Commit | Line | Data |
---|---|---|
b60503ba MW |
1 | /* |
2 | * NVM Express device driver | |
3 | * Copyright (c) 2011, Intel Corporation. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License along with | |
15 | * this program; if not, write to the Free Software Foundation, Inc., | |
16 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
17 | */ | |
18 | ||
19 | #include <linux/nvme.h> | |
20 | #include <linux/bio.h> | |
8de05535 | 21 | #include <linux/bitops.h> |
b60503ba | 22 | #include <linux/blkdev.h> |
fd63e9ce | 23 | #include <linux/delay.h> |
b60503ba MW |
24 | #include <linux/errno.h> |
25 | #include <linux/fs.h> | |
26 | #include <linux/genhd.h> | |
5aff9382 | 27 | #include <linux/idr.h> |
b60503ba MW |
28 | #include <linux/init.h> |
29 | #include <linux/interrupt.h> | |
30 | #include <linux/io.h> | |
31 | #include <linux/kdev_t.h> | |
1fa6aead | 32 | #include <linux/kthread.h> |
b60503ba MW |
33 | #include <linux/kernel.h> |
34 | #include <linux/mm.h> | |
35 | #include <linux/module.h> | |
36 | #include <linux/moduleparam.h> | |
37 | #include <linux/pci.h> | |
be7b6275 | 38 | #include <linux/poison.h> |
b60503ba MW |
39 | #include <linux/sched.h> |
40 | #include <linux/slab.h> | |
41 | #include <linux/types.h> | |
42 | #include <linux/version.h> | |
43 | ||
44 | #define NVME_Q_DEPTH 1024 | |
45 | #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) | |
46 | #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) | |
47 | #define NVME_MINORS 64 | |
e85248e5 MW |
48 | #define IO_TIMEOUT (5 * HZ) |
49 | #define ADMIN_TIMEOUT (60 * HZ) | |
b60503ba MW |
50 | |
51 | static int nvme_major; | |
52 | module_param(nvme_major, int, 0); | |
53 | ||
58ffacb5 MW |
54 | static int use_threaded_interrupts; |
55 | module_param(use_threaded_interrupts, int, 0); | |
56 | ||
1fa6aead MW |
57 | static DEFINE_SPINLOCK(dev_list_lock); |
58 | static LIST_HEAD(dev_list); | |
59 | static struct task_struct *nvme_thread; | |
60 | ||
b60503ba MW |
61 | /* |
62 | * Represents an NVM Express device. Each nvme_dev is a PCI function. | |
63 | */ | |
64 | struct nvme_dev { | |
1fa6aead | 65 | struct list_head node; |
b60503ba MW |
66 | struct nvme_queue **queues; |
67 | u32 __iomem *dbs; | |
68 | struct pci_dev *pci_dev; | |
091b6092 | 69 | struct dma_pool *prp_page_pool; |
99802a7a | 70 | struct dma_pool *prp_small_pool; |
b60503ba MW |
71 | int instance; |
72 | int queue_count; | |
f1938f6e | 73 | int db_stride; |
b60503ba MW |
74 | u32 ctrl_config; |
75 | struct msix_entry *entry; | |
76 | struct nvme_bar __iomem *bar; | |
77 | struct list_head namespaces; | |
51814232 MW |
78 | char serial[20]; |
79 | char model[40]; | |
80 | char firmware_rev[8]; | |
b60503ba MW |
81 | }; |
82 | ||
83 | /* | |
84 | * An NVM Express namespace is equivalent to a SCSI LUN | |
85 | */ | |
86 | struct nvme_ns { | |
87 | struct list_head list; | |
88 | ||
89 | struct nvme_dev *dev; | |
90 | struct request_queue *queue; | |
91 | struct gendisk *disk; | |
92 | ||
93 | int ns_id; | |
94 | int lba_shift; | |
95 | }; | |
96 | ||
97 | /* | |
98 | * An NVM Express queue. Each device has at least two (one for admin | |
99 | * commands and one for I/O commands). | |
100 | */ | |
101 | struct nvme_queue { | |
102 | struct device *q_dmadev; | |
091b6092 | 103 | struct nvme_dev *dev; |
b60503ba MW |
104 | spinlock_t q_lock; |
105 | struct nvme_command *sq_cmds; | |
106 | volatile struct nvme_completion *cqes; | |
107 | dma_addr_t sq_dma_addr; | |
108 | dma_addr_t cq_dma_addr; | |
109 | wait_queue_head_t sq_full; | |
1fa6aead | 110 | wait_queue_t sq_cong_wait; |
b60503ba MW |
111 | struct bio_list sq_cong; |
112 | u32 __iomem *q_db; | |
113 | u16 q_depth; | |
114 | u16 cq_vector; | |
115 | u16 sq_head; | |
116 | u16 sq_tail; | |
117 | u16 cq_head; | |
82123460 | 118 | u16 cq_phase; |
b60503ba MW |
119 | unsigned long cmdid_data[]; |
120 | }; | |
121 | ||
122 | /* | |
123 | * Check we didin't inadvertently grow the command struct | |
124 | */ | |
125 | static inline void _nvme_check_size(void) | |
126 | { | |
127 | BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); | |
128 | BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); | |
129 | BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); | |
130 | BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); | |
131 | BUILD_BUG_ON(sizeof(struct nvme_features) != 64); | |
132 | BUILD_BUG_ON(sizeof(struct nvme_command) != 64); | |
133 | BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); | |
134 | BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); | |
135 | BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); | |
136 | } | |
137 | ||
e85248e5 MW |
138 | struct nvme_cmd_info { |
139 | unsigned long ctx; | |
140 | unsigned long timeout; | |
141 | }; | |
142 | ||
143 | static struct nvme_cmd_info *nvme_cmd_info(struct nvme_queue *nvmeq) | |
144 | { | |
145 | return (void *)&nvmeq->cmdid_data[BITS_TO_LONGS(nvmeq->q_depth)]; | |
146 | } | |
147 | ||
b60503ba | 148 | /** |
714a7a22 MW |
149 | * alloc_cmdid() - Allocate a Command ID |
150 | * @nvmeq: The queue that will be used for this command | |
151 | * @ctx: A pointer that will be passed to the handler | |
152 | * @handler: The ID of the handler to call | |
b60503ba MW |
153 | * |
154 | * Allocate a Command ID for a queue. The data passed in will | |
155 | * be passed to the completion handler. This is implemented by using | |
156 | * the bottom two bits of the ctx pointer to store the handler ID. | |
157 | * Passing in a pointer that's not 4-byte aligned will cause a BUG. | |
158 | * We can change this if it becomes a problem. | |
184d2944 MW |
159 | * |
160 | * May be called with local interrupts disabled and the q_lock held, | |
161 | * or with interrupts enabled and no locks held. | |
b60503ba | 162 | */ |
e85248e5 MW |
163 | static int alloc_cmdid(struct nvme_queue *nvmeq, void *ctx, int handler, |
164 | unsigned timeout) | |
b60503ba | 165 | { |
e6d15f79 | 166 | int depth = nvmeq->q_depth - 1; |
e85248e5 | 167 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); |
b60503ba MW |
168 | int cmdid; |
169 | ||
170 | BUG_ON((unsigned long)ctx & 3); | |
171 | ||
172 | do { | |
173 | cmdid = find_first_zero_bit(nvmeq->cmdid_data, depth); | |
174 | if (cmdid >= depth) | |
175 | return -EBUSY; | |
176 | } while (test_and_set_bit(cmdid, nvmeq->cmdid_data)); | |
177 | ||
e85248e5 MW |
178 | info[cmdid].ctx = (unsigned long)ctx | handler; |
179 | info[cmdid].timeout = jiffies + timeout; | |
b60503ba MW |
180 | return cmdid; |
181 | } | |
182 | ||
183 | static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx, | |
e85248e5 | 184 | int handler, unsigned timeout) |
b60503ba MW |
185 | { |
186 | int cmdid; | |
187 | wait_event_killable(nvmeq->sq_full, | |
e85248e5 | 188 | (cmdid = alloc_cmdid(nvmeq, ctx, handler, timeout)) >= 0); |
b60503ba MW |
189 | return (cmdid < 0) ? -EINTR : cmdid; |
190 | } | |
191 | ||
fa922821 MW |
192 | /* |
193 | * If you need more than four handlers, you'll need to change how | |
be7b6275 MW |
194 | * alloc_cmdid and nvme_process_cq work. Consider using a special |
195 | * CMD_CTX value instead, if that works for your situation. | |
b60503ba MW |
196 | */ |
197 | enum { | |
198 | sync_completion_id = 0, | |
199 | bio_completion_id, | |
200 | }; | |
201 | ||
00df5cb4 | 202 | /* Special values must be a multiple of 4, and less than 0x1000 */ |
be7b6275 | 203 | #define CMD_CTX_BASE (POISON_POINTER_DELTA + sync_completion_id) |
d2d87034 MW |
204 | #define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE) |
205 | #define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE) | |
206 | #define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE) | |
00df5cb4 | 207 | #define CMD_CTX_FLUSH (0x318 + CMD_CTX_BASE) |
be7b6275 | 208 | |
184d2944 MW |
209 | /* |
210 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
211 | */ | |
b60503ba MW |
212 | static unsigned long free_cmdid(struct nvme_queue *nvmeq, int cmdid) |
213 | { | |
214 | unsigned long data; | |
e85248e5 | 215 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); |
b60503ba | 216 | |
e85248e5 | 217 | if (cmdid >= nvmeq->q_depth) |
48e3d398 | 218 | return CMD_CTX_INVALID; |
e85248e5 MW |
219 | data = info[cmdid].ctx; |
220 | info[cmdid].ctx = CMD_CTX_COMPLETED; | |
b60503ba MW |
221 | clear_bit(cmdid, nvmeq->cmdid_data); |
222 | wake_up(&nvmeq->sq_full); | |
223 | return data; | |
224 | } | |
225 | ||
21075bde | 226 | static unsigned long cancel_cmdid(struct nvme_queue *nvmeq, int cmdid) |
3c0cf138 | 227 | { |
21075bde | 228 | unsigned long data; |
e85248e5 | 229 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); |
21075bde | 230 | data = info[cmdid].ctx; |
e85248e5 | 231 | info[cmdid].ctx = CMD_CTX_CANCELLED; |
21075bde | 232 | return data; |
3c0cf138 MW |
233 | } |
234 | ||
b60503ba MW |
235 | static struct nvme_queue *get_nvmeq(struct nvme_ns *ns) |
236 | { | |
9ecdc946 | 237 | return ns->dev->queues[get_cpu() + 1]; |
b60503ba MW |
238 | } |
239 | ||
240 | static void put_nvmeq(struct nvme_queue *nvmeq) | |
241 | { | |
1b23484b | 242 | put_cpu(); |
b60503ba MW |
243 | } |
244 | ||
245 | /** | |
714a7a22 | 246 | * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell |
b60503ba MW |
247 | * @nvmeq: The queue to use |
248 | * @cmd: The command to send | |
249 | * | |
250 | * Safe to use from interrupt context | |
251 | */ | |
252 | static int nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd) | |
253 | { | |
254 | unsigned long flags; | |
255 | u16 tail; | |
b60503ba MW |
256 | spin_lock_irqsave(&nvmeq->q_lock, flags); |
257 | tail = nvmeq->sq_tail; | |
258 | memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); | |
b60503ba MW |
259 | if (++tail == nvmeq->q_depth) |
260 | tail = 0; | |
7547881d | 261 | writel(tail, nvmeq->q_db); |
b60503ba MW |
262 | nvmeq->sq_tail = tail; |
263 | spin_unlock_irqrestore(&nvmeq->q_lock, flags); | |
264 | ||
265 | return 0; | |
266 | } | |
267 | ||
e025344c | 268 | struct nvme_prps { |
0d1bc912 | 269 | int npages; /* 0 means small pool in use */ |
e025344c SMM |
270 | dma_addr_t first_dma; |
271 | __le64 *list[0]; | |
272 | }; | |
273 | ||
d567760c | 274 | static void nvme_free_prps(struct nvme_dev *dev, struct nvme_prps *prps) |
e025344c SMM |
275 | { |
276 | const int last_prp = PAGE_SIZE / 8 - 1; | |
277 | int i; | |
278 | dma_addr_t prp_dma; | |
279 | ||
280 | if (!prps) | |
281 | return; | |
282 | ||
283 | prp_dma = prps->first_dma; | |
99802a7a MW |
284 | |
285 | if (prps->npages == 0) | |
286 | dma_pool_free(dev->prp_small_pool, prps->list[0], prp_dma); | |
e025344c SMM |
287 | for (i = 0; i < prps->npages; i++) { |
288 | __le64 *prp_list = prps->list[i]; | |
289 | dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); | |
091b6092 | 290 | dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); |
e025344c SMM |
291 | prp_dma = next_prp_dma; |
292 | } | |
293 | kfree(prps); | |
294 | } | |
295 | ||
d534df3c | 296 | struct nvme_bio { |
b60503ba MW |
297 | struct bio *bio; |
298 | int nents; | |
e025344c | 299 | struct nvme_prps *prps; |
b60503ba MW |
300 | struct scatterlist sg[0]; |
301 | }; | |
302 | ||
303 | /* XXX: use a mempool */ | |
d534df3c | 304 | static struct nvme_bio *alloc_nbio(unsigned nseg, gfp_t gfp) |
b60503ba | 305 | { |
d534df3c | 306 | return kzalloc(sizeof(struct nvme_bio) + |
b60503ba MW |
307 | sizeof(struct scatterlist) * nseg, gfp); |
308 | } | |
309 | ||
d534df3c | 310 | static void free_nbio(struct nvme_queue *nvmeq, struct nvme_bio *nbio) |
b60503ba | 311 | { |
d567760c | 312 | nvme_free_prps(nvmeq->dev, nbio->prps); |
d534df3c | 313 | kfree(nbio); |
b60503ba MW |
314 | } |
315 | ||
316 | static void bio_completion(struct nvme_queue *nvmeq, void *ctx, | |
317 | struct nvme_completion *cqe) | |
318 | { | |
d534df3c MW |
319 | struct nvme_bio *nbio = ctx; |
320 | struct bio *bio = nbio->bio; | |
b60503ba MW |
321 | u16 status = le16_to_cpup(&cqe->status) >> 1; |
322 | ||
d534df3c | 323 | dma_unmap_sg(nvmeq->q_dmadev, nbio->sg, nbio->nents, |
b60503ba | 324 | bio_data_dir(bio) ? DMA_TO_DEVICE : DMA_FROM_DEVICE); |
d534df3c | 325 | free_nbio(nvmeq, nbio); |
09a58f53 | 326 | if (status) { |
1ad2f893 | 327 | bio_endio(bio, -EIO); |
09a58f53 | 328 | } else if (bio->bi_vcnt > bio->bi_idx) { |
eac623ba MW |
329 | if (bio_list_empty(&nvmeq->sq_cong)) |
330 | add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait); | |
1ad2f893 MW |
331 | bio_list_add(&nvmeq->sq_cong, bio); |
332 | wake_up_process(nvme_thread); | |
333 | } else { | |
334 | bio_endio(bio, 0); | |
335 | } | |
b60503ba MW |
336 | } |
337 | ||
184d2944 | 338 | /* length is in bytes. gfp flags indicates whether we may sleep. */ |
d567760c | 339 | static struct nvme_prps *nvme_setup_prps(struct nvme_dev *dev, |
e025344c | 340 | struct nvme_common_command *cmd, |
b77954cb MW |
341 | struct scatterlist *sg, int *len, |
342 | gfp_t gfp) | |
ff22b54f | 343 | { |
99802a7a | 344 | struct dma_pool *pool; |
b77954cb | 345 | int length = *len; |
ff22b54f MW |
346 | int dma_len = sg_dma_len(sg); |
347 | u64 dma_addr = sg_dma_address(sg); | |
348 | int offset = offset_in_page(dma_addr); | |
e025344c SMM |
349 | __le64 *prp_list; |
350 | dma_addr_t prp_dma; | |
0d1bc912 | 351 | int nprps, npages, i; |
e025344c | 352 | struct nvme_prps *prps = NULL; |
ff22b54f MW |
353 | |
354 | cmd->prp1 = cpu_to_le64(dma_addr); | |
355 | length -= (PAGE_SIZE - offset); | |
356 | if (length <= 0) | |
e025344c | 357 | return prps; |
ff22b54f MW |
358 | |
359 | dma_len -= (PAGE_SIZE - offset); | |
360 | if (dma_len) { | |
361 | dma_addr += (PAGE_SIZE - offset); | |
362 | } else { | |
363 | sg = sg_next(sg); | |
364 | dma_addr = sg_dma_address(sg); | |
365 | dma_len = sg_dma_len(sg); | |
366 | } | |
367 | ||
368 | if (length <= PAGE_SIZE) { | |
369 | cmd->prp2 = cpu_to_le64(dma_addr); | |
e025344c SMM |
370 | return prps; |
371 | } | |
372 | ||
373 | nprps = DIV_ROUND_UP(length, PAGE_SIZE); | |
0d1bc912 | 374 | npages = DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); |
b77954cb MW |
375 | prps = kmalloc(sizeof(*prps) + sizeof(__le64 *) * npages, gfp); |
376 | if (!prps) { | |
377 | cmd->prp2 = cpu_to_le64(dma_addr); | |
378 | *len = (*len - length) + PAGE_SIZE; | |
379 | return prps; | |
380 | } | |
0d1bc912 | 381 | |
99802a7a MW |
382 | if (nprps <= (256 / 8)) { |
383 | pool = dev->prp_small_pool; | |
384 | prps->npages = 0; | |
385 | } else { | |
386 | pool = dev->prp_page_pool; | |
0d1bc912 | 387 | prps->npages = 1; |
99802a7a MW |
388 | } |
389 | ||
b77954cb MW |
390 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
391 | if (!prp_list) { | |
392 | cmd->prp2 = cpu_to_le64(dma_addr); | |
393 | *len = (*len - length) + PAGE_SIZE; | |
394 | kfree(prps); | |
395 | return NULL; | |
396 | } | |
0d1bc912 | 397 | prps->list[0] = prp_list; |
e025344c SMM |
398 | prps->first_dma = prp_dma; |
399 | cmd->prp2 = cpu_to_le64(prp_dma); | |
400 | i = 0; | |
401 | for (;;) { | |
7523d834 | 402 | if (i == PAGE_SIZE / 8) { |
e025344c | 403 | __le64 *old_prp_list = prp_list; |
b77954cb MW |
404 | prp_list = dma_pool_alloc(pool, gfp, &prp_dma); |
405 | if (!prp_list) { | |
406 | *len = (*len - length); | |
407 | return prps; | |
408 | } | |
0d1bc912 | 409 | prps->list[prps->npages++] = prp_list; |
7523d834 MW |
410 | prp_list[0] = old_prp_list[i - 1]; |
411 | old_prp_list[i - 1] = cpu_to_le64(prp_dma); | |
412 | i = 1; | |
e025344c SMM |
413 | } |
414 | prp_list[i++] = cpu_to_le64(dma_addr); | |
415 | dma_len -= PAGE_SIZE; | |
416 | dma_addr += PAGE_SIZE; | |
417 | length -= PAGE_SIZE; | |
418 | if (length <= 0) | |
419 | break; | |
420 | if (dma_len > 0) | |
421 | continue; | |
422 | BUG_ON(dma_len < 0); | |
423 | sg = sg_next(sg); | |
424 | dma_addr = sg_dma_address(sg); | |
425 | dma_len = sg_dma_len(sg); | |
ff22b54f MW |
426 | } |
427 | ||
e025344c | 428 | return prps; |
ff22b54f MW |
429 | } |
430 | ||
1ad2f893 MW |
431 | /* NVMe scatterlists require no holes in the virtual address */ |
432 | #define BIOVEC_NOT_VIRT_MERGEABLE(vec1, vec2) ((vec2)->bv_offset || \ | |
433 | (((vec1)->bv_offset + (vec1)->bv_len) % PAGE_SIZE)) | |
434 | ||
d534df3c | 435 | static int nvme_map_bio(struct device *dev, struct nvme_bio *nbio, |
b60503ba MW |
436 | struct bio *bio, enum dma_data_direction dma_dir, int psegs) |
437 | { | |
76830840 MW |
438 | struct bio_vec *bvec, *bvprv = NULL; |
439 | struct scatterlist *sg = NULL; | |
1ad2f893 | 440 | int i, old_idx, length = 0, nsegs = 0; |
b60503ba | 441 | |
76830840 | 442 | sg_init_table(nbio->sg, psegs); |
1ad2f893 | 443 | old_idx = bio->bi_idx; |
b60503ba | 444 | bio_for_each_segment(bvec, bio, i) { |
76830840 MW |
445 | if (bvprv && BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) { |
446 | sg->length += bvec->bv_len; | |
447 | } else { | |
1ad2f893 MW |
448 | if (bvprv && BIOVEC_NOT_VIRT_MERGEABLE(bvprv, bvec)) |
449 | break; | |
76830840 MW |
450 | sg = sg ? sg + 1 : nbio->sg; |
451 | sg_set_page(sg, bvec->bv_page, bvec->bv_len, | |
452 | bvec->bv_offset); | |
453 | nsegs++; | |
454 | } | |
1ad2f893 | 455 | length += bvec->bv_len; |
76830840 | 456 | bvprv = bvec; |
b60503ba | 457 | } |
1ad2f893 | 458 | bio->bi_idx = i; |
d534df3c | 459 | nbio->nents = nsegs; |
76830840 | 460 | sg_mark_end(sg); |
1ad2f893 MW |
461 | if (dma_map_sg(dev, nbio->sg, nbio->nents, dma_dir) == 0) { |
462 | bio->bi_idx = old_idx; | |
463 | return -ENOMEM; | |
464 | } | |
465 | return length; | |
b60503ba MW |
466 | } |
467 | ||
00df5cb4 MW |
468 | static int nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
469 | int cmdid) | |
470 | { | |
471 | struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; | |
472 | ||
473 | memset(cmnd, 0, sizeof(*cmnd)); | |
474 | cmnd->common.opcode = nvme_cmd_flush; | |
475 | cmnd->common.command_id = cmdid; | |
476 | cmnd->common.nsid = cpu_to_le32(ns->ns_id); | |
477 | ||
478 | if (++nvmeq->sq_tail == nvmeq->q_depth) | |
479 | nvmeq->sq_tail = 0; | |
480 | writel(nvmeq->sq_tail, nvmeq->q_db); | |
481 | ||
482 | return 0; | |
483 | } | |
484 | ||
485 | static int nvme_submit_flush_data(struct nvme_queue *nvmeq, struct nvme_ns *ns) | |
486 | { | |
487 | int cmdid = alloc_cmdid(nvmeq, (void *)CMD_CTX_FLUSH, | |
488 | sync_completion_id, IO_TIMEOUT); | |
489 | if (unlikely(cmdid < 0)) | |
490 | return cmdid; | |
491 | ||
492 | return nvme_submit_flush(nvmeq, ns, cmdid); | |
493 | } | |
494 | ||
184d2944 MW |
495 | /* |
496 | * Called with local interrupts disabled and the q_lock held. May not sleep. | |
497 | */ | |
b60503ba MW |
498 | static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns, |
499 | struct bio *bio) | |
500 | { | |
ff22b54f | 501 | struct nvme_command *cmnd; |
d534df3c | 502 | struct nvme_bio *nbio; |
b60503ba | 503 | enum dma_data_direction dma_dir; |
1ad2f893 | 504 | int cmdid, length, result = -ENOMEM; |
b60503ba MW |
505 | u16 control; |
506 | u32 dsmgmt; | |
b60503ba MW |
507 | int psegs = bio_phys_segments(ns->queue, bio); |
508 | ||
00df5cb4 MW |
509 | if ((bio->bi_rw & REQ_FLUSH) && psegs) { |
510 | result = nvme_submit_flush_data(nvmeq, ns); | |
511 | if (result) | |
512 | return result; | |
513 | } | |
514 | ||
eeee3226 | 515 | nbio = alloc_nbio(psegs, GFP_ATOMIC); |
d534df3c | 516 | if (!nbio) |
eeee3226 | 517 | goto nomem; |
d534df3c | 518 | nbio->bio = bio; |
b60503ba | 519 | |
eeee3226 | 520 | result = -EBUSY; |
d534df3c | 521 | cmdid = alloc_cmdid(nvmeq, nbio, bio_completion_id, IO_TIMEOUT); |
b60503ba | 522 | if (unlikely(cmdid < 0)) |
d534df3c | 523 | goto free_nbio; |
b60503ba | 524 | |
00df5cb4 MW |
525 | if ((bio->bi_rw & REQ_FLUSH) && !psegs) |
526 | return nvme_submit_flush(nvmeq, ns, cmdid); | |
527 | ||
b60503ba MW |
528 | control = 0; |
529 | if (bio->bi_rw & REQ_FUA) | |
530 | control |= NVME_RW_FUA; | |
531 | if (bio->bi_rw & (REQ_FAILFAST_DEV | REQ_RAHEAD)) | |
532 | control |= NVME_RW_LR; | |
533 | ||
534 | dsmgmt = 0; | |
535 | if (bio->bi_rw & REQ_RAHEAD) | |
536 | dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; | |
537 | ||
ff22b54f | 538 | cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail]; |
b60503ba | 539 | |
b8deb62c | 540 | memset(cmnd, 0, sizeof(*cmnd)); |
b60503ba | 541 | if (bio_data_dir(bio)) { |
ff22b54f | 542 | cmnd->rw.opcode = nvme_cmd_write; |
b60503ba MW |
543 | dma_dir = DMA_TO_DEVICE; |
544 | } else { | |
ff22b54f | 545 | cmnd->rw.opcode = nvme_cmd_read; |
b60503ba MW |
546 | dma_dir = DMA_FROM_DEVICE; |
547 | } | |
548 | ||
1ad2f893 MW |
549 | result = nvme_map_bio(nvmeq->q_dmadev, nbio, bio, dma_dir, psegs); |
550 | if (result < 0) | |
eeee3226 | 551 | goto free_nbio; |
1ad2f893 | 552 | length = result; |
b60503ba | 553 | |
ff22b54f MW |
554 | cmnd->rw.command_id = cmdid; |
555 | cmnd->rw.nsid = cpu_to_le32(ns->ns_id); | |
d567760c | 556 | nbio->prps = nvme_setup_prps(nvmeq->dev, &cmnd->common, nbio->sg, |
b77954cb | 557 | &length, GFP_ATOMIC); |
ff22b54f | 558 | cmnd->rw.slba = cpu_to_le64(bio->bi_sector >> (ns->lba_shift - 9)); |
1ad2f893 | 559 | cmnd->rw.length = cpu_to_le16((length >> ns->lba_shift) - 1); |
ff22b54f MW |
560 | cmnd->rw.control = cpu_to_le16(control); |
561 | cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); | |
b60503ba | 562 | |
d8ee9d69 MW |
563 | bio->bi_sector += length >> 9; |
564 | ||
b60503ba MW |
565 | if (++nvmeq->sq_tail == nvmeq->q_depth) |
566 | nvmeq->sq_tail = 0; | |
7547881d | 567 | writel(nvmeq->sq_tail, nvmeq->q_db); |
b60503ba | 568 | |
1974b1ae MW |
569 | return 0; |
570 | ||
d534df3c MW |
571 | free_nbio: |
572 | free_nbio(nvmeq, nbio); | |
eeee3226 MW |
573 | nomem: |
574 | return result; | |
b60503ba MW |
575 | } |
576 | ||
577 | /* | |
578 | * NB: return value of non-zero would mean that we were a stacking driver. | |
579 | * make_request must always succeed. | |
580 | */ | |
581 | static int nvme_make_request(struct request_queue *q, struct bio *bio) | |
582 | { | |
583 | struct nvme_ns *ns = q->queuedata; | |
584 | struct nvme_queue *nvmeq = get_nvmeq(ns); | |
eeee3226 MW |
585 | int result = -EBUSY; |
586 | ||
587 | spin_lock_irq(&nvmeq->q_lock); | |
588 | if (bio_list_empty(&nvmeq->sq_cong)) | |
589 | result = nvme_submit_bio_queue(nvmeq, ns, bio); | |
590 | if (unlikely(result)) { | |
591 | if (bio_list_empty(&nvmeq->sq_cong)) | |
592 | add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait); | |
b60503ba MW |
593 | bio_list_add(&nvmeq->sq_cong, bio); |
594 | } | |
eeee3226 MW |
595 | |
596 | spin_unlock_irq(&nvmeq->q_lock); | |
b60503ba MW |
597 | put_nvmeq(nvmeq); |
598 | ||
599 | return 0; | |
600 | } | |
601 | ||
602 | struct sync_cmd_info { | |
603 | struct task_struct *task; | |
604 | u32 result; | |
605 | int status; | |
606 | }; | |
607 | ||
608 | static void sync_completion(struct nvme_queue *nvmeq, void *ctx, | |
609 | struct nvme_completion *cqe) | |
610 | { | |
611 | struct sync_cmd_info *cmdinfo = ctx; | |
c4270559 | 612 | if (unlikely((unsigned long)cmdinfo == CMD_CTX_CANCELLED)) |
be7b6275 | 613 | return; |
00df5cb4 MW |
614 | if ((unsigned long)cmdinfo == CMD_CTX_FLUSH) |
615 | return; | |
b36235df MW |
616 | if (unlikely((unsigned long)cmdinfo == CMD_CTX_COMPLETED)) { |
617 | dev_warn(nvmeq->q_dmadev, | |
618 | "completed id %d twice on queue %d\n", | |
619 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
620 | return; | |
621 | } | |
48e3d398 MW |
622 | if (unlikely((unsigned long)cmdinfo == CMD_CTX_INVALID)) { |
623 | dev_warn(nvmeq->q_dmadev, | |
624 | "invalid id %d completed on queue %d\n", | |
625 | cqe->command_id, le16_to_cpup(&cqe->sq_id)); | |
626 | return; | |
627 | } | |
b60503ba MW |
628 | cmdinfo->result = le32_to_cpup(&cqe->result); |
629 | cmdinfo->status = le16_to_cpup(&cqe->status) >> 1; | |
630 | wake_up_process(cmdinfo->task); | |
631 | } | |
632 | ||
633 | typedef void (*completion_fn)(struct nvme_queue *, void *, | |
634 | struct nvme_completion *); | |
635 | ||
8de05535 MW |
636 | static const completion_fn nvme_completions[4] = { |
637 | [sync_completion_id] = sync_completion, | |
638 | [bio_completion_id] = bio_completion, | |
639 | }; | |
640 | ||
b60503ba MW |
641 | static irqreturn_t nvme_process_cq(struct nvme_queue *nvmeq) |
642 | { | |
82123460 | 643 | u16 head, phase; |
b60503ba | 644 | |
b60503ba | 645 | head = nvmeq->cq_head; |
82123460 | 646 | phase = nvmeq->cq_phase; |
b60503ba MW |
647 | |
648 | for (;;) { | |
649 | unsigned long data; | |
650 | void *ptr; | |
651 | unsigned char handler; | |
652 | struct nvme_completion cqe = nvmeq->cqes[head]; | |
82123460 | 653 | if ((le16_to_cpu(cqe.status) & 1) != phase) |
b60503ba MW |
654 | break; |
655 | nvmeq->sq_head = le16_to_cpu(cqe.sq_head); | |
656 | if (++head == nvmeq->q_depth) { | |
657 | head = 0; | |
82123460 | 658 | phase = !phase; |
b60503ba MW |
659 | } |
660 | ||
661 | data = free_cmdid(nvmeq, cqe.command_id); | |
662 | handler = data & 3; | |
663 | ptr = (void *)(data & ~3UL); | |
8de05535 | 664 | nvme_completions[handler](nvmeq, ptr, &cqe); |
b60503ba MW |
665 | } |
666 | ||
667 | /* If the controller ignores the cq head doorbell and continuously | |
668 | * writes to the queue, it is theoretically possible to wrap around | |
669 | * the queue twice and mistakenly return IRQ_NONE. Linux only | |
670 | * requires that 0.1% of your interrupts are handled, so this isn't | |
671 | * a big problem. | |
672 | */ | |
82123460 | 673 | if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) |
b60503ba MW |
674 | return IRQ_NONE; |
675 | ||
f1938f6e | 676 | writel(head, nvmeq->q_db + (1 << nvmeq->dev->db_stride)); |
b60503ba | 677 | nvmeq->cq_head = head; |
82123460 | 678 | nvmeq->cq_phase = phase; |
b60503ba MW |
679 | |
680 | return IRQ_HANDLED; | |
681 | } | |
682 | ||
683 | static irqreturn_t nvme_irq(int irq, void *data) | |
58ffacb5 MW |
684 | { |
685 | irqreturn_t result; | |
686 | struct nvme_queue *nvmeq = data; | |
687 | spin_lock(&nvmeq->q_lock); | |
688 | result = nvme_process_cq(nvmeq); | |
689 | spin_unlock(&nvmeq->q_lock); | |
690 | return result; | |
691 | } | |
692 | ||
693 | static irqreturn_t nvme_irq_check(int irq, void *data) | |
694 | { | |
695 | struct nvme_queue *nvmeq = data; | |
696 | struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head]; | |
697 | if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase) | |
698 | return IRQ_NONE; | |
699 | return IRQ_WAKE_THREAD; | |
700 | } | |
701 | ||
3c0cf138 MW |
702 | static void nvme_abort_command(struct nvme_queue *nvmeq, int cmdid) |
703 | { | |
704 | spin_lock_irq(&nvmeq->q_lock); | |
21075bde | 705 | cancel_cmdid(nvmeq, cmdid); |
3c0cf138 MW |
706 | spin_unlock_irq(&nvmeq->q_lock); |
707 | } | |
708 | ||
b60503ba MW |
709 | /* |
710 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
711 | * if the result is positive, it's an NVM Express status code | |
712 | */ | |
3c0cf138 | 713 | static int nvme_submit_sync_cmd(struct nvme_queue *nvmeq, |
e85248e5 | 714 | struct nvme_command *cmd, u32 *result, unsigned timeout) |
b60503ba MW |
715 | { |
716 | int cmdid; | |
717 | struct sync_cmd_info cmdinfo; | |
718 | ||
719 | cmdinfo.task = current; | |
720 | cmdinfo.status = -EINTR; | |
721 | ||
e85248e5 MW |
722 | cmdid = alloc_cmdid_killable(nvmeq, &cmdinfo, sync_completion_id, |
723 | timeout); | |
b60503ba MW |
724 | if (cmdid < 0) |
725 | return cmdid; | |
726 | cmd->common.command_id = cmdid; | |
727 | ||
3c0cf138 MW |
728 | set_current_state(TASK_KILLABLE); |
729 | nvme_submit_cmd(nvmeq, cmd); | |
b60503ba MW |
730 | schedule(); |
731 | ||
3c0cf138 MW |
732 | if (cmdinfo.status == -EINTR) { |
733 | nvme_abort_command(nvmeq, cmdid); | |
734 | return -EINTR; | |
735 | } | |
736 | ||
b60503ba MW |
737 | if (result) |
738 | *result = cmdinfo.result; | |
739 | ||
740 | return cmdinfo.status; | |
741 | } | |
742 | ||
743 | static int nvme_submit_admin_cmd(struct nvme_dev *dev, struct nvme_command *cmd, | |
744 | u32 *result) | |
745 | { | |
e85248e5 | 746 | return nvme_submit_sync_cmd(dev->queues[0], cmd, result, ADMIN_TIMEOUT); |
b60503ba MW |
747 | } |
748 | ||
749 | static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) | |
750 | { | |
751 | int status; | |
752 | struct nvme_command c; | |
753 | ||
754 | memset(&c, 0, sizeof(c)); | |
755 | c.delete_queue.opcode = opcode; | |
756 | c.delete_queue.qid = cpu_to_le16(id); | |
757 | ||
758 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
759 | if (status) | |
760 | return -EIO; | |
761 | return 0; | |
762 | } | |
763 | ||
764 | static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, | |
765 | struct nvme_queue *nvmeq) | |
766 | { | |
767 | int status; | |
768 | struct nvme_command c; | |
769 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; | |
770 | ||
771 | memset(&c, 0, sizeof(c)); | |
772 | c.create_cq.opcode = nvme_admin_create_cq; | |
773 | c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); | |
774 | c.create_cq.cqid = cpu_to_le16(qid); | |
775 | c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
776 | c.create_cq.cq_flags = cpu_to_le16(flags); | |
777 | c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); | |
778 | ||
779 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
780 | if (status) | |
781 | return -EIO; | |
782 | return 0; | |
783 | } | |
784 | ||
785 | static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, | |
786 | struct nvme_queue *nvmeq) | |
787 | { | |
788 | int status; | |
789 | struct nvme_command c; | |
790 | int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; | |
791 | ||
792 | memset(&c, 0, sizeof(c)); | |
793 | c.create_sq.opcode = nvme_admin_create_sq; | |
794 | c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); | |
795 | c.create_sq.sqid = cpu_to_le16(qid); | |
796 | c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); | |
797 | c.create_sq.sq_flags = cpu_to_le16(flags); | |
798 | c.create_sq.cqid = cpu_to_le16(qid); | |
799 | ||
800 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
801 | if (status) | |
802 | return -EIO; | |
803 | return 0; | |
804 | } | |
805 | ||
806 | static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) | |
807 | { | |
808 | return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); | |
809 | } | |
810 | ||
811 | static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) | |
812 | { | |
813 | return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); | |
814 | } | |
815 | ||
bc5fc7e4 MW |
816 | static int nvme_identify(struct nvme_dev *dev, unsigned nsid, unsigned cns, |
817 | dma_addr_t dma_addr) | |
818 | { | |
819 | struct nvme_command c; | |
820 | ||
821 | memset(&c, 0, sizeof(c)); | |
822 | c.identify.opcode = nvme_admin_identify; | |
823 | c.identify.nsid = cpu_to_le32(nsid); | |
824 | c.identify.prp1 = cpu_to_le64(dma_addr); | |
825 | c.identify.cns = cpu_to_le32(cns); | |
826 | ||
827 | return nvme_submit_admin_cmd(dev, &c, NULL); | |
828 | } | |
829 | ||
830 | static int nvme_get_features(struct nvme_dev *dev, unsigned fid, | |
831 | unsigned dword11, dma_addr_t dma_addr, u32 *result) | |
832 | { | |
833 | struct nvme_command c; | |
834 | ||
835 | memset(&c, 0, sizeof(c)); | |
836 | c.features.opcode = nvme_admin_get_features; | |
837 | c.features.prp1 = cpu_to_le64(dma_addr); | |
838 | c.features.fid = cpu_to_le32(fid); | |
839 | c.features.dword11 = cpu_to_le32(dword11); | |
840 | ||
841 | return nvme_submit_admin_cmd(dev, &c, result); | |
842 | } | |
843 | ||
b60503ba MW |
844 | static void nvme_free_queue(struct nvme_dev *dev, int qid) |
845 | { | |
846 | struct nvme_queue *nvmeq = dev->queues[qid]; | |
aba2080f | 847 | int vector = dev->entry[nvmeq->cq_vector].vector; |
b60503ba | 848 | |
aba2080f MW |
849 | irq_set_affinity_hint(vector, NULL); |
850 | free_irq(vector, nvmeq); | |
b60503ba MW |
851 | |
852 | /* Don't tell the adapter to delete the admin queue */ | |
853 | if (qid) { | |
854 | adapter_delete_sq(dev, qid); | |
855 | adapter_delete_cq(dev, qid); | |
856 | } | |
857 | ||
858 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), | |
859 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
860 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
861 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); | |
862 | kfree(nvmeq); | |
863 | } | |
864 | ||
865 | static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, | |
866 | int depth, int vector) | |
867 | { | |
868 | struct device *dmadev = &dev->pci_dev->dev; | |
e85248e5 | 869 | unsigned extra = (depth / 8) + (depth * sizeof(struct nvme_cmd_info)); |
b60503ba MW |
870 | struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq) + extra, GFP_KERNEL); |
871 | if (!nvmeq) | |
872 | return NULL; | |
873 | ||
874 | nvmeq->cqes = dma_alloc_coherent(dmadev, CQ_SIZE(depth), | |
875 | &nvmeq->cq_dma_addr, GFP_KERNEL); | |
876 | if (!nvmeq->cqes) | |
877 | goto free_nvmeq; | |
878 | memset((void *)nvmeq->cqes, 0, CQ_SIZE(depth)); | |
879 | ||
880 | nvmeq->sq_cmds = dma_alloc_coherent(dmadev, SQ_SIZE(depth), | |
881 | &nvmeq->sq_dma_addr, GFP_KERNEL); | |
882 | if (!nvmeq->sq_cmds) | |
883 | goto free_cqdma; | |
884 | ||
885 | nvmeq->q_dmadev = dmadev; | |
091b6092 | 886 | nvmeq->dev = dev; |
b60503ba MW |
887 | spin_lock_init(&nvmeq->q_lock); |
888 | nvmeq->cq_head = 0; | |
82123460 | 889 | nvmeq->cq_phase = 1; |
b60503ba | 890 | init_waitqueue_head(&nvmeq->sq_full); |
1fa6aead | 891 | init_waitqueue_entry(&nvmeq->sq_cong_wait, nvme_thread); |
b60503ba | 892 | bio_list_init(&nvmeq->sq_cong); |
f1938f6e | 893 | nvmeq->q_db = &dev->dbs[qid << (dev->db_stride + 1)]; |
b60503ba MW |
894 | nvmeq->q_depth = depth; |
895 | nvmeq->cq_vector = vector; | |
896 | ||
897 | return nvmeq; | |
898 | ||
899 | free_cqdma: | |
900 | dma_free_coherent(dmadev, CQ_SIZE(nvmeq->q_depth), (void *)nvmeq->cqes, | |
901 | nvmeq->cq_dma_addr); | |
902 | free_nvmeq: | |
903 | kfree(nvmeq); | |
904 | return NULL; | |
905 | } | |
906 | ||
3001082c MW |
907 | static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq, |
908 | const char *name) | |
909 | { | |
58ffacb5 MW |
910 | if (use_threaded_interrupts) |
911 | return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector, | |
ec6ce618 | 912 | nvme_irq_check, nvme_irq, |
58ffacb5 MW |
913 | IRQF_DISABLED | IRQF_SHARED, |
914 | name, nvmeq); | |
3001082c MW |
915 | return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq, |
916 | IRQF_DISABLED | IRQF_SHARED, name, nvmeq); | |
917 | } | |
918 | ||
b60503ba MW |
919 | static __devinit struct nvme_queue *nvme_create_queue(struct nvme_dev *dev, |
920 | int qid, int cq_size, int vector) | |
921 | { | |
922 | int result; | |
923 | struct nvme_queue *nvmeq = nvme_alloc_queue(dev, qid, cq_size, vector); | |
924 | ||
3f85d50b | 925 | if (!nvmeq) |
6f0f5449 | 926 | return ERR_PTR(-ENOMEM); |
3f85d50b | 927 | |
b60503ba MW |
928 | result = adapter_alloc_cq(dev, qid, nvmeq); |
929 | if (result < 0) | |
930 | goto free_nvmeq; | |
931 | ||
932 | result = adapter_alloc_sq(dev, qid, nvmeq); | |
933 | if (result < 0) | |
934 | goto release_cq; | |
935 | ||
3001082c | 936 | result = queue_request_irq(dev, nvmeq, "nvme"); |
b60503ba MW |
937 | if (result < 0) |
938 | goto release_sq; | |
939 | ||
940 | return nvmeq; | |
941 | ||
942 | release_sq: | |
943 | adapter_delete_sq(dev, qid); | |
944 | release_cq: | |
945 | adapter_delete_cq(dev, qid); | |
946 | free_nvmeq: | |
947 | dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), | |
948 | (void *)nvmeq->cqes, nvmeq->cq_dma_addr); | |
949 | dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), | |
950 | nvmeq->sq_cmds, nvmeq->sq_dma_addr); | |
951 | kfree(nvmeq); | |
6f0f5449 | 952 | return ERR_PTR(result); |
b60503ba MW |
953 | } |
954 | ||
955 | static int __devinit nvme_configure_admin_queue(struct nvme_dev *dev) | |
956 | { | |
957 | int result; | |
958 | u32 aqa; | |
22605f96 MW |
959 | u64 cap; |
960 | unsigned long timeout; | |
b60503ba MW |
961 | struct nvme_queue *nvmeq; |
962 | ||
963 | dev->dbs = ((void __iomem *)dev->bar) + 4096; | |
964 | ||
965 | nvmeq = nvme_alloc_queue(dev, 0, 64, 0); | |
3f85d50b MW |
966 | if (!nvmeq) |
967 | return -ENOMEM; | |
b60503ba MW |
968 | |
969 | aqa = nvmeq->q_depth - 1; | |
970 | aqa |= aqa << 16; | |
971 | ||
972 | dev->ctrl_config = NVME_CC_ENABLE | NVME_CC_CSS_NVM; | |
973 | dev->ctrl_config |= (PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT; | |
974 | dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; | |
7f53f9d2 | 975 | dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; |
b60503ba | 976 | |
5911f200 | 977 | writel(0, &dev->bar->cc); |
b60503ba MW |
978 | writel(aqa, &dev->bar->aqa); |
979 | writeq(nvmeq->sq_dma_addr, &dev->bar->asq); | |
980 | writeq(nvmeq->cq_dma_addr, &dev->bar->acq); | |
981 | writel(dev->ctrl_config, &dev->bar->cc); | |
982 | ||
22605f96 MW |
983 | cap = readq(&dev->bar->cap); |
984 | timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
f1938f6e | 985 | dev->db_stride = NVME_CAP_STRIDE(cap); |
22605f96 | 986 | |
b60503ba MW |
987 | while (!(readl(&dev->bar->csts) & NVME_CSTS_RDY)) { |
988 | msleep(100); | |
989 | if (fatal_signal_pending(current)) | |
990 | return -EINTR; | |
22605f96 MW |
991 | if (time_after(jiffies, timeout)) { |
992 | dev_err(&dev->pci_dev->dev, | |
993 | "Device not ready; aborting initialisation\n"); | |
994 | return -ENODEV; | |
995 | } | |
b60503ba MW |
996 | } |
997 | ||
3001082c | 998 | result = queue_request_irq(dev, nvmeq, "nvme admin"); |
b60503ba MW |
999 | dev->queues[0] = nvmeq; |
1000 | return result; | |
1001 | } | |
1002 | ||
7fc3cdab MW |
1003 | static int nvme_map_user_pages(struct nvme_dev *dev, int write, |
1004 | unsigned long addr, unsigned length, | |
1005 | struct scatterlist **sgp) | |
b60503ba | 1006 | { |
36c14ed9 | 1007 | int i, err, count, nents, offset; |
7fc3cdab MW |
1008 | struct scatterlist *sg; |
1009 | struct page **pages; | |
36c14ed9 MW |
1010 | |
1011 | if (addr & 3) | |
1012 | return -EINVAL; | |
7fc3cdab MW |
1013 | if (!length) |
1014 | return -EINVAL; | |
1015 | ||
36c14ed9 | 1016 | offset = offset_in_page(addr); |
7fc3cdab MW |
1017 | count = DIV_ROUND_UP(offset + length, PAGE_SIZE); |
1018 | pages = kcalloc(count, sizeof(*pages), GFP_KERNEL); | |
36c14ed9 MW |
1019 | |
1020 | err = get_user_pages_fast(addr, count, 1, pages); | |
1021 | if (err < count) { | |
1022 | count = err; | |
1023 | err = -EFAULT; | |
1024 | goto put_pages; | |
1025 | } | |
7fc3cdab MW |
1026 | |
1027 | sg = kcalloc(count, sizeof(*sg), GFP_KERNEL); | |
36c14ed9 | 1028 | sg_init_table(sg, count); |
d0ba1e49 MW |
1029 | for (i = 0; i < count; i++) { |
1030 | sg_set_page(&sg[i], pages[i], | |
1031 | min_t(int, length, PAGE_SIZE - offset), offset); | |
1032 | length -= (PAGE_SIZE - offset); | |
1033 | offset = 0; | |
7fc3cdab MW |
1034 | } |
1035 | ||
1036 | err = -ENOMEM; | |
1037 | nents = dma_map_sg(&dev->pci_dev->dev, sg, count, | |
1038 | write ? DMA_TO_DEVICE : DMA_FROM_DEVICE); | |
36c14ed9 MW |
1039 | if (!nents) |
1040 | goto put_pages; | |
b60503ba | 1041 | |
7fc3cdab MW |
1042 | kfree(pages); |
1043 | *sgp = sg; | |
1044 | return nents; | |
b60503ba | 1045 | |
7fc3cdab MW |
1046 | put_pages: |
1047 | for (i = 0; i < count; i++) | |
1048 | put_page(pages[i]); | |
1049 | kfree(pages); | |
1050 | return err; | |
1051 | } | |
b60503ba | 1052 | |
7fc3cdab | 1053 | static void nvme_unmap_user_pages(struct nvme_dev *dev, int write, |
d1a490e0 | 1054 | unsigned long addr, int length, struct scatterlist *sg) |
7fc3cdab MW |
1055 | { |
1056 | int i, count; | |
b60503ba | 1057 | |
7fc3cdab | 1058 | count = DIV_ROUND_UP(offset_in_page(addr) + length, PAGE_SIZE); |
d1a490e0 | 1059 | dma_unmap_sg(&dev->pci_dev->dev, sg, count, DMA_FROM_DEVICE); |
7fc3cdab | 1060 | |
36c14ed9 | 1061 | for (i = 0; i < count; i++) |
7fc3cdab MW |
1062 | put_page(sg_page(&sg[i])); |
1063 | } | |
b60503ba | 1064 | |
a53295b6 MW |
1065 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) |
1066 | { | |
1067 | struct nvme_dev *dev = ns->dev; | |
1068 | struct nvme_queue *nvmeq; | |
1069 | struct nvme_user_io io; | |
1070 | struct nvme_command c; | |
1071 | unsigned length; | |
a53295b6 MW |
1072 | int nents, status; |
1073 | struct scatterlist *sg; | |
e025344c | 1074 | struct nvme_prps *prps; |
a53295b6 MW |
1075 | |
1076 | if (copy_from_user(&io, uio, sizeof(io))) | |
1077 | return -EFAULT; | |
6c7d4945 MW |
1078 | length = (io.nblocks + 1) << ns->lba_shift; |
1079 | ||
1080 | switch (io.opcode) { | |
1081 | case nvme_cmd_write: | |
1082 | case nvme_cmd_read: | |
6bbf1acd | 1083 | case nvme_cmd_compare: |
6c7d4945 MW |
1084 | nents = nvme_map_user_pages(dev, io.opcode & 1, io.addr, |
1085 | length, &sg); | |
6413214c | 1086 | break; |
6c7d4945 | 1087 | default: |
6bbf1acd | 1088 | return -EINVAL; |
6c7d4945 MW |
1089 | } |
1090 | ||
a53295b6 MW |
1091 | if (nents < 0) |
1092 | return nents; | |
1093 | ||
1094 | memset(&c, 0, sizeof(c)); | |
1095 | c.rw.opcode = io.opcode; | |
1096 | c.rw.flags = io.flags; | |
6c7d4945 | 1097 | c.rw.nsid = cpu_to_le32(ns->ns_id); |
a53295b6 | 1098 | c.rw.slba = cpu_to_le64(io.slba); |
6c7d4945 | 1099 | c.rw.length = cpu_to_le16(io.nblocks); |
a53295b6 MW |
1100 | c.rw.control = cpu_to_le16(io.control); |
1101 | c.rw.dsmgmt = cpu_to_le16(io.dsmgmt); | |
6c7d4945 MW |
1102 | c.rw.reftag = io.reftag; |
1103 | c.rw.apptag = io.apptag; | |
1104 | c.rw.appmask = io.appmask; | |
a53295b6 | 1105 | /* XXX: metadata */ |
b77954cb | 1106 | prps = nvme_setup_prps(dev, &c.common, sg, &length, GFP_KERNEL); |
a53295b6 | 1107 | |
d567760c | 1108 | nvmeq = get_nvmeq(ns); |
fa922821 MW |
1109 | /* |
1110 | * Since nvme_submit_sync_cmd sleeps, we can't keep preemption | |
b1ad37ef MW |
1111 | * disabled. We may be preempted at any point, and be rescheduled |
1112 | * to a different CPU. That will cause cacheline bouncing, but no | |
1113 | * additional races since q_lock already protects against other CPUs. | |
1114 | */ | |
a53295b6 | 1115 | put_nvmeq(nvmeq); |
b77954cb MW |
1116 | if (length != (io.nblocks + 1) << ns->lba_shift) |
1117 | status = -ENOMEM; | |
1118 | else | |
1119 | status = nvme_submit_sync_cmd(nvmeq, &c, NULL, IO_TIMEOUT); | |
a53295b6 | 1120 | |
d1a490e0 | 1121 | nvme_unmap_user_pages(dev, io.opcode & 1, io.addr, length, sg); |
d567760c | 1122 | nvme_free_prps(dev, prps); |
a53295b6 MW |
1123 | return status; |
1124 | } | |
1125 | ||
6bbf1acd MW |
1126 | static int nvme_user_admin_cmd(struct nvme_ns *ns, |
1127 | struct nvme_admin_cmd __user *ucmd) | |
6ee44cdc MW |
1128 | { |
1129 | struct nvme_dev *dev = ns->dev; | |
6bbf1acd | 1130 | struct nvme_admin_cmd cmd; |
6ee44cdc | 1131 | struct nvme_command c; |
6bbf1acd | 1132 | int status, length, nents = 0; |
6ee44cdc | 1133 | struct scatterlist *sg; |
6bbf1acd | 1134 | struct nvme_prps *prps = NULL; |
6ee44cdc | 1135 | |
6bbf1acd MW |
1136 | if (!capable(CAP_SYS_ADMIN)) |
1137 | return -EACCES; | |
1138 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
6ee44cdc | 1139 | return -EFAULT; |
6ee44cdc MW |
1140 | |
1141 | memset(&c, 0, sizeof(c)); | |
6bbf1acd MW |
1142 | c.common.opcode = cmd.opcode; |
1143 | c.common.flags = cmd.flags; | |
1144 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
1145 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
1146 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
1147 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
1148 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
1149 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
1150 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
1151 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
1152 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
1153 | ||
1154 | length = cmd.data_len; | |
1155 | if (cmd.data_len) { | |
1156 | nents = nvme_map_user_pages(dev, 1, cmd.addr, length, &sg); | |
1157 | if (nents < 0) | |
1158 | return nents; | |
1159 | prps = nvme_setup_prps(dev, &c.common, sg, &length, GFP_KERNEL); | |
1160 | } | |
1161 | ||
1162 | if (length != cmd.data_len) | |
b77954cb MW |
1163 | status = -ENOMEM; |
1164 | else | |
1165 | status = nvme_submit_admin_cmd(dev, &c, NULL); | |
6bbf1acd | 1166 | if (cmd.data_len) { |
d1a490e0 | 1167 | nvme_unmap_user_pages(dev, 0, cmd.addr, cmd.data_len, sg); |
6bbf1acd MW |
1168 | nvme_free_prps(dev, prps); |
1169 | } | |
6ee44cdc MW |
1170 | return status; |
1171 | } | |
1172 | ||
b60503ba MW |
1173 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, |
1174 | unsigned long arg) | |
1175 | { | |
1176 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
1177 | ||
1178 | switch (cmd) { | |
6bbf1acd MW |
1179 | case NVME_IOCTL_ID: |
1180 | return ns->ns_id; | |
1181 | case NVME_IOCTL_ADMIN_CMD: | |
1182 | return nvme_user_admin_cmd(ns, (void __user *)arg); | |
a53295b6 MW |
1183 | case NVME_IOCTL_SUBMIT_IO: |
1184 | return nvme_submit_io(ns, (void __user *)arg); | |
b60503ba MW |
1185 | default: |
1186 | return -ENOTTY; | |
1187 | } | |
1188 | } | |
1189 | ||
1190 | static const struct block_device_operations nvme_fops = { | |
1191 | .owner = THIS_MODULE, | |
1192 | .ioctl = nvme_ioctl, | |
49481682 | 1193 | .compat_ioctl = nvme_ioctl, |
b60503ba MW |
1194 | }; |
1195 | ||
8de05535 MW |
1196 | static void nvme_timeout_ios(struct nvme_queue *nvmeq) |
1197 | { | |
1198 | int depth = nvmeq->q_depth - 1; | |
1199 | struct nvme_cmd_info *info = nvme_cmd_info(nvmeq); | |
1200 | unsigned long now = jiffies; | |
1201 | int cmdid; | |
1202 | ||
1203 | for_each_set_bit(cmdid, nvmeq->cmdid_data, depth) { | |
1204 | unsigned long data; | |
1205 | void *ptr; | |
1206 | unsigned char handler; | |
1207 | static struct nvme_completion cqe = { .status = cpu_to_le16(NVME_SC_ABORT_REQ) << 1, }; | |
1208 | ||
1209 | if (!time_after(now, info[cmdid].timeout)) | |
1210 | continue; | |
1211 | dev_warn(nvmeq->q_dmadev, "Timing out I/O %d\n", cmdid); | |
1212 | data = cancel_cmdid(nvmeq, cmdid); | |
1213 | handler = data & 3; | |
1214 | ptr = (void *)(data & ~3UL); | |
1215 | nvme_completions[handler](nvmeq, ptr, &cqe); | |
1216 | } | |
1217 | } | |
1218 | ||
1fa6aead MW |
1219 | static void nvme_resubmit_bios(struct nvme_queue *nvmeq) |
1220 | { | |
1221 | while (bio_list_peek(&nvmeq->sq_cong)) { | |
1222 | struct bio *bio = bio_list_pop(&nvmeq->sq_cong); | |
1223 | struct nvme_ns *ns = bio->bi_bdev->bd_disk->private_data; | |
1224 | if (nvme_submit_bio_queue(nvmeq, ns, bio)) { | |
1225 | bio_list_add_head(&nvmeq->sq_cong, bio); | |
1226 | break; | |
1227 | } | |
3cb967c0 MW |
1228 | if (bio_list_empty(&nvmeq->sq_cong)) |
1229 | remove_wait_queue(&nvmeq->sq_full, | |
1230 | &nvmeq->sq_cong_wait); | |
1fa6aead MW |
1231 | } |
1232 | } | |
1233 | ||
1234 | static int nvme_kthread(void *data) | |
1235 | { | |
1236 | struct nvme_dev *dev; | |
1237 | ||
1238 | while (!kthread_should_stop()) { | |
1239 | __set_current_state(TASK_RUNNING); | |
1240 | spin_lock(&dev_list_lock); | |
1241 | list_for_each_entry(dev, &dev_list, node) { | |
1242 | int i; | |
1243 | for (i = 0; i < dev->queue_count; i++) { | |
1244 | struct nvme_queue *nvmeq = dev->queues[i]; | |
740216fc MW |
1245 | if (!nvmeq) |
1246 | continue; | |
1fa6aead MW |
1247 | spin_lock_irq(&nvmeq->q_lock); |
1248 | if (nvme_process_cq(nvmeq)) | |
1249 | printk("process_cq did something\n"); | |
8de05535 | 1250 | nvme_timeout_ios(nvmeq); |
1fa6aead MW |
1251 | nvme_resubmit_bios(nvmeq); |
1252 | spin_unlock_irq(&nvmeq->q_lock); | |
1253 | } | |
1254 | } | |
1255 | spin_unlock(&dev_list_lock); | |
1256 | set_current_state(TASK_INTERRUPTIBLE); | |
1257 | schedule_timeout(HZ); | |
1258 | } | |
1259 | return 0; | |
1260 | } | |
1261 | ||
5aff9382 MW |
1262 | static DEFINE_IDA(nvme_index_ida); |
1263 | ||
1264 | static int nvme_get_ns_idx(void) | |
1265 | { | |
1266 | int index, error; | |
1267 | ||
1268 | do { | |
1269 | if (!ida_pre_get(&nvme_index_ida, GFP_KERNEL)) | |
1270 | return -1; | |
1271 | ||
1272 | spin_lock(&dev_list_lock); | |
1273 | error = ida_get_new(&nvme_index_ida, &index); | |
1274 | spin_unlock(&dev_list_lock); | |
1275 | } while (error == -EAGAIN); | |
1276 | ||
1277 | if (error) | |
1278 | index = -1; | |
1279 | return index; | |
1280 | } | |
1281 | ||
1282 | static void nvme_put_ns_idx(int index) | |
1283 | { | |
1284 | spin_lock(&dev_list_lock); | |
1285 | ida_remove(&nvme_index_ida, index); | |
1286 | spin_unlock(&dev_list_lock); | |
1287 | } | |
1288 | ||
1289 | static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, int nsid, | |
b60503ba MW |
1290 | struct nvme_id_ns *id, struct nvme_lba_range_type *rt) |
1291 | { | |
1292 | struct nvme_ns *ns; | |
1293 | struct gendisk *disk; | |
1294 | int lbaf; | |
1295 | ||
1296 | if (rt->attributes & NVME_LBART_ATTRIB_HIDE) | |
1297 | return NULL; | |
1298 | ||
1299 | ns = kzalloc(sizeof(*ns), GFP_KERNEL); | |
1300 | if (!ns) | |
1301 | return NULL; | |
1302 | ns->queue = blk_alloc_queue(GFP_KERNEL); | |
1303 | if (!ns->queue) | |
1304 | goto out_free_ns; | |
1305 | ns->queue->queue_flags = QUEUE_FLAG_DEFAULT | QUEUE_FLAG_NOMERGES | | |
1306 | QUEUE_FLAG_NONROT | QUEUE_FLAG_DISCARD; | |
1307 | blk_queue_make_request(ns->queue, nvme_make_request); | |
1308 | ns->dev = dev; | |
1309 | ns->queue->queuedata = ns; | |
1310 | ||
1311 | disk = alloc_disk(NVME_MINORS); | |
1312 | if (!disk) | |
1313 | goto out_free_queue; | |
5aff9382 | 1314 | ns->ns_id = nsid; |
b60503ba MW |
1315 | ns->disk = disk; |
1316 | lbaf = id->flbas & 0xf; | |
1317 | ns->lba_shift = id->lbaf[lbaf].ds; | |
1318 | ||
1319 | disk->major = nvme_major; | |
1320 | disk->minors = NVME_MINORS; | |
5aff9382 | 1321 | disk->first_minor = NVME_MINORS * nvme_get_ns_idx(); |
b60503ba MW |
1322 | disk->fops = &nvme_fops; |
1323 | disk->private_data = ns; | |
1324 | disk->queue = ns->queue; | |
388f037f | 1325 | disk->driverfs_dev = &dev->pci_dev->dev; |
5aff9382 | 1326 | sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid); |
b60503ba MW |
1327 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); |
1328 | ||
1329 | return ns; | |
1330 | ||
1331 | out_free_queue: | |
1332 | blk_cleanup_queue(ns->queue); | |
1333 | out_free_ns: | |
1334 | kfree(ns); | |
1335 | return NULL; | |
1336 | } | |
1337 | ||
1338 | static void nvme_ns_free(struct nvme_ns *ns) | |
1339 | { | |
5aff9382 | 1340 | int index = ns->disk->first_minor / NVME_MINORS; |
b60503ba | 1341 | put_disk(ns->disk); |
5aff9382 | 1342 | nvme_put_ns_idx(index); |
b60503ba MW |
1343 | blk_cleanup_queue(ns->queue); |
1344 | kfree(ns); | |
1345 | } | |
1346 | ||
b3b06812 | 1347 | static int set_queue_count(struct nvme_dev *dev, int count) |
b60503ba MW |
1348 | { |
1349 | int status; | |
1350 | u32 result; | |
b3b06812 | 1351 | u32 q_count = (count - 1) | ((count - 1) << 16); |
b60503ba | 1352 | |
bc5fc7e4 MW |
1353 | status = nvme_get_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0, |
1354 | &result); | |
b60503ba MW |
1355 | if (status) |
1356 | return -EIO; | |
1357 | return min(result & 0xffff, result >> 16) + 1; | |
1358 | } | |
1359 | ||
b60503ba MW |
1360 | static int __devinit nvme_setup_io_queues(struct nvme_dev *dev) |
1361 | { | |
f1938f6e | 1362 | int result, cpu, i, nr_io_queues, db_bar_size; |
b60503ba | 1363 | |
b348b7d5 MW |
1364 | nr_io_queues = num_online_cpus(); |
1365 | result = set_queue_count(dev, nr_io_queues); | |
1b23484b MW |
1366 | if (result < 0) |
1367 | return result; | |
b348b7d5 MW |
1368 | if (result < nr_io_queues) |
1369 | nr_io_queues = result; | |
b60503ba | 1370 | |
1b23484b MW |
1371 | /* Deregister the admin queue's interrupt */ |
1372 | free_irq(dev->entry[0].vector, dev->queues[0]); | |
1373 | ||
f1938f6e MW |
1374 | db_bar_size = 4096 + ((nr_io_queues + 1) << (dev->db_stride + 3)); |
1375 | if (db_bar_size > 8192) { | |
1376 | iounmap(dev->bar); | |
1377 | dev->bar = ioremap(pci_resource_start(dev->pci_dev, 0), | |
1378 | db_bar_size); | |
1379 | dev->dbs = ((void __iomem *)dev->bar) + 4096; | |
1380 | dev->queues[0]->q_db = dev->dbs; | |
1381 | } | |
1382 | ||
b348b7d5 | 1383 | for (i = 0; i < nr_io_queues; i++) |
1b23484b MW |
1384 | dev->entry[i].entry = i; |
1385 | for (;;) { | |
b348b7d5 MW |
1386 | result = pci_enable_msix(dev->pci_dev, dev->entry, |
1387 | nr_io_queues); | |
1b23484b MW |
1388 | if (result == 0) { |
1389 | break; | |
1390 | } else if (result > 0) { | |
b348b7d5 | 1391 | nr_io_queues = result; |
1b23484b MW |
1392 | continue; |
1393 | } else { | |
b348b7d5 | 1394 | nr_io_queues = 1; |
1b23484b MW |
1395 | break; |
1396 | } | |
1397 | } | |
1398 | ||
1399 | result = queue_request_irq(dev, dev->queues[0], "nvme admin"); | |
1400 | /* XXX: handle failure here */ | |
1401 | ||
1402 | cpu = cpumask_first(cpu_online_mask); | |
b348b7d5 | 1403 | for (i = 0; i < nr_io_queues; i++) { |
1b23484b MW |
1404 | irq_set_affinity_hint(dev->entry[i].vector, get_cpu_mask(cpu)); |
1405 | cpu = cpumask_next(cpu, cpu_online_mask); | |
1406 | } | |
1407 | ||
b348b7d5 | 1408 | for (i = 0; i < nr_io_queues; i++) { |
1b23484b MW |
1409 | dev->queues[i + 1] = nvme_create_queue(dev, i + 1, |
1410 | NVME_Q_DEPTH, i); | |
6f0f5449 MW |
1411 | if (IS_ERR(dev->queues[i + 1])) |
1412 | return PTR_ERR(dev->queues[i + 1]); | |
1b23484b MW |
1413 | dev->queue_count++; |
1414 | } | |
b60503ba | 1415 | |
9ecdc946 MW |
1416 | for (; i < num_possible_cpus(); i++) { |
1417 | int target = i % rounddown_pow_of_two(dev->queue_count - 1); | |
1418 | dev->queues[i + 1] = dev->queues[target + 1]; | |
1419 | } | |
1420 | ||
b60503ba MW |
1421 | return 0; |
1422 | } | |
1423 | ||
1424 | static void nvme_free_queues(struct nvme_dev *dev) | |
1425 | { | |
1426 | int i; | |
1427 | ||
1428 | for (i = dev->queue_count - 1; i >= 0; i--) | |
1429 | nvme_free_queue(dev, i); | |
1430 | } | |
1431 | ||
1432 | static int __devinit nvme_dev_add(struct nvme_dev *dev) | |
1433 | { | |
1434 | int res, nn, i; | |
1435 | struct nvme_ns *ns, *next; | |
51814232 | 1436 | struct nvme_id_ctrl *ctrl; |
bc5fc7e4 MW |
1437 | struct nvme_id_ns *id_ns; |
1438 | void *mem; | |
b60503ba | 1439 | dma_addr_t dma_addr; |
b60503ba MW |
1440 | |
1441 | res = nvme_setup_io_queues(dev); | |
1442 | if (res) | |
1443 | return res; | |
1444 | ||
bc5fc7e4 | 1445 | mem = dma_alloc_coherent(&dev->pci_dev->dev, 8192, &dma_addr, |
b60503ba MW |
1446 | GFP_KERNEL); |
1447 | ||
bc5fc7e4 | 1448 | res = nvme_identify(dev, 0, 1, dma_addr); |
b60503ba MW |
1449 | if (res) { |
1450 | res = -EIO; | |
1451 | goto out_free; | |
1452 | } | |
1453 | ||
bc5fc7e4 | 1454 | ctrl = mem; |
51814232 MW |
1455 | nn = le32_to_cpup(&ctrl->nn); |
1456 | memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn)); | |
1457 | memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn)); | |
1458 | memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr)); | |
b60503ba | 1459 | |
bc5fc7e4 | 1460 | id_ns = mem; |
2b2c1896 | 1461 | for (i = 1; i <= nn; i++) { |
bc5fc7e4 | 1462 | res = nvme_identify(dev, i, 0, dma_addr); |
b60503ba MW |
1463 | if (res) |
1464 | continue; | |
1465 | ||
bc5fc7e4 | 1466 | if (id_ns->ncap == 0) |
b60503ba MW |
1467 | continue; |
1468 | ||
bc5fc7e4 MW |
1469 | res = nvme_get_features(dev, NVME_FEAT_LBA_RANGE, i, |
1470 | dma_addr + 4096, NULL); | |
b60503ba MW |
1471 | if (res) |
1472 | continue; | |
1473 | ||
bc5fc7e4 | 1474 | ns = nvme_alloc_ns(dev, i, mem, mem + 4096); |
b60503ba MW |
1475 | if (ns) |
1476 | list_add_tail(&ns->list, &dev->namespaces); | |
1477 | } | |
1478 | list_for_each_entry(ns, &dev->namespaces, list) | |
1479 | add_disk(ns->disk); | |
1480 | ||
bc5fc7e4 | 1481 | goto out; |
b60503ba MW |
1482 | |
1483 | out_free: | |
1484 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) { | |
1485 | list_del(&ns->list); | |
1486 | nvme_ns_free(ns); | |
1487 | } | |
1488 | ||
bc5fc7e4 | 1489 | out: |
684f5c20 | 1490 | dma_free_coherent(&dev->pci_dev->dev, 8192, mem, dma_addr); |
b60503ba MW |
1491 | return res; |
1492 | } | |
1493 | ||
1494 | static int nvme_dev_remove(struct nvme_dev *dev) | |
1495 | { | |
1496 | struct nvme_ns *ns, *next; | |
1497 | ||
1fa6aead MW |
1498 | spin_lock(&dev_list_lock); |
1499 | list_del(&dev->node); | |
1500 | spin_unlock(&dev_list_lock); | |
1501 | ||
b60503ba MW |
1502 | /* TODO: wait all I/O finished or cancel them */ |
1503 | ||
1504 | list_for_each_entry_safe(ns, next, &dev->namespaces, list) { | |
1505 | list_del(&ns->list); | |
1506 | del_gendisk(ns->disk); | |
1507 | nvme_ns_free(ns); | |
1508 | } | |
1509 | ||
1510 | nvme_free_queues(dev); | |
1511 | ||
1512 | return 0; | |
1513 | } | |
1514 | ||
091b6092 MW |
1515 | static int nvme_setup_prp_pools(struct nvme_dev *dev) |
1516 | { | |
1517 | struct device *dmadev = &dev->pci_dev->dev; | |
1518 | dev->prp_page_pool = dma_pool_create("prp list page", dmadev, | |
1519 | PAGE_SIZE, PAGE_SIZE, 0); | |
1520 | if (!dev->prp_page_pool) | |
1521 | return -ENOMEM; | |
1522 | ||
99802a7a MW |
1523 | /* Optimisation for I/Os between 4k and 128k */ |
1524 | dev->prp_small_pool = dma_pool_create("prp list 256", dmadev, | |
1525 | 256, 256, 0); | |
1526 | if (!dev->prp_small_pool) { | |
1527 | dma_pool_destroy(dev->prp_page_pool); | |
1528 | return -ENOMEM; | |
1529 | } | |
091b6092 MW |
1530 | return 0; |
1531 | } | |
1532 | ||
1533 | static void nvme_release_prp_pools(struct nvme_dev *dev) | |
1534 | { | |
1535 | dma_pool_destroy(dev->prp_page_pool); | |
99802a7a | 1536 | dma_pool_destroy(dev->prp_small_pool); |
091b6092 MW |
1537 | } |
1538 | ||
b60503ba MW |
1539 | /* XXX: Use an ida or something to let remove / add work correctly */ |
1540 | static void nvme_set_instance(struct nvme_dev *dev) | |
1541 | { | |
1542 | static int instance; | |
1543 | dev->instance = instance++; | |
1544 | } | |
1545 | ||
1546 | static void nvme_release_instance(struct nvme_dev *dev) | |
1547 | { | |
1548 | } | |
1549 | ||
1550 | static int __devinit nvme_probe(struct pci_dev *pdev, | |
1551 | const struct pci_device_id *id) | |
1552 | { | |
574e8b95 | 1553 | int bars, result = -ENOMEM; |
b60503ba MW |
1554 | struct nvme_dev *dev; |
1555 | ||
1556 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
1557 | if (!dev) | |
1558 | return -ENOMEM; | |
1559 | dev->entry = kcalloc(num_possible_cpus(), sizeof(*dev->entry), | |
1560 | GFP_KERNEL); | |
1561 | if (!dev->entry) | |
1562 | goto free; | |
1b23484b MW |
1563 | dev->queues = kcalloc(num_possible_cpus() + 1, sizeof(void *), |
1564 | GFP_KERNEL); | |
b60503ba MW |
1565 | if (!dev->queues) |
1566 | goto free; | |
1567 | ||
0ee5a7d7 SMM |
1568 | if (pci_enable_device_mem(pdev)) |
1569 | goto free; | |
f64d3365 | 1570 | pci_set_master(pdev); |
574e8b95 MW |
1571 | bars = pci_select_bars(pdev, IORESOURCE_MEM); |
1572 | if (pci_request_selected_regions(pdev, bars, "nvme")) | |
1573 | goto disable; | |
0ee5a7d7 | 1574 | |
b60503ba MW |
1575 | INIT_LIST_HEAD(&dev->namespaces); |
1576 | dev->pci_dev = pdev; | |
1577 | pci_set_drvdata(pdev, dev); | |
2930353f MW |
1578 | dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)); |
1579 | dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64)); | |
b60503ba | 1580 | nvme_set_instance(dev); |
53c9577e | 1581 | dev->entry[0].vector = pdev->irq; |
b60503ba | 1582 | |
091b6092 MW |
1583 | result = nvme_setup_prp_pools(dev); |
1584 | if (result) | |
1585 | goto disable_msix; | |
1586 | ||
b60503ba MW |
1587 | dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); |
1588 | if (!dev->bar) { | |
1589 | result = -ENOMEM; | |
574e8b95 | 1590 | goto disable_msix; |
b60503ba MW |
1591 | } |
1592 | ||
1593 | result = nvme_configure_admin_queue(dev); | |
1594 | if (result) | |
1595 | goto unmap; | |
1596 | dev->queue_count++; | |
1597 | ||
1fa6aead MW |
1598 | spin_lock(&dev_list_lock); |
1599 | list_add(&dev->node, &dev_list); | |
1600 | spin_unlock(&dev_list_lock); | |
1601 | ||
740216fc MW |
1602 | result = nvme_dev_add(dev); |
1603 | if (result) | |
1604 | goto delete; | |
1605 | ||
b60503ba MW |
1606 | return 0; |
1607 | ||
1608 | delete: | |
740216fc MW |
1609 | spin_lock(&dev_list_lock); |
1610 | list_del(&dev->node); | |
1611 | spin_unlock(&dev_list_lock); | |
1612 | ||
b60503ba MW |
1613 | nvme_free_queues(dev); |
1614 | unmap: | |
1615 | iounmap(dev->bar); | |
574e8b95 | 1616 | disable_msix: |
b60503ba MW |
1617 | pci_disable_msix(pdev); |
1618 | nvme_release_instance(dev); | |
091b6092 | 1619 | nvme_release_prp_pools(dev); |
574e8b95 | 1620 | disable: |
0ee5a7d7 | 1621 | pci_disable_device(pdev); |
574e8b95 | 1622 | pci_release_regions(pdev); |
b60503ba MW |
1623 | free: |
1624 | kfree(dev->queues); | |
1625 | kfree(dev->entry); | |
1626 | kfree(dev); | |
1627 | return result; | |
1628 | } | |
1629 | ||
1630 | static void __devexit nvme_remove(struct pci_dev *pdev) | |
1631 | { | |
1632 | struct nvme_dev *dev = pci_get_drvdata(pdev); | |
1633 | nvme_dev_remove(dev); | |
1634 | pci_disable_msix(pdev); | |
1635 | iounmap(dev->bar); | |
1636 | nvme_release_instance(dev); | |
091b6092 | 1637 | nvme_release_prp_pools(dev); |
0ee5a7d7 | 1638 | pci_disable_device(pdev); |
574e8b95 | 1639 | pci_release_regions(pdev); |
b60503ba MW |
1640 | kfree(dev->queues); |
1641 | kfree(dev->entry); | |
1642 | kfree(dev); | |
1643 | } | |
1644 | ||
1645 | /* These functions are yet to be implemented */ | |
1646 | #define nvme_error_detected NULL | |
1647 | #define nvme_dump_registers NULL | |
1648 | #define nvme_link_reset NULL | |
1649 | #define nvme_slot_reset NULL | |
1650 | #define nvme_error_resume NULL | |
1651 | #define nvme_suspend NULL | |
1652 | #define nvme_resume NULL | |
1653 | ||
1654 | static struct pci_error_handlers nvme_err_handler = { | |
1655 | .error_detected = nvme_error_detected, | |
1656 | .mmio_enabled = nvme_dump_registers, | |
1657 | .link_reset = nvme_link_reset, | |
1658 | .slot_reset = nvme_slot_reset, | |
1659 | .resume = nvme_error_resume, | |
1660 | }; | |
1661 | ||
1662 | /* Move to pci_ids.h later */ | |
1663 | #define PCI_CLASS_STORAGE_EXPRESS 0x010802 | |
1664 | ||
1665 | static DEFINE_PCI_DEVICE_TABLE(nvme_id_table) = { | |
1666 | { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, | |
1667 | { 0, } | |
1668 | }; | |
1669 | MODULE_DEVICE_TABLE(pci, nvme_id_table); | |
1670 | ||
1671 | static struct pci_driver nvme_driver = { | |
1672 | .name = "nvme", | |
1673 | .id_table = nvme_id_table, | |
1674 | .probe = nvme_probe, | |
1675 | .remove = __devexit_p(nvme_remove), | |
1676 | .suspend = nvme_suspend, | |
1677 | .resume = nvme_resume, | |
1678 | .err_handler = &nvme_err_handler, | |
1679 | }; | |
1680 | ||
1681 | static int __init nvme_init(void) | |
1682 | { | |
1fa6aead MW |
1683 | int result = -EBUSY; |
1684 | ||
1685 | nvme_thread = kthread_run(nvme_kthread, NULL, "nvme"); | |
1686 | if (IS_ERR(nvme_thread)) | |
1687 | return PTR_ERR(nvme_thread); | |
b60503ba MW |
1688 | |
1689 | nvme_major = register_blkdev(nvme_major, "nvme"); | |
1690 | if (nvme_major <= 0) | |
1fa6aead | 1691 | goto kill_kthread; |
b60503ba MW |
1692 | |
1693 | result = pci_register_driver(&nvme_driver); | |
1fa6aead MW |
1694 | if (result) |
1695 | goto unregister_blkdev; | |
1696 | return 0; | |
b60503ba | 1697 | |
1fa6aead | 1698 | unregister_blkdev: |
b60503ba | 1699 | unregister_blkdev(nvme_major, "nvme"); |
1fa6aead MW |
1700 | kill_kthread: |
1701 | kthread_stop(nvme_thread); | |
b60503ba MW |
1702 | return result; |
1703 | } | |
1704 | ||
1705 | static void __exit nvme_exit(void) | |
1706 | { | |
1707 | pci_unregister_driver(&nvme_driver); | |
1708 | unregister_blkdev(nvme_major, "nvme"); | |
1fa6aead | 1709 | kthread_stop(nvme_thread); |
b60503ba MW |
1710 | } |
1711 | ||
1712 | MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); | |
1713 | MODULE_LICENSE("GPL"); | |
ce38c149 | 1714 | MODULE_VERSION("0.7"); |
b60503ba MW |
1715 | module_init(nvme_init); |
1716 | module_exit(nvme_exit); |