pci.c 86.5 KB
Newer Older
M
Matthew Wilcox 已提交
1 2
/*
 * NVM Express device driver
3
 * Copyright (c) 2011-2014, Intel Corporation.
M
Matthew Wilcox 已提交
4 5 6 7 8 9 10 11 12 13 14
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 */

15
#include <linux/bitops.h>
M
Matthew Wilcox 已提交
16
#include <linux/blkdev.h>
M
Matias Bjørling 已提交
17
#include <linux/blk-mq.h>
K
Keith Busch 已提交
18
#include <linux/cpu.h>
19
#include <linux/delay.h>
M
Matthew Wilcox 已提交
20 21 22
#include <linux/errno.h>
#include <linux/fs.h>
#include <linux/genhd.h>
K
Keith Busch 已提交
23
#include <linux/hdreg.h>
24
#include <linux/idr.h>
M
Matthew Wilcox 已提交
25 26 27 28
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/io.h>
#include <linux/kdev_t.h>
29
#include <linux/kthread.h>
M
Matthew Wilcox 已提交
30
#include <linux/kernel.h>
K
Keith Busch 已提交
31
#include <linux/list_sort.h>
M
Matthew Wilcox 已提交
32 33 34 35
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/pci.h>
36
#include <linux/poison.h>
37
#include <linux/ptrace.h>
M
Matthew Wilcox 已提交
38 39
#include <linux/sched.h>
#include <linux/slab.h>
K
Keith Busch 已提交
40
#include <linux/t10-pi.h>
M
Matthew Wilcox 已提交
41
#include <linux/types.h>
K
Keith Busch 已提交
42
#include <linux/pr.h>
V
Vishal Verma 已提交
43
#include <scsi/sg.h>
44
#include <linux/io-64-nonatomic-lo-hi.h>
K
Keith Busch 已提交
45
#include <asm/unaligned.h>
46

47
#include <uapi/linux/nvme_ioctl.h>
48 49
#include "nvme.h"

50
#define NVME_MINORS		(1U << MINORBITS)
51
#define NVME_Q_DEPTH		1024
J
Jens Axboe 已提交
52
#define NVME_AQ_DEPTH		256
M
Matthew Wilcox 已提交
53 54
#define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
#define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
55
#define ADMIN_TIMEOUT		(admin_timeout * HZ)
56
#define SHUTDOWN_TIMEOUT	(shutdown_timeout * HZ)
57 58 59 60

static unsigned char admin_timeout = 60;
module_param(admin_timeout, byte, 0644);
MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
M
Matthew Wilcox 已提交
61

62 63
unsigned char nvme_io_timeout = 30;
module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
64
MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
M
Matthew Wilcox 已提交
65

66 67 68 69
static unsigned char shutdown_timeout = 5;
module_param(shutdown_timeout, byte, 0644);
MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");

M
Matthew Wilcox 已提交
70 71 72
static int nvme_major;
module_param(nvme_major, int, 0);

73 74 75
static int nvme_char_major;
module_param(nvme_char_major, int, 0);

76 77 78
static int use_threaded_interrupts;
module_param(use_threaded_interrupts, int, 0);

79 80 81 82
static bool use_cmb_sqes = true;
module_param(use_cmb_sqes, bool, 0644);
MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");

83 84 85
static DEFINE_SPINLOCK(dev_list_lock);
static LIST_HEAD(dev_list);
static struct task_struct *nvme_thread;
K
Keith Busch 已提交
86
static struct workqueue_struct *nvme_workq;
87
static wait_queue_head_t nvme_kthread_wait;
88

89 90
static struct class *nvme_class;

91
static int __nvme_reset(struct nvme_dev *dev);
92
static int nvme_reset(struct nvme_dev *dev);
J
Jens Axboe 已提交
93
static void nvme_process_cq(struct nvme_queue *nvmeq);
94
static void nvme_dead_ctrl(struct nvme_dev *dev);
95

K
Keith Busch 已提交
96 97 98
struct async_cmd_info {
	struct kthread_work work;
	struct kthread_worker *worker;
M
Matias Bjørling 已提交
99
	struct request *req;
K
Keith Busch 已提交
100 101 102 103
	u32 result;
	int status;
	void *ctx;
};
104

M
Matthew Wilcox 已提交
105 106 107 108 109 110
/*
 * An NVM Express queue.  Each device has at least two (one for admin
 * commands and one for I/O commands).
 */
struct nvme_queue {
	struct device *q_dmadev;
M
Matthew Wilcox 已提交
111
	struct nvme_dev *dev;
112
	char irqname[24];	/* nvme4294967295-65535\0 */
M
Matthew Wilcox 已提交
113 114
	spinlock_t q_lock;
	struct nvme_command *sq_cmds;
115
	struct nvme_command __iomem *sq_cmds_io;
M
Matthew Wilcox 已提交
116
	volatile struct nvme_completion *cqes;
117
	struct blk_mq_tags **tags;
M
Matthew Wilcox 已提交
118 119 120 121
	dma_addr_t sq_dma_addr;
	dma_addr_t cq_dma_addr;
	u32 __iomem *q_db;
	u16 q_depth;
J
Jens Axboe 已提交
122
	s16 cq_vector;
M
Matthew Wilcox 已提交
123 124 125
	u16 sq_head;
	u16 sq_tail;
	u16 cq_head;
K
Keith Busch 已提交
126
	u16 qid;
127 128
	u8 cq_phase;
	u8 cqe_seen;
K
Keith Busch 已提交
129
	struct async_cmd_info cmdinfo;
M
Matthew Wilcox 已提交
130 131 132 133 134 135 136 137 138 139 140 141
};

/*
 * Check we didin't inadvertently grow the command struct
 */
static inline void _nvme_check_size(void)
{
	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
	BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64);
	BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64);
	BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
142
	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
K
Keith Busch 已提交
143
	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
M
Matthew Wilcox 已提交
144 145 146 147
	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
K
Keith Busch 已提交
148
	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
M
Matthew Wilcox 已提交
149 150
}

151
typedef void (*nvme_completion_fn)(struct nvme_queue *, void *,
152 153
						struct nvme_completion *);

154
struct nvme_cmd_info {
155 156
	nvme_completion_fn fn;
	void *ctx;
K
Keith Busch 已提交
157
	int aborted;
M
Matias Bjørling 已提交
158
	struct nvme_queue *nvmeq;
159
	struct nvme_iod iod[0];
160 161
};

162 163 164 165 166
/*
 * Max size of iod being embedded in the request payload
 */
#define NVME_INT_PAGES		2
#define NVME_INT_BYTES(dev)	(NVME_INT_PAGES * (dev)->page_size)
C
Chong Yuan 已提交
167
#define NVME_INT_MASK		0x01
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190

/*
 * Will slightly overestimate the number of pages needed.  This is OK
 * as it only leads to a small amount of wasted memory for the lifetime of
 * the I/O.
 */
static int nvme_npages(unsigned size, struct nvme_dev *dev)
{
	unsigned nprps = DIV_ROUND_UP(size + dev->page_size, dev->page_size);
	return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
}

static unsigned int nvme_cmd_size(struct nvme_dev *dev)
{
	unsigned int ret = sizeof(struct nvme_cmd_info);

	ret += sizeof(struct nvme_iod);
	ret += sizeof(__le64 *) * nvme_npages(NVME_INT_BYTES(dev), dev);
	ret += sizeof(struct scatterlist) * NVME_INT_PAGES;

	return ret;
}

M
Matias Bjørling 已提交
191 192
static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
				unsigned int hctx_idx)
193
{
M
Matias Bjørling 已提交
194 195 196
	struct nvme_dev *dev = data;
	struct nvme_queue *nvmeq = dev->queues[0];

197 198 199 200
	WARN_ON(hctx_idx != 0);
	WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
	WARN_ON(nvmeq->tags);

M
Matias Bjørling 已提交
201
	hctx->driver_data = nvmeq;
202
	nvmeq->tags = &dev->admin_tagset.tags[0];
M
Matias Bjørling 已提交
203
	return 0;
204 205
}

206 207 208 209 210 211 212
static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
{
	struct nvme_queue *nvmeq = hctx->driver_data;

	nvmeq->tags = NULL;
}

M
Matias Bjørling 已提交
213 214 215
static int nvme_admin_init_request(void *data, struct request *req,
				unsigned int hctx_idx, unsigned int rq_idx,
				unsigned int numa_node)
216
{
M
Matias Bjørling 已提交
217 218 219 220 221 222 223
	struct nvme_dev *dev = data;
	struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = dev->queues[0];

	BUG_ON(!nvmeq);
	cmd->nvmeq = nvmeq;
	return 0;
224 225
}

M
Matias Bjørling 已提交
226 227
static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
			  unsigned int hctx_idx)
M
Matthew Wilcox 已提交
228
{
M
Matias Bjørling 已提交
229
	struct nvme_dev *dev = data;
230
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
M
Matias Bjørling 已提交
231

232 233
	if (!nvmeq->tags)
		nvmeq->tags = &dev->tagset.tags[hctx_idx];
M
Matthew Wilcox 已提交
234

235
	WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
M
Matias Bjørling 已提交
236 237
	hctx->driver_data = nvmeq;
	return 0;
M
Matthew Wilcox 已提交
238 239
}

M
Matias Bjørling 已提交
240 241 242
static int nvme_init_request(void *data, struct request *req,
				unsigned int hctx_idx, unsigned int rq_idx,
				unsigned int numa_node)
M
Matthew Wilcox 已提交
243
{
M
Matias Bjørling 已提交
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258
	struct nvme_dev *dev = data;
	struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];

	BUG_ON(!nvmeq);
	cmd->nvmeq = nvmeq;
	return 0;
}

static void nvme_set_info(struct nvme_cmd_info *cmd, void *ctx,
				nvme_completion_fn handler)
{
	cmd->fn = handler;
	cmd->ctx = ctx;
	cmd->aborted = 0;
K
Keith Busch 已提交
259
	blk_mq_start_request(blk_mq_rq_from_pdu(cmd));
M
Matthew Wilcox 已提交
260 261
}

262 263 264 265 266 267 268 269 270 271
static void *iod_get_private(struct nvme_iod *iod)
{
	return (void *) (iod->private & ~0x1UL);
}

/*
 * If bit 0 is set, the iod is embedded in the request payload.
 */
static bool iod_should_kfree(struct nvme_iod *iod)
{
C
Chong Yuan 已提交
272
	return (iod->private & NVME_INT_MASK) == 0;
273 274
}

275 276
/* Special values must be less than 0x1000 */
#define CMD_CTX_BASE		((void *)POISON_POINTER_DELTA)
277 278 279
#define CMD_CTX_CANCELLED	(0x30C + CMD_CTX_BASE)
#define CMD_CTX_COMPLETED	(0x310 + CMD_CTX_BASE)
#define CMD_CTX_INVALID		(0x314 + CMD_CTX_BASE)
280

281
static void special_completion(struct nvme_queue *nvmeq, void *ctx,
282 283 284 285 286
						struct nvme_completion *cqe)
{
	if (ctx == CMD_CTX_CANCELLED)
		return;
	if (ctx == CMD_CTX_COMPLETED) {
287
		dev_warn(nvmeq->q_dmadev,
288 289 290 291 292
				"completed id %d twice on queue %d\n",
				cqe->command_id, le16_to_cpup(&cqe->sq_id));
		return;
	}
	if (ctx == CMD_CTX_INVALID) {
293
		dev_warn(nvmeq->q_dmadev,
294 295 296 297
				"invalid id %d completed on queue %d\n",
				cqe->command_id, le16_to_cpup(&cqe->sq_id));
		return;
	}
298
	dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx);
299 300
}

M
Matias Bjørling 已提交
301
static void *cancel_cmd_info(struct nvme_cmd_info *cmd, nvme_completion_fn *fn)
M
Matthew Wilcox 已提交
302
{
303
	void *ctx;
M
Matthew Wilcox 已提交
304

305
	if (fn)
M
Matias Bjørling 已提交
306 307 308 309
		*fn = cmd->fn;
	ctx = cmd->ctx;
	cmd->fn = special_completion;
	cmd->ctx = CMD_CTX_CANCELLED;
310
	return ctx;
M
Matthew Wilcox 已提交
311 312
}

M
Matias Bjørling 已提交
313 314
static void async_req_completion(struct nvme_queue *nvmeq, void *ctx,
						struct nvme_completion *cqe)
315
{
M
Matias Bjørling 已提交
316 317 318 319 320
	u32 result = le32_to_cpup(&cqe->result);
	u16 status = le16_to_cpup(&cqe->status) >> 1;

	if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ)
		++nvmeq->dev->event_limit;
K
Keith Busch 已提交
321 322 323 324 325 326 327 328 329 330
	if (status != NVME_SC_SUCCESS)
		return;

	switch (result & 0xff07) {
	case NVME_AER_NOTICE_NS_CHANGED:
		dev_info(nvmeq->q_dmadev, "rescanning\n");
		schedule_work(&nvmeq->dev->scan_work);
	default:
		dev_warn(nvmeq->q_dmadev, "async event result %08x\n", result);
	}
M
Matthew Wilcox 已提交
331 332
}

M
Matias Bjørling 已提交
333 334
static void abort_completion(struct nvme_queue *nvmeq, void *ctx,
						struct nvme_completion *cqe)
335
{
M
Matias Bjørling 已提交
336 337 338 339
	struct request *req = ctx;

	u16 status = le16_to_cpup(&cqe->status) >> 1;
	u32 result = le32_to_cpup(&cqe->result);
340

341
	blk_mq_free_request(req);
342

M
Matias Bjørling 已提交
343 344
	dev_warn(nvmeq->q_dmadev, "Abort status:%x result:%x", status, result);
	++nvmeq->dev->abort_limit;
345 346
}

M
Matias Bjørling 已提交
347 348
static void async_completion(struct nvme_queue *nvmeq, void *ctx,
						struct nvme_completion *cqe)
M
Matthew Wilcox 已提交
349
{
M
Matias Bjørling 已提交
350 351 352 353
	struct async_cmd_info *cmdinfo = ctx;
	cmdinfo->result = le32_to_cpup(&cqe->result);
	cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
	queue_kthread_work(cmdinfo->worker, &cmdinfo->work);
354
	blk_mq_free_request(cmdinfo->req);
M
Matthew Wilcox 已提交
355 356
}

M
Matias Bjørling 已提交
357 358
static inline struct nvme_cmd_info *get_cmd_from_tag(struct nvme_queue *nvmeq,
				  unsigned int tag)
M
Matthew Wilcox 已提交
359
{
360
	struct request *req = blk_mq_tag_to_rq(*nvmeq->tags, tag);
361

M
Matias Bjørling 已提交
362
	return blk_mq_rq_to_pdu(req);
363 364
}

M
Matias Bjørling 已提交
365 366 367 368 369
/*
 * Called with local interrupts disabled and the q_lock held.  May not sleep.
 */
static void *nvme_finish_cmd(struct nvme_queue *nvmeq, int tag,
						nvme_completion_fn *fn)
370
{
M
Matias Bjørling 已提交
371 372 373 374 375 376 377 378 379 380 381 382
	struct nvme_cmd_info *cmd = get_cmd_from_tag(nvmeq, tag);
	void *ctx;
	if (tag >= nvmeq->q_depth) {
		*fn = special_completion;
		return CMD_CTX_INVALID;
	}
	if (fn)
		*fn = cmd->fn;
	ctx = cmd->ctx;
	cmd->fn = special_completion;
	cmd->ctx = CMD_CTX_COMPLETED;
	return ctx;
M
Matthew Wilcox 已提交
383 384 385
}

/**
386
 * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
M
Matthew Wilcox 已提交
387 388 389 390 391
 * @nvmeq: The queue to use
 * @cmd: The command to send
 *
 * Safe to use from interrupt context
 */
392 393
static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
						struct nvme_command *cmd)
M
Matthew Wilcox 已提交
394
{
M
Matias Bjørling 已提交
395 396
	u16 tail = nvmeq->sq_tail;

397 398 399 400 401
	if (nvmeq->sq_cmds_io)
		memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd));
	else
		memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));

M
Matthew Wilcox 已提交
402 403
	if (++tail == nvmeq->q_depth)
		tail = 0;
404
	writel(tail, nvmeq->q_db);
M
Matthew Wilcox 已提交
405 406 407
	nvmeq->sq_tail = tail;
}

408
static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
M
Matias Bjørling 已提交
409 410 411
{
	unsigned long flags;
	spin_lock_irqsave(&nvmeq->q_lock, flags);
412
	__nvme_submit_cmd(nvmeq, cmd);
M
Matias Bjørling 已提交
413 414 415
	spin_unlock_irqrestore(&nvmeq->q_lock, flags);
}

416
static __le64 **iod_list(struct nvme_iod *iod)
417
{
418
	return ((void *)iod) + iod->offset;
419 420
}

421 422
static inline void iod_init(struct nvme_iod *iod, unsigned nbytes,
			    unsigned nseg, unsigned long private)
423
{
424 425 426 427 428
	iod->private = private;
	iod->offset = offsetof(struct nvme_iod, sg[nseg]);
	iod->npages = -1;
	iod->length = nbytes;
	iod->nents = 0;
429
}
M
Matthew Wilcox 已提交
430

431
static struct nvme_iod *
432 433
__nvme_alloc_iod(unsigned nseg, unsigned bytes, struct nvme_dev *dev,
		 unsigned long priv, gfp_t gfp)
M
Matthew Wilcox 已提交
434
{
435
	struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) +
436
				sizeof(__le64 *) * nvme_npages(bytes, dev) +
437 438
				sizeof(struct scatterlist) * nseg, gfp);

439 440
	if (iod)
		iod_init(iod, bytes, nseg, priv);
441 442

	return iod;
M
Matthew Wilcox 已提交
443 444
}

445 446 447 448 449 450 451 452 453 454 455 456 457
static struct nvme_iod *nvme_alloc_iod(struct request *rq, struct nvme_dev *dev,
			               gfp_t gfp)
{
	unsigned size = !(rq->cmd_flags & REQ_DISCARD) ? blk_rq_bytes(rq) :
                                                sizeof(struct nvme_dsm_range);
	struct nvme_iod *iod;

	if (rq->nr_phys_segments <= NVME_INT_PAGES &&
	    size <= NVME_INT_BYTES(dev)) {
		struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(rq);

		iod = cmd->iod;
		iod_init(iod, size, rq->nr_phys_segments,
C
Chong Yuan 已提交
458
				(unsigned long) rq | NVME_INT_MASK);
459 460 461 462 463 464 465
		return iod;
	}

	return __nvme_alloc_iod(rq->nr_phys_segments, size, dev,
				(unsigned long) rq, gfp);
}

466
static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod)
M
Matthew Wilcox 已提交
467
{
468
	const int last_prp = dev->page_size / 8 - 1;
469 470 471 472 473 474 475 476 477 478 479 480
	int i;
	__le64 **list = iod_list(iod);
	dma_addr_t prp_dma = iod->first_dma;

	if (iod->npages == 0)
		dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
	for (i = 0; i < iod->npages; i++) {
		__le64 *prp_list = list[i];
		dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
		dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
		prp_dma = next_prp_dma;
	}
481 482 483

	if (iod_should_kfree(iod))
		kfree(iod);
M
Matthew Wilcox 已提交
484 485
}

K
Keith Busch 已提交
486 487 488 489 490 491 492 493 494 495 496 497
static int nvme_error_status(u16 status)
{
	switch (status & 0x7ff) {
	case NVME_SC_SUCCESS:
		return 0;
	case NVME_SC_CAP_EXCEEDED:
		return -ENOSPC;
	default:
		return -EIO;
	}
}

498
#ifdef CONFIG_BLK_DEV_INTEGRITY
K
Keith Busch 已提交
499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542
static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
{
	if (be32_to_cpu(pi->ref_tag) == v)
		pi->ref_tag = cpu_to_be32(p);
}

static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
{
	if (be32_to_cpu(pi->ref_tag) == p)
		pi->ref_tag = cpu_to_be32(v);
}

/**
 * nvme_dif_remap - remaps ref tags to bip seed and physical lba
 *
 * The virtual start sector is the one that was originally submitted by the
 * block layer.	Due to partitioning, MD/DM cloning, etc. the actual physical
 * start sector may be different. Remap protection information to match the
 * physical LBA on writes, and back to the original seed on reads.
 *
 * Type 0 and 3 do not have a ref tag, so no remapping required.
 */
static void nvme_dif_remap(struct request *req,
			void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
{
	struct nvme_ns *ns = req->rq_disk->private_data;
	struct bio_integrity_payload *bip;
	struct t10_pi_tuple *pi;
	void *p, *pmap;
	u32 i, nlb, ts, phys, virt;

	if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3)
		return;

	bip = bio_integrity(req->bio);
	if (!bip)
		return;

	pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset;

	p = pmap;
	virt = bip_get_seed(bip);
	phys = nvme_block_nr(ns, blk_rq_pos(req));
	nlb = (blk_rq_bytes(req) >> ns->lba_shift);
543
	ts = ns->disk->queue->integrity.tuple_size;
K
Keith Busch 已提交
544 545 546 547 548 549 550 551 552

	for (i = 0; i < nlb; i++, virt++, phys++) {
		pi = (struct t10_pi_tuple *)p;
		dif_swap(phys, virt, pi);
		p += ts;
	}
	kunmap_atomic(pmap);
}

553 554 555 556 557 558
static void nvme_init_integrity(struct nvme_ns *ns)
{
	struct blk_integrity integrity;

	switch (ns->pi_type) {
	case NVME_NS_DPS_PI_TYPE3:
559
		integrity.profile = &t10_pi_type3_crc;
560 561 562
		break;
	case NVME_NS_DPS_PI_TYPE1:
	case NVME_NS_DPS_PI_TYPE2:
563
		integrity.profile = &t10_pi_type1_crc;
564 565
		break;
	default:
566
		integrity.profile = NULL;
567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
		break;
	}
	integrity.tuple_size = ns->ms;
	blk_integrity_register(ns->disk, &integrity);
	blk_queue_max_integrity_segments(ns->queue, 1);
}
#else /* CONFIG_BLK_DEV_INTEGRITY */
static void nvme_dif_remap(struct request *req,
			void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
{
}
static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
{
}
static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
{
}
static void nvme_init_integrity(struct nvme_ns *ns)
{
}
#endif

M
Matias Bjørling 已提交
589
static void req_completion(struct nvme_queue *nvmeq, void *ctx,
M
Matthew Wilcox 已提交
590 591
						struct nvme_completion *cqe)
{
592
	struct nvme_iod *iod = ctx;
593
	struct request *req = iod_get_private(iod);
M
Matias Bjørling 已提交
594
	struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
M
Matthew Wilcox 已提交
595
	u16 status = le16_to_cpup(&cqe->status) >> 1;
596
	bool requeue = false;
597
	int error = 0;
M
Matthew Wilcox 已提交
598

599
	if (unlikely(status)) {
M
Matias Bjørling 已提交
600 601
		if (!(status & NVME_SC_DNR || blk_noretry_request(req))
		    && (jiffies - req->start_time) < req->timeout) {
602 603
			unsigned long flags;

604
			requeue = true;
M
Matias Bjørling 已提交
605
			blk_mq_requeue_request(req);
606 607 608 609
			spin_lock_irqsave(req->q->queue_lock, flags);
			if (!blk_queue_stopped(req->q))
				blk_mq_kick_requeue_list(req->q);
			spin_unlock_irqrestore(req->q->queue_lock, flags);
610
			goto release_iod;
611
		}
612

613
		if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
614
			if (cmd_rq->ctx == CMD_CTX_CANCELLED)
615 616 617
				error = -EINTR;
			else
				error = status;
618
		} else {
619
			error = nvme_error_status(status);
620
		}
621 622
	}

K
Keith Busch 已提交
623 624 625 626
	if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
		u32 result = le32_to_cpup(&cqe->result);
		req->special = (void *)(uintptr_t)result;
	}
M
Matias Bjørling 已提交
627 628

	if (cmd_rq->aborted)
629
		dev_warn(nvmeq->dev->dev,
M
Matias Bjørling 已提交
630
			"completing aborted command with status:%04x\n",
631
			error);
M
Matias Bjørling 已提交
632

633
release_iod:
K
Keith Busch 已提交
634
	if (iod->nents) {
635
		dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents,
M
Matias Bjørling 已提交
636
			rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
K
Keith Busch 已提交
637 638 639
		if (blk_integrity_rq(req)) {
			if (!rq_data_dir(req))
				nvme_dif_remap(req, nvme_dif_complete);
640
			dma_unmap_sg(nvmeq->dev->dev, iod->meta_sg, 1,
K
Keith Busch 已提交
641 642 643
				rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
		}
	}
644
	nvme_free_iod(nvmeq->dev, iod);
K
Keith Busch 已提交
645

646 647
	if (likely(!requeue))
		blk_mq_complete_request(req, error);
M
Matthew Wilcox 已提交
648 649
}

650
/* length is in bytes.  gfp flags indicates whether we may sleep. */
651 652
static int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod,
		int total_len, gfp_t gfp)
M
Matthew Wilcox 已提交
653
{
654
	struct dma_pool *pool;
655 656
	int length = total_len;
	struct scatterlist *sg = iod->sg;
M
Matthew Wilcox 已提交
657 658
	int dma_len = sg_dma_len(sg);
	u64 dma_addr = sg_dma_address(sg);
659 660
	u32 page_size = dev->page_size;
	int offset = dma_addr & (page_size - 1);
661
	__le64 *prp_list;
662
	__le64 **list = iod_list(iod);
663
	dma_addr_t prp_dma;
664
	int nprps, i;
M
Matthew Wilcox 已提交
665

666
	length -= (page_size - offset);
M
Matthew Wilcox 已提交
667
	if (length <= 0)
668
		return total_len;
M
Matthew Wilcox 已提交
669

670
	dma_len -= (page_size - offset);
M
Matthew Wilcox 已提交
671
	if (dma_len) {
672
		dma_addr += (page_size - offset);
M
Matthew Wilcox 已提交
673 674 675 676 677 678
	} else {
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
	}

679
	if (length <= page_size) {
680
		iod->first_dma = dma_addr;
681
		return total_len;
682 683
	}

684
	nprps = DIV_ROUND_UP(length, page_size);
685 686
	if (nprps <= (256 / 8)) {
		pool = dev->prp_small_pool;
687
		iod->npages = 0;
688 689
	} else {
		pool = dev->prp_page_pool;
690
		iod->npages = 1;
691 692
	}

693 694
	prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
	if (!prp_list) {
695
		iod->first_dma = dma_addr;
696
		iod->npages = -1;
697
		return (total_len - length) + page_size;
698
	}
699 700
	list[0] = prp_list;
	iod->first_dma = prp_dma;
701 702
	i = 0;
	for (;;) {
703
		if (i == page_size >> 3) {
704
			__le64 *old_prp_list = prp_list;
705
			prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
706 707 708
			if (!prp_list)
				return total_len - length;
			list[iod->npages++] = prp_list;
709 710 711
			prp_list[0] = old_prp_list[i - 1];
			old_prp_list[i - 1] = cpu_to_le64(prp_dma);
			i = 1;
712 713
		}
		prp_list[i++] = cpu_to_le64(dma_addr);
714 715 716
		dma_len -= page_size;
		dma_addr += page_size;
		length -= page_size;
717 718 719 720 721 722 723 724
		if (length <= 0)
			break;
		if (dma_len > 0)
			continue;
		BUG_ON(dma_len < 0);
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
M
Matthew Wilcox 已提交
725 726
	}

727
	return total_len;
M
Matthew Wilcox 已提交
728 729
}

730 731 732
static void nvme_submit_priv(struct nvme_queue *nvmeq, struct request *req,
		struct nvme_iod *iod)
{
733
	struct nvme_command cmnd;
734

735 736
	memcpy(&cmnd, req->cmd, sizeof(cmnd));
	cmnd.rw.command_id = req->tag;
737
	if (req->nr_phys_segments) {
738 739
		cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
		cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
740 741
	}

742
	__nvme_submit_cmd(nvmeq, &cmnd);
743 744
}

M
Matias Bjørling 已提交
745 746 747 748 749 750 751
/*
 * We reuse the small pool to allocate the 16-byte range here as it is not
 * worth having a special pool for these or additional cases to handle freeing
 * the iod.
 */
static void nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns,
		struct request *req, struct nvme_iod *iod)
752
{
753 754
	struct nvme_dsm_range *range =
				(struct nvme_dsm_range *)iod_list(iod)[0];
755
	struct nvme_command cmnd;
756 757

	range->cattr = cpu_to_le32(0);
M
Matias Bjørling 已提交
758 759
	range->nlb = cpu_to_le32(blk_rq_bytes(req) >> ns->lba_shift);
	range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
760

761 762 763 764 765 766 767
	memset(&cmnd, 0, sizeof(cmnd));
	cmnd.dsm.opcode = nvme_cmd_dsm;
	cmnd.dsm.command_id = req->tag;
	cmnd.dsm.nsid = cpu_to_le32(ns->ns_id);
	cmnd.dsm.prp1 = cpu_to_le64(iod->first_dma);
	cmnd.dsm.nr = 0;
	cmnd.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
768

769
	__nvme_submit_cmd(nvmeq, &cmnd);
770 771
}

M
Matias Bjørling 已提交
772
static void nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns,
M
Matthew Wilcox 已提交
773 774
								int cmdid)
{
775
	struct nvme_command cmnd;
M
Matthew Wilcox 已提交
776

777 778 779 780
	memset(&cmnd, 0, sizeof(cmnd));
	cmnd.common.opcode = nvme_cmd_flush;
	cmnd.common.command_id = cmdid;
	cmnd.common.nsid = cpu_to_le32(ns->ns_id);
M
Matthew Wilcox 已提交
781

782
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matthew Wilcox 已提交
783 784
}

M
Matias Bjørling 已提交
785 786
static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod,
							struct nvme_ns *ns)
M
Matthew Wilcox 已提交
787
{
788
	struct request *req = iod_get_private(iod);
789
	struct nvme_command cmnd;
M
Matias Bjørling 已提交
790 791
	u16 control = 0;
	u32 dsmgmt = 0;
M
Matthew Wilcox 已提交
792

M
Matias Bjørling 已提交
793
	if (req->cmd_flags & REQ_FUA)
M
Matthew Wilcox 已提交
794
		control |= NVME_RW_FUA;
M
Matias Bjørling 已提交
795
	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
M
Matthew Wilcox 已提交
796 797
		control |= NVME_RW_LR;

M
Matias Bjørling 已提交
798
	if (req->cmd_flags & REQ_RAHEAD)
M
Matthew Wilcox 已提交
799 800
		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;

801 802 803 804 805 806 807 808
	memset(&cmnd, 0, sizeof(cmnd));
	cmnd.rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
	cmnd.rw.command_id = req->tag;
	cmnd.rw.nsid = cpu_to_le32(ns->ns_id);
	cmnd.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
	cmnd.rw.prp2 = cpu_to_le64(iod->first_dma);
	cmnd.rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
	cmnd.rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
M
Matthew Wilcox 已提交
809

810
	if (ns->ms) {
K
Keith Busch 已提交
811 812 813 814 815 816 817 818
		switch (ns->pi_type) {
		case NVME_NS_DPS_PI_TYPE3:
			control |= NVME_RW_PRINFO_PRCHK_GUARD;
			break;
		case NVME_NS_DPS_PI_TYPE1:
		case NVME_NS_DPS_PI_TYPE2:
			control |= NVME_RW_PRINFO_PRCHK_GUARD |
					NVME_RW_PRINFO_PRCHK_REF;
819
			cmnd.rw.reftag = cpu_to_le32(
K
Keith Busch 已提交
820 821 822
					nvme_block_nr(ns, blk_rq_pos(req)));
			break;
		}
823 824 825 826 827 828
		if (blk_integrity_rq(req))
			cmnd.rw.metadata =
				cpu_to_le64(sg_dma_address(iod->meta_sg));
		else
			control |= NVME_RW_PRINFO_PRACT;
	}
K
Keith Busch 已提交
829

830 831
	cmnd.rw.control = cpu_to_le16(control);
	cmnd.rw.dsmgmt = cpu_to_le32(dsmgmt);
M
Matthew Wilcox 已提交
832

833
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matthew Wilcox 已提交
834

835
	return 0;
836 837
}

838 839 840
/*
 * NOTE: ns is NULL when called on the admin queue.
 */
M
Matias Bjørling 已提交
841 842
static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
			 const struct blk_mq_queue_data *bd)
843
{
M
Matias Bjørling 已提交
844 845
	struct nvme_ns *ns = hctx->queue->queuedata;
	struct nvme_queue *nvmeq = hctx->driver_data;
846
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
847 848
	struct request *req = bd->rq;
	struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
849
	struct nvme_iod *iod;
M
Matias Bjørling 已提交
850
	enum dma_data_direction dma_dir;
851

K
Keith Busch 已提交
852 853 854 855 856
	/*
	 * If formated with metadata, require the block layer provide a buffer
	 * unless this namespace is formated such that the metadata can be
	 * stripped/generated by the controller with PRACT=1.
	 */
857
	if (ns && ns->ms && !blk_integrity_rq(req)) {
858 859
		if (!(ns->pi_type && ns->ms == 8) &&
					req->cmd_type != REQ_TYPE_DRV_PRIV) {
860
			blk_mq_complete_request(req, -EFAULT);
K
Keith Busch 已提交
861 862 863 864
			return BLK_MQ_RQ_QUEUE_OK;
		}
	}

865
	iod = nvme_alloc_iod(req, dev, GFP_ATOMIC);
866
	if (!iod)
867
		return BLK_MQ_RQ_QUEUE_BUSY;
M
Matias Bjørling 已提交
868 869

	if (req->cmd_flags & REQ_DISCARD) {
870 871 872 873 874 875
		void *range;
		/*
		 * We reuse the small pool to allocate the 16-byte range here
		 * as it is not worth having a special pool for these or
		 * additional cases to handle freeing the iod.
		 */
876
		range = dma_pool_alloc(dev->prp_small_pool, GFP_ATOMIC,
877
						&iod->first_dma);
M
Matias Bjørling 已提交
878
		if (!range)
879
			goto retry_cmd;
880 881
		iod_list(iod)[0] = (__le64 *)range;
		iod->npages = 0;
882
	} else if (req->nr_phys_segments) {
M
Matias Bjørling 已提交
883 884
		dma_dir = rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;

885
		sg_init_table(iod->sg, req->nr_phys_segments);
M
Matias Bjørling 已提交
886
		iod->nents = blk_rq_map_sg(req->q, req, iod->sg);
887 888
		if (!iod->nents)
			goto error_cmd;
M
Matias Bjørling 已提交
889 890

		if (!dma_map_sg(nvmeq->q_dmadev, iod->sg, iod->nents, dma_dir))
891
			goto retry_cmd;
M
Matias Bjørling 已提交
892

893
		if (blk_rq_bytes(req) !=
894 895
                    nvme_setup_prps(dev, iod, blk_rq_bytes(req), GFP_ATOMIC)) {
			dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
896 897
			goto retry_cmd;
		}
K
Keith Busch 已提交
898
		if (blk_integrity_rq(req)) {
899 900 901
			if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) {
				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
						dma_dir);
K
Keith Busch 已提交
902
				goto error_cmd;
903
			}
K
Keith Busch 已提交
904 905 906

			sg_init_table(iod->meta_sg, 1);
			if (blk_rq_map_integrity_sg(
907 908 909
					req->q, req->bio, iod->meta_sg) != 1) {
				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
						dma_dir);
K
Keith Busch 已提交
910
				goto error_cmd;
911
			}
K
Keith Busch 已提交
912 913 914 915

			if (rq_data_dir(req))
				nvme_dif_remap(req, nvme_dif_prep);

916 917 918
			if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir)) {
				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
						dma_dir);
K
Keith Busch 已提交
919
				goto error_cmd;
920
			}
K
Keith Busch 已提交
921
		}
922
	}
923

K
Keith Busch 已提交
924
	nvme_set_info(cmd, iod, req_completion);
M
Matias Bjørling 已提交
925
	spin_lock_irq(&nvmeq->q_lock);
926 927 928
	if (req->cmd_type == REQ_TYPE_DRV_PRIV)
		nvme_submit_priv(nvmeq, req, iod);
	else if (req->cmd_flags & REQ_DISCARD)
M
Matias Bjørling 已提交
929 930 931 932 933 934 935 936 937 938
		nvme_submit_discard(nvmeq, ns, req, iod);
	else if (req->cmd_flags & REQ_FLUSH)
		nvme_submit_flush(nvmeq, ns, req->tag);
	else
		nvme_submit_iod(nvmeq, iod, ns);

	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
	return BLK_MQ_RQ_QUEUE_OK;

939
 error_cmd:
940
	nvme_free_iod(dev, iod);
941 942
	return BLK_MQ_RQ_QUEUE_ERROR;
 retry_cmd:
943
	nvme_free_iod(dev, iod);
944
	return BLK_MQ_RQ_QUEUE_BUSY;
M
Matthew Wilcox 已提交
945 946
}

J
Jens Axboe 已提交
947
static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag)
M
Matthew Wilcox 已提交
948
{
M
Matthew Wilcox 已提交
949
	u16 head, phase;
M
Matthew Wilcox 已提交
950 951

	head = nvmeq->cq_head;
M
Matthew Wilcox 已提交
952
	phase = nvmeq->cq_phase;
M
Matthew Wilcox 已提交
953 954

	for (;;) {
955 956
		void *ctx;
		nvme_completion_fn fn;
M
Matthew Wilcox 已提交
957
		struct nvme_completion cqe = nvmeq->cqes[head];
M
Matthew Wilcox 已提交
958
		if ((le16_to_cpu(cqe.status) & 1) != phase)
M
Matthew Wilcox 已提交
959 960 961 962
			break;
		nvmeq->sq_head = le16_to_cpu(cqe.sq_head);
		if (++head == nvmeq->q_depth) {
			head = 0;
M
Matthew Wilcox 已提交
963
			phase = !phase;
M
Matthew Wilcox 已提交
964
		}
J
Jens Axboe 已提交
965 966
		if (tag && *tag == cqe.command_id)
			*tag = -1;
M
Matias Bjørling 已提交
967
		ctx = nvme_finish_cmd(nvmeq, cqe.command_id, &fn);
968
		fn(nvmeq, ctx, &cqe);
M
Matthew Wilcox 已提交
969 970 971 972 973 974 975 976
	}

	/* If the controller ignores the cq head doorbell and continuously
	 * writes to the queue, it is theoretically possible to wrap around
	 * the queue twice and mistakenly return IRQ_NONE.  Linux only
	 * requires that 0.1% of your interrupts are handled, so this isn't
	 * a big problem.
	 */
M
Matthew Wilcox 已提交
977
	if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
J
Jens Axboe 已提交
978
		return;
M
Matthew Wilcox 已提交
979

980 981
	if (likely(nvmeq->cq_vector >= 0))
		writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
M
Matthew Wilcox 已提交
982
	nvmeq->cq_head = head;
M
Matthew Wilcox 已提交
983
	nvmeq->cq_phase = phase;
M
Matthew Wilcox 已提交
984

985
	nvmeq->cqe_seen = 1;
J
Jens Axboe 已提交
986 987 988 989 990
}

static void nvme_process_cq(struct nvme_queue *nvmeq)
{
	__nvme_process_cq(nvmeq, NULL);
M
Matthew Wilcox 已提交
991 992 993
}

static irqreturn_t nvme_irq(int irq, void *data)
994 995 996 997
{
	irqreturn_t result;
	struct nvme_queue *nvmeq = data;
	spin_lock(&nvmeq->q_lock);
998 999 1000
	nvme_process_cq(nvmeq);
	result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
	nvmeq->cqe_seen = 0;
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
	spin_unlock(&nvmeq->q_lock);
	return result;
}

static irqreturn_t nvme_irq_check(int irq, void *data)
{
	struct nvme_queue *nvmeq = data;
	struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head];
	if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase)
		return IRQ_NONE;
	return IRQ_WAKE_THREAD;
}

J
Jens Axboe 已提交
1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030
static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
{
	struct nvme_queue *nvmeq = hctx->driver_data;

	if ((le16_to_cpu(nvmeq->cqes[nvmeq->cq_head].status) & 1) ==
	    nvmeq->cq_phase) {
		spin_lock_irq(&nvmeq->q_lock);
		__nvme_process_cq(nvmeq, &tag);
		spin_unlock_irq(&nvmeq->q_lock);

		if (tag == -1)
			return 1;
	}

	return 0;
}

M
Matthew Wilcox 已提交
1031 1032 1033 1034
/*
 * Returns 0 on success.  If the result is negative, it's a Linux error code;
 * if the result is positive, it's an NVM Express status code
 */
1035 1036 1037
int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
		void *buffer, void __user *ubuffer, unsigned bufflen,
		u32 *result, unsigned timeout)
M
Matthew Wilcox 已提交
1038
{
1039 1040
	bool write = cmd->common.opcode & 1;
	struct bio *bio = NULL;
1041
	struct request *req;
1042
	int ret;
M
Matthew Wilcox 已提交
1043

1044
	req = blk_mq_alloc_request(q, write, GFP_KERNEL, false);
1045 1046
	if (IS_ERR(req))
		return PTR_ERR(req);
M
Matthew Wilcox 已提交
1047

1048
	req->cmd_type = REQ_TYPE_DRV_PRIV;
1049
	req->cmd_flags |= REQ_FAILFAST_DRIVER;
1050 1051 1052
	req->__data_len = 0;
	req->__sector = (sector_t) -1;
	req->bio = req->biotail = NULL;
M
Matthew Wilcox 已提交
1053

1054
	req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
M
Matias Bjørling 已提交
1055

1056 1057
	req->cmd = (unsigned char *)cmd;
	req->cmd_len = sizeof(struct nvme_command);
K
Keith Busch 已提交
1058
	req->special = (void *)0;
M
Matthew Wilcox 已提交
1059

1060
	if (buffer && bufflen) {
1061 1062
		ret = blk_rq_map_kern(q, req, buffer, bufflen,
				      __GFP_DIRECT_RECLAIM);
1063 1064 1065
		if (ret)
			goto out;
	} else if (ubuffer && bufflen) {
1066 1067
		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
				      __GFP_DIRECT_RECLAIM);
1068 1069 1070 1071
		if (ret)
			goto out;
		bio = req->bio;
	}
1072

1073 1074 1075
	blk_execute_rq(req->q, NULL, req, 0);
	if (bio)
		blk_rq_unmap_user(bio);
M
Matthew Wilcox 已提交
1076
	if (result)
K
Keith Busch 已提交
1077
		*result = (u32)(uintptr_t)req->special;
1078 1079
	ret = req->errors;
 out:
1080
	blk_mq_free_request(req);
1081
	return ret;
1082 1083
}

1084 1085
int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
		void *buffer, unsigned bufflen)
1086
{
1087
	return __nvme_submit_sync_cmd(q, cmd, buffer, NULL, bufflen, NULL, 0);
M
Matthew Wilcox 已提交
1088 1089
}

M
Matias Bjørling 已提交
1090 1091 1092 1093 1094 1095 1096
static int nvme_submit_async_admin_req(struct nvme_dev *dev)
{
	struct nvme_queue *nvmeq = dev->queues[0];
	struct nvme_command c;
	struct nvme_cmd_info *cmd_info;
	struct request *req;

1097
	req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC, true);
1098 1099
	if (IS_ERR(req))
		return PTR_ERR(req);
M
Matias Bjørling 已提交
1100

K
Keith Busch 已提交
1101
	req->cmd_flags |= REQ_NO_TIMEOUT;
M
Matias Bjørling 已提交
1102
	cmd_info = blk_mq_rq_to_pdu(req);
1103
	nvme_set_info(cmd_info, NULL, async_req_completion);
M
Matias Bjørling 已提交
1104 1105 1106 1107 1108

	memset(&c, 0, sizeof(c));
	c.common.opcode = nvme_admin_async_event;
	c.common.command_id = req->tag;

1109
	blk_mq_free_request(req);
1110 1111
	__nvme_submit_cmd(nvmeq, &c);
	return 0;
M
Matias Bjørling 已提交
1112 1113 1114
}

static int nvme_submit_admin_async_cmd(struct nvme_dev *dev,
K
Keith Busch 已提交
1115 1116 1117
			struct nvme_command *cmd,
			struct async_cmd_info *cmdinfo, unsigned timeout)
{
M
Matias Bjørling 已提交
1118 1119 1120
	struct nvme_queue *nvmeq = dev->queues[0];
	struct request *req;
	struct nvme_cmd_info *cmd_rq;
K
Keith Busch 已提交
1121

M
Matias Bjørling 已提交
1122
	req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_KERNEL, false);
1123 1124
	if (IS_ERR(req))
		return PTR_ERR(req);
M
Matias Bjørling 已提交
1125 1126 1127 1128 1129

	req->timeout = timeout;
	cmd_rq = blk_mq_rq_to_pdu(req);
	cmdinfo->req = req;
	nvme_set_info(cmd_rq, cmdinfo, async_completion);
K
Keith Busch 已提交
1130
	cmdinfo->status = -EINTR;
M
Matias Bjørling 已提交
1131 1132 1133

	cmd->common.command_id = req->tag;

1134 1135
	nvme_submit_cmd(nvmeq, cmd);
	return 0;
K
Keith Busch 已提交
1136 1137
}

M
Matthew Wilcox 已提交
1138 1139 1140 1141 1142 1143 1144 1145
static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
{
	struct nvme_command c;

	memset(&c, 0, sizeof(c));
	c.delete_queue.opcode = opcode;
	c.delete_queue.qid = cpu_to_le16(id);

1146
	return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
1147 1148 1149 1150 1151 1152 1153 1154
}

static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
						struct nvme_queue *nvmeq)
{
	struct nvme_command c;
	int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;

1155 1156 1157 1158
	/*
	 * Note: we (ab)use the fact the the prp fields survive if no data
	 * is attached to the request.
	 */
M
Matthew Wilcox 已提交
1159 1160 1161 1162 1163 1164 1165 1166
	memset(&c, 0, sizeof(c));
	c.create_cq.opcode = nvme_admin_create_cq;
	c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr);
	c.create_cq.cqid = cpu_to_le16(qid);
	c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
	c.create_cq.cq_flags = cpu_to_le16(flags);
	c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector);

1167
	return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
1168 1169 1170 1171 1172 1173 1174 1175
}

static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
						struct nvme_queue *nvmeq)
{
	struct nvme_command c;
	int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM;

1176 1177 1178 1179
	/*
	 * Note: we (ab)use the fact the the prp fields survive if no data
	 * is attached to the request.
	 */
M
Matthew Wilcox 已提交
1180 1181 1182 1183 1184 1185 1186 1187
	memset(&c, 0, sizeof(c));
	c.create_sq.opcode = nvme_admin_create_sq;
	c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr);
	c.create_sq.sqid = cpu_to_le16(qid);
	c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
	c.create_sq.sq_flags = cpu_to_le16(flags);
	c.create_sq.cqid = cpu_to_le16(qid);

1188
	return nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
}

static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid)
{
	return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid);
}

static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid)
{
	return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid);
}

1201
int nvme_identify_ctrl(struct nvme_dev *dev, struct nvme_id_ctrl **id)
1202
{
1203
	struct nvme_command c = { };
1204
	int error;
1205

1206 1207 1208 1209
	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
	c.identify.opcode = nvme_admin_identify;
	c.identify.cns = cpu_to_le32(1);

1210 1211 1212
	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
	if (!*id)
		return -ENOMEM;
1213

1214 1215 1216 1217 1218 1219 1220 1221 1222 1223
	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
			sizeof(struct nvme_id_ctrl));
	if (error)
		kfree(*id);
	return error;
}

int nvme_identify_ns(struct nvme_dev *dev, unsigned nsid,
		struct nvme_id_ns **id)
{
1224
	struct nvme_command c = { };
1225
	int error;
1226

1227 1228 1229 1230
	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
	c.identify.opcode = nvme_admin_identify,
	c.identify.nsid = cpu_to_le32(nsid),

1231 1232 1233 1234 1235 1236 1237 1238 1239
	*id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
	if (!*id)
		return -ENOMEM;

	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
			sizeof(struct nvme_id_ns));
	if (error)
		kfree(*id);
	return error;
1240 1241
}

V
Vishal Verma 已提交
1242
int nvme_get_features(struct nvme_dev *dev, unsigned fid, unsigned nsid,
1243
					dma_addr_t dma_addr, u32 *result)
1244 1245 1246 1247 1248
{
	struct nvme_command c;

	memset(&c, 0, sizeof(c));
	c.features.opcode = nvme_admin_get_features;
1249
	c.features.nsid = cpu_to_le32(nsid);
1250 1251 1252
	c.features.prp1 = cpu_to_le64(dma_addr);
	c.features.fid = cpu_to_le32(fid);

1253 1254
	return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0,
			result, 0);
1255 1256
}

V
Vishal Verma 已提交
1257 1258
int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11,
					dma_addr_t dma_addr, u32 *result)
1259 1260 1261 1262 1263 1264 1265 1266 1267
{
	struct nvme_command c;

	memset(&c, 0, sizeof(c));
	c.features.opcode = nvme_admin_set_features;
	c.features.prp1 = cpu_to_le64(dma_addr);
	c.features.fid = cpu_to_le32(fid);
	c.features.dword11 = cpu_to_le32(dword11);

1268 1269 1270 1271 1272 1273
	return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, NULL, 0,
			result, 0);
}

int nvme_get_log_page(struct nvme_dev *dev, struct nvme_smart_log **log)
{
1274 1275 1276 1277 1278 1279
	struct nvme_command c = { };
	int error;

	c.common.opcode = nvme_admin_get_log_page,
	c.common.nsid = cpu_to_le32(0xFFFFFFFF),
	c.common.cdw10[0] = cpu_to_le32(
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291
			(((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
			 NVME_LOG_SMART),

	*log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
	if (!*log)
		return -ENOMEM;

	error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
			sizeof(struct nvme_smart_log));
	if (error)
		kfree(*log);
	return error;
1292 1293
}

K
Keith Busch 已提交
1294
/**
M
Matias Bjørling 已提交
1295
 * nvme_abort_req - Attempt aborting a request
K
Keith Busch 已提交
1296 1297 1298 1299
 *
 * Schedule controller reset if the command was already aborted once before and
 * still hasn't been returned to the driver, or if this is the admin queue.
 */
M
Matias Bjørling 已提交
1300
static void nvme_abort_req(struct request *req)
K
Keith Busch 已提交
1301
{
M
Matias Bjørling 已提交
1302 1303
	struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = cmd_rq->nvmeq;
K
Keith Busch 已提交
1304
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
1305 1306 1307
	struct request *abort_req;
	struct nvme_cmd_info *abort_cmd;
	struct nvme_command cmd;
K
Keith Busch 已提交
1308

M
Matias Bjørling 已提交
1309
	if (!nvmeq->qid || cmd_rq->aborted) {
1310 1311 1312 1313 1314 1315 1316
		spin_lock(&dev_list_lock);
		if (!__nvme_reset(dev)) {
			dev_warn(dev->dev,
				 "I/O %d QID %d timeout, reset controller\n",
				 req->tag, nvmeq->qid);
		}
		spin_unlock(&dev_list_lock);
K
Keith Busch 已提交
1317 1318 1319 1320 1321 1322
		return;
	}

	if (!dev->abort_limit)
		return;

M
Matias Bjørling 已提交
1323 1324
	abort_req = blk_mq_alloc_request(dev->admin_q, WRITE, GFP_ATOMIC,
									false);
1325
	if (IS_ERR(abort_req))
K
Keith Busch 已提交
1326 1327
		return;

M
Matias Bjørling 已提交
1328 1329 1330
	abort_cmd = blk_mq_rq_to_pdu(abort_req);
	nvme_set_info(abort_cmd, abort_req, abort_completion);

K
Keith Busch 已提交
1331 1332
	memset(&cmd, 0, sizeof(cmd));
	cmd.abort.opcode = nvme_admin_abort_cmd;
M
Matias Bjørling 已提交
1333
	cmd.abort.cid = req->tag;
K
Keith Busch 已提交
1334
	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);
M
Matias Bjørling 已提交
1335
	cmd.abort.command_id = abort_req->tag;
K
Keith Busch 已提交
1336 1337

	--dev->abort_limit;
M
Matias Bjørling 已提交
1338
	cmd_rq->aborted = 1;
K
Keith Busch 已提交
1339

M
Matias Bjørling 已提交
1340
	dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", req->tag,
K
Keith Busch 已提交
1341
							nvmeq->qid);
1342
	nvme_submit_cmd(dev->queues[0], &cmd);
K
Keith Busch 已提交
1343 1344
}

1345
static void nvme_cancel_queue_ios(struct request *req, void *data, bool reserved)
1346
{
M
Matias Bjørling 已提交
1347 1348 1349 1350
	struct nvme_queue *nvmeq = data;
	void *ctx;
	nvme_completion_fn fn;
	struct nvme_cmd_info *cmd;
K
Keith Busch 已提交
1351 1352 1353 1354
	struct nvme_completion cqe;

	if (!blk_mq_request_started(req))
		return;
1355

M
Matias Bjørling 已提交
1356
	cmd = blk_mq_rq_to_pdu(req);
1357

M
Matias Bjørling 已提交
1358 1359 1360
	if (cmd->ctx == CMD_CTX_CANCELLED)
		return;

K
Keith Busch 已提交
1361 1362 1363 1364 1365 1366
	if (blk_queue_dying(req->q))
		cqe.status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1);
	else
		cqe.status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);


M
Matias Bjørling 已提交
1367 1368 1369 1370
	dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n",
						req->tag, nvmeq->qid);
	ctx = cancel_cmd_info(cmd, &fn);
	fn(nvmeq, ctx, &cqe);
1371 1372
}

M
Matias Bjørling 已提交
1373
static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
1374
{
M
Matias Bjørling 已提交
1375 1376 1377 1378 1379
	struct nvme_cmd_info *cmd = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = cmd->nvmeq;

	dev_warn(nvmeq->q_dmadev, "Timeout I/O %d QID %d\n", req->tag,
							nvmeq->qid);
1380
	spin_lock_irq(&nvmeq->q_lock);
1381
	nvme_abort_req(req);
1382
	spin_unlock_irq(&nvmeq->q_lock);
M
Matias Bjørling 已提交
1383

1384 1385 1386 1387 1388 1389
	/*
	 * The aborted req will be completed on receiving the abort req.
	 * We enable the timer again. If hit twice, it'll cause a device reset,
	 * as the device then is in a faulty state.
	 */
	return BLK_EH_RESET_TIMER;
M
Matias Bjørling 已提交
1390
}
1391

M
Matias Bjørling 已提交
1392 1393
static void nvme_free_queue(struct nvme_queue *nvmeq)
{
1394 1395
	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
1396 1397
	if (nvmeq->sq_cmds)
		dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
1398 1399 1400 1401
					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
	kfree(nvmeq);
}

1402
static void nvme_free_queues(struct nvme_dev *dev, int lowest)
1403 1404 1405
{
	int i;

1406
	for (i = dev->queue_count - 1; i >= lowest; i--) {
M
Matias Bjørling 已提交
1407
		struct nvme_queue *nvmeq = dev->queues[i];
1408
		dev->queue_count--;
M
Matias Bjørling 已提交
1409
		dev->queues[i] = NULL;
K
Keith Busch 已提交
1410
		nvme_free_queue(nvmeq);
1411
	}
1412 1413
}

K
Keith Busch 已提交
1414 1415 1416 1417 1418
/**
 * nvme_suspend_queue - put queue into suspended state
 * @nvmeq - queue to suspend
 */
static int nvme_suspend_queue(struct nvme_queue *nvmeq)
M
Matthew Wilcox 已提交
1419
{
K
Keith Busch 已提交
1420
	int vector;
M
Matthew Wilcox 已提交
1421

1422
	spin_lock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
1423 1424 1425 1426 1427
	if (nvmeq->cq_vector == -1) {
		spin_unlock_irq(&nvmeq->q_lock);
		return 1;
	}
	vector = nvmeq->dev->entry[nvmeq->cq_vector].vector;
K
Keith Busch 已提交
1428
	nvmeq->dev->online_queues--;
K
Keith Busch 已提交
1429
	nvmeq->cq_vector = -1;
1430 1431
	spin_unlock_irq(&nvmeq->q_lock);

1432 1433 1434
	if (!nvmeq->qid && nvmeq->dev->admin_q)
		blk_mq_freeze_queue_start(nvmeq->dev->admin_q);

M
Matthew Wilcox 已提交
1435 1436
	irq_set_affinity_hint(vector, NULL);
	free_irq(vector, nvmeq);
M
Matthew Wilcox 已提交
1437

K
Keith Busch 已提交
1438 1439
	return 0;
}
M
Matthew Wilcox 已提交
1440

K
Keith Busch 已提交
1441 1442
static void nvme_clear_queue(struct nvme_queue *nvmeq)
{
1443
	spin_lock_irq(&nvmeq->q_lock);
1444 1445
	if (nvmeq->tags && *nvmeq->tags)
		blk_mq_all_tag_busy_iter(*nvmeq->tags, nvme_cancel_queue_ios, nvmeq);
1446
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1447 1448
}

K
Keith Busch 已提交
1449 1450
static void nvme_disable_queue(struct nvme_dev *dev, int qid)
{
M
Matias Bjørling 已提交
1451
	struct nvme_queue *nvmeq = dev->queues[qid];
K
Keith Busch 已提交
1452 1453 1454 1455 1456 1457

	if (!nvmeq)
		return;
	if (nvme_suspend_queue(nvmeq))
		return;

K
Keith Busch 已提交
1458 1459 1460
	/* Don't tell the adapter to delete the admin queue.
	 * Don't tell a removed adapter to delete IO queues. */
	if (qid && readl(&dev->bar->csts) != -1) {
M
Matthew Wilcox 已提交
1461 1462 1463
		adapter_delete_sq(dev, qid);
		adapter_delete_cq(dev, qid);
	}
1464 1465 1466 1467

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1468 1469
}

1470 1471 1472 1473 1474 1475 1476
static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
				int entry_size)
{
	int q_depth = dev->q_depth;
	unsigned q_size_aligned = roundup(q_depth * entry_size, dev->page_size);

	if (q_size_aligned * nr_io_queues > dev->cmb_size) {
1477 1478 1479
		u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
		mem_per_q = round_down(mem_per_q, dev->page_size);
		q_depth = div_u64(mem_per_q, entry_size);
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510

		/*
		 * Ensure the reduced q_depth is above some threshold where it
		 * would be better to map queues in system memory with the
		 * original depth
		 */
		if (q_depth < 64)
			return -ENOMEM;
	}

	return q_depth;
}

static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq,
				int qid, int depth)
{
	if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) {
		unsigned offset = (qid - 1) *
					roundup(SQ_SIZE(depth), dev->page_size);
		nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset;
		nvmeq->sq_cmds_io = dev->cmb + offset;
	} else {
		nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth),
					&nvmeq->sq_dma_addr, GFP_KERNEL);
		if (!nvmeq->sq_cmds)
			return -ENOMEM;
	}

	return 0;
}

M
Matthew Wilcox 已提交
1511
static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
K
Keith Busch 已提交
1512
							int depth)
M
Matthew Wilcox 已提交
1513
{
M
Matias Bjørling 已提交
1514
	struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
M
Matthew Wilcox 已提交
1515 1516 1517
	if (!nvmeq)
		return NULL;

1518
	nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
J
Joe Perches 已提交
1519
					  &nvmeq->cq_dma_addr, GFP_KERNEL);
M
Matthew Wilcox 已提交
1520 1521 1522
	if (!nvmeq->cqes)
		goto free_nvmeq;

1523
	if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
M
Matthew Wilcox 已提交
1524 1525
		goto free_cqdma;

1526
	nvmeq->q_dmadev = dev->dev;
M
Matthew Wilcox 已提交
1527
	nvmeq->dev = dev;
1528 1529
	snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
			dev->instance, qid);
M
Matthew Wilcox 已提交
1530 1531
	spin_lock_init(&nvmeq->q_lock);
	nvmeq->cq_head = 0;
M
Matthew Wilcox 已提交
1532
	nvmeq->cq_phase = 1;
1533
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
M
Matthew Wilcox 已提交
1534
	nvmeq->q_depth = depth;
K
Keith Busch 已提交
1535
	nvmeq->qid = qid;
1536
	nvmeq->cq_vector = -1;
M
Matias Bjørling 已提交
1537
	dev->queues[qid] = nvmeq;
M
Matthew Wilcox 已提交
1538

1539 1540 1541 1542
	/* make sure queue descriptor is set before queue count, for kthread */
	mb();
	dev->queue_count++;

M
Matthew Wilcox 已提交
1543 1544 1545
	return nvmeq;

 free_cqdma:
1546
	dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
M
Matthew Wilcox 已提交
1547 1548 1549 1550 1551 1552
							nvmeq->cq_dma_addr);
 free_nvmeq:
	kfree(nvmeq);
	return NULL;
}

1553 1554 1555
static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq,
							const char *name)
{
1556 1557
	if (use_threaded_interrupts)
		return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector,
1558
					nvme_irq_check, nvme_irq, IRQF_SHARED,
1559
					name, nvmeq);
1560
	return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq,
1561
				IRQF_SHARED, name, nvmeq);
1562 1563
}

1564
static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
M
Matthew Wilcox 已提交
1565
{
1566
	struct nvme_dev *dev = nvmeq->dev;
M
Matthew Wilcox 已提交
1567

1568
	spin_lock_irq(&nvmeq->q_lock);
1569 1570 1571
	nvmeq->sq_tail = 0;
	nvmeq->cq_head = 0;
	nvmeq->cq_phase = 1;
1572
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
1573
	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
K
Keith Busch 已提交
1574
	dev->online_queues++;
1575
	spin_unlock_irq(&nvmeq->q_lock);
1576 1577 1578 1579 1580 1581
}

static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
{
	struct nvme_dev *dev = nvmeq->dev;
	int result;
1582

K
Keith Busch 已提交
1583
	nvmeq->cq_vector = qid - 1;
M
Matthew Wilcox 已提交
1584 1585
	result = adapter_alloc_cq(dev, qid, nvmeq);
	if (result < 0)
1586
		return result;
M
Matthew Wilcox 已提交
1587 1588 1589 1590 1591

	result = adapter_alloc_sq(dev, qid, nvmeq);
	if (result < 0)
		goto release_cq;

1592
	result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
M
Matthew Wilcox 已提交
1593 1594 1595
	if (result < 0)
		goto release_sq;

1596 1597
	nvme_init_queue(nvmeq, qid);
	return result;
M
Matthew Wilcox 已提交
1598 1599 1600 1601 1602

 release_sq:
	adapter_delete_sq(dev, qid);
 release_cq:
	adapter_delete_cq(dev, qid);
1603
	return result;
M
Matthew Wilcox 已提交
1604 1605
}

1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617
static int nvme_wait_ready(struct nvme_dev *dev, u64 cap, bool enabled)
{
	unsigned long timeout;
	u32 bit = enabled ? NVME_CSTS_RDY : 0;

	timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;

	while ((readl(&dev->bar->csts) & NVME_CSTS_RDY) != bit) {
		msleep(100);
		if (fatal_signal_pending(current))
			return -EINTR;
		if (time_after(jiffies, timeout)) {
1618
			dev_err(dev->dev,
M
Matthew Wilcox 已提交
1619 1620
				"Device not ready; aborting %s\n", enabled ?
						"initialisation" : "reset");
1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
			return -ENODEV;
		}
	}

	return 0;
}

/*
 * If the device has been passed off to us in an enabled state, just clear
 * the enabled bit.  The spec says we should set the 'shutdown notification
 * bits', but doing so may cause the device to complete commands to the
 * admin queue ... and we don't know what memory that might be pointing at!
 */
static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap)
{
1636 1637 1638
	dev->ctrl_config &= ~NVME_CC_SHN_MASK;
	dev->ctrl_config &= ~NVME_CC_ENABLE;
	writel(dev->ctrl_config, &dev->bar->cc);
1639

1640 1641 1642 1643 1644
	return nvme_wait_ready(dev, cap, false);
}

static int nvme_enable_ctrl(struct nvme_dev *dev, u64 cap)
{
1645 1646 1647 1648
	dev->ctrl_config &= ~NVME_CC_SHN_MASK;
	dev->ctrl_config |= NVME_CC_ENABLE;
	writel(dev->ctrl_config, &dev->bar->cc);

1649 1650 1651
	return nvme_wait_ready(dev, cap, true);
}

K
Keith Busch 已提交
1652 1653 1654 1655
static int nvme_shutdown_ctrl(struct nvme_dev *dev)
{
	unsigned long timeout;

1656 1657 1658 1659
	dev->ctrl_config &= ~NVME_CC_SHN_MASK;
	dev->ctrl_config |= NVME_CC_SHN_NORMAL;

	writel(dev->ctrl_config, &dev->bar->cc);
K
Keith Busch 已提交
1660

1661
	timeout = SHUTDOWN_TIMEOUT + jiffies;
K
Keith Busch 已提交
1662 1663 1664 1665 1666 1667
	while ((readl(&dev->bar->csts) & NVME_CSTS_SHST_MASK) !=
							NVME_CSTS_SHST_CMPLT) {
		msleep(100);
		if (fatal_signal_pending(current))
			return -EINTR;
		if (time_after(jiffies, timeout)) {
1668
			dev_err(dev->dev,
K
Keith Busch 已提交
1669 1670 1671 1672 1673 1674 1675 1676
				"Device shutdown incomplete; abort shutdown\n");
			return -ENODEV;
		}
	}

	return 0;
}

M
Matias Bjørling 已提交
1677
static struct blk_mq_ops nvme_mq_admin_ops = {
1678
	.queue_rq	= nvme_queue_rq,
M
Matias Bjørling 已提交
1679 1680
	.map_queue	= blk_mq_map_queue,
	.init_hctx	= nvme_admin_init_hctx,
1681
	.exit_hctx      = nvme_admin_exit_hctx,
M
Matias Bjørling 已提交
1682 1683 1684 1685 1686 1687 1688 1689 1690 1691
	.init_request	= nvme_admin_init_request,
	.timeout	= nvme_timeout,
};

static struct blk_mq_ops nvme_mq_ops = {
	.queue_rq	= nvme_queue_rq,
	.map_queue	= blk_mq_map_queue,
	.init_hctx	= nvme_init_hctx,
	.init_request	= nvme_init_request,
	.timeout	= nvme_timeout,
J
Jens Axboe 已提交
1692
	.poll		= nvme_poll,
M
Matias Bjørling 已提交
1693 1694
};

1695 1696 1697 1698 1699 1700 1701 1702
static void nvme_dev_remove_admin(struct nvme_dev *dev)
{
	if (dev->admin_q && !blk_queue_dying(dev->admin_q)) {
		blk_cleanup_queue(dev->admin_q);
		blk_mq_free_tag_set(&dev->admin_tagset);
	}
}

M
Matias Bjørling 已提交
1703 1704 1705 1706 1707 1708
static int nvme_alloc_admin_tags(struct nvme_dev *dev)
{
	if (!dev->admin_q) {
		dev->admin_tagset.ops = &nvme_mq_admin_ops;
		dev->admin_tagset.nr_hw_queues = 1;
		dev->admin_tagset.queue_depth = NVME_AQ_DEPTH - 1;
1709
		dev->admin_tagset.reserved_tags = 1;
M
Matias Bjørling 已提交
1710
		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
1711
		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
1712
		dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
M
Matias Bjørling 已提交
1713 1714 1715 1716 1717 1718
		dev->admin_tagset.driver_data = dev;

		if (blk_mq_alloc_tag_set(&dev->admin_tagset))
			return -ENOMEM;

		dev->admin_q = blk_mq_init_queue(&dev->admin_tagset);
1719
		if (IS_ERR(dev->admin_q)) {
M
Matias Bjørling 已提交
1720 1721 1722
			blk_mq_free_tag_set(&dev->admin_tagset);
			return -ENOMEM;
		}
1723 1724
		if (!blk_get_queue(dev->admin_q)) {
			nvme_dev_remove_admin(dev);
1725
			dev->admin_q = NULL;
1726 1727
			return -ENODEV;
		}
K
Keith Busch 已提交
1728 1729
	} else
		blk_mq_unfreeze_queue(dev->admin_q);
M
Matias Bjørling 已提交
1730 1731 1732 1733

	return 0;
}

1734
static int nvme_configure_admin_queue(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1735
{
1736
	int result;
M
Matthew Wilcox 已提交
1737
	u32 aqa;
1738
	u64 cap = lo_hi_readq(&dev->bar->cap);
M
Matthew Wilcox 已提交
1739
	struct nvme_queue *nvmeq;
1740 1741 1742 1743 1744 1745
	/*
	 * default to a 4K page size, with the intention to update this
	 * path in the future to accomodate architectures with differing
	 * kernel and IO page sizes.
	 */
	unsigned page_shift = 12;
1746 1747 1748
	unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12;

	if (page_shift < dev_page_min) {
1749
		dev_err(dev->dev,
1750 1751 1752 1753 1754
				"Minimum device page size (%u) too large for "
				"host (%u)\n", 1 << dev_page_min,
				1 << page_shift);
		return -ENODEV;
	}
M
Matthew Wilcox 已提交
1755

1756 1757 1758 1759 1760 1761
	dev->subsystem = readl(&dev->bar->vs) >= NVME_VS(1, 1) ?
						NVME_CAP_NSSRC(cap) : 0;

	if (dev->subsystem && (readl(&dev->bar->csts) & NVME_CSTS_NSSRO))
		writel(NVME_CSTS_NSSRO, &dev->bar->csts);

1762 1763 1764
	result = nvme_disable_ctrl(dev, cap);
	if (result < 0)
		return result;
M
Matthew Wilcox 已提交
1765

M
Matias Bjørling 已提交
1766
	nvmeq = dev->queues[0];
1767
	if (!nvmeq) {
K
Keith Busch 已提交
1768
		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
1769 1770 1771
		if (!nvmeq)
			return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1772 1773 1774 1775

	aqa = nvmeq->q_depth - 1;
	aqa |= aqa << 16;

1776 1777
	dev->page_size = 1 << page_shift;

1778
	dev->ctrl_config = NVME_CC_CSS_NVM;
1779
	dev->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
M
Matthew Wilcox 已提交
1780
	dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1781
	dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
M
Matthew Wilcox 已提交
1782 1783

	writel(aqa, &dev->bar->aqa);
1784 1785
	lo_hi_writeq(nvmeq->sq_dma_addr, &dev->bar->asq);
	lo_hi_writeq(nvmeq->cq_dma_addr, &dev->bar->acq);
M
Matthew Wilcox 已提交
1786

1787
	result = nvme_enable_ctrl(dev, cap);
1788
	if (result)
M
Matias Bjørling 已提交
1789 1790
		goto free_nvmeq;

K
Keith Busch 已提交
1791
	nvmeq->cq_vector = 0;
1792
	result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
1793 1794
	if (result) {
		nvmeq->cq_vector = -1;
K
Keith Busch 已提交
1795
		goto free_nvmeq;
1796
	}
1797

M
Matthew Wilcox 已提交
1798
	return result;
M
Matias Bjørling 已提交
1799 1800 1801 1802

 free_nvmeq:
	nvme_free_queues(dev, 0);
	return result;
M
Matthew Wilcox 已提交
1803 1804
}

M
Matthew Wilcox 已提交
1805 1806 1807 1808 1809
static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
{
	struct nvme_dev *dev = ns->dev;
	struct nvme_user_io io;
	struct nvme_command c;
1810
	unsigned length, meta_len;
1811 1812 1813
	int status, write;
	dma_addr_t meta_dma = 0;
	void *meta = NULL;
A
Arnd Bergmann 已提交
1814
	void __user *metadata;
M
Matthew Wilcox 已提交
1815 1816 1817

	if (copy_from_user(&io, uio, sizeof(io)))
		return -EFAULT;
1818 1819 1820 1821

	switch (io.opcode) {
	case nvme_cmd_write:
	case nvme_cmd_read:
M
Matthew Wilcox 已提交
1822
	case nvme_cmd_compare:
M
Matthew Wilcox 已提交
1823
		break;
1824
	default:
M
Matthew Wilcox 已提交
1825
		return -EINVAL;
1826 1827
	}

1828 1829
	length = (io.nblocks + 1) << ns->lba_shift;
	meta_len = (io.nblocks + 1) * ns->ms;
A
Arnd Bergmann 已提交
1830
	metadata = (void __user *)(uintptr_t)io.metadata;
1831
	write = io.opcode & 1;
M
Matthew Wilcox 已提交
1832

1833 1834 1835
	if (ns->ext) {
		length += meta_len;
		meta_len = 0;
1836 1837
	}
	if (meta_len) {
1838 1839 1840
		if (((io.metadata & 3) || !io.metadata) && !ns->ext)
			return -EINVAL;

1841
		meta = dma_alloc_coherent(dev->dev, meta_len,
1842
						&meta_dma, GFP_KERNEL);
A
Arnd Bergmann 已提交
1843

1844 1845 1846 1847 1848
		if (!meta) {
			status = -ENOMEM;
			goto unmap;
		}
		if (write) {
A
Arnd Bergmann 已提交
1849
			if (copy_from_user(meta, metadata, meta_len)) {
1850 1851 1852 1853 1854 1855
				status = -EFAULT;
				goto unmap;
			}
		}
	}

M
Matthew Wilcox 已提交
1856 1857 1858
	memset(&c, 0, sizeof(c));
	c.rw.opcode = io.opcode;
	c.rw.flags = io.flags;
1859
	c.rw.nsid = cpu_to_le32(ns->ns_id);
M
Matthew Wilcox 已提交
1860
	c.rw.slba = cpu_to_le64(io.slba);
1861
	c.rw.length = cpu_to_le16(io.nblocks);
M
Matthew Wilcox 已提交
1862
	c.rw.control = cpu_to_le16(io.control);
1863 1864 1865 1866
	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
	c.rw.reftag = cpu_to_le32(io.reftag);
	c.rw.apptag = cpu_to_le16(io.apptag);
	c.rw.appmask = cpu_to_le16(io.appmask);
1867
	c.rw.metadata = cpu_to_le64(meta_dma);
1868 1869

	status = __nvme_submit_sync_cmd(ns->queue, &c, NULL,
A
Arnd Bergmann 已提交
1870
			(void __user *)(uintptr_t)io.addr, length, NULL, 0);
1871
 unmap:
1872 1873
	if (meta) {
		if (status == NVME_SC_SUCCESS && !write) {
A
Arnd Bergmann 已提交
1874
			if (copy_to_user(metadata, meta, meta_len))
1875 1876
				status = -EFAULT;
		}
1877
		dma_free_coherent(dev->dev, meta_len, meta, meta_dma);
1878
	}
M
Matthew Wilcox 已提交
1879 1880 1881
	return status;
}

M
Matias Bjørling 已提交
1882 1883
static int nvme_user_cmd(struct nvme_dev *dev, struct nvme_ns *ns,
			struct nvme_passthru_cmd __user *ucmd)
1884
{
1885
	struct nvme_passthru_cmd cmd;
1886
	struct nvme_command c;
1887 1888
	unsigned timeout = 0;
	int status;
1889

M
Matthew Wilcox 已提交
1890 1891 1892
	if (!capable(CAP_SYS_ADMIN))
		return -EACCES;
	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1893 1894 1895
		return -EFAULT;

	memset(&c, 0, sizeof(c));
M
Matthew Wilcox 已提交
1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907
	c.common.opcode = cmd.opcode;
	c.common.flags = cmd.flags;
	c.common.nsid = cpu_to_le32(cmd.nsid);
	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
	c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
	c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
	c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
	c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
	c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
	c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);

1908 1909
	if (cmd.timeout_ms)
		timeout = msecs_to_jiffies(cmd.timeout_ms);
1910

1911
	status = __nvme_submit_sync_cmd(ns ? ns->queue : dev->admin_q, &c,
A
Arnd Bergmann 已提交
1912
			NULL, (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1913 1914 1915 1916
			&cmd.result, timeout);
	if (status >= 0) {
		if (put_user(cmd.result, &ucmd->result))
			return -EFAULT;
M
Matthew Wilcox 已提交
1917
	}
1918

1919 1920 1921
	return status;
}

1922 1923 1924 1925 1926 1927 1928 1929 1930
static int nvme_subsys_reset(struct nvme_dev *dev)
{
	if (!dev->subsystem)
		return -ENOTTY;

	writel(0x4E564D65, &dev->bar->nssr); /* "NVMe" */
	return 0;
}

M
Matthew Wilcox 已提交
1931 1932 1933 1934 1935 1936
static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd,
							unsigned long arg)
{
	struct nvme_ns *ns = bdev->bd_disk->private_data;

	switch (cmd) {
M
Matthew Wilcox 已提交
1937
	case NVME_IOCTL_ID:
1938
		force_successful_syscall_return();
M
Matthew Wilcox 已提交
1939 1940
		return ns->ns_id;
	case NVME_IOCTL_ADMIN_CMD:
M
Matias Bjørling 已提交
1941
		return nvme_user_cmd(ns->dev, NULL, (void __user *)arg);
1942
	case NVME_IOCTL_IO_CMD:
M
Matias Bjørling 已提交
1943
		return nvme_user_cmd(ns->dev, ns, (void __user *)arg);
M
Matthew Wilcox 已提交
1944 1945
	case NVME_IOCTL_SUBMIT_IO:
		return nvme_submit_io(ns, (void __user *)arg);
V
Vishal Verma 已提交
1946 1947 1948 1949
	case SG_GET_VERSION_NUM:
		return nvme_sg_get_version_num((void __user *)arg);
	case SG_IO:
		return nvme_sg_io(ns, (void __user *)arg);
M
Matthew Wilcox 已提交
1950 1951 1952 1953 1954
	default:
		return -ENOTTY;
	}
}

K
Keith Busch 已提交
1955 1956 1957 1958 1959 1960
#ifdef CONFIG_COMPAT
static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
					unsigned int cmd, unsigned long arg)
{
	switch (cmd) {
	case SG_IO:
1961
		return -ENOIOCTLCMD;
K
Keith Busch 已提交
1962 1963 1964 1965 1966 1967 1968
	}
	return nvme_ioctl(bdev, mode, cmd, arg);
}
#else
#define nvme_compat_ioctl	NULL
#endif

1969
static void nvme_free_dev(struct kref *kref);
1970 1971 1972 1973
static void nvme_free_ns(struct kref *kref)
{
	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);

M
Matias Bjørling 已提交
1974 1975 1976
	if (ns->type == NVME_NS_LIGHTNVM)
		nvme_nvm_unregister(ns->queue, ns->disk->disk_name);

1977 1978 1979 1980
	spin_lock(&dev_list_lock);
	ns->disk->private_data = NULL;
	spin_unlock(&dev_list_lock);

1981
	kref_put(&ns->dev->kref, nvme_free_dev);
1982 1983 1984 1985
	put_disk(ns->disk);
	kfree(ns);
}

1986 1987
static int nvme_open(struct block_device *bdev, fmode_t mode)
{
1988 1989
	int ret = 0;
	struct nvme_ns *ns;
1990

1991 1992 1993 1994
	spin_lock(&dev_list_lock);
	ns = bdev->bd_disk->private_data;
	if (!ns)
		ret = -ENXIO;
1995
	else if (!kref_get_unless_zero(&ns->kref))
1996 1997 1998 1999
		ret = -ENXIO;
	spin_unlock(&dev_list_lock);

	return ret;
2000 2001 2002 2003 2004
}

static void nvme_release(struct gendisk *disk, fmode_t mode)
{
	struct nvme_ns *ns = disk->private_data;
2005
	kref_put(&ns->kref, nvme_free_ns);
2006 2007
}

K
Keith Busch 已提交
2008 2009 2010 2011 2012 2013 2014 2015 2016
static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo)
{
	/* some standard values */
	geo->heads = 1 << 6;
	geo->sectors = 1 << 5;
	geo->cylinders = get_capacity(bd->bd_disk) >> 11;
	return 0;
}

K
Keith Busch 已提交
2017 2018 2019 2020 2021 2022
static void nvme_config_discard(struct nvme_ns *ns)
{
	u32 logical_block_size = queue_logical_block_size(ns->queue);
	ns->queue->limits.discard_zeroes_data = 0;
	ns->queue->limits.discard_alignment = logical_block_size;
	ns->queue->limits.discard_granularity = logical_block_size;
2023
	blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
K
Keith Busch 已提交
2024 2025 2026
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
}

K
Keith Busch 已提交
2027 2028 2029 2030 2031
static int nvme_revalidate_disk(struct gendisk *disk)
{
	struct nvme_ns *ns = disk->private_data;
	struct nvme_dev *dev = ns->dev;
	struct nvme_id_ns *id;
2032 2033
	u8 lbaf, pi_type;
	u16 old_ms;
K
Keith Busch 已提交
2034
	unsigned short bs;
K
Keith Busch 已提交
2035

2036
	if (nvme_identify_ns(dev, ns->ns_id, &id)) {
K
Keith Busch 已提交
2037 2038 2039
		dev_warn(dev->dev, "%s: Identify failure nvme%dn%d\n", __func__,
						dev->instance, ns->ns_id);
		return -ENODEV;
K
Keith Busch 已提交
2040
	}
K
Keith Busch 已提交
2041 2042 2043
	if (id->ncap == 0) {
		kfree(id);
		return -ENODEV;
K
Keith Busch 已提交
2044
	}
K
Keith Busch 已提交
2045

M
Matias Bjørling 已提交
2046 2047 2048 2049 2050 2051 2052 2053 2054 2055
	if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
		if (nvme_nvm_register(ns->queue, disk->disk_name)) {
			dev_warn(dev->dev,
				"%s: LightNVM init failure\n", __func__);
			kfree(id);
			return -ENODEV;
		}
		ns->type = NVME_NS_LIGHTNVM;
	}

K
Keith Busch 已提交
2056 2057
	old_ms = ns->ms;
	lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
K
Keith Busch 已提交
2058
	ns->lba_shift = id->lbaf[lbaf].ds;
K
Keith Busch 已提交
2059
	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
2060
	ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
K
Keith Busch 已提交
2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073

	/*
	 * If identify namespace failed, use default 512 byte block size so
	 * block layer can use before failing read/write for 0 capacity.
	 */
	if (ns->lba_shift == 0)
		ns->lba_shift = 9;
	bs = 1 << ns->lba_shift;

	/* XXX: PI implementation requires metadata equal t10 pi tuple size */
	pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
					id->dps & NVME_NS_DPS_PI_MASK : 0;

2074
	blk_mq_freeze_queue(disk->queue);
2075 2076
	if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
				ns->ms != old_ms ||
K
Keith Busch 已提交
2077
				bs != queue_logical_block_size(disk->queue) ||
2078
				(ns->ms && ns->ext)))
K
Keith Busch 已提交
2079 2080 2081 2082 2083
		blk_integrity_unregister(disk);

	ns->pi_type = pi_type;
	blk_queue_logical_block_size(ns->queue, bs);

2084
	if (ns->ms && !ns->ext)
K
Keith Busch 已提交
2085 2086
		nvme_init_integrity(ns);

M
Matias Bjørling 已提交
2087 2088 2089
	if ((ns->ms && !(ns->ms == 8 && ns->pi_type) &&
						!blk_get_integrity(disk)) ||
						ns->type == NVME_NS_LIGHTNVM)
K
Keith Busch 已提交
2090 2091 2092 2093 2094 2095
		set_capacity(disk, 0);
	else
		set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));

	if (dev->oncs & NVME_CTRL_ONCS_DSM)
		nvme_config_discard(ns);
2096
	blk_mq_unfreeze_queue(disk->queue);
K
Keith Busch 已提交
2097

2098
	kfree(id);
K
Keith Busch 已提交
2099 2100 2101
	return 0;
}

K
Keith Busch 已提交
2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133
static char nvme_pr_type(enum pr_type type)
{
	switch (type) {
	case PR_WRITE_EXCLUSIVE:
		return 1;
	case PR_EXCLUSIVE_ACCESS:
		return 2;
	case PR_WRITE_EXCLUSIVE_REG_ONLY:
		return 3;
	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
		return 4;
	case PR_WRITE_EXCLUSIVE_ALL_REGS:
		return 5;
	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
		return 6;
	default:
		return 0;
	}
};

static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
				u64 key, u64 sa_key, u8 op)
{
	struct nvme_ns *ns = bdev->bd_disk->private_data;
	struct nvme_command c;
	u8 data[16] = { 0, };

	put_unaligned_le64(key, &data[0]);
	put_unaligned_le64(sa_key, &data[8]);

	memset(&c, 0, sizeof(c));
	c.common.opcode = op;
2134 2135
	c.common.nsid = cpu_to_le32(ns->ns_id);
	c.common.cdw10[0] = cpu_to_le32(cdw10);
K
Keith Busch 已提交
2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175

	return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
}

static int nvme_pr_register(struct block_device *bdev, u64 old,
		u64 new, unsigned flags)
{
	u32 cdw10;

	if (flags & ~PR_FL_IGNORE_KEY)
		return -EOPNOTSUPP;

	cdw10 = old ? 2 : 0;
	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
}

static int nvme_pr_reserve(struct block_device *bdev, u64 key,
		enum pr_type type, unsigned flags)
{
	u32 cdw10;

	if (flags & ~PR_FL_IGNORE_KEY)
		return -EOPNOTSUPP;

	cdw10 = nvme_pr_type(type) << 8;
	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
}

static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
		enum pr_type type, bool abort)
{
	u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
}

static int nvme_pr_clear(struct block_device *bdev, u64 key)
{
2176
	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
K
Keith Busch 已提交
2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
}

static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
{
	u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
}

static const struct pr_ops nvme_pr_ops = {
	.pr_register	= nvme_pr_register,
	.pr_reserve	= nvme_pr_reserve,
	.pr_release	= nvme_pr_release,
	.pr_preempt	= nvme_pr_preempt,
	.pr_clear	= nvme_pr_clear,
};

M
Matthew Wilcox 已提交
2194 2195 2196
static const struct block_device_operations nvme_fops = {
	.owner		= THIS_MODULE,
	.ioctl		= nvme_ioctl,
K
Keith Busch 已提交
2197
	.compat_ioctl	= nvme_compat_ioctl,
2198 2199
	.open		= nvme_open,
	.release	= nvme_release,
K
Keith Busch 已提交
2200
	.getgeo		= nvme_getgeo,
K
Keith Busch 已提交
2201
	.revalidate_disk= nvme_revalidate_disk,
K
Keith Busch 已提交
2202
	.pr_ops		= &nvme_pr_ops,
M
Matthew Wilcox 已提交
2203 2204
};

2205 2206
static int nvme_kthread(void *data)
{
2207
	struct nvme_dev *dev, *next;
2208 2209

	while (!kthread_should_stop()) {
2210
		set_current_state(TASK_INTERRUPTIBLE);
2211
		spin_lock(&dev_list_lock);
2212
		list_for_each_entry_safe(dev, next, &dev_list, node) {
2213
			int i;
2214 2215 2216 2217
			u32 csts = readl(&dev->bar->csts);

			if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
							csts & NVME_CSTS_CFS) {
2218 2219 2220 2221 2222
				if (!__nvme_reset(dev)) {
					dev_warn(dev->dev,
						"Failed status: %x, reset controller\n",
						readl(&dev->bar->csts));
				}
2223 2224
				continue;
			}
2225
			for (i = 0; i < dev->queue_count; i++) {
M
Matias Bjørling 已提交
2226
				struct nvme_queue *nvmeq = dev->queues[i];
2227 2228
				if (!nvmeq)
					continue;
2229
				spin_lock_irq(&nvmeq->q_lock);
2230
				nvme_process_cq(nvmeq);
K
Keith Busch 已提交
2231 2232

				while ((i == 0) && (dev->event_limit > 0)) {
M
Matias Bjørling 已提交
2233
					if (nvme_submit_async_admin_req(dev))
K
Keith Busch 已提交
2234 2235 2236
						break;
					dev->event_limit--;
				}
2237 2238 2239 2240
				spin_unlock_irq(&nvmeq->q_lock);
			}
		}
		spin_unlock(&dev_list_lock);
2241
		schedule_timeout(round_jiffies_relative(HZ));
2242 2243 2244 2245
	}
	return 0;
}

K
Keith Busch 已提交
2246
static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
M
Matthew Wilcox 已提交
2247 2248 2249
{
	struct nvme_ns *ns;
	struct gendisk *disk;
2250
	int node = dev_to_node(dev->dev);
M
Matthew Wilcox 已提交
2251

M
Matias Bjørling 已提交
2252
	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2253
	if (!ns)
K
Keith Busch 已提交
2254 2255
		return;

M
Matias Bjørling 已提交
2256
	ns->queue = blk_mq_init_queue(&dev->tagset);
2257
	if (IS_ERR(ns->queue))
M
Matthew Wilcox 已提交
2258
		goto out_free_ns;
M
Matthew Wilcox 已提交
2259 2260
	queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
M
Matthew Wilcox 已提交
2261 2262 2263
	ns->dev = dev;
	ns->queue->queuedata = ns;

M
Matias Bjørling 已提交
2264
	disk = alloc_disk_node(0, node);
M
Matthew Wilcox 已提交
2265 2266
	if (!disk)
		goto out_free_queue;
M
Matias Bjørling 已提交
2267

2268
	kref_init(&ns->kref);
2269
	ns->ns_id = nsid;
M
Matthew Wilcox 已提交
2270
	ns->disk = disk;
K
Keith Busch 已提交
2271 2272 2273
	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
	list_add_tail(&ns->list, &dev->namespaces);

2274
	blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
K
Keith Busch 已提交
2275
	if (dev->max_hw_sectors) {
K
Keith Busch 已提交
2276
		blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors);
K
Keith Busch 已提交
2277
		blk_queue_max_segments(ns->queue,
2278
			(dev->max_hw_sectors / (dev->page_size >> 9)) + 1);
K
Keith Busch 已提交
2279
	}
M
Matias Bjørling 已提交
2280 2281
	if (dev->stripe_size)
		blk_queue_chunk_sectors(ns->queue, dev->stripe_size >> 9);
2282 2283
	if (dev->vwc & NVME_CTRL_VWC_PRESENT)
		blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
2284
	blk_queue_virt_boundary(ns->queue, dev->page_size - 1);
M
Matthew Wilcox 已提交
2285 2286

	disk->major = nvme_major;
2287
	disk->first_minor = 0;
M
Matthew Wilcox 已提交
2288 2289 2290
	disk->fops = &nvme_fops;
	disk->private_data = ns;
	disk->queue = ns->queue;
2291
	disk->driverfs_dev = dev->device;
2292
	disk->flags = GENHD_FL_EXT_DEVT;
2293
	sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid);
M
Matthew Wilcox 已提交
2294

K
Keith Busch 已提交
2295 2296 2297 2298 2299 2300 2301
	/*
	 * Initialize capacity to 0 until we establish the namespace format and
	 * setup integrity extentions if necessary. The revalidate_disk after
	 * add_disk allows the driver to register with integrity if the format
	 * requires it.
	 */
	set_capacity(disk, 0);
K
Keith Busch 已提交
2302 2303 2304
	if (nvme_revalidate_disk(ns->disk))
		goto out_free_disk;

2305
	kref_get(&dev->kref);
M
Matias Bjørling 已提交
2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317
	if (ns->type != NVME_NS_LIGHTNVM) {
		add_disk(ns->disk);
		if (ns->ms) {
			struct block_device *bd = bdget_disk(ns->disk, 0);
			if (!bd)
				return;
			if (blkdev_get(bd, FMODE_READ, NULL)) {
				bdput(bd);
				return;
			}
			blkdev_reread_part(bd);
			blkdev_put(bd, FMODE_READ);
2318 2319
		}
	}
K
Keith Busch 已提交
2320
	return;
K
Keith Busch 已提交
2321 2322 2323
 out_free_disk:
	kfree(disk);
	list_del(&ns->list);
M
Matthew Wilcox 已提交
2324 2325 2326 2327 2328 2329
 out_free_queue:
	blk_cleanup_queue(ns->queue);
 out_free_ns:
	kfree(ns);
}

2330 2331 2332 2333 2334 2335 2336
/*
 * Create I/O queues.  Failing to create an I/O queue is not an issue,
 * we can continue with less than the desired amount of queues, and
 * even a controller without I/O queues an still be used to issue
 * admin commands.  This might be useful to upgrade a buggy firmware
 * for example.
 */
K
Keith Busch 已提交
2337 2338
static void nvme_create_io_queues(struct nvme_dev *dev)
{
M
Matias Bjørling 已提交
2339
	unsigned i;
K
Keith Busch 已提交
2340

M
Matias Bjørling 已提交
2341
	for (i = dev->queue_count; i <= dev->max_qid; i++)
K
Keith Busch 已提交
2342
		if (!nvme_alloc_queue(dev, i, dev->q_depth))
K
Keith Busch 已提交
2343 2344
			break;

M
Matias Bjørling 已提交
2345
	for (i = dev->online_queues; i <= dev->queue_count - 1; i++)
2346 2347
		if (nvme_create_queue(dev->queues[i], i)) {
			nvme_free_queues(dev, i);
K
Keith Busch 已提交
2348
			break;
2349
		}
K
Keith Busch 已提交
2350 2351
}

2352
static int set_queue_count(struct nvme_dev *dev, int count)
M
Matthew Wilcox 已提交
2353 2354 2355
{
	int status;
	u32 result;
2356
	u32 q_count = (count - 1) | ((count - 1) << 16);
M
Matthew Wilcox 已提交
2357

2358
	status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0,
2359
								&result);
M
Matthew Wilcox 已提交
2360 2361 2362
	if (status < 0)
		return status;
	if (status > 0) {
2363
		dev_err(dev->dev, "Could not set queue count (%d)\n", status);
2364
		return 0;
M
Matthew Wilcox 已提交
2365
	}
M
Matthew Wilcox 已提交
2366 2367 2368
	return min(result & 0xffff, result >> 16) + 1;
}

2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420
static void __iomem *nvme_map_cmb(struct nvme_dev *dev)
{
	u64 szu, size, offset;
	u32 cmbloc;
	resource_size_t bar_size;
	struct pci_dev *pdev = to_pci_dev(dev->dev);
	void __iomem *cmb;
	dma_addr_t dma_addr;

	if (!use_cmb_sqes)
		return NULL;

	dev->cmbsz = readl(&dev->bar->cmbsz);
	if (!(NVME_CMB_SZ(dev->cmbsz)))
		return NULL;

	cmbloc = readl(&dev->bar->cmbloc);

	szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
	size = szu * NVME_CMB_SZ(dev->cmbsz);
	offset = szu * NVME_CMB_OFST(cmbloc);
	bar_size = pci_resource_len(pdev, NVME_CMB_BIR(cmbloc));

	if (offset > bar_size)
		return NULL;

	/*
	 * Controllers may support a CMB size larger than their BAR,
	 * for example, due to being behind a bridge. Reduce the CMB to
	 * the reported size of the BAR
	 */
	if (size > bar_size - offset)
		size = bar_size - offset;

	dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(cmbloc)) + offset;
	cmb = ioremap_wc(dma_addr, size);
	if (!cmb)
		return NULL;

	dev->cmb_dma_addr = dma_addr;
	dev->cmb_size = size;
	return cmb;
}

static inline void nvme_release_cmb(struct nvme_dev *dev)
{
	if (dev->cmb) {
		iounmap(dev->cmb);
		dev->cmb = NULL;
	}
}

K
Keith Busch 已提交
2421 2422
static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
{
2423
	return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
K
Keith Busch 已提交
2424 2425
}

2426
static int nvme_setup_io_queues(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
2427
{
M
Matias Bjørling 已提交
2428
	struct nvme_queue *adminq = dev->queues[0];
2429
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
2430
	int result, i, vecs, nr_io_queues, size;
M
Matthew Wilcox 已提交
2431

K
Keith Busch 已提交
2432
	nr_io_queues = num_possible_cpus();
2433
	result = set_queue_count(dev, nr_io_queues);
2434
	if (result <= 0)
M
Matthew Wilcox 已提交
2435
		return result;
2436 2437
	if (result < nr_io_queues)
		nr_io_queues = result;
M
Matthew Wilcox 已提交
2438

2439 2440 2441 2442 2443 2444 2445 2446 2447
	if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) {
		result = nvme_cmb_qdepth(dev, nr_io_queues,
				sizeof(struct nvme_command));
		if (result > 0)
			dev->q_depth = result;
		else
			nvme_release_cmb(dev);
	}

K
Keith Busch 已提交
2448 2449
	size = db_bar_size(dev, nr_io_queues);
	if (size > 8192) {
2450
		iounmap(dev->bar);
K
Keith Busch 已提交
2451 2452 2453 2454 2455 2456 2457 2458
		do {
			dev->bar = ioremap(pci_resource_start(pdev, 0), size);
			if (dev->bar)
				break;
			if (!--nr_io_queues)
				return -ENOMEM;
			size = db_bar_size(dev, nr_io_queues);
		} while (1);
2459
		dev->dbs = ((void __iomem *)dev->bar) + 4096;
2460
		adminq->q_db = dev->dbs;
2461 2462
	}

K
Keith Busch 已提交
2463
	/* Deregister the admin queue's interrupt */
2464
	free_irq(dev->entry[0].vector, adminq);
K
Keith Busch 已提交
2465

2466 2467 2468 2469 2470 2471 2472
	/*
	 * If we enable msix early due to not intx, disable it again before
	 * setting up the full range we need.
	 */
	if (!pdev->irq)
		pci_disable_msix(pdev);

2473
	for (i = 0; i < nr_io_queues; i++)
M
Matthew Wilcox 已提交
2474
		dev->entry[i].entry = i;
2475 2476 2477 2478 2479 2480 2481 2482
	vecs = pci_enable_msix_range(pdev, dev->entry, 1, nr_io_queues);
	if (vecs < 0) {
		vecs = pci_enable_msi_range(pdev, 1, min(nr_io_queues, 32));
		if (vecs < 0) {
			vecs = 1;
		} else {
			for (i = 0; i < vecs; i++)
				dev->entry[i].vector = i + pdev->irq;
R
Ramachandra Rao Gajula 已提交
2483 2484 2485
		}
	}

2486 2487 2488 2489 2490 2491 2492
	/*
	 * Should investigate if there's a performance win from allocating
	 * more queues than interrupt vectors; it might allow the submission
	 * path to scale better, even if the receive path is limited by the
	 * number of interrupts.
	 */
	nr_io_queues = vecs;
K
Keith Busch 已提交
2493
	dev->max_qid = nr_io_queues;
2494

2495
	result = queue_request_irq(dev, adminq, adminq->irqname);
2496 2497
	if (result) {
		adminq->cq_vector = -1;
2498
		goto free_queues;
2499
	}
M
Matthew Wilcox 已提交
2500

2501
	/* Free previously allocated queues that are no longer usable */
K
Keith Busch 已提交
2502
	nvme_free_queues(dev, nr_io_queues + 1);
M
Matias Bjørling 已提交
2503
	nvme_create_io_queues(dev);
M
Matthew Wilcox 已提交
2504

2505
	return 0;
M
Matthew Wilcox 已提交
2506

2507
 free_queues:
2508
	nvme_free_queues(dev, 1);
2509
	return result;
M
Matthew Wilcox 已提交
2510 2511
}

K
Keith Busch 已提交
2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544
static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
{
	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);

	return nsa->ns_id - nsb->ns_id;
}

static struct nvme_ns *nvme_find_ns(struct nvme_dev *dev, unsigned nsid)
{
	struct nvme_ns *ns;

	list_for_each_entry(ns, &dev->namespaces, list) {
		if (ns->ns_id == nsid)
			return ns;
		if (ns->ns_id > nsid)
			break;
	}
	return NULL;
}

static inline bool nvme_io_incapable(struct nvme_dev *dev)
{
	return (!dev->bar || readl(&dev->bar->csts) & NVME_CSTS_CFS ||
							dev->online_queues < 2);
}

static void nvme_ns_remove(struct nvme_ns *ns)
{
	bool kill = nvme_io_incapable(ns->dev) && !blk_queue_dying(ns->queue);

	if (kill)
		blk_set_queue_dying(ns->queue);
2545
	if (ns->disk->flags & GENHD_FL_UP)
K
Keith Busch 已提交
2546 2547 2548 2549
		del_gendisk(ns->disk);
	if (kill || !blk_queue_dying(ns->queue)) {
		blk_mq_abort_requeue_list(ns->queue);
		blk_cleanup_queue(ns->queue);
2550 2551 2552
	}
	list_del_init(&ns->list);
	kref_put(&ns->kref, nvme_free_ns);
K
Keith Busch 已提交
2553 2554 2555 2556 2557 2558 2559 2560 2561 2562
}

static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn)
{
	struct nvme_ns *ns, *next;
	unsigned i;

	for (i = 1; i <= nn; i++) {
		ns = nvme_find_ns(dev, i);
		if (ns) {
2563
			if (revalidate_disk(ns->disk))
K
Keith Busch 已提交
2564 2565 2566 2567 2568
				nvme_ns_remove(ns);
		} else
			nvme_alloc_ns(dev, i);
	}
	list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
2569
		if (ns->ns_id > nn)
K
Keith Busch 已提交
2570 2571 2572 2573 2574
			nvme_ns_remove(ns);
	}
	list_sort(NULL, &dev->namespaces, ns_cmp);
}

2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590
static void nvme_set_irq_hints(struct nvme_dev *dev)
{
	struct nvme_queue *nvmeq;
	int i;

	for (i = 0; i < dev->online_queues; i++) {
		nvmeq = dev->queues[i];

		if (!nvmeq->tags || !(*nvmeq->tags))
			continue;

		irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector,
					blk_mq_tags_cpumask(*nvmeq->tags));
	}
}

K
Keith Busch 已提交
2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601
static void nvme_dev_scan(struct work_struct *work)
{
	struct nvme_dev *dev = container_of(work, struct nvme_dev, scan_work);
	struct nvme_id_ctrl *ctrl;

	if (!dev->tagset.tags)
		return;
	if (nvme_identify_ctrl(dev, &ctrl))
		return;
	nvme_scan_namespaces(dev, le32_to_cpup(&ctrl->nn));
	kfree(ctrl);
2602
	nvme_set_irq_hints(dev);
K
Keith Busch 已提交
2603 2604
}

2605 2606 2607 2608 2609 2610
/*
 * Return: error value if an error occurred setting up the queues or calling
 * Identify Device.  0 if these succeeded, even if adding some of the
 * namespaces failed.  At the moment, these failures are silent.  TBD which
 * failures should be reported.
 */
2611
static int nvme_dev_add(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
2612
{
2613
	struct pci_dev *pdev = to_pci_dev(dev->dev);
2614
	int res;
2615
	struct nvme_id_ctrl *ctrl;
2616
	int shift = NVME_CAP_MPSMIN(lo_hi_readq(&dev->bar->cap)) + 12;
M
Matthew Wilcox 已提交
2617

2618
	res = nvme_identify_ctrl(dev, &ctrl);
M
Matthew Wilcox 已提交
2619
	if (res) {
2620
		dev_err(dev->dev, "Identify Controller failed (%d)\n", res);
K
Keith Busch 已提交
2621
		return -EIO;
M
Matthew Wilcox 已提交
2622 2623
	}

2624
	dev->oncs = le16_to_cpup(&ctrl->oncs);
K
Keith Busch 已提交
2625
	dev->abort_limit = ctrl->acl + 1;
2626
	dev->vwc = ctrl->vwc;
2627 2628 2629
	memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn));
	memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn));
	memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
2630
	if (ctrl->mdts)
K
Keith Busch 已提交
2631
		dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9);
2632 2633
	else
		dev->max_hw_sectors = UINT_MAX;
2634
	if ((pdev->vendor == PCI_VENDOR_ID_INTEL) &&
M
Matias Bjørling 已提交
2635 2636 2637
			(pdev->device == 0x0953) && ctrl->vs[3]) {
		unsigned int max_hw_sectors;

2638
		dev->stripe_size = 1 << (ctrl->vs[3] + shift);
M
Matias Bjørling 已提交
2639 2640 2641 2642 2643 2644 2645
		max_hw_sectors = dev->stripe_size >> (shift - 9);
		if (dev->max_hw_sectors) {
			dev->max_hw_sectors = min(max_hw_sectors,
							dev->max_hw_sectors);
		} else
			dev->max_hw_sectors = max_hw_sectors;
	}
2646
	kfree(ctrl);
M
Matias Bjørling 已提交
2647

2648 2649 2650 2651 2652 2653
	if (!dev->tagset.tags) {
		dev->tagset.ops = &nvme_mq_ops;
		dev->tagset.nr_hw_queues = dev->online_queues - 1;
		dev->tagset.timeout = NVME_IO_TIMEOUT;
		dev->tagset.numa_node = dev_to_node(dev->dev);
		dev->tagset.queue_depth =
M
Matias Bjørling 已提交
2654
				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
2655 2656 2657
		dev->tagset.cmd_size = nvme_cmd_size(dev);
		dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
		dev->tagset.driver_data = dev;
M
Matthew Wilcox 已提交
2658

2659 2660 2661
		if (blk_mq_alloc_tag_set(&dev->tagset))
			return 0;
	}
K
Keith Busch 已提交
2662
	schedule_work(&dev->scan_work);
K
Keith Busch 已提交
2663
	return 0;
M
Matthew Wilcox 已提交
2664 2665
}

2666 2667
static int nvme_dev_map(struct nvme_dev *dev)
{
K
Keith Busch 已提交
2668
	u64 cap;
2669
	int bars, result = -ENOMEM;
2670
	struct pci_dev *pdev = to_pci_dev(dev->dev);
2671 2672 2673 2674 2675 2676 2677

	if (pci_enable_device_mem(pdev))
		return result;

	dev->entry[0].vector = pdev->irq;
	pci_set_master(pdev);
	bars = pci_select_bars(pdev, IORESOURCE_MEM);
2678 2679 2680
	if (!bars)
		goto disable_pci;

2681 2682 2683
	if (pci_request_selected_regions(pdev, bars, "nvme"))
		goto disable_pci;

2684 2685
	if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
	    dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
2686
		goto disable;
2687 2688 2689 2690

	dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
	if (!dev->bar)
		goto disable;
2691

K
Keith Busch 已提交
2692 2693 2694 2695
	if (readl(&dev->bar->csts) == -1) {
		result = -ENODEV;
		goto unmap;
	}
2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706

	/*
	 * Some devices don't advertse INTx interrupts, pre-enable a single
	 * MSIX vec for setup. We'll adjust this later.
	 */
	if (!pdev->irq) {
		result = pci_enable_msix(pdev, dev->entry, 1);
		if (result < 0)
			goto unmap;
	}

2707
	cap = lo_hi_readq(&dev->bar->cap);
K
Keith Busch 已提交
2708 2709
	dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
	dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
2710
	dev->dbs = ((void __iomem *)dev->bar) + 4096;
2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722

	/*
	 * Temporary fix for the Apple controller found in the MacBook8,1 and
	 * some MacBook7,1 to avoid controller resets and data loss.
	 */
	if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) {
		dev->q_depth = 2;
		dev_warn(dev->dev, "detected Apple NVMe controller, set "
			"queue depth=%u to work around controller resets\n",
			dev->q_depth);
	}

2723 2724
	if (readl(&dev->bar->vs) >= NVME_VS(1, 2))
		dev->cmb = nvme_map_cmb(dev);
2725 2726 2727

	return 0;

K
Keith Busch 已提交
2728 2729 2730
 unmap:
	iounmap(dev->bar);
	dev->bar = NULL;
2731 2732 2733 2734 2735 2736 2737 2738 2739
 disable:
	pci_release_regions(pdev);
 disable_pci:
	pci_disable_device(pdev);
	return result;
}

static void nvme_dev_unmap(struct nvme_dev *dev)
{
2740 2741 2742 2743 2744 2745
	struct pci_dev *pdev = to_pci_dev(dev->dev);

	if (pdev->msi_enabled)
		pci_disable_msi(pdev);
	else if (pdev->msix_enabled)
		pci_disable_msix(pdev);
2746 2747 2748 2749

	if (dev->bar) {
		iounmap(dev->bar);
		dev->bar = NULL;
2750
		pci_release_regions(pdev);
2751 2752
	}

2753 2754
	if (pci_is_enabled(pdev))
		pci_disable_device(pdev);
2755 2756
}

K
Keith Busch 已提交
2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773
struct nvme_delq_ctx {
	struct task_struct *waiter;
	struct kthread_worker *worker;
	atomic_t refcount;
};

static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev)
{
	dq->waiter = current;
	mb();

	for (;;) {
		set_current_state(TASK_KILLABLE);
		if (!atomic_read(&dq->refcount))
			break;
		if (!schedule_timeout(ADMIN_TIMEOUT) ||
					fatal_signal_pending(current)) {
K
Keith Busch 已提交
2774 2775 2776 2777 2778 2779 2780
			/*
			 * Disable the controller first since we can't trust it
			 * at this point, but leave the admin queue enabled
			 * until all queue deletion requests are flushed.
			 * FIXME: This may take a while if there are more h/w
			 * queues than admin tags.
			 */
K
Keith Busch 已提交
2781
			set_current_state(TASK_RUNNING);
2782
			nvme_disable_ctrl(dev, lo_hi_readq(&dev->bar->cap));
K
Keith Busch 已提交
2783
			nvme_clear_queue(dev->queues[0]);
K
Keith Busch 已提交
2784
			flush_kthread_worker(dq->worker);
K
Keith Busch 已提交
2785
			nvme_disable_queue(dev, 0);
K
Keith Busch 已提交
2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808
			return;
		}
	}
	set_current_state(TASK_RUNNING);
}

static void nvme_put_dq(struct nvme_delq_ctx *dq)
{
	atomic_dec(&dq->refcount);
	if (dq->waiter)
		wake_up_process(dq->waiter);
}

static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq)
{
	atomic_inc(&dq->refcount);
	return dq;
}

static void nvme_del_queue_end(struct nvme_queue *nvmeq)
{
	struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx;
	nvme_put_dq(dq);
2809 2810 2811 2812

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824
}

static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode,
						kthread_work_func_t fn)
{
	struct nvme_command c;

	memset(&c, 0, sizeof(c));
	c.delete_queue.opcode = opcode;
	c.delete_queue.qid = cpu_to_le16(nvmeq->qid);

	init_kthread_work(&nvmeq->cmdinfo.work, fn);
M
Matias Bjørling 已提交
2825 2826
	return nvme_submit_admin_async_cmd(nvmeq->dev, &c, &nvmeq->cmdinfo,
								ADMIN_TIMEOUT);
K
Keith Busch 已提交
2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876
}

static void nvme_del_cq_work_handler(struct kthread_work *work)
{
	struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
							cmdinfo.work);
	nvme_del_queue_end(nvmeq);
}

static int nvme_delete_cq(struct nvme_queue *nvmeq)
{
	return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq,
						nvme_del_cq_work_handler);
}

static void nvme_del_sq_work_handler(struct kthread_work *work)
{
	struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
							cmdinfo.work);
	int status = nvmeq->cmdinfo.status;

	if (!status)
		status = nvme_delete_cq(nvmeq);
	if (status)
		nvme_del_queue_end(nvmeq);
}

static int nvme_delete_sq(struct nvme_queue *nvmeq)
{
	return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq,
						nvme_del_sq_work_handler);
}

static void nvme_del_queue_start(struct kthread_work *work)
{
	struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
							cmdinfo.work);
	if (nvme_delete_sq(nvmeq))
		nvme_del_queue_end(nvmeq);
}

static void nvme_disable_io_queues(struct nvme_dev *dev)
{
	int i;
	DEFINE_KTHREAD_WORKER_ONSTACK(worker);
	struct nvme_delq_ctx dq;
	struct task_struct *kworker_task = kthread_run(kthread_worker_fn,
					&worker, "nvme%d", dev->instance);

	if (IS_ERR(kworker_task)) {
2877
		dev_err(dev->dev,
K
Keith Busch 已提交
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887
			"Failed to create queue del task\n");
		for (i = dev->queue_count - 1; i > 0; i--)
			nvme_disable_queue(dev, i);
		return;
	}

	dq.waiter = NULL;
	atomic_set(&dq.refcount, 0);
	dq.worker = &worker;
	for (i = dev->queue_count - 1; i > 0; i--) {
M
Matias Bjørling 已提交
2888
		struct nvme_queue *nvmeq = dev->queues[i];
K
Keith Busch 已提交
2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900

		if (nvme_suspend_queue(nvmeq))
			continue;
		nvmeq->cmdinfo.ctx = nvme_get_dq(&dq);
		nvmeq->cmdinfo.worker = dq.worker;
		init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start);
		queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work);
	}
	nvme_wait_dq(&dq, dev);
	kthread_stop(kworker_task);
}

2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920
/*
* Remove the node from the device list and check
* for whether or not we need to stop the nvme_thread.
*/
static void nvme_dev_list_remove(struct nvme_dev *dev)
{
	struct task_struct *tmp = NULL;

	spin_lock(&dev_list_lock);
	list_del_init(&dev->node);
	if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) {
		tmp = nvme_thread;
		nvme_thread = NULL;
	}
	spin_unlock(&dev_list_lock);

	if (tmp)
		kthread_stop(tmp);
}

2921 2922 2923 2924 2925 2926 2927
static void nvme_freeze_queues(struct nvme_dev *dev)
{
	struct nvme_ns *ns;

	list_for_each_entry(ns, &dev->namespaces, list) {
		blk_mq_freeze_queue_start(ns->queue);

2928
		spin_lock_irq(ns->queue->queue_lock);
2929
		queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
2930
		spin_unlock_irq(ns->queue->queue_lock);
2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948

		blk_mq_cancel_requeue_work(ns->queue);
		blk_mq_stop_hw_queues(ns->queue);
	}
}

static void nvme_unfreeze_queues(struct nvme_dev *dev)
{
	struct nvme_ns *ns;

	list_for_each_entry(ns, &dev->namespaces, list) {
		queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
		blk_mq_unfreeze_queue(ns->queue);
		blk_mq_start_stopped_hw_queues(ns->queue, true);
		blk_mq_kick_requeue_list(ns->queue);
	}
}

2949
static void nvme_dev_shutdown(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
2950
{
2951
	int i;
2952
	u32 csts = -1;
2953

2954
	nvme_dev_list_remove(dev);
2955

2956 2957
	if (dev->bar) {
		nvme_freeze_queues(dev);
2958
		csts = readl(&dev->bar->csts);
2959
	}
2960
	if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) {
K
Keith Busch 已提交
2961
		for (i = dev->queue_count - 1; i >= 0; i--) {
M
Matias Bjørling 已提交
2962
			struct nvme_queue *nvmeq = dev->queues[i];
K
Keith Busch 已提交
2963 2964 2965 2966
			nvme_suspend_queue(nvmeq);
		}
	} else {
		nvme_disable_io_queues(dev);
K
Keith Busch 已提交
2967
		nvme_shutdown_ctrl(dev);
K
Keith Busch 已提交
2968 2969
		nvme_disable_queue(dev, 0);
	}
2970
	nvme_dev_unmap(dev);
2971 2972 2973

	for (i = dev->queue_count - 1; i >= 0; i--)
		nvme_clear_queue(dev->queues[i]);
2974 2975 2976 2977
}

static void nvme_dev_remove(struct nvme_dev *dev)
{
2978
	struct nvme_ns *ns, *next;
2979

2980
	list_for_each_entry_safe(ns, next, &dev->namespaces, list)
K
Keith Busch 已提交
2981
		nvme_ns_remove(ns);
M
Matthew Wilcox 已提交
2982 2983
}

M
Matthew Wilcox 已提交
2984 2985
static int nvme_setup_prp_pools(struct nvme_dev *dev)
{
2986
	dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
M
Matthew Wilcox 已提交
2987 2988 2989 2990
						PAGE_SIZE, PAGE_SIZE, 0);
	if (!dev->prp_page_pool)
		return -ENOMEM;

2991
	/* Optimisation for I/Os between 4k and 128k */
2992
	dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
2993 2994 2995 2996 2997
						256, 256, 0);
	if (!dev->prp_small_pool) {
		dma_pool_destroy(dev->prp_page_pool);
		return -ENOMEM;
	}
M
Matthew Wilcox 已提交
2998 2999 3000 3001 3002 3003
	return 0;
}

static void nvme_release_prp_pools(struct nvme_dev *dev)
{
	dma_pool_destroy(dev->prp_page_pool);
3004
	dma_pool_destroy(dev->prp_small_pool);
M
Matthew Wilcox 已提交
3005 3006
}

3007 3008 3009
static DEFINE_IDA(nvme_instance_ida);

static int nvme_set_instance(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
3010
{
3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026
	int instance, error;

	do {
		if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
			return -ENODEV;

		spin_lock(&dev_list_lock);
		error = ida_get_new(&nvme_instance_ida, &instance);
		spin_unlock(&dev_list_lock);
	} while (error == -EAGAIN);

	if (error)
		return -ENODEV;

	dev->instance = instance;
	return 0;
M
Matthew Wilcox 已提交
3027 3028 3029 3030
}

static void nvme_release_instance(struct nvme_dev *dev)
{
3031 3032 3033
	spin_lock(&dev_list_lock);
	ida_remove(&nvme_instance_ida, dev->instance);
	spin_unlock(&dev_list_lock);
M
Matthew Wilcox 已提交
3034 3035
}

3036 3037 3038
static void nvme_free_dev(struct kref *kref)
{
	struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
3039

3040
	put_device(dev->dev);
3041
	put_device(dev->device);
3042
	nvme_release_instance(dev);
3043 3044 3045 3046
	if (dev->tagset.tags)
		blk_mq_free_tag_set(&dev->tagset);
	if (dev->admin_q)
		blk_put_queue(dev->admin_q);
3047 3048 3049 3050 3051 3052 3053
	kfree(dev->queues);
	kfree(dev->entry);
	kfree(dev);
}

static int nvme_dev_open(struct inode *inode, struct file *f)
{
3054 3055 3056 3057 3058 3059 3060
	struct nvme_dev *dev;
	int instance = iminor(inode);
	int ret = -ENODEV;

	spin_lock(&dev_list_lock);
	list_for_each_entry(dev, &dev_list, node) {
		if (dev->instance == instance) {
3061 3062 3063 3064
			if (!dev->admin_q) {
				ret = -EWOULDBLOCK;
				break;
			}
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074
			if (!kref_get_unless_zero(&dev->kref))
				break;
			f->private_data = dev;
			ret = 0;
			break;
		}
	}
	spin_unlock(&dev_list_lock);

	return ret;
3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086
}

static int nvme_dev_release(struct inode *inode, struct file *f)
{
	struct nvme_dev *dev = f->private_data;
	kref_put(&dev->kref, nvme_free_dev);
	return 0;
}

static long nvme_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
{
	struct nvme_dev *dev = f->private_data;
M
Matias Bjørling 已提交
3087 3088
	struct nvme_ns *ns;

3089 3090
	switch (cmd) {
	case NVME_IOCTL_ADMIN_CMD:
M
Matias Bjørling 已提交
3091
		return nvme_user_cmd(dev, NULL, (void __user *)arg);
3092
	case NVME_IOCTL_IO_CMD:
M
Matias Bjørling 已提交
3093 3094 3095 3096
		if (list_empty(&dev->namespaces))
			return -ENOTTY;
		ns = list_first_entry(&dev->namespaces, struct nvme_ns, list);
		return nvme_user_cmd(dev, ns, (void __user *)arg);
3097 3098 3099
	case NVME_IOCTL_RESET:
		dev_warn(dev->dev, "resetting controller\n");
		return nvme_reset(dev);
3100 3101
	case NVME_IOCTL_SUBSYS_RESET:
		return nvme_subsys_reset(dev);
3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114
	default:
		return -ENOTTY;
	}
}

static const struct file_operations nvme_dev_fops = {
	.owner		= THIS_MODULE,
	.open		= nvme_dev_open,
	.release	= nvme_dev_release,
	.unlocked_ioctl	= nvme_dev_ioctl,
	.compat_ioctl	= nvme_dev_ioctl,
};

3115
static void nvme_probe_work(struct work_struct *work)
3116
{
3117
	struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work);
3118
	bool start_thread = false;
3119
	int result;
3120 3121 3122

	result = nvme_dev_map(dev);
	if (result)
3123
		goto out;
3124 3125 3126 3127 3128 3129

	result = nvme_configure_admin_queue(dev);
	if (result)
		goto unmap;

	spin_lock(&dev_list_lock);
3130 3131 3132 3133
	if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) {
		start_thread = true;
		nvme_thread = NULL;
	}
3134 3135 3136
	list_add(&dev->node, &dev_list);
	spin_unlock(&dev_list_lock);

3137 3138
	if (start_thread) {
		nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
3139
		wake_up_all(&nvme_kthread_wait);
3140 3141 3142 3143 3144 3145 3146
	} else
		wait_event_killable(nvme_kthread_wait, nvme_thread);

	if (IS_ERR_OR_NULL(nvme_thread)) {
		result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR;
		goto disable;
	}
M
Matias Bjørling 已提交
3147 3148

	nvme_init_queue(dev->queues[0], 0);
K
Keith Busch 已提交
3149 3150 3151
	result = nvme_alloc_admin_tags(dev);
	if (result)
		goto disable;
3152

3153
	result = nvme_setup_io_queues(dev);
3154
	if (result)
K
Keith Busch 已提交
3155
		goto free_tags;
3156

3157
	dev->event_limit = 1;
3158

3159 3160 3161 3162
	/*
	 * Keep the controller around but remove all namespaces if we don't have
	 * any working I/O queue.
	 */
3163 3164 3165 3166 3167 3168 3169 3170 3171
	if (dev->online_queues < 2) {
		dev_warn(dev->dev, "IO queues not created\n");
		nvme_dev_remove(dev);
	} else {
		nvme_unfreeze_queues(dev);
		nvme_dev_add(dev);
	}

	return;
3172

K
Keith Busch 已提交
3173 3174
 free_tags:
	nvme_dev_remove_admin(dev);
3175 3176 3177
	blk_put_queue(dev->admin_q);
	dev->admin_q = NULL;
	dev->queues[0]->tags = NULL;
3178
 disable:
3179
	nvme_disable_queue(dev, 0);
3180
	nvme_dev_list_remove(dev);
3181 3182
 unmap:
	nvme_dev_unmap(dev);
3183 3184 3185
 out:
	if (!work_busy(&dev->reset_work))
		nvme_dead_ctrl(dev);
3186 3187
}

K
Keith Busch 已提交
3188 3189 3190
static int nvme_remove_dead_ctrl(void *arg)
{
	struct nvme_dev *dev = (struct nvme_dev *)arg;
3191
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
3192 3193

	if (pci_get_drvdata(pdev))
3194
		pci_stop_and_remove_bus_device_locked(pdev);
K
Keith Busch 已提交
3195 3196 3197 3198
	kref_put(&dev->kref, nvme_free_dev);
	return 0;
}

3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210
static void nvme_dead_ctrl(struct nvme_dev *dev)
{
	dev_warn(dev->dev, "Device failed to resume\n");
	kref_get(&dev->kref);
	if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d",
						dev->instance))) {
		dev_err(dev->dev,
			"Failed to start controller remove task\n");
		kref_put(&dev->kref, nvme_free_dev);
	}
}

3211
static void nvme_reset_work(struct work_struct *ws)
K
Keith Busch 已提交
3212
{
3213
	struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
3214 3215
	bool in_probe = work_busy(&dev->probe_work);

K
Keith Busch 已提交
3216
	nvme_dev_shutdown(dev);
3217 3218 3219 3220 3221 3222 3223 3224

	/* Synchronize with device probe so that work will see failure status
	 * and exit gracefully without trying to schedule another reset */
	flush_work(&dev->probe_work);

	/* Fail this device if reset occured during probe to avoid
	 * infinite initialization loops. */
	if (in_probe) {
3225
		nvme_dead_ctrl(dev);
3226
		return;
K
Keith Busch 已提交
3227
	}
3228 3229 3230
	/* Schedule device resume asynchronously so the reset work is available
	 * to cleanup errors that may occur during reinitialization */
	schedule_work(&dev->probe_work);
K
Keith Busch 已提交
3231 3232
}

3233
static int __nvme_reset(struct nvme_dev *dev)
T
Tejun Heo 已提交
3234
{
3235 3236 3237 3238 3239
	if (work_pending(&dev->reset_work))
		return -EBUSY;
	list_del_init(&dev->node);
	queue_work(nvme_workq, &dev->reset_work);
	return 0;
T
Tejun Heo 已提交
3240 3241
}

3242 3243
static int nvme_reset(struct nvme_dev *dev)
{
3244
	int ret;
3245 3246 3247 3248 3249

	if (!dev->admin_q || blk_queue_dying(dev->admin_q))
		return -ENODEV;

	spin_lock(&dev_list_lock);
3250
	ret = __nvme_reset(dev);
3251 3252 3253 3254
	spin_unlock(&dev_list_lock);

	if (!ret) {
		flush_work(&dev->reset_work);
3255
		flush_work(&dev->probe_work);
3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276
		return 0;
	}

	return ret;
}

static ssize_t nvme_sysfs_reset(struct device *dev,
				struct device_attribute *attr, const char *buf,
				size_t count)
{
	struct nvme_dev *ndev = dev_get_drvdata(dev);
	int ret;

	ret = nvme_reset(ndev);
	if (ret < 0)
		return ret;

	return count;
}
static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);

3277
static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
M
Matthew Wilcox 已提交
3278
{
M
Matias Bjørling 已提交
3279
	int node, result = -ENOMEM;
M
Matthew Wilcox 已提交
3280 3281
	struct nvme_dev *dev;

M
Matias Bjørling 已提交
3282 3283 3284 3285 3286
	node = dev_to_node(&pdev->dev);
	if (node == NUMA_NO_NODE)
		set_dev_node(&pdev->dev, 0);

	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
3287 3288
	if (!dev)
		return -ENOMEM;
M
Matias Bjørling 已提交
3289 3290
	dev->entry = kzalloc_node(num_possible_cpus() * sizeof(*dev->entry),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
3291 3292
	if (!dev->entry)
		goto free;
M
Matias Bjørling 已提交
3293 3294
	dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
3295 3296 3297 3298
	if (!dev->queues)
		goto free;

	INIT_LIST_HEAD(&dev->namespaces);
3299
	INIT_WORK(&dev->reset_work, nvme_reset_work);
3300
	dev->dev = get_device(&pdev->dev);
K
Keith Busch 已提交
3301
	pci_set_drvdata(pdev, dev);
3302 3303
	result = nvme_set_instance(dev);
	if (result)
K
Keith Busch 已提交
3304
		goto put_pci;
M
Matthew Wilcox 已提交
3305

M
Matthew Wilcox 已提交
3306 3307
	result = nvme_setup_prp_pools(dev);
	if (result)
3308
		goto release;
M
Matthew Wilcox 已提交
3309

3310
	kref_init(&dev->kref);
3311 3312 3313 3314 3315
	dev->device = device_create(nvme_class, &pdev->dev,
				MKDEV(nvme_char_major, dev->instance),
				dev, "nvme%d", dev->instance);
	if (IS_ERR(dev->device)) {
		result = PTR_ERR(dev->device);
3316
		goto release_pools;
3317 3318
	}
	get_device(dev->device);
3319 3320 3321 3322 3323
	dev_set_drvdata(dev->device, dev);

	result = device_create_file(dev->device, &dev_attr_reset_controller);
	if (result)
		goto put_dev;
3324

3325
	INIT_LIST_HEAD(&dev->node);
K
Keith Busch 已提交
3326
	INIT_WORK(&dev->scan_work, nvme_dev_scan);
3327
	INIT_WORK(&dev->probe_work, nvme_probe_work);
3328
	schedule_work(&dev->probe_work);
M
Matthew Wilcox 已提交
3329 3330
	return 0;

3331 3332 3333
 put_dev:
	device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
	put_device(dev->device);
3334
 release_pools:
M
Matthew Wilcox 已提交
3335
	nvme_release_prp_pools(dev);
3336 3337
 release:
	nvme_release_instance(dev);
K
Keith Busch 已提交
3338
 put_pci:
3339
	put_device(dev->dev);
M
Matthew Wilcox 已提交
3340 3341 3342 3343 3344 3345 3346
 free:
	kfree(dev->queues);
	kfree(dev->entry);
	kfree(dev);
	return result;
}

3347 3348
static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
{
K
Keith Busch 已提交
3349
	struct nvme_dev *dev = pci_get_drvdata(pdev);
3350

K
Keith Busch 已提交
3351 3352 3353
	if (prepare)
		nvme_dev_shutdown(dev);
	else
3354
		schedule_work(&dev->probe_work);
3355 3356
}

3357 3358 3359 3360 3361 3362
static void nvme_shutdown(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
	nvme_dev_shutdown(dev);
}

3363
static void nvme_remove(struct pci_dev *pdev)
M
Matthew Wilcox 已提交
3364 3365
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
K
Keith Busch 已提交
3366 3367 3368 3369 3370 3371

	spin_lock(&dev_list_lock);
	list_del_init(&dev->node);
	spin_unlock(&dev_list_lock);

	pci_set_drvdata(pdev, NULL);
3372
	flush_work(&dev->probe_work);
K
Keith Busch 已提交
3373
	flush_work(&dev->reset_work);
K
Keith Busch 已提交
3374
	flush_work(&dev->scan_work);
3375
	device_remove_file(dev->device, &dev_attr_reset_controller);
3376
	nvme_dev_remove(dev);
3377
	nvme_dev_shutdown(dev);
M
Matias Bjørling 已提交
3378
	nvme_dev_remove_admin(dev);
3379
	device_destroy(nvme_class, MKDEV(nvme_char_major, dev->instance));
3380
	nvme_free_queues(dev, 0);
3381
	nvme_release_cmb(dev);
K
Keith Busch 已提交
3382
	nvme_release_prp_pools(dev);
3383
	kref_put(&dev->kref, nvme_free_dev);
M
Matthew Wilcox 已提交
3384 3385 3386 3387 3388 3389 3390 3391
}

/* These functions are yet to be implemented */
#define nvme_error_detected NULL
#define nvme_dump_registers NULL
#define nvme_link_reset NULL
#define nvme_slot_reset NULL
#define nvme_error_resume NULL
3392

3393
#ifdef CONFIG_PM_SLEEP
3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407
static int nvme_suspend(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

	nvme_dev_shutdown(ndev);
	return 0;
}

static int nvme_resume(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

3408
	schedule_work(&ndev->probe_work);
K
Keith Busch 已提交
3409
	return 0;
3410
}
3411
#endif
3412 3413

static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
M
Matthew Wilcox 已提交
3414

3415
static const struct pci_error_handlers nvme_err_handler = {
M
Matthew Wilcox 已提交
3416 3417 3418 3419 3420
	.error_detected	= nvme_error_detected,
	.mmio_enabled	= nvme_dump_registers,
	.link_reset	= nvme_link_reset,
	.slot_reset	= nvme_slot_reset,
	.resume		= nvme_error_resume,
3421
	.reset_notify	= nvme_reset_notify,
M
Matthew Wilcox 已提交
3422 3423 3424 3425 3426
};

/* Move to pci_ids.h later */
#define PCI_CLASS_STORAGE_EXPRESS	0x010802

3427
static const struct pci_device_id nvme_id_table[] = {
M
Matthew Wilcox 已提交
3428
	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
3429
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
M
Matthew Wilcox 已提交
3430 3431 3432 3433 3434 3435 3436 3437
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, nvme_id_table);

static struct pci_driver nvme_driver = {
	.name		= "nvme",
	.id_table	= nvme_id_table,
	.probe		= nvme_probe,
3438
	.remove		= nvme_remove,
3439
	.shutdown	= nvme_shutdown,
3440 3441 3442
	.driver		= {
		.pm	= &nvme_dev_pm_ops,
	},
M
Matthew Wilcox 已提交
3443 3444 3445 3446 3447
	.err_handler	= &nvme_err_handler,
};

static int __init nvme_init(void)
{
3448
	int result;
3449

3450
	init_waitqueue_head(&nvme_kthread_wait);
M
Matthew Wilcox 已提交
3451

K
Keith Busch 已提交
3452 3453
	nvme_workq = create_singlethread_workqueue("nvme");
	if (!nvme_workq)
3454
		return -ENOMEM;
K
Keith Busch 已提交
3455

3456 3457
	result = register_blkdev(nvme_major, "nvme");
	if (result < 0)
K
Keith Busch 已提交
3458
		goto kill_workq;
3459
	else if (result > 0)
3460
		nvme_major = result;
M
Matthew Wilcox 已提交
3461

3462 3463 3464 3465 3466 3467 3468 3469
	result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
							&nvme_dev_fops);
	if (result < 0)
		goto unregister_blkdev;
	else if (result > 0)
		nvme_char_major = result;

	nvme_class = class_create(THIS_MODULE, "nvme");
3470 3471
	if (IS_ERR(nvme_class)) {
		result = PTR_ERR(nvme_class);
3472
		goto unregister_chrdev;
3473
	}
3474

3475 3476
	result = pci_register_driver(&nvme_driver);
	if (result)
3477
		goto destroy_class;
3478
	return 0;
M
Matthew Wilcox 已提交
3479

3480 3481 3482 3483
 destroy_class:
	class_destroy(nvme_class);
 unregister_chrdev:
	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
3484
 unregister_blkdev:
M
Matthew Wilcox 已提交
3485
	unregister_blkdev(nvme_major, "nvme");
K
Keith Busch 已提交
3486 3487
 kill_workq:
	destroy_workqueue(nvme_workq);
M
Matthew Wilcox 已提交
3488 3489 3490 3491 3492 3493 3494
	return result;
}

static void __exit nvme_exit(void)
{
	pci_unregister_driver(&nvme_driver);
	unregister_blkdev(nvme_major, "nvme");
K
Keith Busch 已提交
3495
	destroy_workqueue(nvme_workq);
3496 3497
	class_destroy(nvme_class);
	__unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
3498
	BUG_ON(nvme_thread && !IS_ERR(nvme_thread));
3499
	_nvme_check_size();
M
Matthew Wilcox 已提交
3500 3501 3502 3503
}

MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
MODULE_LICENSE("GPL");
3504
MODULE_VERSION("1.0");
M
Matthew Wilcox 已提交
3505 3506
module_init(nvme_init);
module_exit(nvme_exit);