[PATCH v7 00/11] VP9 codec V4L2 control interface

[Andrzej Pietrasiewicz]

Hi Hans,

W dniu 17.11.2021 o 11:51, Andrzej Pietrasiewicz pisze:
> Hi again,
> 
> W dniu 17.11.2021 o 11:49, Andrzej Pietrasiewicz pisze:
>> Hi,
>>
>> W dniu 17.11.2021 o 10:59, Hans Verkuil pisze:
>>> On 16/11/2021 14:14, Andrzej Pietrasiewicz wrote:
>>>> Hi,
>>>>
>>>> W dniu 16.11.2021 o 09:21, Hans Verkuil pisze:
>>>>> On 16/11/2021 09:09, Andrzej Pietrasiewicz wrote:
>>>>>> Hi Hans,
>>>>>>
>>>>>> W dniu 15.11.2021 o 22:16, Hans Verkuil pisze:
>>>>>>> On 15/11/2021 18:14, Andrzej Pietrasiewicz wrote:
>>>>>>>> Hi Hans,
>>>>>>>>
>>>>>>>> W dniu 15.11.2021 o 16:07, Hans Verkuil pisze:
>>>>>>>>> Andrzej,
>>>>>>>>>
>>>>>>>>> Can you rebase this series on top of the master branch of
>>>>>>>>> https://git.linuxtv.org/media_stage.git/ ? Unfortunately this v7 no longer
>>>>>>>>> applies. Specifically "rkvdec: Add the VP9 backend" failed in a 
>>>>>>>>> non-trivial
>>>>>>>>> manner.
>>>>>>>>
>>>>>>>> This is a branch for you:
>>>>>>>>
>>>>>>>> https://gitlab.collabora.com/linux/for-upstream/-/tree/vp9-uapi
>>>>>>>
>>>>>>> I'm getting a bunch of sparse/smatch warnings:
>>>>>>>
>>>>>>
>>>>>> Thanks for finding this, I will re-create the branch and let you know on irc.
>>>>>> Some of the below are "false positives, namely:
>>>>>>
>>>>>> drivers/media/platform/omap3isp/omap3isp.h
>>>>>> drivers/media/platform/qcom/venus/core.h
>>>>>
>>>>> Ah, sorry, I though I had filtered those out. Obviously you can ignore those.
>>>>>
>>>>> Please post a v8. That way the series is archived on lore. And it works better
>>>>> with patchwork.
>>>>
>>>> Sure, no problem. Also please see below.
>>>>
>>>>>
>>>>> Regards,
>>>>>
>>>>>     Hans
>>>>>
>>>>>>
>>>>>> which are not touched by the series.
>>>>>>
>>>>>> Regards,
>>>>>>
>>>>>> Andrzej
>>>>>>
>>>>>>> sparse:
>>>>>>> rkvdec/rkvdec-vp9.c:190:43: warning: variable 'dec_params' set but not 
>>>>>>> used [-Wunused-but-set-variable]
>>>>>>> rkvdec/rkvdec-vp9.c:245:43: warning: variable 'dec_params' set but not 
>>>>>>> used [-Wunused-but-set-variable]
>>>>>>> SPARSE:hantro/hantro_postproc.c hantro/hantro_postproc.c:37:35: warning: 
>>>>>>> symbol 'hantro_g1_postproc_regs' was not declared. Should it be static?
>>>>>>>
>>>>>>> smatch:
>>>>>>> rkvdec/rkvdec-vp9.c:190:43: warning: variable 'dec_params' set but not 
>>>>>>> used [-Wunused-but-set-variable]
>>>>>>> rkvdec/rkvdec-vp9.c:245:43: warning: variable 'dec_params' set but not 
>>>>>>> used [-Wunused-but-set-variable]
>>>>>>> rkvdec/rkvdec-vp9.c: rkvdec/rkvdec-vp9.c:236 init_intra_only_probs() 
>>>>>>> error: buffer overflow 'ptr' 90 <= 91
>>>>
>>>> this looks a false positive.
>>>>
>>>> A portion of memory pointed to by ptr is indexed with i * 23 + m,
>>>> where i ranges from 0 to 3, inclusive, and m ranges from 0 to 22,
>>>> inclusive if i < 3, otherwise m ranges from 0 to 20, inclusive.
>>>> So the largest index value we compute equals 89 (3 * 23 + 20).
>>>> Because ptr points to something that is at least 90 bytes large,
>>>> 89 is a valid index and no greater index will be ever computed.
>>>
>>> But we do need to get rid of this smatch warning, otherwise it will pollute the
>>> list of smatch warnings.
>>>
>>> I was looking at the code and wonder if it wouldn't make more sense to
>>> move writing to rkprobs->intra_mode[i].uv_mode[] into a separate for loop:
>>>
>>>          for (i = 0; i < ARRAY_SIZE(v4l2_vp9_kf_uv_mode_prob); i++)
>>>                  rkprobs->intra_mode[i / 23].uv_mode[i % 23] = 
>>> v4l2_vp9_kf_uv_mode_prob[i];
>>>
>>> Wouldn't that do the same as the current code? It looks simpler as well.
>>>
>>
>> I think it would, but I would slightly change the loop:
>>
>>      for (i = 0; i < ARRAY_SIZE(v4l2_vp9_kf_uv_mode_prob); i++) {
> 
> actually, sizeof(v4l2_vp9_kf_uv_mode_prob)
> 
> 
> 
>>          const u8 *ptr = (const u8 *)v4l2_vp9_kf_uv_mode_prob;
>>
>>          rkprobs->intra_mode[i / 23].uv_mode[i % 23] = ptr[i];
>>      }
>>
>> because v4l2_vp9_kf_uv_mode_prob is actually a u8[10][9].
>>
>> I will make such a change locally and test whether it causes regressions.

This worked, no regressions:

	for (i = 0; i < sizeof(v4l2_vp9_kf_uv_mode_prob); ++i) {
		const u8 *ptr = (const u8 *)v4l2_vp9_kf_uv_mode_prob;

		rkprobs->intra_mode[i / 23].uv_mode[i % 23] = ptr[i];
	}

Andrzej