video: tegra: dc: Separate allocations for U and V

Currently, dc_ext only takes a single nvmap memory ID per overlay, even
in the YUV case (the U and V planes are expected to be differentiated
using an offset from the beginning of the nvmap allocation).

This is problematic for some software flows, such as certain video
interlacing algorithms that will vary the luma plane while keeping the
chrome plane constant.

This change allows dc_ext clients to specify a different nvmap
allocation for each of the Y, U, and V planes.  If a YUV surface is
used and no U or V plane allocation is specified, the old behavior is
preserved: the U and V offsets are assumed to be within the same
allocation as Y.

Note: this changes the behavior of the offset parameter: the old code
added offset to offset_u and offset_v when using it.  The new code
treats all three offsets as relative to the beginning of the allocation.
It also fixes a bug in the code where offset was applied twice to the Y
plane.  I believe this is safe because the presence of this bug means
that no existing clients are using offset != 0 (or if they are, they're
already broken).

Signed-off-by: Robert Morell <rmorell@nvidia.com>

Bug 850882

Original-Change-Id: I230e03db25baaae73a3bdc0d45a2aec162b87fa4
Reviewed-on: http://git-master/r/41471
Reviewed-by: Varun Colbert <vcolbert@nvidia.com>
Tested-by: Varun Colbert <vcolbert@nvidia.com>

Rebase-Id: Ra6dd17a50de7150edf104d2a6c9b3b9949919022

7 files changed, 85 insertions, 34 deletions

diff --git a/arch/arm/mach-tegra/include/mach/dc.h b/arch/arm/mach-tegra/include/mach/dc.h
index ec24c3277260..5aac57586e70 100644
--- a/arch/arm/mach-tegra/include/mach/dc.h
+++ b/arch/arm/mach-tegra/include/mach/dc.h
@@ -370,9 +370,8 @@ struct tegra_dc_win {
 
 	void			*virt_addr;
 	dma_addr_t		phys_addr;
-	unsigned		offset;
-	unsigned		offset_u;
-	unsigned		offset_v;
+	dma_addr_t		phys_addr_u;
+	dma_addr_t		phys_addr_v;
 	unsigned		stride;
 	unsigned		stride_uv;
 	fixed20_12		x;
diff --git a/drivers/video/tegra/dc/dc.c b/drivers/video/tegra/dc/dc.c
index c996a38d1f74..50b761c4389f 100644
--- a/drivers/video/tegra/dc/dc.c
+++ b/drivers/video/tegra/dc/dc.c
@@ -1030,20 +1030,17 @@ int tegra_dc_update_windows(struct tegra_dc_win *windows[], int n)
 		tegra_dc_writel(dc, 0, DC_WIN_BUF_STRIDE);
 		tegra_dc_writel(dc, 0, DC_WIN_UV_BUF_STRIDE);
 		tegra_dc_writel(dc,
-				(unsigned long)win->phys_addr +
-				(unsigned long)win->offset,
+				(unsigned long)win->phys_addr,
 				DC_WINBUF_START_ADDR);
 
 		if (!yuvp) {
 			tegra_dc_writel(dc, win->stride, DC_WIN_LINE_STRIDE);
 		} else {
 			tegra_dc_writel(dc,
-					(unsigned long)win->phys_addr +
-					(unsigned long)win->offset_u,
+					(unsigned long)win->phys_addr_u,
 					DC_WINBUF_START_ADDR_U);
 			tegra_dc_writel(dc,
-					(unsigned long)win->phys_addr +
-					(unsigned long)win->offset_v,
+					(unsigned long)win->phys_addr_v,
 					DC_WINBUF_START_ADDR_V);
 			tegra_dc_writel(dc,
 					LINE_STRIDE(win->stride) |
diff --git a/drivers/video/tegra/dc/ext/dev.c b/drivers/video/tegra/dc/ext/dev.c
index 6e1110725aa0..01454f5a8235 100644
--- a/drivers/video/tegra/dc/ext/dev.c
+++ b/drivers/video/tegra/dc/ext/dev.c
@@ -44,9 +44,10 @@ static int head_count;
 
 struct tegra_dc_ext_flip_win {
 	struct tegra_dc_ext_flip_windowattr	attr;
-	struct nvmap_handle_ref			*handle;
-	/* ugh. is this really necessary */
+	struct nvmap_handle_ref			*handle[TEGRA_DC_NUM_PLANES];
 	dma_addr_t				phys_addr;
+	dma_addr_t				phys_addr_u;
+	dma_addr_t				phys_addr_v;
 	u32					syncpt_max;
 };
 
@@ -177,9 +178,9 @@ static int tegra_dc_ext_set_windowattr(struct tegra_dc_ext *ext,
 {
 	struct tegra_dc_ext_win *ext_win = &ext->win[win->idx];
 
-	if (flip_win->handle == NULL) {
+	if (flip_win->handle[TEGRA_DC_Y] == NULL) {
 		win->flags = 0;
-		ext_win->cur_handle = NULL;
+		memset(ext_win->cur_handle, 0, sizeof(ext_win->cur_handle));
 		return 0;
 	}
 
@@ -199,12 +200,20 @@ static int tegra_dc_ext_set_windowattr(struct tegra_dc_ext *ext,
 	win->out_w = flip_win->attr.out_w;
 	win->out_h = flip_win->attr.out_h;
 	win->z = flip_win->attr.z;
-	ext_win->cur_handle = flip_win->handle;
+	memcpy(ext_win->cur_handle, flip_win->handle,
+	       sizeof(ext_win->cur_handle));
 
 	/* XXX verify that this won't read outside of the surface */
 	win->phys_addr = flip_win->phys_addr + flip_win->attr.offset;
-	win->offset_u = flip_win->attr.offset_u + flip_win->attr.offset;
-	win->offset_v = flip_win->attr.offset_v + flip_win->attr.offset;
+
+	win->phys_addr_u = flip_win->handle[TEGRA_DC_U] ?
+		flip_win->phys_addr_u : flip_win->phys_addr;
+	win->phys_addr_u += flip_win->attr.offset_u;
+
+	win->phys_addr_v = flip_win->handle[TEGRA_DC_V] ?
+		flip_win->phys_addr_v : flip_win->phys_addr;
+	win->phys_addr_v += flip_win->attr.offset_v;
+
 	win->stride = flip_win->attr.stride;
 	win->stride_uv = flip_win->attr.stride_uv;
 
@@ -225,7 +234,8 @@ static void tegra_dc_ext_flip_worker(struct work_struct *work)
 		container_of(work, struct tegra_dc_ext_flip_data, work);
 	struct tegra_dc_ext *ext = data->ext;
 	struct tegra_dc_win *wins[DC_N_WINDOWS];
-	struct nvmap_handle_ref *unpin_handles[DC_N_WINDOWS];
+	struct nvmap_handle_ref *unpin_handles[DC_N_WINDOWS *
+					       TEGRA_DC_NUM_PLANES];
 	int i, nr_unpin = 0, nr_win = 0;
 
 	for (i = 0; i < DC_N_WINDOWS; i++) {
@@ -240,9 +250,16 @@ static void tegra_dc_ext_flip_worker(struct work_struct *work)
 		win = tegra_dc_get_window(ext->dc, index);
 		ext_win = &ext->win[index];
 
-		if ((win->flags & TEGRA_WIN_FLAG_ENABLED) &&
-		    ext_win->cur_handle)
-			unpin_handles[nr_unpin++] = ext_win->cur_handle;
+		if (win->flags & TEGRA_WIN_FLAG_ENABLED) {
+			int j;
+			for (j = 0; j < TEGRA_DC_NUM_PLANES; j++) {
+				if (!ext_win->cur_handle[j])
+					continue;
+
+				unpin_handles[nr_unpin++] =
+					ext_win->cur_handle[j];
+			}
+		}
 
 		tegra_dc_ext_set_windowattr(ext, win, &data->win[i]);
 
@@ -383,10 +400,34 @@ static int tegra_dc_ext_flip(struct tegra_dc_ext_user *user,
 			continue;
 
 		ret = tegra_dc_ext_pin_window(user, flip_win->attr.buff_id,
-					      &flip_win->handle,
+					      &flip_win->handle[TEGRA_DC_Y],
 					      &flip_win->phys_addr);
 		if (ret)
 			goto fail_pin;
+
+		if (flip_win->attr.buff_id_u) {
+			ret = tegra_dc_ext_pin_window(user,
+					      flip_win->attr.buff_id_u,
+					      &flip_win->handle[TEGRA_DC_U],
+					      &flip_win->phys_addr_u);
+			if (ret)
+				goto fail_pin;
+		} else {
+			flip_win->handle[TEGRA_DC_U] = NULL;
+			flip_win->phys_addr_u = 0;
+		}
+
+		if (flip_win->attr.buff_id_v) {
+			ret = tegra_dc_ext_pin_window(user,
+					      flip_win->attr.buff_id_v,
+					      &flip_win->handle[TEGRA_DC_V],
+					      &flip_win->phys_addr_v);
+			if (ret)
+				goto fail_pin;
+		} else {
+			flip_win->handle[TEGRA_DC_V] = NULL;
+			flip_win->phys_addr_v = 0;
+		}
 	}
 
 	ret = lock_windows_for_flip(user, args);
@@ -427,12 +468,15 @@ unlock:
 	unlock_windows_for_flip(user, args);
 
 fail_pin:
-	while (i--) {
-		if (!data->win[i].handle)
-			continue;
+	for (i = 0; i < DC_N_WINDOWS; i++) {
+		int j;
+		for (j = 0; j < TEGRA_DC_NUM_PLANES; j++) {
+			if (!data->win[i].handle[j])
+				continue;
 
-		nvmap_unpin(ext->nvmap, data->win[i].handle);
-		nvmap_free(ext->nvmap, data->win[i].handle);
+			nvmap_unpin(ext->nvmap, data->win[i].handle[j]);
+			nvmap_free(ext->nvmap, data->win[i].handle[j]);
+		}
 	}
 	kfree(data);
 
diff --git a/drivers/video/tegra/dc/ext/tegra_dc_ext_priv.h b/drivers/video/tegra/dc/ext/tegra_dc_ext_priv.h
index 6f0d0e4e732b..54a10b2c8682 100644
--- a/drivers/video/tegra/dc/ext/tegra_dc_ext_priv.h
+++ b/drivers/video/tegra/dc/ext/tegra_dc_ext_priv.h
@@ -36,6 +36,13 @@ struct tegra_dc_ext_user {
 	struct nvmap_client	*nvmap;
 };
 
+enum {
+	TEGRA_DC_Y,
+	TEGRA_DC_U,
+	TEGRA_DC_V,
+	TEGRA_DC_NUM_PLANES,
+};
+
 struct tegra_dc_ext_win {
 	struct tegra_dc_ext	*ext;
 
@@ -45,7 +52,8 @@ struct tegra_dc_ext_win {
 
 	struct mutex		lock;
 
-	struct nvmap_handle_ref	*cur_handle;
+	/* Current nvmap handle (if any) for Y, U, V planes */
+	struct nvmap_handle_ref	*cur_handle[TEGRA_DC_NUM_PLANES];
 
 	struct workqueue_struct	*flip_wq;
 };
diff --git a/drivers/video/tegra/dc/overlay.c b/drivers/video/tegra/dc/overlay.c
index 623c26ae98d4..057241e86c81 100644
--- a/drivers/video/tegra/dc/overlay.c
+++ b/drivers/video/tegra/dc/overlay.c
@@ -213,8 +213,8 @@ static int tegra_overlay_set_windowattr(struct tegra_overlay_info *overlay,
 
 	/* STOPSHIP verify that this won't read outside of the surface */
 	win->phys_addr = flip_win->phys_addr + flip_win->attr.offset;
-	win->offset_u = flip_win->attr.offset_u + flip_win->attr.offset;
-	win->offset_v = flip_win->attr.offset_v + flip_win->attr.offset;
+	win->phys_addr_u = flip_win->phys_addr + flip_win->attr.offset_u;
+	win->phys_addr_v = flip_win->phys_addr + flip_win->attr.offset_v;
 	win->stride = flip_win->attr.stride;
 	win->stride_uv = flip_win->attr.stride_uv;
 
diff --git a/drivers/video/tegra/fb.c b/drivers/video/tegra/fb.c
index b7acecc8f562..b170bcdcba2c 100644
--- a/drivers/video/tegra/fb.c
+++ b/drivers/video/tegra/fb.c
@@ -111,8 +111,8 @@ static int tegra_fb_set_par(struct fb_info *info)
 		tegra_fb->win->stride = round_up(info->fix.line_length,
 						TEGRA_LINEAR_PITCH_ALIGNMENT);
 		tegra_fb->win->stride_uv = 0;
-		tegra_fb->win->offset_u = 0;
-		tegra_fb->win->offset_v = 0;
+		tegra_fb->win->phys_addr_u = 0;
+		tegra_fb->win->phys_addr_v = 0;
 	}
 
 	if (var->pixclock) {
@@ -436,8 +436,8 @@ struct tegra_fb_info *tegra_fb_register(struct nvhost_device *ndev,
 	win->z = 0;
 	win->phys_addr = fb_phys;
 	win->virt_addr = fb_base;
-	win->offset_u = 0;
-	win->offset_v = 0;
+	win->phys_addr_u = 0;
+	win->phys_addr_v = 0;
 	win->stride = fb_data->xres * fb_data->bits_per_pixel / 8;
 	/* Pad the stride to 16-byte boundary. */
 	win->stride = round_up(win->stride, TEGRA_LINEAR_PITCH_ALIGNMENT);
diff --git a/include/video/tegra_dc_ext.h b/include/video/tegra_dc_ext.h
index 118184849d1b..824afeaf147b 100644
--- a/include/video/tegra_dc_ext.h
+++ b/include/video/tegra_dc_ext.h
@@ -82,8 +82,11 @@ struct tegra_dc_ext_flip_windowattr {
 	struct timespec timestamp;
 	__u32	pre_syncpt_id;
 	__u32	pre_syncpt_val;
+	/* These are optional; if zero, U and V are taken from buff_id */
+	__u32	buff_id_u;
+	__u32	buff_id_v;
 	/* Leave some wiggle room for future expansion */
-	__u32   pad[8];
+	__u32   pad[6];
 };
 
 #define TEGRA_DC_EXT_FLIP_N_WINDOWS	3