Re: [PATCH 4.4 030/103] drm/amdgpu: Make display watermark calculations more accurate

From: Mario Kleiner
Date: Tue Jun 06 2017 - 13:46:21 EST


On Thu, Jun 1, 2017 at 1:13 PM, Ben Hutchings
<ben.hutchings@xxxxxxxxxxxxxxx> wrote:
> On Tue, 2017-05-23 at 22:08 +0200, Greg Kroah-Hartman wrote:
>> 4.4-stable review patch. If anyone has any objections, please let me know.
>>
>> ------------------
>>
>> From: Mario Kleiner <mario.kleiner.de@xxxxxxxxx>
>>
>> commit d63c277dc672e0c568481af043359420fa9d4736 upstream.
>>
>> Avoid big roundoff errors in scanline/hactive durations for
>> high pixel clocks, especially for >= 500 Mhz, and thereby
>> program more accurate display fifo watermarks.
> [...]
>> --- a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c
>> @@ -1237,14 +1237,14 @@ static void dce_v10_0_program_watermarks
>> {
>> struct drm_display_mode *mode = &amdgpu_crtc->base.mode;
>> struct dce10_wm_params wm_low, wm_high;
>> - u32 pixel_period;
>> + u32 active_time;
>> u32 line_time = 0;
>> u32 latency_watermark_a = 0, latency_watermark_b = 0;
>> u32 tmp, wm_mask, lb_vblank_lead_lines = 0;
>>
>> if (amdgpu_crtc->base.enabled && num_heads && mode) {
>> - pixel_period = 1000000 / (u32)mode->clock;
>> - line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
>> + active_time = 1000000UL * (u32)mode->crtc_hdisplay / (u32)mode->clock;
>> + line_time = min((u32) (1000000UL * (u32)mode->crtc_htotal / (u32)mode->clock), (u32)65535);
> [...]
>
> Won't these multiplications overflow if a >4K display is attached to a
> 32-bit machine?
>
> Ben.
>

Yes, indeed > 4k causes a new overflow problem! Thanks for catching this.
I will prepare a follow-up patch on top of this one, to use ...

active_time = (u32) div_u64((u64) mode->crtc_hdisplay * 1000000,
(u32)mode->clock);
line_time = (u32) div_u64((u64) mode->crtc_htotal * 1000000, (u32)mode->clock);
line_time = min(line_time, (u32) 65535);

...instead.

Ok?
-mario

> --
> Ben Hutchings
> Software Developer, Codethink Ltd.
>
>