[PATCH] OMAP2/3: PM: remove unnecessary wakeup/sleep dependency clear

[PATCH] OMAP2/3: PM: remove unnecessary wakeup/sleep dependency clear

[Paul Walmsley]

The OMAP2 and OMAP3 PM code clears clockdomain wakeup and sleep
dependencies.  This is unnecessary after commit
6f7f63cc9adf3192e6fcac4e8bed5cc10fd924aa ("OMAP clockdomain:
initialize clockdomain registers when the clockdomain layer starts")
which clears these dependencies during clockdomain init.

Signed-off-by: Paul Walmsley <paul@pwsan.com>
Cc: Kevin Hilman <khilman@deeprootsystems.com>
---
Full-chip retention idle entry tested on OMAP35xx Beagleboard.

 arch/arm/mach-omap2/pm24xx.c |    8 +-------
 arch/arm/mach-omap2/pm34xx.c |   15 ---------------
 2 files changed, 1 insertions(+), 22 deletions(-)

diff --git a/arch/arm/mach-omap2/pm24xx.c b/arch/arm/mach-omap2/pm24xx.c
index 97feb3a..10f8747 100644
--- a/arch/arm/mach-omap2/pm24xx.c
+++ b/arch/arm/mach-omap2/pm24xx.c
@@ -363,9 +363,6 @@ static const struct platform_suspend_ops __initdata omap_pm_ops;
 /* XXX This function should be shareable between OMAP2xxx and OMAP3 */
 static int __init clkdms_setup(struct clockdomain *clkdm, void *unused)
 {
-	clkdm_clear_all_wkdeps(clkdm);
-	clkdm_clear_all_sleepdeps(clkdm);
-
 	if (clkdm->flags & CLKDM_CAN_ENABLE_AUTO)
 		omap2_clkdm_allow_idle(clkdm);
 	else if (clkdm->flags & CLKDM_CAN_FORCE_SLEEP &&
@@ -411,10 +408,7 @@ static void __init prcm_setup_regs(void)
 	pwrdm_set_next_pwrst(pwrdm, PWRDM_POWER_OFF);
 	omap2_clkdm_sleep(gfx_clkdm);
 
-	/*
-	 * Clear clockdomain wakeup dependencies and enable
-	 * hardware-supervised idle for all clkdms
-	 */
+	/* Enable hardware-supervised idle for all clkdms */
 	clkdm_for_each(clkdms_setup, NULL);
 	clkdm_add_wkdep(mpu_clkdm, wkup_clkdm);
 
diff --git a/arch/arm/mach-omap2/pm34xx.c b/arch/arm/mach-omap2/pm34xx.c
index a4aa192..0eb9738 100644
--- a/arch/arm/mach-omap2/pm34xx.c
+++ b/arch/arm/mach-omap2/pm34xx.c
@@ -694,21 +694,6 @@ static void __init prcm_setup_regs(void)
 	u32 omap3630_grpsel_uart4_mask = cpu_is_omap3630() ?
 					OMAP3630_GRPSEL_UART4_MASK : 0;
 
-
-	/* XXX Reset all wkdeps. This should be done when initializing
-	 * powerdomains */
-	omap2_prm_write_mod_reg(0, OMAP3430_IVA2_MOD, PM_WKDEP);
-	omap2_prm_write_mod_reg(0, MPU_MOD, PM_WKDEP);
-	omap2_prm_write_mod_reg(0, OMAP3430_DSS_MOD, PM_WKDEP);
-	omap2_prm_write_mod_reg(0, OMAP3430_NEON_MOD, PM_WKDEP);
-	omap2_prm_write_mod_reg(0, OMAP3430_CAM_MOD, PM_WKDEP);
-	omap2_prm_write_mod_reg(0, OMAP3430_PER_MOD, PM_WKDEP);
-	if (omap_rev() > OMAP3430_REV_ES1_0) {
-		omap2_prm_write_mod_reg(0, OMAP3430ES2_SGX_MOD, PM_WKDEP);
-		omap2_prm_write_mod_reg(0, OMAP3430ES2_USBHOST_MOD, PM_WKDEP);
-	} else
-		omap2_prm_write_mod_reg(0, GFX_MOD, PM_WKDEP);
-
 	/*
 	 * Enable interface clock autoidle for all modules.
 	 * Note that in the long run this should be done by clockfw
-- 
1.7.2.3

[PATCH] OMAP2/3: PM: remove unnecessary wakeup/sleep dependency clear

[Kevin Hilman]

Paul Walmsley <paul@pwsan.com> writes:

> The OMAP2 and OMAP3 PM code clears clockdomain wakeup and sleep
> dependencies.  This is unnecessary after commit
> 6f7f63cc9adf3192e6fcac4e8bed5cc10fd924aa ("OMAP clockdomain:
> initialize clockdomain registers when the clockdomain layer starts")
> which clears these dependencies during clockdomain init.
>
> Signed-off-by: Paul Walmsley <paul@pwsan.com>
> Cc: Kevin Hilman <khilman@deeprootsystems.com>
> ---
> Full-chip retention idle entry tested on OMAP35xx Beagleboard.

Also tested on 3630/Zoom3.

Queueing for 2.6.39.  Thanks,

Kevin

>  arch/arm/mach-omap2/pm24xx.c |    8 +-------
>  arch/arm/mach-omap2/pm34xx.c |   15 ---------------
>  2 files changed, 1 insertions(+), 22 deletions(-)
>
> diff --git a/arch/arm/mach-omap2/pm24xx.c b/arch/arm/mach-omap2/pm24xx.c
> index 97feb3a..10f8747 100644
> --- a/arch/arm/mach-omap2/pm24xx.c
> +++ b/arch/arm/mach-omap2/pm24xx.c
> @@ -363,9 +363,6 @@ static const struct platform_suspend_ops __initdata omap_pm_ops;
>  /* XXX This function should be shareable between OMAP2xxx and OMAP3 */
>  static int __init clkdms_setup(struct clockdomain *clkdm, void *unused)
>  {
> -	clkdm_clear_all_wkdeps(clkdm);
> -	clkdm_clear_all_sleepdeps(clkdm);
> -
>  	if (clkdm->flags & CLKDM_CAN_ENABLE_AUTO)
>  		omap2_clkdm_allow_idle(clkdm);
>  	else if (clkdm->flags & CLKDM_CAN_FORCE_SLEEP &&
> @@ -411,10 +408,7 @@ static void __init prcm_setup_regs(void)
>  	pwrdm_set_next_pwrst(pwrdm, PWRDM_POWER_OFF);
>  	omap2_clkdm_sleep(gfx_clkdm);
>  
> -	/*
> -	 * Clear clockdomain wakeup dependencies and enable
> -	 * hardware-supervised idle for all clkdms
> -	 */
> +	/* Enable hardware-supervised idle for all clkdms */
>  	clkdm_for_each(clkdms_setup, NULL);
>  	clkdm_add_wkdep(mpu_clkdm, wkup_clkdm);
>  
> diff --git a/arch/arm/mach-omap2/pm34xx.c b/arch/arm/mach-omap2/pm34xx.c
> index a4aa192..0eb9738 100644
> --- a/arch/arm/mach-omap2/pm34xx.c
> +++ b/arch/arm/mach-omap2/pm34xx.c
> @@ -694,21 +694,6 @@ static void __init prcm_setup_regs(void)
>  	u32 omap3630_grpsel_uart4_mask = cpu_is_omap3630() ?
>  					OMAP3630_GRPSEL_UART4_MASK : 0;
>  
> -
> -	/* XXX Reset all wkdeps. This should be done when initializing
> -	 * powerdomains */
> -	omap2_prm_write_mod_reg(0, OMAP3430_IVA2_MOD, PM_WKDEP);
> -	omap2_prm_write_mod_reg(0, MPU_MOD, PM_WKDEP);
> -	omap2_prm_write_mod_reg(0, OMAP3430_DSS_MOD, PM_WKDEP);
> -	omap2_prm_write_mod_reg(0, OMAP3430_NEON_MOD, PM_WKDEP);
> -	omap2_prm_write_mod_reg(0, OMAP3430_CAM_MOD, PM_WKDEP);
> -	omap2_prm_write_mod_reg(0, OMAP3430_PER_MOD, PM_WKDEP);
> -	if (omap_rev() > OMAP3430_REV_ES1_0) {
> -		omap2_prm_write_mod_reg(0, OMAP3430ES2_SGX_MOD, PM_WKDEP);
> -		omap2_prm_write_mod_reg(0, OMAP3430ES2_USBHOST_MOD, PM_WKDEP);
> -	} else
> -		omap2_prm_write_mod_reg(0, GFX_MOD, PM_WKDEP);
> -
>  	/*
>  	 * Enable interface clock autoidle for all modules.
>  	 * Note that in the long run this should be done by clockfw

[PATCH] OMAP2/3: PM: remove unnecessary wakeup/sleep dependency clear

[Kevin Hilman]

Hi Paul,

Kevin Hilman <khilman@ti.com> writes:

> Paul Walmsley <paul@pwsan.com> writes:
>
>> The OMAP2 and OMAP3 PM code clears clockdomain wakeup and sleep
>> dependencies.  This is unnecessary after commit
>> 6f7f63cc9adf3192e6fcac4e8bed5cc10fd924aa ("OMAP clockdomain:
>> initialize clockdomain registers when the clockdomain layer starts")
>> which clears these dependencies during clockdomain init.
>>
>> Signed-off-by: Paul Walmsley <paul@pwsan.com>
>> Cc: Kevin Hilman <khilman@deeprootsystems.com>
>> ---
>> Full-chip retention idle entry tested on OMAP35xx Beagleboard.
>
> Also tested on 3630/Zoom3.
>
> Queueing for 2.6.39.  Thanks,
>

While testing, I found out this patch breaks suspend/resume and idle on
at least 34xx.

A first suspend/resume works fine, but on subsequent ones SGX
powerdomain does not hit retention (and thus CORE powerdomain does not
either....

More below...

>
>>  arch/arm/mach-omap2/pm24xx.c |    8 +-------
>>  arch/arm/mach-omap2/pm34xx.c |   15 ---------------
>>  2 files changed, 1 insertions(+), 22 deletions(-)
>>
>> diff --git a/arch/arm/mach-omap2/pm24xx.c b/arch/arm/mach-omap2/pm24xx.c
>> index 97feb3a..10f8747 100644
>> --- a/arch/arm/mach-omap2/pm24xx.c
>> +++ b/arch/arm/mach-omap2/pm24xx.c
>> @@ -363,9 +363,6 @@ static const struct platform_suspend_ops __initdata omap_pm_ops;
>>  /* XXX This function should be shareable between OMAP2xxx and OMAP3 */
>>  static int __init clkdms_setup(struct clockdomain *clkdm, void *unused)
>>  {
>> -	clkdm_clear_all_wkdeps(clkdm);
>> -	clkdm_clear_all_sleepdeps(clkdm);
>> -
>>  	if (clkdm->flags & CLKDM_CAN_ENABLE_AUTO)
>>  		omap2_clkdm_allow_idle(clkdm);
>>  	else if (clkdm->flags & CLKDM_CAN_FORCE_SLEEP &&
>> @@ -411,10 +408,7 @@ static void __init prcm_setup_regs(void)
>>  	pwrdm_set_next_pwrst(pwrdm, PWRDM_POWER_OFF);
>>  	omap2_clkdm_sleep(gfx_clkdm);
>>  
>> -	/*
>> -	 * Clear clockdomain wakeup dependencies and enable
>> -	 * hardware-supervised idle for all clkdms
>> -	 */
>> +	/* Enable hardware-supervised idle for all clkdms */
>>  	clkdm_for_each(clkdms_setup, NULL);
>>  	clkdm_add_wkdep(mpu_clkdm, wkup_clkdm);
>>  
>> diff --git a/arch/arm/mach-omap2/pm34xx.c b/arch/arm/mach-omap2/pm34xx.c
>> index a4aa192..0eb9738 100644
>> --- a/arch/arm/mach-omap2/pm34xx.c
>> +++ b/arch/arm/mach-omap2/pm34xx.c
>> @@ -694,21 +694,6 @@ static void __init prcm_setup_regs(void)
>>  	u32 omap3630_grpsel_uart4_mask = cpu_is_omap3630() ?
>>  					OMAP3630_GRPSEL_UART4_MASK : 0;
>>  
>> -
>> -	/* XXX Reset all wkdeps. This should be done when initializing
>> -	 * powerdomains */
>> -	omap2_prm_write_mod_reg(0, OMAP3430_IVA2_MOD, PM_WKDEP);
>> -	omap2_prm_write_mod_reg(0, MPU_MOD, PM_WKDEP);
>> -	omap2_prm_write_mod_reg(0, OMAP3430_DSS_MOD, PM_WKDEP);
>> -	omap2_prm_write_mod_reg(0, OMAP3430_NEON_MOD, PM_WKDEP);
>> -	omap2_prm_write_mod_reg(0, OMAP3430_CAM_MOD, PM_WKDEP);
>> -	omap2_prm_write_mod_reg(0, OMAP3430_PER_MOD, PM_WKDEP);
>> -	if (omap_rev() > OMAP3430_REV_ES1_0) {
>> -		omap2_prm_write_mod_reg(0, OMAP3430ES2_SGX_MOD, PM_WKDEP);
>> -		omap2_prm_write_mod_reg(0, OMAP3430ES2_USBHOST_MOD, PM_WKDEP);
>> -	} else
>> -		omap2_prm_write_mod_reg(0, GFX_MOD, PM_WKDEP);


Simply adding back the above 5 lines (the SGX ones) gets things working
again, which suggests something isn't quite right with SGX, but I
haven't found the exact reason yet.

Kevin


>>  	/*
>>  	 * Enable interface clock autoidle for all modules.
>>  	 * Note that in the long run this should be done by clockfw

[PATCH] OMAP2/3: PM: remove unnecessary wakeup/sleep dependency clear

[Paul Walmsley]

Hi Kevin,

On Thu, 10 Feb 2011, Kevin Hilman wrote:

> While testing, I found out this patch breaks suspend/resume and idle on
> at least 34xx.
> 
> A first suspend/resume works fine, but on subsequent ones SGX
> powerdomain does not hit retention (and thus CORE powerdomain does not
> either....

...

> Simply adding back the above 5 lines (the SGX ones) gets things working
> again, which suggests something isn't quite right with SGX, but I
> haven't found the exact reason yet.

Thanks for the report, I will take a look.


- Paul

[PATCH] OMAP2/3: PM: remove unnecessary wakeup/sleep dependency clear

[Paul Walmsley]

On Thu, 10 Feb 2011, Paul Walmsley wrote:

> On Thu, 10 Feb 2011, Kevin Hilman wrote:
> 
> > While testing, I found out this patch breaks suspend/resume and idle on
> > at least 34xx.
> > 
> > A first suspend/resume works fine, but on subsequent ones SGX
> > powerdomain does not hit retention (and thus CORE powerdomain does not
> > either....
> 
> ...
> 
> > Simply adding back the above 5 lines (the SGX ones) gets things working
> > again, which suggests something isn't quite right with SGX, but I
> > haven't found the exact reason yet.
> 
> Thanks for the report, I will take a look.

Kevin and I did some debugging on this.  It turns out that there is a bug 
in the recent clockdomain abstraction code in the git.pwsan.com 
integration branch that caused the underlying problem.  Will follow up 
with comments on the threads of the original patches.

Kevin is planning to keep the patch under discussion in this thread in 
his 2.6.39 queue...


- Paul