On Sat, Aug 12, 2006 at 01:40:29PM +0200, Peter Zijlstra (a.p.zijlstra@chello.nl) wrote:

On Sat, 2006-08-12 at 14:42 +0400, Evgeniy Polyakov wrote:

quoted

When network uses the same allocator, it depends on it, and thus it is
possible to have (cut by you) a situation when reserve (which depends on
SLAB and it's OOM too) is not filled or even does not exist.

No, the reserve does not depend on SLAB, and I totally short-circuit the
SLAB allocator for skbs and related on memory pressure.

The memalloc reserve is on the page allocator level and is only
accessable for PF_MEMALLOC processes or __GFP_MEMALLOC (new in my
patches) allocations. (arguably there could be some more deadlocks wrt.
PF_MEMALLOC where the code under PF_MEMALLOC is not properly bounded,
those would be bugs and should be fixed if present/found)

By SLAB I always mean allocator which is used to get the memory.
In your design main allocator is used to reserve the memory, which then
can be used by your own allocator.

quoted

If transferred to your implementation, then just steal some pages from
SLAB when new network device is added and use them when OOM happens.
It is much simpler and can help in the most of situations.

SLAB reclaim is painfull and has been tried by the time you OOM.

Just never return reserved pages, provide kernel parameter of how large
your reserve is and get pages from there when you detected OOM.
SLAB (main allocator) can do anything it want, but your reserve will never 
be touched by it.

-- 
	Evgeniy Polyakov