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 if (pblk_io_aligned(pblk, nr_secs))
92 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
94 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
96 #ifdef CONFIG_NVM_DEBUG
97 atomic_long_add(nr_secs, &pblk->inflight_reads);
101 static int pblk_submit_read_io(struct pblk *pblk, struct nvm_rq *rqd)
105 err = pblk_submit_io(pblk, rqd);
112 static void pblk_end_io_read(struct nvm_rq *rqd)
114 struct pblk *pblk = rqd->private;
115 struct nvm_tgt_dev *dev = pblk->dev;
116 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
117 struct bio *bio = rqd->bio;
120 pblk_log_read_err(pblk, rqd);
121 #ifdef CONFIG_NVM_DEBUG
123 WARN_ONCE(bio->bi_status, "pblk: corrupted read error\n");
126 nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
129 if (r_ctx->private) {
130 struct bio *orig_bio = r_ctx->private;
132 #ifdef CONFIG_NVM_DEBUG
133 WARN_ONCE(orig_bio->bi_status, "pblk: corrupted read bio\n");
139 #ifdef CONFIG_NVM_DEBUG
140 atomic_long_add(rqd->nr_ppas, &pblk->sync_reads);
141 atomic_long_sub(rqd->nr_ppas, &pblk->inflight_reads);
144 pblk_free_rqd(pblk, rqd, READ);
145 atomic_dec(&pblk->inflight_io);
148 static int pblk_fill_partial_read_bio(struct pblk *pblk, struct nvm_rq *rqd,
149 unsigned int bio_init_idx,
150 unsigned long *read_bitmap)
152 struct bio *new_bio, *bio = rqd->bio;
153 struct bio_vec src_bv, dst_bv;
154 void *ppa_ptr = NULL;
156 dma_addr_t dma_ppa_list = 0;
157 int nr_secs = rqd->nr_ppas;
158 int nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
160 DECLARE_COMPLETION_ONSTACK(wait);
162 new_bio = bio_alloc(GFP_KERNEL, nr_holes);
164 pr_err("pblk: could not alloc read bio\n");
168 if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
171 if (nr_holes != new_bio->bi_vcnt) {
172 pr_err("pblk: malformed bio\n");
176 new_bio->bi_iter.bi_sector = 0; /* internal bio */
177 bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
178 new_bio->bi_private = &wait;
179 new_bio->bi_end_io = pblk_end_bio_sync;
182 rqd->nr_ppas = nr_holes;
183 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
186 if (unlikely(nr_secs > 1 && nr_holes == 1)) {
187 ppa_ptr = rqd->ppa_list;
188 dma_ppa_list = rqd->dma_ppa_list;
189 rqd->ppa_addr = rqd->ppa_list[0];
192 ret = pblk_submit_read_io(pblk, rqd);
195 pr_err("pblk: read IO submission failed\n");
199 if (!wait_for_completion_io_timeout(&wait,
200 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
201 pr_err("pblk: partial read I/O timed out\n");
205 atomic_long_inc(&pblk->read_failed);
206 #ifdef CONFIG_NVM_DEBUG
207 pblk_print_failed_rqd(pblk, rqd, rqd->error);
211 if (unlikely(nr_secs > 1 && nr_holes == 1)) {
212 rqd->ppa_list = ppa_ptr;
213 rqd->dma_ppa_list = dma_ppa_list;
216 /* Fill the holes in the original bio */
218 hole = find_first_zero_bit(read_bitmap, nr_secs);
220 src_bv = new_bio->bi_io_vec[i++];
221 dst_bv = bio->bi_io_vec[bio_init_idx + hole];
223 src_p = kmap_atomic(src_bv.bv_page);
224 dst_p = kmap_atomic(dst_bv.bv_page);
226 memcpy(dst_p + dst_bv.bv_offset,
227 src_p + src_bv.bv_offset,
228 PBLK_EXPOSED_PAGE_SIZE);
230 kunmap_atomic(src_p);
231 kunmap_atomic(dst_p);
233 mempool_free(src_bv.bv_page, pblk->page_pool);
235 hole = find_next_zero_bit(read_bitmap, nr_secs, hole + 1);
236 } while (hole < nr_secs);
240 /* Complete the original bio and associated request */
242 rqd->nr_ppas = nr_secs;
246 pblk_end_io_read(rqd);
250 /* Free allocated pages in new bio */
251 pblk_bio_free_pages(pblk, bio, 0, new_bio->bi_vcnt);
253 pblk_end_io_read(rqd);
257 static void pblk_read_rq(struct pblk *pblk, struct nvm_rq *rqd,
258 unsigned long *read_bitmap)
260 struct bio *bio = rqd->bio;
262 sector_t lba = pblk_get_lba(bio);
264 /* logic error: lba out-of-bounds. Ignore read request */
265 if (lba >= pblk->rl.nr_secs) {
266 WARN(1, "pblk: read lba out of bounds\n");
270 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
272 #ifdef CONFIG_NVM_DEBUG
273 atomic_long_inc(&pblk->inflight_reads);
277 if (pblk_ppa_empty(ppa)) {
278 WARN_ON(test_and_set_bit(0, read_bitmap));
282 /* Try to read from write buffer. The address is later checked on the
283 * write buffer to prevent retrieving overwritten data.
285 if (pblk_addr_in_cache(ppa)) {
286 if (!pblk_read_from_cache(pblk, bio, lba, ppa, 0)) {
287 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
290 WARN_ON(test_and_set_bit(0, read_bitmap));
291 #ifdef CONFIG_NVM_DEBUG
292 atomic_long_inc(&pblk->cache_reads);
298 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
301 int pblk_submit_read(struct pblk *pblk, struct bio *bio)
303 struct nvm_tgt_dev *dev = pblk->dev;
304 unsigned int nr_secs = pblk_get_secs(bio);
306 unsigned long read_bitmap; /* Max 64 ppas per request */
307 unsigned int bio_init_idx;
308 int ret = NVM_IO_ERR;
310 if (nr_secs > PBLK_MAX_REQ_ADDRS)
313 bitmap_zero(&read_bitmap, nr_secs);
315 rqd = pblk_alloc_rqd(pblk, READ);
317 pr_err_ratelimited("pblk: not able to alloc rqd");
321 rqd->opcode = NVM_OP_PREAD;
323 rqd->nr_ppas = nr_secs;
325 rqd->end_io = pblk_end_io_read;
327 /* Save the index for this bio's start. This is needed in case
328 * we need to fill a partial read.
330 bio_init_idx = pblk_get_bi_idx(bio);
332 rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
333 &rqd->dma_meta_list);
334 if (!rqd->meta_list) {
335 pr_err("pblk: not able to allocate ppa list\n");
340 rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size;
341 rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size;
343 pblk_read_ppalist_rq(pblk, rqd, &read_bitmap);
345 pblk_read_rq(pblk, rqd, &read_bitmap);
349 if (bitmap_full(&read_bitmap, nr_secs)) {
351 atomic_inc(&pblk->inflight_io);
352 pblk_end_io_read(rqd);
356 /* All sectors are to be read from the device */
357 if (bitmap_empty(&read_bitmap, rqd->nr_ppas)) {
358 struct bio *int_bio = NULL;
359 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
361 /* Clone read bio to deal with read errors internally */
362 int_bio = bio_clone_fast(bio, GFP_KERNEL, pblk_bio_set);
364 pr_err("pblk: could not clone read bio\n");
369 r_ctx->private = bio;
371 ret = pblk_submit_read_io(pblk, rqd);
373 pr_err("pblk: read IO submission failed\n");
382 /* The read bio request could be partially filled by the write buffer,
383 * but there are some holes that need to be read from the drive.
385 ret = pblk_fill_partial_read_bio(pblk, rqd, bio_init_idx, &read_bitmap);
387 pr_err("pblk: failed to perform partial read\n");
394 pblk_free_rqd(pblk, rqd, READ);
398 static int read_ppalist_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
399 struct pblk_line *line, u64 *lba_list,
400 unsigned int nr_secs)
402 struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
406 pblk_lookup_l2p_rand(pblk, ppas, lba_list, nr_secs);
408 for (i = 0; i < nr_secs; i++) {
409 if (pblk_addr_in_cache(ppas[i]) || ppas[i].g.blk != line->id ||
410 pblk_ppa_empty(ppas[i])) {
411 lba_list[i] = ADDR_EMPTY;
415 rqd->ppa_list[valid_secs++] = ppas[i];
418 #ifdef CONFIG_NVM_DEBUG
419 atomic_long_add(valid_secs, &pblk->inflight_reads);
424 static int read_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
425 struct pblk_line *line, sector_t lba)
430 if (lba == ADDR_EMPTY)
433 /* logic error: lba out-of-bounds */
434 if (lba >= pblk->rl.nr_secs) {
435 WARN(1, "pblk: read lba out of bounds\n");
439 spin_lock(&pblk->trans_lock);
440 ppa = pblk_trans_map_get(pblk, lba);
441 spin_unlock(&pblk->trans_lock);
443 /* Ignore updated values until the moment */
444 if (pblk_addr_in_cache(ppa) || ppa.g.blk != line->id ||
451 #ifdef CONFIG_NVM_DEBUG
452 atomic_long_inc(&pblk->inflight_reads);
459 int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
460 unsigned int nr_secs, unsigned int *secs_to_gc,
461 struct pblk_line *line)
463 struct nvm_tgt_dev *dev = pblk->dev;
464 struct nvm_geo *geo = &dev->geo;
468 DECLARE_COMPLETION_ONSTACK(wait);
470 memset(&rqd, 0, sizeof(struct nvm_rq));
472 rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
478 rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
479 rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
481 *secs_to_gc = read_ppalist_rq_gc(pblk, &rqd, line, lba_list,
483 if (*secs_to_gc == 1)
484 rqd.ppa_addr = rqd.ppa_list[0];
486 *secs_to_gc = read_rq_gc(pblk, &rqd, line, lba_list[0]);
492 data_len = (*secs_to_gc) * geo->sec_size;
493 bio = pblk_bio_map_addr(pblk, data, *secs_to_gc, data_len,
494 PBLK_KMALLOC_META, GFP_KERNEL);
496 pr_err("pblk: could not allocate GC bio (%lu)\n", PTR_ERR(bio));
500 bio->bi_iter.bi_sector = 0; /* internal bio */
501 bio_set_op_attrs(bio, REQ_OP_READ, 0);
503 rqd.opcode = NVM_OP_PREAD;
504 rqd.end_io = pblk_end_io_sync;
506 rqd.nr_ppas = *secs_to_gc;
507 rqd.flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
510 ret = pblk_submit_read_io(pblk, &rqd);
513 pr_err("pblk: GC read request failed\n");
517 if (!wait_for_completion_io_timeout(&wait,
518 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
519 pr_err("pblk: GC read I/O timed out\n");
521 atomic_dec(&pblk->inflight_io);
524 atomic_long_inc(&pblk->read_failed_gc);
525 #ifdef CONFIG_NVM_DEBUG
526 pblk_print_failed_rqd(pblk, &rqd, rqd.error);
530 #ifdef CONFIG_NVM_DEBUG
531 atomic_long_add(*secs_to_gc, &pblk->sync_reads);
532 atomic_long_add(*secs_to_gc, &pblk->recov_gc_reads);
533 atomic_long_sub(*secs_to_gc, &pblk->inflight_reads);
537 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
541 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);