LinuxLists.cc - [PATCH] PCI: PM: Consolidate runtime resume and system resume paths

2019-10-14 10:56:22

Subject: [PATCH] PCI: PM: Consolidate runtime resume and system resume paths

From: Rafael J. Wysocki <[email protected]>

There is an arbitrary difference between the system resume and
runtime resume code paths for PCI devices regarding the delay to
apply when switching the devices from D3cold to D0.

Namely, pci_restore_standard_config() used in the runtime resume
code path calls pci_set_power_state() which in turn invokes
__pci_start_power_transition() to power up the device through the
platform firmware and that function applies the transition delay
(as per PCI Express Base Specification Revision 2.0, Section 6.6.1).
However, pci_pm_default_resume_early() used in the system resume
code path calls pci_power_up() which doesn't apply the delay at
all and that causes issues to occur during resume from
suspend-to-idle on some systems where the delay is required.

Since there is no reason for that difference to exist, modify
pci_pm_default_resume_early() to invoke pci_restore_standard_config()
instead of pci_power_up() and drop the latter, but in order to
prevent the ACPI power state values (cached by the ACPI layer) from
becoming stale in some cases during resume from suspend-to-RAM
(ACPI S3), as per commit cc2893b6af52 ("PCI: Ensure we re-enable
devices on resume"), refresh the ACPI power state information in
pci_pm_default_resume_early() in that case.

[Note that while this change should take the issue originally
addressed by commit cc2893b6af52 ("PCI: Ensure we re-enable devices
on resume") into account in a generally safer way, an alternative
would be to make pci_power_up() use __pci_start_power_transition()
instead of calling platform_pci_set_power_state() directly.]

Fixes: db288c9c5f9d ("PCI / PM: restore the original behavior of pci_set_power_state()")
Reported-by: Daniel Drake <[email protected]>
Tested-by: Daniel Drake <[email protected]>
Link: https://lore.kernel.org/linux-pm/CAD8Lp44TYxrMgPLkHCqF9hv6smEurMXvmmvmtyFhZ6Q4SE+dig@mail.gmail.com/T/#m21be74af263c6a34f36e0fc5c77c5449d9406925
Signed-off-by: Rafael J. Wysocki <[email protected]>
---
drivers/pci/pci-driver.c | 8 +++++---
drivers/pci/pci.c | 15 ---------------
drivers/pci/pci.h | 1 -
3 files changed, 5 insertions(+), 19 deletions(-)

Index: linux-pm/drivers/pci/pci-driver.c
===================================================================
--- linux-pm.orig/drivers/pci/pci-driver.c
+++ linux-pm/drivers/pci/pci-driver.c
@@ -523,9 +523,10 @@ static int pci_restore_standard_config(s

static void pci_pm_default_resume_early(struct pci_dev *pci_dev)
{
- pci_power_up(pci_dev);
- pci_restore_state(pci_dev);
- pci_pme_restore(pci_dev);
+ if (pm_resume_via_firmware())
+ pci_refresh_power_state(pci_dev);
+
+ pci_restore_standard_config(pci_dev);
}

/*
@@ -713,6 +714,7 @@ static void pci_pm_complete(struct devic
pci_power_t pre_sleep_state = pci_dev->current_state;

pci_refresh_power_state(pci_dev);
+ pci_update_current_state(pci_dev, pci_dev->current_state);
/*
* On platforms with ACPI this check may also trigger for
* devices sharing power resources if one of those power
Index: linux-pm/drivers/pci/pci.c
===================================================================
--- linux-pm.orig/drivers/pci/pci.c
+++ linux-pm/drivers/pci/pci.c
@@ -954,21 +954,6 @@ void pci_refresh_power_state(struct pci_
{
if (platform_pci_power_manageable(dev))
platform_pci_refresh_power_state(dev);
-
- pci_update_current_state(dev, dev->current_state);
-}
-
-/**
- * pci_power_up - Put the given device into D0 forcibly
- * @dev: PCI device to power up
- */
-void pci_power_up(struct pci_dev *dev)
-{
- if (platform_pci_power_manageable(dev))
- platform_pci_set_power_state(dev, PCI_D0);
-
- pci_raw_set_power_state(dev, PCI_D0);
- pci_update_current_state(dev, PCI_D0);
}

/**
Index: linux-pm/drivers/pci/pci.h
===================================================================
--- linux-pm.orig/drivers/pci/pci.h
+++ linux-pm/drivers/pci/pci.h
@@ -85,7 +85,6 @@ struct pci_platform_pm_ops {
int pci_set_platform_pm(const struct pci_platform_pm_ops *ops);
void pci_update_current_state(struct pci_dev *dev, pci_power_t state);
void pci_refresh_power_state(struct pci_dev *dev);
-void pci_power_up(struct pci_dev *dev);
void pci_disable_enabled_device(struct pci_dev *dev);
int pci_finish_runtime_suspend(struct pci_dev *dev);
void pcie_clear_root_pme_status(struct pci_dev *dev);

2019-10-14 11:21:53

by Rafael J. Wysocki

[permalink] [raw]

Subject: Re: [PATCH] PCI: PM: Consolidate runtime resume and system resume paths

On Monday, October 14, 2019 12:51:31 PM CEST Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <[email protected]>
>
> There is an arbitrary difference between the system resume and
> runtime resume code paths for PCI devices regarding the delay to
> apply when switching the devices from D3cold to D0.
>
> Namely, pci_restore_standard_config() used in the runtime resume
> code path calls pci_set_power_state() which in turn invokes
> __pci_start_power_transition() to power up the device through the
> platform firmware and that function applies the transition delay
> (as per PCI Express Base Specification Revision 2.0, Section 6.6.1).
> However, pci_pm_default_resume_early() used in the system resume
> code path calls pci_power_up() which doesn't apply the delay at
> all and that causes issues to occur during resume from
> suspend-to-idle on some systems where the delay is required.
>
> Since there is no reason for that difference to exist, modify
> pci_pm_default_resume_early() to invoke pci_restore_standard_config()
> instead of pci_power_up() and drop the latter, but in order to
> prevent the ACPI power state values (cached by the ACPI layer) from
> becoming stale in some cases during resume from suspend-to-RAM
> (ACPI S3), as per commit cc2893b6af52 ("PCI: Ensure we re-enable
> devices on resume"), refresh the ACPI power state information in
> pci_pm_default_resume_early() in that case.
>
> [Note that while this change should take the issue originally
> addressed by commit cc2893b6af52 ("PCI: Ensure we re-enable devices
> on resume") into account in a generally safer way, an alternative
> would be to make pci_power_up() use __pci_start_power_transition()
> instead of calling platform_pci_set_power_state() directly.]
>
> Fixes: db288c9c5f9d ("PCI / PM: restore the original behavior of pci_set_power_state()")
> Reported-by: Daniel Drake <[email protected]>
> Tested-by: Daniel Drake <[email protected]>
> Link: https://lore.kernel.org/linux-pm/CAD8Lp44TYxrMgPLkHCqF9hv6smEurMXvmmvmtyFhZ6Q4SE+dig@mail.gmail.com/T/#m21be74af263c6a34f36e0fc5c77c5449d9406925
> Signed-off-by: Rafael J. Wysocki <[email protected]>
> ---
> drivers/pci/pci-driver.c | 8 +++++---
> drivers/pci/pci.c | 15 ---------------
> drivers/pci/pci.h | 1 -
> 3 files changed, 5 insertions(+), 19 deletions(-)
>
> Index: linux-pm/drivers/pci/pci-driver.c
> ===================================================================
> --- linux-pm.orig/drivers/pci/pci-driver.c
> +++ linux-pm/drivers/pci/pci-driver.c
> @@ -523,9 +523,10 @@ static int pci_restore_standard_config(s
>
> static void pci_pm_default_resume_early(struct pci_dev *pci_dev)
> {
> - pci_power_up(pci_dev);
> - pci_restore_state(pci_dev);
> - pci_pme_restore(pci_dev);
> + if (pm_resume_via_firmware())
> + pci_refresh_power_state(pci_dev);

Well, this is still not going to work if the ACPI power state after the update
above is not D0, but the pci_update_current_state() in pci_restore_standard_config()
returns D0, which was the case that triggered commit cc2893b6af52 IIRC.

So scratch this one, please, and I'll submit the safer option.

> +
> + pci_restore_standard_config(pci_dev);
> }
>
> /*
> @@ -713,6 +714,7 @@ static void pci_pm_complete(struct devic
> pci_power_t pre_sleep_state = pci_dev->current_state;
>
> pci_refresh_power_state(pci_dev);
> + pci_update_current_state(pci_dev, pci_dev->current_state);
> /*
> * On platforms with ACPI this check may also trigger for
> * devices sharing power resources if one of those power

2019-10-14 11:25:31

by Rafael J. Wysocki

[permalink] [raw]

Subject: [PATCH] PCI: PM: Fix pci_power_up()

2019-10-15 05:41:26

by Daniel Drake

[permalink] [raw]

Subject: Re: [PATCH] PCI: PM: Fix pci_power_up()

On Mon, Oct 14, 2019 at 7:25 PM Rafael J. Wysocki <[email protected]> wrote:
> Since there is no reason for that difference to exist, modify
> pci_power_up() to follow pci_set_power_state() more closely and
> invoke __pci_start_power_transition() from there to call the
> platform firmware to power up the device (in case that's necessary).
>
> Fixes: db288c9c5f9d ("PCI / PM: restore the original behavior of pci_set_power_state()")
> Reported-by: Daniel Drake <[email protected]>
> Link: https://lore.kernel.org/linux-pm/CAD8Lp44TYxrMgPLkHCqF9hv6smEurMXvmmvmtyFhZ6Q4SE+dig@mail.gmail.com/T/#m21be74af263c6a34f36e0fc5c77c5449d9406925
> Signed-off-by: Rafael J. Wysocki <[email protected]>
> ---
>
> Daniel, please test this one.

This one is working too, thanks

Daniel

2019-10-15 11:05:34

by Rafael J. Wysocki

[permalink] [raw]

Subject: Re: [PATCH] PCI: PM: Fix pci_power_up()

On Tue, Oct 15, 2019 at 7:11 AM Daniel Drake <[email protected]> wrote:
>
> On Mon, Oct 14, 2019 at 7:25 PM Rafael J. Wysocki <[email protected]> wrote:
> > Since there is no reason for that difference to exist, modify
> > pci_power_up() to follow pci_set_power_state() more closely and
> > invoke __pci_start_power_transition() from there to call the
> > platform firmware to power up the device (in case that's necessary).
> >
> > Fixes: db288c9c5f9d ("PCI / PM: restore the original behavior of pci_set_power_state()")
> > Reported-by: Daniel Drake <[email protected]>
> > Link: https://lore.kernel.org/linux-pm/CAD8Lp44TYxrMgPLkHCqF9hv6smEurMXvmmvmtyFhZ6Q4SE+dig@mail.gmail.com/T/#m21be74af263c6a34f36e0fc5c77c5449d9406925
> > Signed-off-by: Rafael J. Wysocki <[email protected]>
> > ---
> >
> > Daniel, please test this one.
>
> This one is working too, thanks

Thank you!

Bjorn, any concerns?

2019-10-16 02:01:37

by Bjorn Helgaas

[permalink] [raw]

Subject: Re: [PATCH] PCI: PM: Fix pci_power_up()

On Mon, Oct 14, 2019 at 01:25:00PM +0200, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <[email protected]>
>
> There is an arbitrary difference between the system resume and
> runtime resume code paths for PCI devices regarding the delay to
> apply when switching the devices from D3cold to D0.
>
> Namely, pci_restore_standard_config() used in the runtime resume
> code path calls pci_set_power_state() which in turn invokes
> __pci_start_power_transition() to power up the device through the
> platform firmware and that function applies the transition delay
> (as per PCI Express Base Specification Revision 2.0, Section 6.6.1).
> However, pci_pm_default_resume_early() used in the system resume
> code path calls pci_power_up() which doesn't apply the delay at
> all and that causes issues to occur during resume from
> suspend-to-idle on some systems where the delay is required.
>
> Since there is no reason for that difference to exist, modify
> pci_power_up() to follow pci_set_power_state() more closely and
> invoke __pci_start_power_transition() from there to call the
> platform firmware to power up the device (in case that's necessary).
>
> Fixes: db288c9c5f9d ("PCI / PM: restore the original behavior of pci_set_power_state()")
> Reported-by: Daniel Drake <[email protected]>
> Link: https://lore.kernel.org/linux-pm/CAD8Lp44TYxrMgPLkHCqF9hv6smEurMXvmmvmtyFhZ6Q4SE+dig@mail.gmail.com/T/#m21be74af263c6a34f36e0fc5c77c5449d9406925
> Signed-off-by: Rafael J. Wysocki <[email protected]>
> ---
>
> Daniel, please test this one.
>
> ---
> drivers/pci/pci.c | 24 +++++++++++-------------
> 1 file changed, 11 insertions(+), 13 deletions(-)
>
> Index: linux-pm/drivers/pci/pci.c
> ===================================================================
> --- linux-pm.orig/drivers/pci/pci.c
> +++ linux-pm/drivers/pci/pci.c
> @@ -959,19 +959,6 @@ void pci_refresh_power_state(struct pci_
> }
>
> /**
> - * pci_power_up - Put the given device into D0 forcibly
> - * @dev: PCI device to power up
> - */
> -void pci_power_up(struct pci_dev *dev)
> -{
> - if (platform_pci_power_manageable(dev))
> - platform_pci_set_power_state(dev, PCI_D0);
> -
> - pci_raw_set_power_state(dev, PCI_D0);
> - pci_update_current_state(dev, PCI_D0);
> -}
> -
> -/**
> * pci_platform_power_transition - Use platform to change device power state
> * @dev: PCI device to handle.
> * @state: State to put the device into.
> @@ -1154,6 +1141,17 @@ int pci_set_power_state(struct pci_dev *
> EXPORT_SYMBOL(pci_set_power_state);
>
> /**
> + * pci_power_up - Put the given device into D0 forcibly

Not specifically for this patch, but what does "forcibly" mean?

> + * @dev: PCI device to power up
> + */
> +void pci_power_up(struct pci_dev *dev)
> +{
> + __pci_start_power_transition(dev, PCI_D0);
> + pci_raw_set_power_state(dev, PCI_D0);
> + pci_update_current_state(dev, PCI_D0);

There's not very much difference between:

pci_power_up(dev);

and

pci_set_power_state(dev, PCI_D0);

It looks like the main difference is that pci_set_power_state() calls
__pci_complete_power_transition(), which ultimately calls
acpi_pci_set_power_state() (for ACPI systems).

So maybe "forcibly" means something like "ignoring any platform power
management methods"? It's not obvious to me when we should skip the
platform stuff or whether the skipping should be done at the high
level (like calling either pci_power_up() or pci_set_power_state()) or
at a lower level (e.g., if everybody called pci_set_power_state() and
it could internally tell whether we're skipping the platform part).

If we could unify the paths as much as possible, that would be nice,
but if it's not feasible, it's not feasible. If you'd like me to push
this for v5.4, let me know, otherwise you can apply my:

Acked-by: Bjorn Helgaas <[email protected]>

> +}
> +
> +/**
> * pci_choose_state - Choose the power state of a PCI device
> * @dev: PCI device to be suspended
> * @state: target sleep state for the whole system. This is the value
>
>
>

2019-10-16 07:09:01

by Rafael J. Wysocki

[permalink] [raw]

Subject: Re: [PATCH] PCI: PM: Fix pci_power_up()

On Tue, Oct 15, 2019 at 9:20 PM Bjorn Helgaas <[email protected]> wrote:
>
> On Mon, Oct 14, 2019 at 01:25:00PM +0200, Rafael J. Wysocki wrote:
> > From: Rafael J. Wysocki <[email protected]>
> >
> > There is an arbitrary difference between the system resume and
> > runtime resume code paths for PCI devices regarding the delay to
> > apply when switching the devices from D3cold to D0.
> >
> > Namely, pci_restore_standard_config() used in the runtime resume
> > code path calls pci_set_power_state() which in turn invokes
> > __pci_start_power_transition() to power up the device through the
> > platform firmware and that function applies the transition delay
> > (as per PCI Express Base Specification Revision 2.0, Section 6.6.1).
> > However, pci_pm_default_resume_early() used in the system resume
> > code path calls pci_power_up() which doesn't apply the delay at
> > all and that causes issues to occur during resume from
> > suspend-to-idle on some systems where the delay is required.
> >
> > Since there is no reason for that difference to exist, modify
> > pci_power_up() to follow pci_set_power_state() more closely and
> > invoke __pci_start_power_transition() from there to call the
> > platform firmware to power up the device (in case that's necessary).
> >
> > Fixes: db288c9c5f9d ("PCI / PM: restore the original behavior of pci_set_power_state()")
> > Reported-by: Daniel Drake <[email protected]>
> > Link: https://lore.kernel.org/linux-pm/CAD8Lp44TYxrMgPLkHCqF9hv6smEurMXvmmvmtyFhZ6Q4SE+dig@mail.gmail.com/T/#m21be74af263c6a34f36e0fc5c77c5449d9406925
> > Signed-off-by: Rafael J. Wysocki <[email protected]>
> > ---
> >
> > Daniel, please test this one.
> >
> > ---
> > drivers/pci/pci.c | 24 +++++++++++-------------
> > 1 file changed, 11 insertions(+), 13 deletions(-)
> >
> > Index: linux-pm/drivers/pci/pci.c
> > ===================================================================
> > --- linux-pm.orig/drivers/pci/pci.c
> > +++ linux-pm/drivers/pci/pci.c
> > @@ -959,19 +959,6 @@ void pci_refresh_power_state(struct pci_
> > }
> >
> > /**
> > - * pci_power_up - Put the given device into D0 forcibly
> > - * @dev: PCI device to power up
> > - */
> > -void pci_power_up(struct pci_dev *dev)
> > -{
> > - if (platform_pci_power_manageable(dev))
> > - platform_pci_set_power_state(dev, PCI_D0);
> > -
> > - pci_raw_set_power_state(dev, PCI_D0);
> > - pci_update_current_state(dev, PCI_D0);
> > -}
> > -
> > -/**
> > * pci_platform_power_transition - Use platform to change device power state
> > * @dev: PCI device to handle.
> > * @state: State to put the device into.
> > @@ -1154,6 +1141,17 @@ int pci_set_power_state(struct pci_dev *
> > EXPORT_SYMBOL(pci_set_power_state);
> >
> > /**
> > + * pci_power_up - Put the given device into D0 forcibly
>
> Not specifically for this patch, but what does "forcibly" mean?
>
> > + * @dev: PCI device to power up
> > + */
> > +void pci_power_up(struct pci_dev *dev)
> > +{
> > + __pci_start_power_transition(dev, PCI_D0);
> > + pci_raw_set_power_state(dev, PCI_D0);
> > + pci_update_current_state(dev, PCI_D0);
>
> There's not very much difference between:
>
> pci_power_up(dev);
>
> and
>
> pci_set_power_state(dev, PCI_D0);
>
> It looks like the main difference is that pci_set_power_state() calls
> __pci_complete_power_transition(), which ultimately calls
> acpi_pci_set_power_state() (for ACPI systems).

Yes, it does, for power states deeper than D0, which is not the case here.

The main difference is the dev->current_state == state check in
pci_set_power_state(), but in the resume case specifically
dev->current_state == PCI_D0 doesn't matter, because the real power
state of the device may be different.

> So maybe "forcibly" means something like "ignoring any platform power
> management methods"?

It means "go into D0 no matter what the current cached value is".

> It's not obvious to me when we should skip the
> platform stuff or whether the skipping should be done at the high
> level (like calling either pci_power_up() or pci_set_power_state()) or
> at a lower level (e.g., if everybody called pci_set_power_state() and
> it could internally tell whether we're skipping the platform part).

For transitions into D0 __pci_start_power_transition() is the platform
stuff, so we don't skip it and the other things that are present in
pci_set_power_state() and are not there in pci_power_up() are simply
unnecessary for transitions to D0.

> If we could unify the paths as much as possible, that would be nice,
> but if it's not feasible, it's not feasible.

It kind of is, but I'd prefer to do it on top of this patch.

First, the pci_update_current_state() in pci_power_up() can be moved
to pci_pm_default_resume_early() which is the only caller of
pci_power_up(). [The role of that pci_update_current_state() is to
change the current_state value to D3cold if the device is not
accessible (or the platform firmware says that it is D3cold, which may
be the case after a failing attempt to use it to switch the device
over to D0).] Next, if pci_power_up() is modified to return the
return value of pci_raw_set_power_state(), pci_set_power_state() can
be implemented (roughly) as

sanitize the state argument

if (dev->current_state == state)
return 0;

if (state == PCI_D0)
return pci_power_up();

carry out a transition into a deeper power state.

And so pci_power_up() will be used by pci_set_power_state(), for
transitions into D0, and (directly) by pci_pm_default_resume_early().

How does that sound?

> If you'd like me to push this for v5.4, let me know, otherwise you
> can apply my:
>
> Acked-by: Bjorn Helgaas <[email protected]>

I will, thanks!