lkml.org 
[lkml]   [2022]   [Jul]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH net-next 1/1] net: dsa: mv88e6xxx: allow reading FID when handling ATU violations
On 2022-07-07 12:28, Vladimir Oltean wrote:
> On Wed, Jul 06, 2022 at 02:25:02PM +0200, Hans Schultz wrote:
>> For convenience the function mv88e6xxx_g1_atu_op() has been used to
>> read
>> ATU violations, but the function has other purposes and does not
>> enable
>> the possibility to read the FID when reading ATU violations.
>>
>> The FID is needed to get hold of which VID was involved in the
>> violation,
>> thus the need for future purposes to be able to read the FID.
>
> Make no mistake, the existing code doesn't disallow reading back the
> FID
> during an ATU Get/Clear Violation operation, and your patch isn't
> "allowing" something that wasn't disallowed.

It would only read 0 the way it worked. And I don't understand why
mv88e6xxx_g1_atu_op() writes the FID?

>
> The documentation for the ATU FID register says that its contents is
> ignored before the operation starts, and it contains the returned ATU
> entry's FID after the operation completes.
>
> So the change simply says: don't bother to write the ATU FID register
> with zero, it doesn't matter what this contains. This is probably true,
> but the patch needs to do what's written on the box.

Writing 0 to the ATU fID register resulted in a read giving zero of
course.

>
> Please note that this only even matters at all for switches with
> mv88e6xxx_num_databases(chip) > 256, where MV88E6352_G1_ATU_FID is a
> dedicated register which this patch avoids writing. For other switches,
> the FID is embedded within MV88E6XXX_G1_ATU_CTL or MV88E6XXX_G1_ATU_OP.
> So _practically_, for those switches, you are still emitting the
> GET_CLR_VIOLATION ATU op with a FID of 0 whether you like it or not,
> and
> this patch introduces a (most likely irrelevant) discrepancy between
> the
> access methods for various switches.
>
> Please note that this observation is relevant for your future changes
> to
> read back the FID too. As I said here:
> https://patchwork.kernel.org/project/netdevbpf/patch/20220524152144.40527-4-schultz.hans+netdev@gmail.com/#24912482
> you can't just assume that the FID lies within the MV88E6352_G1_ATU_FID
> register, just look at the way it is packed within
> mv88e6xxx_g1_atu_op().
> You'll need to unpack it in the same way.

So I need a new function to read the FID that mimics
mv88e6xxx_g1_atu_op()
as far as I understand?

\
 
 \ /
  Last update: 2022-07-08 09:14    [W:0.069 / U:0.392 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site