Patchwork [RFC] clk: divider: Tolerate 0 divider for one based dividers

login
register
mail settings
Submitter Soren Brinkmann
Date March 11, 2013, 9:13 p.m.
Message ID <dbceb80f-0303-4ba0-acd8-336dd85444ae@DB8EHSMHS017.ehs.local>
Download mbox | patch
Permalink /patch/226703/
State New
Headers show

Comments

Soren Brinkmann - March 11, 2013, 9:13 p.m.
Handle a zero divider value as one/bypass for dividers which have the
CLK_DIVIDER_ONE_BASED flag set.

Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com>
---
In Zynq we have a lot of dividers which are one based, but at the same time
zero is a valid value which is handled as one/bypass. Also, the reset value of
some of these registers is zero, resulting in warnings when the clock framework
encounters this.

So, my question here is: Are our dividers odd? Does it make sense to allow zero
for all one based dividers, as shown in this patch? Or does this behavior
qualify for another flag for the divider clocks (e.g. CLK_DIVIDER_ZERO_OKAY)?

	Thanks,
	Sören

 drivers/clk/clk-divider.c    | 5 +++--
 include/linux/clk-provider.h | 4 ++--
 2 files changed, 5 insertions(+), 4 deletions(-)
Mike Turquette - March 22, 2013, 5:11 p.m.
Quoting Soren Brinkmann (2013-03-11 14:13:37)
> Handle a zero divider value as one/bypass for dividers which have the
> CLK_DIVIDER_ONE_BASED flag set.
> 
> Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com>
> ---
> In Zynq we have a lot of dividers which are one based, but at the same time
> zero is a valid value which is handled as one/bypass. Also, the reset value of
> some of these registers is zero, resulting in warnings when the clock framework
> encounters this.
> 
> So, my question here is: Are our dividers odd? Does it make sense to allow zero
> for all one based dividers, as shown in this patch? Or does this behavior
> qualify for another flag for the divider clocks (e.g. CLK_DIVIDER_ZERO_OKAY)?
> 

Everyone's dividers are odd ;)

For the case where a bitfield value of zero or one translates into
divide-by-one, then I think a flag can be introduced.

But handling bypass and reset is likely to not be as generic or common
across implementation, so you might need your own custom clk_hw_foo for
that.

The basic clk_divider type is meant to do exactly what the name says:
divide an incoming clock rate.  If your clock does more than that then
you might need to cook up your own clock type.

>         Thanks,
>         Sören
> 
>  drivers/clk/clk-divider.c    | 5 +++--
>  include/linux/clk-provider.h | 4 ++--
>  2 files changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/clk/clk-divider.c b/drivers/clk/clk-divider.c
> index 68b4021..6c2a431 100644
> --- a/drivers/clk/clk-divider.c
> +++ b/drivers/clk/clk-divider.c
> @@ -109,8 +109,9 @@ static unsigned long clk_divider_recalc_rate(struct clk_hw *hw,
>  
>         div = _get_div(divider, val);
>         if (!div) {
> -               WARN(1, "%s: Invalid divisor for clock %s\n", __func__,
> -                                               __clk_get_name(hw->clk));
> +               WARN(!(divider->flags & CLK_DIVIDER_ONE_BASED),
> +                               "%s: Invalid divisor for clock %s\n", __func__,
> +                               __clk_get_name(hw->clk));

div in this code is not the value programmed into the divider, but the
numerical divider applied to the parent clock rate.  Thus dividing by
zero never makes sense here.

Regards,
Mike

>                 return parent_rate;
>         }
>  
> diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
> index 7f197d7..5b19b13 100644
> --- a/include/linux/clk-provider.h
> +++ b/include/linux/clk-provider.h
> @@ -238,8 +238,8 @@ struct clk_div_table {
>   * Flags:
>   * CLK_DIVIDER_ONE_BASED - by default the divisor is the value read from the
>   *     register plus one.  If CLK_DIVIDER_ONE_BASED is set then the divider is
> - *     the raw value read from the register, with the value of zero considered
> - *     invalid
> + *     the raw value read from the register. A zero divider is considered to be
> + *     the same as the a value of one.
>   * CLK_DIVIDER_POWER_OF_TWO - clock divisor is 2 raised to the value read from
>   *     the hardware register
>   */
> -- 
> 1.8.1.5
Soren Brinkmann - March 22, 2013, 6 p.m.
Hi Mike,

On Fri, Mar 22, 2013 at 10:11:41AM -0700, Mike Turquette wrote:
> Quoting Soren Brinkmann (2013-03-11 14:13:37)
> > Handle a zero divider value as one/bypass for dividers which have the
> > CLK_DIVIDER_ONE_BASED flag set.
> > 
> > Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com>
> > ---
> > In Zynq we have a lot of dividers which are one based, but at the same time
> > zero is a valid value which is handled as one/bypass. Also, the reset value of
> > some of these registers is zero, resulting in warnings when the clock framework
> > encounters this.
> > 
> > So, my question here is: Are our dividers odd? Does it make sense to allow zero
> > for all one based dividers, as shown in this patch? Or does this behavior
> > qualify for another flag for the divider clocks (e.g. CLK_DIVIDER_ZERO_OKAY)?
> > 
> 
> Everyone's dividers are odd ;)
> 
> For the case where a bitfield value of zero or one translates into
> divide-by-one, then I think a flag can be introduced.
> 
> But handling bypass and reset is likely to not be as generic or common
> across implementation, so you might need your own custom clk_hw_foo for
> that.
What I mean is that 0 corresponds to not modify the input frequency.
Whether hardware realizes this through bypass or divide by 1 is an irrelevant
implementation detail. I don't want to handle reset. Just avoid the warning
about a zero divider in case the CLK_DIVIDER_ONE_BASED flag is set.

> 
> The basic clk_divider type is meant to do exactly what the name says:
> divide an incoming clock rate.  If your clock does more than that then
> you might need to cook up your own clock type.
> 
> >         Thanks,
> >         Sören
> > 
> >  drivers/clk/clk-divider.c    | 5 +++--
> >  include/linux/clk-provider.h | 4 ++--
> >  2 files changed, 5 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/clk/clk-divider.c b/drivers/clk/clk-divider.c
> > index 68b4021..6c2a431 100644
> > --- a/drivers/clk/clk-divider.c
> > +++ b/drivers/clk/clk-divider.c
> > @@ -109,8 +109,9 @@ static unsigned long clk_divider_recalc_rate(struct clk_hw *hw,
> >  
> >         div = _get_div(divider, val);
> >         if (!div) {
> > -               WARN(1, "%s: Invalid divisor for clock %s\n", __func__,
> > -                                               __clk_get_name(hw->clk));
> > +               WARN(!(divider->flags & CLK_DIVIDER_ONE_BASED),
> > +                               "%s: Invalid divisor for clock %s\n", __func__,
> > +                               __clk_get_name(hw->clk));
> 
> div in this code is not the value programmed into the divider, but the
> numerical divider applied to the parent clock rate.  Thus dividing by
> zero never makes sense here.
Right, and for dividers which have the ONE_BASED flags set, this value
is just the register value read from HW. Our HW, allows the register
to be 0 and the resulting output frequency is its input frequency - just
as if it was set to 1.
Since some of our dividers have a reset value of 0, I see this warning
during boot up. My intention here is to get rid of the warning for clocks
which deem 0 a valid divider value.
I thought, reusing the ONE_BASED flag might be okay, since all such
dividers have this redundant 0 state and might handle it similar.
Otherwise a new flag might be required.

	Sören
Mike Turquette - April 1, 2013, 6:25 p.m.
Quoting Sören Brinkmann (2013-03-22 11:00:15)
> Since some of our dividers have a reset value of 0, I see this warning
> during boot up. My intention here is to get rid of the warning for clocks
> which deem 0 a valid divider value.
> I thought, reusing the ONE_BASED flag might be okay, since all such
> dividers have this redundant 0 state and might handle it similar.
> Otherwise a new flag might be required.
> 

Hi Soren,

A flag will be necessary.  I just checked some documentation for the
divider outputs for OMAP's PLLs and (which use the common clk_divider
type) and programming zero into those dividers is not allowed.  At reset
the divider registers are set to 1.

So it is unsafe to simply reuse the ONE_BASED flag.

Let's keep the current behavior as the default and introduce a new flag
to handle the special case of a zero divider.  Would you like to take a
crack at it?  It might also be nice to update the WARN message with a
hint to take a look at your new flag in case future platforms hit the
same issue as you.

Regards,
Mike

>         Sören
Soren Brinkmann - April 1, 2013, 6:31 p.m.
Hi Mike,

On Mon, Apr 01, 2013 at 11:25:22AM -0700, Mike Turquette wrote:
> Quoting Sören Brinkmann (2013-03-22 11:00:15)
> > Since some of our dividers have a reset value of 0, I see this warning
> > during boot up. My intention here is to get rid of the warning for clocks
> > which deem 0 a valid divider value.
> > I thought, reusing the ONE_BASED flag might be okay, since all such
> > dividers have this redundant 0 state and might handle it similar.
> > Otherwise a new flag might be required.
> > 
> 
> Hi Soren,
> 
> A flag will be necessary.  I just checked some documentation for the
> divider outputs for OMAP's PLLs and (which use the common clk_divider
> type) and programming zero into those dividers is not allowed.  At reset
> the divider registers are set to 1.
> 
> So it is unsafe to simply reuse the ONE_BASED flag.
> 
> Let's keep the current behavior as the default and introduce a new flag
> to handle the special case of a zero divider.  Would you like to take a
> crack at it?  It might also be nice to update the WARN message with a
> hint to take a look at your new flag in case future platforms hit the
> same issue as you.
Okay, sounds good. I should be able to get something ready within this
week.
Does anybody have suggestions for naming the new flag?
CLK_DIVIDER_ZERO_OKAY was my original suggestionn I think.

	Sören
Mike Turquette - April 1, 2013, 9:35 p.m.
Quoting Sören Brinkmann (2013-04-01 11:31:32)
> Hi Mike,
> 
> On Mon, Apr 01, 2013 at 11:25:22AM -0700, Mike Turquette wrote:
> > Quoting Sören Brinkmann (2013-03-22 11:00:15)
> > > Since some of our dividers have a reset value of 0, I see this warning
> > > during boot up. My intention here is to get rid of the warning for clocks
> > > which deem 0 a valid divider value.
> > > I thought, reusing the ONE_BASED flag might be okay, since all such
> > > dividers have this redundant 0 state and might handle it similar.
> > > Otherwise a new flag might be required.
> > > 
> > 
> > Hi Soren,
> > 
> > A flag will be necessary.  I just checked some documentation for the
> > divider outputs for OMAP's PLLs and (which use the common clk_divider
> > type) and programming zero into those dividers is not allowed.  At reset
> > the divider registers are set to 1.
> > 
> > So it is unsafe to simply reuse the ONE_BASED flag.
> > 
> > Let's keep the current behavior as the default and introduce a new flag
> > to handle the special case of a zero divider.  Would you like to take a
> > crack at it?  It might also be nice to update the WARN message with a
> > hint to take a look at your new flag in case future platforms hit the
> > same issue as you.
> Okay, sounds good. I should be able to get something ready within this
> week.
> Does anybody have suggestions for naming the new flag?
> CLK_DIVIDER_ZERO_OKAY was my original suggestionn I think.
> 

CLK_DIVIDER_ALLOW_ZERO sounds betterer to me.

Regards,
Mike

>         Sören

Patch

diff --git a/drivers/clk/clk-divider.c b/drivers/clk/clk-divider.c
index 68b4021..6c2a431 100644
--- a/drivers/clk/clk-divider.c
+++ b/drivers/clk/clk-divider.c
@@ -109,8 +109,9 @@  static unsigned long clk_divider_recalc_rate(struct clk_hw *hw,
 
 	div = _get_div(divider, val);
 	if (!div) {
-		WARN(1, "%s: Invalid divisor for clock %s\n", __func__,
-						__clk_get_name(hw->clk));
+		WARN(!(divider->flags & CLK_DIVIDER_ONE_BASED),
+				"%s: Invalid divisor for clock %s\n", __func__,
+				__clk_get_name(hw->clk));
 		return parent_rate;
 	}
 
diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
index 7f197d7..5b19b13 100644
--- a/include/linux/clk-provider.h
+++ b/include/linux/clk-provider.h
@@ -238,8 +238,8 @@  struct clk_div_table {
  * Flags:
  * CLK_DIVIDER_ONE_BASED - by default the divisor is the value read from the
  * 	register plus one.  If CLK_DIVIDER_ONE_BASED is set then the divider is
- * 	the raw value read from the register, with the value of zero considered
- * 	invalid
+ *	the raw value read from the register. A zero divider is considered to be
+ *	the same as the a value of one.
  * CLK_DIVIDER_POWER_OF_TWO - clock divisor is 2 raised to the value read from
  * 	the hardware register
  */