Re: [PATCH 01/26] drm: modeset-helper: Export dev_pm_ops for simple drivers

[Paul Cercueil <paul@crapouillou.net>]

Hi Thomas,

Le mar. 8 nov. 2022 à 13:10:49 +0100, Thomas Zimmermann 
<tzimmermann@suse.de> a écrit :
> Hi Paul,
> 
> thanks for cleaning up. Please see my comments below.
> 
> Am 07.11.22 um 18:50 schrieb Paul Cercueil:
>> Export a dev_pm_ops meant to be used with simple drivers, which have
>> their struct drm_device registered as their struct device's drvdata, 
>> and
>> only call drm_mode_config_pm_{suspend,resume}.
>> 
>> The symbol is conditionally exported if IS_ENABLED(CONFIG_PM_SLEEP), 
>> and
>> therefore it should always be referenced using the pm_sleep_ptr() 
>> macro.
>> 
>> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
>> ---
>>   drivers/gpu/drm/drm_modeset_helper.c | 32 
>> ++++++++++++++++++++++++++++
>>   include/drm/drm_modeset_helper.h     |  4 ++++
>>   2 files changed, 36 insertions(+)
>> 
>> diff --git a/drivers/gpu/drm/drm_modeset_helper.c 
>> b/drivers/gpu/drm/drm_modeset_helper.c
>> index f858dfedf2cf..0bc9f9228a60 100644
>> --- a/drivers/gpu/drm/drm_modeset_helper.c
>> +++ b/drivers/gpu/drm/drm_modeset_helper.c
>> @@ -20,6 +20,9 @@
>>    * OF THIS SOFTWARE.
>>    */
>> 
>> +#include <linux/device.h>
>> +#include <linux/pm.h>
>> +
>>   #include <drm/drm_atomic_helper.h>
>>   #include <drm/drm_fb_helper.h>
>>   #include <drm/drm_fourcc.h>
>> @@ -244,3 +247,32 @@ int drm_mode_config_helper_resume(struct 
>> drm_device *dev)
>>   	return ret;
>>   }
>>   EXPORT_SYMBOL(drm_mode_config_helper_resume);
>> +
>> +static int drm_mode_config_pm_suspend(struct device *dev)
>> +{
>> +	struct drm_device *drm = dev_get_drvdata(dev);
>> +
>> +	return drm_mode_config_helper_suspend(drm);
>> +}
>> +
>> +static int drm_mode_config_pm_resume(struct device *dev)
>> +{
>> +	struct drm_device *drm = dev_get_drvdata(dev);
>> +
>> +	return drm_mode_config_helper_resume(drm);
>> +}
>> +
>> +/**
>> + * drm_mode_config_pm_ops - Exported dev_pm_ops helper for simple 
>> drivers
>> + *
>> + * This dev_pm_ops can be used for simple drivers that would 
>> otherwise only call
>> + * drm_mode_config_helper_suspend / drm_mode_config_helper_resume 
>> in their PM
>> + * callbacks. It is only valid if the driver's drm_device has been 
>> registered as
>> + * the struct device's drvdata.
>> + *
>> + * The exported symbol must always be used with the pm_sleep_ptr() 
>> macro, like
>> + * this:
>> + * .pm = pm_sleep_ptr(&drm_mode_config_pm_ops),
>> + */
>> +EXPORT_SIMPLE_DEV_PM_OPS(drm_mode_config_pm_ops,
>> +			 drm_mode_config_pm_suspend, drm_mode_config_pm_resume);
>> diff --git a/include/drm/drm_modeset_helper.h 
>> b/include/drm/drm_modeset_helper.h
>> index 995fd981cab0..85f29637e9c1 100644
>> --- a/include/drm/drm_modeset_helper.h
>> +++ b/include/drm/drm_modeset_helper.h
>> @@ -23,6 +23,8 @@
>>   #ifndef __DRM_KMS_HELPER_H__
>>   #define __DRM_KMS_HELPER_H__
> 
> I like that you clean up the driver, but not how it's done TBH.
> 
>> 
>> +#include <linux/pm.h>
>> +
> 
> Half of DRM somehow incudes drm_kms_helper.h. So this include 
> statements
> affects more or less everything.
> 
>>   struct drm_crtc;
>>   struct drm_crtc_funcs;
>>   struct drm_device;
>> @@ -41,4 +43,6 @@ int drm_crtc_init(struct drm_device *dev, struct 
>> drm_crtc *crtc,
>>   int drm_mode_config_helper_suspend(struct drm_device *dev);
>>   int drm_mode_config_helper_resume(struct drm_device *dev);
>> 
>> +extern const struct dev_pm_ops drm_mode_config_pm_ops;
>> +
> 
> That's maybe subjective, but I don't like exporting such _funcs and 
> _ops
> instances. They are like blackboxes.  And they pollute the symbol
> namespace unnecessarily.
> 
> I propose a solution similar to DEFINE_DRM_GEM_FOPS [1] or
> drm_module_pci_driver. [2]
> 
> Define a macro in the header to create the _ops instance, such as
> 
> #if defined(CONFIG_PM)
> #define DEFINE_DRM_MODE_CONFIG_HELPER_PM_OPS(_name)  \
>   static __ ## _name ## _suspend() {  \
>     call drm_mode_config_helper_suspend()  \
>   } \
>   static __ ## _name ## _resume() {  \
>     call drm_mode_config_helper_resume()  \
>   }  \
>   static SIMPLE_DEV_PM_OPS(_name, __ ## _name ## _suspend, __ ## _name
> ## _resume);
> #else
> #define DEFINE_DRM_MODE_CONFIG_HELPER_PM_OPS(_name)
> #endif
> 
> Drivers can then keep the instance and the include for pm.h internal
> within their _drv.c file. The small callback functions are within the
> source file as well.  If CONFIG_PM has been disabled, nothing is being
> generated.

Another alternative would be to make the exported 
"drm_mode_config_pm_ops" a pointer. Then it can be opaque and you 
wouldn't need the <linux/pm.h> include in the header.

Your solution works too, I guess. I find it inelegant, but I won't 
fight you hard on this.

Cheers,
-Paul

> 
> Best regards
> Thomas
> 
> [1]
> https://elixir.bootlin.com/linux/v6.1-rc4/source/include/drm/drm_gem.h#L396
> [2]
> https://elixir.bootlin.com/linux/v6.1-rc4/source/include/drm/drm_module.h#L58
> 
> 
>>   #endif
> 
> --
> Thomas Zimmermann
> Graphics Driver Developer
> SUSE Software Solutions Germany GmbH
> Maxfeldstr. 5, 90409 Nürnberg, Germany
> (HRB 36809, AG Nürnberg)
> Geschäftsführer: Ivo Totev