With apic enabled on this board, some normal amount of disk access
(including building a kernel) is successful. However, irqs are messed up.
The eepro100 module for the onboard 82559 nic loads, but prints this after
the revision history: PCI: No IRQ known for interrupt pin A of device
00:07.0. Probably buggy MP table.
Also, when booting with apic, I get this (with both MP Spec 1.1 and 1.4)
... IO APIC version: 0011
WARNING: unexpected IO-APIC, please mail to [email protected]
dmesg for both MP Spec 1.1 and 1.4 for ac8 boot are attached (it looks
like there's one more irq in one of the modes, eepro has same error on
both). ac22 has the same io-apic warning and eepro100 load-time messages.
with noapic, dmesg includes this nugget of information after eepro100
revision history and before all the usual diagnostic s and self-tests:
PCI: Enabling device 00:07.0 (0014 ->0017)
PCI: Assigned IRQ 5 for device 00:07.0
This is all with the latest bios vdls1010. The changelog lists "Support
Solaris 8.0 dual processor environment" so I thought it was worth a shot,
but doesn't seem to fix the problem.
On Wed, 27 Jun 2001, J. Nick Koston wrote:
> There seems to be a major problem with this board and 2.4.x kernels.
> I consistantly get SCSI Input/Output errors on multiple drives that I
> know are good when running a SMP kernel. These errors do no happen
> with a UP kernel. This is happening on multiple systems and with
> multiple know good scsi drives of all speeds and sizes.
(modified)
> Test#1:
> dd if=/dev/sdc of=/dev/null
> (fails with input/output error)
When running this (kernel booted with noapic), I get no io errors (only
ran it once), but there is a bit of disk access on sda. It seems the
kernel (2.4.6-pre8) has decided to use 4MB of swap in addition to the
200+MB of physical ram. I guess this is another weird system situation
that the vm could be optimized for, but at the expense of good performance
for the other 99% of the world.
Mem: 255436K av, 250848 used, 4588k free, 0k shared, 222712k buff
swap: 2050000k av, 4192k used, 2045808k free 644k cached
bdflush and kswapd are taking 1-5% cpu each
vmstat: 1 0 0 4160 3572 219376 4232 4 2 5275 795 148 3684 2 9 89
justin