i965: Move front buffer rendering fields from intel_context to brw.
[mesa.git] / src / mesa / drivers / dri / i965 / intel_context.c
index 7ec316aa8ae88c1186a3a6de5e60b97fd4688313..8a8d2e2e4389a45a43da363b299313baec31245f 100644 (file)
  **************************************************************************/
 
 
-#include "glheader.h"
-#include "context.h"
-#include "matrix.h"
-#include "simple_list.h"
-#include "extensions.h"
-#include "framebuffer.h"
-#include "imports.h"
-#include "points.h"
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/extensions.h"
+#include "main/fbobject.h"
+#include "main/framebuffer.h"
+#include "main/imports.h"
+#include "main/renderbuffer.h"
 
 #include "swrast/swrast.h"
 #include "swrast_setup/swrast_setup.h"
 #include "tnl/tnl.h"
-#include "vbo/vbo.h"
-
-#include "tnl/t_pipeline.h"
-#include "tnl/t_vertex.h"
-
 #include "drivers/common/driverfuncs.h"
+#include "drivers/common/meta.h"
 
-#include "intel_screen.h"
 #include "intel_chipset.h"
-
-#include "i830_dri.h"
-#include "i830_common.h"
-
+#include "intel_buffers.h"
 #include "intel_tex.h"
-#include "intel_span.h"
-#include "intel_ioctl.h"
 #include "intel_batchbuffer.h"
-#include "intel_blit.h"
+#include "intel_pixel.h"
 #include "intel_regions.h"
 #include "intel_buffer_objects.h"
-
-#include "bufmgr.h"
+#include "intel_fbo.h"
+#include "intel_bufmgr.h"
+#include "intel_screen.h"
+#include "intel_mipmap_tree.h"
 
 #include "utils.h"
-#include "vblank.h"
+#include "../glsl/ralloc.h"
+
 #ifndef INTEL_DEBUG
 int INTEL_DEBUG = (0);
 #endif
 
-#define need_GL_NV_point_sprite
-#define need_GL_ARB_multisample
-#define need_GL_ARB_point_parameters
-#define need_GL_ARB_texture_compression
-#define need_GL_ARB_vertex_buffer_object
-#define need_GL_ARB_vertex_program
-#define need_GL_ARB_window_pos
-#define need_GL_ARB_occlusion_query
-#define need_GL_EXT_blend_color
-#define need_GL_EXT_blend_equation_separate
-#define need_GL_EXT_blend_func_separate
-#define need_GL_EXT_blend_minmax
-#define need_GL_EXT_cull_vertex
-#define need_GL_EXT_fog_coord
-#define need_GL_EXT_multi_draw_arrays
-#define need_GL_EXT_secondary_color
-#define need_GL_EXT_point_parameters
-#include "extension_helper.h"
-
-#ifndef VERBOSE
-int VERBOSE = 0;
-#endif
-
-/***************************************
- * Mesa's Driver Functions
- ***************************************/
-
-#define DRIVER_VERSION                     "4.1.3002"
 
-static const GLubyte *intelGetString( GLcontext *ctx, GLenum name )
+static const GLubyte *
+intelGetString(struct gl_context * ctx, GLenum name)
 {
-   const char * chipset;
+   const struct intel_context *const intel = intel_context(ctx);
+   const char *chipset;
    static char buffer[128];
 
    switch (name) {
    case GL_VENDOR:
-      return (GLubyte *)"Tungsten Graphics, Inc";
+      return (GLubyte *) "Intel Open Source Technology Center";
       break;
-      
+
    case GL_RENDERER:
-      switch (intel_context(ctx)->intelScreen->deviceID) {
-      case PCI_CHIP_I965_Q:
-        chipset = "Intel(R) 965Q";
-         break;
-      case PCI_CHIP_I965_G:
-      case PCI_CHIP_I965_G_1:
-        chipset = "Intel(R) 965G";
-         break;
-      case PCI_CHIP_I946_GZ:
-        chipset = "Intel(R) 946GZ";
-         break;
-      case PCI_CHIP_I965_GM:
-        chipset = "Intel(R) 965GM";
-         break;
-      case PCI_CHIP_I965_GME:
-        chipset = "Intel(R) 965GME/GLE";
-         break;
+      switch (intel->intelScreen->deviceID) {
+#undef CHIPSET
+#define CHIPSET(id, symbol, str) case id: chipset = str; break;
+#include "pci_ids/i965_pci_ids.h"
       default:
-        chipset = "Unknown Intel Chipset";
+         chipset = "Unknown Intel Chipset";
+         break;
       }
 
-      (void) driGetRendererString( buffer, chipset, DRIVER_VERSION, 0 );
+      (void) driGetRendererString(buffer, chipset, 0);
       return (GLubyte *) buffer;
 
    default:
@@ -136,580 +90,819 @@ static const GLubyte *intelGetString( GLcontext *ctx, GLenum name )
    }
 }
 
+void
+intel_resolve_for_dri2_flush(struct brw_context *brw,
+                             __DRIdrawable *drawable)
+{
+   struct intel_context *intel = &brw->intel;
+   if (intel->gen < 6) {
+      /* MSAA and fast color clear are not supported, so don't waste time
+       * checking whether a resolve is needed.
+       */
+      return;
+   }
 
-/**
- * Extension strings exported by the intel driver.
- *
- * \note
- * It appears that ARB_texture_env_crossbar has "disappeared" compared to the
- * old i830-specific driver.
- */
-const struct dri_extension card_extensions[] =
+   struct gl_framebuffer *fb = drawable->driverPrivate;
+   struct intel_renderbuffer *rb;
+
+   /* Usually, only the back buffer will need to be downsampled. However,
+    * the front buffer will also need it if the user has rendered into it.
+    */
+   static const gl_buffer_index buffers[2] = {
+         BUFFER_BACK_LEFT,
+         BUFFER_FRONT_LEFT,
+   };
+
+   for (int i = 0; i < 2; ++i) {
+      rb = intel_get_renderbuffer(fb, buffers[i]);
+      if (rb == NULL || rb->mt == NULL)
+         continue;
+      if (rb->mt->num_samples <= 1)
+         intel_miptree_resolve_color(brw, rb->mt);
+      else
+         intel_miptree_downsample(brw, rb->mt);
+   }
+}
+
+static void
+intel_flush_front(struct gl_context *ctx)
 {
-    { "GL_ARB_multisample",                GL_ARB_multisample_functions },
-    { "GL_ARB_multitexture",               NULL },
-    { "GL_ARB_point_parameters",           GL_ARB_point_parameters_functions },
-    { "GL_NV_point_sprite",                GL_NV_point_sprite_functions },
-    { "GL_ARB_texture_border_clamp",       NULL },
-    { "GL_ARB_texture_compression",        GL_ARB_texture_compression_functions },
-    { "GL_ARB_texture_cube_map",           NULL },
-    { "GL_ARB_texture_env_add",            NULL },
-    { "GL_ARB_texture_env_combine",        NULL },
-    { "GL_ARB_texture_env_dot3",           NULL },
-    { "GL_ARB_texture_mirrored_repeat",    NULL },
-    { "GL_ARB_texture_non_power_of_two",   NULL },
-    { "GL_ARB_texture_rectangle",          NULL },
-    { "GL_NV_texture_rectangle",           NULL },
-    { "GL_EXT_texture_rectangle",          NULL },
-    { "GL_ARB_texture_rectangle",          NULL },
-    { "GL_ARB_point_sprite",               NULL},
-    { "GL_ARB_point_parameters",          NULL }, 
-    { "GL_ARB_vertex_buffer_object",       GL_ARB_vertex_buffer_object_functions },
-    { "GL_ARB_vertex_program",             GL_ARB_vertex_program_functions },
-    { "GL_ARB_window_pos",                 GL_ARB_window_pos_functions },
-    { "GL_EXT_blend_color",                GL_EXT_blend_color_functions },
-    { "GL_EXT_blend_equation_separate",    GL_EXT_blend_equation_separate_functions },
-    { "GL_EXT_blend_func_separate",        GL_EXT_blend_func_separate_functions },
-    { "GL_EXT_blend_minmax",               GL_EXT_blend_minmax_functions },
-    { "GL_EXT_blend_logic_op",             NULL },
-    { "GL_EXT_blend_subtract",             NULL },
-    { "GL_EXT_cull_vertex",                GL_EXT_cull_vertex_functions },
-    { "GL_EXT_fog_coord",                  GL_EXT_fog_coord_functions },
-    { "GL_EXT_multi_draw_arrays",          GL_EXT_multi_draw_arrays_functions },
-    { "GL_EXT_secondary_color",            GL_EXT_secondary_color_functions },
-    { "GL_EXT_stencil_wrap",               NULL },
-    { "GL_EXT_texture_edge_clamp",         NULL },
-    { "GL_EXT_texture_env_combine",        NULL },
-    { "GL_EXT_texture_env_dot3",           NULL },
-    { "GL_EXT_texture_filter_anisotropic", NULL },
-    { "GL_EXT_texture_lod_bias",           NULL },
-    { "GL_EXT_texture_sRGB",               NULL },
-    { "GL_3DFX_texture_compression_FXT1",  NULL },
-    { "GL_APPLE_client_storage",           NULL },
-    { "GL_MESA_pack_invert",               NULL },
-    { "GL_MESA_ycbcr_texture",             NULL },
-    { "GL_NV_blend_square",                NULL },
-    { "GL_SGIS_generate_mipmap",           NULL },
-    { NULL,                                NULL }
-};
+   struct brw_context *brw = brw_context(ctx);
+   struct intel_context *intel = intel_context(ctx);
+    __DRIcontext *driContext = intel->driContext;
+    __DRIdrawable *driDrawable = driContext->driDrawablePriv;
+    __DRIscreen *const screen = intel->intelScreen->driScrnPriv;
+
+    if (brw->front_buffer_dirty && _mesa_is_winsys_fbo(ctx->DrawBuffer)) {
+      if (screen->dri2.loader->flushFrontBuffer != NULL &&
+          driDrawable &&
+          driDrawable->loaderPrivate) {
+
+         /* Resolve before flushing FAKE_FRONT_LEFT to FRONT_LEFT.
+          *
+          * This potentially resolves both front and back buffer. It
+          * is unnecessary to resolve the back, but harms nothing except
+          * performance. And no one cares about front-buffer render
+          * performance.
+          */
+         intel_resolve_for_dri2_flush(brw, driDrawable);
 
-const struct dri_extension arb_oc_extension = 
-    { "GL_ARB_occlusion_query",            GL_ARB_occlusion_query_functions};
+         screen->dri2.loader->flushFrontBuffer(driDrawable,
+                                               driDrawable->loaderPrivate);
 
-void intelInitExtensions(GLcontext *ctx, GLboolean enable_imaging)
-{           
-       struct intel_context *intel = ctx?intel_context(ctx):NULL;
-       driInitExtensions(ctx, card_extensions, enable_imaging);
-       if (!ctx || intel->intelScreen->drmMinor >= 8)
-               driInitSingleExtension (ctx, &arb_oc_extension);
+        /* We set the dirty bit in intel_prepare_render() if we're
+         * front buffer rendering once we get there.
+         */
+        brw->front_buffer_dirty = false;
+      }
+   }
 }
 
-static const struct dri_debug_control debug_control[] =
+static unsigned
+intel_bits_per_pixel(const struct intel_renderbuffer *rb)
 {
-    { "fall",  DEBUG_FALLBACKS },
-    { "tex",   DEBUG_TEXTURE },
-    { "ioctl", DEBUG_IOCTL },
-    { "prim",  DEBUG_PRIMS },
-    { "vert",  DEBUG_VERTS },
-    { "state", DEBUG_STATE },
-    { "verb",  DEBUG_VERBOSE },
-    { "dri",   DEBUG_DRI },
-    { "dma",   DEBUG_DMA },
-    { "san",   DEBUG_SANITY },
-    { "sync",  DEBUG_SYNC },
-    { "sleep", DEBUG_SLEEP },
-    { "pix",   DEBUG_PIXEL },
-    { "buf",   DEBUG_BUFMGR },
-    { "stats", DEBUG_STATS },
-    { "tile",  DEBUG_TILE },
-    { "sing",  DEBUG_SINGLE_THREAD },
-    { "thre",  DEBUG_SINGLE_THREAD },
-    { "wm",    DEBUG_WM },
-    { "vs",    DEBUG_VS },
-    { "bat",   DEBUG_BATCH },
-    { NULL,    0 }
-};
+   return _mesa_get_format_bytes(intel_rb_format(rb)) * 8;
+}
 
+static void
+intel_query_dri2_buffers(struct brw_context *brw,
+                        __DRIdrawable *drawable,
+                        __DRIbuffer **buffers,
+                        int *count);
 
-static void intelInvalidateState( GLcontext *ctx, GLuint new_state )
+static void
+intel_process_dri2_buffer(struct brw_context *brw,
+                         __DRIdrawable *drawable,
+                         __DRIbuffer *buffer,
+                         struct intel_renderbuffer *rb,
+                         const char *buffer_name);
+
+void
+intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable)
 {
-   struct intel_context *intel = intel_context(ctx);
+   struct gl_framebuffer *fb = drawable->driverPrivate;
+   struct intel_renderbuffer *rb;
+   struct brw_context *brw = context->driverPrivate;
+   struct intel_context *intel = &brw->intel;
+   __DRIbuffer *buffers = NULL;
+   int i, count;
+   const char *region_name;
+
+   /* Set this up front, so that in case our buffers get invalidated
+    * while we're getting new buffers, we don't clobber the stamp and
+    * thus ignore the invalidate. */
+   drawable->lastStamp = drawable->dri2.stamp;
+
+   if (unlikely(INTEL_DEBUG & DEBUG_DRI))
+      fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
+
+   intel_query_dri2_buffers(brw, drawable, &buffers, &count);
+
+   if (buffers == NULL)
+      return;
+
+   for (i = 0; i < count; i++) {
+       switch (buffers[i].attachment) {
+       case __DRI_BUFFER_FRONT_LEFT:
+          rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
+          region_name = "dri2 front buffer";
+          break;
+
+       case __DRI_BUFFER_FAKE_FRONT_LEFT:
+          rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
+          region_name = "dri2 fake front buffer";
+          break;
+
+       case __DRI_BUFFER_BACK_LEFT:
+          rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
+          region_name = "dri2 back buffer";
+          break;
+
+       case __DRI_BUFFER_DEPTH:
+       case __DRI_BUFFER_HIZ:
+       case __DRI_BUFFER_DEPTH_STENCIL:
+       case __DRI_BUFFER_STENCIL:
+       case __DRI_BUFFER_ACCUM:
+       default:
+          fprintf(stderr,
+                  "unhandled buffer attach event, attachment type %d\n",
+                  buffers[i].attachment);
+          return;
+       }
+
+       intel_process_dri2_buffer(brw, drawable, &buffers[i], rb, region_name);
+   }
 
-   _swrast_InvalidateState( ctx, new_state );
-   _swsetup_InvalidateState( ctx, new_state );
-   _vbo_InvalidateState( ctx, new_state );
-   _tnl_InvalidateState( ctx, new_state );
-   _tnl_invalidate_vertex_state( ctx, new_state );
-   
-   intel->NewGLState |= new_state;
+   driUpdateFramebufferSize(&intel->ctx, drawable);
+}
+
+/**
+ * intel_prepare_render should be called anywhere that curent read/drawbuffer
+ * state is required.
+ */
+void
+intel_prepare_render(struct brw_context *brw)
+{
+   struct intel_context *intel = &brw->intel;
+   __DRIcontext *driContext = intel->driContext;
+   __DRIdrawable *drawable;
+
+   drawable = driContext->driDrawablePriv;
+   if (drawable && drawable->dri2.stamp != driContext->dri2.draw_stamp) {
+      if (drawable->lastStamp != drawable->dri2.stamp)
+        intel_update_renderbuffers(driContext, drawable);
+      driContext->dri2.draw_stamp = drawable->dri2.stamp;
+   }
+
+   drawable = driContext->driReadablePriv;
+   if (drawable && drawable->dri2.stamp != driContext->dri2.read_stamp) {
+      if (drawable->lastStamp != drawable->dri2.stamp)
+        intel_update_renderbuffers(driContext, drawable);
+      driContext->dri2.read_stamp = drawable->dri2.stamp;
+   }
+
+   /* If we're currently rendering to the front buffer, the rendering
+    * that will happen next will probably dirty the front buffer.  So
+    * mark it as dirty here.
+    */
+   if (brw->is_front_buffer_rendering)
+      brw->front_buffer_dirty = true;
 
-   if (intel->vtbl.invalidate_state)
-      intel->vtbl.invalidate_state( intel, new_state );
+   /* Wait for the swapbuffers before the one we just emitted, so we
+    * don't get too many swaps outstanding for apps that are GPU-heavy
+    * but not CPU-heavy.
+    *
+    * We're using intelDRI2Flush (called from the loader before
+    * swapbuffer) and glFlush (for front buffer rendering) as the
+    * indicator that a frame is done and then throttle when we get
+    * here as we prepare to render the next frame.  At this point for
+    * round trips for swap/copy and getting new buffers are done and
+    * we'll spend less time waiting on the GPU.
+    *
+    * Unfortunately, we don't have a handle to the batch containing
+    * the swap, and getting our hands on that doesn't seem worth it,
+    * so we just us the first batch we emitted after the last swap.
+    */
+   if (intel->need_throttle && intel->first_post_swapbuffers_batch) {
+      if (!intel->disable_throttling)
+         drm_intel_bo_wait_rendering(intel->first_post_swapbuffers_batch);
+      drm_intel_bo_unreference(intel->first_post_swapbuffers_batch);
+      intel->first_post_swapbuffers_batch = NULL;
+      intel->need_throttle = false;
+   }
 }
 
+static const struct dri_debug_control debug_control[] = {
+   { "tex",   DEBUG_TEXTURE},
+   { "state", DEBUG_STATE},
+   { "ioctl", DEBUG_IOCTL},
+   { "blit",  DEBUG_BLIT},
+   { "mip",   DEBUG_MIPTREE},
+   { "fall",  DEBUG_PERF},
+   { "perf",  DEBUG_PERF},
+   { "bat",   DEBUG_BATCH},
+   { "pix",   DEBUG_PIXEL},
+   { "buf",   DEBUG_BUFMGR},
+   { "reg",   DEBUG_REGION},
+   { "fbo",   DEBUG_FBO},
+   { "fs",    DEBUG_WM },
+   { "gs",    DEBUG_GS},
+   { "sync",  DEBUG_SYNC},
+   { "prim",  DEBUG_PRIMS },
+   { "vert",  DEBUG_VERTS },
+   { "dri",   DEBUG_DRI },
+   { "sf",    DEBUG_SF },
+   { "stats", DEBUG_STATS },
+   { "wm",    DEBUG_WM },
+   { "urb",   DEBUG_URB },
+   { "vs",    DEBUG_VS },
+   { "clip",  DEBUG_CLIP },
+   { "aub",   DEBUG_AUB },
+   { "shader_time", DEBUG_SHADER_TIME },
+   { "no16",  DEBUG_NO16 },
+   { "blorp", DEBUG_BLORP },
+   { NULL,    0 }
+};
+
 
-void intelFlush( GLcontext *ctx )
+static void
+intelInvalidateState(struct gl_context * ctx, GLuint new_state)
 {
-   struct intel_context *intel = intel_context( ctx );
+    struct intel_context *intel = intel_context(ctx);
+
+    if (ctx->swrast_context)
+       _swrast_InvalidateState(ctx, new_state);
+   _vbo_InvalidateState(ctx, new_state);
 
-   bmLockAndFence(intel);
+   intel->NewGLState |= new_state;
 }
 
-void intelFinish( GLcontext *ctx ) 
+void
+_intel_flush(struct gl_context *ctx, const char *file, int line)
 {
-   struct intel_context *intel = intel_context( ctx );
+   struct brw_context *brw = brw_context(ctx);
+   struct intel_context *intel = intel_context(ctx);
 
-   bmFinishFence(intel, bmLockAndFence(intel));
+   if (intel->batch.used)
+      _intel_batchbuffer_flush(brw, file, line);
 }
 
 static void
-intelBeginQuery(GLcontext *ctx, GLenum target, struct gl_query_object *q)
+intel_glFlush(struct gl_context *ctx)
 {
-       struct intel_context *intel = intel_context( ctx );
-       drmI830MMIO io = {
-               .read_write = MMIO_READ,
-               .reg = MMIO_REGS_PS_DEPTH_COUNT,
-               .data = &q->Result 
-       };
-       intel->stats_wm++;
-       intelFinish(&intel->ctx);
-       drmCommandWrite(intel->driFd, DRM_I830_MMIO, &io, sizeof(io));
+   struct brw_context *brw = brw_context(ctx);
+   struct intel_context *intel = intel_context(ctx);
+
+   intel_flush(ctx);
+   intel_flush_front(ctx);
+   if (brw->is_front_buffer_rendering)
+      intel->need_throttle = true;
 }
 
-static void
-intelEndQuery(GLcontext *ctx, GLenum target, struct gl_query_object *q)
+void
+intelFinish(struct gl_context * ctx)
 {
-       struct intel_context *intel = intel_context( ctx );
-       GLuint64EXT tmp;        
-       drmI830MMIO io = {
-               .read_write = MMIO_READ,
-               .reg = MMIO_REGS_PS_DEPTH_COUNT,
-               .data = &tmp
-       };
-       intelFinish(&intel->ctx);
-       drmCommandWrite(intel->driFd, DRM_I830_MMIO, &io, sizeof(io));
-       q->Result = tmp - q->Result;
-       q->Ready = GL_TRUE;
-       intel->stats_wm--;
-}
+   struct intel_context *intel = intel_context(ctx);
+
+   intel_flush(ctx);
+   intel_flush_front(ctx);
 
+   if (intel->batch.last_bo)
+      drm_intel_bo_wait_rendering(intel->batch.last_bo);
+}
 
-void intelInitDriverFunctions( struct dd_function_table *functions )
+void
+intelInitDriverFunctions(struct dd_function_table *functions)
 {
-   _mesa_init_driver_functions( functions );
+   _mesa_init_driver_functions(functions);
 
-   functions->Flush = intelFlush;
+   functions->Flush = intel_glFlush;
    functions->Finish = intelFinish;
    functions->GetString = intelGetString;
    functions->UpdateState = intelInvalidateState;
-   functions->BeginQuery = intelBeginQuery;
-   functions->EndQuery = intelEndQuery;
 
-   /* CopyPixels can be accelerated even with the current memory
-    * manager:
-    */
-   if (!getenv("INTEL_NO_BLIT")) {
-      functions->CopyPixels = intelCopyPixels;
-      functions->Bitmap = intelBitmap;
+   intelInitTextureFuncs(functions);
+   intelInitTextureImageFuncs(functions);
+   intelInitTextureSubImageFuncs(functions);
+   intelInitTextureCopyImageFuncs(functions);
+   intelInitClearFuncs(functions);
+   intelInitBufferFuncs(functions);
+   intelInitPixelFuncs(functions);
+   intelInitBufferObjectFuncs(functions);
+   intel_init_syncobj_functions(functions);
+}
+
+static bool
+validate_context_version(struct intel_screen *screen,
+                         int mesa_api,
+                         unsigned major_version,
+                         unsigned minor_version,
+                         unsigned *dri_ctx_error)
+{
+   unsigned req_version = 10 * major_version + minor_version;
+   unsigned max_version = 0;
+
+   switch (mesa_api) {
+   case API_OPENGL_COMPAT:
+      max_version = screen->max_gl_compat_version;
+      break;
+   case API_OPENGL_CORE:
+      max_version = screen->max_gl_core_version;
+      break;
+   case API_OPENGLES:
+      max_version = screen->max_gl_es1_version;
+      break;
+   case API_OPENGLES2:
+      max_version = screen->max_gl_es2_version;
+      break;
+   default:
+      max_version = 0;
+      break;
    }
 
-   intelInitTextureFuncs( functions );
-   intelInitStateFuncs( functions );
-   intelInitBufferFuncs( functions );
+   if (max_version == 0) {
+      *dri_ctx_error = __DRI_CTX_ERROR_BAD_API;
+      return false;
+   } else if (req_version > max_version) {
+      *dri_ctx_error = __DRI_CTX_ERROR_BAD_VERSION;
+      return false;
+   }
+
+   return true;
 }
 
+bool
+intelInitContext(struct brw_context *brw,
+                 int api,
+                 unsigned major_version,
+                 unsigned minor_version,
+                 const struct gl_config * mesaVis,
+                 __DRIcontext * driContextPriv,
+                 void *sharedContextPrivate,
+                 struct dd_function_table *functions,
+                 unsigned *dri_ctx_error)
+{
+   struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &intel->ctx;
+   struct gl_context *shareCtx = (struct gl_context *) sharedContextPrivate;
+   __DRIscreen *sPriv = driContextPriv->driScreenPriv;
+   struct intel_screen *intelScreen = sPriv->driverPrivate;
+   int bo_reuse_mode;
+   struct gl_config visual;
+
+   /* we can't do anything without a connection to the device */
+   if (intelScreen->bufmgr == NULL) {
+      *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
+      return false;
+   }
 
+   if (!validate_context_version(intelScreen,
+                                 api, major_version, minor_version,
+                                 dri_ctx_error))
+      return false;
 
-GLboolean intelInitContext( struct intel_context *intel,
-                           const __GLcontextModes *mesaVis,
-                           __DRIcontextPrivate *driContextPriv,
-                           void *sharedContextPrivate,
-                           struct dd_function_table *functions )
-{
-   GLcontext *ctx = &intel->ctx;
-   GLcontext *shareCtx = (GLcontext *) sharedContextPrivate;
-   __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
-   intelScreenPrivate *intelScreen = (intelScreenPrivate *)sPriv->private;
-   volatile drmI830Sarea *saPriv = (volatile drmI830Sarea *)
-      (((GLubyte *)sPriv->pSAREA)+intelScreen->sarea_priv_offset);
-
-   if (!_mesa_initialize_context(&intel->ctx,
-                                mesaVis, shareCtx, 
-                                functions,
-                                (void*) intel)) {
-      _mesa_printf("%s: failed to init mesa context\n", __FUNCTION__);
-      return GL_FALSE;
+   if (mesaVis == NULL) {
+      memset(&visual, 0, sizeof visual);
+      mesaVis = &visual;
    }
 
-   driContextPriv->driverPrivate = intel;
    intel->intelScreen = intelScreen;
-   intel->driScreen = sPriv;
-   intel->sarea = saPriv;
-
-   driParseConfigFiles (&intel->optionCache, &intelScreen->optionCache,
-                  intel->driScreen->myNum, "i965");
 
-   intel->vblank_flags = (intel->intelScreen->irq_active != 0)
-          ? driGetDefaultVBlankFlags(&intel->optionCache) : VBLANK_FLAG_NO_IRQ;
+   if (!_mesa_initialize_context(&intel->ctx, api, mesaVis, shareCtx,
+                                 functions)) {
+      *dri_ctx_error = __DRI_CTX_ERROR_NO_MEMORY;
+      printf("%s: failed to init mesa context\n", __FUNCTION__);
+      return false;
+   }
 
-   ctx->Const.MaxTextureMaxAnisotropy = 2.0;
+   driContextPriv->driverPrivate = intel;
+   intel->driContext = driContextPriv;
+   intel->driFd = sPriv->fd;
 
-   if (getenv("INTEL_STRICT_CONFORMANCE")) {
-      intel->strict_conformance = 1;
+   intel->gen = intelScreen->gen;
+
+   const int devID = intelScreen->deviceID;
+   if (IS_SNB_GT1(devID) || IS_IVB_GT1(devID) || IS_HSW_GT1(devID))
+      intel->gt = 1;
+   else if (IS_SNB_GT2(devID) || IS_IVB_GT2(devID) || IS_HSW_GT2(devID))
+      intel->gt = 2;
+   else if (IS_HSW_GT3(devID))
+      intel->gt = 3;
+   else
+      intel->gt = 0;
+
+   if (IS_HASWELL(devID)) {
+      intel->is_haswell = true;
+   } else if (IS_BAYTRAIL(devID)) {
+      intel->is_baytrail = true;
+      intel->gt = 1;
+   } else if (IS_G4X(devID)) {
+      intel->is_g4x = true;
    }
 
-   if (intel->strict_conformance) {
-      ctx->Const.MinLineWidth = 1.0;
-      ctx->Const.MinLineWidthAA = 1.0;
-      ctx->Const.MaxLineWidth = 1.0;
-      ctx->Const.MaxLineWidthAA = 1.0;
-      ctx->Const.LineWidthGranularity = 1.0;
-   }
-   else {
-      ctx->Const.MinLineWidth = 1.0;
-      ctx->Const.MinLineWidthAA = 1.0;
-      ctx->Const.MaxLineWidth = 5.0;
-      ctx->Const.MaxLineWidthAA = 5.0;
-      ctx->Const.LineWidthGranularity = 0.5;
-   }
+   intel->has_separate_stencil = intel->intelScreen->hw_has_separate_stencil;
+   intel->must_use_separate_stencil = intel->intelScreen->hw_must_use_separate_stencil;
+   intel->has_hiz = intel->gen >= 6;
+   intel->has_llc = intel->intelScreen->hw_has_llc;
+   intel->has_swizzling = intel->intelScreen->hw_has_swizzling;
 
-   ctx->Const.MinPointSize = 1.0;
-   ctx->Const.MinPointSizeAA = 1.0;
-   ctx->Const.MaxPointSize = 255.0;
-   ctx->Const.MaxPointSizeAA = 3.0;
-   ctx->Const.PointSizeGranularity = 1.0;
+   memset(&ctx->TextureFormatSupported,
+         0, sizeof(ctx->TextureFormatSupported));
 
-   /* reinitialize the context point state.
-    * It depend on constants in __GLcontextRec::Const
-    */
-   _mesa_init_point(ctx);
+   driParseConfigFiles(&brw->optionCache, &intelScreen->optionCache,
+                       sPriv->myNum, "i965");
 
-   /* Initialize the software rasterizer and helper modules. */
-   _swrast_CreateContext( ctx );
-   _vbo_CreateContext( ctx );
-   _tnl_CreateContext( ctx );
-   _swsetup_CreateContext( ctx );
+   /* Estimate the size of the mappable aperture into the GTT.  There's an
+    * ioctl to get the whole GTT size, but not one to get the mappable subset.
+    * It turns out it's basically always 256MB, though some ancient hardware
+    * was smaller.
+    */
+   uint32_t gtt_size = 256 * 1024 * 1024;
 
-   TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
+   /* We don't want to map two objects such that a memcpy between them would
+    * just fault one mapping in and then the other over and over forever.  So
+    * we would need to divide the GTT size by 2.  Additionally, some GTT is
+    * taken up by things like the framebuffer and the ringbuffer and such, so
+    * be more conservative.
+    */
+   intel->max_gtt_map_object_size = gtt_size / 4;
 
-   /* Configure swrast to match hardware characteristics: */
-   _swrast_allow_pixel_fog( ctx, GL_FALSE );
-   _swrast_allow_vertex_fog( ctx, GL_TRUE );
+   intel->bufmgr = intelScreen->bufmgr;
 
-   /* Dri stuff */
-   intel->hHWContext = driContextPriv->hHWContext;
-   intel->driFd = sPriv->fd;
-   intel->driHwLock = (drmLock *) &sPriv->pSAREA->lock;
-
-   intel->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24;
-   intel->hw_stipple = 1;
-
-   switch(mesaVis->depthBits) {
-   case 0:                     /* what to do in this case? */
-   case 16:
-      intel->depth_scale = 1.0/0xffff;
-      intel->polygon_offset_scale = 1.0/0xffff;
-      intel->depth_clear_mask = ~0;
-      intel->ClearDepth = 0xffff;
-      break;
-   case 24:
-      intel->depth_scale = 1.0/0xffffff;
-      intel->polygon_offset_scale = 2.0/0xffffff; /* req'd to pass glean */
-      intel->depth_clear_mask = 0x00ffffff;
-      intel->stencil_clear_mask = 0xff000000;
-      intel->ClearDepth = 0x00ffffff;
+   bo_reuse_mode = driQueryOptioni(&brw->optionCache, "bo_reuse");
+   switch (bo_reuse_mode) {
+   case DRI_CONF_BO_REUSE_DISABLED:
       break;
-   default:
-      assert(0); 
+   case DRI_CONF_BO_REUSE_ALL:
+      intel_bufmgr_gem_enable_reuse(intel->bufmgr);
       break;
    }
 
-   /* Initialize swrast, tnl driver tables: */
-   intelInitSpanFuncs( ctx );
+   /* Initialize the software rasterizer and helper modules.
+    *
+    * As of GL 3.1 core, the gen4+ driver doesn't need the swrast context for
+    * software fallbacks (which we have to support on legacy GL to do weird
+    * glDrawPixels(), glBitmap(), and other functions).
+    */
+   if (api != API_OPENGL_CORE) {
+      _swrast_CreateContext(ctx);
+   }
 
-   intel->no_hw = getenv("INTEL_NO_HW") != NULL;
+   _vbo_CreateContext(ctx);
+   if (ctx->swrast_context) {
+      _tnl_CreateContext(ctx);
+      _swsetup_CreateContext(ctx);
 
-   if (!intel->intelScreen->irq_active) {
-      _mesa_printf("IRQs not active.  Exiting\n");
-      exit(1);
+      /* Configure swrast to match hardware characteristics: */
+      _swrast_allow_pixel_fog(ctx, false);
+      _swrast_allow_vertex_fog(ctx, true);
    }
-   intelInitExtensions(ctx, GL_TRUE); 
 
-   INTEL_DEBUG  = driParseDebugString( getenv( "INTEL_DEBUG" ),
-                                      debug_control );
+   _mesa_meta_init(ctx);
 
+   intelInitExtensions(ctx);
 
-   /* Buffer manager: 
-    */
-   intel->bm = bm_fake_intel_Attach( intel );
+   INTEL_DEBUG = driParseDebugString(getenv("INTEL_DEBUG"), debug_control);
+   if (INTEL_DEBUG & DEBUG_BUFMGR)
+      dri_bufmgr_set_debug(intel->bufmgr, true);
+   if ((INTEL_DEBUG & DEBUG_SHADER_TIME) && intel->gen < 7) {
+      fprintf(stderr,
+              "shader_time debugging requires gen7 (Ivybridge) or better.\n");
+      INTEL_DEBUG &= ~DEBUG_SHADER_TIME;
+   }
+   if (INTEL_DEBUG & DEBUG_PERF)
+      intel->perf_debug = true;
 
+   if (INTEL_DEBUG & DEBUG_AUB)
+      drm_intel_bufmgr_gem_set_aub_dump(intel->bufmgr, true);
 
-   bmInitPool(intel,
-             intel->intelScreen->tex.offset, /* low offset */
-             intel->intelScreen->tex.map, /* low virtual */
-             intel->intelScreen->tex.size,
-             BM_MEM_AGP);
+   intel_batchbuffer_init(brw);
 
-   /* These are still static, but create regions for them.  
-    */
-   intel->front_region = 
-      intel_region_create_static(intel,
-                                BM_MEM_AGP,
-                                intelScreen->front.offset,
-                                intelScreen->front.map,
-                                intelScreen->cpp,
-                                intelScreen->front.pitch / intelScreen->cpp,
-                                intelScreen->height,
-                                intelScreen->front.size,
-                                intelScreen->front.tiled != 0);
-
-   intel->back_region = 
-      intel_region_create_static(intel,
-                                BM_MEM_AGP,
-                                intelScreen->back.offset,
-                                intelScreen->back.map,
-                                intelScreen->cpp,
-                                intelScreen->back.pitch / intelScreen->cpp,
-                                intelScreen->height,
-                                intelScreen->back.size,
-                                 intelScreen->back.tiled != 0);
-
-   /* Still assuming front.cpp == depth.cpp
-    *
-    * XXX: Setting tiling to false because Depth tiling only supports
-    * YMAJOR but the blitter only supports XMAJOR tiling.  Have to
-    * resolve later.
-    */
-   intel->depth_region = 
-      intel_region_create_static(intel,
-                                BM_MEM_AGP,
-                                intelScreen->depth.offset,
-                                intelScreen->depth.map,
-                                intelScreen->cpp,
-                                intelScreen->depth.pitch / intelScreen->cpp,
-                                intelScreen->height,
-                                intelScreen->depth.size,
-                                 intelScreen->depth.tiled != 0);
-   
-   intel_bufferobj_init( intel );
-   intel->batch = intel_batchbuffer_alloc( intel );
-
-   if (intel->ctx.Mesa_DXTn) {
-      _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
-      _mesa_enable_extension( ctx, "GL_S3_s3tc" );
-   }
-   else if (driQueryOptionb (&intel->optionCache, "force_s3tc_enable")) {
-      _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
-   }
+   intel_fbo_init(brw);
 
-/*    driInitTextureObjects( ctx, & intel->swapped, */
-/*                       DRI_TEXMGR_DO_TEXTURE_1D | */
-/*                       DRI_TEXMGR_DO_TEXTURE_2D |  */
-/*                       DRI_TEXMGR_DO_TEXTURE_RECT ); */
+   if (!driQueryOptionb(&brw->optionCache, "hiz")) {
+       intel->has_hiz = false;
+       /* On gen6, you can only do separate stencil with HIZ. */
+       if (intel->gen == 6)
+         intel->has_separate_stencil = false;
+   }
 
+   if (driQueryOptionb(&brw->optionCache, "always_flush_batch")) {
+      fprintf(stderr, "flushing batchbuffer before/after each draw call\n");
+      intel->always_flush_batch = 1;
+   }
 
-   if (getenv("INTEL_NO_RAST")) {
-      fprintf(stderr, "disabling 3D rasterization\n");
-      intel->no_rast = 1;
+   if (driQueryOptionb(&brw->optionCache, "always_flush_cache")) {
+      fprintf(stderr, "flushing GPU caches before/after each draw call\n");
+      intel->always_flush_cache = 1;
    }
 
+   if (driQueryOptionb(&brw->optionCache, "disable_throttling")) {
+      fprintf(stderr, "disabling flush throttling\n");
+      intel->disable_throttling = 1;
+   }
 
-   return GL_TRUE;
+   return true;
 }
 
-void intelDestroyContext(__DRIcontextPrivate *driContextPriv)
+void
+intelDestroyContext(__DRIcontext * driContextPriv)
 {
-   struct intel_context *intel = (struct intel_context *) driContextPriv->driverPrivate;
+   struct brw_context *brw =
+      (struct brw_context *) driContextPriv->driverPrivate;
+   struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &intel->ctx;
 
-   assert(intel); /* should never be null */
+   assert(intel);               /* should never be null */
    if (intel) {
-      GLboolean   release_texture_heaps;
+      /* Dump a final BMP in case the application doesn't call SwapBuffers */
+      if (INTEL_DEBUG & DEBUG_AUB) {
+         intel_batchbuffer_flush(brw);
+        aub_dump_bmp(&intel->ctx);
+      }
 
+      _mesa_meta_free(&intel->ctx);
 
-      intel->vtbl.destroy( intel );
+      brw->vtbl.destroy(brw);
 
-      release_texture_heaps = (intel->ctx.Shared->RefCount == 1);
-      _swsetup_DestroyContext (&intel->ctx);
-      _tnl_DestroyContext (&intel->ctx);
-      _vbo_DestroyContext (&intel->ctx);
+      if (ctx->swrast_context) {
+         _swsetup_DestroyContext(&intel->ctx);
+         _tnl_DestroyContext(&intel->ctx);
+      }
+      _vbo_DestroyContext(&intel->ctx);
 
-      _swrast_DestroyContext (&intel->ctx);
-      intel->Fallback = 0;     /* don't call _swrast_Flush later */
-      intel_batchbuffer_free(intel->batch);
-      intel->batch = NULL;
-      
+      if (ctx->swrast_context)
+         _swrast_DestroyContext(&intel->ctx);
 
-      if ( release_texture_heaps ) {
-         /* This share group is about to go away, free our private
-          * texture object data.
-          */
+      intel_batchbuffer_free(brw);
 
-        /* XXX: destroy the shared bufmgr struct here?
-         */
-      }
+      drm_intel_bo_unreference(intel->first_post_swapbuffers_batch);
+      intel->first_post_swapbuffers_batch = NULL;
 
-      /* Free the regions created to describe front/back/depth
-       * buffers:
-       */
-#if 0
-      intel_region_release(intel, &intel->front_region);
-      intel_region_release(intel, &intel->back_region);
-      intel_region_release(intel, &intel->depth_region);
-      intel_region_release(intel, &intel->draw_region);
-#endif
+      driDestroyOptionCache(&brw->optionCache);
 
       /* free the Mesa context */
-      _mesa_destroy_context(&intel->ctx);
-   }
+      _mesa_free_context_data(&intel->ctx);
 
-   driContextPriv->driverPrivate = NULL;
+      ralloc_free(intel);
+      driContextPriv->driverPrivate = NULL;
+   }
 }
 
-GLboolean intelUnbindContext(__DRIcontextPrivate *driContextPriv)
+GLboolean
+intelUnbindContext(__DRIcontext * driContextPriv)
 {
-   return GL_TRUE;
+   /* Unset current context and dispath table */
+   _mesa_make_current(NULL, NULL, NULL);
+
+   return true;
 }
 
-GLboolean intelMakeCurrent(__DRIcontextPrivate *driContextPriv,
-                         __DRIdrawablePrivate *driDrawPriv,
-                         __DRIdrawablePrivate *driReadPriv)
+/**
+ * Fixes up the context for GLES23 with our default-to-sRGB-capable behavior
+ * on window system framebuffers.
+ *
+ * Desktop GL is fairly reasonable in its handling of sRGB: You can ask if
+ * your renderbuffer can do sRGB encode, and you can flip a switch that does
+ * sRGB encode if the renderbuffer can handle it.  You can ask specifically
+ * for a visual where you're guaranteed to be capable, but it turns out that
+ * everyone just makes all their ARGB8888 visuals capable and doesn't offer
+ * incapable ones, becuase there's no difference between the two in resources
+ * used.  Applications thus get built that accidentally rely on the default
+ * visual choice being sRGB, so we make ours sRGB capable.  Everything sounds
+ * great...
+ *
+ * But for GLES2/3, they decided that it was silly to not turn on sRGB encode
+ * for sRGB renderbuffers you made with the GL_EXT_texture_sRGB equivalent.
+ * So they removed the enable knob and made it "if the renderbuffer is sRGB
+ * capable, do sRGB encode".  Then, for your window system renderbuffers, you
+ * can ask for sRGB visuals and get sRGB encode, or not ask for sRGB visuals
+ * and get no sRGB encode (assuming that both kinds of visual are available).
+ * Thus our choice to support sRGB by default on our visuals for desktop would
+ * result in broken rendering of GLES apps that aren't expecting sRGB encode.
+ *
+ * Unfortunately, renderbuffer setup happens before a context is created.  So
+ * in intel_screen.c we always set up sRGB, and here, if you're a GLES2/3
+ * context (without an sRGB visual, though we don't have sRGB visuals exposed
+ * yet), we go turn that back off before anyone finds out.
+ */
+static void
+intel_gles3_srgb_workaround(struct brw_context *brw,
+                            struct gl_framebuffer *fb)
 {
+   struct intel_context *intel = &brw->intel;
+   struct gl_context *ctx = &intel->ctx;
 
-   if (driContextPriv) {
-      struct intel_context *intel = (struct intel_context *) driContextPriv->driverPrivate;
-
-      if (intel->driReadDrawable != driReadPriv) {
-          intel->driReadDrawable = driReadPriv;
-      }
-
-      if ( intel->driDrawable != driDrawPriv ) {
-        /* Shouldn't the readbuffer be stored also? */
-        driDrawableInitVBlank( driDrawPriv, intel->vblank_flags,
-                     &intel->vbl_seq );
+   if (_mesa_is_desktop_gl(ctx) || !fb->Visual.sRGBCapable)
+      return;
 
-        intel->driDrawable = driDrawPriv;
-        intelWindowMoved( intel );
+   /* Some day when we support the sRGB capable bit on visuals available for
+    * GLES, we'll need to respect that and not disable things here.
+    */
+   fb->Visual.sRGBCapable = false;
+   for (int i = 0; i < BUFFER_COUNT; i++) {
+      if (fb->Attachment[i].Renderbuffer &&
+          fb->Attachment[i].Renderbuffer->Format == MESA_FORMAT_SARGB8) {
+         fb->Attachment[i].Renderbuffer->Format = MESA_FORMAT_ARGB8888;
       }
-
-      _mesa_make_current(&intel->ctx,
-                        (GLframebuffer *) driDrawPriv->driverPrivate,
-                        (GLframebuffer *) driReadPriv->driverPrivate);
-
-      intel->ctx.Driver.DrawBuffer( &intel->ctx, intel->ctx.Color.DrawBuffer[0] );
-   } else {
-      _mesa_make_current(NULL, NULL, NULL);
    }
-
-   return GL_TRUE;
 }
 
-
-static void intelContendedLock( struct intel_context *intel, GLuint flags )
+GLboolean
+intelMakeCurrent(__DRIcontext * driContextPriv,
+                 __DRIdrawable * driDrawPriv,
+                 __DRIdrawable * driReadPriv)
 {
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   __DRIscreenPrivate *sPriv = intel->driScreen;
-   volatile drmI830Sarea * sarea = intel->sarea;
-   int me = intel->hHWContext;
-   int my_bufmgr = bmCtxId(intel);
+   struct brw_context *brw;
+   GET_CURRENT_CONTEXT(curCtx);
 
-   drmGetLock(intel->driFd, intel->hHWContext, flags);
+   if (driContextPriv)
+      brw = (struct brw_context *) driContextPriv->driverPrivate;
+   else
+      brw = NULL;
 
-   /* If the window moved, may need to set a new cliprect now.
-    *
-    * NOTE: This releases and regains the hw lock, so all state
-    * checking must be done *after* this call:
+   /* According to the glXMakeCurrent() man page: "Pending commands to
+    * the previous context, if any, are flushed before it is released."
+    * But only flush if we're actually changing contexts.
     */
-   if (dPriv)
-      DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);
+   if (brw_context(curCtx) && brw_context(curCtx) != brw) {
+      _mesa_flush(curCtx);
+   }
 
+   struct intel_context *intel = &brw->intel;
 
-   intel->locked = 1;
-   intel->need_flush = 1;
+   if (driContextPriv) {
+      struct gl_context *ctx = &intel->ctx;
+      struct gl_framebuffer *fb, *readFb;
+      
+      if (driDrawPriv == NULL && driReadPriv == NULL) {
+        fb = _mesa_get_incomplete_framebuffer();
+        readFb = _mesa_get_incomplete_framebuffer();
+      } else {
+        fb = driDrawPriv->driverPrivate;
+        readFb = driReadPriv->driverPrivate;
+        driContextPriv->dri2.draw_stamp = driDrawPriv->dri2.stamp - 1;
+        driContextPriv->dri2.read_stamp = driReadPriv->dri2.stamp - 1;
+      }
 
-   /* Lost context?
-    */
-   if (sarea->ctxOwner != me) {
-      DBG("Lost Context: sarea->ctxOwner %x me %x\n", sarea->ctxOwner, me);
-      sarea->ctxOwner = me;
-      intel->vtbl.lost_hardware( intel );
-   }
+      intel_prepare_render(brw);
+      _mesa_make_current(ctx, fb, readFb);
 
-   /* As above, but don't evict the texture data on transitions
-    * between contexts which all share a local buffer manager.
-    */
-   if (sarea->texAge != my_bufmgr) {
-      DBG("Lost Textures: sarea->texAge %x my_bufmgr %x\n", sarea->ctxOwner, my_bufmgr);
-      sarea->texAge = my_bufmgr;
-      bm_fake_NotifyContendedLockTake( intel ); 
+      intel_gles3_srgb_workaround(brw, ctx->WinSysDrawBuffer);
+      intel_gles3_srgb_workaround(brw, ctx->WinSysReadBuffer);
    }
-
-   /* Drawable changed?
-    */
-   if (dPriv && intel->lastStamp != dPriv->lastStamp) {
-      intelWindowMoved( intel );
-      intel->lastStamp = dPriv->lastStamp;
+   else {
+      _mesa_make_current(NULL, NULL, NULL);
    }
-}
 
-_glthread_DECLARE_STATIC_MUTEX(lockMutex);
+   return true;
+}
 
-/* Lock the hardware and validate our state.  
+/**
+ * \brief Query DRI2 to obtain a DRIdrawable's buffers.
+ *
+ * To determine which DRI buffers to request, examine the renderbuffers
+ * attached to the drawable's framebuffer. Then request the buffers with
+ * DRI2GetBuffers() or DRI2GetBuffersWithFormat().
+ *
+ * This is called from intel_update_renderbuffers().
+ *
+ * \param drawable      Drawable whose buffers are queried.
+ * \param buffers       [out] List of buffers returned by DRI2 query.
+ * \param buffer_count  [out] Number of buffers returned.
+ *
+ * \see intel_update_renderbuffers()
+ * \see DRI2GetBuffers()
+ * \see DRI2GetBuffersWithFormat()
  */
-void LOCK_HARDWARE( struct intel_context *intel )
+static void
+intel_query_dri2_buffers(struct brw_context *brw,
+                        __DRIdrawable *drawable,
+                        __DRIbuffer **buffers,
+                        int *buffer_count)
 {
-    char __ret=0;
+   struct intel_context *intel = &brw->intel;
+   __DRIscreen *screen = intel->intelScreen->driScrnPriv;
+   struct gl_framebuffer *fb = drawable->driverPrivate;
+   int i = 0;
+   unsigned attachments[8];
+
+   struct intel_renderbuffer *front_rb;
+   struct intel_renderbuffer *back_rb;
+
+   front_rb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
+   back_rb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
+
+   memset(attachments, 0, sizeof(attachments));
+   if ((brw->is_front_buffer_rendering ||
+       brw->is_front_buffer_reading ||
+       !back_rb) && front_rb) {
+      /* If a fake front buffer is in use, then querying for
+       * __DRI_BUFFER_FRONT_LEFT will cause the server to copy the image from
+       * the real front buffer to the fake front buffer.  So before doing the
+       * query, we need to make sure all the pending drawing has landed in the
+       * real front buffer.
+       */
+      intel_flush(&intel->ctx);
+      intel_flush_front(&intel->ctx);
+
+      attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
+      attachments[i++] = intel_bits_per_pixel(front_rb);
+   } else if (front_rb && brw->front_buffer_dirty) {
+      /* We have pending front buffer rendering, but we aren't querying for a
+       * front buffer.  If the front buffer we have is a fake front buffer,
+       * the X server is going to throw it away when it processes the query.
+       * So before doing the query, make sure all the pending drawing has
+       * landed in the real front buffer.
+       */
+      intel_flush(&intel->ctx);
+      intel_flush_front(&intel->ctx);
+   }
 
-    _glthread_LOCK_MUTEX(lockMutex);
-    assert(!intel->locked);
+   if (back_rb) {
+      attachments[i++] = __DRI_BUFFER_BACK_LEFT;
+      attachments[i++] = intel_bits_per_pixel(back_rb);
+   }
 
+   assert(i <= ARRAY_SIZE(attachments));
 
-    DRM_CAS(intel->driHwLock, intel->hHWContext,
-           (DRM_LOCK_HELD|intel->hHWContext), __ret);
-    if (__ret)
-        intelContendedLock( intel, 0 );
+   *buffers = screen->dri2.loader->getBuffersWithFormat(drawable,
+                                                       &drawable->w,
+                                                       &drawable->h,
+                                                       attachments, i / 2,
+                                                       buffer_count,
+                                                       drawable->loaderPrivate);
+}
 
-   intel->locked = 1;
+/**
+ * \brief Assign a DRI buffer's DRM region to a renderbuffer.
+ *
+ * This is called from intel_update_renderbuffers().
+ *
+ * \par Note:
+ *    DRI buffers whose attachment point is DRI2BufferStencil or
+ *    DRI2BufferDepthStencil are handled as special cases.
+ *
+ * \param buffer_name is a human readable name, such as "dri2 front buffer",
+ *        that is passed to intel_region_alloc_for_handle().
+ *
+ * \see intel_update_renderbuffers()
+ * \see intel_region_alloc_for_handle()
+ */
+static void
+intel_process_dri2_buffer(struct brw_context *brw,
+                         __DRIdrawable *drawable,
+                         __DRIbuffer *buffer,
+                         struct intel_renderbuffer *rb,
+                         const char *buffer_name)
+{
+   struct intel_context *intel = &brw->intel;
+   struct intel_region *region = NULL;
 
-   if (bmError(intel)) {
-      bmEvictAll(intel);
-      intel->vtbl.lost_hardware( intel );
-   }
+   if (!rb)
+      return;
 
-   /* Make sure nothing has been emitted prior to getting the lock:
-    */
-   assert(intel->batch->map == 0);
+   unsigned num_samples = rb->Base.Base.NumSamples;
 
-   /* XXX: postpone, may not be needed:
+   /* We try to avoid closing and reopening the same BO name, because the first
+    * use of a mapping of the buffer involves a bunch of page faulting which is
+    * moderately expensive.
     */
-   if (!intel_batchbuffer_map(intel->batch)) {
-      bmEvictAll(intel);
-      intel->vtbl.lost_hardware( intel );
-
-      /* This could only fail if the batchbuffer was greater in size
-       * than the available texture memory:
-       */
-      if (!intel_batchbuffer_map(intel->batch)) {
-        _mesa_printf("double failure to map batchbuffer\n");
-        assert(0);
-      }
+   if (num_samples == 0) {
+       if (rb->mt &&
+           rb->mt->region &&
+           rb->mt->region->name == buffer->name)
+          return;
+   } else {
+       if (rb->mt &&
+           rb->mt->singlesample_mt &&
+           rb->mt->singlesample_mt->region &&
+           rb->mt->singlesample_mt->region->name == buffer->name)
+          return;
    }
-}
-  
-/* Unlock the hardware using the global current context 
- */
-void UNLOCK_HARDWARE( struct intel_context *intel )
-{
-   /* Make sure everything has been released: 
-    */
-   assert(intel->batch->ptr == intel->batch->map + intel->batch->offset);
 
-   intel_batchbuffer_unmap(intel->batch);
-   intel->vtbl.note_unlock( intel );
-   intel->locked = 0;
+   if (unlikely(INTEL_DEBUG & DEBUG_DRI)) {
+      fprintf(stderr,
+             "attaching buffer %d, at %d, cpp %d, pitch %d\n",
+             buffer->name, buffer->attachment,
+             buffer->cpp, buffer->pitch);
+   }
 
-   DRM_UNLOCK(intel->driFd, intel->driHwLock, intel->hHWContext);
-   _glthread_UNLOCK_MUTEX(lockMutex); 
+   intel_miptree_release(&rb->mt);
+   region = intel_region_alloc_for_handle(intel->intelScreen,
+                                          buffer->cpp,
+                                          drawable->w,
+                                          drawable->h,
+                                          buffer->pitch,
+                                          buffer->name,
+                                          buffer_name);
+   if (!region)
+      return;
+
+   rb->mt = intel_miptree_create_for_dri2_buffer(brw,
+                                                 buffer->attachment,
+                                                 intel_rb_format(rb),
+                                                 num_samples,
+                                                 region);
+   intel_region_release(&region);
 }
-
-