[dpdk-dev] [PATCH] optimize vhost enqueue

[Wang, Zhihong]

Thanks Maxime and Yuanhan for your review and suggestions!
Please help review the v2 of this patch.


> -----Original Message-----
> From: Yuanhan Liu [mailto:yuanhan.liu at linux.intel.com]
> Sent: Wednesday, August 17, 2016 5:51 PM
> To: Maxime Coquelin <maxime.coquelin at redhat.com>
> Cc: Wang, Zhihong <zhihong.wang at intel.com>; dev at dpdk.org
> Subject: Re: [dpdk-dev] [PATCH] optimize vhost enqueue
> 
> On Wed, Aug 17, 2016 at 11:17:46AM +0200, Maxime Coquelin wrote:
> > >>>This is something I've thought about while writing the code, the reason I
> > >>>keep it as one function body is that:
> > >>>
> > >>> 1. This function is very performance sensitive, and we need full control of
> > >>>    code ordering (You can compare with the current performance with
> the
> > >>>    mrg_rxbuf feature turned on to see the difference).
> > >>
> > >>Will inline functions help?
> > >
> > >
> > >Optimization in this patch actually reorganizes the code from its logic,
> > >so it's not suitable for making separated functions.
> > >
> > >I'll explain this in v2.
> >
> > I agree with Yuanhan.
> > Inline functions should not break the optimizations.
> > IMHO, this is mandatory for the patch to be accepted.
> 
> Yes.
> 
> > It seems you are not the only one facing the issue:
> > https://github.com/YanVugenfirer/kvm-guest-drivers-windows/issues/70
> >
> > So a dedicated fix is really important.
> 
> Yes.
> 
> >
> > >This patch doesn't try to fix this issue, it rewrites the logic totally,
> > >and somehow fixes this issue.
> > >
> > >Do you think integrating this whole patch into the stable branch will work?
> > >Personally I think it makes more sense.
> >
> > No.
> > We don't even know why/how it fixes the Windows issue, which would be
> > the first thing to understand before integrating a fix in stable branch.
> 
> Yes.
> 
> >
> > And the stable branch is not meant for integrating such big reworks,
> > it is only meant to fix bugs.
> 
> Yes.
> 
> > The risk of regressions have to be avoided as much as possible.
> 
> Yes.
> 
> 	--yliu