[Linux-stm32] [PATCH net-next v2 04/12] net: stmmac: Add Split Header support and enable it in XGMAC cores

[Jakub Kicinski]

On Tue, 13 Aug 2019 08:30:47 +0000, Jose Abreu wrote:
> ++ Jakub
> 
> From: David Miller <davem at davemloft.net>
> Date: Aug/12/2019, 22:06:18 (UTC+00:00)
> 
> > From: Jose Abreu <Jose.Abreu at synopsys.com>
> > Date: Mon, 12 Aug 2019 11:44:03 +0200
> >   
> > > 	- Add performance info (David)  
> > 
> > Ummm...
> > 
> > Whilst cpu utilization is interesting, I might be mainly interested in
> > how this effects "networking" performance.  I find it very surprising
> > that it isn't obvious that this is what I wanted.
> > 
> > Do you not do performance testing on the networking level when you
> > make fundamental changes to how packets are processed by the
> > hardware/driver?  
> 
> I do.
> 
> In my HW this feature does not impact performance neither improves it as 
> I'm already on max of theoretical bandwidth.

I think that'd be a good clarification in the commit message.
Like perhaps stating that the throughput is XYZ in both cases
(MTU is also worth clarifying).

> I do expect it to reduce CPU usage and memory footprint because we no 
> longer have to memcpy the entire buffer to SKB and instead we just copy 
> the header and then append payload as page which is passed directly to 
> net layer.
> 
> This feature is already used in some drivers and is part of GRO 
> offloading I think.
> 
> Jakub, as David is off can you please comment on how can we proceed with 
> this series ? I can add more information in commit log for this patch 
> ...

I had some suggestions/questions on the earlier patch, I think a respin
once those are resolved would be good.