909534 - asm.js signal handler should _exit() instead of exit()

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Running at/on_exit functions is a bad idea.

Comment 1

[Luke Wagner [:luke]]

Ah, good point.  The exit() call is intended to be, as the comment explains, a backstop and never actually executed.  OOC, are you actually seeing this executed?  In which branch?

Comment 2

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

ckitching was seeing it on Android after hitting a previous fatal assertion.  Log coming up.

Comment 3

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Attachment: 2920510.txt

Comment 4

[Luke Wagner [:luke]]

Could you 'p prevHandler->sa_sigaction' in the AsmJSFaultHandler (I'm hoping for a symbolic name here).

Comment 5

[Luke Wagner [:luke]]

Also, looking at this handler again, it seems wrong to call exit() in the SA_SIGINFO branch: what if the chained handler fixes up the fault (like AsmJSFaultHandler does) or calls sigaction(..., NULL) and returns (like AsmJSFaultHandler does in the 'else' branch; so that the fault happens at the faulting instruction.

I'll patch.

Comment 6

[Luke Wagner [:luke]]

Attachment: don't call exit() at all

Comment 7

[Luke Wagner [:luke]]

Attachment: only handle SIGSEGV

This is a small related cleanup I've been meaning to do: SIGBUS was only necessary for OSX and x86 segment prefixes.

Comment 8

[Luke Wagner [:luke]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/4543cebdc8a0
https://hg.mozilla.org/integration/mozilla-inbound/rev/e89815fe247d

Comment 9

[Luke Wagner [:luke]]

Kyle: does this fix the problem for you?

Comment 10

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Chris?

Comment 11

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/4543cebdc8a0
https://hg.mozilla.org/mozilla-central/rev/e89815fe247d

Comment 12

[Chris Kitching [:ckitching]]

So it appears that by default Bugzilla stops showing the needinfo flag when the bug is closed...

You're too efficient! :P

Yes, this seems to have fixed the problem. My more recent work isn't producing unbearably huge and confusing sequences of segfaults any more. (At least, not of length >1)