pci.c 63.7 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"

M
Matthew Wilcox 已提交
38 39
#define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
#define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
40

41 42 43 44
/*
 * We handle AEN commands ourselves and don't even let the
 * block layer know about them.
 */
45
#define NVME_AQ_BLKMQ_DEPTH	(NVME_AQ_DEPTH - NVME_NR_AERS)
46

47 48 49
static int use_threaded_interrupts;
module_param(use_threaded_interrupts, int, 0);

50 51 52 53
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");

54 55 56 57
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)");
58

59 60 61 62 63 64 65 66 67 68
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");

69 70
struct nvme_dev;
struct nvme_queue;
71

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

75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
/*
 * 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;
91
	unsigned long bar_mapped_size;
92
	struct work_struct remove_work;
93
	struct mutex shutdown_lock;
94 95 96 97 98
	bool subsystem;
	void __iomem *cmb;
	dma_addr_t cmb_dma_addr;
	u64 cmb_size;
	u32 cmbsz;
99
	u32 cmbloc;
100
	struct nvme_ctrl ctrl;
K
Keith Busch 已提交
101
	struct completion ioq_wait;
102 103

	/* shadow doorbell buffer support: */
104 105 106 107
	u32 *dbbuf_dbs;
	dma_addr_t dbbuf_dbs_dma_addr;
	u32 *dbbuf_eis;
	dma_addr_t dbbuf_eis_dma_addr;
108 109 110 111 112 113

	/* 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
114
};
115

116 117 118 119 120 121 122 123 124 125 126
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);
}

127 128 129 130 131 132 133 134 135 136
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;
}

137 138 139 140 141
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
142 143 144 145 146 147
/*
 * 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
148
	struct nvme_dev *dev;
M
Matthew Wilcox 已提交
149 150
	spinlock_t q_lock;
	struct nvme_command *sq_cmds;
151
	struct nvme_command __iomem *sq_cmds_io;
M
Matthew Wilcox 已提交
152
	volatile struct nvme_completion *cqes;
153
	struct blk_mq_tags **tags;
M
Matthew Wilcox 已提交
154 155 156 157
	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
158
	s16 cq_vector;
M
Matthew Wilcox 已提交
159 160
	u16 sq_tail;
	u16 cq_head;
K
Keith Busch 已提交
161
	u16 qid;
162 163
	u8 cq_phase;
	u8 cqe_seen;
164 165 166 167
	u32 *dbbuf_sq_db;
	u32 *dbbuf_cq_db;
	u32 *dbbuf_sq_ei;
	u32 *dbbuf_cq_ei;
M
Matthew Wilcox 已提交
168 169
};

170 171 172
/*
 * 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
173
 * me express that.  Use nvme_init_iod to ensure there's enough space
174 175 176
 * allocated to store the PRP list.
 */
struct nvme_iod {
177
	struct nvme_request req;
C
Christoph Hellwig 已提交
178 179
	struct nvme_queue *nvmeq;
	int aborted;
180 181 182 183
	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;
184
	struct scatterlist meta_sg; /* metadata requires single contiguous buffer */
C
Christoph Hellwig 已提交
185 186
	struct scatterlist *sg;
	struct scatterlist inline_sg[0];
M
Matthew Wilcox 已提交
187 188 189 190 191 192 193 194 195 196 197 198
};

/*
 * 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);
199
	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
K
Keith Busch 已提交
200
	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
M
Matthew Wilcox 已提交
201
	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
202 203
	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
204
	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
K
Keith Busch 已提交
205
	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279
	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)) {
280
		dev_warn(dev->ctrl.device, "unable to set dbbuf\n");
281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
		/* 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
312 313
}

314 315 316 317
/*
 * Max size of iod being embedded in the request payload
 */
#define NVME_INT_PAGES		2
318
#define NVME_INT_BYTES(dev)	(NVME_INT_PAGES * (dev)->ctrl.page_size)
319 320 321 322 323 324 325 326

/*
 * 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)
{
327 328
	unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size,
				      dev->ctrl.page_size);
329 330 331
	return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
}

C
Christoph Hellwig 已提交
332 333
static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
		unsigned int size, unsigned int nseg)
334
{
C
Christoph Hellwig 已提交
335 336 337
	return sizeof(__le64 *) * nvme_npages(size, dev) +
			sizeof(struct scatterlist) * nseg;
}
338

C
Christoph Hellwig 已提交
339 340 341 342
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);
343 344
}

M
Matias Bjørling 已提交
345 346
static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
				unsigned int hctx_idx)
347
{
M
Matias Bjørling 已提交
348 349 350
	struct nvme_dev *dev = data;
	struct nvme_queue *nvmeq = dev->queues[0];

351 352 353 354
	WARN_ON(hctx_idx != 0);
	WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
	WARN_ON(nvmeq->tags);

M
Matias Bjørling 已提交
355
	hctx->driver_data = nvmeq;
356
	nvmeq->tags = &dev->admin_tagset.tags[0];
M
Matias Bjørling 已提交
357
	return 0;
358 359
}

360 361 362 363 364 365 366
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
367 368
static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
			  unsigned int hctx_idx)
M
Matthew Wilcox 已提交
369
{
M
Matias Bjørling 已提交
370
	struct nvme_dev *dev = data;
371
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
M
Matias Bjørling 已提交
372

373 374
	if (!nvmeq->tags)
		nvmeq->tags = &dev->tagset.tags[hctx_idx];
M
Matthew Wilcox 已提交
375

376
	WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
M
Matias Bjørling 已提交
377 378
	hctx->driver_data = nvmeq;
	return 0;
M
Matthew Wilcox 已提交
379 380
}

381 382
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
383
{
384
	struct nvme_dev *dev = set->driver_data;
C
Christoph Hellwig 已提交
385
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
386 387
	int queue_idx = (set == &dev->tagset) ? hctx_idx + 1 : 0;
	struct nvme_queue *nvmeq = dev->queues[queue_idx];
M
Matias Bjørling 已提交
388 389

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
390
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
391 392 393
	return 0;
}

394 395 396 397 398 399 400
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
401
/**
402
 * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
M
Matthew Wilcox 已提交
403 404 405 406 407
 * @nvmeq: The queue to use
 * @cmd: The command to send
 *
 * Safe to use from interrupt context
 */
408 409
static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
						struct nvme_command *cmd)
M
Matthew Wilcox 已提交
410
{
M
Matias Bjørling 已提交
411 412
	u16 tail = nvmeq->sq_tail;

413 414 415 416 417
	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
418 419
	if (++tail == nvmeq->q_depth)
		tail = 0;
420 421 422
	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
423 424 425
	nvmeq->sq_tail = tail;
}

C
Christoph Hellwig 已提交
426
static __le64 **iod_list(struct request *req)
M
Matthew Wilcox 已提交
427
{
C
Christoph Hellwig 已提交
428
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
429
	return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req));
M
Matthew Wilcox 已提交
430 431
}

432
static blk_status_t nvme_init_iod(struct request *rq, struct nvme_dev *dev)
433
{
C
Christoph Hellwig 已提交
434
	struct nvme_iod *iod = blk_mq_rq_to_pdu(rq);
435
	int nseg = blk_rq_nr_phys_segments(rq);
436
	unsigned int size = blk_rq_payload_bytes(rq);
437

C
Christoph Hellwig 已提交
438 439 440
	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)
441
			return BLK_STS_RESOURCE;
C
Christoph Hellwig 已提交
442 443
	} else {
		iod->sg = iod->inline_sg;
444 445
	}

C
Christoph Hellwig 已提交
446 447 448 449
	iod->aborted = 0;
	iod->npages = -1;
	iod->nents = 0;
	iod->length = size;
K
Keith Busch 已提交
450

451
	return BLK_STS_OK;
452 453
}

C
Christoph Hellwig 已提交
454
static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
455
{
C
Christoph Hellwig 已提交
456
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
457
	const int last_prp = dev->ctrl.page_size / 8 - 1;
458
	int i;
C
Christoph Hellwig 已提交
459
	__le64 **list = iod_list(req);
460 461 462 463 464 465 466 467 468 469
	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;
	}
470

C
Christoph Hellwig 已提交
471 472
	if (iod->sg != iod->inline_sg)
		kfree(iod->sg);
K
Keith Busch 已提交
473 474
}

475
#ifdef CONFIG_BLK_DEV_INTEGRITY
K
Keith Busch 已提交
476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519
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);
520
	ts = ns->disk->queue->integrity.tuple_size;
K
Keith Busch 已提交
521 522 523 524 525 526 527 528

	for (i = 0; i < nlb; i++, virt++, phys++) {
		pi = (struct t10_pi_tuple *)p;
		dif_swap(phys, virt, pi);
		p += ts;
	}
	kunmap_atomic(pmap);
}
529 530 531 532 533 534 535 536 537 538 539 540 541
#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

542
static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
543
{
C
Christoph Hellwig 已提交
544
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
545
	struct dma_pool *pool;
546
	int length = blk_rq_payload_bytes(req);
547
	struct scatterlist *sg = iod->sg;
M
Matthew Wilcox 已提交
548 549
	int dma_len = sg_dma_len(sg);
	u64 dma_addr = sg_dma_address(sg);
550
	u32 page_size = dev->ctrl.page_size;
551
	int offset = dma_addr & (page_size - 1);
552
	__le64 *prp_list;
C
Christoph Hellwig 已提交
553
	__le64 **list = iod_list(req);
554
	dma_addr_t prp_dma;
555
	int nprps, i;
M
Matthew Wilcox 已提交
556

557
	length -= (page_size - offset);
M
Matthew Wilcox 已提交
558
	if (length <= 0)
559
		return true;
M
Matthew Wilcox 已提交
560

561
	dma_len -= (page_size - offset);
M
Matthew Wilcox 已提交
562
	if (dma_len) {
563
		dma_addr += (page_size - offset);
M
Matthew Wilcox 已提交
564 565 566 567 568 569
	} else {
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
	}

570
	if (length <= page_size) {
571
		iod->first_dma = dma_addr;
572
		return true;
573 574
	}

575
	nprps = DIV_ROUND_UP(length, page_size);
576 577
	if (nprps <= (256 / 8)) {
		pool = dev->prp_small_pool;
578
		iod->npages = 0;
579 580
	} else {
		pool = dev->prp_page_pool;
581
		iod->npages = 1;
582 583
	}

584
	prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
585
	if (!prp_list) {
586
		iod->first_dma = dma_addr;
587
		iod->npages = -1;
588
		return false;
589
	}
590 591
	list[0] = prp_list;
	iod->first_dma = prp_dma;
592 593
	i = 0;
	for (;;) {
594
		if (i == page_size >> 3) {
595
			__le64 *old_prp_list = prp_list;
596
			prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
597
			if (!prp_list)
598
				return false;
599
			list[iod->npages++] = prp_list;
600 601 602
			prp_list[0] = old_prp_list[i - 1];
			old_prp_list[i - 1] = cpu_to_le64(prp_dma);
			i = 1;
603 604
		}
		prp_list[i++] = cpu_to_le64(dma_addr);
605 606 607
		dma_len -= page_size;
		dma_addr += page_size;
		length -= page_size;
608 609 610 611 612 613 614 615
		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
616 617
	}

618
	return true;
M
Matthew Wilcox 已提交
619 620
}

621
static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req,
622
		struct nvme_command *cmnd)
623
{
C
Christoph Hellwig 已提交
624
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
C
Christoph Hellwig 已提交
625 626 627
	struct request_queue *q = req->q;
	enum dma_data_direction dma_dir = rq_data_dir(req) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE;
628
	blk_status_t ret = BLK_STS_IOERR;
629

630
	sg_init_table(iod->sg, blk_rq_nr_phys_segments(req));
C
Christoph Hellwig 已提交
631 632 633
	iod->nents = blk_rq_map_sg(q, req, iod->sg);
	if (!iod->nents)
		goto out;
634

635
	ret = BLK_STS_RESOURCE;
636 637
	if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir,
				DMA_ATTR_NO_WARN))
C
Christoph Hellwig 已提交
638
		goto out;
639

640
	if (!nvme_setup_prps(dev, req))
C
Christoph Hellwig 已提交
641
		goto out_unmap;
642

643
	ret = BLK_STS_IOERR;
C
Christoph Hellwig 已提交
644 645 646
	if (blk_integrity_rq(req)) {
		if (blk_rq_count_integrity_sg(q, req->bio) != 1)
			goto out_unmap;
647

648 649
		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
650
			goto out_unmap;
651

C
Christoph Hellwig 已提交
652 653
		if (rq_data_dir(req))
			nvme_dif_remap(req, nvme_dif_prep);
654

655
		if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir))
C
Christoph Hellwig 已提交
656
			goto out_unmap;
657
	}
M
Matthew Wilcox 已提交
658

659 660
	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
661
	if (blk_integrity_rq(req))
662
		cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
663
	return BLK_STS_OK;
M
Matthew Wilcox 已提交
664

C
Christoph Hellwig 已提交
665 666 667 668
out_unmap:
	dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
out:
	return ret;
M
Matthew Wilcox 已提交
669 670
}

C
Christoph Hellwig 已提交
671
static void nvme_unmap_data(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
672
{
C
Christoph Hellwig 已提交
673
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
674 675 676 677 678 679 680 681
	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);
682
			dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir);
K
Keith Busch 已提交
683
		}
684
	}
K
Keith Busch 已提交
685

686
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
687
	nvme_free_iod(dev, req);
688
}
M
Matthew Wilcox 已提交
689

690 691 692
/*
 * NOTE: ns is NULL when called on the admin queue.
 */
693
static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
M
Matias Bjørling 已提交
694
			 const struct blk_mq_queue_data *bd)
695
{
M
Matias Bjørling 已提交
696 697
	struct nvme_ns *ns = hctx->queue->queuedata;
	struct nvme_queue *nvmeq = hctx->driver_data;
698
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
699
	struct request *req = bd->rq;
C
Christoph Hellwig 已提交
700
	struct nvme_command cmnd;
701
	blk_status_t ret;
K
Keith Busch 已提交
702

703
	ret = nvme_setup_cmd(ns, req, &cmnd);
704
	if (ret)
C
Christoph Hellwig 已提交
705
		return ret;
M
Matias Bjørling 已提交
706

707
	ret = nvme_init_iod(req, dev);
708
	if (ret)
709
		goto out_free_cmd;
M
Matias Bjørling 已提交
710

711
	if (blk_rq_nr_phys_segments(req)) {
712
		ret = nvme_map_data(dev, req, &cmnd);
713 714 715
		if (ret)
			goto out_cleanup_iod;
	}
M
Matias Bjørling 已提交
716

717
	blk_mq_start_request(req);
M
Matias Bjørling 已提交
718

C
Christoph Hellwig 已提交
719
	spin_lock_irq(&nvmeq->q_lock);
720
	if (unlikely(nvmeq->cq_vector < 0)) {
721
		ret = BLK_STS_IOERR;
722
		spin_unlock_irq(&nvmeq->q_lock);
723
		goto out_cleanup_iod;
724
	}
C
Christoph Hellwig 已提交
725
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matias Bjørling 已提交
726 727
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
728
	return BLK_STS_OK;
729
out_cleanup_iod:
C
Christoph Hellwig 已提交
730
	nvme_free_iod(dev, req);
731 732
out_free_cmd:
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
733
	return ret;
M
Matthew Wilcox 已提交
734
}
K
Keith Busch 已提交
735

736
static void nvme_pci_complete_rq(struct request *req)
737
{
C
Christoph Hellwig 已提交
738
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
739

740 741
	nvme_unmap_data(iod->nvmeq->dev, req);
	nvme_complete_rq(req);
M
Matthew Wilcox 已提交
742 743
}

744 745 746 747 748 749 750
/* 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;
}

751
static inline void nvme_ring_cq_doorbell(struct nvme_queue *nvmeq)
M
Matthew Wilcox 已提交
752
{
753
	u16 head = nvmeq->cq_head;
754

755 756 757 758 759 760
	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);
	}
}
761

762 763 764 765
static inline void nvme_handle_cqe(struct nvme_queue *nvmeq,
		struct nvme_completion *cqe)
{
	struct request *req;
766

767 768 769 770 771
	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
772 773
	}

774 775 776 777 778 779 780 781 782 783
	/*
	 * 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
784
		return;
785
	}
M
Matthew Wilcox 已提交
786

787 788 789
	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
790

791 792
static inline bool nvme_read_cqe(struct nvme_queue *nvmeq,
		struct nvme_completion *cqe)
M
Matthew Wilcox 已提交
793
{
794 795
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) {
		*cqe = nvmeq->cqes[nvmeq->cq_head];
796

797 798 799
		if (++nvmeq->cq_head == nvmeq->q_depth) {
			nvmeq->cq_head = 0;
			nvmeq->cq_phase = !nvmeq->cq_phase;
M
Matthew Wilcox 已提交
800
		}
801
		return true;
M
Matthew Wilcox 已提交
802
	}
803
	return false;
J
Jens Axboe 已提交
804 805 806 807
}

static void nvme_process_cq(struct nvme_queue *nvmeq)
{
808 809
	struct nvme_completion cqe;
	int consumed = 0;
M
Matthew Wilcox 已提交
810

811 812 813 814
	while (nvme_read_cqe(nvmeq, &cqe)) {
		nvme_handle_cqe(nvmeq, &cqe);
		consumed++;
	}
815

816 817 818 819
	if (consumed) {
		nvme_ring_cq_doorbell(nvmeq);
		nvmeq->cqe_seen = 1;
	}
M
Matthew Wilcox 已提交
820 821 822
}

static irqreturn_t nvme_irq(int irq, void *data)
823 824 825 826
{
	irqreturn_t result;
	struct nvme_queue *nvmeq = data;
	spin_lock(&nvmeq->q_lock);
827 828 829
	nvme_process_cq(nvmeq);
	result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
	nvmeq->cqe_seen = 0;
830 831 832 833 834 835 836
	spin_unlock(&nvmeq->q_lock);
	return result;
}

static irqreturn_t nvme_irq_check(int irq, void *data)
{
	struct nvme_queue *nvmeq = data;
837 838 839
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return IRQ_WAKE_THREAD;
	return IRQ_NONE;
840 841
}

K
Keith Busch 已提交
842
static int __nvme_poll(struct nvme_queue *nvmeq, unsigned int tag)
J
Jens Axboe 已提交
843
{
844 845
	struct nvme_completion cqe;
	int found = 0, consumed = 0;
J
Jens Axboe 已提交
846

847 848
	if (!nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return 0;
J
Jens Axboe 已提交
849

850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
	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
866 867
}

K
Keith Busch 已提交
868 869 870 871 872 873 874
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);
}

875
static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx)
M
Matthew Wilcox 已提交
876
{
877
	struct nvme_dev *dev = to_nvme_dev(ctrl);
878
	struct nvme_queue *nvmeq = dev->queues[0];
M
Matias Bjørling 已提交
879
	struct nvme_command c;
M
Matthew Wilcox 已提交
880

M
Matias Bjørling 已提交
881 882
	memset(&c, 0, sizeof(c));
	c.common.opcode = nvme_admin_async_event;
883
	c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx;
884

885
	spin_lock_irq(&nvmeq->q_lock);
886
	__nvme_submit_cmd(nvmeq, &c);
887
	spin_unlock_irq(&nvmeq->q_lock);
888 889
}

M
Matthew Wilcox 已提交
890
static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
891
{
M
Matthew Wilcox 已提交
892 893 894 895 896 897
	struct nvme_command c;

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

898
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
899 900 901 902 903 904 905 906
}

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;

907 908 909 910
	/*
	 * 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
911 912 913 914 915 916 917 918
	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);

919
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
920 921 922 923 924 925
}

static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
						struct nvme_queue *nvmeq)
{
	struct nvme_command c;
926
	int flags = NVME_QUEUE_PHYS_CONTIG;
M
Matthew Wilcox 已提交
927

928 929 930 931
	/*
	 * 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
932 933 934 935 936 937 938 939
	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);

940
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
941 942 943 944 945 946 947 948 949 950 951 952
}

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);
}

953
static void abort_endio(struct request *req, blk_status_t error)
954
{
C
Christoph Hellwig 已提交
955 956
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
957

958 959
	dev_warn(nvmeq->dev->ctrl.device,
		 "Abort status: 0x%x", nvme_req(req)->status);
960 961
	atomic_inc(&nvmeq->dev->ctrl.abort_limit);
	blk_mq_free_request(req);
962 963
}

K
Keith Busch 已提交
964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008
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);
}

1009
static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
K
Keith Busch 已提交
1010
{
C
Christoph Hellwig 已提交
1011 1012
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
K
Keith Busch 已提交
1013
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
1014 1015
	struct request *abort_req;
	struct nvme_command cmd;
K
Keith Busch 已提交
1016 1017 1018 1019 1020 1021 1022 1023
	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);
1024
		nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
1025 1026
		return BLK_EH_HANDLED;
	}
K
Keith Busch 已提交
1027

K
Keith Busch 已提交
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
	/*
	 * 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;
	}

1038
	/*
1039 1040 1041 1042 1043
	 * 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.
	 */
1044
	if (dev->ctrl.state == NVME_CTRL_RESETTING) {
1045
		dev_warn(dev->ctrl.device,
1046 1047
			 "I/O %d QID %d timeout, disable controller\n",
			 req->tag, nvmeq->qid);
1048
		nvme_dev_disable(dev, false);
1049
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1050
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1051 1052
	}

1053 1054 1055 1056
	/*
 	 * 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.
1057
	 */
C
Christoph Hellwig 已提交
1058
	if (!nvmeq->qid || iod->aborted) {
1059
		dev_warn(dev->ctrl.device,
1060 1061
			 "I/O %d QID %d timeout, reset controller\n",
			 req->tag, nvmeq->qid);
1062
		nvme_dev_disable(dev, false);
1063
		nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
1064

1065 1066 1067 1068
		/*
		 * Mark the request as handled, since the inline shutdown
		 * forces all outstanding requests to complete.
		 */
1069
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1070
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1071 1072
	}

1073
	if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) {
1074
		atomic_inc(&dev->ctrl.abort_limit);
1075
		return BLK_EH_RESET_TIMER;
1076
	}
1077
	iod->aborted = 1;
M
Matias Bjørling 已提交
1078

K
Keith Busch 已提交
1079 1080
	memset(&cmd, 0, sizeof(cmd));
	cmd.abort.opcode = nvme_admin_abort_cmd;
M
Matias Bjørling 已提交
1081
	cmd.abort.cid = req->tag;
K
Keith Busch 已提交
1082 1083
	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);

1084 1085 1086
	dev_warn(nvmeq->dev->ctrl.device,
		"I/O %d QID %d timeout, aborting\n",
		 req->tag, nvmeq->qid);
1087 1088

	abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
1089
			BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
1090 1091 1092 1093 1094 1095 1096 1097
	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
1098

1099 1100 1101 1102 1103 1104
	/*
	 * 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
1105 1106
}

M
Matias Bjørling 已提交
1107 1108
static void nvme_free_queue(struct nvme_queue *nvmeq)
{
1109 1110
	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
1111 1112
	if (nvmeq->sq_cmds)
		dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
1113 1114 1115 1116
					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
	kfree(nvmeq);
}

1117
static void nvme_free_queues(struct nvme_dev *dev, int lowest)
1118 1119 1120
{
	int i;

1121
	for (i = dev->ctrl.queue_count - 1; i >= lowest; i--) {
M
Matias Bjørling 已提交
1122
		struct nvme_queue *nvmeq = dev->queues[i];
1123
		dev->ctrl.queue_count--;
M
Matias Bjørling 已提交
1124
		dev->queues[i] = NULL;
K
Keith Busch 已提交
1125
		nvme_free_queue(nvmeq);
1126
	}
1127 1128
}

K
Keith Busch 已提交
1129 1130 1131 1132 1133
/**
 * 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
1134
{
K
Keith Busch 已提交
1135
	int vector;
M
Matthew Wilcox 已提交
1136

1137
	spin_lock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
1138 1139 1140 1141
	if (nvmeq->cq_vector == -1) {
		spin_unlock_irq(&nvmeq->q_lock);
		return 1;
	}
1142
	vector = nvmeq->cq_vector;
K
Keith Busch 已提交
1143
	nvmeq->dev->online_queues--;
K
Keith Busch 已提交
1144
	nvmeq->cq_vector = -1;
1145 1146
	spin_unlock_irq(&nvmeq->q_lock);

1147
	if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
1148
		blk_mq_quiesce_queue(nvmeq->dev->ctrl.admin_q);
1149

1150
	pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq);
M
Matthew Wilcox 已提交
1151

K
Keith Busch 已提交
1152 1153
	return 0;
}
M
Matthew Wilcox 已提交
1154

1155
static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
K
Keith Busch 已提交
1156
{
1157
	struct nvme_queue *nvmeq = dev->queues[0];
K
Keith Busch 已提交
1158 1159 1160 1161 1162 1163

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

1164 1165 1166
	if (shutdown)
		nvme_shutdown_ctrl(&dev->ctrl);
	else
1167
		nvme_disable_ctrl(&dev->ctrl, dev->ctrl.cap);
1168 1169 1170 1171

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1172 1173
}

1174 1175 1176 1177
static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
				int entry_size)
{
	int q_depth = dev->q_depth;
1178 1179
	unsigned q_size_aligned = roundup(q_depth * entry_size,
					  dev->ctrl.page_size);
1180 1181

	if (q_size_aligned * nr_io_queues > dev->cmb_size) {
1182
		u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1183
		mem_per_q = round_down(mem_per_q, dev->ctrl.page_size);
1184
		q_depth = div_u64(mem_per_q, entry_size);
1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201

		/*
		 * 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)) {
1202 1203
		unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth),
						      dev->ctrl.page_size);
1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215
		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
1216
static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
1217
							int depth, int node)
M
Matthew Wilcox 已提交
1218
{
1219 1220
	struct nvme_queue *nvmeq = kzalloc_node(sizeof(*nvmeq), GFP_KERNEL,
							node);
M
Matthew Wilcox 已提交
1221 1222 1223
	if (!nvmeq)
		return NULL;

1224
	nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
J
Joe Perches 已提交
1225
					  &nvmeq->cq_dma_addr, GFP_KERNEL);
M
Matthew Wilcox 已提交
1226 1227 1228
	if (!nvmeq->cqes)
		goto free_nvmeq;

1229
	if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
M
Matthew Wilcox 已提交
1230 1231
		goto free_cqdma;

1232
	nvmeq->q_dmadev = dev->dev;
M
Matthew Wilcox 已提交
1233
	nvmeq->dev = dev;
M
Matthew Wilcox 已提交
1234 1235
	spin_lock_init(&nvmeq->q_lock);
	nvmeq->cq_head = 0;
M
Matthew Wilcox 已提交
1236
	nvmeq->cq_phase = 1;
1237
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
M
Matthew Wilcox 已提交
1238
	nvmeq->q_depth = depth;
K
Keith Busch 已提交
1239
	nvmeq->qid = qid;
1240
	nvmeq->cq_vector = -1;
M
Matias Bjørling 已提交
1241
	dev->queues[qid] = nvmeq;
1242
	dev->ctrl.queue_count++;
1243

M
Matthew Wilcox 已提交
1244 1245 1246
	return nvmeq;

 free_cqdma:
1247
	dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
M
Matthew Wilcox 已提交
1248 1249 1250 1251 1252 1253
							nvmeq->cq_dma_addr);
 free_nvmeq:
	kfree(nvmeq);
	return NULL;
}

1254
static int queue_request_irq(struct nvme_queue *nvmeq)
1255
{
1256 1257 1258 1259 1260 1261 1262 1263 1264 1265
	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);
	}
1266 1267
}

1268
static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
M
Matthew Wilcox 已提交
1269
{
1270
	struct nvme_dev *dev = nvmeq->dev;
M
Matthew Wilcox 已提交
1271

1272
	spin_lock_irq(&nvmeq->q_lock);
1273 1274 1275
	nvmeq->sq_tail = 0;
	nvmeq->cq_head = 0;
	nvmeq->cq_phase = 1;
1276
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
1277
	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
1278
	nvme_dbbuf_init(dev, nvmeq, qid);
K
Keith Busch 已提交
1279
	dev->online_queues++;
1280
	spin_unlock_irq(&nvmeq->q_lock);
1281 1282 1283 1284 1285 1286
}

static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
{
	struct nvme_dev *dev = nvmeq->dev;
	int result;
1287

K
Keith Busch 已提交
1288
	nvmeq->cq_vector = qid - 1;
M
Matthew Wilcox 已提交
1289 1290
	result = adapter_alloc_cq(dev, qid, nvmeq);
	if (result < 0)
1291
		return result;
M
Matthew Wilcox 已提交
1292 1293 1294 1295 1296

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

1297
	result = queue_request_irq(nvmeq);
M
Matthew Wilcox 已提交
1298 1299 1300
	if (result < 0)
		goto release_sq;

1301 1302
	nvme_init_queue(nvmeq, qid);
	return result;
M
Matthew Wilcox 已提交
1303 1304 1305 1306 1307

 release_sq:
	adapter_delete_sq(dev, qid);
 release_cq:
	adapter_delete_cq(dev, qid);
1308
	return result;
M
Matthew Wilcox 已提交
1309 1310
}

1311
static const struct blk_mq_ops nvme_mq_admin_ops = {
1312
	.queue_rq	= nvme_queue_rq,
1313
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1314
	.init_hctx	= nvme_admin_init_hctx,
1315
	.exit_hctx      = nvme_admin_exit_hctx,
1316
	.init_request	= nvme_init_request,
M
Matias Bjørling 已提交
1317 1318 1319
	.timeout	= nvme_timeout,
};

1320
static const struct blk_mq_ops nvme_mq_ops = {
M
Matias Bjørling 已提交
1321
	.queue_rq	= nvme_queue_rq,
1322
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1323 1324
	.init_hctx	= nvme_init_hctx,
	.init_request	= nvme_init_request,
1325
	.map_queues	= nvme_pci_map_queues,
M
Matias Bjørling 已提交
1326
	.timeout	= nvme_timeout,
J
Jens Axboe 已提交
1327
	.poll		= nvme_poll,
M
Matias Bjørling 已提交
1328 1329
};

1330 1331
static void nvme_dev_remove_admin(struct nvme_dev *dev)
{
1332
	if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) {
1333 1334 1335 1336 1337
		/*
		 * 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.
		 */
1338
		blk_mq_unquiesce_queue(dev->ctrl.admin_q);
1339
		blk_cleanup_queue(dev->ctrl.admin_q);
1340 1341 1342 1343
		blk_mq_free_tag_set(&dev->admin_tagset);
	}
}

M
Matias Bjørling 已提交
1344 1345
static int nvme_alloc_admin_tags(struct nvme_dev *dev)
{
1346
	if (!dev->ctrl.admin_q) {
M
Matias Bjørling 已提交
1347 1348
		dev->admin_tagset.ops = &nvme_mq_admin_ops;
		dev->admin_tagset.nr_hw_queues = 1;
K
Keith Busch 已提交
1349 1350 1351 1352 1353 1354

		/*
		 * 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
1355
		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
1356
		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
1357
		dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
1358
		dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED;
M
Matias Bjørling 已提交
1359 1360 1361 1362 1363
		dev->admin_tagset.driver_data = dev;

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

1364 1365
		dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset);
		if (IS_ERR(dev->ctrl.admin_q)) {
M
Matias Bjørling 已提交
1366 1367 1368
			blk_mq_free_tag_set(&dev->admin_tagset);
			return -ENOMEM;
		}
1369
		if (!blk_get_queue(dev->ctrl.admin_q)) {
1370
			nvme_dev_remove_admin(dev);
1371
			dev->ctrl.admin_q = NULL;
1372 1373
			return -ENODEV;
		}
K
Keith Busch 已提交
1374
	} else
1375
		blk_mq_unquiesce_queue(dev->ctrl.admin_q);
M
Matias Bjørling 已提交
1376 1377 1378 1379

	return 0;
}

1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
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;
}

1406
static int nvme_pci_configure_admin_queue(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1407
{
1408
	int result;
M
Matthew Wilcox 已提交
1409 1410 1411
	u32 aqa;
	struct nvme_queue *nvmeq;

1412 1413 1414 1415
	result = nvme_remap_bar(dev, db_bar_size(dev, 0));
	if (result < 0)
		return result;

1416
	dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ?
1417
				NVME_CAP_NSSRC(dev->ctrl.cap) : 0;
1418

1419 1420 1421
	if (dev->subsystem &&
	    (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO))
		writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS);
1422

1423
	result = nvme_disable_ctrl(&dev->ctrl, dev->ctrl.cap);
1424 1425
	if (result < 0)
		return result;
M
Matthew Wilcox 已提交
1426

M
Matias Bjørling 已提交
1427
	nvmeq = dev->queues[0];
1428
	if (!nvmeq) {
1429 1430
		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH,
					dev_to_node(dev->dev));
1431 1432 1433
		if (!nvmeq)
			return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1434 1435 1436 1437

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

1438 1439 1440
	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
1441

1442
	result = nvme_enable_ctrl(&dev->ctrl, dev->ctrl.cap);
1443
	if (result)
K
Keith Busch 已提交
1444
		return result;
M
Matias Bjørling 已提交
1445

K
Keith Busch 已提交
1446
	nvmeq->cq_vector = 0;
1447
	result = queue_request_irq(nvmeq);
1448 1449
	if (result) {
		nvmeq->cq_vector = -1;
K
Keith Busch 已提交
1450
		return result;
1451
	}
1452

M
Matthew Wilcox 已提交
1453 1454 1455
	return result;
}

1456
static int nvme_create_io_queues(struct nvme_dev *dev)
K
Keith Busch 已提交
1457
{
1458
	unsigned i, max;
1459
	int ret = 0;
K
Keith Busch 已提交
1460

1461
	for (i = dev->ctrl.queue_count; i <= dev->max_qid; i++) {
1462 1463 1464
		/* 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))) {
1465
			ret = -ENOMEM;
K
Keith Busch 已提交
1466
			break;
1467 1468
		}
	}
K
Keith Busch 已提交
1469

1470
	max = min(dev->max_qid, dev->ctrl.queue_count - 1);
1471
	for (i = dev->online_queues; i <= max; i++) {
1472
		ret = nvme_create_queue(dev->queues[i], i);
K
Keith Busch 已提交
1473
		if (ret)
K
Keith Busch 已提交
1474
			break;
M
Matthew Wilcox 已提交
1475
	}
1476 1477 1478 1479 1480 1481 1482 1483

	/*
	 * 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
1484 1485
}

1486 1487 1488 1489 1490 1491
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));

1492
	return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz  : x%08x\n",
1493 1494 1495 1496
		       ndev->cmbloc, ndev->cmbsz);
}
static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL);

1497 1498 1499 1500 1501 1502 1503 1504
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;

1505
	dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ);
1506 1507
	if (!(NVME_CMB_SZ(dev->cmbsz)))
		return NULL;
1508
	dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC);
1509

1510 1511
	if (!use_cmb_sqes)
		return NULL;
1512 1513 1514

	szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
	size = szu * NVME_CMB_SZ(dev->cmbsz);
1515 1516
	offset = szu * NVME_CMB_OFST(dev->cmbloc);
	bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc));
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528

	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;

1529
	dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset;
1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543
	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;
1544 1545 1546 1547 1548
		if (dev->cmbsz) {
			sysfs_remove_file_from_group(&dev->ctrl.device->kobj,
						     &dev_attr_cmb.attr, NULL);
			dev->cmbsz = 0;
		}
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
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)
K
Keith Busch 已提交
1603
{
1604
	struct nvme_host_mem_buf_desc *descs;
1605 1606
	u32 chunk_size, max_entries;
	int i = 0;
1607
	void **bufs;
1608
	u64 size = 0, tmp;
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 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706

	/* 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);
K
Keith Busch 已提交
1707 1708
}

1709
static int nvme_setup_io_queues(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1710
{
M
Matias Bjørling 已提交
1711
	struct nvme_queue *adminq = dev->queues[0];
1712
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1713 1714
	int result, nr_io_queues;
	unsigned long size;
M
Matthew Wilcox 已提交
1715

1716
	nr_io_queues = num_present_cpus();
C
Christoph Hellwig 已提交
1717 1718
	result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
	if (result < 0)
M
Matthew Wilcox 已提交
1719
		return result;
C
Christoph Hellwig 已提交
1720

1721
	if (nr_io_queues == 0)
1722
		return 0;
M
Matthew Wilcox 已提交
1723

1724 1725 1726 1727 1728 1729 1730 1731 1732
	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);
	}

1733 1734 1735 1736 1737 1738 1739 1740 1741
	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;
1742

K
Keith Busch 已提交
1743
	/* Deregister the admin queue's interrupt */
1744
	pci_free_irq(pdev, 0, adminq);
K
Keith Busch 已提交
1745

1746 1747 1748 1749
	/*
	 * If we enable msix early due to not intx, disable it again before
	 * setting up the full range we need.
	 */
1750 1751 1752 1753 1754 1755
	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
1756

1757 1758 1759 1760 1761 1762 1763
	/*
	 * 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.
	 */

1764
	result = queue_request_irq(adminq);
1765 1766
	if (result) {
		adminq->cq_vector = -1;
K
Keith Busch 已提交
1767
		return result;
1768
	}
1769
	return nvme_create_io_queues(dev);
M
Matthew Wilcox 已提交
1770 1771
}

1772
static void nvme_del_queue_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1773
{
K
Keith Busch 已提交
1774
	struct nvme_queue *nvmeq = req->end_io_data;
1775

K
Keith Busch 已提交
1776 1777
	blk_mq_free_request(req);
	complete(&nvmeq->dev->ioq_wait);
K
Keith Busch 已提交
1778 1779
}

1780
static void nvme_del_cq_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1781
{
K
Keith Busch 已提交
1782
	struct nvme_queue *nvmeq = req->end_io_data;
K
Keith Busch 已提交
1783

K
Keith Busch 已提交
1784 1785 1786
	if (!error) {
		unsigned long flags;

1787 1788 1789 1790 1791 1792 1793
		/*
		 * 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
1794 1795
		nvme_process_cq(nvmeq);
		spin_unlock_irqrestore(&nvmeq->q_lock, flags);
K
Keith Busch 已提交
1796
	}
K
Keith Busch 已提交
1797 1798

	nvme_del_queue_end(req, error);
K
Keith Busch 已提交
1799 1800
}

K
Keith Busch 已提交
1801
static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
1802
{
K
Keith Busch 已提交
1803 1804 1805
	struct request_queue *q = nvmeq->dev->ctrl.admin_q;
	struct request *req;
	struct nvme_command cmd;
1806

K
Keith Busch 已提交
1807 1808 1809
	memset(&cmd, 0, sizeof(cmd));
	cmd.delete_queue.opcode = opcode;
	cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
1810

1811
	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
K
Keith Busch 已提交
1812 1813
	if (IS_ERR(req))
		return PTR_ERR(req);
1814

K
Keith Busch 已提交
1815 1816 1817 1818 1819 1820 1821
	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;
1822 1823
}

1824
static void nvme_disable_io_queues(struct nvme_dev *dev, int queues)
K
Keith Busch 已提交
1825
{
1826
	int pass;
K
Keith Busch 已提交
1827 1828
	unsigned long timeout;
	u8 opcode = nvme_admin_delete_sq;
K
Keith Busch 已提交
1829

K
Keith Busch 已提交
1830
	for (pass = 0; pass < 2; pass++) {
K
Keith Busch 已提交
1831
		int sent = 0, i = queues;
K
Keith Busch 已提交
1832 1833 1834 1835

		reinit_completion(&dev->ioq_wait);
 retry:
		timeout = ADMIN_TIMEOUT;
1836 1837
		for (; i > 0; i--, sent++)
			if (nvme_delete_queue(dev->queues[i], opcode))
K
Keith Busch 已提交
1838
				break;
1839

K
Keith Busch 已提交
1840 1841 1842 1843 1844 1845 1846 1847 1848
		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
1849 1850
}

1851 1852 1853 1854 1855 1856
/*
 * 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.
 */
1857
static int nvme_dev_add(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1858
{
1859
	if (!dev->ctrl.tagset) {
1860 1861 1862 1863 1864
		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
1865
				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
1866 1867 1868
		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
1869

1870 1871
		if (blk_mq_alloc_tag_set(&dev->tagset))
			return 0;
1872
		dev->ctrl.tagset = &dev->tagset;
1873 1874

		nvme_dbbuf_set(dev);
1875 1876 1877 1878 1879
	} 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);
1880
	}
1881

K
Keith Busch 已提交
1882
	return 0;
M
Matthew Wilcox 已提交
1883 1884
}

1885
static int nvme_pci_enable(struct nvme_dev *dev)
1886
{
1887
	int result = -ENOMEM;
1888
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1889 1890 1891 1892 1893 1894

	if (pci_enable_device_mem(pdev))
		return result;

	pci_set_master(pdev);

1895 1896
	if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
	    dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
1897
		goto disable;
1898

1899
	if (readl(dev->bar + NVME_REG_CSTS) == -1) {
K
Keith Busch 已提交
1900
		result = -ENODEV;
1901
		goto disable;
K
Keith Busch 已提交
1902
	}
1903 1904

	/*
1905 1906 1907
	 * 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.
1908
	 */
1909 1910 1911
	result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
	if (result < 0)
		return result;
1912

1913
	dev->ctrl.cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
1914

1915
	dev->q_depth = min_t(int, NVME_CAP_MQES(dev->ctrl.cap) + 1,
1916
				io_queue_depth);
1917
	dev->db_stride = 1 << NVME_CAP_STRIDE(dev->ctrl.cap);
1918
	dev->dbs = dev->bar + 4096;
1919 1920 1921 1922 1923 1924 1925

	/*
	 * 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;
1926 1927
		dev_warn(dev->ctrl.device, "detected Apple NVMe controller, "
			"set queue depth=%u to work around controller resets\n",
1928
			dev->q_depth);
1929 1930
	} else if (pdev->vendor == PCI_VENDOR_ID_SAMSUNG &&
		   (pdev->device == 0xa821 || pdev->device == 0xa822) &&
1931
		   NVME_CAP_MQES(dev->ctrl.cap) == 0) {
1932 1933 1934
		dev->q_depth = 64;
		dev_err(dev->ctrl.device, "detected PM1725 NVMe controller, "
                        "set queue depth=%u\n", dev->q_depth);
1935 1936
	}

1937 1938 1939 1940 1941 1942 1943 1944
	/*
	 * 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.
	 */

1945
	if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) {
1946
		dev->cmb = nvme_map_cmb(dev);
1947

1948 1949 1950
		if (dev->cmbsz) {
			if (sysfs_add_file_to_group(&dev->ctrl.device->kobj,
						    &dev_attr_cmb.attr, NULL))
1951
				dev_warn(dev->ctrl.device,
1952 1953 1954 1955
					 "failed to add sysfs attribute for CMB\n");
		}
	}

K
Keith Busch 已提交
1956 1957
	pci_enable_pcie_error_reporting(pdev);
	pci_save_state(pdev);
1958 1959 1960 1961 1962 1963 1964 1965
	return 0;

 disable:
	pci_disable_device(pdev);
	return result;
}

static void nvme_dev_unmap(struct nvme_dev *dev)
1966 1967 1968
{
	if (dev->bar)
		iounmap(dev->bar);
1969
	pci_release_mem_regions(to_pci_dev(dev->dev));
1970 1971 1972
}

static void nvme_pci_disable(struct nvme_dev *dev)
1973
{
1974 1975
	struct pci_dev *pdev = to_pci_dev(dev->dev);

1976
	nvme_release_cmb(dev);
1977
	pci_free_irq_vectors(pdev);
1978

K
Keith Busch 已提交
1979 1980
	if (pci_is_enabled(pdev)) {
		pci_disable_pcie_error_reporting(pdev);
1981
		pci_disable_device(pdev);
K
Keith Busch 已提交
1982 1983 1984
	}
}

1985
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
M
Matthew Wilcox 已提交
1986
{
1987
	int i, queues;
K
Keith Busch 已提交
1988 1989
	bool dead = true;
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1990

1991
	mutex_lock(&dev->shutdown_lock);
K
Keith Busch 已提交
1992 1993 1994
	if (pci_is_enabled(pdev)) {
		u32 csts = readl(dev->bar + NVME_REG_CSTS);

K
Keith Busch 已提交
1995 1996
		if (dev->ctrl.state == NVME_CTRL_LIVE ||
		    dev->ctrl.state == NVME_CTRL_RESETTING)
K
Keith Busch 已提交
1997 1998 1999
			nvme_start_freeze(&dev->ctrl);
		dead = !!((csts & NVME_CSTS_CFS) || !(csts & NVME_CSTS_RDY) ||
			pdev->error_state  != pci_channel_io_normal);
2000
	}
2001

K
Keith Busch 已提交
2002 2003 2004 2005
	/*
	 * Give the controller a chance to complete all entered requests if
	 * doing a safe shutdown.
	 */
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019
	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
2020 2021
	nvme_stop_queues(&dev->ctrl);

2022
	queues = dev->online_queues - 1;
2023
	for (i = dev->ctrl.queue_count - 1; i > 0; i--)
2024 2025
		nvme_suspend_queue(dev->queues[i]);

K
Keith Busch 已提交
2026
	if (dead) {
2027 2028 2029 2030
		/* A device might become IO incapable very soon during
		 * probe, before the admin queue is configured. Thus,
		 * queue_count can be 0 here.
		 */
2031
		if (dev->ctrl.queue_count)
2032
			nvme_suspend_queue(dev->queues[0]);
K
Keith Busch 已提交
2033
	} else {
2034
		nvme_disable_io_queues(dev, queues);
2035
		nvme_disable_admin_queue(dev, shutdown);
K
Keith Busch 已提交
2036
	}
2037
	nvme_pci_disable(dev);
2038

2039 2040
	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
2041 2042 2043 2044 2045 2046 2047 2048

	/*
	 * 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);
2049
	mutex_unlock(&dev->shutdown_lock);
M
Matthew Wilcox 已提交
2050 2051
}

M
Matthew Wilcox 已提交
2052 2053
static int nvme_setup_prp_pools(struct nvme_dev *dev)
{
2054
	dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
M
Matthew Wilcox 已提交
2055 2056 2057 2058
						PAGE_SIZE, PAGE_SIZE, 0);
	if (!dev->prp_page_pool)
		return -ENOMEM;

2059
	/* Optimisation for I/Os between 4k and 128k */
2060
	dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
2061 2062 2063 2064 2065
						256, 256, 0);
	if (!dev->prp_small_pool) {
		dma_pool_destroy(dev->prp_page_pool);
		return -ENOMEM;
	}
M
Matthew Wilcox 已提交
2066 2067 2068 2069 2070 2071
	return 0;
}

static void nvme_release_prp_pools(struct nvme_dev *dev)
{
	dma_pool_destroy(dev->prp_page_pool);
2072
	dma_pool_destroy(dev->prp_small_pool);
M
Matthew Wilcox 已提交
2073 2074
}

2075
static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
2076
{
2077
	struct nvme_dev *dev = to_nvme_dev(ctrl);
2078

2079
	nvme_dbbuf_dma_free(dev);
2080
	put_device(dev->dev);
2081 2082
	if (dev->tagset.tags)
		blk_mq_free_tag_set(&dev->tagset);
2083 2084
	if (dev->ctrl.admin_q)
		blk_put_queue(dev->ctrl.admin_q);
2085
	kfree(dev->queues);
2086
	free_opal_dev(dev->ctrl.opal_dev);
2087 2088 2089
	kfree(dev);
}

2090 2091
static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status)
{
2092
	dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status);
2093 2094

	kref_get(&dev->ctrl.kref);
2095
	nvme_dev_disable(dev, false);
2096 2097 2098 2099
	if (!schedule_work(&dev->remove_work))
		nvme_put_ctrl(&dev->ctrl);
}

2100
static void nvme_reset_work(struct work_struct *work)
2101
{
2102 2103
	struct nvme_dev *dev =
		container_of(work, struct nvme_dev, ctrl.reset_work);
2104
	bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL);
2105
	int result = -ENODEV;
2106

2107
	if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING))
2108
		goto out;
2109

2110 2111 2112 2113
	/*
	 * If we're called to reset a live controller first shut it down before
	 * moving on.
	 */
2114
	if (dev->ctrl.ctrl_config & NVME_CC_ENABLE)
2115
		nvme_dev_disable(dev, false);
2116

2117
	result = nvme_pci_enable(dev);
2118
	if (result)
2119
		goto out;
2120

2121
	result = nvme_pci_configure_admin_queue(dev);
2122
	if (result)
2123
		goto out;
2124

M
Matias Bjørling 已提交
2125
	nvme_init_queue(dev->queues[0], 0);
K
Keith Busch 已提交
2126 2127
	result = nvme_alloc_admin_tags(dev);
	if (result)
2128
		goto out;
2129

2130 2131
	result = nvme_init_identify(&dev->ctrl);
	if (result)
2132
		goto out;
2133

2134 2135 2136 2137 2138 2139 2140 2141 2142
	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;
2143
	}
2144

2145 2146 2147 2148 2149 2150 2151
	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");
	}

2152 2153 2154
	if (dev->ctrl.hmpre)
		nvme_setup_host_mem(dev);

2155
	result = nvme_setup_io_queues(dev);
2156
	if (result)
2157
		goto out;
2158

2159 2160 2161 2162
	/*
	 * Keep the controller around but remove all namespaces if we don't have
	 * any working I/O queue.
	 */
2163
	if (dev->online_queues < 2) {
2164
		dev_warn(dev->ctrl.device, "IO queues not created\n");
2165
		nvme_kill_queues(&dev->ctrl);
2166
		nvme_remove_namespaces(&dev->ctrl);
2167
	} else {
2168
		nvme_start_queues(&dev->ctrl);
K
Keith Busch 已提交
2169
		nvme_wait_freeze(&dev->ctrl);
2170
		nvme_dev_add(dev);
K
Keith Busch 已提交
2171
		nvme_unfreeze(&dev->ctrl);
2172 2173
	}

2174 2175 2176 2177
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) {
		dev_warn(dev->ctrl.device, "failed to mark controller live\n");
		goto out;
	}
2178

2179
	nvme_start_ctrl(&dev->ctrl);
2180
	return;
2181

2182
 out:
2183
	nvme_remove_dead_ctrl(dev, result);
2184 2185
}

2186
static void nvme_remove_dead_ctrl_work(struct work_struct *work)
K
Keith Busch 已提交
2187
{
2188
	struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
2189
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
2190

2191
	nvme_kill_queues(&dev->ctrl);
K
Keith Busch 已提交
2192
	if (pci_get_drvdata(pdev))
K
Keith Busch 已提交
2193
		device_release_driver(&pdev->dev);
2194
	nvme_put_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2195 2196
}

2197
static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
T
Tejun Heo 已提交
2198
{
2199
	*val = readl(to_nvme_dev(ctrl)->bar + off);
2200
	return 0;
T
Tejun Heo 已提交
2201 2202
}

2203
static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
2204
{
2205 2206 2207
	writel(val, to_nvme_dev(ctrl)->bar + off);
	return 0;
}
2208

2209 2210 2211 2212
static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
{
	*val = readq(to_nvme_dev(ctrl)->bar + off);
	return 0;
2213 2214
}

2215
static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
M
Ming Lin 已提交
2216
	.name			= "pcie",
2217
	.module			= THIS_MODULE,
2218
	.flags			= NVME_F_METADATA_SUPPORTED,
2219
	.reg_read32		= nvme_pci_reg_read32,
2220
	.reg_write32		= nvme_pci_reg_write32,
2221
	.reg_read64		= nvme_pci_reg_read64,
2222
	.free_ctrl		= nvme_pci_free_ctrl,
2223
	.submit_async_event	= nvme_pci_submit_async_event,
2224
};
2225

2226 2227 2228 2229
static int nvme_dev_map(struct nvme_dev *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev->dev);

2230
	if (pci_request_mem_regions(pdev, "nvme"))
2231 2232
		return -ENODEV;

2233
	if (nvme_remap_bar(dev, NVME_REG_DBS + 4096))
2234 2235
		goto release;

M
Max Gurtovoy 已提交
2236
	return 0;
2237
  release:
M
Max Gurtovoy 已提交
2238 2239
	pci_release_mem_regions(pdev);
	return -ENODEV;
2240 2241
}

2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261
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;
}

2262
static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
M
Matthew Wilcox 已提交
2263
{
M
Matias Bjørling 已提交
2264
	int node, result = -ENOMEM;
M
Matthew Wilcox 已提交
2265
	struct nvme_dev *dev;
2266
	unsigned long quirks = id->driver_data;
M
Matthew Wilcox 已提交
2267

M
Matias Bjørling 已提交
2268 2269
	node = dev_to_node(&pdev->dev);
	if (node == NUMA_NO_NODE)
2270
		set_dev_node(&pdev->dev, first_memory_node);
M
Matias Bjørling 已提交
2271 2272

	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2273 2274
	if (!dev)
		return -ENOMEM;
M
Matias Bjørling 已提交
2275 2276
	dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2277 2278 2279
	if (!dev->queues)
		goto free;

2280
	dev->dev = get_device(&pdev->dev);
K
Keith Busch 已提交
2281
	pci_set_drvdata(pdev, dev);
2282

2283 2284 2285 2286
	result = nvme_dev_map(dev);
	if (result)
		goto free;

2287
	INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work);
2288
	INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
2289
	mutex_init(&dev->shutdown_lock);
K
Keith Busch 已提交
2290
	init_completion(&dev->ioq_wait);
M
Matthew Wilcox 已提交
2291

M
Matthew Wilcox 已提交
2292 2293
	result = nvme_setup_prp_pools(dev);
	if (result)
K
Keith Busch 已提交
2294
		goto put_pci;
2295

2296 2297
	quirks |= check_dell_samsung_bug(pdev);

2298
	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
2299
			quirks);
2300
	if (result)
2301
		goto release_pools;
2302

2303
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING);
2304 2305
	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));

2306
	queue_work(nvme_wq, &dev->ctrl.reset_work);
M
Matthew Wilcox 已提交
2307 2308
	return 0;

2309
 release_pools:
M
Matthew Wilcox 已提交
2310
	nvme_release_prp_pools(dev);
K
Keith Busch 已提交
2311
 put_pci:
2312
	put_device(dev->dev);
2313
	nvme_dev_unmap(dev);
M
Matthew Wilcox 已提交
2314 2315 2316 2317 2318 2319
 free:
	kfree(dev->queues);
	kfree(dev);
	return result;
}

2320
static void nvme_reset_prepare(struct pci_dev *pdev)
2321
{
K
Keith Busch 已提交
2322
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2323
	nvme_dev_disable(dev, false);
2324
}
2325

2326 2327
static void nvme_reset_done(struct pci_dev *pdev)
{
2328 2329
	struct nvme_dev *dev = pci_get_drvdata(pdev);
	nvme_reset_ctrl(&dev->ctrl);
2330 2331
}

2332 2333 2334
static void nvme_shutdown(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2335
	nvme_dev_disable(dev, true);
2336 2337
}

2338 2339 2340 2341 2342
/*
 * 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.
 */
2343
static void nvme_remove(struct pci_dev *pdev)
M
Matthew Wilcox 已提交
2344 2345
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
K
Keith Busch 已提交
2346

2347 2348
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING);

2349
	cancel_work_sync(&dev->ctrl.reset_work);
K
Keith Busch 已提交
2350
	pci_set_drvdata(pdev, NULL);
2351

2352
	if (!pci_device_is_present(pdev)) {
2353
		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
2354 2355
		nvme_dev_disable(dev, false);
	}
2356

2357
	flush_work(&dev->ctrl.reset_work);
2358 2359
	nvme_stop_ctrl(&dev->ctrl);
	nvme_remove_namespaces(&dev->ctrl);
2360
	nvme_dev_disable(dev, true);
2361
	nvme_free_host_mem(dev);
M
Matias Bjørling 已提交
2362
	nvme_dev_remove_admin(dev);
2363
	nvme_free_queues(dev, 0);
2364
	nvme_uninit_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2365
	nvme_release_prp_pools(dev);
2366
	nvme_dev_unmap(dev);
2367
	nvme_put_ctrl(&dev->ctrl);
M
Matthew Wilcox 已提交
2368 2369
}

K
Keith Busch 已提交
2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387
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;
}

2388
#ifdef CONFIG_PM_SLEEP
2389 2390 2391 2392 2393
static int nvme_suspend(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

2394
	nvme_dev_disable(ndev, true);
2395 2396 2397 2398 2399 2400 2401 2402
	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);

2403
	nvme_reset_ctrl(&ndev->ctrl);
K
Keith Busch 已提交
2404
	return 0;
2405
}
2406
#endif
2407 2408

static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
M
Matthew Wilcox 已提交
2409

K
Keith Busch 已提交
2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423
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
2424 2425
		dev_warn(dev->ctrl.device,
			"frozen state error detected, reset controller\n");
2426
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2427 2428
		return PCI_ERS_RESULT_NEED_RESET;
	case pci_channel_io_perm_failure:
K
Keith Busch 已提交
2429 2430
		dev_warn(dev->ctrl.device,
			"failure state error detected, request disconnect\n");
K
Keith Busch 已提交
2431 2432 2433 2434 2435 2436 2437 2438 2439
		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);

2440
	dev_info(dev->ctrl.device, "restart after slot reset\n");
K
Keith Busch 已提交
2441
	pci_restore_state(pdev);
2442
	nvme_reset_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2443 2444 2445 2446 2447 2448 2449 2450
	return PCI_ERS_RESULT_RECOVERED;
}

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

2451
static const struct pci_error_handlers nvme_err_handler = {
M
Matthew Wilcox 已提交
2452 2453 2454
	.error_detected	= nvme_error_detected,
	.slot_reset	= nvme_slot_reset,
	.resume		= nvme_error_resume,
2455 2456
	.reset_prepare	= nvme_reset_prepare,
	.reset_done	= nvme_reset_done,
M
Matthew Wilcox 已提交
2457 2458
};

2459
static const struct pci_device_id nvme_id_table[] = {
2460
	{ PCI_VDEVICE(INTEL, 0x0953),
2461
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2462
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2463 2464
	{ PCI_VDEVICE(INTEL, 0x0a53),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2465
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2466 2467
	{ PCI_VDEVICE(INTEL, 0x0a54),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2468
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2469 2470
	{ PCI_VDEVICE(INTEL, 0xf1a5),	/* Intel 600P/P3100 */
		.driver_data = NVME_QUIRK_NO_DEEPEST_PS },
2471 2472
	{ PCI_VDEVICE(INTEL, 0x5845),	/* Qemu emulated controller */
		.driver_data = NVME_QUIRK_IDENTIFY_CNS, },
2473 2474
	{ PCI_DEVICE(0x1c58, 0x0003),	/* HGST adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2475 2476
	{ PCI_DEVICE(0x1c5f, 0x0540),	/* Memblaze Pblaze4 adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2477 2478 2479 2480
	{ 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, },
M
Matthew Wilcox 已提交
2481
	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
2482
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
2483
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) },
M
Matthew Wilcox 已提交
2484 2485 2486 2487 2488 2489 2490 2491
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, nvme_id_table);

static struct pci_driver nvme_driver = {
	.name		= "nvme",
	.id_table	= nvme_id_table,
	.probe		= nvme_probe,
2492
	.remove		= nvme_remove,
2493
	.shutdown	= nvme_shutdown,
2494 2495 2496
	.driver		= {
		.pm	= &nvme_dev_pm_ops,
	},
K
Keith Busch 已提交
2497
	.sriov_configure = nvme_pci_sriov_configure,
M
Matthew Wilcox 已提交
2498 2499 2500 2501 2502
	.err_handler	= &nvme_err_handler,
};

static int __init nvme_init(void)
{
2503
	return pci_register_driver(&nvme_driver);
M
Matthew Wilcox 已提交
2504 2505 2506 2507 2508
}

static void __exit nvme_exit(void)
{
	pci_unregister_driver(&nvme_driver);
2509
	_nvme_check_size();
M
Matthew Wilcox 已提交
2510 2511 2512 2513
}

MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
MODULE_LICENSE("GPL");
2514
MODULE_VERSION("1.0");
M
Matthew Wilcox 已提交
2515 2516
module_init(nvme_init);
module_exit(nvme_exit);