Re: [PATCH net-next 2/3] dt-bindings: net: micrel: Configure latency values and timestamping check for LAN8814 phy

[Allan W. Nielsen]

On Sun, Mar 13, 2022 at 04:07:24PM +0100, Andrew Lunn wrote:
> On Sat, Mar 12, 2022 at 06:46:46PM -0800, Richard Cochran wrote:
> > On Sat, Mar 12, 2022 at 09:04:31PM +0100, Andrew Lunn wrote:
> > > Do these get passed to the kernel so the hardware can act on them, or
> > > are they used purely in userspace by ptp4l?
> >
> > user space only.
I'm wondering if one-step will work if these correction values are not
applied to HW.

> > > If they has passed to the kernel, could we provide a getter as well as
> > > a setter, so the defaults hard coded in the driver can be read back?
> >
> > Any hard coded defaults in the kernel are a nuisance.
> >
> > I mean, do you want user space to say,
> >
> >    "okay, so I know the correct value is X.  But the drivers may offer
> >    random values according to kernel version.  So, I'll read out the
> >    driver value Y, and then apply X-Y."
> >
> > Insanity.
> 
> No, i would not suggests that at all.
> 
> You quoted the man page and it says the default it zero. If there was
> an API to ask the driver what correction it is doing, and an API to
> offload the delay correction to the hardware, i would simply remove
> the comment about the default being zero. If these calls return
> -EOPNOTSUPP, then user space stays the same, and does actually use a
> default of 0. If offload is supported, you can show the user the
> current absolute values, and allow the user to set the absolute
> values.
This sounds like a good approach to me (but I know it is not my opinion
you are asking for).

In all cases, if there is a desire to have such APIs, and let drivers
advertise default compensation values in this way, we can work on that.

> Anyway, it is clear you don't want the driver doing any correction, so
> lets stop this discussion.
> 
>      Andrew

-- 
/Allan