Sorry, other commitments mean I will have very limited time to devote to this for at least the next week, maybe the next two weeks.
There are a number of inconsistencies in the accounts so far:
1. Alex and Pierre both say they are using fxp0 and fxp1 but fxp1 and fxp2 show significant interrupt rates for Alex while Pierre sees significant interrupt rates for fxp0 and fxp1.
2. Pierre reports interrupt storm on irq10 with my suggested modifications to /boot/loader.conf.local but the IPSO startup reports only one device on irq10: fxp1 and fxp1 was assigned to irq10 in my suggested modifications.
Maybe the boxes have different motherboards. Maybe the IPSO startup output we have is from a different box
From memory, the front panels of these devices designate the onboard (not on expansion slot) NICs as eth1, eth2, eth3 and eth4 but both Pierre and Alex report devices fxp0 and fxp1 in use. To get this to work correctly we need appropriate IPSO source code (used to be available from Nokia on request if I recall correctly; don't know about Checkpoint who took over the firewalls from Nokia) or an accurate association between front panel port name, FreeBSD device name and IRQ. Both Pierre and Alex have made associations between front panel port names and FreeBSD device names. I want to know how these associations were derived. If they have been assumed then they can't be trusted and they will need to be derived from the source code or worked out by experiment..
If anyone can track down the source code for a version of IPSO that supported IP380 I'll take a look at it to try to derive the association I requested.
There is something strange with this output: irq6, irq12 and irq15 are not set as I expected, but... it works.
All the hw.pci....irq entries for fxp NICs and dc NICs need to specify INTA because these devices request interrupts on their INTA line. The INTx reported by IPSO refer to interrupt lines on the motherboard which is different from interrupt lines the actual devices use. More detailed explanation will have to wait.