Re: [RFC 1/3] acpi-video: Add an acpi_video_unregister_backlight function

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 05/13/2014 05:11 PM, Aaron Lu wrote:
> On 05/13/2014 02:03 AM, Hans de Goede wrote:
>> Add an acpi_video_unregister_backlight function, which only unregisters
>> the backlight device, and leaves the acpi_notifier in place. Some acpi_vendor
>> driver need this as they don't want the acpi_video# backlight device, but do
>> need the acpi-video driver for hotkey handling.
>>
>> Chances are that this new acpi_video_unregister_backlight() is actually
>> what existing acpi_vendor drivers have wanted all along. Currently acpi_vendor
>> drivers which want to disable the acpi_video# backlight device, make 2 calls:
>>
>> acpi_video_dmi_promote_vendor();
>> acpi_video_unregister();
>>
>> The intention here is to make things independent of when acpi_video_register()
>> gets called. As acpi_video_register() will get called on acpi-video load time
>> on non intel gfx machines, while it gets called on i915 load time on intel
>> gfx machines.
>>
>> This leads to the following 2 interesting scenarios:
>>
>> a) intel gfx:
>> 1) acpi-video module gets loaded (as it is a dependency of acpi_vendor and i915)
>> 2) acpi-video does NOT call acpi_video_register()
>> 3) acpi_vendor loads (lets assume it loads before i915), calls
>> acpi_video_dmi_promote_vendor(); which sets ACPI_VIDEO_BACKLIGHT_DMI_VENDOR
>> 4) calls acpi_video_unregister -> not registered, nop
>> 5) i915 loads, calls acpi_video_register
>> 6) acpi_video_register registers the acpi_notifier for the hotkeys,
>>    does NOT register a backlight device because of ACPI_VIDEO_BACKLIGHT_DMI_VENDOR
>>
>> b) non intel gfx
>> 1) acpi-video module gets loaded (as it is a dependency acpi_vendor)
>> 2) acpi-video calls acpi_video_register()
>> 3) acpi_video_register registers the acpi_notifier for the hotkeys,
>>    and a backlight device
>> 4) acpi_vendor loads, calls acpi_video_dmi_promote_vendor()
>> 5) calls acpi_video_unregister, this unregisters BOTH the acpi_notifier for
>>    the hotkeys AND the backlight device
>>
>> So here we have possibly the same acpi_vendor module, making the same calls,
>> but with different results, in one cases acpi-video does handle hotkeys,
>> in the other it does not.
>>
>> Note that the a) scenario turns into b) if we assume the i915 module loads
>> before the vendor_acpi module, so we also have different behavior depending
>> on module loading order!
>>
>> So as said I believe that quite a few existing acpi_vendor modules really
>> always want the behavior of a), which calling acpi_video_unregister_backlight()
>> instead of acpi_video_unregister() will give them.
>>
>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>> ---
>>  drivers/acpi/video.c | 31 ++++++++++++++++++++++---------
>>  include/acpi/video.h |  2 ++
>>  2 files changed, 24 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/acpi/video.c b/drivers/acpi/video.c
>> index 6a2099d..80a6759 100644
>> --- a/drivers/acpi/video.c
>> +++ b/drivers/acpi/video.c
>> @@ -1843,7 +1843,7 @@ static void acpi_video_dev_register_backlight(struct acpi_video_device *device)
>>  	}
>>  }
>>  
>> -static int acpi_video_bus_register_backlight(struct acpi_video_bus *video)
>> +static void acpi_video_bus_register_backlight(struct acpi_video_bus *video)
>>  {
>>  	struct acpi_video_device *dev;
>>  
>> @@ -1851,10 +1851,6 @@ static int acpi_video_bus_register_backlight(struct acpi_video_bus *video)
>>  	list_for_each_entry(dev, &video->video_device_list, entry)
>>  		acpi_video_dev_register_backlight(dev);
>>  	mutex_unlock(&video->device_list_lock);
>> -
>> -	video->pm_nb.notifier_call = acpi_video_resume;
>> -	video->pm_nb.priority = 0;
>> -	return register_pm_notifier(&video->pm_nb);
> 
> Hmm, I don't understand this. The pm notifier is used to restore
> backlight level on system resume, so should be there when we have
> created the backlight interface and gone when we unregistered the
> backlight interface. It doesn't have anything to do with hotkey
> processing.

A valid point, I did things this way because I wanted the new
acpi_video_unregister_backlight to result in the same behavior as
having set the acpi_video=vendor / called acpi_video_dmi_promote_vendor()
before acpi_video_register runs.

So this means undoing what-ever acpi_video_register() does, which it
would not have done if acpi_video_dmi_promote_vendor() came first.

If you look at the current code (so without this patch) then
the pm notifier gets installed unconditionally by
acpi_video_bus_register_backlight() which gets called unconditionally
by acpi_video_bus_add(). So even with acpi_video=vendor it still gets
installed.

acpi_video=vendor / acpi_video_dmi_promote_vendor() influences
acpi_video_backlight_support() which only gets called by
acpi_video_verify_backlight_support() which only gets called by
acpi_video_dev_register_backlight(). So acpi_video=vendor only causes
the backlight devices to not register, the pm notifier will still be
installed. My intend was to behave the same independent of module
loading / init order hence I moved the pm notifier install / uninstall
to keep the pm notifier installed when calling the new
acpi_video_unregister_backlight().

Looking at the code you're right that this is not really sensible, since
the acpi_video_resume call done by the notifier will be a nop when there
are no backlight devices registered.

So I can split this patch into 2 patches, 1 to not install the pm notifier
if we're not going to register backlight devices anyways, and a 2nd patch
adding the new acpi_video_unregister_backlight().

Does that sound like a plan ?

Regards,

Hans




> 
> Thanks,
> Aaron
> 
>>  }
>>  
>>  static void acpi_video_dev_unregister_backlight(struct acpi_video_device *device)
>> @@ -1876,17 +1872,14 @@ static void acpi_video_dev_unregister_backlight(struct acpi_video_device *device
>>  	}
>>  }
>>  
>> -static int acpi_video_bus_unregister_backlight(struct acpi_video_bus *video)
>> +static void acpi_video_bus_unregister_backlight(struct acpi_video_bus *video)
>>  {
>>  	struct acpi_video_device *dev;
>> -	int error = unregister_pm_notifier(&video->pm_nb);
>>  
>>  	mutex_lock(&video->device_list_lock);
>>  	list_for_each_entry(dev, &video->video_device_list, entry)
>>  		acpi_video_dev_unregister_backlight(dev);
>>  	mutex_unlock(&video->device_list_lock);
>> -
>> -	return error;
>>  }
>>  
>>  static void acpi_video_dev_add_notify_handler(struct acpi_video_device *device)
>> @@ -2059,6 +2052,11 @@ static int acpi_video_bus_add(struct acpi_device *device)
>>  	mutex_unlock(&video_list_lock);
>>  
>>  	acpi_video_bus_register_backlight(video);
>> +
>> +	video->pm_nb.notifier_call = acpi_video_resume;
>> +	video->pm_nb.priority = 0;
>> +	register_pm_notifier(&video->pm_nb);
>> +
>>  	acpi_video_bus_add_notify_handler(video);
>>  
>>  	return 0;
>> @@ -2084,6 +2082,7 @@ static int acpi_video_bus_remove(struct acpi_device *device)
>>  	video = acpi_driver_data(device);
>>  
>>  	acpi_video_bus_remove_notify_handler(video);
>> +	unregister_pm_notifier(&video->pm_nb);
>>  	acpi_video_bus_unregister_backlight(video);
>>  	acpi_video_bus_put_devices(video);
>>  
>> @@ -2173,6 +2172,20 @@ void acpi_video_unregister(void)
>>  }
>>  EXPORT_SYMBOL(acpi_video_unregister);
>>  
>> +void acpi_video_unregister_backlight(void)
>> +{
>> +	struct acpi_video_bus *video;
>> +
>> +	if (!register_count)
>> +		return;
>> +
>> +	mutex_lock(&video_list_lock);
>> +	list_for_each_entry(video, &video_bus_head, entry)
>> +		acpi_video_bus_unregister_backlight(video);
>> +	mutex_unlock(&video_list_lock);
>> +}
>> +EXPORT_SYMBOL(acpi_video_unregister_backlight);
>> +
>>  /*
>>   * This is kind of nasty. Hardware using Intel chipsets may require
>>   * the video opregion code to be run first in order to initialise
>> diff --git a/include/acpi/video.h b/include/acpi/video.h
>> index 61109f2..4722c06 100644
>> --- a/include/acpi/video.h
>> +++ b/include/acpi/video.h
>> @@ -19,11 +19,13 @@ struct acpi_device;
>>  #if (defined CONFIG_ACPI_VIDEO || defined CONFIG_ACPI_VIDEO_MODULE)
>>  extern int acpi_video_register(void);
>>  extern void acpi_video_unregister(void);
>> +extern void acpi_video_unregister_backlight(void);
>>  extern int acpi_video_get_edid(struct acpi_device *device, int type,
>>  			       int device_id, void **edid);
>>  #else
>>  static inline int acpi_video_register(void) { return 0; }
>>  static inline void acpi_video_unregister(void) { return; }
>> +extern inline void acpi_video_unregister_backlight(void); { return; }
>>  static inline int acpi_video_get_edid(struct acpi_device *device, int type,
>>  				      int device_id, void **edid)
>>  {
>>
>