Re: [PATCH v3 2/3] clk: renesas: Add family-specific clock driver for RZ/V2H(P)

[Geert Uytterhoeven]

Hi Prabhakar,

On Mon, Jul 15, 2024 at 10:44 AM Lad, Prabhakar
<prabhakar.csengg@xxxxxxxxx> wrote:
> On Fri, Jul 12, 2024 at 6:11 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > On Fri, Jul 12, 2024 at 5:29 PM Lad, Prabhakar
> > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > On Fri, Jul 12, 2024 at 4:23 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > > On Fri, Jul 12, 2024 at 5:14 PM Lad, Prabhakar
> > > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > > On Fri, Jul 12, 2024 at 12:59 PM Geert Uytterhoeven
> > > > > > On Thu, Jun 27, 2024 at 6:14 PM Prabhakar <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > > > > From: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > > > > > >
> > > > > > > Add family-specific clock driver for RZ/V2H(P) SoCs.
> > > > > > >
> > > > > > > Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > > > > > > ---
> > > > > > > v2->v3
> > > > > > > - Dropped num_hw_resets from struct rzv2h_cpg_priv
> > > > > > > - Dropped range_check for module clocks
> > > > > > > - Made mon_index to s8 instead of u8 in struct rzv2h_mod_clk
> > > > > > > - Added support for critical module clocks with DEF_MOD_CRITICAL
> > > > > > > - Added check for mon_index in rzv2h_mod_clock_endisable and
> > > > > > >   rzv2h_mod_clock_is_enabled()
> > > >
> > > > > > > --- /dev/null
> > > > > > > +++ b/drivers/clk/renesas/rzv2h-cpg.h
> > > >
> > > > > > > +/**
> > > > > > > + * struct rzv2h_reset - Reset definitions
> > > > > > > + *
> > > > > > > + * @reset_index: reset register index
> > > > > > > + * @reset_bit: reset bit
> > > > > > > + * @mon_index: monitor register index
> > > > > > > + * @mon_bit: monitor bit
> > > > > > > + */
> > > > > > > +struct rzv2h_reset {
> > > > > > > +       u8 reset_index;
> > > > > > > +       u8 reset_bit;
> > > > > > > +       u8 mon_index;
> > > > > > > +       u8 mon_bit;
> > > > > > > +};
> > > > > > > +
> > > > > > > +#define RST_ID(x, y)   ((((x) * 16)) + (y))
> > > > > > > +
> > > > > > > +#define DEF_RST_BASE(_id, _resindex, _resbit, _monindex, _monbit)      \
> > > > > > > +       [_id] = { \
> > > > > >
> > > > > > Indexing by _id means the reset array will be very sparse.  E.g. the
> > > > > > innocent-looking r9a09g057_resets[] with only a single entry takes
> > > > > > 600 bytes.
> > > > > >
> > > > > > If you do need the full array for indexing, please allocate and
> > > > > > populate it at runtime.
> > > > > >
> > > > > OK, I will use the radix tree for resets (is that OK)?
> > > >
> > > > You mean XArray? include/linux/radix-tree.h has:
> > > >
> > > >     /* Keep unconverted code working */
> > > >     #define radix_tree_root         xarray
> > > >     #define radix_tree_node         xa_node
> > > >
> > > Yes, I meant the above.
> > >
> > > > Given a single xa_node is already 576 bytes, just allocating the full
> > > > linear reset array at runtime is probably better.
> > > >
> > > Agreed, I will create a linear reset array and loop through the array
> > > based on reset index and reset bit to match with id whenever required.
> >
> > With a full allocated linear reset array you do not need to loop,
> > but you can just index it by the reset ID??
> >
> Instead of having a sparse array, to save memory I was thinking
> something like below:
>
> /**
>  * struct rzv2h_reset - Reset definitions
>  *
>  * @reset_index: reset register index
>  * @reset_bit: reset bit
>  * @mon_index: monitor register index
>  * @mon_bit: monitor bit
>  */
> struct rzv2h_reset {
>     u8 reset_index;
>     u8 reset_bit;
>     u8 mon_index;
>     u8 mon_bit;
> };
>
> #define DEF_RST_BASE(_resindex, _resbit, _monindex, _monbit)    \
>     { \
>         .reset_index = (_resindex), \
>         .reset_bit = (_resbit), \
>         .mon_index = (_monindex), \
>         .mon_bit = (_monbit), \
>     }
>
> #define DEF_RST(_resindex, _resbit, _monindex, _monbit)    \
>     DEF_RST_BASE(_resindex, _resbit, _monindex, _monbit)
>
>
> in rzv2h_cpg_probe() (.num_resets = ARRAY_SIZE(r9a09g057_resets))
>
>     resets = devm_kmalloc_array(dev, info->num_resets, sizeof(struct
> rzv2h_reset), GFP_KERNEL);
>     if (!resets)
>         return -ENOMEM;
>
>     for (i = 0; i < priv->num_resets; i++)
>         memcpy(&resets[i], &info->resets[i], sizeof(struct rzv2h_reset));

You can combine both using devm_kmemdup().

> And have the below xlate function that will convert id into index ie
> index into rests array.
>
> static int rzv2h_get_reset_index(struct rzv2h_cpg_priv *priv,
>                  unsigned long id)
> {
>     u8 reset_index = id / 16;
>     u8 reset_bit = id % 16;
>     unsigned int i;
>
>     for (i = 0; i < priv->num_resets; i++) {
>         if (priv->resets[i].reset_index == reset_index &&
>             priv->resets[i].reset_bit == reset_bit)
>             return i;
>     }
>
>     return -EINVAL;
> }
>
> static int rzv2h_cpg_reset_xlate(struct reset_controller_dev *rcdev,
>                  const struct of_phandle_args *reset_spec)
> {
>     struct rzv2h_cpg_priv *priv = rcdev_to_priv(rcdev);
>     unsigned int id = reset_spec->args[0];
>     int index = rzv2h_get_reset_index(priv, id);
>
>     if (index < 0) {
>         dev_err(rcdev->dev, "Invalid reset index %u\n", id);
>         return -EINVAL;
>     }
>
>     return index;
> }
>
>
> rzv2h_cpg_assert() and rzv2h_cpg_deassert() which will use an id that
> can directly index into resets[] array.
>
> Please let me know if this is OK.

That would work, too, at the expense of needing a loop for look-up
(traditional trade-off between memory and time ;-)
But look-up is only done once (per device), so that should be fine.

It all depends on how many resets you will end up using...
Memory allocation also has a granularity, so once you have more
than a specific number of resets, you better use a sparse array,
and simple indexing.

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