pci.c 65.0 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>
20
#include <linux/dmi.h>
M
Matthew Wilcox 已提交
21 22 23 24 25
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/io.h>
#include <linux/mm.h>
#include <linux/module.h>
26
#include <linux/mutex.h>
27
#include <linux/once.h>
M
Matthew Wilcox 已提交
28
#include <linux/pci.h>
29
#include <linux/poison.h>
K
Keith Busch 已提交
30
#include <linux/t10-pi.h>
31
#include <linux/timer.h>
M
Matthew Wilcox 已提交
32
#include <linux/types.h>
33
#include <linux/io-64-nonatomic-lo-hi.h>
K
Keith Busch 已提交
34
#include <asm/unaligned.h>
35
#include <linux/sed-opal.h>
36

37 38
#include "nvme.h"

M
Matthew Wilcox 已提交
39 40
#define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
#define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
41

42 43 44 45
/*
 * We handle AEN commands ourselves and don't even let the
 * block layer know about them.
 */
46
#define NVME_AQ_BLKMQ_DEPTH	(NVME_AQ_DEPTH - NVME_NR_AERS)
47

48 49 50
static int use_threaded_interrupts;
module_param(use_threaded_interrupts, int, 0);

51 52 53 54
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");

55 56 57 58
static unsigned int max_host_mem_size_mb = 128;
module_param(max_host_mem_size_mb, uint, 0444);
MODULE_PARM_DESC(max_host_mem_size_mb,
	"Maximum Host Memory Buffer (HMB) size per controller (in MiB)");
59

60 61 62 63 64 65 66 67 68 69
static int io_queue_depth_set(const char *val, const struct kernel_param *kp);
static const struct kernel_param_ops io_queue_depth_ops = {
	.set = io_queue_depth_set,
	.get = param_get_int,
};

static int io_queue_depth = 1024;
module_param_cb(io_queue_depth, &io_queue_depth_ops, &io_queue_depth, 0644);
MODULE_PARM_DESC(io_queue_depth, "set io queue depth, should >= 2");

70 71
struct nvme_dev;
struct nvme_queue;
72

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
/*
 * 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 online_queues;
	unsigned max_qid;
	int q_depth;
	u32 db_stride;
	void __iomem *bar;
92
	unsigned long bar_mapped_size;
93
	struct work_struct remove_work;
94
	struct mutex shutdown_lock;
95 96 97 98 99
	bool subsystem;
	void __iomem *cmb;
	dma_addr_t cmb_dma_addr;
	u64 cmb_size;
	u32 cmbsz;
100
	u32 cmbloc;
101
	struct nvme_ctrl ctrl;
K
Keith Busch 已提交
102
	struct completion ioq_wait;
103 104

	/* shadow doorbell buffer support: */
105 106 107 108
	u32 *dbbuf_dbs;
	dma_addr_t dbbuf_dbs_dma_addr;
	u32 *dbbuf_eis;
	dma_addr_t dbbuf_eis_dma_addr;
109 110 111 112

	/* host memory buffer support: */
	u64 host_mem_size;
	u32 nr_host_mem_descs;
113
	dma_addr_t host_mem_descs_dma;
114 115
	struct nvme_host_mem_buf_desc *host_mem_descs;
	void **host_mem_desc_bufs;
K
Keith Busch 已提交
116
};
117

118 119 120 121 122 123 124 125 126 127 128
static int io_queue_depth_set(const char *val, const struct kernel_param *kp)
{
	int n = 0, ret;

	ret = kstrtoint(val, 10, &n);
	if (ret != 0 || n < 2)
		return -EINVAL;

	return param_set_int(val, kp);
}

129 130 131 132 133 134 135 136 137 138
static inline unsigned int sq_idx(unsigned int qid, u32 stride)
{
	return qid * 2 * stride;
}

static inline unsigned int cq_idx(unsigned int qid, u32 stride)
{
	return (qid * 2 + 1) * stride;
}

139 140 141 142 143
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
144 145 146 147 148 149
/*
 * 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
150
	struct nvme_dev *dev;
M
Matthew Wilcox 已提交
151 152
	spinlock_t q_lock;
	struct nvme_command *sq_cmds;
153
	struct nvme_command __iomem *sq_cmds_io;
M
Matthew Wilcox 已提交
154
	volatile struct nvme_completion *cqes;
155
	struct blk_mq_tags **tags;
M
Matthew Wilcox 已提交
156 157 158 159
	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
160
	s16 cq_vector;
M
Matthew Wilcox 已提交
161 162
	u16 sq_tail;
	u16 cq_head;
K
Keith Busch 已提交
163
	u16 qid;
164 165
	u8 cq_phase;
	u8 cqe_seen;
166 167 168 169
	u32 *dbbuf_sq_db;
	u32 *dbbuf_cq_db;
	u32 *dbbuf_sq_ei;
	u32 *dbbuf_cq_ei;
M
Matthew Wilcox 已提交
170 171
};

172 173 174
/*
 * 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
175
 * me express that.  Use nvme_init_iod to ensure there's enough space
176 177 178
 * allocated to store the PRP list.
 */
struct nvme_iod {
179
	struct nvme_request req;
C
Christoph Hellwig 已提交
180 181
	struct nvme_queue *nvmeq;
	int aborted;
182 183 184 185
	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;
186
	struct scatterlist meta_sg; /* metadata requires single contiguous buffer */
C
Christoph Hellwig 已提交
187 188
	struct scatterlist *sg;
	struct scatterlist inline_sg[0];
M
Matthew Wilcox 已提交
189 190 191 192 193 194 195 196 197 198 199 200
};

/*
 * 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);
201
	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
K
Keith Busch 已提交
202
	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
M
Matthew Wilcox 已提交
203
	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
204 205
	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
M
Matthew Wilcox 已提交
206
	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
K
Keith Busch 已提交
207
	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281
	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
}

static inline unsigned int nvme_dbbuf_size(u32 stride)
{
	return ((num_possible_cpus() + 1) * 8 * stride);
}

static int nvme_dbbuf_dma_alloc(struct nvme_dev *dev)
{
	unsigned int mem_size = nvme_dbbuf_size(dev->db_stride);

	if (dev->dbbuf_dbs)
		return 0;

	dev->dbbuf_dbs = dma_alloc_coherent(dev->dev, mem_size,
					    &dev->dbbuf_dbs_dma_addr,
					    GFP_KERNEL);
	if (!dev->dbbuf_dbs)
		return -ENOMEM;
	dev->dbbuf_eis = dma_alloc_coherent(dev->dev, mem_size,
					    &dev->dbbuf_eis_dma_addr,
					    GFP_KERNEL);
	if (!dev->dbbuf_eis) {
		dma_free_coherent(dev->dev, mem_size,
				  dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr);
		dev->dbbuf_dbs = NULL;
		return -ENOMEM;
	}

	return 0;
}

static void nvme_dbbuf_dma_free(struct nvme_dev *dev)
{
	unsigned int mem_size = nvme_dbbuf_size(dev->db_stride);

	if (dev->dbbuf_dbs) {
		dma_free_coherent(dev->dev, mem_size,
				  dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr);
		dev->dbbuf_dbs = NULL;
	}
	if (dev->dbbuf_eis) {
		dma_free_coherent(dev->dev, mem_size,
				  dev->dbbuf_eis, dev->dbbuf_eis_dma_addr);
		dev->dbbuf_eis = NULL;
	}
}

static void nvme_dbbuf_init(struct nvme_dev *dev,
			    struct nvme_queue *nvmeq, int qid)
{
	if (!dev->dbbuf_dbs || !qid)
		return;

	nvmeq->dbbuf_sq_db = &dev->dbbuf_dbs[sq_idx(qid, dev->db_stride)];
	nvmeq->dbbuf_cq_db = &dev->dbbuf_dbs[cq_idx(qid, dev->db_stride)];
	nvmeq->dbbuf_sq_ei = &dev->dbbuf_eis[sq_idx(qid, dev->db_stride)];
	nvmeq->dbbuf_cq_ei = &dev->dbbuf_eis[cq_idx(qid, dev->db_stride)];
}

static void nvme_dbbuf_set(struct nvme_dev *dev)
{
	struct nvme_command c;

	if (!dev->dbbuf_dbs)
		return;

	memset(&c, 0, sizeof(c));
	c.dbbuf.opcode = nvme_admin_dbbuf;
	c.dbbuf.prp1 = cpu_to_le64(dev->dbbuf_dbs_dma_addr);
	c.dbbuf.prp2 = cpu_to_le64(dev->dbbuf_eis_dma_addr);

	if (nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0)) {
282
		dev_warn(dev->ctrl.device, "unable to set dbbuf\n");
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313
		/* Free memory and continue on */
		nvme_dbbuf_dma_free(dev);
	}
}

static inline int nvme_dbbuf_need_event(u16 event_idx, u16 new_idx, u16 old)
{
	return (u16)(new_idx - event_idx - 1) < (u16)(new_idx - old);
}

/* Update dbbuf and return true if an MMIO is required */
static bool nvme_dbbuf_update_and_check_event(u16 value, u32 *dbbuf_db,
					      volatile u32 *dbbuf_ei)
{
	if (dbbuf_db) {
		u16 old_value;

		/*
		 * Ensure that the queue is written before updating
		 * the doorbell in memory
		 */
		wmb();

		old_value = *dbbuf_db;
		*dbbuf_db = value;

		if (!nvme_dbbuf_need_event(*dbbuf_ei, value, old_value))
			return false;
	}

	return true;
M
Matthew Wilcox 已提交
314 315
}

316 317 318 319
/*
 * Max size of iod being embedded in the request payload
 */
#define NVME_INT_PAGES		2
320
#define NVME_INT_BYTES(dev)	(NVME_INT_PAGES * (dev)->ctrl.page_size)
321 322 323 324 325 326 327 328

/*
 * 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)
{
329 330
	unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size,
				      dev->ctrl.page_size);
331 332 333
	return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
}

C
Christoph Hellwig 已提交
334 335
static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
		unsigned int size, unsigned int nseg)
336
{
C
Christoph Hellwig 已提交
337 338 339
	return sizeof(__le64 *) * nvme_npages(size, dev) +
			sizeof(struct scatterlist) * nseg;
}
340

C
Christoph Hellwig 已提交
341 342 343 344
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);
345 346
}

M
Matias Bjørling 已提交
347 348
static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
				unsigned int hctx_idx)
349
{
M
Matias Bjørling 已提交
350 351 352
	struct nvme_dev *dev = data;
	struct nvme_queue *nvmeq = dev->queues[0];

353 354 355 356
	WARN_ON(hctx_idx != 0);
	WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
	WARN_ON(nvmeq->tags);

M
Matias Bjørling 已提交
357
	hctx->driver_data = nvmeq;
358
	nvmeq->tags = &dev->admin_tagset.tags[0];
M
Matias Bjørling 已提交
359
	return 0;
360 361
}

362 363 364 365 366 367 368
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
369 370
static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
			  unsigned int hctx_idx)
M
Matthew Wilcox 已提交
371
{
M
Matias Bjørling 已提交
372
	struct nvme_dev *dev = data;
373
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
M
Matias Bjørling 已提交
374

375 376
	if (!nvmeq->tags)
		nvmeq->tags = &dev->tagset.tags[hctx_idx];
M
Matthew Wilcox 已提交
377

378
	WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
M
Matias Bjørling 已提交
379 380
	hctx->driver_data = nvmeq;
	return 0;
M
Matthew Wilcox 已提交
381 382
}

383 384
static int nvme_init_request(struct blk_mq_tag_set *set, struct request *req,
		unsigned int hctx_idx, unsigned int numa_node)
M
Matthew Wilcox 已提交
385
{
386
	struct nvme_dev *dev = set->driver_data;
C
Christoph Hellwig 已提交
387
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
388 389
	int queue_idx = (set == &dev->tagset) ? hctx_idx + 1 : 0;
	struct nvme_queue *nvmeq = dev->queues[queue_idx];
M
Matias Bjørling 已提交
390 391

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
392
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
393 394 395
	return 0;
}

396 397 398 399 400 401 402
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
403
/**
404
 * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
M
Matthew Wilcox 已提交
405 406 407 408 409
 * @nvmeq: The queue to use
 * @cmd: The command to send
 *
 * Safe to use from interrupt context
 */
410 411
static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
						struct nvme_command *cmd)
M
Matthew Wilcox 已提交
412
{
M
Matias Bjørling 已提交
413 414
	u16 tail = nvmeq->sq_tail;

415 416 417 418 419
	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
420 421
	if (++tail == nvmeq->q_depth)
		tail = 0;
422 423 424
	if (nvme_dbbuf_update_and_check_event(tail, nvmeq->dbbuf_sq_db,
					      nvmeq->dbbuf_sq_ei))
		writel(tail, nvmeq->q_db);
M
Matthew Wilcox 已提交
425 426 427
	nvmeq->sq_tail = tail;
}

C
Christoph Hellwig 已提交
428
static __le64 **iod_list(struct request *req)
M
Matthew Wilcox 已提交
429
{
C
Christoph Hellwig 已提交
430
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
431
	return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req));
M
Matthew Wilcox 已提交
432 433
}

434
static blk_status_t nvme_init_iod(struct request *rq, struct nvme_dev *dev)
435
{
C
Christoph Hellwig 已提交
436
	struct nvme_iod *iod = blk_mq_rq_to_pdu(rq);
437
	int nseg = blk_rq_nr_phys_segments(rq);
438
	unsigned int size = blk_rq_payload_bytes(rq);
439

C
Christoph Hellwig 已提交
440 441 442
	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)
443
			return BLK_STS_RESOURCE;
C
Christoph Hellwig 已提交
444 445
	} else {
		iod->sg = iod->inline_sg;
446 447
	}

C
Christoph Hellwig 已提交
448 449 450 451
	iod->aborted = 0;
	iod->npages = -1;
	iod->nents = 0;
	iod->length = size;
K
Keith Busch 已提交
452

453
	return BLK_STS_OK;
454 455
}

C
Christoph Hellwig 已提交
456
static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
457
{
C
Christoph Hellwig 已提交
458
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
459
	const int last_prp = dev->ctrl.page_size / 8 - 1;
460
	int i;
C
Christoph Hellwig 已提交
461
	__le64 **list = iod_list(req);
462 463 464 465 466 467 468 469 470 471
	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;
	}
472

C
Christoph Hellwig 已提交
473 474
	if (iod->sg != iod->inline_sg)
		kfree(iod->sg);
K
Keith Busch 已提交
475 476
}

477
#ifdef CONFIG_BLK_DEV_INTEGRITY
K
Keith Busch 已提交
478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521
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);
522
	ts = ns->disk->queue->integrity.tuple_size;
K
Keith Busch 已提交
523 524 525 526 527 528 529 530

	for (i = 0; i < nlb; i++, virt++, phys++) {
		pi = (struct t10_pi_tuple *)p;
		dif_swap(phys, virt, pi);
		p += ts;
	}
	kunmap_atomic(pmap);
}
531 532 533 534 535 536 537 538 539 540 541 542 543
#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

544 545 546 547 548 549 550 551 552 553 554 555 556 557
static void nvme_print_sgl(struct scatterlist *sgl, int nents)
{
	int i;
	struct scatterlist *sg;

	for_each_sg(sgl, sg, nents, i) {
		dma_addr_t phys = sg_phys(sg);
		pr_warn("sg[%d] phys_addr:%pad offset:%d length:%d "
			"dma_address:%pad dma_length:%d\n",
			i, &phys, sg->offset, sg->length, &sg_dma_address(sg),
			sg_dma_len(sg));
	}
}

558
static blk_status_t nvme_setup_prps(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
559
{
C
Christoph Hellwig 已提交
560
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
561
	struct dma_pool *pool;
562
	int length = blk_rq_payload_bytes(req);
563
	struct scatterlist *sg = iod->sg;
M
Matthew Wilcox 已提交
564 565
	int dma_len = sg_dma_len(sg);
	u64 dma_addr = sg_dma_address(sg);
566
	u32 page_size = dev->ctrl.page_size;
567
	int offset = dma_addr & (page_size - 1);
568
	__le64 *prp_list;
C
Christoph Hellwig 已提交
569
	__le64 **list = iod_list(req);
570
	dma_addr_t prp_dma;
571
	int nprps, i;
M
Matthew Wilcox 已提交
572

573
	length -= (page_size - offset);
574 575
	if (length <= 0) {
		iod->first_dma = 0;
576
		return BLK_STS_OK;
577
	}
M
Matthew Wilcox 已提交
578

579
	dma_len -= (page_size - offset);
M
Matthew Wilcox 已提交
580
	if (dma_len) {
581
		dma_addr += (page_size - offset);
M
Matthew Wilcox 已提交
582 583 584 585 586 587
	} else {
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
	}

588
	if (length <= page_size) {
589
		iod->first_dma = dma_addr;
590
		return BLK_STS_OK;
591 592
	}

593
	nprps = DIV_ROUND_UP(length, page_size);
594 595
	if (nprps <= (256 / 8)) {
		pool = dev->prp_small_pool;
596
		iod->npages = 0;
597 598
	} else {
		pool = dev->prp_page_pool;
599
		iod->npages = 1;
600 601
	}

602
	prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
603
	if (!prp_list) {
604
		iod->first_dma = dma_addr;
605
		iod->npages = -1;
606
		return BLK_STS_RESOURCE;
607
	}
608 609
	list[0] = prp_list;
	iod->first_dma = prp_dma;
610 611
	i = 0;
	for (;;) {
612
		if (i == page_size >> 3) {
613
			__le64 *old_prp_list = prp_list;
614
			prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
615
			if (!prp_list)
616
				return BLK_STS_RESOURCE;
617
			list[iod->npages++] = prp_list;
618 619 620
			prp_list[0] = old_prp_list[i - 1];
			old_prp_list[i - 1] = cpu_to_le64(prp_dma);
			i = 1;
621 622
		}
		prp_list[i++] = cpu_to_le64(dma_addr);
623 624 625
		dma_len -= page_size;
		dma_addr += page_size;
		length -= page_size;
626 627 628 629
		if (length <= 0)
			break;
		if (dma_len > 0)
			continue;
630 631
		if (unlikely(dma_len < 0))
			goto bad_sgl;
632 633 634
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
M
Matthew Wilcox 已提交
635 636
	}

637 638 639
	return BLK_STS_OK;

 bad_sgl:
640 641 642
	WARN(DO_ONCE(nvme_print_sgl, iod->sg, iod->nents),
			"Invalid SGL for payload:%d nents:%d\n",
			blk_rq_payload_bytes(req), iod->nents);
643
	return BLK_STS_IOERR;
M
Matthew Wilcox 已提交
644 645
}

646
static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req,
647
		struct nvme_command *cmnd)
648
{
C
Christoph Hellwig 已提交
649
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
C
Christoph Hellwig 已提交
650 651 652
	struct request_queue *q = req->q;
	enum dma_data_direction dma_dir = rq_data_dir(req) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE;
653
	blk_status_t ret = BLK_STS_IOERR;
654

655
	sg_init_table(iod->sg, blk_rq_nr_phys_segments(req));
C
Christoph Hellwig 已提交
656 657 658
	iod->nents = blk_rq_map_sg(q, req, iod->sg);
	if (!iod->nents)
		goto out;
659

660
	ret = BLK_STS_RESOURCE;
661 662
	if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir,
				DMA_ATTR_NO_WARN))
C
Christoph Hellwig 已提交
663
		goto out;
664

665 666
	ret = nvme_setup_prps(dev, req);
	if (ret != BLK_STS_OK)
C
Christoph Hellwig 已提交
667
		goto out_unmap;
668

669
	ret = BLK_STS_IOERR;
C
Christoph Hellwig 已提交
670 671 672
	if (blk_integrity_rq(req)) {
		if (blk_rq_count_integrity_sg(q, req->bio) != 1)
			goto out_unmap;
673

674 675
		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
676
			goto out_unmap;
677

678
		if (req_op(req) == REQ_OP_WRITE)
C
Christoph Hellwig 已提交
679
			nvme_dif_remap(req, nvme_dif_prep);
680

681
		if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir))
C
Christoph Hellwig 已提交
682
			goto out_unmap;
683
	}
M
Matthew Wilcox 已提交
684

685 686
	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
687
	if (blk_integrity_rq(req))
688
		cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
689
	return BLK_STS_OK;
M
Matthew Wilcox 已提交
690

C
Christoph Hellwig 已提交
691 692 693 694
out_unmap:
	dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
out:
	return ret;
M
Matthew Wilcox 已提交
695 696
}

C
Christoph Hellwig 已提交
697
static void nvme_unmap_data(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
698
{
C
Christoph Hellwig 已提交
699
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
700 701 702 703 704 705
	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)) {
706
			if (req_op(req) == REQ_OP_READ)
707
				nvme_dif_remap(req, nvme_dif_complete);
708
			dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir);
K
Keith Busch 已提交
709
		}
710
	}
K
Keith Busch 已提交
711

712
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
713
	nvme_free_iod(dev, req);
714
}
M
Matthew Wilcox 已提交
715

716 717 718
/*
 * NOTE: ns is NULL when called on the admin queue.
 */
719
static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
M
Matias Bjørling 已提交
720
			 const struct blk_mq_queue_data *bd)
721
{
M
Matias Bjørling 已提交
722 723
	struct nvme_ns *ns = hctx->queue->queuedata;
	struct nvme_queue *nvmeq = hctx->driver_data;
724
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
725
	struct request *req = bd->rq;
C
Christoph Hellwig 已提交
726
	struct nvme_command cmnd;
727
	blk_status_t ret;
K
Keith Busch 已提交
728

729
	ret = nvme_setup_cmd(ns, req, &cmnd);
730
	if (ret)
C
Christoph Hellwig 已提交
731
		return ret;
M
Matias Bjørling 已提交
732

733
	ret = nvme_init_iod(req, dev);
734
	if (ret)
735
		goto out_free_cmd;
M
Matias Bjørling 已提交
736

737
	if (blk_rq_nr_phys_segments(req)) {
738
		ret = nvme_map_data(dev, req, &cmnd);
739 740 741
		if (ret)
			goto out_cleanup_iod;
	}
M
Matias Bjørling 已提交
742

743
	blk_mq_start_request(req);
M
Matias Bjørling 已提交
744

C
Christoph Hellwig 已提交
745
	spin_lock_irq(&nvmeq->q_lock);
746
	if (unlikely(nvmeq->cq_vector < 0)) {
747
		ret = BLK_STS_IOERR;
748
		spin_unlock_irq(&nvmeq->q_lock);
749
		goto out_cleanup_iod;
750
	}
C
Christoph Hellwig 已提交
751
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matias Bjørling 已提交
752 753
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
754
	return BLK_STS_OK;
755
out_cleanup_iod:
C
Christoph Hellwig 已提交
756
	nvme_free_iod(dev, req);
757 758
out_free_cmd:
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
759
	return ret;
M
Matthew Wilcox 已提交
760
}
K
Keith Busch 已提交
761

762
static void nvme_pci_complete_rq(struct request *req)
763
{
C
Christoph Hellwig 已提交
764
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
765

766 767
	nvme_unmap_data(iod->nvmeq->dev, req);
	nvme_complete_rq(req);
M
Matthew Wilcox 已提交
768 769
}

770 771 772 773 774 775 776
/* 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;
}

777
static inline void nvme_ring_cq_doorbell(struct nvme_queue *nvmeq)
M
Matthew Wilcox 已提交
778
{
779
	u16 head = nvmeq->cq_head;
780

781 782 783 784 785 786
	if (likely(nvmeq->cq_vector >= 0)) {
		if (nvme_dbbuf_update_and_check_event(head, nvmeq->dbbuf_cq_db,
						      nvmeq->dbbuf_cq_ei))
			writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
	}
}
787

788 789 790 791
static inline void nvme_handle_cqe(struct nvme_queue *nvmeq,
		struct nvme_completion *cqe)
{
	struct request *req;
792

793 794 795 796 797
	if (unlikely(cqe->command_id >= nvmeq->q_depth)) {
		dev_warn(nvmeq->dev->ctrl.device,
			"invalid id %d completed on queue %d\n",
			cqe->command_id, le16_to_cpu(cqe->sq_id));
		return;
M
Matthew Wilcox 已提交
798 799
	}

800 801 802 803 804 805 806 807 808 809
	/*
	 * 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)) {
		nvme_complete_async_event(&nvmeq->dev->ctrl,
				cqe->status, &cqe->result);
J
Jens Axboe 已提交
810
		return;
811
	}
M
Matthew Wilcox 已提交
812

813
	nvmeq->cqe_seen = 1;
814 815 816
	req = blk_mq_tag_to_rq(*nvmeq->tags, cqe->command_id);
	nvme_end_request(req, cqe->status, cqe->result);
}
M
Matthew Wilcox 已提交
817

818 819
static inline bool nvme_read_cqe(struct nvme_queue *nvmeq,
		struct nvme_completion *cqe)
M
Matthew Wilcox 已提交
820
{
821 822
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) {
		*cqe = nvmeq->cqes[nvmeq->cq_head];
823

824 825 826
		if (++nvmeq->cq_head == nvmeq->q_depth) {
			nvmeq->cq_head = 0;
			nvmeq->cq_phase = !nvmeq->cq_phase;
M
Matthew Wilcox 已提交
827
		}
828
		return true;
M
Matthew Wilcox 已提交
829
	}
830
	return false;
J
Jens Axboe 已提交
831 832 833 834
}

static void nvme_process_cq(struct nvme_queue *nvmeq)
{
835 836
	struct nvme_completion cqe;
	int consumed = 0;
M
Matthew Wilcox 已提交
837

838 839 840 841
	while (nvme_read_cqe(nvmeq, &cqe)) {
		nvme_handle_cqe(nvmeq, &cqe);
		consumed++;
	}
842

843
	if (consumed)
844
		nvme_ring_cq_doorbell(nvmeq);
M
Matthew Wilcox 已提交
845 846 847
}

static irqreturn_t nvme_irq(int irq, void *data)
848 849 850 851
{
	irqreturn_t result;
	struct nvme_queue *nvmeq = data;
	spin_lock(&nvmeq->q_lock);
852 853 854
	nvme_process_cq(nvmeq);
	result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
	nvmeq->cqe_seen = 0;
855 856 857 858 859 860 861
	spin_unlock(&nvmeq->q_lock);
	return result;
}

static irqreturn_t nvme_irq_check(int irq, void *data)
{
	struct nvme_queue *nvmeq = data;
862 863 864
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return IRQ_WAKE_THREAD;
	return IRQ_NONE;
865 866
}

K
Keith Busch 已提交
867
static int __nvme_poll(struct nvme_queue *nvmeq, unsigned int tag)
J
Jens Axboe 已提交
868
{
869 870
	struct nvme_completion cqe;
	int found = 0, consumed = 0;
J
Jens Axboe 已提交
871

872 873
	if (!nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return 0;
J
Jens Axboe 已提交
874

875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890
	spin_lock_irq(&nvmeq->q_lock);
	while (nvme_read_cqe(nvmeq, &cqe)) {
		nvme_handle_cqe(nvmeq, &cqe);
		consumed++;

		if (tag == cqe.command_id) {
			found = 1;
			break;
		}
       }

	if (consumed)
		nvme_ring_cq_doorbell(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);

	return found;
J
Jens Axboe 已提交
891 892
}

K
Keith Busch 已提交
893 894 895 896 897 898 899
static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
{
	struct nvme_queue *nvmeq = hctx->driver_data;

	return __nvme_poll(nvmeq, tag);
}

900
static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx)
M
Matthew Wilcox 已提交
901
{
902
	struct nvme_dev *dev = to_nvme_dev(ctrl);
903
	struct nvme_queue *nvmeq = dev->queues[0];
M
Matias Bjørling 已提交
904
	struct nvme_command c;
M
Matthew Wilcox 已提交
905

M
Matias Bjørling 已提交
906 907
	memset(&c, 0, sizeof(c));
	c.common.opcode = nvme_admin_async_event;
908
	c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx;
909

910
	spin_lock_irq(&nvmeq->q_lock);
911
	__nvme_submit_cmd(nvmeq, &c);
912
	spin_unlock_irq(&nvmeq->q_lock);
913 914
}

M
Matthew Wilcox 已提交
915
static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
916
{
M
Matthew Wilcox 已提交
917 918 919 920 921 922
	struct nvme_command c;

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

923
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
924 925 926 927 928 929 930 931
}

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;

932 933 934 935
	/*
	 * 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
936 937 938 939 940 941 942 943
	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);

944
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
945 946 947 948 949 950
}

static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
						struct nvme_queue *nvmeq)
{
	struct nvme_command c;
951
	int flags = NVME_QUEUE_PHYS_CONTIG;
M
Matthew Wilcox 已提交
952

953 954 955 956
	/*
	 * 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
957 958 959 960 961 962 963 964
	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);

965
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
966 967 968 969 970 971 972 973 974 975 976 977
}

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);
}

978
static void abort_endio(struct request *req, blk_status_t error)
979
{
C
Christoph Hellwig 已提交
980 981
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
982

983 984
	dev_warn(nvmeq->dev->ctrl.device,
		 "Abort status: 0x%x", nvme_req(req)->status);
985 986
	atomic_inc(&nvmeq->dev->ctrl.abort_limit);
	blk_mq_free_request(req);
987 988
}

K
Keith Busch 已提交
989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
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 (dev->ctrl.state == NVME_CTRL_RESETTING)
		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;
}

static void nvme_warn_reset(struct nvme_dev *dev, u32 csts)
{
	/* Read a config register to help see what died. */
	u16 pci_status;
	int result;

	result = pci_read_config_word(to_pci_dev(dev->dev), PCI_STATUS,
				      &pci_status);
	if (result == PCIBIOS_SUCCESSFUL)
		dev_warn(dev->ctrl.device,
			 "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n",
			 csts, pci_status);
	else
		dev_warn(dev->ctrl.device,
			 "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n",
			 csts, result);
}

1034
static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
K
Keith Busch 已提交
1035
{
C
Christoph Hellwig 已提交
1036 1037
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
K
Keith Busch 已提交
1038
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
1039 1040
	struct request *abort_req;
	struct nvme_command cmd;
K
Keith Busch 已提交
1041 1042 1043 1044 1045 1046 1047 1048
	u32 csts = readl(dev->bar + NVME_REG_CSTS);

	/*
	 * Reset immediately if the controller is failed
	 */
	if (nvme_should_reset(dev, csts)) {
		nvme_warn_reset(dev, csts);
		nvme_dev_disable(dev, false);
1049
		nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
1050 1051
		return BLK_EH_HANDLED;
	}
K
Keith Busch 已提交
1052

K
Keith Busch 已提交
1053 1054 1055 1056 1057 1058 1059 1060 1061 1062
	/*
	 * Did we miss an interrupt?
	 */
	if (__nvme_poll(nvmeq, req->tag)) {
		dev_warn(dev->ctrl.device,
			 "I/O %d QID %d timeout, completion polled\n",
			 req->tag, nvmeq->qid);
		return BLK_EH_HANDLED;
	}

1063
	/*
1064 1065 1066 1067 1068
	 * 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.
	 */
1069
	if (dev->ctrl.state == NVME_CTRL_RESETTING) {
1070
		dev_warn(dev->ctrl.device,
1071 1072
			 "I/O %d QID %d timeout, disable controller\n",
			 req->tag, nvmeq->qid);
1073
		nvme_dev_disable(dev, false);
1074
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1075
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1076 1077
	}

1078 1079 1080 1081
	/*
 	 * 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.
1082
	 */
C
Christoph Hellwig 已提交
1083
	if (!nvmeq->qid || iod->aborted) {
1084
		dev_warn(dev->ctrl.device,
1085 1086
			 "I/O %d QID %d timeout, reset controller\n",
			 req->tag, nvmeq->qid);
1087
		nvme_dev_disable(dev, false);
1088
		nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
1089

1090 1091 1092 1093
		/*
		 * Mark the request as handled, since the inline shutdown
		 * forces all outstanding requests to complete.
		 */
1094
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1095
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1096 1097
	}

1098
	if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) {
1099
		atomic_inc(&dev->ctrl.abort_limit);
1100
		return BLK_EH_RESET_TIMER;
1101
	}
1102
	iod->aborted = 1;
M
Matias Bjørling 已提交
1103

K
Keith Busch 已提交
1104 1105
	memset(&cmd, 0, sizeof(cmd));
	cmd.abort.opcode = nvme_admin_abort_cmd;
M
Matias Bjørling 已提交
1106
	cmd.abort.cid = req->tag;
K
Keith Busch 已提交
1107 1108
	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);

1109 1110 1111
	dev_warn(nvmeq->dev->ctrl.device,
		"I/O %d QID %d timeout, aborting\n",
		 req->tag, nvmeq->qid);
1112 1113

	abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
1114
			BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
1115 1116 1117 1118 1119 1120 1121 1122
	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
1123

1124 1125 1126 1127 1128 1129
	/*
	 * 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
1130 1131
}

M
Matias Bjørling 已提交
1132 1133
static void nvme_free_queue(struct nvme_queue *nvmeq)
{
1134 1135
	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
1136 1137
	if (nvmeq->sq_cmds)
		dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
1138 1139 1140 1141
					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
	kfree(nvmeq);
}

1142
static void nvme_free_queues(struct nvme_dev *dev, int lowest)
1143 1144 1145
{
	int i;

1146
	for (i = dev->ctrl.queue_count - 1; i >= lowest; i--) {
M
Matias Bjørling 已提交
1147
		struct nvme_queue *nvmeq = dev->queues[i];
1148
		dev->ctrl.queue_count--;
M
Matias Bjørling 已提交
1149
		dev->queues[i] = NULL;
K
Keith Busch 已提交
1150
		nvme_free_queue(nvmeq);
1151
	}
1152 1153
}

K
Keith Busch 已提交
1154 1155 1156 1157 1158
/**
 * 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
1159
{
K
Keith Busch 已提交
1160
	int vector;
M
Matthew Wilcox 已提交
1161

1162
	spin_lock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
1163 1164 1165 1166
	if (nvmeq->cq_vector == -1) {
		spin_unlock_irq(&nvmeq->q_lock);
		return 1;
	}
1167
	vector = nvmeq->cq_vector;
K
Keith Busch 已提交
1168
	nvmeq->dev->online_queues--;
K
Keith Busch 已提交
1169
	nvmeq->cq_vector = -1;
1170 1171
	spin_unlock_irq(&nvmeq->q_lock);

1172
	if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
1173
		blk_mq_quiesce_queue(nvmeq->dev->ctrl.admin_q);
1174

1175
	pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq);
M
Matthew Wilcox 已提交
1176

K
Keith Busch 已提交
1177 1178
	return 0;
}
M
Matthew Wilcox 已提交
1179

1180
static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
K
Keith Busch 已提交
1181
{
1182
	struct nvme_queue *nvmeq = dev->queues[0];
K
Keith Busch 已提交
1183 1184 1185 1186 1187 1188

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

1189 1190 1191
	if (shutdown)
		nvme_shutdown_ctrl(&dev->ctrl);
	else
1192
		nvme_disable_ctrl(&dev->ctrl, dev->ctrl.cap);
1193 1194 1195 1196

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1197 1198
}

1199 1200 1201 1202
static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
				int entry_size)
{
	int q_depth = dev->q_depth;
1203 1204
	unsigned q_size_aligned = roundup(q_depth * entry_size,
					  dev->ctrl.page_size);
1205 1206

	if (q_size_aligned * nr_io_queues > dev->cmb_size) {
1207
		u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1208
		mem_per_q = round_down(mem_per_q, dev->ctrl.page_size);
1209
		q_depth = div_u64(mem_per_q, entry_size);
1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226

		/*
		 * 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)) {
1227 1228
		unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth),
						      dev->ctrl.page_size);
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240
		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
1241
static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
1242
							int depth, int node)
M
Matthew Wilcox 已提交
1243
{
1244 1245
	struct nvme_queue *nvmeq = kzalloc_node(sizeof(*nvmeq), GFP_KERNEL,
							node);
M
Matthew Wilcox 已提交
1246 1247 1248
	if (!nvmeq)
		return NULL;

1249
	nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
J
Joe Perches 已提交
1250
					  &nvmeq->cq_dma_addr, GFP_KERNEL);
M
Matthew Wilcox 已提交
1251 1252 1253
	if (!nvmeq->cqes)
		goto free_nvmeq;

1254
	if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
M
Matthew Wilcox 已提交
1255 1256
		goto free_cqdma;

1257
	nvmeq->q_dmadev = dev->dev;
M
Matthew Wilcox 已提交
1258
	nvmeq->dev = dev;
M
Matthew Wilcox 已提交
1259 1260
	spin_lock_init(&nvmeq->q_lock);
	nvmeq->cq_head = 0;
M
Matthew Wilcox 已提交
1261
	nvmeq->cq_phase = 1;
1262
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
M
Matthew Wilcox 已提交
1263
	nvmeq->q_depth = depth;
K
Keith Busch 已提交
1264
	nvmeq->qid = qid;
1265
	nvmeq->cq_vector = -1;
M
Matias Bjørling 已提交
1266
	dev->queues[qid] = nvmeq;
1267
	dev->ctrl.queue_count++;
1268

M
Matthew Wilcox 已提交
1269 1270 1271
	return nvmeq;

 free_cqdma:
1272
	dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
M
Matthew Wilcox 已提交
1273 1274 1275 1276 1277 1278
							nvmeq->cq_dma_addr);
 free_nvmeq:
	kfree(nvmeq);
	return NULL;
}

1279
static int queue_request_irq(struct nvme_queue *nvmeq)
1280
{
1281 1282 1283 1284 1285 1286 1287 1288 1289 1290
	struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev);
	int nr = nvmeq->dev->ctrl.instance;

	if (use_threaded_interrupts) {
		return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq_check,
				nvme_irq, nvmeq, "nvme%dq%d", nr, nvmeq->qid);
	} else {
		return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq,
				NULL, nvmeq, "nvme%dq%d", nr, nvmeq->qid);
	}
1291 1292
}

1293
static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
M
Matthew Wilcox 已提交
1294
{
1295
	struct nvme_dev *dev = nvmeq->dev;
M
Matthew Wilcox 已提交
1296

1297
	spin_lock_irq(&nvmeq->q_lock);
1298 1299 1300
	nvmeq->sq_tail = 0;
	nvmeq->cq_head = 0;
	nvmeq->cq_phase = 1;
1301
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
1302
	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
1303
	nvme_dbbuf_init(dev, nvmeq, qid);
K
Keith Busch 已提交
1304
	dev->online_queues++;
1305
	spin_unlock_irq(&nvmeq->q_lock);
1306 1307 1308 1309 1310 1311
}

static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
{
	struct nvme_dev *dev = nvmeq->dev;
	int result;
1312

K
Keith Busch 已提交
1313
	nvmeq->cq_vector = qid - 1;
M
Matthew Wilcox 已提交
1314 1315
	result = adapter_alloc_cq(dev, qid, nvmeq);
	if (result < 0)
1316
		return result;
M
Matthew Wilcox 已提交
1317 1318 1319 1320 1321

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

1322
	nvme_init_queue(nvmeq, qid);
1323
	result = queue_request_irq(nvmeq);
M
Matthew Wilcox 已提交
1324 1325 1326
	if (result < 0)
		goto release_sq;

1327
	return result;
M
Matthew Wilcox 已提交
1328 1329 1330 1331 1332

 release_sq:
	adapter_delete_sq(dev, qid);
 release_cq:
	adapter_delete_cq(dev, qid);
1333
	return result;
M
Matthew Wilcox 已提交
1334 1335
}

1336
static const struct blk_mq_ops nvme_mq_admin_ops = {
1337
	.queue_rq	= nvme_queue_rq,
1338
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1339
	.init_hctx	= nvme_admin_init_hctx,
1340
	.exit_hctx      = nvme_admin_exit_hctx,
1341
	.init_request	= nvme_init_request,
M
Matias Bjørling 已提交
1342 1343 1344
	.timeout	= nvme_timeout,
};

1345
static const struct blk_mq_ops nvme_mq_ops = {
M
Matias Bjørling 已提交
1346
	.queue_rq	= nvme_queue_rq,
1347
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1348 1349
	.init_hctx	= nvme_init_hctx,
	.init_request	= nvme_init_request,
1350
	.map_queues	= nvme_pci_map_queues,
M
Matias Bjørling 已提交
1351
	.timeout	= nvme_timeout,
J
Jens Axboe 已提交
1352
	.poll		= nvme_poll,
M
Matias Bjørling 已提交
1353 1354
};

1355 1356
static void nvme_dev_remove_admin(struct nvme_dev *dev)
{
1357
	if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) {
1358 1359 1360 1361 1362
		/*
		 * 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.
		 */
1363
		blk_mq_unquiesce_queue(dev->ctrl.admin_q);
1364
		blk_cleanup_queue(dev->ctrl.admin_q);
1365 1366 1367 1368
		blk_mq_free_tag_set(&dev->admin_tagset);
	}
}

M
Matias Bjørling 已提交
1369 1370
static int nvme_alloc_admin_tags(struct nvme_dev *dev)
{
1371
	if (!dev->ctrl.admin_q) {
M
Matias Bjørling 已提交
1372 1373
		dev->admin_tagset.ops = &nvme_mq_admin_ops;
		dev->admin_tagset.nr_hw_queues = 1;
K
Keith Busch 已提交
1374 1375 1376 1377 1378 1379

		/*
		 * 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
1380
		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
1381
		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
1382
		dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
1383
		dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED;
M
Matias Bjørling 已提交
1384 1385 1386 1387
		dev->admin_tagset.driver_data = dev;

		if (blk_mq_alloc_tag_set(&dev->admin_tagset))
			return -ENOMEM;
1388
		dev->ctrl.admin_tagset = &dev->admin_tagset;
M
Matias Bjørling 已提交
1389

1390 1391
		dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset);
		if (IS_ERR(dev->ctrl.admin_q)) {
M
Matias Bjørling 已提交
1392 1393 1394
			blk_mq_free_tag_set(&dev->admin_tagset);
			return -ENOMEM;
		}
1395
		if (!blk_get_queue(dev->ctrl.admin_q)) {
1396
			nvme_dev_remove_admin(dev);
1397
			dev->ctrl.admin_q = NULL;
1398 1399
			return -ENODEV;
		}
K
Keith Busch 已提交
1400
	} else
1401
		blk_mq_unquiesce_queue(dev->ctrl.admin_q);
M
Matias Bjørling 已提交
1402 1403 1404 1405

	return 0;
}

1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431
static unsigned long db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
{
	return NVME_REG_DBS + ((nr_io_queues + 1) * 8 * dev->db_stride);
}

static int nvme_remap_bar(struct nvme_dev *dev, unsigned long size)
{
	struct pci_dev *pdev = to_pci_dev(dev->dev);

	if (size <= dev->bar_mapped_size)
		return 0;
	if (size > pci_resource_len(pdev, 0))
		return -ENOMEM;
	if (dev->bar)
		iounmap(dev->bar);
	dev->bar = ioremap(pci_resource_start(pdev, 0), size);
	if (!dev->bar) {
		dev->bar_mapped_size = 0;
		return -ENOMEM;
	}
	dev->bar_mapped_size = size;
	dev->dbs = dev->bar + NVME_REG_DBS;

	return 0;
}

1432
static int nvme_pci_configure_admin_queue(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1433
{
1434
	int result;
M
Matthew Wilcox 已提交
1435 1436 1437
	u32 aqa;
	struct nvme_queue *nvmeq;

1438 1439 1440 1441
	result = nvme_remap_bar(dev, db_bar_size(dev, 0));
	if (result < 0)
		return result;

1442
	dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ?
1443
				NVME_CAP_NSSRC(dev->ctrl.cap) : 0;
1444

1445 1446 1447
	if (dev->subsystem &&
	    (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO))
		writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS);
1448

1449
	result = nvme_disable_ctrl(&dev->ctrl, dev->ctrl.cap);
1450 1451
	if (result < 0)
		return result;
M
Matthew Wilcox 已提交
1452

M
Matias Bjørling 已提交
1453
	nvmeq = dev->queues[0];
1454
	if (!nvmeq) {
1455 1456
		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH,
					dev_to_node(dev->dev));
1457 1458 1459
		if (!nvmeq)
			return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1460 1461 1462 1463

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

1464 1465 1466
	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
1467

1468
	result = nvme_enable_ctrl(&dev->ctrl, dev->ctrl.cap);
1469
	if (result)
K
Keith Busch 已提交
1470
		return result;
M
Matias Bjørling 已提交
1471

K
Keith Busch 已提交
1472
	nvmeq->cq_vector = 0;
1473
	nvme_init_queue(nvmeq, 0);
1474
	result = queue_request_irq(nvmeq);
1475 1476
	if (result) {
		nvmeq->cq_vector = -1;
K
Keith Busch 已提交
1477
		return result;
1478
	}
1479

M
Matthew Wilcox 已提交
1480 1481 1482
	return result;
}

1483
static int nvme_create_io_queues(struct nvme_dev *dev)
K
Keith Busch 已提交
1484
{
1485
	unsigned i, max;
1486
	int ret = 0;
K
Keith Busch 已提交
1487

1488
	for (i = dev->ctrl.queue_count; i <= dev->max_qid; i++) {
1489 1490 1491
		/* vector == qid - 1, match nvme_create_queue */
		if (!nvme_alloc_queue(dev, i, dev->q_depth,
		     pci_irq_get_node(to_pci_dev(dev->dev), i - 1))) {
1492
			ret = -ENOMEM;
K
Keith Busch 已提交
1493
			break;
1494 1495
		}
	}
K
Keith Busch 已提交
1496

1497
	max = min(dev->max_qid, dev->ctrl.queue_count - 1);
1498
	for (i = dev->online_queues; i <= max; i++) {
1499
		ret = nvme_create_queue(dev->queues[i], i);
K
Keith Busch 已提交
1500
		if (ret)
K
Keith Busch 已提交
1501
			break;
M
Matthew Wilcox 已提交
1502
	}
1503 1504 1505 1506 1507 1508 1509 1510

	/*
	 * 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
1511 1512
}

1513 1514 1515 1516 1517 1518
static ssize_t nvme_cmb_show(struct device *dev,
			     struct device_attribute *attr,
			     char *buf)
{
	struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev));

1519
	return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz  : x%08x\n",
1520 1521 1522 1523
		       ndev->cmbloc, ndev->cmbsz);
}
static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL);

1524 1525 1526 1527 1528 1529 1530 1531
static void __iomem *nvme_map_cmb(struct nvme_dev *dev)
{
	u64 szu, size, offset;
	resource_size_t bar_size;
	struct pci_dev *pdev = to_pci_dev(dev->dev);
	void __iomem *cmb;
	dma_addr_t dma_addr;

1532
	dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ);
1533 1534
	if (!(NVME_CMB_SZ(dev->cmbsz)))
		return NULL;
1535
	dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC);
1536

1537 1538
	if (!use_cmb_sqes)
		return NULL;
1539 1540 1541

	szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
	size = szu * NVME_CMB_SZ(dev->cmbsz);
1542 1543
	offset = szu * NVME_CMB_OFST(dev->cmbloc);
	bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc));
1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555

	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;

1556
	dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset;
1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570
	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;
1571 1572 1573
		sysfs_remove_file_from_group(&dev->ctrl.device->kobj,
					     &dev_attr_cmb.attr, NULL);
		dev->cmbsz = 0;
1574 1575 1576
	}
}

1577 1578
static int nvme_set_host_mem(struct nvme_dev *dev, u32 bits)
{
1579
	u64 dma_addr = dev->host_mem_descs_dma;
1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615
	struct nvme_command c;
	int ret;

	memset(&c, 0, sizeof(c));
	c.features.opcode	= nvme_admin_set_features;
	c.features.fid		= cpu_to_le32(NVME_FEAT_HOST_MEM_BUF);
	c.features.dword11	= cpu_to_le32(bits);
	c.features.dword12	= cpu_to_le32(dev->host_mem_size >>
					      ilog2(dev->ctrl.page_size));
	c.features.dword13	= cpu_to_le32(lower_32_bits(dma_addr));
	c.features.dword14	= cpu_to_le32(upper_32_bits(dma_addr));
	c.features.dword15	= cpu_to_le32(dev->nr_host_mem_descs);

	ret = nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
	if (ret) {
		dev_warn(dev->ctrl.device,
			 "failed to set host mem (err %d, flags %#x).\n",
			 ret, bits);
	}
	return ret;
}

static void nvme_free_host_mem(struct nvme_dev *dev)
{
	int i;

	for (i = 0; i < dev->nr_host_mem_descs; i++) {
		struct nvme_host_mem_buf_desc *desc = &dev->host_mem_descs[i];
		size_t size = le32_to_cpu(desc->size) * dev->ctrl.page_size;

		dma_free_coherent(dev->dev, size, dev->host_mem_desc_bufs[i],
				le64_to_cpu(desc->addr));
	}

	kfree(dev->host_mem_desc_bufs);
	dev->host_mem_desc_bufs = NULL;
1616 1617 1618
	dma_free_coherent(dev->dev,
			dev->nr_host_mem_descs * sizeof(*dev->host_mem_descs),
			dev->host_mem_descs, dev->host_mem_descs_dma);
1619 1620 1621
	dev->host_mem_descs = NULL;
}

1622 1623
static int __nvme_alloc_host_mem(struct nvme_dev *dev, u64 preferred,
		u32 chunk_size)
K
Keith Busch 已提交
1624
{
1625
	struct nvme_host_mem_buf_desc *descs;
1626
	u32 max_entries, len;
1627
	dma_addr_t descs_dma;
1628
	int i = 0;
1629
	void **bufs;
1630
	u64 size = 0, tmp;
1631 1632 1633 1634

	tmp = (preferred + chunk_size - 1);
	do_div(tmp, chunk_size);
	max_entries = tmp;
1635 1636 1637 1638

	if (dev->ctrl.hmmaxd && dev->ctrl.hmmaxd < max_entries)
		max_entries = dev->ctrl.hmmaxd;

1639 1640
	descs = dma_zalloc_coherent(dev->dev, max_entries * sizeof(*descs),
			&descs_dma, GFP_KERNEL);
1641 1642 1643 1644 1645 1646 1647
	if (!descs)
		goto out;

	bufs = kcalloc(max_entries, sizeof(*bufs), GFP_KERNEL);
	if (!bufs)
		goto out_free_descs;

1648
	for (size = 0; size < preferred; size += len) {
1649 1650
		dma_addr_t dma_addr;

1651
		len = min_t(u64, chunk_size, preferred - size);
1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
		bufs[i] = dma_alloc_attrs(dev->dev, len, &dma_addr, GFP_KERNEL,
				DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN);
		if (!bufs[i])
			break;

		descs[i].addr = cpu_to_le64(dma_addr);
		descs[i].size = cpu_to_le32(len / dev->ctrl.page_size);
		i++;
	}

1662
	if (!size)
1663 1664 1665 1666 1667
		goto out_free_bufs;

	dev->nr_host_mem_descs = i;
	dev->host_mem_size = size;
	dev->host_mem_descs = descs;
1668
	dev->host_mem_descs_dma = descs_dma;
1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681
	dev->host_mem_desc_bufs = bufs;
	return 0;

out_free_bufs:
	while (--i >= 0) {
		size_t size = le32_to_cpu(descs[i].size) * dev->ctrl.page_size;

		dma_free_coherent(dev->dev, size, bufs[i],
				le64_to_cpu(descs[i].addr));
	}

	kfree(bufs);
out_free_descs:
1682 1683
	dma_free_coherent(dev->dev, max_entries * sizeof(*descs), descs,
			descs_dma);
1684 1685 1686 1687 1688
out:
	dev->host_mem_descs = NULL;
	return -ENOMEM;
}

1689 1690 1691 1692 1693
static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred)
{
	u32 chunk_size;

	/* start big and work our way down */
1694
	for (chunk_size = min_t(u64, preferred, PAGE_SIZE * MAX_ORDER_NR_PAGES);
1695
	     chunk_size >= max_t(u32, dev->ctrl.hmminds * 4096, PAGE_SIZE * 2);
1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706
	     chunk_size /= 2) {
		if (!__nvme_alloc_host_mem(dev, preferred, chunk_size)) {
			if (!min || dev->host_mem_size >= min)
				return 0;
			nvme_free_host_mem(dev);
		}
	}

	return -ENOMEM;
}

1707
static int nvme_setup_host_mem(struct nvme_dev *dev)
1708 1709 1710 1711 1712
{
	u64 max = (u64)max_host_mem_size_mb * SZ_1M;
	u64 preferred = (u64)dev->ctrl.hmpre * 4096;
	u64 min = (u64)dev->ctrl.hmmin * 4096;
	u32 enable_bits = NVME_HOST_MEM_ENABLE;
1713
	int ret = 0;
1714 1715 1716 1717 1718 1719 1720

	preferred = min(preferred, max);
	if (min > max) {
		dev_warn(dev->ctrl.device,
			"min host memory (%lld MiB) above limit (%d MiB).\n",
			min >> ilog2(SZ_1M), max_host_mem_size_mb);
		nvme_free_host_mem(dev);
1721
		return 0;
1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734
	}

	/*
	 * If we already have a buffer allocated check if we can reuse it.
	 */
	if (dev->host_mem_descs) {
		if (dev->host_mem_size >= min)
			enable_bits |= NVME_HOST_MEM_RETURN;
		else
			nvme_free_host_mem(dev);
	}

	if (!dev->host_mem_descs) {
1735 1736 1737
		if (nvme_alloc_host_mem(dev, min, preferred)) {
			dev_warn(dev->ctrl.device,
				"failed to allocate host memory buffer.\n");
1738
			return 0; /* controller must work without HMB */
1739 1740 1741 1742 1743
		}

		dev_info(dev->ctrl.device,
			"allocated %lld MiB host memory buffer.\n",
			dev->host_mem_size >> ilog2(SZ_1M));
1744 1745
	}

1746 1747
	ret = nvme_set_host_mem(dev, enable_bits);
	if (ret)
1748
		nvme_free_host_mem(dev);
1749
	return ret;
K
Keith Busch 已提交
1750 1751
}

1752
static int nvme_setup_io_queues(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1753
{
M
Matias Bjørling 已提交
1754
	struct nvme_queue *adminq = dev->queues[0];
1755
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1756 1757
	int result, nr_io_queues;
	unsigned long size;
M
Matthew Wilcox 已提交
1758

1759
	nr_io_queues = num_present_cpus();
C
Christoph Hellwig 已提交
1760 1761
	result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
	if (result < 0)
M
Matthew Wilcox 已提交
1762
		return result;
C
Christoph Hellwig 已提交
1763

1764
	if (nr_io_queues == 0)
1765
		return 0;
M
Matthew Wilcox 已提交
1766

1767 1768 1769 1770 1771 1772 1773 1774 1775
	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);
	}

1776 1777 1778 1779 1780 1781 1782 1783 1784
	do {
		size = db_bar_size(dev, nr_io_queues);
		result = nvme_remap_bar(dev, size);
		if (!result)
			break;
		if (!--nr_io_queues)
			return -ENOMEM;
	} while (1);
	adminq->q_db = dev->dbs;
1785

K
Keith Busch 已提交
1786
	/* Deregister the admin queue's interrupt */
1787
	pci_free_irq(pdev, 0, adminq);
K
Keith Busch 已提交
1788

1789 1790 1791 1792
	/*
	 * If we enable msix early due to not intx, disable it again before
	 * setting up the full range we need.
	 */
1793 1794 1795 1796 1797 1798
	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
1799

1800 1801 1802 1803 1804 1805 1806
	/*
	 * 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.
	 */

1807
	result = queue_request_irq(adminq);
1808 1809
	if (result) {
		adminq->cq_vector = -1;
K
Keith Busch 已提交
1810
		return result;
1811
	}
1812
	return nvme_create_io_queues(dev);
M
Matthew Wilcox 已提交
1813 1814
}

1815
static void nvme_del_queue_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1816
{
K
Keith Busch 已提交
1817
	struct nvme_queue *nvmeq = req->end_io_data;
1818

K
Keith Busch 已提交
1819 1820
	blk_mq_free_request(req);
	complete(&nvmeq->dev->ioq_wait);
K
Keith Busch 已提交
1821 1822
}

1823
static void nvme_del_cq_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1824
{
K
Keith Busch 已提交
1825
	struct nvme_queue *nvmeq = req->end_io_data;
K
Keith Busch 已提交
1826

K
Keith Busch 已提交
1827 1828 1829
	if (!error) {
		unsigned long flags;

1830 1831 1832 1833 1834 1835 1836
		/*
		 * 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
1837 1838
		nvme_process_cq(nvmeq);
		spin_unlock_irqrestore(&nvmeq->q_lock, flags);
K
Keith Busch 已提交
1839
	}
K
Keith Busch 已提交
1840 1841

	nvme_del_queue_end(req, error);
K
Keith Busch 已提交
1842 1843
}

K
Keith Busch 已提交
1844
static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
1845
{
K
Keith Busch 已提交
1846 1847 1848
	struct request_queue *q = nvmeq->dev->ctrl.admin_q;
	struct request *req;
	struct nvme_command cmd;
1849

K
Keith Busch 已提交
1850 1851 1852
	memset(&cmd, 0, sizeof(cmd));
	cmd.delete_queue.opcode = opcode;
	cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
1853

1854
	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
K
Keith Busch 已提交
1855 1856
	if (IS_ERR(req))
		return PTR_ERR(req);
1857

K
Keith Busch 已提交
1858 1859 1860 1861 1862 1863 1864
	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;
1865 1866
}

1867
static void nvme_disable_io_queues(struct nvme_dev *dev, int queues)
K
Keith Busch 已提交
1868
{
1869
	int pass;
K
Keith Busch 已提交
1870 1871
	unsigned long timeout;
	u8 opcode = nvme_admin_delete_sq;
K
Keith Busch 已提交
1872

K
Keith Busch 已提交
1873
	for (pass = 0; pass < 2; pass++) {
K
Keith Busch 已提交
1874
		int sent = 0, i = queues;
K
Keith Busch 已提交
1875 1876 1877 1878

		reinit_completion(&dev->ioq_wait);
 retry:
		timeout = ADMIN_TIMEOUT;
1879 1880
		for (; i > 0; i--, sent++)
			if (nvme_delete_queue(dev->queues[i], opcode))
K
Keith Busch 已提交
1881
				break;
1882

K
Keith Busch 已提交
1883 1884 1885 1886 1887 1888 1889 1890 1891
		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
1892 1893
}

1894 1895 1896 1897 1898 1899
/*
 * 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.
 */
1900
static int nvme_dev_add(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1901
{
1902
	if (!dev->ctrl.tagset) {
1903 1904 1905 1906 1907
		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
1908
				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
1909 1910 1911
		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
1912

1913 1914
		if (blk_mq_alloc_tag_set(&dev->tagset))
			return 0;
1915
		dev->ctrl.tagset = &dev->tagset;
1916 1917

		nvme_dbbuf_set(dev);
1918 1919 1920 1921 1922
	} 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);
1923
	}
1924

K
Keith Busch 已提交
1925
	return 0;
M
Matthew Wilcox 已提交
1926 1927
}

1928
static int nvme_pci_enable(struct nvme_dev *dev)
1929
{
1930
	int result = -ENOMEM;
1931
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1932 1933 1934 1935 1936 1937

	if (pci_enable_device_mem(pdev))
		return result;

	pci_set_master(pdev);

1938 1939
	if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
	    dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
1940
		goto disable;
1941

1942
	if (readl(dev->bar + NVME_REG_CSTS) == -1) {
K
Keith Busch 已提交
1943
		result = -ENODEV;
1944
		goto disable;
K
Keith Busch 已提交
1945
	}
1946 1947

	/*
1948 1949 1950
	 * 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.
1951
	 */
1952 1953 1954
	result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
	if (result < 0)
		return result;
1955

1956
	dev->ctrl.cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
1957

1958
	dev->q_depth = min_t(int, NVME_CAP_MQES(dev->ctrl.cap) + 1,
1959
				io_queue_depth);
1960
	dev->db_stride = 1 << NVME_CAP_STRIDE(dev->ctrl.cap);
1961
	dev->dbs = dev->bar + 4096;
1962 1963 1964 1965 1966 1967 1968

	/*
	 * 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;
1969 1970
		dev_warn(dev->ctrl.device, "detected Apple NVMe controller, "
			"set queue depth=%u to work around controller resets\n",
1971
			dev->q_depth);
1972 1973
	} else if (pdev->vendor == PCI_VENDOR_ID_SAMSUNG &&
		   (pdev->device == 0xa821 || pdev->device == 0xa822) &&
1974
		   NVME_CAP_MQES(dev->ctrl.cap) == 0) {
1975 1976 1977
		dev->q_depth = 64;
		dev_err(dev->ctrl.device, "detected PM1725 NVMe controller, "
                        "set queue depth=%u\n", dev->q_depth);
1978 1979
	}

1980 1981
	/*
	 * CMBs can currently only exist on >=1.2 PCIe devices. We only
1982 1983 1984
	 * populate sysfs if a CMB is implemented. Since nvme_dev_attrs_group
	 * has no name we can pass NULL as final argument to
	 * sysfs_add_file_to_group.
1985 1986
	 */

1987
	if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) {
1988
		dev->cmb = nvme_map_cmb(dev);
1989
		if (dev->cmb) {
1990 1991
			if (sysfs_add_file_to_group(&dev->ctrl.device->kobj,
						    &dev_attr_cmb.attr, NULL))
1992
				dev_warn(dev->ctrl.device,
1993 1994 1995 1996
					 "failed to add sysfs attribute for CMB\n");
		}
	}

K
Keith Busch 已提交
1997 1998
	pci_enable_pcie_error_reporting(pdev);
	pci_save_state(pdev);
1999 2000 2001 2002 2003 2004 2005 2006
	return 0;

 disable:
	pci_disable_device(pdev);
	return result;
}

static void nvme_dev_unmap(struct nvme_dev *dev)
2007 2008 2009
{
	if (dev->bar)
		iounmap(dev->bar);
2010
	pci_release_mem_regions(to_pci_dev(dev->dev));
2011 2012 2013
}

static void nvme_pci_disable(struct nvme_dev *dev)
2014
{
2015 2016
	struct pci_dev *pdev = to_pci_dev(dev->dev);

2017
	nvme_release_cmb(dev);
2018
	pci_free_irq_vectors(pdev);
2019

K
Keith Busch 已提交
2020 2021
	if (pci_is_enabled(pdev)) {
		pci_disable_pcie_error_reporting(pdev);
2022
		pci_disable_device(pdev);
K
Keith Busch 已提交
2023 2024 2025
	}
}

2026
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
M
Matthew Wilcox 已提交
2027
{
2028
	int i, queues;
K
Keith Busch 已提交
2029 2030
	bool dead = true;
	struct pci_dev *pdev = to_pci_dev(dev->dev);
2031

2032
	mutex_lock(&dev->shutdown_lock);
K
Keith Busch 已提交
2033 2034 2035
	if (pci_is_enabled(pdev)) {
		u32 csts = readl(dev->bar + NVME_REG_CSTS);

K
Keith Busch 已提交
2036 2037
		if (dev->ctrl.state == NVME_CTRL_LIVE ||
		    dev->ctrl.state == NVME_CTRL_RESETTING)
K
Keith Busch 已提交
2038 2039 2040
			nvme_start_freeze(&dev->ctrl);
		dead = !!((csts & NVME_CSTS_CFS) || !(csts & NVME_CSTS_RDY) ||
			pdev->error_state  != pci_channel_io_normal);
2041
	}
2042

K
Keith Busch 已提交
2043 2044 2045 2046
	/*
	 * Give the controller a chance to complete all entered requests if
	 * doing a safe shutdown.
	 */
2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060
	if (!dead) {
		if (shutdown)
			nvme_wait_freeze_timeout(&dev->ctrl, NVME_IO_TIMEOUT);

		/*
		 * If the controller is still alive tell it to stop using the
		 * host memory buffer.  In theory the shutdown / reset should
		 * make sure that it doesn't access the host memoery anymore,
		 * but I'd rather be safe than sorry..
		 */
		if (dev->host_mem_descs)
			nvme_set_host_mem(dev, 0);

	}
K
Keith Busch 已提交
2061 2062
	nvme_stop_queues(&dev->ctrl);

2063
	queues = dev->online_queues - 1;
2064
	for (i = dev->ctrl.queue_count - 1; i > 0; i--)
2065 2066
		nvme_suspend_queue(dev->queues[i]);

K
Keith Busch 已提交
2067
	if (dead) {
2068 2069 2070 2071
		/* A device might become IO incapable very soon during
		 * probe, before the admin queue is configured. Thus,
		 * queue_count can be 0 here.
		 */
2072
		if (dev->ctrl.queue_count)
2073
			nvme_suspend_queue(dev->queues[0]);
K
Keith Busch 已提交
2074
	} else {
2075
		nvme_disable_io_queues(dev, queues);
2076
		nvme_disable_admin_queue(dev, shutdown);
K
Keith Busch 已提交
2077
	}
2078
	nvme_pci_disable(dev);
2079

2080 2081
	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);
K
Keith Busch 已提交
2082 2083 2084 2085 2086 2087 2088 2089

	/*
	 * The driver will not be starting up queues again if shutting down so
	 * must flush all entered requests to their failed completion to avoid
	 * deadlocking blk-mq hot-cpu notifier.
	 */
	if (shutdown)
		nvme_start_queues(&dev->ctrl);
2090
	mutex_unlock(&dev->shutdown_lock);
M
Matthew Wilcox 已提交
2091 2092
}

M
Matthew Wilcox 已提交
2093 2094
static int nvme_setup_prp_pools(struct nvme_dev *dev)
{
2095
	dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
M
Matthew Wilcox 已提交
2096 2097 2098 2099
						PAGE_SIZE, PAGE_SIZE, 0);
	if (!dev->prp_page_pool)
		return -ENOMEM;

2100
	/* Optimisation for I/Os between 4k and 128k */
2101
	dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
2102 2103 2104 2105 2106
						256, 256, 0);
	if (!dev->prp_small_pool) {
		dma_pool_destroy(dev->prp_page_pool);
		return -ENOMEM;
	}
M
Matthew Wilcox 已提交
2107 2108 2109 2110 2111 2112
	return 0;
}

static void nvme_release_prp_pools(struct nvme_dev *dev)
{
	dma_pool_destroy(dev->prp_page_pool);
2113
	dma_pool_destroy(dev->prp_small_pool);
M
Matthew Wilcox 已提交
2114 2115
}

2116
static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
2117
{
2118
	struct nvme_dev *dev = to_nvme_dev(ctrl);
2119

2120
	nvme_dbbuf_dma_free(dev);
2121
	put_device(dev->dev);
2122 2123
	if (dev->tagset.tags)
		blk_mq_free_tag_set(&dev->tagset);
2124 2125
	if (dev->ctrl.admin_q)
		blk_put_queue(dev->ctrl.admin_q);
2126
	kfree(dev->queues);
2127
	free_opal_dev(dev->ctrl.opal_dev);
2128 2129 2130
	kfree(dev);
}

2131 2132
static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status)
{
2133
	dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status);
2134 2135

	kref_get(&dev->ctrl.kref);
2136
	nvme_dev_disable(dev, false);
2137 2138 2139 2140
	if (!schedule_work(&dev->remove_work))
		nvme_put_ctrl(&dev->ctrl);
}

2141
static void nvme_reset_work(struct work_struct *work)
2142
{
2143 2144
	struct nvme_dev *dev =
		container_of(work, struct nvme_dev, ctrl.reset_work);
2145
	bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL);
2146
	int result = -ENODEV;
2147

2148
	if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING))
2149
		goto out;
2150

2151 2152 2153 2154
	/*
	 * If we're called to reset a live controller first shut it down before
	 * moving on.
	 */
2155
	if (dev->ctrl.ctrl_config & NVME_CC_ENABLE)
2156
		nvme_dev_disable(dev, false);
2157

2158
	result = nvme_pci_enable(dev);
2159
	if (result)
2160
		goto out;
2161

2162
	result = nvme_pci_configure_admin_queue(dev);
2163
	if (result)
2164
		goto out;
2165

K
Keith Busch 已提交
2166 2167
	result = nvme_alloc_admin_tags(dev);
	if (result)
2168
		goto out;
2169

2170 2171
	result = nvme_init_identify(&dev->ctrl);
	if (result)
2172
		goto out;
2173

2174 2175 2176 2177 2178 2179 2180 2181 2182
	if (dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) {
		if (!dev->ctrl.opal_dev)
			dev->ctrl.opal_dev =
				init_opal_dev(&dev->ctrl, &nvme_sec_submit);
		else if (was_suspend)
			opal_unlock_from_suspend(dev->ctrl.opal_dev);
	} else {
		free_opal_dev(dev->ctrl.opal_dev);
		dev->ctrl.opal_dev = NULL;
2183
	}
2184

2185 2186 2187 2188 2189 2190 2191
	if (dev->ctrl.oacs & NVME_CTRL_OACS_DBBUF_SUPP) {
		result = nvme_dbbuf_dma_alloc(dev);
		if (result)
			dev_warn(dev->dev,
				 "unable to allocate dma for dbbuf\n");
	}

2192 2193 2194 2195 2196
	if (dev->ctrl.hmpre) {
		result = nvme_setup_host_mem(dev);
		if (result < 0)
			goto out;
	}
2197

2198
	result = nvme_setup_io_queues(dev);
2199
	if (result)
2200
		goto out;
2201

2202 2203 2204 2205
	/*
	 * Keep the controller around but remove all namespaces if we don't have
	 * any working I/O queue.
	 */
2206
	if (dev->online_queues < 2) {
2207
		dev_warn(dev->ctrl.device, "IO queues not created\n");
2208
		nvme_kill_queues(&dev->ctrl);
2209
		nvme_remove_namespaces(&dev->ctrl);
2210
	} else {
2211
		nvme_start_queues(&dev->ctrl);
K
Keith Busch 已提交
2212
		nvme_wait_freeze(&dev->ctrl);
2213
		nvme_dev_add(dev);
K
Keith Busch 已提交
2214
		nvme_unfreeze(&dev->ctrl);
2215 2216
	}

2217 2218 2219 2220
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) {
		dev_warn(dev->ctrl.device, "failed to mark controller live\n");
		goto out;
	}
2221

2222
	nvme_start_ctrl(&dev->ctrl);
2223
	return;
2224

2225
 out:
2226
	nvme_remove_dead_ctrl(dev, result);
2227 2228
}

2229
static void nvme_remove_dead_ctrl_work(struct work_struct *work)
K
Keith Busch 已提交
2230
{
2231
	struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
2232
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
2233

2234
	nvme_kill_queues(&dev->ctrl);
K
Keith Busch 已提交
2235
	if (pci_get_drvdata(pdev))
K
Keith Busch 已提交
2236
		device_release_driver(&pdev->dev);
2237
	nvme_put_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2238 2239
}

2240
static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
T
Tejun Heo 已提交
2241
{
2242
	*val = readl(to_nvme_dev(ctrl)->bar + off);
2243
	return 0;
T
Tejun Heo 已提交
2244 2245
}

2246
static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
2247
{
2248 2249 2250
	writel(val, to_nvme_dev(ctrl)->bar + off);
	return 0;
}
2251

2252 2253 2254 2255
static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
{
	*val = readq(to_nvme_dev(ctrl)->bar + off);
	return 0;
2256 2257
}

2258
static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
M
Ming Lin 已提交
2259
	.name			= "pcie",
2260
	.module			= THIS_MODULE,
2261
	.flags			= NVME_F_METADATA_SUPPORTED,
2262
	.reg_read32		= nvme_pci_reg_read32,
2263
	.reg_write32		= nvme_pci_reg_write32,
2264
	.reg_read64		= nvme_pci_reg_read64,
2265
	.free_ctrl		= nvme_pci_free_ctrl,
2266
	.submit_async_event	= nvme_pci_submit_async_event,
2267
};
2268

2269 2270 2271 2272
static int nvme_dev_map(struct nvme_dev *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev->dev);

2273
	if (pci_request_mem_regions(pdev, "nvme"))
2274 2275
		return -ENODEV;

2276
	if (nvme_remap_bar(dev, NVME_REG_DBS + 4096))
2277 2278
		goto release;

M
Max Gurtovoy 已提交
2279
	return 0;
2280
  release:
M
Max Gurtovoy 已提交
2281 2282
	pci_release_mem_regions(pdev);
	return -ENODEV;
2283 2284
}

2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304
static unsigned long check_dell_samsung_bug(struct pci_dev *pdev)
{
	if (pdev->vendor == 0x144d && pdev->device == 0xa802) {
		/*
		 * Several Samsung devices seem to drop off the PCIe bus
		 * randomly when APST is on and uses the deepest sleep state.
		 * This has been observed on a Samsung "SM951 NVMe SAMSUNG
		 * 256GB", a "PM951 NVMe SAMSUNG 512GB", and a "Samsung SSD
		 * 950 PRO 256GB", but it seems to be restricted to two Dell
		 * laptops.
		 */
		if (dmi_match(DMI_SYS_VENDOR, "Dell Inc.") &&
		    (dmi_match(DMI_PRODUCT_NAME, "XPS 15 9550") ||
		     dmi_match(DMI_PRODUCT_NAME, "Precision 5510")))
			return NVME_QUIRK_NO_DEEPEST_PS;
	}

	return 0;
}

2305
static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
M
Matthew Wilcox 已提交
2306
{
M
Matias Bjørling 已提交
2307
	int node, result = -ENOMEM;
M
Matthew Wilcox 已提交
2308
	struct nvme_dev *dev;
2309
	unsigned long quirks = id->driver_data;
M
Matthew Wilcox 已提交
2310

M
Matias Bjørling 已提交
2311 2312
	node = dev_to_node(&pdev->dev);
	if (node == NUMA_NO_NODE)
2313
		set_dev_node(&pdev->dev, first_memory_node);
M
Matias Bjørling 已提交
2314 2315

	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2316 2317
	if (!dev)
		return -ENOMEM;
M
Matias Bjørling 已提交
2318 2319
	dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2320 2321 2322
	if (!dev->queues)
		goto free;

2323
	dev->dev = get_device(&pdev->dev);
K
Keith Busch 已提交
2324
	pci_set_drvdata(pdev, dev);
2325

2326 2327
	result = nvme_dev_map(dev);
	if (result)
2328
		goto put_pci;
2329

2330
	INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work);
2331
	INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
2332
	mutex_init(&dev->shutdown_lock);
K
Keith Busch 已提交
2333
	init_completion(&dev->ioq_wait);
M
Matthew Wilcox 已提交
2334

M
Matthew Wilcox 已提交
2335 2336
	result = nvme_setup_prp_pools(dev);
	if (result)
2337
		goto unmap;
2338

2339 2340
	quirks |= check_dell_samsung_bug(pdev);

2341
	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
2342
			quirks);
2343
	if (result)
2344
		goto release_pools;
2345

2346
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING);
2347 2348
	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));

2349
	queue_work(nvme_wq, &dev->ctrl.reset_work);
M
Matthew Wilcox 已提交
2350 2351
	return 0;

2352
 release_pools:
M
Matthew Wilcox 已提交
2353
	nvme_release_prp_pools(dev);
2354 2355
 unmap:
	nvme_dev_unmap(dev);
K
Keith Busch 已提交
2356
 put_pci:
2357
	put_device(dev->dev);
M
Matthew Wilcox 已提交
2358 2359 2360 2361 2362 2363
 free:
	kfree(dev->queues);
	kfree(dev);
	return result;
}

2364
static void nvme_reset_prepare(struct pci_dev *pdev)
2365
{
K
Keith Busch 已提交
2366
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2367
	nvme_dev_disable(dev, false);
2368
}
2369

2370 2371
static void nvme_reset_done(struct pci_dev *pdev)
{
2372 2373
	struct nvme_dev *dev = pci_get_drvdata(pdev);
	nvme_reset_ctrl(&dev->ctrl);
2374 2375
}

2376 2377 2378
static void nvme_shutdown(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2379
	nvme_dev_disable(dev, true);
2380 2381
}

2382 2383 2384 2385 2386
/*
 * 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.
 */
2387
static void nvme_remove(struct pci_dev *pdev)
M
Matthew Wilcox 已提交
2388 2389
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
K
Keith Busch 已提交
2390

2391 2392
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING);

2393
	cancel_work_sync(&dev->ctrl.reset_work);
K
Keith Busch 已提交
2394
	pci_set_drvdata(pdev, NULL);
2395

2396
	if (!pci_device_is_present(pdev)) {
2397
		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
2398 2399
		nvme_dev_disable(dev, false);
	}
2400

2401
	flush_work(&dev->ctrl.reset_work);
2402 2403
	nvme_stop_ctrl(&dev->ctrl);
	nvme_remove_namespaces(&dev->ctrl);
2404
	nvme_dev_disable(dev, true);
2405
	nvme_free_host_mem(dev);
M
Matias Bjørling 已提交
2406
	nvme_dev_remove_admin(dev);
2407
	nvme_free_queues(dev, 0);
2408
	nvme_uninit_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2409
	nvme_release_prp_pools(dev);
2410
	nvme_dev_unmap(dev);
2411
	nvme_put_ctrl(&dev->ctrl);
M
Matthew Wilcox 已提交
2412 2413
}

K
Keith Busch 已提交
2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431
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;
}

2432
#ifdef CONFIG_PM_SLEEP
2433 2434 2435 2436 2437
static int nvme_suspend(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

2438
	nvme_dev_disable(ndev, true);
2439 2440 2441 2442 2443 2444 2445 2446
	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);

2447
	nvme_reset_ctrl(&ndev->ctrl);
K
Keith Busch 已提交
2448
	return 0;
2449
}
2450
#endif
2451 2452

static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
M
Matthew Wilcox 已提交
2453

K
Keith Busch 已提交
2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467
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
2468 2469
		dev_warn(dev->ctrl.device,
			"frozen state error detected, reset controller\n");
2470
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2471 2472
		return PCI_ERS_RESULT_NEED_RESET;
	case pci_channel_io_perm_failure:
K
Keith Busch 已提交
2473 2474
		dev_warn(dev->ctrl.device,
			"failure state error detected, request disconnect\n");
K
Keith Busch 已提交
2475 2476 2477 2478 2479 2480 2481 2482 2483
		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);

2484
	dev_info(dev->ctrl.device, "restart after slot reset\n");
K
Keith Busch 已提交
2485
	pci_restore_state(pdev);
2486
	nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2487 2488 2489 2490 2491 2492 2493 2494
	return PCI_ERS_RESULT_RECOVERED;
}

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

2495
static const struct pci_error_handlers nvme_err_handler = {
M
Matthew Wilcox 已提交
2496 2497 2498
	.error_detected	= nvme_error_detected,
	.slot_reset	= nvme_slot_reset,
	.resume		= nvme_error_resume,
2499 2500
	.reset_prepare	= nvme_reset_prepare,
	.reset_done	= nvme_reset_done,
M
Matthew Wilcox 已提交
2501 2502
};

2503
static const struct pci_device_id nvme_id_table[] = {
2504
	{ PCI_VDEVICE(INTEL, 0x0953),
2505
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2506
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2507 2508
	{ PCI_VDEVICE(INTEL, 0x0a53),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2509
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2510 2511
	{ PCI_VDEVICE(INTEL, 0x0a54),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2512
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2513 2514 2515
	{ PCI_VDEVICE(INTEL, 0x0a55),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2516 2517
	{ PCI_VDEVICE(INTEL, 0xf1a5),	/* Intel 600P/P3100 */
		.driver_data = NVME_QUIRK_NO_DEEPEST_PS },
2518 2519
	{ PCI_VDEVICE(INTEL, 0x5845),	/* Qemu emulated controller */
		.driver_data = NVME_QUIRK_IDENTIFY_CNS, },
2520 2521
	{ PCI_DEVICE(0x1c58, 0x0003),	/* HGST adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2522 2523
	{ PCI_DEVICE(0x1c5f, 0x0540),	/* Memblaze Pblaze4 adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2524 2525 2526 2527
	{ PCI_DEVICE(0x144d, 0xa821),   /* Samsung PM1725 */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
	{ PCI_DEVICE(0x144d, 0xa822),   /* Samsung PM1725a */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
C
Christoph Hellwig 已提交
2528 2529 2530 2531
	{ PCI_DEVICE(0x1d1d, 0x1f1f),	/* LighNVM qemu device */
		.driver_data = NVME_QUIRK_LIGHTNVM, },
	{ PCI_DEVICE(0x1d1d, 0x2807),	/* CNEX WL */
		.driver_data = NVME_QUIRK_LIGHTNVM, },
M
Matthew Wilcox 已提交
2532
	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
2533
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
2534
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) },
M
Matthew Wilcox 已提交
2535 2536 2537 2538 2539 2540 2541 2542
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, nvme_id_table);

static struct pci_driver nvme_driver = {
	.name		= "nvme",
	.id_table	= nvme_id_table,
	.probe		= nvme_probe,
2543
	.remove		= nvme_remove,
2544
	.shutdown	= nvme_shutdown,
2545 2546 2547
	.driver		= {
		.pm	= &nvme_dev_pm_ops,
	},
K
Keith Busch 已提交
2548
	.sriov_configure = nvme_pci_sriov_configure,
M
Matthew Wilcox 已提交
2549 2550 2551 2552 2553
	.err_handler	= &nvme_err_handler,
};

static int __init nvme_init(void)
{
2554
	return pci_register_driver(&nvme_driver);
M
Matthew Wilcox 已提交
2555 2556 2557 2558 2559
}

static void __exit nvme_exit(void)
{
	pci_unregister_driver(&nvme_driver);
2560
	_nvme_check_size();
M
Matthew Wilcox 已提交
2561 2562 2563 2564
}

MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
MODULE_LICENSE("GPL");
2565
MODULE_VERSION("1.0");
M
Matthew Wilcox 已提交
2566 2567
module_init(nvme_init);
module_exit(nvme_exit);