So I finally reconnected my O2 to the network and installed a new kernel
after Tsutsui-san merged Chris' latest changes and figured I'd see how
well a userland build went (which also served to keep the machine a bit
busy and see how stable it was).

When running off the local disk, everything goes well except (a) it's
awfully slow without the hack of ahc_pci.c (this was built from a ~
clean tree, so it didn't have the hack to force the card to ignore
current device settings), and (b) the disk is now too small to host
src, obj, new destdir and the stuff installed.

No big deal, I've got a lot of disk available on the network, so I
restarted the build from scratch over NFS.  I've had the assembler
die with a SIGSEGV twice (I think in the same place... building in
gnu/lib/libgcc right after make includes -- out of the do-gnu-lib-
libgcc target, I'd guess -- compiling __dummy.c).  Looking back
at the build to local disk, it actually completed everything and
ran out of space installing stuff to the new tempdir, so the sources
are in a sane state (in both cases, I'm building from the same tree
on the local disk -- it's just all the output, including new tools
which are on NFS in this case).

Anybody seen issues building over NFS?  The kernel running is from
10/9 and from ident output does have Tsutsui-san's bus.c changes;
the NIC is a 3c905C-TXM and I *am* doing IP/TCP/UDP checksums on
the card (say I with a surprised tone :-).  My kernel does not
have COMPAT_16 (I only thought about that this morning and did not
get a chance to install a new kernel), but I doubt that makes a
difference given the fact that it all worked when output was on
local disk...

Thanks!
--rafal

----
Rafal Boni                                                     rafal@pobox.com
  We are all worms.  But I do believe I am a glowworm.  -- Winston Churchill