957487 - |mach valgrind-test|: Use --vex-iropt-register-updates=allregs-at-mem-access

Status: RESOLVED FIXED

(Testing :: General, defect)

Description

[Nicholas Nethercote [inactive]]

|mach valgrind-test| currently uses Valgrind's --vex-iropt-register-updates=allregs-at-each-insn flag (due to bug 947671). Julian suggested that --vex-iropt-register-update=allregs-at-mem-access would suffice.

So I did some try pushes. 

- First, I removed the --vex-iropt-register-updates flag altogether, which causes the default value of unwindregs-at-mem-access to be used. Of the 9 runs that didn't suffer infrastructure problems, three of them crashed:

  https://tbpl.mozilla.org/?tree=Try&rev=ea2ea9606fc2

- Then, I tried allregs-at-mem-access. 23 of the runs didn't suffer infrastructure problems, and none of those crashed:

  https://tbpl.mozilla.org/?tree=Try&rev=f3468a7aaf86

So it looks like allregs-at-mem-access is safe to use.

Comment 1

[Nicholas Nethercote [inactive]]

Attachment: |mach valgrind-test|: Use --vex-iropt-register-updates=allregs-at-mem-access.

Locally, this reduced the time for |mach valgrind-test| from 4m9s to 4m0s,
which is less than I hoped for. But having got this far, I might as well finish
it.

Comment 2

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

I'd like to first run this by :bc, who has run Valgrind on the Firefox browser as well.

Bob, did you happen to try both the --vex-iropt-register-update=allregs-at-mem-access and --vex-iropt-register-updates=allregs-at-each-insn flags in the past, and if so, notice any difference?

Comment 3

[Bob Clary [:bc] (inactive)]

I believe I only ran with allregs-at-mem-access for a short time before I switched to allregs-at-each-insn for everything, so no I can't really say I had enough data to compare.

Comment 4

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

Comment on attachment 8356981 [details] [diff] [review]
|mach valgrind-test|: Use --vex-iropt-register-updates=allregs-at-mem-access.

Okay, in that case let's do the switch and see if we hit anything moving forward.

Comment 5

[Nicholas Nethercote [inactive]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/550e27c525de

Comment 6

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/550e27c525de