Re: [RFC][PATCH 2/9] deadlock prevention core

[David Miller]

From: Andrew Morton <akpm@xxxxxxxx>
Date: Mon, 14 Aug 2006 00:07:36 -0700

> What is a "socket wait queue" and how/why can it consume so much memory?
> 
> Can it be prevented from doing that?
>
> If this refers to the socket buffers, they're mostly allocated with
> at least __GFP_WAIT, aren't they?

He's talking about the fact that, once we've tied a receive buffer to
a socket, we can't liberate that memory in any way until the user
reads in the data (which can be whenever it likes) especially if we've
ACK'd the data for protocols such as TCP.

Receive buffers are allocated by the device, usually in interrupt of
software interrupt context, to refill it's RX ring using GFP_ATOMIC or
similar.

Send buffers are usually allocated with GFP_KERNEL, but that can be
modified via the sk->sk_allocation socket member.  This is used by
things like sunrpc and other cases which need to allocate socket
buffers with GFP_ATOMIC or with GFP_NOFS for NFS's sake.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/