1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2015 IT University of Copenhagen. All rights reserved.
4 * Initial release: Matias Bjorling <m@bjorling.me>
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 */
32 struct nvm_ch_map *chnls;
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))
49 static bool nvm_target_exists(const char *name)
52 struct nvm_target *tgt;
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)) {
61 mutex_unlock(&dev->mlock);
65 mutex_unlock(&dev->mlock);
73 static int nvm_reserve_luns(struct nvm_dev *dev, int lun_begin, int lun_end)
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);
86 while (--i >= lun_begin)
87 clear_bit(i, dev->lun_map);
92 static void nvm_release_luns_err(struct nvm_dev *dev, int lun_begin,
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;
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;
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,
122 kfree(ch_map->lun_offs);
125 kfree(dev_map->chnls);
128 kfree(tgt_dev->luns);
132 static struct nvm_tgt_dev *nvm_create_tgt_dev(struct nvm_dev *dev,
133 u16 lun_begin, u16 lun_end,
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;
147 int lun_balanced = 1;
148 int sec_per_lun, prev_num_lun;
151 num_ch = (num_ch_mod == 0) ? num_ch : num_ch + 1;
153 dev_map = kmalloc(sizeof(struct nvm_dev_map), GFP_KERNEL);
157 dev_map->chnls = kcalloc(num_ch, sizeof(struct nvm_ch_map), GFP_KERNEL);
161 luns = kcalloc(num_lun, sizeof(struct ppa_addr), GFP_KERNEL);
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];
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)
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);
185 for (j = 0; j < luns_in_chnl; j++) {
187 luns[lunid].a.ch = i;
188 luns[lunid++].a.lun = j;
191 lun_roffs[j + blun] = blun;
194 ch_map->lun_offs = lun_offs;
196 /* when starting a new channel, lun offset is reset */
198 luns_left -= luns_in_chnl;
201 dev_map->num_ch = num_ch;
203 tgt_dev = kmalloc(sizeof(struct nvm_tgt_dev), GFP_KERNEL);
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;
222 tgt_dev->map = dev_map;
223 tgt_dev->luns = luns;
224 tgt_dev->parent = dev;
229 kfree(dev_map->chnls[i].lun_offs);
232 kfree(dev_map->chnls);
239 static const struct block_device_operations nvm_fops = {
240 .owner = THIS_MODULE,
243 static struct nvm_tgt_type *__nvm_find_target_type(const char *name)
245 struct nvm_tgt_type *tt;
247 list_for_each_entry(tt, &nvm_tgt_types, list)
248 if (!strcmp(name, tt->name))
254 static struct nvm_tgt_type *nvm_find_target_type(const char *name)
256 struct nvm_tgt_type *tt;
258 down_write(&nvm_tgtt_lock);
259 tt = __nvm_find_target_type(name);
260 up_write(&nvm_tgtt_lock);
265 static int nvm_config_check_luns(struct nvm_geo *geo, int lun_begin,
268 if (lun_begin > lun_end || lun_end >= geo->all_luns) {
269 pr_err("lun out of bound (%u:%u > %u)\n",
270 lun_begin, lun_end, geo->all_luns - 1);
277 static int __nvm_config_simple(struct nvm_dev *dev,
278 struct nvm_ioctl_create_simple *s)
280 struct nvm_geo *geo = &dev->geo;
282 if (s->lun_begin == -1 && s->lun_end == -1) {
284 s->lun_end = geo->all_luns - 1;
287 return nvm_config_check_luns(geo, s->lun_begin, s->lun_end);
290 static int __nvm_config_extended(struct nvm_dev *dev,
291 struct nvm_ioctl_create_extended *e)
293 if (e->lun_begin == 0xFFFF && e->lun_end == 0xFFFF) {
295 e->lun_end = dev->geo.all_luns - 1;
298 /* op not set falls into target's default */
299 if (e->op == 0xFFFF) {
300 e->op = NVM_TARGET_DEFAULT_OP;
301 } else if (e->op < NVM_TARGET_MIN_OP || e->op > NVM_TARGET_MAX_OP) {
302 pr_err("invalid over provisioning value\n");
306 return nvm_config_check_luns(&dev->geo, e->lun_begin, e->lun_end);
309 static int nvm_create_tgt(struct nvm_dev *dev, struct nvm_ioctl_create *create)
311 struct nvm_ioctl_create_extended e;
312 struct request_queue *tqueue;
313 struct gendisk *tdisk;
314 struct nvm_tgt_type *tt;
315 struct nvm_target *t;
316 struct nvm_tgt_dev *tgt_dev;
321 switch (create->conf.type) {
322 case NVM_CONFIG_TYPE_SIMPLE:
323 ret = __nvm_config_simple(dev, &create->conf.s);
327 e.lun_begin = create->conf.s.lun_begin;
328 e.lun_end = create->conf.s.lun_end;
329 e.op = NVM_TARGET_DEFAULT_OP;
331 case NVM_CONFIG_TYPE_EXTENDED:
332 ret = __nvm_config_extended(dev, &create->conf.e);
339 pr_err("config type not valid\n");
343 tt = nvm_find_target_type(create->tgttype);
345 pr_err("target type %s not found\n", create->tgttype);
349 if ((tt->flags & NVM_TGT_F_HOST_L2P) != (dev->geo.dom & NVM_RSP_L2P)) {
350 pr_err("device is incompatible with target L2P type.\n");
354 if (nvm_target_exists(create->tgtname)) {
355 pr_err("target name already exists (%s)\n",
360 ret = nvm_reserve_luns(dev, e.lun_begin, e.lun_end);
364 t = kmalloc(sizeof(struct nvm_target), GFP_KERNEL);
370 tgt_dev = nvm_create_tgt_dev(dev, e.lun_begin, e.lun_end, e.op);
372 pr_err("could not create target device\n");
377 tdisk = alloc_disk(0);
383 tqueue = blk_alloc_queue(tt->make_rq, dev->q->node);
389 strlcpy(tdisk->disk_name, create->tgtname, sizeof(tdisk->disk_name));
390 tdisk->flags = GENHD_FL_EXT_DEVT;
392 tdisk->first_minor = 0;
393 tdisk->fops = &nvm_fops;
394 tdisk->queue = tqueue;
396 targetdata = tt->init(tgt_dev, tdisk, create->flags);
397 if (IS_ERR(targetdata)) {
398 ret = PTR_ERR(targetdata);
402 tdisk->private_data = targetdata;
403 tqueue->queuedata = targetdata;
405 mdts = (dev->geo.csecs >> 9) * NVM_MAX_VLBA;
407 mdts = min_t(u32, dev->geo.mdts,
408 (dev->geo.csecs >> 9) * NVM_MAX_VLBA);
410 blk_queue_max_hw_sectors(tqueue, mdts);
412 set_capacity(tdisk, tt->capacity(targetdata));
415 if (tt->sysfs_init && tt->sysfs_init(tdisk)) {
424 mutex_lock(&dev->mlock);
425 list_add_tail(&t->list, &dev->targets);
426 mutex_unlock(&dev->mlock);
428 __module_get(tt->owner);
433 tt->exit(targetdata, true);
435 blk_cleanup_queue(tqueue);
440 nvm_remove_tgt_dev(tgt_dev, 0);
444 nvm_release_luns_err(dev, e.lun_begin, e.lun_end);
448 static void __nvm_remove_target(struct nvm_target *t, bool graceful)
450 struct nvm_tgt_type *tt = t->type;
451 struct gendisk *tdisk = t->disk;
452 struct request_queue *q = tdisk->queue;
455 blk_cleanup_queue(q);
458 tt->sysfs_exit(tdisk);
461 tt->exit(tdisk->private_data, graceful);
463 nvm_remove_tgt_dev(t->dev, 1);
465 module_put(t->type->owner);
472 * nvm_remove_tgt - Removes a target from the media manager
473 * @remove: ioctl structure with target name to remove.
480 static int nvm_remove_tgt(struct nvm_ioctl_remove *remove)
482 struct nvm_target *t = NULL;
485 down_read(&nvm_lock);
486 list_for_each_entry(dev, &nvm_devices, devices) {
487 mutex_lock(&dev->mlock);
488 t = nvm_find_target(dev, remove->tgtname);
490 mutex_unlock(&dev->mlock);
493 mutex_unlock(&dev->mlock);
498 pr_err("failed to remove target %s\n",
503 __nvm_remove_target(t, true);
504 kref_put(&dev->ref, nvm_free);
509 static int nvm_register_map(struct nvm_dev *dev)
511 struct nvm_dev_map *rmap;
514 rmap = kmalloc(sizeof(struct nvm_dev_map), GFP_KERNEL);
518 rmap->chnls = kcalloc(dev->geo.num_ch, sizeof(struct nvm_ch_map),
523 for (i = 0; i < dev->geo.num_ch; i++) {
524 struct nvm_ch_map *ch_rmap;
526 int luns_in_chnl = dev->geo.num_lun;
528 ch_rmap = &rmap->chnls[i];
530 ch_rmap->ch_off = -1;
531 ch_rmap->num_lun = luns_in_chnl;
533 lun_roffs = kcalloc(luns_in_chnl, sizeof(int), GFP_KERNEL);
537 for (j = 0; j < luns_in_chnl; j++)
540 ch_rmap->lun_offs = lun_roffs;
548 kfree(rmap->chnls[i].lun_offs);
555 static void nvm_unregister_map(struct nvm_dev *dev)
557 struct nvm_dev_map *rmap = dev->rmap;
560 for (i = 0; i < dev->geo.num_ch; i++)
561 kfree(rmap->chnls[i].lun_offs);
567 static void nvm_map_to_dev(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *p)
569 struct nvm_dev_map *dev_map = tgt_dev->map;
570 struct nvm_ch_map *ch_map = &dev_map->chnls[p->a.ch];
571 int lun_off = ch_map->lun_offs[p->a.lun];
573 p->a.ch += ch_map->ch_off;
577 static void nvm_map_to_tgt(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *p)
579 struct nvm_dev *dev = tgt_dev->parent;
580 struct nvm_dev_map *dev_rmap = dev->rmap;
581 struct nvm_ch_map *ch_rmap = &dev_rmap->chnls[p->a.ch];
582 int lun_roff = ch_rmap->lun_offs[p->a.lun];
584 p->a.ch -= ch_rmap->ch_off;
585 p->a.lun -= lun_roff;
588 static void nvm_ppa_tgt_to_dev(struct nvm_tgt_dev *tgt_dev,
589 struct ppa_addr *ppa_list, int nr_ppas)
593 for (i = 0; i < nr_ppas; i++) {
594 nvm_map_to_dev(tgt_dev, &ppa_list[i]);
595 ppa_list[i] = generic_to_dev_addr(tgt_dev->parent, ppa_list[i]);
599 static void nvm_ppa_dev_to_tgt(struct nvm_tgt_dev *tgt_dev,
600 struct ppa_addr *ppa_list, int nr_ppas)
604 for (i = 0; i < nr_ppas; i++) {
605 ppa_list[i] = dev_to_generic_addr(tgt_dev->parent, ppa_list[i]);
606 nvm_map_to_tgt(tgt_dev, &ppa_list[i]);
610 static void nvm_rq_tgt_to_dev(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
612 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
614 nvm_ppa_tgt_to_dev(tgt_dev, ppa_list, rqd->nr_ppas);
617 static void nvm_rq_dev_to_tgt(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd)
619 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
621 nvm_ppa_dev_to_tgt(tgt_dev, ppa_list, rqd->nr_ppas);
624 int nvm_register_tgt_type(struct nvm_tgt_type *tt)
628 down_write(&nvm_tgtt_lock);
629 if (__nvm_find_target_type(tt->name))
632 list_add(&tt->list, &nvm_tgt_types);
633 up_write(&nvm_tgtt_lock);
637 EXPORT_SYMBOL(nvm_register_tgt_type);
639 void nvm_unregister_tgt_type(struct nvm_tgt_type *tt)
644 down_write(&nvm_tgtt_lock);
646 up_write(&nvm_tgtt_lock);
648 EXPORT_SYMBOL(nvm_unregister_tgt_type);
650 void *nvm_dev_dma_alloc(struct nvm_dev *dev, gfp_t mem_flags,
651 dma_addr_t *dma_handler)
653 return dev->ops->dev_dma_alloc(dev, dev->dma_pool, mem_flags,
656 EXPORT_SYMBOL(nvm_dev_dma_alloc);
658 void nvm_dev_dma_free(struct nvm_dev *dev, void *addr, dma_addr_t dma_handler)
660 dev->ops->dev_dma_free(dev->dma_pool, addr, dma_handler);
662 EXPORT_SYMBOL(nvm_dev_dma_free);
664 static struct nvm_dev *nvm_find_nvm_dev(const char *name)
668 list_for_each_entry(dev, &nvm_devices, devices)
669 if (!strcmp(name, dev->name))
675 static int nvm_set_rqd_ppalist(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd,
676 const struct ppa_addr *ppas, int nr_ppas)
678 struct nvm_dev *dev = tgt_dev->parent;
679 struct nvm_geo *geo = &tgt_dev->geo;
680 int i, plane_cnt, pl_idx;
683 if (geo->pln_mode == NVM_PLANE_SINGLE && nr_ppas == 1) {
684 rqd->nr_ppas = nr_ppas;
685 rqd->ppa_addr = ppas[0];
690 rqd->nr_ppas = nr_ppas;
691 rqd->ppa_list = nvm_dev_dma_alloc(dev, GFP_KERNEL, &rqd->dma_ppa_list);
692 if (!rqd->ppa_list) {
693 pr_err("failed to allocate dma memory\n");
697 plane_cnt = geo->pln_mode;
698 rqd->nr_ppas *= plane_cnt;
700 for (i = 0; i < nr_ppas; i++) {
701 for (pl_idx = 0; pl_idx < plane_cnt; pl_idx++) {
704 rqd->ppa_list[(pl_idx * nr_ppas) + i] = ppa;
711 static void nvm_free_rqd_ppalist(struct nvm_tgt_dev *tgt_dev,
717 nvm_dev_dma_free(tgt_dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
720 static int nvm_set_flags(struct nvm_geo *geo, struct nvm_rq *rqd)
724 if (geo->version == NVM_OCSSD_SPEC_20)
728 flags |= geo->pln_mode >> 1;
730 if (rqd->opcode == NVM_OP_PREAD)
731 flags |= (NVM_IO_SCRAMBLE_ENABLE | NVM_IO_SUSPEND);
732 else if (rqd->opcode == NVM_OP_PWRITE)
733 flags |= NVM_IO_SCRAMBLE_ENABLE;
738 int nvm_submit_io(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd, void *buf)
740 struct nvm_dev *dev = tgt_dev->parent;
743 if (!dev->ops->submit_io)
746 nvm_rq_tgt_to_dev(tgt_dev, rqd);
749 rqd->flags = nvm_set_flags(&tgt_dev->geo, rqd);
751 /* In case of error, fail with right address format */
752 ret = dev->ops->submit_io(dev, rqd, buf);
754 nvm_rq_dev_to_tgt(tgt_dev, rqd);
757 EXPORT_SYMBOL(nvm_submit_io);
759 static void nvm_sync_end_io(struct nvm_rq *rqd)
761 struct completion *waiting = rqd->private;
766 static int nvm_submit_io_wait(struct nvm_dev *dev, struct nvm_rq *rqd,
769 DECLARE_COMPLETION_ONSTACK(wait);
772 rqd->end_io = nvm_sync_end_io;
773 rqd->private = &wait;
775 ret = dev->ops->submit_io(dev, rqd, buf);
779 wait_for_completion_io(&wait);
784 int nvm_submit_io_sync(struct nvm_tgt_dev *tgt_dev, struct nvm_rq *rqd,
787 struct nvm_dev *dev = tgt_dev->parent;
790 if (!dev->ops->submit_io)
793 nvm_rq_tgt_to_dev(tgt_dev, rqd);
796 rqd->flags = nvm_set_flags(&tgt_dev->geo, rqd);
798 ret = nvm_submit_io_wait(dev, rqd, buf);
802 EXPORT_SYMBOL(nvm_submit_io_sync);
804 void nvm_end_io(struct nvm_rq *rqd)
806 struct nvm_tgt_dev *tgt_dev = rqd->dev;
808 /* Convert address space */
810 nvm_rq_dev_to_tgt(tgt_dev, rqd);
815 EXPORT_SYMBOL(nvm_end_io);
817 static int nvm_submit_io_sync_raw(struct nvm_dev *dev, struct nvm_rq *rqd)
819 if (!dev->ops->submit_io)
823 rqd->flags = nvm_set_flags(&dev->geo, rqd);
825 return nvm_submit_io_wait(dev, rqd, NULL);
828 static int nvm_bb_chunk_sense(struct nvm_dev *dev, struct ppa_addr ppa)
830 struct nvm_rq rqd = { NULL };
832 struct bio_vec bio_vec;
836 page = alloc_page(GFP_KERNEL);
840 bio_init(&bio, &bio_vec, 1);
841 bio_add_page(&bio, page, PAGE_SIZE, 0);
842 bio_set_op_attrs(&bio, REQ_OP_READ, 0);
845 rqd.opcode = NVM_OP_PREAD;
848 rqd.ppa_addr = generic_to_dev_addr(dev, ppa);
850 ret = nvm_submit_io_sync_raw(dev, &rqd);
860 * Scans a 1.2 chunk first and last page to determine if its state.
861 * If the chunk is found to be open, also scan it to update the write
864 static int nvm_bb_chunk_scan(struct nvm_dev *dev, struct ppa_addr ppa,
865 struct nvm_chk_meta *meta)
867 struct nvm_geo *geo = &dev->geo;
870 /* sense first page */
871 ret = nvm_bb_chunk_sense(dev, ppa);
872 if (ret < 0) /* io error */
874 else if (ret == 0) /* valid data */
875 meta->state = NVM_CHK_ST_OPEN;
878 * If empty page, the chunk is free, else it is an
879 * actual io error. In that case, mark it offline.
882 case NVM_RSP_ERR_EMPTYPAGE:
883 meta->state = NVM_CHK_ST_FREE;
885 case NVM_RSP_ERR_FAILCRC:
886 case NVM_RSP_ERR_FAILECC:
887 case NVM_RSP_WARN_HIGHECC:
888 meta->state = NVM_CHK_ST_OPEN;
891 return -ret; /* other io error */
895 /* sense last page */
896 ppa.g.pg = geo->num_pg - 1;
897 ppa.g.pl = geo->num_pln - 1;
899 ret = nvm_bb_chunk_sense(dev, ppa);
900 if (ret < 0) /* io error */
902 else if (ret == 0) { /* Chunk fully written */
903 meta->state = NVM_CHK_ST_CLOSED;
904 meta->wp = geo->clba;
906 } else if (ret > 0) {
908 case NVM_RSP_ERR_EMPTYPAGE:
909 case NVM_RSP_ERR_FAILCRC:
910 case NVM_RSP_ERR_FAILECC:
911 case NVM_RSP_WARN_HIGHECC:
912 meta->state = NVM_CHK_ST_OPEN;
915 return -ret; /* other io error */
921 * chunk is open, we scan sequentially to update the write pointer.
922 * We make the assumption that targets write data across all planes
923 * before moving to the next page.
925 for (pg = 0; pg < geo->num_pg; pg++) {
926 for (pl = 0; pl < geo->num_pln; pl++) {
930 ret = nvm_bb_chunk_sense(dev, ppa);
931 if (ret < 0) /* io error */
934 meta->wp += geo->ws_min;
935 } else if (ret > 0) {
937 case NVM_RSP_ERR_EMPTYPAGE:
939 case NVM_RSP_ERR_FAILCRC:
940 case NVM_RSP_ERR_FAILECC:
941 case NVM_RSP_WARN_HIGHECC:
942 meta->wp += geo->ws_min;
945 return -ret; /* other io error */
955 * folds a bad block list from its plane representation to its
956 * chunk representation.
958 * If any of the planes status are bad or grown bad, the chunk is marked
959 * offline. If not bad, the first plane state acts as the chunk state.
961 static int nvm_bb_to_chunk(struct nvm_dev *dev, struct ppa_addr ppa,
962 u8 *blks, int nr_blks, struct nvm_chk_meta *meta)
964 struct nvm_geo *geo = &dev->geo;
965 int ret, blk, pl, offset, blktype;
967 for (blk = 0; blk < geo->num_chk; blk++) {
968 offset = blk * geo->pln_mode;
969 blktype = blks[offset];
971 for (pl = 0; pl < geo->pln_mode; pl++) {
972 if (blks[offset + pl] &
973 (NVM_BLK_T_BAD|NVM_BLK_T_GRWN_BAD)) {
974 blktype = blks[offset + pl];
982 meta->type = NVM_CHK_TP_W_SEQ;
984 meta->slba = generic_to_dev_addr(dev, ppa).ppa;
985 meta->cnlb = dev->geo.clba;
987 if (blktype == NVM_BLK_T_FREE) {
988 ret = nvm_bb_chunk_scan(dev, ppa, meta);
992 meta->state = NVM_CHK_ST_OFFLINE;
1001 static int nvm_get_bb_meta(struct nvm_dev *dev, sector_t slba,
1002 int nchks, struct nvm_chk_meta *meta)
1004 struct nvm_geo *geo = &dev->geo;
1005 struct ppa_addr ppa;
1007 int ch, lun, nr_blks;
1011 ppa = dev_to_generic_addr(dev, ppa);
1016 if ((nchks % geo->num_chk) != 0)
1019 nr_blks = geo->num_chk * geo->pln_mode;
1021 blks = kmalloc(nr_blks, GFP_KERNEL);
1025 for (ch = ppa.g.ch; ch < geo->num_ch; ch++) {
1026 for (lun = ppa.g.lun; lun < geo->num_lun; lun++) {
1027 struct ppa_addr ppa_gen, ppa_dev;
1034 ppa_gen.g.lun = lun;
1035 ppa_dev = generic_to_dev_addr(dev, ppa_gen);
1037 ret = dev->ops->get_bb_tbl(dev, ppa_dev, blks);
1041 ret = nvm_bb_to_chunk(dev, ppa_gen, blks, nr_blks,
1046 meta += geo->num_chk;
1047 nchks -= geo->num_chk;
1055 int nvm_get_chunk_meta(struct nvm_tgt_dev *tgt_dev, struct ppa_addr ppa,
1056 int nchks, struct nvm_chk_meta *meta)
1058 struct nvm_dev *dev = tgt_dev->parent;
1060 nvm_ppa_tgt_to_dev(tgt_dev, &ppa, 1);
1062 if (dev->geo.version == NVM_OCSSD_SPEC_12)
1063 return nvm_get_bb_meta(dev, (sector_t)ppa.ppa, nchks, meta);
1065 return dev->ops->get_chk_meta(dev, (sector_t)ppa.ppa, nchks, meta);
1067 EXPORT_SYMBOL_GPL(nvm_get_chunk_meta);
1069 int nvm_set_chunk_meta(struct nvm_tgt_dev *tgt_dev, struct ppa_addr *ppas,
1070 int nr_ppas, int type)
1072 struct nvm_dev *dev = tgt_dev->parent;
1076 if (dev->geo.version == NVM_OCSSD_SPEC_20)
1079 if (nr_ppas > NVM_MAX_VLBA) {
1080 pr_err("unable to update all blocks atomically\n");
1084 memset(&rqd, 0, sizeof(struct nvm_rq));
1086 nvm_set_rqd_ppalist(tgt_dev, &rqd, ppas, nr_ppas);
1087 nvm_rq_tgt_to_dev(tgt_dev, &rqd);
1089 ret = dev->ops->set_bb_tbl(dev, &rqd.ppa_addr, rqd.nr_ppas, type);
1090 nvm_free_rqd_ppalist(tgt_dev, &rqd);
1096 EXPORT_SYMBOL_GPL(nvm_set_chunk_meta);
1098 static int nvm_core_init(struct nvm_dev *dev)
1100 struct nvm_geo *geo = &dev->geo;
1103 dev->lun_map = kcalloc(BITS_TO_LONGS(geo->all_luns),
1104 sizeof(unsigned long), GFP_KERNEL);
1108 INIT_LIST_HEAD(&dev->area_list);
1109 INIT_LIST_HEAD(&dev->targets);
1110 mutex_init(&dev->mlock);
1111 spin_lock_init(&dev->lock);
1113 ret = nvm_register_map(dev);
1119 kfree(dev->lun_map);
1123 static void nvm_free(struct kref *ref)
1125 struct nvm_dev *dev = container_of(ref, struct nvm_dev, ref);
1128 dev->ops->destroy_dma_pool(dev->dma_pool);
1131 nvm_unregister_map(dev);
1133 kfree(dev->lun_map);
1137 static int nvm_init(struct nvm_dev *dev)
1139 struct nvm_geo *geo = &dev->geo;
1142 if (dev->ops->identity(dev)) {
1143 pr_err("device could not be identified\n");
1147 pr_debug("ver:%u.%u nvm_vendor:%x\n", geo->major_ver_id,
1148 geo->minor_ver_id, geo->vmnt);
1150 ret = nvm_core_init(dev);
1152 pr_err("could not initialize core structures.\n");
1156 pr_info("registered %s [%u/%u/%u/%u/%u]\n",
1157 dev->name, dev->geo.ws_min, dev->geo.ws_opt,
1158 dev->geo.num_chk, dev->geo.all_luns,
1162 pr_err("failed to initialize nvm\n");
1166 struct nvm_dev *nvm_alloc_dev(int node)
1168 struct nvm_dev *dev;
1170 dev = kzalloc_node(sizeof(struct nvm_dev), GFP_KERNEL, node);
1172 kref_init(&dev->ref);
1176 EXPORT_SYMBOL(nvm_alloc_dev);
1178 int nvm_register(struct nvm_dev *dev)
1180 int ret, exp_pool_size;
1182 if (!dev->q || !dev->ops) {
1183 kref_put(&dev->ref, nvm_free);
1187 ret = nvm_init(dev);
1189 kref_put(&dev->ref, nvm_free);
1193 exp_pool_size = max_t(int, PAGE_SIZE,
1194 (NVM_MAX_VLBA * (sizeof(u64) + dev->geo.sos)));
1195 exp_pool_size = round_up(exp_pool_size, PAGE_SIZE);
1197 dev->dma_pool = dev->ops->create_dma_pool(dev, "ppalist",
1199 if (!dev->dma_pool) {
1200 pr_err("could not create dma pool\n");
1201 kref_put(&dev->ref, nvm_free);
1205 /* register device with a supported media manager */
1206 down_write(&nvm_lock);
1207 list_add(&dev->devices, &nvm_devices);
1208 up_write(&nvm_lock);
1212 EXPORT_SYMBOL(nvm_register);
1214 void nvm_unregister(struct nvm_dev *dev)
1216 struct nvm_target *t, *tmp;
1218 mutex_lock(&dev->mlock);
1219 list_for_each_entry_safe(t, tmp, &dev->targets, list) {
1220 if (t->dev->parent != dev)
1222 __nvm_remove_target(t, false);
1223 kref_put(&dev->ref, nvm_free);
1225 mutex_unlock(&dev->mlock);
1227 down_write(&nvm_lock);
1228 list_del(&dev->devices);
1229 up_write(&nvm_lock);
1231 kref_put(&dev->ref, nvm_free);
1233 EXPORT_SYMBOL(nvm_unregister);
1235 static int __nvm_configure_create(struct nvm_ioctl_create *create)
1237 struct nvm_dev *dev;
1240 down_write(&nvm_lock);
1241 dev = nvm_find_nvm_dev(create->dev);
1242 up_write(&nvm_lock);
1245 pr_err("device not found\n");
1249 kref_get(&dev->ref);
1250 ret = nvm_create_tgt(dev, create);
1252 kref_put(&dev->ref, nvm_free);
1257 static long nvm_ioctl_info(struct file *file, void __user *arg)
1259 struct nvm_ioctl_info *info;
1260 struct nvm_tgt_type *tt;
1263 info = memdup_user(arg, sizeof(struct nvm_ioctl_info));
1267 info->version[0] = NVM_VERSION_MAJOR;
1268 info->version[1] = NVM_VERSION_MINOR;
1269 info->version[2] = NVM_VERSION_PATCH;
1271 down_write(&nvm_tgtt_lock);
1272 list_for_each_entry(tt, &nvm_tgt_types, list) {
1273 struct nvm_ioctl_info_tgt *tgt = &info->tgts[tgt_iter];
1275 tgt->version[0] = tt->version[0];
1276 tgt->version[1] = tt->version[1];
1277 tgt->version[2] = tt->version[2];
1278 strncpy(tgt->tgtname, tt->name, NVM_TTYPE_NAME_MAX);
1283 info->tgtsize = tgt_iter;
1284 up_write(&nvm_tgtt_lock);
1286 if (copy_to_user(arg, info, sizeof(struct nvm_ioctl_info))) {
1295 static long nvm_ioctl_get_devices(struct file *file, void __user *arg)
1297 struct nvm_ioctl_get_devices *devices;
1298 struct nvm_dev *dev;
1301 devices = kzalloc(sizeof(struct nvm_ioctl_get_devices), GFP_KERNEL);
1305 down_write(&nvm_lock);
1306 list_for_each_entry(dev, &nvm_devices, devices) {
1307 struct nvm_ioctl_device_info *info = &devices->info[i];
1309 strlcpy(info->devname, dev->name, sizeof(info->devname));
1311 /* kept for compatibility */
1312 info->bmversion[0] = 1;
1313 info->bmversion[1] = 0;
1314 info->bmversion[2] = 0;
1315 strlcpy(info->bmname, "gennvm", sizeof(info->bmname));
1319 pr_err("max 31 devices can be reported.\n");
1323 up_write(&nvm_lock);
1325 devices->nr_devices = i;
1327 if (copy_to_user(arg, devices,
1328 sizeof(struct nvm_ioctl_get_devices))) {
1337 static long nvm_ioctl_dev_create(struct file *file, void __user *arg)
1339 struct nvm_ioctl_create create;
1341 if (copy_from_user(&create, arg, sizeof(struct nvm_ioctl_create)))
1344 if (create.conf.type == NVM_CONFIG_TYPE_EXTENDED &&
1345 create.conf.e.rsv != 0) {
1346 pr_err("reserved config field in use\n");
1350 create.dev[DISK_NAME_LEN - 1] = '\0';
1351 create.tgttype[NVM_TTYPE_NAME_MAX - 1] = '\0';
1352 create.tgtname[DISK_NAME_LEN - 1] = '\0';
1354 if (create.flags != 0) {
1355 __u32 flags = create.flags;
1357 /* Check for valid flags */
1358 if (flags & NVM_TARGET_FACTORY)
1359 flags &= ~NVM_TARGET_FACTORY;
1362 pr_err("flag not supported\n");
1367 return __nvm_configure_create(&create);
1370 static long nvm_ioctl_dev_remove(struct file *file, void __user *arg)
1372 struct nvm_ioctl_remove remove;
1374 if (copy_from_user(&remove, arg, sizeof(struct nvm_ioctl_remove)))
1377 remove.tgtname[DISK_NAME_LEN - 1] = '\0';
1379 if (remove.flags != 0) {
1380 pr_err("no flags supported\n");
1384 return nvm_remove_tgt(&remove);
1387 /* kept for compatibility reasons */
1388 static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
1390 struct nvm_ioctl_dev_init init;
1392 if (copy_from_user(&init, arg, sizeof(struct nvm_ioctl_dev_init)))
1395 if (init.flags != 0) {
1396 pr_err("no flags supported\n");
1403 /* Kept for compatibility reasons */
1404 static long nvm_ioctl_dev_factory(struct file *file, void __user *arg)
1406 struct nvm_ioctl_dev_factory fact;
1408 if (copy_from_user(&fact, arg, sizeof(struct nvm_ioctl_dev_factory)))
1411 fact.dev[DISK_NAME_LEN - 1] = '\0';
1413 if (fact.flags & ~(NVM_FACTORY_NR_BITS - 1))
1419 static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
1421 void __user *argp = (void __user *)arg;
1423 if (!capable(CAP_SYS_ADMIN))
1428 return nvm_ioctl_info(file, argp);
1429 case NVM_GET_DEVICES:
1430 return nvm_ioctl_get_devices(file, argp);
1431 case NVM_DEV_CREATE:
1432 return nvm_ioctl_dev_create(file, argp);
1433 case NVM_DEV_REMOVE:
1434 return nvm_ioctl_dev_remove(file, argp);
1436 return nvm_ioctl_dev_init(file, argp);
1437 case NVM_DEV_FACTORY:
1438 return nvm_ioctl_dev_factory(file, argp);
1443 static const struct file_operations _ctl_fops = {
1444 .open = nonseekable_open,
1445 .unlocked_ioctl = nvm_ctl_ioctl,
1446 .owner = THIS_MODULE,
1447 .llseek = noop_llseek,
1450 static struct miscdevice _nvm_misc = {
1451 .minor = MISC_DYNAMIC_MINOR,
1453 .nodename = "lightnvm/control",
1456 builtin_misc_device(_nvm_misc);