Prevent registration of duplicate "struct regulator" names.
They'd be unavailable, and clearly indicate something wrong.

[Edited to remove check for NULL consumer device until we have a
solution for things like cpufreq -- broonie]

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

There are some minor textual changes in here as well, mostly to enable()
and disable() but the primary goal of these changes is to fix
misrenderings of the kerneldoc documentation for the regulator API.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

This is only the documentation that the kerneldoc system warns about.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Remove kerneldoc warnings that don't relate to missing documentation,
mostly by renaming parameters in the documentation to match their
actual names.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Some of the internal structures have no kerneldoc but the ** at the start
of the comment marking them for documentation.  Remove the annotation
until some is added.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Calling set_machine_constraints before regulator device initialization
causes crash when constraints have apply_uV set.

Signed-off-by: Mike Rapoport <mike@compulab.co.il>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Clean up the sysfs interface to regulators by only exposing the
attributes that can be properly displayed.  For example: when a
particular regulator method is needed to display the value, only
create that attribute when that method exists.

This cleaned-up interface is much more comprehensible.  Most
regulators only support a subset of the possible methods, so
often more than half the attributes would be meaningless.  Many
"not defined" values are no longer necessary.  (But handling
of out-of-range values still looks a bit iffy.)

Documentation is updated to reflect that few of the attributes
are *always* present, and to briefly explain why a regulator may
not have a given attribute.

This adds object code, about a dozen bytes more than was removed
by the preceding patch, but saves a bunch of per-regulator data
associated with the now-removed attributes.  So there's a net
reduction in memory footprint.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Shrink regulator core by removing duplication in attribute printing
and probe() cleanup paths.  Saves about 340 bytes (object) on ARM.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Minor bugfixes in handling of regulator modes:

 - have the routine verifying regulator modes check against
   the set of legal modes (!);

 - have regulator_set_optimum_mode() verify the return value
   of regulator_ops.get_optimum_mode(), like drms_uA_update();

 - one call to regulator_ops.set_mode() treated zero as a
   failure code; make this consistent with other callers.

Both regulator_set_mode() and regulator_set_optimum_mode() now
require valid_ops_mask to include REGULATOR_CHANGE_MODE; that
seems like a bugfix too.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>

Make the <linux/regulator.h> framework treat enable/disable call
pairs like the <linux/clk.h> and <linux/interrupt.h> frameworks do:
they're refcounted, so that different parts of a driver don't need
to put work into coordination that frameworks normally handle.
It's a minor object code shrink.

It also makes the regulator_is_disabled() kerneldoc say what it's
actually returning:  return value is not a refcount, and may report
an error (e.g. I/O error from I2C).

It also fixes some minor regulator_put() goofage:  removing unlocked
access to the enable state.  (But still not making regulator put/get
match the refcounting pattern they invoke.)

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>