Re: [Intel-gfx] [PATCH 2/5] drm/i915/fbc: Split plane stride checks per-platform

From: "Ville Syrjälä" <ville.syrjala@linux.intel.com>
To: "Govindapillai, Vinod" <vinod.govindapillai@intel.com>
Cc: "intel-gfx@lists.freedesktop.org" <intel-gfx@lists.freedesktop.org>
Subject: Re: [Intel-gfx] [PATCH 2/5] drm/i915/fbc: Split plane stride checks per-platform
Date: Mon, 2 Oct 2023 10:02:41 +0300	[thread overview]
Message-ID: <ZRprEaR3ldipEySa@intel.com> (raw)
In-Reply-To: <9eb7236c1b81fe534032e94bbb43b20dd2a7e879.camel@intel.com>

On Sun, Oct 01, 2023 at 10:53:37AM +0000, Govindapillai, Vinod wrote:
> Hi Ville,
> 
> 
> On Thu, 2023-09-14 at 14:38 +0300, Ville Syrjala wrote:
> > From: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > 
> > Carve up stride_is_valid() into per-platform variants to
> > make it easier to see what limits are actually being imposed.
> > 
> > TODO: maybe go for vfuncs later
> > 
> > Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > ---
> >  drivers/gpu/drm/i915/display/intel_fbc.c | 68 ++++++++++++++++++------
> >  1 file changed, 51 insertions(+), 17 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/display/intel_fbc.c b/drivers/gpu/drm/i915/display/intel_fbc.c
> > index 1b3358a0fbfb..4c4626c84666 100644
> > --- a/drivers/gpu/drm/i915/display/intel_fbc.c
> > +++ b/drivers/gpu/drm/i915/display/intel_fbc.c
> > @@ -848,6 +848,47 @@ void intel_fbc_cleanup(struct drm_i915_private *i915)
> >         }
> >  }
> >  
> > +static bool i8xx_fbc_stride_is_valid(const struct intel_plane_state *plane_state)
> > +{
> > +       const struct drm_framebuffer *fb = plane_state->hw.fb;
> > +       unsigned int stride = intel_fbc_plane_stride(plane_state) *
> > +               fb->format->cpp[0];
> > +
> > +       return stride == 4096 || stride == 8192;
> > +}
> > +
> > +static bool i965_fbc_stride_is_valid(const struct intel_plane_state *plane_state)
> > +{
> > +       const struct drm_framebuffer *fb = plane_state->hw.fb;
> > +       unsigned int stride = intel_fbc_plane_stride(plane_state) *
> > +               fb->format->cpp[0];
> > +
> > +       return stride >= 2048 && stride <= 16384;
> > +}
> > +
> > +static bool g4x_fbc_stride_is_valid(const struct intel_plane_state *plane_state)
> > +{
> > +       return true;
> > +}
> > +
> > +static bool skl_fbc_stride_is_valid(const struct intel_plane_state *plane_state)
> > +{
> > +       const struct drm_framebuffer *fb = plane_state->hw.fb;
> > +       unsigned int stride = intel_fbc_plane_stride(plane_state) *
> > +               fb->format->cpp[0];
> > +
> > +       /* Display WA #1105: skl,bxt,kbl,cfl,glk */
> > +       if (fb->modifier == DRM_FORMAT_MOD_LINEAR && stride & 511)
> > +               return false;
> > +
> > +       return true;
> > +}
> > +
> > +static bool icl_fbc_stride_is_valid(const struct intel_plane_state *plane_state)
> > +{
> > +       return true;
> > +}
> > +
> >  static bool stride_is_valid(const struct intel_plane_state *plane_state)
> >  {
> >         struct drm_i915_private *i915 = to_i915(plane_state->uapi.plane->dev);
> > @@ -859,23 +900,16 @@ static bool stride_is_valid(const struct intel_plane_state *plane_state)
> >         if (drm_WARN_ON_ONCE(&i915->drm, (stride & (64 - 1)) != 0))
> >                 return false;
> >  
> > -       /* Below are the additional FBC restrictions. */
> > -       if (stride < 512)
> > -               return false;
> Is this check not required anymore for ICL+ and G4x?

Pre-skl FBC only supports X-tile which is a multiple of 512 bytes
anyway, so the check is redundant there.

And skl+ can support smaller strides with modifiers that have
smaller tile width (minus the linear stride w/a on skl/bxt/glk).

Perhaps removing this check should be a separate patch...
and we could remove the "multiple of 64 bytes" check too
since that is always true on any platform/modifier.

> 
> > -
> > -       if (DISPLAY_VER(i915) == 2 || DISPLAY_VER(i915) == 3)
> > -               return stride == 4096 || stride == 8192;
> > -
> > -       if (DISPLAY_VER(i915) == 4 && !IS_G4X(i915) &&
> > -           (stride < 2048 || stride > 16384))
> > -               return false;
> > -
> > -       /* Display WA #1105: skl,bxt,kbl,cfl,glk */
> > -       if ((DISPLAY_VER(i915) == 9 || IS_GEMINILAKE(i915)) &&
> > -           fb->modifier == DRM_FORMAT_MOD_LINEAR && stride & 511)
> > -               return false;
> > -
> > -       return true;
> > +       if (DISPLAY_VER(i915) >= 11)
> > +               return icl_fbc_stride_is_valid(plane_state);
> > +       else if (DISPLAY_VER(i915) >= 9)
> > +               return skl_fbc_stride_is_valid(plane_state);
> > +       else if (DISPLAY_VER(i915) >= 5 || IS_G4X(i915))
> > +               return g4x_fbc_stride_is_valid(plane_state);
> > +       else if (DISPLAY_VER(i915) == 4)
> > +               return i965_fbc_stride_is_valid(plane_state);
> > +       else
> > +               return i8xx_fbc_stride_is_valid(plane_state);
> Also I guess we  could pass "stride" as parameter to these functions for clarity and simplify.

We need more than the stride there.

> 
> There as some IGT CI failures related to bad_stride tests.

Yeah, I need to nuke that subtest.

> 
> BR
> Vinod
> >  }
> >  
> >  static bool pixel_format_is_valid(const struct intel_plane_state *plane_state)
> 

-- 
Ville Syrjälä
Intel