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-write.c - pblk's write path from write buffer to media
16 */
17
18 #include "pblk.h"
19
pblk_end_w_bio(struct pblk * pblk,struct nvm_rq * rqd,struct pblk_c_ctx * c_ctx)20 static unsigned long pblk_end_w_bio(struct pblk *pblk, struct nvm_rq *rqd,
21 struct pblk_c_ctx *c_ctx)
22 {
23 struct bio *original_bio;
24 struct pblk_rb *rwb = &pblk->rwb;
25 unsigned long ret;
26 int i;
27
28 for (i = 0; i < c_ctx->nr_valid; i++) {
29 struct pblk_w_ctx *w_ctx;
30 int pos = c_ctx->sentry + i;
31 int flags;
32
33 w_ctx = pblk_rb_w_ctx(rwb, pos);
34 flags = READ_ONCE(w_ctx->flags);
35
36 if (flags & PBLK_FLUSH_ENTRY) {
37 flags &= ~PBLK_FLUSH_ENTRY;
38 /* Release flags on context. Protect from writes */
39 smp_store_release(&w_ctx->flags, flags);
40
41 #ifdef CONFIG_NVM_PBLK_DEBUG
42 atomic_dec(&rwb->inflight_flush_point);
43 #endif
44 }
45
46 while ((original_bio = bio_list_pop(&w_ctx->bios)))
47 bio_endio(original_bio);
48 }
49
50 if (c_ctx->nr_padded)
51 pblk_bio_free_pages(pblk, rqd->bio, c_ctx->nr_valid,
52 c_ctx->nr_padded);
53
54 #ifdef CONFIG_NVM_PBLK_DEBUG
55 atomic_long_add(rqd->nr_ppas, &pblk->sync_writes);
56 #endif
57
58 ret = pblk_rb_sync_advance(&pblk->rwb, c_ctx->nr_valid);
59
60 bio_put(rqd->bio);
61 pblk_free_rqd(pblk, rqd, PBLK_WRITE);
62
63 return ret;
64 }
65
pblk_end_queued_w_bio(struct pblk * pblk,struct nvm_rq * rqd,struct pblk_c_ctx * c_ctx)66 static unsigned long pblk_end_queued_w_bio(struct pblk *pblk,
67 struct nvm_rq *rqd,
68 struct pblk_c_ctx *c_ctx)
69 {
70 list_del(&c_ctx->list);
71 return pblk_end_w_bio(pblk, rqd, c_ctx);
72 }
73
pblk_complete_write(struct pblk * pblk,struct nvm_rq * rqd,struct pblk_c_ctx * c_ctx)74 static void pblk_complete_write(struct pblk *pblk, struct nvm_rq *rqd,
75 struct pblk_c_ctx *c_ctx)
76 {
77 struct pblk_c_ctx *c, *r;
78 unsigned long flags;
79 unsigned long pos;
80
81 #ifdef CONFIG_NVM_PBLK_DEBUG
82 atomic_long_sub(c_ctx->nr_valid, &pblk->inflight_writes);
83 #endif
84
85 pblk_up_rq(pblk, rqd->ppa_list, rqd->nr_ppas, c_ctx->lun_bitmap);
86
87 pos = pblk_rb_sync_init(&pblk->rwb, &flags);
88 if (pos == c_ctx->sentry) {
89 pos = pblk_end_w_bio(pblk, rqd, c_ctx);
90
91 retry:
92 list_for_each_entry_safe(c, r, &pblk->compl_list, list) {
93 rqd = nvm_rq_from_c_ctx(c);
94 if (c->sentry == pos) {
95 pos = pblk_end_queued_w_bio(pblk, rqd, c);
96 goto retry;
97 }
98 }
99 } else {
100 WARN_ON(nvm_rq_from_c_ctx(c_ctx) != rqd);
101 list_add_tail(&c_ctx->list, &pblk->compl_list);
102 }
103 pblk_rb_sync_end(&pblk->rwb, &flags);
104 }
105
106 /* Map remaining sectors in chunk, starting from ppa */
pblk_map_remaining(struct pblk * pblk,struct ppa_addr * ppa)107 static void pblk_map_remaining(struct pblk *pblk, struct ppa_addr *ppa)
108 {
109 struct nvm_tgt_dev *dev = pblk->dev;
110 struct nvm_geo *geo = &dev->geo;
111 struct pblk_line *line;
112 struct ppa_addr map_ppa = *ppa;
113 u64 paddr;
114 int done = 0;
115
116 line = &pblk->lines[pblk_ppa_to_line(*ppa)];
117 spin_lock(&line->lock);
118
119 while (!done) {
120 paddr = pblk_dev_ppa_to_line_addr(pblk, map_ppa);
121
122 if (!test_and_set_bit(paddr, line->map_bitmap))
123 line->left_msecs--;
124
125 if (!test_and_set_bit(paddr, line->invalid_bitmap))
126 le32_add_cpu(line->vsc, -1);
127
128 if (geo->version == NVM_OCSSD_SPEC_12) {
129 map_ppa.ppa++;
130 if (map_ppa.g.pg == geo->num_pg)
131 done = 1;
132 } else {
133 map_ppa.m.sec++;
134 if (map_ppa.m.sec == geo->clba)
135 done = 1;
136 }
137 }
138
139 line->w_err_gc->has_write_err = 1;
140 spin_unlock(&line->lock);
141 }
142
pblk_prepare_resubmit(struct pblk * pblk,unsigned int sentry,unsigned int nr_entries)143 static void pblk_prepare_resubmit(struct pblk *pblk, unsigned int sentry,
144 unsigned int nr_entries)
145 {
146 struct pblk_rb *rb = &pblk->rwb;
147 struct pblk_rb_entry *entry;
148 struct pblk_line *line;
149 struct pblk_w_ctx *w_ctx;
150 struct ppa_addr ppa_l2p;
151 int flags;
152 unsigned int pos, i;
153
154 spin_lock(&pblk->trans_lock);
155 pos = sentry;
156 for (i = 0; i < nr_entries; i++) {
157 entry = &rb->entries[pos];
158 w_ctx = &entry->w_ctx;
159
160 /* Check if the lba has been overwritten */
161 ppa_l2p = pblk_trans_map_get(pblk, w_ctx->lba);
162 if (!pblk_ppa_comp(ppa_l2p, entry->cacheline))
163 w_ctx->lba = ADDR_EMPTY;
164
165 /* Mark up the entry as submittable again */
166 flags = READ_ONCE(w_ctx->flags);
167 flags |= PBLK_WRITTEN_DATA;
168 /* Release flags on write context. Protect from writes */
169 smp_store_release(&w_ctx->flags, flags);
170
171 /* Decrese the reference count to the line as we will
172 * re-map these entries
173 */
174 line = &pblk->lines[pblk_ppa_to_line(w_ctx->ppa)];
175 kref_put(&line->ref, pblk_line_put);
176
177 pos = (pos + 1) & (rb->nr_entries - 1);
178 }
179 spin_unlock(&pblk->trans_lock);
180 }
181
pblk_queue_resubmit(struct pblk * pblk,struct pblk_c_ctx * c_ctx)182 static void pblk_queue_resubmit(struct pblk *pblk, struct pblk_c_ctx *c_ctx)
183 {
184 struct pblk_c_ctx *r_ctx;
185
186 r_ctx = kzalloc(sizeof(struct pblk_c_ctx), GFP_KERNEL);
187 if (!r_ctx)
188 return;
189
190 r_ctx->lun_bitmap = NULL;
191 r_ctx->sentry = c_ctx->sentry;
192 r_ctx->nr_valid = c_ctx->nr_valid;
193 r_ctx->nr_padded = c_ctx->nr_padded;
194
195 spin_lock(&pblk->resubmit_lock);
196 list_add_tail(&r_ctx->list, &pblk->resubmit_list);
197 spin_unlock(&pblk->resubmit_lock);
198
199 #ifdef CONFIG_NVM_PBLK_DEBUG
200 atomic_long_add(c_ctx->nr_valid, &pblk->recov_writes);
201 #endif
202 }
203
pblk_submit_rec(struct work_struct * work)204 static void pblk_submit_rec(struct work_struct *work)
205 {
206 struct pblk_rec_ctx *recovery =
207 container_of(work, struct pblk_rec_ctx, ws_rec);
208 struct pblk *pblk = recovery->pblk;
209 struct nvm_rq *rqd = recovery->rqd;
210 struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
211 struct ppa_addr *ppa_list;
212
213 pblk_log_write_err(pblk, rqd);
214
215 if (rqd->nr_ppas == 1)
216 ppa_list = &rqd->ppa_addr;
217 else
218 ppa_list = rqd->ppa_list;
219
220 pblk_map_remaining(pblk, ppa_list);
221 pblk_queue_resubmit(pblk, c_ctx);
222
223 pblk_up_rq(pblk, rqd->ppa_list, rqd->nr_ppas, c_ctx->lun_bitmap);
224 if (c_ctx->nr_padded)
225 pblk_bio_free_pages(pblk, rqd->bio, c_ctx->nr_valid,
226 c_ctx->nr_padded);
227 bio_put(rqd->bio);
228 pblk_free_rqd(pblk, rqd, PBLK_WRITE);
229 mempool_free(recovery, &pblk->rec_pool);
230
231 atomic_dec(&pblk->inflight_io);
232 }
233
234
pblk_end_w_fail(struct pblk * pblk,struct nvm_rq * rqd)235 static void pblk_end_w_fail(struct pblk *pblk, struct nvm_rq *rqd)
236 {
237 struct pblk_rec_ctx *recovery;
238
239 recovery = mempool_alloc(&pblk->rec_pool, GFP_ATOMIC);
240 if (!recovery) {
241 pblk_err(pblk, "could not allocate recovery work\n");
242 return;
243 }
244
245 recovery->pblk = pblk;
246 recovery->rqd = rqd;
247
248 INIT_WORK(&recovery->ws_rec, pblk_submit_rec);
249 queue_work(pblk->close_wq, &recovery->ws_rec);
250 }
251
pblk_end_io_write(struct nvm_rq * rqd)252 static void pblk_end_io_write(struct nvm_rq *rqd)
253 {
254 struct pblk *pblk = rqd->private;
255 struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
256
257 if (rqd->error) {
258 pblk_end_w_fail(pblk, rqd);
259 return;
260 }
261 #ifdef CONFIG_NVM_PBLK_DEBUG
262 else
263 WARN_ONCE(rqd->bio->bi_status, "pblk: corrupted write error\n");
264 #endif
265
266 pblk_complete_write(pblk, rqd, c_ctx);
267 atomic_dec(&pblk->inflight_io);
268 }
269
pblk_end_io_write_meta(struct nvm_rq * rqd)270 static void pblk_end_io_write_meta(struct nvm_rq *rqd)
271 {
272 struct pblk *pblk = rqd->private;
273 struct pblk_g_ctx *m_ctx = nvm_rq_to_pdu(rqd);
274 struct pblk_line *line = m_ctx->private;
275 struct pblk_emeta *emeta = line->emeta;
276 int sync;
277
278 pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
279
280 if (rqd->error) {
281 pblk_log_write_err(pblk, rqd);
282 pblk_err(pblk, "metadata I/O failed. Line %d\n", line->id);
283 line->w_err_gc->has_write_err = 1;
284 }
285
286 sync = atomic_add_return(rqd->nr_ppas, &emeta->sync);
287 if (sync == emeta->nr_entries)
288 pblk_gen_run_ws(pblk, line, NULL, pblk_line_close_ws,
289 GFP_ATOMIC, pblk->close_wq);
290
291 pblk_free_rqd(pblk, rqd, PBLK_WRITE_INT);
292
293 atomic_dec(&pblk->inflight_io);
294 }
295
pblk_alloc_w_rq(struct pblk * pblk,struct nvm_rq * rqd,unsigned int nr_secs,nvm_end_io_fn (* end_io))296 static int pblk_alloc_w_rq(struct pblk *pblk, struct nvm_rq *rqd,
297 unsigned int nr_secs,
298 nvm_end_io_fn(*end_io))
299 {
300 struct nvm_tgt_dev *dev = pblk->dev;
301
302 /* Setup write request */
303 rqd->opcode = NVM_OP_PWRITE;
304 rqd->nr_ppas = nr_secs;
305 rqd->flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
306 rqd->private = pblk;
307 rqd->end_io = end_io;
308
309 rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
310 &rqd->dma_meta_list);
311 if (!rqd->meta_list)
312 return -ENOMEM;
313
314 rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size;
315 rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size;
316
317 return 0;
318 }
319
pblk_setup_w_rq(struct pblk * pblk,struct nvm_rq * rqd,struct ppa_addr * erase_ppa)320 static int pblk_setup_w_rq(struct pblk *pblk, struct nvm_rq *rqd,
321 struct ppa_addr *erase_ppa)
322 {
323 struct pblk_line_meta *lm = &pblk->lm;
324 struct pblk_line *e_line = pblk_line_get_erase(pblk);
325 struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
326 unsigned int valid = c_ctx->nr_valid;
327 unsigned int padded = c_ctx->nr_padded;
328 unsigned int nr_secs = valid + padded;
329 unsigned long *lun_bitmap;
330 int ret;
331
332 lun_bitmap = kzalloc(lm->lun_bitmap_len, GFP_KERNEL);
333 if (!lun_bitmap)
334 return -ENOMEM;
335 c_ctx->lun_bitmap = lun_bitmap;
336
337 ret = pblk_alloc_w_rq(pblk, rqd, nr_secs, pblk_end_io_write);
338 if (ret) {
339 kfree(lun_bitmap);
340 return ret;
341 }
342
343 if (likely(!e_line || !atomic_read(&e_line->left_eblks)))
344 pblk_map_rq(pblk, rqd, c_ctx->sentry, lun_bitmap, valid, 0);
345 else
346 pblk_map_erase_rq(pblk, rqd, c_ctx->sentry, lun_bitmap,
347 valid, erase_ppa);
348
349 return 0;
350 }
351
pblk_calc_secs_to_sync(struct pblk * pblk,unsigned int secs_avail,unsigned int secs_to_flush)352 static int pblk_calc_secs_to_sync(struct pblk *pblk, unsigned int secs_avail,
353 unsigned int secs_to_flush)
354 {
355 int secs_to_sync;
356
357 secs_to_sync = pblk_calc_secs(pblk, secs_avail, secs_to_flush);
358
359 #ifdef CONFIG_NVM_PBLK_DEBUG
360 if ((!secs_to_sync && secs_to_flush)
361 || (secs_to_sync < 0)
362 || (secs_to_sync > secs_avail && !secs_to_flush)) {
363 pblk_err(pblk, "bad sector calculation (a:%d,s:%d,f:%d)\n",
364 secs_avail, secs_to_sync, secs_to_flush);
365 }
366 #endif
367
368 return secs_to_sync;
369 }
370
pblk_submit_meta_io(struct pblk * pblk,struct pblk_line * meta_line)371 int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line)
372 {
373 struct nvm_tgt_dev *dev = pblk->dev;
374 struct nvm_geo *geo = &dev->geo;
375 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
376 struct pblk_line_meta *lm = &pblk->lm;
377 struct pblk_emeta *emeta = meta_line->emeta;
378 struct pblk_g_ctx *m_ctx;
379 struct bio *bio;
380 struct nvm_rq *rqd;
381 void *data;
382 u64 paddr;
383 int rq_ppas = pblk->min_write_pgs;
384 int id = meta_line->id;
385 int rq_len;
386 int i, j;
387 int ret;
388
389 rqd = pblk_alloc_rqd(pblk, PBLK_WRITE_INT);
390
391 m_ctx = nvm_rq_to_pdu(rqd);
392 m_ctx->private = meta_line;
393
394 rq_len = rq_ppas * geo->csecs;
395 data = ((void *)emeta->buf) + emeta->mem;
396
397 bio = pblk_bio_map_addr(pblk, data, rq_ppas, rq_len,
398 l_mg->emeta_alloc_type, GFP_KERNEL);
399 if (IS_ERR(bio)) {
400 pblk_err(pblk, "failed to map emeta io");
401 ret = PTR_ERR(bio);
402 goto fail_free_rqd;
403 }
404 bio->bi_iter.bi_sector = 0; /* internal bio */
405 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
406 rqd->bio = bio;
407
408 ret = pblk_alloc_w_rq(pblk, rqd, rq_ppas, pblk_end_io_write_meta);
409 if (ret)
410 goto fail_free_bio;
411
412 for (i = 0; i < rqd->nr_ppas; ) {
413 spin_lock(&meta_line->lock);
414 paddr = __pblk_alloc_page(pblk, meta_line, rq_ppas);
415 spin_unlock(&meta_line->lock);
416 for (j = 0; j < rq_ppas; j++, i++, paddr++)
417 rqd->ppa_list[i] = addr_to_gen_ppa(pblk, paddr, id);
418 }
419
420 emeta->mem += rq_len;
421 if (emeta->mem >= lm->emeta_len[0]) {
422 spin_lock(&l_mg->close_lock);
423 list_del(&meta_line->list);
424 spin_unlock(&l_mg->close_lock);
425 }
426
427 pblk_down_page(pblk, rqd->ppa_list, rqd->nr_ppas);
428
429 ret = pblk_submit_io(pblk, rqd);
430 if (ret) {
431 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
432 goto fail_rollback;
433 }
434
435 return NVM_IO_OK;
436
437 fail_rollback:
438 pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
439 spin_lock(&l_mg->close_lock);
440 pblk_dealloc_page(pblk, meta_line, rq_ppas);
441 list_add(&meta_line->list, &meta_line->list);
442 spin_unlock(&l_mg->close_lock);
443 fail_free_bio:
444 bio_put(bio);
445 fail_free_rqd:
446 pblk_free_rqd(pblk, rqd, PBLK_WRITE_INT);
447 return ret;
448 }
449
pblk_valid_meta_ppa(struct pblk * pblk,struct pblk_line * meta_line,struct nvm_rq * data_rqd)450 static inline bool pblk_valid_meta_ppa(struct pblk *pblk,
451 struct pblk_line *meta_line,
452 struct nvm_rq *data_rqd)
453 {
454 struct nvm_tgt_dev *dev = pblk->dev;
455 struct nvm_geo *geo = &dev->geo;
456 struct pblk_c_ctx *data_c_ctx = nvm_rq_to_pdu(data_rqd);
457 struct pblk_line *data_line = pblk_line_get_data(pblk);
458 struct ppa_addr ppa, ppa_opt;
459 u64 paddr;
460 int pos_opt;
461
462 /* Schedule a metadata I/O that is half the distance from the data I/O
463 * with regards to the number of LUNs forming the pblk instance. This
464 * balances LUN conflicts across every I/O.
465 *
466 * When the LUN configuration changes (e.g., due to GC), this distance
467 * can align, which would result on metadata and data I/Os colliding. In
468 * this case, modify the distance to not be optimal, but move the
469 * optimal in the right direction.
470 */
471 paddr = pblk_lookup_page(pblk, meta_line);
472 ppa = addr_to_gen_ppa(pblk, paddr, 0);
473 ppa_opt = addr_to_gen_ppa(pblk, paddr + data_line->meta_distance, 0);
474 pos_opt = pblk_ppa_to_pos(geo, ppa_opt);
475
476 if (test_bit(pos_opt, data_c_ctx->lun_bitmap) ||
477 test_bit(pos_opt, data_line->blk_bitmap))
478 return true;
479
480 if (unlikely(pblk_ppa_comp(ppa_opt, ppa)))
481 data_line->meta_distance--;
482
483 return false;
484 }
485
pblk_should_submit_meta_io(struct pblk * pblk,struct nvm_rq * data_rqd)486 static struct pblk_line *pblk_should_submit_meta_io(struct pblk *pblk,
487 struct nvm_rq *data_rqd)
488 {
489 struct pblk_line_meta *lm = &pblk->lm;
490 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
491 struct pblk_line *meta_line;
492
493 spin_lock(&l_mg->close_lock);
494 retry:
495 if (list_empty(&l_mg->emeta_list)) {
496 spin_unlock(&l_mg->close_lock);
497 return NULL;
498 }
499 meta_line = list_first_entry(&l_mg->emeta_list, struct pblk_line, list);
500 if (meta_line->emeta->mem >= lm->emeta_len[0])
501 goto retry;
502 spin_unlock(&l_mg->close_lock);
503
504 if (!pblk_valid_meta_ppa(pblk, meta_line, data_rqd))
505 return NULL;
506
507 return meta_line;
508 }
509
pblk_submit_io_set(struct pblk * pblk,struct nvm_rq * rqd)510 static int pblk_submit_io_set(struct pblk *pblk, struct nvm_rq *rqd)
511 {
512 struct ppa_addr erase_ppa;
513 struct pblk_line *meta_line;
514 int err;
515
516 pblk_ppa_set_empty(&erase_ppa);
517
518 /* Assign lbas to ppas and populate request structure */
519 err = pblk_setup_w_rq(pblk, rqd, &erase_ppa);
520 if (err) {
521 pblk_err(pblk, "could not setup write request: %d\n", err);
522 return NVM_IO_ERR;
523 }
524
525 meta_line = pblk_should_submit_meta_io(pblk, rqd);
526
527 /* Submit data write for current data line */
528 err = pblk_submit_io(pblk, rqd);
529 if (err) {
530 pblk_err(pblk, "data I/O submission failed: %d\n", err);
531 return NVM_IO_ERR;
532 }
533
534 if (!pblk_ppa_empty(erase_ppa)) {
535 /* Submit erase for next data line */
536 if (pblk_blk_erase_async(pblk, erase_ppa)) {
537 struct pblk_line *e_line = pblk_line_get_erase(pblk);
538 struct nvm_tgt_dev *dev = pblk->dev;
539 struct nvm_geo *geo = &dev->geo;
540 int bit;
541
542 atomic_inc(&e_line->left_eblks);
543 bit = pblk_ppa_to_pos(geo, erase_ppa);
544 WARN_ON(!test_and_clear_bit(bit, e_line->erase_bitmap));
545 }
546 }
547
548 if (meta_line) {
549 /* Submit metadata write for previous data line */
550 err = pblk_submit_meta_io(pblk, meta_line);
551 if (err) {
552 pblk_err(pblk, "metadata I/O submission failed: %d",
553 err);
554 return NVM_IO_ERR;
555 }
556 }
557
558 return NVM_IO_OK;
559 }
560
pblk_free_write_rqd(struct pblk * pblk,struct nvm_rq * rqd)561 static void pblk_free_write_rqd(struct pblk *pblk, struct nvm_rq *rqd)
562 {
563 struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
564 struct bio *bio = rqd->bio;
565
566 if (c_ctx->nr_padded)
567 pblk_bio_free_pages(pblk, bio, c_ctx->nr_valid,
568 c_ctx->nr_padded);
569 }
570
pblk_submit_write(struct pblk * pblk)571 static int pblk_submit_write(struct pblk *pblk)
572 {
573 struct bio *bio;
574 struct nvm_rq *rqd;
575 unsigned int secs_avail, secs_to_sync, secs_to_com;
576 unsigned int secs_to_flush;
577 unsigned long pos;
578 unsigned int resubmit;
579
580 spin_lock(&pblk->resubmit_lock);
581 resubmit = !list_empty(&pblk->resubmit_list);
582 spin_unlock(&pblk->resubmit_lock);
583
584 /* Resubmit failed writes first */
585 if (resubmit) {
586 struct pblk_c_ctx *r_ctx;
587
588 spin_lock(&pblk->resubmit_lock);
589 r_ctx = list_first_entry(&pblk->resubmit_list,
590 struct pblk_c_ctx, list);
591 list_del(&r_ctx->list);
592 spin_unlock(&pblk->resubmit_lock);
593
594 secs_avail = r_ctx->nr_valid;
595 pos = r_ctx->sentry;
596
597 pblk_prepare_resubmit(pblk, pos, secs_avail);
598 secs_to_sync = pblk_calc_secs_to_sync(pblk, secs_avail,
599 secs_avail);
600
601 kfree(r_ctx);
602 } else {
603 /* If there are no sectors in the cache,
604 * flushes (bios without data) will be cleared on
605 * the cache threads
606 */
607 secs_avail = pblk_rb_read_count(&pblk->rwb);
608 if (!secs_avail)
609 return 1;
610
611 secs_to_flush = pblk_rb_flush_point_count(&pblk->rwb);
612 if (!secs_to_flush && secs_avail < pblk->min_write_pgs)
613 return 1;
614
615 secs_to_sync = pblk_calc_secs_to_sync(pblk, secs_avail,
616 secs_to_flush);
617 if (secs_to_sync > pblk->max_write_pgs) {
618 pblk_err(pblk, "bad buffer sync calculation\n");
619 return 1;
620 }
621
622 secs_to_com = (secs_to_sync > secs_avail) ?
623 secs_avail : secs_to_sync;
624 pos = pblk_rb_read_commit(&pblk->rwb, secs_to_com);
625 }
626
627 bio = bio_alloc(GFP_KERNEL, secs_to_sync);
628
629 bio->bi_iter.bi_sector = 0; /* internal bio */
630 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
631
632 rqd = pblk_alloc_rqd(pblk, PBLK_WRITE);
633 rqd->bio = bio;
634
635 if (pblk_rb_read_to_bio(&pblk->rwb, rqd, pos, secs_to_sync,
636 secs_avail)) {
637 pblk_err(pblk, "corrupted write bio\n");
638 goto fail_put_bio;
639 }
640
641 if (pblk_submit_io_set(pblk, rqd))
642 goto fail_free_bio;
643
644 #ifdef CONFIG_NVM_PBLK_DEBUG
645 atomic_long_add(secs_to_sync, &pblk->sub_writes);
646 #endif
647
648 return 0;
649
650 fail_free_bio:
651 pblk_free_write_rqd(pblk, rqd);
652 fail_put_bio:
653 bio_put(bio);
654 pblk_free_rqd(pblk, rqd, PBLK_WRITE);
655
656 return 1;
657 }
658
pblk_write_ts(void * data)659 int pblk_write_ts(void *data)
660 {
661 struct pblk *pblk = data;
662
663 while (!kthread_should_stop()) {
664 if (!pblk_submit_write(pblk))
665 continue;
666 set_current_state(TASK_INTERRUPTIBLE);
667 io_schedule();
668 }
669
670 return 0;
671 }
672
