mm/zsmalloc: add statistics support
authorGanesh Mahendran <opensource.ganesh@gmail.com>
Thu, 12 Feb 2015 23:00:54 +0000 (15:00 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 13 Feb 2015 02:54:12 +0000 (18:54 -0800)
Keeping fragmentation of zsmalloc in a low level is our target.  But now
we still need to add the debug code in zsmalloc to get the quantitative
data.

This patch adds a new configuration CONFIG_ZSMALLOC_STAT to enable the
statistics collection for developers.  Currently only the objects
statatitics in each class are collected.  User can get the information via
debugfs.

     cat /sys/kernel/debug/zsmalloc/zram0/...

For example:

After I copied "jdk-8u25-linux-x64.tar.gz" to zram with ext4 filesystem:
 class  size obj_allocated   obj_used pages_used
     0    32             0          0          0
     1    48           256         12          3
     2    64            64         14          1
     3    80            51          7          1
     4    96           128          5          3
     5   112            73          5          2
     6   128            32          4          1
     7   144             0          0          0
     8   160             0          0          0
     9   176             0          0          0
    10   192             0          0          0
    11   208             0          0          0
    12   224             0          0          0
    13   240             0          0          0
    14   256            16          1          1
    15   272            15          9          1
    16   288             0          0          0
    17   304             0          0          0
    18   320             0          0          0
    19   336             0          0          0
    20   352             0          0          0
    21   368             0          0          0
    22   384             0          0          0
    23   400             0          0          0
    24   416             0          0          0
    25   432             0          0          0
    26   448             0          0          0
    27   464             0          0          0
    28   480             0          0          0
    29   496            33          1          4
    30   512             0          0          0
    31   528             0          0          0
    32   544             0          0          0
    33   560             0          0          0
    34   576             0          0          0
    35   592             0          0          0
    36   608             0          0          0
    37   624             0          0          0
    38   640             0          0          0
    40   672             0          0          0
    42   704             0          0          0
    43   720            17          1          3
    44   736             0          0          0
    46   768             0          0          0
    49   816             0          0          0
    51   848             0          0          0
    52   864            14          1          3
    54   896             0          0          0
    57   944            13          1          3
    58   960             0          0          0
    62  1024             4          1          1
    66  1088            15          2          4
    67  1104             0          0          0
    71  1168             0          0          0
    74  1216             0          0          0
    76  1248             0          0          0
    83  1360             3          1          1
    91  1488            11          1          4
    94  1536             0          0          0
   100  1632             5          1          2
   107  1744             0          0          0
   111  1808             9          1          4
   126  2048             4          4          2
   144  2336             7          3          4
   151  2448             0          0          0
   168  2720            15         15         10
   190  3072            28         27         21
   202  3264             0          0          0
   254  4096         36209      36209      36209

 Total               37022      36326      36288

We can calculate the overall fragentation by the last line:
    Total               37022      36326      36288
    (37022 - 36326) / 37022 = 1.87%

Also by analysing objects alocated in every class we know why we got so
low fragmentation: Most of the allocated objects is in <class 254>.  And
there is only 1 page in class 254 zspage.  So, No fragmentation will be
introduced by allocating objs in class 254.

And in future, we can collect other zsmalloc statistics as we need and
analyse them.

Signed-off-by: Ganesh Mahendran <opensource.ganesh@gmail.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/Kconfig
mm/zsmalloc.c

index 4395b12..de5239c 100644 (file)
@@ -602,6 +602,16 @@ config PGTABLE_MAPPING
          You can check speed with zsmalloc benchmark:
          https://github.com/spartacus06/zsmapbench
 
+config ZSMALLOC_STAT
+       bool "Export zsmalloc statistics"
+       depends on ZSMALLOC
+       select DEBUG_FS
+       help
+         This option enables code in the zsmalloc to collect various
+         statistics about whats happening in zsmalloc and exports that
+         information to userspace via debugfs.
+         If unsure, say N.
+
 config GENERIC_EARLY_IOREMAP
        bool
 
index 2359e61..0dec1fa 100644 (file)
@@ -91,6 +91,7 @@
 #include <linux/hardirq.h>
 #include <linux/spinlock.h>
 #include <linux/types.h>
+#include <linux/debugfs.h>
 #include <linux/zsmalloc.h>
 #include <linux/zpool.h>
 
@@ -168,6 +169,22 @@ enum fullness_group {
        ZS_FULL
 };
 
+enum zs_stat_type {
+       OBJ_ALLOCATED,
+       OBJ_USED,
+       NR_ZS_STAT_TYPE,
+};
+
+#ifdef CONFIG_ZSMALLOC_STAT
+
+static struct dentry *zs_stat_root;
+
+struct zs_size_stat {
+       unsigned long objs[NR_ZS_STAT_TYPE];
+};
+
+#endif
+
 /*
  * number of size_classes
  */
@@ -200,6 +217,10 @@ struct size_class {
        /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */
        int pages_per_zspage;
 
+#ifdef CONFIG_ZSMALLOC_STAT
+       struct zs_size_stat stats;
+#endif
+
        spinlock_t lock;
 
        struct page *fullness_list[_ZS_NR_FULLNESS_GROUPS];
@@ -217,10 +238,16 @@ struct link_free {
 };
 
 struct zs_pool {
+       char *name;
+
        struct size_class **size_class;
 
        gfp_t flags;    /* allocation flags used when growing pool */
        atomic_long_t pages_allocated;
+
+#ifdef CONFIG_ZSMALLOC_STAT
+       struct dentry *stat_dentry;
+#endif
 };
 
 /*
@@ -942,6 +969,166 @@ static bool can_merge(struct size_class *prev, int size, int pages_per_zspage)
        return true;
 }
 
+#ifdef CONFIG_ZSMALLOC_STAT
+
+static inline void zs_stat_inc(struct size_class *class,
+                               enum zs_stat_type type, unsigned long cnt)
+{
+       class->stats.objs[type] += cnt;
+}
+
+static inline void zs_stat_dec(struct size_class *class,
+                               enum zs_stat_type type, unsigned long cnt)
+{
+       class->stats.objs[type] -= cnt;
+}
+
+static inline unsigned long zs_stat_get(struct size_class *class,
+                               enum zs_stat_type type)
+{
+       return class->stats.objs[type];
+}
+
+static int __init zs_stat_init(void)
+{
+       if (!debugfs_initialized())
+               return -ENODEV;
+
+       zs_stat_root = debugfs_create_dir("zsmalloc", NULL);
+       if (!zs_stat_root)
+               return -ENOMEM;
+
+       return 0;
+}
+
+static void __exit zs_stat_exit(void)
+{
+       debugfs_remove_recursive(zs_stat_root);
+}
+
+static int zs_stats_size_show(struct seq_file *s, void *v)
+{
+       int i;
+       struct zs_pool *pool = s->private;
+       struct size_class *class;
+       int objs_per_zspage;
+       unsigned long obj_allocated, obj_used, pages_used;
+       unsigned long total_objs = 0, total_used_objs = 0, total_pages = 0;
+
+       seq_printf(s, " %5s %5s %13s %10s %10s\n", "class", "size",
+                               "obj_allocated", "obj_used", "pages_used");
+
+       for (i = 0; i < zs_size_classes; i++) {
+               class = pool->size_class[i];
+
+               if (class->index != i)
+                       continue;
+
+               spin_lock(&class->lock);
+               obj_allocated = zs_stat_get(class, OBJ_ALLOCATED);
+               obj_used = zs_stat_get(class, OBJ_USED);
+               spin_unlock(&class->lock);
+
+               objs_per_zspage = get_maxobj_per_zspage(class->size,
+                               class->pages_per_zspage);
+               pages_used = obj_allocated / objs_per_zspage *
+                               class->pages_per_zspage;
+
+               seq_printf(s, " %5u %5u    %10lu %10lu %10lu\n", i,
+                       class->size, obj_allocated, obj_used, pages_used);
+
+               total_objs += obj_allocated;
+               total_used_objs += obj_used;
+               total_pages += pages_used;
+       }
+
+       seq_puts(s, "\n");
+       seq_printf(s, " %5s %5s    %10lu %10lu %10lu\n", "Total", "",
+                       total_objs, total_used_objs, total_pages);
+
+       return 0;
+}
+
+static int zs_stats_size_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, zs_stats_size_show, inode->i_private);
+}
+
+static const struct file_operations zs_stat_size_ops = {
+       .open           = zs_stats_size_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static int zs_pool_stat_create(char *name, struct zs_pool *pool)
+{
+       struct dentry *entry;
+
+       if (!zs_stat_root)
+               return -ENODEV;
+
+       entry = debugfs_create_dir(name, zs_stat_root);
+       if (!entry) {
+               pr_warn("debugfs dir <%s> creation failed\n", name);
+               return -ENOMEM;
+       }
+       pool->stat_dentry = entry;
+
+       entry = debugfs_create_file("obj_in_classes", S_IFREG | S_IRUGO,
+                       pool->stat_dentry, pool, &zs_stat_size_ops);
+       if (!entry) {
+               pr_warn("%s: debugfs file entry <%s> creation failed\n",
+                               name, "obj_in_classes");
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+static void zs_pool_stat_destroy(struct zs_pool *pool)
+{
+       debugfs_remove_recursive(pool->stat_dentry);
+}
+
+#else /* CONFIG_ZSMALLOC_STAT */
+
+static inline void zs_stat_inc(struct size_class *class,
+                               enum zs_stat_type type, unsigned long cnt)
+{
+}
+
+static inline void zs_stat_dec(struct size_class *class,
+                               enum zs_stat_type type, unsigned long cnt)
+{
+}
+
+static inline unsigned long zs_stat_get(struct size_class *class,
+                               enum zs_stat_type type)
+{
+       return 0;
+}
+
+static int __init zs_stat_init(void)
+{
+       return 0;
+}
+
+static void __exit zs_stat_exit(void)
+{
+}
+
+static inline int zs_pool_stat_create(char *name, struct zs_pool *pool)
+{
+       return 0;
+}
+
+static inline void zs_pool_stat_destroy(struct zs_pool *pool)
+{
+}
+
+#endif
+
 unsigned long zs_get_total_pages(struct zs_pool *pool)
 {
        return atomic_long_read(&pool->pages_allocated);
@@ -1074,7 +1261,10 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size)
                set_zspage_mapping(first_page, class->index, ZS_EMPTY);
                atomic_long_add(class->pages_per_zspage,
                                        &pool->pages_allocated);
+
                spin_lock(&class->lock);
+               zs_stat_inc(class, OBJ_ALLOCATED, get_maxobj_per_zspage(
+                               class->size, class->pages_per_zspage));
        }
 
        obj = (unsigned long)first_page->freelist;
@@ -1088,6 +1278,7 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size)
        kunmap_atomic(vaddr);
 
        first_page->inuse++;
+       zs_stat_inc(class, OBJ_USED, 1);
        /* Now move the zspage to another fullness group, if required */
        fix_fullness_group(pool, first_page);
        spin_unlock(&class->lock);
@@ -1128,6 +1319,12 @@ void zs_free(struct zs_pool *pool, unsigned long obj)
 
        first_page->inuse--;
        fullness = fix_fullness_group(pool, first_page);
+
+       zs_stat_dec(class, OBJ_USED, 1);
+       if (fullness == ZS_EMPTY)
+               zs_stat_dec(class, OBJ_ALLOCATED, get_maxobj_per_zspage(
+                               class->size, class->pages_per_zspage));
+
        spin_unlock(&class->lock);
 
        if (fullness == ZS_EMPTY) {
@@ -1158,9 +1355,16 @@ struct zs_pool *zs_create_pool(char *name, gfp_t flags)
        if (!pool)
                return NULL;
 
+       pool->name = kstrdup(name, GFP_KERNEL);
+       if (!pool->name) {
+               kfree(pool);
+               return NULL;
+       }
+
        pool->size_class = kcalloc(zs_size_classes, sizeof(struct size_class *),
                        GFP_KERNEL);
        if (!pool->size_class) {
+               kfree(pool->name);
                kfree(pool);
                return NULL;
        }
@@ -1210,6 +1414,9 @@ struct zs_pool *zs_create_pool(char *name, gfp_t flags)
 
        pool->flags = flags;
 
+       if (zs_pool_stat_create(name, pool))
+               goto err;
+
        return pool;
 
 err:
@@ -1222,6 +1429,8 @@ void zs_destroy_pool(struct zs_pool *pool)
 {
        int i;
 
+       zs_pool_stat_destroy(pool);
+
        for (i = 0; i < zs_size_classes; i++) {
                int fg;
                struct size_class *class = pool->size_class[i];
@@ -1242,6 +1451,7 @@ void zs_destroy_pool(struct zs_pool *pool)
        }
 
        kfree(pool->size_class);
+       kfree(pool->name);
        kfree(pool);
 }
 EXPORT_SYMBOL_GPL(zs_destroy_pool);
@@ -1250,17 +1460,30 @@ static int __init zs_init(void)
 {
        int ret = zs_register_cpu_notifier();
 
-       if (ret) {
-               zs_unregister_cpu_notifier();
-               return ret;
-       }
+       if (ret)
+               goto notifier_fail;
 
        init_zs_size_classes();
 
 #ifdef CONFIG_ZPOOL
        zpool_register_driver(&zs_zpool_driver);
 #endif
+
+       ret = zs_stat_init();
+       if (ret) {
+               pr_err("zs stat initialization failed\n");
+               goto stat_fail;
+       }
        return 0;
+
+stat_fail:
+#ifdef CONFIG_ZPOOL
+       zpool_unregister_driver(&zs_zpool_driver);
+#endif
+notifier_fail:
+       zs_unregister_cpu_notifier();
+
+       return ret;
 }
 
 static void __exit zs_exit(void)
@@ -1269,6 +1492,8 @@ static void __exit zs_exit(void)
        zpool_unregister_driver(&zs_zpool_driver);
 #endif
        zs_unregister_cpu_notifier();
+
+       zs_stat_exit();
 }
 
 module_init(zs_init);