Re: [PATCH 3/3] clk: samsung: allow runtime PM with auto clock gating

["André Draszik" <andre.draszik@linaro.org>]

Hi Krzysztof,

On Sat, 2026-01-17 at 20:25 +0100, Krzysztof Kozlowski wrote:
> On 09/01/2026 18:27, André Draszik wrote:
> > When automatic clock gating is enabled, runtime PM (RPM) isn't entered
> > even if enabled for a CMU if a sysreg clock exists and is provided by
> > this CMU (as is generally the case).
> > 
> > The reason is that this driver acquires a CMU's sysreg registers using
> > syscon_regmap_lookup_by_phandle() which ends up preparing the sysreg
> > clock. Given the sysreg clock is provided by this CMU, this CMU's usage
> > count is therefore bumped and RPM can not be entered as this CMU never
> > becomes idle.
> > 
> > Switch to using device_node_to_regmap() which doesn't handle resources
> > (the clock), leaving the CMU's usage count unaffected.
> > 
> > Note1: sysreg clock handling is completely removed with this commit
> 
> I miss where do you remove in this commit the sysreg clock handling?

The CMU driver originally used syscon_regmap_lookup_by_phandle() for sysreg,
which does clk_get() and clk_prepare(), and then implicitly and for each
register access clk_enable()/disable() for the clock specified in DT for the
(sysreg) node.

This commit changes it to using device_node_to_regmap(), which does nothing
with the clock (or any other resources). I opted to have the CMU driver still
do a clk_get(), so that the relationship is still visible, e.g. in
$debugfs/clk/clk_summary.

> > because sysreg register access is only required during suspend/resume.
> > In the runtime suspend case, we would have to enable the clock to read
> > the registers, but we can not do that as that would cause a resume of
> > this driver which is not allowed. This is not a problem because we
> > would only need to handle the clock manually if automatic clock gating
> > wasn't enabled in the first. This code is only relevant if automatic
> > clock gating is enabled, though.
> > 
> > Fixes: 298fac4f4b96 ("clk: samsung: Implement automatic clock gating mode for CMUs")
> > Signed-off-by: André Draszik <andre.draszik@linaro.org>
> > ---
> >  drivers/clk/samsung/clk.c | 92 +++++++++++++++++++++++++++++++++++------------
> >  drivers/clk/samsung/clk.h |  2 ++
> >  2 files changed, 71 insertions(+), 23 deletions(-)
> > 
> > diff --git a/drivers/clk/samsung/clk.c b/drivers/clk/samsung/clk.c
> > index 9f68f079fd552f8dfb6898dbfb47dec0e84c626c..6515df81fcbc79b90f5262843e67575f6a4e0dda 100644
> > --- a/drivers/clk/samsung/clk.c
> > +++ b/drivers/clk/samsung/clk.c
> > @@ -9,11 +9,13 @@
> >   */
> >  
> >  #include <linux/slab.h>
> > +#include <linux/clk.h>
> >  #include <linux/clkdev.h>
> >  #include <linux/clk-provider.h>
> >  #include <linux/io.h>
> >  #include <linux/mfd/syscon.h>
> >  #include <linux/mod_devicetable.h>
> > +#include <linux/of.h>
> >  #include <linux/of_address.h>
> >  #include <linux/regmap.h>
> >  #include <linux/syscore_ops.h>
> > @@ -489,6 +491,50 @@ void __init samsung_cmu_register_clocks(struct samsung_clk_provider *ctx,
> >  		samsung_clk_register_cpu(ctx, cmu->cpu_clks, cmu->nr_cpu_clks);
> >  }
> >  
> > +static int samsung_get_sysreg_regmap(struct device_node *np,
> > +				     struct samsung_clk_provider *ctx)
> > +{
> > +	struct device_node *sysreg_np;
> > +	struct clk *sysreg_clk;
> > +	struct regmap *regmap;
> > +	int ret;
> > +
> > +	sysreg_np = of_parse_phandle(np, "samsung,sysreg", 0);
> > +	if (!sysreg_np)
> > +		return -ENODEV;
> > +
> > +	sysreg_clk = of_clk_get(sysreg_np, 0);
> 
> I don't think you should be poking clock of some other device. This
> clearly breaks encapsulation. What sysreg is needs to do to access
> registers, is only business of sysreg. No other drivers should
> re-implement this.

Similarly, to how other drivers (like UFS and peric/USI) also poke in
the sysreg region in a similar way, the CMU related parts of sysreg are
handled in this driver, though, as only the CMU driver knows how to
configure them based on the gating (auto vs. manual).

No separate driver was sysreg was added when this code was introduced.
The clk_get() used to be part of syscon_regmap_lookup_by_phandle(), and
I chose to keep only that, as per above.

I propose to add an actual sysreg driver once the need arises, then
this can also move, but for now keeping it here seems straight forward.

Is that OK with you?

Cheers,
Andre'