blob: 642ee00e914302d0bec9b16b9c1765dc3b9af115 [file] [log] [blame]
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001/*
2 * NVM Express device driver
Matthew Wilcox6eb0d692014-03-24 10:11:22 -04003 * Copyright (c) 2011-2014, Intel Corporation.
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05004 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050013 */
14
Keith Buscha0a34082015-12-07 15:30:31 -070015#include <linux/aer.h>
Matthew Wilcox8de05532011-05-12 13:50:28 -040016#include <linux/bitops.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050017#include <linux/blkdev.h>
Matias Bjørlinga4aea562014-11-04 08:20:14 -070018#include <linux/blk-mq.h>
Christoph Hellwigdca51e72016-09-14 16:18:57 +020019#include <linux/blk-mq-pci.h>
Keith Busch42f61422014-03-24 10:46:25 -060020#include <linux/cpu.h>
Matthew Wilcoxfd63e9ce2011-05-06 08:37:54 -040021#include <linux/delay.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050022#include <linux/errno.h>
23#include <linux/fs.h>
24#include <linux/genhd.h>
Keith Busch4cc09e22014-04-02 15:45:37 -060025#include <linux/hdreg.h>
Matthew Wilcox5aff9382011-05-06 08:45:47 -040026#include <linux/idr.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050027#include <linux/init.h>
28#include <linux/interrupt.h>
29#include <linux/io.h>
30#include <linux/kdev_t.h>
31#include <linux/kernel.h>
32#include <linux/mm.h>
33#include <linux/module.h>
34#include <linux/moduleparam.h>
Keith Busch77bf25e2015-11-26 12:21:29 +010035#include <linux/mutex.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050036#include <linux/pci.h>
Matthew Wilcoxbe7b6272011-02-06 07:53:23 -050037#include <linux/poison.h>
Matthew Wilcoxc3bfe712013-07-08 17:26:25 -040038#include <linux/ptrace.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050039#include <linux/sched.h>
40#include <linux/slab.h>
Keith Busche1e5e562015-02-19 13:39:03 -070041#include <linux/t10-pi.h>
Christoph Hellwig2d55cd52016-02-29 15:59:46 +010042#include <linux/timer.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050043#include <linux/types.h>
Linus Torvalds9cf5c092015-11-06 14:22:15 -080044#include <linux/io-64-nonatomic-lo-hi.h>
Keith Busch1d277a62015-10-15 14:10:52 +020045#include <asm/unaligned.h>
Hitoshi Mitake797a7962012-02-07 11:45:33 +090046
Christoph Hellwigf11bb3e2015-10-03 15:46:41 +020047#include "nvme.h"
48
Keith Busch9d43cf62014-05-13 11:42:02 -060049#define NVME_Q_DEPTH 1024
Jens Axboed31af0a2015-03-06 12:56:13 -070050#define NVME_AQ_DEPTH 256
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050051#define SQ_SIZE(depth) (depth * sizeof(struct nvme_command))
52#define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion))
Christoph Hellwigadf68f22015-11-28 15:42:28 +010053
54/*
55 * We handle AEN commands ourselves and don't even let the
56 * block layer know about them.
57 */
Christoph Hellwigf866fc42016-04-26 13:52:00 +020058#define NVME_AQ_BLKMQ_DEPTH (NVME_AQ_DEPTH - NVME_NR_AERS)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050059
Matthew Wilcox58ffacb2011-02-06 07:28:06 -050060static int use_threaded_interrupts;
61module_param(use_threaded_interrupts, int, 0);
62
Jon Derrick8ffaadf2015-07-20 10:14:09 -060063static bool use_cmb_sqes = true;
64module_param(use_cmb_sqes, bool, 0644);
65MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");
66
Keith Busch9a6b9452013-12-10 13:10:36 -070067static struct workqueue_struct *nvme_workq;
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -050068
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010069struct nvme_dev;
70struct nvme_queue;
Keith Buschb3fffde2015-02-03 11:21:42 -070071
Keith Busch4cc06522015-06-05 10:30:08 -060072static int nvme_reset(struct nvme_dev *dev);
Jens Axboea0fa9642015-11-03 20:37:26 -070073static void nvme_process_cq(struct nvme_queue *nvmeq);
Keith Buscha5cdb682016-01-12 14:41:18 -070074static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
Keith Buschd4b4ff82013-12-10 13:10:37 -070075
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050076/*
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010077 * Represents an NVM Express device. Each nvme_dev is a PCI function.
78 */
79struct nvme_dev {
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010080 struct nvme_queue **queues;
81 struct blk_mq_tag_set tagset;
82 struct blk_mq_tag_set admin_tagset;
83 u32 __iomem *dbs;
84 struct device *dev;
85 struct dma_pool *prp_page_pool;
86 struct dma_pool *prp_small_pool;
87 unsigned queue_count;
88 unsigned online_queues;
89 unsigned max_qid;
90 int q_depth;
91 u32 db_stride;
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010092 void __iomem *bar;
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010093 struct work_struct reset_work;
Christoph Hellwig5c8809e2015-11-26 12:35:49 +010094 struct work_struct remove_work;
Christoph Hellwig2d55cd52016-02-29 15:59:46 +010095 struct timer_list watchdog_timer;
Keith Busch77bf25e2015-11-26 12:21:29 +010096 struct mutex shutdown_lock;
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010097 bool subsystem;
Christoph Hellwig1c63dc62015-11-26 10:06:56 +010098 void __iomem *cmb;
Christoph Hellwigc83bbed2017-10-01 09:37:35 +020099 pci_bus_addr_t cmb_bus_addr;
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100100 u64 cmb_size;
101 u32 cmbsz;
Stephen Bates202021c2016-10-05 20:01:12 -0600102 u32 cmbloc;
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100103 struct nvme_ctrl ctrl;
Keith Buschdb3cbff2016-01-12 14:41:17 -0700104 struct completion ioq_wait;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500105};
106
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100107static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl)
108{
109 return container_of(ctrl, struct nvme_dev, ctrl);
110}
111
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500112/*
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500113 * An NVM Express queue. Each device has at least two (one for admin
114 * commands and one for I/O commands).
115 */
116struct nvme_queue {
117 struct device *q_dmadev;
Matthew Wilcox091b6092011-02-10 09:56:01 -0500118 struct nvme_dev *dev;
Matthew Wilcox3193f072014-01-27 15:57:22 -0500119 char irqname[24]; /* nvme4294967295-65535\0 */
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500120 spinlock_t q_lock;
121 struct nvme_command *sq_cmds;
Jon Derrick8ffaadf2015-07-20 10:14:09 -0600122 struct nvme_command __iomem *sq_cmds_io;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500123 volatile struct nvme_completion *cqes;
Keith Busch42483222015-06-01 09:29:54 -0600124 struct blk_mq_tags **tags;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500125 dma_addr_t sq_dma_addr;
126 dma_addr_t cq_dma_addr;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500127 u32 __iomem *q_db;
128 u16 q_depth;
Jens Axboe6222d172015-01-15 15:19:10 -0700129 s16 cq_vector;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500130 u16 sq_tail;
131 u16 cq_head;
Keith Buschc30341d2013-12-10 13:10:38 -0700132 u16 qid;
Matthew Wilcoxe9539f42013-06-24 11:47:34 -0400133 u8 cq_phase;
134 u8 cqe_seen;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500135};
136
137/*
Christoph Hellwig71bd1502015-10-16 07:58:32 +0200138 * The nvme_iod describes the data in an I/O, including the list of PRP
139 * entries. You can't see it in this data structure because C doesn't let
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100140 * me express that. Use nvme_init_iod to ensure there's enough space
Christoph Hellwig71bd1502015-10-16 07:58:32 +0200141 * allocated to store the PRP list.
142 */
143struct nvme_iod {
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100144 struct nvme_queue *nvmeq;
145 int aborted;
Christoph Hellwig71bd1502015-10-16 07:58:32 +0200146 int npages; /* In the PRP list. 0 means small pool in use */
Christoph Hellwig71bd1502015-10-16 07:58:32 +0200147 int nents; /* Used in scatterlist */
148 int length; /* Of data, in bytes */
149 dma_addr_t first_dma;
Christoph Hellwigbf684052015-10-26 17:12:51 +0900150 struct scatterlist meta_sg; /* metadata requires single contiguous buffer */
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100151 struct scatterlist *sg;
152 struct scatterlist inline_sg[0];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500153};
154
155/*
156 * Check we didin't inadvertently grow the command struct
157 */
158static inline void _nvme_check_size(void)
159{
160 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
161 BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64);
162 BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64);
163 BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
164 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
Vishal Vermaf8ebf842013-03-27 07:13:41 -0400165 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
Keith Buschc30341d2013-12-10 13:10:38 -0700166 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500167 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
168 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
169 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
170 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
Keith Busch6ecec742012-09-26 12:49:27 -0600171 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500172}
173
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700174/*
175 * Max size of iod being embedded in the request payload
176 */
177#define NVME_INT_PAGES 2
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +0100178#define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->ctrl.page_size)
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700179
180/*
181 * Will slightly overestimate the number of pages needed. This is OK
182 * as it only leads to a small amount of wasted memory for the lifetime of
183 * the I/O.
184 */
185static int nvme_npages(unsigned size, struct nvme_dev *dev)
186{
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +0100187 unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size,
188 dev->ctrl.page_size);
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700189 return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
190}
191
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100192static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
193 unsigned int size, unsigned int nseg)
194{
195 return sizeof(__le64 *) * nvme_npages(size, dev) +
196 sizeof(struct scatterlist) * nseg;
197}
198
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700199static unsigned int nvme_cmd_size(struct nvme_dev *dev)
200{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100201 return sizeof(struct nvme_iod) +
202 nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES);
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700203}
204
Christoph Hellwigdca51e72016-09-14 16:18:57 +0200205static int nvmeq_irq(struct nvme_queue *nvmeq)
206{
207 return pci_irq_vector(to_pci_dev(nvmeq->dev->dev), nvmeq->cq_vector);
208}
209
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700210static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
211 unsigned int hctx_idx)
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500212{
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700213 struct nvme_dev *dev = data;
214 struct nvme_queue *nvmeq = dev->queues[0];
215
Keith Busch42483222015-06-01 09:29:54 -0600216 WARN_ON(hctx_idx != 0);
217 WARN_ON(dev->admin_tagset.tags[0] != hctx->tags);
218 WARN_ON(nvmeq->tags);
219
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700220 hctx->driver_data = nvmeq;
Keith Busch42483222015-06-01 09:29:54 -0600221 nvmeq->tags = &dev->admin_tagset.tags[0];
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700222 return 0;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500223}
224
Keith Busch4af0e212015-06-08 10:08:13 -0600225static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
226{
227 struct nvme_queue *nvmeq = hctx->driver_data;
228
229 nvmeq->tags = NULL;
230}
231
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700232static int nvme_admin_init_request(void *data, struct request *req,
233 unsigned int hctx_idx, unsigned int rq_idx,
234 unsigned int numa_node)
Keith Busch22404272013-07-15 15:02:20 -0600235{
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700236 struct nvme_dev *dev = data;
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100237 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700238 struct nvme_queue *nvmeq = dev->queues[0];
239
240 BUG_ON(!nvmeq);
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100241 iod->nvmeq = nvmeq;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700242 return 0;
Keith Busch22404272013-07-15 15:02:20 -0600243}
244
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700245static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
246 unsigned int hctx_idx)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500247{
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700248 struct nvme_dev *dev = data;
Keith Busch42483222015-06-01 09:29:54 -0600249 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500250
Keith Busch42483222015-06-01 09:29:54 -0600251 if (!nvmeq->tags)
252 nvmeq->tags = &dev->tagset.tags[hctx_idx];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500253
Keith Busch42483222015-06-01 09:29:54 -0600254 WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags);
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700255 hctx->driver_data = nvmeq;
256 return 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500257}
258
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700259static int nvme_init_request(void *data, struct request *req,
260 unsigned int hctx_idx, unsigned int rq_idx,
261 unsigned int numa_node)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500262{
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700263 struct nvme_dev *dev = data;
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100264 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700265 struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1];
266
267 BUG_ON(!nvmeq);
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100268 iod->nvmeq = nvmeq;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700269 return 0;
270}
271
Christoph Hellwigdca51e72016-09-14 16:18:57 +0200272static int nvme_pci_map_queues(struct blk_mq_tag_set *set)
273{
274 struct nvme_dev *dev = set->driver_data;
275
276 return blk_mq_pci_map_queues(set, to_pci_dev(dev->dev));
277}
278
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500279/**
Christoph Hellwigadf68f22015-11-28 15:42:28 +0100280 * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500281 * @nvmeq: The queue to use
282 * @cmd: The command to send
283 *
284 * Safe to use from interrupt context
285 */
Sunad Bhandarye3f879b2015-07-31 18:56:58 +0530286static void __nvme_submit_cmd(struct nvme_queue *nvmeq,
287 struct nvme_command *cmd)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500288{
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700289 u16 tail = nvmeq->sq_tail;
290
Jon Derrick8ffaadf2015-07-20 10:14:09 -0600291 if (nvmeq->sq_cmds_io)
292 memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd));
293 else
294 memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));
295
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500296 if (++tail == nvmeq->q_depth)
297 tail = 0;
Matthew Wilcox75478812011-02-16 09:59:59 -0500298 writel(tail, nvmeq->q_db);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500299 nvmeq->sq_tail = tail;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500300}
301
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100302static __le64 **iod_list(struct request *req)
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700303{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100304 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
305 return (__le64 **)(iod->sg + req->nr_phys_segments);
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700306}
307
Ming Lin58b45602016-03-22 00:24:43 -0700308static int nvme_init_iod(struct request *rq, unsigned size,
309 struct nvme_dev *dev)
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500310{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100311 struct nvme_iod *iod = blk_mq_rq_to_pdu(rq);
312 int nseg = rq->nr_phys_segments;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500313
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100314 if (nseg > NVME_INT_PAGES || size > NVME_INT_BYTES(dev)) {
315 iod->sg = kmalloc(nvme_iod_alloc_size(dev, size, nseg), GFP_ATOMIC);
316 if (!iod->sg)
317 return BLK_MQ_RQ_QUEUE_BUSY;
318 } else {
319 iod->sg = iod->inline_sg;
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700320 }
321
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100322 iod->aborted = 0;
323 iod->npages = -1;
324 iod->nents = 0;
325 iod->length = size;
Keith Buschf80ec962016-07-12 16:20:31 -0700326
327 if (!(rq->cmd_flags & REQ_DONTPREP)) {
328 rq->retries = 0;
329 rq->cmd_flags |= REQ_DONTPREP;
330 }
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100331 return 0;
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700332}
333
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100334static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500335{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100336 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +0100337 const int last_prp = dev->ctrl.page_size / 8 - 1;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500338 int i;
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100339 __le64 **list = iod_list(req);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500340 dma_addr_t prp_dma = iod->first_dma;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500341
Ming Lin69042422016-04-25 14:33:20 -0700342 nvme_cleanup_cmd(req);
Ming Lin03b59292016-03-22 00:24:45 -0700343
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500344 if (iod->npages == 0)
345 dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
346 for (i = 0; i < iod->npages; i++) {
347 __le64 *prp_list = list[i];
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500348 dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
Matthew Wilcox091b6092011-02-10 09:56:01 -0500349 dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500350 prp_dma = next_prp_dma;
351 }
Jens Axboeac3dd5b2015-01-22 12:07:58 -0700352
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100353 if (iod->sg != iod->inline_sg)
354 kfree(iod->sg);
Keith Buschb4ff9c82014-08-29 09:06:12 -0600355}
356
Keith Busch52b68d72015-02-23 09:16:21 -0700357#ifdef CONFIG_BLK_DEV_INTEGRITY
Keith Busche1e5e562015-02-19 13:39:03 -0700358static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
359{
360 if (be32_to_cpu(pi->ref_tag) == v)
361 pi->ref_tag = cpu_to_be32(p);
362}
363
364static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
365{
366 if (be32_to_cpu(pi->ref_tag) == p)
367 pi->ref_tag = cpu_to_be32(v);
368}
369
370/**
371 * nvme_dif_remap - remaps ref tags to bip seed and physical lba
372 *
373 * The virtual start sector is the one that was originally submitted by the
374 * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical
375 * start sector may be different. Remap protection information to match the
376 * physical LBA on writes, and back to the original seed on reads.
377 *
378 * Type 0 and 3 do not have a ref tag, so no remapping required.
379 */
380static void nvme_dif_remap(struct request *req,
381 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
382{
383 struct nvme_ns *ns = req->rq_disk->private_data;
384 struct bio_integrity_payload *bip;
385 struct t10_pi_tuple *pi;
386 void *p, *pmap;
387 u32 i, nlb, ts, phys, virt;
388
389 if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3)
390 return;
391
392 bip = bio_integrity(req->bio);
393 if (!bip)
394 return;
395
396 pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset;
Keith Busche1e5e562015-02-19 13:39:03 -0700397
398 p = pmap;
399 virt = bip_get_seed(bip);
400 phys = nvme_block_nr(ns, blk_rq_pos(req));
401 nlb = (blk_rq_bytes(req) >> ns->lba_shift);
Dan Williamsac6fc482015-10-21 13:20:18 -0400402 ts = ns->disk->queue->integrity.tuple_size;
Keith Busche1e5e562015-02-19 13:39:03 -0700403
404 for (i = 0; i < nlb; i++, virt++, phys++) {
405 pi = (struct t10_pi_tuple *)p;
406 dif_swap(phys, virt, pi);
407 p += ts;
408 }
409 kunmap_atomic(pmap);
410}
Keith Busch52b68d72015-02-23 09:16:21 -0700411#else /* CONFIG_BLK_DEV_INTEGRITY */
412static void nvme_dif_remap(struct request *req,
413 void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi))
414{
415}
416static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi)
417{
418}
419static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi)
420{
421}
Keith Busch52b68d72015-02-23 09:16:21 -0700422#endif
423
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100424static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req,
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200425 int total_len)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500426{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100427 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Matthew Wilcox99802a72011-02-10 10:30:34 -0500428 struct dma_pool *pool;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500429 int length = total_len;
430 struct scatterlist *sg = iod->sg;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500431 int dma_len = sg_dma_len(sg);
432 u64 dma_addr = sg_dma_address(sg);
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +0100433 u32 page_size = dev->ctrl.page_size;
Murali Iyerf137e0f2015-03-26 11:07:51 -0500434 int offset = dma_addr & (page_size - 1);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500435 __le64 *prp_list;
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100436 __le64 **list = iod_list(req);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500437 dma_addr_t prp_dma;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500438 int nprps, i;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500439
Keith Busch1d090622014-06-23 11:34:01 -0600440 length -= (page_size - offset);
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500441 if (length <= 0)
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200442 return true;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500443
Keith Busch1d090622014-06-23 11:34:01 -0600444 dma_len -= (page_size - offset);
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500445 if (dma_len) {
Keith Busch1d090622014-06-23 11:34:01 -0600446 dma_addr += (page_size - offset);
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500447 } else {
448 sg = sg_next(sg);
449 dma_addr = sg_dma_address(sg);
450 dma_len = sg_dma_len(sg);
451 }
452
Keith Busch1d090622014-06-23 11:34:01 -0600453 if (length <= page_size) {
Keith Buschedd10d32014-04-03 16:45:23 -0600454 iod->first_dma = dma_addr;
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200455 return true;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500456 }
457
Keith Busch1d090622014-06-23 11:34:01 -0600458 nprps = DIV_ROUND_UP(length, page_size);
Matthew Wilcox99802a72011-02-10 10:30:34 -0500459 if (nprps <= (256 / 8)) {
460 pool = dev->prp_small_pool;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500461 iod->npages = 0;
Matthew Wilcox99802a72011-02-10 10:30:34 -0500462 } else {
463 pool = dev->prp_page_pool;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500464 iod->npages = 1;
Matthew Wilcox99802a72011-02-10 10:30:34 -0500465 }
466
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200467 prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
Matthew Wilcoxb77954c2011-05-12 13:51:41 -0400468 if (!prp_list) {
Keith Buschedd10d32014-04-03 16:45:23 -0600469 iod->first_dma = dma_addr;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500470 iod->npages = -1;
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200471 return false;
Matthew Wilcoxb77954c2011-05-12 13:51:41 -0400472 }
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500473 list[0] = prp_list;
474 iod->first_dma = prp_dma;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500475 i = 0;
476 for (;;) {
Keith Busch1d090622014-06-23 11:34:01 -0600477 if (i == page_size >> 3) {
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500478 __le64 *old_prp_list = prp_list;
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200479 prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500480 if (!prp_list)
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200481 return false;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500482 list[iod->npages++] = prp_list;
Matthew Wilcox7523d832011-03-16 16:43:40 -0400483 prp_list[0] = old_prp_list[i - 1];
484 old_prp_list[i - 1] = cpu_to_le64(prp_dma);
485 i = 1;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500486 }
487 prp_list[i++] = cpu_to_le64(dma_addr);
Keith Busch1d090622014-06-23 11:34:01 -0600488 dma_len -= page_size;
489 dma_addr += page_size;
490 length -= page_size;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500491 if (length <= 0)
492 break;
493 if (dma_len > 0)
494 continue;
495 BUG_ON(dma_len < 0);
496 sg = sg_next(sg);
497 dma_addr = sg_dma_address(sg);
498 dma_len = sg_dma_len(sg);
499 }
500
Christoph Hellwig69d2b572015-10-16 07:58:37 +0200501 return true;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500502}
503
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100504static int nvme_map_data(struct nvme_dev *dev, struct request *req,
Ming Lin03b59292016-03-22 00:24:45 -0700505 unsigned size, struct nvme_command *cmnd)
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200506{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100507 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200508 struct request_queue *q = req->q;
509 enum dma_data_direction dma_dir = rq_data_dir(req) ?
510 DMA_TO_DEVICE : DMA_FROM_DEVICE;
511 int ret = BLK_MQ_RQ_QUEUE_ERROR;
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200512
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200513 sg_init_table(iod->sg, req->nr_phys_segments);
514 iod->nents = blk_rq_map_sg(q, req, iod->sg);
515 if (!iod->nents)
516 goto out;
517
518 ret = BLK_MQ_RQ_QUEUE_BUSY;
Mauricio Faria de Oliveira2b6b5352016-10-11 13:54:20 -0700519 if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir,
520 DMA_ATTR_NO_WARN))
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200521 goto out;
522
Ming Lin03b59292016-03-22 00:24:45 -0700523 if (!nvme_setup_prps(dev, req, size))
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200524 goto out_unmap;
525
526 ret = BLK_MQ_RQ_QUEUE_ERROR;
527 if (blk_integrity_rq(req)) {
528 if (blk_rq_count_integrity_sg(q, req->bio) != 1)
529 goto out_unmap;
530
Christoph Hellwigbf684052015-10-26 17:12:51 +0900531 sg_init_table(&iod->meta_sg, 1);
532 if (blk_rq_map_integrity_sg(q, req->bio, &iod->meta_sg) != 1)
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200533 goto out_unmap;
534
535 if (rq_data_dir(req))
536 nvme_dif_remap(req, nvme_dif_prep);
537
Christoph Hellwigbf684052015-10-26 17:12:51 +0900538 if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir))
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200539 goto out_unmap;
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200540 }
541
Christoph Hellwigeb793e22016-06-13 16:45:25 +0200542 cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
543 cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma);
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200544 if (blk_integrity_rq(req))
Christoph Hellwigbf684052015-10-26 17:12:51 +0900545 cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200546 return BLK_MQ_RQ_QUEUE_OK;
547
548out_unmap:
549 dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
550out:
551 return ret;
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200552}
553
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100554static void nvme_unmap_data(struct nvme_dev *dev, struct request *req)
Christoph Hellwigd4f6c3a2015-11-26 10:51:23 +0100555{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100556 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Christoph Hellwigd4f6c3a2015-11-26 10:51:23 +0100557 enum dma_data_direction dma_dir = rq_data_dir(req) ?
558 DMA_TO_DEVICE : DMA_FROM_DEVICE;
559
560 if (iod->nents) {
561 dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir);
562 if (blk_integrity_rq(req)) {
563 if (!rq_data_dir(req))
564 nvme_dif_remap(req, nvme_dif_complete);
Christoph Hellwigbf684052015-10-26 17:12:51 +0900565 dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir);
Christoph Hellwigd4f6c3a2015-11-26 10:51:23 +0100566 }
567 }
568
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100569 nvme_free_iod(dev, req);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500570}
571
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700572/*
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200573 * NOTE: ns is NULL when called on the admin queue.
574 */
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700575static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
576 const struct blk_mq_queue_data *bd)
Keith Busch53562be2014-04-29 11:41:29 -0600577{
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700578 struct nvme_ns *ns = hctx->queue->queuedata;
579 struct nvme_queue *nvmeq = hctx->driver_data;
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200580 struct nvme_dev *dev = nvmeq->dev;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700581 struct request *req = bd->rq;
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200582 struct nvme_command cmnd;
Ming Lin58b45602016-03-22 00:24:43 -0700583 unsigned map_len;
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200584 int ret = BLK_MQ_RQ_QUEUE_OK;
Keith Buschedd10d32014-04-03 16:45:23 -0600585
Keith Busche1e5e562015-02-19 13:39:03 -0700586 /*
587 * If formated with metadata, require the block layer provide a buffer
588 * unless this namespace is formated such that the metadata can be
589 * stripped/generated by the controller with PRACT=1.
590 */
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200591 if (ns && ns->ms && !blk_integrity_rq(req)) {
Keith Busch71feb362015-06-19 11:07:30 -0600592 if (!(ns->pi_type && ns->ms == 8) &&
593 req->cmd_type != REQ_TYPE_DRV_PRIV) {
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100594 blk_mq_end_request(req, -EFAULT);
Keith Busche1e5e562015-02-19 13:39:03 -0700595 return BLK_MQ_RQ_QUEUE_OK;
596 }
597 }
598
Ming Lin58b45602016-03-22 00:24:43 -0700599 map_len = nvme_map_len(req);
600 ret = nvme_init_iod(req, map_len, dev);
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100601 if (ret)
602 return ret;
Keith Buschedd10d32014-04-03 16:45:23 -0600603
Ming Lin8093f7c2016-04-12 13:10:14 -0600604 ret = nvme_setup_cmd(ns, req, &cmnd);
Ming Lin03b59292016-03-22 00:24:45 -0700605 if (ret)
606 goto out;
Keith Buschedd10d32014-04-03 16:45:23 -0600607
Ming Lin03b59292016-03-22 00:24:45 -0700608 if (req->nr_phys_segments)
609 ret = nvme_map_data(dev, req, map_len, &cmnd);
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700610
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200611 if (ret)
612 goto out;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700613
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200614 cmnd.common.command_id = req->tag;
Christoph Hellwigaae239e2015-11-26 12:59:50 +0100615 blk_mq_start_request(req);
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200616
617 spin_lock_irq(&nvmeq->q_lock);
Keith Buschae1fba22016-02-11 13:05:42 -0700618 if (unlikely(nvmeq->cq_vector < 0)) {
Keith Busch69d9a992016-02-24 09:15:56 -0700619 if (ns && !test_bit(NVME_NS_DEAD, &ns->flags))
620 ret = BLK_MQ_RQ_QUEUE_BUSY;
621 else
622 ret = BLK_MQ_RQ_QUEUE_ERROR;
Keith Buschae1fba22016-02-11 13:05:42 -0700623 spin_unlock_irq(&nvmeq->q_lock);
624 goto out;
625 }
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200626 __nvme_submit_cmd(nvmeq, &cmnd);
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700627 nvme_process_cq(nvmeq);
628 spin_unlock_irq(&nvmeq->q_lock);
629 return BLK_MQ_RQ_QUEUE_OK;
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200630out:
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100631 nvme_free_iod(dev, req);
Christoph Hellwigba1ca372015-10-16 07:58:38 +0200632 return ret;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500633}
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500634
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100635static void nvme_complete_rq(struct request *req)
636{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100637 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
638 struct nvme_dev *dev = iod->nvmeq->dev;
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100639 int error = 0;
640
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100641 nvme_unmap_data(dev, req);
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100642
643 if (unlikely(req->errors)) {
644 if (nvme_req_needs_retry(req, req->errors)) {
Keith Buschf80ec962016-07-12 16:20:31 -0700645 req->retries++;
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100646 nvme_requeue_req(req);
647 return;
648 }
649
650 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
651 error = req->errors;
652 else
653 error = nvme_error_status(req->errors);
654 }
655
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100656 if (unlikely(iod->aborted)) {
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -0700657 dev_warn(dev->ctrl.device,
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100658 "completing aborted command with status: %04x\n",
659 req->errors);
660 }
661
662 blk_mq_end_request(req, error);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500663}
664
Marta Rybczynskad783e0b2016-03-22 16:02:06 +0100665/* We read the CQE phase first to check if the rest of the entry is valid */
666static inline bool nvme_cqe_valid(struct nvme_queue *nvmeq, u16 head,
667 u16 phase)
668{
669 return (le16_to_cpu(nvmeq->cqes[head].status) & 1) == phase;
670}
671
Jens Axboea0fa9642015-11-03 20:37:26 -0700672static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500673{
Matthew Wilcox82123462011-01-20 13:24:06 -0500674 u16 head, phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500675
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500676 head = nvmeq->cq_head;
Matthew Wilcox82123462011-01-20 13:24:06 -0500677 phase = nvmeq->cq_phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500678
Marta Rybczynskad783e0b2016-03-22 16:02:06 +0100679 while (nvme_cqe_valid(nvmeq, head, phase)) {
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500680 struct nvme_completion cqe = nvmeq->cqes[head];
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100681 struct request *req;
Christoph Hellwigadf68f22015-11-28 15:42:28 +0100682
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500683 if (++head == nvmeq->q_depth) {
684 head = 0;
Matthew Wilcox82123462011-01-20 13:24:06 -0500685 phase = !phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500686 }
Christoph Hellwigadf68f22015-11-28 15:42:28 +0100687
Jens Axboea0fa9642015-11-03 20:37:26 -0700688 if (tag && *tag == cqe.command_id)
689 *tag = -1;
Christoph Hellwigadf68f22015-11-28 15:42:28 +0100690
Christoph Hellwigaae239e2015-11-26 12:59:50 +0100691 if (unlikely(cqe.command_id >= nvmeq->q_depth)) {
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -0700692 dev_warn(nvmeq->dev->ctrl.device,
Christoph Hellwigaae239e2015-11-26 12:59:50 +0100693 "invalid id %d completed on queue %d\n",
694 cqe.command_id, le16_to_cpu(cqe.sq_id));
695 continue;
696 }
697
Christoph Hellwigadf68f22015-11-28 15:42:28 +0100698 /*
699 * AEN requests are special as they don't time out and can
700 * survive any kind of queue freeze and often don't respond to
701 * aborts. We don't even bother to allocate a struct request
702 * for them but rather special case them here.
703 */
704 if (unlikely(nvmeq->qid == 0 &&
705 cqe.command_id >= NVME_AQ_BLKMQ_DEPTH)) {
Christoph Hellwigf866fc42016-04-26 13:52:00 +0200706 nvme_complete_async_event(&nvmeq->dev->ctrl, &cqe);
Christoph Hellwigadf68f22015-11-28 15:42:28 +0100707 continue;
708 }
709
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100710 req = blk_mq_tag_to_rq(*nvmeq->tags, cqe.command_id);
Christoph Hellwig1cb3cce2016-02-29 15:59:47 +0100711 if (req->cmd_type == REQ_TYPE_DRV_PRIV && req->special)
712 memcpy(req->special, &cqe, sizeof(cqe));
Marta Rybczynskad783e0b2016-03-22 16:02:06 +0100713 blk_mq_complete_request(req, le16_to_cpu(cqe.status) >> 1);
Christoph Hellwigeee417b2015-11-26 13:03:13 +0100714
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500715 }
716
717 /* If the controller ignores the cq head doorbell and continuously
718 * writes to the queue, it is theoretically possible to wrap around
719 * the queue twice and mistakenly return IRQ_NONE. Linux only
720 * requires that 0.1% of your interrupts are handled, so this isn't
721 * a big problem.
722 */
Matthew Wilcox82123462011-01-20 13:24:06 -0500723 if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
Jens Axboea0fa9642015-11-03 20:37:26 -0700724 return;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500725
Keith Busch604e8c82015-11-20 08:38:13 -0700726 if (likely(nvmeq->cq_vector >= 0))
727 writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500728 nvmeq->cq_head = head;
Matthew Wilcox82123462011-01-20 13:24:06 -0500729 nvmeq->cq_phase = phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500730
Matthew Wilcoxe9539f42013-06-24 11:47:34 -0400731 nvmeq->cqe_seen = 1;
Jens Axboea0fa9642015-11-03 20:37:26 -0700732}
733
734static void nvme_process_cq(struct nvme_queue *nvmeq)
735{
736 __nvme_process_cq(nvmeq, NULL);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500737}
738
739static irqreturn_t nvme_irq(int irq, void *data)
740{
Matthew Wilcox58ffacb2011-02-06 07:28:06 -0500741 irqreturn_t result;
742 struct nvme_queue *nvmeq = data;
743 spin_lock(&nvmeq->q_lock);
Matthew Wilcoxe9539f42013-06-24 11:47:34 -0400744 nvme_process_cq(nvmeq);
745 result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE;
746 nvmeq->cqe_seen = 0;
Matthew Wilcox58ffacb2011-02-06 07:28:06 -0500747 spin_unlock(&nvmeq->q_lock);
748 return result;
749}
750
751static irqreturn_t nvme_irq_check(int irq, void *data)
752{
753 struct nvme_queue *nvmeq = data;
Marta Rybczynskad783e0b2016-03-22 16:02:06 +0100754 if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase))
755 return IRQ_WAKE_THREAD;
756 return IRQ_NONE;
Matthew Wilcox58ffacb2011-02-06 07:28:06 -0500757}
758
Jens Axboea0fa9642015-11-03 20:37:26 -0700759static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
760{
761 struct nvme_queue *nvmeq = hctx->driver_data;
762
Marta Rybczynskad783e0b2016-03-22 16:02:06 +0100763 if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) {
Jens Axboea0fa9642015-11-03 20:37:26 -0700764 spin_lock_irq(&nvmeq->q_lock);
765 __nvme_process_cq(nvmeq, &tag);
766 spin_unlock_irq(&nvmeq->q_lock);
767
768 if (tag == -1)
769 return 1;
770 }
771
772 return 0;
773}
774
Christoph Hellwigf866fc42016-04-26 13:52:00 +0200775static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500776{
Christoph Hellwigf866fc42016-04-26 13:52:00 +0200777 struct nvme_dev *dev = to_nvme_dev(ctrl);
Christoph Hellwig9396dec2016-02-29 15:59:44 +0100778 struct nvme_queue *nvmeq = dev->queues[0];
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700779 struct nvme_command c;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700780
781 memset(&c, 0, sizeof(c));
782 c.common.opcode = nvme_admin_async_event;
Christoph Hellwigf866fc42016-04-26 13:52:00 +0200783 c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700784
Christoph Hellwig9396dec2016-02-29 15:59:44 +0100785 spin_lock_irq(&nvmeq->q_lock);
Christoph Hellwigf866fc42016-04-26 13:52:00 +0200786 __nvme_submit_cmd(nvmeq, &c);
Christoph Hellwig9396dec2016-02-29 15:59:44 +0100787 spin_unlock_irq(&nvmeq->q_lock);
Keith Busch4d115422013-12-10 13:10:40 -0700788}
789
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500790static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
791{
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500792 struct nvme_command c;
793
794 memset(&c, 0, sizeof(c));
795 c.delete_queue.opcode = opcode;
796 c.delete_queue.qid = cpu_to_le16(id);
797
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100798 return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500799}
800
801static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
802 struct nvme_queue *nvmeq)
803{
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500804 struct nvme_command c;
805 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;
806
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200807 /*
808 * Note: we (ab)use the fact the the prp fields survive if no data
809 * is attached to the request.
810 */
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500811 memset(&c, 0, sizeof(c));
812 c.create_cq.opcode = nvme_admin_create_cq;
813 c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr);
814 c.create_cq.cqid = cpu_to_le16(qid);
815 c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
816 c.create_cq.cq_flags = cpu_to_le16(flags);
817 c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector);
818
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100819 return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500820}
821
822static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
823 struct nvme_queue *nvmeq)
824{
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500825 struct nvme_command c;
826 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM;
827
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200828 /*
829 * Note: we (ab)use the fact the the prp fields survive if no data
830 * is attached to the request.
831 */
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500832 memset(&c, 0, sizeof(c));
833 c.create_sq.opcode = nvme_admin_create_sq;
834 c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr);
835 c.create_sq.sqid = cpu_to_le16(qid);
836 c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
837 c.create_sq.sq_flags = cpu_to_le16(flags);
838 c.create_sq.cqid = cpu_to_le16(qid);
839
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100840 return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500841}
842
843static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid)
844{
845 return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid);
846}
847
848static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid)
849{
850 return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid);
851}
852
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100853static void abort_endio(struct request *req, int error)
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400854{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100855 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
856 struct nvme_queue *nvmeq = iod->nvmeq;
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100857 u16 status = req->errors;
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400858
Christoph Hellwig1cb3cce2016-02-29 15:59:47 +0100859 dev_warn(nvmeq->dev->ctrl.device, "Abort status: 0x%x", status);
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100860 atomic_inc(&nvmeq->dev->ctrl.abort_limit);
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100861 blk_mq_free_request(req);
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200862}
863
Christoph Hellwig31c7c7d2015-10-22 14:03:35 +0200864static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved)
Christoph Hellwigd29ec822015-05-22 11:12:46 +0200865{
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100866 struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
867 struct nvme_queue *nvmeq = iod->nvmeq;
Keith Buschc30341d2013-12-10 13:10:38 -0700868 struct nvme_dev *dev = nvmeq->dev;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700869 struct request *abort_req;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700870 struct nvme_command cmd;
Keith Buschc30341d2013-12-10 13:10:38 -0700871
Christoph Hellwig31c7c7d2015-10-22 14:03:35 +0200872 /*
Christoph Hellwigfd634f412015-11-26 12:42:26 +0100873 * Shutdown immediately if controller times out while starting. The
874 * reset work will see the pci device disabled when it gets the forced
875 * cancellation error. All outstanding requests are completed on
876 * shutdown, so we return BLK_EH_HANDLED.
877 */
Christoph Hellwigbb8d2612016-04-26 13:51:57 +0200878 if (dev->ctrl.state == NVME_CTRL_RESETTING) {
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -0700879 dev_warn(dev->ctrl.device,
Christoph Hellwigfd634f412015-11-26 12:42:26 +0100880 "I/O %d QID %d timeout, disable controller\n",
881 req->tag, nvmeq->qid);
Keith Buscha5cdb682016-01-12 14:41:18 -0700882 nvme_dev_disable(dev, false);
Christoph Hellwigfd634f412015-11-26 12:42:26 +0100883 req->errors = NVME_SC_CANCELLED;
884 return BLK_EH_HANDLED;
Keith Buschc30341d2013-12-10 13:10:38 -0700885 }
886
Christoph Hellwigfd634f412015-11-26 12:42:26 +0100887 /*
888 * Shutdown the controller immediately and schedule a reset if the
889 * command was already aborted once before and still hasn't been
890 * returned to the driver, or if this is the admin queue.
Christoph Hellwig31c7c7d2015-10-22 14:03:35 +0200891 */
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100892 if (!nvmeq->qid || iod->aborted) {
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -0700893 dev_warn(dev->ctrl.device,
Keith Busche1569a12015-11-26 12:11:07 +0100894 "I/O %d QID %d timeout, reset controller\n",
895 req->tag, nvmeq->qid);
Keith Buscha5cdb682016-01-12 14:41:18 -0700896 nvme_dev_disable(dev, false);
Keith Buschc5f6ce92016-10-05 16:32:45 -0400897 nvme_reset(dev);
Keith Buschc30341d2013-12-10 13:10:38 -0700898
Keith Busche1569a12015-11-26 12:11:07 +0100899 /*
900 * Mark the request as handled, since the inline shutdown
901 * forces all outstanding requests to complete.
902 */
903 req->errors = NVME_SC_CANCELLED;
904 return BLK_EH_HANDLED;
Keith Buschc30341d2013-12-10 13:10:38 -0700905 }
Keith Buschc30341d2013-12-10 13:10:38 -0700906
Christoph Hellwigf4800d62015-11-28 15:43:10 +0100907 iod->aborted = 1;
Keith Buschc30341d2013-12-10 13:10:38 -0700908
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100909 if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) {
910 atomic_inc(&dev->ctrl.abort_limit);
911 return BLK_EH_RESET_TIMER;
912 }
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700913
Keith Buschc30341d2013-12-10 13:10:38 -0700914 memset(&cmd, 0, sizeof(cmd));
915 cmd.abort.opcode = nvme_admin_abort_cmd;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700916 cmd.abort.cid = req->tag;
Keith Buschc30341d2013-12-10 13:10:38 -0700917 cmd.abort.sqid = cpu_to_le16(nvmeq->qid);
Keith Buschc30341d2013-12-10 13:10:38 -0700918
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -0700919 dev_warn(nvmeq->dev->ctrl.device,
920 "I/O %d QID %d timeout, aborting\n",
921 req->tag, nvmeq->qid);
Keith Buschc30341d2013-12-10 13:10:38 -0700922
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100923 abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd,
Christoph Hellwigeb71f432016-06-13 16:45:23 +0200924 BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
Christoph Hellwig6bf25d12015-11-20 09:36:44 +0100925 if (IS_ERR(abort_req)) {
926 atomic_inc(&dev->ctrl.abort_limit);
Christoph Hellwig31c7c7d2015-10-22 14:03:35 +0200927 return BLK_EH_RESET_TIMER;
Christoph Hellwig6bf25d12015-11-20 09:36:44 +0100928 }
Keith Buschc30341d2013-12-10 13:10:38 -0700929
Christoph Hellwige7a2a872015-11-16 10:39:48 +0100930 abort_req->timeout = ADMIN_TIMEOUT;
931 abort_req->end_io_data = NULL;
932 blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio);
Keith Busch07836e62015-02-19 10:34:48 -0700933
Keith Busch7a509a62015-01-07 18:55:53 -0700934 /*
935 * The aborted req will be completed on receiving the abort req.
936 * We enable the timer again. If hit twice, it'll cause a device reset,
937 * as the device then is in a faulty state.
938 */
Keith Busch07836e62015-02-19 10:34:48 -0700939 return BLK_EH_RESET_TIMER;
Matthew Wilcoxa09115b2012-08-07 15:56:23 -0400940}
941
Keith Buschf435c282014-07-07 09:14:42 -0600942static void nvme_free_queue(struct nvme_queue *nvmeq)
Matthew Wilcox9e866772012-08-03 13:55:56 -0400943{
944 dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
945 (void *)nvmeq->cqes, nvmeq->cq_dma_addr);
Jon Derrick8ffaadf2015-07-20 10:14:09 -0600946 if (nvmeq->sq_cmds)
947 dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
Matthew Wilcox9e866772012-08-03 13:55:56 -0400948 nvmeq->sq_cmds, nvmeq->sq_dma_addr);
949 kfree(nvmeq);
950}
951
Keith Buscha1a5ef92013-12-16 13:50:00 -0500952static void nvme_free_queues(struct nvme_dev *dev, int lowest)
Keith Busch22404272013-07-15 15:02:20 -0600953{
954 int i;
955
Keith Buscha1a5ef92013-12-16 13:50:00 -0500956 for (i = dev->queue_count - 1; i >= lowest; i--) {
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700957 struct nvme_queue *nvmeq = dev->queues[i];
Keith Busch22404272013-07-15 15:02:20 -0600958 dev->queue_count--;
Matias Bjørlinga4aea562014-11-04 08:20:14 -0700959 dev->queues[i] = NULL;
Keith Buschf435c282014-07-07 09:14:42 -0600960 nvme_free_queue(nvmeq);
kaoudis121c7ad2015-01-14 21:01:58 -0700961 }
Keith Busch22404272013-07-15 15:02:20 -0600962}
963
Keith Busch4d115422013-12-10 13:10:40 -0700964/**
965 * nvme_suspend_queue - put queue into suspended state
966 * @nvmeq - queue to suspend
Keith Busch4d115422013-12-10 13:10:40 -0700967 */
968static int nvme_suspend_queue(struct nvme_queue *nvmeq)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500969{
Keith Busch2b25d982014-12-22 12:59:04 -0700970 int vector;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500971
Matthew Wilcoxa09115b2012-08-07 15:56:23 -0400972 spin_lock_irq(&nvmeq->q_lock);
Keith Busch2b25d982014-12-22 12:59:04 -0700973 if (nvmeq->cq_vector == -1) {
974 spin_unlock_irq(&nvmeq->q_lock);
975 return 1;
976 }
Christoph Hellwigdca51e72016-09-14 16:18:57 +0200977 vector = nvmeq_irq(nvmeq);
Keith Busch42f61422014-03-24 10:46:25 -0600978 nvmeq->dev->online_queues--;
Keith Busch2b25d982014-12-22 12:59:04 -0700979 nvmeq->cq_vector = -1;
Matthew Wilcoxa09115b2012-08-07 15:56:23 -0400980 spin_unlock_irq(&nvmeq->q_lock);
981
Christoph Hellwig1c63dc62015-11-26 10:06:56 +0100982 if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q)
Keith Busch25646262016-01-04 09:10:57 -0700983 blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q);
Keith Busch6df3dbc2015-03-26 13:49:33 -0600984
Matthew Wilcoxaba20802011-03-27 08:52:06 -0400985 free_irq(vector, nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500986
Keith Busch4d115422013-12-10 13:10:40 -0700987 return 0;
988}
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500989
Keith Buscha5cdb682016-01-12 14:41:18 -0700990static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
Keith Busch4d115422013-12-10 13:10:40 -0700991{
Keith Buscha5cdb682016-01-12 14:41:18 -0700992 struct nvme_queue *nvmeq = dev->queues[0];
Keith Busch4d115422013-12-10 13:10:40 -0700993
994 if (!nvmeq)
995 return;
996 if (nvme_suspend_queue(nvmeq))
997 return;
998
Keith Buscha5cdb682016-01-12 14:41:18 -0700999 if (shutdown)
1000 nvme_shutdown_ctrl(&dev->ctrl);
1001 else
1002 nvme_disable_ctrl(&dev->ctrl, lo_hi_readq(
1003 dev->bar + NVME_REG_CAP));
Keith Busch07836e62015-02-19 10:34:48 -07001004
1005 spin_lock_irq(&nvmeq->q_lock);
1006 nvme_process_cq(nvmeq);
1007 spin_unlock_irq(&nvmeq->q_lock);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001008}
1009
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001010static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,
1011 int entry_size)
1012{
1013 int q_depth = dev->q_depth;
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001014 unsigned q_size_aligned = roundup(q_depth * entry_size,
1015 dev->ctrl.page_size);
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001016
1017 if (q_size_aligned * nr_io_queues > dev->cmb_size) {
Jon Derrickc45f5c92015-07-21 15:08:13 -06001018 u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues);
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001019 mem_per_q = round_down(mem_per_q, dev->ctrl.page_size);
Jon Derrickc45f5c92015-07-21 15:08:13 -06001020 q_depth = div_u64(mem_per_q, entry_size);
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001021
1022 /*
1023 * Ensure the reduced q_depth is above some threshold where it
1024 * would be better to map queues in system memory with the
1025 * original depth
1026 */
1027 if (q_depth < 64)
1028 return -ENOMEM;
1029 }
1030
1031 return q_depth;
1032}
1033
1034static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq,
1035 int qid, int depth)
1036{
1037 if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) {
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001038 unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth),
1039 dev->ctrl.page_size);
Christoph Hellwigc83bbed2017-10-01 09:37:35 +02001040 nvmeq->sq_dma_addr = dev->cmb_bus_addr + offset;
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001041 nvmeq->sq_cmds_io = dev->cmb + offset;
1042 } else {
1043 nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth),
1044 &nvmeq->sq_dma_addr, GFP_KERNEL);
1045 if (!nvmeq->sq_cmds)
1046 return -ENOMEM;
1047 }
1048
1049 return 0;
1050}
1051
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001052static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
Keith Busch2b25d982014-12-22 12:59:04 -07001053 int depth)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001054{
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001055 struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001056 if (!nvmeq)
1057 return NULL;
1058
Christoph Hellwige75ec752015-05-22 11:12:39 +02001059 nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth),
Joe Perches4d51abf2014-06-15 13:37:33 -07001060 &nvmeq->cq_dma_addr, GFP_KERNEL);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001061 if (!nvmeq->cqes)
1062 goto free_nvmeq;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001063
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001064 if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth))
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001065 goto free_cqdma;
1066
Christoph Hellwige75ec752015-05-22 11:12:39 +02001067 nvmeq->q_dmadev = dev->dev;
Matthew Wilcox091b6092011-02-10 09:56:01 -05001068 nvmeq->dev = dev;
Matthew Wilcox3193f072014-01-27 15:57:22 -05001069 snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001070 dev->ctrl.instance, qid);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001071 spin_lock_init(&nvmeq->q_lock);
1072 nvmeq->cq_head = 0;
Matthew Wilcox82123462011-01-20 13:24:06 -05001073 nvmeq->cq_phase = 1;
Haiyan Hub80d5cc2013-09-10 11:25:37 +08001074 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001075 nvmeq->q_depth = depth;
Keith Buschc30341d2013-12-10 13:10:38 -07001076 nvmeq->qid = qid;
Jon Derrick758dd7f2015-06-30 11:22:52 -06001077 nvmeq->cq_vector = -1;
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001078 dev->queues[qid] = nvmeq;
Jon Derrick36a7e992015-05-27 12:26:23 -06001079 dev->queue_count++;
1080
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001081 return nvmeq;
1082
1083 free_cqdma:
Christoph Hellwige75ec752015-05-22 11:12:39 +02001084 dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001085 nvmeq->cq_dma_addr);
1086 free_nvmeq:
1087 kfree(nvmeq);
1088 return NULL;
1089}
1090
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001091static int queue_request_irq(struct nvme_queue *nvmeq)
Matthew Wilcox30010822011-01-20 09:10:15 -05001092{
Matthew Wilcox58ffacb2011-02-06 07:28:06 -05001093 if (use_threaded_interrupts)
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001094 return request_threaded_irq(nvmeq_irq(nvmeq), nvme_irq_check,
1095 nvme_irq, IRQF_SHARED, nvmeq->irqname, nvmeq);
1096 else
1097 return request_irq(nvmeq_irq(nvmeq), nvme_irq, IRQF_SHARED,
1098 nvmeq->irqname, nvmeq);
Matthew Wilcox30010822011-01-20 09:10:15 -05001099}
1100
Keith Busch22404272013-07-15 15:02:20 -06001101static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001102{
Keith Busch22404272013-07-15 15:02:20 -06001103 struct nvme_dev *dev = nvmeq->dev;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001104
Keith Busch7be50e92014-09-10 15:48:47 -06001105 spin_lock_irq(&nvmeq->q_lock);
Keith Busch22404272013-07-15 15:02:20 -06001106 nvmeq->sq_tail = 0;
1107 nvmeq->cq_head = 0;
1108 nvmeq->cq_phase = 1;
Haiyan Hub80d5cc2013-09-10 11:25:37 +08001109 nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
Keith Busch22404272013-07-15 15:02:20 -06001110 memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
Keith Busch42f61422014-03-24 10:46:25 -06001111 dev->online_queues++;
Keith Busch7be50e92014-09-10 15:48:47 -06001112 spin_unlock_irq(&nvmeq->q_lock);
Keith Busch22404272013-07-15 15:02:20 -06001113}
1114
1115static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
1116{
1117 struct nvme_dev *dev = nvmeq->dev;
1118 int result;
Matthew Wilcox3f85d502011-02-01 08:39:04 -05001119
Keith Busch2b25d982014-12-22 12:59:04 -07001120 nvmeq->cq_vector = qid - 1;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001121 result = adapter_alloc_cq(dev, qid, nvmeq);
1122 if (result < 0)
Jianchao Wang23f9fb02018-02-15 19:13:41 +08001123 goto release_vector;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001124
1125 result = adapter_alloc_sq(dev, qid, nvmeq);
1126 if (result < 0)
1127 goto release_cq;
1128
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001129 result = queue_request_irq(nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001130 if (result < 0)
1131 goto release_sq;
1132
Keith Busch22404272013-07-15 15:02:20 -06001133 nvme_init_queue(nvmeq, qid);
Keith Busch22404272013-07-15 15:02:20 -06001134 return result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001135
1136 release_sq:
Jianchao Wang23f9fb02018-02-15 19:13:41 +08001137 dev->online_queues--;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001138 adapter_delete_sq(dev, qid);
1139 release_cq:
1140 adapter_delete_cq(dev, qid);
Jianchao Wang23f9fb02018-02-15 19:13:41 +08001141 release_vector:
1142 nvmeq->cq_vector = -1;
Keith Busch22404272013-07-15 15:02:20 -06001143 return result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001144}
1145
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001146static struct blk_mq_ops nvme_mq_admin_ops = {
Christoph Hellwigd29ec822015-05-22 11:12:46 +02001147 .queue_rq = nvme_queue_rq,
Christoph Hellwigeee417b2015-11-26 13:03:13 +01001148 .complete = nvme_complete_rq,
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001149 .init_hctx = nvme_admin_init_hctx,
Keith Busch4af0e212015-06-08 10:08:13 -06001150 .exit_hctx = nvme_admin_exit_hctx,
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001151 .init_request = nvme_admin_init_request,
1152 .timeout = nvme_timeout,
1153};
1154
1155static struct blk_mq_ops nvme_mq_ops = {
1156 .queue_rq = nvme_queue_rq,
Christoph Hellwigeee417b2015-11-26 13:03:13 +01001157 .complete = nvme_complete_rq,
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001158 .init_hctx = nvme_init_hctx,
1159 .init_request = nvme_init_request,
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001160 .map_queues = nvme_pci_map_queues,
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001161 .timeout = nvme_timeout,
Jens Axboea0fa9642015-11-03 20:37:26 -07001162 .poll = nvme_poll,
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001163};
1164
Keith Buschea191d22015-01-07 18:55:49 -07001165static void nvme_dev_remove_admin(struct nvme_dev *dev)
1166{
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001167 if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) {
Keith Busch69d9a992016-02-24 09:15:56 -07001168 /*
1169 * If the controller was reset during removal, it's possible
1170 * user requests may be waiting on a stopped queue. Start the
1171 * queue to flush these to completion.
1172 */
1173 blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true);
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001174 blk_cleanup_queue(dev->ctrl.admin_q);
Keith Buschea191d22015-01-07 18:55:49 -07001175 blk_mq_free_tag_set(&dev->admin_tagset);
1176 }
1177}
1178
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001179static int nvme_alloc_admin_tags(struct nvme_dev *dev)
1180{
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001181 if (!dev->ctrl.admin_q) {
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001182 dev->admin_tagset.ops = &nvme_mq_admin_ops;
1183 dev->admin_tagset.nr_hw_queues = 1;
Keith Busche3e9d502016-01-04 09:10:55 -07001184
1185 /*
1186 * Subtract one to leave an empty queue entry for 'Full Queue'
1187 * condition. See NVM-Express 1.2 specification, section 4.1.2.
1188 */
1189 dev->admin_tagset.queue_depth = NVME_AQ_BLKMQ_DEPTH - 1;
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001190 dev->admin_tagset.timeout = ADMIN_TIMEOUT;
Christoph Hellwige75ec752015-05-22 11:12:39 +02001191 dev->admin_tagset.numa_node = dev_to_node(dev->dev);
Jens Axboeac3dd5b2015-01-22 12:07:58 -07001192 dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001193 dev->admin_tagset.driver_data = dev;
1194
1195 if (blk_mq_alloc_tag_set(&dev->admin_tagset))
1196 return -ENOMEM;
1197
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001198 dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset);
1199 if (IS_ERR(dev->ctrl.admin_q)) {
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001200 blk_mq_free_tag_set(&dev->admin_tagset);
1201 return -ENOMEM;
1202 }
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001203 if (!blk_get_queue(dev->ctrl.admin_q)) {
Keith Buschea191d22015-01-07 18:55:49 -07001204 nvme_dev_remove_admin(dev);
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001205 dev->ctrl.admin_q = NULL;
Keith Buschea191d22015-01-07 18:55:49 -07001206 return -ENODEV;
1207 }
Keith Busch0fb59cb2015-01-07 18:55:50 -07001208 } else
Keith Busch25646262016-01-04 09:10:57 -07001209 blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true);
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001210
1211 return 0;
1212}
1213
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001214static int nvme_configure_admin_queue(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001215{
Matthew Wilcoxba47e382013-05-04 06:43:16 -04001216 int result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001217 u32 aqa;
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001218 u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001219 struct nvme_queue *nvmeq;
Keith Busch1d090622014-06-23 11:34:01 -06001220
Gabriel Krisman Bertazi8ef20742016-10-19 09:51:05 -06001221 dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ?
Keith Buschdfbac8c2015-08-10 15:20:40 -06001222 NVME_CAP_NSSRC(cap) : 0;
1223
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001224 if (dev->subsystem &&
1225 (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO))
1226 writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS);
Keith Buschdfbac8c2015-08-10 15:20:40 -06001227
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001228 result = nvme_disable_ctrl(&dev->ctrl, cap);
Matthew Wilcoxba47e382013-05-04 06:43:16 -04001229 if (result < 0)
1230 return result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001231
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001232 nvmeq = dev->queues[0];
Keith Buschcd638942013-07-15 15:02:23 -06001233 if (!nvmeq) {
Keith Busch2b25d982014-12-22 12:59:04 -07001234 nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH);
Keith Buschcd638942013-07-15 15:02:23 -06001235 if (!nvmeq)
1236 return -ENOMEM;
Keith Buschcd638942013-07-15 15:02:23 -06001237 }
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001238
1239 aqa = nvmeq->q_depth - 1;
1240 aqa |= aqa << 16;
1241
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001242 writel(aqa, dev->bar + NVME_REG_AQA);
1243 lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ);
1244 lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ);
Keith Busch1d090622014-06-23 11:34:01 -06001245
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001246 result = nvme_enable_ctrl(&dev->ctrl, cap);
Keith Busch025c5572013-05-01 13:07:51 -06001247 if (result)
Keith Buschd4875622016-11-15 15:56:26 -05001248 return result;
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001249
Keith Busch2b25d982014-12-22 12:59:04 -07001250 nvmeq->cq_vector = 0;
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001251 result = queue_request_irq(nvmeq);
Jon Derrick758dd7f2015-06-30 11:22:52 -06001252 if (result) {
1253 nvmeq->cq_vector = -1;
Keith Buschd4875622016-11-15 15:56:26 -05001254 return result;
Jon Derrick758dd7f2015-06-30 11:22:52 -06001255 }
Keith Busch025c5572013-05-01 13:07:51 -06001256
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001257 return result;
1258}
1259
Guilherme G. Piccolic875a702016-04-13 11:08:20 -03001260static bool nvme_should_reset(struct nvme_dev *dev, u32 csts)
1261{
1262
1263 /* If true, indicates loss of adapter communication, possibly by a
1264 * NVMe Subsystem reset.
1265 */
1266 bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO);
1267
1268 /* If there is a reset ongoing, we shouldn't reset again. */
Rakesh Pandit383f15e2017-06-05 14:43:11 +03001269 if (dev->ctrl.state == NVME_CTRL_RESETTING)
Guilherme G. Piccolic875a702016-04-13 11:08:20 -03001270 return false;
1271
1272 /* We shouldn't reset unless the controller is on fatal error state
1273 * _or_ if we lost the communication with it.
1274 */
1275 if (!(csts & NVME_CSTS_CFS) && !nssro)
1276 return false;
1277
1278 /* If PCI error recovery process is happening, we cannot reset or
1279 * the recovery mechanism will surely fail.
1280 */
1281 if (pci_channel_offline(to_pci_dev(dev->dev)))
1282 return false;
1283
1284 return true;
1285}
1286
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001287static void nvme_watchdog_timer(unsigned long data)
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001288{
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001289 struct nvme_dev *dev = (struct nvme_dev *)data;
1290 u32 csts = readl(dev->bar + NVME_REG_CSTS);
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001291
Guilherme G. Piccolic875a702016-04-13 11:08:20 -03001292 /* Skip controllers under certain specific conditions. */
1293 if (nvme_should_reset(dev, csts)) {
Keith Buschc5f6ce92016-10-05 16:32:45 -04001294 if (!nvme_reset(dev))
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001295 dev_warn(dev->dev,
1296 "Failed status: 0x%x, reset controller.\n",
1297 csts);
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001298 return;
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001299 }
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001300
1301 mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001302}
1303
Christoph Hellwig749941f2015-11-26 11:46:39 +01001304static int nvme_create_io_queues(struct nvme_dev *dev)
Keith Busch42f61422014-03-24 10:46:25 -06001305{
Keith Busch949928c2015-12-17 17:08:15 -07001306 unsigned i, max;
Christoph Hellwig749941f2015-11-26 11:46:39 +01001307 int ret = 0;
Keith Busch42f61422014-03-24 10:46:25 -06001308
Christoph Hellwig749941f2015-11-26 11:46:39 +01001309 for (i = dev->queue_count; i <= dev->max_qid; i++) {
1310 if (!nvme_alloc_queue(dev, i, dev->q_depth)) {
1311 ret = -ENOMEM;
Keith Busch42f61422014-03-24 10:46:25 -06001312 break;
Christoph Hellwig749941f2015-11-26 11:46:39 +01001313 }
1314 }
Keith Busch42f61422014-03-24 10:46:25 -06001315
Keith Busch949928c2015-12-17 17:08:15 -07001316 max = min(dev->max_qid, dev->queue_count - 1);
1317 for (i = dev->online_queues; i <= max; i++) {
Christoph Hellwig749941f2015-11-26 11:46:39 +01001318 ret = nvme_create_queue(dev->queues[i], i);
Keith Buschd4875622016-11-15 15:56:26 -05001319 if (ret)
Keith Busch42f61422014-03-24 10:46:25 -06001320 break;
Matthew Wilcox27e81662014-04-11 11:58:45 -04001321 }
Christoph Hellwig749941f2015-11-26 11:46:39 +01001322
1323 /*
1324 * Ignore failing Create SQ/CQ commands, we can continue with less
1325 * than the desired aount of queues, and even a controller without
1326 * I/O queues an still be used to issue admin commands. This might
1327 * be useful to upgrade a buggy firmware for example.
1328 */
1329 return ret >= 0 ? 0 : ret;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001330}
1331
Stephen Bates202021c2016-10-05 20:01:12 -06001332static ssize_t nvme_cmb_show(struct device *dev,
1333 struct device_attribute *attr,
1334 char *buf)
1335{
1336 struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev));
1337
1338 return snprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz : x%08x\n",
1339 ndev->cmbloc, ndev->cmbsz);
1340}
1341static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL);
1342
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001343static void __iomem *nvme_map_cmb(struct nvme_dev *dev)
1344{
1345 u64 szu, size, offset;
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001346 resource_size_t bar_size;
1347 struct pci_dev *pdev = to_pci_dev(dev->dev);
1348 void __iomem *cmb;
Christoph Hellwigc83bbed2017-10-01 09:37:35 +02001349 int bar;
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001350
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001351 dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ);
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001352 if (!(NVME_CMB_SZ(dev->cmbsz)))
1353 return NULL;
Stephen Bates202021c2016-10-05 20:01:12 -06001354 dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC);
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001355
Stephen Bates202021c2016-10-05 20:01:12 -06001356 if (!use_cmb_sqes)
1357 return NULL;
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001358
1359 szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz));
1360 size = szu * NVME_CMB_SZ(dev->cmbsz);
Stephen Bates202021c2016-10-05 20:01:12 -06001361 offset = szu * NVME_CMB_OFST(dev->cmbloc);
Christoph Hellwigc83bbed2017-10-01 09:37:35 +02001362 bar = NVME_CMB_BIR(dev->cmbloc);
1363 bar_size = pci_resource_len(pdev, bar);
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001364
1365 if (offset > bar_size)
1366 return NULL;
1367
1368 /*
1369 * Controllers may support a CMB size larger than their BAR,
1370 * for example, due to being behind a bridge. Reduce the CMB to
1371 * the reported size of the BAR
1372 */
1373 if (size > bar_size - offset)
1374 size = bar_size - offset;
1375
Christoph Hellwigc83bbed2017-10-01 09:37:35 +02001376 cmb = ioremap_wc(pci_resource_start(pdev, bar) + offset, size);
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001377 if (!cmb)
1378 return NULL;
1379
Christoph Hellwigc83bbed2017-10-01 09:37:35 +02001380 dev->cmb_bus_addr = pci_bus_address(pdev, bar) + offset;
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001381 dev->cmb_size = size;
1382 return cmb;
1383}
1384
1385static inline void nvme_release_cmb(struct nvme_dev *dev)
1386{
1387 if (dev->cmb) {
1388 iounmap(dev->cmb);
1389 dev->cmb = NULL;
Jon Derrick6d6a43a2017-05-05 14:52:06 -06001390 if (dev->cmbsz) {
1391 sysfs_remove_file_from_group(&dev->ctrl.device->kobj,
1392 &dev_attr_cmb.attr, NULL);
1393 dev->cmbsz = 0;
1394 }
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001395 }
1396}
1397
Keith Busch9d713c22013-07-15 15:02:24 -06001398static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
1399{
Haiyan Hub80d5cc2013-09-10 11:25:37 +08001400 return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
Keith Busch9d713c22013-07-15 15:02:24 -06001401}
1402
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001403static int nvme_setup_io_queues(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001404{
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001405 struct nvme_queue *adminq = dev->queues[0];
Christoph Hellwige75ec752015-05-22 11:12:39 +02001406 struct pci_dev *pdev = to_pci_dev(dev->dev);
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001407 int result, nr_io_queues, size;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001408
Keith Busch2800b8e2016-05-13 12:38:09 -06001409 nr_io_queues = num_online_cpus();
Christoph Hellwig9a0be7a2015-11-26 11:09:06 +01001410 result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues);
1411 if (result < 0)
Matthew Wilcox1b234842011-01-20 13:01:49 -05001412 return result;
Christoph Hellwig9a0be7a2015-11-26 11:09:06 +01001413
Christoph Hellwigf5fa90d2016-06-06 23:20:50 +02001414 if (nr_io_queues == 0)
Keith Buscha5229052016-04-08 16:09:10 -06001415 return 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001416
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001417 if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) {
1418 result = nvme_cmb_qdepth(dev, nr_io_queues,
1419 sizeof(struct nvme_command));
1420 if (result > 0)
1421 dev->q_depth = result;
1422 else
1423 nvme_release_cmb(dev);
1424 }
1425
Keith Busch9d713c22013-07-15 15:02:24 -06001426 size = db_bar_size(dev, nr_io_queues);
1427 if (size > 8192) {
Matthew Wilcoxf1938f62011-10-20 17:00:41 -04001428 iounmap(dev->bar);
Keith Busch9d713c22013-07-15 15:02:24 -06001429 do {
1430 dev->bar = ioremap(pci_resource_start(pdev, 0), size);
1431 if (dev->bar)
1432 break;
1433 if (!--nr_io_queues)
1434 return -ENOMEM;
1435 size = db_bar_size(dev, nr_io_queues);
1436 } while (1);
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001437 dev->dbs = dev->bar + 4096;
Keith Busch5a92e702014-02-21 14:13:44 -07001438 adminq->q_db = dev->dbs;
Matthew Wilcoxf1938f62011-10-20 17:00:41 -04001439 }
1440
Keith Busch9d713c22013-07-15 15:02:24 -06001441 /* Deregister the admin queue's interrupt */
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001442 free_irq(pci_irq_vector(pdev, 0), adminq);
Keith Busch9d713c22013-07-15 15:02:24 -06001443
Jens Axboee32efbf2014-11-14 09:49:26 -07001444 /*
1445 * If we enable msix early due to not intx, disable it again before
1446 * setting up the full range we need.
1447 */
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001448 pci_free_irq_vectors(pdev);
1449 nr_io_queues = pci_alloc_irq_vectors(pdev, 1, nr_io_queues,
1450 PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY);
1451 if (nr_io_queues <= 0)
1452 return -EIO;
1453 dev->max_qid = nr_io_queues;
Matthew Wilcox1b234842011-01-20 13:01:49 -05001454
Matthew Wilcox063a8092013-06-20 10:53:48 -04001455 /*
1456 * Should investigate if there's a performance win from allocating
1457 * more queues than interrupt vectors; it might allow the submission
1458 * path to scale better, even if the receive path is limited by the
1459 * number of interrupts.
1460 */
Ramachandra Rao Gajulafa08a392013-05-11 15:19:31 -07001461
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001462 result = queue_request_irq(adminq);
Jon Derrick758dd7f2015-06-30 11:22:52 -06001463 if (result) {
1464 adminq->cq_vector = -1;
Keith Buschd4875622016-11-15 15:56:26 -05001465 return result;
Jon Derrick758dd7f2015-06-30 11:22:52 -06001466 }
Christoph Hellwig749941f2015-11-26 11:46:39 +01001467 return nvme_create_io_queues(dev);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001468}
1469
Keith Buschdb3cbff2016-01-12 14:41:17 -07001470static void nvme_del_queue_end(struct request *req, int error)
1471{
1472 struct nvme_queue *nvmeq = req->end_io_data;
1473
1474 blk_mq_free_request(req);
1475 complete(&nvmeq->dev->ioq_wait);
1476}
1477
1478static void nvme_del_cq_end(struct request *req, int error)
1479{
1480 struct nvme_queue *nvmeq = req->end_io_data;
1481
1482 if (!error) {
1483 unsigned long flags;
1484
Ming Lin2e39e0f2016-04-05 10:32:04 -07001485 /*
1486 * We might be called with the AQ q_lock held
1487 * and the I/O queue q_lock should always
1488 * nest inside the AQ one.
1489 */
1490 spin_lock_irqsave_nested(&nvmeq->q_lock, flags,
1491 SINGLE_DEPTH_NESTING);
Keith Buschdb3cbff2016-01-12 14:41:17 -07001492 nvme_process_cq(nvmeq);
1493 spin_unlock_irqrestore(&nvmeq->q_lock, flags);
1494 }
1495
1496 nvme_del_queue_end(req, error);
1497}
1498
1499static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode)
1500{
1501 struct request_queue *q = nvmeq->dev->ctrl.admin_q;
1502 struct request *req;
1503 struct nvme_command cmd;
1504
1505 memset(&cmd, 0, sizeof(cmd));
1506 cmd.delete_queue.opcode = opcode;
1507 cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid);
1508
Christoph Hellwigeb71f432016-06-13 16:45:23 +02001509 req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY);
Keith Buschdb3cbff2016-01-12 14:41:17 -07001510 if (IS_ERR(req))
1511 return PTR_ERR(req);
1512
1513 req->timeout = ADMIN_TIMEOUT;
1514 req->end_io_data = nvmeq;
1515
1516 blk_execute_rq_nowait(q, NULL, req, false,
1517 opcode == nvme_admin_delete_cq ?
1518 nvme_del_cq_end : nvme_del_queue_end);
1519 return 0;
1520}
1521
Keith Busch70659062016-10-12 09:22:16 -06001522static void nvme_disable_io_queues(struct nvme_dev *dev, int queues)
Keith Buschdb3cbff2016-01-12 14:41:17 -07001523{
Keith Busch70659062016-10-12 09:22:16 -06001524 int pass;
Keith Buschdb3cbff2016-01-12 14:41:17 -07001525 unsigned long timeout;
1526 u8 opcode = nvme_admin_delete_sq;
1527
1528 for (pass = 0; pass < 2; pass++) {
Keith Busch014a0d62016-05-06 11:50:52 -06001529 int sent = 0, i = queues;
Keith Buschdb3cbff2016-01-12 14:41:17 -07001530
1531 reinit_completion(&dev->ioq_wait);
1532 retry:
1533 timeout = ADMIN_TIMEOUT;
Gabriel Krisman Bertazic21377f2016-08-11 09:35:57 -06001534 for (; i > 0; i--, sent++)
1535 if (nvme_delete_queue(dev->queues[i], opcode))
Keith Buschdb3cbff2016-01-12 14:41:17 -07001536 break;
Gabriel Krisman Bertazic21377f2016-08-11 09:35:57 -06001537
Keith Buschdb3cbff2016-01-12 14:41:17 -07001538 while (sent--) {
1539 timeout = wait_for_completion_io_timeout(&dev->ioq_wait, timeout);
1540 if (timeout == 0)
1541 return;
1542 if (i)
1543 goto retry;
1544 }
1545 opcode = nvme_admin_delete_cq;
1546 }
1547}
1548
Matthew Wilcox422ef0c2013-04-16 11:22:36 -04001549/*
1550 * Return: error value if an error occurred setting up the queues or calling
1551 * Identify Device. 0 if these succeeded, even if adding some of the
1552 * namespaces failed. At the moment, these failures are silent. TBD which
1553 * failures should be reported.
1554 */
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001555static int nvme_dev_add(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001556{
Christoph Hellwig5bae7f72015-11-28 15:39:07 +01001557 if (!dev->ctrl.tagset) {
Keith Buschffe77042015-06-08 10:08:15 -06001558 dev->tagset.ops = &nvme_mq_ops;
1559 dev->tagset.nr_hw_queues = dev->online_queues - 1;
1560 dev->tagset.timeout = NVME_IO_TIMEOUT;
1561 dev->tagset.numa_node = dev_to_node(dev->dev);
1562 dev->tagset.queue_depth =
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001563 min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
Keith Buschffe77042015-06-08 10:08:15 -06001564 dev->tagset.cmd_size = nvme_cmd_size(dev);
1565 dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
1566 dev->tagset.driver_data = dev;
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001567
Keith Buschffe77042015-06-08 10:08:15 -06001568 if (blk_mq_alloc_tag_set(&dev->tagset))
1569 return 0;
Christoph Hellwig5bae7f72015-11-28 15:39:07 +01001570 dev->ctrl.tagset = &dev->tagset;
Keith Busch949928c2015-12-17 17:08:15 -07001571 } else {
1572 blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1);
1573
1574 /* Free previously allocated queues that are no longer usable */
1575 nvme_free_queues(dev, dev->online_queues);
Keith Buschffe77042015-06-08 10:08:15 -06001576 }
Keith Busch949928c2015-12-17 17:08:15 -07001577
Keith Busche1e5e562015-02-19 13:39:03 -07001578 return 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001579}
1580
Keith Buschb00a7262016-02-24 09:15:52 -07001581static int nvme_pci_enable(struct nvme_dev *dev)
Keith Busch0877cb02013-07-15 15:02:19 -06001582{
Keith Busch42f61422014-03-24 10:46:25 -06001583 u64 cap;
Keith Buschb00a7262016-02-24 09:15:52 -07001584 int result = -ENOMEM;
Christoph Hellwige75ec752015-05-22 11:12:39 +02001585 struct pci_dev *pdev = to_pci_dev(dev->dev);
Keith Busch0877cb02013-07-15 15:02:19 -06001586
1587 if (pci_enable_device_mem(pdev))
1588 return result;
1589
Keith Busch0877cb02013-07-15 15:02:19 -06001590 pci_set_master(pdev);
Keith Busch0877cb02013-07-15 15:02:19 -06001591
Christoph Hellwige75ec752015-05-22 11:12:39 +02001592 if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) &&
1593 dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32)))
Russell King052d0ef2013-06-26 23:49:11 +01001594 goto disable;
Keith Busch0877cb02013-07-15 15:02:19 -06001595
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001596 if (readl(dev->bar + NVME_REG_CSTS) == -1) {
Keith Busch0e53d182013-12-10 13:10:39 -07001597 result = -ENODEV;
Keith Buschb00a7262016-02-24 09:15:52 -07001598 goto disable;
Keith Busch0e53d182013-12-10 13:10:39 -07001599 }
Jens Axboee32efbf2014-11-14 09:49:26 -07001600
1601 /*
Keith Buscha5229052016-04-08 16:09:10 -06001602 * Some devices and/or platforms don't advertise or work with INTx
1603 * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll
1604 * adjust this later.
Jens Axboee32efbf2014-11-14 09:49:26 -07001605 */
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001606 result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
1607 if (result < 0)
1608 return result;
Jens Axboee32efbf2014-11-14 09:49:26 -07001609
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001610 cap = lo_hi_readq(dev->bar + NVME_REG_CAP);
1611
Keith Busch42f61422014-03-24 10:46:25 -06001612 dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
1613 dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001614 dev->dbs = dev->bar + 4096;
Stephan Günther1f390c12015-12-01 13:23:22 -07001615
1616 /*
1617 * Temporary fix for the Apple controller found in the MacBook8,1 and
1618 * some MacBook7,1 to avoid controller resets and data loss.
1619 */
1620 if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) {
1621 dev->q_depth = 2;
1622 dev_warn(dev->dev, "detected Apple NVMe controller, set "
1623 "queue depth=%u to work around controller resets\n",
1624 dev->q_depth);
1625 }
1626
Stephen Bates202021c2016-10-05 20:01:12 -06001627 /*
1628 * CMBs can currently only exist on >=1.2 PCIe devices. We only
1629 * populate sysfs if a CMB is implemented. Note that we add the
1630 * CMB attribute to the nvme_ctrl kobj which removes the need to remove
1631 * it on exit. Since nvme_dev_attrs_group has no name we can pass
1632 * NULL as final argument to sysfs_add_file_to_group.
1633 */
1634
Gabriel Krisman Bertazi8ef20742016-10-19 09:51:05 -06001635 if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) {
Jon Derrick8ffaadf2015-07-20 10:14:09 -06001636 dev->cmb = nvme_map_cmb(dev);
Keith Busch0877cb02013-07-15 15:02:19 -06001637
Stephen Bates202021c2016-10-05 20:01:12 -06001638 if (dev->cmbsz) {
1639 if (sysfs_add_file_to_group(&dev->ctrl.device->kobj,
1640 &dev_attr_cmb.attr, NULL))
1641 dev_warn(dev->dev,
1642 "failed to add sysfs attribute for CMB\n");
1643 }
1644 }
1645
Keith Buscha0a34082015-12-07 15:30:31 -07001646 pci_enable_pcie_error_reporting(pdev);
1647 pci_save_state(pdev);
Keith Busch0877cb02013-07-15 15:02:19 -06001648 return 0;
1649
1650 disable:
Keith Busch0877cb02013-07-15 15:02:19 -06001651 pci_disable_device(pdev);
1652 return result;
1653}
1654
1655static void nvme_dev_unmap(struct nvme_dev *dev)
1656{
Keith Buschb00a7262016-02-24 09:15:52 -07001657 if (dev->bar)
1658 iounmap(dev->bar);
Johannes Thumshirna1f447b2016-06-07 09:44:02 +02001659 pci_release_mem_regions(to_pci_dev(dev->dev));
Keith Buschb00a7262016-02-24 09:15:52 -07001660}
1661
1662static void nvme_pci_disable(struct nvme_dev *dev)
1663{
Christoph Hellwige75ec752015-05-22 11:12:39 +02001664 struct pci_dev *pdev = to_pci_dev(dev->dev);
1665
Jon Derrick6d6a43a2017-05-05 14:52:06 -06001666 nvme_release_cmb(dev);
Christoph Hellwigdca51e72016-09-14 16:18:57 +02001667 pci_free_irq_vectors(pdev);
Keith Busch0877cb02013-07-15 15:02:19 -06001668
Keith Buscha0a34082015-12-07 15:30:31 -07001669 if (pci_is_enabled(pdev)) {
1670 pci_disable_pcie_error_reporting(pdev);
Christoph Hellwige75ec752015-05-22 11:12:39 +02001671 pci_disable_device(pdev);
Keith Busch4d115422013-12-10 13:10:40 -07001672 }
Keith Busch4d115422013-12-10 13:10:40 -07001673}
1674
Keith Buscha5cdb682016-01-12 14:41:18 -07001675static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001676{
Keith Busch70659062016-10-12 09:22:16 -06001677 int i, queues;
Keith Busch7c1b2452014-06-25 11:18:12 -06001678 u32 csts = -1;
Keith Busch22404272013-07-15 15:02:20 -06001679
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001680 del_timer_sync(&dev->watchdog_timer);
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001681
Keith Busch77bf25e2015-11-26 12:21:29 +01001682 mutex_lock(&dev->shutdown_lock);
Keith Buschb00a7262016-02-24 09:15:52 -07001683 if (pci_is_enabled(to_pci_dev(dev->dev))) {
Keith Busch25646262016-01-04 09:10:57 -07001684 nvme_stop_queues(&dev->ctrl);
Christoph Hellwig7a67cbe2015-11-20 08:58:10 +01001685 csts = readl(dev->bar + NVME_REG_CSTS);
Keith Buschc9d3bf82015-01-07 18:55:52 -07001686 }
Gabriel Krisman Bertazic21377f2016-08-11 09:35:57 -06001687
Keith Busch70659062016-10-12 09:22:16 -06001688 queues = dev->online_queues - 1;
Gabriel Krisman Bertazic21377f2016-08-11 09:35:57 -06001689 for (i = dev->queue_count - 1; i > 0; i--)
1690 nvme_suspend_queue(dev->queues[i]);
1691
Keith Busch7c1b2452014-06-25 11:18:12 -06001692 if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) {
Gabriel Krisman Bertazi82469c52016-09-06 17:39:13 -03001693 /* A device might become IO incapable very soon during
1694 * probe, before the admin queue is configured. Thus,
1695 * queue_count can be 0 here.
1696 */
1697 if (dev->queue_count)
1698 nvme_suspend_queue(dev->queues[0]);
Keith Busch4d115422013-12-10 13:10:40 -07001699 } else {
Keith Busch70659062016-10-12 09:22:16 -06001700 nvme_disable_io_queues(dev, queues);
Keith Buscha5cdb682016-01-12 14:41:18 -07001701 nvme_disable_admin_queue(dev, shutdown);
Keith Busch4d115422013-12-10 13:10:40 -07001702 }
Keith Buschb00a7262016-02-24 09:15:52 -07001703 nvme_pci_disable(dev);
Keith Busch07836e62015-02-19 10:34:48 -07001704
Ming Line1958e62016-05-18 14:05:01 -07001705 blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl);
1706 blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl);
Keith Busch77bf25e2015-11-26 12:21:29 +01001707 mutex_unlock(&dev->shutdown_lock);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001708}
1709
Matthew Wilcox091b6092011-02-10 09:56:01 -05001710static int nvme_setup_prp_pools(struct nvme_dev *dev)
1711{
Christoph Hellwige75ec752015-05-22 11:12:39 +02001712 dev->prp_page_pool = dma_pool_create("prp list page", dev->dev,
Matthew Wilcox091b6092011-02-10 09:56:01 -05001713 PAGE_SIZE, PAGE_SIZE, 0);
1714 if (!dev->prp_page_pool)
1715 return -ENOMEM;
1716
Matthew Wilcox99802a72011-02-10 10:30:34 -05001717 /* Optimisation for I/Os between 4k and 128k */
Christoph Hellwige75ec752015-05-22 11:12:39 +02001718 dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev,
Matthew Wilcox99802a72011-02-10 10:30:34 -05001719 256, 256, 0);
1720 if (!dev->prp_small_pool) {
1721 dma_pool_destroy(dev->prp_page_pool);
1722 return -ENOMEM;
1723 }
Matthew Wilcox091b6092011-02-10 09:56:01 -05001724 return 0;
1725}
1726
1727static void nvme_release_prp_pools(struct nvme_dev *dev)
1728{
1729 dma_pool_destroy(dev->prp_page_pool);
Matthew Wilcox99802a72011-02-10 10:30:34 -05001730 dma_pool_destroy(dev->prp_small_pool);
Matthew Wilcox091b6092011-02-10 09:56:01 -05001731}
1732
Christoph Hellwig1673f1f2015-11-26 10:54:19 +01001733static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001734{
Christoph Hellwig1673f1f2015-11-26 10:54:19 +01001735 struct nvme_dev *dev = to_nvme_dev(ctrl);
Keith Busch9ac27092014-01-31 16:53:39 -07001736
Christoph Hellwige75ec752015-05-22 11:12:39 +02001737 put_device(dev->dev);
Keith Busch4af0e212015-06-08 10:08:13 -06001738 if (dev->tagset.tags)
1739 blk_mq_free_tag_set(&dev->tagset);
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001740 if (dev->ctrl.admin_q)
1741 blk_put_queue(dev->ctrl.admin_q);
Keith Busch5e82e952013-02-19 10:17:58 -07001742 kfree(dev->queues);
Keith Busch5e82e952013-02-19 10:17:58 -07001743 kfree(dev);
1744}
1745
Keith Buschf58944e2016-02-24 09:15:55 -07001746static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status)
1747{
Linus Torvalds237045f2016-03-18 17:13:31 -07001748 dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status);
Keith Buschf58944e2016-02-24 09:15:55 -07001749
1750 kref_get(&dev->ctrl.kref);
Keith Busch69d9a992016-02-24 09:15:56 -07001751 nvme_dev_disable(dev, false);
Keith Buschf58944e2016-02-24 09:15:55 -07001752 if (!schedule_work(&dev->remove_work))
1753 nvme_put_ctrl(&dev->ctrl);
1754}
1755
Christoph Hellwigfd634f412015-11-26 12:42:26 +01001756static void nvme_reset_work(struct work_struct *work)
Keith Busch5e82e952013-02-19 10:17:58 -07001757{
Christoph Hellwigfd634f412015-11-26 12:42:26 +01001758 struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work);
Keith Buschf58944e2016-02-24 09:15:55 -07001759 int result = -ENODEV;
Keith Buschf0b50732013-07-15 15:02:21 -06001760
Rakesh Pandit383f15e2017-06-05 14:43:11 +03001761 if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING))
Christoph Hellwigfd634f412015-11-26 12:42:26 +01001762 goto out;
1763
1764 /*
1765 * If we're called to reset a live controller first shut it down before
1766 * moving on.
1767 */
Keith Buschb00a7262016-02-24 09:15:52 -07001768 if (dev->ctrl.ctrl_config & NVME_CC_ENABLE)
Keith Buscha5cdb682016-01-12 14:41:18 -07001769 nvme_dev_disable(dev, false);
Christoph Hellwigfd634f412015-11-26 12:42:26 +01001770
Keith Buschb00a7262016-02-24 09:15:52 -07001771 result = nvme_pci_enable(dev);
Keith Buschf0b50732013-07-15 15:02:21 -06001772 if (result)
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001773 goto out;
Keith Buschf0b50732013-07-15 15:02:21 -06001774
1775 result = nvme_configure_admin_queue(dev);
1776 if (result)
Keith Buschf58944e2016-02-24 09:15:55 -07001777 goto out;
Keith Buschf0b50732013-07-15 15:02:21 -06001778
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001779 nvme_init_queue(dev->queues[0], 0);
Keith Busch0fb59cb2015-01-07 18:55:50 -07001780 result = nvme_alloc_admin_tags(dev);
1781 if (result)
Keith Buschf58944e2016-02-24 09:15:55 -07001782 goto out;
Dan McLeranb9afca32014-04-07 17:10:11 -06001783
Christoph Hellwigce4541f2015-10-16 07:58:46 +02001784 result = nvme_init_identify(&dev->ctrl);
1785 if (result)
Keith Buschf58944e2016-02-24 09:15:55 -07001786 goto out;
Christoph Hellwigce4541f2015-10-16 07:58:46 +02001787
Keith Buschf0b50732013-07-15 15:02:21 -06001788 result = nvme_setup_io_queues(dev);
Keith Buschbadc34d2014-06-23 14:25:35 -06001789 if (result)
Keith Buschf58944e2016-02-24 09:15:55 -07001790 goto out;
Keith Buschf0b50732013-07-15 15:02:21 -06001791
Keith Busch21f033f2016-04-12 11:13:11 -06001792 /*
1793 * A controller that can not execute IO typically requires user
1794 * intervention to correct. For such degraded controllers, the driver
1795 * should not submit commands the user did not request, so skip
1796 * registering for asynchronous event notification on this condition.
1797 */
Christoph Hellwigf866fc42016-04-26 13:52:00 +02001798 if (dev->online_queues > 1)
1799 nvme_queue_async_events(&dev->ctrl);
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001800
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001801 mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001802
Christoph Hellwig2659e572015-10-02 18:51:31 +02001803 /*
1804 * Keep the controller around but remove all namespaces if we don't have
1805 * any working I/O queue.
1806 */
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001807 if (dev->online_queues < 2) {
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -07001808 dev_warn(dev->ctrl.device, "IO queues not created\n");
Keith Busch3b247742016-04-27 15:51:18 -06001809 nvme_kill_queues(&dev->ctrl);
Christoph Hellwig5bae7f72015-11-28 15:39:07 +01001810 nvme_remove_namespaces(&dev->ctrl);
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001811 } else {
Keith Busch25646262016-01-04 09:10:57 -07001812 nvme_start_queues(&dev->ctrl);
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001813 nvme_dev_add(dev);
1814 }
1815
Christoph Hellwigbb8d2612016-04-26 13:51:57 +02001816 if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) {
1817 dev_warn(dev->ctrl.device, "failed to mark controller live\n");
1818 goto out;
1819 }
Christoph Hellwig92911a52016-04-26 13:51:58 +02001820
1821 if (dev->online_queues > 1)
Christoph Hellwig5955be22016-04-26 13:51:59 +02001822 nvme_queue_scan(&dev->ctrl);
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001823 return;
Keith Buschf0b50732013-07-15 15:02:21 -06001824
Christoph Hellwig3cf519b2015-10-03 09:49:23 +02001825 out:
Keith Buschf58944e2016-02-24 09:15:55 -07001826 nvme_remove_dead_ctrl(dev, result);
Keith Buschf0b50732013-07-15 15:02:21 -06001827}
1828
Christoph Hellwig5c8809e2015-11-26 12:35:49 +01001829static void nvme_remove_dead_ctrl_work(struct work_struct *work)
Keith Busch9a6b9452013-12-10 13:10:36 -07001830{
Christoph Hellwig5c8809e2015-11-26 12:35:49 +01001831 struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
Christoph Hellwige75ec752015-05-22 11:12:39 +02001832 struct pci_dev *pdev = to_pci_dev(dev->dev);
Keith Busch9a6b9452013-12-10 13:10:36 -07001833
Keith Busch69d9a992016-02-24 09:15:56 -07001834 nvme_kill_queues(&dev->ctrl);
Keith Busch9a6b9452013-12-10 13:10:36 -07001835 if (pci_get_drvdata(pdev))
Keith Busch921920a2016-03-28 16:03:21 -06001836 device_release_driver(&pdev->dev);
Christoph Hellwig1673f1f2015-11-26 10:54:19 +01001837 nvme_put_ctrl(&dev->ctrl);
Keith Busch9a6b9452013-12-10 13:10:36 -07001838}
1839
Keith Busch4cc06522015-06-05 10:30:08 -06001840static int nvme_reset(struct nvme_dev *dev)
1841{
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001842 if (!dev->ctrl.admin_q || blk_queue_dying(dev->ctrl.admin_q))
Keith Busch4cc06522015-06-05 10:30:08 -06001843 return -ENODEV;
Rakesh Pandit383f15e2017-06-05 14:43:11 +03001844 if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING))
1845 return -EBUSY;
Christoph Hellwig846cc052015-11-26 12:10:29 +01001846 if (!queue_work(nvme_workq, &dev->reset_work))
1847 return -EBUSY;
Christoph Hellwig846cc052015-11-26 12:10:29 +01001848 return 0;
Keith Busch4cc06522015-06-05 10:30:08 -06001849}
1850
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001851static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
Keith Busch4cc06522015-06-05 10:30:08 -06001852{
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001853 *val = readl(to_nvme_dev(ctrl)->bar + off);
1854 return 0;
Keith Busch4cc06522015-06-05 10:30:08 -06001855}
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001856
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001857static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
1858{
1859 writel(val, to_nvme_dev(ctrl)->bar + off);
1860 return 0;
1861}
1862
Christoph Hellwig7fd89302015-11-28 15:37:52 +01001863static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
1864{
1865 *val = readq(to_nvme_dev(ctrl)->bar + off);
1866 return 0;
1867}
1868
Christoph Hellwigf3ca80f2015-11-28 15:40:19 +01001869static int nvme_pci_reset_ctrl(struct nvme_ctrl *ctrl)
1870{
Keith Buschc5f6ce92016-10-05 16:32:45 -04001871 struct nvme_dev *dev = to_nvme_dev(ctrl);
1872 int ret = nvme_reset(dev);
1873
1874 if (!ret)
1875 flush_work(&dev->reset_work);
1876 return ret;
Christoph Hellwigf3ca80f2015-11-28 15:40:19 +01001877}
1878
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001879static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = {
Ming Lin1a353d82016-06-13 16:45:24 +02001880 .name = "pcie",
Sagi Grimberge439bb12016-02-10 10:03:29 -08001881 .module = THIS_MODULE,
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001882 .reg_read32 = nvme_pci_reg_read32,
Christoph Hellwig5fd4ce12015-11-28 15:03:49 +01001883 .reg_write32 = nvme_pci_reg_write32,
Christoph Hellwig7fd89302015-11-28 15:37:52 +01001884 .reg_read64 = nvme_pci_reg_read64,
Christoph Hellwigf3ca80f2015-11-28 15:40:19 +01001885 .reset_ctrl = nvme_pci_reset_ctrl,
Christoph Hellwig1673f1f2015-11-26 10:54:19 +01001886 .free_ctrl = nvme_pci_free_ctrl,
Christoph Hellwigf866fc42016-04-26 13:52:00 +02001887 .submit_async_event = nvme_pci_submit_async_event,
Christoph Hellwig1c63dc62015-11-26 10:06:56 +01001888};
Keith Busch4cc06522015-06-05 10:30:08 -06001889
Keith Buschb00a7262016-02-24 09:15:52 -07001890static int nvme_dev_map(struct nvme_dev *dev)
1891{
Keith Buschb00a7262016-02-24 09:15:52 -07001892 struct pci_dev *pdev = to_pci_dev(dev->dev);
1893
Johannes Thumshirna1f447b2016-06-07 09:44:02 +02001894 if (pci_request_mem_regions(pdev, "nvme"))
Keith Buschb00a7262016-02-24 09:15:52 -07001895 return -ENODEV;
1896
1897 dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
1898 if (!dev->bar)
1899 goto release;
1900
1901 return 0;
1902 release:
Johannes Thumshirna1f447b2016-06-07 09:44:02 +02001903 pci_release_mem_regions(pdev);
Keith Buschb00a7262016-02-24 09:15:52 -07001904 return -ENODEV;
1905}
1906
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001907static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001908{
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001909 int node, result = -ENOMEM;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001910 struct nvme_dev *dev;
1911
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001912 node = dev_to_node(&pdev->dev);
1913 if (node == NUMA_NO_NODE)
Masayoshi Mizuma2fa84352016-06-20 09:33:17 +09001914 set_dev_node(&pdev->dev, first_memory_node);
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001915
1916 dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001917 if (!dev)
1918 return -ENOMEM;
Matias Bjørlinga4aea562014-11-04 08:20:14 -07001919 dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *),
1920 GFP_KERNEL, node);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001921 if (!dev->queues)
1922 goto free;
1923
Christoph Hellwige75ec752015-05-22 11:12:39 +02001924 dev->dev = get_device(&pdev->dev);
Keith Busch9a6b9452013-12-10 13:10:36 -07001925 pci_set_drvdata(pdev, dev);
Keith Buschb3fffde2015-02-03 11:21:42 -07001926
Keith Buschb00a7262016-02-24 09:15:52 -07001927 result = nvme_dev_map(dev);
1928 if (result)
1929 goto free;
1930
Christoph Hellwigf3ca80f2015-11-28 15:40:19 +01001931 INIT_WORK(&dev->reset_work, nvme_reset_work);
Christoph Hellwig5c8809e2015-11-26 12:35:49 +01001932 INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
Christoph Hellwig2d55cd52016-02-29 15:59:46 +01001933 setup_timer(&dev->watchdog_timer, nvme_watchdog_timer,
1934 (unsigned long)dev);
Keith Busch77bf25e2015-11-26 12:21:29 +01001935 mutex_init(&dev->shutdown_lock);
Keith Buschdb3cbff2016-01-12 14:41:17 -07001936 init_completion(&dev->ioq_wait);
Christoph Hellwigf3ca80f2015-11-28 15:40:19 +01001937
1938 result = nvme_setup_prp_pools(dev);
1939 if (result)
1940 goto put_pci;
1941
1942 result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
1943 id->driver_data);
1944 if (result)
1945 goto release_pools;
1946
Rakesh Pandit383f15e2017-06-05 14:43:11 +03001947 nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING);
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -07001948 dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));
1949
Keith Busch92f7a162015-10-23 11:42:02 -06001950 queue_work(nvme_workq, &dev->reset_work);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001951 return 0;
1952
Keith Busch0877cb02013-07-15 15:02:19 -06001953 release_pools:
Matthew Wilcox091b6092011-02-10 09:56:01 -05001954 nvme_release_prp_pools(dev);
Keith Buscha96d4f52014-08-19 19:15:59 -06001955 put_pci:
Christoph Hellwige75ec752015-05-22 11:12:39 +02001956 put_device(dev->dev);
Keith Buschb00a7262016-02-24 09:15:52 -07001957 nvme_dev_unmap(dev);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001958 free:
1959 kfree(dev->queues);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001960 kfree(dev);
1961 return result;
1962}
1963
Keith Buschf0d54a52014-05-02 10:40:43 -06001964static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
1965{
Keith Buscha6739472014-06-23 16:03:21 -06001966 struct nvme_dev *dev = pci_get_drvdata(pdev);
Keith Buschf0d54a52014-05-02 10:40:43 -06001967
Keith Buscha6739472014-06-23 16:03:21 -06001968 if (prepare)
Keith Buscha5cdb682016-01-12 14:41:18 -07001969 nvme_dev_disable(dev, false);
Keith Buscha6739472014-06-23 16:03:21 -06001970 else
Keith Buschc5f6ce92016-10-05 16:32:45 -04001971 nvme_reset(dev);
Keith Buschf0d54a52014-05-02 10:40:43 -06001972}
1973
Keith Busch09ece142014-01-27 11:29:40 -05001974static void nvme_shutdown(struct pci_dev *pdev)
1975{
1976 struct nvme_dev *dev = pci_get_drvdata(pdev);
Keith Buscha5cdb682016-01-12 14:41:18 -07001977 nvme_dev_disable(dev, true);
Keith Busch09ece142014-01-27 11:29:40 -05001978}
1979
Keith Buschf58944e2016-02-24 09:15:55 -07001980/*
1981 * The driver's remove may be called on a device in a partially initialized
1982 * state. This function must not have any dependencies on the device state in
1983 * order to proceed.
1984 */
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001985static void nvme_remove(struct pci_dev *pdev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001986{
1987 struct nvme_dev *dev = pci_get_drvdata(pdev);
Keith Busch9a6b9452013-12-10 13:10:36 -07001988
Christoph Hellwigbb8d2612016-04-26 13:51:57 +02001989 nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING);
1990
Rakesh Pandit383f15e2017-06-05 14:43:11 +03001991 cancel_work_sync(&dev->reset_work);
Keith Busch9a6b9452013-12-10 13:10:36 -07001992 pci_set_drvdata(pdev, NULL);
Keith Busch0ff9d4e2016-05-12 08:37:14 -06001993
Keith Busch02b23e02017-02-10 18:15:49 -05001994 if (!pci_device_is_present(pdev)) {
Keith Busch0ff9d4e2016-05-12 08:37:14 -06001995 nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
Keith Busch02b23e02017-02-10 18:15:49 -05001996 nvme_dev_disable(dev, false);
1997 }
Keith Busch0ff9d4e2016-05-12 08:37:14 -06001998
Keith Busch9bf2b972016-04-08 16:11:02 -06001999 flush_work(&dev->reset_work);
Keith Busch53029b02015-11-28 15:41:02 +01002000 nvme_uninit_ctrl(&dev->ctrl);
Keith Buscha5cdb682016-01-12 14:41:18 -07002001 nvme_dev_disable(dev, true);
Matias Bjørlinga4aea562014-11-04 08:20:14 -07002002 nvme_dev_remove_admin(dev);
2003 nvme_free_queues(dev, 0);
Keith Busch9a6b9452013-12-10 13:10:36 -07002004 nvme_release_prp_pools(dev);
Keith Buschb00a7262016-02-24 09:15:52 -07002005 nvme_dev_unmap(dev);
Christoph Hellwig1673f1f2015-11-26 10:54:19 +01002006 nvme_put_ctrl(&dev->ctrl);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002007}
2008
Keith Busch13880f52016-06-20 09:41:06 -06002009static int nvme_pci_sriov_configure(struct pci_dev *pdev, int numvfs)
2010{
2011 int ret = 0;
2012
2013 if (numvfs == 0) {
2014 if (pci_vfs_assigned(pdev)) {
2015 dev_warn(&pdev->dev,
2016 "Cannot disable SR-IOV VFs while assigned\n");
2017 return -EPERM;
2018 }
2019 pci_disable_sriov(pdev);
2020 return 0;
2021 }
2022
2023 ret = pci_enable_sriov(pdev, numvfs);
2024 return ret ? ret : numvfs;
2025}
2026
Jingoo Han671a6012014-02-13 11:19:14 +09002027#ifdef CONFIG_PM_SLEEP
Keith Buschcd638942013-07-15 15:02:23 -06002028static int nvme_suspend(struct device *dev)
2029{
2030 struct pci_dev *pdev = to_pci_dev(dev);
2031 struct nvme_dev *ndev = pci_get_drvdata(pdev);
2032
Keith Buscha5cdb682016-01-12 14:41:18 -07002033 nvme_dev_disable(ndev, true);
Keith Buschcd638942013-07-15 15:02:23 -06002034 return 0;
2035}
2036
2037static int nvme_resume(struct device *dev)
2038{
2039 struct pci_dev *pdev = to_pci_dev(dev);
2040 struct nvme_dev *ndev = pci_get_drvdata(pdev);
Keith Buschcd638942013-07-15 15:02:23 -06002041
Keith Buschc5f6ce92016-10-05 16:32:45 -04002042 nvme_reset(ndev);
Keith Busch9a6b9452013-12-10 13:10:36 -07002043 return 0;
Keith Buschcd638942013-07-15 15:02:23 -06002044}
Jingoo Han671a6012014-02-13 11:19:14 +09002045#endif
Keith Buschcd638942013-07-15 15:02:23 -06002046
2047static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002048
Keith Buscha0a34082015-12-07 15:30:31 -07002049static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev,
2050 pci_channel_state_t state)
2051{
2052 struct nvme_dev *dev = pci_get_drvdata(pdev);
2053
2054 /*
2055 * A frozen channel requires a reset. When detected, this method will
2056 * shutdown the controller to quiesce. The controller will be restarted
2057 * after the slot reset through driver's slot_reset callback.
2058 */
Keith Buscha0a34082015-12-07 15:30:31 -07002059 switch (state) {
2060 case pci_channel_io_normal:
2061 return PCI_ERS_RESULT_CAN_RECOVER;
2062 case pci_channel_io_frozen:
Keith Buschd011fb32016-04-04 15:07:41 -06002063 dev_warn(dev->ctrl.device,
2064 "frozen state error detected, reset controller\n");
Keith Buscha5cdb682016-01-12 14:41:18 -07002065 nvme_dev_disable(dev, false);
Keith Buscha0a34082015-12-07 15:30:31 -07002066 return PCI_ERS_RESULT_NEED_RESET;
2067 case pci_channel_io_perm_failure:
Keith Buschd011fb32016-04-04 15:07:41 -06002068 dev_warn(dev->ctrl.device,
2069 "failure state error detected, request disconnect\n");
Keith Buscha0a34082015-12-07 15:30:31 -07002070 return PCI_ERS_RESULT_DISCONNECT;
2071 }
2072 return PCI_ERS_RESULT_NEED_RESET;
2073}
2074
2075static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev)
2076{
2077 struct nvme_dev *dev = pci_get_drvdata(pdev);
2078
Sagi Grimberg1b3c47c2016-02-10 08:51:15 -07002079 dev_info(dev->ctrl.device, "restart after slot reset\n");
Keith Buscha0a34082015-12-07 15:30:31 -07002080 pci_restore_state(pdev);
Keith Buschc5f6ce92016-10-05 16:32:45 -04002081 nvme_reset(dev);
Keith Buscha0a34082015-12-07 15:30:31 -07002082 return PCI_ERS_RESULT_RECOVERED;
2083}
2084
2085static void nvme_error_resume(struct pci_dev *pdev)
2086{
2087 pci_cleanup_aer_uncorrect_error_status(pdev);
2088}
2089
Stephen Hemminger1d352032012-09-07 09:33:17 -07002090static const struct pci_error_handlers nvme_err_handler = {
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002091 .error_detected = nvme_error_detected,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002092 .slot_reset = nvme_slot_reset,
2093 .resume = nvme_error_resume,
Keith Buschf0d54a52014-05-02 10:40:43 -06002094 .reset_notify = nvme_reset_notify,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002095};
2096
2097/* Move to pci_ids.h later */
2098#define PCI_CLASS_STORAGE_EXPRESS 0x010802
2099
Matthew Wilcox6eb0d692014-03-24 10:11:22 -04002100static const struct pci_device_id nvme_id_table[] = {
Christoph Hellwig106198e2015-11-26 10:07:41 +01002101 { PCI_VDEVICE(INTEL, 0x0953),
Keith Busch08095e72016-03-04 13:15:17 -07002102 .driver_data = NVME_QUIRK_STRIPE_SIZE |
2103 NVME_QUIRK_DISCARD_ZEROES, },
Keith Busch99466e72016-05-02 15:14:24 -06002104 { PCI_VDEVICE(INTEL, 0x0a53),
2105 .driver_data = NVME_QUIRK_STRIPE_SIZE |
2106 NVME_QUIRK_DISCARD_ZEROES, },
2107 { PCI_VDEVICE(INTEL, 0x0a54),
2108 .driver_data = NVME_QUIRK_STRIPE_SIZE |
2109 NVME_QUIRK_DISCARD_ZEROES, },
Keith Busch540c8012015-10-22 15:45:06 -06002110 { PCI_VDEVICE(INTEL, 0x5845), /* Qemu emulated controller */
2111 .driver_data = NVME_QUIRK_IDENTIFY_CNS, },
Guilherme G. Piccoli54adc012016-06-14 18:22:41 -03002112 { PCI_DEVICE(0x1c58, 0x0003), /* HGST adapter */
2113 .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
Jeff Lienf425b052017-11-21 10:44:37 -06002114 { PCI_DEVICE(0x1c58, 0x0023), /* WDC SN200 adapter */
2115 .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
Wenbo Wang015282c2016-09-08 12:12:11 -04002116 { PCI_DEVICE(0x1c5f, 0x0540), /* Memblaze Pblaze4 adapter */
2117 .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, },
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002118 { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
Stephan Güntherc74dc782015-11-04 00:49:45 +01002119 { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) },
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002120 { 0, }
2121};
2122MODULE_DEVICE_TABLE(pci, nvme_id_table);
2123
2124static struct pci_driver nvme_driver = {
2125 .name = "nvme",
2126 .id_table = nvme_id_table,
2127 .probe = nvme_probe,
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08002128 .remove = nvme_remove,
Keith Busch09ece142014-01-27 11:29:40 -05002129 .shutdown = nvme_shutdown,
Keith Buschcd638942013-07-15 15:02:23 -06002130 .driver = {
2131 .pm = &nvme_dev_pm_ops,
2132 },
Keith Busch13880f52016-06-20 09:41:06 -06002133 .sriov_configure = nvme_pci_sriov_configure,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002134 .err_handler = &nvme_err_handler,
2135};
2136
2137static int __init nvme_init(void)
2138{
Matthew Wilcox0ac13142012-07-31 13:31:15 -04002139 int result;
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05002140
Keith Busch92f7a162015-10-23 11:42:02 -06002141 nvme_workq = alloc_workqueue("nvme", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
Keith Busch9a6b9452013-12-10 13:10:36 -07002142 if (!nvme_workq)
Dan McLeranb9afca32014-04-07 17:10:11 -06002143 return -ENOMEM;
Keith Busch9a6b9452013-12-10 13:10:36 -07002144
Keith Buschf3db22f2014-06-11 11:51:35 -06002145 result = pci_register_driver(&nvme_driver);
2146 if (result)
Ming Lin576d55d2016-02-10 10:03:32 -08002147 destroy_workqueue(nvme_workq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002148 return result;
2149}
2150
2151static void __exit nvme_exit(void)
2152{
2153 pci_unregister_driver(&nvme_driver);
Keith Busch9a6b9452013-12-10 13:10:36 -07002154 destroy_workqueue(nvme_workq);
Matthew Wilcox21bd78b2014-05-09 22:42:26 -04002155 _nvme_check_size();
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002156}
2157
2158MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
2159MODULE_LICENSE("GPL");
Keith Buschc78b47132014-11-21 15:16:32 -07002160MODULE_VERSION("1.0");
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05002161module_init(nvme_init);
2162module_exit(nvme_exit);