Merge remote branch 'origin/mesa_7_7_branch'

38 files changed, 743 insertions, 740 deletions

diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 94d000a5ac..5f60139968 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -120,7 +120,7 @@ softpipe_destroy( struct pipe_context *pipe )
  * if (the texture is being used as a framebuffer surface)
  *    return PIPE_REFERENCED_FOR_WRITE
  * else if (the texture is a bound texture source)
- *    return PIPE_REFERENCED_FOR_READ  XXX not done yet
+ *    return PIPE_REFERENCED_FOR_READ
  * else
  *    return PIPE_UNREFERENCED
  */
@@ -132,6 +132,7 @@ softpipe_is_texture_referenced( struct pipe_context *pipe,
    struct softpipe_context *softpipe = softpipe_context( pipe );
    unsigned i;
 
+   /* check if any of the bound drawing surfaces are this texture */
    if (softpipe->dirty_render_cache) {
       for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) {
          if (softpipe->framebuffer.cbufs[i] && 
@@ -145,7 +146,12 @@ softpipe_is_texture_referenced( struct pipe_context *pipe,
       }
    }
    
-   /* FIXME: we also need to do the same for the texture cache */
+   /* check if any of the tex_cache textures are this texture */
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      if (softpipe->tex_cache[i] &&
+          softpipe->tex_cache[i]->texture == texture)
+         return PIPE_REFERENCED_FOR_READ;
+   }
    
    return PIPE_UNREFERENCED;
 }
@@ -243,9 +249,9 @@ softpipe_create( struct pipe_screen *screen )
 
 
    /* setup quad rendering stages */
-      softpipe->quad.shade = sp_quad_shade_stage(softpipe);
-      softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe);
-      softpipe->quad.blend = sp_quad_blend_stage(softpipe);
+   softpipe->quad.shade = sp_quad_shade_stage(softpipe);
+   softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe);
+   softpipe->quad.blend = sp_quad_blend_stage(softpipe);
 
 
    /*
@@ -275,7 +281,6 @@ softpipe_create( struct pipe_screen *screen )
    draw_set_render(softpipe->draw, softpipe->vbuf_backend);
 
 
-
    /* plug in AA line/point stages */
    draw_install_aaline_stage(softpipe->draw, &softpipe->pipe);
    draw_install_aapoint_stage(softpipe->draw, &softpipe->pipe);
@@ -291,4 +296,3 @@ softpipe_create( struct pipe_screen *screen )
    softpipe_destroy(&softpipe->pipe);
    return NULL;
 }
-
diff --git a/src/gallium/state_trackers/xorg/xorg_composite.c b/src/gallium/state_trackers/xorg/xorg_composite.c
index 733bd53fca..86a52077c3 100644
--- a/src/gallium/state_trackers/xorg/xorg_composite.c
+++ b/src/gallium/state_trackers/xorg/xorg_composite.c
@@ -152,7 +152,8 @@ render_filter_to_gallium(int xrender_filter, int *out_filter)
       *out_filter = PIPE_TEX_FILTER_LINEAR;
       break;
    default:
-      debug_printf("Unkown xrender filter");
+      debug_printf("Unkown xrender filter\n");
+   case PictFilterConvolution:
       *out_filter = PIPE_TEX_FILTER_NEAREST;
       return FALSE;
    }
diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c
index c41a7cd639..ca3c712dcd 100644
--- a/src/gallium/state_trackers/xorg/xorg_dri2.c
+++ b/src/gallium/state_trackers/xorg/xorg_dri2.c
@@ -42,6 +42,12 @@
 
 #include "util/u_rect.h"
 
+/* Make all the #if cases in the code esier to read */
+/* XXX can it be set to 1? */
+#ifndef DRI2INFOREC_VERSION
+#define DRI2INFOREC_VERSION 0
+#endif
+
 typedef struct {
     PixmapPtr pPixmap;
     struct pipe_texture *tex;
@@ -79,7 +85,7 @@ driDoCreateBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int format)
     case DRI2BufferFrontLeft:
 	break;
     case DRI2BufferStencil:
-#if defined(DRI2INFOREC_VERSION) && DRI2INFOREC_VERSION > 2
+#if DRI2INFOREC_VERSION >= 3
     case DRI2BufferDepthStencil:
 #else
     /* Works on old X servers because sanity checking is for the weak */
@@ -121,9 +127,12 @@ driDoCreateBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int format)
     }
 
     if (!tex) {
+	/* First call to make sure we have a pixmap private */
 	exaMoveInPixmap(private->pPixmap);
 	xorg_exa_set_shared_usage(private->pPixmap);
 	pScreen->ModifyPixmapHeader(private->pPixmap, 0, 0, 0, 0, 0, NULL);
+	/* Second call to make sure texture has valid contents */
+	exaMoveInPixmap(private->pPixmap);
 	tex = xorg_exa_get_texture(private->pPixmap);
     }
 
@@ -137,6 +146,11 @@ driDoCreateBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int format)
     buffer->cpp = 4;
     buffer->driverPrivate = private;
     buffer->flags = 0; /* not tiled */
+#if DRI2INFOREC_VERSION == 2
+    ((DRI2Buffer2Ptr)buffer)->format = 0;
+#elif DRI2INFOREC_VERSION >= 3
+    buffer->format = 0;
+#endif
     private->tex = tex;
 
     return TRUE;
@@ -157,12 +171,12 @@ driDoDestroyBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer)
     (*pScreen->DestroyPixmap)(private->pPixmap);
 }
 
-#if defined(DRI2INFOREC_VERSION) && DRI2INFOREC_VERSION > 2
+#if DRI2INFOREC_VERSION >= 2
 
-static DRI2BufferPtr
+static DRI2Buffer2Ptr
 driCreateBuffer(DrawablePtr pDraw, unsigned int attachment, unsigned int format)
 {
-    DRI2BufferPtr buffer;
+    DRI2Buffer2Ptr buffer;
     BufferPrivatePtr private;
 
     buffer = xcalloc(1, sizeof *buffer);
@@ -177,7 +191,8 @@ driCreateBuffer(DrawablePtr pDraw, unsigned int attachment, unsigned int format)
     buffer->attachment = attachment;
     buffer->driverPrivate = private;
 
-    if (driDoCreateBuffer(pDraw, buffer, format))
+    /* So far it is safe to downcast a DRI2Buffer2Ptr to DRI2BufferPtr */
+    if (driDoCreateBuffer(pDraw, (DRI2BufferPtr)buffer, format))
 	return buffer;
 
     xfree(private);
@@ -187,15 +202,16 @@ fail:
 }
 
 static void
-driDestroyBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer)
+driDestroyBuffer(DrawablePtr pDraw, DRI2Buffer2Ptr buffer)
 {
-    driDoDestroyBuffer(pDraw, buffer);
+    /* So far it is safe to downcast a DRI2Buffer2Ptr to DRI2BufferPtr */
+    driDoDestroyBuffer(pDraw, (DRI2BufferPtr)buffer);
 
     xfree(buffer->driverPrivate);
     xfree(buffer);
 }
 
-#else /* DRI2INFOREC_VERSION <= 2 */
+#else /* DRI2INFOREC_VERSION < 2 */
 
 static DRI2BufferPtr
 driCreateBuffers(DrawablePtr pDraw, unsigned int *attachments, int count)
@@ -245,7 +261,7 @@ driDestroyBuffers(DrawablePtr pDraw, DRI2BufferPtr buffers, int count)
     }
 }
 
-#endif /* DRI2INFOREC_VERSION */
+#endif /* DRI2INFOREC_VERSION >= 2 */
 
 static void
 driCopyRegion(DrawablePtr pDraw, RegionPtr pRegion,
@@ -260,6 +276,7 @@ driCopyRegion(DrawablePtr pDraw, RegionPtr pRegion,
     PixmapPtr dst_pixmap;
     GCPtr gc;
     RegionPtr copy_clip;
+    Bool save_accel;
 
     /*
      * In driCreateBuffers we dewrap windows into the
@@ -325,8 +342,11 @@ driCopyRegion(DrawablePtr pDraw, RegionPtr pRegion,
 	}
     }
 
+    save_accel = ms->exa->accel;
+    ms->exa->accel = TRUE;
     (*gc->ops->CopyArea)(&src_pixmap->drawable, &dst_pixmap->drawable, gc,
 			 0, 0, pDraw->width, pDraw->height, 0, 0);
+    ms->exa->accel = save_accel;
 
     FreeScratchGC(gc);
 
@@ -342,7 +362,7 @@ driScreenInit(ScreenPtr pScreen)
     modesettingPtr ms = modesettingPTR(pScrn);
     DRI2InfoRec dri2info;
 
-#if defined(DRI2INFOREC_VERSION)
+#if DRI2INFOREC_VERSION >= 2
     dri2info.version = DRI2INFOREC_VERSION;
 #else
     dri2info.version = 1;
@@ -352,7 +372,7 @@ driScreenInit(ScreenPtr pScreen)
     dri2info.driverName = pScrn->driverName;
     dri2info.deviceName = "/dev/dri/card0"; /* FIXME */
 
-#if defined(DRI2INFOREC_VERSION) && DRI2INFOREC_VERSION > 2
+#if DRI2INFOREC_VERSION >= 2
     dri2info.CreateBuffer = driCreateBuffer;
     dri2info.DestroyBuffer = driDestroyBuffer;
 #else
diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c
index 26cf2dd772..d949167adc 100644
--- a/src/gallium/state_trackers/xorg/xorg_driver.c
+++ b/src/gallium/state_trackers/xorg/xorg_driver.c
@@ -75,10 +75,12 @@ static Bool PreInit(ScrnInfoPtr pScrn, int flags);
 typedef enum
 {
     OPTION_SW_CURSOR,
+    OPTION_2D_ACCEL,
 } modesettingOpts;
 
 static const OptionInfoRec Options[] = {
     {OPTION_SW_CURSOR, "SWcursor", OPTV_BOOLEAN, {0}, FALSE},
+    {OPTION_2D_ACCEL, "2DAccel", OPTV_BOOLEAN, {0}, FALSE},
     {-1, NULL, OPTV_NONE, {0}, FALSE}
 };
 
@@ -609,7 +611,8 @@ ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 
     xf86SetBlackWhitePixels(pScreen);
 
-    ms->exa = xorg_exa_init(pScrn);
+    ms->exa = xorg_exa_init(pScrn, xf86ReturnOptValBool(ms->Options,
+                                                        OPTION_2D_ACCEL, TRUE));
     ms->debug_fallback = debug_get_bool_option("XORG_DEBUG_FALLBACK", TRUE);
 
     xorg_init_video(pScreen);
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index 6fa274eb0a..3a51ad2d59 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -46,7 +46,6 @@
 #include "util/u_rect.h"
 
 #define DEBUG_PRINT 0
-#define ACCEL_ENABLED TRUE
 
 /*
  * Helper functions
@@ -170,15 +169,7 @@ xorg_exa_common_done(struct exa_context *exa)
 static void
 ExaWaitMarker(ScreenPtr pScreen, int marker)
 {
-   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
-   modesettingPtr ms = modesettingPTR(pScrn);
-   struct exa_context *exa = ms->exa;
-
-#if 0
-   xorg_exa_flush(exa, PIPE_FLUSH_RENDER_CACHE, NULL);
-#else
-   xorg_exa_finish(exa);
-#endif
+   /* Nothing to do, handled in the PrepareAccess hook */
 }
 
 static int
@@ -384,7 +375,7 @@ ExaPrepareSolid(PixmapPtr pPixmap, int alu, Pixel planeMask, Pixel fg)
 	XORG_FALLBACK("format %s", pf_name(priv->tex->format));
     }
 
-    return ACCEL_ENABLED && xorg_solid_bind_state(exa, priv, fg);
+    return exa->accel && xorg_solid_bind_state(exa, priv, fg);
 }
 
 static void
@@ -443,7 +434,7 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir,
     exa->copy.src = src_priv;
     exa->copy.dst = priv;
 
-    return ACCEL_ENABLED;
+    return exa->accel;
 }
 
 static void
@@ -572,7 +563,7 @@ ExaPrepareComposite(int op, PicturePtr pSrcPicture,
                        render_format_name(pMaskPicture->format));
    }
 
-   return ACCEL_ENABLED &&
+   return exa->accel &&
           xorg_composite_bind_state(exa, op, pSrcPicture, pMaskPicture,
                                     pDstPicture,
                                     pSrc ? exaGetPixmapDriverPrivate(pSrc) : NULL,
@@ -605,6 +596,9 @@ ExaCheckComposite(int op,
 		  PicturePtr pSrcPicture, PicturePtr pMaskPicture,
 		  PicturePtr pDstPicture)
 {
+   ScrnInfoPtr pScrn = xf86Screens[pDstPicture->pDrawable->pScreen->myNum];
+   modesettingPtr ms = modesettingPTR(pScrn);
+   struct exa_context *exa = ms->exa;
    boolean accelerated = xorg_composite_accelerated(op,
                                                     pSrcPicture,
                                                     pMaskPicture,
@@ -613,7 +607,7 @@ ExaCheckComposite(int op,
    debug_printf("ExaCheckComposite(%d, %p, %p, %p) = %d\n",
                 op, pSrcPicture, pMaskPicture, pDstPicture, accelerated);
 #endif
-   return ACCEL_ENABLED && accelerated;
+   return exa->accel && accelerated;
 }
 
 static void *
@@ -751,10 +745,11 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height,
 			     bitsPerPixel, devKind, NULL);
 
     /* Deal with screen resize */
-    if (!priv->tex ||
-        (priv->tex->width[0] != width ||
-         priv->tex->height[0] != height ||
-         priv->tex_flags != priv->flags)) {
+    if ((exa->accel || priv->flags) &&
+        (!priv->tex ||
+         (priv->tex->width[0] != width ||
+          priv->tex->height[0] != height ||
+          priv->tex_flags != priv->flags))) {
 	struct pipe_texture *texture = NULL;
 	struct pipe_texture template;
 
@@ -869,7 +864,7 @@ xorg_exa_close(ScrnInfoPtr pScrn)
 }
 
 void *
-xorg_exa_init(ScrnInfoPtr pScrn)
+xorg_exa_init(ScrnInfoPtr pScrn, Bool accel)
 {
    modesettingPtr ms = modesettingPTR(pScrn);
    struct exa_context *exa;
@@ -934,6 +929,7 @@ xorg_exa_init(ScrnInfoPtr pScrn)
    ms->ctx = exa->pipe;
 
    exa->renderer = renderer_create(exa->pipe);
+   exa->accel = accel;
 
    return (void *)exa;
 
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.h b/src/gallium/state_trackers/xorg/xorg_exa.h
index 7f4aebb9c3..15cc29d662 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.h
+++ b/src/gallium/state_trackers/xorg/xorg_exa.h
@@ -24,6 +24,8 @@ struct exa_context
    float solid_color[4];
    boolean has_solid_color;
 
+   boolean accel;
+
    /* float[9] projective matrix bound to pictures */
    struct {
       float    src[9];
diff --git a/src/gallium/state_trackers/xorg/xorg_tracker.h b/src/gallium/state_trackers/xorg/xorg_tracker.h
index 6130cf6621..20c9259c7b 100644
--- a/src/gallium/state_trackers/xorg/xorg_tracker.h
+++ b/src/gallium/state_trackers/xorg/xorg_tracker.h
@@ -131,7 +131,7 @@ xorg_exa_create_root_texture(ScrnInfoPtr pScrn,
 			     int depth, int bpp);
 
 void *
-xorg_exa_init(ScrnInfoPtr pScrn);
+xorg_exa_init(ScrnInfoPtr pScrn, Bool accel);
 
 void
 xorg_exa_close(ScrnInfoPtr pScrn);
diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c
index 5794395771..76cd3ac1aa 100644
--- a/src/gallium/state_trackers/xorg/xorg_xv.c
+++ b/src/gallium/state_trackers/xorg/xorg_xv.c
@@ -257,7 +257,7 @@ copy_packed_data(ScrnInfoPtr pScrn,
    switch (id) {
    case FOURCC_YV12: {
       for (i = 0; i < w; ++i) {
-         for (j = 0; i < h; ++j) {
+         for (j = 0; j < h; ++j) {
             /*XXX use src? */
             y1  = buf[j*w + i];
             u   = buf[(j/2) * (w/2) + i/2 + y_array_size];
@@ -447,6 +447,11 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
    int x, y, w, h;
    struct exa_pixmap_priv *dst = exaGetPixmapDriverPrivate(pPixmap);
 
+   if (!dst->tex) {
+	xorg_exa_set_shared_usage(pPixmap);
+	pScrn->pScreen->ModifyPixmapHeader(pPixmap, 0, 0, 0, 0, 0, NULL);
+   }
+
    if (!dst || !dst->tex)
       XORG_FALLBACK("Xv destination %s", !dst ? "!dst" : "!dst->tex");
 
@@ -470,6 +475,9 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
    setup_vs_video_constants(pPriv->r, dst);
    setup_fs_video_constants(pPriv->r, hdtv);
 
+   exaMoveInPixmap(pPixmap);
+   DamageDamageRegion(&pPixmap->drawable, dstRegion);
+
    while (nbox--) {
       int box_x1 = pbox->x1;
       int box_y1 = pbox->y1;
@@ -477,8 +485,8 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
       int box_y2 = pbox->y2;
       float diff_x = (float)src_w / (float)dst_w;
       float diff_y = (float)src_h / (float)dst_h;
-      int offset_x = box_x1 - dstX;
-      int offset_y = box_y1 - dstY;
+      int offset_x = box_x1 - dstX + pPixmap->screen_x;
+      int offset_y = box_y1 - dstY + pPixmap->screen_y;
       int offset_w;
       int offset_h;
 
@@ -496,7 +504,7 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
 
       pbox++;
    }
-   DamageDamageRegion(&pPixmap->drawable, dstRegion);
+   DamageRegionProcessPending(&pPixmap->drawable);
 
    return TRUE;
 }
diff --git a/src/mesa/drivers/dri/r200/r200_state_init.c b/src/mesa/drivers/dri/r200/r200_state_init.c
index 68bfeea701..e53fd72290 100644
--- a/src/mesa/drivers/dri/r200/r200_state_init.c
+++ b/src/mesa/drivers/dri/r200/r200_state_init.c
@@ -640,7 +640,7 @@ static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom)
    OUT_BATCH_TABLE(atom->cmd, 10);
 
    if (t && t->mt && !t->image_override) {
-     OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+     OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, get_base_teximage_offset(t),
 		  RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
    } else if (!t) {
      /* workaround for old CS mechanism */
diff --git a/src/mesa/drivers/dri/r200/r200_tex.c b/src/mesa/drivers/dri/r200/r200_tex.c
index 5a21a8b9c5..a417721553 100644
--- a/src/mesa/drivers/dri/r200/r200_tex.c
+++ b/src/mesa/drivers/dri/r200/r200_tex.c
@@ -385,16 +385,7 @@ static void r200TexParameter( GLcontext *ctx, GLenum target,
    case GL_TEXTURE_MAX_LEVEL:
    case GL_TEXTURE_MIN_LOD:
    case GL_TEXTURE_MAX_LOD:
-      /* This isn't the most efficient solution but there doesn't appear to
-       * be a nice alternative.  Since there's no LOD clamping,
-       * we just have to rely on loading the right subset of mipmap levels
-       * to simulate a clamped LOD.
-       */
-      if (t->mt) {
-         radeon_miptree_unreference(t->mt);
-	 t->mt = 0;
-	 t->validated = GL_FALSE;
-      }
+      t->validated = GL_FALSE;
       break;
 
    default:
@@ -413,7 +404,7 @@ static void r200DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 	      (void *)texObj,
 	      _mesa_lookup_enum_by_nr(texObj->Target));
    }
-   
+
    if (rmesa) {
       int i;
       radeon_firevertices(&rmesa->radeon);
@@ -425,11 +416,9 @@ static void r200DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 	 }
       }      
    }
-   
-   if (t->mt) {
-      radeon_miptree_unreference(t->mt);
-      t->mt = 0;
-   }
+
+   radeon_miptree_unreference(&t->mt);
+
    _mesa_delete_texture_object(ctx, texObj);
 }
 
diff --git a/src/mesa/drivers/dri/r200/r200_texstate.c b/src/mesa/drivers/dri/r200/r200_texstate.c
index 7d0afa1add..7782404a79 100644
--- a/src/mesa/drivers/dri/r200/r200_texstate.c
+++ b/src/mesa/drivers/dri/r200/r200_texstate.c
@@ -824,14 +824,10 @@ void r200SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_fo
 		radeon_bo_unref(rImage->bo);
 		rImage->bo = NULL;
 	}
-	if (t->mt) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = NULL;
-	}
-	if (rImage->mt) {
-		radeon_miptree_unreference(rImage->mt);
-		rImage->mt = NULL;
-	}
+
+	radeon_miptree_unreference(&t->mt);
+	radeon_miptree_unreference(&rImage->mt);
+
 	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
 				   rb->base.Width, rb->base.Height, 1, 0, rb->cpp);
 	texImage->RowStride = rb->pitch / rb->cpp;
@@ -1423,10 +1419,9 @@ void set_re_cntl_d3d( GLcontext *ctx, int unit, GLboolean use_d3d )
  */
 static void setup_hardware_state(r200ContextPtr rmesa, radeonTexObj *t)
 {
-   int firstlevel = t->mt ? t->mt->firstLevel : 0;
-   const struct gl_texture_image *firstImage = t->base.Image[0][firstlevel];
+   const struct gl_texture_image *firstImage = t->base.Image[0][t->minLod];
    GLint log2Width, log2Height, log2Depth, texelBytes;
-   
+
    if ( t->bo ) {
        return;
    }
@@ -1454,9 +1449,9 @@ static void setup_hardware_state(r200ContextPtr rmesa, radeonTexObj *t)
 	 return;
       }
    }
-   
+
    t->pp_txfilter &= ~R200_MAX_MIP_LEVEL_MASK;
-   t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << R200_MAX_MIP_LEVEL_SHIFT;
+   t->pp_txfilter |= (t->maxLod - t->minLod) << R200_MAX_MIP_LEVEL_SHIFT;
 	
    t->pp_txformat &= ~(R200_TXFORMAT_WIDTH_MASK |
 		       R200_TXFORMAT_HEIGHT_MASK |
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index c7996eb76a..ad8db6e68e 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -46,14 +46,12 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "r300_context.h"
 #include "r300_ioctl.h"
-#include "radeon_reg.h"
 #include "r300_reg.h"
 #include "r300_cmdbuf.h"
 #include "r300_emit.h"
 #include "radeon_bocs_wrapper.h"
 #include "radeon_mipmap_tree.h"
 #include "r300_state.h"
-#include "radeon_reg.h"
 #include "radeon_queryobj.h"
 
 /** # of dwords reserved for additional instructions that may need to be written
@@ -171,7 +169,7 @@ static void emit_tex_offsets(GLcontext *ctx, struct radeon_state_atom * atom)
 		if (t && !t->image_override) {
 			BEGIN_BATCH_NO_AUTOSTATE(4);
 			OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1);
-			OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+			OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, get_base_teximage_offset(t),
 					RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
 			END_BATCH();
 		} else if (!t) {
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index 6f66e970e4..5f07b95634 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -439,11 +439,11 @@ static void r300InitGLExtensions(GLcontext *ctx)
 	if (r300->options.stencil_two_side_disabled)
 		_mesa_disable_extension(ctx, "GL_EXT_stencil_two_side");
 
-	if (r300->options.s3tc_force_enabled) {
+	if (r300->options.s3tc_force_disabled) {
+		_mesa_disable_extension(ctx, "GL_EXT_texture_compression_s3tc");
+	} else if (ctx->Mesa_DXTn || r300->options.s3tc_force_enabled) {
 		_mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
 		_mesa_enable_extension(ctx, "GL_S3_s3tc");
-	} else if (r300->options.s3tc_force_disabled) {
-		_mesa_disable_extension(ctx, "GL_EXT_texture_compression_s3tc");
 	}
 
 	if (!r300->radeon.radeonScreen->drmSupportsOcclusionQueries) {
diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h
index 8e57e354d1..a456d8867c 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.h
+++ b/src/mesa/drivers/dri/r300/r300_emit.h
@@ -42,7 +42,6 @@
 #include "main/glheader.h"
 #include "r300_context.h"
 #include "r300_cmdbuf.h"
-#include "radeon_reg.h"
 
 static INLINE uint32_t cmdpacket0(struct radeon_screen *rscrn,
                                   int reg, int count)
diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index 3cd38753b8..4ae593cbe7 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -67,8 +67,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "vbo/vbo_split.h"
 #include "tnl/tnl.h"
 #include "tnl/t_vp_build.h"
-#include "radeon_reg.h"
-#include "radeon_macros.h"
 #include "r300_context.h"
 #include "r300_ioctl.h"
 #include "r300_state.h"
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index ac20c08e20..1fd32d497b 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1741,7 +1741,8 @@ static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state)
 		r300SetPolygonOffsetState(ctx, state);
 		break;
 	case GL_SCISSOR_TEST:
-		radeon_firevertices(&rmesa->radeon);
+		if (!rmesa->radeon.radeonScreen->kernel_mm)
+			radeon_firevertices(&rmesa->radeon);
 		rmesa->radeon.state.scissor.enabled = state;
 		radeonUpdateScissor( ctx );
 		break;
diff --git a/src/mesa/drivers/dri/r300/r300_tex.c b/src/mesa/drivers/dri/r300/r300_tex.c
index 27b78a912f..726b3ff98e 100644
--- a/src/mesa/drivers/dri/r300/r300_tex.c
+++ b/src/mesa/drivers/dri/r300/r300_tex.c
@@ -223,16 +223,7 @@ static void r300TexParameter(GLcontext * ctx, GLenum target,
 	case GL_TEXTURE_MAX_LEVEL:
 	case GL_TEXTURE_MIN_LOD:
 	case GL_TEXTURE_MAX_LOD:
-		/* This isn't the most efficient solution but there doesn't appear to
-		 * be a nice alternative.  Since there's no LOD clamping,
-		 * we just have to rely on loading the right subset of mipmap levels
-		 * to simulate a clamped LOD.
-		 */
-		if (t->mt) {
-			radeon_miptree_unreference(t->mt);
-			t->mt = 0;
-			t->validated = GL_FALSE;
-		}
+		t->validated = GL_FALSE;
 		break;
 
 	case GL_DEPTH_TEXTURE_MODE:
@@ -270,7 +261,11 @@ static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 
 	if (rmesa) {
 		int i;
-		radeon_firevertices(&rmesa->radeon);
+		struct radeon_bo *bo;
+		bo = !t->mt ? t->bo : t->mt->bo;
+		if (bo && radeon_bo_is_referenced_by_cs(bo, rmesa->radeon.cmdbuf.cs)) {
+			radeon_firevertices(&rmesa->radeon);
+		}
 
 		for(i = 0; i < R300_MAX_TEXTURE_UNITS; ++i)
 			if (rmesa->hw.textures[i] == t)
@@ -282,10 +277,8 @@ static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 		t->bo = NULL;
 	}
 
-	if (t->mt) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = 0;
-	}
+	radeon_miptree_unreference(&t->mt);
+
 	_mesa_delete_texture_object(ctx, texObj);
 }
 
diff --git a/src/mesa/drivers/dri/r300/r300_texstate.c b/src/mesa/drivers/dri/r300/r300_texstate.c
index 44ca24daf8..e6f2c0c1a7 100644
--- a/src/mesa/drivers/dri/r300/r300_texstate.c
+++ b/src/mesa/drivers/dri/r300/r300_texstate.c
@@ -83,6 +83,7 @@ static const struct tx_table {
 	_ASSIGN(ARGB8888, R300_EASY_TX_FORMAT(W, Z, Y, X, W8Z8Y8X8)),
 	_ASSIGN(ARGB8888_REV, R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8)),
 #endif
+	_ASSIGN(XRGB8888, R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8)),
 	_ASSIGN(RGB888, R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8)),
 	_ASSIGN(RGB565, R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5)),
 	_ASSIGN(RGB565_REV, R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5)),
@@ -202,9 +203,7 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj)
 static void setup_hardware_state(r300ContextPtr rmesa, radeonTexObj *t)
 {
 	const struct gl_texture_image *firstImage;
-	int firstlevel = t->mt ? t->mt->firstLevel : 0;
-	    
-	firstImage = t->base.Image[0][firstlevel];
+	firstImage = t->base.Image[0][t->minLod];
 
 	if (!t->image_override
 	    && VALID_FORMAT(firstImage->TexFormat)) {
@@ -227,7 +226,7 @@ static void setup_hardware_state(r300ContextPtr rmesa, radeonTexObj *t)
 	t->pp_txsize = (((R300_TX_WIDTHMASK_MASK & ((firstImage->Width - 1) << R300_TX_WIDTHMASK_SHIFT)))
 			| ((R300_TX_HEIGHTMASK_MASK & ((firstImage->Height - 1) << R300_TX_HEIGHTMASK_SHIFT)))
 			| ((R300_TX_DEPTHMASK_MASK & ((firstImage->DepthLog2) << R300_TX_DEPTHMASK_SHIFT)))
-			| ((R300_TX_MAX_MIP_LEVEL_MASK & ((t->mt->lastLevel - t->mt->firstLevel) << R300_TX_MAX_MIP_LEVEL_SHIFT))));
+			| ((R300_TX_MAX_MIP_LEVEL_MASK & ((t->maxLod - t->minLod) << R300_TX_MAX_MIP_LEVEL_SHIFT))));
 
 	t->tile_bits = 0;
 
@@ -238,7 +237,7 @@ static void setup_hardware_state(r300ContextPtr rmesa, radeonTexObj *t)
 
 
 	if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) {
-		unsigned int align = (64 / t->mt->bpp) - 1;
+		unsigned int align = (64 / _mesa_get_format_bytes(firstImage->TexFormat)) - 1;
 		t->pp_txsize |= R300_TX_SIZE_TXPITCH_EN;
 		if (!t->image_override)
 			t->pp_txpitch = ((firstImage->Width + align) & ~align) - 1;
@@ -437,14 +436,10 @@ void r300SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_fo
 		radeon_bo_unref(rImage->bo);
 		rImage->bo = NULL;
 	}
-	if (t->mt) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = NULL;
-	}
-	if (rImage->mt) {
-		radeon_miptree_unreference(rImage->mt);
-		rImage->mt = NULL;
-	}
+
+	radeon_miptree_unreference(&t->mt);
+	radeon_miptree_unreference(&rImage->mt);
+
 	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
 				   rb->base.Width, rb->base.Height, 1, 0, rb->cpp);
 	texImage->RowStride = rb->pitch / rb->cpp;
diff --git a/src/mesa/drivers/dri/r600/r600_tex.c b/src/mesa/drivers/dri/r600/r600_tex.c
index 20965bb3c8..9d83a64e22 100644
--- a/src/mesa/drivers/dri/r600/r600_tex.c
+++ b/src/mesa/drivers/dri/r600/r600_tex.c
@@ -312,16 +312,7 @@ static void r600TexParameter(GLcontext * ctx, GLenum target,
 	case GL_TEXTURE_MAX_LEVEL:
 	case GL_TEXTURE_MIN_LOD:
 	case GL_TEXTURE_MAX_LOD:
-		/* This isn't the most efficient solution but there doesn't appear to
-		 * be a nice alternative.  Since there's no LOD clamping,
-		 * we just have to rely on loading the right subset of mipmap levels
-		 * to simulate a clamped LOD.
-		 */
-		if (t->mt) {
-			radeon_miptree_unreference(t->mt);
-			t->mt = 0;
-			t->validated = GL_FALSE;
-		}
+		t->validated = GL_FALSE;
 		break;
 
 	case GL_DEPTH_TEXTURE_MODE:
@@ -369,10 +360,8 @@ static void r600DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj)
 		t->bo = NULL;
 	}
 
-	if (t->mt) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = 0;
-	}
+	radeon_miptree_unreference(&t->mt);
+
 	_mesa_delete_texture_object(ctx, texObj);
 }
 
diff --git a/src/mesa/drivers/dri/r600/r600_texstate.c b/src/mesa/drivers/dri/r600/r600_texstate.c
index 27c8354923..4ec315b78c 100644
--- a/src/mesa/drivers/dri/r600/r600_texstate.c
+++ b/src/mesa/drivers/dri/r600/r600_texstate.c
@@ -649,7 +649,6 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 {
 	radeonTexObj *t = radeon_tex_obj(texObj);
 	const struct gl_texture_image *firstImage;
-	int firstlevel = t->mt ? t->mt->firstLevel : 0;
 	GLuint uTexelPitch, row_align;
 
 	if (rmesa->radeon.radeonScreen->driScreen->dri2.enabled &&
@@ -657,7 +656,7 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 	    t->bo)
 		return;
 
-	firstImage = t->base.Image[0][firstlevel];
+	firstImage = t->base.Image[0][t->minLod];
 
 	if (!t->image_override) {
 		if (!r600GetTexFormat(texObj, firstImage->TexFormat)) {
@@ -692,7 +691,8 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 	}
 
 	row_align = rmesa->radeon.texture_row_align - 1;
-	uTexelPitch = ((firstImage->Width * t->mt->bpp + row_align) & ~row_align) / t->mt->bpp;
+	uTexelPitch = (_mesa_format_row_stride(firstImage->TexFormat, firstImage->Width) + row_align) & ~row_align;
+	uTexelPitch = uTexelPitch / _mesa_get_format_bytes(firstImage->TexFormat);
 	uTexelPitch = (uTexelPitch + R700_TEXEL_PITCH_ALIGNMENT_MASK)
 		& ~R700_TEXEL_PITCH_ALIGNMENT_MASK;
 
@@ -706,10 +706,10 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 	SETfield(t->SQ_TEX_RESOURCE1, firstImage->Height - 1,
 		 TEX_HEIGHT_shift, TEX_HEIGHT_mask);
 
-	if ((t->mt->lastLevel - t->mt->firstLevel) > 0) {
-		t->SQ_TEX_RESOURCE3 = t->mt->levels[0].size / 256;
-		SETfield(t->SQ_TEX_RESOURCE4, t->mt->firstLevel, BASE_LEVEL_shift, BASE_LEVEL_mask);
-		SETfield(t->SQ_TEX_RESOURCE5, t->mt->lastLevel, LAST_LEVEL_shift, LAST_LEVEL_mask);
+	if ((t->maxLod - t->minLod) > 0) {
+		t->SQ_TEX_RESOURCE3 = t->mt->levels[t->minLod].size / 256;
+		SETfield(t->SQ_TEX_RESOURCE4, 0, BASE_LEVEL_shift, BASE_LEVEL_mask);
+		SETfield(t->SQ_TEX_RESOURCE5, t->maxLod - t->minLod, LAST_LEVEL_shift, LAST_LEVEL_mask);
 	}
 }
 
@@ -808,9 +808,8 @@ void r600SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 	struct gl_texture_object *tObj =
 	    _mesa_lookup_texture(rmesa->radeon.glCtx, texname);
 	radeonTexObjPtr t = radeon_tex_obj(tObj);
-	int firstlevel = t->mt ? t->mt->firstLevel : 0;
 	const struct gl_texture_image *firstImage;
-	uint32_t pitch_val, size, row_align, bpp;
+	uint32_t pitch_val, size, row_align;
 
 	if (!tObj)
 		return;
@@ -820,13 +819,9 @@ void r600SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 	if (!offset)
 		return;
 
-	bpp = depth / 8;
-	if (bpp == 3) 
-		bpp = 4;
-
-	firstImage = t->base.Image[0][firstlevel];
+	firstImage = t->base.Image[0][t->minLod];
 	row_align = rmesa->radeon.texture_row_align - 1;
-	size = ((firstImage->Width * bpp + row_align) & ~row_align) * firstImage->Height;
+	size = ((_mesa_format_row_stride(firstImage->TexFormat, firstImage->Width) + row_align) & ~row_align) * firstImage->Height;
 	if (t->bo) {
 		radeon_bo_unref(t->bo);
 		t->bo = NULL;
@@ -949,14 +944,10 @@ void r600SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_fo
 		radeon_bo_unref(rImage->bo);
 		rImage->bo = NULL;
 	}
-	if (t->mt) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = NULL;
-	}
-	if (rImage->mt) {
-		radeon_miptree_unreference(rImage->mt);
-		rImage->mt = NULL;
-	}
+
+	radeon_miptree_unreference(&t->mt);
+	radeon_miptree_unreference(&rImage->mt);
+
 	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
 				   rb->base.Width, rb->base.Height, 1, 0, rb->cpp);
 	texImage->RowStride = rb->pitch / rb->cpp;
diff --git a/src/mesa/drivers/dri/r600/r700_chip.c b/src/mesa/drivers/dri/r600/r700_chip.c
index 705b5738ed..8126777bf4 100644
--- a/src/mesa/drivers/dri/r600/r700_chip.c
+++ b/src/mesa/drivers/dri/r600/r700_chip.c
@@ -54,11 +54,15 @@ static void r700SendTexState(GLcontext *ctx, struct radeon_state_atom *atom)
 	for (i = 0; i < R700_TEXTURE_NUMBERUNITS; i++) {
 		if (ctx->Texture.Unit[i]._ReallyEnabled) {
 			radeonTexObj *t = r700->textures[i];
+			uint32_t offset;
 			if (t) {
-				if (!t->image_override)
+				if (!t->image_override) {
 					bo = t->mt->bo;
-				else
+					offset = get_base_teximage_offset(t);
+				} else {
 					bo = t->bo;
+					offset = 0;
+				}
 				if (bo) {
 
 					r700SyncSurf(context, bo,
@@ -77,7 +81,7 @@ static void r700SendTexState(GLcontext *ctx, struct radeon_state_atom *atom)
 					R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE6);
 					R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE2,
 							     bo,
-							     0,
+							     offset,
 							     RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
 					R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE3,
 							     bo,
diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_drm.h b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h
index 7141371633..46e30b905a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_bo_drm.h
+++ b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h
@@ -39,6 +39,7 @@
 #define RADEON_BO_FLAGS_MICRO_TILE  2
 
 struct radeon_bo_manager;
+struct radeon_cs;
 
 struct radeon_bo {
     uint32_t                    alignment;
@@ -74,6 +75,7 @@ struct radeon_bo_funcs {
     int (*bo_get_tiling)(struct radeon_bo *bo, uint32_t *tiling_flags,
 			  uint32_t *pitch);
     int (*bo_is_busy)(struct radeon_bo *bo, uint32_t *domain);
+    int (*bo_is_referenced_by_cs)(struct radeon_bo *bo, struct radeon_cs *cs);
 };
 
 struct radeon_bo_manager {
@@ -199,6 +201,15 @@ static inline int radeon_bo_is_static(struct radeon_bo *bo)
 	return 0;
 }
 
+static inline int _radeon_bo_is_referenced_by_cs(struct radeon_bo *bo,
+                                                 struct radeon_cs *cs,
+                                                 const char *file,
+                                                 const char *func,
+                                                 unsigned line)
+{
+    return bo->cref > 1;
+}
+
 #define radeon_bo_open(bom, h, s, a, d, f)\
     _radeon_bo_open(bom, h, s, a, d, f, __FILE__, __FUNCTION__, __LINE__)
 #define radeon_bo_ref(bo)\
@@ -215,5 +226,7 @@ static inline int radeon_bo_is_static(struct radeon_bo *bo)
     _radeon_bo_wait(bo, __FILE__, __func__, __LINE__)
 #define radeon_bo_is_busy(bo, domain) \
     _radeon_bo_is_busy(bo, domain, __FILE__, __func__, __LINE__)
+#define radeon_bo_is_referenced_by_cs(bo, cs) \
+    _radeon_bo_is_referenced_by_cs(bo, cs, __FILE__, __FUNCTION__, __LINE__)
 
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c
index 8fac5c6c51..99d3ec7005 100644
--- a/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c
+++ b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c
@@ -136,8 +136,13 @@ radeonBufferSubData(GLcontext * ctx,
                     const GLvoid * data,
                     struct gl_buffer_object *obj)
 {
+    radeonContextPtr radeon = RADEON_CONTEXT(ctx);
     struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj);
 
+    if (radeon_bo_is_referenced_by_cs(radeon_obj->bo, radeon->cmdbuf.cs)) {
+        radeon_firevertices(radeon);
+    }
+
     radeon_bo_map(radeon_obj->bo, GL_TRUE);
 
     _mesa_memcpy(radeon_obj->bo->ptr + offset, data, size);
diff --git a/src/mesa/drivers/dri/radeon/radeon_common.c b/src/mesa/drivers/dri/radeon/radeon_common.c
index 097ab7cf61..3b4366aa61 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common.c
+++ b/src/mesa/drivers/dri/radeon/radeon_common.c
@@ -257,7 +257,9 @@ void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h)
 	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 	if (ctx->Scissor.Enabled) {
 		/* We don't pipeline cliprect changes */
-		radeon_firevertices(radeon);
+		if (!radeon->radeonScreen->kernel_mm) {
+			radeon_firevertices(radeon);
+		}
 		radeonUpdateScissor(ctx);
 	}
 }
@@ -1123,8 +1125,6 @@ void radeonFlush(GLcontext *ctx)
 	if (radeon->dma.flush)
 		radeon->dma.flush( ctx );
 
-	radeonEmitState(radeon);
-
 	if (radeon->cmdbuf.cs->cdw)
 		rcommonFlushCmdBuf(radeon, __FUNCTION__);
 
@@ -1147,9 +1147,6 @@ void radeonFlush(GLcontext *ctx)
 			}
 		}
 	}
-
-	make_empty_list(&radeon->query.not_flushed_head);
-
 }
 
 /* Make sure all commands have been sent to the hardware and have
diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.c b/src/mesa/drivers/dri/radeon/radeon_common_context.c
index fe99644907..2a38c4599c 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common_context.c
+++ b/src/mesa/drivers/dri/radeon/radeon_common_context.c
@@ -265,7 +265,6 @@ GLboolean radeonInitContext(radeonContextPtr radeon,
 		radeon->texture_compressed_row_align = 64;
 	}
 
-	make_empty_list(&radeon->query.not_flushed_head);
 	radeon_init_dma(radeon);
 
 	return GL_TRUE;
diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.h b/src/mesa/drivers/dri/radeon/radeon_common_context.h
index 0309345393..ded81fff29 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common_context.h
+++ b/src/mesa/drivers/dri/radeon/radeon_common_context.h
@@ -208,6 +208,10 @@ struct radeon_tex_obj {
 	 * and so on.
 	 */
 	GLboolean validated;
+	/* Minimum LOD to be used during rendering */
+	unsigned minLod;
+	/* Miximum LOD to be used during rendering */
+	unsigned maxLod;
 
 	GLuint override_offset;
 	GLboolean image_override; /* Image overridden by GLX_EXT_tfp */
@@ -502,7 +506,6 @@ struct radeon_context {
 
    struct {
 	struct radeon_query_object *current;
-	struct radeon_query_object not_flushed_head;
 	struct radeon_state_atom queryobj;
    } query;
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
index dadc72f4c1..0497fa7db5 100644
--- a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
+++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (C) 2009 Maciej Cencora.
  * Copyright (C) 2008 Nicolai Haehnle.
  *
  * All Rights Reserved.
@@ -32,10 +33,13 @@
 
 #include "main/simple_list.h"
 #include "main/texcompress.h"
+#include "main/teximage.h"
+#include "main/texobj.h"
+#include "radeon_texture.h"
 
 static GLuint radeon_compressed_texture_size(GLcontext *ctx,
 		GLsizei width, GLsizei height, GLsizei depth,
-		GLuint mesaFormat)
+		gl_format mesaFormat)
 {
 	GLuint size = _mesa_format_image_size(mesaFormat, width, height, depth);
 
@@ -55,29 +59,6 @@ static GLuint radeon_compressed_texture_size(GLcontext *ctx,
 	return size;
 }
 
-
-static int radeon_compressed_num_bytes(GLuint mesaFormat)
-{
-   int bytes = 0;
-   switch(mesaFormat) {
-     
-   case MESA_FORMAT_RGB_FXT1:
-   case MESA_FORMAT_RGBA_FXT1:
-   case MESA_FORMAT_RGB_DXT1:
-   case MESA_FORMAT_RGBA_DXT1:
-     bytes = 2;
-     break;
-     
-   case MESA_FORMAT_RGBA_DXT3:
-   case MESA_FORMAT_RGBA_DXT5:
-     bytes = 4;
-   default:
-     break;
-   }
-   
-   return bytes;
-}
-
 /**
  * Compute sizes and fill in offset and blit information for the given
  * image (determined by \p face and \p level).
@@ -92,25 +73,24 @@ static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree
 	uint32_t row_align;
 
 	/* Find image size in bytes */
-	if (mt->compressed) {
+	if (_mesa_is_format_compressed(mt->mesaFormat)) {
 		/* TODO: Is this correct? Need test cases for compressed textures! */
 		row_align = rmesa->texture_compressed_row_align - 1;
-		lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align;
-		lvl->size = radeon_compressed_texture_size(mt->radeon->glCtx,
-							   lvl->width, lvl->height, lvl->depth, mt->compressed);
+		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
+		lvl->size = radeon_compressed_texture_size(rmesa->glCtx, lvl->width, lvl->height, lvl->depth, mt->mesaFormat);
 	} else if (mt->target == GL_TEXTURE_RECTANGLE_NV) {
 		row_align = rmesa->texture_rect_row_align - 1;
-		lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align;
+		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
 		lvl->size = lvl->rowstride * lvl->height;
 	} else if (mt->tilebits & RADEON_TXO_MICRO_TILE) {
 		/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
 		 * though the actual offset may be different (if texture is less than
 		 * 32 bytes width) to the untiled case */
-		lvl->rowstride = (lvl->width * mt->bpp * 2 + 31) & ~31;
+		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) * 2 + 31) & ~31;
 		lvl->size = lvl->rowstride * ((lvl->height + 1) / 2) * lvl->depth;
 	} else {
 		row_align = rmesa->texture_row_align - 1;
-		lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align;
+		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
 		lvl->size = lvl->rowstride * lvl->height * lvl->depth;
 	}
 	assert(lvl->size > 0);
@@ -138,17 +118,15 @@ static GLuint minify(GLuint size, GLuint levels)
 static void calculate_miptree_layout_r100(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
 {
 	GLuint curOffset;
-	GLuint numLevels;
 	GLuint i;
 	GLuint face;
 
-	numLevels = mt->lastLevel - mt->firstLevel + 1;
-	assert(numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
+	assert(mt->numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
 
 	curOffset = 0;
 	for(face = 0; face < mt->faces; face++) {
 
-		for(i = 0; i < numLevels; i++) {
+		for(i = 0; i < mt->numLevels; i++) {
 			mt->levels[i].width = minify(mt->width0, i);
 			mt->levels[i].height = minify(mt->height0, i);
 			mt->levels[i].depth = minify(mt->depth0, i);
@@ -163,14 +141,12 @@ static void calculate_miptree_layout_r100(radeonContextPtr rmesa, radeon_mipmap_
 static void calculate_miptree_layout_r300(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
 {
 	GLuint curOffset;
-	GLuint numLevels;
 	GLuint i;
 
-	numLevels = mt->lastLevel - mt->firstLevel + 1;
-	assert(numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
+	assert(mt->numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
 
 	curOffset = 0;
-	for(i = 0; i < numLevels; i++) {
+	for(i = 0; i < mt->numLevels; i++) {
 		GLuint face;
 
 		mt->levels[i].width = minify(mt->width0, i);
@@ -188,27 +164,22 @@ static void calculate_miptree_layout_r300(radeonContextPtr rmesa, radeon_mipmap_
 /**
  * Create a new mipmap tree, calculate its layout and allocate memory.
  */
-radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *t,
-		GLenum target, GLenum internal_format, GLuint firstLevel, GLuint lastLevel,
-		GLuint width0, GLuint height0, GLuint depth0,
-		GLuint bpp, GLuint tilebits, GLuint compressed)
+static radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa,
+		GLenum target, gl_format mesaFormat, GLuint baseLevel, GLuint numLevels,
+		GLuint width0, GLuint height0, GLuint depth0, GLuint tilebits)
 {
 	radeon_mipmap_tree *mt = CALLOC_STRUCT(_radeon_mipmap_tree);
 
-	mt->radeon = rmesa;
-	mt->internal_format = internal_format;
+	mt->mesaFormat = mesaFormat;
 	mt->refcount = 1;
-	mt->t = t;
 	mt->target = target;
 	mt->faces = (target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
-	mt->firstLevel = firstLevel;
-	mt->lastLevel = lastLevel;
+	mt->baseLevel = baseLevel;
+	mt->numLevels = numLevels;
 	mt->width0 = width0;
 	mt->height0 = height0;
 	mt->depth0 = depth0;
-	mt->bpp = compressed ? radeon_compressed_num_bytes(compressed) : bpp;
 	mt->tilebits = tilebits;
-	mt->compressed = compressed;
 
 	if (rmesa->radeonScreen->chip_family >= CHIP_FAMILY_R300)
 		calculate_miptree_layout_r300(rmesa, mt);
@@ -223,53 +194,43 @@ radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *
 	return mt;
 }
 
-void radeon_miptree_reference(radeon_mipmap_tree *mt)
+void radeon_miptree_reference(radeon_mipmap_tree *mt, radeon_mipmap_tree **ptr)
 {
+	assert(!*ptr);
+
 	mt->refcount++;
 	assert(mt->refcount > 0);
+
+	*ptr = mt;
 }
 
-void radeon_miptree_unreference(radeon_mipmap_tree *mt)
+void radeon_miptree_unreference(radeon_mipmap_tree **ptr)
 {
+	radeon_mipmap_tree *mt = *ptr;
 	if (!mt)
 		return;
 
 	assert(mt->refcount > 0);
+
 	mt->refcount--;
 	if (!mt->refcount) {
 		radeon_bo_unref(mt->bo);
 		free(mt);
 	}
-}
 
+	*ptr = 0;
+}
 
 /**
- * Calculate first and last mip levels for the given texture object,
- * where the dimensions are taken from the given texture image at
- * the given level.
- *
- * Note: level is the OpenGL level number, which is not necessarily the same
- * as the first level that is actually present.
- *
- * The base level image of the given texture face must be non-null,
- * or this will fail.
+ * Calculate min and max LOD for the given texture object.
+ * @param[in] tObj texture object whose LOD values to calculate
+ * @param[out] pminLod minimal LOD
+ * @param[out] pmaxLod maximal LOD
  */
-static void calculate_first_last_level(struct gl_texture_object *tObj,
-				       GLuint *pfirstLevel, GLuint *plastLevel,
-				       GLuint face, GLuint level)
+static void calculate_min_max_lod(struct gl_texture_object *tObj,
+				       unsigned *pminLod, unsigned *pmaxLod)
 {
-	const struct gl_texture_image * const baseImage =
-		tObj->Image[face][level];
-
-	assert(baseImage);
-	
-	/* These must be signed values.  MinLod and MaxLod can be negative numbers,
-	* and having firstLevel and lastLevel as signed prevents the need for
-	* extra sign checks.
-	*/
-	int   firstLevel;
-	int   lastLevel;
-
+	int minLod, maxLod;
 	/* Yes, this looks overly complicated, but it's all needed.
 	*/
 	switch (tObj->Target) {
@@ -280,55 +241,46 @@ static void calculate_first_last_level(struct gl_texture_object *tObj,
 		if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) {
 			/* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
 			*/
-			firstLevel = lastLevel = tObj->BaseLevel;
+			minLod = maxLod = tObj->BaseLevel;
 		} else {
-			firstLevel = tObj->BaseLevel + (GLint)(tObj->MinLod + 0.5);
-			firstLevel = MAX2(firstLevel, tObj->BaseLevel);
-			firstLevel = MIN2(firstLevel, level + baseImage->MaxLog2);
-			lastLevel = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5);
-			lastLevel = MAX2(lastLevel, tObj->BaseLevel);
-			lastLevel = MIN2(lastLevel, level + baseImage->MaxLog2);
-			lastLevel = MIN2(lastLevel, tObj->MaxLevel);
-			lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
+			minLod = tObj->BaseLevel + (GLint)(tObj->MinLod);
+			minLod = MAX2(minLod, tObj->BaseLevel);
+			minLod = MIN2(minLod, tObj->MaxLevel);
+			maxLod = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5);
+			maxLod = MIN2(maxLod, tObj->MaxLevel);
+			maxLod = MIN2(maxLod, tObj->Image[0][minLod]->MaxLog2 + minLod);
+			maxLod = MAX2(maxLod, minLod); /* need at least one level */
 		}
 		break;
 	case GL_TEXTURE_RECTANGLE_NV:
 	case GL_TEXTURE_4D_SGIS:
-		firstLevel = lastLevel = 0;
+		minLod = maxLod = 0;
 		break;
 	default:
 		return;
 	}
 
 	/* save these values */
-	*pfirstLevel = firstLevel;
-	*plastLevel = lastLevel;
+	*pminLod = minLod;
+	*pmaxLod = maxLod;
 }
 
-
 /**
  * Checks whether the given miptree can hold the given texture image at the
  * given face and level.
  */
 GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
-		struct gl_texture_image *texImage, GLuint face, GLuint level)
+		struct gl_texture_image *texImage, GLuint face, GLuint mtLevel)
 {
-	GLboolean isCompressed = _mesa_is_format_compressed(texImage->TexFormat);
 	radeon_mipmap_level *lvl;
 
-	if (face >= mt->faces || level < mt->firstLevel || level > mt->lastLevel)
-		return GL_FALSE;
-
-	if (texImage->InternalFormat != mt->internal_format ||
-	    isCompressed != mt->compressed)
+	if (face >= mt->faces || mtLevel > mt->numLevels)
 		return GL_FALSE;
 
-	if (!isCompressed &&
-	    !mt->compressed &&
-	    _mesa_get_format_bytes(texImage->TexFormat) != mt->bpp)
+	if (texImage->TexFormat != mt->mesaFormat)
 		return GL_FALSE;
 
-	lvl = &mt->levels[level - mt->firstLevel];
+	lvl = &mt->levels[mtLevel];
 	if (lvl->width != texImage->Width ||
 	    lvl->height != texImage->Height ||
 	    lvl->depth != texImage->Depth)
@@ -337,64 +289,72 @@ GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
 	return GL_TRUE;
 }
 
-
 /**
  * Checks whether the given miptree has the right format to store the given texture object.
  */
-GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj)
+static GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj)
 {
 	struct gl_texture_image *firstImage;
-	GLuint compressed;
-	GLuint numfaces = 1;
-	GLuint firstLevel, lastLevel;
-	GLuint texelBytes;
-
-	calculate_first_last_level(texObj, &firstLevel, &lastLevel, 0, texObj->BaseLevel);
-	if (texObj->Target == GL_TEXTURE_CUBE_MAP)
-		numfaces = 6;
-
-	firstImage = texObj->Image[0][firstLevel];
-	compressed = _mesa_is_format_compressed(firstImage->TexFormat) ? firstImage->TexFormat : 0;
-	texelBytes = _mesa_get_format_bytes(firstImage->TexFormat);
-
-	return (mt->firstLevel == firstLevel &&
-	        mt->lastLevel == lastLevel &&
-	        mt->width0 == firstImage->Width &&
-	        mt->height0 == firstImage->Height &&
-	        mt->depth0 == firstImage->Depth &&
-	        mt->compressed == compressed &&
-	        (!mt->compressed ? (mt->bpp == texelBytes) : 1));
-}
+	unsigned numLevels;
+	radeon_mipmap_level *mtBaseLevel;
 
+	if (texObj->BaseLevel < mt->baseLevel)
+		return GL_FALSE;
+
+	mtBaseLevel = &mt->levels[texObj->BaseLevel - mt->baseLevel];
+	firstImage = texObj->Image[0][texObj->BaseLevel];
+	numLevels = MIN2(texObj->MaxLevel - texObj->BaseLevel + 1, firstImage->MaxLog2 + 1);
+
+	if (RADEON_DEBUG & RADEON_TEXTURE) {
+		fprintf(stderr, "Checking if miptree %p matches texObj %p\n", mt, texObj);
+		fprintf(stderr, "target %d vs %d\n", mt->target, texObj->Target);
+		fprintf(stderr, "format %d vs %d\n", mt->mesaFormat, firstImage->TexFormat);
+		fprintf(stderr, "numLevels %d vs %d\n", mt->numLevels, numLevels);
+		fprintf(stderr, "width0 %d vs %d\n", mtBaseLevel->width, firstImage->Width);
+		fprintf(stderr, "height0 %d vs %d\n", mtBaseLevel->height, firstImage->Height);
+		fprintf(stderr, "depth0 %d vs %d\n", mtBaseLevel->depth, firstImage->Depth);
+		if (mt->target == texObj->Target &&
+	        mt->mesaFormat == firstImage->TexFormat &&
+	        mt->numLevels >= numLevels &&
+	        mtBaseLevel->width == firstImage->Width &&
+	        mtBaseLevel->height == firstImage->Height &&
+	        mtBaseLevel->depth == firstImage->Depth) {
+			fprintf(stderr, "MATCHED\n");
+		} else {
+			fprintf(stderr, "NOT MATCHED\n");
+		}
+	}
+
+	return (mt->target == texObj->Target &&
+	        mt->mesaFormat == firstImage->TexFormat &&
+	        mt->numLevels >= numLevels &&
+	        mtBaseLevel->width == firstImage->Width &&
+	        mtBaseLevel->height == firstImage->Height &&
+	        mtBaseLevel->depth == firstImage->Depth);
+}
 
 /**
- * Try to allocate a mipmap tree for the given texture that will fit the
- * given image in the given position.
+ * Try to allocate a mipmap tree for the given texture object.
+ * @param[in] rmesa radeon context
+ * @param[in] t radeon texture object
  */
-void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t,
-		radeon_texture_image *image, GLuint face, GLuint level)
+void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t)
 {
-	GLuint compressed = _mesa_is_format_compressed(image->base.TexFormat) ? image->base.TexFormat : 0;
-	GLuint numfaces = 1;
-	GLuint firstLevel, lastLevel;
-	GLuint texelBytes;
+	struct gl_texture_object *texObj = &t->base;
+	struct gl_texture_image *texImg = texObj->Image[0][texObj->BaseLevel];
+	GLuint numLevels;
 
 	assert(!t->mt);
 
-	calculate_first_last_level(&t->base, &firstLevel, &lastLevel, face, level);
-	if (t->base.Target == GL_TEXTURE_CUBE_MAP)
-		numfaces = 6;
-
-	if (level != firstLevel || face >= numfaces)
+	if (!texImg)
 		return;
 
-	texelBytes = _mesa_get_format_bytes(image->base.TexFormat);
+	numLevels = MIN2(texObj->MaxLevel - texObj->BaseLevel + 1, texImg->MaxLog2 + 1);
 
-	t->mt = radeon_miptree_create(rmesa, t, t->base.Target,
-		image->base.InternalFormat,
-		firstLevel, lastLevel,
-		image->base.Width, image->base.Height, image->base.Depth,
-		texelBytes, t->tile_bits, compressed);
+	t->mt = radeon_miptree_create(rmesa, t->base.Target,
+		texImg->TexFormat, texObj->BaseLevel,
+		numLevels, texImg->Width, texImg->Height,
+		texImg->Depth, t->tile_bits);
 }
 
 /* Although we use the image_offset[] array to store relative offsets
@@ -406,21 +366,239 @@ void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t,
 void
 radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets)
 {
-     if (mt->target != GL_TEXTURE_3D || mt->faces == 1)
-        offsets[0] = 0;
-     else {
-	int i;
-	for (i = 0; i < 6; i++)
-		offsets[i] = mt->levels[level].faces[i].offset;
-     }
+	if (mt->target != GL_TEXTURE_3D || mt->faces == 1) {
+		offsets[0] = 0;
+	} else {
+		int i;
+		for (i = 0; i < 6; i++) {
+			offsets[i] = mt->levels[level].faces[i].offset;
+		}
+	}
 }
 
 GLuint
 radeon_miptree_image_offset(radeon_mipmap_tree *mt,
 			    GLuint face, GLuint level)
 {
-   if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
-      return (mt->levels[level].faces[face].offset);
-   else
-      return mt->levels[level].faces[0].offset;
+	if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
+		return (mt->levels[level].faces[face].offset);
+	else
+		return mt->levels[level].faces[0].offset;
+}
+
+/**
+ * Convert radeon miptree texture level to GL texture level
+ * @param[in] tObj texture object whom level is to be converted
+ * @param[in] level radeon miptree texture level
+ * @return GL texture level
+ */
+unsigned radeon_miptree_level_to_gl_level(struct gl_texture_object *tObj, unsigned level)
+{
+	return level + tObj->BaseLevel;
 }
+
+/**
+ * Convert GL texture level to radeon miptree texture level
+ * @param[in] tObj texture object whom level is to be converted
+ * @param[in] level GL texture level
+ * @return radeon miptree texture level
+ */
+unsigned radeon_gl_level_to_miptree_level(struct gl_texture_object *tObj, unsigned level)
+{
+	return level - tObj->BaseLevel;
+}
+
+/**
+ * Ensure that the given image is stored in the given miptree from now on.
+ */
+static void migrate_image_to_miptree(radeon_mipmap_tree *mt,
+									 radeon_texture_image *image,
+									 int face, int mtLevel)
+{
+	radeon_mipmap_level *dstlvl = &mt->levels[mtLevel];
+	unsigned char *dest;
+
+	assert(image->mt != mt);
+	assert(dstlvl->width == image->base.Width);
+	assert(dstlvl->height == image->base.Height);
+	assert(dstlvl->depth == image->base.Depth);
+
+	radeon_bo_map(mt->bo, GL_TRUE);
+	dest = mt->bo->ptr + dstlvl->faces[face].offset;
+
+	if (image->mt) {
+		/* Format etc. should match, so we really just need a memcpy().
+		 * In fact, that memcpy() could be done by the hardware in many
+		 * cases, provided that we have a proper memory manager.
+		 */
+		assert(mt->mesaFormat == image->base.TexFormat);
+
+		radeon_mipmap_level *srclvl = &image->mt->levels[image->mtlevel];
+
+		assert(srclvl->size == dstlvl->size);
+		assert(srclvl->rowstride == dstlvl->rowstride);
+
+		radeon_bo_map(image->mt->bo, GL_FALSE);
+
+		memcpy(dest,
+			image->mt->bo->ptr + srclvl->faces[face].offset,
+			dstlvl->size);
+		radeon_bo_unmap(image->mt->bo);
+
+		radeon_miptree_unreference(&image->mt);
+	} else {
+		/* need to confirm this value is correct */
+		if (_mesa_is_format_compressed(image->base.TexFormat)) {
+			unsigned size = _mesa_format_image_size(image->base.TexFormat,
+													image->base.Width,
+													image->base.Height,
+													image->base.Depth);
+			memcpy(dest, image->base.Data, size);
+		} else {
+			uint32_t srcrowstride;
+			uint32_t height;
+
+			height = image->base.Height * image->base.Depth;
+			srcrowstride = image->base.Width * _mesa_get_format_bytes(image->base.TexFormat);
+			copy_rows(dest, dstlvl->rowstride, image->base.Data, srcrowstride,
+					height, srcrowstride);
+		}
+
+		_mesa_free_texmemory(image->base.Data);
+		image->base.Data = 0;
+	}
+
+	radeon_bo_unmap(mt->bo);
+
+	radeon_miptree_reference(mt, &image->mt);
+	image->mtface = face;
+	image->mtlevel = mtLevel;
+}
+
+/**
+ * Filter matching miptrees, and select one with the most of data.
+ * @param[in] texObj radeon texture object
+ * @param[in] firstLevel first texture level to check
+ * @param[in] lastLevel last texture level to check
+ */
+static radeon_mipmap_tree * get_biggest_matching_miptree(radeonTexObj *texObj,
+														 unsigned firstLevel,
+														 unsigned lastLevel)
+{
+	const unsigned numLevels = lastLevel - firstLevel;
+	unsigned *mtSizes = calloc(numLevels, sizeof(unsigned));
+	radeon_mipmap_tree **mts = calloc(numLevels, sizeof(radeon_mipmap_tree *));
+	unsigned mtCount = 0;
+	unsigned maxMtIndex = 0;
+
+	for (unsigned level = firstLevel; level <= lastLevel; ++level) {
+		radeon_texture_image *img = get_radeon_texture_image(texObj->base.Image[0][level]);
+		unsigned found = 0;
+		// TODO: why this hack??
+		if (!img)
+			break;
+
+		if (!img->mt || !radeon_miptree_matches_texture(img->mt, &texObj->base))
+			continue;
+
+		for (int i = 0; i < mtCount; ++i) {
+			if (mts[i] == img->mt) {
+				found = 1;
+				mtSizes[i] += img->mt->levels[img->mtlevel].size;
+				break;
+			}
+		}
+
+		if (!found) {
+			mtSizes[mtCount] += img->mt->levels[img->mtlevel].size;
+			mts[mtCount++] = img->mt;
+			mtCount++;
+		}
+	}
+
+	if (mtCount == 0) {
+		return NULL;
+	}
+
+	for (int i = 1; i < mtCount; ++i) {
+		if (mtSizes[i] > mtSizes[maxMtIndex]) {
+			maxMtIndex = i;
+		}
+	}
+
+	return mts[maxMtIndex];
+}
+
+/**
+ * Validate texture mipmap tree.
+ * If individual images are stored in different mipmap trees
+ * use the mipmap tree that has the most of the correct data.
+ */
+int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj)
+{
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+	radeonTexObj *t = radeon_tex_obj(texObj);
+
+	if (t->validated || t->image_override) {
+		return GL_TRUE;
+	}
+
+	if (texObj->Image[0][texObj->BaseLevel]->Border > 0)
+		return GL_FALSE;
+
+	_mesa_test_texobj_completeness(rmesa->glCtx, texObj);
+	if (!texObj->_Complete) {
+		return GL_FALSE;
+	}
+
+	calculate_min_max_lod(&t->base, &t->minLod, &t->maxLod);
+
+	if (RADEON_DEBUG & RADEON_TEXTURE)
+		fprintf(stderr, "%s: Validating texture %p now, minLod = %d, maxLod = %d\n",
+				__FUNCTION__, texObj ,t->minLod, t->maxLod);
+
+	radeon_mipmap_tree *dst_miptree;
+	dst_miptree = get_biggest_matching_miptree(t, t->minLod, t->maxLod);
+
+	if (!dst_miptree) {
+		radeon_miptree_unreference(&t->mt);
+		radeon_try_alloc_miptree(rmesa, t);
+		dst_miptree = t->mt;
+	}
+
+	const unsigned faces = texObj->Target == GL_TEXTURE_CUBE_MAP ? 6 : 1;
+	unsigned face, level;
+	radeon_texture_image *img;
+	/* Validate only the levels that will actually be used during rendering */
+	for (face = 0; face < faces; ++face) {
+		for (level = t->minLod; level <= t->maxLod; ++level) {
+			img = get_radeon_texture_image(texObj->Image[face][level]);
+
+			if (RADEON_DEBUG & RADEON_TEXTURE) {
+				fprintf(stderr, "Checking image level %d, face %d, mt %p ... ", level, face, img->mt);
+			}
+			
+			if (img->mt != dst_miptree) {
+				if (RADEON_DEBUG & RADEON_TEXTURE) {
+					fprintf(stderr, "MIGRATING\n");
+				}
+				migrate_image_to_miptree(dst_miptree, img, face, radeon_gl_level_to_miptree_level(texObj, level));
+			} else if (RADEON_DEBUG & RADEON_TEXTURE) {
+				fprintf(stderr, "OK\n");
+			}
+		}
+	}
+
+	t->validated = GL_TRUE;
+
+	return GL_TRUE;
+}
+
+uint32_t get_base_teximage_offset(radeonTexObj *texObj)
+{
+	if (!texObj->mt) {
+		return 0;
+	} else {
+		return radeon_miptree_image_offset(texObj->mt, 0, texObj->minLod);
+	}
+}
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
index db28252da3..28b8485095 100644
--- a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
+++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
@@ -59,43 +59,38 @@ struct _radeon_mipmap_level {
  * changed.
  */
 struct _radeon_mipmap_tree {
-	radeonContextPtr radeon;
-	radeonTexObj *t;
 	struct radeon_bo *bo;
 	GLuint refcount;
 
 	GLuint totalsize; /** total size of the miptree, in bytes */
 
 	GLenum target; /** GL_TEXTURE_xxx */
-	GLenum internal_format;
+	GLenum mesaFormat; /** MESA_FORMAT_xxx */
 	GLuint faces; /** # of faces: 6 for cubemaps, 1 otherwise */
-	GLuint firstLevel; /** First mip level stored in this mipmap tree */
-	GLuint lastLevel; /** Last mip level stored in this mipmap tree */
+	GLuint baseLevel; /** gl_texture_object->baseLevel it was created for */
+	GLuint numLevels; /** Number of mip levels stored in this mipmap tree */
 
 	GLuint width0; /** Width of firstLevel image */
 	GLuint height0; /** Height of firstLevel image */
 	GLuint depth0; /** Depth of firstLevel image */
 
-	GLuint bpp; /** Bytes per texel */
 	GLuint tilebits; /** RADEON_TXO_xxx_TILE */
-	GLuint compressed; /** MESA_FORMAT_xxx indicating a compressed format, or 0 if uncompressed */
 
 	radeon_mipmap_level levels[RADEON_MIPTREE_MAX_TEXTURE_LEVELS];
 };
 
-radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj *t,
-		GLenum target, GLenum internal_format, GLuint firstLevel, GLuint lastLevel,
-		GLuint width0, GLuint height0, GLuint depth0,
-		GLuint bpp, GLuint tilebits, GLuint compressed);
-void radeon_miptree_reference(radeon_mipmap_tree *mt);
-void radeon_miptree_unreference(radeon_mipmap_tree *mt);
+void radeon_miptree_reference(radeon_mipmap_tree *mt, radeon_mipmap_tree **ptr);
+void radeon_miptree_unreference(radeon_mipmap_tree **ptr);
 
 GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
 		struct gl_texture_image *texImage, GLuint face, GLuint level);
-GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj);
-void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t,
-			      radeon_texture_image *texImage, GLuint face, GLuint level);
+void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t);
 GLuint radeon_miptree_image_offset(radeon_mipmap_tree *mt,
 				   GLuint face, GLuint level);
 void radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets);
+
+unsigned radeon_miptree_level_to_gl_level(struct gl_texture_object *tObj, unsigned level);
+unsigned radeon_gl_level_to_miptree_level(struct gl_texture_object *tObj, unsigned level);
+
+uint32_t get_base_teximage_offset(radeonTexObj *texObj);
 #endif /* __RADEON_MIPMAP_TREE_H_ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_queryobj.c b/src/mesa/drivers/dri/radeon/radeon_queryobj.c
index 452d0446f9..98117cdfc1 100644
--- a/src/mesa/drivers/dri/radeon/radeon_queryobj.c
+++ b/src/mesa/drivers/dri/radeon/radeon_queryobj.c
@@ -31,20 +31,6 @@
 #include "main/imports.h"
 #include "main/simple_list.h"
 
-static int radeonQueryIsFlushed(GLcontext *ctx, struct gl_query_object *q)
-{
-	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
-	struct radeon_query_object *tmp, *query = (struct radeon_query_object *)q;
-
-	foreach(tmp, &radeon->query.not_flushed_head) {
-		if (tmp == query) {
-			return 0;
-		}
-	}
-
-	return 1;
-}
-
 static void radeonQueryGetResult(GLcontext *ctx, struct gl_query_object *q)
 {
 	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
@@ -122,10 +108,11 @@ static void radeonDeleteQuery(GLcontext *ctx, struct gl_query_object *q)
 
 static void radeonWaitQuery(GLcontext *ctx, struct gl_query_object *q)
 {
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 	struct radeon_query_object *query = (struct radeon_query_object *)q;
 
 	/* If the cmdbuf with packets for this query hasn't been flushed yet, do it now */
-	if (!radeonQueryIsFlushed(ctx, q))
+	if (radeon_bo_is_referenced_by_cs(query->bo, radeon->cmdbuf.cs))
 		ctx->Driver.Flush(ctx);
 
 	radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, q->Id, query->bo, query->curr_offset);
@@ -157,8 +144,6 @@ static void radeonBeginQuery(GLcontext *ctx, struct gl_query_object *q)
 
 	radeon->query.queryobj.dirty = GL_TRUE;
 	radeon->hw.is_dirty = GL_TRUE;
-	insert_at_tail(&radeon->query.not_flushed_head, query);
-
 }
 
 void radeonEmitQueryEnd(GLcontext *ctx)
@@ -206,7 +191,7 @@ static void radeonCheckQuery(GLcontext *ctx, struct gl_query_object *q)
 		uint32_t domain;
 
 		/* Need to perform a flush, as per ARB_occlusion_query spec */
-		if (!radeonQueryIsFlushed(ctx, q)) {
+		if (radeon_bo_is_referenced_by_cs(query->bo, radeon->cmdbuf.cs)) {
 			ctx->Driver.Flush(ctx);
 		}
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c
index 2d19220d8a..dd82888254 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state_init.c
+++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c
@@ -645,11 +645,11 @@ static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom)
      OUT_BATCH(CP_PACKET0(RADEON_PP_TXOFFSET_0 + (24 * i), 0));
      if (t->mt && !t->image_override) {
         if ((ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_CUBE_BIT)) {
-            lvl = &t->mt->levels[0];
+            lvl = &t->mt->levels[t->minLod];
 	    OUT_BATCH_RELOC(lvl->faces[5].offset, t->mt->bo, lvl->faces[5].offset,
 			RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
         } else {
-           OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0,
+           OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, get_base_teximage_offset(t),
 		     RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0);
         }
       } else {
diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.c b/src/mesa/drivers/dri/radeon/radeon_tex.c
index 60981aada2..749ab75f20 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tex.c
+++ b/src/mesa/drivers/dri/radeon/radeon_tex.c
@@ -348,17 +348,7 @@ static void radeonTexParameter( GLcontext *ctx, GLenum target,
    case GL_TEXTURE_MAX_LEVEL:
    case GL_TEXTURE_MIN_LOD:
    case GL_TEXTURE_MAX_LOD:
-
-      /* This isn't the most efficient solution but there doesn't appear to
-       * be a nice alternative.  Since there's no LOD clamping,
-       * we just have to rely on loading the right subset of mipmap levels
-       * to simulate a clamped LOD.
-       */
-      if (t->mt) {
-         radeon_miptree_unreference(t->mt);
-	 t->mt = 0;
-	 t->validated = GL_FALSE;
-      }
+      t->validated = GL_FALSE;
       break;
 
    default:
@@ -388,10 +378,8 @@ static void radeonDeleteTexture( GLcontext *ctx,
      }
    }
 
-   if (t->mt) {
-      radeon_miptree_unreference(t->mt);
-      t->mt = 0;
-   }
+   radeon_miptree_unreference(&t->mt);
+
    /* Free mipmap images and the texture object itself */
    _mesa_delete_texture_object(ctx, texObj);
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_texstate.c b/src/mesa/drivers/dri/radeon/radeon_texstate.c
index 429977a8bc..3cbe3b4725 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texstate.c
+++ b/src/mesa/drivers/dri/radeon/radeon_texstate.c
@@ -699,14 +699,10 @@ void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_
 		radeon_bo_unref(rImage->bo);
 		rImage->bo = NULL;
 	}
-	if (t->mt) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = NULL;
-	}
-	if (rImage->mt) {
-		radeon_miptree_unreference(rImage->mt);
-		rImage->mt = NULL;
-	}
+
+	radeon_miptree_unreference(&t->mt);
+	radeon_miptree_unreference(&rImage->mt);
+
 	_mesa_init_teximage_fields(radeon->glCtx, target, texImage,
 				   rb->base.Width, rb->base.Height, 1, 0, rb->cpp);
 	texImage->RowStride = rb->pitch / rb->cpp;
@@ -718,8 +714,6 @@ void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_
 	t->tile_bits = 0;
 	t->image_override = GL_TRUE;
 	t->override_offset = 0;
-	t->pp_txpitch &= (1 << 13) -1;
-	pitch_val = rb->pitch;
 	switch (rb->cpp) {
 	case 4:
 		if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT)
@@ -738,12 +732,17 @@ void radeonSetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_
 		t->pp_txfilter |= tx_table[MESA_FORMAT_RGB565].filter;
 		break;
 	}
-        t->pp_txsize = ((rb->base.Width - 1) << RADEON_TEX_USIZE_SHIFT)
-		   | ((rb->base.Height - 1) << RADEON_TEX_VSIZE_SHIFT);
-        t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2;
-	t->pp_txpitch = pitch_val;
-        t->pp_txpitch -= 32;
 
+	t->pp_txpitch &= (1 << 13) -1;
+	pitch_val = rb->pitch;
+
+        t->pp_txsize = ((rb->base.Width - 1) << RADEON_TEX_USIZE_SHIFT)
+		| ((rb->base.Height - 1) << RADEON_TEX_VSIZE_SHIFT);
+	if (target == GL_TEXTURE_RECTANGLE_NV) {
+		t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2;
+		t->pp_txpitch = pitch_val;
+		t->pp_txpitch -= 32;
+	}
 	t->validated = GL_TRUE;
 	_mesa_unlock_texture(radeon->glCtx, texObj);
 	return;
@@ -1021,7 +1020,7 @@ static GLboolean setup_hardware_state(r100ContextPtr rmesa, radeonTexObj *t, int
 	return GL_TRUE;
    }
 
-   firstImage = t->base.Image[0][t->mt->firstLevel];   
+   firstImage = t->base.Image[0][t->minLod];
 
    if (firstImage->Border > 0) {
       fprintf(stderr, "%s: border\n", __FUNCTION__);
@@ -1049,9 +1048,9 @@ static GLboolean setup_hardware_state(r100ContextPtr rmesa, radeonTexObj *t, int
 	 return GL_FALSE;
       }
    }
-   
+
    t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
-   t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << RADEON_MAX_MIP_LEVEL_SHIFT;
+   t->pp_txfilter |= (t->maxLod - t->minLod) << RADEON_MAX_MIP_LEVEL_SHIFT;
 	
    t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
 		       RADEON_TXFORMAT_HEIGHT_MASK |
@@ -1060,9 +1059,9 @@ static GLboolean setup_hardware_state(r100ContextPtr rmesa, radeonTexObj *t, int
 		       RADEON_TXFORMAT_F5_HEIGHT_MASK);
    t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
 		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
-   
+
    t->tile_bits = 0;
-   
+
    if (t->base.Target == GL_TEXTURE_CUBE_MAP) {
       ASSERT(log2Width == log2Height);
       t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_F5_WIDTH_SHIFT) |
diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.c b/src/mesa/drivers/dri/radeon/radeon_texture.c
index baa99b752b..6f11f1fa4a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texture.c
+++ b/src/mesa/drivers/dri/radeon/radeon_texture.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (C) 2009 Maciej Cencora.
  * Copyright (C) 2008 Nicolai Haehnle.
  * Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
  *
@@ -46,7 +47,7 @@
 #include "radeon_mipmap_tree.h"
 
 
-static void copy_rows(void* dst, GLuint dststride, const void* src, GLuint srcstride,
+void copy_rows(void* dst, GLuint dststride, const void* src, GLuint srcstride,
 	GLuint numrows, GLuint rowsize)
 {
 	assert(rowsize <= dststride);
@@ -81,8 +82,7 @@ void radeonFreeTexImageData(GLcontext *ctx, struct gl_texture_image *timage)
 	radeon_texture_image* image = get_radeon_texture_image(timage);
 
 	if (image->mt) {
-		radeon_miptree_unreference(image->mt);
-		image->mt = 0;
+		radeon_miptree_unreference(&image->mt);
 		assert(!image->base.Data);
 	} else {
 		_mesa_free_texture_image_data(ctx, timage);
@@ -108,7 +108,7 @@ static void teximage_set_map_data(radeon_texture_image *image)
 	lvl = &image->mt->levels[image->mtlevel];
 
 	image->base.Data = image->mt->bo->ptr + lvl->faces[image->mtface].offset;
-	image->base.RowStride = lvl->rowstride / image->mt->bpp;
+	image->base.RowStride = lvl->rowstride / _mesa_get_format_bytes(image->base.TexFormat);
 }
 
 
@@ -174,7 +174,7 @@ void radeonMapTexture(GLcontext *ctx, struct gl_texture_object *texObj)
 
 	radeon_bo_map(t->mt->bo, GL_FALSE);
 	for(face = 0; face < t->mt->faces; ++face) {
-		for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level)
+		for(level = t->minLod; level <= t->maxLod; ++level)
 			teximage_set_map_data(get_radeon_texture_image(texObj->Image[face][level]));
 	}
 }
@@ -191,7 +191,7 @@ void radeonUnmapTexture(GLcontext *ctx, struct gl_texture_object *texObj)
 	  return;
 
 	for(face = 0; face < t->mt->faces; ++face) {
-		for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level)
+		for(level = t->minLod; level <= t->maxLod; ++level)
 			texObj->Image[face][level]->Data = 0;
 	}
 	radeon_bo_unmap(t->mt->bo);
@@ -240,8 +240,7 @@ static void radeon_generate_mipmap(GLcontext *ctx, GLenum target,
 			image->mtlevel = i;
 			image->mtface = face;
 
-			radeon_miptree_unreference(image->mt);
-			image->mt = NULL;
+			radeon_miptree_unreference(&image->mt);
 		}
 	}
 	
@@ -510,6 +509,135 @@ gl_format radeonChooseTextureFormat(GLcontext * ctx,
 	return MESA_FORMAT_NONE;		/* never get here */
 }
 
+static void teximage_assign_miptree(radeonContextPtr rmesa,
+	struct gl_texture_object *texObj,
+	struct gl_texture_image *texImage,
+	unsigned face,
+	unsigned level)
+{
+	radeonTexObj *t = radeon_tex_obj(texObj);
+	radeon_texture_image* image = get_radeon_texture_image(texImage);
+
+	/* Try using current miptree, or create new if there isn't any */
+	if (!t->mt || !radeon_miptree_matches_image(t->mt, texImage, face,
+					radeon_gl_level_to_miptree_level(texObj, level))) {
+		radeon_miptree_unreference(&t->mt);
+		radeon_try_alloc_miptree(rmesa, t);
+	}
+
+	/* Miptree alocation may have failed,
+	 * when there was no image for baselevel specified */
+	if (t->mt) {
+		image->mtface = face;
+		image->mtlevel = radeon_gl_level_to_miptree_level(texObj, level);
+		radeon_miptree_reference(t->mt, &image->mt);
+	}
+}
+
+static GLuint * allocate_image_offsets(GLcontext *ctx,
+	unsigned alignedWidth,
+	unsigned height,
+	unsigned depth)
+{
+	int i;
+	GLuint *offsets;
+
+	offsets = _mesa_malloc(depth * sizeof(GLuint)) ;
+	if (!offsets) {
+		_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTex[Sub]Image");
+		return NULL;
+	}
+
+	for (i = 0; i < depth; ++i) {
+		offsets[i] = alignedWidth * height * i;
+	}
+
+	return offsets;
+}
+
+/**
+ * Update a subregion of the given texture image.
+ */
+static void radeon_store_teximage(GLcontext* ctx, int dims,
+		GLint xoffset, GLint yoffset, GLint zoffset,
+		GLsizei width, GLsizei height, GLsizei depth,
+		GLsizei imageSize,
+		GLenum format, GLenum type,
+		const GLvoid * pixels,
+		const struct gl_pixelstore_attrib *packing,
+		struct gl_texture_object *texObj,
+		struct gl_texture_image *texImage,
+		int compressed)
+{
+	radeonTexObj *t = radeon_tex_obj(texObj);
+	radeon_texture_image* image = get_radeon_texture_image(texImage);
+
+	GLuint dstRowStride;
+	GLuint *dstImageOffsets;
+
+	if (image->mt) {
+		dstRowStride = image->mt->levels[image->mtlevel].rowstride;
+	} else if (t->bo) {
+		/* TFP case */
+		/* TODO */
+		assert(0);
+	} else {
+		dstRowStride = _mesa_format_row_stride(texImage->TexFormat, width);
+	}
+
+	if (dims == 3) {
+		unsigned alignedWidth = dstRowStride/_mesa_get_format_bytes(texImage->TexFormat);
+		dstImageOffsets = allocate_image_offsets(ctx, alignedWidth, height, depth);
+		if (!dstImageOffsets) {
+			return;
+		}
+	} else {
+		dstImageOffsets = texImage->ImageOffsets;
+	}
+
+	radeon_teximage_map(image, GL_TRUE);
+
+	if (compressed) {
+		uint32_t srcRowStride, bytesPerRow, rows;
+		GLubyte *img_start;
+		if (!image->mt) {
+			dstRowStride = _mesa_format_row_stride(texImage->TexFormat, texImage->Width);
+			img_start = _mesa_compressed_image_address(xoffset, yoffset, 0,
+									texImage->TexFormat,
+									texImage->Width, texImage->Data);
+		}
+		else {
+			uint32_t blocks_x, block_width, block_height;
+			_mesa_get_format_block_size(image->mt->mesaFormat, &block_width, &block_height);
+			blocks_x = dstRowStride / block_width;
+			img_start = texImage->Data + _mesa_get_format_bytes(image->mt->mesaFormat) * 4 * (blocks_x * (yoffset / 4) + xoffset / 4);
+		}
+		srcRowStride = _mesa_format_row_stride(texImage->TexFormat, width);
+		bytesPerRow = srcRowStride;
+		rows = (height + 3) / 4;
+
+		copy_rows(img_start, dstRowStride,  pixels, srcRowStride, rows,  bytesPerRow);
+
+	}
+	else {
+		if (!_mesa_texstore(ctx, dims, texImage->_BaseFormat,
+					texImage->TexFormat, texImage->Data,
+					xoffset, yoffset, zoffset,
+					dstRowStride,
+					dstImageOffsets,
+					width, height, depth,
+					format, type, pixels, packing)) {
+			_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage");
+		}
+	}
+
+	if (dims == 3) {
+		_mesa_free(dstImageOffsets);
+	}
+
+	radeon_teximage_unmap(image);
+}
+
 /**
  * All glTexImage calls go through this function.
  */
@@ -528,13 +656,17 @@ static void radeon_teximage(
 	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 	radeonTexObj* t = radeon_tex_obj(texObj);
 	radeon_texture_image* image = get_radeon_texture_image(texImage);
-	GLuint dstRowStride;
 	GLint postConvWidth = width;
 	GLint postConvHeight = height;
-	GLuint texelBytes;
 	GLuint face = radeon_face_for_target(target);
 
-	radeon_firevertices(rmesa);
+	{
+		struct radeon_bo *bo;
+		bo = !image->mt ? image->bo : image->mt->bo;
+		if (bo && radeon_bo_is_referenced_by_cs(bo, rmesa->cmdbuf.cs)) {
+			radeon_firevertices(rmesa);
+		}
+	}
 
 	t->validated = GL_FALSE;
 
@@ -543,53 +675,30 @@ static void radeon_teximage(
 						  &postConvHeight);
 	}
 
-	if (_mesa_is_format_compressed(texImage->TexFormat)) {
-		texelBytes = 0;
-	} else {
-		texelBytes = _mesa_get_format_bytes(texImage->TexFormat);
+	if (!_mesa_is_format_compressed(texImage->TexFormat)) {
+		GLuint texelBytes = _mesa_get_format_bytes(texImage->TexFormat);
 		/* Minimum pitch of 32 bytes */
 		if (postConvWidth * texelBytes < 32) {
-		  postConvWidth = 32 / texelBytes;
-		  texImage->RowStride = postConvWidth;
+			postConvWidth = 32 / texelBytes;
+			texImage->RowStride = postConvWidth;
 		}
-		if (!image->mt) {      
+		if (!image->mt) {
 			assert(texImage->RowStride == postConvWidth);
 		}
 	}
 
-	/* Allocate memory for image */
-	radeonFreeTexImageData(ctx, texImage); /* Mesa core only clears texImage->Data but not image->mt */
+	/* Mesa core only clears texImage->Data but not image->mt */
+	radeonFreeTexImageData(ctx, texImage);
 
-	if (t->mt &&
-	    t->mt->firstLevel == level &&
-	    t->mt->lastLevel == level &&
-	    t->mt->target != GL_TEXTURE_CUBE_MAP_ARB &&
-	    !radeon_miptree_matches_image(t->mt, texImage, face, level)) {
-	  radeon_miptree_unreference(t->mt);
-	  t->mt = NULL;
-	}
-
-	if (!t->mt)
-		radeon_try_alloc_miptree(rmesa, t, image, face, level);
-	if (t->mt && radeon_miptree_matches_image(t->mt, texImage, face, level)) {
-		radeon_mipmap_level *lvl;
-		image->mt = t->mt;
-		image->mtlevel = level - t->mt->firstLevel;
-		image->mtface = face;
-		radeon_miptree_reference(t->mt);
-		lvl = &image->mt->levels[image->mtlevel];
-		dstRowStride = lvl->rowstride;
-	} else {
-		int size;
-		if (_mesa_is_format_compressed(texImage->TexFormat)) {
-			size = _mesa_format_image_size(texImage->TexFormat,
-						       texImage->Width,
-						       texImage->Height,
-						       texImage->Depth);
-		} else {
-			size = texImage->Width * texImage->Height * texImage->Depth * _mesa_get_format_bytes(texImage->TexFormat);
+	if (!t->bo) {
+		teximage_assign_miptree(rmesa, texObj, texImage, face, level);
+		if (!t->mt) {
+			int size = _mesa_format_image_size(texImage->TexFormat,
+								texImage->Width,
+								texImage->Height,
+								texImage->Depth);
+			texImage->Data = _mesa_alloc_texmemory(size);
 		}
-		texImage->Data = _mesa_alloc_texmemory(size);
 	}
 
 	/* Upload texture image; note that the spec allows pixels to be NULL */
@@ -603,65 +712,16 @@ static void radeon_teximage(
 	}
 
 	if (pixels) {
-		radeon_teximage_map(image, GL_TRUE);
-		if (compressed) {
-			if (image->mt) {
-				uint32_t srcRowStride, bytesPerRow, rows;
-				srcRowStride = _mesa_format_row_stride(texImage->TexFormat, width);
-				bytesPerRow = srcRowStride;
-				rows = (height + 3) / 4;
-				copy_rows(texImage->Data, image->mt->levels[level].rowstride,
-					  pixels, srcRowStride, rows, bytesPerRow);
-			} else {
-				memcpy(texImage->Data, pixels, imageSize);
-			}
-		} else {
-			GLuint dstRowStride;
-			GLuint *dstImageOffsets;
-
-			if (image->mt) {
-				radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel];
-				dstRowStride = lvl->rowstride;
-			} else {
-				dstRowStride = texImage->Width * _mesa_get_format_bytes(texImage->TexFormat);
-			}
-
-			if (dims == 3) {
-				int i;
-
-				dstImageOffsets = _mesa_malloc(depth * sizeof(GLuint)) ;
-				if (!dstImageOffsets)
-					_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
-
-				for (i = 0; i < depth; ++i) {
-					dstImageOffsets[i] = dstRowStride/_mesa_get_format_bytes(texImage->TexFormat) * height * i;
-				}
-			} else {
-				dstImageOffsets = texImage->ImageOffsets;
-			}
-
-			if (!_mesa_texstore(ctx, dims,
-					    texImage->_BaseFormat,
-					    texImage->TexFormat,
-					    texImage->Data, 0, 0, 0, /* dstX/Y/Zoffset */
-					    dstRowStride,
-					    dstImageOffsets,
-					    width, height, depth,
-					    format, type, pixels, packing)) {
-				_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
-			}
-
-			if (dims == 3)
-				_mesa_free(dstImageOffsets);
-		}
+		radeon_store_teximage(ctx, dims,
+			0, 0, 0,
+			width, height, depth,
+			imageSize, format, type,
+			pixels, packing,
+			texObj, texImage,
+			compressed);
 	}
 
 	_mesa_unmap_teximage_pbo(ctx, packing);
-
-	if (pixels)
-	  radeon_teximage_unmap(image);
-
-
 }
 
 void radeonTexImage1D(GLcontext * ctx, GLenum target, GLint level,
@@ -714,7 +774,7 @@ void radeonTexImage3D(GLcontext * ctx, GLenum target, GLint level,
 }
 
 /**
- * Update a subregion of the given texture image.
+ * All glTexSubImage calls go through this function.
  */
 static void radeon_texsubimage(GLcontext* ctx, int dims, GLenum target, int level,
 		GLint xoffset, GLint yoffset, GLint zoffset,
@@ -731,66 +791,34 @@ static void radeon_texsubimage(GLcontext* ctx, int dims, GLenum target, int leve
 	radeonTexObj* t = radeon_tex_obj(texObj);
 	radeon_texture_image* image = get_radeon_texture_image(texImage);
 
-	radeon_firevertices(rmesa);
+	{
+		struct radeon_bo *bo;
+		bo = !image->mt ? image->bo : image->mt->bo;
+		if (bo && radeon_bo_is_referenced_by_cs(bo, rmesa->cmdbuf.cs)) {
+			radeon_firevertices(rmesa);
+		}
+	}
 
 	t->validated = GL_FALSE;
 	if (compressed) {
 		pixels = _mesa_validate_pbo_compressed_teximage(
-			ctx, imageSize, pixels, packing, "glCompressedTexImage");
+			ctx, imageSize, pixels, packing, "glCompressedTexSubImage");
 	} else {
 		pixels = _mesa_validate_pbo_teximage(ctx, dims,
-			width, height, depth, format, type, pixels, packing, "glTexSubImage1D");
+			width, height, depth, format, type, pixels, packing, "glTexSubImage");
 	}
 
 	if (pixels) {
-		GLint dstRowStride;
-		radeon_teximage_map(image, GL_TRUE);
-
-		if (image->mt) {
-			radeon_mipmap_level *lvl = &image->mt->levels[image->mtlevel];
-			dstRowStride = lvl->rowstride;
-		} else {
-			dstRowStride = texImage->RowStride * _mesa_get_format_bytes(texImage->TexFormat);
-		}
-
-		if (compressed) {
-			uint32_t srcRowStride, bytesPerRow, rows;
-			GLubyte *img_start;
-			if (!image->mt) {
-				dstRowStride = _mesa_format_row_stride(texImage->TexFormat, texImage->Width);
-				img_start = _mesa_compressed_image_address(xoffset, yoffset, 0,
-									   texImage->TexFormat,
-									   texImage->Width, texImage->Data);
-			}
-			else {
-				uint32_t blocks_x = dstRowStride / (image->mt->bpp * 4);
-				img_start = texImage->Data + image->mt->bpp * 4 * (blocks_x * (yoffset / 4) + xoffset / 4);
-			}
-			srcRowStride = _mesa_format_row_stride(texImage->TexFormat, width);
-			bytesPerRow = srcRowStride;
-			rows = (height + 3) / 4;
-
-			copy_rows(img_start, dstRowStride,  pixels, srcRowStride, rows,  bytesPerRow);
-			
-		}
-		else {
-			if (!_mesa_texstore(ctx, dims, texImage->_BaseFormat,
-					    texImage->TexFormat, texImage->Data,
-					    xoffset, yoffset, zoffset,
-					    dstRowStride,
-					    texImage->ImageOffsets,
-					    width, height, depth,
-					    format, type, pixels, packing)) {
-				_mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage");
-			}
-		}
+		radeon_store_teximage(ctx, dims,
+			xoffset, yoffset, zoffset,
+			width, height, depth,
+			imageSize, format, type,
+			pixels, packing,
+			texObj, texImage,
+			compressed);
 	}
 
-	radeon_teximage_unmap(image);
-
 	_mesa_unmap_teximage_pbo(ctx, packing);
-
-
 }
 
 void radeonTexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
@@ -846,143 +874,6 @@ void radeonTexSubImage3D(GLcontext * ctx, GLenum target, GLint level,
 		format, type, pixels, packing, texObj, texImage, 0);
 }
 
-
-
-/**
- * Ensure that the given image is stored in the given miptree from now on.
- */
-static void migrate_image_to_miptree(radeon_mipmap_tree *mt, radeon_texture_image *image, int face, int level)
-{
-	radeon_mipmap_level *dstlvl = &mt->levels[level - mt->firstLevel];
-	unsigned char *dest;
-
-	assert(image->mt != mt);
-	assert(dstlvl->width == image->base.Width);
-	assert(dstlvl->height == image->base.Height);
-	assert(dstlvl->depth == image->base.Depth);
-
-
-	radeon_bo_map(mt->bo, GL_TRUE);
-	dest = mt->bo->ptr + dstlvl->faces[face].offset;
-
-	if (image->mt) {
-		/* Format etc. should match, so we really just need a memcpy().
-		 * In fact, that memcpy() could be done by the hardware in many
-		 * cases, provided that we have a proper memory manager.
-		 */
-		radeon_mipmap_level *srclvl = &image->mt->levels[image->mtlevel-image->mt->firstLevel];
-
-		assert(srclvl->size == dstlvl->size);
-		assert(srclvl->rowstride == dstlvl->rowstride);
-
-		radeon_bo_map(image->mt->bo, GL_FALSE);
-
-		memcpy(dest,
-			image->mt->bo->ptr + srclvl->faces[face].offset,
-			dstlvl->size);
-		radeon_bo_unmap(image->mt->bo);
-
-		radeon_miptree_unreference(image->mt);
-	} else {
-		uint32_t srcrowstride;
-		uint32_t height;
-		/* need to confirm this value is correct */
-		if (mt->compressed) {
-			height = (image->base.Height + 3) / 4;
-			srcrowstride = _mesa_format_row_stride(image->base.TexFormat, image->base.Width);
-		} else {
-			height = image->base.Height * image->base.Depth;
-			srcrowstride = image->base.Width * _mesa_get_format_bytes(image->base.TexFormat);
-		}
-
-//		if (mt->tilebits)
-//			WARN_ONCE("%s: tiling not supported yet", __FUNCTION__);
-
-		copy_rows(dest, dstlvl->rowstride, image->base.Data, srcrowstride,
-			  height, srcrowstride);
-
-		_mesa_free_texmemory(image->base.Data);
-		image->base.Data = 0;
-	}
-
-	radeon_bo_unmap(mt->bo);
-
-	image->mt = mt;
-	image->mtface = face;
-	image->mtlevel = level;
-	radeon_miptree_reference(image->mt);
-}
-
-int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj)
-{
-	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-	radeonTexObj *t = radeon_tex_obj(texObj);
-	radeon_texture_image *baseimage = get_radeon_texture_image(texObj->Image[0][texObj->BaseLevel]);
-	int face, level;
-
-	if (t->validated || t->image_override)
-		return GL_TRUE;
-
-	if (RADEON_DEBUG & RADEON_TEXTURE)
-		fprintf(stderr, "%s: Validating texture %p now\n", __FUNCTION__, texObj);
-
-	if (baseimage->base.Border > 0)
-		return GL_FALSE;
-
-	/* Ensure a matching miptree exists.
-	 *
-	 * Differing mipmap trees can result when the app uses TexImage to
-	 * change texture dimensions.
-	 *
-	 * Prefer to use base image's miptree if it
-	 * exists, since that most likely contains more valid data (remember
-	 * that the base level is usually significantly larger than the rest
-	 * of the miptree, so cubemaps are the only possible exception).
-	 */
-	if (baseimage->mt &&
-	    baseimage->mt != t->mt &&
-	    radeon_miptree_matches_texture(baseimage->mt, &t->base)) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = baseimage->mt;
-		radeon_miptree_reference(t->mt);
-	} else if (t->mt && !radeon_miptree_matches_texture(t->mt, &t->base)) {
-		radeon_miptree_unreference(t->mt);
-		t->mt = 0;
-	}
-
-	if (!t->mt) {
-		if (RADEON_DEBUG & RADEON_TEXTURE)
-			fprintf(stderr, " Allocate new miptree\n");
-		radeon_try_alloc_miptree(rmesa, t, baseimage, 0, texObj->BaseLevel);
-		if (!t->mt) {
-			_mesa_problem(ctx, "radeon_validate_texture failed to alloc miptree");
-			return GL_FALSE;
-		}
-	}
-
-	/* Ensure all images are stored in the single main miptree */
-	for(face = 0; face < t->mt->faces; ++face) {
-		for(level = t->mt->firstLevel; level <= t->mt->lastLevel; ++level) {
-			radeon_texture_image *image = get_radeon_texture_image(texObj->Image[face][level]);
-			if (RADEON_DEBUG & RADEON_TEXTURE)
-				fprintf(stderr, " face %i, level %i... %p vs %p ", face, level, t->mt, image->mt);
-			if (t->mt == image->mt || (!image->mt && !image->base.Data)) {
-				if (RADEON_DEBUG & RADEON_TEXTURE)
-					fprintf(stderr, "OK\n");
-
-				continue;
-			}
-
-			if (RADEON_DEBUG & RADEON_TEXTURE)
-				fprintf(stderr, "migrating\n");
-			migrate_image_to_miptree(t->mt, image, face, level);
-		}
-	}
-
-	return GL_TRUE;
-}
-
-
 /**
  * Need to map texture image into memory before copying image data,
  * then unmap it.
diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.h b/src/mesa/drivers/dri/radeon/radeon_texture.h
index 8995546d77..906daf12d0 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texture.h
+++ b/src/mesa/drivers/dri/radeon/radeon_texture.h
@@ -33,7 +33,8 @@
 
 #include "main/formats.h"
 
-
+void copy_rows(void* dst, GLuint dststride, const void* src, GLuint srcstride,
+	GLuint numrows, GLuint rowsize);
 struct gl_texture_image *radeonNewTextureImage(GLcontext *ctx);
 void radeonFreeTexImageData(GLcontext *ctx, struct gl_texture_image *timage);
 
diff --git a/src/mesa/main/texstore.c b/src/mesa/main/texstore.c
index abb4ed2663..5387eb1283 100644
--- a/src/mesa/main/texstore.c
+++ b/src/mesa/main/texstore.c
@@ -2198,8 +2198,8 @@ _mesa_texstore_al1616(TEXSTORE_PARAMS)
             if (dstFormat == MESA_FORMAT_AL1616) {
                for (col = 0; col < srcWidth; col++) {
                   /* src[0] is luminance, src[1] is alpha */
-                 dstUI[col] = PACK_COLOR_88( FLOAT_TO_USHORT(src[1]),
-                                             FLOAT_TO_USHORT(src[0]) );
+                 dstUI[col] = PACK_COLOR_1616( FLOAT_TO_USHORT(src[1]),
+					       FLOAT_TO_USHORT(src[0]) );
                  src += 2;
                }
             }
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index 9186db76e1..d4630a514f 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -93,51 +93,6 @@ gl_target_to_pipe(GLenum target)
 }
 
 
-/**
- * Return nominal bytes per texel for a compressed format, 0 for non-compressed
- * format.
- */
-static GLuint
-compressed_num_bytes(gl_format format)
-{
-   switch (format) {
-#if FEATURE_texture_fxt1
-   case MESA_FORMAT_RGB_FXT1:
-   case MESA_FORMAT_RGBA_FXT1:
-#endif
-#if FEATURE_texture_s3tc
-   case MESA_FORMAT_RGB_DXT1:
-   case MESA_FORMAT_RGBA_DXT1:
-      return 2;
-   case MESA_FORMAT_RGBA_DXT3:
-   case MESA_FORMAT_RGBA_DXT5:
-      return 4;
-#endif
-   default:
-      return 0;
-   }
-}
-
-
-static GLboolean
-is_compressed_mesa_format(gl_format format)
-{
-   switch (format) {
-   case MESA_FORMAT_RGB_DXT1:
-   case MESA_FORMAT_RGBA_DXT1:
-   case MESA_FORMAT_RGBA_DXT3:
-   case MESA_FORMAT_RGBA_DXT5:
-   case MESA_FORMAT_SRGB_DXT1:
-   case MESA_FORMAT_SRGBA_DXT1:
-   case MESA_FORMAT_SRGBA_DXT3:
-   case MESA_FORMAT_SRGBA_DXT5:
-      return GL_TRUE;
-   default:
-      return GL_FALSE;
-   }
-}
-
-
 /** called via ctx->Driver.NewTextureImage() */
 static struct gl_texture_image *
 st_NewTextureImage(GLcontext * ctx)
@@ -663,7 +618,7 @@ st_TexImage(GLcontext * ctx,
     */
    if (!compressed_src &&
        !ctx->Mesa_DXTn &&
-       is_compressed_mesa_format(texImage->TexFormat) &&
+       _mesa_is_format_compressed(texImage->TexFormat) &&
        screen->is_format_supported(screen,
                                    stImage->pt->format,
                                    stImage->pt->target,
@@ -1066,7 +1021,7 @@ st_TexSubimage(GLcontext *ctx, GLint dims, GLenum target, GLint level,
    /* See if we can do texture compression with a blit/render.
     */
    if (!ctx->Mesa_DXTn &&
-       is_compressed_mesa_format(texImage->TexFormat) &&
+       _mesa_is_format_compressed(texImage->TexFormat) &&
        screen->is_format_supported(screen,
                                    stImage->pt->format,
                                    stImage->pt->target,
@@ -1724,8 +1679,6 @@ copy_image_data_to_texture(struct st_context *st,
       pipe_texture_reference(&stImage->pt, NULL);
    }
    else if (stImage->base.Data) {
-      assert(stImage->base.Data != NULL);
-
       /* More straightforward upload.  
        */
 
@@ -1764,7 +1717,7 @@ st_finalize_texture(GLcontext *ctx,
 {
    struct st_texture_object *stObj = st_texture_object(tObj);
    const GLuint nr_faces = (stObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
-   GLuint cpp, face;
+   GLuint blockSize, face;
    struct st_texture_image *firstImage;
 
    *needFlush = GL_FALSE;
@@ -1796,13 +1749,8 @@ st_finalize_texture(GLcontext *ctx,
       pipe_texture_reference(&stObj->pt, firstImage->pt);
    }
 
-   /* FIXME: determine format block instead of cpp */
-   if (_mesa_is_format_compressed(firstImage->base.TexFormat)) {
-      cpp = compressed_num_bytes(firstImage->base.TexFormat);
-   }
-   else {
-      cpp = _mesa_get_format_bytes(firstImage->base.TexFormat);
-   }
+   /* bytes per pixel block (blocks are usually 1x1) */
+   blockSize = _mesa_get_format_bytes(firstImage->base.TexFormat);
 
    /* If we already have a gallium texture, check that it matches the texture
     * object's format, target, size, num_levels, etc.
@@ -1816,8 +1764,7 @@ st_finalize_texture(GLcontext *ctx,
           stObj->pt->width[0] != firstImage->base.Width2 ||
           stObj->pt->height[0] != firstImage->base.Height2 ||
           stObj->pt->depth[0] != firstImage->base.Depth2 ||
-          /* Nominal bytes per pixel: */
-          stObj->pt->block.size / stObj->pt->block.width != cpp)
+          stObj->pt->block.size != blockSize)
       {
          pipe_texture_reference(&stObj->pt, NULL);
          ctx->st->dirty.st |= ST_NEW_FRAMEBUFFER;
diff --git a/src/mesa/state_tracker/st_texture.c b/src/mesa/state_tracker/st_texture.c
index 3945822f66..10f1351283 100644
--- a/src/mesa/state_tracker/st_texture.c
+++ b/src/mesa/state_tracker/st_texture.c
@@ -588,5 +588,5 @@ st_teximage_flush_before_map(struct st_context *st,
 
    if (referenced && ((referenced & PIPE_REFERENCED_FOR_WRITE) ||
 		      (usage & PIPE_TRANSFER_WRITE)))
-      st_flush(st, PIPE_FLUSH_RENDER_CACHE, NULL);
+      st->pipe->flush(st->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 }
diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c
index f72d2d84f3..c90565eae8 100644
--- a/src/mesa/vbo/vbo_exec_api.c
+++ b/src/mesa/vbo/vbo_exec_api.c
@@ -867,15 +867,27 @@ void vbo_exec_FlushVertices_internal( GLcontext *ctx, GLboolean unmap )
 }
 
 
-
+/**
+ * \param flags  bitmask of FLUSH_STORED_VERTICES, FLUSH_UPDATE_CURRENT
+ */
 void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags )
 {
    struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
 
+#ifdef DEBUG
+   /* debug check: make sure we don't get called recursively */
+   static GLuint callDepth = 0;
+   callDepth++;
+   assert(callDepth == 1);
+#endif
+
    if (0) _mesa_printf("%s\n", __FUNCTION__);
 
    if (exec->ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END) {
       if (0) _mesa_printf("%s - inside begin/end\n", __FUNCTION__);
+#ifdef DEBUG
+      callDepth--;
+#endif
       return;
    }
 
@@ -889,6 +901,10 @@ void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags )
    }
 
    exec->ctx->Driver.NeedFlush &= ~flags;
+
+#ifdef DEBUG
+   callDepth--;
+#endif
 }