(Reported to Ingo Molnar who told me to send it here too.)
I've just put a new computer together with an AMD Athlon 64 3200+ and a
Chaintech ZNF3-150 motherboard. This has the Nvidia NForce 3 chipset.
Booting 2.6.4, from the debian kernel-image-2.6.4-1-k7 package, I get
reproducible errors from the onboard BCM5788 NIC and a Realtek 8139 NIC
when under load.
When the disk is under load, I get reproducible DMA errors from any IDE
disk that is under load.
When I boot with "noapic", these problems go away.
I'm happy to help debug this any way necessary.
Thanks,
Doug.
(Please CC: me, I am not subscribed)
--
http://adju.st | It's nice to be important but
6973E2CF: 2C95 66AD | more important to be nice
1596 37D2 41FC 609F | -- Scooter
76C0 A4EC 6973 E2CF |
On Tue, 23 Mar 2004 10:32:45 +0000, Doug Winter wrote:
>I've just put a new computer together with an AMD Athlon 64 3200+ and a
>Chaintech ZNF3-150 motherboard. This has the Nvidia NForce 3 chipset.
>
>Booting 2.6.4, from the debian kernel-image-2.6.4-1-k7 package, I get
>reproducible errors from the onboard BCM5788 NIC and a Realtek 8139 NIC
>when under load.
>
>When the disk is under load, I get reproducible DMA errors from any IDE
>disk that is under load.
>
>When I boot with "noapic", these problems go away.
NVidia chipsets are known to have problems when used with
local APIC or I/O-APIC. We don't know exactly what happens,
but it looks like a hardware or BIOS problem. The only
known cure is to try combinations of:
upgrade BIOS
pci=noacpi
acpi=off
noapic
nolapic
FWIW, VIA's K8T800 chipset seems to work very well.
> On Tue, 23 Mar 2004 10:32:45 +0000, Doug Winter wrote:
> >I've just put a new computer together with an AMD Athlon 64 3200+ and a
> >Chaintech ZNF3-150 motherboard. This has the Nvidia NForce 3 chipset.
> >
> >Booting 2.6.4, from the debian kernel-image-2.6.4-1-k7 package, I get
> >reproducible errors from the onboard BCM5788 NIC and a Realtek 8139 NIC
> >when under load.
> >
> >When the disk is under load, I get reproducible DMA errors from any IDE
> >disk that is under load.
> >
> >When I boot with "noapic", these problems go away.
> NVidia chipsets are known to have problems when used with
> local APIC or I/O-APIC. We don't know exactly what happens,
> but it looks like a hardware or BIOS problem. The only
> known cure is to try combinations of:
> upgrade BIOS
> pci=noacpi
> acpi=off
> noapic
> nolapic
>FWIW, VIA's K8T800 chipset seems to work very well.
Perhaps I should ask firstly how your timer interrupts are routed?
io-apic edge, local apic, or xtpic - does it still have one of these?
And you could check and see whether this patch has made it through to your
current kernel (I just remember the posting - I don't have an nforce3).
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-02/1648.html
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-02/1604.html
I don't know about idle states for K8 and nforce3 but you could also try my C1
idle halt patch (for K7 and Nforce2 - also SiS740). It may have no effect on
nforce3 but you might get lucky (don't forget the kernel arg or it won't work).
http://linux.derkeiler.com/Mailing-Lists/Kernel/2004-02/6520.html
The io-apic patch there may also help route the timer interrupts via IO-apic
assuming nvidia followed nforce2 methods.
Please CC me with your responses.
Regards
Ross.