On Mon, Feb 06, 2017 at 04:56:55PM -0800, Dmitry Torokhov wrote:

> I can definitely select the severity based on acpi_disabled flag, but I
> think you should reconsider: from what I have seen in the kernel quite a
> few driversi, even OF-specific ones, select regulator_get_optional() for
> no good reason other than to avoid this "scary" warning. If its severity
> were reduced quite a few drivers could be switched to normal
> regulator_get().

We had problems with people writing obviously problematic code long
before we had the warning (or optional regulators), I'm really
unconvinced that the warning has anything much to do with it.  There
will also be some code that predates dummy regulators, or where that
support was new enough that people weren't aware of it.  We get exactly
the same sort of problem with people writing constraints that are clear
nonsense (things like a supply named with a specific voltage but a large
voltage range).

> Candidates for conversion (not complete, just examples):

> drivers/input/touchscreen/zforce_ts.c
> drivers/input/touchscreen/tsc200x-core.c

These are broken, yes.  The zforce code at least is old enough that it
might predate dummy supplies.

> drivers/power/avs/rockchip-io-domain.c

This clearly cares about the voltages and could never work with a dummy
regulator, it is potentially a good use of optional supplies as far as I
can see (though I'm unsure about what an I/O domain without a reference
supply would mean).

> drivers/mfd/stmpe.c

This is again just obviously not good, it just logs but doesn't act on a
failure to enable the power supplies.