drm/komeda: Add layer split support
authorjames qian wang (Arm Technology China) <james.qian.wang@arm.com>
Mon, 10 Jun 2019 10:16:08 +0000 (11:16 +0100)
committerLiviu Dudau <Liviu.Dudau@arm.com>
Wed, 19 Jun 2019 10:42:17 +0000 (11:42 +0100)
Komeda supports two types of layer split:
- none-scaling split
- scaling split
Since D71 merger only support scaler as input, so for none-scaling split,
the two layer dflow will be output to compiz directly. for scaling_split,
the data flow will be merged by merger firstly, then output the merged
data flow to compiz.

Komeda handles the split in kernel completely to hide the detailed and
complicated split calcualtion to user mode, for user only need to set the
layer_split property to enable/disable it.

v2: Rebase

Signed-off-by: James Qian Wang (Arm Technology China) <james.qian.wang@arm.com>
Signed-off-by: Liviu Dudau <liviu.dudau@arm.com>
drivers/gpu/drm/arm/display/komeda/komeda_kms.c
drivers/gpu/drm/arm/display/komeda/komeda_kms.h
drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c
drivers/gpu/drm/arm/display/komeda/komeda_pipeline.h
drivers/gpu/drm/arm/display/komeda/komeda_pipeline_state.c
drivers/gpu/drm/arm/display/komeda/komeda_plane.c

index 0ec76656cd1f2de1a92b65872200d8adebe210f6..5d10c55c9c40e7bf4166f81e35570e63934af6ce 100644 (file)
@@ -173,6 +173,14 @@ static int komeda_crtc_normalize_zpos(struct drm_crtc *crtc,
                plane = plane_st->plane;
 
                plane_st->normalized_zpos = order++;
+               /* When layer_split has been enabled, one plane will be handled
+                * by two separated komeda layers (left/right), which may needs
+                * two zorders.
+                * - zorder: for left_layer for left display part.
+                * - zorder + 1: will be reserved for right layer.
+                */
+               if (to_kplane_st(plane_st)->layer_split)
+                       order++;
 
                DRM_DEBUG_ATOMIC("[PLANE:%d:%s] zpos:%d, normalized zpos: %d\n",
                                 plane->base.id, plane->name,
index 5f71e669d92bca698b5b09c5c480eeb8b59c8aa8..9dcfe5a01e23114ec29de4c94794cb58410e8aaf 100644 (file)
@@ -36,6 +36,8 @@ struct komeda_plane {
 
        /** @prop_img_enhancement: for on/off image enhancement */
        struct drm_property *prop_img_enhancement;
+       /** @prop_layer_split: for on/off layer_split */
+       struct drm_property *prop_layer_split;
 };
 
 /**
@@ -50,8 +52,11 @@ struct komeda_plane_state {
        /** @zlist_node: zorder list node */
        struct list_head zlist_node;
 
-       /* @img_enhancement: on/off image enhancement */
-       u8 img_enhancement : 1;
+       /* @img_enhancement: on/off image enhancement
+        * @layer_split: on/off layer_split
+        */
+       u8 img_enhancement : 1,
+          layer_split : 1;
 };
 
 /**
@@ -155,6 +160,19 @@ is_only_changed_connector(struct drm_crtc_state *st, struct drm_connector *conn)
        return BIT(drm_connector_index(conn)) == changed_connectors;
 }
 
+static inline bool has_flip_h(u32 rot)
+{
+       u32 rotation = drm_rotation_simplify(rot,
+                                            DRM_MODE_ROTATE_0 |
+                                            DRM_MODE_ROTATE_90 |
+                                            DRM_MODE_REFLECT_MASK);
+
+       if (rotation & DRM_MODE_ROTATE_90)
+               return !!(rotation & DRM_MODE_REFLECT_Y);
+       else
+               return !!(rotation & DRM_MODE_REFLECT_X);
+}
+
 unsigned long komeda_calc_aclk(struct komeda_crtc_state *kcrtc_st);
 
 int komeda_kms_setup_crtcs(struct komeda_kms_dev *kms, struct komeda_dev *mdev);
index 543ecc80703fd147dec3c4036251448083fb005f..eb9e0c0af8f3b1b22b06977b7f53e58fb0030a00 100644 (file)
@@ -265,16 +265,35 @@ static void komeda_component_verify_inputs(struct komeda_component *c)
        }
 }
 
+static struct komeda_layer *
+komeda_get_layer_split_right_layer(struct komeda_pipeline *pipe,
+                                  struct komeda_layer *left)
+{
+       int index = left->base.id - KOMEDA_COMPONENT_LAYER0;
+       int i;
+
+       for (i = index + 1; i < pipe->n_layers; i++)
+               if (left->layer_type == pipe->layers[i]->layer_type)
+                       return pipe->layers[i];
+       return NULL;
+}
+
 static void komeda_pipeline_assemble(struct komeda_pipeline *pipe)
 {
        struct komeda_component *c;
-       int id;
+       struct komeda_layer *layer;
+       int i, id;
 
        dp_for_each_set_bit(id, pipe->avail_comps) {
                c = komeda_pipeline_get_component(pipe, id);
-
                komeda_component_verify_inputs(c);
        }
+       /* calculate right layer for the layer split */
+       for (i = 0; i < pipe->n_layers; i++) {
+               layer = pipe->layers[i];
+
+               layer->right = komeda_get_layer_split_right_layer(pipe, layer);
+       }
 }
 
 int komeda_assemble_pipelines(struct komeda_dev *mdev)
index 857be3f64b3248a2c3504a0781581e3a87f0a566..f6a4a51cb5f723d7d96c63f765ce31baa11b3167 100644 (file)
@@ -228,6 +228,12 @@ struct komeda_layer {
        struct malidp_range hsize_in, vsize_in;
        u32 layer_type; /* RICH, SIMPLE or WB */
        u32 supported_rots;
+       /* komeda supports layer split which splits a whole image to two parts
+        * left and right and handle them by two individual layer processors
+        * Note: left/right are always according to the final display rect,
+        * not the source buffer.
+        */
+       struct komeda_layer *right;
 };
 
 struct komeda_layer_state {
@@ -339,6 +345,7 @@ struct komeda_data_flow_cfg {
        u8 en_scaling : 1,
           en_img_enhancement : 1,
           en_split : 1,
+          is_yuv : 1,
           right_part : 1; /* right part of display image if split enabled */
 };
 
@@ -493,6 +500,11 @@ int komeda_build_wb_data_flow(struct komeda_layer *wb_layer,
 int komeda_build_display_data_flow(struct komeda_crtc *kcrtc,
                                   struct komeda_crtc_state *kcrtc_st);
 
+int komeda_build_layer_split_data_flow(struct komeda_layer *left,
+                                      struct komeda_plane_state *kplane_st,
+                                      struct komeda_crtc_state *kcrtc_st,
+                                      struct komeda_data_flow_cfg *dflow);
+
 int komeda_release_unclaimed_resources(struct komeda_pipeline *pipe,
                                       struct komeda_crtc_state *kcrtc_st);
 
index 714961a0f91e22cc58c63bc1d976e78253cededf..23182edeb09a56b1264195bc7a52a74d83461371 100644 (file)
@@ -211,13 +211,14 @@ komeda_component_check_input(struct komeda_component_state *state,
        struct komeda_component *c = state->component;
 
        if ((idx < 0) || (idx >= c->max_active_inputs)) {
-               DRM_DEBUG_ATOMIC("%s invalid input id: %d.\n", c->name, idx);
+               DRM_DEBUG_ATOMIC("%s required an invalid %s-input[%d].\n",
+                                input->component->name, c->name, idx);
                return -EINVAL;
        }
 
        if (has_bit(idx, state->active_inputs)) {
-               DRM_DEBUG_ATOMIC("%s required input_id: %d has been occupied already.\n",
-                                c->name, idx);
+               DRM_DEBUG_ATOMIC("%s required %s-input[%d] has been occupied already.\n",
+                                input->component->name, c->name, idx);
                return -EINVAL;
        }
 
@@ -269,6 +270,15 @@ komeda_component_get_avail_scaler(struct komeda_component *c,
        return to_scaler(c);
 }
 
+static void
+komeda_rotate_data_flow(struct komeda_data_flow_cfg *dflow, u32 rot)
+{
+       if (drm_rotation_90_or_270(rot)) {
+               swap(dflow->in_h, dflow->in_w);
+               swap(dflow->total_in_h, dflow->total_in_w);
+       }
+}
+
 static int
 komeda_layer_check_cfg(struct komeda_layer *layer,
                       struct komeda_fb *kfb,
@@ -360,8 +370,7 @@ komeda_layer_validate(struct komeda_layer *layer,
         * The rotation has been handled by layer, so adjusted the data flow for
         * the next stage.
         */
-       if (drm_rotation_90_or_270(st->rot))
-               swap(dflow->in_h, dflow->in_w);
+       komeda_rotate_data_flow(dflow, st->rot);
 
        return 0;
 }
@@ -525,6 +534,52 @@ komeda_scaler_validate(void *user,
        return err;
 }
 
+static int
+komeda_merger_validate(struct komeda_merger *merger,
+                      void *user,
+                      struct komeda_crtc_state *kcrtc_st,
+                      struct komeda_data_flow_cfg *left_input,
+                      struct komeda_data_flow_cfg *right_input,
+                      struct komeda_data_flow_cfg *output)
+{
+       struct komeda_component_state *c_st;
+       struct komeda_merger_state *st;
+       int err = 0;
+
+       if (!merger) {
+               DRM_DEBUG_ATOMIC("No merger is available");
+               return -EINVAL;
+       }
+
+       if (!in_range(&merger->hsize_merged, output->out_w)) {
+               DRM_DEBUG_ATOMIC("merged_w: %d is out of the accepted range.\n",
+                                output->out_w);
+               return -EINVAL;
+       }
+
+       if (!in_range(&merger->vsize_merged, output->out_h)) {
+               DRM_DEBUG_ATOMIC("merged_h: %d is out of the accepted range.\n",
+                                output->out_h);
+               return -EINVAL;
+       }
+
+       c_st = komeda_component_get_state_and_set_user(&merger->base,
+                       kcrtc_st->base.state, kcrtc_st->base.crtc, kcrtc_st->base.crtc);
+
+       if (IS_ERR(c_st))
+               return PTR_ERR(c_st);
+
+       st = to_merger_st(c_st);
+       st->hsize_merged = output->out_w;
+       st->vsize_merged = output->out_h;
+
+       komeda_component_add_input(c_st, &left_input->input, 0);
+       komeda_component_add_input(c_st, &right_input->input, 1);
+       komeda_component_set_output(&output->input, &merger->base, 0);
+
+       return err;
+}
+
 void pipeline_composition_size(struct komeda_crtc_state *kcrtc_st,
                               u16 *hsize, u16 *vsize)
 {
@@ -583,6 +638,7 @@ komeda_compiz_set_input(struct komeda_compiz *compiz,
                c_st->changed_active_inputs |= BIT(idx);
 
        komeda_component_add_input(c_st, &dflow->input, idx);
+       komeda_component_set_output(&dflow->input, &compiz->base, 0);
 
        return 0;
 }
@@ -690,6 +746,16 @@ void komeda_complete_data_flow_cfg(struct komeda_data_flow_cfg *dflow,
                swap(w, h);
 
        dflow->en_scaling = (w != dflow->out_w) || (h != dflow->out_h);
+       dflow->is_yuv = fb->format->is_yuv;
+}
+
+static bool merger_is_available(struct komeda_pipeline *pipe,
+                               struct komeda_data_flow_cfg *dflow)
+{
+       u32 avail_inputs = pipe->merger ?
+                          pipe->merger->base.supported_inputs : 0;
+
+       return has_bit(dflow->input.component->id, avail_inputs);
 }
 
 int komeda_build_layer_data_flow(struct komeda_layer *layer,
@@ -714,6 +780,230 @@ int komeda_build_layer_data_flow(struct komeda_layer *layer,
        if (err)
                return err;
 
+       /* if split, check if can put the data flow into merger */
+       if (dflow->en_split && merger_is_available(pipe, dflow))
+               return 0;
+
+       err = komeda_compiz_set_input(pipe->compiz, kcrtc_st, dflow);
+
+       return err;
+}
+
+/*
+ * Split is introduced for workaround scaler's input/output size limitation.
+ * The idea is simple, if one scaler can not fit the requirement, use two.
+ * So split splits the big source image to two half parts (left/right) and do
+ * the scaling by two scaler separately and independently.
+ * But split also imports an edge problem in the middle of the image when
+ * scaling, to avoid it, split isn't a simple half-and-half, but add an extra
+ * pixels (overlap) to both side, after split the left/right will be:
+ * - left: [0, src_length/2 + overlap]
+ * - right: [src_length/2 - overlap, src_length]
+ * The extra overlap do eliminate the edge problem, but which may also generates
+ * unnecessary pixels when scaling, we need to crop them before scaler output
+ * the result to the next stage. and for the how to crop, it depends on the
+ * unneeded pixels, another words the position where overlay has been added.
+ * - left: crop the right
+ * - right: crop the left
+ *
+ * The diagram for how to do the split
+ *
+ *  <---------------------left->out_w ---------------->
+ * |--------------------------------|---right_crop-----| <- left after split
+ *  \                                \                /
+ *   \                                \<--overlap--->/
+ *   |-----------------|-------------|(Middle)------|-----------------| <- src
+ *                     /<---overlap--->\                               \
+ *                    /                 \                               \
+ * right after split->|-----left_crop---|--------------------------------|
+ *                    ^<------------------- right->out_w --------------->^
+ *
+ * NOTE: To consistent with HW the output_w always contains the crop size.
+ */
+
+static void komeda_split_data_flow(struct komeda_scaler *scaler,
+                                  struct komeda_data_flow_cfg *dflow,
+                                  struct komeda_data_flow_cfg *l_dflow,
+                                  struct komeda_data_flow_cfg *r_dflow)
+{
+       bool r90 = drm_rotation_90_or_270(dflow->rot);
+       bool flip_h = has_flip_h(dflow->rot);
+       u32 l_out, r_out, overlap;
+
+       memcpy(l_dflow, dflow, sizeof(*dflow));
+       memcpy(r_dflow, dflow, sizeof(*dflow));
+
+       l_dflow->right_part = false;
+       r_dflow->right_part = true;
+       r_dflow->blending_zorder = dflow->blending_zorder + 1;
+
+       overlap = 0;
+       if (dflow->en_scaling && scaler)
+               overlap += scaler->scaling_split_overlap;
+
+       /* original dflow may fed into splitter, and which doesn't need
+        * enhancement overlap
+        */
+       dflow->overlap = overlap;
+
+       if (dflow->en_img_enhancement && scaler)
+               overlap += scaler->enh_split_overlap;
+
+       l_dflow->overlap = overlap;
+       r_dflow->overlap = overlap;
+
+       /* split the origin content */
+       /* left/right here always means the left/right part of display image,
+        * not the source Image
+        */
+       /* DRM rotation is anti-clockwise */
+       if (r90) {
+               if (dflow->en_scaling) {
+                       l_dflow->in_h = ALIGN(dflow->in_h, 2) / 2 + l_dflow->overlap;
+                       r_dflow->in_h = l_dflow->in_h;
+               } else if (dflow->en_img_enhancement) {
+                       /* enhancer only */
+                       l_dflow->in_h = ALIGN(dflow->in_h, 2) / 2 + l_dflow->overlap;
+                       r_dflow->in_h = dflow->in_h / 2 + r_dflow->overlap;
+               } else {
+                       /* split without scaler, no overlap */
+                       l_dflow->in_h = ALIGN(((dflow->in_h + 1) >> 1), 2);
+                       r_dflow->in_h = dflow->in_h - l_dflow->in_h;
+               }
+
+               /* Consider YUV format, after split, the split source w/h
+                * may not aligned to 2. we have two choices for such case.
+                * 1. scaler is enabled (overlap != 0), we can do a alignment
+                *    both left/right and crop the extra data by scaler.
+                * 2. scaler is not enabled, only align the split left
+                *    src/disp, and the rest part assign to right
+                */
+               if ((overlap != 0) && dflow->is_yuv) {
+                       l_dflow->in_h = ALIGN(l_dflow->in_h, 2);
+                       r_dflow->in_h = ALIGN(r_dflow->in_h, 2);
+               }
+
+               if (flip_h)
+                       l_dflow->in_y = dflow->in_y + dflow->in_h - l_dflow->in_h;
+               else
+                       r_dflow->in_y = dflow->in_y + dflow->in_h - r_dflow->in_h;
+       } else {
+               if (dflow->en_scaling) {
+                       l_dflow->in_w = ALIGN(dflow->in_w, 2) / 2 + l_dflow->overlap;
+                       r_dflow->in_w = l_dflow->in_w;
+               } else if (dflow->en_img_enhancement) {
+                       l_dflow->in_w = ALIGN(dflow->in_w, 2) / 2 + l_dflow->overlap;
+                       r_dflow->in_w = dflow->in_w / 2 + r_dflow->overlap;
+               } else {
+                       l_dflow->in_w = ALIGN(((dflow->in_w + 1) >> 1), 2);
+                       r_dflow->in_w = dflow->in_w - l_dflow->in_w;
+               }
+
+               /* do YUV alignment when scaler enabled */
+               if ((overlap != 0) && dflow->is_yuv) {
+                       l_dflow->in_w = ALIGN(l_dflow->in_w, 2);
+                       r_dflow->in_w = ALIGN(r_dflow->in_w, 2);
+               }
+
+               /* on flip_h, the left display content from the right-source */
+               if (flip_h)
+                       l_dflow->in_x = dflow->in_w + dflow->in_x - l_dflow->in_w;
+               else
+                       r_dflow->in_x = dflow->in_w + dflow->in_x - r_dflow->in_w;
+       }
+
+       /* split the disp_rect */
+       if (dflow->en_scaling || dflow->en_img_enhancement)
+               l_dflow->out_w = ((dflow->out_w + 1) >> 1);
+       else
+               l_dflow->out_w = ALIGN(((dflow->out_w + 1) >> 1), 2);
+
+       r_dflow->out_w = dflow->out_w - l_dflow->out_w;
+
+       l_dflow->out_x = dflow->out_x;
+       r_dflow->out_x = l_dflow->out_w + l_dflow->out_x;
+
+       /* calculate the scaling crop */
+       /* left scaler output more data and do crop */
+       if (r90) {
+               l_out = (dflow->out_w * l_dflow->in_h) / dflow->in_h;
+               r_out = (dflow->out_w * r_dflow->in_h) / dflow->in_h;
+       } else {
+               l_out = (dflow->out_w * l_dflow->in_w) / dflow->in_w;
+               r_out = (dflow->out_w * r_dflow->in_w) / dflow->in_w;
+       }
+
+       l_dflow->left_crop  = 0;
+       l_dflow->right_crop = l_out - l_dflow->out_w;
+       r_dflow->left_crop  = r_out - r_dflow->out_w;
+       r_dflow->right_crop = 0;
+
+       /* out_w includes the crop length */
+       l_dflow->out_w += l_dflow->right_crop + l_dflow->left_crop;
+       r_dflow->out_w += r_dflow->right_crop + r_dflow->left_crop;
+}
+
+/* For layer split, a plane state will be split to two data flows and handled
+ * by two separated komeda layer input pipelines. komeda supports two types of
+ * layer split:
+ * - none-scaling split:
+ *             / layer-left -> \
+ * plane_state                  compiz-> ...
+ *             \ layer-right-> /
+ *
+ * - scaling split:
+ *             / layer-left -> scaler->\
+ * plane_state                          merger -> compiz-> ...
+ *             \ layer-right-> scaler->/
+ *
+ * Since merger only supports scaler as input, so for none-scaling split, two
+ * layer data flows will be output to compiz directly. for scaling_split, two
+ * data flow will be merged by merger firstly, then merger outputs one merged
+ * data flow to compiz.
+ */
+int komeda_build_layer_split_data_flow(struct komeda_layer *left,
+                                      struct komeda_plane_state *kplane_st,
+                                      struct komeda_crtc_state *kcrtc_st,
+                                      struct komeda_data_flow_cfg *dflow)
+{
+       struct drm_plane *plane = kplane_st->base.plane;
+       struct komeda_pipeline *pipe = left->base.pipeline;
+       struct komeda_layer *right = left->right;
+       struct komeda_data_flow_cfg l_dflow, r_dflow;
+       int err;
+
+       komeda_split_data_flow(pipe->scalers[0], dflow, &l_dflow, &r_dflow);
+
+       DRM_DEBUG_ATOMIC("Assign %s + %s to [PLANE:%d:%s]: "
+                        "src[x/y:%d/%d, w/h:%d/%d] disp[x/y:%d/%d, w/h:%d/%d]",
+                        left->base.name, right->base.name,
+                        plane->base.id, plane->name,
+                        dflow->in_x, dflow->in_y, dflow->in_w, dflow->in_h,
+                        dflow->out_x, dflow->out_y, dflow->out_w, dflow->out_h);
+
+       err = komeda_build_layer_data_flow(left, kplane_st, kcrtc_st, &l_dflow);
+       if (err)
+               return err;
+
+       err = komeda_build_layer_data_flow(right, kplane_st, kcrtc_st, &r_dflow);
+       if (err)
+               return err;
+
+       /* The rotation has been handled by layer, so adjusted the data flow */
+       komeda_rotate_data_flow(dflow, dflow->rot);
+
+       /* left and right dflow has been merged to compiz already,
+        * no need merger to merge them anymore.
+        */
+       if (r_dflow.input.component == l_dflow.input.component)
+               return 0;
+
+       /* line merger path */
+       err = komeda_merger_validate(pipe->merger, plane, kcrtc_st,
+                                    &l_dflow, &r_dflow, dflow);
+       if (err)
+               return err;
+
        err = komeda_compiz_set_input(pipe->compiz, kcrtc_st, dflow);
 
        return err;
index 9d29820345b9fdc251aa7b1555c50aabc50711d8..d1c58a88d9e2582d788adc02145e5b3d7f70542c 100644 (file)
@@ -45,7 +45,8 @@ komeda_plane_init_data_flow(struct drm_plane_state *st,
                return -EINVAL;
        }
 
-       dflow->en_img_enhancement = kplane_st->img_enhancement;
+       dflow->en_img_enhancement = !!kplane_st->img_enhancement;
+       dflow->en_split = !!kplane_st->layer_split;
 
        komeda_complete_data_flow_cfg(dflow, fb);
 
@@ -91,7 +92,12 @@ komeda_plane_atomic_check(struct drm_plane *plane,
        if (err)
                return err;
 
-       err = komeda_build_layer_data_flow(layer, kplane_st, kcrtc_st, &dflow);
+       if (dflow.en_split)
+               err = komeda_build_layer_split_data_flow(layer,
+                               kplane_st, kcrtc_st, &dflow);
+       else
+               err = komeda_build_layer_data_flow(layer,
+                               kplane_st, kcrtc_st, &dflow);
 
        return err;
 }
@@ -181,6 +187,8 @@ komeda_plane_atomic_get_property(struct drm_plane *plane,
 
        if (property == kplane->prop_img_enhancement)
                *val = st->img_enhancement;
+       else if (property == kplane->prop_layer_split)
+               *val = st->layer_split;
        else
                return -EINVAL;
 
@@ -198,6 +206,8 @@ komeda_plane_atomic_set_property(struct drm_plane *plane,
 
        if (property == kplane->prop_img_enhancement)
                st->img_enhancement = !!val;
+       else if (property == kplane->prop_layer_split)
+               st->layer_split = !!val;
        else
                return -EINVAL;
 
@@ -247,6 +257,16 @@ komeda_plane_create_layer_properties(struct komeda_plane *kplane,
                kplane->prop_img_enhancement = prop;
        }
 
+       /* property: layer split */
+       if (layer->right) {
+               prop = drm_property_create_bool(drm, DRM_MODE_PROP_ATOMIC,
+                                               "layer_split");
+               if (!prop)
+                       return -ENOMEM;
+               kplane->prop_layer_split = prop;
+               drm_object_attach_property(&plane->base, prop, 0);
+       }
+
        return 0;
 }