Summary: | net-libs/webkit-gtk-2.42.4-r600: crash in late build phase due to SIGILL in libatomic.so | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | satmd <satmd> |
Component: | Current packages | Assignee: | Gentoo Toolchain Maintainers <toolchain> |
Status: | RESOLVED INVALID | ||
Severity: | normal | CC: | gnome, satmd |
Priority: | Normal | ||
Version: | unspecified | ||
Hardware: | AMD64 | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: |
build.log for -r410 (file is broken)
build.log (xz) of net-libs/webkit-gtk-2.42.4-r600 failed command |
Description
satmd
2024-01-20 11:41:46 UTC
Created attachment 882681 [details]
build.log for -r410 (file is broken)
dmesgs from attempts with -r410 (WebKit-2.4.1) and -r600 (WebKit-6.0): [ 9275.775625] traps: WebKit-6.0[14581] trap invalid opcode ip:7f484fc853c4 sp:7fffdf6c3ee8 error:0 in libatomic.so.1.2.0[7f484fc83000+3000] [138650.963824] traps: WebKit2-4.1[22115] trap invalid opcode ip:7f7dfa8ad3c4 sp:7fff7e477eb8 error:0 in libatomic.so.1.2.0[7f7dfa8ab000+3000] [221134.185616] traps: WebKit-6.0[2303] trap invalid opcode ip:7fca35c5d3c4 sp:7fff4009a658 error:0 in libatomic.so.1.2.0[7fca35c5b000+3000] [729778.098678] traps: WebKit2-4.1[10057] trap invalid opcode ip:7fcc35eaa3c4 sp:7ffdec038a98 error:0 in libatomic.so.1.2.0[7fcc35ea8000+3000] can you run the failing command manually in the build dir under gdb and get a bt and disassembly? Thanks As I said in the chat today, a cronjob started building webkit-gtk again which causes build.log to be truncated. I will provide new log and try gdb when the build fails again. Sorry, building webkit-gtk on this machine takes a while. Created attachment 882806 [details]
build.log (xz) of net-libs/webkit-gtk-2.42.4-r600
Created attachment 882807 [details]
failed command
I'm unable to provide gdb with an exact stable command, because something is being dynamically created during the command leading up to the SIGILL situation.
I have been able to pinpoint the command during the build that will have the SIGILL in one of its child processes.
I'm able to run that command through strace and through "gdb sh", but they don't provide useful information.
There's a hint though at the children switching to 32bit mode. I noticed having "Disable 32 bit vdso" in my kernel and try toggling this.
Also using CFLAGS+=" -g " next time.
Also updated the bug description to -r600, because I'm now debugging using the latest version again. I have been able to solve this - kinda. The solution was actually found when another package failed: gimp. gimp build crashes in gegl-0.4 and since I had a persistent binary on my file system, I could run debugging on it. This turned up some results in the web search related to a kernel option that seems highly unstable with certain software. Unsetting CONFIG_GDS_FORCE_MITIGATION in my kernel allows me to continue further in the build of gimp. I will give feedback after I have attempted another build of webkit-gtk. I expect that to succeed now. Question: Should we warn about that kernel option when building webkit and/or software which need gegl binaries to build? You probably built your system with -march=native, which includes AVX, then you built a kernel using forced mitigations, and then suddenly binaries needing AVX stop working. |