Patchwork sctp: add support for busy polling to sctp protocol

login
register
mail settings
Submitter Neil Horman
Date April 17, 2014, 7:26 p.m.
Message ID <1397762811-2989-1-git-send-email-nhorman@tuxdriver.com>
Download mbox | patch
Permalink /patch/340014/
State Accepted
Delegated to: David Miller
Headers show

Comments

Neil Horman - April 17, 2014, 7:26 p.m.
The busy polling socket option adds support for sockets to busy wait on data
arriving on the napi queue from which they have most recently received a frame.
Currently only tcp and udp support this feature, but theres no reason sctp can't
do so as well.  Add it in so appliations can take advantage of it

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Vlad Yasevich <vyasevich@gmail.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Daniel Borkmann <dborkman@redhat.com>
CC: netdev@vger.kernel.org
---
 net/sctp/socket.c   | 5 +++++
 net/sctp/ulpqueue.c | 4 ++++
 2 files changed, 9 insertions(+)
Vlad Yasevich - April 17, 2014, 9:37 p.m.
On 04/17/2014 03:26 PM, Neil Horman wrote:
> The busy polling socket option adds support for sockets to busy wait on data
> arriving on the napi queue from which they have most recently received a frame.
> Currently only tcp and udp support this feature, but theres no reason sctp can't
> do so as well.  Add it in so appliations can take advantage of it
> 
> Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
> CC: Vlad Yasevich <vyasevich@gmail.com>
> CC: "David S. Miller" <davem@davemloft.net>
> CC: Daniel Borkmann <dborkman@redhat.com>
> CC: netdev@vger.kernel.org

Hi Neil

Do we gain anything by polling?
This is probably more appropriate for net-next...

Acked-by: Vlad Yasevich <vyasevich@gmail.com>

Thanks
-vlad

> ---
>  net/sctp/socket.c   | 5 +++++
>  net/sctp/ulpqueue.c | 4 ++++
>  2 files changed, 9 insertions(+)
> 
> diff --git a/net/sctp/socket.c b/net/sctp/socket.c
> index 981aaf8..9788e07 100644
> --- a/net/sctp/socket.c
> +++ b/net/sctp/socket.c
> @@ -71,6 +71,7 @@
>  #include <net/route.h>
>  #include <net/ipv6.h>
>  #include <net/inet_common.h>
> +#include <net/busy_poll.h>
>  
>  #include <linux/socket.h> /* for sa_family_t */
>  #include <linux/export.h>
> @@ -6553,6 +6554,10 @@ static struct sk_buff *sctp_skb_recv_datagram(struct sock *sk, int flags,
>  		if (sk->sk_shutdown & RCV_SHUTDOWN)
>  			break;
>  
> +		if (sk_can_busy_loop(sk) &&
> +		    sk_busy_loop(sk, noblock))
> +			continue;
> +
>  		/* User doesn't want to wait.  */
>  		error = -EAGAIN;
>  		if (!timeo)
> diff --git a/net/sctp/ulpqueue.c b/net/sctp/ulpqueue.c
> index 5dc9411..a4df012 100644
> --- a/net/sctp/ulpqueue.c
> +++ b/net/sctp/ulpqueue.c
> @@ -38,6 +38,7 @@
>  #include <linux/types.h>
>  #include <linux/skbuff.h>
>  #include <net/sock.h>
> +#include <net/busy_poll.h>
>  #include <net/sctp/structs.h>
>  #include <net/sctp/sctp.h>
>  #include <net/sctp/sm.h>
> @@ -204,6 +205,9 @@ int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event)
>  	if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN))
>  		goto out_free;
>  
> +	if (!sctp_ulpevent_is_notification(event))
> +		sk_mark_napi_id(sk, skb);
> +
>  	/* Check if the user wishes to receive this event.  */
>  	if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe))
>  		goto out_free;
> 

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Daniel Borkmann - April 18, 2014, 9:39 a.m.
On 04/17/2014 09:26 PM, Neil Horman wrote:
> The busy polling socket option adds support for sockets to busy wait on data
> arriving on the napi queue from which they have most recently received a frame.
> Currently only tcp and udp support this feature, but theres no reason sctp can't
> do so as well.  Add it in so appliations can take advantage of it
>
> Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
> CC: Vlad Yasevich <vyasevich@gmail.com>
> CC: "David S. Miller" <davem@davemloft.net>
> CC: Daniel Borkmann <dborkman@redhat.com>
> CC: netdev@vger.kernel.org

Acked-by: Daniel Borkmann <dborkman@redhat.com>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Neil Horman - April 18, 2014, 11:53 a.m.
On Thu, Apr 17, 2014 at 05:37:29PM -0400, Vlad Yasevich wrote:
> On 04/17/2014 03:26 PM, Neil Horman wrote:
> > The busy polling socket option adds support for sockets to busy wait on data
> > arriving on the napi queue from which they have most recently received a frame.
> > Currently only tcp and udp support this feature, but theres no reason sctp can't
> > do so as well.  Add it in so appliations can take advantage of it
> > 
> > Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
> > CC: Vlad Yasevich <vyasevich@gmail.com>
> > CC: "David S. Miller" <davem@davemloft.net>
> > CC: Daniel Borkmann <dborkman@redhat.com>
> > CC: netdev@vger.kernel.org
> 
> Hi Neil
> 
> Do we gain anything by polling?
Yes, for high througput applications, we save the latency introduced by
processing the interrupt and waiting for the scheduled napi poll to run.  See
the initial busy poll patch series for intel performance numbers.

Neil

> This is probably more appropriate for net-next...
> 
> Acked-by: Vlad Yasevich <vyasevich@gmail.com>
> 
> Thanks
> -vlad
> 
> > ---
> >  net/sctp/socket.c   | 5 +++++
> >  net/sctp/ulpqueue.c | 4 ++++
> >  2 files changed, 9 insertions(+)
> > 
> > diff --git a/net/sctp/socket.c b/net/sctp/socket.c
> > index 981aaf8..9788e07 100644
> > --- a/net/sctp/socket.c
> > +++ b/net/sctp/socket.c
> > @@ -71,6 +71,7 @@
> >  #include <net/route.h>
> >  #include <net/ipv6.h>
> >  #include <net/inet_common.h>
> > +#include <net/busy_poll.h>
> >  
> >  #include <linux/socket.h> /* for sa_family_t */
> >  #include <linux/export.h>
> > @@ -6553,6 +6554,10 @@ static struct sk_buff *sctp_skb_recv_datagram(struct sock *sk, int flags,
> >  		if (sk->sk_shutdown & RCV_SHUTDOWN)
> >  			break;
> >  
> > +		if (sk_can_busy_loop(sk) &&
> > +		    sk_busy_loop(sk, noblock))
> > +			continue;
> > +
> >  		/* User doesn't want to wait.  */
> >  		error = -EAGAIN;
> >  		if (!timeo)
> > diff --git a/net/sctp/ulpqueue.c b/net/sctp/ulpqueue.c
> > index 5dc9411..a4df012 100644
> > --- a/net/sctp/ulpqueue.c
> > +++ b/net/sctp/ulpqueue.c
> > @@ -38,6 +38,7 @@
> >  #include <linux/types.h>
> >  #include <linux/skbuff.h>
> >  #include <net/sock.h>
> > +#include <net/busy_poll.h>
> >  #include <net/sctp/structs.h>
> >  #include <net/sctp/sctp.h>
> >  #include <net/sctp/sm.h>
> > @@ -204,6 +205,9 @@ int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event)
> >  	if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN))
> >  		goto out_free;
> >  
> > +	if (!sctp_ulpevent_is_notification(event))
> > +		sk_mark_napi_id(sk, skb);
> > +
> >  	/* Check if the user wishes to receive this event.  */
> >  	if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe))
> >  		goto out_free;
> > 
> 
> 
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Miller - April 20, 2014, 10:20 p.m.
From: Neil Horman <nhorman@tuxdriver.com>
Date: Thu, 17 Apr 2014 15:26:51 -0400

> The busy polling socket option adds support for sockets to busy wait on data
> arriving on the napi queue from which they have most recently received a frame.
> Currently only tcp and udp support this feature, but theres no reason sctp can't
> do so as well.  Add it in so appliations can take advantage of it
> 
> Signed-off-by: Neil Horman <nhorman@tuxdriver.com>

Applied.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 981aaf8..9788e07 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -71,6 +71,7 @@ 
 #include <net/route.h>
 #include <net/ipv6.h>
 #include <net/inet_common.h>
+#include <net/busy_poll.h>
 
 #include <linux/socket.h> /* for sa_family_t */
 #include <linux/export.h>
@@ -6553,6 +6554,10 @@  static struct sk_buff *sctp_skb_recv_datagram(struct sock *sk, int flags,
 		if (sk->sk_shutdown & RCV_SHUTDOWN)
 			break;
 
+		if (sk_can_busy_loop(sk) &&
+		    sk_busy_loop(sk, noblock))
+			continue;
+
 		/* User doesn't want to wait.  */
 		error = -EAGAIN;
 		if (!timeo)
diff --git a/net/sctp/ulpqueue.c b/net/sctp/ulpqueue.c
index 5dc9411..a4df012 100644
--- a/net/sctp/ulpqueue.c
+++ b/net/sctp/ulpqueue.c
@@ -38,6 +38,7 @@ 
 #include <linux/types.h>
 #include <linux/skbuff.h>
 #include <net/sock.h>
+#include <net/busy_poll.h>
 #include <net/sctp/structs.h>
 #include <net/sctp/sctp.h>
 #include <net/sctp/sm.h>
@@ -204,6 +205,9 @@  int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event)
 	if (sock_flag(sk, SOCK_DEAD) || (sk->sk_shutdown & RCV_SHUTDOWN))
 		goto out_free;
 
+	if (!sctp_ulpevent_is_notification(event))
+		sk_mark_napi_id(sk, skb);
+
 	/* Check if the user wishes to receive this event.  */
 	if (!sctp_ulpevent_is_enabled(event, &sctp_sk(sk)->subscribe))
 		goto out_free;