radeon/r200/r300/r600: make bo mapping be explicit

This moves the bo mapping outside the DMA layer and makes it explicit,
this should in theory make it simpler to split the clean up the dma/cmdbuf
linkage that I created before that is broken.

Tested on: r600, rv380 (tcl/no-tcl), rv200 (tcl/no-tcl)

Signed-off-by: Dave Airlie <airlied@redhat.com>

10 files changed, 60 insertions, 23 deletions

diff --git a/src/mesa/drivers/dri/r200/r200_maos_arrays.c b/src/mesa/drivers/dri/r200/r200_maos_arrays.c
index 383a0c4b0d..249c0bbc11 100644
--- a/src/mesa/drivers/dri/r200/r200_maos_arrays.c
+++ b/src/mesa/drivers/dri/r200/r200_maos_arrays.c
@@ -90,12 +90,14 @@ static void r200_emit_vecfog(GLcontext *ctx, struct radeon_aos *aos,
 	aos->components = size;
 	aos->count = count;
 
+	radeon_bo_map(aos->bo, 1);
 	out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
 	for (i = 0; i < count; i++) {
 	  out[0] = r200ComputeFogBlendFactor( ctx, *(GLfloat *)data );
 	  out++;
 	  data += stride;
 	}
+	radeon_bo_unmap(aos->bo);
 }
 
 /* Emit any changed arrays to new GART memory, re-emit a packet to
diff --git a/src/mesa/drivers/dri/r200/r200_swtcl.c b/src/mesa/drivers/dri/r200/r200_swtcl.c
index fadc766b49..4596912ddc 100644
--- a/src/mesa/drivers/dri/r200/r200_swtcl.c
+++ b/src/mesa/drivers/dri/r200/r200_swtcl.c
@@ -297,7 +297,7 @@ void r200_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
    radeonEmitState(&rmesa->radeon);
    r200EmitVertexAOS( rmesa,
 		      rmesa->radeon.swtcl.vertex_size,
-		      first_elem(&rmesa->radeon.dma.reserved)->bo,
+		      rmesa->radeon.swtcl.bo,
 		      current_offset);
 
 
diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
index e9968f9ffe..3dcd986e22 100644
--- a/src/mesa/drivers/dri/r300/r300_draw.c
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -100,7 +100,7 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 		GLubyte *in = (GLubyte *)src_ptr;
 
 		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4);
-
+		radeon_bo_map(r300->ind_buf.bo, 1);
 		assert(r300->ind_buf.bo->ptr != NULL);
 		out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 
@@ -111,7 +111,7 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 		if (i < mesa_ind_buf->count) {
 			*out++ = in[i];
 		}
-
+		radeon_bo_unmap(r300->ind_buf.bo);
 #if MESA_BIG_ENDIAN
 	} else { /* if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) */
 		GLushort *in = (GLushort *)src_ptr;
@@ -120,6 +120,7 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo,
 				     &r300->ind_buf.bo_offset, size, 4);
 
+		radeon_bo_map(r300->ind_buf.bo, 1);
 		assert(r300->ind_buf.bo->ptr != NULL);
 		out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 
@@ -130,6 +131,7 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 		if (i < mesa_ind_buf->count) {
 			*out++ = in[i];
 		}
+		radeon_bo_unmap(r300->ind_buf.bo);
 #endif
 	}
 
@@ -173,10 +175,12 @@ static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 
 		radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4);
 
+		radeon_bo_map(r300->ind_buf.bo, 1);
 		assert(r300->ind_buf.bo->ptr != NULL);
 		dst_ptr = ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset);
 		_mesa_memcpy(dst_ptr, src_ptr, size);
 
+		radeon_bo_unmap(r300->ind_buf.bo);
 		r300->ind_buf.is_32bit = (mesa_ind_buf->type == GL_UNSIGNED_INT);
 		r300->ind_buf.count = mesa_ind_buf->count;
 
@@ -242,6 +246,7 @@ static void r300ConvertAttrib(GLcontext *ctx, int count, const struct gl_client_
 	}
 
 	radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, sizeof(GLfloat) * input->Size * count, 32);
+	radeon_bo_map(attr->bo, 1);
 	dst_ptr = (GLfloat *)ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
 
 	radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
@@ -280,6 +285,7 @@ static void r300ConvertAttrib(GLcontext *ctx, int count, const struct gl_client_
 			break;
 	}
 
+	radeon_bo_unmap(attr->bo);
 	if (mapped_named_bo) {
 		ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj);
 	}
@@ -294,6 +300,8 @@ static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *i
 
 	radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, size, 32);
 
+	radeon_bo_map(attr->bo, 1);
+
 	if (!input->BufferObj->Pointer) {
 		ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj);
 		mapped_named_bo = GL_TRUE;
@@ -317,6 +325,7 @@ static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *i
 		ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj);
 	}
 
+	radeon_bo_unmap(attr->bo);
 	attr->stride = dst_stride;
 }
 
@@ -527,6 +536,7 @@ static void r300AllocDmaRegions(GLcontext *ctx, const struct gl_client_array *in
 				}
 
 				radeonAllocDmaRegion(&r300->radeon, &vbuf->attribs[index].bo, &vbuf->attribs[index].bo_offset, size, 32);
+				radeon_bo_map(vbuf->attribs[index].bo, 1);
 				assert(vbuf->attribs[index].bo->ptr != NULL);
 				dst = (uint32_t *)ADD_POINTERS(vbuf->attribs[index].bo->ptr, vbuf->attribs[index].bo_offset);
 				switch (vbuf->attribs[index].dwords) {
@@ -536,6 +546,7 @@ static void r300AllocDmaRegions(GLcontext *ctx, const struct gl_client_array *in
 					case 4: radeonEmitVec16(dst, input[i]->Ptr, input[i]->StrideB, local_count); break;
 					default: assert(0); break;
 				}
+				radeon_bo_unmap(vbuf->attribs[index].bo);
 
 			}
 		}
diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c
index 99bd22edac..383c8a274b 100644
--- a/src/mesa/drivers/dri/r300/r300_swtcl.c
+++ b/src/mesa/drivers/dri/r300/r300_swtcl.c
@@ -665,11 +665,11 @@ void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
 	r300EmitCacheFlush(rmesa);
 
 	radeonEmitState(&rmesa->radeon);
-    r300_emit_scissor(ctx);
+	r300_emit_scissor(ctx);
 	r300EmitVertexAOS(rmesa,
-			rmesa->radeon.swtcl.vertex_size,
-			first_elem(&rmesa->radeon.dma.reserved)->bo,
-			current_offset);
+			  rmesa->radeon.swtcl.vertex_size,
+			  rmesa->radeon.swtcl.bo,
+			  current_offset);
 
 	r300EmitVbufPrim(rmesa,
 		   rmesa->radeon.swtcl.hw_primitive,
diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c
index 47f89c91f8..eab27cbd84 100644
--- a/src/mesa/drivers/dri/r600/r700_render.c
+++ b/src/mesa/drivers/dri/r600/r700_render.c
@@ -526,6 +526,9 @@ static void r700ConvertAttrib(GLcontext *ctx, int count,
 
     radeonAllocDmaRegion(&context->radeon, &attr->bo, &attr->bo_offset, 
                          sizeof(GLfloat) * input->Size * count, 32);
+
+    radeon_bo_map(attr->bo, 1);
+
     dst_ptr = (GLfloat *)ADD_POINTERS(attr->bo->ptr, attr->bo_offset);
 
     assert(src_ptr != NULL);
@@ -559,6 +562,8 @@ static void r700ConvertAttrib(GLcontext *ctx, int count,
             break;
     }
 
+    radeon_bo_unmap(attr->bo);
+
     if (mapped_named_bo) 
     {
         ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj);
@@ -577,6 +582,8 @@ static void r700AlignDataToDword(GLcontext *ctx,
 
     radeonAllocDmaRegion(&context->radeon, &attr->bo, &attr->bo_offset, size, 32);
 
+    radeon_bo_map(attr->bo, 1);
+
     if (!input->BufferObj->Pointer) 
     {
         ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj);
@@ -596,6 +603,7 @@ static void r700AlignDataToDword(GLcontext *ctx,
         }
     }
 
+    radeon_bo_unmap(attr->bo);
     if (mapped_named_bo) 
     {
         ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj);
@@ -664,14 +672,18 @@ static void r700SetupStreams(GLcontext *ctx, const struct gl_client_array *input
 
                 radeonAllocDmaRegion(&context->radeon, &context->stream_desc[index].bo, 
                                      &context->stream_desc[index].bo_offset, size, 32);
+
+                radeon_bo_map(context->stream_desc[index].bo, 1);
                 assert(context->stream_desc[index].bo->ptr != NULL);
+
+
                 dst = (uint32_t *)ADD_POINTERS(context->stream_desc[index].bo->ptr, 
                                                context->stream_desc[index].bo_offset);
 
                 switch (context->stream_desc[index].dwords) 
                 {
                 case 1:                     
-                    radeonEmitVec4(dst, input[i]->Ptr, input[i]->StrideB, local_count);                         
+                    radeonEmitVec4(dst, input[i]->Ptr, input[i]->StrideB, local_count);
                     break;
                 case 2: 
                     radeonEmitVec8(dst, input[i]->Ptr, input[i]->StrideB, local_count); 
@@ -686,6 +698,7 @@ static void r700SetupStreams(GLcontext *ctx, const struct gl_client_array *input
                     assert(0); 
                     break;
                 }
+		radeon_bo_unmap(context->stream_desc[index].bo);
             }
         }
 
@@ -757,6 +770,7 @@ static void r700FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 	radeonAllocDmaRegion(&context->radeon, &context->ind_buf.bo,
 			     &context->ind_buf.bo_offset, size, 4);
 
+	radeon_bo_map(context->ind_buf.bo, 1);
 	assert(context->ind_buf.bo->ptr != NULL);
 	out = (GLuint *)ADD_POINTERS(context->ind_buf.bo->ptr, context->ind_buf.bo_offset);
 
@@ -770,6 +784,7 @@ static void r700FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
             *out++ = in[i];
         }
 
+	radeon_bo_unmap(context->ind_buf.bo);
 #if MESA_BIG_ENDIAN
     }
     else
@@ -780,6 +795,7 @@ static void r700FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 	radeonAllocDmaRegion(&context->radeon, &context->ind_buf.bo,
 			     &context->ind_buf.bo_offset, size, 4);
 
+	radeon_bo_map(context->ind_buf.bo, 1);
 	assert(context->ind_buf.bo->ptr != NULL);
 	out = (GLuint *)ADD_POINTERS(context->ind_buf.bo->ptr, context->ind_buf.bo_offset);
 
@@ -792,6 +808,7 @@ static void r700FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
         {
             *out++ = in[i];
         }
+	radeon_bo_unmap(context->ind_buf.bo);
 #endif
     }
 
@@ -837,11 +854,13 @@ static void r700SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 
 	radeonAllocDmaRegion(&context->radeon, &context->ind_buf.bo,
 			     &context->ind_buf.bo_offset, size, 4);
+	radeon_bo_map(context->ind_buf.bo, 1);
 	assert(context->ind_buf.bo->ptr != NULL);
 	dst_ptr = ADD_POINTERS(context->ind_buf.bo->ptr, context->ind_buf.bo_offset);
 
         _mesa_memcpy(dst_ptr, src_ptr, size);
 
+	radeon_bo_unmap(context->ind_buf.bo);
         context->ind_buf.is_32bit = (mesa_ind_buf->type == GL_UNSIGNED_INT);
         context->ind_buf.count = mesa_ind_buf->count;
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.h b/src/mesa/drivers/dri/radeon/radeon_common_context.h
index ded81fff29..ad953ddbb5 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common_context.h
+++ b/src/mesa/drivers/dri/radeon/radeon_common_context.h
@@ -328,6 +328,7 @@ struct radeon_swtcl_info {
 	GLuint vertex_attr_count;
 
 	GLuint emit_prediction;
+        struct radeon_bo *bo;
 };
 
 #define RADEON_MAX_AOS_ARRAYS		16
diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c
index c6edbae9a1..2a1bd7357a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_dma.c
+++ b/src/mesa/drivers/dri/radeon/radeon_dma.c
@@ -151,6 +151,7 @@ void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 	aos->components = size;
 	aos->count = count;
 
+	radeon_bo_map(aos->bo, 1);
 	out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
 	switch (size) {
 	case 1: radeonEmitVec4(out, data, stride, count); break;
@@ -161,6 +162,7 @@ void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos,
 		assert(0);
 		break;
 	}
+	radeon_bo_unmap(aos->bo);
 }
 
 void radeon_init_dma(radeonContextPtr rmesa)
@@ -183,10 +185,6 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size)
 			__FUNCTION__, size, rmesa->dma.minimum_size);
 
 
-	/* unmap old reserved bo */
-	if (!is_empty_list(&rmesa->dma.reserved))
-		radeon_bo_unmap(first_elem(&rmesa->dma.reserved)->bo);
-
 	if (is_empty_list(&rmesa->dma.free)
 	      || last_elem(&rmesa->dma.free)->bo->size < size) {
 		dma_bo = CALLOC_STRUCT(radeon_dma_bo);
@@ -223,8 +221,6 @@ again_alloc:
         /* Cmd buff have been flushed in radeon_revalidate_bos */
 		goto again_alloc;
 	}
-
-	radeon_bo_map(first_elem(&rmesa->dma.reserved)->bo, 1);
 }
 
 /* Allocates a region from rmesa->dma.current.  If there isn't enough
@@ -281,7 +277,6 @@ void radeonFreeDmaRegions(radeonContextPtr rmesa)
 
 	foreach_s(dma_bo, temp, &rmesa->dma.reserved) {
 		remove_from_list(dma_bo);
-		radeon_bo_unmap(dma_bo->bo);
 	        radeon_bo_unref(dma_bo->bo);
 		FREE(dma_bo);
 	}
@@ -367,9 +362,6 @@ void radeonReleaseDmaRegions(radeonContextPtr rmesa)
 		insert_at_tail(&rmesa->dma.free, dma_bo);
 	}
 
-	/* unmap the last dma region */
-	if (!is_empty_list(&rmesa->dma.reserved))
-		radeon_bo_unmap(first_elem(&rmesa->dma.reserved)->bo);
 	/* move reserved to wait list */
 	foreach_s(dma_bo, temp, &rmesa->dma.reserved) {
 		/* free objects that are too small to be used because of large request */
@@ -403,11 +395,12 @@ void rcommon_flush_last_swtcl_prim( GLcontext *ctx  )
 	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 	struct radeon_dma *dma = &rmesa->dma;
 		
-
 	if (RADEON_DEBUG & RADEON_IOCTL)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 	dma->flush = NULL;
 
+	radeon_bo_unmap(rmesa->swtcl.bo);
+
 	if (!is_empty_list(&dma->reserved)) {
 	    GLuint current_offset = dma->current_used;
 
@@ -422,6 +415,8 @@ void rcommon_flush_last_swtcl_prim( GLcontext *ctx  )
 	    }
 	    rmesa->swtcl.numverts = 0;
 	}
+	radeon_bo_unref(rmesa->swtcl.bo);
+	rmesa->swtcl.bo = NULL;
 }
 /* Alloc space in the current dma region.
  */
@@ -432,6 +427,7 @@ rcommonAllocDmaLowVerts( radeonContextPtr rmesa, int nverts, int vsize )
 	void *head;
 	if (RADEON_DEBUG & RADEON_IOCTL)
 		fprintf(stderr, "%s\n", __FUNCTION__);
+
 	if(is_empty_list(&rmesa->dma.reserved)
 	      ||rmesa->dma.current_vertexptr + bytes > first_elem(&rmesa->dma.reserved)->bo->size) {
 		if (rmesa->dma.flush) {
@@ -455,7 +451,11 @@ rcommonAllocDmaLowVerts( radeonContextPtr rmesa, int nverts, int vsize )
                 rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
                 rmesa->dma.current_vertexptr );
 
-	head = (first_elem(&rmesa->dma.reserved)->bo->ptr + rmesa->dma.current_vertexptr);
+	rmesa->swtcl.bo = first_elem(&rmesa->dma.reserved)->bo;
+	radeon_bo_ref(rmesa->swtcl.bo);
+	radeon_bo_map(rmesa->swtcl.bo, 1);
+
+	head = (rmesa->swtcl.bo->ptr + rmesa->dma.current_vertexptr);
 	rmesa->dma.current_vertexptr += bytes;
 	rmesa->swtcl.numverts += nverts;
 	return head;
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
index de18d2ddd6..d810e6080e 100644
--- a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
@@ -76,12 +76,14 @@ static void emit_vecfog(GLcontext *ctx, struct radeon_aos *aos,
 
    /* Emit the data
     */
+   radeon_bo_map(aos->bo, 1);
    out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
    for (i = 0; i < count; i++) {
       out[0] = radeonComputeFogBlendFactor( ctx, *(GLfloat *)data );
       out++;
       data += stride;
    }
+   radeon_bo_unmap(aos->bo);
 }
 
 static void emit_s0_vec(uint32_t *out, GLvoid *data, int stride, int count)
@@ -151,6 +153,7 @@ static void emit_tex_vector(GLcontext *ctx, struct radeon_aos *aos,
 
    /* Emit the data
     */
+   radeon_bo_map(aos->bo, 1);
    out = (uint32_t*)((char*)aos->bo->ptr + aos->offset);
    switch (size) {
    case 1:
@@ -170,6 +173,7 @@ static void emit_tex_vector(GLcontext *ctx, struct radeon_aos *aos,
       exit(1);
       break;
    }
+   radeon_bo_unmap(aos->bo);
 }
 
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
index 5ed11d0a9d..98f96ff2a7 100644
--- a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
@@ -420,10 +420,10 @@ void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
    }
 
 
-
+   radeon_bo_map(rmesa->radeon.tcl.aos[0].bo, 1);
    setup_tab[i].emit( ctx, 0, VB->Count, 
 		      rmesa->radeon.tcl.aos[0].bo->ptr + rmesa->radeon.tcl.aos[0].offset);
-
+   radeon_bo_unmap(rmesa->radeon.tcl.aos[0].bo);
    //   rmesa->radeon.tcl.aos[0].size = setup_tab[i].vertex_size;
    rmesa->radeon.tcl.aos[0].stride = setup_tab[i].vertex_size;
    rmesa->tcl.vertex_format = setup_tab[i].vertex_format;
diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.c b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
index 6bbe8e252e..8bf1bfbc57 100644
--- a/src/mesa/drivers/dri/radeon/radeon_swtcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
@@ -309,7 +309,7 @@ void r100_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
    radeonEmitState(&rmesa->radeon);
    radeonEmitVertexAOS( rmesa,
 			rmesa->radeon.swtcl.vertex_size,
-			first_elem(&rmesa->radeon.dma.reserved)->bo,
+			rmesa->radeon.swtcl.bo,
 			current_offset);