So, i have had a weird phenomenon. On a first-gen (i.e. no USB, etc)
miata, 433au model, in both 1.4.x and 1.5.x, i have noticed that if there
is more than a gig of core in the system, it is rather flaky. It can use
all the memory, no problem, but at least in this system, the problem shows
up in the de driver, which will reliably panic the machine if one brings
the interface up and down twice, even over wuite a span of time, and will 
lose use of the interface after a while (roughly 30-45 minutes, it seems
that heavier use breaks it sooner) without a panic, but an attempt to
reset the interface from e.g. the console will panic it immediately.
Removing the memory above 1g stops the problem. It is not a matter of bad
memory, I've tried 4 different sets of dimms in many different
combinations of slots, and that doesn't change the 1g memory size being
the cause of the problem. 

When the interface just locks up, the kernel keeps printing this message: 

de1: unable to load tx map, error = 22



and the output from the most recent panic: 

fatal kernel trap:

    trap entry = 0x2 (memory management fault)
    a0         = 0x68
    a1         = 0x1
    a2         = 0x0
    pc         = 0xfffffc0000480b1c
    ra         = 0xfffffc0000480b50
    curproc    = 0xfffffc000b0fb408
        pid = 225, comm = ifconfig

panic: trap
syncing disks... 8 8 7 7 7 6 6 5 5 4 4 4 4 4 3 3 3 3 3 2 2 2 1 1 1 done

dump to dev 8,25 not possible
rebooting...

(it did not drop to ddb, it went straight to a reboot)
any ideas? is more than a gig of core just a no go on this machine? 
(it doesnt matter if the ethernet is in a 32bit or 64 bit slot, either,
ive moved it all around with no change in behavior)
(it is also reproducible on a second 433au of similar vintage)
the ethernet is a basic DE500, of course (actual dec make, not a clone)

happy hacking,
Isildur