Re: [PATCH] net: Select PTP_1588_CLOCK in PTP-specific drivers

[Arnd Bergmann]

On Tue, Apr 28, 2020 at 11:21 AM Geert Uytterhoeven
<geert@xxxxxxxxxxxxxx> wrote:
> On Tue, Apr 28, 2020 at 11:14 AM Clay McClure <clay@xxxxxxxxxxx> wrote:
> > Commit d1cbfd771ce8 ("ptp_clock: Allow for it to be optional") changed
> > all PTP-capable Ethernet drivers from `select PTP_1588_CLOCK` to `imply
> > PTP_1588_CLOCK`, "in order to break the hard dependency between the PTP
> > clock subsystem and ethernet drivers capable of being clock providers."
> > As a result it is possible to build PTP-capable Ethernet drivers without
> > the PTP subsystem by deselecting PTP_1588_CLOCK. Drivers are required to
> > handle the missing dependency gracefully.
> >
> > Some PTP-capable Ethernet drivers (e.g., TI_CPSW) factor their PTP code
> > out into separate drivers (e.g., TI_CPTS_MOD). The above commit also
> > changed these PTP-specific drivers to `imply PTP_1588_CLOCK`, making it
> > possible to build them without the PTP subsystem. But as Grygorii
> > Strashko noted in [1]:
> >
> > On Wed, Apr 22, 2020 at 02:16:11PM +0300, Grygorii Strashko wrote:
> >
> > > Another question is that CPTS completely nonfunctional in this case and
> > > it was never expected that somebody will even try to use/run such
> > > configuration (except for random build purposes).
> >
> > In my view, enabling a PTP-specific driver without the PTP subsystem is
> > a configuration error made possible by the above commit. Kconfig should
> > not allow users to create a configuration with missing dependencies that
> > results in "completely nonfunctional" drivers.
> >
> > I audited all network drivers that call ptp_clock_register() and found
> > six that look like PTP-specific drivers that are likely nonfunctional
> > without PTP_1588_CLOCK:
> >
> >     NET_DSA_MV88E6XXX_PTP
> >     NET_DSA_SJA1105_PTP
> >     MACB_USE_HWSTAMP
> >     CAVIUM_PTP
> >     TI_CPTS_MOD
> >     PTP_1588_CLOCK_IXP46X
> >
> > Note how they all reference PTP or timestamping in their name; this is a
> > clue that they depend on PTP_1588_CLOCK.
> >
> > Change these drivers back [2] to `select PTP_1588_CLOCK`. Note that this
> > requires also selecting POSIX_TIMERS, a transitive dependency of
> > PTP_1588_CLOCK.
>
> If these drivers have a hard dependency on PTP_1588_CLOCK, IMHO they
> should depend on PTP_1588_CLOCK, not select PTP_1588_CLOCK.

Agreed. Note that for drivers that only optionally use the PTP_1588_CLOCK
support, we probably want 'depends on PTP_1588_CLOCK ||
!PTP_1588_CLOCK' (or the syntax replacing it eventually), to avoid the
case where a built-in driver fails to use a modular ptp implementation.

     Arnd