On śro, 2014-07-02 at 12:13 +0200, Javier Martinez Canillas wrote:
> Hello Mike,
>
> On 07/01/2014 07:26 PM, Mike Turquette wrote:
> > Quoting Yadwinder Singh Brar (2014-06-29 21:01:36)
> >> Hi Javier,
> >>
> >> On Thu, Jun 26, 2014 at 11:45 PM, Javier Martinez Canillas
> >> <javier.martinez at collabora.co.uk> wrote:
> >> > Maxim Integrated Power Management ICs are very similar with
> >> > regard to their clock outputs. Most of the clock drivers for
> >> > these chips are duplicating code and are simpler enough that
> >> > can be converted to use a generic driver to consolidate code
> >> > and avoid duplication.
> >> >
> >> > Signed-off-by: Javier Martinez Canillas <javier.martinez at collabora.co.uk>
> >> > Reviewed-by: Krzysztof Kozlowski <k.kozlowski at samsung.com>
> >> > ---
> >> >
> >> > Changes since v4:
> >> > - Return recalc 0 if clock isn't enabled in Suggested by Yadwinder Singh Brar.
> >> >
> >>
> >> It seems you didn't implement or posted same patch again :) .
> >>
> >> > Changes since v3:
> >> > - Add current copyright information. Suggested by Krzysztof Kozlowski
> >> > - Do a single allocation for struct max_gen_clk. Suggested by Krzysztof Kozlowski
> >> > - Add EXPORT_SYMBOL() for exported symbols. Suggested by Krzysztof Kozlowski
> >> >
> >> > drivers/clk/Kconfig | 3 +
> >> > drivers/clk/Makefile | 1 +
> >> > drivers/clk/clk-max-gen.c | 195 ++++++++++++++++++++++++++++++++++++++++++++++
> >> > drivers/clk/clk-max-gen.h | 32 ++++++++
> >> > 4 files changed, 231 insertions(+)
> >> > create mode 100644 drivers/clk/clk-max-gen.c
> >> > create mode 100644 drivers/clk/clk-max-gen.h
> >> >
> >>
> >> [ .. ]
> >>
> >> > +
> >> > +static unsigned long max_gen_recalc_rate(struct clk_hw *hw,
> >> > + unsigned long parent_rate)
> >> > +{
> >> > + return 32768;
> >> > +}
> >>
> >> Its still same here.
> >
> > Changing this would be a new behavior. I do not know of any other clock
> > drivers that conditionally returns a rate of 0 based on whether or not
> > the clock is gated.
> >
>
> After Yadwinder feedback I searched for clock drivers that returned 0 when the
> clock was not enabled/prepared and found for example drivers/clk/clk-s2mps11.c:
>
>
> static unsigned long s2mps11_clk_recalc_rate(struct clk_hw *hw,
> unsigned long parent_rate)
> {
> struct s2mps11_clk *s2mps11 = to_s2mps11_clk(hw);
> if (s2mps11->enabled)
> return 32768;
> else
> return 0;
> }
>
> > It is also buggy since calls to clk_enable and clk_disable do not invoke
> > .recalc_rate, so the rate of your clock would not be updated from the
> > framework's perspective until some later point where you call
> > clk_set_rate or something.
> >
>
> s2ps11->enabled is set in the driver's clk_ops .prepare and .unprepare function
> handlers and calls to clk_prepare and clk_unprepare also don't seems to invoke
> .recalc_rate so I guess that driver is wrong as well and should just return the
> clock rate unconditionally?
The s2mps11 may be not a best example for proper clock driver :).
Karol Wrona already sent a patch for s2mps11:
https://lkml.org/lkml/2014/7/1/389
Best regards,
Krzysztof
>
> > If your driver needs to know whether or not the clock is enabled then we
> > could introduce a new bool clk_is_enabled(struct clk *clk); to clk.h,
> > but I'd rather not do that. Instead if a driver needs a clock then it
> > calls clk_enable on it without any knowledge about the internal state of
> > the clock enable_count.
> >
> > Regards,
> > Mike
> >
>
> Thanks a lot for the explanation, I'll revert that change then and return the
> clock rate unconditionally on the next version of the patch-set.
>
> Best regards,
> Javier