626437 - New shark code only works once, then fails, then hangs

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Boris Zbarsky [:bzbarsky]]

BUILD: Current tip TM build

STEPS TO REPRODUCE:

1)  Put the following in ~/test.js: "startProfiling(); stopProfiling();"
2)  Start shark and put it in programmatic control mode
3)  Build an optimized js shell with --enable-shark
4)  Run |js ~/test.js|
5)  Run |js ~/test.js|
6)  Run |js ~/test.js|

(yes, do it three times).

EXPECTED RESULTS: 3 profiles generated

ACTUAL RESULTS: Step 4 generates a profile.  Step 5 doesn't generate a profile.  Step 6 hangs the shell; attaching a debugger shows this stack:

#0  0x00007fff846c82da in mach_msg_trap ()
#1  0x00007fff846c894d in mach_msg ()
#2  0x000000010018022b in Shark::Start ()
#3  0x0000000100104ce9 in js::Probes::startProfiling ()
#4  0x000000010003e961 in StartProfiling ()

This severely limits the usefulness of the new profiler hooks, unfortunately.  :(

Comment 1

[Andreas Gal :gal]

Patrick does this happen with your code too or did I break something?

Comment 2

[Andreas Gal :gal]

(Patrick is not in today, we will look tomorrow.)

Comment 3

[Boris Zbarsky [:bzbarsky]]

More data:

a) Step 2 in comment 0 needs to include setting the sampling interval to 20us.
   If I leave it at 1ms, then shark gives an alert about no samples, and the js
   shell hangs.
b) The hang from step 6 above and the hang from (a) above both stop if I quit
   shark.

Comment 4

[Andreas Gal :gal]

Attachment: patch

Embarrassing.

Comment 5

[Andreas Gal :gal]

bz want to give this a spin?

Comment 6

[Patrick Walton (:pcwalton)]

Comment on attachment 505537 [details] [diff] [review]
patch

Works for me.

Comment 7

[Boris Zbarsky [:bzbarsky]]

This fixes the steps in comment 0, but only with the 20us setting.  If I use a sample interval large enough that there are no samples, then the shell still hangs waiting on shark.  I'd tell you where it hangs, but trying to attach gdb to it hangs gdb.... 

Separate bug on that?

Comment 8

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 505537 [details] [diff] [review]
patch

r=me; the other issue is more likely a problem in the sharkctl stuff...

Comment 9

[Patrick Walton (:pcwalton)]

(In reply to comment #7)
> This fixes the steps in comment 0, but only with the 20us setting.  If I use a
> sample interval large enough that there are no samples, then the shell still
> hangs waiting on shark.  I'd tell you where it hangs, but trying to attach gdb
> to it hangs gdb.... 
> 
> Separate bug on that?

Did this happen with the old CHUD stuff?

Comment 10

[Andreas Gal :gal]

http://hg.mozilla.org/tracemonkey/rev/33313f209f4d

Comment 11

[Boris Zbarsky [:bzbarsky]]

> Did this happen with the old CHUD stuff?

Not that I recall.  I certainly recall getting that popup from Shark in some cases, but I never recall the browser freezing in the process.

Comment 12

[Patrick Walton (:pcwalton)]

(In reply to comment #11)
> > Did this happen with the old CHUD stuff?
> 
> Not that I recall.  I certainly recall getting that popup from Shark in some
> cases, but I never recall the browser freezing in the process.

Okay, if you could file a followup bug then that'd be awesome. If I had to guess I'd say that Shark is trying to send us an error code, but I haven't reverse engineered that part of the protocol yet, so we end up blocking on a Mach message that never arrives.

Comment 13

[Chris Leary [:cdleary] (not checking bugmail)]

cdleary-bot mozilla-central merge info:
http://hg.mozilla.org/mozilla-central/rev/33313f209f4d

Comment 14

[Boris Zbarsky [:bzbarsky]]

> Okay, if you could file a followup bug then that'd be awesome

Filed bug 628153