On Tue 29 Jun 15:23 CDT 2021, Stephen Boyd wrote:
> Quoting Bjorn Andersson (2021-06-29 08:57:29)
> > On Tue 29 Jun 08:36 CDT 2021, Shawn Guo wrote:
> >
> > > On Sun, Jun 27, 2021 at 05:28:34PM -0700, Stephen Boyd wrote:
> > > > Quoting Shawn Guo (2021-05-03 22:28:42)
> > > > > Unlike MSM8916 which has only one APCS clock, MSM8939 gets three for
> > > > > Cluster0 (little cores), Cluster1 (big cores) and CCI (Cache Coherent
> > > > > Interconnect). Instead of hard coding APCS (and A53PLL) clock name,
> > > > > retrieve the name from DT, so that multiple APCS clocks can be
> > > > > registered.
> > > > >
> > > > > Signed-off-by: Shawn Guo <shawn.guo@xxxxxxxxxx>
> > > > > ---
> > > > > drivers/clk/qcom/a53-pll.c | 5 ++++-
> > > > > drivers/clk/qcom/apcs-msm8916.c | 5 ++++-
> > > > > 2 files changed, 8 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/drivers/clk/qcom/a53-pll.c b/drivers/clk/qcom/a53-pll.c
> > > > > index 8614b0b0e82c..964f5ab7d02f 100644
> > > > > --- a/drivers/clk/qcom/a53-pll.c
> > > > > +++ b/drivers/clk/qcom/a53-pll.c
> > > > > @@ -42,6 +42,7 @@ static int qcom_a53pll_probe(struct platform_device *pdev)
> > > > > struct clk_pll *pll;
> > > > > void __iomem *base;
> > > > > struct clk_init_data init = { };
> > > > > + const char *clk_name = NULL;
> > > > > int ret;
> > > > >
> > > > > pll = devm_kzalloc(dev, sizeof(*pll), GFP_KERNEL);
> > > > > @@ -66,7 +67,9 @@ static int qcom_a53pll_probe(struct platform_device *pdev)
> > > > > pll->status_bit = 16;
> > > > > pll->freq_tbl = a53pll_freq;
> > > > >
> > > > > - init.name = "a53pll";
> > > > > + of_property_read_string(pdev->dev.of_node, "clock-output-names",
> > > > > + &clk_name);
> > > >
> > > > Please no? Is there any use for this? Why not just generate the name as
> > > > a53pll@<MMIO ADDRESS>?
> > >
> > > There is no other use for this than getting different names. I will do
> > > what you suggest here. Thanks!
> > >
> >
> > I have exactly the same problem with my two DP PHYs (in
> > phy_dp_clks_register()), so I'm in favor of us setting some sort of
> > standard for this (not for anyone to rely on, but to avoid everyone
> > coming up with their own scheme).
> >
> > But unfortunately I don't have easy access to the phy block's base
> > address in phy_dp_clks_register().
>
> It really doesn't matter what name you use as it's basically only for
> debugging. The problem is uniqueness. I've wondered if leaving the name
> as NULL and then passing in a dev would be sufficient to generate a clk
> name at runtime. Basically dev_name() plus an incrementing global
> numberspace would probably work fine. Debugging would be annoying in
> that case, but maybe it wouldn't matter.
Something like "%s:link" and "%s:vco_div" based on dev_name() would be
quite nice in the case of DP. Probably more enjoyable to read than
dev_name():N and dev_name():N+1.
It comes with a cost of a few extra lines of code in each driver, but if
it's only a few drivers I don't think it warrants the extra logic in the
core.
Regards,
Bjorn
