]> git.baikalelectronics.ru Git - kernel.git/commitdiff
drm/i915/skl: distribute DDB based on panel resolution
authorMahesh Kumar <mahesh1.kumar@intel.com>
Wed, 1 Aug 2018 15:11:13 +0000 (20:41 +0530)
committerMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Tue, 7 Aug 2018 08:25:30 +0000 (10:25 +0200)
We distribute DDB equally among all pipes irrespective of display
buffer requirement of each pipe. This leads to a situation where high
resolution y-tiled display can not be enabled with 2 low resolution
displays.

Main contributing factor for DDB requirement is width of the display.
This patch make changes to distribute ddb based on display width.
So display with higher width will get bigger chunk of DDB.

Changes Since V1:
 - pipe_size/ddb_size will not overflow u16 so use appropriate
   data-types during computation (Chris)
Changes Since V2:
 - avoid redundancy and possible truncation errors (Chris)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107113
Cc: raviraj.p.sitaram@intel.com
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mahesh Kumar <mahesh1.kumar@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180801151113.5337-1-mahesh1.kumar@intel.com
drivers/gpu/drm/i915/intel_pm.c

index 0e4b8328a971c1c36b320d3bee82b3d75b9baebb..03654f5f68c3e8552a3ebbbfc61e11cebccd95a7 100644 (file)
@@ -3814,8 +3814,12 @@ skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
        struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
        struct drm_i915_private *dev_priv = to_i915(dev);
        struct drm_crtc *for_crtc = cstate->base.crtc;
-       u16 pipe_size, ddb_size;
-       int nth_active_pipe;
+       const struct drm_crtc_state *crtc_state;
+       const struct drm_crtc *crtc;
+       u32 pipe_width = 0, total_width = 0, width_before_pipe = 0;
+       enum pipe for_pipe = to_intel_crtc(for_crtc)->pipe;
+       u16 ddb_size;
+       u32 i;
 
        if (WARN_ON(!state) || !cstate->base.active) {
                alloc->start = 0;
@@ -3833,14 +3837,14 @@ skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
                                      *num_active, ddb);
 
        /*
-        * If the state doesn't change the active CRTC's, then there's
-        * no need to recalculate; the existing pipe allocation limits
-        * should remain unchanged.  Note that we're safe from racing
-        * commits since any racing commit that changes the active CRTC
-        * list would need to grab _all_ crtc locks, including the one
-        * we currently hold.
+        * If the state doesn't change the active CRTC's or there is no
+        * modeset request, then there's no need to recalculate;
+        * the existing pipe allocation limits should remain unchanged.
+        * Note that we're safe from racing commits since any racing commit
+        * that changes the active CRTC list or do modeset would need to
+        * grab _all_ crtc locks, including the one we currently hold.
         */
-       if (!intel_state->active_pipe_changes) {
+       if (!intel_state->active_pipe_changes && !intel_state->modeset) {
                /*
                 * alloc may be cleared by clear_intel_crtc_state,
                 * copy from old state to be sure
@@ -3849,11 +3853,32 @@ skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
                return;
        }
 
-       nth_active_pipe = hweight32(intel_state->active_crtcs &
-                                   (drm_crtc_mask(for_crtc) - 1));
-       pipe_size = ddb_size / hweight32(intel_state->active_crtcs);
-       alloc->start = nth_active_pipe * ddb_size / *num_active;
-       alloc->end = alloc->start + pipe_size;
+       /*
+        * Watermark/ddb requirement highly depends upon width of the
+        * framebuffer, So instead of allocating DDB equally among pipes
+        * distribute DDB based on resolution/width of the display.
+        */
+       for_each_new_crtc_in_state(state, crtc, crtc_state, i) {
+               const struct drm_display_mode *adjusted_mode;
+               int hdisplay, vdisplay;
+               enum pipe pipe;
+
+               if (!crtc_state->enable)
+                       continue;
+
+               pipe = to_intel_crtc(crtc)->pipe;
+               adjusted_mode = &crtc_state->adjusted_mode;
+               drm_mode_get_hv_timing(adjusted_mode, &hdisplay, &vdisplay);
+               total_width += hdisplay;
+
+               if (pipe < for_pipe)
+                       width_before_pipe += hdisplay;
+               else if (pipe == for_pipe)
+                       pipe_width = hdisplay;
+       }
+
+       alloc->start = ddb_size * width_before_pipe / total_width;
+       alloc->end = ddb_size * (width_before_pipe + pipe_width) / total_width;
 }
 
 static unsigned int skl_cursor_allocation(int num_active)
@@ -5254,7 +5279,7 @@ skl_ddb_add_affected_pipes(struct drm_atomic_state *state, bool *changed)
         * any other display updates race with this transaction, so we need
         * to grab the lock on *all* CRTC's.
         */
-       if (intel_state->active_pipe_changes) {
+       if (intel_state->active_pipe_changes || intel_state->modeset) {
                realloc_pipes = ~0;
                intel_state->wm_results.dirty_pipes = ~0;
        }