Re: [PATCH net-next] MIPS: generic: Add switchdev, pinctrl and fit to ocelot_defconfig

[Horatiu Vultur]

Hi Paul,

The 04/03/2019 23:23, Paul Burton wrote:
> External E-Mail
> 
> 
> Hi Horatiu,
> 
> On Wed, Apr 03, 2019 at 05:27:36PM +0200, Horatiu Vultur wrote:
> > diff --git a/arch/mips/configs/generic/board-ocelot.config b/arch/mips/configs/generic/board-ocelot.config
> > index f607888..3215741 100644
> > --- a/arch/mips/configs/generic/board-ocelot.config
> > +++ b/arch/mips/configs/generic/board-ocelot.config
> >%
> > +# CONFIG_HID is not set
> > +# CONFIG_USB_SUPPORT is not set
> > +# CONFIG_VIRTIO_MENU is not set
> > +# CONFIG_SCSI is not set
> 
> Unfortunately this part won't work so well. If board-ocelot.config
> disables these things, then what should happen if another board that's
> also included in a generic kernel enables them?
> 
> eg. if you run 'make ARCH=mips 32r2el_defconfig' then we merge all of
> the following:
> 
>   board-boston.config enables USB
>   board-sead-3.config enables USB
>   board-ocelot.config disables USB

I didn't think about this scenario, because I didn't expect that
building a generic configuration will bring together all the board
configurations.

Anyway, I will send a new patch in which I will remove these
configurations.

> 
> These are mutually exclusive, and it seems that on my system we
> currently end up disabling USB due to board-ocelot.config. That will of
> course break USB support for Boston or SEAD-3 which are also supported
> by the same kernel binary. In practice which one 'wins' will depend on
> the order the files are listed by make's wildcard function - so far as
> I'm aware that doesn't guarantee any particular order so if it ends up
> depending on the order the filesystem lists the files or something like
> that then configurations might even differ when used on different
> machines.
> 
> So to avoid that the best we can do is leave these enabled and the
> general rule is that board-*.config files can only enable extra things,
> not disable them.
> 
> You might be tempted to disable the options in generic_defconfig &
> update any board configs that actually need them to enable them, but
> that doesn't work too well for things which are 'default y' because
> kconfig then warns about the conflict between generic_defconfig & the
> board config being merged with it. That applies to the first 3 of the
> entries you disable, leaving only CONFIG_SCSI that could potentially be
> dealt with that way...
> 
> Thanks,
>     Paul
> 

-- 
/Horatiu