Re: [PATCH 2/2] mm/zsmalloc: add statistics support
From: Ganesh Mahendran
Date: Thu Dec 18 2014 - 20:57:07 EST
2014-12-19 9:30 GMT+08:00 Minchan Kim <minchan@xxxxxxxxxx>:
> On Fri, Dec 19, 2014 at 09:21:52AM +0800, Ganesh Mahendran wrote:
>> Hello, Minchan
>>
>> 2014-12-19 7:44 GMT+08:00 Minchan Kim <minchan@xxxxxxxxxx>:
>> > Hello Ganesh,
>> >
>> > On Wed, Dec 17, 2014 at 03:05:19PM +0800, Ganesh Mahendran wrote:
>> >> Hello, Minchan
>> >>
>> >> Thanks for your review.
>> >>
>> >> 2014-12-16 10:45 GMT+08:00 Minchan Kim <minchan@xxxxxxxxxx>:
>> >> > On Sat, Dec 13, 2014 at 09:45:14PM +0800, Ganesh Mahendran wrote:
>> >> >> As a ram based memory allocator, keep the fragmentation in a low level
>> >> >
>> >> > Just say, zsmalloc.
>> >>
>> >> Ok.
>> >>
>> >> >
>> >> >> is our target. But now we still need to add the debug code in zsmalloc
>> >> >> to get the quantitative data.
>> >> >>
>> >> >> After the RFC patch [1], Minchan Kim gave some suggestions.
>> >> >> [1] https://patchwork.kernel.org/patch/5469301/
>> >> >>
>> >> >> This patch adds a new configuration CONFIG_ZSMALLOC_STAT to enable the statistics
>> >> >> collection for developers. Currently only the objects information in each class
>> >> >> are collected. User can get the information via debugfs. For example:
>> >> >>
>> >> >> After I copy file 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 see the overall fragentation is:
>> >> >> (37022 - 36326) / 37022 = 1.87%
>> >> >>
>> >> >> Also from the statistics we know why we got so low fragmentation:
>> >> >> Most of the objects is in class 254 with size 4096 Bytes. The pages in
>> >> >> zspage is 1. And there is only one object in a page. So, No fragmentation
>> >> >> will be produced.
>> >> >>
>> >> >> Also we can collect other information and show it to user in the future.
>> >> >
>> >> > So, could you make zs
>> >> Ok
>> >> >>
>> >> >> Signed-off-by: Ganesh Mahendran <opensource.ganesh@xxxxxxxxx>
>> >> >> Suggested-by: Minchan Kim <minchan@xxxxxxxxxx>
>> >> >> ---
>> >> >> mm/Kconfig | 10 ++++
>> >> >> mm/zsmalloc.c | 164 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>> >> >> 2 files changed, 174 insertions(+)
>> >> >>
>> >> >> diff --git a/mm/Kconfig b/mm/Kconfig
>> >> >> index 1d1ae6b..95c5728 100644
>> >> >> --- a/mm/Kconfig
>> >> >> +++ b/mm/Kconfig
>> >> >> @@ -601,6 +601,16 @@ 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
>> >> >>
>> >> >> diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
>> >> >> index b724039..a8d0020 100644
>> >> >> --- a/mm/zsmalloc.c
>> >> >> +++ b/mm/zsmalloc.c
>> >> >> @@ -168,6 +168,8 @@ enum fullness_group {
>> >> >> ZS_FULL
>> >> >> };
>> >> >>
>> >> >> +static int zs_pool_num;
>> >> >
>> >> > What's this? What protects the race?
>> >> It is the pool index. Yes, there is problem here.
>> >> I will change it to atomic and increased every time a new zs pool created.
>> >> And then name the /sys/kernel/debug/pool-x using this index.
>> >>
>> >> static atomic_t zs_pool_index = ATOMIC_INIT(0);
>> >> ...
>> >> pool->index = atomic_inc_return(&zs_pool_index);
>> >>
>> >> > It means description.
>> >> >
>> >> >> +
>> >> >> /*
>> >> >> * number of size_classes
>> >> >> */
>> >> >> @@ -200,6 +202,11 @@ struct size_class {
>> >> >> /* Number of PAGE_SIZE sized pages to combine to form a 'zspage' */
>> >> >> int pages_per_zspage;
>> >> >>
>> >> >> +#ifdef CONFIG_ZSMALLOC_STAT
>> >> >> + unsigned long obj_allocated;
>> >> >> + unsigned long obj_used;
>> >> >> +#endif
>> >> >
>> >> > I perfer creating new struct.
>> >> >
>> >> > struct zs_size_stat {
>> >> > unsigend long obj_allocated;
>> >> > unsignged long obj_used;
>> >> > };
>> >>
>> >> Got it, I will redo this.
>> >>
>> >> >
>> >> >> +
>> >> >> spinlock_t lock;
>> >> >>
>> >> >> struct page *fullness_list[_ZS_NR_FULLNESS_GROUPS];
>> >> >> @@ -221,6 +228,10 @@ struct zs_pool {
>> >> >>
>> >> >> 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 +953,132 @@ static bool can_merge(struct size_class *prev, int size, int pages_per_zspage)
>> >> >> return true;
>> >> >> }
>> >> >>
>> >> >> +
>> >> >> +#ifdef CONFIG_ZSMALLOC_STAT
>> >> >> +#include <linux/debugfs.h>
>> >> >
>> >> > A question:
>> >> > Why "#include" is here instead of top on the source file?
>> >>
>> >> Yes, the "#include ..." should be on the top of the source file.
>> >> I will modify it.
>> >>
>> >> >
>> >> >> +
>> >> >> +static struct dentry *zs_stat_root;
>> >> >> +
>> >> >> +static int __init zs_stat_init(void)
>> >> >> +{
>> >> >> + if (!debugfs_initialized())
>> >> >> + return -ENODEV;
>> >> >
>> >> > Do we need above check?
>> >> Yes, I think we need this check.
>> >> When debugfs module init failed, we should not go ahead here.
>> >>
>> >> >
>> >> > When I read comment of debugfs_create_dir, it says
>> >> > "If debugfs is not enabled in the kernel, the value -%ENODEV will be
>> >> > returned."
>> >>
>> >> This check is not for the situation when the debugfs is not enabled.
>> >> But for if we failed in debugfs_init(), then
>> >> we should not use any API of debugfs.
>> >>
>> >> And I think "-%ENODEV will be returned" means below code in "debugfs.h"
>> >>
>> >> static inline struct dentry *debugfs_create_dir(const char *name,
>> >> struct dentry *parent)
>> >> {
>> >> return ERR_PTR(-ENODEV);
>> >> }
>> >>
>> >> >
>> >> >> +
>> >> >> + zs_stat_root = debugfs_create_dir("zsmalloc", NULL);
>> >> >> + if (!zs_stat_root)
>> >> >> + return -ENOMEM;
>> >> >
>> >> > On null return of debugfs_create_dir, it means always ENOMEM?
>> >>
>> >> Yes, you are right. -ENOMEM is not the only reason for the failure.
>> >> But debugfs_create_dir does not bring back the errno.
>> >> And for zsmalloc, we indeed have the permission(-EPERM) to create the entry and
>> >> also we will not create duplicate(-EEXIST) entry in debufs.
>> >>
>> >> So, I think -ENOMEM is suitable.
>> >
>> > It seems you are saying why debugfs_create_dir can fail but zsmalloc will not
>> > fail by such reasons.
>> > I don't know the internal of debugfs_create_dir but description just says
>> > "If an error occurs, %NULL will be returned" but everyone returns ENOMEM
>> > blindly then. Hmm, I don't think it's good but I'm okay too because it's not
>> > our fault and others have been in there. :(
>> >
>> > When I look at zs_init, you don't propagate the error to user. Just missing?
>>
>> As the statistics collection is not essential to user, and I think we
>> do not need to break
>> the zsmalloc module loading when error happens in
>> zs_stat_init()/debugfs_create_dir().
>> So I do not propagate the error to user and just give the message:
>> pr_warn("zs stat initialization failed\n");
>
> Hmm? So you want to work without stat although user want?
It seems we should allow user to know there is something wrong
happened in module loading.
In the beginning, I just considered that it was only a debug option.
When developers enable
this option in the end-user's machine, It should not affect zsmalloc's
basic function.
> It means every zs_stat_[inc|dec] should check if it was initialized properly?
In zs_stat_[inc|dec], these functions only handle the fields in struct
size_class{}. So we do not need
to check whether it was .initialized properly. But when failed in
zs_stat_init(), there is
no way to show these statistics to user.
>
>>
>> Thanks.
>>
>> >
>> >>
>> >> >
>> >> >> +
>> >> >> + return 0;
>> >> >> +}
>> >> >> +
>> >> >> +static void __exit zs_stat_exit(void)
>> >> >> +{
>> >> >> + debugfs_remove_recursive(zs_stat_root);
>> >> >> +}
>> >> >> +
>> >> >> +static int zs_stats_show(struct seq_file *s, void *v)
>> >> >> +{
>> >> >> + int i;
>> >> >> + struct zs_pool *pool = (struct zs_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 = class->obj_allocated;
>> >> >> + obj_used = class->obj_used;
>> >> >> + objs_per_zspage = get_maxobj_per_zspage(class->size,
>> >> >> + class->pages_per_zspage);
>> >> >> + pages_used = obj_allocated / objs_per_zspage *
>> >> >> + class->pages_per_zspage;
>> >> >
>> >> > I think We don't need to protect class->pages_per_zspage with class->lock.
>> >>
>> >> Yes, you are right.
>> >>
>> >> >
>> >> >> +
>> >> >> + spin_unlock(&class->lock);
>> >> >> +
>> >> >> + seq_printf(s, " %5u %5u %10lu %10lu %10lu\n", i, class->size,
>> >> >> + obj_allocated, obj_used, pages_used);
>> >> >> +
>> >> >> + total_objs += class->obj_allocated;
>> >> >> + total_used_objs += class->obj_used;
>> >> >
>> >> > You couldn't access class->fields without class lock.
>> >> > Please, assign them into local variable under the lock and sum them without the lock.
>> >>
>> >> Got it. I will redo this.
>> >>
>> >> >
>> >> >> + 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_open(struct inode *inode, struct file *file)
>> >> >> +{
>> >> >> + return single_open(file, zs_stats_show, inode->i_private);
>> >> >> +}
>> >> >> +
>> >> >> +static const struct file_operations zs_stats_operations = {
>> >> >> + .open = zs_stats_open,
>> >> >> + .read = seq_read,
>> >> >> + .llseek = seq_lseek,
>> >> >> + .release = single_release,
>> >> >> +};
>> >> >> +
>> >> >> +static int zs_pool_stat_create(struct zs_pool *pool, int index)
>> >> >> +{
>> >> >> + char name[10];
>> >> >> + int ret = 0;
>> >> >> +
>> >> >> + if (!zs_stat_root) {
>> >> >> + ret = -ENODEV;
>> >> >> + goto out;
>> >> >> + }
>> >> >> +
>> >> >> + snprintf(name, sizeof(name), "pool-%d", index);
>> >> >
>> >> > Hmm, how does admin know any zsmalloc instance is associated with
>> >> > any block device?
>> >> > Maybe we need export zspool index to the client and print it
>> >> > when pool is populated.
>> >>
>> >> Thanks for your suggestion.
>> >>
>> >> >
>> >> >> + pool->stat_dentry = debugfs_create_dir(name, zs_stat_root);
>> >> >> + if (!pool->stat_dentry) {
>> >> >> + ret = -ENOMEM;
>> >> >> + goto out;
>> >> >> + }
>> >> >> +
>> >> >> + debugfs_create_file("obj_in_classes", S_IFREG | S_IRUGO,
>> >> >> + pool->stat_dentry, pool, &zs_stats_operations);
>> >> >
>> >> > No need to check return?
>> >>
>> >> It is better to check the return value and give user some information
>> >> about the failure.
>> >>
>> >> >
>> >> >> +
>> >> >> +out:
>> >> >> + return ret;
>> >> >> +}
>> >> >> +
>> >> >> +static void zs_pool_stat_destroy(struct zs_pool *pool)
>> >> >> +{
>> >> >> + debugfs_remove_recursive(pool->stat_dentry);
>> >> >> +}
>> >> >> +
>> >> >> +#else /* CONFIG_ZSMALLOC_STAT */
>> >> >> +
>> >> >> +static int __init zs_stat_init(void)
>> >> >> +{
>> >> >> + return 0;
>> >> >> +}
>> >> >> +
>> >> >> +static void __exit zs_stat_exit(void) { }
>> >> >> +
>> >> >> +static inline int zs_pool_stat_create(struct zs_pool *pool, int index)
>> >> >> +{
>> >> >> + 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);
>> >> >> @@ -1075,6 +1212,10 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size)
>> >> >> atomic_long_add(class->pages_per_zspage,
>> >> >> &pool->pages_allocated);
>> >> >> spin_lock(&class->lock);
>> >> >> +#ifdef CONFIG_ZSMALLOC_STAT
>> >> >> + class->obj_allocated += get_maxobj_per_zspage(class->size,
>> >> >> + class->pages_per_zspage);
>> >> >> +#endif
>> >> >
>> >> > I prefer zs_stat_inc(class, OBJ_ALLOCATED, get_max_obj());
>> >>
>> >> Got it. thanks
>> >>
>> >> >
>> >> >> }
>> >> >>
>> >> >> obj = (unsigned long)first_page->freelist;
>> >> >> @@ -1088,6 +1229,9 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size)
>> >> >> kunmap_atomic(vaddr);
>> >> >>
>> >> >> first_page->inuse++;
>> >> >> +#ifdef CONFIG_ZSMALLOC_STAT
>> >> >> + class->obj_used++;
>> >> >> +#endif
>> >> >
>> >> > zs_stat_inc(class, OBJ_USED, 1)
>> >>
>> >> OK
>> >>
>> >> >
>> >> >
>> >> >> /* Now move the zspage to another fullness group, if required */
>> >> >> fix_fullness_group(pool, first_page);
>> >> >> spin_unlock(&class->lock);
>> >> >> @@ -1127,12 +1271,19 @@ void zs_free(struct zs_pool *pool, unsigned long obj)
>> >> >> first_page->freelist = (void *)obj;
>> >> >>
>> >> >> first_page->inuse--;
>> >> >> +#ifdef CONFIG_ZSMALLOC_STAT
>> >> >> + class->obj_used--;
>> >> >> +#endif
>> >> >
>> >> > zs_stat_dec(class, OBJ_USED, 1)
>> >>
>> >> OK
>> >>
>> >> >
>> >> >> fullness = fix_fullness_group(pool, first_page);
>> >> >> spin_unlock(&class->lock);
>> >> >>
>> >> >> if (fullness == ZS_EMPTY) {
>> >> >> atomic_long_sub(class->pages_per_zspage,
>> >> >> &pool->pages_allocated);
>> >> >> +#ifdef CONFIG_ZSMALLOC_STAT
>> >> >> + class->obj_allocated -= get_maxobj_per_zspage(class->size,
>> >> >> + class->pages_per_zspage);
>> >> >> +#endif
>> >> >
>> >> > zs_stat_dec(class, OBJ_ALLOCATED, get_max_obj());
>> >> >
>> >> >> free_zspage(first_page);
>> >> >> }
>> >> >> }
>> >> >> @@ -1209,6 +1360,10 @@ struct zs_pool *zs_create_pool(gfp_t flags)
>> >> >> }
>> >> >>
>> >> >> pool->flags = flags;
>> >> >> + zs_pool_num++;
>> >> >
>> >> > Who protect the race?
>> >> > And manybe we should keep the index in zspool and export it to the user
>> >> > to let them know what zsmalloc instance is thiers.
>> >>
>> >> OK
>> >>
>> >> >
>> >> >> +
>> >> >> + if (zs_pool_stat_create(pool, zs_pool_num))
>> >> >> + pr_warn("zs pool %d stat initialization failed\n", zs_pool_num);
>> >> >>
>> >> >> return pool;
>> >> >>
>> >> >> @@ -1241,6 +1396,9 @@ void zs_destroy_pool(struct zs_pool *pool)
>> >> >> kfree(class);
>> >> >> }
>> >> >>
>> >> >> + zs_pool_stat_destroy(pool);
>> >> >> + zs_pool_num--;
>> >> >> +
>> >> >> kfree(pool->size_class);
>> >> >> kfree(pool);
>> >> >> }
>> >> >> @@ -1260,6 +1418,10 @@ static int __init zs_init(void)
>> >> >> #ifdef CONFIG_ZPOOL
>> >> >> zpool_register_driver(&zs_zpool_driver);
>> >> >> #endif
>> >> >> +
>> >> >> + if (zs_stat_init())
>> >> >> + pr_warn("zs stat initialization failed\n");
>> >> >> +
>> >> >> return 0;
>> >> >> }
>> >> >>
>> >> >> @@ -1269,6 +1431,8 @@ static void __exit zs_exit(void)
>> >> >> zpool_unregister_driver(&zs_zpool_driver);
>> >> >> #endif
>> >> >> zs_unregister_cpu_notifier();
>> >> >> +
>> >> >> + zs_stat_exit();
>> >> >> }
>> >> >>
>> >> >> module_init(zs_init);
>> >> >> --
>> >> >> 1.7.9.5
>> >> >>
>> >>
>> >> --
>> >> To unsubscribe, send a message with 'unsubscribe linux-mm' in
>> >> the body to majordomo@xxxxxxxxxx For more info on Linux MM,
>> >> see: http://www.linux-mm.org/ .
>> >> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
>> >
>> > --
>> > Kind regards,
>> > Minchan Kim
>>
>> --
>> To unsubscribe, send a message with 'unsubscribe linux-mm' in
>> the body to majordomo@xxxxxxxxxx For more info on Linux MM,
>> see: http://www.linux-mm.org/ .
>> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
>
> --
> Kind regards,
> Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/