I'd been successfully using the patch from bug 1469496 on Linux with an NVIDIA Quadro P400 (with gfx.webrender.all=true), using the proprietary drivers, but it recently stopped working; I think a driver upgrade (to version 390.77) was the relevant thing that changed. (This is all on mozilla-central builds.) I tried attaching with gdb and setting a conditional breakpoint on the call to fGetGraphicsResetStatus[1] returning nonzero, but it never did (nor was a simple breakpoint inside the `if` statement reached). The call itself was reached, which implies IsSupported(gl::GLFeature::robustness) was true. [1] https://searchfox.org/mozilla-central/rev/847b64cc28b74b44c379f9bff4f415b97da1c6d7/gfx/webrender_bindings/RendererOGL.cpp#161

:nical, can you comment to the bug?

I'm not sure what to do about this. I wasn't actually able to trigger the memory purge events on my computers so I pretty much wrote the code as suggested by the NV_robustness_video_memory_purge without being able to test. The only other place we call fGetGraphicsResetStatus is in WebGL code (so a differen gl context) which rules out another call site flushing out the error code. It could be that the problem you are seeing is not caused by a memory purge or that the driver isn't properly reporting the event. Jed, could you describe in details what kind of corruption you are seeing, if/how you manage to trigger it and if there is anything that looks interesting in the stdout/stderr ?

My STR is running `echo mem > /sys/power/state` as root, and then pressing a key to resume. I'll attach a screenshot. Newly loaded tabs aren't affected (including by closing a tab and restoring it with Ctrl-Shift-T, but reloading a tab doesn't fix it).

Debian Testing, Gnome on Xorg, GTX 1060, proprietary Nvidia driver 390.77 1. mozregression --launch 2018-08-20 --pref gfx.webrender.all:true 2. As root in a seperate terminal: echo mem > /sys/power/state 3. Hit a key to wake up. 4. Started screencast.

This completes the patch from bug 1469496, actually requesting a context reset after a memory purge event. Tested on my NVIDIA laptop.

OK that extension is braindead: NV's driver should definitely be always generating context-loss on these events. I /guess/ we have to take this patch, but we do so under duress.

https://mozilla.logbot.info/gfx/20180829#c15236290 > hi, I don't have commit access, so could someone land the patch in bug 1484782 for me? Comment 8 seems to agree ("OK", "should definitely", "but we do so under duress"). https://wiki.mozilla.org/Sheriffing/How_To/Landing_checkin-needed_patches

Pushed by ebalazs@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/88973a92ac86 Enable Video Memory Purge extension if available. r=jgilbert

Works for me. Thanks!

Update: it works on the Quadro P400 I mentioned in comment #0, but not a GeForce GTX 960 (same driver version). I can file a new bug on that when I have some time to look into it.

(In reply to Jed Davis [:jld] (⏰UTC-6) from comment #13) > Update: it works on the Quadro P400 I mentioned in comment #0, but not a > GeForce GTX 960 (same driver version). I can file a new bug on that when I > have some time to look into it. The reason is that I was accidentally running an older build of Firefox than I thought. Apologies for the false alarm.

(Sorry, phabricator hickup)