merged drm-filp-0-1-branchdrm-filp-1-20020328-merge

11 files changed, 285 insertions, 308 deletions

diff --git a/shared-core/mga_dma.c b/shared-core/mga_dma.c
index 22bd61f0..5e95c9f9 100644
--- a/shared-core/mga_dma.c
+++ b/shared-core/mga_dma.c
@@ -686,7 +686,7 @@ int mga_dma_flush( DRM_IOCTL_ARGS )
 	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
 	drm_lock_t lock;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( lock, (drm_lock_t *)data, sizeof(lock) );
 
@@ -720,7 +720,7 @@ int mga_dma_reset( DRM_IOCTL_ARGS )
 	DRM_DEVICE;
 	drm_mga_private_t *dev_priv = (drm_mga_private_t *)dev->dev_private;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	return mga_do_dma_reset( dev_priv );
 }
@@ -730,7 +730,8 @@ int mga_dma_reset( DRM_IOCTL_ARGS )
  * DMA buffer management
  */
 
-static int mga_dma_get_buffers( drm_device_t *dev, drm_dma_t *d )
+static int mga_dma_get_buffers( DRMFILE filp,
+				drm_device_t *dev, drm_dma_t *d )
 {
 	drm_buf_t *buf;
 	int i;
@@ -739,7 +740,7 @@ static int mga_dma_get_buffers( drm_device_t *dev, drm_dma_t *d )
 		buf = mga_freelist_get( dev );
 		if ( !buf ) return DRM_ERR(EAGAIN);
 
-		buf->pid = DRM_CURRENTPID;
+		buf->filp = filp;
 
 		if ( DRM_COPY_TO_USER( &d->request_indices[i],
 				   &buf->idx, sizeof(buf->idx) ) )
@@ -761,7 +762,7 @@ int mga_dma_buffers( DRM_IOCTL_ARGS )
 	drm_dma_t d;
 	int ret = 0;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( d, (drm_dma_t *)data, sizeof(d) );
 
@@ -786,7 +787,7 @@ int mga_dma_buffers( DRM_IOCTL_ARGS )
 	d.granted_count = 0;
 
 	if ( d.request_count ) {
-		ret = mga_dma_get_buffers( dev, &d );
+		ret = mga_dma_get_buffers( filp, dev, &d );
 	}
 
 	DRM_COPY_TO_USER_IOCTL( (drm_dma_t *)data, d, sizeof(d) );
diff --git a/shared-core/mga_drv.h b/shared-core/mga_drv.h
index a5085b06..7efc89bc 100644
--- a/shared-core/mga_drv.h
+++ b/shared-core/mga_drv.h
@@ -90,14 +90,14 @@ typedef struct drm_mga_private {
 	unsigned int texture_offset;
 	unsigned int texture_size;
 
-	drm_map_t *sarea;
-	drm_map_t *fb;
-	drm_map_t *mmio;
-	drm_map_t *status;
-	drm_map_t *warp;
-	drm_map_t *primary;
-	drm_map_t *buffers;
-	drm_map_t *agp_textures;
+	drm_local_map_t *sarea;
+	drm_local_map_t *fb;
+	drm_local_map_t *mmio;
+	drm_local_map_t *status;
+	drm_local_map_t *warp;
+	drm_local_map_t *primary;
+	drm_local_map_t *buffers;
+	drm_local_map_t *agp_textures;
 } drm_mga_private_t;
 
 				/* mga_dma.c */
@@ -131,32 +131,30 @@ extern int  mga_getparam( DRM_IOCTL_ARGS );
 extern int mga_warp_install_microcode( drm_mga_private_t *dev_priv );
 extern int mga_warp_init( drm_mga_private_t *dev_priv );
 
-#define mga_flush_write_combine()	DRM_WRITEMEMORYBARRIER()
-
+#define mga_flush_write_combine()	DRM_WRITEMEMORYBARRIER(dev_priv->primary)
 
+#if defined(__linux__) && defined(__alpha__)
 #define MGA_BASE( reg )		((unsigned long)(dev_priv->mmio->handle))
 #define MGA_ADDR( reg )		(MGA_BASE(reg) + reg)
 
 #define MGA_DEREF( reg )	*(volatile u32 *)MGA_ADDR( reg )
 #define MGA_DEREF8( reg )	*(volatile u8 *)MGA_ADDR( reg )
 
-#ifdef __alpha__
 #define MGA_READ( reg )		(_MGA_READ((u32 *)MGA_ADDR(reg)))
 #define MGA_READ8( reg )	(_MGA_READ((u8 *)MGA_ADDR(reg)))
-#define MGA_WRITE( reg, val )	do { DRM_WRITEMEMORYBARRIER(); MGA_DEREF( reg ) = val; } while (0)
-#define MGA_WRITE8( reg, val )  do { DRM_WRITEMEMORYBARRIER(); MGA_DEREF8( reg ) = val; } while (0)
+#define MGA_WRITE( reg, val )	do { DRM_WRITEMEMORYBARRIER(dev_priv->mmio); MGA_DEREF( reg ) = val; } while (0)
+#define MGA_WRITE8( reg, val )  do { DRM_WRITEMEMORYBARRIER(dev_priv->mmio); MGA_DEREF8( reg ) = val; } while (0)
 
 static inline u32 _MGA_READ(u32 *addr)
 {
-	DRM_READMEMORYBARRIER();
+	DRM_READMEMORYBARRIER(dev_priv->mmio);
 	return *(volatile u32 *)addr;
 }
-
 #else
-#define MGA_READ( reg )		MGA_DEREF( reg )
-#define MGA_READ8( reg )	MGA_DEREF8( reg )
-#define MGA_WRITE( reg, val )	do { MGA_DEREF( reg ) = val; } while (0)
-#define MGA_WRITE8( reg, val )  do { MGA_DEREF8( reg ) = val; } while (0)
+#define MGA_READ8( reg )	DRM_READ8(dev_priv->mmio, (reg))
+#define MGA_READ( reg )		DRM_READ32(dev_priv->mmio, (reg))
+#define MGA_WRITE8( reg, val )  DRM_WRITE8(dev_priv->mmio, (reg), (val))
+#define MGA_WRITE( reg, val )	DRM_WRITE32(dev_priv->mmio, (reg), (val))
 #endif
 
 #define DWGREG0 	0x1c00
@@ -186,16 +184,6 @@ do {									\
 	}								\
 } while (0)
 
-#define LOCK_TEST_WITH_RETURN( dev )					\
-do {									\
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||		\
-	     dev->lock.pid != DRM_CURRENTPID ) {				\
-		DRM_ERROR( "%s called without lock held\n",		\
-			   __FUNCTION__ );					\
-		return DRM_ERR(EINVAL);				\
-	}								\
-} while (0)
-
 #define WRAP_TEST_WITH_RETURN( dev_priv )				\
 do {									\
 	if ( test_bit( 0, &dev_priv->prim.wrapped ) ) {			\
diff --git a/shared-core/mga_state.c b/shared-core/mga_state.c
index 61077220..256dd47d 100644
--- a/shared-core/mga_state.c
+++ b/shared-core/mga_state.c
@@ -887,7 +887,7 @@ int mga_dma_clear( DRM_IOCTL_ARGS )
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	drm_mga_clear_t clear;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( clear, (drm_mga_clear_t *)data, sizeof(clear) );
 
@@ -911,7 +911,7 @@ int mga_dma_swap( DRM_IOCTL_ARGS )
 	drm_mga_private_t *dev_priv = dev->dev_private;
 	drm_mga_sarea_t *sarea_priv = dev_priv->sarea_priv;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( sarea_priv->nbox > MGA_NR_SAREA_CLIPRECTS )
 		sarea_priv->nbox = MGA_NR_SAREA_CLIPRECTS;
@@ -936,7 +936,7 @@ int mga_dma_vertex( DRM_IOCTL_ARGS )
 	drm_mga_buf_priv_t *buf_priv;
 	drm_mga_vertex_t vertex;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( vertex,
 			     (drm_mga_vertex_t *)data,
@@ -975,7 +975,7 @@ int mga_dma_indices( DRM_IOCTL_ARGS )
 	drm_mga_buf_priv_t *buf_priv;
 	drm_mga_indices_t indices;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( indices,
 			     (drm_mga_indices_t *)data,
@@ -1015,7 +1015,7 @@ int mga_dma_iload( DRM_IOCTL_ARGS )
 	drm_mga_iload_t iload;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( iload, (drm_mga_iload_t *)data, sizeof(iload) );
 
@@ -1055,7 +1055,7 @@ int mga_dma_blit( DRM_IOCTL_ARGS )
 	drm_mga_blit_t blit;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( blit, (drm_mga_blit_t *)data, sizeof(blit) );
 
diff --git a/shared-core/r128_cce.c b/shared-core/r128_cce.c
index 5175885e..7f0f4325 100644
--- a/shared-core/r128_cce.c
+++ b/shared-core/r128_cce.c
@@ -579,6 +579,7 @@ static int r128_do_init_cce( drm_device_t *dev, drm_r128_init_t *init )
 		(dev_priv->ring.size / sizeof(u32)) - 1;
 
 	dev_priv->ring.high_mark = 128;
+	dev_priv->ring.ring_rptr = dev_priv->ring_rptr;
 
 	dev_priv->sarea_priv->last_frame = 0;
 	R128_WRITE( R128_LAST_FRAME_REG, dev_priv->sarea_priv->last_frame );
@@ -663,7 +664,7 @@ int r128_cce_start( DRM_IOCTL_ARGS )
 	drm_r128_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( dev_priv->cce_running || dev_priv->cce_mode == R128_PM4_NONPM4 ) {
 		DRM_DEBUG( "%s while CCE running\n", __FUNCTION__ );
@@ -686,7 +687,7 @@ int r128_cce_stop( DRM_IOCTL_ARGS )
 	int ret;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL(stop, (drm_r128_cce_stop_t *)data, sizeof(stop) );
 
@@ -725,7 +726,7 @@ int r128_cce_reset( DRM_IOCTL_ARGS )
 	drm_r128_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_DEBUG( "%s called before init done\n", __FUNCTION__ );
@@ -746,7 +747,7 @@ int r128_cce_idle( DRM_IOCTL_ARGS )
 	drm_r128_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( dev_priv->cce_running ) {
 		r128_do_cce_flush( dev_priv );
@@ -760,7 +761,7 @@ int r128_engine_reset( DRM_IOCTL_ARGS )
 	DRM_DEVICE;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	return r128_do_engine_reset( dev );
 }
@@ -807,7 +808,7 @@ int r128_fullscreen( DRM_IOCTL_ARGS )
 	DRM_DEVICE;
 	drm_r128_fullscreen_t fs;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( fs, (drm_r128_fullscreen_t *)data, sizeof(fs) );
 
@@ -889,7 +890,7 @@ drm_buf_t *r128_freelist_get( drm_device_t *dev )
 	for ( i = 0 ; i < dma->buf_count ; i++ ) {
 		buf = dma->buflist[i];
 		buf_priv = buf->dev_private;
-		if ( buf->pid == 0 )
+		if ( buf->filp == 0 )
 			return buf;
 	}
 
@@ -948,7 +949,7 @@ int r128_wait_ring( drm_r128_private_t *dev_priv, int n )
 	return DRM_ERR(EBUSY);
 }
 
-static int r128_cce_get_buffers( drm_device_t *dev, drm_dma_t *d )
+static int r128_cce_get_buffers( DRMFILE filp, drm_device_t *dev, drm_dma_t *d )
 {
 	int i;
 	drm_buf_t *buf;
@@ -957,7 +958,7 @@ static int r128_cce_get_buffers( drm_device_t *dev, drm_dma_t *d )
 		buf = r128_freelist_get( dev );
 		if ( !buf ) return DRM_ERR(EAGAIN);
 
-		buf->pid = DRM_CURRENTPID;
+		buf->filp = filp;
 
 		if ( DRM_COPY_TO_USER( &d->request_indices[i], &buf->idx,
 				   sizeof(buf->idx) ) )
@@ -978,7 +979,7 @@ int r128_cce_buffers( DRM_IOCTL_ARGS )
 	int ret = 0;
 	drm_dma_t d;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( d, (drm_dma_t *) data, sizeof(d) );
 
@@ -1001,7 +1002,7 @@ int r128_cce_buffers( DRM_IOCTL_ARGS )
 	d.granted_count = 0;
 
 	if ( d.request_count ) {
-		ret = r128_cce_get_buffers( dev, &d );
+		ret = r128_cce_get_buffers( filp, dev, &d );
 	}
 
 	DRM_COPY_TO_USER_IOCTL((drm_dma_t *) data, d, sizeof(d) );
diff --git a/shared-core/r128_drv.h b/shared-core/r128_drv.h
index 763fcb3a..bd913878 100644
--- a/shared-core/r128_drv.h
+++ b/shared-core/r128_drv.h
@@ -34,8 +34,8 @@
 #ifndef __R128_DRV_H__
 #define __R128_DRV_H__
 
-#define GET_RING_HEAD(ring)		DRM_READ32(  (volatile u32 *) (ring)->head )
-#define SET_RING_HEAD(ring,val)		DRM_WRITE32( (volatile u32 *) (ring)->head, (val) )
+#define GET_RING_HEAD(ring)		DRM_READ32(  (ring)->ring_rptr, 0 ) /* (ring)->head */
+#define SET_RING_HEAD(ring,val)		DRM_WRITE32( (ring)->ring_rptr, 0, (val) ) /* (ring)->head */
 
 typedef struct drm_r128_freelist {
    	unsigned int age;
@@ -56,6 +56,7 @@ typedef struct drm_r128_ring_buffer {
 	int space;
 
 	int high_mark;
+	drm_local_map_t *ring_rptr;
 } drm_r128_ring_buffer_t;
 
 typedef struct drm_r128_private {
@@ -98,13 +99,13 @@ typedef struct drm_r128_private {
 	u32 depth_pitch_offset_c;
 	u32 span_pitch_offset_c;
 
-	drm_map_t *sarea;
-	drm_map_t *fb;
-	drm_map_t *mmio;
-	drm_map_t *cce_ring;
-	drm_map_t *ring_rptr;
-	drm_map_t *buffers;
-	drm_map_t *agp_textures;
+	drm_local_map_t *sarea;
+	drm_local_map_t *fb;
+	drm_local_map_t *mmio;
+	drm_local_map_t *cce_ring;
+	drm_local_map_t *ring_rptr;
+	drm_local_map_t *buffers;
+	drm_local_map_t *agp_textures;
 } drm_r128_private_t;
 
 typedef struct drm_r128_buf_priv {
@@ -370,15 +371,10 @@ extern int r128_cce_indirect( DRM_IOCTL_ARGS );
 
 #define R128_PERFORMANCE_BOXES		0
 
-
-#define R128_BASE(reg)		((unsigned long)(dev_priv->mmio->handle))
-#define R128_ADDR(reg)		(R128_BASE( reg ) + reg)
-
-#define R128_READ(reg)		DRM_READ32(  (volatile u32 *) R128_ADDR(reg) )
-#define R128_WRITE(reg,val)	DRM_WRITE32( (volatile u32 *) R128_ADDR(reg), (val) )
-
-#define R128_READ8(reg)		DRM_READ8(  (volatile u8 *) R128_ADDR(reg) )
-#define R128_WRITE8(reg,val)	DRM_WRITE8( (volatile u8 *) R128_ADDR(reg), (val) )
+#define R128_READ(reg)		DRM_READ32(  dev_priv->mmio, (reg) )
+#define R128_WRITE(reg,val)	DRM_WRITE32( dev_priv->mmio, (reg), (val) )
+#define R128_READ8(reg)		DRM_READ8(   dev_priv->mmio, (reg) )
+#define R128_WRITE8(reg,val)	DRM_WRITE8(  dev_priv->mmio, (reg), (val) )
 
 #define R128_WRITE_PLL(addr,val)					\
 do {									\
@@ -403,15 +399,6 @@ extern int R128_READ_PLL(drm_device_t *dev, int addr);
  * Misc helper macros
  */
 
-#define LOCK_TEST_WITH_RETURN( dev )					\
-do {									\
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||		\
-	     dev->lock.pid != DRM_CURRENTPID ) {			\
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );	\
-		return DRM_ERR(EINVAL);				\
-	}								\
-} while (0)
-
 #define RING_SPACE_TEST_WITH_RETURN( dev_priv )				\
 do {									\
 	drm_r128_ring_buffer_t *ring = &dev_priv->ring; int i;		\
@@ -453,7 +440,7 @@ do {									\
 #if defined(__powerpc__)
 #define r128_flush_write_combine()	(void) GET_RING_HEAD( &dev_priv->ring )
 #else
-#define r128_flush_write_combine()	DRM_WRITEMEMORYBARRIER()
+#define r128_flush_write_combine()	DRM_WRITEMEMORYBARRIER(dev_priv->ring_rptr)
 #endif
 
 
diff --git a/shared-core/r128_state.c b/shared-core/r128_state.c
index 68f73061..12c4a0ee 100644
--- a/shared-core/r128_state.c
+++ b/shared-core/r128_state.c
@@ -778,7 +778,8 @@ static void r128_cce_dispatch_indices( drm_device_t *dev,
 	sarea_priv->nbox = 0;
 }
 
-static int r128_cce_dispatch_blit( drm_device_t *dev,
+static int r128_cce_dispatch_blit( DRMFILE filp,
+				   drm_device_t *dev,
 				   drm_r128_blit_t *blit )
 {
 	drm_r128_private_t *dev_priv = dev->dev_private;
@@ -829,9 +830,9 @@ static int r128_cce_dispatch_blit( drm_device_t *dev,
 	buf = dma->buflist[blit->idx];
 	buf_priv = buf->dev_private;
 
-	if ( buf->pid != DRM_CURRENTPID ) {
+	if ( buf->filp != filp ) {
 		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
@@ -896,7 +897,7 @@ static int r128_cce_dispatch_write_span( drm_device_t *dev,
 	int count, x, y;
 	u32 *buffer;
 	u8 *mask;
-	int i;
+	int i, buffer_size, mask_size;
 	RING_LOCALS;
 	DRM_DEBUG( "\n" );
 
@@ -908,25 +909,25 @@ static int r128_cce_dispatch_write_span( drm_device_t *dev,
 		return DRM_ERR(EFAULT);
 	}
 
-	buffer = DRM_MALLOC( depth->n * sizeof(u32) );
+	buffer_size = depth->n * sizeof(u32);
+	buffer = DRM_MALLOC( buffer_size );
 	if ( buffer == NULL )
 		return DRM_ERR(ENOMEM);
-	if ( DRM_COPY_FROM_USER( buffer, depth->buffer,
-			     depth->n * sizeof(u32) ) ) {
-		DRM_FREE( buffer );
+	if ( DRM_COPY_FROM_USER( buffer, depth->buffer, buffer_size ) ) {
+		DRM_FREE( buffer, buffer_size);
 		return DRM_ERR(EFAULT);
 	}
 
+	mask_size = depth->n * sizeof(u8);
 	if ( depth->mask ) {
-		mask = DRM_MALLOC( depth->n * sizeof(u8) );
+		mask = DRM_MALLOC( mask_size );
 		if ( mask == NULL ) {
-			DRM_FREE( buffer );
+			DRM_FREE( buffer, buffer_size );
 			return DRM_ERR(ENOMEM);
 		}
-		if ( DRM_COPY_FROM_USER( mask, depth->mask,
-				     depth->n * sizeof(u8) ) ) {
-			DRM_FREE( buffer );
-			DRM_FREE( mask );
+		if ( DRM_COPY_FROM_USER( mask, depth->mask, mask_size ) ) {
+			DRM_FREE( buffer, buffer_size );
+			DRM_FREE( mask, mask_size );
 			return DRM_ERR(EFAULT);
 		}
 
@@ -953,7 +954,7 @@ static int r128_cce_dispatch_write_span( drm_device_t *dev,
 			}
 		}
 
-		DRM_FREE( mask );
+		DRM_FREE( mask, mask_size );
 	} else {
 		for ( i = 0 ; i < count ; i++, x++ ) {
 			BEGIN_RING( 6 );
@@ -977,7 +978,7 @@ static int r128_cce_dispatch_write_span( drm_device_t *dev,
 		}
 	}
 
-	DRM_FREE( buffer );
+	DRM_FREE( buffer, buffer_size );
 
 	return 0;
 }
@@ -989,60 +990,62 @@ static int r128_cce_dispatch_write_pixels( drm_device_t *dev,
 	int count, *x, *y;
 	u32 *buffer;
 	u8 *mask;
-	int i;
+	int i, xbuf_size, ybuf_size, buffer_size, mask_size;
 	RING_LOCALS;
 	DRM_DEBUG( "\n" );
 
 	count = depth->n;
 
-	x = DRM_MALLOC( count * sizeof(*x) );
+	xbuf_size = count * sizeof(*x);
+	ybuf_size = count * sizeof(*y);
+	x = DRM_MALLOC( xbuf_size );
 	if ( x == NULL ) {
 		return DRM_ERR(ENOMEM);
 	}
-	y = DRM_MALLOC( count * sizeof(*y) );
+	y = DRM_MALLOC( ybuf_size );
 	if ( y == NULL ) {
-		DRM_FREE( x );
+		DRM_FREE( x, xbuf_size );
 		return DRM_ERR(ENOMEM);
 	}
-	if ( DRM_COPY_FROM_USER( x, depth->x, count * sizeof(int) ) ) {
-		DRM_FREE( x );
-		DRM_FREE( y );
+	if ( DRM_COPY_FROM_USER( x, depth->x, xbuf_size ) ) {
+		DRM_FREE( x, xbuf_size );
+		DRM_FREE( y, ybuf_size );
 		return DRM_ERR(EFAULT);
 	}
-	if ( DRM_COPY_FROM_USER( y, depth->y, count * sizeof(int) ) ) {
-		DRM_FREE( x );
-		DRM_FREE( y );
+	if ( DRM_COPY_FROM_USER( y, depth->y, xbuf_size ) ) {
+		DRM_FREE( x, xbuf_size );
+		DRM_FREE( y, ybuf_size );
 		return DRM_ERR(EFAULT);
 	}
 
-	buffer = DRM_MALLOC( depth->n * sizeof(u32) );
+	buffer_size = depth->n * sizeof(u32);
+	buffer = DRM_MALLOC( buffer_size );
 	if ( buffer == NULL ) {
-		DRM_FREE( x );
-		DRM_FREE( y );
+		DRM_FREE( x, xbuf_size );
+		DRM_FREE( y, ybuf_size );
 		return DRM_ERR(ENOMEM);
 	}
-	if ( DRM_COPY_FROM_USER( buffer, depth->buffer,
-			     depth->n * sizeof(u32) ) ) {
-		DRM_FREE( x );
-		DRM_FREE( y );
-		DRM_FREE( buffer );
+	if ( DRM_COPY_FROM_USER( buffer, depth->buffer, buffer_size ) ) {
+		DRM_FREE( x, xbuf_size );
+		DRM_FREE( y, ybuf_size );
+		DRM_FREE( buffer, buffer_size );
 		return DRM_ERR(EFAULT);
 	}
 
 	if ( depth->mask ) {
-		mask = DRM_MALLOC( depth->n * sizeof(u8) );
+		mask_size = depth->n * sizeof(u8);
+		mask = DRM_MALLOC( mask_size );
 		if ( mask == NULL ) {
-			DRM_FREE( x );
-			DRM_FREE( y );
-			DRM_FREE( buffer );
+			DRM_FREE( x, xbuf_size );
+			DRM_FREE( y, ybuf_size );
+			DRM_FREE( buffer, buffer_size );
 			return DRM_ERR(ENOMEM);
 		}
-		if ( DRM_COPY_FROM_USER( mask, depth->mask,
-				     depth->n * sizeof(u8) ) ) {
-			DRM_FREE( x );
-			DRM_FREE( y );
-			DRM_FREE( buffer );
-			DRM_FREE( mask );
+		if ( DRM_COPY_FROM_USER( mask, depth->mask, mask_size ) ) {
+			DRM_FREE( x, xbuf_size );
+			DRM_FREE( y, ybuf_size );
+			DRM_FREE( buffer, buffer_size );
+			DRM_FREE( mask, mask_size );
 			return DRM_ERR(EFAULT);
 		}
 
@@ -1069,7 +1072,7 @@ static int r128_cce_dispatch_write_pixels( drm_device_t *dev,
 			}
 		}
 
-		DRM_FREE( mask );
+		DRM_FREE( mask, mask_size );
 	} else {
 		for ( i = 0 ; i < count ; i++ ) {
 			BEGIN_RING( 6 );
@@ -1093,9 +1096,9 @@ static int r128_cce_dispatch_write_pixels( drm_device_t *dev,
 		}
 	}
 
-	DRM_FREE( x );
-	DRM_FREE( y );
-	DRM_FREE( buffer );
+	DRM_FREE( x, xbuf_size );
+	DRM_FREE( y, ybuf_size );
+	DRM_FREE( buffer, buffer_size );
 
 	return 0;
 }
@@ -1146,7 +1149,7 @@ static int r128_cce_dispatch_read_pixels( drm_device_t *dev,
 {
 	drm_r128_private_t *dev_priv = dev->dev_private;
 	int count, *x, *y;
-	int i;
+	int i, xbuf_size, ybuf_size;
 	RING_LOCALS;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
@@ -1155,23 +1158,25 @@ static int r128_cce_dispatch_read_pixels( drm_device_t *dev,
 		count = dev_priv->depth_pitch;
 	}
 
-	x = DRM_MALLOC( count * sizeof(*x) );
+	xbuf_size = count * sizeof(*x);
+	ybuf_size = count * sizeof(*y);
+	x = DRM_MALLOC( xbuf_size );
 	if ( x == NULL ) {
 		return DRM_ERR(ENOMEM);
 	}
-	y = DRM_MALLOC( count * sizeof(*y) );
+	y = DRM_MALLOC( ybuf_size );
 	if ( y == NULL ) {
-		DRM_FREE( x );
+		DRM_FREE( x, xbuf_size );
 		return DRM_ERR(ENOMEM);
 	}
-	if ( DRM_COPY_FROM_USER( x, depth->x, count * sizeof(int) ) ) {
-		DRM_FREE( x );
-		DRM_FREE( y );
+	if ( DRM_COPY_FROM_USER( x, depth->x, xbuf_size ) ) {
+		DRM_FREE( x, xbuf_size );
+		DRM_FREE( y, ybuf_size );
 		return DRM_ERR(EFAULT);
 	}
-	if ( DRM_COPY_FROM_USER( y, depth->y, count * sizeof(int) ) ) {
-		DRM_FREE( x );
-		DRM_FREE( y );
+	if ( DRM_COPY_FROM_USER( y, depth->y, ybuf_size ) ) {
+		DRM_FREE( x, xbuf_size );
+		DRM_FREE( y, ybuf_size );
 		return DRM_ERR(EFAULT);
 	}
 
@@ -1199,8 +1204,8 @@ static int r128_cce_dispatch_read_pixels( drm_device_t *dev,
 		ADVANCE_RING();
 	}
 
-	DRM_FREE( x );
-	DRM_FREE( y );
+	DRM_FREE( x, xbuf_size );
+	DRM_FREE( y, ybuf_size );
 
 	return 0;
 }
@@ -1240,7 +1245,7 @@ int r128_cce_clear( DRM_IOCTL_ARGS )
 	drm_r128_clear_t clear;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( clear, (drm_r128_clear_t *) data,
 			     sizeof(clear) );
@@ -1266,7 +1271,7 @@ int r128_cce_swap( DRM_IOCTL_ARGS )
 	drm_r128_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	DRM_DEBUG( "%s\n", __FUNCTION__ );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 
@@ -1293,7 +1298,7 @@ int r128_cce_vertex( DRM_IOCTL_ARGS )
 	drm_r128_buf_priv_t *buf_priv;
 	drm_r128_vertex_t vertex;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1324,9 +1329,9 @@ int r128_cce_vertex( DRM_IOCTL_ARGS )
 	buf = dma->buflist[vertex.idx];
 	buf_priv = buf->dev_private;
 
-	if ( buf->pid != DRM_CURRENTPID ) {
+	if ( buf->filp != filp ) {
 		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
@@ -1353,7 +1358,7 @@ int r128_cce_indices( DRM_IOCTL_ARGS )
 	drm_r128_indices_t elts;
 	int count;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1383,9 +1388,9 @@ int r128_cce_indices( DRM_IOCTL_ARGS )
 	buf = dma->buflist[elts.idx];
 	buf_priv = buf->dev_private;
 
-	if ( buf->pid != DRM_CURRENTPID ) {
+	if ( buf->filp != filp ) {
 		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
@@ -1421,7 +1426,7 @@ int r128_cce_blit( DRM_IOCTL_ARGS )
 	drm_r128_private_t *dev_priv = dev->dev_private;
 	drm_r128_blit_t blit;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( blit, (drm_r128_blit_t *) data,
 			     sizeof(blit) );
@@ -1437,7 +1442,7 @@ int r128_cce_blit( DRM_IOCTL_ARGS )
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 	VB_AGE_TEST_WITH_RETURN( dev_priv );
 
-	return r128_cce_dispatch_blit( dev, &blit );
+	return r128_cce_dispatch_blit( filp, dev, &blit );
 }
 
 int r128_cce_depth( DRM_IOCTL_ARGS )
@@ -1446,7 +1451,7 @@ int r128_cce_depth( DRM_IOCTL_ARGS )
 	drm_r128_private_t *dev_priv = dev->dev_private;
 	drm_r128_depth_t depth;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( depth, (drm_r128_depth_t *) data,
 			     sizeof(depth) );
@@ -1474,7 +1479,7 @@ int r128_cce_stipple( DRM_IOCTL_ARGS )
 	drm_r128_stipple_t stipple;
 	u32 mask[32];
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( stipple, (drm_r128_stipple_t *) data,
 			     sizeof(stipple) );
@@ -1502,7 +1507,7 @@ int r128_cce_indirect( DRM_IOCTL_ARGS )
 	RING_LOCALS;
 #endif
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1525,9 +1530,9 @@ int r128_cce_indirect( DRM_IOCTL_ARGS )
 	buf = dma->buflist[indirect.idx];
 	buf_priv = buf->dev_private;
 
-	if ( buf->pid != DRM_CURRENTPID ) {
+	if ( buf->filp != filp ) {
 		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
diff --git a/shared-core/radeon_cp.c b/shared-core/radeon_cp.c
index b4d0e4b6..3ec8dfd1 100644
--- a/shared-core/radeon_cp.c
+++ b/shared-core/radeon_cp.c
@@ -36,6 +36,11 @@
 
 #define RADEON_FIFO_DEBUG	0
 
+#if defined(__alpha__) || defined(__powerpc__)
+# define PCIGART_ENABLED
+#else
+# undef PCIGART_ENABLED
+#endif
 
 
 /* CP microcode (from ATI) */
@@ -880,7 +885,6 @@ static void radeon_cp_init_ring_buffer( drm_device_t *dev,
 
 	/* Set the write pointer delay */
 	RADEON_WRITE( RADEON_CP_RB_WPTR_DELAY, 0 );
-	RADEON_READ( RADEON_CP_RB_WPTR_DELAY ); /* read back to propagate */
 
 	/* Initialize the ring buffer's read and write pointers */
 	cur_read_ptr = RADEON_READ( RADEON_CP_RB_RPTR );
@@ -922,11 +926,11 @@ static void radeon_cp_init_ring_buffer( drm_device_t *dev,
 	RADEON_WRITE( RADEON_SCRATCH_UMSK, 0x7 );
 
 	/* Writeback doesn't seem to work everywhere, test it first */
-	DRM_WRITE32( &dev_priv->scratch[1], 0 );
+	DRM_WRITE32( dev_priv->ring_rptr, RADEON_SCRATCHOFF(1), 0 );
 	RADEON_WRITE( RADEON_SCRATCH_REG1, 0xdeadbeef );
 
 	for ( tmp = 0 ; tmp < dev_priv->usec_timeout ; tmp++ ) {
-		if ( DRM_READ32( &dev_priv->scratch[1] ) == 0xdeadbeef )
+		if ( DRM_READ32( dev_priv->ring_rptr, RADEON_SCRATCHOFF(1) ) == 0xdeadbeef )
 			break;
 		DRM_UDELAY( 1 );
 	}
@@ -986,6 +990,17 @@ static int radeon_do_init_cp( drm_device_t *dev, drm_radeon_init_t *init )
 
 	dev_priv->is_pci = init->is_pci;
 
+#if !defined(PCIGART_ENABLED)
+	/* PCI support is not 100% working, so we disable it here.
+	 */
+	if ( dev_priv->is_pci ) {
+		DRM_ERROR( "PCI GART not yet supported for Radeon!\n" );
+		dev->dev_private = (void *)dev_priv;
+		radeon_do_cleanup_cp(dev);
+		return DRM_ERR(EINVAL);
+	}
+#endif
+
 	if ( dev_priv->is_pci && !dev->sg ) {
 		DRM_ERROR( "PCI GART memory not allocated!\n" );
 		dev->dev_private = (void *)dev_priv;
@@ -1202,6 +1217,7 @@ static int radeon_do_init_cp( drm_device_t *dev, drm_radeon_init_t *init )
 		(dev_priv->ring.size / sizeof(u32)) - 1;
 
 	dev_priv->ring.high_mark = RADEON_RING_HIGH_MARK;
+	dev_priv->ring.ring_rptr = dev_priv->ring_rptr;
 
 #if __REALLY_HAVE_SG
 	if ( dev_priv->is_pci ) {
@@ -1307,7 +1323,7 @@ int radeon_cp_start( DRM_IOCTL_ARGS )
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( dev_priv->cp_running ) {
 		DRM_DEBUG( "%s while CP running\n", __FUNCTION__ );
@@ -1335,10 +1351,13 @@ int radeon_cp_stop( DRM_IOCTL_ARGS )
 	int ret;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( stop, (drm_radeon_cp_stop_t *)data, sizeof(stop) );
 
+	if (!dev_priv->cp_running)
+		return 0;
+
 	/* Flush any pending CP commands.  This ensures any outstanding
 	 * commands are exectuted by the engine before we turn it off.
 	 */
@@ -1366,6 +1385,39 @@ int radeon_cp_stop( DRM_IOCTL_ARGS )
 	return 0;
 }
 
+
+void radeon_do_release( drm_device_t *dev )
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	int ret;
+
+	if (dev_priv) {
+		if (dev_priv->cp_running) {
+			/* Stop the cp */
+			while ((ret = radeon_do_cp_idle( dev_priv )) != 0) {
+				DRM_DEBUG("radeon_do_cp_idle %d\n", ret);
+#ifdef __linux__
+				schedule();
+#else
+				tsleep(&ret, PZERO, "rdnrel", 1);
+#endif
+			}
+			radeon_do_cp_stop( dev_priv );
+			radeon_do_engine_reset( dev );
+		}
+
+		/* Disable *all* interrupts */
+		RADEON_WRITE( RADEON_GEN_INT_CNTL, 0 );
+
+		/* Free memory heap structures */
+		radeon_mem_takedown( &(dev_priv->agp_heap) );
+		radeon_mem_takedown( &(dev_priv->fb_heap) );
+
+		/* deallocate kernel resources */
+		radeon_do_cleanup_cp( dev );
+	}
+}
+
 /* Just reset the CP ring.  Called as part of an X Server engine reset.
  */
 int radeon_cp_reset( DRM_IOCTL_ARGS )
@@ -1374,7 +1426,7 @@ int radeon_cp_reset( DRM_IOCTL_ARGS )
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_DEBUG( "%s called before init done\n", __FUNCTION__ );
@@ -1395,10 +1447,7 @@ int radeon_cp_idle( DRM_IOCTL_ARGS )
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
-
-/* 	if (dev->irq)  */
-/* 		radeon_emit_and_wait_irq( dev ); */
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	return radeon_do_cp_idle( dev_priv );
 }
@@ -1408,7 +1457,7 @@ int radeon_engine_reset( DRM_IOCTL_ARGS )
 	DRM_DEVICE;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	return radeon_do_engine_reset( dev );
 }
@@ -1467,7 +1516,7 @@ drm_buf_t *radeon_freelist_get( drm_device_t *dev )
 		for ( i = start ; i < dma->buf_count ; i++ ) {
 			buf = dma->buflist[i];
 			buf_priv = buf->dev_private;
-			if ( buf->pid == 0 || (buf->pending && 
+			if ( buf->filp == 0 || (buf->pending && 
 					       buf_priv->age <= done_age) ) {
 				dev_priv->stats.requested_bufs++;
 				buf->pending = 0;
@@ -1494,7 +1543,7 @@ drm_buf_t *radeon_freelist_get( drm_device_t *dev )
 	drm_buf_t *buf;
 	int i, t;
 	int start;
-	u32 done_age = DRM_READ32(&dev_priv->scratch[1]);
+	u32 done_age = DRM_READ32(dev_priv->ring_rptr, RADEON_SCRATCHOFF(1));
 
 	if ( ++dev_priv->last_buf >= dma->buf_count )
 		dev_priv->last_buf = 0;
@@ -1506,7 +1555,7 @@ drm_buf_t *radeon_freelist_get( drm_device_t *dev )
 		for ( i = start ; i < dma->buf_count ; i++ ) {
 			buf = dma->buflist[i];
 			buf_priv = buf->dev_private;
-			if ( buf->pid == 0 || (buf->pending && 
+			if ( buf->filp == 0 || (buf->pending && 
 					       buf_priv->age <= done_age) ) {
 				dev_priv->stats.requested_bufs++;
 				buf->pending = 0;
@@ -1571,7 +1620,7 @@ int radeon_wait_ring( drm_radeon_private_t *dev_priv, int n )
 	return DRM_ERR(EBUSY);
 }
 
-static int radeon_cp_get_buffers( drm_device_t *dev, drm_dma_t *d )
+static int radeon_cp_get_buffers( DRMFILE filp, drm_device_t *dev, drm_dma_t *d )
 {
 	int i;
 	drm_buf_t *buf;
@@ -1580,7 +1629,7 @@ static int radeon_cp_get_buffers( drm_device_t *dev, drm_dma_t *d )
 		buf = radeon_freelist_get( dev );
 		if ( !buf ) return DRM_ERR(EBUSY); /* NOTE: broken client */
 
-		buf->pid = DRM_CURRENTPID;
+		buf->filp = filp;
 
 		if ( DRM_COPY_TO_USER( &d->request_indices[i], &buf->idx,
 				   sizeof(buf->idx) ) )
@@ -1601,7 +1650,7 @@ int radeon_cp_buffers( DRM_IOCTL_ARGS )
 	int ret = 0;
 	drm_dma_t d;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( d, (drm_dma_t *)data, sizeof(d) );
 
@@ -1624,7 +1673,7 @@ int radeon_cp_buffers( DRM_IOCTL_ARGS )
 	d.granted_count = 0;
 
 	if ( d.request_count ) {
-		ret = radeon_cp_get_buffers( dev, &d );
+		ret = radeon_cp_get_buffers( filp, dev, &d );
 	}
 
 	DRM_COPY_TO_USER_IOCTL( (drm_dma_t *)data, d, sizeof(d) );
diff --git a/shared-core/radeon_drv.h b/shared-core/radeon_drv.h
index 635ad14b..7faffa7a 100644
--- a/shared-core/radeon_drv.h
+++ b/shared-core/radeon_drv.h
@@ -31,8 +31,8 @@
 #ifndef __RADEON_DRV_H__
 #define __RADEON_DRV_H__
 
-#define GET_RING_HEAD(ring)		DRM_READ32(  (volatile u32 *) (ring)->head )
-#define SET_RING_HEAD(ring,val)		DRM_WRITE32( (volatile u32 *) (ring)->head , (val))
+#define GET_RING_HEAD(ring)		DRM_READ32(  (ring)->ring_rptr, 0 ) /* (ring)->head */
+#define SET_RING_HEAD(ring,val)		DRM_WRITE32( (ring)->ring_rptr, 0, (val) ) /* (ring)->head */
 
 typedef struct drm_radeon_freelist {
    	unsigned int age;
@@ -53,6 +53,7 @@ typedef struct drm_radeon_ring_buffer {
 	int space;
 
 	int high_mark;
+	drm_local_map_t *ring_rptr;
 } drm_radeon_ring_buffer_t;
 
 typedef struct drm_radeon_depth_clear_t {
@@ -67,7 +68,7 @@ struct mem_block {
 	struct mem_block *prev;
 	int start;
 	int size;
-	int pid;		/* 0: free, -1: heap, other: real pids */
+	DRMFILE filp;		/* 0: free, -1: heap, other: real files */
 };
 
 typedef struct drm_radeon_private {
@@ -126,13 +127,13 @@ typedef struct drm_radeon_private {
 
 	drm_radeon_depth_clear_t depth_clear;
 
-	drm_map_t *sarea;
-	drm_map_t *fb;
-	drm_map_t *mmio;
-	drm_map_t *cp_ring;
-	drm_map_t *ring_rptr;
-	drm_map_t *buffers;
-	drm_map_t *agp_textures;
+	drm_local_map_t *sarea;
+	drm_local_map_t *fb;
+	drm_local_map_t *mmio;
+	drm_local_map_t *cp_ring;
+	drm_local_map_t *ring_rptr;
+	drm_local_map_t *buffers;
+	drm_local_map_t *agp_textures;
 
 	struct mem_block *agp_heap;
 	struct mem_block *fb_heap;
@@ -183,7 +184,7 @@ extern int radeon_mem_alloc( DRM_IOCTL_ARGS );
 extern int radeon_mem_free( DRM_IOCTL_ARGS );
 extern int radeon_mem_init_heap( DRM_IOCTL_ARGS );
 extern void radeon_mem_takedown( struct mem_block **heap );
-extern void radeon_mem_release( struct mem_block *heap );
+extern void radeon_mem_release( DRMFILE filp, struct mem_block *heap );
 
 				/* radeon_irq.c */
 extern int radeon_irq_emit( DRM_IOCTL_ARGS );
@@ -193,6 +194,7 @@ extern int radeon_emit_and_wait_irq(drm_device_t *dev);
 extern int radeon_wait_irq(drm_device_t *dev, int swi_nr);
 extern int radeon_emit_irq(drm_device_t *dev);
 
+extern void radeon_do_release(drm_device_t *dev);
 
 /* Flags for stats.boxes
  */
@@ -266,8 +268,10 @@ extern int radeon_emit_irq(drm_device_t *dev);
 #define RADEON_SCRATCH_UMSK		0x0770
 #define RADEON_SCRATCH_ADDR		0x0774
 
+#define RADEON_SCRATCHOFF( x )		(RADEON_SCRATCH_REG_OFFSET + 4*(x))
+
 #define GET_SCRATCH( x )	(dev_priv->writeback_works			\
-				? DRM_READ32( &dev_priv->scratch[(x)] )		\
+				? DRM_READ32( dev_priv->ring_rptr, RADEON_SCRATCHOFF(x) ) \
 				: RADEON_READ( RADEON_SCRATCH_REG0 + 4*(x) ) )
 
 
@@ -686,15 +690,10 @@ extern int radeon_emit_irq(drm_device_t *dev);
 
 #define RADEON_RING_HIGH_MARK		128
 
-
-#define RADEON_BASE(reg)	((unsigned long)(dev_priv->mmio->handle))
-#define RADEON_ADDR(reg)	(RADEON_BASE( reg ) + reg)
-
-#define RADEON_READ(reg)	DRM_READ32(  (volatile u32 *) RADEON_ADDR(reg) )
-#define RADEON_WRITE(reg,val)	DRM_WRITE32( (volatile u32 *) RADEON_ADDR(reg), (val) )
-
-#define RADEON_READ8(reg)	DRM_READ8(  (volatile u8 *) RADEON_ADDR(reg) )
-#define RADEON_WRITE8(reg,val)	DRM_WRITE8( (volatile u8 *) RADEON_ADDR(reg), (val) )
+#define RADEON_READ(reg)	DRM_READ32(  dev_priv->mmio, (reg) )
+#define RADEON_WRITE(reg,val)	DRM_WRITE32( dev_priv->mmio, (reg), (val) )
+#define RADEON_READ8(reg)	DRM_READ8(  dev_priv->mmio, (reg) )
+#define RADEON_WRITE8(reg,val)	DRM_WRITE8( dev_priv->mmio, (reg), (val) )
 
 #define RADEON_WRITE_PLL( addr, val )					\
 do {									\
@@ -771,16 +770,6 @@ extern int RADEON_READ_PLL( drm_device_t *dev, int addr );
  * Misc helper macros
  */
 
-#define LOCK_TEST_WITH_RETURN( dev )					\
-do {									\
-	if ( !_DRM_LOCK_IS_HELD( dev->lock.hw_lock->lock ) ||		\
-	     dev->lock.pid != DRM_CURRENTPID ) {			\
-		DRM_ERROR( "%s called without lock held\n", __FUNCTION__ );	\
-		return DRM_ERR(EINVAL);				\
-	}								\
-} while (0)
-
-
 /* Perfbox functionality only.  
  */
 #define RING_SPACE_TEST_WITH_RETURN( dev_priv )				\
@@ -827,43 +816,10 @@ do {									\
 
 #define RING_LOCALS	int write, _nr; unsigned int mask; u32 *ring;
 
-#if defined(__alpha__)
-#  define RADEON_PAD_RING 16 /* pad ring requests to 16 lw boundaries */
-#else
-#  define RADEON_PAD_RING 0
-#endif
-
-#if RADEON_PAD_RING
-# define radeon_pad_size(n) 	\
-	(((RADEON_PAD_RING) - ((n) % (RADEON_PAD_RING))) % (RADEON_PAD_RING))
-# define radeon_pad_ring() do {						\
-	if (RADEON_VERBOSE) {						\
-		DRM_INFO("Padding ring from %d (%x) with %d words\n", 	\
-			 write, write, radeon_pad_size(write));		\
-	}								\
-	switch (radeon_pad_size(write)) {			      	\
-	case 0:  /* aligned */					      	\
-		break;						      	\
-	case 1:  /* 1 word */					      	\
-		OUT_RING(CP_PACKET2());				      	\
-		break;						      	\
-	default: /* >= 2 words */				      	\
-		OUT_RING(CP_PACKET3(0x1000, radeon_pad_size(write) - 1));\
-		write = (write + radeon_pad_size(write)) & mask;	\
-		write &= mask;						\
-		break;						      	\
-	}							      	\
-} while(0)
-#else
-# define radeon_pad_size(n) 0
-# define radeon_pad_ring()
-#endif
-
-#define BEGIN_RING( req_n ) do {					\
-	int n = req_n + radeon_pad_size(req_n);				\
+#define BEGIN_RING( n ) do {						\
 	if ( RADEON_VERBOSE ) {						\
-		DRM_INFO( "BEGIN_RING( %d (%d) ) in %s\n",		\
-			   n, req_n, __FUNCTION__ );			\
+		DRM_INFO( "BEGIN_RING( %d ) in %s\n",			\
+			   n, __FUNCTION__ );				\
 	}								\
 	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {		\
                 COMMIT_RING();						\
@@ -880,7 +836,6 @@ do {									\
 		DRM_INFO( "ADVANCE_RING() wr=0x%06x tail=0x%06x\n",	\
 			  write, dev_priv->ring.tail );			\
 	}								\
-	radeon_pad_ring();						\
 	if (((dev_priv->ring.tail + _nr) & mask) != write) {		\
 		DRM_ERROR( 						\
 			"ADVANCE_RING(): mismatch: nr: %x write: %x line: %d\n",	\
@@ -892,7 +847,7 @@ do {									\
 
 #define COMMIT_RING() do {						\
 	/* Flush writes to ring */					\
-	DRM_READMEMORYBARRIER();					\
+	DRM_READMEMORYBARRIER(dev_priv->mmio);					\
 	GET_RING_HEAD( &dev_priv->ring );				\
 	RADEON_WRITE( RADEON_CP_RB_WPTR, dev_priv->ring.tail );		\
 	/* read from PCI bus to ensure correct posting */		\
diff --git a/shared-core/radeon_irq.c b/shared-core/radeon_irq.c
index 596706bf..9199fbbf 100644
--- a/shared-core/radeon_irq.c
+++ b/shared-core/radeon_irq.c
@@ -179,7 +179,7 @@ int radeon_irq_emit( DRM_IOCTL_ARGS )
 	drm_radeon_irq_emit_t emit;
 	int result;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
diff --git a/shared-core/radeon_mem.c b/shared-core/radeon_mem.c
index 5c07c1af..c3cbd3a9 100644
--- a/shared-core/radeon_mem.c
+++ b/shared-core/radeon_mem.c
@@ -40,7 +40,7 @@
  */
 
 static struct mem_block *split_block(struct mem_block *p, int start, int size,
-				     int pid )
+				     DRMFILE filp )
 {
 	/* Maybe cut off the start of an existing block */
 	if (start > p->start) {
@@ -49,7 +49,7 @@ static struct mem_block *split_block(struct mem_block *p, int start, int size,
 			goto out;
 		newblock->start = start;
 		newblock->size = p->size - (start - p->start);
-		newblock->pid = 0;
+		newblock->filp = 0;
 		newblock->next = p->next;
 		newblock->prev = p;
 		p->next->prev = newblock;
@@ -65,7 +65,7 @@ static struct mem_block *split_block(struct mem_block *p, int start, int size,
 			goto out;
 		newblock->start = start + size;
 		newblock->size = p->size - size;
-		newblock->pid = 0;
+		newblock->filp = 0;
 		newblock->next = p->next;
 		newblock->prev = p;
 		p->next->prev = newblock;
@@ -75,20 +75,20 @@ static struct mem_block *split_block(struct mem_block *p, int start, int size,
 
  out:
 	/* Our block is in the middle */
-	p->pid = pid;
+	p->filp = filp;
 	return p;
 }
 
 static struct mem_block *alloc_block( struct mem_block *heap, int size, 
-				      int align2, int pid )
+				      int align2, DRMFILE filp )
 {
 	struct mem_block *p;
 	int mask = (1 << align2)-1;
 
 	for (p = heap->next ; p != heap ; p = p->next) {
 		int start = (p->start + mask) & ~mask;
-		if (p->pid == 0 && start + size <= p->start + p->size)
-			return split_block( p, start, size, pid );
+		if (p->filp == 0 && start + size <= p->start + p->size)
+			return split_block( p, start, size, filp );
 	}
 
 	return NULL;
@@ -108,38 +108,28 @@ static struct mem_block *find_block( struct mem_block *heap, int start )
 
 static void free_block( struct mem_block *p )
 {
-	p->pid = 0;
+	p->filp = 0;
 
-	/* Assumes a single contiguous range.  Needs a special pid in
+	/* Assumes a single contiguous range.  Needs a special filp in
 	 * 'heap' to stop it being subsumed.
 	 */
-	if (p->next->pid == 0) {
+	if (p->next->filp == 0) {
 		struct mem_block *q = p->next;
 		p->size += q->size;
 		p->next = q->next;
 		p->next->prev = p;
-		DRM_FREE(q);
+		DRM_FREE(q, sizeof(*q));
 	}
 
-	if (p->prev->pid == 0) {
+	if (p->prev->filp == 0) {
 		struct mem_block *q = p->prev;
 		q->size += p->size;
 		q->next = p->next;
 		q->next->prev = q;
-		DRM_FREE(p);
+		DRM_FREE(p, sizeof(*q));
 	}
 }
 
-static void print_heap( struct mem_block *heap )
-{
-	struct mem_block *p;
-
-	for (p = heap->next ; p != heap ; p = p->next) 
-		DRM_DEBUG("0x%x..0x%x (0x%x) -- owner %d\n",
-			  p->start, p->start + p->size,
-			  p->size, p->pid);
-}
-
 /* Initialize.  How to check for an uninitialized heap?
  */
 static int init_heap(struct mem_block **heap, int start, int size)
@@ -151,47 +141,46 @@ static int init_heap(struct mem_block **heap, int start, int size)
 	
 	*heap = DRM_MALLOC(sizeof(**heap));
 	if (!*heap) {
-		DRM_FREE( blocks );
+		DRM_FREE( blocks, sizeof(*blocks) );
 		return -ENOMEM;
 	}
 
 	blocks->start = start;
 	blocks->size = size;
-	blocks->pid = 0;
+	blocks->filp = 0;
 	blocks->next = blocks->prev = *heap;
 
 	memset( *heap, 0, sizeof(**heap) );
-	(*heap)->pid = -1;
+	(*heap)->filp = (DRMFILE) -1;
 	(*heap)->next = (*heap)->prev = blocks;
 	return 0;
 }
 
 
-/* Free all blocks associated with the releasing pid.
+/* Free all blocks associated with the releasing file.
  */
-void radeon_mem_release( struct mem_block *heap )
+void radeon_mem_release( DRMFILE filp, struct mem_block *heap )
 {
-	int pid = DRM_CURRENTPID;
 	struct mem_block *p;
 
 	if (!heap || !heap->next)
 		return;
 
 	for (p = heap->next ; p != heap ; p = p->next) {
-		if (p->pid == pid) 
-			p->pid = 0;
+		if (p->filp == filp) 
+			p->filp = 0;
 	}
 
-	/* Assumes a single contiguous range.  Needs a special pid in
+	/* Assumes a single contiguous range.  Needs a special filp in
 	 * 'heap' to stop it being subsumed.
 	 */
 	for (p = heap->next ; p != heap ; p = p->next) {
-		while (p->pid == 0 && p->next->pid == 0) {
+		while (p->filp == 0 && p->next->filp == 0) {
 			struct mem_block *q = p->next;
 			p->size += q->size;
 			p->next = q->next;
 			p->next->prev = p;
-			DRM_FREE(q);
+			DRM_FREE(q, sizeof(*q));
 		}
 	}
 }
@@ -208,10 +197,10 @@ void radeon_mem_takedown( struct mem_block **heap )
 	for (p = (*heap)->next ; p != *heap ; ) {
 		struct mem_block *q = p;
 		p = p->next;
-		DRM_FREE(q);
+		DRM_FREE(q, sizeof(*q));
 	}
 
-	DRM_FREE( *heap );
+	DRM_FREE( *heap, sizeof(**heap) );
 	*heap = 0;
 }
 
@@ -258,7 +247,7 @@ int radeon_mem_alloc( DRM_IOCTL_ARGS )
 		alloc.alignment = 12;
 
 	block = alloc_block( *heap, alloc.size, alloc.alignment,
-			     DRM_CURRENTPID );
+			     filp );
 
 	if (!block) 
 		return DRM_ERR(ENOMEM);
@@ -297,7 +286,7 @@ int radeon_mem_free( DRM_IOCTL_ARGS )
 	if (!block)
 		return DRM_ERR(EFAULT);
 
-	if (block->pid != DRM_CURRENTPID)
+	if (block->filp != filp)
 		return DRM_ERR(EPERM);
 
 	free_block( block );	
diff --git a/shared-core/radeon_state.c b/shared-core/radeon_state.c
index d9bc948c..86cbead5 100644
--- a/shared-core/radeon_state.c
+++ b/shared-core/radeon_state.c
@@ -1063,7 +1063,8 @@ static void radeon_cp_dispatch_indices( drm_device_t *dev,
 
 #define RADEON_MAX_TEXTURE_SIZE (RADEON_BUFFER_SIZE - 8 * sizeof(u32))
 
-static int radeon_cp_dispatch_texture( drm_device_t *dev,
+static int radeon_cp_dispatch_texture( DRMFILE filp,
+				       drm_device_t *dev,
 				       drm_radeon_texture_t *tex,
 				       drm_radeon_tex_image_t *image )
 {
@@ -1216,7 +1217,7 @@ static int radeon_cp_dispatch_texture( drm_device_t *dev,
 			}
 		}
 
-		buf->pid = DRM_CURRENTPID;
+		buf->filp = filp;
 		buf->used = (dwords + 8) * sizeof(u32);
 		radeon_cp_dispatch_indirect( dev, buf, 0, buf->used );
 		radeon_cp_discard_buffer( dev, buf );
@@ -1273,7 +1274,7 @@ int radeon_cp_clear( DRM_IOCTL_ARGS )
 	drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( clear, (drm_radeon_clear_t *)data,
 			     sizeof(clear) );
@@ -1342,7 +1343,7 @@ int radeon_cp_flip( DRM_IOCTL_ARGS )
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 
@@ -1362,7 +1363,7 @@ int radeon_cp_swap( DRM_IOCTL_ARGS )
 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
 	DRM_DEBUG( "\n" );
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 
@@ -1386,7 +1387,7 @@ int radeon_cp_vertex( DRM_IOCTL_ARGS )
 	drm_radeon_vertex_t vertex;
 	drm_radeon_tcl_prim_t prim;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1416,9 +1417,9 @@ int radeon_cp_vertex( DRM_IOCTL_ARGS )
 
 	buf = dma->buflist[vertex.idx];
 
-	if ( buf->pid != DRM_CURRENTPID ) {
-		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+	if ( buf->filp != filp ) {
+		DRM_ERROR( "process %d using buffer owned by %p\n",
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
@@ -1473,7 +1474,7 @@ int radeon_cp_indices( DRM_IOCTL_ARGS )
 	drm_radeon_tcl_prim_t prim;
 	int count;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1503,9 +1504,9 @@ int radeon_cp_indices( DRM_IOCTL_ARGS )
 
 	buf = dma->buflist[elts.idx];
 
-	if ( buf->pid != DRM_CURRENTPID ) {
-		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+	if ( buf->filp != filp ) {
+		DRM_ERROR( "process %d using buffer owned by %p\n",
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
@@ -1568,7 +1569,7 @@ int radeon_cp_texture( DRM_IOCTL_ARGS )
 	drm_radeon_tex_image_t image;
 	int ret;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( tex, (drm_radeon_texture_t *)data, sizeof(tex) );
 
@@ -1585,7 +1586,7 @@ int radeon_cp_texture( DRM_IOCTL_ARGS )
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 	VB_AGE_TEST_WITH_RETURN( dev_priv );
 
-	ret = radeon_cp_dispatch_texture( dev, &tex, &image );
+	ret = radeon_cp_dispatch_texture( filp, dev, &tex, &image );
 
 	COMMIT_RING();
 	return ret;
@@ -1598,7 +1599,7 @@ int radeon_cp_stipple( DRM_IOCTL_ARGS )
 	drm_radeon_stipple_t stipple;
 	u32 mask[32];
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	DRM_COPY_FROM_USER_IOCTL( stipple, (drm_radeon_stipple_t *)data,
 			     sizeof(stipple) );
@@ -1623,7 +1624,7 @@ int radeon_cp_indirect( DRM_IOCTL_ARGS )
 	drm_radeon_indirect_t indirect;
 	RING_LOCALS;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1645,9 +1646,9 @@ int radeon_cp_indirect( DRM_IOCTL_ARGS )
 
 	buf = dma->buflist[indirect.idx];
 
-	if ( buf->pid != DRM_CURRENTPID ) {
-		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+	if ( buf->filp != filp ) {
+		DRM_ERROR( "process %d using buffer owned by %p\n",
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 	if ( buf->pending ) {
@@ -1700,7 +1701,7 @@ int radeon_cp_vertex2( DRM_IOCTL_ARGS )
 	int i;
 	unsigned char laststate;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -1725,9 +1726,9 @@ int radeon_cp_vertex2( DRM_IOCTL_ARGS )
 
 	buf = dma->buflist[vertex.idx];
 
-	if ( buf->pid != DRM_CURRENTPID ) {
-		DRM_ERROR( "process %d using buffer owned by %d\n",
-			   DRM_CURRENTPID, buf->pid );
+	if ( buf->filp != filp ) {
+		DRM_ERROR( "process %d using buffer owned by %p\n",
+			   DRM_CURRENTPID, buf->filp );
 		return DRM_ERR(EINVAL);
 	}
 
@@ -2027,7 +2028,7 @@ int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
 	drm_radeon_cmd_header_t header;
 	int orig_nbox;
 
-	LOCK_TEST_WITH_RETURN( dev );
+	LOCK_TEST_WITH_RETURN( dev, filp );
 
 	if ( !dev_priv ) {
 		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
@@ -2096,8 +2097,9 @@ int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
 			}
 
 			buf = dma->buflist[idx];
-			if ( buf->pid != DRM_CURRENTPID || buf->pending ) {
-				DRM_ERROR( "bad buffer\n" );
+			if ( buf->filp != filp || buf->pending ) {
+				DRM_ERROR( "bad buffer %p %p %d\n",
+					   buf->filp, filp, buf->pending);
 				return DRM_ERR(EINVAL);
 			}