When I build ovs-dpdk without mlx4/mlx5 driver, the configure log is as
follows:
checking for DPDK... yes
checking for faulty pkg-config version... no
checking for rte_config.h... yes
checking whether RTE_LIBRTE_VHOST_NUMA is declared... yes
checking for library containing get_mempolicy... -lnuma
In case of native tunnel with bfd enabled, if the MAC address of the
remote end's interface changes (e.g. because it got rebooted, and the
MAC address is allocated dynamically), the BFD session will never be
re-established.
This happens because the local tunnel neigh entry doesn't get updated,
with the command is now possible to change the aging time of the
cache entries.
For the existing entries the aging time is updated only if the
current expiration is greater than the new one. In any case, the next
refresh will set it to the new value.
This is intended mostly for debugging
This is a minor issue but visible e.g. when you try to flush the neigh
cache while the ARP flow is still present in the datapath, triggering
the revalidation of the datapath flows which subsequently
refreshes/adds the entry in the cache.
Signed-off-by: Paolo Valerio
---
v2:
- renamed update ->
Expires is modified in different threads (revalidator, pmd-rx, bfd-tx).
It's better to use atomics for such potentially parallel write.
Signed-off-by: Paolo Valerio
---
v2:
- modified commit description
- added _MS suffix to NEIGH_ENTRY_DEFAULT_IDLE_TIME
- renamed local variable expired ->
v2:
- rebased against master
- turned 'ageing' -> 'aging'
- further details of v1 -> v2 respin has been added
to each patch
The series is composed of the following patches:
#1: Expires is modified in different contexts (revalidator, pmd-rx, bfd-tx).
It's probably not very racy for many
Gaëtan Rivet writes:
> On Mon, Nov 8, 2021, at 20:30, Paolo Valerio wrote:
> [...]
+
+atomic_read_relaxed(>expires, );
>>>
>>> I'm having doubts about unfenced read / writes on expires.
>>> Technically on lookup there would be reads then writes without barriers.
>>> I'm not
On Wed, Nov 3, 2021 at 3:17 PM svc.eng.git-mail
wrote:
>
> Hi Lorenzo,
>
> > what will it happen if pinctrl learns a different MAC for the given IP
> > after the CMS has
> configured the static binding? Will it be overwritten? If so I guess we need a
> flag to avoid it.
>
> 1. Static binding
On Mon, Nov 8, 2021 at 6:55 AM Dumitru Ceara wrote:
>
> On 11/8/21 12:45 PM, Adrian Moreno wrote:
> >
> >
> > On 11/5/21 17:12, Dumitru Ceara wrote:
> >> On 10/21/21 11:10 AM, Adrian Moreno wrote:
> >>> Currently, if "make" is run after the project is built, the root
> >>> manpage (ovn-detrace.1)
On Tue, Nov 9, 2021 at 11:12 AM Numan Siddique wrote:
>
> On Tue, Nov 9, 2021 at 1:38 AM Frode Nordahl
> wrote:
> >
> > On Mon, Nov 8, 2021 at 11:08 PM Numan Siddique wrote:
> > >
> > > On Fri, Nov 5, 2021 at 5:32 PM Frode Nordahl
> > > wrote:
> > > >
> > > > fre. 5. nov. 2021, 20:43 skrev Han
Bleep bloop. Greetings Mark Gray, I am a robot and I have tried out your patch.
Thanks for your contribution.
I encountered some error that I wasn't expecting. See the details below.
checkpatch:
WARNING: Comment with 'xxx' marker
#1500 FILE: northd/northd.c:14516:
/* XXX Having to
From: Mark Gray
Add an additional node that initially does nothing. This serves as a
template for how to add a new node. This node is inserted after
the northd_node.
This node will be updated in a later commit to generate logical
flows for the SBDB.
Signed-off-by: Mark Gray
---
From: Mark Gray
'struct northd_data' is used to hold the global state data for the
incremental processing node 'en_northd'. This structure will also hold
'struct northd_input' which will hold references to output data from
its input nodes. In particular, this will hold references to database
From: Mark Gray
In order to remove the IDL loop variable from the engine context,
we do not calculate the database sequence numbers incrementally.
Signed-off-by: Mark Gray
---
lib/inc-proc-eng.h | 2 -
northd/en-northd.c | 3 +-
northd/inc-proc-northd.c | 2 -
From: Mark Gray
Initial implementation adds a single node (northd). This single
node executes the northd processing pipeline but does not do so
incrementally.
In order to develop incremental processing for northd, the code
will be organised with a .c/.h file for each I-P node following
the
From: Mark Gray
As 'mcast_groups' and 'igmp_groups' need to be modified by
build_lflows(), we move build_mcast_groups() to build_flows(). This
is in preparation for a later commit which will create an lflow
node. If these structures form part of the input of the new lflow
we should not modify
From: Mark Gray
destroy_nat_entries() iterates over nat_entries using 'n_nat'
as the number of NAT entries from the NB database. This behaviour can be
incorrect as it assumes that there are 'n_nat' 'nat_entries'. 'struct
ovn_datapath' should maintain a count of 'nat_entries' in 'struct
Add the 'inc-proc-eng' framework to northd. This does *not*
add any incremental processing at this stage but provides the
framework to do so. Even in this base configuration, we see an
advantage as northd no longer processes the databases if it has
been woken only to handle, for example, a unixctl
On Fri, Nov 5, 2021 at 5:21 AM Dumitru Ceara wrote:
>
> There are various costs (e.g., not being able to perform hardware
> offload in some cases) when using check_pkt_larger() so the CMS
> can now limit the impact by bypassing the packet length checks for
> specific types of traffic (e.g., TCP).
On Tue, Nov 9, 2021 at 1:38 AM Frode Nordahl
wrote:
>
> On Mon, Nov 8, 2021 at 11:08 PM Numan Siddique wrote:
> >
> > On Fri, Nov 5, 2021 at 5:32 PM Frode Nordahl
> > wrote:
> > >
> > > fre. 5. nov. 2021, 20:43 skrev Han Zhou :
> > >
> > > >
> > > >
> > > > On Fri, Nov 5, 2021 at 7:00 AM Frode
On Mon, Nov 1, 2021 at 9:36 AM Dumitru Ceara wrote:
>
> Since 0038579d1928 ("northd: Optimize ct nat for load balancer
> traffic.") calls to 'ct()' and 'ct(nat)' are merged because 'ct(nat)'
> implies 'ct()'. However ovn-trace was not setting ct_state when
> processing 'ct(nat)'. This was
It was not the intent to change the range of tunnel keys for vtep
setups. If that happens, we should fix it. Perhaps the range is not
touched but the range validation is activated in vtep case. If so, the
validation should be disabled.
Ihar
On 11/8/21 4:57 PM, Vladislav Odintsov wrote:
Hi,
On Wed, Nov 3, 2021 at 4:11 PM Numan Siddique wrote:
>
> On Tue, Nov 2, 2021 at 3:35 PM Dumitru Ceara wrote:
> >
> > On 11/2/21 8:30 PM, Dumitru Ceara wrote:
> > > This series backports the load balancer performance improvements to
> > > stable branch-21.09.
> > >
> > > - patch 1/3 changes the
On Tue, Nov 9, 2021 at 7:42 AM Dumitru Ceara wrote:
>
> On 11/8/21 6:10 PM, num...@ovn.org wrote:
> > From: Numan Siddique
> >
> > If a logical switch is used to join multiple gateway
> > routers and if each of these gateway routers are
> > configured with the same load balancer, then after the
That sounds okay.
But the problem is that if one have at least one chassis, which even doesn’t
act in OVN IC, it will break ovn interconnected switches and LRs.
Regards,
Vladislav Odintsov
> On 9 Nov 2021, at 18:46, Numan Siddique wrote:
>
> On Mon, Nov 8, 2021 at 5:05 PM Numan Siddique
On Tue, Nov 9, 2021 at 7:43 AM Dumitru Ceara wrote:
>
> This change is backported from upstream ovn-kubernetes as we're seeing
> the same issue in OVN runs too:
>
> https://github.com/ovn-org/ovn-kubernetes/commit/c2ef04188
>
> Signed-off-by: Dumitru Ceara
Thanks for fixing this.
Applied to
vhost ops is now prefixed with rte_ [1].
1: https://git.dpdk.org/dpdk/commit/?id=e7c727c307d7
Signed-off-by: David Marchand
---
lib/netdev-dpdk.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index f45575e9a5..d1fd6c766e 100644
---
PKT_[RT]X_* and other mbuf macros have been prefixed with RTE_MBUF_ [1].
Update accordingly.
1: https://git.dpdk.org/dpdk/commit/?id=daa02b5cddbb
Signed-off-by: David Marchand
---
lib/dp-packet.h | 26 ++
lib/netdev-dpdk.c | 18 +-
2 files changed, 23
The shared attribute has been deprecated some time ago and is removed
in 21.11 [1].
Since its value was 0, let's rely on implicit init by not mentionning it.
1: https://git.dpdk.org/dpdk/commit/?id=92ef4b8f1688
Signed-off-by: David Marchand
---
lib/netdev-dpdk.c | 2 +-
1 file changed, 1
Hello Ian,
This series applies on a dpdk-latest branch rebased on current master [1].
It builds and passes checks in GHA, see [2].
Fyi, the lcore API is marked stable, I'll repost my patch for OVS as a
followup of this series.
v21.11-rc2 was tagged last night.
There will be no other API change
Instead of dereferencing DPDK ethdev internals, we can list "sibling"
ports [1]: such ports share the underlying rte_device object.
That API was experimental and went stable in 21.11 [2].
1: https://git.dpdk.org/dpdk/commit/?id=7f98942886bc
2: https://git.dpdk.org/dpdk/commit/?id=b37ed6def367
While testing OVS-windows flows for the Ip fragments, the traffic will be
dropped
As it may match incorrect OVS flow. From the code, after the Ipv4 fragments are
Reassembled, it willl still use the flow key of the last Ipv4 fragments to match
CT causing match error.
On Mon, Nov 8, 2021 at 5:05 PM Numan Siddique wrote:
>
> + Ihar if he has any comments.
>
> Thanks
> Numan
>
Due to the limitations of VxLAN, I don't think OVN interconnect is
supported with VxLAN.
Please see the ovn-architecture man pages.
--
When VXLAN tunneling is enabled in an OVN
num...@ovn.org writes:
> From: Numan Siddique
>
> xlate_check_pkt_larger() sets ctx->exit to 'true' at the end
> causing the translation to stop. This results in incomplete
> datapath rules.
>
> For example, for the below OF rules configured on a bridge,
>
> table=0,in_port=1
>
This reverts commit c645550bb249 ("odp-util: Always report
ODP_FIT_TOO_LITTLE for IGMP.")
Always forcing a slow path action can result in some over-broad
flows which swallow all traffic and force them to userspace, as reported
in the thread at
This change is backported from upstream ovn-kubernetes as we're seeing
the same issue in OVN runs too:
https://github.com/ovn-org/ovn-kubernetes/commit/c2ef04188
Signed-off-by: Dumitru Ceara
---
.github/workflows/ovn-kubernetes.yml | 14 --
1 file changed, 14 deletions(-)
diff
On 11/8/21 6:10 PM, num...@ovn.org wrote:
> From: Numan Siddique
>
> If a logical router is associated with load balancer VIPs which
> are unreachable from it, we don't need to add ARP responder flows
> in the L2_LKUP stage of ingress pipeline of the logical switches
> connected to the router.
On 11/8/21 6:10 PM, num...@ovn.org wrote:
> From: Numan Siddique
>
> If a logical switch is used to join multiple gateway
> routers and if each of these gateway routers are
> configured with the same load balancer, then after the
> commit [1] it's results in a flow explosion in
> the
On Mon, Nov 8, 2021, at 20:30, Paolo Valerio wrote:
[...]
>>> +new_exp = time_msec() + ageing;
>>> +
>>> +CMAP_FOR_EACH (neigh, cmap_node, ) {
>>> +atomic_read_relaxed(>expires, _exp);
>>> +if (new_exp < curr_exp) {
>>> +atomic_store_relaxed(>expires, new_exp);
On Mon, Nov 8, 2021, at 20:30, Paolo Valerio wrote:
[...]
>>> +
>>> +atomic_read_relaxed(>expires, );
>>
>> I'm having doubts about unfenced read / writes on expires.
>> Technically on lookup there would be reads then writes without barriers.
>> I'm not convinced it makes a difference, WDYT?
On Tue, Nov 9, 2021 at 11:21 AM 0-day Robot wrote:
>
> Bleep bloop. Greetings Frode Nordahl, I am a robot and I have tried out your
> patch.
> Thanks for your contribution.
>
> I encountered some error that I wasn't expecting. See the details below.
>
>
> checkpatch:
> WARNING: Line is 113
Bleep bloop. Greetings Frode Nordahl, I am a robot and I have tried out your
patch.
Thanks for your contribution.
I encountered some error that I wasn't expecting. See the details below.
checkpatch:
WARNING: Line is 113 characters long (recommended limit is 79)
#29 FILE: ovn-nb.xml:2824:
The Logical Router Port gateway_mtu option is currently only
documented together with implementation details in the ovn-northd
documentation.
As this is CMS API we ought to document this as part of the OVN
Northbound Database as well.
Reported-at: https://github.com/ovn-org/ovn/issues/78
Fixes:
The Logical Router Port gateway_mtu option is currently only
documented together with implementation details in the ovn-northd
documentation.
As this is CMS API we ought to document this as part of the OVN
Northbound Database as well.
Reported-at: https://github.com/ovn-org/ovn/issues/78
Fixes:
On Tue, Nov 9, 2021 at 10:40 AM 0-day Robot wrote:
>
> Bleep bloop. Greetings David Marchand, I am a robot and I have tried out
> your patch.
> Thanks for your contribution.
>
> I encountered some error that I wasn't expecting. See the details below.
>
>
> checkpatch:
> ERROR: Improper
Bleep bloop. Greetings David Marchand, I am a robot and I have tried out your
patch.
Thanks for your contribution.
I encountered some error that I wasn't expecting. See the details below.
checkpatch:
ERROR: Improper whitespace around control block
#74 FILE: lib/netdev-dpdk.c:1450:
Fix a typo which made us build against stable API only in GHA.
Note: this should not be merged to master, intended for dpdk-latest
branch only.
Fixes: e5cdd2651661 ("travis: Check compilation with DPDK experimental API.")
Signed-off-by: David Marchand
---
Ian,
At the next dpdk-latest rebase,
vhost ops is now prefixed with rte_ [1].
1: https://git.dpdk.org/dpdk/commit/?id=e7c727c307d7
Signed-off-by: David Marchand
---
lib/netdev-dpdk.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index 2eb5e2e1bc..882b138cfa 100644
---
PKT_[RT]X_* and other mbuf macros have been prefixed with RTE_MBUF_ [1].
Update accordingly.
1: https://git.dpdk.org/dpdk/commit/?id=daa02b5cddbb
Signed-off-by: David Marchand
---
lib/dp-packet.h | 26 ++
lib/netdev-dpdk.c | 18 +-
2 files changed, 23
The shared attribute has been deprecated some time ago and is removed
in 21.11 [1].
Since its value was 0, let's rely on implicit init by not mentionning it.
1: https://git.dpdk.org/dpdk/commit/?id=92ef4b8f1688
Signed-off-by: David Marchand
---
lib/netdev-dpdk.c | 2 +-
1 file changed, 1
Instead of dereferencing DPDK ethdev internals, we can list "sibling"
ports [1]: such ports share the underlying rte_device object.
That API was experimental and went stable in 21.11 [2].
1: https://git.dpdk.org/dpdk/commit/?id=7f98942886bc
2: https://git.dpdk.org/dpdk/commit/?id=b37ed6def367
Hello Ian,
This series applies on a dpdk-latest branch rebased on current master [1].
It builds and passes checks in GHA, see [2].
Fyi, the lcore API is marked stable, I'll repost my patch for OVS as a
followup of this series.
v21.11-rc2 was tagged last night.
There will be no other API change
On Tue, Sep 7, 2021 at 11:35 AM Maxime Coquelin
wrote:
> On 9/7/21 10:23 AM, David Marchand wrote:
> > Instead of dereferencing DPDK ethdev internals, we can list "sibling"
> > ports [1]: such ports share the underlying rte_device object.
> >
> > This API was experimental so far, but it will go
While testing OVS-windows flows for the Ip fragments, the traffic will be
dropped
As it may match incorrect OVS flow. From the code, after the Ipv4 fragments are
Reassembled, it willl still use the flow key of the last Ipv4 fragments to match
CT causing match error.
54 matches
Mail list logo