> Looks like that pre-2.0.31-9 has no problems with "can not get a free page"
> in low memory situation.
Unfortunately it does have it. On an old i38640/8MB 10meg swap, it still
gives that message when compiling something large or exiting X. It happens
even though MM_DEBUG in mm/vmscan.c has been undefined. On a 32meg P150
nothing like this happens even on a heavily loaded system.
> I have tried both SMP and non SMP kernel booted in single user mode on 8 MB
> RAM. I was able to do mke2fs (120 MB), badblocks -vw (120 MB), bonnie -s 150
> without any sign of problem. The same on 4 MB RAM. System quite responsive
> (max 10-20 s wait, usually 1-5 s).
That's true, even compiling GhostScript with 'make -j -l 7.0' leaves the
system quite useable (higher -l value kills the i386 ;-))