Merge tag 'docs-4.12-2' of git://git.lwn.net/linux
[sfrench/cifs-2.6.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 #include <linux/sched/sysctl.h>
30 #include <uapi/linux/lightnvm.h>
31
32 enum nvme_nvm_admin_opcode {
33         nvme_nvm_admin_identity         = 0xe2,
34         nvme_nvm_admin_get_l2p_tbl      = 0xea,
35         nvme_nvm_admin_get_bb_tbl       = 0xf2,
36         nvme_nvm_admin_set_bb_tbl       = 0xf1,
37 };
38
39 struct nvme_nvm_hb_rw {
40         __u8                    opcode;
41         __u8                    flags;
42         __u16                   command_id;
43         __le32                  nsid;
44         __u64                   rsvd2;
45         __le64                  metadata;
46         __le64                  prp1;
47         __le64                  prp2;
48         __le64                  spba;
49         __le16                  length;
50         __le16                  control;
51         __le32                  dsmgmt;
52         __le64                  slba;
53 };
54
55 struct nvme_nvm_ph_rw {
56         __u8                    opcode;
57         __u8                    flags;
58         __u16                   command_id;
59         __le32                  nsid;
60         __u64                   rsvd2;
61         __le64                  metadata;
62         __le64                  prp1;
63         __le64                  prp2;
64         __le64                  spba;
65         __le16                  length;
66         __le16                  control;
67         __le32                  dsmgmt;
68         __le64                  resv;
69 };
70
71 struct nvme_nvm_identity {
72         __u8                    opcode;
73         __u8                    flags;
74         __u16                   command_id;
75         __le32                  nsid;
76         __u64                   rsvd[2];
77         __le64                  prp1;
78         __le64                  prp2;
79         __le32                  chnl_off;
80         __u32                   rsvd11[5];
81 };
82
83 struct nvme_nvm_l2ptbl {
84         __u8                    opcode;
85         __u8                    flags;
86         __u16                   command_id;
87         __le32                  nsid;
88         __le32                  cdw2[4];
89         __le64                  prp1;
90         __le64                  prp2;
91         __le64                  slba;
92         __le32                  nlb;
93         __le16                  cdw14[6];
94 };
95
96 struct nvme_nvm_getbbtbl {
97         __u8                    opcode;
98         __u8                    flags;
99         __u16                   command_id;
100         __le32                  nsid;
101         __u64                   rsvd[2];
102         __le64                  prp1;
103         __le64                  prp2;
104         __le64                  spba;
105         __u32                   rsvd4[4];
106 };
107
108 struct nvme_nvm_setbbtbl {
109         __u8                    opcode;
110         __u8                    flags;
111         __u16                   command_id;
112         __le32                  nsid;
113         __le64                  rsvd[2];
114         __le64                  prp1;
115         __le64                  prp2;
116         __le64                  spba;
117         __le16                  nlb;
118         __u8                    value;
119         __u8                    rsvd3;
120         __u32                   rsvd4[3];
121 };
122
123 struct nvme_nvm_erase_blk {
124         __u8                    opcode;
125         __u8                    flags;
126         __u16                   command_id;
127         __le32                  nsid;
128         __u64                   rsvd[2];
129         __le64                  prp1;
130         __le64                  prp2;
131         __le64                  spba;
132         __le16                  length;
133         __le16                  control;
134         __le32                  dsmgmt;
135         __le64                  resv;
136 };
137
138 struct nvme_nvm_command {
139         union {
140                 struct nvme_common_command common;
141                 struct nvme_nvm_identity identity;
142                 struct nvme_nvm_hb_rw hb_rw;
143                 struct nvme_nvm_ph_rw ph_rw;
144                 struct nvme_nvm_l2ptbl l2p;
145                 struct nvme_nvm_getbbtbl get_bb;
146                 struct nvme_nvm_setbbtbl set_bb;
147                 struct nvme_nvm_erase_blk erase;
148         };
149 };
150
151 #define NVME_NVM_LP_MLC_PAIRS 886
152 struct nvme_nvm_lp_mlc {
153         __le16                  num_pairs;
154         __u8                    pairs[NVME_NVM_LP_MLC_PAIRS];
155 };
156
157 struct nvme_nvm_lp_tbl {
158         __u8                    id[8];
159         struct nvme_nvm_lp_mlc  mlc;
160 };
161
162 struct nvme_nvm_id_group {
163         __u8                    mtype;
164         __u8                    fmtype;
165         __le16                  res16;
166         __u8                    num_ch;
167         __u8                    num_lun;
168         __u8                    num_pln;
169         __u8                    rsvd1;
170         __le16                  num_blk;
171         __le16                  num_pg;
172         __le16                  fpg_sz;
173         __le16                  csecs;
174         __le16                  sos;
175         __le16                  rsvd2;
176         __le32                  trdt;
177         __le32                  trdm;
178         __le32                  tprt;
179         __le32                  tprm;
180         __le32                  tbet;
181         __le32                  tbem;
182         __le32                  mpos;
183         __le32                  mccap;
184         __le16                  cpar;
185         __u8                    reserved[10];
186         struct nvme_nvm_lp_tbl lptbl;
187 } __packed;
188
189 struct nvme_nvm_addr_format {
190         __u8                    ch_offset;
191         __u8                    ch_len;
192         __u8                    lun_offset;
193         __u8                    lun_len;
194         __u8                    pln_offset;
195         __u8                    pln_len;
196         __u8                    blk_offset;
197         __u8                    blk_len;
198         __u8                    pg_offset;
199         __u8                    pg_len;
200         __u8                    sect_offset;
201         __u8                    sect_len;
202         __u8                    res[4];
203 } __packed;
204
205 struct nvme_nvm_id {
206         __u8                    ver_id;
207         __u8                    vmnt;
208         __u8                    cgrps;
209         __u8                    res;
210         __le32                  cap;
211         __le32                  dom;
212         struct nvme_nvm_addr_format ppaf;
213         __u8                    resv[228];
214         struct nvme_nvm_id_group groups[4];
215 } __packed;
216
217 struct nvme_nvm_bb_tbl {
218         __u8    tblid[4];
219         __le16  verid;
220         __le16  revid;
221         __le32  rvsd1;
222         __le32  tblks;
223         __le32  tfact;
224         __le32  tgrown;
225         __le32  tdresv;
226         __le32  thresv;
227         __le32  rsvd2[8];
228         __u8    blk[0];
229 };
230
231 /*
232  * Check we didn't inadvertently grow the command struct
233  */
234 static inline void _nvme_nvm_check_size(void)
235 {
236         BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
237         BUILD_BUG_ON(sizeof(struct nvme_nvm_hb_rw) != 64);
238         BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
239         BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
240         BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
241         BUILD_BUG_ON(sizeof(struct nvme_nvm_l2ptbl) != 64);
242         BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
243         BUILD_BUG_ON(sizeof(struct nvme_nvm_id_group) != 960);
244         BUILD_BUG_ON(sizeof(struct nvme_nvm_addr_format) != 16);
245         BUILD_BUG_ON(sizeof(struct nvme_nvm_id) != 4096);
246         BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 64);
247 }
248
249 static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
250 {
251         struct nvme_nvm_id_group *src;
252         struct nvm_id_group *dst;
253
254         if (nvme_nvm_id->cgrps != 1)
255                 return -EINVAL;
256
257         src = &nvme_nvm_id->groups[0];
258         dst = &nvm_id->grp;
259
260         dst->mtype = src->mtype;
261         dst->fmtype = src->fmtype;
262         dst->num_ch = src->num_ch;
263         dst->num_lun = src->num_lun;
264         dst->num_pln = src->num_pln;
265
266         dst->num_pg = le16_to_cpu(src->num_pg);
267         dst->num_blk = le16_to_cpu(src->num_blk);
268         dst->fpg_sz = le16_to_cpu(src->fpg_sz);
269         dst->csecs = le16_to_cpu(src->csecs);
270         dst->sos = le16_to_cpu(src->sos);
271
272         dst->trdt = le32_to_cpu(src->trdt);
273         dst->trdm = le32_to_cpu(src->trdm);
274         dst->tprt = le32_to_cpu(src->tprt);
275         dst->tprm = le32_to_cpu(src->tprm);
276         dst->tbet = le32_to_cpu(src->tbet);
277         dst->tbem = le32_to_cpu(src->tbem);
278         dst->mpos = le32_to_cpu(src->mpos);
279         dst->mccap = le32_to_cpu(src->mccap);
280
281         dst->cpar = le16_to_cpu(src->cpar);
282
283         if (dst->fmtype == NVM_ID_FMTYPE_MLC) {
284                 memcpy(dst->lptbl.id, src->lptbl.id, 8);
285                 dst->lptbl.mlc.num_pairs =
286                                 le16_to_cpu(src->lptbl.mlc.num_pairs);
287
288                 if (dst->lptbl.mlc.num_pairs > NVME_NVM_LP_MLC_PAIRS) {
289                         pr_err("nvm: number of MLC pairs not supported\n");
290                         return -EINVAL;
291                 }
292
293                 memcpy(dst->lptbl.mlc.pairs, src->lptbl.mlc.pairs,
294                                         dst->lptbl.mlc.num_pairs);
295         }
296
297         return 0;
298 }
299
300 static int nvme_nvm_identity(struct nvm_dev *nvmdev, struct nvm_id *nvm_id)
301 {
302         struct nvme_ns *ns = nvmdev->q->queuedata;
303         struct nvme_nvm_id *nvme_nvm_id;
304         struct nvme_nvm_command c = {};
305         int ret;
306
307         c.identity.opcode = nvme_nvm_admin_identity;
308         c.identity.nsid = cpu_to_le32(ns->ns_id);
309         c.identity.chnl_off = 0;
310
311         nvme_nvm_id = kmalloc(sizeof(struct nvme_nvm_id), GFP_KERNEL);
312         if (!nvme_nvm_id)
313                 return -ENOMEM;
314
315         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
316                                 nvme_nvm_id, sizeof(struct nvme_nvm_id));
317         if (ret) {
318                 ret = -EIO;
319                 goto out;
320         }
321
322         nvm_id->ver_id = nvme_nvm_id->ver_id;
323         nvm_id->vmnt = nvme_nvm_id->vmnt;
324         nvm_id->cap = le32_to_cpu(nvme_nvm_id->cap);
325         nvm_id->dom = le32_to_cpu(nvme_nvm_id->dom);
326         memcpy(&nvm_id->ppaf, &nvme_nvm_id->ppaf,
327                                         sizeof(struct nvm_addr_format));
328
329         ret = init_grps(nvm_id, nvme_nvm_id);
330 out:
331         kfree(nvme_nvm_id);
332         return ret;
333 }
334
335 static int nvme_nvm_get_l2p_tbl(struct nvm_dev *nvmdev, u64 slba, u32 nlb,
336                                 nvm_l2p_update_fn *update_l2p, void *priv)
337 {
338         struct nvme_ns *ns = nvmdev->q->queuedata;
339         struct nvme_nvm_command c = {};
340         u32 len = queue_max_hw_sectors(ns->ctrl->admin_q) << 9;
341         u32 nlb_pr_rq = len / sizeof(u64);
342         u64 cmd_slba = slba;
343         void *entries;
344         int ret = 0;
345
346         c.l2p.opcode = nvme_nvm_admin_get_l2p_tbl;
347         c.l2p.nsid = cpu_to_le32(ns->ns_id);
348         entries = kmalloc(len, GFP_KERNEL);
349         if (!entries)
350                 return -ENOMEM;
351
352         while (nlb) {
353                 u32 cmd_nlb = min(nlb_pr_rq, nlb);
354                 u64 elba = slba + cmd_nlb;
355
356                 c.l2p.slba = cpu_to_le64(cmd_slba);
357                 c.l2p.nlb = cpu_to_le32(cmd_nlb);
358
359                 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q,
360                                 (struct nvme_command *)&c, entries, len);
361                 if (ret) {
362                         dev_err(ns->ctrl->device,
363                                 "L2P table transfer failed (%d)\n", ret);
364                         ret = -EIO;
365                         goto out;
366                 }
367
368                 if (unlikely(elba > nvmdev->total_secs)) {
369                         pr_err("nvm: L2P data from device is out of bounds!\n");
370                         ret = -EINVAL;
371                         goto out;
372                 }
373
374                 /* Transform physical address to target address space */
375                 nvm_part_to_tgt(nvmdev, entries, cmd_nlb);
376
377                 if (update_l2p(cmd_slba, cmd_nlb, entries, priv)) {
378                         ret = -EINTR;
379                         goto out;
380                 }
381
382                 cmd_slba += cmd_nlb;
383                 nlb -= cmd_nlb;
384         }
385
386 out:
387         kfree(entries);
388         return ret;
389 }
390
391 static int nvme_nvm_get_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr ppa,
392                                                                 u8 *blks)
393 {
394         struct request_queue *q = nvmdev->q;
395         struct nvm_geo *geo = &nvmdev->geo;
396         struct nvme_ns *ns = q->queuedata;
397         struct nvme_ctrl *ctrl = ns->ctrl;
398         struct nvme_nvm_command c = {};
399         struct nvme_nvm_bb_tbl *bb_tbl;
400         int nr_blks = geo->blks_per_lun * geo->plane_mode;
401         int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blks;
402         int ret = 0;
403
404         c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
405         c.get_bb.nsid = cpu_to_le32(ns->ns_id);
406         c.get_bb.spba = cpu_to_le64(ppa.ppa);
407
408         bb_tbl = kzalloc(tblsz, GFP_KERNEL);
409         if (!bb_tbl)
410                 return -ENOMEM;
411
412         ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
413                                                                 bb_tbl, tblsz);
414         if (ret) {
415                 dev_err(ctrl->device, "get bad block table failed (%d)\n", ret);
416                 ret = -EIO;
417                 goto out;
418         }
419
420         if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
421                 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
422                 dev_err(ctrl->device, "bbt format mismatch\n");
423                 ret = -EINVAL;
424                 goto out;
425         }
426
427         if (le16_to_cpu(bb_tbl->verid) != 1) {
428                 ret = -EINVAL;
429                 dev_err(ctrl->device, "bbt version not supported\n");
430                 goto out;
431         }
432
433         if (le32_to_cpu(bb_tbl->tblks) != nr_blks) {
434                 ret = -EINVAL;
435                 dev_err(ctrl->device,
436                                 "bbt unsuspected blocks returned (%u!=%u)",
437                                 le32_to_cpu(bb_tbl->tblks), nr_blks);
438                 goto out;
439         }
440
441         memcpy(blks, bb_tbl->blk, geo->blks_per_lun * geo->plane_mode);
442 out:
443         kfree(bb_tbl);
444         return ret;
445 }
446
447 static int nvme_nvm_set_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr *ppas,
448                                                         int nr_ppas, int type)
449 {
450         struct nvme_ns *ns = nvmdev->q->queuedata;
451         struct nvme_nvm_command c = {};
452         int ret = 0;
453
454         c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
455         c.set_bb.nsid = cpu_to_le32(ns->ns_id);
456         c.set_bb.spba = cpu_to_le64(ppas->ppa);
457         c.set_bb.nlb = cpu_to_le16(nr_ppas - 1);
458         c.set_bb.value = type;
459
460         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
461                                                                 NULL, 0);
462         if (ret)
463                 dev_err(ns->ctrl->device, "set bad block table failed (%d)\n",
464                                                                         ret);
465         return ret;
466 }
467
468 static inline void nvme_nvm_rqtocmd(struct nvm_rq *rqd, struct nvme_ns *ns,
469                                     struct nvme_nvm_command *c)
470 {
471         c->ph_rw.opcode = rqd->opcode;
472         c->ph_rw.nsid = cpu_to_le32(ns->ns_id);
473         c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
474         c->ph_rw.metadata = cpu_to_le64(rqd->dma_meta_list);
475         c->ph_rw.control = cpu_to_le16(rqd->flags);
476         c->ph_rw.length = cpu_to_le16(rqd->nr_ppas - 1);
477
478         if (rqd->opcode == NVM_OP_HBWRITE || rqd->opcode == NVM_OP_HBREAD)
479                 c->hb_rw.slba = cpu_to_le64(nvme_block_nr(ns,
480                                         rqd->bio->bi_iter.bi_sector));
481 }
482
483 static void nvme_nvm_end_io(struct request *rq, int error)
484 {
485         struct nvm_rq *rqd = rq->end_io_data;
486
487         rqd->ppa_status = le64_to_cpu(nvme_req(rq)->result.u64);
488         rqd->error = nvme_req(rq)->status;
489         nvm_end_io(rqd);
490
491         kfree(nvme_req(rq)->cmd);
492         blk_mq_free_request(rq);
493 }
494
495 static 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         cmd = kzalloc(sizeof(struct nvme_nvm_command), GFP_KERNEL);
504         if (!cmd)
505                 return -ENOMEM;
506
507         nvme_nvm_rqtocmd(rqd, ns, cmd);
508
509         rq = nvme_alloc_request(q, (struct nvme_command *)cmd, 0, NVME_QID_ANY);
510         if (IS_ERR(rq)) {
511                 kfree(cmd);
512                 return -ENOMEM;
513         }
514         rq->cmd_flags &= ~REQ_FAILFAST_DRIVER;
515
516         if (bio) {
517                 blk_init_request_from_bio(rq, bio);
518         } else {
519                 rq->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
520                 rq->__data_len = 0;
521         }
522
523         rq->end_io_data = rqd;
524
525         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
526
527         return 0;
528 }
529
530 static void *nvme_nvm_create_dma_pool(struct nvm_dev *nvmdev, char *name)
531 {
532         struct nvme_ns *ns = nvmdev->q->queuedata;
533
534         return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
535 }
536
537 static void nvme_nvm_destroy_dma_pool(void *pool)
538 {
539         struct dma_pool *dma_pool = pool;
540
541         dma_pool_destroy(dma_pool);
542 }
543
544 static void *nvme_nvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
545                                     gfp_t mem_flags, dma_addr_t *dma_handler)
546 {
547         return dma_pool_alloc(pool, mem_flags, dma_handler);
548 }
549
550 static void nvme_nvm_dev_dma_free(void *pool, void *addr,
551                                                         dma_addr_t dma_handler)
552 {
553         dma_pool_free(pool, addr, dma_handler);
554 }
555
556 static struct nvm_dev_ops nvme_nvm_dev_ops = {
557         .identity               = nvme_nvm_identity,
558
559         .get_l2p_tbl            = nvme_nvm_get_l2p_tbl,
560
561         .get_bb_tbl             = nvme_nvm_get_bb_tbl,
562         .set_bb_tbl             = nvme_nvm_set_bb_tbl,
563
564         .submit_io              = nvme_nvm_submit_io,
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 static void nvme_nvm_end_user_vio(struct request *rq, int error)
575 {
576         struct completion *waiting = rq->end_io_data;
577
578         complete(waiting);
579 }
580
581 static int nvme_nvm_submit_user_cmd(struct request_queue *q,
582                                 struct nvme_ns *ns,
583                                 struct nvme_nvm_command *vcmd,
584                                 void __user *ubuf, unsigned int bufflen,
585                                 void __user *meta_buf, unsigned int meta_len,
586                                 void __user *ppa_buf, unsigned int ppa_len,
587                                 u32 *result, u64 *status, unsigned int timeout)
588 {
589         bool write = nvme_is_write((struct nvme_command *)vcmd);
590         struct nvm_dev *dev = ns->ndev;
591         struct gendisk *disk = ns->disk;
592         struct request *rq;
593         struct bio *bio = NULL;
594         __le64 *ppa_list = NULL;
595         dma_addr_t ppa_dma;
596         __le64 *metadata = NULL;
597         dma_addr_t metadata_dma;
598         DECLARE_COMPLETION_ONSTACK(wait);
599         int ret = 0;
600
601         rq = nvme_alloc_request(q, (struct nvme_command *)vcmd, 0,
602                         NVME_QID_ANY);
603         if (IS_ERR(rq)) {
604                 ret = -ENOMEM;
605                 goto err_cmd;
606         }
607
608         rq->timeout = timeout ? timeout : ADMIN_TIMEOUT;
609
610         rq->cmd_flags &= ~REQ_FAILFAST_DRIVER;
611         rq->end_io_data = &wait;
612
613         if (ppa_buf && ppa_len) {
614                 ppa_list = dma_pool_alloc(dev->dma_pool, GFP_KERNEL, &ppa_dma);
615                 if (!ppa_list) {
616                         ret = -ENOMEM;
617                         goto err_rq;
618                 }
619                 if (copy_from_user(ppa_list, (void __user *)ppa_buf,
620                                                 sizeof(u64) * (ppa_len + 1))) {
621                         ret = -EFAULT;
622                         goto err_ppa;
623                 }
624                 vcmd->ph_rw.spba = cpu_to_le64(ppa_dma);
625         } else {
626                 vcmd->ph_rw.spba = cpu_to_le64((uintptr_t)ppa_buf);
627         }
628
629         if (ubuf && bufflen) {
630                 ret = blk_rq_map_user(q, rq, NULL, ubuf, bufflen, GFP_KERNEL);
631                 if (ret)
632                         goto err_ppa;
633                 bio = rq->bio;
634
635                 if (meta_buf && meta_len) {
636                         metadata = dma_pool_alloc(dev->dma_pool, GFP_KERNEL,
637                                                                 &metadata_dma);
638                         if (!metadata) {
639                                 ret = -ENOMEM;
640                                 goto err_map;
641                         }
642
643                         if (write) {
644                                 if (copy_from_user(metadata,
645                                                 (void __user *)meta_buf,
646                                                 meta_len)) {
647                                         ret = -EFAULT;
648                                         goto err_meta;
649                                 }
650                         }
651                         vcmd->ph_rw.metadata = cpu_to_le64(metadata_dma);
652                 }
653
654                 if (!disk)
655                         goto submit;
656
657                 bio->bi_bdev = bdget_disk(disk, 0);
658                 if (!bio->bi_bdev) {
659                         ret = -ENODEV;
660                         goto err_meta;
661                 }
662         }
663
664 submit:
665         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_user_vio);
666
667         wait_for_completion_io(&wait);
668
669         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
670                 ret = -EINTR;
671         else if (nvme_req(rq)->status & 0x7ff)
672                 ret = -EIO;
673         if (result)
674                 *result = nvme_req(rq)->status & 0x7ff;
675         if (status)
676                 *status = le64_to_cpu(nvme_req(rq)->result.u64);
677
678         if (metadata && !ret && !write) {
679                 if (copy_to_user(meta_buf, (void *)metadata, meta_len))
680                         ret = -EFAULT;
681         }
682 err_meta:
683         if (meta_buf && meta_len)
684                 dma_pool_free(dev->dma_pool, metadata, metadata_dma);
685 err_map:
686         if (bio) {
687                 if (disk && bio->bi_bdev)
688                         bdput(bio->bi_bdev);
689                 blk_rq_unmap_user(bio);
690         }
691 err_ppa:
692         if (ppa_buf && ppa_len)
693                 dma_pool_free(dev->dma_pool, ppa_list, ppa_dma);
694 err_rq:
695         blk_mq_free_request(rq);
696 err_cmd:
697         return ret;
698 }
699
700 static int nvme_nvm_submit_vio(struct nvme_ns *ns,
701                                         struct nvm_user_vio __user *uvio)
702 {
703         struct nvm_user_vio vio;
704         struct nvme_nvm_command c;
705         unsigned int length;
706         int ret;
707
708         if (copy_from_user(&vio, uvio, sizeof(vio)))
709                 return -EFAULT;
710         if (vio.flags)
711                 return -EINVAL;
712
713         memset(&c, 0, sizeof(c));
714         c.ph_rw.opcode = vio.opcode;
715         c.ph_rw.nsid = cpu_to_le32(ns->ns_id);
716         c.ph_rw.control = cpu_to_le16(vio.control);
717         c.ph_rw.length = cpu_to_le16(vio.nppas);
718
719         length = (vio.nppas + 1) << ns->lba_shift;
720
721         ret = nvme_nvm_submit_user_cmd(ns->queue, ns, &c,
722                         (void __user *)(uintptr_t)vio.addr, length,
723                         (void __user *)(uintptr_t)vio.metadata,
724                                                         vio.metadata_len,
725                         (void __user *)(uintptr_t)vio.ppa_list, vio.nppas,
726                         &vio.result, &vio.status, 0);
727
728         if (ret && copy_to_user(uvio, &vio, sizeof(vio)))
729                 return -EFAULT;
730
731         return ret;
732 }
733
734 static int nvme_nvm_user_vcmd(struct nvme_ns *ns, int admin,
735                                         struct nvm_passthru_vio __user *uvcmd)
736 {
737         struct nvm_passthru_vio vcmd;
738         struct nvme_nvm_command c;
739         struct request_queue *q;
740         unsigned int timeout = 0;
741         int ret;
742
743         if (copy_from_user(&vcmd, uvcmd, sizeof(vcmd)))
744                 return -EFAULT;
745         if ((vcmd.opcode != 0xF2) && (!capable(CAP_SYS_ADMIN)))
746                 return -EACCES;
747         if (vcmd.flags)
748                 return -EINVAL;
749
750         memset(&c, 0, sizeof(c));
751         c.common.opcode = vcmd.opcode;
752         c.common.nsid = cpu_to_le32(ns->ns_id);
753         c.common.cdw2[0] = cpu_to_le32(vcmd.cdw2);
754         c.common.cdw2[1] = cpu_to_le32(vcmd.cdw3);
755         /* cdw11-12 */
756         c.ph_rw.length = cpu_to_le16(vcmd.nppas);
757         c.ph_rw.control  = cpu_to_le16(vcmd.control);
758         c.common.cdw10[3] = cpu_to_le32(vcmd.cdw13);
759         c.common.cdw10[4] = cpu_to_le32(vcmd.cdw14);
760         c.common.cdw10[5] = cpu_to_le32(vcmd.cdw15);
761
762         if (vcmd.timeout_ms)
763                 timeout = msecs_to_jiffies(vcmd.timeout_ms);
764
765         q = admin ? ns->ctrl->admin_q : ns->queue;
766
767         ret = nvme_nvm_submit_user_cmd(q, ns,
768                         (struct nvme_nvm_command *)&c,
769                         (void __user *)(uintptr_t)vcmd.addr, vcmd.data_len,
770                         (void __user *)(uintptr_t)vcmd.metadata,
771                                                         vcmd.metadata_len,
772                         (void __user *)(uintptr_t)vcmd.ppa_list, vcmd.nppas,
773                         &vcmd.result, &vcmd.status, timeout);
774
775         if (ret && copy_to_user(uvcmd, &vcmd, sizeof(vcmd)))
776                 return -EFAULT;
777
778         return ret;
779 }
780
781 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg)
782 {
783         switch (cmd) {
784         case NVME_NVM_IOCTL_ADMIN_VIO:
785                 return nvme_nvm_user_vcmd(ns, 1, (void __user *)arg);
786         case NVME_NVM_IOCTL_IO_VIO:
787                 return nvme_nvm_user_vcmd(ns, 0, (void __user *)arg);
788         case NVME_NVM_IOCTL_SUBMIT_VIO:
789                 return nvme_nvm_submit_vio(ns, (void __user *)arg);
790         default:
791                 return -ENOTTY;
792         }
793 }
794
795 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node)
796 {
797         struct request_queue *q = ns->queue;
798         struct nvm_dev *dev;
799
800         _nvme_nvm_check_size();
801
802         dev = nvm_alloc_dev(node);
803         if (!dev)
804                 return -ENOMEM;
805
806         dev->q = q;
807         memcpy(dev->name, disk_name, DISK_NAME_LEN);
808         dev->ops = &nvme_nvm_dev_ops;
809         dev->private_data = ns;
810         ns->ndev = dev;
811
812         return nvm_register(dev);
813 }
814
815 void nvme_nvm_unregister(struct nvme_ns *ns)
816 {
817         nvm_unregister(ns->ndev);
818 }
819
820 static ssize_t nvm_dev_attr_show(struct device *dev,
821                                  struct device_attribute *dattr, char *page)
822 {
823         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
824         struct nvm_dev *ndev = ns->ndev;
825         struct nvm_id *id;
826         struct nvm_id_group *grp;
827         struct attribute *attr;
828
829         if (!ndev)
830                 return 0;
831
832         id = &ndev->identity;
833         grp = &id->grp;
834         attr = &dattr->attr;
835
836         if (strcmp(attr->name, "version") == 0) {
837                 return scnprintf(page, PAGE_SIZE, "%u\n", id->ver_id);
838         } else if (strcmp(attr->name, "vendor_opcode") == 0) {
839                 return scnprintf(page, PAGE_SIZE, "%u\n", id->vmnt);
840         } else if (strcmp(attr->name, "capabilities") == 0) {
841                 return scnprintf(page, PAGE_SIZE, "%u\n", id->cap);
842         } else if (strcmp(attr->name, "device_mode") == 0) {
843                 return scnprintf(page, PAGE_SIZE, "%u\n", id->dom);
844         /* kept for compatibility */
845         } else if (strcmp(attr->name, "media_manager") == 0) {
846                 return scnprintf(page, PAGE_SIZE, "%s\n", "gennvm");
847         } else if (strcmp(attr->name, "ppa_format") == 0) {
848                 return scnprintf(page, PAGE_SIZE,
849                         "0x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
850                         id->ppaf.ch_offset, id->ppaf.ch_len,
851                         id->ppaf.lun_offset, id->ppaf.lun_len,
852                         id->ppaf.pln_offset, id->ppaf.pln_len,
853                         id->ppaf.blk_offset, id->ppaf.blk_len,
854                         id->ppaf.pg_offset, id->ppaf.pg_len,
855                         id->ppaf.sect_offset, id->ppaf.sect_len);
856         } else if (strcmp(attr->name, "media_type") == 0) {     /* u8 */
857                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->mtype);
858         } else if (strcmp(attr->name, "flash_media_type") == 0) {
859                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fmtype);
860         } else if (strcmp(attr->name, "num_channels") == 0) {
861                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_ch);
862         } else if (strcmp(attr->name, "num_luns") == 0) {
863                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_lun);
864         } else if (strcmp(attr->name, "num_planes") == 0) {
865                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pln);
866         } else if (strcmp(attr->name, "num_blocks") == 0) {     /* u16 */
867                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_blk);
868         } else if (strcmp(attr->name, "num_pages") == 0) {
869                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pg);
870         } else if (strcmp(attr->name, "page_size") == 0) {
871                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fpg_sz);
872         } else if (strcmp(attr->name, "hw_sector_size") == 0) {
873                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->csecs);
874         } else if (strcmp(attr->name, "oob_sector_size") == 0) {/* u32 */
875                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->sos);
876         } else if (strcmp(attr->name, "read_typ") == 0) {
877                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdt);
878         } else if (strcmp(attr->name, "read_max") == 0) {
879                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdm);
880         } else if (strcmp(attr->name, "prog_typ") == 0) {
881                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprt);
882         } else if (strcmp(attr->name, "prog_max") == 0) {
883                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprm);
884         } else if (strcmp(attr->name, "erase_typ") == 0) {
885                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbet);
886         } else if (strcmp(attr->name, "erase_max") == 0) {
887                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbem);
888         } else if (strcmp(attr->name, "multiplane_modes") == 0) {
889                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mpos);
890         } else if (strcmp(attr->name, "media_capabilities") == 0) {
891                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mccap);
892         } else if (strcmp(attr->name, "max_phys_secs") == 0) {
893                 return scnprintf(page, PAGE_SIZE, "%u\n",
894                                 ndev->ops->max_phys_sect);
895         } else {
896                 return scnprintf(page,
897                                  PAGE_SIZE,
898                                  "Unhandled attr(%s) in `nvm_dev_attr_show`\n",
899                                  attr->name);
900         }
901 }
902
903 #define NVM_DEV_ATTR_RO(_name)                                          \
904         DEVICE_ATTR(_name, S_IRUGO, nvm_dev_attr_show, NULL)
905
906 static NVM_DEV_ATTR_RO(version);
907 static NVM_DEV_ATTR_RO(vendor_opcode);
908 static NVM_DEV_ATTR_RO(capabilities);
909 static NVM_DEV_ATTR_RO(device_mode);
910 static NVM_DEV_ATTR_RO(ppa_format);
911 static NVM_DEV_ATTR_RO(media_manager);
912
913 static NVM_DEV_ATTR_RO(media_type);
914 static NVM_DEV_ATTR_RO(flash_media_type);
915 static NVM_DEV_ATTR_RO(num_channels);
916 static NVM_DEV_ATTR_RO(num_luns);
917 static NVM_DEV_ATTR_RO(num_planes);
918 static NVM_DEV_ATTR_RO(num_blocks);
919 static NVM_DEV_ATTR_RO(num_pages);
920 static NVM_DEV_ATTR_RO(page_size);
921 static NVM_DEV_ATTR_RO(hw_sector_size);
922 static NVM_DEV_ATTR_RO(oob_sector_size);
923 static NVM_DEV_ATTR_RO(read_typ);
924 static NVM_DEV_ATTR_RO(read_max);
925 static NVM_DEV_ATTR_RO(prog_typ);
926 static NVM_DEV_ATTR_RO(prog_max);
927 static NVM_DEV_ATTR_RO(erase_typ);
928 static NVM_DEV_ATTR_RO(erase_max);
929 static NVM_DEV_ATTR_RO(multiplane_modes);
930 static NVM_DEV_ATTR_RO(media_capabilities);
931 static NVM_DEV_ATTR_RO(max_phys_secs);
932
933 static struct attribute *nvm_dev_attrs[] = {
934         &dev_attr_version.attr,
935         &dev_attr_vendor_opcode.attr,
936         &dev_attr_capabilities.attr,
937         &dev_attr_device_mode.attr,
938         &dev_attr_media_manager.attr,
939
940         &dev_attr_ppa_format.attr,
941         &dev_attr_media_type.attr,
942         &dev_attr_flash_media_type.attr,
943         &dev_attr_num_channels.attr,
944         &dev_attr_num_luns.attr,
945         &dev_attr_num_planes.attr,
946         &dev_attr_num_blocks.attr,
947         &dev_attr_num_pages.attr,
948         &dev_attr_page_size.attr,
949         &dev_attr_hw_sector_size.attr,
950         &dev_attr_oob_sector_size.attr,
951         &dev_attr_read_typ.attr,
952         &dev_attr_read_max.attr,
953         &dev_attr_prog_typ.attr,
954         &dev_attr_prog_max.attr,
955         &dev_attr_erase_typ.attr,
956         &dev_attr_erase_max.attr,
957         &dev_attr_multiplane_modes.attr,
958         &dev_attr_media_capabilities.attr,
959         &dev_attr_max_phys_secs.attr,
960         NULL,
961 };
962
963 static const struct attribute_group nvm_dev_attr_group = {
964         .name           = "lightnvm",
965         .attrs          = nvm_dev_attrs,
966 };
967
968 int nvme_nvm_register_sysfs(struct nvme_ns *ns)
969 {
970         return sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
971                                         &nvm_dev_attr_group);
972 }
973
974 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns)
975 {
976         sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
977                                         &nvm_dev_attr_group);
978 }
979
980 /* move to shared place when used in multiple places. */
981 #define PCI_VENDOR_ID_CNEX 0x1d1d
982 #define PCI_DEVICE_ID_CNEX_WL 0x2807
983 #define PCI_DEVICE_ID_CNEX_QEMU 0x1f1f
984
985 int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
986 {
987         struct nvme_ctrl *ctrl = ns->ctrl;
988         /* XXX: this is poking into PCI structures from generic code! */
989         struct pci_dev *pdev = to_pci_dev(ctrl->dev);
990
991         /* QEMU NVMe simulator - PCI ID + Vendor specific bit */
992         if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
993                                 pdev->device == PCI_DEVICE_ID_CNEX_QEMU &&
994                                                         id->vs[0] == 0x1)
995                 return 1;
996
997         /* CNEX Labs - PCI ID + Vendor specific bit */
998         if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
999                                 pdev->device == PCI_DEVICE_ID_CNEX_WL &&
1000                                                         id->vs[0] == 0x1)
1001                 return 1;
1002
1003         return 0;
1004 }