+++ This bug was initially created as a clone of Bug #613274 +++ Loading the galactic demo in the 2011/5/8 nightly causes my X server process to spin at 100% CPU for minutes at a time, resulting in a completely unusable desktop. I'm running Xorg 1.9.0 with the binary NVIDIA driver version 260.19.06. I'm not sure there's a lot we can do here; more filing this bug for tracking/brainstorming purposes. The demo itself is fundamentally wrong for shoehorning a webgl use case into canvas, and is apparently badly written to boot, but I guess it's Webby. Judging by what |top| said while I was SSH'd into the DoS'd machine, firefox-bin is staying responsive. It looks like firefox-bin is firing off commands to Xorg that force it into very slow CPU fallback paths. (Possibly related to the non-rectangular clips discussed in bug 613274.) If there's nothing we can optimize within firefox-bin, I wonder if we could fire off a tracer ping to the X server to check its responsiveness, like what window managers would send to firefox-bin. If we don't get a response within K seconds, maybe we could suspend script execution a la slow-script timeout? A problem here is that other applications hogging the X server would trip our detector, and we can't determine with 100% certainty which particular web pages are responsible for the DoS. Maybe this is something we'll just have to wait for Wayland to fix.
I also wondered about checking that the server had caught up with the requests sent for bug 559110, but I'm tempted to think the X server is the place to fix this. Perhaps it could either push back on the client sending requests (Bug 559110 comment 12) or prioritize input events. FWIW here (X.Org X Server 1.10.1, radeon 6.14.1) the X workspace certainly slows down considerably but I can still interact with apps.
Sounds like this might be a bug/missing-feature in the nvidia driver.
You need to log in before you can comment on or make changes to this bug.