Re: [PATCH v3 5/8] clk: sunxi-ng: nkm: Support finding closest rate

[Frank Oltmanns]

On 2023-07-03 at 09:25:59 +0200, Maxime Ripard <maxime@xxxxxxxxxx> wrote:
> [[PGP Signed Part:Undecided]]
> On Mon, Jul 03, 2023 at 09:17:43AM +0200, Frank Oltmanns wrote:
>>
>> On 2023-07-02 at 19:55:24 +0200, Frank Oltmanns <frank@xxxxxxxxxxxx> wrote:
>> > When finding the best rate for a NKM clock, consider rates that are
>> > higher than the requested rate, if the CCU_FEATURE_CLOSEST_RATE flag is
>> > set.
>> >
>> > Accommodate ccu_mux_helper_determine_rate to this change.
>> >
>> > Signed-off-by: Frank Oltmanns <frank@xxxxxxxxxxxx>
>> > ---
>> >  drivers/clk/sunxi-ng/ccu_mux.c | 23 +++++++++++++++-----
>> >  drivers/clk/sunxi-ng/ccu_nkm.c | 48 +++++++++++++++++++++++++++++++-----------
>> >  2 files changed, 54 insertions(+), 17 deletions(-)
>> >
>> > diff --git a/drivers/clk/sunxi-ng/ccu_mux.c b/drivers/clk/sunxi-ng/ccu_mux.c
>> > index 1d557e323169..8594d6a4addd 100644
>> > --- a/drivers/clk/sunxi-ng/ccu_mux.c
>> > +++ b/drivers/clk/sunxi-ng/ccu_mux.c
>> > @@ -113,7 +113,7 @@ int ccu_mux_helper_determine_rate(struct ccu_common *common,
>> >  	}
>> >
>> >  	for (i = 0; i < clk_hw_get_num_parents(hw); i++) {
>> > -		unsigned long tmp_rate, parent_rate;
>> > +		unsigned long tmp_rate, parent_rate, best_diff = ULONG_MAX;
>> >  		struct clk_hw *parent;
>> >
>> >  		parent = clk_hw_get_parent_by_index(hw, i);
>> > @@ -139,10 +139,23 @@ int ccu_mux_helper_determine_rate(struct ccu_common *common,
>> >  			goto out;
>> >  		}
>> >
>> > -		if ((req->rate - tmp_rate) < (req->rate - best_rate)) {
>> > -			best_rate = tmp_rate;
>> > -			best_parent_rate = parent_rate;
>> > -			best_parent = parent;
>> > +		if (common->features & CCU_FEATURE_CLOSEST_RATE) {
>> > +			unsigned long tmp_diff = req->rate > tmp_rate ?
>> > +						 req->rate - tmp_rate :
>> > +						 tmp_rate - req->rate;
>> > +
>> > +			if (tmp_diff < best_diff) {
>> > +				best_rate = tmp_rate;
>> > +				best_parent_rate = parent_rate;
>> > +				best_parent = parent;
>> > +				best_diff = tmp_diff;
>> > +			}
>> > +		} else {
>> > +			if ((req->rate - tmp_rate) < (req->rate - best_rate)) {
>> > +				best_rate = tmp_rate;
>> > +				best_parent_rate = parent_rate;
>> > +				best_parent = parent;
>> > +			}
>> >  		}
>> >  	}
>> >
>> > diff --git a/drivers/clk/sunxi-ng/ccu_nkm.c b/drivers/clk/sunxi-ng/ccu_nkm.c
>> > index d83843e69c25..36d9e987e4d8 100644
>> > --- a/drivers/clk/sunxi-ng/ccu_nkm.c
>> > +++ b/drivers/clk/sunxi-ng/ccu_nkm.c
>> > @@ -18,9 +18,11 @@ struct _ccu_nkm {
>> >  };
>> >
>> >  static unsigned long ccu_nkm_find_best_with_parent_adj(unsigned long *parent, unsigned long rate,
>> > -						       struct _ccu_nkm *nkm, struct clk_hw *phw)
>> > +						       struct _ccu_nkm *nkm, struct clk_hw *phw,
>> > +						       unsigned long features)
>> >  {
>> > -	unsigned long best_rate = 0, best_parent_rate = *parent, tmp_parent = *parent;
>> > +	unsigned long best_rate = 0, best_parent_rate = 0, tmp_parent = *parent;
>> > +	unsigned long best_diff = ULONG_MAX;
>> >  	unsigned long best_n = 0, best_k = 0, best_m = 0;
>> >  	unsigned long _n, _k, _m;
>> >
>> > @@ -28,16 +30,26 @@ static unsigned long ccu_nkm_find_best_with_parent_adj(unsigned long *parent, un
>> >  		for (_n = nkm->min_n; _n <= nkm->max_n; _n++) {
>> >  			for (_m = nkm->min_m; _m <= nkm->max_m; _m++) {
>> >  				unsigned long tmp_rate;
>> > +				unsigned long tmp_diff;
>> >
>> >  				tmp_parent = clk_hw_round_rate(phw, rate * _m / (_n * _k));
>> >
>> >  				tmp_rate = tmp_parent * _n * _k / _m;
>> > -				if (tmp_rate > rate)
>> > -					continue;
>> >
>> > -				if ((rate - tmp_rate) < (rate - best_rate)) {
>> > +				if (features & CCU_FEATURE_CLOSEST_RATE) {
>> > +					tmp_diff = rate > tmp_rate ?
>> > +						   rate - tmp_rate :
>> > +						   tmp_rate - rate;
>> > +				} else {
>> > +					if (tmp_rate > rate)
>> > +						continue;
>> > +					tmp_diff = rate - tmp_diff;
>>
>> Sorry, this should of course be tmp_diff = rate - tmp_rate. I'll fix
>> that in v4. Also I'll do tests on my phone where
>> CCU_FEATURE_CLOSEST_RATE is not set (i.e., without PATCH 8), so see if
>> it replicates the old behaviour. I'll also look into adding kunit tests,
>> so that this doesn't happen again. I'm not sure if this is feasible, but
>> I'll ask here for advise, if/when I encounter obstacles.
>
> While this would obviously be great, I don't think we have the
> infrastructure just yet to allow to easily add kunit tests for entire
> clocks.

I think, clk_test.c provides a good blueprint. I tried to do that for
clk-fractional-divider [1], but Stephen wanted to go a different route,
so I dropped it. You could look at clk_fd_test_init() in [1]. A similar
approach might work for the sunxi-ng clocks. I don't see any real
blockers, but maybe that's me being naive.

[1]: https://lore.kernel.org/all/20230614185521.477924-3-frank@xxxxxxxxxxxx/

Best regards,
  Frank

>
> Maxime
>
> [[End of PGP Signed Part]]