PCI / ACPI / PM: Resume bridges w/o drivers on suspend-to-RAM

Commit Message

Rafael J. Wysocki June 26, 2018, 10:06 a.m. UTC

From: Rafael J. Wysocki <[e-mail]>

It is reported that commit c62ec4610c40 (PM / core: Fix direct_complete
handling for devices with no callbacks) introduced a system suspend
regression on Samsung 305V4A by allowing a PCI bridge (not a PCIe
port) to stay in D3 over suspend-to-RAM, which is a side effect of
setting power.direct_complete for the children of that bridge that
have no PM callbacks.

On the majority of systems PCI bridges are not allowed to be
runtime-suspended (the power/control sysfs attribute is set to "on"
for them by default), but user space can change that setting and if
it does so and a given bridge has no children with PM callbacks, the
direct_complete optimization will be applied to it and it will stay
in suspend over system suspend.  Apparently, that confuses the
platform firmware on the affected machine and that may very well
happen elsewhere, so avoid the direct_complete optimization for
PCI bridges with no drivers (if there is a driver, it should take
care of the PM handling) on suspend-to-RAM altogether (that should
not matter for suspend-to-idle as platform firmware is not involved
in it).

Fixes: c62ec4610c40 (PM / core: Fix direct_complete handling for devices with no callbacks)
Link: https://bugzilla.kernel.org/show_bug.cgi?id=199941
Reported-by: [e-mail]
Tested-by: [e-mail]
Signed-off-by: Rafael J. Wysocki <[e-mail]>
---
 drivers/pci/pci-acpi.c |    8 ++++++++
 1 file changed, 8 insertions(+)


--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Mika Westerberg June 26, 2018, 10:33 a.m. UTC | #1

On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <[e-mail]>
> 
> It is reported that commit c62ec4610c40 (PM / core: Fix direct_complete
> handling for devices with no callbacks) introduced a system suspend
> regression on Samsung 305V4A by allowing a PCI bridge (not a PCIe
> port) to stay in D3 over suspend-to-RAM, which is a side effect of
> setting power.direct_complete for the children of that bridge that
> have no PM callbacks.
> 
> On the majority of systems PCI bridges are not allowed to be
> runtime-suspended (the power/control sysfs attribute is set to "on"
> for them by default), but user space can change that setting and if
> it does so and a given bridge has no children with PM callbacks, the
> direct_complete optimization will be applied to it and it will stay
> in suspend over system suspend.  Apparently, that confuses the
> platform firmware on the affected machine and that may very well
> happen elsewhere, so avoid the direct_complete optimization for
> PCI bridges with no drivers (if there is a driver, it should take
> care of the PM handling) on suspend-to-RAM altogether (that should
> not matter for suspend-to-idle as platform firmware is not involved
> in it).
> 
> Fixes: c62ec4610c40 (PM / core: Fix direct_complete handling for devices with no callbacks)
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=199941
> Reported-by: [e-mail]
> Tested-by: [e-mail]
> Signed-off-by: Rafael J. Wysocki <[e-mail]>

Reviewed-by: Mika Westerberg <[e-mail]>
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Bjorn Helgaas June 26, 2018, 2:01 p.m. UTC | #2

On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <[e-mail]>
> 
> It is reported that commit c62ec4610c40 (PM / core: Fix direct_complete
> handling for devices with no callbacks) introduced a system suspend
> regression on Samsung 305V4A by allowing a PCI bridge (not a PCIe
> port) to stay in D3 over suspend-to-RAM, which is a side effect of
> setting power.direct_complete for the children of that bridge that
> have no PM callbacks.
> 
> On the majority of systems PCI bridges are not allowed to be
> runtime-suspended (the power/control sysfs attribute is set to "on"
> for them by default), but user space can change that setting and if
> it does so and a given bridge has no children with PM callbacks, the
> direct_complete optimization will be applied to it and it will stay
> in suspend over system suspend.  Apparently, that confuses the

"stay in D3 over system suspend"? (just to be explicit about what "in
suspend" means)

> platform firmware on the affected machine and that may very well
> happen elsewhere, so avoid the direct_complete optimization for
> PCI bridges with no drivers (if there is a driver, it should take
> care of the PM handling) on suspend-to-RAM altogether (that should
> not matter for suspend-to-idle as platform firmware is not involved
> in it).
> 
> Fixes: c62ec4610c40 (PM / core: Fix direct_complete handling for devices with no callbacks)
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=199941
> Reported-by: [e-mail]
> Tested-by: [e-mail]
> Signed-off-by: Rafael J. Wysocki <[e-mail]>
> ---
>  drivers/pci/pci-acpi.c |    8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> Index: linux-pm/drivers/pci/pci-acpi.c
> ===================================================================
> --- linux-pm.orig/drivers/pci/pci-acpi.c
> +++ linux-pm/drivers/pci/pci-acpi.c
> @@ -638,6 +638,14 @@ static bool acpi_pci_need_resume(struct
>  	if (acpi_target_system_state() == ACPI_STATE_S0)
>  		return false;
>  
> +	/*
> +	 * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
> +	 * confuses the platform firmware, so avoid doing that, unless the
> +	 * bridge has a driver that should take care of PM handling.
> +	 */
> +	if (pci_is_bridge(dev) && !dev->driver)
> +		return true;

It sounds like the question of whether leaving a bridge in D3 confuses
the firmware has a platform-specific answer.  How does the driver PM
handling know how to do the right thing?  Does it need to know whether
it's safe to put the device in D3?  Or maybe a device is never put in
D3 for system suspend if it has a driver?

But I'm just kibbitzing; since you merged c62ec4610c40, I'm happy if
you also merge this:

Acked-by: Bjorn Helgaas <[e-mail]>

>  	return !!adev->power.flags.dsw_present;
>  }
>  
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
> the body of a message to [e-mail]
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Rafael J. Wysocki June 26, 2018, 2:22 p.m. UTC | #3

On Tue, Jun 26, 2018 at 4:01 PM, Bjorn Helgaas <[e-mail]> wrote:
> On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
>> From: Rafael J. Wysocki <[e-mail]>
>>
>> It is reported that commit c62ec4610c40 (PM / core: Fix direct_complete
>> handling for devices with no callbacks) introduced a system suspend
>> regression on Samsung 305V4A by allowing a PCI bridge (not a PCIe
>> port) to stay in D3 over suspend-to-RAM, which is a side effect of
>> setting power.direct_complete for the children of that bridge that
>> have no PM callbacks.
>>
>> On the majority of systems PCI bridges are not allowed to be
>> runtime-suspended (the power/control sysfs attribute is set to "on"
>> for them by default), but user space can change that setting and if
>> it does so and a given bridge has no children with PM callbacks, the
>> direct_complete optimization will be applied to it and it will stay
>> in suspend over system suspend.  Apparently, that confuses the
>
> "stay in D3 over system suspend"? (just to be explicit about what "in
> suspend" means)

OK

>> platform firmware on the affected machine and that may very well
>> happen elsewhere, so avoid the direct_complete optimization for
>> PCI bridges with no drivers (if there is a driver, it should take
>> care of the PM handling) on suspend-to-RAM altogether (that should
>> not matter for suspend-to-idle as platform firmware is not involved
>> in it).
>>
>> Fixes: c62ec4610c40 (PM / core: Fix direct_complete handling for devices with no callbacks)
>> Link: https://bugzilla.kernel.org/show_bug.cgi?id=199941
>> Reported-by: [e-mail]
>> Tested-by: [e-mail]
>> Signed-off-by: Rafael J. Wysocki <[e-mail]>
>> ---
>>  drivers/pci/pci-acpi.c |    8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> Index: linux-pm/drivers/pci/pci-acpi.c
>> ===================================================================
>> --- linux-pm.orig/drivers/pci/pci-acpi.c
>> +++ linux-pm/drivers/pci/pci-acpi.c
>> @@ -638,6 +638,14 @@ static bool acpi_pci_need_resume(struct
>>       if (acpi_target_system_state() == ACPI_STATE_S0)
>>               return false;
>>
>> +     /*
>> +      * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
>> +      * confuses the platform firmware, so avoid doing that, unless the
>> +      * bridge has a driver that should take care of PM handling.
>> +      */
>> +     if (pci_is_bridge(dev) && !dev->driver)
>> +             return true;
>
> It sounds like the question of whether leaving a bridge in D3 confuses
> the firmware has a platform-specific answer.

Well, it may confuse the platform firmware in general.

> How does the driver PM handling know how to do the right thing?

For endpoints this is not an issue as they always have been expected
to be in D3 before passing control to the platform firmware on S3
entry, but we've never done that for bridges by default, except for
PCIe ports with PM enabled (in which case the driver decides whether
or not to enable it).

> Does it need to know whether
> it's safe to put the device in D3?  Or maybe a device is never put in
> D3 for system suspend if it has a driver?

As per the above, endpoints should be OK (or they are quirky if not),
but bridges are sort of a gray area.  PCIe ports for which we enable
PM should be fine, for the rest it's better to be conservative and
resume them IMO.

> But I'm just kibbitzing; since you merged c62ec4610c40, I'm happy if
> you also merge this:
>
> Acked-by: Bjorn Helgaas <[e-mail]>
>
>>       return !!adev->power.flags.dsw_present;
>>  }

Thanks!
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Bjorn Helgaas June 26, 2018, 5:14 p.m. UTC | #4

On Tue, Jun 26, 2018 at 04:22:00PM +0200, Rafael J. Wysocki wrote:
> On Tue, Jun 26, 2018 at 4:01 PM, Bjorn Helgaas <[e-mail]> wrote:
> > On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
> >> From: Rafael J. Wysocki <[e-mail]>

> >> +     /*
> >> +      * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
> >> +      * confuses the platform firmware, so avoid doing that, unless the
> >> +      * bridge has a driver that should take care of PM handling.
> >> +      */
> >> +     if (pci_is_bridge(dev) && !dev->driver)
> >> +             return true;
> >
> > It sounds like the question of whether leaving a bridge in D3 confuses
> > the firmware has a platform-specific answer.
> 
> Well, it may confuse the platform firmware in general.
> 
> > How does the driver PM handling know how to do the right thing?
> 
> For endpoints this is not an issue as they always have been expected
> to be in D3 before passing control to the platform firmware on S3
> entry, but we've never done that for bridges by default, except for
> PCIe ports with PM enabled (in which case the driver decides whether
> or not to enable it).

If there's any spec reference for the expected power states of devices
when entering S3, that would be useful here.  I can't tell if there's
any guidance for this or if it's just figured out experimentally.

> > Does it need to know whether
> > it's safe to put the device in D3?  Or maybe a device is never put in
> > D3 for system suspend if it has a driver?
> 
> As per the above, endpoints should be OK (or they are quirky if not),
> but bridges are sort of a gray area.  PCIe ports for which we enable
> PM should be fine, for the rest it's better to be conservative and
> resume them IMO.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Rafael J. Wysocki June 26, 2018, 5:19 p.m. UTC | #5

On Tue, Jun 26, 2018 at 7:14 PM, Bjorn Helgaas <[e-mail]> wrote:
> On Tue, Jun 26, 2018 at 04:22:00PM +0200, Rafael J. Wysocki wrote:
>> On Tue, Jun 26, 2018 at 4:01 PM, Bjorn Helgaas <[e-mail]> wrote:
>> > On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
>> >> From: Rafael J. Wysocki <[e-mail]>
>
>> >> +     /*
>> >> +      * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
>> >> +      * confuses the platform firmware, so avoid doing that, unless the
>> >> +      * bridge has a driver that should take care of PM handling.
>> >> +      */
>> >> +     if (pci_is_bridge(dev) && !dev->driver)
>> >> +             return true;
>> >
>> > It sounds like the question of whether leaving a bridge in D3 confuses
>> > the firmware has a platform-specific answer.
>>
>> Well, it may confuse the platform firmware in general.
>>
>> > How does the driver PM handling know how to do the right thing?
>>
>> For endpoints this is not an issue as they always have been expected
>> to be in D3 before passing control to the platform firmware on S3
>> entry, but we've never done that for bridges by default, except for
>> PCIe ports with PM enabled (in which case the driver decides whether
>> or not to enable it).
>
> If there's any spec reference for the expected power states of devices
> when entering S3, that would be useful here.  I can't tell if there's
> any guidance for this or if it's just figured out experimentally.

It is not direct, but Section 16.1.6 of ACPI 6.2 says this in step 4
of the system suspend outline:

OSPM places all device drivers into their respective Dx state. If the
device is enabled for wake,
it enters the Dx state associated with the wake capability. If the
device is not enabled to wake
the system, it enters the D3 state.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Bjorn Helgaas June 26, 2018, 8:32 p.m. UTC | #6

On Tue, Jun 26, 2018 at 07:19:29PM +0200, Rafael J. Wysocki wrote:
> On Tue, Jun 26, 2018 at 7:14 PM, Bjorn Helgaas <[e-mail]> wrote:
> > On Tue, Jun 26, 2018 at 04:22:00PM +0200, Rafael J. Wysocki wrote:
> >> On Tue, Jun 26, 2018 at 4:01 PM, Bjorn Helgaas <[e-mail]> wrote:
> >> > On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
> >> >> From: Rafael J. Wysocki <[e-mail]>
> >
> >> >> +     /*
> >> >> +      * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
> >> >> +      * confuses the platform firmware, so avoid doing that, unless the
> >> >> +      * bridge has a driver that should take care of PM handling.
> >> >> +      */
> >> >> +     if (pci_is_bridge(dev) && !dev->driver)
> >> >> +             return true;
> >> >
> >> > It sounds like the question of whether leaving a bridge in D3 confuses
> >> > the firmware has a platform-specific answer.
> >>
> >> Well, it may confuse the platform firmware in general.
> >>
> >> > How does the driver PM handling know how to do the right thing?
> >>
> >> For endpoints this is not an issue as they always have been expected
> >> to be in D3 before passing control to the platform firmware on S3
> >> entry, but we've never done that for bridges by default, except for
> >> PCIe ports with PM enabled (in which case the driver decides whether
> >> or not to enable it).
> >
> > If there's any spec reference for the expected power states of devices
> > when entering S3, that would be useful here.  I can't tell if there's
> > any guidance for this or if it's just figured out experimentally.
> 
> It is not direct, but Section 16.1.6 of ACPI 6.2 says this in step 4
> of the system suspend outline:
> 
> OSPM places all device drivers into their respective Dx state. If the
> device is enabled for wake,
> it enters the Dx state associated with the wake capability. If the
> device is not enabled to wake
> the system, it enters the D3 state.

Thanks, that's a very useful citation!

Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Rafael J. Wysocki June 27, 2018, 7:16 a.m. UTC | #7

On Tue, Jun 26, 2018 at 10:32 PM, Bjorn Helgaas <[e-mail]> wrote:
> On Tue, Jun 26, 2018 at 07:19:29PM +0200, Rafael J. Wysocki wrote:
>> On Tue, Jun 26, 2018 at 7:14 PM, Bjorn Helgaas <[e-mail]> wrote:
>> > On Tue, Jun 26, 2018 at 04:22:00PM +0200, Rafael J. Wysocki wrote:
>> >> On Tue, Jun 26, 2018 at 4:01 PM, Bjorn Helgaas <[e-mail]> wrote:
>> >> > On Tue, Jun 26, 2018 at 12:06:01PM +0200, Rafael J. Wysocki wrote:
>> >> >> From: Rafael J. Wysocki <[e-mail]>
>> >
>> >> >> +     /*
>> >> >> +      * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
>> >> >> +      * confuses the platform firmware, so avoid doing that, unless the
>> >> >> +      * bridge has a driver that should take care of PM handling.
>> >> >> +      */
>> >> >> +     if (pci_is_bridge(dev) && !dev->driver)
>> >> >> +             return true;
>> >> >
>> >> > It sounds like the question of whether leaving a bridge in D3 confuses
>> >> > the firmware has a platform-specific answer.
>> >>
>> >> Well, it may confuse the platform firmware in general.
>> >>
>> >> > How does the driver PM handling know how to do the right thing?
>> >>
>> >> For endpoints this is not an issue as they always have been expected
>> >> to be in D3 before passing control to the platform firmware on S3
>> >> entry, but we've never done that for bridges by default, except for
>> >> PCIe ports with PM enabled (in which case the driver decides whether
>> >> or not to enable it).
>> >
>> > If there's any spec reference for the expected power states of devices
>> > when entering S3, that would be useful here.  I can't tell if there's
>> > any guidance for this or if it's just figured out experimentally.
>>
>> It is not direct, but Section 16.1.6 of ACPI 6.2 says this in step 4
>> of the system suspend outline:
>>
>> OSPM places all device drivers into their respective Dx state. If the
>> device is enabled for wake,
>> it enters the Dx state associated with the wake capability. If the
>> device is not enabled to wake
>> the system, it enters the D3 state.
>
> Thanks, that's a very useful citation!

You're welcome! :-)

I actually need to withdraw the $subject patch as it doesn't help (I
misread the reporters response) and the one that does work is a bit
too intrusive IMO, so I'll investigate this a bit more.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

Index: linux-pm/drivers/pci/pci-acpi.c
===================================================================
--- linux-pm.orig/drivers/pci/pci-acpi.c
+++ linux-pm/drivers/pci/pci-acpi.c
@@ -638,6 +638,14 @@  static bool acpi_pci_need_resume(struct
 	if (acpi_target_system_state() == ACPI_STATE_S0)
 		return false;
 
+	/*
+	 * In some cases (eg. Samsung 305V4A) leaving a bridge in suspend
+	 * confuses the platform firmware, so avoid doing that, unless the
+	 * bridge has a driver that should take care of PM handling.
+	 */
+	if (pci_is_bridge(dev) && !dev->driver)
+		return true;
+
 	return !!adev->power.flags.dsw_present;
 }