Trying to make headway into finding the Xid crashes source...

[Dabljuh]

Ok I think I have found the definite solution to the problem.

Its not DPMS. That merely coincided with some other alterations I did.

The solution is to make sure the GFX card has its own, unshared interrupt. As the problem in fact seems to be about interrupt handling, I went to assign the different PCI slots distinct IRQs and turn acpi off. now cat /proc/interrupts shows that nvidia has its own unshared interrupt, where as before it would share with the sound card - emu10k1 (SB Live)
The SB Live is famous for being a real ressource hog on the PCI bus.

I'm not too much into assigning IRQs in linux, if someone knows how to do it the correct way (hard-assign IRQs for different pci devices via linux / acpi) please post.

gl-117 works no problem, tested high gfx settings and low. what used to crash me the fastest was the doom 3 menu, which works fine too

the IRQ error has been proofed by going back to having sb live share IRQ with the nvidia card and get the well known crashes back. I don't know if its an AGPGart problem something else with my hardware, but my guess is the problem lies in the nvidia drivers.

[Darkvater]

I don't think it is a PCI IRQ problem. This is what I have:

           CPU0
  0:    5277376          XT-PIC  timer
  1:       6152          XT-PIC  i8042
  2:          0          XT-PIC  cascade
  5:      20092          XT-PIC  uhci_hcd, uhci_hcd, eth0, EMU10K1
  7:      53522          XT-PIC  libata
  9:          0          XT-PIC  acpi
 11:     318353          XT-PIC  nvidia
 12:      68695          XT-PIC  i8042
 14:        520          XT-PIC  ide0
 15:      38827          XT-PIC  ide1
NMI:          0
LOC:          0
ERR:          0
MIS:          0

which if I am correct has IRQ 11 for only Nvidia. Of course I have no more Xid's anymore thank god, but that is only since I hardcoded AGP to 2x instead of 4x and turned off 'AGP Fast Wires'. Altought IRQ sharing might still be a problem cause my system intermittedly crashes when I plug in an USB stick (and its irq is shared if I read this table correctly).

I am no linux expert though, just telling my experiences.

[Dabljuh]

What I see is that you have an unshared IRQ for your card and you don't have the problem this thread is about.

Oh, and fast writes always caused crashes for me and still does. No matter the OS or the card or the driver.

[gilboa]

Ouch... If I recall correctly the nVidia is sharing an IRQ with my USB2 controller.
However, my A7M266D IRQ sharing is hard-coded (Which leaves me with no choice.) so I can't really change it.

I'll try and google for a way to over-ride the BIOS defaults (by using ACPI, I assume)

Cheers,

[Dabljuh]

You'll probably need to fiddle around a bit. On my system, noacpi... isn't really. The interrupts seem a bit random, in fact the settings in the BIOS only seem to partially correlate with the what cat /proc/interrupts reports. Make however sure you have the latest beta version of your BIOS.

After all, just as an experiment, you can try disabling USB in the bios and use PS/2 connectors for the mouse and see if that helps.

[gilboa]

The A7M266D is an old dual AthlonMP workstation board; Asus stopped releasing BIOS updates for it ages ago... :/
Disabling stuff won't help, PCI slot 3 (where the USB2 controller sits) is tied to the AGP port.
As I have no other free PCI slots, I cannot even move the device and free the slot...

Cheers,

[tobiasoed]

I have the same symptoms as Darkvater: with acpi eanabled and nvidia holding it's own interrupt the system is unstable. Only limiting agp to 2 in the bios makes things stable.
It seems as though everybody who solved their problem by assigning an interrupt to nvidia disabled acpi to get it done. My question is: Is it an acpi problem or an interrupt sharing one. Does anyone with acpi eanabled have a stable system now?
Tobias.
ps: recompiling the kernel to try out the NvAGP right now. Why is agpgart built into the fc3 kernels?

[Dabljuh]

agpgart is built into most kernels because, well, its the agp support, you know. Nvidia is just about the only gfx driver that comes with its own, optional agp.

[Nyago]

Bah. I woke up this morning to a frozen display.

I'm playing around with interrupts now. My bios allows me to set them explicitly per pci slot.

[gilboa]

God I hate Linux sometimes.

The problem is not the board... it's the f****n ACPI driver.

[root@gilboa-home-dev linux]# cat /proc/interrupts
           CPU0       CPU1
  0:     623449     608710    IO-APIC-edge  timer
  1:        871       1047    IO-APIC-edge  i8042
  8:          0          1    IO-APIC-edge  rtc
  9:          1          0   IO-APIC-level  acpi
 14:       5157       5295    IO-APIC-edge  ide0
 15:       5489       4962    IO-APIC-edge  ide1
169:      15703      31399   IO-APIC-level  aic7xxx
177:       6990         13   IO-APIC-level  CMI8738-MC6, ehci_hcd, eth0
185:          0          0   IO-APIC-level  EMU10K1
193:        164         18   IO-APIC-level  ohci_hcd
201:        163       1307   IO-APIC-level  ohci_hcd, nvidia
NMI:          0          0
LOC:    1232159    1232160
ERR:          0
MIS:          0

Not matter what I do I cannot disable the CONFIG_ACPI_BOOT flag, which controls the ACPI-on-boot function, which in-turn, does the god-awful IRQ allocation seen above. (I assume that it's hard-wired to SMP support or APIC)
The IRQ problem is not BIOS related (The BIOS manual IRQ allocation is being ignored completely). This is a pure Linux problem.
Adding noacpi, acpi=off, acpi=noirq or pci=biosirq does nothing.

Any ideas how can I control the ACPI IRQ routing functions?
Or: how can I force the OS to use the BIOS assigned IRQs?

[tobiasoed]

You may want to try APIC instead of ACPI. One stand for advanced programable interrupt controller the other for advanced configuration and power interface IIRC. There are a few kernel parameters related to apic (in particular nolapic). Have a look at Documentation/kernel-parameters.txt wherever that's on your system. (for fc3 it's in below /usr/share/doc/kernel-doc-xx/ for fc2 it used to be below /usr/src/linux-xx/).
Tobias
rant: Why in the world did they got rid of kernel-sources in fc3, I'm having a heck of a time getting the kernel to compile using the source package. Similarly, with agpgart being built in: I have machines that don't ever start X, I have no need for agp or framebuffers on them. fc is getting bloated.

[gilboa]

Outch... you are right... I did confuse ACPI And APIC... I wrote the email at 3am and guess I was way-too-sleepy to function

As for apic, if I recall correctly, you can only disable it; not control it. And last time I tried running with noapic my SCSI controller went to hell... (MPI

Side note: I disabled the USB2 part of the USB controller and left the machine running overnight with a couple of OGL screen-saver; Didn't crash. Tried playing some RO/UT2K4 and ETF. Didn't crash.
It usually crashed within minutes, so even if it didn't solve the problem; the machine is indeed more stable.

Cheers,