On Sun, 22 Aug 2010, Pekka Enberg wrote:Looking at the stack trace of the oops, I think Stan has CONFIG_SLAB which doesn't have order-0 fallback.
In Stan's case, it's a order-1 GFP_ATOMIC allocation but there areThis is the TCP slab? Best fix would be in the page allocator. However,
only order-0 pages available. Mel, any recent page allocator fixes in
2.6.35 or 2.6.36-rc1 that Stan/Mikael should test?
in this particular case the slub allocator would be able to fall back to
an order 0 allocation and still satisfy the request.