[Qemu-devel] Bug in mips user-linux

[Qemu-devel] Bug in mips user-linux

[Torbjörn Granlund]

[-- Attachment #1: Type: text/plain, Size: 889 bytes --]

This is a long-standing issue; I've reproduced it in 2.5.0, 2.6.2, and
2.7.0.

Command used:  qemu-mipsn32 ./reuse

The binary file 'reuse' is attached.

Funnily enough, it does not fail every time.  (I would expect user-level
emulation to be pretty deterministic, but alas, this is not.)

When it fails, it either says,

  qemu: uncaught target signal 11 (Segmentation fault) - core dumped
  Segmentation fault

or unfrequently,

  mipsn: /var/tmp/build.2567/qemu-2.6.2/translate-all.c:1884: page_set_flags: Assertion `start < end' failed.
  qemu: uncaught target signal 6 (Aborted) - core dumped

while the same binary runs fine on hardware as well as
emulated systems running under qemu-system-mips64.

The failures only happen for the n32 ABI (not o32 or 64).

This test case comes from GMP (default gcc compile of
gmp/tests/mpz/reuse.c).

(Host and guest systems all run Debian 8.)



[-- Attachment #2: reuse.xz --]
[-- Type: application/octet-stream, Size: 357368 bytes --]

[-- Attachment #3: Type: text/plain, Size: 55 bytes --]



-- 
Torbjörn
Please encrypt, key id 0xC8601622

Re: [Qemu-devel] Bug in mips user-linux

[Thomas Huth]

On 10.10.2016 03:01, Torbjörn Granlund wrote:
> This is a long-standing issue; I've reproduced it in 2.5.0, 2.6.2, and
> 2.7.0.
> 
> Command used:  qemu-mipsn32 ./reuse
> 
> The binary file 'reuse' is attached.
> 
> Funnily enough, it does not fail every time.  (I would expect user-level
> emulation to be pretty deterministic, but alas, this is not.)
> 
> When it fails, it either says,
> 
>   qemu: uncaught target signal 11 (Segmentation fault) - core dumped
>   Segmentation fault
> 
> or unfrequently,
> 
>   mipsn: /var/tmp/build.2567/qemu-2.6.2/translate-all.c:1884: page_set_flags: Assertion `start < end' failed.
>   qemu: uncaught target signal 6 (Aborted) - core dumped
> 
> while the same binary runs fine on hardware as well as
> emulated systems running under qemu-system-mips64.
> 
> The failures only happen for the n32 ABI (not o32 or 64).
> 
> This test case comes from GMP (default gcc compile of
> gmp/tests/mpz/reuse.c).
> 
> (Host and guest systems all run Debian 8.)

Could you maybe get a proper backtrace of the crash with gdb? I guess
that would help to track down the problem.
And please make sure to copy the MIPS maintainers on CC: when reporting
such issues (see MAINTAINERS file), otherwise your mail might get lost
in the high traffic of the qemu-devel mailing list.

 Thanks,
  Thomas

Re: [Qemu-devel] Bug in mips user-linux

[Yongbok Kim]

On 19/10/2016 09:26, Thomas Huth wrote:
> On 10.10.2016 03:01, Torbjörn Granlund wrote:
>> This is a long-standing issue; I've reproduced it in 2.5.0, 2.6.2, and
>> 2.7.0.
>>
>> Command used:  qemu-mipsn32 ./reuse
>>
>> The binary file 'reuse' is attached.
>>
>> Funnily enough, it does not fail every time.  (I would expect user-level
>> emulation to be pretty deterministic, but alas, this is not.)
>>
>> When it fails, it either says,
>>
>>   qemu: uncaught target signal 11 (Segmentation fault) - core dumped
>>   Segmentation fault
>>
>> or unfrequently,
>>
>>   mipsn: /var/tmp/build.2567/qemu-2.6.2/translate-all.c:1884: page_set_flags: Assertion `start < end' failed.
>>   qemu: uncaught target signal 6 (Aborted) - core dumped
>>
>> while the same binary runs fine on hardware as well as
>> emulated systems running under qemu-system-mips64.
>>
>> The failures only happen for the n32 ABI (not o32 or 64).
>>
>> This test case comes from GMP (default gcc compile of
>> gmp/tests/mpz/reuse.c).
>>
>> (Host and guest systems all run Debian 8.)
> 
> Could you maybe get a proper backtrace of the crash with gdb? I guess
> that would help to track down the problem.
> And please make sure to copy the MIPS maintainers on CC: when reporting
> such issues (see MAINTAINERS file), otherwise your mail might get lost
> in the high traffic of the qemu-devel mailing list.
> 
>  Thanks,
>   Thomas
> 

Hi I've tested the binary with the upstream but couldn't reproduce the
problem. Would you please provide more details?

Regards,
Yongbok

Re: [Qemu-devel] Bug in mips user-linux

[Torbjörn Granlund]

Yongbok Kim <yongbok.kim@imgtec.com> writes:

  Hi I've tested the binary with the upstream but couldn't reproduce the
  problem. Would you please provide more details?
  
What qemu version did you use?

Specifically, did you use any of the versions which I listed as failing?

If you used something newer than 2.7.0, then perhaps the bug is fixed.

I trust that you noticed from my report that this bug does not happen
each run.  The "rate of intermittence" might vary, I suppose, so please
rerun my binary a fair number of times before you conclude that it is
not easy to reproduce.

-- 
Torbjörn
Please encrypt, key id 0xC8601622