[llvm-dev] Unstable UBSan tests on AArch64

Adhemerval Zanella via llvm-dev llvm-dev at lists.llvm.org
Mon Nov 2 11:48:55 PST 2015



On 02-11-2015 17:22, Renato Golin wrote:
> On 2 November 2015 at 19:11, Adhemerval Zanella
> <adhemerval.zanella at linaro.org> wrote:
>> Right, it could be a case where the thread stack is being places in below
>> 0x7f00000000. On the TSAN unification patch I also lower the high memory
>> start from 0x7e00000000 to 0x7d00000000, which can give more VMA available.
>> Strange that I have not yet see any similar issue in my environment.
> 
> This was happening once every few builds. It could be that you never got it.
> 
> We may have to wait a few builds after the VMA unification to move the
> bot back to active duty.

It could, although still this timeout in a such simple tests is strange. I
am playing here with multiple thread and 0x7f... seem a high limit that
usually this is not reached by the thread stack (GLIBC usually sets 8MB for
each thread stack plus a guard page).

However I do see that with higher thread usage and also higher memory
utilization some mmap will eventually reach the high memory start limit
(it does not explain the test failure although).

Warn me if you see a failing test with abort (it would be good if you could
generate a core at least).

> 
> --renato
> 


More information about the llvm-dev mailing list