Re: [PATCH net-next v13 1/2] ptp: ocp: add TAP CPLD access for ADVA TimeCard X1

[Sagi Maimon <maimon.sagi@gmail.com>]

On Fri, 11 Sep 2026 16:52:43 -0700 Jakub Kicinski wrote:
> > +               done=<0|1> busy=<0|1> failed=<0|1>  
> 
> I think the sysfs rules are "one value per file"
> We can stretch it a bit and say that this file contains flags
> 
> So format like done or busy or failed or multiple eg "busy failed"
> But the X=[0|1] is too much

Changed for v14.  The file now contains the names of the flags that are
set, space separated, and an empty line when none is:

  # cat cpld_status
  done

  # cat cpld_status     # during programming
  busy

The ABI entry lists the three names and what each means.

> Also I think the AI is right this one commit does too many things.

Split for v14.  The two pre-existing fixes that were buried in it are
separate patches now, with Fixes: tags:

  1/4 ptp: ocp: unregister devlink before detach on probe error
  2/4 ptp: ocp: fix dpll cleanup on probe error
  3/4 ptp: ocp: add TAP CPLD access for ADVA TimeCard X1
  4/4 ptp: ocp: add TAP CPLD flashing via devlink

2/4 is a second bug in the same error path, found while looking at the
first: out_dpll never called dpll_device_unregister(), so a failing
dpll_pin_get() or dpll_pin_register() left the dpll device registered
with a priv pointer into storage devlink_free() releases a few lines
later, and a failing dpll_device_register() leaked the reference
dpll_device_get() had taken.

1 and 2 fix 09eeb3aecc6c, which is in released kernels, but they only
trigger when probe fails, so I did not mark them for stable.  They are in
this series because 3/4 adds a mutex that ptp_ocp_detach() destroys, so
devlink has to be unregistered first.

If you would rather see 3/4 split further - the bus arbitration alone,
then the two read-only interfaces on top - I am happy to do that.

v14 posted at: https://lore.kernel.org/netdev/20260914071531.11640-1-maimon.sagi@gmail.com/T/#u

Thanks for the review,
Sagi