A set of fixes for net.
First bug is related flow-table management. Second one is in sample
action. Third is related flow stats and last one add gre-err handler for ovs.
The following changes since commit dba63115ce0c888fcb4cdec3f8a4ba97d144afaf:
powerpc: bpf: Fix the broken LD_VLAN_TAG_PRESEN
Is there any chance we could get some unit tests written for IPFIX as part
of this effort?
The netflow tests occasionally help to prevent regressions, so it would be
good to have something similar for IPFIX.
___
dev mailing list
dev@openvswitch.org
http:
Yes, I was thinking the same thing. I'm looking forward to trying it out.
Especially if I can submit patches via https://github.com/sflow/ovs. That
will make life easier. I have one outstanding that lays some groundwork
for the sFlow tunnel extensions. If that process goes smoothly then I'll
On Mon, Jun 30, 2014 at 05:50:20PM -0700, Gurucharan Shetty wrote:
> For RHEL, Fedora and Xenserver, run unit tests while
> building rpms. This may catch some cross-platform bugs.
>
> The commit also allows the users to optionally skip unit tests.
> (On debian, the default is to run unit tests. F
This will generate HTML output of the testsuite coverage in
_build-gcc/coverage/*.
After running "ovs-dev.py coverage", the source will be configured to
collect coverage information. This may cause testsuite failures,
particularly if running with high parallelism. Re-run "ovs-dev.py conf"
to remov
Signed-off-by: Joe Stringer
---
utilities/ovs-dev.py | 21 ++---
1 file changed, 10 insertions(+), 11 deletions(-)
diff --git a/utilities/ovs-dev.py b/utilities/ovs-dev.py
index 74143ca..91732e1 100755
--- a/utilities/ovs-dev.py
+++ b/utilities/ovs-dev.py
@@ -47,6 +47,14 @@ def
Commit acaa8dac49 (revalidator: Eliminate duplicate flow handling.)
ensured that a ukey will always exist for a given flow, even if it is
about to be deleted. This means that push_dump_ops__() no longer needs
to handle the case where there is no ukey. This commit removes the
redundant code.
Signed
>
> The correct place to put it is in %check section of the spec file.
> It's not common to see specs using %check section, but it could be
> because most of the projects don't have a testsuite.
>
> Anyway, this is the package guide lines for Fedora:
> http://fedoraproject.org/wiki/Packaging:Guidel
For RHEL, Fedora and Xenserver, run unit tests while
building rpms. This may catch some cross-platform bugs.
The commit also allows the users to optionally skip unit tests.
(On debian, the default is to run unit tests. For consistency,
do the same for rpms.)
VMware-BZ: 1267127
CC: Flavio Leitne
The upstream u64_stats API has been changed to remove the _bh()
versions and switch all consumers to use IRQ safe variants instead.
This was done to be safe for netpoll generated packets, which can
occur in hard IRQ context. From a safety perspective, this doesn't
directly affect OVS since it doesn
Thx Pravin, applied the branch-2.0 branch-2.1 patches with your suggestion.
Still need a Ack for the master/branch-2.3 patch.
Thanks,
Alex Wang,
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
Just to be clear, this is a v2 on the patch 12/12 of the series, due to the
batching changes Ethan just pushed. The 1-11 seem to still apply.
Jarno
On Jun 30, 2014, at 4:37 PM, Jarno Rajahalme wrote:
> Now that all the relevant classifier structures use RCU and internal
> mutual exclusion fo
Now that all the relevant classifier structures use RCU and internal
mutual exclusion for modifications, we can remove the fat-rwlock and
thus make the classifier lookups lockless.
As the readers are operating concurrently with the writers, a
concurrent reader may or may not see a new rule being a
On Mon, Jun 30, 2014 at 02:43:35PM -0700, Gurucharan Shetty wrote:
> On Mon, Jun 30, 2014 at 2:29 PM, Flavio Leitner wrote:
> >
> > Hi Gurucharan,
> >
> > I've tried to do the same with Fedora's spec but the time to run
> > "make check" when doing frequent builds is considerable.
> >
> > I had qui
On Mon, Jun 30, 2014 at 2:36 PM, Alex Wang wrote:
> Due to the race condition in userspace, there is chance that two
> overlapping megaflows could be installed in datapath. And this
> causes userspace unable to delete the less inclusive megaflow flow
> even after it timeout, since the flow_del lo
The upstream kernel net/netlink/af_netlink.c netlink_recvmsg() contains the
following code to refill the Netlink socket buffer with more dump skbs
while a dump is in progress:
if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
ret = netlink_dump(sk);
Due to the race condition in userspace, there is chance that two
overlapping megaflows could be installed in datapath. And this
causes userspace unable to delete the less inclusive megaflow flow
even after it timeout, since the flow_del logic will stop at the
first match of masked flow.
This comm
Typically, kernel datapath threads send upcalls to userspace where
handler threads process the upcalls. For TAP and DPDK devices, the
datapath threads operate in userspace, so there is no need for
separate handler threads.
This patch allows userspace datapath threads to directly call the
ofproto u
I was actually thinking that when the limit is reached, we would stop
creating new batches, and simply execute any packets which don't fit
into the existing set one by one. Not sure which way is better.
Anyways I'll send out a proposal and we can decide how to do it in the
review.
Ethan
On Mon,
On Jun 30, 2014, at 2:32 PM, Ethan Jackson wrote:
>> Would it make sense to batch the upcalls as well, i.e. collect them up and
>> then issue them with a single dp_netdev_output_userspace()?
>
> Yep, Ryan is working on that.
>
>> It would seem that starting from the newest batch and looping b
On Mon, Jun 30, 2014 at 2:29 PM, Flavio Leitner wrote:
>
> Hi Gurucharan,
>
> I've tried to do the same with Fedora's spec but the time to run
> "make check" when doing frequent builds is considerable.
>
> I had quite a few false-positives during parallel execution of
> "make check" and that break
Due to the race condition in userspace, there is chance that two
overlapping megaflows could be installed in datapath. And this
causes userspace unable to delete the less inclusive megaflow flow
even after it timeout, since the flow_del logic will stop at the
first match of masked flow.
This comm
Due to the race condition in userspace, there is chance that two
overlapping megaflows could be installed in datapath. And this
causes userspace unable to delete the less inclusive megaflow flow
even after it timeout, since the flow_del logic will stop at the
first match of masked flow.
This comm
> Would it make sense to batch the upcalls as well, i.e. collect them up and
> then issue them with a single dp_netdev_output_userspace()?
Yep, Ryan is working on that.
> It would seem that starting from the newest batch and looping backwards might
> find the batch sooner, especially if the pac
Hi Gurucharan,
I've tried to do the same with Fedora's spec but the time to run
"make check" when doing frequent builds is considerable.
I had quite a few false-positives during parallel execution of
"make check" and that breaks the entire rpm construction.
I believe most of developers are run
Minor comments below, otherwise:
Acked-by: Jarno Rajahalme
Jarno
On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
> Signed-off-by: Ethan Jackson
> ---
> lib/dpif-netdev.c | 96 +++
> 1 file changed, 54 insertions(+), 42 deletions(-)
>
>
On Mon, Jun 30, 2014 at 11:59:31AM -0700, Ben Pfaff wrote:
> If this seems like an improvement, I can do the same thing for the RHEL
> version.
>
> CC: Flavio Leitner
> Signed-off-by: Ben Pfaff
I think it's more clear, thanks Ben!
Acked-by: Flavio Leitner
> ---
> INSTALL.Fedora | 24 ++
weird, git am still give me same error, but removing headers and applying diff
using git apply works, anyways thanks.
LGTM.
Regards,
Pritesh
On Jun 30, 2014, at 1:56 PM, Jesse Gross wrote:
> Yes, it worked fine for me (using git am).
>
> On Mon, Jun 30, 2014 at 1:52 PM, Ben Pfaff wrote:
>> T
For RHEL, Fedora and Xenserver, run unit tests while
building rpms. This may catch some cross-platform bugs.
The commit also allows the users to optionally skip unit tests.
(On debian, the default is to run unit tests. For consistency,
do the same for rpms.)
VMware-BZ: 1267127
CC: Flavio Leitne
Yes, it worked fine for me (using git am).
On Mon, Jun 30, 2014 at 1:52 PM, Ben Pfaff wrote:
> That's odd. Jesse, did this apply for you?
>
> On Mon, Jun 30, 2014 at 08:47:08PM +, Pritesh Kothari (pritkoth) wrote:
>> LGTM but give error on master:
>> error: patch failed: datapath/flow_netlin
That's odd. Jesse, did this apply for you?
On Mon, Jun 30, 2014 at 08:47:08PM +, Pritesh Kothari (pritkoth) wrote:
> LGTM but give error on master:
> error: patch failed: datapath/flow_netlink.c:524
> error: datapath/flow_netlink.c: patch does not apply
>
> Regards,
> Pritesh
>
> On Jun 30,
I've made the suggested changes. Thanks for the review.
Ethan
On Mon, Jun 30, 2014 at 1:40 PM, Jarno Rajahalme wrote:
> Minor comments below, otherwise:
>
> Acked-by: Jarno Rajahalme
>
> Jarno
>
> On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
>
>> Used in a future patch.
>>
>> Signed-of
On Mon, Jun 30, 2014 at 01:45:06PM -0700, Jesse Gross wrote:
> On Mon, Jun 30, 2014 at 1:37 PM, Ben Pfaff wrote:
> > Found by inspection.
> >
> > Signed-off-by: Ben Pfaff
>
> Acked-by: Jesse Gross
Thanks, applied to master.
___
dev mailing list
dev@o
Yes my bad. Appears I'm rusty with my patch management . . .
Ethan
On Mon, Jun 30, 2014 at 1:28 PM, Jarno Rajahalme wrote:
> Did not review this as this seems to be already reviewed.
>
> Jarno
>
> On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
>
>> From: Gurucharan Shetty
>>
>> This just
On Mon, Jun 30, 2014 at 11:22 AM, Ryan Wilson 76511 wrote:
> You are right, turns out this was just an anomaly in performance. Please
> ignore the comment in the commit message.
>
> I still recommend this be pushed as these annotations also exist in
> netdev_dpdk_send() for similar functionality.
LGTM but give error on master:
error: patch failed: datapath/flow_netlink.c:524
error: datapath/flow_netlink.c: patch does not apply
Regards,
Pritesh
On Jun 30, 2014, at 1:37 PM, Ben Pfaff wrote:
> Found by inspection.
>
> Signed-off-by: Ben Pfaff
> ---
> datapath/flow_netlink.c |5 +++--
On Mon, Jun 30, 2014 at 1:37 PM, Ben Pfaff wrote:
> Found by inspection.
>
> Signed-off-by: Ben Pfaff
Acked-by: Jesse Gross
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
Minor comments below, otherwise:
Acked-by: Jarno Rajahalme
Jarno
On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
> Used in a future patch.
>
> Signed-off-by: Ethan Jackson
> ---
> lib/classifier.c | 47 ++-
> lib/classifier.h | 5 +++--
> lib
Found by inspection.
Signed-off-by: Ben Pfaff
---
datapath/flow_netlink.c |5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/datapath/flow_netlink.c b/datapath/flow_netlink.c
index 5a1a487..5ee3150 100644
--- a/datapath/flow_netlink.c
+++ b/datapath/flow_netlink.c
@@ -1,
Acked-by: Jarno Rajahalme
Jarno
On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
> Signed-off-by: Ethan Jackson
> ---
> lib/dpif-netdev.c | 8 +---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
> index 8954e1b..4deb763 100644
>
Acked-by: Jarno Rajahalme
On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
> The new name "packet_batch" is a bit more straight forward.
>
> Signed-off-by: Ethan Jackson
> ---
> lib/dpif-netdev.c | 10 +-
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/lib/dpif-net
Did not review this as this seems to be already reviewed.
Jarno
On Jun 30, 2014, at 1:11 PM, Ethan Jackson wrote:
> From: Gurucharan Shetty
>
> This just makes ovs-benchmark compile on windows.
> This lets us go ahead with just a 'make' instead of
> picking and choosing executables that are
The new name "packet_batch" is a bit more straight forward.
Signed-off-by: Ethan Jackson
---
lib/dpif-netdev.c | 10 +-
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index fce2650..8954e1b 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif
Signed-off-by: Ethan Jackson
---
lib/dpif-netdev.c | 96 +++
1 file changed, 54 insertions(+), 42 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 3a2b68f..58c629b 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -
Used in a future patch.
Signed-off-by: Ethan Jackson
---
lib/classifier.c | 47 ++-
lib/classifier.h | 5 +++--
lib/dpif-netdev.c | 2 +-
3 files changed, 34 insertions(+), 20 deletions(-)
diff --git a/lib/classifier.c b/lib/classifier.c
index 330
From: Gurucharan Shetty
This just makes ovs-benchmark compile on windows.
This lets us go ahead with just a 'make' instead of
picking and choosing executables that are tested to work on
windows as arguments for make.
This commit does not make ovs-benchmark a supported utility
on windows.
Signed
Signed-off-by: Ethan Jackson
---
lib/dpif-netdev.c | 8 +---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 8954e1b..4deb763 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -1073,15 +1073,13 @@ dp_netdev_flow_cast(const stru
Oops my bad.
On Mon, Jun 30, 2014 at 12:54 PM, Jarno Rajahalme wrote:
> This patch was already applied, and the only patch I would get to apply is
> 2/5. Seems like you need to rebase to master and repost.
>
> Jarno
>
> On Jun 29, 2014, at 9:24 PM, Ethan Jackson wrote:
>
>> This causes warnin
Thanks.
I'd like Flavio's opinion also.
On Mon, Jun 30, 2014 at 07:05:18PM +, Pritesh Kothari (pritkoth) wrote:
> Acked-by: Pritesh Kothari
>
> On Jun 30, 2014, at 11:59 AM, Ben Pfaff wrote:
>
> > If this seems like an improvement, I can do the same thing for the RHEL
> > version.
> >
>
This patch was already applied, and the only patch I would get to apply is 2/5.
Seems like you need to rebase to master and repost.
Jarno
On Jun 29, 2014, at 9:24 PM, Ethan Jackson wrote:
> This causes warnings and is useless.
>
> Signed-off-by: Ethan Jackson
> ---
> lib/dpif-netdev.c | 2
Thx Pravin, I'll address your comments.
Also, will send the patch for branch-2.0 -> master
On Mon, Jun 30, 2014 at 12:00 PM, Pravin Shelar wrote:
> Good catch.
>
>
> On Mon, Jun 30, 2014 at 10:59 AM, Alex Wang wrote:
> > Due to the race condition in userspace, there is chance that two
> > ove
Acked-by: Pritesh Kothari
On Jun 30, 2014, at 11:59 AM, Ben Pfaff wrote:
> If this seems like an improvement, I can do the same thing for the RHEL
> version.
>
> CC: Flavio Leitner
> Signed-off-by: Ben Pfaff
> ---
> INSTALL.Fedora | 24 ++--
> 1 file changed, 18 insertio
Good catch.
On Mon, Jun 30, 2014 at 10:59 AM, Alex Wang wrote:
> Due to the race condition in userspace, there is chance that two
> overlapping megaflows could be installed in datapath. And this
> causes userspace unable to delete the less inclusive megaflow flow
> even after it timeout, since
If this seems like an improvement, I can do the same thing for the RHEL
version.
CC: Flavio Leitner
Signed-off-by: Ben Pfaff
---
INSTALL.Fedora | 24 ++--
1 file changed, 18 insertions(+), 6 deletions(-)
diff --git a/INSTALL.Fedora b/INSTALL.Fedora
index d711e24..ae0eb28
On Mon, Jun 30, 2014 at 10:26:01AM -0700, Gurucharan Shetty wrote:
> After the change, both of them compile. test-netflow related
> unit tests pass.
>
> test-sflow related tests do not pass because
> of LOOPBACK_INTERFACE constraints for 'agent'.
> (It should be revisited later.)
>
> Signed-off-b
On Mon, Jun 30, 2014 at 08:02:40AM -0700, Gurucharan Shetty wrote:
> Signed-off-by: Gurucharan Shetty
Acked-by: Ben Pfaff
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
Sorry, this is only for branch-2.1. I'll start back/forward porting...
On Mon, Jun 30, 2014 at 11:28 AM, Andy Zhou wrote:
> Nice fix! Does this patch suppose to apply for branch-2.0? It only
> applies to branch-2.1 for me.
>
> So for branch 2.1:
>
> Acked-by: Andy Zhou
>
> On Mon, Jun 30, 20
Nice fix! Does this patch suppose to apply for branch-2.0? It only
applies to branch-2.1 for me.
So for branch 2.1:
Acked-by: Andy Zhou
On Mon, Jun 30, 2014 at 10:59 AM, Alex Wang wrote:
> Due to the race condition in userspace, there is chance that two
> overlapping megaflows could be instal
You are right, turns out this was just an anomaly in performance. Please
ignore the comment in the commit message.
I still recommend this be pushed as these annotations also exist in
netdev_dpdk_send() for similar functionality. (Again, remove the comment
on performance in my commit message if you
On Jun 30, 2014, at 11:08 AM, Pravin Shelar wrote:
> On Mon, Jun 30, 2014 at 9:56 AM, Daniele Di Proietto
> wrote:
>> I was happier with the previous approach, i.e. setting the timestamp on the
>> first queued packet.
>> Maybe there are issues I don’t see, can you elaborate a little?
>>
>
>
After the change, both of them compile. test-netflow related
unit tests pass.
test-sflow related tests do not pass because
of LOOPBACK_INTERFACE constraints for 'agent'.
(It should be revisited later.)
Signed-off-by: Gurucharan Shetty
---
v2-v3:
* Skip creating Windows events from the client. T
On Thu, Jun 26, 2014 at 6:16 PM, Ryan Wilson wrote:
> Since dropped packets due to large packet size or lack of memory
> are unlikely, it is best to add OVS_UNLIKELY annotations to these
> conditions.
>
> With DPDK fast path forwarding test, this increased throughtput
> from 4.12 to 4.16 million p
On Mon, Jun 30, 2014 at 10:53 AM, Ryan Wilson 76511 wrote:
> I'll add the inline to both dpdk_queue_flush() and dpdk_queue_flush__()
> and repost this patch.
>
Actually I did it and pushed the patch. I also added unlikely in flush function.
Thanks,
Pravin.
> Cheers,
>
> Ryan
>
> On 6/30/14 9:52
On Mon, Jun 30, 2014 at 9:58 AM, Daniele Di Proietto
wrote:
> This makes a lot of sense, thanks!
>
> Acked-by: Daniele Di Proietto
>
> On Jun 26, 2014, at 6:16 PM, Ryan Wilson wrote:
>
>> This patch fixes a bug where rte_pktmbuf_alloc() would fail and
>> packets which succeeded to allocate memor
On Mon, Jun 30, 2014 at 9:56 AM, Daniele Di Proietto
wrote:
> I was happier with the previous approach, i.e. setting the timestamp on the
> first queued packet.
> Maybe there are issues I don’t see, can you elaborate a little?
>
tsc is set conditionally when count is one, but the count might not
Hi,
Are you interested in acquiring ITSM User's Email list for your marketing
campaigns?
Titles like - IT Decision Makers, CEO, CFO, IT Engineers and many more.
We provide the top Decision Makers Name, Email, Title, Phone and company
details across North America, EMEA, ASIA Pacific and
Due to the race condition in userspace, there is chance that two
overlapping megaflows could be installed in datapath. And this
causes userspace unable to delete the less inclusive megaflow flow
even after it timeout, since the flow_del logic will stop at the
first match of masked flow.
This comm
I'll add the inline to both dpdk_queue_flush() and dpdk_queue_flush__()
and repost this patch.
Cheers,
Ryan
On 6/30/14 9:52 AM, "Daniele Di Proietto" wrote:
>I don¹t know if we want to keep the inline attribute here (like Pravin
>said for commit 4/5).
>
>Otherwise LGTM
>
>Daniele
>
>On Jun 26,
These are already in the git repository code.
Mike Polehn
-Original Message-
From: dev [mailto:dev-boun...@openvswitch.org] On Behalf Of Daniele Di Proietto
Sent: Monday, June 30, 2014 10:00 AM
To: Ryan Wilson
Cc: dev@openvswitch.org
Subject: Re: [ovs-dev] [PATCH v2 5/5] netdev-dpdk: Add
On Mon, Jun 30, 2014 at 10:13:54AM -0700, Gurucharan Shetty wrote:
> On Thu, Jun 26, 2014 at 1:26 PM, Ben Pfaff wrote:
> > On Tue, Jun 24, 2014 at 11:52:54AM -0700, Gurucharan Shetty wrote:
> >> On Windows ECONNRESET is WSAECONNRESET.
> >>
> >> Also, "unix" connections are done through TCP sockets
On Thu, Jun 26, 2014 at 1:26 PM, Ben Pfaff wrote:
> On Tue, Jun 24, 2014 at 11:52:54AM -0700, Gurucharan Shetty wrote:
>> On Windows ECONNRESET is WSAECONNRESET.
>>
>> Also, "unix" connections are done through TCP sockets.
>> For the 'refuse-connection' test, the error message for Windows
>> is WS
Acked-by: Daniele Di Proietto
On Jun 26, 2014, at 7:17 PM, Ben Pfaff wrote:
> Great, thanks. Looks good.
>
> I'll leave it to whoever reviews the series as a whole to push this.
> On Jun 26, 2014 6:36 PM, "Ryan Wilson 76511" wrote:
>
>> Crap, its late in the day and I can't think / type appa
This makes a lot of sense, thanks!
Acked-by: Daniele Di Proietto
On Jun 26, 2014, at 6:16 PM, Ryan Wilson wrote:
> This patch fixes a bug where rte_pktmbuf_alloc() would fail and
> packets which succeeded to allocate memory with rte_pktmbuf_alloc()
> would not be sent and leak memory.
>
> Als
I was happier with the previous approach, i.e. setting the timestamp on the
first queued packet.
Maybe there are issues I don’t see, can you elaborate a little?
Thanks,
Daniele
On Jun 26, 2014, at 6:16 PM, Ryan Wilson wrote:
> The current timestamp should be set every time the queue is flushe
I don’t know if we want to keep the inline attribute here (like Pravin said for
commit 4/5).
Otherwise LGTM
Daniele
On Jun 26, 2014, at 6:16 PM, Ryan Wilson wrote:
> This patch refactors dpdk_queue_flush() to reuse code in
> dpdk_queue_pkts().
>
> Signed-off-by: Ryan Wilson
> ---
> lib/netd
Looks good, thanks!
On Jun 30, 2014 8:52 AM, "Gurucharan Shetty" wrote:
> Signed-off-by: Gurucharan Shetty
> ---
> INSTALL.RHEL |2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/INSTALL.RHEL b/INSTALL.RHEL
> index 01b95f7..de85199 100644
> --- a/INSTALL.RHEL
> +++ b/IN
Signed-off-by: Gurucharan Shetty
---
INSTALL.RHEL |2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/INSTALL.RHEL b/INSTALL.RHEL
index 01b95f7..de85199 100644
--- a/INSTALL.RHEL
+++ b/INSTALL.RHEL
@@ -22,7 +22,7 @@ the top-level:
5. SRPMS/ Where the newly created source packa
On Fri, Jun 27, 2014 at 5:40 PM, Ben Pfaff wrote:
> On Fri, Jun 27, 2014 at 02:32:51PM -0700, Gurucharan Shetty wrote:
>> We currently have a poll_fd_wait_event(fd, wevent, events) function that
>> is used at places common to Windows and Linux where we have to wait on
>> sockets. On Linux, 'weven
On Sun, Jun 29, 2014 at 09:21:05PM -0300, Flavio Leitner wrote:
> Signed-off-by: Flavio Leitner
Thanks a lot. Applied to master.
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
This requires less locking and makes lockless classifier use a simpler.
Signed-off-by: Jarno Rajahalme
---
lib/dpif-netdev.c | 51 ---
1 file changed, 24 insertions(+), 27 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index fce
Add an internal mutex to the struct cls_classifier, and reorganize
classifier internal structures according to the user of each field,
marking the fields that need to be protected by the mutex. This makes
locking requirements easier to track, and may make lookup more memory
efficient.
After this
cls_set_prefix_fields() now synchronizes explicitly with the readers,
waiting them to finish using the old configuration before changing to
the new configuration.
Signed-off-by: Jarno Rajahalme
---
lib/classifier.c| 214 +--
lib/classifier.h
Now that all the relevant classifier structures use RCU and internal
mutual exclusion for modifications, we can remove the fat-rwlock and
thus make the classifier lookups lockless.
As the readers are operating concurrently with the writers, a
concurrent reader may or may not see a new rule being a
Rename 'nbits' as 'n_bits' to be more consistent with other count-like
fields.
Signed-off-by: Jarno Rajahalme
---
lib/classifier.c | 60 +++---
1 file changed, 30 insertions(+), 30 deletions(-)
diff --git a/lib/classifier.c b/lib/classifier.c
in
The following patch will add the first external user.
Signed-off-by: Jarno Rajahalme
---
lib/ovs-rcu.c |3 +--
lib/ovs-rcu.h |4
2 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/lib/ovs-rcu.c b/lib/ovs-rcu.c
index 62fe614..57f550c 100644
--- a/lib/ovs-rcu.c
+++ b/lib/o
Hide the cursor from the classifier iteration users and move locking to
the iterators. This will make following RCU changes simpler, as the call
sites of the iterators need not be changed at that point.
Signed-off-by: Jarno Rajahalme
---
lib/classifier.c| 70 ++
Signed-off-by: Jarno Rajahalme
---
lib/classifier.h |2 --
1 file changed, 2 deletions(-)
diff --git a/lib/classifier.h b/lib/classifier.h
index 326ca08..4c2f39b 100644
--- a/lib/classifier.h
+++ b/lib/classifier.h
@@ -293,8 +293,6 @@ bool classifier_rule_overlaps(const struct classifier *cl
Use cmap instead of hmap & hindex in classifier. Performance impact
with current locking strategy is not yet tested. Later patches will
introduce RCU into the classifier.
Signed-off-by: Jarno Rajahalme
Acked-by: Ben Pfaff
---
lib/classifier.c| 194 +---
This is a prerequisite step in making the classifier lookups lockless.
If taking a reference fails, we do the lookup again, as a new (lower
priority) rule may now match instead.
Also remove unwildcarding dl_type and nw_frag, as these are already
taken care of by xlate_actions().
Signed-off-by: Ja
Updating the reference count only requires atomicity, but no memory
ordering with respect to any other loads or stores. Avoiding the
overhead of the default memory_order_seq_cst can make these more
efficient.
Signed-off-by: Jarno Rajahalme
---
lib/ovs-atomic.h |8 +---
1 file changed, 5
This series makes the classifier use OVS-RCU and thus makes classifier
lookups lockless. No performance measurements have been made yet.
Patch 6 ("Use cmap.") has been previously reviewed.
Jarno Rajahalme (12):
dpif-netdev: Use cmap instead of hmap.
ovs-atomic: Use memory_order_relaxed for o
Add support for atomic compare_exchange operations.
Add ovs_refcount_try_ref(), which takes a reference only if the
refcount is non-zero and returns true if a reference was taken, false
otherwise. This can be used in combined RCU/refcount scenarios where
we have an RCU protected reference to an r
The original message was received at Mon, 30 Jun 2014 19:33:31 +0700 from
44.244.122.121
- The following addresses had permanent fatal errors -
dev@openvswitch.org
- Transcript of session follows -
... while talking to 84.71.62.212:
554 5.0.0 Service unavailable; [81.24.131.90] b
93 matches
Mail list logo
