Re: [PATCH v6 7/7] clk: mediatek: Enable critical clocks for MT2701

From: James Liao
Date: Tue Feb 23 2016 - 21:26:26 EST


Hi Mike,

On Mon, 2016-02-15 at 17:50 +0800, James Liao wrote:
> On Thu, 2016-02-11 at 13:35 -0800, Michael Turquette wrote:
> > Hi James,
> >
> > Quoting James Liao (2016-02-05 01:37:30)
> > > Some system clocks should be turned on by default on MT2701.
> > > This patch enable these clocks when related clocks have
> > > been registered.
> > >
> > > Signed-off-by: James Liao <jamesjj.liao@xxxxxxxxxxxx>
> > > ---
> > > drivers/clk/mediatek/clk-mt2701.c | 23 +++++++++++++++++++++--
> > > 1 file changed, 21 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/clk/mediatek/clk-mt2701.c b/drivers/clk/mediatek/clk-mt2701.c
> > > index 01722e0..f7b4d52 100644
> > > --- a/drivers/clk/mediatek/clk-mt2701.c
> > > +++ b/drivers/clk/mediatek/clk-mt2701.c
> > > @@ -573,6 +573,21 @@ static const struct mtk_gate top_clks[] __initconst = {
> > > GATE_TOP_AUD(CLK_TOP_AUD_I2S6_MCLK, "aud_i2s6_mclk", "aud_k6_src_div", 28),
> > > };
> > >
> > > +static struct clk_onecell_data *top_clk_data __initdata;
> > > +static struct clk_onecell_data *pll_clk_data __initdata;
> > > +
> > > +static void __init mtk_clk_enable_critical(void)
> > > +{
> > > + if (!top_clk_data || !pll_clk_data)
> > > + return;
> > > +
> > > + clk_prepare_enable(pll_clk_data->clks[CLK_APMIXED_ARMPLL]);
> > > + clk_prepare_enable(top_clk_data->clks[CLK_TOP_AXI_SEL]);
> > > + clk_prepare_enable(top_clk_data->clks[CLK_TOP_MEM_SEL]);
> > > + clk_prepare_enable(top_clk_data->clks[CLK_TOP_DDRPHYCFG_SEL]);
> > > + clk_prepare_enable(top_clk_data->clks[CLK_TOP_RTC_SEL]);
> > > +}
> >
> > I think we're close to having a better solution. Please see this patch
> > series[0] and let me know if it provides a better way for you to enable
> > these critical clocks instead of the open-coded solution.
> >
> > [0] http://lkml.kernel.org/r/<1455225554-13267-1-git-send-email-mturquette@xxxxxxxxxxxx>
> >
>
> I took a quick view of this patchset, CLK_ENABLE_HAND_OFF is not
> suitable for these critical clocks. I'm not sure CLK_IS_CRITICAL
> suitable or not. A clock with CLK_IS_CRITICAL will be prepare_enable
> when it is registered. But if its parent is not registered into CCF, it
> will not be enabled even the parent clock register into CCF later.
>
> > > +
> > > static void __init mtk_topckgen_init(struct device_node *node)
> > > {
> > > struct clk_onecell_data *clk_data;
> > > @@ -585,7 +600,7 @@ static void __init mtk_topckgen_init(struct device_node *node)
> > > return;
> > > }
> > >
> > > - clk_data = mtk_alloc_clk_data(CLK_TOP_NR);
> > > + top_clk_data = clk_data = mtk_alloc_clk_data(CLK_TOP_NR);
> > >
> > > mtk_clk_register_fixed_clks(top_fixed_clks, ARRAY_SIZE(top_fixed_clks),
> > > clk_data);
> > > @@ -606,6 +621,8 @@ static void __init mtk_topckgen_init(struct device_node *node)
> > > if (r)
> > > pr_err("%s(): could not register clock provider: %d\n",
> > > __func__, r);
> > > +
> > > + mtk_clk_enable_critical();
> >
> > So you call the function here, and ...
> > > }
> > > CLK_OF_DECLARE(mtk_topckgen, "mediatek,mt2701-topckgen", mtk_topckgen_init);
> > >
> > > @@ -1201,7 +1218,7 @@ static void __init mtk_apmixedsys_init(struct device_node *node)
> > > struct clk_onecell_data *clk_data;
> > > int r;
> > >
> > > - clk_data = mtk_alloc_clk_data(ARRAY_SIZE(apmixed_plls));
> > > + pll_clk_data = clk_data = mtk_alloc_clk_data(ARRAY_SIZE(apmixed_plls));
> > > if (!clk_data)
> > > return;
> > >
> > > @@ -1212,6 +1229,8 @@ static void __init mtk_apmixedsys_init(struct device_node *node)
> > > if (r)
> > > pr_err("%s(): could not register clock provider: %d\n",
> > > __func__, r);
> > > +
> > > + mtk_clk_enable_critical();
> >
> > ... here as well. So the prepare_count and enable_count will be 2? I'm
> > not sure that this makes sense. If you have different clocks that need
> > to be enabled at different times (due to the split registration scheme
> > that you employ in this driver) then why not pass an array of those
> > clocks into mtk_clk_enable_critical?
>
> No. After topckgen and apmixedsys initialized, critical clocks'
> reference count will be 1.
>
> There is a guard at the beginning of mtk_clk_enable_critical():
>
> if (!top_clk_data || !pll_clk_data)
> return;
>
> So this function will be only executed when topckgen and apmixedsys init
> done.
>
> We need to enable critical clocks:
>
> 1. As early as possible.
> 2. After critical clocks and their parents are registered into CCF.
>
> We need to wait all necessary clock providers because parents of
> critical clocks must be enabled at the same time. For example,
> CLK_TOP_AXI_SEL may have different parent on different platforms. We
> also need to keep new parent clock on if CLK_TOP_AXI_SEL change its
> parent dynamically.
>
> Because we can't ensure the init ordering of clock providers, we can
> only check all necessary providers are registered or not in each init
> function.
>
> > Anyways, like I mentioned above I think we're close to a solution to
> > handle this in the framework.
> >
> > One question: do you ever want these critical clocks to be gated by a
> > driver, or do you want them always on, forever?
>
> In this patch, these clocks should be always on (in SW view) even if
> there are other drivers controlling them.
>
> If these is a common way to keep critical clocks and their parents on, I
> would like to use the common way. But the solution in your patch series
> [0] looks not handle parent clocks well. What's your suggestion to
> implement critical clocks base on [0]?

Do you have comments for this patch series?


Best regards,

James