Re: TCP and reordering

[David Woodhouse <dwmw2@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 1045 bytes --]

On Wed, 2012-11-28 at 08:09 -0800, Eric Dumazet wrote:
> My point was that if you limit number of in flight packet to 1,
> its relatively easy to add glue in the priv dev data, so that you chain
> the destructor without adding yet another fields in all skbs.

Hm, true. I do think we'll need to be able to have at least two packets
in-flight though. There's a fair amount of network stack to be navigated
by a skb once we release it, in the PPPoE and especially L2TP cases. The
latency involved in that, if we only allow one packet at a time, will
surely be noticeable. You're basically guaranteeing that a 'TX done' IRQ
won't have another packet in the device's queue ready to send, and you
won't be able to saturate the uplink.

I suppose your principle applies even for more than one skb at a time;
we can have an array of {skb, old_destructor} tuples in our per-channel
private state, rather than having to keep it with the skb itself... but
that's slightly less simple.

I'll go and knock something up...

-- 
dwmw2


[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 6171 bytes --]