[ovs-dev] 答复: [RFC PATCH] netdev-dpdk: Narrow down txq critical section.

Li,Rongqing lirongqing at baidu.com
Wed Dec 4 02:03:38 UTC 2019



> -----邮件原件-----
> 发件人: David Marchand [mailto:david.marchand at redhat.com]
> 发送时间: 2019年12月3日 20:40
> 收件人: Ilya Maximets <i.maximets at ovn.org>
> 抄送: ovs dev <dev at openvswitch.org>; Ian Stokes <ian.stokes at intel.com>;
> Li,Rongqing <lirongqing at baidu.com>
> 主题: Re: [RFC PATCH] netdev-dpdk: Narrow down txq critical section.
> 
> On Tue, Dec 3, 2019 at 12:17 PM Ilya Maximets <i.maximets at ovn.org> wrote:
> >
> > On 02.12.2019 17:24, David Marchand wrote:
> > > tx_lock protects the NIC/vhost queue from concurrent access.
> > > Move it closer to the parts it protects and let packets duplication
> > > (when source is not DPDK) and the egress policer run out of it.
> > >
> > > Signed-off-by: David Marchand <david.marchand at redhat.com>
> > > ---
> > > I caught this by code review, but I imagine this could make the
> > > contention on the tx lock even worse if broadcasting packets.
> >
> > Idea is OK, I think.  I didn't look a t the code, but, IIRC, there was
> > already a similar patch in a mail-list.. Hard to find.
> 
> Ah indeed, I found this:
> https://mail.openvswitch.org/pipermail/ovs-dev/2019-January/355743.html
> But I did not find it in patchwork, not sure why, copied its author.
> 
> My RFC patch goes one step further, since it also moves the allocations out of
> the lock.
> 

you can continue your RFC , and I have no time to rework this patch

And I support your RFC, it maybe unable to benchmark, but always give little improvement.

-LI
> 
> --
> David Marchand



More information about the dev mailing list