Created attachment 30512 [details] dmesg without pci=nomsi 1/ Without compiz X starts correctly, but trying to activate compiz give windows lags every 3 seconds and it is not usable. 2/ Using the latest git code for radeon make glxgears a black screen and compiz activation give nothing but blackscreen and cursor with kernel 2.6.35 and 2.6.36. Workaround is using 2.6.34 kernel or passing the vblank_mode=0 environnement parameter. This not appends with ubuntu 10.04 stocks package. If I disable Radeon KMS in the kernel, everything works fine. With KMS everythings works normally with pci=nomsi parameter: I was tell need of pci=nomsi is significant of a kernel bug, I was thinking it as a bios or chipset issue. From git drm-radeon-testing, compiled the 17 Sept.2010 kernel-2.6.36-rc3 From ppa xorg-edgers:ddx,mesa,libdrm (but same issue with compiled from git 17 Sept. 2010) Graphic card: ATI HD4700, rv740 Motherboard: 4coredual-sata2-R2.0
Created attachment 30522 [details] lspci -vnn without pci=nomsi
Created attachment 30532 [details] proc/interrupts without pci=nomsi
Created attachment 30542 [details] Xorg log without pci=nomsi
Created attachment 30552 [details] dmesg without pci=nomsi
Created attachment 30562 [details] dmesg with pci=nomsi
Created attachment 30572 [details] lspci -vnn with pci=nomsi
Created attachment 30582 [details] proc/interrupts with pci=nomsi
Created attachment 30592 [details] Xorg.0.log with pci=nomsi
Can you post dmesg without pci=nomsi option please
Nevermind
Re-reading me I see it's a little ambigious, the workaround vblank_mode=0 doesn't solve the problem 1/ And with pci=nomsi, vblank_mode=0 is no more needed even with 2.6.35 and 2.6.36. Rafael.
What's your motherboard chipset? Is it NVidia by chance?
The motherboard chipset is VIA. Motherboard: 4coredual-Sata2 R2.0 Northbridge: VIA PT880 / PT880 Ultra Southbridge: VIA VT8237S
Seems like your motherboard chipset doesn't support MSIs properly. Do any other devices on your system have problems with MSIs? I notice a lot of hda-intel spam for example.
Is this actually a regression? If so what was the last working version? Note that UMS and KMS are different drivers and UMS does not support interrupts at all on your asic and KMS supported MSIs before your asic got interrupt support at all.
Yes, there is a sound problem, since forever with this card: but I 'kind' of solve it through the "single_command" mode in ( mode use only in debug normally), so no more spurious response. This is this bug: https://bugzilla.kernel.org/show_bug.cgi?id=18512 I don't think it is related to MSI, cause using pci=nomsi or the module option enable_msi=0 with snd-hda-intel do nothing. The fact that vblank_mode make the desktop appear since 2.6.35 (but not solving the real problem -lagging window, intermittent screen freeze-) ... I thought it was wise to indicate it but I feel more like it is a side effect of the real problem which occurs with all kernel I've tested since 2.6.32.
Dropping from the list of recent regressions. It's possible that the motherboard requires a PCI quirk for MSI to work correctly. Please use the pci=nomsi workaround for now, unless you still have problems with it.
Ok stuck with pci=nomsi, no problem so far.