2 * Copyright (C) 2016 CNEX Labs
3 * Initial release: Javier Gonzalez <javier@cnexlabs.com>
4 * Matias Bjorling <matias@cnexlabs.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License version
8 * 2 as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * pblk-core.c - pblk's core functionality
21 static void pblk_line_mark_bb(struct work_struct *work)
23 struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
25 struct pblk *pblk = line_ws->pblk;
26 struct nvm_tgt_dev *dev = pblk->dev;
27 struct ppa_addr *ppa = line_ws->priv;
30 ret = nvm_set_tgt_bb_tbl(dev, ppa, 1, NVM_BLK_T_GRWN_BAD);
32 struct pblk_line *line;
35 line = &pblk->lines[pblk_ppa_to_line(*ppa)];
36 pos = pblk_ppa_to_pos(&dev->geo, *ppa);
38 pblk_err(pblk, "failed to mark bb, line:%d, pos:%d\n",
43 mempool_free(line_ws, &pblk->gen_ws_pool);
46 static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
47 struct ppa_addr ppa_addr)
49 struct nvm_tgt_dev *dev = pblk->dev;
50 struct nvm_geo *geo = &dev->geo;
52 int pos = pblk_ppa_to_pos(geo, ppa_addr);
54 pblk_debug(pblk, "erase failed: line:%d, pos:%d\n", line->id, pos);
55 atomic_long_inc(&pblk->erase_failed);
57 atomic_dec(&line->blk_in_line);
58 if (test_and_set_bit(pos, line->blk_bitmap))
59 pblk_err(pblk, "attempted to erase bb: line:%d, pos:%d\n",
62 /* Not necessary to mark bad blocks on 2.0 spec. */
63 if (geo->version == NVM_OCSSD_SPEC_20)
66 ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
71 pblk_gen_run_ws(pblk, NULL, ppa, pblk_line_mark_bb,
72 GFP_ATOMIC, pblk->bb_wq);
75 static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
77 struct nvm_tgt_dev *dev = pblk->dev;
78 struct nvm_geo *geo = &dev->geo;
79 struct nvm_chk_meta *chunk;
80 struct pblk_line *line;
83 line = &pblk->lines[pblk_ppa_to_line(rqd->ppa_addr)];
84 pos = pblk_ppa_to_pos(geo, rqd->ppa_addr);
85 chunk = &line->chks[pos];
87 atomic_dec(&line->left_seblks);
90 chunk->state = NVM_CHK_ST_OFFLINE;
91 pblk_mark_bb(pblk, line, rqd->ppa_addr);
93 chunk->state = NVM_CHK_ST_FREE;
96 atomic_dec(&pblk->inflight_io);
99 /* Erase completion assumes that only one block is erased at the time */
100 static void pblk_end_io_erase(struct nvm_rq *rqd)
102 struct pblk *pblk = rqd->private;
104 __pblk_end_io_erase(pblk, rqd);
105 mempool_free(rqd, &pblk->e_rq_pool);
109 * Get information for all chunks from the device.
111 * The caller is responsible for freeing the returned structure
113 struct nvm_chk_meta *pblk_chunk_get_info(struct pblk *pblk)
115 struct nvm_tgt_dev *dev = pblk->dev;
116 struct nvm_geo *geo = &dev->geo;
117 struct nvm_chk_meta *meta;
124 len = geo->all_chunks * sizeof(*meta);
125 meta = kzalloc(len, GFP_KERNEL);
127 return ERR_PTR(-ENOMEM);
129 ret = nvm_get_chunk_meta(dev, meta, ppa, geo->all_chunks);
132 return ERR_PTR(-EIO);
138 struct nvm_chk_meta *pblk_chunk_get_off(struct pblk *pblk,
139 struct nvm_chk_meta *meta,
142 struct nvm_tgt_dev *dev = pblk->dev;
143 struct nvm_geo *geo = &dev->geo;
144 int ch_off = ppa.m.grp * geo->num_chk * geo->num_lun;
145 int lun_off = ppa.m.pu * geo->num_chk;
146 int chk_off = ppa.m.chk;
148 return meta + ch_off + lun_off + chk_off;
151 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
154 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
155 struct list_head *move_list = NULL;
157 /* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
158 * table is modified with reclaimed sectors, a check is done to endure
159 * that newer updates are not overwritten.
161 spin_lock(&line->lock);
162 WARN_ON(line->state == PBLK_LINESTATE_FREE);
164 if (test_and_set_bit(paddr, line->invalid_bitmap)) {
165 WARN_ONCE(1, "pblk: double invalidate\n");
166 spin_unlock(&line->lock);
169 le32_add_cpu(line->vsc, -1);
171 if (line->state == PBLK_LINESTATE_CLOSED)
172 move_list = pblk_line_gc_list(pblk, line);
173 spin_unlock(&line->lock);
176 spin_lock(&l_mg->gc_lock);
177 spin_lock(&line->lock);
178 /* Prevent moving a line that has just been chosen for GC */
179 if (line->state == PBLK_LINESTATE_GC) {
180 spin_unlock(&line->lock);
181 spin_unlock(&l_mg->gc_lock);
184 spin_unlock(&line->lock);
186 list_move_tail(&line->list, move_list);
187 spin_unlock(&l_mg->gc_lock);
191 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
193 struct pblk_line *line;
197 #ifdef CONFIG_NVM_PBLK_DEBUG
198 /* Callers must ensure that the ppa points to a device address */
199 BUG_ON(pblk_addr_in_cache(ppa));
200 BUG_ON(pblk_ppa_empty(ppa));
203 line_id = pblk_ppa_to_line(ppa);
204 line = &pblk->lines[line_id];
205 paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
207 __pblk_map_invalidate(pblk, line, paddr);
210 static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
211 unsigned int nr_secs)
215 spin_lock(&pblk->trans_lock);
216 for (lba = slba; lba < slba + nr_secs; lba++) {
219 ppa = pblk_trans_map_get(pblk, lba);
221 if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
222 pblk_map_invalidate(pblk, ppa);
224 pblk_ppa_set_empty(&ppa);
225 pblk_trans_map_set(pblk, lba, ppa);
227 spin_unlock(&pblk->trans_lock);
230 /* Caller must guarantee that the request is a valid type */
231 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type)
240 pool = &pblk->w_rq_pool;
241 rq_size = pblk_w_rq_size;
244 pool = &pblk->r_rq_pool;
245 rq_size = pblk_g_rq_size;
248 pool = &pblk->e_rq_pool;
249 rq_size = pblk_g_rq_size;
252 rqd = mempool_alloc(pool, GFP_KERNEL);
253 memset(rqd, 0, rq_size);
258 /* Typically used on completion path. Cannot guarantee request consistency */
259 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type)
261 struct nvm_tgt_dev *dev = pblk->dev;
266 kfree(((struct pblk_c_ctx *)nvm_rq_to_pdu(rqd))->lun_bitmap);
269 pool = &pblk->w_rq_pool;
272 pool = &pblk->r_rq_pool;
275 pool = &pblk->e_rq_pool;
278 pblk_err(pblk, "trying to free unknown rqd type\n");
283 nvm_dev_dma_free(dev->parent, rqd->meta_list,
285 mempool_free(rqd, pool);
288 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
294 WARN_ON(off + nr_pages != bio->bi_vcnt);
296 for (i = off; i < nr_pages + off; i++) {
297 bv = bio->bi_io_vec[i];
298 mempool_free(bv.bv_page, &pblk->page_bio_pool);
302 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
305 struct request_queue *q = pblk->dev->q;
309 for (i = 0; i < nr_pages; i++) {
310 page = mempool_alloc(&pblk->page_bio_pool, flags);
312 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
313 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
314 pblk_err(pblk, "could not add page to bio\n");
315 mempool_free(page, &pblk->page_bio_pool);
322 pblk_bio_free_pages(pblk, bio, (bio->bi_vcnt - i), i);
326 void pblk_write_kick(struct pblk *pblk)
328 wake_up_process(pblk->writer_ts);
329 mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
332 void pblk_write_timer_fn(struct timer_list *t)
334 struct pblk *pblk = from_timer(pblk, t, wtimer);
336 /* kick the write thread every tick to flush outstanding data */
337 pblk_write_kick(pblk);
340 void pblk_write_should_kick(struct pblk *pblk)
342 unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
344 if (secs_avail >= pblk->min_write_pgs)
345 pblk_write_kick(pblk);
348 static void pblk_wait_for_meta(struct pblk *pblk)
351 if (!atomic_read(&pblk->inflight_io))
358 static void pblk_flush_writer(struct pblk *pblk)
360 pblk_rb_flush(&pblk->rwb);
362 if (!pblk_rb_sync_count(&pblk->rwb))
365 pblk_write_kick(pblk);
370 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
372 struct pblk_line_meta *lm = &pblk->lm;
373 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
374 struct list_head *move_list = NULL;
375 int vsc = le32_to_cpu(*line->vsc);
377 lockdep_assert_held(&line->lock);
379 if (line->w_err_gc->has_write_err) {
380 if (line->gc_group != PBLK_LINEGC_WERR) {
381 line->gc_group = PBLK_LINEGC_WERR;
382 move_list = &l_mg->gc_werr_list;
383 pblk_rl_werr_line_in(&pblk->rl);
386 if (line->gc_group != PBLK_LINEGC_FULL) {
387 line->gc_group = PBLK_LINEGC_FULL;
388 move_list = &l_mg->gc_full_list;
390 } else if (vsc < lm->high_thrs) {
391 if (line->gc_group != PBLK_LINEGC_HIGH) {
392 line->gc_group = PBLK_LINEGC_HIGH;
393 move_list = &l_mg->gc_high_list;
395 } else if (vsc < lm->mid_thrs) {
396 if (line->gc_group != PBLK_LINEGC_MID) {
397 line->gc_group = PBLK_LINEGC_MID;
398 move_list = &l_mg->gc_mid_list;
400 } else if (vsc < line->sec_in_line) {
401 if (line->gc_group != PBLK_LINEGC_LOW) {
402 line->gc_group = PBLK_LINEGC_LOW;
403 move_list = &l_mg->gc_low_list;
405 } else if (vsc == line->sec_in_line) {
406 if (line->gc_group != PBLK_LINEGC_EMPTY) {
407 line->gc_group = PBLK_LINEGC_EMPTY;
408 move_list = &l_mg->gc_empty_list;
411 line->state = PBLK_LINESTATE_CORRUPT;
412 line->gc_group = PBLK_LINEGC_NONE;
413 move_list = &l_mg->corrupt_list;
414 pblk_err(pblk, "corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
417 lm->high_thrs, lm->mid_thrs);
423 void pblk_discard(struct pblk *pblk, struct bio *bio)
425 sector_t slba = pblk_get_lba(bio);
426 sector_t nr_secs = pblk_get_secs(bio);
428 pblk_invalidate_range(pblk, slba, nr_secs);
431 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
433 atomic_long_inc(&pblk->write_failed);
434 #ifdef CONFIG_NVM_PBLK_DEBUG
435 pblk_print_failed_rqd(pblk, rqd, rqd->error);
439 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
441 /* Empty page read is not necessarily an error (e.g., L2P recovery) */
442 if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
443 atomic_long_inc(&pblk->read_empty);
447 switch (rqd->error) {
448 case NVM_RSP_WARN_HIGHECC:
449 atomic_long_inc(&pblk->read_high_ecc);
451 case NVM_RSP_ERR_FAILECC:
452 case NVM_RSP_ERR_FAILCRC:
453 atomic_long_inc(&pblk->read_failed);
456 pblk_err(pblk, "unknown read error:%d\n", rqd->error);
458 #ifdef CONFIG_NVM_PBLK_DEBUG
459 pblk_print_failed_rqd(pblk, rqd, rqd->error);
463 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write)
465 pblk->sec_per_write = sec_per_write;
468 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
470 struct nvm_tgt_dev *dev = pblk->dev;
472 atomic_inc(&pblk->inflight_io);
474 #ifdef CONFIG_NVM_PBLK_DEBUG
475 if (pblk_check_io(pblk, rqd))
479 return nvm_submit_io(dev, rqd);
482 int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd)
484 struct nvm_tgt_dev *dev = pblk->dev;
486 atomic_inc(&pblk->inflight_io);
488 #ifdef CONFIG_NVM_PBLK_DEBUG
489 if (pblk_check_io(pblk, rqd))
493 return nvm_submit_io_sync(dev, rqd);
496 static void pblk_bio_map_addr_endio(struct bio *bio)
501 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
502 unsigned int nr_secs, unsigned int len,
503 int alloc_type, gfp_t gfp_mask)
505 struct nvm_tgt_dev *dev = pblk->dev;
511 if (alloc_type == PBLK_KMALLOC_META)
512 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
514 bio = bio_kmalloc(gfp_mask, nr_secs);
516 return ERR_PTR(-ENOMEM);
518 for (i = 0; i < nr_secs; i++) {
519 page = vmalloc_to_page(kaddr);
521 pblk_err(pblk, "could not map vmalloc bio\n");
523 bio = ERR_PTR(-ENOMEM);
527 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
528 if (ret != PAGE_SIZE) {
529 pblk_err(pblk, "could not add page to bio\n");
531 bio = ERR_PTR(-ENOMEM);
538 bio->bi_end_io = pblk_bio_map_addr_endio;
543 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
544 unsigned long secs_to_flush)
546 int max = pblk->sec_per_write;
547 int min = pblk->min_write_pgs;
548 int secs_to_sync = 0;
550 if (secs_avail >= max)
552 else if (secs_avail >= min)
553 secs_to_sync = min * (secs_avail / min);
554 else if (secs_to_flush)
560 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
565 spin_lock(&line->lock);
566 addr = find_next_zero_bit(line->map_bitmap,
567 pblk->lm.sec_per_line, line->cur_sec);
568 line->cur_sec = addr - nr_secs;
570 for (i = 0; i < nr_secs; i++, line->cur_sec--)
571 WARN_ON(!test_and_clear_bit(line->cur_sec, line->map_bitmap));
572 spin_unlock(&line->lock);
575 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
580 lockdep_assert_held(&line->lock);
582 /* logic error: ppa out-of-bounds. Prevent generating bad address */
583 if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
584 WARN(1, "pblk: page allocation out of bounds\n");
585 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
588 line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
589 pblk->lm.sec_per_line, line->cur_sec);
590 for (i = 0; i < nr_secs; i++, line->cur_sec++)
591 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
596 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
600 /* Lock needed in case a write fails and a recovery needs to remap
601 * failed write buffer entries
603 spin_lock(&line->lock);
604 addr = __pblk_alloc_page(pblk, line, nr_secs);
605 line->left_msecs -= nr_secs;
606 WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
607 spin_unlock(&line->lock);
612 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line)
616 spin_lock(&line->lock);
617 paddr = find_next_zero_bit(line->map_bitmap,
618 pblk->lm.sec_per_line, line->cur_sec);
619 spin_unlock(&line->lock);
625 * Submit emeta to one LUN in the raid line at the time to avoid a deadlock when
626 * taking the per LUN semaphore.
628 static int pblk_line_submit_emeta_io(struct pblk *pblk, struct pblk_line *line,
629 void *emeta_buf, u64 paddr, int dir)
631 struct nvm_tgt_dev *dev = pblk->dev;
632 struct nvm_geo *geo = &dev->geo;
633 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
634 struct pblk_line_meta *lm = &pblk->lm;
635 void *ppa_list, *meta_list;
638 dma_addr_t dma_ppa_list, dma_meta_list;
639 int min = pblk->min_write_pgs;
640 int left_ppas = lm->emeta_sec[0];
647 if (dir == PBLK_WRITE) {
648 bio_op = REQ_OP_WRITE;
649 cmd_op = NVM_OP_PWRITE;
650 } else if (dir == PBLK_READ) {
651 bio_op = REQ_OP_READ;
652 cmd_op = NVM_OP_PREAD;
656 meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
661 ppa_list = meta_list + pblk_dma_meta_size;
662 dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
665 memset(&rqd, 0, sizeof(struct nvm_rq));
667 rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
668 rq_len = rq_ppas * geo->csecs;
670 bio = pblk_bio_map_addr(pblk, emeta_buf, rq_ppas, rq_len,
671 l_mg->emeta_alloc_type, GFP_KERNEL);
677 bio->bi_iter.bi_sector = 0; /* internal bio */
678 bio_set_op_attrs(bio, bio_op, 0);
681 rqd.meta_list = meta_list;
682 rqd.ppa_list = ppa_list;
683 rqd.dma_meta_list = dma_meta_list;
684 rqd.dma_ppa_list = dma_ppa_list;
686 rqd.nr_ppas = rq_ppas;
688 if (dir == PBLK_WRITE) {
689 struct pblk_sec_meta *meta_list = rqd.meta_list;
692 for (i = 0; i < rqd.nr_ppas; ) {
693 spin_lock(&line->lock);
694 paddr = __pblk_alloc_page(pblk, line, min);
695 spin_unlock(&line->lock);
696 for (j = 0; j < min; j++, i++, paddr++) {
697 meta_list[i].lba = cpu_to_le64(ADDR_EMPTY);
699 addr_to_gen_ppa(pblk, paddr, id);
703 for (i = 0; i < rqd.nr_ppas; ) {
704 struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, id);
705 int pos = pblk_ppa_to_pos(geo, ppa);
707 if (pblk_io_aligned(pblk, rq_ppas))
710 while (test_bit(pos, line->blk_bitmap)) {
712 if (pblk_boundary_paddr_checks(pblk, paddr)) {
713 pblk_err(pblk, "corrupt emeta line:%d\n",
720 ppa = addr_to_gen_ppa(pblk, paddr, id);
721 pos = pblk_ppa_to_pos(geo, ppa);
724 if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
725 pblk_err(pblk, "corrupt emeta line:%d\n",
732 for (j = 0; j < min; j++, i++, paddr++)
734 addr_to_gen_ppa(pblk, paddr, line->id);
738 ret = pblk_submit_io_sync(pblk, &rqd);
740 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
745 atomic_dec(&pblk->inflight_io);
748 if (dir == PBLK_WRITE)
749 pblk_log_write_err(pblk, &rqd);
751 pblk_log_read_err(pblk, &rqd);
755 left_ppas -= rq_ppas;
759 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
763 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
765 struct nvm_tgt_dev *dev = pblk->dev;
766 struct nvm_geo *geo = &dev->geo;
767 struct pblk_line_meta *lm = &pblk->lm;
770 /* This usually only happens on bad lines */
771 bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
772 if (bit >= lm->blk_per_line)
775 return bit * geo->ws_opt;
778 static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
781 struct nvm_tgt_dev *dev = pblk->dev;
782 struct pblk_line_meta *lm = &pblk->lm;
785 __le64 *lba_list = NULL;
789 if (dir == PBLK_WRITE) {
790 bio_op = REQ_OP_WRITE;
791 cmd_op = NVM_OP_PWRITE;
792 lba_list = emeta_to_lbas(pblk, line->emeta->buf);
793 } else if (dir == PBLK_READ_RECOV || dir == PBLK_READ) {
794 bio_op = REQ_OP_READ;
795 cmd_op = NVM_OP_PREAD;
799 memset(&rqd, 0, sizeof(struct nvm_rq));
801 rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
806 rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
807 rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
809 bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
815 bio->bi_iter.bi_sector = 0; /* internal bio */
816 bio_set_op_attrs(bio, bio_op, 0);
821 rqd.nr_ppas = lm->smeta_sec;
823 for (i = 0; i < lm->smeta_sec; i++, paddr++) {
824 struct pblk_sec_meta *meta_list = rqd.meta_list;
826 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
828 if (dir == PBLK_WRITE) {
829 __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
831 meta_list[i].lba = lba_list[paddr] = addr_empty;
836 * This I/O is sent by the write thread when a line is replace. Since
837 * the write thread is the only one sending write and erase commands,
838 * there is no need to take the LUN semaphore.
840 ret = pblk_submit_io_sync(pblk, &rqd);
842 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
847 atomic_dec(&pblk->inflight_io);
850 if (dir == PBLK_WRITE) {
851 pblk_log_write_err(pblk, &rqd);
853 } else if (dir == PBLK_READ)
854 pblk_log_read_err(pblk, &rqd);
858 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
863 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
865 u64 bpaddr = pblk_line_smeta_start(pblk, line);
867 return pblk_line_submit_smeta_io(pblk, line, bpaddr, PBLK_READ_RECOV);
870 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
873 return pblk_line_submit_emeta_io(pblk, line, emeta_buf,
874 line->emeta_ssec, PBLK_READ);
877 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
880 rqd->opcode = NVM_OP_ERASE;
887 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
889 struct nvm_rq rqd = {NULL};
892 pblk_setup_e_rq(pblk, &rqd, ppa);
894 /* The write thread schedules erases so that it minimizes disturbances
895 * with writes. Thus, there is no need to take the LUN semaphore.
897 ret = pblk_submit_io_sync(pblk, &rqd);
899 __pblk_end_io_erase(pblk, &rqd);
904 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
906 struct pblk_line_meta *lm = &pblk->lm;
910 /* Erase only good blocks, one at a time */
912 spin_lock(&line->lock);
913 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
915 if (bit >= lm->blk_per_line) {
916 spin_unlock(&line->lock);
920 ppa = pblk->luns[bit].bppa; /* set ch and lun */
921 ppa.a.blk = line->id;
923 atomic_dec(&line->left_eblks);
924 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
925 spin_unlock(&line->lock);
927 ret = pblk_blk_erase_sync(pblk, ppa);
929 pblk_err(pblk, "failed to erase line %d\n", line->id);
937 static void pblk_line_setup_metadata(struct pblk_line *line,
938 struct pblk_line_mgmt *l_mg,
939 struct pblk_line_meta *lm)
943 lockdep_assert_held(&l_mg->free_lock);
946 meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
947 if (meta_line == PBLK_DATA_LINES) {
948 spin_unlock(&l_mg->free_lock);
950 spin_lock(&l_mg->free_lock);
954 set_bit(meta_line, &l_mg->meta_bitmap);
955 line->meta_line = meta_line;
957 line->smeta = l_mg->sline_meta[meta_line];
958 line->emeta = l_mg->eline_meta[meta_line];
960 memset(line->smeta, 0, lm->smeta_len);
961 memset(line->emeta->buf, 0, lm->emeta_len[0]);
963 line->emeta->mem = 0;
964 atomic_set(&line->emeta->sync, 0);
967 /* For now lines are always assumed full lines. Thus, smeta former and current
968 * lun bitmaps are omitted.
970 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
971 struct pblk_line *cur)
973 struct nvm_tgt_dev *dev = pblk->dev;
974 struct nvm_geo *geo = &dev->geo;
975 struct pblk_line_meta *lm = &pblk->lm;
976 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
977 struct pblk_emeta *emeta = line->emeta;
978 struct line_emeta *emeta_buf = emeta->buf;
979 struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
982 /* After erasing the line, new bad blocks might appear and we risk
983 * having an invalid line
985 nr_blk_line = lm->blk_per_line -
986 bitmap_weight(line->blk_bitmap, lm->blk_per_line);
987 if (nr_blk_line < lm->min_blk_line) {
988 spin_lock(&l_mg->free_lock);
989 spin_lock(&line->lock);
990 line->state = PBLK_LINESTATE_BAD;
991 spin_unlock(&line->lock);
993 list_add_tail(&line->list, &l_mg->bad_list);
994 spin_unlock(&l_mg->free_lock);
996 pblk_debug(pblk, "line %d is bad\n", line->id);
1001 /* Run-time metadata */
1002 line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1004 /* Mark LUNs allocated in this line (all for now) */
1005 bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1007 smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1008 memcpy(smeta_buf->header.uuid, pblk->instance_uuid, 16);
1009 smeta_buf->header.id = cpu_to_le32(line->id);
1010 smeta_buf->header.type = cpu_to_le16(line->type);
1011 smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1012 smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1014 /* Start metadata */
1015 smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1016 smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1018 /* Fill metadata among lines */
1020 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1021 smeta_buf->prev_id = cpu_to_le32(cur->id);
1022 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1024 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1027 /* All smeta must be set at this point */
1028 smeta_buf->header.crc = cpu_to_le32(
1029 pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1030 smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1033 memcpy(&emeta_buf->header, &smeta_buf->header,
1034 sizeof(struct line_header));
1036 emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1037 emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1038 emeta_buf->header.crc = cpu_to_le32(
1039 pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1041 emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1042 emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1043 emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1044 emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1045 emeta_buf->crc = cpu_to_le32(0);
1046 emeta_buf->prev_id = smeta_buf->prev_id;
1051 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1053 struct pblk_line_meta *lm = &pblk->lm;
1055 line->map_bitmap = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
1056 if (!line->map_bitmap)
1059 /* will be initialized using bb info from map_bitmap */
1060 line->invalid_bitmap = kmalloc(lm->sec_bitmap_len, GFP_KERNEL);
1061 if (!line->invalid_bitmap) {
1062 kfree(line->map_bitmap);
1063 line->map_bitmap = NULL;
1070 /* For now lines are always assumed full lines. Thus, smeta former and current
1071 * lun bitmaps are omitted.
1073 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1076 struct nvm_tgt_dev *dev = pblk->dev;
1077 struct nvm_geo *geo = &dev->geo;
1078 struct pblk_line_meta *lm = &pblk->lm;
1079 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1084 line->sec_in_line = lm->sec_per_line;
1086 /* Capture bad block information on line mapping bitmaps */
1087 while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1088 bit + 1)) < lm->blk_per_line) {
1089 off = bit * geo->ws_opt;
1090 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1092 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1094 line->sec_in_line -= geo->clba;
1097 /* Mark smeta metadata sectors as bad sectors */
1098 bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1099 off = bit * geo->ws_opt;
1100 bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1101 line->sec_in_line -= lm->smeta_sec;
1102 line->smeta_ssec = off;
1103 line->cur_sec = off + lm->smeta_sec;
1105 if (init && pblk_line_submit_smeta_io(pblk, line, off, PBLK_WRITE)) {
1106 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1110 bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1112 /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1113 * blocks to make sure that there are enough sectors to store emeta
1115 emeta_secs = lm->emeta_sec[0];
1116 off = lm->sec_per_line;
1117 while (emeta_secs) {
1119 if (!test_bit(off, line->invalid_bitmap)) {
1120 bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1121 emeta_secs -= geo->ws_opt;
1125 line->emeta_ssec = off;
1126 line->sec_in_line -= lm->emeta_sec[0];
1127 line->nr_valid_lbas = 0;
1128 line->left_msecs = line->sec_in_line;
1129 *line->vsc = cpu_to_le32(line->sec_in_line);
1131 if (lm->sec_per_line - line->sec_in_line !=
1132 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1133 spin_lock(&line->lock);
1134 line->state = PBLK_LINESTATE_BAD;
1135 spin_unlock(&line->lock);
1137 list_add_tail(&line->list, &l_mg->bad_list);
1138 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1146 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1148 struct pblk_line_meta *lm = &pblk->lm;
1149 struct nvm_tgt_dev *dev = pblk->dev;
1150 struct nvm_geo *geo = &dev->geo;
1151 int blk_to_erase = atomic_read(&line->blk_in_line);
1154 for (i = 0; i < lm->blk_per_line; i++) {
1155 struct pblk_lun *rlun = &pblk->luns[i];
1156 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1157 int state = line->chks[pos].state;
1159 /* Free chunks should not be erased */
1160 if (state & NVM_CHK_ST_FREE) {
1161 set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1162 line->erase_bitmap);
1167 return blk_to_erase;
1170 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1172 struct pblk_line_meta *lm = &pblk->lm;
1173 int blk_in_line = atomic_read(&line->blk_in_line);
1176 /* Bad blocks do not need to be erased */
1177 bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1179 spin_lock(&line->lock);
1181 /* If we have not written to this line, we need to mark up free chunks
1184 if (line->state == PBLK_LINESTATE_NEW) {
1185 blk_to_erase = pblk_prepare_new_line(pblk, line);
1186 line->state = PBLK_LINESTATE_FREE;
1188 blk_to_erase = blk_in_line;
1191 if (blk_in_line < lm->min_blk_line) {
1192 spin_unlock(&line->lock);
1196 if (line->state != PBLK_LINESTATE_FREE) {
1197 WARN(1, "pblk: corrupted line %d, state %d\n",
1198 line->id, line->state);
1199 spin_unlock(&line->lock);
1203 line->state = PBLK_LINESTATE_OPEN;
1205 atomic_set(&line->left_eblks, blk_to_erase);
1206 atomic_set(&line->left_seblks, blk_to_erase);
1208 line->meta_distance = lm->meta_distance;
1209 spin_unlock(&line->lock);
1211 kref_init(&line->ref);
1216 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1218 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1221 spin_lock(&l_mg->free_lock);
1222 l_mg->data_line = line;
1223 list_del(&line->list);
1225 ret = pblk_line_prepare(pblk, line);
1227 list_add(&line->list, &l_mg->free_list);
1228 spin_unlock(&l_mg->free_lock);
1231 spin_unlock(&l_mg->free_lock);
1233 ret = pblk_line_alloc_bitmaps(pblk, line);
1237 if (!pblk_line_init_bb(pblk, line, 0)) {
1238 list_add(&line->list, &l_mg->free_list);
1242 pblk_rl_free_lines_dec(&pblk->rl, line, true);
1246 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1248 kfree(line->map_bitmap);
1249 line->map_bitmap = NULL;
1254 static void pblk_line_reinit(struct pblk_line *line)
1256 *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1258 line->map_bitmap = NULL;
1259 line->invalid_bitmap = NULL;
1264 void pblk_line_free(struct pblk_line *line)
1266 kfree(line->map_bitmap);
1267 kfree(line->invalid_bitmap);
1269 pblk_line_reinit(line);
1272 struct pblk_line *pblk_line_get(struct pblk *pblk)
1274 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1275 struct pblk_line_meta *lm = &pblk->lm;
1276 struct pblk_line *line;
1279 lockdep_assert_held(&l_mg->free_lock);
1282 if (list_empty(&l_mg->free_list)) {
1283 pblk_err(pblk, "no free lines\n");
1287 line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1288 list_del(&line->list);
1289 l_mg->nr_free_lines--;
1291 bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1292 if (unlikely(bit >= lm->blk_per_line)) {
1293 spin_lock(&line->lock);
1294 line->state = PBLK_LINESTATE_BAD;
1295 spin_unlock(&line->lock);
1297 list_add_tail(&line->list, &l_mg->bad_list);
1299 pblk_debug(pblk, "line %d is bad\n", line->id);
1303 ret = pblk_line_prepare(pblk, line);
1307 list_add(&line->list, &l_mg->bad_list);
1310 list_add(&line->list, &l_mg->corrupt_list);
1313 pblk_err(pblk, "failed to prepare line %d\n", line->id);
1314 list_add(&line->list, &l_mg->free_list);
1315 l_mg->nr_free_lines++;
1323 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1324 struct pblk_line *line)
1326 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1327 struct pblk_line *retry_line;
1330 spin_lock(&l_mg->free_lock);
1331 retry_line = pblk_line_get(pblk);
1333 l_mg->data_line = NULL;
1334 spin_unlock(&l_mg->free_lock);
1338 retry_line->map_bitmap = line->map_bitmap;
1339 retry_line->invalid_bitmap = line->invalid_bitmap;
1340 retry_line->smeta = line->smeta;
1341 retry_line->emeta = line->emeta;
1342 retry_line->meta_line = line->meta_line;
1344 pblk_line_reinit(line);
1346 l_mg->data_line = retry_line;
1347 spin_unlock(&l_mg->free_lock);
1349 pblk_rl_free_lines_dec(&pblk->rl, line, false);
1351 if (pblk_line_erase(pblk, retry_line))
1357 static void pblk_set_space_limit(struct pblk *pblk)
1359 struct pblk_rl *rl = &pblk->rl;
1361 atomic_set(&rl->rb_space, 0);
1364 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1366 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1367 struct pblk_line *line;
1369 spin_lock(&l_mg->free_lock);
1370 line = pblk_line_get(pblk);
1372 spin_unlock(&l_mg->free_lock);
1376 line->seq_nr = l_mg->d_seq_nr++;
1377 line->type = PBLK_LINETYPE_DATA;
1378 l_mg->data_line = line;
1380 pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1382 /* Allocate next line for preparation */
1383 l_mg->data_next = pblk_line_get(pblk);
1384 if (!l_mg->data_next) {
1385 /* If we cannot get a new line, we need to stop the pipeline.
1386 * Only allow as many writes in as we can store safely and then
1389 pblk_set_space_limit(pblk);
1391 l_mg->data_next = NULL;
1393 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1394 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1396 spin_unlock(&l_mg->free_lock);
1398 if (pblk_line_alloc_bitmaps(pblk, line))
1401 if (pblk_line_erase(pblk, line)) {
1402 line = pblk_line_retry(pblk, line);
1408 if (!pblk_line_init_metadata(pblk, line, NULL)) {
1409 line = pblk_line_retry(pblk, line);
1416 if (!pblk_line_init_bb(pblk, line, 1)) {
1417 line = pblk_line_retry(pblk, line);
1424 pblk_rl_free_lines_dec(&pblk->rl, line, true);
1429 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1431 lockdep_assert_held(&pblk->l_mg.free_lock);
1433 pblk_set_space_limit(pblk);
1434 pblk->state = PBLK_STATE_STOPPING;
1437 static void pblk_line_close_meta_sync(struct pblk *pblk)
1439 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1440 struct pblk_line_meta *lm = &pblk->lm;
1441 struct pblk_line *line, *tline;
1444 spin_lock(&l_mg->close_lock);
1445 if (list_empty(&l_mg->emeta_list)) {
1446 spin_unlock(&l_mg->close_lock);
1450 list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1451 spin_unlock(&l_mg->close_lock);
1453 list_for_each_entry_safe(line, tline, &list, list) {
1454 struct pblk_emeta *emeta = line->emeta;
1456 while (emeta->mem < lm->emeta_len[0]) {
1459 ret = pblk_submit_meta_io(pblk, line);
1461 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1468 pblk_wait_for_meta(pblk);
1469 flush_workqueue(pblk->close_wq);
1472 void __pblk_pipeline_flush(struct pblk *pblk)
1474 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1477 spin_lock(&l_mg->free_lock);
1478 if (pblk->state == PBLK_STATE_RECOVERING ||
1479 pblk->state == PBLK_STATE_STOPPED) {
1480 spin_unlock(&l_mg->free_lock);
1483 pblk->state = PBLK_STATE_RECOVERING;
1484 spin_unlock(&l_mg->free_lock);
1486 pblk_flush_writer(pblk);
1487 pblk_wait_for_meta(pblk);
1489 ret = pblk_recov_pad(pblk);
1491 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1495 flush_workqueue(pblk->bb_wq);
1496 pblk_line_close_meta_sync(pblk);
1499 void __pblk_pipeline_stop(struct pblk *pblk)
1501 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1503 spin_lock(&l_mg->free_lock);
1504 pblk->state = PBLK_STATE_STOPPED;
1505 l_mg->data_line = NULL;
1506 l_mg->data_next = NULL;
1507 spin_unlock(&l_mg->free_lock);
1510 void pblk_pipeline_stop(struct pblk *pblk)
1512 __pblk_pipeline_flush(pblk);
1513 __pblk_pipeline_stop(pblk);
1516 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1518 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1519 struct pblk_line *cur, *new = NULL;
1520 unsigned int left_seblks;
1522 cur = l_mg->data_line;
1523 new = l_mg->data_next;
1526 l_mg->data_line = new;
1528 spin_lock(&l_mg->free_lock);
1529 pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1530 spin_unlock(&l_mg->free_lock);
1533 left_seblks = atomic_read(&new->left_seblks);
1535 /* If line is not fully erased, erase it */
1536 if (atomic_read(&new->left_eblks)) {
1537 if (pblk_line_erase(pblk, new))
1545 if (pblk_line_alloc_bitmaps(pblk, new))
1549 if (!pblk_line_init_metadata(pblk, new, cur)) {
1550 new = pblk_line_retry(pblk, new);
1557 if (!pblk_line_init_bb(pblk, new, 1)) {
1558 new = pblk_line_retry(pblk, new);
1565 pblk_rl_free_lines_dec(&pblk->rl, new, true);
1567 /* Allocate next line for preparation */
1568 spin_lock(&l_mg->free_lock);
1569 l_mg->data_next = pblk_line_get(pblk);
1570 if (!l_mg->data_next) {
1571 /* If we cannot get a new line, we need to stop the pipeline.
1572 * Only allow as many writes in as we can store safely and then
1575 pblk_stop_writes(pblk, new);
1576 l_mg->data_next = NULL;
1578 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1579 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1581 spin_unlock(&l_mg->free_lock);
1587 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1589 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1590 struct pblk_gc *gc = &pblk->gc;
1592 spin_lock(&line->lock);
1593 WARN_ON(line->state != PBLK_LINESTATE_GC);
1594 line->state = PBLK_LINESTATE_FREE;
1595 line->gc_group = PBLK_LINEGC_NONE;
1596 pblk_line_free(line);
1598 if (line->w_err_gc->has_write_err) {
1599 pblk_rl_werr_line_out(&pblk->rl);
1600 line->w_err_gc->has_write_err = 0;
1603 spin_unlock(&line->lock);
1604 atomic_dec(&gc->pipeline_gc);
1606 spin_lock(&l_mg->free_lock);
1607 list_add_tail(&line->list, &l_mg->free_list);
1608 l_mg->nr_free_lines++;
1609 spin_unlock(&l_mg->free_lock);
1611 pblk_rl_free_lines_inc(&pblk->rl, line);
1614 static void pblk_line_put_ws(struct work_struct *work)
1616 struct pblk_line_ws *line_put_ws = container_of(work,
1617 struct pblk_line_ws, ws);
1618 struct pblk *pblk = line_put_ws->pblk;
1619 struct pblk_line *line = line_put_ws->line;
1621 __pblk_line_put(pblk, line);
1622 mempool_free(line_put_ws, &pblk->gen_ws_pool);
1625 void pblk_line_put(struct kref *ref)
1627 struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1628 struct pblk *pblk = line->pblk;
1630 __pblk_line_put(pblk, line);
1633 void pblk_line_put_wq(struct kref *ref)
1635 struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1636 struct pblk *pblk = line->pblk;
1637 struct pblk_line_ws *line_put_ws;
1639 line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1643 line_put_ws->pblk = pblk;
1644 line_put_ws->line = line;
1645 line_put_ws->priv = NULL;
1647 INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1648 queue_work(pblk->r_end_wq, &line_put_ws->ws);
1651 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1656 rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1658 pblk_setup_e_rq(pblk, rqd, ppa);
1660 rqd->end_io = pblk_end_io_erase;
1661 rqd->private = pblk;
1663 /* The write thread schedules erases so that it minimizes disturbances
1664 * with writes. Thus, there is no need to take the LUN semaphore.
1666 err = pblk_submit_io(pblk, rqd);
1668 struct nvm_tgt_dev *dev = pblk->dev;
1669 struct nvm_geo *geo = &dev->geo;
1671 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1672 pblk_ppa_to_line(ppa),
1673 pblk_ppa_to_pos(geo, ppa));
1679 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1681 return pblk->l_mg.data_line;
1684 /* For now, always erase next line */
1685 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1687 return pblk->l_mg.data_next;
1690 int pblk_line_is_full(struct pblk_line *line)
1692 return (line->left_msecs == 0);
1695 static void pblk_line_should_sync_meta(struct pblk *pblk)
1697 if (pblk_rl_is_limit(&pblk->rl))
1698 pblk_line_close_meta_sync(pblk);
1701 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1703 struct nvm_tgt_dev *dev = pblk->dev;
1704 struct nvm_geo *geo = &dev->geo;
1705 struct pblk_line_meta *lm = &pblk->lm;
1706 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1707 struct list_head *move_list;
1710 #ifdef CONFIG_NVM_PBLK_DEBUG
1711 WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1712 "pblk: corrupt closed line %d\n", line->id);
1715 spin_lock(&l_mg->free_lock);
1716 WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1717 spin_unlock(&l_mg->free_lock);
1719 spin_lock(&l_mg->gc_lock);
1720 spin_lock(&line->lock);
1721 WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1722 line->state = PBLK_LINESTATE_CLOSED;
1723 move_list = pblk_line_gc_list(pblk, line);
1725 list_add_tail(&line->list, move_list);
1727 kfree(line->map_bitmap);
1728 line->map_bitmap = NULL;
1732 for (i = 0; i < lm->blk_per_line; i++) {
1733 struct pblk_lun *rlun = &pblk->luns[i];
1734 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1735 int state = line->chks[pos].state;
1737 if (!(state & NVM_CHK_ST_OFFLINE))
1738 state = NVM_CHK_ST_CLOSED;
1741 spin_unlock(&line->lock);
1742 spin_unlock(&l_mg->gc_lock);
1745 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1747 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1748 struct pblk_line_meta *lm = &pblk->lm;
1749 struct pblk_emeta *emeta = line->emeta;
1750 struct line_emeta *emeta_buf = emeta->buf;
1751 struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1753 /* No need for exact vsc value; avoid a big line lock and take aprox. */
1754 memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1755 memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1757 wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1758 wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1759 wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1761 emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1762 emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1764 spin_lock(&l_mg->close_lock);
1765 spin_lock(&line->lock);
1767 /* Update the in-memory start address for emeta, in case it has
1768 * shifted due to write errors
1770 if (line->emeta_ssec != line->cur_sec)
1771 line->emeta_ssec = line->cur_sec;
1773 list_add_tail(&line->list, &l_mg->emeta_list);
1774 spin_unlock(&line->lock);
1775 spin_unlock(&l_mg->close_lock);
1777 pblk_line_should_sync_meta(pblk);
1782 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1784 struct pblk_line_meta *lm = &pblk->lm;
1785 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1786 unsigned int lba_list_size = lm->emeta_len[2];
1787 struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1788 struct pblk_emeta *emeta = line->emeta;
1790 w_err_gc->lba_list = pblk_malloc(lba_list_size,
1791 l_mg->emeta_alloc_type, GFP_KERNEL);
1792 memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1796 void pblk_line_close_ws(struct work_struct *work)
1798 struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1800 struct pblk *pblk = line_ws->pblk;
1801 struct pblk_line *line = line_ws->line;
1802 struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1804 /* Write errors makes the emeta start address stored in smeta invalid,
1805 * so keep a copy of the lba list until we've gc'd the line
1807 if (w_err_gc->has_write_err)
1808 pblk_save_lba_list(pblk, line);
1810 pblk_line_close(pblk, line);
1811 mempool_free(line_ws, &pblk->gen_ws_pool);
1814 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1815 void (*work)(struct work_struct *), gfp_t gfp_mask,
1816 struct workqueue_struct *wq)
1818 struct pblk_line_ws *line_ws;
1820 line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1822 line_ws->pblk = pblk;
1823 line_ws->line = line;
1824 line_ws->priv = priv;
1826 INIT_WORK(&line_ws->ws, work);
1827 queue_work(wq, &line_ws->ws);
1830 static void __pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list,
1831 int nr_ppas, int pos)
1833 struct pblk_lun *rlun = &pblk->luns[pos];
1837 * Only send one inflight I/O per LUN. Since we map at a page
1838 * granurality, all ppas in the I/O will map to the same LUN
1840 #ifdef CONFIG_NVM_PBLK_DEBUG
1843 for (i = 1; i < nr_ppas; i++)
1844 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1845 ppa_list[0].a.ch != ppa_list[i].a.ch);
1848 ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1849 if (ret == -ETIME || ret == -EINTR)
1850 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1854 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1856 struct nvm_tgt_dev *dev = pblk->dev;
1857 struct nvm_geo *geo = &dev->geo;
1858 int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1860 __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1863 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1864 unsigned long *lun_bitmap)
1866 struct nvm_tgt_dev *dev = pblk->dev;
1867 struct nvm_geo *geo = &dev->geo;
1868 int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1870 /* If the LUN has been locked for this same request, do no attempt to
1873 if (test_and_set_bit(pos, lun_bitmap))
1876 __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1879 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1881 struct nvm_tgt_dev *dev = pblk->dev;
1882 struct nvm_geo *geo = &dev->geo;
1883 struct pblk_lun *rlun;
1884 int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1886 #ifdef CONFIG_NVM_PBLK_DEBUG
1889 for (i = 1; i < nr_ppas; i++)
1890 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1891 ppa_list[0].a.ch != ppa_list[i].a.ch);
1894 rlun = &pblk->luns[pos];
1898 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1899 unsigned long *lun_bitmap)
1901 struct nvm_tgt_dev *dev = pblk->dev;
1902 struct nvm_geo *geo = &dev->geo;
1903 struct pblk_lun *rlun;
1904 int num_lun = geo->all_luns;
1907 while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
1908 rlun = &pblk->luns[bit];
1913 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1915 struct ppa_addr ppa_l2p;
1917 /* logic error: lba out-of-bounds. Ignore update */
1918 if (!(lba < pblk->rl.nr_secs)) {
1919 WARN(1, "pblk: corrupted L2P map request\n");
1923 spin_lock(&pblk->trans_lock);
1924 ppa_l2p = pblk_trans_map_get(pblk, lba);
1926 if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
1927 pblk_map_invalidate(pblk, ppa_l2p);
1929 pblk_trans_map_set(pblk, lba, ppa);
1930 spin_unlock(&pblk->trans_lock);
1933 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1936 #ifdef CONFIG_NVM_PBLK_DEBUG
1937 /* Callers must ensure that the ppa points to a cache address */
1938 BUG_ON(!pblk_addr_in_cache(ppa));
1939 BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1942 pblk_update_map(pblk, lba, ppa);
1945 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
1946 struct pblk_line *gc_line, u64 paddr_gc)
1948 struct ppa_addr ppa_l2p, ppa_gc;
1951 #ifdef CONFIG_NVM_PBLK_DEBUG
1952 /* Callers must ensure that the ppa points to a cache address */
1953 BUG_ON(!pblk_addr_in_cache(ppa_new));
1954 BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
1957 /* logic error: lba out-of-bounds. Ignore update */
1958 if (!(lba < pblk->rl.nr_secs)) {
1959 WARN(1, "pblk: corrupted L2P map request\n");
1963 spin_lock(&pblk->trans_lock);
1964 ppa_l2p = pblk_trans_map_get(pblk, lba);
1965 ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
1967 if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
1968 spin_lock(&gc_line->lock);
1969 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
1970 "pblk: corrupted GC update");
1971 spin_unlock(&gc_line->lock);
1977 pblk_trans_map_set(pblk, lba, ppa_new);
1979 spin_unlock(&pblk->trans_lock);
1983 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
1984 struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
1986 struct ppa_addr ppa_l2p;
1988 #ifdef CONFIG_NVM_PBLK_DEBUG
1989 /* Callers must ensure that the ppa points to a device address */
1990 BUG_ON(pblk_addr_in_cache(ppa_mapped));
1992 /* Invalidate and discard padded entries */
1993 if (lba == ADDR_EMPTY) {
1994 atomic64_inc(&pblk->pad_wa);
1995 #ifdef CONFIG_NVM_PBLK_DEBUG
1996 atomic_long_inc(&pblk->padded_wb);
1998 if (!pblk_ppa_empty(ppa_mapped))
1999 pblk_map_invalidate(pblk, ppa_mapped);
2003 /* logic error: lba out-of-bounds. Ignore update */
2004 if (!(lba < pblk->rl.nr_secs)) {
2005 WARN(1, "pblk: corrupted L2P map request\n");
2009 spin_lock(&pblk->trans_lock);
2010 ppa_l2p = pblk_trans_map_get(pblk, lba);
2012 /* Do not update L2P if the cacheline has been updated. In this case,
2013 * the mapped ppa must be invalidated
2015 if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2016 if (!pblk_ppa_empty(ppa_mapped))
2017 pblk_map_invalidate(pblk, ppa_mapped);
2021 #ifdef CONFIG_NVM_PBLK_DEBUG
2022 WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2025 pblk_trans_map_set(pblk, lba, ppa_mapped);
2027 spin_unlock(&pblk->trans_lock);
2030 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2031 sector_t blba, int nr_secs)
2035 spin_lock(&pblk->trans_lock);
2036 for (i = 0; i < nr_secs; i++) {
2037 struct ppa_addr ppa;
2039 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2041 /* If the L2P entry maps to a line, the reference is valid */
2042 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2043 int line_id = pblk_ppa_to_line(ppa);
2044 struct pblk_line *line = &pblk->lines[line_id];
2046 kref_get(&line->ref);
2049 spin_unlock(&pblk->trans_lock);
2052 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2053 u64 *lba_list, int nr_secs)
2058 spin_lock(&pblk->trans_lock);
2059 for (i = 0; i < nr_secs; i++) {
2061 if (lba != ADDR_EMPTY) {
2062 /* logic error: lba out-of-bounds. Ignore update */
2063 if (!(lba < pblk->rl.nr_secs)) {
2064 WARN(1, "pblk: corrupted L2P map request\n");
2067 ppas[i] = pblk_trans_map_get(pblk, lba);
2070 spin_unlock(&pblk->trans_lock);