[PATCH 06/10] compositor-drm: Handle more than two output crtcs per card

[Mario Kleiner]

On 06/29/2015 10:20 PM, Bill Spitzak wrote:
> On Fri, Jun 26, 2015 at 11:34 AM, Derek Foreman <derekf at osg.samsung.com
> <mailto:derekf at osg.samsung.com>> wrote:
>
>
>      > +static unsigned int drm_waitvblank_pipe(struct drm_output *output)
>      > +{
>      > +     if (output->pipe > 1)
>      > +             return (output->pipe << DRM_VBLANK_HIGH_CRTC_SHIFT) &
>      > +                             DRM_VBLANK_HIGH_CRTC_MASK;
>
>
> This does not seem right as the individual bits in the pipe number
> cannot really change things in any useful way. From the header file I
> see this:
>
> +    /* bits 1-6 are reserved for high crtcs */
> +    DRM_VBLANK_HIGH_CRTDC_MASK = 0x0000003e,
> +#define DRM_VBLANK_HIGH_CRTC_SHIFT 1
>
> My best guess is that the intended code was something like:
>
>    (1 << output->pipe - 2 + DRM_VBLANK_HIGH_CRTC_SHIFT) & DRM_VBLANK_HIGH_CRTC_MASK
>
> Or is output->pipe always a power of 2?
>

The bits are not a bit mask where a specific bit position corresponds to 
a specific pipe, but those bits encode a 5 bits wide pipe number between 
0 and (2^5)-1 ie. 0-31. The comment in the code snippet you show about 
bits 1-6 is a bit inaccurate, as it is only bits 1-5, looking at the 
DRM_VBLANK_HIGH_CRTC_MASK.

You could address output->pipe 0 and 1 also with that code. The reason 
for the special case of pipe 0 and 1 is backwards compatibility with 
older kernels which didn't support passing in a pipe number, but instead 
had "no special flags" for pipe 0 and the DRM_VBLANK_SECONDARY flag for 
pipe 1. A leftover from the days when there were only single/dual-head 
graphis cards.

But apart from that this is just code transplanted from the x-org video 
drivers which i tested at some point both on NVidia quad-head (nouveau) 
and AMD six-head gpu's (radeon).

-mario