2.4.10-pre15 -> final breaks IOAPIC on UP?

2.4.10-pre15 -> final breaks IOAPIC on UP?

[Matthias Andree]

Hi,

I recently reported difficulties with several 2.4.9-ac and
2.4.10-vanilla versions that froze solid when exiting an X11 display and
returning to text-mode display (XFree 4.1.0, Diamond V550 (Riva TNT)
with nv driver), not even SysRq would help.

After some research (took several hours worth of trail & error), I think
I found the culprit, but first look at the figures. All share the same
configuration which I can provide on request, which includes IOAPIC
support on Uniprocessor machines (Duron/700 in Gigabyte 7ZX-R here):

2.4.9: ok
2.4.10-pre7: ok
2.4.10-pre12: ok
2.4.10-pre14: ok
2.4.10-pre15: ok
2.4.10-final: broken, freezes
2.4.10-final: ok if compiled without IOAPIC support on UP.

So, I believe that any IOAPIC related change between 2.4.10-pre15 and
2.4.10-final breaks my X11 here.

Any comments?

-- 
Matthias Andree

"Those who give up essential liberties for temporary safety deserve
neither liberty nor safety." - Benjamin Franklin

Re: 2.4.10-pre15 -> final breaks IOAPIC on UP?

[Alan Cox]

> So, I believe that any IOAPIC related change between 2.4.10-pre15 and
> 2.4.10-final breaks my X11 here.

Uniprocessor IO-APIC only works for some machines. It also subtly changes
IRQ delivery timing properties which may be worth checking too

Alan

Re: 2.4.10-pre15 -> final breaks IOAPIC on UP?

[Eyal Lebedinsky]

Alan Cox wrote:
> 
> > So, I believe that any IOAPIC related change between 2.4.10-pre15 and
> > 2.4.10-final breaks my X11 here.
> 
> Uniprocessor IO-APIC only works for some machines. It also subtly changes
> IRQ delivery timing properties which may be worth checking too

I have an SMP machine (Epox D3VA, 2xPIII/933) which needs "noapic" boot
or it fails to read the "partition check" (lost interrupt). It is this
way since 2.4.8 and around 2.4.9-ac12 I checked and it had the same
problem. The problem is 100% reproducible.

I believe there are still some problems with apic even on SMP. Maybe
only on some mobos?

--
Eyal Lebedinsky (eyal@eyal.emu.id.au) <http://samba.anu.edu.au/eyal/>

Re: 2.4.10-pre15 -> final breaks IOAPIC on UP?

[Eric W. Biederman]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

> > So, I believe that any IOAPIC related change between 2.4.10-pre15 and
> > 2.4.10-final breaks my X11 here.
> 
> Uniprocessor IO-APIC only works for some machines. It also subtly changes
> IRQ delivery timing properties which may be worth checking too

All athlon motherboards with the via686 southbridge and an AMD northbridge
should fail according to the AMD errata.  This may also happen with a via
northbridge.  The AMD apic bus is subtely different from the intel apic
bus in implementation.

Eric

Re: 2.4.10-pre15 -> final breaks IOAPIC on UP?

[Alan Cox]

> > Uniprocessor IO-APIC only works for some machines. It also subtly changes
> > IRQ delivery timing properties which may be worth checking too
> 
> All athlon motherboards with the via686 southbridge and an AMD northbridge
> should fail according to the AMD errata.  This may also happen with a via
> northbridge.  The AMD apic bus is subtely different from the intel apic
> bus in implementation.

Care to knock up a PCI quirks patch for it ?