Merge branch 'drm-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6

* 'drm-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm: don't set the signal blocker on the master process.
  drm: don't call the vblank tasklet with irqs disabled.
  r300: Fix cliprect emit
  drm/radeon: r300_cmdbuf: Always emit INDX_BUFFER immediately after DRAW_INDEX
  radeon: fix some hard lockups on r3/4/500s

6 files changed, 225 insertions, 86 deletions

diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
index 089c015c01d..53f0e5af1cc 100644
--- a/drivers/gpu/drm/drm_irq.c
+++ b/drivers/gpu/drm/drm_irq.c
@@ -400,27 +400,31 @@ static void drm_locked_tasklet_func(unsigned long data)
 {
 	struct drm_device *dev = (struct drm_device *)data;
 	unsigned long irqflags;
-
+	void (*tasklet_func)(struct drm_device *);
+	
 	spin_lock_irqsave(&dev->tasklet_lock, irqflags);
+	tasklet_func = dev->locked_tasklet_func;
+	spin_unlock_irqrestore(&dev->tasklet_lock, irqflags);
 
-	if (!dev->locked_tasklet_func ||
+	if (!tasklet_func ||
 	    !drm_lock_take(&dev->lock,
 			   DRM_KERNEL_CONTEXT)) {
-		spin_unlock_irqrestore(&dev->tasklet_lock, irqflags);
 		return;
 	}
 
 	dev->lock.lock_time = jiffies;
 	atomic_inc(&dev->counts[_DRM_STAT_LOCKS]);
 
-	dev->locked_tasklet_func(dev);
+	spin_lock_irqsave(&dev->tasklet_lock, irqflags);
+	tasklet_func = dev->locked_tasklet_func;
+	dev->locked_tasklet_func = NULL;
+	spin_unlock_irqrestore(&dev->tasklet_lock, irqflags);
+	
+	if (tasklet_func != NULL)
+		tasklet_func(dev);
 
 	drm_lock_free(&dev->lock,
 		      DRM_KERNEL_CONTEXT);
-
-	dev->locked_tasklet_func = NULL;
-
-	spin_unlock_irqrestore(&dev->tasklet_lock, irqflags);
 }
 
 /**
diff --git a/drivers/gpu/drm/drm_lock.c b/drivers/gpu/drm/drm_lock.c
index 0998723cde7..a4caf95485d 100644
--- a/drivers/gpu/drm/drm_lock.c
+++ b/drivers/gpu/drm/drm_lock.c
@@ -105,14 +105,19 @@ int drm_lock(struct drm_device *dev, void *data, struct drm_file *file_priv)
 		  ret ? "interrupted" : "has lock");
 	if (ret) return ret;
 
-	sigemptyset(&dev->sigmask);
-	sigaddset(&dev->sigmask, SIGSTOP);
-	sigaddset(&dev->sigmask, SIGTSTP);
-	sigaddset(&dev->sigmask, SIGTTIN);
-	sigaddset(&dev->sigmask, SIGTTOU);
-	dev->sigdata.context = lock->context;
-	dev->sigdata.lock = dev->lock.hw_lock;
-	block_all_signals(drm_notifier, &dev->sigdata, &dev->sigmask);
+	/* don't set the block all signals on the master process for now 
+	 * really probably not the correct answer but lets us debug xkb
+ 	 * xserver for now */
+	if (!file_priv->master) {
+		sigemptyset(&dev->sigmask);
+		sigaddset(&dev->sigmask, SIGSTOP);
+		sigaddset(&dev->sigmask, SIGTSTP);
+		sigaddset(&dev->sigmask, SIGTTIN);
+		sigaddset(&dev->sigmask, SIGTTOU);
+		dev->sigdata.context = lock->context;
+		dev->sigdata.lock = dev->lock.hw_lock;
+		block_all_signals(drm_notifier, &dev->sigdata, &dev->sigmask);
+	}
 
 	if (dev->driver->dma_ready && (lock->flags & _DRM_LOCK_READY))
 		dev->driver->dma_ready(dev);
@@ -150,6 +155,7 @@ int drm_unlock(struct drm_device *dev, void *data, struct drm_file *file_priv)
 {
 	struct drm_lock *lock = data;
 	unsigned long irqflags;
+	void (*tasklet_func)(struct drm_device *);
 
 	if (lock->context == DRM_KERNEL_CONTEXT) {
 		DRM_ERROR("Process %d using kernel context %d\n",
@@ -158,14 +164,11 @@ int drm_unlock(struct drm_device *dev, void *data, struct drm_file *file_priv)
 	}
 
 	spin_lock_irqsave(&dev->tasklet_lock, irqflags);
-
-	if (dev->locked_tasklet_func) {
-		dev->locked_tasklet_func(dev);
-
-		dev->locked_tasklet_func = NULL;
-	}
-
+	tasklet_func = dev->locked_tasklet_func;
+	dev->locked_tasklet_func = NULL;
 	spin_unlock_irqrestore(&dev->tasklet_lock, irqflags);
+	if (tasklet_func != NULL)
+		tasklet_func(dev);
 
 	atomic_inc(&dev->counts[_DRM_STAT_UNLOCKS]);
 
diff --git a/drivers/gpu/drm/radeon/r300_cmdbuf.c b/drivers/gpu/drm/radeon/r300_cmdbuf.c
index 702df45320f..4b27d9abb7b 100644
--- a/drivers/gpu/drm/radeon/r300_cmdbuf.c
+++ b/drivers/gpu/drm/radeon/r300_cmdbuf.c
@@ -77,6 +77,9 @@ static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
 				return -EFAULT;
 			}
 
+			box.x2--; /* Hardware expects inclusive bottom-right corner */
+			box.y2--;
+
 			if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
 				box.x1 = (box.x1) &
 					R300_CLIPRECT_MASK;
@@ -95,8 +98,8 @@ static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
 					R300_CLIPRECT_MASK;
 				box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) &
 					R300_CLIPRECT_MASK;
-
 			}
+
 			OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
 				 (box.y1 << R300_CLIPRECT_Y_SHIFT));
 			OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
@@ -136,6 +139,18 @@ static int r300_emit_cliprects(drm_radeon_private_t *dev_priv,
 		ADVANCE_RING();
 	}
 
+	/* flus cache and wait idle clean after cliprect change */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FLUSH);
+	ADVANCE_RING();
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	ADVANCE_RING();
+	/* set flush flag */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
 	return 0;
 }
 
@@ -166,13 +181,13 @@ void r300_init_reg_flags(struct drm_device *dev)
 	ADD_RANGE(0x21DC, 1);
 	ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
 	ADD_RANGE(R300_VAP_CLIP_X_0, 4);
-	ADD_RANGE(R300_VAP_PVS_WAITIDLE, 1);
+	ADD_RANGE(R300_VAP_PVS_STATE_FLUSH_REG, 1);
 	ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
 	ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
 	ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
 	ADD_RANGE(R300_GB_ENABLE, 1);
 	ADD_RANGE(R300_GB_MSPOS0, 5);
-	ADD_RANGE(R300_TX_CNTL, 1);
+	ADD_RANGE(R300_TX_INVALTAGS, 1);
 	ADD_RANGE(R300_TX_ENABLE, 1);
 	ADD_RANGE(0x4200, 4);
 	ADD_RANGE(0x4214, 1);
@@ -388,15 +403,28 @@ static __inline__ int r300_emit_vpu(drm_radeon_private_t *dev_priv,
 	if (sz * 16 > cmdbuf->bufsz)
 		return -EINVAL;
 
-	BEGIN_RING(5 + sz * 4);
-	/* Wait for VAP to come to senses.. */
-	/* there is no need to emit it multiple times, (only once before VAP is programmed,
-	   but this optimization is for later */
-	OUT_RING_REG(R300_VAP_PVS_WAITIDLE, 0);
+	/* VAP is very sensitive so we purge cache before we program it
+	 * and we also flush its state before & after */
+	BEGIN_RING(6);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FLUSH);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
+	/* set flush flag */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
+	BEGIN_RING(3 + sz * 4);
 	OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
 	OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
 	OUT_RING_TABLE((int *)cmdbuf->buf, sz * 4);
+	ADVANCE_RING();
 
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
+	OUT_RING(0);
 	ADVANCE_RING();
 
 	cmdbuf->buf += sz * 16;
@@ -424,6 +452,15 @@ static __inline__ int r300_emit_clear(drm_radeon_private_t *dev_priv,
 	OUT_RING_TABLE((int *)cmdbuf->buf, 8);
 	ADVANCE_RING();
 
+	BEGIN_RING(4);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FLUSH);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	ADVANCE_RING();
+	/* set flush flag */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
 	cmdbuf->buf += 8 * 4;
 	cmdbuf->bufsz -= 8 * 4;
 
@@ -543,22 +580,23 @@ static __inline__ int r300_emit_bitblt_multi(drm_radeon_private_t *dev_priv,
 	return 0;
 }
 
-static __inline__ int r300_emit_indx_buffer(drm_radeon_private_t *dev_priv,
-					     drm_radeon_kcmd_buffer_t *cmdbuf)
+static __inline__ int r300_emit_draw_indx_2(drm_radeon_private_t *dev_priv,
+					    drm_radeon_kcmd_buffer_t *cmdbuf)
 {
-	u32 *cmd = (u32 *) cmdbuf->buf;
-	int count, ret;
+	u32 *cmd;
+	int count;
+	int expected_count;
 	RING_LOCALS;
 
-	count=(cmd[0]>>16) & 0x3fff;
+	cmd = (u32 *) cmdbuf->buf;
+	count = (cmd[0]>>16) & 0x3fff;
+	expected_count = cmd[1] >> 16;
+	if (!(cmd[1] & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
+		expected_count = (expected_count+1)/2;
 
-	if ((cmd[1] & 0x8000ffff) != 0x80000810) {
-		DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
-		return -EINVAL;
-	}
-	ret = !radeon_check_offset(dev_priv, cmd[2]);
-	if (ret) {
-		DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+	if (count && count != expected_count) {
+		DRM_ERROR("3D_DRAW_INDX_2: packet size %i, expected %i\n",
+			count, expected_count);
 		return -EINVAL;
 	}
 
@@ -570,6 +608,50 @@ static __inline__ int r300_emit_indx_buffer(drm_radeon_private_t *dev_priv,
 	cmdbuf->buf += (count+2)*4;
 	cmdbuf->bufsz -= (count+2)*4;
 
+	if (!count) {
+		drm_r300_cmd_header_t header;
+
+		if (cmdbuf->bufsz < 4*4 + sizeof(header)) {
+			DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER, but stream is too short.\n");
+			return -EINVAL;
+		}
+
+		header.u = *(unsigned int *)cmdbuf->buf;
+
+		cmdbuf->buf += sizeof(header);
+		cmdbuf->bufsz -= sizeof(header);
+		cmd = (u32 *) cmdbuf->buf;
+
+		if (header.header.cmd_type != R300_CMD_PACKET3 ||
+		    header.packet3.packet != R300_CMD_PACKET3_RAW ||
+		    cmd[0] != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) {
+			DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER.\n");
+			return -EINVAL;
+		}
+
+		if ((cmd[1] & 0x8000ffff) != 0x80000810) {
+			DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
+			return -EINVAL;
+		}
+		if (!radeon_check_offset(dev_priv, cmd[2])) {
+			DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+			return -EINVAL;
+		}
+		if (cmd[3] != expected_count) {
+			DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n",
+				cmd[3], expected_count);
+			return -EINVAL;
+		}
+
+		BEGIN_RING(4);
+		OUT_RING(cmd[0]);
+		OUT_RING_TABLE((int *)(cmdbuf->buf + 4), 3);
+		ADVANCE_RING();
+
+		cmdbuf->buf += 4*4;
+		cmdbuf->bufsz -= 4*4;
+	}
+
 	return 0;
 }
 
@@ -613,11 +695,22 @@ static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t *dev_priv,
 	case RADEON_CNTL_BITBLT_MULTI:
 		return r300_emit_bitblt_multi(dev_priv, cmdbuf);
 
-	case RADEON_CP_INDX_BUFFER:	/* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
-		return r300_emit_indx_buffer(dev_priv, cmdbuf);
-	case RADEON_CP_3D_DRAW_IMMD_2:	/* triggers drawing using in-packet vertex data */
-	case RADEON_CP_3D_DRAW_VBUF_2:	/* triggers drawing of vertex buffers setup elsewhere */
-	case RADEON_CP_3D_DRAW_INDX_2:	/* triggers drawing using indices to vertex buffer */
+	case RADEON_CP_INDX_BUFFER:
+		DRM_ERROR("packet3 INDX_BUFFER without preceding 3D_DRAW_INDX_2 is illegal.\n");
+		return -EINVAL;
+	case RADEON_CP_3D_DRAW_IMMD_2:
+		/* triggers drawing using in-packet vertex data */
+	case RADEON_CP_3D_DRAW_VBUF_2:
+		/* triggers drawing of vertex buffers setup elsewhere */
+		dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
+					   RADEON_PURGE_EMITED);
+		break;
+	case RADEON_CP_3D_DRAW_INDX_2:
+		/* triggers drawing using indices to vertex buffer */
+		/* whenever we send vertex we clear flush & purge */
+		dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
+					   RADEON_PURGE_EMITED);
+		return r300_emit_draw_indx_2(dev_priv, cmdbuf);
 	case RADEON_WAIT_FOR_IDLE:
 	case RADEON_CP_NOP:
 		/* these packets are safe */
@@ -713,17 +806,53 @@ static __inline__ int r300_emit_packet3(drm_radeon_private_t *dev_priv,
  */
 static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
 {
+	uint32_t cache_z, cache_3d, cache_2d;
 	RING_LOCALS;
 
-	BEGIN_RING(6);
-	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
-	OUT_RING(R300_RB3D_DSTCACHE_UNKNOWN_0A);
+	cache_z = R300_ZC_FLUSH;
+	cache_2d = R300_RB2D_DC_FLUSH;
+	cache_3d = R300_RB3D_DC_FLUSH;
+	if (!(dev_priv->track_flush & RADEON_PURGE_EMITED)) {
+		/* we can purge, primitive where draw since last purge */
+		cache_z |= R300_ZC_FREE;
+		cache_2d |= R300_RB2D_DC_FREE;
+		cache_3d |= R300_RB3D_DC_FREE;
+	}
+
+	/* flush & purge zbuffer */
+	BEGIN_RING(2);
 	OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));
-	OUT_RING(R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE|
-		 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
-	OUT_RING(CP_PACKET3(RADEON_CP_NOP, 0));
-	OUT_RING(0x0);
+	OUT_RING(cache_z);
+	ADVANCE_RING();
+	/* flush & purge 3d */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(cache_3d);
+	ADVANCE_RING();
+	/* flush & purge texture */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_TX_INVALTAGS, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
+	/* FIXME: is this one really needed ? */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(R300_RB3D_AARESOLVE_CTL, 0));
+	OUT_RING(0);
+	ADVANCE_RING();
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+	ADVANCE_RING();
+	/* flush & purge 2d through E2 as RB2D will trigger lockup */
+	BEGIN_RING(4);
+	OUT_RING(CP_PACKET0(R300_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(cache_2d);
+	OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+	OUT_RING(RADEON_WAIT_2D_IDLECLEAN |
+		 RADEON_WAIT_HOST_IDLECLEAN);
 	ADVANCE_RING();
+	/* set flush & purge flags */
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED;
 }
 
 /**
@@ -905,8 +1034,7 @@ int r300_do_cp_cmdbuf(struct drm_device *dev,
 
 	DRM_DEBUG("\n");
 
-	/* See the comment above r300_emit_begin3d for why this call must be here,
-	 * and what the cleanup gotos are for. */
+	/* pacify */
 	r300_pacify(dev_priv);
 
 	if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
diff --git a/drivers/gpu/drm/radeon/r300_reg.h b/drivers/gpu/drm/radeon/r300_reg.h
index a6802f26afc..ee6f811599a 100644
--- a/drivers/gpu/drm/radeon/r300_reg.h
+++ b/drivers/gpu/drm/radeon/r300_reg.h
@@ -317,7 +317,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  * Therefore, I suspect writing zero to 0x2284 synchronizes the engine and
  * avoids bugs caused by still running shaders reading bad data from memory.
  */
-#define R300_VAP_PVS_WAITIDLE               0x2284 /* GUESS */
+#define R300_VAP_PVS_STATE_FLUSH_REG        0x2284
 
 /* Absolutely no clue what this register is about. */
 #define R300_VAP_UNKNOWN_2288               0x2288
@@ -513,7 +513,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* gap */
 
 /* Zero to flush caches. */
-#define R300_TX_CNTL                        0x4100
+#define R300_TX_INVALTAGS                   0x4100
 #define R300_TX_FLUSH                       0x0
 
 /* The upper enable bits are guessed, based on fglrx reported limits. */
@@ -1362,6 +1362,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_RB3D_COLORPITCH2               0x4E40 /* GUESS */
 #define R300_RB3D_COLORPITCH3               0x4E44 /* GUESS */
 
+#define R300_RB3D_AARESOLVE_CTL             0x4E88
 /* gap */
 
 /* Guess by Vladimir.
diff --git a/drivers/gpu/drm/radeon/radeon_cp.c b/drivers/gpu/drm/radeon/radeon_cp.c
index f0de81a5689..3331f88dcfb 100644
--- a/drivers/gpu/drm/radeon/radeon_cp.c
+++ b/drivers/gpu/drm/radeon/radeon_cp.c
@@ -40,6 +40,7 @@
 #define RADEON_FIFO_DEBUG	0
 
 static int radeon_do_cleanup_cp(struct drm_device * dev);
+static void radeon_do_cp_start(drm_radeon_private_t * dev_priv);
 
 static u32 R500_READ_MCIND(drm_radeon_private_t *dev_priv, int addr)
 {
@@ -198,23 +199,8 @@ static int radeon_do_pixcache_flush(drm_radeon_private_t * dev_priv)
 			DRM_UDELAY(1);
 		}
 	} else {
-		/* 3D */
-		tmp = RADEON_READ(R300_RB3D_DSTCACHE_CTLSTAT);
-		tmp |= RADEON_RB3D_DC_FLUSH_ALL;
-		RADEON_WRITE(R300_RB3D_DSTCACHE_CTLSTAT, tmp);
-
-		/* 2D */
-		tmp = RADEON_READ(R300_DSTCACHE_CTLSTAT);
-		tmp |= RADEON_RB3D_DC_FLUSH_ALL;
-		RADEON_WRITE(R300_DSTCACHE_CTLSTAT, tmp);
-
-		for (i = 0; i < dev_priv->usec_timeout; i++) {
-			if (!(RADEON_READ(R300_DSTCACHE_CTLSTAT)
-			  & RADEON_RB3D_DC_BUSY)) {
-				return 0;
-			}
-			DRM_UDELAY(1);
-		}
+		/* don't flush or purge cache here or lockup */
+		return 0;
 	}
 
 #if RADEON_FIFO_DEBUG
@@ -237,6 +223,9 @@ static int radeon_do_wait_for_fifo(drm_radeon_private_t * dev_priv, int entries)
 			return 0;
 		DRM_UDELAY(1);
 	}
+	DRM_INFO("wait for fifo failed status : 0x%08X 0x%08X\n",
+		 RADEON_READ(RADEON_RBBM_STATUS),
+		 RADEON_READ(R300_VAP_CNTL_STATUS));
 
 #if RADEON_FIFO_DEBUG
 	DRM_ERROR("failed!\n");
@@ -263,6 +252,9 @@ static int radeon_do_wait_for_idle(drm_radeon_private_t * dev_priv)
 		}
 		DRM_UDELAY(1);
 	}
+	DRM_INFO("wait idle failed status : 0x%08X 0x%08X\n",
+		 RADEON_READ(RADEON_RBBM_STATUS),
+		 RADEON_READ(R300_VAP_CNTL_STATUS));
 
 #if RADEON_FIFO_DEBUG
 	DRM_ERROR("failed!\n");
@@ -443,14 +435,20 @@ static void radeon_do_cp_start(drm_radeon_private_t * dev_priv)
 
 	dev_priv->cp_running = 1;
 
-	BEGIN_RING(6);
-
+	BEGIN_RING(8);
+	/* isync can only be written through cp on r5xx write it here */
+	OUT_RING(CP_PACKET0(RADEON_ISYNC_CNTL, 0));
+	OUT_RING(RADEON_ISYNC_ANY2D_IDLE3D |
+		 RADEON_ISYNC_ANY3D_IDLE2D |
+		 RADEON_ISYNC_WAIT_IDLEGUI |
+		 RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 	RADEON_PURGE_CACHE();
 	RADEON_PURGE_ZCACHE();
 	RADEON_WAIT_UNTIL_IDLE();
-
 	ADVANCE_RING();
 	COMMIT_RING();
+
+	dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED;
 }
 
 /* Reset the Command Processor.  This will not flush any pending
diff --git a/drivers/gpu/drm/radeon/radeon_drv.h b/drivers/gpu/drm/radeon/radeon_drv.h
index 3f0eca957aa..09938169317 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.h
+++ b/drivers/gpu/drm/radeon/radeon_drv.h
@@ -220,6 +220,9 @@ struct radeon_virt_surface {
 	struct drm_file *file_priv;
 };
 
+#define RADEON_FLUSH_EMITED	(1 < 0)
+#define RADEON_PURGE_EMITED	(1 < 1)
+
 typedef struct drm_radeon_private {
 	drm_radeon_ring_buffer_t ring;
 	drm_radeon_sarea_t *sarea_priv;
@@ -311,6 +314,7 @@ typedef struct drm_radeon_private {
 	unsigned long fb_aper_offset;
 
 	int num_gb_pipes;
+	int track_flush;
 } drm_radeon_private_t;
 
 typedef struct drm_radeon_buf_priv {
@@ -693,7 +697,6 @@ extern int r300_do_cp_cmdbuf(struct drm_device * dev,
 #define R300_ZB_ZCACHE_CTLSTAT                  0x4f18
 #	define R300_ZC_FLUSH		        (1 << 0)
 #	define R300_ZC_FREE		        (1 << 1)
-#	define R300_ZC_FLUSH_ALL		0x3
 #	define R300_ZC_BUSY		        (1 << 31)
 #define RADEON_RB3D_DSTCACHE_CTLSTAT	0x325c
 #	define RADEON_RB3D_DC_FLUSH		(3 << 0)
@@ -701,6 +704,8 @@ extern int r300_do_cp_cmdbuf(struct drm_device * dev,
 #	define RADEON_RB3D_DC_FLUSH_ALL		0xf
 #	define RADEON_RB3D_DC_BUSY		(1 << 31)
 #define R300_RB3D_DSTCACHE_CTLSTAT              0x4e4c
+#	define R300_RB3D_DC_FLUSH		(2 << 0)
+#	define R300_RB3D_DC_FREE		(2 << 2)
 #	define R300_RB3D_DC_FINISH		(1 << 4)
 #define RADEON_RB3D_ZSTENCILCNTL	0x1c2c
 #	define RADEON_Z_TEST_MASK		(7 << 4)
@@ -1246,17 +1251,17 @@ do {									\
 		OUT_RING(RADEON_RB3D_DC_FLUSH);				\
 	} else {                                                        \
 		OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));	\
-		OUT_RING(RADEON_RB3D_DC_FLUSH);				\
+		OUT_RING(R300_RB3D_DC_FLUSH);				\
 	}                                                               \
 } while (0)
 
 #define RADEON_PURGE_CACHE() do {					\
 	if ((dev_priv->flags & RADEON_FAMILY_MASK) <= CHIP_RV280) {	\
 		OUT_RING(CP_PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));	\
-		OUT_RING(RADEON_RB3D_DC_FLUSH_ALL);			\
+		OUT_RING(RADEON_RB3D_DC_FLUSH | RADEON_RB3D_DC_FREE);	\
 	} else {                                                        \
 		OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));	\
-		OUT_RING(RADEON_RB3D_DC_FLUSH_ALL);			\
+		OUT_RING(R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);	\
 	}                                                               \
 } while (0)
 
@@ -1273,10 +1278,10 @@ do {									\
 #define RADEON_PURGE_ZCACHE() do {					\
 	if ((dev_priv->flags & RADEON_FAMILY_MASK) <= CHIP_RV280) {	\
 		OUT_RING(CP_PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));	\
-		OUT_RING(RADEON_RB3D_ZC_FLUSH_ALL);			\
+		OUT_RING(RADEON_RB3D_ZC_FLUSH | RADEON_RB3D_ZC_FREE);			\
 	} else {                                                        \
-		OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));	\
-		OUT_RING(R300_ZC_FLUSH_ALL);				\
+		OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));	\
+		OUT_RING(R300_ZC_FLUSH | R300_ZC_FREE);				\
 	}                                                               \
 } while (0)