I take back some of my "works OK" comment. 

I have a pair of Alphas with a point-point link between 82558 cards.  
It doesn't work well with heavy load in both directions.  It doesn't 
crash, just gets horrible throughput.  That's with 1.4.

It seems ok when the traffic is mostly in one direction. 

A similar setup on an Intel platform works fine.

I'll be glad to help track this down if anybody has any good ideas.


-----

Somebody half-offered to post this, but their machine wasn't available... 

This is from a Compaq NC3131 and 3132 on a NetBSD 1.4 system.  It's 
an Alpha/Miata. 


ppb0 at pci0 dev 11 function 0: Digital Equipment DECchip 21154 PCI-PCI Bridge (rev. 0x02)
pci1 at ppb0 bus 1
pci1: i/o enabled, memory enabled
fxp0 at pci1 dev 4 function 0: Intel EtherExpress Pro 10+/100B Ethernet
fxp0: interrupting at dec 550 irq 12
fxp0: Ethernet address 00:08:c7:91:b6:d8
inphy0 at fxp0 phy 1: i82555 10/100 media interface, rev. 0
inphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto
fxp1 at pci1 dev 5 function 0: Intel EtherExpress Pro 10+/100B Ethernet
fxp1: interrupting at dec 550 irq 13
fxp1: Ethernet address 00:08:c7:91:b6:d9
inphy1 at fxp1 phy 1: i82555 10/100 media interface, rev. 0
inphy1: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto
fxp2 at pci1 dev 6 function 0: Intel EtherExpress Pro 10+/100B Ethernet
fxp2: interrupting at dec 550 irq 14
fxp2: Ethernet address 00:08:c7:e6:0f:a0
inphy2 at fxp2 phy 1: i82555 10/100 media interface, rev. 0
inphy2: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto
fxp3 at pci1 dev 7 function 0: Intel EtherExpress Pro 10+/100B Ethernet
fxp3: interrupting at dec 550 irq 15
fxp3: Ethernet address 00:08:c7:e6:0f:a1
inphy3 at fxp3 phy 1: i82555 10/100 media interface, rev. 0
inphy3: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto