Re: kernel 3.2.27 on arm: WARNING: at mm/page_alloc.c:2109 __alloc_pages_nodemask+0x1d4/0x68c()

[Maxime Bizon <mbizon@freebox.fr>]

On Thu, 2012-10-04 at 18:50 +0200, Eric Dumazet wrote:

> > Since skb_recycle() resets skb->data using (skb->head + NET_SKB_PAD), a
> > recycled skb going multiple times through a path that needs to expand
> > skb head will get bigger and bigger each time, and you eventually end up
> > with an allocation failure.
> > 
> 
> Because there is not enough headroom ?

yes, on ipv6 forward path the default NET_SKB_PAD is too small, so each
packet forwarded has its headroom expanded, it is then recycled and gets
its original default headroom back, then it gets forwarded,
expanded, ...

> > An idea to fix this would be to pass needed skb size to skb_resize() and
> > set skb->data to MIN(NET_SKB_PAD, (skb->end - skb->head - skb_size) / 2)
> 
> I am trying to decode this but I cant ;)
> 
> What is skb_resize() ?
> and what do you mean setting skb->data to MIN(NET_SKB_PAD, (skb->end -
> skb->head - skb_size) / 2)
> 
> Care to explain again your idea ?

damn typo, I meant skb_recycle(), and my formula is probably wrong.

skb_size is passed to skb_recycle_check() to ensure the skb is at least
that big.

The idea is it to pass the same value to skb_recycle(), allowing it to
set skb->data somewhat at the middle of skb head space (after honoring
NET_SKB_PAD), that way the recycled skb won't have its head expanded
again if it takes the same path.

-- 
Maxime