Re: [Y2038] [PATCH] drm/sti: Use 64-bit timestamps

[Vincent ABRIOU]

On 04/17/2016 01:39 AM, Arnd Bergmann wrote:
> On Wednesday 13 April 2016 02:28:02 Tina Ruchandani wrote:
>> 'struct timespec' uses a 32-bit field for seconds, which
>> will overflow in year 2038 and beyond. This patch is part
>> of a larger attempt to remove instances of timeval, timespec
>> and time_t, all of which suffer from the y2038 issue, from the
>> kernel.
>>
>> Signed-off-by: Tina Ruchandani <ruchandani.tina@xxxxxxxxx>
>
> Looks good in principle. Two small points:
>
>>   void sti_plane_update_fps(struct sti_plane *plane,
>>   			  bool new_frame,
>>   			  bool new_field)
>>   {
>> -	struct timespec now;
>> +	ktime_t now;
>>   	struct sti_fps_info *fps;
>>   	int fpks, fipks, ms_since_last, num_frames, num_fields;
>>
>> -	getrawmonotonic(&now);
>> +	now = ktime_get();
>
> It's unclear why the driver was using getrawmonotonic() here rather
> than ktime_get_ts(). The code is fairly new, so Vincent can
> probably explain this.
>
> If it was intentional, we should use ktime_get_raw() instead of
> ktime_get().
>

getrawmonotonic comes from a legacy code so the use is not intentional.
Honestly, it is not clear to me the difference between monotonic and 
rawmonotonic. But in the debug context in which it is used, ktime_get 
and ktime_get_raw will deliver the same level of information we need. So 
implementation done by Tina is fine for me.

Vincent

>> @@ -76,7 +66,7 @@ void sti_plane_update_fps(struct sti_plane *plane,
>>   		return;
>>
>>   	fps->curr_frame_counter++;
>> -	ms_since_last = sti_plane_timespec_ms_diff(now, fps->last_timestamp);
>> +	ms_since_last = ktime_to_ms(ktime_sub(now, fps->last_timestamp));
>>   	num_frames = fps->curr_frame_counter - fps->last_frame_counter;
>
> This could be expressed in a more compact way using ktime_ms_delta().
>
> 	Arnd
>