Adjusted the code as suggested, and applied to master~ !
Thx very much for the help and explanation throughout the time,
Alex Wang,
On Mon, Apr 14, 2014 at 10:24 PM, Pravin Shelar wrote:
> On Tue, Apr 15, 2014 at 6:42 AM, Alex Wang wrote:
> > In order to allow handlers directly read upcalls
On April 14, 2014 at 8:42:54 PM, Neelakantam Gaddam (neelugad...@gmail.com)
wrote:
> Can we have control on flow timeout in datapath in the latest 2.1 OVS
> version?
The next release will have a "max-idle" that can increase the amount of time
flows will stay when idle. However, ovs-vswitchd ca
On 15 April 2014 16:34, YAMAMOTO Takashi wrote:
> > +if (ukey) {
> > +ovs_mutex_lock(&ukey->mutex);
> > +ukey->mark = ukey->flow_exists = mark;
> > +ovs_mutex_unlock(&ukey->mutex);
> > }
> >
> > -if (!ukey) {
> > -ukey = uke
On Tue, Apr 15, 2014 at 6:42 AM, Alex Wang wrote:
> In order to allow handlers directly read upcalls from datapath,
> we need to support per-handler netlink socket for each vport in
> datapath. This commit makes this happen. Also, it is guaranteed
> backward and forward compatibility with previo
> +if (ukey) {
> +ovs_mutex_lock(&ukey->mutex);
> +ukey->mark = ukey->flow_exists = mark;
> +ovs_mutex_unlock(&ukey->mutex);
> }
>
> -if (!ukey) {
> -ukey = ukey_create(udump->key, udump->key_len, used);
> -hmap_
On Tue, Apr 15, 2014 at 6:01 AM, Alex Wang wrote:
> Thx Pravin for the review, please see my comments inline,
>
>>
>> > + */
>> > +int ovs_vport_set_upcall_portids(struct vport *vport, struct nlattr
>> > *ids)
>> > +{
>> > + struct vport_portids *old, *vport_portids;
>> > +
>> > + if
Hi Pettit,
Can we have control on flow timeout in datapath in the latest 2.1 OVS
version?
Is there any way to modify the userspace daemon so that datapath flows stay
longer duration?
On Mon, Apr 14, 2014 at 11:21 PM, Justin Pettit wrote:
> idle_timeout and hard_timeout are not related to dat
> Signed-off-by: Jarno Rajahalme
Reviewed-by: YAMAMOTO Takashi
> ---
> lib/odp-util.c |2 +-
> tests/odp.at |1 +
> 2 files changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/lib/odp-util.c b/lib/odp-util.c
> index b58f1c0..10bf925 100644
> --- a/lib/odp-util.c
> +++ b/lib/odp
> @@ -3597,26 +3589,44 @@ commit_odp_tunnel_action(const struct flow *flow,
> struct flow *base,
> static void
> commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
> struct ofpbuf *odp_actions,
> - struct flow_wildca
> Signed-off-by: Jarno Rajahalme
Reviewed-by: YAMAMOTO Takashi
> ---
> ofproto/ofproto-dpif-xlate.c |8 +-
> ofproto/ofproto-dpif-xlate.h |3 ++-
> ofproto/ofproto-dpif.c | 58
> +-
> 3 files changed, 66 insertions(+), 3 deletions(-)
In order to allow handlers directly read upcalls from datapath,
we need to support per-handler netlink socket for each vport in
datapath. This commit makes this happen. Also, it is guaranteed
backward and forward compatibility with previous branch.
Signed-off-by: Alex Wang
Acked-by: Thomas Graf
He ill-starred veteran's joy. I se
Ually the whole five acts from start to finish. He acted the words as he read,
modulating his voice to suit the various characters, stamping and storming, and
to adjust his black skullcap--it _would_ tumble off at the pathetic
parts--dealing himself a success
It is nicer than the old one. We have reserved a spot for you. :-)
On Mon, Apr 14, 2014 at 5:37 PM, Joe Stringer wrote:
> Thanks! Hope you like the new office :)
>
>
> On 15 April 2014 12:32, Andy Zhou wrote:
>>
>> Pushed. Sorry it took a while because of the office move.
>>
>> On Mon, Apr 14,
Thanks! Hope you like the new office :)
On 15 April 2014 12:32, Andy Zhou wrote:
> Pushed. Sorry it took a while because of the office move.
>
> On Mon, Apr 14, 2014 at 1:20 AM, Andy Zhou wrote:
> > Not a problem I was planing to push both patches in the morning (PST_
> > if we can resolve 2n
Pushed. Thanks for the fix!
On Mon, Apr 14, 2014 at 3:36 AM, Joe Stringer wrote:
> From: Joe Stringer
>
> dcf00ba35a0 (ofproto/bond: Implement bond megaflow using recirculation)
> allowed bond_entry statistics to be modified while holding a readlock.
> This patch modifies bond_entry_account() to
Pushed. Sorry it took a while because of the office move.
On Mon, Apr 14, 2014 at 1:20 AM, Andy Zhou wrote:
> Not a problem I was planing to push both patches in the morning (PST_
> if we can resolve 2nd patch soon. At any rate, I will push this patch
> in the morning.
>
> On Mon, Apr 14, 2014
Thx Pravin for the review, please see my comments inline,
> > + */
> > +int ovs_vport_set_upcall_portids(struct vport *vport, struct nlattr
> *ids)
> > +{
> > + struct vport_portids *old, *vport_portids;
> > +
> > + if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
> > +
The latest version is on patchwork:
http://patchwork.openvswitch.org/patch/3628/
On 10 April 2014 04:59, Ben Pfaff wrote:
> On Wed, Apr 09, 2014 at 04:36:50PM +1200, Joe Stringer wrote:
> > Thanks Alex.
> >
> > There is also one minor fix to be rolled in:
> >
> > @@ -1403,8 +1416,8 @@ revalida
On Fri, Apr 11, 2014 at 9:16 AM, Zoltan Kiss wrote:
> This is the ported version of commit 36d5fe6a with the same name from
> net-next.
> Apart from the small datapath.c changes it adjust the compat layer files as
> well. This is the original commit message:
>
> "skb_zerocopy can copy elements of
Supporting 9100, or even 8100 would be nice if they are easy. If we
have to make a choice,
I'd rank 88a8, followed by 9100, then 8100. My understanding is that
9100 was fairly popular
in older networking equipments. Openflow1.1 only requires 88a8.
On Mon, Apr 14, 2014 at 8:05 AM, Thomas F Herbert
It would be great if your next rebased patch also provide some
information on what has been tested.
Some higher level feedback follows, Minor feedback inline.
Patch format:
For ease of up-streaming to Linux kenrel, we should split the patch
into two parts. One
for kernel datapath, one for OVS us
Thanks for fixing this! Pushed to master.
On Sun, Apr 13, 2014 at 5:03 PM, Simon Horman wrote:
> This corrects a typo that was added by adcf00ba35a0ce9c
> ("ofproto/bond: Implement bond megaflow using recirculation").
>
> Cc: Andy Zhou
> Signed-off-by: Simon Horman
> ---
> ofproto/ofproto-dpif
LGTM. pushed to master. Thanks!
On Sun, Apr 13, 2014 at 5:03 PM, Simon Horman wrote:
> This corrects an editing error that was added by 6b83a3c5c37a0711
> ("ofproto: Support OF version-specific table-miss behaviours").
>
> Signed-off-by: Simon Horman
> ---
> ofproto/ofproto-dpif.c | 1 +
> 1 f
> On Apr 14, 2014, at 1:00 PM, Justin Pettit wrote:
>
> It looks like there are a few other places that need to get updated, too.
> Did you want me to take a pass at them?
>
Why not, seems you found them already, so please go ahead :-)
Jarno
> For this, though:
>
> Acked-by: Justin Pett
This should optimize port masks for megaflows for typical port usage
in matches. Each subtable has it's own trie if the subtable matches
any of the ports bits. This trie is consulted only after failing
lookup to determine the number of bits that needs to be unwildcarded
to guarantee that any pack
It looks like there are a few other places that need to get updated, too. Did
you want me to take a pass at them?
For this, though:
Acked-by: Justin Pettit
--Justin
On April 14, 2014 at 10:51:08 AM, Jarno Rajahalme (jrajaha...@nicira.com) wrote:
> We recently renamed ofpbuf's 'l2' member as
On 04/12/2014 12:00 AM, Andy Zhou wrote:
Implements Linux kernel datapath hash action. Hash action computes
hash and stores it into current packet key.
Signed-off-by: Andy Zhou
hash fix
^^ What's this? ;-)
---
datapath/actions.c | 19 +++
datapath/flow.h |
On 04/12/2014 10:09 PM, Thomas F Herbert wrote:
diff --git a/lib/packets.c b/lib/packets.c
index 6244c3f..385da79 100644
--- a/lib/packets.c
+++ b/lib/packets.c
@@ -198,8 +198,9 @@ eth_pop_vlan(struct ofpbuf *packet)
{
struct vlan_eth_header *veh = ofpbuf_l2(packet);
-if (veh && ofpb
On Apr 14, 2014, at 1:14 AM, YAMAMOTO Takashi wrote:
>> Diagnostic code path dereferenced a NULL pointer.
>>
>> Signed-off-by: Jarno Rajahalme
>
> Reviewed-by: YAMAMOTO Takashi
Thanks for your review!
Pushed to master,
Jarno
___
dev mailing l
idle_timeout and hard_timeout are not related to datapath flows; they're
OpenFlow concepts. In 1.10, you can't control how long a flow stays in the
datapath. The userspace daemon manages those flows and tries to find the
optimal time to leave them in place.
--Justin
On April 14, 2014 at 9:3
We recently renamed ofpbuf's 'l2' member as 'frame'.
Signed-off-by: Jarno Rajahalme
---
lib/ofp-actions.h |4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/lib/ofp-actions.h b/lib/ofp-actions.h
index 89bf867..1db3bde 100644
--- a/lib/ofp-actions.h
+++ b/lib/ofp-actions.h
*Дорогой друг, я адвокат ALVIS HILBASIN, адвокат юридической, мне нужна
ваша помощь, чтобы представлять моего покойного клиента, он хранение сумму
$ (US $ 13,580,000.00) Миллион Соединенные долларов в банке, прежде чем он
умер, и директор банке / Руководство посоветовал мне связаться с любым из
его
On 04/14/2014 05:59 PM, Pravin Shelar wrote:
On Mon, Apr 14, 2014 at 8:19 PM, Thomas Graf wrote:
On 04/14/2014 08:54 AM, Pravin Shelar wrote:
On Wed, Apr 9, 2014 at 4:05 PM, Thomas Graf wrote:
I believe skb_get_hash() can be 0 and that would result in mistaking any
empty slot to be a cache
No, that's a job for the buildds.
On Apr 14, 2014 9:24 AM, "Powers, Joshua" wrote:
> Ben,
>
> >Architecture: source i386 all
>
> Will you also be building the amd64 packages?
>
> Thanks,
> Josh
>
___
dev mailing list
dev@openvswitch.org
http://openvswit
Hi All,
I am using openvswitch-1.10 version.
Is there any way to change the values of idle_timeout and hard_timeout for
the datapath flows ?
My requirement is to increase the flow activity period in the kernel
datapath.
Please suggest me the best method to get this done.
--
Thanks & Regards
Ben,
>Architecture: source i386 all
Will you also be building the amd64 packages?
Thanks,
Josh
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
Your message dated Mon, 14 Apr 2014 16:06:17 +
with message-id
and subject line Bug#744769: fixed in openvswitch 2.1.0+git20140411-2
has caused the Debian Bug report #744769,
regarding openvswitch-switch missing directory hangs install
to be marked as done.
This means that you claim that the
Accepted:
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA256
Format: 1.8
Date: Mon, 14 Apr 2014 08:29:33 -0700
Source: openvswitch
Binary: openvswitch-datapath-source openvswitch-datapath-dkms
openvswitch-common openvswitch-switch openvswitch-ipsec openvswitch-pki
openvswitch-dbg python-openvswi
On Mon, Apr 14, 2014 at 8:19 PM, Thomas Graf wrote:
> On 04/14/2014 08:54 AM, Pravin Shelar wrote:
>>
>> On Wed, Apr 9, 2014 at 4:05 PM, Thomas Graf wrote:
>>>
>>> I believe skb_get_hash() can be 0 and that would result in mistaking any
>>> empty slot to be a cached entry and we would only look a
openvswitch_2.1.0+git20140411-2_i386.changes uploaded successfully to localhost
along with the files:
openvswitch_2.1.0+git20140411-2.dsc
openvswitch_2.1.0+git20140411-2.debian.tar.xz
openvswitch-common_2.1.0+git20140411-2_i386.deb
openvswitch-switch_2.1.0+git20140411-2_i386.deb
openvswit
On Apr 14, 2014, at 1:11 AM, yamam...@valinux.co.jp (YAMAMOTO Takashi) wrote:
>> diff --git a/lib/odp-execute.c b/lib/odp-execute.c
>> index 37e44e3..127df85 100644
>> --- a/lib/odp-execute.c
>> +++ b/lib/odp-execute.c
>> @@ -18,6 +18,7 @@
>> #include
>> #include "odp-execute.h"
>> #include
>>
Andy and Manikanta,
I agree with Andy/ I don't think there is a necessity for any design
changes in OVS for qinq at least for Open Flow 1.1. I am working on
debugging/testing the patch now.
Right now I am working on debugging my double tagging patch as specified
in 802.1ad with an outer TPI
Package: openvswitch-switch
Version: 2.1.0+git20140411-1
Openvswitch-switch (amd64) install will hang due to a missing directory. This
occurred on the previous version dated 0325 as well. The user will see the
following output during the install:
Setting up openvswitch-common (2.1.0+git20140411-
On 04/14/2014 08:54 AM, Pravin Shelar wrote:
On Wed, Apr 9, 2014 at 4:05 PM, Thomas Graf wrote:
I believe skb_get_hash() can be 0 and that would result in mistaking any
empty slot to be a cached entry and we would only look at the first flow
mask.
software skb_get_hash() returns zero for erro
On Mon, Apr 14, 2014 at 4:27 AM, Manikanta Srinivas
wrote:
> Hi,
>
> Thanks for your efforts. We are interested in QinQ implementation of
> openvswitch. After going through the patch, we are left with following
> queries.
>
> 1. We think there should be a separate mode to support qinq tunnel. This
On Mon, Apr 14, 2014 at 2:43 AM, Joe Stringer wrote:
> On 14 April 2014 21:04, Andy Zhou wrote:
>>
>> > Looking closer, there are two things that this patch achieves:
>> > (1) Allow statistics to be attributed to two different bonds at the same
>> > time.
>> If this is the goal, should we also co
On 4/14/14, 3:00 PM, thomas.mo...@orange.com wrote:
Hi Lori,
2014-04-14, Lori Jakab:
On 4/14/14, 11:01 AM, thomas.mo...@orange.com wrote:
2014-04-11, Lori Jakab:
On 4/11/14, 4:24 PM, thomas.mo...@orange.com wrote:
The result is the following:
- an option is added to GRE tunnel port, to allow
Hi Lori,
2014-04-14, Lori Jakab:
> On 4/14/14, 11:01 AM, thomas.mo...@orange.com wrote:
>> 2014-04-11, Lori Jakab:
>>> On 4/11/14, 4:24 PM, thomas.mo...@orange.com wrote:
The result is the following:
- an option is added to GRE tunnel port, to allow making them OVS
l3ports
>>> My fi
Hi,
Thanks for your efforts. We are interested in QinQ implementation of
openvswitch. After going through the patch, we are left with following queries.
1. We think there should be a separate mode to support qinq tunnel. This can be
achieved by implementing a new configuration parameter
On 4/14/14, 11:01 AM, thomas.mo...@orange.com wrote:
Hi Lori,
2014-04-11, Lori Jakab:
On 4/11/14, 4:24 PM, thomas.mo...@orange.com wrote:
The result is the following:
- an option is added to GRE tunnel port, to allow making them OVS l3ports
My first impression is that starting with my patches
From: Joe Stringer
dcf00ba35a0 (ofproto/bond: Implement bond megaflow using recirculation)
allowed bond_entry statistics to be modified while holding a readlock.
This patch modifies bond_entry_account() to get a writelock before
modifying the statistics and adds thread-safety annotations to these
On 14 April 2014 21:04, Andy Zhou wrote:
> > Looking closer, there are two things that this patch achieves:
> > (1) Allow statistics to be attributed to two different bonds at the same
> > time.
> If this is the goal, should we also consider each bond have its own
> read/write lock?
>
If reading/
On Mon, Apr 14, 2014 at 1:15 AM, Joe Stringer wrote:
> Thanks for looking over this. I agree with you that simply changing it to
> take a write lock would be simpler, but let me explain my thinking.
That would be simpler bug fix patch. I see that you want to go further.
>
> * Any thread that cal
Not a problem I was planing to push both patches in the morning (PST_
if we can resolve 2nd patch soon. At any rate, I will push this patch
in the morning.
On Mon, Apr 14, 2014 at 12:46 AM, Joe Stringer wrote:
> Thanks. I don't have push access, could you take care of that?
>
>
> On 14 April 201
Thanks for looking over this. I agree with you that simply changing it to
take a write lock would be simpler, but let me explain my thinking.
* Any thread that calls xlate_actions() could account stats for a bond.
This could be handlers or revalidators.
* Handler threads account the first set of s
> Diagnostic code path dereferenced a NULL pointer.
>
> Signed-off-by: Jarno Rajahalme
Reviewed-by: YAMAMOTO Takashi
> ---
> lib/odp-util.c | 16
> 1 file changed, 8 insertions(+), 8 deletions(-)
>
> diff --git a/lib/odp-util.c b/lib/odp-util.c
> index 10bf925..f61c2b2 100
> diff --git a/lib/odp-execute.c b/lib/odp-execute.c
> index 37e44e3..127df85 100644
> --- a/lib/odp-execute.c
> +++ b/lib/odp-execute.c
> @@ -18,6 +18,7 @@
> #include
> #include "odp-execute.h"
> #include
> +#include
> #include
> #include
>
on netbsd-6, arpa/inet.h is required before
Hi Lori,
2014-04-11, Lori Jakab:
> On 4/11/14, 4:24 PM, thomas.mo...@orange.com wrote:
>>
>> The result is the following:
>> - an option is added to GRE tunnel port, to allow making them OVS l3ports
>
> My first impression is that starting with my patches and now yours there
> are a lot of differe
Thanks. I don't have push access, could you take care of that?
On 14 April 2014 18:47, Andy Zhou wrote:
> Looks good. Thanks.
> Acked-by: Andy Zhou
>
> On Sun, Apr 13, 2014 at 10:30 PM, Joe Stringer wrote:
> > From: Joe Stringer
> >
> > Add a constant for the number of bond buckets, and othe
On Sat, Apr 5, 2014 at 5:23 AM, Alex Wang wrote:
> In order to allow handlers directly read upcalls from datapath,
> we need to support per-handler netlink socket for each vport in
> datapath. This commit makes this happen. Also, it is guaranteed
> backward and forward compatibility with previou
> Currently recirculation action can optionally compute hash. This patch
> adds a hash action that is independent of the recirc action, which
> no longer computes hash. For megaflow bond with recirc, the output
> to a bond port action will look like:
>
> hash(hash_l4(0)), recric()
>
> Obviou
On Wed, Apr 9, 2014 at 4:59 PM, Thomas Graf wrote:
> On 04/08/2014 12:00 AM, Pravin wrote:
>>
>> +struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
>> + const struct sw_flow_key *key,
>> + u32 skb_hash,
It could be I am still missing something here. It seems most of the
writes are protected by the write lock, with the exception
of bond_recirculation_account(), which is the main issue this patch
fixes. So would it be simpler to just make it acquire the write lock
instead of read lock? Unless it i
63 matches
Mail list logo
