Comment 48 for bug 877789

I'd like to chime in here: I am on a Thinkpad T60 (core duo -> 32 bit) with a radeon X1300 and switched to KMS + gallium recently and, while everything is working remarkably nice and stable in general (big kudos to the developers), I've got some negative side effects which look like the IRQ issues observed by the others in this thread to me. Symptoms are:

1) SATA is being reconfigured and even reset from time to time. This can be reproduced reliably by switiching VTs which causes

ata1.00: configured for UDMA/133
ata1: EH complete

to appear in dmesg most of the time. Under heavy GPU load, the link will be reset from time to time:

ata1: hard resetting link
ata1: SATA link up 1.5 Gbps (SStatus 113 SControl 300)
ata1.00: ACPI cmd ef/02:00:00:00:00:a0 (SET FEATURES) succeeded
ata1.00: ACPI cmd f5/00:00:00:00:00:a0 (SECURITY FREEZE LOCK) filtered out
ata1.00: ACPI cmd ef/10:03:00:00:00:a0 (SET FEATURES) filtered out
ata1.00: ACPI cmd ef/02:00:00:00:00:a0 (SET FEATURES) succeeded
ata1.00: ACPI cmd f5/00:00:00:00:00:a0 (SECURITY FREEZE LOCK) filtered out
ata1.00: ACPI cmd ef/10:03:00:00:00:a0 (SET FEATURES) filtered out
ata1.00: configured for UDMA/133
ata1.00: configured for UDMA/133
ata1: EH complete

Very occasionally, the controller will lock for some seconds before the reset occurs.

2) Under heavy GPU load, sound (intel HDA) has a tendency to crackle. I am not using a sound daemon, just plain alsa. The "snd-hda-intel timing workaround activated, blablabla..." message appears in dmesg.

3) I get NMIs from time to time (however, I don't see how those might be related):

Uhhuh. NMI received for unknown reason b1 on CPU 0.

I am running gentoo (mostly stable) with kernel 2.6.35 (gentoo patchset), xf86-video-ati 6.13.1, mesa bleeding-edge git (master), xorg 1.7.7 and libdrm git (master again). I've been trying different versions of everything (including vanilla 2.6.36-rc1) as well as different preemtion settings without any effect on those symptoms. As I never had such issues before switching to KMS, I stronly suspect KMS to be the culprit (however, SATA reconfiguration messages _have_ appeared before from time to time, so it might be that KMS is aggrevating a timing glitch already present in the hardware). To me, this looks somewhat like interrupts being masked overly long somewhere in the KMS code.