debian 2.0, keyboard conflict with PCMCIA

[jesse@spine.com]

I've successfully installed debian 2.0 on my thinkpad 560e, but I'm
still getting frequent crashes.  It boots and loads fine, and is
network accessible, but seconds after I start typing on the keyboard
it unloads the pcmcia card and reloads it.  This behavior ONLY happens 
if I type on the machine's keyboard, I can telnet in from elsewhere
and type all I like.

I've tried this with two different cards, a 3com 3c589 and a motorola
mariner, each exhibits the same behavior, and both work on a similar
laptop in slackware (and on the failing laptop in windows).

My log/messages file gets a whole series of these:

Aug 30 19:57:49 ankle kernel: loading device 'eth0'...
Aug 30 19:57:49 ankle kernel: eth0: 3Com 3c589, port 0x300, irq 3,
     Auto port, hw_addr 00:60:97:41:81:78
Aug 30 19:57:49 ankle kernel: eth0: switched to 10baseT port

And eventually it crashes the machine.

Spread out in the log is also a random sampling of other errors, very
inconsistent:

Aug 30 15:01:31 ankle kernel: eth0: 91c92 hardware Tx buffer allocation failed, status 0xff. 
Aug 30 15:02:03 ankle kernel: eth0: SMC91c92 transmit timed out, Tx_status ff status ffff. 
Aug 30 15:02:03 ankle kernel: eth0: 91c92 hardware Tx buffer allocation failed, status 0xff. 
Aug 30 15:02:12 ankle kernel: smc91c92_cs: Unable to find hardware address. 
Aug 30 15:02:15 ankle kernel: smc91c92_cs: using 8-bit IO window. 
Aug 30 15:02:15 ankle kernel: smc91c92_cs: Yikes!  Bad chip signature! 
Aug 30 15:02:15 ankle kernel: smc91c92_cs: Yikes!  Bad chip signature! 
Aug 30 15:02:15 ankle kernel: eth0: 91c92 hardware Tx buffer allocation failed, status 0xff. 
Aug 30 15:02:45 ankle last message repeated 4 times
Aug 30 15:03:00 ankle last message repeated 3 times
Aug 30 16:05:34 ankle kernel: memory_cs: mem0: anonymous: unknown size 

But in spite of these ominous looking errors, the ethernet stays up
fine (I was able to install debian via FTP!) as long as I don't type
on the keyboard.

The errors above are pretty rare, mostly it just cycles through the
unloading/loading messages without any errors.   

I've tried excluding the IRQ/port, but that had no effect.
/proc/interrupts and /proc/ioports look okay anyway:

# cat /proc/interrupts
 0:    3960057   timer
 1:          4   keyboard
 2:          0   cascade
 3:       3119   3c589_cs
11:          1   i82365
12:         44   PS/2 Mouse
13:          1   math error
14:     210238 + ide0
# cat /proc/ioports
0000-001f : dma1
0020-003f : pic1
0040-005f : timer
0060-006f : keyboard
0080-008f : dma page reg
00a0-00bf : pic2
00c0-00df : dma2
00f0-00ff : npu
01f0-01f7 : ide0
0300-030f : 3c589_cs
03c0-03df : vga+
03e0-03e1 : i82365
03f0-03f5 : floppy
03f6-03f6 : ide0
03f7-03f7 : floppy DIR
03f8-03ff : serial(set)

(for completeness) 
 
# uname -a
Linux ankle 2.0.34 #13 Sun Aug 30 21:17:55 PDT 1998 i586 unknown

# lsmod
Module         Pages    Used by
psaux              1            1 (autoclean)
3c589_cs           2            1
ds                 2    [3c589_cs]      3
i82365             4            2
pcmcia_core        8    [3c589_cs ds i82365]    4

I've read everything I can find, and the only person I found with the
same error gave up on his machine.  Does anyone have any ideas?