Re: f2fs page flag set with private but page.private pointer is NULL

From: Jaegeuk Kim <jaegeuk@kernel.org>
To: Kassey <kassey1216@gmail.com>
Cc: linux-f2fs-devel@lists.sourceforge.net, Jonathan Corbet <corbet@lwn.net>
Subject: Re: f2fs page flag set with private but page.private pointer is NULL
Date: Mon, 17 Sep 2018 19:32:27 -0700	[thread overview]
Message-ID: <20180918023227.GA85192@jaegeuk-macbookpro.roam.corp.google.com> (raw)
In-Reply-To: <CAKwPUoy9QQfbsi7P+_-zyX_dg=mw-gmvSWgLGDDYoVcmh-_=nw@mail.gmail.com>

Hi Kassey,

On 09/11, Kassey wrote:
> hi, Jaegeuk:
> 
>    we got some easy reproduced issue when doing reboot test.
>    that the android init is sleep on a page to be writebback.
>    from the ftrace we can confirm wb_workfn got running each 5s.
>    but init has stuck for quite long time because below page can not
> be writeback done.

Could you please share kernel version and how many patches you have done
cherry-pick from f2fs-stable or android-common?

> 
>    by check the page's member(a_ops) , we can see that is  f2fs page
> and private is NULL.
> 
>    can you help to give some suggest  ?
> 
> 
>    crash> kmem 0xFFFFFFBF01B6C840
> 
>       PAGE               PHYSICAL      MAPPING       INDEX CNT FLAGS
> 
> ffffffbf01b6c840         6db21000 ffffffc0964d1bc0       5c  2 1292c
> referenced,uptodate,lru,owner_priv_1,private,writeback,mappedtodisk
> 
> crash> struct page  0xFFFFFFBF01B6C840
> struct page {
>   flags = 76076,
>   {
>     mapping = 0xffffffc0964d1bc0,

Can you check mapping->host->i_ino?

>     s_mem = 0xffffffc0964d1bc0,
>     compound_mapcount = {
>       counter = -1773331520
>     }
>   },
>   {
>     index = 92,
>     freelist = 0x5c
>   },
>   {
>     counters = 12884901887,
>     {
>       {
>         _mapcount = {
>           counter = -1
>         },
>         active = 4294967295,
>         {
>           inuse = 65535,
>           objects = 32767,
>           frozen = 1
>         },
>         units = -1
>       },
>       _refcount = {
>         counter = 2
>       }
>     }
>   },
>   {
>     lru = {
>       next = 0xffffffbf02bd9520,
>       prev = 0xffffffbf01b66960
>     },
>     pgmap = 0xffffffbf02bd9520,
>     {
>       next = 0xffffffbf02bd9520,
>       pages = 28731744,
>       pobjects = -65
>     },
>     callback_head = {
>       next = 0xffffffbf02bd9520,
>       func = 0xffffffbf01b66960
>     },
>     {
>       compound_head = 18446743794582656288,
>       compound_dtor = 28731744,
>       compound_order = 4294967231
>     }
>   },
>   {
>     private = 0,

F2FS sets Private for most of pages, which is quite weird to me.

>     ptl = 0x0,
>     slab_cache = 0x0
>   }
> }
> 
> 
> init:
> 
> 
> -000|__switch_to()
> 
> -001|__schedule()
> 
> -002|need_resched(inline)
> 
> -002|schedule()
> 
> -003|schedule_timeout()
> 
> -004|get_current(inline)
> 
> -004|io_schedule_timeout()
> 
> -005|bit_wait_io()
> 
> -006|__wait_on_bit()
> 
> -007|wait_on_page_bit()

Could you check whether any IOs were pending in block layer or UFS/eMMC layer?

Thasnk,

> 
> -008|PageWriteback(inline)
> 
> -008|wait_on_page_writeback(inline)
> 
> -008|__filemap_fdatawait_range()
> 
> -009|filemap_fdatawait_keep_errors()
> 
> -010|sync_inodes_sb()
> 
> -011|__sync_filesystem(inline)
> 
> -011|sync_filesystem()
> 
> -012|generic_shutdown_super()
> 
> -013|kill_block_super()
> 
> -014|kill_f2fs_super()
> 
> -015|deactivate_locked_super()
> 
> -016|deactivate_super()
> 
> -017|mnt_free_id(inline)
> 
> -017|cleanup_mnt()
> 
> -018|__cleanup_mnt()
> 
> -019|task_work_run()
> 
> -020|do_notify_resume()
> 
> -021|work_pending(asm)
> 
> -->|exception
> 
> -022|NUX:0x539E58(asm)
> 
> ---|end of frame
> 
> 
> -- 
> Best regards
> Kassey