[Update][PATCH 1/2] ACPI: Allow ACPI handles of devices to be initialized in advance

[Rafael J. Wysocki]

On Monday, November 19, 2012 11:42:34 AM Mika Westerberg wrote:
> On Sun, Nov 18, 2012 at 10:12:52PM +0100, Rafael J. Wysocki wrote:
> > From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > 
> > Currently, the ACPI handles of devices are initialized from within
> > device_add(), by acpi_bind_one() called from acpi_platform_notify()
> > which first uses the .find_device() routine provided by the device's
> > bus type to find the matching device node in the ACPI namespace.
> > This is a source of some computational overhead and, moreover, the
> > correctness of the result depends on the implementation of
> > .find_device() which is known to fail occasionally for some bus types
> > (e.g. PCI).  In some cases, however, the corresponding ACPI device
> > node is known already before calling device_add() for the given
> > struct device object and the whole .find_device() dance in
> > acpi_platform_notify() is then simply unnecessary.
> > 
> > For this reason, make it possible to initialize the ACPI handles of
> > devices before calling device_add() for them.  Modify
> > acpi_platform_notify() to call acpi_bind_one() in advance to check
> > the device's existing ACPI handle and skip the .find_device()
> > search if that is successful.  Change acpi_bind_one() accordingly.
> > 
> > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > ---
> >  drivers/acpi/glue.c |   42 +++++++++++++++++++++++++++++++++---------
> >  1 file changed, 33 insertions(+), 9 deletions(-)
> > 
> > Index: linux/drivers/acpi/glue.c
> > ===================================================================
> > --- linux.orig/drivers/acpi/glue.c
> > +++ linux/drivers/acpi/glue.c
> > @@ -135,41 +135,54 @@ static int acpi_bind_one(struct device *
> >  	int retval = -EINVAL;
> >  
> >  	if (dev->acpi_handle) {
> > -		dev_warn(dev, "Drivers changed 'acpi_handle'\n");
> > -		return -EINVAL;
> > +		if (handle) {
> > +			dev_warn(dev, "ACPI handle is already set\n");
> > +			return -EINVAL;
> > +		} else {
> > +			handle = dev->acpi_handle;
> > +		}
> >  	}
> > +	if (!handle)
> > +		return -EINVAL;
> >  
> >  	get_device(dev);
> >  	status = acpi_bus_get_device(handle, &acpi_dev);
> >  	if (ACPI_FAILURE(status))
> >  		goto err;
> >  
> > -	physical_node = kzalloc(sizeof(struct acpi_device_physical_node),
> > -		GFP_KERNEL);
> > +	physical_node = kzalloc(sizeof(*physical_node), GFP_KERNEL);
> 
> Here we allocate memory for the physical node...
> 
> >  	if (!physical_node) {
> >  		retval = -ENOMEM;
> >  		goto err;
> >  	}
> >  
> >  	mutex_lock(&acpi_dev->physical_node_lock);
> > +
> > +	/* Sanity check. */
> > +	list_for_each_entry(physical_node, &acpi_dev->physical_node_list, node)
> 
> .. and overwrite it here ;-)

Ah, good catch!

> Maybe using a different variable for the sanity check?

Yeah, I wanted to be overly smart. :-)

> I've changed the SPI/I2C patches to use this as well and they got a lot
> smaller as we don't have to do the .find_device() magic.
> 
> Once you have fixed the above, you can add my
> 
> Reviewed-by: Mika Westerberg <mika.westerberg@xxxxxxxxxxxxxxx>
> 
> to these two patches, if you like.

I will.

In the meantime, updated patch is appended.

Thanks,
Rafael


---
From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
Subject: ACPI: Allow ACPI handles of devices to be initialized in advance

Currently, the ACPI handles of devices are initialized from within
device_add(), by acpi_bind_one() called from acpi_platform_notify()
which first uses the .find_device() routine provided by the device's
bus type to find the matching device node in the ACPI namespace.
This is a source of some computational overhead and, moreover, the
correctness of the result depends on the implementation of
.find_device() which is known to fail occasionally for some bus types
(e.g. PCI).  In some cases, however, the corresponding ACPI device
node is known already before calling device_add() for the given
struct device object and the whole .find_device() dance in
acpi_platform_notify() is then simply unnecessary.

For this reason, make it possible to initialize the ACPI handles of
devices before calling device_add() for them.  Modify
acpi_platform_notify() to call acpi_bind_one() in advance to check
the device's existing ACPI handle and skip the .find_device()
search if that is successful.  Change acpi_bind_one() accordingly.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
---
 drivers/acpi/glue.c |   44 ++++++++++++++++++++++++++++++++++----------
 1 file changed, 34 insertions(+), 10 deletions(-)

Index: linux/drivers/acpi/glue.c
===================================================================
--- linux.orig/drivers/acpi/glue.c
+++ linux/drivers/acpi/glue.c
@@ -130,46 +130,59 @@ static int acpi_bind_one(struct device *
 {
 	struct acpi_device *acpi_dev;
 	acpi_status status;
-	struct acpi_device_physical_node *physical_node;
+	struct acpi_device_physical_node *physical_node, *pn;
 	char physical_node_name[sizeof(PHYSICAL_NODE_STRING) + 2];
 	int retval = -EINVAL;
 
 	if (dev->acpi_handle) {
-		dev_warn(dev, "Drivers changed 'acpi_handle'\n");
-		return -EINVAL;
+		if (handle) {
+			dev_warn(dev, "ACPI handle is already set\n");
+			return -EINVAL;
+		} else {
+			handle = dev->acpi_handle;
+		}
 	}
+	if (!handle)
+		return -EINVAL;
 
 	get_device(dev);
 	status = acpi_bus_get_device(handle, &acpi_dev);
 	if (ACPI_FAILURE(status))
 		goto err;
 
-	physical_node = kzalloc(sizeof(struct acpi_device_physical_node),
-		GFP_KERNEL);
+	physical_node = kzalloc(sizeof(*physical_node), GFP_KERNEL);
 	if (!physical_node) {
 		retval = -ENOMEM;
 		goto err;
 	}
 
 	mutex_lock(&acpi_dev->physical_node_lock);
+
+	/* Sanity check. */
+	list_for_each_entry(pn, &acpi_dev->physical_node_list, node)
+		if (pn->dev == dev) {
+			dev_warn(dev, "Already associated with ACPI node\n");
+			goto err_free;
+		}
+
 	/* allocate physical node id according to physical_node_id_bitmap */
 	physical_node->node_id =
 		find_first_zero_bit(acpi_dev->physical_node_id_bitmap,
 		ACPI_MAX_PHYSICAL_NODE);
 	if (physical_node->node_id >= ACPI_MAX_PHYSICAL_NODE) {
 		retval = -ENOSPC;
-		mutex_unlock(&acpi_dev->physical_node_lock);
-		kfree(physical_node);
-		goto err;
+		goto err_free;
 	}
 
 	set_bit(physical_node->node_id, acpi_dev->physical_node_id_bitmap);
 	physical_node->dev = dev;
 	list_add_tail(&physical_node->node, &acpi_dev->physical_node_list);
 	acpi_dev->physical_node_count++;
+
 	mutex_unlock(&acpi_dev->physical_node_lock);
 
-	dev->acpi_handle = handle;
+	if (!dev->acpi_handle)
+		dev->acpi_handle = handle;
 
 	if (!physical_node->node_id)
 		strcpy(physical_node_name, PHYSICAL_NODE_STRING);
@@ -187,8 +200,14 @@ static int acpi_bind_one(struct device *
 	return 0;
 
  err:
+	dev->acpi_handle = NULL;
 	put_device(dev);
 	return retval;
+
+ err_free:
+	mutex_unlock(&acpi_dev->physical_node_lock);
+	kfree(physical_node);
+	goto err;
 }
 
 static int acpi_unbind_one(struct device *dev)
@@ -247,6 +266,10 @@ static int acpi_platform_notify(struct d
 	acpi_handle handle;
 	int ret = -EINVAL;
 
+	ret = acpi_bind_one(dev, NULL);
+	if (!ret)
+		goto out;
+
 	if (!dev->bus || !dev->parent) {
 		/* bridge devices genernally haven't bus or parent */
 		ret = acpi_find_bridge_device(dev, &handle);
@@ -260,10 +283,11 @@ static int acpi_platform_notify(struct d
 	}
 	if ((ret = type->find_device(dev, &handle)) != 0)
 		DBG("Can't get handler for %s\n", dev_name(dev));
-      end:
+ end:
 	if (!ret)
 		acpi_bind_one(dev, handle);
 
+ out:
 #if ACPI_GLUE_DEBUG
 	if (!ret) {
 		struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/