[PATCH] ACPI: Don't let acpi_pad needlessly mark LAPIC unstable

[PATCH] ACPI: Don't let acpi_pad needlessly mark LAPIC unstable

[Chen Gong]

according to Venki's suggestion in the commit 0dc698b,
adding LAPIC unstable detection in the acpi_pad drvier too.

Signed-off-by: Chen Gong <gong.chen@linux.intel.com>
---
 drivers/acpi/acpi_pad.c |   36 ++++++++++++++++++++++++------------
 1 files changed, 24 insertions(+), 12 deletions(-)

diff --git a/drivers/acpi/acpi_pad.c b/drivers/acpi/acpi_pad.c
index d269a8f..446aced 100644
--- a/drivers/acpi/acpi_pad.c
+++ b/drivers/acpi/acpi_pad.c
@@ -46,6 +46,8 @@ static unsigned long power_saving_mwait_eax;
 
 static unsigned char tsc_detected_unstable;
 static unsigned char tsc_marked_unstable;
+static unsigned char lapic_detected_unstable;
+static unsigned char lapic_marked_unstable;
 
 static void power_saving_mwait_init(void)
 {
@@ -75,9 +77,6 @@ static void power_saving_mwait_init(void)
 	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
 		(highest_subcstate - 1);
 
-	for_each_online_cpu(i)
-		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ON, &i);
-
 #if defined(CONFIG_GENERIC_TIME) && defined(CONFIG_X86)
 	switch (boot_cpu_data.x86_vendor) {
 	case X86_VENDOR_AMD:
@@ -86,13 +85,15 @@ static void power_saving_mwait_init(void)
 		 * AMD Fam10h TSC will tick in all
 		 * C/P/S0/S1 states when this bit is set.
 		 */
-		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
-			return;
-
-		/*FALL THROUGH*/
+		if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
+			tsc_detected_unstable = 1;
+		if (!boot_cpu_has(X86_FEATURE_ARAT))
+			lapic_detected_unstable = 1;
+		break;
 	default:
-		/* TSC could halt in idle */
+		/* TSC & LAPIC could halt in idle */
 		tsc_detected_unstable = 1;
+		lapic_detected_unstable = 1;
 	}
 #endif
 }
@@ -180,10 +181,20 @@ static int power_saving_thread(void *data)
 				mark_tsc_unstable("TSC halts in idle");
 				tsc_marked_unstable = 1;
 			}
+			if (lapic_detected_unstable && !lapic_marked_unstable) {
+				int i;
+				/* LAPIC could halt in idle, so notify users */
+				for_each_online_cpu(i)
+					clockevents_notify(
+						CLOCK_EVT_NOTIFY_BROADCAST_ON,
+						&i);
+				lapic_marked_unstable = 1;
+			}
 			local_irq_disable();
 			cpu = smp_processor_id();
-			clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
-				&cpu);
+			if (lapic_marked_unstable)
+				clockevents_notify(
+					CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 			stop_critical_timings();
 
 			__monitor((void *)&current_thread_info()->flags, 0, 0);
@@ -192,8 +203,9 @@ static int power_saving_thread(void *data)
 				__mwait(power_saving_mwait_eax, 1);
 
 			start_critical_timings();
-			clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
-				&cpu);
+			if (lapic_marked_unstable)
+				clockevents_notify(
+					CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 			local_irq_enable();
 
 			if (jiffies > expire_time) {
-- 
1.7.0.2.279.gf1ba1c

Re: [PATCH] ACPI: Don't let acpi_pad needlessly mark LAPIC unstable

[chen gong]

On 2010-5-30 11:37, Chen Gong wrote:
> according to Venki's suggestion in the commit 0dc698b,
> adding LAPIC unstable detection in the acpi_pad drvier too.
> 
> Signed-off-by: Chen Gong <gong.chen@linux.intel.com>
> ---
>  drivers/acpi/acpi_pad.c |   36 ++++++++++++++++++++++++------------
>  1 files changed, 24 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/acpi/acpi_pad.c b/drivers/acpi/acpi_pad.c
> index d269a8f..446aced 100644
> --- a/drivers/acpi/acpi_pad.c
> +++ b/drivers/acpi/acpi_pad.c
> @@ -46,6 +46,8 @@ static unsigned long power_saving_mwait_eax;
>  
>  static unsigned char tsc_detected_unstable;
>  static unsigned char tsc_marked_unstable;
> +static unsigned char lapic_detected_unstable;
> +static unsigned char lapic_marked_unstable;
>  
>  static void power_saving_mwait_init(void)
>  {
> @@ -75,9 +77,6 @@ static void power_saving_mwait_init(void)
>  	power_saving_mwait_eax = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
>  		(highest_subcstate - 1);
>  
> -	for_each_online_cpu(i)
> -		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ON, &i);
> -
>  #if defined(CONFIG_GENERIC_TIME) && defined(CONFIG_X86)
>  	switch (boot_cpu_data.x86_vendor) {
>  	case X86_VENDOR_AMD:
> @@ -86,13 +85,15 @@ static void power_saving_mwait_init(void)
>  		 * AMD Fam10h TSC will tick in all
>  		 * C/P/S0/S1 states when this bit is set.
>  		 */
> -		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
> -			return;
> -
> -		/*FALL THROUGH*/
> +		if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
> +			tsc_detected_unstable = 1;
> +		if (!boot_cpu_has(X86_FEATURE_ARAT))
> +			lapic_detected_unstable = 1;
> +		break;
>  	default:
> -		/* TSC could halt in idle */
> +		/* TSC & LAPIC could halt in idle */
>  		tsc_detected_unstable = 1;
> +		lapic_detected_unstable = 1;
>  	}
>  #endif
>  }
> @@ -180,10 +181,20 @@ static int power_saving_thread(void *data)
>  				mark_tsc_unstable("TSC halts in idle");
>  				tsc_marked_unstable = 1;
>  			}
> +			if (lapic_detected_unstable && !lapic_marked_unstable) {
> +				int i;
> +				/* LAPIC could halt in idle, so notify users */
> +				for_each_online_cpu(i)
> +					clockevents_notify(
> +						CLOCK_EVT_NOTIFY_BROADCAST_ON,
> +						&i);
> +				lapic_marked_unstable = 1;
> +			}
>  			local_irq_disable();
>  			cpu = smp_processor_id();
> -			clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER,
> -				&cpu);
> +			if (lapic_marked_unstable)
> +				clockevents_notify(
> +					CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
>  			stop_critical_timings();
>  
>  			__monitor((void *)&current_thread_info()->flags, 0, 0);
> @@ -192,8 +203,9 @@ static int power_saving_thread(void *data)
>  				__mwait(power_saving_mwait_eax, 1);
>  
>  			start_critical_timings();
> -			clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
> -				&cpu);
> +			if (lapic_marked_unstable)
> +				clockevents_notify(
> +					CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
>  			local_irq_enable();
>  
>  			if (jiffies > expire_time) {

Ping again.
Is it reasonable ?