On 64-bit Gentoo, with X.Org 7.4 (XServer 1.5.3) and kernels 2.6.26 and 2.6.27, running on Catalyst 8.11 (ati-drivers-8.552-r2) results in data corruption that sometimes manifests in segfaulting application. emerging not-so-small applications (mozilla-thunderbird) has gcc, as (assembler) and python segfault at times. That is, if you're lucky. If you're not lucky, broken machine code can be produced without you ever knowing, fscking up your whole system. Everything went stable again after reverting back to xf86-video-radeonhd. Of course I don't know if everyone is affected by this. I am. Reproducible: Sometimes Steps to Reproduce: 1. emerge ati-drivers-8.552-r2 2. emerge something large (mozilla-thunderbird in my case) Actual Results: You might get segfaults of gcc, as or python or broken resulting binaries. Expected Results: No segfaults, correctly compiled binaries. I'm attaching emerge --info
Created attachment 172606 [details] emerge --info
Looks suspiciously like a hardware problem resulting from power loss caused by the extra workload the accellerated gfx card puts on the PSU. Please reopen when you have actual evidence that the driver is corrupting your memory directly.
(In reply to comment #2) > Looks suspiciously like a hardware problem resulting from power loss caused by > the extra workload the accellerated gfx card puts on the PSU. Please reopen > when you have actual evidence that the driver is corrupting your memory > directly. I would be glad if you could tell me how to get such evidence. All I can come up with right now is: 1) fglrx results in *less* power consumption rather than more since it underclocks and undervoltages the card. xf86-video-radeonhd does not do this resulting in more power consumption and the card becomes hotter with it. 2) My PSU is a quality part (600W ThermalTake PurePower RX) rated for two cards in Crossfire while I have only one card (HD4870). It is rated for 48Amps combined output. It has 4 12V rails, 18Amps each. 3) My system passes stress tests without problems. Prime95 8 hours stable (not referring to crashes, but correct, validated data) while also generating GPU load at the same time (ATI Tray Tools.) All this on Windows. No such stress-testing tools seem to exist for Linux. 4) It happens no matter if the system is overclocked or not. Of course something else than fglrx could be at fault, but fglrx is what triggers the data corruption here. Do you still think this is invalid and not worth testing?
I forgot to mention that your justification for marking this as INVALID is actually the invalid thing here (because of what I wrote above and also because there isn't anything "accelerated" running while the segfaults occur), not my bug report.
