvme_class)) {
3324 result = PTR_ERR(nvme_class);
b3fffdef 3325 goto unregister_chrdev;
c727040b 3326 }
b3fffdef 3327
f3db22fe
KB		 3328 result = pci_register_driver(&nvme_driver);
3329 if (result)
b3fffdef 3330 goto destroy_class;
1fa6aead 3331 return 0;
b60503ba 3332
b3fffdef
KB		 3333 destroy_class:
3334 class_destroy(nvme_class);
3335 unregister_chrdev:
3336 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1fa6aead 3337 unregister_blkdev:
b60503ba 3338 unregister_blkdev(nvme_major, "nvme");
9a6b9458
KB		 3339 kill_workq:
3340 destroy_workqueue(nvme_workq);
b60503ba
MW		 3341 return result;
3342}
3343
3344static void __exit nvme_exit(void)
3345{
3346 pci_unregister_driver(&nvme_driver);
3347 unregister_blkdev(nvme_major, "nvme");
9a6b9458 3348 destroy_workqueue(nvme_workq);
b3fffdef
KB		 3349 class_destroy(nvme_class);
3350 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
b9afca3e 3351 BUG_ON(nvme_thread && !IS_ERR(nvme_thread));
21bd78bc 3352 _nvme_check_size();
b60503ba
MW		 3353}
3354
3355MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
3356MODULE_LICENSE("GPL");
c78b4713 3357MODULE_VERSION("1.0");
b60503ba
MW		 3358module_init(nvme_init);
3359module_exit(nvme_exit);
Linux 6.x block layer and io_uring tree(s)