Re: [PATCH net-next v2 2/9] phy: add hwtstamp_get callback to phy drivers

["Russell King (Oracle)" <linux@armlinux.org.uk>]

On Mon, Nov 17, 2025 at 05:39:13PM +0000, Vadim Fedorenko wrote:
> On 13/11/2025 12:24, Russell King (Oracle) wrote:
> > On Thu, Nov 13, 2025 at 12:12:44PM +0000, Vadim Fedorenko wrote:
> > > On 13/11/2025 12:02, Russell King (Oracle) wrote:
> > > > On Thu, Nov 13, 2025 at 11:32:00AM +0000, Vadim Fedorenko wrote:
> > > > > PHY devices had lack of hwtstamp_get callback even though most of them
> > > > > are tracking configuration info. Introduce new call back to
> > > > > mii_timestamper.
> > > > > 
> > > > > Signed-off-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
> > > > 
> > > > As part of my Marvell PTP work, I have a similar patch, but it's
> > > > way simpler. Is this not sufficient?
> > > > 
> > > > __phy_hwtstamp_get() is called via phylib_stubs struct and
> > > > phy_hwtstamp_get(), dev_get_hwtstamp_phylib(), dev_get_hwtstamp(),
> > > > and dev_ifsioc().
> > > > 
> > > > Using the phylib ioctl handler means we're implementing a path that
> > > > is already marked as legacy - see dev_get_hwtstamp():
> > > > 
> > > >           if (!ops->ndo_hwtstamp_get)
> > > >                   return dev_eth_ioctl(dev, ifr, SIOCGHWTSTAMP); /* legacy */
> > > > 
> > > > So, I think the below would be the preferred implementation.
> > > 
> > > You mean do not add SIOCGHWTSTAMP case in phy_mii_ioctl() as we should
> > > never reach this legacy option?
> > 
> > We _can_ reach phy_mii_ioctl() for SIOCGHWTSTAMP where drivers do not
> > provide the ndo_hwtstamp_get() method. However, as this is legacy code,
> > the question is: should we add it?
> > 
> > > Technically, some drivers are (yet) not
> > > converted to ndo_hwtstamp callbacks and this part can potentially work
> > > for bnx2x driver, until the other series lands.
> > 
> > Right, but providing new features to legacy paths gives less reason for
> > people to stop using the legacy paths.
> > 
> > > I was planning to remove SIOCSHWTSTAMP/SIOCGHWTSTAMP dev_eth_ioctl calls
> > > later once everything has landed and we have tests confirming that ioctl
> > > and netlink interfaces work exactly the same way.
> > 
> > However, implementations that do populate non-legacy ndo_hwtstamp_get()
> > won't work correctly with your conversion, since we'll fall through to
> > the path which calls __phy_hwtstamp_get() which won't do anything.
> > 
> > So I disagree with your patch - it only adds support for legacy net
> > drivers to get the hwtstamp settings from the PHY. Non-legacy won't be
> > supported.
> > 
> > At minimum, we should be adding support for non-legacy, and _possibly_
> > legacy.
> > 
> > Let's wait for others to comment on my point about adding this for the
> > legacy drivers/code path.
> As there was no conversation for a couple of days, and Andrew expressed
> the same thought about not going to implement SIOCGHWTSTAMP, I'm going
> to publish new version with this part removed.

Let's see the code, because I stopped engaging with you because I
couldn't understand what you were saying.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!