wl.c 43.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
/*
 * Copyright (c) International Business Machines Corp., 2006
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * 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.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
 *
 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
 */

/*
Artem Bityutskiy's avatar
Artem Bityutskiy committed
22
 * UBI wear-leveling sub-system.
23
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
24 25 26 27 28 29
 * This sub-system is responsible for wear-leveling. It works in terms of
 * physical* eraseblocks and erase counters and knows nothing about logical
 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
 * eraseblocks are of two types - used and free. Used physical eraseblocks are
 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
30 31
 *
 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
Artem Bityutskiy's avatar
Artem Bityutskiy committed
32
 * header. The rest of the physical eraseblock contains only %0xFF bytes.
33
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
34
 * When physical eraseblocks are returned to the WL sub-system by means of the
35 36
 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
 * done asynchronously in context of the per-UBI device background thread,
Artem Bityutskiy's avatar
Artem Bityutskiy committed
37
 * which is also managed by the WL sub-system.
38 39 40 41 42 43 44 45
 *
 * The wear-leveling is ensured by means of moving the contents of used
 * physical eraseblocks with low erase counter to free physical eraseblocks
 * with high erase counter.
 *
 * The 'ubi_wl_get_peb()' function accepts data type hints which help to pick
 * an "optimal" physical eraseblock. For example, when it is known that the
 * physical eraseblock will be "put" soon because it contains short-term data,
Artem Bityutskiy's avatar
Artem Bityutskiy committed
46 47
 * the WL sub-system may pick a free physical eraseblock with low erase
 * counter, and so forth.
48
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
49 50
 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
 * bad.
51
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
52 53 54
 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
 * in a physical eraseblock, it has to be moved. Technically this is the same
 * as moving it for wear-leveling reasons.
55
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
56 57 58
 * As it was said, for the UBI sub-system all physical eraseblocks are either
 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
 * used eraseblocks are kept in a set of different RB-trees: @wl->used,
59 60 61 62 63
 * @wl->prot.pnum, @wl->prot.aec, and @wl->scrub.
 *
 * Note, in this implementation, we keep a small in-RAM object for each physical
 * eraseblock. This is surely not a scalable solution. But it appears to be good
 * enough for moderately large flashes and it is simple. In future, one may
Artem Bityutskiy's avatar
Artem Bityutskiy committed
64
 * re-work this sub-system and make it more scalable.
65
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
66 67 68
 * At the moment this sub-system does not utilize the sequence number, which
 * was introduced relatively recently. But it would be wise to do this because
 * the sequence number of a logical eraseblock characterizes how old is it. For
69 70 71
 * example, when we move a PEB with low erase counter, and we need to pick the
 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
 * pick target PEB with an average EC if our PEB is not very "old". This is a
Artem Bityutskiy's avatar
Artem Bityutskiy committed
72
 * room for future re-works of the WL sub-system.
73
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
74 75
 * Note: the stuff with protection trees looks too complex and is difficult to
 * understand. Should be fixed.
76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96
 */

#include <linux/slab.h>
#include <linux/crc32.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
#include "ubi.h"

/* Number of physical eraseblocks reserved for wear-leveling purposes */
#define WL_RESERVED_PEBS 1

/*
 * How many erase cycles are short term, unknown, and long term physical
 * eraseblocks protected.
 */
#define ST_PROTECTION 16
#define U_PROTECTION  10
#define LT_PROTECTION 4

/*
 * Maximum difference between two erase counters. If this threshold is
Artem Bityutskiy's avatar
Artem Bityutskiy committed
97 98 99
 * exceeded, the WL sub-system starts moving data from used physical
 * eraseblocks with low erase counter to free physical eraseblocks with high
 * erase counter.
100 101 102 103
 */
#define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD

/*
Artem Bityutskiy's avatar
Artem Bityutskiy committed
104
 * When a physical eraseblock is moved, the WL sub-system has to pick the target
105 106 107 108 109
 * physical eraseblock to move to. The simplest way would be just to pick the
 * one with the highest erase counter. But in certain workloads this could lead
 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
 * situation when the picked physical eraseblock is constantly erased after the
 * data is written to it. So, we have a constant which limits the highest erase
Artem Bityutskiy's avatar
Artem Bityutskiy committed
110 111
 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
 * does not pick eraseblocks with erase counter greater then the lowest erase
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
 * counter plus %WL_FREE_MAX_DIFF.
 */
#define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)

/*
 * Maximum number of consecutive background thread failures which is enough to
 * switch to read-only mode.
 */
#define WL_MAX_FAILURES 32

/**
 * struct ubi_wl_prot_entry - PEB protection entry.
 * @rb_pnum: link in the @wl->prot.pnum RB-tree
 * @rb_aec: link in the @wl->prot.aec RB-tree
 * @abs_ec: the absolute erase counter value when the protection ends
 * @e: the wear-leveling entry of the physical eraseblock under protection
 *
Artem Bityutskiy's avatar
Artem Bityutskiy committed
129 130 131 132 133
 * When the WL sub-system returns a physical eraseblock, the physical
 * eraseblock is protected from being moved for some "time". For this reason,
 * the physical eraseblock is not directly moved from the @wl->free tree to the
 * @wl->used tree. There is one more tree in between where this physical
 * eraseblock is temporarily stored (@wl->prot).
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
 *
 * All this protection stuff is needed because:
 *  o we don't want to move physical eraseblocks just after we have given them
 *    to the user; instead, we first want to let users fill them up with data;
 *
 *  o there is a chance that the user will put the physical eraseblock very
 *    soon, so it makes sense not to move it for some time, but wait; this is
 *    especially important in case of "short term" physical eraseblocks.
 *
 * Physical eraseblocks stay protected only for limited time. But the "time" is
 * measured in erase cycles in this case. This is implemented with help of the
 * absolute erase counter (@wl->abs_ec). When it reaches certain value, the
 * physical eraseblocks are moved from the protection trees (@wl->prot.*) to
 * the @wl->used tree.
 *
 * Protected physical eraseblocks are searched by physical eraseblock number
 * (when they are put) and by the absolute erase counter (to check if it is
 * time to move them to the @wl->used tree). So there are actually 2 RB-trees
 * storing the protected physical eraseblocks: @wl->prot.pnum and
 * @wl->prot.aec. They are referred to as the "protection" trees. The
 * first one is indexed by the physical eraseblock number. The second one is
 * indexed by the absolute erase counter. Both trees store
 * &struct ubi_wl_prot_entry objects.
 *
 * Each physical eraseblock has 2 main states: free and used. The former state
 * corresponds to the @wl->free tree. The latter state is split up on several
 * sub-states:
 * o the WL movement is allowed (@wl->used tree);
 * o the WL movement is temporarily prohibited (@wl->prot.pnum and
 * @wl->prot.aec trees);
 * o scrubbing is needed (@wl->scrub tree).
 *
 * Depending on the sub-state, wear-leveling entries of the used physical
 * eraseblocks may be kept in one of those trees.
 */
struct ubi_wl_prot_entry {
	struct rb_node rb_pnum;
	struct rb_node rb_aec;
	unsigned long long abs_ec;
	struct ubi_wl_entry *e;
};

/**
 * struct ubi_work - UBI work description data structure.
 * @list: a link in the list of pending works
 * @func: worker function
 * @priv: private data of the worker function
 * @e: physical eraseblock to erase
 * @torture: if the physical eraseblock has to be tortured
 *
 * The @func pointer points to the worker function. If the @cancel argument is
 * not zero, the worker has to free the resources and exit immediately. The
 * worker has to return zero in case of success and a negative error code in
 * case of failure.
 */
struct ubi_work {
	struct list_head list;
	int (*func)(struct ubi_device *ubi, struct ubi_work *wrk, int cancel);
	/* The below fields are only relevant to erasure works */
	struct ubi_wl_entry *e;
	int torture;
};

#ifdef CONFIG_MTD_UBI_DEBUG_PARANOID
198
static int paranoid_check_ec(struct ubi_device *ubi, int pnum, int ec);
199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253
static int paranoid_check_in_wl_tree(struct ubi_wl_entry *e,
				     struct rb_root *root);
#else
#define paranoid_check_ec(ubi, pnum, ec) 0
#define paranoid_check_in_wl_tree(e, root)
#endif

/**
 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
 * @e: the wear-leveling entry to add
 * @root: the root of the tree
 *
 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
 * the @ubi->used and @ubi->free RB-trees.
 */
static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
{
	struct rb_node **p, *parent = NULL;

	p = &root->rb_node;
	while (*p) {
		struct ubi_wl_entry *e1;

		parent = *p;
		e1 = rb_entry(parent, struct ubi_wl_entry, rb);

		if (e->ec < e1->ec)
			p = &(*p)->rb_left;
		else if (e->ec > e1->ec)
			p = &(*p)->rb_right;
		else {
			ubi_assert(e->pnum != e1->pnum);
			if (e->pnum < e1->pnum)
				p = &(*p)->rb_left;
			else
				p = &(*p)->rb_right;
		}
	}

	rb_link_node(&e->rb, parent, p);
	rb_insert_color(&e->rb, root);
}

/**
 * do_work - do one pending work.
 * @ubi: UBI device description object
 *
 * This function returns zero in case of success and a negative error code in
 * case of failure.
 */
static int do_work(struct ubi_device *ubi)
{
	int err;
	struct ubi_work *wrk;

254 255
	cond_resched();

Artem Bityutskiy's avatar
Artem Bityutskiy committed
256 257 258 259 260 261 262
	/*
	 * @ubi->work_sem is used to synchronize with the workers. Workers take
	 * it in read mode, so many of them may be doing works at a time. But
	 * the queue flush code has to be sure the whole queue of works is
	 * done, and it takes the mutex in write mode.
	 */
	down_read(&ubi->work_sem);
263 264 265
	spin_lock(&ubi->wl_lock);
	if (list_empty(&ubi->works)) {
		spin_unlock(&ubi->wl_lock);
Artem Bityutskiy's avatar
Artem Bityutskiy committed
266
		up_read(&ubi->work_sem);
267 268 269 270 271
		return 0;
	}

	wrk = list_entry(ubi->works.next, struct ubi_work, list);
	list_del(&wrk->list);
Artem Bityutskiy's avatar
Artem Bityutskiy committed
272 273
	ubi->works_count -= 1;
	ubi_assert(ubi->works_count >= 0);
274 275 276 277 278 279 280 281 282 283
	spin_unlock(&ubi->wl_lock);

	/*
	 * Call the worker function. Do not touch the work structure
	 * after this call as it will have been freed or reused by that
	 * time by the worker function.
	 */
	err = wrk->func(ubi, wrk, 0);
	if (err)
		ubi_err("work failed with error code %d", err);
Artem Bityutskiy's avatar
Artem Bityutskiy committed
284
	up_read(&ubi->work_sem);
Artem Bityutskiy's avatar
Artem Bityutskiy committed
285

286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
	return err;
}

/**
 * produce_free_peb - produce a free physical eraseblock.
 * @ubi: UBI device description object
 *
 * This function tries to make a free PEB by means of synchronous execution of
 * pending works. This may be needed if, for example the background thread is
 * disabled. Returns zero in case of success and a negative error code in case
 * of failure.
 */
static int produce_free_peb(struct ubi_device *ubi)
{
	int err;

	spin_lock(&ubi->wl_lock);
303
	while (!ubi->free.rb_node) {
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452
		spin_unlock(&ubi->wl_lock);

		dbg_wl("do one work synchronously");
		err = do_work(ubi);
		if (err)
			return err;

		spin_lock(&ubi->wl_lock);
	}
	spin_unlock(&ubi->wl_lock);

	return 0;
}

/**
 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
 * @e: the wear-leveling entry to check
 * @root: the root of the tree
 *
 * This function returns non-zero if @e is in the @root RB-tree and zero if it
 * is not.
 */
static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
{
	struct rb_node *p;

	p = root->rb_node;
	while (p) {
		struct ubi_wl_entry *e1;

		e1 = rb_entry(p, struct ubi_wl_entry, rb);

		if (e->pnum == e1->pnum) {
			ubi_assert(e == e1);
			return 1;
		}

		if (e->ec < e1->ec)
			p = p->rb_left;
		else if (e->ec > e1->ec)
			p = p->rb_right;
		else {
			ubi_assert(e->pnum != e1->pnum);
			if (e->pnum < e1->pnum)
				p = p->rb_left;
			else
				p = p->rb_right;
		}
	}

	return 0;
}

/**
 * prot_tree_add - add physical eraseblock to protection trees.
 * @ubi: UBI device description object
 * @e: the physical eraseblock to add
 * @pe: protection entry object to use
 * @abs_ec: absolute erase counter value when this physical eraseblock has
 * to be removed from the protection trees.
 *
 * @wl->lock has to be locked.
 */
static void prot_tree_add(struct ubi_device *ubi, struct ubi_wl_entry *e,
			  struct ubi_wl_prot_entry *pe, int abs_ec)
{
	struct rb_node **p, *parent = NULL;
	struct ubi_wl_prot_entry *pe1;

	pe->e = e;
	pe->abs_ec = ubi->abs_ec + abs_ec;

	p = &ubi->prot.pnum.rb_node;
	while (*p) {
		parent = *p;
		pe1 = rb_entry(parent, struct ubi_wl_prot_entry, rb_pnum);

		if (e->pnum < pe1->e->pnum)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}
	rb_link_node(&pe->rb_pnum, parent, p);
	rb_insert_color(&pe->rb_pnum, &ubi->prot.pnum);

	p = &ubi->prot.aec.rb_node;
	parent = NULL;
	while (*p) {
		parent = *p;
		pe1 = rb_entry(parent, struct ubi_wl_prot_entry, rb_aec);

		if (pe->abs_ec < pe1->abs_ec)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}
	rb_link_node(&pe->rb_aec, parent, p);
	rb_insert_color(&pe->rb_aec, &ubi->prot.aec);
}

/**
 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
 * @root: the RB-tree where to look for
 * @max: highest possible erase counter
 *
 * This function looks for a wear leveling entry with erase counter closest to
 * @max and less then @max.
 */
static struct ubi_wl_entry *find_wl_entry(struct rb_root *root, int max)
{
	struct rb_node *p;
	struct ubi_wl_entry *e;

	e = rb_entry(rb_first(root), struct ubi_wl_entry, rb);
	max += e->ec;

	p = root->rb_node;
	while (p) {
		struct ubi_wl_entry *e1;

		e1 = rb_entry(p, struct ubi_wl_entry, rb);
		if (e1->ec >= max)
			p = p->rb_left;
		else {
			p = p->rb_right;
			e = e1;
		}
	}

	return e;
}

/**
 * ubi_wl_get_peb - get a physical eraseblock.
 * @ubi: UBI device description object
 * @dtype: type of data which will be stored in this physical eraseblock
 *
 * This function returns a physical eraseblock in case of success and a
 * negative error code in case of failure. Might sleep.
 */
int ubi_wl_get_peb(struct ubi_device *ubi, int dtype)
{
	int err, protect, medium_ec;
	struct ubi_wl_entry *e, *first, *last;
	struct ubi_wl_prot_entry *pe;

	ubi_assert(dtype == UBI_LONGTERM || dtype == UBI_SHORTTERM ||
		   dtype == UBI_UNKNOWN);

453
	pe = kmalloc(sizeof(struct ubi_wl_prot_entry), GFP_NOFS);
454 455 456 457 458
	if (!pe)
		return -ENOMEM;

retry:
	spin_lock(&ubi->wl_lock);
459
	if (!ubi->free.rb_node) {
460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477
		if (ubi->works_count == 0) {
			ubi_assert(list_empty(&ubi->works));
			ubi_err("no free eraseblocks");
			spin_unlock(&ubi->wl_lock);
			kfree(pe);
			return -ENOSPC;
		}
		spin_unlock(&ubi->wl_lock);

		err = produce_free_peb(ubi);
		if (err < 0) {
			kfree(pe);
			return err;
		}
		goto retry;
	}

	switch (dtype) {
478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
	case UBI_LONGTERM:
		/*
		 * For long term data we pick a physical eraseblock with high
		 * erase counter. But the highest erase counter we can pick is
		 * bounded by the the lowest erase counter plus
		 * %WL_FREE_MAX_DIFF.
		 */
		e = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF);
		protect = LT_PROTECTION;
		break;
	case UBI_UNKNOWN:
		/*
		 * For unknown data we pick a physical eraseblock with medium
		 * erase counter. But we by no means can pick a physical
		 * eraseblock with erase counter greater or equivalent than the
		 * lowest erase counter plus %WL_FREE_MAX_DIFF.
		 */
		first = rb_entry(rb_first(&ubi->free), struct ubi_wl_entry, rb);
		last = rb_entry(rb_last(&ubi->free), struct ubi_wl_entry, rb);
497

498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
		if (last->ec - first->ec < WL_FREE_MAX_DIFF)
			e = rb_entry(ubi->free.rb_node,
					struct ubi_wl_entry, rb);
		else {
			medium_ec = (first->ec + WL_FREE_MAX_DIFF)/2;
			e = find_wl_entry(&ubi->free, medium_ec);
		}
		protect = U_PROTECTION;
		break;
	case UBI_SHORTTERM:
		/*
		 * For short term data we pick a physical eraseblock with the
		 * lowest erase counter as we expect it will be erased soon.
		 */
		e = rb_entry(rb_first(&ubi->free), struct ubi_wl_entry, rb);
		protect = ST_PROTECTION;
		break;
	default:
		protect = 0;
		e = NULL;
		BUG();
519 520 521 522 523 524
	}

	/*
	 * Move the physical eraseblock to the protection trees where it will
	 * be protected from being moved for some time.
	 */
525 526
	paranoid_check_in_wl_tree(e, &ubi->free);
	rb_erase(&e->rb, &ubi->free);
527 528 529 530 531 532 533 534 535 536 537 538
	prot_tree_add(ubi, e, pe, protect);

	dbg_wl("PEB %d EC %d, protection %d", e->pnum, e->ec, protect);
	spin_unlock(&ubi->wl_lock);

	return e->pnum;
}

/**
 * prot_tree_del - remove a physical eraseblock from the protection trees
 * @ubi: UBI device description object
 * @pnum: the physical eraseblock to remove
539 540 541 542
 *
 * This function returns PEB @pnum from the protection trees and returns zero
 * in case of success and %-ENODEV if the PEB was not found in the protection
 * trees.
543
 */
544
static int prot_tree_del(struct ubi_device *ubi, int pnum)
545 546 547 548 549 550 551 552 553 554
{
	struct rb_node *p;
	struct ubi_wl_prot_entry *pe = NULL;

	p = ubi->prot.pnum.rb_node;
	while (p) {

		pe = rb_entry(p, struct ubi_wl_prot_entry, rb_pnum);

		if (pnum == pe->e->pnum)
555
			goto found;
556 557 558 559 560 561 562

		if (pnum < pe->e->pnum)
			p = p->rb_left;
		else
			p = p->rb_right;
	}

563 564 565
	return -ENODEV;

found:
566 567 568 569
	ubi_assert(pe->e->pnum == pnum);
	rb_erase(&pe->rb_aec, &ubi->prot.aec);
	rb_erase(&pe->rb_pnum, &ubi->prot.pnum);
	kfree(pe);
570
	return 0;
571 572 573 574 575 576 577 578 579 580 581
}

/**
 * sync_erase - synchronously erase a physical eraseblock.
 * @ubi: UBI device description object
 * @e: the the physical eraseblock to erase
 * @torture: if the physical eraseblock has to be tortured
 *
 * This function returns zero in case of success and a negative error code in
 * case of failure.
 */
582 583
static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
		      int torture)
584 585 586 587 588 589 590 591 592 593 594
{
	int err;
	struct ubi_ec_hdr *ec_hdr;
	unsigned long long ec = e->ec;

	dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);

	err = paranoid_check_ec(ubi, e->pnum, e->ec);
	if (err > 0)
		return -EINVAL;

595
	ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616
	if (!ec_hdr)
		return -ENOMEM;

	err = ubi_io_sync_erase(ubi, e->pnum, torture);
	if (err < 0)
		goto out_free;

	ec += err;
	if (ec > UBI_MAX_ERASECOUNTER) {
		/*
		 * Erase counter overflow. Upgrade UBI and use 64-bit
		 * erase counters internally.
		 */
		ubi_err("erase counter overflow at PEB %d, EC %llu",
			e->pnum, ec);
		err = -EINVAL;
		goto out_free;
	}

	dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);

617
	ec_hdr->ec = cpu_to_be64(ec);
618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634

	err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
	if (err)
		goto out_free;

	e->ec = ec;
	spin_lock(&ubi->wl_lock);
	if (e->ec > ubi->max_ec)
		ubi->max_ec = e->ec;
	spin_unlock(&ubi->wl_lock);

out_free:
	kfree(ec_hdr);
	return err;
}

/**
635
 * check_protection_over - check if it is time to stop protecting some PEBs.
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
 * @ubi: UBI device description object
 *
 * This function is called after each erase operation, when the absolute erase
 * counter is incremented, to check if some physical eraseblock  have not to be
 * protected any longer. These physical eraseblocks are moved from the
 * protection trees to the used tree.
 */
static void check_protection_over(struct ubi_device *ubi)
{
	struct ubi_wl_prot_entry *pe;

	/*
	 * There may be several protected physical eraseblock to remove,
	 * process them all.
	 */
	while (1) {
		spin_lock(&ubi->wl_lock);
653
		if (!ubi->prot.aec.rb_node) {
654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669
			spin_unlock(&ubi->wl_lock);
			break;
		}

		pe = rb_entry(rb_first(&ubi->prot.aec),
			      struct ubi_wl_prot_entry, rb_aec);

		if (pe->abs_ec > ubi->abs_ec) {
			spin_unlock(&ubi->wl_lock);
			break;
		}

		dbg_wl("PEB %d protection over, abs_ec %llu, PEB abs_ec %llu",
		       pe->e->pnum, ubi->abs_ec, pe->abs_ec);
		rb_erase(&pe->rb_aec, &ubi->prot.aec);
		rb_erase(&pe->rb_pnum, &ubi->prot.pnum);
670
		wl_tree_add(pe->e, &ubi->used);
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716
		spin_unlock(&ubi->wl_lock);

		kfree(pe);
		cond_resched();
	}
}

/**
 * schedule_ubi_work - schedule a work.
 * @ubi: UBI device description object
 * @wrk: the work to schedule
 *
 * This function enqueues a work defined by @wrk to the tail of the pending
 * works list.
 */
static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
{
	spin_lock(&ubi->wl_lock);
	list_add_tail(&wrk->list, &ubi->works);
	ubi_assert(ubi->works_count >= 0);
	ubi->works_count += 1;
	if (ubi->thread_enabled)
		wake_up_process(ubi->bgt_thread);
	spin_unlock(&ubi->wl_lock);
}

static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
			int cancel);

/**
 * schedule_erase - schedule an erase work.
 * @ubi: UBI device description object
 * @e: the WL entry of the physical eraseblock to erase
 * @torture: if the physical eraseblock has to be tortured
 *
 * This function returns zero in case of success and a %-ENOMEM in case of
 * failure.
 */
static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
			  int torture)
{
	struct ubi_work *wl_wrk;

	dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
	       e->pnum, e->ec, torture);

717
	wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741
	if (!wl_wrk)
		return -ENOMEM;

	wl_wrk->func = &erase_worker;
	wl_wrk->e = e;
	wl_wrk->torture = torture;

	schedule_ubi_work(ubi, wl_wrk);
	return 0;
}

/**
 * wear_leveling_worker - wear-leveling worker function.
 * @ubi: UBI device description object
 * @wrk: the work object
 * @cancel: non-zero if the worker has to free memory and exit
 *
 * This function copies a more worn out physical eraseblock to a less worn out
 * one. Returns zero in case of success and a negative error code in case of
 * failure.
 */
static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
				int cancel)
{
742
	int err, put = 0, scrubbing = 0, protect = 0;
Artem Bityutskiy's avatar
Artem Bityutskiy committed
743
	struct ubi_wl_prot_entry *uninitialized_var(pe);
744 745 746 747 748 749 750 751
	struct ubi_wl_entry *e1, *e2;
	struct ubi_vid_hdr *vid_hdr;

	kfree(wrk);

	if (cancel)
		return 0;

752
	vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
753 754 755
	if (!vid_hdr)
		return -ENOMEM;

756
	mutex_lock(&ubi->move_mutex);
757
	spin_lock(&ubi->wl_lock);
758 759
	ubi_assert(!ubi->move_from && !ubi->move_to);
	ubi_assert(!ubi->move_to_put);
760

761
	if (!ubi->free.rb_node ||
762
	    (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
763
		/*
764 765 766
		 * No free physical eraseblocks? Well, they must be waiting in
		 * the queue to be erased. Cancel movement - it will be
		 * triggered again when a free physical eraseblock appears.
767 768 769 770 771 772 773
		 *
		 * No used physical eraseblocks? They must be temporarily
		 * protected from being moved. They will be moved to the
		 * @ubi->used tree later and the wear-leveling will be
		 * triggered again.
		 */
		dbg_wl("cancel WL, a list is empty: free %d, used %d",
774
		       !ubi->free.rb_node, !ubi->used.rb_node);
775
		goto out_cancel;
776 777
	}

778
	if (!ubi->scrub.rb_node) {
779 780 781 782 783 784 785 786 787 788 789
		/*
		 * Now pick the least worn-out used physical eraseblock and a
		 * highly worn-out free physical eraseblock. If the erase
		 * counters differ much enough, start wear-leveling.
		 */
		e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, rb);
		e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF);

		if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
			dbg_wl("no WL needed: min used EC %d, max free EC %d",
			       e1->ec, e2->ec);
790
			goto out_cancel;
791
		}
792 793
		paranoid_check_in_wl_tree(e1, &ubi->used);
		rb_erase(&e1->rb, &ubi->used);
794 795 796
		dbg_wl("move PEB %d EC %d to PEB %d EC %d",
		       e1->pnum, e1->ec, e2->pnum, e2->ec);
	} else {
797 798
		/* Perform scrubbing */
		scrubbing = 1;
799 800
		e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, rb);
		e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF);
801
		paranoid_check_in_wl_tree(e1, &ubi->scrub);
Artem Bityutskiy's avatar
Artem Bityutskiy committed
802
		rb_erase(&e1->rb, &ubi->scrub);
803 804 805
		dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
	}

806 807
	paranoid_check_in_wl_tree(e2, &ubi->free);
	rb_erase(&e2->rb, &ubi->free);
808 809 810 811 812 813 814 815 816
	ubi->move_from = e1;
	ubi->move_to = e2;
	spin_unlock(&ubi->wl_lock);

	/*
	 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
	 * We so far do not know which logical eraseblock our physical
	 * eraseblock (@e1) belongs to. We have to read the volume identifier
	 * header first.
817 818 819 820
	 *
	 * Note, we are protected from this PEB being unmapped and erased. The
	 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
	 * which is being moved was unmapped.
821 822 823 824 825 826 827 828 829 830 831 832 833 834
	 */

	err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
	if (err && err != UBI_IO_BITFLIPS) {
		if (err == UBI_IO_PEB_FREE) {
			/*
			 * We are trying to move PEB without a VID header. UBI
			 * always write VID headers shortly after the PEB was
			 * given, so we have a situation when it did not have
			 * chance to write it down because it was preempted.
			 * Just re-schedule the work, so that next time it will
			 * likely have the VID header in place.
			 */
			dbg_wl("PEB %d has no VID header", e1->pnum);
835
			goto out_not_moved;
836
		}
837 838 839 840 841 842

		ubi_err("error %d while reading VID header from PEB %d",
			err, e1->pnum);
		if (err > 0)
			err = -EIO;
		goto out_error;
843 844 845 846
	}

	err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
	if (err) {
847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867

		if (err < 0)
			goto out_error;
		if (err == 1)
			goto out_not_moved;

		/*
		 * For some reason the LEB was not moved - it might be because
		 * the volume is being deleted. We should prevent this PEB from
		 * being selected for wear-levelling movement for some "time",
		 * so put it to the protection tree.
		 */

		dbg_wl("cancelled moving PEB %d", e1->pnum);
		pe = kmalloc(sizeof(struct ubi_wl_prot_entry), GFP_NOFS);
		if (!pe) {
			err = -ENOMEM;
			goto out_error;
		}

		protect = 1;
868 869 870
	}

	ubi_free_vid_hdr(ubi, vid_hdr);
871 872 873 874
	if (scrubbing && !protect)
		ubi_msg("scrubbed PEB %d, data moved to PEB %d",
			e1->pnum, e2->pnum);

875
	spin_lock(&ubi->wl_lock);
876 877
	if (protect)
		prot_tree_add(ubi, e1, pe, protect);
878
	if (!ubi->move_to_put)
879
		wl_tree_add(e2, &ubi->used);
880 881 882
	else
		put = 1;
	ubi->move_from = ubi->move_to = NULL;
883
	ubi->move_to_put = ubi->wl_scheduled = 0;
884 885 886 887 888 889 890 891 892
	spin_unlock(&ubi->wl_lock);

	if (put) {
		/*
		 * Well, the target PEB was put meanwhile, schedule it for
		 * erasure.
		 */
		dbg_wl("PEB %d was put meanwhile, erase", e2->pnum);
		err = schedule_erase(ubi, e2, 0);
893 894
		if (err)
			goto out_error;
895 896
	}

897 898 899 900
	if (!protect) {
		err = schedule_erase(ubi, e1, 0);
		if (err)
			goto out_error;
901 902
	}

903

904
	dbg_wl("done");
905 906
	mutex_unlock(&ubi->move_mutex);
	return 0;
907 908

	/*
909 910 911
	 * For some reasons the LEB was not moved, might be an error, might be
	 * something else. @e1 was not changed, so return it back. @e2 might
	 * be changed, schedule it for erasure.
912
	 */
913
out_not_moved:
914 915
	ubi_free_vid_hdr(ubi, vid_hdr);
	spin_lock(&ubi->wl_lock);
916 917
	if (scrubbing)
		wl_tree_add(e1, &ubi->scrub);
918
	else
919
		wl_tree_add(e1, &ubi->used);
920
	ubi->move_from = ubi->move_to = NULL;
921
	ubi->move_to_put = ubi->wl_scheduled = 0;
922 923 924
	spin_unlock(&ubi->wl_lock);

	err = schedule_erase(ubi, e2, 0);
925 926 927 928 929 930 931 932 933
	if (err)
		goto out_error;

	mutex_unlock(&ubi->move_mutex);
	return 0;

out_error:
	ubi_err("error %d while moving PEB %d to PEB %d",
		err, e1->pnum, e2->pnum);
934

935 936 937 938 939 940 941 942 943 944 945
	ubi_free_vid_hdr(ubi, vid_hdr);
	spin_lock(&ubi->wl_lock);
	ubi->move_from = ubi->move_to = NULL;
	ubi->move_to_put = ubi->wl_scheduled = 0;
	spin_unlock(&ubi->wl_lock);

	kmem_cache_free(ubi_wl_entry_slab, e1);
	kmem_cache_free(ubi_wl_entry_slab, e2);
	ubi_ro_mode(ubi);

	mutex_unlock(&ubi->move_mutex);
946
	return err;
947 948 949 950 951 952 953

out_cancel:
	ubi->wl_scheduled = 0;
	spin_unlock(&ubi->wl_lock);
	mutex_unlock(&ubi->move_mutex);
	ubi_free_vid_hdr(ubi, vid_hdr);
	return 0;
954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979
}

/**
 * ensure_wear_leveling - schedule wear-leveling if it is needed.
 * @ubi: UBI device description object
 *
 * This function checks if it is time to start wear-leveling and schedules it
 * if yes. This function returns zero in case of success and a negative error
 * code in case of failure.
 */
static int ensure_wear_leveling(struct ubi_device *ubi)
{
	int err = 0;
	struct ubi_wl_entry *e1;
	struct ubi_wl_entry *e2;
	struct ubi_work *wrk;

	spin_lock(&ubi->wl_lock);
	if (ubi->wl_scheduled)
		/* Wear-leveling is already in the work queue */
		goto out_unlock;

	/*
	 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
	 * the WL worker has to be scheduled anyway.
	 */
980 981
	if (!ubi->scrub.rb_node) {
		if (!ubi->used.rb_node || !ubi->free.rb_node)
982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002
			/* No physical eraseblocks - no deal */
			goto out_unlock;

		/*
		 * We schedule wear-leveling only if the difference between the
		 * lowest erase counter of used physical eraseblocks and a high
		 * erase counter of free physical eraseblocks is greater then
		 * %UBI_WL_THRESHOLD.
		 */
		e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, rb);
		e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF);

		if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
			goto out_unlock;
		dbg_wl("schedule wear-leveling");
	} else
		dbg_wl("schedule scrubbing");

	ubi->wl_scheduled = 1;
	spin_unlock(&ubi->wl_lock);

1003
	wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
	if (!wrk) {
		err = -ENOMEM;
		goto out_cancel;
	}

	wrk->func = &wear_leveling_worker;
	schedule_ubi_work(ubi, wrk);
	return err;

out_cancel:
	spin_lock(&ubi->wl_lock);
	ubi->wl_scheduled = 0;
out_unlock:
	spin_unlock(&ubi->wl_lock);
	return err;
}

/**
 * erase_worker - physical eraseblock erase worker function.
 * @ubi: UBI device description object
 * @wl_wrk: the work object
 * @cancel: non-zero if the worker has to free memory and exit
 *
 * This function erases a physical eraseblock and perform torture testing if
 * needed. It also takes care about marking the physical eraseblock bad if
 * needed. Returns zero in case of success and a negative error code in case of
 * failure.
 */
static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
			int cancel)
{
	struct ubi_wl_entry *e = wl_wrk->e;
1036
	int pnum = e->pnum, err, need;
1037 1038 1039 1040

	if (cancel) {
		dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
		kfree(wl_wrk);
1041
		kmem_cache_free(ubi_wl_entry_slab, e);
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053
		return 0;
	}

	dbg_wl("erase PEB %d EC %d", pnum, e->ec);

	err = sync_erase(ubi, e, wl_wrk->torture);
	if (!err) {
		/* Fine, we've erased it successfully */
		kfree(wl_wrk);

		spin_lock(&ubi->wl_lock);
		ubi->abs_ec += 1;
1054
		wl_tree_add(e, &ubi->free);
1055 1056 1057
		spin_unlock(&ubi->wl_lock);

		/*
1058 1059
		 * One more erase operation has happened, take care about
		 * protected physical eraseblocks.
1060 1061 1062 1063 1064 1065 1066 1067
		 */
		check_protection_over(ubi);

		/* And take care about wear-leveling */
		err = ensure_wear_leveling(ubi);
		return err;
	}

Artem Bityutskiy's avatar
Artem Bityutskiy committed
1068
	ubi_err("failed to erase PEB %d, error %d", pnum, err);
1069
	kfree(wl_wrk);
1070
	kmem_cache_free(ubi_wl_entry_slab, e);
1071

1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
	if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
	    err == -EBUSY) {
		int err1;

		/* Re-schedule the LEB for erasure */
		err1 = schedule_erase(ubi, e, 0);
		if (err1) {
			err = err1;
			goto out_ro;
		}
		return err;
	} else if (err != -EIO) {
1084 1085 1086 1087 1088
		/*
		 * If this is not %-EIO, we have no idea what to do. Scheduling
		 * this physical eraseblock for erasure again would cause
		 * errors again and again. Well, lets switch to RO mode.
		 */
1089
		goto out_ro;
1090 1091 1092 1093 1094 1095
	}

	/* It is %-EIO, the PEB went bad */

	if (!ubi->bad_allowed) {
		ubi_err("bad physical eraseblock %d detected", pnum);
1096 1097
		goto out_ro;
	}
1098

1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
	spin_lock(&ubi->volumes_lock);
	need = ubi->beb_rsvd_level - ubi->beb_rsvd_pebs + 1;
	if (need > 0) {
		need = ubi->avail_pebs >= need ? need : ubi->avail_pebs;
		ubi->avail_pebs -= need;
		ubi->rsvd_pebs += need;
		ubi->beb_rsvd_pebs += need;
		if (need > 0)
			ubi_msg("reserve more %d PEBs", need);
	}
1109

1110
	if (ubi->beb_rsvd_pebs == 0) {
1111
		spin_unlock(&ubi->volumes_lock);
1112 1113 1114
		ubi_err("no reserved physical eraseblocks");
		goto out_ro;
	}
1115

1116 1117
	spin_unlock(&ubi->volumes_lock);
	ubi_msg("mark PEB %d as bad", pnum);
1118

1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
	err = ubi_io_mark_bad(ubi, pnum);
	if (err)
		goto out_ro;

	spin_lock(&ubi->volumes_lock);
	ubi->beb_rsvd_pebs -= 1;
	ubi->bad_peb_count += 1;
	ubi->good_peb_count -= 1;
	ubi_calculate_reserved(ubi);
	if (ubi->beb_rsvd_pebs == 0)
		ubi_warn("last PEB from the reserved pool was used");
	spin_unlock(&ubi->volumes_lock);

	return err;
1133

1134 1135
out_ro:
	ubi_ro_mode(ubi);
1136 1137 1138 1139
	return err;
}

/**
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1140
 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1141 1142 1143 1144 1145 1146 1147
 * @ubi: UBI device description object
 * @pnum: physical eraseblock to return
 * @torture: if this physical eraseblock has to be tortured
 *
 * This function is called to return physical eraseblock @pnum to the pool of
 * free physical eraseblocks. The @torture flag has to be set if an I/O error
 * occurred to this @pnum and it has to be tested. This function returns zero
1148
 * in case of success, and a negative error code in case of failure.
1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
 */
int ubi_wl_put_peb(struct ubi_device *ubi, int pnum, int torture)
{
	int err;
	struct ubi_wl_entry *e;

	dbg_wl("PEB %d", pnum);
	ubi_assert(pnum >= 0);
	ubi_assert(pnum < ubi->peb_count);

1159
retry:
1160 1161 1162 1163 1164 1165 1166 1167
	spin_lock(&ubi->wl_lock);
	e = ubi->lookuptbl[pnum];
	if (e == ubi->move_from) {
		/*
		 * User is putting the physical eraseblock which was selected to
		 * be moved. It will be scheduled for erasure in the
		 * wear-leveling worker.
		 */
1168
		dbg_wl("PEB %d is being moved, wait", pnum);
1169
		spin_unlock(&ubi->wl_lock);
1170 1171 1172 1173 1174

		/* Wait for the WL worker by taking the @ubi->move_mutex */
		mutex_lock(&ubi->move_mutex);
		mutex_unlock(&ubi->move_mutex);
		goto retry;
1175 1176 1177 1178
	} else if (e == ubi->move_to) {
		/*
		 * User is putting the physical eraseblock which was selected
		 * as the target the data is moved to. It may happen if the EBA
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1179 1180 1181 1182 1183
		 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
		 * but the WL sub-system has not put the PEB to the "used" tree
		 * yet, but it is about to do this. So we just set a flag which
		 * will tell the WL worker that the PEB is not needed anymore
		 * and should be scheduled for erasure.
1184 1185 1186 1187 1188 1189 1190
		 */
		dbg_wl("PEB %d is the target of data moving", pnum);
		ubi_assert(!ubi->move_to_put);
		ubi->move_to_put = 1;
		spin_unlock(&ubi->wl_lock);
		return 0;
	} else {
1191 1192 1193 1194 1195 1196
		if (in_wl_tree(e, &ubi->used)) {
			paranoid_check_in_wl_tree(e, &ubi->used);
			rb_erase(&e->rb, &ubi->used);
		} else if (in_wl_tree(e, &ubi->scrub)) {
			paranoid_check_in_wl_tree(e, &ubi->scrub);
			rb_erase(&e->rb, &ubi->scrub);
1197 1198 1199 1200 1201 1202 1203 1204 1205
		} else {
			err = prot_tree_del(ubi, e->pnum);
			if (err) {
				ubi_err("PEB %d not found", pnum);
				ubi_ro_mode(ubi);
				spin_unlock(&ubi->wl_lock);
				return err;
			}
		}
1206 1207 1208 1209 1210 1211
	}
	spin_unlock(&ubi->wl_lock);

	err = schedule_erase(ubi, e, torture);
	if (err) {
		spin_lock(&ubi->wl_lock);
1212
		wl_tree_add(e, &ubi->used);
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
		spin_unlock(&ubi->wl_lock);
	}

	return err;
}

/**
 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
 * @ubi: UBI device description object
 * @pnum: the physical eraseblock to schedule
 *
 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
 * needs scrubbing. This function schedules a physical eraseblock for
 * scrubbing which is done in background. This function returns zero in case of
 * success and a negative error code in case of failure.
 */
int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
{
	struct ubi_wl_entry *e;

1233
	dbg_msg("schedule PEB %d for scrubbing", pnum);
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255

retry:
	spin_lock(&ubi->wl_lock);
	e = ubi->lookuptbl[pnum];
	if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub)) {
		spin_unlock(&ubi->wl_lock);
		return 0;
	}

	if (e == ubi->move_to) {
		/*
		 * This physical eraseblock was used to move data to. The data
		 * was moved but the PEB was not yet inserted to the proper
		 * tree. We should just wait a little and let the WL worker
		 * proceed.
		 */
		spin_unlock(&ubi->wl_lock);
		dbg_wl("the PEB %d is not in proper tree, retry", pnum);
		yield();
		goto retry;
	}

1256 1257 1258
	if (in_wl_tree(e, &ubi->used)) {
		paranoid_check_in_wl_tree(e, &ubi->used);
		rb_erase(&e->rb, &ubi->used);
1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
	} else {
		int err;

		err = prot_tree_del(ubi, e->pnum);
		if (err) {
			ubi_err("PEB %d not found", pnum);
			ubi_ro_mode(ubi);
			spin_unlock(&ubi->wl_lock);
			return err;
		}
	}
1270

1271
	wl_tree_add(e, &ubi->scrub);
1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289
	spin_unlock(&ubi->wl_lock);

	/*
	 * Technically scrubbing is the same as wear-leveling, so it is done
	 * by the WL worker.
	 */
	return ensure_wear_leveling(ubi);
}

/**
 * ubi_wl_flush - flush all pending works.
 * @ubi: UBI device description object
 *
 * This function returns zero in case of success and a negative error code in
 * case of failure.
 */
int ubi_wl_flush(struct ubi_device *ubi)
{
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1290
	int err;
1291 1292 1293 1294 1295

	/*
	 * Erase while the pending works queue is not empty, but not more then
	 * the number of currently pending works.
	 */
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315
	dbg_wl("flush (%d pending works)", ubi->works_count);
	while (ubi->works_count) {
		err = do_work(ubi);
		if (err)
			return err;
	}

	/*
	 * Make sure all the works which have been done in parallel are
	 * finished.
	 */
	down_write(&ubi->work_sem);
	up_write(&ubi->work_sem);

	/*
	 * And in case last was the WL worker and it cancelled the LEB
	 * movement, flush again.
	 */
	while (ubi->works_count) {
		dbg_wl("flush more (%d pending works)", ubi->works_count);
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
		err = do_work(ubi);
		if (err)
			return err;
	}

	return 0;
}

/**
 * tree_destroy - destroy an RB-tree.
 * @root: the root of the tree to destroy
 */
static void tree_destroy(struct rb_root *root)
{
	struct rb_node *rb;
	struct ubi_wl_entry *e;

	rb = root->rb_node;
	while (rb) {
		if (rb->rb_left)
			rb = rb->rb_left;
		else if (rb->rb_right)
			rb = rb->rb_right;
		else {
			e = rb_entry(rb, struct ubi_wl_entry, rb);

			rb = rb_parent(rb);
			if (rb) {
				if (rb->rb_left == &e->rb)
					rb->rb_left = NULL;
				else
					rb->rb_right = NULL;
			}

1350
			kmem_cache_free(ubi_wl_entry_slab, e);
1351 1352 1353 1354 1355 1356 1357 1358
		}
	}
}

/**
 * ubi_thread - UBI background thread.
 * @u: the UBI device description object pointer
 */
1359
int ubi_thread(void *u)
1360 1361 1362 1363 1364
{
	int failures = 0;
	struct ubi_device *ubi = u;

	ubi_msg("background thread \"%s\" started, PID %d",
1365
		ubi->bgt_name, task_pid_nr(current));
1366

1367
	set_freezable();
1368 1369 1370 1371
	for (;;) {
		int err;

		if (kthread_should_stop())
1372
			break;
1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398

		if (try_to_freeze())
			continue;

		spin_lock(&ubi->wl_lock);
		if (list_empty(&ubi->works) || ubi->ro_mode ||
			       !ubi->thread_enabled) {
			set_current_state(TASK_INTERRUPTIBLE);
			spin_unlock(&ubi->wl_lock);
			schedule();
			continue;
		}
		spin_unlock(&ubi->wl_lock);

		err = do_work(ubi);
		if (err) {
			ubi_err("%s: work failed with error code %d",
				ubi->bgt_name, err);
			if (failures++ > WL_MAX_FAILURES) {
				/*
				 * Too many failures, disable the thread and
				 * switch to read-only mode.
				 */
				ubi_msg("%s: %d consecutive failures",
					ubi->bgt_name, WL_MAX_FAILURES);
				ubi_ro_mode(ubi);
1399 1400
				ubi->thread_enabled = 0;
				continue;
1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429
			}
		} else
			failures = 0;

		cond_resched();
	}

	dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
	return 0;
}

/**
 * cancel_pending - cancel all pending works.
 * @ubi: UBI device description object
 */
static void cancel_pending(struct ubi_device *ubi)
{
	while (!list_empty(&ubi->works)) {
		struct ubi_work *wrk;

		wrk = list_entry(ubi->works.next, struct ubi_work, list);
		list_del(&wrk->list);
		wrk->func(ubi, wrk, 1);
		ubi->works_count -= 1;
		ubi_assert(ubi->works_count >= 0);
	}
}

/**
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1430
 * ubi_wl_init_scan - initialize the WL sub-system using scanning information.
1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448
 * @ubi: UBI device description object
 * @si: scanning information
 *
 * This function returns zero in case of success, and a negative error code in
 * case of failure.
 */
int ubi_wl_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si)
{
	int err;
	struct rb_node *rb1, *rb2;
	struct ubi_scan_volume *sv;
	struct ubi_scan_leb *seb, *tmp;
	struct ubi_wl_entry *e;


	ubi->used = ubi->free = ubi->scrub = RB_ROOT;
	ubi->prot.pnum = ubi->prot.aec = RB_ROOT;
	spin_lock_init(&ubi->wl_lock);
1449
	mutex_init(&ubi->move_mutex);
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1450
	init_rwsem(&ubi->work_sem);
1451 1452 1453 1454 1455 1456 1457 1458
	ubi->max_ec = si->max_ec;
	INIT_LIST_HEAD(&ubi->works);

	sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);

	err = -ENOMEM;
	ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
	if (!ubi->lookuptbl)
1459
		return err;
1460 1461 1462 1463

	list_for_each_entry_safe(seb, tmp, &si->erase, u.list) {
		cond_resched();

1464
		e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1465 1466 1467 1468 1469 1470 1471
		if (!e)
			goto out_free;

		e->pnum = seb->pnum;
		e->ec = seb->ec;
		ubi->lookuptbl[e->pnum] = e;
		if (schedule_erase(ubi, e, 0)) {
1472
			kmem_cache_free(ubi_wl_entry_slab, e);
1473 1474 1475 1476 1477 1478 1479
			goto out_free;
		}
	}

	list_for_each_entry(seb, &si->free, u.list) {
		cond_resched();

1480
		e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1481 1482 1483 1484 1485 1486
		if (!e)
			goto out_free;

		e->pnum = seb->pnum;
		e->ec = seb->ec;
		ubi_assert(e->ec >= 0);
1487
		wl_tree_add(e, &ubi->free);
1488 1489 1490 1491 1492 1493
		ubi->lookuptbl[e->pnum] = e;
	}

	list_for_each_entry(seb, &si->corr, u.list) {
		cond_resched();

1494
		e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1495 1496 1497 1498 1499 1500 1501
		if (!e)
			goto out_free;

		e->pnum = seb->pnum;
		e->ec = seb->ec;
		ubi->lookuptbl[e->pnum] = e;
		if (schedule_erase(ubi, e, 0)) {
1502
			kmem_cache_free(ubi_wl_entry_slab, e);
1503 1504 1505 1506 1507 1508 1509 1510
			goto out_free;
		}
	}

	ubi_rb_for_each_entry(rb1, sv, &si->volumes, rb) {
		ubi_rb_for_each_entry(rb2, seb, &sv->root, u.rb) {
			cond_resched();

1511
			e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1512 1513 1514 1515 1516 1517 1518 1519 1520
			if (!e)
				goto out_free;

			e->pnum = seb->pnum;
			e->ec = seb->ec;
			ubi->lookuptbl[e->pnum] = e;
			if (!seb->scrub) {
				dbg_wl("add PEB %d EC %d to the used tree",
				       e->pnum, e->ec);
1521
				wl_tree_add(e, &ubi->used);
1522 1523 1524
			} else {
				dbg_wl("add PEB %d EC %d to the scrub tree",
				       e->pnum, e->ec);
1525
				wl_tree_add(e, &ubi->scrub);
1526 1527 1528 1529
			}
		}
	}

1530
	if (ubi->avail_pebs < WL_RESERVED_PEBS) {
1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579
		ubi_err("no enough physical eraseblocks (%d, need %d)",
			ubi->avail_pebs, WL_RESERVED_PEBS);
		goto out_free;
	}
	ubi->avail_pebs -= WL_RESERVED_PEBS;
	ubi->rsvd_pebs += WL_RESERVED_PEBS;

	/* Schedule wear-leveling if needed */
	err = ensure_wear_leveling(ubi);
	if (err)
		goto out_free;

	return 0;

out_free:
	cancel_pending(ubi);
	tree_destroy(&ubi->used);
	tree_destroy(&ubi->free);
	tree_destroy(&ubi->scrub);
	kfree(ubi->lookuptbl);
	return err;
}

/**
 * protection_trees_destroy - destroy the protection RB-trees.
 * @ubi: UBI device description object
 */
static void protection_trees_destroy(struct ubi_device *ubi)
{
	struct rb_node *rb;
	struct ubi_wl_prot_entry *pe;

	rb = ubi->prot.aec.rb_node;
	while (rb) {
		if (rb->rb_left)
			rb = rb->rb_left;
		else if (rb->rb_right)
			rb = rb->rb_right;
		else {
			pe = rb_entry(rb, struct ubi_wl_prot_entry, rb_aec);

			rb = rb_parent(rb);
			if (rb) {
				if (rb->rb_left == &pe->rb_aec)
					rb->rb_left = NULL;
				else
					rb->rb_right = NULL;
			}

1580
			kmem_cache_free(ubi_wl_entry_slab, pe->e);
1581 1582 1583 1584 1585 1586
			kfree(pe);
		}
	}
}

/**
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1587
 * ubi_wl_close - close the wear-leveling sub-system.
1588 1589 1590 1591
 * @ubi: UBI device description object
 */
void ubi_wl_close(struct ubi_device *ubi)
{
Artem Bityutskiy's avatar
Artem Bityutskiy committed
1592
	dbg_wl("close the WL sub-system");
1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603
	cancel_pending(ubi);
	protection_trees_destroy(ubi);
	tree_destroy(&ubi->used);
	tree_destroy(&ubi->free);
	tree_destroy(&ubi->scrub);
	kfree(ubi->lookuptbl);
}

#ifdef CONFIG_MTD_UBI_DEBUG_PARANOID

/**
1604
 * paranoid_check_ec - make sure that the erase counter of a PEB is correct.
1605 1606 1607 1608 1609 1610 1611 1612
 * @ubi: UBI device description object
 * @pnum: the physical eraseblock number to check
 * @ec: the erase counter to check
 *
 * This function returns zero if the erase counter of physical eraseblock @pnum
 * is equivalent to @ec, %1 if not, and a negative error code if an error
 * occurred.
 */
1613
static int paranoid_check_ec(struct ubi_device *ubi, int pnum, int ec)
1614 1615 1616 1617 1618
{
	int err;
	long long read_ec;
	struct ubi_ec_hdr *ec_hdr;

1619
	ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629
	if (!ec_hdr)
		return -ENOMEM;

	err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
	if (err && err != UBI_IO_BITFLIPS) {
		/* The header does not have to exist */
		err = 0;
		goto out_free;
	}

1630
	read_ec = be64_to_cpu(ec_hdr->ec);
1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644
	if (ec != read_ec) {
		ubi_err("paranoid check failed for PEB %d", pnum);
		ubi_err("read EC is %lld, should be %d", read_ec, ec);
		ubi_dbg_dump_stack();
		err = 1;
	} else
		err = 0;

out_free:
	kfree(ec_hdr);
	return err;
}

/**
1645
 * paranoid_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1646 1647 1648
 * @e: the wear-leveling entry to check
 * @root: the root of the tree
 *
1649 1650
 * This function returns zero if @e is in the @root RB-tree and %1 if it is
 * not.
1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664
 */
static int paranoid_check_in_wl_tree(struct ubi_wl_entry *e,
				     struct rb_root *root)
{
	if (in_wl_tree(e, root))
		return 0;

	ubi_err("paranoid check failed for PEB %d, EC %d, RB-tree %p ",
		e->pnum, e->ec, root);
	ubi_dbg_dump_stack();
	return 1;
}

#endif /* CONFIG_MTD_UBI_DEBUG_PARANOID */