Closed Bug 494762 Opened 12 years ago Closed 12 years ago

large increase in number of FF 3.0.10 crashes after 10.5.7 update

Categories

(Firefox :: General, defect)

3.0 Branch
x86
macOS
defect
Not set
critical

Tracking

()

RESOLVED INVALID

People

(Reporter: jtd, Unassigned)

References

()

Details

After Apple released 10.5.7 around May 12, 2009, there's been a large increase in the number of FF 3.0.10 mac crashes.  The URL link shows the number of catch-all objc_msgSend crashes, the top mac crasher, by OS version.  Note the strange spike in 10.5.6 crashes and the steady rise of 10.5.7 crashes while other versions seem relatively constant.

Overall summary

https://spreadsheets.google.com/pub?key=rdbgpMikWLN2f2BR1Ym12jw&hl=en

The first sheet contains the overview, the other sheets a breakdown for objc_msgSend crashes which shot up to over 9000 last week.  The rate of Flash_EnforceLocalSecurity crashes more than tripled from the week before.  The chart is a bit hard to read in the published version, so here's a better screenshot:

http://people.mozilla.org/~jdaggett/increaseincrashes.png

Note: the 500 maximum is probably an artifact of a crash reporter limit, my guess is these numbers are probably higher.

This is based on pulling down crash reports for the last 30 days and analyzing the stack crawls.  This is just for objc_msgSend crashes.

There seems to be a lot of problems when printing, especially for folks using Canon printers.  But the rate of other bugs seems to have increased also which makes me suspect there's some sort of memory corruption problem that Apple introduced with this update.

My guess is that this is an Apple-related problem but I logged this just in case there's something we need to do to ameliorate the problem.
Looks like May 12, 2009 is a release date of an update of Adobe Reader.
http://www.adobe.com/support/downloads/product.jsp?product=10&platform=Macintosh

Can plug-ins contaminate the memory?
(In reply to comment #1)
> Looks like May 12, 2009 is a release date of an update of Adobe Reader.
> http://www.adobe.com/support/downloads/product.jsp?product=10&platform=Macintosh

Last I knew, the Adobe Reader plug-in could only be loaded by Safari, so that seems unlikely.

Marcia's already filed a bug on the CFReadStreamGetStatus crash (it's Flash in all the other frames, in a non-main thread), bug 494448.
Also possibly related:

As of OS X 10.5.7, Apple seems to have updated their bundled version of
the Flash plugin to 10.0r22 (from 9.0r151).  And the new version seems
to have at least some Cocoa code (while 9.0 was purely a Carbon app)
-- class-dump returns a non-empty result on the 10.0r22's executable
in Contents/MacOS, while it returns nothing on the 9.0r151 executable.

If the new Flash plugin is a Cocoa app, some of the objc_msgSend
crashes may be in the Flash plugin.
> Looks like May 12, 2009 is a release date of an update of Adobe Reader.
> http://www.adobe.com/support/downloads/product.jsp?product=10&platform=Macintosh
> 
> Can plug-ins contaminate the memory?

Definitely possible but it's hard to imagine that Adobe Reader use is so
widespread as to have caused this.  Besides, by default the PDF viewer
is Preview.app, not Adobe Reader on the Mac.

> If the new Flash plugin is a Cocoa app, some of the objc_msgSend
> crashes may be in the Flash plugin.

I checked this.  Of the objc_msgSend errors only a small number had
Flash in the crashing thread stack crawl and the number didn't change
after the update.  Click on the spreadsheet, then on "chart of crash
types".  Crashes in objc_msgSend with Flash on the stack are in red.

There was still a big increase in non-objc_msgSend Flash crashes, so it's definitely possible the Cocoa-ey nature of Flash is part of the problem.
>> If the new Flash plugin is a Cocoa app, some of the objc_msgSend
>> crashes may be in the Flash plugin.
>
> I checked this.  Of the objc_msgSend errors only a small number had
> Flash in the crashing thread stack crawl and the number didn't
> change after the update.  Click on the spreadsheet, then on "chart
> of crash types".  Crashes in objc_msgSend with Flash on the stack
> are in red.

OK.  I hadn't read your comment #0 carefully enough, and didn't
realize that your figures for objc_msgSend crashes include the totals
for all the other "crash types" (i.e. that what you've provided in
your 2nd through 4th sheets is a breakdown of different kinds of
objc_msgSend crashes).

By the way, I also find the simultaneous increases in crashes on
10.5.6 and 10.5.7 utterly baffling.  Baffling enough to wonder if this
isn't some kind of artifact in Socorro's data (or in how you've
analyzed it).
>  There seems to be a lot of problems when printing, especially for folks using
>  Canon printers.

A Canon Printer driver bug has been around for awhile 
https://bugzilla.mozilla.org/show_bug.cgi?id=460963
https://bugzilla.mozilla.org/show_bug.cgi?id=493655

I wonder if we have any contacts at Apple that could confirm addtional problems with the udpate, or if they could give advice on any APIs changes that we need to watch out for or correct.
Nothing really happened with this bug and we're not really sure what caused it. Later in June, we saw an uptick in the number of Flash on Mac crashes. Those were handled in a number of bugs with reports to Adobe (after we found reproducible sites). I haven't seen anything else that would cause an uptick and I'm not sure what to do with this bug.
There are bugs on most of the specific crashes.  This bug isn't useful any more.
Status: NEW → RESOLVED
Closed: 12 years ago
Resolution: --- → INVALID
You need to log in before you can comment on or make changes to this bug.