[PATCH/RFC] ACPICA: disable ACPI 2.0 _GTS/_BFS support

[PATCH/RFC] ACPICA: disable ACPI 2.0 _GTS/_BFS support

[Len Brown]

From: Len Brown <len.brown@intel.com>
Organization: Intel Open Source Technology Center

ACPI 2.0 defined two new methods in the suspend/resume sequence,
_GTS (Going to Sleep) and _BFS (Back from Sleep)

They are optional methods, but if the BIOS supplies them,
the OS is supposed to evaluate them immediately before
writing the register to sleep, and immediately after waking up --
a time when interrupts are disabled.

The spec says that they must be self-contained
methods, not calling any other methods, perhaps
because that they are run under unique conditions?

These methodds are evaluated in Linux by
acpi_evaluate_object(), which always kmalloc's
a return structure to conserve stack space.
But kmalloc with interrupts off is problematic --
do we really want to insist on GFP_ATOMIC here?

Now, several years after ACPI 2.0 was released,
we have yet to observe a single implementation of
_GTS/_BFS in the field -- suggesting that they will
never actually be deployed.

So lets keep Linux simple by removing this
theoretical support for _GTS/_BFS, the only
AML methods that mandated being evaluated
with interrupts disabled.

Signed-off-by: Len Brown <len.brown@intel.com>
---
 drivers/acpi/hardware/hwsleep.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/drivers/acpi/hardware/hwsleep.c b/drivers/acpi/hardware/hwsleep.c
index 25dccdf..c209434 100644
--- a/drivers/acpi/hardware/hwsleep.c
+++ b/drivers/acpi/hardware/hwsleep.c
@@ -289,6 +289,7 @@ acpi_status asmlinkage acpi_enter_sleep_state(u8 sleep_state)
 		return_ACPI_STATUS(status);
 	}
 
+#ifdef	ACPI_20_GTS_BFS
 	/* Execute the _GTS method */
 
 	arg_list.count = 1;
@@ -300,6 +301,7 @@ acpi_status asmlinkage acpi_enter_sleep_state(u8 sleep_state)
 	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
 		return_ACPI_STATUS(status);
 	}
+#endif
 
 	/* Get current value of PM1A control */
 
@@ -535,6 +537,7 @@ acpi_status acpi_leave_sleep_state_prep(u8 sleep_state)
 		}
 	}
 
+#ifdef	ACPI_20_GTS_BFS
 	/* Execute the _BFS method */
 
 	arg_list.count = 1;
@@ -546,6 +549,7 @@ acpi_status acpi_leave_sleep_state_prep(u8 sleep_state)
 	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
 		ACPI_EXCEPTION((AE_INFO, status, "During Method _BFS"));
 	}
+#endif
 
 	return_ACPI_STATUS(status);
 }
-- 
1.6.1.rc3.35.gc0ceb

Re: [PATCH/RFC] ACPICA: disable ACPI 2.0 _GTS/_BFS support

[Bjorn Helgaas]

On Friday 19 December 2008 02:23:49 am Len Brown wrote:
> ACPI 2.0 defined two new methods in the suspend/resume sequence,
> _GTS (Going to Sleep) and _BFS (Back from Sleep)
> 
> They are optional methods, but if the BIOS supplies them,
> the OS is supposed to evaluate them immediately before
> writing the register to sleep, and immediately after waking up --
> a time when interrupts are disabled.
> 
> ...
> Now, several years after ACPI 2.0 was released,
> we have yet to observe a single implementation of
> _GTS/_BFS in the field -- suggesting that they will
> never actually be deployed.

There are actually some HP ia64 systems that have _GTS.  All
the rx7600-, rx8600-, and Superdome-class systems I looked at
have it.  But in each case, the method is empty, so I don't
know why they even bothered to implement it.

> So lets keep Linux simple by removing this
> theoretical support for _GTS/_BFS, the only
> AML methods that mandated being evaluated
> with interrupts disabled.

Do you want to print a note that _GTS/_BFS exists, but
we're ignoring it?  If some platform comes along that
uses them, a dmesg note might help debug problems.

Bjorn

Re: [PATCH/RFC] ACPICA: disable ACPI 2.0 _GTS/_BFS support

[Rafael J. Wysocki]

On Friday, 19 of December 2008, Len Brown wrote:
> From: Len Brown <len.brown@intel.com>
> Organization: Intel Open Source Technology Center
> 
> ACPI 2.0 defined two new methods in the suspend/resume sequence,
> _GTS (Going to Sleep) and _BFS (Back from Sleep)
> 
> They are optional methods, but if the BIOS supplies them,
> the OS is supposed to evaluate them immediately before
> writing the register to sleep, and immediately after waking up --
> a time when interrupts are disabled.
> 
> The spec says that they must be self-contained
> methods, not calling any other methods, perhaps
> because that they are run under unique conditions?
> 
> These methodds are evaluated in Linux by
> acpi_evaluate_object(), which always kmalloc's
> a return structure to conserve stack space.
> But kmalloc with interrupts off is problematic --
> do we really want to insist on GFP_ATOMIC here?

Well, I think we should.

Alternatively, we can keep a preallocated buffer for this purpose.

> Now, several years after ACPI 2.0 was released,
> we have yet to observe a single implementation of
> _GTS/_BFS in the field -- suggesting that they will
> never actually be deployed.
> 
> So lets keep Linux simple by removing this
> theoretical support for _GTS/_BFS, the only
> AML methods that mandated being evaluated
> with interrupts disabled.

I'm not sure if I agree with this approach.

My opinion is that we should be able to execute AML code with interrupts
disabled and there should be means to do that.

The entire irqrouter_resume() thing requires us to evaluate AML with interrupts
disabled and in that particular case it also really _makes_ _sense_.

So, IMO dropping _GTS/_BFS doesn't really buy us anything.

Thanks,
Rafael

Re: [PATCH/RFC] ACPICA: disable ACPI 2.0 _GTS/_BFS support

[Zhao Yakui]

On Fri, 2008-12-19 at 17:23 +0800, Len Brown wrote:
> From: Len Brown <len.brown@intel.com>
> Organization: Intel Open Source Technology Center
> 
> ACPI 2.0 defined two new methods in the suspend/resume sequence,
> _GTS (Going to Sleep) and _BFS (Back from Sleep)

Is the purpose of this patch that in course of suspend/resume the ACPI
object won't be evaluated again when irq is disabled? And this can avoid
the preemption point after evaluating the ACPI object.
   
   If so, this issue can be fixed by the patch
   http://bugzilla.kernel.org/attachment.cgi?id=19370&action=view


> 
> They are optional methods, but if the BIOS supplies them,
> the OS is supposed to evaluate them immediately before
> writing the register to sleep, and immediately after waking up --
> a time when interrupts are disabled.
> 
> The spec says that they must be self-contained
> methods, not calling any other methods, perhaps
> because that they are run under unique conditions?
> 
> These methodds are evaluated in Linux by
> acpi_evaluate_object(), which always kmalloc's
> a return structure to conserve stack space.
> But kmalloc with interrupts off is problematic --
> do we really want to insist on GFP_ATOMIC here?
> 
> Now, several years after ACPI 2.0 was released,
> we have yet to observe a single implementation of
> _GTS/_BFS in the field -- suggesting that they will
> never actually be deployed.
> 
> So lets keep Linux simple by removing this
> theoretical support for _GTS/_BFS, the only
> AML methods that mandated being evaluated
> with interrupts disabled.
> 
> Signed-off-by: Len Brown <len.brown@intel.com>
> ---
>  drivers/acpi/hardware/hwsleep.c |    4 ++++
>  1 files changed, 4 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/acpi/hardware/hwsleep.c b/drivers/acpi/hardware/hwsleep.c
> index 25dccdf..c209434 100644
> --- a/drivers/acpi/hardware/hwsleep.c
> +++ b/drivers/acpi/hardware/hwsleep.c
> @@ -289,6 +289,7 @@ acpi_status asmlinkage acpi_enter_sleep_state(u8 sleep_state)
>  		return_ACPI_STATUS(status);
>  	}
>  
> +#ifdef	ACPI_20_GTS_BFS
>  	/* Execute the _GTS method */
>  
>  	arg_list.count = 1;
> @@ -300,6 +301,7 @@ acpi_status asmlinkage acpi_enter_sleep_state(u8 sleep_state)
>  	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
>  		return_ACPI_STATUS(status);
>  	}
> +#endif
>  
>  	/* Get current value of PM1A control */
>  
> @@ -535,6 +537,7 @@ acpi_status acpi_leave_sleep_state_prep(u8 sleep_state)
>  		}
>  	}
>  
> +#ifdef	ACPI_20_GTS_BFS
>  	/* Execute the _BFS method */
>  
>  	arg_list.count = 1;
> @@ -546,6 +549,7 @@ acpi_status acpi_leave_sleep_state_prep(u8 sleep_state)
>  	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
>  		ACPI_EXCEPTION((AE_INFO, status, "During Method _BFS"));
>  	}
> +#endif
>  
>  	return_ACPI_STATUS(status);
>  }

Re: [PATCH/RFC] ACPICA: disable ACPI 2.0 _GTS/_BFS support

[Len Brown]

On Fri, 19 Dec 2008, Bjorn Helgaas wrote:

> On Friday 19 December 2008 02:23:49 am Len Brown wrote:
> > ACPI 2.0 defined two new methods in the suspend/resume sequence,
> > _GTS (Going to Sleep) and _BFS (Back from Sleep)
> > 
> > They are optional methods, but if the BIOS supplies them,
> > the OS is supposed to evaluate them immediately before
> > writing the register to sleep, and immediately after waking up --
> > a time when interrupts are disabled.
> > 
> > ...
> > Now, several years after ACPI 2.0 was released,
> > we have yet to observe a single implementation of
> > _GTS/_BFS in the field -- suggesting that they will
> > never actually be deployed.
> 
> There are actually some HP ia64 systems that have _GTS.  All
> the rx7600-, rx8600-, and Superdome-class systems I looked at
> have it.  But in each case, the method is empty, so I don't
> know why they even bothered to implement it.

I've seen "place holders" in AML many times.
_SCP is another popular one that is almost never implemented,
but sometimes exists and does nothing -- which is
quite misleading...

> > So lets keep Linux simple by removing this
> > theoretical support for _GTS/_BFS, the only
> > AML methods that mandated being evaluated
> > with interrupts disabled.
> 
> Do you want to print a note that _GTS/_BFS exists, but
> we're ignoring it?  If some platform comes along that
> uses them, a dmesg note might help debug problems.

Good idea.

Probably the smart thing to do would be to print a
message if they exist and are non-empty.
Otherwise we'd have to DMI those HP boxes with
empty methods.

-- Len Brown, Intel Open Source Technology Center