video: tegra: Allow fractional input rects

This change makes the input rect for Tegra windows be a 20.12
fixed-point number instead of an integer.  This allows software to
specify sub-pixel precision.

bug 818525

Change-Id: I130f63b68159ed896d1113ea537307997875ca40
Signed-off-by: Robert Morell <rmorell@nvidia.com>
Reviewed-on: http://git-master/r/40526
Reviewed-by: Varun Colbert <vcolbert@nvidia.com>
Tested-by: Varun Colbert <vcolbert@nvidia.com>

6 files changed, 101 insertions, 37 deletions

diff --git a/arch/arm/mach-tegra/include/mach/dc.h b/arch/arm/mach-tegra/include/mach/dc.h
index 848711feccbb..f9378a46be42 100644
--- a/arch/arm/mach-tegra/include/mach/dc.h
+++ b/arch/arm/mach-tegra/include/mach/dc.h
@@ -24,6 +24,7 @@
 
 #include <linux/pm.h>
 #include <linux/types.h>
+#include <drm/drm_fixed.h>
 
 #define TEGRA_MAX_DC		2
 #define DC_N_WINDOWS		3
@@ -360,10 +361,10 @@ struct tegra_dc_win {
 	unsigned		offset_v;
 	unsigned		stride;
 	unsigned		stride_uv;
-	unsigned		x;
-	unsigned		y;
-	unsigned		w;
-	unsigned		h;
+	fixed20_12		x;
+	fixed20_12		y;
+	fixed20_12		w;
+	fixed20_12		h;
 	unsigned		out_x;
 	unsigned		out_y;
 	unsigned		out_w;
diff --git a/drivers/video/tegra/dc/dc.c b/drivers/video/tegra/dc/dc.c
index 6d2c268d9647..cc47993e1b35 100644
--- a/drivers/video/tegra/dc/dc.c
+++ b/drivers/video/tegra/dc/dc.c
@@ -34,6 +34,7 @@
 #include <linux/seq_file.h>
 #include <linux/backlight.h>
 #include <linux/switch.h>
+#include <drm/drm_fixed.h>
 
 #include <mach/clk.h>
 #include <mach/dc.h>
@@ -83,12 +84,12 @@ static const struct {
 static inline bool win_use_v_filter(const struct tegra_dc_win *win)
 {
 	return can_filter[win->idx].v &&
-		win->h != win->out_h;
+		win->h.full != dfixed_const(win->out_h);
 }
 static inline bool win_use_h_filter(const struct tegra_dc_win *win)
 {
 	return can_filter[win->idx].h &&
-		win->w != win->out_w;
+		win->w.full != dfixed_const(win->out_w);
 }
 
 static inline int tegra_dc_fmt_bpp(int fmt)
@@ -776,7 +777,8 @@ static unsigned long tegra_dc_calc_win_bandwidth(struct tegra_dc *dc,
 	if (!WIN_IS_ENABLED(w))
 		return 0;
 
-	if (w->w == 0 || w->h == 0 || w->out_w == 0 || w->out_h == 0)
+	if (dfixed_trunc(w->w) == 0 || dfixed_trunc(w->h) == 0 ||
+	    w->out_w == 0 || w->out_h == 0)
 		return 0;
 
 	tiled_windows_bw_multiplier =
@@ -791,7 +793,7 @@ static unsigned long tegra_dc_calc_win_bandwidth(struct tegra_dc *dc,
 	 * to prevent overflow of long. */
 	ret = (unsigned long)(dc->pixel_clk >> 16) *
 		bpp / 8 *
-		(win_use_v_filter(w) ? 2 : 1) * w->w / w->out_w *
+		(win_use_v_filter(w) ? 2 : 1) * dfixed_trunc(w->w) / w->out_w *
 		(WIN_IS_TILED(w) ? tiled_windows_bw_multiplier : 1);
 
 /*
@@ -864,6 +866,53 @@ static int tegra_dc_set_dynamic_emc(struct tegra_dc_win *windows[], int n)
 	return 0;
 }
 
+static inline u32 compute_dda_inc(fixed20_12 in, unsigned out_int,
+				  bool v, unsigned Bpp)
+{
+	/*
+	 * min(round((prescaled_size_in_pixels - 1) * 0x1000 /
+	 *	     (post_scaled_size_in_pixels - 1)), MAX)
+	 * Where the value of MAX is as follows:
+	 * For V_DDA_INCREMENT: 15.0 (0xF000)
+	 * For H_DDA_INCREMENT:  4.0 (0x4000) for 4 Bytes/pix formats.
+	 *			 8.0 (0x8000) for 2 Bytes/pix formats.
+	 */
+
+	fixed20_12 out = dfixed_init(out_int);
+	u32 dda_inc;
+	int max;
+
+	if (v) {
+		max = 15;
+	} else {
+		switch (Bpp) {
+		default:
+			WARN_ON_ONCE(1);
+			/* fallthrough */
+		case 4:
+			max = 4;
+			break;
+		case 2:
+			max = 8;
+			break;
+		}
+	}
+
+	out.full = max_t(u32, out.full - dfixed_const(1), dfixed_const(1));
+	in.full -= dfixed_const(1);
+
+	dda_inc = dfixed_div(in, out);
+
+	dda_inc = min_t(u32, dda_inc, dfixed_const(max));
+
+	return dda_inc;
+}
+
+static inline u32 compute_initial_dda(fixed20_12 in)
+{
+	return dfixed_frac(in);
+}
+
 /* does not support updating windows on multiple dcs in one call */
 int tegra_dc_update_windows(struct tegra_dc_win *windows[], int n)
 {
@@ -891,11 +940,13 @@ int tegra_dc_update_windows(struct tegra_dc_win *windows[], int n)
 		struct tegra_dc_win *win = windows[i];
 		unsigned h_dda;
 		unsigned v_dda;
-		unsigned h_offset;
-		unsigned v_offset;
+		fixed20_12 h_offset, v_offset;
 		bool invert_h = (win->flags & TEGRA_WIN_FLAG_INVERT_H) != 0;
 		bool invert_v = (win->flags & TEGRA_WIN_FLAG_INVERT_V) != 0;
 		bool yuvp = tegra_dc_is_yuv_planar(win->fmt);
+		unsigned Bpp = tegra_dc_fmt_bpp(win->fmt) / 8;
+		/* Bytes per pixel of bandwidth, used for dda_inc calculation */
+		unsigned Bpp_bw = Bpp * (yuvp ? 2 : 1);
 		const bool filter_h = win_use_h_filter(win);
 		const bool filter_v = win_use_v_filter(win);
 
@@ -931,16 +982,18 @@ int tegra_dc_update_windows(struct tegra_dc_win *windows[], int n)
 				V_SIZE(win->out_h) | H_SIZE(win->out_w),
 				DC_WIN_SIZE);
 		tegra_dc_writel(dc,
-				V_PRESCALED_SIZE(win->h) |
-				H_PRESCALED_SIZE(win->w * tegra_dc_fmt_bpp(win->fmt) / 8),
+				V_PRESCALED_SIZE(dfixed_trunc(win->h)) |
+				H_PRESCALED_SIZE(dfixed_trunc(win->w) * Bpp),
 				DC_WIN_PRESCALED_SIZE);
 
-		h_dda = ((win->w - 1) * 0x1000) / max_t(int, win->out_w - 1, 1);
-		v_dda = ((win->h - 1) * 0x1000) / max_t(int, win->out_h - 1, 1);
+		h_dda = compute_dda_inc(win->w, win->out_w, false, Bpp_bw);
+		v_dda = compute_dda_inc(win->h, win->out_h, true, Bpp_bw);
 		tegra_dc_writel(dc, V_DDA_INC(v_dda) | H_DDA_INC(h_dda),
 				DC_WIN_DDA_INCREMENT);
-		tegra_dc_writel(dc, 0, DC_WIN_H_INITIAL_DDA);
-		tegra_dc_writel(dc, 0, DC_WIN_V_INITIAL_DDA);
+		h_dda = compute_initial_dda(win->x);
+		v_dda = compute_initial_dda(win->y);
+		tegra_dc_writel(dc, h_dda, DC_WIN_H_INITIAL_DDA);
+		tegra_dc_writel(dc, v_dda, DC_WIN_V_INITIAL_DDA);
 
 		tegra_dc_writel(dc, 0, DC_WIN_BUF_STRIDE);
 		tegra_dc_writel(dc, 0, DC_WIN_UV_BUF_STRIDE);
@@ -968,17 +1021,18 @@ int tegra_dc_update_windows(struct tegra_dc_win *windows[], int n)
 
 		h_offset = win->x;
 		if (invert_h) {
-			h_offset += win->w - 1;
+			h_offset.full += win->w.full - dfixed_const(1);
 		}
-		h_offset *= tegra_dc_fmt_bpp(win->fmt) / 8;
 
 		v_offset = win->y;
 		if (invert_v) {
-			v_offset += win->h - 1;
+			v_offset.full += win->h.full - dfixed_const(1);
 		}
 
-		tegra_dc_writel(dc, h_offset, DC_WINBUF_ADDR_H_OFFSET);
-		tegra_dc_writel(dc, v_offset, DC_WINBUF_ADDR_V_OFFSET);
+		tegra_dc_writel(dc, dfixed_trunc(h_offset) * Bpp,
+				DC_WINBUF_ADDR_H_OFFSET);
+		tegra_dc_writel(dc, dfixed_trunc(v_offset),
+				DC_WINBUF_ADDR_V_OFFSET);
 
 		if (WIN_IS_TILED(win))
 			tegra_dc_writel(dc,
diff --git a/drivers/video/tegra/dc/ext/dev.c b/drivers/video/tegra/dc/ext/dev.c
index 05569ea74d0a..788593ea0dcf 100644
--- a/drivers/video/tegra/dc/ext/dev.c
+++ b/drivers/video/tegra/dc/ext/dev.c
@@ -189,10 +189,10 @@ static int tegra_dc_ext_set_windowattr(struct tegra_dc_ext *ext,
 	else if (flip_win->attr.blend == TEGRA_DC_EXT_BLEND_COVERAGE)
 		win->flags |= TEGRA_WIN_FLAG_BLEND_COVERAGE;
 	win->fmt = flip_win->attr.pixformat;
-	win->x = flip_win->attr.x;
-	win->y = flip_win->attr.y;
-	win->w = flip_win->attr.w;
-	win->h = flip_win->attr.h;
+	win->x.full = flip_win->attr.x;
+	win->y.full = flip_win->attr.y;
+	win->w.full = flip_win->attr.w;
+	win->h.full = flip_win->attr.h;
 	/* XXX verify that this doesn't go outside display's active region */
 	win->out_x = flip_win->attr.out_x;
 	win->out_y = flip_win->attr.out_y;
diff --git a/drivers/video/tegra/dc/overlay.c b/drivers/video/tegra/dc/overlay.c
index 3444b2f9fa34..6f42242a7102 100644
--- a/drivers/video/tegra/dc/overlay.c
+++ b/drivers/video/tegra/dc/overlay.c
@@ -27,6 +27,7 @@
 #include <linux/spinlock.h>
 #include <linux/tegra_overlay.h>
 #include <linux/uaccess.h>
+#include <drm/drm_fixed.h>
 
 #include <asm/atomic.h>
 
@@ -159,10 +160,10 @@ static int tegra_overlay_set_windowattr(struct tegra_overlay_info *overlay,
 		win->flags |= TEGRA_WIN_FLAG_TILED;
 
 	win->fmt = flip_win->attr.pixformat;
-	win->x = flip_win->attr.x;
-	win->y = flip_win->attr.y;
-	win->w = flip_win->attr.w;
-	win->h = flip_win->attr.h;
+	win->x.full = dfixed_const(flip_win->attr.x);
+	win->y.full = dfixed_const(flip_win->attr.y);
+	win->w.full = dfixed_const(flip_win->attr.w);
+	win->h.full = dfixed_const(flip_win->attr.h);
 	win->out_x = flip_win->attr.out_x;
 	win->out_y = flip_win->attr.out_y;
 	win->out_w = flip_win->attr.out_w;
@@ -183,12 +184,16 @@ static int tegra_overlay_set_windowattr(struct tegra_overlay_info *overlay,
 
 	if (((win->out_x + win->out_w) > xres) && (win->out_x < xres)) {
 		long new_w = xres - win->out_x;
-		win->w = win->w * new_w / win->out_w;
+		u64 in_w = win->w.full * new_w;
+		do_div(in_w, win->out_w);
+		win->w.full = lower_32_bits(in_w);
 	        win->out_w = new_w;
 	}
 	if (((win->out_y + win->out_h) > yres) && (win->out_y < yres)) {
 		long new_h = yres - win->out_y;
-		win->h = win->h * new_h / win->out_h;
+		u64 in_h = win->h.full * new_h;
+		do_div(in_h, win->out_h);
+		win->h.full = lower_32_bits(in_h);
 	        win->out_h = new_h;
 	}
 
diff --git a/drivers/video/tegra/fb.c b/drivers/video/tegra/fb.c
index a3d5c26289dd..6aff1cab30d4 100644
--- a/drivers/video/tegra/fb.c
+++ b/drivers/video/tegra/fb.c
@@ -138,8 +138,8 @@ static int tegra_fb_set_par(struct fb_info *info)
 
 		tegra_dc_set_fb_mode(tegra_fb->win->dc, info->mode, stereo);
 
-		tegra_fb->win->w = info->mode->xres;
-		tegra_fb->win->h = info->mode->yres;
+		tegra_fb->win->w.full = dfixed_const(info->mode->xres);
+		tegra_fb->win->h.full = dfixed_const(info->mode->yres);
 		tegra_fb->win->out_w = info->mode->xres;
 		tegra_fb->win->out_h = info->mode->yres;
 	}
@@ -446,10 +446,10 @@ struct tegra_fb_info *tegra_fb_register(struct nvhost_device *ndev,
 	info->var.vsync_len		= 0;
 	info->var.vmode			= FB_VMODE_NONINTERLACED;
 
-	win->x = 0;
-	win->y = 0;
-	win->w = fb_data->xres;
-	win->h = fb_data->yres;
+	win->x.full = dfixed_const(0);
+	win->y.full = dfixed_const(0);
+	win->w.full = dfixed_const(fb_data->xres);
+	win->h.full = dfixed_const(fb_data->yres);
 	/* TODO: set to output res dc */
 	win->out_x = 0;
 	win->out_y = 0;
diff --git a/include/video/tegra_dc_ext.h b/include/video/tegra_dc_ext.h
index 6f43ee689b26..2e89a12add3e 100644
--- a/include/video/tegra_dc_ext.h
+++ b/include/video/tegra_dc_ext.h
@@ -65,6 +65,10 @@ struct tegra_dc_ext_flip_windowattr {
 	__u32	stride;
 	__u32	stride_uv;
 	__u32	pixformat;
+	/*
+	 * x, y, w, h are fixed-point: 20 bits of integer (MSB) and 12 bits of
+	 * fractional (LSB)
+	 */
 	__u32	x;
 	__u32	y;
 	__u32	w;