Depending on the platform and/or plane type we can get away
with a bit less alignment in the VT-d w/a. Reduce the numbers
accordingly.
Note that it's not actually clear in VLV/CHV need this w/a,
and if they do we don't actually know what kind of alignment
is sufficient. Leave the 256k alignment in place for now, but
toss in a FIXME.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250122151755.6928-3-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
if (intel_plane_can_async_flip(plane, fb->modifier))
return 256 * 1024;
+ /* FIXME undocumented so not sure what's actually needed */
if (intel_scanout_needs_vtd_wa(i915))
return 256 * 1024;
struct drm_i915_private *i915 = to_i915(plane->base.dev);
if (intel_scanout_needs_vtd_wa(i915))
- return 256 * 1024;
+ return 64 * 1024;
return 4 * 1024; /* physical for i915/i945 */
}
struct drm_i915_private *i915 = to_i915(plane->base.dev);
if (intel_scanout_needs_vtd_wa(i915))
- return 256 * 1024;
+ return 128 * 1024;
return 4 * 1024;
}