radeon command submission start

take code from Jerome munge into a TTM IB re-use

9 files changed, 450 insertions, 51 deletions

diff --git a/linux-core/Makefile.kernel b/linux-core/Makefile.kernel
index d3668dfb..404944bf 100644
--- a/linux-core/Makefile.kernel
+++ b/linux-core/Makefile.kernel
@@ -41,7 +41,7 @@ nouveau-objs := nouveau_drv.o nouveau_state.o nouveau_fifo.o nouveau_mem.o \
 		nv50_kms_wrapper.o \
 		nv50_fbcon.o
 radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o r300_cmdbuf.o radeon_gem.o \
-	 radeon_buffer.o radeon_fence.o atom.o radeon_display.o radeon_atombios.o radeon_i2c.o radeon_connectors.o \
+	 radeon_buffer.o radeon_fence.o atom.o radeon_display.o radeon_atombios.o radeon_i2c.o radeon_connectors.o radeon_cs.o \
 	atombios_crtc.o radeon_encoders.o radeon_fb.o radeon_combios.o
 sis-objs    := sis_drv.o sis_mm.o
 ffb-objs    := ffb_drv.o ffb_context.o
diff --git a/linux-core/radeon_cs.c b/linux-core/radeon_cs.c
new file mode 120000
index 00000000..0e3fa8ee
--- /dev/null
+++ b/linux-core/radeon_cs.c
@@ -0,0 +1 @@
+../shared-core/radeon_cs.c
\ No newline at end of file
diff --git a/linux-core/radeon_gem.c b/linux-core/radeon_gem.c
index d536aed2..021bb9f9 100644
--- a/linux-core/radeon_gem.c
+++ b/linux-core/radeon_gem.c
@@ -27,6 +27,9 @@
 #include "radeon_drm.h"
 #include "radeon_drv.h"
 
+static int radeon_gem_ib_init(struct drm_device *dev);
+static int radeon_gem_ib_destroy(struct drm_device *dev);
+
 int radeon_gem_init_object(struct drm_gem_object *obj)
 {
 	struct drm_radeon_gem_object *obj_priv;
@@ -513,19 +516,19 @@ static int radeon_gart_init(struct drm_device *dev)
 		ret = drm_buffer_object_create(dev, RADEON_PCIGART_TABLE_SIZE,
 					       drm_bo_type_kernel,
 					       DRM_BO_FLAG_READ | DRM_BO_FLAG_MEM_VRAM | DRM_BO_FLAG_MAPPABLE | DRM_BO_FLAG_NO_EVICT,
-					       0, 1, 0, &dev_priv->mm.pcie_table);
+					       0, 1, 0, &dev_priv->mm.pcie_table.bo);
 		if (ret)
 			return -EINVAL;
 
-		DRM_DEBUG("pcie table bo created %p, %x\n", dev_priv->mm.pcie_table, dev_priv->mm.pcie_table->offset);
-		ret = drm_bo_kmap(dev_priv->mm.pcie_table, 0, RADEON_PCIGART_TABLE_SIZE >> PAGE_SHIFT,
-				  &dev_priv->mm.pcie_table_map);
+		DRM_DEBUG("pcie table bo created %p, %x\n", dev_priv->mm.pcie_table.bo, dev_priv->mm.pcie_table.bo->offset);
+		ret = drm_bo_kmap(dev_priv->mm.pcie_table.bo, 0, RADEON_PCIGART_TABLE_SIZE >> PAGE_SHIFT,
+				  &dev_priv->mm.pcie_table.kmap);
 		if (ret)
 			return -EINVAL;
 
 		dev_priv->pcigart_offset_set = 2;
-		dev_priv->gart_info.bus_addr =  dev_priv->fb_location + dev_priv->mm.pcie_table->offset;
-		dev_priv->gart_info.addr = dev_priv->mm.pcie_table_map.virtual;
+		dev_priv->gart_info.bus_addr =  dev_priv->fb_location + dev_priv->mm.pcie_table.bo->offset;
+		dev_priv->gart_info.addr = dev_priv->mm.pcie_table.kmap.virtual;
 		dev_priv->gart_info.gart_reg_if = DRM_ATI_GART_PCIE;
 		dev_priv->gart_info.gart_table_location = DRM_ATI_GART_FB;
 		memset(dev_priv->gart_info.addr, 0, RADEON_PCIGART_TABLE_SIZE);
@@ -566,14 +569,14 @@ int radeon_alloc_gart_objects(struct drm_device *dev)
 				       drm_bo_type_kernel,
 				       DRM_BO_FLAG_READ | DRM_BO_FLAG_MEM_TT |
 				       DRM_BO_FLAG_MAPPABLE | DRM_BO_FLAG_NO_EVICT,
-				       0, 1, 0, &dev_priv->mm.ring);
+				       0, 1, 0, &dev_priv->mm.ring.bo);
 	if (ret) {
 		DRM_ERROR("failed to allocate ring\n");
 		return -EINVAL;
 	}
 
-	ret = drm_bo_kmap(dev_priv->mm.ring, 0, RADEON_DEFAULT_RING_SIZE >> PAGE_SHIFT,
-			  &dev_priv->mm.ring_map);
+	ret = drm_bo_kmap(dev_priv->mm.ring.bo, 0, RADEON_DEFAULT_RING_SIZE >> PAGE_SHIFT,
+			  &dev_priv->mm.ring.kmap);
 	if (ret) {
 		DRM_ERROR("failed to map ring\n");
 		return -EINVAL;
@@ -583,24 +586,26 @@ int radeon_alloc_gart_objects(struct drm_device *dev)
 				       drm_bo_type_kernel,
 				       DRM_BO_FLAG_WRITE |DRM_BO_FLAG_READ | DRM_BO_FLAG_MEM_TT |
 				       DRM_BO_FLAG_MAPPABLE | DRM_BO_FLAG_NO_EVICT,
-				       0, 1, 0, &dev_priv->mm.ring_read_ptr);
+				       0, 1, 0, &dev_priv->mm.ring_read.bo);
 	if (ret) {
 		DRM_ERROR("failed to allocate ring read\n");
 		return -EINVAL;
 	}
 
-	ret = drm_bo_kmap(dev_priv->mm.ring_read_ptr, 0,
+	ret = drm_bo_kmap(dev_priv->mm.ring_read.bo, 0,
 			  PAGE_SIZE >> PAGE_SHIFT,
-			  &dev_priv->mm.ring_read_ptr_map);
+			  &dev_priv->mm.ring_read.kmap);
 	if (ret) {
 		DRM_ERROR("failed to map ring read\n");
 		return -EINVAL;
 	}
 
 	DRM_DEBUG("Ring ptr %p mapped at %d %p, read ptr %p maped at %d %p\n",
-		  dev_priv->mm.ring, dev_priv->mm.ring->offset, dev_priv->mm.ring_map.virtual,
-		  dev_priv->mm.ring_read_ptr, dev_priv->mm.ring_read_ptr->offset, dev_priv->mm.ring_read_ptr_map.virtual);
+		  dev_priv->mm.ring.bo, dev_priv->mm.ring.bo->offset, dev_priv->mm.ring.kmap.virtual,
+		  dev_priv->mm.ring_read.bo, dev_priv->mm.ring_read.bo->offset, dev_priv->mm.ring_read.kmap.virtual);
 
+	/* init the indirect buffers */
+	radeon_gem_ib_init(dev);
 	return 0;			  
 
 }
@@ -634,6 +639,8 @@ int radeon_gem_mm_init(struct drm_device *dev)
 	ret = radeon_alloc_gart_objects(dev);
 	if (ret)
 		return -EINVAL;
+
+	
 	return 0;
 }
 
@@ -641,16 +648,19 @@ void radeon_gem_mm_fini(struct drm_device *dev)
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;
 
+	radeon_gem_ib_destroy(dev);
+
 	mutex_lock(&dev->struct_mutex);
 
-	if (dev_priv->mm.ring_read_ptr) {
-		drm_bo_kunmap(&dev_priv->mm.ring_read_ptr_map);
-		drm_bo_usage_deref_locked(&dev_priv->mm.ring_read_ptr);
+	
+	if (dev_priv->mm.ring_read.bo) {
+		drm_bo_kunmap(&dev_priv->mm.ring_read.kmap);
+		drm_bo_usage_deref_locked(&dev_priv->mm.ring_read.bo);
 	}
 
-	if (dev_priv->mm.ring) {
-		drm_bo_kunmap(&dev_priv->mm.ring_map);
-		drm_bo_usage_deref_locked(&dev_priv->mm.ring);
+	if (dev_priv->mm.ring.bo) {
+		drm_bo_kunmap(&dev_priv->mm.ring.kmap);
+		drm_bo_usage_deref_locked(&dev_priv->mm.ring.bo);
 	}
 
 	if (drm_bo_clean_mm(dev, DRM_BO_MEM_TT, 1)) {
@@ -658,9 +668,9 @@ void radeon_gem_mm_fini(struct drm_device *dev)
 	}
 
 	if (dev_priv->flags & RADEON_IS_PCIE) {
-		if (dev_priv->mm.pcie_table) {
-			drm_bo_kunmap(&dev_priv->mm.pcie_table_map);
-			drm_bo_usage_deref_locked(&dev_priv->mm.pcie_table);
+		if (dev_priv->mm.pcie_table.bo) {
+			drm_bo_kunmap(&dev_priv->mm.pcie_table.kmap);
+			drm_bo_usage_deref_locked(&dev_priv->mm.pcie_table.bo);
 		}
 	}
 
@@ -685,3 +695,158 @@ int radeon_gem_object_pin(struct drm_gem_object *obj,
 	return ret;
 }
 
+#define RADEON_IB_MEMORY (1*1024*1024)
+#define RADEON_IB_SIZE (65536)
+
+#define RADEON_NUM_IB (RADEON_IB_MEMORY / RADEON_IB_SIZE)
+
+int radeon_gem_ib_get(struct drm_device *dev, void **ib, uint32_t dwords)
+{
+	int i, index = -1;
+	int ret;
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+
+	for (i = 0; i < RADEON_NUM_IB; i++) {
+		if (!(dev_priv->ib_alloc_bitmap & (1 << i))){
+			index = i;
+			break;
+		}
+	}
+
+	/* if all in use we need to wait */
+	if (index == -1) {
+		for (i = 0; i < RADEON_NUM_IB; i++) {
+			if (dev_priv->ib_alloc_bitmap & (1 << i)) {
+				mutex_lock(&dev_priv->ib_objs[i]->bo->mutex);
+				ret = drm_bo_wait(dev_priv->ib_objs[i]->bo, 0, 1, 0, 0);
+				mutex_unlock(&dev_priv->ib_objs[i]->bo->mutex);
+				if (ret)
+					continue;
+				dev_priv->ib_alloc_bitmap &= ~(1 << i);
+				index = i;
+				break;
+			}
+		}
+	}
+
+	if (index == -1) {
+		DRM_ERROR("Major case fail to allocate IB from freelist %x\n", dev_priv->ib_alloc_bitmap);
+		return -EINVAL;
+	}
+		
+
+	if (dwords > RADEON_IB_SIZE / sizeof(uint32_t))
+		return -EINVAL;
+
+	*ib = dev_priv->ib_objs[index]->kmap.virtual;
+	dev_priv->ib_alloc_bitmap |= (1 << i);
+	return 0;
+}
+
+static void radeon_gem_ib_free(struct drm_device *dev, void *ib, uint32_t dwords)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	struct drm_fence_object *fence;
+	int ret;
+	int i;
+	RING_LOCALS;
+
+	for (i = 0; i < RADEON_NUM_IB; i++) {
+
+		if (dev_priv->ib_objs[i]->kmap.virtual == ib) {
+			ret = drm_bo_do_validate(dev_priv->ib_objs[i]->bo, 0,
+				 DRM_BO_FLAG_NO_EVICT,
+				 0, 0, NULL);
+			if (ret)
+				DRM_ERROR("FAiled to validate\n");
+
+			DRM_DEBUG("validated IB %x, %d\n", dev_priv->ib_objs[i]->bo->offset, dwords);
+			BEGIN_RING(4);
+			
+			OUT_RING(CP_PACKET0(RADEON_CP_IB_BASE, 1));
+			OUT_RING(dev_priv->gart_vm_start + dev_priv->ib_objs[i]->bo->offset);
+			OUT_RING(dwords);
+			OUT_RING(CP_PACKET2());
+
+			ADVANCE_RING();
+
+			COMMIT_RING();
+
+			/* emit a fence object */
+			ret = drm_fence_buffer_objects(dev, NULL, 0, NULL, &fence);
+			if (ret) {
+				
+				drm_putback_buffer_objects(dev);
+			}
+			/* dereference the fence object */
+			if (fence)
+				drm_fence_usage_deref_unlocked(&fence);
+		}
+	}
+
+}
+
+static int radeon_gem_ib_destroy(struct drm_device *dev)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	int i;
+
+	if (dev_priv->ib_objs) {
+		for (i = 0; i < RADEON_NUM_IB; i++) {
+			if (dev_priv->ib_objs[i]) {
+				drm_bo_kunmap(&dev_priv->ib_objs[i]->kmap);
+				drm_bo_usage_deref_unlocked(&dev_priv->ib_objs[i]->bo);
+			}
+			drm_free(dev_priv->ib_objs[i], sizeof(struct radeon_mm_obj), DRM_MEM_DRIVER);
+		}
+		drm_free(dev_priv->ib_objs, RADEON_NUM_IB*sizeof(struct radeon_mm_obj *), DRM_MEM_DRIVER);
+	}
+	dev_priv->ib_objs = NULL;
+	return 0;
+}
+
+/* allocate 1MB of 64k IBs the the kernel can keep mapped */
+static int radeon_gem_ib_init(struct drm_device *dev)
+{
+
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	int i;
+	int ret;
+
+	dev_priv->ib_objs = drm_calloc(RADEON_NUM_IB, sizeof(struct radeon_mm_obj *), DRM_MEM_DRIVER);
+	if (!dev_priv->ib_objs)
+		goto free_all;
+
+	for (i = 0; i  < RADEON_NUM_IB; i++) {
+		dev_priv->ib_objs[i] = drm_calloc(1, sizeof(struct radeon_mm_obj), DRM_MEM_DRIVER);
+		if (!dev_priv->ib_objs[i])
+			goto free_all;
+
+		ret = drm_buffer_object_create(dev, RADEON_IB_SIZE,
+					       drm_bo_type_kernel,
+					       DRM_BO_FLAG_READ | DRM_BO_FLAG_MEM_TT |
+					       DRM_BO_FLAG_MAPPABLE, 0,
+					       0, 0, &dev_priv->ib_objs[i]->bo);
+		if (ret)
+			goto free_all;
+
+		ret = drm_bo_kmap(dev_priv->ib_objs[i]->bo, 0, RADEON_IB_SIZE >> PAGE_SHIFT,
+				  &dev_priv->ib_objs[i]->kmap);
+
+		if (ret)
+			goto free_all;
+	}
+
+	dev_priv->ib_alloc_bitmap = 0;
+
+	dev_priv->cs.ib_get = radeon_gem_ib_get;
+	dev_priv->cs.ib_free = radeon_gem_ib_free;
+
+	radeon_cs_init(dev);
+	return 0;
+
+free_all:
+	radeon_gem_ib_destroy(dev);
+	return -ENOMEM;
+}
+
diff --git a/linux-core/radeon_reg.h b/linux-core/radeon_reg.h
index 9b6bf1ec..a5be6a63 100644
--- a/linux-core/radeon_reg.h
+++ b/linux-core/radeon_reg.h
@@ -3086,6 +3086,10 @@
 #       define RADEON_CSQ_PRIPIO_INDBM         (3    << 28)
 #       define RADEON_CSQ_PRIBM_INDBM          (4    << 28)
 #       define RADEON_CSQ_PRIPIO_INDPIO        (15   << 28)
+
+#define R300_CP_RESYNC_ADDR                 0x778
+#define R300_CP_RESYNC_DATA                 0x77c
+
 #define RADEON_CP_CSQ_STAT                  0x07f8
 #       define RADEON_CSQ_RPTR_PRIMARY_MASK    (0xff <<  0)
 #       define RADEON_CSQ_WPTR_PRIMARY_MASK    (0xff <<  8)
diff --git a/shared-core/radeon_cp.c b/shared-core/radeon_cp.c
index 80951e91..46f4290a 100644
--- a/shared-core/radeon_cp.c
+++ b/shared-core/radeon_cp.c
@@ -658,8 +658,8 @@ static void radeon_cp_init_ring_buffer(struct drm_device * dev,
 					 ((dev_priv->gart_vm_start - 1) & 0xffff0000)
 					 | (dev_priv->fb_location >> 16));
 	
-	if (dev_priv->mm.ring) {
-		ring_start = dev_priv->mm.ring->offset +
+	if (dev_priv->mm.ring.bo) {
+		ring_start = dev_priv->mm.ring.bo->offset +
 			dev_priv->gart_vm_start;
 	} else
 #if __OS_HAS_AGP
@@ -692,9 +692,9 @@ static void radeon_cp_init_ring_buffer(struct drm_device * dev,
 	dev_priv->ring.tail = cur_read_ptr;
 
 
-	if (dev_priv->mm.ring_read_ptr) {
+	if (dev_priv->mm.ring_read.bo) {
 		RADEON_WRITE(RADEON_CP_RB_RPTR_ADDR,
-			     dev_priv->mm.ring_read_ptr->offset +
+			     dev_priv->mm.ring_read.bo->offset +
 			     dev_priv->gart_vm_start);
 	} else
 #if __OS_HAS_AGP
@@ -745,9 +745,9 @@ static void radeon_cp_init_ring_buffer(struct drm_device * dev,
 	RADEON_WRITE(RADEON_SCRATCH_ADDR, RADEON_READ(RADEON_CP_RB_RPTR_ADDR)
 		     + RADEON_SCRATCH_REG_OFFSET);
 
-	if (dev_priv->mm.ring_read_ptr)
+	if (dev_priv->mm.ring_read.bo)
 		dev_priv->scratch = ((__volatile__ u32 *)
-				     dev_priv->mm.ring_read_ptr_map.virtual +
+				     dev_priv->mm.ring_read.kmap.virtual +
 				     (RADEON_SCRATCH_REG_OFFSET / sizeof(u32)));
 	else
 		dev_priv->scratch = ((__volatile__ u32 *)
@@ -772,12 +772,18 @@ static void radeon_cp_init_ring_buffer(struct drm_device * dev,
 	radeon_do_wait_for_idle(dev_priv);
 
 	/* Sync everything up */
+	if (dev_priv->chip_family > CHIP_RV280) {
 	RADEON_WRITE(RADEON_ISYNC_CNTL,
 		     (RADEON_ISYNC_ANY2D_IDLE3D |
 		      RADEON_ISYNC_ANY3D_IDLE2D |
 		      RADEON_ISYNC_WAIT_IDLEGUI |
 		      RADEON_ISYNC_CPSCRATCH_IDLEGUI));
-
+	} else {
+	RADEON_WRITE(RADEON_ISYNC_CNTL,
+		     (RADEON_ISYNC_ANY2D_IDLE3D |
+		      RADEON_ISYNC_ANY3D_IDLE2D |
+		      RADEON_ISYNC_WAIT_IDLEGUI));
+	}
 }
 
 static void radeon_test_writeback(drm_radeon_private_t * dev_priv)
@@ -785,8 +791,8 @@ static void radeon_test_writeback(drm_radeon_private_t * dev_priv)
 	u32 tmp;
 	void *ring_read_ptr;
 
-	if (dev_priv->mm.ring_read_ptr)
-		ring_read_ptr = dev_priv->mm.ring_read_ptr_map.virtual;
+	if (dev_priv->mm.ring_read.bo)
+		ring_read_ptr = dev_priv->mm.ring_read.kmap.virtual;
 	else
 		ring_read_ptr = dev_priv->ring_rptr->handle;
 
@@ -2282,8 +2288,8 @@ int radeon_modeset_cp_init(struct drm_device *dev)
 	dev_priv->ring.size = RADEON_DEFAULT_RING_SIZE;
 	dev_priv->cp_mode = RADEON_CSQ_PRIBM_INDBM;
 
-	dev_priv->ring.start = (u32 *)(void *)(unsigned long)dev_priv->mm.ring_map.virtual;
-	dev_priv->ring.end = (u32 *)(void *)(unsigned long)dev_priv->mm.ring_map.virtual +
+	dev_priv->ring.start = (u32 *)(void *)(unsigned long)dev_priv->mm.ring.kmap.virtual;
+	dev_priv->ring.end = (u32 *)(void *)(unsigned long)dev_priv->mm.ring.kmap.virtual +
 		dev_priv->ring.size / sizeof(u32);
 	dev_priv->ring.size_l2qw = drm_order(dev_priv->ring.size / 8);
 	dev_priv->ring.rptr_update = 4096;
@@ -2297,7 +2303,7 @@ int radeon_modeset_cp_init(struct drm_device *dev)
 
 	radeon_cp_load_microcode(dev_priv);
 	
-	DRM_DEBUG("ring offset is %x %x\n", dev_priv->mm.ring->offset, dev_priv->mm.ring_read_ptr->offset);
+	DRM_DEBUG("ring offset is %x %x\n", dev_priv->mm.ring.bo->offset, dev_priv->mm.ring_read.bo->offset);
 
 	radeon_cp_init_ring_buffer(dev, dev_priv);
 
diff --git a/shared-core/radeon_cs.c b/shared-core/radeon_cs.c
new file mode 100644
index 00000000..61fef79d
--- /dev/null
+++ b/shared-core/radeon_cs.c
@@ -0,0 +1,181 @@
+/*
+ * Copyright 2008 Jerome Glisse.
+ * All Rights Reserved.
+ *
+ * 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:
+ *    Jerome Glisse <glisse@freedesktop.org>
+ */
+#include "drmP.h"
+#include "radeon_drm.h"
+#include "radeon_drv.h"
+#include "r300_reg.h"
+
+int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *fpriv)
+{
+	struct drm_radeon_private *radeon = dev->dev_private;
+	struct drm_radeon_cs *cs = data;
+	uint32_t *packets = NULL;
+	uint32_t cs_id;
+	void *ib = NULL;
+	long size;
+	int r;
+
+	/* set command stream id to 0 which is fake id */
+	cs_id = 0;
+	DRM_COPY_TO_USER(&cs->cs_id, &cs_id, sizeof(uint32_t));
+
+	if (radeon == NULL) {
+		DRM_ERROR("called with no initialization\n");
+		return -EINVAL;
+	}
+	if (!cs->dwords) {
+		return 0;
+	}
+	/* limit cs to 64K ib */
+	if (cs->dwords > (16 * 1024)) {
+		return -EINVAL;
+	}
+	/* copy cs from userspace maybe we should copy into ib to save
+	 * one copy but ib will be mapped wc so not good for cmd checking
+	 * somethings worth testing i guess (Jerome)
+	 */
+	size = cs->dwords * sizeof(uint32_t);
+	packets = drm_alloc(size, DRM_MEM_DRIVER);
+	if (packets == NULL) {
+		return -ENOMEM;
+	}
+	if (DRM_COPY_FROM_USER(packets, (void __user *)(unsigned long)cs->packets, size)) {
+		r = -EFAULT;
+		goto out;
+	}
+	/* get ib */
+	r = radeon->cs.ib_get(dev, &ib, cs->dwords);
+	if (r) {
+		goto out;
+	}
+
+	/* now parse command stream */
+	r = radeon->cs.parse(dev, ib, packets, cs->dwords);
+	if (r) {
+		goto out;
+	}
+
+	/* emit cs id sequence */
+	radeon->cs.id_emit(dev, &cs_id);
+	DRM_COPY_TO_USER(&cs->cs_id, &cs_id, sizeof(uint32_t));
+out:
+	radeon->cs.ib_free(dev, ib, cs->dwords);
+	drm_free(packets, size, DRM_MEM_DRIVER);
+	return r;
+}
+
+int radeon_cs_parse(struct drm_device *dev, void *ib,
+		    uint32_t *packets, uint32_t dwords)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	volatile int rb;
+
+	/* copy the packet into the IB */
+	memcpy(ib, packets, dwords * sizeof(uint32_t));
+
+	/* read back last byte to flush WC buffers */
+	rb = readl((ib + (dwords-1) * sizeof(uint32_t)));
+
+	return 0;
+}
+
+uint32_t radeon_cs_id_get(struct drm_radeon_private *radeon)
+{
+	/* FIXME: protect with a spinlock */
+	/* FIXME: check if wrap affect last reported wrap & sequence */
+	radeon->cs.id_scnt = (radeon->cs.id_scnt + 1) & 0x00FFFFFF;
+	if (!radeon->cs.id_scnt) {
+		/* increment wrap counter */
+		radeon->cs.id_wcnt += 0x01000000;
+		/* valid sequence counter start at 1 */
+		radeon->cs.id_scnt = 1;
+	}
+	return (radeon->cs.id_scnt | radeon->cs.id_wcnt);
+}
+
+void r100_cs_id_emit(struct drm_device *dev, uint32_t *id)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	RING_LOCALS;
+
+	/* ISYNC_CNTL should have CPSCRACTH bit set */
+	*id = radeon_cs_id_get(dev_priv);
+	/* emit id in SCRATCH4 (not used yet in old drm) */
+	BEGIN_RING(2);
+	OUT_RING(CP_PACKET0(RADEON_SCRATCH_REG4, 0));
+	OUT_RING(*id);
+	ADVANCE_RING();	
+}
+
+void r300_cs_id_emit(struct drm_device *dev, uint32_t *id)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+	RING_LOCALS;
+
+	/* ISYNC_CNTL should not have CPSCRACTH bit set */
+	*id = radeon_cs_id_get(dev_priv);
+	/* emit id in SCRATCH6 */
+	BEGIN_RING(6);
+	OUT_RING(CP_PACKET0(R300_CP_RESYNC_ADDR, 0));
+	OUT_RING(6);
+	OUT_RING(CP_PACKET0(R300_CP_RESYNC_DATA, 0));
+	OUT_RING(*id);
+	OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+	OUT_RING(R300_RB3D_DC_FINISH);
+	ADVANCE_RING();	
+}
+
+uint32_t r100_cs_id_last_get(struct drm_device *dev)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+
+	return RADEON_READ(RADEON_SCRATCH_REG4);
+}
+
+uint32_t r300_cs_id_last_get(struct drm_device *dev)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+
+	return RADEON_READ(RADEON_SCRATCH_REG6);
+}
+
+int radeon_cs_init(struct drm_device *dev)
+{
+	drm_radeon_private_t *dev_priv = dev->dev_private;
+
+	if (dev_priv->chip_family < CHIP_RV280) {
+		dev_priv->cs.id_emit = r100_cs_id_emit;
+		dev_priv->cs.id_last_get = r100_cs_id_last_get;
+	} else if (dev_priv->chip_family < CHIP_R600) {
+		dev_priv->cs.id_emit = r300_cs_id_emit;
+		dev_priv->cs.id_last_get = r300_cs_id_last_get;
+	}
+
+	dev_priv->cs.parse = radeon_cs_parse;
+	/* ib get depends on memory manager or not so memory manager */
+	return 0;
+}
diff --git a/shared-core/radeon_drm.h b/shared-core/radeon_drm.h
index 7fcf9305..c20de561 100644
--- a/shared-core/radeon_drm.h
+++ b/shared-core/radeon_drm.h
@@ -506,6 +506,7 @@ typedef struct {
 #define DRM_RADEON_GEM_SET_DOMAIN 0x23
 #define DRM_RADEON_GEM_INDIRECT 0x24 // temporary for X server
 
+#define DRM_RADEON_CS           0x25
 
 #define DRM_IOCTL_RADEON_CP_INIT    DRM_IOW( DRM_COMMAND_BASE + DRM_RADEON_CP_INIT, drm_radeon_init_t)
 #define DRM_IOCTL_RADEON_CP_START   DRM_IO(  DRM_COMMAND_BASE + DRM_RADEON_CP_START)
@@ -545,6 +546,7 @@ typedef struct {
 #define DRM_IOCTL_RADEON_GEM_SET_DOMAIN  DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_SET_DOMAIN, struct drm_radeon_gem_set_domain)
 #define DRM_IOCTL_RADEON_GEM_INDIRECT DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_INDIRECT, struct drm_radeon_gem_indirect)
 
+#define DRM_IOCTL_RADEON_CS DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_CS, struct drm_radeon_cs)
 
 
 typedef struct drm_radeon_init {
@@ -861,4 +863,17 @@ struct drm_radeon_gem_indirect {
 	uint32_t used;
 };
 
+/* New interface which obsolete all previous interface.
+ */
+
+
+struct drm_radeon_cs {
+//	uint32_t __user     *packets;
+	uint32_t            dwords;
+	uint32_t            cs_id;
+	uint64_t            packets;
+
+};
+
+
 #endif
diff --git a/shared-core/radeon_drv.h b/shared-core/radeon_drv.h
index b8f49404..c431fe4a 100644
--- a/shared-core/radeon_drv.h
+++ b/shared-core/radeon_drv.h
@@ -195,11 +195,11 @@ enum radeon_mac_model {
 
 
 #define GET_RING_HEAD(dev_priv)	(dev_priv->writeback_works ? \
-				 (dev_priv->mm.ring_read_ptr ? readl(dev_priv->mm.ring_read_ptr_map.virtual + 0) : DRM_READ32((dev_priv)->ring_rptr, 0 )) : \
+				 (dev_priv->mm.ring_read.bo ? readl(dev_priv->mm.ring_read.kmap.virtual + 0) : DRM_READ32((dev_priv)->ring_rptr, 0 )) : \
 				 RADEON_READ(RADEON_CP_RB_RPTR))
 
-#define SET_RING_HEAD(dev_priv,val) (dev_priv->mm.ring_read_ptr ? \
-				     writel((val), dev_priv->mm.ring_read_ptr_map.virtual) : \
+#define SET_RING_HEAD(dev_priv,val) (dev_priv->mm.ring_read.bo ? \
+				     writel((val), dev_priv->mm.ring_read.kmap.virtual) : \
 				     DRM_WRITE32((dev_priv)->ring_rptr, 0, (val)))
 
 typedef struct drm_radeon_freelist {
@@ -261,6 +261,11 @@ struct radeon_virt_surface {
 	struct drm_file *file_priv;
 };
 
+struct radeon_mm_obj {
+	struct drm_buffer_object *bo;
+	struct drm_bo_kmap_obj kmap;
+};
+
 struct radeon_mm_info {
 	uint64_t vram_offset; // Offset into GPU space
 	uint64_t vram_size;
@@ -268,15 +273,10 @@ struct radeon_mm_info {
 	
 	uint64_t gart_start;
 	uint64_t gart_size;
-
-	struct drm_buffer_object *pcie_table;
-	struct drm_bo_kmap_obj pcie_table_map;
-
-	struct drm_buffer_object *ring;
-	struct drm_bo_kmap_obj ring_map;
-
-	struct drm_buffer_object *ring_read_ptr;
-	struct drm_bo_kmap_obj ring_read_ptr_map;
+	
+	struct radeon_mm_obj pcie_table;
+	struct radeon_mm_obj ring;
+	struct radeon_mm_obj ring_read;
 };
 
 #include "radeon_mode.h"
@@ -286,6 +286,25 @@ struct drm_radeon_master_private {
 	drm_radeon_sarea_t *sarea_priv;
 };
 
+/* command submission struct */
+struct drm_radeon_cs_priv {
+	uint32_t id_wcnt;
+	uint32_t id_scnt;
+	uint32_t id_last_wcnt;
+	uint32_t id_last_scnt;
+
+	int (*parse)(struct drm_device *dev, void *ib,
+			uint32_t *packets, uint32_t dwords);
+	void (*id_emit)(struct drm_device *dev, uint32_t *id);
+	uint32_t (*id_last_get)(struct drm_device *dev);
+	/* this ib handling callback are for hidding memory manager drm
+	 * from memory manager less drm, free have to emit ib discard
+	 * sequence into the ring */
+	int (*ib_get)(struct drm_device *dev, void **ib, uint32_t dwords);
+	uint32_t (*ib_get_ptr)(struct drm_device *dev, void *ib);
+	void (*ib_free)(struct drm_device *dev, void *ib, uint32_t dwords);
+};
+
 typedef struct drm_radeon_private {
 
 	drm_radeon_ring_buffer_t ring;
@@ -392,7 +411,11 @@ typedef struct drm_radeon_private {
 	u32 ram_width;
 
 	enum radeon_pll_errata pll_errata;
-	
+
+	struct radeon_mm_obj **ib_objs;
+	/* ib bitmap */
+	uint64_t ib_alloc_bitmap; // TO DO replace with a real bitmap
+	struct drm_radeon_cs_priv cs;
 } drm_radeon_private_t;
 
 typedef struct drm_radeon_buf_priv {
@@ -669,14 +692,15 @@ extern int r300_do_cp_cmdbuf(struct drm_device *dev,
 #define RADEON_SCRATCH_REG3		0x15ec
 #define RADEON_SCRATCH_REG4		0x15f0
 #define RADEON_SCRATCH_REG5		0x15f4
+#define RADEON_SCRATCH_REG6		0x15f8
 #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 ?			\
-				 (dev_priv->mm.ring_read_ptr ? \
-				  readl(dev_priv->mm.ring_read_ptr_map.virtual + RADEON_SCRATCHOFF(0)) : \
+				 (dev_priv->mm.ring_read.bo ? \
+				  readl(dev_priv->mm.ring_read.kmap.virtual + RADEON_SCRATCHOFF(0)) : \
 				  DRM_READ32(dev_priv->ring_rptr, RADEON_SCRATCHOFF(x))) : \
 				 RADEON_READ( RADEON_SCRATCH_REG0 + 4*(x)))
 
@@ -1593,4 +1617,6 @@ extern void radeon_set_pcigart(drm_radeon_private_t * dev_priv, int on);
 extern int radeon_master_create(struct drm_device *dev, struct drm_master *master);
 extern void radeon_master_destroy(struct drm_device *dev, struct drm_master *master);
 extern void radeon_cp_dispatch_flip(struct drm_device * dev, struct drm_master *master);
+extern int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *fpriv);
+extern int radeon_cs_init(struct drm_device *dev);
 #endif				/* __RADEON_DRV_H__ */
diff --git a/shared-core/radeon_state.c b/shared-core/radeon_state.c
index 6de4b135..d56dac19 100644
--- a/shared-core/radeon_state.c
+++ b/shared-core/radeon_state.c
@@ -3279,6 +3279,7 @@ struct drm_ioctl_desc radeon_ioctls[] = {
 	DRM_IOCTL_DEF(DRM_RADEON_GEM_PWRITE, radeon_gem_pwrite_ioctl, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_RADEON_GEM_SET_DOMAIN, radeon_gem_set_domain_ioctl, DRM_AUTH),
 	DRM_IOCTL_DEF(DRM_RADEON_GEM_INDIRECT, radeon_gem_indirect_ioctl, DRM_AUTH),
+	DRM_IOCTL_DEF(DRM_RADEON_CS, radeon_cs_ioctl, DRM_AUTH),
 };
 
 int radeon_max_ioctl = DRM_ARRAY_SIZE(radeon_ioctls);