pci.c 53.1 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.
 */

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

47 48
#include "nvme.h"

49
#define NVME_Q_DEPTH		1024
J
Jens Axboe 已提交
50
#define NVME_AQ_DEPTH		256
M
Matthew Wilcox 已提交
51 52
#define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
#define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
53 54 55 56 57
		
/*
 * We handle AEN commands ourselves and don't even let the
 * block layer know about them.
 */
58
#define NVME_AQ_BLKMQ_DEPTH	(NVME_AQ_DEPTH - NVME_NR_AERS)
59

60 61 62
static int use_threaded_interrupts;
module_param(use_threaded_interrupts, int, 0);

63 64 65 66
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");

K
Keith Busch 已提交
67
static struct workqueue_struct *nvme_workq;
68

69 70
struct nvme_dev;
struct nvme_queue;
71

72
static int nvme_reset(struct nvme_dev *dev);
J
Jens Axboe 已提交
73
static void nvme_process_cq(struct nvme_queue *nvmeq);
74
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
75

76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
/*
 * Represents an NVM Express device.  Each nvme_dev is a PCI function.
 */
struct nvme_dev {
	struct nvme_queue **queues;
	struct blk_mq_tag_set tagset;
	struct blk_mq_tag_set admin_tagset;
	u32 __iomem *dbs;
	struct device *dev;
	struct dma_pool *prp_page_pool;
	struct dma_pool *prp_small_pool;
	unsigned queue_count;
	unsigned online_queues;
	unsigned max_qid;
	int q_depth;
	u32 db_stride;
	void __iomem *bar;
	struct work_struct reset_work;
94
	struct work_struct remove_work;
95
	struct timer_list watchdog_timer;
96
	struct mutex shutdown_lock;
97 98 99 100 101 102
	bool subsystem;
	void __iomem *cmb;
	dma_addr_t cmb_dma_addr;
	u64 cmb_size;
	u32 cmbsz;
	struct nvme_ctrl ctrl;
K
Keith Busch 已提交
103
	struct completion ioq_wait;
K
Keith Busch 已提交
104
};
105

106 107 108 109 110
static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl)
{
	return container_of(ctrl, struct nvme_dev, ctrl);
}

M
Matthew Wilcox 已提交
111 112 113 114 115 116
/*
 * 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
117
	struct nvme_dev *dev;
118
	char irqname[24];	/* nvme4294967295-65535\0 */
M
Matthew Wilcox 已提交
119 120
	spinlock_t q_lock;
	struct nvme_command *sq_cmds;
121
	struct nvme_command __iomem *sq_cmds_io;
M
Matthew Wilcox 已提交
122
	volatile struct nvme_completion *cqes;
123
	struct blk_mq_tags **tags;
M
Matthew Wilcox 已提交
124 125 126 127
	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
128
	s16 cq_vector;
M
Matthew Wilcox 已提交
129 130
	u16 sq_tail;
	u16 cq_head;
K
Keith Busch 已提交
131
	u16 qid;
132 133
	u8 cq_phase;
	u8 cqe_seen;
M
Matthew Wilcox 已提交
134 135
};

136 137 138
/*
 * The nvme_iod describes the data in an I/O, including the list of PRP
 * entries.  You can't see it in this data structure because C doesn't let
C
Christoph Hellwig 已提交
139
 * me express that.  Use nvme_init_iod to ensure there's enough space
140 141 142
 * allocated to store the PRP list.
 */
struct nvme_iod {
C
Christoph Hellwig 已提交
143 144
	struct nvme_queue *nvmeq;
	int aborted;
145 146 147 148
	int npages;		/* In the PRP list. 0 means small pool in use */
	int nents;		/* Used in scatterlist */
	int length;		/* Of data, in bytes */
	dma_addr_t first_dma;
149
	struct scatterlist meta_sg; /* metadata requires single contiguous buffer */
C
Christoph Hellwig 已提交
150 151
	struct scatterlist *sg;
	struct scatterlist inline_sg[0];
M
Matthew Wilcox 已提交
152 153 154 155 156 157 158 159 160 161 162 163
};

/*
 * 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);
164
	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
K
Keith Busch 已提交
165
	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
M
Matthew Wilcox 已提交
166 167 168 169
	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
170
	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
M
Matthew Wilcox 已提交
171 172
}

173 174 175 176
/*
 * Max size of iod being embedded in the request payload
 */
#define NVME_INT_PAGES		2
177
#define NVME_INT_BYTES(dev)	(NVME_INT_PAGES * (dev)->ctrl.page_size)
178 179 180 181 182 183 184 185

/*
 * 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)
{
186 187
	unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size,
				      dev->ctrl.page_size);
188 189 190
	return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
}

C
Christoph Hellwig 已提交
191 192
static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
		unsigned int size, unsigned int nseg)
193
{
C
Christoph Hellwig 已提交
194 195 196
	return sizeof(__le64 *) * nvme_npages(size, dev) +
			sizeof(struct scatterlist) * nseg;
}
197

C
Christoph Hellwig 已提交
198 199 200 201
static unsigned int nvme_cmd_size(struct nvme_dev *dev)
{
	return sizeof(struct nvme_iod) +
		nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES);
202 203
}

204 205 206 207 208
static int nvmeq_irq(struct nvme_queue *nvmeq)
{
	return pci_irq_vector(to_pci_dev(nvmeq->dev->dev), nvmeq->cq_vector);
}

M
Matias Bjørling 已提交
209 210
static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
				unsigned int hctx_idx)
211
{
M
Matias Bjørling 已提交
212 213 214
	struct nvme_dev *dev = data;
	struct nvme_queue *nvmeq = dev->queues[0];

215 216 217 218
	WARN_ON(hctx_idx != 0);
	WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
	WARN_ON(nvmeq->tags);

M
Matias Bjørling 已提交
219
	hctx->driver_data = nvmeq;
220
	nvmeq->tags = &dev->admin_tagset.tags[0];
M
Matias Bjørling 已提交
221
	return 0;
222 223
}

224 225 226 227 228 229 230
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
231 232 233
static int nvme_admin_init_request(void *data, struct request *req,
				unsigned int hctx_idx, unsigned int rq_idx,
				unsigned int numa_node)
234
{
M
Matias Bjørling 已提交
235
	struct nvme_dev *dev = data;
C
Christoph Hellwig 已提交
236
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
237 238 239
	struct nvme_queue *nvmeq = dev->queues[0];

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
240
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
241
	return 0;
242 243
}

M
Matias Bjørling 已提交
244 245
static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
			  unsigned int hctx_idx)
M
Matthew Wilcox 已提交
246
{
M
Matias Bjørling 已提交
247
	struct nvme_dev *dev = data;
248
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
M
Matias Bjørling 已提交
249

250 251
	if (!nvmeq->tags)
		nvmeq->tags = &dev->tagset.tags[hctx_idx];
M
Matthew Wilcox 已提交
252

253
	WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
M
Matias Bjørling 已提交
254 255
	hctx->driver_data = nvmeq;
	return 0;
M
Matthew Wilcox 已提交
256 257
}

M
Matias Bjørling 已提交
258 259 260
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
261
{
M
Matias Bjørling 已提交
262
	struct nvme_dev *dev = data;
C
Christoph Hellwig 已提交
263
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
264 265 266
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
267
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
268 269 270
	return 0;
}

271 272 273 274 275 276 277
static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
{
	struct nvme_dev *dev = set->driver_data;

	return blk_mq_pci_map_queues(set, to_pci_dev(dev->dev));
}

M
Matthew Wilcox 已提交
278
/**
279
 * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
M
Matthew Wilcox 已提交
280 281 282 283 284
 * @nvmeq: The queue to use
 * @cmd: The command to send
 *
 * Safe to use from interrupt context
 */
285 286
static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
						struct nvme_command *cmd)
M
Matthew Wilcox 已提交
287
{
M
Matias Bjørling 已提交
288 289
	u16 tail = nvmeq->sq_tail;

290 291 292 293 294
	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
295 296
	if (++tail == nvmeq->q_depth)
		tail = 0;
297
	writel(tail, nvmeq->q_db);
M
Matthew Wilcox 已提交
298 299 300
	nvmeq->sq_tail = tail;
}

C
Christoph Hellwig 已提交
301
static __le64 **iod_list(struct request *req)
M
Matthew Wilcox 已提交
302
{
C
Christoph Hellwig 已提交
303 304
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	return (__le64 **)(iod->sg + req->nr_phys_segments);
M
Matthew Wilcox 已提交
305 306
}

M
Ming Lin 已提交
307 308
static int nvme_init_iod(struct request *rq, unsigned size,
		struct nvme_dev *dev)
309
{
C
Christoph Hellwig 已提交
310 311
	struct nvme_iod *iod = blk_mq_rq_to_pdu(rq);
	int nseg = rq->nr_phys_segments;
312

C
Christoph Hellwig 已提交
313 314 315 316 317 318
	if (nseg > NVME_INT_PAGES || size > NVME_INT_BYTES(dev)) {
		iod->sg = kmalloc(nvme_iod_alloc_size(dev, size, nseg), GFP_ATOMIC);
		if (!iod->sg)
			return BLK_MQ_RQ_QUEUE_BUSY;
	} else {
		iod->sg = iod->inline_sg;
319 320
	}

C
Christoph Hellwig 已提交
321 322 323 324
	iod->aborted = 0;
	iod->npages = -1;
	iod->nents = 0;
	iod->length = size;
K
Keith Busch 已提交
325 326 327 328 329

	if (!(rq->cmd_flags & REQ_DONTPREP)) {
		rq->retries = 0;
		rq->cmd_flags |= REQ_DONTPREP;
	}
C
Christoph Hellwig 已提交
330
	return 0;
331 332
}

C
Christoph Hellwig 已提交
333
static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
334
{
C
Christoph Hellwig 已提交
335
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
336
	const int last_prp = dev->ctrl.page_size / 8 - 1;
337
	int i;
C
Christoph Hellwig 已提交
338
	__le64 **list = iod_list(req);
339 340
	dma_addr_t prp_dma = iod->first_dma;

M
Ming Lin 已提交
341
	nvme_cleanup_cmd(req);
M
Ming Lin 已提交
342

343 344 345 346 347 348 349 350
	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;
	}
351

C
Christoph Hellwig 已提交
352 353
	if (iod->sg != iod->inline_sg)
		kfree(iod->sg);
K
Keith Busch 已提交
354 355
}

356
#ifdef CONFIG_BLK_DEV_INTEGRITY
K
Keith Busch 已提交
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
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);
401
	ts = ns->disk->queue->integrity.tuple_size;
K
Keith Busch 已提交
402 403 404 405 406 407 408 409

	for (i = 0; i < nlb; i++, virt++, phys++) {
		pi = (struct t10_pi_tuple *)p;
		dif_swap(phys, virt, pi);
		p += ts;
	}
	kunmap_atomic(pmap);
}
410 411 412 413 414 415 416 417 418 419 420 421 422
#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)
{
}
#endif

C
Christoph Hellwig 已提交
423
static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req,
424
		int total_len)
M
Matthew Wilcox 已提交
425
{
C
Christoph Hellwig 已提交
426
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
427
	struct dma_pool *pool;
428 429
	int length = total_len;
	struct scatterlist *sg = iod->sg;
M
Matthew Wilcox 已提交
430 431
	int dma_len = sg_dma_len(sg);
	u64 dma_addr = sg_dma_address(sg);
432
	u32 page_size = dev->ctrl.page_size;
433
	int offset = dma_addr & (page_size - 1);
434
	__le64 *prp_list;
C
Christoph Hellwig 已提交
435
	__le64 **list = iod_list(req);
436
	dma_addr_t prp_dma;
437
	int nprps, i;
M
Matthew Wilcox 已提交
438

439
	length -= (page_size - offset);
M
Matthew Wilcox 已提交
440
	if (length <= 0)
441
		return true;
M
Matthew Wilcox 已提交
442

443
	dma_len -= (page_size - offset);
M
Matthew Wilcox 已提交
444
	if (dma_len) {
445
		dma_addr += (page_size - offset);
M
Matthew Wilcox 已提交
446 447 448 449 450 451
	} else {
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
	}

452
	if (length <= page_size) {
453
		iod->first_dma = dma_addr;
454
		return true;
455 456
	}

457
	nprps = DIV_ROUND_UP(length, page_size);
458 459
	if (nprps <= (256 / 8)) {
		pool = dev->prp_small_pool;
460
		iod->npages = 0;
461 462
	} else {
		pool = dev->prp_page_pool;
463
		iod->npages = 1;
464 465
	}

466
	prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
467
	if (!prp_list) {
468
		iod->first_dma = dma_addr;
469
		iod->npages = -1;
470
		return false;
471
	}
472 473
	list[0] = prp_list;
	iod->first_dma = prp_dma;
474 475
	i = 0;
	for (;;) {
476
		if (i == page_size >> 3) {
477
			__le64 *old_prp_list = prp_list;
478
			prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
479
			if (!prp_list)
480
				return false;
481
			list[iod->npages++] = prp_list;
482 483 484
			prp_list[0] = old_prp_list[i - 1];
			old_prp_list[i - 1] = cpu_to_le64(prp_dma);
			i = 1;
485 486
		}
		prp_list[i++] = cpu_to_le64(dma_addr);
487 488 489
		dma_len -= page_size;
		dma_addr += page_size;
		length -= page_size;
490 491 492 493 494 495 496 497
		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
498 499
	}

500
	return true;
M
Matthew Wilcox 已提交
501 502
}

C
Christoph Hellwig 已提交
503
static int nvme_map_data(struct nvme_dev *dev, struct request *req,
M
Ming Lin 已提交
504
		unsigned size, struct nvme_command *cmnd)
505
{
C
Christoph Hellwig 已提交
506
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
C
Christoph Hellwig 已提交
507 508 509 510
	struct request_queue *q = req->q;
	enum dma_data_direction dma_dir = rq_data_dir(req) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE;
	int ret = BLK_MQ_RQ_QUEUE_ERROR;
511

C
Christoph Hellwig 已提交
512 513 514 515
	sg_init_table(iod->sg, req->nr_phys_segments);
	iod->nents = blk_rq_map_sg(q, req, iod->sg);
	if (!iod->nents)
		goto out;
516

C
Christoph Hellwig 已提交
517
	ret = BLK_MQ_RQ_QUEUE_BUSY;
518 519
	if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir,
				DMA_ATTR_NO_WARN))
C
Christoph Hellwig 已提交
520
		goto out;
521

M
Ming Lin 已提交
522
	if (!nvme_setup_prps(dev, req, size))
C
Christoph Hellwig 已提交
523
		goto out_unmap;
524

C
Christoph Hellwig 已提交
525 526 527 528
	ret = BLK_MQ_RQ_QUEUE_ERROR;
	if (blk_integrity_rq(req)) {
		if (blk_rq_count_integrity_sg(q, req->bio) != 1)
			goto out_unmap;
529

530 531
		sg_init_table(&iod->meta_sg, 1);
		if (blk_rq_map_integrity_sg(q, req->bio, &iod->meta_sg) != 1)
C
Christoph Hellwig 已提交
532
			goto out_unmap;
533

C
Christoph Hellwig 已提交
534 535
		if (rq_data_dir(req))
			nvme_dif_remap(req, nvme_dif_prep);
536

537
		if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir))
C
Christoph Hellwig 已提交
538
			goto out_unmap;
539
	}
M
Matthew Wilcox 已提交
540

541 542
	cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
	cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma);
C
Christoph Hellwig 已提交
543
	if (blk_integrity_rq(req))
544
		cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
C
Christoph Hellwig 已提交
545
	return BLK_MQ_RQ_QUEUE_OK;
M
Matthew Wilcox 已提交
546

C
Christoph Hellwig 已提交
547 548 549 550
out_unmap:
	dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
out:
	return ret;
M
Matthew Wilcox 已提交
551 552
}

C
Christoph Hellwig 已提交
553
static void nvme_unmap_data(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
554
{
C
Christoph Hellwig 已提交
555
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
556 557 558 559 560 561 562 563
	enum dma_data_direction dma_dir = rq_data_dir(req) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE;

	if (iod->nents) {
		dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
		if (blk_integrity_rq(req)) {
			if (!rq_data_dir(req))
				nvme_dif_remap(req, nvme_dif_complete);
564
			dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir);
K
Keith Busch 已提交
565
		}
566
	}
K
Keith Busch 已提交
567

C
Christoph Hellwig 已提交
568
	nvme_free_iod(dev, req);
569
}
M
Matthew Wilcox 已提交
570

571 572 573
/*
 * NOTE: ns is NULL when called on the admin queue.
 */
M
Matias Bjørling 已提交
574 575
static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
			 const struct blk_mq_queue_data *bd)
576
{
M
Matias Bjørling 已提交
577 578
	struct nvme_ns *ns = hctx->queue->queuedata;
	struct nvme_queue *nvmeq = hctx->driver_data;
579
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
580
	struct request *req = bd->rq;
C
Christoph Hellwig 已提交
581
	struct nvme_command cmnd;
M
Ming Lin 已提交
582
	unsigned map_len;
C
Christoph Hellwig 已提交
583
	int ret = BLK_MQ_RQ_QUEUE_OK;
584

K
Keith Busch 已提交
585 586 587 588 589
	/*
	 * 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.
	 */
590
	if (ns && ns->ms && !blk_integrity_rq(req)) {
591 592
		if (!(ns->pi_type && ns->ms == 8) &&
					req->cmd_type != REQ_TYPE_DRV_PRIV) {
593
			blk_mq_end_request(req, -EFAULT);
K
Keith Busch 已提交
594 595 596 597
			return BLK_MQ_RQ_QUEUE_OK;
		}
	}

M
Ming Lin 已提交
598 599
	map_len = nvme_map_len(req);
	ret = nvme_init_iod(req, map_len, dev);
C
Christoph Hellwig 已提交
600 601
	if (ret)
		return ret;
M
Matias Bjørling 已提交
602

M
Ming Lin 已提交
603
	ret = nvme_setup_cmd(ns, req, &cmnd);
M
Ming Lin 已提交
604 605
	if (ret)
		goto out;
M
Matias Bjørling 已提交
606

M
Ming Lin 已提交
607 608
	if (req->nr_phys_segments)
		ret = nvme_map_data(dev, req, map_len, &cmnd);
M
Matias Bjørling 已提交
609

C
Christoph Hellwig 已提交
610 611
	if (ret)
		goto out;
M
Matias Bjørling 已提交
612

C
Christoph Hellwig 已提交
613
	cmnd.common.command_id = req->tag;
614
	blk_mq_start_request(req);
M
Matias Bjørling 已提交
615

C
Christoph Hellwig 已提交
616
	spin_lock_irq(&nvmeq->q_lock);
617
	if (unlikely(nvmeq->cq_vector < 0)) {
618 619 620 621
		if (ns && !test_bit(NVME_NS_DEAD, &ns->flags))
			ret = BLK_MQ_RQ_QUEUE_BUSY;
		else
			ret = BLK_MQ_RQ_QUEUE_ERROR;
622 623 624
		spin_unlock_irq(&nvmeq->q_lock);
		goto out;
	}
C
Christoph Hellwig 已提交
625
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matias Bjørling 已提交
626 627 628
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
	return BLK_MQ_RQ_QUEUE_OK;
C
Christoph Hellwig 已提交
629
out:
C
Christoph Hellwig 已提交
630
	nvme_free_iod(dev, req);
C
Christoph Hellwig 已提交
631
	return ret;
M
Matthew Wilcox 已提交
632
}
K
Keith Busch 已提交
633

634 635
static void nvme_complete_rq(struct request *req)
{
C
Christoph Hellwig 已提交
636 637
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_dev *dev = iod->nvmeq->dev;
638
	int error = 0;
K
Keith Busch 已提交
639

C
Christoph Hellwig 已提交
640
	nvme_unmap_data(dev, req);
K
Keith Busch 已提交
641

642 643
	if (unlikely(req->errors)) {
		if (nvme_req_needs_retry(req, req->errors)) {
K
Keith Busch 已提交
644
			req->retries++;
645 646
			nvme_requeue_req(req);
			return;
K
Keith Busch 已提交
647
		}
648

649 650 651 652 653
		if (req->cmd_type == REQ_TYPE_DRV_PRIV)
			error = req->errors;
		else
			error = nvme_error_status(req->errors);
	}
M
Matias Bjørling 已提交
654

C
Christoph Hellwig 已提交
655
	if (unlikely(iod->aborted)) {
656
		dev_warn(dev->ctrl.device,
657 658 659
			"completing aborted command with status: %04x\n",
			req->errors);
	}
M
Matias Bjørling 已提交
660

661
	blk_mq_end_request(req, error);
M
Matthew Wilcox 已提交
662 663
}

664 665 666 667 668 669 670
/* We read the CQE phase first to check if the rest of the entry is valid */
static inline bool nvme_cqe_valid(struct nvme_queue *nvmeq, u16 head,
		u16 phase)
{
	return (le16_to_cpu(nvmeq->cqes[head].status) & 1) == phase;
}

J
Jens Axboe 已提交
671
static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag)
M
Matthew Wilcox 已提交
672
{
M
Matthew Wilcox 已提交
673
	u16 head, phase;
M
Matthew Wilcox 已提交
674 675

	head = nvmeq->cq_head;
M
Matthew Wilcox 已提交
676
	phase = nvmeq->cq_phase;
M
Matthew Wilcox 已提交
677

678
	while (nvme_cqe_valid(nvmeq, head, phase)) {
M
Matthew Wilcox 已提交
679
		struct nvme_completion cqe = nvmeq->cqes[head];
680
		struct request *req;
681

M
Matthew Wilcox 已提交
682 683
		if (++head == nvmeq->q_depth) {
			head = 0;
M
Matthew Wilcox 已提交
684
			phase = !phase;
M
Matthew Wilcox 已提交
685
		}
686

J
Jens Axboe 已提交
687 688
		if (tag && *tag == cqe.command_id)
			*tag = -1;
689

690
		if (unlikely(cqe.command_id >= nvmeq->q_depth)) {
691
			dev_warn(nvmeq->dev->ctrl.device,
692 693 694 695 696
				"invalid id %d completed on queue %d\n",
				cqe.command_id, le16_to_cpu(cqe.sq_id));
			continue;
		}

697 698 699 700 701 702 703 704
		/*
		 * AEN requests are special as they don't time out and can
		 * survive any kind of queue freeze and often don't respond to
		 * aborts.  We don't even bother to allocate a struct request
		 * for them but rather special case them here.
		 */
		if (unlikely(nvmeq->qid == 0 &&
				cqe.command_id >= NVME_AQ_BLKMQ_DEPTH)) {
705
			nvme_complete_async_event(&nvmeq->dev->ctrl, &cqe);
706 707 708
			continue;
		}

709
		req = blk_mq_tag_to_rq(*nvmeq->tags, cqe.command_id);
710 711
		if (req->cmd_type == REQ_TYPE_DRV_PRIV && req->special)
			memcpy(req->special, &cqe, sizeof(cqe));
712
		blk_mq_complete_request(req, le16_to_cpu(cqe.status) >> 1);
713

M
Matthew Wilcox 已提交
714 715 716 717 718 719 720 721
	}

	/* 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
722
	if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
J
Jens Axboe 已提交
723
		return;
M
Matthew Wilcox 已提交
724

725 726
	if (likely(nvmeq->cq_vector >= 0))
		writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
M
Matthew Wilcox 已提交
727
	nvmeq->cq_head = head;
M
Matthew Wilcox 已提交
728
	nvmeq->cq_phase = phase;
M
Matthew Wilcox 已提交
729

730
	nvmeq->cqe_seen = 1;
J
Jens Axboe 已提交
731 732 733 734 735
}

static void nvme_process_cq(struct nvme_queue *nvmeq)
{
	__nvme_process_cq(nvmeq, NULL);
M
Matthew Wilcox 已提交
736 737 738
}

static irqreturn_t nvme_irq(int irq, void *data)
739 740 741 742
{
	irqreturn_t result;
	struct nvme_queue *nvmeq = data;
	spin_lock(&nvmeq->q_lock);
743 744 745
	nvme_process_cq(nvmeq);
	result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
	nvmeq->cqe_seen = 0;
746 747 748 749 750 751 752
	spin_unlock(&nvmeq->q_lock);
	return result;
}

static irqreturn_t nvme_irq_check(int irq, void *data)
{
	struct nvme_queue *nvmeq = data;
753 754 755
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return IRQ_WAKE_THREAD;
	return IRQ_NONE;
756 757
}

J
Jens Axboe 已提交
758 759 760 761
static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
{
	struct nvme_queue *nvmeq = hctx->driver_data;

762
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) {
J
Jens Axboe 已提交
763 764 765 766 767 768 769 770 771 772 773
		spin_lock_irq(&nvmeq->q_lock);
		__nvme_process_cq(nvmeq, &tag);
		spin_unlock_irq(&nvmeq->q_lock);

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

	return 0;
}

774
static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx)
M
Matthew Wilcox 已提交
775
{
776
	struct nvme_dev *dev = to_nvme_dev(ctrl);
777
	struct nvme_queue *nvmeq = dev->queues[0];
M
Matias Bjørling 已提交
778
	struct nvme_command c;
M
Matthew Wilcox 已提交
779

M
Matias Bjørling 已提交
780 781
	memset(&c, 0, sizeof(c));
	c.common.opcode = nvme_admin_async_event;
782
	c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx;
783

784
	spin_lock_irq(&nvmeq->q_lock);
785
	__nvme_submit_cmd(nvmeq, &c);
786
	spin_unlock_irq(&nvmeq->q_lock);
787 788
}

M
Matthew Wilcox 已提交
789
static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
790
{
M
Matthew Wilcox 已提交
791 792 793 794 795 796
	struct nvme_command c;

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

797
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
798 799 800 801 802 803 804 805
}

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;

806 807 808 809
	/*
	 * 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
810 811 812 813 814 815 816 817
	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);

818
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
819 820 821 822 823 824 825 826
}

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;

827 828 829 830
	/*
	 * 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
831 832 833 834 835 836 837 838
	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);

839
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
840 841 842 843 844 845 846 847 848 849 850 851
}

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);
}

852
static void abort_endio(struct request *req, int error)
853
{
C
Christoph Hellwig 已提交
854 855
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
856
	u16 status = req->errors;
857

858
	dev_warn(nvmeq->dev->ctrl.device, "Abort status: 0x%x", status);
859 860
	atomic_inc(&nvmeq->dev->ctrl.abort_limit);
	blk_mq_free_request(req);
861 862
}

863
static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
K
Keith Busch 已提交
864
{
C
Christoph Hellwig 已提交
865 866
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
K
Keith Busch 已提交
867
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
868 869
	struct request *abort_req;
	struct nvme_command cmd;
K
Keith Busch 已提交
870

871
	/*
872 873 874 875 876
	 * Shutdown immediately if controller times out while starting. The
	 * reset work will see the pci device disabled when it gets the forced
	 * cancellation error. All outstanding requests are completed on
	 * shutdown, so we return BLK_EH_HANDLED.
	 */
877
	if (dev->ctrl.state == NVME_CTRL_RESETTING) {
878
		dev_warn(dev->ctrl.device,
879 880
			 "I/O %d QID %d timeout, disable controller\n",
			 req->tag, nvmeq->qid);
881
		nvme_dev_disable(dev, false);
882 883
		req->errors = NVME_SC_CANCELLED;
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
884 885
	}

886 887 888 889
	/*
 	 * Shutdown the controller immediately and schedule a 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.
890
	 */
C
Christoph Hellwig 已提交
891
	if (!nvmeq->qid || iod->aborted) {
892
		dev_warn(dev->ctrl.device,
893 894
			 "I/O %d QID %d timeout, reset controller\n",
			 req->tag, nvmeq->qid);
895
		nvme_dev_disable(dev, false);
896
		queue_work(nvme_workq, &dev->reset_work);
K
Keith Busch 已提交
897

898 899 900 901 902 903
		/*
		 * Mark the request as handled, since the inline shutdown
		 * forces all outstanding requests to complete.
		 */
		req->errors = NVME_SC_CANCELLED;
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
904 905
	}

C
Christoph Hellwig 已提交
906
	iod->aborted = 1;
K
Keith Busch 已提交
907

908
	if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) {
909
		atomic_inc(&dev->ctrl.abort_limit);
910
		return BLK_EH_RESET_TIMER;
911
	}
M
Matias Bjørling 已提交
912

K
Keith Busch 已提交
913 914
	memset(&cmd, 0, sizeof(cmd));
	cmd.abort.opcode = nvme_admin_abort_cmd;
M
Matias Bjørling 已提交
915
	cmd.abort.cid = req->tag;
K
Keith Busch 已提交
916 917
	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);

918 919 920
	dev_warn(nvmeq->dev->ctrl.device,
		"I/O %d QID %d timeout, aborting\n",
		 req->tag, nvmeq->qid);
921 922

	abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
923
			BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
924 925 926 927 928 929 930 931
	if (IS_ERR(abort_req)) {
		atomic_inc(&dev->ctrl.abort_limit);
		return BLK_EH_RESET_TIMER;
	}

	abort_req->timeout = ADMIN_TIMEOUT;
	abort_req->end_io_data = NULL;
	blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio);
K
Keith Busch 已提交
932

933 934 935 936 937 938
	/*
	 * 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;
K
Keith Busch 已提交
939 940
}

M
Matias Bjørling 已提交
941 942
static void nvme_free_queue(struct nvme_queue *nvmeq)
{
943 944
	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
945 946
	if (nvmeq->sq_cmds)
		dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
947 948 949 950
					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
	kfree(nvmeq);
}

951
static void nvme_free_queues(struct nvme_dev *dev, int lowest)
952 953 954
{
	int i;

955
	for (i = dev->queue_count - 1; i >= lowest; i--) {
M
Matias Bjørling 已提交
956
		struct nvme_queue *nvmeq = dev->queues[i];
957
		dev->queue_count--;
M
Matias Bjørling 已提交
958
		dev->queues[i] = NULL;
K
Keith Busch 已提交
959
		nvme_free_queue(nvmeq);
960
	}
961 962
}

K
Keith Busch 已提交
963 964 965 966 967
/**
 * 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
968
{
K
Keith Busch 已提交
969
	int vector;
M
Matthew Wilcox 已提交
970

971
	spin_lock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
972 973 974 975
	if (nvmeq->cq_vector == -1) {
		spin_unlock_irq(&nvmeq->q_lock);
		return 1;
	}
976
	vector = nvmeq_irq(nvmeq);
K
Keith Busch 已提交
977
	nvmeq->dev->online_queues--;
K
Keith Busch 已提交
978
	nvmeq->cq_vector = -1;
979 980
	spin_unlock_irq(&nvmeq->q_lock);

981
	if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
982
		blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q);
983

M
Matthew Wilcox 已提交
984
	free_irq(vector, nvmeq);
M
Matthew Wilcox 已提交
985

K
Keith Busch 已提交
986 987
	return 0;
}
M
Matthew Wilcox 已提交
988

989
static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
K
Keith Busch 已提交
990
{
991
	struct nvme_queue *nvmeq = dev->queues[0];
K
Keith Busch 已提交
992 993 994 995 996 997

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

998 999 1000 1001 1002
	if (shutdown)
		nvme_shutdown_ctrl(&dev->ctrl);
	else
		nvme_disable_ctrl(&dev->ctrl, lo_hi_readq(
						dev->bar + NVME_REG_CAP));
1003 1004 1005 1006

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1007 1008
}

1009 1010 1011 1012
static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
				int entry_size)
{
	int q_depth = dev->q_depth;
1013 1014
	unsigned q_size_aligned = roundup(q_depth * entry_size,
					  dev->ctrl.page_size);
1015 1016

	if (q_size_aligned * nr_io_queues > dev->cmb_size) {
1017
		u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1018
		mem_per_q = round_down(mem_per_q, dev->ctrl.page_size);
1019
		q_depth = div_u64(mem_per_q, entry_size);
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036

		/*
		 * 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)) {
1037 1038
		unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth),
						      dev->ctrl.page_size);
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
		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
1051
static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
K
Keith Busch 已提交
1052
							int depth)
M
Matthew Wilcox 已提交
1053
{
M
Matias Bjørling 已提交
1054
	struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
M
Matthew Wilcox 已提交
1055 1056 1057
	if (!nvmeq)
		return NULL;

1058
	nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
J
Joe Perches 已提交
1059
					  &nvmeq->cq_dma_addr, GFP_KERNEL);
M
Matthew Wilcox 已提交
1060 1061 1062
	if (!nvmeq->cqes)
		goto free_nvmeq;

1063
	if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
M
Matthew Wilcox 已提交
1064 1065
		goto free_cqdma;

1066
	nvmeq->q_dmadev = dev->dev;
M
Matthew Wilcox 已提交
1067
	nvmeq->dev = dev;
1068
	snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
1069
			dev->ctrl.instance, qid);
M
Matthew Wilcox 已提交
1070 1071
	spin_lock_init(&nvmeq->q_lock);
	nvmeq->cq_head = 0;
M
Matthew Wilcox 已提交
1072
	nvmeq->cq_phase = 1;
1073
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
M
Matthew Wilcox 已提交
1074
	nvmeq->q_depth = depth;
K
Keith Busch 已提交
1075
	nvmeq->qid = qid;
1076
	nvmeq->cq_vector = -1;
M
Matias Bjørling 已提交
1077
	dev->queues[qid] = nvmeq;
1078 1079
	dev->queue_count++;

M
Matthew Wilcox 已提交
1080 1081 1082
	return nvmeq;

 free_cqdma:
1083
	dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
M
Matthew Wilcox 已提交
1084 1085 1086 1087 1088 1089
							nvmeq->cq_dma_addr);
 free_nvmeq:
	kfree(nvmeq);
	return NULL;
}

1090
static int queue_request_irq(struct nvme_queue *nvmeq)
1091
{
1092
	if (use_threaded_interrupts)
1093 1094 1095 1096 1097
		return request_threaded_irq(nvmeq_irq(nvmeq), nvme_irq_check,
				nvme_irq, IRQF_SHARED, nvmeq->irqname, nvmeq);
	else
		return request_irq(nvmeq_irq(nvmeq), nvme_irq, IRQF_SHARED,
				nvmeq->irqname, nvmeq);
1098 1099
}

1100
static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
M
Matthew Wilcox 已提交
1101
{
1102
	struct nvme_dev *dev = nvmeq->dev;
M
Matthew Wilcox 已提交
1103

1104
	spin_lock_irq(&nvmeq->q_lock);
1105 1106 1107
	nvmeq->sq_tail = 0;
	nvmeq->cq_head = 0;
	nvmeq->cq_phase = 1;
1108
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
1109
	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
K
Keith Busch 已提交
1110
	dev->online_queues++;
1111
	spin_unlock_irq(&nvmeq->q_lock);
1112 1113 1114 1115 1116 1117
}

static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
{
	struct nvme_dev *dev = nvmeq->dev;
	int result;
1118

K
Keith Busch 已提交
1119
	nvmeq->cq_vector = qid - 1;
M
Matthew Wilcox 已提交
1120 1121
	result = adapter_alloc_cq(dev, qid, nvmeq);
	if (result < 0)
1122
		return result;
M
Matthew Wilcox 已提交
1123 1124 1125 1126 1127

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

1128
	result = queue_request_irq(nvmeq);
M
Matthew Wilcox 已提交
1129 1130 1131
	if (result < 0)
		goto release_sq;

1132 1133
	nvme_init_queue(nvmeq, qid);
	return result;
M
Matthew Wilcox 已提交
1134 1135 1136 1137 1138

 release_sq:
	adapter_delete_sq(dev, qid);
 release_cq:
	adapter_delete_cq(dev, qid);
1139
	return result;
M
Matthew Wilcox 已提交
1140 1141
}

M
Matias Bjørling 已提交
1142
static struct blk_mq_ops nvme_mq_admin_ops = {
1143
	.queue_rq	= nvme_queue_rq,
1144
	.complete	= nvme_complete_rq,
M
Matias Bjørling 已提交
1145
	.init_hctx	= nvme_admin_init_hctx,
1146
	.exit_hctx      = nvme_admin_exit_hctx,
M
Matias Bjørling 已提交
1147 1148 1149 1150 1151 1152
	.init_request	= nvme_admin_init_request,
	.timeout	= nvme_timeout,
};

static struct blk_mq_ops nvme_mq_ops = {
	.queue_rq	= nvme_queue_rq,
1153
	.complete	= nvme_complete_rq,
M
Matias Bjørling 已提交
1154 1155
	.init_hctx	= nvme_init_hctx,
	.init_request	= nvme_init_request,
1156
	.map_queues	= nvme_pci_map_queues,
M
Matias Bjørling 已提交
1157
	.timeout	= nvme_timeout,
J
Jens Axboe 已提交
1158
	.poll		= nvme_poll,
M
Matias Bjørling 已提交
1159 1160
};

1161 1162
static void nvme_dev_remove_admin(struct nvme_dev *dev)
{
1163
	if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) {
1164 1165 1166 1167 1168 1169
		/*
		 * If the controller was reset during removal, it's possible
		 * user requests may be waiting on a stopped queue. Start the
		 * queue to flush these to completion.
		 */
		blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true);
1170
		blk_cleanup_queue(dev->ctrl.admin_q);
1171 1172 1173 1174
		blk_mq_free_tag_set(&dev->admin_tagset);
	}
}

M
Matias Bjørling 已提交
1175 1176
static int nvme_alloc_admin_tags(struct nvme_dev *dev)
{
1177
	if (!dev->ctrl.admin_q) {
M
Matias Bjørling 已提交
1178 1179
		dev->admin_tagset.ops = &nvme_mq_admin_ops;
		dev->admin_tagset.nr_hw_queues = 1;
K
Keith Busch 已提交
1180 1181 1182 1183 1184 1185

		/*
		 * Subtract one to leave an empty queue entry for 'Full Queue'
		 * condition. See NVM-Express 1.2 specification, section 4.1.2.
		 */
		dev->admin_tagset.queue_depth = NVME_AQ_BLKMQ_DEPTH - 1;
M
Matias Bjørling 已提交
1186
		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
1187
		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
1188
		dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
M
Matias Bjørling 已提交
1189 1190 1191 1192 1193
		dev->admin_tagset.driver_data = dev;

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

1194 1195
		dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset);
		if (IS_ERR(dev->ctrl.admin_q)) {
M
Matias Bjørling 已提交
1196 1197 1198
			blk_mq_free_tag_set(&dev->admin_tagset);
			return -ENOMEM;
		}
1199
		if (!blk_get_queue(dev->ctrl.admin_q)) {
1200
			nvme_dev_remove_admin(dev);
1201
			dev->ctrl.admin_q = NULL;
1202 1203
			return -ENODEV;
		}
K
Keith Busch 已提交
1204
	} else
1205
		blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true);
M
Matias Bjørling 已提交
1206 1207 1208 1209

	return 0;
}

1210
static int nvme_configure_admin_queue(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1211
{
1212
	int result;
M
Matthew Wilcox 已提交
1213
	u32 aqa;
1214
	u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
M
Matthew Wilcox 已提交
1215 1216
	struct nvme_queue *nvmeq;

1217
	dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1) ?
1218 1219
						NVME_CAP_NSSRC(cap) : 0;

1220 1221 1222
	if (dev->subsystem &&
	    (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO))
		writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS);
1223

1224
	result = nvme_disable_ctrl(&dev->ctrl, cap);
1225 1226
	if (result < 0)
		return result;
M
Matthew Wilcox 已提交
1227

M
Matias Bjørling 已提交
1228
	nvmeq = dev->queues[0];
1229
	if (!nvmeq) {
K
Keith Busch 已提交
1230
		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
1231 1232 1233
		if (!nvmeq)
			return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1234 1235 1236 1237

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

1238 1239 1240
	writel(aqa, dev->bar + NVME_REG_AQA);
	lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ);
	lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ);
M
Matthew Wilcox 已提交
1241

1242
	result = nvme_enable_ctrl(&dev->ctrl, cap);
1243
	if (result)
M
Matias Bjørling 已提交
1244 1245
		goto free_nvmeq;

K
Keith Busch 已提交
1246
	nvmeq->cq_vector = 0;
1247
	result = queue_request_irq(nvmeq);
1248 1249
	if (result) {
		nvmeq->cq_vector = -1;
K
Keith Busch 已提交
1250
		goto free_nvmeq;
1251
	}
1252

M
Matthew Wilcox 已提交
1253
	return result;
M
Matias Bjørling 已提交
1254 1255 1256 1257

 free_nvmeq:
	nvme_free_queues(dev, 0);
	return result;
M
Matthew Wilcox 已提交
1258 1259
}

1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286
static bool nvme_should_reset(struct nvme_dev *dev, u32 csts)
{

	/* If true, indicates loss of adapter communication, possibly by a
	 * NVMe Subsystem reset.
	 */
	bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO);

	/* If there is a reset ongoing, we shouldn't reset again. */
	if (work_busy(&dev->reset_work))
		return false;

	/* We shouldn't reset unless the controller is on fatal error state
	 * _or_ if we lost the communication with it.
	 */
	if (!(csts & NVME_CSTS_CFS) && !nssro)
		return false;

	/* If PCI error recovery process is happening, we cannot reset or
	 * the recovery mechanism will surely fail.
	 */
	if (pci_channel_offline(to_pci_dev(dev->dev)))
		return false;

	return true;
}

1287
static void nvme_watchdog_timer(unsigned long data)
1288
{
1289 1290
	struct nvme_dev *dev = (struct nvme_dev *)data;
	u32 csts = readl(dev->bar + NVME_REG_CSTS);
1291

1292 1293 1294
	/* Skip controllers under certain specific conditions. */
	if (nvme_should_reset(dev, csts)) {
		if (queue_work(nvme_workq, &dev->reset_work))
1295 1296 1297 1298
			dev_warn(dev->dev,
				"Failed status: 0x%x, reset controller.\n",
				csts);
		return;
1299
	}
1300 1301

	mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
1302 1303
}

1304
static int nvme_create_io_queues(struct nvme_dev *dev)
K
Keith Busch 已提交
1305
{
1306
	unsigned i, max;
1307
	int ret = 0;
K
Keith Busch 已提交
1308

1309 1310 1311
	for (i = dev->queue_count; i <= dev->max_qid; i++) {
		if (!nvme_alloc_queue(dev, i, dev->q_depth)) {
			ret = -ENOMEM;
K
Keith Busch 已提交
1312
			break;
1313 1314
		}
	}
K
Keith Busch 已提交
1315

1316 1317
	max = min(dev->max_qid, dev->queue_count - 1);
	for (i = dev->online_queues; i <= max; i++) {
1318 1319
		ret = nvme_create_queue(dev->queues[i], i);
		if (ret) {
1320
			nvme_free_queues(dev, i);
K
Keith Busch 已提交
1321
			break;
1322
		}
M
Matthew Wilcox 已提交
1323
	}
1324 1325 1326 1327 1328 1329 1330 1331

	/*
	 * Ignore failing Create SQ/CQ commands, we can continue with less
	 * than the desired aount 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.
	 */
	return ret >= 0 ? 0 : ret;
M
Matthew Wilcox 已提交
1332 1333
}

1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345
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;

1346
	dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ);
1347 1348 1349
	if (!(NVME_CMB_SZ(dev->cmbsz)))
		return NULL;

1350
	cmbloc = readl(dev->bar + NVME_REG_CMBLOC);
1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385

	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
1386 1387
static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
{
1388
	return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
K
Keith Busch 已提交
1389 1390
}

1391
static int nvme_setup_io_queues(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1392
{
M
Matias Bjørling 已提交
1393
	struct nvme_queue *adminq = dev->queues[0];
1394
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1395
	int result, nr_io_queues, size;
M
Matthew Wilcox 已提交
1396

1397
	nr_io_queues = num_online_cpus();
C
Christoph Hellwig 已提交
1398 1399
	result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
	if (result < 0)
M
Matthew Wilcox 已提交
1400
		return result;
C
Christoph Hellwig 已提交
1401

1402
	if (nr_io_queues == 0)
1403
		return 0;
M
Matthew Wilcox 已提交
1404

1405 1406 1407 1408 1409 1410 1411 1412 1413
	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
1414 1415
	size = db_bar_size(dev, nr_io_queues);
	if (size > 8192) {
1416
		iounmap(dev->bar);
K
Keith Busch 已提交
1417 1418 1419 1420 1421 1422 1423 1424
		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);
1425
		dev->dbs = dev->bar + 4096;
1426
		adminq->q_db = dev->dbs;
1427 1428
	}

K
Keith Busch 已提交
1429
	/* Deregister the admin queue's interrupt */
1430
	free_irq(pci_irq_vector(pdev, 0), adminq);
K
Keith Busch 已提交
1431

1432 1433 1434 1435
	/*
	 * If we enable msix early due to not intx, disable it again before
	 * setting up the full range we need.
	 */
1436 1437 1438 1439 1440 1441
	pci_free_irq_vectors(pdev);
	nr_io_queues = pci_alloc_irq_vectors(pdev, 1, nr_io_queues,
			PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY);
	if (nr_io_queues <= 0)
		return -EIO;
	dev->max_qid = nr_io_queues;
R
Ramachandra Rao Gajula 已提交
1442

1443 1444 1445 1446 1447 1448 1449
	/*
	 * 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.
	 */

1450
	result = queue_request_irq(adminq);
1451 1452
	if (result) {
		adminq->cq_vector = -1;
1453
		goto free_queues;
1454
	}
1455
	return nvme_create_io_queues(dev);
M
Matthew Wilcox 已提交
1456

1457
 free_queues:
1458
	nvme_free_queues(dev, 1);
1459
	return result;
M
Matthew Wilcox 已提交
1460 1461
}

K
Keith Busch 已提交
1462
static void nvme_del_queue_end(struct request *req, int error)
K
Keith Busch 已提交
1463
{
K
Keith Busch 已提交
1464
	struct nvme_queue *nvmeq = req->end_io_data;
1465

K
Keith Busch 已提交
1466 1467
	blk_mq_free_request(req);
	complete(&nvmeq->dev->ioq_wait);
K
Keith Busch 已提交
1468 1469
}

K
Keith Busch 已提交
1470
static void nvme_del_cq_end(struct request *req, int error)
K
Keith Busch 已提交
1471
{
K
Keith Busch 已提交
1472
	struct nvme_queue *nvmeq = req->end_io_data;
K
Keith Busch 已提交
1473

K
Keith Busch 已提交
1474 1475 1476
	if (!error) {
		unsigned long flags;

1477 1478 1479 1480 1481 1482 1483
		/*
		 * We might be called with the AQ q_lock held
		 * and the I/O queue q_lock should always
		 * nest inside the AQ one.
		 */
		spin_lock_irqsave_nested(&nvmeq->q_lock, flags,
					SINGLE_DEPTH_NESTING);
K
Keith Busch 已提交
1484 1485
		nvme_process_cq(nvmeq);
		spin_unlock_irqrestore(&nvmeq->q_lock, flags);
K
Keith Busch 已提交
1486
	}
K
Keith Busch 已提交
1487 1488

	nvme_del_queue_end(req, error);
K
Keith Busch 已提交
1489 1490
}

K
Keith Busch 已提交
1491
static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
1492
{
K
Keith Busch 已提交
1493 1494 1495
	struct request_queue *q = nvmeq->dev->ctrl.admin_q;
	struct request *req;
	struct nvme_command cmd;
1496

K
Keith Busch 已提交
1497 1498 1499
	memset(&cmd, 0, sizeof(cmd));
	cmd.delete_queue.opcode = opcode;
	cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
1500

1501
	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
K
Keith Busch 已提交
1502 1503
	if (IS_ERR(req))
		return PTR_ERR(req);
1504

K
Keith Busch 已提交
1505 1506 1507 1508 1509 1510 1511
	req->timeout = ADMIN_TIMEOUT;
	req->end_io_data = nvmeq;

	blk_execute_rq_nowait(q, NULL, req, false,
			opcode == nvme_admin_delete_cq ?
				nvme_del_cq_end : nvme_del_queue_end);
	return 0;
1512 1513
}

K
Keith Busch 已提交
1514
static void nvme_disable_io_queues(struct nvme_dev *dev)
K
Keith Busch 已提交
1515
{
K
Keith Busch 已提交
1516
	int pass, queues = dev->online_queues - 1;
K
Keith Busch 已提交
1517 1518
	unsigned long timeout;
	u8 opcode = nvme_admin_delete_sq;
K
Keith Busch 已提交
1519

K
Keith Busch 已提交
1520
	for (pass = 0; pass < 2; pass++) {
K
Keith Busch 已提交
1521
		int sent = 0, i = queues;
K
Keith Busch 已提交
1522 1523 1524 1525

		reinit_completion(&dev->ioq_wait);
 retry:
		timeout = ADMIN_TIMEOUT;
1526 1527
		for (; i > 0; i--, sent++)
			if (nvme_delete_queue(dev->queues[i], opcode))
K
Keith Busch 已提交
1528
				break;
1529

K
Keith Busch 已提交
1530 1531 1532 1533 1534 1535 1536 1537 1538
		while (sent--) {
			timeout = wait_for_completion_io_timeout(&dev->ioq_wait, timeout);
			if (timeout == 0)
				return;
			if (i)
				goto retry;
		}
		opcode = nvme_admin_delete_cq;
	}
K
Keith Busch 已提交
1539 1540
}

1541 1542 1543 1544 1545 1546
/*
 * 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.
 */
1547
static int nvme_dev_add(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1548
{
1549
	if (!dev->ctrl.tagset) {
1550 1551 1552 1553 1554
		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
1555
				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
1556 1557 1558
		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
1559

1560 1561
		if (blk_mq_alloc_tag_set(&dev->tagset))
			return 0;
1562
		dev->ctrl.tagset = &dev->tagset;
1563 1564 1565 1566 1567
	} else {
		blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1);

		/* Free previously allocated queues that are no longer usable */
		nvme_free_queues(dev, dev->online_queues);
1568
	}
1569

K
Keith Busch 已提交
1570
	return 0;
M
Matthew Wilcox 已提交
1571 1572
}

1573
static int nvme_pci_enable(struct nvme_dev *dev)
1574
{
K
Keith Busch 已提交
1575
	u64 cap;
1576
	int result = -ENOMEM;
1577
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1578 1579 1580 1581 1582 1583

	if (pci_enable_device_mem(pdev))
		return result;

	pci_set_master(pdev);

1584 1585
	if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
	    dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
1586
		goto disable;
1587

1588
	if (readl(dev->bar + NVME_REG_CSTS) == -1) {
K
Keith Busch 已提交
1589
		result = -ENODEV;
1590
		goto disable;
K
Keith Busch 已提交
1591
	}
1592 1593

	/*
1594 1595 1596
	 * Some devices and/or platforms don't advertise or work with INTx
	 * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll
	 * adjust this later.
1597
	 */
1598 1599 1600
	result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
	if (result < 0)
		return result;
1601

1602 1603
	cap = lo_hi_readq(dev->bar + NVME_REG_CAP);

K
Keith Busch 已提交
1604 1605
	dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
	dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
1606
	dev->dbs = dev->bar + 4096;
1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618

	/*
	 * 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);
	}

1619
	if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2))
1620
		dev->cmb = nvme_map_cmb(dev);
1621

K
Keith Busch 已提交
1622 1623
	pci_enable_pcie_error_reporting(pdev);
	pci_save_state(pdev);
1624 1625 1626 1627 1628 1629 1630 1631
	return 0;

 disable:
	pci_disable_device(pdev);
	return result;
}

static void nvme_dev_unmap(struct nvme_dev *dev)
1632 1633 1634
{
	if (dev->bar)
		iounmap(dev->bar);
1635
	pci_release_mem_regions(to_pci_dev(dev->dev));
1636 1637 1638
}

static void nvme_pci_disable(struct nvme_dev *dev)
1639
{
1640 1641
	struct pci_dev *pdev = to_pci_dev(dev->dev);

1642
	pci_free_irq_vectors(pdev);
1643

K
Keith Busch 已提交
1644 1645
	if (pci_is_enabled(pdev)) {
		pci_disable_pcie_error_reporting(pdev);
1646
		pci_disable_device(pdev);
K
Keith Busch 已提交
1647 1648 1649
	}
}

1650
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
M
Matthew Wilcox 已提交
1651
{
1652
	int i;
1653
	u32 csts = -1;
1654

1655
	del_timer_sync(&dev->watchdog_timer);
1656

1657
	mutex_lock(&dev->shutdown_lock);
1658
	if (pci_is_enabled(to_pci_dev(dev->dev))) {
1659
		nvme_stop_queues(&dev->ctrl);
1660
		csts = readl(dev->bar + NVME_REG_CSTS);
1661
	}
1662 1663 1664 1665

	for (i = dev->queue_count - 1; i > 0; i--)
		nvme_suspend_queue(dev->queues[i]);

1666
	if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) {
1667 1668 1669 1670 1671 1672
		/* A device might become IO incapable very soon during
		 * probe, before the admin queue is configured. Thus,
		 * queue_count can be 0 here.
		 */
		if (dev->queue_count)
			nvme_suspend_queue(dev->queues[0]);
K
Keith Busch 已提交
1673 1674
	} else {
		nvme_disable_io_queues(dev);
1675
		nvme_disable_admin_queue(dev, shutdown);
K
Keith Busch 已提交
1676
	}
1677
	nvme_pci_disable(dev);
1678

1679 1680
	blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl);
	blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl);
1681
	mutex_unlock(&dev->shutdown_lock);
M
Matthew Wilcox 已提交
1682 1683
}

M
Matthew Wilcox 已提交
1684 1685
static int nvme_setup_prp_pools(struct nvme_dev *dev)
{
1686
	dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
M
Matthew Wilcox 已提交
1687 1688 1689 1690
						PAGE_SIZE, PAGE_SIZE, 0);
	if (!dev->prp_page_pool)
		return -ENOMEM;

1691
	/* Optimisation for I/Os between 4k and 128k */
1692
	dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
1693 1694 1695 1696 1697
						256, 256, 0);
	if (!dev->prp_small_pool) {
		dma_pool_destroy(dev->prp_page_pool);
		return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1698 1699 1700 1701 1702 1703
	return 0;
}

static void nvme_release_prp_pools(struct nvme_dev *dev)
{
	dma_pool_destroy(dev->prp_page_pool);
1704
	dma_pool_destroy(dev->prp_small_pool);
M
Matthew Wilcox 已提交
1705 1706
}

1707
static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
1708
{
1709
	struct nvme_dev *dev = to_nvme_dev(ctrl);
1710

1711
	put_device(dev->dev);
1712 1713
	if (dev->tagset.tags)
		blk_mq_free_tag_set(&dev->tagset);
1714 1715
	if (dev->ctrl.admin_q)
		blk_put_queue(dev->ctrl.admin_q);
1716 1717 1718 1719
	kfree(dev->queues);
	kfree(dev);
}

1720 1721
static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status)
{
1722
	dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status);
1723 1724

	kref_get(&dev->ctrl.kref);
1725
	nvme_dev_disable(dev, false);
1726 1727 1728 1729
	if (!schedule_work(&dev->remove_work))
		nvme_put_ctrl(&dev->ctrl);
}

1730
static void nvme_reset_work(struct work_struct *work)
1731
{
1732
	struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work);
1733
	int result = -ENODEV;
1734

1735
	if (WARN_ON(dev->ctrl.state == NVME_CTRL_RESETTING))
1736
		goto out;
1737

1738 1739 1740 1741
	/*
	 * If we're called to reset a live controller first shut it down before
	 * moving on.
	 */
1742
	if (dev->ctrl.ctrl_config & NVME_CC_ENABLE)
1743
		nvme_dev_disable(dev, false);
1744

1745
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING))
K
Keith Busch 已提交
1746 1747
		goto out;

1748
	result = nvme_pci_enable(dev);
1749
	if (result)
1750
		goto out;
1751 1752 1753

	result = nvme_configure_admin_queue(dev);
	if (result)
1754
		goto out;
1755

M
Matias Bjørling 已提交
1756
	nvme_init_queue(dev->queues[0], 0);
K
Keith Busch 已提交
1757 1758
	result = nvme_alloc_admin_tags(dev);
	if (result)
1759
		goto out;
1760

1761 1762
	result = nvme_init_identify(&dev->ctrl);
	if (result)
1763
		goto out;
1764

1765
	result = nvme_setup_io_queues(dev);
1766
	if (result)
1767
		goto out;
1768

1769 1770 1771 1772 1773 1774
	/*
	 * A controller that can not execute IO typically requires user
	 * intervention to correct. For such degraded controllers, the driver
	 * should not submit commands the user did not request, so skip
	 * registering for asynchronous event notification on this condition.
	 */
1775 1776
	if (dev->online_queues > 1)
		nvme_queue_async_events(&dev->ctrl);
1777

1778
	mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
1779

1780 1781 1782 1783
	/*
	 * Keep the controller around but remove all namespaces if we don't have
	 * any working I/O queue.
	 */
1784
	if (dev->online_queues < 2) {
1785
		dev_warn(dev->ctrl.device, "IO queues not created\n");
1786
		nvme_kill_queues(&dev->ctrl);
1787
		nvme_remove_namespaces(&dev->ctrl);
1788
	} else {
1789
		nvme_start_queues(&dev->ctrl);
1790 1791 1792
		nvme_dev_add(dev);
	}

1793 1794 1795 1796
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) {
		dev_warn(dev->ctrl.device, "failed to mark controller live\n");
		goto out;
	}
1797 1798

	if (dev->online_queues > 1)
1799
		nvme_queue_scan(&dev->ctrl);
1800
	return;
1801

1802
 out:
1803
	nvme_remove_dead_ctrl(dev, result);
1804 1805
}

1806
static void nvme_remove_dead_ctrl_work(struct work_struct *work)
K
Keith Busch 已提交
1807
{
1808
	struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
1809
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
1810

1811
	nvme_kill_queues(&dev->ctrl);
K
Keith Busch 已提交
1812
	if (pci_get_drvdata(pdev))
K
Keith Busch 已提交
1813
		device_release_driver(&pdev->dev);
1814
	nvme_put_ctrl(&dev->ctrl);
K
Keith Busch 已提交
1815 1816
}

1817
static int nvme_reset(struct nvme_dev *dev)
K
Keith Busch 已提交
1818
{
1819
	if (!dev->ctrl.admin_q || blk_queue_dying(dev->ctrl.admin_q))
1820
		return -ENODEV;
1821

C
Christoph Hellwig 已提交
1822 1823
	if (!queue_work(nvme_workq, &dev->reset_work))
		return -EBUSY;
1824

C
Christoph Hellwig 已提交
1825 1826
	flush_work(&dev->reset_work);
	return 0;
K
Keith Busch 已提交
1827 1828
}

1829
static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
T
Tejun Heo 已提交
1830
{
1831
	*val = readl(to_nvme_dev(ctrl)->bar + off);
1832
	return 0;
T
Tejun Heo 已提交
1833 1834
}

1835
static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
1836
{
1837 1838 1839
	writel(val, to_nvme_dev(ctrl)->bar + off);
	return 0;
}
1840

1841 1842 1843 1844
static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
{
	*val = readq(to_nvme_dev(ctrl)->bar + off);
	return 0;
1845 1846
}

1847 1848 1849
static int nvme_pci_reset_ctrl(struct nvme_ctrl *ctrl)
{
	return nvme_reset(to_nvme_dev(ctrl));
1850
}
1851

1852
static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
M
Ming Lin 已提交
1853
	.name			= "pcie",
1854
	.module			= THIS_MODULE,
1855
	.reg_read32		= nvme_pci_reg_read32,
1856
	.reg_write32		= nvme_pci_reg_write32,
1857
	.reg_read64		= nvme_pci_reg_read64,
1858
	.reset_ctrl		= nvme_pci_reset_ctrl,
1859
	.free_ctrl		= nvme_pci_free_ctrl,
1860
	.submit_async_event	= nvme_pci_submit_async_event,
1861
};
1862

1863 1864 1865 1866
static int nvme_dev_map(struct nvme_dev *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev->dev);

1867
	if (pci_request_mem_regions(pdev, "nvme"))
1868 1869 1870 1871 1872 1873 1874 1875
		return -ENODEV;

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

       return 0;
  release:
1876
       pci_release_mem_regions(pdev);
1877 1878 1879
       return -ENODEV;
}

1880
static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
M
Matthew Wilcox 已提交
1881
{
M
Matias Bjørling 已提交
1882
	int node, result = -ENOMEM;
M
Matthew Wilcox 已提交
1883 1884
	struct nvme_dev *dev;

M
Matias Bjørling 已提交
1885 1886
	node = dev_to_node(&pdev->dev);
	if (node == NUMA_NO_NODE)
1887
		set_dev_node(&pdev->dev, first_memory_node);
M
Matias Bjørling 已提交
1888 1889

	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
1890 1891
	if (!dev)
		return -ENOMEM;
M
Matias Bjørling 已提交
1892 1893
	dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
1894 1895 1896
	if (!dev->queues)
		goto free;

1897
	dev->dev = get_device(&pdev->dev);
K
Keith Busch 已提交
1898
	pci_set_drvdata(pdev, dev);
1899

1900 1901 1902 1903
	result = nvme_dev_map(dev);
	if (result)
		goto free;

1904
	INIT_WORK(&dev->reset_work, nvme_reset_work);
1905
	INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
1906 1907
	setup_timer(&dev->watchdog_timer, nvme_watchdog_timer,
		(unsigned long)dev);
1908
	mutex_init(&dev->shutdown_lock);
K
Keith Busch 已提交
1909
	init_completion(&dev->ioq_wait);
M
Matthew Wilcox 已提交
1910

M
Matthew Wilcox 已提交
1911 1912
	result = nvme_setup_prp_pools(dev);
	if (result)
K
Keith Busch 已提交
1913
		goto put_pci;
1914

1915 1916
	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
			id->driver_data);
1917
	if (result)
1918
		goto release_pools;
1919

1920 1921
	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));

1922
	queue_work(nvme_workq, &dev->reset_work);
M
Matthew Wilcox 已提交
1923 1924
	return 0;

1925
 release_pools:
M
Matthew Wilcox 已提交
1926
	nvme_release_prp_pools(dev);
K
Keith Busch 已提交
1927
 put_pci:
1928
	put_device(dev->dev);
1929
	nvme_dev_unmap(dev);
M
Matthew Wilcox 已提交
1930 1931 1932 1933 1934 1935
 free:
	kfree(dev->queues);
	kfree(dev);
	return result;
}

1936 1937
static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
{
K
Keith Busch 已提交
1938
	struct nvme_dev *dev = pci_get_drvdata(pdev);
1939

K
Keith Busch 已提交
1940
	if (prepare)
1941
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
1942
	else
1943
		queue_work(nvme_workq, &dev->reset_work);
1944 1945
}

1946 1947 1948
static void nvme_shutdown(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
1949
	nvme_dev_disable(dev, true);
1950 1951
}

1952 1953 1954 1955 1956
/*
 * The driver's remove may be called on a device in a partially initialized
 * state. This function must not have any dependencies on the device state in
 * order to proceed.
 */
1957
static void nvme_remove(struct pci_dev *pdev)
M
Matthew Wilcox 已提交
1958 1959
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
K
Keith Busch 已提交
1960

1961 1962
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING);

K
Keith Busch 已提交
1963
	pci_set_drvdata(pdev, NULL);
1964 1965 1966 1967

	if (!pci_device_is_present(pdev))
		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);

K
Keith Busch 已提交
1968
	flush_work(&dev->reset_work);
1969
	nvme_uninit_ctrl(&dev->ctrl);
1970
	nvme_dev_disable(dev, true);
M
Matias Bjørling 已提交
1971
	nvme_dev_remove_admin(dev);
1972
	nvme_free_queues(dev, 0);
1973
	nvme_release_cmb(dev);
K
Keith Busch 已提交
1974
	nvme_release_prp_pools(dev);
1975
	nvme_dev_unmap(dev);
1976
	nvme_put_ctrl(&dev->ctrl);
M
Matthew Wilcox 已提交
1977 1978
}

K
Keith Busch 已提交
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996
static int nvme_pci_sriov_configure(struct pci_dev *pdev, int numvfs)
{
	int ret = 0;

	if (numvfs == 0) {
		if (pci_vfs_assigned(pdev)) {
			dev_warn(&pdev->dev,
				"Cannot disable SR-IOV VFs while assigned\n");
			return -EPERM;
		}
		pci_disable_sriov(pdev);
		return 0;
	}

	ret = pci_enable_sriov(pdev, numvfs);
	return ret ? ret : numvfs;
}

1997
#ifdef CONFIG_PM_SLEEP
1998 1999 2000 2001 2002
static int nvme_suspend(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

2003
	nvme_dev_disable(ndev, true);
2004 2005 2006 2007 2008 2009 2010 2011
	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);

2012
	queue_work(nvme_workq, &ndev->reset_work);
K
Keith Busch 已提交
2013
	return 0;
2014
}
2015
#endif
2016 2017

static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
M
Matthew Wilcox 已提交
2018

K
Keith Busch 已提交
2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032
static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev,
						pci_channel_state_t state)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);

	/*
	 * A frozen channel requires a reset. When detected, this method will
	 * shutdown the controller to quiesce. The controller will be restarted
	 * after the slot reset through driver's slot_reset callback.
	 */
	switch (state) {
	case pci_channel_io_normal:
		return PCI_ERS_RESULT_CAN_RECOVER;
	case pci_channel_io_frozen:
K
Keith Busch 已提交
2033 2034
		dev_warn(dev->ctrl.device,
			"frozen state error detected, reset controller\n");
2035
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2036 2037
		return PCI_ERS_RESULT_NEED_RESET;
	case pci_channel_io_perm_failure:
K
Keith Busch 已提交
2038 2039
		dev_warn(dev->ctrl.device,
			"failure state error detected, request disconnect\n");
K
Keith Busch 已提交
2040 2041 2042 2043 2044 2045 2046 2047 2048
		return PCI_ERS_RESULT_DISCONNECT;
	}
	return PCI_ERS_RESULT_NEED_RESET;
}

static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);

2049
	dev_info(dev->ctrl.device, "restart after slot reset\n");
K
Keith Busch 已提交
2050 2051 2052 2053 2054 2055 2056 2057 2058 2059
	pci_restore_state(pdev);
	queue_work(nvme_workq, &dev->reset_work);
	return PCI_ERS_RESULT_RECOVERED;
}

static void nvme_error_resume(struct pci_dev *pdev)
{
	pci_cleanup_aer_uncorrect_error_status(pdev);
}

2060
static const struct pci_error_handlers nvme_err_handler = {
M
Matthew Wilcox 已提交
2061 2062 2063
	.error_detected	= nvme_error_detected,
	.slot_reset	= nvme_slot_reset,
	.resume		= nvme_error_resume,
2064
	.reset_notify	= nvme_reset_notify,
M
Matthew Wilcox 已提交
2065 2066 2067 2068 2069
};

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

2070
static const struct pci_device_id nvme_id_table[] = {
2071
	{ PCI_VDEVICE(INTEL, 0x0953),
2072 2073
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
				NVME_QUIRK_DISCARD_ZEROES, },
2074 2075 2076 2077 2078 2079
	{ PCI_VDEVICE(INTEL, 0x0a53),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
				NVME_QUIRK_DISCARD_ZEROES, },
	{ PCI_VDEVICE(INTEL, 0x0a54),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
				NVME_QUIRK_DISCARD_ZEROES, },
2080 2081
	{ PCI_VDEVICE(INTEL, 0x5845),	/* Qemu emulated controller */
		.driver_data = NVME_QUIRK_IDENTIFY_CNS, },
2082 2083
	{ PCI_DEVICE(0x1c58, 0x0003),	/* HGST adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2084 2085
	{ PCI_DEVICE(0x1c5f, 0x0540),	/* Memblaze Pblaze4 adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
M
Matthew Wilcox 已提交
2086
	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
2087
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
M
Matthew Wilcox 已提交
2088 2089 2090 2091 2092 2093 2094 2095
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, nvme_id_table);

static struct pci_driver nvme_driver = {
	.name		= "nvme",
	.id_table	= nvme_id_table,
	.probe		= nvme_probe,
2096
	.remove		= nvme_remove,
2097
	.shutdown	= nvme_shutdown,
2098 2099 2100
	.driver		= {
		.pm	= &nvme_dev_pm_ops,
	},
K
Keith Busch 已提交
2101
	.sriov_configure = nvme_pci_sriov_configure,
M
Matthew Wilcox 已提交
2102 2103 2104 2105 2106
	.err_handler	= &nvme_err_handler,
};

static int __init nvme_init(void)
{
2107
	int result;
2108

2109
	nvme_workq = alloc_workqueue("nvme", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
K
Keith Busch 已提交
2110
	if (!nvme_workq)
2111
		return -ENOMEM;
K
Keith Busch 已提交
2112

2113 2114
	result = pci_register_driver(&nvme_driver);
	if (result)
2115
		destroy_workqueue(nvme_workq);
M
Matthew Wilcox 已提交
2116 2117 2118 2119 2120 2121
	return result;
}

static void __exit nvme_exit(void)
{
	pci_unregister_driver(&nvme_driver);
K
Keith Busch 已提交
2122
	destroy_workqueue(nvme_workq);
2123
	_nvme_check_size();
M
Matthew Wilcox 已提交
2124 2125 2126 2127
}

MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
MODULE_LICENSE("GPL");
2128
MODULE_VERSION("1.0");
M
Matthew Wilcox 已提交
2129 2130
module_init(nvme_init);
module_exit(nvme_exit);