*/
/*
- * UBI wear-leveling unit.
+ * UBI wear-leveling sub-system.
*
- * This unit 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 unit'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.
+ * 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.
*
* Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
- * header. The rest of the physical eraseblock contains only 0xFF bytes.
+ * header. The rest of the physical eraseblock contains only %0xFF bytes.
*
- * When physical eraseblocks are returned to the WL unit by means of the
+ * When physical eraseblocks are returned to the WL sub-system by means of the
* 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
* done asynchronously in context of the per-UBI device background thread,
- * which is also managed by the WL unit.
+ * which is also managed by the WL sub-system.
*
* The wear-leveling is ensured by means of moving the contents of used
* physical eraseblocks with low erase counter to free physical eraseblocks
* 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,
- * the WL unit may pick a free physical eraseblock with low erase counter, and
- * so forth.
+ * the WL sub-system may pick a free physical eraseblock with low erase
+ * counter, and so forth.
*
- * If the WL unit fails to erase a physical eraseblock, it marks it as bad.
+ * If the WL sub-system fails to erase a physical eraseblock, it marks it as
+ * bad.
*
- * This unit 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.
+ * 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.
*
- * As it was said, for the UBI unit 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,
+ * 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,
* @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
- * re-work this unit and make it more scalable.
+ * re-work this sub-system and make it more scalable.
*
- * At the moment this unit 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
+ * 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
* 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
- * room for future re-works of the WL unit.
+ * room for future re-works of the WL sub-system.
*
- * FIXME: looks too complex, should be simplified (later).
+ * Note: the stuff with protection trees looks too complex and is difficult to
+ * understand. Should be fixed.
*/
#include <linux/slab.h>
/*
* Maximum difference between two erase counters. If this threshold is
- * exceeded, the WL unit starts moving data from used physical eraseblocks with
- * low erase counter to free physical eraseblocks with high erase counter.
+ * exceeded, the WL sub-system starts moving data from used physical
+ * eraseblocks with low erase counter to free physical eraseblocks with high
+ * erase counter.
*/
#define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
/*
- * When a physical eraseblock is moved, the WL unit has to pick the target
+ * When a physical eraseblock is moved, the WL sub-system has to pick the target
* 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
- * counter of the free physical eraseblock to pick. Namely, the WL unit does
- * not pick eraseblocks with erase counter greater then the lowest erase
+ * 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
* counter plus %WL_FREE_MAX_DIFF.
*/
#define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
* @abs_ec: the absolute erase counter value when the protection ends
* @e: the wear-leveling entry of the physical eraseblock under protection
*
- * When the WL unit 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).
+ * 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).
*
* All this protection stuff is needed because:
* o we don't want to move physical eraseblocks just after we have given them
* @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
*
cond_resched();
+ /*
+ * @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);
spin_lock(&ubi->wl_lock);
-
if (list_empty(&ubi->works)) {
spin_unlock(&ubi->wl_lock);
+ up_read(&ubi->work_sem);
return 0;
}
wrk = list_entry(ubi->works.next, struct ubi_work, list);
list_del(&wrk->list);
+ ubi->works_count -= 1;
+ ubi_assert(ubi->works_count >= 0);
spin_unlock(&ubi->wl_lock);
/*
err = wrk->func(ubi, wrk, 0);
if (err)
ubi_err("work failed with error code %d", err);
+ up_read(&ubi->work_sem);
- spin_lock(&ubi->wl_lock);
- ubi->works_count -= 1;
- ubi_assert(ubi->works_count >= 0);
- spin_unlock(&ubi->wl_lock);
return err;
}
}
switch (dtype) {
- 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);
+ 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);
- 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();
+ 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();
}
/*
* This function returns zero in case of success and a negative error code in
* case of failure.
*/
-static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, int torture)
+static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
+ int torture)
{
int err;
struct ubi_ec_hdr *ec_hdr;
}
/**
- * check_protection_over - check if it is time to stop protecting some
- * physical eraseblocks.
+ * check_protection_over - check if it is time to stop protecting some PEBs.
* @ubi: UBI device description object
*
* This function is called after each erase operation, when the absolute erase
int cancel)
{
int err, put = 0, scrubbing = 0, protect = 0;
- struct ubi_wl_prot_entry *pe;
+ struct ubi_wl_prot_entry *uninitialized_var(pe);
struct ubi_wl_entry *e1, *e2;
struct ubi_vid_hdr *vid_hdr;
}
ubi_free_vid_hdr(ubi, vid_hdr);
+ if (scrubbing && !protect)
+ ubi_msg("scrubbed PEB %d, data moved to PEB %d",
+ e1->pnum, e2->pnum);
+
spin_lock(&ubi->wl_lock);
if (protect)
prot_tree_add(ubi, e1, pe, protect);
spin_unlock(&ubi->wl_lock);
/*
- * One more erase operation has happened, take care about protected
- * physical eraseblocks.
+ * One more erase operation has happened, take care about
+ * protected physical eraseblocks.
*/
check_protection_over(ubi);
}
/**
- * ubi_wl_put_peb - return a physical eraseblock to the wear-leveling unit.
+ * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
* @ubi: UBI device description object
* @pnum: physical eraseblock to return
* @torture: if this physical eraseblock has to be tortured
/*
* User is putting the physical eraseblock which was selected
* as the target the data is moved to. It may happen if the EBA
- * unit already re-mapped the LEB in 'ubi_eba_copy_leb()' but
- * the WL unit 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 sheduled for erasure.
+ * 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.
*/
dbg_wl("PEB %d is the target of data moving", pnum);
ubi_assert(!ubi->move_to_put);
{
struct ubi_wl_entry *e;
- ubi_msg("schedule PEB %d for scrubbing", pnum);
+ dbg_msg("schedule PEB %d for scrubbing", pnum);
retry:
spin_lock(&ubi->wl_lock);
*/
int ubi_wl_flush(struct ubi_device *ubi)
{
- int err, pending_count;
-
- pending_count = ubi->works_count;
-
- dbg_wl("flush (%d pending works)", pending_count);
+ int err;
/*
* Erase while the pending works queue is not empty, but not more then
* the number of currently pending works.
*/
- while (pending_count-- > 0) {
+ 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);
err = do_work(ubi);
if (err)
return err;
* ubi_thread - UBI background thread.
* @u: the UBI device description object pointer
*/
-static int ubi_thread(void *u)
+int ubi_thread(void *u)
{
int failures = 0;
struct ubi_device *ubi = u;
int err;
if (kthread_should_stop())
- goto out;
+ break;
if (try_to_freeze())
continue;
cond_resched();
}
-out:
dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
return 0;
}
}
/**
- * ubi_wl_init_scan - initialize the wear-leveling unit using scanning
- * information.
+ * ubi_wl_init_scan - initialize the WL sub-system using scanning information.
* @ubi: UBI device description object
* @si: scanning information
*
ubi->prot.pnum = ubi->prot.aec = RB_ROOT;
spin_lock_init(&ubi->wl_lock);
mutex_init(&ubi->move_mutex);
+ init_rwsem(&ubi->work_sem);
ubi->max_ec = si->max_ec;
INIT_LIST_HEAD(&ubi->works);
sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
- ubi->bgt_thread = kthread_create(ubi_thread, ubi, ubi->bgt_name);
- if (IS_ERR(ubi->bgt_thread)) {
- err = PTR_ERR(ubi->bgt_thread);
- ubi_err("cannot spawn \"%s\", error %d", ubi->bgt_name,
- err);
- return err;
- }
-
err = -ENOMEM;
ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
if (!ubi->lookuptbl)
- goto out_free;
+ return err;
list_for_each_entry_safe(seb, tmp, &si->erase, u.list) {
cond_resched();
}
/**
- * ubi_wl_close - close the wear-leveling unit.
+ * ubi_wl_close - close the wear-leveling sub-system.
* @ubi: UBI device description object
*/
void ubi_wl_close(struct ubi_device *ubi)
{
- dbg_wl("disable \"%s\"", ubi->bgt_name);
- if (ubi->bgt_thread)
- kthread_stop(ubi->bgt_thread);
-
- dbg_wl("close the UBI wear-leveling unit");
-
+ dbg_wl("close the WL sub-system");
cancel_pending(ubi);
protection_trees_destroy(ubi);
tree_destroy(&ubi->used);
#ifdef CONFIG_MTD_UBI_DEBUG_PARANOID
/**
- * paranoid_check_ec - make sure that the erase counter of a physical eraseblock
- * is correct.
+ * paranoid_check_ec - make sure that the erase counter of a PEB is correct.
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to check
* @ec: the erase counter to check
}
/**
- * paranoid_check_in_wl_tree - make sure that a wear-leveling entry is present
- * in a WL RB-tree.
+ * paranoid_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
* @e: the wear-leveling entry to check
* @root: the root of the tree
*
- * This function returns zero if @e is in the @root RB-tree and %1 if it
- * is not.
+ * This function returns zero if @e is in the @root RB-tree and %1 if it is
+ * not.
*/
static int paranoid_check_in_wl_tree(struct ubi_wl_entry *e,
struct rb_root *root)