X-Git-Url: http://pilppa.org/gitweb/gitweb.cgi?a=blobdiff_plain;f=drivers%2Fmtd%2Fubi%2Feba.c;h=7ce91ca742b136c7b60ac6db25ffb3d3771d2d6b;hb=06b8147c5dbd385b5b97ca74e19f6f3951ebc1cb;hp=1297732f4db95c4275e9f3f78b81345d48787572;hpb=4937ce87959629d31e9b09cf5bdf1e12a305c805;p=linux-2.6-omap-h63xx.git diff --git a/drivers/mtd/ubi/eba.c b/drivers/mtd/ubi/eba.c index 1297732f4db..7ce91ca742b 100644 --- a/drivers/mtd/ubi/eba.c +++ b/drivers/mtd/ubi/eba.c @@ -31,7 +31,7 @@ * logical eraseblock it is locked for reading or writing. The per-logical * eraseblock locking is implemented by means of the lock tree. The lock tree * is an RB-tree which refers all the currently locked logical eraseblocks. The - * lock tree elements are &struct ltree_entry objects. They are indexed by + * lock tree elements are &struct ubi_ltree_entry objects. They are indexed by * (@vol_id, @lnum) pairs. * * EBA also maintains the global sequence counter which is incremented each @@ -49,29 +49,6 @@ /* Number of physical eraseblocks reserved for atomic LEB change operation */ #define EBA_RESERVED_PEBS 1 -/** - * struct ltree_entry - an entry in the lock tree. - * @rb: links RB-tree nodes - * @vol_id: volume ID of the locked logical eraseblock - * @lnum: locked logical eraseblock number - * @users: how many tasks are using this logical eraseblock or wait for it - * @mutex: read/write mutex to implement read/write access serialization to - * the (@vol_id, @lnum) logical eraseblock - * - * When a logical eraseblock is being locked - corresponding &struct ltree_entry - * object is inserted to the lock tree (@ubi->ltree). - */ -struct ltree_entry { - struct rb_node rb; - int vol_id; - int lnum; - int users; - struct rw_semaphore mutex; -}; - -/* Slab cache for lock-tree entries */ -static struct kmem_cache *ltree_slab; - /** * next_sqnum - get next sequence number. * @ubi: UBI device description object @@ -101,7 +78,7 @@ static unsigned long long next_sqnum(struct ubi_device *ubi) */ static int ubi_get_compat(const struct ubi_device *ubi, int vol_id) { - if (vol_id == UBI_LAYOUT_VOL_ID) + if (vol_id == UBI_LAYOUT_VOLUME_ID) return UBI_LAYOUT_VOLUME_COMPAT; return 0; } @@ -112,20 +89,20 @@ static int ubi_get_compat(const struct ubi_device *ubi, int vol_id) * @vol_id: volume ID * @lnum: logical eraseblock number * - * This function returns a pointer to the corresponding &struct ltree_entry + * This function returns a pointer to the corresponding &struct ubi_ltree_entry * object if the logical eraseblock is locked and %NULL if it is not. * @ubi->ltree_lock has to be locked. */ -static struct ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id, - int lnum) +static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id, + int lnum) { struct rb_node *p; p = ubi->ltree.rb_node; while (p) { - struct ltree_entry *le; + struct ubi_ltree_entry *le; - le = rb_entry(p, struct ltree_entry, rb); + le = rb_entry(p, struct ubi_ltree_entry, rb); if (vol_id < le->vol_id) p = p->rb_left; @@ -155,15 +132,17 @@ static struct ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id, * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation * failed. */ -static struct ltree_entry *ltree_add_entry(struct ubi_device *ubi, int vol_id, - int lnum) +static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi, + int vol_id, int lnum) { - struct ltree_entry *le, *le1, *le_free; + struct ubi_ltree_entry *le, *le1, *le_free; - le = kmem_cache_alloc(ltree_slab, GFP_NOFS); + le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS); if (!le) return ERR_PTR(-ENOMEM); + le->users = 0; + init_rwsem(&le->mutex); le->vol_id = vol_id; le->lnum = lnum; @@ -189,7 +168,7 @@ static struct ltree_entry *ltree_add_entry(struct ubi_device *ubi, int vol_id, p = &ubi->ltree.rb_node; while (*p) { parent = *p; - le1 = rb_entry(parent, struct ltree_entry, rb); + le1 = rb_entry(parent, struct ubi_ltree_entry, rb); if (vol_id < le1->vol_id) p = &(*p)->rb_left; @@ -211,7 +190,7 @@ static struct ltree_entry *ltree_add_entry(struct ubi_device *ubi, int vol_id, spin_unlock(&ubi->ltree_lock); if (le_free) - kmem_cache_free(ltree_slab, le_free); + kfree(le_free); return le; } @@ -227,7 +206,7 @@ static struct ltree_entry *ltree_add_entry(struct ubi_device *ubi, int vol_id, */ static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum) { - struct ltree_entry *le; + struct ubi_ltree_entry *le; le = ltree_add_entry(ubi, vol_id, lnum); if (IS_ERR(le)) @@ -245,7 +224,7 @@ static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum) static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum) { int free = 0; - struct ltree_entry *le; + struct ubi_ltree_entry *le; spin_lock(&ubi->ltree_lock); le = ltree_lookup(ubi, vol_id, lnum); @@ -259,7 +238,7 @@ static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum) up_read(&le->mutex); if (free) - kmem_cache_free(ltree_slab, le); + kfree(le); } /** @@ -273,7 +252,7 @@ static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum) */ static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum) { - struct ltree_entry *le; + struct ubi_ltree_entry *le; le = ltree_add_entry(ubi, vol_id, lnum); if (IS_ERR(le)) @@ -282,6 +261,44 @@ static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum) return 0; } +/** + * leb_write_lock - lock logical eraseblock for writing. + * @ubi: UBI device description object + * @vol_id: volume ID + * @lnum: logical eraseblock number + * + * This function locks a logical eraseblock for writing if there is no + * contention and does nothing if there is contention. Returns %0 in case of + * success, %1 in case of contention, and and a negative error code in case of + * failure. + */ +static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum) +{ + int free; + struct ubi_ltree_entry *le; + + le = ltree_add_entry(ubi, vol_id, lnum); + if (IS_ERR(le)) + return PTR_ERR(le); + if (down_write_trylock(&le->mutex)) + return 0; + + /* Contention, cancel */ + spin_lock(&ubi->ltree_lock); + le->users -= 1; + ubi_assert(le->users >= 0); + if (le->users == 0) { + rb_erase(&le->rb, &ubi->ltree); + free = 1; + } else + free = 0; + spin_unlock(&ubi->ltree_lock); + if (free) + kfree(le); + + return 1; +} + /** * leb_write_unlock - unlock logical eraseblock. * @ubi: UBI device description object @@ -291,7 +308,7 @@ static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum) static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum) { int free; - struct ltree_entry *le; + struct ubi_ltree_entry *le; spin_lock(&ubi->ltree_lock); le = ltree_lookup(ubi, vol_id, lnum); @@ -306,23 +323,23 @@ static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum) up_write(&le->mutex); if (free) - kmem_cache_free(ltree_slab, le); + kfree(le); } /** * ubi_eba_unmap_leb - un-map logical eraseblock. * @ubi: UBI device description object - * @vol_id: volume ID + * @vol: volume description object * @lnum: logical eraseblock number * * This function un-maps logical eraseblock @lnum and schedules corresponding * physical eraseblock for erasure. Returns zero in case of success and a * negative error code in case of failure. */ -int ubi_eba_unmap_leb(struct ubi_device *ubi, int vol_id, int lnum) +int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol, + int lnum) { - int idx = vol_id2idx(ubi, vol_id), err, pnum; - struct ubi_volume *vol = ubi->volumes[idx]; + int err, pnum, vol_id = vol->vol_id; if (ubi->ro_mode) return -EROFS; @@ -349,7 +366,7 @@ out_unlock: /** * ubi_eba_read_leb - read data. * @ubi: UBI device description object - * @vol_id: volume ID + * @vol: volume description object * @lnum: logical eraseblock number * @buf: buffer to store the read data * @offset: offset from where to read @@ -365,12 +382,11 @@ out_unlock: * returned for any volume type if an ECC error was detected by the MTD device * driver. Other negative error cored may be returned in case of other errors. */ -int ubi_eba_read_leb(struct ubi_device *ubi, int vol_id, int lnum, void *buf, - int offset, int len, int check) +int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum, + void *buf, int offset, int len, int check) { - int err, pnum, scrub = 0, idx = vol_id2idx(ubi, vol_id); + int err, pnum, scrub = 0, vol_id = vol->vol_id; struct ubi_vid_hdr *vid_hdr; - struct ubi_volume *vol = ubi->volumes[idx]; uint32_t uninitialized_var(crc); err = leb_read_lock(ubi, vol_id, lnum); @@ -578,7 +594,7 @@ write_error: /** * ubi_eba_write_leb - write data to dynamic volume. * @ubi: UBI device description object - * @vol_id: volume ID + * @vol: volume description object * @lnum: logical eraseblock number * @buf: the data to write * @offset: offset within the logical eraseblock where to write @@ -586,15 +602,14 @@ write_error: * @dtype: data type * * This function writes data to logical eraseblock @lnum of a dynamic volume - * @vol_id. Returns zero in case of success and a negative error code in case + * @vol. Returns zero in case of success and a negative error code in case * of failure. In case of error, it is possible that something was still * written to the flash media, but may be some garbage. */ -int ubi_eba_write_leb(struct ubi_device *ubi, int vol_id, int lnum, +int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum, const void *buf, int offset, int len, int dtype) { - int idx = vol_id2idx(ubi, vol_id), err, pnum, tries = 0; - struct ubi_volume *vol = ubi->volumes[idx]; + int err, pnum, tries = 0, vol_id = vol->vol_id; struct ubi_vid_hdr *vid_hdr; if (ubi->ro_mode) @@ -613,7 +628,8 @@ int ubi_eba_write_leb(struct ubi_device *ubi, int vol_id, int lnum, if (err) { ubi_warn("failed to write data to PEB %d", pnum); if (err == -EIO && ubi->bad_allowed) - err = recover_peb(ubi, pnum, vol_id, lnum, buf, offset, len); + err = recover_peb(ubi, pnum, vol_id, lnum, buf, + offset, len); if (err) ubi_ro_mode(ubi); } @@ -656,11 +672,14 @@ retry: goto write_error; } - err = ubi_io_write_data(ubi, buf, pnum, offset, len); - if (err) { - ubi_warn("failed to write %d bytes at offset %d of LEB %d:%d, " - "PEB %d", len, offset, vol_id, lnum, pnum); - goto write_error; + if (len) { + err = ubi_io_write_data(ubi, buf, pnum, offset, len); + if (err) { + ubi_warn("failed to write %d bytes at offset %d of " + "LEB %d:%d, PEB %d", len, offset, vol_id, + lnum, pnum); + goto write_error; + } } vol->eba_tbl[lnum] = pnum; @@ -698,7 +717,7 @@ write_error: /** * ubi_eba_write_leb_st - write data to static volume. * @ubi: UBI device description object - * @vol_id: volume ID + * @vol: volume description object * @lnum: logical eraseblock number * @buf: data to write * @len: how many bytes to write @@ -706,7 +725,7 @@ write_error: * @used_ebs: how many logical eraseblocks will this volume contain * * This function writes data to logical eraseblock @lnum of static volume - * @vol_id. The @used_ebs argument should contain total number of logical + * @vol. The @used_ebs argument should contain total number of logical * eraseblock in this static volume. * * When writing to the last logical eraseblock, the @len argument doesn't have @@ -718,12 +737,11 @@ write_error: * volumes. This function returns zero in case of success and a negative error * code in case of failure. */ -int ubi_eba_write_leb_st(struct ubi_device *ubi, int vol_id, int lnum, - const void *buf, int len, int dtype, int used_ebs) +int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol, + int lnum, const void *buf, int len, int dtype, + int used_ebs) { - int err, pnum, tries = 0, data_size = len; - int idx = vol_id2idx(ubi, vol_id); - struct ubi_volume *vol = ubi->volumes[idx]; + int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id; struct ubi_vid_hdr *vid_hdr; uint32_t crc; @@ -819,7 +837,7 @@ write_error: /* * ubi_eba_atomic_leb_change - change logical eraseblock atomically. * @ubi: UBI device description object - * @vol_id: volume ID + * @vol: volume description object * @lnum: logical eraseblock number * @buf: data to write * @len: how many bytes to write @@ -834,17 +852,27 @@ write_error: * UBI reserves one LEB for the "atomic LEB change" operation, so only one * LEB change may be done at a time. This is ensured by @ubi->alc_mutex. */ -int ubi_eba_atomic_leb_change(struct ubi_device *ubi, int vol_id, int lnum, - const void *buf, int len, int dtype) +int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol, + int lnum, const void *buf, int len, int dtype) { - int err, pnum, tries = 0, idx = vol_id2idx(ubi, vol_id); - struct ubi_volume *vol = ubi->volumes[idx]; + int err, pnum, tries = 0, vol_id = vol->vol_id; struct ubi_vid_hdr *vid_hdr; uint32_t crc; if (ubi->ro_mode) return -EROFS; + if (len == 0) { + /* + * Special case when data length is zero. In this case the LEB + * has to be unmapped and mapped somewhere else. + */ + err = ubi_eba_unmap_leb(ubi, vol, lnum); + if (err) + return err; + return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0, dtype); + } + vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS); if (!vid_hdr) return -ENOMEM; @@ -927,21 +955,6 @@ write_error: goto retry; } -/** - * ltree_entry_ctor - lock tree entries slab cache constructor. - * @obj: the lock-tree entry to construct - * @cache: the lock tree entry slab cache - * @flags: constructor flags - */ -static void ltree_entry_ctor(void *obj, struct kmem_cache *cache, - unsigned long flags) -{ - struct ltree_entry *le = obj; - - le->users = 0; - init_rwsem(&le->mutex); -} - /** * ubi_eba_copy_leb - copy logical eraseblock. * @ubi: UBI device description object @@ -951,14 +964,16 @@ static void ltree_entry_ctor(void *obj, struct kmem_cache *cache, * * This function copies logical eraseblock from physical eraseblock @from to * physical eraseblock @to. The @vid_hdr buffer may be changed by this - * function. Returns zero in case of success, %UBI_IO_BITFLIPS if the operation - * was canceled because bit-flips were detected at the target PEB, and a - * negative error code in case of failure. + * function. Returns: + * o %0 in case of success; + * o %1 if the operation was canceled and should be tried later (e.g., + * because a bit-flip was detected at the target PEB); + * o %2 if the volume is being deleted and this LEB should not be moved. */ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to, struct ubi_vid_hdr *vid_hdr) { - int err, vol_id, lnum, data_size, aldata_size, pnum, idx; + int err, vol_id, lnum, data_size, aldata_size, idx; struct ubi_volume *vol; uint32_t crc; @@ -974,51 +989,67 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to, data_size = aldata_size = ubi->leb_size - be32_to_cpu(vid_hdr->data_pad); - /* - * We do not want anybody to write to this logical eraseblock while we - * are moving it, so we lock it. - */ - err = leb_write_lock(ubi, vol_id, lnum); - if (err) - return err; - - mutex_lock(&ubi->buf_mutex); - - /* - * But the logical eraseblock might have been put by this time. - * Cancel if it is true. - */ idx = vol_id2idx(ubi, vol_id); - + spin_lock(&ubi->volumes_lock); /* - * We may race with volume deletion/re-size, so we have to hold - * @ubi->volumes_lock. + * Note, we may race with volume deletion, which means that the volume + * this logical eraseblock belongs to might be being deleted. Since the + * volume deletion unmaps all the volume's logical eraseblocks, it will + * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish. */ - spin_lock(&ubi->volumes_lock); vol = ubi->volumes[idx]; if (!vol) { - dbg_eba("volume %d was removed meanwhile", vol_id); + /* No need to do further work, cancel */ + dbg_eba("volume %d is being removed, cancel", vol_id); spin_unlock(&ubi->volumes_lock); - goto out_unlock; + return 2; } + spin_unlock(&ubi->volumes_lock); - pnum = vol->eba_tbl[lnum]; - if (pnum != from) { - dbg_eba("LEB %d:%d is no longer mapped to PEB %d, mapped to " - "PEB %d, cancel", vol_id, lnum, from, pnum); - spin_unlock(&ubi->volumes_lock); - goto out_unlock; + /* + * We do not want anybody to write to this logical eraseblock while we + * are moving it, so lock it. + * + * Note, we are using non-waiting locking here, because we cannot sleep + * on the LEB, since it may cause deadlocks. Indeed, imagine a task is + * unmapping the LEB which is mapped to the PEB we are going to move + * (@from). This task locks the LEB and goes sleep in the + * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are + * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the + * LEB is already locked, we just do not move it and return %1. + */ + err = leb_write_trylock(ubi, vol_id, lnum); + if (err) { + dbg_eba("contention on LEB %d:%d, cancel", vol_id, lnum); + return err; } - spin_unlock(&ubi->volumes_lock); - /* OK, now the LEB is locked and we can safely start moving it */ + /* + * The LEB might have been put meanwhile, and the task which put it is + * probably waiting on @ubi->move_mutex. No need to continue the work, + * cancel it. + */ + if (vol->eba_tbl[lnum] != from) { + dbg_eba("LEB %d:%d is no longer mapped to PEB %d, mapped to " + "PEB %d, cancel", vol_id, lnum, from, + vol->eba_tbl[lnum]); + err = 1; + goto out_unlock_leb; + } + /* + * OK, now the LEB is locked and we can safely start moving iy. Since + * this function utilizes thie @ubi->peb1_buf buffer which is shared + * with some other functions, so lock the buffer by taking the + * @ubi->buf_mutex. + */ + mutex_lock(&ubi->buf_mutex); dbg_eba("read %d bytes of data", aldata_size); err = ubi_io_read_data(ubi, ubi->peb_buf1, from, 0, aldata_size); if (err && err != UBI_IO_BITFLIPS) { ubi_warn("error %d while reading data from PEB %d", err, from); - goto out_unlock; + goto out_unlock_buf; } /* @@ -1054,7 +1085,7 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to, err = ubi_io_write_vid_hdr(ubi, to, vid_hdr); if (err) - goto out_unlock; + goto out_unlock_buf; cond_resched(); @@ -1063,13 +1094,15 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to, if (err) { if (err != UBI_IO_BITFLIPS) ubi_warn("cannot read VID header back from PEB %d", to); - goto out_unlock; + else + err = 1; + goto out_unlock_buf; } if (data_size > 0) { err = ubi_io_write_data(ubi, ubi->peb_buf1, to, 0, aldata_size); if (err) - goto out_unlock; + goto out_unlock_buf; cond_resched(); @@ -1083,7 +1116,9 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to, if (err != UBI_IO_BITFLIPS) ubi_warn("cannot read data back from PEB %d", to); - goto out_unlock; + else + err = 1; + goto out_unlock_buf; } cond_resched(); @@ -1091,15 +1126,16 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to, if (memcmp(ubi->peb_buf1, ubi->peb_buf2, aldata_size)) { ubi_warn("read data back from PEB %d - it is different", to); - goto out_unlock; + goto out_unlock_buf; } } ubi_assert(vol->eba_tbl[lnum] == from); vol->eba_tbl[lnum] = to; -out_unlock: +out_unlock_buf: mutex_unlock(&ubi->buf_mutex); +out_unlock_leb: leb_write_unlock(ubi, vol_id, lnum); return err; } @@ -1126,14 +1162,6 @@ int ubi_eba_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si) mutex_init(&ubi->alc_mutex); ubi->ltree = RB_ROOT; - if (ubi_devices_cnt == 0) { - ltree_slab = kmem_cache_create("ubi_ltree_slab", - sizeof(struct ltree_entry), 0, - 0, <ree_entry_ctor); - if (!ltree_slab) - return -ENOMEM; - } - ubi->global_sqnum = si->max_sqnum + 1; num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT; @@ -1169,6 +1197,15 @@ int ubi_eba_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si) } } + if (ubi->avail_pebs < EBA_RESERVED_PEBS) { + ubi_err("no enough physical eraseblocks (%d, need %d)", + ubi->avail_pebs, EBA_RESERVED_PEBS); + err = -ENOSPC; + goto out_free; + } + ubi->avail_pebs -= EBA_RESERVED_PEBS; + ubi->rsvd_pebs += EBA_RESERVED_PEBS; + if (ubi->bad_allowed) { ubi_calculate_reserved(ubi); @@ -1185,15 +1222,6 @@ int ubi_eba_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si) ubi->rsvd_pebs += ubi->beb_rsvd_pebs; } - if (ubi->avail_pebs < EBA_RESERVED_PEBS) { - ubi_err("no enough physical eraseblocks (%d, need %d)", - ubi->avail_pebs, EBA_RESERVED_PEBS); - err = -ENOSPC; - goto out_free; - } - ubi->avail_pebs -= EBA_RESERVED_PEBS; - ubi->rsvd_pebs += EBA_RESERVED_PEBS; - dbg_eba("EBA unit is initialized"); return 0; @@ -1203,8 +1231,6 @@ out_free: continue; kfree(ubi->volumes[i]->eba_tbl); } - if (ubi_devices_cnt == 0) - kmem_cache_destroy(ltree_slab); return err; } @@ -1223,6 +1249,4 @@ void ubi_eba_close(const struct ubi_device *ubi) continue; kfree(ubi->volumes[i]->eba_tbl); } - if (ubi_devices_cnt == 1) - kmem_cache_destroy(ltree_slab); }