wm(4) won't negotiate 1000-base-T with Cisco switches and won't operate with manual media settings

[buhrow%lothlorien.nfbcal.org@localhost (Brian Buhrow)]

        hello.  I've been running the following chips under NetBSD 4.x for a
number of years at 100mbits/sec without any problems at all.  However,
recently, when I tried to link one of these up to a gige port on a Cisco
ME3400 switch, it failed to establish link.  If left to autonegotiate, it
establishes a link with the Cisco at 100mbits/sec, but forcing either side,
or both sides to 1000-base-t causes the network status to be set to no
carrier.  Worse, it appears that seting the media to anything other than
auto or 100baseTX or 100baseTX mediaopt full-duplex doesn't work.
        notes from the FreeBSD Intel E1000 driver hint that this might be one 
of the
problematic chips which fails to establish link with certain link partners,
a sentiment echoed by their notes for the Windows drivers as well.
        To make sure I've covered my bases, I booted a monolithic kernel from
the snapshot at nyftp.netbsd.org dated: 201111220010Z with the same
results.  All works fine if all I need is 100mbits/sec, but if I want to
set the media to anything else, no dice, including 10baseT.
        Does anyone know what the specific bug is in this hardware to cause
the dire notes in Intel's README files?  I spent a good deal of time
looking at and playing with the if_wm(4) driver in an effort to understand
what it's doing, since I did almost this same thing for 1000-base-SX if_wm
cards about 5 years ago, but I'm having a lot of trouble understanding the
relationship between the mii layer, the phy driver and the if_wm driver
itself.  After pouring over the e1000 driver in FreeBSD-8, I'm not
convinced we're doing the right things with the phy, but I'm not sure
enough of the hardware to know if it's a driver bug or a hardware bug.  As
an extra data point, the 1000-base-T mode isn't completely broken in that I
have one site using this chip negotiating 1000-base-T full-duplex with a
Force10 switch.  Again, if we try to nail it, it doesn't work, but autoneg
does the right thing in this instance.

        Does anyone have any ideas, documents i should read, or patches
floating around that I should try?  I can file a pr, and I will, but I
thought I'd ask here first.
        Oh, yes, this is under I386, both NetBSD-4 as of early 2008, and the
NetBSD-current snapshot noted above.

-thanks
-Brian


wm0 at pci0 dev 5 function 0: Intel i82541GI 1000BASE-T Ethernet, rev. 5
wm0: interrupting at ioapic0 pin 16 (irq 5)
wm0: 32-bit 66MHz PCI bus
wm0: 64 word (6 address bits) MicroWire EEPROM
wm0: Ethernet address 00:30:18:xx:xx:xx
igphy0 at wm0 phy 1: Intel IGP01E1000 Gigabit PHY, rev. 0
igphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT,
1000baseT-FDX
, auto