Created attachment 659857 [details] FF troublshooting info User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:16.0) Gecko/20100101 Firefox/16.0 Build ID: 20120904124322 Steps to reproduce: casual browsing, any site, even with only one tab open. easy to reproduce when I go to 9gag.com and scroll down 10-20 posts there Actual results: FF's memory usage rapidly grows to 2GB and it starts to slow down and eventually crashes quite quickly. Happens ever since version 14, although it may have something to do with the installation of Shockwave Flash v11.4 r402 which happened around that time, but: * I disabled all plug-ins (I don't have any other add-ons) and it didn't help. * restarting in safe mode solves the problem (all plug-ins remained enabled), no rapid memory growth and memory remains at a low value, similarly to IE9. Expected results: FF should remain within reasonable memory consumption.
tried resetting FF through the troubleshooting screen - didn't help
Nicholas: Could you take a look at this. I don't understand why the resident memory is a 1.9gb
From your about:memory: 71.81 MB ── explicit 1,963.07 MB ── resident 3,968.38 MB ── vsize That's *really* odd. "explicit" and "resident" should be fairly close to each other. "vsize" is excessive, too. Can you visit about:config and change the "layers.acceleration.disabled" option to "true" by double-clicking on it, and then restart Firefox and see what happens? We've seen hardware acceleration problems cause this kind of problem before and that's the only thing I can think of.
I think a key question is, why would safe-mode fix this problem? Can you please attach the contents of your about:support both in and out of safe mode?
>I think a key question is, why would safe-mode fix this problem? The safemode disables the hardware acceleration and the assumption about the layers-acceleration from Nicholas makes sense. I would have suggested this but it didn't know that hwa could cause memory issues.
Bug 767337 is a prior case where hwaccel caused high memory consumption.
(In reply to Nicholas Nethercote [:njn] from comment #10) > Bug 767337 is a prior case where hwaccel caused high memory consumption. We've only seen those problems on Intel graphics hardware, but it is not hard to imagine that similar problems could occur with some AMD drivers. Are there any websites in particular that seem to cause this problem?
>Are there any websites in particular that seem to cause this problem? The reporter mentioned http://9gag.com in his comment. Reporter: Can you please post the graphic section from about:support here ?
I've set "layers.acceleration.disabled" to "true" and it indeed solved the problem. The first attachment in this bug is actually the about:support with "layers.acceleration.disabled" set to "false". Going to add the same with this setting on "true". Maybe since I have 2 graphics cards in CrossfireX this causes the problem?
So here's what I've done now: I've disabled CrossfireX for FF through the graphics drivers and the problem has been solved. The crazy memory growth does not happen anymore. I don't know whether you're going to try and fix this bug or leave it as a "known bug" but I guess you should at least document it somewhere.
By the way, thank you for the quick responses and the professional handling of the issue so far.
Nick, are you able to look further into this? Should this driver be on the hwaccel blocklist?
(In reply to Nicholas Nethercote [:njn] from comment #17) > Nick, are you able to look further into this? Should this driver be on the > hwaccel blocklist? Sorry, I don't have access to a machine with AMD graphics card, I couldn't recreate on my machine. We might want to block CrossfireX, since we block Optimus and it sounds like it is causing us problems.
Bas, do you have access to a CrossfireX computer?
Reporter: There is a new driver version available from AMD http://support.amd.com/de/kbarticles/Pages/AMDCatalyst129betadriver.aspx the list of fixes includey a fixed crossfire corruption with Firefox. Could you test this driver ?
Is there a bug to blocklist the old version of the driver?
bug 792480 blocked one driver version
The new 1.29 Catalyst driver seems to have solved the problem. I've removed the special Firefox profile that I'd created in the old driver, made sure that CrossfireX was enabled and FF's HWAccel was enabled. No crazy memory growth.
Great! Can we close this now?
You can close it. Thanks, all!
(In reply to Joe Drew (:JOEDREW! \o/) from comment #19) > Bas, do you have access to a CrossfireX computer? I do not, we should get one somewhere though.
(In reply to Matthias Versen (Matti) from comment #2) > btw: Firefox is currently a 32bit process and a 32bit process can allocate > only 2GB of memory. The crash is very likely a OOM crash (out of memory) Firefox is a 32bit process with Large Address Aware enabled. It can use closer to 4GB on a 64bit operating system by default, and on a 32bit OS depends on the userva setting in bcedit or /3GB setting in Boot.ini