linux-pci.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
@ 2020-10-29  9:24 Mika Westerberg
  2020-11-13  6:37 ` Mika Westerberg
                   ` (2 more replies)
  0 siblings, 3 replies; 7+ messages in thread
From: Mika Westerberg @ 2020-10-29  9:24 UTC (permalink / raw)
  To: Bjorn Helgaas, Rafael J. Wysocki
  Cc: Lukas Wunner, David Airlie, Daniel Vetter, Utkarsh Patel,
	Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann, linux-pci,
	Mika Westerberg

When a PCI bridge is runtime resumed from D3cold the underlying bus is
walked and the attached devices are runtime resumed as well. However, in
addition to that we also generate a wakeup event for these devices even
though this actually is not a real wakeup event coming from the
hardware.

Normally this does not cause problems but when combined with
/sys/power/wakeup_count like using the steps below:

  # count=$(cat /sys/power/wakeup_count)
  # echo $count > /sys/power/wakeup_count
  # echo mem > /sys/power/state

The system suspend cycle might get aborted at this point if a PCI bridge
that was runtime suspended (D3cold) was runtime resumed for any reason.
The runtime resume calls pci_wakeup_bus() and that generates wakeup
event increasing wakeup_count.

Since this is not a real wakeup event we can prevent the above from
happening by removing the call to pci_wakeup_event() in
pci_wakeup_bus(). While there rename pci_wakeup_bus() to
pci_resume_bus() to better reflect what it does.

Reported-by: Utkarsh Patel <utkarsh.h.patel@intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
---
 drivers/gpu/vga/vga_switcheroo.c |  2 +-
 drivers/pci/pci.c                | 16 +++++-----------
 include/linux/pci.h              |  2 +-
 3 files changed, 7 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/vga/vga_switcheroo.c b/drivers/gpu/vga/vga_switcheroo.c
index 087304b1a5d7..8843b078ad4e 100644
--- a/drivers/gpu/vga/vga_switcheroo.c
+++ b/drivers/gpu/vga/vga_switcheroo.c
@@ -1039,7 +1039,7 @@ static int vga_switcheroo_runtime_resume(struct device *dev)
 	mutex_lock(&vgasr_mutex);
 	vga_switcheroo_power_switch(pdev, VGA_SWITCHEROO_ON);
 	mutex_unlock(&vgasr_mutex);
-	pci_wakeup_bus(pdev->bus);
+	pci_resume_bus(pdev->bus);
 	ret = dev->bus->pm->runtime_resume(dev);
 	if (ret)
 		return ret;
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 6d4d5a2f923d..b25dfa63eeb9 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -1174,26 +1174,20 @@ int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
 }
 EXPORT_SYMBOL_GPL(pci_platform_power_transition);
 
-/**
- * pci_wakeup - Wake up a PCI device
- * @pci_dev: Device to handle.
- * @ign: ignored parameter
- */
-static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
+static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
 {
-	pci_wakeup_event(pci_dev);
 	pm_request_resume(&pci_dev->dev);
 	return 0;
 }
 
 /**
- * pci_wakeup_bus - Walk given bus and wake up devices on it
+ * pci_resume_bus - Walk given bus and runtime resume devices on it
  * @bus: Top bus of the subtree to walk.
  */
-void pci_wakeup_bus(struct pci_bus *bus)
+void pci_resume_bus(struct pci_bus *bus)
 {
 	if (bus)
-		pci_walk_bus(bus, pci_wakeup, NULL);
+		pci_walk_bus(bus, pci_resume_one, NULL);
 }
 
 static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
@@ -1256,7 +1250,7 @@ int pci_power_up(struct pci_dev *dev)
 		 * may be powered on into D0uninitialized state, resume them to
 		 * give them a chance to suspend again
 		 */
-		pci_wakeup_bus(dev->subordinate);
+		pci_resume_bus(dev->subordinate);
 	}
 
 	return pci_raw_set_power_state(dev, PCI_D0);
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 22207a79762c..9256ef2e4327 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -1271,7 +1271,7 @@ bool pci_dev_run_wake(struct pci_dev *dev);
 void pci_d3cold_enable(struct pci_dev *dev);
 void pci_d3cold_disable(struct pci_dev *dev);
 bool pcie_relaxed_ordering_enabled(struct pci_dev *dev);
-void pci_wakeup_bus(struct pci_bus *bus);
+void pci_resume_bus(struct pci_bus *bus);
 void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state);
 
 /* For use by arch with custom probe code */
-- 
2.28.0


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
  2020-10-29  9:24 [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus Mika Westerberg
@ 2020-11-13  6:37 ` Mika Westerberg
  2020-11-16 16:06   ` Rafael J. Wysocki
  2020-11-16 16:05 ` Rafael J. Wysocki
  2020-11-18 21:22 ` Bjorn Helgaas
  2 siblings, 1 reply; 7+ messages in thread
From: Mika Westerberg @ 2020-11-13  6:37 UTC (permalink / raw)
  To: Bjorn Helgaas, Rafael J. Wysocki
  Cc: Lukas Wunner, David Airlie, Daniel Vetter, Utkarsh Patel,
	Maarten Lankhorst, Maxime Ripard, Thomas Zimmermann, linux-pci

Hi,

On Thu, Oct 29, 2020 at 12:24:53PM +0300, Mika Westerberg wrote:
> When a PCI bridge is runtime resumed from D3cold the underlying bus is
> walked and the attached devices are runtime resumed as well. However, in
> addition to that we also generate a wakeup event for these devices even
> though this actually is not a real wakeup event coming from the
> hardware.
> 
> Normally this does not cause problems but when combined with
> /sys/power/wakeup_count like using the steps below:
> 
>   # count=$(cat /sys/power/wakeup_count)
>   # echo $count > /sys/power/wakeup_count
>   # echo mem > /sys/power/state
> 
> The system suspend cycle might get aborted at this point if a PCI bridge
> that was runtime suspended (D3cold) was runtime resumed for any reason.
> The runtime resume calls pci_wakeup_bus() and that generates wakeup
> event increasing wakeup_count.
> 
> Since this is not a real wakeup event we can prevent the above from
> happening by removing the call to pci_wakeup_event() in
> pci_wakeup_bus(). While there rename pci_wakeup_bus() to
> pci_resume_bus() to better reflect what it does.

Any comments on this?

Thanks!

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
  2020-10-29  9:24 [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus Mika Westerberg
  2020-11-13  6:37 ` Mika Westerberg
@ 2020-11-16 16:05 ` Rafael J. Wysocki
  2020-11-18 21:22 ` Bjorn Helgaas
  2 siblings, 0 replies; 7+ messages in thread
From: Rafael J. Wysocki @ 2020-11-16 16:05 UTC (permalink / raw)
  To: Mika Westerberg
  Cc: Bjorn Helgaas, Lukas Wunner, David Airlie, Daniel Vetter,
	Utkarsh Patel, Maarten Lankhorst, Maxime Ripard,
	Thomas Zimmermann, linux-pci

On Thursday, October 29, 2020 10:24:53 AM CET Mika Westerberg wrote:
> When a PCI bridge is runtime resumed from D3cold the underlying bus is
> walked and the attached devices are runtime resumed as well. However, in
> addition to that we also generate a wakeup event for these devices even
> though this actually is not a real wakeup event coming from the
> hardware.
> 
> Normally this does not cause problems but when combined with
> /sys/power/wakeup_count like using the steps below:
> 
>   # count=$(cat /sys/power/wakeup_count)
>   # echo $count > /sys/power/wakeup_count
>   # echo mem > /sys/power/state
> 
> The system suspend cycle might get aborted at this point if a PCI bridge
> that was runtime suspended (D3cold) was runtime resumed for any reason.
> The runtime resume calls pci_wakeup_bus() and that generates wakeup
> event increasing wakeup_count.
> 
> Since this is not a real wakeup event we can prevent the above from
> happening by removing the call to pci_wakeup_event() in
> pci_wakeup_bus(). While there rename pci_wakeup_bus() to
> pci_resume_bus() to better reflect what it does.
> 
> Reported-by: Utkarsh Patel <utkarsh.h.patel@intel.com>
> Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>

Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

> ---
>  drivers/gpu/vga/vga_switcheroo.c |  2 +-
>  drivers/pci/pci.c                | 16 +++++-----------
>  include/linux/pci.h              |  2 +-
>  3 files changed, 7 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/gpu/vga/vga_switcheroo.c b/drivers/gpu/vga/vga_switcheroo.c
> index 087304b1a5d7..8843b078ad4e 100644
> --- a/drivers/gpu/vga/vga_switcheroo.c
> +++ b/drivers/gpu/vga/vga_switcheroo.c
> @@ -1039,7 +1039,7 @@ static int vga_switcheroo_runtime_resume(struct device *dev)
>  	mutex_lock(&vgasr_mutex);
>  	vga_switcheroo_power_switch(pdev, VGA_SWITCHEROO_ON);
>  	mutex_unlock(&vgasr_mutex);
> -	pci_wakeup_bus(pdev->bus);
> +	pci_resume_bus(pdev->bus);
>  	ret = dev->bus->pm->runtime_resume(dev);
>  	if (ret)
>  		return ret;
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 6d4d5a2f923d..b25dfa63eeb9 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -1174,26 +1174,20 @@ int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
>  }
>  EXPORT_SYMBOL_GPL(pci_platform_power_transition);
>  
> -/**
> - * pci_wakeup - Wake up a PCI device
> - * @pci_dev: Device to handle.
> - * @ign: ignored parameter
> - */
> -static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
> +static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
>  {
> -	pci_wakeup_event(pci_dev);
>  	pm_request_resume(&pci_dev->dev);
>  	return 0;
>  }
>  
>  /**
> - * pci_wakeup_bus - Walk given bus and wake up devices on it
> + * pci_resume_bus - Walk given bus and runtime resume devices on it
>   * @bus: Top bus of the subtree to walk.
>   */
> -void pci_wakeup_bus(struct pci_bus *bus)
> +void pci_resume_bus(struct pci_bus *bus)
>  {
>  	if (bus)
> -		pci_walk_bus(bus, pci_wakeup, NULL);
> +		pci_walk_bus(bus, pci_resume_one, NULL);
>  }
>  
>  static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
> @@ -1256,7 +1250,7 @@ int pci_power_up(struct pci_dev *dev)
>  		 * may be powered on into D0uninitialized state, resume them to
>  		 * give them a chance to suspend again
>  		 */
> -		pci_wakeup_bus(dev->subordinate);
> +		pci_resume_bus(dev->subordinate);
>  	}
>  
>  	return pci_raw_set_power_state(dev, PCI_D0);
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 22207a79762c..9256ef2e4327 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -1271,7 +1271,7 @@ bool pci_dev_run_wake(struct pci_dev *dev);
>  void pci_d3cold_enable(struct pci_dev *dev);
>  void pci_d3cold_disable(struct pci_dev *dev);
>  bool pcie_relaxed_ordering_enabled(struct pci_dev *dev);
> -void pci_wakeup_bus(struct pci_bus *bus);
> +void pci_resume_bus(struct pci_bus *bus);
>  void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state);
>  
>  /* For use by arch with custom probe code */
> 





^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
  2020-11-13  6:37 ` Mika Westerberg
@ 2020-11-16 16:06   ` Rafael J. Wysocki
  0 siblings, 0 replies; 7+ messages in thread
From: Rafael J. Wysocki @ 2020-11-16 16:06 UTC (permalink / raw)
  To: Mika Westerberg
  Cc: Bjorn Helgaas, Lukas Wunner, David Airlie, Daniel Vetter,
	Utkarsh Patel, Maarten Lankhorst, Maxime Ripard,
	Thomas Zimmermann, linux-pci

On Friday, November 13, 2020 7:37:45 AM CET Mika Westerberg wrote:
> Hi,
> 
> On Thu, Oct 29, 2020 at 12:24:53PM +0300, Mika Westerberg wrote:
> > When a PCI bridge is runtime resumed from D3cold the underlying bus is
> > walked and the attached devices are runtime resumed as well. However, in
> > addition to that we also generate a wakeup event for these devices even
> > though this actually is not a real wakeup event coming from the
> > hardware.
> > 
> > Normally this does not cause problems but when combined with
> > /sys/power/wakeup_count like using the steps below:
> > 
> >   # count=$(cat /sys/power/wakeup_count)
> >   # echo $count > /sys/power/wakeup_count
> >   # echo mem > /sys/power/state
> > 
> > The system suspend cycle might get aborted at this point if a PCI bridge
> > that was runtime suspended (D3cold) was runtime resumed for any reason.
> > The runtime resume calls pci_wakeup_bus() and that generates wakeup
> > event increasing wakeup_count.
> > 
> > Since this is not a real wakeup event we can prevent the above from
> > happening by removing the call to pci_wakeup_event() in
> > pci_wakeup_bus(). While there rename pci_wakeup_bus() to
> > pci_resume_bus() to better reflect what it does.
> 
> Any comments on this?

I've missed it, sorry.

Now replied with a R-by, thanks!




^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
  2020-10-29  9:24 [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus Mika Westerberg
  2020-11-13  6:37 ` Mika Westerberg
  2020-11-16 16:05 ` Rafael J. Wysocki
@ 2020-11-18 21:22 ` Bjorn Helgaas
  2020-11-19  7:55   ` Mika Westerberg
  2 siblings, 1 reply; 7+ messages in thread
From: Bjorn Helgaas @ 2020-11-18 21:22 UTC (permalink / raw)
  To: Mika Westerberg
  Cc: Bjorn Helgaas, Rafael J. Wysocki, Lukas Wunner, David Airlie,
	Daniel Vetter, Utkarsh Patel, Maarten Lankhorst, Maxime Ripard,
	Thomas Zimmermann, linux-pci

On Thu, Oct 29, 2020 at 12:24:53PM +0300, Mika Westerberg wrote:
> When a PCI bridge is runtime resumed from D3cold the underlying bus is
> walked and the attached devices are runtime resumed as well. However, in
> addition to that we also generate a wakeup event for these devices even
> though this actually is not a real wakeup event coming from the
> hardware.
> 
> Normally this does not cause problems but when combined with
> /sys/power/wakeup_count like using the steps below:
> 
>   # count=$(cat /sys/power/wakeup_count)
>   # echo $count > /sys/power/wakeup_count
>   # echo mem > /sys/power/state
> 
> The system suspend cycle might get aborted at this point if a PCI bridge
> that was runtime suspended (D3cold) was runtime resumed for any reason.
> The runtime resume calls pci_wakeup_bus() and that generates wakeup
> event increasing wakeup_count.
> 
> Since this is not a real wakeup event we can prevent the above from
> happening by removing the call to pci_wakeup_event() in
> pci_wakeup_bus(). While there rename pci_wakeup_bus() to
> pci_resume_bus() to better reflect what it does.
> 
> Reported-by: Utkarsh Patel <utkarsh.h.patel@intel.com>

Is there a URL to a report on a mailing list or a bugzilla that we can
include here?  What was the actual user-visible issue?  If we can
mention it here, it may be useful to others who encounter the same
issue.  I guess maybe a system suspend fails?

> Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
> ---
>  drivers/gpu/vga/vga_switcheroo.c |  2 +-
>  drivers/pci/pci.c                | 16 +++++-----------
>  include/linux/pci.h              |  2 +-
>  3 files changed, 7 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/gpu/vga/vga_switcheroo.c b/drivers/gpu/vga/vga_switcheroo.c
> index 087304b1a5d7..8843b078ad4e 100644
> --- a/drivers/gpu/vga/vga_switcheroo.c
> +++ b/drivers/gpu/vga/vga_switcheroo.c
> @@ -1039,7 +1039,7 @@ static int vga_switcheroo_runtime_resume(struct device *dev)
>  	mutex_lock(&vgasr_mutex);
>  	vga_switcheroo_power_switch(pdev, VGA_SWITCHEROO_ON);
>  	mutex_unlock(&vgasr_mutex);
> -	pci_wakeup_bus(pdev->bus);
> +	pci_resume_bus(pdev->bus);
>  	ret = dev->bus->pm->runtime_resume(dev);
>  	if (ret)
>  		return ret;
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 6d4d5a2f923d..b25dfa63eeb9 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -1174,26 +1174,20 @@ int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
>  }
>  EXPORT_SYMBOL_GPL(pci_platform_power_transition);
>  
> -/**
> - * pci_wakeup - Wake up a PCI device
> - * @pci_dev: Device to handle.
> - * @ign: ignored parameter
> - */
> -static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
> +static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
>  {
> -	pci_wakeup_event(pci_dev);

IIUC this is the critical change, and all the rest of this patch is
no-op renames.  Can you split this into two patches so the important
change is more obvious?

Then the obvious questions will be why it is safe to remove this, and
where the desired call for the real wakeup is.

>  	pm_request_resume(&pci_dev->dev);
>  	return 0;
>  }
>  
>  /**
> - * pci_wakeup_bus - Walk given bus and wake up devices on it
> + * pci_resume_bus - Walk given bus and runtime resume devices on it
>   * @bus: Top bus of the subtree to walk.
>   */
> -void pci_wakeup_bus(struct pci_bus *bus)
> +void pci_resume_bus(struct pci_bus *bus)
>  {
>  	if (bus)
> -		pci_walk_bus(bus, pci_wakeup, NULL);
> +		pci_walk_bus(bus, pci_resume_one, NULL);
>  }
>  
>  static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
> @@ -1256,7 +1250,7 @@ int pci_power_up(struct pci_dev *dev)
>  		 * may be powered on into D0uninitialized state, resume them to
>  		 * give them a chance to suspend again
>  		 */
> -		pci_wakeup_bus(dev->subordinate);
> +		pci_resume_bus(dev->subordinate);
>  	}
>  
>  	return pci_raw_set_power_state(dev, PCI_D0);
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 22207a79762c..9256ef2e4327 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -1271,7 +1271,7 @@ bool pci_dev_run_wake(struct pci_dev *dev);
>  void pci_d3cold_enable(struct pci_dev *dev);
>  void pci_d3cold_disable(struct pci_dev *dev);
>  bool pcie_relaxed_ordering_enabled(struct pci_dev *dev);
> -void pci_wakeup_bus(struct pci_bus *bus);
> +void pci_resume_bus(struct pci_bus *bus);
>  void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state);
>  
>  /* For use by arch with custom probe code */
> -- 
> 2.28.0
> 

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
  2020-11-18 21:22 ` Bjorn Helgaas
@ 2020-11-19  7:55   ` Mika Westerberg
  2020-11-20  2:27     ` Patel, Utkarsh H
  0 siblings, 1 reply; 7+ messages in thread
From: Mika Westerberg @ 2020-11-19  7:55 UTC (permalink / raw)
  To: Bjorn Helgaas
  Cc: Bjorn Helgaas, Rafael J. Wysocki, Lukas Wunner, David Airlie,
	Daniel Vetter, Utkarsh Patel, Maarten Lankhorst, Maxime Ripard,
	Thomas Zimmermann, linux-pci

Hi Bjorn,

On Wed, Nov 18, 2020 at 03:22:00PM -0600, Bjorn Helgaas wrote:
> On Thu, Oct 29, 2020 at 12:24:53PM +0300, Mika Westerberg wrote:
> > When a PCI bridge is runtime resumed from D3cold the underlying bus is
> > walked and the attached devices are runtime resumed as well. However, in
> > addition to that we also generate a wakeup event for these devices even
> > though this actually is not a real wakeup event coming from the
> > hardware.
> > 
> > Normally this does not cause problems but when combined with
> > /sys/power/wakeup_count like using the steps below:
> > 
> >   # count=$(cat /sys/power/wakeup_count)
> >   # echo $count > /sys/power/wakeup_count
> >   # echo mem > /sys/power/state
> > 
> > The system suspend cycle might get aborted at this point if a PCI bridge
> > that was runtime suspended (D3cold) was runtime resumed for any reason.
> > The runtime resume calls pci_wakeup_bus() and that generates wakeup
> > event increasing wakeup_count.
> > 
> > Since this is not a real wakeup event we can prevent the above from
> > happening by removing the call to pci_wakeup_event() in
> > pci_wakeup_bus(). While there rename pci_wakeup_bus() to
> > pci_resume_bus() to better reflect what it does.
> > 
> > Reported-by: Utkarsh Patel <utkarsh.h.patel@intel.com>
> 
> Is there a URL to a report on a mailing list or a bugzilla that we can
> include here?  What was the actual user-visible issue?  If we can
> mention it here, it may be useful to others who encounter the same
> issue.  I guess maybe a system suspend fails?

I'm not sure if there is bugzilla entry about this. There might be a
Google partner bug but not sure if it is public.

@Utkarsh, if there is one can you share that link with Bjorn?

There are two user visible issues, first is that if you do the above
steps manually the suspend gets aborted (as the above commit log tries
to explain).

The second "user visible" issue is that the ChromeOS suspend stress test
script below fails (as it does the same steps):

  https://chromium.googlesource.com/chromiumos/platform/power_manager/+/refs/heads/master/tools/suspend_stress_test

> > Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
> > ---
> >  drivers/gpu/vga/vga_switcheroo.c |  2 +-
> >  drivers/pci/pci.c                | 16 +++++-----------
> >  include/linux/pci.h              |  2 +-
> >  3 files changed, 7 insertions(+), 13 deletions(-)
> > 
> > diff --git a/drivers/gpu/vga/vga_switcheroo.c b/drivers/gpu/vga/vga_switcheroo.c
> > index 087304b1a5d7..8843b078ad4e 100644
> > --- a/drivers/gpu/vga/vga_switcheroo.c
> > +++ b/drivers/gpu/vga/vga_switcheroo.c
> > @@ -1039,7 +1039,7 @@ static int vga_switcheroo_runtime_resume(struct device *dev)
> >  	mutex_lock(&vgasr_mutex);
> >  	vga_switcheroo_power_switch(pdev, VGA_SWITCHEROO_ON);
> >  	mutex_unlock(&vgasr_mutex);
> > -	pci_wakeup_bus(pdev->bus);
> > +	pci_resume_bus(pdev->bus);
> >  	ret = dev->bus->pm->runtime_resume(dev);
> >  	if (ret)
> >  		return ret;
> > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > index 6d4d5a2f923d..b25dfa63eeb9 100644
> > --- a/drivers/pci/pci.c
> > +++ b/drivers/pci/pci.c
> > @@ -1174,26 +1174,20 @@ int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
> >  }
> >  EXPORT_SYMBOL_GPL(pci_platform_power_transition);
> >  
> > -/**
> > - * pci_wakeup - Wake up a PCI device
> > - * @pci_dev: Device to handle.
> > - * @ign: ignored parameter
> > - */
> > -static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
> > +static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
> >  {
> > -	pci_wakeup_event(pci_dev);
> 
> IIUC this is the critical change, and all the rest of this patch is
> no-op renames.  Can you split this into two patches so the important
> change is more obvious?

Sure.

> Then the obvious questions will be why it is safe to remove this, and
> where the desired call for the real wakeup is.

This is only called on runtime resume path to turn on devices below this
one. However, wakeup is only relevant on system sleep path.

For ACPI backed devices the real wakeup is handled in the
pci_acpi_wake_dev() and in case of PME it is pcie_pme_handle_request().
And then there is the pme_poll thread as well.

^ permalink raw reply	[flat|nested] 7+ messages in thread

* RE: [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus
  2020-11-19  7:55   ` Mika Westerberg
@ 2020-11-20  2:27     ` Patel, Utkarsh H
  0 siblings, 0 replies; 7+ messages in thread
From: Patel, Utkarsh H @ 2020-11-20  2:27 UTC (permalink / raw)
  To: Mika Westerberg, Bjorn Helgaas
  Cc: Bjorn Helgaas, Rafael J. Wysocki, Lukas Wunner, David Airlie,
	Daniel Vetter, Maarten Lankhorst, Maxime Ripard,
	Thomas Zimmermann, linux-pci

Hi Mika and Bjorn,

> -----Original Message-----
> From: Mika Westerberg <mika.westerberg@linux.intel.com>
> Sent: Wednesday, November 18, 2020 11:56 PM
> To: Bjorn Helgaas <helgaas@kernel.org>
> Cc: Bjorn Helgaas <bhelgaas@google.com>; Rafael J. Wysocki
> <rjw@rjwysocki.net>; Lukas Wunner <lukas@wunner.de>; David Airlie
> <airlied@linux.ie>; Daniel Vetter <daniel@ffwll.ch>; Patel, Utkarsh H
> <utkarsh.h.patel@intel.com>; Maarten Lankhorst
> <maarten.lankhorst@linux.intel.com>; Maxime Ripard <mripard@kernel.org>;
> Thomas Zimmermann <tzimmermann@suse.de>; linux-pci@vger.kernel.org
> Subject: Re: [PATCH] PCI/PM: Do not generate wakeup event when runtime
> resuming bus
> 
> Hi Bjorn,
> 
> On Wed, Nov 18, 2020 at 03:22:00PM -0600, Bjorn Helgaas wrote:
> > On Thu, Oct 29, 2020 at 12:24:53PM +0300, Mika Westerberg wrote:
> > > When a PCI bridge is runtime resumed from D3cold the underlying bus
> > > is walked and the attached devices are runtime resumed as well.
> > > However, in addition to that we also generate a wakeup event for
> > > these devices even though this actually is not a real wakeup event
> > > coming from the hardware.
> > >
> > > Normally this does not cause problems but when combined with
> > > /sys/power/wakeup_count like using the steps below:
> > >
> > >   # count=$(cat /sys/power/wakeup_count)
> > >   # echo $count > /sys/power/wakeup_count
> > >   # echo mem > /sys/power/state
> > >
> > > The system suspend cycle might get aborted at this point if a PCI
> > > bridge that was runtime suspended (D3cold) was runtime resumed for any
> reason.
> > > The runtime resume calls pci_wakeup_bus() and that generates wakeup
> > > event increasing wakeup_count.
> > >
> > > Since this is not a real wakeup event we can prevent the above from
> > > happening by removing the call to pci_wakeup_event() in
> > > pci_wakeup_bus(). While there rename pci_wakeup_bus() to
> > > pci_resume_bus() to better reflect what it does.
> > >
> > > Reported-by: Utkarsh Patel <utkarsh.h.patel@intel.com>
> >
> > Is there a URL to a report on a mailing list or a bugzilla that we can
> > include here?  What was the actual user-visible issue?  If we can
> > mention it here, it may be useful to others who encounter the same
> > issue.  I guess maybe a system suspend fails?
> 
> I'm not sure if there is bugzilla entry about this. There might be a Google
> partner bug but not sure if it is public.
> 
> @Utkarsh, if there is one can you share that link with Bjorn?

This is reported only Google partner bug which is private and I am not sure if I can share the link here. 

> 
> There are two user visible issues, first is that if you do the above steps
> manually the suspend gets aborted (as the above commit log tries to explain).
> 
> The second "user visible" issue is that the ChromeOS suspend stress test
> script below fails (as it does the same steps):
> 
> 
> https://chromium.googlesource.com/chromiumos/platform/power_manager/
> +/refs/heads/master/tools/suspend_stress_test
> 
> > > Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
> > > ---
> > >  drivers/gpu/vga/vga_switcheroo.c |  2 +-
> > >  drivers/pci/pci.c                | 16 +++++-----------
> > >  include/linux/pci.h              |  2 +-
> > >  3 files changed, 7 insertions(+), 13 deletions(-)
> > >
> > > diff --git a/drivers/gpu/vga/vga_switcheroo.c
> > > b/drivers/gpu/vga/vga_switcheroo.c
> > > index 087304b1a5d7..8843b078ad4e 100644
> > > --- a/drivers/gpu/vga/vga_switcheroo.c
> > > +++ b/drivers/gpu/vga/vga_switcheroo.c
> > > @@ -1039,7 +1039,7 @@ static int
> vga_switcheroo_runtime_resume(struct device *dev)
> > >  	mutex_lock(&vgasr_mutex);
> > >  	vga_switcheroo_power_switch(pdev, VGA_SWITCHEROO_ON);
> > >  	mutex_unlock(&vgasr_mutex);
> > > -	pci_wakeup_bus(pdev->bus);
> > > +	pci_resume_bus(pdev->bus);
> > >  	ret = dev->bus->pm->runtime_resume(dev);
> > >  	if (ret)
> > >  		return ret;
> > > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c index
> > > 6d4d5a2f923d..b25dfa63eeb9 100644
> > > --- a/drivers/pci/pci.c
> > > +++ b/drivers/pci/pci.c
> > > @@ -1174,26 +1174,20 @@ int pci_platform_power_transition(struct
> > > pci_dev *dev, pci_power_t state)  }
> > > EXPORT_SYMBOL_GPL(pci_platform_power_transition);
> > >
> > > -/**
> > > - * pci_wakeup - Wake up a PCI device
> > > - * @pci_dev: Device to handle.
> > > - * @ign: ignored parameter
> > > - */
> > > -static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
> > > +static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
> > >  {
> > > -	pci_wakeup_event(pci_dev);
> >
> > IIUC this is the critical change, and all the rest of this patch is
> > no-op renames.  Can you split this into two patches so the important
> > change is more obvious?
> 
> Sure.
> 
> > Then the obvious questions will be why it is safe to remove this, and
> > where the desired call for the real wakeup is.
> 
> This is only called on runtime resume path to turn on devices below this one.
> However, wakeup is only relevant on system sleep path.
> 
> For ACPI backed devices the real wakeup is handled in the
> pci_acpi_wake_dev() and in case of PME it is pcie_pme_handle_request().
> And then there is the pme_poll thread as well.

Sincerely,
Utkarsh Patel. 

^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2020-11-20  2:27 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-10-29  9:24 [PATCH] PCI/PM: Do not generate wakeup event when runtime resuming bus Mika Westerberg
2020-11-13  6:37 ` Mika Westerberg
2020-11-16 16:06   ` Rafael J. Wysocki
2020-11-16 16:05 ` Rafael J. Wysocki
2020-11-18 21:22 ` Bjorn Helgaas
2020-11-19  7:55   ` Mika Westerberg
2020-11-20  2:27     ` Patel, Utkarsh H

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).