Re: [PATCH v7 4/9] clk: mediatek: Add MT2701 clock support

[James Liao]

Hi Stephen,

On Mon, 2016-05-09 at 15:29 -0700, Stephen Boyd wrote:
> On 05/09, James Liao wrote:
> > On Fri, 2016-05-06 at 16:11 -0700, Stephen Boyd wrote:
> > > On 04/14, James Liao wrote:
> > > > diff --git a/drivers/clk/mediatek/clk-mt2701.c b/drivers/clk/mediatek/clk-mt2701.c
> > > > new file mode 100644
> > > > index 0000000..b4db141
> > > > --- /dev/null
> > > > +++ b/drivers/clk/mediatek/clk-mt2701.c
> > > > +static void __init mtk_infrasys_init(struct device_node *node)
> > > > +{
> > > > +	struct clk_onecell_data *clk_data;
> > > > +	int r;
> > > > +
> > > > +	clk_data = mtk_alloc_clk_data(CLK_INFRA_NR);
> > > > +
> > > > +	mtk_clk_register_gates(node, infra_clks, ARRAY_SIZE(infra_clks),
> > > > +						clk_data);
> > > > +	mtk_clk_register_factors(infra_fixed_divs, ARRAY_SIZE(infra_fixed_divs),
> > > > +						clk_data);
> > > > +
> > > > +	r = of_clk_add_provider(node, of_clk_src_onecell_get, clk_data);
> > > > +	if (r)
> > > > +		pr_err("%s(): could not register clock provider: %d\n",
> > > > +			__func__, r);
> > > > +}
> > > > +CLK_OF_DECLARE(mtk_infrasys, "mediatek,mt2701-infracfg", mtk_infrasys_init);
> > > 
> > > I'm still lost on the usage of CLK_OF_DECLARE here. What part of
> > > these clk controllers needs to be registered to make the timer
> > > work?
> > 
> > GPT (mtk-timer.c) may need infracfg and topckgen clocks. MT8173 for
> > example:
> > 
> > 	timer: timer@10008000 {
> > 	[...]
> > 		clocks = <&infracfg CLK_INFRA_CLK_13M>,
> > 			 <&topckgen CLK_TOP_RTC_SEL>;
> > 	};
> 
> Ok. It should be possible to only register the clk tree for these
> clks in CLK_OF_DECLARE() and then register the other clks that
> the clk provider provides with a regular driver probe routine.
> That way we can get the advantage of the device framework, etc.
> but still register the clks we need to make the timer work early
> on.

I'll study a new way to separate clock registration of a subsystem into
CLK_OF_DECLARE() and device probe().

> > 
> > > > +	GATE_DISP1(CLK_MM_DPI_DIGL, "mm_dpi_digl", "dpi0_sel", 2),
> > > > +	GATE_DISP1(CLK_MM_DPI_ENGINE, "mm_dpi_eng", "mm_sel", 3),
> > > > +	GATE_DISP1(CLK_MM_DPI1_DIGL, "mm_dpi1_digl", "dpi1_sel", 4),
> > > > +	GATE_DISP1(CLK_MM_DPI1_ENGINE, "mm_dpi1_eng", "mm_sel", 5),
> > > > +	GATE_DISP1(CLK_MM_TVE_OUTPUT, "mm_tve_output", "tve_sel", 6),
> > > > +	GATE_DISP1(CLK_MM_TVE_INPUT, "mm_tve_input", "dpi0_sel", 7),
> > > > +	GATE_DISP1(CLK_MM_HDMI_PIXEL, "mm_hdmi_pixel", "dpi1_sel", 8),
> > > > +	GATE_DISP1(CLK_MM_HDMI_PLL, "mm_hdmi_pll", "hdmi_sel", 9),
> > > > +	GATE_DISP1(CLK_MM_HDMI_AUDIO, "mm_hdmi_audio", "apll_sel", 10),
> > > > +	GATE_DISP1(CLK_MM_HDMI_SPDIF, "mm_hdmi_spdif", "apll_sel", 11),
> > > > +	GATE_DISP1(CLK_MM_TVE_FMM, "mm_tve_fmm", "mm_sel", 14),
> > > > +};
> > > 
> > > I also don't understand why we don't have different files and
> > > drivers for all these different clock controllers? They all have
> > > a similar probe structure, sure, but otherwise these are
> > > different devices with different clks for them. The whole #ifdef
> > > thing in the later patch would go away too.
> > 
> > Yes, you are right. So you prefer to support subsystem clocks in
> > separated files such as clk-mt2701-mm.c, clk-mt2701-bdp.c and so on,
> > right?
> > 
> > But even if we implement subsystem clock in separated files, we still
> > need a way to make them optional. So the config options and #ifdef may
> > not be removed.
> 
> Presumably different files could just not be compiled if the
> config is disabled, thus removing any need for #ifdef.

OK. I'll try to implement these subsystem clocks into separated files in
next patch.

> > 
> > > > +	GATE_BDP1(CLK_BDP_BRG_RT_B, "brg_rt_bclk", "mm_sel", 12),
> > > > +	GATE_BDP1(CLK_BDP_BRG_RT_DRAM, "brg_rt_dram", "mm_sel", 13),
> > > > +	GATE_BDP1(CLK_BDP_LARBRT_DRAM, "larbrt_dram", "mm_sel", 14),
> > > > +	GATE_BDP1(CLK_BDP_TMDS_SYN, "tmds_syn", "hdmi_0_pll340m", 15),
> > > > +	GATE_BDP1(CLK_BDP_HDMI_MON, "hdmi_mon", "hdmi_0_pll340m", 16),
> > > > +};
> > > > +
> > > > +static void __init mtk_bdpsys_init(struct device_node *node)
> > > 
> > > Shouldn't be __init because it's driver probe path.
> > 
> > I use builtin_platform_driver_probe(clk_drv, clk_probe) to register this
> > driver, and it looks can be __init.
> 
> Ok, but that doesn't work with deferred probe right? It may be
> better to avoid it then.

It may be a concern. I'll investigate a proper way to implement the init
functions.


Best regards,

James