]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/lightnvm/pblk-read.c
UBUNTU: Ubuntu-4.15.0-96.97
[mirror_ubuntu-bionic-kernel.git] / drivers / lightnvm / pblk-read.c
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 */
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
41 static void pblk_read_ppalist_rq(struct pblk *pblk, struct nvm_rq *rqd,
42 sector_t blba, unsigned long *read_bitmap)
43 {
44 struct pblk_sec_meta *meta_list = rqd->meta_list;
45 struct bio *bio = rqd->bio;
46 struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
47 int nr_secs = rqd->nr_ppas;
48 bool advanced_bio = false;
49 int i, j = 0;
50
51 pblk_lookup_l2p_seq(pblk, ppas, blba, nr_secs);
52
53 for (i = 0; i < nr_secs; i++) {
54 struct ppa_addr p = ppas[i];
55 sector_t lba = blba + i;
56
57 retry:
58 if (pblk_ppa_empty(p)) {
59 WARN_ON(test_and_set_bit(i, read_bitmap));
60 meta_list[i].lba = cpu_to_le64(ADDR_EMPTY);
61
62 if (unlikely(!advanced_bio)) {
63 bio_advance(bio, (i) * PBLK_EXPOSED_PAGE_SIZE);
64 advanced_bio = true;
65 }
66
67 goto next;
68 }
69
70 /* Try to read from write buffer. The address is later checked
71 * on the write buffer to prevent retrieving overwritten data.
72 */
73 if (pblk_addr_in_cache(p)) {
74 if (!pblk_read_from_cache(pblk, bio, lba, p, i,
75 advanced_bio)) {
76 pblk_lookup_l2p_seq(pblk, &p, lba, 1);
77 goto retry;
78 }
79 WARN_ON(test_and_set_bit(i, read_bitmap));
80 meta_list[i].lba = cpu_to_le64(lba);
81 advanced_bio = true;
82 #ifdef CONFIG_NVM_DEBUG
83 atomic_long_inc(&pblk->cache_reads);
84 #endif
85 } else {
86 /* Read from media non-cached sectors */
87 rqd->ppa_list[j++] = p;
88 }
89
90 next:
91 if (advanced_bio)
92 bio_advance(bio, PBLK_EXPOSED_PAGE_SIZE);
93 }
94
95 if (pblk_io_aligned(pblk, nr_secs))
96 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
97 else
98 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
99
100 #ifdef CONFIG_NVM_DEBUG
101 atomic_long_add(nr_secs, &pblk->inflight_reads);
102 #endif
103 }
104
105 static int pblk_submit_read_io(struct pblk *pblk, struct nvm_rq *rqd)
106 {
107 int err;
108
109 err = pblk_submit_io(pblk, rqd);
110 if (err)
111 return NVM_IO_ERR;
112
113 return NVM_IO_OK;
114 }
115
116 static void pblk_read_check(struct pblk *pblk, struct nvm_rq *rqd,
117 sector_t blba)
118 {
119 struct pblk_sec_meta *meta_list = rqd->meta_list;
120 int nr_lbas = rqd->nr_ppas;
121 int i;
122
123 for (i = 0; i < nr_lbas; i++) {
124 u64 lba = le64_to_cpu(meta_list[i].lba);
125
126 if (lba == ADDR_EMPTY)
127 continue;
128
129 WARN(lba != blba + i, "pblk: corrupted read LBA\n");
130 }
131 }
132
133 static void pblk_read_put_rqd_kref(struct pblk *pblk, struct nvm_rq *rqd)
134 {
135 struct ppa_addr *ppa_list;
136 int i;
137
138 ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
139
140 for (i = 0; i < rqd->nr_ppas; i++) {
141 struct ppa_addr ppa = ppa_list[i];
142 struct pblk_line *line;
143
144 line = &pblk->lines[pblk_dev_ppa_to_line(ppa)];
145 kref_put(&line->ref, pblk_line_put_wq);
146 }
147 }
148
149 static void pblk_end_user_read(struct bio *bio)
150 {
151 #ifdef CONFIG_NVM_DEBUG
152 WARN_ONCE(bio->bi_status, "pblk: corrupted read bio\n");
153 #endif
154 bio_endio(bio);
155 bio_put(bio);
156 }
157
158 static void __pblk_end_io_read(struct pblk *pblk, struct nvm_rq *rqd,
159 bool put_line)
160 {
161 struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
162 struct bio *bio = rqd->bio;
163
164 if (rqd->error)
165 pblk_log_read_err(pblk, rqd);
166 #ifdef CONFIG_NVM_DEBUG
167 else
168 WARN_ONCE(bio->bi_status, "pblk: corrupted read error\n");
169 #endif
170
171 pblk_read_check(pblk, rqd, r_ctx->lba);
172
173 bio_put(bio);
174 if (r_ctx->private)
175 pblk_end_user_read((struct bio *)r_ctx->private);
176
177 if (put_line)
178 pblk_read_put_rqd_kref(pblk, rqd);
179
180 #ifdef CONFIG_NVM_DEBUG
181 atomic_long_add(rqd->nr_ppas, &pblk->sync_reads);
182 atomic_long_sub(rqd->nr_ppas, &pblk->inflight_reads);
183 #endif
184
185 pblk_free_rqd(pblk, rqd, PBLK_READ);
186 atomic_dec(&pblk->inflight_io);
187 }
188
189 static void pblk_end_io_read(struct nvm_rq *rqd)
190 {
191 struct pblk *pblk = rqd->private;
192
193 __pblk_end_io_read(pblk, rqd, true);
194 }
195
196 static int pblk_fill_partial_read_bio(struct pblk *pblk, struct nvm_rq *rqd,
197 unsigned int bio_init_idx,
198 unsigned long *read_bitmap)
199 {
200 struct bio *new_bio, *bio = rqd->bio;
201 struct pblk_sec_meta *meta_list = rqd->meta_list;
202 struct bio_vec src_bv, dst_bv;
203 void *ppa_ptr = NULL;
204 void *src_p, *dst_p;
205 dma_addr_t dma_ppa_list = 0;
206 __le64 *lba_list_mem, *lba_list_media;
207 int nr_secs = rqd->nr_ppas;
208 int nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
209 int i, ret, hole;
210
211 /* Re-use allocated memory for intermediate lbas */
212 lba_list_mem = (((void *)rqd->ppa_list) + pblk_dma_ppa_size);
213 lba_list_media = (((void *)rqd->ppa_list) + 2 * pblk_dma_ppa_size);
214
215 new_bio = bio_alloc(GFP_KERNEL, nr_holes);
216
217 if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
218 goto err;
219
220 if (nr_holes != new_bio->bi_vcnt) {
221 pr_err("pblk: malformed bio\n");
222 goto err;
223 }
224
225 for (i = 0; i < nr_secs; i++)
226 lba_list_mem[i] = meta_list[i].lba;
227
228 new_bio->bi_iter.bi_sector = 0; /* internal bio */
229 bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
230
231 rqd->bio = new_bio;
232 rqd->nr_ppas = nr_holes;
233 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
234
235 if (unlikely(nr_holes == 1)) {
236 ppa_ptr = rqd->ppa_list;
237 dma_ppa_list = rqd->dma_ppa_list;
238 rqd->ppa_addr = rqd->ppa_list[0];
239 }
240
241 ret = pblk_submit_io_sync(pblk, rqd);
242 if (ret) {
243 bio_put(rqd->bio);
244 pr_err("pblk: sync read IO submission failed\n");
245 goto err;
246 }
247
248 if (rqd->error) {
249 atomic_long_inc(&pblk->read_failed);
250 #ifdef CONFIG_NVM_DEBUG
251 pblk_print_failed_rqd(pblk, rqd, rqd->error);
252 #endif
253 }
254
255 if (unlikely(nr_holes == 1)) {
256 struct ppa_addr ppa;
257
258 ppa = rqd->ppa_addr;
259 rqd->ppa_list = ppa_ptr;
260 rqd->dma_ppa_list = dma_ppa_list;
261 rqd->ppa_list[0] = ppa;
262 }
263
264 for (i = 0; i < nr_secs; i++) {
265 lba_list_media[i] = meta_list[i].lba;
266 meta_list[i].lba = lba_list_mem[i];
267 }
268
269 /* Fill the holes in the original bio */
270 i = 0;
271 hole = find_first_zero_bit(read_bitmap, nr_secs);
272 do {
273 int line_id = pblk_dev_ppa_to_line(rqd->ppa_list[i]);
274 struct pblk_line *line = &pblk->lines[line_id];
275
276 kref_put(&line->ref, pblk_line_put);
277
278 meta_list[hole].lba = lba_list_media[i];
279
280 src_bv = new_bio->bi_io_vec[i++];
281 dst_bv = bio->bi_io_vec[bio_init_idx + hole];
282
283 src_p = kmap_atomic(src_bv.bv_page);
284 dst_p = kmap_atomic(dst_bv.bv_page);
285
286 memcpy(dst_p + dst_bv.bv_offset,
287 src_p + src_bv.bv_offset,
288 PBLK_EXPOSED_PAGE_SIZE);
289
290 kunmap_atomic(src_p);
291 kunmap_atomic(dst_p);
292
293 mempool_free(src_bv.bv_page, pblk->page_bio_pool);
294
295 hole = find_next_zero_bit(read_bitmap, nr_secs, hole + 1);
296 } while (hole < nr_secs);
297
298 bio_put(new_bio);
299
300 /* Complete the original bio and associated request */
301 bio_endio(bio);
302 rqd->bio = bio;
303 rqd->nr_ppas = nr_secs;
304
305 __pblk_end_io_read(pblk, rqd, false);
306 return NVM_IO_OK;
307
308 err:
309 /* Free allocated pages in new bio */
310 pblk_bio_free_pages(pblk, bio, 0, new_bio->bi_vcnt);
311 __pblk_end_io_read(pblk, rqd, false);
312 return NVM_IO_ERR;
313 }
314
315 static void pblk_read_rq(struct pblk *pblk, struct nvm_rq *rqd,
316 sector_t lba, unsigned long *read_bitmap)
317 {
318 struct pblk_sec_meta *meta_list = rqd->meta_list;
319 struct bio *bio = rqd->bio;
320 struct ppa_addr ppa;
321
322 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
323
324 #ifdef CONFIG_NVM_DEBUG
325 atomic_long_inc(&pblk->inflight_reads);
326 #endif
327
328 retry:
329 if (pblk_ppa_empty(ppa)) {
330 WARN_ON(test_and_set_bit(0, read_bitmap));
331 meta_list[0].lba = cpu_to_le64(ADDR_EMPTY);
332 return;
333 }
334
335 /* Try to read from write buffer. The address is later checked on the
336 * write buffer to prevent retrieving overwritten data.
337 */
338 if (pblk_addr_in_cache(ppa)) {
339 if (!pblk_read_from_cache(pblk, bio, lba, ppa, 0, 1)) {
340 pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
341 goto retry;
342 }
343
344 WARN_ON(test_and_set_bit(0, read_bitmap));
345 meta_list[0].lba = cpu_to_le64(lba);
346
347 #ifdef CONFIG_NVM_DEBUG
348 atomic_long_inc(&pblk->cache_reads);
349 #endif
350 } else {
351 rqd->ppa_addr = ppa;
352 }
353
354 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
355 }
356
357 int pblk_submit_read(struct pblk *pblk, struct bio *bio)
358 {
359 struct nvm_tgt_dev *dev = pblk->dev;
360 sector_t blba = pblk_get_lba(bio);
361 unsigned int nr_secs = pblk_get_secs(bio);
362 struct pblk_g_ctx *r_ctx;
363 struct nvm_rq *rqd;
364 unsigned int bio_init_idx;
365 unsigned long read_bitmap; /* Max 64 ppas per request */
366 int ret = NVM_IO_ERR;
367
368 /* logic error: lba out-of-bounds. Ignore read request */
369 if (blba >= pblk->rl.nr_secs || nr_secs > PBLK_MAX_REQ_ADDRS) {
370 WARN(1, "pblk: read lba out of bounds (lba:%llu, nr:%d)\n",
371 (unsigned long long)blba, nr_secs);
372 return NVM_IO_ERR;
373 }
374
375 bitmap_zero(&read_bitmap, nr_secs);
376
377 rqd = pblk_alloc_rqd(pblk, PBLK_READ);
378
379 rqd->opcode = NVM_OP_PREAD;
380 rqd->bio = bio;
381 rqd->nr_ppas = nr_secs;
382 rqd->private = pblk;
383 rqd->end_io = pblk_end_io_read;
384
385 r_ctx = nvm_rq_to_pdu(rqd);
386 r_ctx->lba = blba;
387
388 /* Save the index for this bio's start. This is needed in case
389 * we need to fill a partial read.
390 */
391 bio_init_idx = pblk_get_bi_idx(bio);
392
393 rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
394 &rqd->dma_meta_list);
395 if (!rqd->meta_list) {
396 pr_err("pblk: not able to allocate ppa list\n");
397 goto fail_rqd_free;
398 }
399
400 if (nr_secs > 1) {
401 rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size;
402 rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size;
403
404 pblk_read_ppalist_rq(pblk, rqd, blba, &read_bitmap);
405 } else {
406 pblk_read_rq(pblk, rqd, blba, &read_bitmap);
407 }
408
409 bio_get(bio);
410 if (bitmap_full(&read_bitmap, nr_secs)) {
411 bio_endio(bio);
412 atomic_inc(&pblk->inflight_io);
413 __pblk_end_io_read(pblk, rqd, false);
414 return NVM_IO_OK;
415 }
416
417 /* All sectors are to be read from the device */
418 if (bitmap_empty(&read_bitmap, rqd->nr_ppas)) {
419 struct bio *int_bio = NULL;
420
421 /* Clone read bio to deal with read errors internally */
422 int_bio = bio_clone_fast(bio, GFP_KERNEL, pblk_bio_set);
423 if (!int_bio) {
424 pr_err("pblk: could not clone read bio\n");
425 return NVM_IO_ERR;
426 }
427
428 rqd->bio = int_bio;
429 r_ctx->private = bio;
430
431 ret = pblk_submit_read_io(pblk, rqd);
432 if (ret) {
433 pr_err("pblk: read IO submission failed\n");
434 if (int_bio)
435 bio_put(int_bio);
436 return ret;
437 }
438
439 return NVM_IO_OK;
440 }
441
442 /* The read bio request could be partially filled by the write buffer,
443 * but there are some holes that need to be read from the drive.
444 */
445 ret = pblk_fill_partial_read_bio(pblk, rqd, bio_init_idx, &read_bitmap);
446 if (ret) {
447 pr_err("pblk: failed to perform partial read\n");
448 return ret;
449 }
450
451 return NVM_IO_OK;
452
453 fail_rqd_free:
454 pblk_free_rqd(pblk, rqd, PBLK_READ);
455 return ret;
456 }
457
458 static int read_ppalist_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
459 struct pblk_line *line, u64 *lba_list,
460 u64 *paddr_list_gc, unsigned int nr_secs)
461 {
462 struct ppa_addr ppa_list_l2p[PBLK_MAX_REQ_ADDRS];
463 struct ppa_addr ppa_gc;
464 int valid_secs = 0;
465 int i;
466
467 pblk_lookup_l2p_rand(pblk, ppa_list_l2p, lba_list, nr_secs);
468
469 for (i = 0; i < nr_secs; i++) {
470 if (lba_list[i] == ADDR_EMPTY)
471 continue;
472
473 ppa_gc = addr_to_gen_ppa(pblk, paddr_list_gc[i], line->id);
474 if (!pblk_ppa_comp(ppa_list_l2p[i], ppa_gc)) {
475 paddr_list_gc[i] = lba_list[i] = ADDR_EMPTY;
476 continue;
477 }
478
479 rqd->ppa_list[valid_secs++] = ppa_list_l2p[i];
480 }
481
482 #ifdef CONFIG_NVM_DEBUG
483 atomic_long_add(valid_secs, &pblk->inflight_reads);
484 #endif
485
486 return valid_secs;
487 }
488
489 static int read_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
490 struct pblk_line *line, sector_t lba,
491 u64 paddr_gc)
492 {
493 struct ppa_addr ppa_l2p, ppa_gc;
494 int valid_secs = 0;
495
496 if (lba == ADDR_EMPTY)
497 goto out;
498
499 /* logic error: lba out-of-bounds */
500 if (lba >= pblk->rl.nr_secs) {
501 WARN(1, "pblk: read lba out of bounds\n");
502 goto out;
503 }
504
505 spin_lock(&pblk->trans_lock);
506 ppa_l2p = pblk_trans_map_get(pblk, lba);
507 spin_unlock(&pblk->trans_lock);
508
509 ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, line->id);
510 if (!pblk_ppa_comp(ppa_l2p, ppa_gc))
511 goto out;
512
513 rqd->ppa_addr = ppa_l2p;
514 valid_secs = 1;
515
516 #ifdef CONFIG_NVM_DEBUG
517 atomic_long_inc(&pblk->inflight_reads);
518 #endif
519
520 out:
521 return valid_secs;
522 }
523
524 int pblk_submit_read_gc(struct pblk *pblk, struct pblk_gc_rq *gc_rq)
525 {
526 struct nvm_tgt_dev *dev = pblk->dev;
527 struct nvm_geo *geo = &dev->geo;
528 struct bio *bio;
529 struct nvm_rq rqd;
530 int data_len;
531 int ret = NVM_IO_OK;
532
533 memset(&rqd, 0, sizeof(struct nvm_rq));
534
535 rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
536 &rqd.dma_meta_list);
537 if (!rqd.meta_list)
538 return -ENOMEM;
539
540 if (gc_rq->nr_secs > 1) {
541 rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
542 rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
543
544 gc_rq->secs_to_gc = read_ppalist_rq_gc(pblk, &rqd, gc_rq->line,
545 gc_rq->lba_list,
546 gc_rq->paddr_list,
547 gc_rq->nr_secs);
548 if (gc_rq->secs_to_gc == 1)
549 rqd.ppa_addr = rqd.ppa_list[0];
550 } else {
551 gc_rq->secs_to_gc = read_rq_gc(pblk, &rqd, gc_rq->line,
552 gc_rq->lba_list[0],
553 gc_rq->paddr_list[0]);
554 }
555
556 if (!(gc_rq->secs_to_gc))
557 goto out;
558
559 data_len = (gc_rq->secs_to_gc) * geo->sec_size;
560 bio = pblk_bio_map_addr(pblk, gc_rq->data, gc_rq->secs_to_gc, data_len,
561 PBLK_VMALLOC_META, GFP_KERNEL);
562 if (IS_ERR(bio)) {
563 pr_err("pblk: could not allocate GC bio (%lu)\n", PTR_ERR(bio));
564 goto err_free_dma;
565 }
566
567 bio->bi_iter.bi_sector = 0; /* internal bio */
568 bio_set_op_attrs(bio, REQ_OP_READ, 0);
569
570 rqd.opcode = NVM_OP_PREAD;
571 rqd.nr_ppas = gc_rq->secs_to_gc;
572 rqd.flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
573 rqd.bio = bio;
574
575 if (pblk_submit_io_sync(pblk, &rqd)) {
576 ret = -EIO;
577 pr_err("pblk: GC read request failed\n");
578 goto err_free_bio;
579 }
580
581 atomic_dec(&pblk->inflight_io);
582
583 if (rqd.error) {
584 atomic_long_inc(&pblk->read_failed_gc);
585 #ifdef CONFIG_NVM_DEBUG
586 pblk_print_failed_rqd(pblk, &rqd, rqd.error);
587 #endif
588 }
589
590 #ifdef CONFIG_NVM_DEBUG
591 atomic_long_add(gc_rq->secs_to_gc, &pblk->sync_reads);
592 atomic_long_add(gc_rq->secs_to_gc, &pblk->recov_gc_reads);
593 atomic_long_sub(gc_rq->secs_to_gc, &pblk->inflight_reads);
594 #endif
595
596 out:
597 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
598 return ret;
599
600 err_free_bio:
601 bio_put(bio);
602 err_free_dma:
603 nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
604 return ret;
605 }