While initializing an evergreen adapter on a ppc64 machine (64bit
kernel, big endian) I am getting PCI recovery errors (EEH = AER + IOMMU
faults). I'm investigating exactly why those happen, but I get them
while I'm running evergreen_irq_set on evergreen.c.
Since these kinds of errors are only detected by the kernel during a
readl call, I tried instrumenting the offending code with calls to:
RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC0_REGISTER_OFFSET)
But those seem to interfere with the normal flow of adapter configuration.
What call could I use that would be innocuous? Perhaps a thermal sensor
read?
The crash seems to be happening around the time it starts configuring
the CRTC registers on evergreen.c:evergreen_irq_set. (Though again, I've
had instrumenting issues and that might not be the case).
...
WREG32(GRBM_INT_CNTL, grbm_int_cntl);
WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
...
Any other suggestions on how to be sure of this? Also, what would
already be initialized (registered interrupts) by the time it makes this
call? Any suggestions on what else could be causing this?
The system also tries to initialize the intel_hda module upon boot. I
take it it's due to the embedded audio adapter for HDMI, I've taken the
driver down for now (removed it from initrds and blacklisted it for
modprobe). But I wonder how much could fails on the audio initialization
interfere with the graphics part on the radeon driver. Does anyone have
any estimates on this?
Whew, too many questions together :)
Thanks for any help, guys!
--
Lucas Kannebley Tavares
Software Engineer
IBM Linux Technology Center
_______________________________________________
[email protected]: X.Org development
Archives: http://lists.x.org/archives/xorg-devel
Info: http://lists.x.org/mailman/listinfo/xorg-devel
