Merge tag 'drm-vc4-next-2016-02-17' of github.com:anholt/linux into drm-next

This pull request brings in overlay plane support for vc4.

* tag 'drm-vc4-next-2016-02-17' of github.com:anholt/linux:
  drm/vc4: Add support for YUV planes.
  drm/vc4: Add support a few more RGB display plane formats.
  drm/vc4: Add support for scaling of display planes.
  drm/vc4: Fix which value is being used for source image size.
  drm/vc4: Add more display planes to each CRTC.
  drm/vc4: Make the CRTCs cooperate on allocating display lists.
  drm/vc4: Add a proper short-circut path for legacy cursor updates.
  drm/vc4: Move the plane clipping/scaling setup to a separate function.
  drm/vc4: Add missing __iomem annotation to hw_dlist.
  drm/vc4: Improve comments on vc4_plane_state members.

6 files changed, 872 insertions, 122 deletions

diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c
index 9374097..619dc78 100644
--- a/drivers/gpu/drm/vc4/vc4_crtc.c
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -49,22 +49,27 @@ struct vc4_crtc {
 	/* Which HVS channel we're using for our CRTC. */
 	int channel;
 
-	/* Pointer to the actual hardware display list memory for the
-	 * crtc.
-	 */
-	u32 __iomem *dlist;
-
-	u32 dlist_size; /* in dwords */
-
 	struct drm_pending_vblank_event *event;
 };
 
+struct vc4_crtc_state {
+	struct drm_crtc_state base;
+	/* Dlist area for this CRTC configuration. */
+	struct drm_mm_node mm;
+};
+
 static inline struct vc4_crtc *
 to_vc4_crtc(struct drm_crtc *crtc)
 {
 	return (struct vc4_crtc *)crtc;
 }
 
+static inline struct vc4_crtc_state *
+to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
+{
+	return (struct vc4_crtc_state *)crtc_state;
+}
+
 struct vc4_crtc_data {
 	/* Which channel of the HVS this pixelvalve sources from. */
 	int hvs_channel;
@@ -319,11 +324,13 @@ static void vc4_crtc_enable(struct drm_crtc *crtc)
 static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
 				 struct drm_crtc_state *state)
 {
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
 	struct drm_device *dev = crtc->dev;
 	struct vc4_dev *vc4 = to_vc4_dev(dev);
 	struct drm_plane *plane;
-	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	unsigned long flags;
 	u32 dlist_count = 0;
+	int ret;
 
 	/* The pixelvalve can only feed one encoder (and encoders are
 	 * 1:1 with connectors.)
@@ -346,18 +353,12 @@ static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
 
 	dlist_count++; /* Account for SCALER_CTL0_END. */
 
-	if (!vc4_crtc->dlist || dlist_count > vc4_crtc->dlist_size) {
-		vc4_crtc->dlist = ((u32 __iomem *)vc4->hvs->dlist +
-				   HVS_BOOTLOADER_DLIST_END);
-		vc4_crtc->dlist_size = ((SCALER_DLIST_SIZE >> 2) -
-					HVS_BOOTLOADER_DLIST_END);
-
-		if (dlist_count > vc4_crtc->dlist_size) {
-			DRM_DEBUG_KMS("dlist too large for CRTC (%d > %d).\n",
-				      dlist_count, vc4_crtc->dlist_size);
-			return -EINVAL;
-		}
-	}
+	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
+	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
+				 dlist_count, 1, 0);
+	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
+	if (ret)
+		return ret;
 
 	return 0;
 }
@@ -368,47 +369,29 @@ static void vc4_crtc_atomic_flush(struct drm_crtc *crtc,
 	struct drm_device *dev = crtc->dev;
 	struct vc4_dev *vc4 = to_vc4_dev(dev);
 	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
 	struct drm_plane *plane;
 	bool debug_dump_regs = false;
-	u32 __iomem *dlist_next = vc4_crtc->dlist;
+	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
+	u32 __iomem *dlist_next = dlist_start;
 
 	if (debug_dump_regs) {
 		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
 		vc4_hvs_dump_state(dev);
 	}
 
-	/* Copy all the active planes' dlist contents to the hardware dlist.
-	 *
-	 * XXX: If the new display list was large enough that it
-	 * overlapped a currently-read display list, we need to do
-	 * something like disable scanout before putting in the new
-	 * list.  For now, we're safe because we only have the two
-	 * planes.
-	 */
+	/* Copy all the active planes' dlist contents to the hardware dlist. */
 	drm_atomic_crtc_for_each_plane(plane, crtc) {
 		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
 	}
 
-	if (dlist_next == vc4_crtc->dlist) {
-		/* If no planes were enabled, use the SCALER_CTL0_END
-		 * at the start of the display list memory (in the
-		 * bootloader section).  We'll rewrite that
-		 * SCALER_CTL0_END, just in case, though.
-		 */
-		writel(SCALER_CTL0_END, vc4->hvs->dlist);
-		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel), 0);
-	} else {
-		writel(SCALER_CTL0_END, dlist_next);
-		dlist_next++;
-
-		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
-			  (u32 __iomem *)vc4_crtc->dlist -
-			  (u32 __iomem *)vc4->hvs->dlist);
-
-		/* Make the next display list start after ours. */
-		vc4_crtc->dlist_size -= (dlist_next - vc4_crtc->dlist);
-		vc4_crtc->dlist = dlist_next;
-	}
+	writel(SCALER_CTL0_END, dlist_next);
+	dlist_next++;
+
+	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
+
+	HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
+		  vc4_state->mm.start);
 
 	if (debug_dump_regs) {
 		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
@@ -573,6 +556,36 @@ static int vc4_page_flip(struct drm_crtc *crtc,
 		return drm_atomic_helper_page_flip(crtc, fb, event, flags);
 }
 
+static struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc)
+{
+	struct vc4_crtc_state *vc4_state;
+
+	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
+	if (!vc4_state)
+		return NULL;
+
+	__drm_atomic_helper_crtc_duplicate_state(crtc, &vc4_state->base);
+	return &vc4_state->base;
+}
+
+static void vc4_crtc_destroy_state(struct drm_crtc *crtc,
+				   struct drm_crtc_state *state)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(crtc->dev);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
+
+	if (vc4_state->mm.allocated) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
+		drm_mm_remove_node(&vc4_state->mm);
+		spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
+
+	}
+
+	__drm_atomic_helper_crtc_destroy_state(crtc, state);
+}
+
 static const struct drm_crtc_funcs vc4_crtc_funcs = {
 	.set_config = drm_atomic_helper_set_config,
 	.destroy = vc4_crtc_destroy,
@@ -581,8 +594,8 @@ static const struct drm_crtc_funcs vc4_crtc_funcs = {
 	.cursor_set = NULL, /* handled by drm_mode_cursor_universal */
 	.cursor_move = NULL, /* handled by drm_mode_cursor_universal */
 	.reset = drm_atomic_helper_crtc_reset,
-	.atomic_duplicate_state = drm_atomic_helper_crtc_duplicate_state,
-	.atomic_destroy_state = drm_atomic_helper_crtc_destroy_state,
+	.atomic_duplicate_state = vc4_crtc_duplicate_state,
+	.atomic_destroy_state = vc4_crtc_destroy_state,
 };
 
 static const struct drm_crtc_helper_funcs vc4_crtc_helper_funcs = {
@@ -644,9 +657,9 @@ static int vc4_crtc_bind(struct device *dev, struct device *master, void *data)
 	struct vc4_dev *vc4 = to_vc4_dev(drm);
 	struct vc4_crtc *vc4_crtc;
 	struct drm_crtc *crtc;
-	struct drm_plane *primary_plane, *cursor_plane;
+	struct drm_plane *primary_plane, *cursor_plane, *destroy_plane, *temp;
 	const struct of_device_id *match;
-	int ret;
+	int ret, i;
 
 	vc4_crtc = devm_kzalloc(dev, sizeof(*vc4_crtc), GFP_KERNEL);
 	if (!vc4_crtc)
@@ -675,27 +688,49 @@ static int vc4_crtc_bind(struct device *dev, struct device *master, void *data)
 		goto err;
 	}
 
-	cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
-	if (IS_ERR(cursor_plane)) {
-		dev_err(dev, "failed to construct cursor plane\n");
-		ret = PTR_ERR(cursor_plane);
-		goto err_primary;
-	}
-
-	drm_crtc_init_with_planes(drm, crtc, primary_plane, cursor_plane,
+	drm_crtc_init_with_planes(drm, crtc, primary_plane, NULL,
 				  &vc4_crtc_funcs, NULL);
 	drm_crtc_helper_add(crtc, &vc4_crtc_helper_funcs);
 	primary_plane->crtc = crtc;
-	cursor_plane->crtc = crtc;
 	vc4->crtc[drm_crtc_index(crtc)] = vc4_crtc;
 	vc4_crtc->channel = vc4_crtc->data->hvs_channel;
 
+	/* Set up some arbitrary number of planes.  We're not limited
+	 * by a set number of physical registers, just the space in
+	 * the HVS (16k) and how small an plane can be (28 bytes).
+	 * However, each plane we set up takes up some memory, and
+	 * increases the cost of looping over planes, which atomic
+	 * modesetting does quite a bit.  As a result, we pick a
+	 * modest number of planes to expose, that should hopefully
+	 * still cover any sane usecase.
+	 */
+	for (i = 0; i < 8; i++) {
+		struct drm_plane *plane =
+			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY);
+
+		if (IS_ERR(plane))
+			continue;
+
+		plane->possible_crtcs = 1 << drm_crtc_index(crtc);
+	}
+
+	/* Set up the legacy cursor after overlay initialization,
+	 * since we overlay planes on the CRTC in the order they were
+	 * initialized.
+	 */
+	cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
+	if (!IS_ERR(cursor_plane)) {
+		cursor_plane->possible_crtcs = 1 << drm_crtc_index(crtc);
+		cursor_plane->crtc = crtc;
+		crtc->cursor = cursor_plane;
+	}
+
 	CRTC_WRITE(PV_INTEN, 0);
 	CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
 	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
 			       vc4_crtc_irq_handler, 0, "vc4 crtc", vc4_crtc);
 	if (ret)
-		goto err_cursor;
+		goto err_destroy_planes;
 
 	vc4_set_crtc_possible_masks(drm, crtc);
 
@@ -703,10 +738,12 @@ static int vc4_crtc_bind(struct device *dev, struct device *master, void *data)
 
 	return 0;
 
-err_cursor:
-	cursor_plane->funcs->destroy(cursor_plane);
-err_primary:
-	primary_plane->funcs->destroy(primary_plane);
+err_destroy_planes:
+	list_for_each_entry_safe(destroy_plane, temp,
+				 &drm->mode_config.plane_list, head) {
+		if (destroy_plane->possible_crtcs == 1 << drm_crtc_index(crtc))
+		    destroy_plane->funcs->destroy(destroy_plane);
+	}
 err:
 	return ret;
 }
diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h
index 4c734d0..3d1df6b 100644
--- a/drivers/gpu/drm/vc4/vc4_drv.h
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -149,7 +149,17 @@ struct vc4_v3d {
 struct vc4_hvs {
 	struct platform_device *pdev;
 	void __iomem *regs;
-	void __iomem *dlist;
+	u32 __iomem *dlist;
+
+	/* Memory manager for CRTCs to allocate space in the display
+	 * list.  Units are dwords.
+	 */
+	struct drm_mm dlist_mm;
+	/* Memory manager for the LBM memory used by HVS scaling. */
+	struct drm_mm lbm_mm;
+	spinlock_t mm_lock;
+
+	struct drm_mm_node mitchell_netravali_filter;
 };
 
 struct vc4_plane {
diff --git a/drivers/gpu/drm/vc4/vc4_hvs.c b/drivers/gpu/drm/vc4/vc4_hvs.c
index 8098c5b..6fbab1c 100644
--- a/drivers/gpu/drm/vc4/vc4_hvs.c
+++ b/drivers/gpu/drm/vc4/vc4_hvs.c
@@ -100,12 +100,76 @@ int vc4_hvs_debugfs_regs(struct seq_file *m, void *unused)
 }
 #endif
 
+/* The filter kernel is composed of dwords each containing 3 9-bit
+ * signed integers packed next to each other.
+ */
+#define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
+#define VC4_PPF_FILTER_WORD(c0, c1, c2)				\
+	((((c0) & 0x1ff) << 0) |				\
+	 (((c1) & 0x1ff) << 9) |				\
+	 (((c2) & 0x1ff) << 18))
+
+/* The whole filter kernel is arranged as the coefficients 0-16 going
+ * up, then a pad, then 17-31 going down and reversed within the
+ * dwords.  This means that a linear phase kernel (where it's
+ * symmetrical at the boundary between 15 and 16) has the last 5
+ * dwords matching the first 5, but reversed.
+ */
+#define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,	\
+				c9, c10, c11, c12, c13, c14, c15)	\
+	{VC4_PPF_FILTER_WORD(c0, c1, c2),				\
+	 VC4_PPF_FILTER_WORD(c3, c4, c5),				\
+	 VC4_PPF_FILTER_WORD(c6, c7, c8),				\
+	 VC4_PPF_FILTER_WORD(c9, c10, c11),				\
+	 VC4_PPF_FILTER_WORD(c12, c13, c14),				\
+	 VC4_PPF_FILTER_WORD(c15, c15, 0)}
+
+#define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
+#define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
+
+/* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
+ * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
+ */
+static const u32 mitchell_netravali_1_3_1_3_kernel[] =
+	VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
+				50, 82, 119, 155, 187, 213, 227);
+
+static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
+					struct drm_mm_node *space,
+					const u32 *kernel)
+{
+	int ret, i;
+	u32 __iomem *dst_kernel;
+
+	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS, 1,
+				 0);
+	if (ret) {
+		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
+			  ret);
+		return ret;
+	}
+
+	dst_kernel = hvs->dlist + space->start;
+
+	for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
+		if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
+			writel(kernel[i], &dst_kernel[i]);
+		else {
+			writel(kernel[VC4_KERNEL_DWORDS - i - 1],
+			       &dst_kernel[i]);
+		}
+	}
+
+	return 0;
+}
+
 static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
 {
 	struct platform_device *pdev = to_platform_device(dev);
 	struct drm_device *drm = dev_get_drvdata(master);
 	struct vc4_dev *vc4 = drm->dev_private;
 	struct vc4_hvs *hvs = NULL;
+	int ret;
 
 	hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
 	if (!hvs)
@@ -119,6 +183,33 @@ static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
 
 	hvs->dlist = hvs->regs + SCALER_DLIST_START;
 
+	spin_lock_init(&hvs->mm_lock);
+
+	/* Set up the HVS display list memory manager.  We never
+	 * overwrite the setup from the bootloader (just 128b out of
+	 * our 16K), since we don't want to scramble the screen when
+	 * transitioning from the firmware's boot setup to runtime.
+	 */
+	drm_mm_init(&hvs->dlist_mm,
+		    HVS_BOOTLOADER_DLIST_END,
+		    (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
+
+	/* Set up the HVS LBM memory manager.  We could have some more
+	 * complicated data structure that allowed reuse of LBM areas
+	 * between planes when they don't overlap on the screen, but
+	 * for now we just allocate globally.
+	 */
+	drm_mm_init(&hvs->lbm_mm, 0, 96 * 1024);
+
+	/* Upload filter kernels.  We only have the one for now, so we
+	 * keep it around for the lifetime of the driver.
+	 */
+	ret = vc4_hvs_upload_linear_kernel(hvs,
+					   &hvs->mitchell_netravali_filter,
+					   mitchell_netravali_1_3_1_3_kernel);
+	if (ret)
+		return ret;
+
 	vc4->hvs = hvs;
 	return 0;
 }
@@ -129,6 +220,12 @@ static void vc4_hvs_unbind(struct device *dev, struct device *master,
 	struct drm_device *drm = dev_get_drvdata(master);
 	struct vc4_dev *vc4 = drm->dev_private;
 
+	if (vc4->hvs->mitchell_netravali_filter.allocated)
+		drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
+
+	drm_mm_takedown(&vc4->hvs->dlist_mm);
+	drm_mm_takedown(&vc4->hvs->lbm_mm);
+
 	vc4->hvs = NULL;
 }
 
diff --git a/drivers/gpu/drm/vc4/vc4_kms.c b/drivers/gpu/drm/vc4/vc4_kms.c
index f95f2df..4718ae5 100644
--- a/drivers/gpu/drm/vc4/vc4_kms.c
+++ b/drivers/gpu/drm/vc4/vc4_kms.c
@@ -49,6 +49,15 @@ vc4_atomic_complete_commit(struct vc4_commit *c)
 
 	drm_atomic_helper_commit_modeset_enables(dev, state);
 
+	/* Make sure that drm_atomic_helper_wait_for_vblanks()
+	 * actually waits for vblank.  If we're doing a full atomic
+	 * modeset (as opposed to a vc4_update_plane() short circuit),
+	 * then we need to wait for scanout to be done with our
+	 * display lists before we free it and potentially reallocate
+	 * and overwrite the dlist memory with a new modeset.
+	 */
+	state->legacy_cursor_update = false;
+
 	drm_atomic_helper_wait_for_vblanks(dev, state);
 
 	drm_atomic_helper_cleanup_planes(dev, state);
diff --git a/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c
index 0addbad..7b0c72a 100644
--- a/drivers/gpu/drm/vc4/vc4_plane.c
+++ b/drivers/gpu/drm/vc4/vc4_plane.c
@@ -24,19 +24,52 @@
 #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);
diff --git a/drivers/gpu/drm/vc4/vc4_regs.h b/drivers/gpu/drm/vc4/vc4_regs.h
index 4e52a0a..25df20e 100644
--- a/drivers/gpu/drm/vc4/vc4_regs.h
+++ b/drivers/gpu/drm/vc4/vc4_regs.h
@@ -503,7 +503,12 @@ enum hvs_pixel_format {
 	HVS_PIXEL_FORMAT_RGB888 = 5,
 	HVS_PIXEL_FORMAT_RGBA6666 = 6,
 	/* 32bpp */
-	HVS_PIXEL_FORMAT_RGBA8888 = 7
+	HVS_PIXEL_FORMAT_RGBA8888 = 7,
+
+	HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE = 8,
+	HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE = 9,
+	HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE = 10,
+	HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE = 11,
 };
 
 /* Note: the LSB is the rightmost character shown.  Only valid for
@@ -536,6 +541,21 @@ enum hvs_pixel_format {
 #define SCALER_CTL0_ORDER_MASK			VC4_MASK(14, 13)
 #define SCALER_CTL0_ORDER_SHIFT			13
 
+#define SCALER_CTL0_SCL1_MASK			VC4_MASK(10, 8)
+#define SCALER_CTL0_SCL1_SHIFT			8
+
+#define SCALER_CTL0_SCL0_MASK			VC4_MASK(7, 5)
+#define SCALER_CTL0_SCL0_SHIFT			5
+
+#define SCALER_CTL0_SCL_H_PPF_V_PPF		0
+#define SCALER_CTL0_SCL_H_TPZ_V_PPF		1
+#define SCALER_CTL0_SCL_H_PPF_V_TPZ		2
+#define SCALER_CTL0_SCL_H_TPZ_V_TPZ		3
+#define SCALER_CTL0_SCL_H_PPF_V_NONE		4
+#define SCALER_CTL0_SCL_H_NONE_V_PPF		5
+#define SCALER_CTL0_SCL_H_NONE_V_TPZ		6
+#define SCALER_CTL0_SCL_H_TPZ_V_NONE		7
+
 /* Set to indicate no scaling. */
 #define SCALER_CTL0_UNITY			BIT(4)
 
@@ -551,6 +571,12 @@ enum hvs_pixel_format {
 #define SCALER_POS0_START_X_MASK		VC4_MASK(11, 0)
 #define SCALER_POS0_START_X_SHIFT		0
 
+#define SCALER_POS1_SCL_HEIGHT_MASK		VC4_MASK(27, 16)
+#define SCALER_POS1_SCL_HEIGHT_SHIFT		16
+
+#define SCALER_POS1_SCL_WIDTH_MASK		VC4_MASK(11, 0)
+#define SCALER_POS1_SCL_WIDTH_SHIFT		0
+
 #define SCALER_POS2_ALPHA_MODE_MASK		VC4_MASK(31, 30)
 #define SCALER_POS2_ALPHA_MODE_SHIFT		30
 #define SCALER_POS2_ALPHA_MODE_PIPELINE		0
@@ -564,6 +590,80 @@ enum hvs_pixel_format {
 #define SCALER_POS2_WIDTH_MASK			VC4_MASK(11, 0)
 #define SCALER_POS2_WIDTH_SHIFT			0
 
+/* Color Space Conversion words.  Some values are S2.8 signed
+ * integers, except that the 2 integer bits map as {0x0: 0, 0x1: 1,
+ * 0x2: 2, 0x3: -1}
+ */
+/* bottom 8 bits of S2.8 contribution of Cr to Blue */
+#define SCALER_CSC0_COEF_CR_BLU_MASK		VC4_MASK(31, 24)
+#define SCALER_CSC0_COEF_CR_BLU_SHIFT		24
+/* Signed offset to apply to Y before CSC. (Y' = Y + YY_OFS) */
+#define SCALER_CSC0_COEF_YY_OFS_MASK		VC4_MASK(23, 16)
+#define SCALER_CSC0_COEF_YY_OFS_SHIFT		16
+/* Signed offset to apply to CB before CSC (Cb' = Cb - 128 + CB_OFS). */
+#define SCALER_CSC0_COEF_CB_OFS_MASK		VC4_MASK(15, 8)
+#define SCALER_CSC0_COEF_CB_OFS_SHIFT		8
+/* Signed offset to apply to CB before CSC (Cr' = Cr - 128 + CR_OFS). */
+#define SCALER_CSC0_COEF_CR_OFS_MASK		VC4_MASK(7, 0)
+#define SCALER_CSC0_COEF_CR_OFS_SHIFT		0
+#define SCALER_CSC0_ITR_R_601_5			0x00f00000
+#define SCALER_CSC0_ITR_R_709_3			0x00f00000
+#define SCALER_CSC0_JPEG_JFIF			0x00000000
+
+/* S2.8 contribution of Cb to Green */
+#define SCALER_CSC1_COEF_CB_GRN_MASK		VC4_MASK(31, 22)
+#define SCALER_CSC1_COEF_CB_GRN_SHIFT		22
+/* S2.8 contribution of Cr to Green */
+#define SCALER_CSC1_COEF_CR_GRN_MASK		VC4_MASK(21, 12)
+#define SCALER_CSC1_COEF_CR_GRN_SHIFT		12
+/* S2.8 contribution of Y to all of RGB */
+#define SCALER_CSC1_COEF_YY_ALL_MASK		VC4_MASK(11, 2)
+#define SCALER_CSC1_COEF_YY_ALL_SHIFT		2
+/* top 2 bits of S2.8 contribution of Cr to Blue */
+#define SCALER_CSC1_COEF_CR_BLU_MASK		VC4_MASK(1, 0)
+#define SCALER_CSC1_COEF_CR_BLU_SHIFT		0
+#define SCALER_CSC1_ITR_R_601_5			0xe73304a8
+#define SCALER_CSC1_ITR_R_709_3			0xf2b784a8
+#define SCALER_CSC1_JPEG_JFIF			0xea34a400
+
+/* S2.8 contribution of Cb to Red */
+#define SCALER_CSC2_COEF_CB_RED_MASK		VC4_MASK(29, 20)
+#define SCALER_CSC2_COEF_CB_RED_SHIFT		20
+/* S2.8 contribution of Cr to Red */
+#define SCALER_CSC2_COEF_CR_RED_MASK		VC4_MASK(19, 10)
+#define SCALER_CSC2_COEF_CR_RED_SHIFT		10
+/* S2.8 contribution of Cb to Blue */
+#define SCALER_CSC2_COEF_CB_BLU_MASK		VC4_MASK(19, 10)
+#define SCALER_CSC2_COEF_CB_BLU_SHIFT		10
+#define SCALER_CSC2_ITR_R_601_5			0x00066204
+#define SCALER_CSC2_ITR_R_709_3			0x00072a1c
+#define SCALER_CSC2_JPEG_JFIF			0x000599c5
+
+#define SCALER_TPZ0_VERT_RECALC			BIT(31)
+#define SCALER_TPZ0_SCALE_MASK			VC4_MASK(28, 8)
+#define SCALER_TPZ0_SCALE_SHIFT			8
+#define SCALER_TPZ0_IPHASE_MASK			VC4_MASK(7, 0)
+#define SCALER_TPZ0_IPHASE_SHIFT		0
+#define SCALER_TPZ1_RECIP_MASK			VC4_MASK(15, 0)
+#define SCALER_TPZ1_RECIP_SHIFT			0
+
+/* Skips interpolating coefficients to 64 phases, so just 8 are used.
+ * Required for nearest neighbor.
+ */
+#define SCALER_PPF_NOINTERP			BIT(31)
+/* Replaes the highest valued coefficient with one that makes all 4
+ * sum to unity.
+ */
+#define SCALER_PPF_AGC				BIT(30)
+#define SCALER_PPF_SCALE_MASK			VC4_MASK(24, 8)
+#define SCALER_PPF_SCALE_SHIFT			8
+#define SCALER_PPF_IPHASE_MASK			VC4_MASK(6, 0)
+#define SCALER_PPF_IPHASE_SHIFT			0
+
+#define SCALER_PPF_KERNEL_OFFSET_MASK		VC4_MASK(13, 0)
+#define SCALER_PPF_KERNEL_OFFSET_SHIFT		0
+#define SCALER_PPF_KERNEL_UNCACHED		BIT(31)
+
 #define SCALER_SRC_PITCH_MASK			VC4_MASK(15, 0)
 #define SCALER_SRC_PITCH_SHIFT			0