]>
Commit | Line | Data |
---|---|---|
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 | ||
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 nvm_tgt_dev *dev = pblk->dev; | |
24 | struct bio *original_bio; | |
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 | ||
31 | w_ctx = pblk_rb_w_ctx(&pblk->rwb, c_ctx->sentry + i); | |
32 | while ((original_bio = bio_list_pop(&w_ctx->bios))) | |
33 | bio_endio(original_bio); | |
34 | } | |
35 | ||
36 | if (c_ctx->nr_padded) | |
37 | pblk_bio_free_pages(pblk, rqd->bio, c_ctx->nr_valid, | |
38 | c_ctx->nr_padded); | |
39 | ||
40 | #ifdef CONFIG_NVM_DEBUG | |
41 | atomic_long_add(rqd->nr_ppas, &pblk->sync_writes); | |
42 | #endif | |
43 | ||
44 | ret = pblk_rb_sync_advance(&pblk->rwb, c_ctx->nr_valid); | |
45 | ||
46 | nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list); | |
47 | ||
48 | bio_put(rqd->bio); | |
49 | pblk_free_rqd(pblk, rqd, WRITE); | |
50 | ||
51 | return ret; | |
52 | } | |
53 | ||
54 | static unsigned long pblk_end_queued_w_bio(struct pblk *pblk, | |
55 | struct nvm_rq *rqd, | |
56 | struct pblk_c_ctx *c_ctx) | |
57 | { | |
58 | list_del(&c_ctx->list); | |
59 | return pblk_end_w_bio(pblk, rqd, c_ctx); | |
60 | } | |
61 | ||
62 | static void pblk_complete_write(struct pblk *pblk, struct nvm_rq *rqd, | |
63 | struct pblk_c_ctx *c_ctx) | |
64 | { | |
65 | struct pblk_c_ctx *c, *r; | |
66 | unsigned long flags; | |
67 | unsigned long pos; | |
68 | ||
69 | #ifdef CONFIG_NVM_DEBUG | |
70 | atomic_long_sub(c_ctx->nr_valid, &pblk->inflight_writes); | |
71 | #endif | |
72 | ||
73 | pblk_up_rq(pblk, rqd->ppa_list, rqd->nr_ppas, c_ctx->lun_bitmap); | |
74 | ||
75 | pos = pblk_rb_sync_init(&pblk->rwb, &flags); | |
76 | if (pos == c_ctx->sentry) { | |
77 | pos = pblk_end_w_bio(pblk, rqd, c_ctx); | |
78 | ||
79 | retry: | |
80 | list_for_each_entry_safe(c, r, &pblk->compl_list, list) { | |
81 | rqd = nvm_rq_from_c_ctx(c); | |
82 | if (c->sentry == pos) { | |
83 | pos = pblk_end_queued_w_bio(pblk, rqd, c); | |
84 | goto retry; | |
85 | } | |
86 | } | |
87 | } else { | |
88 | WARN_ON(nvm_rq_from_c_ctx(c_ctx) != rqd); | |
89 | list_add_tail(&c_ctx->list, &pblk->compl_list); | |
90 | } | |
91 | pblk_rb_sync_end(&pblk->rwb, &flags); | |
92 | } | |
93 | ||
94 | /* When a write fails, we are not sure whether the block has grown bad or a page | |
95 | * range is more susceptible to write errors. If a high number of pages fail, we | |
96 | * assume that the block is bad and we mark it accordingly. In all cases, we | |
97 | * remap and resubmit the failed entries as fast as possible; if a flush is | |
98 | * waiting on a completion, the whole stack would stall otherwise. | |
99 | */ | |
100 | static void pblk_end_w_fail(struct pblk *pblk, struct nvm_rq *rqd) | |
101 | { | |
102 | void *comp_bits = &rqd->ppa_status; | |
103 | struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd); | |
104 | struct pblk_rec_ctx *recovery; | |
105 | struct ppa_addr *ppa_list = rqd->ppa_list; | |
106 | int nr_ppas = rqd->nr_ppas; | |
107 | unsigned int c_entries; | |
108 | int bit, ret; | |
109 | ||
110 | if (unlikely(nr_ppas == 1)) | |
111 | ppa_list = &rqd->ppa_addr; | |
112 | ||
113 | recovery = mempool_alloc(pblk->rec_pool, GFP_ATOMIC); | |
114 | ||
115 | INIT_LIST_HEAD(&recovery->failed); | |
116 | ||
117 | bit = -1; | |
118 | while ((bit = find_next_bit(comp_bits, nr_ppas, bit + 1)) < nr_ppas) { | |
119 | struct pblk_rb_entry *entry; | |
120 | struct ppa_addr ppa; | |
121 | ||
122 | /* Logic error */ | |
123 | if (bit > c_ctx->nr_valid) { | |
124 | WARN_ONCE(1, "pblk: corrupted write request\n"); | |
125 | mempool_free(recovery, pblk->rec_pool); | |
126 | goto out; | |
127 | } | |
128 | ||
129 | ppa = ppa_list[bit]; | |
130 | entry = pblk_rb_sync_scan_entry(&pblk->rwb, &ppa); | |
131 | if (!entry) { | |
132 | pr_err("pblk: could not scan entry on write failure\n"); | |
133 | mempool_free(recovery, pblk->rec_pool); | |
134 | goto out; | |
135 | } | |
136 | ||
137 | /* The list is filled first and emptied afterwards. No need for | |
138 | * protecting it with a lock | |
139 | */ | |
140 | list_add_tail(&entry->index, &recovery->failed); | |
141 | } | |
142 | ||
143 | c_entries = find_first_bit(comp_bits, nr_ppas); | |
144 | ret = pblk_recov_setup_rq(pblk, c_ctx, recovery, comp_bits, c_entries); | |
145 | if (ret) { | |
146 | pr_err("pblk: could not recover from write failure\n"); | |
147 | mempool_free(recovery, pblk->rec_pool); | |
148 | goto out; | |
149 | } | |
150 | ||
151 | INIT_WORK(&recovery->ws_rec, pblk_submit_rec); | |
152 | queue_work(pblk->close_wq, &recovery->ws_rec); | |
153 | ||
154 | out: | |
155 | pblk_complete_write(pblk, rqd, c_ctx); | |
156 | } | |
157 | ||
158 | static void pblk_end_io_write(struct nvm_rq *rqd) | |
159 | { | |
160 | struct pblk *pblk = rqd->private; | |
161 | struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd); | |
162 | ||
163 | if (rqd->error) { | |
164 | pblk_log_write_err(pblk, rqd); | |
165 | return pblk_end_w_fail(pblk, rqd); | |
166 | } | |
167 | #ifdef CONFIG_NVM_DEBUG | |
168 | else | |
169 | WARN_ONCE(rqd->bio->bi_status, "pblk: corrupted write error\n"); | |
170 | #endif | |
171 | ||
172 | pblk_complete_write(pblk, rqd, c_ctx); | |
173 | atomic_dec(&pblk->inflight_io); | |
174 | } | |
175 | ||
176 | static void pblk_end_io_write_meta(struct nvm_rq *rqd) | |
177 | { | |
178 | struct pblk *pblk = rqd->private; | |
179 | struct nvm_tgt_dev *dev = pblk->dev; | |
180 | struct pblk_g_ctx *m_ctx = nvm_rq_to_pdu(rqd); | |
181 | struct pblk_line *line = m_ctx->private; | |
182 | struct pblk_emeta *emeta = line->emeta; | |
183 | int sync; | |
184 | ||
185 | pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas); | |
186 | ||
187 | if (rqd->error) { | |
188 | pblk_log_write_err(pblk, rqd); | |
189 | pr_err("pblk: metadata I/O failed. Line %d\n", line->id); | |
190 | } | |
191 | #ifdef CONFIG_NVM_DEBUG | |
192 | else | |
193 | WARN_ONCE(rqd->bio->bi_status, "pblk: corrupted write error\n"); | |
194 | #endif | |
195 | ||
196 | sync = atomic_add_return(rqd->nr_ppas, &emeta->sync); | |
197 | if (sync == emeta->nr_entries) | |
198 | pblk_gen_run_ws(pblk, line, NULL, pblk_line_close_ws, | |
199 | GFP_ATOMIC, pblk->close_wq); | |
200 | ||
201 | bio_put(rqd->bio); | |
202 | nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list); | |
203 | pblk_free_rqd(pblk, rqd, READ); | |
204 | ||
205 | atomic_dec(&pblk->inflight_io); | |
206 | } | |
207 | ||
208 | static int pblk_alloc_w_rq(struct pblk *pblk, struct nvm_rq *rqd, | |
209 | unsigned int nr_secs, | |
210 | nvm_end_io_fn(*end_io)) | |
211 | { | |
212 | struct nvm_tgt_dev *dev = pblk->dev; | |
213 | ||
214 | /* Setup write request */ | |
215 | rqd->opcode = NVM_OP_PWRITE; | |
216 | rqd->nr_ppas = nr_secs; | |
217 | rqd->flags = pblk_set_progr_mode(pblk, WRITE); | |
218 | rqd->private = pblk; | |
219 | rqd->end_io = end_io; | |
220 | ||
221 | rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, | |
222 | &rqd->dma_meta_list); | |
223 | if (!rqd->meta_list) | |
224 | return -ENOMEM; | |
225 | ||
226 | rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size; | |
227 | rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size; | |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
232 | static int pblk_setup_w_rq(struct pblk *pblk, struct nvm_rq *rqd, | |
233 | struct pblk_c_ctx *c_ctx, struct ppa_addr *erase_ppa) | |
234 | { | |
235 | struct pblk_line_meta *lm = &pblk->lm; | |
236 | struct pblk_line *e_line = pblk_line_get_erase(pblk); | |
237 | unsigned int valid = c_ctx->nr_valid; | |
238 | unsigned int padded = c_ctx->nr_padded; | |
239 | unsigned int nr_secs = valid + padded; | |
240 | unsigned long *lun_bitmap; | |
241 | int ret = 0; | |
242 | ||
243 | lun_bitmap = kzalloc(lm->lun_bitmap_len, GFP_KERNEL); | |
244 | if (!lun_bitmap) | |
245 | return -ENOMEM; | |
246 | c_ctx->lun_bitmap = lun_bitmap; | |
247 | ||
248 | ret = pblk_alloc_w_rq(pblk, rqd, nr_secs, pblk_end_io_write); | |
249 | if (ret) { | |
250 | kfree(lun_bitmap); | |
251 | return ret; | |
252 | } | |
253 | ||
254 | if (likely(!e_line || !atomic_read(&e_line->left_eblks))) | |
255 | pblk_map_rq(pblk, rqd, c_ctx->sentry, lun_bitmap, valid, 0); | |
256 | else | |
257 | pblk_map_erase_rq(pblk, rqd, c_ctx->sentry, lun_bitmap, | |
258 | valid, erase_ppa); | |
259 | ||
260 | return 0; | |
261 | } | |
262 | ||
263 | int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd, | |
264 | struct pblk_c_ctx *c_ctx) | |
265 | { | |
266 | struct pblk_line_meta *lm = &pblk->lm; | |
267 | unsigned long *lun_bitmap; | |
268 | int ret; | |
269 | ||
270 | lun_bitmap = kzalloc(lm->lun_bitmap_len, GFP_KERNEL); | |
271 | if (!lun_bitmap) | |
272 | return -ENOMEM; | |
273 | ||
274 | c_ctx->lun_bitmap = lun_bitmap; | |
275 | ||
276 | ret = pblk_alloc_w_rq(pblk, rqd, rqd->nr_ppas, pblk_end_io_write); | |
277 | if (ret) | |
278 | return ret; | |
279 | ||
280 | pblk_map_rq(pblk, rqd, c_ctx->sentry, lun_bitmap, c_ctx->nr_valid, 0); | |
281 | ||
282 | rqd->ppa_status = (u64)0; | |
283 | rqd->flags = pblk_set_progr_mode(pblk, WRITE); | |
284 | ||
285 | return ret; | |
286 | } | |
287 | ||
288 | static int pblk_calc_secs_to_sync(struct pblk *pblk, unsigned int secs_avail, | |
289 | unsigned int secs_to_flush) | |
290 | { | |
291 | int secs_to_sync; | |
292 | ||
293 | secs_to_sync = pblk_calc_secs(pblk, secs_avail, secs_to_flush); | |
294 | ||
295 | #ifdef CONFIG_NVM_DEBUG | |
296 | if ((!secs_to_sync && secs_to_flush) | |
297 | || (secs_to_sync < 0) | |
298 | || (secs_to_sync > secs_avail && !secs_to_flush)) { | |
299 | pr_err("pblk: bad sector calculation (a:%d,s:%d,f:%d)\n", | |
300 | secs_avail, secs_to_sync, secs_to_flush); | |
301 | } | |
302 | #endif | |
303 | ||
304 | return secs_to_sync; | |
305 | } | |
306 | ||
307 | static inline int pblk_valid_meta_ppa(struct pblk *pblk, | |
308 | struct pblk_line *meta_line, | |
309 | struct ppa_addr *ppa_list, int nr_ppas) | |
310 | { | |
311 | struct nvm_tgt_dev *dev = pblk->dev; | |
312 | struct nvm_geo *geo = &dev->geo; | |
313 | struct pblk_line *data_line; | |
314 | struct ppa_addr ppa, ppa_opt; | |
315 | u64 paddr; | |
316 | int i; | |
317 | ||
318 | data_line = &pblk->lines[pblk_dev_ppa_to_line(ppa_list[0])]; | |
319 | paddr = pblk_lookup_page(pblk, meta_line); | |
320 | ppa = addr_to_gen_ppa(pblk, paddr, 0); | |
321 | ||
322 | if (test_bit(pblk_ppa_to_pos(geo, ppa), data_line->blk_bitmap)) | |
323 | return 1; | |
324 | ||
325 | /* Schedule a metadata I/O that is half the distance from the data I/O | |
326 | * with regards to the number of LUNs forming the pblk instance. This | |
327 | * balances LUN conflicts across every I/O. | |
328 | * | |
329 | * When the LUN configuration changes (e.g., due to GC), this distance | |
330 | * can align, which would result on a LUN deadlock. In this case, modify | |
331 | * the distance to not be optimal, but allow metadata I/Os to succeed. | |
332 | */ | |
333 | ppa_opt = addr_to_gen_ppa(pblk, paddr + data_line->meta_distance, 0); | |
334 | if (unlikely(ppa_opt.ppa == ppa.ppa)) { | |
335 | data_line->meta_distance--; | |
336 | return 0; | |
337 | } | |
338 | ||
339 | for (i = 0; i < nr_ppas; i += pblk->min_write_pgs) | |
340 | if (ppa_list[i].g.ch == ppa_opt.g.ch && | |
341 | ppa_list[i].g.lun == ppa_opt.g.lun) | |
342 | return 1; | |
343 | ||
344 | if (test_bit(pblk_ppa_to_pos(geo, ppa_opt), data_line->blk_bitmap)) { | |
345 | for (i = 0; i < nr_ppas; i += pblk->min_write_pgs) | |
346 | if (ppa_list[i].g.ch == ppa.g.ch && | |
347 | ppa_list[i].g.lun == ppa.g.lun) | |
348 | return 0; | |
349 | ||
350 | return 1; | |
351 | } | |
352 | ||
353 | return 0; | |
354 | } | |
355 | ||
356 | int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line) | |
357 | { | |
358 | struct nvm_tgt_dev *dev = pblk->dev; | |
359 | struct nvm_geo *geo = &dev->geo; | |
360 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; | |
361 | struct pblk_line_meta *lm = &pblk->lm; | |
362 | struct pblk_emeta *emeta = meta_line->emeta; | |
363 | struct pblk_g_ctx *m_ctx; | |
364 | struct bio *bio; | |
365 | struct nvm_rq *rqd; | |
366 | void *data; | |
367 | u64 paddr; | |
368 | int rq_ppas = pblk->min_write_pgs; | |
369 | int id = meta_line->id; | |
370 | int rq_len; | |
371 | int i, j; | |
372 | int ret; | |
373 | ||
374 | rqd = pblk_alloc_rqd(pblk, READ); | |
375 | ||
376 | m_ctx = nvm_rq_to_pdu(rqd); | |
377 | m_ctx->private = meta_line; | |
378 | ||
379 | rq_len = rq_ppas * geo->sec_size; | |
380 | data = ((void *)emeta->buf) + emeta->mem; | |
381 | ||
382 | bio = pblk_bio_map_addr(pblk, data, rq_ppas, rq_len, | |
383 | l_mg->emeta_alloc_type, GFP_KERNEL); | |
384 | if (IS_ERR(bio)) { | |
385 | ret = PTR_ERR(bio); | |
386 | goto fail_free_rqd; | |
387 | } | |
388 | bio->bi_iter.bi_sector = 0; /* internal bio */ | |
389 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); | |
390 | rqd->bio = bio; | |
391 | ||
392 | ret = pblk_alloc_w_rq(pblk, rqd, rq_ppas, pblk_end_io_write_meta); | |
393 | if (ret) | |
394 | goto fail_free_bio; | |
395 | ||
396 | for (i = 0; i < rqd->nr_ppas; ) { | |
397 | spin_lock(&meta_line->lock); | |
398 | paddr = __pblk_alloc_page(pblk, meta_line, rq_ppas); | |
399 | spin_unlock(&meta_line->lock); | |
400 | for (j = 0; j < rq_ppas; j++, i++, paddr++) | |
401 | rqd->ppa_list[i] = addr_to_gen_ppa(pblk, paddr, id); | |
402 | } | |
403 | ||
404 | emeta->mem += rq_len; | |
405 | if (emeta->mem >= lm->emeta_len[0]) { | |
406 | spin_lock(&l_mg->close_lock); | |
407 | list_del(&meta_line->list); | |
408 | spin_unlock(&l_mg->close_lock); | |
409 | } | |
410 | ||
411 | pblk_down_page(pblk, rqd->ppa_list, rqd->nr_ppas); | |
412 | ||
413 | ret = pblk_submit_io(pblk, rqd); | |
414 | if (ret) { | |
415 | pr_err("pblk: emeta I/O submission failed: %d\n", ret); | |
416 | goto fail_rollback; | |
417 | } | |
418 | ||
419 | return NVM_IO_OK; | |
420 | ||
421 | fail_rollback: | |
422 | pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas); | |
423 | spin_lock(&l_mg->close_lock); | |
424 | pblk_dealloc_page(pblk, meta_line, rq_ppas); | |
425 | list_add(&meta_line->list, &meta_line->list); | |
426 | spin_unlock(&l_mg->close_lock); | |
427 | ||
428 | nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list); | |
429 | fail_free_bio: | |
430 | if (likely(l_mg->emeta_alloc_type == PBLK_VMALLOC_META)) | |
431 | bio_put(bio); | |
432 | fail_free_rqd: | |
433 | pblk_free_rqd(pblk, rqd, READ); | |
434 | return ret; | |
435 | } | |
436 | ||
437 | static int pblk_sched_meta_io(struct pblk *pblk, struct ppa_addr *prev_list, | |
438 | int prev_n) | |
439 | { | |
440 | struct pblk_line_meta *lm = &pblk->lm; | |
441 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; | |
442 | struct pblk_line *meta_line; | |
443 | ||
444 | spin_lock(&l_mg->close_lock); | |
445 | retry: | |
446 | if (list_empty(&l_mg->emeta_list)) { | |
447 | spin_unlock(&l_mg->close_lock); | |
448 | return 0; | |
449 | } | |
450 | meta_line = list_first_entry(&l_mg->emeta_list, struct pblk_line, list); | |
451 | if (meta_line->emeta->mem >= lm->emeta_len[0]) | |
452 | goto retry; | |
453 | spin_unlock(&l_mg->close_lock); | |
454 | ||
455 | if (!pblk_valid_meta_ppa(pblk, meta_line, prev_list, prev_n)) | |
456 | return 0; | |
457 | ||
458 | return pblk_submit_meta_io(pblk, meta_line); | |
459 | } | |
460 | ||
461 | static int pblk_submit_io_set(struct pblk *pblk, struct nvm_rq *rqd) | |
462 | { | |
463 | struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd); | |
464 | struct ppa_addr erase_ppa; | |
465 | int err; | |
466 | ||
467 | ppa_set_empty(&erase_ppa); | |
468 | ||
469 | /* Assign lbas to ppas and populate request structure */ | |
470 | err = pblk_setup_w_rq(pblk, rqd, c_ctx, &erase_ppa); | |
471 | if (err) { | |
472 | pr_err("pblk: could not setup write request: %d\n", err); | |
473 | return NVM_IO_ERR; | |
474 | } | |
475 | ||
476 | if (likely(ppa_empty(erase_ppa))) { | |
477 | /* Submit metadata write for previous data line */ | |
478 | err = pblk_sched_meta_io(pblk, rqd->ppa_list, rqd->nr_ppas); | |
479 | if (err) { | |
480 | pr_err("pblk: metadata I/O submission failed: %d", err); | |
481 | return NVM_IO_ERR; | |
482 | } | |
483 | ||
484 | /* Submit data write for current data line */ | |
485 | err = pblk_submit_io(pblk, rqd); | |
486 | if (err) { | |
487 | pr_err("pblk: data I/O submission failed: %d\n", err); | |
488 | return NVM_IO_ERR; | |
489 | } | |
490 | } else { | |
491 | /* Submit data write for current data line */ | |
492 | err = pblk_submit_io(pblk, rqd); | |
493 | if (err) { | |
494 | pr_err("pblk: data I/O submission failed: %d\n", err); | |
495 | return NVM_IO_ERR; | |
496 | } | |
497 | ||
498 | /* Submit available erase for next data line */ | |
499 | if (pblk_blk_erase_async(pblk, erase_ppa)) { | |
500 | struct pblk_line *e_line = pblk_line_get_erase(pblk); | |
501 | struct nvm_tgt_dev *dev = pblk->dev; | |
502 | struct nvm_geo *geo = &dev->geo; | |
503 | int bit; | |
504 | ||
505 | atomic_inc(&e_line->left_eblks); | |
506 | bit = pblk_ppa_to_pos(geo, erase_ppa); | |
507 | WARN_ON(!test_and_clear_bit(bit, e_line->erase_bitmap)); | |
508 | } | |
509 | } | |
510 | ||
511 | return NVM_IO_OK; | |
512 | } | |
513 | ||
514 | static void pblk_free_write_rqd(struct pblk *pblk, struct nvm_rq *rqd) | |
515 | { | |
516 | struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd); | |
517 | struct bio *bio = rqd->bio; | |
518 | ||
519 | if (c_ctx->nr_padded) | |
520 | pblk_bio_free_pages(pblk, bio, c_ctx->nr_valid, | |
521 | c_ctx->nr_padded); | |
522 | } | |
523 | ||
524 | static int pblk_submit_write(struct pblk *pblk) | |
525 | { | |
526 | struct bio *bio; | |
527 | struct nvm_rq *rqd; | |
528 | unsigned int secs_avail, secs_to_sync, secs_to_com; | |
529 | unsigned int secs_to_flush; | |
530 | unsigned long pos; | |
531 | ||
532 | /* If there are no sectors in the cache, flushes (bios without data) | |
533 | * will be cleared on the cache threads | |
534 | */ | |
535 | secs_avail = pblk_rb_read_count(&pblk->rwb); | |
536 | if (!secs_avail) | |
537 | return 1; | |
538 | ||
539 | secs_to_flush = pblk_rb_sync_point_count(&pblk->rwb); | |
540 | if (!secs_to_flush && secs_avail < pblk->min_write_pgs) | |
541 | return 1; | |
542 | ||
543 | bio = bio_alloc(GFP_KERNEL, pblk->max_write_pgs); | |
544 | ||
545 | bio->bi_iter.bi_sector = 0; /* internal bio */ | |
546 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); | |
547 | ||
548 | rqd = pblk_alloc_rqd(pblk, WRITE); | |
549 | rqd->bio = bio; | |
550 | ||
551 | secs_to_sync = pblk_calc_secs_to_sync(pblk, secs_avail, secs_to_flush); | |
552 | if (secs_to_sync > pblk->max_write_pgs) { | |
553 | pr_err("pblk: bad buffer sync calculation\n"); | |
554 | goto fail_put_bio; | |
555 | } | |
556 | ||
557 | secs_to_com = (secs_to_sync > secs_avail) ? secs_avail : secs_to_sync; | |
558 | pos = pblk_rb_read_commit(&pblk->rwb, secs_to_com); | |
559 | ||
560 | if (pblk_rb_read_to_bio(&pblk->rwb, rqd, bio, pos, secs_to_sync, | |
561 | secs_avail)) { | |
562 | pr_err("pblk: corrupted write bio\n"); | |
563 | goto fail_put_bio; | |
564 | } | |
565 | ||
566 | if (pblk_submit_io_set(pblk, rqd)) | |
567 | goto fail_free_bio; | |
568 | ||
569 | #ifdef CONFIG_NVM_DEBUG | |
570 | atomic_long_add(secs_to_sync, &pblk->sub_writes); | |
571 | #endif | |
572 | ||
573 | return 0; | |
574 | ||
575 | fail_free_bio: | |
576 | pblk_free_write_rqd(pblk, rqd); | |
577 | fail_put_bio: | |
578 | bio_put(bio); | |
579 | pblk_free_rqd(pblk, rqd, WRITE); | |
580 | ||
581 | return 1; | |
582 | } | |
583 | ||
584 | int pblk_write_ts(void *data) | |
585 | { | |
586 | struct pblk *pblk = data; | |
587 | ||
588 | while (!kthread_should_stop()) { | |
589 | if (!pblk_submit_write(pblk)) | |
590 | continue; | |
591 | set_current_state(TASK_INTERRUPTIBLE); | |
592 | io_schedule(); | |
593 | } | |
594 | ||
595 | return 0; | |
596 | } |