path: root/target/linux/bcm27xx/patches-5.4/950-0234-drm-vc4-Add-an-overlay-plane-to-vc4-firmware-kms.patch

From 75e1dce99c260cb1365edd9af68cb5c07487831c Mon Sep 17 00:00:00 2001
From: Dave Stevenson <dave.stevenson@raspberrypi.org>
Date: Wed, 27 Mar 2019 17:45:01 +0000
Subject: [PATCH] drm: vc4: Add an overlay plane to vc4-firmware-kms

This uses a new API that is exposed via the mailbox service
to stick an element straight on the screen using DispmanX.

The primary and cursor planes have also been switched to using
the new plane API, and it supports layering based on the DRM
zpos parameter.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.org>
---
 drivers/gpu/drm/vc4/vc4_firmware_kms.c     | 518 ++++++++++++++-------
 drivers/gpu/drm/vc4/vc4_kms.c              |   1 +
 drivers/gpu/drm/vc4/vc_image_types.h       | 143 ++++++
 include/soc/bcm2835/raspberrypi-firmware.h |   2 +
 4 files changed, 495 insertions(+), 169 deletions(-)
 create mode 100644 drivers/gpu/drm/vc4/vc_image_types.h

--- a/drivers/gpu/drm/vc4/vc4_firmware_kms.c
+++ b/drivers/gpu/drm/vc4/vc4_firmware_kms.c
@@ -28,8 +28,46 @@
 #include "linux/of_device.h"
 #include "vc4_drv.h"
 #include "vc4_regs.h"
+#include "vc_image_types.h"
 #include <soc/bcm2835/raspberrypi-firmware.h>
 
+struct set_plane {
+	u8 display;
+	u8 plane_id;
+	u8 vc_image_type;
+	s8 layer;
+
+	u16 width;
+	u16 height;
+
+	u16 pitch;
+	u16 vpitch;
+
+	u32 src_x;	/* 16p16 */
+	u32 src_y;	/* 16p16 */
+
+	u32 src_w;	/* 16p16 */
+	u32 src_h;	/* 16p16 */
+
+	s16 dst_x;
+	s16 dst_y;
+
+	u16 dst_w;
+	u16 dst_h;
+
+	u8 alpha;
+	u8 num_planes;
+	u8 is_vu;
+	u8 padding;
+
+	u32 planes[4];  /* DMA address of each plane */
+};
+
+struct mailbox_set_plane {
+	struct rpi_firmware_property_tag_header tag;
+	struct set_plane plane;
+};
+
 struct fb_alloc_tags {
 	struct rpi_firmware_property_tag_header tag1;
 	u32 xres, yres;
@@ -49,6 +87,79 @@ struct fb_alloc_tags {
 	u32 layer;
 };
 
+static const struct vc_image_format {
+	u32 drm;	/* DRM_FORMAT_* */
+	u32 vc_image;	/* VC_IMAGE_* */
+	u32 is_vu;
+} vc_image_formats[] = {
+	{
+		.drm = DRM_FORMAT_XRGB8888,
+		.vc_image = VC_IMAGE_XRGB8888,
+	},
+	{
+		.drm = DRM_FORMAT_ARGB8888,
+		.vc_image = VC_IMAGE_ARGB8888,
+	},
+/*
+ *	FIXME: Need to resolve which DRM format goes to which vc_image format
+ *	for the remaining RGBA and RGBX formats.
+ *	{
+ *		.drm = DRM_FORMAT_ABGR8888,
+ *		.vc_image = VC_IMAGE_RGBA8888,
+ *	},
+ *	{
+ *		.drm = DRM_FORMAT_XBGR8888,
+ *		.vc_image = VC_IMAGE_RGBA8888,
+ *	},
+ */
+	{
+		.drm = DRM_FORMAT_RGB565,
+		.vc_image = VC_IMAGE_RGB565,
+	},
+	{
+		.drm = DRM_FORMAT_RGB888,
+		.vc_image = VC_IMAGE_BGR888,
+	},
+	{
+		.drm = DRM_FORMAT_BGR888,
+		.vc_image = VC_IMAGE_RGB888,
+	},
+	{
+		.drm = DRM_FORMAT_YUV422,
+		.vc_image = VC_IMAGE_YUV422PLANAR,
+	},
+	{
+		.drm = DRM_FORMAT_YUV420,
+		.vc_image = VC_IMAGE_YUV420,
+	},
+	{
+		.drm = DRM_FORMAT_YVU420,
+		.vc_image = VC_IMAGE_YUV420,
+		.is_vu = 1,
+	},
+	{
+		.drm = DRM_FORMAT_NV12,
+		.vc_image = VC_IMAGE_YUV420SP,
+	},
+	{
+		.drm = DRM_FORMAT_NV21,
+		.vc_image = VC_IMAGE_YUV420SP,
+		.is_vu = 1,
+	},
+};
+
+static const struct vc_image_format *vc4_get_vc_image_fmt(u32 drm_format)
+{
+	unsigned int i;
+
+	for (i = 0; i < ARRAY_SIZE(vc_image_formats); i++) {
+		if (vc_image_formats[i].drm == drm_format)
+			return &vc_image_formats[i];
+	}
+
+	return NULL;
+}
+
 /* The firmware delivers a vblank interrupt to us through the SMI
  * hardware, which has only this one register.
  */
@@ -115,6 +226,7 @@ struct vc4_fkms_plane {
 	struct fbinfo_s *fbinfo;
 	dma_addr_t fbinfo_bus_addr;
 	u32 pitch;
+	struct mailbox_set_plane mb;
 };
 
 static inline struct vc4_fkms_plane *to_vc4_fkms_plane(struct drm_plane *plane)
@@ -122,165 +234,183 @@ static inline struct vc4_fkms_plane *to_
 	return (struct vc4_fkms_plane *)plane;
 }
 
-/* Turns the display on/off. */
-static int vc4_plane_set_primary_blank(struct drm_plane *plane, bool blank)
+static int vc4_plane_set_blank(struct drm_plane *plane, bool blank)
 {
 	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
+	struct vc4_fkms_plane *vc4_plane = to_vc4_fkms_plane(plane);
+	struct mailbox_set_plane blank_mb = {
+		.tag = { RPI_FIRMWARE_SET_PLANE, sizeof(struct set_plane), 0 },
+		.plane = {
+			.display = vc4_plane->mb.plane.display,
+			.plane_id = vc4_plane->mb.plane.plane_id,
+		}
+	};
+	int ret;
 
-	u32 packet = blank;
-
-	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] primary plane %s",
+	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] overlay plane %s",
 			 plane->base.id, plane->name,
 			 blank ? "blank" : "unblank");
 
-	return rpi_firmware_property(vc4->firmware,
-				     RPI_FIRMWARE_FRAMEBUFFER_BLANK,
-				     &packet, sizeof(packet));
+	if (blank)
+		ret = rpi_firmware_property_list(vc4->firmware, &blank_mb,
+						 sizeof(blank_mb));
+	else
+		ret = rpi_firmware_property_list(vc4->firmware, &vc4_plane->mb,
+						 sizeof(vc4_plane->mb));
+
+	WARN_ONCE(ret, "%s: firmware call failed. Please update your firmware",
+		  __func__);
+	return ret;
 }
 
-static void vc4_primary_plane_atomic_update(struct drm_plane *plane,
-					    struct drm_plane_state *old_state)
+static void vc4_plane_atomic_update(struct drm_plane *plane,
+				    struct drm_plane_state *old_state)
 {
-	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
 	struct drm_plane_state *state = plane->state;
 	struct drm_framebuffer *fb = state->fb;
 	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
-	u32 format = fb->format->format;
-	struct fb_alloc_tags fbinfo = {
-		.tag1 = { RPI_FIRMWARE_FRAMEBUFFER_SET_PHYSICAL_WIDTH_HEIGHT,
-			  8, 0, },
-			.xres = state->crtc_w,
-			.yres = state->crtc_h,
-		.tag2 = { RPI_FIRMWARE_FRAMEBUFFER_SET_VIRTUAL_WIDTH_HEIGHT,
-			  8, 0, },
-			.xres_virtual = state->crtc_w,
-			.yres_virtual = state->crtc_h,
-		.tag3 = { RPI_FIRMWARE_FRAMEBUFFER_SET_DEPTH, 4, 0 },
-			.bpp = 32,
-		.tag4 = { RPI_FIRMWARE_FRAMEBUFFER_SET_VIRTUAL_OFFSET, 8, 0 },
-			.xoffset = 0,
-			.yoffset = 0,
-		.tag5 = { RPI_FIRMWARE_FRAMEBUFFER_ALLOCATE, 8, 0 },
-			.base = bo->paddr + fb->offsets[0],
-			.screen_size = state->crtc_w * state->crtc_h * 4,
-		.tag6 = { RPI_FIRMWARE_FRAMEBUFFER_SET_PITCH, 4, 0 },
-			.pitch = fb->pitches[0],
-		.tag7 = { RPI_FIRMWARE_FRAMEBUFFER_SET_ALPHA_MODE, 4, 0 },
-			.alpha_mode = format == DRM_FORMAT_ARGB8888 ? 0 : 2,
-		.tag8 = { RPI_FIRMWARE_FRAMEBUFFER_SET_LAYER, 4, 0 },
-			.layer = -127,
-	};
-	u32 bpp = 32;
-	int ret;
+	const struct drm_format_info *drm_fmt = fb->format;
+	const struct vc_image_format *vc_fmt =
+					vc4_get_vc_image_fmt(drm_fmt->format);
+	struct vc4_fkms_plane *vc4_plane = to_vc4_fkms_plane(plane);
+	struct mailbox_set_plane *mb = &vc4_plane->mb;
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(state->crtc);
+	int num_planes = fb->format->num_planes;
+	struct drm_display_mode *mode = &state->crtc->mode;
 
-	if (fb->modifier == DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED)
-		fbinfo.bpp |= BIT(31);
+	mb->plane.vc_image_type = vc_fmt->vc_image;
+	mb->plane.width = fb->width;
+	mb->plane.height = fb->height;
+	mb->plane.pitch = fb->pitches[0];
+	mb->plane.src_w = state->src_w;
+	mb->plane.src_h = state->src_h;
+	mb->plane.src_x = state->src_x;
+	mb->plane.src_y = state->src_y;
+	mb->plane.dst_w = state->crtc_w;
+	mb->plane.dst_h = state->crtc_h;
+	mb->plane.dst_x = state->crtc_x;
+	mb->plane.dst_y = state->crtc_y;
+	mb->plane.alpha = state->alpha >> 8;
+	mb->plane.layer = state->normalized_zpos ?
+					state->normalized_zpos : -127;
+	mb->plane.num_planes = num_planes;
+	mb->plane.is_vu = vc_fmt->is_vu;
+	mb->plane.planes[0] = bo->paddr + fb->offsets[0];
 
-	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] primary update %dx%d@%d +%d,%d 0x%pad/%d\n",
+	/* FIXME: If the dest rect goes off screen then clip the src rect so we
+	 * don't have off-screen pixels.
+	 */
+	if (plane->type == DRM_PLANE_TYPE_CURSOR) {
+		/* There is no scaling on the cursor plane, therefore the calcs
+		 * to alter the source crop as the cursor goes off the screen
+		 * are simple.
+		 */
+		if (mb->plane.dst_x + mb->plane.dst_w > mode->hdisplay) {
+			mb->plane.dst_w = mode->hdisplay - mb->plane.dst_x;
+			mb->plane.src_w = (mode->hdisplay - mb->plane.dst_x)
+									<< 16;
+		}
+		if (mb->plane.dst_y + mb->plane.dst_h > mode->vdisplay) {
+			mb->plane.dst_h = mode->vdisplay - mb->plane.dst_y;
+			mb->plane.src_h = (mode->vdisplay - mb->plane.dst_y)
+									<< 16;
+		}
+	}
+
+	if (num_planes > 1) {
+		/* Assume this must be YUV */
+		/* Makes assumptions on the stride for the chroma planes as we
+		 * can't easily plumb in non-standard pitches.
+		 */
+		mb->plane.planes[1] = bo->paddr + fb->offsets[1];
+		if (num_planes > 2)
+			mb->plane.planes[2] = bo->paddr + fb->offsets[2];
+		else
+			mb->plane.planes[2] = 0;
+
+		/* Special case the YUV420 with U and V as line interleaved
+		 * planes as we have special handling for that case.
+		 */
+		if (num_planes == 3 &&
+		    (fb->offsets[2] - fb->offsets[1]) == fb->pitches[1])
+			mb->plane.vc_image_type = VC_IMAGE_YUV420_S;
+	} else {
+		mb->plane.planes[1] = 0;
+		mb->plane.planes[2] = 0;
+	}
+	mb->plane.planes[3] = 0;
+
+	switch (fb->modifier) {
+	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
+		switch (mb->plane.vc_image_type) {
+		case VC_IMAGE_RGBX32:
+			mb->plane.vc_image_type = VC_IMAGE_TF_RGBX32;
+			break;
+		case VC_IMAGE_RGBA32:
+			mb->plane.vc_image_type = VC_IMAGE_TF_RGBA32;
+			break;
+		case VC_IMAGE_RGB565:
+			mb->plane.vc_image_type = VC_IMAGE_TF_RGB565;
+			break;
+		}
+		break;
+	case DRM_FORMAT_MOD_BROADCOM_SAND128:
+		mb->plane.vc_image_type = VC_IMAGE_YUV_UV;
+		mb->plane.pitch = fourcc_mod_broadcom_param(fb->modifier);
+		break;
+	}
+
+	if (vc4_crtc) {
+		mb->plane.dst_x += vc4_crtc->overscan[0];
+		mb->plane.dst_y += vc4_crtc->overscan[1];
+	}
+
+	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] plane update %dx%d@%d +dst(%d,%d, %d,%d) +src(%d,%d, %d,%d) 0x%08x/%08x/%08x/%d, alpha %u zpos %u\n",
 			 plane->base.id, plane->name,
-			 state->crtc_w,
-			 state->crtc_h,
-			 bpp,
+			 mb->plane.width,
+			 mb->plane.height,
+			 mb->plane.vc_image_type,
 			 state->crtc_x,
 			 state->crtc_y,
-			 &fbinfo.base,
-			 fb->pitches[0]);
-
-	ret = rpi_firmware_property_list(vc4->firmware, &fbinfo,
-					 sizeof(fbinfo));
-	WARN_ON_ONCE(fbinfo.pitch != fb->pitches[0]);
-	WARN_ON_ONCE(fbinfo.base != bo->paddr + fb->offsets[0]);
-
-	/* If the CRTC is on (or going to be on) and we're enabled,
+			 state->crtc_w,
+			 state->crtc_h,
+			 mb->plane.src_x,
+			 mb->plane.src_y,
+			 mb->plane.src_w,
+			 mb->plane.src_h,
+			 mb->plane.planes[0],
+			 mb->plane.planes[1],
+			 mb->plane.planes[2],
+			 fb->pitches[0],
+			 state->alpha,
+			 state->normalized_zpos);
+
+	/*
+	 * Do NOT set now, as we haven't checked if the crtc is active or not.
+	 * Set from vc4_plane_set_blank instead.
+	 *
+	 * If the CRTC is on (or going to be on) and we're enabled,
 	 * then unblank.  Otherwise, stay blank until CRTC enable.
-	*/
+	 */
 	if (state->crtc->state->active)
-		vc4_plane_set_primary_blank(plane, false);
+		vc4_plane_set_blank(plane, false);
 }
 
-static void vc4_primary_plane_atomic_disable(struct drm_plane *plane,
-					     struct drm_plane_state *old_state)
+static void vc4_plane_atomic_disable(struct drm_plane *plane,
+				     struct drm_plane_state *old_state)
 {
-	vc4_plane_set_primary_blank(plane, true);
-}
-
-static void vc4_cursor_plane_atomic_update(struct drm_plane *plane,
-					   struct drm_plane_state *old_state)
-{
-	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
+	//struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
 	struct drm_plane_state *state = plane->state;
-	struct vc4_crtc *vc4_crtc = to_vc4_crtc(state->crtc);
-	struct drm_framebuffer *fb = state->fb;
-	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
-	dma_addr_t addr = bo->paddr + fb->offsets[0];
-	int ret;
-	u32 packet_state[] = {
-		state->crtc->state->active,
-		state->crtc_x,
-		state->crtc_y,
-		0
-	};
-	WARN_ON_ONCE(fb->pitches[0] != state->crtc_w * 4);
+	struct vc4_fkms_plane *vc4_plane = to_vc4_fkms_plane(plane);
 
-	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] update %dx%d cursor at %d,%d (0x%pad/%d)",
+	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] plane disable %dx%d@%d +%d,%d\n",
 			 plane->base.id, plane->name,
 			 state->crtc_w,
 			 state->crtc_h,
+			 vc4_plane->mb.plane.vc_image_type,
 			 state->crtc_x,
-			 state->crtc_y,
-			 &addr,
-			 fb->pitches[0]);
-
-	/* add on the top/left offsets when overscan is active */
-	if (vc4_crtc) {
-		packet_state[1] += vc4_crtc->overscan[0];
-		packet_state[2] += vc4_crtc->overscan[1];
-	}
-
-	ret = rpi_firmware_property(vc4->firmware,
-				    RPI_FIRMWARE_SET_CURSOR_STATE,
-				    &packet_state,
-				    sizeof(packet_state));
-	if (ret || packet_state[0] != 0)
-		DRM_ERROR("Failed to set cursor state: 0x%08x\n", packet_state[0]);
-
-	/* Note: When the cursor contents change, the modesetting
-	 * driver calls drm_mode_cursor_univeral() with
-	 * DRM_MODE_CURSOR_BO, which means a new fb will be allocated.
-	 */
-	if (!old_state ||
-	    state->crtc_w != old_state->crtc_w ||
-	    state->crtc_h != old_state->crtc_h ||
-	    fb != old_state->fb) {
-		u32 packet_info[] = { state->crtc_w, state->crtc_h,
-				      0, /* unused */
-				      addr,
-				      0, 0, /* hotx, hoty */};
-
-		ret = rpi_firmware_property(vc4->firmware,
-					    RPI_FIRMWARE_SET_CURSOR_INFO,
-					    &packet_info,
-					    sizeof(packet_info));
-		if (ret || packet_info[0] != 0)
-			DRM_ERROR("Failed to set cursor info: 0x%08x\n", packet_info[0]);
-	}
-}
-
-static void vc4_cursor_plane_atomic_disable(struct drm_plane *plane,
-					    struct drm_plane_state *old_state)
-{
-	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
-	u32 packet_state[] = { false, 0, 0, 0 };
-	int ret;
-
-	DRM_DEBUG_ATOMIC("[PLANE:%d:%s] disabling cursor", plane->base.id, plane->name);
-
-	ret = rpi_firmware_property(vc4->firmware,
-				    RPI_FIRMWARE_SET_CURSOR_STATE,
-				    &packet_state,
-				    sizeof(packet_state));
-	if (ret || packet_state[0] != 0)
-		DRM_ERROR("Failed to set cursor state: 0x%08x\n", packet_state[0]);
+			 state->crtc_y);
+	vc4_plane_set_blank(plane, true);
 }
 
 static int vc4_plane_atomic_check(struct drm_plane *plane,
@@ -302,6 +432,7 @@ static bool vc4_fkms_format_mod_supporte
 	switch (format) {
 	case DRM_FORMAT_XRGB8888:
 	case DRM_FORMAT_ARGB8888:
+	case DRM_FORMAT_RGB565:
 		switch (modifier) {
 		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
 		case DRM_FORMAT_MOD_LINEAR:
@@ -310,8 +441,22 @@ static bool vc4_fkms_format_mod_supporte
 		default:
 			return false;
 		}
+	case DRM_FORMAT_NV12:
+	case DRM_FORMAT_NV21:
+		switch (fourcc_mod_broadcom_mod(modifier)) {
+		case DRM_FORMAT_MOD_LINEAR:
+		case DRM_FORMAT_MOD_BROADCOM_SAND128:
+			return true;
+		default:
+			return false;
+		}
+	case DRM_FORMAT_RGB888:
+	case DRM_FORMAT_BGR888:
+	case DRM_FORMAT_YUV422:
+	case DRM_FORMAT_YUV420:
+	case DRM_FORMAT_YVU420:
 	default:
-		return false;
+		return (modifier == DRM_FORMAT_MOD_LINEAR);
 	}
 }
 
@@ -326,31 +471,24 @@ static const struct drm_plane_funcs vc4_
 	.format_mod_supported = vc4_fkms_format_mod_supported,
 };
 
-static const struct drm_plane_helper_funcs vc4_primary_plane_helper_funcs = {
-	.prepare_fb = drm_gem_fb_prepare_fb,
-	.cleanup_fb = NULL,
-	.atomic_check = vc4_plane_atomic_check,
-	.atomic_update = vc4_primary_plane_atomic_update,
-	.atomic_disable = vc4_primary_plane_atomic_disable,
-};
-
-static const struct drm_plane_helper_funcs vc4_cursor_plane_helper_funcs = {
+static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
 	.prepare_fb = drm_gem_fb_prepare_fb,
 	.cleanup_fb = NULL,
 	.atomic_check = vc4_plane_atomic_check,
-	.atomic_update = vc4_cursor_plane_atomic_update,
-	.atomic_disable = vc4_cursor_plane_atomic_disable,
+	.atomic_update = vc4_plane_atomic_update,
+	.atomic_disable = vc4_plane_atomic_disable,
 };
 
 static struct drm_plane *vc4_fkms_plane_init(struct drm_device *dev,
-					     enum drm_plane_type type)
+					     enum drm_plane_type type,
+					     u8 plane_id)
 {
-	/* Primary and cursor planes only */
 	struct drm_plane *plane = NULL;
 	struct vc4_fkms_plane *vc4_plane;
-	u32 formats[] = {DRM_FORMAT_ARGB8888, DRM_FORMAT_XRGB8888};
+	u32 formats[ARRAY_SIZE(vc_image_formats)];
+	unsigned int default_zpos = 0;
+	u32 num_formats = 0;
 	int ret = 0;
-	bool primary = (type == DRM_PLANE_TYPE_PRIMARY);
 	static const uint64_t modifiers[] = {
 		DRM_FORMAT_MOD_LINEAR,
 		/* VC4_T_TILED should come after linear, because we
@@ -359,6 +497,7 @@ static struct drm_plane *vc4_fkms_plane_
 		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
 		DRM_FORMAT_MOD_INVALID,
 	};
+	int i;
 
 	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
 				 GFP_KERNEL);
@@ -367,19 +506,48 @@ static struct drm_plane *vc4_fkms_plane_
 		goto fail;
 	}
 
+	for (i = 0; i < ARRAY_SIZE(vc_image_formats); i++)
+		formats[num_formats++] = vc_image_formats[i].drm;
+
 	plane = &vc4_plane->base;
 	ret = drm_universal_plane_init(dev, plane, 0xff,
 				       &vc4_plane_funcs,
-				       formats, primary ? 2 : 1, modifiers,
-				       type, primary ? "primary" : "cursor");
+				       formats, num_formats, modifiers,
+				       type, NULL);
 
-	if (type == DRM_PLANE_TYPE_PRIMARY)
-		drm_plane_helper_add(plane, &vc4_primary_plane_helper_funcs);
-	else
-		drm_plane_helper_add(plane, &vc4_cursor_plane_helper_funcs);
+	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
 
 	drm_plane_create_alpha_property(plane);
 
+	/*
+	 * Default frame buffer setup is with FB on -127, and raspistill etc
+	 * tend to drop overlays on layer 2. Cursor plane was on layer +127.
+	 *
+	 * For F-KMS the mailbox call allows for a s8.
+	 * Remap zpos 0 to -127 for the background layer, but leave all the
+	 * other layers as requested by KMS.
+	 */
+	switch (type) {
+	case DRM_PLANE_TYPE_PRIMARY:
+		default_zpos = 0;
+		break;
+	case DRM_PLANE_TYPE_OVERLAY:
+		default_zpos = 1;
+		break;
+	case DRM_PLANE_TYPE_CURSOR:
+		default_zpos = 2;
+		break;
+	}
+	drm_plane_create_zpos_property(plane, default_zpos, 0, 127);
+
+	/* Prepare the static elements of the mailbox structure */
+	vc4_plane->mb.tag.tag = RPI_FIRMWARE_SET_PLANE;
+	vc4_plane->mb.tag.buf_size = sizeof(struct set_plane);
+	vc4_plane->mb.tag.req_resp_size = 0;
+	vc4_plane->mb.plane.display = 0;
+	vc4_plane->mb.plane.plane_id = plane_id;
+	vc4_plane->mb.plane.layer = default_zpos ? default_zpos : -127;
+
 	return plane;
 fail:
 	if (plane)
@@ -401,19 +569,23 @@ static void vc4_crtc_disable(struct drm_
 	 * whether anything scans out at all, but the firmware doesn't
 	 * give us a CRTC-level control for that.
 	 */
-	vc4_cursor_plane_atomic_disable(crtc->cursor, crtc->cursor->state);
-	vc4_plane_set_primary_blank(crtc->primary, true);
+
+	vc4_plane_atomic_disable(crtc->cursor, crtc->cursor->state);
+	vc4_plane_atomic_disable(crtc->primary, crtc->primary->state);
+
+	/* FIXME: Disable overlay planes */
 }
 
 static void vc4_crtc_enable(struct drm_crtc *crtc, struct drm_crtc_state *old_state)
 {
 	/* Unblank the planes (if they're supposed to be displayed). */
+
 	if (crtc->primary->state->fb)
-		vc4_plane_set_primary_blank(crtc->primary, false);
-	if (crtc->cursor->state->fb) {
-		vc4_cursor_plane_atomic_update(crtc->cursor,
-					       crtc->cursor->state);
-	}
+		vc4_plane_set_blank(crtc->primary, false);
+	if (crtc->cursor->state->fb)
+		vc4_plane_set_blank(crtc->cursor, crtc->cursor->state);
+
+	/* FIXME: Enable overlay planes */
 }
 
 static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
@@ -673,8 +845,10 @@ static int vc4_fkms_bind(struct device *
 	struct vc4_crtc *vc4_crtc;
 	struct vc4_fkms_encoder *vc4_encoder;
 	struct drm_crtc *crtc;
-	struct drm_plane *primary_plane, *cursor_plane, *destroy_plane, *temp;
+	struct drm_plane *primary_plane, *overlay_plane, *cursor_plane;
+	struct drm_plane *destroy_plane, *temp;
 	struct device_node *firmware_node;
+	u32 blank = 1;
 	int ret;
 
 	vc4->firmware_kms = true;
@@ -703,20 +877,26 @@ static int vc4_fkms_bind(struct device *
 	if (IS_ERR(vc4_crtc->regs))
 		return PTR_ERR(vc4_crtc->regs);
 
-	/* For now, we create just the primary and the legacy cursor
-	 * planes.  We should be able to stack more planes on easily,
-	 * but to do that we would need to compute the bandwidth
-	 * requirement of the plane configuration, and reject ones
-	 * that will take too much.
-	 */
-	primary_plane = vc4_fkms_plane_init(drm, DRM_PLANE_TYPE_PRIMARY);
+	/* Blank the firmware provided framebuffer */
+	rpi_firmware_property(vc4->firmware,
+			      RPI_FIRMWARE_FRAMEBUFFER_BLANK,
+			      &blank, sizeof(blank));
+
+	primary_plane = vc4_fkms_plane_init(drm, DRM_PLANE_TYPE_PRIMARY, 0);
 	if (IS_ERR(primary_plane)) {
 		dev_err(dev, "failed to construct primary plane\n");
 		ret = PTR_ERR(primary_plane);
 		goto err;
 	}
 
-	cursor_plane = vc4_fkms_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
+	overlay_plane = vc4_fkms_plane_init(drm, DRM_PLANE_TYPE_OVERLAY, 1);
+	if (IS_ERR(overlay_plane)) {
+		dev_err(dev, "failed to construct overlay plane\n");
+		ret = PTR_ERR(overlay_plane);
+		goto err;
+	}
+
+	cursor_plane = vc4_fkms_plane_init(drm, DRM_PLANE_TYPE_CURSOR, 2);
 	if (IS_ERR(cursor_plane)) {
 		dev_err(dev, "failed to construct cursor plane\n");
 		ret = PTR_ERR(cursor_plane);
--- a/drivers/gpu/drm/vc4/vc4_kms.c
+++ b/drivers/gpu/drm/vc4/vc4_kms.c
@@ -542,6 +542,7 @@ int vc4_kms_load(struct drm_device *dev)
 	dev->mode_config.preferred_depth = 24;
 	dev->mode_config.async_page_flip = true;
 	dev->mode_config.allow_fb_modifiers = true;
+	dev->mode_config.normalize_zpos = true;
 
 	drm_modeset_lock_init(&vc4->ctm_state_lock);
 
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc_image_types.h
@@ -0,0 +1,143 @@
+
+/*
+ * Copyright (c) 2012, Broadcom Europe Ltd
+ *
+ * Values taken from vc_image_types.h released by Broadcom at
+ * https://github.com/raspberrypi/userland/blob/master/interface/vctypes/vc_image_types.h
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+enum {
+	VC_IMAGE_MIN = 0, //bounds for error checking
+
+	VC_IMAGE_RGB565 = 1,
+	VC_IMAGE_1BPP,
+	VC_IMAGE_YUV420,
+	VC_IMAGE_48BPP,
+	VC_IMAGE_RGB888,
+	VC_IMAGE_8BPP,
+	/* 4bpp palettised image */
+	VC_IMAGE_4BPP,
+	/* A separated format of 16 colour/light shorts followed by 16 z
+	 * values
+	 */
+	VC_IMAGE_3D32,
+	/* 16 colours followed by 16 z values */
+	VC_IMAGE_3D32B,
+	/* A separated format of 16 material/colour/light shorts followed by
+	 * 16 z values
+	 */
+	VC_IMAGE_3D32MAT,
+	/* 32 bit format containing 18 bits of 6.6.6 RGB, 9 bits per short */
+	VC_IMAGE_RGB2X9,
+	/* 32-bit format holding 18 bits of 6.6.6 RGB */
+	VC_IMAGE_RGB666,
+	/* 4bpp palettised image with embedded palette */
+	VC_IMAGE_PAL4_OBSOLETE,
+	/* 8bpp palettised image with embedded palette */
+	VC_IMAGE_PAL8_OBSOLETE,
+	/* RGB888 with an alpha byte after each pixel */
+	VC_IMAGE_RGBA32,
+	/* a line of Y (32-byte padded), a line of U (16-byte padded), and a
+	 * line of V (16-byte padded)
+	 */
+	VC_IMAGE_YUV422,
+	/* RGB565 with a transparent patch */
+	VC_IMAGE_RGBA565,
+	/* Compressed (4444) version of RGBA32 */
+	VC_IMAGE_RGBA16,
+	/* VCIII codec format */
+	VC_IMAGE_YUV_UV,
+	/* VCIII T-format RGBA8888 */
+	VC_IMAGE_TF_RGBA32,
+	/* VCIII T-format RGBx8888 */
+	VC_IMAGE_TF_RGBX32,
+	/* VCIII T-format float */
+	VC_IMAGE_TF_FLOAT,
+	/* VCIII T-format RGBA4444 */
+	VC_IMAGE_TF_RGBA16,
+	/* VCIII T-format RGB5551 */
+	VC_IMAGE_TF_RGBA5551,
+	/* VCIII T-format RGB565 */
+	VC_IMAGE_TF_RGB565,
+	/* VCIII T-format 8-bit luma and 8-bit alpha */
+	VC_IMAGE_TF_YA88,
+	/* VCIII T-format 8 bit generic sample */
+	VC_IMAGE_TF_BYTE,
+	/* VCIII T-format 8-bit palette */
+	VC_IMAGE_TF_PAL8,
+	/* VCIII T-format 4-bit palette */
+	VC_IMAGE_TF_PAL4,
+	/* VCIII T-format Ericsson Texture Compressed */
+	VC_IMAGE_TF_ETC1,
+	/* RGB888 with R & B swapped */
+	VC_IMAGE_BGR888,
+	/* RGB888 with R & B swapped, but with no pitch, i.e. no padding after
+	 * each row of pixels
+	 */
+	VC_IMAGE_BGR888_NP,
+	/* Bayer image, extra defines which variant is being used */
+	VC_IMAGE_BAYER,
+	/* General wrapper for codec images e.g. JPEG from camera */
+	VC_IMAGE_CODEC,
+	/* VCIII codec format */
+	VC_IMAGE_YUV_UV32,
+	/* VCIII T-format 8-bit luma */
+	VC_IMAGE_TF_Y8,
+	/* VCIII T-format 8-bit alpha */
+	VC_IMAGE_TF_A8,
+	/* VCIII T-format 16-bit generic sample */
+	VC_IMAGE_TF_SHORT,
+	/* VCIII T-format 1bpp black/white */
+	VC_IMAGE_TF_1BPP,
+	VC_IMAGE_OPENGL,
+	/* VCIII-B0 HVS YUV 4:4:4 interleaved samples */
+	VC_IMAGE_YUV444I,
+	/* Y, U, & V planes separately (VC_IMAGE_YUV422 has them interleaved on
+	 * a per line basis)
+	 */
+	VC_IMAGE_YUV422PLANAR,
+	/* 32bpp with 8bit alpha at MS byte, with R, G, B (LS byte) */
+	VC_IMAGE_ARGB8888,
+	/* 32bpp with 8bit unused at MS byte, with R, G, B (LS byte) */
+	VC_IMAGE_XRGB8888,
+
+	/* interleaved 8 bit samples of Y, U, Y, V (4 flavours) */
+	VC_IMAGE_YUV422YUYV,
+	VC_IMAGE_YUV422YVYU,
+	VC_IMAGE_YUV422UYVY,
+	VC_IMAGE_YUV422VYUY,
+
+	/* 32bpp like RGBA32 but with unused alpha */
+	VC_IMAGE_RGBX32,
+	/* 32bpp, corresponding to RGBA with unused alpha */
+	VC_IMAGE_RGBX8888,
+	/* 32bpp, corresponding to BGRA with unused alpha */
+	VC_IMAGE_BGRX8888,
+
+	/* Y as a plane, then UV byte interleaved in plane with with same pitch,
+	 * half height
+	 */
+	VC_IMAGE_YUV420SP,
+
+	/* Y, U, & V planes separately 4:4:4 */
+	VC_IMAGE_YUV444PLANAR,
+
+	/* T-format 8-bit U - same as TF_Y8 buf from U plane */
+	VC_IMAGE_TF_U8,
+	/* T-format 8-bit U - same as TF_Y8 buf from V plane */
+	VC_IMAGE_TF_V8,
+
+	/* YUV4:2:0 planar, 16bit values */
+	VC_IMAGE_YUV420_16,
+	/* YUV4:2:0 codec format, 16bit values */
+	VC_IMAGE_YUV_UV_16,
+	/* YUV4:2:0 with U,V in side-by-side format */
+	VC_IMAGE_YUV420_S,
+
+	VC_IMAGE_MAX,     /* bounds for error checking */
+	VC_IMAGE_FORCE_ENUM_16BIT = 0xffff,
+};
--- a/include/soc/bcm2835/raspberrypi-firmware.h
+++ b/include/soc/bcm2835/raspberrypi-firmware.h
@@ -147,6 +147,8 @@ enum rpi_firmware_property_tag {
 
 	RPI_FIRMWARE_VCHIQ_INIT =                             0x00048010,
 
+	RPI_FIRMWARE_SET_PLANE =                              0x00048015,
+
 	RPI_FIRMWARE_GET_COMMAND_LINE =                       0x00050001,
 	RPI_FIRMWARE_GET_DMA_CHANNELS =                       0x00060001,
 };