1/*
2 * Copyright (C) 2015 IT University of Copenhagen. All rights reserved.
3 * Initial release: Matias Bjorling <m@bjorling.me>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License version
7 * 2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; see the file COPYING. If not, write to
16 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
17 * USA.
18 *
19 */
20
21#include <linux/list.h>
22#include <linux/types.h>
23#include <linux/sem.h>
24#include <linux/bitmap.h>
25#include <linux/module.h>
26#include <linux/moduleparam.h>
27#include <linux/miscdevice.h>
28#include <linux/lightnvm.h>
29#include <linux/sched/sysctl.h>
30
31static LIST_HEAD(nvm_tgt_types);
32static DECLARE_RWSEM(nvm_tgtt_lock);
33static LIST_HEAD(nvm_devices);
34static DECLARE_RWSEM(nvm_lock);
35
36/* Map between virtual and physical channel and lun */
37struct nvm_ch_map {
38 int ch_off;
39 int num_lun;
40 int *lun_offs;
41};
42
43struct nvm_dev_map {
44 struct nvm_ch_map *chnls;
45 int num_ch;
46};
47
48static struct nvm_target *nvm_find_target(struct nvm_dev *dev, const char *name)
49{
50 struct nvm_target *tgt;
51
52 list_for_each_entry(tgt, &dev->targets, list)
53 if (!strcmp(name, tgt->disk->disk_name))
54 return tgt;
55
56 return NULL;
57}
58
59static bool nvm_target_exists(const char *name)
60{
61 struct nvm_dev *dev;
62 struct nvm_target *tgt;
63 bool ret = false;
64
65 down_write(&nvm_lock);
66 list_for_each_entry(dev, &nvm_devices, devices) {
67 mutex_lock(&dev->mlock);
68 list_for_each_entry(tgt, &dev->targets, list) {
69 if (!strcmp(name, tgt->disk->disk_name)) {
70 ret = true;
71 mutex_unlock(&dev->mlock);
72 goto out;
73 }
74 }
75 mutex_unlock(&dev->mlock);
76 }
77
78out:
79 up_write(&nvm_lock);
80 return ret;
81}
82
83static int nvm_reserve_luns(struct nvm_dev *dev, int lun_begin, int lun_end)
84{
85 int i;
86
87 for (i = lun_begin; i <= lun_end; i++) {
88 if (test_and_set_bit(i, dev->lun_map)) {
89 pr_err("nvm: lun %d already allocated\n", i);
90 goto err;
91 }
92 }
93
94 return 0;
95err:
96 while (--i >= lun_begin)
97 clear_bit(i, dev->lun_map);
98
99 return -EBUSY;
100}
101
102static void nvm_release_luns_err(struct nvm_dev *dev, int lun_begin,
103 int lun_end)
104{
105 int i;
106
107 for (i = lun_begin; i <= lun_end; i++)
108 WARN_ON(!test_and_clear_bit(i, dev->lun_map));
109}
110
111static void nvm_remove_tgt_dev(struct nvm_tgt_dev *tgt_dev, int clear)
112{
113 struct nvm_dev *dev = tgt_dev->parent;
114 struct nvm_dev_map *dev_map = tgt_dev->map;
115 int i, j;
116
117 for (i = 0; i < dev_map->num_ch; i++) {
118 struct nvm_ch_map *ch_map = &dev_map->chnls[i];
119 int *lun_offs = ch_map->lun_offs;
120 int ch = i + ch_map->ch_off;
121
122 if (clear) {
123 for (j = 0; j < ch_map->num_lun; j++) {
124 int lun = j + lun_offs[j];
125 int lunid = (ch * dev->geo.num_lun) + lun;
126
127 WARN_ON(!test_and_clear_bit(lunid,
128 dev->lun_map));
129 }
130 }
131
132 kfree(ch_map->lun_offs);
133 }
134
135 kfree(dev_map->chnls);
136 kfree(dev_map);
137
138 kfree(tgt_dev->luns);
139 kfree(tgt_dev);
140}
141
142static struct nvm_tgt_dev *nvm_create_tgt_dev(struct nvm_dev *dev,
143 u16 lun_begin, u16 lun_end,
144 u16 op)
145{
146 struct nvm_tgt_dev *tgt_dev = NULL;
147 struct nvm_dev_map *dev_rmap = dev->rmap;
148 struct nvm_dev_map *dev_map;
149 struct ppa_addr *luns;
150 int num_lun = lun_end - lun_begin + 1;
151 int luns_left = num_lun;
152 int num_ch = num_lun / dev->geo.num_lun;
153 int num_ch_mod = num_lun % dev->geo.num_lun;
154 int bch = lun_begin / dev->geo.num_lun;
155 int blun = lun_begin % dev->geo.num_lun;
156 int lunid = 0;
157 int lun_balanced = 1;
158 int sec_per_lun, prev_num_lun;
159 int i, j;
160
161 num_ch = (num_ch_mod == 0) ? num_ch : num_ch + 1;
162
163 dev_map = kmalloc(sizeof(struct nvm_dev_map), GFP_KERNEL);
164 if (!dev_map)
165 goto err_dev;
166
167 dev_map->chnls = kcalloc(num_ch, sizeof(struct nvm_ch_map), GFP_KERNEL);
168 if (!dev_map->chnls)
169 goto err_chnls;
170
171 luns = kcalloc(num_lun, sizeof(struct ppa_addr), GFP_KERNEL);
172 if (!luns)
173 goto err_luns;
174
175 prev_num_lun = (luns_left > dev->geo.num_lun) ?
176 dev->geo.num_lun : luns_left;
177 for (i = 0; i < num_ch; i++) {
178 struct nvm_ch_map *ch_rmap = &dev_rmap->chnls[i + bch];
179 int *lun_roffs = ch_rmap->lun_offs;
180 struct nvm_ch_map *ch_map = &dev_map->chnls[i];
181 int *lun_offs;
182 int luns_in_chnl = (luns_left > dev->geo.num_lun) ?
183 dev->geo.num_lun : luns_left;
184
185 if (lun_balanced && prev_num_lun != luns_in_chnl)
186 lun_balanced = 0;
187
188 ch_map->ch_off = ch_rmap->ch_off = bch;
189 ch_map->num_lun = luns_in_chnl;
190
191 lun_offs = kcalloc(luns_in_chnl, sizeof(int), GFP_KERNEL);
192 if (!lun_offs)
193 goto err_ch;
194
195 for (j = 0; j < luns_in_chnl; j++) {
196 luns[lunid].ppa = 0;
197 luns[lunid].a.ch = i;
198 luns[lunid++].a.lun = j;
199
200 lun_offs[j] = blun;
201 lun_roffs[j + blun] = blun;
202 }
203
204 ch_map->lun_offs = lun_offs;
205
206 /* when starting a new channel, lun offset is reset */
207 blun = 0;
208 luns_left -= luns_in_chnl;
209 }
210
211 dev_map->num_ch = num_ch;
212
213 tgt_dev = kmalloc(sizeof(struct nvm_tgt_dev), GFP_KERNEL);
214 if (!tgt_dev)
215 goto err_ch;
216
217 /* Inherit device geometry from parent */
218 memcpy(&tgt_dev->geo, &dev->geo, sizeof(struct nvm_geo));
219
220 /* Target device only owns a portion of the physical device */
221 tgt_dev->geo.num_ch = num_ch;
222 tgt_dev->geo.num_lun = (lun_balanced) ? prev_num_lun : -1;
223 tgt_dev->geo.all_luns = num_lun;
224 tgt_dev->geo.all_chunks = num_lun * dev->geo.num_chk;
225
226 tgt_dev->geo.op = op;
227
228 sec_per_lun = dev->geo.clba * dev->geo.num_chk;
229 tgt_dev->geo.total_secs = num_lun * sec_per_lun;
230
231 tgt_dev->q = dev->q;
232 tgt_dev->map = dev_map;
233 tgt_dev->luns = luns;
234 tgt_dev->parent = dev;
235
236 return tgt_dev;
237err_ch:
238 while (--i >= 0)
239 kfree(dev_map->chnls[i].lun_offs);
240 kfree(luns);
241err_luns:
242 kfree(dev_map->chnls);
243err_chnls:
244 kfree(dev_map);
245err_dev:
246 return tgt_dev;
247}
248
249static const struct block_device_operations nvm_fops = {
250 .owner = THIS_MODULE,
251};
252
253static struct nvm_tgt_type *__nvm_find_target_type(const char *name)
254{
255 struct nvm_tgt_type *tt;
256
257 list_for_each_entry(tt, &nvm_tgt_types, list)
258 if (!strcmp(name, tt->name))
259 return tt;
260
261 return NULL;
262}
263
264static struct nvm_tgt_type *nvm_find_target_type(const char *name)
265{
266 struct nvm_tgt_type *tt;
267
268 down_write(&nvm_tgtt_lock);
269 tt = __nvm_find_target_type(name);
270 up_write(&nvm_tgtt_lock);
271
272 return tt;
273}
274
275static int nvm_config_check_luns(struct nvm_geo *geo, int lun_begin,
276 int lun_end)
277{
278 if (lun_begin > lun_end || lun_end >= geo->all_luns) {
279 pr_err("nvm: lun out of bound (%u:%u > %u)\n",
280 lun_begin, lun_end, geo->all_luns - 1);
281 return -EINVAL;
282 }
283
284 return 0;
285}
286
287static int __nvm_config_simple(struct nvm_dev *dev,
288 struct nvm_ioctl_create_simple *s)
289{
290 struct nvm_geo *geo = &dev->geo;
291
292 if (s->lun_begin == -1 && s->lun_end == -1) {
293 s->lun_begin = 0;
294 s->lun_end = geo->all_luns - 1;
295 }
296
297 return nvm_config_check_luns(geo, s->lun_begin, s->lun_end);
298}
299
300static int __nvm_config_extended(struct nvm_dev *dev,
301 struct nvm_ioctl_create_extended *e)
302{
303 if (e->lun_begin == 0xFFFF && e->lun_end == 0xFFFF) {
304 e->lun_begin = 0;
305 e->lun_end = dev->geo.all_luns - 1;
306 }
307
308 /* op not set falls into target's default */
309 if (e->op == 0xFFFF) {
310 e->op = NVM_TARGET_DEFAULT_OP;
311 } else if (e->op < NVM_TARGET_MIN_OP || e->op > NVM_TARGET_MAX_OP) {
312 pr_err("nvm: invalid over provisioning value\n");
313 return -EINVAL;
314 }
315
316 return nvm_config_check_luns(&dev->geo, e->lun_begin, e->lun_end);
317}
318
319static int nvm_create_tgt(struct nvm_dev *dev, struct nvm_ioctl_create *create)
320{
321 struct nvm_ioctl_create_extended e;
322 struct request_queue *tqueue;
323 struct gendisk *tdisk;
324 struct nvm_tgt_type *tt;
325 struct nvm_target *t;
326 struct nvm_tgt_dev *tgt_dev;
327 void *targetdata;
328 int ret;
329
330 switch (create->conf.type) {
331 case NVM_CONFIG_TYPE_SIMPLE:
332 ret = __nvm_config_simple(dev, &create->conf.s);
333 if (ret)
334 return ret;
335
336 e.lun_begin = create->conf.s.lun_begin;
337 e.lun_end = create->conf.s.lun_end;
338 e.op = NVM_TARGET_DEFAULT_OP;
339 break;
340 case NVM_CONFIG_TYPE_EXTENDED:
341 ret = __nvm_config_extended(dev, &create->conf.e);
342 if (ret)
343 return ret;
344
345 e = create->conf.e;
346 break;
347 default:
348 pr_err("nvm: config type not valid\n");
349 return -EINVAL;
350 }
351
352 tt = nvm_find_target_type(create->tgttype);
353 if (!tt) {
354 pr_err("nvm: target type %s not found\n", create->tgttype);
355 return -EINVAL;
356 }
357
358 if ((tt->flags & NVM_TGT_F_HOST_L2P) != (dev->geo.dom & NVM_RSP_L2P)) {
359 pr_err("nvm: device is incompatible with target L2P type.\n");
360 return -EINVAL;
361 }
362
363 if (nvm_target_exists(create->tgtname)) {
364 pr_err("nvm: target name already exists (%s)\n",
365 create->tgtname);
366 return -EINVAL;
367 }
368
369 ret = nvm_reserve_luns(dev, e.lun_begin, e.lun_end);
370 if (ret)
371 return ret;
372
373 t = kmalloc(sizeof(struct nvm_target), GFP_KERNEL);
374 if (!t) {
375 ret = -ENOMEM;
376 goto err_reserve;
377 }
378
379 tgt_dev = nvm_create_tgt_dev(dev, e.lun_begin, e.lun_end, e.op);
380 if (!tgt_dev) {
381 pr_err("nvm: could not create target device\n");
382 ret = -ENOMEM;
383 goto err_t;
384 }
385
386 tdisk = alloc_disk(0);
387 if (!tdisk) {
388 ret = -ENOMEM;
389 goto err_dev;
390 }
391
392 tqueue = blk_alloc_queue_node(GFP_KERNEL, dev->q->node);
393 if (!tqueue) {
394 ret = -ENOMEM;
395 goto err_disk;
396 }
397 blk_queue_make_request(tqueue, tt->make_rq);
398
399 strlcpy(tdisk->disk_name, create->tgtname, sizeof(tdisk->disk_name));
400 tdisk->flags = GENHD_FL_EXT_DEVT;
401 tdisk->major = 0;
402 tdisk->first_minor = 0;
403 tdisk->fops = &nvm_fops;
404 tdisk->queue = tqueue;
405
406 targetdata = tt->init(tgt_dev, tdisk, create->flags);
407 if (IS_ERR(targetdata)) {
408 ret = PTR_ERR(targetdata);
409 goto err_init;
410 }
411
412 tdisk->private_data = targetdata;
413 tqueue->queuedata = targetdata;
414
415 blk_queue_max_hw_sectors(tqueue,
416 (dev->geo.csecs >> 9) * NVM_MAX_VLBA);
417
418 set_capacity(tdisk, tt->capacity(targetdata));
419 add_disk(tdisk);
420
421 if (tt->sysfs_init && tt->sysfs_init(tdisk)) {
422 ret = -ENOMEM;
423 goto err_sysfs;
424 }
425
426 t->type = tt;
427 t->disk = tdisk;
428 t->dev = tgt_dev;
429
430 mutex_lock(&dev->mlock);
431 list_add_tail(&t->list, &dev->targets);
432 mutex_unlock(&dev->mlock);
433
434 __module_get(tt->owner);
435
436 return 0;
437err_sysfs:
438 if (tt->exit)
439 tt->exit(targetdata, true);
440err_init:
441 blk_cleanup_queue(tqueue);
442 tdisk->queue = NULL;
443err_disk:
444 put_disk(tdisk);
445err_dev:
446 nvm_remove_tgt_dev(tgt_dev, 0);
447err_t:
448 kfree(t);
449err_reserve:
450 nvm_release_luns_err(dev, e.lun_begin, e.lun_end);
451 return ret;
452}
453
454static void __nvm_remove_target(struct nvm_target *t, bool graceful)
455{
456 struct nvm_tgt_type *tt = t->type;
457 struct gendisk *tdisk = t->disk;
458 struct request_queue *q = tdisk->queue;
459
460 del_gendisk(tdisk);
461 blk_cleanup_queue(q);
462
463 if (tt->sysfs_exit)
464 tt->sysfs_exit(tdisk);
465
466 if (tt->exit)
467 tt->exit(tdisk->private_data, graceful);
468
469 nvm_remove_tgt_dev(t->dev, 1);
470 put_disk(tdisk);
471 module_put(t->type->owner);
472
473 list_del(&t->list);
474 kfree(t);
475}
476
477/**
478 * nvm_remove_tgt - Removes a target from the media manager
479 * @dev: device
480 * @remove: ioctl structure with target name to remove.
481 *
482 * Returns:
483 * 0: on success
484 * 1: on not found
485 * <0: on error
486 */
487static int nvm_remove_tgt(struct nvm_dev *dev, struct nvm_ioctl_remove *remove)
488{
489 struct nvm_target *t;
490
491 mutex_lock(&dev->mlock);
492 t = nvm_find_target(dev, remove->tgtname);
493 if (!t) {
494 mutex_unlock(&dev->mlock);
495 return 1;
496 }
497 __nvm_remove_target(t, true);
498 mutex_unlock(&dev->mlock);
499
500 return 0;
501}
502
503static int nvm_register_map(struct nvm_dev *dev)
504{
505 struct nvm_dev_map *rmap;
506 int i, j;
507
508 rmap = kmalloc(sizeof(struct nvm_dev_map), GFP_KERNEL);
509 if (!rmap)
510 goto err_rmap;
511
512 rmap->chnls = kcalloc(dev->geo.num_ch, sizeof(struct nvm_ch_map),
513 GFP_KERNEL);
514 if (!rmap->chnls)
515 goto err_chnls;
516
517 for (i = 0; i < dev->geo.num_ch; i++) {
518 struct nvm_ch_map *ch_rmap;
519 int *lun_roffs;
520 int luns_in_chnl = dev->geo.num_lun;
521
522 ch_rmap = &rmap->chnls[i];
523
524 ch_rmap->ch_off = -1;
525 ch_rmap->num_lun = luns_in_chnl;
526
527 lun_roffs = kcalloc(luns_in_chnl, sizeof(int), GFP_KERNEL);
528 if (!lun_roffs)
529 goto err_ch;
530
531 for (j = 0; j < luns_in_chnl; j++)
532 lun_roffs[j] = -1;
533
534 ch_rmap->lun_offs = lun_roffs;
535 }
536
537 dev->rmap = rmap;
538
539 return 0;
540err_ch:
541 while (--i >= 0)
542 kfree(rmap->chnls[i].lun_offs);
543err_chnls:
544 kfree(rmap);
545err_rmap:
546 return -ENOMEM;
547}
548
549static void nvm_unregister_map(struct nvm_dev *dev)
550{
551 struct nvm_dev_map *rmap = dev->rmap;
552 int i;
553
554 for (i = 0; i < dev->geo.num_ch; i++)
555 kfree(rmap->chnls[i].lun_offs);
556
557 kfree(rmap->chnls);
558 kfree(rmap);
559}
560
561static void nvm_map_to_dev(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *p)
562{
563 struct nvm_dev_map *dev_map = tgt_dev->map;
564 struct nvm_ch_map *ch_map = &dev_map->chnls[p->a.ch];
565 int lun_off = ch_map->lun_offs[p->a.lun];
566
567 p->a.ch += ch_map->ch_off;
568 p->a.lun += lun_off;
569}
570
571static void nvm_map_to_tgt(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *p)
572{
573 struct nvm_dev *dev = tgt_dev->parent;
574 struct nvm_dev_map *dev_rmap = dev->rmap;
575 struct nvm_ch_map *ch_rmap = &dev_rmap->chnls[p->a.ch];
576 int lun_roff = ch_rmap->lun_offs[p->a.lun];
577
578 p->a.ch -= ch_rmap->ch_off;
579 p->a.lun -= lun_roff;
580}
581
582static void nvm_ppa_tgt_to_dev(struct nvm_tgt_dev *tgt_dev,
583 struct ppa_addr *ppa_list, int nr_ppas)
584{
585 int i;
586
587 for (i = 0; i < nr_ppas; i++) {
588 nvm_map_to_dev(tgt_dev, &ppa_list[i]);
589 ppa_list[i] = generic_to_dev_addr(tgt_dev->parent, ppa_list[i]);
590 }
591}
592
593static void nvm_ppa_dev_to_tgt(struct nvm_tgt_dev *tgt_dev,
594 struct ppa_addr *ppa_list, int nr_ppas)
595{
596 int i;
597
598 for (i = 0; i < nr_ppas; i++) {
599 ppa_list[i] = dev_to_generic_addr(tgt_dev->parent, ppa_list[i]);
600 nvm_map_to_tgt(tgt_dev, &ppa_list[i]);
601 }
602}
603
604static void nvm_rq_tgt_to_dev(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
605{
606 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
607
608 nvm_ppa_tgt_to_dev(tgt_dev, ppa_list, rqd->nr_ppas);
609}
610
611static void nvm_rq_dev_to_tgt(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
612{
613 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
614
615 nvm_ppa_dev_to_tgt(tgt_dev, ppa_list, rqd->nr_ppas);
616}
617
618int nvm_register_tgt_type(struct nvm_tgt_type *tt)
619{
620 int ret = 0;
621
622 down_write(&nvm_tgtt_lock);
623 if (__nvm_find_target_type(tt->name))
624 ret = -EEXIST;
625 else
626 list_add(&tt->list, &nvm_tgt_types);
627 up_write(&nvm_tgtt_lock);
628
629 return ret;
630}
631EXPORT_SYMBOL(nvm_register_tgt_type);
632
633void nvm_unregister_tgt_type(struct nvm_tgt_type *tt)
634{
635 if (!tt)
636 return;
637
638 down_write(&nvm_tgtt_lock);
639 list_del(&tt->list);
640 up_write(&nvm_tgtt_lock);
641}
642EXPORT_SYMBOL(nvm_unregister_tgt_type);
643
644void *nvm_dev_dma_alloc(struct nvm_dev *dev, gfp_t mem_flags,
645 dma_addr_t *dma_handler)
646{
647 return dev->ops->dev_dma_alloc(dev, dev->dma_pool, mem_flags,
648 dma_handler);
649}
650EXPORT_SYMBOL(nvm_dev_dma_alloc);
651
652void nvm_dev_dma_free(struct nvm_dev *dev, void *addr, dma_addr_t dma_handler)
653{
654 dev->ops->dev_dma_free(dev->dma_pool, addr, dma_handler);
655}
656EXPORT_SYMBOL(nvm_dev_dma_free);
657
658static struct nvm_dev *nvm_find_nvm_dev(const char *name)
659{
660 struct nvm_dev *dev;
661
662 list_for_each_entry(dev, &nvm_devices, devices)
663 if (!strcmp(name, dev->name))
664 return dev;
665
666 return NULL;
667}
668
669static int nvm_set_rqd_ppalist(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd,
670 const struct ppa_addr *ppas, int nr_ppas)
671{
672 struct nvm_dev *dev = tgt_dev->parent;
673 struct nvm_geo *geo = &tgt_dev->geo;
674 int i, plane_cnt, pl_idx;
675 struct ppa_addr ppa;
676
677 if (geo->pln_mode == NVM_PLANE_SINGLE && nr_ppas == 1) {
678 rqd->nr_ppas = nr_ppas;
679 rqd->ppa_addr = ppas[0];
680
681 return 0;
682 }
683
684 rqd->nr_ppas = nr_ppas;
685 rqd->ppa_list = nvm_dev_dma_alloc(dev, GFP_KERNEL, &rqd->dma_ppa_list);
686 if (!rqd->ppa_list) {
687 pr_err("nvm: failed to allocate dma memory\n");
688 return -ENOMEM;
689 }
690
691 plane_cnt = geo->pln_mode;
692 rqd->nr_ppas *= plane_cnt;
693
694 for (i = 0; i < nr_ppas; i++) {
695 for (pl_idx = 0; pl_idx < plane_cnt; pl_idx++) {
696 ppa = ppas[i];
697 ppa.g.pl = pl_idx;
698 rqd->ppa_list[(pl_idx * nr_ppas) + i] = ppa;
699 }
700 }
701
702 return 0;
703}
704
705static void nvm_free_rqd_ppalist(struct nvm_tgt_dev *tgt_dev,
706 struct nvm_rq *rqd)
707{
708 if (!rqd->ppa_list)
709 return;
710
711 nvm_dev_dma_free(tgt_dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
712}
713
714static int nvm_set_flags(struct nvm_geo *geo, struct nvm_rq *rqd)
715{
716 int flags = 0;
717
718 if (geo->version == NVM_OCSSD_SPEC_20)
719 return 0;
720
721 if (rqd->is_seq)
722 flags |= geo->pln_mode >> 1;
723
724 if (rqd->opcode == NVM_OP_PREAD)
725 flags |= (NVM_IO_SCRAMBLE_ENABLE | NVM_IO_SUSPEND);
726 else if (rqd->opcode == NVM_OP_PWRITE)
727 flags |= NVM_IO_SCRAMBLE_ENABLE;
728
729 return flags;
730}
731
732int nvm_submit_io(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
733{
734 struct nvm_dev *dev = tgt_dev->parent;
735 int ret;
736
737 if (!dev->ops->submit_io)
738 return -ENODEV;
739
740 nvm_rq_tgt_to_dev(tgt_dev, rqd);
741
742 rqd->dev = tgt_dev;
743 rqd->flags = nvm_set_flags(&tgt_dev->geo, rqd);
744
745 /* In case of error, fail with right address format */
746 ret = dev->ops->submit_io(dev, rqd);
747 if (ret)
748 nvm_rq_dev_to_tgt(tgt_dev, rqd);
749 return ret;
750}
751EXPORT_SYMBOL(nvm_submit_io);
752
753int nvm_submit_io_sync(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
754{
755 struct nvm_dev *dev = tgt_dev->parent;
756 int ret;
757
758 if (!dev->ops->submit_io_sync)
759 return -ENODEV;
760
761 nvm_rq_tgt_to_dev(tgt_dev, rqd);
762
763 rqd->dev = tgt_dev;
764 rqd->flags = nvm_set_flags(&tgt_dev->geo, rqd);
765
766 /* In case of error, fail with right address format */
767 ret = dev->ops->submit_io_sync(dev, rqd);
768 nvm_rq_dev_to_tgt(tgt_dev, rqd);
769
770 return ret;
771}
772EXPORT_SYMBOL(nvm_submit_io_sync);
773
774void nvm_end_io(struct nvm_rq *rqd)
775{
776 struct nvm_tgt_dev *tgt_dev = rqd->dev;
777
778 /* Convert address space */
779 if (tgt_dev)
780 nvm_rq_dev_to_tgt(tgt_dev, rqd);
781
782 if (rqd->end_io)
783 rqd->end_io(rqd);
784}
785EXPORT_SYMBOL(nvm_end_io);
786
787static int nvm_submit_io_sync_raw(struct nvm_dev *dev, struct nvm_rq *rqd)
788{
789 if (!dev->ops->submit_io_sync)
790 return -ENODEV;
791
792 rqd->flags = nvm_set_flags(&dev->geo, rqd);
793
794 return dev->ops->submit_io_sync(dev, rqd);
795}
796
797static int nvm_bb_chunk_sense(struct nvm_dev *dev, struct ppa_addr ppa)
798{
799 struct nvm_rq rqd = { NULL };
800 struct bio bio;
801 struct bio_vec bio_vec;
802 struct page *page;
803 int ret;
804
805 page = alloc_page(GFP_KERNEL);
806 if (!page)
807 return -ENOMEM;
808
809 bio_init(&bio, &bio_vec, 1);
810 bio_add_page(&bio, page, PAGE_SIZE, 0);
811 bio_set_op_attrs(&bio, REQ_OP_READ, 0);
812
813 rqd.bio = &bio;
814 rqd.opcode = NVM_OP_PREAD;
815 rqd.is_seq = 1;
816 rqd.nr_ppas = 1;
817 rqd.ppa_addr = generic_to_dev_addr(dev, ppa);
818
819 ret = nvm_submit_io_sync_raw(dev, &rqd);
820 if (ret)
821 return ret;
822
823 __free_page(page);
824
825 return rqd.error;
826}
827
828/*
829 * Scans a 1.2 chunk first and last page to determine if its state.
830 * If the chunk is found to be open, also scan it to update the write
831 * pointer.
832 */
833static int nvm_bb_chunk_scan(struct nvm_dev *dev, struct ppa_addr ppa,
834 struct nvm_chk_meta *meta)
835{
836 struct nvm_geo *geo = &dev->geo;
837 int ret, pg, pl;
838
839 /* sense first page */
840 ret = nvm_bb_chunk_sense(dev, ppa);
841 if (ret < 0) /* io error */
842 return ret;
843 else if (ret == 0) /* valid data */
844 meta->state = NVM_CHK_ST_OPEN;
845 else if (ret > 0) {
846 /*
847 * If empty page, the chunk is free, else it is an
848 * actual io error. In that case, mark it offline.
849 */
850 switch (ret) {
851 case NVM_RSP_ERR_EMPTYPAGE:
852 meta->state = NVM_CHK_ST_FREE;
853 return 0;
854 case NVM_RSP_ERR_FAILCRC:
855 case NVM_RSP_ERR_FAILECC:
856 case NVM_RSP_WARN_HIGHECC:
857 meta->state = NVM_CHK_ST_OPEN;
858 goto scan;
859 default:
860 return -ret; /* other io error */
861 }
862 }
863
864 /* sense last page */
865 ppa.g.pg = geo->num_pg - 1;
866 ppa.g.pl = geo->num_pln - 1;
867
868 ret = nvm_bb_chunk_sense(dev, ppa);
869 if (ret < 0) /* io error */
870 return ret;
871 else if (ret == 0) { /* Chunk fully written */
872 meta->state = NVM_CHK_ST_CLOSED;
873 meta->wp = geo->clba;
874 return 0;
875 } else if (ret > 0) {
876 switch (ret) {
877 case NVM_RSP_ERR_EMPTYPAGE:
878 case NVM_RSP_ERR_FAILCRC:
879 case NVM_RSP_ERR_FAILECC:
880 case NVM_RSP_WARN_HIGHECC:
881 meta->state = NVM_CHK_ST_OPEN;
882 break;
883 default:
884 return -ret; /* other io error */
885 }
886 }
887
888scan:
889 /*
890 * chunk is open, we scan sequentially to update the write pointer.
891 * We make the assumption that targets write data across all planes
892 * before moving to the next page.
893 */
894 for (pg = 0; pg < geo->num_pg; pg++) {
895 for (pl = 0; pl < geo->num_pln; pl++) {
896 ppa.g.pg = pg;
897 ppa.g.pl = pl;
898
899 ret = nvm_bb_chunk_sense(dev, ppa);
900 if (ret < 0) /* io error */
901 return ret;
902 else if (ret == 0) {
903 meta->wp += geo->ws_min;
904 } else if (ret > 0) {
905 switch (ret) {
906 case NVM_RSP_ERR_EMPTYPAGE:
907 return 0;
908 case NVM_RSP_ERR_FAILCRC:
909 case NVM_RSP_ERR_FAILECC:
910 case NVM_RSP_WARN_HIGHECC:
911 meta->wp += geo->ws_min;
912 break;
913 default:
914 return -ret; /* other io error */
915 }
916 }
917 }
918 }
919
920 return 0;
921}
922
923/*
924 * folds a bad block list from its plane representation to its
925 * chunk representation.
926 *
927 * If any of the planes status are bad or grown bad, the chunk is marked
928 * offline. If not bad, the first plane state acts as the chunk state.
929 */
930static int nvm_bb_to_chunk(struct nvm_dev *dev, struct ppa_addr ppa,
931 u8 *blks, int nr_blks, struct nvm_chk_meta *meta)
932{
933 struct nvm_geo *geo = &dev->geo;
934 int ret, blk, pl, offset, blktype;
935
936 for (blk = 0; blk < geo->num_chk; blk++) {
937 offset = blk * geo->pln_mode;
938 blktype = blks[offset];
939
940 for (pl = 0; pl < geo->pln_mode; pl++) {
941 if (blks[offset + pl] &
942 (NVM_BLK_T_BAD|NVM_BLK_T_GRWN_BAD)) {
943 blktype = blks[offset + pl];
944 break;
945 }
946 }
947
948 ppa.g.blk = blk;
949
950 meta->wp = 0;
951 meta->type = NVM_CHK_TP_W_SEQ;
952 meta->wi = 0;
953 meta->slba = generic_to_dev_addr(dev, ppa).ppa;
954 meta->cnlb = dev->geo.clba;
955
956 if (blktype == NVM_BLK_T_FREE) {
957 ret = nvm_bb_chunk_scan(dev, ppa, meta);
958 if (ret)
959 return ret;
960 } else {
961 meta->state = NVM_CHK_ST_OFFLINE;
962 }
963
964 meta++;
965 }
966
967 return 0;
968}
969
970static int nvm_get_bb_meta(struct nvm_dev *dev, sector_t slba,
971 int nchks, struct nvm_chk_meta *meta)
972{
973 struct nvm_geo *geo = &dev->geo;
974 struct ppa_addr ppa;
975 u8 *blks;
976 int ch, lun, nr_blks;
977 int ret = 0;
978
979 ppa.ppa = slba;
980 ppa = dev_to_generic_addr(dev, ppa);
981
982 if (ppa.g.blk != 0)
983 return -EINVAL;
984
985 if ((nchks % geo->num_chk) != 0)
986 return -EINVAL;
987
988 nr_blks = geo->num_chk * geo->pln_mode;
989
990 blks = kmalloc(nr_blks, GFP_KERNEL);
991 if (!blks)
992 return -ENOMEM;
993
994 for (ch = ppa.g.ch; ch < geo->num_ch; ch++) {
995 for (lun = ppa.g.lun; lun < geo->num_lun; lun++) {
996 struct ppa_addr ppa_gen, ppa_dev;
997
998 if (!nchks)
999 goto done;
1000
1001 ppa_gen.ppa = 0;
1002 ppa_gen.g.ch = ch;
1003 ppa_gen.g.lun = lun;
1004 ppa_dev = generic_to_dev_addr(dev, ppa_gen);
1005
1006 ret = dev->ops->get_bb_tbl(dev, ppa_dev, blks);
1007 if (ret)
1008 goto done;
1009
1010 ret = nvm_bb_to_chunk(dev, ppa_gen, blks, nr_blks,
1011 meta);
1012 if (ret)
1013 goto done;
1014
1015 meta += geo->num_chk;
1016 nchks -= geo->num_chk;
1017 }
1018 }
1019done:
1020 kfree(blks);
1021 return ret;
1022}
1023
1024int nvm_get_chunk_meta(struct nvm_tgt_dev *tgt_dev, struct ppa_addr ppa,
1025 int nchks, struct nvm_chk_meta *meta)
1026{
1027 struct nvm_dev *dev = tgt_dev->parent;
1028
1029 nvm_ppa_tgt_to_dev(tgt_dev, &ppa, 1);
1030
1031 if (dev->geo.version == NVM_OCSSD_SPEC_12)
1032 return nvm_get_bb_meta(dev, (sector_t)ppa.ppa, nchks, meta);
1033
1034 return dev->ops->get_chk_meta(dev, (sector_t)ppa.ppa, nchks, meta);
1035}
1036EXPORT_SYMBOL_GPL(nvm_get_chunk_meta);
1037
1038int nvm_set_chunk_meta(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *ppas,
1039 int nr_ppas, int type)
1040{
1041 struct nvm_dev *dev = tgt_dev->parent;
1042 struct nvm_rq rqd;
1043 int ret;
1044
1045 if (dev->geo.version == NVM_OCSSD_SPEC_20)
1046 return 0;
1047
1048 if (nr_ppas > NVM_MAX_VLBA) {
1049 pr_err("nvm: unable to update all blocks atomically\n");
1050 return -EINVAL;
1051 }
1052
1053 memset(&rqd, 0, sizeof(struct nvm_rq));
1054
1055 nvm_set_rqd_ppalist(tgt_dev, &rqd, ppas, nr_ppas);
1056 nvm_rq_tgt_to_dev(tgt_dev, &rqd);
1057
1058 ret = dev->ops->set_bb_tbl(dev, &rqd.ppa_addr, rqd.nr_ppas, type);
1059 nvm_free_rqd_ppalist(tgt_dev, &rqd);
1060 if (ret)
1061 return -EINVAL;
1062
1063 return 0;
1064}
1065EXPORT_SYMBOL_GPL(nvm_set_chunk_meta);
1066
1067static int nvm_core_init(struct nvm_dev *dev)
1068{
1069 struct nvm_geo *geo = &dev->geo;
1070 int ret;
1071
1072 dev->lun_map = kcalloc(BITS_TO_LONGS(geo->all_luns),
1073 sizeof(unsigned long), GFP_KERNEL);
1074 if (!dev->lun_map)
1075 return -ENOMEM;
1076
1077 INIT_LIST_HEAD(&dev->area_list);
1078 INIT_LIST_HEAD(&dev->targets);
1079 mutex_init(&dev->mlock);
1080 spin_lock_init(&dev->lock);
1081
1082 ret = nvm_register_map(dev);
1083 if (ret)
1084 goto err_fmtype;
1085
1086 return 0;
1087err_fmtype:
1088 kfree(dev->lun_map);
1089 return ret;
1090}
1091
1092static void nvm_free(struct nvm_dev *dev)
1093{
1094 if (!dev)
1095 return;
1096
1097 if (dev->dma_pool)
1098 dev->ops->destroy_dma_pool(dev->dma_pool);
1099
1100 nvm_unregister_map(dev);
1101 kfree(dev->lun_map);
1102 kfree(dev);
1103}
1104
1105static int nvm_init(struct nvm_dev *dev)
1106{
1107 struct nvm_geo *geo = &dev->geo;
1108 int ret = -EINVAL;
1109
1110 if (dev->ops->identity(dev)) {
1111 pr_err("nvm: device could not be identified\n");
1112 goto err;
1113 }
1114
1115 pr_debug("nvm: ver:%u.%u nvm_vendor:%x\n",
1116 geo->major_ver_id, geo->minor_ver_id,
1117 geo->vmnt);
1118
1119 ret = nvm_core_init(dev);
1120 if (ret) {
1121 pr_err("nvm: could not initialize core structures.\n");
1122 goto err;
1123 }
1124
1125 pr_info("nvm: registered %s [%u/%u/%u/%u/%u]\n",
1126 dev->name, dev->geo.ws_min, dev->geo.ws_opt,
1127 dev->geo.num_chk, dev->geo.all_luns,
1128 dev->geo.num_ch);
1129 return 0;
1130err:
1131 pr_err("nvm: failed to initialize nvm\n");
1132 return ret;
1133}
1134
1135struct nvm_dev *nvm_alloc_dev(int node)
1136{
1137 return kzalloc_node(sizeof(struct nvm_dev), GFP_KERNEL, node);
1138}
1139EXPORT_SYMBOL(nvm_alloc_dev);
1140
1141int nvm_register(struct nvm_dev *dev)
1142{
1143 int ret, exp_pool_size;
1144
1145 if (!dev->q || !dev->ops)
1146 return -EINVAL;
1147
1148 ret = nvm_init(dev);
1149 if (ret)
1150 return ret;
1151
1152 exp_pool_size = max_t(int, PAGE_SIZE,
1153 (NVM_MAX_VLBA * (sizeof(u64) + dev->geo.sos)));
1154 exp_pool_size = round_up(exp_pool_size, PAGE_SIZE);
1155
1156 dev->dma_pool = dev->ops->create_dma_pool(dev, "ppalist",
1157 exp_pool_size);
1158 if (!dev->dma_pool) {
1159 pr_err("nvm: could not create dma pool\n");
1160 nvm_free(dev);
1161 return -ENOMEM;
1162 }
1163
1164 /* register device with a supported media manager */
1165 down_write(&nvm_lock);
1166 list_add(&dev->devices, &nvm_devices);
1167 up_write(&nvm_lock);
1168
1169 return 0;
1170}
1171EXPORT_SYMBOL(nvm_register);
1172
1173void nvm_unregister(struct nvm_dev *dev)
1174{
1175 struct nvm_target *t, *tmp;
1176
1177 mutex_lock(&dev->mlock);
1178 list_for_each_entry_safe(t, tmp, &dev->targets, list) {
1179 if (t->dev->parent != dev)
1180 continue;
1181 __nvm_remove_target(t, false);
1182 }
1183 mutex_unlock(&dev->mlock);
1184
1185 down_write(&nvm_lock);
1186 list_del(&dev->devices);
1187 up_write(&nvm_lock);
1188
1189 nvm_free(dev);
1190}
1191EXPORT_SYMBOL(nvm_unregister);
1192
1193static int __nvm_configure_create(struct nvm_ioctl_create *create)
1194{
1195 struct nvm_dev *dev;
1196
1197 down_write(&nvm_lock);
1198 dev = nvm_find_nvm_dev(create->dev);
1199 up_write(&nvm_lock);
1200
1201 if (!dev) {
1202 pr_err("nvm: device not found\n");
1203 return -EINVAL;
1204 }
1205
1206 return nvm_create_tgt(dev, create);
1207}
1208
1209static long nvm_ioctl_info(struct file *file, void __user *arg)
1210{
1211 struct nvm_ioctl_info *info;
1212 struct nvm_tgt_type *tt;
1213 int tgt_iter = 0;
1214
1215 info = memdup_user(arg, sizeof(struct nvm_ioctl_info));
1216 if (IS_ERR(info))
1217 return -EFAULT;
1218
1219 info->version[0] = NVM_VERSION_MAJOR;
1220 info->version[1] = NVM_VERSION_MINOR;
1221 info->version[2] = NVM_VERSION_PATCH;
1222
1223 down_write(&nvm_tgtt_lock);
1224 list_for_each_entry(tt, &nvm_tgt_types, list) {
1225 struct nvm_ioctl_info_tgt *tgt = &info->tgts[tgt_iter];
1226
1227 tgt->version[0] = tt->version[0];
1228 tgt->version[1] = tt->version[1];
1229 tgt->version[2] = tt->version[2];
1230 strncpy(tgt->tgtname, tt->name, NVM_TTYPE_NAME_MAX);
1231
1232 tgt_iter++;
1233 }
1234
1235 info->tgtsize = tgt_iter;
1236 up_write(&nvm_tgtt_lock);
1237
1238 if (copy_to_user(arg, info, sizeof(struct nvm_ioctl_info))) {
1239 kfree(info);
1240 return -EFAULT;
1241 }
1242
1243 kfree(info);
1244 return 0;
1245}
1246
1247static long nvm_ioctl_get_devices(struct file *file, void __user *arg)
1248{
1249 struct nvm_ioctl_get_devices *devices;
1250 struct nvm_dev *dev;
1251 int i = 0;
1252
1253 devices = kzalloc(sizeof(struct nvm_ioctl_get_devices), GFP_KERNEL);
1254 if (!devices)
1255 return -ENOMEM;
1256
1257 down_write(&nvm_lock);
1258 list_for_each_entry(dev, &nvm_devices, devices) {
1259 struct nvm_ioctl_device_info *info = &devices->info[i];
1260
1261 strlcpy(info->devname, dev->name, sizeof(info->devname));
1262
1263 /* kept for compatibility */
1264 info->bmversion[0] = 1;
1265 info->bmversion[1] = 0;
1266 info->bmversion[2] = 0;
1267 strlcpy(info->bmname, "gennvm", sizeof(info->bmname));
1268 i++;
1269
1270 if (i > 31) {
1271 pr_err("nvm: max 31 devices can be reported.\n");
1272 break;
1273 }
1274 }
1275 up_write(&nvm_lock);
1276
1277 devices->nr_devices = i;
1278
1279 if (copy_to_user(arg, devices,
1280 sizeof(struct nvm_ioctl_get_devices))) {
1281 kfree(devices);
1282 return -EFAULT;
1283 }
1284
1285 kfree(devices);
1286 return 0;
1287}
1288
1289static long nvm_ioctl_dev_create(struct file *file, void __user *arg)
1290{
1291 struct nvm_ioctl_create create;
1292
1293 if (copy_from_user(&create, arg, sizeof(struct nvm_ioctl_create)))
1294 return -EFAULT;
1295
1296 if (create.conf.type == NVM_CONFIG_TYPE_EXTENDED &&
1297 create.conf.e.rsv != 0) {
1298 pr_err("nvm: reserved config field in use\n");
1299 return -EINVAL;
1300 }
1301
1302 create.dev[DISK_NAME_LEN - 1] = '\0';
1303 create.tgttype[NVM_TTYPE_NAME_MAX - 1] = '\0';
1304 create.tgtname[DISK_NAME_LEN - 1] = '\0';
1305
1306 if (create.flags != 0) {
1307 __u32 flags = create.flags;
1308
1309 /* Check for valid flags */
1310 if (flags & NVM_TARGET_FACTORY)
1311 flags &= ~NVM_TARGET_FACTORY;
1312
1313 if (flags) {
1314 pr_err("nvm: flag not supported\n");
1315 return -EINVAL;
1316 }
1317 }
1318
1319 return __nvm_configure_create(&create);
1320}
1321
1322static long nvm_ioctl_dev_remove(struct file *file, void __user *arg)
1323{
1324 struct nvm_ioctl_remove remove;
1325 struct nvm_dev *dev;
1326 int ret = 0;
1327
1328 if (copy_from_user(&remove, arg, sizeof(struct nvm_ioctl_remove)))
1329 return -EFAULT;
1330
1331 remove.tgtname[DISK_NAME_LEN - 1] = '\0';
1332
1333 if (remove.flags != 0) {
1334 pr_err("nvm: no flags supported\n");
1335 return -EINVAL;
1336 }
1337
1338 list_for_each_entry(dev, &nvm_devices, devices) {
1339 ret = nvm_remove_tgt(dev, &remove);
1340 if (!ret)
1341 break;
1342 }
1343
1344 return ret;
1345}
1346
1347/* kept for compatibility reasons */
1348static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
1349{
1350 struct nvm_ioctl_dev_init init;
1351
1352 if (copy_from_user(&init, arg, sizeof(struct nvm_ioctl_dev_init)))
1353 return -EFAULT;
1354
1355 if (init.flags != 0) {
1356 pr_err("nvm: no flags supported\n");
1357 return -EINVAL;
1358 }
1359
1360 return 0;
1361}
1362
1363/* Kept for compatibility reasons */
1364static long nvm_ioctl_dev_factory(struct file *file, void __user *arg)
1365{
1366 struct nvm_ioctl_dev_factory fact;
1367
1368 if (copy_from_user(&fact, arg, sizeof(struct nvm_ioctl_dev_factory)))
1369 return -EFAULT;
1370
1371 fact.dev[DISK_NAME_LEN - 1] = '\0';
1372
1373 if (fact.flags & ~(NVM_FACTORY_NR_BITS - 1))
1374 return -EINVAL;
1375
1376 return 0;
1377}
1378
1379static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
1380{
1381 void __user *argp = (void __user *)arg;
1382
1383 if (!capable(CAP_SYS_ADMIN))
1384 return -EPERM;
1385
1386 switch (cmd) {
1387 case NVM_INFO:
1388 return nvm_ioctl_info(file, argp);
1389 case NVM_GET_DEVICES:
1390 return nvm_ioctl_get_devices(file, argp);
1391 case NVM_DEV_CREATE:
1392 return nvm_ioctl_dev_create(file, argp);
1393 case NVM_DEV_REMOVE:
1394 return nvm_ioctl_dev_remove(file, argp);
1395 case NVM_DEV_INIT:
1396 return nvm_ioctl_dev_init(file, argp);
1397 case NVM_DEV_FACTORY:
1398 return nvm_ioctl_dev_factory(file, argp);
1399 }
1400 return 0;
1401}
1402
1403static const struct file_operations _ctl_fops = {
1404 .open = nonseekable_open,
1405 .unlocked_ioctl = nvm_ctl_ioctl,
1406 .owner = THIS_MODULE,
1407 .llseek = noop_llseek,
1408};
1409
1410static struct miscdevice _nvm_misc = {
1411 .minor = MISC_DYNAMIC_MINOR,
1412 .name = "lightnvm",
1413 .nodename = "lightnvm/control",
1414 .fops = &_ctl_fops,
1415};
1416builtin_misc_device(_nvm_misc);
1417