On an older AMD Phenom II box, firefox-114 is getting built with AVX instructions. Checking with gdb, the offending instructions start with a "vshufps" located in /usr/lib64/firefox/libxul.so Built using gcc-13.1.1_p20230527 lto, pgo, hardened USE flags are set The compiler flags are somewhat agressive, so I'll be rebuilding with something more tame to see if that changes anything. In the meantime though, something is broke with AVX instructions being generated on this platform with march=native Reproducible: Always
I'm guessing this *isn't* the nss-3.90 thing (please double check that's not installed) if it's in libxul.so. Could you give us a backtrace in full when it gets SIGILL'd please, as well as emerge --info? Thanks.
Created attachment 863739 [details] emerge --info
Created attachment 863740 [details] equery uses firefox
(In reply to Sam James from comment #1) > I'm guessing this *isn't* the nss-3.90 thing (please double check that's not > installed) if it's in libxul.so. No, that had already been downgraded to dev-libs/nss-3.89.1 > > Could you give us a backtrace in full when it gets SIGILL'd please, as well > as emerge --info? Thanks. I'm attaching backtrace
Created attachment 863742 [details] backtrace
Could you build firefox+glibc with debugging symbols (-ggdb3 + installsources ideally, see https://wiki.gentoo.org/wiki/Debugging#Per-package) please?
(In reply to Sam James from comment #6) > Could you build firefox+glibc with debugging symbols (-ggdb3 + > installsources ideally, see > https://wiki.gentoo.org/wiki/Debugging#Per-package) please? Yes, will take a little while
Created attachment 863754 [details] Backtrace with debug symbols Hope 13 layers into the stack is sufficient. This machine does not have the memory needed to do a proper backtrace on a codebase this large, so I killed it after churning through swap for 20min.
Hmm wonder if this is skia related, saw they had made tons of updates there (e.g. hopefully fixing bug 838373) https://bugs.gentoo.org/838373#c57
Yeah, I think it is. Can someone hitting this give me the full build.log, per chance?
I had a crashing firefox 114.0.1 on opensuse tumbleweed today after update. https://bugzilla.opensuse.org/show_bug.cgi?id=1212101 https://bugzilla.mozilla.org/show_bug.cgi?id=1838323
(In reply to jospezial from comment #11) > I had a crashing firefox 114.0.1 on opensuse tumbleweed today after update. > https://bugzilla.opensuse.org/show_bug.cgi?id=1212101 > https://bugzilla.mozilla.org/show_bug.cgi?id=1838323 Thank you for sharing those!
firefox works now on opensuse tumbleweed on old CPUs that have no AVX. Changelog: 2023-06-20 - Andreas Stieger <Andreas.Stieger@gmx.de> - Mozilla Firefox 114.0.2: * Several crash fixes * Web Extensions: Fixes for 114 regressions in Native Messaging support 2023-06-20 - Wolfgang Rosenauer <wr@rosenauer.org> - do not enable LTO as it caused crashes now (boo#1212101) The GCC bug develops on.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=03ee26bcb2e7fece7f04f18f39c3cd153f06ce5d commit 03ee26bcb2e7fece7f04f18f39c3cd153f06ce5d Author: Joonas Niilola <juippis@gentoo.org> AuthorDate: 2023-07-04 16:30:32 +0000 Commit: Joonas Niilola <juippis@gentoo.org> CommitDate: 2023-07-04 16:32:02 +0000 www-client/firefox: add 115.0 - drop ~ppc64 keyword due to patches not applying before a rebase. Bug: https://bugs.gentoo.org/908297 Bug: https://bugs.gentoo.org/908979 Bug: https://bugs.gentoo.org/908412 Bug: https://bugs.gentoo.org/908992 Signed-off-by: Joonas Niilola <juippis@gentoo.org> www-client/firefox/Manifest | 101 +++ www-client/firefox/firefox-115.0.ebuild | 1458 +++++++++++++++++++++++++++++++ 2 files changed, 1559 insertions(+)