[RFC PATCH] ACPICA: Improve GPE detect

[RFC PATCH] ACPICA: Improve GPE detect

[Lin Ming]

Hi, Rafael

Currently, the ACPICA GPE interrupt handler reads all the GPE registers to
detect which GPEs, if any, have fired. This is done on each and every SCI,
since there is no way to know up front if any GPEs have been raised.

However, since ACPICA internally keeps a GPE enable mask for every GPE
register, it may be possible to ignore any GPE registers whose internal
GPE enable mask is zero. This would eliminate reading the status and enable
registers for that particular GPE register pair.

http://www.acpica.org/bugzilla/show_bug.cgi?id=884

I have tested suspend/resume and it works OK.
What do you think about below simple patch?

Thanks.
---
 drivers/acpi/acpica/evgpe.c |    8 ++++++++
 1 files changed, 8 insertions(+), 0 deletions(-)

diff --git a/drivers/acpi/acpica/evgpe.c b/drivers/acpi/acpica/evgpe.c
index 7c339d3..49fab43 100644
--- a/drivers/acpi/acpica/evgpe.c
+++ b/drivers/acpi/acpica/evgpe.c
@@ -373,6 +373,14 @@ u32 acpi_ev_gpe_detect(struct acpi_gpe_xrupt_info * gpe_xrupt_list)
 
 			gpe_register_info = &gpe_block->register_info[i];
 
+			if (!gpe_register_info->enable_for_run &&
+				!gpe_register_info->enable_for_wake) {
+
+				/* Disabled for both "runtime" and "wakeup", move on */
+
+				continue;
+			}
+
 			/* Read the Status Register */
 
 			status =

RE: [RFC PATCH] ACPICA: Improve GPE detect

[Moore, Robert]

+			if (!gpe_register_info->enable_for_run &&
+				!gpe_register_info->enable_for_wake) {

I would bitwise OR them together first

+			if (!(gpe_register_info->enable_for_run |
+				gpe_register_info->enable_for_wake)) {

-----Original Message-----
From: Lin, Ming M 
Sent: Wednesday, December 15, 2010 12:00 AM
To: Rafael J. Wysocki
Cc: lenb; Moore, Robert; linux-acpi
Subject: [RFC PATCH] ACPICA: Improve GPE detect

Hi, Rafael

Currently, the ACPICA GPE interrupt handler reads all the GPE registers to
detect which GPEs, if any, have fired. This is done on each and every SCI,
since there is no way to know up front if any GPEs have been raised.

However, since ACPICA internally keeps a GPE enable mask for every GPE
register, it may be possible to ignore any GPE registers whose internal
GPE enable mask is zero. This would eliminate reading the status and enable
registers for that particular GPE register pair.

http://www.acpica.org/bugzilla/show_bug.cgi?id=884

I have tested suspend/resume and it works OK.
What do you think about below simple patch?

Thanks.
---
 drivers/acpi/acpica/evgpe.c |    8 ++++++++
 1 files changed, 8 insertions(+), 0 deletions(-)

diff --git a/drivers/acpi/acpica/evgpe.c b/drivers/acpi/acpica/evgpe.c
index 7c339d3..49fab43 100644
--- a/drivers/acpi/acpica/evgpe.c
+++ b/drivers/acpi/acpica/evgpe.c
@@ -373,6 +373,14 @@ u32 acpi_ev_gpe_detect(struct acpi_gpe_xrupt_info * gpe_xrupt_list)
 
 			gpe_register_info = &gpe_block->register_info[i];
 
+			if (!gpe_register_info->enable_for_run &&
+				!gpe_register_info->enable_for_wake) {
+
+				/* Disabled for both "runtime" and "wakeup", move on */
+
+				continue;
+			}
+
 			/* Read the Status Register */
 
 			status =

Re: [RFC PATCH] ACPICA: Improve GPE detect

[Rafael J. Wysocki]

On Wednesday, December 15, 2010, Moore, Robert wrote:
> +			if (!gpe_register_info->enable_for_run &&
> +				!gpe_register_info->enable_for_wake) {
> 
> I would bitwise OR them together first
> 
> +			if (!(gpe_register_info->enable_for_run |
> +				gpe_register_info->enable_for_wake)) {

Well, I think a || would be a little better (they are treated as flags rather
than bit patterns everywhere else), but I generally agree.

The patch looks good apart from this.

Thanks,
Rafael


> -----Original Message-----
> From: Lin, Ming M 
> Sent: Wednesday, December 15, 2010 12:00 AM
> To: Rafael J. Wysocki
> Cc: lenb; Moore, Robert; linux-acpi
> Subject: [RFC PATCH] ACPICA: Improve GPE detect
> 
> Hi, Rafael
> 
> Currently, the ACPICA GPE interrupt handler reads all the GPE registers to
> detect which GPEs, if any, have fired. This is done on each and every SCI,
> since there is no way to know up front if any GPEs have been raised.
> 
> However, since ACPICA internally keeps a GPE enable mask for every GPE
> register, it may be possible to ignore any GPE registers whose internal
> GPE enable mask is zero. This would eliminate reading the status and enable
> registers for that particular GPE register pair.
> 
> http://www.acpica.org/bugzilla/show_bug.cgi?id=884
> 
> I have tested suspend/resume and it works OK.
> What do you think about below simple patch?
> 
> Thanks.
> ---
>  drivers/acpi/acpica/evgpe.c |    8 ++++++++
>  1 files changed, 8 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/acpi/acpica/evgpe.c b/drivers/acpi/acpica/evgpe.c
> index 7c339d3..49fab43 100644
> --- a/drivers/acpi/acpica/evgpe.c
> +++ b/drivers/acpi/acpica/evgpe.c
> @@ -373,6 +373,14 @@ u32 acpi_ev_gpe_detect(struct acpi_gpe_xrupt_info * gpe_xrupt_list)
>  
>  			gpe_register_info = &gpe_block->register_info[i];
>  
> +			if (!gpe_register_info->enable_for_run &&
> +				!gpe_register_info->enable_for_wake) {
> +
> +				/* Disabled for both "runtime" and "wakeup", move on */
> +
> +				continue;
> +			}
> +
>  			/* Read the Status Register */
>  
>  			status =
> 
> 
> 

RE: [RFC PATCH] ACPICA: Improve GPE detect

[Moore, Robert]

They are 8 enable bits per gpe_register.

So, may as well OR them together first and use a single compare.


-----Original Message-----
From: Rafael J. Wysocki [mailto:rjw@sisk.pl] 
Sent: Friday, December 17, 2010 12:23 PM
To: Moore, Robert
Cc: Lin, Ming M; lenb; linux-acpi
Subject: Re: [RFC PATCH] ACPICA: Improve GPE detect

On Wednesday, December 15, 2010, Moore, Robert wrote:
> +			if (!gpe_register_info->enable_for_run &&
> +				!gpe_register_info->enable_for_wake) {
> 
> I would bitwise OR them together first
> 
> +			if (!(gpe_register_info->enable_for_run |
> +				gpe_register_info->enable_for_wake)) {

Well, I think a || would be a little better (they are treated as flags rather
than bit patterns everywhere else), but I generally agree.

The patch looks good apart from this.

Thanks,
Rafael


> -----Original Message-----
> From: Lin, Ming M 
> Sent: Wednesday, December 15, 2010 12:00 AM
> To: Rafael J. Wysocki
> Cc: lenb; Moore, Robert; linux-acpi
> Subject: [RFC PATCH] ACPICA: Improve GPE detect
> 
> Hi, Rafael
> 
> Currently, the ACPICA GPE interrupt handler reads all the GPE registers to
> detect which GPEs, if any, have fired. This is done on each and every SCI,
> since there is no way to know up front if any GPEs have been raised.
> 
> However, since ACPICA internally keeps a GPE enable mask for every GPE
> register, it may be possible to ignore any GPE registers whose internal
> GPE enable mask is zero. This would eliminate reading the status and enable
> registers for that particular GPE register pair.
> 
> http://www.acpica.org/bugzilla/show_bug.cgi?id=884
> 
> I have tested suspend/resume and it works OK.
> What do you think about below simple patch?
> 
> Thanks.
> ---
>  drivers/acpi/acpica/evgpe.c |    8 ++++++++
>  1 files changed, 8 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/acpi/acpica/evgpe.c b/drivers/acpi/acpica/evgpe.c
> index 7c339d3..49fab43 100644
> --- a/drivers/acpi/acpica/evgpe.c
> +++ b/drivers/acpi/acpica/evgpe.c
> @@ -373,6 +373,14 @@ u32 acpi_ev_gpe_detect(struct acpi_gpe_xrupt_info * gpe_xrupt_list)
>  
>  			gpe_register_info = &gpe_block->register_info[i];
>  
> +			if (!gpe_register_info->enable_for_run &&
> +				!gpe_register_info->enable_for_wake) {
> +
> +				/* Disabled for both "runtime" and "wakeup", move on */
> +
> +				continue;
> +			}
> +
>  			/* Read the Status Register */
>  
>  			status =
> 
> 
> 

Re: [RFC PATCH] ACPICA: Improve GPE detect

[Rafael J. Wysocki]

On Friday, December 17, 2010, Moore, Robert wrote:
> They are 8 enable bits per gpe_register.

That's correct, sorry.

> So, may as well OR them together first and use a single compare.

Exactly.


> -----Original Message-----
> From: Rafael J. Wysocki [mailto:rjw@sisk.pl] 
> Sent: Friday, December 17, 2010 12:23 PM
> To: Moore, Robert
> Cc: Lin, Ming M; lenb; linux-acpi
> Subject: Re: [RFC PATCH] ACPICA: Improve GPE detect
> 
> On Wednesday, December 15, 2010, Moore, Robert wrote:
> > +			if (!gpe_register_info->enable_for_run &&
> > +				!gpe_register_info->enable_for_wake) {
> > 
> > I would bitwise OR them together first
> > 
> > +			if (!(gpe_register_info->enable_for_run |
> > +				gpe_register_info->enable_for_wake)) {
> 
> Well, I think a || would be a little better (they are treated as flags rather
> than bit patterns everywhere else), but I generally agree.
> 
> The patch looks good apart from this.
> 
> Thanks,
> Rafael
> 
> 
> > -----Original Message-----
> > From: Lin, Ming M 
> > Sent: Wednesday, December 15, 2010 12:00 AM
> > To: Rafael J. Wysocki
> > Cc: lenb; Moore, Robert; linux-acpi
> > Subject: [RFC PATCH] ACPICA: Improve GPE detect
> > 
> > Hi, Rafael
> > 
> > Currently, the ACPICA GPE interrupt handler reads all the GPE registers to
> > detect which GPEs, if any, have fired. This is done on each and every SCI,
> > since there is no way to know up front if any GPEs have been raised.
> > 
> > However, since ACPICA internally keeps a GPE enable mask for every GPE
> > register, it may be possible to ignore any GPE registers whose internal
> > GPE enable mask is zero. This would eliminate reading the status and enable
> > registers for that particular GPE register pair.
> > 
> > http://www.acpica.org/bugzilla/show_bug.cgi?id=884
> > 
> > I have tested suspend/resume and it works OK.
> > What do you think about below simple patch?
> > 
> > Thanks.
> > ---
> >  drivers/acpi/acpica/evgpe.c |    8 ++++++++
> >  1 files changed, 8 insertions(+), 0 deletions(-)
> > 
> > diff --git a/drivers/acpi/acpica/evgpe.c b/drivers/acpi/acpica/evgpe.c
> > index 7c339d3..49fab43 100644
> > --- a/drivers/acpi/acpica/evgpe.c
> > +++ b/drivers/acpi/acpica/evgpe.c
> > @@ -373,6 +373,14 @@ u32 acpi_ev_gpe_detect(struct acpi_gpe_xrupt_info * gpe_xrupt_list)
> >  
> >  			gpe_register_info = &gpe_block->register_info[i];
> >  
> > +			if (!gpe_register_info->enable_for_run &&
> > +				!gpe_register_info->enable_for_wake) {
> > +
> > +				/* Disabled for both "runtime" and "wakeup", move on */
> > +
> > +				continue;
> > +			}
> > +
> >  			/* Read the Status Register */
> >  
> >  			status =
> > 
> > 
> > 
> 
>