]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blobdiff - drivers/gpu/drm/vc4/vc4_plane.c
drm: Remove unused drm_device from drm_gem_object_lookup()
[mirror_ubuntu-artful-kernel.git] / drivers / gpu / drm / vc4 / vc4_plane.c
index 0addbad15832bc495335cbc3d6500bdb4a9d854b..7b0c72ae02a035f9f286a21cdb4a76b49f1ba155 100644 (file)
 #include "drm_fb_cma_helper.h"
 #include "drm_plane_helper.h"
 
+enum vc4_scaling_mode {
+       VC4_SCALING_NONE,
+       VC4_SCALING_TPZ,
+       VC4_SCALING_PPF,
+};
+
 struct vc4_plane_state {
        struct drm_plane_state base;
+       /* System memory copy of the display list for this element, computed
+        * at atomic_check time.
+        */
        u32 *dlist;
-       u32 dlist_size; /* Number of dwords in allocated for the display list */
+       u32 dlist_size; /* Number of dwords allocated for the display list */
        u32 dlist_count; /* Number of used dwords in the display list. */
 
-       /* Offset in the dlist to pointer word 0. */
-       u32 pw0_offset;
+       /* Offset in the dlist to various words, for pageflip or
+        * cursor updates.
+        */
+       u32 pos0_offset;
+       u32 pos2_offset;
+       u32 ptr0_offset;
 
        /* Offset where the plane's dlist was last stored in the
-          hardware at vc4_crtc_atomic_flush() time.
-       */
-       u32 *hw_dlist;
+        * hardware at vc4_crtc_atomic_flush() time.
+        */
+       u32 __iomem *hw_dlist;
+
+       /* Clipped coordinates of the plane on the display. */
+       int crtc_x, crtc_y, crtc_w, crtc_h;
+       /* Clipped area being scanned from in the FB. */
+       u32 src_x, src_y;
+
+       u32 src_w[2], src_h[2];
+
+       /* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
+       enum vc4_scaling_mode x_scaling[2], y_scaling[2];
+       bool is_unity;
+       bool is_yuv;
+
+       /* Offset to start scanning out from the start of the plane's
+        * BO.
+        */
+       u32 offsets[3];
+
+       /* Our allocation in LBM for temporary storage during scaling. */
+       struct drm_mm_node lbm;
 };
 
 static inline struct vc4_plane_state *
@@ -50,6 +83,7 @@ static const struct hvs_format {
        u32 hvs; /* HVS_FORMAT_* */
        u32 pixel_order;
        bool has_alpha;
+       bool flip_cbcr;
 } hvs_formats[] = {
        {
                .drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
@@ -59,6 +93,48 @@ static const struct hvs_format {
                .drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
                .pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = true,
        },
+       {
+               .drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
+               .pixel_order = HVS_PIXEL_ORDER_XRGB, .has_alpha = false,
+       },
+       {
+               .drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
+               .pixel_order = HVS_PIXEL_ORDER_XBGR, .has_alpha = false,
+       },
+       {
+               .drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
+               .pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = true,
+       },
+       {
+               .drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
+               .pixel_order = HVS_PIXEL_ORDER_ABGR, .has_alpha = false,
+       },
+       {
+               .drm = DRM_FORMAT_YUV422,
+               .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
+       },
+       {
+               .drm = DRM_FORMAT_YVU422,
+               .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
+               .flip_cbcr = true,
+       },
+       {
+               .drm = DRM_FORMAT_YUV420,
+               .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
+       },
+       {
+               .drm = DRM_FORMAT_YVU420,
+               .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
+               .flip_cbcr = true,
+       },
+       {
+               .drm = DRM_FORMAT_NV12,
+               .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
+       },
+       {
+               .drm = DRM_FORMAT_NV16,
+               .hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
+       },
 };
 
 static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
@@ -73,6 +149,16 @@ static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
        return NULL;
 }
 
+static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
+{
+       if (dst > src)
+               return VC4_SCALING_PPF;
+       else if (dst < src)
+               return VC4_SCALING_TPZ;
+       else
+               return VC4_SCALING_NONE;
+}
+
 static bool plane_enabled(struct drm_plane_state *state)
 {
        return state->fb && state->crtc;
@@ -89,6 +175,8 @@ static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane
        if (!vc4_state)
                return NULL;
 
+       memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
+
        __drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
 
        if (vc4_state->dlist) {
@@ -108,8 +196,17 @@ static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane
 static void vc4_plane_destroy_state(struct drm_plane *plane,
                                    struct drm_plane_state *state)
 {
+       struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
        struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 
+       if (vc4_state->lbm.allocated) {
+               unsigned long irqflags;
+
+               spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
+               drm_mm_remove_node(&vc4_state->lbm);
+               spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
+       }
+
        kfree(vc4_state->dlist);
        __drm_atomic_helper_plane_destroy_state(plane, &vc4_state->base);
        kfree(state);
@@ -148,84 +245,400 @@ static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
        vc4_state->dlist[vc4_state->dlist_count++] = val;
 }
 
+/* Returns the scl0/scl1 field based on whether the dimensions need to
+ * be up/down/non-scaled.
+ *
+ * This is a replication of a table from the spec.
+ */
+static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
+{
+       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+
+       switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
+       case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
+               return SCALER_CTL0_SCL_H_PPF_V_PPF;
+       case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
+               return SCALER_CTL0_SCL_H_TPZ_V_PPF;
+       case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
+               return SCALER_CTL0_SCL_H_PPF_V_TPZ;
+       case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
+               return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
+       case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
+               return SCALER_CTL0_SCL_H_PPF_V_NONE;
+       case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
+               return SCALER_CTL0_SCL_H_NONE_V_PPF;
+       case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
+               return SCALER_CTL0_SCL_H_NONE_V_TPZ;
+       case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
+               return SCALER_CTL0_SCL_H_TPZ_V_NONE;
+       default:
+       case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
+               /* The unity case is independently handled by
+                * SCALER_CTL0_UNITY.
+                */
+               return 0;
+       }
+}
+
+static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
+{
+       struct drm_plane *plane = state->plane;
+       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+       struct drm_framebuffer *fb = state->fb;
+       struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
+       u32 subpixel_src_mask = (1 << 16) - 1;
+       u32 format = fb->pixel_format;
+       int num_planes = drm_format_num_planes(format);
+       u32 h_subsample = 1;
+       u32 v_subsample = 1;
+       int i;
+
+       for (i = 0; i < num_planes; i++)
+               vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
+
+       /* We don't support subpixel source positioning for scaling. */
+       if ((state->src_x & subpixel_src_mask) ||
+           (state->src_y & subpixel_src_mask) ||
+           (state->src_w & subpixel_src_mask) ||
+           (state->src_h & subpixel_src_mask)) {
+               return -EINVAL;
+       }
+
+       vc4_state->src_x = state->src_x >> 16;
+       vc4_state->src_y = state->src_y >> 16;
+       vc4_state->src_w[0] = state->src_w >> 16;
+       vc4_state->src_h[0] = state->src_h >> 16;
+
+       vc4_state->crtc_x = state->crtc_x;
+       vc4_state->crtc_y = state->crtc_y;
+       vc4_state->crtc_w = state->crtc_w;
+       vc4_state->crtc_h = state->crtc_h;
+
+       vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
+                                                      vc4_state->crtc_w);
+       vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
+                                                      vc4_state->crtc_h);
+
+       if (num_planes > 1) {
+               vc4_state->is_yuv = true;
+
+               h_subsample = drm_format_horz_chroma_subsampling(format);
+               v_subsample = drm_format_vert_chroma_subsampling(format);
+               vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
+               vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
+
+               vc4_state->x_scaling[1] =
+                       vc4_get_scaling_mode(vc4_state->src_w[1],
+                                            vc4_state->crtc_w);
+               vc4_state->y_scaling[1] =
+                       vc4_get_scaling_mode(vc4_state->src_h[1],
+                                            vc4_state->crtc_h);
+
+               /* YUV conversion requires that scaling be enabled,
+                * even on a plane that's otherwise 1:1.  Choose TPZ
+                * for simplicity.
+                */
+               if (vc4_state->x_scaling[0] == VC4_SCALING_NONE)
+                       vc4_state->x_scaling[0] = VC4_SCALING_TPZ;
+               if (vc4_state->y_scaling[0] == VC4_SCALING_NONE)
+                       vc4_state->y_scaling[0] = VC4_SCALING_TPZ;
+       }
+
+       vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
+                              vc4_state->y_scaling[0] == VC4_SCALING_NONE &&
+                              vc4_state->x_scaling[1] == VC4_SCALING_NONE &&
+                              vc4_state->y_scaling[1] == VC4_SCALING_NONE);
+
+       /* No configuring scaling on the cursor plane, since it gets
+          non-vblank-synced updates, and scaling requires requires
+          LBM changes which have to be vblank-synced.
+        */
+       if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
+               return -EINVAL;
+
+       /* Clamp the on-screen start x/y to 0.  The hardware doesn't
+        * support negative y, and negative x wastes bandwidth.
+        */
+       if (vc4_state->crtc_x < 0) {
+               for (i = 0; i < num_planes; i++) {
+                       u32 cpp = drm_format_plane_cpp(fb->pixel_format, i);
+                       u32 subs = ((i == 0) ? 1 : h_subsample);
+
+                       vc4_state->offsets[i] += (cpp *
+                                                 (-vc4_state->crtc_x) / subs);
+               }
+               vc4_state->src_w[0] += vc4_state->crtc_x;
+               vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
+               vc4_state->crtc_x = 0;
+       }
+
+       if (vc4_state->crtc_y < 0) {
+               for (i = 0; i < num_planes; i++) {
+                       u32 subs = ((i == 0) ? 1 : v_subsample);
+
+                       vc4_state->offsets[i] += (fb->pitches[i] *
+                                                 (-vc4_state->crtc_y) / subs);
+               }
+               vc4_state->src_h[0] += vc4_state->crtc_y;
+               vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
+               vc4_state->crtc_y = 0;
+       }
+
+       return 0;
+}
+
+static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
+{
+       u32 scale, recip;
+
+       scale = (1 << 16) * src / dst;
+
+       /* The specs note that while the reciprocal would be defined
+        * as (1<<32)/scale, ~0 is close enough.
+        */
+       recip = ~0 / scale;
+
+       vc4_dlist_write(vc4_state,
+                       VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
+                       VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
+       vc4_dlist_write(vc4_state,
+                       VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
+}
+
+static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
+{
+       u32 scale = (1 << 16) * src / dst;
+
+       vc4_dlist_write(vc4_state,
+                       SCALER_PPF_AGC |
+                       VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
+                       VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
+}
+
+static u32 vc4_lbm_size(struct drm_plane_state *state)
+{
+       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+       /* This is the worst case number.  One of the two sizes will
+        * be used depending on the scaling configuration.
+        */
+       u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
+       u32 lbm;
+
+       if (!vc4_state->is_yuv) {
+               if (vc4_state->is_unity)
+                       return 0;
+               else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
+                       lbm = pix_per_line * 8;
+               else {
+                       /* In special cases, this multiplier might be 12. */
+                       lbm = pix_per_line * 16;
+               }
+       } else {
+               /* There are cases for this going down to a multiplier
+                * of 2, but according to the firmware source, the
+                * table in the docs is somewhat wrong.
+                */
+               lbm = pix_per_line * 16;
+       }
+
+       lbm = roundup(lbm, 32);
+
+       return lbm;
+}
+
+static void vc4_write_scaling_parameters(struct drm_plane_state *state,
+                                        int channel)
+{
+       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+
+       /* Ch0 H-PPF Word 0: Scaling Parameters */
+       if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
+               vc4_write_ppf(vc4_state,
+                             vc4_state->src_w[channel], vc4_state->crtc_w);
+       }
+
+       /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
+       if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
+               vc4_write_ppf(vc4_state,
+                             vc4_state->src_h[channel], vc4_state->crtc_h);
+               vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+       }
+
+       /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
+       if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
+               vc4_write_tpz(vc4_state,
+                             vc4_state->src_w[channel], vc4_state->crtc_w);
+       }
+
+       /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
+       if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
+               vc4_write_tpz(vc4_state,
+                             vc4_state->src_h[channel], vc4_state->crtc_h);
+               vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+       }
+}
+
 /* Writes out a full display list for an active plane to the plane's
  * private dlist state.
  */
 static int vc4_plane_mode_set(struct drm_plane *plane,
                              struct drm_plane_state *state)
 {
+       struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
        struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
        struct drm_framebuffer *fb = state->fb;
-       struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
        u32 ctl0_offset = vc4_state->dlist_count;
        const struct hvs_format *format = vc4_get_hvs_format(fb->pixel_format);
-       uint32_t offset = fb->offsets[0];
-       int crtc_x = state->crtc_x;
-       int crtc_y = state->crtc_y;
-       int crtc_w = state->crtc_w;
-       int crtc_h = state->crtc_h;
-
-       if (state->crtc_w << 16 != state->src_w ||
-           state->crtc_h << 16 != state->src_h) {
-               /* We don't support scaling yet, which involves
-                * allocating the LBM memory for scaling temporary
-                * storage, and putting filter kernels in the HVS
-                * context.
-                */
-               return -EINVAL;
+       int num_planes = drm_format_num_planes(format->drm);
+       u32 scl0, scl1;
+       u32 lbm_size;
+       unsigned long irqflags;
+       int ret, i;
+
+       ret = vc4_plane_setup_clipping_and_scaling(state);
+       if (ret)
+               return ret;
+
+       /* Allocate the LBM memory that the HVS will use for temporary
+        * storage due to our scaling/format conversion.
+        */
+       lbm_size = vc4_lbm_size(state);
+       if (lbm_size) {
+               if (!vc4_state->lbm.allocated) {
+                       spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
+                       ret = drm_mm_insert_node(&vc4->hvs->lbm_mm,
+                                                &vc4_state->lbm,
+                                                lbm_size, 32, 0);
+                       spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
+               } else {
+                       WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
+               }
        }
 
-       if (crtc_x < 0) {
-               offset += drm_format_plane_cpp(fb->pixel_format, 0) * -crtc_x;
-               crtc_w += crtc_x;
-               crtc_x = 0;
-       }
+       if (ret)
+               return ret;
 
-       if (crtc_y < 0) {
-               offset += fb->pitches[0] * -crtc_y;
-               crtc_h += crtc_y;
-               crtc_y = 0;
+       /* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
+        * and 4:4:4, scl1 should be set to scl0 so both channels of
+        * the scaler do the same thing.  For YUV, the Y plane needs
+        * to be put in channel 1 and Cb/Cr in channel 0, so we swap
+        * the scl fields here.
+        */
+       if (num_planes == 1) {
+               scl0 = vc4_get_scl_field(state, 1);
+               scl1 = scl0;
+       } else {
+               scl0 = vc4_get_scl_field(state, 1);
+               scl1 = vc4_get_scl_field(state, 0);
        }
 
+       /* Control word */
        vc4_dlist_write(vc4_state,
                        SCALER_CTL0_VALID |
                        (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
                        (format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
-                       SCALER_CTL0_UNITY);
+                       (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
+                       VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
+                       VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
 
        /* Position Word 0: Image Positions and Alpha Value */
+       vc4_state->pos0_offset = vc4_state->dlist_count;
        vc4_dlist_write(vc4_state,
                        VC4_SET_FIELD(0xff, SCALER_POS0_FIXED_ALPHA) |
-                       VC4_SET_FIELD(crtc_x, SCALER_POS0_START_X) |
-                       VC4_SET_FIELD(crtc_y, SCALER_POS0_START_Y));
-
-       /* Position Word 1: Scaled Image Dimensions.
-        * Skipped due to SCALER_CTL0_UNITY scaling.
-        */
+                       VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
+                       VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
+
+       /* Position Word 1: Scaled Image Dimensions. */
+       if (!vc4_state->is_unity) {
+               vc4_dlist_write(vc4_state,
+                               VC4_SET_FIELD(vc4_state->crtc_w,
+                                             SCALER_POS1_SCL_WIDTH) |
+                               VC4_SET_FIELD(vc4_state->crtc_h,
+                                             SCALER_POS1_SCL_HEIGHT));
+       }
 
        /* Position Word 2: Source Image Size, Alpha Mode */
+       vc4_state->pos2_offset = vc4_state->dlist_count;
        vc4_dlist_write(vc4_state,
                        VC4_SET_FIELD(format->has_alpha ?
                                      SCALER_POS2_ALPHA_MODE_PIPELINE :
                                      SCALER_POS2_ALPHA_MODE_FIXED,
                                      SCALER_POS2_ALPHA_MODE) |
-                       VC4_SET_FIELD(crtc_w, SCALER_POS2_WIDTH) |
-                       VC4_SET_FIELD(crtc_h, SCALER_POS2_HEIGHT));
+                       VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
+                       VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
 
        /* Position Word 3: Context.  Written by the HVS. */
        vc4_dlist_write(vc4_state, 0xc0c0c0c0);
 
-       vc4_state->pw0_offset = vc4_state->dlist_count;
 
-       /* Pointer Word 0: RGB / Y Pointer */
-       vc4_dlist_write(vc4_state, bo->paddr + offset);
+       /* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
+        *
+        * The pointers may be any byte address.
+        */
+       vc4_state->ptr0_offset = vc4_state->dlist_count;
+       if (!format->flip_cbcr) {
+               for (i = 0; i < num_planes; i++)
+                       vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
+       } else {
+               WARN_ON_ONCE(num_planes != 3);
+               vc4_dlist_write(vc4_state, vc4_state->offsets[0]);
+               vc4_dlist_write(vc4_state, vc4_state->offsets[2]);
+               vc4_dlist_write(vc4_state, vc4_state->offsets[1]);
+       }
 
-       /* Pointer Context Word 0: Written by the HVS */
-       vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+       /* Pointer Context Word 0/1/2: Written by the HVS */
+       for (i = 0; i < num_planes; i++)
+               vc4_dlist_write(vc4_state, 0xc0c0c0c0);
 
-       /* Pitch word 0: Pointer 0 Pitch */
-       vc4_dlist_write(vc4_state,
-                       VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH));
+       /* Pitch word 0/1/2 */
+       for (i = 0; i < num_planes; i++) {
+               vc4_dlist_write(vc4_state,
+                               VC4_SET_FIELD(fb->pitches[i], SCALER_SRC_PITCH));
+       }
+
+       /* Colorspace conversion words */
+       if (vc4_state->is_yuv) {
+               vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
+               vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
+               vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
+       }
+
+       if (!vc4_state->is_unity) {
+               /* LBM Base Address. */
+               if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
+                   vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
+                       vc4_dlist_write(vc4_state, vc4_state->lbm.start);
+               }
+
+               if (num_planes > 1) {
+                       /* Emit Cb/Cr as channel 0 and Y as channel
+                        * 1. This matches how we set up scl0/scl1
+                        * above.
+                        */
+                       vc4_write_scaling_parameters(state, 1);
+               }
+               vc4_write_scaling_parameters(state, 0);
+
+               /* If any PPF setup was done, then all the kernel
+                * pointers get uploaded.
+                */
+               if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
+                   vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
+                   vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
+                   vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
+                       u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
+                                                  SCALER_PPF_KERNEL_OFFSET);
+
+                       /* HPPF plane 0 */
+                       vc4_dlist_write(vc4_state, kernel);
+                       /* VPPF plane 0 */
+                       vc4_dlist_write(vc4_state, kernel);
+                       /* HPPF plane 1 */
+                       vc4_dlist_write(vc4_state, kernel);
+                       /* VPPF plane 1 */
+                       vc4_dlist_write(vc4_state, kernel);
+               }
+       }
 
        vc4_state->dlist[ctl0_offset] |=
                VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
@@ -303,13 +716,13 @@ void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
         * scanout will start from this address as soon as the FIFO
         * needs to refill with pixels.
         */
-       writel(addr, &vc4_state->hw_dlist[vc4_state->pw0_offset]);
+       writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
 
        /* Also update the CPU-side dlist copy, so that any later
         * atomic updates that don't do a new modeset on our plane
         * also use our updated address.
         */
-       vc4_state->dlist[vc4_state->pw0_offset] = addr;
+       vc4_state->dlist[vc4_state->ptr0_offset] = addr;
 }
 
 static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
@@ -325,8 +738,83 @@ static void vc4_plane_destroy(struct drm_plane *plane)
        drm_plane_cleanup(plane);
 }
 
+/* Implements immediate (non-vblank-synced) updates of the cursor
+ * position, or falls back to the atomic helper otherwise.
+ */
+static int
+vc4_update_plane(struct drm_plane *plane,
+                struct drm_crtc *crtc,
+                struct drm_framebuffer *fb,
+                int crtc_x, int crtc_y,
+                unsigned int crtc_w, unsigned int crtc_h,
+                uint32_t src_x, uint32_t src_y,
+                uint32_t src_w, uint32_t src_h)
+{
+       struct drm_plane_state *plane_state;
+       struct vc4_plane_state *vc4_state;
+
+       if (plane != crtc->cursor)
+               goto out;
+
+       plane_state = plane->state;
+       vc4_state = to_vc4_plane_state(plane_state);
+
+       if (!plane_state)
+               goto out;
+
+       /* If we're changing the cursor contents, do that in the
+        * normal vblank-synced atomic path.
+        */
+       if (fb != plane_state->fb)
+               goto out;
+
+       /* No configuring new scaling in the fast path. */
+       if (crtc_w != plane_state->crtc_w ||
+           crtc_h != plane_state->crtc_h ||
+           src_w != plane_state->src_w ||
+           src_h != plane_state->src_h) {
+               goto out;
+       }
+
+       /* Set the cursor's position on the screen.  This is the
+        * expected change from the drm_mode_cursor_universal()
+        * helper.
+        */
+       plane_state->crtc_x = crtc_x;
+       plane_state->crtc_y = crtc_y;
+
+       /* Allow changing the start position within the cursor BO, if
+        * that matters.
+        */
+       plane_state->src_x = src_x;
+       plane_state->src_y = src_y;
+
+       /* Update the display list based on the new crtc_x/y. */
+       vc4_plane_atomic_check(plane, plane_state);
+
+       /* Note that we can't just call vc4_plane_write_dlist()
+        * because that would smash the context data that the HVS is
+        * currently using.
+        */
+       writel(vc4_state->dlist[vc4_state->pos0_offset],
+              &vc4_state->hw_dlist[vc4_state->pos0_offset]);
+       writel(vc4_state->dlist[vc4_state->pos2_offset],
+              &vc4_state->hw_dlist[vc4_state->pos2_offset]);
+       writel(vc4_state->dlist[vc4_state->ptr0_offset],
+              &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
+
+       return 0;
+
+out:
+       return drm_atomic_helper_update_plane(plane, crtc, fb,
+                                             crtc_x, crtc_y,
+                                             crtc_w, crtc_h,
+                                             src_x, src_y,
+                                             src_w, src_h);
+}
+
 static const struct drm_plane_funcs vc4_plane_funcs = {
-       .update_plane = drm_atomic_helper_update_plane,
+       .update_plane = vc4_update_plane,
        .disable_plane = drm_atomic_helper_disable_plane,
        .destroy = vc4_plane_destroy,
        .set_property = NULL,
@@ -341,6 +829,7 @@ struct drm_plane *vc4_plane_init(struct drm_device *dev,
        struct drm_plane *plane = NULL;
        struct vc4_plane *vc4_plane;
        u32 formats[ARRAY_SIZE(hvs_formats)];
+       u32 num_formats = 0;
        int ret = 0;
        unsigned i;
 
@@ -351,12 +840,20 @@ struct drm_plane *vc4_plane_init(struct drm_device *dev,
                goto fail;
        }
 
-       for (i = 0; i < ARRAY_SIZE(hvs_formats); i++)
-               formats[i] = hvs_formats[i].drm;
+       for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
+               /* Don't allow YUV in cursor planes, since that means
+                * tuning on the scaler, which we don't allow for the
+                * cursor.
+                */
+               if (type != DRM_PLANE_TYPE_CURSOR ||
+                   hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
+                       formats[num_formats++] = hvs_formats[i].drm;
+               }
+       }
        plane = &vc4_plane->base;
        ret = drm_universal_plane_init(dev, plane, 0xff,
                                       &vc4_plane_funcs,
-                                      formats, ARRAY_SIZE(formats),
+                                      formats, num_formats,
                                       type, NULL);
 
        drm_plane_helper_add(plane, &vc4_plane_helper_funcs);