pci.c 62.4 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>
M
Matthew Wilcox 已提交
27
#include <linux/pci.h>
28
#include <linux/poison.h>
K
Keith Busch 已提交
29
#include <linux/t10-pi.h>
30
#include <linux/timer.h>
M
Matthew Wilcox 已提交
31
#include <linux/types.h>
32
#include <linux/io-64-nonatomic-lo-hi.h>
K
Keith Busch 已提交
33
#include <asm/unaligned.h>
34
#include <linux/sed-opal.h>
35

36 37
#include "nvme.h"

38
#define NVME_Q_DEPTH		1024
J
Jens Axboe 已提交
39
#define NVME_AQ_DEPTH		256
M
Matthew Wilcox 已提交
40 41
#define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
#define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
42

43 44 45 46
/*
 * We handle AEN commands ourselves and don't even let the
 * block layer know about them.
 */
47
#define NVME_AQ_BLKMQ_DEPTH	(NVME_AQ_DEPTH - NVME_NR_AERS)
48

49 50 51
static int use_threaded_interrupts;
module_param(use_threaded_interrupts, int, 0);

52 53 54 55
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");

56 57 58 59 60
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)");

61 62
struct nvme_dev;
struct nvme_queue;
63

J
Jens Axboe 已提交
64
static void nvme_process_cq(struct nvme_queue *nvmeq);
65
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
66

67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
/*
 * 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;
84
	unsigned long bar_mapped_size;
85
	struct work_struct remove_work;
86
	struct mutex shutdown_lock;
87 88 89 90 91
	bool subsystem;
	void __iomem *cmb;
	dma_addr_t cmb_dma_addr;
	u64 cmb_size;
	u32 cmbsz;
92
	u32 cmbloc;
93
	struct nvme_ctrl ctrl;
K
Keith Busch 已提交
94
	struct completion ioq_wait;
95 96

	/* shadow doorbell buffer support: */
97 98 99 100
	u32 *dbbuf_dbs;
	dma_addr_t dbbuf_dbs_dma_addr;
	u32 *dbbuf_eis;
	dma_addr_t dbbuf_eis_dma_addr;
101 102 103 104 105 106

	/* host memory buffer support: */
	u64 host_mem_size;
	u32 nr_host_mem_descs;
	struct nvme_host_mem_buf_desc *host_mem_descs;
	void **host_mem_desc_bufs;
K
Keith Busch 已提交
107
};
108

109 110 111 112 113 114 115 116 117 118
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;
}

119 120 121 122 123
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
124 125 126 127 128 129
/*
 * 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
130
	struct nvme_dev *dev;
M
Matthew Wilcox 已提交
131 132
	spinlock_t q_lock;
	struct nvme_command *sq_cmds;
133
	struct nvme_command __iomem *sq_cmds_io;
M
Matthew Wilcox 已提交
134
	volatile struct nvme_completion *cqes;
135
	struct blk_mq_tags **tags;
M
Matthew Wilcox 已提交
136 137 138 139
	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
140
	s16 cq_vector;
M
Matthew Wilcox 已提交
141 142
	u16 sq_tail;
	u16 cq_head;
K
Keith Busch 已提交
143
	u16 qid;
144 145
	u8 cq_phase;
	u8 cqe_seen;
146 147 148 149
	u32 *dbbuf_sq_db;
	u32 *dbbuf_cq_db;
	u32 *dbbuf_sq_ei;
	u32 *dbbuf_cq_ei;
M
Matthew Wilcox 已提交
150 151
};

152 153 154
/*
 * 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
155
 * me express that.  Use nvme_init_iod to ensure there's enough space
156 157 158
 * allocated to store the PRP list.
 */
struct nvme_iod {
159
	struct nvme_request req;
C
Christoph Hellwig 已提交
160 161
	struct nvme_queue *nvmeq;
	int aborted;
162 163 164 165
	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;
166
	struct scatterlist meta_sg; /* metadata requires single contiguous buffer */
C
Christoph Hellwig 已提交
167 168
	struct scatterlist *sg;
	struct scatterlist inline_sg[0];
M
Matthew Wilcox 已提交
169 170 171 172 173 174 175 176 177 178 179 180
};

/*
 * 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);
181
	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
K
Keith Busch 已提交
182
	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
M
Matthew Wilcox 已提交
183
	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
184 185
	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
186
	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
K
Keith Busch 已提交
187
	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
	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)) {
262
		dev_warn(dev->ctrl.device, "unable to set dbbuf\n");
263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293
		/* 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
294 295
}

296 297 298 299
/*
 * Max size of iod being embedded in the request payload
 */
#define NVME_INT_PAGES		2
300
#define NVME_INT_BYTES(dev)	(NVME_INT_PAGES * (dev)->ctrl.page_size)
301 302 303 304 305 306 307 308

/*
 * 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)
{
309 310
	unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size,
				      dev->ctrl.page_size);
311 312 313
	return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
}

C
Christoph Hellwig 已提交
314 315
static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
		unsigned int size, unsigned int nseg)
316
{
C
Christoph Hellwig 已提交
317 318 319
	return sizeof(__le64 *) * nvme_npages(size, dev) +
			sizeof(struct scatterlist) * nseg;
}
320

C
Christoph Hellwig 已提交
321 322 323 324
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);
325 326
}

M
Matias Bjørling 已提交
327 328
static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
				unsigned int hctx_idx)
329
{
M
Matias Bjørling 已提交
330 331 332
	struct nvme_dev *dev = data;
	struct nvme_queue *nvmeq = dev->queues[0];

333 334 335 336
	WARN_ON(hctx_idx != 0);
	WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
	WARN_ON(nvmeq->tags);

M
Matias Bjørling 已提交
337
	hctx->driver_data = nvmeq;
338
	nvmeq->tags = &dev->admin_tagset.tags[0];
M
Matias Bjørling 已提交
339
	return 0;
340 341
}

342 343 344 345 346 347 348
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
349 350
static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
			  unsigned int hctx_idx)
M
Matthew Wilcox 已提交
351
{
M
Matias Bjørling 已提交
352
	struct nvme_dev *dev = data;
353
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
M
Matias Bjørling 已提交
354

355 356
	if (!nvmeq->tags)
		nvmeq->tags = &dev->tagset.tags[hctx_idx];
M
Matthew Wilcox 已提交
357

358
	WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
M
Matias Bjørling 已提交
359 360
	hctx->driver_data = nvmeq;
	return 0;
M
Matthew Wilcox 已提交
361 362
}

363 364
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
365
{
366
	struct nvme_dev *dev = set->driver_data;
C
Christoph Hellwig 已提交
367
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
368 369
	int queue_idx = (set == &dev->tagset) ? hctx_idx + 1 : 0;
	struct nvme_queue *nvmeq = dev->queues[queue_idx];
M
Matias Bjørling 已提交
370 371

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
372
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
373 374 375
	return 0;
}

376 377 378 379 380 381 382
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
383
/**
384
 * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
M
Matthew Wilcox 已提交
385 386 387 388 389
 * @nvmeq: The queue to use
 * @cmd: The command to send
 *
 * Safe to use from interrupt context
 */
390 391
static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
						struct nvme_command *cmd)
M
Matthew Wilcox 已提交
392
{
M
Matias Bjørling 已提交
393 394
	u16 tail = nvmeq->sq_tail;

395 396 397 398 399
	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
400 401
	if (++tail == nvmeq->q_depth)
		tail = 0;
402 403 404
	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
405 406 407
	nvmeq->sq_tail = tail;
}

C
Christoph Hellwig 已提交
408
static __le64 **iod_list(struct request *req)
M
Matthew Wilcox 已提交
409
{
C
Christoph Hellwig 已提交
410
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
411
	return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req));
M
Matthew Wilcox 已提交
412 413
}

414
static blk_status_t nvme_init_iod(struct request *rq, struct nvme_dev *dev)
415
{
C
Christoph Hellwig 已提交
416
	struct nvme_iod *iod = blk_mq_rq_to_pdu(rq);
417
	int nseg = blk_rq_nr_phys_segments(rq);
418
	unsigned int size = blk_rq_payload_bytes(rq);
419

C
Christoph Hellwig 已提交
420 421 422
	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)
423
			return BLK_STS_RESOURCE;
C
Christoph Hellwig 已提交
424 425
	} else {
		iod->sg = iod->inline_sg;
426 427
	}

C
Christoph Hellwig 已提交
428 429 430 431
	iod->aborted = 0;
	iod->npages = -1;
	iod->nents = 0;
	iod->length = size;
K
Keith Busch 已提交
432

433
	return BLK_STS_OK;
434 435
}

C
Christoph Hellwig 已提交
436
static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
437
{
C
Christoph Hellwig 已提交
438
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
439
	const int last_prp = dev->ctrl.page_size / 8 - 1;
440
	int i;
C
Christoph Hellwig 已提交
441
	__le64 **list = iod_list(req);
442 443 444 445 446 447 448 449 450 451
	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;
	}
452

C
Christoph Hellwig 已提交
453 454
	if (iod->sg != iod->inline_sg)
		kfree(iod->sg);
K
Keith Busch 已提交
455 456
}

457
#ifdef CONFIG_BLK_DEV_INTEGRITY
K
Keith Busch 已提交
458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
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);
502
	ts = ns->disk->queue->integrity.tuple_size;
K
Keith Busch 已提交
503 504 505 506 507 508 509 510

	for (i = 0; i < nlb; i++, virt++, phys++) {
		pi = (struct t10_pi_tuple *)p;
		dif_swap(phys, virt, pi);
		p += ts;
	}
	kunmap_atomic(pmap);
}
511 512 513 514 515 516 517 518 519 520 521 522 523
#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

524
static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
525
{
C
Christoph Hellwig 已提交
526
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
527
	struct dma_pool *pool;
528
	int length = blk_rq_payload_bytes(req);
529
	struct scatterlist *sg = iod->sg;
M
Matthew Wilcox 已提交
530 531
	int dma_len = sg_dma_len(sg);
	u64 dma_addr = sg_dma_address(sg);
532
	u32 page_size = dev->ctrl.page_size;
533
	int offset = dma_addr & (page_size - 1);
534
	__le64 *prp_list;
C
Christoph Hellwig 已提交
535
	__le64 **list = iod_list(req);
536
	dma_addr_t prp_dma;
537
	int nprps, i;
M
Matthew Wilcox 已提交
538

539
	length -= (page_size - offset);
M
Matthew Wilcox 已提交
540
	if (length <= 0)
541
		return true;
M
Matthew Wilcox 已提交
542

543
	dma_len -= (page_size - offset);
M
Matthew Wilcox 已提交
544
	if (dma_len) {
545
		dma_addr += (page_size - offset);
M
Matthew Wilcox 已提交
546 547 548 549 550 551
	} else {
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
	}

552
	if (length <= page_size) {
553
		iod->first_dma = dma_addr;
554
		return true;
555 556
	}

557
	nprps = DIV_ROUND_UP(length, page_size);
558 559
	if (nprps <= (256 / 8)) {
		pool = dev->prp_small_pool;
560
		iod->npages = 0;
561 562
	} else {
		pool = dev->prp_page_pool;
563
		iod->npages = 1;
564 565
	}

566
	prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
567
	if (!prp_list) {
568
		iod->first_dma = dma_addr;
569
		iod->npages = -1;
570
		return false;
571
	}
572 573
	list[0] = prp_list;
	iod->first_dma = prp_dma;
574 575
	i = 0;
	for (;;) {
576
		if (i == page_size >> 3) {
577
			__le64 *old_prp_list = prp_list;
578
			prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
579
			if (!prp_list)
580
				return false;
581
			list[iod->npages++] = prp_list;
582 583 584
			prp_list[0] = old_prp_list[i - 1];
			old_prp_list[i - 1] = cpu_to_le64(prp_dma);
			i = 1;
585 586
		}
		prp_list[i++] = cpu_to_le64(dma_addr);
587 588 589
		dma_len -= page_size;
		dma_addr += page_size;
		length -= page_size;
590 591 592 593 594 595 596 597
		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
598 599
	}

600
	return true;
M
Matthew Wilcox 已提交
601 602
}

603
static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req,
604
		struct nvme_command *cmnd)
605
{
C
Christoph Hellwig 已提交
606
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
C
Christoph Hellwig 已提交
607 608 609
	struct request_queue *q = req->q;
	enum dma_data_direction dma_dir = rq_data_dir(req) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE;
610
	blk_status_t ret = BLK_STS_IOERR;
611

612
	sg_init_table(iod->sg, blk_rq_nr_phys_segments(req));
C
Christoph Hellwig 已提交
613 614 615
	iod->nents = blk_rq_map_sg(q, req, iod->sg);
	if (!iod->nents)
		goto out;
616

617
	ret = BLK_STS_RESOURCE;
618 619
	if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir,
				DMA_ATTR_NO_WARN))
C
Christoph Hellwig 已提交
620
		goto out;
621

622
	if (!nvme_setup_prps(dev, req))
C
Christoph Hellwig 已提交
623
		goto out_unmap;
624

625
	ret = BLK_STS_IOERR;
C
Christoph Hellwig 已提交
626 627 628
	if (blk_integrity_rq(req)) {
		if (blk_rq_count_integrity_sg(q, req->bio) != 1)
			goto out_unmap;
629

630 631
		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
632
			goto out_unmap;
633

C
Christoph Hellwig 已提交
634 635
		if (rq_data_dir(req))
			nvme_dif_remap(req, nvme_dif_prep);
636

637
		if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir))
C
Christoph Hellwig 已提交
638
			goto out_unmap;
639
	}
M
Matthew Wilcox 已提交
640

641 642
	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
643
	if (blk_integrity_rq(req))
644
		cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
645
	return BLK_STS_OK;
M
Matthew Wilcox 已提交
646

C
Christoph Hellwig 已提交
647 648 649 650
out_unmap:
	dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
out:
	return ret;
M
Matthew Wilcox 已提交
651 652
}

C
Christoph Hellwig 已提交
653
static void nvme_unmap_data(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
654
{
C
Christoph Hellwig 已提交
655
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
656 657 658 659 660 661 662 663
	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);
664
			dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir);
K
Keith Busch 已提交
665
		}
666
	}
K
Keith Busch 已提交
667

668
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
669
	nvme_free_iod(dev, req);
670
}
M
Matthew Wilcox 已提交
671

672 673 674
/*
 * NOTE: ns is NULL when called on the admin queue.
 */
675
static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
M
Matias Bjørling 已提交
676
			 const struct blk_mq_queue_data *bd)
677
{
M
Matias Bjørling 已提交
678 679
	struct nvme_ns *ns = hctx->queue->queuedata;
	struct nvme_queue *nvmeq = hctx->driver_data;
680
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
681
	struct request *req = bd->rq;
C
Christoph Hellwig 已提交
682
	struct nvme_command cmnd;
683
	blk_status_t ret;
K
Keith Busch 已提交
684

685
	ret = nvme_setup_cmd(ns, req, &cmnd);
686
	if (ret)
C
Christoph Hellwig 已提交
687
		return ret;
M
Matias Bjørling 已提交
688

689
	ret = nvme_init_iod(req, dev);
690
	if (ret)
691
		goto out_free_cmd;
M
Matias Bjørling 已提交
692

693
	if (blk_rq_nr_phys_segments(req)) {
694
		ret = nvme_map_data(dev, req, &cmnd);
695 696 697
		if (ret)
			goto out_cleanup_iod;
	}
M
Matias Bjørling 已提交
698

699
	blk_mq_start_request(req);
M
Matias Bjørling 已提交
700

C
Christoph Hellwig 已提交
701
	spin_lock_irq(&nvmeq->q_lock);
702
	if (unlikely(nvmeq->cq_vector < 0)) {
703
		ret = BLK_STS_IOERR;
704
		spin_unlock_irq(&nvmeq->q_lock);
705
		goto out_cleanup_iod;
706
	}
C
Christoph Hellwig 已提交
707
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matias Bjørling 已提交
708 709
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
710
	return BLK_STS_OK;
711
out_cleanup_iod:
C
Christoph Hellwig 已提交
712
	nvme_free_iod(dev, req);
713 714
out_free_cmd:
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
715
	return ret;
M
Matthew Wilcox 已提交
716
}
K
Keith Busch 已提交
717

718
static void nvme_pci_complete_rq(struct request *req)
719
{
C
Christoph Hellwig 已提交
720
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
721

722 723
	nvme_unmap_data(iod->nvmeq->dev, req);
	nvme_complete_rq(req);
M
Matthew Wilcox 已提交
724 725
}

726 727 728 729 730 731 732
/* 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
733
static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag)
M
Matthew Wilcox 已提交
734
{
M
Matthew Wilcox 已提交
735
	u16 head, phase;
M
Matthew Wilcox 已提交
736 737

	head = nvmeq->cq_head;
M
Matthew Wilcox 已提交
738
	phase = nvmeq->cq_phase;
M
Matthew Wilcox 已提交
739

740
	while (nvme_cqe_valid(nvmeq, head, phase)) {
M
Matthew Wilcox 已提交
741
		struct nvme_completion cqe = nvmeq->cqes[head];
742
		struct request *req;
743

M
Matthew Wilcox 已提交
744 745
		if (++head == nvmeq->q_depth) {
			head = 0;
M
Matthew Wilcox 已提交
746
			phase = !phase;
M
Matthew Wilcox 已提交
747
		}
748

J
Jens Axboe 已提交
749 750
		if (tag && *tag == cqe.command_id)
			*tag = -1;
751

752
		if (unlikely(cqe.command_id >= nvmeq->q_depth)) {
753
			dev_warn(nvmeq->dev->ctrl.device,
754 755 756 757 758
				"invalid id %d completed on queue %d\n",
				cqe.command_id, le16_to_cpu(cqe.sq_id));
			continue;
		}

759 760 761 762 763 764 765 766
		/*
		 * 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)) {
767 768
			nvme_complete_async_event(&nvmeq->dev->ctrl,
					cqe.status, &cqe.result);
769 770 771
			continue;
		}

772
		req = blk_mq_tag_to_rq(*nvmeq->tags, cqe.command_id);
773
		nvme_end_request(req, cqe.status, cqe.result);
M
Matthew Wilcox 已提交
774 775
	}

M
Matthew Wilcox 已提交
776
	if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
J
Jens Axboe 已提交
777
		return;
M
Matthew Wilcox 已提交
778

779
	if (likely(nvmeq->cq_vector >= 0))
780 781 782
		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);
M
Matthew Wilcox 已提交
783
	nvmeq->cq_head = head;
M
Matthew Wilcox 已提交
784
	nvmeq->cq_phase = phase;
M
Matthew Wilcox 已提交
785

786
	nvmeq->cqe_seen = 1;
J
Jens Axboe 已提交
787 788 789 790 791
}

static void nvme_process_cq(struct nvme_queue *nvmeq)
{
	__nvme_process_cq(nvmeq, NULL);
M
Matthew Wilcox 已提交
792 793 794
}

static irqreturn_t nvme_irq(int irq, void *data)
795 796 797 798
{
	irqreturn_t result;
	struct nvme_queue *nvmeq = data;
	spin_lock(&nvmeq->q_lock);
799 800 801
	nvme_process_cq(nvmeq);
	result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
	nvmeq->cqe_seen = 0;
802 803 804 805 806 807 808
	spin_unlock(&nvmeq->q_lock);
	return result;
}

static irqreturn_t nvme_irq_check(int irq, void *data)
{
	struct nvme_queue *nvmeq = data;
809 810 811
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return IRQ_WAKE_THREAD;
	return IRQ_NONE;
812 813
}

K
Keith Busch 已提交
814
static int __nvme_poll(struct nvme_queue *nvmeq, unsigned int tag)
J
Jens Axboe 已提交
815
{
816
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) {
J
Jens Axboe 已提交
817 818 819 820 821 822 823 824 825 826 827
		spin_lock_irq(&nvmeq->q_lock);
		__nvme_process_cq(nvmeq, &tag);
		spin_unlock_irq(&nvmeq->q_lock);

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

	return 0;
}

K
Keith Busch 已提交
828 829 830 831 832 833 834
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);
}

835
static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx)
M
Matthew Wilcox 已提交
836
{
837
	struct nvme_dev *dev = to_nvme_dev(ctrl);
838
	struct nvme_queue *nvmeq = dev->queues[0];
M
Matias Bjørling 已提交
839
	struct nvme_command c;
M
Matthew Wilcox 已提交
840

M
Matias Bjørling 已提交
841 842
	memset(&c, 0, sizeof(c));
	c.common.opcode = nvme_admin_async_event;
843
	c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx;
844

845
	spin_lock_irq(&nvmeq->q_lock);
846
	__nvme_submit_cmd(nvmeq, &c);
847
	spin_unlock_irq(&nvmeq->q_lock);
848 849
}

M
Matthew Wilcox 已提交
850
static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
851
{
M
Matthew Wilcox 已提交
852 853 854 855 856 857
	struct nvme_command c;

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

858
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
859 860 861 862 863 864 865 866
}

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;

867 868 869 870
	/*
	 * 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
871 872 873 874 875 876 877 878
	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);

879
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
880 881 882 883 884 885
}

static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
						struct nvme_queue *nvmeq)
{
	struct nvme_command c;
886
	int flags = NVME_QUEUE_PHYS_CONTIG;
M
Matthew Wilcox 已提交
887

888 889 890 891
	/*
	 * 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
892 893 894 895 896 897 898 899
	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);

900
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
901 902 903 904 905 906 907 908 909 910 911 912
}

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);
}

913
static void abort_endio(struct request *req, blk_status_t error)
914
{
C
Christoph Hellwig 已提交
915 916
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
917

918 919
	dev_warn(nvmeq->dev->ctrl.device,
		 "Abort status: 0x%x", nvme_req(req)->status);
920 921
	atomic_inc(&nvmeq->dev->ctrl.abort_limit);
	blk_mq_free_request(req);
922 923
}

K
Keith Busch 已提交
924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968
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);
}

969
static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
K
Keith Busch 已提交
970
{
C
Christoph Hellwig 已提交
971 972
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
K
Keith Busch 已提交
973
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
974 975
	struct request *abort_req;
	struct nvme_command cmd;
K
Keith Busch 已提交
976 977 978 979 980 981 982 983
	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);
984
		nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
985 986
		return BLK_EH_HANDLED;
	}
K
Keith Busch 已提交
987

K
Keith Busch 已提交
988 989 990 991 992 993 994 995 996 997
	/*
	 * 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;
	}

998
	/*
999 1000 1001 1002 1003
	 * 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.
	 */
1004
	if (dev->ctrl.state == NVME_CTRL_RESETTING) {
1005
		dev_warn(dev->ctrl.device,
1006 1007
			 "I/O %d QID %d timeout, disable controller\n",
			 req->tag, nvmeq->qid);
1008
		nvme_dev_disable(dev, false);
1009
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1010
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1011 1012
	}

1013 1014 1015 1016
	/*
 	 * 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.
1017
	 */
C
Christoph Hellwig 已提交
1018
	if (!nvmeq->qid || iod->aborted) {
1019
		dev_warn(dev->ctrl.device,
1020 1021
			 "I/O %d QID %d timeout, reset controller\n",
			 req->tag, nvmeq->qid);
1022
		nvme_dev_disable(dev, false);
1023
		nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
1024

1025 1026 1027 1028
		/*
		 * Mark the request as handled, since the inline shutdown
		 * forces all outstanding requests to complete.
		 */
1029
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1030
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1031 1032
	}

1033
	if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) {
1034
		atomic_inc(&dev->ctrl.abort_limit);
1035
		return BLK_EH_RESET_TIMER;
1036
	}
1037
	iod->aborted = 1;
M
Matias Bjørling 已提交
1038

K
Keith Busch 已提交
1039 1040
	memset(&cmd, 0, sizeof(cmd));
	cmd.abort.opcode = nvme_admin_abort_cmd;
M
Matias Bjørling 已提交
1041
	cmd.abort.cid = req->tag;
K
Keith Busch 已提交
1042 1043
	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);

1044 1045 1046
	dev_warn(nvmeq->dev->ctrl.device,
		"I/O %d QID %d timeout, aborting\n",
		 req->tag, nvmeq->qid);
1047 1048

	abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
1049
			BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
1050 1051 1052 1053 1054 1055 1056 1057
	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
1058

1059 1060 1061 1062 1063 1064
	/*
	 * 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
1065 1066
}

M
Matias Bjørling 已提交
1067 1068
static void nvme_free_queue(struct nvme_queue *nvmeq)
{
1069 1070
	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
1071 1072
	if (nvmeq->sq_cmds)
		dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
1073 1074 1075 1076
					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
	kfree(nvmeq);
}

1077
static void nvme_free_queues(struct nvme_dev *dev, int lowest)
1078 1079 1080
{
	int i;

1081
	for (i = dev->queue_count - 1; i >= lowest; i--) {
M
Matias Bjørling 已提交
1082
		struct nvme_queue *nvmeq = dev->queues[i];
1083
		dev->queue_count--;
M
Matias Bjørling 已提交
1084
		dev->queues[i] = NULL;
K
Keith Busch 已提交
1085
		nvme_free_queue(nvmeq);
1086
	}
1087 1088
}

K
Keith Busch 已提交
1089 1090 1091 1092 1093
/**
 * 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
1094
{
K
Keith Busch 已提交
1095
	int vector;
M
Matthew Wilcox 已提交
1096

1097
	spin_lock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
1098 1099 1100 1101
	if (nvmeq->cq_vector == -1) {
		spin_unlock_irq(&nvmeq->q_lock);
		return 1;
	}
1102
	vector = nvmeq->cq_vector;
K
Keith Busch 已提交
1103
	nvmeq->dev->online_queues--;
K
Keith Busch 已提交
1104
	nvmeq->cq_vector = -1;
1105 1106
	spin_unlock_irq(&nvmeq->q_lock);

1107
	if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
1108
		blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q);
1109

1110
	pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq);
M
Matthew Wilcox 已提交
1111

K
Keith Busch 已提交
1112 1113
	return 0;
}
M
Matthew Wilcox 已提交
1114

1115
static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
K
Keith Busch 已提交
1116
{
1117
	struct nvme_queue *nvmeq = dev->queues[0];
K
Keith Busch 已提交
1118 1119 1120 1121 1122 1123

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

1124 1125 1126 1127 1128
	if (shutdown)
		nvme_shutdown_ctrl(&dev->ctrl);
	else
		nvme_disable_ctrl(&dev->ctrl, lo_hi_readq(
						dev->bar + NVME_REG_CAP));
1129 1130 1131 1132

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1133 1134
}

1135 1136 1137 1138
static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
				int entry_size)
{
	int q_depth = dev->q_depth;
1139 1140
	unsigned q_size_aligned = roundup(q_depth * entry_size,
					  dev->ctrl.page_size);
1141 1142

	if (q_size_aligned * nr_io_queues > dev->cmb_size) {
1143
		u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1144
		mem_per_q = round_down(mem_per_q, dev->ctrl.page_size);
1145
		q_depth = div_u64(mem_per_q, entry_size);
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162

		/*
		 * 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)) {
1163 1164
		unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth),
						      dev->ctrl.page_size);
1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
		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
1177
static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
1178
							int depth, int node)
M
Matthew Wilcox 已提交
1179
{
1180 1181
	struct nvme_queue *nvmeq = kzalloc_node(sizeof(*nvmeq), GFP_KERNEL,
							node);
M
Matthew Wilcox 已提交
1182 1183 1184
	if (!nvmeq)
		return NULL;

1185
	nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
J
Joe Perches 已提交
1186
					  &nvmeq->cq_dma_addr, GFP_KERNEL);
M
Matthew Wilcox 已提交
1187 1188 1189
	if (!nvmeq->cqes)
		goto free_nvmeq;

1190
	if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
M
Matthew Wilcox 已提交
1191 1192
		goto free_cqdma;

1193
	nvmeq->q_dmadev = dev->dev;
M
Matthew Wilcox 已提交
1194
	nvmeq->dev = dev;
M
Matthew Wilcox 已提交
1195 1196
	spin_lock_init(&nvmeq->q_lock);
	nvmeq->cq_head = 0;
M
Matthew Wilcox 已提交
1197
	nvmeq->cq_phase = 1;
1198
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
M
Matthew Wilcox 已提交
1199
	nvmeq->q_depth = depth;
K
Keith Busch 已提交
1200
	nvmeq->qid = qid;
1201
	nvmeq->cq_vector = -1;
M
Matias Bjørling 已提交
1202
	dev->queues[qid] = nvmeq;
1203 1204
	dev->queue_count++;

M
Matthew Wilcox 已提交
1205 1206 1207
	return nvmeq;

 free_cqdma:
1208
	dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
M
Matthew Wilcox 已提交
1209 1210 1211 1212 1213 1214
							nvmeq->cq_dma_addr);
 free_nvmeq:
	kfree(nvmeq);
	return NULL;
}

1215
static int queue_request_irq(struct nvme_queue *nvmeq)
1216
{
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
	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);
	}
1227 1228
}

1229
static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
M
Matthew Wilcox 已提交
1230
{
1231
	struct nvme_dev *dev = nvmeq->dev;
M
Matthew Wilcox 已提交
1232

1233
	spin_lock_irq(&nvmeq->q_lock);
1234 1235 1236
	nvmeq->sq_tail = 0;
	nvmeq->cq_head = 0;
	nvmeq->cq_phase = 1;
1237
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
1238
	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
1239
	nvme_dbbuf_init(dev, nvmeq, qid);
K
Keith Busch 已提交
1240
	dev->online_queues++;
1241
	spin_unlock_irq(&nvmeq->q_lock);
1242 1243 1244 1245 1246 1247
}

static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
{
	struct nvme_dev *dev = nvmeq->dev;
	int result;
1248

K
Keith Busch 已提交
1249
	nvmeq->cq_vector = qid - 1;
M
Matthew Wilcox 已提交
1250 1251
	result = adapter_alloc_cq(dev, qid, nvmeq);
	if (result < 0)
1252
		return result;
M
Matthew Wilcox 已提交
1253 1254 1255 1256 1257

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

1258
	result = queue_request_irq(nvmeq);
M
Matthew Wilcox 已提交
1259 1260 1261
	if (result < 0)
		goto release_sq;

1262 1263
	nvme_init_queue(nvmeq, qid);
	return result;
M
Matthew Wilcox 已提交
1264 1265 1266 1267 1268

 release_sq:
	adapter_delete_sq(dev, qid);
 release_cq:
	adapter_delete_cq(dev, qid);
1269
	return result;
M
Matthew Wilcox 已提交
1270 1271
}

1272
static const struct blk_mq_ops nvme_mq_admin_ops = {
1273
	.queue_rq	= nvme_queue_rq,
1274
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1275
	.init_hctx	= nvme_admin_init_hctx,
1276
	.exit_hctx      = nvme_admin_exit_hctx,
1277
	.init_request	= nvme_init_request,
M
Matias Bjørling 已提交
1278 1279 1280
	.timeout	= nvme_timeout,
};

1281
static const struct blk_mq_ops nvme_mq_ops = {
M
Matias Bjørling 已提交
1282
	.queue_rq	= nvme_queue_rq,
1283
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1284 1285
	.init_hctx	= nvme_init_hctx,
	.init_request	= nvme_init_request,
1286
	.map_queues	= nvme_pci_map_queues,
M
Matias Bjørling 已提交
1287
	.timeout	= nvme_timeout,
J
Jens Axboe 已提交
1288
	.poll		= nvme_poll,
M
Matias Bjørling 已提交
1289 1290
};

1291 1292
static void nvme_dev_remove_admin(struct nvme_dev *dev)
{
1293
	if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) {
1294 1295 1296 1297 1298 1299
		/*
		 * 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);
1300
		blk_cleanup_queue(dev->ctrl.admin_q);
1301 1302 1303 1304
		blk_mq_free_tag_set(&dev->admin_tagset);
	}
}

M
Matias Bjørling 已提交
1305 1306
static int nvme_alloc_admin_tags(struct nvme_dev *dev)
{
1307
	if (!dev->ctrl.admin_q) {
M
Matias Bjørling 已提交
1308 1309
		dev->admin_tagset.ops = &nvme_mq_admin_ops;
		dev->admin_tagset.nr_hw_queues = 1;
K
Keith Busch 已提交
1310 1311 1312 1313 1314 1315

		/*
		 * 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
1316
		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
1317
		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
1318
		dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
1319
		dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED;
M
Matias Bjørling 已提交
1320 1321 1322 1323 1324
		dev->admin_tagset.driver_data = dev;

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

1325 1326
		dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset);
		if (IS_ERR(dev->ctrl.admin_q)) {
M
Matias Bjørling 已提交
1327 1328 1329
			blk_mq_free_tag_set(&dev->admin_tagset);
			return -ENOMEM;
		}
1330
		if (!blk_get_queue(dev->ctrl.admin_q)) {
1331
			nvme_dev_remove_admin(dev);
1332
			dev->ctrl.admin_q = NULL;
1333 1334
			return -ENODEV;
		}
K
Keith Busch 已提交
1335
	} else
1336
		blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true);
M
Matias Bjørling 已提交
1337 1338 1339 1340

	return 0;
}

1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366
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;
}

1367
static int nvme_configure_admin_queue(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1368
{
1369
	int result;
M
Matthew Wilcox 已提交
1370
	u32 aqa;
1371
	u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
M
Matthew Wilcox 已提交
1372 1373
	struct nvme_queue *nvmeq;

1374 1375 1376 1377
	result = nvme_remap_bar(dev, db_bar_size(dev, 0));
	if (result < 0)
		return result;

1378
	dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ?
1379 1380
						NVME_CAP_NSSRC(cap) : 0;

1381 1382 1383
	if (dev->subsystem &&
	    (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO))
		writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS);
1384

1385
	result = nvme_disable_ctrl(&dev->ctrl, cap);
1386 1387
	if (result < 0)
		return result;
M
Matthew Wilcox 已提交
1388

M
Matias Bjørling 已提交
1389
	nvmeq = dev->queues[0];
1390
	if (!nvmeq) {
1391 1392
		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH,
					dev_to_node(dev->dev));
1393 1394 1395
		if (!nvmeq)
			return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1396 1397 1398 1399

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

1400 1401 1402
	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
1403

1404
	result = nvme_enable_ctrl(&dev->ctrl, cap);
1405
	if (result)
K
Keith Busch 已提交
1406
		return result;
M
Matias Bjørling 已提交
1407

K
Keith Busch 已提交
1408
	nvmeq->cq_vector = 0;
1409
	result = queue_request_irq(nvmeq);
1410 1411
	if (result) {
		nvmeq->cq_vector = -1;
K
Keith Busch 已提交
1412
		return result;
1413
	}
1414

M
Matthew Wilcox 已提交
1415 1416 1417
	return result;
}

1418
static int nvme_create_io_queues(struct nvme_dev *dev)
K
Keith Busch 已提交
1419
{
1420
	unsigned i, max;
1421
	int ret = 0;
K
Keith Busch 已提交
1422

1423
	for (i = dev->queue_count; i <= dev->max_qid; i++) {
1424 1425 1426
		/* 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))) {
1427
			ret = -ENOMEM;
K
Keith Busch 已提交
1428
			break;
1429 1430
		}
	}
K
Keith Busch 已提交
1431

1432 1433
	max = min(dev->max_qid, dev->queue_count - 1);
	for (i = dev->online_queues; i <= max; i++) {
1434
		ret = nvme_create_queue(dev->queues[i], i);
K
Keith Busch 已提交
1435
		if (ret)
K
Keith Busch 已提交
1436
			break;
M
Matthew Wilcox 已提交
1437
	}
1438 1439 1440 1441 1442 1443 1444 1445

	/*
	 * 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
1446 1447
}

1448 1449 1450 1451 1452 1453
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));

1454
	return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz  : x%08x\n",
1455 1456 1457 1458
		       ndev->cmbloc, ndev->cmbsz);
}
static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL);

1459 1460 1461 1462 1463 1464 1465 1466
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;

1467
	dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ);
1468 1469
	if (!(NVME_CMB_SZ(dev->cmbsz)))
		return NULL;
1470
	dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC);
1471

1472 1473
	if (!use_cmb_sqes)
		return NULL;
1474 1475 1476

	szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
	size = szu * NVME_CMB_SZ(dev->cmbsz);
1477 1478
	offset = szu * NVME_CMB_OFST(dev->cmbloc);
	bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc));
1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490

	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;

1491
	dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset;
1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505
	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;
1506 1507 1508 1509 1510
		if (dev->cmbsz) {
			sysfs_remove_file_from_group(&dev->ctrl.device->kobj,
						     &dev_attr_cmb.attr, NULL);
			dev->cmbsz = 0;
		}
1511 1512 1513
	}
}

1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 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 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669
static int nvme_set_host_mem(struct nvme_dev *dev, u32 bits)
{
	size_t len = dev->nr_host_mem_descs * sizeof(*dev->host_mem_descs);
	struct nvme_command c;
	u64 dma_addr;
	int ret;

	dma_addr = dma_map_single(dev->dev, dev->host_mem_descs, len,
			DMA_TO_DEVICE);
	if (dma_mapping_error(dev->dev, dma_addr))
		return -ENOMEM;

	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);
	}
	dma_unmap_single(dev->dev, dma_addr, len, DMA_TO_DEVICE);
	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;
	kfree(dev->host_mem_descs);
	dev->host_mem_descs = NULL;
}

static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred)
{
	struct nvme_host_mem_buf_desc *descs;
	u32 chunk_size, max_entries, i = 0;
	void **bufs;
	u64 size, tmp;

	/* start big and work our way down */
	chunk_size = min(preferred, (u64)PAGE_SIZE << MAX_ORDER);
retry:
	tmp = (preferred + chunk_size - 1);
	do_div(tmp, chunk_size);
	max_entries = tmp;
	descs = kcalloc(max_entries, sizeof(*descs), GFP_KERNEL);
	if (!descs)
		goto out;

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

	for (size = 0; size < preferred; size += chunk_size) {
		u32 len = min_t(u64, chunk_size, preferred - size);
		dma_addr_t dma_addr;

		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++;
	}

	if (!size || (min && size < min)) {
		dev_warn(dev->ctrl.device,
			"failed to allocate host memory buffer.\n");
		goto out_free_bufs;
	}

	dev_info(dev->ctrl.device,
		"allocated %lld MiB host memory buffer.\n",
		size >> ilog2(SZ_1M));
	dev->nr_host_mem_descs = i;
	dev->host_mem_size = size;
	dev->host_mem_descs = descs;
	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:
	kfree(descs);
out:
	/* try a smaller chunk size if we failed early */
	if (chunk_size >= PAGE_SIZE * 2 && (i == 0 || size < min)) {
		chunk_size /= 2;
		goto retry;
	}
	dev->host_mem_descs = NULL;
	return -ENOMEM;
}

static void nvme_setup_host_mem(struct nvme_dev *dev)
{
	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;

	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);
		return;
	}

	/*
	 * 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) {
		if (nvme_alloc_host_mem(dev, min, preferred))
			return;
	}

	if (nvme_set_host_mem(dev, enable_bits))
		nvme_free_host_mem(dev);
}

1670
static int nvme_setup_io_queues(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1671
{
M
Matias Bjørling 已提交
1672
	struct nvme_queue *adminq = dev->queues[0];
1673
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1674 1675
	int result, nr_io_queues;
	unsigned long size;
M
Matthew Wilcox 已提交
1676

1677
	nr_io_queues = num_online_cpus();
C
Christoph Hellwig 已提交
1678 1679
	result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
	if (result < 0)
M
Matthew Wilcox 已提交
1680
		return result;
C
Christoph Hellwig 已提交
1681

1682
	if (nr_io_queues == 0)
1683
		return 0;
M
Matthew Wilcox 已提交
1684

1685 1686 1687 1688 1689 1690 1691 1692 1693
	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);
	}

1694 1695 1696 1697 1698 1699 1700 1701 1702
	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;
1703

K
Keith Busch 已提交
1704
	/* Deregister the admin queue's interrupt */
1705
	pci_free_irq(pdev, 0, adminq);
K
Keith Busch 已提交
1706

1707 1708 1709 1710
	/*
	 * If we enable msix early due to not intx, disable it again before
	 * setting up the full range we need.
	 */
1711 1712 1713 1714 1715 1716
	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
1717

1718 1719 1720 1721 1722 1723 1724
	/*
	 * 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.
	 */

1725
	result = queue_request_irq(adminq);
1726 1727
	if (result) {
		adminq->cq_vector = -1;
K
Keith Busch 已提交
1728
		return result;
1729
	}
1730
	return nvme_create_io_queues(dev);
M
Matthew Wilcox 已提交
1731 1732
}

1733
static void nvme_del_queue_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1734
{
K
Keith Busch 已提交
1735
	struct nvme_queue *nvmeq = req->end_io_data;
1736

K
Keith Busch 已提交
1737 1738
	blk_mq_free_request(req);
	complete(&nvmeq->dev->ioq_wait);
K
Keith Busch 已提交
1739 1740
}

1741
static void nvme_del_cq_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1742
{
K
Keith Busch 已提交
1743
	struct nvme_queue *nvmeq = req->end_io_data;
K
Keith Busch 已提交
1744

K
Keith Busch 已提交
1745 1746 1747
	if (!error) {
		unsigned long flags;

1748 1749 1750 1751 1752 1753 1754
		/*
		 * 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
1755 1756
		nvme_process_cq(nvmeq);
		spin_unlock_irqrestore(&nvmeq->q_lock, flags);
K
Keith Busch 已提交
1757
	}
K
Keith Busch 已提交
1758 1759

	nvme_del_queue_end(req, error);
K
Keith Busch 已提交
1760 1761
}

K
Keith Busch 已提交
1762
static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
1763
{
K
Keith Busch 已提交
1764 1765 1766
	struct request_queue *q = nvmeq->dev->ctrl.admin_q;
	struct request *req;
	struct nvme_command cmd;
1767

K
Keith Busch 已提交
1768 1769 1770
	memset(&cmd, 0, sizeof(cmd));
	cmd.delete_queue.opcode = opcode;
	cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
1771

1772
	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
K
Keith Busch 已提交
1773 1774
	if (IS_ERR(req))
		return PTR_ERR(req);
1775

K
Keith Busch 已提交
1776 1777 1778 1779 1780 1781 1782
	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;
1783 1784
}

1785
static void nvme_disable_io_queues(struct nvme_dev *dev, int queues)
K
Keith Busch 已提交
1786
{
1787
	int pass;
K
Keith Busch 已提交
1788 1789
	unsigned long timeout;
	u8 opcode = nvme_admin_delete_sq;
K
Keith Busch 已提交
1790

K
Keith Busch 已提交
1791
	for (pass = 0; pass < 2; pass++) {
K
Keith Busch 已提交
1792
		int sent = 0, i = queues;
K
Keith Busch 已提交
1793 1794 1795 1796

		reinit_completion(&dev->ioq_wait);
 retry:
		timeout = ADMIN_TIMEOUT;
1797 1798
		for (; i > 0; i--, sent++)
			if (nvme_delete_queue(dev->queues[i], opcode))
K
Keith Busch 已提交
1799
				break;
1800

K
Keith Busch 已提交
1801 1802 1803 1804 1805 1806 1807 1808 1809
		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
1810 1811
}

1812 1813 1814 1815 1816 1817
/*
 * 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.
 */
1818
static int nvme_dev_add(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1819
{
1820
	if (!dev->ctrl.tagset) {
1821 1822 1823 1824 1825
		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
1826
				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
1827 1828 1829
		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
1830

1831 1832
		if (blk_mq_alloc_tag_set(&dev->tagset))
			return 0;
1833
		dev->ctrl.tagset = &dev->tagset;
1834 1835

		nvme_dbbuf_set(dev);
1836 1837 1838 1839 1840
	} 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);
1841
	}
1842

K
Keith Busch 已提交
1843
	return 0;
M
Matthew Wilcox 已提交
1844 1845
}

1846
static int nvme_pci_enable(struct nvme_dev *dev)
1847
{
K
Keith Busch 已提交
1848
	u64 cap;
1849
	int result = -ENOMEM;
1850
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1851 1852 1853 1854 1855 1856

	if (pci_enable_device_mem(pdev))
		return result;

	pci_set_master(pdev);

1857 1858
	if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
	    dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
1859
		goto disable;
1860

1861
	if (readl(dev->bar + NVME_REG_CSTS) == -1) {
K
Keith Busch 已提交
1862
		result = -ENODEV;
1863
		goto disable;
K
Keith Busch 已提交
1864
	}
1865 1866

	/*
1867 1868 1869
	 * 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.
1870
	 */
1871 1872 1873
	result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
	if (result < 0)
		return result;
1874

1875 1876
	cap = lo_hi_readq(dev->bar + NVME_REG_CAP);

K
Keith Busch 已提交
1877 1878
	dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
	dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
1879
	dev->dbs = dev->bar + 4096;
1880 1881 1882 1883 1884 1885 1886

	/*
	 * 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;
1887 1888
		dev_warn(dev->ctrl.device, "detected Apple NVMe controller, "
			"set queue depth=%u to work around controller resets\n",
1889 1890 1891
			dev->q_depth);
	}

1892 1893 1894 1895 1896 1897 1898 1899
	/*
	 * CMBs can currently only exist on >=1.2 PCIe devices. We only
	 * populate sysfs if a CMB is implemented. Note that we add the
	 * CMB attribute to the nvme_ctrl kobj which removes the need to remove
	 * it on exit. Since nvme_dev_attrs_group has no name we can pass
	 * NULL as final argument to sysfs_add_file_to_group.
	 */

1900
	if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) {
1901
		dev->cmb = nvme_map_cmb(dev);
1902

1903 1904 1905
		if (dev->cmbsz) {
			if (sysfs_add_file_to_group(&dev->ctrl.device->kobj,
						    &dev_attr_cmb.attr, NULL))
1906
				dev_warn(dev->ctrl.device,
1907 1908 1909 1910
					 "failed to add sysfs attribute for CMB\n");
		}
	}

K
Keith Busch 已提交
1911 1912
	pci_enable_pcie_error_reporting(pdev);
	pci_save_state(pdev);
1913 1914 1915 1916 1917 1918 1919 1920
	return 0;

 disable:
	pci_disable_device(pdev);
	return result;
}

static void nvme_dev_unmap(struct nvme_dev *dev)
1921 1922 1923
{
	if (dev->bar)
		iounmap(dev->bar);
1924
	pci_release_mem_regions(to_pci_dev(dev->dev));
1925 1926 1927
}

static void nvme_pci_disable(struct nvme_dev *dev)
1928
{
1929 1930
	struct pci_dev *pdev = to_pci_dev(dev->dev);

1931
	nvme_release_cmb(dev);
1932
	pci_free_irq_vectors(pdev);
1933

K
Keith Busch 已提交
1934 1935
	if (pci_is_enabled(pdev)) {
		pci_disable_pcie_error_reporting(pdev);
1936
		pci_disable_device(pdev);
K
Keith Busch 已提交
1937 1938 1939
	}
}

1940
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
M
Matthew Wilcox 已提交
1941
{
1942
	int i, queues;
K
Keith Busch 已提交
1943 1944
	bool dead = true;
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1945

1946
	mutex_lock(&dev->shutdown_lock);
K
Keith Busch 已提交
1947 1948 1949 1950 1951 1952 1953
	if (pci_is_enabled(pdev)) {
		u32 csts = readl(dev->bar + NVME_REG_CSTS);

		if (dev->ctrl.state == NVME_CTRL_LIVE)
			nvme_start_freeze(&dev->ctrl);
		dead = !!((csts & NVME_CSTS_CFS) || !(csts & NVME_CSTS_RDY) ||
			pdev->error_state  != pci_channel_io_normal);
1954
	}
1955

K
Keith Busch 已提交
1956 1957 1958 1959
	/*
	 * Give the controller a chance to complete all entered requests if
	 * doing a safe shutdown.
	 */
1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973
	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
1974 1975
	nvme_stop_queues(&dev->ctrl);

1976
	queues = dev->online_queues - 1;
1977 1978 1979
	for (i = dev->queue_count - 1; i > 0; i--)
		nvme_suspend_queue(dev->queues[i]);

K
Keith Busch 已提交
1980
	if (dead) {
1981 1982 1983 1984 1985 1986
		/* 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
1987
	} else {
1988
		nvme_disable_io_queues(dev, queues);
1989
		nvme_disable_admin_queue(dev, shutdown);
K
Keith Busch 已提交
1990
	}
1991
	nvme_pci_disable(dev);
1992

1993 1994
	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
1995 1996 1997 1998 1999 2000 2001 2002

	/*
	 * 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);
2003
	mutex_unlock(&dev->shutdown_lock);
M
Matthew Wilcox 已提交
2004 2005
}

M
Matthew Wilcox 已提交
2006 2007
static int nvme_setup_prp_pools(struct nvme_dev *dev)
{
2008
	dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
M
Matthew Wilcox 已提交
2009 2010 2011 2012
						PAGE_SIZE, PAGE_SIZE, 0);
	if (!dev->prp_page_pool)
		return -ENOMEM;

2013
	/* Optimisation for I/Os between 4k and 128k */
2014
	dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
2015 2016 2017 2018 2019
						256, 256, 0);
	if (!dev->prp_small_pool) {
		dma_pool_destroy(dev->prp_page_pool);
		return -ENOMEM;
	}
M
Matthew Wilcox 已提交
2020 2021 2022 2023 2024 2025
	return 0;
}

static void nvme_release_prp_pools(struct nvme_dev *dev)
{
	dma_pool_destroy(dev->prp_page_pool);
2026
	dma_pool_destroy(dev->prp_small_pool);
M
Matthew Wilcox 已提交
2027 2028
}

2029
static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
2030
{
2031
	struct nvme_dev *dev = to_nvme_dev(ctrl);
2032

2033
	nvme_dbbuf_dma_free(dev);
2034
	put_device(dev->dev);
2035 2036
	if (dev->tagset.tags)
		blk_mq_free_tag_set(&dev->tagset);
2037 2038
	if (dev->ctrl.admin_q)
		blk_put_queue(dev->ctrl.admin_q);
2039
	kfree(dev->queues);
2040
	free_opal_dev(dev->ctrl.opal_dev);
2041 2042 2043
	kfree(dev);
}

2044 2045
static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status)
{
2046
	dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status);
2047 2048

	kref_get(&dev->ctrl.kref);
2049
	nvme_dev_disable(dev, false);
2050 2051 2052 2053
	if (!schedule_work(&dev->remove_work))
		nvme_put_ctrl(&dev->ctrl);
}

2054
static void nvme_reset_work(struct work_struct *work)
2055
{
2056 2057
	struct nvme_dev *dev =
		container_of(work, struct nvme_dev, ctrl.reset_work);
2058
	bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL);
2059
	int result = -ENODEV;
2060

2061
	if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING))
2062
		goto out;
2063

2064 2065 2066 2067
	/*
	 * If we're called to reset a live controller first shut it down before
	 * moving on.
	 */
2068
	if (dev->ctrl.ctrl_config & NVME_CC_ENABLE)
2069
		nvme_dev_disable(dev, false);
2070

2071
	result = nvme_pci_enable(dev);
2072
	if (result)
2073
		goto out;
2074 2075 2076

	result = nvme_configure_admin_queue(dev);
	if (result)
2077
		goto out;
2078

M
Matias Bjørling 已提交
2079
	nvme_init_queue(dev->queues[0], 0);
K
Keith Busch 已提交
2080 2081
	result = nvme_alloc_admin_tags(dev);
	if (result)
2082
		goto out;
2083

2084 2085
	result = nvme_init_identify(&dev->ctrl);
	if (result)
2086
		goto out;
2087

2088 2089 2090 2091 2092 2093 2094 2095 2096
	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;
2097
	}
2098

2099 2100 2101 2102 2103 2104 2105
	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");
	}

2106 2107 2108
	if (dev->ctrl.hmpre)
		nvme_setup_host_mem(dev);

2109
	result = nvme_setup_io_queues(dev);
2110
	if (result)
2111
		goto out;
2112

2113 2114 2115 2116 2117 2118
	/*
	 * 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.
	 */
2119 2120
	if (dev->online_queues > 1)
		nvme_queue_async_events(&dev->ctrl);
2121

2122 2123 2124 2125
	/*
	 * Keep the controller around but remove all namespaces if we don't have
	 * any working I/O queue.
	 */
2126
	if (dev->online_queues < 2) {
2127
		dev_warn(dev->ctrl.device, "IO queues not created\n");
2128
		nvme_kill_queues(&dev->ctrl);
2129
		nvme_remove_namespaces(&dev->ctrl);
2130
	} else {
2131
		nvme_start_queues(&dev->ctrl);
K
Keith Busch 已提交
2132
		nvme_wait_freeze(&dev->ctrl);
2133
		nvme_dev_add(dev);
K
Keith Busch 已提交
2134
		nvme_unfreeze(&dev->ctrl);
2135 2136
	}

2137 2138 2139 2140
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) {
		dev_warn(dev->ctrl.device, "failed to mark controller live\n");
		goto out;
	}
2141 2142

	if (dev->online_queues > 1)
2143
		nvme_queue_scan(&dev->ctrl);
2144
	return;
2145

2146
 out:
2147
	nvme_remove_dead_ctrl(dev, result);
2148 2149
}

2150
static void nvme_remove_dead_ctrl_work(struct work_struct *work)
K
Keith Busch 已提交
2151
{
2152
	struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
2153
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
2154

2155
	nvme_kill_queues(&dev->ctrl);
K
Keith Busch 已提交
2156
	if (pci_get_drvdata(pdev))
K
Keith Busch 已提交
2157
		device_release_driver(&pdev->dev);
2158
	nvme_put_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2159 2160
}

2161
static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
T
Tejun Heo 已提交
2162
{
2163
	*val = readl(to_nvme_dev(ctrl)->bar + off);
2164
	return 0;
T
Tejun Heo 已提交
2165 2166
}

2167
static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
2168
{
2169 2170 2171
	writel(val, to_nvme_dev(ctrl)->bar + off);
	return 0;
}
2172

2173 2174 2175 2176
static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
{
	*val = readq(to_nvme_dev(ctrl)->bar + off);
	return 0;
2177 2178
}

2179
static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
M
Ming Lin 已提交
2180
	.name			= "pcie",
2181
	.module			= THIS_MODULE,
2182
	.flags			= NVME_F_METADATA_SUPPORTED,
2183
	.reg_read32		= nvme_pci_reg_read32,
2184
	.reg_write32		= nvme_pci_reg_write32,
2185
	.reg_read64		= nvme_pci_reg_read64,
2186
	.free_ctrl		= nvme_pci_free_ctrl,
2187
	.submit_async_event	= nvme_pci_submit_async_event,
2188
};
2189

2190 2191 2192 2193
static int nvme_dev_map(struct nvme_dev *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev->dev);

2194
	if (pci_request_mem_regions(pdev, "nvme"))
2195 2196
		return -ENODEV;

2197
	if (nvme_remap_bar(dev, NVME_REG_DBS + 4096))
2198 2199
		goto release;

M
Max Gurtovoy 已提交
2200
	return 0;
2201
  release:
M
Max Gurtovoy 已提交
2202 2203
	pci_release_mem_regions(pdev);
	return -ENODEV;
2204 2205
}

2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225
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;
}

2226
static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
M
Matthew Wilcox 已提交
2227
{
M
Matias Bjørling 已提交
2228
	int node, result = -ENOMEM;
M
Matthew Wilcox 已提交
2229
	struct nvme_dev *dev;
2230
	unsigned long quirks = id->driver_data;
M
Matthew Wilcox 已提交
2231

M
Matias Bjørling 已提交
2232 2233
	node = dev_to_node(&pdev->dev);
	if (node == NUMA_NO_NODE)
2234
		set_dev_node(&pdev->dev, first_memory_node);
M
Matias Bjørling 已提交
2235 2236

	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2237 2238
	if (!dev)
		return -ENOMEM;
M
Matias Bjørling 已提交
2239 2240
	dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2241 2242 2243
	if (!dev->queues)
		goto free;

2244
	dev->dev = get_device(&pdev->dev);
K
Keith Busch 已提交
2245
	pci_set_drvdata(pdev, dev);
2246

2247 2248 2249 2250
	result = nvme_dev_map(dev);
	if (result)
		goto free;

2251
	INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work);
2252
	INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
2253
	mutex_init(&dev->shutdown_lock);
K
Keith Busch 已提交
2254
	init_completion(&dev->ioq_wait);
M
Matthew Wilcox 已提交
2255

M
Matthew Wilcox 已提交
2256 2257
	result = nvme_setup_prp_pools(dev);
	if (result)
K
Keith Busch 已提交
2258
		goto put_pci;
2259

2260 2261
	quirks |= check_dell_samsung_bug(pdev);

2262
	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
2263
			quirks);
2264
	if (result)
2265
		goto release_pools;
2266

2267
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING);
2268 2269
	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));

2270
	queue_work(nvme_wq, &dev->ctrl.reset_work);
M
Matthew Wilcox 已提交
2271 2272
	return 0;

2273
 release_pools:
M
Matthew Wilcox 已提交
2274
	nvme_release_prp_pools(dev);
K
Keith Busch 已提交
2275
 put_pci:
2276
	put_device(dev->dev);
2277
	nvme_dev_unmap(dev);
M
Matthew Wilcox 已提交
2278 2279 2280 2281 2282 2283
 free:
	kfree(dev->queues);
	kfree(dev);
	return result;
}

2284 2285
static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
{
K
Keith Busch 已提交
2286
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2287

K
Keith Busch 已提交
2288
	if (prepare)
2289
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2290
	else
2291
		nvme_reset_ctrl(&dev->ctrl);
2292 2293
}

2294 2295 2296
static void nvme_shutdown(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2297
	nvme_dev_disable(dev, true);
2298 2299
}

2300 2301 2302 2303 2304
/*
 * 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.
 */
2305
static void nvme_remove(struct pci_dev *pdev)
M
Matthew Wilcox 已提交
2306 2307
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
K
Keith Busch 已提交
2308

2309 2310
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING);

2311
	cancel_work_sync(&dev->ctrl.reset_work);
K
Keith Busch 已提交
2312
	pci_set_drvdata(pdev, NULL);
2313

2314
	if (!pci_device_is_present(pdev)) {
2315
		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
2316 2317
		nvme_dev_disable(dev, false);
	}
2318

2319
	flush_work(&dev->ctrl.reset_work);
2320
	nvme_uninit_ctrl(&dev->ctrl);
2321
	nvme_dev_disable(dev, true);
2322
	nvme_free_host_mem(dev);
M
Matias Bjørling 已提交
2323
	nvme_dev_remove_admin(dev);
2324
	nvme_free_queues(dev, 0);
K
Keith Busch 已提交
2325
	nvme_release_prp_pools(dev);
2326
	nvme_dev_unmap(dev);
2327
	nvme_put_ctrl(&dev->ctrl);
M
Matthew Wilcox 已提交
2328 2329
}

K
Keith Busch 已提交
2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347
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;
}

2348
#ifdef CONFIG_PM_SLEEP
2349 2350 2351 2352 2353
static int nvme_suspend(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

2354
	nvme_dev_disable(ndev, true);
2355 2356 2357 2358 2359 2360 2361 2362
	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);

2363
	nvme_reset_ctrl(&ndev->ctrl);
K
Keith Busch 已提交
2364
	return 0;
2365
}
2366
#endif
2367 2368

static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
M
Matthew Wilcox 已提交
2369

K
Keith Busch 已提交
2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383
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
2384 2385
		dev_warn(dev->ctrl.device,
			"frozen state error detected, reset controller\n");
2386
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2387 2388
		return PCI_ERS_RESULT_NEED_RESET;
	case pci_channel_io_perm_failure:
K
Keith Busch 已提交
2389 2390
		dev_warn(dev->ctrl.device,
			"failure state error detected, request disconnect\n");
K
Keith Busch 已提交
2391 2392 2393 2394 2395 2396 2397 2398 2399
		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);

2400
	dev_info(dev->ctrl.device, "restart after slot reset\n");
K
Keith Busch 已提交
2401
	pci_restore_state(pdev);
2402
	nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2403 2404 2405 2406 2407 2408 2409 2410
	return PCI_ERS_RESULT_RECOVERED;
}

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

2411
static const struct pci_error_handlers nvme_err_handler = {
M
Matthew Wilcox 已提交
2412 2413 2414
	.error_detected	= nvme_error_detected,
	.slot_reset	= nvme_slot_reset,
	.resume		= nvme_error_resume,
2415
	.reset_notify	= nvme_reset_notify,
M
Matthew Wilcox 已提交
2416 2417
};

2418
static const struct pci_device_id nvme_id_table[] = {
2419
	{ PCI_VDEVICE(INTEL, 0x0953),
2420
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2421
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2422 2423
	{ PCI_VDEVICE(INTEL, 0x0a53),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2424
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2425 2426
	{ PCI_VDEVICE(INTEL, 0x0a54),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2427
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2428 2429
	{ PCI_VDEVICE(INTEL, 0xf1a5),	/* Intel 600P/P3100 */
		.driver_data = NVME_QUIRK_NO_DEEPEST_PS },
2430 2431
	{ PCI_VDEVICE(INTEL, 0x5845),	/* Qemu emulated controller */
		.driver_data = NVME_QUIRK_IDENTIFY_CNS, },
2432 2433
	{ PCI_DEVICE(0x1c58, 0x0003),	/* HGST adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2434 2435
	{ PCI_DEVICE(0x1c5f, 0x0540),	/* Memblaze Pblaze4 adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
M
Matthew Wilcox 已提交
2436
	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
2437
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
2438
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) },
M
Matthew Wilcox 已提交
2439 2440 2441 2442 2443 2444 2445 2446
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, nvme_id_table);

static struct pci_driver nvme_driver = {
	.name		= "nvme",
	.id_table	= nvme_id_table,
	.probe		= nvme_probe,
2447
	.remove		= nvme_remove,
2448
	.shutdown	= nvme_shutdown,
2449 2450 2451
	.driver		= {
		.pm	= &nvme_dev_pm_ops,
	},
K
Keith Busch 已提交
2452
	.sriov_configure = nvme_pci_sriov_configure,
M
Matthew Wilcox 已提交
2453 2454 2455 2456 2457
	.err_handler	= &nvme_err_handler,
};

static int __init nvme_init(void)
{
2458
	return pci_register_driver(&nvme_driver);
M
Matthew Wilcox 已提交
2459 2460 2461 2462 2463
}

static void __exit nvme_exit(void)
{
	pci_unregister_driver(&nvme_driver);
2464
	_nvme_check_size();
M
Matthew Wilcox 已提交
2465 2466 2467 2468
}

MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
MODULE_LICENSE("GPL");
2469
MODULE_VERSION("1.0");
M
Matthew Wilcox 已提交
2470 2471
module_init(nvme_init);
module_exit(nvme_exit);