projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: Inline blk_integrity in struct gendisk
[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>
5d0f6131 42#include <scsi/sg.h>
797a796a
HM		 43#include <asm-generic/io-64-nonatomic-lo-hi.h>
44
9d99a8dd 45#include <uapi/linux/nvme_ioctl.h>
f11bb3e2
CH		 46#include "nvme.h"
47
b3fffdef 48#define NVME_MINORS (1U << MINORBITS)
9d43cf64 49#define NVME_Q_DEPTH 1024
d31af0a3 50#define NVME_AQ_DEPTH 256
b60503ba
MW		 51#define SQ_SIZE(depth) (depth * sizeof(struct nvme_command))
52#define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion))
9d43cf64 53#define ADMIN_TIMEOUT (admin_timeout * HZ)
2484f407 54#define SHUTDOWN_TIMEOUT (shutdown_timeout * HZ)
9d43cf64
KB		 55
56static unsigned char admin_timeout = 60;
57module_param(admin_timeout, byte, 0644);
58MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
b60503ba 59
bd67608a
MW		 60unsigned char nvme_io_timeout = 30;
61module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
b355084a 62MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
b60503ba 63
2484f407
DM		 64static unsigned char shutdown_timeout = 5;
65module_param(shutdown_timeout, byte, 0644);
66MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
67
b60503ba
MW		 68static int nvme_major;
69module_param(nvme_major, int, 0);
70
b3fffdef
KB		 71static int nvme_char_major;
72module_param(nvme_char_major, int, 0);
73
58ffacb5
MW		 74static int use_threaded_interrupts;
75module_param(use_threaded_interrupts, int, 0);
76
8ffaadf7
JD		 77static bool use_cmb_sqes = true;
78module_param(use_cmb_sqes, bool, 0644);
79MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");
80
1fa6aead
MW		 81static DEFINE_SPINLOCK(dev_list_lock);
82static LIST_HEAD(dev_list);
83static struct task_struct *nvme_thread;
9a6b9458 84static struct workqueue_struct *nvme_workq;
b9afca3e 85static wait_queue_head_t nvme_kthread_wait;
1fa6aead 86
b3fffdef
KB		 87static struct class *nvme_class;
88
90667892 89static int __nvme_reset(struct nvme_dev *dev);
4cc06521 90static int nvme_reset(struct nvme_dev *dev);
a4aea562 91static int nvme_process_cq(struct nvme_queue *nvmeq);
3cf519b5 92static void nvme_dead_ctrl(struct nvme_dev *dev);
d4b4ff8e 93
4d115420
KB		 94struct async_cmd_info {
95 struct kthread_work work;
96 struct kthread_worker *worker;
a4aea562 97 struct request *req;
4d115420
KB		 98 u32 result;
99 int status;
100 void *ctx;
101};
1fa6aead 102
b60503ba
MW		 103/*
104 * An NVM Express queue. Each device has at least two (one for admin
105 * commands and one for I/O commands).
106 */
107struct nvme_queue {
108 struct device *q_dmadev;
091b6092 109 struct nvme_dev *dev;
3193f07b 110 char irqname[24]; /* nvme4294967295-65535\0 */
b60503ba
MW		 111 spinlock_t q_lock;
112 struct nvme_command *sq_cmds;
8ffaadf7 113 struct nvme_command __iomem *sq_cmds_io;
b60503ba 114 volatile struct nvme_completion *cqes;
42483228 115 struct blk_mq_tags **tags;
b60503ba
MW		 116 dma_addr_t sq_dma_addr;
117 dma_addr_t cq_dma_addr;
b60503ba
MW		 118 u32 __iomem *q_db;
119 u16 q_depth;
6222d172 120 s16 cq_vector;
b60503ba
MW		 121 u16 sq_head;
122 u16 sq_tail;
123 u16 cq_head;
c30341dc 124 u16 qid;
e9539f47
MW		 125 u8 cq_phase;
126 u8 cqe_seen;
4d115420 127 struct async_cmd_info cmdinfo;
b60503ba
MW		 128};
129
130/*
131 * Check we didin't inadvertently grow the command struct
132 */
133static inline void _nvme_check_size(void)
134{
135 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
136 BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64);
137 BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64);
138 BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
139 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
f8ebf840 140 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
c30341dc 141 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
b60503ba
MW		 142 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
143 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
144 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
145 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
6ecec745 146 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
b60503ba
MW		 147}
148
edd10d33 149typedef void (*nvme_completion_fn)(struct nvme_queue *, void *,
c2f5b650
MW		 150 struct nvme_completion *);
151
e85248e5 152struct nvme_cmd_info {
c2f5b650
MW		 153 nvme_completion_fn fn;
154 void *ctx;
c30341dc 155 int aborted;
a4aea562 156 struct nvme_queue *nvmeq;
ac3dd5bd 157 struct nvme_iod iod[0];
e85248e5
MW		 158};
159
ac3dd5bd
JA		 160/*
161 * Max size of iod being embedded in the request payload
162 */
163#define NVME_INT_PAGES 2
164#define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->page_size)
fda631ff 165#define NVME_INT_MASK 0x01
ac3dd5bd
JA		 166
167/*
168 * Will slightly overestimate the number of pages needed. This is OK
169 * as it only leads to a small amount of wasted memory for the lifetime of
170 * the I/O.
171 */
172static int nvme_npages(unsigned size, struct nvme_dev *dev)
173{
174 unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size);
175 return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
176}
177
178static unsigned int nvme_cmd_size(struct nvme_dev *dev)
179{
180 unsigned int ret = sizeof(struct nvme_cmd_info);
181
182 ret += sizeof(struct nvme_iod);
183 ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev);
184 ret += sizeof(struct scatterlist) * NVME_INT_PAGES;
185
186 return ret;
187}
188
a4aea562
MB		 189static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
190 unsigned int hctx_idx)
e85248e5 191{
a4aea562
MB		 192 struct nvme_dev *dev = data;
193 struct nvme_queue *nvmeq = dev->queues[0];
194
42483228
KB		 195 WARN_ON(hctx_idx != 0);
196 WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
197 WARN_ON(nvmeq->tags);
198
a4aea562 199 hctx->driver_data = nvmeq;
42483228 200 nvmeq->tags = &dev->admin_tagset.tags[0];
a4aea562 201 return 0;
e85248e5
MW		 202}
203
4af0e21c
KB		 204static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
205{
206 struct nvme_queue *nvmeq = hctx->driver_data;
207
208 nvmeq->tags = NULL;
209}
210
a4aea562
MB		 211static int nvme_admin_init_request(void *data, struct request *req,
212 unsigned int hctx_idx, unsigned int rq_idx,
213 unsigned int numa_node)
22404274 214{
a4aea562
MB		 215 struct nvme_dev *dev = data;
216 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
217 struct nvme_queue *nvmeq = dev->queues[0];
218
219 BUG_ON(!nvmeq);
220 cmd->nvmeq = nvmeq;
221 return 0;
22404274
KB		 222}
223
a4aea562
MB		 224static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
225 unsigned int hctx_idx)
b60503ba 226{
a4aea562 227 struct nvme_dev *dev = data;
42483228 228 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
a4aea562 229
42483228
KB		 230 if (!nvmeq->tags)
231 nvmeq->tags = &dev->tagset.tags[hctx_idx];
b60503ba 232
42483228 233 WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
a4aea562
MB		 234 hctx->driver_data = nvmeq;
235 return 0;
b60503ba
MW		 236}
237
a4aea562
MB		 238static int nvme_init_request(void *data, struct request *req,
239 unsigned int hctx_idx, unsigned int rq_idx,
240 unsigned int numa_node)
b60503ba 241{
a4aea562
MB		 242 struct nvme_dev *dev = data;
243 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
244 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
245
246 BUG_ON(!nvmeq);
247 cmd->nvmeq = nvmeq;
248 return 0;
249}
250
251static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx,
252 nvme_completion_fn handler)
253{
254 cmd->fn = handler;
255 cmd->ctx = ctx;
256 cmd->aborted = 0;
c917dfe5 257 blk_mq_start_request(blk_mq_rq_from_pdu(cmd));
b60503ba
MW		 258}
259
ac3dd5bd
JA		 260static void *iod_get_private(struct nvme_iod *iod)
261{
262 return (void *) (iod->private & ~0x1UL);
263}
264
265/*
266 * If bit 0 is set, the iod is embedded in the request payload.
267 */
268static bool iod_should_kfree(struct nvme_iod *iod)
269{
fda631ff 270 return (iod->private & NVME_INT_MASK) == 0;
ac3dd5bd
JA		 271}
272
c2f5b650
MW		 273/* Special values must be less than 0x1000 */
274#define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA)
d2d87034
MW		 275#define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE)
276#define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE)
277#define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE)
be7b6275 278
edd10d33 279static void special_completion(struct nvme_queue *nvmeq, void *ctx,
c2f5b650
MW		 280 struct nvme_completion *cqe)
281{
282 if (ctx == CMD_CTX_CANCELLED)
283 return;
c2f5b650 284 if (ctx == CMD_CTX_COMPLETED) {
edd10d33 285 dev_warn(nvmeq->q_dmadev,
c2f5b650
MW		 286 "completed id %d twice on queue %d\n",
287 cqe->command_id, le16_to_cpup(&cqe->sq_id));
288 return;
289 }
290 if (ctx == CMD_CTX_INVALID) {
edd10d33 291 dev_warn(nvmeq->q_dmadev,
c2f5b650
MW		 292 "invalid id %d completed on queue %d\n",
293 cqe->command_id, le16_to_cpup(&cqe->sq_id));
294 return;
295 }
edd10d33 296 dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx);
c2f5b650
MW		 297}
298
a4aea562 299static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn)
b60503ba 300{
c2f5b650 301 void *ctx;
b60503ba 302
859361a2 303 if (fn)
a4aea562
MB		 304 *fn = cmd->fn;
305 ctx = cmd->ctx;
306 cmd->fn = special_completion;
307 cmd->ctx = CMD_CTX_CANCELLED;
c2f5b650 308 return ctx;
b60503ba
MW		 309}
310
a4aea562
MB		 311static void async_req_completion(struct nvme_queue *nvmeq, void *ctx,
312 struct nvme_completion *cqe)
3c0cf138 313{
a4aea562
MB		 314 u32 result = le32_to_cpup(&cqe->result);
315 u16 status = le16_to_cpup(&cqe->status) >> 1;
316
317 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ)
318 ++nvmeq->dev->event_limit;
a5768aa8
KB		 319 if (status != NVME_SC_SUCCESS)
320 return;
321
322 switch (result & 0xff07) {
323 case NVME_AER_NOTICE_NS_CHANGED:
324 dev_info(nvmeq->q_dmadev, "rescanning\n");
325 schedule_work(&nvmeq->dev->scan_work);
326 default:
327 dev_warn(nvmeq->q_dmadev, "async event result %08x\n", result);
328 }
b60503ba
MW		 329}
330
a4aea562
MB		 331static void abort_completion(struct nvme_queue *nvmeq, void *ctx,
332 struct nvme_completion *cqe)
5a92e700 333{
a4aea562
MB		 334 struct request *req = ctx;
335
336 u16 status = le16_to_cpup(&cqe->status) >> 1;
337 u32 result = le32_to_cpup(&cqe->result);
a51afb54 338
42483228 339 blk_mq_free_request(req);
a51afb54 340
a4aea562
MB		 341 dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result);
342 ++nvmeq->dev->abort_limit;
5a92e700
KB		 343}
344
a4aea562
MB		 345static void async_completion(struct nvme_queue *nvmeq, void *ctx,
346 struct nvme_completion *cqe)
b60503ba 347{
a4aea562
MB		 348 struct async_cmd_info *cmdinfo = ctx;
349 cmdinfo->result = le32_to_cpup(&cqe->result);
350 cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
351 queue_kthread_work(cmdinfo->worker, &cmdinfo->work);
42483228 352 blk_mq_free_request(cmdinfo->req);
b60503ba
MW		 353}
354
a4aea562
MB		 355static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq,
356 unsigned int tag)
b60503ba 357{
42483228 358 struct request *req = blk_mq_tag_to_rq(*nvmeq->tags, tag);
a51afb54 359
a4aea562 360 return blk_mq_rq_to_pdu(req);
4f5099af
KB		 361}
362
a4aea562
MB		 363/*
364 * Called with local interrupts disabled and the q_lock held. May not sleep.
365 */
366static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag,
367 nvme_completion_fn *fn)
4f5099af 368{
a4aea562
MB		 369 struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag);
370 void *ctx;
371 if (tag >= nvmeq->q_depth) {
372 *fn = special_completion;
373 return CMD_CTX_INVALID;
374 }
375 if (fn)
376 *fn = cmd->fn;
377 ctx = cmd->ctx;
378 cmd->fn = special_completion;
379 cmd->ctx = CMD_CTX_COMPLETED;
380 return ctx;
b60503ba
MW		 381}
382
383/**
714a7a22 384 * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
b60503ba
MW		 385 * @nvmeq: The queue to use
386 * @cmd: The command to send
387 *
388 * Safe to use from interrupt context
389 */
e3f879bf
SB		 390static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
391 struct nvme_command *cmd)
b60503ba 392{
a4aea562
MB		 393 u16 tail = nvmeq->sq_tail;
394
8ffaadf7
JD		 395 if (nvmeq->sq_cmds_io)
396 memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd));
397 else
398 memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));
399
b60503ba
MW		 400 if (++tail == nvmeq->q_depth)
401 tail = 0;
7547881d 402 writel(tail, nvmeq->q_db);
b60503ba 403 nvmeq->sq_tail = tail;
b60503ba
MW		 404}
405
e3f879bf 406static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
a4aea562
MB		 407{
408 unsigned long flags;
a4aea562 409 spin_lock_irqsave(&nvmeq->q_lock, flags);
e3f879bf 410 __nvme_submit_cmd(nvmeq, cmd);
a4aea562 411 spin_unlock_irqrestore(&nvmeq->q_lock, flags);
a4aea562
MB		 412}
413
eca18b23 414static __le64 **iod_list(struct nvme_iod *iod)
e025344c 415{
eca18b23 416 return ((void *)iod) + iod->offset;
e025344c
SMM		 417}
418
ac3dd5bd
JA		 419static inline void iod_init(struct nvme_iod *iod, unsigned nbytes,
420 unsigned nseg, unsigned long private)
eca18b23 421{
ac3dd5bd
JA		 422 iod->private = private;
423 iod->offset = offsetof(struct nvme_iod, sg[nseg]);
424 iod->npages = -1;
425 iod->length = nbytes;
426 iod->nents = 0;
eca18b23 427}
b60503ba 428
eca18b23 429static struct nvme_iod *
ac3dd5bd
JA		 430__nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev,
431 unsigned long priv, gfp_t gfp)
b60503ba 432{
eca18b23 433 struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) +
ac3dd5bd 434 sizeof(__le64 *) * nvme_npages(bytes, dev) +
eca18b23
MW		 435 sizeof(struct scatterlist) * nseg, gfp);
436
ac3dd5bd
JA		 437 if (iod)
438 iod_init(iod, bytes, nseg, priv);
eca18b23
MW		 439
440 return iod;
b60503ba
MW		 441}
442
ac3dd5bd
JA		 443static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev,
444 gfp_t gfp)
445{
446 unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) :
447 sizeof(struct nvme_dsm_range);
ac3dd5bd
JA		 448 struct nvme_iod *iod;
449
450 if (rq->nr_phys_segments <= NVME_INT_PAGES &&
451 size <= NVME_INT_BYTES(dev)) {
452 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq);
453
454 iod = cmd->iod;
ac3dd5bd 455 iod_init(iod, size, rq->nr_phys_segments,
fda631ff 456 (unsigned long) rq | NVME_INT_MASK);
ac3dd5bd
JA		 457 return iod;
458 }
459
460 return __nvme_alloc_iod(rq->nr_phys_segments, size, dev,
461 (unsigned long) rq, gfp);
462}
463
d29ec824 464static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod)
b60503ba 465{
1d090624 466 const int last_prp = dev->page_size / 8 - 1;
eca18b23
MW		 467 int i;
468 __le64 **list = iod_list(iod);
469 dma_addr_t prp_dma = iod->first_dma;
470
471 if (iod->npages == 0)
472 dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
473 for (i = 0; i < iod->npages; i++) {
474 __le64 *prp_list = list[i];
475 dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
476 dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
477 prp_dma = next_prp_dma;
478 }
ac3dd5bd
JA		 479
480 if (iod_should_kfree(iod))
481 kfree(iod);
b60503ba
MW		 482}
483
b4ff9c8d
KB		 484static int nvme_error_status(u16 status)
485{
486 switch (status & 0x7ff) {
487 case NVME_SC_SUCCESS:
488 return 0;
489 case NVME_SC_CAP_EXCEEDED:
490 return -ENOSPC;
491 default:
492 return -EIO;
493 }
494}
495
52b68d7e 496#ifdef CONFIG_BLK_DEV_INTEGRITY
e1e5e564
KB		 497static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
498{
499 if (be32_to_cpu(pi->ref_tag) == v)
500 pi->ref_tag = cpu_to_be32(p);
501}
502
503static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
504{
505 if (be32_to_cpu(pi->ref_tag) == p)
506 pi->ref_tag = cpu_to_be32(v);
507}
508
509/**
510 * nvme_dif_remap - remaps ref tags to bip seed and physical lba
511 *
512 * The virtual start sector is the one that was originally submitted by the
513 * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical
514 * start sector may be different. Remap protection information to match the
515 * physical LBA on writes, and back to the original seed on reads.
516 *
517 * Type 0 and 3 do not have a ref tag, so no remapping required.
518 */
519static void nvme_dif_remap(struct request *req,
520 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
521{
522 struct nvme_ns *ns = req->rq_disk->private_data;
523 struct bio_integrity_payload *bip;
524 struct t10_pi_tuple *pi;
525 void *p, *pmap;
526 u32 i, nlb, ts, phys, virt;
527
528 if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3)
529 return;
530
531 bip = bio_integrity(req->bio);
532 if (!bip)
533 return;
534
535 pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset;
e1e5e564
KB		 536
537 p = pmap;
538 virt = bip_get_seed(bip);
539 phys = nvme_block_nr(ns, blk_rq_pos(req));
540 nlb = (blk_rq_bytes(req) >> ns->lba_shift);
25520d55 541 ts = ns->disk->integrity.tuple_size;
e1e5e564
KB		 542
543 for (i = 0; i < nlb; i++, virt++, phys++) {
544 pi = (struct t10_pi_tuple *)p;
545 dif_swap(phys, virt, pi);
546 p += ts;
547 }
548 kunmap_atomic(pmap);
549}
550
52b68d7e
KB		 551static int nvme_noop_verify(struct blk_integrity_iter *iter)
552{
553 return 0;
554}
555
556static int nvme_noop_generate(struct blk_integrity_iter *iter)
557{
558 return 0;
559}
560
0f8087ec 561struct blk_integrity_profile nvme_meta_noop = {
52b68d7e
KB		 562 .name = "NVME_META_NOOP",
563 .generate_fn = nvme_noop_generate,
564 .verify_fn = nvme_noop_verify,
565};
566
567static void nvme_init_integrity(struct nvme_ns *ns)
568{
569 struct blk_integrity integrity;
570
571 switch (ns->pi_type) {
572 case NVME_NS_DPS_PI_TYPE3:
0f8087ec 573 integrity.profile = &t10_pi_type3_crc;
52b68d7e
KB		 574 break;
575 case NVME_NS_DPS_PI_TYPE1:
576 case NVME_NS_DPS_PI_TYPE2:
0f8087ec 577 integrity.profile = &t10_pi_type1_crc;
52b68d7e
KB		 578 break;
579 default:
0f8087ec 580 integrity.profile = &nvme_meta_noop;
52b68d7e
KB		 581 break;
582 }
583 integrity.tuple_size = ns->ms;
584 blk_integrity_register(ns->disk, &integrity);
585 blk_queue_max_integrity_segments(ns->queue, 1);
586}
587#else /* CONFIG_BLK_DEV_INTEGRITY */
588static void nvme_dif_remap(struct request *req,
589 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
590{
591}
592static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
593{
594}
595static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
596{
597}
598static void nvme_init_integrity(struct nvme_ns *ns)
599{
600}
601#endif
602
a4aea562 603static void req_completion(struct nvme_queue *nvmeq, void *ctx,
b60503ba
MW		 604 struct nvme_completion *cqe)
605{
eca18b23 606 struct nvme_iod *iod = ctx;
ac3dd5bd 607 struct request *req = iod_get_private(iod);
a4aea562 608 struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
b60503ba 609 u16 status = le16_to_cpup(&cqe->status) >> 1;
ef658fc2 610 int error = 0;
b60503ba 611
edd10d33 612 if (unlikely(status)) {
a4aea562
MB		 613 if (!(status & NVME_SC_DNR || blk_noretry_request(req))
614 && (jiffies - req->start_time) < req->timeout) {
c9d3bf88
KB		 615 unsigned long flags;
616
a4aea562 617 blk_mq_requeue_request(req);
c9d3bf88
KB		 618 spin_lock_irqsave(req->q->queue_lock, flags);
619 if (!blk_queue_stopped(req->q))
620 blk_mq_kick_requeue_list(req->q);
621 spin_unlock_irqrestore(req->q->queue_lock, flags);
edd10d33
KB		 622 return;
623 }
f4829a9b 624
d29ec824 625 if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
17188bb4 626 if (cmd_rq->ctx == CMD_CTX_CANCELLED)
1951feae
CH		 627 error = -EINTR;
628 else
629 error = status;
d29ec824 630 } else {
1951feae 631 error = nvme_error_status(status);
d29ec824 632 }
f4829a9b
CH		 633 }
634
a0a931d6
KB		 635 if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
636 u32 result = le32_to_cpup(&cqe->result);
637 req->special = (void *)(uintptr_t)result;
638 }
a4aea562
MB		 639
640 if (cmd_rq->aborted)
e75ec752 641 dev_warn(nvmeq->dev->dev,
a4aea562 642 "completing aborted command with status:%04x\n",
1951feae 643 error);
a4aea562 644
e1e5e564 645 if (iod->nents) {
e75ec752 646 dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents,
a4aea562 647 rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
e1e5e564
KB		 648 if (blk_integrity_rq(req)) {
649 if (!rq_data_dir(req))
650 nvme_dif_remap(req, nvme_dif_complete);
e75ec752 651 dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1,
e1e5e564
KB		 652 rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
653 }
654 }
edd10d33 655 nvme_free_iod(nvmeq->dev, iod);
3291fa57 656
1951feae 657 blk_mq_complete_request(req, error);
b60503ba
MW		 658}
659
184d2944 660/* length is in bytes. gfp flags indicates whether we may sleep. */
d29ec824
CH		 661static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod,
662 int total_len, gfp_t gfp)
ff22b54f 663{
99802a7a 664 struct dma_pool *pool;
eca18b23
MW		 665 int length = total_len;
666 struct scatterlist *sg = iod->sg;
ff22b54f
MW		 667 int dma_len = sg_dma_len(sg);
668 u64 dma_addr = sg_dma_address(sg);
f137e0f1
MI		 669 u32 page_size = dev->page_size;
670 int offset = dma_addr & (page_size - 1);
e025344c 671 __le64 *prp_list;
eca18b23 672 __le64 **list = iod_list(iod);
e025344c 673 dma_addr_t prp_dma;
eca18b23 674 int nprps, i;
ff22b54f 675
1d090624 676 length -= (page_size - offset);
ff22b54f 677 if (length <= 0)
eca18b23 678 return total_len;
ff22b54f 679
1d090624 680 dma_len -= (page_size - offset);
ff22b54f 681 if (dma_len) {
1d090624 682 dma_addr += (page_size - offset);
ff22b54f
MW		 683 } else {
684 sg = sg_next(sg);
685 dma_addr = sg_dma_address(sg);
686 dma_len = sg_dma_len(sg);
687 }
688
1d090624 689 if (length <= page_size) {
edd10d33 690 iod->first_dma = dma_addr;
eca18b23 691 return total_len;
e025344c
SMM		 692 }
693
1d090624 694 nprps = DIV_ROUND_UP(length, page_size);
99802a7a
MW		 695 if (nprps <= (256 / 8)) {
696 pool = dev->prp_small_pool;
eca18b23 697 iod->npages = 0;
99802a7a
MW		 698 } else {
699 pool = dev->prp_page_pool;
eca18b23 700 iod->npages = 1;
99802a7a
MW		 701 }
702
b77954cb
MW		 703 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
704 if (!prp_list) {
edd10d33 705 iod->first_dma = dma_addr;
eca18b23 706 iod->npages = -1;
1d090624 707 return (total_len - length) + page_size;
b77954cb 708 }
eca18b23
MW		 709 list[0] = prp_list;
710 iod->first_dma = prp_dma;
e025344c
SMM		 711 i = 0;
712 for (;;) {
1d090624 713 if (i == page_size >> 3) {
e025344c 714 __le64 *old_prp_list = prp_list;
b77954cb 715 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
eca18b23
MW		 716 if (!prp_list)
717 return total_len - length;
718 list[iod->npages++] = prp_list;
7523d834
MW		 719 prp_list[0] = old_prp_list[i - 1];
720 old_prp_list[i - 1] = cpu_to_le64(prp_dma);
721 i = 1;
e025344c
SMM		 722 }
723 prp_list[i++] = cpu_to_le64(dma_addr);
1d090624
KB		 724 dma_len -= page_size;
725 dma_addr += page_size;
726 length -= page_size;
e025344c
SMM		 727 if (length <= 0)
728 break;
729 if (dma_len > 0)
730 continue;
731 BUG_ON(dma_len < 0);
732 sg = sg_next(sg);
733 dma_addr = sg_dma_address(sg);
734 dma_len = sg_dma_len(sg);
ff22b54f
MW		 735 }
736
eca18b23 737 return total_len;
ff22b54f
MW		 738}
739
d29ec824
CH		 740static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req,
741 struct nvme_iod *iod)
742{
498c4394 743 struct nvme_command cmnd;
d29ec824 744
498c4394
JD		 745 memcpy(&cmnd, req->cmd, sizeof(cmnd));
746 cmnd.rw.command_id = req->tag;
d29ec824 747 if (req->nr_phys_segments) {
498c4394
JD		 748 cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
749 cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
d29ec824
CH		 750 }
751
498c4394 752 __nvme_submit_cmd(nvmeq, &cmnd);
d29ec824
CH		 753}
754
a4aea562
MB		 755/*
756 * We reuse the small pool to allocate the 16-byte range here as it is not
757 * worth having a special pool for these or additional cases to handle freeing
758 * the iod.
759 */
760static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns,
761 struct request *req, struct nvme_iod *iod)
0e5e4f0e 762{
edd10d33
KB		 763 struct nvme_dsm_range *range =
764 (struct nvme_dsm_range *)iod_list(iod)[0];
498c4394 765 struct nvme_command cmnd;
0e5e4f0e 766
0e5e4f0e 767 range->cattr = cpu_to_le32(0);
a4aea562
MB		 768 range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift);
769 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
0e5e4f0e 770
498c4394
JD		 771 memset(&cmnd, 0, sizeof(cmnd));
772 cmnd.dsm.opcode = nvme_cmd_dsm;
773 cmnd.dsm.command_id = req->tag;
774 cmnd.dsm.nsid = cpu_to_le32(ns->ns_id);
775 cmnd.dsm.prp1 = cpu_to_le64(iod->first_dma);
776 cmnd.dsm.nr = 0;
777 cmnd.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
0e5e4f0e 778
498c4394 779 __nvme_submit_cmd(nvmeq, &cmnd);
0e5e4f0e
KB		 780}
781
a4aea562 782static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns,
00df5cb4
MW		 783 int cmdid)
784{
498c4394 785 struct nvme_command cmnd;
00df5cb4 786
498c4394
JD		 787 memset(&cmnd, 0, sizeof(cmnd));
788 cmnd.common.opcode = nvme_cmd_flush;
789 cmnd.common.command_id = cmdid;
790 cmnd.common.nsid = cpu_to_le32(ns->ns_id);
00df5cb4 791
498c4394 792 __nvme_submit_cmd(nvmeq, &cmnd);
00df5cb4
MW		 793}
794
a4aea562
MB		 795static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod,
796 struct nvme_ns *ns)
b60503ba 797{
ac3dd5bd 798 struct request *req = iod_get_private(iod);
498c4394 799 struct nvme_command cmnd;
a4aea562
MB		 800 u16 control = 0;
801 u32 dsmgmt = 0;
00df5cb4 802
a4aea562 803 if (req->cmd_flags & REQ_FUA)
b60503ba 804 control |= NVME_RW_FUA;
a4aea562 805 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
b60503ba
MW		 806 control |= NVME_RW_LR;
807
a4aea562 808 if (req->cmd_flags & REQ_RAHEAD)
b60503ba
MW		 809 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
810
498c4394
JD		 811 memset(&cmnd, 0, sizeof(cmnd));
812 cmnd.rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
813 cmnd.rw.command_id = req->tag;
814 cmnd.rw.nsid = cpu_to_le32(ns->ns_id);
815 cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
816 cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
817 cmnd.rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
818 cmnd.rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
b60503ba 819
e19b127f 820 if (ns->ms) {
e1e5e564
KB		 821 switch (ns->pi_type) {
822 case NVME_NS_DPS_PI_TYPE3:
823 control |= NVME_RW_PRINFO_PRCHK_GUARD;
824 break;
825 case NVME_NS_DPS_PI_TYPE1:
826 case NVME_NS_DPS_PI_TYPE2:
827 control |= NVME_RW_PRINFO_PRCHK_GUARD |
828 NVME_RW_PRINFO_PRCHK_REF;
498c4394 829 cmnd.rw.reftag = cpu_to_le32(
e1e5e564
KB		 830 nvme_block_nr(ns, blk_rq_pos(req)));
831 break;
832 }
e19b127f
AP		 833 if (blk_integrity_rq(req))
834 cmnd.rw.metadata =
835 cpu_to_le64(sg_dma_address(iod->meta_sg));
836 else
837 control |= NVME_RW_PRINFO_PRACT;
838 }
e1e5e564 839
498c4394
JD		 840 cmnd.rw.control = cpu_to_le16(control);
841 cmnd.rw.dsmgmt = cpu_to_le32(dsmgmt);
b60503ba 842
498c4394 843 __nvme_submit_cmd(nvmeq, &cmnd);
b60503ba 844
1974b1ae 845 return 0;
edd10d33
KB		 846}
847
d29ec824
CH		 848/*
849 * NOTE: ns is NULL when called on the admin queue.
850 */
a4aea562
MB		 851static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
852 const struct blk_mq_queue_data *bd)
edd10d33 853{
a4aea562
MB		 854 struct nvme_ns *ns = hctx->queue->queuedata;
855 struct nvme_queue *nvmeq = hctx->driver_data;
d29ec824 856 struct nvme_dev *dev = nvmeq->dev;
a4aea562
MB		 857 struct request *req = bd->rq;
858 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
edd10d33 859 struct nvme_iod *iod;
a4aea562 860 enum dma_data_direction dma_dir;
edd10d33 861
e1e5e564
KB		 862 /*
863 * If formated with metadata, require the block layer provide a buffer
864 * unless this namespace is formated such that the metadata can be
865 * stripped/generated by the controller with PRACT=1.
866 */
d29ec824 867 if (ns && ns->ms && !blk_integrity_rq(req)) {
71feb364
KB		 868 if (!(ns->pi_type && ns->ms == 8) &&
869 req->cmd_type != REQ_TYPE_DRV_PRIV) {
f4829a9b 870 blk_mq_complete_request(req, -EFAULT);
e1e5e564
KB		 871 return BLK_MQ_RQ_QUEUE_OK;
872 }
873 }
874
d29ec824 875 iod = nvme_alloc_iod(req, dev, GFP_ATOMIC);
edd10d33 876 if (!iod)
fe54303e 877 return BLK_MQ_RQ_QUEUE_BUSY;
a4aea562 878
a4aea562 879 if (req->cmd_flags & REQ_DISCARD) {
edd10d33
KB		 880 void *range;
881 /*
882 * We reuse the small pool to allocate the 16-byte range here
883 * as it is not worth having a special pool for these or
884 * additional cases to handle freeing the iod.
885 */
d29ec824 886 range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC,
edd10d33 887 &iod->first_dma);
a4aea562 888 if (!range)
fe54303e 889 goto retry_cmd;
edd10d33
KB		 890 iod_list(iod)[0] = (__le64 *)range;
891 iod->npages = 0;
ac3dd5bd 892 } else if (req->nr_phys_segments) {
a4aea562
MB		 893 dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
894
ac3dd5bd 895 sg_init_table(iod->sg, req->nr_phys_segments);
a4aea562 896 iod->nents = blk_rq_map_sg(req->q, req, iod->sg);
fe54303e
JA		 897 if (!iod->nents)
898 goto error_cmd;
a4aea562
MB		 899
900 if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir))
fe54303e 901 goto retry_cmd;
a4aea562 902
fe54303e 903 if (blk_rq_bytes(req) !=
d29ec824
CH		 904 nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) {
905 dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
fe54303e
JA		 906 goto retry_cmd;
907 }
e1e5e564
KB		 908 if (blk_integrity_rq(req)) {
909 if (blk_rq_count_integrity_sg(req->q, req->bio) != 1)
910 goto error_cmd;
911
912 sg_init_table(iod->meta_sg, 1);
913 if (blk_rq_map_integrity_sg(
914 req->q, req->bio, iod->meta_sg) != 1)
915 goto error_cmd;
916
917 if (rq_data_dir(req))
918 nvme_dif_remap(req, nvme_dif_prep);
919
920 if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir))
921 goto error_cmd;
922 }
edd10d33 923 }
1974b1ae 924
9af8785a 925 nvme_set_info(cmd, iod, req_completion);
a4aea562 926 spin_lock_irq(&nvmeq->q_lock);
d29ec824
CH		 927 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
928 nvme_submit_priv(nvmeq, req, iod);
929 else if (req->cmd_flags & REQ_DISCARD)
a4aea562
MB		 930 nvme_submit_discard(nvmeq, ns, req, iod);
931 else if (req->cmd_flags & REQ_FLUSH)
932 nvme_submit_flush(nvmeq, ns, req->tag);
933 else
934 nvme_submit_iod(nvmeq, iod, ns);
935
936 nvme_process_cq(nvmeq);
937 spin_unlock_irq(&nvmeq->q_lock);
938 return BLK_MQ_RQ_QUEUE_OK;
939
fe54303e 940 error_cmd:
d29ec824 941 nvme_free_iod(dev, iod);
fe54303e
JA		 942 return BLK_MQ_RQ_QUEUE_ERROR;
943 retry_cmd:
d29ec824 944 nvme_free_iod(dev, iod);
fe54303e 945 return BLK_MQ_RQ_QUEUE_BUSY;
b60503ba
MW		 946}
947
e9539f47 948static int nvme_process_cq(struct nvme_queue *nvmeq)
b60503ba 949{
82123460 950 u16 head, phase;
b60503ba 951
b60503ba 952 head = nvmeq->cq_head;
82123460 953 phase = nvmeq->cq_phase;
b60503ba
MW		 954
955 for (;;) {
c2f5b650
MW		 956 void *ctx;
957 nvme_completion_fn fn;
b60503ba 958 struct nvme_completion cqe = nvmeq->cqes[head];
82123460 959 if ((le16_to_cpu(cqe.status) & 1) != phase)
b60503ba
MW		 960 break;
961 nvmeq->sq_head = le16_to_cpu(cqe.sq_head);
962 if (++head == nvmeq->q_depth) {
963 head = 0;
82123460 964 phase = !phase;
b60503ba 965 }
a4aea562 966 ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn);
edd10d33 967 fn(nvmeq, ctx, &cqe);
b60503ba
MW		 968 }
969
970 /* If the controller ignores the cq head doorbell and continuously
971 * writes to the queue, it is theoretically possible to wrap around
972 * the queue twice and mistakenly return IRQ_NONE. Linux only
973 * requires that 0.1% of your interrupts are handled, so this isn't
974 * a big problem.
975 */
82123460 976 if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
e9539f47 977 return 0;
b60503ba 978
b80d5ccc 979 writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
b60503ba 980 nvmeq->cq_head = head;
82123460 981 nvmeq->cq_phase = phase;
b60503ba 982
e9539f47
MW		 983 nvmeq->cqe_seen = 1;
984 return 1;
b60503ba
MW		 985}
986
987static irqreturn_t nvme_irq(int irq, void *data)
58ffacb5
MW		 988{
989 irqreturn_t result;
990 struct nvme_queue *nvmeq = data;
991 spin_lock(&nvmeq->q_lock);
e9539f47
MW		 992 nvme_process_cq(nvmeq);
993 result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
994 nvmeq->cqe_seen = 0;
58ffacb5
MW		 995 spin_unlock(&nvmeq->q_lock);
996 return result;
997}
998
999static irqreturn_t nvme_irq_check(int irq, void *data)
1000{
1001 struct nvme_queue *nvmeq = data;
1002 struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head];
1003 if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase)
1004 return IRQ_NONE;
1005 return IRQ_WAKE_THREAD;
1006}
1007
b60503ba
MW		 1008/*
1009 * Returns 0 on success. If the result is negative, it's a Linux error code;
1010 * if the result is positive, it's an NVM Express status code
1011 */
d29ec824
CH		 1012int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1013 void *buffer, void __user *ubuffer, unsigned bufflen,
1014 u32 *result, unsigned timeout)
b60503ba 1015{
d29ec824
CH		 1016 bool write = cmd->common.opcode & 1;
1017 struct bio *bio = NULL;
f705f837 1018 struct request *req;
d29ec824 1019 int ret;
b60503ba 1020
d29ec824 1021 req = blk_mq_alloc_request(q, write, GFP_KERNEL, false);
f705f837
CH		 1022 if (IS_ERR(req))
1023 return PTR_ERR(req);
b60503ba 1024
d29ec824 1025 req->cmd_type = REQ_TYPE_DRV_PRIV;
e112af0d 1026 req->cmd_flags |= REQ_FAILFAST_DRIVER;
d29ec824
CH		 1027 req->__data_len = 0;
1028 req->__sector = (sector_t) -1;
1029 req->bio = req->biotail = NULL;
b60503ba 1030
f4ff414a 1031 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
a4aea562 1032
d29ec824
CH		 1033 req->cmd = (unsigned char *)cmd;
1034 req->cmd_len = sizeof(struct nvme_command);
a0a931d6 1035 req->special = (void *)0;
b60503ba 1036
d29ec824
CH		 1037 if (buffer && bufflen) {
1038 ret = blk_rq_map_kern(q, req, buffer, bufflen, __GFP_WAIT);
1039 if (ret)
1040 goto out;
1041 } else if (ubuffer && bufflen) {
1042 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, __GFP_WAIT);
1043 if (ret)
1044 goto out;
1045 bio = req->bio;
1046 }
3c0cf138 1047
d29ec824
CH		 1048 blk_execute_rq(req->q, NULL, req, 0);
1049 if (bio)
1050 blk_rq_unmap_user(bio);
b60503ba 1051 if (result)
a0a931d6 1052 *result = (u32)(uintptr_t)req->special;
d29ec824
CH		 1053 ret = req->errors;
1054 out:
f705f837 1055 blk_mq_free_request(req);
d29ec824 1056 return ret;
f705f837
CH		 1057}
1058
d29ec824
CH		 1059int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1060 void *buffer, unsigned bufflen)
f705f837 1061{
d29ec824 1062 return __nvme_submit_sync_cmd(q, cmd, buffer, NULL, bufflen, NULL, 0);
b60503ba
MW		 1063}
1064
a4aea562
MB		 1065static int nvme_submit_async_admin_req(struct nvme_dev *dev)
1066{
1067 struct nvme_queue *nvmeq = dev->queues[0];
1068 struct nvme_command c;
1069 struct nvme_cmd_info *cmd_info;
1070 struct request *req;
1071
1efccc9d 1072 req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, true);
9f173b33
DC		 1073 if (IS_ERR(req))
1074 return PTR_ERR(req);
a4aea562 1075
c917dfe5 1076 req->cmd_flags |= REQ_NO_TIMEOUT;
a4aea562 1077 cmd_info = blk_mq_rq_to_pdu(req);
1efccc9d 1078 nvme_set_info(cmd_info, NULL, async_req_completion);
a4aea562
MB		 1079
1080 memset(&c, 0, sizeof(c));
1081 c.common.opcode = nvme_admin_async_event;
1082 c.common.command_id = req->tag;
1083
42483228 1084 blk_mq_free_request(req);
e3f879bf
SB		 1085 __nvme_submit_cmd(nvmeq, &c);
1086 return 0;
a4aea562
MB		 1087}
1088
1089static int nvme_submit_admin_async_cmd(struct nvme_dev *dev,
4d115420
KB		 1090 struct nvme_command *cmd,
1091 struct async_cmd_info *cmdinfo, unsigned timeout)
1092{
a4aea562
MB		 1093 struct nvme_queue *nvmeq = dev->queues[0];
1094 struct request *req;
1095 struct nvme_cmd_info *cmd_rq;
4d115420 1096
a4aea562 1097 req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_KERNEL, false);
9f173b33
DC		 1098 if (IS_ERR(req))
1099 return PTR_ERR(req);
a4aea562
MB		 1100
1101 req->timeout = timeout;
1102 cmd_rq = blk_mq_rq_to_pdu(req);
1103 cmdinfo->req = req;
1104 nvme_set_info(cmd_rq, cmdinfo, async_completion);
4d115420 1105 cmdinfo->status = -EINTR;
a4aea562
MB		 1106
1107 cmd->common.command_id = req->tag;
1108
e3f879bf
SB		 1109 nvme_submit_cmd(nvmeq, cmd);
1110 return 0;
4d115420
KB		 1111}
1112
b60503ba
MW		 1113static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
1114{
b60503ba
MW		 1115 struct nvme_command c;
1116
1117 memset(&c, 0, sizeof(c));
1118 c.delete_queue.opcode = opcode;
1119 c.delete_queue.qid = cpu_to_le16(id);
1120
d29ec824 1121 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1122}
1123
1124static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
1125 struct nvme_queue *nvmeq)
1126{
b60503ba
MW		 1127 struct nvme_command c;
1128 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;
1129
d29ec824
CH		 1130 /*
1131 * Note: we (ab)use the fact the the prp fields survive if no data
1132 * is attached to the request.
1133 */
b60503ba
MW		 1134 memset(&c, 0, sizeof(c));
1135 c.create_cq.opcode = nvme_admin_create_cq;
1136 c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr);
1137 c.create_cq.cqid = cpu_to_le16(qid);
1138 c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
1139 c.create_cq.cq_flags = cpu_to_le16(flags);
1140 c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector);
1141
d29ec824 1142 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1143}
1144
1145static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
1146 struct nvme_queue *nvmeq)
1147{
b60503ba
MW		 1148 struct nvme_command c;
1149 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM;
1150
d29ec824
CH		 1151 /*
1152 * Note: we (ab)use the fact the the prp fields survive if no data
1153 * is attached to the request.
1154 */
b60503ba
MW		 1155 memset(&c, 0, sizeof(c));
1156 c.create_sq.opcode = nvme_admin_create_sq;
1157 c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr);
1158 c.create_sq.sqid = cpu_to_le16(qid);
1159 c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
1160 c.create_sq.sq_flags = cpu_to_le16(flags);
1161 c.create_sq.cqid = cpu_to_le16(qid);
1162
d29ec824 1163 return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
b60503ba
MW		 1164}
1165
1166static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid)
1167{
1168 return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid);
1169}
1170
1171static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid)
1172{
1173 return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid);
1174}
1175
d29ec824 1176int nvme_identify_ctrl(struct nvme_dev *dev, struct nvme_id_ctrl **id)
bc5fc7e4 1177{
e44ac588 1178 struct nvme_command c = { };
d29ec824 1179 int error;
bc5fc7e4 1180
e44ac588
AM		 1181 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1182 c.identify.opcode = nvme_admin_identify;
1183 c.identify.cns = cpu_to_le32(1);
1184
d29ec824
CH		 1185 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1186 if (!*id)
1187 return -ENOMEM;
bc5fc7e4 1188
d29ec824
CH		 1189 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1190 sizeof(struct nvme_id_ctrl));
1191 if (error)
1192 kfree(*id);
1193 return error;
1194}
1195
1196int nvme_identify_ns(struct nvme_dev *dev, unsigned nsid,
1197 struct nvme_id_ns **id)
1198{
e44ac588 1199 struct nvme_command c = { };
d29ec824 1200 int error;
bc5fc7e4 1201
e44ac588
AM		 1202 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1203 c.identify.opcode = nvme_admin_identify,
1204 c.identify.nsid = cpu_to_le32(nsid),
1205
d29ec824
CH		 1206 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
1207 if (!*id)
1208 return -ENOMEM;
1209
1210 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1211 sizeof(struct nvme_id_ns));
1212 if (error)
1213 kfree(*id);
1214 return error;
bc5fc7e4
MW		 1215}
1216
5d0f6131 1217int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid,
08df1e05 1218 dma_addr_t dma_addr, u32 *result)
bc5fc7e4
MW		 1219{
1220 struct nvme_command c;
1221
1222 memset(&c, 0, sizeof(c));
1223 c.features.opcode = nvme_admin_get_features;
a42cecce 1224 c.features.nsid = cpu_to_le32(nsid);
bc5fc7e4
MW		 1225 c.features.prp1 = cpu_to_le64(dma_addr);
1226 c.features.fid = cpu_to_le32(fid);
bc5fc7e4 1227
d29ec824
CH		 1228 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0,
1229 result, 0);
df348139
MW		 1230}
1231
5d0f6131
VV		 1232int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11,
1233 dma_addr_t dma_addr, u32 *result)
df348139
MW		 1234{
1235 struct nvme_command c;
1236
1237 memset(&c, 0, sizeof(c));
1238 c.features.opcode = nvme_admin_set_features;
1239 c.features.prp1 = cpu_to_le64(dma_addr);
1240 c.features.fid = cpu_to_le32(fid);
1241 c.features.dword11 = cpu_to_le32(dword11);
1242
d29ec824
CH		 1243 return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0,
1244 result, 0);
1245}
1246
1247int nvme_get_log_page(struct nvme_dev *dev, struct nvme_smart_log **log)
1248{
e44ac588
AM		 1249 struct nvme_command c = { };
1250 int error;
1251
1252 c.common.opcode = nvme_admin_get_log_page,
1253 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
1254 c.common.cdw10[0] = cpu_to_le32(
d29ec824
CH		 1255 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
1256 NVME_LOG_SMART),
d29ec824
CH		 1257
1258 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
1259 if (!*log)
1260 return -ENOMEM;
1261
1262 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
1263 sizeof(struct nvme_smart_log));
1264 if (error)
1265 kfree(*log);
1266 return error;
bc5fc7e4
MW		 1267}
1268
c30341dc 1269/**
a4aea562 1270 * nvme_abort_req - Attempt aborting a request
c30341dc
KB		 1271 *
1272 * Schedule controller reset if the command was already aborted once before and
1273 * still hasn't been returned to the driver, or if this is the admin queue.
1274 */
a4aea562 1275static void nvme_abort_req(struct request *req)
c30341dc 1276{
a4aea562
MB		 1277 struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
1278 struct nvme_queue *nvmeq = cmd_rq->nvmeq;
c30341dc 1279 struct nvme_dev *dev = nvmeq->dev;
a4aea562
MB		 1280 struct request *abort_req;
1281 struct nvme_cmd_info *abort_cmd;
1282 struct nvme_command cmd;
c30341dc 1283
a4aea562 1284 if (!nvmeq->qid || cmd_rq->aborted) {
90667892
CH		 1285 spin_lock(&dev_list_lock);
1286 if (!__nvme_reset(dev)) {
1287 dev_warn(dev->dev,
1288 "I/O %d QID %d timeout, reset controller\n",
1289 req->tag, nvmeq->qid);
1290 }
1291 spin_unlock(&dev_list_lock);
c30341dc
KB		 1292 return;
1293 }
1294
1295 if (!dev->abort_limit)
1296 return;
1297
a4aea562
MB		 1298 abort_req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC,
1299 false);
9f173b33 1300 if (IS_ERR(abort_req))
c30341dc
KB		 1301 return;
1302
a4aea562
MB		 1303 abort_cmd = blk_mq_rq_to_pdu(abort_req);
1304 nvme_set_info(abort_cmd, abort_req, abort_completion);
1305
c30341dc
KB		 1306 memset(&cmd, 0, sizeof(cmd));
1307 cmd.abort.opcode = nvme_admin_abort_cmd;
a4aea562 1308 cmd.abort.cid = req->tag;
c30341dc 1309 cmd.abort.sqid = cpu_to_le16(nvmeq->qid);
a4aea562 1310 cmd.abort.command_id = abort_req->tag;
c30341dc
KB		 1311
1312 --dev->abort_limit;
a4aea562 1313 cmd_rq->aborted = 1;
c30341dc 1314
a4aea562 1315 dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag,
c30341dc 1316 nvmeq->qid);
e3f879bf 1317 nvme_submit_cmd(dev->queues[0], &cmd);
c30341dc
KB		 1318}
1319
42483228 1320static void nvme_cancel_queue_ios(struct request *req, void *data, bool reserved)
a09115b2 1321{
a4aea562
MB		 1322 struct nvme_queue *nvmeq = data;
1323 void *ctx;
1324 nvme_completion_fn fn;
1325 struct nvme_cmd_info *cmd;
cef6a948
KB		 1326 struct nvme_completion cqe;
1327
1328 if (!blk_mq_request_started(req))
1329 return;
a09115b2 1330
a4aea562 1331 cmd = blk_mq_rq_to_pdu(req);
a09115b2 1332
a4aea562
MB		 1333 if (cmd->ctx == CMD_CTX_CANCELLED)
1334 return;
1335
cef6a948
KB		 1336 if (blk_queue_dying(req->q))
1337 cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1);
1338 else
1339 cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);
1340
1341
a4aea562
MB		 1342 dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n",
1343 req->tag, nvmeq->qid);
1344 ctx = cancel_cmd_info(cmd, &fn);
1345 fn(nvmeq, ctx, &cqe);
a09115b2
MW		 1346}
1347
a4aea562 1348static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
9e866774 1349{
a4aea562
MB		 1350 struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
1351 struct nvme_queue *nvmeq = cmd->nvmeq;
1352
1353 dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag,
1354 nvmeq->qid);
7a509a6b 1355 spin_lock_irq(&nvmeq->q_lock);
07836e65 1356 nvme_abort_req(req);
7a509a6b 1357 spin_unlock_irq(&nvmeq->q_lock);
a4aea562 1358
07836e65
KB		 1359 /*
1360 * The aborted req will be completed on receiving the abort req.
1361 * We enable the timer again. If hit twice, it'll cause a device reset,
1362 * as the device then is in a faulty state.
1363 */
1364 return BLK_EH_RESET_TIMER;
a4aea562 1365}
22404274 1366
a4aea562
MB		 1367static void nvme_free_queue(struct nvme_queue *nvmeq)
1368{
9e866774
MW		 1369 dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
1370 (void *)nvmeq->cqes, nvmeq->cq_dma_addr);
8ffaadf7
JD		 1371 if (nvmeq->sq_cmds)
1372 dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
9e866774
MW		 1373 nvmeq->sq_cmds, nvmeq->sq_dma_addr);
1374 kfree(nvmeq);
1375}
1376
a1a5ef99 1377static void nvme_free_queues(struct nvme_dev *dev, int lowest)
22404274
KB		 1378{
1379 int i;
1380
a1a5ef99 1381 for (i = dev->queue_count - 1; i >= lowest; i--) {
a4aea562 1382 struct nvme_queue *nvmeq = dev->queues[i];
22404274 1383 dev->queue_count--;
a4aea562 1384 dev->queues[i] = NULL;
f435c282 1385 nvme_free_queue(nvmeq);
121c7ad4 1386 }
22404274
KB		 1387}
1388
4d115420
KB		 1389/**
1390 * nvme_suspend_queue - put queue into suspended state
1391 * @nvmeq - queue to suspend
4d115420
KB		 1392 */
1393static int nvme_suspend_queue(struct nvme_queue *nvmeq)
b60503ba 1394{
2b25d981 1395 int vector;
b60503ba 1396
a09115b2 1397 spin_lock_irq(&nvmeq->q_lock);
2b25d981
KB		 1398 if (nvmeq->cq_vector == -1) {
1399 spin_unlock_irq(&nvmeq->q_lock);
1400 return 1;
1401 }
1402 vector = nvmeq->dev->entry[nvmeq->cq_vector].vector;
42f61420 1403 nvmeq->dev->online_queues--;
2b25d981 1404 nvmeq->cq_vector = -1;
a09115b2
MW		 1405 spin_unlock_irq(&nvmeq->q_lock);
1406
6df3dbc8
KB		 1407 if (!nvmeq->qid && nvmeq->dev->admin_q)
1408 blk_mq_freeze_queue_start(nvmeq->dev->admin_q);
1409
aba2080f
MW		 1410 irq_set_affinity_hint(vector, NULL);
1411 free_irq(vector, nvmeq);
b60503ba 1412
4d115420
KB		 1413 return 0;
1414}
b60503ba 1415
4d115420
KB		 1416static void nvme_clear_queue(struct nvme_queue *nvmeq)
1417{
22404274 1418 spin_lock_irq(&nvmeq->q_lock);
42483228
KB		 1419 if (nvmeq->tags && *nvmeq->tags)
1420 blk_mq_all_tag_busy_iter(*nvmeq->tags, nvme_cancel_queue_ios, nvmeq);
22404274 1421 spin_unlock_irq(&nvmeq->q_lock);
b60503ba
MW		 1422}
1423
4d115420
KB		 1424static void nvme_disable_queue(struct nvme_dev *dev, int qid)
1425{
a4aea562 1426 struct nvme_queue *nvmeq = dev->queues[qid];
4d115420
KB		 1427
1428 if (!nvmeq)
1429 return;
1430 if (nvme_suspend_queue(nvmeq))
1431 return;
1432
0e53d180
KB		 1433 /* Don't tell the adapter to delete the admin queue.
1434 * Don't tell a removed adapter to delete IO queues. */
1435 if (qid && readl(&dev->bar->csts) != -1) {
b60503ba
MW		 1436 adapter_delete_sq(dev, qid);
1437 adapter_delete_cq(dev, qid);
1438 }
07836e65
KB		 1439
1440 spin_lock_irq(&nvmeq->q_lock);
1441 nvme_process_cq(nvmeq);
1442 spin_unlock_irq(&nvmeq->q_lock);
b60503ba
MW		 1443}
1444
8ffaadf7
JD		 1445static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
1446 int entry_size)
1447{
1448 int q_depth = dev->q_depth;
1449 unsigned q_size_aligned = roundup(q_depth * entry_size, dev->page_size);
1450
1451 if (q_size_aligned * nr_io_queues > dev->cmb_size) {
c45f5c99
JD		 1452 u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1453 mem_per_q = round_down(mem_per_q, dev->page_size);
1454 q_depth = div_u64(mem_per_q, entry_size);
8ffaadf7
JD		 1455
1456 /*
1457 * Ensure the reduced q_depth is above some threshold where it
1458 * would be better to map queues in system memory with the
1459 * original depth
1460 */
1461 if (q_depth < 64)
1462 return -ENOMEM;
1463 }
1464
1465 return q_depth;
1466}
1467
1468static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1469 int qid, int depth)
1470{
1471 if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) {
1472 unsigned offset = (qid - 1) *
1473 roundup(SQ_SIZE(depth), dev->page_size);
1474 nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset;
1475 nvmeq->sq_cmds_io = dev->cmb + offset;
1476 } else {
1477 nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth),
1478 &nvmeq->sq_dma_addr, GFP_KERNEL);
1479 if (!nvmeq->sq_cmds)
1480 return -ENOMEM;
1481 }
1482
1483 return 0;
1484}
1485
b60503ba 1486static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
2b25d981 1487 int depth)
b60503ba 1488{
a4aea562 1489 struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
b60503ba
MW		 1490 if (!nvmeq)
1491 return NULL;
1492
e75ec752 1493 nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
4d51abf9 1494 &nvmeq->cq_dma_addr, GFP_KERNEL);
b60503ba
MW		 1495 if (!nvmeq->cqes)
1496 goto free_nvmeq;
b60503ba 1497
8ffaadf7 1498 if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
b60503ba
MW		 1499 goto free_cqdma;
1500
e75ec752 1501 nvmeq->q_dmadev = dev->dev;
091b6092 1502 nvmeq->dev = dev;
3193f07b
MW		 1503 snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
1504 dev->instance, qid);
b60503ba
MW		 1505 spin_lock_init(&nvmeq->q_lock);
1506 nvmeq->cq_head = 0;
82123460 1507 nvmeq->cq_phase = 1;
b80d5ccc 1508 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
b60503ba 1509 nvmeq->q_depth = depth;
c30341dc 1510 nvmeq->qid = qid;
758dd7fd 1511 nvmeq->cq_vector = -1;
a4aea562 1512 dev->queues[qid] = nvmeq;
b60503ba 1513
36a7e993
JD		 1514 /* make sure queue descriptor is set before queue count, for kthread */
1515 mb();
1516 dev->queue_count++;
1517
b60503ba
MW		 1518 return nvmeq;
1519
1520 free_cqdma:
e75ec752 1521 dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
b60503ba
MW		 1522 nvmeq->cq_dma_addr);
1523 free_nvmeq:
1524 kfree(nvmeq);
1525 return NULL;
1526}
1527
3001082c
MW		 1528static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1529 const char *name)
1530{
58ffacb5
MW		 1531 if (use_threaded_interrupts)
1532 return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector,
481e5bad 1533 nvme_irq_check, nvme_irq, IRQF_SHARED,
58ffacb5 1534 name, nvmeq);
3001082c 1535 return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq,
481e5bad 1536 IRQF_SHARED, name, nvmeq);
3001082c
MW		 1537}
1538
22404274 1539static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
b60503ba 1540{
22404274 1541 struct nvme_dev *dev = nvmeq->dev;
b60503ba 1542
7be50e93 1543 spin_lock_irq(&nvmeq->q_lock);
22404274
KB		 1544 nvmeq->sq_tail = 0;
1545 nvmeq->cq_head = 0;
1546 nvmeq->cq_phase = 1;
b80d5ccc 1547 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
22404274 1548 memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
42f61420 1549 dev->online_queues++;
7be50e93 1550 spin_unlock_irq(&nvmeq->q_lock);
22404274
KB		 1551}
1552
1553static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
1554{
1555 struct nvme_dev *dev = nvmeq->dev;
1556 int result;
3f85d50b 1557
2b25d981 1558 nvmeq->cq_vector = qid - 1;
b60503ba
MW		 1559 result = adapter_alloc_cq(dev, qid, nvmeq);
1560 if (result < 0)
22404274 1561 return result;
b60503ba
MW		 1562
1563 result = adapter_alloc_sq(dev, qid, nvmeq);
1564 if (result < 0)
1565 goto release_cq;
1566
3193f07b 1567 result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
b60503ba
MW		 1568 if (result < 0)
1569 goto release_sq;
1570
22404274 1571 nvme_init_queue(nvmeq, qid);
22404274 1572 return result;
b60503ba
MW		 1573
1574 release_sq:
1575 adapter_delete_sq(dev, qid);
1576 release_cq:
1577 adapter_delete_cq(dev, qid);
22404274 1578 return result;
b60503ba
MW		 1579}
1580
ba47e386
MW		 1581static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled)
1582{
1583 unsigned long timeout;
1584 u32 bit = enabled ? NVME_CSTS_RDY : 0;
1585
1586 timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1587
1588 while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) {
1589 msleep(100);
1590 if (fatal_signal_pending(current))
1591 return -EINTR;
1592 if (time_after(jiffies, timeout)) {
e75ec752 1593 dev_err(dev->dev,
27e8166c
MW		 1594 "Device not ready; aborting %s\n", enabled ?
1595 "initialisation" : "reset");
ba47e386
MW		 1596 return -ENODEV;
1597 }
1598 }
1599
1600 return 0;
1601}
1602
1603/*
1604 * If the device has been passed off to us in an enabled state, just clear
1605 * the enabled bit. The spec says we should set the 'shutdown notification
1606 * bits', but doing so may cause the device to complete commands to the
1607 * admin queue ... and we don't know what memory that might be pointing at!
1608 */
1609static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap)
1610{
01079522
DM		 1611 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1612 dev->ctrl_config &= ~NVME_CC_ENABLE;
1613 writel(dev->ctrl_config, &dev->bar->cc);
44af146a 1614
ba47e386
MW		 1615 return nvme_wait_ready(dev, cap, false);
1616}
1617
1618static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap)
1619{
01079522
DM		 1620 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1621 dev->ctrl_config |= NVME_CC_ENABLE;
1622 writel(dev->ctrl_config, &dev->bar->cc);
1623
ba47e386
MW		 1624 return nvme_wait_ready(dev, cap, true);
1625}
1626
1894d8f1
KB		 1627static int nvme_shutdown_ctrl(struct nvme_dev *dev)
1628{
1629 unsigned long timeout;
1894d8f1 1630
01079522
DM		 1631 dev->ctrl_config &= ~NVME_CC_SHN_MASK;
1632 dev->ctrl_config |= NVME_CC_SHN_NORMAL;
1633
1634 writel(dev->ctrl_config, &dev->bar->cc);
1894d8f1 1635
2484f407 1636 timeout = SHUTDOWN_TIMEOUT + jiffies;
1894d8f1
KB		 1637 while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) !=
1638 NVME_CSTS_SHST_CMPLT) {
1639 msleep(100);
1640 if (fatal_signal_pending(current))
1641 return -EINTR;
1642 if (time_after(jiffies, timeout)) {
e75ec752 1643 dev_err(dev->dev,
1894d8f1
KB		 1644 "Device shutdown incomplete; abort shutdown\n");
1645 return -ENODEV;
1646 }
1647 }
1648
1649 return 0;
1650}
1651
a4aea562 1652static struct blk_mq_ops nvme_mq_admin_ops = {
d29ec824 1653 .queue_rq = nvme_queue_rq,
a4aea562
MB		 1654 .map_queue = blk_mq_map_queue,
1655 .init_hctx = nvme_admin_init_hctx,
4af0e21c 1656 .exit_hctx = nvme_admin_exit_hctx,
a4aea562
MB		 1657 .init_request = nvme_admin_init_request,
1658 .timeout = nvme_timeout,
1659};
1660
1661static struct blk_mq_ops nvme_mq_ops = {
1662 .queue_rq = nvme_queue_rq,
1663 .map_queue = blk_mq_map_queue,
1664 .init_hctx = nvme_init_hctx,
1665 .init_request = nvme_init_request,
1666 .timeout = nvme_timeout,
1667};
1668
ea191d2f
KB		 1669static void nvme_dev_remove_admin(struct nvme_dev *dev)
1670{
1671 if (dev->admin_q && !blk_queue_dying(dev->admin_q)) {
1672 blk_cleanup_queue(dev->admin_q);
1673 blk_mq_free_tag_set(&dev->admin_tagset);
1674 }
1675}
1676
a4aea562
MB		 1677static int nvme_alloc_admin_tags(struct nvme_dev *dev)
1678{
1679 if (!dev->admin_q) {
1680 dev->admin_tagset.ops = &nvme_mq_admin_ops;
1681 dev->admin_tagset.nr_hw_queues = 1;
1682 dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1;
1efccc9d 1683 dev->admin_tagset.reserved_tags = 1;
a4aea562 1684 dev->admin_tagset.timeout = ADMIN_TIMEOUT;
e75ec752 1685 dev->admin_tagset.numa_node = dev_to_node(dev->dev);
ac3dd5bd 1686 dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
a4aea562
MB		 1687 dev->admin_tagset.driver_data = dev;
1688
1689 if (blk_mq_alloc_tag_set(&dev->admin_tagset))
1690 return -ENOMEM;
1691
1692 dev->admin_q = blk_mq_init_queue(&dev->admin_tagset);
35b489d3 1693 if (IS_ERR(dev->admin_q)) {
a4aea562
MB		 1694 blk_mq_free_tag_set(&dev->admin_tagset);
1695 return -ENOMEM;
1696 }
ea191d2f
KB		 1697 if (!blk_get_queue(dev->admin_q)) {
1698 nvme_dev_remove_admin(dev);
4af0e21c 1699 dev->admin_q = NULL;
ea191d2f
KB		 1700 return -ENODEV;
1701 }
0fb59cbc
KB		 1702 } else
1703 blk_mq_unfreeze_queue(dev->admin_q);
a4aea562
MB		 1704
1705 return 0;
1706}
1707
8d85fce7 1708static int nvme_configure_admin_queue(struct nvme_dev *dev)
b60503ba 1709{
ba47e386 1710 int result;
b60503ba 1711 u32 aqa;
ba47e386 1712 u64 cap = readq(&dev->bar->cap);
b60503ba 1713 struct nvme_queue *nvmeq;
1d090624
KB		 1714 unsigned page_shift = PAGE_SHIFT;
1715 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12;
1716 unsigned dev_page_max = NVME_CAP_MPSMAX(cap) + 12;
1717
1718 if (page_shift < dev_page_min) {
e75ec752 1719 dev_err(dev->dev,
1d090624
KB		 1720 "Minimum device page size (%u) too large for "
1721 "host (%u)\n", 1 << dev_page_min,
1722 1 << page_shift);
1723 return -ENODEV;
1724 }
1725 if (page_shift > dev_page_max) {
e75ec752 1726 dev_info(dev->dev,
1d090624
KB		 1727 "Device maximum page size (%u) smaller than "
1728 "host (%u); enabling work-around\n",
1729 1 << dev_page_max, 1 << page_shift);
1730 page_shift = dev_page_max;
1731 }
b60503ba 1732
dfbac8c7
KB		 1733 dev->subsystem = readl(&dev->bar->vs) >= NVME_VS(1, 1) ?
1734 NVME_CAP_NSSRC(cap) : 0;
1735
1736 if (dev->subsystem && (readl(&dev->bar->csts) & NVME_CSTS_NSSRO))
1737 writel(NVME_CSTS_NSSRO, &dev->bar->csts);
1738
ba47e386
MW		 1739 result = nvme_disable_ctrl(dev, cap);
1740 if (result < 0)
1741 return result;
b60503ba 1742
a4aea562 1743 nvmeq = dev->queues[0];
cd638946 1744 if (!nvmeq) {
2b25d981 1745 nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
cd638946
KB		 1746 if (!nvmeq)
1747 return -ENOMEM;
cd638946 1748 }
b60503ba
MW		 1749
1750 aqa = nvmeq->q_depth - 1;
1751 aqa |= aqa << 16;
1752
1d090624
KB		 1753 dev->page_size = 1 << page_shift;
1754
01079522 1755 dev->ctrl_config = NVME_CC_CSS_NVM;
1d090624 1756 dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
b60503ba 1757 dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
7f53f9d2 1758 dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
b60503ba
MW		 1759
1760 writel(aqa, &dev->bar->aqa);
1761 writeq(nvmeq->sq_dma_addr, &dev->bar->asq);
1762 writeq(nvmeq->cq_dma_addr, &dev->bar->acq);
b60503ba 1763
ba47e386 1764 result = nvme_enable_ctrl(dev, cap);
025c557a 1765 if (result)
a4aea562
MB		 1766 goto free_nvmeq;
1767
2b25d981 1768 nvmeq->cq_vector = 0;
3193f07b 1769 result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
758dd7fd
JD		 1770 if (result) {
1771 nvmeq->cq_vector = -1;
0fb59cbc 1772 goto free_nvmeq;
758dd7fd 1773 }
025c557a 1774
b60503ba 1775 return result;
a4aea562 1776
a4aea562
MB		 1777 free_nvmeq:
1778 nvme_free_queues(dev, 0);
1779 return result;
b60503ba
MW		 1780}
1781
a53295b6
MW		 1782static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1783{
1784 struct nvme_dev *dev = ns->dev;
a53295b6
MW		 1785 struct nvme_user_io io;
1786 struct nvme_command c;
d29ec824 1787 unsigned length, meta_len;
a67a9513 1788 int status, write;
a67a9513
KB		 1789 dma_addr_t meta_dma = 0;
1790 void *meta = NULL;
fec558b5 1791 void __user *metadata;
a53295b6
MW		 1792
1793 if (copy_from_user(&io, uio, sizeof(io)))
1794 return -EFAULT;
6c7d4945
MW		 1795
1796 switch (io.opcode) {
1797 case nvme_cmd_write:
1798 case nvme_cmd_read:
6bbf1acd 1799 case nvme_cmd_compare:
6413214c 1800 break;
6c7d4945 1801 default:
6bbf1acd 1802 return -EINVAL;
6c7d4945
MW		 1803 }
1804
d29ec824
CH		 1805 length = (io.nblocks + 1) << ns->lba_shift;
1806 meta_len = (io.nblocks + 1) * ns->ms;
3d42e67f 1807 metadata = (void __user *)(uintptr_t)io.metadata;
d29ec824 1808 write = io.opcode & 1;
a53295b6 1809
71feb364
KB		 1810 if (ns->ext) {
1811 length += meta_len;
1812 meta_len = 0;
a67a9513
KB		 1813 }
1814 if (meta_len) {
d29ec824
CH		 1815 if (((io.metadata & 3) || !io.metadata) && !ns->ext)
1816 return -EINVAL;
1817
e75ec752 1818 meta = dma_alloc_coherent(dev->dev, meta_len,
a67a9513 1819 &meta_dma, GFP_KERNEL);
fec558b5 1820
a67a9513
KB		 1821 if (!meta) {
1822 status = -ENOMEM;
1823 goto unmap;
1824 }
1825 if (write) {
fec558b5 1826 if (copy_from_user(meta, metadata, meta_len)) {
a67a9513
KB		 1827 status = -EFAULT;
1828 goto unmap;
1829 }
1830 }
1831 }
1832
a53295b6
MW		 1833 memset(&c, 0, sizeof(c));
1834 c.rw.opcode = io.opcode;
1835 c.rw.flags = io.flags;
6c7d4945 1836 c.rw.nsid = cpu_to_le32(ns->ns_id);
a53295b6 1837 c.rw.slba = cpu_to_le64(io.slba);
6c7d4945 1838 c.rw.length = cpu_to_le16(io.nblocks);
a53295b6 1839 c.rw.control = cpu_to_le16(io.control);
1c9b5265
MW		 1840 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1841 c.rw.reftag = cpu_to_le32(io.reftag);
1842 c.rw.apptag = cpu_to_le16(io.apptag);
1843 c.rw.appmask = cpu_to_le16(io.appmask);
a67a9513 1844 c.rw.metadata = cpu_to_le64(meta_dma);
d29ec824
CH		 1845
1846 status = __nvme_submit_sync_cmd(ns->queue, &c, NULL,
3d42e67f 1847 (void __user *)(uintptr_t)io.addr, length, NULL, 0);
f410c680 1848 unmap:
a67a9513
KB		 1849 if (meta) {
1850 if (status == NVME_SC_SUCCESS && !write) {
fec558b5 1851 if (copy_to_user(metadata, meta, meta_len))
a67a9513
KB		 1852 status = -EFAULT;
1853 }
e75ec752 1854 dma_free_coherent(dev->dev, meta_len, meta, meta_dma);
f410c680 1855 }
a53295b6
MW		 1856 return status;
1857}
1858
a4aea562
MB		 1859static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns,
1860 struct nvme_passthru_cmd __user *ucmd)
6ee44cdc 1861{
7963e521 1862 struct nvme_passthru_cmd cmd;
6ee44cdc 1863 struct nvme_command c;
d29ec824
CH		 1864 unsigned timeout = 0;
1865 int status;
6ee44cdc 1866
6bbf1acd
MW		 1867 if (!capable(CAP_SYS_ADMIN))
1868 return -EACCES;
1869 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
6ee44cdc 1870 return -EFAULT;
6ee44cdc
MW		 1871
1872 memset(&c, 0, sizeof(c));
6bbf1acd
MW		 1873 c.common.opcode = cmd.opcode;
1874 c.common.flags = cmd.flags;
1875 c.common.nsid = cpu_to_le32(cmd.nsid);
1876 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1877 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1878 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1879 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1880 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1881 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1882 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1883 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1884
d29ec824
CH		 1885 if (cmd.timeout_ms)
1886 timeout = msecs_to_jiffies(cmd.timeout_ms);
eca18b23 1887
f705f837 1888 status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c,
3d42e67f 1889 NULL, (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
d29ec824
CH		 1890 &cmd.result, timeout);
1891 if (status >= 0) {
1892 if (put_user(cmd.result, &ucmd->result))
1893 return -EFAULT;
6bbf1acd 1894 }
f4f117f6 1895
6ee44cdc
MW		 1896 return status;
1897}
1898
81f03fed
JD		 1899static int nvme_subsys_reset(struct nvme_dev *dev)
1900{
1901 if (!dev->subsystem)
1902 return -ENOTTY;
1903
1904 writel(0x4E564D65, &dev->bar->nssr); /* "NVMe" */
1905 return 0;
1906}
1907
b60503ba
MW		 1908static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd,
1909 unsigned long arg)
1910{
1911 struct nvme_ns *ns = bdev->bd_disk->private_data;
1912
1913 switch (cmd) {
6bbf1acd 1914 case NVME_IOCTL_ID:
c3bfe717 1915 force_successful_syscall_return();
6bbf1acd
MW		 1916 return ns->ns_id;
1917 case NVME_IOCTL_ADMIN_CMD:
a4aea562 1918 return nvme_user_cmd(ns->dev, NULL, (void __user *)arg);
7963e521 1919 case NVME_IOCTL_IO_CMD:
a4aea562 1920 return nvme_user_cmd(ns->dev, ns, (void __user *)arg);
a53295b6
MW		 1921 case NVME_IOCTL_SUBMIT_IO:
1922 return nvme_submit_io(ns, (void __user *)arg);
5d0f6131
VV		 1923 case SG_GET_VERSION_NUM:
1924 return nvme_sg_get_version_num((void __user *)arg);
1925 case SG_IO:
1926 return nvme_sg_io(ns, (void __user *)arg);
b60503ba
MW		 1927 default:
1928 return -ENOTTY;
1929 }
1930}
1931
320a3827
KB		 1932#ifdef CONFIG_COMPAT
1933static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1934 unsigned int cmd, unsigned long arg)
1935{
320a3827
KB		 1936 switch (cmd) {
1937 case SG_IO:
e179729a 1938 return -ENOIOCTLCMD;
320a3827
KB		 1939 }
1940 return nvme_ioctl(bdev, mode, cmd, arg);
1941}
1942#else
1943#define nvme_compat_ioctl NULL
1944#endif
1945
5105aa55 1946static void nvme_free_dev(struct kref *kref);
188c3568
KB		 1947static void nvme_free_ns(struct kref *kref)
1948{
1949 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
1950
1951 spin_lock(&dev_list_lock);
1952 ns->disk->private_data = NULL;
1953 spin_unlock(&dev_list_lock);
1954
5105aa55 1955 kref_put(&ns->dev->kref, nvme_free_dev);
188c3568
KB		 1956 put_disk(ns->disk);
1957 kfree(ns);
1958}
1959
9ac27090
KB		 1960static int nvme_open(struct block_device *bdev, fmode_t mode)
1961{
9e60352c
KB		 1962 int ret = 0;
1963 struct nvme_ns *ns;
9ac27090 1964
9e60352c
KB		 1965 spin_lock(&dev_list_lock);
1966 ns = bdev->bd_disk->private_data;
1967 if (!ns)
1968 ret = -ENXIO;
188c3568 1969 else if (!kref_get_unless_zero(&ns->kref))
9e60352c
KB		 1970 ret = -ENXIO;
1971 spin_unlock(&dev_list_lock);
1972
1973 return ret;
9ac27090
KB		 1974}
1975
9ac27090
KB		 1976static void nvme_release(struct gendisk *disk, fmode_t mode)
1977{
1978 struct nvme_ns *ns = disk->private_data;
188c3568 1979 kref_put(&ns->kref, nvme_free_ns);
9ac27090
KB		 1980}
1981
4cc09e2d
KB		 1982static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo)
1983{
1984 /* some standard values */
1985 geo->heads = 1 << 6;
1986 geo->sectors = 1 << 5;
1987 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1988 return 0;
1989}
1990
e1e5e564
KB		 1991static void nvme_config_discard(struct nvme_ns *ns)
1992{
1993 u32 logical_block_size = queue_logical_block_size(ns->queue);
1994 ns->queue->limits.discard_zeroes_data = 0;
1995 ns->queue->limits.discard_alignment = logical_block_size;
1996 ns->queue->limits.discard_granularity = logical_block_size;
2bb4cd5c 1997 blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
e1e5e564
KB		 1998 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
1999}
2000
1b9dbf7f
KB		 2001static int nvme_revalidate_disk(struct gendisk *disk)
2002{
2003 struct nvme_ns *ns = disk->private_data;
2004 struct nvme_dev *dev = ns->dev;
2005 struct nvme_id_ns *id;
a67a9513
KB		 2006 u8 lbaf, pi_type;
2007 u16 old_ms;
e1e5e564 2008 unsigned short bs;
1b9dbf7f 2009
d29ec824 2010 if (nvme_identify_ns(dev, ns->ns_id, &id)) {
a5768aa8
KB		 2011 dev_warn(dev->dev, "%s: Identify failure nvme%dn%d\n", __func__,
2012 dev->instance, ns->ns_id);
2013 return -ENODEV;
1b9dbf7f 2014 }
a5768aa8
KB		 2015 if (id->ncap == 0) {
2016 kfree(id);
2017 return -ENODEV;
e1e5e564 2018 }
1b9dbf7f 2019
e1e5e564
KB		 2020 old_ms = ns->ms;
2021 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1b9dbf7f 2022 ns->lba_shift = id->lbaf[lbaf].ds;
e1e5e564 2023 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
a67a9513 2024 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
e1e5e564
KB		 2025
2026 /*
2027 * If identify namespace failed, use default 512 byte block size so
2028 * block layer can use before failing read/write for 0 capacity.
2029 */
2030 if (ns->lba_shift == 0)
2031 ns->lba_shift = 9;
2032 bs = 1 << ns->lba_shift;
2033
2034 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
2035 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
2036 id->dps & NVME_NS_DPS_PI_MASK : 0;
2037
52b68d7e
KB		 2038 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
2039 ns->ms != old_ms ||
e1e5e564 2040 bs != queue_logical_block_size(disk->queue) ||
a67a9513 2041 (ns->ms && ns->ext)))
e1e5e564
KB		 2042 blk_integrity_unregister(disk);
2043
2044 ns->pi_type = pi_type;
2045 blk_queue_logical_block_size(ns->queue, bs);
2046
25520d55 2047 if (ns->ms && !ns->ext)
e1e5e564
KB		 2048 nvme_init_integrity(ns);
2049
e19b127f 2050 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
e1e5e564
KB		 2051 set_capacity(disk, 0);
2052 else
2053 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
2054
2055 if (dev->oncs & NVME_CTRL_ONCS_DSM)
2056 nvme_config_discard(ns);
1b9dbf7f 2057
d29ec824 2058 kfree(id);
1b9dbf7f
KB		 2059 return 0;
2060}
2061
b60503ba
MW		 2062static const struct block_device_operations nvme_fops = {
2063 .owner = THIS_MODULE,
2064 .ioctl = nvme_ioctl,
320a3827 2065 .compat_ioctl = nvme_compat_ioctl,
9ac27090
KB		 2066 .open = nvme_open,
2067 .release = nvme_release,
4cc09e2d 2068 .getgeo = nvme_getgeo,
1b9dbf7f 2069 .revalidate_disk= nvme_revalidate_disk,
b60503ba
MW		 2070};
2071
1fa6aead
MW		 2072static int nvme_kthread(void *data)
2073{
d4b4ff8e 2074 struct nvme_dev *dev, *next;
1fa6aead
MW		 2075
2076 while (!kthread_should_stop()) {
564a232c 2077 set_current_state(TASK_INTERRUPTIBLE);
1fa6aead 2078 spin_lock(&dev_list_lock);
d4b4ff8e 2079 list_for_each_entry_safe(dev, next, &dev_list, node) {
1fa6aead 2080 int i;
dfbac8c7
KB		 2081 u32 csts = readl(&dev->bar->csts);
2082
2083 if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
2084 csts & NVME_CSTS_CFS) {
90667892
CH		 2085 if (!__nvme_reset(dev)) {
2086 dev_warn(dev->dev,
2087 "Failed status: %x, reset controller\n",
2088 readl(&dev->bar->csts));
2089 }
d4b4ff8e
KB		 2090 continue;
2091 }
1fa6aead 2092 for (i = 0; i < dev->queue_count; i++) {
a4aea562 2093 struct nvme_queue *nvmeq = dev->queues[i];
740216fc
MW		 2094 if (!nvmeq)
2095 continue;
1fa6aead 2096 spin_lock_irq(&nvmeq->q_lock);
bc57a0f7 2097 nvme_process_cq(nvmeq);
6fccf938
KB		 2098
2099 while ((i == 0) && (dev->event_limit > 0)) {
a4aea562 2100 if (nvme_submit_async_admin_req(dev))
6fccf938
KB		 2101 break;
2102 dev->event_limit--;
2103 }
1fa6aead
MW		 2104 spin_unlock_irq(&nvmeq->q_lock);
2105 }
2106 }
2107 spin_unlock(&dev_list_lock);
acb7aa0d 2108 schedule_timeout(round_jiffies_relative(HZ));
1fa6aead
MW		 2109 }
2110 return 0;
2111}
2112
e1e5e564 2113static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
b60503ba
MW		 2114{
2115 struct nvme_ns *ns;
2116 struct gendisk *disk;
e75ec752 2117 int node = dev_to_node(dev->dev);
b60503ba 2118
a4aea562 2119 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
b60503ba 2120 if (!ns)
e1e5e564
KB		 2121 return;
2122
a4aea562 2123 ns->queue = blk_mq_init_queue(&dev->tagset);
9f173b33 2124 if (IS_ERR(ns->queue))
b60503ba 2125 goto out_free_ns;
4eeb9215
MW		 2126 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
2127 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
b60503ba
MW		 2128 ns->dev = dev;
2129 ns->queue->queuedata = ns;
2130
a4aea562 2131 disk = alloc_disk_node(0, node);
b60503ba
MW		 2132 if (!disk)
2133 goto out_free_queue;
a4aea562 2134
188c3568 2135 kref_init(&ns->kref);
5aff9382 2136 ns->ns_id = nsid;
b60503ba 2137 ns->disk = disk;
e1e5e564
KB		 2138 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2139 list_add_tail(&ns->list, &dev->namespaces);
2140
e9ef4636 2141 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
e824410f 2142 if (dev->max_hw_sectors) {
8fc23e03 2143 blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors);
e824410f
KB		 2144 blk_queue_max_segments(ns->queue,
2145 ((dev->max_hw_sectors << 9) / dev->page_size) + 1);
2146 }
a4aea562
MB		 2147 if (dev->stripe_size)
2148 blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9);
a7d2ce28
KB		 2149 if (dev->vwc & NVME_CTRL_VWC_PRESENT)
2150 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
03100aad 2151 blk_queue_virt_boundary(ns->queue, dev->page_size - 1);
b60503ba
MW		 2152
2153 disk->major = nvme_major;
469071a3 2154 disk->first_minor = 0;
b60503ba
MW		 2155 disk->fops = &nvme_fops;
2156 disk->private_data = ns;
2157 disk->queue = ns->queue;
b3fffdef 2158 disk->driverfs_dev = dev->device;
469071a3 2159 disk->flags = GENHD_FL_EXT_DEVT;
5aff9382 2160 sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid);
b60503ba 2161
e1e5e564
KB		 2162 /*
2163 * Initialize capacity to 0 until we establish the namespace format and
2164 * setup integrity extentions if necessary. The revalidate_disk after
2165 * add_disk allows the driver to register with integrity if the format
2166 * requires it.
2167 */
2168 set_capacity(disk, 0);
a5768aa8
KB		 2169 if (nvme_revalidate_disk(ns->disk))
2170 goto out_free_disk;
2171
5105aa55 2172 kref_get(&dev->kref);
e1e5e564 2173 add_disk(ns->disk);
7bee6074
KB		 2174 if (ns->ms) {
2175 struct block_device *bd = bdget_disk(ns->disk, 0);
2176 if (!bd)
2177 return;
2178 if (blkdev_get(bd, FMODE_READ, NULL)) {
2179 bdput(bd);
2180 return;
2181 }
2182 blkdev_reread_part(bd);
2183 blkdev_put(bd, FMODE_READ);
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);
2410 if (ns->disk->flags & GENHD_FL_UP) {
2411 if (blk_get_integrity(ns->disk))
2412 blk_integrity_unregister(ns->disk);
2413 del_gendisk(ns->disk);
2414 }
2415 if (kill || !blk_queue_dying(ns->queue)) {
2416 blk_mq_abort_requeue_list(ns->queue);
2417 blk_cleanup_queue(ns->queue);
5105aa55
KB		 2418 }
2419 list_del_init(&ns->list);
2420 kref_put(&ns->kref, nvme_free_ns);
a5768aa8
KB		 2421}
2422
2423static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn)
2424{
2425 struct nvme_ns *ns, *next;
2426 unsigned i;
2427
2428 for (i = 1; i <= nn; i++) {
2429 ns = nvme_find_ns(dev, i);
2430 if (ns) {
5105aa55 2431 if (revalidate_disk(ns->disk))
a5768aa8 2432 nvme_ns_remove(ns);
a5768aa8
KB		 2433 } else
2434 nvme_alloc_ns(dev, i);
2435 }
2436 list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
5105aa55 2437 if (ns->ns_id > nn)
a5768aa8 2438 nvme_ns_remove(ns);
a5768aa8
KB		 2439 }
2440 list_sort(NULL, &dev->namespaces, ns_cmp);
2441}
2442
bda4e0fb
KB		 2443static void nvme_set_irq_hints(struct nvme_dev *dev)
2444{
2445 struct nvme_queue *nvmeq;
2446 int i;
2447
2448 for (i = 0; i < dev->online_queues; i++) {
2449 nvmeq = dev->queues[i];
2450
2451 if (!nvmeq->tags || !(*nvmeq->tags))
2452 continue;
2453
2454 irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector,
2455 blk_mq_tags_cpumask(*nvmeq->tags));
2456 }
2457}
2458
a5768aa8
KB		 2459static void nvme_dev_scan(struct work_struct *work)
2460{
2461 struct nvme_dev *dev = container_of(work, struct nvme_dev, scan_work);
2462 struct nvme_id_ctrl *ctrl;
2463
2464 if (!dev->tagset.tags)
2465 return;
2466 if (nvme_identify_ctrl(dev, &ctrl))
2467 return;
2468 nvme_scan_namespaces(dev, le32_to_cpup(&ctrl->nn));
2469 kfree(ctrl);
bda4e0fb 2470 nvme_set_irq_hints(dev);
a5768aa8
KB		 2471}
2472
422ef0c7
MW		 2473/*
2474 * Return: error value if an error occurred setting up the queues or calling
2475 * Identify Device. 0 if these succeeded, even if adding some of the
2476 * namespaces failed. At the moment, these failures are silent. TBD which
2477 * failures should be reported.
2478 */
8d85fce7 2479static int nvme_dev_add(struct nvme_dev *dev)
b60503ba 2480{
e75ec752 2481 struct pci_dev *pdev = to_pci_dev(dev->dev);
c3bfe717 2482 int res;
51814232 2483 struct nvme_id_ctrl *ctrl;
159b67d7 2484 int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12;
b60503ba 2485
d29ec824 2486 res = nvme_identify_ctrl(dev, &ctrl);
b60503ba 2487 if (res) {
e75ec752 2488 dev_err(dev->dev, "Identify Controller failed (%d)\n", res);
e1e5e564 2489 return -EIO;
b60503ba
MW		 2490 }
2491
0e5e4f0e 2492 dev->oncs = le16_to_cpup(&ctrl->oncs);
c30341dc 2493 dev->abort_limit = ctrl->acl + 1;
a7d2ce28 2494 dev->vwc = ctrl->vwc;
51814232
MW		 2495 memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn));
2496 memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn));
2497 memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
159b67d7 2498 if (ctrl->mdts)
8fc23e03 2499 dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9);
68608c26 2500 if ((pdev->vendor == PCI_VENDOR_ID_INTEL) &&
a4aea562
MB		 2501 (pdev->device == 0x0953) && ctrl->vs[3]) {
2502 unsigned int max_hw_sectors;
2503
159b67d7 2504 dev->stripe_size = 1 << (ctrl->vs[3] + shift);
a4aea562
MB		 2505 max_hw_sectors = dev->stripe_size >> (shift - 9);
2506 if (dev->max_hw_sectors) {
2507 dev->max_hw_sectors = min(max_hw_sectors,
2508 dev->max_hw_sectors);
2509 } else
2510 dev->max_hw_sectors = max_hw_sectors;
2511 }
d29ec824 2512 kfree(ctrl);
a4aea562 2513
ffe7704d
KB		 2514 if (!dev->tagset.tags) {
2515 dev->tagset.ops = &nvme_mq_ops;
2516 dev->tagset.nr_hw_queues = dev->online_queues - 1;
2517 dev->tagset.timeout = NVME_IO_TIMEOUT;
2518 dev->tagset.numa_node = dev_to_node(dev->dev);
2519 dev->tagset.queue_depth =
a4aea562 2520 min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
ffe7704d
KB		 2521 dev->tagset.cmd_size = nvme_cmd_size(dev);
2522 dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
2523 dev->tagset.driver_data = dev;
b60503ba 2524
ffe7704d
KB		 2525 if (blk_mq_alloc_tag_set(&dev->tagset))
2526 return 0;
2527 }
a5768aa8 2528 schedule_work(&dev->scan_work);
e1e5e564 2529 return 0;
b60503ba
MW		 2530}
2531
0877cb0d
KB		 2532static int nvme_dev_map(struct nvme_dev *dev)
2533{
42f61420 2534 u64 cap;
0877cb0d 2535 int bars, result = -ENOMEM;
e75ec752 2536 struct pci_dev *pdev = to_pci_dev(dev->dev);
0877cb0d
KB		 2537
2538 if (pci_enable_device_mem(pdev))
2539 return result;
2540
2541 dev->entry[0].vector = pdev->irq;
2542 pci_set_master(pdev);
2543 bars = pci_select_bars(pdev, IORESOURCE_MEM);
be7837e8
JA		 2544 if (!bars)
2545 goto disable_pci;
2546
0877cb0d
KB		 2547 if (pci_request_selected_regions(pdev, bars, "nvme"))
2548 goto disable_pci;
2549
e75ec752
CH		 2550 if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
2551 dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
052d0efa 2552 goto disable;
0877cb0d 2553
0877cb0d
KB		 2554 dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
2555 if (!dev->bar)
2556 goto disable;
e32efbfc 2557
0e53d180
KB		 2558 if (readl(&dev->bar->csts) == -1) {
2559 result = -ENODEV;
2560 goto unmap;
2561 }
e32efbfc
JA		 2562
2563 /*
2564 * Some devices don't advertse INTx interrupts, pre-enable a single
2565 * MSIX vec for setup. We'll adjust this later.
2566 */
2567 if (!pdev->irq) {
2568 result = pci_enable_msix(pdev, dev->entry, 1);
2569 if (result < 0)
2570 goto unmap;
2571 }
2572
42f61420
KB		 2573 cap = readq(&dev->bar->cap);
2574 dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
2575 dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
0877cb0d 2576 dev->dbs = ((void __iomem *)dev->bar) + 4096;
8ffaadf7
JD		 2577 if (readl(&dev->bar->vs) >= NVME_VS(1, 2))
2578 dev->cmb = nvme_map_cmb(dev);
0877cb0d
KB		 2579
2580 return 0;
2581
0e53d180
KB		 2582 unmap:
2583 iounmap(dev->bar);
2584 dev->bar = NULL;
0877cb0d
KB		 2585 disable:
2586 pci_release_regions(pdev);
2587 disable_pci:
2588 pci_disable_device(pdev);
2589 return result;
2590}
2591
2592static void nvme_dev_unmap(struct nvme_dev *dev)
2593{
e75ec752
CH		 2594 struct pci_dev *pdev = to_pci_dev(dev->dev);
2595
2596 if (pdev->msi_enabled)
2597 pci_disable_msi(pdev);
2598 else if (pdev->msix_enabled)
2599 pci_disable_msix(pdev);
0877cb0d
KB		 2600
2601 if (dev->bar) {
2602 iounmap(dev->bar);
2603 dev->bar = NULL;
e75ec752 2604 pci_release_regions(pdev);
0877cb0d
KB		 2605 }
2606
e75ec752
CH		 2607 if (pci_is_enabled(pdev))
2608 pci_disable_device(pdev);
0877cb0d
KB		 2609}
2610
4d115420
KB		 2611struct nvme_delq_ctx {
2612 struct task_struct *waiter;
2613 struct kthread_worker *worker;
2614 atomic_t refcount;
2615};
2616
2617static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev)
2618{
2619 dq->waiter = current;
2620 mb();
2621
2622 for (;;) {
2623 set_current_state(TASK_KILLABLE);
2624 if (!atomic_read(&dq->refcount))
2625 break;
2626 if (!schedule_timeout(ADMIN_TIMEOUT) ||
2627 fatal_signal_pending(current)) {
0fb59cbc
KB		 2628 /*
2629 * Disable the controller first since we can't trust it
2630 * at this point, but leave the admin queue enabled
2631 * until all queue deletion requests are flushed.
2632 * FIXME: This may take a while if there are more h/w
2633 * queues than admin tags.
2634 */
4d115420 2635 set_current_state(TASK_RUNNING);
4d115420 2636 nvme_disable_ctrl(dev, readq(&dev->bar->cap));
0fb59cbc 2637 nvme_clear_queue(dev->queues[0]);
4d115420 2638 flush_kthread_worker(dq->worker);
0fb59cbc 2639 nvme_disable_queue(dev, 0);
4d115420
KB		 2640 return;
2641 }
2642 }
2643 set_current_state(TASK_RUNNING);
2644}
2645
2646static void nvme_put_dq(struct nvme_delq_ctx *dq)
2647{
2648 atomic_dec(&dq->refcount);
2649 if (dq->waiter)
2650 wake_up_process(dq->waiter);
2651}
2652
2653static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq)
2654{
2655 atomic_inc(&dq->refcount);
2656 return dq;
2657}
2658
2659static void nvme_del_queue_end(struct nvme_queue *nvmeq)
2660{
2661 struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx;
4d115420
KB		 2662 nvme_put_dq(dq);
2663}
2664
2665static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode,
2666 kthread_work_func_t fn)
2667{
2668 struct nvme_command c;
2669
2670 memset(&c, 0, sizeof(c));
2671 c.delete_queue.opcode = opcode;
2672 c.delete_queue.qid = cpu_to_le16(nvmeq->qid);
2673
2674 init_kthread_work(&nvmeq->cmdinfo.work, fn);
a4aea562
MB		 2675 return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo,
2676 ADMIN_TIMEOUT);
4d115420
KB		 2677}
2678
2679static void nvme_del_cq_work_handler(struct kthread_work *work)
2680{
2681 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2682 cmdinfo.work);
2683 nvme_del_queue_end(nvmeq);
2684}
2685
2686static int nvme_delete_cq(struct nvme_queue *nvmeq)
2687{
2688 return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq,
2689 nvme_del_cq_work_handler);
2690}
2691
2692static void nvme_del_sq_work_handler(struct kthread_work *work)
2693{
2694 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2695 cmdinfo.work);
2696 int status = nvmeq->cmdinfo.status;
2697
2698 if (!status)
2699 status = nvme_delete_cq(nvmeq);
2700 if (status)
2701 nvme_del_queue_end(nvmeq);
2702}
2703
2704static int nvme_delete_sq(struct nvme_queue *nvmeq)
2705{
2706 return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq,
2707 nvme_del_sq_work_handler);
2708}
2709
2710static void nvme_del_queue_start(struct kthread_work *work)
2711{
2712 struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
2713 cmdinfo.work);
4d115420
KB		 2714 if (nvme_delete_sq(nvmeq))
2715 nvme_del_queue_end(nvmeq);
2716}
2717
2718static void nvme_disable_io_queues(struct nvme_dev *dev)
2719{
2720 int i;
2721 DEFINE_KTHREAD_WORKER_ONSTACK(worker);
2722 struct nvme_delq_ctx dq;
2723 struct task_struct *kworker_task = kthread_run(kthread_worker_fn,
2724 &worker, "nvme%d", dev->instance);
2725
2726 if (IS_ERR(kworker_task)) {
e75ec752 2727 dev_err(dev->dev,
4d115420
KB		 2728 "Failed to create queue del task\n");
2729 for (i = dev->queue_count - 1; i > 0; i--)
2730 nvme_disable_queue(dev, i);
2731 return;
2732 }
2733
2734 dq.waiter = NULL;
2735 atomic_set(&dq.refcount, 0);
2736 dq.worker = &worker;
2737 for (i = dev->queue_count - 1; i > 0; i--) {
a4aea562 2738 struct nvme_queue *nvmeq = dev->queues[i];
4d115420
KB		 2739
2740 if (nvme_suspend_queue(nvmeq))
2741 continue;
2742 nvmeq->cmdinfo.ctx = nvme_get_dq(&dq);
2743 nvmeq->cmdinfo.worker = dq.worker;
2744 init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start);
2745 queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work);
2746 }
2747 nvme_wait_dq(&dq, dev);
2748 kthread_stop(kworker_task);
2749}
2750
b9afca3e
DM		 2751/*
2752* Remove the node from the device list and check
2753* for whether or not we need to stop the nvme_thread.
2754*/
2755static void nvme_dev_list_remove(struct nvme_dev *dev)
2756{
2757 struct task_struct *tmp = NULL;
2758
2759 spin_lock(&dev_list_lock);
2760 list_del_init(&dev->node);
2761 if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) {
2762 tmp = nvme_thread;
2763 nvme_thread = NULL;
2764 }
2765 spin_unlock(&dev_list_lock);
2766
2767 if (tmp)
2768 kthread_stop(tmp);
2769}
2770
c9d3bf88
KB		 2771static void nvme_freeze_queues(struct nvme_dev *dev)
2772{
2773 struct nvme_ns *ns;
2774
2775 list_for_each_entry(ns, &dev->namespaces, list) {
2776 blk_mq_freeze_queue_start(ns->queue);
2777
cddcd72b 2778 spin_lock_irq(ns->queue->queue_lock);
c9d3bf88 2779 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
cddcd72b 2780 spin_unlock_irq(ns->queue->queue_lock);
c9d3bf88
KB		 2781
2782 blk_mq_cancel_requeue_work(ns->queue);
2783 blk_mq_stop_hw_queues(ns->queue);
2784 }
2785}
2786
2787static void nvme_unfreeze_queues(struct nvme_dev *dev)
2788{
2789 struct nvme_ns *ns;
2790
2791 list_for_each_entry(ns, &dev->namespaces, list) {
2792 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2793 blk_mq_unfreeze_queue(ns->queue);
2794 blk_mq_start_stopped_hw_queues(ns->queue, true);
2795 blk_mq_kick_requeue_list(ns->queue);
2796 }
2797}
2798
f0b50732 2799static void nvme_dev_shutdown(struct nvme_dev *dev)
b60503ba 2800{
22404274 2801 int i;
7c1b2450 2802 u32 csts = -1;
22404274 2803
b9afca3e 2804 nvme_dev_list_remove(dev);
1fa6aead 2805
c9d3bf88
KB		 2806 if (dev->bar) {
2807 nvme_freeze_queues(dev);
7c1b2450 2808 csts = readl(&dev->bar->csts);
c9d3bf88 2809 }
7c1b2450 2810 if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) {
4d115420 2811 for (i = dev->queue_count - 1; i >= 0; i--) {
a4aea562 2812 struct nvme_queue *nvmeq = dev->queues[i];
4d115420 2813 nvme_suspend_queue(nvmeq);
4d115420
KB		 2814 }
2815 } else {
2816 nvme_disable_io_queues(dev);
1894d8f1 2817 nvme_shutdown_ctrl(dev);
4d115420
KB		 2818 nvme_disable_queue(dev, 0);
2819 }
f0b50732 2820 nvme_dev_unmap(dev);
07836e65
KB		 2821
2822 for (i = dev->queue_count - 1; i >= 0; i--)
2823 nvme_clear_queue(dev->queues[i]);
f0b50732
KB		 2824}
2825
2826static void nvme_dev_remove(struct nvme_dev *dev)
2827{
5105aa55 2828 struct nvme_ns *ns, *next;
f0b50732 2829
5105aa55 2830 list_for_each_entry_safe(ns, next, &dev->namespaces, list)
a5768aa8 2831 nvme_ns_remove(ns);
b60503ba
MW		 2832}
2833
091b6092
MW		 2834static int nvme_setup_prp_pools(struct nvme_dev *dev)
2835{
e75ec752 2836 dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
091b6092
MW		 2837 PAGE_SIZE, PAGE_SIZE, 0);
2838 if (!dev->prp_page_pool)
2839 return -ENOMEM;
2840
99802a7a 2841 /* Optimisation for I/Os between 4k and 128k */
e75ec752 2842 dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
99802a7a
MW		 2843 256, 256, 0);
2844 if (!dev->prp_small_pool) {
2845 dma_pool_destroy(dev->prp_page_pool);
2846 return -ENOMEM;
2847 }
091b6092
MW		 2848 return 0;
2849}
2850
2851static void nvme_release_prp_pools(struct nvme_dev *dev)
2852{
2853 dma_pool_destroy(dev->prp_page_pool);
99802a7a 2854 dma_pool_destroy(dev->prp_small_pool);
091b6092
MW		 2855}
2856
cd58ad7d
QSA		 2857static DEFINE_IDA(nvme_instance_ida);
2858
2859static int nvme_set_instance(struct nvme_dev *dev)
b60503ba 2860{
cd58ad7d
QSA		 2861 int instance, error;
2862
2863 do {
2864 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2865 return -ENODEV;
2866
2867 spin_lock(&dev_list_lock);
2868 error = ida_get_new(&nvme_instance_ida, &instance);
2869 spin_unlock(&dev_list_lock);
2870 } while (error == -EAGAIN);
2871
2872 if (error)
2873 return -ENODEV;
2874
2875 dev->instance = instance;
2876 return 0;
b60503ba
MW		 2877}
2878
2879static void nvme_release_instance(struct nvme_dev *dev)
2880{
cd58ad7d
QSA		 2881 spin_lock(&dev_list_lock);
2882 ida_remove(&nvme_instance_ida, dev->instance);
2883 spin_unlock(&dev_list_lock);
b60503ba
MW		 2884}
2885
5e82e952
KB		 2886static void nvme_free_dev(struct kref *kref)
2887{
2888 struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
9ac27090 2889
e75ec752 2890 put_device(dev->dev);
b3fffdef 2891 put_device(dev->device);
285dffc9 2892 nvme_release_instance(dev);
4af0e21c
KB		 2893 if (dev->tagset.tags)
2894 blk_mq_free_tag_set(&dev->tagset);
2895 if (dev->admin_q)
2896 blk_put_queue(dev->admin_q);
5e82e952
KB		 2897 kfree(dev->queues);
2898 kfree(dev->entry);
2899 kfree(dev);
2900}
2901
2902static int nvme_dev_open(struct inode *inode, struct file *f)
2903{
b3fffdef
KB		 2904 struct nvme_dev *dev;
2905 int instance = iminor(inode);
2906 int ret = -ENODEV;
2907
2908 spin_lock(&dev_list_lock);
2909 list_for_each_entry(dev, &dev_list, node) {
2910 if (dev->instance == instance) {
2e1d8448
KB		 2911 if (!dev->admin_q) {
2912 ret = -EWOULDBLOCK;
2913 break;
2914 }
b3fffdef
KB		 2915 if (!kref_get_unless_zero(&dev->kref))
2916 break;
2917 f->private_data = dev;
2918 ret = 0;
2919 break;
2920 }
2921 }
2922 spin_unlock(&dev_list_lock);
2923
2924 return ret;
5e82e952
KB		 2925}
2926
2927static int nvme_dev_release(struct inode *inode, struct file *f)
2928{
2929 struct nvme_dev *dev = f->private_data;
2930 kref_put(&dev->kref, nvme_free_dev);
2931 return 0;
2932}
2933
2934static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
2935{
2936 struct nvme_dev *dev = f->private_data;
a4aea562
MB		 2937 struct nvme_ns *ns;
2938
5e82e952
KB		 2939 switch (cmd) {
2940 case NVME_IOCTL_ADMIN_CMD:
a4aea562 2941 return nvme_user_cmd(dev, NULL, (void __user *)arg);
7963e521 2942 case NVME_IOCTL_IO_CMD:
a4aea562
MB		 2943 if (list_empty(&dev->namespaces))
2944 return -ENOTTY;
2945 ns = list_first_entry(&dev->namespaces, struct nvme_ns, list);
2946 return nvme_user_cmd(dev, ns, (void __user *)arg);
4cc06521
KB		 2947 case NVME_IOCTL_RESET:
2948 dev_warn(dev->dev, "resetting controller\n");
2949 return nvme_reset(dev);
81f03fed
JD		 2950 case NVME_IOCTL_SUBSYS_RESET:
2951 return nvme_subsys_reset(dev);
5e82e952
KB		 2952 default:
2953 return -ENOTTY;
2954 }
2955}
2956
2957static const struct file_operations nvme_dev_fops = {
2958 .owner = THIS_MODULE,
2959 .open = nvme_dev_open,
2960 .release = nvme_dev_release,
2961 .unlocked_ioctl = nvme_dev_ioctl,
2962 .compat_ioctl = nvme_dev_ioctl,
2963};
2964
3cf519b5 2965static void nvme_probe_work(struct work_struct *work)
f0b50732 2966{
3cf519b5 2967 struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work);
b9afca3e 2968 bool start_thread = false;
3cf519b5 2969 int result;
f0b50732
KB		 2970
2971 result = nvme_dev_map(dev);
2972 if (result)
3cf519b5 2973 goto out;
f0b50732
KB		 2974
2975 result = nvme_configure_admin_queue(dev);
2976 if (result)
2977 goto unmap;
2978
2979 spin_lock(&dev_list_lock);
b9afca3e
DM		 2980 if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) {
2981 start_thread = true;
2982 nvme_thread = NULL;
2983 }
f0b50732
KB		 2984 list_add(&dev->node, &dev_list);
2985 spin_unlock(&dev_list_lock);
2986
b9afca3e
DM		 2987 if (start_thread) {
2988 nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
387caa5a 2989 wake_up_all(&nvme_kthread_wait);
b9afca3e
DM		 2990 } else
2991 wait_event_killable(nvme_kthread_wait, nvme_thread);
2992
2993 if (IS_ERR_OR_NULL(nvme_thread)) {
2994 result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR;
2995 goto disable;
2996 }
a4aea562
MB		 2997
2998 nvme_init_queue(dev->queues[0], 0);
0fb59cbc
KB		 2999 result = nvme_alloc_admin_tags(dev);
3000 if (result)
3001 goto disable;
b9afca3e 3002
f0b50732 3003 result = nvme_setup_io_queues(dev);
badc34d4 3004 if (result)
0fb59cbc 3005 goto free_tags;
f0b50732 3006
1efccc9d 3007 dev->event_limit = 1;
3cf519b5 3008
2659e57b
CH		 3009 /*
3010 * Keep the controller around but remove all namespaces if we don't have
3011 * any working I/O queue.
3012 */
3cf519b5
CH		 3013 if (dev->online_queues < 2) {
3014 dev_warn(dev->dev, "IO queues not created\n");
3cf519b5
CH		 3015 nvme_dev_remove(dev);
3016 } else {
3017 nvme_unfreeze_queues(dev);
3018 nvme_dev_add(dev);
3019 }
3020
3021 return;
f0b50732 3022
0fb59cbc
KB		 3023 free_tags:
3024 nvme_dev_remove_admin(dev);
4af0e21c
KB		 3025 blk_put_queue(dev->admin_q);
3026 dev->admin_q = NULL;
3027 dev->queues[0]->tags = NULL;
f0b50732 3028 disable:
a1a5ef99 3029 nvme_disable_queue(dev, 0);
b9afca3e 3030 nvme_dev_list_remove(dev);
f0b50732
KB		 3031 unmap:
3032 nvme_dev_unmap(dev);
3cf519b5
CH		 3033 out:
3034 if (!work_busy(&dev->reset_work))
3035 nvme_dead_ctrl(dev);
f0b50732
KB		 3036}
3037
9a6b9458
KB		 3038static int nvme_remove_dead_ctrl(void *arg)
3039{
3040 struct nvme_dev *dev = (struct nvme_dev *)arg;
e75ec752 3041 struct pci_dev *pdev = to_pci_dev(dev->dev);
9a6b9458
KB		 3042
3043 if (pci_get_drvdata(pdev))
c81f4975 3044 pci_stop_and_remove_bus_device_locked(pdev);
9a6b9458
KB		 3045 kref_put(&dev->kref, nvme_free_dev);
3046 return 0;
3047}
3048
de3eff2b
KB		 3049static void nvme_dead_ctrl(struct nvme_dev *dev)
3050{
3051 dev_warn(dev->dev, "Device failed to resume\n");
3052 kref_get(&dev->kref);
3053 if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d",
3054 dev->instance))) {
3055 dev_err(dev->dev,
3056 "Failed to start controller remove task\n");
3057 kref_put(&dev->kref, nvme_free_dev);
3058 }
3059}
3060
77b50d9e 3061static void nvme_reset_work(struct work_struct *ws)
9a6b9458 3062{
77b50d9e 3063 struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
ffe7704d
KB		 3064 bool in_probe = work_busy(&dev->probe_work);
3065
9a6b9458 3066 nvme_dev_shutdown(dev);
ffe7704d
KB		 3067
3068 /* Synchronize with device probe so that work will see failure status
3069 * and exit gracefully without trying to schedule another reset */
3070 flush_work(&dev->probe_work);
3071
3072 /* Fail this device if reset occured during probe to avoid
3073 * infinite initialization loops. */
3074 if (in_probe) {
de3eff2b 3075 nvme_dead_ctrl(dev);
ffe7704d 3076 return;
9a6b9458 3077 }
ffe7704d
KB		 3078 /* Schedule device resume asynchronously so the reset work is available
3079 * to cleanup errors that may occur during reinitialization */
3080 schedule_work(&dev->probe_work);
9a6b9458
KB		 3081}
3082
90667892
CH		 3083static int __nvme_reset(struct nvme_dev *dev)
3084{
3085 if (work_pending(&dev->reset_work))
3086 return -EBUSY;
3087 list_del_init(&dev->node);
3088 queue_work(nvme_workq, &dev->reset_work);
3089 return 0;
3090}
3091
4cc06521
KB		 3092static int nvme_reset(struct nvme_dev *dev)
3093{
90667892 3094 int ret;
4cc06521
KB		 3095
3096 if (!dev->admin_q || blk_queue_dying(dev->admin_q))
3097 return -ENODEV;
3098
3099 spin_lock(&dev_list_lock);
90667892 3100 ret = __nvme_reset(dev);
4cc06521
KB		 3101 spin_unlock(&dev_list_lock);
3102
3103 if (!ret) {
3104 flush_work(&dev->reset_work);
ffe7704d 3105 flush_work(&dev->probe_work);
4cc06521
KB		 3106 return 0;
3107 }
3108
3109 return ret;
3110}
3111
3112static ssize_t nvme_sysfs_reset(struct device *dev,
3113 struct device_attribute *attr, const char *buf,
3114 size_t count)
3115{
3116 struct nvme_dev *ndev = dev_get_drvdata(dev);
3117 int ret;
3118
3119 ret = nvme_reset(ndev);
3120 if (ret < 0)
3121 return ret;
3122
3123 return count;
3124}
3125static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3126
8d85fce7 3127static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
b60503ba 3128{
a4aea562 3129 int node, result = -ENOMEM;
b60503ba
MW		 3130 struct nvme_dev *dev;
3131
a4aea562
MB		 3132 node = dev_to_node(&pdev->dev);
3133 if (node == NUMA_NO_NODE)
3134 set_dev_node(&pdev->dev, 0);
3135
3136 dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
b60503ba
MW		 3137 if (!dev)
3138 return -ENOMEM;
a4aea562
MB		 3139 dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry),
3140 GFP_KERNEL, node);
b60503ba
MW		 3141 if (!dev->entry)
3142 goto free;
a4aea562
MB		 3143 dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
3144 GFP_KERNEL, node);
b60503ba
MW		 3145 if (!dev->queues)
3146 goto free;
3147
3148 INIT_LIST_HEAD(&dev->namespaces);
77b50d9e 3149 INIT_WORK(&dev->reset_work, nvme_reset_work);
e75ec752 3150 dev->dev = get_device(&pdev->dev);
9a6b9458 3151 pci_set_drvdata(pdev, dev);
cd58ad7d
QSA		 3152 result = nvme_set_instance(dev);
3153 if (result)
a96d4f5c 3154 goto put_pci;
b60503ba 3155
091b6092
MW		 3156 result = nvme_setup_prp_pools(dev);
3157 if (result)
0877cb0d 3158 goto release;
091b6092 3159
fb35e914 3160 kref_init(&dev->kref);
b3fffdef
KB		 3161 dev->device = device_create(nvme_class, &pdev->dev,
3162 MKDEV(nvme_char_major, dev->instance),
3163 dev, "nvme%d", dev->instance);
3164 if (IS_ERR(dev->device)) {
3165 result = PTR_ERR(dev->device);
2e1d8448 3166 goto release_pools;
b3fffdef
KB		 3167 }
3168 get_device(dev->device);
4cc06521
KB		 3169 dev_set_drvdata(dev->device, dev);
3170
3171 result = device_create_file(dev->device, &dev_attr_reset_controller);
3172 if (result)
3173 goto put_dev;
740216fc 3174
e6e96d73 3175 INIT_LIST_HEAD(&dev->node);
a5768aa8 3176 INIT_WORK(&dev->scan_work, nvme_dev_scan);
3cf519b5 3177 INIT_WORK(&dev->probe_work, nvme_probe_work);
2e1d8448 3178 schedule_work(&dev->probe_work);
b60503ba
MW		 3179 return 0;
3180
4cc06521
KB		 3181 put_dev:
3182 device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
3183 put_device(dev->device);
0877cb0d 3184 release_pools:
091b6092 3185 nvme_release_prp_pools(dev);
0877cb0d
KB		 3186 release:
3187 nvme_release_instance(dev);
a96d4f5c 3188 put_pci:
e75ec752 3189 put_device(dev->dev);
b60503ba
MW		 3190 free:
3191 kfree(dev->queues);
3192 kfree(dev->entry);
3193 kfree(dev);
3194 return result;
3195}
3196
f0d54a54
KB		 3197static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
3198{
a6739479 3199 struct nvme_dev *dev = pci_get_drvdata(pdev);
f0d54a54 3200
a6739479
KB		 3201 if (prepare)
3202 nvme_dev_shutdown(dev);
3203 else
0a7385ad 3204 schedule_work(&dev->probe_work);
f0d54a54
KB		 3205}
3206
09ece142
KB		 3207static void nvme_shutdown(struct pci_dev *pdev)
3208{
3209 struct nvme_dev *dev = pci_get_drvdata(pdev);
3210 nvme_dev_shutdown(dev);
3211}
3212
8d85fce7 3213static void nvme_remove(struct pci_dev *pdev)
b60503ba
MW		 3214{
3215 struct nvme_dev *dev = pci_get_drvdata(pdev);
9a6b9458
KB		 3216
3217 spin_lock(&dev_list_lock);
3218 list_del_init(&dev->node);
3219 spin_unlock(&dev_list_lock);
3220
3221 pci_set_drvdata(pdev, NULL);
2e1d8448 3222 flush_work(&dev->probe_work);
9a6b9458 3223 flush_work(&dev->reset_work);
a5768aa8 3224 flush_work(&dev->scan_work);
4cc06521 3225 device_remove_file(dev->device, &dev_attr_reset_controller);
c9d3bf88 3226 nvme_dev_remove(dev);
3399a3f7 3227 nvme_dev_shutdown(dev);
a4aea562 3228 nvme_dev_remove_admin(dev);
b3fffdef 3229 device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
a1a5ef99 3230 nvme_free_queues(dev, 0);
8ffaadf7 3231 nvme_release_cmb(dev);
9a6b9458 3232 nvme_release_prp_pools(dev);
5e82e952 3233 kref_put(&dev->kref, nvme_free_dev);
b60503ba
MW		 3234}
3235
3236/* These functions are yet to be implemented */
3237#define nvme_error_detected NULL
3238#define nvme_dump_registers NULL
3239#define nvme_link_reset NULL
3240#define nvme_slot_reset NULL
3241#define nvme_error_resume NULL
cd638946 3242
671a6018 3243#ifdef CONFIG_PM_SLEEP
cd638946
KB		 3244static int nvme_suspend(struct device *dev)
3245{
3246 struct pci_dev *pdev = to_pci_dev(dev);
3247 struct nvme_dev *ndev = pci_get_drvdata(pdev);
3248
3249 nvme_dev_shutdown(ndev);
3250 return 0;
3251}
3252
3253static int nvme_resume(struct device *dev)
3254{
3255 struct pci_dev *pdev = to_pci_dev(dev);
3256 struct nvme_dev *ndev = pci_get_drvdata(pdev);
cd638946 3257
0a7385ad 3258 schedule_work(&ndev->probe_work);
9a6b9458 3259 return 0;
cd638946 3260}
671a6018 3261#endif
cd638946
KB		 3262
3263static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
b60503ba 3264
1d352035 3265static const struct pci_error_handlers nvme_err_handler = {
b60503ba
MW		 3266 .error_detected = nvme_error_detected,
3267 .mmio_enabled = nvme_dump_registers,
3268 .link_reset = nvme_link_reset,
3269 .slot_reset = nvme_slot_reset,
3270 .resume = nvme_error_resume,
f0d54a54 3271 .reset_notify = nvme_reset_notify,
b60503ba
MW		 3272};
3273
3274/* Move to pci_ids.h later */
3275#define PCI_CLASS_STORAGE_EXPRESS 0x010802
3276
6eb0d698 3277static const struct pci_device_id nvme_id_table[] = {
b60503ba
MW		 3278 { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
3279 { 0, }
3280};
3281MODULE_DEVICE_TABLE(pci, nvme_id_table);
3282
3283static struct pci_driver nvme_driver = {
3284 .name = "nvme",
3285 .id_table = nvme_id_table,
3286 .probe = nvme_probe,
8d85fce7 3287 .remove = nvme_remove,
09ece142 3288 .shutdown = nvme_shutdown,
cd638946
KB		 3289 .driver = {
3290 .pm = &nvme_dev_pm_ops,
3291 },
b60503ba
MW		 3292 .err_handler = &nvme_err_handler,
3293};
3294
3295static int __init nvme_init(void)
3296{
0ac13140 3297 int result;
1fa6aead 3298
b9afca3e 3299 init_waitqueue_head(&nvme_kthread_wait);
b60503ba 3300
9a6b9458
KB		 3301 nvme_workq = create_singlethread_workqueue("nvme");
3302 if (!nvme_workq)
b9afca3e 3303 return -ENOMEM;
9a6b9458 3304
5c42ea16
KB		 3305 result = register_blkdev(nvme_major, "nvme");
3306 if (result < 0)
9a6b9458 3307 goto kill_workq;
5c42ea16 3308 else if (result > 0)
0ac13140 3309 nvme_major = result;
b60503ba 3310
b3fffdef
KB		 3311 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
3312 &nvme_dev_fops);
3313 if (result < 0)
3314 goto unregister_blkdev;
3315 else if (result > 0)
3316 nvme_char_major = result;
3317
3318 nvme_class = class_create(THIS_MODULE, "nvme");
c727040b
AK		 3319 if (IS_ERR(nvme_class)) {
3320 result = PTR_ERR(nvme_class);
b3fffdef 3321 goto unregister_chrdev;
c727040b 3322 }
b3fffdef 3323
f3db22fe
KB		 3324 result = pci_register_driver(&nvme_driver);
3325 if (result)
b3fffdef 3326 goto destroy_class;
1fa6aead 3327 return 0;
b60503ba 3328
b3fffdef
KB		 3329 destroy_class:
3330 class_destroy(nvme_class);
3331 unregister_chrdev:
3332 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1fa6aead 3333 unregister_blkdev:
b60503ba 3334 unregister_blkdev(nvme_major, "nvme");
9a6b9458
KB		 3335 kill_workq:
3336 destroy_workqueue(nvme_workq);
b60503ba
MW		 3337 return result;
3338}
3339
3340static void __exit nvme_exit(void)
3341{
3342 pci_unregister_driver(&nvme_driver);
3343 unregister_blkdev(nvme_major, "nvme");
9a6b9458 3344 destroy_workqueue(nvme_workq);
b3fffdef
KB		 3345 class_destroy(nvme_class);
3346 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
b9afca3e 3347 BUG_ON(nvme_thread && !IS_ERR(nvme_thread));
21bd78bc 3348 _nvme_check_size();
b60503ba
MW		 3349}
3350
3351MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
3352MODULE_LICENSE("GPL");
c78b4713 3353MODULE_VERSION("1.0");
b60503ba
MW		 3354module_init(nvme_init);
3355module_exit(nvme_exit);
Linux 6.x block layer and io_uring tree(s)