Re: [PATCH net-next v2 3/4] mlxbf_gige: add BlueField-3 Serdes configuration

[Saeed Mahameed <saeed@kernel.org>]

On 11 Nov 21:34, Jakub Kicinski wrote:
>On Thu, 10 Nov 2022 14:33:47 +0100 Andrew Lunn wrote:
>> On Wed, Nov 09, 2022 at 05:47:51PM -0500, David Thompson wrote:
>> > The BlueField-3 out-of-band Ethernet interface requires
>> > SerDes configuration. There are two aspects to this:
>> >
>> > Configuration of PLL:
>> >     1) Initialize UPHY registers to values dependent on p1clk clock
>> >     2) Load PLL best known values via the gateway register
>> >     3) Set the fuses to tune up the SerDes voltage
>> >     4) Lock the PLL
>> >     5) Get the lanes out of functional reset.
>> >     6) Configure the UPHY microcontroller via gateway reads/writes
>> >
>> > Configuration of lanes:
>> >     1) Configure and open TX lanes
>> >     2) Configure and open RX lanes
>>
>> I still don't like all these black magic tables in the driver.
>>
>> But lets see what others say.
>
>Well, the patch was marked as Changes Requested so it seems that DaveM
>concurs :) (I'm slightly desensitized to those tables because they
>happen in WiFi relatively often.)
>
>The recommendation is to come up with a format for a binary file, load
>it via FW loader and then parse in the kernel?

By FW loader you mean request_firmware() functionality ?

I am not advocating for black magic tables of course :), but how do we
avoid them if request_firmware() will be an overkill to configure such a
simple device? Express such data in a developer friendly c structures
with somewhat sensible field names?

>
>We did have a recommendation against parsing FW files in the kernel at
>some point, too, but perhaps this is simple enough to pass.
>
>Should this be shared infra? The problem is fairly common.

Infrastructure to parse vendor Firmware ? we can't get vendors to agree on
ethtool interface, you want them to agree on one firmware format :)?

BTW i don't think the issue here is firmware at all, this is device
specific config space.