pci.c 59.3 KB
Newer Older
M
Matthew Wilcox 已提交
1 2
/*
 * NVM Express device driver
3
 * Copyright (c) 2011-2014, Intel Corporation.
M
Matthew Wilcox 已提交
4 5 6 7 8 9 10 11 12 13 14
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 */

K
Keith Busch 已提交
15
#include <linux/aer.h>
16
#include <linux/bitops.h>
M
Matthew Wilcox 已提交
17
#include <linux/blkdev.h>
M
Matias Bjørling 已提交
18
#include <linux/blk-mq.h>
19
#include <linux/blk-mq-pci.h>
K
Keith Busch 已提交
20
#include <linux/cpu.h>
21
#include <linux/delay.h>
22
#include <linux/dmi.h>
M
Matthew Wilcox 已提交
23 24 25
#include <linux/errno.h>
#include <linux/fs.h>
#include <linux/genhd.h>
K
Keith Busch 已提交
26
#include <linux/hdreg.h>
27
#include <linux/idr.h>
M
Matthew Wilcox 已提交
28 29 30 31 32 33 34 35
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/io.h>
#include <linux/kdev_t.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
36
#include <linux/mutex.h>
M
Matthew Wilcox 已提交
37
#include <linux/pci.h>
38
#include <linux/poison.h>
39
#include <linux/ptrace.h>
M
Matthew Wilcox 已提交
40 41
#include <linux/sched.h>
#include <linux/slab.h>
K
Keith Busch 已提交
42
#include <linux/t10-pi.h>
43
#include <linux/timer.h>
M
Matthew Wilcox 已提交
44
#include <linux/types.h>
45
#include <linux/io-64-nonatomic-lo-hi.h>
K
Keith Busch 已提交
46
#include <asm/unaligned.h>
47
#include <linux/sed-opal.h>
48

49 50
#include "nvme.h"

51
#define NVME_Q_DEPTH		1024
J
Jens Axboe 已提交
52
#define NVME_AQ_DEPTH		256
M
Matthew Wilcox 已提交
53 54
#define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
#define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
55

56 57 58 59
/*
 * We handle AEN commands ourselves and don't even let the
 * block layer know about them.
 */
60
#define NVME_AQ_BLKMQ_DEPTH	(NVME_AQ_DEPTH - NVME_NR_AERS)
61

62 63 64
static int use_threaded_interrupts;
module_param(use_threaded_interrupts, int, 0);

65 66 67 68
static bool use_cmb_sqes = true;
module_param(use_cmb_sqes, bool, 0644);
MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");

K
Keith Busch 已提交
69
static struct workqueue_struct *nvme_workq;
70

71 72
struct nvme_dev;
struct nvme_queue;
73

74
static int nvme_reset(struct nvme_dev *dev);
J
Jens Axboe 已提交
75
static void nvme_process_cq(struct nvme_queue *nvmeq);
76
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
77

78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
/*
 * Represents an NVM Express device.  Each nvme_dev is a PCI function.
 */
struct nvme_dev {
	struct nvme_queue **queues;
	struct blk_mq_tag_set tagset;
	struct blk_mq_tag_set admin_tagset;
	u32 __iomem *dbs;
	struct device *dev;
	struct dma_pool *prp_page_pool;
	struct dma_pool *prp_small_pool;
	unsigned queue_count;
	unsigned online_queues;
	unsigned max_qid;
	int q_depth;
	u32 db_stride;
	void __iomem *bar;
	struct work_struct reset_work;
96
	struct work_struct remove_work;
97
	struct timer_list watchdog_timer;
98
	struct mutex shutdown_lock;
99 100 101 102 103
	bool subsystem;
	void __iomem *cmb;
	dma_addr_t cmb_dma_addr;
	u64 cmb_size;
	u32 cmbsz;
104
	u32 cmbloc;
105
	struct nvme_ctrl ctrl;
K
Keith Busch 已提交
106
	struct completion ioq_wait;
107 108 109 110
	u32 *dbbuf_dbs;
	dma_addr_t dbbuf_dbs_dma_addr;
	u32 *dbbuf_eis;
	dma_addr_t dbbuf_eis_dma_addr;
K
Keith Busch 已提交
111
};
112

113 114 115 116 117 118 119 120 121 122
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;
}

123 124 125 126 127
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
128 129 130 131 132 133
/*
 * 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
134
	struct nvme_dev *dev;
M
Matthew Wilcox 已提交
135 136
	spinlock_t q_lock;
	struct nvme_command *sq_cmds;
137
	struct nvme_command __iomem *sq_cmds_io;
M
Matthew Wilcox 已提交
138
	volatile struct nvme_completion *cqes;
139
	struct blk_mq_tags **tags;
M
Matthew Wilcox 已提交
140 141 142 143
	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
144
	s16 cq_vector;
M
Matthew Wilcox 已提交
145 146
	u16 sq_tail;
	u16 cq_head;
K
Keith Busch 已提交
147
	u16 qid;
148 149
	u8 cq_phase;
	u8 cqe_seen;
150 151 152 153
	u32 *dbbuf_sq_db;
	u32 *dbbuf_cq_db;
	u32 *dbbuf_sq_ei;
	u32 *dbbuf_cq_ei;
M
Matthew Wilcox 已提交
154 155
};

156 157 158
/*
 * 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
159
 * me express that.  Use nvme_init_iod to ensure there's enough space
160 161 162
 * allocated to store the PRP list.
 */
struct nvme_iod {
163
	struct nvme_request req;
C
Christoph Hellwig 已提交
164 165
	struct nvme_queue *nvmeq;
	int aborted;
166 167 168 169
	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;
170
	struct scatterlist meta_sg; /* metadata requires single contiguous buffer */
C
Christoph Hellwig 已提交
171 172
	struct scatterlist *sg;
	struct scatterlist inline_sg[0];
M
Matthew Wilcox 已提交
173 174 175 176 177 178 179 180 181 182 183 184
};

/*
 * 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);
185
	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
K
Keith Busch 已提交
186
	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
M
Matthew Wilcox 已提交
187 188 189 190
	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
K
Keith Busch 已提交
191
	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265
	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)) {
266
		dev_warn(dev->ctrl.device, "unable to set dbbuf\n");
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
		/* 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
298 299
}

300 301 302 303
/*
 * Max size of iod being embedded in the request payload
 */
#define NVME_INT_PAGES		2
304
#define NVME_INT_BYTES(dev)	(NVME_INT_PAGES * (dev)->ctrl.page_size)
305 306 307 308 309 310 311 312

/*
 * 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)
{
313 314
	unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size,
				      dev->ctrl.page_size);
315 316 317
	return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
}

C
Christoph Hellwig 已提交
318 319
static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
		unsigned int size, unsigned int nseg)
320
{
C
Christoph Hellwig 已提交
321 322 323
	return sizeof(__le64 *) * nvme_npages(size, dev) +
			sizeof(struct scatterlist) * nseg;
}
324

C
Christoph Hellwig 已提交
325 326 327 328
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);
329 330
}

M
Matias Bjørling 已提交
331 332
static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
				unsigned int hctx_idx)
333
{
M
Matias Bjørling 已提交
334 335 336
	struct nvme_dev *dev = data;
	struct nvme_queue *nvmeq = dev->queues[0];

337 338 339 340
	WARN_ON(hctx_idx != 0);
	WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
	WARN_ON(nvmeq->tags);

M
Matias Bjørling 已提交
341
	hctx->driver_data = nvmeq;
342
	nvmeq->tags = &dev->admin_tagset.tags[0];
M
Matias Bjørling 已提交
343
	return 0;
344 345
}

346 347 348 349 350 351 352
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;
}

353 354 355
static int nvme_admin_init_request(struct blk_mq_tag_set *set,
		struct request *req, unsigned int hctx_idx,
		unsigned int numa_node)
356
{
357
	struct nvme_dev *dev = set->driver_data;
C
Christoph Hellwig 已提交
358
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
359 360 361
	struct nvme_queue *nvmeq = dev->queues[0];

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
362
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
363
	return 0;
364 365
}

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

372 373
	if (!nvmeq->tags)
		nvmeq->tags = &dev->tagset.tags[hctx_idx];
M
Matthew Wilcox 已提交
374

375
	WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
M
Matias Bjørling 已提交
376 377
	hctx->driver_data = nvmeq;
	return 0;
M
Matthew Wilcox 已提交
378 379
}

380 381
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
382
{
383
	struct nvme_dev *dev = set->driver_data;
C
Christoph Hellwig 已提交
384
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
385 386 387
	struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];

	BUG_ON(!nvmeq);
C
Christoph Hellwig 已提交
388
	iod->nvmeq = nvmeq;
M
Matias Bjørling 已提交
389 390 391
	return 0;
}

392 393 394 395 396 397 398
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
399
/**
400
 * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
M
Matthew Wilcox 已提交
401 402 403 404 405
 * @nvmeq: The queue to use
 * @cmd: The command to send
 *
 * Safe to use from interrupt context
 */
406 407
static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
						struct nvme_command *cmd)
M
Matthew Wilcox 已提交
408
{
M
Matias Bjørling 已提交
409 410
	u16 tail = nvmeq->sq_tail;

411 412 413 414 415
	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
416 417
	if (++tail == nvmeq->q_depth)
		tail = 0;
418 419 420
	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
421 422 423
	nvmeq->sq_tail = tail;
}

C
Christoph Hellwig 已提交
424
static __le64 **iod_list(struct request *req)
M
Matthew Wilcox 已提交
425
{
C
Christoph Hellwig 已提交
426
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
427
	return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req));
M
Matthew Wilcox 已提交
428 429
}

430
static blk_status_t nvme_init_iod(struct request *rq, struct nvme_dev *dev)
431
{
C
Christoph Hellwig 已提交
432
	struct nvme_iod *iod = blk_mq_rq_to_pdu(rq);
433
	int nseg = blk_rq_nr_phys_segments(rq);
434
	unsigned int size = blk_rq_payload_bytes(rq);
435

C
Christoph Hellwig 已提交
436 437 438
	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)
439
			return BLK_STS_RESOURCE;
C
Christoph Hellwig 已提交
440 441
	} else {
		iod->sg = iod->inline_sg;
442 443
	}

C
Christoph Hellwig 已提交
444 445 446 447
	iod->aborted = 0;
	iod->npages = -1;
	iod->nents = 0;
	iod->length = size;
K
Keith Busch 已提交
448

449
	return BLK_STS_OK;
450 451
}

C
Christoph Hellwig 已提交
452
static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
453
{
C
Christoph Hellwig 已提交
454
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
455
	const int last_prp = dev->ctrl.page_size / 8 - 1;
456
	int i;
C
Christoph Hellwig 已提交
457
	__le64 **list = iod_list(req);
458 459 460 461 462 463 464 465 466 467
	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;
	}
468

C
Christoph Hellwig 已提交
469 470
	if (iod->sg != iod->inline_sg)
		kfree(iod->sg);
K
Keith Busch 已提交
471 472
}

473
#ifdef CONFIG_BLK_DEV_INTEGRITY
K
Keith Busch 已提交
474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
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);
518
	ts = ns->disk->queue->integrity.tuple_size;
K
Keith Busch 已提交
519 520 521 522 523 524 525 526

	for (i = 0; i < nlb; i++, virt++, phys++) {
		pi = (struct t10_pi_tuple *)p;
		dif_swap(phys, virt, pi);
		p += ts;
	}
	kunmap_atomic(pmap);
}
527 528 529 530 531 532 533 534 535 536 537 538 539
#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

540
static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
541
{
C
Christoph Hellwig 已提交
542
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
543
	struct dma_pool *pool;
544
	int length = blk_rq_payload_bytes(req);
545
	struct scatterlist *sg = iod->sg;
M
Matthew Wilcox 已提交
546 547
	int dma_len = sg_dma_len(sg);
	u64 dma_addr = sg_dma_address(sg);
548
	u32 page_size = dev->ctrl.page_size;
549
	int offset = dma_addr & (page_size - 1);
550
	__le64 *prp_list;
C
Christoph Hellwig 已提交
551
	__le64 **list = iod_list(req);
552
	dma_addr_t prp_dma;
553
	int nprps, i;
M
Matthew Wilcox 已提交
554

555
	length -= (page_size - offset);
M
Matthew Wilcox 已提交
556
	if (length <= 0)
557
		return true;
M
Matthew Wilcox 已提交
558

559
	dma_len -= (page_size - offset);
M
Matthew Wilcox 已提交
560
	if (dma_len) {
561
		dma_addr += (page_size - offset);
M
Matthew Wilcox 已提交
562 563 564 565 566 567
	} else {
		sg = sg_next(sg);
		dma_addr = sg_dma_address(sg);
		dma_len = sg_dma_len(sg);
	}

568
	if (length <= page_size) {
569
		iod->first_dma = dma_addr;
570
		return true;
571 572
	}

573
	nprps = DIV_ROUND_UP(length, page_size);
574 575
	if (nprps <= (256 / 8)) {
		pool = dev->prp_small_pool;
576
		iod->npages = 0;
577 578
	} else {
		pool = dev->prp_page_pool;
579
		iod->npages = 1;
580 581
	}

582
	prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
583
	if (!prp_list) {
584
		iod->first_dma = dma_addr;
585
		iod->npages = -1;
586
		return false;
587
	}
588 589
	list[0] = prp_list;
	iod->first_dma = prp_dma;
590 591
	i = 0;
	for (;;) {
592
		if (i == page_size >> 3) {
593
			__le64 *old_prp_list = prp_list;
594
			prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
595
			if (!prp_list)
596
				return false;
597
			list[iod->npages++] = prp_list;
598 599 600
			prp_list[0] = old_prp_list[i - 1];
			old_prp_list[i - 1] = cpu_to_le64(prp_dma);
			i = 1;
601 602
		}
		prp_list[i++] = cpu_to_le64(dma_addr);
603 604 605
		dma_len -= page_size;
		dma_addr += page_size;
		length -= page_size;
606 607 608 609 610 611 612 613
		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
614 615
	}

616
	return true;
M
Matthew Wilcox 已提交
617 618
}

619
static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req,
620
		struct nvme_command *cmnd)
621
{
C
Christoph Hellwig 已提交
622
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
C
Christoph Hellwig 已提交
623 624 625
	struct request_queue *q = req->q;
	enum dma_data_direction dma_dir = rq_data_dir(req) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE;
626
	blk_status_t ret = BLK_STS_IOERR;
627

628
	sg_init_table(iod->sg, blk_rq_nr_phys_segments(req));
C
Christoph Hellwig 已提交
629 630 631
	iod->nents = blk_rq_map_sg(q, req, iod->sg);
	if (!iod->nents)
		goto out;
632

633
	ret = BLK_STS_RESOURCE;
634 635
	if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir,
				DMA_ATTR_NO_WARN))
C
Christoph Hellwig 已提交
636
		goto out;
637

638
	if (!nvme_setup_prps(dev, req))
C
Christoph Hellwig 已提交
639
		goto out_unmap;
640

641
	ret = BLK_STS_IOERR;
C
Christoph Hellwig 已提交
642 643 644
	if (blk_integrity_rq(req)) {
		if (blk_rq_count_integrity_sg(q, req->bio) != 1)
			goto out_unmap;
645

646 647
		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
648
			goto out_unmap;
649

C
Christoph Hellwig 已提交
650 651
		if (rq_data_dir(req))
			nvme_dif_remap(req, nvme_dif_prep);
652

653
		if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir))
C
Christoph Hellwig 已提交
654
			goto out_unmap;
655
	}
M
Matthew Wilcox 已提交
656

657 658
	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
659
	if (blk_integrity_rq(req))
660
		cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
661
	return BLK_STS_OK;
M
Matthew Wilcox 已提交
662

C
Christoph Hellwig 已提交
663 664 665 666
out_unmap:
	dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
out:
	return ret;
M
Matthew Wilcox 已提交
667 668
}

C
Christoph Hellwig 已提交
669
static void nvme_unmap_data(struct nvme_dev *dev, struct request *req)
M
Matthew Wilcox 已提交
670
{
C
Christoph Hellwig 已提交
671
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
672 673 674 675 676 677 678 679
	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);
680
			dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir);
K
Keith Busch 已提交
681
		}
682
	}
K
Keith Busch 已提交
683

684
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
685
	nvme_free_iod(dev, req);
686
}
M
Matthew Wilcox 已提交
687

688 689 690
/*
 * NOTE: ns is NULL when called on the admin queue.
 */
691
static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
M
Matias Bjørling 已提交
692
			 const struct blk_mq_queue_data *bd)
693
{
M
Matias Bjørling 已提交
694 695
	struct nvme_ns *ns = hctx->queue->queuedata;
	struct nvme_queue *nvmeq = hctx->driver_data;
696
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
697
	struct request *req = bd->rq;
C
Christoph Hellwig 已提交
698
	struct nvme_command cmnd;
699
	blk_status_t ret = BLK_STS_OK;
700

K
Keith Busch 已提交
701 702 703 704 705
	/*
	 * If formated with metadata, require the block layer provide a buffer
	 * unless this namespace is formated such that the metadata can be
	 * stripped/generated by the controller with PRACT=1.
	 */
706
	if (ns && ns->ms && !blk_integrity_rq(req)) {
707
		if (!(ns->pi_type && ns->ms == 8) &&
708 709
		    !blk_rq_is_passthrough(req))
			return BLK_STS_NOTSUPP;
K
Keith Busch 已提交
710 711
	}

712
	ret = nvme_setup_cmd(ns, req, &cmnd);
713
	if (ret)
C
Christoph Hellwig 已提交
714
		return ret;
M
Matias Bjørling 已提交
715

716
	ret = nvme_init_iod(req, dev);
717
	if (ret)
718
		goto out_free_cmd;
M
Matias Bjørling 已提交
719

720
	if (blk_rq_nr_phys_segments(req)) {
721
		ret = nvme_map_data(dev, req, &cmnd);
722 723 724
		if (ret)
			goto out_cleanup_iod;
	}
M
Matias Bjørling 已提交
725

726
	blk_mq_start_request(req);
M
Matias Bjørling 已提交
727

C
Christoph Hellwig 已提交
728
	spin_lock_irq(&nvmeq->q_lock);
729
	if (unlikely(nvmeq->cq_vector < 0)) {
730
		ret = BLK_STS_IOERR;
731
		spin_unlock_irq(&nvmeq->q_lock);
732
		goto out_cleanup_iod;
733
	}
C
Christoph Hellwig 已提交
734
	__nvme_submit_cmd(nvmeq, &cmnd);
M
Matias Bjørling 已提交
735 736
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
737
	return BLK_STS_OK;
738
out_cleanup_iod:
C
Christoph Hellwig 已提交
739
	nvme_free_iod(dev, req);
740 741
out_free_cmd:
	nvme_cleanup_cmd(req);
C
Christoph Hellwig 已提交
742
	return ret;
M
Matthew Wilcox 已提交
743
}
K
Keith Busch 已提交
744

745
static void nvme_pci_complete_rq(struct request *req)
746
{
C
Christoph Hellwig 已提交
747
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
M
Matias Bjørling 已提交
748

749 750
	nvme_unmap_data(iod->nvmeq->dev, req);
	nvme_complete_rq(req);
M
Matthew Wilcox 已提交
751 752
}

753 754 755 756 757 758 759
/* 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
760
static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag)
M
Matthew Wilcox 已提交
761
{
M
Matthew Wilcox 已提交
762
	u16 head, phase;
M
Matthew Wilcox 已提交
763 764

	head = nvmeq->cq_head;
M
Matthew Wilcox 已提交
765
	phase = nvmeq->cq_phase;
M
Matthew Wilcox 已提交
766

767
	while (nvme_cqe_valid(nvmeq, head, phase)) {
M
Matthew Wilcox 已提交
768
		struct nvme_completion cqe = nvmeq->cqes[head];
769
		struct request *req;
770

M
Matthew Wilcox 已提交
771 772
		if (++head == nvmeq->q_depth) {
			head = 0;
M
Matthew Wilcox 已提交
773
			phase = !phase;
M
Matthew Wilcox 已提交
774
		}
775

J
Jens Axboe 已提交
776 777
		if (tag && *tag == cqe.command_id)
			*tag = -1;
778

779
		if (unlikely(cqe.command_id >= nvmeq->q_depth)) {
780
			dev_warn(nvmeq->dev->ctrl.device,
781 782 783 784 785
				"invalid id %d completed on queue %d\n",
				cqe.command_id, le16_to_cpu(cqe.sq_id));
			continue;
		}

786 787 788 789 790 791 792 793
		/*
		 * 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)) {
794 795
			nvme_complete_async_event(&nvmeq->dev->ctrl,
					cqe.status, &cqe.result);
796 797 798
			continue;
		}

799
		req = blk_mq_tag_to_rq(*nvmeq->tags, cqe.command_id);
800
		nvme_end_request(req, cqe.status, cqe.result);
M
Matthew Wilcox 已提交
801 802
	}

M
Matthew Wilcox 已提交
803
	if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
J
Jens Axboe 已提交
804
		return;
M
Matthew Wilcox 已提交
805

806
	if (likely(nvmeq->cq_vector >= 0))
807 808 809
		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
810
	nvmeq->cq_head = head;
M
Matthew Wilcox 已提交
811
	nvmeq->cq_phase = phase;
M
Matthew Wilcox 已提交
812

813
	nvmeq->cqe_seen = 1;
J
Jens Axboe 已提交
814 815 816 817 818
}

static void nvme_process_cq(struct nvme_queue *nvmeq)
{
	__nvme_process_cq(nvmeq, NULL);
M
Matthew Wilcox 已提交
819 820 821
}

static irqreturn_t nvme_irq(int irq, void *data)
822 823 824 825
{
	irqreturn_t result;
	struct nvme_queue *nvmeq = data;
	spin_lock(&nvmeq->q_lock);
826 827 828
	nvme_process_cq(nvmeq);
	result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
	nvmeq->cqe_seen = 0;
829 830 831 832 833 834 835
	spin_unlock(&nvmeq->q_lock);
	return result;
}

static irqreturn_t nvme_irq_check(int irq, void *data)
{
	struct nvme_queue *nvmeq = data;
836 837 838
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
		return IRQ_WAKE_THREAD;
	return IRQ_NONE;
839 840
}

K
Keith Busch 已提交
841
static int __nvme_poll(struct nvme_queue *nvmeq, unsigned int tag)
J
Jens Axboe 已提交
842
{
843
	if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) {
J
Jens Axboe 已提交
844 845 846 847 848 849 850 851 852 853 854
		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
855 856 857 858 859 860 861
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);
}

862
static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx)
M
Matthew Wilcox 已提交
863
{
864
	struct nvme_dev *dev = to_nvme_dev(ctrl);
865
	struct nvme_queue *nvmeq = dev->queues[0];
M
Matias Bjørling 已提交
866
	struct nvme_command c;
M
Matthew Wilcox 已提交
867

M
Matias Bjørling 已提交
868 869
	memset(&c, 0, sizeof(c));
	c.common.opcode = nvme_admin_async_event;
870
	c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx;
871

872
	spin_lock_irq(&nvmeq->q_lock);
873
	__nvme_submit_cmd(nvmeq, &c);
874
	spin_unlock_irq(&nvmeq->q_lock);
875 876
}

M
Matthew Wilcox 已提交
877
static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
878
{
M
Matthew Wilcox 已提交
879 880 881 882 883 884
	struct nvme_command c;

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

885
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
886 887 888 889 890 891 892 893
}

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;

894 895 896 897
	/*
	 * 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
898 899 900 901 902 903 904 905
	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);

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

static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
						struct nvme_queue *nvmeq)
{
	struct nvme_command c;
913
	int flags = NVME_QUEUE_PHYS_CONTIG;
M
Matthew Wilcox 已提交
914

915 916 917 918
	/*
	 * 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
919 920 921 922 923 924 925 926
	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);

927
	return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
M
Matthew Wilcox 已提交
928 929 930 931 932 933 934 935 936 937 938 939
}

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);
}

940
static void abort_endio(struct request *req, blk_status_t error)
941
{
C
Christoph Hellwig 已提交
942 943
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
944

945 946
	dev_warn(nvmeq->dev->ctrl.device,
		 "Abort status: 0x%x", nvme_req(req)->status);
947 948
	atomic_inc(&nvmeq->dev->ctrl.abort_limit);
	blk_mq_free_request(req);
949 950
}

951
static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
K
Keith Busch 已提交
952
{
C
Christoph Hellwig 已提交
953 954
	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
	struct nvme_queue *nvmeq = iod->nvmeq;
K
Keith Busch 已提交
955
	struct nvme_dev *dev = nvmeq->dev;
M
Matias Bjørling 已提交
956 957
	struct request *abort_req;
	struct nvme_command cmd;
K
Keith Busch 已提交
958

K
Keith Busch 已提交
959 960 961 962 963 964 965 966 967 968
	/*
	 * 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;
	}

969
	/*
970 971 972 973 974
	 * 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.
	 */
975
	if (dev->ctrl.state == NVME_CTRL_RESETTING) {
976
		dev_warn(dev->ctrl.device,
977 978
			 "I/O %d QID %d timeout, disable controller\n",
			 req->tag, nvmeq->qid);
979
		nvme_dev_disable(dev, false);
980
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
981
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
982 983
	}

984 985 986 987
	/*
 	 * 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.
988
	 */
C
Christoph Hellwig 已提交
989
	if (!nvmeq->qid || iod->aborted) {
990
		dev_warn(dev->ctrl.device,
991 992
			 "I/O %d QID %d timeout, reset controller\n",
			 req->tag, nvmeq->qid);
993
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
994
		nvme_reset(dev);
K
Keith Busch 已提交
995

996 997 998 999
		/*
		 * Mark the request as handled, since the inline shutdown
		 * forces all outstanding requests to complete.
		 */
1000
		nvme_req(req)->flags |= NVME_REQ_CANCELLED;
1001
		return BLK_EH_HANDLED;
K
Keith Busch 已提交
1002 1003
	}

1004
	if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) {
1005
		atomic_inc(&dev->ctrl.abort_limit);
1006
		return BLK_EH_RESET_TIMER;
1007
	}
1008
	iod->aborted = 1;
M
Matias Bjørling 已提交
1009

K
Keith Busch 已提交
1010 1011
	memset(&cmd, 0, sizeof(cmd));
	cmd.abort.opcode = nvme_admin_abort_cmd;
M
Matias Bjørling 已提交
1012
	cmd.abort.cid = req->tag;
K
Keith Busch 已提交
1013 1014
	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);

1015 1016 1017
	dev_warn(nvmeq->dev->ctrl.device,
		"I/O %d QID %d timeout, aborting\n",
		 req->tag, nvmeq->qid);
1018 1019

	abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
1020
			BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
1021 1022 1023 1024 1025 1026 1027 1028
	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
1029

1030 1031 1032 1033 1034 1035
	/*
	 * 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
1036 1037
}

M
Matias Bjørling 已提交
1038 1039
static void nvme_free_queue(struct nvme_queue *nvmeq)
{
1040 1041
	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
1042 1043
	if (nvmeq->sq_cmds)
		dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
1044 1045 1046 1047
					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
	kfree(nvmeq);
}

1048
static void nvme_free_queues(struct nvme_dev *dev, int lowest)
1049 1050 1051
{
	int i;

1052
	for (i = dev->queue_count - 1; i >= lowest; i--) {
M
Matias Bjørling 已提交
1053
		struct nvme_queue *nvmeq = dev->queues[i];
1054
		dev->queue_count--;
M
Matias Bjørling 已提交
1055
		dev->queues[i] = NULL;
K
Keith Busch 已提交
1056
		nvme_free_queue(nvmeq);
1057
	}
1058 1059
}

K
Keith Busch 已提交
1060 1061 1062 1063 1064
/**
 * 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
1065
{
K
Keith Busch 已提交
1066
	int vector;
M
Matthew Wilcox 已提交
1067

1068
	spin_lock_irq(&nvmeq->q_lock);
K
Keith Busch 已提交
1069 1070 1071 1072
	if (nvmeq->cq_vector == -1) {
		spin_unlock_irq(&nvmeq->q_lock);
		return 1;
	}
1073
	vector = nvmeq->cq_vector;
K
Keith Busch 已提交
1074
	nvmeq->dev->online_queues--;
K
Keith Busch 已提交
1075
	nvmeq->cq_vector = -1;
1076 1077
	spin_unlock_irq(&nvmeq->q_lock);

1078
	if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
1079
		blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q);
1080

1081
	pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq);
M
Matthew Wilcox 已提交
1082

K
Keith Busch 已提交
1083 1084
	return 0;
}
M
Matthew Wilcox 已提交
1085

1086
static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
K
Keith Busch 已提交
1087
{
1088
	struct nvme_queue *nvmeq = dev->queues[0];
K
Keith Busch 已提交
1089 1090 1091 1092 1093 1094

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

1095 1096 1097 1098 1099
	if (shutdown)
		nvme_shutdown_ctrl(&dev->ctrl);
	else
		nvme_disable_ctrl(&dev->ctrl, lo_hi_readq(
						dev->bar + NVME_REG_CAP));
1100 1101 1102 1103

	spin_lock_irq(&nvmeq->q_lock);
	nvme_process_cq(nvmeq);
	spin_unlock_irq(&nvmeq->q_lock);
M
Matthew Wilcox 已提交
1104 1105
}

1106 1107 1108 1109
static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
				int entry_size)
{
	int q_depth = dev->q_depth;
1110 1111
	unsigned q_size_aligned = roundup(q_depth * entry_size,
					  dev->ctrl.page_size);
1112 1113

	if (q_size_aligned * nr_io_queues > dev->cmb_size) {
1114
		u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
1115
		mem_per_q = round_down(mem_per_q, dev->ctrl.page_size);
1116
		q_depth = div_u64(mem_per_q, entry_size);
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133

		/*
		 * 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)) {
1134 1135
		unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth),
						      dev->ctrl.page_size);
1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
		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
1148
static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
1149
							int depth, int node)
M
Matthew Wilcox 已提交
1150
{
1151 1152
	struct nvme_queue *nvmeq = kzalloc_node(sizeof(*nvmeq), GFP_KERNEL,
							node);
M
Matthew Wilcox 已提交
1153 1154 1155
	if (!nvmeq)
		return NULL;

1156
	nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
J
Joe Perches 已提交
1157
					  &nvmeq->cq_dma_addr, GFP_KERNEL);
M
Matthew Wilcox 已提交
1158 1159 1160
	if (!nvmeq->cqes)
		goto free_nvmeq;

1161
	if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
M
Matthew Wilcox 已提交
1162 1163
		goto free_cqdma;

1164
	nvmeq->q_dmadev = dev->dev;
M
Matthew Wilcox 已提交
1165
	nvmeq->dev = dev;
M
Matthew Wilcox 已提交
1166 1167
	spin_lock_init(&nvmeq->q_lock);
	nvmeq->cq_head = 0;
M
Matthew Wilcox 已提交
1168
	nvmeq->cq_phase = 1;
1169
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
M
Matthew Wilcox 已提交
1170
	nvmeq->q_depth = depth;
K
Keith Busch 已提交
1171
	nvmeq->qid = qid;
1172
	nvmeq->cq_vector = -1;
M
Matias Bjørling 已提交
1173
	dev->queues[qid] = nvmeq;
1174 1175
	dev->queue_count++;

M
Matthew Wilcox 已提交
1176 1177 1178
	return nvmeq;

 free_cqdma:
1179
	dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
M
Matthew Wilcox 已提交
1180 1181 1182 1183 1184 1185
							nvmeq->cq_dma_addr);
 free_nvmeq:
	kfree(nvmeq);
	return NULL;
}

1186
static int queue_request_irq(struct nvme_queue *nvmeq)
1187
{
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
	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);
	}
1198 1199
}

1200
static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
M
Matthew Wilcox 已提交
1201
{
1202
	struct nvme_dev *dev = nvmeq->dev;
M
Matthew Wilcox 已提交
1203

1204
	spin_lock_irq(&nvmeq->q_lock);
1205 1206 1207
	nvmeq->sq_tail = 0;
	nvmeq->cq_head = 0;
	nvmeq->cq_phase = 1;
1208
	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
1209
	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
1210
	nvme_dbbuf_init(dev, nvmeq, qid);
K
Keith Busch 已提交
1211
	dev->online_queues++;
1212
	spin_unlock_irq(&nvmeq->q_lock);
1213 1214 1215 1216 1217 1218
}

static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
{
	struct nvme_dev *dev = nvmeq->dev;
	int result;
1219

K
Keith Busch 已提交
1220
	nvmeq->cq_vector = qid - 1;
M
Matthew Wilcox 已提交
1221 1222
	result = adapter_alloc_cq(dev, qid, nvmeq);
	if (result < 0)
1223
		return result;
M
Matthew Wilcox 已提交
1224 1225 1226 1227 1228

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

1229
	result = queue_request_irq(nvmeq);
M
Matthew Wilcox 已提交
1230 1231 1232
	if (result < 0)
		goto release_sq;

1233 1234
	nvme_init_queue(nvmeq, qid);
	return result;
M
Matthew Wilcox 已提交
1235 1236 1237 1238 1239

 release_sq:
	adapter_delete_sq(dev, qid);
 release_cq:
	adapter_delete_cq(dev, qid);
1240
	return result;
M
Matthew Wilcox 已提交
1241 1242
}

1243
static const struct blk_mq_ops nvme_mq_admin_ops = {
1244
	.queue_rq	= nvme_queue_rq,
1245
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1246
	.init_hctx	= nvme_admin_init_hctx,
1247
	.exit_hctx      = nvme_admin_exit_hctx,
M
Matias Bjørling 已提交
1248 1249 1250 1251
	.init_request	= nvme_admin_init_request,
	.timeout	= nvme_timeout,
};

1252
static const struct blk_mq_ops nvme_mq_ops = {
M
Matias Bjørling 已提交
1253
	.queue_rq	= nvme_queue_rq,
1254
	.complete	= nvme_pci_complete_rq,
M
Matias Bjørling 已提交
1255 1256
	.init_hctx	= nvme_init_hctx,
	.init_request	= nvme_init_request,
1257
	.map_queues	= nvme_pci_map_queues,
M
Matias Bjørling 已提交
1258
	.timeout	= nvme_timeout,
J
Jens Axboe 已提交
1259
	.poll		= nvme_poll,
M
Matias Bjørling 已提交
1260 1261
};

1262 1263
static void nvme_dev_remove_admin(struct nvme_dev *dev)
{
1264
	if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) {
1265 1266 1267 1268 1269 1270
		/*
		 * 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);
1271
		blk_cleanup_queue(dev->ctrl.admin_q);
1272 1273 1274 1275
		blk_mq_free_tag_set(&dev->admin_tagset);
	}
}

M
Matias Bjørling 已提交
1276 1277
static int nvme_alloc_admin_tags(struct nvme_dev *dev)
{
1278
	if (!dev->ctrl.admin_q) {
M
Matias Bjørling 已提交
1279 1280
		dev->admin_tagset.ops = &nvme_mq_admin_ops;
		dev->admin_tagset.nr_hw_queues = 1;
K
Keith Busch 已提交
1281 1282 1283 1284 1285 1286

		/*
		 * 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
1287
		dev->admin_tagset.timeout = ADMIN_TIMEOUT;
1288
		dev->admin_tagset.numa_node = dev_to_node(dev->dev);
1289
		dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
1290
		dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED;
M
Matias Bjørling 已提交
1291 1292 1293 1294 1295
		dev->admin_tagset.driver_data = dev;

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

1296 1297
		dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset);
		if (IS_ERR(dev->ctrl.admin_q)) {
M
Matias Bjørling 已提交
1298 1299 1300
			blk_mq_free_tag_set(&dev->admin_tagset);
			return -ENOMEM;
		}
1301
		if (!blk_get_queue(dev->ctrl.admin_q)) {
1302
			nvme_dev_remove_admin(dev);
1303
			dev->ctrl.admin_q = NULL;
1304 1305
			return -ENODEV;
		}
K
Keith Busch 已提交
1306
	} else
1307
		blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true);
M
Matias Bjørling 已提交
1308 1309 1310 1311

	return 0;
}

1312
static int nvme_configure_admin_queue(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1313
{
1314
	int result;
M
Matthew Wilcox 已提交
1315
	u32 aqa;
1316
	u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
M
Matthew Wilcox 已提交
1317 1318
	struct nvme_queue *nvmeq;

1319
	dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ?
1320 1321
						NVME_CAP_NSSRC(cap) : 0;

1322 1323 1324
	if (dev->subsystem &&
	    (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO))
		writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS);
1325

1326
	result = nvme_disable_ctrl(&dev->ctrl, cap);
1327 1328
	if (result < 0)
		return result;
M
Matthew Wilcox 已提交
1329

M
Matias Bjørling 已提交
1330
	nvmeq = dev->queues[0];
1331
	if (!nvmeq) {
1332 1333
		nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH,
					dev_to_node(dev->dev));
1334 1335 1336
		if (!nvmeq)
			return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1337 1338 1339 1340

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

1341 1342 1343
	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
1344

1345
	result = nvme_enable_ctrl(&dev->ctrl, cap);
1346
	if (result)
K
Keith Busch 已提交
1347
		return result;
M
Matias Bjørling 已提交
1348

K
Keith Busch 已提交
1349
	nvmeq->cq_vector = 0;
1350
	result = queue_request_irq(nvmeq);
1351 1352
	if (result) {
		nvmeq->cq_vector = -1;
K
Keith Busch 已提交
1353
		return result;
1354
	}
1355

M
Matthew Wilcox 已提交
1356 1357 1358
	return result;
}

1359 1360 1361 1362 1363 1364 1365 1366 1367
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. */
1368
	if (dev->ctrl.state == NVME_CTRL_RESETTING)
1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385
		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;
}

1386 1387 1388 1389 1390 1391 1392 1393 1394
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)
1395
		dev_warn(dev->ctrl.device,
1396 1397 1398
			 "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n",
			 csts, pci_status);
	else
1399
		dev_warn(dev->ctrl.device,
1400 1401 1402 1403
			 "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n",
			 csts, result);
}

1404
static void nvme_watchdog_timer(unsigned long data)
1405
{
1406 1407
	struct nvme_dev *dev = (struct nvme_dev *)data;
	u32 csts = readl(dev->bar + NVME_REG_CSTS);
1408

1409 1410
	/* Skip controllers under certain specific conditions. */
	if (nvme_should_reset(dev, csts)) {
K
Keith Busch 已提交
1411
		if (!nvme_reset(dev))
1412
			nvme_warn_reset(dev, csts);
1413
		return;
1414
	}
1415 1416

	mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
1417 1418
}

1419
static int nvme_create_io_queues(struct nvme_dev *dev)
K
Keith Busch 已提交
1420
{
1421
	unsigned i, max;
1422
	int ret = 0;
K
Keith Busch 已提交
1423

1424
	for (i = dev->queue_count; i <= dev->max_qid; i++) {
1425 1426 1427
		/* 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))) {
1428
			ret = -ENOMEM;
K
Keith Busch 已提交
1429
			break;
1430 1431
		}
	}
K
Keith Busch 已提交
1432

1433 1434
	max = min(dev->max_qid, dev->queue_count - 1);
	for (i = dev->online_queues; i <= max; i++) {
1435
		ret = nvme_create_queue(dev->queues[i], i);
K
Keith Busch 已提交
1436
		if (ret)
K
Keith Busch 已提交
1437
			break;
M
Matthew Wilcox 已提交
1438
	}
1439 1440 1441 1442 1443 1444 1445 1446

	/*
	 * 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
1447 1448
}

1449 1450 1451 1452 1453 1454
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));

1455
	return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz  : x%08x\n",
1456 1457 1458 1459
		       ndev->cmbloc, ndev->cmbsz);
}
static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL);

1460 1461 1462 1463 1464 1465 1466 1467
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;

1468
	dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ);
1469 1470
	if (!(NVME_CMB_SZ(dev->cmbsz)))
		return NULL;
1471
	dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC);
1472

1473 1474
	if (!use_cmb_sqes)
		return NULL;
1475 1476 1477

	szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
	size = szu * NVME_CMB_SZ(dev->cmbsz);
1478 1479
	offset = szu * NVME_CMB_OFST(dev->cmbloc);
	bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc));
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491

	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;

1492
	dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset;
1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
	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;
1507 1508 1509 1510 1511
		if (dev->cmbsz) {
			sysfs_remove_file_from_group(&dev->ctrl.device->kobj,
						     &dev_attr_cmb.attr, NULL);
			dev->cmbsz = 0;
		}
1512 1513 1514
	}
}

K
Keith Busch 已提交
1515 1516
static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
{
1517
	return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
K
Keith Busch 已提交
1518 1519
}

1520
static int nvme_setup_io_queues(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1521
{
M
Matias Bjørling 已提交
1522
	struct nvme_queue *adminq = dev->queues[0];
1523
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1524
	int result, nr_io_queues, size;
M
Matthew Wilcox 已提交
1525

1526
	nr_io_queues = num_online_cpus();
C
Christoph Hellwig 已提交
1527 1528
	result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
	if (result < 0)
M
Matthew Wilcox 已提交
1529
		return result;
C
Christoph Hellwig 已提交
1530

1531
	if (nr_io_queues == 0)
1532
		return 0;
M
Matthew Wilcox 已提交
1533

1534 1535 1536 1537 1538 1539 1540 1541 1542
	if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) {
		result = nvme_cmb_qdepth(dev, nr_io_queues,
				sizeof(struct nvme_command));
		if (result > 0)
			dev->q_depth = result;
		else
			nvme_release_cmb(dev);
	}

K
Keith Busch 已提交
1543 1544
	size = db_bar_size(dev, nr_io_queues);
	if (size > 8192) {
1545
		iounmap(dev->bar);
K
Keith Busch 已提交
1546 1547 1548 1549 1550 1551 1552 1553
		do {
			dev->bar = ioremap(pci_resource_start(pdev, 0), size);
			if (dev->bar)
				break;
			if (!--nr_io_queues)
				return -ENOMEM;
			size = db_bar_size(dev, nr_io_queues);
		} while (1);
1554
		dev->dbs = dev->bar + 4096;
1555
		adminq->q_db = dev->dbs;
1556 1557
	}

K
Keith Busch 已提交
1558
	/* Deregister the admin queue's interrupt */
1559
	pci_free_irq(pdev, 0, adminq);
K
Keith Busch 已提交
1560

1561 1562 1563 1564
	/*
	 * If we enable msix early due to not intx, disable it again before
	 * setting up the full range we need.
	 */
1565 1566 1567 1568 1569 1570
	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
1571

1572 1573 1574 1575 1576 1577 1578
	/*
	 * 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.
	 */

1579
	result = queue_request_irq(adminq);
1580 1581
	if (result) {
		adminq->cq_vector = -1;
K
Keith Busch 已提交
1582
		return result;
1583
	}
1584
	return nvme_create_io_queues(dev);
M
Matthew Wilcox 已提交
1585 1586
}

1587
static void nvme_del_queue_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1588
{
K
Keith Busch 已提交
1589
	struct nvme_queue *nvmeq = req->end_io_data;
1590

K
Keith Busch 已提交
1591 1592
	blk_mq_free_request(req);
	complete(&nvmeq->dev->ioq_wait);
K
Keith Busch 已提交
1593 1594
}

1595
static void nvme_del_cq_end(struct request *req, blk_status_t error)
K
Keith Busch 已提交
1596
{
K
Keith Busch 已提交
1597
	struct nvme_queue *nvmeq = req->end_io_data;
K
Keith Busch 已提交
1598

K
Keith Busch 已提交
1599 1600 1601
	if (!error) {
		unsigned long flags;

1602 1603 1604 1605 1606 1607 1608
		/*
		 * 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
1609 1610
		nvme_process_cq(nvmeq);
		spin_unlock_irqrestore(&nvmeq->q_lock, flags);
K
Keith Busch 已提交
1611
	}
K
Keith Busch 已提交
1612 1613

	nvme_del_queue_end(req, error);
K
Keith Busch 已提交
1614 1615
}

K
Keith Busch 已提交
1616
static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
1617
{
K
Keith Busch 已提交
1618 1619 1620
	struct request_queue *q = nvmeq->dev->ctrl.admin_q;
	struct request *req;
	struct nvme_command cmd;
1621

K
Keith Busch 已提交
1622 1623 1624
	memset(&cmd, 0, sizeof(cmd));
	cmd.delete_queue.opcode = opcode;
	cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
1625

1626
	req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
K
Keith Busch 已提交
1627 1628
	if (IS_ERR(req))
		return PTR_ERR(req);
1629

K
Keith Busch 已提交
1630 1631 1632 1633 1634 1635 1636
	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;
1637 1638
}

1639
static void nvme_disable_io_queues(struct nvme_dev *dev, int queues)
K
Keith Busch 已提交
1640
{
1641
	int pass;
K
Keith Busch 已提交
1642 1643
	unsigned long timeout;
	u8 opcode = nvme_admin_delete_sq;
K
Keith Busch 已提交
1644

K
Keith Busch 已提交
1645
	for (pass = 0; pass < 2; pass++) {
K
Keith Busch 已提交
1646
		int sent = 0, i = queues;
K
Keith Busch 已提交
1647 1648 1649 1650

		reinit_completion(&dev->ioq_wait);
 retry:
		timeout = ADMIN_TIMEOUT;
1651 1652
		for (; i > 0; i--, sent++)
			if (nvme_delete_queue(dev->queues[i], opcode))
K
Keith Busch 已提交
1653
				break;
1654

K
Keith Busch 已提交
1655 1656 1657 1658 1659 1660 1661 1662 1663
		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
1664 1665
}

1666 1667 1668 1669 1670 1671
/*
 * 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.
 */
1672
static int nvme_dev_add(struct nvme_dev *dev)
M
Matthew Wilcox 已提交
1673
{
1674
	if (!dev->ctrl.tagset) {
1675 1676 1677 1678 1679
		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
1680
				min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
1681 1682 1683
		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
1684

1685 1686
		if (blk_mq_alloc_tag_set(&dev->tagset))
			return 0;
1687
		dev->ctrl.tagset = &dev->tagset;
1688 1689

		nvme_dbbuf_set(dev);
1690 1691 1692 1693 1694
	} 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);
1695
	}
1696

K
Keith Busch 已提交
1697
	return 0;
M
Matthew Wilcox 已提交
1698 1699
}

1700
static int nvme_pci_enable(struct nvme_dev *dev)
1701
{
K
Keith Busch 已提交
1702
	u64 cap;
1703
	int result = -ENOMEM;
1704
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1705 1706 1707 1708 1709 1710

	if (pci_enable_device_mem(pdev))
		return result;

	pci_set_master(pdev);

1711 1712
	if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
	    dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
1713
		goto disable;
1714

1715
	if (readl(dev->bar + NVME_REG_CSTS) == -1) {
K
Keith Busch 已提交
1716
		result = -ENODEV;
1717
		goto disable;
K
Keith Busch 已提交
1718
	}
1719 1720

	/*
1721 1722 1723
	 * 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.
1724
	 */
1725 1726 1727
	result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
	if (result < 0)
		return result;
1728

1729 1730
	cap = lo_hi_readq(dev->bar + NVME_REG_CAP);

K
Keith Busch 已提交
1731 1732
	dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
	dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
1733
	dev->dbs = dev->bar + 4096;
1734 1735 1736 1737 1738 1739 1740

	/*
	 * 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;
1741 1742
		dev_warn(dev->ctrl.device, "detected Apple NVMe controller, "
			"set queue depth=%u to work around controller resets\n",
1743 1744 1745
			dev->q_depth);
	}

1746 1747 1748 1749 1750 1751 1752 1753
	/*
	 * 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.
	 */

1754
	if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) {
1755
		dev->cmb = nvme_map_cmb(dev);
1756

1757 1758 1759
		if (dev->cmbsz) {
			if (sysfs_add_file_to_group(&dev->ctrl.device->kobj,
						    &dev_attr_cmb.attr, NULL))
1760
				dev_warn(dev->ctrl.device,
1761 1762 1763 1764
					 "failed to add sysfs attribute for CMB\n");
		}
	}

K
Keith Busch 已提交
1765 1766
	pci_enable_pcie_error_reporting(pdev);
	pci_save_state(pdev);
1767 1768 1769 1770 1771 1772 1773 1774
	return 0;

 disable:
	pci_disable_device(pdev);
	return result;
}

static void nvme_dev_unmap(struct nvme_dev *dev)
1775 1776 1777
{
	if (dev->bar)
		iounmap(dev->bar);
1778
	pci_release_mem_regions(to_pci_dev(dev->dev));
1779 1780 1781
}

static void nvme_pci_disable(struct nvme_dev *dev)
1782
{
1783 1784
	struct pci_dev *pdev = to_pci_dev(dev->dev);

1785
	nvme_release_cmb(dev);
1786
	pci_free_irq_vectors(pdev);
1787

K
Keith Busch 已提交
1788 1789
	if (pci_is_enabled(pdev)) {
		pci_disable_pcie_error_reporting(pdev);
1790
		pci_disable_device(pdev);
K
Keith Busch 已提交
1791 1792 1793
	}
}

1794
static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
M
Matthew Wilcox 已提交
1795
{
1796
	int i, queues;
K
Keith Busch 已提交
1797 1798
	bool dead = true;
	struct pci_dev *pdev = to_pci_dev(dev->dev);
1799

1800
	del_timer_sync(&dev->watchdog_timer);
1801

1802
	mutex_lock(&dev->shutdown_lock);
K
Keith Busch 已提交
1803 1804 1805 1806 1807 1808 1809
	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);
1810
	}
1811

K
Keith Busch 已提交
1812 1813 1814 1815 1816 1817 1818 1819
	/*
	 * Give the controller a chance to complete all entered requests if
	 * doing a safe shutdown.
	 */
	if (!dead && shutdown)
		nvme_wait_freeze_timeout(&dev->ctrl, NVME_IO_TIMEOUT);
	nvme_stop_queues(&dev->ctrl);

1820
	queues = dev->online_queues - 1;
1821 1822 1823
	for (i = dev->queue_count - 1; i > 0; i--)
		nvme_suspend_queue(dev->queues[i]);

K
Keith Busch 已提交
1824
	if (dead) {
1825 1826 1827 1828 1829 1830
		/* 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
1831
	} else {
1832
		nvme_disable_io_queues(dev, queues);
1833
		nvme_disable_admin_queue(dev, shutdown);
K
Keith Busch 已提交
1834
	}
1835
	nvme_pci_disable(dev);
1836

1837 1838
	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
1839 1840 1841 1842 1843 1844 1845 1846

	/*
	 * 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);
1847
	mutex_unlock(&dev->shutdown_lock);
M
Matthew Wilcox 已提交
1848 1849
}

M
Matthew Wilcox 已提交
1850 1851
static int nvme_setup_prp_pools(struct nvme_dev *dev)
{
1852
	dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
M
Matthew Wilcox 已提交
1853 1854 1855 1856
						PAGE_SIZE, PAGE_SIZE, 0);
	if (!dev->prp_page_pool)
		return -ENOMEM;

1857
	/* Optimisation for I/Os between 4k and 128k */
1858
	dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
1859 1860 1861 1862 1863
						256, 256, 0);
	if (!dev->prp_small_pool) {
		dma_pool_destroy(dev->prp_page_pool);
		return -ENOMEM;
	}
M
Matthew Wilcox 已提交
1864 1865 1866 1867 1868 1869
	return 0;
}

static void nvme_release_prp_pools(struct nvme_dev *dev)
{
	dma_pool_destroy(dev->prp_page_pool);
1870
	dma_pool_destroy(dev->prp_small_pool);
M
Matthew Wilcox 已提交
1871 1872
}

1873
static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
1874
{
1875
	struct nvme_dev *dev = to_nvme_dev(ctrl);
1876

1877
	nvme_dbbuf_dma_free(dev);
1878
	put_device(dev->dev);
1879 1880
	if (dev->tagset.tags)
		blk_mq_free_tag_set(&dev->tagset);
1881 1882
	if (dev->ctrl.admin_q)
		blk_put_queue(dev->ctrl.admin_q);
1883
	kfree(dev->queues);
1884
	free_opal_dev(dev->ctrl.opal_dev);
1885 1886 1887
	kfree(dev);
}

1888 1889
static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status)
{
1890
	dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status);
1891 1892

	kref_get(&dev->ctrl.kref);
1893
	nvme_dev_disable(dev, false);
1894 1895 1896 1897
	if (!schedule_work(&dev->remove_work))
		nvme_put_ctrl(&dev->ctrl);
}

1898
static void nvme_reset_work(struct work_struct *work)
1899
{
1900
	struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work);
1901
	bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL);
1902
	int result = -ENODEV;
1903

1904
	if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING))
1905
		goto out;
1906

1907 1908 1909 1910
	/*
	 * If we're called to reset a live controller first shut it down before
	 * moving on.
	 */
1911
	if (dev->ctrl.ctrl_config & NVME_CC_ENABLE)
1912
		nvme_dev_disable(dev, false);
1913

1914
	result = nvme_pci_enable(dev);
1915
	if (result)
1916
		goto out;
1917 1918 1919

	result = nvme_configure_admin_queue(dev);
	if (result)
1920
		goto out;
1921

M
Matias Bjørling 已提交
1922
	nvme_init_queue(dev->queues[0], 0);
K
Keith Busch 已提交
1923 1924
	result = nvme_alloc_admin_tags(dev);
	if (result)
1925
		goto out;
1926

1927 1928
	result = nvme_init_identify(&dev->ctrl);
	if (result)
1929
		goto out;
1930

1931 1932 1933 1934 1935 1936 1937 1938 1939
	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;
1940
	}
1941

1942 1943 1944 1945 1946 1947 1948
	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");
	}

1949
	result = nvme_setup_io_queues(dev);
1950
	if (result)
1951
		goto out;
1952

1953 1954 1955 1956 1957 1958
	/*
	 * 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.
	 */
1959 1960
	if (dev->online_queues > 1)
		nvme_queue_async_events(&dev->ctrl);
1961

1962
	mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
1963

1964 1965 1966 1967
	/*
	 * Keep the controller around but remove all namespaces if we don't have
	 * any working I/O queue.
	 */
1968
	if (dev->online_queues < 2) {
1969
		dev_warn(dev->ctrl.device, "IO queues not created\n");
1970
		nvme_kill_queues(&dev->ctrl);
1971
		nvme_remove_namespaces(&dev->ctrl);
1972
	} else {
1973
		nvme_start_queues(&dev->ctrl);
K
Keith Busch 已提交
1974
		nvme_wait_freeze(&dev->ctrl);
1975
		nvme_dev_add(dev);
K
Keith Busch 已提交
1976
		nvme_unfreeze(&dev->ctrl);
1977 1978
	}

1979 1980 1981 1982
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) {
		dev_warn(dev->ctrl.device, "failed to mark controller live\n");
		goto out;
	}
1983 1984

	if (dev->online_queues > 1)
1985
		nvme_queue_scan(&dev->ctrl);
1986
	return;
1987

1988
 out:
1989
	nvme_remove_dead_ctrl(dev, result);
1990 1991
}

1992
static void nvme_remove_dead_ctrl_work(struct work_struct *work)
K
Keith Busch 已提交
1993
{
1994
	struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
1995
	struct pci_dev *pdev = to_pci_dev(dev->dev);
K
Keith Busch 已提交
1996

1997
	nvme_kill_queues(&dev->ctrl);
K
Keith Busch 已提交
1998
	if (pci_get_drvdata(pdev))
K
Keith Busch 已提交
1999
		device_release_driver(&pdev->dev);
2000
	nvme_put_ctrl(&dev->ctrl);
K
Keith Busch 已提交
2001 2002
}

2003
static int nvme_reset(struct nvme_dev *dev)
K
Keith Busch 已提交
2004
{
2005
	if (!dev->ctrl.admin_q || blk_queue_dying(dev->ctrl.admin_q))
2006
		return -ENODEV;
2007 2008
	if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING))
		return -EBUSY;
C
Christoph Hellwig 已提交
2009 2010 2011
	if (!queue_work(nvme_workq, &dev->reset_work))
		return -EBUSY;
	return 0;
K
Keith Busch 已提交
2012 2013
}

2014
static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
T
Tejun Heo 已提交
2015
{
2016
	*val = readl(to_nvme_dev(ctrl)->bar + off);
2017
	return 0;
T
Tejun Heo 已提交
2018 2019
}

2020
static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
2021
{
2022 2023 2024
	writel(val, to_nvme_dev(ctrl)->bar + off);
	return 0;
}
2025

2026 2027 2028 2029
static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
{
	*val = readq(to_nvme_dev(ctrl)->bar + off);
	return 0;
2030 2031
}

2032 2033
static int nvme_pci_reset_ctrl(struct nvme_ctrl *ctrl)
{
K
Keith Busch 已提交
2034 2035 2036 2037 2038 2039
	struct nvme_dev *dev = to_nvme_dev(ctrl);
	int ret = nvme_reset(dev);

	if (!ret)
		flush_work(&dev->reset_work);
	return ret;
2040
}
2041

2042
static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
M
Ming Lin 已提交
2043
	.name			= "pcie",
2044
	.module			= THIS_MODULE,
2045
	.flags			= NVME_F_METADATA_SUPPORTED,
2046
	.reg_read32		= nvme_pci_reg_read32,
2047
	.reg_write32		= nvme_pci_reg_write32,
2048
	.reg_read64		= nvme_pci_reg_read64,
2049
	.reset_ctrl		= nvme_pci_reset_ctrl,
2050
	.free_ctrl		= nvme_pci_free_ctrl,
2051
	.submit_async_event	= nvme_pci_submit_async_event,
2052
};
2053

2054 2055 2056 2057
static int nvme_dev_map(struct nvme_dev *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev->dev);

2058
	if (pci_request_mem_regions(pdev, "nvme"))
2059 2060 2061 2062 2063 2064
		return -ENODEV;

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

M
Max Gurtovoy 已提交
2065
	return 0;
2066
  release:
M
Max Gurtovoy 已提交
2067 2068
	pci_release_mem_regions(pdev);
	return -ENODEV;
2069 2070
}

2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090
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;
}

2091
static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
M
Matthew Wilcox 已提交
2092
{
M
Matias Bjørling 已提交
2093
	int node, result = -ENOMEM;
M
Matthew Wilcox 已提交
2094
	struct nvme_dev *dev;
2095
	unsigned long quirks = id->driver_data;
M
Matthew Wilcox 已提交
2096

M
Matias Bjørling 已提交
2097 2098
	node = dev_to_node(&pdev->dev);
	if (node == NUMA_NO_NODE)
2099
		set_dev_node(&pdev->dev, first_memory_node);
M
Matias Bjørling 已提交
2100 2101

	dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2102 2103
	if (!dev)
		return -ENOMEM;
M
Matias Bjørling 已提交
2104 2105
	dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
							GFP_KERNEL, node);
M
Matthew Wilcox 已提交
2106 2107 2108
	if (!dev->queues)
		goto free;

2109
	dev->dev = get_device(&pdev->dev);
K
Keith Busch 已提交
2110
	pci_set_drvdata(pdev, dev);
2111

2112 2113 2114 2115
	result = nvme_dev_map(dev);
	if (result)
		goto free;

2116
	INIT_WORK(&dev->reset_work, nvme_reset_work);
2117
	INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
2118 2119
	setup_timer(&dev->watchdog_timer, nvme_watchdog_timer,
		(unsigned long)dev);
2120
	mutex_init(&dev->shutdown_lock);
K
Keith Busch 已提交
2121
	init_completion(&dev->ioq_wait);
M
Matthew Wilcox 已提交
2122

M
Matthew Wilcox 已提交
2123 2124
	result = nvme_setup_prp_pools(dev);
	if (result)
K
Keith Busch 已提交
2125
		goto put_pci;
2126

2127 2128
	quirks |= check_dell_samsung_bug(pdev);

2129
	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
2130
			quirks);
2131
	if (result)
2132
		goto release_pools;
2133

2134
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING);
2135 2136
	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));

2137
	queue_work(nvme_workq, &dev->reset_work);
M
Matthew Wilcox 已提交
2138 2139
	return 0;

2140
 release_pools:
M
Matthew Wilcox 已提交
2141
	nvme_release_prp_pools(dev);
K
Keith Busch 已提交
2142
 put_pci:
2143
	put_device(dev->dev);
2144
	nvme_dev_unmap(dev);
M
Matthew Wilcox 已提交
2145 2146 2147 2148 2149 2150
 free:
	kfree(dev->queues);
	kfree(dev);
	return result;
}

2151 2152
static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
{
K
Keith Busch 已提交
2153
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2154

K
Keith Busch 已提交
2155
	if (prepare)
2156
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2157
	else
K
Keith Busch 已提交
2158
		nvme_reset(dev);
2159 2160
}

2161 2162 2163
static void nvme_shutdown(struct pci_dev *pdev)
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
2164
	nvme_dev_disable(dev, true);
2165 2166
}

2167 2168 2169 2170 2171
/*
 * 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.
 */
2172
static void nvme_remove(struct pci_dev *pdev)
M
Matthew Wilcox 已提交
2173 2174
{
	struct nvme_dev *dev = pci_get_drvdata(pdev);
K
Keith Busch 已提交
2175

2176 2177
	nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING);

2178
	cancel_work_sync(&dev->reset_work);
K
Keith Busch 已提交
2179
	pci_set_drvdata(pdev, NULL);
2180

2181
	if (!pci_device_is_present(pdev)) {
2182
		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
2183 2184
		nvme_dev_disable(dev, false);
	}
2185

K
Keith Busch 已提交
2186
	flush_work(&dev->reset_work);
2187
	nvme_uninit_ctrl(&dev->ctrl);
2188
	nvme_dev_disable(dev, true);
M
Matias Bjørling 已提交
2189
	nvme_dev_remove_admin(dev);
2190
	nvme_free_queues(dev, 0);
K
Keith Busch 已提交
2191
	nvme_release_prp_pools(dev);
2192
	nvme_dev_unmap(dev);
2193
	nvme_put_ctrl(&dev->ctrl);
M
Matthew Wilcox 已提交
2194 2195
}

K
Keith Busch 已提交
2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
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;
}

2214
#ifdef CONFIG_PM_SLEEP
2215 2216 2217 2218 2219
static int nvme_suspend(struct device *dev)
{
	struct pci_dev *pdev = to_pci_dev(dev);
	struct nvme_dev *ndev = pci_get_drvdata(pdev);

2220
	nvme_dev_disable(ndev, true);
2221 2222 2223 2224 2225 2226 2227 2228
	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);

K
Keith Busch 已提交
2229
	nvme_reset(ndev);
K
Keith Busch 已提交
2230
	return 0;
2231
}
2232
#endif
2233 2234

static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
M
Matthew Wilcox 已提交
2235

K
Keith Busch 已提交
2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249
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
2250 2251
		dev_warn(dev->ctrl.device,
			"frozen state error detected, reset controller\n");
2252
		nvme_dev_disable(dev, false);
K
Keith Busch 已提交
2253 2254
		return PCI_ERS_RESULT_NEED_RESET;
	case pci_channel_io_perm_failure:
K
Keith Busch 已提交
2255 2256
		dev_warn(dev->ctrl.device,
			"failure state error detected, request disconnect\n");
K
Keith Busch 已提交
2257 2258 2259 2260 2261 2262 2263 2264 2265
		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);

2266
	dev_info(dev->ctrl.device, "restart after slot reset\n");
K
Keith Busch 已提交
2267
	pci_restore_state(pdev);
K
Keith Busch 已提交
2268
	nvme_reset(dev);
K
Keith Busch 已提交
2269 2270 2271 2272 2273 2274 2275 2276
	return PCI_ERS_RESULT_RECOVERED;
}

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

2277
static const struct pci_error_handlers nvme_err_handler = {
M
Matthew Wilcox 已提交
2278 2279 2280
	.error_detected	= nvme_error_detected,
	.slot_reset	= nvme_slot_reset,
	.resume		= nvme_error_resume,
2281
	.reset_notify	= nvme_reset_notify,
M
Matthew Wilcox 已提交
2282 2283
};

2284
static const struct pci_device_id nvme_id_table[] = {
2285
	{ PCI_VDEVICE(INTEL, 0x0953),
2286
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2287
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2288 2289
	{ PCI_VDEVICE(INTEL, 0x0a53),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2290
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2291 2292
	{ PCI_VDEVICE(INTEL, 0x0a54),
		.driver_data = NVME_QUIRK_STRIPE_SIZE |
2293
				NVME_QUIRK_DEALLOCATE_ZEROES, },
2294 2295
	{ PCI_VDEVICE(INTEL, 0xf1a5),	/* Intel 600P/P3100 */
		.driver_data = NVME_QUIRK_NO_DEEPEST_PS },
2296 2297
	{ PCI_VDEVICE(INTEL, 0x5845),	/* Qemu emulated controller */
		.driver_data = NVME_QUIRK_IDENTIFY_CNS, },
2298 2299
	{ PCI_DEVICE(0x1c58, 0x0003),	/* HGST adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
2300 2301
	{ PCI_DEVICE(0x1c5f, 0x0540),	/* Memblaze Pblaze4 adapter */
		.driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
M
Matthew Wilcox 已提交
2302
	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
2303
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
2304
	{ PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) },
M
Matthew Wilcox 已提交
2305 2306 2307 2308 2309 2310 2311 2312
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, nvme_id_table);

static struct pci_driver nvme_driver = {
	.name		= "nvme",
	.id_table	= nvme_id_table,
	.probe		= nvme_probe,
2313
	.remove		= nvme_remove,
2314
	.shutdown	= nvme_shutdown,
2315 2316 2317
	.driver		= {
		.pm	= &nvme_dev_pm_ops,
	},
K
Keith Busch 已提交
2318
	.sriov_configure = nvme_pci_sriov_configure,
M
Matthew Wilcox 已提交
2319 2320 2321 2322 2323
	.err_handler	= &nvme_err_handler,
};

static int __init nvme_init(void)
{
2324
	int result;
2325

2326
	nvme_workq = alloc_workqueue("nvme", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
K
Keith Busch 已提交
2327
	if (!nvme_workq)
2328
		return -ENOMEM;
K
Keith Busch 已提交
2329

2330 2331
	result = pci_register_driver(&nvme_driver);
	if (result)
2332
		destroy_workqueue(nvme_workq);
M
Matthew Wilcox 已提交
2333 2334 2335 2336 2337 2338
	return result;
}

static void __exit nvme_exit(void)
{
	pci_unregister_driver(&nvme_driver);
K
Keith Busch 已提交
2339
	destroy_workqueue(nvme_workq);
2340
	_nvme_check_size();
M
Matthew Wilcox 已提交
2341 2342 2343 2344
}

MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
MODULE_LICENSE("GPL");
2345
MODULE_VERSION("1.0");
M
Matthew Wilcox 已提交
2346 2347
module_init(nvme_init);
module_exit(nvme_exit);