The memory problems still exist in pre-2.0.31.
I found a good way to reproduce it reliably:
I'm booting a machine from floppy with mem=4M.
Now, linuxrc starts two shells on consoles 1 and 2. I activate the swap
partition manually. Then I try to create a filesystem on a 1GB-Partition.
When it comes to "Writing inode tables:", the kernel says at some time
"Couldn't get a free page..." and the mke2fs will stop with this:
Writing inode tables: 29/ 128
Now if I switch to the other console and start some program (for example
"ls"), in some cases the mke2fs continues and successfully creates the
filesystem. In some other cases, it is impossible to start the ls-command,
so the system hangs completely.
mke2fs does not always stop at the same inode table. With 4MB, I had cases
where it would go up to 91/128, and with 5MB RAM it stopped once at
21/128. The problem is the same with 2.0.30 and pre-2.0.31. We even could
observe it on machines with 32MB RAM, but of course not reproducable. But
mke2fs seems to be a good candidate to force the problems even on big
machines. David's patch to "try_to_free_page" was never triggered.
Additionally, we are getting "Couldn't get a free page..." on almost any
machine running 2.0.30. Most of the machines have 64MB RAM or more and
sufficient swap.
Hubert mantel@suse.de