[ovs-dev] [PATCH 2/2] netdev-dpdk: Retry tx/rx queue setup until we don't get any failure.

Ethan Jackson ethan at nicira.com
Tue Aug 4 21:46:03 UTC 2015


Sure let's do that.  If you write an INSTALL.DPDK note, and post it
here.  I'll incorporate it into the patch and merge.  Thanks for your
patience.

Ethan

On Tue, Aug 4, 2015 at 4:48 AM, Stokes, Ian <ian.stokes at intel.com> wrote:
> This issue currently affects Fortville cards on systems with a logical core count greater than 64.
>
> Just to call out however that the number of cores setup by default to forward traffic to a NIC using OVS with DPDK will be the total number of cores detected on the system. This is the default behavior, thus there is no way user can "devote" less cores to forwarding on the NIC.
>
> This means they will always hit this issue when using a system with 72 logical cores and a Fortville card. There is no way they can work around the issue with tx configuration changes.
>
> If this preferable then I can write up a note for theINSTALL.DPDK.md explaining the issue.
>
> Thanks
> Ian
>
>> -----Original Message-----
>> From: Ben Pfaff [mailto:blp at nicira.com]
>> Sent: Monday, August 03, 2015 6:11 PM
>> To: Ethan Jackson
>> Cc: Stokes, Ian; Justin Pettit; Pravin Shelar; Traynor, Kevin; Daniele
>> Di Proietto; dev at openvswitch.org
>> Subject: Re: [ovs-dev] [PATCH 2/2] netdev-dpdk: Retry tx/rx queue setup
>> until we don't get any failure.
>>
>> If that's an accurate description of the problem then that seems fine to
>> me.  We could add a note to INSTALL.DPDK.md describing the issue I
>> suppose, if someone wants to write one up.
>>
>> On Sat, Aug 01, 2015 at 02:40:16PM -0700, Ethan Jackson wrote:
>> > I personally am fine with waiting on this being fixed until December.
>> > My reading is, it only happens on one specific NIC, and even then only
>> > if you devote a huge number of cores to forwarding on that NIC.
>> >
>> > That said, I won't block this if another committer disagrees with me.
>> > Ben Justin Pravin?  What do you think?
>> >
>> > Ethan
>> >
>> > On Thu, Jul 30, 2015 at 1:13 AM, Stokes, Ian <ian.stokes at intel.com>
>> wrote:
>> > > So ideally this will be fixed in a future release of DPDK. We have
>> flagged this. However that solution will not be in place until the DPDK
>> 2.2 release in December at the earliest (DPDK 2.1 is currently in
>> release candidate mode at the moment so it won't make it to that). When
>> this has been changed in DPDK we can revisit the OVS code.
>> > >
>> > > Technically DPDK is doing what it is supposed to with the current
>> implementation i.e. it is returning the max number of queues it
>> supports. From the OVS side I think we need to understand that this has
>> a different connotation to what it had with previously with NICS in
>> terms of how many of those queues are usable.
>> > >
>> > > Unfortunately I don’t see another way to negotiate the tx queue
>> initialization without something like the patch below.
>> > > Not until we have more explicit configuration details available for
>> the HW device from DPDK.
>> > >
>> > > Thanks
>> > > Ian
>> > >
>> > >> -----Original Message-----
>> > >> From: Ethan Jackson [mailto:ethan at nicira.com]
>> > >> Sent: Wednesday, July 29, 2015 10:13 PM
>> > >> To: Stokes, Ian
>> > >> Cc: Traynor, Kevin; Daniele Di Proietto; dev at openvswitch.org
>> > >> Subject: Re: [ovs-dev] [PATCH 2/2] netdev-dpdk: Retry tx/rx queue
>> setup
>> > >> until we don't get any failure.
>> > >>
>> > >> Sorry for taking so long to get to this.  The one question I have
>> is:
>> > >> Is OVS the right layer to be fixing this?  Isn't this really an
>> issue
>> > >> of DPDK reporting a number of available queues that for practical
>> > >> purposes is wrong?  I.E. Shouldn't this be fixed by the DPDK driver
>> of
>> > >> this system?  This patch feels like a hack to me . . .
>> > >>
>> > >> Ethan
>> > >>
>> > >> On Tue, Jul 28, 2015 at 2:36 AM, Stokes, Ian <ian.stokes at intel.com>
>> > >> wrote:
>> > >> > Hi all,
>> > >> >
>> > >> > Just wondering what the status of this patch is? Is there any
>> feedback
>> > >> > or queries we can answer to help?
>> > >> >
>> > >> > Thanks
>> > >> > Ian
>> > >> >
>> > >> >> -----Original Message-----
>> > >> >> From: dev [mailto:dev-bounces at openvswitch.org] On Behalf Of
>> Traynor,
>> > >> >> Kevin
>> > >> >> Sent: Thursday, July 23, 2015 11:28 AM
>> > >> >> To: Daniele Di Proietto; dev at openvswitch.org
>> > >> >> Subject: Re: [ovs-dev] [PATCH 2/2] netdev-dpdk: Retry tx/rx
>> queue
>> > >> setup
>> > >> >> until we don't get any failure.
>> > >> >>
>> > >> >>
>> > >> >> > -----Original Message-----
>> > >> >> > From: dev [mailto:dev-bounces at openvswitch.org] On Behalf Of
>> Daniele
>> > >> Di
>> > >> >> > Proietto
>> > >> >> > Sent: Thursday, July 16, 2015 7:48 PM
>> > >> >> > To: dev at openvswitch.org
>> > >> >> > Subject: [ovs-dev] [PATCH 2/2] netdev-dpdk: Retry tx/rx queue
>> setup
>> > >> >> until we
>> > >> >> > don't get any failure.
>> > >> >> >
>> > >> >> > It has been observed that some DPDK device (e.g intel xl710)
>> report
>> > >> an
>> > >> >> > high number of queues but make some of them available only for
>> > >> special
>> > >> >> > functions (SRIOV).  Therefore the queues will be counted in
>> > >> >> > rte_eth_dev_info_get(), but rte_eth_tx_queue_setup() will
>> fail.
>> > >> >> >
>> > >> >> > This commit works around the issue by retrying the device
>> > >> >> initialization
>> > >> >> > with a smaller number of queues, if a queue fails to setup.
>> > >> >> >
>> > >> >> > Reported-by: Ian Stokes <ian.stokes at intel.com>
>> > >> >> > Signed-off-by: Daniele Di Proietto <diproiettod at vmware.com>
>> > >> >> > ---
>> > >> >> >  lib/netdev-dpdk.c | 100
>> +++++++++++++++++++++++++++++++++++++++---
>> > >> ---
>> > >> >> -------
>> > >> >> > --
>> > >> >> >  1 file changed, 73 insertions(+), 27 deletions(-)
>> > >> >>
>> > >> >>
>> > >> >> Acked-by: Kevin Traynor <kevin.traynor at intel.com>
>> > >> >> _______________________________________________
>> > >> >> dev mailing list
>> > >> >> dev at openvswitch.org
>> > >> >> http://openvswitch.org/mailman/listinfo/dev
>> > >> > _______________________________________________
>> > >> > dev mailing list
>> > >> > dev at openvswitch.org
>> > >> > http://openvswitch.org/mailman/listinfo/dev



More information about the dev mailing list