From 61da612a4f8862e0aac4ff4fc87c133cb8a1c4a5 Mon Sep 17 00:00:00 2001
From: Dave Airlie <airlied@redhat.com>
Date: Wed, 14 Jan 2009 15:12:57 +1000
Subject: [PATCH] r300: start moving new r300 cmdbuf into common code

---
 src/mesa/drivers/dri/r200/Makefile             |   2 +
 src/mesa/drivers/dri/r200/r200_state.c         |   4 +-
 src/mesa/drivers/dri/r300/r300_cmdbuf.c        | 108 +++---------------------
 src/mesa/drivers/dri/r300/r300_cmdbuf.h        |  13 ++-
 src/mesa/drivers/dri/r300/r300_context.c       |   3 +-
 src/mesa/drivers/dri/r300/r300_context.h       |  17 +---
 src/mesa/drivers/dri/r300/r300_ioctl.c         |   6 +-
 src/mesa/drivers/dri/r300/r300_render.c        |  10 +--
 src/mesa/drivers/dri/r300/r300_swtcl.c         |   6 +-
 src/mesa/drivers/dri/r300/radeon_context.h     |   1 -
 src/mesa/drivers/dri/r300/radeon_lock.h        |  74 +---------------
 src/mesa/drivers/dri/radeon/common_context.h   |  29 +++++--
 src/mesa/drivers/dri/radeon/common_lock.c      |   2 +-
 src/mesa/drivers/dri/radeon/common_lock.h      |  23 +++--
 src/mesa/drivers/dri/radeon/common_misc.c      | 112 ++++++++++++++++++++++++-
 src/mesa/drivers/dri/radeon/common_misc.h      |   6 ++
 src/mesa/drivers/dri/radeon/radeon_cs_legacy.h |   2 +-
 src/mesa/drivers/dri/radeon/radeon_screen.c    |   1 +
 18 files changed, 194 insertions(+), 225 deletions(-)

diff --git a/src/mesa/drivers/dri/r200/Makefile b/src/mesa/drivers/dri/r200/Makefile
index 34aa747..5ba1603 100644
--- a/src/mesa/drivers/dri/r200/Makefile
+++ b/src/mesa/drivers/dri/r200/Makefile
@@ -51,6 +51,7 @@ COMMON_SYMLINKS = \
 	radeon_screen.c \
 	radeon_screen.h \
 	radeon_bo_legacy.h \
+	radeon_cs_legacy.h \
 	radeon_buffer.h \
 	common_context.h \
 	common_lock.c \
@@ -58,6 +59,7 @@ COMMON_SYMLINKS = \
 	common_misc.h \
 	common_misc.c
 
+
 ##### TARGETS #####
 
 
diff --git a/src/mesa/drivers/dri/r200/r200_state.c b/src/mesa/drivers/dri/r200/r200_state.c
index 67357fa..997c171 100644
--- a/src/mesa/drivers/dri/r200/r200_state.c
+++ b/src/mesa/drivers/dri/r200/r200_state.c
@@ -1711,7 +1711,7 @@ void r200UpdateViewportOffset( GLcontext *ctx )
       }
    }
 
-   r200UpdateScissor( ctx );
+   radeonUpdateScissor( ctx );
 }
 
 
@@ -2056,7 +2056,7 @@ static void r200Enable( GLcontext *ctx, GLenum cap, GLboolean state )
    case GL_SCISSOR_TEST:
       R200_FIREVERTICES( rmesa );
       rmesa->radeon.state.scissor.enabled = state;
-      r200UpdateScissor( ctx );
+      radeonUpdateScissor( ctx );
       break;
 
    case GL_STENCIL_TEST:
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index 32d0dcf..bbca71d 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -69,70 +69,19 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
 #define SPACE_FOR_FLUSHING	4
 
-/**
- * Send the current command buffer via ioctl to the hardware.
- */
-int r300FlushCmdBufLocked(r300ContextPtr r300, const char *caller)
-{
-	int ret = 0;
-
-	if (r300->cmdbuf.flushing) {
-		fprintf(stderr, "Recursive call into r300FlushCmdBufLocked!\n");
-		exit(-1);
-	}
-	r300->cmdbuf.flushing = 1;
-    if (r300->cmdbuf.cs->cdw) {
-        ret = radeon_cs_emit(r300->cmdbuf.cs);
-        r300->hw.all_dirty = 1;
-    }
-    radeon_cs_erase(r300->cmdbuf.cs);
-	r300->cmdbuf.flushing = 0;
-	return ret;
-}
-
-int r300FlushCmdBuf(r300ContextPtr r300, const char *caller)
-{
-	int ret;
-
-	LOCK_HARDWARE(&r300->radeon);
-	ret = r300FlushCmdBufLocked(r300, caller);
-	UNLOCK_HARDWARE(&r300->radeon);
-
-	if (ret) {
-		fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret);
-		_mesa_exit(ret);
-	}
-
-	return ret;
-}
-
-/**
- * Make sure that enough space is available in the command buffer
- * by flushing if necessary.
- *
- * \param dwords The number of dwords we need to be free on the command buffer
- */
-void r300EnsureCmdBufSpace(r300ContextPtr r300, int dwords, const char *caller)
-{
-	if ((r300->cmdbuf.cs->cdw + dwords + 128) > r300->cmdbuf.size ||
-        radeon_cs_need_flush(r300->cmdbuf.cs)) {
-		r300FlushCmdBuf(r300, caller);
-    }
-}
-
 void r300BeginBatch(r300ContextPtr r300, int n,
 		    int dostate,
                     const char *file,
                     const char *function,
                     int line)
 {
-	r300EnsureCmdBufSpace(r300, n, function);
-	if (!r300->cmdbuf.cs->cdw && dostate) {
+	rcommonEnsureCmdBufSpace(&r300->radeon, n, function);
+	if (!r300->radeon.cmdbuf.cs->cdw && dostate) {
 		if (RADEON_DEBUG & DEBUG_IOCTL)
 			fprintf(stderr, "Reemit state after flush (from %s)\n", function);
 		r300EmitState(r300);
 	}
-    radeon_cs_begin(r300->cmdbuf.cs, n, file, function, line);
+	radeon_cs_begin(r300->radeon.cmdbuf.cs, n, file, function, line);
 }
 
 static void r300PrintStateAtom(r300ContextPtr r300,
@@ -209,15 +158,15 @@ void r300EmitState(r300ContextPtr r300)
 	if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_PRIMS))
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
-	if (r300->cmdbuf.cs->cdw && !r300->hw.is_dirty && !r300->hw.all_dirty)
+	if (r300->radeon.cmdbuf.cs->cdw && !r300->hw.is_dirty && !r300->hw.all_dirty)
 		return;
 
 	/* To avoid going across the entire set of states multiple times, just check
 	 * for enough space for the case of emitting all state.
 	 */
-	r300EnsureCmdBufSpace(r300, r300->hw.max_state_size, __FUNCTION__);
+	rcommonEnsureCmdBufSpace(&r300->radeon, r300->hw.max_state_size, __FUNCTION__);
 
-	if (!r300->cmdbuf.cs->cdw) {
+	if (!r300->radeon.cmdbuf.cs->cdw) {
 		if (RADEON_DEBUG & DEBUG_STATE)
 			fprintf(stderr, "Begin reemit state\n");
 
@@ -473,7 +422,7 @@ int check_r500fp_const(GLcontext *ctx, struct radeon_state_atom *atom)
  */
 void r300InitCmdBuf(r300ContextPtr r300)
 {
-	int size, mtu;
+	int mtu;
 	int has_tcl = 1;
 	int is_r500 = 0;
 	int i;
@@ -770,40 +719,7 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	r300->hw.is_dirty = GL_TRUE;
 	r300->hw.all_dirty = GL_TRUE;
 
-	/* Initialize command buffer */
-	size =
-	    256 * driQueryOptioni(&r300->radeon.optionCache,
-				  "command_buffer_size");
-	if (size < 2 * r300->hw.max_state_size) {
-		size = 2 * r300->hw.max_state_size + 65535;
-	}
-	if (size > 64 * 256)
-		size = 64 * 256;
-
-	size = 64 * 1024 / 4;
-	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) {
-		fprintf(stderr, "sizeof(drm_r300_cmd_header_t)=%zd\n",
-			sizeof(drm_r300_cmd_header_t));
-		fprintf(stderr, "sizeof(drm_radeon_cmd_buffer_t)=%zd\n",
-			sizeof(drm_radeon_cmd_buffer_t));
-		fprintf(stderr,
-			"Allocating %d bytes command buffer (max state is %d bytes)\n",
-			size * 4, r300->hw.max_state_size * 4);
-	}
-
-    if (r300->radeon.radeonScreen->kernel_mm) {
-        int fd = r300->radeon.radeonScreen->driScreen->fd;
-        r300->cmdbuf.csm = radeon_cs_manager_gem_ctor(fd);
-    } else {
-        r300->cmdbuf.csm = radeon_cs_manager_legacy_ctor(&r300->radeon);
-    }
-    if (r300->cmdbuf.csm == NULL) {
-        /* FIXME: fatal error */
-        return;
-    }
-    r300->cmdbuf.cs = radeon_cs_create(r300->cmdbuf.csm, size);
-    assert(r300->cmdbuf.cs != NULL);
-	r300->cmdbuf.size = size;
+	rcommonInitCmdBuf(&r300->radeon, r300->hw.max_state_size);
 }
 
 /**
@@ -813,13 +729,9 @@ void r300DestroyCmdBuf(r300ContextPtr r300)
 {
 	struct radeon_state_atom *atom;
 
-    radeon_cs_destroy(r300->cmdbuf.cs);
 	foreach(atom, &r300->hw.atomlist) {
 		FREE(atom->cmd);
 	}
-    if (r300->radeon.radeonScreen->driScreen->dri2.enabled || r300->radeon.radeonScreen->kernel_mm) {
-        radeon_cs_manager_gem_dtor(r300->cmdbuf.csm);
-    } else {
-        radeon_cs_manager_legacy_dtor(r300->cmdbuf.csm);
-    }
+	rcommonDestroyCmdBuf(&r300->radeon);
+
 }
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.h b/src/mesa/drivers/dri/r300/r300_cmdbuf.h
index f4cc5d3..0c76c6c 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.h
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.h
@@ -39,14 +39,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_context.h"
 #include "radeon_cs.h"
 
-extern int r300FlushCmdBufLocked(r300ContextPtr r300, const char *caller);
-extern int r300FlushCmdBuf(r300ContextPtr r300, const char *caller);
 
 extern void r300EmitState(r300ContextPtr r300);
 
 extern void r300InitCmdBuf(r300ContextPtr r300);
 extern void r300DestroyCmdBuf(r300ContextPtr r300);
-extern void r300EnsureCmdBufSpace(r300ContextPtr r300, int dwords, const char *caller);
 
 void r300BeginBatch(r300ContextPtr r300,
 		    int n,
@@ -78,7 +75,7 @@ void r300BeginBatch(r300ContextPtr r300,
  */
 #define OUT_BATCH(data) \
 	do { \
-        radeon_cs_write_dword(b_l_r300->cmdbuf.cs, data);\
+        radeon_cs_write_dword(b_l_r300->radeon.cmdbuf.cs, data);\
 	} while(0)
 
 /**
@@ -90,8 +87,8 @@ void r300BeginBatch(r300ContextPtr r300,
             fprintf(stderr, "(%s:%s:%d) offset : %d\n",\
             __FILE__, __FUNCTION__, __LINE__, offset);\
         }\
-        radeon_cs_write_dword(b_l_r300->cmdbuf.cs, offset);\
-        radeon_cs_write_reloc(b_l_r300->cmdbuf.cs, \
+        radeon_cs_write_dword(b_l_r300->radeon.cmdbuf.cs, offset);\
+        radeon_cs_write_reloc(b_l_r300->radeon.cmdbuf.cs, \
                               bo, \
                               rd, \
                               wd, \
@@ -105,7 +102,7 @@ void r300BeginBatch(r300ContextPtr r300,
 	do { \
 		int _i; \
         for (_i=0; _i < n; _i++) {\
-            radeon_cs_write_dword(b_l_r300->cmdbuf.cs, ptr[_i]);\
+            radeon_cs_write_dword(b_l_r300->radeon.cmdbuf.cs, ptr[_i]);\
         }\
 	} while(0)
 
@@ -116,7 +113,7 @@ void r300BeginBatch(r300ContextPtr r300,
  */
 #define END_BATCH() \
 	do { \
-        radeon_cs_end(b_l_r300->cmdbuf.cs, __FILE__, __FUNCTION__, __LINE__);\
+        radeon_cs_end(b_l_r300->radeon.cmdbuf.cs, __FILE__, __FUNCTION__, __LINE__);\
 	} while(0)
 
 /**
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index 974cf5f..6b79e7d 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -65,6 +65,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_tex.h"
 #include "r300_emit.h"
 #include "r300_swtcl.h"
+#include "radeon_bo_legacy.h"
 
 
 #include "vblank.h"
@@ -443,7 +444,7 @@ void r300DestroyContext(__DRIcontextPrivate * driContextPriv)
 		_vbo_DestroyContext(r300->radeon.glCtx);
 		_swrast_DestroyContext(r300->radeon.glCtx);
 
-		r300FlushCmdBuf(r300, __FUNCTION__);
+		rcommonFlushCmdBuf(&r300->radeon, __FUNCTION__);
 		r300DestroyCmdBuf(r300);
 
 		if (radeon->state.scissor.pClipRects) {
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index de188c3..1a43016 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -42,6 +42,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_drm.h"
 #include "dri_util.h"
 #include "texmem.h"
+#include "radeon_context.h"
 #include "radeon_bo.h"
 
 #include "main/macros.h"
@@ -505,20 +506,6 @@ struct r300_hw_state {
 };
 
 /**
- * This structure holds the command buffer while it is being constructed.
- *
- * The first batch of commands in the buffer is always the state that needs
- * to be re-emitted when the context is lost. This batch can be skipped
- * otherwise.
- */
-struct r300_cmdbuf {
-    struct radeon_cs_manager    *csm;
-    struct radeon_cs            *cs;
-	int size; /** # of dwords total */
-	unsigned int flushing:1; /** whether we're currently in FlushCmdBufLocked */
-};
-
-/**
  * State cache
  */
 
@@ -838,7 +825,7 @@ struct r300_context {
 	struct radeon_context radeon;	/* parent class, must be first */
 
 	struct r300_hw_state hw;
-	struct r300_cmdbuf cmdbuf;
+
 	struct r300_state state;
 	struct gl_vertex_program *curr_vp;
 	struct r300_vertex_program *selected_vp;
diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c
index e70e5f6..6398963 100644
--- a/src/mesa/drivers/dri/r300/r300_ioctl.c
+++ b/src/mesa/drivers/dri/r300/r300_ioctl.c
@@ -589,7 +589,7 @@ static void r300Clear(GLcontext * ctx, GLbitfield mask)
 	}
 
 	/* Make sure it fits there. */
-	r300EnsureCmdBufSpace(r300, 421 * 3, __FUNCTION__);
+	rcommonEnsureCmdBufSpace(&r300->radeon, 421 * 3, __FUNCTION__);
 	if (flags || bits)
 		r300EmitClearState(ctx);
     rrbd = (void *)fb->Attachment[BUFFER_DEPTH].Renderbuffer;
@@ -623,8 +623,8 @@ void r300Flush(GLcontext * ctx)
         rmesa->swtcl.flush(rmesa);
     }
 
-	if (rmesa->cmdbuf.cs->cdw) {
-		r300FlushCmdBuf(rmesa, __FUNCTION__);
+	if (rmesa->radeon.cmdbuf.cs->cdw) {
+		rcommonFlushCmdBuf(&rmesa->radeon, __FUNCTION__);
 	}
 }
 
diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index ccc00b3..cb6652f 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -212,7 +212,7 @@ static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type)
 	    OUT_BATCH(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
         OUT_BATCH(rmesa->state.elt_dma_offset);
         OUT_BATCH(vertex_count);
-        radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+        radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
                               rmesa->state.elt_dma_bo,
                               RADEON_GEM_DOMAIN_GTT, 0, 0);
     }
@@ -295,13 +295,13 @@ static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
 	for (i = 0; i + 1 < nr; i += 2) {
         voffset =  rmesa->state.aos[i + 0].offset +
                    offset * 4 * rmesa->state.aos[i + 0].stride;
-        radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+        radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
                               rmesa->state.aos[i+0].bo,
                               RADEON_GEM_DOMAIN_GTT,
                               0, 0);
         voffset =  rmesa->state.aos[i + 1].offset +
                    offset * 4 * rmesa->state.aos[i + 1].stride;
-        radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+        radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
                               rmesa->state.aos[i+1].bo,
                               RADEON_GEM_DOMAIN_GTT,
                               0, 0);
@@ -309,7 +309,7 @@ static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
 	if (nr & 1) {
         voffset =  rmesa->state.aos[nr - 1].offset +
                    offset * 4 * rmesa->state.aos[nr - 1].stride;
-        radeon_cs_write_reloc(rmesa->cmdbuf.cs,
+        radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs,
                               rmesa->state.aos[nr-1].bo,
                               RADEON_GEM_DOMAIN_GTT,
                               0, 0);
@@ -346,7 +346,7 @@ static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx,
 	 * This is supposed to ensure that we can get all rendering
 	 * commands into a single command buffer.
 	 */
-	r300EnsureCmdBufSpace(rmesa, 64, __FUNCTION__);
+	rcommonEnsureCmdBufSpace(&rmesa->radeon, 64, __FUNCTION__);
 
 	if (vb->Elts) {
 		if (num_verts > 65535) {
diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c
index fbfa8f4..0108f7c 100644
--- a/src/mesa/drivers/dri/r300/r300_swtcl.c
+++ b/src/mesa/drivers/dri/r300/r300_swtcl.c
@@ -247,9 +247,9 @@ static void flush_last_swtcl_prim( r300ContextPtr rmesa  )
 		fprintf(stderr, "%s\n", __FUNCTION__);
     rmesa->swtcl.flush = NULL;
     radeon_bo_unmap(rmesa->swtcl.bo);
-    r300EnsureCmdBufSpace(rmesa,
-                          rmesa->hw.max_state_size + (12*sizeof(int)),
-                          __FUNCTION__);
+    rcommonEnsureCmdBufSpace(rmesa,
+			     rmesa->hw.max_state_size + (12*sizeof(int)),
+			     __FUNCTION__);
     r300EmitState(rmesa);
     r300EmitVertexAOS(rmesa,
                       rmesa->swtcl.vertex_size,
diff --git a/src/mesa/drivers/dri/r300/radeon_context.h b/src/mesa/drivers/dri/r300/radeon_context.h
index be21a8c..736bf7c 100644
--- a/src/mesa/drivers/dri/r300/radeon_context.h
+++ b/src/mesa/drivers/dri/r300/radeon_context.h
@@ -50,7 +50,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "dri_util.h"
 
 #include "radeon_screen.h"
-#include "common_context.h"
 #include "common_misc.h"
 
 #if R200_MERGED
diff --git a/src/mesa/drivers/dri/r300/radeon_lock.h b/src/mesa/drivers/dri/r300/radeon_lock.h
index eaef49c..f2d17a5 100644
--- a/src/mesa/drivers/dri/r300/radeon_lock.h
+++ b/src/mesa/drivers/dri/r300/radeon_lock.h
@@ -42,78 +42,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifndef __RADEON_LOCK_H__
 #define __RADEON_LOCK_H__
 
-#include "radeon_context.h"
-
-extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags);
-extern void radeonUpdatePageFlipping(radeonContextPtr rmesa);
-
-/* Turn DEBUG_LOCKING on to find locking conflicts.
- */
-#define DEBUG_LOCKING	0
-
-#if DEBUG_LOCKING
-extern char *prevLockFile;
-extern int prevLockLine;
-
-#define DEBUG_LOCK()							\
-   do {									\
-      prevLockFile = (__FILE__);					\
-      prevLockLine = (__LINE__);					\
-   } while (0)
-
-#define DEBUG_RESET()							\
-   do {									\
-      prevLockFile = 0;							\
-      prevLockLine = 0;							\
-   } while (0)
-
-#define DEBUG_CHECK_LOCK()						\
-   do {									\
-      if (prevLockFile) {						\
-	 fprintf(stderr,						\
-		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
-		  prevLockFile, prevLockLine, __FILE__, __LINE__);	\
-	 exit(1);							\
-      }									\
-   } while (0)
-
-#else
-
-#define DEBUG_LOCK()
-#define DEBUG_RESET()
-#define DEBUG_CHECK_LOCK()
-
-#endif
-
-/*
- * !!! We may want to separate locks from locks with validation.  This
- * could be used to improve performance for those things commands that
- * do not do any drawing !!!
- */
-
-/* Lock the hardware and validate our state.
- */
-#define LOCK_HARDWARE( rmesa )						\
-	do {								\
-		char __ret = 0;						\
-		DEBUG_CHECK_LOCK();					\
-        if (!(rmesa)->radeonScreen->driScreen->dri2.enabled) { \
-		DRM_CAS((rmesa)->dri.hwLock, (rmesa)->dri.hwContext,	\
-			(DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret); \
-		if (__ret)						\
-			radeonGetLock((rmesa), 0);			\
-        }\
-		DEBUG_LOCK();						\
-	} while (0)
-
-#define UNLOCK_HARDWARE( rmesa )					\
-	do {								\
-        if (!(rmesa)->radeonScreen->driScreen->dri2.enabled) { \
-		DRM_UNLOCK((rmesa)->dri.fd,				\
-			(rmesa)->dri.hwLock,				\
-			(rmesa)->dri.hwContext);			\
-		DEBUG_RESET();						\
-        }\
-	} while (0)
+#include "common_lock.h"
 
 #endif				/* __RADEON_LOCK_H__ */
diff --git a/src/mesa/drivers/dri/radeon/common_context.h b/src/mesa/drivers/dri/radeon/common_context.h
index e4d2103..eb95dc9 100644
--- a/src/mesa/drivers/dri/radeon/common_context.h
+++ b/src/mesa/drivers/dri/radeon/common_context.h
@@ -1,6 +1,14 @@
 
 #ifndef COMMON_CONTEXT_H
 #define COMMON_CONTEXT_H
+
+#include "main/mm.h"
+#include "math/m_vector.h"
+#include "texmem.h"
+#include "tnl/t_context.h"
+#include "main/colormac.h"
+
+
 /* This union is used to avoid warnings/miscompilation
    with float to uint32_t casts due to strict-aliasing */
 typedef union { GLfloat f; uint32_t ui32; } float_ui32_type;
@@ -9,11 +17,6 @@ struct radeon_context;
 typedef struct radeon_context radeonContextRec;
 typedef struct radeon_context *radeonContextPtr;
 
-#include "main/mm.h"
-#include "math/m_vector.h"
-#include "texmem.h"
-#include "tnl/t_context.h"
-
 
 #define TEX_0   0x1
 #define TEX_1   0x2
@@ -256,6 +259,20 @@ struct radeon_state {
 	struct radeon_stencilbuffer_state stencil;
 };
 
+/**
+ * This structure holds the command buffer while it is being constructed.
+ *
+ * The first batch of commands in the buffer is always the state that needs
+ * to be re-emitted when the context is lost. This batch can be skipped
+ * otherwise.
+ */
+struct radeon_cmdbuf {
+	struct radeon_cs_manager    *csm;
+	struct radeon_cs            *cs;
+	int size; /** # of dwords total */
+	unsigned int flushing:1; /** whether we're currently in FlushCmdBufLocked */
+};
+
 struct radeon_context {
    GLcontext *glCtx;
    radeonScreenPtr radeonScreen;	/* Screen private DRI data */
@@ -308,6 +325,8 @@ struct radeon_context {
     */
    driOptionCache optionCache;
 
+   struct radeon_cmdbuf cmdbuf;
+
    struct {
       void (*get_lock)(radeonContextPtr radeon);
       void (*update_viewport_offset)(GLcontext *ctx);
diff --git a/src/mesa/drivers/dri/radeon/common_lock.c b/src/mesa/drivers/dri/radeon/common_lock.c
index d983e8c..0951745 100644
--- a/src/mesa/drivers/dri/radeon/common_lock.c
+++ b/src/mesa/drivers/dri/radeon/common_lock.c
@@ -84,7 +84,7 @@ void radeonUpdatePageFlipping(radeonContextPtr rmesa)
 #else
 /* Turn on/off page flipping according to the flags in the sarea:
  */
-static void radeonUpdatePageFlipping(radeonContextPtr rmesa)
+void radeonUpdatePageFlipping(radeonContextPtr rmesa)
 {
 	rmesa->doPageFlip = rmesa->sarea->pfState;
 	if (rmesa->glCtx->WinSysDrawBuffer) {
diff --git a/src/mesa/drivers/dri/radeon/common_lock.h b/src/mesa/drivers/dri/radeon/common_lock.h
index 50a4a0b..431b076 100644
--- a/src/mesa/drivers/dri/radeon/common_lock.h
+++ b/src/mesa/drivers/dri/radeon/common_lock.h
@@ -41,6 +41,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #ifndef COMMON_LOCK_H
 #define COMMON_LOCK_H
+
+#include "main/colormac.h"
+#include "radeon_screen.h"
+#include "common_context.h"
+
 extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags);
 
 /* Turn DEBUG_LOCKING on to find locking conflicts.
@@ -93,19 +98,23 @@ extern int prevLockLine;
    do {								\
       char __ret = 0;						\
       DEBUG_CHECK_LOCK();					\
-      DRM_CAS( (rmesa)->dri.hwLock, (rmesa)->dri.hwContext,		\
-	       (DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret );	\
-      if ( __ret )							\
+      if (!(rmesa)->radeonScreen->driScreen->dri2.enabled) {		\
+	DRM_CAS( (rmesa)->dri.hwLock, (rmesa)->dri.hwContext,		\
+		 (DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret );	\
+	if ( __ret )							\
 	  radeonGetLock( (rmesa), 0 );					\
+      }									\
       DEBUG_LOCK();							\
    } while (0)
 
 #define UNLOCK_HARDWARE( rmesa )					\
    do {									\
-      DRM_UNLOCK( (rmesa)->dri.fd,					\
-		  (rmesa)->dri.hwLock,					\
-		  (rmesa)->dri.hwContext );				\
-      DEBUG_RESET();							\
+     if (!(rmesa)->radeonScreen->driScreen->dri2.enabled) {		\
+       DRM_UNLOCK( (rmesa)->dri.fd,					\
+		   (rmesa)->dri.hwLock,					\
+		   (rmesa)->dri.hwContext );				\
+       DEBUG_RESET();							\
+     }									\
    } while (0)
 
 #endif
diff --git a/src/mesa/drivers/dri/radeon/common_misc.c b/src/mesa/drivers/dri/radeon/common_misc.c
index eab9cc7..ba3c76d 100644
--- a/src/mesa/drivers/dri/radeon/common_misc.c
+++ b/src/mesa/drivers/dri/radeon/common_misc.c
@@ -51,11 +51,16 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "drirenderbuffer.h"
 #include "vblank.h"
 
-
+#include "radeon_bo.h"
+#include "radeon_cs.h"
+#include "radeon_bo_legacy.h"
+#include "radeon_cs_legacy.h"
+#include "radeon_bo_gem.h"
+#include "radeon_cs_gem.h"
 #include "dri_util.h"
 #include "radeon_drm.h"
-#include "radeon_screen.h"
 #include "radeon_buffer.h"
+#include "radeon_screen.h"
 #include "common_context.h"
 #include "common_misc.h"
 #include "common_lock.h"
@@ -577,3 +582,106 @@ void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
     }
 }
 
+/* cmdbuffer */
+/**
+ * Send the current command buffer via ioctl to the hardware.
+ */
+int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller)
+{
+	int ret = 0;
+
+	if (rmesa->cmdbuf.flushing) {
+		fprintf(stderr, "Recursive call into r300FlushCmdBufLocked!\n");
+		exit(-1);
+	}
+	rmesa->cmdbuf.flushing = 1;
+	if (rmesa->cmdbuf.cs->cdw) {
+		ret = radeon_cs_emit(rmesa->cmdbuf.cs);
+		rmesa->vtbl.set_all_dirty(rmesa->glCtx);
+	}
+	radeon_cs_erase(rmesa->cmdbuf.cs);
+	rmesa->cmdbuf.flushing = 0;
+	return ret;
+}
+
+int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller)
+{
+	int ret;
+
+	LOCK_HARDWARE(rmesa);
+	ret = rcommonFlushCmdBufLocked(rmesa, caller);
+	UNLOCK_HARDWARE(rmesa);
+
+	if (ret) {
+		fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret);
+		_mesa_exit(ret);
+	}
+
+	return ret;
+}
+
+/**
+ * Make sure that enough space is available in the command buffer
+ * by flushing if necessary.
+ *
+ * \param dwords The number of dwords we need to be free on the command buffer
+ */
+void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller)
+{
+	if ((rmesa->cmdbuf.cs->cdw + dwords + 128) > rmesa->cmdbuf.size ||
+	    radeon_cs_need_flush(rmesa->cmdbuf.cs)) {
+		rcommonFlushCmdBuf(rmesa, caller);
+    }
+}
+
+void rcommonInitCmdBuf(radeonContextPtr rmesa, int max_state_size)
+{
+	GLuint size;
+	/* Initialize command buffer */
+	size = 256 * driQueryOptioni(&rmesa->optionCache,
+				     "command_buffer_size");
+	if (size < 2 * max_state_size) {
+		size = 2 * max_state_size + 65535;
+	}
+	if (size > 64 * 256)
+		size = 64 * 256;
+
+	size = 64 * 1024 / 4;
+
+	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) {
+		fprintf(stderr, "sizeof(drm_r300_cmd_header_t)=%zd\n",
+			sizeof(drm_r300_cmd_header_t));
+		fprintf(stderr, "sizeof(drm_radeon_cmd_buffer_t)=%zd\n",
+			sizeof(drm_radeon_cmd_buffer_t));
+		fprintf(stderr,
+			"Allocating %d bytes command buffer (max state is %d bytes)\n",
+			size * 4, max_state_size * 4);
+	}
+
+	if (rmesa->radeonScreen->kernel_mm) {
+		int fd = rmesa->radeonScreen->driScreen->fd;
+		rmesa->cmdbuf.csm = radeon_cs_manager_gem_ctor(fd);
+	} else {
+		rmesa->cmdbuf.csm = radeon_cs_manager_legacy_ctor(rmesa);
+	}
+	if (rmesa->cmdbuf.csm == NULL) {
+		/* FIXME: fatal error */
+		return;
+	}
+	rmesa->cmdbuf.cs = radeon_cs_create(rmesa->cmdbuf.csm, size);
+	assert(rmesa->cmdbuf.cs != NULL);
+	rmesa->cmdbuf.size = size;
+
+}
+/**
+ * Destroy the command buffer
+ */
+void rcommonDestroyCmdBuf(radeonContextPtr rmesa)
+{
+	radeon_cs_destroy(rmesa->cmdbuf.cs);
+	if (rmesa->radeonScreen->driScreen->dri2.enabled || rmesa->radeonScreen->kernel_mm) {
+		radeon_cs_manager_gem_dtor(rmesa->cmdbuf.csm);
+	} else {
+		radeon_cs_manager_legacy_dtor(rmesa->cmdbuf.csm);
+	}
+}
diff --git a/src/mesa/drivers/dri/radeon/common_misc.h b/src/mesa/drivers/dri/radeon/common_misc.h
index cc4832c..7057ad9 100644
--- a/src/mesa/drivers/dri/radeon/common_misc.h
+++ b/src/mesa/drivers/dri/radeon/common_misc.h
@@ -1,6 +1,7 @@
 #ifndef COMMON_MISC_H
 #define COMMON_MISC_H
 
+#include "common_context.h"
 void radeonRecalcScissorRects(radeonContextPtr radeon);
 void radeonSetCliprects(radeonContextPtr radeon);
 void radeonUpdateScissor( GLcontext *ctx );
@@ -17,4 +18,9 @@ void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
 
 void radeonUpdatePageFlipping(radeonContextPtr rmesa);
 
+void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller);
+int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller);
+int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller);
+void rcommonInitCmdBuf(radeonContextPtr rmesa, int max_state_size);
+void rcommonDestroyCmdBuf(radeonContextPtr rmesa);
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h
index 71a4dad..4da2479 100644
--- a/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h
+++ b/src/mesa/drivers/dri/radeon/radeon_cs_legacy.h
@@ -33,7 +33,7 @@
 #define RADEON_CS_LEGACY_H
 
 #include "radeon_cs.h"
-#include "radeon_context.h"
+#include "common_context.h"
 
 struct radeon_cs_manager *radeon_cs_manager_legacy_ctor(struct radeon_context *ctx);
 void radeon_cs_manager_legacy_dtor(struct radeon_cs_manager *csm);
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c
index c099d0e..e655408 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.c
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.c
@@ -47,6 +47,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_macros.h"
 #include "radeon_screen.h"
 #include "radeon_buffer.h"
+#include "common_misc.h"
 #if !RADEON_COMMON
 #include "radeon_context.h"
 #include "radeon_span.h"
-- 
2.7.4