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

[Mika Westerberg <mika.westerberg@linux.intel.com>]

On Sun, Nov 18, 2012 at 10:12:52PM +0100, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> 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>
> ---
>  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 ;-)

Maybe using a different variable for the sanity check?

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@linux.intel.com>

to these two patches, if you like.