Merge branch 'for-4.5/nvme' of git://git.kernel.dk/linux-block
[deliverable/linux.git] / drivers / nvme / host / lightnvm.c
1 /*
2 * nvme-lightnvm.c - LightNVM NVMe device
3 *
4 * Copyright (C) 2014-2015 IT University of Copenhagen
5 * Initial release: Matias Bjorling <mb@lightnvm.io>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version
9 * 2 as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; see the file COPYING. If not, write to
18 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
19 * USA.
20 *
21 */
22
23 #include "nvme.h"
24
25 #include <linux/nvme.h>
26 #include <linux/bitops.h>
27 #include <linux/lightnvm.h>
28 #include <linux/vmalloc.h>
29
30 enum nvme_nvm_admin_opcode {
31 nvme_nvm_admin_identity = 0xe2,
32 nvme_nvm_admin_get_l2p_tbl = 0xea,
33 nvme_nvm_admin_get_bb_tbl = 0xf2,
34 nvme_nvm_admin_set_bb_tbl = 0xf1,
35 };
36
37 struct nvme_nvm_hb_rw {
38 __u8 opcode;
39 __u8 flags;
40 __u16 command_id;
41 __le32 nsid;
42 __u64 rsvd2;
43 __le64 metadata;
44 __le64 prp1;
45 __le64 prp2;
46 __le64 spba;
47 __le16 length;
48 __le16 control;
49 __le32 dsmgmt;
50 __le64 slba;
51 };
52
53 struct nvme_nvm_ph_rw {
54 __u8 opcode;
55 __u8 flags;
56 __u16 command_id;
57 __le32 nsid;
58 __u64 rsvd2;
59 __le64 metadata;
60 __le64 prp1;
61 __le64 prp2;
62 __le64 spba;
63 __le16 length;
64 __le16 control;
65 __le32 dsmgmt;
66 __le64 resv;
67 };
68
69 struct nvme_nvm_identity {
70 __u8 opcode;
71 __u8 flags;
72 __u16 command_id;
73 __le32 nsid;
74 __u64 rsvd[2];
75 __le64 prp1;
76 __le64 prp2;
77 __le32 chnl_off;
78 __u32 rsvd11[5];
79 };
80
81 struct nvme_nvm_l2ptbl {
82 __u8 opcode;
83 __u8 flags;
84 __u16 command_id;
85 __le32 nsid;
86 __le32 cdw2[4];
87 __le64 prp1;
88 __le64 prp2;
89 __le64 slba;
90 __le32 nlb;
91 __le16 cdw14[6];
92 };
93
94 struct nvme_nvm_getbbtbl {
95 __u8 opcode;
96 __u8 flags;
97 __u16 command_id;
98 __le32 nsid;
99 __u64 rsvd[2];
100 __le64 prp1;
101 __le64 prp2;
102 __le64 spba;
103 __u32 rsvd4[4];
104 };
105
106 struct nvme_nvm_setbbtbl {
107 __u8 opcode;
108 __u8 flags;
109 __u16 command_id;
110 __le32 nsid;
111 __le64 rsvd[2];
112 __le64 prp1;
113 __le64 prp2;
114 __le64 spba;
115 __le16 nlb;
116 __u8 value;
117 __u8 rsvd3;
118 __u32 rsvd4[3];
119 };
120
121 struct nvme_nvm_erase_blk {
122 __u8 opcode;
123 __u8 flags;
124 __u16 command_id;
125 __le32 nsid;
126 __u64 rsvd[2];
127 __le64 prp1;
128 __le64 prp2;
129 __le64 spba;
130 __le16 length;
131 __le16 control;
132 __le32 dsmgmt;
133 __le64 resv;
134 };
135
136 struct nvme_nvm_command {
137 union {
138 struct nvme_common_command common;
139 struct nvme_nvm_identity identity;
140 struct nvme_nvm_hb_rw hb_rw;
141 struct nvme_nvm_ph_rw ph_rw;
142 struct nvme_nvm_l2ptbl l2p;
143 struct nvme_nvm_getbbtbl get_bb;
144 struct nvme_nvm_setbbtbl set_bb;
145 struct nvme_nvm_erase_blk erase;
146 };
147 };
148
149 struct nvme_nvm_lp_mlc {
150 __u16 num_pairs;
151 __u8 pairs[886];
152 };
153
154 struct nvme_nvm_lp_tbl {
155 __u8 id[8];
156 struct nvme_nvm_lp_mlc mlc;
157 };
158
159 struct nvme_nvm_id_group {
160 __u8 mtype;
161 __u8 fmtype;
162 __le16 res16;
163 __u8 num_ch;
164 __u8 num_lun;
165 __u8 num_pln;
166 __u8 rsvd1;
167 __le16 num_blk;
168 __le16 num_pg;
169 __le16 fpg_sz;
170 __le16 csecs;
171 __le16 sos;
172 __le16 rsvd2;
173 __le32 trdt;
174 __le32 trdm;
175 __le32 tprt;
176 __le32 tprm;
177 __le32 tbet;
178 __le32 tbem;
179 __le32 mpos;
180 __le32 mccap;
181 __le16 cpar;
182 __u8 reserved[10];
183 struct nvme_nvm_lp_tbl lptbl;
184 } __packed;
185
186 struct nvme_nvm_addr_format {
187 __u8 ch_offset;
188 __u8 ch_len;
189 __u8 lun_offset;
190 __u8 lun_len;
191 __u8 pln_offset;
192 __u8 pln_len;
193 __u8 blk_offset;
194 __u8 blk_len;
195 __u8 pg_offset;
196 __u8 pg_len;
197 __u8 sect_offset;
198 __u8 sect_len;
199 __u8 res[4];
200 } __packed;
201
202 struct nvme_nvm_id {
203 __u8 ver_id;
204 __u8 vmnt;
205 __u8 cgrps;
206 __u8 res;
207 __le32 cap;
208 __le32 dom;
209 struct nvme_nvm_addr_format ppaf;
210 __u8 resv[228];
211 struct nvme_nvm_id_group groups[4];
212 } __packed;
213
214 struct nvme_nvm_bb_tbl {
215 __u8 tblid[4];
216 __le16 verid;
217 __le16 revid;
218 __le32 rvsd1;
219 __le32 tblks;
220 __le32 tfact;
221 __le32 tgrown;
222 __le32 tdresv;
223 __le32 thresv;
224 __le32 rsvd2[8];
225 __u8 blk[0];
226 };
227
228 /*
229 * Check we didn't inadvertently grow the command struct
230 */
231 static inline void _nvme_nvm_check_size(void)
232 {
233 BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
234 BUILD_BUG_ON(sizeof(struct nvme_nvm_hb_rw) != 64);
235 BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
236 BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
237 BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
238 BUILD_BUG_ON(sizeof(struct nvme_nvm_l2ptbl) != 64);
239 BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
240 BUILD_BUG_ON(sizeof(struct nvme_nvm_id_group) != 960);
241 BUILD_BUG_ON(sizeof(struct nvme_nvm_addr_format) != 128);
242 BUILD_BUG_ON(sizeof(struct nvme_nvm_id) != 4096);
243 BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 512);
244 }
245
246 static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
247 {
248 struct nvme_nvm_id_group *src;
249 struct nvm_id_group *dst;
250 int i, end;
251
252 end = min_t(u32, 4, nvm_id->cgrps);
253
254 for (i = 0; i < end; i++) {
255 src = &nvme_nvm_id->groups[i];
256 dst = &nvm_id->groups[i];
257
258 dst->mtype = src->mtype;
259 dst->fmtype = src->fmtype;
260 dst->num_ch = src->num_ch;
261 dst->num_lun = src->num_lun;
262 dst->num_pln = src->num_pln;
263
264 dst->num_pg = le16_to_cpu(src->num_pg);
265 dst->num_blk = le16_to_cpu(src->num_blk);
266 dst->fpg_sz = le16_to_cpu(src->fpg_sz);
267 dst->csecs = le16_to_cpu(src->csecs);
268 dst->sos = le16_to_cpu(src->sos);
269
270 dst->trdt = le32_to_cpu(src->trdt);
271 dst->trdm = le32_to_cpu(src->trdm);
272 dst->tprt = le32_to_cpu(src->tprt);
273 dst->tprm = le32_to_cpu(src->tprm);
274 dst->tbet = le32_to_cpu(src->tbet);
275 dst->tbem = le32_to_cpu(src->tbem);
276 dst->mpos = le32_to_cpu(src->mpos);
277 dst->mccap = le32_to_cpu(src->mccap);
278
279 dst->cpar = le16_to_cpu(src->cpar);
280
281 if (dst->fmtype == NVM_ID_FMTYPE_MLC) {
282 memcpy(dst->lptbl.id, src->lptbl.id, 8);
283 dst->lptbl.mlc.num_pairs =
284 le16_to_cpu(src->lptbl.mlc.num_pairs);
285 /* 4 bits per pair */
286 memcpy(dst->lptbl.mlc.pairs, src->lptbl.mlc.pairs,
287 dst->lptbl.mlc.num_pairs >> 1);
288 }
289 }
290
291 return 0;
292 }
293
294 static int nvme_nvm_identity(struct nvm_dev *nvmdev, struct nvm_id *nvm_id)
295 {
296 struct nvme_ns *ns = nvmdev->q->queuedata;
297 struct nvme_nvm_id *nvme_nvm_id;
298 struct nvme_nvm_command c = {};
299 int ret;
300
301 c.identity.opcode = nvme_nvm_admin_identity;
302 c.identity.nsid = cpu_to_le32(ns->ns_id);
303 c.identity.chnl_off = 0;
304
305 nvme_nvm_id = kmalloc(sizeof(struct nvme_nvm_id), GFP_KERNEL);
306 if (!nvme_nvm_id)
307 return -ENOMEM;
308
309 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
310 nvme_nvm_id, sizeof(struct nvme_nvm_id));
311 if (ret) {
312 ret = -EIO;
313 goto out;
314 }
315
316 nvm_id->ver_id = nvme_nvm_id->ver_id;
317 nvm_id->vmnt = nvme_nvm_id->vmnt;
318 nvm_id->cgrps = nvme_nvm_id->cgrps;
319 nvm_id->cap = le32_to_cpu(nvme_nvm_id->cap);
320 nvm_id->dom = le32_to_cpu(nvme_nvm_id->dom);
321 memcpy(&nvm_id->ppaf, &nvme_nvm_id->ppaf,
322 sizeof(struct nvme_nvm_addr_format));
323
324 ret = init_grps(nvm_id, nvme_nvm_id);
325 out:
326 kfree(nvme_nvm_id);
327 return ret;
328 }
329
330 static int nvme_nvm_get_l2p_tbl(struct nvm_dev *nvmdev, u64 slba, u32 nlb,
331 nvm_l2p_update_fn *update_l2p, void *priv)
332 {
333 struct nvme_ns *ns = nvmdev->q->queuedata;
334 struct nvme_nvm_command c = {};
335 u32 len = queue_max_hw_sectors(ns->ctrl->admin_q) << 9;
336 u32 nlb_pr_rq = len / sizeof(u64);
337 u64 cmd_slba = slba;
338 void *entries;
339 int ret = 0;
340
341 c.l2p.opcode = nvme_nvm_admin_get_l2p_tbl;
342 c.l2p.nsid = cpu_to_le32(ns->ns_id);
343 entries = kmalloc(len, GFP_KERNEL);
344 if (!entries)
345 return -ENOMEM;
346
347 while (nlb) {
348 u32 cmd_nlb = min(nlb_pr_rq, nlb);
349
350 c.l2p.slba = cpu_to_le64(cmd_slba);
351 c.l2p.nlb = cpu_to_le32(cmd_nlb);
352
353 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q,
354 (struct nvme_command *)&c, entries, len);
355 if (ret) {
356 dev_err(ns->ctrl->dev, "L2P table transfer failed (%d)\n",
357 ret);
358 ret = -EIO;
359 goto out;
360 }
361
362 if (update_l2p(cmd_slba, cmd_nlb, entries, priv)) {
363 ret = -EINTR;
364 goto out;
365 }
366
367 cmd_slba += cmd_nlb;
368 nlb -= cmd_nlb;
369 }
370
371 out:
372 kfree(entries);
373 return ret;
374 }
375
376 static int nvme_nvm_get_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr ppa,
377 int nr_blocks, nvm_bb_update_fn *update_bbtbl,
378 void *priv)
379 {
380 struct request_queue *q = nvmdev->q;
381 struct nvme_ns *ns = q->queuedata;
382 struct nvme_ctrl *ctrl = ns->ctrl;
383 struct nvme_nvm_command c = {};
384 struct nvme_nvm_bb_tbl *bb_tbl;
385 int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blocks;
386 int ret = 0;
387
388 c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
389 c.get_bb.nsid = cpu_to_le32(ns->ns_id);
390 c.get_bb.spba = cpu_to_le64(ppa.ppa);
391
392 bb_tbl = kzalloc(tblsz, GFP_KERNEL);
393 if (!bb_tbl)
394 return -ENOMEM;
395
396 ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
397 bb_tbl, tblsz);
398 if (ret) {
399 dev_err(ctrl->dev, "get bad block table failed (%d)\n", ret);
400 ret = -EIO;
401 goto out;
402 }
403
404 if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
405 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
406 dev_err(ctrl->dev, "bbt format mismatch\n");
407 ret = -EINVAL;
408 goto out;
409 }
410
411 if (le16_to_cpu(bb_tbl->verid) != 1) {
412 ret = -EINVAL;
413 dev_err(ctrl->dev, "bbt version not supported\n");
414 goto out;
415 }
416
417 if (le32_to_cpu(bb_tbl->tblks) != nr_blocks) {
418 ret = -EINVAL;
419 dev_err(ctrl->dev, "bbt unsuspected blocks returned (%u!=%u)",
420 le32_to_cpu(bb_tbl->tblks), nr_blocks);
421 goto out;
422 }
423
424 ppa = dev_to_generic_addr(nvmdev, ppa);
425 ret = update_bbtbl(ppa, nr_blocks, bb_tbl->blk, priv);
426 out:
427 kfree(bb_tbl);
428 return ret;
429 }
430
431 static int nvme_nvm_set_bb_tbl(struct nvm_dev *nvmdev, struct nvm_rq *rqd,
432 int type)
433 {
434 struct nvme_ns *ns = nvmdev->q->queuedata;
435 struct nvme_nvm_command c = {};
436 int ret = 0;
437
438 c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
439 c.set_bb.nsid = cpu_to_le32(ns->ns_id);
440 c.set_bb.spba = cpu_to_le64(rqd->ppa_addr.ppa);
441 c.set_bb.nlb = cpu_to_le16(rqd->nr_pages - 1);
442 c.set_bb.value = type;
443
444 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
445 NULL, 0);
446 if (ret)
447 dev_err(ns->ctrl->dev, "set bad block table failed (%d)\n", ret);
448 return ret;
449 }
450
451 static inline void nvme_nvm_rqtocmd(struct request *rq, struct nvm_rq *rqd,
452 struct nvme_ns *ns, struct nvme_nvm_command *c)
453 {
454 c->ph_rw.opcode = rqd->opcode;
455 c->ph_rw.nsid = cpu_to_le32(ns->ns_id);
456 c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
457 c->ph_rw.control = cpu_to_le16(rqd->flags);
458 c->ph_rw.length = cpu_to_le16(rqd->nr_pages - 1);
459
460 if (rqd->opcode == NVM_OP_HBWRITE || rqd->opcode == NVM_OP_HBREAD)
461 c->hb_rw.slba = cpu_to_le64(nvme_block_nr(ns,
462 rqd->bio->bi_iter.bi_sector));
463 }
464
465 static void nvme_nvm_end_io(struct request *rq, int error)
466 {
467 struct nvm_rq *rqd = rq->end_io_data;
468
469 nvm_end_io(rqd, error);
470
471 kfree(rq->cmd);
472 blk_mq_free_request(rq);
473 }
474
475 static int nvme_nvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
476 {
477 struct request_queue *q = dev->q;
478 struct nvme_ns *ns = q->queuedata;
479 struct request *rq;
480 struct bio *bio = rqd->bio;
481 struct nvme_nvm_command *cmd;
482
483 rq = blk_mq_alloc_request(q, bio_rw(bio), 0);
484 if (IS_ERR(rq))
485 return -ENOMEM;
486
487 cmd = kzalloc(sizeof(struct nvme_nvm_command), GFP_KERNEL);
488 if (!cmd) {
489 blk_mq_free_request(rq);
490 return -ENOMEM;
491 }
492
493 rq->cmd_type = REQ_TYPE_DRV_PRIV;
494 rq->ioprio = bio_prio(bio);
495
496 if (bio_has_data(bio))
497 rq->nr_phys_segments = bio_phys_segments(q, bio);
498
499 rq->__data_len = bio->bi_iter.bi_size;
500 rq->bio = rq->biotail = bio;
501
502 nvme_nvm_rqtocmd(rq, rqd, ns, cmd);
503
504 rq->cmd = (unsigned char *)cmd;
505 rq->cmd_len = sizeof(struct nvme_nvm_command);
506 rq->special = (void *)0;
507
508 rq->end_io_data = rqd;
509
510 blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
511
512 return 0;
513 }
514
515 static int nvme_nvm_erase_block(struct nvm_dev *dev, struct nvm_rq *rqd)
516 {
517 struct request_queue *q = dev->q;
518 struct nvme_ns *ns = q->queuedata;
519 struct nvme_nvm_command c = {};
520
521 c.erase.opcode = NVM_OP_ERASE;
522 c.erase.nsid = cpu_to_le32(ns->ns_id);
523 c.erase.spba = cpu_to_le64(rqd->ppa_addr.ppa);
524 c.erase.length = cpu_to_le16(rqd->nr_pages - 1);
525
526 return nvme_submit_sync_cmd(q, (struct nvme_command *)&c, NULL, 0);
527 }
528
529 static void *nvme_nvm_create_dma_pool(struct nvm_dev *nvmdev, char *name)
530 {
531 struct nvme_ns *ns = nvmdev->q->queuedata;
532
533 return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
534 }
535
536 static void nvme_nvm_destroy_dma_pool(void *pool)
537 {
538 struct dma_pool *dma_pool = pool;
539
540 dma_pool_destroy(dma_pool);
541 }
542
543 static void *nvme_nvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
544 gfp_t mem_flags, dma_addr_t *dma_handler)
545 {
546 return dma_pool_alloc(pool, mem_flags, dma_handler);
547 }
548
549 static void nvme_nvm_dev_dma_free(void *pool, void *ppa_list,
550 dma_addr_t dma_handler)
551 {
552 dma_pool_free(pool, ppa_list, dma_handler);
553 }
554
555 static struct nvm_dev_ops nvme_nvm_dev_ops = {
556 .identity = nvme_nvm_identity,
557
558 .get_l2p_tbl = nvme_nvm_get_l2p_tbl,
559
560 .get_bb_tbl = nvme_nvm_get_bb_tbl,
561 .set_bb_tbl = nvme_nvm_set_bb_tbl,
562
563 .submit_io = nvme_nvm_submit_io,
564 .erase_block = nvme_nvm_erase_block,
565
566 .create_dma_pool = nvme_nvm_create_dma_pool,
567 .destroy_dma_pool = nvme_nvm_destroy_dma_pool,
568 .dev_dma_alloc = nvme_nvm_dev_dma_alloc,
569 .dev_dma_free = nvme_nvm_dev_dma_free,
570
571 .max_phys_sect = 64,
572 };
573
574 int nvme_nvm_register(struct request_queue *q, char *disk_name)
575 {
576 return nvm_register(q, disk_name, &nvme_nvm_dev_ops);
577 }
578
579 void nvme_nvm_unregister(struct request_queue *q, char *disk_name)
580 {
581 nvm_unregister(disk_name);
582 }
583
584 /* move to shared place when used in multiple places. */
585 #define PCI_VENDOR_ID_CNEX 0x1d1d
586 #define PCI_DEVICE_ID_CNEX_WL 0x2807
587 #define PCI_DEVICE_ID_CNEX_QEMU 0x1f1f
588
589 int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
590 {
591 struct nvme_ctrl *ctrl = ns->ctrl;
592 /* XXX: this is poking into PCI structures from generic code! */
593 struct pci_dev *pdev = to_pci_dev(ctrl->dev);
594
595 /* QEMU NVMe simulator - PCI ID + Vendor specific bit */
596 if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
597 pdev->device == PCI_DEVICE_ID_CNEX_QEMU &&
598 id->vs[0] == 0x1)
599 return 1;
600
601 /* CNEX Labs - PCI ID + Vendor specific bit */
602 if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
603 pdev->device == PCI_DEVICE_ID_CNEX_WL &&
604 id->vs[0] == 0x1)
605 return 1;
606
607 return 0;
608 }
This page took 0.04243 seconds and 5 git commands to generate.