* Russell King - ARM Linux [off-list ref] [160115 01:15]:

On Thu, Jan 14, 2016 at 11:40:25PM +0100, Sebastian Reichel wrote:

quoted

On Thu, Jan 14, 2016 at 05:25:49PM +0000, Russell King - ARM Linux wrote:

quoted

There are very good reasons not to do this: that will result in
configurations where MMC_OMAP_HS was set but without REGULATOR_PBIAS
ending up with MMC_OMAP_HS being disabled.  That doesn't help the
root problem, which is "why has the kernel boot regressed for my
previous working configuration?"

The solution proposed here adds a message to the boot which points
out fair and square what needs to be done to rectify the boot
failure.  Adding a dependency just brings up the question "where
has my MMC driver gone?"

The best thing would be to have no regression. Just printing a
message means I have to build another kernel. But more importantly
the message may not be visible by the user - e.g. if the display has
not yet been initialized.

I agree in principle, but that's not possible here (see Tony's mails
on why the PBIAS stuff needs to be optional.)  I'd agree with changing
the Kconfig if PBIAS were a hard and fast requirement, but it isn't.

Yeah. Probably the best solution for cases like this is to boot with
a small initrd. That way you can get usb console and networking going
before mounting the real root and you will see kernel messages with
only minimal things built in. You can build a small initramfs with
CONFIG_INITRAMFS_SOURCE that is appended to the kernel so you can
also boot the legacy user space too with that if needed.

Regards,

Tony