Re: [PATCH] net: phy: broadcom: Wire suspend/resume for BCM54612E

[Andrew Lunn]

On Fri, Nov 03, 2023 at 02:47:38AM +0100, Marco von Rosenberg wrote:
> On Wednesday, November 1, 2023 11:06:56 PM CET Andrew Lunn wrote:
> > On Wed, Nov 01, 2023 at 10:42:52PM +0100, Marco von Rosenberg wrote:
> > > On Tuesday, October 31, 2023 1:31:11 AM CET Andrew Lunn wrote:
> > > > Are we talking about a device which as been suspended? The PHY has
> > > > been left running because there is no suspend callback? Something then
> > > > triggers a resume. The bootloader then suspends the active PHY? Linux
> > > > then boots, detects its a resume, so does not touch the hardware
> > > > because there is no resume callback? The suspended PHY is then
> > > > useless.
> > > 
> > > Hi Andrew,
> > > 
> > > thanks for your feedback. I guess a bit of context is missing here. The
> > > issue has nothing to do with an ordinary suspension of the OS. The main
> > > point is that on initial power-up, the bootloader suspends the PHY before
> > > booting Linux. With a resume callback defined, Linux would call it on
> > > boot and make the PHY usable.
> > 
> > Ah, so you rely on phy_attach_direct() calling phy_resume(phydev).
> > 
> > This seems an odd way to solve the problem. It was not Linux which
> > suspend the PHY, so using resume is asymmetric.
> > 
> > I think soft_reset() or config_init() should be taking the PHY out of
> > suspend.
> 
> I agree with all of your points. This is just one way which happens to solve
> this specific problem. Of course it might be asymmetric to see the patch as
> a solution to my problem. However is there anything fundamentally wrong with
> adding suspend/resume callbacks?

No, there is nothing wrong with that at all, if you want to support
suspend/resume. I do however see that as a different use case to what
you describe as your problem. It fixing your problem is more of a side
effect.

We can go with this fix, but please change your justification in the
commit message. Also, its unlikely, but resume could be made
conditional in phy_attach_direct(), and you would then be back to a
broken PHY on boot. Fixing this in config_init() is the correct way
for your use case.

       Andrew