View Single Post
Old 08-10-07, 07:02 AM   #17
cybasheep
Registered User
 
Join Date: Aug 2007
Posts: 3
Default Re: 6.1-Stable Freeze with 6600GT

I experimented a bit and it seems that there really is a big problem with interrupt delivery and IRQ sharing.

Disabling the ioapic (and SMP of course) results in some massive IRQ sharing, as expected:

Code:
hint.apic.0.disabled="1"
kern.smp.disabled="1"
Code:
 0: *timer*
 1: atkbdc0 <Keyboard controller (i8042)>
    atkbd0 <AT Keyboard>
    atkbdc0 <Keyboard controller (i8042)>
    atkbd0 <AT Keyboard>
 2: N.A.
 3: uhci3 <VIA 83C572 USB controller>
    sio1
    uhci3 <VIA 83C572 USB controller>
    sio1
 4: sio0
    sio0 <16550A-compatible COM port>
    sio0
    sio0 <16550A-compatible COM port>
 5: bktr0 <BrookTree 878>
    uhci2 <VIA 83C572 USB controller>
    ehci0 <VIA VT6202 USB 2.0 controller>
    bktr0 <BrookTree 878>
    uhci2 <VIA 83C572 USB controller>
    ehci0 <VIA VT6202 USB 2.0 controller>
 6: fdc0 <floppy drive controller (FDE)>
    fdc0 <floppy drive controller (FDE)>
 7: ppc0 <ECP parallel printer port>
    ppc0 <ECP parallel printer port>
 8: *rtc*
 9: free
10: pcib2 <ACPI PCI-PCI bridge>
    nvidia0 <GeForce 7600 GT>
    pcib3 <ACPI PCI-PCI bridge>
    pcib4 <ACPI PCI-PCI bridge>
    pcib5 <ACPI PCI-PCI bridge>
    re0 <RealTek 8168/8111B PCIe Gigabit Ethernet>
    pcib6 <ACPI PCI-PCI bridge>
    uhci0 <VIA 83C572 USB controller>
    pcm1 <Envy24 audio (Terratec DMX 6fire)>
    pcib2 <ACPI PCI-PCI bridge>
    nvidia0 <GeForce 7600 GT>
    pcib3 <ACPI PCI-PCI bridge>
    pcib4 <ACPI PCI-PCI bridge>
    pcib5 <ACPI PCI-PCI bridge>
    re0 <RealTek 8168/8111B PCIe Gigabit Ethernet>
    pcib6 <ACPI PCI-PCI bridge>
    uhci0 <VIA 83C572 USB controller>
    pcm1 <Envy24 audio (Terratec DMX 6fire)>
11: atapci0 <VIA 8237A SATA150 controller>
    uhci1 <VIA 83C572 USB controller>
    pcm0 <VIA VT8251/8237A High Definition Audio Controller>
    atapci0 <VIA 8237A SATA150 controller>
    uhci1 <VIA 83C572 USB controller>
    pcm0 <VIA VT8251/8237A High Definition Audio Controller>
12: free
13: *npx*
14: free
15: free
... and with that, it takes no longer than five minutes of running X before the problem appears. Normally I'd be quite happy to just blame PC hardware as usual, but I tried stressing some other devices sharing IRQ 10 for testing: playing music on pcm1 while continuously rsyncing data over re0 (which is a PCIe device as well) from a memory card in a reader attached to a port whose parent is uhci0 and none of them went haywire, plus it-works-in-windows, so there is some hope left this could be actually fixed by other means than shuffling pci cards (or even motherboards) around.
cybasheep is offline   Reply With Quote