1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
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
30enum 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
37struct 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
53struct 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
69struct 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
81struct 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
94struct 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
106struct 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
121struct 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
136struct 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
149struct nvme_nvm_completion {
150 __le64 result;
151 __le16 sq_head;
152 __le16 sq_id;
153 __u16 command_id;
154 __le16 status;
155};
156
157#define NVME_NVM_LP_MLC_PAIRS 886
158struct nvme_nvm_lp_mlc {
159 __le16 num_pairs;
160 __u8 pairs[NVME_NVM_LP_MLC_PAIRS];
161};
162
163struct nvme_nvm_lp_tbl {
164 __u8 id[8];
165 struct nvme_nvm_lp_mlc mlc;
166};
167
168struct nvme_nvm_id_group {
169 __u8 mtype;
170 __u8 fmtype;
171 __le16 res16;
172 __u8 num_ch;
173 __u8 num_lun;
174 __u8 num_pln;
175 __u8 rsvd1;
176 __le16 num_blk;
177 __le16 num_pg;
178 __le16 fpg_sz;
179 __le16 csecs;
180 __le16 sos;
181 __le16 rsvd2;
182 __le32 trdt;
183 __le32 trdm;
184 __le32 tprt;
185 __le32 tprm;
186 __le32 tbet;
187 __le32 tbem;
188 __le32 mpos;
189 __le32 mccap;
190 __le16 cpar;
191 __u8 reserved[10];
192 struct nvme_nvm_lp_tbl lptbl;
193} __packed;
194
195struct nvme_nvm_addr_format {
196 __u8 ch_offset;
197 __u8 ch_len;
198 __u8 lun_offset;
199 __u8 lun_len;
200 __u8 pln_offset;
201 __u8 pln_len;
202 __u8 blk_offset;
203 __u8 blk_len;
204 __u8 pg_offset;
205 __u8 pg_len;
206 __u8 sect_offset;
207 __u8 sect_len;
208 __u8 res[4];
209} __packed;
210
211struct nvme_nvm_id {
212 __u8 ver_id;
213 __u8 vmnt;
214 __u8 cgrps;
215 __u8 res;
216 __le32 cap;
217 __le32 dom;
218 struct nvme_nvm_addr_format ppaf;
219 __u8 resv[228];
220 struct nvme_nvm_id_group groups[4];
221} __packed;
222
223struct nvme_nvm_bb_tbl {
224 __u8 tblid[4];
225 __le16 verid;
226 __le16 revid;
227 __le32 rvsd1;
228 __le32 tblks;
229 __le32 tfact;
230 __le32 tgrown;
231 __le32 tdresv;
232 __le32 thresv;
233 __le32 rsvd2[8];
234 __u8 blk[0];
235};
236
237
238
239
240static inline void _nvme_nvm_check_size(void)
241{
242 BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
243 BUILD_BUG_ON(sizeof(struct nvme_nvm_hb_rw) != 64);
244 BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
245 BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
246 BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
247 BUILD_BUG_ON(sizeof(struct nvme_nvm_l2ptbl) != 64);
248 BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
249 BUILD_BUG_ON(sizeof(struct nvme_nvm_id_group) != 960);
250 BUILD_BUG_ON(sizeof(struct nvme_nvm_addr_format) != 128);
251 BUILD_BUG_ON(sizeof(struct nvme_nvm_id) != 4096);
252 BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 512);
253}
254
255static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
256{
257 struct nvme_nvm_id_group *src;
258 struct nvm_id_group *dst;
259 int i, end;
260
261 end = min_t(u32, 4, nvm_id->cgrps);
262
263 for (i = 0; i < end; i++) {
264 src = &nvme_nvm_id->groups[i];
265 dst = &nvm_id->groups[i];
266
267 dst->mtype = src->mtype;
268 dst->fmtype = src->fmtype;
269 dst->num_ch = src->num_ch;
270 dst->num_lun = src->num_lun;
271 dst->num_pln = src->num_pln;
272
273 dst->num_pg = le16_to_cpu(src->num_pg);
274 dst->num_blk = le16_to_cpu(src->num_blk);
275 dst->fpg_sz = le16_to_cpu(src->fpg_sz);
276 dst->csecs = le16_to_cpu(src->csecs);
277 dst->sos = le16_to_cpu(src->sos);
278
279 dst->trdt = le32_to_cpu(src->trdt);
280 dst->trdm = le32_to_cpu(src->trdm);
281 dst->tprt = le32_to_cpu(src->tprt);
282 dst->tprm = le32_to_cpu(src->tprm);
283 dst->tbet = le32_to_cpu(src->tbet);
284 dst->tbem = le32_to_cpu(src->tbem);
285 dst->mpos = le32_to_cpu(src->mpos);
286 dst->mccap = le32_to_cpu(src->mccap);
287
288 dst->cpar = le16_to_cpu(src->cpar);
289
290 if (dst->fmtype == NVM_ID_FMTYPE_MLC) {
291 memcpy(dst->lptbl.id, src->lptbl.id, 8);
292 dst->lptbl.mlc.num_pairs =
293 le16_to_cpu(src->lptbl.mlc.num_pairs);
294
295 if (dst->lptbl.mlc.num_pairs > NVME_NVM_LP_MLC_PAIRS) {
296 pr_err("nvm: number of MLC pairs not supported\n");
297 return -EINVAL;
298 }
299
300 memcpy(dst->lptbl.mlc.pairs, src->lptbl.mlc.pairs,
301 dst->lptbl.mlc.num_pairs);
302 }
303 }
304
305 return 0;
306}
307
308static int nvme_nvm_identity(struct nvm_dev *nvmdev, struct nvm_id *nvm_id)
309{
310 struct nvme_ns *ns = nvmdev->q->queuedata;
311 struct nvme_nvm_id *nvme_nvm_id;
312 struct nvme_nvm_command c = {};
313 int ret;
314
315 c.identity.opcode = nvme_nvm_admin_identity;
316 c.identity.nsid = cpu_to_le32(ns->ns_id);
317 c.identity.chnl_off = 0;
318
319 nvme_nvm_id = kmalloc(sizeof(struct nvme_nvm_id), GFP_KERNEL);
320 if (!nvme_nvm_id)
321 return -ENOMEM;
322
323 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
324 nvme_nvm_id, sizeof(struct nvme_nvm_id));
325 if (ret) {
326 ret = -EIO;
327 goto out;
328 }
329
330 nvm_id->ver_id = nvme_nvm_id->ver_id;
331 nvm_id->vmnt = nvme_nvm_id->vmnt;
332 nvm_id->cgrps = nvme_nvm_id->cgrps;
333 nvm_id->cap = le32_to_cpu(nvme_nvm_id->cap);
334 nvm_id->dom = le32_to_cpu(nvme_nvm_id->dom);
335 memcpy(&nvm_id->ppaf, &nvme_nvm_id->ppaf,
336 sizeof(struct nvme_nvm_addr_format));
337
338 ret = init_grps(nvm_id, nvme_nvm_id);
339out:
340 kfree(nvme_nvm_id);
341 return ret;
342}
343
344static int nvme_nvm_get_l2p_tbl(struct nvm_dev *nvmdev, u64 slba, u32 nlb,
345 nvm_l2p_update_fn *update_l2p, void *priv)
346{
347 struct nvme_ns *ns = nvmdev->q->queuedata;
348 struct nvme_nvm_command c = {};
349 u32 len = queue_max_hw_sectors(ns->ctrl->admin_q) << 9;
350 u32 nlb_pr_rq = len / sizeof(u64);
351 u64 cmd_slba = slba;
352 void *entries;
353 int ret = 0;
354
355 c.l2p.opcode = nvme_nvm_admin_get_l2p_tbl;
356 c.l2p.nsid = cpu_to_le32(ns->ns_id);
357 entries = kmalloc(len, GFP_KERNEL);
358 if (!entries)
359 return -ENOMEM;
360
361 while (nlb) {
362 u32 cmd_nlb = min(nlb_pr_rq, nlb);
363
364 c.l2p.slba = cpu_to_le64(cmd_slba);
365 c.l2p.nlb = cpu_to_le32(cmd_nlb);
366
367 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q,
368 (struct nvme_command *)&c, entries, len);
369 if (ret) {
370 dev_err(ns->ctrl->device,
371 "L2P table transfer failed (%d)\n", ret);
372 ret = -EIO;
373 goto out;
374 }
375
376 if (update_l2p(cmd_slba, cmd_nlb, entries, priv)) {
377 ret = -EINTR;
378 goto out;
379 }
380
381 cmd_slba += cmd_nlb;
382 nlb -= cmd_nlb;
383 }
384
385out:
386 kfree(entries);
387 return ret;
388}
389
390static int nvme_nvm_get_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr ppa,
391 u8 *blks)
392{
393 struct request_queue *q = nvmdev->q;
394 struct nvme_ns *ns = q->queuedata;
395 struct nvme_ctrl *ctrl = ns->ctrl;
396 struct nvme_nvm_command c = {};
397 struct nvme_nvm_bb_tbl *bb_tbl;
398 int nr_blks = nvmdev->blks_per_lun * nvmdev->plane_mode;
399 int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blks;
400 int ret = 0;
401
402 c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
403 c.get_bb.nsid = cpu_to_le32(ns->ns_id);
404 c.get_bb.spba = cpu_to_le64(ppa.ppa);
405
406 bb_tbl = kzalloc(tblsz, GFP_KERNEL);
407 if (!bb_tbl)
408 return -ENOMEM;
409
410 ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
411 bb_tbl, tblsz);
412 if (ret) {
413 dev_err(ctrl->device, "get bad block table failed (%d)\n", ret);
414 ret = -EIO;
415 goto out;
416 }
417
418 if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
419 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
420 dev_err(ctrl->device, "bbt format mismatch\n");
421 ret = -EINVAL;
422 goto out;
423 }
424
425 if (le16_to_cpu(bb_tbl->verid) != 1) {
426 ret = -EINVAL;
427 dev_err(ctrl->device, "bbt version not supported\n");
428 goto out;
429 }
430
431 if (le32_to_cpu(bb_tbl->tblks) != nr_blks) {
432 ret = -EINVAL;
433 dev_err(ctrl->device,
434 "bbt unsuspected blocks returned (%u!=%u)",
435 le32_to_cpu(bb_tbl->tblks), nr_blks);
436 goto out;
437 }
438
439 memcpy(blks, bb_tbl->blk, nvmdev->blks_per_lun * nvmdev->plane_mode);
440out:
441 kfree(bb_tbl);
442 return ret;
443}
444
445static int nvme_nvm_set_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr *ppas,
446 int nr_ppas, int type)
447{
448 struct nvme_ns *ns = nvmdev->q->queuedata;
449 struct nvme_nvm_command c = {};
450 int ret = 0;
451
452 c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
453 c.set_bb.nsid = cpu_to_le32(ns->ns_id);
454 c.set_bb.spba = cpu_to_le64(ppas->ppa);
455 c.set_bb.nlb = cpu_to_le16(nr_ppas - 1);
456 c.set_bb.value = type;
457
458 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
459 NULL, 0);
460 if (ret)
461 dev_err(ns->ctrl->device, "set bad block table failed (%d)\n",
462 ret);
463 return ret;
464}
465
466static inline void nvme_nvm_rqtocmd(struct request *rq, struct nvm_rq *rqd,
467 struct nvme_ns *ns, struct nvme_nvm_command *c)
468{
469 c->ph_rw.opcode = rqd->opcode;
470 c->ph_rw.nsid = cpu_to_le32(ns->ns_id);
471 c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
472 c->ph_rw.metadata = cpu_to_le64(rqd->dma_meta_list);
473 c->ph_rw.control = cpu_to_le16(rqd->flags);
474 c->ph_rw.length = cpu_to_le16(rqd->nr_ppas - 1);
475
476 if (rqd->opcode == NVM_OP_HBWRITE || rqd->opcode == NVM_OP_HBREAD)
477 c->hb_rw.slba = cpu_to_le64(nvme_block_nr(ns,
478 rqd->bio->bi_iter.bi_sector));
479}
480
481static void nvme_nvm_end_io(struct request *rq, int error)
482{
483 struct nvm_rq *rqd = rq->end_io_data;
484 struct nvme_nvm_completion *cqe = rq->special;
485
486 if (cqe)
487 rqd->ppa_status = le64_to_cpu(cqe->result);
488
489 nvm_end_io(rqd, error);
490
491 kfree(rq->cmd);
492 blk_mq_free_request(rq);
493}
494
495static int nvme_nvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
496{
497 struct request_queue *q = dev->q;
498 struct nvme_ns *ns = q->queuedata;
499 struct request *rq;
500 struct bio *bio = rqd->bio;
501 struct nvme_nvm_command *cmd;
502
503 rq = blk_mq_alloc_request(q, bio_data_dir(bio), 0);
504 if (IS_ERR(rq))
505 return -ENOMEM;
506
507 cmd = kzalloc(sizeof(struct nvme_nvm_command) +
508 sizeof(struct nvme_nvm_completion), GFP_KERNEL);
509 if (!cmd) {
510 blk_mq_free_request(rq);
511 return -ENOMEM;
512 }
513
514 rq->cmd_type = REQ_TYPE_DRV_PRIV;
515 rq->ioprio = bio_prio(bio);
516
517 if (bio_has_data(bio))
518 rq->nr_phys_segments = bio_phys_segments(q, bio);
519
520 rq->__data_len = bio->bi_iter.bi_size;
521 rq->bio = rq->biotail = bio;
522
523 nvme_nvm_rqtocmd(rq, rqd, ns, cmd);
524
525 rq->cmd = (unsigned char *)cmd;
526 rq->cmd_len = sizeof(struct nvme_nvm_command);
527 rq->special = cmd + 1;
528
529 rq->end_io_data = rqd;
530
531 blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
532
533 return 0;
534}
535
536static int nvme_nvm_erase_block(struct nvm_dev *dev, struct nvm_rq *rqd)
537{
538 struct request_queue *q = dev->q;
539 struct nvme_ns *ns = q->queuedata;
540 struct nvme_nvm_command c = {};
541
542 c.erase.opcode = NVM_OP_ERASE;
543 c.erase.nsid = cpu_to_le32(ns->ns_id);
544 c.erase.spba = cpu_to_le64(rqd->ppa_addr.ppa);
545 c.erase.length = cpu_to_le16(rqd->nr_ppas - 1);
546
547 return nvme_submit_sync_cmd(q, (struct nvme_command *)&c, NULL, 0);
548}
549
550static void *nvme_nvm_create_dma_pool(struct nvm_dev *nvmdev, char *name)
551{
552 struct nvme_ns *ns = nvmdev->q->queuedata;
553
554 return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
555}
556
557static void nvme_nvm_destroy_dma_pool(void *pool)
558{
559 struct dma_pool *dma_pool = pool;
560
561 dma_pool_destroy(dma_pool);
562}
563
564static void *nvme_nvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
565 gfp_t mem_flags, dma_addr_t *dma_handler)
566{
567 return dma_pool_alloc(pool, mem_flags, dma_handler);
568}
569
570static void nvme_nvm_dev_dma_free(void *pool, void *addr,
571 dma_addr_t dma_handler)
572{
573 dma_pool_free(pool, addr, dma_handler);
574}
575
576static struct nvm_dev_ops nvme_nvm_dev_ops = {
577 .identity = nvme_nvm_identity,
578
579 .get_l2p_tbl = nvme_nvm_get_l2p_tbl,
580
581 .get_bb_tbl = nvme_nvm_get_bb_tbl,
582 .set_bb_tbl = nvme_nvm_set_bb_tbl,
583
584 .submit_io = nvme_nvm_submit_io,
585 .erase_block = nvme_nvm_erase_block,
586
587 .create_dma_pool = nvme_nvm_create_dma_pool,
588 .destroy_dma_pool = nvme_nvm_destroy_dma_pool,
589 .dev_dma_alloc = nvme_nvm_dev_dma_alloc,
590 .dev_dma_free = nvme_nvm_dev_dma_free,
591
592 .max_phys_sect = 64,
593};
594
595int nvme_nvm_register(struct request_queue *q, char *disk_name)
596{
597 return nvm_register(q, disk_name, &nvme_nvm_dev_ops);
598}
599
600void nvme_nvm_unregister(struct request_queue *q, char *disk_name)
601{
602 nvm_unregister(disk_name);
603}
604
605
606#define PCI_VENDOR_ID_CNEX 0x1d1d
607#define PCI_DEVICE_ID_CNEX_WL 0x2807
608#define PCI_DEVICE_ID_CNEX_QEMU 0x1f1f
609
610int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
611{
612 struct nvme_ctrl *ctrl = ns->ctrl;
613
614 struct pci_dev *pdev = to_pci_dev(ctrl->dev);
615
616
617 if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
618 pdev->device == PCI_DEVICE_ID_CNEX_QEMU &&
619 id->vs[0] == 0x1)
620 return 1;
621
622
623 if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
624 pdev->device == PCI_DEVICE_ID_CNEX_WL &&
625 id->vs[0] == 0x1)
626 return 1;
627
628 return 0;
629}
630
