Patchwork [-v5,3/3] PCIe, PM, Add PCIe runtime D3cold support

login
register
mail settings
Submitter Huang Ying
Date June 11, 2012, 2:47 a.m.
Message ID <1339382833-7928-4-git-send-email-ying.huang@intel.com>
Download mbox | patch
Permalink /patch/164053/
State Superseded
Headers show

Comments

Huang Ying - June 11, 2012, 2:47 a.m.
This patch adds runtime D3cold support to PCIe bus.  D3cold is the
deepest power saving state for PCIe devices.  Where the main PCIe link
will be powered off totally, so before the PCIe main link is powered
on again, you can not access the device, even the configuration space,
which is still accessible in D3hot.  Because the main PCIe link is not
available, the PCI PM registers can not be used to put device into/out
of D3cold state, that will be done by platform logic such as ACPI
_PR3.

To support remote wake up in D3cold, aux power is supplied, and a
side-band pin: WAKE# is used to notify remote wake up event, the pin
is usually connected to platform logic such as ACPI GPE.  This is
quite different with other power saving states, where remote wake up
is notified via PME message through the PCIe main link.

PCIe devices in plug-in slot usually has no direct platform logic, for
example, there is usually no ACPI _PR3 for them.  The D3cold support
for these devices can be done via the PCIe port connected to it.  When
power on/off the PCIe port, the corresponding PCIe devices are powered
on/off too.  And the remote wake up event from PCIe device will be
notified to the corresponding PCIe port.

The PCI subsystem D3cold support and corresponding ACPI platform
support is implemented in the patch.  Including the support for PCIe
devices in plug-in slot.

For more information about PCIe D3cold and corresponding ACPI support,
please refer to:

- PCI Express Base Specification Revision 2.0
- Advanced Configuration and Power Interface Specification Revision 5.0

Originally-by: Zheng Yan <zheng.z.yan@intel.com>
Signed-off-by: Huang Ying <ying.huang@intel.com>
---
 drivers/pci/pci-acpi.c         |   23 ++++++--
 drivers/pci/pci-driver.c       |   10 +++
 drivers/pci/pci-sysfs.c        |   29 ++++++++++
 drivers/pci/pci.c              |  114 +++++++++++++++++++++++++++++++++++++----
 drivers/pci/pci.h              |    1 
 drivers/pci/pcie/portdrv_pci.c |   44 ++++++++++++++-
 include/linux/pci.h            |   16 ++++-
 7 files changed, 215 insertions(+), 22 deletions(-)

--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Rafael J. Wysocki - June 15, 2012, 9:21 p.m.
On Monday, June 11, 2012, Huang Ying wrote:
> This patch adds runtime D3cold support to PCIe bus.  D3cold is the
> deepest power saving state for PCIe devices.  Where the main PCIe link
> will be powered off totally, so before the PCIe main link is powered
> on again, you can not access the device, even the configuration space,
> which is still accessible in D3hot.  Because the main PCIe link is not
> available, the PCI PM registers can not be used to put device into/out
> of D3cold state, that will be done by platform logic such as ACPI
> _PR3.
> 
> To support remote wake up in D3cold, aux power is supplied, and a
> side-band pin: WAKE# is used to notify remote wake up event, the pin
> is usually connected to platform logic such as ACPI GPE.  This is
> quite different with other power saving states, where remote wake up
> is notified via PME message through the PCIe main link.
> 
> PCIe devices in plug-in slot usually has no direct platform logic, for
> example, there is usually no ACPI _PR3 for them.  The D3cold support
> for these devices can be done via the PCIe port connected to it.  When
> power on/off the PCIe port, the corresponding PCIe devices are powered
> on/off too.  And the remote wake up event from PCIe device will be
> notified to the corresponding PCIe port.
> 
> The PCI subsystem D3cold support and corresponding ACPI platform
> support is implemented in the patch.  Including the support for PCIe
> devices in plug-in slot.
> 
> For more information about PCIe D3cold and corresponding ACPI support,
> please refer to:
> 
> - PCI Express Base Specification Revision 2.0
> - Advanced Configuration and Power Interface Specification Revision 5.0
> 
> Originally-by: Zheng Yan <zheng.z.yan@intel.com>
> Signed-off-by: Huang Ying <ying.huang@intel.com>
> ---
>  drivers/pci/pci-acpi.c         |   23 ++++++--
>  drivers/pci/pci-driver.c       |   10 +++
>  drivers/pci/pci-sysfs.c        |   29 ++++++++++
>  drivers/pci/pci.c              |  114 +++++++++++++++++++++++++++++++++++++----
>  drivers/pci/pci.h              |    1 
>  drivers/pci/pcie/portdrv_pci.c |   44 ++++++++++++++-
>  include/linux/pci.h            |   16 ++++-
>  7 files changed, 215 insertions(+), 22 deletions(-)
> 
> --- a/drivers/pci/pci-acpi.c
> +++ b/drivers/pci/pci-acpi.c
> @@ -48,6 +48,12 @@ static void pci_acpi_wake_dev(acpi_handl
>  	if (event != ACPI_NOTIFY_DEVICE_WAKE || !pci_dev)
>  		return;
>  
> +	if (pci_dev->current_state == PCI_D3cold) {
> +		pci_wakeup_event(pci_dev);
> +		pm_runtime_resume(&pci_dev->dev);
> +		return;
> +	}
> +
>  	if (!pci_dev->pm_cap || !pci_dev->pme_support
>  	     || pci_check_pme_status(pci_dev)) {
>  		if (pci_dev->pme_poll)
> @@ -187,10 +193,13 @@ acpi_status pci_acpi_remove_pm_notifier(
>  
>  static pci_power_t acpi_pci_choose_state(struct pci_dev *pdev)
>  {
> -	int acpi_state;
> +	int acpi_state, d_max;
>  
> -	acpi_state = acpi_pm_device_sleep_state(&pdev->dev, NULL,
> -						ACPI_STATE_D3);
> +	if (pdev->no_d3cold)
> +		d_max = ACPI_STATE_D3_HOT;
> +	else
> +		d_max = ACPI_STATE_D3_COLD;
> +	acpi_state = acpi_pm_device_sleep_state(&pdev->dev, NULL, d_max);
>  	if (acpi_state < 0)
>  		return PCI_POWER_ERROR;
>  
> @@ -297,7 +306,13 @@ static void acpi_pci_propagate_run_wake(
>  
>  static int acpi_pci_run_wake(struct pci_dev *dev, bool enable)
>  {
> -	if (dev->pme_interrupt)
> +	/*
> +	 * Per PCI Express Base Specification Revision 2.0 section
> +	 * 5.3.3.2 Link Wakeup, platform support is needed for D3cold
> +	 * waking up to power on the main link even if there is PME
> +	 * support for D3cold
> +	 */
> +	if (dev->pme_interrupt && !dev->runtime_d3cold)
>  		return 0;
>  
>  	if (!acpi_pm_device_run_wake(&dev->dev, enable))
> --- a/drivers/pci/pci-driver.c
> +++ b/drivers/pci/pci-driver.c
> @@ -1019,10 +1019,13 @@ static int pci_pm_runtime_suspend(struct
>  	if (!pm || !pm->runtime_suspend)
>  		return -ENOSYS;
>  
> +	pci_dev->no_d3cold = false;
>  	error = pm->runtime_suspend(dev);
>  	suspend_report_result(pm->runtime_suspend, error);
>  	if (error)
>  		return error;
> +	if (!pci_dev->d3cold_allowed)
> +		pci_dev->no_d3cold = true;
>  
>  	pci_fixup_device(pci_fixup_suspend, pci_dev);
>  
> @@ -1044,6 +1047,7 @@ static int pci_pm_runtime_suspend(struct
>  
>  static int pci_pm_runtime_resume(struct device *dev)
>  {
> +	int rc;
>  	struct pci_dev *pci_dev = to_pci_dev(dev);
>  	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
>  
> @@ -1054,7 +1058,11 @@ static int pci_pm_runtime_resume(struct
>  	__pci_enable_wake(pci_dev, PCI_D0, true, false);
>  	pci_fixup_device(pci_fixup_resume, pci_dev);
>  
> -	return pm->runtime_resume(dev);
> +	rc = pm->runtime_resume(dev);
> +
> +	pci_dev->runtime_d3cold = false;
> +
> +	return rc;
>  }
>  
>  static int pci_pm_runtime_idle(struct device *dev)
> --- a/drivers/pci/pci-sysfs.c
> +++ b/drivers/pci/pci-sysfs.c
> @@ -28,6 +28,7 @@
>  #include <linux/pci-aspm.h>
>  #include <linux/slab.h>
>  #include <linux/vgaarb.h>
> +#include <linux/pm_runtime.h>
>  #include "pci.h"
>  
>  static int sysfs_initialized;	/* = 0 */
> @@ -378,6 +379,31 @@ dev_bus_rescan_store(struct device *dev,
>  
>  #endif
>  
> +#if defined(CONFIG_PM_RUNTIME) && defined(CONFIG_ACPI)
> +static ssize_t d3cold_allowed_store(struct device *dev,
> +				    struct device_attribute *attr,
> +				    const char *buf, size_t count)
> +{
> +	struct pci_dev *pdev = to_pci_dev(dev);
> +	unsigned long val;
> +
> +	if (strict_strtoul(buf, 0, &val) < 0)
> +		return -EINVAL;
> +
> +	pdev->d3cold_allowed = !!val;
> +	pm_runtime_resume(dev);
> +
> +	return count;
> +}
> +
> +static ssize_t d3cold_allowed_show(struct device *dev,
> +				   struct device_attribute *attr, char *buf)
> +{
> +	struct pci_dev *pdev = to_pci_dev(dev);
> +	return sprintf (buf, "%u\n", pdev->d3cold_allowed);
> +}
> +#endif
> +
>  struct device_attribute pci_dev_attrs[] = {
>  	__ATTR_RO(resource),
>  	__ATTR_RO(vendor),
> @@ -402,6 +428,9 @@ struct device_attribute pci_dev_attrs[]
>  	__ATTR(remove, (S_IWUSR|S_IWGRP), NULL, remove_store),
>  	__ATTR(rescan, (S_IWUSR|S_IWGRP), NULL, dev_rescan_store),
>  #endif
> +#if defined(CONFIG_PM_RUNTIME) && defined(CONFIG_ACPI)
> +	__ATTR(d3cold_allowed, 0644, d3cold_allowed_show, d3cold_allowed_store),
> +#endif
>  	__ATTR_NULL,
>  };
>  
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -622,7 +622,8 @@ static int pci_raw_set_power_state(struc
>  		dev_info(&dev->dev, "Refused to change power state, "
>  			"currently in D%d\n", dev->current_state);
>  
> -	/* According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
> +	/*
> +	 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
>  	 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
>  	 * from D3hot to D0 _may_ perform an internal reset, thereby
>  	 * going to "D0 Uninitialized" rather than "D0 Initialized".
> @@ -654,6 +655,16 @@ void pci_update_current_state(struct pci
>  	if (dev->pm_cap) {
>  		u16 pmcsr;
>  
> +		/*
> +		 * Configuration space is not accessible for device in
> +		 * D3cold, so just keep or set D3cold for safety
> +		 */
> +		if (dev->current_state == PCI_D3cold)
> +			return;
> +		if (state == PCI_D3cold) {
> +			dev->current_state = PCI_D3cold;
> +			return;
> +		}
>  		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
>  		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
>  	} else {
> @@ -694,8 +705,50 @@ static int pci_platform_power_transition
>   */
>  static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
>  {
> -	if (state == PCI_D0)
> +	if (state == PCI_D0) {
>  		pci_platform_power_transition(dev, PCI_D0);
> +		/*
> +		 * Mandatory power management transition delays, see
> +		 * PCI Express Base Specification Revision 2.0 Section
> +		 * 6.6.1: Conventional Reset.  Do not delay for
> +		 * devices powered on/off by corresponding bridge,
> +		 * because have already delayed for the bridge.
> +		 */
> +		if (dev->runtime_d3cold) {
> +			msleep(dev->d3cold_delay);
> +			/*
> +			 * When powering on a bridge from D3cold, the
> +			 * whole hierarchy may be powered on into
> +			 * D0uninitialized state, resume them to give
> +			 * them a chance to suspend again
> +			 */
> +			pci_wakeup_bus(dev->subordinate);
> +		}
> +	}
> +}
> +
> +/**
> + * __pci_dev_set_current_state - Set current state of a PCI device
> + * @dev: Device to handle
> + * @data: pointer to state to be set
> + */
> +static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
> +{
> +	pci_power_t state = *(pci_power_t *)data;
> +
> +	dev->current_state = state;
> +	return 0;
> +}
> +
> +/**
> + * __pci_bus_set_current_state - Walk given bus and set current state of devices
> + * @bus: Top bus of the subtree to walk.
> + * @state: state to be set
> + */
> +static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
> +{
> +	if (bus)
> +		pci_walk_bus(bus, __pci_dev_set_current_state, &state);
>  }
>  
>  /**
> @@ -707,8 +760,15 @@ static void __pci_start_power_transition
>   */
>  int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
>  {
> -	return state >= PCI_D0 ?
> -			pci_platform_power_transition(dev, state) : -EINVAL;
> +	int ret;
> +
> +	if (state < PCI_D0)
> +		return -EINVAL;
> +	ret = pci_platform_power_transition(dev, state);
> +	/* Power off the bridge may power off the whole hierarchy */
> +	if (!ret && state == PCI_D3cold)
> +		__pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
> +	return ret;
>  }
>  EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
>  
> @@ -731,9 +791,13 @@ int pci_set_power_state(struct pci_dev *
>  {
>  	int error;
>  
> -	/* bound the state we're entering */
> -	if (state > PCI_D3hot)
> +	/* D3cold during system suspend/hibernate is not supported */
> +	if (!dev->runtime_d3cold && state > PCI_D3hot)

I don't think this is the right place for the dev->runtime_d3cold check,
I've already told you that and haven't heard anything from you causing me
to change my mind.

If you don't want D3cold to be used during system sleep transitions,
please make sure that D3cold is not passed to pci_set_power_state() in
those cases.

Thanks,
Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
huang ying - June 16, 2012, 3:30 a.m.
Sorry for bothering if you received this message twice.  I have some
issue with my mailing system.

On Sat, Jun 16, 2012 at 5:21 AM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
[snip]
>
> > @@ -731,9 +791,13 @@ int pci_set_power_state(struct pci_dev *
> >  {
> >       int error;
> >
> > -     /* bound the state we're entering */
> > -     if (state > PCI_D3hot)
> > +     /* D3cold during system suspend/hibernate is not supported */
> > +     if (!dev->runtime_d3cold && state > PCI_D3hot)
>
> I don't think this is the right place for the dev->runtime_d3cold check,
> I've already told you that and haven't heard anything from you causing me
> to change my mind.
>
> If you don't want D3cold to be used during system sleep transitions,
> please make sure that D3cold is not passed to pci_set_power_state() in
> those cases.

OK.  Checked pci_set_power_state usage with following command line:

$ grep 'pci_set_power_state' -r . | grep -v PCI_D3hot | grep -v PCI_D0
| grep -v pci_choose_state
$ grep 'pci_set_power_state' -r . | grep -v PCI_D3hot | grep -v PCI_D0
| grep -v pci_choose_state | grep -i d3cold

Found that, almost all pci_set_power_state usage is:

1) set to PCI_Dx (not PCI_D3cold)
2) set to the state that choosed by pci_choose_state()
3) driver use pci_prepare_to_sleep(), which will call pci_set_power_state()

The only two exceptions are as follow:

./drivers/misc/cb710/core.c:            pci_set_power_state(pdev, PCI_D3cold);
./arch/x86/pci/mrst.c:  pci_set_power_state(dev, PCI_D3cold);

And pci_choose_state() will not return PCI_D3cold for system suspend.

So what do you think about the following solution?

a) Change pci_prepare_to_sleep(), so that it will not put device into
PCI_D3cold.
b) Change PCI_D3cold to PCI_D3hot in the following two places,

./drivers/misc/cb710/core.c:            pci_set_power_state(pdev, PCI_D3cold);
./arch/x86/pci/mrst.c:  pci_set_power_state(dev, PCI_D3cold);

Best Regards,
Huang Ying
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Rafael J. Wysocki - June 16, 2012, 9:09 p.m.
On Saturday, June 16, 2012, huang ying wrote:
> On Sat, Jun 16, 2012 at 5:21 AM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
> [snip]
> >
> > > @@ -731,9 +791,13 @@ int pci_set_power_state(struct pci_dev *
> > >  {
> > >       int error;
> > >
> > > -     /* bound the state we're entering */
> > > -     if (state > PCI_D3hot)
> > > +     /* D3cold during system suspend/hibernate is not supported */
> > > +     if (!dev->runtime_d3cold && state > PCI_D3hot)
> >
> > I don't think this is the right place for the dev->runtime_d3cold check,
> > I've already told you that and haven't heard anything from you causing me
> > to change my mind.
> >
> > If you don't want D3cold to be used during system sleep transitions,
> > please make sure that D3cold is not passed to pci_set_power_state() in
> > those cases.
> >
> 
> OK.  Checked pci_set_power_state usage with following command line:
> 
> $ grep 'pci_set_power_state' -r . | grep -v PCI_D3hot | grep -v PCI_D0 |
> grep -v pci_choose_state
> $ grep 'pci_set_power_state' -r . | grep -v PCI_D3hot | grep -v PCI_D0 |
> grep -v pci_choose_state | grep -i d3cold
> 
> Found that, almost all pci_set_power_state usage is:
> 
> 1) set to PCI_Dx (not PCI_D3cold)
> 2) set to the state that choosed by pci_choose_state()
> 3) driver use pci_prepare_to_sleep(), which will call pci_set_power_state()
> 
> The only two exceptions are as follow:
> 
> ./drivers/misc/cb710/core.c:            pci_set_power_state(pdev,
> PCI_D3cold);
> ./arch/x86/pci/mrst.c:  pci_set_power_state(dev, PCI_D3cold);

Well, no one is supposed to pass PCI_D3cold to pci_set_power_state() before
the $subject patch, so they should be fixed anyway.

> And pci_choose_state() will not return PCI_D3cold for system suspend.
> 
> So what do you think about the following solution?
> 
> a) Change pci_prepare_to_sleep(), so that it will not put device into
> PCI_D3cold.
> b) Change PCI_D3cold to PCI_D3hot in the following two places,
> 
> ./drivers/misc/cb710/core.c:            pci_set_power_state(pdev,
> PCI_D3cold);
> ./arch/x86/pci/mrst.c:  pci_set_power_state(dev, PCI_D3cold);

Yes, that will do.

Thanks,
Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

--- a/drivers/pci/pci-acpi.c
+++ b/drivers/pci/pci-acpi.c
@@ -48,6 +48,12 @@  static void pci_acpi_wake_dev(acpi_handl
 	if (event != ACPI_NOTIFY_DEVICE_WAKE || !pci_dev)
 		return;
 
+	if (pci_dev->current_state == PCI_D3cold) {
+		pci_wakeup_event(pci_dev);
+		pm_runtime_resume(&pci_dev->dev);
+		return;
+	}
+
 	if (!pci_dev->pm_cap || !pci_dev->pme_support
 	     || pci_check_pme_status(pci_dev)) {
 		if (pci_dev->pme_poll)
@@ -187,10 +193,13 @@  acpi_status pci_acpi_remove_pm_notifier(
 
 static pci_power_t acpi_pci_choose_state(struct pci_dev *pdev)
 {
-	int acpi_state;
+	int acpi_state, d_max;
 
-	acpi_state = acpi_pm_device_sleep_state(&pdev->dev, NULL,
-						ACPI_STATE_D3);
+	if (pdev->no_d3cold)
+		d_max = ACPI_STATE_D3_HOT;
+	else
+		d_max = ACPI_STATE_D3_COLD;
+	acpi_state = acpi_pm_device_sleep_state(&pdev->dev, NULL, d_max);
 	if (acpi_state < 0)
 		return PCI_POWER_ERROR;
 
@@ -297,7 +306,13 @@  static void acpi_pci_propagate_run_wake(
 
 static int acpi_pci_run_wake(struct pci_dev *dev, bool enable)
 {
-	if (dev->pme_interrupt)
+	/*
+	 * Per PCI Express Base Specification Revision 2.0 section
+	 * 5.3.3.2 Link Wakeup, platform support is needed for D3cold
+	 * waking up to power on the main link even if there is PME
+	 * support for D3cold
+	 */
+	if (dev->pme_interrupt && !dev->runtime_d3cold)
 		return 0;
 
 	if (!acpi_pm_device_run_wake(&dev->dev, enable))
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -1019,10 +1019,13 @@  static int pci_pm_runtime_suspend(struct
 	if (!pm || !pm->runtime_suspend)
 		return -ENOSYS;
 
+	pci_dev->no_d3cold = false;
 	error = pm->runtime_suspend(dev);
 	suspend_report_result(pm->runtime_suspend, error);
 	if (error)
 		return error;
+	if (!pci_dev->d3cold_allowed)
+		pci_dev->no_d3cold = true;
 
 	pci_fixup_device(pci_fixup_suspend, pci_dev);
 
@@ -1044,6 +1047,7 @@  static int pci_pm_runtime_suspend(struct
 
 static int pci_pm_runtime_resume(struct device *dev)
 {
+	int rc;
 	struct pci_dev *pci_dev = to_pci_dev(dev);
 	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
 
@@ -1054,7 +1058,11 @@  static int pci_pm_runtime_resume(struct
 	__pci_enable_wake(pci_dev, PCI_D0, true, false);
 	pci_fixup_device(pci_fixup_resume, pci_dev);
 
-	return pm->runtime_resume(dev);
+	rc = pm->runtime_resume(dev);
+
+	pci_dev->runtime_d3cold = false;
+
+	return rc;
 }
 
 static int pci_pm_runtime_idle(struct device *dev)
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -28,6 +28,7 @@ 
 #include <linux/pci-aspm.h>
 #include <linux/slab.h>
 #include <linux/vgaarb.h>
+#include <linux/pm_runtime.h>
 #include "pci.h"
 
 static int sysfs_initialized;	/* = 0 */
@@ -378,6 +379,31 @@  dev_bus_rescan_store(struct device *dev,
 
 #endif
 
+#if defined(CONFIG_PM_RUNTIME) && defined(CONFIG_ACPI)
+static ssize_t d3cold_allowed_store(struct device *dev,
+				    struct device_attribute *attr,
+				    const char *buf, size_t count)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	unsigned long val;
+
+	if (strict_strtoul(buf, 0, &val) < 0)
+		return -EINVAL;
+
+	pdev->d3cold_allowed = !!val;
+	pm_runtime_resume(dev);
+
+	return count;
+}
+
+static ssize_t d3cold_allowed_show(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	return sprintf (buf, "%u\n", pdev->d3cold_allowed);
+}
+#endif
+
 struct device_attribute pci_dev_attrs[] = {
 	__ATTR_RO(resource),
 	__ATTR_RO(vendor),
@@ -402,6 +428,9 @@  struct device_attribute pci_dev_attrs[]
 	__ATTR(remove, (S_IWUSR|S_IWGRP), NULL, remove_store),
 	__ATTR(rescan, (S_IWUSR|S_IWGRP), NULL, dev_rescan_store),
 #endif
+#if defined(CONFIG_PM_RUNTIME) && defined(CONFIG_ACPI)
+	__ATTR(d3cold_allowed, 0644, d3cold_allowed_show, d3cold_allowed_store),
+#endif
 	__ATTR_NULL,
 };
 
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -622,7 +622,8 @@  static int pci_raw_set_power_state(struc
 		dev_info(&dev->dev, "Refused to change power state, "
 			"currently in D%d\n", dev->current_state);
 
-	/* According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
+	/*
+	 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
 	 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
 	 * from D3hot to D0 _may_ perform an internal reset, thereby
 	 * going to "D0 Uninitialized" rather than "D0 Initialized".
@@ -654,6 +655,16 @@  void pci_update_current_state(struct pci
 	if (dev->pm_cap) {
 		u16 pmcsr;
 
+		/*
+		 * Configuration space is not accessible for device in
+		 * D3cold, so just keep or set D3cold for safety
+		 */
+		if (dev->current_state == PCI_D3cold)
+			return;
+		if (state == PCI_D3cold) {
+			dev->current_state = PCI_D3cold;
+			return;
+		}
 		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
 	} else {
@@ -694,8 +705,50 @@  static int pci_platform_power_transition
  */
 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
 {
-	if (state == PCI_D0)
+	if (state == PCI_D0) {
 		pci_platform_power_transition(dev, PCI_D0);
+		/*
+		 * Mandatory power management transition delays, see
+		 * PCI Express Base Specification Revision 2.0 Section
+		 * 6.6.1: Conventional Reset.  Do not delay for
+		 * devices powered on/off by corresponding bridge,
+		 * because have already delayed for the bridge.
+		 */
+		if (dev->runtime_d3cold) {
+			msleep(dev->d3cold_delay);
+			/*
+			 * When powering on a bridge from D3cold, the
+			 * whole hierarchy may be powered on into
+			 * D0uninitialized state, resume them to give
+			 * them a chance to suspend again
+			 */
+			pci_wakeup_bus(dev->subordinate);
+		}
+	}
+}
+
+/**
+ * __pci_dev_set_current_state - Set current state of a PCI device
+ * @dev: Device to handle
+ * @data: pointer to state to be set
+ */
+static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
+{
+	pci_power_t state = *(pci_power_t *)data;
+
+	dev->current_state = state;
+	return 0;
+}
+
+/**
+ * __pci_bus_set_current_state - Walk given bus and set current state of devices
+ * @bus: Top bus of the subtree to walk.
+ * @state: state to be set
+ */
+static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
+{
+	if (bus)
+		pci_walk_bus(bus, __pci_dev_set_current_state, &state);
 }
 
 /**
@@ -707,8 +760,15 @@  static void __pci_start_power_transition
  */
 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
 {
-	return state >= PCI_D0 ?
-			pci_platform_power_transition(dev, state) : -EINVAL;
+	int ret;
+
+	if (state < PCI_D0)
+		return -EINVAL;
+	ret = pci_platform_power_transition(dev, state);
+	/* Power off the bridge may power off the whole hierarchy */
+	if (!ret && state == PCI_D3cold)
+		__pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
+	return ret;
 }
 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
 
@@ -731,9 +791,13 @@  int pci_set_power_state(struct pci_dev *
 {
 	int error;
 
-	/* bound the state we're entering */
-	if (state > PCI_D3hot)
+	/* D3cold during system suspend/hibernate is not supported */
+	if (!dev->runtime_d3cold && state > PCI_D3hot)
 		state = PCI_D3hot;
+
+	/* bound the state we're entering */
+	if (state > PCI_D3cold)
+		state = PCI_D3cold;
 	else if (state < PCI_D0)
 		state = PCI_D0;
 	else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
@@ -748,10 +812,15 @@  int pci_set_power_state(struct pci_dev *
 
 	/* This device is quirked not to be put into D3, so
 	   don't put it in D3 */
-	if (state == PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
+	if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
 		return 0;
 
-	error = pci_raw_set_power_state(dev, state);
+	/*
+	 * To put device in D3cold, we put device into D3hot in native
+	 * way, then put device into D3cold with platform ops
+	 */
+	error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
+					PCI_D3hot : state);
 
 	if (!__pci_complete_power_transition(dev, state))
 		error = 0;
@@ -1498,6 +1567,28 @@  void pci_pme_wakeup_bus(struct pci_bus *
 }
 
 /**
+ * pci_wakeup - Wake up a PCI device
+ * @dev: Device to handle.
+ * @ign: ignored parameter
+ */
+static int pci_wakeup(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
+ * @bus: Top bus of the subtree to walk.
+ */
+void pci_wakeup_bus(struct pci_bus *bus)
+{
+	if (bus)
+		pci_walk_bus(bus, pci_wakeup, NULL);
+}
+
+/**
  * pci_pme_capable - check the capability of PCI device to generate PME#
  * @dev: PCI device to handle.
  * @state: PCI state from which device will issue PME#.
@@ -1791,12 +1882,16 @@  int pci_finish_runtime_suspend(struct pc
 	if (target_state == PCI_POWER_ERROR)
 		return -EIO;
 
+	dev->runtime_d3cold = target_state == PCI_D3cold;
+
 	__pci_enable_wake(dev, target_state, true, pci_dev_run_wake(dev));
 
 	error = pci_set_power_state(dev, target_state);
 
-	if (error)
+	if (error) {
 		__pci_enable_wake(dev, target_state, true, false);
+		dev->runtime_d3cold = false;
+	}
 
 	return error;
 }
@@ -1866,6 +1961,7 @@  void pci_pm_init(struct pci_dev *dev)
 
 	dev->pm_cap = pm;
 	dev->d3_delay = PCI_PM_D3_WAIT;
+	dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
 
 	dev->d1_support = false;
 	dev->d2_support = false;
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -70,6 +70,7 @@  extern void pci_update_current_state(str
 extern void pci_disable_enabled_device(struct pci_dev *dev);
 extern int pci_finish_runtime_suspend(struct pci_dev *dev);
 extern int __pci_pme_wakeup(struct pci_dev *dev, void *ign);
+extern void pci_wakeup_bus(struct pci_bus *bus);
 extern void pci_pm_init(struct pci_dev *dev);
 extern void platform_pci_wakeup_init(struct pci_dev *dev);
 extern void pci_allocate_cap_save_buffers(struct pci_dev *dev);
--- a/drivers/pci/pcie/portdrv_pci.c
+++ b/drivers/pci/pcie/portdrv_pci.c
@@ -101,12 +101,48 @@  static int pcie_port_resume_noirq(struct
 }
 
 #ifdef CONFIG_PM_RUNTIME
-static int pcie_port_runtime_pm(struct device *dev)
+struct d3cold_info {
+	bool no_d3cold;
+	unsigned int d3cold_delay;
+};
+
+static int pci_dev_d3cold_info(struct pci_dev *pdev, void *data)
+{
+	struct d3cold_info *info = data;
+
+	info->d3cold_delay = max_t(unsigned int, pdev->d3cold_delay,
+				   info->d3cold_delay);
+	if (pdev->no_d3cold)
+		info->no_d3cold = true;
+	return 0;
+}
+
+static int pcie_port_runtime_suspend(struct device *dev)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct d3cold_info d3cold_info = {
+		.no_d3cold	= false,
+		.d3cold_delay	= PCI_PM_D3_WAIT,
+	};
+
+	/*
+	 * If any subordinate device disable D3cold, we should not put
+	 * the port into D3cold.  The D3cold delay of port should be
+	 * the max of that of all subordinate devices.
+	 */
+	pci_walk_bus(pdev->subordinate, pci_dev_d3cold_info, &d3cold_info);
+	pdev->no_d3cold = d3cold_info.no_d3cold;
+	pdev->d3cold_delay = d3cold_info.d3cold_delay;
+	return 0;
+}
+
+static int pcie_port_runtime_resume(struct device *dev)
 {
 	return 0;
 }
 #else
-#define pcie_port_runtime_pm	NULL
+#define pcie_port_runtime_suspend	NULL
+#define pcie_port_runtime_resume	NULL
 #endif
 
 static const struct dev_pm_ops pcie_portdrv_pm_ops = {
@@ -117,8 +153,8 @@  static const struct dev_pm_ops pcie_port
 	.poweroff	= pcie_port_device_suspend,
 	.restore	= pcie_port_device_resume,
 	.resume_noirq	= pcie_port_resume_noirq,
-	.runtime_suspend = pcie_port_runtime_pm,
-	.runtime_resume = pcie_port_runtime_pm,
+	.runtime_suspend = pcie_port_runtime_suspend,
+	.runtime_resume = pcie_port_runtime_resume,
 };
 
 #define PCIE_PORTDRV_PM_OPS	(&pcie_portdrv_pm_ops)
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -132,9 +132,10 @@  static inline const char *pci_power_name
 	return pci_power_names[1 + (int) state];
 }
 
-#define PCI_PM_D2_DELAY	200
-#define PCI_PM_D3_WAIT	10
-#define PCI_PM_BUS_WAIT	50
+#define PCI_PM_D2_DELAY		200
+#define PCI_PM_D3_WAIT		10
+#define PCI_PM_D3COLD_WAIT	100
+#define PCI_PM_BUS_WAIT		50
 
 /** The pci_channel state describes connectivity between the CPU and
  *  the pci device.  If some PCI bus between here and the pci device
@@ -278,11 +279,18 @@  struct pci_dev {
 	unsigned int	pme_poll:1;	/* Poll device's PME status bit */
 	unsigned int	d1_support:1;	/* Low power state D1 is supported */
 	unsigned int	d2_support:1;	/* Low power state D2 is supported */
-	unsigned int	no_d1d2:1;	/* Only allow D0 and D3 */
+	unsigned int	no_d1d2:1;	/* D1 and D2 are forbidden */
+	unsigned int	no_d3cold:1;	/* D3cold is forbidden */
+	unsigned int	d3cold_allowed:1;	/* D3cold is allowed by user */
 	unsigned int	mmio_always_on:1;	/* disallow turning off io/mem
 						   decoding during bar sizing */
 	unsigned int	wakeup_prepared:1;
+	unsigned int	runtime_d3cold:1;	/* whether go through runtime
+						   D3cold, not set for devices
+						   powered on/off by the
+						   corresponding bridge */
 	unsigned int	d3_delay;	/* D3->D0 transition time in ms */
+	unsigned int	d3cold_delay;	/* D3cold->D0 transition time in ms */
 
 #ifdef CONFIG_PCIEASPM
 	struct pcie_link_state	*link_state;	/* ASPM link state. */