1529355 - emit unwinding directives for aarch64 in dump_syms

Status: RESOLVED MOVED

(Toolkit :: Crash Reporting, enhancement)

Description

[Nathan Froyd [:froydnj]]

+++ This bug was initially created as a clone of Bug #1517739 +++

+++ This bug was initially created as a clone of Bug #1513284 +++

See discussion in bug 1517739, which is almost, but not quite, the same thing.

Comment 1

[Nathan Froyd [:froydnj]]

Attachment: Bug 1529355 - emit STACK CFI lines for aarch64 windows; r=gsvelto

aarch64 windows uses a custom unwind information format to describe
how to unwind from within functions. We need to parse that
information to describe to Breakpad how to unwind stacks for minidump
analysis.

Comment 2

[Gabriele Svelto [:gsvelto]]

Thanks for this, I'll review it tomorrow first thing. Since these changes happen into the part of Breakpad we haven't forked of it would be nice to push them upstream but I'm OK with landing them first directly in mozilla-central because we need them ASAP.

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/b171bfdb308f
emit STACK CFI lines for aarch64 windows; r=gsvelto

Comment 4

[Dorel Luca [:dluca]]

Backed out changeset b171bfdb308f (bug 1529355) for build bustage

Log:
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=229905958&repo=autoland&lineNumber=27553

Push with failures:
https://treeherder.mozilla.org/#/jobs?repo=autoland&selectedJob=229905958&revision=b171bfdb308f94aea2193530ef670f2c5a350a84

Backout:
https://hg.mozilla.org/integration/autoland/rev/ef2adef9bbccc3c90945ff3cc1648488bcf58b7e

Comment 5

[Nathan Froyd [:froydnj]]

Works on my machine, but I guess that's not sufficient proof. Will also fix up the assertion errors...

Comment 6

[Gabriele Svelto [:gsvelto]]

Nathan, do you have time to finish this? I can take over it (and push it upstream) if you're too busy.

Comment 7

[Nathan Froyd [:froydnj]]

(In reply to Gabriele Svelto [:gsvelto] from comment #6)

Nathan, do you have time to finish this? I can take over it (and push it upstream) if you're too busy.

I do have time to finish this; the problem I ran into was that clang (and, presumably, Rust) emit bogus unwind information for some functions, which causes dump_syms to wander off into the weeds. I haven't had time to dive into LLVM yet, though the bug I filed (https://bugs.llvm.org/show_bug.cgi?id=40876) sounds like it might not be too hard to fix.

Comment 8

[Gabriele Svelto [:gsvelto]]

I see, we're hitting all sort of corner cases lately with debuginfo :-/

Comment 9

[Gabriele Svelto [:gsvelto]]

We moved this work to: https://github.com/mozilla/dump_syms/issues/5