Re: [PATCH 0/2] memcg: add the soft_limit reclaim in global direct reclaim

[Ying Han <yinghan@google.com>]

On Fri, Apr 29, 2011 at 10:19 AM, Ying Han <yinghan@google.com> wrote:
> On Fri, Apr 29, 2011 at 9:44 AM, Minchan Kim <minchan.kim@gmail.com> wrote:
>> Hi Ying,
>>
>> On Thu, Apr 28, 2011 at 03:37:04PM -0700, Ying Han wrote:
>>> We recently added the change in global background reclaim which counts the
>>> return value of soft_limit reclaim. Now this patch adds the similar logic
>>> on global direct reclaim.
>>>
>>> We should skip scanning global LRU on shrink_zone if soft_limit reclaim does
>>> enough work. This is the first step where we start with counting the nr_scanned
>>> and nr_reclaimed from soft_limit reclaim into global scan_control.
>>>
>>> The patch is based on mmotm-04-14 and i triggered kernel BUG at mm/vmscan.c:1058!
>>
>> Could you tell me exact patches?
>> mmtom-04-14 + just 2 patch of this? or + something?
>>
>> These day, You and Kame produces many patches.
>> Do I have to apply something of them?
> No, I applied my patch on top of mmotm and here is the last commit
> before my patch.
>
> commit 66a3827927351e0f88dc391919cf0cda10d42dd7
> Author: Andrew Morton <akpm@linux-foundation.org>
> Date:   Thu Apr 14 15:51:34 2011 -0700

sorry, please ignore the last post. I learned that the mmotm i posted
is based on
tag: mmotm-2011-04-14-15-08.

--Ying


>
>>
>>>
>>> [  938.242033] kernel BUG at mm/vmscan.c:1058!
>>> [  938.242033] invalid opcode: 0000 [#1] SMP·
>>> [  938.242033] last sysfs file: /sys/devices/pci0000:00/0000:00:1f.2/device
>>> [  938.242033] Pid: 546, comm: kswapd0 Tainted: G        W   2.6.39-smp-direct_reclaim
>>> [  938.242033] RIP: 0010:[<ffffffff810ed174>]  [<ffffffff810ed174>] isolate_pages_global+0x18c/0x34f
>>> [  938.242033] RSP: 0018:ffff88082f83bb50  EFLAGS: 00010082
>>> [  938.242033] RAX: 00000000ffffffea RBX: ffff88082f83bc90 RCX: 0000000000000401
>>> [  938.242033] RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffffea001ca653e8
>>> [  938.242033] RBP: ffff88082f83bc20 R08: 0000000000000000 R09: ffff88085ffb6e00
>>> [  938.242033] R10: ffff88085ffb73d0 R11: ffff88085ffb6e00 R12: ffff88085ffb6e00
>>> [  938.242033] R13: ffffea001ca65410 R14: 0000000000000001 R15: ffffea001ca653e8
>>> [  938.242033] FS:  0000000000000000(0000) GS:ffff88085fd00000(0000) knlGS:0000000000000000
>>> [  938.242033] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
>>> [  938.242033] CR2: 00007f5c3405c320 CR3: 0000000001803000 CR4: 00000000000006e0
>>> [  938.242033] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>>> [  938.242033] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
>>> [  938.242033] Process kswapd0 (pid: 546, threadinfo ffff88082f83a000, task ffff88082fe52080)
>>> [  938.242033] Stack:
>>> [  938.242033]  ffff88085ffb6e00 ffffea0000000002 0000000000000021 0000000000000000
>>> [  938.242033]  0000000000000000 ffff88082f83bcb8 ffffea00108eec80 ffffea00108eecb8
>>> [  938.242033]  ffffea00108eecf0 0000000000000004 fffffffffffffffc 0000000000000020
>>> [  938.242033] Call Trace:
>>> [  938.242033]  [<ffffffff810ee8a5>] shrink_inactive_list+0x185/0x418
>>> [  938.242033]  [<ffffffff810366cc>] ? __switch_to+0xea/0x212
>>> [  938.242033]  [<ffffffff810e8b35>] ? determine_dirtyable_memory+0x1a/0x2c
>>> [  938.242033]  [<ffffffff810ef19b>] shrink_zone+0x380/0x44d
>>> [  938.242033]  [<ffffffff810e5188>] ? zone_watermark_ok_safe+0xa1/0xae
>>> [  938.242033]  [<ffffffff810efbd8>] kswapd+0x41b/0x76b
>>> [  938.242033]  [<ffffffff810ef7bd>] ? zone_reclaim+0x2fb/0x2fb
>>> [  938.242033]  [<ffffffff81088569>] kthread+0x82/0x8a
>>> [  938.242033]  [<ffffffff8141b0d4>] kernel_thread_helper+0x4/0x10
>>> [  938.242033]  [<ffffffff810884e7>] ? kthread_worker_fn+0x112/0x112
>>> [  938.242033]  [<ffffffff8141b0d0>] ? gs_change+0xb/0xb
>>>
>>
>> It seems there is active page in inactive list.
>> As I look deactivate_page, lru_deactivate_fn clears PageActive before
>> add_page_to_lru_list and it should be protected by zone->lru_lock.
>> In addiion, PageLRU would protect with race with isolation functions.
>>
>> Hmm, I don't have any clue now.
>> Is it reproducible easily?
> I can manage to reproduce it on my host by adding lots of memory
> pressure and then trigger the global
> reclaim.
>
>>
>> Could you apply below debugging patch and report the result?
>>
>> diff --git a/include/linux/mm_inline.h b/include/linux/mm_inline.h
>> index 8f7d247..f39b53a 100644
>> --- a/include/linux/mm_inline.h
>> +++ b/include/linux/mm_inline.h
>> @@ -25,6 +25,8 @@ static inline void
>>  __add_page_to_lru_list(struct zone *zone, struct page *page, enum lru_list l,
>>                       struct list_head *head)
>>  {
>> +       VM_BUG_ON(PageActive(page) && (
>> +                       l == LRU_INACTIVE_ANON || l == LRU_INACTIVE_FILE));
>>        list_add(&page->lru, head);
>>        __mod_zone_page_state(zone, NR_LRU_BASE + l, hpage_nr_pages(page));
>>        mem_cgroup_add_lru_list(page, l);
>> diff --git a/mm/swap.c b/mm/swap.c
>> index a83ec5a..5f7c3c8 100644
>> --- a/mm/swap.c
>> +++ b/mm/swap.c
>> @@ -454,6 +454,8 @@ static void lru_deactivate_fn(struct page *page, void *arg)
>>                 * The page's writeback ends up during pagevec
>>                 * We moves tha page into tail of inactive.
>>                 */
>> +               VM_BUG_ON(PageActive(page) && (
>> +                       lru == LRU_INACTIVE_ANON || lru == LRU_INACTIVE_FILE));
>>                list_move_tail(&page->lru, &zone->lru[lru].list);
>>                mem_cgroup_rotate_reclaimable_page(page);
>>                __count_vm_event(PGROTATED);
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index b3a569f..3415896 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -963,7 +963,7 @@ int __isolate_lru_page(struct page *page, int mode, int file)
>>
>>        /* Only take pages on the LRU. */
>>        if (!PageLRU(page))
>> -               return ret;
>> +               return 1;
>>
>>        /*
>>         * When checking the active state, we need to be sure we are
>> @@ -971,10 +971,10 @@ int __isolate_lru_page(struct page *page, int mode, int file)
>>         * of each.
>>         */
>>        if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
>> -               return ret;
>> +               return 2;
>>
>>        if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
>> -               return ret;
>> +               return 3;
>>
>>        /*
>>         * When this function is being called for lumpy reclaim, we
>> @@ -982,7 +982,7 @@ int __isolate_lru_page(struct page *page, int mode, int file)
>>         * unevictable; only give shrink_page_list evictable pages.
>>         */
>>        if (PageUnevictable(page))
>> -               return ret;
>> +               return 4;
>>
>>        ret = -EBUSY;
>>
>> @@ -1035,13 +1035,14 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
>>                unsigned long end_pfn;
>>                unsigned long page_pfn;
>>                int zone_id;
>> +               int ret;
>>
>>                page = lru_to_page(src);
>>                prefetchw_prev_lru_page(page, src, flags);
>>
>>                VM_BUG_ON(!PageLRU(page));
>>
>> -               switch (__isolate_lru_page(page, mode, file)) {
>> +               switch (ret = __isolate_lru_page(page, mode, file)) {
>>                case 0:
>>                        list_move(&page->lru, dst);
>>                        mem_cgroup_del_lru(page);
>> @@ -1055,6 +1056,7 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
>>                        continue;
>>
>>                default:
>> +                       printk(KERN_ERR "ret %d\n", ret);
>>                        BUG();
>>                }
>>
>>> Thank you Minchan for the pointer. I reverted the following commit and I
>>> haven't seen the problem with the same operation. I haven't looked deeply
>>> on the patch yet, but figured it would be a good idea to post the dump.
>>> The dump looks not directly related to this patchset, but ppl can use it to
>>> reproduce the problem.
>>
>> I tested the patch with rsync + fadvise several times
>> in my machine(2P, 2G DRAM) but I didn't have ever seen the BUG.
>> But I didn't test it in memcg. As I look dump, it seems not related to memcg.
>> Anyway, I tried it to reproduce it in my machine.
>> Maybe I will start testing after next week. Sorry.
>>
>> I hope my debugging patch givse some clues.
>> Thanks for the reporting, Ying.
>
> Sure, i will try the patch and post the result.
>
> --Ying
>
>> --
>> Kind regards,
>> Minchan Kim
>>
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>