platform/x86: wmi: Always evaluate _WED when receiving an event
authorArmin Wolf <W_Armin@gmx.de>
Mon, 19 Feb 2024 11:59:17 +0000 (12:59 +0100)
committerIlpo Järvinen <ilpo.jarvinen@linux.intel.com>
Tue, 27 Feb 2024 12:44:25 +0000 (14:44 +0200)
The ACPI WMI specification states:

"The _WED control method is evaluated by the mapper in
 response to receiving a notification from a control
 method."

This means that _WED should be evaluated unconditionally even
if no WMI event consumers are present.
Some firmware implementations actually depend on this behavior
by storing the event data inside a queue which will fill up if
the WMI core stops retrieving event data items due to no
consumers being present

Fix this by always evaluating _WED even if no WMI event consumers
are present.

Signed-off-by: Armin Wolf <W_Armin@gmx.de>
Link: https://lore.kernel.org/r/20240219115919.16526-4-W_Armin@gmx.de
Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
drivers/platform/x86/wmi.c

index ff4742c40cc3996fd4e952ee1e933e599d58092e..abd0183c4107848edb637d17edc4757784f4fa68 100644 (file)
@@ -1206,37 +1206,46 @@ acpi_wmi_ec_space_handler(u32 function, acpi_physical_address address,
        }
 }
 
-static void wmi_notify_driver(struct wmi_block *wblock)
+static int wmi_get_notify_data(struct wmi_block *wblock, union acpi_object **obj)
 {
-       struct wmi_driver *driver = drv_to_wdrv(wblock->dev.dev.driver);
        struct acpi_buffer data = { ACPI_ALLOCATE_BUFFER, NULL };
-       union acpi_object *obj = NULL;
        acpi_status status;
 
-       if (!driver->no_notify_data) {
-               status = get_event_data(wblock, &data);
-               if (ACPI_FAILURE(status)) {
-                       dev_warn(&wblock->dev.dev, "Failed to get event data\n");
-                       return;
-               }
+       if (test_bit(WMI_NO_EVENT_DATA, &wblock->flags)) {
+               *obj = NULL;
+               return 0;
+       }
 
-               obj = data.pointer;
-               if (!obj) {
-                       dev_warn(&wblock->dev.dev, "Event contains no event data\n");
-                       return;
-               }
+       status = get_event_data(wblock, &data);
+       if (ACPI_FAILURE(status)) {
+               dev_warn(&wblock->dev.dev, "Failed to get event data\n");
+               return -EIO;
+       }
+
+       *obj = data.pointer;
+
+       return 0;
+}
+
+static void wmi_notify_driver(struct wmi_block *wblock, union acpi_object *obj)
+{
+       struct wmi_driver *driver = drv_to_wdrv(wblock->dev.dev.driver);
+
+       if (!obj && !driver->no_notify_data) {
+               dev_warn(&wblock->dev.dev, "Event contains no event data\n");
+               return;
        }
 
        if (driver->notify)
                driver->notify(&wblock->dev, obj);
-
-       kfree(obj);
 }
 
 static int wmi_notify_device(struct device *dev, void *data)
 {
        struct wmi_block *wblock = dev_to_wblock(dev);
+       union acpi_object *obj;
        u32 *event = data;
+       int ret;
 
        if (!(wblock->gblock.flags & ACPI_WMI_EVENT && wblock->gblock.notify_id == *event))
                return 0;
@@ -1246,10 +1255,32 @@ static int wmi_notify_device(struct device *dev, void *data)
         * Because of this the WMI driver notify handler takes precedence.
         */
        if (wblock->dev.dev.driver && wblock->driver_ready) {
-               wmi_notify_driver(wblock);
+               ret = wmi_get_notify_data(wblock, &obj);
+               if (ret >= 0) {
+                       wmi_notify_driver(wblock, obj);
+                       kfree(obj);
+               }
        } else {
-               if (wblock->handler)
+               if (wblock->handler) {
                        wblock->handler(*event, wblock->handler_data);
+               } else {
+                       /* The ACPI WMI specification says that _WED should be
+                        * evaluated every time an notification is received, even
+                        * if no consumers are present.
+                        *
+                        * Some firmware implementations actually depend on this
+                        * by using a queue for events which will fill up if the
+                        * WMI driver core stops evaluating _WED due to missing
+                        * WMI event consumers.
+                        *
+                        * Because of this we need this seemingly useless call to
+                        * wmi_get_notify_data() which in turn evaluates _WED.
+                        */
+                       ret = wmi_get_notify_data(wblock, &obj);
+                       if (ret >= 0)
+                               kfree(obj);
+               }
+
        }
        up_read(&wblock->notify_lock);