[PATCH 17/17] ARM: i.MX Phytec physom: Add deep-probe support

[Marco Felsch]

On 21-04-26 11:36, Ahmad Fatoum wrote:
> Hi,
> 
> On 26.04.21 11:20, Marco Felsch wrote:
> >> How about changing BAREBOX_DEEP_PROBE_ENABLE so BAREBOX_DEEP_PROBE_ENABLE(physom_imx6_match);
> >> works? Updated boards should be using board drivers anyway, so make it easier for them to
> >> enable deep probe.
> > 
> > Hi Ahmad,
> > 
> > Why do I need to chnage it to board drivers if I only wanna fix a probe
> > order issue? But you're right adding the _match table should be possible
> > to :)
> 
> Either you have board code at a single init level and changing to a board driver
> will be very straight-forward. Or you have multiple initcalls, which expect
> drivers at prior initcall levels to have probed a device, which will be broken
> by a move to deep probe.
> 
> >> I'd go even further and require board drivers by having it take a driver_d * as argument
> >> and introduce deep_probe_board_driver() that expands to postcore_platform_driver and
> >> BAREBOX_DEEP_PROBE_ENABLE.
> > 
> > How about this:
> >  1st) adding a probe_enum like kernel and introduce PROBE_PREFER_DEEP
> >  2nd) adding a deep_probe_enable_board() function to common/deep_probe.c
> >  3th) checking during register_driver() if the flag is set and call the
> >       deep_probe_enable_board().
> > 
> > This allows use to still use the macro mechanism and the boards already
> > converted to the board driver style can use this simple enum to mark all
> > boards as deep-probable.
> 
> I don't see how that would work. deep_probe_is_supported() is called on oftree
> registration, which is prior to the board driver's register_driver.

Arg.. you're right, didn't noticed that call during
barebox_register_of().

Regards,
  Marco