* ip_summed setting for TCP pure-ACK packets
@ 2010-11-08 16:35 Ben Hutchings
2010-11-08 16:56 ` Eric Dumazet
0 siblings, 1 reply; 4+ messages in thread
From: Ben Hutchings @ 2010-11-08 16:35 UTC (permalink / raw)
To: David Miller; +Cc: sf-linux-drivers, netdev
As we discussed at LPC:
Current controllers handled by the sfc driver have a per-queue (rather
than per-packet) option for checksum generation. Currently pure-ACK
packets sent by TCP have ip_summed == CHECKSUM_NONE and we must send
them on hardware queues with checksum generation disabled. To support
this, we allocate 2 hardware queues per core TX queue.
To reduce the risk of reordering (and possibly the number of hardware TX
queues required), it would be helpful for TCP to set ip_summed ==
CHECKSUM_PARTIAL on pure-ACK packets when the output device is known to
support checksum generation.
Ben.
--
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: ip_summed setting for TCP pure-ACK packets
2010-11-08 16:35 ip_summed setting for TCP pure-ACK packets Ben Hutchings
@ 2010-11-08 16:56 ` Eric Dumazet
2010-11-09 16:24 ` Ben Hutchings
0 siblings, 1 reply; 4+ messages in thread
From: Eric Dumazet @ 2010-11-08 16:56 UTC (permalink / raw)
To: Ben Hutchings; +Cc: David Miller, sf-linux-drivers, netdev
Le lundi 08 novembre 2010 à 16:35 +0000, Ben Hutchings a écrit :
> As we discussed at LPC:
>
> Current controllers handled by the sfc driver have a per-queue (rather
> than per-packet) option for checksum generation. Currently pure-ACK
> packets sent by TCP have ip_summed == CHECKSUM_NONE and we must send
> them on hardware queues with checksum generation disabled. To support
> this, we allocate 2 hardware queues per core TX queue.
>
> To reduce the risk of reordering (and possibly the number of hardware TX
> queues required), it would be helpful for TCP to set ip_summed ==
> CHECKSUM_PARTIAL on pure-ACK packets when the output device is known to
> support checksum generation.
>
> Ben.
>
Hmm
Do you mean commit 2e8e18ef52e7dd1af0a3bd1 is not enough ?
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: ip_summed setting for TCP pure-ACK packets
2010-11-08 16:56 ` Eric Dumazet
@ 2010-11-09 16:24 ` Ben Hutchings
2010-11-09 16:28 ` Eric Dumazet
0 siblings, 1 reply; 4+ messages in thread
From: Ben Hutchings @ 2010-11-09 16:24 UTC (permalink / raw)
To: Eric Dumazet; +Cc: David Miller, sf-linux-drivers, netdev
On Mon, 2010-11-08 at 17:56 +0100, Eric Dumazet wrote:
> Le lundi 08 novembre 2010 à 16:35 +0000, Ben Hutchings a écrit :
> > As we discussed at LPC:
> >
> > Current controllers handled by the sfc driver have a per-queue (rather
> > than per-packet) option for checksum generation. Currently pure-ACK
> > packets sent by TCP have ip_summed == CHECKSUM_NONE and we must send
> > them on hardware queues with checksum generation disabled. To support
> > this, we allocate 2 hardware queues per core TX queue.
> >
> > To reduce the risk of reordering (and possibly the number of hardware TX
> > queues required), it would be helpful for TCP to set ip_summed ==
> > CHECKSUM_PARTIAL on pure-ACK packets when the output device is known to
> > support checksum generation.
> >
> > Ben.
> >
>
> Hmm
>
> Do you mean commit 2e8e18ef52e7dd1af0a3bd1 is not enough ?
It might well be... I must admit I hadn't thought to check whether this
issue had gone away.
Yes, that does the trick. Sorry for wasting people's time on this.
Ben.
--
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: ip_summed setting for TCP pure-ACK packets
2010-11-09 16:24 ` Ben Hutchings
@ 2010-11-09 16:28 ` Eric Dumazet
0 siblings, 0 replies; 4+ messages in thread
From: Eric Dumazet @ 2010-11-09 16:28 UTC (permalink / raw)
To: Ben Hutchings; +Cc: David Miller, sf-linux-drivers, netdev
Le mardi 09 novembre 2010 à 16:24 +0000, Ben Hutchings a écrit :
> On Mon, 2010-11-08 at 17:56 +0100, Eric Dumazet wrote:
> > Le lundi 08 novembre 2010 à 16:35 +0000, Ben Hutchings a écrit :
> > > As we discussed at LPC:
> > >
> > > Current controllers handled by the sfc driver have a per-queue (rather
> > > than per-packet) option for checksum generation. Currently pure-ACK
> > > packets sent by TCP have ip_summed == CHECKSUM_NONE and we must send
> > > them on hardware queues with checksum generation disabled. To support
> > > this, we allocate 2 hardware queues per core TX queue.
> > >
> > > To reduce the risk of reordering (and possibly the number of hardware TX
> > > queues required), it would be helpful for TCP to set ip_summed ==
> > > CHECKSUM_PARTIAL on pure-ACK packets when the output device is known to
> > > support checksum generation.
> > >
> > > Ben.
> > >
> >
> > Hmm
> >
> > Do you mean commit 2e8e18ef52e7dd1af0a3bd1 is not enough ?
>
> It might well be... I must admit I hadn't thought to check whether this
> issue had gone away.
>
> Yes, that does the trick. Sorry for wasting people's time on this.
>
You're welcome ;)
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2010-11-09 16:28 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2010-11-08 16:35 ip_summed setting for TCP pure-ACK packets Ben Hutchings
2010-11-08 16:56 ` Eric Dumazet
2010-11-09 16:24 ` Ben Hutchings
2010-11-09 16:28 ` Eric Dumazet
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox