Re: [PATCH v3] clk: renesas: r8a774c0: Add RPC clocks

[Geert Uytterhoeven]

Hi Biju,

On Mon, Nov 16, 2020 at 10:03 AM Biju Das <biju.das.jz@xxxxxxxxxxxxxx> wrote:
> > Subject: Re: [PATCH v3] clk: renesas: r8a774c0: Add RPC clocks
> > On Tue, Nov 10, 2020 at 1:56 PM Lad Prabhakar <prabhakar.mahadev-
> > lad.rj@xxxxxxxxxxxxxx> wrote:
> > > Describe the RPCSRC internal clock and the RPC[D2] clocks derived from
> > > it, as well as the RPC-IF module clock, in the RZ/G2E (R8A774C0)
> > > CPG/MSSR driver.
> > >
> > > Add new clk type CLK_TYPE_GEN3E3_RPCSRC to register rpcsrc as a fixed
> > > clock on R-Car Gen3 E3 (and also RZ/G2E which is identical to E3 SoC),
> > > parent and the divider is set based on the register value
> > > CPG_RPCCKCR[4:3] (parent is cross verified against MD[4:1] pins) which
> > > has been set prior to booting the kernel.
> > >
> > > MD[4] MD[3] MD[2] MD[1]
> > >   0     0     0    1     -> RPCSRC CLK source is PLL1
> > >   0     0     1    1     -> RPCSRC CLK source is PLL1
> > >   0     1     0    0     -> RPCSRC CLK source is PLL1
> > >   1     0     1    1     -> RPCSRC CLK source is PLL1
> > >   x     x     x    x     -> For any other values RPCSRC CLK source is
> > PLL0
> > >
> > > Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > > Reviewed-by: Biju Das <biju.das.jz@xxxxxxxxxxxxxx>
> >
> > Thanks for your patch!
> >
> > > ---
> > > v2->v3
> > > * Implemented as a fixed clock
> >
> > Sounds fine to me.  If we ever need to configure this clock from Linux,
> > the driver can be changed.
> >
> > > --- a/drivers/clk/renesas/rcar-gen3-cpg.c
> > > +++ b/drivers/clk/renesas/rcar-gen3-cpg.c

> > > @@ -696,6 +709,42 @@ struct clk * __init
> > rcar_gen3_cpg_clk_register(struct device *dev,
> > >                                                   cpg_rpcsrc_div_table,
> > >                                                   &cpg_lock);
> > >
> > > +       case CLK_TYPE_GEN3E3_RPCSRC:
> > > +               /*
> > > +                * Register RPCSRC as fixed factor clock based on the
> > > +                * MD[4:1] pins and CPG_RPCCKCR[4:3] register value for
> > > +                * which has been set prior to booting the kernel.
> > > +                */
> > > +
> > > +               value = (readl(base + CPG_RPCCKCR) & GENMASK(4, 3)) >>
> > 3;
> > > +               if (cpg_rpcsrc_e3_parent_is_pll0(cpg_mode)) {
> > > +                       if (value != 2)
> > > +                               return ERR_PTR(-EINVAL);
> > > +               } else {
> > > +                       if (value == 2)
> > > +                               return ERR_PTR(-EINVAL);
> > > +               }
> >
> > IMHO this cross-verification is not needed, and harmful: it prevents the
> > boot loader from changing the configuration, which I think is a valid use
> > case.
>
> But this check validates, whether bootloader done wrong configuration or not?
> For eg:- PLL1 and setting wrong divider value in RPCCKCR.
>
> It allows bootloader for changing right configurations. I may be wrong. Please correct me if I am wrong.

What is a wrong configuration? According to the RPCSRC docs, DIV[4:3]
select both the parent clock and the divider.
All four possible values are valid.

MD[4:1] select the boot device, and determine the _initial values_ of
the DIV[4:0] bits.  Nothing in the documentation says they cannot be
changed later, after which the DIV[4:3] bits no longer match MD[4:1].

If you want to be really sure, you can change the bits, and measure the
impact on the RPC clock signal.

A long time ago, I did a similar thing for PLL0CR.CKSEL on R-Car D3, and
looked at the impact on the serial console (albeit with remote access,
i.e. no console output if kernel and hardware didn't agree ;-)

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds