[Intel-gfx] [PATCH] drm/i915/tv: avoid possible division by zero

[Su Hui]

On 2023/7/18 19:28, Andrzej Hajda wrote:
> On 18.07.2023 12:10, Su Hui wrote:
>> On 2023/7/18 13:39, Dan Carpenter wrote:
>>> On Mon, Jul 17, 2023 at 04:52:51PM +0200, Andrzej Hajda wrote:
>>>> On 17.07.2023 08:22, Su Hui wrote:
>>>>> Clang warning: drivers/gpu/drm/i915/display/intel_tv.c:
>>>>> line 991, column 22 Division by zero.
>>>>> Assuming tv_mode->oversample=1 and (!tv_mode->progressive)=1,
>>>>> then division by zero will happen.
>>>>>
>>>>> Fixes: 1bba5543e4fe ("drm/i915: Fix TV encoder clock computation")
>>>>> Signed-off-by: Su Hui <suhui at nfschina.com>
>>>>> ---
>>>>>    drivers/gpu/drm/i915/display/intel_tv.c | 3 ++-
>>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/i915/display/intel_tv.c 
>>>>> b/drivers/gpu/drm/i915/display/intel_tv.c
>>>>> index 36b479b46b60..82b54af51f23 100644
>>>>> --- a/drivers/gpu/drm/i915/display/intel_tv.c
>>>>> +++ b/drivers/gpu/drm/i915/display/intel_tv.c
>>>>> @@ -988,7 +988,8 @@ intel_tv_mode_to_mode(struct drm_display_mode 
>>>>> *mode,
>>>>>                  const struct tv_mode *tv_mode,
>>>>>                  int clock)
>>>>>    {
>>>>> -    mode->clock = clock / (tv_mode->oversample >> 
>>>>> !tv_mode->progressive);
>>>>> +    mode->clock = clock / (tv_mode->oversample != 1 ?
>>>>> +            tv_mode->oversample >> !tv_mode->progressive : 1);
>>>> Seems too smart to me, why not just:
>>>> mode->clock = clock / tv_mode->oversample;
>>>> if (!tv_mode->progressive)
>>>>      mode->clock <<= 1;
>>> This is nice.
>>
>> mode->clock = clock / tv_mode->oversample << !tv_mode->progressive;
>>
>> But I think this one is much better,  it has less code and run faster.
>> Should I resend v3 to add some explanation or follow Dan's advice?
>
> Speed gain here is irrelevant here, and disputable.
>
> One thing which could be problematic is that we could loose the least 
> significant bit in mode->clock,
> in case non-progressive, but I am not sure if it really matters, as 
> mode->clock is not precise value anyway.
> Alternatively we could 1st shift, then divide, but in this case 
> overflow can occur, at least in theory - I suspect there are no such 
> big clocks (in kHz).
>
> Finally I would agree with Dan, readability is better with conditional.
>
How about this one?

-       mode->clock = clock / (tv_mode->oversample >> !tv_mode->progressive);
+       mode->clock = clock;
+       if (tv_mode->oversample >> !tv_mode->progressive)
+               mode->clock /= tv_mode->oversample >> 1;

Prevent loss of accuracy and also make it more readable.
If it's OK, I will send v3 patch.

By the way, do we need to print some error messages or do some things  when
"tv_mode->oversample << !tv_mode->progressive" is zero? I'm not sure about
this.

Su Hui

> Regards
> Andrzej