Re: [PATCH v2 4/4] clk: qcom: Add Global Clock controller (GCC) driver for SDM845

[Amit Nischal]

On 2018-04-06 04:27, Stephen Boyd wrote:
> Quoting Amit Nischal (2018-04-03 05:24:41)
> > On 2018-03-20 06:12, Stephen Boyd wrote:
> > > Quoting Amit Nischal (2018-03-07 23:18:15)
> > >> +};
> > >> +
> > >> +static struct clk_rcg2 gcc_sdcc4_apps_clk_src = {
> > >> +       .cmd_rcgr = 0x1600c,
> > >> +       .mnd_width = 8,
> > >> +       .hid_width = 5,
> > >> +       .parent_map = gcc_parent_map_0,
> > >> +       .freq_tbl = ftbl_gcc_sdcc4_apps_clk_src,
> > >> +       .safe_src_freq_tbl = &cxo_safe_src_f,
> > >
> > > Why does sdcc have safe src stuff? Is something turning on the sdcc clk
> > > outside of our control?
> >
> > I will get more details on this and will get back.
>
> Any news?

I am removing the safe src for SDCC, but I am trying to get details from
teams as to why this was added, if it would be required I will add back
the safe src index again and submit the patch.

> > >
> > >> +       .clkr.hw.init = &(struct clk_init_data){
> > >> +               .name = "gcc_sdcc4_apps_clk_src",
> > >> +               .parent_names = gcc_parent_names_0,
> > >> +               .num_parents = 4,
> > >> +               .flags = CLK_SET_RATE_PARENT,
> > >> +               .ops = &clk_rcg2_shared_ops,
> > >> +       },
> > >> +};
> > >> +
> > > [...]
> > >> +
> > >> +static struct clk_branch gcc_video_xo_clk = {
> > >> +       .halt_reg = 0xb028,
> > >> +       .halt_check = BRANCH_HALT,
> > >> +       .clkr = {
> > >> +               .enable_reg = 0xb028,
> > >> +               .enable_mask = BIT(0),
> > >> +               .hw.init = &(struct clk_init_data){
> > >> +                       .name = "gcc_video_xo_clk",
> > >> +                       .flags = CLK_IS_CRITICAL,
> > >> +                       .ops = &clk_branch2_ops,
> > >> +
> > >
> > > These things have no parents and we mark them critical. Why are we
> > > even exposing them to the kernel? Are they not on by default? Are we
> > > going to change these to non-critical at some point in the future?
> >
> > These clocks are not enabled by default and going to video or other
> > multimedia cores so we are marking them as critical and need to expose
> > to the kernel. As of now, there is no plan to change these to
> > non-critical.
>
> Ok. Can we open code enabling these branches in the driver probe then?
> Still seems wasteful if nobody uses these.
>
> Put another way, either a driver (or other clk controller) should be
> toggling these gates at runtime or we should enable them once and leave
> them out of the framework. If the driver approach is taken, then the
> drivers should be able to turn the clks on and off to save some power.

As of now, no client driver is taking care of toggling these gates at
runtime. We want these clocks to be always on and that's why marked
them as CRITICAL so that if any user tries to unprepare/disable then
it won't happen and framework generates the warning.
Once the client drivers will take care of above, then we will submit
a cleanup patch.

> --
> To unsubscribe from this list: send the line "unsubscribe linux-clk" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html