1 /*
2 * Copyright (C) 2016 CNEX Labs
3 * Initial release: Javier Gonzalez <javier@cnexlabs.com>
4 * Matias Bjorling <matias@cnexlabs.com>
5 *
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.
9 *
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.
14 *
15 * pblk-read.c - pblk's read path
16 */
17
18 #include "pblk.h"
19
20 /*
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
25 * issued.
26 */
pblk_read_from_cache(struct pblk * pblk,struct bio * bio,sector_t lba,struct ppa_addr ppa,int bio_iter,bool advanced_bio)27 static int pblk_read_from_cache(struct pblk *pblk, struct bio *bio,
28 sector_t lba, struct ppa_addr ppa,
29 int bio_iter, bool advanced_bio)
30 {
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));
35 #endif
36
37 return pblk_rb_copy_to_bio(&pblk->rwb, bio, lba, ppa,
38 bio_iter, advanced_bio);
39 }
40
pblk_read_ppalist_rq(struct pblk * pblk,struct nvm_rq * rqd,unsigned long * read_bitmap)41 static void pblk_read_ppalist_rq(struct pblk *pblk, struct nvm_rq *rqd,
42 unsigned long *read_bitmap)
43 {
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;
48 bool advanced_bio = false;
49 int i, j = 0;
50
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");
54 return;
55 }
56
57 pblk_lookup_l2p_seq(pblk, ppas, blba, nr_secs);
58
59 for (i = 0; i < nr_secs; i++) {
60 struct ppa_addr p = ppas[i];
61 sector_t lba = blba + i;
62
63 retry:
64 if (pblk_ppa_empty(p)) {
65 WARN_ON(test_and_set_bit(i, read_bitmap));
66
67 if (unlikely(!advanced_bio)) {
68 bio_advance(bio, (i) * PBLK_EXPOSED_PAGE_SIZE);
69 advanced_bio = true;
70 }
71
72 goto next;
73 }
74
75 /* Try to read from write buffer. The address is later checked
76 * on the write buffer to prevent retrieving overwritten data.
77 */
78 if (pblk_addr_in_cache(p)) {
79 if (!pblk_read_from_cache(pblk, bio, lba, p, i,
80 advanced_bio)) {
81 pblk_lookup_l2p_seq(pblk, &p, lba, 1);
82 goto retry;
83 }
84 WARN_ON(test_and_set_bit(i, read_bitmap));
85 advanced_bio = true;
86 #ifdef CONFIG_NVM_DEBUG
87 atomic_long_inc(&pblk->cache_reads);
88 #endif
89 } else {
90 /* Read from media non-cached sectors */
91 rqd->ppa_list[j++] = p;
92 }
93
94 next:
95 if (advanced_bio)
96 bio_advance(bio, PBLK_EXPOSED_PAGE_SIZE);
97 }
98
99 if (pblk_io_aligned(pblk, nr_secs))
100 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
101 else
102 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
103
104 #ifdef CONFIG_NVM_DEBUG
105 atomic_long_add(nr_secs, &pblk->inflight_reads);
106 #endif
107 }
108
pblk_submit_read_io(struct pblk * pblk,struct nvm_rq * rqd)109 static int pblk_submit_read_io(struct pblk *pblk, struct nvm_rq *rqd)
110 {
111 int err;
112
113 err = pblk_submit_io(pblk, rqd);
114 if (err)
115 return NVM_IO_ERR;
116
117 return NVM_IO_OK;
118 }
119
pblk_end_io_read(struct nvm_rq * rqd)120 static void pblk_end_io_read(struct nvm_rq *rqd)
121 {
122 struct pblk *pblk = rqd->private;
123 struct nvm_tgt_dev *dev = pblk->dev;
124 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
125 struct bio *bio = rqd->bio;
126
127 if (rqd->error)
128 pblk_log_read_err(pblk, rqd);
129 #ifdef CONFIG_NVM_DEBUG
130 else
131 WARN_ONCE(bio->bi_status, "pblk: corrupted read error\n");
132 #endif
133
134 nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
135
136 bio_put(bio);
137 if (r_ctx->private) {
138 struct bio *orig_bio = r_ctx->private;
139
140 #ifdef CONFIG_NVM_DEBUG
141 WARN_ONCE(orig_bio->bi_status, "pblk: corrupted read bio\n");
142 #endif
143 bio_endio(orig_bio);
144 bio_put(orig_bio);
145 }
146
147 #ifdef CONFIG_NVM_DEBUG
148 atomic_long_add(rqd->nr_ppas, &pblk->sync_reads);
149 atomic_long_sub(rqd->nr_ppas, &pblk->inflight_reads);
150 #endif
151
152 pblk_free_rqd(pblk, rqd, READ);
153 atomic_dec(&pblk->inflight_io);
154 }
155
pblk_fill_partial_read_bio(struct pblk * pblk,struct nvm_rq * rqd,unsigned int bio_init_idx,unsigned long * read_bitmap)156 static int pblk_fill_partial_read_bio(struct pblk *pblk, struct nvm_rq *rqd,
157 unsigned int bio_init_idx,
158 unsigned long *read_bitmap)
159 {
160 struct bio *new_bio, *bio = rqd->bio;
161 struct bio_vec src_bv, dst_bv;
162 void *ppa_ptr = NULL;
163 void *src_p, *dst_p;
164 dma_addr_t dma_ppa_list = 0;
165 int nr_secs = rqd->nr_ppas;
166 int nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
167 int i, ret, hole;
168 DECLARE_COMPLETION_ONSTACK(wait);
169
170 new_bio = bio_alloc(GFP_KERNEL, nr_holes);
171 if (!new_bio) {
172 pr_err("pblk: could not alloc read bio\n");
173 return NVM_IO_ERR;
174 }
175
176 if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
177 goto err;
178
179 if (nr_holes != new_bio->bi_vcnt) {
180 pr_err("pblk: malformed bio\n");
181 goto err;
182 }
183
184 new_bio->bi_iter.bi_sector = 0; /* internal bio */
185 bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
186 new_bio->bi_private = &wait;
187 new_bio->bi_end_io = pblk_end_bio_sync;
188
189 rqd->bio = new_bio;
190 rqd->nr_ppas = nr_holes;
191 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
192 rqd->end_io = NULL;
193
194 if (unlikely(nr_secs > 1 && nr_holes == 1)) {
195 ppa_ptr = rqd->ppa_list;
196 dma_ppa_list = rqd->dma_ppa_list;
197 rqd->ppa_addr = rqd->ppa_list[0];
198 }
199
200 ret = pblk_submit_read_io(pblk, rqd);
201 if (ret) {
202 bio_put(rqd->bio);
203 pr_err("pblk: read IO submission failed\n");
204 goto err;
205 }
206
207 if (!wait_for_completion_io_timeout(&wait,
208 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
209 pr_err("pblk: partial read I/O timed out\n");
210 }
211
212 if (rqd->error) {
213 atomic_long_inc(&pblk->read_failed);
214 #ifdef CONFIG_NVM_DEBUG
215 pblk_print_failed_rqd(pblk, rqd, rqd->error);
216 #endif
217 }
218
219 if (unlikely(nr_secs > 1 && nr_holes == 1)) {
220 rqd->ppa_list = ppa_ptr;
221 rqd->dma_ppa_list = dma_ppa_list;
222 }
223
224 /* Fill the holes in the original bio */
225 i = 0;
226 hole = find_first_zero_bit(read_bitmap, nr_secs);
227 do {
228 src_bv = new_bio->bi_io_vec[i++];
229 dst_bv = bio->bi_io_vec[bio_init_idx + hole];
230
231 src_p = kmap_atomic(src_bv.bv_page);
232 dst_p = kmap_atomic(dst_bv.bv_page);
233
234 memcpy(dst_p + dst_bv.bv_offset,
235 src_p + src_bv.bv_offset,
236 PBLK_EXPOSED_PAGE_SIZE);
237
238 kunmap_atomic(src_p);
239 kunmap_atomic(dst_p);
240
241 mempool_free(src_bv.bv_page, pblk->page_bio_pool);
242
243 hole = find_next_zero_bit(read_bitmap, nr_secs, hole + 1);
244 } while (hole < nr_secs);
245
246 bio_put(new_bio);
247
248 /* Complete the original bio and associated request */
249 rqd->bio = bio;
250 rqd->nr_ppas = nr_secs;
251 rqd->private = pblk;
252
253 bio_endio(bio);
254 pblk_end_io_read(rqd);
255 return NVM_IO_OK;
256
257 err:
258 /* Free allocated pages in new bio */
259 pblk_bio_free_pages(pblk, bio, 0, new_bio->bi_vcnt);
260 rqd->private = pblk;
261 pblk_end_io_read(rqd);
262 return NVM_IO_ERR;
263 }
264
pblk_read_rq(struct pblk * pblk,struct nvm_rq * rqd,unsigned long * read_bitmap)265 static void pblk_read_rq(struct pblk *pblk, struct nvm_rq *rqd,
266 unsigned long *read_bitmap)
267 {
268 struct bio *bio = rqd->bio;
269 struct ppa_addr ppa;
270 sector_t lba = pblk_get_lba(bio);
271
272 /* logic error: lba out-of-bounds. Ignore read request */
273 if (lba >= pblk->rl.nr_secs) {
274 WARN(1, "pblk: read lba out of bounds\n");
275 return;
276 }
277
278 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
279
280 #ifdef CONFIG_NVM_DEBUG
281 atomic_long_inc(&pblk->inflight_reads);
282 #endif
283
284 retry:
285 if (pblk_ppa_empty(ppa)) {
286 WARN_ON(test_and_set_bit(0, read_bitmap));
287 return;
288 }
289
290 /* Try to read from write buffer. The address is later checked on the
291 * write buffer to prevent retrieving overwritten data.
292 */
293 if (pblk_addr_in_cache(ppa)) {
294 if (!pblk_read_from_cache(pblk, bio, lba, ppa, 0, 1)) {
295 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
296 goto retry;
297 }
298 WARN_ON(test_and_set_bit(0, read_bitmap));
299 #ifdef CONFIG_NVM_DEBUG
300 atomic_long_inc(&pblk->cache_reads);
301 #endif
302 } else {
303 rqd->ppa_addr = ppa;
304 }
305
306 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
307 }
308
pblk_submit_read(struct pblk * pblk,struct bio * bio)309 int pblk_submit_read(struct pblk *pblk, struct bio *bio)
310 {
311 struct nvm_tgt_dev *dev = pblk->dev;
312 unsigned int nr_secs = pblk_get_secs(bio);
313 struct nvm_rq *rqd;
314 unsigned long read_bitmap; /* Max 64 ppas per request */
315 unsigned int bio_init_idx;
316 int ret = NVM_IO_ERR;
317
318 if (nr_secs > PBLK_MAX_REQ_ADDRS)
319 return NVM_IO_ERR;
320
321 bitmap_zero(&read_bitmap, nr_secs);
322
323 rqd = pblk_alloc_rqd(pblk, READ);
324 if (IS_ERR(rqd)) {
325 pr_err_ratelimited("pblk: not able to alloc rqd");
326 return NVM_IO_ERR;
327 }
328
329 rqd->opcode = NVM_OP_PREAD;
330 rqd->bio = bio;
331 rqd->nr_ppas = nr_secs;
332 rqd->private = pblk;
333 rqd->end_io = pblk_end_io_read;
334
335 /* Save the index for this bio's start. This is needed in case
336 * we need to fill a partial read.
337 */
338 bio_init_idx = pblk_get_bi_idx(bio);
339
340 rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
341 &rqd->dma_meta_list);
342 if (!rqd->meta_list) {
343 pr_err("pblk: not able to allocate ppa list\n");
344 goto fail_rqd_free;
345 }
346
347 if (nr_secs > 1) {
348 rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size;
349 rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size;
350
351 pblk_read_ppalist_rq(pblk, rqd, &read_bitmap);
352 } else {
353 pblk_read_rq(pblk, rqd, &read_bitmap);
354 }
355
356 bio_get(bio);
357 if (bitmap_full(&read_bitmap, nr_secs)) {
358 bio_endio(bio);
359 atomic_inc(&pblk->inflight_io);
360 pblk_end_io_read(rqd);
361 return NVM_IO_OK;
362 }
363
364 /* All sectors are to be read from the device */
365 if (bitmap_empty(&read_bitmap, rqd->nr_ppas)) {
366 struct bio *int_bio = NULL;
367 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
368
369 /* Clone read bio to deal with read errors internally */
370 int_bio = bio_clone_fast(bio, GFP_KERNEL, pblk_bio_set);
371 if (!int_bio) {
372 pr_err("pblk: could not clone read bio\n");
373 return NVM_IO_ERR;
374 }
375
376 rqd->bio = int_bio;
377 r_ctx->private = bio;
378
379 ret = pblk_submit_read_io(pblk, rqd);
380 if (ret) {
381 pr_err("pblk: read IO submission failed\n");
382 if (int_bio)
383 bio_put(int_bio);
384 return ret;
385 }
386
387 return NVM_IO_OK;
388 }
389
390 /* The read bio request could be partially filled by the write buffer,
391 * but there are some holes that need to be read from the drive.
392 */
393 ret = pblk_fill_partial_read_bio(pblk, rqd, bio_init_idx, &read_bitmap);
394 if (ret) {
395 pr_err("pblk: failed to perform partial read\n");
396 return ret;
397 }
398
399 return NVM_IO_OK;
400
401 fail_rqd_free:
402 pblk_free_rqd(pblk, rqd, READ);
403 return ret;
404 }
405
read_ppalist_rq_gc(struct pblk * pblk,struct nvm_rq * rqd,struct pblk_line * line,u64 * lba_list,unsigned int nr_secs)406 static int read_ppalist_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
407 struct pblk_line *line, u64 *lba_list,
408 unsigned int nr_secs)
409 {
410 struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
411 int valid_secs = 0;
412 int i;
413
414 pblk_lookup_l2p_rand(pblk, ppas, lba_list, nr_secs);
415
416 for (i = 0; i < nr_secs; i++) {
417 if (pblk_addr_in_cache(ppas[i]) || ppas[i].g.blk != line->id ||
418 pblk_ppa_empty(ppas[i])) {
419 lba_list[i] = ADDR_EMPTY;
420 continue;
421 }
422
423 rqd->ppa_list[valid_secs++] = ppas[i];
424 }
425
426 #ifdef CONFIG_NVM_DEBUG
427 atomic_long_add(valid_secs, &pblk->inflight_reads);
428 #endif
429 return valid_secs;
430 }
431
read_rq_gc(struct pblk * pblk,struct nvm_rq * rqd,struct pblk_line * line,sector_t lba)432 static int read_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
433 struct pblk_line *line, sector_t lba)
434 {
435 struct ppa_addr ppa;
436 int valid_secs = 0;
437
438 if (lba == ADDR_EMPTY)
439 goto out;
440
441 /* logic error: lba out-of-bounds */
442 if (lba >= pblk->rl.nr_secs) {
443 WARN(1, "pblk: read lba out of bounds\n");
444 goto out;
445 }
446
447 spin_lock(&pblk->trans_lock);
448 ppa = pblk_trans_map_get(pblk, lba);
449 spin_unlock(&pblk->trans_lock);
450
451 /* Ignore updated values until the moment */
452 if (pblk_addr_in_cache(ppa) || ppa.g.blk != line->id ||
453 pblk_ppa_empty(ppa))
454 goto out;
455
456 rqd->ppa_addr = ppa;
457 valid_secs = 1;
458
459 #ifdef CONFIG_NVM_DEBUG
460 atomic_long_inc(&pblk->inflight_reads);
461 #endif
462
463 out:
464 return valid_secs;
465 }
466
pblk_submit_read_gc(struct pblk * pblk,u64 * lba_list,void * data,unsigned int nr_secs,unsigned int * secs_to_gc,struct pblk_line * line)467 int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
468 unsigned int nr_secs, unsigned int *secs_to_gc,
469 struct pblk_line *line)
470 {
471 struct nvm_tgt_dev *dev = pblk->dev;
472 struct nvm_geo *geo = &dev->geo;
473 struct bio *bio;
474 struct nvm_rq rqd;
475 int ret, data_len;
476 DECLARE_COMPLETION_ONSTACK(wait);
477
478 memset(&rqd, 0, sizeof(struct nvm_rq));
479
480 rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
481 &rqd.dma_meta_list);
482 if (!rqd.meta_list)
483 return NVM_IO_ERR;
484
485 if (nr_secs > 1) {
486 rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
487 rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
488
489 *secs_to_gc = read_ppalist_rq_gc(pblk, &rqd, line, lba_list,
490 nr_secs);
491 if (*secs_to_gc == 1)
492 rqd.ppa_addr = rqd.ppa_list[0];
493 } else {
494 *secs_to_gc = read_rq_gc(pblk, &rqd, line, lba_list[0]);
495 }
496
497 if (!(*secs_to_gc))
498 goto out;
499
500 data_len = (*secs_to_gc) * geo->sec_size;
501 bio = pblk_bio_map_addr(pblk, data, *secs_to_gc, data_len,
502 PBLK_VMALLOC_META, GFP_KERNEL);
503 if (IS_ERR(bio)) {
504 pr_err("pblk: could not allocate GC bio (%lu)\n", PTR_ERR(bio));
505 goto err_free_dma;
506 }
507
508 bio->bi_iter.bi_sector = 0; /* internal bio */
509 bio_set_op_attrs(bio, REQ_OP_READ, 0);
510
511 rqd.opcode = NVM_OP_PREAD;
512 rqd.end_io = pblk_end_io_sync;
513 rqd.private = &wait;
514 rqd.nr_ppas = *secs_to_gc;
515 rqd.flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
516 rqd.bio = bio;
517
518 ret = pblk_submit_read_io(pblk, &rqd);
519 if (ret) {
520 bio_endio(bio);
521 pr_err("pblk: GC read request failed\n");
522 goto err_free_bio;
523 }
524
525 if (!wait_for_completion_io_timeout(&wait,
526 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
527 pr_err("pblk: GC read I/O timed out\n");
528 }
529 atomic_dec(&pblk->inflight_io);
530
531 if (rqd.error) {
532 atomic_long_inc(&pblk->read_failed_gc);
533 #ifdef CONFIG_NVM_DEBUG
534 pblk_print_failed_rqd(pblk, &rqd, rqd.error);
535 #endif
536 }
537
538 #ifdef CONFIG_NVM_DEBUG
539 atomic_long_add(*secs_to_gc, &pblk->sync_reads);
540 atomic_long_add(*secs_to_gc, &pblk->recov_gc_reads);
541 atomic_long_sub(*secs_to_gc, &pblk->inflight_reads);
542 #endif
543
544 bio_put(bio);
545 out:
546 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
547 return NVM_IO_OK;
548
549 err_free_bio:
550 bio_put(bio);
551 err_free_dma:
552 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
553 return NVM_IO_ERR;
554 }
555