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-read.c - pblk's read path
21 * There is no guarantee that the value read from cache has not been updated and
22 * resides at another location in the cache. We guarantee though that if the
23 * value is read from the cache, it belongs to the mapped lba. In order to
24 * guarantee and order between writes and reads are ordered, a flush must be
27 static int pblk_read_from_cache(struct pblk *pblk, struct bio *bio,
28 sector_t lba, struct ppa_addr ppa,
31 #ifdef CONFIG_NVM_DEBUG
32 /* Callers must ensure that the ppa points to a cache address */
33 BUG_ON(pblk_ppa_empty(ppa));
34 BUG_ON(!pblk_addr_in_cache(ppa));
37 return pblk_rb_copy_to_bio(&pblk->rwb, bio, lba,
38 pblk_addr_to_cacheline(ppa), bio_iter);
41 static void pblk_read_ppalist_rq(struct pblk *pblk, struct nvm_rq *rqd,
42 unsigned long *read_bitmap)
44 struct bio *bio = rqd->bio;
45 struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
46 sector_t blba = pblk_get_lba(bio);
47 int nr_secs = rqd->nr_ppas;
51 /* logic error: lba out-of-bounds. Ignore read request */
52 if (blba + nr_secs >= pblk->rl.nr_secs) {
53 WARN(1, "pblk: read lbas out of bounds\n");
57 pblk_lookup_l2p_seq(pblk, ppas, blba, nr_secs);
59 for (i = 0; i < nr_secs; i++) {
60 struct ppa_addr p = ppas[i];
61 sector_t lba = blba + i;
64 if (pblk_ppa_empty(p)) {
65 WARN_ON(test_and_set_bit(i, read_bitmap));
69 /* Try to read from write buffer. The address is later checked
70 * on the write buffer to prevent retrieving overwritten data.
72 if (pblk_addr_in_cache(p)) {
73 if (!pblk_read_from_cache(pblk, bio, lba, p, i)) {
74 pblk_lookup_l2p_seq(pblk, &p, lba, 1);
77 WARN_ON(test_and_set_bit(i, read_bitmap));
79 #ifdef CONFIG_NVM_DEBUG
80 atomic_long_inc(&pblk->cache_reads);
83 /* Read from media non-cached sectors */
84 rqd->ppa_list[j++] = p;
88 bio_advance(bio, PBLK_EXPOSED_PAGE_SIZE);
91 #ifdef CONFIG_NVM_DEBUG
92 atomic_long_add(nr_secs, &pblk->inflight_reads);
96 static int pblk_submit_read_io(struct pblk *pblk, struct nvm_rq *rqd)
100 rqd->flags = pblk_set_read_mode(pblk);
102 err = pblk_submit_io(pblk, rqd);
109 static void pblk_end_io_read(struct nvm_rq *rqd)
111 struct pblk *pblk = rqd->private;
112 struct nvm_tgt_dev *dev = pblk->dev;
113 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
114 struct bio *bio = rqd->bio;
117 pblk_log_read_err(pblk, rqd);
118 #ifdef CONFIG_NVM_DEBUG
120 WARN_ONCE(bio->bi_status, "pblk: corrupted read error\n");
123 if (rqd->nr_ppas > 1)
124 nvm_dev_dma_free(dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
127 if (r_ctx->private) {
128 struct bio *orig_bio = r_ctx->private;
130 #ifdef CONFIG_NVM_DEBUG
131 WARN_ONCE(orig_bio->bi_status, "pblk: corrupted read bio\n");
137 #ifdef CONFIG_NVM_DEBUG
138 atomic_long_add(rqd->nr_ppas, &pblk->sync_reads);
139 atomic_long_sub(rqd->nr_ppas, &pblk->inflight_reads);
142 pblk_free_rqd(pblk, rqd, READ);
145 static int pblk_fill_partial_read_bio(struct pblk *pblk, struct nvm_rq *rqd,
146 unsigned int bio_init_idx,
147 unsigned long *read_bitmap)
149 struct bio *new_bio, *bio = rqd->bio;
150 struct bio_vec src_bv, dst_bv;
151 void *ppa_ptr = NULL;
153 dma_addr_t dma_ppa_list = 0;
154 int nr_secs = rqd->nr_ppas;
155 int nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
157 DECLARE_COMPLETION_ONSTACK(wait);
159 new_bio = bio_alloc(GFP_KERNEL, nr_holes);
161 pr_err("pblk: could not alloc read bio\n");
165 if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
168 if (nr_holes != new_bio->bi_vcnt) {
169 pr_err("pblk: malformed bio\n");
173 new_bio->bi_iter.bi_sector = 0; /* internal bio */
174 bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
175 new_bio->bi_private = &wait;
176 new_bio->bi_end_io = pblk_end_bio_sync;
179 rqd->nr_ppas = nr_holes;
182 if (unlikely(nr_secs > 1 && nr_holes == 1)) {
183 ppa_ptr = rqd->ppa_list;
184 dma_ppa_list = rqd->dma_ppa_list;
185 rqd->ppa_addr = rqd->ppa_list[0];
188 ret = pblk_submit_read_io(pblk, rqd);
191 pr_err("pblk: read IO submission failed\n");
195 if (!wait_for_completion_io_timeout(&wait,
196 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
197 pr_err("pblk: partial read I/O timed out\n");
201 atomic_long_inc(&pblk->read_failed);
202 #ifdef CONFIG_NVM_DEBUG
203 pblk_print_failed_rqd(pblk, rqd, rqd->error);
207 if (unlikely(nr_secs > 1 && nr_holes == 1)) {
208 rqd->ppa_list = ppa_ptr;
209 rqd->dma_ppa_list = dma_ppa_list;
212 /* Fill the holes in the original bio */
214 hole = find_first_zero_bit(read_bitmap, nr_secs);
216 src_bv = new_bio->bi_io_vec[i++];
217 dst_bv = bio->bi_io_vec[bio_init_idx + hole];
219 src_p = kmap_atomic(src_bv.bv_page);
220 dst_p = kmap_atomic(dst_bv.bv_page);
222 memcpy(dst_p + dst_bv.bv_offset,
223 src_p + src_bv.bv_offset,
224 PBLK_EXPOSED_PAGE_SIZE);
226 kunmap_atomic(src_p);
227 kunmap_atomic(dst_p);
229 mempool_free(src_bv.bv_page, pblk->page_pool);
231 hole = find_next_zero_bit(read_bitmap, nr_secs, hole + 1);
232 } while (hole < nr_secs);
236 /* Complete the original bio and associated request */
238 rqd->nr_ppas = nr_secs;
242 pblk_end_io_read(rqd);
246 /* Free allocated pages in new bio */
247 pblk_bio_free_pages(pblk, bio, 0, new_bio->bi_vcnt);
249 pblk_end_io_read(rqd);
253 static void pblk_read_rq(struct pblk *pblk, struct nvm_rq *rqd,
254 unsigned long *read_bitmap)
256 struct bio *bio = rqd->bio;
258 sector_t lba = pblk_get_lba(bio);
260 /* logic error: lba out-of-bounds. Ignore read request */
261 if (lba >= pblk->rl.nr_secs) {
262 WARN(1, "pblk: read lba out of bounds\n");
266 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
268 #ifdef CONFIG_NVM_DEBUG
269 atomic_long_inc(&pblk->inflight_reads);
273 if (pblk_ppa_empty(ppa)) {
274 WARN_ON(test_and_set_bit(0, read_bitmap));
278 /* Try to read from write buffer. The address is later checked on the
279 * write buffer to prevent retrieving overwritten data.
281 if (pblk_addr_in_cache(ppa)) {
282 if (!pblk_read_from_cache(pblk, bio, lba, ppa, 0)) {
283 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
286 WARN_ON(test_and_set_bit(0, read_bitmap));
287 #ifdef CONFIG_NVM_DEBUG
288 atomic_long_inc(&pblk->cache_reads);
295 int pblk_submit_read(struct pblk *pblk, struct bio *bio)
297 struct nvm_tgt_dev *dev = pblk->dev;
298 unsigned int nr_secs = pblk_get_secs(bio);
300 unsigned long read_bitmap; /* Max 64 ppas per request */
301 unsigned int bio_init_idx;
302 int ret = NVM_IO_ERR;
304 if (nr_secs > PBLK_MAX_REQ_ADDRS)
307 bitmap_zero(&read_bitmap, nr_secs);
309 rqd = pblk_alloc_rqd(pblk, READ);
311 pr_err_ratelimited("pblk: not able to alloc rqd");
315 rqd->opcode = NVM_OP_PREAD;
317 rqd->nr_ppas = nr_secs;
319 rqd->end_io = pblk_end_io_read;
321 /* Save the index for this bio's start. This is needed in case
322 * we need to fill a partial read.
324 bio_init_idx = pblk_get_bi_idx(bio);
327 rqd->ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
329 if (!rqd->ppa_list) {
330 pr_err("pblk: not able to allocate ppa list\n");
334 pblk_read_ppalist_rq(pblk, rqd, &read_bitmap);
336 pblk_read_rq(pblk, rqd, &read_bitmap);
340 if (bitmap_full(&read_bitmap, nr_secs)) {
342 pblk_end_io_read(rqd);
346 /* All sectors are to be read from the device */
347 if (bitmap_empty(&read_bitmap, rqd->nr_ppas)) {
348 struct bio *int_bio = NULL;
349 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
351 /* Clone read bio to deal with read errors internally */
352 int_bio = bio_clone_fast(bio, GFP_KERNEL, pblk_bio_set);
354 pr_err("pblk: could not clone read bio\n");
359 r_ctx->private = bio;
361 ret = pblk_submit_read_io(pblk, rqd);
363 pr_err("pblk: read IO submission failed\n");
372 /* The read bio request could be partially filled by the write buffer,
373 * but there are some holes that need to be read from the drive.
375 ret = pblk_fill_partial_read_bio(pblk, rqd, bio_init_idx, &read_bitmap);
377 pr_err("pblk: failed to perform partial read\n");
384 pblk_free_rqd(pblk, rqd, READ);
388 static int read_ppalist_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
389 struct pblk_line *line, u64 *lba_list,
390 unsigned int nr_secs)
392 struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
396 pblk_lookup_l2p_rand(pblk, ppas, lba_list, nr_secs);
398 for (i = 0; i < nr_secs; i++) {
399 if (pblk_addr_in_cache(ppas[i]) || ppas[i].g.blk != line->id ||
400 pblk_ppa_empty(ppas[i])) {
401 lba_list[i] = ADDR_EMPTY;
405 rqd->ppa_list[valid_secs++] = ppas[i];
408 #ifdef CONFIG_NVM_DEBUG
409 atomic_long_add(valid_secs, &pblk->inflight_reads);
414 static int read_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
415 struct pblk_line *line, sector_t lba)
420 if (lba == ADDR_EMPTY)
423 /* logic error: lba out-of-bounds */
424 if (lba >= pblk->rl.nr_secs) {
425 WARN(1, "pblk: read lba out of bounds\n");
429 spin_lock(&pblk->trans_lock);
430 ppa = pblk_trans_map_get(pblk, lba);
431 spin_unlock(&pblk->trans_lock);
433 /* Ignore updated values until the moment */
434 if (pblk_addr_in_cache(ppa) || ppa.g.blk != line->id ||
441 #ifdef CONFIG_NVM_DEBUG
442 atomic_long_inc(&pblk->inflight_reads);
449 int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
450 unsigned int nr_secs, unsigned int *secs_to_gc,
451 struct pblk_line *line)
453 struct nvm_tgt_dev *dev = pblk->dev;
454 struct nvm_geo *geo = &dev->geo;
455 struct request_queue *q = dev->q;
459 DECLARE_COMPLETION_ONSTACK(wait);
461 memset(&rqd, 0, sizeof(struct nvm_rq));
464 rqd.ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
469 *secs_to_gc = read_ppalist_rq_gc(pblk, &rqd, line, lba_list,
471 if (*secs_to_gc == 1) {
474 ppa = rqd.ppa_list[0];
475 nvm_dev_dma_free(dev->parent, rqd.ppa_list,
480 *secs_to_gc = read_rq_gc(pblk, &rqd, line, lba_list[0]);
486 data_len = (*secs_to_gc) * geo->sec_size;
487 bio = bio_map_kern(q, data, data_len, GFP_KERNEL);
489 pr_err("pblk: could not allocate GC bio (%lu)\n", PTR_ERR(bio));
493 bio->bi_iter.bi_sector = 0; /* internal bio */
494 bio_set_op_attrs(bio, REQ_OP_READ, 0);
496 rqd.opcode = NVM_OP_PREAD;
497 rqd.end_io = pblk_end_io_sync;
499 rqd.nr_ppas = *secs_to_gc;
502 ret = pblk_submit_read_io(pblk, &rqd);
505 pr_err("pblk: GC read request failed\n");
509 if (!wait_for_completion_io_timeout(&wait,
510 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
511 pr_err("pblk: GC read I/O timed out\n");
515 atomic_long_inc(&pblk->read_failed_gc);
516 #ifdef CONFIG_NVM_DEBUG
517 pblk_print_failed_rqd(pblk, &rqd, rqd.error);
521 #ifdef CONFIG_NVM_DEBUG
522 atomic_long_add(*secs_to_gc, &pblk->sync_reads);
523 atomic_long_add(*secs_to_gc, &pblk->recov_gc_reads);
524 atomic_long_sub(*secs_to_gc, &pblk->inflight_reads);
529 nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);
534 nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);