On Wed, Jul 22, 2015 at 10:31:34PM +0800, Henry Chen wrote:
> On Tue, 2015-07-21 at 18:25 +0100, Mark Brown wrote:

> > OK, so the issue here is that when we fall back to regmap_read() we may
> > do so because we have reg_read() and reg_write() functions which in turn
> > imply no formatting.  The expectation here is that val must be an array
> > of int.  The code doesn't completely take that into account though and
> > the user you're pointing at is assuming it's an array of 16 bit values
> > which isn't totally unreasonable if it did specify val_bits (we don't
> > check for that).

> So, could I call regmap_bulk_read() on rtc-mt6307.c, should I need to
> change it ?

It should be fine but you may need to change to pass an array of
unsigned int instead of an array of u16 in.

> > > Maybe it was not the good fix for this, but should be a problem need to
> > > be reported, or should I need to give the regmap_bus on mtk_pmic_wrap.c?

> > That file isn't in mainline...

> oh...it's mtk-pmic-wrap.c, sorry about that.

Found it - thanks.

> > memcpy() is definitely not a safe way to move from an unsigned int to a
> > u16 which is what your specific use case is trying to do.  I'll need to
> > do an audit of existing users (or someone else will!) to figure out what
> > people are doing with .val_bits in drivers using reg_read() and
> > reg_write() but I think what we should be doing here is probably
> > providing appropriate conversion functions based on val_bits on init.

> Ok, got it, memcpy() should not be used here anymore.

Right.  We just need to do a survey of existing users and figure out
what the least disruptive format function to provide is.  That way we
don't have to special case other code that uses formatting.