diff mbox

[08/13] OMAP3: cpufreq driver changes for DVFS support

Message ID 1295618465-15234-9-git-send-email-vishwanath.bs@ti.com (mailing list archive)
State Changes Requested
Delegated to: Kevin Hilman
Headers show

Commit Message

Sripathy, Vishwanath Jan. 21, 2011, 2:01 p.m. UTC
None

Comments

Jarkko Nikula April 13, 2011, 2:13 p.m. UTC | #1
Hi

A few comments below in case this is still under development as I was
playing a bit with this version on omap3.

On Fri, 21 Jan 2011 19:31:00 +0530
Vishwanath BS <vishwanath.bs@ti.com> wrote:

> Changes in the omap cpufreq driver for DVFS support.
> 
> Signed-off-by: Vishwanath BS <vishwanath.bs@ti.com>
> Cc: Santosh Shilimkar <santosh.shilimkar@ti.com>
> ---
>  arch/arm/plat-omap/cpu-omap.c |   35 ++++++++++++++++++++++++++++++++---
>  1 files changed, 32 insertions(+), 3 deletions(-)
> 
...
> @@ -85,11 +87,11 @@ static int omap_target(struct cpufreq_policy *policy,
>  		       unsigned int target_freq,
>  		       unsigned int relation)
>  {
> -#ifdef CONFIG_ARCH_OMAP1
>  	struct cpufreq_freqs freqs;
> -#endif
>  #if defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
>  	unsigned long freq;
> +	int i;
> +	struct cpufreq_freqs freqs_notify;
>  	struct device *mpu_dev = omap2_get_mpuss_device();
>  #endif
>  	int ret = 0;
> @@ -116,9 +118,36 @@ static int omap_target(struct cpufreq_policy *policy,
>  	ret = clk_set_rate(mpu_clk, freqs.new * 1000);
>  	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
>  #elif defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
> +	freqs.old = omap_getspeed(policy->cpu);;
> +	freqs_notify.new = clk_round_rate(mpu_clk, target_freq * 1000) / 1000;
> +	freqs.cpu = policy->cpu;
> +
> +	if (freqs.old == freqs.new)
> +		return ret;
> +

Here freqs.new is uninitialized which very likely means that code falls
always through, sets the correct target frequency though, but can
populate the wrong frequency through the cpufreq_notify_transition when
running the pre notifiers below.

I think the code above meant

freqs_notify.new = clk_round_rate(mpu_clk, target_freq * 1000) / 1000;
->
freqs.new = clk_round_rate(mpu_clk, target_freq * 1000) / 1000;

as the freqs_notify is otherwise unused?

However, that doesn't work as the clk_round_rate returns the current
rate for mpu_clk on omap3 and "freqs.old == freqs.new" is always true.
Looks like there is no round_rate function for arm_fck.

I replaced that with "freqs.new = target_freq;" but this means there
will be unnecessary fall throughs if the real frequency (eg 124800 kHz)
will differ from opp table frequency (eg 125000 kHz) and no real
changes are happening e.g. with on-demand governor.

> +	/* pre notifiers */
> +	for_each_cpu(i, policy->cpus) {
> +		freqs.cpu = i;
> +		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
> +	}
Sripathy, Vishwanath April 13, 2011, 5:57 p.m. UTC | #2
> -----Original Message-----
> From: Jarkko Nikula [mailto:jhnikula@gmail.com]
> Sent: Wednesday, April 13, 2011 7:13 AM
> To: Vishwanath BS
> Cc: linux-omap@vger.kernel.org; patches@linaro.org; Santosh
> Shilimkar
> Subject: Re: [PATCH 08/13] OMAP3: cpufreq driver changes for DVFS
> support
>
> Hi
>
> A few comments below in case this is still under development as I
> was
> playing a bit with this version on omap3.
There is an updated version of omap cpufreq patches where this issue is
fixed.
https://patchwork.kernel.org/patch/632781/

Regards
Vishwa

>
> On Fri, 21 Jan 2011 19:31:00 +0530
> Vishwanath BS <vishwanath.bs@ti.com> wrote:
>
> > Changes in the omap cpufreq driver for DVFS support.
> >
> > Signed-off-by: Vishwanath BS <vishwanath.bs@ti.com>
> > Cc: Santosh Shilimkar <santosh.shilimkar@ti.com>
> > ---
> >  arch/arm/plat-omap/cpu-omap.c |   35
> ++++++++++++++++++++++++++++++++---
> >  1 files changed, 32 insertions(+), 3 deletions(-)
> >
> ...
> > @@ -85,11 +87,11 @@ static int omap_target(struct cpufreq_policy
> *policy,
> >  		       unsigned int target_freq,
> >  		       unsigned int relation)
> >  {
> > -#ifdef CONFIG_ARCH_OMAP1
> >  	struct cpufreq_freqs freqs;
> > -#endif
> >  #if defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
> >  	unsigned long freq;
> > +	int i;
> > +	struct cpufreq_freqs freqs_notify;
> >  	struct device *mpu_dev = omap2_get_mpuss_device();
> >  #endif
> >  	int ret = 0;
> > @@ -116,9 +118,36 @@ static int omap_target(struct cpufreq_policy
> *policy,
> >  	ret = clk_set_rate(mpu_clk, freqs.new * 1000);
> >  	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
> >  #elif defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
> > +	freqs.old = omap_getspeed(policy->cpu);;
> > +	freqs_notify.new = clk_round_rate(mpu_clk, target_freq * 1000)
> / 1000;
> > +	freqs.cpu = policy->cpu;
> > +
> > +	if (freqs.old == freqs.new)
> > +		return ret;
> > +
>
> Here freqs.new is uninitialized which very likely means that code
> falls
> always through, sets the correct target frequency though, but can
> populate the wrong frequency through the cpufreq_notify_transition
> when
> running the pre notifiers below.
>
> I think the code above meant
>
> freqs_notify.new = clk_round_rate(mpu_clk, target_freq * 1000) /
> 1000;
> ->
> freqs.new = clk_round_rate(mpu_clk, target_freq * 1000) / 1000;
>
> as the freqs_notify is otherwise unused?
>
> However, that doesn't work as the clk_round_rate returns the current
> rate for mpu_clk on omap3 and "freqs.old == freqs.new" is always
> true.
> Looks like there is no round_rate function for arm_fck.
>
> I replaced that with "freqs.new = target_freq;" but this means there
> will be unnecessary fall throughs if the real frequency (eg 124800
> kHz)
> will differ from opp table frequency (eg 125000 kHz) and no real
> changes are happening e.g. with on-demand governor.
>
> > +	/* pre notifiers */
> > +	for_each_cpu(i, policy->cpus) {
> > +		freqs.cpu = i;
> > +		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
> > +	}
>
> --
> Jarkko
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Jarkko Nikula April 14, 2011, 12:28 p.m. UTC | #3
On Wed, 13 Apr 2011 10:57:54 -0700
Vishwanath Sripathy <vishwanath.bs@ti.com> wrote:

> > -----Original Message-----
> > From: Jarkko Nikula [mailto:jhnikula@gmail.com]
> > Sent: Wednesday, April 13, 2011 7:13 AM
> > To: Vishwanath BS
> > Cc: linux-omap@vger.kernel.org; patches@linaro.org; Santosh
> > Shilimkar
> > Subject: Re: [PATCH 08/13] OMAP3: cpufreq driver changes for DVFS
> > support
> >
> > Hi
> >
> > A few comments below in case this is still under development as I
> > was
> > playing a bit with this version on omap3.
> There is an updated version of omap cpufreq patches where this issue is
> fixed.
> https://patchwork.kernel.org/patch/632781/
> 
Thanks, I didn't notice that patch fixes the issues I noticed. Now
also the clk_round_rate works fine since the dpll1_ck does have the
round_clock defined and the former arm_fck is anyway derived from it.
diff mbox

Patch

diff --git a/arch/arm/plat-omap/cpu-omap.c b/arch/arm/plat-omap/cpu-omap.c
index 1c1b80b..d965220 100644
--- a/arch/arm/plat-omap/cpu-omap.c
+++ b/arch/arm/plat-omap/cpu-omap.c
@@ -30,10 +30,12 @@ 
 #include <mach/hardware.h>
 #include <plat/clock.h>
 #include <asm/system.h>
+#include <asm/cpu.h>
 
 #if defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
 #include <plat/omap-pm.h>
 #include <plat/common.h>
+#include <plat/dvfs.h>
 #endif
 
 #define VERY_HI_RATE	900000000
@@ -85,11 +87,11 @@  static int omap_target(struct cpufreq_policy *policy,
 		       unsigned int target_freq,
 		       unsigned int relation)
 {
-#ifdef CONFIG_ARCH_OMAP1
 	struct cpufreq_freqs freqs;
-#endif
 #if defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
 	unsigned long freq;
+	int i;
+	struct cpufreq_freqs freqs_notify;
 	struct device *mpu_dev = omap2_get_mpuss_device();
 #endif
 	int ret = 0;
@@ -116,9 +118,36 @@  static int omap_target(struct cpufreq_policy *policy,
 	ret = clk_set_rate(mpu_clk, freqs.new * 1000);
 	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 #elif defined(CONFIG_ARCH_OMAP3) && !defined(CONFIG_OMAP_PM_NONE)
+	freqs.old = omap_getspeed(policy->cpu);;
+	freqs_notify.new = clk_round_rate(mpu_clk, target_freq * 1000) / 1000;
+	freqs.cpu = policy->cpu;
+
+	if (freqs.old == freqs.new)
+		return ret;
+
+	/* pre notifiers */
+	for_each_cpu(i, policy->cpus) {
+		freqs.cpu = i;
+		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+	}
+
+	/* scale the frequency */
 	freq = target_freq * 1000;
 	if (opp_find_freq_ceil(mpu_dev, &freq))
-		omap_pm_cpu_set_freq(freq);
+		omap_device_scale(mpu_dev, mpu_dev, freq);
+
+	/* Update loops per jiffy */
+	freqs.new = omap_getspeed(policy->cpu);
+	for_each_cpu(i, policy->cpus)
+		per_cpu(cpu_data, i).loops_per_jiffy =
+		cpufreq_scale(per_cpu(cpu_data, i).loops_per_jiffy,
+				freqs.old, freqs.new);
+
+	/* post notifiers */
+	for_each_cpu(i, policy->cpus) {
+		freqs.cpu = i;
+		cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+	}
 #endif
 	return ret;
 }