Re: [net-next] net: dsa: qca8k: only change the MIB_EN bit in MODULE_EN register

On Tue, Nov 9, 2021 at 12:38 AM Vladimir Oltean [off-list ref] wrote:

On Mon, Nov 08, 2021 at 11:13:30PM +0100, Robert Marko wrote:

quoted

quoted

The driver keeps state. If the switch just resets by itself, what do you
think will continue to work fine afterwards? The code path needs testing.
I am not convinced that a desynchronized software state is any better
than a lockup.

It's really unpredictable, as QCA doesn't specify what does the software reset
actually does, as I doubt that they are completely resetting the
switch to HW defaults.
But since I was not able to trigger the QM error and the resulting
reset, it's hard to tell.
Phylink would probably see the ports going down and trigger the MAC
configuration again,
this should at least allow using the ports and forwarding to CPU again.
However, it may also reset the forwarding config to basically flooding
all ports which is the default
which is not great.

But I do agree that it may not be a lot better than a lockup.

I'm not sure what you expect going forward. You haven't proven an issue
with the actual code structure, or an improvement brought by your change.
Allowing the hardware to autonomously reconfigure itself, even if
partially, is out of the question (of course, that's if and only if I
understand correctly the info that you've presented).

After this discussion, I think that John clears the bits intentionally,
I still don't think its really the best practice to do so in the MIB enablement
without documenting it.
However, since this doesn't seem to be hurting anyone I will drop it and rather
focus on IPQ4019 support which is slowly shaping into something upstreamable.

Regards,
Robert

-- 
Robert Marko
Staff Embedded Linux Engineer
Sartura Ltd.
Lendavska ulica 16a
10000 Zagreb, Croatia
Email: robert.marko@sartura.hr
Web: www.sartura.hr