On Fri Jul 26, 2024 at 7:11 PM AEST, Thomas Huth wrote: > On 25/07/2024 17.40, Nicholas Piggin wrote: > > In Gitlab CI, some ppc64 multi-threaded tcg tests crash when run in the > > clang-user job with an assertion failure in glibc that seems to > > indicate corruption: > > > > signals: allocatestack.c:223: allocate_stack: > > Assertion `powerof2 (pagesize_m1 + 1)' failed. > > > > Disable these tests for now. > > > > Signed-off-by: Nicholas Piggin <npig...@gmail.com> > > --- > > tests/tcg/ppc64/Makefile.target | 12 ++++++++++++ > > 1 file changed, 12 insertions(+) > > > > diff --git a/tests/tcg/ppc64/Makefile.target > > b/tests/tcg/ppc64/Makefile.target > > index 8c3e4e4038..509a20be2b 100644 > > --- a/tests/tcg/ppc64/Makefile.target > > +++ b/tests/tcg/ppc64/Makefile.target > > @@ -11,6 +11,18 @@ config-cc.mak: Makefile > > > > -include config-cc.mak > > > > +# multi-threaded tests are known to fail (e.g., clang-user CI job) > > +# See: https://gitlab.com/qemu-project/qemu/-/issues/2456 > > +run-signals: signals > > + $(call skip-test, $<, "BROKEN (flaky with clang) ") > > +run-plugin-signals-with-%: > > + $(call skip-test, $<, "BROKEN (flaky with clang) ") > > + > > +run-threadcount: threadcount > > + $(call skip-test, $<, "BROKEN (flaky with clang) ") > > +run-plugin-threadcount-with-%: > > + $(call skip-test, $<, "BROKEN (flaky with clang) ") > > + > > ifneq ($(CROSS_CC_HAS_POWER8_VECTOR),) > > PPC64_TESTS=bcdsub non_signalling_xscv > > endif > > Could you please check whether this is already fixed by Richard's patch: > > https://gitlab.com/qemu-project/qemu/-/commit/8e466dd092469e5ab0f355775c57 > > ?
No, doesn't seem to unfortunately. Here is the same fail - https://gitlab.com/npiggin/qemu/-/jobs/7436325582 I did build it with clang and using the same sanitize flags on my local system, and could not reproduce. So not sure. I might try run the clang-user without any sanitize flags. I sent this patch with ppc pull request already, I think we just do that for now to get clang-user passing. Simple enough to revert once fixed. Thanks, Nick