merged r200-0-2-branch to trunk

7 files changed, 785 insertions, 20 deletions

diff --git a/shared/radeon.h b/shared/radeon.h
index 885e4297..4cb00986 100644
--- a/shared/radeon.h
+++ b/shared/radeon.h
@@ -48,10 +48,10 @@
 
 #define DRIVER_NAME		"radeon"
 #define DRIVER_DESC		"ATI Radeon"
-#define DRIVER_DATE		"20020611"
+#define DRIVER_DATE		"20020828"
 
 #define DRIVER_MAJOR		1
-#define DRIVER_MINOR		5
+#define DRIVER_MINOR		6
 #define DRIVER_PATCHLEVEL	0
 
 /* Interface history:
@@ -68,6 +68,11 @@
  * 1.5 - Add r200 packets to cmdbuf ioctl
  *     - Add r200 function to init ioctl
  *     - Add 'scalar2' instruction to cmdbuf
+ * 1.6 - Add static agp memory manager
+ *       Add irq handler (won't be turned on unless X server knows to)
+ *       Add irq ioctls and irq_active getparam.
+ *       Add wait command for cmdbuf ioctl
+ *       Add agp offset query for getparam
  */
 #define DRIVER_IOCTLS							     \
  [DRM_IOCTL_NR(DRM_IOCTL_DMA)]               = { radeon_cp_buffers,  1, 0 }, \
@@ -88,9 +93,18 @@
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_VERTEX2)]    = { radeon_cp_vertex2,  1, 0 }, \
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_CMDBUF)]     = { radeon_cp_cmdbuf,   1, 0 }, \
  [DRM_IOCTL_NR(DRM_IOCTL_RADEON_GETPARAM)]   = { radeon_cp_getparam, 1, 0 }, \
- [DRM_IOCTL_NR(DRM_IOCTL_RADEON_FLIP)]       = { radeon_cp_flip,     1, 0 }, 
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_FLIP)]       = { radeon_cp_flip,     1, 0 }, \
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_ALLOC)]      = { radeon_mem_alloc,   1, 0 }, \
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_FREE)]       = { radeon_mem_free,    1, 0 }, \
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_INIT_HEAP)]  = { radeon_mem_init_heap, 1, 1 }, \
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_IRQ_EMIT)]   = { radeon_irq_emit, 1, 0 }, \
+ [DRM_IOCTL_NR(DRM_IOCTL_RADEON_IRQ_WAIT)]   = { radeon_irq_wait, 1, 0 },
 
-/* Driver customization:
+/* When a client dies:
+ *    - Check for and clean up flipped page state
+ *    - Free any alloced agp memory.
+ *
+ * DRM infrastructure takes care of reclaiming dma buffers.
  */
 #define DRIVER_PRERELEASE() do {					\
 	if ( dev->dev_private ) {					\
@@ -98,22 +112,78 @@
 		if ( dev_priv->page_flipping ) {			\
 			radeon_do_cleanup_pageflip( dev );		\
 		}							\
+                radeon_mem_release( dev_priv->agp_heap );		\
 	}								\
 } while (0)
 
+/* On unloading the module:
+ *    - Free memory heap structure
+ *    - Remove mappings made at startup and free dev_private.
+ */
 #define DRIVER_PRETAKEDOWN() do {					\
-	if ( dev->dev_private ) radeon_do_cleanup_cp( dev );		\
+	if ( dev->dev_private ) {					\
+		drm_radeon_private_t *dev_priv = dev->dev_private;	\
+		radeon_mem_takedown( &(dev_priv->agp_heap) );		\
+		radeon_do_cleanup_cp( dev );				\
+	}								\
 } while (0)
 
 /* DMA customization:
  */
 #define __HAVE_DMA		1
 
+
+#define __HAVE_DMA_IRQ		1
+#define __HAVE_DMA_IRQ_BH	1 
+#define __HAVE_SHARED_IRQ       1
+
+#define DRIVER_PREINSTALL() do {				\
+	drm_radeon_private_t *dev_priv =			\
+		(drm_radeon_private_t *)dev->dev_private;	\
+	u32 tmp;						\
+								\
+	/* Clear bit if it's already high: */			\
+   	tmp = RADEON_READ( RADEON_GEN_INT_STATUS );		\
+   	tmp = tmp & RADEON_SW_INT_TEST_ACK;			\
+   	RADEON_WRITE( RADEON_GEN_INT_STATUS, tmp );		\
+								\
+ 	/* Disable *all* interrupts */				\
+      	RADEON_WRITE( RADEON_GEN_INT_CNTL, 0 );			\
+} while (0)
+
+#ifdef __linux__ 
+#define IWH(x) init_waitqueue_head(x) 
+#else 
+#define IWH(x) 
+#endif 
+
+#define DRIVER_POSTINSTALL() do {				\
+	drm_radeon_private_t *dev_priv =			\
+		(drm_radeon_private_t *)dev->dev_private;	\
+								\
+   	atomic_set(&dev_priv->irq_received, 0);			\
+   	atomic_set(&dev_priv->irq_emitted, 0);			\
+	IWH(&dev_priv->irq_queue);		\
+								\
+	/* Turn on SW_INT only */				\
+   	RADEON_WRITE( RADEON_GEN_INT_CNTL,			\
+		      RADEON_SW_INT_ENABLE );			\
+} while (0)
+
+#define DRIVER_UNINSTALL() do {					\
+	drm_radeon_private_t *dev_priv =			\
+		(drm_radeon_private_t *)dev->dev_private;	\
+	if ( dev_priv ) {					\
+		/* Disable *all* interrupts */			\
+		RADEON_WRITE( RADEON_GEN_INT_CNTL, 0 );		\
+	}							\
+} while (0)
+
 /* Buffer customization:
  */
 #define DRIVER_BUF_PRIV_T	drm_radeon_buf_priv_t
 
-#define DRIVER_AGP_BUFFERS_MAP( dev )					\
+#define DRIVER_AGP_BUFFERS_MAP( dev )				\
 	((drm_radeon_private_t *)((dev)->dev_private))->buffers
 
 #endif
diff --git a/shared/radeon_cp.c b/shared/radeon_cp.c
index 86ea01bc..c0fb63ef 100644
--- a/shared/radeon_cp.c
+++ b/shared/radeon_cp.c
@@ -1412,6 +1412,9 @@ int radeon_cp_idle( DRM_IOCTL_ARGS )
 
 	LOCK_TEST_WITH_RETURN( dev );
 
+/* 	if (dev->irq)  */
+/* 		radeon_emit_and_wait_irq( dev ); */
+
 	return radeon_do_cp_idle( dev_priv );
 }
 
diff --git a/shared/radeon_drm.h b/shared/radeon_drm.h
index 6469bfb8..9a747e04 100644
--- a/shared/radeon_drm.h
+++ b/shared/radeon_drm.h
@@ -142,6 +142,9 @@
 #define RADEON_CMD_PACKET3     5 /* emit hw packet */
 #define RADEON_CMD_PACKET3_CLIP 6 /* emit hw packet wrapped in cliprects */
 #define RADEON_CMD_SCALARS2     7 /* r200 stopgap */
+#define RADEON_CMD_WAIT         8 /* emit hw wait commands -- note:
+				   *  doesn't make the cpu wait, just
+				   *  the graphics hardware */
 
 
 typedef union {
@@ -161,8 +164,14 @@ typedef union {
 	struct { 
 		unsigned char cmd_type, buf_idx, pad0, pad1; 
 	} dma;
+	struct { 
+		unsigned char cmd_type, flags, pad0, pad1; 
+	} wait;
 } drm_radeon_cmd_header_t;
 
+#define RADEON_WAIT_2D  0x1
+#define RADEON_WAIT_3D  0x2
+
 
 #define RADEON_FRONT			0x1
 #define RADEON_BACK			0x2
@@ -364,6 +373,11 @@ typedef struct {
 #define DRM_IOCTL_RADEON_CMDBUF     DRM_IOW( 0x50, drm_radeon_cmd_buffer_t)
 #define DRM_IOCTL_RADEON_GETPARAM   DRM_IOWR(0x51, drm_radeon_getparam_t)
 #define DRM_IOCTL_RADEON_FLIP	    DRM_IO(  0x52)
+#define DRM_IOCTL_RADEON_ALLOC      DRM_IOWR( 0x53, drm_radeon_mem_alloc_t)
+#define DRM_IOCTL_RADEON_FREE       DRM_IOW( 0x54, drm_radeon_mem_free_t)
+#define DRM_IOCTL_RADEON_INIT_HEAP  DRM_IOW( 0x55, drm_radeon_mem_init_heap_t)
+#define DRM_IOCTL_RADEON_IRQ_EMIT   DRM_IOWR( 0x56, drm_radeon_irq_emit_t)
+#define DRM_IOCTL_RADEON_IRQ_WAIT   DRM_IOW( 0x57, drm_radeon_irq_wait_t)
 
 typedef struct drm_radeon_init {
 	enum {
@@ -499,14 +513,51 @@ typedef struct drm_radeon_indirect {
 /* 1.3: An ioctl to get parameters that aren't available to the 3d
  * client any other way.  
  */
-#define RADEON_PARAM_AGP_BUFFER_OFFSET 0x1
-#define RADEON_PARAM_LAST_FRAME 0x2
-#define RADEON_PARAM_LAST_DISPATCH 0x3
-#define RADEON_PARAM_LAST_CLEAR 0x4
+#define RADEON_PARAM_AGP_BUFFER_OFFSET     1 /* card offset of 1st agp buffer */
+#define RADEON_PARAM_LAST_FRAME            2
+#define RADEON_PARAM_LAST_DISPATCH         3
+#define RADEON_PARAM_LAST_CLEAR            4
+#define RADEON_PARAM_IRQ_ACTIVE            5
+#define RADEON_PARAM_AGP_BASE              6 /* card offset of agp base */
 
 typedef struct drm_radeon_getparam {
 	int param;
 	int *value;
 } drm_radeon_getparam_t;
 
+/* 1.6: Set up a memory manager for regions of shared memory:
+ */
+#define RADEON_MEM_REGION_AGP 1
+#define RADEON_MEM_REGION_FB  2
+
+typedef struct drm_radeon_mem_alloc {
+	int region;
+	int alignment;
+	int size;
+	int *region_offset;	/* offset from start of fb or agp */
+} drm_radeon_mem_alloc_t;
+
+typedef struct drm_radeon_mem_free {
+	int region;
+	int region_offset;
+} drm_radeon_mem_free_t;
+
+typedef struct drm_radeon_mem_init_heap {
+	int region;
+	int size;
+	int start;	
+} drm_radeon_mem_init_heap_t;
+
+
+/* 1.6: Userspace can request & wait on irq's:
+ */
+typedef struct drm_radeon_irq_emit {
+	int *irq_seq;
+} drm_radeon_irq_emit_t;
+
+typedef struct drm_radeon_irq_wait {
+	int irq_seq;
+} drm_radeon_irq_wait_t;
+
+
 #endif
diff --git a/shared/radeon_drv.h b/shared/radeon_drv.h
index e012bbbd..efe3020c 100644
--- a/shared/radeon_drv.h
+++ b/shared/radeon_drv.h
@@ -61,6 +61,15 @@ typedef struct drm_radeon_depth_clear_t {
 	u32 se_cntl;
 } drm_radeon_depth_clear_t;
 
+
+struct mem_block {
+	struct mem_block *next;
+	struct mem_block *prev;
+	int start;
+	int size;
+	int pid;		/* 0: free, -1: heap, other: real pids */
+};
+
 typedef struct drm_radeon_private {
 	drm_radeon_ring_buffer_t ring;
 	drm_radeon_sarea_t *sarea_priv;
@@ -126,6 +135,14 @@ typedef struct drm_radeon_private {
 	drm_map_t *ring_rptr;
 	drm_map_t *buffers;
 	drm_map_t *agp_textures;
+
+	struct mem_block *agp_heap;
+	struct mem_block *fb_heap;
+
+   	wait_queue_head_t irq_queue;
+   	atomic_t irq_received;
+   	atomic_t irq_emitted;
+
 } drm_radeon_private_t;
 
 typedef struct drm_radeon_buf_priv {
@@ -164,6 +181,20 @@ extern int radeon_cp_cmdbuf( DRM_IOCTL_ARGS );
 extern int radeon_cp_getparam( DRM_IOCTL_ARGS );
 extern int radeon_cp_flip( DRM_IOCTL_ARGS );
 
+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 int radeon_irq_emit( DRM_IOCTL_ARGS );
+extern int radeon_irq_wait( DRM_IOCTL_ARGS );
+
+extern int radeon_emit_and_wait_irq(drm_device_t *dev);
+extern int radeon_wait_irq(drm_device_t *dev, int irq_nr);
+extern int radeon_emit_irq(drm_device_t *dev);
+
+
 /* Flags for stats.boxes
  */
 #define RADEON_BOX_DMA_IDLE      0x1
@@ -238,6 +269,16 @@ extern int radeon_cp_flip( DRM_IOCTL_ARGS );
 				? DRM_READ32( &dev_priv->scratch[(x)] )		\
 				: RADEON_READ( RADEON_SCRATCH_REG0 + 4*(x) ) )
 
+
+#define RADEON_GEN_INT_CNTL		0x0040
+#	define RADEON_GUI_IDLE_INT_ENABLE	(1 << 19)
+#	define RADEON_SW_INT_ENABLE		(1 << 25)
+
+#define RADEON_GEN_INT_STATUS		0x0044
+#	define RADEON_GUI_IDLE_INT_TEST_ACK     (1 << 19)
+#	define RADEON_SW_INT_TEST_ACK   	(1 << 25)
+#	define RADEON_SW_INT_FIRE		(1 << 26)
+
 #define RADEON_HOST_PATH_CNTL		0x0130
 #	define RADEON_HDP_SOFT_RESET		(1 << 26)
 #	define RADEON_HDP_WC_TIMEOUT_MASK	(7 << 28)
@@ -526,6 +567,8 @@ extern int radeon_cp_flip( DRM_IOCTL_ARGS );
 #define RADEON_TXFORMAT_ARGB4444	5
 #define RADEON_TXFORMAT_ARGB8888	6
 #define RADEON_TXFORMAT_RGBA8888	7
+#define RADEON_TXFORMAT_VYUY422         10
+#define RADEON_TXFORMAT_YVYU422         11
 
 #define R200_PP_TXCBLEND_0                0x2f00
 #define R200_PP_TXCBLEND_1                0x2f10
diff --git a/shared/radeon_irq.c b/shared/radeon_irq.c
new file mode 100644
index 00000000..11609393
--- /dev/null
+++ b/shared/radeon_irq.c
@@ -0,0 +1,214 @@
+/* radeon_mem.c -- Simple agp/fb memory manager for radeon -*- linux-c -*-
+ *
+ * Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+ * 
+ * The Weather Channel (TM) funded Tungsten Graphics to develop the
+ * initial release of the Radeon 8500 driver under the XFree86 license.
+ * This notice must be preserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "radeon.h"
+#include "drmP.h"
+#include "drm.h"
+#include "radeon_drm.h"
+#include "radeon_drv.h"
+
+/* Interrupts - Used for device synchronization and flushing in the
+ * following circumstances:
+ *
+ * - Exclusive FB access with hw idle:
+ *    - Wait for GUI Idle (?) interrupt, then do normal flush.
+ *
+ * - Frame throttling, NV_fence:
+ *    - Drop marker irq's into command stream ahead of time.
+ *    - Wait on irq's with lock *not held*
+ *    - Check each for termination condition
+ *
+ * - Internally in cp_getbuffer, etc:
+ *    - as above, but wait with lock held???
+ *
+ * NOTE: These functions are misleadingly named -- the irq's aren't
+ * tied to dma at all, this is just a hangover from dri prehistory.
+ */
+
+void DRM(dma_service)( DRM_IRQ_ARGS )
+{
+	drm_device_t *dev = (drm_device_t *) arg;
+	drm_radeon_private_t *dev_priv = 
+	   (drm_radeon_private_t *)dev->dev_private;
+   	u32 temp;
+
+	/* Need to wait for fifo to drain?
+	 */
+	temp = RADEON_READ(RADEON_GEN_INT_STATUS);  
+	temp = temp & RADEON_SW_INT_TEST_ACK;  
+	if (temp == 0) return;  
+	RADEON_WRITE(RADEON_GEN_INT_STATUS, temp);  
+
+	atomic_inc(&dev_priv->irq_received);
+#ifdef __linux__
+	queue_task(&dev->tq, &tq_immediate);  
+	mark_bh(IMMEDIATE_BH);  
+#endif /* __linux__ */
+#ifdef __FreeBSD__
+	taskqueue_enqueue(taskqueue_swi, &dev->task);
+#endif /* __FreeBSD__ */
+}
+
+void DRM(dma_immediate_bh)( DRM_TASKQUEUE_ARGS )
+{
+	drm_device_t *dev = (drm_device_t *) arg;
+	drm_radeon_private_t *dev_priv = 
+	   (drm_radeon_private_t *)dev->dev_private;
+
+#ifdef __linux__
+	wake_up_interruptible(&dev_priv->irq_queue); 
+#endif /* __linux__ */
+#ifdef __FreeBSD__
+	wakeup( &dev_priv->irq_queue );
+#endif
+}
+
+
+int radeon_emit_irq(drm_device_t *dev)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	RING_LOCALS;
+
+	atomic_inc(&dev_priv->irq_emitted);
+
+	BEGIN_RING(2); 
+	OUT_RING( CP_PACKET0( RADEON_GEN_INT_STATUS, 0 ) );
+	OUT_RING( RADEON_SW_INT_FIRE );
+	ADVANCE_RING(); 
+ 	COMMIT_RING();
+
+	return atomic_read(&dev_priv->irq_emitted);
+}
+
+
+int radeon_wait_irq(drm_device_t *dev, int irq_nr)
+{
+  	drm_radeon_private_t *dev_priv = 
+	   (drm_radeon_private_t *)dev->dev_private;
+#ifdef __linux__
+	DECLARE_WAITQUEUE(entry, current);
+	unsigned long end = jiffies + HZ*3;
+#endif /* __linux__ */
+	int ret = 0;
+
+ 	if (atomic_read(&dev_priv->irq_received) >= irq_nr)  
+ 		return 0; 
+
+	dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
+
+#ifdef __linux__
+	add_wait_queue(&dev_priv->irq_queue, &entry);
+
+	for (;;) {
+		current->state = TASK_INTERRUPTIBLE;
+	   	if (atomic_read(&dev_priv->irq_received) >= irq_nr) 
+		   break;
+		if((signed)(end - jiffies) <= 0) {
+		   	ret = -EBUSY;	/* Lockup?  Missed irq? */
+			break;
+		}
+	      	schedule_timeout(HZ*3);
+	      	if (signal_pending(current)) {
+		   	ret = -EINTR;
+			break;
+		}
+	}
+
+	current->state = TASK_RUNNING;
+	remove_wait_queue(&dev_priv->irq_queue, &entry);
+	return ret;
+#endif /* __linux__ */
+	
+#ifdef __FreeBSD__
+	ret = tsleep( &dev_priv->irq_queue, PZERO | PCATCH, \
+		"rdnirq", 3*hz );
+	if ( (ret == EWOULDBLOCK) || (ret == EINTR) )
+		return DRM_ERR(EBUSY);
+	return ret;
+#endif /* __FreeBSD__ */
+}
+
+
+int radeon_emit_and_wait_irq(drm_device_t *dev)
+{
+	return radeon_wait_irq( dev, radeon_emit_irq(dev) );
+}
+
+
+/* Needs the lock as it touches the ring.
+ */
+int radeon_irq_emit( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	drm_radeon_irq_emit_t emit;
+	int result;
+
+	LOCK_TEST_WITH_RETURN( dev );
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( emit, (drm_radeon_irq_emit_t *)data,
+				  sizeof(emit) );
+
+	result = radeon_emit_irq( dev );
+
+	if ( DRM_COPY_TO_USER( emit.irq_seq, &result, sizeof(int) ) ) {
+		DRM_ERROR( "copy_to_user\n" );
+		return DRM_ERR(EFAULT);
+	}
+
+	return 0;
+}
+
+
+/* Doesn't need the hardware lock.
+ */
+int radeon_irq_wait( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	drm_radeon_irq_wait_t irqwait;
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( irqwait, (drm_radeon_irq_wait_t *)data,
+				  sizeof(irqwait) );
+
+	return radeon_wait_irq( dev, irqwait.irq_seq );
+}
+
diff --git a/shared/radeon_mem.c b/shared/radeon_mem.c
new file mode 100644
index 00000000..d77c60b9
--- /dev/null
+++ b/shared/radeon_mem.c
@@ -0,0 +1,334 @@
+/* radeon_mem.c -- Simple agp/fb memory manager for radeon -*- linux-c -*-
+ *
+ * Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+ * 
+ * The Weather Channel (TM) funded Tungsten Graphics to develop the
+ * initial release of the Radeon 8500 driver under the XFree86 license.
+ * This notice must be preserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "radeon.h"
+#include "drmP.h"
+#include "drm.h"
+#include "radeon_drm.h"
+#include "radeon_drv.h"
+
+/* Very simple allocator for agp memory, working on a static range
+ * already mapped into each client's address space.  
+ */
+
+static struct mem_block *split_block(struct mem_block *p, int start, int size,
+				     int pid )
+{
+	/* Maybe cut off the start of an existing block */
+	if (start > p->start) {
+		struct mem_block *newblock = DRM_MALLOC(sizeof(*newblock));
+		if (!newblock) 
+			goto out;
+		newblock->start = start;
+		newblock->size = p->size - (start - p->start);
+		newblock->pid = 0;
+		newblock->next = p->next;
+		newblock->prev = p;
+		p->next->prev = newblock;
+		p->next = newblock;
+		p->size -= newblock->size;
+		p = newblock;
+	}
+   
+	/* Maybe cut off the end of an existing block */
+	if (size < p->size) {
+		struct mem_block *newblock = DRM_MALLOC(sizeof(*newblock));
+		if (!newblock)
+			goto out;
+		newblock->start = start + size;
+		newblock->size = p->size - size;
+		newblock->pid = 0;
+		newblock->next = p->next;
+		newblock->prev = p;
+		p->next->prev = newblock;
+		p->next = newblock;
+		p->size = size;
+	}
+
+ out:
+	/* Our block is in the middle */
+	p->pid = pid;
+	return p;
+}
+
+static struct mem_block *alloc_block( struct mem_block *heap, int size, 
+				      int align2, int pid )
+{
+	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 );
+	}
+
+	return NULL;
+}
+
+static struct mem_block *find_block( struct mem_block *heap, int start )
+{
+	struct mem_block *p;
+
+	for (p = heap->next ; p != heap ; p = p->next) 
+		if (p->start == start)
+			return p;
+
+	return NULL;
+}
+
+
+static void free_block( struct mem_block *p )
+{
+	p->pid = 0;
+
+	/* Assumes a single contiguous range.  Needs a special pid in
+	 * 'heap' to stop it being subsumed.
+	 */
+	if (p->next->pid == 0) {
+		struct mem_block *q = p->next;
+		p->size += q->size;
+		p->next = q->next;
+		p->next->prev = p;
+		DRM_FREE(p);
+	}
+
+	if (p->prev->pid == 0) {
+		struct mem_block *q = p->prev;
+		q->size += p->size;
+		q->next = p->next;
+		q->next->prev = q;
+		DRM_FREE(p);
+	}
+}
+
+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)
+{
+	struct mem_block *blocks = DRM_MALLOC(sizeof(*blocks));
+
+	if (!blocks) 
+		return -ENOMEM;
+	
+	*heap = DRM_MALLOC(sizeof(**heap));
+	if (!*heap) {
+		DRM_FREE( blocks );
+		return -ENOMEM;
+	}
+
+	blocks->start = start;
+	blocks->size = size;
+	blocks->pid = 0;
+	blocks->next = blocks->prev = *heap;
+
+	memset( *heap, 0, sizeof(**heap) );
+	(*heap)->pid = -1;
+	(*heap)->next = (*heap)->prev = blocks;
+	return 0;
+}
+
+
+/* Free all blocks associated with the releasing pid.
+ */
+void radeon_mem_release( 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;
+	}
+
+	/* Assumes a single contiguous range.  Needs a special pid in
+	 * 'heap' to stop it being subsumed.
+	 */
+	for (p = heap->next ; p != heap ; p = p->next) {
+		while (p->pid == 0 && p->next->pid == 0) {
+			struct mem_block *q = p->next;
+			p->size += q->size;
+			p->next = q->next;
+			p->next->prev = p;
+			DRM_FREE(q);
+		}
+	}
+}
+
+/* Shutdown.
+ */
+void radeon_mem_takedown( struct mem_block **heap )
+{
+	struct mem_block *p;
+	
+	if (!*heap)
+		return;
+
+	for (p = (*heap)->next ; p != *heap ; ) {
+		struct mem_block *q = p;
+		p = p->next;
+		DRM_FREE(q);
+	}
+
+	DRM_FREE( *heap );
+	*heap = 0;
+}
+
+
+
+/* IOCTL HANDLERS */
+
+static struct mem_block **get_heap( drm_radeon_private_t *dev_priv,
+				   int region )
+{
+	switch( region ) {
+	case RADEON_MEM_REGION_AGP:
+ 		return &dev_priv->agp_heap; 
+	case RADEON_MEM_REGION_FB:
+		return &dev_priv->fb_heap;
+	default:
+		return 0;
+	}
+}
+
+int radeon_mem_alloc( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	drm_radeon_mem_alloc_t alloc;
+	struct mem_block *block, **heap;
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( alloc, (drm_radeon_mem_alloc_t *)data,
+				  sizeof(alloc) );
+
+	heap = get_heap( dev_priv, alloc.region );
+	if (!heap || !*heap)
+		return DRM_ERR(EFAULT);
+	
+	/* Make things easier on ourselves: all allocations at least
+	 * 4k aligned.
+	 */
+	if (alloc.alignment < 12)
+		alloc.alignment = 12;
+
+	block = alloc_block( *heap, alloc.size, alloc.alignment,
+			     DRM_CURRENTPID );
+
+	if (!block) 
+		return DRM_ERR(ENOMEM);
+
+	if ( DRM_COPY_TO_USER( alloc.region_offset, &block->start, 
+			       sizeof(int) ) ) {
+		DRM_ERROR( "copy_to_user\n" );
+		return DRM_ERR(EFAULT);
+	}
+	
+	return 0;
+}
+
+
+
+int radeon_mem_free( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	drm_radeon_mem_free_t memfree;
+	struct mem_block *block, **heap;
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( memfree, (drm_radeon_mem_free_t *)data,
+				  sizeof(memfree) );
+
+	heap = get_heap( dev_priv, memfree.region );
+	if (!heap || !*heap)
+		return DRM_ERR(EFAULT);
+	
+	block = find_block( *heap, memfree.region_offset );
+	if (!block)
+		return DRM_ERR(EFAULT);
+
+	if (block->pid != DRM_CURRENTPID)
+		return DRM_ERR(EPERM);
+
+	free_block( block );	
+	return 0;
+}
+
+int radeon_mem_init_heap( DRM_IOCTL_ARGS )
+{
+	DRM_DEVICE;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	drm_radeon_mem_init_heap_t initheap;
+	struct mem_block **heap;
+
+	if ( !dev_priv ) {
+		DRM_ERROR( "%s called with no initialization\n", __FUNCTION__ );
+		return DRM_ERR(EINVAL);
+	}
+
+	DRM_COPY_FROM_USER_IOCTL( initheap, (drm_radeon_mem_init_heap_t *)data,
+				  sizeof(initheap) );
+
+	heap = get_heap( dev_priv, initheap.region );
+	if (!heap) 
+		return DRM_ERR(EFAULT);
+	
+	if (*heap) {
+		DRM_ERROR("heap already initialized?");
+		return DRM_ERR(EFAULT);
+	}
+		
+	return init_heap( heap, initheap.start, initheap.size );
+}
+
+
diff --git a/shared/radeon_state.c b/shared/radeon_state.c
index 6dbf6093..0172128b 100644
--- a/shared/radeon_state.c
+++ b/shared/radeon_state.c
@@ -719,7 +719,6 @@ static void radeon_cp_dispatch_swap( drm_device_t *dev )
 	RING_LOCALS;
 	DRM_DEBUG( "\n" );
 
-
 	/* Do some trivial performance monitoring...
 	 */
 	if (dev_priv->do_boxes)
@@ -1088,6 +1087,8 @@ static int radeon_cp_dispatch_texture( drm_device_t *dev,
 	case RADEON_TXFORMAT_ARGB1555:
 	case RADEON_TXFORMAT_RGB565:
 	case RADEON_TXFORMAT_ARGB4444:
+	case RADEON_TXFORMAT_VYUY422:
+	case RADEON_TXFORMAT_YVYU422:
 		format = RADEON_COLOR_FORMAT_RGB565;
 		tex_width = tex->width * 2;
 		blit_width = image->width * 2;
@@ -1226,6 +1227,7 @@ static int radeon_cp_dispatch_texture( drm_device_t *dev,
 	return ret;
 }
 
+
 static void radeon_cp_dispatch_stipple( drm_device_t *dev, u32 *stipple )
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
@@ -1306,6 +1308,9 @@ static int radeon_do_init_pageflip( drm_device_t *dev )
 	return 0;
 }
 
+/* Called whenever a client dies, from DRM(release).
+ * NOTE:  Lock isn't necessarily held when this is called!
+ */
 int radeon_do_cleanup_pageflip( drm_device_t *dev )
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
@@ -1936,14 +1941,18 @@ static int radeon_emit_packet3_cliprect( drm_device_t *dev,
 		if ( i < cmdbuf->nbox ) {
 			if (DRM_COPY_FROM_USER_UNCHECKED( &box, &boxes[i], sizeof(box) ))
 				return DRM_ERR(EFAULT);
-			/* FIXME The second and subsequent times round this loop, send a
-			 * WAIT_UNTIL_3D_IDLE before calling emit_clip_rect(). This
-			 * fixes a lockup on fast machines when sending several
-			 * cliprects with a cmdbuf, as when waving a 2D window over
-			 * a 3D window. Something in the commands from user space
-			 * seems to hang the card when they're sent several times
-			 * in a row. That would be the correct place to fix it but
-			 * this works around it until I can figure that out - Tim Smith */
+			/* FIXME The second and subsequent times round
+			 * this loop, send a WAIT_UNTIL_3D_IDLE before
+			 * calling emit_clip_rect(). This fixes a
+			 * lockup on fast machines when sending
+			 * several cliprects with a cmdbuf, as when
+			 * waving a 2D window over a 3D
+			 * window. Something in the commands from user
+			 * space seems to hang the card when they're
+			 * sent several times in a row. That would be
+			 * the correct place to fix it but this works
+			 * around it until I can figure that out - Tim
+			 * Smith */
 			if ( i ) {
 				BEGIN_RING( 2 );
 				RADEON_WAIT_UNTIL_3D_IDLE();
@@ -1967,6 +1976,34 @@ static int radeon_emit_packet3_cliprect( drm_device_t *dev,
 }
 
 
+static int radeon_emit_wait( drm_device_t *dev, int flags )
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	RING_LOCALS;
+
+	DRM_DEBUG("%s: %x\n", __FUNCTION__, flags);
+	switch (flags) {
+	case RADEON_WAIT_2D:
+		BEGIN_RING( 2 );
+		RADEON_WAIT_UNTIL_2D_IDLE(); 
+		ADVANCE_RING();
+		break;
+	case RADEON_WAIT_3D:
+		BEGIN_RING( 2 );
+		RADEON_WAIT_UNTIL_3D_IDLE(); 
+		ADVANCE_RING();
+		break;
+	case RADEON_WAIT_2D|RADEON_WAIT_3D:
+		BEGIN_RING( 2 );
+		RADEON_WAIT_UNTIL_IDLE(); 
+		ADVANCE_RING();
+		break;
+	default:
+		return DRM_ERR(EINVAL);
+	}
+
+	return 0;
+}
 
 int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
 {
@@ -1989,7 +2026,6 @@ int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
 	DRM_COPY_FROM_USER_IOCTL( cmdbuf, (drm_radeon_cmd_buffer_t *)data,
 			     sizeof(cmdbuf) );
 
-	DRM_DEBUG( "pid=%d\n", DRM_CURRENTPID );
 	RING_SPACE_TEST_WITH_RETURN( dev_priv );
 	VB_AGE_TEST_WITH_RETURN( dev_priv );
 
@@ -2080,6 +2116,14 @@ int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
 				return DRM_ERR(EINVAL);
 			}
 			break;
+
+		case RADEON_CMD_WAIT:
+			DRM_DEBUG("RADEON_CMD_WAIT\n");
+			if (radeon_emit_wait( dev, header.wait.flags )) {
+				DRM_ERROR("radeon_emit_wait failed\n");
+				return DRM_ERR(EINVAL);
+			}
+			break;
 		default:
 			DRM_ERROR("bad cmd_type %d at %p\n", 
 				  header.header.cmd_type,
@@ -2128,6 +2172,12 @@ int radeon_cp_getparam( DRM_IOCTL_ARGS )
 		dev_priv->stats.last_clear_reads++;
 		value = GET_SCRATCH( 2 );
 		break;
+	case RADEON_PARAM_IRQ_ACTIVE:
+		value = dev->irq ? 1 : 0;
+		break;
+	case RADEON_PARAM_AGP_BASE:
+		value = dev_priv->agp_vm_start;
+		break;
 	default:
 		return DRM_ERR(EINVAL);
 	}