741272 - Implement a wrapper around raise() for Android and Gonk

Status: RESOLVED FIXED

(Firefox OS Graveyard :: General, defect)

Description

[Cervantes Yu [:cyu] [:cervantes]]

Due to the bug in bionic libc's implementation of raise(), we will need a wrapper around it. Bug 730229 shows a typical symptom of the bug: when one thread hits an assertion, a random thread receives SIGABRT instead of the thread that calls raise(). The implementation in android 4.0 at the time being is even worse, which mixes gettid() and kill() and could in theory kill another process.

As POSIX specifies that raise(signo) is equivalent to pthread_kill(pthread_self(), signo), I would suggest implementing the wrapper in this way if there's no other concern. It should work correctly under bionic's pthread implementation.

Comment 1

[Cervantes Yu [:cyu] [:cervantes]]

Attachment: Wrap raise() for android and gonk

Implement raise() using pthread_kill() so the signal will be sent to the the
caller thread instead of a random thread in the thread group due to the bug in bionic libc.

Comment 2

[Mike Hommey [:glandium]]

Why don't you fix bionic in gonk instead? That would fix it for more than just gecko.

Comment 3

[Cervantes Yu [:cyu] [:cervantes]]

Per discussion with Michael Wu, the fix will be needed for both gonk and android. Firefox for android will need the fix for some time.

I also sent a patch to bionic, but we don't know how long we need to wait before the fix will be available. And there is no guarantee the fix will be backported to previous releases of Android.

Given the above reasons, it's suggested that we wrap the call in gecko.
This is the discussion thread for your reference: https://github.com/mozilla-b2g/platform_bionic/pull/2

Comment 4

[Ryan VanderMeulen (PTO, back 6-April)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/dde6ab4aa2c3

I have no idea what Target Milestone to use for B2G...

Comment 5

[(no longer active)]

https://hg.mozilla.org/mozilla-central/rev/dde6ab4aa2c3