1 files changed, 540 insertions, 172 deletions

diff --git a/linux-core/i810_dma.c b/linux-core/i810_dma.c
index 6ac4d2ee..f041f209 100644
--- a/linux-core/i810_dma.c
+++ b/linux-core/i810_dma.c
@@ -37,8 +37,12 @@
 
 #include <linux/interrupt.h>	/* For task queue support */
 
-#define I810_BUF_FREE		1
-#define I810_BUF_USED		0
+#define I810_BUF_FREE		2
+#define I810_BUF_CLIENT		1
+#define I810_BUF_HARDWARE      	0
+
+#define I810_BUF_UNMAPPED 0
+#define I810_BUF_MAPPED   1
 
 #define I810_REG(reg)		2
 #define I810_BASE(reg)		((unsigned long) \
@@ -91,7 +95,7 @@ static inline void i810_print_status_page(drm_device_t *dev)
       	DRM_DEBUG(  "hw_status: Reserved : %x\n", temp[3]);
    	DRM_DEBUG(  "hw_status: Driver Counter : %d\n", temp[5]);
    	for(i = 6; i < dma->buf_count + 6; i++) {
-	   	DRM_DEBUG(  "buffer status idx : %d used: %d\n", i - 6, temp[i]);
+	   	DRM_DEBUG( "buffer status idx : %d used: %d\n", i - 6, temp[i]);
 	}
 }
 
@@ -108,7 +112,7 @@ static drm_buf_t *i810_freelist_get(drm_device_t *dev)
 	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
 		/* In use is already a pointer */
 	   	used = cmpxchg(buf_priv->in_use, I810_BUF_FREE, 
-			       I810_BUF_USED);
+			       I810_BUF_CLIENT);
 	   	if(used == I810_BUF_FREE) {
 			return buf;
 		}
@@ -126,8 +130,8 @@ static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
    	int used;
    
    	/* In use is already a pointer */
-   	used = cmpxchg(buf_priv->in_use, I810_BUF_USED, I810_BUF_FREE);
-   	if(used != I810_BUF_USED) {
+   	used = cmpxchg(buf_priv->in_use, I810_BUF_CLIENT, I810_BUF_FREE);
+   	if(used != I810_BUF_CLIENT) {
 	   	DRM_ERROR("Freeing buffer thats not in use : %d\n", buf->idx);
 	   	return -EINVAL;
 	}
@@ -135,26 +139,114 @@ static int i810_freelist_put(drm_device_t *dev, drm_buf_t *buf)
    	return 0;
 }
 
-static int i810_dma_get_buffers(drm_device_t *dev, drm_dma_t *d)
+static struct file_operations i810_buffer_fops = {
+	open:	 i810_open,
+	flush:	 drm_flush,
+	release: i810_release,
+	ioctl:	 i810_ioctl,
+	mmap:	 i810_mmap_buffers,
+	read:	 drm_read,
+	fasync:	 drm_fasync,
+      	poll:	 drm_poll,
+};
+
+int i810_mmap_buffers(struct file *filp, struct vm_area_struct *vma)
+{
+	drm_file_t	  *priv	  = filp->private_data;
+	drm_device_t	  *dev	  = priv->dev;
+	drm_i810_private_t *dev_priv = dev->dev_private;
+	drm_buf_t *buf = dev_priv->mmap_buffer;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+   
+	vma->vm_flags |= (VM_IO | VM_DONTCOPY);
+	vma->vm_file = filp;
+   
+   	buf_priv->currently_mapped = I810_BUF_MAPPED;
+
+	if (remap_page_range(vma->vm_start,
+			     VM_OFFSET(vma),
+			     vma->vm_end - vma->vm_start,
+			     vma->vm_page_prot)) return -EAGAIN;
+	return 0;
+}
+
+static int i810_map_buffer(drm_buf_t *buf, struct file *filp)
 {
-	int		  i;
+	drm_file_t	  *priv	  = filp->private_data;
+	drm_device_t	  *dev	  = priv->dev;
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+      	drm_i810_private_t *dev_priv = dev->dev_private;
+   	struct file_operations *old_fops;
+	int retcode = 0;
+
+	if(buf_priv->currently_mapped == I810_BUF_MAPPED) return -EINVAL;
+	down(&current->mm->mmap_sem);
+   	old_fops = filp->f_op;
+	filp->f_op = &i810_buffer_fops;
+	dev_priv->mmap_buffer = buf;
+	buf_priv->virtual = (void *)do_mmap(filp, 0, buf->total, 
+					    PROT_READ|PROT_WRITE,
+					    MAP_SHARED, 
+					    buf->bus_address);
+	dev_priv->mmap_buffer = NULL;
+   	filp->f_op = old_fops;
+	if ((unsigned long)buf_priv->virtual > -1024UL) {
+		/* Real error */
+		DRM_DEBUG("mmap error\n");
+		retcode = (signed int)buf_priv->virtual;
+		buf_priv->virtual = 0;
+	}
+   	up(&current->mm->mmap_sem);
+	return retcode;
+}
+
+static int i810_unmap_buffer(drm_buf_t *buf)
+{
+	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	int retcode = 0;
+
+	if(buf_priv->currently_mapped != I810_BUF_MAPPED) return -EINVAL;
+	down(&current->mm->mmap_sem);
+        retcode = do_munmap((unsigned long)buf_priv->virtual, 
+			    (size_t) buf->total);
+   	buf_priv->currently_mapped = I810_BUF_UNMAPPED;
+   	buf_priv->virtual = 0;
+   	up(&current->mm->mmap_sem);
+
+	return retcode;
+}
+
+static int i810_dma_get_buffer(drm_device_t *dev, drm_i810_dma_t *d, 
+			       struct file *filp)
+{
+	drm_file_t	  *priv	  = filp->private_data;
 	drm_buf_t	  *buf;
+	drm_i810_buf_priv_t *buf_priv;
+	int retcode = 0;
 
-	for (i = d->granted_count; i < d->request_count; i++) {
-		buf = i810_freelist_get(dev);
-		if (!buf) break;
-		buf->pid     = current->pid;
-		copy_to_user_ret(&d->request_indices[i],
-				 &buf->idx,
-				 sizeof(buf->idx),
-				 -EFAULT);
-		copy_to_user_ret(&d->request_sizes[i],
-				 &buf->total,
-				 sizeof(buf->total),
-				 -EFAULT);
-		++d->granted_count;
+	buf = i810_freelist_get(dev);
+	if (!buf) {
+		retcode = -ENOMEM;
+	   	DRM_DEBUG("%s retcode %d\n", __FUNCTION__, retcode);
+		goto out_get_buf;
 	}
-	return 0;
+   
+	retcode = i810_map_buffer(buf, filp);
+	if(retcode) {
+		i810_freelist_put(dev, buf);
+	   	DRM_DEBUG("mapbuf failed in %s retcode %d\n", 
+			  __FUNCTION__, retcode);
+	   	goto out_get_buf;
+	}
+	buf->pid     = priv->pid;
+	buf_priv = buf->dev_private;	
+	d->granted = 1;
+   	d->request_idx = buf->idx;
+   	d->request_size = buf->total;
+   	d->virtual = buf_priv->virtual;
+
+out_get_buf:
+	return retcode;
 }
 
 static unsigned long i810_alloc_page(drm_device_t *dev)
@@ -185,7 +277,10 @@ static void i810_free_page(drm_device_t *dev, unsigned long page)
 
 static int i810_dma_cleanup(drm_device_t *dev)
 {
+	drm_device_dma_t *dma = dev->dma;
+
 	if(dev->dev_private) {
+		int i;
 	   	drm_i810_private_t *dev_priv = 
 	     		(drm_i810_private_t *) dev->dev_private;
 	   
@@ -201,6 +296,12 @@ static int i810_dma_cleanup(drm_device_t *dev)
 	   	drm_free(dev->dev_private, sizeof(drm_i810_private_t), 
 			 DRM_MEM_DRIVER);
 	   	dev->dev_private = NULL;
+
+		for (i = 0; i < dma->buf_count; i++) {
+			drm_buf_t *buf = dma->buflist[ i ];
+			drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+			drm_ioremapfree(buf_priv->kernel_virtual, buf->total);
+		}
 	}
    	return 0;
 }
@@ -211,6 +312,7 @@ static int i810_wait_ring(drm_device_t *dev, int n)
    	drm_i810_ring_buffer_t *ring = &(dev_priv->ring);
    	int iters = 0;
    	unsigned long end;
+	unsigned int last_head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
 
 	end = jiffies + (HZ*3);
    	while (ring->space < n) {
@@ -218,9 +320,11 @@ static int i810_wait_ring(drm_device_t *dev, int n)
 	
 	   	ring->head = I810_READ(LP_RING + RING_HEAD) & HEAD_ADDR;
 	   	ring->space = ring->head - (ring->tail+8);
+		if (ring->space < 0) ring->space += ring->Size;
 	   
-	   	if (ring->space < 0) ring->space += ring->Size;
-	   
+		if (ring->head != last_head)
+		   end = jiffies + (HZ*3);
+	  
 	   	iters++;
 		if((signed)(end - jiffies) <= 0) {
 		   	DRM_ERROR("space: %d wanted %d\n", ring->space, n);
@@ -250,9 +354,9 @@ static int i810_freelist_init(drm_device_t *dev)
 {
       	drm_device_dma_t *dma = dev->dma;
    	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
-   	u8 *hw_status = (u8 *)dev_priv->hw_status_page;
-   	int i;
    	int my_idx = 24;
+   	u32 *hw_status = (u32 *)(dev_priv->hw_status_page + my_idx);
+   	int i;
    
    	if(dma->buf_count > 1019) {
 	   	/* Not enough space in the status page for the freelist */
@@ -263,11 +367,14 @@ static int i810_freelist_init(drm_device_t *dev)
 	   	drm_buf_t *buf = dma->buflist[ i ];
 	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
 	   
-	   	buf_priv->in_use = hw_status + my_idx;
-	   	DRM_DEBUG("buf_priv->in_use : %p\n", buf_priv->in_use);
-	   	*buf_priv->in_use = I810_BUF_FREE;
+	   	buf_priv->in_use = hw_status++;
 	   	buf_priv->my_use_idx = my_idx;
 	   	my_idx += 4;
+
+	   	*buf_priv->in_use = I810_BUF_FREE;
+
+		buf_priv->kernel_virtual = drm_ioremap(buf->bus_address, 
+						       buf->total);
 	}
 	return 0;
 }
@@ -301,9 +408,11 @@ static int i810_dma_initialize(drm_device_t *dev,
    	dev_priv->ring.Start = init->ring_start;
    	dev_priv->ring.End = init->ring_end;
    	dev_priv->ring.Size = init->ring_size;
+
    	dev_priv->ring.virtual_start = drm_ioremap(dev->agp->base + 
 						   init->ring_start, 
 						   init->ring_size);
+
    	dev_priv->ring.tail_mask = dev_priv->ring.Size - 1;
    
    	if (dev_priv->ring.virtual_start == NULL) {
@@ -312,6 +421,17 @@ static int i810_dma_initialize(drm_device_t *dev,
 			  " ring buffer\n");
 	   	return -ENOMEM;
 	}
+
+	dev_priv->w = init->w;
+	dev_priv->h = init->h;
+	dev_priv->pitch = init->pitch;
+	dev_priv->back_offset = init->back_offset;
+	dev_priv->depth_offset = init->depth_offset;
+
+	dev_priv->front_di1 = init->front_offset | init->pitch_bits;
+	dev_priv->back_di1 = init->back_offset | init->pitch_bits;
+	dev_priv->zi1 = init->depth_offset | init->pitch_bits;
+	
    
    	/* Program Hardware Status Page */
    	dev_priv->hw_status_page = i810_alloc_page(dev);
@@ -366,37 +486,270 @@ int i810_dma_init(struct inode *inode, struct file *filp,
    	return retcode;
 }
 
-static void i810_dma_dispatch_general(drm_device_t *dev, drm_buf_t *buf,
-				      int used )
+
+
+/* Most efficient way to verify state for the i810 is as it is
+ * emitted.  Non-conformant state is silently dropped.
+ *
+ * Use 'volatile' & local var tmp to force the emitted values to be
+ * identical to the verified ones.
+ */
+static void i810EmitContextVerified( drm_device_t *dev, 
+				     volatile unsigned int *code ) 
+{	
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	int i, j = 0;
+	unsigned int tmp;
+	RING_LOCALS;
+
+	BEGIN_LP_RING( I810_CTX_SETUP_SIZE );
+
+	OUT_RING( GFX_OP_COLOR_FACTOR );
+	OUT_RING( code[I810_CTXREG_CF1] );
+
+	OUT_RING( GFX_OP_STIPPLE );
+	OUT_RING( code[I810_CTXREG_ST1] );
+
+	for ( i = 4 ; i < I810_CTX_SETUP_SIZE ; i++ ) {
+		tmp = code[i];
+
+		if ((tmp & (7<<29)) == (3<<29) &&
+		    (tmp & (0x1f<<24)) < (0x1d<<24)) 
+		{
+			OUT_RING( tmp ); 
+			j++;
+		} 
+	}
+
+	if (j & 1) 
+		OUT_RING( 0 ); 
+
+	ADVANCE_LP_RING();
+}
+
+static void i810EmitTexVerified( drm_device_t *dev, 
+				 volatile unsigned int *code ) 
+{	
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	int i, j = 0;
+	unsigned int tmp;
+	RING_LOCALS;
+
+	BEGIN_LP_RING( I810_TEX_SETUP_SIZE );
+
+	OUT_RING( GFX_OP_MAP_INFO );
+	OUT_RING( code[I810_TEXREG_MI1] );
+	OUT_RING( code[I810_TEXREG_MI2] );
+	OUT_RING( code[I810_TEXREG_MI3] );
+
+	for ( i = 4 ; i < I810_TEX_SETUP_SIZE ; i++ ) {
+		tmp = code[i];
+
+		if ((tmp & (7<<29)) == (3<<29) &&
+		    (tmp & (0x1f<<24)) < (0x1d<<24)) 
+		{
+			OUT_RING( tmp ); 
+			j++;
+		}
+	} 
+		
+	if (j & 1) 
+		OUT_RING( 0 ); 
+
+	ADVANCE_LP_RING();
+}
+
+
+/* Need to do some additional checking when setting the dest buffer.
+ */
+static void i810EmitDestVerified( drm_device_t *dev, 
+				  volatile unsigned int *code ) 
+{	
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+	unsigned int tmp;
+	RING_LOCALS;
+
+	BEGIN_LP_RING( I810_DEST_SETUP_SIZE + 2 );
+
+	tmp = code[I810_DESTREG_DI1];
+	if (tmp == dev_priv->front_di1 || tmp == dev_priv->back_di1) {
+		OUT_RING( CMD_OP_DESTBUFFER_INFO );
+		OUT_RING( tmp );
+	} else
+	   DRM_DEBUG("bad di1 %x (allow %x or %x)\n",
+		     tmp, dev_priv->front_di1, dev_priv->back_di1);
+
+	/* invarient:
+	 */
+	OUT_RING( CMD_OP_Z_BUFFER_INFO );
+	OUT_RING( dev_priv->zi1 );
+
+	OUT_RING( GFX_OP_DESTBUFFER_VARS );
+	OUT_RING( code[I810_DESTREG_DV1] );
+
+	OUT_RING( GFX_OP_DRAWRECT_INFO );
+	OUT_RING( code[I810_DESTREG_DR1] );
+	OUT_RING( code[I810_DESTREG_DR2] );
+	OUT_RING( code[I810_DESTREG_DR3] );
+	OUT_RING( code[I810_DESTREG_DR4] );
+	OUT_RING( 0 );
+
+	ADVANCE_LP_RING();
+}
+
+
+
+static void i810EmitState( drm_device_t *dev )
 {
-      	drm_i810_private_t *dev_priv = dev->dev_private;
-   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
-	unsigned long address = (unsigned long)buf->bus_address;
-	unsigned long start = address - dev->agp->base;
-   	RING_LOCALS;
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+      	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	unsigned int dirty = sarea_priv->dirty;
 
-   	dev_priv->counter++;
-   	DRM_DEBUG(  "dispatch counter : %ld\n", dev_priv->counter);
-   	DRM_DEBUG(  "i810_dma_dispatch\n");
-   	DRM_DEBUG(  "start : 0x%lx\n", start);
-	DRM_DEBUG(  "used : 0x%x\n", used);
-   	DRM_DEBUG(  "start + used - 4 : 0x%lx\n", start + used - 4);
-   	i810_kernel_lost_context(dev);
+	if (dirty & I810_UPLOAD_BUFFERS) {
+		i810EmitDestVerified( dev, sarea_priv->BufferState );
+		sarea_priv->dirty &= ~I810_UPLOAD_BUFFERS;
+	}
 
-   	BEGIN_LP_RING(10);
-   	OUT_RING( CMD_OP_BATCH_BUFFER );
-   	OUT_RING( start | BB1_PROTECTED );
-   	OUT_RING( start + used - 4 );
-      	OUT_RING( CMD_STORE_DWORD_IDX );
-   	OUT_RING( 20 );
-   	OUT_RING( dev_priv->counter );
-   	OUT_RING( CMD_STORE_DWORD_IDX );
-   	OUT_RING( buf_priv->my_use_idx );
-   	OUT_RING( I810_BUF_FREE );   
-      	OUT_RING( CMD_REPORT_HEAD );
-      	ADVANCE_LP_RING();
+	if (dirty & I810_UPLOAD_CTX) {
+		i810EmitContextVerified( dev, sarea_priv->ContextState );
+		sarea_priv->dirty &= ~I810_UPLOAD_CTX;
+	}
+
+	if (dirty & I810_UPLOAD_TEX0) {
+		i810EmitTexVerified( dev, sarea_priv->TexState[0] );
+		sarea_priv->dirty &= ~I810_UPLOAD_TEX0;
+	}
+
+	if (dirty & I810_UPLOAD_TEX1) {
+		i810EmitTexVerified( dev, sarea_priv->TexState[1] );
+		sarea_priv->dirty &= ~I810_UPLOAD_TEX1;
+	}
+}
+
+
+
+/* need to verify 
+ */
+static void i810_dma_dispatch_clear( drm_device_t *dev, int flags, 
+				     unsigned int clear_color,
+				     unsigned int clear_zval )
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+      	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	int pitch = dev_priv->pitch;
+	int cpp = 2;
+	int i;
+	RING_LOCALS;
+
+  	i810_kernel_lost_context(dev);
+
+      	if (nbox > I810_NR_SAREA_CLIPRECTS)
+     		nbox = I810_NR_SAREA_CLIPRECTS;
+
+	for (i = 0 ; i < nbox ; i++, pbox++) {
+		unsigned int x = pbox->x1;
+		unsigned int y = pbox->y1;
+		unsigned int width = (pbox->x2 - x) * cpp;
+		unsigned int height = pbox->y2 - y;
+		unsigned int start = y * pitch + x * cpp;
+
+		if (pbox->x1 > pbox->x2 ||
+		    pbox->y1 > pbox->y2 ||
+		    pbox->x2 > dev_priv->w ||
+		    pbox->y2 > dev_priv->h)
+			continue;
+
+	   	if ( flags & I810_FRONT ) {	    
+		   	DRM_DEBUG("clear front\n");
+			BEGIN_LP_RING( 6 );	    
+			OUT_RING( BR00_BITBLT_CLIENT | 
+				  BR00_OP_COLOR_BLT | 0x3 );
+			OUT_RING( BR13_SOLID_PATTERN | (0xF0 << 16) | pitch );
+			OUT_RING( (height << 16) | width );
+			OUT_RING( start );
+			OUT_RING( clear_color );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+		}
+
+		if ( flags & I810_BACK ) {
+			DRM_DEBUG("clear back\n");
+			BEGIN_LP_RING( 6 );	    
+			OUT_RING( BR00_BITBLT_CLIENT | 
+				  BR00_OP_COLOR_BLT | 0x3 );
+			OUT_RING( BR13_SOLID_PATTERN | (0xF0 << 16) | pitch );
+			OUT_RING( (height << 16) | width );
+			OUT_RING( dev_priv->back_offset + start );
+			OUT_RING( clear_color );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+		}
+
+		if ( flags & I810_DEPTH ) {
+			DRM_DEBUG("clear depth\n");
+			BEGIN_LP_RING( 6 );	    
+			OUT_RING( BR00_BITBLT_CLIENT | 
+				  BR00_OP_COLOR_BLT | 0x3 );
+			OUT_RING( BR13_SOLID_PATTERN | (0xF0 << 16) | pitch );
+			OUT_RING( (height << 16) | width );
+			OUT_RING( dev_priv->depth_offset + start );
+			OUT_RING( clear_zval );
+			OUT_RING( 0 );
+			ADVANCE_LP_RING();
+		}
+	}
+}
+
+static void i810_dma_dispatch_swap( drm_device_t *dev )
+{
+   	drm_i810_private_t *dev_priv = dev->dev_private;
+      	drm_i810_sarea_t *sarea_priv = dev_priv->sarea_priv;
+	int nbox = sarea_priv->nbox;
+	drm_clip_rect_t *pbox = sarea_priv->boxes;
+	int pitch = dev_priv->pitch;
+	int cpp = 2;
+	int ofs = dev_priv->back_offset;
+	int i;
+	RING_LOCALS;
+
+	DRM_DEBUG("swapbuffers\n");
+
+  	i810_kernel_lost_context(dev);
+
+      	if (nbox > I810_NR_SAREA_CLIPRECTS)
+     		nbox = I810_NR_SAREA_CLIPRECTS;
+
+	for (i = 0 ; i < nbox; i++, pbox++) 
+	{
+		unsigned int w = pbox->x2 - pbox->x1;
+		unsigned int h = pbox->y2 - pbox->y1;
+		unsigned int dst = pbox->x1*cpp + pbox->y1*pitch;
+		unsigned int start = ofs + dst;
+
+		if (pbox->x1 > pbox->x2 ||
+		    pbox->y1 > pbox->y2 ||
+		    pbox->x2 > dev_priv->w ||
+		    pbox->y2 > dev_priv->h)
+			continue;
+ 
+	   	DRM_DEBUG("dispatch swap %d,%d-%d,%d!\n",
+			  pbox[i].x1, pbox[i].y1,
+			  pbox[i].x2, pbox[i].y2);
+
+		BEGIN_LP_RING( 6 );
+		OUT_RING( BR00_BITBLT_CLIENT | BR00_OP_SRC_COPY_BLT | 0x4 );
+		OUT_RING( pitch | (0xCC << 16));
+		OUT_RING( (h << 16) | (w * cpp));
+		OUT_RING( dst );
+		OUT_RING( pitch );	
+		OUT_RING( start );
+		ADVANCE_LP_RING();
+	}
 }
 
+
 static void i810_dma_dispatch_vertex(drm_device_t *dev, 
 				     drm_buf_t *buf,
 				     int discard,
@@ -409,14 +762,29 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
    	int nbox = sarea_priv->nbox;
 	unsigned long address = (unsigned long)buf->bus_address;
 	unsigned long start = address - dev->agp->base;     
-	int i = 0;
+	int i = 0, u;
    	RING_LOCALS;
 
-   
+   	i810_kernel_lost_context(dev);
+
    	if (nbox > I810_NR_SAREA_CLIPRECTS) 
 		nbox = I810_NR_SAREA_CLIPRECTS;
-   
-   	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
+
+	if (discard) {
+		u = cmpxchg(buf_priv->in_use, I810_BUF_CLIENT, 
+			    I810_BUF_HARDWARE);
+		if(u != I810_BUF_CLIENT) {
+			DRM_DEBUG("xxxx 2\n");
+		}
+	}
+
+	if (used > 4*1024) 
+		used = 0;
+
+	if (sarea_priv->dirty)
+	   i810EmitState( dev );
+
+  	DRM_DEBUG("dispatch vertex addr 0x%lx, used 0x%x nbox %d\n", 
 		  address, used, nbox);
 
    	dev_priv->counter++;
@@ -425,8 +793,20 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
    	DRM_DEBUG(  "start : %lx\n", start);
 	DRM_DEBUG(  "used : %d\n", used);
    	DRM_DEBUG(  "start + used - 4 : %ld\n", start + used - 4);
-   	i810_kernel_lost_context(dev);
 
+	if (buf_priv->currently_mapped == I810_BUF_MAPPED) {
+		*(u32 *)buf_priv->virtual = (GFX_OP_PRIMITIVE |
+					     sarea_priv->vertex_prim |
+					     ((used/4)-2));
+		
+		if (used & 4) {
+			*(u32 *)((u32)buf_priv->virtual + used) = 0;
+			used += 4;
+		}
+
+		i810_unmap_buffer(buf);
+	}
+		   
 	if (used) {
 		do {
 			if (i < nbox) {
@@ -434,7 +814,7 @@ static void i810_dma_dispatch_vertex(drm_device_t *dev,
 				OUT_RING( GFX_OP_SCISSOR | SC_UPDATE_SCISSOR | 
 					  SC_ENABLE );
 				OUT_RING( GFX_OP_SCISSOR_INFO );
-				OUT_RING( box[i].x1 | (box[i].y1 << 16) );
+				OUT_RING( box[i].x1 | (box[i].y1<<16) );
 				OUT_RING( (box[i].x2-1) | ((box[i].y2-1)<<16) );
 				ADVANCE_LP_RING();
 			}
@@ -479,7 +859,9 @@ static void i810_dma_service(int irq, void *device, struct pt_regs *regs)
    	temp = temp & ~(0x6000);
    	if(temp != 0) I810_WRITE16(I810REG_INT_IDENTITY_R, 
 				   temp); /* Clear all interrupts */
-   
+	else
+	   return;
+ 
    	queue_task(&dev->tq, &tq_immediate);
    	mark_bh(IMMEDIATE_BH);
 }
@@ -539,7 +921,7 @@ int i810_irq_install(drm_device_t *dev, int irq)
 				/* Install handler */
 	if ((retcode = request_irq(dev->irq,
 				   i810_dma_service,
-				   0,
+				   SA_SHIRQ,
 				   dev->devname,
 				   dev))) {
 		down(&dev->struct_sem);
@@ -560,6 +942,9 @@ int i810_irq_uninstall(drm_device_t *dev)
 	int irq;
    	u16 temp;
 
+
+/*  	return 0; */
+
 	down(&dev->struct_sem);
 	irq	 = dev->irq;
 	dev->irq = 0;
@@ -618,10 +1003,15 @@ static inline void i810_dma_emit_flush(drm_device_t *dev)
    	RING_LOCALS;
 
    	i810_kernel_lost_context(dev);
+
    	BEGIN_LP_RING(2);
       	OUT_RING( CMD_REPORT_HEAD );
-   	OUT_RING( GFX_OP_USER_INTERRUPT );
+      	OUT_RING( GFX_OP_USER_INTERRUPT );
       	ADVANCE_LP_RING();
+
+/*  	i810_wait_ring( dev, dev_priv->ring.Size - 8 ); */
+/*     	atomic_set(&dev_priv->flush_done, 1); */
+/*     	wake_up_interruptible(&dev_priv->flush_queue); */
 }
 
 static inline void i810_dma_quiescent_emit(drm_device_t *dev)
@@ -630,13 +1020,17 @@ static inline void i810_dma_quiescent_emit(drm_device_t *dev)
    	RING_LOCALS;
 
   	i810_kernel_lost_context(dev);
-   	BEGIN_LP_RING(4);
 
+   	BEGIN_LP_RING(4);
    	OUT_RING( INST_PARSER_CLIENT | INST_OP_FLUSH | INST_FLUSH_MAP_CACHE );
    	OUT_RING( CMD_REPORT_HEAD );
-   	OUT_RING( GFX_OP_USER_INTERRUPT );
-   	OUT_RING( 0 );
+      	OUT_RING( 0 );
+      	OUT_RING( GFX_OP_USER_INTERRUPT );
    	ADVANCE_LP_RING();
+
+/*  	i810_wait_ring( dev, dev_priv->ring.Size - 8 ); */
+/*     	atomic_set(&dev_priv->flush_done, 1); */
+/*     	wake_up_interruptible(&dev_priv->flush_queue); */
 }
 
 static void i810_dma_quiescent(drm_device_t *dev)
@@ -676,8 +1070,9 @@ static int i810_flush_queue(drm_device_t *dev)
 {
    	DECLARE_WAITQUEUE(entry, current);
   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
 	unsigned long end;
-   	int ret = 0;      
+   	int i, ret = 0;      
 
    	if(dev_priv == NULL) {
 	   	return 0;
@@ -702,7 +1097,21 @@ static int i810_flush_queue(drm_device_t *dev)
    
    	current->state = TASK_RUNNING;
    	remove_wait_queue(&dev_priv->flush_queue, &entry);
-   
+
+
+   	for (i = 0; i < dma->buf_count; i++) {
+	   	drm_buf_t *buf = dma->buflist[ i ];
+	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
+	   
+		int used = cmpxchg(buf_priv->in_use, I810_BUF_HARDWARE, 
+				   I810_BUF_FREE);
+
+		if (used == I810_BUF_HARDWARE)
+			DRM_DEBUG("reclaimed from HARDWARE\n");
+		if (used == I810_BUF_CLIENT)
+			DRM_DEBUG("still on client HARDWARE\n");
+	}
+
    	return ret;
 }
 
@@ -713,20 +1122,23 @@ void i810_reclaim_buffers(drm_device_t *dev, pid_t pid)
 	int		 i;
 
 	if (!dma) return;
-      	if(dev->dev_private == NULL) return;
-	if(dma->buflist == NULL) return;
+      	if (!dev->dev_private) return;
+	if (!dma->buflist) return;
+
         i810_flush_queue(dev);
 
 	for (i = 0; i < dma->buf_count; i++) {
 	   	drm_buf_t *buf = dma->buflist[ i ];
 	   	drm_i810_buf_priv_t *buf_priv = buf->dev_private;
 	   
-		/* Only buffers that need to get reclaimed ever 
-		 * get set to free 
-		 */
 		if (buf->pid == pid && buf_priv) {
-		   	cmpxchg(buf_priv->in_use, 
-				I810_BUF_USED, I810_BUF_FREE);
+			int used = cmpxchg(buf_priv->in_use, I810_BUF_CLIENT, 
+					   I810_BUF_FREE);
+
+			if (used == I810_BUF_CLIENT)
+				DRM_DEBUG("reclaimed from client\n");
+		   	if(buf_priv->currently_mapped == I810_BUF_MAPPED)
+		     		buf_priv->currently_mapped = I810_BUF_UNMAPPED;
 		}
 	}
 }
@@ -760,19 +1172,6 @@ int i810_lock(struct inode *inode, struct file *filp, unsigned int cmd,
 	 */
 
 	if (!ret) {
-#if 0
-	   	if (_DRM_LOCKING_CONTEXT(dev->lock.hw_lock->lock)
-		    != lock.context) {
-			long j = jiffies - dev->lock.lock_time;
-
-			if (j > 0 && j <= DRM_LOCK_SLICE) {
-				/* Can't take lock if we just had it and
-				   there is contention. */
-				current->state = TASK_INTERRUPTIBLE;
-				schedule_timeout(j);
-			}
-		}
-#endif
 		add_wait_queue(&dev->lock.lock_queue, &entry);
 		for (;;) {
 			if (!dev->lock.hw_lock) {
@@ -829,90 +1228,80 @@ int i810_flush_ioctl(struct inode *inode, struct file *filp,
    	return 0;
 }
 
-static int i810DmaGeneral(drm_device_t *dev, drm_i810_general_t *args)
-{
-	drm_device_dma_t *dma = dev->dma;
-   	drm_buf_t *buf = dma->buflist[ args->idx ];
-   	
-	if (!args->used) {
-	   	i810_freelist_put(dev, buf);
-	} else {  
-		i810_dma_dispatch_general( dev, buf, args->used );
-		atomic_add(args->used, &dma->total_bytes);
-		atomic_inc(&dma->total_dmas);
-	}
-	return 0; 
-}
-
-static int i810DmaVertex(drm_device_t *dev, drm_i810_vertex_t *args)
-{
-	drm_device_dma_t *dma = dev->dma;
-	drm_buf_t *buf = dma->buflist[ args->idx ];
-   	i810_dma_dispatch_vertex( dev, buf, args->discard, args->used );
-   	atomic_add(args->used, &dma->total_bytes);
-	atomic_inc(&dma->total_dmas);
-   	return 0;
-}
 
-int i810_dma_general(struct inode *inode, struct file *filp,
-		     unsigned int cmd, unsigned long arg)
+int i810_dma_vertex(struct inode *inode, struct file *filp,
+	       unsigned int cmd, unsigned long arg)
 {
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->dev;
-	drm_i810_general_t general;
-      	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
+	drm_device_dma_t *dma = dev->dma;
+   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
       	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
    	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
      					dev_priv->sarea_priv; 
+	drm_i810_vertex_t vertex;
 
-	int retcode = 0;
-	
-	copy_from_user_ret(&general, (drm_i810_general_t *)arg, sizeof(general),
+	copy_from_user_ret(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
 			   -EFAULT);
 
-	DRM_DEBUG("i810 dma general idx %d used %d\n",
-		  general.idx, general.used);
-   
    	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-		DRM_ERROR("i810_dma_general called without lock held\n");
+		DRM_ERROR("i810_dma_vertex called without lock held\n");
 		return -EINVAL;
 	}
 
-	retcode = i810DmaGeneral(dev, &general);
+	DRM_DEBUG("i810 dma vertex, idx %d used %d discard %d\n",
+		  vertex.idx, vertex.used, vertex.discard);
+
+	i810_dma_dispatch_vertex( dev, 
+				  dma->buflist[ vertex.idx ], 
+				  vertex.discard, vertex.used );
+
+   	atomic_add(vertex.used, &dma->total_bytes);
+	atomic_inc(&dma->total_dmas);
 	sarea_priv->last_enqueue = dev_priv->counter-1;
    	sarea_priv->last_dispatch = (int) hw_status[5];
    
-	return retcode;
+	return 0;
 }
 
-int i810_dma_vertex(struct inode *inode, struct file *filp,
-	       unsigned int cmd, unsigned long arg)
+
+
+int i810_clear_bufs(struct inode *inode, struct file *filp,
+		   unsigned int cmd, unsigned long arg)
 {
 	drm_file_t *priv = filp->private_data;
 	drm_device_t *dev = priv->dev;
-   	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
-      	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
-   	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
-     					dev_priv->sarea_priv; 
-	drm_i810_vertex_t vertex;
-	int retcode = 0;
+	drm_i810_clear_t clear;
 
-	copy_from_user_ret(&vertex, (drm_i810_vertex_t *)arg, sizeof(vertex),
+   	copy_from_user_ret(&clear, (drm_i810_clear_t *)arg, sizeof(clear), 
 			   -EFAULT);
+   
    	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
-		DRM_ERROR("i810_dma_vertex called without lock held\n");
+		DRM_ERROR("i810_clear_bufs called without lock held\n");
 		return -EINVAL;
 	}
 
-	DRM_DEBUG("i810 dma vertex, idx %d used %d discard %d\n",
-		  vertex.idx, vertex.used, vertex.discard);
+	i810_dma_dispatch_clear( dev, clear.flags, 
+				 clear.clear_color, 
+				 clear.clear_depth );
+   	return 0;
+}
 
-	retcode = i810DmaVertex(dev, &vertex);
-	sarea_priv->last_enqueue = dev_priv->counter-1;
-   	sarea_priv->last_dispatch = (int) hw_status[5];
+int i810_swap_bufs(struct inode *inode, struct file *filp,
+		  unsigned int cmd, unsigned long arg)
+{
+	drm_file_t *priv = filp->private_data;
+	drm_device_t *dev = priv->dev;
    
-	return retcode;
+	DRM_DEBUG("i810_swap_bufs\n");
 
+   	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
+		DRM_ERROR("i810_swap_buf called without lock held\n");
+		return -EINVAL;
+	}
+
+	i810_dma_dispatch_swap( dev );
+   	return 0;
 }
 
 int i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
@@ -929,53 +1318,32 @@ int i810_getage(struct inode *inode, struct file *filp, unsigned int cmd,
 	return 0;
 }
 
-int i810_dma(struct inode *inode, struct file *filp, unsigned int cmd,
-	    unsigned long arg)
+int i810_getbuf(struct inode *inode, struct file *filp, unsigned int cmd,
+		unsigned long arg)
 {
 	drm_file_t	  *priv	    = filp->private_data;
 	drm_device_t	  *dev	    = priv->dev;
-	drm_device_dma_t  *dma	    = dev->dma;
 	int		  retcode   = 0;
-	drm_dma_t	  d;
+	drm_i810_dma_t	  d;
    	drm_i810_private_t *dev_priv = (drm_i810_private_t *)dev->dev_private;
    	u32 *hw_status = (u32 *)dev_priv->hw_status_page;
    	drm_i810_sarea_t *sarea_priv = (drm_i810_sarea_t *) 
      					dev_priv->sarea_priv; 
 
-
-   	copy_from_user_ret(&d, (drm_dma_t *)arg, sizeof(d), -EFAULT);
-	DRM_DEBUG("%d %d: %d send, %d req\n",
-		  current->pid, d.context, d.send_count, d.request_count);
+	DRM_DEBUG("getbuf\n");
+   	copy_from_user_ret(&d, (drm_i810_dma_t *)arg, sizeof(d), -EFAULT);
    
 	if(!_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock)) {
 		DRM_ERROR("i810_dma called without lock held\n");
 		return -EINVAL;
 	}
-
-	/* Please don't send us buffers.
-	 */
-	if (d.send_count != 0) {
-		DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
-			  current->pid, d.send_count);
-		return -EINVAL;
-	}
 	
-	/* We'll send you buffers.
-	 */
-	if (d.request_count < 0 || d.request_count > dma->buf_count) {
-		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
-			  current->pid, d.request_count, dma->buf_count);
-		return -EINVAL;
-	}
-	
-	d.granted_count = 0;
+	d.granted = 0;
 
-	if (!retcode && d.request_count) {
-		retcode = i810_dma_get_buffers(dev, &d);
-	}
+	retcode = i810_dma_get_buffer(dev, &d, filp);
 
-	DRM_DEBUG("i810_dma: %d returning, granted = %d\n",
-		  current->pid, d.granted_count);
+	DRM_DEBUG("i810_dma: %d returning %d, granted = %d\n",
+		  current->pid, retcode, d.granted);
 
 	copy_to_user_ret((drm_dma_t *)arg, &d, sizeof(d), -EFAULT);   
    	sarea_priv->last_dispatch = (int) hw_status[5];