]> git.openfabrics.org - ~emulex/infiniband.git/commitdiff
ACPI: Allow ACPI handles of devices to be initialized in advance
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 20 Nov 2012 23:21:39 +0000 (00:21 +0100)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 20 Nov 2012 23:21:39 +0000 (00:21 +0100)
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@intel.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
drivers/acpi/glue.c

index 2f3849aedc97d150271623b0439e21ebe600a2c2..3e75d6e5a4697c949c249aa5cc1f67d17aeece41 100644 (file)
@@ -130,46 +130,59 @@ static int acpi_bind_one(struct device *dev, acpi_handle handle)
 {
        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 *dev, acpi_handle handle)
        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 device *dev)
        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 device *dev)
        }
        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 };