pblk-write.c 15.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
/*
 * Copyright (C) 2016 CNEX Labs
 * Initial release: Javier Gonzalez <javier@cnexlabs.com>
 *                  Matias Bjorling <matias@cnexlabs.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License version
 * 2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * pblk-write.c - pblk's write path from write buffer to media
 */

#include "pblk.h"

static unsigned long pblk_end_w_bio(struct pblk *pblk, struct nvm_rq *rqd,
				    struct pblk_c_ctx *c_ctx)
{
	struct nvm_tgt_dev *dev = pblk->dev;
	struct bio *original_bio;
	unsigned long ret;
	int i;

	for (i = 0; i < c_ctx->nr_valid; i++) {
		struct pblk_w_ctx *w_ctx;

		w_ctx = pblk_rb_w_ctx(&pblk->rwb, c_ctx->sentry + i);
		while ((original_bio = bio_list_pop(&w_ctx->bios)))
			bio_endio(original_bio);
	}

36 37 38 39
	if (c_ctx->nr_padded)
		pblk_bio_free_pages(pblk, rqd->bio, c_ctx->nr_valid,
							c_ctx->nr_padded);

40
#ifdef CONFIG_NVM_DEBUG
41
	atomic_long_add(rqd->nr_ppas, &pblk->sync_writes);
42 43 44 45
#endif

	ret = pblk_rb_sync_advance(&pblk->rwb, c_ctx->nr_valid);

46
	nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113

	bio_put(rqd->bio);
	pblk_free_rqd(pblk, rqd, WRITE);

	return ret;
}

static unsigned long pblk_end_queued_w_bio(struct pblk *pblk,
					   struct nvm_rq *rqd,
					   struct pblk_c_ctx *c_ctx)
{
	list_del(&c_ctx->list);
	return pblk_end_w_bio(pblk, rqd, c_ctx);
}

static void pblk_complete_write(struct pblk *pblk, struct nvm_rq *rqd,
				struct pblk_c_ctx *c_ctx)
{
	struct pblk_c_ctx *c, *r;
	unsigned long flags;
	unsigned long pos;

#ifdef CONFIG_NVM_DEBUG
	atomic_long_sub(c_ctx->nr_valid, &pblk->inflight_writes);
#endif

	pblk_up_rq(pblk, rqd->ppa_list, rqd->nr_ppas, c_ctx->lun_bitmap);

	pos = pblk_rb_sync_init(&pblk->rwb, &flags);
	if (pos == c_ctx->sentry) {
		pos = pblk_end_w_bio(pblk, rqd, c_ctx);

retry:
		list_for_each_entry_safe(c, r, &pblk->compl_list, list) {
			rqd = nvm_rq_from_c_ctx(c);
			if (c->sentry == pos) {
				pos = pblk_end_queued_w_bio(pblk, rqd, c);
				goto retry;
			}
		}
	} else {
		WARN_ON(nvm_rq_from_c_ctx(c_ctx) != rqd);
		list_add_tail(&c_ctx->list, &pblk->compl_list);
	}
	pblk_rb_sync_end(&pblk->rwb, &flags);
}

/* When a write fails, we are not sure whether the block has grown bad or a page
 * range is more susceptible to write errors. If a high number of pages fail, we
 * assume that the block is bad and we mark it accordingly. In all cases, we
 * remap and resubmit the failed entries as fast as possible; if a flush is
 * waiting on a completion, the whole stack would stall otherwise.
 */
static void pblk_end_w_fail(struct pblk *pblk, struct nvm_rq *rqd)
{
	void *comp_bits = &rqd->ppa_status;
	struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
	struct pblk_rec_ctx *recovery;
	struct ppa_addr *ppa_list = rqd->ppa_list;
	int nr_ppas = rqd->nr_ppas;
	unsigned int c_entries;
	int bit, ret;

	if (unlikely(nr_ppas == 1))
		ppa_list = &rqd->ppa_addr;

	recovery = mempool_alloc(pblk->rec_pool, GFP_ATOMIC);
114

115 116 117 118 119 120 121 122 123
	INIT_LIST_HEAD(&recovery->failed);

	bit = -1;
	while ((bit = find_next_bit(comp_bits, nr_ppas, bit + 1)) < nr_ppas) {
		struct pblk_rb_entry *entry;
		struct ppa_addr ppa;

		/* Logic error */
		if (bit > c_ctx->nr_valid) {
124
			WARN_ONCE(1, "pblk: corrupted write request\n");
125
			mempool_free(recovery, pblk->rec_pool);
126 127 128 129 130 131 132
			goto out;
		}

		ppa = ppa_list[bit];
		entry = pblk_rb_sync_scan_entry(&pblk->rwb, &ppa);
		if (!entry) {
			pr_err("pblk: could not scan entry on write failure\n");
133
			mempool_free(recovery, pblk->rec_pool);
134 135 136 137 138 139 140 141 142 143 144 145 146
			goto out;
		}

		/* The list is filled first and emptied afterwards. No need for
		 * protecting it with a lock
		 */
		list_add_tail(&entry->index, &recovery->failed);
	}

	c_entries = find_first_bit(comp_bits, nr_ppas);
	ret = pblk_recov_setup_rq(pblk, c_ctx, recovery, comp_bits, c_entries);
	if (ret) {
		pr_err("pblk: could not recover from write failure\n");
147
		mempool_free(recovery, pblk->rec_pool);
148 149 150 151
		goto out;
	}

	INIT_WORK(&recovery->ws_rec, pblk_submit_rec);
152
	queue_work(pblk->close_wq, &recovery->ws_rec);
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168

out:
	pblk_complete_write(pblk, rqd, c_ctx);
}

static void pblk_end_io_write(struct nvm_rq *rqd)
{
	struct pblk *pblk = rqd->private;
	struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);

	if (rqd->error) {
		pblk_log_write_err(pblk, rqd);
		return pblk_end_w_fail(pblk, rqd);
	}
#ifdef CONFIG_NVM_DEBUG
	else
169
		WARN_ONCE(rqd->bio->bi_status, "pblk: corrupted write error\n");
170 171 172
#endif

	pblk_complete_write(pblk, rqd, c_ctx);
173
	atomic_dec(&pblk->inflight_io);
174 175
}

176 177 178 179 180 181 182 183 184
static void pblk_end_io_write_meta(struct nvm_rq *rqd)
{
	struct pblk *pblk = rqd->private;
	struct nvm_tgt_dev *dev = pblk->dev;
	struct pblk_g_ctx *m_ctx = nvm_rq_to_pdu(rqd);
	struct pblk_line *line = m_ctx->private;
	struct pblk_emeta *emeta = line->emeta;
	int sync;

185
	pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
186 187 188

	if (rqd->error) {
		pblk_log_write_err(pblk, rqd);
189
		pr_err("pblk: metadata I/O failed. Line %d\n", line->id);
190 191 192 193
	}

	sync = atomic_add_return(rqd->nr_ppas, &emeta->sync);
	if (sync == emeta->nr_entries)
194 195
		pblk_gen_run_ws(pblk, line, NULL, pblk_line_close_ws,
						GFP_ATOMIC, pblk->close_wq);
196

197
	nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
198
	pblk_free_rqd(pblk, rqd, READ);
199 200

	atomic_dec(&pblk->inflight_io);
201 202
}

203
static int pblk_alloc_w_rq(struct pblk *pblk, struct nvm_rq *rqd,
204 205
			   unsigned int nr_secs,
			   nvm_end_io_fn(*end_io))
206 207 208 209 210 211 212 213
{
	struct nvm_tgt_dev *dev = pblk->dev;

	/* Setup write request */
	rqd->opcode = NVM_OP_PWRITE;
	rqd->nr_ppas = nr_secs;
	rqd->flags = pblk_set_progr_mode(pblk, WRITE);
	rqd->private = pblk;
214
	rqd->end_io = end_io;
215 216 217 218 219 220 221 222 223 224 225 226 227

	rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
							&rqd->dma_meta_list);
	if (!rqd->meta_list)
		return -ENOMEM;

	rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size;
	rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size;

	return 0;
}

static int pblk_setup_w_rq(struct pblk *pblk, struct nvm_rq *rqd,
228
			   struct pblk_c_ctx *c_ctx, struct ppa_addr *erase_ppa)
229 230
{
	struct pblk_line_meta *lm = &pblk->lm;
231
	struct pblk_line *e_line = pblk_line_get_erase(pblk);
232 233 234 235 236 237 238
	unsigned int valid = c_ctx->nr_valid;
	unsigned int padded = c_ctx->nr_padded;
	unsigned int nr_secs = valid + padded;
	unsigned long *lun_bitmap;
	int ret = 0;

	lun_bitmap = kzalloc(lm->lun_bitmap_len, GFP_KERNEL);
239 240
	if (!lun_bitmap)
		return -ENOMEM;
241 242
	c_ctx->lun_bitmap = lun_bitmap;

243
	ret = pblk_alloc_w_rq(pblk, rqd, nr_secs, pblk_end_io_write);
244 245
	if (ret) {
		kfree(lun_bitmap);
246
		return ret;
247 248
	}

249
	if (likely(!e_line || !atomic_read(&e_line->left_eblks)))
250 251 252
		pblk_map_rq(pblk, rqd, c_ctx->sentry, lun_bitmap, valid, 0);
	else
		pblk_map_erase_rq(pblk, rqd, c_ctx->sentry, lun_bitmap,
253
							valid, erase_ppa);
254

255
	return 0;
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270
}

int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
			struct pblk_c_ctx *c_ctx)
{
	struct pblk_line_meta *lm = &pblk->lm;
	unsigned long *lun_bitmap;
	int ret;

	lun_bitmap = kzalloc(lm->lun_bitmap_len, GFP_KERNEL);
	if (!lun_bitmap)
		return -ENOMEM;

	c_ctx->lun_bitmap = lun_bitmap;

271
	ret = pblk_alloc_w_rq(pblk, rqd, rqd->nr_ppas, pblk_end_io_write);
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301
	if (ret)
		return ret;

	pblk_map_rq(pblk, rqd, c_ctx->sentry, lun_bitmap, c_ctx->nr_valid, 0);

	rqd->ppa_status = (u64)0;
	rqd->flags = pblk_set_progr_mode(pblk, WRITE);

	return ret;
}

static int pblk_calc_secs_to_sync(struct pblk *pblk, unsigned int secs_avail,
				  unsigned int secs_to_flush)
{
	int secs_to_sync;

	secs_to_sync = pblk_calc_secs(pblk, secs_avail, secs_to_flush);

#ifdef CONFIG_NVM_DEBUG
	if ((!secs_to_sync && secs_to_flush)
			|| (secs_to_sync < 0)
			|| (secs_to_sync > secs_avail && !secs_to_flush)) {
		pr_err("pblk: bad sector calculation (a:%d,s:%d,f:%d)\n",
				secs_avail, secs_to_sync, secs_to_flush);
	}
#endif

	return secs_to_sync;
}

302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
static inline int pblk_valid_meta_ppa(struct pblk *pblk,
				      struct pblk_line *meta_line,
				      struct ppa_addr *ppa_list, int nr_ppas)
{
	struct nvm_tgt_dev *dev = pblk->dev;
	struct nvm_geo *geo = &dev->geo;
	struct pblk_line *data_line;
	struct ppa_addr ppa, ppa_opt;
	u64 paddr;
	int i;

	data_line = &pblk->lines[pblk_dev_ppa_to_line(ppa_list[0])];
	paddr = pblk_lookup_page(pblk, meta_line);
	ppa = addr_to_gen_ppa(pblk, paddr, 0);

	if (test_bit(pblk_ppa_to_pos(geo, ppa), data_line->blk_bitmap))
		return 1;

	/* Schedule a metadata I/O that is half the distance from the data I/O
	 * with regards to the number of LUNs forming the pblk instance. This
	 * balances LUN conflicts across every I/O.
	 *
	 * When the LUN configuration changes (e.g., due to GC), this distance
	 * can align, which would result on a LUN deadlock. In this case, modify
	 * the distance to not be optimal, but allow metadata I/Os to succeed.
	 */
	ppa_opt = addr_to_gen_ppa(pblk, paddr + data_line->meta_distance, 0);
	if (unlikely(ppa_opt.ppa == ppa.ppa)) {
		data_line->meta_distance--;
		return 0;
	}

	for (i = 0; i < nr_ppas; i += pblk->min_write_pgs)
		if (ppa_list[i].g.ch == ppa_opt.g.ch &&
					ppa_list[i].g.lun == ppa_opt.g.lun)
			return 1;

	if (test_bit(pblk_ppa_to_pos(geo, ppa_opt), data_line->blk_bitmap)) {
		for (i = 0; i < nr_ppas; i += pblk->min_write_pgs)
			if (ppa_list[i].g.ch == ppa.g.ch &&
						ppa_list[i].g.lun == ppa.g.lun)
				return 0;

		return 1;
	}

	return 0;
}

int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line)
{
	struct nvm_tgt_dev *dev = pblk->dev;
	struct nvm_geo *geo = &dev->geo;
	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
	struct pblk_line_meta *lm = &pblk->lm;
	struct pblk_emeta *emeta = meta_line->emeta;
	struct pblk_g_ctx *m_ctx;
	struct bio *bio;
	struct nvm_rq *rqd;
	void *data;
	u64 paddr;
	int rq_ppas = pblk->min_write_pgs;
	int id = meta_line->id;
	int rq_len;
	int i, j;
	int ret;

	rqd = pblk_alloc_rqd(pblk, READ);
370

371 372 373 374 375 376
	m_ctx = nvm_rq_to_pdu(rqd);
	m_ctx->private = meta_line;

	rq_len = rq_ppas * geo->sec_size;
	data = ((void *)emeta->buf) + emeta->mem;

377 378
	bio = pblk_bio_map_addr(pblk, data, rq_ppas, rq_len,
					l_mg->emeta_alloc_type, GFP_KERNEL);
379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405
	if (IS_ERR(bio)) {
		ret = PTR_ERR(bio);
		goto fail_free_rqd;
	}
	bio->bi_iter.bi_sector = 0; /* internal bio */
	bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
	rqd->bio = bio;

	ret = pblk_alloc_w_rq(pblk, rqd, rq_ppas, pblk_end_io_write_meta);
	if (ret)
		goto fail_free_bio;

	for (i = 0; i < rqd->nr_ppas; ) {
		spin_lock(&meta_line->lock);
		paddr = __pblk_alloc_page(pblk, meta_line, rq_ppas);
		spin_unlock(&meta_line->lock);
		for (j = 0; j < rq_ppas; j++, i++, paddr++)
			rqd->ppa_list[i] = addr_to_gen_ppa(pblk, paddr, id);
	}

	emeta->mem += rq_len;
	if (emeta->mem >= lm->emeta_len[0]) {
		spin_lock(&l_mg->close_lock);
		list_del(&meta_line->list);
		spin_unlock(&l_mg->close_lock);
	}

406 407
	pblk_down_page(pblk, rqd->ppa_list, rqd->nr_ppas);

408 409 410 411 412 413 414 415 416
	ret = pblk_submit_io(pblk, rqd);
	if (ret) {
		pr_err("pblk: emeta I/O submission failed: %d\n", ret);
		goto fail_rollback;
	}

	return NVM_IO_OK;

fail_rollback:
417
	pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
418 419 420 421
	spin_lock(&l_mg->close_lock);
	pblk_dealloc_page(pblk, meta_line, rq_ppas);
	list_add(&meta_line->list, &meta_line->list);
	spin_unlock(&l_mg->close_lock);
422 423

	nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
424
fail_free_bio:
425
	bio_put(bio);
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444
fail_free_rqd:
	pblk_free_rqd(pblk, rqd, READ);
	return ret;
}

static int pblk_sched_meta_io(struct pblk *pblk, struct ppa_addr *prev_list,
			       int prev_n)
{
	struct pblk_line_meta *lm = &pblk->lm;
	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
	struct pblk_line *meta_line;

	spin_lock(&l_mg->close_lock);
retry:
	if (list_empty(&l_mg->emeta_list)) {
		spin_unlock(&l_mg->close_lock);
		return 0;
	}
	meta_line = list_first_entry(&l_mg->emeta_list, struct pblk_line, list);
445
	if (meta_line->emeta->mem >= lm->emeta_len[0])
446 447 448 449 450 451 452 453 454
		goto retry;
	spin_unlock(&l_mg->close_lock);

	if (!pblk_valid_meta_ppa(pblk, meta_line, prev_list, prev_n))
		return 0;

	return pblk_submit_meta_io(pblk, meta_line);
}

455 456 457 458 459 460 461 462 463 464 465 466 467 468 469
static int pblk_submit_io_set(struct pblk *pblk, struct nvm_rq *rqd)
{
	struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
	struct ppa_addr erase_ppa;
	int err;

	ppa_set_empty(&erase_ppa);

	/* Assign lbas to ppas and populate request structure */
	err = pblk_setup_w_rq(pblk, rqd, c_ctx, &erase_ppa);
	if (err) {
		pr_err("pblk: could not setup write request: %d\n", err);
		return NVM_IO_ERR;
	}

470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490
	if (likely(ppa_empty(erase_ppa))) {
		/* Submit metadata write for previous data line */
		err = pblk_sched_meta_io(pblk, rqd->ppa_list, rqd->nr_ppas);
		if (err) {
			pr_err("pblk: metadata I/O submission failed: %d", err);
			return NVM_IO_ERR;
		}

		/* Submit data write for current data line */
		err = pblk_submit_io(pblk, rqd);
		if (err) {
			pr_err("pblk: data I/O submission failed: %d\n", err);
			return NVM_IO_ERR;
		}
	} else {
		/* Submit data write for current data line */
		err = pblk_submit_io(pblk, rqd);
		if (err) {
			pr_err("pblk: data I/O submission failed: %d\n", err);
			return NVM_IO_ERR;
		}
491

492 493 494 495 496 497 498 499 500 501 502
		/* Submit available erase for next data line */
		if (pblk_blk_erase_async(pblk, erase_ppa)) {
			struct pblk_line *e_line = pblk_line_get_erase(pblk);
			struct nvm_tgt_dev *dev = pblk->dev;
			struct nvm_geo *geo = &dev->geo;
			int bit;

			atomic_inc(&e_line->left_eblks);
			bit = pblk_ppa_to_pos(geo, erase_ppa);
			WARN_ON(!test_and_clear_bit(bit, e_line->erase_bitmap));
		}
503 504 505 506 507 508 509 510 511 512 513
	}

	return NVM_IO_OK;
}

static void pblk_free_write_rqd(struct pblk *pblk, struct nvm_rq *rqd)
{
	struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
	struct bio *bio = rqd->bio;

	if (c_ctx->nr_padded)
514 515
		pblk_bio_free_pages(pblk, bio, c_ctx->nr_valid,
							c_ctx->nr_padded);
516 517
}

518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537
static int pblk_submit_write(struct pblk *pblk)
{
	struct bio *bio;
	struct nvm_rq *rqd;
	unsigned int secs_avail, secs_to_sync, secs_to_com;
	unsigned int secs_to_flush;
	unsigned long pos;

	/* If there are no sectors in the cache, flushes (bios without data)
	 * will be cleared on the cache threads
	 */
	secs_avail = pblk_rb_read_count(&pblk->rwb);
	if (!secs_avail)
		return 1;

	secs_to_flush = pblk_rb_sync_point_count(&pblk->rwb);
	if (!secs_to_flush && secs_avail < pblk->min_write_pgs)
		return 1;

	bio = bio_alloc(GFP_KERNEL, pblk->max_write_pgs);
538

539 540
	bio->bi_iter.bi_sector = 0; /* internal bio */
	bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
541 542

	rqd = pblk_alloc_rqd(pblk, WRITE);
543 544 545 546 547 548 549 550 551 552 553
	rqd->bio = bio;

	secs_to_sync = pblk_calc_secs_to_sync(pblk, secs_avail, secs_to_flush);
	if (secs_to_sync > pblk->max_write_pgs) {
		pr_err("pblk: bad buffer sync calculation\n");
		goto fail_put_bio;
	}

	secs_to_com = (secs_to_sync > secs_avail) ? secs_avail : secs_to_sync;
	pos = pblk_rb_read_commit(&pblk->rwb, secs_to_com);

554 555
	if (pblk_rb_read_to_bio(&pblk->rwb, rqd, bio, pos, secs_to_sync,
								secs_avail)) {
556 557 558 559
		pr_err("pblk: corrupted write bio\n");
		goto fail_put_bio;
	}

560
	if (pblk_submit_io_set(pblk, rqd))
561 562 563 564 565 566 567 568 569
		goto fail_free_bio;

#ifdef CONFIG_NVM_DEBUG
	atomic_long_add(secs_to_sync, &pblk->sub_writes);
#endif

	return 0;

fail_free_bio:
570
	pblk_free_write_rqd(pblk, rqd);
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590
fail_put_bio:
	bio_put(bio);
	pblk_free_rqd(pblk, rqd, WRITE);

	return 1;
}

int pblk_write_ts(void *data)
{
	struct pblk *pblk = data;

	while (!kthread_should_stop()) {
		if (!pblk_submit_write(pblk))
			continue;
		set_current_state(TASK_INTERRUPTIBLE);
		io_schedule();
	}

	return 0;
}