On Tue, May 13, 2014 at 4:18 AM, Ben Pfaff wrote:
> On Sun, May 11, 2014 at 09:39:32PM +0300, Alexandru Copot wrote:
>> On Sat, May 10, 2014 at 5:40 AM, Ben Pfaff wrote:
>> > Signed-off-by: Ben Pfaff
>>
>> >
>> > diff --git a/lib/ofp-msgs.h b/lib/ofp-msgs.h
>> > index df7569e..b548f6b 100644
>>
The terminology we used for subtable ordering ('splice', 'right
before') was inherited from an earlier use of a linked list, and
turned out to be confusing when applied to an array. Also, we only
ever move one subtable earlier or later within the array, so we can
simplify the code as well.
Signed
On Mon, May 12, 2014 at 08:52:52AM -0700, Ben Pfaff wrote:
> On Mon, May 12, 2014 at 08:56:36AM +0900, Simon Horman wrote:
> > Some concern has been raised by Ben Pfaff that atomic_uint64_t may not
> > be portable. Accordingly, use atomic_ulong instead of atomic_uint64_t
> > in struct ofproto.
> >
On Mon, May 12, 2014 at 04:32:29PM -0700, Daniele Di Proietto wrote:
> You’re right, of course.
>
> Sorry I didn’t see it the first time.
>
> On May 12, 2014, at 4:28 PM, Jesse Gross wrote:
>
> > On Mon, May 12, 2014 at 4:06 PM, Daniele Di Proietto
> > wrote:
> >> If I understand correctly, th
On 13 May 2014 16:44, Andy Zhou wrote:
>
> > diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
> > index 926f3d6..bb16e7b 100644
> > --- a/ofproto/ofproto-dpif.c
> > +++ b/ofproto/ofproto-dpif.c
> > @@ -3211,10 +3211,11 @@ rule_dpif_lookup(struct ofproto_dpif *ofproto,
> struct flow *f
On Mon, May 12, 2014 at 4:20 PM, Joe Stringer wrote:
> The userspace and kernel datapaths previously differed on their
> treatment of the recirc_id and dp_hash fields when sending upcalls.
> While the kernel datapath would always serialise these fields, the
> userspace would not. When using the us
Their is no use case of partial mask in the user space.
Now that both user space and kernel data path always export recirc_id
and dp_hash, may be we can drop the else if (is_mask) case
for recirc_id?
On Mon, May 12, 2014 at 4:00 PM, Joe Stringer wrote:
> On 25 March 2014 14:58, Andy Zhou wrote:
Shift datapath flow test macros and "ofproto-dpif - dummy interface" out
into a separate file.
Signed-off-by: Joe Stringer
---
tests/automake.mk |1 +
tests/dpif-netdev.at | 53 +
tests/ofproto-dpif.at | 51
Ensure that upcall key matches flow install and flow_dump for userspace
datapath. This was previously assumed, but not tested. This makes the
assumption more explicit.
Signed-off-by: Joe Stringer
---
v2: Shift to dpif-netdev.at.
Depends on "dpif-netdev: Always serialise recirculation in upcall k
Signed-off-by: Joe Stringer
---
tests/ofproto-dpif.at |3 ---
1 file changed, 3 deletions(-)
diff --git a/tests/ofproto-dpif.at b/tests/ofproto-dpif.at
index 3757318..dd5f4fd 100644
--- a/tests/ofproto-dpif.at
+++ b/tests/ofproto-dpif.at
@@ -2,9 +2,6 @@ AT_BANNER([ofproto-dpif])
# Strips
Thanks for the review, I plan to push these soon (and backport this patch
to 2.2)
On 13 May 2014 13:02, Ben Pfaff wrote:
> On Tue, May 13, 2014 at 11:20:31AM +1200, Joe Stringer wrote:
> > The userspace and kernel datapaths previously differed on their
> > treatment of the recirc_id and dp_hash
> On Sat, May 10, 2014 at 08:17:16AM -0700, Ben Pfaff wrote:
>> On Sat, May 10, 2014 at 02:17:26PM +0900, YAMAMOTO Takashi wrote:
>> > btw, --enable-Werror CC=clang doesn't work.
>> > (clang complains about -g for linking)
>> > is it only for me? a workaround is LDFLAGS=-Qunused-argument.
>
> Thi
I think it's more related to dpif-netdev, although these tests point out
failures that may affect many of the megaflow tests in ofproto-dpif.
I'll resend this patch with the tests shifted to the new file.
On 13 May 2014 13:03, Ben Pfaff wrote:
> On Tue, May 13, 2014 at 11:20:32AM +1200, Joe St
On Sat, May 10, 2014 at 08:17:16AM -0700, Ben Pfaff wrote:
> On Sat, May 10, 2014 at 02:17:26PM +0900, YAMAMOTO Takashi wrote:
> > btw, --enable-Werror CC=clang doesn't work.
> > (clang complains about -g for linking)
> > is it only for me? a workaround is LDFLAGS=-Qunused-argument.
Thinking abou
> On Tue, May 13, 2014 at 09:42:02AM +1200, Joe Stringer wrote:
>> Is there any general principles behind fixing these? I'm wondering if
>> there's a way I can prevent these in future.
probably the most typical case is to wait for upcall handler
after netdev-dummy/recieve.
another typical one is
On Sun, May 11, 2014 at 09:39:32PM +0300, Alexandru Copot wrote:
> On Sat, May 10, 2014 at 5:40 AM, Ben Pfaff wrote:
> > Signed-off-by: Ben Pfaff
>
> >
> > diff --git a/lib/ofp-msgs.h b/lib/ofp-msgs.h
> > index df7569e..b548f6b 100644
> > --- a/lib/ofp-msgs.h
> > +++ b/lib/ofp-msgs.h
> > @@ -162
On Sun, May 11, 2014 at 09:39:32PM +0300, Alexandru Copot wrote:
> On Sat, May 10, 2014 at 5:40 AM, Ben Pfaff wrote:
> > Signed-off-by: Ben Pfaff
>
> >
> > diff --git a/lib/ofp-msgs.h b/lib/ofp-msgs.h
> > index df7569e..b548f6b 100644
> > --- a/lib/ofp-msgs.h
> > +++ b/lib/ofp-msgs.h
> > @@ -162
On Tue, May 13, 2014 at 09:42:02AM +1200, Joe Stringer wrote:
> Is there any general principles behind fixing these? I'm wondering if
> there's a way I can prevent these in future.
If we come up with a general principle like that, maybe we can come up
with a systematic way to fix the races without
On Tue, May 13, 2014 at 11:20:32AM +1200, Joe Stringer wrote:
> Ensure that upcall key matches flow install and flow_dump for userspace
> datapath. This was previously assumed, but not tested. This makes the
> assumption more explicit.
>
> Signed-off-by: Joe Stringer
Do you think that this is a
On Tue, May 13, 2014 at 11:20:31AM +1200, Joe Stringer wrote:
> The userspace and kernel datapaths previously differed on their
> treatment of the recirc_id and dp_hash fields when sending upcalls.
> While the kernel datapath would always serialise these fields, the
> userspace would not. When usin
On Tue, May 13, 2014 at 11:20:30AM +1200, Joe Stringer wrote:
> The comment was very specific to one user of the function, but the
> effect of the case is wider.
Fixes a spelling error too.
Acked-by: Ben Pfaff
___
dev mailing list
dev@openvswitch.org
h
On Mon, May 12, 2014 at 12:47:47PM -0700, Daniele Di Proietto wrote:
> On some systems libintel_dpdk.a fails to link with libopenvswitch
> unless -ldl is used. This should address the issue
>
> Signed-off-by: Daniele Di Proietto
Hmm.
I have never noticed "unset" used in this way in Autoconf tes
On Sun, May 11, 2014 at 11:55:01PM -0700, Jarno Rajahalme wrote:
> Array splicing was broken when multiple elements were being moved,
> resulting in the priority order being mixed. This came up when the
> highest priority rule from a subtable was removed and the subtable
> needed to be moved down
You’re right, of course.
Sorry I didn’t see it the first time.
On May 12, 2014, at 4:28 PM, Jesse Gross wrote:
> On Mon, May 12, 2014 at 4:06 PM, Daniele Di Proietto
> wrote:
>> If I understand correctly, the new code frees the skb only if
>> execute_recirc() _returns_ an error and this can h
On Mon, May 12, 2014 at 4:06 PM, Daniele Di Proietto
wrote:
> If I understand correctly, the new code frees the skb only if
> execute_recirc() _returns_ an error and this can happen only if
> ovs_flow_extract() returns an error, in which case recirc_skb only gets freed
> once, but maybe I’m wro
Ensure that upcall key matches flow install and flow_dump for userspace
datapath. This was previously assumed, but not tested. This makes the
assumption more explicit.
Signed-off-by: Joe Stringer
---
tests/ofproto-dpif.at | 78 +
1 file changed,
The comment was very specific to one user of the function, but the
effect of the case is wider.
Signed-off-by: Joe Stringer
---
lib/odp-util.c |2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/odp-util.c b/lib/odp-util.c
index af464a0..34ac5cd 100644
--- a/lib/odp-util.c
The userspace and kernel datapaths previously differed on their
treatment of the recirc_id and dp_hash fields when sending upcalls.
While the kernel datapath would always serialise these fields, the
userspace would not. When using the userspace datapath, this would cause
a mismatch between the odp
On Sat, May 10, 2014 at 7:46 PM, Simon Horman wrote:
> diff --git a/datapath/actions.c b/datapath/actions.c
> index 56d..4dc7c3e 100644
> --- a/datapath/actions.c
> +++ b/datapath/actions.c
> @@ -456,8 +462,27 @@ static int sample(struct datapath *dp, struct sk_buff
> *skb,
>
If I understand correctly, the new code frees the skb only if execute_recirc()
_returns_ an error and this can happen only if ovs_flow_extract() returns an
error, in which case recirc_skb only gets freed once, but maybe I’m wrong.
I agree that it probably makes sense to have execute_recirc() tak
On 25 March 2014 14:58, Andy Zhou wrote:
>
> /* Metadata. */
> +if (present_attrs & (UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID)) {
> +flow->recirc_id = nl_attr_get_u32(attrs[OVS_KEY_ATTR_RECIRC_ID]);
> +expected_attrs |= UINT64_C(1) << OVS_KEY_ATTR_RECIRC_ID;
> +} else if
On Sat, May 10, 2014 at 7:46 PM, Simon Horman wrote:
> It is my understanding that on error execute_recirc() does not free the
> skb passed to it.
>
> Assuming this is true then on error skb should always be freed
> if an error occurs in execute_recirc().
>
> And if recirc_skb differs from skb, be
Is there any general principles behind fixing these? I'm wondering if
there's a way I can prevent these in future.
For example, always sleep after sending packets if no flow exists in the
datapath.
___
dev mailing list
dev@openvswitch.org
http://openvswi
> On Mon, May 12, 2014 at 11:49:03PM +0900, YAMAMOTO Takashi wrote:
>> These races were exposed on my environment by commit c2a77f33.
>> ("tests/ofproto-dpif: Use vlog to test dpif behaviour.")
>>
>> Signed-off-by: YAMAMOTO Takashi
>
> Acked-by: Ben Pfaff
applied the set, thanks
YAMAMOTO Taka
On 5/10/14, 12:48 AM, Jesse Gross wrote:
On Fri, Apr 25, 2014 at 2:55 AM, Lori Jakab wrote:
On 4/25/14, 5:19 AM, Jesse Gross wrote:
+noethernet:
if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
output->eth.type))
goto nla_put_failure;
Does it still make sense to
Signed-off-by: Daniele Di Proietto
---
utilities/ovs-dev.py | 19 +++
1 file changed, 19 insertions(+)
diff --git a/utilities/ovs-dev.py b/utilities/ovs-dev.py
index 7a4425f..814a9d0 100755
--- a/utilities/ovs-dev.py
+++ b/utilities/ovs-dev.py
@@ -224,6 +224,12 @@ def run():
On some systems libintel_dpdk.a fails to link with libopenvswitch
unless -ldl is used. This should address the issue
Signed-off-by: Daniele Di Proietto
---
acinclude.m4 | 27 ---
1 file changed, 24 insertions(+), 3 deletions(-)
diff --git a/acinclude.m4 b/acinclude.m4
in
I’ll make use of your code and reimplement the function using AC_SEARCH_LIBS.
I’l post another patch soon.
Thank you for your help
Daniele
On May 9, 2014, at 5:23 PM, Ben Pfaff wrote:
> On Fri, May 09, 2014 at 04:56:57PM -0700, Daniele Di Proietto wrote:
>> libintel_dpdk.a refuses to link with
Without this patch, the match passed into to
ofproto_dpif_add_internal_flow() are modified. The mask of dl_type will
always be converted from wildcarded match into exact match due to
calling rule_dpif_lookup_in_table(). The fix makes sure
ofproto_dpif_add_internal_flow() does not change the origina
Acked-by: Daniele Di Proietto
On May 10, 2014, at 7:46 PM, Simon Horman wrote:
>
> v2
> * As suggested by Jesse Gross
> - Do not add attributes to noisily break compatibility with user-space
> - (Try to) streamline cloning of skbs by sample and recirculate actions
> ---
> datapath/actions.c |
Acked-by: Daniele Di Proietto
On May 10, 2014, at 7:46 PM, Simon Horman wrote:
> Signed-off-by: Simon Horman
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On May 12, 2014, at 8:48 AM, Ben Pfaff wrote:
> On Mon, May 12, 2014 at 08:43:10AM -0700, Jarno Rajahalme wrote:
>>
>> On May 12, 2014, at 8:36 AM, Ben Pfaff wrote:
>>
>>> On Sun, May 11, 2014 at 11:55:01PM -0700, Jarno Rajahalme wrote:
Array splicing was broken when multiple elements we
On Mon, May 12, 2014 at 08:56:36AM +0900, Simon Horman wrote:
> Some concern has been raised by Ben Pfaff that atomic_uint64_t may not
> be portable. Accordingly, use atomic_ulong instead of atomic_uint64_t
> in struct ofproto.
>
> This is in preparation for removing atomic_uint64_t entirely.
>
>
On Mon, May 12, 2014 at 08:56:37AM +0900, Simon Horman wrote:
> Some concern has been raised by Ben Pfaff that atomic_uint64_t may not
> be portable. In particular on 32bit platforms that do not have atomic
> 64bit integers.
>
> Now that there are no longer any users of atomic_uint64_t remove it
>
On Mon, May 12, 2014 at 08:43:10AM -0700, Jarno Rajahalme wrote:
>
> On May 12, 2014, at 8:36 AM, Ben Pfaff wrote:
>
> > On Sun, May 11, 2014 at 11:55:01PM -0700, Jarno Rajahalme wrote:
> >> Array splicing was broken when multiple elements were being moved,
> >> resulting in the priority order b
On Sun, May 11, 2014 at 10:05:07PM +0300, Rami Rosen wrote:
> This trivial patch fixes a typo in lib/vconn-provider.h.
>
> Signed-off-by: Rami Rosen
Applied, thanks!
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On May 12, 2014, at 8:36 AM, Ben Pfaff wrote:
> On Sun, May 11, 2014 at 11:55:01PM -0700, Jarno Rajahalme wrote:
>> Array splicing was broken when multiple elements were being moved,
>> resulting in the priority order being mixed. This came up when the
>> highest priority rule from a subtable w
On Mon, May 12, 2014 at 11:49:09PM +0900, YAMAMOTO Takashi wrote:
> Signed-off-by: YAMAMOTO Takashi
Acked-by: Ben Pfaff
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On Mon, May 12, 2014 at 11:49:08PM +0900, YAMAMOTO Takashi wrote:
> Signed-off-by: YAMAMOTO Takashi
Acked-by: Ben Pfaff
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On Mon, May 12, 2014 at 11:49:07PM +0900, YAMAMOTO Takashi wrote:
> Simplify code and update comments after commit 61e7deb1.
> ("dpif-netdev: Use RCU to protect data.")
>
> Signed-off-by: YAMAMOTO Takashi
Code gets simpler. Awesome!
Acked-by: Ben Pfaff
On Mon, May 12, 2014 at 11:49:05PM +0900, YAMAMOTO Takashi wrote:
> Signed-off-by: YAMAMOTO Takashi
Acked-by: Ben Pfaff
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On Mon, May 12, 2014 at 11:49:06PM +0900, YAMAMOTO Takashi wrote:
> Signed-off-by: YAMAMOTO Takashi
Acked-by: Ben Pfaff
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
On Mon, May 12, 2014 at 11:49:03PM +0900, YAMAMOTO Takashi wrote:
> These races were exposed on my environment by commit c2a77f33.
> ("tests/ofproto-dpif: Use vlog to test dpif behaviour.")
>
> Signed-off-by: YAMAMOTO Takashi
Acked-by: Ben Pfaff
___
d
On Mon, May 12, 2014 at 11:49:04PM +0900, YAMAMOTO Takashi wrote:
> Leftover of commit 0c473314.
> ("ofproto-dpif.at: Wait for the monitor's pidfile disappears where necessary")
>
> Signed-off-by: YAMAMOTO Takashi
Acked-by: Ben Pfaff
___
dev mailing l
On Sun, May 11, 2014 at 11:55:01PM -0700, Jarno Rajahalme wrote:
> Array splicing was broken when multiple elements were being moved,
> resulting in the priority order being mixed. This came up when the
> highest priority rule from a subtable was removed and the subtable
> needed to be moved down
On Sun, May 11, 2014 at 12:12:32AM +, Wilfried Klaebe wrote:
> net: get rid of SET_ETHTOOL_OPS
>
> Dave Miller mentioned he'd like to see SET_ETHTOOL_OPS gone.
> This does that.
>
> Mostly done via coccinelle script:
> @@
> struct ethtool_ops *ops;
> struct net_device *dev;
> @@
> - SET
These races were exposed on my environment by commit c2a77f33.
("tests/ofproto-dpif: Use vlog to test dpif behaviour.")
Signed-off-by: YAMAMOTO Takashi
---
tests/ofproto-dpif.at | 12 +++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/tests/ofproto-dpif.at b/tests/ofproto
Simplify code and update comments after commit 61e7deb1.
("dpif-netdev: Use RCU to protect data.")
Signed-off-by: YAMAMOTO Takashi
---
lib/dpif-netdev.c | 37 -
1 file changed, 8 insertions(+), 29 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.
Signed-off-by: YAMAMOTO Takashi
---
tests/ofproto-dpif.at | 1 +
1 file changed, 1 insertion(+)
diff --git a/tests/ofproto-dpif.at b/tests/ofproto-dpif.at
index 84453d8..4dcaf63 100644
--- a/tests/ofproto-dpif.at
+++ b/tests/ofproto-dpif.at
@@ -3817,6 +3817,7 @@ done
AT_CHECK([ovs-appctl time/w
Signed-off-by: YAMAMOTO Takashi
---
tests/ofproto-dpif.at | 6 --
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/tests/ofproto-dpif.at b/tests/ofproto-dpif.at
index 14293a5..84453d8 100644
--- a/tests/ofproto-dpif.at
+++ b/tests/ofproto-dpif.at
@@ -3677,15 +3677,17 @@ OVS_VSWIT
this is a repost of trivial patches which failed to get reviewed
in a few weeks but still are valid.
what's the recommended procedure for unreviewed patches?
- drop
- push (i suspect this is the right thing to do for trivial ones like these)
- repost (this is what i'm doing)
YAMAMOTO Takashi
YAM
Signed-off-by: YAMAMOTO Takashi
---
ofproto/ofproto-dpif.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/ofproto/ofproto-dpif.h b/ofproto/ofproto-dpif.h
index d4ad624..679a41e 100644
--- a/ofproto/ofproto-dpif.h
+++ b/ofproto/ofproto-dpif.h
@@ -81,8 +81,8 @@ extern struc
Signed-off-by: YAMAMOTO Takashi
---
ofproto/ofproto-dpif.c | 8
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index e50b4fe..926f3d6 100644
--- a/ofproto/ofproto-dpif.c
+++ b/ofproto/ofproto-dpif.c
@@ -1241,13 +1241,13 @@ ad
Leftover of commit 0c473314.
("ofproto-dpif.at: Wait for the monitor's pidfile disappears where necessary")
Signed-off-by: YAMAMOTO Takashi
---
tests/ofproto-dpif.at | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tests/ofproto-dpif.at b/tests/ofproto-dpif.at
index b2be768..1
This patch implements vxlan-soe:
http://tools.ietf.org/html/draft-zhou-li-vxlan-soe-01
Tested VXLAN throughput between two hypervisors, and the performance
gain of vxlan-soe is significant.
netperf TCP_STREAM test result:
Before the change: 2.62 Gbits/sec
After the change: 6.68 Gbits/sec
Spee
65 matches
Mail list logo