Re: [BUG] Bad page map in process after migration (Debian #711249)

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 06/11/2013 01:01 PM, konrad wilk wrote:
>
> On 6/11/2013 12:44 PM, Ian Campbell wrote:
>> A Debian user has reported this with the 3.2 kernel, it sounds vaguely
>> familiar, but neither I nor google can quite put my finger on it. Does
>> it ring any bells?
>>
>>          [1295549.783990] BUG: Bad page map in process java 
>> pte:00000166 pmd:3bacb067
>>          [1295549.784005] page:ffffea0000000000 count:-1 
>> mapcount:-128 mapping:          (null) index:0x0
>>          [1295549.784036] page flags: 0x14(referenced|dirty)
>>          [1295549.784044] addr:00007f879a6a2000 vm_flags:00100077 
>> anon_vma:ffff88003cb472a0 mapping: (null) index:7f879a6a2
>>          [1295549.784052] Pid: 13198, comm: java Not tainted 
>> 3.2.0-4-amd64 #1 Debian 3.2.41-2+deb7u2
>>          [1295549.784058] Call Trace:
>>          [1295549.784070]  [<ffffffff810ce1f5>] ? 
>> print_bad_pte+0x1a5/0x1bd
>>          [1295549.784075]  [<ffffffff810cde55>] ? pte_pfn+0x5/0xe
>>          [1295549.784083]  [<ffffffff810cfe5c>] ? unmap_vmas+0x4d5/0x6ca
>>          [1295549.784095]  [<ffffffff810d00e1>] ? 
>> zap_page_range+0x90/0xb8
>>          [1295549.784108]  [<ffffffff810d5ce0>] ? 
>> mmap_region+0x353/0x44a
>>          [1295549.784122]  [<ffffffff81036428>] ? 
>> should_resched+0x5/0x23
>>          [1295549.784133]  [<ffffffff810cdba2>] ? 
>> sys_madvise+0x3d2/0x5fc
>>          [1295549.784147]  [<ffffffff81053e6c>] ? 
>> recalc_sigpending+0x23/0x3c
>>          [1295549.784156]  [<ffffffff81054518>] ? 
>> __set_task_blocked+0x5e/0x65
>>          [1295549.784168]  [<ffffffff81052e40>] ? 
>> spin_unlock_irq+0xa/0xb
>>          [1295549.784178]  [<ffffffff8105601c>] ? sigprocmask+0x5c/0x63
>>          [1295549.784191]  [<ffffffff813529d2>] ? 
>> system_call_fastpath+0x16/0x1b
>>          [1295549.784201] Disabling lock debugging due to kernel taint
>>
>> There's a bit more info and a full dmesg in
>> http://bugs.debian.org/711249
>
> Thank you.
> Boris,
> Did you fix a similar issue in v3.9? There was the cgroup one which I 
> thought had a similar back-track?


The only thing I can think of is the issue related to lazy MMU flush 
that involved zap_page_range(). The signature was slightly different 
though. It's commit 1160c2779b826c6f5c08e5cc542de58fd1f667d5.

-boris