[PATCH -v3 0/2] PCIe: Add PCIe runtime D3cold support

[PATCH -v3 0/2] PCIe: Add PCIe runtime D3cold support

[Huang Ying]

This patchset is based on the following patch:

[PATCH 1/5] ACPI, PM, Specify lowest allowed state for device sleep state

Changelog:

v3:

- Drop device.power.power_must_be_on, use pci_dev.no_d3cold instead
- Drop device.power.power_off_user, use pci_dev.d3cold_allowed instead
- Use black list instead of white list in pcie port runtime support per request from Bjorn
- Various fixes based on comments from Rafael


v2:

- Refreshed based on comments from Rafael

[PATCH -v3 1/2] PCIe: Add runtime PM support to PCIe port
[PATCH -v3 2/2] PCIe: Add PCIe runtime D3cold support

Best Regards,
Huang Ying

[PATCH -v3 1/2] PCIe: Add runtime PM support to PCIe port

[Huang Ying]

From: "Yan, Zheng" <zheng.z.yan@intel.com>

This patch adds runtime PM support to PCIe port.  This is needed by
PCIe D3cold support, where PCIe device without ACPI node may be
powered on/off by PCIe port.

Because runtime suspend is broken for some chipsets, a black list is
used to disable runtime PM support for these chipsets.

Signed-off-by: Zheng Yan <zheng.z.yan@intel.com>
---
 drivers/pci/pci.c              |   10 ++++++++++
 drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++++++++++++++++++++++
 2 files changed, 40 insertions(+)

--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -1517,6 +1517,16 @@ static void pci_pme_list_scan(struct wor
 	if (!list_empty(&pci_pme_list)) {
 		list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
 			if (pme_dev->dev->pme_poll) {
+				struct pci_dev *bridge;
+
+				bridge = pme_dev->dev->bus->self;
+				/*
+				 * If bridge is in low power state, the
+				 * configuration space of subordinate devices
+				 * may be not accessible
+				 */
+				if (bridge && bridge->current_state != PCI_D0)
+					continue;
 				pci_pme_wakeup(pme_dev->dev, NULL);
 			} else {
 				list_del(&pme_dev->list);
--- a/drivers/pci/pcie/portdrv_pci.c
+++ b/drivers/pci/pcie/portdrv_pci.c
@@ -11,6 +11,7 @@
 #include <linux/kernel.h>
 #include <linux/errno.h>
 #include <linux/pm.h>
+#include <linux/pm_runtime.h>
 #include <linux/init.h>
 #include <linux/pcieport_if.h>
 #include <linux/aer.h>
@@ -99,6 +100,21 @@ static int pcie_port_resume_noirq(struct
 	return 0;
 }
 
+#ifdef CONFIG_PM_RUNTIME
+static int pcie_port_runtime_suspend(struct device *dev)
+{
+	return 0;
+}
+
+static int pcie_port_runtime_resume(struct device *dev)
+{
+	return 0;
+}
+#else
+#define pcie_port_runtime_suspend	NULL
+#define pcie_port_runtime_resume	NULL
+#endif
+
 static const struct dev_pm_ops pcie_portdrv_pm_ops = {
 	.suspend	= pcie_port_device_suspend,
 	.resume		= pcie_port_device_resume,
@@ -107,6 +123,8 @@ 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_suspend,
+	.runtime_resume = pcie_port_runtime_resume,
 };
 
 #define PCIE_PORTDRV_PM_OPS	(&pcie_portdrv_pm_ops)
@@ -117,6 +135,14 @@ static const struct dev_pm_ops pcie_port
 #endif /* !PM */
 
 /*
+ * PCIe port runtime suspend is broken for some chipsets, so use a
+ * black list to disable runtime PM for these chipsets.
+ */
+static const struct pci_device_id port_runtime_pm_black_list[] = {
+	{ /* end: all zeroes */ }
+};
+
+/*
  * pcie_portdrv_probe - Probe PCI-Express port devices
  * @dev: PCI-Express port device being probed
  *
@@ -144,12 +170,16 @@ static int __devinit pcie_portdrv_probe(
 		return status;
 
 	pci_save_state(dev);
+	if (!pci_match_id(port_runtime_pm_black_list, dev))
+		pm_runtime_put_noidle(&dev->dev);
 
 	return 0;
 }
 
 static void pcie_portdrv_remove(struct pci_dev *dev)
 {
+	if (!pci_match_id(port_runtime_pm_black_list, dev))
+		pm_runtime_get_noresume(&dev->dev);
 	pcie_port_device_remove(dev);
 	pci_disable_device(dev);
 }

Re: [PATCH -v3 1/2] PCIe: Add runtime PM support to PCIe port

[Rafael J. Wysocki]

On Tuesday, May 15, 2012, Huang Ying wrote:
> From: "Yan, Zheng" <zheng.z.yan@intel.com>
> 
> This patch adds runtime PM support to PCIe port.  This is needed by
> PCIe D3cold support, where PCIe device without ACPI node may be
> powered on/off by PCIe port.
> 
> Because runtime suspend is broken for some chipsets, a black list is
> used to disable runtime PM support for these chipsets.
> 
> Signed-off-by: Zheng Yan <zheng.z.yan@intel.com>
> ---
>  drivers/pci/pci.c              |   10 ++++++++++
>  drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++++++++++++++++++++++
>  2 files changed, 40 insertions(+)
> 
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -1517,6 +1517,16 @@ static void pci_pme_list_scan(struct wor
>  	if (!list_empty(&pci_pme_list)) {
>  		list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
>  			if (pme_dev->dev->pme_poll) {
> +				struct pci_dev *bridge;
> +
> +				bridge = pme_dev->dev->bus->self;
> +				/*
> +				 * If bridge is in low power state, the
> +				 * configuration space of subordinate devices
> +				 * may be not accessible
> +				 */
> +				if (bridge && bridge->current_state != PCI_D0)
> +					continue;
>  				pci_pme_wakeup(pme_dev->dev, NULL);
>  			} else {
>  				list_del(&pme_dev->list);
> --- a/drivers/pci/pcie/portdrv_pci.c
> +++ b/drivers/pci/pcie/portdrv_pci.c
> @@ -11,6 +11,7 @@
>  #include <linux/kernel.h>
>  #include <linux/errno.h>
>  #include <linux/pm.h>
> +#include <linux/pm_runtime.h>
>  #include <linux/init.h>
>  #include <linux/pcieport_if.h>
>  #include <linux/aer.h>
> @@ -99,6 +100,21 @@ static int pcie_port_resume_noirq(struct
>  	return 0;
>  }
>  
> +#ifdef CONFIG_PM_RUNTIME
> +static int pcie_port_runtime_suspend(struct device *dev)
> +{
> +	return 0;
> +}
> +
> +static int pcie_port_runtime_resume(struct device *dev)
> +{
> +	return 0;
> +}

Do we need those empty definitions?  If so, it might be just one
function.

The patch looks good otherwise.

Thanks,
Rafael


> +#else
> +#define pcie_port_runtime_suspend	NULL
> +#define pcie_port_runtime_resume	NULL
> +#endif
> +
>  static const struct dev_pm_ops pcie_portdrv_pm_ops = {
>  	.suspend	= pcie_port_device_suspend,
>  	.resume		= pcie_port_device_resume,
> @@ -107,6 +123,8 @@ 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_suspend,
> +	.runtime_resume = pcie_port_runtime_resume,
>  };
>  
>  #define PCIE_PORTDRV_PM_OPS	(&pcie_portdrv_pm_ops)
> @@ -117,6 +135,14 @@ static const struct dev_pm_ops pcie_port
>  #endif /* !PM */
>  
>  /*
> + * PCIe port runtime suspend is broken for some chipsets, so use a
> + * black list to disable runtime PM for these chipsets.
> + */
> +static const struct pci_device_id port_runtime_pm_black_list[] = {
> +	{ /* end: all zeroes */ }
> +};
> +
> +/*
>   * pcie_portdrv_probe - Probe PCI-Express port devices
>   * @dev: PCI-Express port device being probed
>   *
> @@ -144,12 +170,16 @@ static int __devinit pcie_portdrv_probe(
>  		return status;
>  
>  	pci_save_state(dev);
> +	if (!pci_match_id(port_runtime_pm_black_list, dev))
> +		pm_runtime_put_noidle(&dev->dev);
>  
>  	return 0;
>  }
>  
>  static void pcie_portdrv_remove(struct pci_dev *dev)
>  {
> +	if (!pci_match_id(port_runtime_pm_black_list, dev))
> +		pm_runtime_get_noresume(&dev->dev);
>  	pcie_port_device_remove(dev);
>  	pci_disable_device(dev);
>  }
> 
> 

Re: [PATCH -v3 1/2] PCIe: Add runtime PM support to PCIe port

[Huang Ying]

On Tue, 2012-05-15 at 20:53 +0200, Rafael J. Wysocki wrote:
> On Tuesday, May 15, 2012, Huang Ying wrote:
> > From: "Yan, Zheng" <zheng.z.yan@intel.com>
> > 
> > This patch adds runtime PM support to PCIe port.  This is needed by
> > PCIe D3cold support, where PCIe device without ACPI node may be
> > powered on/off by PCIe port.
> > 
> > Because runtime suspend is broken for some chipsets, a black list is
> > used to disable runtime PM support for these chipsets.
> > 
> > Signed-off-by: Zheng Yan <zheng.z.yan@intel.com>
> > ---
> >  drivers/pci/pci.c              |   10 ++++++++++
> >  drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++++++++++++++++++++++
> >  2 files changed, 40 insertions(+)
> > 
> > --- a/drivers/pci/pci.c
> > +++ b/drivers/pci/pci.c
> > @@ -1517,6 +1517,16 @@ static void pci_pme_list_scan(struct wor
> >  	if (!list_empty(&pci_pme_list)) {
> >  		list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
> >  			if (pme_dev->dev->pme_poll) {
> > +				struct pci_dev *bridge;
> > +
> > +				bridge = pme_dev->dev->bus->self;
> > +				/*
> > +				 * If bridge is in low power state, the
> > +				 * configuration space of subordinate devices
> > +				 * may be not accessible
> > +				 */
> > +				if (bridge && bridge->current_state != PCI_D0)
> > +					continue;
> >  				pci_pme_wakeup(pme_dev->dev, NULL);
> >  			} else {
> >  				list_del(&pme_dev->list);
> > --- a/drivers/pci/pcie/portdrv_pci.c
> > +++ b/drivers/pci/pcie/portdrv_pci.c
> > @@ -11,6 +11,7 @@
> >  #include <linux/kernel.h>
> >  #include <linux/errno.h>
> >  #include <linux/pm.h>
> > +#include <linux/pm_runtime.h>
> >  #include <linux/init.h>
> >  #include <linux/pcieport_if.h>
> >  #include <linux/aer.h>
> > @@ -99,6 +100,21 @@ static int pcie_port_resume_noirq(struct
> >  	return 0;
> >  }
> >  
> > +#ifdef CONFIG_PM_RUNTIME
> > +static int pcie_port_runtime_suspend(struct device *dev)
> > +{
> > +	return 0;
> > +}
> > +
> > +static int pcie_port_runtime_resume(struct device *dev)
> > +{
> > +	return 0;
> > +}
> 
> Do we need those empty definitions?  If so, it might be just one
> function.

Yes.  The empty definition is necessary.  I will use just one function
here.

Best Regards,
Huang Ying

[PATCH -v3 2/2] PCIe: Add PCIe runtime D3cold support

[Huang Ying]

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         |   26 +++++++--
 drivers/pci/pci-driver.c       |    3 +
 drivers/pci/pci-sysfs.c        |   29 ++++++++++
 drivers/pci/pci.c              |  117 +++++++++++++++++++++++++++++++++++++----
 drivers/pci/pci.h              |    1 
 drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++
 include/linux/pci.h            |   16 ++++-
 7 files changed, 204 insertions(+), 18 deletions(-)

--- a/drivers/pci/pci-acpi.c
+++ b/drivers/pci/pci-acpi.c
@@ -48,6 +48,15 @@ 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_get_sync(&pci_dev->dev);
+		/* Powering on bridge needs to resume whole hierarchy */
+		pci_wakeup_bus(pci_dev->subordinate, true);
+		pm_runtime_put(&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 +196,13 @@ 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 +309,13 @@ 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
@@ -1013,10 +1013,13 @@ 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);
 
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -27,6 +27,7 @@
 #include <linux/security.h>
 #include <linux/pci-aspm.h>
 #include <linux/slab.h>
+#include <linux/pm_runtime.h>
 #include "pci.h"
 
 static int sysfs_initialized;	/* = 0 */
@@ -377,6 +378,31 @@ 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),
@@ -401,6 +427,9 @@ 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
@@ -653,6 +653,12 @@ void pci_update_current_state(struct pci
 	if (dev->pm_cap) {
 		u16 pmcsr;
 
+		/*
+		 * Configuration space is not accessible for device in
+		 * D3cold, so keep in D3cold for safety
+		 */
+		if (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 {
@@ -671,8 +677,12 @@ static int pci_platform_power_transition
 
 	if (platform_pci_power_manageable(dev)) {
 		error = platform_pci_set_power_state(dev, state);
-		if (!error)
-			pci_update_current_state(dev, state);
+		if (!error) {
+			if (state == PCI_D3cold)
+				dev->current_state = PCI_D3cold;
+			else
+				pci_update_current_state(dev, state);
+		}
 		/* Fall back to PCI_D0 if native PM is not supported */
 		if (!dev->pm_cap)
 			dev->current_state = PCI_D0;
@@ -693,8 +703,49 @@ 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, false);
+		}
+	}
+}
+
+/**
+ * __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);
 }
 
 /**
@@ -706,8 +757,15 @@ 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);
 
@@ -731,8 +789,8 @@ int pci_set_power_state(struct pci_dev *
 	int error;
 
 	/* bound the state we're entering */
-	if (state > PCI_D3hot)
-		state = PCI_D3hot;
+	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))
@@ -747,10 +805,15 @@ 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;
@@ -1497,6 +1560,35 @@ void pci_pme_wakeup_bus(struct pci_bus *
 }
 
 /**
+ * pci_wakeup - Wake up a PCI device
+ * @dev: Device to handle.
+ * @data: whether to resume first device synchronously
+ */
+static int pci_wakeup(struct pci_dev *pci_dev, void *data)
+{
+	bool *sync_first = data;
+
+	pci_wakeup_event(pci_dev);
+	if (*sync_first) {
+		pm_runtime_resume(&pci_dev->dev);
+		*sync_first = false;
+	} else
+		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.
+ * @sync_first: whether to resume first device synchronously
+ */
+void pci_wakeup_bus(struct pci_bus *bus, bool sync_first)
+{
+	if (bus)
+		pci_walk_bus(bus, pci_wakeup, &sync_first);
+}
+
+/**
  * 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#.
@@ -1790,12 +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;
 }
@@ -1865,6 +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, bool sync_first);
 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
@@ -101,8 +101,38 @@ static int pcie_port_resume_noirq(struct
 }
 
 #ifdef CONFIG_PM_RUNTIME
+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;
 }
 
--- 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. */

Re: [PATCH -v3 2/2] PCIe: Add PCIe runtime D3cold support

[Rafael J. Wysocki]

On Tuesday, May 15, 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         |   26 +++++++--
>  drivers/pci/pci-driver.c       |    3 +
>  drivers/pci/pci-sysfs.c        |   29 ++++++++++
>  drivers/pci/pci.c              |  117 +++++++++++++++++++++++++++++++++++++----
>  drivers/pci/pci.h              |    1 
>  drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++
>  include/linux/pci.h            |   16 ++++-
>  7 files changed, 204 insertions(+), 18 deletions(-)
> 
> --- a/drivers/pci/pci-acpi.c
> +++ b/drivers/pci/pci-acpi.c
> @@ -48,6 +48,15 @@ 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_get_sync(&pci_dev->dev);
> +		/* Powering on bridge needs to resume whole hierarchy */
> +		pci_wakeup_bus(pci_dev->subordinate, true);
> +		pm_runtime_put(&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 +196,13 @@ 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 +309,13 @@ 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
> @@ -1013,10 +1013,13 @@ 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);
>  
> --- a/drivers/pci/pci-sysfs.c
> +++ b/drivers/pci/pci-sysfs.c
> @@ -27,6 +27,7 @@
>  #include <linux/security.h>
>  #include <linux/pci-aspm.h>
>  #include <linux/slab.h>
> +#include <linux/pm_runtime.h>
>  #include "pci.h"
>  
>  static int sysfs_initialized;	/* = 0 */
> @@ -377,6 +378,31 @@ 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),
> @@ -401,6 +427,9 @@ 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
> @@ -653,6 +653,12 @@ void pci_update_current_state(struct pci
>  	if (dev->pm_cap) {
>  		u16 pmcsr;
>  
> +		/*
> +		 * Configuration space is not accessible for device in
> +		 * D3cold, so keep in D3cold for safety
> +		 */
> +		if (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 {
> @@ -671,8 +677,12 @@ static int pci_platform_power_transition
>  
>  	if (platform_pci_power_manageable(dev)) {
>  		error = platform_pci_set_power_state(dev, state);
> -		if (!error)
> -			pci_update_current_state(dev, state);
> +		if (!error) {
> +			if (state == PCI_D3cold)
> +				dev->current_state = PCI_D3cold;

I would put the above into pci_update_current_state(), so that this entire
hunk wouldn't be necessary.

> +			else
> +				pci_update_current_state(dev, state);
> +		}
>  		/* Fall back to PCI_D0 if native PM is not supported */
>  		if (!dev->pm_cap)
>  			dev->current_state = PCI_D0;
> @@ -693,8 +703,49 @@ 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

Put the /* into a separate line, please.

> +			 * whole hierarchy may be powered on into
> +			 * D0uninitialized state, resume them to give
> +			 * them a chance to suspend again
> +			 */
> +			pci_wakeup_bus(dev->subordinate, false);
> +		}
> +	}
> +}
> +
> +/**
> + * __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);
>  }
>  
>  /**
> @@ -706,8 +757,15 @@ 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);
>  
> @@ -731,8 +789,8 @@ int pci_set_power_state(struct pci_dev *
>  	int error;
>  
>  	/* bound the state we're entering */
> -	if (state > PCI_D3hot)
> -		state = PCI_D3hot;
> +	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))
> @@ -747,10 +805,15 @@ 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);

I wonder if we really need to transition the device into PCI_D3hot before
removing power from it?  If not, this whole thing might be simplified.

>  
>  	if (!__pci_complete_power_transition(dev, state))
>  		error = 0;
> @@ -1497,6 +1560,35 @@ void pci_pme_wakeup_bus(struct pci_bus *
>  }
>  
>  /**
> + * pci_wakeup - Wake up a PCI device
> + * @dev: Device to handle.
> + * @data: whether to resume first device synchronously
> + */
> +static int pci_wakeup(struct pci_dev *pci_dev, void *data)
> +{
> +	bool *sync_first = data;
> +
> +	pci_wakeup_event(pci_dev);
> +	if (*sync_first) {

Why exactly do we need to do this?

> +		pm_runtime_resume(&pci_dev->dev);
> +		*sync_first = false;
> +	} else
> +		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.
> + * @sync_first: whether to resume first device synchronously
> + */
> +void pci_wakeup_bus(struct pci_bus *bus, bool sync_first)
> +{
> +	if (bus)
> +		pci_walk_bus(bus, pci_wakeup, &sync_first);
> +}
> +
> +/**
>   * 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#.
> @@ -1790,12 +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;
>  }
> @@ -1865,6 +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, bool sync_first);
>  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
> @@ -101,8 +101,38 @@ static int pcie_port_resume_noirq(struct
>  }
>  
>  #ifdef CONFIG_PM_RUNTIME
> +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;
>  }
>  
> --- 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. */

Looks good apart from the comments above.

Thanks,
Rafael

Re: [PATCH -v3 2/2] PCIe: Add PCIe runtime D3cold support

[Huang Ying]

On Tue, 2012-05-15 at 21:35 +0200, Rafael J. Wysocki wrote:
> On Tuesday, May 15, 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         |   26 +++++++--
> >  drivers/pci/pci-driver.c       |    3 +
> >  drivers/pci/pci-sysfs.c        |   29 ++++++++++
> >  drivers/pci/pci.c              |  117 +++++++++++++++++++++++++++++++++++++----
> >  drivers/pci/pci.h              |    1 
> >  drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++
> >  include/linux/pci.h            |   16 ++++-
> >  7 files changed, 204 insertions(+), 18 deletions(-)
> > 
> > --- a/drivers/pci/pci-acpi.c
> > +++ b/drivers/pci/pci-acpi.c
> > @@ -48,6 +48,15 @@ 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_get_sync(&pci_dev->dev);
> > +		/* Powering on bridge needs to resume whole hierarchy */
> > +		pci_wakeup_bus(pci_dev->subordinate, true);
> > +		pm_runtime_put(&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 +196,13 @@ 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 +309,13 @@ 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
> > @@ -1013,10 +1013,13 @@ 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);
> >  
> > --- a/drivers/pci/pci-sysfs.c
> > +++ b/drivers/pci/pci-sysfs.c
> > @@ -27,6 +27,7 @@
> >  #include <linux/security.h>
> >  #include <linux/pci-aspm.h>
> >  #include <linux/slab.h>
> > +#include <linux/pm_runtime.h>
> >  #include "pci.h"
> >  
> >  static int sysfs_initialized;	/* = 0 */
> > @@ -377,6 +378,31 @@ 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),
> > @@ -401,6 +427,9 @@ 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
> > @@ -653,6 +653,12 @@ void pci_update_current_state(struct pci
> >  	if (dev->pm_cap) {
> >  		u16 pmcsr;
> >  
> > +		/*
> > +		 * Configuration space is not accessible for device in
> > +		 * D3cold, so keep in D3cold for safety
> > +		 */
> > +		if (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 {
> > @@ -671,8 +677,12 @@ static int pci_platform_power_transition
> >  
> >  	if (platform_pci_power_manageable(dev)) {
> >  		error = platform_pci_set_power_state(dev, state);
> > -		if (!error)
> > -			pci_update_current_state(dev, state);
> > +		if (!error) {
> > +			if (state == PCI_D3cold)
> > +				dev->current_state = PCI_D3cold;
> 
> I would put the above into pci_update_current_state(), so that this entire
> hunk wouldn't be necessary.

OK.  Will do that.

> > +			else
> > +				pci_update_current_state(dev, state);
> > +		}
> >  		/* Fall back to PCI_D0 if native PM is not supported */
> >  		if (!dev->pm_cap)
> >  			dev->current_state = PCI_D0;
> > @@ -693,8 +703,49 @@ 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
> 
> Put the /* into a separate line, please.

Sorry, will change it.

> > +			 * whole hierarchy may be powered on into
> > +			 * D0uninitialized state, resume them to give
> > +			 * them a chance to suspend again
> > +			 */
> > +			pci_wakeup_bus(dev->subordinate, false);
> > +		}
> > +	}
> > +}
> > +
> > +/**
> > + * __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);
> >  }
> >  
> >  /**
> > @@ -706,8 +757,15 @@ 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);
> >  
> > @@ -731,8 +789,8 @@ int pci_set_power_state(struct pci_dev *
> >  	int error;
> >  
> >  	/* bound the state we're entering */
> > -	if (state > PCI_D3hot)
> > -		state = PCI_D3hot;
> > +	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))
> > @@ -747,10 +805,15 @@ 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);
> 
> I wonder if we really need to transition the device into PCI_D3hot before
> removing power from it?  If not, this whole thing might be simplified.

>From a book named PCI Express System Architecture, Figure 16-6. PCI
Express Function Power management State Transitions.  The D3 Cold can be
transition to only from D3 Hot.  But did not find the similar figure in
PCI express spec.

>From the practical point of view.  If the power resources to power
on/off the PCIe device is shared by several devices.  If we do not
transition the device into PCI_D3hot before removing power from it, and
another device share the same power resources is not powered off, we may
waste quite some energy.

Here is another possible situation where the underlying ACPI may not
really power off the device.  If the device has only _PS3 and no _PR3,
for safety, we treat the state after executing _PS3 as D3cold, but in
fact, it may be just D3hot, we need transition the device into D3hot
here.

> >  
> >  	if (!__pci_complete_power_transition(dev, state))
> >  		error = 0;
> > @@ -1497,6 +1560,35 @@ void pci_pme_wakeup_bus(struct pci_bus *
> >  }
> >  
> >  /**
> > + * pci_wakeup - Wake up a PCI device
> > + * @dev: Device to handle.
> > + * @data: whether to resume first device synchronously
> > + */
> > +static int pci_wakeup(struct pci_dev *pci_dev, void *data)
> > +{
> > +	bool *sync_first = data;
> > +
> > +	pci_wakeup_event(pci_dev);
> > +	if (*sync_first) {
> 
> Why exactly do we need to do this?

This is used to avoid the the following situation:

1) The devices under the PCIe port and the PCIe port itself are in
D3cold state
2) Remote wakeup of a device is triggered, so the #WAKE pin of the
device is asserted
3) The #WAKE pin is connected with GPE logic, so Lxx handler will be
executed. In Lxx handler, Notify the ACPI handler of the PCIe port
4) In the PM handler of the PCIe port (pci_acpi_wake_dev), the PCIe port
will be resumed, so that the PCIe port and the devices are powered on to
D0uninitialized state.  So the devices under the PCIe port should be
resumed too (with pci_wakeup_bus).

Without sync_first logic introduced here, all the devices under the PCIe
port are resumed asynchronously, and during the resume process of the
PCIe port, a suspending request is pending for the PCIe port too.  So it
is possible for the PCIe port is resumed, suspended, then resumed again.

With sync_fist logic, one of the devices under the PCIe port are resumed
synchronously, so that the PCIe port is guaranteed to be kept in ACTIVE
state.

If you think this situation is so seldom that we need not care too much,
I will remove the sync_first logic at least for now.

Best Regards,
Huang Ying

Re: [PATCH -v3 2/2] PCIe: Add PCIe runtime D3cold support

[Rafael J. Wysocki]

On Thursday, May 17, 2012, Huang Ying wrote:
> On Tue, 2012-05-15 at 21:35 +0200, Rafael J. Wysocki wrote:
> > On Tuesday, May 15, 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         |   26 +++++++--
> > >  drivers/pci/pci-driver.c       |    3 +
> > >  drivers/pci/pci-sysfs.c        |   29 ++++++++++
> > >  drivers/pci/pci.c              |  117 +++++++++++++++++++++++++++++++++++++----
> > >  drivers/pci/pci.h              |    1 
> > >  drivers/pci/pcie/portdrv_pci.c |   30 ++++++++++
> > >  include/linux/pci.h            |   16 ++++-
> > >  7 files changed, 204 insertions(+), 18 deletions(-)
> > > 
> > > --- a/drivers/pci/pci-acpi.c
> > > +++ b/drivers/pci/pci-acpi.c
> > > @@ -48,6 +48,15 @@ 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_get_sync(&pci_dev->dev);
> > > +		/* Powering on bridge needs to resume whole hierarchy */
> > > +		pci_wakeup_bus(pci_dev->subordinate, true);
> > > +		pm_runtime_put(&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 +196,13 @@ 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 +309,13 @@ 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
> > > @@ -1013,10 +1013,13 @@ 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);
> > >  
> > > --- a/drivers/pci/pci-sysfs.c
> > > +++ b/drivers/pci/pci-sysfs.c
> > > @@ -27,6 +27,7 @@
> > >  #include <linux/security.h>
> > >  #include <linux/pci-aspm.h>
> > >  #include <linux/slab.h>
> > > +#include <linux/pm_runtime.h>
> > >  #include "pci.h"
> > >  
> > >  static int sysfs_initialized;	/* = 0 */
> > > @@ -377,6 +378,31 @@ 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),
> > > @@ -401,6 +427,9 @@ 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
> > > @@ -653,6 +653,12 @@ void pci_update_current_state(struct pci
> > >  	if (dev->pm_cap) {
> > >  		u16 pmcsr;
> > >  
> > > +		/*
> > > +		 * Configuration space is not accessible for device in
> > > +		 * D3cold, so keep in D3cold for safety
> > > +		 */
> > > +		if (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 {
> > > @@ -671,8 +677,12 @@ static int pci_platform_power_transition
> > >  
> > >  	if (platform_pci_power_manageable(dev)) {
> > >  		error = platform_pci_set_power_state(dev, state);
> > > -		if (!error)
> > > -			pci_update_current_state(dev, state);
> > > +		if (!error) {
> > > +			if (state == PCI_D3cold)
> > > +				dev->current_state = PCI_D3cold;
> > 
> > I would put the above into pci_update_current_state(), so that this entire
> > hunk wouldn't be necessary.
> 
> OK.  Will do that.
> 
> > > +			else
> > > +				pci_update_current_state(dev, state);
> > > +		}
> > >  		/* Fall back to PCI_D0 if native PM is not supported */
> > >  		if (!dev->pm_cap)
> > >  			dev->current_state = PCI_D0;
> > > @@ -693,8 +703,49 @@ 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
> > 
> > Put the /* into a separate line, please.
> 
> Sorry, will change it.
> 
> > > +			 * whole hierarchy may be powered on into
> > > +			 * D0uninitialized state, resume them to give
> > > +			 * them a chance to suspend again
> > > +			 */
> > > +			pci_wakeup_bus(dev->subordinate, false);
> > > +		}
> > > +	}
> > > +}
> > > +
> > > +/**
> > > + * __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);
> > >  }
> > >  
> > >  /**
> > > @@ -706,8 +757,15 @@ 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);
> > >  
> > > @@ -731,8 +789,8 @@ int pci_set_power_state(struct pci_dev *
> > >  	int error;
> > >  
> > >  	/* bound the state we're entering */
> > > -	if (state > PCI_D3hot)
> > > -		state = PCI_D3hot;
> > > +	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))
> > > @@ -747,10 +805,15 @@ 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);
> > 
> > I wonder if we really need to transition the device into PCI_D3hot before
> > removing power from it?  If not, this whole thing might be simplified.
> 
> From a book named PCI Express System Architecture, Figure 16-6. PCI
> Express Function Power management State Transitions.  The D3 Cold can be
> transition to only from D3 Hot.  But did not find the similar figure in
> PCI express spec.
> 
> From the practical point of view.  If the power resources to power
> on/off the PCIe device is shared by several devices.  If we do not
> transition the device into PCI_D3hot before removing power from it, and
> another device share the same power resources is not powered off, we may
> waste quite some energy.
> 
> Here is another possible situation where the underlying ACPI may not
> really power off the device.  If the device has only _PS3 and no _PR3,
> for safety, we treat the state after executing _PS3 as D3cold, but in
> fact, it may be just D3hot, we need transition the device into D3hot
> here.
> 
> > >  
> > >  	if (!__pci_complete_power_transition(dev, state))
> > >  		error = 0;
> > > @@ -1497,6 +1560,35 @@ void pci_pme_wakeup_bus(struct pci_bus *
> > >  }
> > >  
> > >  /**
> > > + * pci_wakeup - Wake up a PCI device
> > > + * @dev: Device to handle.
> > > + * @data: whether to resume first device synchronously
> > > + */
> > > +static int pci_wakeup(struct pci_dev *pci_dev, void *data)
> > > +{
> > > +	bool *sync_first = data;
> > > +
> > > +	pci_wakeup_event(pci_dev);
> > > +	if (*sync_first) {
> > 
> > Why exactly do we need to do this?
> 
> This is used to avoid the the following situation:
> 
> 1) The devices under the PCIe port and the PCIe port itself are in
> D3cold state
> 2) Remote wakeup of a device is triggered, so the #WAKE pin of the
> device is asserted
> 3) The #WAKE pin is connected with GPE logic, so Lxx handler will be
> executed. In Lxx handler, Notify the ACPI handler of the PCIe port
> 4) In the PM handler of the PCIe port (pci_acpi_wake_dev), the PCIe port
> will be resumed, so that the PCIe port and the devices are powered on to
> D0uninitialized state.  So the devices under the PCIe port should be
> resumed too (with pci_wakeup_bus).
> 
> Without sync_first logic introduced here, all the devices under the PCIe
> port are resumed asynchronously, and during the resume process of the
> PCIe port, a suspending request is pending for the PCIe port too.  So it
> is possible for the PCIe port is resumed, suspended, then resumed again.
> 
> With sync_fist logic, one of the devices under the PCIe port are resumed
> synchronously, so that the PCIe port is guaranteed to be kept in ACTIVE
> state.
> 
> If you think this situation is so seldom that we need not care too much,
> I will remove the sync_first logic at least for now.

Yes, please.  I think we can add it later if need be.

Thanks,
Rafael