> Hi, kernel hackers.
>
> And vm hackers please listen.
Ears open... =)
> Linux is expected to keep nr_free_pages > reserved_pages. Right? So,
> if system is idle, and you wait some time and then ask for atomic
> memory, you'll get some. Right?
To a point... If the system runs out of memory (physical) it will no
longer be able to allocate memory for buffer heads... hence it will be
unable to swap. Thankfully Bill Hawes (hope that's right :) has some
patches which reserve a number of buffer heads for async io - we'll no
longer need to allocate memory to swap; yippie!
...
> seriously broken in there... There was enough swap, so that memory
> _should_ get freed. 100% reproducible in here...
>
> Pavel
Yes kswapd as currently implemented has some problems... Mark Hemment
and I currently have a bit of a friendly competition going ;) for an
improved implementation. It should be interesting to see what we come up
with!
-ben