Firefox crashes at startup because of invalid instruction
Categories
(Core :: Machine Learning: General, defect)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox-esr115 | --- | unaffected |
| firefox-esr128 | --- | unaffected |
| firefox-esr140 | --- | unaffected |
| firefox142 | --- | unaffected |
| firefox143 | --- | unaffected |
| firefox144 | blocking | fixed |
People
(Reporter: viktor_jaegerskuepper, Assigned: padenot)
References
(Regression)
Details
(Keywords: perf-alert, regression)
Crash Data
Attachments
(3 files)
The latest Firefox Nightly crashes at startup on my system (x86-64, Intel Core 2 Duo) because of an "invalid instruction" (translated from German).
I ran mozregression and got:
Last good revision: 8e649e2c1476f102a2bbe699c1cf4949e557640b
First bad revision: 4c630ff1b3255fb7e6cb13eb8018652aa6703bef
Pushlog:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=8e649e2c1476f102a2bbe699c1cf4949e557640b&tochange=4c630ff1b3255fb7e6cb13eb8018652aa6703bef
Bug 1971158 seems relevant because it has a patch with the description "Compile llama.cpp w/ AVX2 on x86_64 ..." and my CPU doesn't support AVX2.
Sandor, please check if this is a regression of Bug 1971158 too and/or move it to the appropriate component.
|
||
Comment 2•10 months ago
|
||
(In reply to 08xjcec48 from comment #1)
Sandor, please check if this is a regression of Bug 1971158 too and/or move it to the appropriate component.
talked with @padenot, will fix it
|
Assignee | |
Comment 3•10 months ago
|
||
Viktor, sorry about this, I have put in some code specifically for this. Do you mind getting us a link to a crash report? You can likely open another firefox version with the same profile, and find it in about:crashes.
|
|
Assignee | |
Comment 4•10 months ago
|
||
|
||
Updated•10 months ago
|
|
Reporter | |
Comment 5•10 months ago
|
||
Unfortunately there aren't any reports for this particular crash in about:crashes because the crash occurs too early (?). I will find out how to get a backtrace or if I can get this information from a coredump which was saved by systemd.
|
|
Assignee | |
Comment 6•10 months ago
|
||
Viktor, what OS are you running? I've added a patch that could fix it, but I'm quite surprised the code we've added runs at startup.
|
|
Reporter | |
Comment 7•10 months ago
|
||
I'm running Arch Linux.
|
|
Assignee | |
Comment 8•10 months ago
•
|
||
Superb, then you can probably just do gdb /path/to/firefox-nightly and it will crash, then bt can print a stack trace properly, potentially unsymbolicated (but I can symbolicate afterward). Are you running our builds, or something arch is building maybe?
|
|
Reporter | |
Comment 9•10 months ago
|
||
I am running the Mozilla builds. I ran gdb with debuginfod enabled (I guess to use debug info for Arch Linux packages) and then entered bt, I'm attaching the output.
|
|
Assignee | |
Comment 10•10 months ago
|
||
Thanks, it's very clear, I'll get a patch in asap.
|
||
Comment 11•10 months ago
|
||
08xjcec48: It's better to set a needinfo to the assignee of the bug which one thinks might be the regressor. The landing (merge) of the change to mozilla-central / firefox-main is done by people not involved with the code which landed in the bug.
|
||
Updated•10 months ago
|
|
|
||
Updated•10 months ago
|
|
|
||
Updated•10 months ago
|
|
||
Updated•10 months ago
|
|
||
Comment 12•10 months ago
|
||
|
|
Assignee | |
Comment 13•10 months ago
|
||
|
|
||
Comment 14•10 months ago
|
||
|
|
||
Comment 15•10 months ago
|
||
|
|
||
Comment 16•10 months ago
|
||
| bugherder | ||
https://hg.mozilla.org/mozilla-central/rev/ba5ad04d9df5
https://hg.mozilla.org/mozilla-central/rev/2065c7dd1223
|
|
Assignee | |
Comment 17•10 months ago
|
||
Viktor, thanks a lot for taking the time to open a bug. I think you'll have to download a new nightly (when it is published), but it should be fixed. Please let us know here, or directly to me at padenot@mozilla.com if there's any problem, and sorry for the inconvenience.
|
|
||
Comment 18•10 months ago
|
||
A Nightly update with this fix is now available.
|
|
Reporter | |
Comment 19•10 months ago
|
||
The issue was fixed by the second patch (Remove more static initializers from llama.cpp) and the new Nightly build is running fine. Thanks!
|
||
Comment 20•10 months ago
|
||
| bugherder | ||
|
||
Comment 21•10 months ago
|
||
(In reply to Pulsebot from comment #14)
Pushed by padenot@mozilla.com:
https://github.com/mozilla-firefox/firefox/commit/ad918290746f
https://hg.mozilla.org/integration/autoland/rev/09242258e16e
Remove more static initializers from llama.cpp. r=atossou
Perfherder has detected a build_metrics performance change from push 09242258e16ef379e696a42afecf69f23ea84fac.
If you have any questions, please reach out to a performance sheriff. Alternatively, you can find help on Slack by joining #perf-help, and on Matrix you can find help by joining #perftest.
Improvements:
| Ratio | Test | Platform | Options | Absolute values (old vs new) |
|---|---|---|---|---|
| 4% | compiler_metrics num_static_constructors | osx-cross | 96.00 -> 92.00 | |
| 4% | compiler_metrics num_static_constructors | linux64 | 106.00 -> 102.00 | |
| 4% | compiler_metrics num_static_constructors | linux64-shippable | 106.00 -> 102.00 | |
| 4% | compiler_metrics num_static_constructors | linux64 | 110.00 -> 106.00 | |
| 4% | compiler_metrics num_static_constructors | linux64-shippable | instrumented | 112.00 -> 108.00 |
Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests.
If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a performance sheriff to do that for you.
You can run all of these tests on try with ./mach try perf --alert 46614
The following documentation link provides more information about this command.
|
||
Updated•10 months ago
|
Description
•