Re: [PATCH net-next 2/5] net: dsa: add out-of-band tagging protocol

[Maxime Chevallier]

Hello Florian,

On Fri, 22 Apr 2022 11:28:30 -0700
Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:

Thanks for the review :)

> On 4/22/22 11:03, Maxime Chevallier wrote:
> > This tagging protocol is designed for the situation where the link
> > between the MAC and the Switch is designed such that the Destination
> > Port, which is usually embedded in some part of the Ethernet
> > Header, is sent out-of-band, and isn't present at all in the
> > Ethernet frame.
> > 
> > This can happen when the MAC and Switch are tightly integrated on an
> > SoC, as is the case with the Qualcomm IPQ4019 for example, where
> > the DSA tag is inserted directly into the DMA descriptors. In that
> > case, the MAC driver is responsible for sending the tag to the
> > switch using the out-of-band medium. To do so, the MAC driver needs
> > to have the information of the destination port for that skb.
> > 
> > This tagging protocol relies on a new set of fields in skb->shinfo
> > to transmit the dsa tagging information to and from the MAC driver.
> > 
> > Signed-off-by: Maxime Chevallier <maxime.chevallier@xxxxxxxxxxx>  
> 
> First off, I am not a big fan of expanding skb::shared_info because
> it is sensitive to cache line sizes and is critical for performance
> at much higher speeds, I would expect Eric and Jakub to not be
> terribly happy about it.

No problem, I'm testing with the skb->cb approach as you suggested and
see how it goes.

> The Broadcom systemport (bcmsysport.c) has a mode where it can
> extract the Broadcom tag and put it in front of the actual packet
> contents which appears to be very similar here. From there on, you
> can have two strategies:
> 
> - have the Ethernet controller mangle the packet contents such that
> the QCA tag is located in front of the actual Ethernet frame and
> create a new tagging protocol variant for QCA, similar to the
> TAG_BRCM versus TAG_BRCM_PREPEND
> 
> - provide the necessary information for the tagger to work using an
> out of band mechanism, which is what you have done, in which case,
> maybe you can use skb->cb[] instead of using skb::shared_info?

One of the reason why I chose the second is to support possible future
cases where another controller would face a similar situation, and also
make use of the out-of-band tagger.

I understand that it's not very elegant in the sense that this breaks
the nice tagging model we have, but adding/removing data before the
payload also seems convoluted to achieve the same thing :) It seems
that this approach comes with a bit of an overhead since it implies
mangling the skb a bit, but I've yet to test this myself.

That's actually what I wanted your opinion on, it also seems like
Andrew likes the idea of putting the tag ahead of the frame to stick
with the actual model.

I don't have strong feelings myself on the way of doing this, I'm
looking for an approach that is efficient but yet easily maintainable.

Thanks,

Maxime