Debugging ethernet issues on a PowerMac G4 Cube

[Tom Lane <tgl%sss.pgh.pa.us@localhost>]

I have a NetBSD-current installation from about a month ago
(202208250910Z snapshot) running successfully on a G4 Cube,
but there's some nagging flakiness in the Ethernet interface.
dmsg reports this hardware as 

...
[     1.000000] pci2 at uninorth2 bus 0
[     1.000000] pci2: i/o space, memory space enabled
[     1.000000] pchb2 at pci2 dev 11 function 0
[     1.000000] pchb2: Apple Computer UniNorth Host-PCI Bridge (rev. 0x00)
[     1.000000] gem0 at pci2 dev 15 function 0: Apple Computer GMAC Ethernet (rev. 0x01)
[     1.000000] gem0: interrupting at irq 41
[     1.000000] lxtphy0 at gem0 phy 0: LXT971/2 10/100 media interface, rev. 1
[     1.000000] lxtphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto
[     1.000000] gem0: Ethernet address 00:30:65:b2:53:ea, 10KB RX fifo, 4KB TX fifo
[     1.000000] vmmask 7fe00000 schedmask 7fe00000 highmask 7fe00000
[     1.000000] timecounter: Timecounter "mftb" frequency 24907667 Hz quality 100
...

What I'm seeing is that incoming data seems to work fine, but
putting any heavy outbound load on it leads to performance
problems.  The kernel reports "device timeout":

[ 1479795.451779] gem0: device timeout

after which it seems to disconnect from and reconnect to the
network (which I guess is not surprising: a quick look at the
gem code shows that a device reset is what it does after a
timeout).  It seems that TCP connections manage to resume
afterwards, but there's a few seconds where nothing gets in
or out.

I'm not seeing similar issues on another G4 Mac with the same
kernel and same GMAC revision but newer PHY (10/100/1000 capable).
Maybe that says the problem is at the PHY level, but not sure.
The other Mac is a good bit speedier, so it might be a timing bug.

If anyone has tips on debugging this, I'd be grateful.

			regards, tom lane