Re: BUG: Bad page state in process (current git)

[Hugh Dickins]

On Mon, Nov 15, 2010 at 11:54 AM, Markus Trippelsdorf
<markus@xxxxxxxxxxxxxxx> wrote:
> On 2010.11.15 at 13:38 +0100, Markus Trippelsdorf wrote:
>> On 2010.11.12 at 13:20 +0100, Markus Trippelsdorf wrote:
>> >
>> > Yes. Fortunately the BUG is gone since I pulled the upcoming drm fixes
>>
>> No. I happend again today (with those fixes already applied):
>>
>> BUG: Bad page state in process knode Âpfn:7f0a8
>> page:ffffea0001bca4c0 count:0 mapcount:0 mapping: Â Â Â Â Â(null) index:0x0
>> page flags: 0x4000000000000008(uptodate)
>> Pid: 18310, comm: knode Not tainted 2.6.37-rc1-00549-gae712bf-dirty #16
>> Call Trace:
>> Â[<ffffffff810a9022>] ? bad_page+0x92/0xe0
>> Â[<ffffffff810aa240>] ? get_page_from_freelist+0x4b0/0x570
>> Â[<ffffffff8102e50e>] ? apic_timer_interrupt+0xe/0x20
>> Â[<ffffffff810aa413>] ? __alloc_pages_nodemask+0x113/0x6b0
>> Â[<ffffffff810a2dd4>] ? file_read_actor+0xc4/0x190
>> Â[<ffffffff810a4a70>] ? generic_file_aio_read+0x560/0x6b0
>> Â[<ffffffff810bdf8d>] ? handle_mm_fault+0x6bd/0x970
>> Â[<ffffffff8104b1d0>] ? do_page_fault+0x120/0x410
>> Â[<ffffffff810c3d85>] ? do_brk+0x275/0x360
>> Â[<ffffffff81452d8f>] ? page_fault+0x1f/0x30
>> Disabling lock debugging due to kernel taint
>
> And another one. But this time it seems to point to ext4:
>
> BUG: Bad page state in process rm Âpfn:52e54
> page:ffffea0001222260 count:0 mapcount:0 mapping: Â Â Â Â Â(null) index:0x0
> page flags: 0x4000000000000008(uptodate)
> Pid: 2084, comm: rm Not tainted 2.6.37-rc1-00549-gae712bf-dirty #23
> Call Trace:
> Â[<ffffffff810a9022>] ? bad_page+0x92/0xe0
> Â[<ffffffff810aa240>] ? get_page_from_freelist+0x4b0/0x570
> Â[<ffffffff81142ae6>] ? ext4_ext_put_in_cache+0x46/0x90
> Â[<ffffffff810aa413>] ? __alloc_pages_nodemask+0x113/0x6b0
> Â[<ffffffff8118f0c7>] ? number.clone.2+0x2b7/0x2f0
> Â[<ffffffff810a38d5>] ? find_get_page+0x75/0xb0
> Â[<ffffffff810a4011>] ? find_or_create_page+0x51/0xb0
> Â[<ffffffff810ff4d7>] ? __getblk+0xd7/0x260
> Â[<ffffffff8113158f>] ? ext4_getblk+0x8f/0x1e0
> Â[<ffffffff811316ed>] ? ext4_bread+0xd/0x70
> Â[<ffffffff811369f4>] ? htree_dirblock_to_tree+0x34/0x190
> Â[<ffffffff8113870f>] ? ext4_htree_fill_tree+0x9f/0x250
> Â[<ffffffff810e109d>] ? do_filp_open+0x12d/0x5e0
> Â[<ffffffff811289ed>] ? ext4_readdir+0x14d/0x5a0
> Â[<ffffffff810e4e80>] ? filldir+0x0/0xd0
> Â[<ffffffff810e50a8>] ? vfs_readdir+0xa8/0xd0
> Â[<ffffffff810e4e80>] ? filldir+0x0/0xd0
> Â[<ffffffff810e51b1>] ? sys_getdents+0x81/0xf0
> Â[<ffffffff8102dc2b>] ? system_call_fastpath+0x16/0x1b
> Disabling lock debugging due to kernel taint
>
> I don't know. Could a possible bug in linux/fs/ext4/page-io.c be
> responsible for something like this?

I do think you're right: every one of your "Bad page state" reports
has been complaining only about the PageUptodate bit being set, and
that SetPageUpdate() in ext4_end_bio() does look suspicious, coming
after the put_page().

The more suspicious given that other races have been noticed in
precisely that area, and fixed with put_io_page() in the current git
tree.

Perhaps that fixes your problem, but my guess would be not: I suspect
the "if (!partial_write) SetPageUpdate(page);" should be done before
the block (or put_io_page) which does the put_page().

Hugh
--
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/