[Unichrome-users] Unichrome X release 24

[Phil D. <deb...@gm...>]

On Sun, 22 Aug 2004 16:34:30 -0400 (EDT), Dan Morrill
<mor...@ny...> wrote:
> On Sun, August 22, 2004 12:27 pm, Phil DeBecker said:
> > However:  On these laptops, the Unichrome chipset uses IRQ 9, which
> > ACPI also uses.  If you allow the X driver to turn on the interrupt in
> > the kernel driver, something bad happens to ACPI and you lose IRQ
> > steering for IRQ 11.  Then you have no ethernet or wifi.
...
> But as you say, that's not what's going on.  Or at least, not entirely;
> possibly video ints are still confusing ACPI which breaks IRQ steering
> for 11... sounds like a bit of a stretch though.

Like I said, the problem happens when the via.ko driver installs its
interrupt handler for IRQ 9.  The scenario you describe with
report_bad_irq() is indeed what happens, to the extent that it looks
as if the IRQ 11's are not being serviced.  But since the via is on
IRQ 9, the only thing that makes sense is that ACPI IRQ steering is
getting broken somehow.  The workaround is to prevent the IRQ handler
from getting installed in the first place.  The only thing you lose
from this is the ability to use sync-on-vblank.

> However, this DOES sound more and more like a different Via chipset
> problem reported against Fedora Core on Red Hat's bugzilla site.  That
> one involved IRQ 12 IIRC, but involved some weird effect taking down
> something on a different IRQ.  Has anyone seen that bug?  Maybe I'll
> go find the URL; I think they had a fix for it that might be
> applicable.

Sure, by all means post it.  This whole IRQ bug is really weird.

> As a more general question, is this an ACPI DSDT issue?  We already
> know that the DSDT for the 32xx is a bit flaky; is it also affecting
> this?  I don't know much about the guts of ACPI at the moment, but might
> that not be something to look at?

Possibly.  The DSDT is a mess.  S1 suspend isn't available, and the
thermal zone temperature reporting don't work either.  Someone posted
a "fixed" DSDT for this at the linux acpi project on sourceforge, but
I had no luck getting it to work.  In any event, I don't know enough
about ACPI to tell if this could be fixed in the DSDT or if this is a
kernel driver issue.

Phil DeBecker