Commit | Line | Data |
---|---|---|
5f37396d | 1 | // SPDX-License-Identifier: GPL-2.0 |
b60503ba MW |
2 | /* |
3 | * NVM Express device driver | |
6eb0d698 | 4 | * Copyright (c) 2011-2014, Intel Corporation. |
b60503ba MW |
5 | */ |
6 | ||
a0a3408e | 7 | #include <linux/aer.h> |
18119775 | 8 | #include <linux/async.h> |
b60503ba | 9 | #include <linux/blkdev.h> |
a4aea562 | 10 | #include <linux/blk-mq.h> |
dca51e78 | 11 | #include <linux/blk-mq-pci.h> |
ff5350a8 | 12 | #include <linux/dmi.h> |
b60503ba MW |
13 | #include <linux/init.h> |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/io.h> | |
b60503ba MW |
16 | #include <linux/mm.h> |
17 | #include <linux/module.h> | |
77bf25ea | 18 | #include <linux/mutex.h> |
d0877473 | 19 | #include <linux/once.h> |
b60503ba | 20 | #include <linux/pci.h> |
d916b1be | 21 | #include <linux/suspend.h> |
e1e5e564 | 22 | #include <linux/t10-pi.h> |
b60503ba | 23 | #include <linux/types.h> |
2f8e2c87 | 24 | #include <linux/io-64-nonatomic-lo-hi.h> |
a98e58e5 | 25 | #include <linux/sed-opal.h> |
0f238ff5 | 26 | #include <linux/pci-p2pdma.h> |
797a796a | 27 | |
604c01d5 | 28 | #include "trace.h" |
f11bb3e2 CH |
29 | #include "nvme.h" |
30 | ||
c1e0cc7e | 31 | #define SQ_SIZE(q) ((q)->q_depth << (q)->sqes) |
8a1d09a6 | 32 | #define CQ_SIZE(q) ((q)->q_depth * sizeof(struct nvme_completion)) |
c965809c | 33 | |
a7a7cbe3 | 34 | #define SGES_PER_PAGE (PAGE_SIZE / sizeof(struct nvme_sgl_desc)) |
9d43cf64 | 35 | |
943e942e JA |
36 | /* |
37 | * These can be higher, but we need to ensure that any command doesn't | |
38 | * require an sg allocation that needs more than a page of data. | |
39 | */ | |
40 | #define NVME_MAX_KB_SZ 4096 | |
41 | #define NVME_MAX_SEGS 127 | |
42 | ||
58ffacb5 MW |
43 | static int use_threaded_interrupts; |
44 | module_param(use_threaded_interrupts, int, 0); | |
45 | ||
8ffaadf7 | 46 | static bool use_cmb_sqes = true; |
69f4eb9f | 47 | module_param(use_cmb_sqes, bool, 0444); |
8ffaadf7 JD |
48 | MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes"); |
49 | ||
87ad72a5 CH |
50 | static unsigned int max_host_mem_size_mb = 128; |
51 | module_param(max_host_mem_size_mb, uint, 0444); | |
52 | MODULE_PARM_DESC(max_host_mem_size_mb, | |
53 | "Maximum Host Memory Buffer (HMB) size per controller (in MiB)"); | |
1fa6aead | 54 | |
a7a7cbe3 CK |
55 | static unsigned int sgl_threshold = SZ_32K; |
56 | module_param(sgl_threshold, uint, 0644); | |
57 | MODULE_PARM_DESC(sgl_threshold, | |
58 | "Use SGLs when average request segment size is larger or equal to " | |
59 | "this size. Use 0 to disable SGLs."); | |
60 | ||
b27c1e68 | 61 | static int io_queue_depth_set(const char *val, const struct kernel_param *kp); |
62 | static const struct kernel_param_ops io_queue_depth_ops = { | |
63 | .set = io_queue_depth_set, | |
61f3b896 | 64 | .get = param_get_uint, |
b27c1e68 | 65 | }; |
66 | ||
61f3b896 | 67 | static unsigned int io_queue_depth = 1024; |
b27c1e68 | 68 | module_param_cb(io_queue_depth, &io_queue_depth_ops, &io_queue_depth, 0644); |
69 | MODULE_PARM_DESC(io_queue_depth, "set io queue depth, should >= 2"); | |
70 | ||
9c9e76d5 WZ |
71 | static int io_queue_count_set(const char *val, const struct kernel_param *kp) |
72 | { | |
73 | unsigned int n; | |
74 | int ret; | |
75 | ||
76 | ret = kstrtouint(val, 10, &n); | |
77 | if (ret != 0 || n > num_possible_cpus()) | |
78 | return -EINVAL; | |
79 | return param_set_uint(val, kp); | |
80 | } | |
81 | ||
82 | static const struct kernel_param_ops io_queue_count_ops = { | |
83 | .set = io_queue_count_set, | |
84 | .get = param_get_uint, | |
85 | }; | |
86 | ||
3f68baf7 | 87 | static unsigned int write_queues; |
9c9e76d5 | 88 | module_param_cb(write_queues, &io_queue_count_ops, &write_queues, 0644); |
3b6592f7 JA |
89 | MODULE_PARM_DESC(write_queues, |
90 | "Number of queues to use for writes. If not set, reads and writes " | |
91 | "will share a queue set."); | |
92 | ||
3f68baf7 | 93 | static unsigned int poll_queues; |
9c9e76d5 | 94 | module_param_cb(poll_queues, &io_queue_count_ops, &poll_queues, 0644); |
4b04cc6a JA |
95 | MODULE_PARM_DESC(poll_queues, "Number of queues to use for polled IO."); |
96 | ||
1c63dc66 CH |
97 | struct nvme_dev; |
98 | struct nvme_queue; | |
b3fffdef | 99 | |
a5cdb68c | 100 | static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown); |
8fae268b | 101 | static bool __nvme_disable_io_queues(struct nvme_dev *dev, u8 opcode); |
d4b4ff8e | 102 | |
1c63dc66 CH |
103 | /* |
104 | * Represents an NVM Express device. Each nvme_dev is a PCI function. | |
105 | */ | |
106 | struct nvme_dev { | |
147b27e4 | 107 | struct nvme_queue *queues; |
1c63dc66 CH |
108 | struct blk_mq_tag_set tagset; |
109 | struct blk_mq_tag_set admin_tagset; | |
110 | u32 __iomem *dbs; | |
111 | struct device *dev; | |
112 | struct dma_pool *prp_page_pool; | |
113 | struct dma_pool *prp_small_pool; | |
1c63dc66 CH |
114 | unsigned online_queues; |
115 | unsigned max_qid; | |
e20ba6e1 | 116 | unsigned io_queues[HCTX_MAX_TYPES]; |
22b55601 | 117 | unsigned int num_vecs; |
61f3b896 | 118 | u16 q_depth; |
c1e0cc7e | 119 | int io_sqes; |
1c63dc66 | 120 | u32 db_stride; |
1c63dc66 | 121 | void __iomem *bar; |
97f6ef64 | 122 | unsigned long bar_mapped_size; |
5c8809e6 | 123 | struct work_struct remove_work; |
77bf25ea | 124 | struct mutex shutdown_lock; |
1c63dc66 | 125 | bool subsystem; |
1c63dc66 | 126 | u64 cmb_size; |
0f238ff5 | 127 | bool cmb_use_sqes; |
1c63dc66 | 128 | u32 cmbsz; |
202021c1 | 129 | u32 cmbloc; |
1c63dc66 | 130 | struct nvme_ctrl ctrl; |
d916b1be | 131 | u32 last_ps; |
87ad72a5 | 132 | |
943e942e JA |
133 | mempool_t *iod_mempool; |
134 | ||
87ad72a5 | 135 | /* shadow doorbell buffer support: */ |
f9f38e33 HK |
136 | u32 *dbbuf_dbs; |
137 | dma_addr_t dbbuf_dbs_dma_addr; | |
138 | u32 *dbbuf_eis; | |
139 | dma_addr_t dbbuf_eis_dma_addr; | |
87ad72a5 CH |
140 | |
141 | /* host memory buffer support: */ | |
142 | u64 host_mem_size; | |
143 | u32 nr_host_mem_descs; | |
4033f35d | 144 | dma_addr_t host_mem_descs_dma; |
87ad72a5 CH |
145 | struct nvme_host_mem_buf_desc *host_mem_descs; |
146 | void **host_mem_desc_bufs; | |
2a5bcfdd WZ |
147 | unsigned int nr_allocated_queues; |
148 | unsigned int nr_write_queues; | |
149 | unsigned int nr_poll_queues; | |
4d115420 | 150 | }; |
1fa6aead | 151 | |
b27c1e68 | 152 | static int io_queue_depth_set(const char *val, const struct kernel_param *kp) |
153 | { | |
61f3b896 CK |
154 | int ret; |
155 | u16 n; | |
b27c1e68 | 156 | |
61f3b896 | 157 | ret = kstrtou16(val, 10, &n); |
b27c1e68 | 158 | if (ret != 0 || n < 2) |
159 | return -EINVAL; | |
160 | ||
61f3b896 | 161 | return param_set_ushort(val, kp); |
b27c1e68 | 162 | } |
163 | ||
f9f38e33 HK |
164 | static inline unsigned int sq_idx(unsigned int qid, u32 stride) |
165 | { | |
166 | return qid * 2 * stride; | |
167 | } | |
168 | ||
169 | static inline unsigned int cq_idx(unsigned int qid, u32 stride) | |
170 | { | |
171 | return (qid * 2 + 1) * stride; | |
172 | } | |
173 | ||
1c63dc66 CH |
174 | static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl) |
175 | { | |
176 | return container_of(ctrl, struct nvme_dev, ctrl); | |
177 | } | |
178 | ||
b60503ba MW |
179 | /* |
180 | * An NVM Express queue. Each device has at least two (one for admin | |
181 | * commands and one for I/O commands). | |
182 | */ | |
183 | struct nvme_queue { | |
091b6092 | 184 | struct nvme_dev *dev; |
1ab0cd69 | 185 | spinlock_t sq_lock; |
c1e0cc7e | 186 | void *sq_cmds; |
3a7afd8e CH |
187 | /* only used for poll queues: */ |
188 | spinlock_t cq_poll_lock ____cacheline_aligned_in_smp; | |
74943d45 | 189 | struct nvme_completion *cqes; |
b60503ba MW |
190 | dma_addr_t sq_dma_addr; |
191 | dma_addr_t cq_dma_addr; | |
b60503ba MW |
192 | u32 __iomem *q_db; |
193 | u16 q_depth; | |
7c349dde | 194 | u16 cq_vector; |
b60503ba MW |
195 | u16 sq_tail; |
196 | u16 cq_head; | |
c30341dc | 197 | u16 qid; |
e9539f47 | 198 | u8 cq_phase; |
c1e0cc7e | 199 | u8 sqes; |
4e224106 CH |
200 | unsigned long flags; |
201 | #define NVMEQ_ENABLED 0 | |
63223078 | 202 | #define NVMEQ_SQ_CMB 1 |
d1ed6aa1 | 203 | #define NVMEQ_DELETE_ERROR 2 |
7c349dde | 204 | #define NVMEQ_POLLED 3 |
f9f38e33 HK |
205 | u32 *dbbuf_sq_db; |
206 | u32 *dbbuf_cq_db; | |
207 | u32 *dbbuf_sq_ei; | |
208 | u32 *dbbuf_cq_ei; | |
d1ed6aa1 | 209 | struct completion delete_done; |
b60503ba MW |
210 | }; |
211 | ||
71bd150c | 212 | /* |
9b048119 CH |
213 | * The nvme_iod describes the data in an I/O. |
214 | * | |
215 | * The sg pointer contains the list of PRP/SGL chunk allocations in addition | |
216 | * to the actual struct scatterlist. | |
71bd150c CH |
217 | */ |
218 | struct nvme_iod { | |
d49187e9 | 219 | struct nvme_request req; |
f4800d6d | 220 | struct nvme_queue *nvmeq; |
a7a7cbe3 | 221 | bool use_sgl; |
f4800d6d | 222 | int aborted; |
71bd150c | 223 | int npages; /* In the PRP list. 0 means small pool in use */ |
71bd150c | 224 | int nents; /* Used in scatterlist */ |
71bd150c | 225 | dma_addr_t first_dma; |
dff824b2 | 226 | unsigned int dma_len; /* length of single DMA segment mapping */ |
783b94bd | 227 | dma_addr_t meta_dma; |
f4800d6d | 228 | struct scatterlist *sg; |
b60503ba MW |
229 | }; |
230 | ||
2a5bcfdd | 231 | static inline unsigned int nvme_dbbuf_size(struct nvme_dev *dev) |
3b6592f7 | 232 | { |
2a5bcfdd | 233 | return dev->nr_allocated_queues * 8 * dev->db_stride; |
f9f38e33 HK |
234 | } |
235 | ||
236 | static int nvme_dbbuf_dma_alloc(struct nvme_dev *dev) | |
237 | { | |
2a5bcfdd | 238 | unsigned int mem_size = nvme_dbbuf_size(dev); |
f9f38e33 HK |
239 | |
240 | if (dev->dbbuf_dbs) | |
241 | return 0; | |
242 | ||
243 | dev->dbbuf_dbs = dma_alloc_coherent(dev->dev, mem_size, | |
244 | &dev->dbbuf_dbs_dma_addr, | |
245 | GFP_KERNEL); | |
246 | if (!dev->dbbuf_dbs) | |
247 | return -ENOMEM; | |
248 | dev->dbbuf_eis = dma_alloc_coherent(dev->dev, mem_size, | |
249 | &dev->dbbuf_eis_dma_addr, | |
250 | GFP_KERNEL); | |
251 | if (!dev->dbbuf_eis) { | |
252 | dma_free_coherent(dev->dev, mem_size, | |
253 | dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr); | |
254 | dev->dbbuf_dbs = NULL; | |
255 | return -ENOMEM; | |
256 | } | |
257 | ||
258 | return 0; | |
259 | } | |
260 | ||
261 | static void nvme_dbbuf_dma_free(struct nvme_dev *dev) | |
262 | { | |
2a5bcfdd | 263 | unsigned int mem_size = nvme_dbbuf_size(dev); |
f9f38e33 HK |
264 | |
265 | if (dev->dbbuf_dbs) { | |
266 | dma_free_coherent(dev->dev, mem_size, | |
267 | dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr); | |
268 | dev->dbbuf_dbs = NULL; | |
269 | } | |
270 | if (dev->dbbuf_eis) { | |
271 | dma_free_coherent(dev->dev, mem_size, | |
272 | dev->dbbuf_eis, dev->dbbuf_eis_dma_addr); | |
273 | dev->dbbuf_eis = NULL; | |
274 | } | |
275 | } | |
276 | ||
277 | static void nvme_dbbuf_init(struct nvme_dev *dev, | |
278 | struct nvme_queue *nvmeq, int qid) | |
279 | { | |
280 | if (!dev->dbbuf_dbs || !qid) | |
281 | return; | |
282 | ||
283 | nvmeq->dbbuf_sq_db = &dev->dbbuf_dbs[sq_idx(qid, dev->db_stride)]; | |
284 | nvmeq->dbbuf_cq_db = &dev->dbbuf_dbs[cq_idx(qid, dev->db_stride)]; | |
285 | nvmeq->dbbuf_sq_ei = &dev->dbbuf_eis[sq_idx(qid, dev->db_stride)]; | |
286 | nvmeq->dbbuf_cq_ei = &dev->dbbuf_eis[cq_idx(qid, dev->db_stride)]; | |
287 | } | |
288 | ||
289 | static void nvme_dbbuf_set(struct nvme_dev *dev) | |
290 | { | |
291 | struct nvme_command c; | |
292 | ||
293 | if (!dev->dbbuf_dbs) | |
294 | return; | |
295 | ||
296 | memset(&c, 0, sizeof(c)); | |
297 | c.dbbuf.opcode = nvme_admin_dbbuf; | |
298 | c.dbbuf.prp1 = cpu_to_le64(dev->dbbuf_dbs_dma_addr); | |
299 | c.dbbuf.prp2 = cpu_to_le64(dev->dbbuf_eis_dma_addr); | |
300 | ||
301 | if (nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0)) { | |
9bdcfb10 | 302 | dev_warn(dev->ctrl.device, "unable to set dbbuf\n"); |
f9f38e33 HK |
303 | /* Free memory and continue on */ |
304 | nvme_dbbuf_dma_free(dev); | |
305 | } | |
306 | } | |
307 | ||
308 | static inline int nvme_dbbuf_need_event(u16 event_idx, u16 new_idx, u16 old) | |
309 | { | |
310 | return (u16)(new_idx - event_idx - 1) < (u16)(new_idx - old); | |
311 | } | |
312 | ||
313 | /* Update dbbuf and return true if an MMIO is required */ | |
314 | static bool nvme_dbbuf_update_and_check_event(u16 value, u32 *dbbuf_db, | |
315 | volatile u32 *dbbuf_ei) | |
316 | { | |
317 | if (dbbuf_db) { | |
318 | u16 old_value; | |
319 | ||
320 | /* | |
321 | * Ensure that the queue is written before updating | |
322 | * the doorbell in memory | |
323 | */ | |
324 | wmb(); | |
325 | ||
326 | old_value = *dbbuf_db; | |
327 | *dbbuf_db = value; | |
328 | ||
f1ed3df2 MW |
329 | /* |
330 | * Ensure that the doorbell is updated before reading the event | |
331 | * index from memory. The controller needs to provide similar | |
332 | * ordering to ensure the envent index is updated before reading | |
333 | * the doorbell. | |
334 | */ | |
335 | mb(); | |
336 | ||
f9f38e33 HK |
337 | if (!nvme_dbbuf_need_event(*dbbuf_ei, value, old_value)) |
338 | return false; | |
339 | } | |
340 | ||
341 | return true; | |
b60503ba MW |
342 | } |
343 | ||
ac3dd5bd JA |
344 | /* |
345 | * Will slightly overestimate the number of pages needed. This is OK | |
346 | * as it only leads to a small amount of wasted memory for the lifetime of | |
347 | * the I/O. | |
348 | */ | |
349 | static int nvme_npages(unsigned size, struct nvme_dev *dev) | |
350 | { | |
5fd4ce1b CH |
351 | unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size, |
352 | dev->ctrl.page_size); | |
ac3dd5bd JA |
353 | return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); |
354 | } | |
355 | ||
a7a7cbe3 CK |
356 | /* |
357 | * Calculates the number of pages needed for the SGL segments. For example a 4k | |
358 | * page can accommodate 256 SGL descriptors. | |
359 | */ | |
360 | static int nvme_pci_npages_sgl(unsigned int num_seg) | |
ac3dd5bd | 361 | { |
a7a7cbe3 | 362 | return DIV_ROUND_UP(num_seg * sizeof(struct nvme_sgl_desc), PAGE_SIZE); |
f4800d6d | 363 | } |
ac3dd5bd | 364 | |
9056fc9f | 365 | static size_t nvme_pci_iod_alloc_size(struct nvme_dev *dev, |
a7a7cbe3 | 366 | unsigned int size, unsigned int nseg, bool use_sgl) |
f4800d6d | 367 | { |
a7a7cbe3 CK |
368 | size_t alloc_size; |
369 | ||
370 | if (use_sgl) | |
371 | alloc_size = sizeof(__le64 *) * nvme_pci_npages_sgl(nseg); | |
372 | else | |
373 | alloc_size = sizeof(__le64 *) * nvme_npages(size, dev); | |
374 | ||
375 | return alloc_size + sizeof(struct scatterlist) * nseg; | |
f4800d6d | 376 | } |
ac3dd5bd | 377 | |
a4aea562 MB |
378 | static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
379 | unsigned int hctx_idx) | |
e85248e5 | 380 | { |
a4aea562 | 381 | struct nvme_dev *dev = data; |
147b27e4 | 382 | struct nvme_queue *nvmeq = &dev->queues[0]; |
a4aea562 | 383 | |
42483228 KB |
384 | WARN_ON(hctx_idx != 0); |
385 | WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); | |
42483228 | 386 | |
a4aea562 MB |
387 | hctx->driver_data = nvmeq; |
388 | return 0; | |
e85248e5 MW |
389 | } |
390 | ||
a4aea562 MB |
391 | static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
392 | unsigned int hctx_idx) | |
b60503ba | 393 | { |
a4aea562 | 394 | struct nvme_dev *dev = data; |
147b27e4 | 395 | struct nvme_queue *nvmeq = &dev->queues[hctx_idx + 1]; |
a4aea562 | 396 | |
42483228 | 397 | WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); |
a4aea562 MB |
398 | hctx->driver_data = nvmeq; |
399 | return 0; | |
b60503ba MW |
400 | } |
401 | ||
d6296d39 CH |
402 | static int nvme_init_request(struct blk_mq_tag_set *set, struct request *req, |
403 | unsigned int hctx_idx, unsigned int numa_node) | |
b60503ba | 404 | { |
d6296d39 | 405 | struct nvme_dev *dev = set->driver_data; |
f4800d6d | 406 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
0350815a | 407 | int queue_idx = (set == &dev->tagset) ? hctx_idx + 1 : 0; |
147b27e4 | 408 | struct nvme_queue *nvmeq = &dev->queues[queue_idx]; |
a4aea562 MB |
409 | |
410 | BUG_ON(!nvmeq); | |
f4800d6d | 411 | iod->nvmeq = nvmeq; |
59e29ce6 SG |
412 | |
413 | nvme_req(req)->ctrl = &dev->ctrl; | |
a4aea562 MB |
414 | return 0; |
415 | } | |
416 | ||
3b6592f7 JA |
417 | static int queue_irq_offset(struct nvme_dev *dev) |
418 | { | |
419 | /* if we have more than 1 vec, admin queue offsets us by 1 */ | |
420 | if (dev->num_vecs > 1) | |
421 | return 1; | |
422 | ||
423 | return 0; | |
424 | } | |
425 | ||
dca51e78 CH |
426 | static int nvme_pci_map_queues(struct blk_mq_tag_set *set) |
427 | { | |
428 | struct nvme_dev *dev = set->driver_data; | |
3b6592f7 JA |
429 | int i, qoff, offset; |
430 | ||
431 | offset = queue_irq_offset(dev); | |
432 | for (i = 0, qoff = 0; i < set->nr_maps; i++) { | |
433 | struct blk_mq_queue_map *map = &set->map[i]; | |
434 | ||
435 | map->nr_queues = dev->io_queues[i]; | |
436 | if (!map->nr_queues) { | |
e20ba6e1 | 437 | BUG_ON(i == HCTX_TYPE_DEFAULT); |
7e849dd9 | 438 | continue; |
3b6592f7 JA |
439 | } |
440 | ||
4b04cc6a JA |
441 | /* |
442 | * The poll queue(s) doesn't have an IRQ (and hence IRQ | |
443 | * affinity), so use the regular blk-mq cpu mapping | |
444 | */ | |
3b6592f7 | 445 | map->queue_offset = qoff; |
cb9e0e50 | 446 | if (i != HCTX_TYPE_POLL && offset) |
4b04cc6a JA |
447 | blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset); |
448 | else | |
449 | blk_mq_map_queues(map); | |
3b6592f7 JA |
450 | qoff += map->nr_queues; |
451 | offset += map->nr_queues; | |
452 | } | |
453 | ||
454 | return 0; | |
dca51e78 CH |
455 | } |
456 | ||
54b2fcee | 457 | static inline void nvme_write_sq_db(struct nvme_queue *nvmeq) |
04f3eafd | 458 | { |
04f3eafd JA |
459 | if (nvme_dbbuf_update_and_check_event(nvmeq->sq_tail, |
460 | nvmeq->dbbuf_sq_db, nvmeq->dbbuf_sq_ei)) | |
461 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
04f3eafd JA |
462 | } |
463 | ||
b60503ba | 464 | /** |
90ea5ca4 | 465 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
466 | * @nvmeq: The queue to use |
467 | * @cmd: The command to send | |
04f3eafd | 468 | * @write_sq: whether to write to the SQ doorbell |
b60503ba | 469 | */ |
04f3eafd JA |
470 | static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd, |
471 | bool write_sq) | |
b60503ba | 472 | { |
90ea5ca4 | 473 | spin_lock(&nvmeq->sq_lock); |
c1e0cc7e BH |
474 | memcpy(nvmeq->sq_cmds + (nvmeq->sq_tail << nvmeq->sqes), |
475 | cmd, sizeof(*cmd)); | |
90ea5ca4 CH |
476 | if (++nvmeq->sq_tail == nvmeq->q_depth) |
477 | nvmeq->sq_tail = 0; | |
54b2fcee KB |
478 | if (write_sq) |
479 | nvme_write_sq_db(nvmeq); | |
04f3eafd JA |
480 | spin_unlock(&nvmeq->sq_lock); |
481 | } | |
482 | ||
483 | static void nvme_commit_rqs(struct blk_mq_hw_ctx *hctx) | |
484 | { | |
485 | struct nvme_queue *nvmeq = hctx->driver_data; | |
486 | ||
487 | spin_lock(&nvmeq->sq_lock); | |
54b2fcee | 488 | nvme_write_sq_db(nvmeq); |
90ea5ca4 | 489 | spin_unlock(&nvmeq->sq_lock); |
b60503ba MW |
490 | } |
491 | ||
a7a7cbe3 | 492 | static void **nvme_pci_iod_list(struct request *req) |
b60503ba | 493 | { |
f4800d6d | 494 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
a7a7cbe3 | 495 | return (void **)(iod->sg + blk_rq_nr_phys_segments(req)); |
b60503ba MW |
496 | } |
497 | ||
955b1b5a MI |
498 | static inline bool nvme_pci_use_sgls(struct nvme_dev *dev, struct request *req) |
499 | { | |
500 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
20469a37 | 501 | int nseg = blk_rq_nr_phys_segments(req); |
955b1b5a MI |
502 | unsigned int avg_seg_size; |
503 | ||
20469a37 | 504 | avg_seg_size = DIV_ROUND_UP(blk_rq_payload_bytes(req), nseg); |
955b1b5a MI |
505 | |
506 | if (!(dev->ctrl.sgls & ((1 << 0) | (1 << 1)))) | |
507 | return false; | |
508 | if (!iod->nvmeq->qid) | |
509 | return false; | |
510 | if (!sgl_threshold || avg_seg_size < sgl_threshold) | |
511 | return false; | |
512 | return true; | |
513 | } | |
514 | ||
7fe07d14 | 515 | static void nvme_unmap_data(struct nvme_dev *dev, struct request *req) |
b60503ba | 516 | { |
f4800d6d | 517 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
a7a7cbe3 CK |
518 | const int last_prp = dev->ctrl.page_size / sizeof(__le64) - 1; |
519 | dma_addr_t dma_addr = iod->first_dma, next_dma_addr; | |
eca18b23 | 520 | int i; |
eca18b23 | 521 | |
dff824b2 | 522 | if (iod->dma_len) { |
f2fa006f IR |
523 | dma_unmap_page(dev->dev, dma_addr, iod->dma_len, |
524 | rq_dma_dir(req)); | |
dff824b2 | 525 | return; |
7fe07d14 CH |
526 | } |
527 | ||
dff824b2 CH |
528 | WARN_ON_ONCE(!iod->nents); |
529 | ||
7f73eac3 LG |
530 | if (is_pci_p2pdma_page(sg_page(iod->sg))) |
531 | pci_p2pdma_unmap_sg(dev->dev, iod->sg, iod->nents, | |
532 | rq_dma_dir(req)); | |
533 | else | |
dff824b2 CH |
534 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, rq_dma_dir(req)); |
535 | ||
536 | ||
eca18b23 | 537 | if (iod->npages == 0) |
a7a7cbe3 CK |
538 | dma_pool_free(dev->prp_small_pool, nvme_pci_iod_list(req)[0], |
539 | dma_addr); | |
540 | ||
eca18b23 | 541 | for (i = 0; i < iod->npages; i++) { |
a7a7cbe3 CK |
542 | void *addr = nvme_pci_iod_list(req)[i]; |
543 | ||
544 | if (iod->use_sgl) { | |
545 | struct nvme_sgl_desc *sg_list = addr; | |
546 | ||
547 | next_dma_addr = | |
548 | le64_to_cpu((sg_list[SGES_PER_PAGE - 1]).addr); | |
549 | } else { | |
550 | __le64 *prp_list = addr; | |
551 | ||
552 | next_dma_addr = le64_to_cpu(prp_list[last_prp]); | |
553 | } | |
554 | ||
555 | dma_pool_free(dev->prp_page_pool, addr, dma_addr); | |
556 | dma_addr = next_dma_addr; | |
eca18b23 | 557 | } |
ac3dd5bd | 558 | |
d43f1ccf | 559 | mempool_free(iod->sg, dev->iod_mempool); |
b4ff9c8d KB |
560 | } |
561 | ||
d0877473 KB |
562 | static void nvme_print_sgl(struct scatterlist *sgl, int nents) |
563 | { | |
564 | int i; | |
565 | struct scatterlist *sg; | |
566 | ||
567 | for_each_sg(sgl, sg, nents, i) { | |
568 | dma_addr_t phys = sg_phys(sg); | |
569 | pr_warn("sg[%d] phys_addr:%pad offset:%d length:%d " | |
570 | "dma_address:%pad dma_length:%d\n", | |
571 | i, &phys, sg->offset, sg->length, &sg_dma_address(sg), | |
572 | sg_dma_len(sg)); | |
573 | } | |
574 | } | |
575 | ||
a7a7cbe3 CK |
576 | static blk_status_t nvme_pci_setup_prps(struct nvme_dev *dev, |
577 | struct request *req, struct nvme_rw_command *cmnd) | |
ff22b54f | 578 | { |
f4800d6d | 579 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
99802a7a | 580 | struct dma_pool *pool; |
b131c61d | 581 | int length = blk_rq_payload_bytes(req); |
eca18b23 | 582 | struct scatterlist *sg = iod->sg; |
ff22b54f MW |
583 | int dma_len = sg_dma_len(sg); |
584 | u64 dma_addr = sg_dma_address(sg); | |
5fd4ce1b | 585 | u32 page_size = dev->ctrl.page_size; |
f137e0f1 | 586 | int offset = dma_addr & (page_size - 1); |
e025344c | 587 | __le64 *prp_list; |
a7a7cbe3 | 588 | void **list = nvme_pci_iod_list(req); |
e025344c | 589 | dma_addr_t prp_dma; |
eca18b23 | 590 | int nprps, i; |
ff22b54f | 591 | |
1d090624 | 592 | length -= (page_size - offset); |
5228b328 JS |
593 | if (length <= 0) { |
594 | iod->first_dma = 0; | |
a7a7cbe3 | 595 | goto done; |
5228b328 | 596 | } |
ff22b54f | 597 | |
1d090624 | 598 | dma_len -= (page_size - offset); |
ff22b54f | 599 | if (dma_len) { |
1d090624 | 600 | dma_addr += (page_size - offset); |
ff22b54f MW |
601 | } else { |
602 | sg = sg_next(sg); | |
603 | dma_addr = sg_dma_address(sg); | |
604 | dma_len = sg_dma_len(sg); | |
605 | } | |
606 | ||
1d090624 | 607 | if (length <= page_size) { |
edd10d33 | 608 | iod->first_dma = dma_addr; |
a7a7cbe3 | 609 | goto done; |
e025344c SMM |
610 | } |
611 | ||
1d090624 | 612 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
613 | if (nprps <= (256 / 8)) { |
614 | pool = dev->prp_small_pool; | |
eca18b23 | 615 | iod->npages = 0; |
99802a7a MW |
616 | } else { |
617 | pool = dev->prp_page_pool; | |
eca18b23 | 618 | iod->npages = 1; |
99802a7a MW |
619 | } |
620 | ||
69d2b571 | 621 | prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); |
b77954cb | 622 | if (!prp_list) { |
edd10d33 | 623 | iod->first_dma = dma_addr; |
eca18b23 | 624 | iod->npages = -1; |
86eea289 | 625 | return BLK_STS_RESOURCE; |
b77954cb | 626 | } |
eca18b23 MW |
627 | list[0] = prp_list; |
628 | iod->first_dma = prp_dma; | |
e025344c SMM |
629 | i = 0; |
630 | for (;;) { | |
1d090624 | 631 | if (i == page_size >> 3) { |
e025344c | 632 | __le64 *old_prp_list = prp_list; |
69d2b571 | 633 | prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); |
eca18b23 | 634 | if (!prp_list) |
86eea289 | 635 | return BLK_STS_RESOURCE; |
eca18b23 | 636 | list[iod->npages++] = prp_list; |
7523d834 MW |
637 | prp_list[0] = old_prp_list[i - 1]; |
638 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
639 | i = 1; | |
e025344c SMM |
640 | } |
641 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
642 | dma_len -= page_size; |
643 | dma_addr += page_size; | |
644 | length -= page_size; | |
e025344c SMM |
645 | if (length <= 0) |
646 | break; | |
647 | if (dma_len > 0) | |
648 | continue; | |
86eea289 KB |
649 | if (unlikely(dma_len < 0)) |
650 | goto bad_sgl; | |
e025344c SMM |
651 | sg = sg_next(sg); |
652 | dma_addr = sg_dma_address(sg); | |
653 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
654 | } |
655 | ||
a7a7cbe3 CK |
656 | done: |
657 | cmnd->dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
658 | cmnd->dptr.prp2 = cpu_to_le64(iod->first_dma); | |
659 | ||
86eea289 KB |
660 | return BLK_STS_OK; |
661 | ||
662 | bad_sgl: | |
d0877473 KB |
663 | WARN(DO_ONCE(nvme_print_sgl, iod->sg, iod->nents), |
664 | "Invalid SGL for payload:%d nents:%d\n", | |
665 | blk_rq_payload_bytes(req), iod->nents); | |
86eea289 | 666 | return BLK_STS_IOERR; |
ff22b54f MW |
667 | } |
668 | ||
a7a7cbe3 CK |
669 | static void nvme_pci_sgl_set_data(struct nvme_sgl_desc *sge, |
670 | struct scatterlist *sg) | |
671 | { | |
672 | sge->addr = cpu_to_le64(sg_dma_address(sg)); | |
673 | sge->length = cpu_to_le32(sg_dma_len(sg)); | |
674 | sge->type = NVME_SGL_FMT_DATA_DESC << 4; | |
675 | } | |
676 | ||
677 | static void nvme_pci_sgl_set_seg(struct nvme_sgl_desc *sge, | |
678 | dma_addr_t dma_addr, int entries) | |
679 | { | |
680 | sge->addr = cpu_to_le64(dma_addr); | |
681 | if (entries < SGES_PER_PAGE) { | |
682 | sge->length = cpu_to_le32(entries * sizeof(*sge)); | |
683 | sge->type = NVME_SGL_FMT_LAST_SEG_DESC << 4; | |
684 | } else { | |
685 | sge->length = cpu_to_le32(PAGE_SIZE); | |
686 | sge->type = NVME_SGL_FMT_SEG_DESC << 4; | |
687 | } | |
688 | } | |
689 | ||
690 | static blk_status_t nvme_pci_setup_sgls(struct nvme_dev *dev, | |
b0f2853b | 691 | struct request *req, struct nvme_rw_command *cmd, int entries) |
a7a7cbe3 CK |
692 | { |
693 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
a7a7cbe3 CK |
694 | struct dma_pool *pool; |
695 | struct nvme_sgl_desc *sg_list; | |
696 | struct scatterlist *sg = iod->sg; | |
a7a7cbe3 | 697 | dma_addr_t sgl_dma; |
b0f2853b | 698 | int i = 0; |
a7a7cbe3 | 699 | |
a7a7cbe3 CK |
700 | /* setting the transfer type as SGL */ |
701 | cmd->flags = NVME_CMD_SGL_METABUF; | |
702 | ||
b0f2853b | 703 | if (entries == 1) { |
a7a7cbe3 CK |
704 | nvme_pci_sgl_set_data(&cmd->dptr.sgl, sg); |
705 | return BLK_STS_OK; | |
706 | } | |
707 | ||
708 | if (entries <= (256 / sizeof(struct nvme_sgl_desc))) { | |
709 | pool = dev->prp_small_pool; | |
710 | iod->npages = 0; | |
711 | } else { | |
712 | pool = dev->prp_page_pool; | |
713 | iod->npages = 1; | |
714 | } | |
715 | ||
716 | sg_list = dma_pool_alloc(pool, GFP_ATOMIC, &sgl_dma); | |
717 | if (!sg_list) { | |
718 | iod->npages = -1; | |
719 | return BLK_STS_RESOURCE; | |
720 | } | |
721 | ||
722 | nvme_pci_iod_list(req)[0] = sg_list; | |
723 | iod->first_dma = sgl_dma; | |
724 | ||
725 | nvme_pci_sgl_set_seg(&cmd->dptr.sgl, sgl_dma, entries); | |
726 | ||
727 | do { | |
728 | if (i == SGES_PER_PAGE) { | |
729 | struct nvme_sgl_desc *old_sg_desc = sg_list; | |
730 | struct nvme_sgl_desc *link = &old_sg_desc[i - 1]; | |
731 | ||
732 | sg_list = dma_pool_alloc(pool, GFP_ATOMIC, &sgl_dma); | |
733 | if (!sg_list) | |
734 | return BLK_STS_RESOURCE; | |
735 | ||
736 | i = 0; | |
737 | nvme_pci_iod_list(req)[iod->npages++] = sg_list; | |
738 | sg_list[i++] = *link; | |
739 | nvme_pci_sgl_set_seg(link, sgl_dma, entries); | |
740 | } | |
741 | ||
742 | nvme_pci_sgl_set_data(&sg_list[i++], sg); | |
a7a7cbe3 | 743 | sg = sg_next(sg); |
b0f2853b | 744 | } while (--entries > 0); |
a7a7cbe3 | 745 | |
a7a7cbe3 CK |
746 | return BLK_STS_OK; |
747 | } | |
748 | ||
dff824b2 CH |
749 | static blk_status_t nvme_setup_prp_simple(struct nvme_dev *dev, |
750 | struct request *req, struct nvme_rw_command *cmnd, | |
751 | struct bio_vec *bv) | |
752 | { | |
753 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
a4f40484 KH |
754 | unsigned int offset = bv->bv_offset & (dev->ctrl.page_size - 1); |
755 | unsigned int first_prp_len = dev->ctrl.page_size - offset; | |
dff824b2 CH |
756 | |
757 | iod->first_dma = dma_map_bvec(dev->dev, bv, rq_dma_dir(req), 0); | |
758 | if (dma_mapping_error(dev->dev, iod->first_dma)) | |
759 | return BLK_STS_RESOURCE; | |
760 | iod->dma_len = bv->bv_len; | |
761 | ||
762 | cmnd->dptr.prp1 = cpu_to_le64(iod->first_dma); | |
763 | if (bv->bv_len > first_prp_len) | |
764 | cmnd->dptr.prp2 = cpu_to_le64(iod->first_dma + first_prp_len); | |
359c1f88 | 765 | return BLK_STS_OK; |
dff824b2 CH |
766 | } |
767 | ||
29791057 CH |
768 | static blk_status_t nvme_setup_sgl_simple(struct nvme_dev *dev, |
769 | struct request *req, struct nvme_rw_command *cmnd, | |
770 | struct bio_vec *bv) | |
771 | { | |
772 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
773 | ||
774 | iod->first_dma = dma_map_bvec(dev->dev, bv, rq_dma_dir(req), 0); | |
775 | if (dma_mapping_error(dev->dev, iod->first_dma)) | |
776 | return BLK_STS_RESOURCE; | |
777 | iod->dma_len = bv->bv_len; | |
778 | ||
049bf372 | 779 | cmnd->flags = NVME_CMD_SGL_METABUF; |
29791057 CH |
780 | cmnd->dptr.sgl.addr = cpu_to_le64(iod->first_dma); |
781 | cmnd->dptr.sgl.length = cpu_to_le32(iod->dma_len); | |
782 | cmnd->dptr.sgl.type = NVME_SGL_FMT_DATA_DESC << 4; | |
359c1f88 | 783 | return BLK_STS_OK; |
29791057 CH |
784 | } |
785 | ||
fc17b653 | 786 | static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req, |
b131c61d | 787 | struct nvme_command *cmnd) |
d29ec824 | 788 | { |
f4800d6d | 789 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
70479b71 | 790 | blk_status_t ret = BLK_STS_RESOURCE; |
b0f2853b | 791 | int nr_mapped; |
d29ec824 | 792 | |
dff824b2 CH |
793 | if (blk_rq_nr_phys_segments(req) == 1) { |
794 | struct bio_vec bv = req_bvec(req); | |
795 | ||
796 | if (!is_pci_p2pdma_page(bv.bv_page)) { | |
797 | if (bv.bv_offset + bv.bv_len <= dev->ctrl.page_size * 2) | |
798 | return nvme_setup_prp_simple(dev, req, | |
799 | &cmnd->rw, &bv); | |
29791057 CH |
800 | |
801 | if (iod->nvmeq->qid && | |
802 | dev->ctrl.sgls & ((1 << 0) | (1 << 1))) | |
803 | return nvme_setup_sgl_simple(dev, req, | |
804 | &cmnd->rw, &bv); | |
dff824b2 CH |
805 | } |
806 | } | |
807 | ||
808 | iod->dma_len = 0; | |
d43f1ccf CH |
809 | iod->sg = mempool_alloc(dev->iod_mempool, GFP_ATOMIC); |
810 | if (!iod->sg) | |
811 | return BLK_STS_RESOURCE; | |
f9d03f96 | 812 | sg_init_table(iod->sg, blk_rq_nr_phys_segments(req)); |
70479b71 | 813 | iod->nents = blk_rq_map_sg(req->q, req, iod->sg); |
ba1ca37e CH |
814 | if (!iod->nents) |
815 | goto out; | |
d29ec824 | 816 | |
e0596ab2 | 817 | if (is_pci_p2pdma_page(sg_page(iod->sg))) |
2b9f4bb2 LG |
818 | nr_mapped = pci_p2pdma_map_sg_attrs(dev->dev, iod->sg, |
819 | iod->nents, rq_dma_dir(req), DMA_ATTR_NO_WARN); | |
e0596ab2 LG |
820 | else |
821 | nr_mapped = dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, | |
70479b71 | 822 | rq_dma_dir(req), DMA_ATTR_NO_WARN); |
b0f2853b | 823 | if (!nr_mapped) |
ba1ca37e | 824 | goto out; |
d29ec824 | 825 | |
70479b71 | 826 | iod->use_sgl = nvme_pci_use_sgls(dev, req); |
955b1b5a | 827 | if (iod->use_sgl) |
b0f2853b | 828 | ret = nvme_pci_setup_sgls(dev, req, &cmnd->rw, nr_mapped); |
a7a7cbe3 CK |
829 | else |
830 | ret = nvme_pci_setup_prps(dev, req, &cmnd->rw); | |
4aedb705 | 831 | out: |
86eea289 | 832 | if (ret != BLK_STS_OK) |
4aedb705 CH |
833 | nvme_unmap_data(dev, req); |
834 | return ret; | |
835 | } | |
3045c0d0 | 836 | |
4aedb705 CH |
837 | static blk_status_t nvme_map_metadata(struct nvme_dev *dev, struct request *req, |
838 | struct nvme_command *cmnd) | |
839 | { | |
840 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
00df5cb4 | 841 | |
4aedb705 CH |
842 | iod->meta_dma = dma_map_bvec(dev->dev, rq_integrity_vec(req), |
843 | rq_dma_dir(req), 0); | |
844 | if (dma_mapping_error(dev->dev, iod->meta_dma)) | |
845 | return BLK_STS_IOERR; | |
846 | cmnd->rw.metadata = cpu_to_le64(iod->meta_dma); | |
359c1f88 | 847 | return BLK_STS_OK; |
00df5cb4 MW |
848 | } |
849 | ||
d29ec824 CH |
850 | /* |
851 | * NOTE: ns is NULL when called on the admin queue. | |
852 | */ | |
fc17b653 | 853 | static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx, |
a4aea562 | 854 | const struct blk_mq_queue_data *bd) |
edd10d33 | 855 | { |
a4aea562 MB |
856 | struct nvme_ns *ns = hctx->queue->queuedata; |
857 | struct nvme_queue *nvmeq = hctx->driver_data; | |
d29ec824 | 858 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 | 859 | struct request *req = bd->rq; |
9b048119 | 860 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
ba1ca37e | 861 | struct nvme_command cmnd; |
ebe6d874 | 862 | blk_status_t ret; |
e1e5e564 | 863 | |
9b048119 CH |
864 | iod->aborted = 0; |
865 | iod->npages = -1; | |
866 | iod->nents = 0; | |
867 | ||
d1f06f4a JA |
868 | /* |
869 | * We should not need to do this, but we're still using this to | |
870 | * ensure we can drain requests on a dying queue. | |
871 | */ | |
4e224106 | 872 | if (unlikely(!test_bit(NVMEQ_ENABLED, &nvmeq->flags))) |
d1f06f4a JA |
873 | return BLK_STS_IOERR; |
874 | ||
f9d03f96 | 875 | ret = nvme_setup_cmd(ns, req, &cmnd); |
fc17b653 | 876 | if (ret) |
f4800d6d | 877 | return ret; |
a4aea562 | 878 | |
fc17b653 | 879 | if (blk_rq_nr_phys_segments(req)) { |
b131c61d | 880 | ret = nvme_map_data(dev, req, &cmnd); |
fc17b653 | 881 | if (ret) |
9b048119 | 882 | goto out_free_cmd; |
fc17b653 | 883 | } |
a4aea562 | 884 | |
4aedb705 CH |
885 | if (blk_integrity_rq(req)) { |
886 | ret = nvme_map_metadata(dev, req, &cmnd); | |
887 | if (ret) | |
888 | goto out_unmap_data; | |
889 | } | |
890 | ||
aae239e1 | 891 | blk_mq_start_request(req); |
04f3eafd | 892 | nvme_submit_cmd(nvmeq, &cmnd, bd->last); |
fc17b653 | 893 | return BLK_STS_OK; |
4aedb705 CH |
894 | out_unmap_data: |
895 | nvme_unmap_data(dev, req); | |
f9d03f96 CH |
896 | out_free_cmd: |
897 | nvme_cleanup_cmd(req); | |
ba1ca37e | 898 | return ret; |
b60503ba | 899 | } |
e1e5e564 | 900 | |
77f02a7a | 901 | static void nvme_pci_complete_rq(struct request *req) |
eee417b0 | 902 | { |
f4800d6d | 903 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
4aedb705 | 904 | struct nvme_dev *dev = iod->nvmeq->dev; |
a4aea562 | 905 | |
4aedb705 CH |
906 | if (blk_integrity_rq(req)) |
907 | dma_unmap_page(dev->dev, iod->meta_dma, | |
908 | rq_integrity_vec(req)->bv_len, rq_data_dir(req)); | |
b15c592d | 909 | if (blk_rq_nr_phys_segments(req)) |
4aedb705 | 910 | nvme_unmap_data(dev, req); |
77f02a7a | 911 | nvme_complete_rq(req); |
b60503ba MW |
912 | } |
913 | ||
d783e0bd | 914 | /* We read the CQE phase first to check if the rest of the entry is valid */ |
750dde44 | 915 | static inline bool nvme_cqe_pending(struct nvme_queue *nvmeq) |
d783e0bd | 916 | { |
74943d45 KB |
917 | struct nvme_completion *hcqe = &nvmeq->cqes[nvmeq->cq_head]; |
918 | ||
919 | return (le16_to_cpu(READ_ONCE(hcqe->status)) & 1) == nvmeq->cq_phase; | |
d783e0bd MR |
920 | } |
921 | ||
eb281c82 | 922 | static inline void nvme_ring_cq_doorbell(struct nvme_queue *nvmeq) |
b60503ba | 923 | { |
eb281c82 | 924 | u16 head = nvmeq->cq_head; |
adf68f21 | 925 | |
397c699f KB |
926 | if (nvme_dbbuf_update_and_check_event(head, nvmeq->dbbuf_cq_db, |
927 | nvmeq->dbbuf_cq_ei)) | |
928 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); | |
eb281c82 | 929 | } |
aae239e1 | 930 | |
cfa27356 CH |
931 | static inline struct blk_mq_tags *nvme_queue_tagset(struct nvme_queue *nvmeq) |
932 | { | |
933 | if (!nvmeq->qid) | |
934 | return nvmeq->dev->admin_tagset.tags[0]; | |
935 | return nvmeq->dev->tagset.tags[nvmeq->qid - 1]; | |
936 | } | |
937 | ||
5cb525c8 | 938 | static inline void nvme_handle_cqe(struct nvme_queue *nvmeq, u16 idx) |
83a12fb7 | 939 | { |
74943d45 | 940 | struct nvme_completion *cqe = &nvmeq->cqes[idx]; |
83a12fb7 | 941 | struct request *req; |
adf68f21 | 942 | |
83a12fb7 SG |
943 | if (unlikely(cqe->command_id >= nvmeq->q_depth)) { |
944 | dev_warn(nvmeq->dev->ctrl.device, | |
945 | "invalid id %d completed on queue %d\n", | |
946 | cqe->command_id, le16_to_cpu(cqe->sq_id)); | |
947 | return; | |
b60503ba MW |
948 | } |
949 | ||
83a12fb7 SG |
950 | /* |
951 | * AEN requests are special as they don't time out and can | |
952 | * survive any kind of queue freeze and often don't respond to | |
953 | * aborts. We don't even bother to allocate a struct request | |
954 | * for them but rather special case them here. | |
955 | */ | |
58a8df67 | 956 | if (unlikely(nvme_is_aen_req(nvmeq->qid, cqe->command_id))) { |
83a12fb7 SG |
957 | nvme_complete_async_event(&nvmeq->dev->ctrl, |
958 | cqe->status, &cqe->result); | |
a0fa9647 | 959 | return; |
83a12fb7 | 960 | } |
b60503ba | 961 | |
cfa27356 | 962 | req = blk_mq_tag_to_rq(nvme_queue_tagset(nvmeq), cqe->command_id); |
604c01d5 | 963 | trace_nvme_sq(req, cqe->sq_head, nvmeq->sq_tail); |
ff029451 CH |
964 | if (!nvme_end_request(req, cqe->status, cqe->result)) |
965 | nvme_pci_complete_rq(req); | |
83a12fb7 | 966 | } |
b60503ba | 967 | |
5cb525c8 JA |
968 | static inline void nvme_update_cq_head(struct nvme_queue *nvmeq) |
969 | { | |
a8de6639 AD |
970 | u16 tmp = nvmeq->cq_head + 1; |
971 | ||
972 | if (tmp == nvmeq->q_depth) { | |
5cb525c8 | 973 | nvmeq->cq_head = 0; |
e2a366a4 | 974 | nvmeq->cq_phase ^= 1; |
a8de6639 AD |
975 | } else { |
976 | nvmeq->cq_head = tmp; | |
b60503ba | 977 | } |
a0fa9647 JA |
978 | } |
979 | ||
324b494c | 980 | static inline int nvme_process_cq(struct nvme_queue *nvmeq) |
a0fa9647 | 981 | { |
1052b8ac | 982 | int found = 0; |
b60503ba | 983 | |
1052b8ac | 984 | while (nvme_cqe_pending(nvmeq)) { |
bf392a5d | 985 | found++; |
b69e2ef2 KB |
986 | /* |
987 | * load-load control dependency between phase and the rest of | |
988 | * the cqe requires a full read memory barrier | |
989 | */ | |
990 | dma_rmb(); | |
324b494c | 991 | nvme_handle_cqe(nvmeq, nvmeq->cq_head); |
5cb525c8 | 992 | nvme_update_cq_head(nvmeq); |
920d13a8 | 993 | } |
eb281c82 | 994 | |
324b494c | 995 | if (found) |
920d13a8 | 996 | nvme_ring_cq_doorbell(nvmeq); |
5cb525c8 | 997 | return found; |
b60503ba MW |
998 | } |
999 | ||
1000 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 | 1001 | { |
58ffacb5 | 1002 | struct nvme_queue *nvmeq = data; |
68fa9dbe | 1003 | irqreturn_t ret = IRQ_NONE; |
5cb525c8 | 1004 | |
3a7afd8e CH |
1005 | /* |
1006 | * The rmb/wmb pair ensures we see all updates from a previous run of | |
1007 | * the irq handler, even if that was on another CPU. | |
1008 | */ | |
1009 | rmb(); | |
324b494c KB |
1010 | if (nvme_process_cq(nvmeq)) |
1011 | ret = IRQ_HANDLED; | |
3a7afd8e | 1012 | wmb(); |
5cb525c8 | 1013 | |
68fa9dbe | 1014 | return ret; |
58ffacb5 MW |
1015 | } |
1016 | ||
1017 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
1018 | { | |
1019 | struct nvme_queue *nvmeq = data; | |
4e523547 | 1020 | |
750dde44 | 1021 | if (nvme_cqe_pending(nvmeq)) |
d783e0bd MR |
1022 | return IRQ_WAKE_THREAD; |
1023 | return IRQ_NONE; | |
58ffacb5 MW |
1024 | } |
1025 | ||
0b2a8a9f | 1026 | /* |
fa059b85 | 1027 | * Poll for completions for any interrupt driven queue |
0b2a8a9f CH |
1028 | * Can be called from any context. |
1029 | */ | |
fa059b85 | 1030 | static void nvme_poll_irqdisable(struct nvme_queue *nvmeq) |
a0fa9647 | 1031 | { |
3a7afd8e | 1032 | struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev); |
a0fa9647 | 1033 | |
fa059b85 | 1034 | WARN_ON_ONCE(test_bit(NVMEQ_POLLED, &nvmeq->flags)); |
442e19b7 | 1035 | |
fa059b85 KB |
1036 | disable_irq(pci_irq_vector(pdev, nvmeq->cq_vector)); |
1037 | nvme_process_cq(nvmeq); | |
1038 | enable_irq(pci_irq_vector(pdev, nvmeq->cq_vector)); | |
a0fa9647 JA |
1039 | } |
1040 | ||
9743139c | 1041 | static int nvme_poll(struct blk_mq_hw_ctx *hctx) |
dabcefab JA |
1042 | { |
1043 | struct nvme_queue *nvmeq = hctx->driver_data; | |
dabcefab JA |
1044 | bool found; |
1045 | ||
1046 | if (!nvme_cqe_pending(nvmeq)) | |
1047 | return 0; | |
1048 | ||
3a7afd8e | 1049 | spin_lock(&nvmeq->cq_poll_lock); |
324b494c | 1050 | found = nvme_process_cq(nvmeq); |
3a7afd8e | 1051 | spin_unlock(&nvmeq->cq_poll_lock); |
dabcefab | 1052 | |
dabcefab JA |
1053 | return found; |
1054 | } | |
1055 | ||
ad22c355 | 1056 | static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl) |
b60503ba | 1057 | { |
f866fc42 | 1058 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
147b27e4 | 1059 | struct nvme_queue *nvmeq = &dev->queues[0]; |
a4aea562 | 1060 | struct nvme_command c; |
b60503ba | 1061 | |
a4aea562 MB |
1062 | memset(&c, 0, sizeof(c)); |
1063 | c.common.opcode = nvme_admin_async_event; | |
ad22c355 | 1064 | c.common.command_id = NVME_AQ_BLK_MQ_DEPTH; |
04f3eafd | 1065 | nvme_submit_cmd(nvmeq, &c, true); |
f705f837 CH |
1066 | } |
1067 | ||
b60503ba | 1068 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
f705f837 | 1069 | { |
b60503ba MW |
1070 | struct nvme_command c; |
1071 | ||
1072 | memset(&c, 0, sizeof(c)); | |
1073 | c.delete_queue.opcode = opcode; | |
1074 | c.delete_queue.qid = cpu_to_le16(id); | |
1075 | ||
1c63dc66 | 1076 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
1077 | } |
1078 | ||
b60503ba | 1079 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, |
a8e3e0bb | 1080 | struct nvme_queue *nvmeq, s16 vector) |
b60503ba | 1081 | { |
b60503ba | 1082 | struct nvme_command c; |
4b04cc6a JA |
1083 | int flags = NVME_QUEUE_PHYS_CONTIG; |
1084 | ||
7c349dde | 1085 | if (!test_bit(NVMEQ_POLLED, &nvmeq->flags)) |
4b04cc6a | 1086 | flags |= NVME_CQ_IRQ_ENABLED; |
b60503ba | 1087 | |
d29ec824 | 1088 | /* |
16772ae6 | 1089 | * Note: we (ab)use the fact that the prp fields survive if no data |
d29ec824 CH |
1090 | * is attached to the request. |
1091 | */ | |
b60503ba MW |
1092 | memset(&c, 0, sizeof(c)); |
1093 | c.create_cq.opcode = nvme_admin_create_cq; | |
1094 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
1095 | c.create_cq.cqid = cpu_to_le16(qid); | |
1096 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1097 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
7c349dde | 1098 | c.create_cq.irq_vector = cpu_to_le16(vector); |
b60503ba | 1099 | |
1c63dc66 | 1100 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
1101 | } |
1102 | ||
1103 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
1104 | struct nvme_queue *nvmeq) | |
1105 | { | |
9abd68ef | 1106 | struct nvme_ctrl *ctrl = &dev->ctrl; |
b60503ba | 1107 | struct nvme_command c; |
81c1cd98 | 1108 | int flags = NVME_QUEUE_PHYS_CONTIG; |
b60503ba | 1109 | |
9abd68ef JA |
1110 | /* |
1111 | * Some drives have a bug that auto-enables WRRU if MEDIUM isn't | |
1112 | * set. Since URGENT priority is zeroes, it makes all queues | |
1113 | * URGENT. | |
1114 | */ | |
1115 | if (ctrl->quirks & NVME_QUIRK_MEDIUM_PRIO_SQ) | |
1116 | flags |= NVME_SQ_PRIO_MEDIUM; | |
1117 | ||
d29ec824 | 1118 | /* |
16772ae6 | 1119 | * Note: we (ab)use the fact that the prp fields survive if no data |
d29ec824 CH |
1120 | * is attached to the request. |
1121 | */ | |
b60503ba MW |
1122 | memset(&c, 0, sizeof(c)); |
1123 | c.create_sq.opcode = nvme_admin_create_sq; | |
1124 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
1125 | c.create_sq.sqid = cpu_to_le16(qid); | |
1126 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1127 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
1128 | c.create_sq.cqid = cpu_to_le16(qid); | |
1129 | ||
1c63dc66 | 1130 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
1131 | } |
1132 | ||
1133 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
1134 | { | |
1135 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
1136 | } | |
1137 | ||
1138 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
1139 | { | |
1140 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
1141 | } | |
1142 | ||
2a842aca | 1143 | static void abort_endio(struct request *req, blk_status_t error) |
bc5fc7e4 | 1144 | { |
f4800d6d CH |
1145 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
1146 | struct nvme_queue *nvmeq = iod->nvmeq; | |
e44ac588 | 1147 | |
27fa9bc5 CH |
1148 | dev_warn(nvmeq->dev->ctrl.device, |
1149 | "Abort status: 0x%x", nvme_req(req)->status); | |
e7a2a87d | 1150 | atomic_inc(&nvmeq->dev->ctrl.abort_limit); |
e7a2a87d | 1151 | blk_mq_free_request(req); |
bc5fc7e4 MW |
1152 | } |
1153 | ||
b2a0eb1a KB |
1154 | static bool nvme_should_reset(struct nvme_dev *dev, u32 csts) |
1155 | { | |
b2a0eb1a KB |
1156 | /* If true, indicates loss of adapter communication, possibly by a |
1157 | * NVMe Subsystem reset. | |
1158 | */ | |
1159 | bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO); | |
1160 | ||
ad70062c JW |
1161 | /* If there is a reset/reinit ongoing, we shouldn't reset again. */ |
1162 | switch (dev->ctrl.state) { | |
1163 | case NVME_CTRL_RESETTING: | |
ad6a0a52 | 1164 | case NVME_CTRL_CONNECTING: |
b2a0eb1a | 1165 | return false; |
ad70062c JW |
1166 | default: |
1167 | break; | |
1168 | } | |
b2a0eb1a KB |
1169 | |
1170 | /* We shouldn't reset unless the controller is on fatal error state | |
1171 | * _or_ if we lost the communication with it. | |
1172 | */ | |
1173 | if (!(csts & NVME_CSTS_CFS) && !nssro) | |
1174 | return false; | |
1175 | ||
b2a0eb1a KB |
1176 | return true; |
1177 | } | |
1178 | ||
1179 | static void nvme_warn_reset(struct nvme_dev *dev, u32 csts) | |
1180 | { | |
1181 | /* Read a config register to help see what died. */ | |
1182 | u16 pci_status; | |
1183 | int result; | |
1184 | ||
1185 | result = pci_read_config_word(to_pci_dev(dev->dev), PCI_STATUS, | |
1186 | &pci_status); | |
1187 | if (result == PCIBIOS_SUCCESSFUL) | |
1188 | dev_warn(dev->ctrl.device, | |
1189 | "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n", | |
1190 | csts, pci_status); | |
1191 | else | |
1192 | dev_warn(dev->ctrl.device, | |
1193 | "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n", | |
1194 | csts, result); | |
1195 | } | |
1196 | ||
31c7c7d2 | 1197 | static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) |
c30341dc | 1198 | { |
f4800d6d CH |
1199 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
1200 | struct nvme_queue *nvmeq = iod->nvmeq; | |
c30341dc | 1201 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 | 1202 | struct request *abort_req; |
a4aea562 | 1203 | struct nvme_command cmd; |
b2a0eb1a KB |
1204 | u32 csts = readl(dev->bar + NVME_REG_CSTS); |
1205 | ||
651438bb WX |
1206 | /* If PCI error recovery process is happening, we cannot reset or |
1207 | * the recovery mechanism will surely fail. | |
1208 | */ | |
1209 | mb(); | |
1210 | if (pci_channel_offline(to_pci_dev(dev->dev))) | |
1211 | return BLK_EH_RESET_TIMER; | |
1212 | ||
b2a0eb1a KB |
1213 | /* |
1214 | * Reset immediately if the controller is failed | |
1215 | */ | |
1216 | if (nvme_should_reset(dev, csts)) { | |
1217 | nvme_warn_reset(dev, csts); | |
1218 | nvme_dev_disable(dev, false); | |
d86c4d8e | 1219 | nvme_reset_ctrl(&dev->ctrl); |
db8c48e4 | 1220 | return BLK_EH_DONE; |
b2a0eb1a | 1221 | } |
c30341dc | 1222 | |
7776db1c KB |
1223 | /* |
1224 | * Did we miss an interrupt? | |
1225 | */ | |
fa059b85 KB |
1226 | if (test_bit(NVMEQ_POLLED, &nvmeq->flags)) |
1227 | nvme_poll(req->mq_hctx); | |
1228 | else | |
1229 | nvme_poll_irqdisable(nvmeq); | |
1230 | ||
bf392a5d | 1231 | if (blk_mq_request_completed(req)) { |
7776db1c KB |
1232 | dev_warn(dev->ctrl.device, |
1233 | "I/O %d QID %d timeout, completion polled\n", | |
1234 | req->tag, nvmeq->qid); | |
db8c48e4 | 1235 | return BLK_EH_DONE; |
7776db1c KB |
1236 | } |
1237 | ||
31c7c7d2 | 1238 | /* |
fd634f41 CH |
1239 | * Shutdown immediately if controller times out while starting. The |
1240 | * reset work will see the pci device disabled when it gets the forced | |
1241 | * cancellation error. All outstanding requests are completed on | |
db8c48e4 | 1242 | * shutdown, so we return BLK_EH_DONE. |
fd634f41 | 1243 | */ |
4244140d KB |
1244 | switch (dev->ctrl.state) { |
1245 | case NVME_CTRL_CONNECTING: | |
2036f726 KB |
1246 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); |
1247 | /* fall through */ | |
1248 | case NVME_CTRL_DELETING: | |
b9cac43c | 1249 | dev_warn_ratelimited(dev->ctrl.device, |
fd634f41 CH |
1250 | "I/O %d QID %d timeout, disable controller\n", |
1251 | req->tag, nvmeq->qid); | |
2036f726 | 1252 | nvme_dev_disable(dev, true); |
27fa9bc5 | 1253 | nvme_req(req)->flags |= NVME_REQ_CANCELLED; |
db8c48e4 | 1254 | return BLK_EH_DONE; |
39a9dd81 KB |
1255 | case NVME_CTRL_RESETTING: |
1256 | return BLK_EH_RESET_TIMER; | |
4244140d KB |
1257 | default: |
1258 | break; | |
c30341dc KB |
1259 | } |
1260 | ||
fd634f41 | 1261 | /* |
ee0d96d3 BW |
1262 | * Shutdown the controller immediately and schedule a reset if the |
1263 | * command was already aborted once before and still hasn't been | |
1264 | * returned to the driver, or if this is the admin queue. | |
31c7c7d2 | 1265 | */ |
f4800d6d | 1266 | if (!nvmeq->qid || iod->aborted) { |
1b3c47c1 | 1267 | dev_warn(dev->ctrl.device, |
e1569a16 KB |
1268 | "I/O %d QID %d timeout, reset controller\n", |
1269 | req->tag, nvmeq->qid); | |
a5cdb68c | 1270 | nvme_dev_disable(dev, false); |
d86c4d8e | 1271 | nvme_reset_ctrl(&dev->ctrl); |
c30341dc | 1272 | |
27fa9bc5 | 1273 | nvme_req(req)->flags |= NVME_REQ_CANCELLED; |
db8c48e4 | 1274 | return BLK_EH_DONE; |
c30341dc | 1275 | } |
c30341dc | 1276 | |
e7a2a87d | 1277 | if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) { |
6bf25d16 | 1278 | atomic_inc(&dev->ctrl.abort_limit); |
31c7c7d2 | 1279 | return BLK_EH_RESET_TIMER; |
6bf25d16 | 1280 | } |
7bf7d778 | 1281 | iod->aborted = 1; |
a4aea562 | 1282 | |
c30341dc KB |
1283 | memset(&cmd, 0, sizeof(cmd)); |
1284 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
a4aea562 | 1285 | cmd.abort.cid = req->tag; |
c30341dc | 1286 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); |
c30341dc | 1287 | |
1b3c47c1 SG |
1288 | dev_warn(nvmeq->dev->ctrl.device, |
1289 | "I/O %d QID %d timeout, aborting\n", | |
1290 | req->tag, nvmeq->qid); | |
e7a2a87d CH |
1291 | |
1292 | abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd, | |
eb71f435 | 1293 | BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); |
e7a2a87d CH |
1294 | if (IS_ERR(abort_req)) { |
1295 | atomic_inc(&dev->ctrl.abort_limit); | |
1296 | return BLK_EH_RESET_TIMER; | |
1297 | } | |
1298 | ||
1299 | abort_req->timeout = ADMIN_TIMEOUT; | |
1300 | abort_req->end_io_data = NULL; | |
1301 | blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio); | |
c30341dc | 1302 | |
31c7c7d2 CH |
1303 | /* |
1304 | * The aborted req will be completed on receiving the abort req. | |
1305 | * We enable the timer again. If hit twice, it'll cause a device reset, | |
1306 | * as the device then is in a faulty state. | |
1307 | */ | |
1308 | return BLK_EH_RESET_TIMER; | |
c30341dc KB |
1309 | } |
1310 | ||
a4aea562 MB |
1311 | static void nvme_free_queue(struct nvme_queue *nvmeq) |
1312 | { | |
8a1d09a6 | 1313 | dma_free_coherent(nvmeq->dev->dev, CQ_SIZE(nvmeq), |
9e866774 | 1314 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); |
63223078 CH |
1315 | if (!nvmeq->sq_cmds) |
1316 | return; | |
0f238ff5 | 1317 | |
63223078 | 1318 | if (test_and_clear_bit(NVMEQ_SQ_CMB, &nvmeq->flags)) { |
88a041f4 | 1319 | pci_free_p2pmem(to_pci_dev(nvmeq->dev->dev), |
8a1d09a6 | 1320 | nvmeq->sq_cmds, SQ_SIZE(nvmeq)); |
63223078 | 1321 | } else { |
8a1d09a6 | 1322 | dma_free_coherent(nvmeq->dev->dev, SQ_SIZE(nvmeq), |
63223078 | 1323 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); |
0f238ff5 | 1324 | } |
9e866774 MW |
1325 | } |
1326 | ||
a1a5ef99 | 1327 | static void nvme_free_queues(struct nvme_dev *dev, int lowest) |
22404274 KB |
1328 | { |
1329 | int i; | |
1330 | ||
d858e5f0 | 1331 | for (i = dev->ctrl.queue_count - 1; i >= lowest; i--) { |
d858e5f0 | 1332 | dev->ctrl.queue_count--; |
147b27e4 | 1333 | nvme_free_queue(&dev->queues[i]); |
121c7ad4 | 1334 | } |
22404274 KB |
1335 | } |
1336 | ||
4d115420 KB |
1337 | /** |
1338 | * nvme_suspend_queue - put queue into suspended state | |
40581d1a | 1339 | * @nvmeq: queue to suspend |
4d115420 KB |
1340 | */ |
1341 | static int nvme_suspend_queue(struct nvme_queue *nvmeq) | |
b60503ba | 1342 | { |
4e224106 | 1343 | if (!test_and_clear_bit(NVMEQ_ENABLED, &nvmeq->flags)) |
2b25d981 | 1344 | return 1; |
a09115b2 | 1345 | |
4e224106 | 1346 | /* ensure that nvme_queue_rq() sees NVMEQ_ENABLED cleared */ |
d1f06f4a | 1347 | mb(); |
a09115b2 | 1348 | |
4e224106 | 1349 | nvmeq->dev->online_queues--; |
1c63dc66 | 1350 | if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q) |
c81545f9 | 1351 | blk_mq_quiesce_queue(nvmeq->dev->ctrl.admin_q); |
7c349dde KB |
1352 | if (!test_and_clear_bit(NVMEQ_POLLED, &nvmeq->flags)) |
1353 | pci_free_irq(to_pci_dev(nvmeq->dev->dev), nvmeq->cq_vector, nvmeq); | |
4d115420 KB |
1354 | return 0; |
1355 | } | |
b60503ba | 1356 | |
8fae268b KB |
1357 | static void nvme_suspend_io_queues(struct nvme_dev *dev) |
1358 | { | |
1359 | int i; | |
1360 | ||
1361 | for (i = dev->ctrl.queue_count - 1; i > 0; i--) | |
1362 | nvme_suspend_queue(&dev->queues[i]); | |
1363 | } | |
1364 | ||
a5cdb68c | 1365 | static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown) |
4d115420 | 1366 | { |
147b27e4 | 1367 | struct nvme_queue *nvmeq = &dev->queues[0]; |
4d115420 | 1368 | |
a5cdb68c KB |
1369 | if (shutdown) |
1370 | nvme_shutdown_ctrl(&dev->ctrl); | |
1371 | else | |
b5b05048 | 1372 | nvme_disable_ctrl(&dev->ctrl); |
07836e65 | 1373 | |
bf392a5d | 1374 | nvme_poll_irqdisable(nvmeq); |
b60503ba MW |
1375 | } |
1376 | ||
fa46c6fb KB |
1377 | /* |
1378 | * Called only on a device that has been disabled and after all other threads | |
9210c075 DZ |
1379 | * that can check this device's completion queues have synced, except |
1380 | * nvme_poll(). This is the last chance for the driver to see a natural | |
1381 | * completion before nvme_cancel_request() terminates all incomplete requests. | |
fa46c6fb KB |
1382 | */ |
1383 | static void nvme_reap_pending_cqes(struct nvme_dev *dev) | |
1384 | { | |
fa46c6fb KB |
1385 | int i; |
1386 | ||
9210c075 DZ |
1387 | for (i = dev->ctrl.queue_count - 1; i > 0; i--) { |
1388 | spin_lock(&dev->queues[i].cq_poll_lock); | |
324b494c | 1389 | nvme_process_cq(&dev->queues[i]); |
9210c075 DZ |
1390 | spin_unlock(&dev->queues[i].cq_poll_lock); |
1391 | } | |
fa46c6fb KB |
1392 | } |
1393 | ||
8ffaadf7 JD |
1394 | static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, |
1395 | int entry_size) | |
1396 | { | |
1397 | int q_depth = dev->q_depth; | |
5fd4ce1b CH |
1398 | unsigned q_size_aligned = roundup(q_depth * entry_size, |
1399 | dev->ctrl.page_size); | |
8ffaadf7 JD |
1400 | |
1401 | if (q_size_aligned * nr_io_queues > dev->cmb_size) { | |
c45f5c99 | 1402 | u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); |
4e523547 | 1403 | |
5fd4ce1b | 1404 | mem_per_q = round_down(mem_per_q, dev->ctrl.page_size); |
c45f5c99 | 1405 | q_depth = div_u64(mem_per_q, entry_size); |
8ffaadf7 JD |
1406 | |
1407 | /* | |
1408 | * Ensure the reduced q_depth is above some threshold where it | |
1409 | * would be better to map queues in system memory with the | |
1410 | * original depth | |
1411 | */ | |
1412 | if (q_depth < 64) | |
1413 | return -ENOMEM; | |
1414 | } | |
1415 | ||
1416 | return q_depth; | |
1417 | } | |
1418 | ||
1419 | static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, | |
8a1d09a6 | 1420 | int qid) |
8ffaadf7 | 1421 | { |
0f238ff5 LG |
1422 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
1423 | ||
1424 | if (qid && dev->cmb_use_sqes && (dev->cmbsz & NVME_CMBSZ_SQS)) { | |
8a1d09a6 | 1425 | nvmeq->sq_cmds = pci_alloc_p2pmem(pdev, SQ_SIZE(nvmeq)); |
bfac8e9f AM |
1426 | if (nvmeq->sq_cmds) { |
1427 | nvmeq->sq_dma_addr = pci_p2pmem_virt_to_bus(pdev, | |
1428 | nvmeq->sq_cmds); | |
1429 | if (nvmeq->sq_dma_addr) { | |
1430 | set_bit(NVMEQ_SQ_CMB, &nvmeq->flags); | |
1431 | return 0; | |
1432 | } | |
1433 | ||
8a1d09a6 | 1434 | pci_free_p2pmem(pdev, nvmeq->sq_cmds, SQ_SIZE(nvmeq)); |
63223078 | 1435 | } |
0f238ff5 | 1436 | } |
8ffaadf7 | 1437 | |
8a1d09a6 | 1438 | nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(nvmeq), |
63223078 | 1439 | &nvmeq->sq_dma_addr, GFP_KERNEL); |
815c6704 KB |
1440 | if (!nvmeq->sq_cmds) |
1441 | return -ENOMEM; | |
8ffaadf7 JD |
1442 | return 0; |
1443 | } | |
1444 | ||
a6ff7262 | 1445 | static int nvme_alloc_queue(struct nvme_dev *dev, int qid, int depth) |
b60503ba | 1446 | { |
147b27e4 | 1447 | struct nvme_queue *nvmeq = &dev->queues[qid]; |
b60503ba | 1448 | |
62314e40 KB |
1449 | if (dev->ctrl.queue_count > qid) |
1450 | return 0; | |
b60503ba | 1451 | |
c1e0cc7e | 1452 | nvmeq->sqes = qid ? dev->io_sqes : NVME_ADM_SQES; |
8a1d09a6 BH |
1453 | nvmeq->q_depth = depth; |
1454 | nvmeq->cqes = dma_alloc_coherent(dev->dev, CQ_SIZE(nvmeq), | |
750afb08 | 1455 | &nvmeq->cq_dma_addr, GFP_KERNEL); |
b60503ba MW |
1456 | if (!nvmeq->cqes) |
1457 | goto free_nvmeq; | |
b60503ba | 1458 | |
8a1d09a6 | 1459 | if (nvme_alloc_sq_cmds(dev, nvmeq, qid)) |
b60503ba MW |
1460 | goto free_cqdma; |
1461 | ||
091b6092 | 1462 | nvmeq->dev = dev; |
1ab0cd69 | 1463 | spin_lock_init(&nvmeq->sq_lock); |
3a7afd8e | 1464 | spin_lock_init(&nvmeq->cq_poll_lock); |
b60503ba | 1465 | nvmeq->cq_head = 0; |
82123460 | 1466 | nvmeq->cq_phase = 1; |
b80d5ccc | 1467 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
c30341dc | 1468 | nvmeq->qid = qid; |
d858e5f0 | 1469 | dev->ctrl.queue_count++; |
36a7e993 | 1470 | |
147b27e4 | 1471 | return 0; |
b60503ba MW |
1472 | |
1473 | free_cqdma: | |
8a1d09a6 BH |
1474 | dma_free_coherent(dev->dev, CQ_SIZE(nvmeq), (void *)nvmeq->cqes, |
1475 | nvmeq->cq_dma_addr); | |
b60503ba | 1476 | free_nvmeq: |
147b27e4 | 1477 | return -ENOMEM; |
b60503ba MW |
1478 | } |
1479 | ||
dca51e78 | 1480 | static int queue_request_irq(struct nvme_queue *nvmeq) |
3001082c | 1481 | { |
0ff199cb CH |
1482 | struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev); |
1483 | int nr = nvmeq->dev->ctrl.instance; | |
1484 | ||
1485 | if (use_threaded_interrupts) { | |
1486 | return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq_check, | |
1487 | nvme_irq, nvmeq, "nvme%dq%d", nr, nvmeq->qid); | |
1488 | } else { | |
1489 | return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq, | |
1490 | NULL, nvmeq, "nvme%dq%d", nr, nvmeq->qid); | |
1491 | } | |
3001082c MW |
1492 | } |
1493 | ||
22404274 | 1494 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1495 | { |
22404274 | 1496 | struct nvme_dev *dev = nvmeq->dev; |
b60503ba | 1497 | |
22404274 KB |
1498 | nvmeq->sq_tail = 0; |
1499 | nvmeq->cq_head = 0; | |
1500 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1501 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
8a1d09a6 | 1502 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq)); |
f9f38e33 | 1503 | nvme_dbbuf_init(dev, nvmeq, qid); |
42f61420 | 1504 | dev->online_queues++; |
3a7afd8e | 1505 | wmb(); /* ensure the first interrupt sees the initialization */ |
22404274 KB |
1506 | } |
1507 | ||
4b04cc6a | 1508 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled) |
22404274 KB |
1509 | { |
1510 | struct nvme_dev *dev = nvmeq->dev; | |
1511 | int result; | |
7c349dde | 1512 | u16 vector = 0; |
3f85d50b | 1513 | |
d1ed6aa1 CH |
1514 | clear_bit(NVMEQ_DELETE_ERROR, &nvmeq->flags); |
1515 | ||
22b55601 KB |
1516 | /* |
1517 | * A queue's vector matches the queue identifier unless the controller | |
1518 | * has only one vector available. | |
1519 | */ | |
4b04cc6a JA |
1520 | if (!polled) |
1521 | vector = dev->num_vecs == 1 ? 0 : qid; | |
1522 | else | |
7c349dde | 1523 | set_bit(NVMEQ_POLLED, &nvmeq->flags); |
4b04cc6a | 1524 | |
a8e3e0bb | 1525 | result = adapter_alloc_cq(dev, qid, nvmeq, vector); |
ded45505 KB |
1526 | if (result) |
1527 | return result; | |
b60503ba MW |
1528 | |
1529 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1530 | if (result < 0) | |
ded45505 | 1531 | return result; |
c80b36cd | 1532 | if (result) |
b60503ba MW |
1533 | goto release_cq; |
1534 | ||
a8e3e0bb | 1535 | nvmeq->cq_vector = vector; |
161b8be2 | 1536 | nvme_init_queue(nvmeq, qid); |
4b04cc6a | 1537 | |
7c349dde | 1538 | if (!polled) { |
4b04cc6a JA |
1539 | result = queue_request_irq(nvmeq); |
1540 | if (result < 0) | |
1541 | goto release_sq; | |
1542 | } | |
b60503ba | 1543 | |
4e224106 | 1544 | set_bit(NVMEQ_ENABLED, &nvmeq->flags); |
22404274 | 1545 | return result; |
b60503ba | 1546 | |
a8e3e0bb | 1547 | release_sq: |
f25a2dfc | 1548 | dev->online_queues--; |
b60503ba | 1549 | adapter_delete_sq(dev, qid); |
a8e3e0bb | 1550 | release_cq: |
b60503ba | 1551 | adapter_delete_cq(dev, qid); |
22404274 | 1552 | return result; |
b60503ba MW |
1553 | } |
1554 | ||
f363b089 | 1555 | static const struct blk_mq_ops nvme_mq_admin_ops = { |
d29ec824 | 1556 | .queue_rq = nvme_queue_rq, |
77f02a7a | 1557 | .complete = nvme_pci_complete_rq, |
a4aea562 | 1558 | .init_hctx = nvme_admin_init_hctx, |
0350815a | 1559 | .init_request = nvme_init_request, |
a4aea562 MB |
1560 | .timeout = nvme_timeout, |
1561 | }; | |
1562 | ||
f363b089 | 1563 | static const struct blk_mq_ops nvme_mq_ops = { |
376f7ef8 CH |
1564 | .queue_rq = nvme_queue_rq, |
1565 | .complete = nvme_pci_complete_rq, | |
1566 | .commit_rqs = nvme_commit_rqs, | |
1567 | .init_hctx = nvme_init_hctx, | |
1568 | .init_request = nvme_init_request, | |
1569 | .map_queues = nvme_pci_map_queues, | |
1570 | .timeout = nvme_timeout, | |
1571 | .poll = nvme_poll, | |
dabcefab JA |
1572 | }; |
1573 | ||
ea191d2f KB |
1574 | static void nvme_dev_remove_admin(struct nvme_dev *dev) |
1575 | { | |
1c63dc66 | 1576 | if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) { |
69d9a99c KB |
1577 | /* |
1578 | * If the controller was reset during removal, it's possible | |
1579 | * user requests may be waiting on a stopped queue. Start the | |
1580 | * queue to flush these to completion. | |
1581 | */ | |
c81545f9 | 1582 | blk_mq_unquiesce_queue(dev->ctrl.admin_q); |
1c63dc66 | 1583 | blk_cleanup_queue(dev->ctrl.admin_q); |
ea191d2f KB |
1584 | blk_mq_free_tag_set(&dev->admin_tagset); |
1585 | } | |
1586 | } | |
1587 | ||
a4aea562 MB |
1588 | static int nvme_alloc_admin_tags(struct nvme_dev *dev) |
1589 | { | |
1c63dc66 | 1590 | if (!dev->ctrl.admin_q) { |
a4aea562 MB |
1591 | dev->admin_tagset.ops = &nvme_mq_admin_ops; |
1592 | dev->admin_tagset.nr_hw_queues = 1; | |
e3e9d50c | 1593 | |
38dabe21 | 1594 | dev->admin_tagset.queue_depth = NVME_AQ_MQ_TAG_DEPTH; |
a4aea562 | 1595 | dev->admin_tagset.timeout = ADMIN_TIMEOUT; |
d4ec47f1 | 1596 | dev->admin_tagset.numa_node = dev->ctrl.numa_node; |
d43f1ccf | 1597 | dev->admin_tagset.cmd_size = sizeof(struct nvme_iod); |
d3484991 | 1598 | dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED; |
a4aea562 MB |
1599 | dev->admin_tagset.driver_data = dev; |
1600 | ||
1601 | if (blk_mq_alloc_tag_set(&dev->admin_tagset)) | |
1602 | return -ENOMEM; | |
34b6c231 | 1603 | dev->ctrl.admin_tagset = &dev->admin_tagset; |
a4aea562 | 1604 | |
1c63dc66 CH |
1605 | dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset); |
1606 | if (IS_ERR(dev->ctrl.admin_q)) { | |
a4aea562 MB |
1607 | blk_mq_free_tag_set(&dev->admin_tagset); |
1608 | return -ENOMEM; | |
1609 | } | |
1c63dc66 | 1610 | if (!blk_get_queue(dev->ctrl.admin_q)) { |
ea191d2f | 1611 | nvme_dev_remove_admin(dev); |
1c63dc66 | 1612 | dev->ctrl.admin_q = NULL; |
ea191d2f KB |
1613 | return -ENODEV; |
1614 | } | |
0fb59cbc | 1615 | } else |
c81545f9 | 1616 | blk_mq_unquiesce_queue(dev->ctrl.admin_q); |
a4aea562 MB |
1617 | |
1618 | return 0; | |
1619 | } | |
1620 | ||
97f6ef64 XY |
1621 | static unsigned long db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
1622 | { | |
1623 | return NVME_REG_DBS + ((nr_io_queues + 1) * 8 * dev->db_stride); | |
1624 | } | |
1625 | ||
1626 | static int nvme_remap_bar(struct nvme_dev *dev, unsigned long size) | |
1627 | { | |
1628 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
1629 | ||
1630 | if (size <= dev->bar_mapped_size) | |
1631 | return 0; | |
1632 | if (size > pci_resource_len(pdev, 0)) | |
1633 | return -ENOMEM; | |
1634 | if (dev->bar) | |
1635 | iounmap(dev->bar); | |
1636 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
1637 | if (!dev->bar) { | |
1638 | dev->bar_mapped_size = 0; | |
1639 | return -ENOMEM; | |
1640 | } | |
1641 | dev->bar_mapped_size = size; | |
1642 | dev->dbs = dev->bar + NVME_REG_DBS; | |
1643 | ||
1644 | return 0; | |
1645 | } | |
1646 | ||
01ad0990 | 1647 | static int nvme_pci_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1648 | { |
ba47e386 | 1649 | int result; |
b60503ba MW |
1650 | u32 aqa; |
1651 | struct nvme_queue *nvmeq; | |
1652 | ||
97f6ef64 XY |
1653 | result = nvme_remap_bar(dev, db_bar_size(dev, 0)); |
1654 | if (result < 0) | |
1655 | return result; | |
1656 | ||
8ef2074d | 1657 | dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ? |
20d0dfe6 | 1658 | NVME_CAP_NSSRC(dev->ctrl.cap) : 0; |
dfbac8c7 | 1659 | |
7a67cbea CH |
1660 | if (dev->subsystem && |
1661 | (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO)) | |
1662 | writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS); | |
dfbac8c7 | 1663 | |
b5b05048 | 1664 | result = nvme_disable_ctrl(&dev->ctrl); |
ba47e386 MW |
1665 | if (result < 0) |
1666 | return result; | |
b60503ba | 1667 | |
a6ff7262 | 1668 | result = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); |
147b27e4 SG |
1669 | if (result) |
1670 | return result; | |
b60503ba | 1671 | |
635333e4 MG |
1672 | dev->ctrl.numa_node = dev_to_node(dev->dev); |
1673 | ||
147b27e4 | 1674 | nvmeq = &dev->queues[0]; |
b60503ba MW |
1675 | aqa = nvmeq->q_depth - 1; |
1676 | aqa |= aqa << 16; | |
1677 | ||
7a67cbea CH |
1678 | writel(aqa, dev->bar + NVME_REG_AQA); |
1679 | lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ); | |
1680 | lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ); | |
b60503ba | 1681 | |
c0f2f45b | 1682 | result = nvme_enable_ctrl(&dev->ctrl); |
025c557a | 1683 | if (result) |
d4875622 | 1684 | return result; |
a4aea562 | 1685 | |
2b25d981 | 1686 | nvmeq->cq_vector = 0; |
161b8be2 | 1687 | nvme_init_queue(nvmeq, 0); |
dca51e78 | 1688 | result = queue_request_irq(nvmeq); |
758dd7fd | 1689 | if (result) { |
7c349dde | 1690 | dev->online_queues--; |
d4875622 | 1691 | return result; |
758dd7fd | 1692 | } |
025c557a | 1693 | |
4e224106 | 1694 | set_bit(NVMEQ_ENABLED, &nvmeq->flags); |
b60503ba MW |
1695 | return result; |
1696 | } | |
1697 | ||
749941f2 | 1698 | static int nvme_create_io_queues(struct nvme_dev *dev) |
42f61420 | 1699 | { |
4b04cc6a | 1700 | unsigned i, max, rw_queues; |
749941f2 | 1701 | int ret = 0; |
42f61420 | 1702 | |
d858e5f0 | 1703 | for (i = dev->ctrl.queue_count; i <= dev->max_qid; i++) { |
a6ff7262 | 1704 | if (nvme_alloc_queue(dev, i, dev->q_depth)) { |
749941f2 | 1705 | ret = -ENOMEM; |
42f61420 | 1706 | break; |
749941f2 CH |
1707 | } |
1708 | } | |
42f61420 | 1709 | |
d858e5f0 | 1710 | max = min(dev->max_qid, dev->ctrl.queue_count - 1); |
e20ba6e1 CH |
1711 | if (max != 1 && dev->io_queues[HCTX_TYPE_POLL]) { |
1712 | rw_queues = dev->io_queues[HCTX_TYPE_DEFAULT] + | |
1713 | dev->io_queues[HCTX_TYPE_READ]; | |
4b04cc6a JA |
1714 | } else { |
1715 | rw_queues = max; | |
1716 | } | |
1717 | ||
949928c1 | 1718 | for (i = dev->online_queues; i <= max; i++) { |
4b04cc6a JA |
1719 | bool polled = i > rw_queues; |
1720 | ||
1721 | ret = nvme_create_queue(&dev->queues[i], i, polled); | |
d4875622 | 1722 | if (ret) |
42f61420 | 1723 | break; |
27e8166c | 1724 | } |
749941f2 CH |
1725 | |
1726 | /* | |
1727 | * Ignore failing Create SQ/CQ commands, we can continue with less | |
8adb8c14 MI |
1728 | * than the desired amount of queues, and even a controller without |
1729 | * I/O queues can still be used to issue admin commands. This might | |
749941f2 CH |
1730 | * be useful to upgrade a buggy firmware for example. |
1731 | */ | |
1732 | return ret >= 0 ? 0 : ret; | |
b60503ba MW |
1733 | } |
1734 | ||
202021c1 SB |
1735 | static ssize_t nvme_cmb_show(struct device *dev, |
1736 | struct device_attribute *attr, | |
1737 | char *buf) | |
1738 | { | |
1739 | struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev)); | |
1740 | ||
c965809c | 1741 | return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz : x%08x\n", |
202021c1 SB |
1742 | ndev->cmbloc, ndev->cmbsz); |
1743 | } | |
1744 | static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL); | |
1745 | ||
88de4598 | 1746 | static u64 nvme_cmb_size_unit(struct nvme_dev *dev) |
8ffaadf7 | 1747 | { |
88de4598 CH |
1748 | u8 szu = (dev->cmbsz >> NVME_CMBSZ_SZU_SHIFT) & NVME_CMBSZ_SZU_MASK; |
1749 | ||
1750 | return 1ULL << (12 + 4 * szu); | |
1751 | } | |
1752 | ||
1753 | static u32 nvme_cmb_size(struct nvme_dev *dev) | |
1754 | { | |
1755 | return (dev->cmbsz >> NVME_CMBSZ_SZ_SHIFT) & NVME_CMBSZ_SZ_MASK; | |
1756 | } | |
1757 | ||
f65efd6d | 1758 | static void nvme_map_cmb(struct nvme_dev *dev) |
8ffaadf7 | 1759 | { |
88de4598 | 1760 | u64 size, offset; |
8ffaadf7 JD |
1761 | resource_size_t bar_size; |
1762 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
8969f1f8 | 1763 | int bar; |
8ffaadf7 | 1764 | |
9fe5c59f KB |
1765 | if (dev->cmb_size) |
1766 | return; | |
1767 | ||
7a67cbea | 1768 | dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ); |
f65efd6d CH |
1769 | if (!dev->cmbsz) |
1770 | return; | |
202021c1 | 1771 | dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC); |
8ffaadf7 | 1772 | |
88de4598 CH |
1773 | size = nvme_cmb_size_unit(dev) * nvme_cmb_size(dev); |
1774 | offset = nvme_cmb_size_unit(dev) * NVME_CMB_OFST(dev->cmbloc); | |
8969f1f8 CH |
1775 | bar = NVME_CMB_BIR(dev->cmbloc); |
1776 | bar_size = pci_resource_len(pdev, bar); | |
8ffaadf7 JD |
1777 | |
1778 | if (offset > bar_size) | |
f65efd6d | 1779 | return; |
8ffaadf7 JD |
1780 | |
1781 | /* | |
1782 | * Controllers may support a CMB size larger than their BAR, | |
1783 | * for example, due to being behind a bridge. Reduce the CMB to | |
1784 | * the reported size of the BAR | |
1785 | */ | |
1786 | if (size > bar_size - offset) | |
1787 | size = bar_size - offset; | |
1788 | ||
0f238ff5 LG |
1789 | if (pci_p2pdma_add_resource(pdev, bar, size, offset)) { |
1790 | dev_warn(dev->ctrl.device, | |
1791 | "failed to register the CMB\n"); | |
f65efd6d | 1792 | return; |
0f238ff5 LG |
1793 | } |
1794 | ||
8ffaadf7 | 1795 | dev->cmb_size = size; |
0f238ff5 LG |
1796 | dev->cmb_use_sqes = use_cmb_sqes && (dev->cmbsz & NVME_CMBSZ_SQS); |
1797 | ||
1798 | if ((dev->cmbsz & (NVME_CMBSZ_WDS | NVME_CMBSZ_RDS)) == | |
1799 | (NVME_CMBSZ_WDS | NVME_CMBSZ_RDS)) | |
1800 | pci_p2pmem_publish(pdev, true); | |
f65efd6d CH |
1801 | |
1802 | if (sysfs_add_file_to_group(&dev->ctrl.device->kobj, | |
1803 | &dev_attr_cmb.attr, NULL)) | |
1804 | dev_warn(dev->ctrl.device, | |
1805 | "failed to add sysfs attribute for CMB\n"); | |
8ffaadf7 JD |
1806 | } |
1807 | ||
1808 | static inline void nvme_release_cmb(struct nvme_dev *dev) | |
1809 | { | |
0f238ff5 | 1810 | if (dev->cmb_size) { |
1c78f773 MG |
1811 | sysfs_remove_file_from_group(&dev->ctrl.device->kobj, |
1812 | &dev_attr_cmb.attr, NULL); | |
0f238ff5 | 1813 | dev->cmb_size = 0; |
8ffaadf7 JD |
1814 | } |
1815 | } | |
1816 | ||
87ad72a5 CH |
1817 | static int nvme_set_host_mem(struct nvme_dev *dev, u32 bits) |
1818 | { | |
4033f35d | 1819 | u64 dma_addr = dev->host_mem_descs_dma; |
87ad72a5 | 1820 | struct nvme_command c; |
87ad72a5 CH |
1821 | int ret; |
1822 | ||
87ad72a5 CH |
1823 | memset(&c, 0, sizeof(c)); |
1824 | c.features.opcode = nvme_admin_set_features; | |
1825 | c.features.fid = cpu_to_le32(NVME_FEAT_HOST_MEM_BUF); | |
1826 | c.features.dword11 = cpu_to_le32(bits); | |
1827 | c.features.dword12 = cpu_to_le32(dev->host_mem_size >> | |
1828 | ilog2(dev->ctrl.page_size)); | |
1829 | c.features.dword13 = cpu_to_le32(lower_32_bits(dma_addr)); | |
1830 | c.features.dword14 = cpu_to_le32(upper_32_bits(dma_addr)); | |
1831 | c.features.dword15 = cpu_to_le32(dev->nr_host_mem_descs); | |
1832 | ||
1833 | ret = nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); | |
1834 | if (ret) { | |
1835 | dev_warn(dev->ctrl.device, | |
1836 | "failed to set host mem (err %d, flags %#x).\n", | |
1837 | ret, bits); | |
1838 | } | |
87ad72a5 CH |
1839 | return ret; |
1840 | } | |
1841 | ||
1842 | static void nvme_free_host_mem(struct nvme_dev *dev) | |
1843 | { | |
1844 | int i; | |
1845 | ||
1846 | for (i = 0; i < dev->nr_host_mem_descs; i++) { | |
1847 | struct nvme_host_mem_buf_desc *desc = &dev->host_mem_descs[i]; | |
1848 | size_t size = le32_to_cpu(desc->size) * dev->ctrl.page_size; | |
1849 | ||
cc667f6d LD |
1850 | dma_free_attrs(dev->dev, size, dev->host_mem_desc_bufs[i], |
1851 | le64_to_cpu(desc->addr), | |
1852 | DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); | |
87ad72a5 CH |
1853 | } |
1854 | ||
1855 | kfree(dev->host_mem_desc_bufs); | |
1856 | dev->host_mem_desc_bufs = NULL; | |
4033f35d CH |
1857 | dma_free_coherent(dev->dev, |
1858 | dev->nr_host_mem_descs * sizeof(*dev->host_mem_descs), | |
1859 | dev->host_mem_descs, dev->host_mem_descs_dma); | |
87ad72a5 | 1860 | dev->host_mem_descs = NULL; |
7e5dd57e | 1861 | dev->nr_host_mem_descs = 0; |
87ad72a5 CH |
1862 | } |
1863 | ||
92dc6895 CH |
1864 | static int __nvme_alloc_host_mem(struct nvme_dev *dev, u64 preferred, |
1865 | u32 chunk_size) | |
9d713c2b | 1866 | { |
87ad72a5 | 1867 | struct nvme_host_mem_buf_desc *descs; |
92dc6895 | 1868 | u32 max_entries, len; |
4033f35d | 1869 | dma_addr_t descs_dma; |
2ee0e4ed | 1870 | int i = 0; |
87ad72a5 | 1871 | void **bufs; |
6fbcde66 | 1872 | u64 size, tmp; |
87ad72a5 | 1873 | |
87ad72a5 CH |
1874 | tmp = (preferred + chunk_size - 1); |
1875 | do_div(tmp, chunk_size); | |
1876 | max_entries = tmp; | |
044a9df1 CH |
1877 | |
1878 | if (dev->ctrl.hmmaxd && dev->ctrl.hmmaxd < max_entries) | |
1879 | max_entries = dev->ctrl.hmmaxd; | |
1880 | ||
750afb08 LC |
1881 | descs = dma_alloc_coherent(dev->dev, max_entries * sizeof(*descs), |
1882 | &descs_dma, GFP_KERNEL); | |
87ad72a5 CH |
1883 | if (!descs) |
1884 | goto out; | |
1885 | ||
1886 | bufs = kcalloc(max_entries, sizeof(*bufs), GFP_KERNEL); | |
1887 | if (!bufs) | |
1888 | goto out_free_descs; | |
1889 | ||
244a8fe4 | 1890 | for (size = 0; size < preferred && i < max_entries; size += len) { |
87ad72a5 CH |
1891 | dma_addr_t dma_addr; |
1892 | ||
50cdb7c6 | 1893 | len = min_t(u64, chunk_size, preferred - size); |
87ad72a5 CH |
1894 | bufs[i] = dma_alloc_attrs(dev->dev, len, &dma_addr, GFP_KERNEL, |
1895 | DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); | |
1896 | if (!bufs[i]) | |
1897 | break; | |
1898 | ||
1899 | descs[i].addr = cpu_to_le64(dma_addr); | |
1900 | descs[i].size = cpu_to_le32(len / dev->ctrl.page_size); | |
1901 | i++; | |
1902 | } | |
1903 | ||
92dc6895 | 1904 | if (!size) |
87ad72a5 | 1905 | goto out_free_bufs; |
87ad72a5 | 1906 | |
87ad72a5 CH |
1907 | dev->nr_host_mem_descs = i; |
1908 | dev->host_mem_size = size; | |
1909 | dev->host_mem_descs = descs; | |
4033f35d | 1910 | dev->host_mem_descs_dma = descs_dma; |
87ad72a5 CH |
1911 | dev->host_mem_desc_bufs = bufs; |
1912 | return 0; | |
1913 | ||
1914 | out_free_bufs: | |
1915 | while (--i >= 0) { | |
1916 | size_t size = le32_to_cpu(descs[i].size) * dev->ctrl.page_size; | |
1917 | ||
cc667f6d LD |
1918 | dma_free_attrs(dev->dev, size, bufs[i], |
1919 | le64_to_cpu(descs[i].addr), | |
1920 | DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); | |
87ad72a5 CH |
1921 | } |
1922 | ||
1923 | kfree(bufs); | |
1924 | out_free_descs: | |
4033f35d CH |
1925 | dma_free_coherent(dev->dev, max_entries * sizeof(*descs), descs, |
1926 | descs_dma); | |
87ad72a5 | 1927 | out: |
87ad72a5 CH |
1928 | dev->host_mem_descs = NULL; |
1929 | return -ENOMEM; | |
1930 | } | |
1931 | ||
92dc6895 CH |
1932 | static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred) |
1933 | { | |
9dc54a0d CK |
1934 | u64 min_chunk = min_t(u64, preferred, PAGE_SIZE * MAX_ORDER_NR_PAGES); |
1935 | u64 hmminds = max_t(u32, dev->ctrl.hmminds * 4096, PAGE_SIZE * 2); | |
1936 | u64 chunk_size; | |
92dc6895 CH |
1937 | |
1938 | /* start big and work our way down */ | |
9dc54a0d | 1939 | for (chunk_size = min_chunk; chunk_size >= hmminds; chunk_size /= 2) { |
92dc6895 CH |
1940 | if (!__nvme_alloc_host_mem(dev, preferred, chunk_size)) { |
1941 | if (!min || dev->host_mem_size >= min) | |
1942 | return 0; | |
1943 | nvme_free_host_mem(dev); | |
1944 | } | |
1945 | } | |
1946 | ||
1947 | return -ENOMEM; | |
1948 | } | |
1949 | ||
9620cfba | 1950 | static int nvme_setup_host_mem(struct nvme_dev *dev) |
87ad72a5 CH |
1951 | { |
1952 | u64 max = (u64)max_host_mem_size_mb * SZ_1M; | |
1953 | u64 preferred = (u64)dev->ctrl.hmpre * 4096; | |
1954 | u64 min = (u64)dev->ctrl.hmmin * 4096; | |
1955 | u32 enable_bits = NVME_HOST_MEM_ENABLE; | |
6fbcde66 | 1956 | int ret; |
87ad72a5 CH |
1957 | |
1958 | preferred = min(preferred, max); | |
1959 | if (min > max) { | |
1960 | dev_warn(dev->ctrl.device, | |
1961 | "min host memory (%lld MiB) above limit (%d MiB).\n", | |
1962 | min >> ilog2(SZ_1M), max_host_mem_size_mb); | |
1963 | nvme_free_host_mem(dev); | |
9620cfba | 1964 | return 0; |
87ad72a5 CH |
1965 | } |
1966 | ||
1967 | /* | |
1968 | * If we already have a buffer allocated check if we can reuse it. | |
1969 | */ | |
1970 | if (dev->host_mem_descs) { | |
1971 | if (dev->host_mem_size >= min) | |
1972 | enable_bits |= NVME_HOST_MEM_RETURN; | |
1973 | else | |
1974 | nvme_free_host_mem(dev); | |
1975 | } | |
1976 | ||
1977 | if (!dev->host_mem_descs) { | |
92dc6895 CH |
1978 | if (nvme_alloc_host_mem(dev, min, preferred)) { |
1979 | dev_warn(dev->ctrl.device, | |
1980 | "failed to allocate host memory buffer.\n"); | |
9620cfba | 1981 | return 0; /* controller must work without HMB */ |
92dc6895 CH |
1982 | } |
1983 | ||
1984 | dev_info(dev->ctrl.device, | |
1985 | "allocated %lld MiB host memory buffer.\n", | |
1986 | dev->host_mem_size >> ilog2(SZ_1M)); | |
87ad72a5 CH |
1987 | } |
1988 | ||
9620cfba CH |
1989 | ret = nvme_set_host_mem(dev, enable_bits); |
1990 | if (ret) | |
87ad72a5 | 1991 | nvme_free_host_mem(dev); |
9620cfba | 1992 | return ret; |
9d713c2b KB |
1993 | } |
1994 | ||
612b7286 ML |
1995 | /* |
1996 | * nirqs is the number of interrupts available for write and read | |
1997 | * queues. The core already reserved an interrupt for the admin queue. | |
1998 | */ | |
1999 | static void nvme_calc_irq_sets(struct irq_affinity *affd, unsigned int nrirqs) | |
3b6592f7 | 2000 | { |
612b7286 | 2001 | struct nvme_dev *dev = affd->priv; |
2a5bcfdd | 2002 | unsigned int nr_read_queues, nr_write_queues = dev->nr_write_queues; |
3b6592f7 JA |
2003 | |
2004 | /* | |
ee0d96d3 | 2005 | * If there is no interrupt available for queues, ensure that |
612b7286 ML |
2006 | * the default queue is set to 1. The affinity set size is |
2007 | * also set to one, but the irq core ignores it for this case. | |
2008 | * | |
2009 | * If only one interrupt is available or 'write_queue' == 0, combine | |
2010 | * write and read queues. | |
2011 | * | |
2012 | * If 'write_queues' > 0, ensure it leaves room for at least one read | |
2013 | * queue. | |
3b6592f7 | 2014 | */ |
612b7286 ML |
2015 | if (!nrirqs) { |
2016 | nrirqs = 1; | |
2017 | nr_read_queues = 0; | |
2a5bcfdd | 2018 | } else if (nrirqs == 1 || !nr_write_queues) { |
612b7286 | 2019 | nr_read_queues = 0; |
2a5bcfdd | 2020 | } else if (nr_write_queues >= nrirqs) { |
612b7286 | 2021 | nr_read_queues = 1; |
3b6592f7 | 2022 | } else { |
2a5bcfdd | 2023 | nr_read_queues = nrirqs - nr_write_queues; |
3b6592f7 | 2024 | } |
612b7286 ML |
2025 | |
2026 | dev->io_queues[HCTX_TYPE_DEFAULT] = nrirqs - nr_read_queues; | |
2027 | affd->set_size[HCTX_TYPE_DEFAULT] = nrirqs - nr_read_queues; | |
2028 | dev->io_queues[HCTX_TYPE_READ] = nr_read_queues; | |
2029 | affd->set_size[HCTX_TYPE_READ] = nr_read_queues; | |
2030 | affd->nr_sets = nr_read_queues ? 2 : 1; | |
3b6592f7 JA |
2031 | } |
2032 | ||
6451fe73 | 2033 | static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues) |
3b6592f7 JA |
2034 | { |
2035 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
3b6592f7 | 2036 | struct irq_affinity affd = { |
9cfef55b | 2037 | .pre_vectors = 1, |
612b7286 ML |
2038 | .calc_sets = nvme_calc_irq_sets, |
2039 | .priv = dev, | |
3b6592f7 | 2040 | }; |
6451fe73 JA |
2041 | unsigned int irq_queues, this_p_queues; |
2042 | ||
2043 | /* | |
2044 | * Poll queues don't need interrupts, but we need at least one IO | |
2045 | * queue left over for non-polled IO. | |
2046 | */ | |
2a5bcfdd | 2047 | this_p_queues = dev->nr_poll_queues; |
6451fe73 JA |
2048 | if (this_p_queues >= nr_io_queues) { |
2049 | this_p_queues = nr_io_queues - 1; | |
2050 | irq_queues = 1; | |
2051 | } else { | |
7e4c6b9a | 2052 | irq_queues = nr_io_queues - this_p_queues + 1; |
6451fe73 JA |
2053 | } |
2054 | dev->io_queues[HCTX_TYPE_POLL] = this_p_queues; | |
3b6592f7 | 2055 | |
612b7286 ML |
2056 | /* Initialize for the single interrupt case */ |
2057 | dev->io_queues[HCTX_TYPE_DEFAULT] = 1; | |
2058 | dev->io_queues[HCTX_TYPE_READ] = 0; | |
3b6592f7 | 2059 | |
66341331 BH |
2060 | /* |
2061 | * Some Apple controllers require all queues to use the | |
2062 | * first vector. | |
2063 | */ | |
2064 | if (dev->ctrl.quirks & NVME_QUIRK_SINGLE_VECTOR) | |
2065 | irq_queues = 1; | |
2066 | ||
612b7286 ML |
2067 | return pci_alloc_irq_vectors_affinity(pdev, 1, irq_queues, |
2068 | PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd); | |
3b6592f7 JA |
2069 | } |
2070 | ||
8fae268b KB |
2071 | static void nvme_disable_io_queues(struct nvme_dev *dev) |
2072 | { | |
2073 | if (__nvme_disable_io_queues(dev, nvme_admin_delete_sq)) | |
2074 | __nvme_disable_io_queues(dev, nvme_admin_delete_cq); | |
2075 | } | |
2076 | ||
2a5bcfdd WZ |
2077 | static unsigned int nvme_max_io_queues(struct nvme_dev *dev) |
2078 | { | |
2079 | return num_possible_cpus() + dev->nr_write_queues + dev->nr_poll_queues; | |
2080 | } | |
2081 | ||
8d85fce7 | 2082 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 2083 | { |
147b27e4 | 2084 | struct nvme_queue *adminq = &dev->queues[0]; |
e75ec752 | 2085 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2a5bcfdd | 2086 | unsigned int nr_io_queues; |
97f6ef64 | 2087 | unsigned long size; |
2a5bcfdd | 2088 | int result; |
b60503ba | 2089 | |
2a5bcfdd WZ |
2090 | /* |
2091 | * Sample the module parameters once at reset time so that we have | |
2092 | * stable values to work with. | |
2093 | */ | |
2094 | dev->nr_write_queues = write_queues; | |
2095 | dev->nr_poll_queues = poll_queues; | |
d38e9f04 BH |
2096 | |
2097 | /* | |
2098 | * If tags are shared with admin queue (Apple bug), then | |
2099 | * make sure we only use one IO queue. | |
2100 | */ | |
2101 | if (dev->ctrl.quirks & NVME_QUIRK_SHARED_TAGS) | |
2102 | nr_io_queues = 1; | |
2a5bcfdd WZ |
2103 | else |
2104 | nr_io_queues = min(nvme_max_io_queues(dev), | |
2105 | dev->nr_allocated_queues - 1); | |
d38e9f04 | 2106 | |
9a0be7ab CH |
2107 | result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues); |
2108 | if (result < 0) | |
1b23484b | 2109 | return result; |
9a0be7ab | 2110 | |
f5fa90dc | 2111 | if (nr_io_queues == 0) |
a5229050 | 2112 | return 0; |
4e224106 CH |
2113 | |
2114 | clear_bit(NVMEQ_ENABLED, &adminq->flags); | |
b60503ba | 2115 | |
0f238ff5 | 2116 | if (dev->cmb_use_sqes) { |
8ffaadf7 JD |
2117 | result = nvme_cmb_qdepth(dev, nr_io_queues, |
2118 | sizeof(struct nvme_command)); | |
2119 | if (result > 0) | |
2120 | dev->q_depth = result; | |
2121 | else | |
0f238ff5 | 2122 | dev->cmb_use_sqes = false; |
8ffaadf7 JD |
2123 | } |
2124 | ||
97f6ef64 XY |
2125 | do { |
2126 | size = db_bar_size(dev, nr_io_queues); | |
2127 | result = nvme_remap_bar(dev, size); | |
2128 | if (!result) | |
2129 | break; | |
2130 | if (!--nr_io_queues) | |
2131 | return -ENOMEM; | |
2132 | } while (1); | |
2133 | adminq->q_db = dev->dbs; | |
f1938f6e | 2134 | |
8fae268b | 2135 | retry: |
9d713c2b | 2136 | /* Deregister the admin queue's interrupt */ |
0ff199cb | 2137 | pci_free_irq(pdev, 0, adminq); |
9d713c2b | 2138 | |
e32efbfc JA |
2139 | /* |
2140 | * If we enable msix early due to not intx, disable it again before | |
2141 | * setting up the full range we need. | |
2142 | */ | |
dca51e78 | 2143 | pci_free_irq_vectors(pdev); |
3b6592f7 JA |
2144 | |
2145 | result = nvme_setup_irqs(dev, nr_io_queues); | |
22b55601 | 2146 | if (result <= 0) |
dca51e78 | 2147 | return -EIO; |
3b6592f7 | 2148 | |
22b55601 | 2149 | dev->num_vecs = result; |
4b04cc6a | 2150 | result = max(result - 1, 1); |
e20ba6e1 | 2151 | dev->max_qid = result + dev->io_queues[HCTX_TYPE_POLL]; |
fa08a396 | 2152 | |
063a8096 MW |
2153 | /* |
2154 | * Should investigate if there's a performance win from allocating | |
2155 | * more queues than interrupt vectors; it might allow the submission | |
2156 | * path to scale better, even if the receive path is limited by the | |
2157 | * number of interrupts. | |
2158 | */ | |
dca51e78 | 2159 | result = queue_request_irq(adminq); |
7c349dde | 2160 | if (result) |
d4875622 | 2161 | return result; |
4e224106 | 2162 | set_bit(NVMEQ_ENABLED, &adminq->flags); |
8fae268b KB |
2163 | |
2164 | result = nvme_create_io_queues(dev); | |
2165 | if (result || dev->online_queues < 2) | |
2166 | return result; | |
2167 | ||
2168 | if (dev->online_queues - 1 < dev->max_qid) { | |
2169 | nr_io_queues = dev->online_queues - 1; | |
2170 | nvme_disable_io_queues(dev); | |
2171 | nvme_suspend_io_queues(dev); | |
2172 | goto retry; | |
2173 | } | |
2174 | dev_info(dev->ctrl.device, "%d/%d/%d default/read/poll queues\n", | |
2175 | dev->io_queues[HCTX_TYPE_DEFAULT], | |
2176 | dev->io_queues[HCTX_TYPE_READ], | |
2177 | dev->io_queues[HCTX_TYPE_POLL]); | |
2178 | return 0; | |
b60503ba MW |
2179 | } |
2180 | ||
2a842aca | 2181 | static void nvme_del_queue_end(struct request *req, blk_status_t error) |
a5768aa8 | 2182 | { |
db3cbfff | 2183 | struct nvme_queue *nvmeq = req->end_io_data; |
b5875222 | 2184 | |
db3cbfff | 2185 | blk_mq_free_request(req); |
d1ed6aa1 | 2186 | complete(&nvmeq->delete_done); |
a5768aa8 KB |
2187 | } |
2188 | ||
2a842aca | 2189 | static void nvme_del_cq_end(struct request *req, blk_status_t error) |
a5768aa8 | 2190 | { |
db3cbfff | 2191 | struct nvme_queue *nvmeq = req->end_io_data; |
a5768aa8 | 2192 | |
d1ed6aa1 CH |
2193 | if (error) |
2194 | set_bit(NVMEQ_DELETE_ERROR, &nvmeq->flags); | |
db3cbfff KB |
2195 | |
2196 | nvme_del_queue_end(req, error); | |
a5768aa8 KB |
2197 | } |
2198 | ||
db3cbfff | 2199 | static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode) |
bda4e0fb | 2200 | { |
db3cbfff KB |
2201 | struct request_queue *q = nvmeq->dev->ctrl.admin_q; |
2202 | struct request *req; | |
2203 | struct nvme_command cmd; | |
bda4e0fb | 2204 | |
db3cbfff KB |
2205 | memset(&cmd, 0, sizeof(cmd)); |
2206 | cmd.delete_queue.opcode = opcode; | |
2207 | cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
bda4e0fb | 2208 | |
eb71f435 | 2209 | req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); |
db3cbfff KB |
2210 | if (IS_ERR(req)) |
2211 | return PTR_ERR(req); | |
bda4e0fb | 2212 | |
db3cbfff KB |
2213 | req->timeout = ADMIN_TIMEOUT; |
2214 | req->end_io_data = nvmeq; | |
2215 | ||
d1ed6aa1 | 2216 | init_completion(&nvmeq->delete_done); |
db3cbfff KB |
2217 | blk_execute_rq_nowait(q, NULL, req, false, |
2218 | opcode == nvme_admin_delete_cq ? | |
2219 | nvme_del_cq_end : nvme_del_queue_end); | |
2220 | return 0; | |
bda4e0fb KB |
2221 | } |
2222 | ||
8fae268b | 2223 | static bool __nvme_disable_io_queues(struct nvme_dev *dev, u8 opcode) |
a5768aa8 | 2224 | { |
5271edd4 | 2225 | int nr_queues = dev->online_queues - 1, sent = 0; |
db3cbfff | 2226 | unsigned long timeout; |
a5768aa8 | 2227 | |
db3cbfff | 2228 | retry: |
5271edd4 CH |
2229 | timeout = ADMIN_TIMEOUT; |
2230 | while (nr_queues > 0) { | |
2231 | if (nvme_delete_queue(&dev->queues[nr_queues], opcode)) | |
2232 | break; | |
2233 | nr_queues--; | |
2234 | sent++; | |
db3cbfff | 2235 | } |
d1ed6aa1 CH |
2236 | while (sent) { |
2237 | struct nvme_queue *nvmeq = &dev->queues[nr_queues + sent]; | |
2238 | ||
2239 | timeout = wait_for_completion_io_timeout(&nvmeq->delete_done, | |
5271edd4 CH |
2240 | timeout); |
2241 | if (timeout == 0) | |
2242 | return false; | |
d1ed6aa1 | 2243 | |
d1ed6aa1 | 2244 | sent--; |
5271edd4 CH |
2245 | if (nr_queues) |
2246 | goto retry; | |
2247 | } | |
2248 | return true; | |
a5768aa8 KB |
2249 | } |
2250 | ||
5d02a5c1 | 2251 | static void nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 2252 | { |
2b1b7e78 JW |
2253 | int ret; |
2254 | ||
5bae7f73 | 2255 | if (!dev->ctrl.tagset) { |
376f7ef8 | 2256 | dev->tagset.ops = &nvme_mq_ops; |
ffe7704d | 2257 | dev->tagset.nr_hw_queues = dev->online_queues - 1; |
8fe34be1 | 2258 | dev->tagset.nr_maps = 2; /* default + read */ |
ed92ad37 CH |
2259 | if (dev->io_queues[HCTX_TYPE_POLL]) |
2260 | dev->tagset.nr_maps++; | |
ffe7704d | 2261 | dev->tagset.timeout = NVME_IO_TIMEOUT; |
d4ec47f1 | 2262 | dev->tagset.numa_node = dev->ctrl.numa_node; |
61f3b896 CK |
2263 | dev->tagset.queue_depth = min_t(unsigned int, dev->q_depth, |
2264 | BLK_MQ_MAX_DEPTH) - 1; | |
d43f1ccf | 2265 | dev->tagset.cmd_size = sizeof(struct nvme_iod); |
ffe7704d KB |
2266 | dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; |
2267 | dev->tagset.driver_data = dev; | |
b60503ba | 2268 | |
d38e9f04 BH |
2269 | /* |
2270 | * Some Apple controllers requires tags to be unique | |
2271 | * across admin and IO queue, so reserve the first 32 | |
2272 | * tags of the IO queue. | |
2273 | */ | |
2274 | if (dev->ctrl.quirks & NVME_QUIRK_SHARED_TAGS) | |
2275 | dev->tagset.reserved_tags = NVME_AQ_DEPTH; | |
2276 | ||
2b1b7e78 JW |
2277 | ret = blk_mq_alloc_tag_set(&dev->tagset); |
2278 | if (ret) { | |
2279 | dev_warn(dev->ctrl.device, | |
2280 | "IO queues tagset allocation failed %d\n", ret); | |
5d02a5c1 | 2281 | return; |
2b1b7e78 | 2282 | } |
5bae7f73 | 2283 | dev->ctrl.tagset = &dev->tagset; |
949928c1 KB |
2284 | } else { |
2285 | blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1); | |
2286 | ||
2287 | /* Free previously allocated queues that are no longer usable */ | |
2288 | nvme_free_queues(dev, dev->online_queues); | |
ffe7704d | 2289 | } |
949928c1 | 2290 | |
e8fd41bb | 2291 | nvme_dbbuf_set(dev); |
b60503ba MW |
2292 | } |
2293 | ||
b00a726a | 2294 | static int nvme_pci_enable(struct nvme_dev *dev) |
0877cb0d | 2295 | { |
b00a726a | 2296 | int result = -ENOMEM; |
e75ec752 | 2297 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
0877cb0d KB |
2298 | |
2299 | if (pci_enable_device_mem(pdev)) | |
2300 | return result; | |
2301 | ||
0877cb0d | 2302 | pci_set_master(pdev); |
0877cb0d | 2303 | |
4fe06923 | 2304 | if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64))) |
052d0efa | 2305 | goto disable; |
0877cb0d | 2306 | |
7a67cbea | 2307 | if (readl(dev->bar + NVME_REG_CSTS) == -1) { |
0e53d180 | 2308 | result = -ENODEV; |
b00a726a | 2309 | goto disable; |
0e53d180 | 2310 | } |
e32efbfc JA |
2311 | |
2312 | /* | |
a5229050 KB |
2313 | * Some devices and/or platforms don't advertise or work with INTx |
2314 | * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll | |
2315 | * adjust this later. | |
e32efbfc | 2316 | */ |
dca51e78 CH |
2317 | result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES); |
2318 | if (result < 0) | |
2319 | return result; | |
e32efbfc | 2320 | |
20d0dfe6 | 2321 | dev->ctrl.cap = lo_hi_readq(dev->bar + NVME_REG_CAP); |
7a67cbea | 2322 | |
61f3b896 | 2323 | dev->q_depth = min_t(u16, NVME_CAP_MQES(dev->ctrl.cap) + 1, |
b27c1e68 | 2324 | io_queue_depth); |
aa22c8e6 | 2325 | dev->ctrl.sqsize = dev->q_depth - 1; /* 0's based queue depth */ |
20d0dfe6 | 2326 | dev->db_stride = 1 << NVME_CAP_STRIDE(dev->ctrl.cap); |
7a67cbea | 2327 | dev->dbs = dev->bar + 4096; |
1f390c1f | 2328 | |
66341331 BH |
2329 | /* |
2330 | * Some Apple controllers require a non-standard SQE size. | |
2331 | * Interestingly they also seem to ignore the CC:IOSQES register | |
2332 | * so we don't bother updating it here. | |
2333 | */ | |
2334 | if (dev->ctrl.quirks & NVME_QUIRK_128_BYTES_SQES) | |
2335 | dev->io_sqes = 7; | |
2336 | else | |
2337 | dev->io_sqes = NVME_NVM_IOSQES; | |
1f390c1f SG |
2338 | |
2339 | /* | |
2340 | * Temporary fix for the Apple controller found in the MacBook8,1 and | |
2341 | * some MacBook7,1 to avoid controller resets and data loss. | |
2342 | */ | |
2343 | if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) { | |
2344 | dev->q_depth = 2; | |
9bdcfb10 CH |
2345 | dev_warn(dev->ctrl.device, "detected Apple NVMe controller, " |
2346 | "set queue depth=%u to work around controller resets\n", | |
1f390c1f | 2347 | dev->q_depth); |
d554b5e1 MP |
2348 | } else if (pdev->vendor == PCI_VENDOR_ID_SAMSUNG && |
2349 | (pdev->device == 0xa821 || pdev->device == 0xa822) && | |
20d0dfe6 | 2350 | NVME_CAP_MQES(dev->ctrl.cap) == 0) { |
d554b5e1 MP |
2351 | dev->q_depth = 64; |
2352 | dev_err(dev->ctrl.device, "detected PM1725 NVMe controller, " | |
2353 | "set queue depth=%u\n", dev->q_depth); | |
1f390c1f SG |
2354 | } |
2355 | ||
d38e9f04 BH |
2356 | /* |
2357 | * Controllers with the shared tags quirk need the IO queue to be | |
2358 | * big enough so that we get 32 tags for the admin queue | |
2359 | */ | |
2360 | if ((dev->ctrl.quirks & NVME_QUIRK_SHARED_TAGS) && | |
2361 | (dev->q_depth < (NVME_AQ_DEPTH + 2))) { | |
2362 | dev->q_depth = NVME_AQ_DEPTH + 2; | |
2363 | dev_warn(dev->ctrl.device, "IO queue depth clamped to %d\n", | |
2364 | dev->q_depth); | |
2365 | } | |
2366 | ||
2367 | ||
f65efd6d | 2368 | nvme_map_cmb(dev); |
202021c1 | 2369 | |
a0a3408e KB |
2370 | pci_enable_pcie_error_reporting(pdev); |
2371 | pci_save_state(pdev); | |
0877cb0d KB |
2372 | return 0; |
2373 | ||
2374 | disable: | |
0877cb0d KB |
2375 | pci_disable_device(pdev); |
2376 | return result; | |
2377 | } | |
2378 | ||
2379 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
b00a726a KB |
2380 | { |
2381 | if (dev->bar) | |
2382 | iounmap(dev->bar); | |
a1f447b3 | 2383 | pci_release_mem_regions(to_pci_dev(dev->dev)); |
b00a726a KB |
2384 | } |
2385 | ||
2386 | static void nvme_pci_disable(struct nvme_dev *dev) | |
0877cb0d | 2387 | { |
e75ec752 CH |
2388 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2389 | ||
dca51e78 | 2390 | pci_free_irq_vectors(pdev); |
0877cb0d | 2391 | |
a0a3408e KB |
2392 | if (pci_is_enabled(pdev)) { |
2393 | pci_disable_pcie_error_reporting(pdev); | |
e75ec752 | 2394 | pci_disable_device(pdev); |
4d115420 | 2395 | } |
4d115420 KB |
2396 | } |
2397 | ||
a5cdb68c | 2398 | static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown) |
b60503ba | 2399 | { |
e43269e6 | 2400 | bool dead = true, freeze = false; |
302ad8cc | 2401 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
22404274 | 2402 | |
77bf25ea | 2403 | mutex_lock(&dev->shutdown_lock); |
302ad8cc KB |
2404 | if (pci_is_enabled(pdev)) { |
2405 | u32 csts = readl(dev->bar + NVME_REG_CSTS); | |
2406 | ||
ebef7368 | 2407 | if (dev->ctrl.state == NVME_CTRL_LIVE || |
e43269e6 KB |
2408 | dev->ctrl.state == NVME_CTRL_RESETTING) { |
2409 | freeze = true; | |
302ad8cc | 2410 | nvme_start_freeze(&dev->ctrl); |
e43269e6 | 2411 | } |
302ad8cc KB |
2412 | dead = !!((csts & NVME_CSTS_CFS) || !(csts & NVME_CSTS_RDY) || |
2413 | pdev->error_state != pci_channel_io_normal); | |
c9d3bf88 | 2414 | } |
c21377f8 | 2415 | |
302ad8cc KB |
2416 | /* |
2417 | * Give the controller a chance to complete all entered requests if | |
2418 | * doing a safe shutdown. | |
2419 | */ | |
e43269e6 KB |
2420 | if (!dead && shutdown && freeze) |
2421 | nvme_wait_freeze_timeout(&dev->ctrl, NVME_IO_TIMEOUT); | |
9a915a5b JW |
2422 | |
2423 | nvme_stop_queues(&dev->ctrl); | |
87ad72a5 | 2424 | |
64ee0ac0 | 2425 | if (!dead && dev->ctrl.queue_count > 0) { |
8fae268b | 2426 | nvme_disable_io_queues(dev); |
a5cdb68c | 2427 | nvme_disable_admin_queue(dev, shutdown); |
4d115420 | 2428 | } |
8fae268b KB |
2429 | nvme_suspend_io_queues(dev); |
2430 | nvme_suspend_queue(&dev->queues[0]); | |
b00a726a | 2431 | nvme_pci_disable(dev); |
fa46c6fb | 2432 | nvme_reap_pending_cqes(dev); |
07836e65 | 2433 | |
e1958e65 ML |
2434 | blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl); |
2435 | blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl); | |
622b8b68 ML |
2436 | blk_mq_tagset_wait_completed_request(&dev->tagset); |
2437 | blk_mq_tagset_wait_completed_request(&dev->admin_tagset); | |
302ad8cc KB |
2438 | |
2439 | /* | |
2440 | * The driver will not be starting up queues again if shutting down so | |
2441 | * must flush all entered requests to their failed completion to avoid | |
2442 | * deadlocking blk-mq hot-cpu notifier. | |
2443 | */ | |
c8e9e9b7 | 2444 | if (shutdown) { |
302ad8cc | 2445 | nvme_start_queues(&dev->ctrl); |
c8e9e9b7 KB |
2446 | if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) |
2447 | blk_mq_unquiesce_queue(dev->ctrl.admin_q); | |
2448 | } | |
77bf25ea | 2449 | mutex_unlock(&dev->shutdown_lock); |
b60503ba MW |
2450 | } |
2451 | ||
c1ac9a4b KB |
2452 | static int nvme_disable_prepare_reset(struct nvme_dev *dev, bool shutdown) |
2453 | { | |
2454 | if (!nvme_wait_reset(&dev->ctrl)) | |
2455 | return -EBUSY; | |
2456 | nvme_dev_disable(dev, shutdown); | |
2457 | return 0; | |
2458 | } | |
2459 | ||
091b6092 MW |
2460 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
2461 | { | |
e75ec752 | 2462 | dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, |
091b6092 MW |
2463 | PAGE_SIZE, PAGE_SIZE, 0); |
2464 | if (!dev->prp_page_pool) | |
2465 | return -ENOMEM; | |
2466 | ||
99802a7a | 2467 | /* Optimisation for I/Os between 4k and 128k */ |
e75ec752 | 2468 | dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, |
99802a7a MW |
2469 | 256, 256, 0); |
2470 | if (!dev->prp_small_pool) { | |
2471 | dma_pool_destroy(dev->prp_page_pool); | |
2472 | return -ENOMEM; | |
2473 | } | |
091b6092 MW |
2474 | return 0; |
2475 | } | |
2476 | ||
2477 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
2478 | { | |
2479 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 2480 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
2481 | } |
2482 | ||
770597ec KB |
2483 | static void nvme_free_tagset(struct nvme_dev *dev) |
2484 | { | |
2485 | if (dev->tagset.tags) | |
2486 | blk_mq_free_tag_set(&dev->tagset); | |
2487 | dev->ctrl.tagset = NULL; | |
2488 | } | |
2489 | ||
1673f1f0 | 2490 | static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl) |
5e82e952 | 2491 | { |
1673f1f0 | 2492 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
9ac27090 | 2493 | |
f9f38e33 | 2494 | nvme_dbbuf_dma_free(dev); |
770597ec | 2495 | nvme_free_tagset(dev); |
1c63dc66 CH |
2496 | if (dev->ctrl.admin_q) |
2497 | blk_put_queue(dev->ctrl.admin_q); | |
e286bcfc | 2498 | free_opal_dev(dev->ctrl.opal_dev); |
943e942e | 2499 | mempool_destroy(dev->iod_mempool); |
253fd4ac IR |
2500 | put_device(dev->dev); |
2501 | kfree(dev->queues); | |
5e82e952 KB |
2502 | kfree(dev); |
2503 | } | |
2504 | ||
7c1ce408 | 2505 | static void nvme_remove_dead_ctrl(struct nvme_dev *dev) |
f58944e2 | 2506 | { |
c1ac9a4b KB |
2507 | /* |
2508 | * Set state to deleting now to avoid blocking nvme_wait_reset(), which | |
2509 | * may be holding this pci_dev's device lock. | |
2510 | */ | |
2511 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); | |
d22524a4 | 2512 | nvme_get_ctrl(&dev->ctrl); |
69d9a99c | 2513 | nvme_dev_disable(dev, false); |
9f9cafc1 | 2514 | nvme_kill_queues(&dev->ctrl); |
03e0f3a6 | 2515 | if (!queue_work(nvme_wq, &dev->remove_work)) |
f58944e2 KB |
2516 | nvme_put_ctrl(&dev->ctrl); |
2517 | } | |
2518 | ||
fd634f41 | 2519 | static void nvme_reset_work(struct work_struct *work) |
5e82e952 | 2520 | { |
d86c4d8e CH |
2521 | struct nvme_dev *dev = |
2522 | container_of(work, struct nvme_dev, ctrl.reset_work); | |
a98e58e5 | 2523 | bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL); |
e71afda4 | 2524 | int result; |
5e82e952 | 2525 | |
e71afda4 CK |
2526 | if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING)) { |
2527 | result = -ENODEV; | |
fd634f41 | 2528 | goto out; |
e71afda4 | 2529 | } |
5e82e952 | 2530 | |
fd634f41 CH |
2531 | /* |
2532 | * If we're called to reset a live controller first shut it down before | |
2533 | * moving on. | |
2534 | */ | |
b00a726a | 2535 | if (dev->ctrl.ctrl_config & NVME_CC_ENABLE) |
a5cdb68c | 2536 | nvme_dev_disable(dev, false); |
d6135c3a | 2537 | nvme_sync_queues(&dev->ctrl); |
5e82e952 | 2538 | |
5c959d73 | 2539 | mutex_lock(&dev->shutdown_lock); |
b00a726a | 2540 | result = nvme_pci_enable(dev); |
f0b50732 | 2541 | if (result) |
4726bcf3 | 2542 | goto out_unlock; |
f0b50732 | 2543 | |
01ad0990 | 2544 | result = nvme_pci_configure_admin_queue(dev); |
f0b50732 | 2545 | if (result) |
4726bcf3 | 2546 | goto out_unlock; |
f0b50732 | 2547 | |
0fb59cbc KB |
2548 | result = nvme_alloc_admin_tags(dev); |
2549 | if (result) | |
4726bcf3 | 2550 | goto out_unlock; |
b9afca3e | 2551 | |
943e942e JA |
2552 | /* |
2553 | * Limit the max command size to prevent iod->sg allocations going | |
2554 | * over a single page. | |
2555 | */ | |
7637de31 CH |
2556 | dev->ctrl.max_hw_sectors = min_t(u32, |
2557 | NVME_MAX_KB_SZ << 1, dma_max_mapping_size(dev->dev) >> 9); | |
943e942e | 2558 | dev->ctrl.max_segments = NVME_MAX_SEGS; |
a48bc520 CH |
2559 | |
2560 | /* | |
2561 | * Don't limit the IOMMU merged segment size. | |
2562 | */ | |
2563 | dma_set_max_seg_size(dev->dev, 0xffffffff); | |
2564 | ||
5c959d73 KB |
2565 | mutex_unlock(&dev->shutdown_lock); |
2566 | ||
2567 | /* | |
2568 | * Introduce CONNECTING state from nvme-fc/rdma transports to mark the | |
2569 | * initializing procedure here. | |
2570 | */ | |
2571 | if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_CONNECTING)) { | |
2572 | dev_warn(dev->ctrl.device, | |
2573 | "failed to mark controller CONNECTING\n"); | |
cee6c269 | 2574 | result = -EBUSY; |
5c959d73 KB |
2575 | goto out; |
2576 | } | |
943e942e | 2577 | |
95093350 MG |
2578 | /* |
2579 | * We do not support an SGL for metadata (yet), so we are limited to a | |
2580 | * single integrity segment for the separate metadata pointer. | |
2581 | */ | |
2582 | dev->ctrl.max_integrity_segments = 1; | |
2583 | ||
ce4541f4 CH |
2584 | result = nvme_init_identify(&dev->ctrl); |
2585 | if (result) | |
f58944e2 | 2586 | goto out; |
ce4541f4 | 2587 | |
e286bcfc SB |
2588 | if (dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) { |
2589 | if (!dev->ctrl.opal_dev) | |
2590 | dev->ctrl.opal_dev = | |
2591 | init_opal_dev(&dev->ctrl, &nvme_sec_submit); | |
2592 | else if (was_suspend) | |
2593 | opal_unlock_from_suspend(dev->ctrl.opal_dev); | |
2594 | } else { | |
2595 | free_opal_dev(dev->ctrl.opal_dev); | |
2596 | dev->ctrl.opal_dev = NULL; | |
4f1244c8 | 2597 | } |
a98e58e5 | 2598 | |
f9f38e33 HK |
2599 | if (dev->ctrl.oacs & NVME_CTRL_OACS_DBBUF_SUPP) { |
2600 | result = nvme_dbbuf_dma_alloc(dev); | |
2601 | if (result) | |
2602 | dev_warn(dev->dev, | |
2603 | "unable to allocate dma for dbbuf\n"); | |
2604 | } | |
2605 | ||
9620cfba CH |
2606 | if (dev->ctrl.hmpre) { |
2607 | result = nvme_setup_host_mem(dev); | |
2608 | if (result < 0) | |
2609 | goto out; | |
2610 | } | |
87ad72a5 | 2611 | |
f0b50732 | 2612 | result = nvme_setup_io_queues(dev); |
badc34d4 | 2613 | if (result) |
f58944e2 | 2614 | goto out; |
f0b50732 | 2615 | |
2659e57b CH |
2616 | /* |
2617 | * Keep the controller around but remove all namespaces if we don't have | |
2618 | * any working I/O queue. | |
2619 | */ | |
3cf519b5 | 2620 | if (dev->online_queues < 2) { |
1b3c47c1 | 2621 | dev_warn(dev->ctrl.device, "IO queues not created\n"); |
3b24774e | 2622 | nvme_kill_queues(&dev->ctrl); |
5bae7f73 | 2623 | nvme_remove_namespaces(&dev->ctrl); |
770597ec | 2624 | nvme_free_tagset(dev); |
3cf519b5 | 2625 | } else { |
25646264 | 2626 | nvme_start_queues(&dev->ctrl); |
302ad8cc | 2627 | nvme_wait_freeze(&dev->ctrl); |
5d02a5c1 | 2628 | nvme_dev_add(dev); |
302ad8cc | 2629 | nvme_unfreeze(&dev->ctrl); |
3cf519b5 CH |
2630 | } |
2631 | ||
2b1b7e78 JW |
2632 | /* |
2633 | * If only admin queue live, keep it to do further investigation or | |
2634 | * recovery. | |
2635 | */ | |
5d02a5c1 | 2636 | if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) { |
2b1b7e78 | 2637 | dev_warn(dev->ctrl.device, |
5d02a5c1 | 2638 | "failed to mark controller live state\n"); |
e71afda4 | 2639 | result = -ENODEV; |
bb8d261e CH |
2640 | goto out; |
2641 | } | |
92911a55 | 2642 | |
d09f2b45 | 2643 | nvme_start_ctrl(&dev->ctrl); |
3cf519b5 | 2644 | return; |
f0b50732 | 2645 | |
4726bcf3 KB |
2646 | out_unlock: |
2647 | mutex_unlock(&dev->shutdown_lock); | |
3cf519b5 | 2648 | out: |
7c1ce408 CK |
2649 | if (result) |
2650 | dev_warn(dev->ctrl.device, | |
2651 | "Removing after probe failure status: %d\n", result); | |
2652 | nvme_remove_dead_ctrl(dev); | |
f0b50732 KB |
2653 | } |
2654 | ||
5c8809e6 | 2655 | static void nvme_remove_dead_ctrl_work(struct work_struct *work) |
9a6b9458 | 2656 | { |
5c8809e6 | 2657 | struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work); |
e75ec752 | 2658 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
9a6b9458 KB |
2659 | |
2660 | if (pci_get_drvdata(pdev)) | |
921920ab | 2661 | device_release_driver(&pdev->dev); |
1673f1f0 | 2662 | nvme_put_ctrl(&dev->ctrl); |
9a6b9458 KB |
2663 | } |
2664 | ||
1c63dc66 | 2665 | static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) |
9ca97374 | 2666 | { |
1c63dc66 | 2667 | *val = readl(to_nvme_dev(ctrl)->bar + off); |
90667892 | 2668 | return 0; |
9ca97374 TH |
2669 | } |
2670 | ||
5fd4ce1b | 2671 | static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) |
4cc06521 | 2672 | { |
5fd4ce1b CH |
2673 | writel(val, to_nvme_dev(ctrl)->bar + off); |
2674 | return 0; | |
2675 | } | |
4cc06521 | 2676 | |
7fd8930f CH |
2677 | static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) |
2678 | { | |
3a8ecc93 | 2679 | *val = lo_hi_readq(to_nvme_dev(ctrl)->bar + off); |
7fd8930f | 2680 | return 0; |
4cc06521 KB |
2681 | } |
2682 | ||
97c12223 KB |
2683 | static int nvme_pci_get_address(struct nvme_ctrl *ctrl, char *buf, int size) |
2684 | { | |
2685 | struct pci_dev *pdev = to_pci_dev(to_nvme_dev(ctrl)->dev); | |
2686 | ||
2db24e4a | 2687 | return snprintf(buf, size, "%s\n", dev_name(&pdev->dev)); |
97c12223 KB |
2688 | } |
2689 | ||
1c63dc66 | 2690 | static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = { |
1a353d85 | 2691 | .name = "pcie", |
e439bb12 | 2692 | .module = THIS_MODULE, |
e0596ab2 LG |
2693 | .flags = NVME_F_METADATA_SUPPORTED | |
2694 | NVME_F_PCI_P2PDMA, | |
1c63dc66 | 2695 | .reg_read32 = nvme_pci_reg_read32, |
5fd4ce1b | 2696 | .reg_write32 = nvme_pci_reg_write32, |
7fd8930f | 2697 | .reg_read64 = nvme_pci_reg_read64, |
1673f1f0 | 2698 | .free_ctrl = nvme_pci_free_ctrl, |
f866fc42 | 2699 | .submit_async_event = nvme_pci_submit_async_event, |
97c12223 | 2700 | .get_address = nvme_pci_get_address, |
1c63dc66 | 2701 | }; |
4cc06521 | 2702 | |
b00a726a KB |
2703 | static int nvme_dev_map(struct nvme_dev *dev) |
2704 | { | |
b00a726a KB |
2705 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2706 | ||
a1f447b3 | 2707 | if (pci_request_mem_regions(pdev, "nvme")) |
b00a726a KB |
2708 | return -ENODEV; |
2709 | ||
97f6ef64 | 2710 | if (nvme_remap_bar(dev, NVME_REG_DBS + 4096)) |
b00a726a KB |
2711 | goto release; |
2712 | ||
9fa196e7 | 2713 | return 0; |
b00a726a | 2714 | release: |
9fa196e7 MG |
2715 | pci_release_mem_regions(pdev); |
2716 | return -ENODEV; | |
b00a726a KB |
2717 | } |
2718 | ||
8427bbc2 | 2719 | static unsigned long check_vendor_combination_bug(struct pci_dev *pdev) |
ff5350a8 AL |
2720 | { |
2721 | if (pdev->vendor == 0x144d && pdev->device == 0xa802) { | |
2722 | /* | |
2723 | * Several Samsung devices seem to drop off the PCIe bus | |
2724 | * randomly when APST is on and uses the deepest sleep state. | |
2725 | * This has been observed on a Samsung "SM951 NVMe SAMSUNG | |
2726 | * 256GB", a "PM951 NVMe SAMSUNG 512GB", and a "Samsung SSD | |
2727 | * 950 PRO 256GB", but it seems to be restricted to two Dell | |
2728 | * laptops. | |
2729 | */ | |
2730 | if (dmi_match(DMI_SYS_VENDOR, "Dell Inc.") && | |
2731 | (dmi_match(DMI_PRODUCT_NAME, "XPS 15 9550") || | |
2732 | dmi_match(DMI_PRODUCT_NAME, "Precision 5510"))) | |
2733 | return NVME_QUIRK_NO_DEEPEST_PS; | |
8427bbc2 KHF |
2734 | } else if (pdev->vendor == 0x144d && pdev->device == 0xa804) { |
2735 | /* | |
2736 | * Samsung SSD 960 EVO drops off the PCIe bus after system | |
467c77d4 JJ |
2737 | * suspend on a Ryzen board, ASUS PRIME B350M-A, as well as |
2738 | * within few minutes after bootup on a Coffee Lake board - | |
2739 | * ASUS PRIME Z370-A | |
8427bbc2 KHF |
2740 | */ |
2741 | if (dmi_match(DMI_BOARD_VENDOR, "ASUSTeK COMPUTER INC.") && | |
467c77d4 JJ |
2742 | (dmi_match(DMI_BOARD_NAME, "PRIME B350M-A") || |
2743 | dmi_match(DMI_BOARD_NAME, "PRIME Z370-A"))) | |
8427bbc2 | 2744 | return NVME_QUIRK_NO_APST; |
1fae37ac S |
2745 | } else if ((pdev->vendor == 0x144d && (pdev->device == 0xa801 || |
2746 | pdev->device == 0xa808 || pdev->device == 0xa809)) || | |
2747 | (pdev->vendor == 0x1e0f && pdev->device == 0x0001)) { | |
2748 | /* | |
2749 | * Forcing to use host managed nvme power settings for | |
2750 | * lowest idle power with quick resume latency on | |
2751 | * Samsung and Toshiba SSDs based on suspend behavior | |
2752 | * on Coffee Lake board for LENOVO C640 | |
2753 | */ | |
2754 | if ((dmi_match(DMI_BOARD_VENDOR, "LENOVO")) && | |
2755 | dmi_match(DMI_BOARD_NAME, "LNVNB161216")) | |
2756 | return NVME_QUIRK_SIMPLE_SUSPEND; | |
ff5350a8 AL |
2757 | } |
2758 | ||
2759 | return 0; | |
2760 | } | |
2761 | ||
18119775 KB |
2762 | static void nvme_async_probe(void *data, async_cookie_t cookie) |
2763 | { | |
2764 | struct nvme_dev *dev = data; | |
80f513b5 | 2765 | |
bd46a906 | 2766 | flush_work(&dev->ctrl.reset_work); |
18119775 | 2767 | flush_work(&dev->ctrl.scan_work); |
80f513b5 | 2768 | nvme_put_ctrl(&dev->ctrl); |
18119775 KB |
2769 | } |
2770 | ||
8d85fce7 | 2771 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 2772 | { |
a4aea562 | 2773 | int node, result = -ENOMEM; |
b60503ba | 2774 | struct nvme_dev *dev; |
ff5350a8 | 2775 | unsigned long quirks = id->driver_data; |
943e942e | 2776 | size_t alloc_size; |
b60503ba | 2777 | |
a4aea562 MB |
2778 | node = dev_to_node(&pdev->dev); |
2779 | if (node == NUMA_NO_NODE) | |
2fa84351 | 2780 | set_dev_node(&pdev->dev, first_memory_node); |
a4aea562 MB |
2781 | |
2782 | dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); | |
b60503ba MW |
2783 | if (!dev) |
2784 | return -ENOMEM; | |
147b27e4 | 2785 | |
2a5bcfdd WZ |
2786 | dev->nr_write_queues = write_queues; |
2787 | dev->nr_poll_queues = poll_queues; | |
2788 | dev->nr_allocated_queues = nvme_max_io_queues(dev) + 1; | |
2789 | dev->queues = kcalloc_node(dev->nr_allocated_queues, | |
2790 | sizeof(struct nvme_queue), GFP_KERNEL, node); | |
b60503ba MW |
2791 | if (!dev->queues) |
2792 | goto free; | |
2793 | ||
e75ec752 | 2794 | dev->dev = get_device(&pdev->dev); |
9a6b9458 | 2795 | pci_set_drvdata(pdev, dev); |
1c63dc66 | 2796 | |
b00a726a KB |
2797 | result = nvme_dev_map(dev); |
2798 | if (result) | |
b00c9b7a | 2799 | goto put_pci; |
b00a726a | 2800 | |
d86c4d8e | 2801 | INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work); |
5c8809e6 | 2802 | INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work); |
77bf25ea | 2803 | mutex_init(&dev->shutdown_lock); |
b60503ba | 2804 | |
091b6092 MW |
2805 | result = nvme_setup_prp_pools(dev); |
2806 | if (result) | |
b00c9b7a | 2807 | goto unmap; |
4cc06521 | 2808 | |
8427bbc2 | 2809 | quirks |= check_vendor_combination_bug(pdev); |
ff5350a8 | 2810 | |
943e942e JA |
2811 | /* |
2812 | * Double check that our mempool alloc size will cover the biggest | |
2813 | * command we support. | |
2814 | */ | |
2815 | alloc_size = nvme_pci_iod_alloc_size(dev, NVME_MAX_KB_SZ, | |
2816 | NVME_MAX_SEGS, true); | |
2817 | WARN_ON_ONCE(alloc_size > PAGE_SIZE); | |
2818 | ||
2819 | dev->iod_mempool = mempool_create_node(1, mempool_kmalloc, | |
2820 | mempool_kfree, | |
2821 | (void *) alloc_size, | |
2822 | GFP_KERNEL, node); | |
2823 | if (!dev->iod_mempool) { | |
2824 | result = -ENOMEM; | |
2825 | goto release_pools; | |
2826 | } | |
2827 | ||
b6e44b4c KB |
2828 | result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops, |
2829 | quirks); | |
2830 | if (result) | |
2831 | goto release_mempool; | |
2832 | ||
1b3c47c1 SG |
2833 | dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev)); |
2834 | ||
bd46a906 | 2835 | nvme_reset_ctrl(&dev->ctrl); |
18119775 | 2836 | async_schedule(nvme_async_probe, dev); |
4caff8fc | 2837 | |
b60503ba MW |
2838 | return 0; |
2839 | ||
b6e44b4c KB |
2840 | release_mempool: |
2841 | mempool_destroy(dev->iod_mempool); | |
0877cb0d | 2842 | release_pools: |
091b6092 | 2843 | nvme_release_prp_pools(dev); |
b00c9b7a CJ |
2844 | unmap: |
2845 | nvme_dev_unmap(dev); | |
a96d4f5c | 2846 | put_pci: |
e75ec752 | 2847 | put_device(dev->dev); |
b60503ba MW |
2848 | free: |
2849 | kfree(dev->queues); | |
b60503ba MW |
2850 | kfree(dev); |
2851 | return result; | |
2852 | } | |
2853 | ||
775755ed | 2854 | static void nvme_reset_prepare(struct pci_dev *pdev) |
f0d54a54 | 2855 | { |
a6739479 | 2856 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
c1ac9a4b KB |
2857 | |
2858 | /* | |
2859 | * We don't need to check the return value from waiting for the reset | |
2860 | * state as pci_dev device lock is held, making it impossible to race | |
2861 | * with ->remove(). | |
2862 | */ | |
2863 | nvme_disable_prepare_reset(dev, false); | |
2864 | nvme_sync_queues(&dev->ctrl); | |
775755ed | 2865 | } |
f0d54a54 | 2866 | |
775755ed CH |
2867 | static void nvme_reset_done(struct pci_dev *pdev) |
2868 | { | |
f263fbb8 | 2869 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
c1ac9a4b KB |
2870 | |
2871 | if (!nvme_try_sched_reset(&dev->ctrl)) | |
2872 | flush_work(&dev->ctrl.reset_work); | |
f0d54a54 KB |
2873 | } |
2874 | ||
09ece142 KB |
2875 | static void nvme_shutdown(struct pci_dev *pdev) |
2876 | { | |
2877 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
4e523547 | 2878 | |
c1ac9a4b | 2879 | nvme_disable_prepare_reset(dev, true); |
09ece142 KB |
2880 | } |
2881 | ||
f58944e2 KB |
2882 | /* |
2883 | * The driver's remove may be called on a device in a partially initialized | |
2884 | * state. This function must not have any dependencies on the device state in | |
2885 | * order to proceed. | |
2886 | */ | |
8d85fce7 | 2887 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
2888 | { |
2889 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 | 2890 | |
bb8d261e | 2891 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); |
9a6b9458 | 2892 | pci_set_drvdata(pdev, NULL); |
0ff9d4e1 | 2893 | |
6db28eda | 2894 | if (!pci_device_is_present(pdev)) { |
0ff9d4e1 | 2895 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD); |
1d39e692 | 2896 | nvme_dev_disable(dev, true); |
cb4bfda6 | 2897 | nvme_dev_remove_admin(dev); |
6db28eda | 2898 | } |
0ff9d4e1 | 2899 | |
d86c4d8e | 2900 | flush_work(&dev->ctrl.reset_work); |
d09f2b45 SG |
2901 | nvme_stop_ctrl(&dev->ctrl); |
2902 | nvme_remove_namespaces(&dev->ctrl); | |
a5cdb68c | 2903 | nvme_dev_disable(dev, true); |
9fe5c59f | 2904 | nvme_release_cmb(dev); |
87ad72a5 | 2905 | nvme_free_host_mem(dev); |
a4aea562 | 2906 | nvme_dev_remove_admin(dev); |
a1a5ef99 | 2907 | nvme_free_queues(dev, 0); |
9a6b9458 | 2908 | nvme_release_prp_pools(dev); |
b00a726a | 2909 | nvme_dev_unmap(dev); |
726612b6 | 2910 | nvme_uninit_ctrl(&dev->ctrl); |
b60503ba MW |
2911 | } |
2912 | ||
671a6018 | 2913 | #ifdef CONFIG_PM_SLEEP |
d916b1be KB |
2914 | static int nvme_get_power_state(struct nvme_ctrl *ctrl, u32 *ps) |
2915 | { | |
2916 | return nvme_get_features(ctrl, NVME_FEAT_POWER_MGMT, 0, NULL, 0, ps); | |
2917 | } | |
2918 | ||
2919 | static int nvme_set_power_state(struct nvme_ctrl *ctrl, u32 ps) | |
2920 | { | |
2921 | return nvme_set_features(ctrl, NVME_FEAT_POWER_MGMT, ps, NULL, 0, NULL); | |
2922 | } | |
2923 | ||
2924 | static int nvme_resume(struct device *dev) | |
2925 | { | |
2926 | struct nvme_dev *ndev = pci_get_drvdata(to_pci_dev(dev)); | |
2927 | struct nvme_ctrl *ctrl = &ndev->ctrl; | |
2928 | ||
4eaefe8c | 2929 | if (ndev->last_ps == U32_MAX || |
d916b1be | 2930 | nvme_set_power_state(ctrl, ndev->last_ps) != 0) |
c1ac9a4b | 2931 | return nvme_try_sched_reset(&ndev->ctrl); |
d916b1be KB |
2932 | return 0; |
2933 | } | |
2934 | ||
cd638946 KB |
2935 | static int nvme_suspend(struct device *dev) |
2936 | { | |
2937 | struct pci_dev *pdev = to_pci_dev(dev); | |
2938 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
d916b1be KB |
2939 | struct nvme_ctrl *ctrl = &ndev->ctrl; |
2940 | int ret = -EBUSY; | |
2941 | ||
4eaefe8c RW |
2942 | ndev->last_ps = U32_MAX; |
2943 | ||
d916b1be KB |
2944 | /* |
2945 | * The platform does not remove power for a kernel managed suspend so | |
2946 | * use host managed nvme power settings for lowest idle power if | |
2947 | * possible. This should have quicker resume latency than a full device | |
2948 | * shutdown. But if the firmware is involved after the suspend or the | |
2949 | * device does not support any non-default power states, shut down the | |
2950 | * device fully. | |
4eaefe8c RW |
2951 | * |
2952 | * If ASPM is not enabled for the device, shut down the device and allow | |
2953 | * the PCI bus layer to put it into D3 in order to take the PCIe link | |
2954 | * down, so as to allow the platform to achieve its minimum low-power | |
2955 | * state (which may not be possible if the link is up). | |
b97120b1 CH |
2956 | * |
2957 | * If a host memory buffer is enabled, shut down the device as the NVMe | |
2958 | * specification allows the device to access the host memory buffer in | |
2959 | * host DRAM from all power states, but hosts will fail access to DRAM | |
2960 | * during S3. | |
d916b1be | 2961 | */ |
4eaefe8c | 2962 | if (pm_suspend_via_firmware() || !ctrl->npss || |
cb32de1b | 2963 | !pcie_aspm_enabled(pdev) || |
b97120b1 | 2964 | ndev->nr_host_mem_descs || |
c1ac9a4b KB |
2965 | (ndev->ctrl.quirks & NVME_QUIRK_SIMPLE_SUSPEND)) |
2966 | return nvme_disable_prepare_reset(ndev, true); | |
d916b1be KB |
2967 | |
2968 | nvme_start_freeze(ctrl); | |
2969 | nvme_wait_freeze(ctrl); | |
2970 | nvme_sync_queues(ctrl); | |
2971 | ||
5d02a5c1 | 2972 | if (ctrl->state != NVME_CTRL_LIVE) |
d916b1be KB |
2973 | goto unfreeze; |
2974 | ||
d916b1be KB |
2975 | ret = nvme_get_power_state(ctrl, &ndev->last_ps); |
2976 | if (ret < 0) | |
2977 | goto unfreeze; | |
2978 | ||
7cbb5c6f ML |
2979 | /* |
2980 | * A saved state prevents pci pm from generically controlling the | |
2981 | * device's power. If we're using protocol specific settings, we don't | |
2982 | * want pci interfering. | |
2983 | */ | |
2984 | pci_save_state(pdev); | |
2985 | ||
d916b1be KB |
2986 | ret = nvme_set_power_state(ctrl, ctrl->npss); |
2987 | if (ret < 0) | |
2988 | goto unfreeze; | |
2989 | ||
2990 | if (ret) { | |
7cbb5c6f ML |
2991 | /* discard the saved state */ |
2992 | pci_load_saved_state(pdev, NULL); | |
2993 | ||
d916b1be KB |
2994 | /* |
2995 | * Clearing npss forces a controller reset on resume. The | |
05d3046f | 2996 | * correct value will be rediscovered then. |
d916b1be | 2997 | */ |
c1ac9a4b | 2998 | ret = nvme_disable_prepare_reset(ndev, true); |
d916b1be | 2999 | ctrl->npss = 0; |
d916b1be | 3000 | } |
d916b1be KB |
3001 | unfreeze: |
3002 | nvme_unfreeze(ctrl); | |
3003 | return ret; | |
3004 | } | |
3005 | ||
3006 | static int nvme_simple_suspend(struct device *dev) | |
3007 | { | |
3008 | struct nvme_dev *ndev = pci_get_drvdata(to_pci_dev(dev)); | |
4e523547 | 3009 | |
c1ac9a4b | 3010 | return nvme_disable_prepare_reset(ndev, true); |
cd638946 KB |
3011 | } |
3012 | ||
d916b1be | 3013 | static int nvme_simple_resume(struct device *dev) |
cd638946 KB |
3014 | { |
3015 | struct pci_dev *pdev = to_pci_dev(dev); | |
3016 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 3017 | |
c1ac9a4b | 3018 | return nvme_try_sched_reset(&ndev->ctrl); |
cd638946 KB |
3019 | } |
3020 | ||
21774222 | 3021 | static const struct dev_pm_ops nvme_dev_pm_ops = { |
d916b1be KB |
3022 | .suspend = nvme_suspend, |
3023 | .resume = nvme_resume, | |
3024 | .freeze = nvme_simple_suspend, | |
3025 | .thaw = nvme_simple_resume, | |
3026 | .poweroff = nvme_simple_suspend, | |
3027 | .restore = nvme_simple_resume, | |
3028 | }; | |
3029 | #endif /* CONFIG_PM_SLEEP */ | |
b60503ba | 3030 | |
a0a3408e KB |
3031 | static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev, |
3032 | pci_channel_state_t state) | |
3033 | { | |
3034 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
3035 | ||
3036 | /* | |
3037 | * A frozen channel requires a reset. When detected, this method will | |
3038 | * shutdown the controller to quiesce. The controller will be restarted | |
3039 | * after the slot reset through driver's slot_reset callback. | |
3040 | */ | |
a0a3408e KB |
3041 | switch (state) { |
3042 | case pci_channel_io_normal: | |
3043 | return PCI_ERS_RESULT_CAN_RECOVER; | |
3044 | case pci_channel_io_frozen: | |
d011fb31 KB |
3045 | dev_warn(dev->ctrl.device, |
3046 | "frozen state error detected, reset controller\n"); | |
a5cdb68c | 3047 | nvme_dev_disable(dev, false); |
a0a3408e KB |
3048 | return PCI_ERS_RESULT_NEED_RESET; |
3049 | case pci_channel_io_perm_failure: | |
d011fb31 KB |
3050 | dev_warn(dev->ctrl.device, |
3051 | "failure state error detected, request disconnect\n"); | |
a0a3408e KB |
3052 | return PCI_ERS_RESULT_DISCONNECT; |
3053 | } | |
3054 | return PCI_ERS_RESULT_NEED_RESET; | |
3055 | } | |
3056 | ||
3057 | static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev) | |
3058 | { | |
3059 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
3060 | ||
1b3c47c1 | 3061 | dev_info(dev->ctrl.device, "restart after slot reset\n"); |
a0a3408e | 3062 | pci_restore_state(pdev); |
d86c4d8e | 3063 | nvme_reset_ctrl(&dev->ctrl); |
a0a3408e KB |
3064 | return PCI_ERS_RESULT_RECOVERED; |
3065 | } | |
3066 | ||
3067 | static void nvme_error_resume(struct pci_dev *pdev) | |
3068 | { | |
72cd4cc2 KB |
3069 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
3070 | ||
3071 | flush_work(&dev->ctrl.reset_work); | |
a0a3408e KB |
3072 | } |
3073 | ||
1d352035 | 3074 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba | 3075 | .error_detected = nvme_error_detected, |
b60503ba MW |
3076 | .slot_reset = nvme_slot_reset, |
3077 | .resume = nvme_error_resume, | |
775755ed CH |
3078 | .reset_prepare = nvme_reset_prepare, |
3079 | .reset_done = nvme_reset_done, | |
b60503ba MW |
3080 | }; |
3081 | ||
6eb0d698 | 3082 | static const struct pci_device_id nvme_id_table[] = { |
972b13e2 | 3083 | { PCI_VDEVICE(INTEL, 0x0953), /* Intel 750/P3500/P3600/P3700 */ |
08095e70 | 3084 | .driver_data = NVME_QUIRK_STRIPE_SIZE | |
e850fd16 | 3085 | NVME_QUIRK_DEALLOCATE_ZEROES, }, |
972b13e2 | 3086 | { PCI_VDEVICE(INTEL, 0x0a53), /* Intel P3520 */ |
99466e70 | 3087 | .driver_data = NVME_QUIRK_STRIPE_SIZE | |
e850fd16 | 3088 | NVME_QUIRK_DEALLOCATE_ZEROES, }, |
972b13e2 | 3089 | { PCI_VDEVICE(INTEL, 0x0a54), /* Intel P4500/P4600 */ |
99466e70 | 3090 | .driver_data = NVME_QUIRK_STRIPE_SIZE | |
e850fd16 | 3091 | NVME_QUIRK_DEALLOCATE_ZEROES, }, |
972b13e2 | 3092 | { PCI_VDEVICE(INTEL, 0x0a55), /* Dell Express Flash P4600 */ |
f99cb7af DWF |
3093 | .driver_data = NVME_QUIRK_STRIPE_SIZE | |
3094 | NVME_QUIRK_DEALLOCATE_ZEROES, }, | |
50af47d0 | 3095 | { PCI_VDEVICE(INTEL, 0xf1a5), /* Intel 600P/P3100 */ |
9abd68ef | 3096 | .driver_data = NVME_QUIRK_NO_DEEPEST_PS | |
6c6aa2f2 AM |
3097 | NVME_QUIRK_MEDIUM_PRIO_SQ | |
3098 | NVME_QUIRK_NO_TEMP_THRESH_CHANGE }, | |
6299358d JD |
3099 | { PCI_VDEVICE(INTEL, 0xf1a6), /* Intel 760p/Pro 7600p */ |
3100 | .driver_data = NVME_QUIRK_IGNORE_DEV_SUBNQN, }, | |
540c801c | 3101 | { PCI_VDEVICE(INTEL, 0x5845), /* Qemu emulated controller */ |
7b210e4e CH |
3102 | .driver_data = NVME_QUIRK_IDENTIFY_CNS | |
3103 | NVME_QUIRK_DISABLE_WRITE_ZEROES, }, | |
0302ae60 MP |
3104 | { PCI_DEVICE(0x1bb1, 0x0100), /* Seagate Nytro Flash Storage */ |
3105 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
54adc010 GP |
3106 | { PCI_DEVICE(0x1c58, 0x0003), /* HGST adapter */ |
3107 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
8c97eecc JL |
3108 | { PCI_DEVICE(0x1c58, 0x0023), /* WDC SN200 adapter */ |
3109 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
015282c9 WW |
3110 | { PCI_DEVICE(0x1c5f, 0x0540), /* Memblaze Pblaze4 adapter */ |
3111 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
d554b5e1 MP |
3112 | { PCI_DEVICE(0x144d, 0xa821), /* Samsung PM1725 */ |
3113 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
3114 | { PCI_DEVICE(0x144d, 0xa822), /* Samsung PM1725a */ | |
3115 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
608cc4b1 CH |
3116 | { PCI_DEVICE(0x1d1d, 0x1f1f), /* LighNVM qemu device */ |
3117 | .driver_data = NVME_QUIRK_LIGHTNVM, }, | |
3118 | { PCI_DEVICE(0x1d1d, 0x2807), /* CNEX WL */ | |
3119 | .driver_data = NVME_QUIRK_LIGHTNVM, }, | |
ea48e877 WX |
3120 | { PCI_DEVICE(0x1d1d, 0x2601), /* CNEX Granby */ |
3121 | .driver_data = NVME_QUIRK_LIGHTNVM, }, | |
08b903b5 MN |
3122 | { PCI_DEVICE(0x10ec, 0x5762), /* ADATA SX6000LNP */ |
3123 | .driver_data = NVME_QUIRK_IGNORE_DEV_SUBNQN, }, | |
f03e42c6 GC |
3124 | { PCI_DEVICE(0x1cc1, 0x8201), /* ADATA SX8200PNP 512GB */ |
3125 | .driver_data = NVME_QUIRK_NO_DEEPEST_PS | | |
3126 | NVME_QUIRK_IGNORE_DEV_SUBNQN, }, | |
b60503ba | 3127 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
98f7b86a AS |
3128 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001), |
3129 | .driver_data = NVME_QUIRK_SINGLE_VECTOR }, | |
124298bd | 3130 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) }, |
66341331 BH |
3131 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2005), |
3132 | .driver_data = NVME_QUIRK_SINGLE_VECTOR | | |
d38e9f04 BH |
3133 | NVME_QUIRK_128_BYTES_SQES | |
3134 | NVME_QUIRK_SHARED_TAGS }, | |
b60503ba MW |
3135 | { 0, } |
3136 | }; | |
3137 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
3138 | ||
3139 | static struct pci_driver nvme_driver = { | |
3140 | .name = "nvme", | |
3141 | .id_table = nvme_id_table, | |
3142 | .probe = nvme_probe, | |
8d85fce7 | 3143 | .remove = nvme_remove, |
09ece142 | 3144 | .shutdown = nvme_shutdown, |
d916b1be | 3145 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
3146 | .driver = { |
3147 | .pm = &nvme_dev_pm_ops, | |
3148 | }, | |
d916b1be | 3149 | #endif |
74d986ab | 3150 | .sriov_configure = pci_sriov_configure_simple, |
b60503ba MW |
3151 | .err_handler = &nvme_err_handler, |
3152 | }; | |
3153 | ||
3154 | static int __init nvme_init(void) | |
3155 | { | |
81101540 CH |
3156 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); |
3157 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
3158 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
612b7286 | 3159 | BUILD_BUG_ON(IRQ_AFFINITY_MAX_SETS < 2); |
17c33167 | 3160 | |
9a6327d2 | 3161 | return pci_register_driver(&nvme_driver); |
b60503ba MW |
3162 | } |
3163 | ||
3164 | static void __exit nvme_exit(void) | |
3165 | { | |
3166 | pci_unregister_driver(&nvme_driver); | |
03e0f3a6 | 3167 | flush_workqueue(nvme_wq); |
b60503ba MW |
3168 | } |
3169 | ||
3170 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
3171 | MODULE_LICENSE("GPL"); | |
c78b4713 | 3172 | MODULE_VERSION("1.0"); |
b60503ba MW |
3173 | module_init(nvme_init); |
3174 | module_exit(nvme_exit); |