r600g: avoid using pb* helper we are loosing previous cpu cycle with it

r600g is up to a point where all small CPU cycle matter and pb* turn
high on profile. It's mostly because pb try to be generic and thus
trigger unecessary check for r600g driver. To avoid having too much
abstraction & too much depth in the call embedded everythings into
r600_bo. Make code simpler & faster. The performance win highly depend
on the CPU & application considered being more important on slower CPU
and marginal/unoticeable on faster one.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>

9 files changed, 364 insertions, 388 deletions

diff --git a/src/gallium/winsys/r600/drm/Makefile b/src/gallium/winsys/r600/drm/Makefile
index a396205f89..91c65012c8 100644
--- a/src/gallium/winsys/r600/drm/Makefile
+++ b/src/gallium/winsys/r600/drm/Makefile
@@ -8,12 +8,12 @@ C_SOURCES = \
 	bof.c \
 	evergreen_hw_context.c \
 	radeon_bo.c \
-	radeon_bo_pb.c \
 	radeon_pciid.c \
 	r600.c \
 	r600_bo.c \
 	r600_drm.c \
-	r600_hw_context.c
+	r600_hw_context.c \
+	r600_bomgr.c
 
 LIBRARY_INCLUDES = -I$(TOP)/src/gallium/drivers/r600 \
 		   $(shell pkg-config libdrm --cflags-only-I)
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index e1f163eab6..2175d578ec 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -36,7 +36,6 @@
 #include "pipe/p_compiler.h"
 #include "util/u_inlines.h"
 #include "util/u_memory.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600_priv.h"
 
 #define GROUP_FORCE_NEW_BLOCK	0
diff --git a/src/gallium/winsys/r600/drm/r600.c b/src/gallium/winsys/r600/drm/r600.c
index f5e53e21f5..b88733f80f 100644
--- a/src/gallium/winsys/r600/drm/r600.c
+++ b/src/gallium/winsys/r600/drm/r600.c
@@ -27,7 +27,6 @@
 #include "radeon_drm.h"
 #include "pipe/p_compiler.h"
 #include "util/u_inlines.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600_priv.h"
 
 enum radeon_family r600_get_family(struct radeon *r600)
diff --git a/src/gallium/winsys/r600/drm/r600_bo.c b/src/gallium/winsys/r600/drm/r600_bo.c
index 251f009a6b..933b169935 100644
--- a/src/gallium/winsys/r600/drm/r600_bo.c
+++ b/src/gallium/winsys/r600/drm/r600_bo.c
@@ -36,142 +36,153 @@ struct r600_bo *r600_bo(struct radeon *radeon,
 			unsigned size, unsigned alignment,
 			unsigned binding, unsigned usage)
 {
-	struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo));
-	struct pb_desc desc;
-	struct pb_manager *man;
+	struct r600_bo *bo;
+	struct radeon_bo *rbo;
 
-	desc.alignment = alignment;
-	desc.usage = (PB_USAGE_CPU_READ_WRITE | PB_USAGE_GPU_READ_WRITE);
-	ws_bo->size = size;
+	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) {
+		bo = r600_bomgr_bo_create(radeon->bomgr, size, alignment, *radeon->cfence);
+		if (bo) {
+			return bo;
+		}
+	}
 
-	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER))
-		man = radeon->cman;
-	else
-		man = radeon->kman;
+	rbo = radeon_bo(radeon, 0, size, alignment);
+	if (rbo == NULL) {
+		return NULL;
+	}
+
+	bo = calloc(1, sizeof(struct r600_bo));
+	bo->size = size;
+	bo->alignment = alignment;
+	bo->bo = rbo;
+	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) {
+		r600_bomgr_bo_init(radeon->bomgr, bo);
+	}
 
 	/* Staging resources particpate in transfers and blits only
 	 * and are used for uploads and downloads from regular
 	 * resources.  We generate them internally for some transfers.
 	 */
 	if (usage == PIPE_USAGE_STAGING)
-                ws_bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT;
-        else
-                ws_bo->domains = (RADEON_GEM_DOMAIN_CPU |
-                                  RADEON_GEM_DOMAIN_GTT |
-                                  RADEON_GEM_DOMAIN_VRAM);
-
-
-	ws_bo->pb = man->create_buffer(man, size, &desc);
-	if (ws_bo->pb == NULL) {
-		free(ws_bo);
-		return NULL;
-	}
+		bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT;
+	else
+		bo->domains = (RADEON_GEM_DOMAIN_CPU |
+				RADEON_GEM_DOMAIN_GTT |
+				RADEON_GEM_DOMAIN_VRAM);
 
-	pipe_reference_init(&ws_bo->reference, 1);
-	return ws_bo;
+	pipe_reference_init(&bo->reference, 1);
+	return bo;
 }
 
 struct r600_bo *r600_bo_handle(struct radeon *radeon,
 			       unsigned handle, unsigned *array_mode)
 {
-	struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo));
-	struct radeon_bo *bo;
+	struct r600_bo *bo = calloc(1, sizeof(struct r600_bo));
+	struct radeon_bo *rbo;
 
-	ws_bo->pb = radeon_bo_pb_create_buffer_from_handle(radeon->kman, handle);
-	if (!ws_bo->pb) {
-		free(ws_bo);
+	rbo = bo->bo = radeon_bo(radeon, handle, 0, 0);
+	if (rbo == NULL) {
+		free(bo);
 		return NULL;
 	}
-	bo = radeon_bo_pb_get_bo(ws_bo->pb);
-	ws_bo->size = bo->size;
-	ws_bo->domains = (RADEON_GEM_DOMAIN_CPU |
-			  RADEON_GEM_DOMAIN_GTT |
-			  RADEON_GEM_DOMAIN_VRAM);
+	bo->size = bo->size;
+	bo->domains = (RADEON_GEM_DOMAIN_CPU |
+			RADEON_GEM_DOMAIN_GTT |
+			RADEON_GEM_DOMAIN_VRAM);
 
-	pipe_reference_init(&ws_bo->reference, 1);
+	pipe_reference_init(&bo->reference, 1);
 
-	radeon_bo_get_tiling_flags(radeon, bo, &ws_bo->tiling_flags,
-				   &ws_bo->kernel_pitch);
+	radeon_bo_get_tiling_flags(radeon, rbo, &bo->tiling_flags, &bo->kernel_pitch);
 	if (array_mode) {
-		if (ws_bo->tiling_flags) {
-			if (ws_bo->tiling_flags & RADEON_TILING_MICRO)
+		if (bo->tiling_flags) {
+			if (bo->tiling_flags & RADEON_TILING_MICRO)
 				*array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
-			if ((ws_bo->tiling_flags & (RADEON_TILING_MICRO | RADEON_TILING_MACRO)) ==
+			if ((bo->tiling_flags & (RADEON_TILING_MICRO | RADEON_TILING_MACRO)) ==
 			    (RADEON_TILING_MICRO | RADEON_TILING_MACRO))
 				*array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
 		} else {
 			*array_mode = 0;
 		}
 	}
-	return ws_bo;
+	return bo;
 }
 
 void *r600_bo_map(struct radeon *radeon, struct r600_bo *bo, unsigned usage, void *ctx)
 {
-	return pb_map(bo->pb, usage, ctx);
+	struct pipe_context *pctx = ctx;
+
+	if (usage & PB_USAGE_UNSYNCHRONIZED) {
+		radeon_bo_map(radeon, bo->bo);
+		return bo->bo->data + bo->offset;
+	}
+
+	if (p_atomic_read(&bo->bo->reference.count) > 1) {
+		if (usage & PB_USAGE_DONTBLOCK) {
+			return NULL;
+		}
+		if (ctx) {
+			pctx->flush(pctx, 0, NULL);
+		}
+	}
+
+	if (usage & PB_USAGE_DONTBLOCK) {
+		uint32_t domain;
+
+		if (radeon_bo_busy(radeon, bo->bo, &domain))
+			return NULL;
+		if (radeon_bo_map(radeon, bo->bo)) {
+			return NULL;
+		}
+		goto out;
+	}
+
+	radeon_bo_map(radeon, bo->bo);
+	if (radeon_bo_wait(radeon, bo->bo)) {
+		radeon_bo_unmap(radeon, bo->bo);
+		return NULL;
+	}
+
+out:
+	return bo->bo->data + bo->offset;
 }
 
 void r600_bo_unmap(struct radeon *radeon, struct r600_bo *bo)
 {
-	pb_unmap(bo->pb);
+	radeon_bo_unmap(radeon, bo->bo);
 }
 
-static void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo)
+void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo)
 {
-	if (bo->pb)
-		pb_reference(&bo->pb, NULL);
+	if (bo->manager_id) {
+		if (!r600_bomgr_bo_destroy(radeon->bomgr, bo)) {
+			/* destroy is delayed by buffer manager */
+			return;
+		}
+	}
+	radeon_bo_reference(radeon, &bo->bo, NULL);
 	free(bo);
 }
 
-void r600_bo_reference(struct radeon *radeon, struct r600_bo **dst,
-			    struct r600_bo *src)
+void r600_bo_reference(struct radeon *radeon, struct r600_bo **dst, struct r600_bo *src)
 {
 	struct r600_bo *old = *dst;
- 		
+
 	if (pipe_reference(&(*dst)->reference, &src->reference)) {
 		r600_bo_destroy(radeon, old);
 	}
 	*dst = src;
 }
 
-unsigned r600_bo_get_handle(struct r600_bo *pb_bo)
-{
-	struct radeon_bo *bo;
-
-	bo = radeon_bo_pb_get_bo(pb_bo->pb);
-	if (!bo)
-		return 0;
-
-	return bo->handle;
-}
-
-unsigned r600_bo_get_size(struct r600_bo *pb_bo)
-{
-	struct radeon_bo *bo;
-
-	bo = radeon_bo_pb_get_bo(pb_bo->pb);
-	if (!bo)
-		return 0;
-
-	return bo->size;
-}
-
-boolean r600_bo_get_winsys_handle(struct radeon *radeon, struct r600_bo *pb_bo,
+boolean r600_bo_get_winsys_handle(struct radeon *radeon, struct r600_bo *bo,
 				unsigned stride, struct winsys_handle *whandle)
 {
-	struct radeon_bo *bo;
-
-	bo = radeon_bo_pb_get_bo(pb_bo->pb);
-	if (!bo)
-		return FALSE;
-
 	whandle->stride = stride;
 	switch(whandle->type) {
 	case DRM_API_HANDLE_TYPE_KMS:
-		whandle->handle = r600_bo_get_handle(pb_bo);
+		whandle->handle = r600_bo_get_handle(bo);
 		break;
 	case DRM_API_HANDLE_TYPE_SHARED:
-		if (radeon_bo_get_name(radeon, bo, &whandle->handle))
+		if (radeon_bo_get_name(radeon, bo->bo, &whandle->handle))
 			return FALSE;
 		break;
 	default:
diff --git a/src/gallium/winsys/r600/drm/r600_bomgr.c b/src/gallium/winsys/r600/drm/r600_bomgr.c
new file mode 100644
index 0000000000..446ef0f9cf
--- /dev/null
+++ b/src/gallium/winsys/r600/drm/r600_bomgr.c
@@ -0,0 +1,161 @@
+/*
+ * Copyright 2010 VMWare.
+ * Copyright 2010 Red Hat Inc.
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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:
+ *      Jose Fonseca <jrfonseca-at-vmware-dot-com>
+ *      Thomas Hellström <thomas-at-vmware-dot-com>
+ *      Jerome Glisse <jglisse@redhat.com>
+ */
+#include <util/u_memory.h>
+#include <util/u_double_list.h>
+#include <util/u_time.h>
+#include <pipebuffer/pb_bufmgr.h>
+#include "r600_priv.h"
+
+static void r600_bomgr_timeout_flush(struct r600_bomgr *mgr)
+{
+	struct r600_bo *bo, *tmp;
+	int64_t now;
+
+	now = os_time_get();
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) {
+		if(!os_time_timeout(bo->start, bo->end, now))
+			break;
+
+		mgr->num_delayed--;
+		bo->manager_id = 0;
+		LIST_DEL(&bo->list);
+		r600_bo_destroy(mgr->radeon, bo);
+	}
+}
+
+static INLINE int r600_bo_is_compat(struct r600_bomgr *mgr,
+					struct r600_bo *bo,
+					unsigned size,
+					unsigned alignment,
+					unsigned cfence)
+{
+	if(bo->size < size) {
+		return 0;
+	}
+
+	/* be lenient with size */
+	if(bo->size >= 2*size) {
+		return 0;
+	}
+
+	if(!pb_check_alignment(alignment, bo->alignment)) {
+		return 0;
+	}
+
+	if (!fence_is_after(cfence, bo->fence)) {
+		return 0;
+	}
+
+	return 1;
+}
+
+struct r600_bo *r600_bomgr_bo_create(struct r600_bomgr *mgr,
+					unsigned size,
+					unsigned alignment,
+					unsigned cfence)
+{
+	struct r600_bo *bo, *tmp;
+	int64_t now;
+
+
+	pipe_mutex_lock(mgr->mutex);
+
+	now = os_time_get();
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) {
+		if(r600_bo_is_compat(mgr, bo, size, alignment, cfence)) {
+			LIST_DEL(&bo->list);
+			--mgr->num_delayed;
+			r600_bomgr_timeout_flush(mgr);
+			pipe_mutex_unlock(mgr->mutex);
+			LIST_INITHEAD(&bo->list);
+			pipe_reference_init(&bo->reference, 1);
+			return bo;
+		}
+
+		if(os_time_timeout(bo->start, bo->end, now)) {
+			mgr->num_delayed--;
+			bo->manager_id = 0;
+			LIST_DEL(&bo->list);
+			r600_bo_destroy(mgr->radeon, bo);
+		}
+	}
+
+	pipe_mutex_unlock(mgr->mutex);
+	return NULL;
+}
+
+void r600_bomgr_bo_init(struct r600_bomgr *mgr, struct r600_bo *bo)
+{
+	LIST_INITHEAD(&bo->list);
+	bo->manager_id = 1;
+}
+
+bool r600_bomgr_bo_destroy(struct r600_bomgr *mgr, struct r600_bo *bo)
+{
+	bo->start = os_time_get();
+	bo->end = bo->start + mgr->usecs;
+	pipe_mutex_lock(mgr->mutex);
+	LIST_ADDTAIL(&bo->list, &mgr->delayed);
+	++mgr->num_delayed;
+	pipe_mutex_unlock(mgr->mutex);
+	return FALSE;
+}
+
+void r600_bomgr_destroy(struct r600_bomgr *mgr)
+{
+	struct r600_bo *bo, *tmp;
+
+	pipe_mutex_lock(mgr->mutex);
+	LIST_FOR_EACH_ENTRY_SAFE(bo, tmp, &mgr->delayed, list) {
+		mgr->num_delayed--;
+		bo->manager_id = 0;
+		LIST_DEL(&bo->list);
+		r600_bo_destroy(mgr->radeon, bo);
+	}
+	pipe_mutex_unlock(mgr->mutex);
+
+	FREE(mgr);
+}
+
+struct r600_bomgr *r600_bomgr_create(struct radeon *radeon, unsigned usecs)
+{
+	struct r600_bomgr *mgr;
+
+	mgr = CALLOC_STRUCT(r600_bomgr);
+	if (mgr == NULL)
+		return NULL;
+
+	mgr->radeon = radeon;
+	mgr->usecs = usecs;
+	LIST_INITHEAD(&mgr->delayed);
+	mgr->num_delayed = 0;
+	pipe_mutex_init(mgr->mutex);
+
+	return mgr;
+}
diff --git a/src/gallium/winsys/r600/drm/r600_drm.c b/src/gallium/winsys/r600/drm/r600_drm.c
index 8c847122f8..3cbbf91878 100644
--- a/src/gallium/winsys/r600/drm/r600_drm.c
+++ b/src/gallium/winsys/r600/drm/r600_drm.c
@@ -30,7 +30,6 @@
 #include <sys/ioctl.h>
 #include "util/u_inlines.h"
 #include "util/u_debug.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600.h"
 #include "r600_priv.h"
 #include "r600_drm_public.h"
@@ -230,12 +229,10 @@ static struct radeon *radeon_new(int fd, unsigned device)
 		if (radeon_drm_get_tiling(radeon))
 			return NULL;
 	}
-	radeon->kman = radeon_bo_pbmgr_create(radeon);
-	if (!radeon->kman)
-		return NULL;
-	radeon->cman = pb_cache_manager_create(radeon->kman, 1000000);
-	if (!radeon->cman)
+	radeon->bomgr = r600_bomgr_create(radeon, 1000000);
+	if (radeon->bomgr == NULL) {
 		return NULL;
+	}
 	return radeon;
 }
 
@@ -252,11 +249,8 @@ struct radeon *radeon_decref(struct radeon *radeon)
 		return NULL;
 	}
 
-	if (radeon->cman)
-		radeon->cman->destroy(radeon->cman);
-
-	if (radeon->kman)
-		radeon->kman->destroy(radeon->kman);
+	if (radeon->bomgr)
+		r600_bomgr_destroy(radeon->bomgr);
 
 	if (radeon->fd >= 0)
 		drmClose(radeon->fd);
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 0f2724f61c..d01ec3ee9b 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -28,16 +28,15 @@
 #include <string.h>
 #include <stdlib.h>
 #include <assert.h>
+#include <pipe/p_compiler.h>
+#include <util/u_inlines.h>
+#include <util/u_memory.h>
+#include <pipebuffer/pb_bufmgr.h>
 #include "xf86drm.h"
-#include "r600.h"
-#include "r600d.h"
 #include "radeon_drm.h"
-#include "bof.h"
-#include "pipe/p_compiler.h"
-#include "util/u_inlines.h"
-#include "util/u_memory.h"
-#include <pipebuffer/pb_bufmgr.h>
 #include "r600_priv.h"
+#include "bof.h"
+#include "r600d.h"
 
 #define GROUP_FORCE_NEW_BLOCK	0
 
@@ -50,6 +49,7 @@ int r600_context_init_fence(struct r600_context *ctx)
 	}
 	ctx->cfence = r600_bo_map(ctx->radeon, ctx->fence_bo, PB_USAGE_UNSYNCHRONIZED, NULL);
 	*ctx->cfence = 0;
+	ctx->radeon->cfence = ctx->cfence;
 	LIST_INITHEAD(&ctx->fenced_bo);
 	return 0;
 }
@@ -814,6 +814,7 @@ void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *r
 	ctx->reloc[ctx->creloc].write_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
 	ctx->reloc[ctx->creloc].flags = 0;
 	radeon_bo_reference(ctx->radeon, &ctx->bo[ctx->creloc], bo);
+	rbo->fence = ctx->fence;
 	ctx->creloc++;
 	/* set PKT3 to point to proper reloc */
 	*pm4 = bo->reloc_id;
@@ -836,6 +837,7 @@ void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_stat
 			/* find relocation */
 			id = block->pm4_bo_index[id];
 			r600_bo_reference(ctx->radeon, &block->reloc[id].bo, state->regs[i].bo);
+			state->regs[i].bo->fence = ctx->fence;
 		}
 		if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
 			block->status |= R600_BLOCK_STATUS_ENABLED;
@@ -875,10 +877,13 @@ static inline void r600_context_pipe_state_set_resource(struct r600_context *ctx
 		 */
 		r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[0].bo);
 		r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[0].bo);
+		state->regs[0].bo->fence = ctx->fence;
 	} else {
 		/* TEXTURE RESOURCE */
 		r600_bo_reference(ctx->radeon, &block->reloc[1].bo, state->regs[2].bo);
 		r600_bo_reference(ctx->radeon, &block->reloc[2].bo, state->regs[3].bo);
+		state->regs[2].bo->fence = ctx->fence;
+		state->regs[3].bo->fence = ctx->fence;
 	}
 	if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
 		block->status |= R600_BLOCK_STATUS_ENABLED;
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index 193af984f8..056d0255be 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -30,24 +30,24 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <assert.h>
-#include <pipebuffer/pb_bufmgr.h>
-#include "util/u_double_list.h"
+#include <util/u_double_list.h>
+#include <util/u_inlines.h>
+#include <os/os_thread.h>
 #include "r600.h"
 
+struct r600_bomgr;
+
 struct radeon {
 	int				fd;
 	int				refcount;
 	unsigned			device;
 	unsigned			family;
 	enum chip_class			chip_class;
-	struct pb_manager *kman; /* kernel bo manager */
-	struct pb_manager *cman; /* cached bo manager */
-	struct r600_tiling_info tiling_info;
+	struct r600_tiling_info		tiling_info;
+	struct r600_bomgr		*bomgr;
+	unsigned			*cfence;
 };
 
-struct radeon *r600_new(int fd, unsigned device);
-void r600_delete(struct radeon *r600);
-
 struct r600_reg {
 	unsigned			opcode;
 	unsigned			offset_base;
@@ -75,25 +75,49 @@ struct radeon_bo {
 
 struct r600_bo {
 	struct pipe_reference		reference;
-	struct pb_buffer		*pb;
 	unsigned			size;
 	unsigned			tiling_flags;
 	unsigned			kernel_pitch;
 	unsigned			domains;
+	struct radeon_bo		*bo;
+	unsigned			fence;
+	/* manager data */
+	struct list_head		list;
+	unsigned			manager_id;
+	unsigned			alignment;
+	unsigned			offset;
+	int64_t				start;
+	int64_t				end;
+};
+
+struct r600_bomgr {
+	struct radeon			*radeon;
+	unsigned			usecs;
+	pipe_mutex			mutex;
+	struct list_head		delayed;
+	unsigned			num_delayed;
 };
 
+/*
+ * r600_drm.c
+ */
+struct radeon *r600_new(int fd, unsigned device);
+void r600_delete(struct radeon *r600);
 
-/* radeon_pciid.c */
+/*
+ * radeon_pciid.c
+ */
 unsigned radeon_family_from_device(unsigned device);
 
-/* radeon_bo.c */
+/*
+ * radeon_bo.c
+ */
 struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
 			    unsigned size, unsigned alignment);
 void radeon_bo_reference(struct radeon *radeon, struct radeon_bo **dst,
 			 struct radeon_bo *src);
 int radeon_bo_wait(struct radeon *radeon, struct radeon_bo *bo);
 int radeon_bo_busy(struct radeon *radeon, struct radeon_bo *bo, uint32_t *domain);
-void radeon_bo_pbmgr_flush_maps(struct pb_manager *_mgr);
 int radeon_bo_fencelist(struct radeon *radeon, struct radeon_bo **bolist, uint32_t num_bo);
 int radeon_bo_get_tiling_flags(struct radeon *radeon,
 			       struct radeon_bo *bo,
@@ -103,13 +127,9 @@ int radeon_bo_get_name(struct radeon *radeon,
 		       struct radeon_bo *bo,
 		       uint32_t *name);
 
-/* radeon_bo_pb.c */
-struct radeon_bo *radeon_bo_pb_get_bo(struct pb_buffer *_buf);
-struct pb_manager *radeon_bo_pbmgr_create(struct radeon *radeon);
-struct pb_buffer *radeon_bo_pb_create_buffer_from_handle(struct pb_manager *_mgr,
-							 uint32_t handle);
-
-/* r600_hw_context.c */
+/*
+ * r600_hw_context.c
+ */
 int r600_context_init_fence(struct r600_context *ctx);
 void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *rbo);
 void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
@@ -117,14 +137,27 @@ void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
 struct r600_bo *r600_context_reg_bo(struct r600_context *ctx, unsigned offset);
 int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg, unsigned nreg);
 
-/* r600_bo.c */
-unsigned r600_bo_get_handle(struct r600_bo *bo);
-unsigned r600_bo_get_size(struct r600_bo *bo);
-static INLINE struct radeon_bo *r600_bo_get_bo(struct r600_bo *bo)
-{
-	return radeon_bo_pb_get_bo(bo->pb);
-}
+/*
+ * r600_bo.c
+ */
+void r600_bo_destroy(struct radeon *radeon, struct r600_bo *bo);
 
+/*
+ * r600_bomgr.c
+ */
+struct r600_bomgr *r600_bomgr_create(struct radeon *radeon, unsigned usecs);
+void r600_bomgr_destroy(struct r600_bomgr *mgr);
+bool r600_bomgr_bo_destroy(struct r600_bomgr *mgr, struct r600_bo *bo);
+void r600_bomgr_bo_init(struct r600_bomgr *mgr, struct r600_bo *bo);
+struct r600_bo *r600_bomgr_bo_create(struct r600_bomgr *mgr,
+					unsigned size,
+					unsigned alignment,
+					unsigned cfence);
+
+
+/*
+ * helpers
+ */
 #define CTX_RANGE_ID(ctx, offset) (((offset) >> (ctx)->hash_shift) & 255)
 #define CTX_BLOCK_ID(ctx, offset) ((offset) & ((1 << (ctx)->hash_shift) - 1))
 
@@ -172,6 +205,9 @@ static inline void r600_context_block_emit_dirty(struct r600_context *ctx, struc
 	LIST_DELINIT(&block->list);
 }
 
+/*
+ * radeon_bo.c
+ */
 static inline int radeon_bo_map(struct radeon *radeon, struct radeon_bo *bo)
 {
 	bo->map_count++;
@@ -184,4 +220,35 @@ static inline void radeon_bo_unmap(struct radeon *radeon, struct radeon_bo *bo)
 	assert(bo->map_count >= 0);
 }
 
+/*
+ * r600_bo
+ */
+static inline struct radeon_bo *r600_bo_get_bo(struct r600_bo *bo)
+{
+	return bo->bo;
+}
+
+static unsigned inline r600_bo_get_handle(struct r600_bo *bo)
+{
+	return bo->bo->handle;
+}
+
+static unsigned inline r600_bo_get_size(struct r600_bo *bo)
+{
+	return bo->size;
+}
+
+/*
+ * fence
+ */
+static inline bool fence_is_after(unsigned fence, unsigned ofence)
+{
+	/* handle wrap around */
+	if (fence < 0x80000000 && ofence > 0x80000000)
+		return TRUE;
+	if (fence > ofence)
+		return TRUE;
+	return FALSE;
+}
+
 #endif
diff --git a/src/gallium/winsys/r600/drm/radeon_bo_pb.c b/src/gallium/winsys/r600/drm/radeon_bo_pb.c
deleted file mode 100644
index 4bd3ae3ca1..0000000000
--- a/src/gallium/winsys/r600/drm/radeon_bo_pb.c
+++ /dev/null
@@ -1,260 +0,0 @@
-/*
- * Copyright 2010 Dave Airlie
- *
- * 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
- * on the rights to use, copy, modify, merge, publish, distribute, sub
- * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE AUTHOR(S) AND/OR THEIR 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:
- *      Dave Airlie
- */
-#include <util/u_inlines.h>
-#include <util/u_memory.h>
-#include <util/u_double_list.h>
-#include <pipebuffer/pb_buffer.h>
-#include <pipebuffer/pb_bufmgr.h>
-#include "r600_priv.h"
-
-struct radeon_bo_pb {
-	struct pb_buffer b;
-	struct radeon_bo *bo;
-
-	struct radeon_bo_pbmgr *mgr;
-};
-
-extern const struct pb_vtbl radeon_bo_pb_vtbl;
-
-static INLINE struct radeon_bo_pb *radeon_bo_pb(struct pb_buffer *buf)
-{
-	assert(buf);
-	assert(buf->vtbl == &radeon_bo_pb_vtbl);
-	return (struct radeon_bo_pb *)buf;
-}
-
-struct radeon_bo_pbmgr {
-	struct pb_manager b;
-	struct radeon *radeon;
-};
-
-static INLINE struct radeon_bo_pbmgr *radeon_bo_pbmgr(struct pb_manager *mgr)
-{
-	assert(mgr);
-	return (struct radeon_bo_pbmgr *)mgr;
-}
-
-static void radeon_bo_pb_destroy(struct pb_buffer *_buf)
-{
-	struct radeon_bo_pb *buf = radeon_bo_pb(_buf);
-
-	/* If this buffer is on the list of buffers to unmap,
-	 * do the unmapping now.
-	 */
-	radeon_bo_unmap(buf->mgr->radeon, buf->bo);
-	radeon_bo_reference(buf->mgr->radeon, &buf->bo, NULL);
-	FREE(buf);
-}
-
-static void *
-radeon_bo_pb_map_internal(struct pb_buffer *_buf,
-			  unsigned flags, void *ctx)
-{
-	struct radeon_bo_pb *buf = radeon_bo_pb(_buf);
-	struct pipe_context *pctx = ctx;
-
-	if (flags & PB_USAGE_UNSYNCHRONIZED) {
-		if (radeon_bo_map(buf->mgr->radeon, buf->bo)) {
-			return NULL;
-		}
-		return buf->bo->data;
-	}
-
-	if (p_atomic_read(&buf->bo->reference.count) > 1) {
-		if (flags & PB_USAGE_DONTBLOCK) {
-			return NULL;
-		}
-		if (ctx) {
-			pctx->flush(pctx, 0, NULL);
-		}
-	}
-
-	if (flags & PB_USAGE_DONTBLOCK) {
-		uint32_t domain;
-		if (radeon_bo_busy(buf->mgr->radeon, buf->bo, &domain))
-			return NULL;
-		if (radeon_bo_map(buf->mgr->radeon, buf->bo)) {
-			return NULL;
-		}
-		goto out;
-	}
-
-	if (radeon_bo_map(buf->mgr->radeon, buf->bo)) {
-		return NULL;
-	}
-	if (radeon_bo_wait(buf->mgr->radeon, buf->bo)) {
-		radeon_bo_unmap(buf->mgr->radeon, buf->bo);
-		return NULL;
-	}
-out:
-	return buf->bo->data;
-}
-
-static void radeon_bo_pb_unmap_internal(struct pb_buffer *_buf)
-{
-}
-
-static void
-radeon_bo_pb_get_base_buffer(struct pb_buffer *buf,
-			     struct pb_buffer **base_buf,
-			     unsigned *offset)
-{
-	*base_buf = buf;
-	*offset = 0;
-}
-
-static enum pipe_error
-radeon_bo_pb_validate(struct pb_buffer *_buf, 
-		      struct pb_validate *vl,
-		      unsigned flags)
-{
-	/* Always pinned */
-	return PIPE_OK;
-}
-
-static void
-radeon_bo_pb_fence(struct pb_buffer *buf,
-		   struct pipe_fence_handle *fence)
-{
-}
-
-const struct pb_vtbl radeon_bo_pb_vtbl = {
-    radeon_bo_pb_destroy,
-    radeon_bo_pb_map_internal,
-    radeon_bo_pb_unmap_internal,
-    radeon_bo_pb_validate,
-    radeon_bo_pb_fence,
-    radeon_bo_pb_get_base_buffer,
-};
-
-struct pb_buffer *
-radeon_bo_pb_create_buffer_from_handle(struct pb_manager *_mgr,
-				       uint32_t handle)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	struct radeon *radeon = mgr->radeon;
-	struct radeon_bo_pb *bo;
-	struct radeon_bo *hw_bo;
-
-	hw_bo = radeon_bo(radeon, handle, 0, 0);
-	if (hw_bo == NULL)
-		return NULL;
-
-	bo = CALLOC_STRUCT(radeon_bo_pb);
-	if (!bo) {
-		radeon_bo_reference(radeon, &hw_bo, NULL);
-		return NULL;
-	}
-
-	pipe_reference_init(&bo->b.base.reference, 1);
-	bo->b.base.alignment = 0;
-	bo->b.base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
-	bo->b.base.size = hw_bo->size;
-	bo->b.vtbl = &radeon_bo_pb_vtbl;
-	bo->mgr = mgr;
-
-	bo->bo = hw_bo;
-
-	return &bo->b;
-}
-
-static struct pb_buffer *
-radeon_bo_pb_create_buffer(struct pb_manager *_mgr,
-			   pb_size size,
-			   const struct pb_desc *desc)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	struct radeon *radeon = mgr->radeon;
-	struct radeon_bo_pb *bo;
-
-	bo = CALLOC_STRUCT(radeon_bo_pb);
-	if (!bo)
-		goto error1;
-
-	pipe_reference_init(&bo->b.base.reference, 1);
-	bo->b.base.alignment = desc->alignment;
-	bo->b.base.usage = desc->usage;
-	bo->b.base.size = size;
-	bo->b.vtbl = &radeon_bo_pb_vtbl;
-	bo->mgr = mgr;
-
-	bo->bo = radeon_bo(radeon, 0, size, desc->alignment);
-	if (bo->bo == NULL)
-		goto error2;
-	return &bo->b;
-
-error2:
-	FREE(bo);
-error1:
-	return NULL;
-}
-
-static void
-radeon_bo_pbmgr_flush(struct pb_manager *mgr)
-{
-    /* NOP */
-}
-
-static void
-radeon_bo_pbmgr_destroy(struct pb_manager *_mgr)
-{
-	struct radeon_bo_pbmgr *mgr = radeon_bo_pbmgr(_mgr);
-	FREE(mgr);
-}
-
-struct pb_manager *radeon_bo_pbmgr_create(struct radeon *radeon)
-{
-	struct radeon_bo_pbmgr *mgr;
-
-	mgr = CALLOC_STRUCT(radeon_bo_pbmgr);
-	if (!mgr)
-		return NULL;
-
-	mgr->b.destroy = radeon_bo_pbmgr_destroy;
-	mgr->b.create_buffer = radeon_bo_pb_create_buffer;
-	mgr->b.flush = radeon_bo_pbmgr_flush;
-
-	mgr->radeon = radeon;
-	return &mgr->b;
-}
-
-struct radeon_bo *radeon_bo_pb_get_bo(struct pb_buffer *_buf)
-{
-	struct radeon_bo_pb *buf;
-	if (_buf->vtbl == &radeon_bo_pb_vtbl) {
-		buf = radeon_bo_pb(_buf);
-		return buf->bo;
-	} else {
-		struct pb_buffer *base_buf;
-		pb_size offset;
-		pb_get_base_buffer(_buf, &base_buf, &offset);
-		if (base_buf->vtbl == &radeon_bo_pb_vtbl) {
-			buf = radeon_bo_pb(base_buf);
-			return buf->bo;
-		}
-	}
-	return NULL;
-}