Ticket #3497 (closed defect: notsugar)

Opened 13 months ago

Last modified 12 months ago

Browse crashes a lot on 1.75

Reported by: erikos Owned by: erikos
Priority: Immediate Milestone: 0.96
Component: Browse Version: Git as of bugdate
Severity: Blocker Keywords: 12.1.0, upstream
Cc: tonyforster, dsd Distribution/OS: OLPC
Bug Status: New

Description (last modified by manuq) (diff)

I have seen Browse crashing a lot of times on the XO 1.75 with os8. No concrete steps yet to reproduce. I will close and then we can test in build.

Attachments

browse-trace5.txt.asc Download (30.5 KB) - added by manuq 13 months ago.
browse-trace7.txt.asc Download (29.3 KB) - added by manuq 13 months ago.
ephy-trace.txt Download (5.4 KB) - added by manuq 13 months ago.
browse-trace8.txt Download (40.5 KB) - added by manuq 13 months ago.

Change History

  Changed 13 months ago by manuq

Same here. Epiphany crashes in the same machine, I will try a trace.

Changed 13 months ago by manuq

  Changed 13 months ago by manuq

Attached a backtrace, it ends with:

Backtrace stopped: previous frame identical to this frame (corrupt stack?)

  Changed 13 months ago by manuq

After installing a lot of packages via debuginfo-install (more than 1.4G) I still got a slim backtrace. Attaching as browse-trace7.txt.

Changed 13 months ago by manuq

  Changed 13 months ago by manuq

Traced Epiphany, same slim backtrace. Attaching..

Changed 13 months ago by manuq

  Changed 13 months ago by tonyforster

  • cc tonyforster added

  Changed 13 months ago by dsd

Thats painful. Next time please also run "info threads", see if there is more than 1 thread running. If so, get traces from all of them. e.g.

t 1
bt
t 2
bt
t 3
bt

  Changed 13 months ago by dsd

  • cc dsd added

I also have a suspicion here that this is caused by the v8 javascript engine, that compiles javascript into ARM instructions and then executes them. Perhaps it is generating a bad instruction sequence.

If someone has time to do some exploratory work it would be interesting to know if there are ways to disable v8 and go back to "traditional" javascript execution, or what debugging options are available for v8 (can you make it dump the compiled code to disk before executing it or something?).

  Changed 13 months ago by dsd

From NZ testing group

Browse is still very unstable. Logging in to google docs causes it to
crash on the first document list screen after login.
 http://blog.makezine.com causes it to crash. Last time I tried
slashdot also caused it to crash, but this time it didn’t, so I guess
the problem does depend on exactly what the website is showing. In
every case where browse crashed, it still appeared as the active
activity in the center of the home screen even though it had
disappeared.

Changed 13 months ago by manuq

  Changed 13 months ago by manuq

Attached a trace with a backtrace for each thread.

Thanks a lot to the NZ testing group. Yes google docs is a good testing for the crash. Tested in XO-1.5 and it doesn't crash on it (doesn't open the document in a new tab but that is other issue).

follow-up: ↓ 11   Changed 13 months ago by dsd

rm /usr/lib/mozilla/plugins/*

seems to "fix" the crashes for me. Can anyone confirm?

in reply to: ↑ 10   Changed 13 months ago by manuq

Replying to dsd:

{{{
rm /usr/lib/mozilla/plugins/*
}}}

seems to "fix" the crashes for me. Can anyone confirm?

It seems so, I will keep testing.

  Changed 13 months ago by dsd

It doesn't fix it.

Here is a webkitgtk3 build with the JIT disabled at configure-time:
 http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=764667

Please help me test to see if the crash is gone.

It also includes an update to webkitgtk3-1.8.1 (compared to 1.8.0 which we tested with earlier); we'll have to do a plain 1.8.1 build to compare to really have a conclusive test

  Changed 13 months ago by dsd

Oops, that build has the JIT enabled still (and it is crashing). New one coming up.

  Changed 13 months ago by dsd

  Changed 13 months ago by dsd

  Changed 13 months ago by dsd

Seems to fix the crash.

  Changed 13 months ago by manuq

  • keywords 12.1.0, olpc-test-pending added; 12.1.0 removed
  • status changed from new to closed
  • resolution set to fixed
  • description modified (diff)

Yes, I think I tested enough (using modern web apps like gmail or google docs) and it fixes it.

  Changed 13 months ago by manuq

  • status changed from closed to reopened
  • resolution fixed deleted

  Changed 13 months ago by dsd

  • keywords 12.1.0 added; 12.1.0, olpc-test-pending removed

For now disabling the JIT is just a useful test. Ideally we will fix the bug in the JIT and leave it enabled. Commented on the upstream bug, and asked for help on the fedora-arm list. (Plan B is disabling the JIT in fedora for ARM).

  Changed 13 months ago by manuq

  • keywords 12.1.0, upstream added; 12.1.0 removed

  Changed 12 months ago by dsd

  • status changed from reopened to closed
  • resolution set to notsugar

We've disabled the ARM JIT in Fedora and now things are working. Will hopefully be able to re-enable it in future.

  Changed 12 months ago by manuq

This is tracked now in OLPC trac:  http://dev.laptop.org/ticket/11875

Note: See TracTickets for help on using tickets.