Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / lightnvm / core.c
blobc1bcac71008c67374ad76e9ea455cc220f8bba31
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2015 IT University of Copenhagen. All rights reserved.
4 * Initial release: Matias Bjorling <m@bjorling.me>
5 */
7 #define pr_fmt(fmt) "nvm: " fmt
9 #include <linux/list.h>
10 #include <linux/types.h>
11 #include <linux/sem.h>
12 #include <linux/bitmap.h>
13 #include <linux/module.h>
14 #include <linux/moduleparam.h>
15 #include <linux/miscdevice.h>
16 #include <linux/lightnvm.h>
17 #include <linux/sched/sysctl.h>
19 static LIST_HEAD(nvm_tgt_types);
20 static DECLARE_RWSEM(nvm_tgtt_lock);
21 static LIST_HEAD(nvm_devices);
22 static DECLARE_RWSEM(nvm_lock);
24 /* Map between virtual and physical channel and lun */
25 struct nvm_ch_map {
26 int ch_off;
27 int num_lun;
28 int *lun_offs;
31 struct nvm_dev_map {
32 struct nvm_ch_map *chnls;
33 int num_ch;
36 static void nvm_free(struct kref *ref);
38 static struct nvm_target *nvm_find_target(struct nvm_dev *dev, const char *name)
40 struct nvm_target *tgt;
42 list_for_each_entry(tgt, &dev->targets, list)
43 if (!strcmp(name, tgt->disk->disk_name))
44 return tgt;
46 return NULL;
49 static bool nvm_target_exists(const char *name)
51 struct nvm_dev *dev;
52 struct nvm_target *tgt;
53 bool ret = false;
55 down_write(&nvm_lock);
56 list_for_each_entry(dev, &nvm_devices, devices) {
57 mutex_lock(&dev->mlock);
58 list_for_each_entry(tgt, &dev->targets, list) {
59 if (!strcmp(name, tgt->disk->disk_name)) {
60 ret = true;
61 mutex_unlock(&dev->mlock);
62 goto out;
65 mutex_unlock(&dev->mlock);
68 out:
69 up_write(&nvm_lock);
70 return ret;
73 static int nvm_reserve_luns(struct nvm_dev *dev, int lun_begin, int lun_end)
75 int i;
77 for (i = lun_begin; i <= lun_end; i++) {
78 if (test_and_set_bit(i, dev->lun_map)) {
79 pr_err("lun %d already allocated\n", i);
80 goto err;
84 return 0;
85 err:
86 while (--i >= lun_begin)
87 clear_bit(i, dev->lun_map);
89 return -EBUSY;
92 static void nvm_release_luns_err(struct nvm_dev *dev, int lun_begin,
93 int lun_end)
95 int i;
97 for (i = lun_begin; i <= lun_end; i++)
98 WARN_ON(!test_and_clear_bit(i, dev->lun_map));
101 static void nvm_remove_tgt_dev(struct nvm_tgt_dev *tgt_dev, int clear)
103 struct nvm_dev *dev = tgt_dev->parent;
104 struct nvm_dev_map *dev_map = tgt_dev->map;
105 int i, j;
107 for (i = 0; i < dev_map->num_ch; i++) {
108 struct nvm_ch_map *ch_map = &dev_map->chnls[i];
109 int *lun_offs = ch_map->lun_offs;
110 int ch = i + ch_map->ch_off;
112 if (clear) {
113 for (j = 0; j < ch_map->num_lun; j++) {
114 int lun = j + lun_offs[j];
115 int lunid = (ch * dev->geo.num_lun) + lun;
117 WARN_ON(!test_and_clear_bit(lunid,
118 dev->lun_map));
122 kfree(ch_map->lun_offs);
125 kfree(dev_map->chnls);
126 kfree(dev_map);
128 kfree(tgt_dev->luns);
129 kfree(tgt_dev);
132 static struct nvm_tgt_dev *nvm_create_tgt_dev(struct nvm_dev *dev,
133 u16 lun_begin, u16 lun_end,
134 u16 op)
136 struct nvm_tgt_dev *tgt_dev = NULL;
137 struct nvm_dev_map *dev_rmap = dev->rmap;
138 struct nvm_dev_map *dev_map;
139 struct ppa_addr *luns;
140 int num_lun = lun_end - lun_begin + 1;
141 int luns_left = num_lun;
142 int num_ch = num_lun / dev->geo.num_lun;
143 int num_ch_mod = num_lun % dev->geo.num_lun;
144 int bch = lun_begin / dev->geo.num_lun;
145 int blun = lun_begin % dev->geo.num_lun;
146 int lunid = 0;
147 int lun_balanced = 1;
148 int sec_per_lun, prev_num_lun;
149 int i, j;
151 num_ch = (num_ch_mod == 0) ? num_ch : num_ch + 1;
153 dev_map = kmalloc(sizeof(struct nvm_dev_map), GFP_KERNEL);
154 if (!dev_map)
155 goto err_dev;
157 dev_map->chnls = kcalloc(num_ch, sizeof(struct nvm_ch_map), GFP_KERNEL);
158 if (!dev_map->chnls)
159 goto err_chnls;
161 luns = kcalloc(num_lun, sizeof(struct ppa_addr), GFP_KERNEL);
162 if (!luns)
163 goto err_luns;
165 prev_num_lun = (luns_left > dev->geo.num_lun) ?
166 dev->geo.num_lun : luns_left;
167 for (i = 0; i < num_ch; i++) {
168 struct nvm_ch_map *ch_rmap = &dev_rmap->chnls[i + bch];
169 int *lun_roffs = ch_rmap->lun_offs;
170 struct nvm_ch_map *ch_map = &dev_map->chnls[i];
171 int *lun_offs;
172 int luns_in_chnl = (luns_left > dev->geo.num_lun) ?
173 dev->geo.num_lun : luns_left;
175 if (lun_balanced && prev_num_lun != luns_in_chnl)
176 lun_balanced = 0;
178 ch_map->ch_off = ch_rmap->ch_off = bch;
179 ch_map->num_lun = luns_in_chnl;
181 lun_offs = kcalloc(luns_in_chnl, sizeof(int), GFP_KERNEL);
182 if (!lun_offs)
183 goto err_ch;
185 for (j = 0; j < luns_in_chnl; j++) {
186 luns[lunid].ppa = 0;
187 luns[lunid].a.ch = i;
188 luns[lunid++].a.lun = j;
190 lun_offs[j] = blun;
191 lun_roffs[j + blun] = blun;
194 ch_map->lun_offs = lun_offs;
196 /* when starting a new channel, lun offset is reset */
197 blun = 0;
198 luns_left -= luns_in_chnl;
201 dev_map->num_ch = num_ch;
203 tgt_dev = kmalloc(sizeof(struct nvm_tgt_dev), GFP_KERNEL);
204 if (!tgt_dev)
205 goto err_ch;
207 /* Inherit device geometry from parent */
208 memcpy(&tgt_dev->geo, &dev->geo, sizeof(struct nvm_geo));
210 /* Target device only owns a portion of the physical device */
211 tgt_dev->geo.num_ch = num_ch;
212 tgt_dev->geo.num_lun = (lun_balanced) ? prev_num_lun : -1;
213 tgt_dev->geo.all_luns = num_lun;
214 tgt_dev->geo.all_chunks = num_lun * dev->geo.num_chk;
216 tgt_dev->geo.op = op;
218 sec_per_lun = dev->geo.clba * dev->geo.num_chk;
219 tgt_dev->geo.total_secs = num_lun * sec_per_lun;
221 tgt_dev->q = dev->q;
222 tgt_dev->map = dev_map;
223 tgt_dev->luns = luns;
224 tgt_dev->parent = dev;
226 return tgt_dev;
227 err_ch:
228 while (--i >= 0)
229 kfree(dev_map->chnls[i].lun_offs);
230 kfree(luns);
231 err_luns:
232 kfree(dev_map->chnls);
233 err_chnls:
234 kfree(dev_map);
235 err_dev:
236 return tgt_dev;
239 static struct nvm_tgt_type *__nvm_find_target_type(const char *name)
241 struct nvm_tgt_type *tt;
243 list_for_each_entry(tt, &nvm_tgt_types, list)
244 if (!strcmp(name, tt->name))
245 return tt;
247 return NULL;
250 static struct nvm_tgt_type *nvm_find_target_type(const char *name)
252 struct nvm_tgt_type *tt;
254 down_write(&nvm_tgtt_lock);
255 tt = __nvm_find_target_type(name);
256 up_write(&nvm_tgtt_lock);
258 return tt;
261 static int nvm_config_check_luns(struct nvm_geo *geo, int lun_begin,
262 int lun_end)
264 if (lun_begin > lun_end || lun_end >= geo->all_luns) {
265 pr_err("lun out of bound (%u:%u > %u)\n",
266 lun_begin, lun_end, geo->all_luns - 1);
267 return -EINVAL;
270 return 0;
273 static int __nvm_config_simple(struct nvm_dev *dev,
274 struct nvm_ioctl_create_simple *s)
276 struct nvm_geo *geo = &dev->geo;
278 if (s->lun_begin == -1 && s->lun_end == -1) {
279 s->lun_begin = 0;
280 s->lun_end = geo->all_luns - 1;
283 return nvm_config_check_luns(geo, s->lun_begin, s->lun_end);
286 static int __nvm_config_extended(struct nvm_dev *dev,
287 struct nvm_ioctl_create_extended *e)
289 if (e->lun_begin == 0xFFFF && e->lun_end == 0xFFFF) {
290 e->lun_begin = 0;
291 e->lun_end = dev->geo.all_luns - 1;
294 /* op not set falls into target's default */
295 if (e->op == 0xFFFF) {
296 e->op = NVM_TARGET_DEFAULT_OP;
297 } else if (e->op < NVM_TARGET_MIN_OP || e->op > NVM_TARGET_MAX_OP) {
298 pr_err("invalid over provisioning value\n");
299 return -EINVAL;
302 return nvm_config_check_luns(&dev->geo, e->lun_begin, e->lun_end);
305 static int nvm_create_tgt(struct nvm_dev *dev, struct nvm_ioctl_create *create)
307 struct nvm_ioctl_create_extended e;
308 struct request_queue *tqueue;
309 struct gendisk *tdisk;
310 struct nvm_tgt_type *tt;
311 struct nvm_target *t;
312 struct nvm_tgt_dev *tgt_dev;
313 void *targetdata;
314 unsigned int mdts;
315 int ret;
317 switch (create->conf.type) {
318 case NVM_CONFIG_TYPE_SIMPLE:
319 ret = __nvm_config_simple(dev, &create->conf.s);
320 if (ret)
321 return ret;
323 e.lun_begin = create->conf.s.lun_begin;
324 e.lun_end = create->conf.s.lun_end;
325 e.op = NVM_TARGET_DEFAULT_OP;
326 break;
327 case NVM_CONFIG_TYPE_EXTENDED:
328 ret = __nvm_config_extended(dev, &create->conf.e);
329 if (ret)
330 return ret;
332 e = create->conf.e;
333 break;
334 default:
335 pr_err("config type not valid\n");
336 return -EINVAL;
339 tt = nvm_find_target_type(create->tgttype);
340 if (!tt) {
341 pr_err("target type %s not found\n", create->tgttype);
342 return -EINVAL;
345 if ((tt->flags & NVM_TGT_F_HOST_L2P) != (dev->geo.dom & NVM_RSP_L2P)) {
346 pr_err("device is incompatible with target L2P type.\n");
347 return -EINVAL;
350 if (nvm_target_exists(create->tgtname)) {
351 pr_err("target name already exists (%s)\n",
352 create->tgtname);
353 return -EINVAL;
356 ret = nvm_reserve_luns(dev, e.lun_begin, e.lun_end);
357 if (ret)
358 return ret;
360 t = kmalloc(sizeof(struct nvm_target), GFP_KERNEL);
361 if (!t) {
362 ret = -ENOMEM;
363 goto err_reserve;
366 tgt_dev = nvm_create_tgt_dev(dev, e.lun_begin, e.lun_end, e.op);
367 if (!tgt_dev) {
368 pr_err("could not create target device\n");
369 ret = -ENOMEM;
370 goto err_t;
373 tdisk = alloc_disk(0);
374 if (!tdisk) {
375 ret = -ENOMEM;
376 goto err_dev;
379 tqueue = blk_alloc_queue(dev->q->node);
380 if (!tqueue) {
381 ret = -ENOMEM;
382 goto err_disk;
385 strlcpy(tdisk->disk_name, create->tgtname, sizeof(tdisk->disk_name));
386 tdisk->flags = GENHD_FL_EXT_DEVT;
387 tdisk->major = 0;
388 tdisk->first_minor = 0;
389 tdisk->fops = tt->bops;
390 tdisk->queue = tqueue;
392 targetdata = tt->init(tgt_dev, tdisk, create->flags);
393 if (IS_ERR(targetdata)) {
394 ret = PTR_ERR(targetdata);
395 goto err_init;
398 tdisk->private_data = targetdata;
399 tqueue->queuedata = targetdata;
401 mdts = (dev->geo.csecs >> 9) * NVM_MAX_VLBA;
402 if (dev->geo.mdts) {
403 mdts = min_t(u32, dev->geo.mdts,
404 (dev->geo.csecs >> 9) * NVM_MAX_VLBA);
406 blk_queue_max_hw_sectors(tqueue, mdts);
408 set_capacity(tdisk, tt->capacity(targetdata));
409 add_disk(tdisk);
411 if (tt->sysfs_init && tt->sysfs_init(tdisk)) {
412 ret = -ENOMEM;
413 goto err_sysfs;
416 t->type = tt;
417 t->disk = tdisk;
418 t->dev = tgt_dev;
420 mutex_lock(&dev->mlock);
421 list_add_tail(&t->list, &dev->targets);
422 mutex_unlock(&dev->mlock);
424 __module_get(tt->owner);
426 return 0;
427 err_sysfs:
428 if (tt->exit)
429 tt->exit(targetdata, true);
430 err_init:
431 blk_cleanup_queue(tqueue);
432 tdisk->queue = NULL;
433 err_disk:
434 put_disk(tdisk);
435 err_dev:
436 nvm_remove_tgt_dev(tgt_dev, 0);
437 err_t:
438 kfree(t);
439 err_reserve:
440 nvm_release_luns_err(dev, e.lun_begin, e.lun_end);
441 return ret;
444 static void __nvm_remove_target(struct nvm_target *t, bool graceful)
446 struct nvm_tgt_type *tt = t->type;
447 struct gendisk *tdisk = t->disk;
448 struct request_queue *q = tdisk->queue;
450 del_gendisk(tdisk);
451 blk_cleanup_queue(q);
453 if (tt->sysfs_exit)
454 tt->sysfs_exit(tdisk);
456 if (tt->exit)
457 tt->exit(tdisk->private_data, graceful);
459 nvm_remove_tgt_dev(t->dev, 1);
460 put_disk(tdisk);
461 module_put(t->type->owner);
463 list_del(&t->list);
464 kfree(t);
468 * nvm_remove_tgt - Removes a target from the media manager
469 * @remove: ioctl structure with target name to remove.
471 * Returns:
472 * 0: on success
473 * 1: on not found
474 * <0: on error
476 static int nvm_remove_tgt(struct nvm_ioctl_remove *remove)
478 struct nvm_target *t = NULL;
479 struct nvm_dev *dev;
481 down_read(&nvm_lock);
482 list_for_each_entry(dev, &nvm_devices, devices) {
483 mutex_lock(&dev->mlock);
484 t = nvm_find_target(dev, remove->tgtname);
485 if (t) {
486 mutex_unlock(&dev->mlock);
487 break;
489 mutex_unlock(&dev->mlock);
491 up_read(&nvm_lock);
493 if (!t) {
494 pr_err("failed to remove target %s\n",
495 remove->tgtname);
496 return 1;
499 __nvm_remove_target(t, true);
500 kref_put(&dev->ref, nvm_free);
502 return 0;
505 static int nvm_register_map(struct nvm_dev *dev)
507 struct nvm_dev_map *rmap;
508 int i, j;
510 rmap = kmalloc(sizeof(struct nvm_dev_map), GFP_KERNEL);
511 if (!rmap)
512 goto err_rmap;
514 rmap->chnls = kcalloc(dev->geo.num_ch, sizeof(struct nvm_ch_map),
515 GFP_KERNEL);
516 if (!rmap->chnls)
517 goto err_chnls;
519 for (i = 0; i < dev->geo.num_ch; i++) {
520 struct nvm_ch_map *ch_rmap;
521 int *lun_roffs;
522 int luns_in_chnl = dev->geo.num_lun;
524 ch_rmap = &rmap->chnls[i];
526 ch_rmap->ch_off = -1;
527 ch_rmap->num_lun = luns_in_chnl;
529 lun_roffs = kcalloc(luns_in_chnl, sizeof(int), GFP_KERNEL);
530 if (!lun_roffs)
531 goto err_ch;
533 for (j = 0; j < luns_in_chnl; j++)
534 lun_roffs[j] = -1;
536 ch_rmap->lun_offs = lun_roffs;
539 dev->rmap = rmap;
541 return 0;
542 err_ch:
543 while (--i >= 0)
544 kfree(rmap->chnls[i].lun_offs);
545 err_chnls:
546 kfree(rmap);
547 err_rmap:
548 return -ENOMEM;
551 static void nvm_unregister_map(struct nvm_dev *dev)
553 struct nvm_dev_map *rmap = dev->rmap;
554 int i;
556 for (i = 0; i < dev->geo.num_ch; i++)
557 kfree(rmap->chnls[i].lun_offs);
559 kfree(rmap->chnls);
560 kfree(rmap);
563 static void nvm_map_to_dev(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *p)
565 struct nvm_dev_map *dev_map = tgt_dev->map;
566 struct nvm_ch_map *ch_map = &dev_map->chnls[p->a.ch];
567 int lun_off = ch_map->lun_offs[p->a.lun];
569 p->a.ch += ch_map->ch_off;
570 p->a.lun += lun_off;
573 static void nvm_map_to_tgt(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *p)
575 struct nvm_dev *dev = tgt_dev->parent;
576 struct nvm_dev_map *dev_rmap = dev->rmap;
577 struct nvm_ch_map *ch_rmap = &dev_rmap->chnls[p->a.ch];
578 int lun_roff = ch_rmap->lun_offs[p->a.lun];
580 p->a.ch -= ch_rmap->ch_off;
581 p->a.lun -= lun_roff;
584 static void nvm_ppa_tgt_to_dev(struct nvm_tgt_dev *tgt_dev,
585 struct ppa_addr *ppa_list, int nr_ppas)
587 int i;
589 for (i = 0; i < nr_ppas; i++) {
590 nvm_map_to_dev(tgt_dev, &ppa_list[i]);
591 ppa_list[i] = generic_to_dev_addr(tgt_dev->parent, ppa_list[i]);
595 static void nvm_ppa_dev_to_tgt(struct nvm_tgt_dev *tgt_dev,
596 struct ppa_addr *ppa_list, int nr_ppas)
598 int i;
600 for (i = 0; i < nr_ppas; i++) {
601 ppa_list[i] = dev_to_generic_addr(tgt_dev->parent, ppa_list[i]);
602 nvm_map_to_tgt(tgt_dev, &ppa_list[i]);
606 static void nvm_rq_tgt_to_dev(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
608 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
610 nvm_ppa_tgt_to_dev(tgt_dev, ppa_list, rqd->nr_ppas);
613 static void nvm_rq_dev_to_tgt(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
615 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
617 nvm_ppa_dev_to_tgt(tgt_dev, ppa_list, rqd->nr_ppas);
620 int nvm_register_tgt_type(struct nvm_tgt_type *tt)
622 int ret = 0;
624 down_write(&nvm_tgtt_lock);
625 if (__nvm_find_target_type(tt->name))
626 ret = -EEXIST;
627 else
628 list_add(&tt->list, &nvm_tgt_types);
629 up_write(&nvm_tgtt_lock);
631 return ret;
633 EXPORT_SYMBOL(nvm_register_tgt_type);
635 void nvm_unregister_tgt_type(struct nvm_tgt_type *tt)
637 if (!tt)
638 return;
640 down_write(&nvm_tgtt_lock);
641 list_del(&tt->list);
642 up_write(&nvm_tgtt_lock);
644 EXPORT_SYMBOL(nvm_unregister_tgt_type);
646 void *nvm_dev_dma_alloc(struct nvm_dev *dev, gfp_t mem_flags,
647 dma_addr_t *dma_handler)
649 return dev->ops->dev_dma_alloc(dev, dev->dma_pool, mem_flags,
650 dma_handler);
652 EXPORT_SYMBOL(nvm_dev_dma_alloc);
654 void nvm_dev_dma_free(struct nvm_dev *dev, void *addr, dma_addr_t dma_handler)
656 dev->ops->dev_dma_free(dev->dma_pool, addr, dma_handler);
658 EXPORT_SYMBOL(nvm_dev_dma_free);
660 static struct nvm_dev *nvm_find_nvm_dev(const char *name)
662 struct nvm_dev *dev;
664 list_for_each_entry(dev, &nvm_devices, devices)
665 if (!strcmp(name, dev->name))
666 return dev;
668 return NULL;
671 static int nvm_set_rqd_ppalist(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd,
672 const struct ppa_addr *ppas, int nr_ppas)
674 struct nvm_dev *dev = tgt_dev->parent;
675 struct nvm_geo *geo = &tgt_dev->geo;
676 int i, plane_cnt, pl_idx;
677 struct ppa_addr ppa;
679 if (geo->pln_mode == NVM_PLANE_SINGLE && nr_ppas == 1) {
680 rqd->nr_ppas = nr_ppas;
681 rqd->ppa_addr = ppas[0];
683 return 0;
686 rqd->nr_ppas = nr_ppas;
687 rqd->ppa_list = nvm_dev_dma_alloc(dev, GFP_KERNEL, &rqd->dma_ppa_list);
688 if (!rqd->ppa_list) {
689 pr_err("failed to allocate dma memory\n");
690 return -ENOMEM;
693 plane_cnt = geo->pln_mode;
694 rqd->nr_ppas *= plane_cnt;
696 for (i = 0; i < nr_ppas; i++) {
697 for (pl_idx = 0; pl_idx < plane_cnt; pl_idx++) {
698 ppa = ppas[i];
699 ppa.g.pl = pl_idx;
700 rqd->ppa_list[(pl_idx * nr_ppas) + i] = ppa;
704 return 0;
707 static void nvm_free_rqd_ppalist(struct nvm_tgt_dev *tgt_dev,
708 struct nvm_rq *rqd)
710 if (!rqd->ppa_list)
711 return;
713 nvm_dev_dma_free(tgt_dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
716 static int nvm_set_flags(struct nvm_geo *geo, struct nvm_rq *rqd)
718 int flags = 0;
720 if (geo->version == NVM_OCSSD_SPEC_20)
721 return 0;
723 if (rqd->is_seq)
724 flags |= geo->pln_mode >> 1;
726 if (rqd->opcode == NVM_OP_PREAD)
727 flags |= (NVM_IO_SCRAMBLE_ENABLE | NVM_IO_SUSPEND);
728 else if (rqd->opcode == NVM_OP_PWRITE)
729 flags |= NVM_IO_SCRAMBLE_ENABLE;
731 return flags;
734 int nvm_submit_io(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd, void *buf)
736 struct nvm_dev *dev = tgt_dev->parent;
737 int ret;
739 if (!dev->ops->submit_io)
740 return -ENODEV;
742 nvm_rq_tgt_to_dev(tgt_dev, rqd);
744 rqd->dev = tgt_dev;
745 rqd->flags = nvm_set_flags(&tgt_dev->geo, rqd);
747 /* In case of error, fail with right address format */
748 ret = dev->ops->submit_io(dev, rqd, buf);
749 if (ret)
750 nvm_rq_dev_to_tgt(tgt_dev, rqd);
751 return ret;
753 EXPORT_SYMBOL(nvm_submit_io);
755 static void nvm_sync_end_io(struct nvm_rq *rqd)
757 struct completion *waiting = rqd->private;
759 complete(waiting);
762 static int nvm_submit_io_wait(struct nvm_dev *dev, struct nvm_rq *rqd,
763 void *buf)
765 DECLARE_COMPLETION_ONSTACK(wait);
766 int ret = 0;
768 rqd->end_io = nvm_sync_end_io;
769 rqd->private = &wait;
771 ret = dev->ops->submit_io(dev, rqd, buf);
772 if (ret)
773 return ret;
775 wait_for_completion_io(&wait);
777 return 0;
780 int nvm_submit_io_sync(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd,
781 void *buf)
783 struct nvm_dev *dev = tgt_dev->parent;
784 int ret;
786 if (!dev->ops->submit_io)
787 return -ENODEV;
789 nvm_rq_tgt_to_dev(tgt_dev, rqd);
791 rqd->dev = tgt_dev;
792 rqd->flags = nvm_set_flags(&tgt_dev->geo, rqd);
794 ret = nvm_submit_io_wait(dev, rqd, buf);
796 return ret;
798 EXPORT_SYMBOL(nvm_submit_io_sync);
800 void nvm_end_io(struct nvm_rq *rqd)
802 struct nvm_tgt_dev *tgt_dev = rqd->dev;
804 /* Convert address space */
805 if (tgt_dev)
806 nvm_rq_dev_to_tgt(tgt_dev, rqd);
808 if (rqd->end_io)
809 rqd->end_io(rqd);
811 EXPORT_SYMBOL(nvm_end_io);
813 static int nvm_submit_io_sync_raw(struct nvm_dev *dev, struct nvm_rq *rqd)
815 if (!dev->ops->submit_io)
816 return -ENODEV;
818 rqd->dev = NULL;
819 rqd->flags = nvm_set_flags(&dev->geo, rqd);
821 return nvm_submit_io_wait(dev, rqd, NULL);
824 static int nvm_bb_chunk_sense(struct nvm_dev *dev, struct ppa_addr ppa)
826 struct nvm_rq rqd = { NULL };
827 struct bio bio;
828 struct bio_vec bio_vec;
829 struct page *page;
830 int ret;
832 page = alloc_page(GFP_KERNEL);
833 if (!page)
834 return -ENOMEM;
836 bio_init(&bio, &bio_vec, 1);
837 bio_add_page(&bio, page, PAGE_SIZE, 0);
838 bio_set_op_attrs(&bio, REQ_OP_READ, 0);
840 rqd.bio = &bio;
841 rqd.opcode = NVM_OP_PREAD;
842 rqd.is_seq = 1;
843 rqd.nr_ppas = 1;
844 rqd.ppa_addr = generic_to_dev_addr(dev, ppa);
846 ret = nvm_submit_io_sync_raw(dev, &rqd);
847 if (ret)
848 return ret;
850 __free_page(page);
852 return rqd.error;
856 * Scans a 1.2 chunk first and last page to determine if its state.
857 * If the chunk is found to be open, also scan it to update the write
858 * pointer.
860 static int nvm_bb_chunk_scan(struct nvm_dev *dev, struct ppa_addr ppa,
861 struct nvm_chk_meta *meta)
863 struct nvm_geo *geo = &dev->geo;
864 int ret, pg, pl;
866 /* sense first page */
867 ret = nvm_bb_chunk_sense(dev, ppa);
868 if (ret < 0) /* io error */
869 return ret;
870 else if (ret == 0) /* valid data */
871 meta->state = NVM_CHK_ST_OPEN;
872 else if (ret > 0) {
874 * If empty page, the chunk is free, else it is an
875 * actual io error. In that case, mark it offline.
877 switch (ret) {
878 case NVM_RSP_ERR_EMPTYPAGE:
879 meta->state = NVM_CHK_ST_FREE;
880 return 0;
881 case NVM_RSP_ERR_FAILCRC:
882 case NVM_RSP_ERR_FAILECC:
883 case NVM_RSP_WARN_HIGHECC:
884 meta->state = NVM_CHK_ST_OPEN;
885 goto scan;
886 default:
887 return -ret; /* other io error */
891 /* sense last page */
892 ppa.g.pg = geo->num_pg - 1;
893 ppa.g.pl = geo->num_pln - 1;
895 ret = nvm_bb_chunk_sense(dev, ppa);
896 if (ret < 0) /* io error */
897 return ret;
898 else if (ret == 0) { /* Chunk fully written */
899 meta->state = NVM_CHK_ST_CLOSED;
900 meta->wp = geo->clba;
901 return 0;
902 } else if (ret > 0) {
903 switch (ret) {
904 case NVM_RSP_ERR_EMPTYPAGE:
905 case NVM_RSP_ERR_FAILCRC:
906 case NVM_RSP_ERR_FAILECC:
907 case NVM_RSP_WARN_HIGHECC:
908 meta->state = NVM_CHK_ST_OPEN;
909 break;
910 default:
911 return -ret; /* other io error */
915 scan:
917 * chunk is open, we scan sequentially to update the write pointer.
918 * We make the assumption that targets write data across all planes
919 * before moving to the next page.
921 for (pg = 0; pg < geo->num_pg; pg++) {
922 for (pl = 0; pl < geo->num_pln; pl++) {
923 ppa.g.pg = pg;
924 ppa.g.pl = pl;
926 ret = nvm_bb_chunk_sense(dev, ppa);
927 if (ret < 0) /* io error */
928 return ret;
929 else if (ret == 0) {
930 meta->wp += geo->ws_min;
931 } else if (ret > 0) {
932 switch (ret) {
933 case NVM_RSP_ERR_EMPTYPAGE:
934 return 0;
935 case NVM_RSP_ERR_FAILCRC:
936 case NVM_RSP_ERR_FAILECC:
937 case NVM_RSP_WARN_HIGHECC:
938 meta->wp += geo->ws_min;
939 break;
940 default:
941 return -ret; /* other io error */
947 return 0;
951 * folds a bad block list from its plane representation to its
952 * chunk representation.
954 * If any of the planes status are bad or grown bad, the chunk is marked
955 * offline. If not bad, the first plane state acts as the chunk state.
957 static int nvm_bb_to_chunk(struct nvm_dev *dev, struct ppa_addr ppa,
958 u8 *blks, int nr_blks, struct nvm_chk_meta *meta)
960 struct nvm_geo *geo = &dev->geo;
961 int ret, blk, pl, offset, blktype;
963 for (blk = 0; blk < geo->num_chk; blk++) {
964 offset = blk * geo->pln_mode;
965 blktype = blks[offset];
967 for (pl = 0; pl < geo->pln_mode; pl++) {
968 if (blks[offset + pl] &
969 (NVM_BLK_T_BAD|NVM_BLK_T_GRWN_BAD)) {
970 blktype = blks[offset + pl];
971 break;
975 ppa.g.blk = blk;
977 meta->wp = 0;
978 meta->type = NVM_CHK_TP_W_SEQ;
979 meta->wi = 0;
980 meta->slba = generic_to_dev_addr(dev, ppa).ppa;
981 meta->cnlb = dev->geo.clba;
983 if (blktype == NVM_BLK_T_FREE) {
984 ret = nvm_bb_chunk_scan(dev, ppa, meta);
985 if (ret)
986 return ret;
987 } else {
988 meta->state = NVM_CHK_ST_OFFLINE;
991 meta++;
994 return 0;
997 static int nvm_get_bb_meta(struct nvm_dev *dev, sector_t slba,
998 int nchks, struct nvm_chk_meta *meta)
1000 struct nvm_geo *geo = &dev->geo;
1001 struct ppa_addr ppa;
1002 u8 *blks;
1003 int ch, lun, nr_blks;
1004 int ret = 0;
1006 ppa.ppa = slba;
1007 ppa = dev_to_generic_addr(dev, ppa);
1009 if (ppa.g.blk != 0)
1010 return -EINVAL;
1012 if ((nchks % geo->num_chk) != 0)
1013 return -EINVAL;
1015 nr_blks = geo->num_chk * geo->pln_mode;
1017 blks = kmalloc(nr_blks, GFP_KERNEL);
1018 if (!blks)
1019 return -ENOMEM;
1021 for (ch = ppa.g.ch; ch < geo->num_ch; ch++) {
1022 for (lun = ppa.g.lun; lun < geo->num_lun; lun++) {
1023 struct ppa_addr ppa_gen, ppa_dev;
1025 if (!nchks)
1026 goto done;
1028 ppa_gen.ppa = 0;
1029 ppa_gen.g.ch = ch;
1030 ppa_gen.g.lun = lun;
1031 ppa_dev = generic_to_dev_addr(dev, ppa_gen);
1033 ret = dev->ops->get_bb_tbl(dev, ppa_dev, blks);
1034 if (ret)
1035 goto done;
1037 ret = nvm_bb_to_chunk(dev, ppa_gen, blks, nr_blks,
1038 meta);
1039 if (ret)
1040 goto done;
1042 meta += geo->num_chk;
1043 nchks -= geo->num_chk;
1046 done:
1047 kfree(blks);
1048 return ret;
1051 int nvm_get_chunk_meta(struct nvm_tgt_dev *tgt_dev, struct ppa_addr ppa,
1052 int nchks, struct nvm_chk_meta *meta)
1054 struct nvm_dev *dev = tgt_dev->parent;
1056 nvm_ppa_tgt_to_dev(tgt_dev, &ppa, 1);
1058 if (dev->geo.version == NVM_OCSSD_SPEC_12)
1059 return nvm_get_bb_meta(dev, (sector_t)ppa.ppa, nchks, meta);
1061 return dev->ops->get_chk_meta(dev, (sector_t)ppa.ppa, nchks, meta);
1063 EXPORT_SYMBOL_GPL(nvm_get_chunk_meta);
1065 int nvm_set_chunk_meta(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *ppas,
1066 int nr_ppas, int type)
1068 struct nvm_dev *dev = tgt_dev->parent;
1069 struct nvm_rq rqd;
1070 int ret;
1072 if (dev->geo.version == NVM_OCSSD_SPEC_20)
1073 return 0;
1075 if (nr_ppas > NVM_MAX_VLBA) {
1076 pr_err("unable to update all blocks atomically\n");
1077 return -EINVAL;
1080 memset(&rqd, 0, sizeof(struct nvm_rq));
1082 nvm_set_rqd_ppalist(tgt_dev, &rqd, ppas, nr_ppas);
1083 nvm_rq_tgt_to_dev(tgt_dev, &rqd);
1085 ret = dev->ops->set_bb_tbl(dev, &rqd.ppa_addr, rqd.nr_ppas, type);
1086 nvm_free_rqd_ppalist(tgt_dev, &rqd);
1087 if (ret)
1088 return -EINVAL;
1090 return 0;
1092 EXPORT_SYMBOL_GPL(nvm_set_chunk_meta);
1094 static int nvm_core_init(struct nvm_dev *dev)
1096 struct nvm_geo *geo = &dev->geo;
1097 int ret;
1099 dev->lun_map = kcalloc(BITS_TO_LONGS(geo->all_luns),
1100 sizeof(unsigned long), GFP_KERNEL);
1101 if (!dev->lun_map)
1102 return -ENOMEM;
1104 INIT_LIST_HEAD(&dev->area_list);
1105 INIT_LIST_HEAD(&dev->targets);
1106 mutex_init(&dev->mlock);
1107 spin_lock_init(&dev->lock);
1109 ret = nvm_register_map(dev);
1110 if (ret)
1111 goto err_fmtype;
1113 return 0;
1114 err_fmtype:
1115 kfree(dev->lun_map);
1116 return ret;
1119 static void nvm_free(struct kref *ref)
1121 struct nvm_dev *dev = container_of(ref, struct nvm_dev, ref);
1123 if (dev->dma_pool)
1124 dev->ops->destroy_dma_pool(dev->dma_pool);
1126 if (dev->rmap)
1127 nvm_unregister_map(dev);
1129 kfree(dev->lun_map);
1130 kfree(dev);
1133 static int nvm_init(struct nvm_dev *dev)
1135 struct nvm_geo *geo = &dev->geo;
1136 int ret = -EINVAL;
1138 if (dev->ops->identity(dev)) {
1139 pr_err("device could not be identified\n");
1140 goto err;
1143 pr_debug("ver:%u.%u nvm_vendor:%x\n", geo->major_ver_id,
1144 geo->minor_ver_id, geo->vmnt);
1146 ret = nvm_core_init(dev);
1147 if (ret) {
1148 pr_err("could not initialize core structures.\n");
1149 goto err;
1152 pr_info("registered %s [%u/%u/%u/%u/%u]\n",
1153 dev->name, dev->geo.ws_min, dev->geo.ws_opt,
1154 dev->geo.num_chk, dev->geo.all_luns,
1155 dev->geo.num_ch);
1156 return 0;
1157 err:
1158 pr_err("failed to initialize nvm\n");
1159 return ret;
1162 struct nvm_dev *nvm_alloc_dev(int node)
1164 struct nvm_dev *dev;
1166 dev = kzalloc_node(sizeof(struct nvm_dev), GFP_KERNEL, node);
1167 if (dev)
1168 kref_init(&dev->ref);
1170 return dev;
1172 EXPORT_SYMBOL(nvm_alloc_dev);
1174 int nvm_register(struct nvm_dev *dev)
1176 int ret, exp_pool_size;
1178 if (!dev->q || !dev->ops) {
1179 kref_put(&dev->ref, nvm_free);
1180 return -EINVAL;
1183 ret = nvm_init(dev);
1184 if (ret) {
1185 kref_put(&dev->ref, nvm_free);
1186 return ret;
1189 exp_pool_size = max_t(int, PAGE_SIZE,
1190 (NVM_MAX_VLBA * (sizeof(u64) + dev->geo.sos)));
1191 exp_pool_size = round_up(exp_pool_size, PAGE_SIZE);
1193 dev->dma_pool = dev->ops->create_dma_pool(dev, "ppalist",
1194 exp_pool_size);
1195 if (!dev->dma_pool) {
1196 pr_err("could not create dma pool\n");
1197 kref_put(&dev->ref, nvm_free);
1198 return -ENOMEM;
1201 /* register device with a supported media manager */
1202 down_write(&nvm_lock);
1203 list_add(&dev->devices, &nvm_devices);
1204 up_write(&nvm_lock);
1206 return 0;
1208 EXPORT_SYMBOL(nvm_register);
1210 void nvm_unregister(struct nvm_dev *dev)
1212 struct nvm_target *t, *tmp;
1214 mutex_lock(&dev->mlock);
1215 list_for_each_entry_safe(t, tmp, &dev->targets, list) {
1216 if (t->dev->parent != dev)
1217 continue;
1218 __nvm_remove_target(t, false);
1219 kref_put(&dev->ref, nvm_free);
1221 mutex_unlock(&dev->mlock);
1223 down_write(&nvm_lock);
1224 list_del(&dev->devices);
1225 up_write(&nvm_lock);
1227 kref_put(&dev->ref, nvm_free);
1229 EXPORT_SYMBOL(nvm_unregister);
1231 static int __nvm_configure_create(struct nvm_ioctl_create *create)
1233 struct nvm_dev *dev;
1234 int ret;
1236 down_write(&nvm_lock);
1237 dev = nvm_find_nvm_dev(create->dev);
1238 up_write(&nvm_lock);
1240 if (!dev) {
1241 pr_err("device not found\n");
1242 return -EINVAL;
1245 kref_get(&dev->ref);
1246 ret = nvm_create_tgt(dev, create);
1247 if (ret)
1248 kref_put(&dev->ref, nvm_free);
1250 return ret;
1253 static long nvm_ioctl_info(struct file *file, void __user *arg)
1255 struct nvm_ioctl_info *info;
1256 struct nvm_tgt_type *tt;
1257 int tgt_iter = 0;
1259 info = memdup_user(arg, sizeof(struct nvm_ioctl_info));
1260 if (IS_ERR(info))
1261 return -EFAULT;
1263 info->version[0] = NVM_VERSION_MAJOR;
1264 info->version[1] = NVM_VERSION_MINOR;
1265 info->version[2] = NVM_VERSION_PATCH;
1267 down_write(&nvm_tgtt_lock);
1268 list_for_each_entry(tt, &nvm_tgt_types, list) {
1269 struct nvm_ioctl_info_tgt *tgt = &info->tgts[tgt_iter];
1271 tgt->version[0] = tt->version[0];
1272 tgt->version[1] = tt->version[1];
1273 tgt->version[2] = tt->version[2];
1274 strncpy(tgt->tgtname, tt->name, NVM_TTYPE_NAME_MAX);
1276 tgt_iter++;
1279 info->tgtsize = tgt_iter;
1280 up_write(&nvm_tgtt_lock);
1282 if (copy_to_user(arg, info, sizeof(struct nvm_ioctl_info))) {
1283 kfree(info);
1284 return -EFAULT;
1287 kfree(info);
1288 return 0;
1291 static long nvm_ioctl_get_devices(struct file *file, void __user *arg)
1293 struct nvm_ioctl_get_devices *devices;
1294 struct nvm_dev *dev;
1295 int i = 0;
1297 devices = kzalloc(sizeof(struct nvm_ioctl_get_devices), GFP_KERNEL);
1298 if (!devices)
1299 return -ENOMEM;
1301 down_write(&nvm_lock);
1302 list_for_each_entry(dev, &nvm_devices, devices) {
1303 struct nvm_ioctl_device_info *info = &devices->info[i];
1305 strlcpy(info->devname, dev->name, sizeof(info->devname));
1307 /* kept for compatibility */
1308 info->bmversion[0] = 1;
1309 info->bmversion[1] = 0;
1310 info->bmversion[2] = 0;
1311 strlcpy(info->bmname, "gennvm", sizeof(info->bmname));
1312 i++;
1314 if (i >= ARRAY_SIZE(devices->info)) {
1315 pr_err("max %zd devices can be reported.\n",
1316 ARRAY_SIZE(devices->info));
1317 break;
1320 up_write(&nvm_lock);
1322 devices->nr_devices = i;
1324 if (copy_to_user(arg, devices,
1325 sizeof(struct nvm_ioctl_get_devices))) {
1326 kfree(devices);
1327 return -EFAULT;
1330 kfree(devices);
1331 return 0;
1334 static long nvm_ioctl_dev_create(struct file *file, void __user *arg)
1336 struct nvm_ioctl_create create;
1338 if (copy_from_user(&create, arg, sizeof(struct nvm_ioctl_create)))
1339 return -EFAULT;
1341 if (create.conf.type == NVM_CONFIG_TYPE_EXTENDED &&
1342 create.conf.e.rsv != 0) {
1343 pr_err("reserved config field in use\n");
1344 return -EINVAL;
1347 create.dev[DISK_NAME_LEN - 1] = '\0';
1348 create.tgttype[NVM_TTYPE_NAME_MAX - 1] = '\0';
1349 create.tgtname[DISK_NAME_LEN - 1] = '\0';
1351 if (create.flags != 0) {
1352 __u32 flags = create.flags;
1354 /* Check for valid flags */
1355 if (flags & NVM_TARGET_FACTORY)
1356 flags &= ~NVM_TARGET_FACTORY;
1358 if (flags) {
1359 pr_err("flag not supported\n");
1360 return -EINVAL;
1364 return __nvm_configure_create(&create);
1367 static long nvm_ioctl_dev_remove(struct file *file, void __user *arg)
1369 struct nvm_ioctl_remove remove;
1371 if (copy_from_user(&remove, arg, sizeof(struct nvm_ioctl_remove)))
1372 return -EFAULT;
1374 remove.tgtname[DISK_NAME_LEN - 1] = '\0';
1376 if (remove.flags != 0) {
1377 pr_err("no flags supported\n");
1378 return -EINVAL;
1381 return nvm_remove_tgt(&remove);
1384 /* kept for compatibility reasons */
1385 static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
1387 struct nvm_ioctl_dev_init init;
1389 if (copy_from_user(&init, arg, sizeof(struct nvm_ioctl_dev_init)))
1390 return -EFAULT;
1392 if (init.flags != 0) {
1393 pr_err("no flags supported\n");
1394 return -EINVAL;
1397 return 0;
1400 /* Kept for compatibility reasons */
1401 static long nvm_ioctl_dev_factory(struct file *file, void __user *arg)
1403 struct nvm_ioctl_dev_factory fact;
1405 if (copy_from_user(&fact, arg, sizeof(struct nvm_ioctl_dev_factory)))
1406 return -EFAULT;
1408 fact.dev[DISK_NAME_LEN - 1] = '\0';
1410 if (fact.flags & ~(NVM_FACTORY_NR_BITS - 1))
1411 return -EINVAL;
1413 return 0;
1416 static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
1418 void __user *argp = (void __user *)arg;
1420 if (!capable(CAP_SYS_ADMIN))
1421 return -EPERM;
1423 switch (cmd) {
1424 case NVM_INFO:
1425 return nvm_ioctl_info(file, argp);
1426 case NVM_GET_DEVICES:
1427 return nvm_ioctl_get_devices(file, argp);
1428 case NVM_DEV_CREATE:
1429 return nvm_ioctl_dev_create(file, argp);
1430 case NVM_DEV_REMOVE:
1431 return nvm_ioctl_dev_remove(file, argp);
1432 case NVM_DEV_INIT:
1433 return nvm_ioctl_dev_init(file, argp);
1434 case NVM_DEV_FACTORY:
1435 return nvm_ioctl_dev_factory(file, argp);
1437 return 0;
1440 static const struct file_operations _ctl_fops = {
1441 .open = nonseekable_open,
1442 .unlocked_ioctl = nvm_ctl_ioctl,
1443 .owner = THIS_MODULE,
1444 .llseek = noop_llseek,
1447 static struct miscdevice _nvm_misc = {
1448 .minor = MISC_DYNAMIC_MINOR,
1449 .name = "lightnvm",
1450 .nodename = "lightnvm/control",
1451 .fops = &_ctl_fops,
1453 builtin_misc_device(_nvm_misc);