[2/2] clk: rockchip: rk3399: fix returning correct value on clk_i2sout get_rate

Commit Message

Anthony Brandon July 6, 2018, 1:18 p.m. UTC

From: Alberto Panizzo <[e-mail]>

clk_i2sout can be used as codec mclk.
On simple audio card clk_i2sout is just enabled/disabled while the rate
is decided on parent clock by i2s driver.
Without setting CLK_GET_RATE_NOCACHE flag, the get_rate function on
clk_i2sout would return incorrect values after clk_i2sout's parents
update.

Signed-off-by: Alberto Panizzo <[e-mail]>
Signed-off-by: Anthony Brandon <[e-mail]>
---
 drivers/clk/rockchip/clk-rk3399.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Heiko Stuebner July 8, 2018, 9:32 a.m. UTC | #1

Hi,

Am Freitag, 6. Juli 2018, 15:18:52 CEST schrieb Anthony Brandon:
> From: Alberto Panizzo <[e-mail]>
> 
> clk_i2sout can be used as codec mclk.
> On simple audio card clk_i2sout is just enabled/disabled while the rate
> is decided on parent clock by i2s driver.
> Without setting CLK_GET_RATE_NOCACHE flag, the get_rate function on
> clk_i2sout would return incorrect values after clk_i2sout's parents
> update.

Can you elaborate a bit more on the issue you see please?
Because as far as I remember the clock-framework should already
update child-clocks when the rate of their parent changed.

So even the cached rate should be correct after the parent changes,
so I don't really understand in what case you would get a wrong rate.


Heiko


> Signed-off-by: Alberto Panizzo <[e-mail]>
> Signed-off-by: Anthony Brandon <[e-mail]>
> ---
>  drivers/clk/rockchip/clk-rk3399.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/rockchip/clk-rk3399.c b/drivers/clk/rockchip/clk-rk3399.c
> index 2a8634a..6073479 100644
> --- a/drivers/clk/rockchip/clk-rk3399.c
> +++ b/drivers/clk/rockchip/clk-rk3399.c
> @@ -630,7 +630,8 @@ static struct rockchip_clk_branch rk3399_clk_branches[] __initdata = {
>  
>  	MUX(0, "clk_i2sout_src", mux_i2sch_p, CLK_SET_RATE_PARENT,
>  			RK3399_CLKSEL_CON(31), 0, 2, MFLAGS),
> -	COMPOSITE_NODIV(SCLK_I2S_8CH_OUT, "clk_i2sout", mux_i2sout_p, CLK_SET_RATE_PARENT,
> +	COMPOSITE_NODIV(SCLK_I2S_8CH_OUT, "clk_i2sout", mux_i2sout_p,
> +			CLK_SET_RATE_PARENT | CLK_GET_RATE_NOCACHE,
>  			RK3399_CLKSEL_CON(31), 2, 1, MFLAGS,
>  			RK3399_CLKGATE_CON(8), 12, GFLAGS),
>  
>

Alberto Panizzo July 9, 2018, 4:16 p.m. UTC | #2

Hi Heiko,
On Sun, Jul 08, 2018 at 11:32:19AM +0200, Heiko Stuebner wrote:
> Hi,
> 
> Am Freitag, 6. Juli 2018, 15:18:52 CEST schrieb Anthony Brandon:
> > From: Alberto Panizzo <[e-mail]>
> > 
> > clk_i2sout can be used as codec mclk.
> > On simple audio card clk_i2sout is just enabled/disabled while the rate
> > is decided on parent clock by i2s driver.
> > Without setting CLK_GET_RATE_NOCACHE flag, the get_rate function on
> > clk_i2sout would return incorrect values after clk_i2sout's parents
> > update.
> 
> Can you elaborate a bit more on the issue you see please?
> Because as far as I remember the clock-framework should already
> update child-clocks when the rate of their parent changed.
> 
> So even the cached rate should be correct after the parent changes,
> so I don't really understand in what case you would get a wrong rate.
> 

You are right, I'm sorry this patch were coming from days of long test and
checks to understand why clk_get_rate were returning 0 while parents clocks
were set.
(And working with a codec which does not offer deterministic results)

Especially, after having found and fixed the bits misconfigurations, a
clk_get_rate on clk_i2sout before the first clk_i2s0 set_rate is
returning 0.

But clk_i2s0_mux results unparented before first clk_i2s0 set_rate and with
or without NOCACHE, until a set_rate is called on clk_i2s0, a get_rate on
clk_i2sout will return 0.

This does not prevent good behavior after first _hw_params()
so please, drop this patch.

But patch 1/2 is necessary, please apply.

Thanks,

Alberto Panizzo
--
Amarula Solutions SRL                     Via le Canevare 30 31100 Treviso Italy
Amarula Solutions BV           Cruquiuskade 47 Amsterdam 1018 AM The Netherlands
Phone. [phone] Fax. [phone]             www.amarulasolutions.com
 
> 
> 
> > Signed-off-by: Alberto Panizzo <[e-mail]>
> > Signed-off-by: Anthony Brandon <[e-mail]>
> > ---
> >  drivers/clk/rockchip/clk-rk3399.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/clk/rockchip/clk-rk3399.c b/drivers/clk/rockchip/clk-rk3399.c
> > index 2a8634a..6073479 100644
> > --- a/drivers/clk/rockchip/clk-rk3399.c
> > +++ b/drivers/clk/rockchip/clk-rk3399.c
> > @@ -630,7 +630,8 @@ static struct rockchip_clk_branch rk3399_clk_branches[] __initdata = {
> >  
> >  	MUX(0, "clk_i2sout_src", mux_i2sch_p, CLK_SET_RATE_PARENT,
> >  			RK3399_CLKSEL_CON(31), 0, 2, MFLAGS),
> > -	COMPOSITE_NODIV(SCLK_I2S_8CH_OUT, "clk_i2sout", mux_i2sout_p, CLK_SET_RATE_PARENT,
> > +	COMPOSITE_NODIV(SCLK_I2S_8CH_OUT, "clk_i2sout", mux_i2sout_p,
> > +			CLK_SET_RATE_PARENT | CLK_GET_RATE_NOCACHE,
> >  			RK3399_CLKSEL_CON(31), 2, 1, MFLAGS,
> >  			RK3399_CLKGATE_CON(8), 12, GFLAGS),
> >  
> > 
> 
> 
> 
>

Heiko Stuebner July 9, 2018, 5:37 p.m. UTC | #3

Am Montag, 9. Juli 2018, 18:16:21 CEST schrieb Alberto Panizzo:
> Hi Heiko,
> 
> On Sun, Jul 08, 2018 at 11:32:19AM +0200, Heiko Stuebner wrote:
> > Hi,
> > 
> > Am Freitag, 6. Juli 2018, 15:18:52 CEST schrieb Anthony Brandon:
> > > From: Alberto Panizzo <[e-mail]>
> > > 
> > > clk_i2sout can be used as codec mclk.
> > > On simple audio card clk_i2sout is just enabled/disabled while the rate
> > > is decided on parent clock by i2s driver.
> > > Without setting CLK_GET_RATE_NOCACHE flag, the get_rate function on
> > > clk_i2sout would return incorrect values after clk_i2sout's parents
> > > update.
> > 
> > Can you elaborate a bit more on the issue you see please?
> > Because as far as I remember the clock-framework should already
> > update child-clocks when the rate of their parent changed.
> > 
> > So even the cached rate should be correct after the parent changes,
> > so I don't really understand in what case you would get a wrong rate.
> 
> You are right, I'm sorry this patch were coming from days of long test and
> checks to understand why clk_get_rate were returning 0 while parents clocks
> were set.
> (And working with a codec which does not offer deterministic results)
> 
> Especially, after having found and fixed the bits misconfigurations, a
> clk_get_rate on clk_i2sout before the first clk_i2s0 set_rate is
> returning 0.
> 
> But clk_i2s0_mux results unparented before first clk_i2s0 set_rate and with
> or without NOCACHE, until a set_rate is called on clk_i2s0, a get_rate on
> clk_i2sout will return 0.
> 
> This does not prevent good behavior after first _hw_params()
> so please, drop this patch.

great to hear that :-)

> But patch 1/2 is necessary, please apply.

already done yesterday.


Heiko

Patch

diff --git a/drivers/clk/rockchip/clk-rk3399.c b/drivers/clk/rockchip/clk-rk3399.c
index 2a8634a..6073479 100644
--- a/drivers/clk/rockchip/clk-rk3399.c
+++ b/drivers/clk/rockchip/clk-rk3399.c
@@ -630,7 +630,8 @@  static struct rockchip_clk_branch rk3399_clk_branches[] __initdata = {
 
 	MUX(0, "clk_i2sout_src", mux_i2sch_p, CLK_SET_RATE_PARENT,
 			RK3399_CLKSEL_CON(31), 0, 2, MFLAGS),
-	COMPOSITE_NODIV(SCLK_I2S_8CH_OUT, "clk_i2sout", mux_i2sout_p, CLK_SET_RATE_PARENT,
+	COMPOSITE_NODIV(SCLK_I2S_8CH_OUT, "clk_i2sout", mux_i2sout_p,
+			CLK_SET_RATE_PARENT | CLK_GET_RATE_NOCACHE,
 			RK3399_CLKSEL_CON(31), 2, 1, MFLAGS,
 			RK3399_CLKGATE_CON(8), 12, GFLAGS),