i915: Improved page flipping support, including triple buffering.

Pages are tracked independently on each pipe.

Bump the minor version for 3D clients to know page flipping is usable, and
bump driver date.

4 files changed, 122 insertions, 33 deletions

diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c
index 18fe0881..dee381e6 100644
--- a/shared-core/i915_dma.c
+++ b/shared-core/i915_dma.c
@@ -165,8 +165,6 @@ static int i915_initialize(drm_device_t * dev,
 	dev_priv->ring.virtual_start = dev_priv->ring.map.handle;
 
 	dev_priv->cpp = init->cpp;
-	dev_priv->back_offset = init->back_offset;
-	dev_priv->front_offset = init->front_offset;
 	dev_priv->current_page = 0;
 	dev_priv->sarea_priv->pf_current_page = dev_priv->current_page;
 
@@ -553,16 +551,74 @@ static int i915_dispatch_batchbuffer(drm_device_t * dev,
 	return 0;
 }
 
-static int i915_dispatch_flip(drm_device_t * dev)
+static void i915_do_dispatch_flip(drm_device_t * dev, int pipe)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
+	u32 num_pages, current_page, next_page, dspbase;
+	int shift = 2 * pipe, x, y;
 	RING_LOCALS;
 
-	DRM_DEBUG("%s: page=%d pfCurrentPage=%d\n",
+	/* Calculate display base offset */
+	num_pages = dev_priv->sarea_priv->third_handle ? 3 : 2;
+	current_page = (dev_priv->current_page >> shift) & 0x3;
+	next_page = (current_page + 1) % num_pages;
+
+	switch (next_page) {
+	default:
+	case 0:
+		dspbase = dev_priv->sarea_priv->front_offset;
+		break;
+	case 1:
+		dspbase = dev_priv->sarea_priv->back_offset;
+		break;
+	case 2:
+		dspbase = dev_priv->sarea_priv->third_offset;
+		break;
+	}
+
+	if (pipe == 0) {
+		x = dev_priv->sarea_priv->pipeA_x;
+		y = dev_priv->sarea_priv->pipeA_y;
+	} else {
+		x = dev_priv->sarea_priv->pipeB_x;
+		y = dev_priv->sarea_priv->pipeB_y;
+	}
+
+	dspbase += (y * dev_priv->sarea_priv->pitch + x) * dev_priv->cpp;
+
+	DRM_DEBUG("pipe=%d current_page=%d dspbase=0x%x\n", pipe, current_page,
+		  dspbase);
+
+	BEGIN_LP_RING(4);
+	OUT_RING(CMD_OP_DISPLAYBUFFER_INFO | ASYNC_FLIP |
+		 (pipe ? DISPLAY_PLANE_B : DISPLAY_PLANE_A));
+	OUT_RING(0);
+	OUT_RING(dspbase);
+	OUT_RING(0);
+	ADVANCE_LP_RING();
+
+	dev_priv->current_page &= ~(0x3 << shift);
+	dev_priv->current_page |= next_page << shift;
+}
+
+static void i915_dispatch_flip(drm_device_t * dev, int pipes)
+{
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	u32 mi_wait = MI_WAIT_FOR_EVENT;
+	int i;
+	RING_LOCALS;
+
+	DRM_DEBUG("%s: pipes=0x%x page=%d pfCurrentPage=%d\n",
 		  __FUNCTION__,
-		  dev_priv->current_page,
+		  pipes, dev_priv->current_page,
 		  dev_priv->sarea_priv->pf_current_page);
 
+	if (pipes & 0x1)
+		mi_wait |= MI_WAIT_FOR_PLANE_A_FLIP;
+
+	if (pipes & 0x2)
+		mi_wait |= MI_WAIT_FOR_PLANE_B_FLIP;
+
 	i915_kernel_lost_context(dev);
 
 	BEGIN_LP_RING(2);
@@ -570,24 +626,15 @@ static int i915_dispatch_flip(drm_device_t * dev)
 	OUT_RING(0);
 	ADVANCE_LP_RING();
 
-	/* Wait for a pending flip to take effect */
+	/* Wait for pending flips to take effect */
 	BEGIN_LP_RING(2);
-	OUT_RING(MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_A_FLIP);
+	OUT_RING(mi_wait);
 	OUT_RING(0);
 	ADVANCE_LP_RING();
 
-	BEGIN_LP_RING(6);
-	OUT_RING(CMD_OP_DISPLAYBUFFER_INFO | ASYNC_FLIP);
-	OUT_RING(0);
-	if (dev_priv->current_page == 0) {
-		OUT_RING(dev_priv->back_offset);
-		dev_priv->current_page = 1;
-	} else {
-		OUT_RING(dev_priv->front_offset);
-		dev_priv->current_page = 0;
-	}
-	OUT_RING(0);
-	ADVANCE_LP_RING();
+	for (i = 0; i < 2; i++)
+		if (pipes & (1 << i))
+			i915_do_dispatch_flip(dev, i);
 
 	i915_emit_breadcrumb(dev);
 #ifdef I915_HAVE_FENCE
@@ -595,7 +642,6 @@ static int i915_dispatch_flip(drm_device_t * dev)
 #endif
 
 	dev_priv->sarea_priv->pf_current_page = dev_priv->current_page;
-	return 0;
 }
 
 static int i915_quiescent(drm_device_t * dev)
@@ -688,10 +734,19 @@ static int i915_cmdbuffer(DRM_IOCTL_ARGS)
 static int i915_do_cleanup_pageflip(drm_device_t * dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
+	int j;
 
 	DRM_DEBUG("%s\n", __FUNCTION__);
-	if (dev_priv->current_page != 0)
-		i915_dispatch_flip(dev);
+
+	for (j = 0; j < 2 && dev_priv->current_page != 0; j++) {
+		int i, pipes;
+
+		for (i = 0, pipes = 0; i < 2; i++)
+			if (dev_priv->current_page & (0x3 << (2 * i)))
+				pipes |= 1 << i;
+
+		i915_dispatch_flip(dev, pipes);
+	}
 
 	return 0;
 }
@@ -699,12 +754,24 @@ static int i915_do_cleanup_pageflip(drm_device_t * dev)
 static int i915_flip_bufs(DRM_IOCTL_ARGS)
 {
 	DRM_DEVICE;
+	drm_i915_flip_t param;
 
 	DRM_DEBUG("%s\n", __FUNCTION__);
 
 	LOCK_TEST_WITH_RETURN(dev, filp);
 
-	return i915_dispatch_flip(dev);
+	DRM_COPY_FROM_USER_IOCTL(param, (drm_i915_flip_t __user *) data,
+				 sizeof(param));
+
+	if (param.pipes & ~0x3) {
+		DRM_ERROR("Invalid pipes 0x%x, only <= 0x3 is valid\n",
+			  param.pipes);
+		return DRM_ERR(EINVAL);
+	}
+
+	i915_dispatch_flip(dev, param.pipes);
+
+	return 0;
 }
 
 
diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h
index 22a81d14..1f32313e 100644
--- a/shared-core/i915_drm.h
+++ b/shared-core/i915_drm.h
@@ -113,6 +113,12 @@ typedef struct _drm_i915_sarea {
 	int pipeB_y;
 	int pipeB_w;
 	int pipeB_h;
+
+	/* Triple buffering */
+	drm_handle_t third_handle;
+	int third_offset;
+	int third_size;
+	unsigned int third_tiled;
 } drm_i915_sarea_t;
 
 /* Driver specific fence types and classes.
@@ -156,7 +162,7 @@ typedef struct _drm_i915_sarea {
 
 #define DRM_IOCTL_I915_INIT		DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t)
 #define DRM_IOCTL_I915_FLUSH		DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH)
-#define DRM_IOCTL_I915_FLIP		DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLIP)
+#define DRM_IOCTL_I915_FLIP		DRM_IOW( DRM_COMMAND_BASE + DRM_I915_FLIP, drm_i915_flip_t)
 #define DRM_IOCTL_I915_BATCHBUFFER	DRM_IOW( DRM_COMMAND_BASE + DRM_I915_BATCHBUFFER, drm_i915_batchbuffer_t)
 #define DRM_IOCTL_I915_IRQ_EMIT         DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_IRQ_EMIT, drm_i915_irq_emit_t)
 #define DRM_IOCTL_I915_IRQ_WAIT         DRM_IOW( DRM_COMMAND_BASE + DRM_I915_IRQ_WAIT, drm_i915_irq_wait_t)
@@ -172,6 +178,12 @@ typedef struct _drm_i915_sarea {
 #define DRM_IOCTL_I915_VBLANK_SWAP	DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_VBLANK_SWAP, drm_i915_vblank_swap_t)
 
 
+/* Asynchronous page flipping:
+ */
+typedef struct drm_i915_flip {
+	int pipes;
+} drm_i915_flip_t;
+
 /* Allow drivers to submit batchbuffers directly to hardware, relying
  * on the security mechanisms provided by hardware.
  */
diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h
index a81653a7..1b261d6f 100644
--- a/shared-core/i915_drv.h
+++ b/shared-core/i915_drv.h
@@ -37,7 +37,7 @@
 
 #define DRIVER_NAME		"i915"
 #define DRIVER_DESC		"Intel Graphics"
-#define DRIVER_DATE		"20060929"
+#define DRIVER_DATE		"20070209"
 
 /* Interface history:
  *
@@ -49,9 +49,10 @@
  * 1.6: - New ioctl for scheduling buffer swaps on vertical blank
  *      - Support vertical blank on secondary display pipe
  * 1.8: New ioctl for ARB_Occlusion_Query
+ * 1.9: Usable page flipping and triple buffering
  */
 #define DRIVER_MAJOR		1
-#define DRIVER_MINOR		8
+#define DRIVER_MINOR		9
 #define DRIVER_PATCHLEVEL	0
 
 #if defined(__linux__)
@@ -99,8 +100,6 @@ typedef struct drm_i915_private {
 	uint32_t counter;
 
 	unsigned int cpp;
-	int back_offset;
-	int front_offset;
 	int current_page;
 	int use_mi_batchbuffer_start;
 
@@ -352,6 +351,7 @@ extern int i915_wait_ring(drm_device_t * dev, int n, const char *caller);
 #define MI_BATCH_NON_SECURE	(1)
 
 #define MI_WAIT_FOR_EVENT       ((0x3<<23))
+#define MI_WAIT_FOR_PLANE_B_FLIP      (1<<6)
 #define MI_WAIT_FOR_PLANE_A_FLIP      (1<<2)
 #define MI_WAIT_FOR_PLANE_A_SCANLINES (1<<1)
 
@@ -359,6 +359,8 @@ extern int i915_wait_ring(drm_device_t * dev, int n, const char *caller);
 
 #define CMD_OP_DISPLAYBUFFER_INFO ((0x0<<29)|(0x14<<23)|2)
 #define ASYNC_FLIP                (1<<22)
+#define DISPLAY_PLANE_A           (0<<20)
+#define DISPLAY_PLANE_B           (1<<20)
 
 #define CMD_OP_DESTBUFFER_INFO	 ((0x3<<29)|(0x1d<<24)|(0x8e<<16)|1)
 
diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c
index e373a8df..cd2adbf3 100644
--- a/shared-core/i915_irq.c
+++ b/shared-core/i915_irq.c
@@ -47,12 +47,12 @@ static void i915_vblank_tasklet(drm_device_t *dev)
 	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
 	unsigned long irqflags;
 	struct list_head *list, *tmp, hits, *hit;
-	int nhits, nrects, slice[2], upper[2], lower[2], i;
+	int nhits, nrects, slice[2], upper[2], lower[2], i, num_pages;
 	unsigned counter[2] = { atomic_read(&dev->vbl_received),
 				atomic_read(&dev->vbl_received2) };
 	drm_drawable_info_t *drw;
 	drm_i915_sarea_t *sarea_priv = dev_priv->sarea_priv;
-	u32 cpp = dev_priv->cpp;
+	u32 cpp = dev_priv->cpp,  offsets[3];
 	u32 cmd = (cpp == 4) ? (XY_SRC_COPY_BLT_CMD |
 				XY_SRC_COPY_BLT_WRITE_ALPHA |
 				XY_SRC_COPY_BLT_WRITE_RGB)
@@ -144,6 +144,11 @@ static void i915_vblank_tasklet(drm_device_t *dev)
 	lower[0] = sarea_priv->pipeA_y + slice[0];
 	lower[1] = sarea_priv->pipeB_y + slice[0];
 
+	offsets[0] = sarea_priv->front_offset;
+	offsets[1] = sarea_priv->back_offset;
+	offsets[2] = sarea_priv->third_offset;
+	num_pages = sarea_priv->third_handle ? 3 : 2;
+
 	spin_lock(&dev->drw_lock);
 
 	/* Emit blits for buffer swaps, partitioning both outputs into as many
@@ -161,7 +166,7 @@ static void i915_vblank_tasklet(drm_device_t *dev)
 			drm_i915_vbl_swap_t *swap_hit =
 				list_entry(hit, drm_i915_vbl_swap_t, head);
 			drm_clip_rect_t *rect;
-			int num_rects, pipe;
+			int num_rects, pipe, front, back;
 			unsigned short top, bottom;
 
 			drw = drm_get_drawable_info(dev, swap_hit->drw_id);
@@ -174,6 +179,9 @@ static void i915_vblank_tasklet(drm_device_t *dev)
 			top = upper[pipe];
 			bottom = lower[pipe];
 
+			front = (dev_priv->current_page >> (2 * pipe)) & 0x3;
+			back = (front + 1) % num_pages;
+
 			for (num_rects = drw->num_rects; num_rects--; rect++) {
 				int y1 = max(rect->y1, top);
 				int y2 = min(rect->y2, bottom);
@@ -187,10 +195,10 @@ static void i915_vblank_tasklet(drm_device_t *dev)
 				OUT_RING(pitchropcpp);
 				OUT_RING((y1 << 16) | rect->x1);
 				OUT_RING((y2 << 16) | rect->x2);
-				OUT_RING(sarea_priv->front_offset);
+				OUT_RING(offsets[front]);
 				OUT_RING((y1 << 16) | rect->x1);
 				OUT_RING(pitchropcpp & 0xffff);
-				OUT_RING(sarea_priv->back_offset);
+				OUT_RING(offsets[back]);
 
 				ADVANCE_LP_RING();
 			}