[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20220304081834.552ae666@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
Date: Fri, 4 Mar 2022 08:18:34 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: "Jonathan Lemon" <jonathan.lemon@...il.com>
Cc: netdev@...r.kernel.org, richardcochran@...il.com,
davem@...emloft.net, kernel-team@...com
Subject: Re: [PATCH net-next 1/2] ptp: ocp: add nvmem interface for
accessing eeprom
On Thu, 03 Mar 2022 21:39:48 -0800 Jonathan Lemon wrote:
> On 3 Mar 2022, at 21:01, Jakub Kicinski wrote:
> > On Thu, 3 Mar 2022 15:38:00 -0800 Jonathan Lemon wrote:
> >> manufacturer
> >
> > The generic string is for manufacture, i.e. fab; that's different
> > from manufacture*r* i.e. vendor. It's when you multi-source a single
> > board design at multiple factories.
>
> The documentation seems unclear:
>
> board.manufacture
> -----------------
> An identifier of the company or the facility which produced the part.
Yeah, so this is for standard NICs. Say you have a NIC made by
Chelsio (just picking a random company that's unlikely to have its
own fabs), the vendor is Chelsio but they will contract out building
the boards to whatever contractors. The contractor just puts the board
together and runs manufacturing tests, tho, no real IP work.
> There isn’t a board.vendor (or manufacturer) in devlink.h.
>
> The board design is open source, there’s several variants of
> the design being produced, so I’m looking for a simple way to
> identify the design (other than the opaque board id)
And all of them use Facebook PCI_ID, hm. But AFAIU the cards are not
identical, right? Are they using the same exact board design or
something derived from the reference board design that matches
the OCP spec?
And AFAIU the company delivering the card writes / assembles the
firmware, you can't take FW load from company A and flash it onto
company B's card, no?
Powered by blists - more mailing lists