zram: get rid of lockdep warning
authorMinchan Kim <minchan@kernel.org>
Tue, 5 Feb 2013 23:48:53 +0000 (08:48 +0900)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 6 Feb 2013 02:01:24 +0000 (18:01 -0800)
Lockdep complains about recursive deadlock of zram->init_lock.
[1] made it false positive because we can't request IO to zram
before setting disksize. Anyway, we should shut lockdep up to
avoid many reporting from user.

[1] : zram: force disksize setting before using zram

Acked-by: Jerome Marchand <jmarchan@redhat.com>
Acked-by: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/zram/zram_drv.c
drivers/staging/zram/zram_drv.h
drivers/staging/zram/zram_sysfs.c

index 3318b0f..9ac140b 100644 (file)
@@ -61,22 +61,22 @@ static void zram_stat64_inc(struct zram *zram, u64 *v)
        zram_stat64_add(zram, v, 1);
 }
 
-static int zram_test_flag(struct zram *zram, u32 index,
+static int zram_test_flag(struct zram_meta *meta, u32 index,
                        enum zram_pageflags flag)
 {
-       return zram->table[index].flags & BIT(flag);
+       return meta->table[index].flags & BIT(flag);
 }
 
-static void zram_set_flag(struct zram *zram, u32 index,
+static void zram_set_flag(struct zram_meta *meta, u32 index,
                        enum zram_pageflags flag)
 {
-       zram->table[index].flags |= BIT(flag);
+       meta->table[index].flags |= BIT(flag);
 }
 
-static void zram_clear_flag(struct zram *zram, u32 index,
+static void zram_clear_flag(struct zram_meta *meta, u32 index,
                        enum zram_pageflags flag)
 {
-       zram->table[index].flags &= ~BIT(flag);
+       meta->table[index].flags &= ~BIT(flag);
 }
 
 static int page_zero_filled(void *ptr)
@@ -96,16 +96,17 @@ static int page_zero_filled(void *ptr)
 
 static void zram_free_page(struct zram *zram, size_t index)
 {
-       unsigned long handle = zram->table[index].handle;
-       u16 size = zram->table[index].size;
+       struct zram_meta *meta = zram->meta;
+       unsigned long handle = meta->table[index].handle;
+       u16 size = meta->table[index].size;
 
        if (unlikely(!handle)) {
                /*
                 * No memory is allocated for zero filled pages.
                 * Simply clear zero page flag.
                 */
-               if (zram_test_flag(zram, index, ZRAM_ZERO)) {
-                       zram_clear_flag(zram, index, ZRAM_ZERO);
+               if (zram_test_flag(meta, index, ZRAM_ZERO)) {
+                       zram_clear_flag(meta, index, ZRAM_ZERO);
                        zram->stats.pages_zero--;
                }
                return;
@@ -114,17 +115,17 @@ static void zram_free_page(struct zram *zram, size_t index)
        if (unlikely(size > max_zpage_size))
                zram->stats.bad_compress--;
 
-       zs_free(zram->mem_pool, handle);
+       zs_free(meta->mem_pool, handle);
 
        if (size <= PAGE_SIZE / 2)
                zram->stats.good_compress--;
 
        zram_stat64_sub(zram, &zram->stats.compr_size,
-                       zram->table[index].size);
+                       meta->table[index].size);
        zram->stats.pages_stored--;
 
-       zram->table[index].handle = 0;
-       zram->table[index].size = 0;
+       meta->table[index].handle = 0;
+       meta->table[index].size = 0;
 }
 
 static void handle_zero_page(struct bio_vec *bvec)
@@ -149,20 +150,21 @@ static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
        int ret = LZO_E_OK;
        size_t clen = PAGE_SIZE;
        unsigned char *cmem;
-       unsigned long handle = zram->table[index].handle;
+       struct zram_meta *meta = zram->meta;
+       unsigned long handle = meta->table[index].handle;
 
-       if (!handle || zram_test_flag(zram, index, ZRAM_ZERO)) {
+       if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
                memset(mem, 0, PAGE_SIZE);
                return 0;
        }
 
-       cmem = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
-       if (zram->table[index].size == PAGE_SIZE)
+       cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
+       if (meta->table[index].size == PAGE_SIZE)
                memcpy(mem, cmem, PAGE_SIZE);
        else
-               ret = lzo1x_decompress_safe(cmem, zram->table[index].size,
+               ret = lzo1x_decompress_safe(cmem, meta->table[index].size,
                                                mem, &clen);
-       zs_unmap_object(zram->mem_pool, handle);
+       zs_unmap_object(meta->mem_pool, handle);
 
        /* Should NEVER happen. Return bio error if it does. */
        if (unlikely(ret != LZO_E_OK)) {
@@ -180,11 +182,11 @@ static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
        int ret;
        struct page *page;
        unsigned char *user_mem, *uncmem = NULL;
-
+       struct zram_meta *meta = zram->meta;
        page = bvec->bv_page;
 
-       if (unlikely(!zram->table[index].handle) ||
-                       zram_test_flag(zram, index, ZRAM_ZERO)) {
+       if (unlikely(!meta->table[index].handle) ||
+                       zram_test_flag(meta, index, ZRAM_ZERO)) {
                handle_zero_page(bvec);
                return 0;
        }
@@ -232,9 +234,10 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
        unsigned long handle;
        struct page *page;
        unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
+       struct zram_meta *meta = zram->meta;
 
        page = bvec->bv_page;
-       src = zram->compress_buffer;
+       src = meta->compress_buffer;
 
        if (is_partial_io(bvec)) {
                /*
@@ -256,8 +259,8 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
         * System overwrites unused sectors. Free memory associated
         * with this sector now.
         */
-       if (zram->table[index].handle ||
-           zram_test_flag(zram, index, ZRAM_ZERO))
+       if (meta->table[index].handle ||
+           zram_test_flag(meta, index, ZRAM_ZERO))
                zram_free_page(zram, index);
 
        user_mem = kmap_atomic(page);
@@ -276,13 +279,13 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
                if (is_partial_io(bvec))
                        kfree(uncmem);
                zram->stats.pages_zero++;
-               zram_set_flag(zram, index, ZRAM_ZERO);
+               zram_set_flag(meta, index, ZRAM_ZERO);
                ret = 0;
                goto out;
        }
 
        ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
-                              zram->compress_workmem);
+                              meta->compress_workmem);
 
        if (!is_partial_io(bvec)) {
                kunmap_atomic(user_mem);
@@ -303,14 +306,14 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
                        src = uncmem;
        }
 
-       handle = zs_malloc(zram->mem_pool, clen);
+       handle = zs_malloc(meta->mem_pool, clen);
        if (!handle) {
                pr_info("Error allocating memory for compressed "
                        "page: %u, size=%zu\n", index, clen);
                ret = -ENOMEM;
                goto out;
        }
-       cmem = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
+       cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 
        if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
                src = kmap_atomic(page);
@@ -318,10 +321,10 @@ static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
        if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
                kunmap_atomic(src);
 
-       zs_unmap_object(zram->mem_pool, handle);
+       zs_unmap_object(meta->mem_pool, handle);
 
-       zram->table[index].handle = handle;
-       zram->table[index].size = clen;
+       meta->table[index].handle = handle;
+       meta->table[index].size = clen;
 
        /* Update stats */
        zram_stat64_add(zram, &zram->stats.compr_size, clen);
@@ -464,34 +467,25 @@ error:
 void __zram_reset_device(struct zram *zram)
 {
        size_t index;
+       struct zram_meta *meta;
 
        if (!zram->init_done)
                return;
 
+       meta = zram->meta;
        zram->init_done = 0;
 
-       /* Free various per-device buffers */
-       kfree(zram->compress_workmem);
-       free_pages((unsigned long)zram->compress_buffer, 1);
-
-       zram->compress_workmem = NULL;
-       zram->compress_buffer = NULL;
-
        /* Free all pages that are still in this zram device */
        for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
-               unsigned long handle = zram->table[index].handle;
+               unsigned long handle = meta->table[index].handle;
                if (!handle)
                        continue;
 
-               zs_free(zram->mem_pool, handle);
+               zs_free(meta->mem_pool, handle);
        }
 
-       vfree(zram->table);
-       zram->table = NULL;
-
-       zs_destroy_pool(zram->mem_pool);
-       zram->mem_pool = NULL;
-
+       zram_meta_free(zram->meta);
+       zram->meta = NULL;
        /* Reset stats */
        memset(&zram->stats, 0, sizeof(zram->stats));
 
@@ -506,12 +500,65 @@ void zram_reset_device(struct zram *zram)
        up_write(&zram->init_lock);
 }
 
-/* zram->init_lock should be held */
-int zram_init_device(struct zram *zram)
+void zram_meta_free(struct zram_meta *meta)
+{
+       zs_destroy_pool(meta->mem_pool);
+       kfree(meta->compress_workmem);
+       free_pages((unsigned long)meta->compress_buffer, 1);
+       vfree(meta->table);
+       kfree(meta);
+}
+
+struct zram_meta *zram_meta_alloc(u64 disksize)
 {
-       int ret;
        size_t num_pages;
+       struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
+       if (!meta)
+               goto out;
+
+       meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
+       if (!meta->compress_workmem) {
+               pr_err("Error allocating compressor working memory!\n");
+               goto free_meta;
+       }
+
+       meta->compress_buffer =
+               (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
+       if (!meta->compress_buffer) {
+               pr_err("Error allocating compressor buffer space\n");
+               goto free_workmem;
+       }
+
+       num_pages = disksize >> PAGE_SHIFT;
+       meta->table = vzalloc(num_pages * sizeof(*meta->table));
+       if (!meta->table) {
+               pr_err("Error allocating zram address table\n");
+               goto free_buffer;
+       }
+
+       meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
+       if (!meta->mem_pool) {
+               pr_err("Error creating memory pool\n");
+               goto free_table;
+       }
+
+       return meta;
+
+free_table:
+       vfree(meta->table);
+free_buffer:
+       free_pages((unsigned long)meta->compress_buffer, 1);
+free_workmem:
+       kfree(meta->compress_workmem);
+free_meta:
+       kfree(meta);
+       meta = NULL;
+out:
+       return meta;
+}
 
+void zram_init_device(struct zram *zram, struct zram_meta *meta)
+{
        if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
                pr_info(
                "There is little point creating a zram of greater than "
@@ -526,51 +573,13 @@ int zram_init_device(struct zram *zram)
                );
        }
 
-       zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
-       if (!zram->compress_workmem) {
-               pr_err("Error allocating compressor working memory!\n");
-               ret = -ENOMEM;
-               goto fail_no_table;
-       }
-
-       zram->compress_buffer =
-               (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
-       if (!zram->compress_buffer) {
-               pr_err("Error allocating compressor buffer space\n");
-               ret = -ENOMEM;
-               goto fail_no_table;
-       }
-
-       num_pages = zram->disksize >> PAGE_SHIFT;
-       zram->table = vzalloc(num_pages * sizeof(*zram->table));
-       if (!zram->table) {
-               pr_err("Error allocating zram address table\n");
-               ret = -ENOMEM;
-               goto fail_no_table;
-       }
-
        /* zram devices sort of resembles non-rotational disks */
        queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 
-       zram->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
-       if (!zram->mem_pool) {
-               pr_err("Error creating memory pool\n");
-               ret = -ENOMEM;
-               goto fail;
-       }
-
+       zram->meta = meta;
        zram->init_done = 1;
 
        pr_debug("Initialization done!\n");
-       return 0;
-
-fail_no_table:
-       /* To prevent accessing table entries during cleanup */
-       zram->disksize = 0;
-fail:
-       __zram_reset_device(zram);
-       pr_err("Initialization failed: err=%d\n", ret);
-       return ret;
 }
 
 static void zram_slot_free_notify(struct block_device *bdev,
index 5b671d1..2d1a3f1 100644 (file)
@@ -83,11 +83,15 @@ struct zram_stats {
        u32 bad_compress;       /* % of pages with compression ratio>=75% */
 };
 
-struct zram {
-       struct zs_pool *mem_pool;
+struct zram_meta {
        void *compress_workmem;
        void *compress_buffer;
        struct table *table;
+       struct zs_pool *mem_pool;
+};
+
+struct zram {
+       struct zram_meta *meta;
        spinlock_t stat64_lock; /* protect 64-bit stats */
        struct rw_semaphore lock; /* protect compression buffers and table
                                   * against concurrent read and writes */
@@ -111,7 +115,9 @@ unsigned int zram_get_num_devices(void);
 extern struct attribute_group zram_disk_attr_group;
 #endif
 
-extern int zram_init_device(struct zram *zram);
 extern void zram_reset_device(struct zram *zram);
+extern struct zram_meta *zram_meta_alloc(u64 disksize);
+extern void zram_meta_free(struct zram_meta *meta);
+extern void zram_init_device(struct zram *zram, struct zram_meta *meta);
 
 #endif
index 369db12..e6a929d 100644 (file)
@@ -56,22 +56,26 @@ static ssize_t disksize_store(struct device *dev,
                struct device_attribute *attr, const char *buf, size_t len)
 {
        u64 disksize;
+       struct zram_meta *meta;
        struct zram *zram = dev_to_zram(dev);
 
        disksize = memparse(buf, NULL);
        if (!disksize)
                return -EINVAL;
 
+       disksize = PAGE_ALIGN(disksize);
+       meta = zram_meta_alloc(disksize);
        down_write(&zram->init_lock);
        if (zram->init_done) {
                up_write(&zram->init_lock);
+               zram_meta_free(meta);
                pr_info("Cannot change disksize for initialized device\n");
                return -EBUSY;
        }
 
-       zram->disksize = PAGE_ALIGN(disksize);
+       zram->disksize = disksize;
        set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
-       zram_init_device(zram);
+       zram_init_device(zram, meta);
        up_write(&zram->init_lock);
 
        return len;
@@ -182,9 +186,10 @@ static ssize_t mem_used_total_show(struct device *dev,
 {
        u64 val = 0;
        struct zram *zram = dev_to_zram(dev);
+       struct zram_meta *meta = zram->meta;
 
        if (zram->init_done)
-               val = zs_get_total_size_bytes(zram->mem_pool);
+               val = zs_get_total_size_bytes(meta->mem_pool);
 
        return sprintf(buf, "%llu\n", val);
 }