David Miller a écrit :

From: Eric Dumazet <redacted>
Date: Sat, 01 Mar 2008 11:26:17 +0100

quoted

You are absolutly right. We should guard against recursion, using a new field 
in "pcpu_lstats" (cheap access in a hot cache line as we have to update stats 
anyway)

 ...

quoted

[PATCH] loopback: calls netif_receive_skb() instead of netif_rx()

I'm willing to seriously entertain this change and stick it
into net-2.6.26 if you will perform a reasonable deep stack
test.

For example, create an XFS filesystem, and mount it NFS over
loopback.  Then stress it like crazy.

See if this generates stack overflows or weird crashes.

Fair enough :)

I'll do my best to stress it on various situations, with 4K stacks on i386.

Thank you