Re: [PATCH] net/skbuff: silence warnings under memory pressure

[Eric Dumazet]

On 9/5/19 4:09 PM, Qian Cai wrote:

> Instead of repeatedly make generalize statements, could you enlighten me with
> some concrete examples that have the similar properties which would trigger a
> livelock,
> 
> - guaranteed GFP_ATOMIC allocations when processing softirq batches.
> - the allocation has a fallback mechanism that is unnecessary to warn a failure.
> 
> I thought "skb" is a special-case here as every packet sent or received is
> handled using this data structure.
>

Just  'git grep GFP_ATOMIC -- net' and carefully study all the places.

You will discover many allocations done for incoming packets.

All of them can fail and trigger a trace.

Please fix the problem for good, do not pretend addressing the skb allocations
will solve it.

The skb allocation can succeed, then the following allocation might fail.

skb are one of the many objects that networking need to allocate dynamically.