I have been trying 2.5.35 and found a few problems with the stock 2.5.35.
(But it is so nice to be starting from the stock version)
ubd_kern.c has the problem with the 2.5 model where if more than the first
device has a partition the partition tables become confused. This occurs
when the new partition layer allocs the table and inserts it into the
gendisk structure which it it assumes is per device (not driver) and
overwrites any earlier data. so if ubd0 is partitioned it works but if ubd1+
is in use things go rapidly ill. This patch adds the extra
ubd_gendisk/fake_gendisk structures and drops out the 2.4 partition stuff.
arch/um/kernel/Makefile did not get the obj-y += user_syms.o in 2.5.35 so
modules have problems if they user any of the user side symbols.
__do_copy_to_user escaped being exported :(
The for_each_task I converted to do_each_thread...while_each_thread like
i386 does this may be wrong as my system stalls at killall on shutdown.
for_each_process may be needed instead.
I found one include mixup where percpu.h was useing the same ifndef flag as
cacheflush.h this really confused me for a while as I could not figure out
how I was not getting the defines when I could see the include right there.
pgtable.h should now define pte_addr_t it appears
timex.h has gained a couple of defines... there may be a better way to do
somehow the merge missed the do_mounts.c entries for ubd, This makes it
very hard to have a good root= :)