Re: [PATCH] __alloc_pages cleanup -R6 Was: Re: Memory Problem in 2.4.10-pre2 / __alloc_pages failed

[Russell King <rmk@arm.linux.org.uk>]

On Fri, Aug 31, 2001 at 01:53:24AM +0200, Roger Larsson wrote:
> Some ideas implemented in this code:
> * Reserve memory below min for atomic and recursive allocations.
> * When being min..low on free pages, free one more than you want to allocate.
> * When being low..high on free pages, free one less than wanted.
> * When above high - don't free anything.
> * First select zones with more than high free memory.
> * Then those with more than high 'free + inactive_clean - inactive_target'
> * When freeing - do it properly. Don't steal direct reclaimed pages

Hmm, I wonder.

I have a 1MB DMA zone, and 31MB of normal memory.

The machine has been running lots of programs for some time, but not under
any VM pressure.  I now come to open a device which requires 64K in 8K
pages from the DMA zone.  What happens?

I suspect that the chances of it failing will be significantly higher with
this algorithm - do you have any thoughts for this?

I don't think we should purely select the allocation zone based purely on
how much free it contains, but also if it's special (like the DMA zone).

You can't clean in-use slab pages out on demand like you can for fs
cache/user pages.

--
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html