diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index b127c7da192..5f3a9bd5d7b 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -331,9 +331,7 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
 		ctx->b.streamout.suspended = true;
 	}
 
-	/* flush is needed to avoid lockups on some chips with user fences
-	 * this will also flush the framebuffer cache
-	 */
+	/* flush the framebuffer cache */
 	ctx->b.flags |= R600_CONTEXT_FLUSH_AND_INV |
 		      R600_CONTEXT_FLUSH_AND_INV_CB |
 		      R600_CONTEXT_FLUSH_AND_INV_DB |
@@ -434,35 +432,6 @@ void r600_begin_new_cs(struct r600_context *ctx)
 	ctx->initial_gfx_cs_size = ctx->b.rings.gfx.cs->cdw;
 }
 
-void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
-{
-	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
-	uint64_t va;
-
-	r600_need_cs_space(ctx, 10, FALSE);
-
-	va = r600_resource_va(&ctx->screen->b.b, (void*)fence_bo);
-	va = va + (offset << 2);
-
-	/* Use of WAIT_UNTIL is deprecated on Cayman+ */
-	if (ctx->b.family >= CHIP_CAYMAN) {
-		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
-		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
-	} else {
-		r600_write_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
-	}
-
-	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
-	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
-	cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL;       /* ADDRESS_LO */
-	/* DATA_SEL | INT_EN | ADDRESS_HI */
-	cs->buf[cs->cdw++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF);
-	cs->buf[cs->cdw++] = value;                   /* DATA_LO */
-	cs->buf[cs->cdw++] = 0;                       /* DATA_HI */
-	cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
-	cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx, fence_bo, RADEON_USAGE_WRITE);
-}
-
 /* The max number of bytes to copy per packet. */
 #define CP_DMA_MAX_BYTE_COUNT ((1 << 21) - 8)
 
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index e091b084a15..9da42d17d22 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -68,81 +68,6 @@ static const struct debug_named_value r600_debug_options[] = {
 /*
  * pipe_context
  */
-static struct r600_fence *r600_create_fence(struct r600_context *rctx)
-{
-	struct r600_screen *rscreen = rctx->screen;
-	struct r600_fence *fence = NULL;
-
-	pipe_mutex_lock(rscreen->fences.mutex);
-
-	if (!rscreen->fences.bo) {
-		/* Create the shared buffer object */
-		rscreen->fences.bo = (struct r600_resource*)
-			pipe_buffer_create(&rscreen->b.b, PIPE_BIND_CUSTOM,
-					   PIPE_USAGE_STAGING, 4096);
-		if (!rscreen->fences.bo) {
-			R600_ERR("r600: failed to create bo for fence objects\n");
-			goto out;
-		}
-		rscreen->fences.data = r600_buffer_map_sync_with_rings(&rctx->b, rscreen->fences.bo, PIPE_TRANSFER_READ_WRITE);
-	}
-
-	if (!LIST_IS_EMPTY(&rscreen->fences.pool)) {
-		struct r600_fence *entry;
-
-		/* Try to find a freed fence that has been signalled */
-		LIST_FOR_EACH_ENTRY(entry, &rscreen->fences.pool, head) {
-			if (rscreen->fences.data[entry->index] != 0) {
-				LIST_DELINIT(&entry->head);
-				fence = entry;
-				break;
-			}
-		}
-	}
-
-	if (!fence) {
-		/* Allocate a new fence */
-		struct r600_fence_block *block;
-		unsigned index;
-
-		if ((rscreen->fences.next_index + 1) >= 1024) {
-			R600_ERR("r600: too many concurrent fences\n");
-			goto out;
-		}
-
-		index = rscreen->fences.next_index++;
-
-		if (!(index % FENCE_BLOCK_SIZE)) {
-			/* Allocate a new block */
-			block = CALLOC_STRUCT(r600_fence_block);
-			if (block == NULL)
-				goto out;
-
-			LIST_ADD(&block->head, &rscreen->fences.blocks);
-		} else {
-			block = LIST_ENTRY(struct r600_fence_block, rscreen->fences.blocks.next, head);
-		}
-
-		fence = &block->fences[index % FENCE_BLOCK_SIZE];
-		fence->index = index;
-	}
-
-	pipe_reference_init(&fence->reference, 1);
-
-	rscreen->fences.data[fence->index] = 0;
-	r600_context_emit_fence(rctx, rscreen->fences.bo, fence->index, 1);
-
-	/* Create a dummy BO so that fence_finish without a timeout can sleep waiting for completion */
-	fence->sleep_bo = (struct r600_resource*)
-			pipe_buffer_create(&rctx->screen->b.b, PIPE_BIND_CUSTOM,
-					   PIPE_USAGE_STAGING, 1);
-	/* Add the fence as a dummy relocation. */
-	r600_context_bo_reloc(&rctx->b, &rctx->b.rings.gfx, fence->sleep_bo, RADEON_USAGE_READWRITE);
-
-out:
-	pipe_mutex_unlock(rscreen->fences.mutex);
-	return fence;
-}
 
 static void r600_flush(struct pipe_context *ctx, unsigned flags)
 {
@@ -180,12 +105,11 @@ static void r600_flush_from_st(struct pipe_context *ctx,
 			       unsigned flags)
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
-	struct r600_fence **rfence = (struct r600_fence**)fence;
 	unsigned fflags;
 
 	fflags = flags & PIPE_FLUSH_END_OF_FRAME ? RADEON_FLUSH_END_OF_FRAME : 0;
-	if (rfence) {
-		*rfence = r600_create_fence(rctx);
+	if (fence) {
+		*fence = rctx->b.ws->cs_create_fence(rctx->b.rings.gfx.cs);
 	}
 	/* flush gfx & dma ring, order does not matter as only one can be live */
 	if (rctx->b.rings.dma.cs) {
@@ -888,98 +812,15 @@ static void r600_destroy_screen(struct pipe_screen* pscreen)
 		compute_memory_pool_delete(rscreen->global_pool);
 	}
 
-	if (rscreen->fences.bo) {
-		struct r600_fence_block *entry, *tmp;
-
-		LIST_FOR_EACH_ENTRY_SAFE(entry, tmp, &rscreen->fences.blocks, head) {
-			LIST_DEL(&entry->head);
-			FREE(entry);
-		}
-
-		rscreen->b.ws->buffer_unmap(rscreen->fences.bo->cs_buf);
-		pipe_resource_reference((struct pipe_resource**)&rscreen->fences.bo, NULL);
-	}
 	if (rscreen->trace_bo) {
 		rscreen->b.ws->buffer_unmap(rscreen->trace_bo->cs_buf);
 		pipe_resource_reference((struct pipe_resource**)&rscreen->trace_bo, NULL);
 	}
-	pipe_mutex_destroy(rscreen->fences.mutex);
 
 	rscreen->b.ws->destroy(rscreen->b.ws);
 	FREE(rscreen);
 }
 
-static void r600_fence_reference(struct pipe_screen *pscreen,
-                                 struct pipe_fence_handle **ptr,
-                                 struct pipe_fence_handle *fence)
-{
-	struct r600_fence **oldf = (struct r600_fence**)ptr;
-	struct r600_fence *newf = (struct r600_fence*)fence;
-
-	if (pipe_reference(&(*oldf)->reference, &newf->reference)) {
-		struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-		pipe_mutex_lock(rscreen->fences.mutex);
-		pipe_resource_reference((struct pipe_resource**)&(*oldf)->sleep_bo, NULL);
-		LIST_ADDTAIL(&(*oldf)->head, &rscreen->fences.pool);
-		pipe_mutex_unlock(rscreen->fences.mutex);
-	}
-
-	*ptr = fence;
-}
-
-static boolean r600_fence_signalled(struct pipe_screen *pscreen,
-                                    struct pipe_fence_handle *fence)
-{
-	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-	struct r600_fence *rfence = (struct r600_fence*)fence;
-
-	return rscreen->fences.data[rfence->index] != 0;
-}
-
-static boolean r600_fence_finish(struct pipe_screen *pscreen,
-                                 struct pipe_fence_handle *fence,
-                                 uint64_t timeout)
-{
-	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-	struct r600_fence *rfence = (struct r600_fence*)fence;
-	int64_t start_time = 0;
-	unsigned spins = 0;
-
-	if (timeout != PIPE_TIMEOUT_INFINITE) {
-		start_time = os_time_get();
-
-		/* Convert to microseconds. */
-		timeout /= 1000;
-	}
-
-	while (rscreen->fences.data[rfence->index] == 0) {
-		/* Special-case infinite timeout - wait for the dummy BO to become idle */
-		if (timeout == PIPE_TIMEOUT_INFINITE) {
-			rscreen->b.ws->buffer_wait(rfence->sleep_bo->buf, RADEON_USAGE_READWRITE);
-			break;
-		}
-
-		/* The dummy BO will be busy until the CS including the fence has completed, or
-		 * the GPU is reset. Don't bother continuing to spin when the BO is idle. */
-		if (!rscreen->b.ws->buffer_is_busy(rfence->sleep_bo->buf, RADEON_USAGE_READWRITE))
-			break;
-
-		if (++spins % 256)
-			continue;
-#ifdef PIPE_OS_UNIX
-		sched_yield();
-#else
-		os_time_sleep(10);
-#endif
-		if (timeout != PIPE_TIMEOUT_INFINITE &&
-		    os_time_get() - start_time >= timeout) {
-			break;
-		}
-	}
-
-	return rscreen->fences.data[rfence->index] != 0;
-}
-
 static uint64_t r600_get_timestamp(struct pipe_screen *screen)
 {
 	struct r600_screen *rscreen = (struct r600_screen*)screen;
@@ -1035,9 +876,6 @@ struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 	} else {
 		rscreen->b.b.is_format_supported = r600_is_format_supported;
 	}
-	rscreen->b.b.fence_reference = r600_fence_reference;
-	rscreen->b.b.fence_signalled = r600_fence_signalled;
-	rscreen->b.b.fence_finish = r600_fence_finish;
 	rscreen->b.b.get_driver_query_info = r600_get_driver_query_info;
 	if (rscreen->b.info.has_uvd) {
 		rscreen->b.b.get_video_param = ruvd_get_video_param;
@@ -1113,13 +951,6 @@ struct pipe_screen *r600_screen_create(struct radeon_winsys *ws)
 	rscreen->has_cp_dma = rscreen->b.info.drm_minor >= 27 &&
 			      !(rscreen->b.debug_flags & DBG_NO_CP_DMA);
 
-	rscreen->fences.bo = NULL;
-	rscreen->fences.data = NULL;
-	rscreen->fences.next_index = 0;
-	LIST_INITHEAD(&rscreen->fences.pool);
-	LIST_INITHEAD(&rscreen->fences.blocks);
-	pipe_mutex_init(rscreen->fences.mutex);
-
 	rscreen->global_pool = compute_memory_pool_new(rscreen);
 
 	rscreen->cs_count = 0;
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 75a721cb99b..6aa944c54a0 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -187,17 +187,6 @@ struct r600_viewport_state {
 	struct pipe_viewport_state state;
 };
 
-struct r600_pipe_fences {
-	struct r600_resource		*bo;
-	unsigned			*data;
-	unsigned			next_index;
-	/* linked list of preallocated blocks */
-	struct list_head		blocks;
-	/* linked list of freed fences */
-	struct list_head		pool;
-	pipe_mutex			mutex;
-};
-
 /* This must start from 16. */
 /* features */
 #define DBG_NO_LLVM		(1 << 17)
@@ -220,7 +209,6 @@ struct r600_screen {
 	bool				has_msaa;
 	bool				has_cp_dma;
 	bool				has_compressed_msaa_texturing;
-	struct r600_pipe_fences		fences;
 
 	/*for compute global memory binding, we allocate stuff here, instead of
 	 * buffers.
@@ -341,20 +329,6 @@ struct r600_textures_info {
 	uint32_t			*buffer_constants;
 };
 
-struct r600_fence {
-	struct pipe_reference		reference;
-	unsigned			index; /* in the shared bo */
-	struct r600_resource		*sleep_bo;
-	struct list_head		head;
-};
-
-#define FENCE_BLOCK_SIZE 16
-
-struct r600_fence_block {
-	struct r600_fence		fences[FENCE_BLOCK_SIZE];
-	struct list_head		head;
-};
-
 struct r600_constbuf_state
 {
 	struct r600_atom		atom;
@@ -672,8 +646,6 @@ void r600_update_db_shader_control(struct r600_context * rctx);
 void r600_get_backend_mask(struct r600_context *ctx);
 void r600_context_flush(struct r600_context *ctx, unsigned flags);
 void r600_begin_new_cs(struct r600_context *ctx);
-void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence,
-                             unsigned offset, unsigned value);
 void r600_flush_emit(struct r600_context *ctx);
 void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw, boolean count_draw_in);
 void r600_need_dma_space(struct r600_context *ctx, unsigned num_dw);
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c
index f83c7e42a9e..5674f819466 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.c
+++ b/src/gallium/drivers/radeon/r600_pipe_common.c
@@ -47,6 +47,32 @@ static const struct debug_named_value common_debug_options[] = {
 	DEBUG_NAMED_VALUE_END /* must be last */
 };
 
+static void r600_fence_reference(struct pipe_screen *screen,
+				 struct pipe_fence_handle **ptr,
+				 struct pipe_fence_handle *fence)
+{
+	struct radeon_winsys *rws = ((struct r600_common_screen*)screen)->ws;
+
+	rws->fence_reference(ptr, fence);
+}
+
+static boolean r600_fence_signalled(struct pipe_screen *screen,
+				    struct pipe_fence_handle *fence)
+{
+	struct radeon_winsys *rws = ((struct r600_common_screen*)screen)->ws;
+
+	return rws->fence_wait(rws, fence, 0);
+}
+
+static boolean r600_fence_finish(struct pipe_screen *screen,
+				 struct pipe_fence_handle *fence,
+				 uint64_t timeout)
+{
+	struct radeon_winsys *rws = ((struct r600_common_screen*)screen)->ws;
+
+	return rws->fence_wait(rws, fence, timeout);
+}
+
 static bool r600_interpret_tiling(struct r600_common_screen *rscreen,
 				  uint32_t tiling_config)
 {
@@ -164,6 +190,10 @@ bool r600_common_screen_init(struct r600_common_screen *rscreen,
 {
 	ws->query_info(ws, &rscreen->info);
 
+	rscreen->b.fence_finish = r600_fence_finish;
+	rscreen->b.fence_reference = r600_fence_reference;
+	rscreen->b.fence_signalled = r600_fence_signalled;
+
 	rscreen->ws = ws;
 	rscreen->family = rscreen->info.family;
 	rscreen->chip_class = rscreen->info.chip_class;
diff --git a/src/gallium/drivers/radeonsi/r600.h b/src/gallium/drivers/radeonsi/r600.h
index 0a9380bf03f..13bbad4b369 100644
--- a/src/gallium/drivers/radeonsi/r600.h
+++ b/src/gallium/drivers/radeonsi/r600.h
@@ -78,8 +78,6 @@ void r600_context_queries_suspend(struct r600_context *ctx);
 void r600_context_queries_resume(struct r600_context *ctx);
 void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation,
 			    int flag_wait);
-void si_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence,
-                           unsigned offset, unsigned value);
 
 bool si_is_timer_query(unsigned type);
 bool si_query_needs_begin(unsigned type);
diff --git a/src/gallium/drivers/radeonsi/r600_hw_context.c b/src/gallium/drivers/radeonsi/r600_hw_context.c
index b6e7a0f45be..3003dad5e8a 100644
--- a/src/gallium/drivers/radeonsi/r600_hw_context.c
+++ b/src/gallium/drivers/radeonsi/r600_hw_context.c
@@ -161,9 +161,6 @@ void si_need_cs_space(struct r600_context *ctx, unsigned num_dw,
 	/* Count in framebuffer cache flushes at the end of CS. */
 	num_dw += ctx->atoms.cache_flush->num_dw;
 
-	/* Save 16 dwords for the fence mechanism. */
-	num_dw += 16;
-
 #if R600_TRACE_CS
 	if (ctx->screen->trace_bo) {
 		num_dw += R600_TRACE_CS_DWORDS;
@@ -203,7 +200,7 @@ void si_context_flush(struct r600_context *ctx, unsigned flags)
 			R600_CONTEXT_INV_TEX_CACHE;
 	si_emit_cache_flush(&ctx->b, NULL);
 
-	/* partial flush is needed to avoid lockups on some chips with user fences */
+	/* this is probably not needed anymore */
 	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
 	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
 
@@ -279,29 +276,6 @@ void si_begin_new_cs(struct r600_context *ctx)
 	si_all_descriptors_begin_new_cs(ctx);
 }
 
-void si_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
-{
-	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
-	uint64_t va;
-
-	si_need_cs_space(ctx, 10, FALSE);
-
-	va = r600_resource_va(&ctx->screen->b.b, (void*)fence_bo);
-	va = va + (offset << 2);
-
-	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
-	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
-	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
-	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
-	cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL;       /* ADDRESS_LO */
-	/* DATA_SEL | INT_EN | ADDRESS_HI */
-	cs->buf[cs->cdw++] = (1 << 29) | (0 << 24) | ((va >> 32UL) & 0xFF);
-	cs->buf[cs->cdw++] = value;                   /* DATA_LO */
-	cs->buf[cs->cdw++] = 0;                       /* DATA_HI */
-	cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, 0);
-	cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx, fence_bo, RADEON_USAGE_WRITE);
-}
-
 static unsigned r600_query_read_result(char *map, unsigned start_index, unsigned end_index,
 				       bool test_status_bit)
 {
diff --git a/src/gallium/drivers/radeonsi/radeonsi_pipe.c b/src/gallium/drivers/radeonsi/radeonsi_pipe.c
index 121fd186fe7..b2f0866900b 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_pipe.c
+++ b/src/gallium/drivers/radeonsi/radeonsi_pipe.c
@@ -54,95 +54,17 @@
 /*
  * pipe_context
  */
-static struct r600_fence *r600_create_fence(struct r600_context *rctx)
-{
-	struct r600_screen *rscreen = rctx->screen;
-	struct r600_fence *fence = NULL;
-
-	pipe_mutex_lock(rscreen->fences.mutex);
-
-	if (!rscreen->fences.bo) {
-		/* Create the shared buffer object */
-		rscreen->fences.bo = r600_resource_create_custom(&rscreen->b.b,
-							       PIPE_USAGE_STAGING,
-							       4096);
-		if (!rscreen->fences.bo) {
-			R600_ERR("r600: failed to create bo for fence objects\n");
-			goto out;
-		}
-		rscreen->fences.data = rctx->b.ws->buffer_map(rscreen->fences.bo->cs_buf,
-							   rctx->b.rings.gfx.cs,
-							   PIPE_TRANSFER_READ_WRITE);
-	}
-
-	if (!LIST_IS_EMPTY(&rscreen->fences.pool)) {
-		struct r600_fence *entry;
-
-		/* Try to find a freed fence that has been signalled */
-		LIST_FOR_EACH_ENTRY(entry, &rscreen->fences.pool, head) {
-			if (rscreen->fences.data[entry->index] != 0) {
-				LIST_DELINIT(&entry->head);
-				fence = entry;
-				break;
-			}
-		}
-	}
-
-	if (!fence) {
-		/* Allocate a new fence */
-		struct r600_fence_block *block;
-		unsigned index;
-
-		if ((rscreen->fences.next_index + 1) >= 1024) {
-			R600_ERR("r600: too many concurrent fences\n");
-			goto out;
-		}
-
-		index = rscreen->fences.next_index++;
-
-		if (!(index % FENCE_BLOCK_SIZE)) {
-			/* Allocate a new block */
-			block = CALLOC_STRUCT(r600_fence_block);
-			if (block == NULL)
-				goto out;
-
-			LIST_ADD(&block->head, &rscreen->fences.blocks);
-		} else {
-			block = LIST_ENTRY(struct r600_fence_block, rscreen->fences.blocks.next, head);
-		}
-
-		fence = &block->fences[index % FENCE_BLOCK_SIZE];
-		fence->index = index;
-	}
-
-	pipe_reference_init(&fence->reference, 1);
-
-	rscreen->fences.data[fence->index] = 0;
-	si_context_emit_fence(rctx, rscreen->fences.bo, fence->index, 1);
-
-	/* Create a dummy BO so that fence_finish without a timeout can sleep waiting for completion */
-	fence->sleep_bo = r600_resource_create_custom(&rctx->screen->b.b, PIPE_USAGE_STAGING, 1);
-
-	/* Add the fence as a dummy relocation. */
-	r600_context_bo_reloc(&rctx->b, &rctx->b.rings.gfx, fence->sleep_bo, RADEON_USAGE_READWRITE);
-
-out:
-	pipe_mutex_unlock(rscreen->fences.mutex);
-	return fence;
-}
-
-
 void radeonsi_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
 		    unsigned flags)
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
-	struct r600_fence **rfence = (struct r600_fence**)fence;
 	struct pipe_query *render_cond = NULL;
 	boolean render_cond_cond = FALSE;
 	unsigned render_cond_mode = 0;
 
-	if (rfence)
-		*rfence = r600_create_fence(rctx);
+	if (fence) {
+		*fence = rctx->b.ws->cs_create_fence(rctx->b.rings.gfx.cs);
+	}
 
 	/* Disable render condition. */
 	if (rctx->current_render_cond) {
@@ -658,18 +580,6 @@ static void r600_destroy_screen(struct pipe_screen* pscreen)
 
 	r600_common_screen_cleanup(&rscreen->b);
 
-	if (rscreen->fences.bo) {
-		struct r600_fence_block *entry, *tmp;
-
-		LIST_FOR_EACH_ENTRY_SAFE(entry, tmp, &rscreen->fences.blocks, head) {
-			LIST_DEL(&entry->head);
-			FREE(entry);
-		}
-
-		rscreen->b.ws->buffer_unmap(rscreen->fences.bo->cs_buf);
-		r600_resource_reference(&rscreen->fences.bo, NULL);
-	}
-
 #if R600_TRACE_CS
 	if (rscreen->trace_bo) {
 		rscreen->ws->buffer_unmap(rscreen->trace_bo->cs_buf);
@@ -677,83 +587,10 @@ static void r600_destroy_screen(struct pipe_screen* pscreen)
 	}
 #endif
 
-	pipe_mutex_destroy(rscreen->fences.mutex);
-
 	rscreen->b.ws->destroy(rscreen->b.ws);
 	FREE(rscreen);
 }
 
-static void r600_fence_reference(struct pipe_screen *pscreen,
-                                 struct pipe_fence_handle **ptr,
-                                 struct pipe_fence_handle *fence)
-{
-	struct r600_fence **oldf = (struct r600_fence**)ptr;
-	struct r600_fence *newf = (struct r600_fence*)fence;
-
-	if (pipe_reference(&(*oldf)->reference, &newf->reference)) {
-		struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-		pipe_mutex_lock(rscreen->fences.mutex);
-		r600_resource_reference(&(*oldf)->sleep_bo, NULL);
-		LIST_ADDTAIL(&(*oldf)->head, &rscreen->fences.pool);
-		pipe_mutex_unlock(rscreen->fences.mutex);
-	}
-
-	*ptr = fence;
-}
-
-static boolean r600_fence_signalled(struct pipe_screen *pscreen,
-                                    struct pipe_fence_handle *fence)
-{
-	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-	struct r600_fence *rfence = (struct r600_fence*)fence;
-
-	return rscreen->fences.data[rfence->index] != 0;
-}
-
-static boolean r600_fence_finish(struct pipe_screen *pscreen,
-                                 struct pipe_fence_handle *fence,
-                                 uint64_t timeout)
-{
-	struct r600_screen *rscreen = (struct r600_screen *)pscreen;
-	struct r600_fence *rfence = (struct r600_fence*)fence;
-	int64_t start_time = 0;
-	unsigned spins = 0;
-
-	if (timeout != PIPE_TIMEOUT_INFINITE) {
-		start_time = os_time_get();
-
-		/* Convert to microseconds. */
-		timeout /= 1000;
-	}
-
-	while (rscreen->fences.data[rfence->index] == 0) {
-		/* Special-case infinite timeout - wait for the dummy BO to become idle */
-		if (timeout == PIPE_TIMEOUT_INFINITE) {
-			rscreen->b.ws->buffer_wait(rfence->sleep_bo->buf, RADEON_USAGE_READWRITE);
-			break;
-		}
-
-		/* The dummy BO will be busy until the CS including the fence has completed, or
-		 * the GPU is reset. Don't bother continuing to spin when the BO is idle. */
-		if (!rscreen->b.ws->buffer_is_busy(rfence->sleep_bo->buf, RADEON_USAGE_READWRITE))
-			break;
-
-		if (++spins % 256)
-			continue;
-#ifdef PIPE_OS_UNIX
-		sched_yield();
-#else
-		os_time_sleep(10);
-#endif
-		if (timeout != PIPE_TIMEOUT_INFINITE &&
-		    os_time_get() - start_time >= timeout) {
-			break;
-		}
-	}
-
-	return rscreen->fences.data[rfence->index] != 0;
-}
-
 static uint64_t r600_get_timestamp(struct pipe_screen *screen)
 {
 	struct r600_screen *rscreen = (struct r600_screen*)screen;
@@ -782,9 +619,6 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 	rscreen->b.b.get_compute_param = r600_get_compute_param;
 	rscreen->b.b.get_timestamp = r600_get_timestamp;
 	rscreen->b.b.is_format_supported = si_is_format_supported;
-	rscreen->b.b.fence_reference = r600_fence_reference;
-	rscreen->b.b.fence_signalled = r600_fence_signalled;
-	rscreen->b.b.fence_finish = r600_fence_finish;
 	if (rscreen->b.info.has_uvd) {
 		rscreen->b.b.get_video_param = ruvd_get_video_param;
 		rscreen->b.b.is_video_format_supported = ruvd_is_format_supported;
@@ -802,13 +636,6 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 	if (debug_get_bool_option("RADEON_DUMP_SHADERS", FALSE))
 		rscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
 
-	rscreen->fences.bo = NULL;
-	rscreen->fences.data = NULL;
-	rscreen->fences.next_index = 0;
-	LIST_INITHEAD(&rscreen->fences.pool);
-	LIST_INITHEAD(&rscreen->fences.blocks);
-	pipe_mutex_init(rscreen->fences.mutex);
-
 #if R600_TRACE_CS
 	rscreen->cs_count = 0;
 	if (rscreen->info.drm_minor >= 28) {
diff --git a/src/gallium/drivers/radeonsi/radeonsi_pipe.h b/src/gallium/drivers/radeonsi/radeonsi_pipe.h
index 26f7e09aebc..1d4a91bf8ad 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_pipe.h
+++ b/src/gallium/drivers/radeonsi/radeonsi_pipe.h
@@ -53,20 +53,8 @@
 
 struct si_pipe_compute;
 
-struct r600_pipe_fences {
-	struct r600_resource		*bo;
-	unsigned			*data;
-	unsigned			next_index;
-	/* linked list of preallocated blocks */
-	struct list_head		blocks;
-	/* linked list of freed fences */
-	struct list_head		pool;
-	pipe_mutex			mutex;
-};
-
 struct r600_screen {
 	struct r600_common_screen	b;
-	struct r600_pipe_fences		fences;
 #if R600_TRACE_CS
 	struct r600_resource		*trace_bo;
 	uint32_t			*trace_ptr;
@@ -99,20 +87,6 @@ struct r600_textures_info {
 	unsigned			n_samplers;
 };
 
-struct r600_fence {
-	struct pipe_reference		reference;
-	unsigned			index; /* in the shared bo */
-	struct r600_resource            *sleep_bo;
-	struct list_head		head;
-};
-
-#define FENCE_BLOCK_SIZE 16
-
-struct r600_fence_block {
-	struct r600_fence		fences[FENCE_BLOCK_SIZE];
-	struct list_head		head;
-};
-
 #define SI_NUM_ATOMS(rctx) (sizeof((rctx)->atoms)/sizeof((rctx)->atoms.array[0]))
 #define SI_NUM_SHADERS (PIPE_SHADER_FRAGMENT+1)