> > We're trying to track down the source of a problem that occurs
> > whenever the atl1 network driver is activated on a 32-bit 2.6.21-rc4
>
> and -rc5, -rc6, 2.6.20.x, 2.6.19.3, and probably others.
>
> > We can load the driver just fine, but whenever we activate the
> > network, we see APIC errors (a sample of them are shown here,
> > captured from a serial console):
> >
> > [root@hawk ~]# echo 8 > /proc/sys/kernel/printk
> > [root@hawk ~]# [ 93.942012] process `sysctl' is using deprecated
> > sysctl (sysc.
> > [ 94.396609] atl1: eth0 link is up 1000 Mbps full duplex
> > [ 94.498887] APIC error on CPU0: 00(08)
> > [ 94.498534] APIC error on CPU1: 00(08)
> > [ 94.550079] APIC error on CPU0: 08(08)
> > [ 94.549725] APIC error on CPU1: 08(08)
> > [ 94.600915] APIC error on CPU1: 08(08)
> > [ 94.601276] APIC error on CPU0: 08(08)
> > [ 94.652108] APIC error on CPU1: 08(08)
> > [ 94.652470] APIC error on CPU0: 08(08)
> > [ 94.703659] APIC error on CPU0: 08(08)
> > [ 94.703305] APIC error on CPU1: 08(08)
> > [ 94.754852] APIC error on CPU0: 08(40)
> > [ 94.806045] APIC error on CPU0: 40(08)
/* Here is what the APIC error bits mean:
0: Send CS error
1: Receive CS error
2: Send accept error
3: Receive accept error
4: Reserved
5: Send illegal vector
6: Received illegal vector
7: Illegal register address
*/
So the 40 means the APIC got an illegal vector.
Certainly this is consistent with the fact that
the errors start when a specific device is being
used. I assume that device is using MSI?
Curious that it is different in 32-bit and 64-bit mode.
> > [ 94.805692] APIC error on CPU1: 08(08)
> > [ 94.857238] APIC error on CPU0: 08(08)
> > [ 94.856884] APIC error on CPU1: 08(08)
> > [ 94.908432] APIC error on CPU0: 08(08)
> > [ 94.908078] APIC error on CPU1: 08(08)
> > [snip, more of the same]
> > [ 98.901156] APIC error on CPU1: 08(08)
> > [ 98.952702] APIC error on CPU0: 08(08)
> > [ 98.952349] APIC error on CPU1: 08(08)
> > [ 99.003895] APIC error on CPU0: 08(08)
> > [ 99.003542] APIC error on CPU1: 08(08)
> >
> > The machine hangs for about 5-10 seconds, then spontaneously reboots
> > without further console output.
>
> I can prompt an oops by pinging my router while the apic errors are
> scrolling by.
>
> >
> > This is an Asus M2V (Via K8T890) motherboard.
> >
> > The problem does not occur on a 32-bit kernel if we boot with
> > pci=nomsi, and it doesn't occur at all on a 64-bit kernel on the same
> > motherboard.
pci=nomsi, works, okay...
> > We also do not see this problem on Intel-based motherboards, with
> > either 32- or 64-bit kernels.
>
> A full raft of documentation -- including acpidump and
> linux-firmware-kit output, console capture, kernel config, lspci -vvxxx
> (with apic=debug boot option), dmesg, and /proc/interrupts -- is
> available at http://www.hogchain.net/m2v/apic-problem/
[06Dh 109 2] Boot Architecture Flags : 0003
for what it is worth, the bit in ACPI that is used to
disable MSI support is not set -- so as far as the BIOS
is concerned, this system should support MSI.
Is it an add-in card, or lan-on-motherboard?
-Len
Thank you very much for looking at this, Len.
On Fri, 11 May 2007 23:28:58 -0400
Len Brown <[email protected]> wrote:
> > > [ 94.754852] APIC error on CPU0: 08(40)
> > > [ 94.806045] APIC error on CPU0: 40(08)
>
> /* Here is what the APIC error bits mean:
> 0: Send CS error
> 1: Receive CS error
> 2: Send accept error
> 3: Receive accept error
> 4: Reserved
> 5: Send illegal vector
> 6: Received illegal vector
> 7: Illegal register address
> */
>
> So the 40 means the APIC got an illegal vector.
> Certainly this is consistent with the fact that
> the errors start when a specific device is being
> used. I assume that device is using MSI?
Yes, the device is using MSI.
> Curious that it is different in 32-bit and 64-bit mode.
Agreed, although I had one user back in March report APIC errors on the
Asus M2V board while running Debian x86_64. I personally have never
encountered the problem under a 64-bit kernel, but I admit that just
might be random luck.
> > > We also do not see this problem on Intel-based motherboards, with
> > > either 32- or 64-bit kernels.
> >
> > A full raft of documentation -- including acpidump and
> > linux-firmware-kit output, console capture, kernel config, lspci
> > -vvxxx (with apic=debug boot option), dmesg, and /proc/interrupts
> > -- is available at http://www.hogchain.net/m2v/apic-problem/
>
>
> [06Dh 109 2] Boot Architecture Flags : 0003
>
> for what it is worth, the bit in ACPI that is used to
> disable MSI support is not set -- so as far as the BIOS
> is concerned, this system should support MSI.
>
> Is it an add-in card, or lan-on-motherboard?
This is a PCIe LAN-on-motherboard.
My goal is to understand whether this is a problem in the atl1 driver,
or a problem on the motherboard. If it's the former, obviously I want
to fix it. If it's the latter, then I want to disable MSI in the driver
when we discover we're running on this motherboard.
Thanks again for taking time to look at this. Any advice or hints you
provide will be greatly appreciated.
Jay