Re: [PATCH v12 4/6] clk: qcom: Add NSS clock Controller driver for IPQ9574

[Marek Behún]

On Tue, Mar 18, 2025 at 03:50:31PM -0700, Stephen Boyd wrote:
> Quoting Marek Behún (2025-03-17 07:08:16)
> > On Thu, Mar 13, 2025 at 04:33:57PM +0530, Manikanta Mylavarapu wrote:
> > 
> > > +static struct clk_rcg2 nss_cc_clc_clk_src = {
> > > +     .cmd_rcgr = 0x28604,
> > > +     .mnd_width = 0,
> > > +     .hid_width = 5,
> > > +     .parent_map = nss_cc_parent_map_6,
> > > +     .freq_tbl = ftbl_nss_cc_clc_clk_src,
> > > +     .clkr.hw.init = &(const struct clk_init_data) {
> > > +             .name = "nss_cc_clc_clk_src",
> > > +             .parent_data = nss_cc_parent_data_6,
> > > +             .num_parents = ARRAY_SIZE(nss_cc_parent_data_6),
> > > +             .ops = &clk_rcg2_ops,
> > > +     },
> > > +};
> > 
> > This structure definition gets repeated many times in this driver,
> > with only slight changes. (This also happens in other qualcomm clock
> > drivers.)
> > 
> > Would it be possible to refactor it into a macro, to avoid the
> > insane code repetition?
> > 
> 
> We have this discussion every couple years or so. The short answer is
> no. The long answer is that it makes it harder to read because we don't
> know what argument to the macro corresponds to the struct members.
> 
> It could probably use the CLK_HW_INIT_PARENTS_DATA macro though.
> 
> static struct clk_rcg2 nss_cc_clc_clk_src = {
>      .cmd_rcgr = 0x28604,
>      .mnd_width = 0,
>      .hid_width = 5,
>      .parent_map = nss_cc_parent_map_6,
>      .freq_tbl = ftbl_nss_cc_clc_clk_src,
>      .clkr.hw.init = CLK_HW_INIT_PARENTS_DATA("nss_cc_clc_clk_src",
>                                               nss_cc_parent_data_6,
> 					      &clk_rcg2_ops, 0),
>      },
> };
> 
> but then we lose the const. Oh well.
> 
> The whole qcom clk driver probably needs an overhaul to just have
> descriptors that populate a bunch of clks that are allocated at probe
> time so that the memory footprint is smaller if you have multiple clk
> drivers loaded and so that we can probe the driver again without
> unloading the whole kernel module.

Okay then, in that case just ignore me :)