Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

Sorry for the delayed reply, I tested this patch yesterday, found that this is not right fix, Turns out, dp_netdev_reset_pmd_threads() just creates (or restores) pmd threads... and we call dp_netdev_destroy_all_pmds() at the beginning of the function. So, when purge_cb() is called, the flows in

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

On 17.08.2015 20:35, Alex Wang wrote: > > > On Mon, Aug 17, 2015 at 10:22 AM, Daniele Di Proietto > wrote: > > The patch looks good to me. > > I think that losing some packets when the thread are reconfigured > is unavoidable, considering that often

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

On 17 August 2015 at 10:22, Daniele Di Proietto wrote: > The patch looks good to me. > > I think that losing some packets when the thread are reconfigured > is unavoidable, considering that often the NIC must be stopped > to change the number of rxqs. Does this make sense? > > Related to this, I'

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

On Mon, Aug 17, 2015 at 10:22 AM, Daniele Di Proietto < diproiet...@vmware.com> wrote: > The patch looks good to me. > > I think that losing some packets when the thread are reconfigured > is unavoidable, considering that often the NIC must be stopped > to change the number of rxqs. Does this mak

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

The patch looks good to me. I think that losing some packets when the thread are reconfigured is unavoidable, considering that often the NIC must be stopped to change the number of rxqs. Does this make sense? Related to this, I've had an offline discussion with Ethan and we realized that ukeys a

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

Alex, I have lost at least 32 packets while testing this patch yesterday(~15 configuration changes). Is it bad? What is worse: loosing of ~1-2 bursts of packets or a little stats about threads? Best regards, Ilya Maximets. On 12.08.2015 23:09, Alex Wang wrote: > > > On Wed, Aug 12, 2015 at 1:

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

On Wed, Aug 12, 2015 at 1:10 AM, Ilya Maximets wrote: > Ok, but why don't just put dp->dp_purge_cb(dp->dp_purge_aux) right after > dp_netdev_destroy_all_pmds(dp) in dpif_netdev_pmd_set() ? In that case, > disabling of upcalls will not be necessary. > > Hey Ilya, Calling it before 'dp_netdev_dest

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

Ok, but why don't just put dp->dp_purge_cb(dp->dp_purge_aux) right after dp_netdev_destroy_all_pmds(dp) in dpif_netdev_pmd_set() ? In that case, disabling of upcalls will not be necessary. P.S. The second letter of my name is 'L'. On 11.08.2015 21:06, Alex Wang wrote: > Thx, IIya for the review,

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

Thx, IIya for the review, please see my reply below, On Tue, Aug 11, 2015 at 4:41 AM, Ilya Maximets wrote: > On 11.08.2015 05:47, Alex Wang wrote: > > When dpdk configuration changes, all pmd threads are recreated > > and rx queues of each port are reloaded. After this process, > > rx queue c

Re: [ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

On 11.08.2015 05:47, Alex Wang wrote: > When dpdk configuration changes, all pmd threads are recreated > and rx queues of each port are reloaded. After this process, > rx queue could be mapped to a different pmd thread other than > the one before reconfiguration. However, this is totally > transp

[ovs-dev] [PATCH 1/2] dpif-netdev: Purge all ukeys when reconfigure pmd.

When dpdk configuration changes, all pmd threads are recreated and rx queues of each port are reloaded. After this process, rx queue could be mapped to a different pmd thread other than the one before reconfiguration. However, this is totally transparent to ofproto layer modules. So, if the ofpr