Re: [PATCH 42/43] drm/fbdev-generic: Convert to fbdev-ttm

[Jon Hunter <jonathanh@nvidia.com>]

On 22/10/2024 15:55, Thomas Zimmermann wrote:
> Hi
> 
> Am 22.10.24 um 15:31 schrieb Jon Hunter:
>> Hi Thomas,
>>
>> On 12/03/2024 15:45, Thomas Zimmermann wrote:
>>> Only TTM-based drivers use fbdev-generic. Rename it to fbdev-ttm and
>>> change the symbol infix from _generic_ to _ttm_. Link the source file
>>> into TTM helpers, so that it is only build if TTM-based drivers have
>>> been selected. Select DRM_TTM_HELPER for loongson.
>>>
>>> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
>>> ---
>>>   Documentation/gpu/drm-kms-helpers.rst         |  2 +-
>>>   drivers/gpu/drm/Makefile                      |  5 +-
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c       |  6 +-
>>>   .../{drm_fbdev_generic.c => drm_fbdev_ttm.c}  | 80 +++++++++----------
>>>   .../gpu/drm/hisilicon/hibmc/hibmc_drm_drv.c   |  4 +-
>>>   drivers/gpu/drm/loongson/Kconfig              |  1 +
>>>   drivers/gpu/drm/loongson/lsdc_drv.c           |  4 +-
>>>   drivers/gpu/drm/nouveau/nouveau_drm.c         |  6 +-
>>>   drivers/gpu/drm/qxl/qxl_drv.c                 |  4 +-
>>>   drivers/gpu/drm/tiny/bochs.c                  |  4 +-
>>>   drivers/gpu/drm/vboxvideo/vbox_drv.c          |  4 +-
>>>   drivers/gpu/drm/vmwgfx/vmwgfx_drv.c           |  4 +-
>>>   include/drm/drm_fbdev_generic.h               | 15 ----
>>>   include/drm/drm_fbdev_ttm.h                   | 15 ++++
>>>   14 files changed, 77 insertions(+), 77 deletions(-)
>>>   rename drivers/gpu/drm/{drm_fbdev_generic.c => drm_fbdev_ttm.c} (76%)
>>>   delete mode 100644 include/drm/drm_fbdev_generic.h
>>>   create mode 100644 include/drm/drm_fbdev_ttm.h
>>
>> ...
>>
>>> diff --git a/include/drm/drm_fbdev_ttm.h b/include/drm/drm_fbdev_ttm.h
>>> new file mode 100644
>>> index 0000000000000..9e6c3bdf35376
>>> --- /dev/null
>>> +++ b/include/drm/drm_fbdev_ttm.h
>>> @@ -0,0 +1,15 @@
>>> +/* SPDX-License-Identifier: MIT */
>>> +
>>> +#ifndef DRM_FBDEV_TTM_H
>>> +#define DRM_FBDEV_TTM_H
>>> +
>>> +struct drm_device;
>>> +
>>> +#ifdef CONFIG_DRM_FBDEV_EMULATION
>>> +void drm_fbdev_ttm_setup(struct drm_device *dev, unsigned int 
>>> preferred_bpp);
>>> +#else
>>> +static inline void drm_fbdev_ttm_setup(struct drm_device *dev, 
>>> unsigned int preferred_bpp)
>>> +{ }
>>> +#endif
>>> +
>>> +#endif
>>
>>
>> I recently noticed that with Linux v6.11 it is possible to enable
>> CONFIG_DRM_FBDEV_EMULATION without enabling CONFIG_DRM_TTM_HELPER. Now
>> while this does not currently cause any build issues, I believe that if
>> the function drm_fbdev_ttm_setup() is ever used somewhere that does not
>> explicitly select CONFIG_DRM_TTM_HELPER, then you could get a 'ERROR:
>> modpost: "drm_fbdev_ttm_setup"'.
>>
>> So I was thinking that may be this should be ...
>>
>> diff --git a/include/drm/drm_fbdev_ttm.h b/include/drm/drm_fbdev_ttm.h
>> index 9e6c3bdf3537..5b6723a37caa 100644
>> --- a/include/drm/drm_fbdev_ttm.h
>> +++ b/include/drm/drm_fbdev_ttm.h
>> @@ -5,7 +5,7 @@
>>
>>  struct drm_device;
>>
>> -#ifdef CONFIG_DRM_FBDEV_EMULATION
>> +#if defined(CONFIG_DRM_FBDEV_EMULATION) && 
>> defined(CONFIG_DRM_TTM_HELPER)
>>  void drm_fbdev_ttm_setup(struct drm_device *dev, unsigned int 
>> preferred_bpp);
>>  #else
>>  static inline void drm_fbdev_ttm_setup(struct drm_device *dev, 
>> unsigned int preferred_bpp)
>>
>>
>> The above function has been removed now in -next, but I believe we could
>> have the same problem with drm_fbdev_ttm_helper_fb_probe()?
> 
> We'd turn a linker/modpost error into a compiler error. Likely makes no 
> difference. And AFAICT every driver that selects TTM also selects 
> TTM_HELPER. Drivers without TTM should not use this header.


Yes I also noted that all the current drivers select TTM_HELPER and so 
we don't run into this. However, it still seems a bit odd that we don't 
expose a proper stub if TTM_HELPER is disabled (especially seeing as 
there is a stub defined).

Jon

-- 
nvpublic