Solaris using wrong IRQ with iprb

Question

P4 machine, intergrated Intel i815/82559 controller.
Installed Solaris 8 2/02, and applied most recent recommended patch cluster.
iprb0 is not able to communicate with anything on the network. snoop shows arp packets going out, but nothing comming in. In the past when I have seen this, this has been the result of an IRQ conflict with the NIC and something else. This time, the integrated NIC is using IRQ 11 according to the configuration diskette, but the system cofigures the device as if it were using IRQ 8.
How can I fix this?
# ls -l /dev/iprb*
lrwxrwxrwx 1 root other 53 Jan 16 09:16 /dev/iprb -> ../devices/pci@0,0/pci8086,244e@1e/pci1014,234@8:iprb
lrwxrwxrwx 1 root other 54 Jan 16 09:16 /dev/iprb0 -> ../devices/pci@0,0/pci8086,244e@1e/pci1014,234@8:iprb0
# prtconf -vp
Node 0x195e90
assigned-addresses: 82024010.00000000.f2000000.00000000.00001000.81024014.00000000.00002000.00000000.00000040
class-code: 00020000
compatible: 31696370.2c343130.00343332.31696370.2c343130.00343332.38696370.2c363830.39343432.69637000.73616c63.32302c73.30303030.00000000
device-id: 00002449
devsel-speed: 00000001
fast-back-to-back:
interrupts: 00000001
max-latency: 00000038
min-grant: 00000008
model: 'PCI: 1014,234 - Intel i815/82559 Ethernet'
name: 'pci1014,234'
power-consumption: 00000001.00000001
reg: 00024000.00000000.00000000.00000000.00000000.02024010.00000000.00000000.00000000.00001000.01024014.00000000.00000000.00000000.00000040
revision-id: 00000003
slot: 00000000
subsystem-id: 00000234
subsystem-vendor-id: 00001014
unit-address: '8'
vendor-id: 00008086
#prtconf -Dv
pci1014,234, instance #0 (driver name: iprb)
Driver properties:
name <chosen-interrupt> length <8>
value <0x01000000fecaddba>.
Hardware properties:
name <interrupts> length <4>
value <0x01000000>.
name <device-id> length <4>
value <0x49240000>.
name <vendor-id> length <4>
value <0x86800000>.
name <reg> length <60>
value <0x004002000000000000000000000000000000000010400202000000000000000000000000001000001440020100000000000000000000000040000000>.
Interrupt Specifications:
Interrupt Priority=0x6 (ipl 6), vector=0xb (11)

Answer 1

So if I turn off the APIC, how does the machine respond?
Let me give an example. I have a Tyan x86 system on Sol8. The chipset is Intel 7501 so it has an advanced APIC system. It also has a 7209 U320 SCSI controller. This should give better disk I/O with an U320 disk than the older U160 controller but where the older system took 45 seconds to write a 2 Gig file of zeros, the newer system took 15 minutes. So I turned of the IOAPIC and Extended IOAPIC in the BIOS. Now the disk test gives 30 seconds instead of 15 minutes. So far so good.
The sytems are used as application servers and are not disk bound (either NIC or CPU bound). They currently run OK with either BIOS setting. However, the load with the APIC enabled is under 1 with 'top' showing the application taking 15% of the CPU and the system 5%. With the APIC disabled, the application takes 50% of the CPU and the system takes 3%. I haven't done any comparative load testing (what is the max load in each configuration) because I can't. All 'load testing' is done extreeeeemely carefully in production and a load test environment is not even contemplated much less defined.
So any informed guesses as to what is going on would be appreciated. Is the CPU now handling all the interrupts that the IOAPIC system formerly took care of hence the increased load? But vmstat shows system interrupts are down from 15K to 5K with IOAPIC disabled. Could use a little help in figuring out if this is a good change or not. Suggestions for tests to run also cheerfully accepted.