The only user is get_dpifindex(), no need to redirect via the port
operations.
Signed-off-by: Thomas Graf
---
net/openvswitch/datapath.c | 7 ---
net/openvswitch/vport-internal_dev.c | 1 -
net/openvswitch/vport-netdev.c | 7 ---
net/openvswitch/vport-netdev.h | 1
The only user is get_dpifindex(), no need to redirect via the port
operations.
Signed-off-by: Thomas Graf
---
V2 - Fix stylistic issue
net/openvswitch/datapath.c | 3 ++-
net/openvswitch/vport-internal_dev.c | 1 -
net/openvswitch/vport-netdev.c | 7 ---
net/openvswitch
Posting this to netdev due to dependency on first patch.
Thomas Graf (2):
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Use zerocopy if applicable when performing the upcall
include/linux/skbuff.h | 2 ++
net/core/skbuff.c| 46
From: Thomas Graf
Avoids a memcpy() that is expensive for large packets:
from:
4.51% [kernel] [k] memcpy
to:
1.20% [kernel] [k] memcpy
1.01% [kernel] [k] skb_zerocopy
Signed-off-by: Thomas Graf
---
net/openvswitch
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
---
include/linux/skbuff.h | 2 ++
net/core/skbuff.c| 46 +++
net/netfilter/nfnetlink_queue_core.c | 47
On 05/24/13 at 08:31am, Eric Dumazet wrote:
> Hmm. Big warning here : In the nfnetlink case, we specifically made sure
> that this path was taken only if skb->head_frag was set (all incoming
> packets have this property, but not outgoing ones yet)
>
> This should be documented if we move it to net
On 05/24/13 at 03:18pm, Jesse Gross wrote:
> On Fri, May 24, 2013 at 2:57 PM, Eric Dumazet wrote:
> >> My guess is that there isn't a real different for small packets since
> >> everything will be in the cache but it seems worth checking given that
> >> this is optimizing a rare case at the expens
On 05/27/13 at 10:28am, Jesse Gross wrote:
> On Saturday, May 25, 2013, Eric Dumazet wrote:
>
> > On Sat, 2013-05-25 at 08:02 +0100, Thomas Graf wrote:
> >
> > > I ran TCP_CRR to verify the SYN/ACK use case and I did not
> > > observe a difference. If you ha
Jesse,
On 06/29/2013 12:44 AM, Jesse Gross wrote:
Pravin pointed out that the RHEL6 openvswitch_handle_frame_hook
unregister is happening from the RCU callback, which means that we
immediately free the vport after removing the handler. This seems
racy. I also noticed that we have asymmetric use
Thomas Graf (3):
datapath: Rename skb_network_protocol() to __skb_network_protocol()
datapath: rhel: Account for RHEL specific backports
datapath: rhel: Move RHEL OVS hook registration to
netdev_rx_handler_register() backport
acinclude.m4| 3
is present which was removed under RTNL
protection before the RCU callback.
Signed-off-by: Thomas Graf
---
datapath/linux/compat/include/linux/netdevice.h | 21 -
datapath/vport-netdev.c | 17 +
2 files changed, 21 insertions(+), 17
The function skb_network_protocol() is already defined in
upstream but not an exported symbol.
Rename the OVS internal implementation to work around this.
Signed-off-by: Thomas Graf
---
datapath/linux/compat/gso.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a
The following symbols have been backported to RHEL and the kernel
version is no longer an accurate indicator for their presence:
- skb_gso_segment
- netif_skb_features
- netif_needs_gso
Signed-off-by: Thomas Graf
---
acinclude.m4| 3 +++
datapath/linux
On 07/09/2013 06:38 PM, Pravin Shelar wrote:
On Tue, Jul 9, 2013 at 9:00 AM, Thomas Graf wrote:
The function skb_network_protocol() is already defined in
upstream but not an exported symbol.
Rename the OVS internal implementation to work around this.
Are you seeing problem with RHEL kernel
On 07/09/2013 08:06 PM, Jesse Gross wrote:
On Tue, Jul 9, 2013 at 9:00 AM, Thomas Graf wrote:
diff --git a/datapath/vport-netdev.c b/datapath/vport-netdev.c
index fe7e359..d043eae 100644
--- a/datapath/vport-netdev.c
+++ b/datapath/vport-netdev.c
@@ -46,9 +46,8 @@ MODULE_PARM_DESC(vlan_tso
is present which was removed under RTNL
protection before the RCU callback.
Signed-off-by: Thomas Graf
---
V2: - Move nr_bridges to compat/netdevice.c
- Don't use atomic_t for nr_bridges
datapath/linux/compat/include/linux/netdevice.h | 21 -
datapath/linux/c
On 07/15/2013 09:54 PM, Jesse Gross wrote:
On Mon, Jul 15, 2013 at 12:45 PM, Pravin B Shelar wrote:
This reverts commit 752378e1cd1f133a8366fbacec3b281a45ff8268
(datapath: rhel: Account for RHEL specific backports).
Change related to netif_needs_gso() is cuasing panic on RHEL
and Xen platforms.
for CHECKSUM_PARTIAL tx case
Thomas Graf (2):
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Use skb_zerocopy() to prepare skb for upcall
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter/nfnetlink_queue_core.c | 59
From: Thomas Graf
Use of skb_zerocopy() avoids the expensive call to memcpy() when
copying the packet data into the Netlink skb. Completes checksum
through skb_checksum_help() if needed (typicall packet input from
software device) which invalidates some of the gains again.
Stock-RX
- 38.30
On 07/25/13 at 06:39pm, Jesse Gross wrote:
> On Thu, Jul 25, 2013 at 5:43 AM, Thomas Graf wrote:
> > From: Thomas Graf
> >
> > Use of skb_zerocopy() avoids the expensive call to memcpy() when
> > copying the packet data into the Netlink skb. Completes checksum
> &g
rx_handler_data rather than rx_handler. This was
> introduced by commit "3e35fe3 datapath: rhel: Move RHEL OVS hook
> registration to netdev_rx_handler_register() backport".
>
> Reported-by: Andrei Andone
> Signed-off-by: Chris Wright
> Cc: T
Signed-off-by: Thomas Graf
---
ovsdb/ovsdb-server.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/ovsdb/ovsdb-server.c b/ovsdb/ovsdb-server.c
index ab44b3a..f64122e 100644
--- a/ovsdb/ovsdb-server.c
+++ b/ovsdb/ovsdb-server.c
@@ -983,6 +983,7 @@ ovsdb_server_compact(struct unixctl_conn
disabled)
Signed-off-by: Thomas Graf
---
net/openvswitch/datapath.c | 52 +++---
1 file changed, 45 insertions(+), 7 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index 1408adc..3f170e3 100644
--- a/net/openvswitch
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter/nfnetlink_queue_core.c | 59
ist case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (2):
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Use skb_zerocopy() for upcall
include/linux/skbuff.h | 3 ++
net/core/skbuff.c|
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter/nfnetlink_queue_core.c | 59
disabled)
Signed-off-by: Thomas Graf
Cc: Jesse Gross
Cc: Eric Dumazet
---
net/openvswitch/datapath.c | 52 +++---
1 file changed, 45 insertions(+), 7 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index 1408adc
calculate headroom of destination
buffer. This also takes care of the from->head_frag issue.
- Attribute alignment for frag_list case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (2):
net: Export skb_zerocopy() to zerocopy from one skb
On 11/09/13 at 10:02pm, Ben Hutchings wrote:
> On Fri, 2013-11-08 at 10:15 +0100, Thomas Graf wrote:
> > Use of skb_zerocopy() avoids the expensive call to memcpy() when
> > copying the packet data into the Netlink skb. Completes checksum
> > through skb_checksum_help() if nee
There is no reason to enforce padding after the last attribute.
Dropping this enforcement will ease efforts to implement zerocopy
upcall.
Signed-off-by: Thomas Graf
---
lib/netlink.h | 14 +++---
1 file changed, 11 insertions(+), 3 deletions(-)
diff --git a/lib/netlink.h b/lib
Following commit (''netlink: Do not enforce alignment of last Netlink
attribute''), signal the ability to receive unaligned Netlink messages
to the datapath to enable utilization of zerocopy optimizations.
Signed-off-by: Thomas Graf
---
include/linux/openvswitch.h | 4 ++
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter/nfnetlink_queue_core.c | 59
sue.
- Attribute alignment for frag_list case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (3):
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Allow user space to announce ability to accept unaligned
Netlink messa
Signed-off-by: Thomas Graf
---
include/uapi/linux/openvswitch.h | 4
net/openvswitch/datapath.c | 4
net/openvswitch/datapath.h | 2 ++
3 files changed, 10 insertions(+)
diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
index d120f9f..07ef2c3
+ 2.81% vhost-8471 [k] csum_partial_copy_generic
to:
+ 5.72% ovs-vswitchd [k] memcpy
+ 3.32% vhost-5153 [k] memcpy
+ 0.68% vhost-5153 [k] skb_zerocopy
(megaflows disabled)
Signed-off-by: Thomas Graf
---
net/openvswitch/datapath.c | 54
On 11/11/2013 04:50 PM, Ben Pfaff wrote:
On Mon, Nov 11, 2013 at 04:36:24PM +0100, Thomas Graf wrote:
Following commit (''netlink: Do not enforce alignment of last Netlink
attribute''), signal the ability to receive unaligned Netlink messages
to the datapath to enable uti
On 11/13/2013 07:11 AM, Jesse Gross wrote:
On Mon, Nov 11, 2013 at 11:53 PM, Thomas Graf wrote:
On 11/11/2013 04:50 PM, Ben Pfaff wrote:
On Mon, Nov 11, 2013 at 04:36:24PM +0100, Thomas Graf wrote:
Following commit (''netlink: Do not enforce alignment of last Netlink
attribute&
On 11/15/2013 10:32 AM, Jesse Gross wrote:
On Wed, Nov 13, 2013 at 5:46 PM, Thomas Graf wrote:
On 11/13/2013 07:11 AM, Jesse Gross wrote:
On Mon, Nov 11, 2013 at 11:53 PM, Thomas Graf wrote:
On 11/11/2013 04:50 PM, Ben Pfaff wrote:
On Mon, Nov 11, 2013 at 04:36:24PM +0100, Thomas Graf
Signed-off-by: Thomas Graf
---
lib/dpif-linux.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c
index 42958d3..0442f77 100644
--- a/lib/dpif-linux.c
+++ b/lib/dpif-linux.c
@@ -72,7 +72,7 @@ struct dpif_linux_dp {
/* Attributes
ned-off-by: Lorand Jakab
Reviewed-by: Thomas Graf
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
;v2:
Add more content to the commit message.
Reviewed-by: Thomas Graf
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On 11/19/2013 10:50 PM, Pravin B Shelar wrote:
Some distribution kernel has backported skb->rxhash (e.g. RHEL)
Following patch allows use precalculated rxhash.
Signed-off-by: Pravin B Shelar
Reviewed-by: Thomas Graf
___
dev mailing list
On 11/19/2013 10:50 PM, Pravin B Shelar wrote:
Rather than using complete flow hash, we can use skb->rxhash for
calculating source port. Similar calculation is done by vxlan.
Signed-off-by: Pravin B Shelar
---
datapath/vport-lisp.c |9 +++--
1 files changed, 3 insertions(+), 6 deleti
: Pravin B Shelar
Reviewed-by: Thomas Graf
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On 11/20/2013 10:14 AM, Jesse Gross wrote:
On Wed, Nov 20, 2013 at 1:08 AM, Thomas Graf wrote:
On 11/19/2013 10:50 PM, Pravin B Shelar wrote:
Rather than using complete flow hash, we can use skb->rxhash for
calculating source port. Similar calculation is done by vxlan.
Signed-off-by: Pra
On 11/21/2013 01:40 AM, Jesse Gross wrote:
On Wed, Nov 20, 2013 at 1:43 AM, Thomas Graf wrote:
On 11/20/2013 10:14 AM, Jesse Gross wrote:
On Wed, Nov 20, 2013 at 1:08 AM, Thomas Graf wrote:
I might be missing something but what about the rxhash == 0 case?
VXLAN does:
hash
On 11/21/13 at 06:23pm, Ben Hutchings wrote:
> On Thu, 2013-11-21 at 19:13 +0100, Thomas Graf wrote:
> > +
> > +/**
> > + * V2:
>
> This is not kernel-doc format so don't use '/**'.
I was hoping kernel-doc would pick it up but it doesn't.
I&#x
% vhost-5153 [k] skb_zerocopy
(megaflows disabled)
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/openvswitch/datapath.c | 54 +++---
1 file changed, 37 insertions(+), 17 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net
Consolidates ovs_dp_cmd_new() and ovs_dp_cmd_set() to simplify
handling and avoid code duplication.
Allows user space to specify NLM_F_REPLACE with OVS_DP_CMD_NEW and
overwrite the settings such as the user features of an existing
datapath.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel
Allocates a new sk_buff large enough to cover the specified payload
plus required Netlink headers. Will check receiving socket for
memory mapped i/o capability and use it if enabled. Will fall back
to non-mapped skb if message size exceeds the frame size of the ring.
Signed-of-by: Thomas Graf
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 4
net/openvswitch/datapath.c | 14 ++
net/openvswitch/datapath.h | 2 ++
3 files changed, 20 insertions(+)
diff --git a/include/uapi/linux/openvswitch.h b/include/uapi
ist case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (8):
genl: Add genlmsg_new_unicast() for unicast message allocation
netlink: Avoid netlink mmap alloc if msg size exceeds frame size
openvswitch: Enable memory mapped Netlink i/o
net: Export skb_zeroc
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter
An insufficent ring frame size configuration can lead to an
unnecessary skb allocation for every Netlink message. Check frame
size before taking the queue lock and allocating the skb and
re-check with lock to be safe.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/netlink
cpy
+ 1.09% kpktgend_1 [k] memcpy
+ 1.04% kpktgend_3 [k] memcpy
+ 0.96% ovs-vswitchd [k] copy_user_generic_string
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/openvswitch/datapath.c | 56 --
1 f
Drop user features if an outdated user space instance that does not
understand the concept of user_features attempted to create a new
datapath.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 11 ++-
net/openvswitch/datapath.c | 10
Following commit (''netlink: Do not enforce alignment of last Netlink
attribute''), signal the ability to receive unaligned Netlink messages
to the datapath to enable utilization of zerocopy optimizations.
Signed-off-by: Thomas Graf
---
include/linux/openvswitch.h | 15 +
hus avoids copying. The functionality is enabled on
sockets used for unicast traffic.
Further optimizations are possible by avoiding the copy into the
ofpbuf after reading.
Cc: Cong Wang
Signed-off-by: Thomas Graf
---
AUTHORS| 1 +
lib/dpif-linux.c | 6 +-
lib/netd
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter
tination
buffer. This also takes care of the from->head_frag issue.
- Attribute alignment for frag_list case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (8):
genl: Add genlmsg_new_unicast() for unicast message allocation
netlink: Avoid netlink mm
Allocates a new sk_buff large enough to cover the specified payload
plus required Netlink headers. Will check receiving socket for
memory mapped i/o capability and use it if enabled. Will fall back
to non-mapped skb if message size exceeds the frame size of the ring.
Signed-of-by: Thomas Graf
An insufficent ring frame size configuration can lead to an
unnecessary skb allocation for every Netlink message. Check frame
size before taking the queue lock and allocating the skb and
re-check with lock to be safe.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/netlink
Consolidates ovs_dp_cmd_new() and ovs_dp_cmd_set() to simplify
handling and avoid code duplication.
Allows user space to specify NLM_F_REPLACE with OVS_DP_CMD_NEW and
overwrite the settings such as the user features of an existing
datapath.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 4
net/openvswitch/datapath.c | 14 ++
net/openvswitch/datapath.h | 2 ++
3 files changed, 20 insertions(+)
diff --git a/include/uapi/linux/openvswitch.h b/include/uapi
Drop user features if an outdated user space instance that does not
understand the concept of user_features attempted to create a new
datapath.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 10 +-
net/openvswitch/datapath.c | 9
% vhost-5153 [k] skb_zerocopy
(megaflows disabled)
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/openvswitch/datapath.c | 54 +++---
1 file changed, 37 insertions(+), 17 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net
cpy
+ 1.09% kpktgend_1 [k] memcpy
+ 1.04% kpktgend_3 [k] memcpy
+ 0.96% ovs-vswitchd [k] copy_user_generic_string
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/openvswitch/datapath.c | 56 --
1 f
evice refcnt since netdev private data (dev1->ax25_ptr) is
NULL.
---
Therefore I have pushed his patch to branch-1.10.
The following is also needed:
commit 0d1d26db42b5d206ebb75cc6dc93507f4e5b6a8a
Author: Chris Wright
Date: Tue Aug 13 18:02:48 2013 -0700
On Tue, Jul 9, 2013 at 11:36
On 11/25/13 at 01:23pm, Jesse Gross wrote:
> On Fri, Nov 22, 2013 at 8:56 AM, Thomas Graf wrote:
> > diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
> > index 95d4424..3f1fb87 100644
> > --- a/net/openvswitch/datapath.c
> > +++ b/net/openvswitch/d
On 11/25/13 at 03:40pm, Jesse Gross wrote:
> On Fri, Nov 22, 2013 at 8:56 AM, Thomas Graf wrote:
> > diff --git a/include/uapi/linux/openvswitch.h
> > b/include/uapi/linux/openvswitch.h
> > index 07ef2c3..a6863e2 100644
> > --- a/include/uapi/linux/openvswitch.h
&
On 11/26/2013 12:41 AM, Jesse Gross wrote:
On Thu, Nov 21, 2013 at 10:15 AM, Thomas Graf wrote:
diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c
index 1dcf321..5c952bc 100644
--- a/lib/dpif-linux.c
+++ b/lib/dpif-linux.c
@@ -232,6 +233,7 @@ dpif_linux_open(const struct dpif_class *class
On 11/26/2013 12:02 AM, Ben Pfaff wrote:
On Thu, Nov 21, 2013 at 07:16:54PM +0100, Thomas Graf wrote:
Based on the initial patch by Cong Wang posted a couple of months
ago.
This is the user space counterpart needed for the kernel patch
'[PATCH net-next 3/8] openvswitch: Enable memory m
On 11/28/2013 08:24 PM, Ben Pfaff wrote:
On Wed, Nov 27, 2013 at 11:51:20PM +0100, Thomas Graf wrote:
On 11/26/2013 12:02 AM, Ben Pfaff wrote:
On Thu, Nov 21, 2013 at 07:16:54PM +0100, Thomas Graf wrote:
Based on the initial patch by Cong Wang posted a couple of months
ago.
This is the user
Allocates a new sk_buff large enough to cover the specified payload
plus required Netlink headers. Will check receiving socket for
memory mapped i/o capability and use it if enabled. Will fall back
to non-mapped skb if message size exceeds the frame size of the ring.
Signed-of-by: Thomas Graf
An insufficent ring frame size configuration can lead to an
unnecessary skb allocation for every Netlink message. Check frame
size before taking the queue lock and allocating the skb and
re-check with lock to be safe.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/netlink
Make the skb zerocopy logic written for nfnetlink queue available for
use by other modules.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/linux/skbuff.h | 3 ++
net/core/skbuff.c| 85
net/netfilter
% vhost-5153 [k] skb_zerocopy
(megaflows disabled)
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/openvswitch/datapath.c | 54 +++---
1 file changed, 37 insertions(+), 17 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net
Drop user features if an outdated user space instance that does not
understand the concept of user_features attempted to create a new
datapath.
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 10 +-
net/openvswitch/datapath.c | 21
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 4
net/openvswitch/datapath.c | 14 ++
net/openvswitch/datapath.h | 2 ++
3 files changed, 20 insertions(+)
diff --git a/include/uapi/linux/openvswitch.h b/include/uapi
g_list case
- API documentation
- performance data for CHECKSUM_PARTIAL tx case
Thomas Graf (7):
genl: Add genlmsg_new_unicast() for unicast message allocation
netlink: Avoid netlink mmap alloc if msg size exceeds frame size
openvswitch: Enable memory mapped Netlink i/o
net: Exp
cpy
+ 1.09% kpktgend_1 [k] memcpy
+ 1.04% kpktgend_3 [k] memcpy
+ 0.96% ovs-vswitchd [k] copy_user_generic_string
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
net/openvswitch/datapath.c | 56 --
1 f
Following commit (''netlink: Do not enforce alignment of last Netlink
attribute''), signal the ability to receive unaligned Netlink messages
to the datapath to enable utilization of zerocopy optimizations.
Signed-off-by: Thomas Graf
---
V2: - Only provide OVS_DP_A
hus avoids copying. The functionality is enabled on
sockets used for unicast traffic.
Further optimizations are possible by avoiding the copy into the
ofpbuf after reading.
Signed-off-by: Thomas Graf
---
V2: - Provide required definitions in netlink-protocol.h if
does not contain them.
On 11/30/13 at 01:35pm, Florian Westphal wrote:
> Thomas Graf wrote:
> > Benchmark
> > * pktgen -> ovs internal port
> > * 5M pkts, 5M flows
> > * 4 threads, 8 cores
> >
> > Before:
> > Result: OK: 67418743(c67108212+d310530) usec, 500
On 12/02/2013 10:23 PM, Ben Pfaff wrote:
On Sat, Nov 30, 2013 at 01:27:33PM +0100, Thomas Graf wrote:
Based on the initial patch by Cong Wang posted a couple of months
ago.
This is the user space counterpart needed for the kernel patch
'[PATCH net-next 3/8] openvswitch: Enable memory m
Gross
Reviewed-by: Thomas Graf
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
hus avoids copying. The functionality is enabled on
sockets used for unicast traffic.
Further optimizations are possible by avoiding the copy into the
ofpbuf after reading.
Signed-off-by: Thomas Graf
---
V3: - Provide __ALIGN_KERNEL in case is not available
- Silence Clang alignment problem fals
On 12/04/2013 02:54 AM, Zhuangyuxin wrote:
> In this patch, each frame takes 16k mem. For sockets used for upcall, does it
> make more sense to calculate frame size based on MTU of the netdev on the
> port?
I assume you mean max(MTU) over all ports? We would have to replace the
ring after each M
On 12/04/2013 05:33 PM, Ben Pfaff wrote:
If I'm doing the calculations correctly, this mmaps 8 MB per ring-based
Netlink socket on a system with 4 kB pages. OVS currently creates one
Netlink socket for each datapath port. With 1000 ports (a moderate
number; we sometimes test with more), that is
On 12/04/2013 07:08 PM, Ben Pfaff wrote:
On Wed, Dec 04, 2013 at 06:20:53PM +0100, Thomas Graf wrote:
How about we limit the number of mmaped sockets to a configurable
maximum that defaults to 16 or 32?
Maybe you mean that we should only mmap some of the sockets that we
create. If so, this
On 12/03/13 at 09:43pm, Jesse Gross wrote:
Thanks for merging a first set of patches
> On Sat, Nov 30, 2013 at 4:21 AM, Thomas Graf wrote:
> > diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
> > index 8eaa39a..867edf1 100644
> > --- a/net/openvswitch
On 12/04/2013 11:20 PM, Jesse Gross wrote:
If enabling rings on demand can be done cleanly that might be best
solution. To me, it seems difficult to generalize the upcall
characteristics based on port type.
It would require to reopen sockets but I don't see that as a major
obstacle.
16K ports
On 12/10/2013 08:36 PM, David Miller wrote:
From: Jesse Gross
Date: Tue, 10 Dec 2013 11:28:08 -0800
I think this is definitely a good optimization to do given that so
much of the work that OVS does is hashing. However, isn't there a
library where there would be a more appropriate place to put
Series implementing a zerocopy method for OVS upcall messages. Based
on top of commit: (''openvswitch: Enable memory mapped Netlink i/o'')
Thomas Graf (6):
net: Export skb_zerocopy() to zerocopy from one skb to another
openvswitch: Allow user space to announce ability
Drop user features if an outdated user space instance that does not
understand the concept of user_features attempted to create a new
datapath.
Signed-off-by: Thomas Graf
---
include/uapi/linux/openvswitch.h | 10 +-
net/openvswitch/datapath.c | 21 +
2 files
The copy & csum optimization is no longer present with zerocopy
enabled. Compute the checksum in skb_gso_segment() directly by
dropping the HW CSUM capability from the features passed in.
Signed-off-by: Thomas Graf
---
net/openvswitch/datapath.c | 2 +-
1 file changed, 1 insertion(+
Signed-off-by: Thomas Graf
Reviewed-by: Daniel Borkmann
---
include/uapi/linux/openvswitch.h | 4
net/openvswitch/datapath.c | 14 ++
net/openvswitch/datapath.h | 2 ++
3 files changed, 20 insertions(+)
diff --git a/include/uapi/linux/openvswitch.h b/include/uapi
% vhost-5153 [k] memcpy
+ 0.68% vhost-5153 [k] skb_zerocopy
(megaflows disabled)
Signed-off-by: Thomas Graf
---
net/openvswitch/datapath.c | 37 +
1 file changed, 29 insertions(+), 8 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net/openvswitch
Allows removing the net and dp_ifindex argument and simplify the
code.
Signed-off-by: Thomas Graf
---
net/openvswitch/datapath.c | 34 ++
1 file changed, 14 insertions(+), 20 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index
901 - 1000 of 1023 matches
Mail list logo
