]>
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, | |
64 | .get = param_get_int, | |
65 | }; | |
66 | ||
67 | static int io_queue_depth = 1024; | |
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 | ||
3f68baf7 KB |
71 | static unsigned int write_queues; |
72 | module_param(write_queues, uint, 0644); | |
3b6592f7 JA |
73 | MODULE_PARM_DESC(write_queues, |
74 | "Number of queues to use for writes. If not set, reads and writes " | |
75 | "will share a queue set."); | |
76 | ||
3f68baf7 KB |
77 | static unsigned int poll_queues; |
78 | module_param(poll_queues, uint, 0644); | |
4b04cc6a JA |
79 | MODULE_PARM_DESC(poll_queues, "Number of queues to use for polled IO."); |
80 | ||
1c63dc66 CH |
81 | struct nvme_dev; |
82 | struct nvme_queue; | |
b3fffdef | 83 | |
a5cdb68c | 84 | static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown); |
8fae268b | 85 | static bool __nvme_disable_io_queues(struct nvme_dev *dev, u8 opcode); |
d4b4ff8e | 86 | |
1c63dc66 CH |
87 | /* |
88 | * Represents an NVM Express device. Each nvme_dev is a PCI function. | |
89 | */ | |
90 | struct nvme_dev { | |
147b27e4 | 91 | struct nvme_queue *queues; |
1c63dc66 CH |
92 | struct blk_mq_tag_set tagset; |
93 | struct blk_mq_tag_set admin_tagset; | |
94 | u32 __iomem *dbs; | |
95 | struct device *dev; | |
96 | struct dma_pool *prp_page_pool; | |
97 | struct dma_pool *prp_small_pool; | |
1c63dc66 CH |
98 | unsigned online_queues; |
99 | unsigned max_qid; | |
e20ba6e1 | 100 | unsigned io_queues[HCTX_MAX_TYPES]; |
22b55601 | 101 | unsigned int num_vecs; |
1c63dc66 | 102 | int q_depth; |
c1e0cc7e | 103 | int io_sqes; |
1c63dc66 | 104 | u32 db_stride; |
1c63dc66 | 105 | void __iomem *bar; |
97f6ef64 | 106 | unsigned long bar_mapped_size; |
5c8809e6 | 107 | struct work_struct remove_work; |
77bf25ea | 108 | struct mutex shutdown_lock; |
1c63dc66 | 109 | bool subsystem; |
1c63dc66 | 110 | u64 cmb_size; |
0f238ff5 | 111 | bool cmb_use_sqes; |
1c63dc66 | 112 | u32 cmbsz; |
202021c1 | 113 | u32 cmbloc; |
1c63dc66 | 114 | struct nvme_ctrl ctrl; |
d916b1be | 115 | u32 last_ps; |
87ad72a5 | 116 | |
943e942e JA |
117 | mempool_t *iod_mempool; |
118 | ||
87ad72a5 | 119 | /* shadow doorbell buffer support: */ |
f9f38e33 HK |
120 | u32 *dbbuf_dbs; |
121 | dma_addr_t dbbuf_dbs_dma_addr; | |
122 | u32 *dbbuf_eis; | |
123 | dma_addr_t dbbuf_eis_dma_addr; | |
87ad72a5 CH |
124 | |
125 | /* host memory buffer support: */ | |
126 | u64 host_mem_size; | |
127 | u32 nr_host_mem_descs; | |
4033f35d | 128 | dma_addr_t host_mem_descs_dma; |
87ad72a5 CH |
129 | struct nvme_host_mem_buf_desc *host_mem_descs; |
130 | void **host_mem_desc_bufs; | |
4d115420 | 131 | }; |
1fa6aead | 132 | |
b27c1e68 | 133 | static int io_queue_depth_set(const char *val, const struct kernel_param *kp) |
134 | { | |
135 | int n = 0, ret; | |
136 | ||
137 | ret = kstrtoint(val, 10, &n); | |
138 | if (ret != 0 || n < 2) | |
139 | return -EINVAL; | |
140 | ||
141 | return param_set_int(val, kp); | |
142 | } | |
143 | ||
f9f38e33 HK |
144 | static inline unsigned int sq_idx(unsigned int qid, u32 stride) |
145 | { | |
146 | return qid * 2 * stride; | |
147 | } | |
148 | ||
149 | static inline unsigned int cq_idx(unsigned int qid, u32 stride) | |
150 | { | |
151 | return (qid * 2 + 1) * stride; | |
152 | } | |
153 | ||
1c63dc66 CH |
154 | static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl) |
155 | { | |
156 | return container_of(ctrl, struct nvme_dev, ctrl); | |
157 | } | |
158 | ||
b60503ba MW |
159 | /* |
160 | * An NVM Express queue. Each device has at least two (one for admin | |
161 | * commands and one for I/O commands). | |
162 | */ | |
163 | struct nvme_queue { | |
091b6092 | 164 | struct nvme_dev *dev; |
1ab0cd69 | 165 | spinlock_t sq_lock; |
c1e0cc7e | 166 | void *sq_cmds; |
3a7afd8e CH |
167 | /* only used for poll queues: */ |
168 | spinlock_t cq_poll_lock ____cacheline_aligned_in_smp; | |
b60503ba MW |
169 | volatile struct nvme_completion *cqes; |
170 | dma_addr_t sq_dma_addr; | |
171 | dma_addr_t cq_dma_addr; | |
b60503ba MW |
172 | u32 __iomem *q_db; |
173 | u16 q_depth; | |
7c349dde | 174 | u16 cq_vector; |
b60503ba | 175 | u16 sq_tail; |
04f3eafd | 176 | u16 last_sq_tail; |
b60503ba | 177 | u16 cq_head; |
c30341dc | 178 | u16 qid; |
e9539f47 | 179 | u8 cq_phase; |
c1e0cc7e | 180 | u8 sqes; |
4e224106 CH |
181 | unsigned long flags; |
182 | #define NVMEQ_ENABLED 0 | |
63223078 | 183 | #define NVMEQ_SQ_CMB 1 |
d1ed6aa1 | 184 | #define NVMEQ_DELETE_ERROR 2 |
7c349dde | 185 | #define NVMEQ_POLLED 3 |
f9f38e33 HK |
186 | u32 *dbbuf_sq_db; |
187 | u32 *dbbuf_cq_db; | |
188 | u32 *dbbuf_sq_ei; | |
189 | u32 *dbbuf_cq_ei; | |
d1ed6aa1 | 190 | struct completion delete_done; |
b60503ba MW |
191 | }; |
192 | ||
71bd150c | 193 | /* |
9b048119 CH |
194 | * The nvme_iod describes the data in an I/O. |
195 | * | |
196 | * The sg pointer contains the list of PRP/SGL chunk allocations in addition | |
197 | * to the actual struct scatterlist. | |
71bd150c CH |
198 | */ |
199 | struct nvme_iod { | |
d49187e9 | 200 | struct nvme_request req; |
f4800d6d | 201 | struct nvme_queue *nvmeq; |
a7a7cbe3 | 202 | bool use_sgl; |
f4800d6d | 203 | int aborted; |
71bd150c | 204 | int npages; /* In the PRP list. 0 means small pool in use */ |
71bd150c | 205 | int nents; /* Used in scatterlist */ |
71bd150c | 206 | dma_addr_t first_dma; |
dff824b2 | 207 | unsigned int dma_len; /* length of single DMA segment mapping */ |
783b94bd | 208 | dma_addr_t meta_dma; |
f4800d6d | 209 | struct scatterlist *sg; |
b60503ba MW |
210 | }; |
211 | ||
3b6592f7 JA |
212 | static unsigned int max_io_queues(void) |
213 | { | |
4b04cc6a | 214 | return num_possible_cpus() + write_queues + poll_queues; |
3b6592f7 JA |
215 | } |
216 | ||
217 | static unsigned int max_queue_count(void) | |
218 | { | |
219 | /* IO queues + admin queue */ | |
220 | return 1 + max_io_queues(); | |
221 | } | |
222 | ||
f9f38e33 HK |
223 | static inline unsigned int nvme_dbbuf_size(u32 stride) |
224 | { | |
3b6592f7 | 225 | return (max_queue_count() * 8 * stride); |
f9f38e33 HK |
226 | } |
227 | ||
228 | static int nvme_dbbuf_dma_alloc(struct nvme_dev *dev) | |
229 | { | |
230 | unsigned int mem_size = nvme_dbbuf_size(dev->db_stride); | |
231 | ||
232 | if (dev->dbbuf_dbs) | |
233 | return 0; | |
234 | ||
235 | dev->dbbuf_dbs = dma_alloc_coherent(dev->dev, mem_size, | |
236 | &dev->dbbuf_dbs_dma_addr, | |
237 | GFP_KERNEL); | |
238 | if (!dev->dbbuf_dbs) | |
239 | return -ENOMEM; | |
240 | dev->dbbuf_eis = dma_alloc_coherent(dev->dev, mem_size, | |
241 | &dev->dbbuf_eis_dma_addr, | |
242 | GFP_KERNEL); | |
243 | if (!dev->dbbuf_eis) { | |
244 | dma_free_coherent(dev->dev, mem_size, | |
245 | dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr); | |
246 | dev->dbbuf_dbs = NULL; | |
247 | return -ENOMEM; | |
248 | } | |
249 | ||
250 | return 0; | |
251 | } | |
252 | ||
253 | static void nvme_dbbuf_dma_free(struct nvme_dev *dev) | |
254 | { | |
255 | unsigned int mem_size = nvme_dbbuf_size(dev->db_stride); | |
256 | ||
257 | if (dev->dbbuf_dbs) { | |
258 | dma_free_coherent(dev->dev, mem_size, | |
259 | dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr); | |
260 | dev->dbbuf_dbs = NULL; | |
261 | } | |
262 | if (dev->dbbuf_eis) { | |
263 | dma_free_coherent(dev->dev, mem_size, | |
264 | dev->dbbuf_eis, dev->dbbuf_eis_dma_addr); | |
265 | dev->dbbuf_eis = NULL; | |
266 | } | |
267 | } | |
268 | ||
269 | static void nvme_dbbuf_init(struct nvme_dev *dev, | |
270 | struct nvme_queue *nvmeq, int qid) | |
271 | { | |
272 | if (!dev->dbbuf_dbs || !qid) | |
273 | return; | |
274 | ||
275 | nvmeq->dbbuf_sq_db = &dev->dbbuf_dbs[sq_idx(qid, dev->db_stride)]; | |
276 | nvmeq->dbbuf_cq_db = &dev->dbbuf_dbs[cq_idx(qid, dev->db_stride)]; | |
277 | nvmeq->dbbuf_sq_ei = &dev->dbbuf_eis[sq_idx(qid, dev->db_stride)]; | |
278 | nvmeq->dbbuf_cq_ei = &dev->dbbuf_eis[cq_idx(qid, dev->db_stride)]; | |
279 | } | |
280 | ||
281 | static void nvme_dbbuf_set(struct nvme_dev *dev) | |
282 | { | |
283 | struct nvme_command c; | |
284 | ||
285 | if (!dev->dbbuf_dbs) | |
286 | return; | |
287 | ||
288 | memset(&c, 0, sizeof(c)); | |
289 | c.dbbuf.opcode = nvme_admin_dbbuf; | |
290 | c.dbbuf.prp1 = cpu_to_le64(dev->dbbuf_dbs_dma_addr); | |
291 | c.dbbuf.prp2 = cpu_to_le64(dev->dbbuf_eis_dma_addr); | |
292 | ||
293 | if (nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0)) { | |
9bdcfb10 | 294 | dev_warn(dev->ctrl.device, "unable to set dbbuf\n"); |
f9f38e33 HK |
295 | /* Free memory and continue on */ |
296 | nvme_dbbuf_dma_free(dev); | |
297 | } | |
298 | } | |
299 | ||
300 | static inline int nvme_dbbuf_need_event(u16 event_idx, u16 new_idx, u16 old) | |
301 | { | |
302 | return (u16)(new_idx - event_idx - 1) < (u16)(new_idx - old); | |
303 | } | |
304 | ||
305 | /* Update dbbuf and return true if an MMIO is required */ | |
306 | static bool nvme_dbbuf_update_and_check_event(u16 value, u32 *dbbuf_db, | |
307 | volatile u32 *dbbuf_ei) | |
308 | { | |
309 | if (dbbuf_db) { | |
310 | u16 old_value; | |
311 | ||
312 | /* | |
313 | * Ensure that the queue is written before updating | |
314 | * the doorbell in memory | |
315 | */ | |
316 | wmb(); | |
317 | ||
318 | old_value = *dbbuf_db; | |
319 | *dbbuf_db = value; | |
320 | ||
f1ed3df2 MW |
321 | /* |
322 | * Ensure that the doorbell is updated before reading the event | |
323 | * index from memory. The controller needs to provide similar | |
324 | * ordering to ensure the envent index is updated before reading | |
325 | * the doorbell. | |
326 | */ | |
327 | mb(); | |
328 | ||
f9f38e33 HK |
329 | if (!nvme_dbbuf_need_event(*dbbuf_ei, value, old_value)) |
330 | return false; | |
331 | } | |
332 | ||
333 | return true; | |
b60503ba MW |
334 | } |
335 | ||
ac3dd5bd JA |
336 | /* |
337 | * Will slightly overestimate the number of pages needed. This is OK | |
338 | * as it only leads to a small amount of wasted memory for the lifetime of | |
339 | * the I/O. | |
340 | */ | |
341 | static int nvme_npages(unsigned size, struct nvme_dev *dev) | |
342 | { | |
5fd4ce1b CH |
343 | unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size, |
344 | dev->ctrl.page_size); | |
ac3dd5bd JA |
345 | return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); |
346 | } | |
347 | ||
a7a7cbe3 CK |
348 | /* |
349 | * Calculates the number of pages needed for the SGL segments. For example a 4k | |
350 | * page can accommodate 256 SGL descriptors. | |
351 | */ | |
352 | static int nvme_pci_npages_sgl(unsigned int num_seg) | |
ac3dd5bd | 353 | { |
a7a7cbe3 | 354 | return DIV_ROUND_UP(num_seg * sizeof(struct nvme_sgl_desc), PAGE_SIZE); |
f4800d6d | 355 | } |
ac3dd5bd | 356 | |
a7a7cbe3 CK |
357 | static unsigned int nvme_pci_iod_alloc_size(struct nvme_dev *dev, |
358 | unsigned int size, unsigned int nseg, bool use_sgl) | |
f4800d6d | 359 | { |
a7a7cbe3 CK |
360 | size_t alloc_size; |
361 | ||
362 | if (use_sgl) | |
363 | alloc_size = sizeof(__le64 *) * nvme_pci_npages_sgl(nseg); | |
364 | else | |
365 | alloc_size = sizeof(__le64 *) * nvme_npages(size, dev); | |
366 | ||
367 | return alloc_size + sizeof(struct scatterlist) * nseg; | |
f4800d6d | 368 | } |
ac3dd5bd | 369 | |
a4aea562 MB |
370 | static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
371 | unsigned int hctx_idx) | |
e85248e5 | 372 | { |
a4aea562 | 373 | struct nvme_dev *dev = data; |
147b27e4 | 374 | struct nvme_queue *nvmeq = &dev->queues[0]; |
a4aea562 | 375 | |
42483228 KB |
376 | WARN_ON(hctx_idx != 0); |
377 | WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); | |
42483228 | 378 | |
a4aea562 MB |
379 | hctx->driver_data = nvmeq; |
380 | return 0; | |
e85248e5 MW |
381 | } |
382 | ||
a4aea562 MB |
383 | static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, |
384 | unsigned int hctx_idx) | |
b60503ba | 385 | { |
a4aea562 | 386 | struct nvme_dev *dev = data; |
147b27e4 | 387 | struct nvme_queue *nvmeq = &dev->queues[hctx_idx + 1]; |
a4aea562 | 388 | |
42483228 | 389 | WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); |
a4aea562 MB |
390 | hctx->driver_data = nvmeq; |
391 | return 0; | |
b60503ba MW |
392 | } |
393 | ||
d6296d39 CH |
394 | static int nvme_init_request(struct blk_mq_tag_set *set, struct request *req, |
395 | unsigned int hctx_idx, unsigned int numa_node) | |
b60503ba | 396 | { |
d6296d39 | 397 | struct nvme_dev *dev = set->driver_data; |
f4800d6d | 398 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
0350815a | 399 | int queue_idx = (set == &dev->tagset) ? hctx_idx + 1 : 0; |
147b27e4 | 400 | struct nvme_queue *nvmeq = &dev->queues[queue_idx]; |
a4aea562 MB |
401 | |
402 | BUG_ON(!nvmeq); | |
f4800d6d | 403 | iod->nvmeq = nvmeq; |
59e29ce6 SG |
404 | |
405 | nvme_req(req)->ctrl = &dev->ctrl; | |
a4aea562 MB |
406 | return 0; |
407 | } | |
408 | ||
3b6592f7 JA |
409 | static int queue_irq_offset(struct nvme_dev *dev) |
410 | { | |
411 | /* if we have more than 1 vec, admin queue offsets us by 1 */ | |
412 | if (dev->num_vecs > 1) | |
413 | return 1; | |
414 | ||
415 | return 0; | |
416 | } | |
417 | ||
dca51e78 CH |
418 | static int nvme_pci_map_queues(struct blk_mq_tag_set *set) |
419 | { | |
420 | struct nvme_dev *dev = set->driver_data; | |
3b6592f7 JA |
421 | int i, qoff, offset; |
422 | ||
423 | offset = queue_irq_offset(dev); | |
424 | for (i = 0, qoff = 0; i < set->nr_maps; i++) { | |
425 | struct blk_mq_queue_map *map = &set->map[i]; | |
426 | ||
427 | map->nr_queues = dev->io_queues[i]; | |
428 | if (!map->nr_queues) { | |
e20ba6e1 | 429 | BUG_ON(i == HCTX_TYPE_DEFAULT); |
7e849dd9 | 430 | continue; |
3b6592f7 JA |
431 | } |
432 | ||
4b04cc6a JA |
433 | /* |
434 | * The poll queue(s) doesn't have an IRQ (and hence IRQ | |
435 | * affinity), so use the regular blk-mq cpu mapping | |
436 | */ | |
3b6592f7 | 437 | map->queue_offset = qoff; |
cb9e0e50 | 438 | if (i != HCTX_TYPE_POLL && offset) |
4b04cc6a JA |
439 | blk_mq_pci_map_queues(map, to_pci_dev(dev->dev), offset); |
440 | else | |
441 | blk_mq_map_queues(map); | |
3b6592f7 JA |
442 | qoff += map->nr_queues; |
443 | offset += map->nr_queues; | |
444 | } | |
445 | ||
446 | return 0; | |
dca51e78 CH |
447 | } |
448 | ||
04f3eafd JA |
449 | /* |
450 | * Write sq tail if we are asked to, or if the next command would wrap. | |
451 | */ | |
452 | static inline void nvme_write_sq_db(struct nvme_queue *nvmeq, bool write_sq) | |
453 | { | |
454 | if (!write_sq) { | |
455 | u16 next_tail = nvmeq->sq_tail + 1; | |
456 | ||
457 | if (next_tail == nvmeq->q_depth) | |
458 | next_tail = 0; | |
459 | if (next_tail != nvmeq->last_sq_tail) | |
460 | return; | |
461 | } | |
462 | ||
463 | if (nvme_dbbuf_update_and_check_event(nvmeq->sq_tail, | |
464 | nvmeq->dbbuf_sq_db, nvmeq->dbbuf_sq_ei)) | |
465 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
466 | nvmeq->last_sq_tail = nvmeq->sq_tail; | |
467 | } | |
468 | ||
b60503ba | 469 | /** |
90ea5ca4 | 470 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
471 | * @nvmeq: The queue to use |
472 | * @cmd: The command to send | |
04f3eafd | 473 | * @write_sq: whether to write to the SQ doorbell |
b60503ba | 474 | */ |
04f3eafd JA |
475 | static void nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd, |
476 | bool write_sq) | |
b60503ba | 477 | { |
90ea5ca4 | 478 | spin_lock(&nvmeq->sq_lock); |
c1e0cc7e BH |
479 | memcpy(nvmeq->sq_cmds + (nvmeq->sq_tail << nvmeq->sqes), |
480 | cmd, sizeof(*cmd)); | |
90ea5ca4 CH |
481 | if (++nvmeq->sq_tail == nvmeq->q_depth) |
482 | nvmeq->sq_tail = 0; | |
04f3eafd JA |
483 | nvme_write_sq_db(nvmeq, write_sq); |
484 | spin_unlock(&nvmeq->sq_lock); | |
485 | } | |
486 | ||
487 | static void nvme_commit_rqs(struct blk_mq_hw_ctx *hctx) | |
488 | { | |
489 | struct nvme_queue *nvmeq = hctx->driver_data; | |
490 | ||
491 | spin_lock(&nvmeq->sq_lock); | |
492 | if (nvmeq->sq_tail != nvmeq->last_sq_tail) | |
493 | nvme_write_sq_db(nvmeq, true); | |
90ea5ca4 | 494 | spin_unlock(&nvmeq->sq_lock); |
b60503ba MW |
495 | } |
496 | ||
a7a7cbe3 | 497 | static void **nvme_pci_iod_list(struct request *req) |
b60503ba | 498 | { |
f4800d6d | 499 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
a7a7cbe3 | 500 | return (void **)(iod->sg + blk_rq_nr_phys_segments(req)); |
b60503ba MW |
501 | } |
502 | ||
955b1b5a MI |
503 | static inline bool nvme_pci_use_sgls(struct nvme_dev *dev, struct request *req) |
504 | { | |
505 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
20469a37 | 506 | int nseg = blk_rq_nr_phys_segments(req); |
955b1b5a MI |
507 | unsigned int avg_seg_size; |
508 | ||
20469a37 KB |
509 | if (nseg == 0) |
510 | return false; | |
511 | ||
512 | avg_seg_size = DIV_ROUND_UP(blk_rq_payload_bytes(req), nseg); | |
955b1b5a MI |
513 | |
514 | if (!(dev->ctrl.sgls & ((1 << 0) | (1 << 1)))) | |
515 | return false; | |
516 | if (!iod->nvmeq->qid) | |
517 | return false; | |
518 | if (!sgl_threshold || avg_seg_size < sgl_threshold) | |
519 | return false; | |
520 | return true; | |
521 | } | |
522 | ||
7fe07d14 | 523 | static void nvme_unmap_data(struct nvme_dev *dev, struct request *req) |
b60503ba | 524 | { |
f4800d6d | 525 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
a7a7cbe3 CK |
526 | const int last_prp = dev->ctrl.page_size / sizeof(__le64) - 1; |
527 | dma_addr_t dma_addr = iod->first_dma, next_dma_addr; | |
eca18b23 | 528 | int i; |
eca18b23 | 529 | |
dff824b2 | 530 | if (iod->dma_len) { |
f2fa006f IR |
531 | dma_unmap_page(dev->dev, dma_addr, iod->dma_len, |
532 | rq_dma_dir(req)); | |
dff824b2 | 533 | return; |
7fe07d14 CH |
534 | } |
535 | ||
dff824b2 CH |
536 | WARN_ON_ONCE(!iod->nents); |
537 | ||
7f73eac3 LG |
538 | if (is_pci_p2pdma_page(sg_page(iod->sg))) |
539 | pci_p2pdma_unmap_sg(dev->dev, iod->sg, iod->nents, | |
540 | rq_dma_dir(req)); | |
541 | else | |
dff824b2 CH |
542 | dma_unmap_sg(dev->dev, iod->sg, iod->nents, rq_dma_dir(req)); |
543 | ||
544 | ||
eca18b23 | 545 | if (iod->npages == 0) |
a7a7cbe3 CK |
546 | dma_pool_free(dev->prp_small_pool, nvme_pci_iod_list(req)[0], |
547 | dma_addr); | |
548 | ||
eca18b23 | 549 | for (i = 0; i < iod->npages; i++) { |
a7a7cbe3 CK |
550 | void *addr = nvme_pci_iod_list(req)[i]; |
551 | ||
552 | if (iod->use_sgl) { | |
553 | struct nvme_sgl_desc *sg_list = addr; | |
554 | ||
555 | next_dma_addr = | |
556 | le64_to_cpu((sg_list[SGES_PER_PAGE - 1]).addr); | |
557 | } else { | |
558 | __le64 *prp_list = addr; | |
559 | ||
560 | next_dma_addr = le64_to_cpu(prp_list[last_prp]); | |
561 | } | |
562 | ||
563 | dma_pool_free(dev->prp_page_pool, addr, dma_addr); | |
564 | dma_addr = next_dma_addr; | |
eca18b23 | 565 | } |
ac3dd5bd | 566 | |
d43f1ccf | 567 | mempool_free(iod->sg, dev->iod_mempool); |
b4ff9c8d KB |
568 | } |
569 | ||
d0877473 KB |
570 | static void nvme_print_sgl(struct scatterlist *sgl, int nents) |
571 | { | |
572 | int i; | |
573 | struct scatterlist *sg; | |
574 | ||
575 | for_each_sg(sgl, sg, nents, i) { | |
576 | dma_addr_t phys = sg_phys(sg); | |
577 | pr_warn("sg[%d] phys_addr:%pad offset:%d length:%d " | |
578 | "dma_address:%pad dma_length:%d\n", | |
579 | i, &phys, sg->offset, sg->length, &sg_dma_address(sg), | |
580 | sg_dma_len(sg)); | |
581 | } | |
582 | } | |
583 | ||
a7a7cbe3 CK |
584 | static blk_status_t nvme_pci_setup_prps(struct nvme_dev *dev, |
585 | struct request *req, struct nvme_rw_command *cmnd) | |
ff22b54f | 586 | { |
f4800d6d | 587 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
99802a7a | 588 | struct dma_pool *pool; |
b131c61d | 589 | int length = blk_rq_payload_bytes(req); |
eca18b23 | 590 | struct scatterlist *sg = iod->sg; |
ff22b54f MW |
591 | int dma_len = sg_dma_len(sg); |
592 | u64 dma_addr = sg_dma_address(sg); | |
5fd4ce1b | 593 | u32 page_size = dev->ctrl.page_size; |
f137e0f1 | 594 | int offset = dma_addr & (page_size - 1); |
e025344c | 595 | __le64 *prp_list; |
a7a7cbe3 | 596 | void **list = nvme_pci_iod_list(req); |
e025344c | 597 | dma_addr_t prp_dma; |
eca18b23 | 598 | int nprps, i; |
ff22b54f | 599 | |
1d090624 | 600 | length -= (page_size - offset); |
5228b328 JS |
601 | if (length <= 0) { |
602 | iod->first_dma = 0; | |
a7a7cbe3 | 603 | goto done; |
5228b328 | 604 | } |
ff22b54f | 605 | |
1d090624 | 606 | dma_len -= (page_size - offset); |
ff22b54f | 607 | if (dma_len) { |
1d090624 | 608 | dma_addr += (page_size - offset); |
ff22b54f MW |
609 | } else { |
610 | sg = sg_next(sg); | |
611 | dma_addr = sg_dma_address(sg); | |
612 | dma_len = sg_dma_len(sg); | |
613 | } | |
614 | ||
1d090624 | 615 | if (length <= page_size) { |
edd10d33 | 616 | iod->first_dma = dma_addr; |
a7a7cbe3 | 617 | goto done; |
e025344c SMM |
618 | } |
619 | ||
1d090624 | 620 | nprps = DIV_ROUND_UP(length, page_size); |
99802a7a MW |
621 | if (nprps <= (256 / 8)) { |
622 | pool = dev->prp_small_pool; | |
eca18b23 | 623 | iod->npages = 0; |
99802a7a MW |
624 | } else { |
625 | pool = dev->prp_page_pool; | |
eca18b23 | 626 | iod->npages = 1; |
99802a7a MW |
627 | } |
628 | ||
69d2b571 | 629 | prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); |
b77954cb | 630 | if (!prp_list) { |
edd10d33 | 631 | iod->first_dma = dma_addr; |
eca18b23 | 632 | iod->npages = -1; |
86eea289 | 633 | return BLK_STS_RESOURCE; |
b77954cb | 634 | } |
eca18b23 MW |
635 | list[0] = prp_list; |
636 | iod->first_dma = prp_dma; | |
e025344c SMM |
637 | i = 0; |
638 | for (;;) { | |
1d090624 | 639 | if (i == page_size >> 3) { |
e025344c | 640 | __le64 *old_prp_list = prp_list; |
69d2b571 | 641 | prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); |
eca18b23 | 642 | if (!prp_list) |
86eea289 | 643 | return BLK_STS_RESOURCE; |
eca18b23 | 644 | list[iod->npages++] = prp_list; |
7523d834 MW |
645 | prp_list[0] = old_prp_list[i - 1]; |
646 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
647 | i = 1; | |
e025344c SMM |
648 | } |
649 | prp_list[i++] = cpu_to_le64(dma_addr); | |
1d090624 KB |
650 | dma_len -= page_size; |
651 | dma_addr += page_size; | |
652 | length -= page_size; | |
e025344c SMM |
653 | if (length <= 0) |
654 | break; | |
655 | if (dma_len > 0) | |
656 | continue; | |
86eea289 KB |
657 | if (unlikely(dma_len < 0)) |
658 | goto bad_sgl; | |
e025344c SMM |
659 | sg = sg_next(sg); |
660 | dma_addr = sg_dma_address(sg); | |
661 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
662 | } |
663 | ||
a7a7cbe3 CK |
664 | done: |
665 | cmnd->dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); | |
666 | cmnd->dptr.prp2 = cpu_to_le64(iod->first_dma); | |
667 | ||
86eea289 KB |
668 | return BLK_STS_OK; |
669 | ||
670 | bad_sgl: | |
d0877473 KB |
671 | WARN(DO_ONCE(nvme_print_sgl, iod->sg, iod->nents), |
672 | "Invalid SGL for payload:%d nents:%d\n", | |
673 | blk_rq_payload_bytes(req), iod->nents); | |
86eea289 | 674 | return BLK_STS_IOERR; |
ff22b54f MW |
675 | } |
676 | ||
a7a7cbe3 CK |
677 | static void nvme_pci_sgl_set_data(struct nvme_sgl_desc *sge, |
678 | struct scatterlist *sg) | |
679 | { | |
680 | sge->addr = cpu_to_le64(sg_dma_address(sg)); | |
681 | sge->length = cpu_to_le32(sg_dma_len(sg)); | |
682 | sge->type = NVME_SGL_FMT_DATA_DESC << 4; | |
683 | } | |
684 | ||
685 | static void nvme_pci_sgl_set_seg(struct nvme_sgl_desc *sge, | |
686 | dma_addr_t dma_addr, int entries) | |
687 | { | |
688 | sge->addr = cpu_to_le64(dma_addr); | |
689 | if (entries < SGES_PER_PAGE) { | |
690 | sge->length = cpu_to_le32(entries * sizeof(*sge)); | |
691 | sge->type = NVME_SGL_FMT_LAST_SEG_DESC << 4; | |
692 | } else { | |
693 | sge->length = cpu_to_le32(PAGE_SIZE); | |
694 | sge->type = NVME_SGL_FMT_SEG_DESC << 4; | |
695 | } | |
696 | } | |
697 | ||
698 | static blk_status_t nvme_pci_setup_sgls(struct nvme_dev *dev, | |
b0f2853b | 699 | struct request *req, struct nvme_rw_command *cmd, int entries) |
a7a7cbe3 CK |
700 | { |
701 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
a7a7cbe3 CK |
702 | struct dma_pool *pool; |
703 | struct nvme_sgl_desc *sg_list; | |
704 | struct scatterlist *sg = iod->sg; | |
a7a7cbe3 | 705 | dma_addr_t sgl_dma; |
b0f2853b | 706 | int i = 0; |
a7a7cbe3 | 707 | |
a7a7cbe3 CK |
708 | /* setting the transfer type as SGL */ |
709 | cmd->flags = NVME_CMD_SGL_METABUF; | |
710 | ||
b0f2853b | 711 | if (entries == 1) { |
a7a7cbe3 CK |
712 | nvme_pci_sgl_set_data(&cmd->dptr.sgl, sg); |
713 | return BLK_STS_OK; | |
714 | } | |
715 | ||
716 | if (entries <= (256 / sizeof(struct nvme_sgl_desc))) { | |
717 | pool = dev->prp_small_pool; | |
718 | iod->npages = 0; | |
719 | } else { | |
720 | pool = dev->prp_page_pool; | |
721 | iod->npages = 1; | |
722 | } | |
723 | ||
724 | sg_list = dma_pool_alloc(pool, GFP_ATOMIC, &sgl_dma); | |
725 | if (!sg_list) { | |
726 | iod->npages = -1; | |
727 | return BLK_STS_RESOURCE; | |
728 | } | |
729 | ||
730 | nvme_pci_iod_list(req)[0] = sg_list; | |
731 | iod->first_dma = sgl_dma; | |
732 | ||
733 | nvme_pci_sgl_set_seg(&cmd->dptr.sgl, sgl_dma, entries); | |
734 | ||
735 | do { | |
736 | if (i == SGES_PER_PAGE) { | |
737 | struct nvme_sgl_desc *old_sg_desc = sg_list; | |
738 | struct nvme_sgl_desc *link = &old_sg_desc[i - 1]; | |
739 | ||
740 | sg_list = dma_pool_alloc(pool, GFP_ATOMIC, &sgl_dma); | |
741 | if (!sg_list) | |
742 | return BLK_STS_RESOURCE; | |
743 | ||
744 | i = 0; | |
745 | nvme_pci_iod_list(req)[iod->npages++] = sg_list; | |
746 | sg_list[i++] = *link; | |
747 | nvme_pci_sgl_set_seg(link, sgl_dma, entries); | |
748 | } | |
749 | ||
750 | nvme_pci_sgl_set_data(&sg_list[i++], sg); | |
a7a7cbe3 | 751 | sg = sg_next(sg); |
b0f2853b | 752 | } while (--entries > 0); |
a7a7cbe3 | 753 | |
a7a7cbe3 CK |
754 | return BLK_STS_OK; |
755 | } | |
756 | ||
dff824b2 CH |
757 | static blk_status_t nvme_setup_prp_simple(struct nvme_dev *dev, |
758 | struct request *req, struct nvme_rw_command *cmnd, | |
759 | struct bio_vec *bv) | |
760 | { | |
761 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
a4f40484 KH |
762 | unsigned int offset = bv->bv_offset & (dev->ctrl.page_size - 1); |
763 | unsigned int first_prp_len = dev->ctrl.page_size - offset; | |
dff824b2 CH |
764 | |
765 | iod->first_dma = dma_map_bvec(dev->dev, bv, rq_dma_dir(req), 0); | |
766 | if (dma_mapping_error(dev->dev, iod->first_dma)) | |
767 | return BLK_STS_RESOURCE; | |
768 | iod->dma_len = bv->bv_len; | |
769 | ||
770 | cmnd->dptr.prp1 = cpu_to_le64(iod->first_dma); | |
771 | if (bv->bv_len > first_prp_len) | |
772 | cmnd->dptr.prp2 = cpu_to_le64(iod->first_dma + first_prp_len); | |
773 | return 0; | |
774 | } | |
775 | ||
29791057 CH |
776 | static blk_status_t nvme_setup_sgl_simple(struct nvme_dev *dev, |
777 | struct request *req, struct nvme_rw_command *cmnd, | |
778 | struct bio_vec *bv) | |
779 | { | |
780 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
781 | ||
782 | iod->first_dma = dma_map_bvec(dev->dev, bv, rq_dma_dir(req), 0); | |
783 | if (dma_mapping_error(dev->dev, iod->first_dma)) | |
784 | return BLK_STS_RESOURCE; | |
785 | iod->dma_len = bv->bv_len; | |
786 | ||
049bf372 | 787 | cmnd->flags = NVME_CMD_SGL_METABUF; |
29791057 CH |
788 | cmnd->dptr.sgl.addr = cpu_to_le64(iod->first_dma); |
789 | cmnd->dptr.sgl.length = cpu_to_le32(iod->dma_len); | |
790 | cmnd->dptr.sgl.type = NVME_SGL_FMT_DATA_DESC << 4; | |
791 | return 0; | |
792 | } | |
793 | ||
fc17b653 | 794 | static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req, |
b131c61d | 795 | struct nvme_command *cmnd) |
d29ec824 | 796 | { |
f4800d6d | 797 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
70479b71 | 798 | blk_status_t ret = BLK_STS_RESOURCE; |
b0f2853b | 799 | int nr_mapped; |
d29ec824 | 800 | |
dff824b2 CH |
801 | if (blk_rq_nr_phys_segments(req) == 1) { |
802 | struct bio_vec bv = req_bvec(req); | |
803 | ||
804 | if (!is_pci_p2pdma_page(bv.bv_page)) { | |
805 | if (bv.bv_offset + bv.bv_len <= dev->ctrl.page_size * 2) | |
806 | return nvme_setup_prp_simple(dev, req, | |
807 | &cmnd->rw, &bv); | |
29791057 CH |
808 | |
809 | if (iod->nvmeq->qid && | |
810 | dev->ctrl.sgls & ((1 << 0) | (1 << 1))) | |
811 | return nvme_setup_sgl_simple(dev, req, | |
812 | &cmnd->rw, &bv); | |
dff824b2 CH |
813 | } |
814 | } | |
815 | ||
816 | iod->dma_len = 0; | |
d43f1ccf CH |
817 | iod->sg = mempool_alloc(dev->iod_mempool, GFP_ATOMIC); |
818 | if (!iod->sg) | |
819 | return BLK_STS_RESOURCE; | |
f9d03f96 | 820 | sg_init_table(iod->sg, blk_rq_nr_phys_segments(req)); |
70479b71 | 821 | iod->nents = blk_rq_map_sg(req->q, req, iod->sg); |
ba1ca37e CH |
822 | if (!iod->nents) |
823 | goto out; | |
d29ec824 | 824 | |
e0596ab2 | 825 | if (is_pci_p2pdma_page(sg_page(iod->sg))) |
2b9f4bb2 LG |
826 | nr_mapped = pci_p2pdma_map_sg_attrs(dev->dev, iod->sg, |
827 | iod->nents, rq_dma_dir(req), DMA_ATTR_NO_WARN); | |
e0596ab2 LG |
828 | else |
829 | nr_mapped = dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, | |
70479b71 | 830 | rq_dma_dir(req), DMA_ATTR_NO_WARN); |
b0f2853b | 831 | if (!nr_mapped) |
ba1ca37e | 832 | goto out; |
d29ec824 | 833 | |
70479b71 | 834 | iod->use_sgl = nvme_pci_use_sgls(dev, req); |
955b1b5a | 835 | if (iod->use_sgl) |
b0f2853b | 836 | ret = nvme_pci_setup_sgls(dev, req, &cmnd->rw, nr_mapped); |
a7a7cbe3 CK |
837 | else |
838 | ret = nvme_pci_setup_prps(dev, req, &cmnd->rw); | |
4aedb705 | 839 | out: |
86eea289 | 840 | if (ret != BLK_STS_OK) |
4aedb705 CH |
841 | nvme_unmap_data(dev, req); |
842 | return ret; | |
843 | } | |
3045c0d0 | 844 | |
4aedb705 CH |
845 | static blk_status_t nvme_map_metadata(struct nvme_dev *dev, struct request *req, |
846 | struct nvme_command *cmnd) | |
847 | { | |
848 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); | |
00df5cb4 | 849 | |
4aedb705 CH |
850 | iod->meta_dma = dma_map_bvec(dev->dev, rq_integrity_vec(req), |
851 | rq_dma_dir(req), 0); | |
852 | if (dma_mapping_error(dev->dev, iod->meta_dma)) | |
853 | return BLK_STS_IOERR; | |
854 | cmnd->rw.metadata = cpu_to_le64(iod->meta_dma); | |
855 | return 0; | |
00df5cb4 MW |
856 | } |
857 | ||
d29ec824 CH |
858 | /* |
859 | * NOTE: ns is NULL when called on the admin queue. | |
860 | */ | |
fc17b653 | 861 | static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx, |
a4aea562 | 862 | const struct blk_mq_queue_data *bd) |
edd10d33 | 863 | { |
a4aea562 MB |
864 | struct nvme_ns *ns = hctx->queue->queuedata; |
865 | struct nvme_queue *nvmeq = hctx->driver_data; | |
d29ec824 | 866 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 | 867 | struct request *req = bd->rq; |
9b048119 | 868 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
ba1ca37e | 869 | struct nvme_command cmnd; |
ebe6d874 | 870 | blk_status_t ret; |
e1e5e564 | 871 | |
9b048119 CH |
872 | iod->aborted = 0; |
873 | iod->npages = -1; | |
874 | iod->nents = 0; | |
875 | ||
d1f06f4a JA |
876 | /* |
877 | * We should not need to do this, but we're still using this to | |
878 | * ensure we can drain requests on a dying queue. | |
879 | */ | |
4e224106 | 880 | if (unlikely(!test_bit(NVMEQ_ENABLED, &nvmeq->flags))) |
d1f06f4a JA |
881 | return BLK_STS_IOERR; |
882 | ||
f9d03f96 | 883 | ret = nvme_setup_cmd(ns, req, &cmnd); |
fc17b653 | 884 | if (ret) |
f4800d6d | 885 | return ret; |
a4aea562 | 886 | |
fc17b653 | 887 | if (blk_rq_nr_phys_segments(req)) { |
b131c61d | 888 | ret = nvme_map_data(dev, req, &cmnd); |
fc17b653 | 889 | if (ret) |
9b048119 | 890 | goto out_free_cmd; |
fc17b653 | 891 | } |
a4aea562 | 892 | |
4aedb705 CH |
893 | if (blk_integrity_rq(req)) { |
894 | ret = nvme_map_metadata(dev, req, &cmnd); | |
895 | if (ret) | |
896 | goto out_unmap_data; | |
897 | } | |
898 | ||
aae239e1 | 899 | blk_mq_start_request(req); |
04f3eafd | 900 | nvme_submit_cmd(nvmeq, &cmnd, bd->last); |
fc17b653 | 901 | return BLK_STS_OK; |
4aedb705 CH |
902 | out_unmap_data: |
903 | nvme_unmap_data(dev, req); | |
f9d03f96 CH |
904 | out_free_cmd: |
905 | nvme_cleanup_cmd(req); | |
ba1ca37e | 906 | return ret; |
b60503ba | 907 | } |
e1e5e564 | 908 | |
77f02a7a | 909 | static void nvme_pci_complete_rq(struct request *req) |
eee417b0 | 910 | { |
f4800d6d | 911 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
4aedb705 | 912 | struct nvme_dev *dev = iod->nvmeq->dev; |
a4aea562 | 913 | |
4aedb705 CH |
914 | if (blk_integrity_rq(req)) |
915 | dma_unmap_page(dev->dev, iod->meta_dma, | |
916 | rq_integrity_vec(req)->bv_len, rq_data_dir(req)); | |
b15c592d | 917 | if (blk_rq_nr_phys_segments(req)) |
4aedb705 | 918 | nvme_unmap_data(dev, req); |
77f02a7a | 919 | nvme_complete_rq(req); |
b60503ba MW |
920 | } |
921 | ||
d783e0bd | 922 | /* We read the CQE phase first to check if the rest of the entry is valid */ |
750dde44 | 923 | static inline bool nvme_cqe_pending(struct nvme_queue *nvmeq) |
d783e0bd | 924 | { |
750dde44 CH |
925 | return (le16_to_cpu(nvmeq->cqes[nvmeq->cq_head].status) & 1) == |
926 | nvmeq->cq_phase; | |
d783e0bd MR |
927 | } |
928 | ||
eb281c82 | 929 | static inline void nvme_ring_cq_doorbell(struct nvme_queue *nvmeq) |
b60503ba | 930 | { |
eb281c82 | 931 | u16 head = nvmeq->cq_head; |
adf68f21 | 932 | |
397c699f KB |
933 | if (nvme_dbbuf_update_and_check_event(head, nvmeq->dbbuf_cq_db, |
934 | nvmeq->dbbuf_cq_ei)) | |
935 | writel(head, nvmeq->q_db + nvmeq->dev->db_stride); | |
eb281c82 | 936 | } |
aae239e1 | 937 | |
cfa27356 CH |
938 | static inline struct blk_mq_tags *nvme_queue_tagset(struct nvme_queue *nvmeq) |
939 | { | |
940 | if (!nvmeq->qid) | |
941 | return nvmeq->dev->admin_tagset.tags[0]; | |
942 | return nvmeq->dev->tagset.tags[nvmeq->qid - 1]; | |
943 | } | |
944 | ||
5cb525c8 | 945 | static inline void nvme_handle_cqe(struct nvme_queue *nvmeq, u16 idx) |
83a12fb7 | 946 | { |
5cb525c8 | 947 | volatile struct nvme_completion *cqe = &nvmeq->cqes[idx]; |
83a12fb7 | 948 | struct request *req; |
adf68f21 | 949 | |
83a12fb7 SG |
950 | if (unlikely(cqe->command_id >= nvmeq->q_depth)) { |
951 | dev_warn(nvmeq->dev->ctrl.device, | |
952 | "invalid id %d completed on queue %d\n", | |
953 | cqe->command_id, le16_to_cpu(cqe->sq_id)); | |
954 | return; | |
b60503ba MW |
955 | } |
956 | ||
83a12fb7 SG |
957 | /* |
958 | * AEN requests are special as they don't time out and can | |
959 | * survive any kind of queue freeze and often don't respond to | |
960 | * aborts. We don't even bother to allocate a struct request | |
961 | * for them but rather special case them here. | |
962 | */ | |
58a8df67 | 963 | if (unlikely(nvme_is_aen_req(nvmeq->qid, cqe->command_id))) { |
83a12fb7 SG |
964 | nvme_complete_async_event(&nvmeq->dev->ctrl, |
965 | cqe->status, &cqe->result); | |
a0fa9647 | 966 | return; |
83a12fb7 | 967 | } |
b60503ba | 968 | |
cfa27356 | 969 | req = blk_mq_tag_to_rq(nvme_queue_tagset(nvmeq), cqe->command_id); |
604c01d5 | 970 | trace_nvme_sq(req, cqe->sq_head, nvmeq->sq_tail); |
83a12fb7 SG |
971 | nvme_end_request(req, cqe->status, cqe->result); |
972 | } | |
b60503ba | 973 | |
5cb525c8 JA |
974 | static inline void nvme_update_cq_head(struct nvme_queue *nvmeq) |
975 | { | |
a8de6639 AD |
976 | u16 tmp = nvmeq->cq_head + 1; |
977 | ||
978 | if (tmp == nvmeq->q_depth) { | |
5cb525c8 | 979 | nvmeq->cq_head = 0; |
e2a366a4 | 980 | nvmeq->cq_phase ^= 1; |
a8de6639 AD |
981 | } else { |
982 | nvmeq->cq_head = tmp; | |
b60503ba | 983 | } |
a0fa9647 JA |
984 | } |
985 | ||
324b494c | 986 | static inline int nvme_process_cq(struct nvme_queue *nvmeq) |
a0fa9647 | 987 | { |
1052b8ac | 988 | int found = 0; |
b60503ba | 989 | |
1052b8ac | 990 | while (nvme_cqe_pending(nvmeq)) { |
bf392a5d | 991 | found++; |
b69e2ef2 KB |
992 | /* |
993 | * load-load control dependency between phase and the rest of | |
994 | * the cqe requires a full read memory barrier | |
995 | */ | |
996 | dma_rmb(); | |
324b494c | 997 | nvme_handle_cqe(nvmeq, nvmeq->cq_head); |
5cb525c8 | 998 | nvme_update_cq_head(nvmeq); |
920d13a8 | 999 | } |
eb281c82 | 1000 | |
324b494c | 1001 | if (found) |
920d13a8 | 1002 | nvme_ring_cq_doorbell(nvmeq); |
5cb525c8 | 1003 | return found; |
b60503ba MW |
1004 | } |
1005 | ||
1006 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 | 1007 | { |
58ffacb5 | 1008 | struct nvme_queue *nvmeq = data; |
68fa9dbe | 1009 | irqreturn_t ret = IRQ_NONE; |
5cb525c8 | 1010 | |
3a7afd8e CH |
1011 | /* |
1012 | * The rmb/wmb pair ensures we see all updates from a previous run of | |
1013 | * the irq handler, even if that was on another CPU. | |
1014 | */ | |
1015 | rmb(); | |
324b494c KB |
1016 | if (nvme_process_cq(nvmeq)) |
1017 | ret = IRQ_HANDLED; | |
3a7afd8e | 1018 | wmb(); |
5cb525c8 | 1019 | |
68fa9dbe | 1020 | return ret; |
58ffacb5 MW |
1021 | } |
1022 | ||
1023 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
1024 | { | |
1025 | struct nvme_queue *nvmeq = data; | |
750dde44 | 1026 | if (nvme_cqe_pending(nvmeq)) |
d783e0bd MR |
1027 | return IRQ_WAKE_THREAD; |
1028 | return IRQ_NONE; | |
58ffacb5 MW |
1029 | } |
1030 | ||
0b2a8a9f | 1031 | /* |
fa059b85 | 1032 | * Poll for completions for any interrupt driven queue |
0b2a8a9f CH |
1033 | * Can be called from any context. |
1034 | */ | |
fa059b85 | 1035 | static void nvme_poll_irqdisable(struct nvme_queue *nvmeq) |
a0fa9647 | 1036 | { |
3a7afd8e | 1037 | struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev); |
a0fa9647 | 1038 | |
fa059b85 | 1039 | WARN_ON_ONCE(test_bit(NVMEQ_POLLED, &nvmeq->flags)); |
442e19b7 | 1040 | |
fa059b85 KB |
1041 | disable_irq(pci_irq_vector(pdev, nvmeq->cq_vector)); |
1042 | nvme_process_cq(nvmeq); | |
1043 | enable_irq(pci_irq_vector(pdev, nvmeq->cq_vector)); | |
a0fa9647 JA |
1044 | } |
1045 | ||
9743139c | 1046 | static int nvme_poll(struct blk_mq_hw_ctx *hctx) |
dabcefab JA |
1047 | { |
1048 | struct nvme_queue *nvmeq = hctx->driver_data; | |
dabcefab JA |
1049 | bool found; |
1050 | ||
1051 | if (!nvme_cqe_pending(nvmeq)) | |
1052 | return 0; | |
1053 | ||
3a7afd8e | 1054 | spin_lock(&nvmeq->cq_poll_lock); |
324b494c | 1055 | found = nvme_process_cq(nvmeq); |
3a7afd8e | 1056 | spin_unlock(&nvmeq->cq_poll_lock); |
dabcefab | 1057 | |
dabcefab JA |
1058 | return found; |
1059 | } | |
1060 | ||
ad22c355 | 1061 | static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl) |
b60503ba | 1062 | { |
f866fc42 | 1063 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
147b27e4 | 1064 | struct nvme_queue *nvmeq = &dev->queues[0]; |
a4aea562 | 1065 | struct nvme_command c; |
b60503ba | 1066 | |
a4aea562 MB |
1067 | memset(&c, 0, sizeof(c)); |
1068 | c.common.opcode = nvme_admin_async_event; | |
ad22c355 | 1069 | c.common.command_id = NVME_AQ_BLK_MQ_DEPTH; |
04f3eafd | 1070 | nvme_submit_cmd(nvmeq, &c, true); |
f705f837 CH |
1071 | } |
1072 | ||
b60503ba | 1073 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) |
f705f837 | 1074 | { |
b60503ba MW |
1075 | struct nvme_command c; |
1076 | ||
1077 | memset(&c, 0, sizeof(c)); | |
1078 | c.delete_queue.opcode = opcode; | |
1079 | c.delete_queue.qid = cpu_to_le16(id); | |
1080 | ||
1c63dc66 | 1081 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
1082 | } |
1083 | ||
b60503ba | 1084 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, |
a8e3e0bb | 1085 | struct nvme_queue *nvmeq, s16 vector) |
b60503ba | 1086 | { |
b60503ba | 1087 | struct nvme_command c; |
4b04cc6a JA |
1088 | int flags = NVME_QUEUE_PHYS_CONTIG; |
1089 | ||
7c349dde | 1090 | if (!test_bit(NVMEQ_POLLED, &nvmeq->flags)) |
4b04cc6a | 1091 | flags |= NVME_CQ_IRQ_ENABLED; |
b60503ba | 1092 | |
d29ec824 | 1093 | /* |
16772ae6 | 1094 | * Note: we (ab)use the fact that the prp fields survive if no data |
d29ec824 CH |
1095 | * is attached to the request. |
1096 | */ | |
b60503ba MW |
1097 | memset(&c, 0, sizeof(c)); |
1098 | c.create_cq.opcode = nvme_admin_create_cq; | |
1099 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
1100 | c.create_cq.cqid = cpu_to_le16(qid); | |
1101 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1102 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
7c349dde | 1103 | c.create_cq.irq_vector = cpu_to_le16(vector); |
b60503ba | 1104 | |
1c63dc66 | 1105 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
1106 | } |
1107 | ||
1108 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
1109 | struct nvme_queue *nvmeq) | |
1110 | { | |
9abd68ef | 1111 | struct nvme_ctrl *ctrl = &dev->ctrl; |
b60503ba | 1112 | struct nvme_command c; |
81c1cd98 | 1113 | int flags = NVME_QUEUE_PHYS_CONTIG; |
b60503ba | 1114 | |
9abd68ef JA |
1115 | /* |
1116 | * Some drives have a bug that auto-enables WRRU if MEDIUM isn't | |
1117 | * set. Since URGENT priority is zeroes, it makes all queues | |
1118 | * URGENT. | |
1119 | */ | |
1120 | if (ctrl->quirks & NVME_QUIRK_MEDIUM_PRIO_SQ) | |
1121 | flags |= NVME_SQ_PRIO_MEDIUM; | |
1122 | ||
d29ec824 | 1123 | /* |
16772ae6 | 1124 | * Note: we (ab)use the fact that the prp fields survive if no data |
d29ec824 CH |
1125 | * is attached to the request. |
1126 | */ | |
b60503ba MW |
1127 | memset(&c, 0, sizeof(c)); |
1128 | c.create_sq.opcode = nvme_admin_create_sq; | |
1129 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
1130 | c.create_sq.sqid = cpu_to_le16(qid); | |
1131 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
1132 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
1133 | c.create_sq.cqid = cpu_to_le16(qid); | |
1134 | ||
1c63dc66 | 1135 | return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); |
b60503ba MW |
1136 | } |
1137 | ||
1138 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
1139 | { | |
1140 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
1141 | } | |
1142 | ||
1143 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
1144 | { | |
1145 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
1146 | } | |
1147 | ||
2a842aca | 1148 | static void abort_endio(struct request *req, blk_status_t error) |
bc5fc7e4 | 1149 | { |
f4800d6d CH |
1150 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
1151 | struct nvme_queue *nvmeq = iod->nvmeq; | |
e44ac588 | 1152 | |
27fa9bc5 CH |
1153 | dev_warn(nvmeq->dev->ctrl.device, |
1154 | "Abort status: 0x%x", nvme_req(req)->status); | |
e7a2a87d | 1155 | atomic_inc(&nvmeq->dev->ctrl.abort_limit); |
e7a2a87d | 1156 | blk_mq_free_request(req); |
bc5fc7e4 MW |
1157 | } |
1158 | ||
b2a0eb1a KB |
1159 | static bool nvme_should_reset(struct nvme_dev *dev, u32 csts) |
1160 | { | |
1161 | ||
1162 | /* If true, indicates loss of adapter communication, possibly by a | |
1163 | * NVMe Subsystem reset. | |
1164 | */ | |
1165 | bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO); | |
1166 | ||
ad70062c JW |
1167 | /* If there is a reset/reinit ongoing, we shouldn't reset again. */ |
1168 | switch (dev->ctrl.state) { | |
1169 | case NVME_CTRL_RESETTING: | |
ad6a0a52 | 1170 | case NVME_CTRL_CONNECTING: |
b2a0eb1a | 1171 | return false; |
ad70062c JW |
1172 | default: |
1173 | break; | |
1174 | } | |
b2a0eb1a KB |
1175 | |
1176 | /* We shouldn't reset unless the controller is on fatal error state | |
1177 | * _or_ if we lost the communication with it. | |
1178 | */ | |
1179 | if (!(csts & NVME_CSTS_CFS) && !nssro) | |
1180 | return false; | |
1181 | ||
b2a0eb1a KB |
1182 | return true; |
1183 | } | |
1184 | ||
1185 | static void nvme_warn_reset(struct nvme_dev *dev, u32 csts) | |
1186 | { | |
1187 | /* Read a config register to help see what died. */ | |
1188 | u16 pci_status; | |
1189 | int result; | |
1190 | ||
1191 | result = pci_read_config_word(to_pci_dev(dev->dev), PCI_STATUS, | |
1192 | &pci_status); | |
1193 | if (result == PCIBIOS_SUCCESSFUL) | |
1194 | dev_warn(dev->ctrl.device, | |
1195 | "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n", | |
1196 | csts, pci_status); | |
1197 | else | |
1198 | dev_warn(dev->ctrl.device, | |
1199 | "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n", | |
1200 | csts, result); | |
1201 | } | |
1202 | ||
31c7c7d2 | 1203 | static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) |
c30341dc | 1204 | { |
f4800d6d CH |
1205 | struct nvme_iod *iod = blk_mq_rq_to_pdu(req); |
1206 | struct nvme_queue *nvmeq = iod->nvmeq; | |
c30341dc | 1207 | struct nvme_dev *dev = nvmeq->dev; |
a4aea562 | 1208 | struct request *abort_req; |
a4aea562 | 1209 | struct nvme_command cmd; |
b2a0eb1a KB |
1210 | u32 csts = readl(dev->bar + NVME_REG_CSTS); |
1211 | ||
651438bb WX |
1212 | /* If PCI error recovery process is happening, we cannot reset or |
1213 | * the recovery mechanism will surely fail. | |
1214 | */ | |
1215 | mb(); | |
1216 | if (pci_channel_offline(to_pci_dev(dev->dev))) | |
1217 | return BLK_EH_RESET_TIMER; | |
1218 | ||
b2a0eb1a KB |
1219 | /* |
1220 | * Reset immediately if the controller is failed | |
1221 | */ | |
1222 | if (nvme_should_reset(dev, csts)) { | |
1223 | nvme_warn_reset(dev, csts); | |
1224 | nvme_dev_disable(dev, false); | |
d86c4d8e | 1225 | nvme_reset_ctrl(&dev->ctrl); |
db8c48e4 | 1226 | return BLK_EH_DONE; |
b2a0eb1a | 1227 | } |
c30341dc | 1228 | |
7776db1c KB |
1229 | /* |
1230 | * Did we miss an interrupt? | |
1231 | */ | |
fa059b85 KB |
1232 | if (test_bit(NVMEQ_POLLED, &nvmeq->flags)) |
1233 | nvme_poll(req->mq_hctx); | |
1234 | else | |
1235 | nvme_poll_irqdisable(nvmeq); | |
1236 | ||
bf392a5d | 1237 | if (blk_mq_request_completed(req)) { |
7776db1c KB |
1238 | dev_warn(dev->ctrl.device, |
1239 | "I/O %d QID %d timeout, completion polled\n", | |
1240 | req->tag, nvmeq->qid); | |
db8c48e4 | 1241 | return BLK_EH_DONE; |
7776db1c KB |
1242 | } |
1243 | ||
31c7c7d2 | 1244 | /* |
fd634f41 CH |
1245 | * Shutdown immediately if controller times out while starting. The |
1246 | * reset work will see the pci device disabled when it gets the forced | |
1247 | * cancellation error. All outstanding requests are completed on | |
db8c48e4 | 1248 | * shutdown, so we return BLK_EH_DONE. |
fd634f41 | 1249 | */ |
4244140d KB |
1250 | switch (dev->ctrl.state) { |
1251 | case NVME_CTRL_CONNECTING: | |
2036f726 KB |
1252 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); |
1253 | /* fall through */ | |
1254 | case NVME_CTRL_DELETING: | |
b9cac43c | 1255 | dev_warn_ratelimited(dev->ctrl.device, |
fd634f41 CH |
1256 | "I/O %d QID %d timeout, disable controller\n", |
1257 | req->tag, nvmeq->qid); | |
2036f726 | 1258 | nvme_dev_disable(dev, true); |
27fa9bc5 | 1259 | nvme_req(req)->flags |= NVME_REQ_CANCELLED; |
db8c48e4 | 1260 | return BLK_EH_DONE; |
39a9dd81 KB |
1261 | case NVME_CTRL_RESETTING: |
1262 | return BLK_EH_RESET_TIMER; | |
4244140d KB |
1263 | default: |
1264 | break; | |
c30341dc KB |
1265 | } |
1266 | ||
fd634f41 CH |
1267 | /* |
1268 | * Shutdown the controller immediately and schedule a reset if the | |
1269 | * command was already aborted once before and still hasn't been | |
1270 | * returned to the driver, or if this is the admin queue. | |
31c7c7d2 | 1271 | */ |
f4800d6d | 1272 | if (!nvmeq->qid || iod->aborted) { |
1b3c47c1 | 1273 | dev_warn(dev->ctrl.device, |
e1569a16 KB |
1274 | "I/O %d QID %d timeout, reset controller\n", |
1275 | req->tag, nvmeq->qid); | |
a5cdb68c | 1276 | nvme_dev_disable(dev, false); |
d86c4d8e | 1277 | nvme_reset_ctrl(&dev->ctrl); |
c30341dc | 1278 | |
27fa9bc5 | 1279 | nvme_req(req)->flags |= NVME_REQ_CANCELLED; |
db8c48e4 | 1280 | return BLK_EH_DONE; |
c30341dc | 1281 | } |
c30341dc | 1282 | |
e7a2a87d | 1283 | if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) { |
6bf25d16 | 1284 | atomic_inc(&dev->ctrl.abort_limit); |
31c7c7d2 | 1285 | return BLK_EH_RESET_TIMER; |
6bf25d16 | 1286 | } |
7bf7d778 | 1287 | iod->aborted = 1; |
a4aea562 | 1288 | |
c30341dc KB |
1289 | memset(&cmd, 0, sizeof(cmd)); |
1290 | cmd.abort.opcode = nvme_admin_abort_cmd; | |
a4aea562 | 1291 | cmd.abort.cid = req->tag; |
c30341dc | 1292 | cmd.abort.sqid = cpu_to_le16(nvmeq->qid); |
c30341dc | 1293 | |
1b3c47c1 SG |
1294 | dev_warn(nvmeq->dev->ctrl.device, |
1295 | "I/O %d QID %d timeout, aborting\n", | |
1296 | req->tag, nvmeq->qid); | |
e7a2a87d CH |
1297 | |
1298 | abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd, | |
eb71f435 | 1299 | BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); |
e7a2a87d CH |
1300 | if (IS_ERR(abort_req)) { |
1301 | atomic_inc(&dev->ctrl.abort_limit); | |
1302 | return BLK_EH_RESET_TIMER; | |
1303 | } | |
1304 | ||
1305 | abort_req->timeout = ADMIN_TIMEOUT; | |
1306 | abort_req->end_io_data = NULL; | |
1307 | blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio); | |
c30341dc | 1308 | |
31c7c7d2 CH |
1309 | /* |
1310 | * The aborted req will be completed on receiving the abort req. | |
1311 | * We enable the timer again. If hit twice, it'll cause a device reset, | |
1312 | * as the device then is in a faulty state. | |
1313 | */ | |
1314 | return BLK_EH_RESET_TIMER; | |
c30341dc KB |
1315 | } |
1316 | ||
a4aea562 MB |
1317 | static void nvme_free_queue(struct nvme_queue *nvmeq) |
1318 | { | |
8a1d09a6 | 1319 | dma_free_coherent(nvmeq->dev->dev, CQ_SIZE(nvmeq), |
9e866774 | 1320 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); |
63223078 CH |
1321 | if (!nvmeq->sq_cmds) |
1322 | return; | |
0f238ff5 | 1323 | |
63223078 | 1324 | if (test_and_clear_bit(NVMEQ_SQ_CMB, &nvmeq->flags)) { |
88a041f4 | 1325 | pci_free_p2pmem(to_pci_dev(nvmeq->dev->dev), |
8a1d09a6 | 1326 | nvmeq->sq_cmds, SQ_SIZE(nvmeq)); |
63223078 | 1327 | } else { |
8a1d09a6 | 1328 | dma_free_coherent(nvmeq->dev->dev, SQ_SIZE(nvmeq), |
63223078 | 1329 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); |
0f238ff5 | 1330 | } |
9e866774 MW |
1331 | } |
1332 | ||
a1a5ef99 | 1333 | static void nvme_free_queues(struct nvme_dev *dev, int lowest) |
22404274 KB |
1334 | { |
1335 | int i; | |
1336 | ||
d858e5f0 | 1337 | for (i = dev->ctrl.queue_count - 1; i >= lowest; i--) { |
d858e5f0 | 1338 | dev->ctrl.queue_count--; |
147b27e4 | 1339 | nvme_free_queue(&dev->queues[i]); |
121c7ad4 | 1340 | } |
22404274 KB |
1341 | } |
1342 | ||
4d115420 KB |
1343 | /** |
1344 | * nvme_suspend_queue - put queue into suspended state | |
40581d1a | 1345 | * @nvmeq: queue to suspend |
4d115420 KB |
1346 | */ |
1347 | static int nvme_suspend_queue(struct nvme_queue *nvmeq) | |
b60503ba | 1348 | { |
4e224106 | 1349 | if (!test_and_clear_bit(NVMEQ_ENABLED, &nvmeq->flags)) |
2b25d981 | 1350 | return 1; |
a09115b2 | 1351 | |
4e224106 | 1352 | /* ensure that nvme_queue_rq() sees NVMEQ_ENABLED cleared */ |
d1f06f4a | 1353 | mb(); |
a09115b2 | 1354 | |
4e224106 | 1355 | nvmeq->dev->online_queues--; |
1c63dc66 | 1356 | if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q) |
c81545f9 | 1357 | blk_mq_quiesce_queue(nvmeq->dev->ctrl.admin_q); |
7c349dde KB |
1358 | if (!test_and_clear_bit(NVMEQ_POLLED, &nvmeq->flags)) |
1359 | pci_free_irq(to_pci_dev(nvmeq->dev->dev), nvmeq->cq_vector, nvmeq); | |
4d115420 KB |
1360 | return 0; |
1361 | } | |
b60503ba | 1362 | |
8fae268b KB |
1363 | static void nvme_suspend_io_queues(struct nvme_dev *dev) |
1364 | { | |
1365 | int i; | |
1366 | ||
1367 | for (i = dev->ctrl.queue_count - 1; i > 0; i--) | |
1368 | nvme_suspend_queue(&dev->queues[i]); | |
1369 | } | |
1370 | ||
a5cdb68c | 1371 | static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown) |
4d115420 | 1372 | { |
147b27e4 | 1373 | struct nvme_queue *nvmeq = &dev->queues[0]; |
4d115420 | 1374 | |
a5cdb68c KB |
1375 | if (shutdown) |
1376 | nvme_shutdown_ctrl(&dev->ctrl); | |
1377 | else | |
b5b05048 | 1378 | nvme_disable_ctrl(&dev->ctrl); |
07836e65 | 1379 | |
bf392a5d | 1380 | nvme_poll_irqdisable(nvmeq); |
b60503ba MW |
1381 | } |
1382 | ||
fa46c6fb KB |
1383 | /* |
1384 | * Called only on a device that has been disabled and after all other threads | |
9210c075 DZ |
1385 | * that can check this device's completion queues have synced, except |
1386 | * nvme_poll(). This is the last chance for the driver to see a natural | |
1387 | * completion before nvme_cancel_request() terminates all incomplete requests. | |
fa46c6fb KB |
1388 | */ |
1389 | static void nvme_reap_pending_cqes(struct nvme_dev *dev) | |
1390 | { | |
fa46c6fb KB |
1391 | int i; |
1392 | ||
9210c075 DZ |
1393 | for (i = dev->ctrl.queue_count - 1; i > 0; i--) { |
1394 | spin_lock(&dev->queues[i].cq_poll_lock); | |
324b494c | 1395 | nvme_process_cq(&dev->queues[i]); |
9210c075 DZ |
1396 | spin_unlock(&dev->queues[i].cq_poll_lock); |
1397 | } | |
fa46c6fb KB |
1398 | } |
1399 | ||
8ffaadf7 JD |
1400 | static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, |
1401 | int entry_size) | |
1402 | { | |
1403 | int q_depth = dev->q_depth; | |
5fd4ce1b CH |
1404 | unsigned q_size_aligned = roundup(q_depth * entry_size, |
1405 | dev->ctrl.page_size); | |
8ffaadf7 JD |
1406 | |
1407 | if (q_size_aligned * nr_io_queues > dev->cmb_size) { | |
c45f5c99 | 1408 | u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); |
5fd4ce1b | 1409 | mem_per_q = round_down(mem_per_q, dev->ctrl.page_size); |
c45f5c99 | 1410 | q_depth = div_u64(mem_per_q, entry_size); |
8ffaadf7 JD |
1411 | |
1412 | /* | |
1413 | * Ensure the reduced q_depth is above some threshold where it | |
1414 | * would be better to map queues in system memory with the | |
1415 | * original depth | |
1416 | */ | |
1417 | if (q_depth < 64) | |
1418 | return -ENOMEM; | |
1419 | } | |
1420 | ||
1421 | return q_depth; | |
1422 | } | |
1423 | ||
1424 | static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, | |
8a1d09a6 | 1425 | int qid) |
8ffaadf7 | 1426 | { |
0f238ff5 LG |
1427 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
1428 | ||
1429 | if (qid && dev->cmb_use_sqes && (dev->cmbsz & NVME_CMBSZ_SQS)) { | |
8a1d09a6 | 1430 | nvmeq->sq_cmds = pci_alloc_p2pmem(pdev, SQ_SIZE(nvmeq)); |
bfac8e9f AM |
1431 | if (nvmeq->sq_cmds) { |
1432 | nvmeq->sq_dma_addr = pci_p2pmem_virt_to_bus(pdev, | |
1433 | nvmeq->sq_cmds); | |
1434 | if (nvmeq->sq_dma_addr) { | |
1435 | set_bit(NVMEQ_SQ_CMB, &nvmeq->flags); | |
1436 | return 0; | |
1437 | } | |
1438 | ||
8a1d09a6 | 1439 | pci_free_p2pmem(pdev, nvmeq->sq_cmds, SQ_SIZE(nvmeq)); |
63223078 | 1440 | } |
0f238ff5 | 1441 | } |
8ffaadf7 | 1442 | |
8a1d09a6 | 1443 | nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(nvmeq), |
63223078 | 1444 | &nvmeq->sq_dma_addr, GFP_KERNEL); |
815c6704 KB |
1445 | if (!nvmeq->sq_cmds) |
1446 | return -ENOMEM; | |
8ffaadf7 JD |
1447 | return 0; |
1448 | } | |
1449 | ||
a6ff7262 | 1450 | static int nvme_alloc_queue(struct nvme_dev *dev, int qid, int depth) |
b60503ba | 1451 | { |
147b27e4 | 1452 | struct nvme_queue *nvmeq = &dev->queues[qid]; |
b60503ba | 1453 | |
62314e40 KB |
1454 | if (dev->ctrl.queue_count > qid) |
1455 | return 0; | |
b60503ba | 1456 | |
c1e0cc7e | 1457 | nvmeq->sqes = qid ? dev->io_sqes : NVME_ADM_SQES; |
8a1d09a6 BH |
1458 | nvmeq->q_depth = depth; |
1459 | nvmeq->cqes = dma_alloc_coherent(dev->dev, CQ_SIZE(nvmeq), | |
750afb08 | 1460 | &nvmeq->cq_dma_addr, GFP_KERNEL); |
b60503ba MW |
1461 | if (!nvmeq->cqes) |
1462 | goto free_nvmeq; | |
b60503ba | 1463 | |
8a1d09a6 | 1464 | if (nvme_alloc_sq_cmds(dev, nvmeq, qid)) |
b60503ba MW |
1465 | goto free_cqdma; |
1466 | ||
091b6092 | 1467 | nvmeq->dev = dev; |
1ab0cd69 | 1468 | spin_lock_init(&nvmeq->sq_lock); |
3a7afd8e | 1469 | spin_lock_init(&nvmeq->cq_poll_lock); |
b60503ba | 1470 | nvmeq->cq_head = 0; |
82123460 | 1471 | nvmeq->cq_phase = 1; |
b80d5ccc | 1472 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
c30341dc | 1473 | nvmeq->qid = qid; |
d858e5f0 | 1474 | dev->ctrl.queue_count++; |
36a7e993 | 1475 | |
147b27e4 | 1476 | return 0; |
b60503ba MW |
1477 | |
1478 | free_cqdma: | |
8a1d09a6 BH |
1479 | dma_free_coherent(dev->dev, CQ_SIZE(nvmeq), (void *)nvmeq->cqes, |
1480 | nvmeq->cq_dma_addr); | |
b60503ba | 1481 | free_nvmeq: |
147b27e4 | 1482 | return -ENOMEM; |
b60503ba MW |
1483 | } |
1484 | ||
dca51e78 | 1485 | static int queue_request_irq(struct nvme_queue *nvmeq) |
3001082c | 1486 | { |
0ff199cb CH |
1487 | struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev); |
1488 | int nr = nvmeq->dev->ctrl.instance; | |
1489 | ||
1490 | if (use_threaded_interrupts) { | |
1491 | return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq_check, | |
1492 | nvme_irq, nvmeq, "nvme%dq%d", nr, nvmeq->qid); | |
1493 | } else { | |
1494 | return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq, | |
1495 | NULL, nvmeq, "nvme%dq%d", nr, nvmeq->qid); | |
1496 | } | |
3001082c MW |
1497 | } |
1498 | ||
22404274 | 1499 | static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) |
b60503ba | 1500 | { |
22404274 | 1501 | struct nvme_dev *dev = nvmeq->dev; |
b60503ba | 1502 | |
22404274 | 1503 | nvmeq->sq_tail = 0; |
04f3eafd | 1504 | nvmeq->last_sq_tail = 0; |
22404274 KB |
1505 | nvmeq->cq_head = 0; |
1506 | nvmeq->cq_phase = 1; | |
b80d5ccc | 1507 | nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; |
8a1d09a6 | 1508 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq)); |
f9f38e33 | 1509 | nvme_dbbuf_init(dev, nvmeq, qid); |
42f61420 | 1510 | dev->online_queues++; |
3a7afd8e | 1511 | wmb(); /* ensure the first interrupt sees the initialization */ |
22404274 KB |
1512 | } |
1513 | ||
4b04cc6a | 1514 | static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled) |
22404274 KB |
1515 | { |
1516 | struct nvme_dev *dev = nvmeq->dev; | |
1517 | int result; | |
7c349dde | 1518 | u16 vector = 0; |
3f85d50b | 1519 | |
d1ed6aa1 CH |
1520 | clear_bit(NVMEQ_DELETE_ERROR, &nvmeq->flags); |
1521 | ||
22b55601 KB |
1522 | /* |
1523 | * A queue's vector matches the queue identifier unless the controller | |
1524 | * has only one vector available. | |
1525 | */ | |
4b04cc6a JA |
1526 | if (!polled) |
1527 | vector = dev->num_vecs == 1 ? 0 : qid; | |
1528 | else | |
7c349dde | 1529 | set_bit(NVMEQ_POLLED, &nvmeq->flags); |
4b04cc6a | 1530 | |
a8e3e0bb | 1531 | result = adapter_alloc_cq(dev, qid, nvmeq, vector); |
ded45505 KB |
1532 | if (result) |
1533 | return result; | |
b60503ba MW |
1534 | |
1535 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
1536 | if (result < 0) | |
ded45505 | 1537 | return result; |
c80b36cd | 1538 | if (result) |
b60503ba MW |
1539 | goto release_cq; |
1540 | ||
a8e3e0bb | 1541 | nvmeq->cq_vector = vector; |
161b8be2 | 1542 | nvme_init_queue(nvmeq, qid); |
4b04cc6a | 1543 | |
7c349dde | 1544 | if (!polled) { |
4b04cc6a JA |
1545 | result = queue_request_irq(nvmeq); |
1546 | if (result < 0) | |
1547 | goto release_sq; | |
1548 | } | |
b60503ba | 1549 | |
4e224106 | 1550 | set_bit(NVMEQ_ENABLED, &nvmeq->flags); |
22404274 | 1551 | return result; |
b60503ba | 1552 | |
a8e3e0bb | 1553 | release_sq: |
f25a2dfc | 1554 | dev->online_queues--; |
b60503ba | 1555 | adapter_delete_sq(dev, qid); |
a8e3e0bb | 1556 | release_cq: |
b60503ba | 1557 | adapter_delete_cq(dev, qid); |
22404274 | 1558 | return result; |
b60503ba MW |
1559 | } |
1560 | ||
f363b089 | 1561 | static const struct blk_mq_ops nvme_mq_admin_ops = { |
d29ec824 | 1562 | .queue_rq = nvme_queue_rq, |
77f02a7a | 1563 | .complete = nvme_pci_complete_rq, |
a4aea562 | 1564 | .init_hctx = nvme_admin_init_hctx, |
0350815a | 1565 | .init_request = nvme_init_request, |
a4aea562 MB |
1566 | .timeout = nvme_timeout, |
1567 | }; | |
1568 | ||
f363b089 | 1569 | static const struct blk_mq_ops nvme_mq_ops = { |
376f7ef8 CH |
1570 | .queue_rq = nvme_queue_rq, |
1571 | .complete = nvme_pci_complete_rq, | |
1572 | .commit_rqs = nvme_commit_rqs, | |
1573 | .init_hctx = nvme_init_hctx, | |
1574 | .init_request = nvme_init_request, | |
1575 | .map_queues = nvme_pci_map_queues, | |
1576 | .timeout = nvme_timeout, | |
1577 | .poll = nvme_poll, | |
dabcefab JA |
1578 | }; |
1579 | ||
ea191d2f KB |
1580 | static void nvme_dev_remove_admin(struct nvme_dev *dev) |
1581 | { | |
1c63dc66 | 1582 | if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) { |
69d9a99c KB |
1583 | /* |
1584 | * If the controller was reset during removal, it's possible | |
1585 | * user requests may be waiting on a stopped queue. Start the | |
1586 | * queue to flush these to completion. | |
1587 | */ | |
c81545f9 | 1588 | blk_mq_unquiesce_queue(dev->ctrl.admin_q); |
1c63dc66 | 1589 | blk_cleanup_queue(dev->ctrl.admin_q); |
ea191d2f KB |
1590 | blk_mq_free_tag_set(&dev->admin_tagset); |
1591 | } | |
1592 | } | |
1593 | ||
a4aea562 MB |
1594 | static int nvme_alloc_admin_tags(struct nvme_dev *dev) |
1595 | { | |
1c63dc66 | 1596 | if (!dev->ctrl.admin_q) { |
a4aea562 MB |
1597 | dev->admin_tagset.ops = &nvme_mq_admin_ops; |
1598 | dev->admin_tagset.nr_hw_queues = 1; | |
e3e9d50c | 1599 | |
38dabe21 | 1600 | dev->admin_tagset.queue_depth = NVME_AQ_MQ_TAG_DEPTH; |
a4aea562 | 1601 | dev->admin_tagset.timeout = ADMIN_TIMEOUT; |
e75ec752 | 1602 | dev->admin_tagset.numa_node = dev_to_node(dev->dev); |
d43f1ccf | 1603 | dev->admin_tagset.cmd_size = sizeof(struct nvme_iod); |
d3484991 | 1604 | dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED; |
a4aea562 MB |
1605 | dev->admin_tagset.driver_data = dev; |
1606 | ||
1607 | if (blk_mq_alloc_tag_set(&dev->admin_tagset)) | |
1608 | return -ENOMEM; | |
34b6c231 | 1609 | dev->ctrl.admin_tagset = &dev->admin_tagset; |
a4aea562 | 1610 | |
1c63dc66 CH |
1611 | dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset); |
1612 | if (IS_ERR(dev->ctrl.admin_q)) { | |
a4aea562 MB |
1613 | blk_mq_free_tag_set(&dev->admin_tagset); |
1614 | return -ENOMEM; | |
1615 | } | |
1c63dc66 | 1616 | if (!blk_get_queue(dev->ctrl.admin_q)) { |
ea191d2f | 1617 | nvme_dev_remove_admin(dev); |
1c63dc66 | 1618 | dev->ctrl.admin_q = NULL; |
ea191d2f KB |
1619 | return -ENODEV; |
1620 | } | |
0fb59cbc | 1621 | } else |
c81545f9 | 1622 | blk_mq_unquiesce_queue(dev->ctrl.admin_q); |
a4aea562 MB |
1623 | |
1624 | return 0; | |
1625 | } | |
1626 | ||
97f6ef64 XY |
1627 | static unsigned long db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) |
1628 | { | |
1629 | return NVME_REG_DBS + ((nr_io_queues + 1) * 8 * dev->db_stride); | |
1630 | } | |
1631 | ||
1632 | static int nvme_remap_bar(struct nvme_dev *dev, unsigned long size) | |
1633 | { | |
1634 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
1635 | ||
1636 | if (size <= dev->bar_mapped_size) | |
1637 | return 0; | |
1638 | if (size > pci_resource_len(pdev, 0)) | |
1639 | return -ENOMEM; | |
1640 | if (dev->bar) | |
1641 | iounmap(dev->bar); | |
1642 | dev->bar = ioremap(pci_resource_start(pdev, 0), size); | |
1643 | if (!dev->bar) { | |
1644 | dev->bar_mapped_size = 0; | |
1645 | return -ENOMEM; | |
1646 | } | |
1647 | dev->bar_mapped_size = size; | |
1648 | dev->dbs = dev->bar + NVME_REG_DBS; | |
1649 | ||
1650 | return 0; | |
1651 | } | |
1652 | ||
01ad0990 | 1653 | static int nvme_pci_configure_admin_queue(struct nvme_dev *dev) |
b60503ba | 1654 | { |
ba47e386 | 1655 | int result; |
b60503ba MW |
1656 | u32 aqa; |
1657 | struct nvme_queue *nvmeq; | |
1658 | ||
97f6ef64 XY |
1659 | result = nvme_remap_bar(dev, db_bar_size(dev, 0)); |
1660 | if (result < 0) | |
1661 | return result; | |
1662 | ||
8ef2074d | 1663 | dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ? |
20d0dfe6 | 1664 | NVME_CAP_NSSRC(dev->ctrl.cap) : 0; |
dfbac8c7 | 1665 | |
7a67cbea CH |
1666 | if (dev->subsystem && |
1667 | (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO)) | |
1668 | writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS); | |
dfbac8c7 | 1669 | |
b5b05048 | 1670 | result = nvme_disable_ctrl(&dev->ctrl); |
ba47e386 MW |
1671 | if (result < 0) |
1672 | return result; | |
b60503ba | 1673 | |
a6ff7262 | 1674 | result = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); |
147b27e4 SG |
1675 | if (result) |
1676 | return result; | |
b60503ba | 1677 | |
147b27e4 | 1678 | nvmeq = &dev->queues[0]; |
b60503ba MW |
1679 | aqa = nvmeq->q_depth - 1; |
1680 | aqa |= aqa << 16; | |
1681 | ||
7a67cbea CH |
1682 | writel(aqa, dev->bar + NVME_REG_AQA); |
1683 | lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ); | |
1684 | lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ); | |
b60503ba | 1685 | |
c0f2f45b | 1686 | result = nvme_enable_ctrl(&dev->ctrl); |
025c557a | 1687 | if (result) |
d4875622 | 1688 | return result; |
a4aea562 | 1689 | |
2b25d981 | 1690 | nvmeq->cq_vector = 0; |
161b8be2 | 1691 | nvme_init_queue(nvmeq, 0); |
dca51e78 | 1692 | result = queue_request_irq(nvmeq); |
758dd7fd | 1693 | if (result) { |
7c349dde | 1694 | dev->online_queues--; |
d4875622 | 1695 | return result; |
758dd7fd | 1696 | } |
025c557a | 1697 | |
4e224106 | 1698 | set_bit(NVMEQ_ENABLED, &nvmeq->flags); |
b60503ba MW |
1699 | return result; |
1700 | } | |
1701 | ||
749941f2 | 1702 | static int nvme_create_io_queues(struct nvme_dev *dev) |
42f61420 | 1703 | { |
4b04cc6a | 1704 | unsigned i, max, rw_queues; |
749941f2 | 1705 | int ret = 0; |
42f61420 | 1706 | |
d858e5f0 | 1707 | for (i = dev->ctrl.queue_count; i <= dev->max_qid; i++) { |
a6ff7262 | 1708 | if (nvme_alloc_queue(dev, i, dev->q_depth)) { |
749941f2 | 1709 | ret = -ENOMEM; |
42f61420 | 1710 | break; |
749941f2 CH |
1711 | } |
1712 | } | |
42f61420 | 1713 | |
d858e5f0 | 1714 | max = min(dev->max_qid, dev->ctrl.queue_count - 1); |
e20ba6e1 CH |
1715 | if (max != 1 && dev->io_queues[HCTX_TYPE_POLL]) { |
1716 | rw_queues = dev->io_queues[HCTX_TYPE_DEFAULT] + | |
1717 | dev->io_queues[HCTX_TYPE_READ]; | |
4b04cc6a JA |
1718 | } else { |
1719 | rw_queues = max; | |
1720 | } | |
1721 | ||
949928c1 | 1722 | for (i = dev->online_queues; i <= max; i++) { |
4b04cc6a JA |
1723 | bool polled = i > rw_queues; |
1724 | ||
1725 | ret = nvme_create_queue(&dev->queues[i], i, polled); | |
d4875622 | 1726 | if (ret) |
42f61420 | 1727 | break; |
27e8166c | 1728 | } |
749941f2 CH |
1729 | |
1730 | /* | |
1731 | * Ignore failing Create SQ/CQ commands, we can continue with less | |
8adb8c14 MI |
1732 | * than the desired amount of queues, and even a controller without |
1733 | * I/O queues can still be used to issue admin commands. This might | |
749941f2 CH |
1734 | * be useful to upgrade a buggy firmware for example. |
1735 | */ | |
1736 | return ret >= 0 ? 0 : ret; | |
b60503ba MW |
1737 | } |
1738 | ||
202021c1 SB |
1739 | static ssize_t nvme_cmb_show(struct device *dev, |
1740 | struct device_attribute *attr, | |
1741 | char *buf) | |
1742 | { | |
1743 | struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev)); | |
1744 | ||
c965809c | 1745 | return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz : x%08x\n", |
202021c1 SB |
1746 | ndev->cmbloc, ndev->cmbsz); |
1747 | } | |
1748 | static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL); | |
1749 | ||
88de4598 | 1750 | static u64 nvme_cmb_size_unit(struct nvme_dev *dev) |
8ffaadf7 | 1751 | { |
88de4598 CH |
1752 | u8 szu = (dev->cmbsz >> NVME_CMBSZ_SZU_SHIFT) & NVME_CMBSZ_SZU_MASK; |
1753 | ||
1754 | return 1ULL << (12 + 4 * szu); | |
1755 | } | |
1756 | ||
1757 | static u32 nvme_cmb_size(struct nvme_dev *dev) | |
1758 | { | |
1759 | return (dev->cmbsz >> NVME_CMBSZ_SZ_SHIFT) & NVME_CMBSZ_SZ_MASK; | |
1760 | } | |
1761 | ||
f65efd6d | 1762 | static void nvme_map_cmb(struct nvme_dev *dev) |
8ffaadf7 | 1763 | { |
88de4598 | 1764 | u64 size, offset; |
8ffaadf7 JD |
1765 | resource_size_t bar_size; |
1766 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
8969f1f8 | 1767 | int bar; |
8ffaadf7 | 1768 | |
9fe5c59f KB |
1769 | if (dev->cmb_size) |
1770 | return; | |
1771 | ||
7a67cbea | 1772 | dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ); |
f65efd6d CH |
1773 | if (!dev->cmbsz) |
1774 | return; | |
202021c1 | 1775 | dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC); |
8ffaadf7 | 1776 | |
88de4598 CH |
1777 | size = nvme_cmb_size_unit(dev) * nvme_cmb_size(dev); |
1778 | offset = nvme_cmb_size_unit(dev) * NVME_CMB_OFST(dev->cmbloc); | |
8969f1f8 CH |
1779 | bar = NVME_CMB_BIR(dev->cmbloc); |
1780 | bar_size = pci_resource_len(pdev, bar); | |
8ffaadf7 JD |
1781 | |
1782 | if (offset > bar_size) | |
f65efd6d | 1783 | return; |
8ffaadf7 JD |
1784 | |
1785 | /* | |
1786 | * Controllers may support a CMB size larger than their BAR, | |
1787 | * for example, due to being behind a bridge. Reduce the CMB to | |
1788 | * the reported size of the BAR | |
1789 | */ | |
1790 | if (size > bar_size - offset) | |
1791 | size = bar_size - offset; | |
1792 | ||
0f238ff5 LG |
1793 | if (pci_p2pdma_add_resource(pdev, bar, size, offset)) { |
1794 | dev_warn(dev->ctrl.device, | |
1795 | "failed to register the CMB\n"); | |
f65efd6d | 1796 | return; |
0f238ff5 LG |
1797 | } |
1798 | ||
8ffaadf7 | 1799 | dev->cmb_size = size; |
0f238ff5 LG |
1800 | dev->cmb_use_sqes = use_cmb_sqes && (dev->cmbsz & NVME_CMBSZ_SQS); |
1801 | ||
1802 | if ((dev->cmbsz & (NVME_CMBSZ_WDS | NVME_CMBSZ_RDS)) == | |
1803 | (NVME_CMBSZ_WDS | NVME_CMBSZ_RDS)) | |
1804 | pci_p2pmem_publish(pdev, true); | |
f65efd6d CH |
1805 | |
1806 | if (sysfs_add_file_to_group(&dev->ctrl.device->kobj, | |
1807 | &dev_attr_cmb.attr, NULL)) | |
1808 | dev_warn(dev->ctrl.device, | |
1809 | "failed to add sysfs attribute for CMB\n"); | |
8ffaadf7 JD |
1810 | } |
1811 | ||
1812 | static inline void nvme_release_cmb(struct nvme_dev *dev) | |
1813 | { | |
0f238ff5 | 1814 | if (dev->cmb_size) { |
1c78f773 MG |
1815 | sysfs_remove_file_from_group(&dev->ctrl.device->kobj, |
1816 | &dev_attr_cmb.attr, NULL); | |
0f238ff5 | 1817 | dev->cmb_size = 0; |
8ffaadf7 JD |
1818 | } |
1819 | } | |
1820 | ||
87ad72a5 CH |
1821 | static int nvme_set_host_mem(struct nvme_dev *dev, u32 bits) |
1822 | { | |
4033f35d | 1823 | u64 dma_addr = dev->host_mem_descs_dma; |
87ad72a5 | 1824 | struct nvme_command c; |
87ad72a5 CH |
1825 | int ret; |
1826 | ||
87ad72a5 CH |
1827 | memset(&c, 0, sizeof(c)); |
1828 | c.features.opcode = nvme_admin_set_features; | |
1829 | c.features.fid = cpu_to_le32(NVME_FEAT_HOST_MEM_BUF); | |
1830 | c.features.dword11 = cpu_to_le32(bits); | |
1831 | c.features.dword12 = cpu_to_le32(dev->host_mem_size >> | |
1832 | ilog2(dev->ctrl.page_size)); | |
1833 | c.features.dword13 = cpu_to_le32(lower_32_bits(dma_addr)); | |
1834 | c.features.dword14 = cpu_to_le32(upper_32_bits(dma_addr)); | |
1835 | c.features.dword15 = cpu_to_le32(dev->nr_host_mem_descs); | |
1836 | ||
1837 | ret = nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); | |
1838 | if (ret) { | |
1839 | dev_warn(dev->ctrl.device, | |
1840 | "failed to set host mem (err %d, flags %#x).\n", | |
1841 | ret, bits); | |
1842 | } | |
87ad72a5 CH |
1843 | return ret; |
1844 | } | |
1845 | ||
1846 | static void nvme_free_host_mem(struct nvme_dev *dev) | |
1847 | { | |
1848 | int i; | |
1849 | ||
1850 | for (i = 0; i < dev->nr_host_mem_descs; i++) { | |
1851 | struct nvme_host_mem_buf_desc *desc = &dev->host_mem_descs[i]; | |
1852 | size_t size = le32_to_cpu(desc->size) * dev->ctrl.page_size; | |
1853 | ||
cc667f6d LD |
1854 | dma_free_attrs(dev->dev, size, dev->host_mem_desc_bufs[i], |
1855 | le64_to_cpu(desc->addr), | |
1856 | DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); | |
87ad72a5 CH |
1857 | } |
1858 | ||
1859 | kfree(dev->host_mem_desc_bufs); | |
1860 | dev->host_mem_desc_bufs = NULL; | |
4033f35d CH |
1861 | dma_free_coherent(dev->dev, |
1862 | dev->nr_host_mem_descs * sizeof(*dev->host_mem_descs), | |
1863 | dev->host_mem_descs, dev->host_mem_descs_dma); | |
87ad72a5 | 1864 | dev->host_mem_descs = NULL; |
7e5dd57e | 1865 | dev->nr_host_mem_descs = 0; |
87ad72a5 CH |
1866 | } |
1867 | ||
92dc6895 CH |
1868 | static int __nvme_alloc_host_mem(struct nvme_dev *dev, u64 preferred, |
1869 | u32 chunk_size) | |
9d713c2b | 1870 | { |
87ad72a5 | 1871 | struct nvme_host_mem_buf_desc *descs; |
92dc6895 | 1872 | u32 max_entries, len; |
4033f35d | 1873 | dma_addr_t descs_dma; |
2ee0e4ed | 1874 | int i = 0; |
87ad72a5 | 1875 | void **bufs; |
6fbcde66 | 1876 | u64 size, tmp; |
87ad72a5 | 1877 | |
87ad72a5 CH |
1878 | tmp = (preferred + chunk_size - 1); |
1879 | do_div(tmp, chunk_size); | |
1880 | max_entries = tmp; | |
044a9df1 CH |
1881 | |
1882 | if (dev->ctrl.hmmaxd && dev->ctrl.hmmaxd < max_entries) | |
1883 | max_entries = dev->ctrl.hmmaxd; | |
1884 | ||
750afb08 LC |
1885 | descs = dma_alloc_coherent(dev->dev, max_entries * sizeof(*descs), |
1886 | &descs_dma, GFP_KERNEL); | |
87ad72a5 CH |
1887 | if (!descs) |
1888 | goto out; | |
1889 | ||
1890 | bufs = kcalloc(max_entries, sizeof(*bufs), GFP_KERNEL); | |
1891 | if (!bufs) | |
1892 | goto out_free_descs; | |
1893 | ||
244a8fe4 | 1894 | for (size = 0; size < preferred && i < max_entries; size += len) { |
87ad72a5 CH |
1895 | dma_addr_t dma_addr; |
1896 | ||
50cdb7c6 | 1897 | len = min_t(u64, chunk_size, preferred - size); |
87ad72a5 CH |
1898 | bufs[i] = dma_alloc_attrs(dev->dev, len, &dma_addr, GFP_KERNEL, |
1899 | DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); | |
1900 | if (!bufs[i]) | |
1901 | break; | |
1902 | ||
1903 | descs[i].addr = cpu_to_le64(dma_addr); | |
1904 | descs[i].size = cpu_to_le32(len / dev->ctrl.page_size); | |
1905 | i++; | |
1906 | } | |
1907 | ||
92dc6895 | 1908 | if (!size) |
87ad72a5 | 1909 | goto out_free_bufs; |
87ad72a5 | 1910 | |
87ad72a5 CH |
1911 | dev->nr_host_mem_descs = i; |
1912 | dev->host_mem_size = size; | |
1913 | dev->host_mem_descs = descs; | |
4033f35d | 1914 | dev->host_mem_descs_dma = descs_dma; |
87ad72a5 CH |
1915 | dev->host_mem_desc_bufs = bufs; |
1916 | return 0; | |
1917 | ||
1918 | out_free_bufs: | |
1919 | while (--i >= 0) { | |
1920 | size_t size = le32_to_cpu(descs[i].size) * dev->ctrl.page_size; | |
1921 | ||
cc667f6d LD |
1922 | dma_free_attrs(dev->dev, size, bufs[i], |
1923 | le64_to_cpu(descs[i].addr), | |
1924 | DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); | |
87ad72a5 CH |
1925 | } |
1926 | ||
1927 | kfree(bufs); | |
1928 | out_free_descs: | |
4033f35d CH |
1929 | dma_free_coherent(dev->dev, max_entries * sizeof(*descs), descs, |
1930 | descs_dma); | |
87ad72a5 | 1931 | out: |
87ad72a5 CH |
1932 | dev->host_mem_descs = NULL; |
1933 | return -ENOMEM; | |
1934 | } | |
1935 | ||
92dc6895 CH |
1936 | static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred) |
1937 | { | |
1938 | u32 chunk_size; | |
1939 | ||
1940 | /* start big and work our way down */ | |
30f92d62 | 1941 | for (chunk_size = min_t(u64, preferred, PAGE_SIZE * MAX_ORDER_NR_PAGES); |
044a9df1 | 1942 | chunk_size >= max_t(u32, dev->ctrl.hmminds * 4096, PAGE_SIZE * 2); |
92dc6895 CH |
1943 | chunk_size /= 2) { |
1944 | if (!__nvme_alloc_host_mem(dev, preferred, chunk_size)) { | |
1945 | if (!min || dev->host_mem_size >= min) | |
1946 | return 0; | |
1947 | nvme_free_host_mem(dev); | |
1948 | } | |
1949 | } | |
1950 | ||
1951 | return -ENOMEM; | |
1952 | } | |
1953 | ||
9620cfba | 1954 | static int nvme_setup_host_mem(struct nvme_dev *dev) |
87ad72a5 CH |
1955 | { |
1956 | u64 max = (u64)max_host_mem_size_mb * SZ_1M; | |
1957 | u64 preferred = (u64)dev->ctrl.hmpre * 4096; | |
1958 | u64 min = (u64)dev->ctrl.hmmin * 4096; | |
1959 | u32 enable_bits = NVME_HOST_MEM_ENABLE; | |
6fbcde66 | 1960 | int ret; |
87ad72a5 CH |
1961 | |
1962 | preferred = min(preferred, max); | |
1963 | if (min > max) { | |
1964 | dev_warn(dev->ctrl.device, | |
1965 | "min host memory (%lld MiB) above limit (%d MiB).\n", | |
1966 | min >> ilog2(SZ_1M), max_host_mem_size_mb); | |
1967 | nvme_free_host_mem(dev); | |
9620cfba | 1968 | return 0; |
87ad72a5 CH |
1969 | } |
1970 | ||
1971 | /* | |
1972 | * If we already have a buffer allocated check if we can reuse it. | |
1973 | */ | |
1974 | if (dev->host_mem_descs) { | |
1975 | if (dev->host_mem_size >= min) | |
1976 | enable_bits |= NVME_HOST_MEM_RETURN; | |
1977 | else | |
1978 | nvme_free_host_mem(dev); | |
1979 | } | |
1980 | ||
1981 | if (!dev->host_mem_descs) { | |
92dc6895 CH |
1982 | if (nvme_alloc_host_mem(dev, min, preferred)) { |
1983 | dev_warn(dev->ctrl.device, | |
1984 | "failed to allocate host memory buffer.\n"); | |
9620cfba | 1985 | return 0; /* controller must work without HMB */ |
92dc6895 CH |
1986 | } |
1987 | ||
1988 | dev_info(dev->ctrl.device, | |
1989 | "allocated %lld MiB host memory buffer.\n", | |
1990 | dev->host_mem_size >> ilog2(SZ_1M)); | |
87ad72a5 CH |
1991 | } |
1992 | ||
9620cfba CH |
1993 | ret = nvme_set_host_mem(dev, enable_bits); |
1994 | if (ret) | |
87ad72a5 | 1995 | nvme_free_host_mem(dev); |
9620cfba | 1996 | return ret; |
9d713c2b KB |
1997 | } |
1998 | ||
612b7286 ML |
1999 | /* |
2000 | * nirqs is the number of interrupts available for write and read | |
2001 | * queues. The core already reserved an interrupt for the admin queue. | |
2002 | */ | |
2003 | static void nvme_calc_irq_sets(struct irq_affinity *affd, unsigned int nrirqs) | |
3b6592f7 | 2004 | { |
612b7286 ML |
2005 | struct nvme_dev *dev = affd->priv; |
2006 | unsigned int nr_read_queues; | |
3b6592f7 JA |
2007 | |
2008 | /* | |
612b7286 ML |
2009 | * If there is no interupt available for queues, ensure that |
2010 | * the default queue is set to 1. The affinity set size is | |
2011 | * also set to one, but the irq core ignores it for this case. | |
2012 | * | |
2013 | * If only one interrupt is available or 'write_queue' == 0, combine | |
2014 | * write and read queues. | |
2015 | * | |
2016 | * If 'write_queues' > 0, ensure it leaves room for at least one read | |
2017 | * queue. | |
3b6592f7 | 2018 | */ |
612b7286 ML |
2019 | if (!nrirqs) { |
2020 | nrirqs = 1; | |
2021 | nr_read_queues = 0; | |
2022 | } else if (nrirqs == 1 || !write_queues) { | |
2023 | nr_read_queues = 0; | |
2024 | } else if (write_queues >= nrirqs) { | |
2025 | nr_read_queues = 1; | |
3b6592f7 | 2026 | } else { |
612b7286 | 2027 | nr_read_queues = nrirqs - write_queues; |
3b6592f7 | 2028 | } |
612b7286 ML |
2029 | |
2030 | dev->io_queues[HCTX_TYPE_DEFAULT] = nrirqs - nr_read_queues; | |
2031 | affd->set_size[HCTX_TYPE_DEFAULT] = nrirqs - nr_read_queues; | |
2032 | dev->io_queues[HCTX_TYPE_READ] = nr_read_queues; | |
2033 | affd->set_size[HCTX_TYPE_READ] = nr_read_queues; | |
2034 | affd->nr_sets = nr_read_queues ? 2 : 1; | |
3b6592f7 JA |
2035 | } |
2036 | ||
6451fe73 | 2037 | static int nvme_setup_irqs(struct nvme_dev *dev, unsigned int nr_io_queues) |
3b6592f7 JA |
2038 | { |
2039 | struct pci_dev *pdev = to_pci_dev(dev->dev); | |
3b6592f7 | 2040 | struct irq_affinity affd = { |
9cfef55b | 2041 | .pre_vectors = 1, |
612b7286 ML |
2042 | .calc_sets = nvme_calc_irq_sets, |
2043 | .priv = dev, | |
3b6592f7 | 2044 | }; |
6451fe73 JA |
2045 | unsigned int irq_queues, this_p_queues; |
2046 | ||
2047 | /* | |
2048 | * Poll queues don't need interrupts, but we need at least one IO | |
2049 | * queue left over for non-polled IO. | |
2050 | */ | |
2051 | this_p_queues = poll_queues; | |
2052 | if (this_p_queues >= nr_io_queues) { | |
2053 | this_p_queues = nr_io_queues - 1; | |
2054 | irq_queues = 1; | |
2055 | } else { | |
7e4c6b9a | 2056 | irq_queues = nr_io_queues - this_p_queues + 1; |
6451fe73 JA |
2057 | } |
2058 | dev->io_queues[HCTX_TYPE_POLL] = this_p_queues; | |
3b6592f7 | 2059 | |
612b7286 ML |
2060 | /* Initialize for the single interrupt case */ |
2061 | dev->io_queues[HCTX_TYPE_DEFAULT] = 1; | |
2062 | dev->io_queues[HCTX_TYPE_READ] = 0; | |
3b6592f7 | 2063 | |
66341331 BH |
2064 | /* |
2065 | * Some Apple controllers require all queues to use the | |
2066 | * first vector. | |
2067 | */ | |
2068 | if (dev->ctrl.quirks & NVME_QUIRK_SINGLE_VECTOR) | |
2069 | irq_queues = 1; | |
2070 | ||
612b7286 ML |
2071 | return pci_alloc_irq_vectors_affinity(pdev, 1, irq_queues, |
2072 | PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY, &affd); | |
3b6592f7 JA |
2073 | } |
2074 | ||
8fae268b KB |
2075 | static void nvme_disable_io_queues(struct nvme_dev *dev) |
2076 | { | |
2077 | if (__nvme_disable_io_queues(dev, nvme_admin_delete_sq)) | |
2078 | __nvme_disable_io_queues(dev, nvme_admin_delete_cq); | |
2079 | } | |
2080 | ||
8d85fce7 | 2081 | static int nvme_setup_io_queues(struct nvme_dev *dev) |
b60503ba | 2082 | { |
147b27e4 | 2083 | struct nvme_queue *adminq = &dev->queues[0]; |
e75ec752 | 2084 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
97f6ef64 XY |
2085 | int result, nr_io_queues; |
2086 | unsigned long size; | |
b60503ba | 2087 | |
3b6592f7 | 2088 | nr_io_queues = max_io_queues(); |
d38e9f04 BH |
2089 | |
2090 | /* | |
2091 | * If tags are shared with admin queue (Apple bug), then | |
2092 | * make sure we only use one IO queue. | |
2093 | */ | |
2094 | if (dev->ctrl.quirks & NVME_QUIRK_SHARED_TAGS) | |
2095 | nr_io_queues = 1; | |
2096 | ||
9a0be7ab CH |
2097 | result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues); |
2098 | if (result < 0) | |
1b23484b | 2099 | return result; |
9a0be7ab | 2100 | |
f5fa90dc | 2101 | if (nr_io_queues == 0) |
a5229050 | 2102 | return 0; |
4e224106 CH |
2103 | |
2104 | clear_bit(NVMEQ_ENABLED, &adminq->flags); | |
b60503ba | 2105 | |
0f238ff5 | 2106 | if (dev->cmb_use_sqes) { |
8ffaadf7 JD |
2107 | result = nvme_cmb_qdepth(dev, nr_io_queues, |
2108 | sizeof(struct nvme_command)); | |
2109 | if (result > 0) | |
2110 | dev->q_depth = result; | |
2111 | else | |
0f238ff5 | 2112 | dev->cmb_use_sqes = false; |
8ffaadf7 JD |
2113 | } |
2114 | ||
97f6ef64 XY |
2115 | do { |
2116 | size = db_bar_size(dev, nr_io_queues); | |
2117 | result = nvme_remap_bar(dev, size); | |
2118 | if (!result) | |
2119 | break; | |
2120 | if (!--nr_io_queues) | |
2121 | return -ENOMEM; | |
2122 | } while (1); | |
2123 | adminq->q_db = dev->dbs; | |
f1938f6e | 2124 | |
8fae268b | 2125 | retry: |
9d713c2b | 2126 | /* Deregister the admin queue's interrupt */ |
0ff199cb | 2127 | pci_free_irq(pdev, 0, adminq); |
9d713c2b | 2128 | |
e32efbfc JA |
2129 | /* |
2130 | * If we enable msix early due to not intx, disable it again before | |
2131 | * setting up the full range we need. | |
2132 | */ | |
dca51e78 | 2133 | pci_free_irq_vectors(pdev); |
3b6592f7 JA |
2134 | |
2135 | result = nvme_setup_irqs(dev, nr_io_queues); | |
22b55601 | 2136 | if (result <= 0) |
dca51e78 | 2137 | return -EIO; |
3b6592f7 | 2138 | |
22b55601 | 2139 | dev->num_vecs = result; |
4b04cc6a | 2140 | result = max(result - 1, 1); |
e20ba6e1 | 2141 | dev->max_qid = result + dev->io_queues[HCTX_TYPE_POLL]; |
fa08a396 | 2142 | |
063a8096 MW |
2143 | /* |
2144 | * Should investigate if there's a performance win from allocating | |
2145 | * more queues than interrupt vectors; it might allow the submission | |
2146 | * path to scale better, even if the receive path is limited by the | |
2147 | * number of interrupts. | |
2148 | */ | |
dca51e78 | 2149 | result = queue_request_irq(adminq); |
7c349dde | 2150 | if (result) |
d4875622 | 2151 | return result; |
4e224106 | 2152 | set_bit(NVMEQ_ENABLED, &adminq->flags); |
8fae268b KB |
2153 | |
2154 | result = nvme_create_io_queues(dev); | |
2155 | if (result || dev->online_queues < 2) | |
2156 | return result; | |
2157 | ||
2158 | if (dev->online_queues - 1 < dev->max_qid) { | |
2159 | nr_io_queues = dev->online_queues - 1; | |
2160 | nvme_disable_io_queues(dev); | |
2161 | nvme_suspend_io_queues(dev); | |
2162 | goto retry; | |
2163 | } | |
2164 | dev_info(dev->ctrl.device, "%d/%d/%d default/read/poll queues\n", | |
2165 | dev->io_queues[HCTX_TYPE_DEFAULT], | |
2166 | dev->io_queues[HCTX_TYPE_READ], | |
2167 | dev->io_queues[HCTX_TYPE_POLL]); | |
2168 | return 0; | |
b60503ba MW |
2169 | } |
2170 | ||
2a842aca | 2171 | static void nvme_del_queue_end(struct request *req, blk_status_t error) |
a5768aa8 | 2172 | { |
db3cbfff | 2173 | struct nvme_queue *nvmeq = req->end_io_data; |
b5875222 | 2174 | |
db3cbfff | 2175 | blk_mq_free_request(req); |
d1ed6aa1 | 2176 | complete(&nvmeq->delete_done); |
a5768aa8 KB |
2177 | } |
2178 | ||
2a842aca | 2179 | static void nvme_del_cq_end(struct request *req, blk_status_t error) |
a5768aa8 | 2180 | { |
db3cbfff | 2181 | struct nvme_queue *nvmeq = req->end_io_data; |
a5768aa8 | 2182 | |
d1ed6aa1 CH |
2183 | if (error) |
2184 | set_bit(NVMEQ_DELETE_ERROR, &nvmeq->flags); | |
db3cbfff KB |
2185 | |
2186 | nvme_del_queue_end(req, error); | |
a5768aa8 KB |
2187 | } |
2188 | ||
db3cbfff | 2189 | static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode) |
bda4e0fb | 2190 | { |
db3cbfff KB |
2191 | struct request_queue *q = nvmeq->dev->ctrl.admin_q; |
2192 | struct request *req; | |
2193 | struct nvme_command cmd; | |
bda4e0fb | 2194 | |
db3cbfff KB |
2195 | memset(&cmd, 0, sizeof(cmd)); |
2196 | cmd.delete_queue.opcode = opcode; | |
2197 | cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid); | |
bda4e0fb | 2198 | |
eb71f435 | 2199 | req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); |
db3cbfff KB |
2200 | if (IS_ERR(req)) |
2201 | return PTR_ERR(req); | |
bda4e0fb | 2202 | |
db3cbfff KB |
2203 | req->timeout = ADMIN_TIMEOUT; |
2204 | req->end_io_data = nvmeq; | |
2205 | ||
d1ed6aa1 | 2206 | init_completion(&nvmeq->delete_done); |
db3cbfff KB |
2207 | blk_execute_rq_nowait(q, NULL, req, false, |
2208 | opcode == nvme_admin_delete_cq ? | |
2209 | nvme_del_cq_end : nvme_del_queue_end); | |
2210 | return 0; | |
bda4e0fb KB |
2211 | } |
2212 | ||
8fae268b | 2213 | static bool __nvme_disable_io_queues(struct nvme_dev *dev, u8 opcode) |
a5768aa8 | 2214 | { |
5271edd4 | 2215 | int nr_queues = dev->online_queues - 1, sent = 0; |
db3cbfff | 2216 | unsigned long timeout; |
a5768aa8 | 2217 | |
db3cbfff | 2218 | retry: |
5271edd4 CH |
2219 | timeout = ADMIN_TIMEOUT; |
2220 | while (nr_queues > 0) { | |
2221 | if (nvme_delete_queue(&dev->queues[nr_queues], opcode)) | |
2222 | break; | |
2223 | nr_queues--; | |
2224 | sent++; | |
db3cbfff | 2225 | } |
d1ed6aa1 CH |
2226 | while (sent) { |
2227 | struct nvme_queue *nvmeq = &dev->queues[nr_queues + sent]; | |
2228 | ||
2229 | timeout = wait_for_completion_io_timeout(&nvmeq->delete_done, | |
5271edd4 CH |
2230 | timeout); |
2231 | if (timeout == 0) | |
2232 | return false; | |
d1ed6aa1 | 2233 | |
d1ed6aa1 | 2234 | sent--; |
5271edd4 CH |
2235 | if (nr_queues) |
2236 | goto retry; | |
2237 | } | |
2238 | return true; | |
a5768aa8 KB |
2239 | } |
2240 | ||
5d02a5c1 | 2241 | static void nvme_dev_add(struct nvme_dev *dev) |
b60503ba | 2242 | { |
2b1b7e78 JW |
2243 | int ret; |
2244 | ||
5bae7f73 | 2245 | if (!dev->ctrl.tagset) { |
376f7ef8 | 2246 | dev->tagset.ops = &nvme_mq_ops; |
ffe7704d | 2247 | dev->tagset.nr_hw_queues = dev->online_queues - 1; |
8fe34be1 | 2248 | dev->tagset.nr_maps = 2; /* default + read */ |
ed92ad37 CH |
2249 | if (dev->io_queues[HCTX_TYPE_POLL]) |
2250 | dev->tagset.nr_maps++; | |
ffe7704d KB |
2251 | dev->tagset.timeout = NVME_IO_TIMEOUT; |
2252 | dev->tagset.numa_node = dev_to_node(dev->dev); | |
2253 | dev->tagset.queue_depth = | |
a4aea562 | 2254 | min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; |
d43f1ccf | 2255 | dev->tagset.cmd_size = sizeof(struct nvme_iod); |
ffe7704d KB |
2256 | dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; |
2257 | dev->tagset.driver_data = dev; | |
b60503ba | 2258 | |
d38e9f04 BH |
2259 | /* |
2260 | * Some Apple controllers requires tags to be unique | |
2261 | * across admin and IO queue, so reserve the first 32 | |
2262 | * tags of the IO queue. | |
2263 | */ | |
2264 | if (dev->ctrl.quirks & NVME_QUIRK_SHARED_TAGS) | |
2265 | dev->tagset.reserved_tags = NVME_AQ_DEPTH; | |
2266 | ||
2b1b7e78 JW |
2267 | ret = blk_mq_alloc_tag_set(&dev->tagset); |
2268 | if (ret) { | |
2269 | dev_warn(dev->ctrl.device, | |
2270 | "IO queues tagset allocation failed %d\n", ret); | |
5d02a5c1 | 2271 | return; |
2b1b7e78 | 2272 | } |
5bae7f73 | 2273 | dev->ctrl.tagset = &dev->tagset; |
949928c1 KB |
2274 | } else { |
2275 | blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1); | |
2276 | ||
2277 | /* Free previously allocated queues that are no longer usable */ | |
2278 | nvme_free_queues(dev, dev->online_queues); | |
ffe7704d | 2279 | } |
949928c1 | 2280 | |
e8fd41bb | 2281 | nvme_dbbuf_set(dev); |
b60503ba MW |
2282 | } |
2283 | ||
b00a726a | 2284 | static int nvme_pci_enable(struct nvme_dev *dev) |
0877cb0d | 2285 | { |
b00a726a | 2286 | int result = -ENOMEM; |
e75ec752 | 2287 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
0877cb0d KB |
2288 | |
2289 | if (pci_enable_device_mem(pdev)) | |
2290 | return result; | |
2291 | ||
0877cb0d | 2292 | pci_set_master(pdev); |
0877cb0d | 2293 | |
4fe06923 | 2294 | if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64))) |
052d0efa | 2295 | goto disable; |
0877cb0d | 2296 | |
7a67cbea | 2297 | if (readl(dev->bar + NVME_REG_CSTS) == -1) { |
0e53d180 | 2298 | result = -ENODEV; |
b00a726a | 2299 | goto disable; |
0e53d180 | 2300 | } |
e32efbfc JA |
2301 | |
2302 | /* | |
a5229050 KB |
2303 | * Some devices and/or platforms don't advertise or work with INTx |
2304 | * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll | |
2305 | * adjust this later. | |
e32efbfc | 2306 | */ |
dca51e78 CH |
2307 | result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES); |
2308 | if (result < 0) | |
2309 | return result; | |
e32efbfc | 2310 | |
20d0dfe6 | 2311 | dev->ctrl.cap = lo_hi_readq(dev->bar + NVME_REG_CAP); |
7a67cbea | 2312 | |
20d0dfe6 | 2313 | dev->q_depth = min_t(int, NVME_CAP_MQES(dev->ctrl.cap) + 1, |
b27c1e68 | 2314 | io_queue_depth); |
aa22c8e6 | 2315 | dev->ctrl.sqsize = dev->q_depth - 1; /* 0's based queue depth */ |
20d0dfe6 | 2316 | dev->db_stride = 1 << NVME_CAP_STRIDE(dev->ctrl.cap); |
7a67cbea | 2317 | dev->dbs = dev->bar + 4096; |
1f390c1f | 2318 | |
66341331 BH |
2319 | /* |
2320 | * Some Apple controllers require a non-standard SQE size. | |
2321 | * Interestingly they also seem to ignore the CC:IOSQES register | |
2322 | * so we don't bother updating it here. | |
2323 | */ | |
2324 | if (dev->ctrl.quirks & NVME_QUIRK_128_BYTES_SQES) | |
2325 | dev->io_sqes = 7; | |
2326 | else | |
2327 | dev->io_sqes = NVME_NVM_IOSQES; | |
1f390c1f SG |
2328 | |
2329 | /* | |
2330 | * Temporary fix for the Apple controller found in the MacBook8,1 and | |
2331 | * some MacBook7,1 to avoid controller resets and data loss. | |
2332 | */ | |
2333 | if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) { | |
2334 | dev->q_depth = 2; | |
9bdcfb10 CH |
2335 | dev_warn(dev->ctrl.device, "detected Apple NVMe controller, " |
2336 | "set queue depth=%u to work around controller resets\n", | |
1f390c1f | 2337 | dev->q_depth); |
d554b5e1 MP |
2338 | } else if (pdev->vendor == PCI_VENDOR_ID_SAMSUNG && |
2339 | (pdev->device == 0xa821 || pdev->device == 0xa822) && | |
20d0dfe6 | 2340 | NVME_CAP_MQES(dev->ctrl.cap) == 0) { |
d554b5e1 MP |
2341 | dev->q_depth = 64; |
2342 | dev_err(dev->ctrl.device, "detected PM1725 NVMe controller, " | |
2343 | "set queue depth=%u\n", dev->q_depth); | |
1f390c1f SG |
2344 | } |
2345 | ||
d38e9f04 BH |
2346 | /* |
2347 | * Controllers with the shared tags quirk need the IO queue to be | |
2348 | * big enough so that we get 32 tags for the admin queue | |
2349 | */ | |
2350 | if ((dev->ctrl.quirks & NVME_QUIRK_SHARED_TAGS) && | |
2351 | (dev->q_depth < (NVME_AQ_DEPTH + 2))) { | |
2352 | dev->q_depth = NVME_AQ_DEPTH + 2; | |
2353 | dev_warn(dev->ctrl.device, "IO queue depth clamped to %d\n", | |
2354 | dev->q_depth); | |
2355 | } | |
2356 | ||
2357 | ||
f65efd6d | 2358 | nvme_map_cmb(dev); |
202021c1 | 2359 | |
a0a3408e KB |
2360 | pci_enable_pcie_error_reporting(pdev); |
2361 | pci_save_state(pdev); | |
0877cb0d KB |
2362 | return 0; |
2363 | ||
2364 | disable: | |
0877cb0d KB |
2365 | pci_disable_device(pdev); |
2366 | return result; | |
2367 | } | |
2368 | ||
2369 | static void nvme_dev_unmap(struct nvme_dev *dev) | |
b00a726a KB |
2370 | { |
2371 | if (dev->bar) | |
2372 | iounmap(dev->bar); | |
a1f447b3 | 2373 | pci_release_mem_regions(to_pci_dev(dev->dev)); |
b00a726a KB |
2374 | } |
2375 | ||
2376 | static void nvme_pci_disable(struct nvme_dev *dev) | |
0877cb0d | 2377 | { |
e75ec752 CH |
2378 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2379 | ||
dca51e78 | 2380 | pci_free_irq_vectors(pdev); |
0877cb0d | 2381 | |
a0a3408e KB |
2382 | if (pci_is_enabled(pdev)) { |
2383 | pci_disable_pcie_error_reporting(pdev); | |
e75ec752 | 2384 | pci_disable_device(pdev); |
4d115420 | 2385 | } |
4d115420 KB |
2386 | } |
2387 | ||
a5cdb68c | 2388 | static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown) |
b60503ba | 2389 | { |
e43269e6 | 2390 | bool dead = true, freeze = false; |
302ad8cc | 2391 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
22404274 | 2392 | |
77bf25ea | 2393 | mutex_lock(&dev->shutdown_lock); |
302ad8cc KB |
2394 | if (pci_is_enabled(pdev)) { |
2395 | u32 csts = readl(dev->bar + NVME_REG_CSTS); | |
2396 | ||
ebef7368 | 2397 | if (dev->ctrl.state == NVME_CTRL_LIVE || |
e43269e6 KB |
2398 | dev->ctrl.state == NVME_CTRL_RESETTING) { |
2399 | freeze = true; | |
302ad8cc | 2400 | nvme_start_freeze(&dev->ctrl); |
e43269e6 | 2401 | } |
302ad8cc KB |
2402 | dead = !!((csts & NVME_CSTS_CFS) || !(csts & NVME_CSTS_RDY) || |
2403 | pdev->error_state != pci_channel_io_normal); | |
c9d3bf88 | 2404 | } |
c21377f8 | 2405 | |
302ad8cc KB |
2406 | /* |
2407 | * Give the controller a chance to complete all entered requests if | |
2408 | * doing a safe shutdown. | |
2409 | */ | |
e43269e6 KB |
2410 | if (!dead && shutdown && freeze) |
2411 | nvme_wait_freeze_timeout(&dev->ctrl, NVME_IO_TIMEOUT); | |
9a915a5b JW |
2412 | |
2413 | nvme_stop_queues(&dev->ctrl); | |
87ad72a5 | 2414 | |
64ee0ac0 | 2415 | if (!dead && dev->ctrl.queue_count > 0) { |
8fae268b | 2416 | nvme_disable_io_queues(dev); |
a5cdb68c | 2417 | nvme_disable_admin_queue(dev, shutdown); |
4d115420 | 2418 | } |
8fae268b KB |
2419 | nvme_suspend_io_queues(dev); |
2420 | nvme_suspend_queue(&dev->queues[0]); | |
b00a726a | 2421 | nvme_pci_disable(dev); |
fa46c6fb | 2422 | nvme_reap_pending_cqes(dev); |
07836e65 | 2423 | |
e1958e65 ML |
2424 | blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl); |
2425 | blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl); | |
622b8b68 ML |
2426 | blk_mq_tagset_wait_completed_request(&dev->tagset); |
2427 | blk_mq_tagset_wait_completed_request(&dev->admin_tagset); | |
302ad8cc KB |
2428 | |
2429 | /* | |
2430 | * The driver will not be starting up queues again if shutting down so | |
2431 | * must flush all entered requests to their failed completion to avoid | |
2432 | * deadlocking blk-mq hot-cpu notifier. | |
2433 | */ | |
c8e9e9b7 | 2434 | if (shutdown) { |
302ad8cc | 2435 | nvme_start_queues(&dev->ctrl); |
c8e9e9b7 KB |
2436 | if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) |
2437 | blk_mq_unquiesce_queue(dev->ctrl.admin_q); | |
2438 | } | |
77bf25ea | 2439 | mutex_unlock(&dev->shutdown_lock); |
b60503ba MW |
2440 | } |
2441 | ||
c1ac9a4b KB |
2442 | static int nvme_disable_prepare_reset(struct nvme_dev *dev, bool shutdown) |
2443 | { | |
2444 | if (!nvme_wait_reset(&dev->ctrl)) | |
2445 | return -EBUSY; | |
2446 | nvme_dev_disable(dev, shutdown); | |
2447 | return 0; | |
2448 | } | |
2449 | ||
091b6092 MW |
2450 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
2451 | { | |
e75ec752 | 2452 | dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, |
091b6092 MW |
2453 | PAGE_SIZE, PAGE_SIZE, 0); |
2454 | if (!dev->prp_page_pool) | |
2455 | return -ENOMEM; | |
2456 | ||
99802a7a | 2457 | /* Optimisation for I/Os between 4k and 128k */ |
e75ec752 | 2458 | dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, |
99802a7a MW |
2459 | 256, 256, 0); |
2460 | if (!dev->prp_small_pool) { | |
2461 | dma_pool_destroy(dev->prp_page_pool); | |
2462 | return -ENOMEM; | |
2463 | } | |
091b6092 MW |
2464 | return 0; |
2465 | } | |
2466 | ||
2467 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
2468 | { | |
2469 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 2470 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
2471 | } |
2472 | ||
770597ec KB |
2473 | static void nvme_free_tagset(struct nvme_dev *dev) |
2474 | { | |
2475 | if (dev->tagset.tags) | |
2476 | blk_mq_free_tag_set(&dev->tagset); | |
2477 | dev->ctrl.tagset = NULL; | |
2478 | } | |
2479 | ||
1673f1f0 | 2480 | static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl) |
5e82e952 | 2481 | { |
1673f1f0 | 2482 | struct nvme_dev *dev = to_nvme_dev(ctrl); |
9ac27090 | 2483 | |
f9f38e33 | 2484 | nvme_dbbuf_dma_free(dev); |
770597ec | 2485 | nvme_free_tagset(dev); |
1c63dc66 CH |
2486 | if (dev->ctrl.admin_q) |
2487 | blk_put_queue(dev->ctrl.admin_q); | |
e286bcfc | 2488 | free_opal_dev(dev->ctrl.opal_dev); |
943e942e | 2489 | mempool_destroy(dev->iod_mempool); |
253fd4ac IR |
2490 | put_device(dev->dev); |
2491 | kfree(dev->queues); | |
5e82e952 KB |
2492 | kfree(dev); |
2493 | } | |
2494 | ||
7c1ce408 | 2495 | static void nvme_remove_dead_ctrl(struct nvme_dev *dev) |
f58944e2 | 2496 | { |
c1ac9a4b KB |
2497 | /* |
2498 | * Set state to deleting now to avoid blocking nvme_wait_reset(), which | |
2499 | * may be holding this pci_dev's device lock. | |
2500 | */ | |
2501 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); | |
d22524a4 | 2502 | nvme_get_ctrl(&dev->ctrl); |
69d9a99c | 2503 | nvme_dev_disable(dev, false); |
9f9cafc1 | 2504 | nvme_kill_queues(&dev->ctrl); |
03e0f3a6 | 2505 | if (!queue_work(nvme_wq, &dev->remove_work)) |
f58944e2 KB |
2506 | nvme_put_ctrl(&dev->ctrl); |
2507 | } | |
2508 | ||
fd634f41 | 2509 | static void nvme_reset_work(struct work_struct *work) |
5e82e952 | 2510 | { |
d86c4d8e CH |
2511 | struct nvme_dev *dev = |
2512 | container_of(work, struct nvme_dev, ctrl.reset_work); | |
a98e58e5 | 2513 | bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL); |
e71afda4 | 2514 | int result; |
5e82e952 | 2515 | |
e71afda4 CK |
2516 | if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING)) { |
2517 | result = -ENODEV; | |
fd634f41 | 2518 | goto out; |
e71afda4 | 2519 | } |
5e82e952 | 2520 | |
fd634f41 CH |
2521 | /* |
2522 | * If we're called to reset a live controller first shut it down before | |
2523 | * moving on. | |
2524 | */ | |
b00a726a | 2525 | if (dev->ctrl.ctrl_config & NVME_CC_ENABLE) |
a5cdb68c | 2526 | nvme_dev_disable(dev, false); |
d6135c3a | 2527 | nvme_sync_queues(&dev->ctrl); |
5e82e952 | 2528 | |
5c959d73 | 2529 | mutex_lock(&dev->shutdown_lock); |
b00a726a | 2530 | result = nvme_pci_enable(dev); |
f0b50732 | 2531 | if (result) |
4726bcf3 | 2532 | goto out_unlock; |
f0b50732 | 2533 | |
01ad0990 | 2534 | result = nvme_pci_configure_admin_queue(dev); |
f0b50732 | 2535 | if (result) |
4726bcf3 | 2536 | goto out_unlock; |
f0b50732 | 2537 | |
0fb59cbc KB |
2538 | result = nvme_alloc_admin_tags(dev); |
2539 | if (result) | |
4726bcf3 | 2540 | goto out_unlock; |
b9afca3e | 2541 | |
943e942e JA |
2542 | /* |
2543 | * Limit the max command size to prevent iod->sg allocations going | |
2544 | * over a single page. | |
2545 | */ | |
7637de31 CH |
2546 | dev->ctrl.max_hw_sectors = min_t(u32, |
2547 | NVME_MAX_KB_SZ << 1, dma_max_mapping_size(dev->dev) >> 9); | |
943e942e | 2548 | dev->ctrl.max_segments = NVME_MAX_SEGS; |
a48bc520 CH |
2549 | |
2550 | /* | |
2551 | * Don't limit the IOMMU merged segment size. | |
2552 | */ | |
2553 | dma_set_max_seg_size(dev->dev, 0xffffffff); | |
2554 | ||
5c959d73 KB |
2555 | mutex_unlock(&dev->shutdown_lock); |
2556 | ||
2557 | /* | |
2558 | * Introduce CONNECTING state from nvme-fc/rdma transports to mark the | |
2559 | * initializing procedure here. | |
2560 | */ | |
2561 | if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_CONNECTING)) { | |
2562 | dev_warn(dev->ctrl.device, | |
2563 | "failed to mark controller CONNECTING\n"); | |
cee6c269 | 2564 | result = -EBUSY; |
5c959d73 KB |
2565 | goto out; |
2566 | } | |
943e942e | 2567 | |
ce4541f4 CH |
2568 | result = nvme_init_identify(&dev->ctrl); |
2569 | if (result) | |
f58944e2 | 2570 | goto out; |
ce4541f4 | 2571 | |
e286bcfc SB |
2572 | if (dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) { |
2573 | if (!dev->ctrl.opal_dev) | |
2574 | dev->ctrl.opal_dev = | |
2575 | init_opal_dev(&dev->ctrl, &nvme_sec_submit); | |
2576 | else if (was_suspend) | |
2577 | opal_unlock_from_suspend(dev->ctrl.opal_dev); | |
2578 | } else { | |
2579 | free_opal_dev(dev->ctrl.opal_dev); | |
2580 | dev->ctrl.opal_dev = NULL; | |
4f1244c8 | 2581 | } |
a98e58e5 | 2582 | |
f9f38e33 HK |
2583 | if (dev->ctrl.oacs & NVME_CTRL_OACS_DBBUF_SUPP) { |
2584 | result = nvme_dbbuf_dma_alloc(dev); | |
2585 | if (result) | |
2586 | dev_warn(dev->dev, | |
2587 | "unable to allocate dma for dbbuf\n"); | |
2588 | } | |
2589 | ||
9620cfba CH |
2590 | if (dev->ctrl.hmpre) { |
2591 | result = nvme_setup_host_mem(dev); | |
2592 | if (result < 0) | |
2593 | goto out; | |
2594 | } | |
87ad72a5 | 2595 | |
f0b50732 | 2596 | result = nvme_setup_io_queues(dev); |
badc34d4 | 2597 | if (result) |
f58944e2 | 2598 | goto out; |
f0b50732 | 2599 | |
2659e57b CH |
2600 | /* |
2601 | * Keep the controller around but remove all namespaces if we don't have | |
2602 | * any working I/O queue. | |
2603 | */ | |
3cf519b5 | 2604 | if (dev->online_queues < 2) { |
1b3c47c1 | 2605 | dev_warn(dev->ctrl.device, "IO queues not created\n"); |
3b24774e | 2606 | nvme_kill_queues(&dev->ctrl); |
5bae7f73 | 2607 | nvme_remove_namespaces(&dev->ctrl); |
770597ec | 2608 | nvme_free_tagset(dev); |
3cf519b5 | 2609 | } else { |
25646264 | 2610 | nvme_start_queues(&dev->ctrl); |
302ad8cc | 2611 | nvme_wait_freeze(&dev->ctrl); |
5d02a5c1 | 2612 | nvme_dev_add(dev); |
302ad8cc | 2613 | nvme_unfreeze(&dev->ctrl); |
3cf519b5 CH |
2614 | } |
2615 | ||
2b1b7e78 JW |
2616 | /* |
2617 | * If only admin queue live, keep it to do further investigation or | |
2618 | * recovery. | |
2619 | */ | |
5d02a5c1 | 2620 | if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) { |
2b1b7e78 | 2621 | dev_warn(dev->ctrl.device, |
5d02a5c1 | 2622 | "failed to mark controller live state\n"); |
e71afda4 | 2623 | result = -ENODEV; |
bb8d261e CH |
2624 | goto out; |
2625 | } | |
92911a55 | 2626 | |
d09f2b45 | 2627 | nvme_start_ctrl(&dev->ctrl); |
3cf519b5 | 2628 | return; |
f0b50732 | 2629 | |
4726bcf3 KB |
2630 | out_unlock: |
2631 | mutex_unlock(&dev->shutdown_lock); | |
3cf519b5 | 2632 | out: |
7c1ce408 CK |
2633 | if (result) |
2634 | dev_warn(dev->ctrl.device, | |
2635 | "Removing after probe failure status: %d\n", result); | |
2636 | nvme_remove_dead_ctrl(dev); | |
f0b50732 KB |
2637 | } |
2638 | ||
5c8809e6 | 2639 | static void nvme_remove_dead_ctrl_work(struct work_struct *work) |
9a6b9458 | 2640 | { |
5c8809e6 | 2641 | struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work); |
e75ec752 | 2642 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
9a6b9458 KB |
2643 | |
2644 | if (pci_get_drvdata(pdev)) | |
921920ab | 2645 | device_release_driver(&pdev->dev); |
1673f1f0 | 2646 | nvme_put_ctrl(&dev->ctrl); |
9a6b9458 KB |
2647 | } |
2648 | ||
1c63dc66 | 2649 | static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) |
9ca97374 | 2650 | { |
1c63dc66 | 2651 | *val = readl(to_nvme_dev(ctrl)->bar + off); |
90667892 | 2652 | return 0; |
9ca97374 TH |
2653 | } |
2654 | ||
5fd4ce1b | 2655 | static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) |
4cc06521 | 2656 | { |
5fd4ce1b CH |
2657 | writel(val, to_nvme_dev(ctrl)->bar + off); |
2658 | return 0; | |
2659 | } | |
4cc06521 | 2660 | |
7fd8930f CH |
2661 | static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) |
2662 | { | |
3a8ecc93 | 2663 | *val = lo_hi_readq(to_nvme_dev(ctrl)->bar + off); |
7fd8930f | 2664 | return 0; |
4cc06521 KB |
2665 | } |
2666 | ||
97c12223 KB |
2667 | static int nvme_pci_get_address(struct nvme_ctrl *ctrl, char *buf, int size) |
2668 | { | |
2669 | struct pci_dev *pdev = to_pci_dev(to_nvme_dev(ctrl)->dev); | |
2670 | ||
2db24e4a | 2671 | return snprintf(buf, size, "%s\n", dev_name(&pdev->dev)); |
97c12223 KB |
2672 | } |
2673 | ||
1c63dc66 | 2674 | static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = { |
1a353d85 | 2675 | .name = "pcie", |
e439bb12 | 2676 | .module = THIS_MODULE, |
e0596ab2 LG |
2677 | .flags = NVME_F_METADATA_SUPPORTED | |
2678 | NVME_F_PCI_P2PDMA, | |
1c63dc66 | 2679 | .reg_read32 = nvme_pci_reg_read32, |
5fd4ce1b | 2680 | .reg_write32 = nvme_pci_reg_write32, |
7fd8930f | 2681 | .reg_read64 = nvme_pci_reg_read64, |
1673f1f0 | 2682 | .free_ctrl = nvme_pci_free_ctrl, |
f866fc42 | 2683 | .submit_async_event = nvme_pci_submit_async_event, |
97c12223 | 2684 | .get_address = nvme_pci_get_address, |
1c63dc66 | 2685 | }; |
4cc06521 | 2686 | |
b00a726a KB |
2687 | static int nvme_dev_map(struct nvme_dev *dev) |
2688 | { | |
b00a726a KB |
2689 | struct pci_dev *pdev = to_pci_dev(dev->dev); |
2690 | ||
a1f447b3 | 2691 | if (pci_request_mem_regions(pdev, "nvme")) |
b00a726a KB |
2692 | return -ENODEV; |
2693 | ||
97f6ef64 | 2694 | if (nvme_remap_bar(dev, NVME_REG_DBS + 4096)) |
b00a726a KB |
2695 | goto release; |
2696 | ||
9fa196e7 | 2697 | return 0; |
b00a726a | 2698 | release: |
9fa196e7 MG |
2699 | pci_release_mem_regions(pdev); |
2700 | return -ENODEV; | |
b00a726a KB |
2701 | } |
2702 | ||
8427bbc2 | 2703 | static unsigned long check_vendor_combination_bug(struct pci_dev *pdev) |
ff5350a8 AL |
2704 | { |
2705 | if (pdev->vendor == 0x144d && pdev->device == 0xa802) { | |
2706 | /* | |
2707 | * Several Samsung devices seem to drop off the PCIe bus | |
2708 | * randomly when APST is on and uses the deepest sleep state. | |
2709 | * This has been observed on a Samsung "SM951 NVMe SAMSUNG | |
2710 | * 256GB", a "PM951 NVMe SAMSUNG 512GB", and a "Samsung SSD | |
2711 | * 950 PRO 256GB", but it seems to be restricted to two Dell | |
2712 | * laptops. | |
2713 | */ | |
2714 | if (dmi_match(DMI_SYS_VENDOR, "Dell Inc.") && | |
2715 | (dmi_match(DMI_PRODUCT_NAME, "XPS 15 9550") || | |
2716 | dmi_match(DMI_PRODUCT_NAME, "Precision 5510"))) | |
2717 | return NVME_QUIRK_NO_DEEPEST_PS; | |
8427bbc2 KHF |
2718 | } else if (pdev->vendor == 0x144d && pdev->device == 0xa804) { |
2719 | /* | |
2720 | * Samsung SSD 960 EVO drops off the PCIe bus after system | |
467c77d4 JJ |
2721 | * suspend on a Ryzen board, ASUS PRIME B350M-A, as well as |
2722 | * within few minutes after bootup on a Coffee Lake board - | |
2723 | * ASUS PRIME Z370-A | |
8427bbc2 KHF |
2724 | */ |
2725 | if (dmi_match(DMI_BOARD_VENDOR, "ASUSTeK COMPUTER INC.") && | |
467c77d4 JJ |
2726 | (dmi_match(DMI_BOARD_NAME, "PRIME B350M-A") || |
2727 | dmi_match(DMI_BOARD_NAME, "PRIME Z370-A"))) | |
8427bbc2 | 2728 | return NVME_QUIRK_NO_APST; |
1fae37ac S |
2729 | } else if ((pdev->vendor == 0x144d && (pdev->device == 0xa801 || |
2730 | pdev->device == 0xa808 || pdev->device == 0xa809)) || | |
2731 | (pdev->vendor == 0x1e0f && pdev->device == 0x0001)) { | |
2732 | /* | |
2733 | * Forcing to use host managed nvme power settings for | |
2734 | * lowest idle power with quick resume latency on | |
2735 | * Samsung and Toshiba SSDs based on suspend behavior | |
2736 | * on Coffee Lake board for LENOVO C640 | |
2737 | */ | |
2738 | if ((dmi_match(DMI_BOARD_VENDOR, "LENOVO")) && | |
2739 | dmi_match(DMI_BOARD_NAME, "LNVNB161216")) | |
2740 | return NVME_QUIRK_SIMPLE_SUSPEND; | |
ff5350a8 AL |
2741 | } |
2742 | ||
2743 | return 0; | |
2744 | } | |
2745 | ||
18119775 KB |
2746 | static void nvme_async_probe(void *data, async_cookie_t cookie) |
2747 | { | |
2748 | struct nvme_dev *dev = data; | |
80f513b5 | 2749 | |
bd46a906 | 2750 | flush_work(&dev->ctrl.reset_work); |
18119775 | 2751 | flush_work(&dev->ctrl.scan_work); |
80f513b5 | 2752 | nvme_put_ctrl(&dev->ctrl); |
18119775 KB |
2753 | } |
2754 | ||
8d85fce7 | 2755 | static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
b60503ba | 2756 | { |
a4aea562 | 2757 | int node, result = -ENOMEM; |
b60503ba | 2758 | struct nvme_dev *dev; |
ff5350a8 | 2759 | unsigned long quirks = id->driver_data; |
943e942e | 2760 | size_t alloc_size; |
b60503ba | 2761 | |
a4aea562 MB |
2762 | node = dev_to_node(&pdev->dev); |
2763 | if (node == NUMA_NO_NODE) | |
2fa84351 | 2764 | set_dev_node(&pdev->dev, first_memory_node); |
a4aea562 MB |
2765 | |
2766 | dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); | |
b60503ba MW |
2767 | if (!dev) |
2768 | return -ENOMEM; | |
147b27e4 | 2769 | |
3b6592f7 JA |
2770 | dev->queues = kcalloc_node(max_queue_count(), sizeof(struct nvme_queue), |
2771 | GFP_KERNEL, node); | |
b60503ba MW |
2772 | if (!dev->queues) |
2773 | goto free; | |
2774 | ||
e75ec752 | 2775 | dev->dev = get_device(&pdev->dev); |
9a6b9458 | 2776 | pci_set_drvdata(pdev, dev); |
1c63dc66 | 2777 | |
b00a726a KB |
2778 | result = nvme_dev_map(dev); |
2779 | if (result) | |
b00c9b7a | 2780 | goto put_pci; |
b00a726a | 2781 | |
d86c4d8e | 2782 | INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work); |
5c8809e6 | 2783 | INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work); |
77bf25ea | 2784 | mutex_init(&dev->shutdown_lock); |
b60503ba | 2785 | |
091b6092 MW |
2786 | result = nvme_setup_prp_pools(dev); |
2787 | if (result) | |
b00c9b7a | 2788 | goto unmap; |
4cc06521 | 2789 | |
8427bbc2 | 2790 | quirks |= check_vendor_combination_bug(pdev); |
ff5350a8 | 2791 | |
943e942e JA |
2792 | /* |
2793 | * Double check that our mempool alloc size will cover the biggest | |
2794 | * command we support. | |
2795 | */ | |
2796 | alloc_size = nvme_pci_iod_alloc_size(dev, NVME_MAX_KB_SZ, | |
2797 | NVME_MAX_SEGS, true); | |
2798 | WARN_ON_ONCE(alloc_size > PAGE_SIZE); | |
2799 | ||
2800 | dev->iod_mempool = mempool_create_node(1, mempool_kmalloc, | |
2801 | mempool_kfree, | |
2802 | (void *) alloc_size, | |
2803 | GFP_KERNEL, node); | |
2804 | if (!dev->iod_mempool) { | |
2805 | result = -ENOMEM; | |
2806 | goto release_pools; | |
2807 | } | |
2808 | ||
b6e44b4c KB |
2809 | result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops, |
2810 | quirks); | |
2811 | if (result) | |
2812 | goto release_mempool; | |
2813 | ||
1b3c47c1 SG |
2814 | dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev)); |
2815 | ||
bd46a906 | 2816 | nvme_reset_ctrl(&dev->ctrl); |
18119775 | 2817 | async_schedule(nvme_async_probe, dev); |
4caff8fc | 2818 | |
b60503ba MW |
2819 | return 0; |
2820 | ||
b6e44b4c KB |
2821 | release_mempool: |
2822 | mempool_destroy(dev->iod_mempool); | |
0877cb0d | 2823 | release_pools: |
091b6092 | 2824 | nvme_release_prp_pools(dev); |
b00c9b7a CJ |
2825 | unmap: |
2826 | nvme_dev_unmap(dev); | |
a96d4f5c | 2827 | put_pci: |
e75ec752 | 2828 | put_device(dev->dev); |
b60503ba MW |
2829 | free: |
2830 | kfree(dev->queues); | |
b60503ba MW |
2831 | kfree(dev); |
2832 | return result; | |
2833 | } | |
2834 | ||
775755ed | 2835 | static void nvme_reset_prepare(struct pci_dev *pdev) |
f0d54a54 | 2836 | { |
a6739479 | 2837 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
c1ac9a4b KB |
2838 | |
2839 | /* | |
2840 | * We don't need to check the return value from waiting for the reset | |
2841 | * state as pci_dev device lock is held, making it impossible to race | |
2842 | * with ->remove(). | |
2843 | */ | |
2844 | nvme_disable_prepare_reset(dev, false); | |
2845 | nvme_sync_queues(&dev->ctrl); | |
775755ed | 2846 | } |
f0d54a54 | 2847 | |
775755ed CH |
2848 | static void nvme_reset_done(struct pci_dev *pdev) |
2849 | { | |
f263fbb8 | 2850 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
c1ac9a4b KB |
2851 | |
2852 | if (!nvme_try_sched_reset(&dev->ctrl)) | |
2853 | flush_work(&dev->ctrl.reset_work); | |
f0d54a54 KB |
2854 | } |
2855 | ||
09ece142 KB |
2856 | static void nvme_shutdown(struct pci_dev *pdev) |
2857 | { | |
2858 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
c1ac9a4b | 2859 | nvme_disable_prepare_reset(dev, true); |
09ece142 KB |
2860 | } |
2861 | ||
f58944e2 KB |
2862 | /* |
2863 | * The driver's remove may be called on a device in a partially initialized | |
2864 | * state. This function must not have any dependencies on the device state in | |
2865 | * order to proceed. | |
2866 | */ | |
8d85fce7 | 2867 | static void nvme_remove(struct pci_dev *pdev) |
b60503ba MW |
2868 | { |
2869 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
9a6b9458 | 2870 | |
bb8d261e | 2871 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); |
9a6b9458 | 2872 | pci_set_drvdata(pdev, NULL); |
0ff9d4e1 | 2873 | |
6db28eda | 2874 | if (!pci_device_is_present(pdev)) { |
0ff9d4e1 | 2875 | nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD); |
1d39e692 | 2876 | nvme_dev_disable(dev, true); |
cb4bfda6 | 2877 | nvme_dev_remove_admin(dev); |
6db28eda | 2878 | } |
0ff9d4e1 | 2879 | |
d86c4d8e | 2880 | flush_work(&dev->ctrl.reset_work); |
d09f2b45 SG |
2881 | nvme_stop_ctrl(&dev->ctrl); |
2882 | nvme_remove_namespaces(&dev->ctrl); | |
a5cdb68c | 2883 | nvme_dev_disable(dev, true); |
9fe5c59f | 2884 | nvme_release_cmb(dev); |
87ad72a5 | 2885 | nvme_free_host_mem(dev); |
a4aea562 | 2886 | nvme_dev_remove_admin(dev); |
a1a5ef99 | 2887 | nvme_free_queues(dev, 0); |
9a6b9458 | 2888 | nvme_release_prp_pools(dev); |
b00a726a | 2889 | nvme_dev_unmap(dev); |
726612b6 | 2890 | nvme_uninit_ctrl(&dev->ctrl); |
b60503ba MW |
2891 | } |
2892 | ||
671a6018 | 2893 | #ifdef CONFIG_PM_SLEEP |
d916b1be KB |
2894 | static int nvme_get_power_state(struct nvme_ctrl *ctrl, u32 *ps) |
2895 | { | |
2896 | return nvme_get_features(ctrl, NVME_FEAT_POWER_MGMT, 0, NULL, 0, ps); | |
2897 | } | |
2898 | ||
2899 | static int nvme_set_power_state(struct nvme_ctrl *ctrl, u32 ps) | |
2900 | { | |
2901 | return nvme_set_features(ctrl, NVME_FEAT_POWER_MGMT, ps, NULL, 0, NULL); | |
2902 | } | |
2903 | ||
2904 | static int nvme_resume(struct device *dev) | |
2905 | { | |
2906 | struct nvme_dev *ndev = pci_get_drvdata(to_pci_dev(dev)); | |
2907 | struct nvme_ctrl *ctrl = &ndev->ctrl; | |
2908 | ||
4eaefe8c | 2909 | if (ndev->last_ps == U32_MAX || |
d916b1be | 2910 | nvme_set_power_state(ctrl, ndev->last_ps) != 0) |
c1ac9a4b | 2911 | return nvme_try_sched_reset(&ndev->ctrl); |
d916b1be KB |
2912 | return 0; |
2913 | } | |
2914 | ||
cd638946 KB |
2915 | static int nvme_suspend(struct device *dev) |
2916 | { | |
2917 | struct pci_dev *pdev = to_pci_dev(dev); | |
2918 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
d916b1be KB |
2919 | struct nvme_ctrl *ctrl = &ndev->ctrl; |
2920 | int ret = -EBUSY; | |
2921 | ||
4eaefe8c RW |
2922 | ndev->last_ps = U32_MAX; |
2923 | ||
d916b1be KB |
2924 | /* |
2925 | * The platform does not remove power for a kernel managed suspend so | |
2926 | * use host managed nvme power settings for lowest idle power if | |
2927 | * possible. This should have quicker resume latency than a full device | |
2928 | * shutdown. But if the firmware is involved after the suspend or the | |
2929 | * device does not support any non-default power states, shut down the | |
2930 | * device fully. | |
4eaefe8c RW |
2931 | * |
2932 | * If ASPM is not enabled for the device, shut down the device and allow | |
2933 | * the PCI bus layer to put it into D3 in order to take the PCIe link | |
2934 | * down, so as to allow the platform to achieve its minimum low-power | |
2935 | * state (which may not be possible if the link is up). | |
d916b1be | 2936 | */ |
4eaefe8c | 2937 | if (pm_suspend_via_firmware() || !ctrl->npss || |
cb32de1b | 2938 | !pcie_aspm_enabled(pdev) || |
c1ac9a4b KB |
2939 | (ndev->ctrl.quirks & NVME_QUIRK_SIMPLE_SUSPEND)) |
2940 | return nvme_disable_prepare_reset(ndev, true); | |
d916b1be KB |
2941 | |
2942 | nvme_start_freeze(ctrl); | |
2943 | nvme_wait_freeze(ctrl); | |
2944 | nvme_sync_queues(ctrl); | |
2945 | ||
5d02a5c1 | 2946 | if (ctrl->state != NVME_CTRL_LIVE) |
d916b1be KB |
2947 | goto unfreeze; |
2948 | ||
d916b1be KB |
2949 | ret = nvme_get_power_state(ctrl, &ndev->last_ps); |
2950 | if (ret < 0) | |
2951 | goto unfreeze; | |
2952 | ||
7cbb5c6f ML |
2953 | /* |
2954 | * A saved state prevents pci pm from generically controlling the | |
2955 | * device's power. If we're using protocol specific settings, we don't | |
2956 | * want pci interfering. | |
2957 | */ | |
2958 | pci_save_state(pdev); | |
2959 | ||
d916b1be KB |
2960 | ret = nvme_set_power_state(ctrl, ctrl->npss); |
2961 | if (ret < 0) | |
2962 | goto unfreeze; | |
2963 | ||
2964 | if (ret) { | |
7cbb5c6f ML |
2965 | /* discard the saved state */ |
2966 | pci_load_saved_state(pdev, NULL); | |
2967 | ||
d916b1be KB |
2968 | /* |
2969 | * Clearing npss forces a controller reset on resume. The | |
05d3046f | 2970 | * correct value will be rediscovered then. |
d916b1be | 2971 | */ |
c1ac9a4b | 2972 | ret = nvme_disable_prepare_reset(ndev, true); |
d916b1be | 2973 | ctrl->npss = 0; |
d916b1be | 2974 | } |
d916b1be KB |
2975 | unfreeze: |
2976 | nvme_unfreeze(ctrl); | |
2977 | return ret; | |
2978 | } | |
2979 | ||
2980 | static int nvme_simple_suspend(struct device *dev) | |
2981 | { | |
2982 | struct nvme_dev *ndev = pci_get_drvdata(to_pci_dev(dev)); | |
c1ac9a4b | 2983 | return nvme_disable_prepare_reset(ndev, true); |
cd638946 KB |
2984 | } |
2985 | ||
d916b1be | 2986 | static int nvme_simple_resume(struct device *dev) |
cd638946 KB |
2987 | { |
2988 | struct pci_dev *pdev = to_pci_dev(dev); | |
2989 | struct nvme_dev *ndev = pci_get_drvdata(pdev); | |
cd638946 | 2990 | |
c1ac9a4b | 2991 | return nvme_try_sched_reset(&ndev->ctrl); |
cd638946 KB |
2992 | } |
2993 | ||
21774222 | 2994 | static const struct dev_pm_ops nvme_dev_pm_ops = { |
d916b1be KB |
2995 | .suspend = nvme_suspend, |
2996 | .resume = nvme_resume, | |
2997 | .freeze = nvme_simple_suspend, | |
2998 | .thaw = nvme_simple_resume, | |
2999 | .poweroff = nvme_simple_suspend, | |
3000 | .restore = nvme_simple_resume, | |
3001 | }; | |
3002 | #endif /* CONFIG_PM_SLEEP */ | |
b60503ba | 3003 | |
a0a3408e KB |
3004 | static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev, |
3005 | pci_channel_state_t state) | |
3006 | { | |
3007 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
3008 | ||
3009 | /* | |
3010 | * A frozen channel requires a reset. When detected, this method will | |
3011 | * shutdown the controller to quiesce. The controller will be restarted | |
3012 | * after the slot reset through driver's slot_reset callback. | |
3013 | */ | |
a0a3408e KB |
3014 | switch (state) { |
3015 | case pci_channel_io_normal: | |
3016 | return PCI_ERS_RESULT_CAN_RECOVER; | |
3017 | case pci_channel_io_frozen: | |
d011fb31 KB |
3018 | dev_warn(dev->ctrl.device, |
3019 | "frozen state error detected, reset controller\n"); | |
a5cdb68c | 3020 | nvme_dev_disable(dev, false); |
a0a3408e KB |
3021 | return PCI_ERS_RESULT_NEED_RESET; |
3022 | case pci_channel_io_perm_failure: | |
d011fb31 KB |
3023 | dev_warn(dev->ctrl.device, |
3024 | "failure state error detected, request disconnect\n"); | |
a0a3408e KB |
3025 | return PCI_ERS_RESULT_DISCONNECT; |
3026 | } | |
3027 | return PCI_ERS_RESULT_NEED_RESET; | |
3028 | } | |
3029 | ||
3030 | static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev) | |
3031 | { | |
3032 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
3033 | ||
1b3c47c1 | 3034 | dev_info(dev->ctrl.device, "restart after slot reset\n"); |
a0a3408e | 3035 | pci_restore_state(pdev); |
d86c4d8e | 3036 | nvme_reset_ctrl(&dev->ctrl); |
a0a3408e KB |
3037 | return PCI_ERS_RESULT_RECOVERED; |
3038 | } | |
3039 | ||
3040 | static void nvme_error_resume(struct pci_dev *pdev) | |
3041 | { | |
72cd4cc2 KB |
3042 | struct nvme_dev *dev = pci_get_drvdata(pdev); |
3043 | ||
3044 | flush_work(&dev->ctrl.reset_work); | |
a0a3408e KB |
3045 | } |
3046 | ||
1d352035 | 3047 | static const struct pci_error_handlers nvme_err_handler = { |
b60503ba | 3048 | .error_detected = nvme_error_detected, |
b60503ba MW |
3049 | .slot_reset = nvme_slot_reset, |
3050 | .resume = nvme_error_resume, | |
775755ed CH |
3051 | .reset_prepare = nvme_reset_prepare, |
3052 | .reset_done = nvme_reset_done, | |
b60503ba MW |
3053 | }; |
3054 | ||
6eb0d698 | 3055 | static const struct pci_device_id nvme_id_table[] = { |
106198ed | 3056 | { PCI_VDEVICE(INTEL, 0x0953), |
08095e70 | 3057 | .driver_data = NVME_QUIRK_STRIPE_SIZE | |
e850fd16 | 3058 | NVME_QUIRK_DEALLOCATE_ZEROES, }, |
99466e70 KB |
3059 | { PCI_VDEVICE(INTEL, 0x0a53), |
3060 | .driver_data = NVME_QUIRK_STRIPE_SIZE | | |
e850fd16 | 3061 | NVME_QUIRK_DEALLOCATE_ZEROES, }, |
99466e70 KB |
3062 | { PCI_VDEVICE(INTEL, 0x0a54), |
3063 | .driver_data = NVME_QUIRK_STRIPE_SIZE | | |
e850fd16 | 3064 | NVME_QUIRK_DEALLOCATE_ZEROES, }, |
f99cb7af DWF |
3065 | { PCI_VDEVICE(INTEL, 0x0a55), |
3066 | .driver_data = NVME_QUIRK_STRIPE_SIZE | | |
3067 | NVME_QUIRK_DEALLOCATE_ZEROES, }, | |
50af47d0 | 3068 | { PCI_VDEVICE(INTEL, 0xf1a5), /* Intel 600P/P3100 */ |
9abd68ef | 3069 | .driver_data = NVME_QUIRK_NO_DEEPEST_PS | |
6c6aa2f2 AM |
3070 | NVME_QUIRK_MEDIUM_PRIO_SQ | |
3071 | NVME_QUIRK_NO_TEMP_THRESH_CHANGE }, | |
6299358d JD |
3072 | { PCI_VDEVICE(INTEL, 0xf1a6), /* Intel 760p/Pro 7600p */ |
3073 | .driver_data = NVME_QUIRK_IGNORE_DEV_SUBNQN, }, | |
540c801c | 3074 | { PCI_VDEVICE(INTEL, 0x5845), /* Qemu emulated controller */ |
7b210e4e CH |
3075 | .driver_data = NVME_QUIRK_IDENTIFY_CNS | |
3076 | NVME_QUIRK_DISABLE_WRITE_ZEROES, }, | |
0302ae60 MP |
3077 | { PCI_DEVICE(0x1bb1, 0x0100), /* Seagate Nytro Flash Storage */ |
3078 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
54adc010 GP |
3079 | { PCI_DEVICE(0x1c58, 0x0003), /* HGST adapter */ |
3080 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
8c97eecc JL |
3081 | { PCI_DEVICE(0x1c58, 0x0023), /* WDC SN200 adapter */ |
3082 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
015282c9 WW |
3083 | { PCI_DEVICE(0x1c5f, 0x0540), /* Memblaze Pblaze4 adapter */ |
3084 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
d554b5e1 MP |
3085 | { PCI_DEVICE(0x144d, 0xa821), /* Samsung PM1725 */ |
3086 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
3087 | { PCI_DEVICE(0x144d, 0xa822), /* Samsung PM1725a */ | |
3088 | .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, | |
608cc4b1 CH |
3089 | { PCI_DEVICE(0x1d1d, 0x1f1f), /* LighNVM qemu device */ |
3090 | .driver_data = NVME_QUIRK_LIGHTNVM, }, | |
3091 | { PCI_DEVICE(0x1d1d, 0x2807), /* CNEX WL */ | |
3092 | .driver_data = NVME_QUIRK_LIGHTNVM, }, | |
ea48e877 WX |
3093 | { PCI_DEVICE(0x1d1d, 0x2601), /* CNEX Granby */ |
3094 | .driver_data = NVME_QUIRK_LIGHTNVM, }, | |
08b903b5 MN |
3095 | { PCI_DEVICE(0x10ec, 0x5762), /* ADATA SX6000LNP */ |
3096 | .driver_data = NVME_QUIRK_IGNORE_DEV_SUBNQN, }, | |
f03e42c6 GC |
3097 | { PCI_DEVICE(0x1cc1, 0x8201), /* ADATA SX8200PNP 512GB */ |
3098 | .driver_data = NVME_QUIRK_NO_DEEPEST_PS | | |
3099 | NVME_QUIRK_IGNORE_DEV_SUBNQN, }, | |
b60503ba | 3100 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, |
98f7b86a AS |
3101 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001), |
3102 | .driver_data = NVME_QUIRK_SINGLE_VECTOR }, | |
124298bd | 3103 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) }, |
66341331 BH |
3104 | { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2005), |
3105 | .driver_data = NVME_QUIRK_SINGLE_VECTOR | | |
d38e9f04 BH |
3106 | NVME_QUIRK_128_BYTES_SQES | |
3107 | NVME_QUIRK_SHARED_TAGS }, | |
b60503ba MW |
3108 | { 0, } |
3109 | }; | |
3110 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
3111 | ||
3112 | static struct pci_driver nvme_driver = { | |
3113 | .name = "nvme", | |
3114 | .id_table = nvme_id_table, | |
3115 | .probe = nvme_probe, | |
8d85fce7 | 3116 | .remove = nvme_remove, |
09ece142 | 3117 | .shutdown = nvme_shutdown, |
d916b1be | 3118 | #ifdef CONFIG_PM_SLEEP |
cd638946 KB |
3119 | .driver = { |
3120 | .pm = &nvme_dev_pm_ops, | |
3121 | }, | |
d916b1be | 3122 | #endif |
74d986ab | 3123 | .sriov_configure = pci_sriov_configure_simple, |
b60503ba MW |
3124 | .err_handler = &nvme_err_handler, |
3125 | }; | |
3126 | ||
3127 | static int __init nvme_init(void) | |
3128 | { | |
81101540 CH |
3129 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); |
3130 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
3131 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
612b7286 | 3132 | BUILD_BUG_ON(IRQ_AFFINITY_MAX_SETS < 2); |
17c33167 KB |
3133 | |
3134 | write_queues = min(write_queues, num_possible_cpus()); | |
3135 | poll_queues = min(poll_queues, num_possible_cpus()); | |
9a6327d2 | 3136 | return pci_register_driver(&nvme_driver); |
b60503ba MW |
3137 | } |
3138 | ||
3139 | static void __exit nvme_exit(void) | |
3140 | { | |
3141 | pci_unregister_driver(&nvme_driver); | |
03e0f3a6 | 3142 | flush_workqueue(nvme_wq); |
b60503ba MW |
3143 | } |
3144 | ||
3145 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
3146 | MODULE_LICENSE("GPL"); | |
c78b4713 | 3147 | MODULE_VERSION("1.0"); |
b60503ba MW |
3148 | module_init(nvme_init); |
3149 | module_exit(nvme_exit); |