llvm dev - Nov 2015 - Unstable UBSan tests on AArch64

Hi Adhemerval,

Some UBSan tests are timing out randomly.

http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full

ex:
http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/902
http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/894
http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/906
http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/902

The timeout happens consistently here:

PASS: UBSan-TSan-aarch64 :: TestCases/Misc/enum.cpp (25310 of 25316)
PASS: UBSan-TSan-aarch64 :: TestCases/TypeCheck/null.cpp (25311 of 25316)
PASS: UBSan-TSan-aarch64 ::
TestCases/TypeCheck/vptr-virtual-base-construction.cpp (25312 of
25316)
PASS: UBSan-TSan-aarch64 :: TestCases/TypeCheck/vptr-virtual-base.cpp
(25313 of 25316)
PASS: UBSan-TSan-aarch64 :: TestCases/TypeCheck/misaligned.cpp (25314 of 25316)
PASS: UBSan-TSan-aarch64 :: TestCases/Integer/shift.cpp (25315 of 25316)

command timed out: 1200 seconds without output running ['ninja',
'check-all', '-j8'], attempting to kill

I can imagine two scenarios:

 1. The test takes a long time to run, close to the timeout, so that
it's too long some times, and others not.
 2. There is a random codegen/instrumentation problem that only shows
some times. Address randomization could be an issue, here.

I've moved it to the silent build master for now:

http://lab.llvm.org:8014/builders/clang-cmake-aarch64-full

cheers,
--renato

Is it 39 or 42-bit VMA? I noted a 42-bit issue in segment definition
that I have fixed on my TSAN unification mapping patch [1]:

lib/tsan/rtl/tsan_platform.h:
142 const uptr kHiAppMemBeg   = 0x3ff00000000ull;
143 const uptr kHiAppMemEnd   = 0x3fffff00000ull;

The 'kHiAppMemEnd' should be 0x3ffffffffff since I noted that with ASLR
off kernel usually maps some segments up to 0x3fffff00000 (I have not
see with ASLR on). I decided to not send an patch and incorporate it on
the TSAN reword for 39 and 42 bit.

[1] http://reviews.llvm.org/D14199

On 02-11-2015 10:20, Renato Golin wrote:
> Hi Adhemerval,
> 
> Some UBSan tests are timing out randomly.
> 
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full
> 
> ex:
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/902
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/894
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/906
> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/902
> 
> The timeout happens consistently here:
> 
> PASS: UBSan-TSan-aarch64 :: TestCases/Misc/enum.cpp (25310 of 25316)
> PASS: UBSan-TSan-aarch64 :: TestCases/TypeCheck/null.cpp (25311 of 25316)
> PASS: UBSan-TSan-aarch64 ::
> TestCases/TypeCheck/vptr-virtual-base-construction.cpp (25312 of
> 25316)
> PASS: UBSan-TSan-aarch64 :: TestCases/TypeCheck/vptr-virtual-base.cpp
> (25313 of 25316)
> PASS: UBSan-TSan-aarch64 :: TestCases/TypeCheck/misaligned.cpp (25314 of
25316)
> PASS: UBSan-TSan-aarch64 :: TestCases/Integer/shift.cpp (25315 of 25316)
> 
> command timed out: 1200 seconds without output running ['ninja',
> 'check-all', '-j8'], attempting to kill
> 
> I can imagine two scenarios:
> 
>  1. The test takes a long time to run, close to the timeout, so that
> it's too long some times, and others not.
>  2. There is a random codegen/instrumentation problem that only shows
> some times. Address randomization could be an issue, here.
> 
> I've moved it to the silent build master for now:
> 
> http://lab.llvm.org:8014/builders/clang-cmake-aarch64-full
> 
> cheers,
> --renato
>

On 2 November 2015 at 18:40, Adhemerval Zanella
<adhemerval.zanella at linaro.org> wrote:
> Is it 39 or 42-bit VMA? I noted a 42-bit issue in segment definition
> that I have fixed on my TSAN unification mapping patch [1]:
No, that's 39-bit.

cheers,
--renato