Re: Severe performance regression in "net: macsec: preserve ingress frame ordering"

On Tue, Aug 25, 2020 at 01:09:31PM +0000, Van Leeuwen, Pascal wrote:

quoted

-----Original Message-----
From: Andrew Lunn <andrew@lunn.ch>
Sent: Monday, August 24, 2020 3:02 PM
To: Van Leeuwen, Pascal <redacted>
Cc: Sabrina Dubroca <sd@queasysnail.net>; Scott Dial <redacted>; linux-crypto@vger.kernel.org; Ryan Cox
[off-list ref]; netdev@vger.kernel.org; davem@davemloft.net; Antoine Tenart [off-list ref];
ebiggers@google.com
Subject: Re: Severe performance regression in "net: macsec: preserve ingress frame ordering"

<<< External Email >>>
On Mon, Aug 24, 2020 at 09:07:26AM +0000, Van Leeuwen, Pascal wrote:

quoted

No need to point this out to me as we're the number one supplier of inline MACsec IP :-)
In fact, the Microsemi PHY solution you mention is ours, major parts of that design were
even created by these 2 hands here.

Oh,  O.K.

Do you know of other silicon vendors which are using the same IP?

I do, there are many. But unfortunately, I cannot disclose our customers unless this is already
public information, e.g. due to some press release or whatever.

O.K. Maybe i should flip the question around. If somebody was to
submit a driver, how would i quickly determine it is your IP? Any
particularly patterns i should look for.

quoted

Maybe we can encourage them to share the driver, rather than re-invent
the wheel, which often happens when nobody realises it is basically
the same core with a different wrapper.

Yes, that could save a lot of duplication of code and effort.

It would save a lot of effort. But not code duplication. Because if i
or one of the other maintainers notices it is just your IP with a
different wrapper, we would NACK the patch and tell them to refactor
the MSCC driver. There is a long established precedence for that.

The problem is: who will do it? We can't do it, because we have no
access to the actual HW.

Microsemi are very friendly. If you ask them, i'm sure they would send
you a board. I assume you also have some sort of FPGA setup you use
for your own testing? That gives you two platforms. And if there are
many PHYs using your IP, it should not be too hard to just go buy a
reference design kit from a vendor.

And there is the marketing aspect for Rambus. You can say your IP is
easy to use, the core code is already in the kernel, supported and
well tested, you just need to add a few wrapper functions in your
driver. No vendor crap driver needed.

	Andrew