Re: [PATCH v3] ARM: EXYNOS: Fix suspend/resume sequences

Re: [PATCH v3] ARM: EXYNOS: Fix suspend/resume sequences

[Tomasz Figa]

Forgot to CC Daniel and linux-pm. Sorry for the noise.

On 15.07.2014 16:24, Tomasz Figa wrote:
> Due to recent consolidation of Exynos suspend and cpuidle code, some
> parts of suspend and resume sequences are executed two times, once from
> exynos_pm_syscore_ops and then from exynos_cpu_pm_notifier() and thus it
> breaks suspend, at least on Exynos4-based boards. In addition, simple
> core power down from a cpuidle driver could, in case of CPU 0 could
> result in calling functions that are specific to suspend and deeper idle
> states.
> 
> This patch fixes the issue by moving those operations outside the CPU PM
> notifier into suspend and AFTR code paths. This leads to a bit of code
> duplication, but allows additional code simplification, so in the end
> more code is removed than added.
> 
> Signed-off-by: Tomasz Figa <t.figa@samsung.com>
> ---
>  arch/arm/mach-exynos/pm.c        | 164 ++++++++++++++++++---------------------
>  drivers/cpuidle/cpuidle-exynos.c |  25 +-----
>  2 files changed, 80 insertions(+), 109 deletions(-)
> 
> diff --git a/arch/arm/mach-exynos/pm.c b/arch/arm/mach-exynos/pm.c
> index 202ca73..e3b04b4 100644
> --- a/arch/arm/mach-exynos/pm.c
> +++ b/arch/arm/mach-exynos/pm.c
> @@ -176,26 +176,6 @@ int exynos_cluster_power_state(int cluster)
>  #define S5P_CHECK_AFTR  0xFCBA0D10
>  #define S5P_CHECK_SLEEP 0x00000BAD
>  
> -/* Ext-GIC nIRQ/nFIQ is the only wakeup source in AFTR */
> -static void exynos_set_wakeupmask(long mask)
> -{
> -	__raw_writel(mask, S5P_WAKEUP_MASK);
> -}
> -
> -static void exynos_cpu_set_boot_vector(long flags)
> -{
> -	__raw_writel(virt_to_phys(exynos_cpu_resume), EXYNOS_BOOT_VECTOR_ADDR);
> -	__raw_writel(flags, EXYNOS_BOOT_VECTOR_FLAG);
> -}
> -
> -void exynos_enter_aftr(void)
> -{
> -	exynos_set_wakeupmask(0x0000ff3e);
> -	exynos_cpu_set_boot_vector(S5P_CHECK_AFTR);
> -	/* Set value of power down register for aftr mode */
> -	exynos_sys_powerdown_conf(SYS_AFTR);
> -}
> -
>  /* For Cortex-A9 Diagnostic and Power control register */
>  static unsigned int save_arm_register[2];
>  
> @@ -235,6 +215,82 @@ static void exynos_cpu_restore_register(void)
>  		      : "cc");
>  }
>  
> +static void exynos_pm_central_suspend(void)
> +{
> +	unsigned long tmp;
> +
> +	/* Setting Central Sequence Register for power down mode */
> +	tmp = __raw_readl(S5P_CENTRAL_SEQ_CONFIGURATION);
> +	tmp &= ~S5P_CENTRAL_LOWPWR_CFG;
> +	__raw_writel(tmp, S5P_CENTRAL_SEQ_CONFIGURATION);
> +}
> +
> +static int exynos_pm_central_resume(void)
> +{
> +	unsigned long tmp;
> +
> +	/*
> +	 * If PMU failed while entering sleep mode, WFI will be
> +	 * ignored by PMU and then exiting cpu_do_idle().
> +	 * S5P_CENTRAL_LOWPWR_CFG bit will not be set automatically
> +	 * in this situation.
> +	 */
> +	tmp = __raw_readl(S5P_CENTRAL_SEQ_CONFIGURATION);
> +	if (!(tmp & S5P_CENTRAL_LOWPWR_CFG)) {
> +		tmp |= S5P_CENTRAL_LOWPWR_CFG;
> +		__raw_writel(tmp, S5P_CENTRAL_SEQ_CONFIGURATION);
> +		/* clear the wakeup state register */
> +		__raw_writel(0x0, S5P_WAKEUP_STAT);
> +		/* No need to perform below restore code */
> +		return -1;
> +	}
> +
> +	return 0;
> +}
> +
> +/* Ext-GIC nIRQ/nFIQ is the only wakeup source in AFTR */
> +static void exynos_set_wakeupmask(long mask)
> +{
> +	__raw_writel(mask, S5P_WAKEUP_MASK);
> +}
> +
> +static void exynos_cpu_set_boot_vector(long flags)
> +{
> +	__raw_writel(virt_to_phys(exynos_cpu_resume), EXYNOS_BOOT_VECTOR_ADDR);
> +	__raw_writel(flags, EXYNOS_BOOT_VECTOR_FLAG);
> +}
> +
> +static int exynos_aftr_finisher(unsigned long flags)
> +{
> +	exynos_set_wakeupmask(0x0000ff3e);
> +	exynos_cpu_set_boot_vector(S5P_CHECK_AFTR);
> +	/* Set value of power down register for aftr mode */
> +	exynos_sys_powerdown_conf(SYS_AFTR);
> +	cpu_do_idle();
> +
> +	return 0;
> +}
> +
> +void exynos_enter_aftr(void)
> +{
> +	cpu_pm_enter();
> +
> +	exynos_pm_central_suspend();
> +	if (read_cpuid_part_number() == ARM_CPU_PART_CORTEX_A9)
> +		exynos_cpu_save_register();
> +
> +	cpu_suspend(0, exynos_aftr_finisher);
> +
> +	if (read_cpuid_part_number() == ARM_CPU_PART_CORTEX_A9) {
> +		scu_enable(S5P_VA_SCU);
> +		exynos_cpu_restore_register();
> +	}
> +
> +	exynos_pm_central_resume();
> +
> +	cpu_pm_exit();
> +}
> +
>  static int exynos_cpu_suspend(unsigned long arg)
>  {
>  #ifdef CONFIG_CACHE_L2X0
> @@ -279,16 +335,6 @@ static void exynos_pm_prepare(void)
>  	__raw_writel(virt_to_phys(exynos_cpu_resume), S5P_INFORM0);
>  }
>  
> -static void exynos_pm_central_suspend(void)
> -{
> -	unsigned long tmp;
> -
> -	/* Setting Central Sequence Register for power down mode */
> -	tmp = __raw_readl(S5P_CENTRAL_SEQ_CONFIGURATION);
> -	tmp &= ~S5P_CENTRAL_LOWPWR_CFG;
> -	__raw_writel(tmp, S5P_CENTRAL_SEQ_CONFIGURATION);
> -}
> -
>  static int exynos_pm_suspend(void)
>  {
>  	unsigned long tmp;
> @@ -306,29 +352,6 @@ static int exynos_pm_suspend(void)
>  	return 0;
>  }
>  
> -static int exynos_pm_central_resume(void)
> -{
> -	unsigned long tmp;
> -
> -	/*
> -	 * If PMU failed while entering sleep mode, WFI will be
> -	 * ignored by PMU and then exiting cpu_do_idle().
> -	 * S5P_CENTRAL_LOWPWR_CFG bit will not be set automatically
> -	 * in this situation.
> -	 */
> -	tmp = __raw_readl(S5P_CENTRAL_SEQ_CONFIGURATION);
> -	if (!(tmp & S5P_CENTRAL_LOWPWR_CFG)) {
> -		tmp |= S5P_CENTRAL_LOWPWR_CFG;
> -		__raw_writel(tmp, S5P_CENTRAL_SEQ_CONFIGURATION);
> -		/* clear the wakeup state register */
> -		__raw_writel(0x0, S5P_WAKEUP_STAT);
> -		/* No need to perform below restore code */
> -		return -1;
> -	}
> -
> -	return 0;
> -}
> -
>  static void exynos_pm_resume(void)
>  {
>  	if (exynos_pm_central_resume())
> @@ -431,45 +454,10 @@ static const struct platform_suspend_ops exynos_suspend_ops = {
>  	.valid		= suspend_valid_only_mem,
>  };
>  
> -static int exynos_cpu_pm_notifier(struct notifier_block *self,
> -				  unsigned long cmd, void *v)
> -{
> -	int cpu = smp_processor_id();
> -
> -	switch (cmd) {
> -	case CPU_PM_ENTER:
> -		if (cpu == 0) {
> -			exynos_pm_central_suspend();
> -			if (read_cpuid_part_number() == ARM_CPU_PART_CORTEX_A9)
> -				exynos_cpu_save_register();
> -		}
> -		break;
> -
> -	case CPU_PM_EXIT:
> -		if (cpu == 0) {
> -			if (read_cpuid_part_number() ==
> -					ARM_CPU_PART_CORTEX_A9) {
> -				scu_enable(S5P_VA_SCU);
> -				exynos_cpu_restore_register();
> -			}
> -			exynos_pm_central_resume();
> -		}
> -		break;
> -	}
> -
> -	return NOTIFY_OK;
> -}
> -
> -static struct notifier_block exynos_cpu_pm_notifier_block = {
> -	.notifier_call = exynos_cpu_pm_notifier,
> -};
> -
>  void __init exynos_pm_init(void)
>  {
>  	u32 tmp;
>  
> -	cpu_pm_register_notifier(&exynos_cpu_pm_notifier_block);
> -
>  	/* Platform-specific GIC callback */
>  	gic_arch_extn.irq_set_wake = exynos_irq_set_wake;
>  
> diff --git a/drivers/cpuidle/cpuidle-exynos.c b/drivers/cpuidle/cpuidle-exynos.c
> index 7c01512..ba9b34b 100644
> --- a/drivers/cpuidle/cpuidle-exynos.c
> +++ b/drivers/cpuidle/cpuidle-exynos.c
> @@ -20,25 +20,6 @@
>  
>  static void (*exynos_enter_aftr)(void);
>  
> -static int idle_finisher(unsigned long flags)
> -{
> -	exynos_enter_aftr();
> -	cpu_do_idle();
> -
> -	return 1;
> -}
> -
> -static int exynos_enter_core0_aftr(struct cpuidle_device *dev,
> -				struct cpuidle_driver *drv,
> -				int index)
> -{
> -	cpu_pm_enter();
> -	cpu_suspend(0, idle_finisher);
> -	cpu_pm_exit();
> -
> -	return index;
> -}
> -
>  static int exynos_enter_lowpower(struct cpuidle_device *dev,
>  				struct cpuidle_driver *drv,
>  				int index)
> @@ -51,8 +32,10 @@ static int exynos_enter_lowpower(struct cpuidle_device *dev,
>  
>  	if (new_index == 0)
>  		return arm_cpuidle_simple_enter(dev, drv, new_index);
> -	else
> -		return exynos_enter_core0_aftr(dev, drv, new_index);
> +
> +	exynos_enter_aftr();
> +
> +	return new_index;
>  }
>  
>  static struct cpuidle_driver exynos_idle_driver = {
> 

Re: [PATCH v3] ARM: EXYNOS: Fix suspend/resume sequences

[Tomasz Figa]

Hi Kukjin,

On 15.07.2014 16:26, Tomasz Figa wrote:
> Forgot to CC Daniel and linux-pm. Sorry for the noise.
> 
> On 15.07.2014 16:24, Tomasz Figa wrote:
>> Due to recent consolidation of Exynos suspend and cpuidle code, some
>> parts of suspend and resume sequences are executed two times, once from
>> exynos_pm_syscore_ops and then from exynos_cpu_pm_notifier() and thus it
>> breaks suspend, at least on Exynos4-based boards. In addition, simple
>> core power down from a cpuidle driver could, in case of CPU 0 could
>> result in calling functions that are specific to suspend and deeper idle
>> states.
>>
>> This patch fixes the issue by moving those operations outside the CPU PM
>> notifier into suspend and AFTR code paths. This leads to a bit of code
>> duplication, but allows additional code simplification, so in the end
>> more code is removed than added.
>>
>> Signed-off-by: Tomasz Figa <t.figa@samsung.com>
>> ---
>>  arch/arm/mach-exynos/pm.c        | 164 ++++++++++++++++++---------------------
>>  drivers/cpuidle/cpuidle-exynos.c |  25 +-----
>>  2 files changed, 80 insertions(+), 109 deletions(-)
>>

Please consider this patch for next pull request with rc fixes. It
replaces following patches posted in this thread:

[PATCH 5/6] ARM: EXYNOS: Fix suspend/resume sequencies
[PATCH v2 5/6] ARM: EXYNOS: Fix suspend/resume sequences

and also similar patch by Chander:

[PATCH 2/2] cpuidle: Exynos: fix cpuidle for all states

Best regards,
Tomasz

Re: [PATCH v3] ARM: EXYNOS: Fix suspend/resume sequences

[Tomasz Figa]

Hi,

On 15.07.2014 16:26, Tomasz Figa wrote:
> Forgot to CC Daniel and linux-pm. Sorry for the noise.
> 
> On 15.07.2014 16:24, Tomasz Figa wrote:
>> Due to recent consolidation of Exynos suspend and cpuidle code, some
>> parts of suspend and resume sequences are executed two times, once from
>> exynos_pm_syscore_ops and then from exynos_cpu_pm_notifier() and thus it
>> breaks suspend, at least on Exynos4-based boards. In addition, simple
>> core power down from a cpuidle driver could, in case of CPU 0 could
>> result in calling functions that are specific to suspend and deeper idle
>> states.
>>
>> This patch fixes the issue by moving those operations outside the CPU PM
>> notifier into suspend and AFTR code paths. This leads to a bit of code
>> duplication, but allows additional code simplification, so in the end
>> more code is removed than added.
>>
>> Signed-off-by: Tomasz Figa <t.figa@samsung.com>
>> ---
>>  arch/arm/mach-exynos/pm.c        | 164 ++++++++++++++++++---------------------
>>  drivers/cpuidle/cpuidle-exynos.c |  25 +-----
>>  2 files changed, 80 insertions(+), 109 deletions(-)
>>

This is an important regression fix for 3.16 and today 3.16-rc6 was
already released...

Kukjin, could you pick up this patch and few others you missed and make
sure that it is included in 3.16-rc7? If you are busy, maybe Arnd or
Olof could take them directly?

Best regards,
Tomasz