[Intel-gfx] [PATCH v18 18/18] drm/i915: Keep plane size mult of 4 for NV12

[Maarten Lankhorst]

Op 29-03-18 om 10:06 schreef Vidya Srinivas:
> As per display WA 1106, to avoid corruption issues
> NV12 plane height needs to be multiplier of 4
> Hence we modify the fb src and destination height
> and width to be multiples of 4. Without this, pipe
> fifo underruns were seen on APL and KBL.
>
> Credits-to: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
> Signed-off-by: Vidya Srinivas <vidya.srinivas at intel.com>
> ---
>  drivers/gpu/drm/i915/intel_drv.h    | 2 ++
>  drivers/gpu/drm/i915/intel_sprite.c | 8 ++++++++
>  2 files changed, 10 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
> index 9c58da0..a1f718d 100644
> --- a/drivers/gpu/drm/i915/intel_drv.h
> +++ b/drivers/gpu/drm/i915/intel_drv.h
> @@ -159,6 +159,8 @@
>  #define INTEL_I2C_BUS_DVO 1
>  #define INTEL_I2C_BUS_SDVO 2
>  
> +#define MULT4(x) ((x + 3) & ~0x03)
> +
>  /* these are outputs from the chip - integrated only
>     external chips are via DVO or SDVO output */
>  enum intel_output_type {
> diff --git a/drivers/gpu/drm/i915/intel_sprite.c b/drivers/gpu/drm/i915/intel_sprite.c
> index 538d938..9f466c6 100644
> --- a/drivers/gpu/drm/i915/intel_sprite.c
> +++ b/drivers/gpu/drm/i915/intel_sprite.c
> @@ -261,6 +261,14 @@ skl_update_plane(struct intel_plane *plane,
>  	crtc_w--;
>  	crtc_h--;
>  
> +	if (fb->format->format == DRM_FORMAT_NV12) {
> +		src_w = MULT4(src_w);
> +		src_h = MULT4(src_h);
> +		crtc_w = MULT4(crtc_w);
> +		crtc_h = MULT4(crtc_h);
> +		DRM_ERROR("%d %d %d %d\n", src_w, src_h, crtc_w, crtc_h);
> +	}
> +
>  	spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
>  
>  	if (INTEL_GEN(dev_priv) >= 10 || IS_GEMINILAKE(dev_priv))

Nearly there!

Do we have limitations for width too? But I think we shouldn't ever adjust src for any format.
This means that we should probably get rid of the drm_rect_adjust_size call in intel_check_sprite_plane.

If any limitations of NV12 are hit, we should reject with -EINVAL instead so userspace can decide what to do.
The best place to put those checks is probably in skl_update_scaler, where they will be checked by the primary plane too.

This will mean the tests fail, but that can be fixed by selecting 16 as width/height for NV12 in IGT. If you change it to 16 you can put my r-b on it.

Also I think we should put the same limitations for width and height being a multiple in intel_framebuffer_init.

And on a final note for patch ordering, put the workaround and gen10 patch before enabling nv12 support.

~Maarten