meta-browser
meta-browser copied to clipboard
chromium: support for 32 bit
try to build chromium-x11 using lib32
for intel-x86-64
as an image for wrlinux-image-std
There is a warning for relocation and the resulting binary segfaults when run
Any thoughts on supporting 32-bit machines?
root@intel-x86-64:~# chromium --no-sandbox --disable-gpu
[1465:1478:1216/060845.676256:FATAL:gpu_data_manager_impl_private.cc(445)] GPU process isn't usable. Goodbye.
Trace/breakpoint trap
root@intel-x86-64:~# tail -n10 /var/log/syslog
2020-12-16T06:15:50.463854+00:00 intel-x86-64 kernel: VizCompositorTh[2301]: segfault at 0 ip 0000000000000000 sp 00000000f2b5a590 error 14 in chromium-bin[565fe000+104a000]
EDIT: this doesn't happen with a non-root user but i see issues like this at runtime
../../sandbox/linux/seccomp-bpf-helpers/sigsys_handlers.cc:**CRASHING**:seccomp-bpf
failure in syscall 0412
and this at build time
WARNING: lib32-chromium-x11-87.0.4280.141-r0 do_package_qa: QA Issue: lib32-chromium-x11: ELF binary /usr/lib/chromium/chromium-bin has relocations in .text [textrel]
Is there an intent to support 32 bit?
That's strange. I'm building for 32bit armv7 and get no such issues.
Is there an intent to support 32 bit?
Upstream itself stopped providing official 32-bit Linux packages long ago, and I don't think this setup is being actively tested by them. Can you share more information about your setup? Is this a 32-bit Chromium build running in a x86-64 kernel + userland, or a x86-64 kernel with a fully 32-bit userland?
I think Its 32 bit chromium running in a x86-64 kernel with everything else being 32 bit (userspace)
basically i built something similar to a lib32-core-image-sato but using wrlinux.
Windriver has no customer interest in intel86 32 bit machines, and i don't think anyone would want to throw out the 64bit userspace for a 32bit one.
I'm building for 32bit armv7 and get no such issues.
I have heard that 32bit arm has more interest.
If there is no one that cares about intel 32bit, then i won't be digging around to fixing some of these issues the testing team and I found.
I wouldn't go as far as saying "nobody cares", support for it here (and maybe upstream) just doesn't seem as high-priority as x86-64, but we and upstream would be happy to review whatever fixes you may have come up with.
its not my priority either, Windriver testing team went through the common cases and uncovered these kind of issues. given the low priority given to 32 bit intel (both here, upstream and windriver), i will leave it as it is
I'm reopening this one for a while because there's also a discussion about issues with 32-bit Linux in the chromium-packagers mailing list; if some patches end up landing we might be able to revisit this.
qemux86 no longer have "binary relocation errors" with latest builds.