Fix ring space calculations, tests. Based on patch by Bruce Stockwell.

8 files changed, 27 insertions, 88 deletions

diff --git a/linux/r128_cce.c b/linux/r128_cce.c
index d1790497..5c67bec9 100644
--- a/linux/r128_cce.c
+++ b/linux/r128_cce.c
@@ -839,13 +839,9 @@ int r128_wait_ring( drm_r128_private_t *dev_priv, int n )
 	int i;
 
 	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
-		ring->space = *ring->head - ring->tail;
-		if ( ring->space <= 0 )
-			ring->space += ring->size;
-
+		r128_update_ring_snapshot( ring );
 		if ( ring->space >= n )
 			return 0;
-
 		udelay( 1 );
 	}
 
@@ -854,19 +850,6 @@ int r128_wait_ring( drm_r128_private_t *dev_priv, int n )
 	return -EBUSY;
 }
 
-void r128_update_ring_snapshot( drm_r128_private_t *dev_priv )
-{
-	drm_r128_ring_buffer_t *ring = &dev_priv->ring;
-
-	ring->space = *ring->head - ring->tail;
-#if R128_PERFORMANCE_BOXES
-	if ( ring->space == 0 )
-		atomic_inc( &dev_priv->idle_count );
-#endif
-	if ( ring->space <= 0 )
-		ring->space += ring->size;
-}
-
 static int r128_cce_get_buffers( drm_device_t *dev, drm_dma_t *d )
 {
 	int i;
diff --git a/linux/r128_drv.c b/linux/r128_drv.c
index e42868ed..daae0f87 100644
--- a/linux/r128_drv.c
+++ b/linux/r128_drv.c
@@ -38,11 +38,11 @@
 
 #define DRIVER_NAME		"r128"
 #define DRIVER_DESC		"ATI Rage 128"
-#define DRIVER_DATE		"20010216"
+#define DRIVER_DATE		"20010308"
 
 #define DRIVER_MAJOR		2
 #define DRIVER_MINOR		1
-#define DRIVER_PATCHLEVEL	4
+#define DRIVER_PATCHLEVEL	5
 
 #define DRIVER_IOCTLS							    \
    [DRM_IOCTL_NR(DRM_IOCTL_DMA)]             = { r128_cce_buffers,  1, 0 }, \
diff --git a/linux/r128_drv.h b/linux/r128_drv.h
index 9f53746f..6216b9a1 100644
--- a/linux/r128_drv.h
+++ b/linux/r128_drv.h
@@ -130,7 +130,14 @@ extern void r128_freelist_reset( drm_device_t *dev );
 extern drm_buf_t *r128_freelist_get( drm_device_t *dev );
 
 extern int r128_wait_ring( drm_r128_private_t *dev_priv, int n );
-extern void r128_update_ring_snapshot( drm_r128_private_t *dev_priv );
+
+static inline void
+r128_update_ring_snapshot( drm_r128_ring_buffer_t *ring )
+{
+	ring->space = (*(volatile int *)ring->head - ring->tail) * sizeof(u32);
+	if ( ring->space <= 0 )
+		ring->space += ring->size;
+}
 
 extern int r128_do_cce_idle( drm_r128_private_t *dev_priv );
 extern int r128_do_cleanup_cce( drm_device_t *dev );
@@ -415,9 +422,7 @@ do {									\
 	drm_r128_ring_buffer_t *ring = &dev_priv->ring; int i;		\
 	if ( ring->space < ring->high_mark ) {				\
 		for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {	\
-			ring->space = *ring->head - ring->tail;		\
-			if ( ring->space <= 0 )				\
-				ring->space += ring->size;		\
+			r128_update_ring_snapshot( ring );		\
 			if ( ring->space >= ring->high_mark )		\
 				goto __ring_space_done;			\
 			udelay( 1 );					\
@@ -462,7 +467,7 @@ do {									\
 		DRM_INFO( "BEGIN_RING( %d ) in %s\n",			\
 			   (n), __FUNCTION__ );				\
 	}								\
-	if ( dev_priv->ring.space < (n) * sizeof(u32) ) {		\
+	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {		\
 		r128_wait_ring( dev_priv, (n) * sizeof(u32) );		\
 	}								\
 	dev_priv->ring.space -= (n) * sizeof(u32);			\
diff --git a/linux/r128_state.c b/linux/r128_state.c
index fa16f3ce..9fc6b485 100644
--- a/linux/r128_state.c
+++ b/linux/r128_state.c
@@ -372,8 +372,6 @@ static void r128_cce_dispatch_clear( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	if ( dev_priv->page_flipping && dev_priv->current_page == 1 ) {
 		unsigned int tmp = flags;
 
@@ -477,8 +475,6 @@ static void r128_cce_dispatch_swap( drm_device_t *dev )
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 #if R128_PERFORMANCE_BOXES
 	/* Do some trivial performance monitoring...
 	 */
@@ -535,8 +531,6 @@ static void r128_cce_dispatch_flip( drm_device_t *dev )
 	RING_LOCALS;
 	DRM_DEBUG( "%s: page=%d\n", __FUNCTION__, dev_priv->current_page );
 
-	r128_update_ring_snapshot( dev_priv );
-
 #if R128_PERFORMANCE_BOXES
 	/* Do some trivial performance monitoring...
 	 */
@@ -587,8 +581,6 @@ static void r128_cce_dispatch_vertex( drm_device_t *dev,
 	DRM_DEBUG( "%s: buf=%d nbox=%d\n",
 		   __FUNCTION__, buf->idx, sarea_priv->nbox );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	if ( 0 )
 		r128_print_dirty( "dispatch_vertex", sarea_priv->dirty );
 
@@ -656,8 +648,6 @@ static void r128_cce_dispatch_indirect( drm_device_t *dev,
 	DRM_DEBUG( "indirect: buf=%d s=0x%x e=0x%x\n",
 		   buf->idx, start, end );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	if ( start != end ) {
 		int offset = buf->bus_address + start;
 		int dwords = (end - start + 3) / sizeof(u32);
@@ -722,8 +712,6 @@ static void r128_cce_dispatch_indices( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "indices: s=%d e=%d c=%d\n", start, end, count );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	if ( 0 )
 		r128_print_dirty( "dispatch_indices", sarea_priv->dirty );
 
@@ -799,8 +787,6 @@ static int r128_cce_dispatch_blit( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	/* The compiler won't optimize away a division by a variable,
 	 * even if the only legal values are powers of two.  Thus, we'll
 	 * use a shift instead.
@@ -911,8 +897,6 @@ static int r128_cce_dispatch_write_span( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	count = depth->n;
 	if ( copy_from_user( &x, depth->x, sizeof(x) ) ) {
 		return -EFAULT;
@@ -1006,8 +990,6 @@ static int r128_cce_dispatch_write_pixels( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	count = depth->n;
 
 	x = kmalloc( count * sizeof(*x), 0 );
@@ -1123,8 +1105,6 @@ static int r128_cce_dispatch_read_span( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	count = depth->n;
 	if ( copy_from_user( &x, depth->x, sizeof(x) ) ) {
 		return -EFAULT;
@@ -1167,8 +1147,6 @@ static int r128_cce_dispatch_read_pixels( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	count = depth->n;
 	if ( count > dev_priv->depth_pitch ) {
 		count = dev_priv->depth_pitch;
@@ -1236,8 +1214,6 @@ static void r128_cce_dispatch_stipple( drm_device_t *dev, u32 *stipple )
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	r128_update_ring_snapshot( dev_priv );
-
 	BEGIN_RING( 33 );
 
 	OUT_RING( CCE_PACKET0( R128_BRUSH_DATA0, 31 ) );
diff --git a/linux/radeon_cp.c b/linux/radeon_cp.c
index 3feb4b24..cea8c327 100644
--- a/linux/radeon_cp.c
+++ b/linux/radeon_cp.c
@@ -1166,13 +1166,9 @@ int radeon_wait_ring( drm_radeon_private_t *dev_priv, int n )
 	int i;
 
 	for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {
-		ring->space = *ring->head - ring->tail;
-		if ( ring->space <= 0 )
-			ring->space += ring->size;
-
-		if ( ring->space >= n )
+		radeon_update_ring_snapshot( ring );
+		if ( ring->space > n )
 			return 0;
-
 		udelay( 1 );
 	}
 
@@ -1181,17 +1177,6 @@ int radeon_wait_ring( drm_radeon_private_t *dev_priv, int n )
 	return -EBUSY;
 }
 
-void radeon_update_ring_snapshot( drm_radeon_private_t *dev_priv )
-{
-	drm_radeon_ring_buffer_t *ring = &dev_priv->ring;
-
-	ring->space = *ring->head - ring->tail;
-	if ( ring->space == 0 )
-		atomic_inc( &dev_priv->idle_count );
-	if ( ring->space <= 0 )
-		ring->space += ring->size;
-}
-
 static int radeon_cp_get_buffers( drm_device_t *dev, drm_dma_t *d )
 {
 	int i;
diff --git a/linux/radeon_drv.c b/linux/radeon_drv.c
index 1aa889ae..dba2037f 100644
--- a/linux/radeon_drv.c
+++ b/linux/radeon_drv.c
@@ -36,11 +36,11 @@
 
 #define DRIVER_NAME		"radeon"
 #define DRIVER_DESC		"ATI Radeon"
-#define DRIVER_DATE		"20010305"
+#define DRIVER_DATE		"20010308"
 
 #define DRIVER_MAJOR		1
 #define DRIVER_MINOR		0
-#define DRIVER_PATCHLEVEL	0
+#define DRIVER_PATCHLEVEL	1
 
 #define DRIVER_IOCTLS							     \
  [DRM_IOCTL_NR(DRM_IOCTL_DMA)]               = { radeon_cp_buffers,  1, 0 }, \
diff --git a/linux/radeon_drv.h b/linux/radeon_drv.h
index d279b046..f176bb56 100644
--- a/linux/radeon_drv.h
+++ b/linux/radeon_drv.h
@@ -146,7 +146,14 @@ extern void radeon_freelist_reset( drm_device_t *dev );
 extern drm_buf_t *radeon_freelist_get( drm_device_t *dev );
 
 extern int radeon_wait_ring( drm_radeon_private_t *dev_priv, int n );
-extern void radeon_update_ring_snapshot( drm_radeon_private_t *dev_priv );
+
+static inline void
+radeon_update_ring_snapshot( drm_radeon_ring_buffer_t *ring )
+{
+	ring->space = (*(volatile int *)ring->head - ring->tail) * sizeof(u32);
+	if ( ring->space <= 0 )
+		ring->space += ring->size;
+}
 
 extern int radeon_do_cp_idle( drm_radeon_private_t *dev_priv );
 extern int radeon_do_cleanup_cp( drm_device_t *dev );
@@ -601,9 +608,7 @@ do {									\
 	drm_radeon_ring_buffer_t *ring = &dev_priv->ring; int i;	\
 	if ( ring->space < ring->high_mark ) {				\
 		for ( i = 0 ; i < dev_priv->usec_timeout ; i++ ) {	\
-			ring->space = *ring->head - ring->tail;		\
-			if ( ring->space <= 0 )				\
-				ring->space += ring->size;		\
+			radeon_update_ring_snapshot( ring );		\
 			if ( ring->space >= ring->high_mark )		\
 				goto __ring_space_done;			\
 			udelay( 1 );					\
@@ -657,7 +662,7 @@ do {									\
 		DRM_INFO( "BEGIN_RING( %d ) in %s\n",			\
 			   n, __FUNCTION__ );				\
 	}								\
-	if ( dev_priv->ring.space < (n) * sizeof(u32) ) {		\
+	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {		\
 		radeon_wait_ring( dev_priv, (n) * sizeof(u32) );	\
 	}								\
 	dev_priv->ring.space -= (n) * sizeof(u32);			\
diff --git a/linux/radeon_state.c b/linux/radeon_state.c
index ff1b3512..816e63bd 100644
--- a/linux/radeon_state.c
+++ b/linux/radeon_state.c
@@ -498,8 +498,6 @@ static void radeon_cp_dispatch_clear( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 	if ( dev_priv->page_flipping && dev_priv->current_page == 1 ) {
 		unsigned int tmp = flags;
 
@@ -656,8 +654,6 @@ static void radeon_cp_dispatch_swap( drm_device_t *dev )
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 #if RADEON_PERFORMANCE_BOXES
 	/* Do some trivial performance monitoring...
 	 */
@@ -725,8 +721,6 @@ static void radeon_cp_dispatch_flip( drm_device_t *dev )
 	RING_LOCALS;
 	DRM_DEBUG( "%s: page=%d\n", __FUNCTION__, dev_priv->current_page );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 #if RADEON_PERFORMANCE_BOXES
 	/* Do some trivial performance monitoring...
 	 */
@@ -777,8 +771,6 @@ static void radeon_cp_dispatch_vertex( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "%s: nbox=%d\n", __FUNCTION__, sarea_priv->nbox );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 	if ( 0 )
 		radeon_print_dirty( "dispatch_vertex", sarea_priv->dirty );
 
@@ -845,8 +837,6 @@ static void radeon_cp_dispatch_indirect( drm_device_t *dev,
 	DRM_DEBUG( "indirect: buf=%d s=0x%x e=0x%x\n",
 		   buf->idx, start, end );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 	if ( start != end ) {
 		int offset = (dev_priv->agp_buffers_offset
 			      + buf->offset + start);
@@ -909,8 +899,6 @@ static void radeon_cp_dispatch_indices( drm_device_t *dev,
 	RING_LOCALS;
 	DRM_DEBUG( "indices: s=%d e=%d c=%d\n", start, end, count );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 	if ( 0 )
 		radeon_print_dirty( "dispatch_indices", sarea_priv->dirty );
 
@@ -1143,8 +1131,6 @@ static void radeon_cp_dispatch_stipple( drm_device_t *dev, u32 *stipple )
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	radeon_update_ring_snapshot( dev_priv );
-
 	BEGIN_RING( 35 );
 
 	OUT_RING( CP_PACKET0( RADEON_RE_STIPPLE_ADDR, 0 ) );
@@ -1180,7 +1166,6 @@ int radeon_cp_clear( struct inode *inode, struct file *filp,
 			     sizeof(clear) ) )
 		return -EFAULT;
 
-
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 
 	if ( sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS )