Re: [PATCH v4] platform/x86: alienware-wmi-base: Transition to new WMI API

[Kurt Borja]

On Tue Apr 28, 2026 at 2:27 PM -05, Armin Wolf wrote:
> Am 28.04.26 um 18:37 schrieb Kurt Borja:
>
>> Transition to the new wmi_buffer based WMI API.
>>
>> Signed-off-by: Kurt Borja <kuurtb@xxxxxxxxx>
>> ---
>> v4:
>>    - Rebase to use the new wmidev_invoke_method() with size check
>>
>> v3: https://patch.msgid.link/20260331-aw-new-api-v3-1-ef03b94529d8@xxxxxxxxx
>>    - Use __free() instead of manual cleanup.
>>    - Include <linux/types.h>
>>
>> v2: https://patch.msgid.link/20260331-aw-new-api-v2-1-3b0d33bf8d22@xxxxxxxxx
>>    - Cast wmi_buffer data to __le32 and then use le32_to_cpu() before
>>      returning
>>
>> v1: https://patch.msgid.link/20260330-aw-new-api-v1-1-95910bfa1b38@xxxxxxxxx
>> ---
>>   drivers/platform/x86/dell/alienware-wmi-base.c | 31 ++++++++++++++++----------
>>   1 file changed, 19 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/platform/x86/dell/alienware-wmi-base.c b/drivers/platform/x86/dell/alienware-wmi-base.c
>> index 64562b92314f..21d3acfbd73d 100644
>> --- a/drivers/platform/x86/dell/alienware-wmi-base.c
>> +++ b/drivers/platform/x86/dell/alienware-wmi-base.c
>> @@ -12,8 +12,13 @@
>>   #include <linux/cleanup.h>
>>   #include <linux/module.h>
>>   #include <linux/platform_device.h>
>> +#include <linux/slab.h>
>> +#include <linux/types.h>
>>   #include <linux/dmi.h>
>>   #include <linux/leds.h>
>> +
>> +#include <asm/byteorder.h>
>> +
>>   #include "alienware-wmi.h"
>>   
>>   MODULE_AUTHOR("Mario Limonciello <mario.limonciello@xxxxxxxxxxx>");
>> @@ -150,21 +155,23 @@ u8 alienware_interface;
>>   int alienware_wmi_command(struct wmi_device *wdev, u32 method_id,
>>   			  void *in_args, size_t in_size, u32 *out_data)
>>   {
>> -	struct acpi_buffer out = {ACPI_ALLOCATE_BUFFER, NULL};
>> -	struct acpi_buffer in = {in_size, in_args};
>> -	acpi_status ret;
>> +	struct wmi_buffer out, in = {
>> +		.data = in_args,
>> +		.length = in_size,
>> +	};
>> +	int ret;
>>   
>> -	ret = wmidev_evaluate_method(wdev, 0, method_id, &in, out_data ? &out : NULL);
>> -	if (ACPI_FAILURE(ret))
>> -		return -EIO;
>> -
>> -	union acpi_object *obj __free(kfree) = out.pointer;
>> +	if (out_data)
>> +		ret = wmidev_invoke_method(wdev, 0, method_id, &in, &out,
>> +					   sizeof(*out_data));
>> +	else
>> +		ret = wmidev_invoke_method(wdev, 0, method_id, &in, NULL, 0);
>> +	if (ret)
>> +		return ret;
>>   
>>   	if (out_data) {
>> -		if (obj && obj->type == ACPI_TYPE_INTEGER)
>> -			*out_data = (u32)obj->integer.value;
>> -		else
>> -			return -ENOMSG;
>> +		__le32 *data __free(kfree) = out.data;
>> +		*out_data = le32_to_cpu(*data);
>>   	}
>
> Hi,
>
> calling wmidev_invoke_method() with a NULL output buffer will likely lead to crashes. This happens
> because wmidev_invoke_method() demands that a valid pointer to a output buffer is provided by the
> caller. I suggest that you use wmidev_invoke_procedure() instead:
>
> 	__le32 *data __free(kfree) = NULL;
>
> 	if (!out_data)
> 		return wmidev_invoke_procedure(wdev, 0, method_id, &in);
>
> 	ret = wmidev_invoke_method(wdev, 0, method_id, &in, &out, sizeof(*data));
> 	if (ret < 0)
> 		return ret;
>
> 	result = out.data
> 	...

Hi Armin,

You're right, I thought the only difference between _procedure and
_method was the size check.

Also, Sashiko [1] reported that out.data might be ZERO_SIZE_PTR even if
wmidev_invoke_method() succeeds. I thought it was a false positive but I
looked further and it was actually a bug. I submitted a fix for this.

[1] https://sashiko.dev/#/patchset/20260428-aw-new-api-v4-1-d2c06241d7ad%40gmail.com

-- 
Thanks,
 ~ Kurt