[compiler-rt] r350139 - Add support for background thread on NetBSD in ASan

Diana Picus via llvm-commits llvm-commits at lists.llvm.org
Wed Jan 9 00:02:41 PST 2019


Hi,

Sorry about the delay, this slipped past my inbox somehow. I'll try to
have a look. Could you please add the fallback or mark as XFAIL until
we figure it out?

Thanks,
Diana

On Mon, 7 Jan 2019 at 19:59, Yvan Roux <yvan.roux at linaro.org> wrote:
>
> On Mon, 7 Jan 2019 at 18:32, Kamil Rytarowski <n54 at gmx.com> wrote:
> >
> > On 07.01.2019 17:13, Kamil Rytarowski wrote:
> > > On 07.01.2019 16:38, Yvan Roux wrote:
> > >> Kamil, Diana,
> > >>
> > >> On Mon, 7 Jan 2019 at 15:43, Kamil Rytarowski via llvm-commits
> > >> <llvm-commits at lists.llvm.org> wrote:
> > >>>
> > >>> I've already replied to Yvan Roux 4 days ago:
> > >>
> > >> Sorry, I was preempted I didn't had time to investigate further.
> > >>
> > >>> It's not possible to guess what's wrong.
> > >>>  - Is the background thread started at all?
> > >>>  - Is __local_asan_dyninit optimized away or not executed?
> > >>>  - Is there a timing issue that test needs more time?
> > >>
> > >> Given that these test cases were OK in Thumbv8 build #1576 and the
> > >> commits involved into build #1577, my guess is that the issue is
> > >> related to __local_asan_dyninit.
> > >>
> > >
> > > Can you debug it or give me remote access to a machine with this setup?
> > >
> >
> > If the answer is "no", then I will add a fallback for your setup to use
> > older branch.
>
> Diana do you have some time to give a look ? I'll not have time to do
> it before end of the week and granting access to these machines will
> not be done quickly either.
>
> > >> build #1576 ninja check logs:
> > >> http://lab.llvm.org:8011/builders/clang-cmake-thumbv8-full-sh/builds/1576/steps/ninja%20check%202/logs/stdio
> > >>
> > >>> I don't have access to setup to reproduce it myself.
> > >>>
> > >>> The patch fixed Android setup and enabled NetBSD.
> > >>>
> > >>> I think that the problem is not caused by my change, but it's a side
> > >>> effect of other bug/optimization. I can mark the test as XFAIL for your
> > >>> setup or add ifdef for your Linux configuration to start the background
> > >>> thread in the same way as before.
> > >>>
> > >>>
> >


More information about the llvm-commits mailing list