From: Minchan Kim <minchan.kim@gmail.com>
To: Andrea Arcangeli <aarcange@redhat.com>
Cc: linux-mm <linux-mm@kvack.org>
Subject: Re: [BUG]thp: BUG at mm/huge_memory.c:1350
Date: Fri, 21 Jan 2011 01:41:02 +0900 [thread overview]
Message-ID: <AANLkTikHNcD3aOWKJdPtCqdJi9C34iLPxj5-L8=gqBFc@mail.gmail.com> (raw)
In-Reply-To: <20110120161436.GB21494@random.random>
On Fri, Jan 21, 2011 at 1:14 AM, Andrea Arcangeli <aarcange@redhat.com> wrote:
> Hello Minchan,
>
> On Fri, Jan 21, 2011 at 12:49:35AM +0900, Minchan Kim wrote:
>> Hi Andrea,
>>
>> I hit thg BUG 2 time during 5 booting.
>> I applied khugepaged: fix pte_unmap for highpte x86_32 based on 2.6.38-rc1.
>
> This is again 32bit which rings a bell because clearly it didn't have
> a whole lot of testing (not remotely comparable to the amount of
> testing x86-64 had), but it's not necessarily 32bit related. It still
> would be worth to know if it still happens after you disable
> CONFIG_HIGHMEM (to rule out 32bit issues in not well tested kmaps).
>
I wii try it at tomorrow.
> The rmap walk simply didn't find an hugepmd pointing to the page. Or
> the mapcount was left pinned after the page got somehow unmapped.
>
> I wonder why pages are added to swap and the system is so low on
> memory during boot. How much memory do you have in the 32bit system?
That was my curiosity, too.
> Do you boot with something like mem=256m? This is also the Xorg
No. I didn't limit mem size. My system has a 2G memory.
> process, which is a bit more special than most and it may have device
> drivers attached to the pages.
Both bugs are hit by Xorg.
I doubt it.
>
> One critical thing is that split_huge_page must be called when
> splitting vmas, see split_huge_page_address and
> __vma_adjust_trans_huge, right now it's called from vma_adjust
> only. If anything is wrong in that function, or if any place adjusting
> the vma is not covered by that function, it may result in exactly the
> problem you run into. If drivers are mangling over the vmas that would
> also explain it.
>
> If you happen to have crash dump with vmlinux that would be the best
> debug option for this, also if you can reproduce it in a VM that will
> make it easier to reproduce without your hardware. Otherwise we'll
> find another way.
I will investigate it after out of office at tomorrow. Sorry but here
is 1:40 am. :)
If you have a any guess or good method to investigate the bug, please reply
next prev parent reply other threads:[~2011-01-20 16:41 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-01-20 15:49 Minchan Kim
2011-01-20 16:14 ` Andrea Arcangeli
2011-01-20 16:41 ` Minchan Kim [this message]
2011-01-21 17:58 ` Minchan Kim
2011-01-21 18:14 ` Andrea Arcangeli
2011-01-22 0:59 ` Minchan Kim
2011-01-22 1:08 ` Andrea Arcangeli
2011-01-22 2:16 ` Andrea Arcangeli
2011-01-22 2:20 ` Minchan Kim
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='AANLkTikHNcD3aOWKJdPtCqdJi9C34iLPxj5-L8=gqBFc@mail.gmail.com' \
--to=minchan.kim@gmail.com \
--cc=aarcange@redhat.com \
--cc=linux-mm@kvack.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox