[RFC][PATCH 3/7 updated] PM: Asynchronous resume of I/O devices

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Monday 17 August 2009, Rafael J. Wysocki wrote:
> Theoretically, the total time of system sleep transitions (suspend
> to RAM, hibernation) can be reduced by running suspend and resume
> callbacks of device drivers in parallel with each other.  However,
> there are dependencies between devices such that, for example, we may
> not be allowed to put one device into a low power state before
> anohter one has been suspended (e.g. we cannot suspend a bridge
> before suspending all devices behind it).  In particular, we're not
> allowed to suspend the parent of a device before suspending the
> device itself.  Analogously, we're not allowed to resume a device
> before resuming its parent.
> 
> Thus, to make it possible to execute suspend and resume callbacks
> provided by device drivers in parallel with each other, we need to
> provide a synchronization mechanism preventing the dependencies
> between devices from being violated.
> 
> The patch below introduces a mechanism allowing some devices to be
> resumed asynchronously, using completions with the following rules:
> (1) There is a completion, dev->power.comp, for each device object.
> (2) All of these completions are reset before suspend as well as
>     each resume stage (dpm_resume_noirq(), dpm_resume()).
> (3) If dev->power.async_suspend is set for dev or for one of devices
>     it depends on, the PM core waits for the "master" device's
>     completion before attempting to run the resume callbacks,
>     appropriate for this particular stage of resume, for dev.
> (4) dev->power.comp is completed for each device after running its
>     resume callbacks.

I have an update here as well.  The patch below doesn't use completions
any more, but it uses the wait_queue that we already have in struct device for
runtime PM.  It also tries to do more things in parallel.

Thanks,
Rafael

---
Theoretically, the total time of system sleep transitions (suspend
to RAM, hibernation) can be reduced by running suspend and resume
callbacks of device drivers in parallel with each other.  However,
there are dependencies between devices such that, for example, we may
not be allowed to put one device into a low power state before
anohter one has been suspended (e.g. we cannot suspend a bridge
before suspending all devices behind it).  In particular, we're not
allowed to suspend the parent of a device before suspending the
device itself.  Analogously, we're not allowed to resume a device
before resuming its parent.

Thus, to make it possible to execute suspend and resume callbacks
provided by device drivers in parallel with each other, we need to
provide a synchronization mechanism preventing the dependencies
between devices from being violated.

The patch below introduces a mechanism allowing some devices to be
resumed asynchronously, with the following rules:
(1) There is a wait queue head, dev->power.wait_queue, and an
    "operation complete" flag, dev->power.op_complete for each device
    object.
(2) All of the power.op_complete flags are reset before suspend as
    well as after each resume stage (dpm_resume_noirq(),
    dpm_resume()).
(3) If power.async_suspend is set for dev or for one of devices it
    depends on, the PM core waits for the "master" device's
    power.op_complete flag to be set before attempting to run the
    resume callbacks, appropriate for this particular stage of
    resume, for dev.
(4) dev->power.op_complete is set for each device after running its
    resume callbacks at each stage of resume (dpm_resume_noirq(),
    dpm_resume()).

With this mechanism in place, the drivers wanting their resume
callbacks to be executed asynchronously can set
dev->power.async_suspend for them, with the help of
device_enable_async_suspend().  In addition to that, the PM off-tree
dependencies between devices have to be represented by
'struct pm_link' objects introduced by the previous patch.

In this version of the patch the async threads started to execute
the resume callbacks of specific device don't exit immediately having
done that, but search dpm_list for devices whose PM dependencies have
already been satisfied and execute their callbacks without waiting.

---
 drivers/base/power/common.c |    5 
 drivers/base/power/main.c   |  226 +++++++++++++++++++++++++++++++++++++++++---
 include/linux/device.h      |    6 +
 include/linux/pm.h          |    6 -
 4 files changed, 228 insertions(+), 15 deletions(-)

Index: linux-2.6/include/linux/pm.h
===================================================================
--- linux-2.6.orig/include/linux/pm.h
+++ linux-2.6/include/linux/pm.h
@@ -26,6 +26,7 @@
 #include <linux/spinlock.h>
 #include <linux/wait.h>
 #include <linux/timer.h>
+#include <linux/completion.h>
 
 /*
  * Callbacks for platform drivers to implement.
@@ -409,20 +410,23 @@ enum rpm_request {
 
 struct dev_pm_info {
 	spinlock_t		lock;
+	wait_queue_head_t	wait_queue;
 	struct list_head	master_links;
 	struct list_head	slave_links;
 	pm_message_t		power_state;
 	unsigned int		can_wakeup:1;
 	unsigned int		should_wakeup:1;
+	unsigned int		async_suspend:1;
 	enum dpm_state		status;		/* Owned by the PM core */
 #ifdef CONFIG_PM_SLEEP
 	struct list_head	entry;
+	unsigned int		op_started:1;
+	unsigned int		op_complete:1;
 #endif
 #ifdef CONFIG_PM_RUNTIME
 	struct timer_list	suspend_timer;
 	unsigned long		timer_expires;
 	struct work_struct	work;
-	wait_queue_head_t	wait_queue;
 	atomic_t		usage_count;
 	atomic_t		child_count;
 	unsigned int		disable_depth:3;
Index: linux-2.6/include/linux/device.h
===================================================================
--- linux-2.6.orig/include/linux/device.h
+++ linux-2.6/include/linux/device.h
@@ -472,6 +472,12 @@ static inline int device_is_registered(s
 	return dev->kobj.state_in_sysfs;
 }
 
+static inline void device_enable_async_suspend(struct device *dev, bool enable)
+{
+	if (dev->power.status == DPM_ON)
+		dev->power.async_suspend = enable;
+}
+
 void driver_init(void);
 
 /*
Index: linux-2.6/drivers/base/power/main.c
===================================================================
--- linux-2.6.orig/drivers/base/power/main.c
+++ linux-2.6/drivers/base/power/main.c
@@ -26,6 +26,8 @@
 #include <linux/resume-trace.h>
 #include <linux/rwsem.h>
 #include <linux/interrupt.h>
+#include <linux/async.h>
+#include <linux/completion.h>
 
 #include "../base.h"
 #include "power.h"
@@ -43,6 +45,7 @@
 LIST_HEAD(dpm_list);
 
 static DEFINE_MUTEX(dpm_list_mtx);
+static pm_message_t pm_transition;
 
 /*
  * Set once the preparation of devices for a PM transition has started, reset
@@ -144,6 +147,81 @@ void device_pm_move_last(struct device *
 	list_move_tail(&dev->power.entry, &dpm_list);
 }
 
+static void dpm_reset(struct device *dev)
+{
+	dev->power.op_started = false;
+	dev->power.op_complete = false;
+}
+
+static void dpm_reset_all(void)
+{
+	struct device *dev;
+
+	list_for_each_entry(dev, &dpm_list, power.entry)
+		dpm_reset(dev);
+}
+
+static void dpm_synchronize_noirq(void)
+{
+	async_synchronize_full();
+	dpm_reset_all();
+}
+
+static void dpm_synchronize(void)
+{
+	async_synchronize_full();
+	mutex_lock(&dpm_list_mtx);
+	dpm_reset_all();
+	mutex_unlock(&dpm_list_mtx);
+}
+
+static void device_pm_wait(struct device *sub, struct device *dev)
+{
+	if (!dev)
+		return;
+
+	if (!(sub->power.async_suspend || dev->power.async_suspend))
+		return;
+
+	if (!dev->power.op_complete) {
+		dev_dbg(sub, "PM: Waiting for %s %s\n", dev_driver_string(dev),
+			dev_name(dev));
+		wait_event(dev->power.wait_queue, !!dev->power.op_complete);
+	}
+}
+
+static int device_pm_wait_fn(struct device *dev, void *data)
+{
+	device_pm_wait((struct device *)data, dev);
+	return 0;
+}
+
+static void device_pm_wait_for_masters(struct device *slave)
+{
+	if (!pm_trace_enabled)
+		device_for_each_master(slave, slave, device_pm_wait_fn);
+}
+
+static bool device_pm_check(struct device *dev)
+{
+	int ret = 0;
+
+	if (dev)
+		ret = !dev->power.op_complete;
+
+	return ret;
+}
+
+static int device_pm_check_fn(struct device *dev, void *data)
+{
+	return device_pm_check(dev);
+}
+
+static int device_pm_check_masters(struct device *slave)
+{
+	return device_for_each_master(slave, NULL, device_pm_check_fn);
+}
+
 /**
  * pm_op - Execute the PM operation appropriate for given PM event.
  * @dev: Device to handle.
@@ -269,6 +347,20 @@ static int pm_noirq_op(struct device *de
 	return error;
 }
 
+static bool pm_op_started(struct device *dev)
+{
+	bool ret = false;
+
+	spin_lock_irq(&dev->power.lock);
+	if (dev->power.op_started)
+		ret = true;
+	else
+		dev->power.op_started = true;
+	spin_unlock_irq(&dev->power.lock);
+
+	return ret;
+}
+
 static char *pm_verb(int event)
 {
 	switch (event) {
@@ -310,32 +402,78 @@ static void pm_dev_err(struct device *de
 /*------------------------- Resume routines -------------------------*/
 
 /**
- * device_resume_noirq - Execute an "early resume" callback for given device.
+ * __device_resume_noirq - Execute an "early resume" callback for given device.
  * @dev: Device to handle.
  * @state: PM transition of the system being carried out.
  *
  * The driver of @dev will not receive interrupts while this function is being
  * executed.
  */
-static int device_resume_noirq(struct device *dev, pm_message_t state)
+static int __device_resume_noirq(struct device *dev, pm_message_t state)
 {
 	int error = 0;
 
 	TRACE_DEVICE(dev);
 	TRACE_RESUME(0);
 
-	if (!dev->bus)
-		goto End;
-
-	if (dev->bus->pm) {
+	if (dev->bus && dev->bus->pm) {
 		pm_dev_dbg(dev, state, "EARLY ");
 		error = pm_noirq_op(dev, dev->bus->pm, state);
 	}
- End:
+
+	dev->power.op_complete = true;
+	wake_up_all(&dev->power.wait_queue);
+
 	TRACE_RESUME(error);
 	return error;
 }
 
+static void async_device_resume_noirq(struct device *dev)
+{
+	int error;
+
+	pm_dev_dbg(dev, pm_transition, "async EARLY ");
+	error = __device_resume_noirq(dev, pm_transition);
+	if (error)
+		pm_dev_err(dev, pm_transition, " async EARLY", error);
+}
+
+static void async_resume_noirq(void *data, async_cookie_t cookie)
+{
+	struct device *dev = (struct device *)data;
+
+	device_pm_wait_for_masters(dev);
+	async_device_resume_noirq(dev);
+
+	list_for_each_entry_continue(dev, &dpm_list, power.entry) {
+		if (!dev->power.async_suspend || dev->power.status <= DPM_OFF)
+			continue;
+
+		if (device_pm_check_masters(dev))
+			continue;
+
+		if (pm_op_started(dev))
+			continue;
+
+		pm_dev_dbg(dev, pm_transition, "out of order EARLY ");
+		async_device_resume_noirq(dev);
+	}
+}
+
+static int device_resume_noirq(struct device *dev)
+{
+	if (pm_op_started(dev))
+		return 0;
+
+	if (dev->power.async_suspend && !pm_trace_enabled) {
+		async_schedule(async_resume_noirq, dev);
+		return 0;
+	}
+
+	device_pm_wait_for_masters(dev);
+	return __device_resume_noirq(dev, pm_transition);
+}
+
 /**
  * dpm_resume_noirq - Execute "early resume" callbacks for non-sysdev devices.
  * @state: PM transition of the system being carried out.
@@ -349,26 +487,28 @@ void dpm_resume_noirq(pm_message_t state
 
 	mutex_lock(&dpm_list_mtx);
 	transition_started = false;
+	pm_transition = state;
 	list_for_each_entry(dev, &dpm_list, power.entry)
 		if (dev->power.status > DPM_OFF) {
 			int error;
 
 			dev->power.status = DPM_OFF;
-			error = device_resume_noirq(dev, state);
+			error = device_resume_noirq(dev);
 			if (error)
-				pm_dev_err(dev, state, " early", error);
+				pm_dev_err(dev, state, " EARLY", error);
 		}
+	dpm_synchronize_noirq();
 	mutex_unlock(&dpm_list_mtx);
 	resume_device_irqs();
 }
 EXPORT_SYMBOL_GPL(dpm_resume_noirq);
 
 /**
- * device_resume - Execute "resume" callbacks for given device.
+ * __device_resume - Execute "resume" callbacks for given device.
  * @dev: Device to handle.
  * @state: PM transition of the system being carried out.
  */
-static int device_resume(struct device *dev, pm_message_t state)
+static int __device_resume(struct device *dev, pm_message_t state)
 {
 	int error = 0;
 
@@ -409,11 +549,67 @@ static int device_resume(struct device *
 	}
  End:
 	up(&dev->sem);
+	dev->power.op_complete = true;
+	wake_up_all(&dev->power.wait_queue);
 
 	TRACE_RESUME(error);
 	return error;
 }
 
+static void async_device_resume(struct device *dev)
+{
+	int error;
+
+	pm_dev_dbg(dev, pm_transition, "async ");
+	error = __device_resume(dev, pm_transition);
+	if (error)
+		pm_dev_err(dev, pm_transition, " async", error);
+}
+
+static void async_resume(void *data, async_cookie_t cookie)
+{
+	struct device *dev = (struct device *)data;
+
+	device_pm_wait_for_masters(dev);
+
+ repeat:
+	async_device_resume(dev);
+	put_device(dev);
+
+	mutex_lock(&dpm_list_mtx);
+	list_for_each_entry(dev, &dpm_list, power.entry) {
+		if (!dev->power.async_suspend || dev->power.status < DPM_OFF)
+			continue;
+
+		if (device_pm_check_masters(dev))
+			continue;
+
+		if (pm_op_started(dev))
+			continue;
+
+		get_device(dev);
+		mutex_unlock(&dpm_list_mtx);
+		pm_dev_dbg(dev, pm_transition, "out of order ");
+		goto repeat;
+	}
+	mutex_unlock(&dpm_list_mtx);
+}
+
+static int device_resume(struct device *dev)
+{
+	if (pm_op_started(dev))
+		return 0;
+
+	if (dev->power.async_suspend && !pm_trace_enabled) {
+		get_device(dev);
+		async_schedule(async_resume, dev);
+		return 0;
+	}
+
+	device_pm_wait_for_masters(dev);
+	return __device_resume(dev, pm_transition);
+}
+
 /**
  * dpm_resume - Execute "resume" callbacks for non-sysdev devices.
  * @state: PM transition of the system being carried out.
@@ -427,6 +623,7 @@ static void dpm_resume(pm_message_t stat
 
 	INIT_LIST_HEAD(&list);
 	mutex_lock(&dpm_list_mtx);
+	pm_transition = state;
 	while (!list_empty(&dpm_list)) {
 		struct device *dev = to_device(dpm_list.next);
 
@@ -437,7 +634,7 @@ static void dpm_resume(pm_message_t stat
 			dev->power.status = DPM_RESUMING;
 			mutex_unlock(&dpm_list_mtx);
 
-			error = device_resume(dev, state);
+			error = device_resume(dev);
 
 			mutex_lock(&dpm_list_mtx);
 			if (error)
@@ -452,6 +649,7 @@ static void dpm_resume(pm_message_t stat
 	}
 	list_splice(&list, &dpm_list);
 	mutex_unlock(&dpm_list_mtx);
+	dpm_synchronize();
 }
 
 /**
@@ -775,8 +973,10 @@ static int dpm_prepare(pm_message_t stat
 			break;
 		}
 		dev->power.status = DPM_SUSPENDING;
-		if (!list_empty(&dev->power.entry))
+		if (!list_empty(&dev->power.entry)) {
 			list_move_tail(&dev->power.entry, &list);
+			dpm_reset(dev);
+		}
 		put_device(dev);
 	}
 	list_splice(&list, &dpm_list);
Index: linux-2.6/drivers/base/power/common.c
===================================================================
--- linux-2.6.orig/drivers/base/power/common.c
+++ linux-2.6/drivers/base/power/common.c
@@ -19,10 +19,11 @@
  */
 void device_pm_init(struct device *dev)
 {
-	dev->power.status = DPM_ON;
 	spin_lock_init(&dev->power.lock);
+	init_waitqueue_head(&dev->power.wait_queue);
 	INIT_LIST_HEAD(&dev->power.master_links);
 	INIT_LIST_HEAD(&dev->power.slave_links);
+	dev->power.status = DPM_ON;
 	pm_runtime_init(dev);
 }
 
@@ -81,6 +82,8 @@ int pm_link_add(struct device *slave, st
 	return 0;
 
  err_link:
+	master->power.async_suspend = false;
+	slave->power.async_suspend = false;
 	error = -ENOMEM;
 	put_device(slave);
 
_______________________________________________
linux-pm mailing list
linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/linux-pm

[Index of Archives]     [Linux ACPI]     [Netdev]     [Ethernet Bridging]     [Linux Wireless]     [CPU Freq]     [Kernel Newbies]     [Fedora Kernel]     [Security]     [Linux for Hams]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux RAID]     [Linux Admin]     [Samba]

  Powered by Linux