DRM accel and debugfs/sysfs

[Maíra Canal]

On 2/8/23 15:13, Oded Gabbay wrote:
> On Wed, Feb 8, 2023 at 8:07 PM Daniel Vetter <daniel at ffwll.ch> wrote:
>>
>> On Tue, Feb 07, 2023 at 01:17:47PM -0300, Maíra Canal wrote:
>>> On 2/7/23 12:43, Jeffrey Hugo wrote:
>>>> On 2/7/2023 4:31 AM, Maíra Canal wrote:
>>>>> Hi Stanislaw,
>>>>>
>>>>> On 2/1/23 12:20, Stanislaw Gruszka wrote:
>>>>>> Hi
>>>>>>
>>>>>> I was about to send debugfs support for ivpu and noticed that there
>>>>>> are current changes that deprecate drm_devel->debugfs_init callback.
>>>>>>
>>>>>> Further I looked at this commit [1], that stated we should not
>>>>>> use drm_minor for debugfs and sysfs. What is quite contrary to
>>>>>> what drm accel framework did in the first place.
>>>>>>
>>>>>> So my question is how we should use debugfs/sysfs in accel?
>>>>>> Use it with old fashioned minor-centric way or change
>>>>>> the framework somehow ?
>>>>>
>>>>> As we are trying to replace drm_debugfs_create_files() [1], it would
>>>>> be nice to see the accel debugfs support use the new debugfs API. This
>>>>> would mean using the debugfs_list from the drm_device, deprecating
>>>>> the debugfs_init callback, and adding the a similar code snippet to
>>>>> accel_debugfs_init:
>>>>>
>>>>> list_for_each_entry_safe(entry, tmp, &dev->debugfs_list, list) {
>>>>>       debugfs_create_file(entry->file.name, 0444,
>>>>>                   minor->debugfs_root, entry, &drm_debugfs_entry_fops);
>>>>>       list_del(&entry->list);
>>>>>
>>>>> Maybe Daniel has some more thoughts on this matter, but I guess it
>>>>> would be better to drop the use of the old-fashioned minor-centric
>>>>> implementation in accel.
>>
>> It was a simple case of two things landing in parallel and not being
>> synchronized. Would be good if accel could be adapted to use the new
>> debugfs infra, now that both accel and the new debugfs stuff have landed.
>> -Daniel
> Yes, definitely.
> Does anyone volunteer to send a patch to align ?
> If not, we will do it internally and send a patch.

Christian sent today a patchset addressing some problems in the debugfs
API [1]. As he is planning to remove the debugfs_list, I guess it would
be better to wait a while to align accel with the debugfs API,
considering that it currently changing.

Also, it would be nice to get feedback from the accel side.

[1] https://lore.kernel.org/dri-devel/613b9aec-7105-ca2d-13cd-16ddd85a6fda@igalia.com/T/

Best Regards,
- Maíra Canal

> 
> Oded
>>
>>>>>
>>>>> [1] https://cgit.freedesktop.org/drm/drm-misc/tree/Documentation/gpu/todo.rst#n511
>>>>>
>>>>> Best Regards,
>>>>> - Maíra Canal
>>>>
>>>> Thank you for the details Maira.  It helps to explain what the todo is suggesting.  Is there an example of a driver/drm_device that uses debugfs_list which you can easily point to?
>>>
>>> The implementation of this device-centered infrastructure is linked in [1]
>>> and an example of the conversion of debugfs APIs is linked in [2], and other
>>> drivers such as v3d, vkms, vc4 and gud use this new API as well.
>>>
>>> [1] https://cgit.freedesktop.org/drm/drm-misc/commit/?id=1c9cacbea880513a896aee65a5c58007bcb55653
>>> [2] https://cgit.freedesktop.org/drm/drm-misc/commit/?id=2e3ab8a6994f265bbd4dbd00448b84548f18464c
>>>
>>> Best Regards,
>>> - Maíra Canal
>>>
>>>>
>>>> -Jeff
>>>>
>>>>>
>>>>>>
>>>>>> [1] https://cgit.freedesktop.org/drm/drm-misc/commit/?id=99845faae7099cd704ebf67514c1157c26960a26
>>>>>>
>>>>>> Regards
>>>>>> Stanislaw
>>>>>>
>>>>
>>
>> --
>> Daniel Vetter
>> Software Engineer, Intel Corporation
>> http://blog.ffwll.ch