Merge remote branch 'upstream/mesa_7_7_branch' into debian-unstable

37 files changed, 431 insertions, 155 deletions

diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c
index be5eb87e474..bb1f378e728 100644
--- a/src/gallium/auxiliary/util/u_debug.c
+++ b/src/gallium/auxiliary/util/u_debug.c
@@ -644,6 +644,57 @@ void debug_dump_image(const char *prefix,
    }
       
    EngUnmapFile(iFile);
+#elif defined(PIPE_OS_UNIX)
+   /* write a ppm file */
+   char filename[256];
+   FILE *f;
+
+   util_snprintf(filename, sizeof(filename), "%s.ppm", prefix);
+
+   f = fopen(filename, "w");
+   if (f) {
+      int i, x, y;
+      int r, g, b;
+      const uint8_t *ptr = (uint8_t *) data;
+
+      /* XXX this is a hack */
+      switch (format) {
+      case PIPE_FORMAT_B8G8R8A8_UNORM:
+         r = 2;
+         g = 1;
+         b = 0;
+         break;
+      case PIPE_FORMAT_A8R8G8B8_UNORM:
+         b = 0;
+         g = 1;
+         r = 2;
+         break;
+      default:
+         r = 0;
+         g = 1;
+         b = 2;
+      }
+
+      fprintf(f, "P6\n");
+      fprintf(f, "# ppm-file created by osdemo.c\n");
+      fprintf(f, "%i %i\n", width, height);
+      fprintf(f, "255\n");
+      fclose(f);
+
+      f = fopen(filename, "ab");  /* reopen in binary append mode */
+      for (y = 0; y < height; y++) {
+         for (x = 0; x < width; x++) {
+            i = y * stride + x * cpp;
+            fputc(ptr[i + r], f); /* write red */
+            fputc(ptr[i + g], f); /* write green */
+            fputc(ptr[i + b], f); /* write blue */
+         }
+      }
+      fclose(f);
+   }
+   else {
+      fprintf(stderr, "Can't open %s for writing\n", filename);
+   }
 #endif
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
index 65872cecc4f..4edc44e8c23 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -297,13 +297,14 @@ sp_tile_cache_flush_clear(struct softpipe_tile_cache *tc)
                               x, y, TILE_SIZE, TILE_SIZE,
                               tc->tile.data.color32, 0/*STRIDE*/);
 
-            /* do this? */
-            clear_clear_flag(tc->clear_flags, addr);
-
             numCleared++;
          }
       }
    }
+
+   /* reset all clear flags to zero */
+   memset(tc->clear_flags, 0, sizeof(tc->clear_flags));
+
 #if 0
    debug_printf("num cleared: %u\n", numCleared);
 #endif
diff --git a/src/gallium/drivers/svga/svga_context.c b/src/gallium/drivers/svga/svga_context.c
index 1e76592ec60..5a7311c485e 100644
--- a/src/gallium/drivers/svga/svga_context.c
+++ b/src/gallium/drivers/svga/svga_context.c
@@ -284,3 +284,8 @@ void svga_hwtnl_flush_retry( struct svga_context *svga )
    assert(ret == 0);
 }
 
+struct svga_winsys_context *
+svga_winsys_context( struct pipe_context *pipe )
+{
+   return svga_context( pipe )->swc;
+}
diff --git a/src/gallium/drivers/svga/svga_state_rss.c b/src/gallium/drivers/svga/svga_state_rss.c
index aa210697199..e81c3c63352 100644
--- a/src/gallium/drivers/svga/svga_state_rss.c
+++ b/src/gallium/drivers/svga/svga_state_rss.c
@@ -192,15 +192,24 @@ static int emit_rss( struct svga_context *svga,
    }
 
 
-   if (dirty & SVGA_NEW_RAST)
+   if (dirty & (SVGA_NEW_RAST | SVGA_NEW_NEED_PIPELINE))
    {
       const struct svga_rasterizer_state *curr = svga->curr.rast; 
+      unsigned cullmode = curr->cullmode;
 
       /* Shademode: still need to rearrange index list to move
        * flat-shading PV first vertex.
        */
       EMIT_RS( svga, curr->shademode, SHADEMODE, fail );
-      EMIT_RS( svga, curr->cullmode, CULLMODE, fail );
+
+      /* Don't do culling while the software pipeline is active.  It
+       * does it for us, and additionally introduces potentially
+       * back-facing triangles.
+       */
+      if (svga->state.sw.need_pipeline)
+         cullmode = SVGA3D_FACE_NONE;
+
+      EMIT_RS( svga, cullmode, CULLMODE, fail );
       EMIT_RS( svga, curr->scissortestenable, SCISSORTESTENABLE, fail );
       EMIT_RS( svga, curr->multisampleantialias, MULTISAMPLEANTIALIAS, fail );
       EMIT_RS( svga, curr->lastpixel, LASTPIXEL, fail );
diff --git a/src/gallium/drivers/svga/svga_winsys.h b/src/gallium/drivers/svga/svga_winsys.h
index 27b99fe4c10..054d388c0b8 100644
--- a/src/gallium/drivers/svga/svga_winsys.h
+++ b/src/gallium/drivers/svga/svga_winsys.h
@@ -281,6 +281,9 @@ svga_screen_create(struct svga_winsys_screen *sws);
 struct svga_winsys_screen *
 svga_winsys_screen(struct pipe_screen *screen);
 
+struct svga_winsys_context *
+svga_winsys_context(struct pipe_context *context);
+
 struct pipe_buffer *
 svga_screen_buffer_wrap_surface(struct pipe_screen *screen,
 				enum SVGA3dSurfaceFormat format,
diff --git a/src/gallium/state_trackers/dri/dri_screen.c b/src/gallium/state_trackers/dri/dri_screen.c
index c55daf9bca8..66a0e7128ed 100644
--- a/src/gallium/state_trackers/dri/dri_screen.c
+++ b/src/gallium/state_trackers/dri/dri_screen.c
@@ -80,6 +80,9 @@ dri_fill_in_modes(struct dri_screen *screen,
 		  unsigned pixel_bits)
 {
    __DRIconfig **configs = NULL;
+   __DRIconfig **configs_r5g6b5 = NULL;
+   __DRIconfig **configs_a8r8g8b8 = NULL;
+   __DRIconfig **configs_x8r8g8b8 = NULL;
    unsigned num_modes;
    uint8_t depth_bits_array[5];
    uint8_t stencil_bits_array[5];
@@ -117,25 +120,23 @@ dri_fill_in_modes(struct dri_screen *screen,
    pf_x8r8g8b8 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_X8R8G8B8_UNORM,
 					       PIPE_TEXTURE_2D,
 					       PIPE_TEXTURE_USAGE_RENDER_TARGET, 0);
-
-   /* we support buffers with different depths only if we can tell the driver
-    * the actual depth of each of them. */
-   if (screen->sPriv->dri2.loader
-       && (screen->sPriv->dri2.loader->base.version > 2)
-       && (screen->sPriv->dri2.loader->getBuffersWithFormat != NULL)) {
+   pf_r5g6b5 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_R5G6B5_UNORM,
+					     PIPE_TEXTURE_2D,
+					     PIPE_TEXTURE_USAGE_RENDER_TARGET, 0);
+
+   /* We can only get a 16 or 32 bit depth buffer with getBuffersWithFormat */
+   if (screen->sPriv->dri2.loader &&
+       (screen->sPriv->dri2.loader->base.version > 2) &&
+       (screen->sPriv->dri2.loader->getBuffersWithFormat != NULL)) {
       pf_z16 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_Z16_UNORM,
                                              PIPE_TEXTURE_2D,
                                              PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0);
       pf_z32 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_Z32_UNORM,
                                              PIPE_TEXTURE_2D,
                                              PIPE_TEXTURE_USAGE_DEPTH_STENCIL, 0);
-      pf_r5g6b5 = p_screen->is_format_supported(p_screen, PIPE_FORMAT_R5G6B5_UNORM,
-                                                PIPE_TEXTURE_2D,
-                                                PIPE_TEXTURE_USAGE_RENDER_TARGET, 0);
    } else {
       pf_z16 = FALSE;
       pf_z32 = FALSE;
-      pf_r5g6b5 = FALSE;
    }
 
    if (pf_z16) {
@@ -164,41 +165,43 @@ dri_fill_in_modes(struct dri_screen *screen,
    num_modes =
       depth_buffer_factor * back_buffer_factor * msaa_samples_factor * 4;
 
-   if (pixel_bits == 16 && pf_r5g6b5) {
-      configs = driCreateConfigs(GL_RGB, GL_UNSIGNED_SHORT_5_6_5,
-				 depth_bits_array, stencil_bits_array,
-				 depth_buffer_factor, back_buffer_modes,
-				 back_buffer_factor,
-				 msaa_samples_array, 1);
+   if (pf_r5g6b5)
+      configs_r5g6b5 = driCreateConfigs(GL_RGB, GL_UNSIGNED_SHORT_5_6_5,
+                                        depth_bits_array, stencil_bits_array,
+                                        depth_buffer_factor, back_buffer_modes,
+                                        back_buffer_factor,
+                                        msaa_samples_array, 1);
+
+   if (pf_a8r8g8b8)
+      configs_a8r8g8b8 = driCreateConfigs(GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV,
+                                          depth_bits_array,
+                                          stencil_bits_array,
+                                          depth_buffer_factor,
+                                          back_buffer_modes,
+                                          back_buffer_factor,
+                                          msaa_samples_array, 1);
+
+   if (pf_x8r8g8b8)
+      configs_x8r8g8b8 = driCreateConfigs(GL_BGR, GL_UNSIGNED_INT_8_8_8_8_REV,
+                                          depth_bits_array,
+                                          stencil_bits_array,
+                                          depth_buffer_factor,
+                                          back_buffer_modes,
+                                          back_buffer_factor,
+                                          msaa_samples_array, 1);
+
+   if (pixel_bits == 16) {
+      configs = configs_r5g6b5;
+      if (configs_a8r8g8b8)
+         configs = configs ? driConcatConfigs(configs, configs_a8r8g8b8) : configs_a8r8g8b8;
+      if (configs_x8r8g8b8)
+	 configs = configs ? driConcatConfigs(configs, configs_x8r8g8b8) : configs_x8r8g8b8;
    } else {
-      __DRIconfig **configs_a8r8g8b8 = NULL;
-      __DRIconfig **configs_x8r8g8b8 = NULL;
-
-      if (pf_a8r8g8b8)
-	 configs_a8r8g8b8 = driCreateConfigs(GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV,
-					     depth_bits_array,
-					     stencil_bits_array,
-					     depth_buffer_factor,
-					     back_buffer_modes,
-					     back_buffer_factor,
-					     msaa_samples_array, 1);
-      if (pf_x8r8g8b8)
-	 configs_x8r8g8b8 = driCreateConfigs(GL_BGR, GL_UNSIGNED_INT_8_8_8_8_REV,
-					     depth_bits_array,
-					     stencil_bits_array,
-					     depth_buffer_factor,
-					     back_buffer_modes,
-					     back_buffer_factor,
-					     msaa_samples_array, 1);
-
-      if (configs_a8r8g8b8 && configs_x8r8g8b8)
-	 configs = driConcatConfigs(configs_x8r8g8b8, configs_a8r8g8b8);
-      else if (configs_a8r8g8b8)
-	 configs = configs_a8r8g8b8;
-      else if (configs_x8r8g8b8)
-	 configs = configs_x8r8g8b8;
-      else
-	 configs = NULL;
+      configs = configs_a8r8g8b8;
+      if (configs_x8r8g8b8)
+	 configs = configs ? driConcatConfigs(configs, configs_x8r8g8b8) : configs_x8r8g8b8;
+      if (configs_r5g6b5)
+         configs = configs ? driConcatConfigs(configs, configs_r5g6b5) : configs_r5g6b5;
    }
 
    if (configs == NULL) {
diff --git a/src/gallium/state_trackers/xorg/xorg_crtc.c b/src/gallium/state_trackers/xorg/xorg_crtc.c
index daa9f8b8204..cba9b057a5b 100644
--- a/src/gallium/state_trackers/xorg/xorg_crtc.c
+++ b/src/gallium/state_trackers/xorg/xorg_crtc.c
@@ -39,6 +39,7 @@
 #include <xf86.h>
 #include <xf86i2c.h>
 #include <xf86Crtc.h>
+#include <cursorstr.h>
 #include "xorg_tracker.h"
 #include "xf86Modes.h"
 
@@ -146,6 +147,7 @@ crtc_gamma_set(xf86CrtcPtr crtc, CARD16 * red, CARD16 * green, CARD16 * blue,
     /* XXX: hockup */
 }
 
+#if 0 /* Implement and enable to enable rotation and reflection. */
 static void *
 crtc_shadow_allocate(xf86CrtcPtr crtc, int width, int height)
 {
@@ -168,6 +170,8 @@ crtc_shadow_destroy(xf86CrtcPtr crtc, PixmapPtr rotate_pixmap, void *data)
     /* ScrnInfoPtr pScrn = crtc->scrn; */
 }
 
+#endif
+
 /*
  * Cursor functions
  */
@@ -276,7 +280,21 @@ err_bo_destroy:
 static void
 crtc_load_cursor_argb(xf86CrtcPtr crtc, CARD32 * image)
 {
+    xf86CrtcConfigPtr config = XF86_CRTC_CONFIG_PTR(crtc->scrn);
     modesettingPtr ms = modesettingPTR(crtc->scrn);
+
+    /* Older X servers have cursor reference counting bugs leading to use of
+     * freed memory and consequently random crashes. Should be fixed as of
+     * xserver 1.8, but this workaround shouldn't hurt anyway.
+     */
+    if (config->cursor)
+       config->cursor->refcnt++;
+
+    if (ms->cursor)
+       FreeCursor(ms->cursor, None);
+
+    ms->cursor = config->cursor;
+
     if (ms->screen)
 	crtc_load_cursor_argb_ga3d(crtc, image);
 #ifdef HAVE_LIBKMS
@@ -348,9 +366,9 @@ static const xf86CrtcFuncsRec crtc_funcs = {
     .hide_cursor = crtc_hide_cursor,
     .load_cursor_argb = crtc_load_cursor_argb,
 
-    .shadow_create = crtc_shadow_create,
-    .shadow_allocate = crtc_shadow_allocate,
-    .shadow_destroy = crtc_shadow_destroy,
+    .shadow_create = NULL,
+    .shadow_allocate = NULL,
+    .shadow_destroy = NULL,
 
     .gamma_set = crtc_gamma_set,
     .destroy = crtc_destroy,
diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c
index fd11c60a902..af2273026a7 100644
--- a/src/gallium/state_trackers/xorg/xorg_dri2.c
+++ b/src/gallium/state_trackers/xorg/xorg_dri2.c
@@ -295,6 +295,7 @@ dri2_copy_region(DrawablePtr pDraw, RegionPtr pRegion,
     GCPtr gc;
     RegionPtr copy_clip;
     Bool save_accel;
+    CustomizerPtr cust = ms->cust;
 
     /*
      * In driCreateBuffers we dewrap windows into the
@@ -348,7 +349,8 @@ dri2_copy_region(DrawablePtr pDraw, RegionPtr pRegion,
     ValidateGC(dst_draw, gc);
 
     /* If this is a full buffer swap, throttle on the previous one */
-    if (dst_priv->fence && REGION_NUM_RECTS(pRegion) == 1) {
+    if (ms->swapThrottling &&
+	dst_priv->fence && REGION_NUM_RECTS(pRegion) == 1) {
 	BoxPtr extents = REGION_EXTENTS(pScreen, pRegion);
 
 	if (extents->x1 == 0 && extents->y1 == 0 &&
@@ -370,6 +372,9 @@ dri2_copy_region(DrawablePtr pDraw, RegionPtr pRegion,
     DamageRegionAppend(src_draw, pRegion);
     DamageRegionProcessPending(src_draw);
 
+   if (cust && cust->winsys_context_throttle)
+       cust->winsys_context_throttle(cust, ms->ctx, THROTTLE_SWAP);
+
     (*gc->ops->CopyArea)(src_draw, dst_draw, gc,
 			 0, 0, pDraw->width, pDraw->height, 0, 0);
     ms->exa->accel = save_accel;
@@ -377,8 +382,13 @@ dri2_copy_region(DrawablePtr pDraw, RegionPtr pRegion,
     FreeScratchGC(gc);
 
     ms->ctx->flush(ms->ctx, PIPE_FLUSH_SWAPBUFFERS,
-		   pDestBuffer->attachment == DRI2BufferFrontLeft ?
+		   (pDestBuffer->attachment == DRI2BufferFrontLeft
+		    && ms->swapThrottling) ?
 		   &dst_priv->fence : NULL);
+
+   if (cust && cust->winsys_context_throttle)
+       cust->winsys_context_throttle(cust, ms->ctx, THROTTLE_RENDER);
+
 }
 
 Bool
diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c
index 4f99e0001ba..2aedbb0978c 100644
--- a/src/gallium/state_trackers/xorg/xorg_driver.c
+++ b/src/gallium/state_trackers/xorg/xorg_driver.c
@@ -79,12 +79,16 @@ typedef enum
     OPTION_SW_CURSOR,
     OPTION_2D_ACCEL,
     OPTION_DEBUG_FALLBACK,
+    OPTION_THROTTLE_SWAP,
+    OPTION_THROTTLE_DIRTY
 } drv_option_enums;
 
 static const OptionInfoRec drv_options[] = {
     {OPTION_SW_CURSOR, "SWcursor", OPTV_BOOLEAN, {0}, FALSE},
     {OPTION_2D_ACCEL, "2DAccel", OPTV_BOOLEAN, {0}, FALSE},
     {OPTION_DEBUG_FALLBACK, "DebugFallback", OPTV_BOOLEAN, {0}, FALSE},
+    {OPTION_THROTTLE_SWAP, "SwapThrottling", OPTV_BOOLEAN, {0}, FALSE},
+    {OPTION_THROTTLE_DIRTY, "DirtyThrottling", OPTV_BOOLEAN, {0}, FALSE},
     {-1, NULL, OPTV_NONE, {0}, FALSE}
 };
 
@@ -183,10 +187,12 @@ drv_probe_ddc(ScrnInfoPtr pScrn, int index)
 static Bool
 drv_crtc_resize(ScrnInfoPtr pScrn, int width, int height)
 {
+    xf86CrtcConfigPtr xf86_config = XF86_CRTC_CONFIG_PTR(pScrn);
     modesettingPtr ms = modesettingPTR(pScrn);
     ScreenPtr pScreen = pScrn->pScreen;
     int old_width, old_height;
     PixmapPtr rootPixmap;
+    int i;
 
     if (width == pScrn->virtualX && height == pScrn->virtualY)
 	return TRUE;
@@ -204,13 +210,29 @@ drv_crtc_resize(ScrnInfoPtr pScrn, int width, int height)
 
     pScrn->displayWidth = rootPixmap->devKind / (rootPixmap->drawable.bitsPerPixel / 8);
 
-    if (ms->create_front_buffer(pScrn) && ms->bind_front_buffer(pScrn))
-	return TRUE;
+    if (!ms->create_front_buffer(pScrn) || !ms->bind_front_buffer(pScrn))
+	goto error_create;
 
     /*
-     * This is the error recovery path.
+     * create && bind will turn off all crtc(s) in the kernel so we need to
+     * re-enable all the crtcs again. For real HW we might want to do this
+     * before destroying the old framebuffer.
      */
+    for (i = 0; i < xf86_config->num_crtc; i++) {
+	xf86CrtcPtr crtc = xf86_config->crtc[i];
+
+	if (!crtc->enabled)
+	    continue;
+
+	crtc->funcs->set_mode_major(crtc, &crtc->mode, crtc->rotation, crtc->x, crtc->y);
+    }
+
+    return TRUE;
 
+    /*
+     * This is the error recovery path.
+     */
+error_create:
     if (!pScreen->ModifyPixmapHeader(rootPixmap, old_width, old_height, -1, -1, -1, NULL))
 	FatalError("failed to resize rootPixmap error path\n");
 
@@ -532,23 +554,29 @@ static void drv_block_handler(int i, pointer blockData, pointer pTimeout,
     if (ms->ctx) {
        int j;
 
-       ms->ctx->flush(ms->ctx, PIPE_FLUSH_RENDER_CACHE, &ms->fence[XORG_NR_FENCES-1]);
+       ms->ctx->flush(ms->ctx, PIPE_FLUSH_RENDER_CACHE,
+		      ms->dirtyThrottling ?
+		      &ms->fence[XORG_NR_FENCES-1] :
+		      NULL);
        
-       if (ms->fence[0])
-          ms->ctx->screen->fence_finish(ms->ctx->screen, ms->fence[0], 0);
+       if (ms->dirtyThrottling) {
+	   if (ms->fence[0])
+	       ms->ctx->screen->fence_finish(ms->ctx->screen,
+					     ms->fence[0], 0);
   
-       /* The amount of rendering generated by a block handler can be
-        * quite small.  Let us get a fair way ahead of hardware before
-        * throttling.
-        */
-       for (j = 0; j < XORG_NR_FENCES - 1; j++)
-          ms->screen->fence_reference(ms->screen,
-                                      &ms->fence[j],
-                                      ms->fence[j+1]);
-
-       ms->screen->fence_reference(ms->screen,
-                                   &ms->fence[XORG_NR_FENCES-1],
-                                   NULL);
+	   /* The amount of rendering generated by a block handler can be
+	    * quite small.  Let us get a fair way ahead of hardware before
+	    * throttling.
+	    */
+	   for (j = 0; j < XORG_NR_FENCES - 1; j++)
+	       ms->screen->fence_reference(ms->screen,
+					   &ms->fence[j],
+					   ms->fence[j+1]);
+
+	   ms->screen->fence_reference(ms->screen,
+				       &ms->fence[XORG_NR_FENCES-1],
+				       NULL);
+       }
     }
         
 
@@ -632,6 +660,8 @@ drv_screen_init(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
     unsigned max_width, max_height;
     VisualPtr visual;
     CustomizerPtr cust = ms->cust;
+    MessageType from_st;
+    MessageType from_dt;
 
     if (!drv_init_drm(pScrn)) {
 	FatalError("Could not init DRM");
@@ -719,6 +749,19 @@ drv_screen_init(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
     ms->accelerate_2d = xf86ReturnOptValBool(ms->Options, OPTION_2D_ACCEL, FALSE);
     ms->debug_fallback = xf86ReturnOptValBool(ms->Options, OPTION_DEBUG_FALLBACK, ms->accelerate_2d);
 
+    if (cust && cust->winsys_screen_init)
+	cust->winsys_screen_init(cust, ms->fd);
+
+    ms->swapThrottling = cust ?  cust->swap_throttling : TRUE;
+    from_st = xf86GetOptValBool(ms->Options, OPTION_THROTTLE_SWAP,
+				&ms->swapThrottling) ?
+	X_CONFIG : X_DEFAULT;
+
+    ms->dirtyThrottling = cust ?  cust->dirty_throttling : TRUE;
+    from_dt = xf86GetOptValBool(ms->Options, OPTION_THROTTLE_DIRTY,
+				&ms->dirtyThrottling) ?
+	X_CONFIG : X_DEFAULT;
+
     if (ms->screen) {
 	ms->exa = xorg_exa_init(pScrn, ms->accelerate_2d);
 
@@ -743,6 +786,11 @@ drv_screen_init(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
 #else
     xf86DrvMsg(pScrn->scrnIndex, X_INFO, "3D Acceleration is disabled\n");
 #endif
+    xf86DrvMsg(pScrn->scrnIndex, from_st, "Swap Throttling is %s.\n",
+	       ms->swapThrottling ? "enabled" : "disabled");
+    xf86DrvMsg(pScrn->scrnIndex, from_dt, "Dirty Throttling is %s.\n",
+	       ms->dirtyThrottling ? "enabled" : "disabled");
+
     xf86DrvMsg(pScrn->scrnIndex, X_INFO, "##################################\n");
 
     miInitializeBackingStore(pScreen);
@@ -775,9 +823,6 @@ drv_screen_init(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
     if (serverGeneration == 1)
 	xf86ShowUnusedOptions(pScrn->scrnIndex, pScrn->options);
 
-    if (cust && cust->winsys_screen_init)
-	cust->winsys_screen_init(cust, ms->fd);
-
     return drv_enter_vt(scrnIndex, 1);
 }
 
@@ -904,6 +949,11 @@ drv_close_screen(int scrnIndex, ScreenPtr pScreen)
 	drv_leave_vt(scrnIndex, 0);
     }
 
+    if (ms->cursor) {
+       FreeCursor(ms->cursor, None);
+       ms->cursor = NULL;
+    }
+
     if (cust && cust->winsys_screen_close)
 	cust->winsys_screen_close(cust);
 
@@ -955,7 +1005,7 @@ drv_destroy_front_buffer_ga3d(ScrnInfoPtr pScrn)
 {
     modesettingPtr ms = modesettingPTR(pScrn);
 
-    if (!ms->root_bo)
+    if (!ms->root_texture)
 	return TRUE;
 
     if (ms->fb_id != -1) {
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index 6d89fcaa9f1..3f8b36703a0 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -1035,6 +1035,7 @@ xorg_exa_init(ScrnInfoPtr pScrn, Bool accel)
    modesettingPtr ms = modesettingPTR(pScrn);
    struct exa_context *exa;
    ExaDriverPtr pExa;
+   CustomizerPtr cust = ms->cust;
 
    exa = xcalloc(1, sizeof(struct exa_context));
    if (!exa)
@@ -1093,6 +1094,8 @@ xorg_exa_init(ScrnInfoPtr pScrn, Bool accel)
    exa->pipe = ms->api->create_context(ms->api, exa->scrn);
    /* Share context with DRI */
    ms->ctx = exa->pipe;
+   if (cust && cust->winsys_context_throttle)
+       cust->winsys_context_throttle(cust, ms->ctx, THROTTLE_RENDER);
 
    exa->renderer = renderer_create(exa->pipe);
    exa->accel = accel;
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.c b/src/gallium/state_trackers/xorg/xorg_renderer.c
index 37c8942cff9..e920eb7b133 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.c
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.c
@@ -586,7 +586,7 @@ void renderer_copy_pixmap(struct xorg_renderer *r,
 
 
 void renderer_draw_yuv(struct xorg_renderer *r,
-                       int src_x, int src_y, int src_w, int src_h,
+                       float src_x, float src_y, float src_w, float src_h,
                        int dst_x, int dst_y, int dst_w, int dst_h,
                        struct pipe_texture **textures)
 {
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.h b/src/gallium/state_trackers/xorg/xorg_renderer.h
index 5272cde2b3f..6b9f0917825 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.h
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.h
@@ -53,7 +53,7 @@ void renderer_set_constants(struct xorg_renderer *r,
 
 
 void renderer_draw_yuv(struct xorg_renderer *r,
-                       int src_x, int src_y, int src_w, int src_h,
+                       float src_x, float src_y, float src_w, float src_h,
                        int dst_x, int dst_y, int dst_w, int dst_h,
                        struct pipe_texture **textures);
 
diff --git a/src/gallium/state_trackers/xorg/xorg_tracker.h b/src/gallium/state_trackers/xorg/xorg_tracker.h
index 33e8427612e..570368108cf 100644
--- a/src/gallium/state_trackers/xorg/xorg_tracker.h
+++ b/src/gallium/state_trackers/xorg/xorg_tracker.h
@@ -65,12 +65,22 @@ typedef struct
 
 #define XORG_NR_FENCES 3
 
+enum xorg_throttling_reason {
+    THROTTLE_RENDER,
+    THROTTLE_SWAP
+};
+
 typedef struct _CustomizerRec
 {
+    Bool dirty_throttling;
+    Bool swap_throttling;
     Bool (*winsys_screen_init)(struct _CustomizerRec *cust, int fd);
     Bool (*winsys_screen_close)(struct _CustomizerRec *cust);
     Bool (*winsys_enter_vt)(struct _CustomizerRec *cust);
     Bool (*winsys_leave_vt)(struct _CustomizerRec *cust);
+    void (*winsys_context_throttle)(struct _CustomizerRec *cust,
+				    struct pipe_context *pipe,
+				    enum xorg_throttling_reason reason);
 } CustomizerRec, *CustomizerPtr;
 
 typedef struct _modesettingRec
@@ -88,6 +98,9 @@ typedef struct _modesettingRec
 
     Bool noAccel;
     Bool SWCursor;
+    CursorPtr cursor;
+    Bool swapThrottling;
+    Bool dirtyThrottling;
     CloseScreenProcPtr CloseScreen;
 
     /* Broken-out options. */
diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c
index eb5ce882ec6..4f0f8de37ec 100644
--- a/src/gallium/state_trackers/xorg/xorg_xv.c
+++ b/src/gallium/state_trackers/xorg/xorg_xv.c
@@ -384,7 +384,7 @@ setup_fs_video_constants(struct xorg_renderer *r, boolean hdtv)
 
 static void
 draw_yuv(struct xorg_xv_port_priv *port,
-         int src_x, int src_y, int src_w, int src_h,
+         float src_x, float src_y, float src_w, float src_h,
          int dst_x, int dst_y, int dst_w, int dst_h)
 {
    struct pipe_texture **textures = port->yuv[port->current_set];
@@ -533,10 +533,10 @@ 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 + pPixmap->screen_x;
-      int offset_y = box_y1 - dstY + pPixmap->screen_y;
-      int offset_w;
-      int offset_h;
+      float offset_x = box_x1 - dstX + pPixmap->screen_x;
+      float offset_y = box_y1 - dstY + pPixmap->screen_y;
+      float offset_w;
+      float offset_h;
 
       x = box_x1;
       y = box_y1;
@@ -547,8 +547,8 @@ display_video(ScrnInfoPtr pScrn, struct xorg_xv_port_priv *pPriv, int id,
       offset_h = dst_h - h;
 
       draw_yuv(pPriv,
-               src_x + offset_x*diff_x, src_y + offset_y*diff_y,
-               src_w - offset_w*diff_x, src_h - offset_h*diff_y,
+               (float) src_x + offset_x*diff_x, (float) src_y + offset_y*diff_y,
+               (float) src_w - offset_w*diff_x, (float) src_h - offset_h*diff_y,
                x, y, w, h);
 
       pbox++;
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_context.c b/src/gallium/winsys/drm/vmware/core/vmw_context.c
index b5fd4f5a6a1..b48295425f9 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_context.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_context.c
@@ -103,6 +103,9 @@ struct vmw_svga_winsys_context
     * referred.
     */
    boolean preemptive_flush;
+
+   boolean throttle_set;
+   uint32_t throttle_us;
 };
 
 
@@ -122,6 +125,7 @@ vmw_swc_flush(struct svga_winsys_context *swc,
    struct pipe_fence_handle *fence = NULL;
    unsigned i;
    enum pipe_error ret;
+   uint32_t throttle_us;
 
    ret = pb_validate_validate(vswc->validate);
    assert(ret == PIPE_OK);
@@ -140,8 +144,13 @@ vmw_swc_flush(struct svga_winsys_context *swc,
          *reloc->where = ptr;
       }
 
+      throttle_us = vswc->throttle_set ?
+	 vswc->throttle_us : vswc->vws->default_throttle_us;
+
       if (vswc->command.used)
          vmw_ioctl_command(vswc->vws,
+			   vswc->base.cid,
+			   throttle_us,
                            vswc->command.buffer,
                            vswc->command.used,
                            &vswc->last_fence);
@@ -385,3 +394,14 @@ vmw_svga_context_create(struct pipe_screen *screen)
 {
    return svga_context_create(screen);
 }
+
+void
+vmw_svga_context_set_throttling(struct pipe_context *pipe,
+				uint32_t throttle_us)
+{
+   struct svga_winsys_context *swc = svga_winsys_context(pipe);
+   struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
+
+   vswc->throttle_us = throttle_us;
+   vswc->throttle_set = TRUE;
+}
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_context.h b/src/gallium/winsys/drm/vmware/core/vmw_context.h
index 305ce9b5bec..4e62bc9a64b 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_context.h
+++ b/src/gallium/winsys/drm/vmware/core/vmw_context.h
@@ -55,5 +55,8 @@ vmw_svga_winsys_context_create(struct svga_winsys_screen *sws);
 struct pipe_context *
 vmw_svga_context_create(struct pipe_screen *screen);
 
+void
+vmw_svga_context_set_throttling(struct pipe_context *pipe,
+				uint32_t throttle_us);
 
 #endif /* VMW_CONTEXT_H_ */
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen.c b/src/gallium/winsys/drm/vmware/core/vmw_screen.c
index 6cc9b382932..51a4c55e5a2 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen.c
@@ -75,3 +75,13 @@ vmw_winsys_destroy(struct vmw_winsys_screen *vws)
    vmw_ioctl_cleanup(vws);
    FREE(vws);
 }
+
+void
+vmw_winsys_screen_set_throttling(struct pipe_screen *screen,
+				 uint32_t throttle_us)
+{
+   struct vmw_winsys_screen  *vws =
+      vmw_winsys_screen(svga_winsys_screen(screen));
+
+   vws->default_throttle_us = throttle_us;
+}
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen.h b/src/gallium/winsys/drm/vmware/core/vmw_screen.h
index d3f2c2c7f56..b3de72df881 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen.h
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen.h
@@ -53,6 +53,7 @@ struct vmw_winsys_screen
    struct svga_winsys_screen base;
 
    boolean use_old_scanout_flag;
+   uint32_t default_throttle_us;
 
    struct {
       volatile uint32_t *fifo_map;
@@ -96,9 +97,11 @@ vmw_ioctl_surface_destroy(struct vmw_winsys_screen *vws,
 
 void
 vmw_ioctl_command(struct vmw_winsys_screen *vws,
-                       void *commands,
-                       uint32_t size,
-                       uint32_t *fence);
+		  int32_t cid,
+		  uint32_t throttle_us,
+		  void *commands,
+		  uint32_t size,
+		  uint32_t *fence);
 
 struct vmw_region *
 vmw_ioctl_region_create(struct vmw_winsys_screen *vws, uint32_t size);
@@ -135,6 +138,7 @@ void vmw_pools_cleanup(struct vmw_winsys_screen *vws);
 
 struct vmw_winsys_screen *vmw_winsys_create(int fd, boolean use_old_scanout_flag);
 void vmw_winsys_destroy(struct vmw_winsys_screen *sws);
-
+void vmw_winsys_screen_set_throttling(struct pipe_screen *screen,
+				      uint32_t throttle_us);
 
 #endif /* VMW_SCREEN_H_ */
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
index a9a34ea7302..f369bd8fe43 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
@@ -238,7 +238,7 @@ vmw_dri1_present_locked(struct pipe_context *locked_pipe,
 		   cmd.rect.y,
 		   cmd.rect.w, cmd.rect.h, cmd.rect.srcx, cmd.rect.srcy);
 
-      vmw_ioctl_command(vws, &cmd, sizeof cmd.header + cmd.header.size,
+      vmw_ioctl_command(vws, -1, 0, &cmd, sizeof cmd.header + cmd.header.size,
                         &fence_seq);
       visible = TRUE;
    }
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen_ioctl.c b/src/gallium/winsys/drm/vmware/core/vmw_screen_ioctl.c
index 5d81fa8c4a6..d92ba389d35 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen_ioctl.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen_ioctl.c
@@ -241,8 +241,9 @@ vmw_ioctl_surface_destroy(struct vmw_winsys_screen *vws, uint32 sid)
 }
 
 void
-vmw_ioctl_command(struct vmw_winsys_screen *vws, void *commands, uint32_t size,
-		       uint32_t * pfence)
+vmw_ioctl_command(struct vmw_winsys_screen *vws, int32_t cid,
+		  uint32_t throttle_us, void *commands, uint32_t size,
+		  uint32_t *pfence)
 {
    struct drm_vmw_execbuf_arg arg;
    struct drm_vmw_fence_rep rep;
@@ -275,6 +276,7 @@ vmw_ioctl_command(struct vmw_winsys_screen *vws, void *commands, uint32_t size,
    arg.fence_rep = (unsigned long)&rep;
    arg.commands = (unsigned long)commands;
    arg.command_size = size;
+   arg.throttle_us = throttle_us;
 
    do {
        ret = drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_EXECBUF, &arg, sizeof(arg));
diff --git a/src/gallium/winsys/drm/vmware/xorg/vmw_screen.c b/src/gallium/winsys/drm/vmware/xorg/vmw_screen.c
index f43f91e5c0d..68ec1115a43 100644
--- a/src/gallium/winsys/drm/vmware/xorg/vmw_screen.c
+++ b/src/gallium/winsys/drm/vmware/xorg/vmw_screen.c
@@ -32,9 +32,14 @@
 
 #include "vmw_hook.h"
 #include "vmw_driver.h"
+#include <pipe/p_context.h>
 
 #include "cursorstr.h"
 
+void vmw_winsys_screen_set_throttling(struct pipe_screen *screen,
+				      uint32_t throttle_us);
+
+
 /* modified version of crtc functions */
 xf86CrtcFuncsRec vmw_screen_crtc_funcs;
 
@@ -83,14 +88,55 @@ vmw_screen_cursor_close(struct vmw_customizer *vmw)
 	config->crtc[i]->funcs = vmw->cursor_priv;
 }
 
+static void
+vmw_context_throttle(CustomizerPtr cust,
+		     struct pipe_context *pipe,
+		     enum xorg_throttling_reason reason)
+{
+    switch (reason) {
+    case THROTTLE_RENDER:
+	vmw_winsys_screen_set_throttling(pipe->screen, 20000);
+	break;
+    default:
+      vmw_winsys_screen_set_throttling(pipe->screen, 0);
+    }
+}
+
+static void
+vmw_context_no_throttle(CustomizerPtr cust,
+		     struct pipe_context *pipe,
+		     enum xorg_throttling_reason reason)
+{
+    vmw_winsys_screen_set_throttling(pipe->screen, 0);
+}
+
 static Bool
 vmw_screen_init(CustomizerPtr cust, int fd)
 {
     struct vmw_customizer *vmw = vmw_customizer(cust);
+    drmVersionPtr ver;
 
     vmw->fd = fd;
+    ver = drmGetVersion(fd);
+    if (ver == NULL ||
+	(ver->version_major == 1 && ver->version_minor < 1)) {
+	cust->swap_throttling = TRUE;
+	cust->dirty_throttling = TRUE;
+	cust->winsys_context_throttle = vmw_context_no_throttle;
+    } else {
+	cust->swap_throttling = TRUE;
+	cust->dirty_throttling = FALSE;
+	cust->winsys_context_throttle = vmw_context_throttle;
+	debug_printf("%s: Enabling kernel throttling.\n", __func__);
+    }
+
+    if (ver)
+	drmFreeVersion(ver);
+
     vmw_screen_cursor_init(vmw);
 
+
+
     /* if gallium is used then we don't need to do anything more. */
     if (xorg_has_gallium(vmw->pScrn))
 	return TRUE;
diff --git a/src/glx/x11/glx_pbuffer.c b/src/glx/x11/glx_pbuffer.c
index 501500afc3d..79404b008d3 100644
--- a/src/glx/x11/glx_pbuffer.c
+++ b/src/glx/x11/glx_pbuffer.c
@@ -385,7 +385,8 @@ CreateDrawable(Display * dpy, const __GLcontextModes * fbconfig,
    req->glxwindow = (GLXWindow) XAllocID(dpy);
    req->numAttribs = (CARD32) i;
 
-   memcpy(data, attrib_list, 8 * i);
+   if (attrib_list)
+      memcpy(data, attrib_list, 8 * i);
 
    UnlockDisplay(dpy);
    SyncHandle();
diff --git a/src/glx/x11/indirect_vertex_array.c b/src/glx/x11/indirect_vertex_array.c
index ad9882528ff..ec0e654ceae 100644
--- a/src/glx/x11/indirect_vertex_array.c
+++ b/src/glx/x11/indirect_vertex_array.c
@@ -291,7 +291,8 @@ __glXInitVertexArrayState(__GLXcontext * gc)
 
    arrays->stack_index = 0;
    arrays->stack = malloc(sizeof(struct array_stack_state)
-                          * arrays->num_arrays);
+                          * arrays->num_arrays
+                          * __GL_CLIENT_ATTRIB_STACK_DEPTH);
 }
 
 
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index 66b57e58a71..a1f259084ad 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -794,12 +794,14 @@ static void r300PointParameter(GLcontext * ctx, GLenum pname, const GLfloat * pa
 		R300_STATECHANGE(r300, ga_point_minmax);
 		r300->hw.ga_point_minmax.cmd[1] &= ~R300_GA_POINT_MINMAX_MIN_MASK;
 		r300->hw.ga_point_minmax.cmd[1] |= (GLuint)(ctx->Point.MinSize * 6.0);
+		r300PointSize(ctx, ctx->Point.Size);
 		break;
 	case GL_POINT_SIZE_MAX:
 		R300_STATECHANGE(r300, ga_point_minmax);
 		r300->hw.ga_point_minmax.cmd[1] &= ~R300_GA_POINT_MINMAX_MAX_MASK;
 		r300->hw.ga_point_minmax.cmd[1] |= (GLuint)(ctx->Point.MaxSize * 6.0)
 			<< R300_GA_POINT_MINMAX_MAX_SHIFT;
+		r300PointSize(ctx, ctx->Point.Size);
 		break;
 	case GL_POINT_DISTANCE_ATTENUATION:
 		break;
@@ -1762,8 +1764,6 @@ static void r300ResetHwState(r300ContextPtr r300)
 	if (RADEON_DEBUG & RADEON_STATE)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
-	radeon_firevertices(&r300->radeon);
-
 	r300ColorMask(ctx,
 		      ctx->Color.ColorMask[RCOMP],
 		      ctx->Color.ColorMask[GCOMP],
@@ -1984,23 +1984,6 @@ void r300UpdateShaders(r300ContextPtr rmesa)
 	if (rmesa->options.hw_tcl_enabled) {
 		struct r300_vertex_program *vp;
 
-		if (rmesa->radeon.NewGLState) {
-			int i;
-			for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
-				rmesa->temp_attrib[i] =
-				    TNL_CONTEXT(ctx)->vb.AttribPtr[i];
-				TNL_CONTEXT(ctx)->vb.AttribPtr[i] =
-				    &rmesa->dummy_attrib[i];
-			}
-
-			_tnl_UpdateFixedFunctionProgram(ctx);
-
-			for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
-				TNL_CONTEXT(ctx)->vb.AttribPtr[i] =
-				    rmesa->temp_attrib[i];
-			}
-		}
-
 		vp = r300SelectAndTranslateVertexShader(ctx);
 
 		r300SwitchFallback(ctx, R300_FALLBACK_VERTEX_PROGRAM, vp->error);
diff --git a/src/mesa/drivers/dri/r600/r600_context.c b/src/mesa/drivers/dri/r600/r600_context.c
index 5cfa490a7e0..82d70ccb473 100644
--- a/src/mesa/drivers/dri/r600/r600_context.c
+++ b/src/mesa/drivers/dri/r600/r600_context.c
@@ -320,9 +320,12 @@ static void r600InitGLExtensions(GLcontext *ctx)
 		_mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
 	}
 
-	/* XXX: RV740 only seems to report results from half of its DBs */
-	if (r600->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV740)
-		_mesa_disable_extension(ctx, "GL_ARB_occlusion_query");
+	/* RV740 had a broken pipe config prior to drm 1.32 */
+	if (!r600->radeon.radeonScreen->kernel_mm) {
+		if ((r600->radeon.dri.drmMinor < 32) &&
+		    (r600->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV740))
+			_mesa_disable_extension(ctx, "GL_ARB_occlusion_query");
+	}
 }
 
 /* Create the device specific rendering context.
diff --git a/src/mesa/drivers/dri/r600/r600_texstate.c b/src/mesa/drivers/dri/r600/r600_texstate.c
index 8fdca0bab15..18810817ffc 100644
--- a/src/mesa/drivers/dri/r600/r600_texstate.c
+++ b/src/mesa/drivers/dri/r600/r600_texstate.c
@@ -374,17 +374,11 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, gl_format mesa
 		  break;
 		*/
 	case MESA_FORMAT_RGB_DXT1: /* not supported yet */
-
-		break;
 	case MESA_FORMAT_RGBA_DXT1: /* not supported yet */
-
-		break;
 	case MESA_FORMAT_RGBA_DXT3: /* not supported yet */
-
-		break;
 	case MESA_FORMAT_RGBA_DXT5: /* not supported yet */
+	        return GL_FALSE;
 
-		break;
 	case MESA_FORMAT_RGBA_FLOAT32:
 		SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_32_32_FLOAT,
 			 SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask);
@@ -671,8 +665,8 @@ void r600SetDepthTexMode(struct gl_texture_object *tObj)
 
 	t = radeon_tex_obj(tObj);
 
-	r600GetTexFormat(tObj, tObj->Image[0][tObj->BaseLevel]->TexFormat);
-
+	if(!r600GetTexFormat(tObj, tObj->Image[0][tObj->BaseLevel]->TexFormat))
+	  t->validated = GL_FALSE;
 }
 
 /**
@@ -681,7 +675,7 @@ void r600SetDepthTexMode(struct gl_texture_object *tObj)
  * \param rmesa Context pointer
  * \param t the r300 texture object
  */
-static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *texObj)
+static GLboolean setup_hardware_state(context_t *rmesa, struct gl_texture_object *texObj)
 {
 	radeonTexObj *t = radeon_tex_obj(texObj);
 	const struct gl_texture_image *firstImage;
@@ -690,15 +684,15 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 	if (rmesa->radeon.radeonScreen->driScreen->dri2.enabled &&
 	    t->image_override &&
 	    t->bo)
-		return;
+		return GL_TRUE;
 
 	firstImage = t->base.Image[0][t->minLod];
 
 	if (!t->image_override) {
 		if (!r600GetTexFormat(texObj, firstImage->TexFormat)) {
-			radeon_error("unexpected texture format in %s\n",
-				      __FUNCTION__);
-			return;
+			radeon_warning("unsupported texture format in %s\n",
+				       __FUNCTION__);
+			return GL_FALSE;
 		}
 	}
 
@@ -723,7 +717,7 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 		break;
         default:
 		radeon_error("unexpected texture target type in %s\n", __FUNCTION__);
-		return;
+		return GL_FALSE;
 	}
 
 	row_align = rmesa->radeon.texture_row_align - 1;
@@ -749,6 +743,7 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex
 		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);
 	}
+	return GL_TRUE;
 }
 
 /**
@@ -766,7 +761,8 @@ static GLboolean r600_validate_texture(GLcontext * ctx, struct gl_texture_object
 
 	/* Configure the hardware registers (more precisely, the cached version
 	 * of the hardware registers). */
-	setup_hardware_state(rmesa, texObj);
+ 	if (!setup_hardware_state(rmesa, texObj))
+ 	        return GL_FALSE;
 
 	t->validated = GL_TRUE;
 	return GL_TRUE;
diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c
index 47f89c91f87..5d5b107e5dc 100644
--- a/src/mesa/drivers/dri/r600/r700_render.c
+++ b/src/mesa/drivers/dri/r600/r700_render.c
@@ -815,11 +815,10 @@ static void r700SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
 
 #if MESA_BIG_ENDIAN
     if (mesa_ind_buf->type == GL_UNSIGNED_INT)
-    {
 #else
     if (mesa_ind_buf->type != GL_UNSIGNED_BYTE)
-    {
 #endif
+    {
         const GLvoid *src_ptr;
         GLvoid *dst_ptr;
         GLboolean mapped_named_bo = GL_FALSE;
@@ -856,6 +855,14 @@ static void r700SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer
     }
 }
 
+static GLboolean check_fallbacks(GLcontext *ctx)
+{
+	if (ctx->RenderMode != GL_RENDER)
+		return GL_TRUE;
+
+	return GL_FALSE;
+}
+
 static GLboolean r700TryDrawPrims(GLcontext *ctx,
 				  const struct gl_client_array *arrays[],
 				  const struct _mesa_prim *prim,
@@ -872,6 +879,9 @@ static GLboolean r700TryDrawPrims(GLcontext *ctx,
     if (ctx->NewState)
         _mesa_update_state( ctx );
 
+    if (check_fallbacks(ctx))
+	    return GL_FALSE;
+
     _tnl_UpdateFixedFunctionProgram(ctx);
     r700SetVertexFormat(ctx, arrays, max_index + 1);
     /* shaders need to be updated before buffers are validated */
@@ -966,8 +976,10 @@ static void r700DrawPrims(GLcontext *ctx,
 	retval = r700TryDrawPrims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 
 	/* If failed run tnl pipeline - it should take care of fallbacks */
-	if (!retval)
+	if (!retval) {
+		_swsetup_Wakeup(ctx);
 		_tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+	}
 }
 
 void r700InitDraw(GLcontext *ctx)
diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c
index 82bd48b97b4..11b298bc96a 100644
--- a/src/mesa/drivers/dri/r600/r700_state.c
+++ b/src/mesa/drivers/dri/r600/r700_state.c
@@ -616,7 +616,7 @@ static GLuint translate_logicop(GLenum logicop)
 	case GL_XOR:
 		return 0x66;
 	case GL_EQUIV:
-		return 0xaa;
+		return 0x99;
 	case GL_AND_REVERSE:
 		return 0x44;
 	case GL_AND_INVERTED:
@@ -910,10 +910,12 @@ static void r700PointParameter(GLcontext * ctx, GLenum pname, const GLfloat * pa
 	case GL_POINT_SIZE_MIN:
 		SETfield(r700->PA_SU_POINT_MINMAX.u32All, (int)(ctx->Point.MinSize * 8.0),
 			 MIN_SIZE_shift, MIN_SIZE_mask);
+		r700PointSize(ctx, ctx->Point.Size);
 		break;
 	case GL_POINT_SIZE_MAX:
 		SETfield(r700->PA_SU_POINT_MINMAX.u32All, (int)(ctx->Point.MaxSize * 8.0),
 			 MAX_SIZE_shift, MAX_SIZE_mask);
+		r700PointSize(ctx, ctx->Point.Size);
 		break;
 	case GL_POINT_DISTANCE_ATTENUATION:
 		break;
@@ -1625,8 +1627,6 @@ void r700InitState(GLcontext * ctx) //-------------------
     R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
     int id = 0;
 
-    radeon_firevertices(&context->radeon);
-
     r700->TA_CNTL_AUX.u32All = 0;
     SETfield(r700->TA_CNTL_AUX.u32All, 28, TD_FIFO_CREDIT_shift, TD_FIFO_CREDIT_mask);
     r700->VC_ENHANCE.u32All = 0;
diff --git a/src/mesa/drivers/dri/radeon/radeon_chipset.h b/src/mesa/drivers/dri/radeon/radeon_chipset.h
index 46a9cd5ff84..f17a305bce8 100644
--- a/src/mesa/drivers/dri/radeon/radeon_chipset.h
+++ b/src/mesa/drivers/dri/radeon/radeon_chipset.h
@@ -340,6 +340,7 @@
 #define PCI_CHIP_RS880_9712             0x9712
 #define PCI_CHIP_RS880_9713             0x9713
 #define PCI_CHIP_RS880_9714             0x9714
+#define PCI_CHIP_RS880_9715             0x9715
 
 #define PCI_CHIP_RV770_9440             0x9440
 #define PCI_CHIP_RV770_9441             0x9441
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c
index be2d8365ef5..9eb19b29187 100644
--- a/src/mesa/drivers/dri/radeon/radeon_screen.c
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.c
@@ -828,6 +828,7 @@ static int radeon_set_screen_flags(radeonScreenPtr screen, int device_id)
    case PCI_CHIP_RS880_9712:
    case PCI_CHIP_RS880_9713:
    case PCI_CHIP_RS880_9714:
+   case PCI_CHIP_RS880_9715:
       screen->chip_family = CHIP_FAMILY_RS880;
       screen->chip_flags = RADEON_CHIPSET_TCL;
       break;
diff --git a/src/mesa/main/extensions.c b/src/mesa/main/extensions.c
index fd705ea5b76..07526d09a20 100644
--- a/src/mesa/main/extensions.c
+++ b/src/mesa/main/extensions.c
@@ -45,6 +45,18 @@ static const struct {
    const char *name;
    int flag_offset;
 } default_extensions[] = {
+   /*
+    * TODO: Sort by year, to ensure that old extensions are listed first, so
+    * that old applications that truncate the extension string get the basic
+    * extensions, roughly corresponding to what existed then.
+    *
+    * The best example is quake3demo, which often misses
+    * GL_EXT_compiled_vertex_array because it gets truncated.
+    */
+   { ON,  "GL_EXT_compiled_vertex_array",      F(EXT_compiled_vertex_array) },
+   { OFF, "GL_EXT_texture_compression_s3tc",   F(EXT_texture_compression_s3tc) },
+   { OFF, "GL_EXT_texture_env_add",            F(EXT_texture_env_add) },
+
    { OFF, "GL_ARB_copy_buffer",                F(ARB_copy_buffer) },
    { OFF, "GL_ARB_depth_texture",              F(ARB_depth_texture) },
    { OFF, "GL_ARB_depth_clamp",                F(ARB_depth_clamp) },
@@ -99,7 +111,6 @@ static const struct {
    { OFF, "GL_EXT_blend_subtract",             F(EXT_blend_subtract) },
    { OFF, "GL_EXT_clip_volume_hint",           F(EXT_clip_volume_hint) },
    { OFF, "GL_EXT_cull_vertex",                F(EXT_cull_vertex) },
-   { ON,  "GL_EXT_compiled_vertex_array",      F(EXT_compiled_vertex_array) },
    { OFF, "GL_EXT_convolution",                F(EXT_convolution) },
    { ON,  "GL_EXT_copy_texture",               F(EXT_copy_texture) },
    { OFF, "GL_EXT_depth_bounds_test",          F(EXT_depth_bounds_test) },
@@ -128,10 +139,8 @@ static const struct {
    { ON,  "GL_EXT_subtexture",                 F(EXT_subtexture) },
    { ON,  "GL_EXT_texture",                    F(EXT_texture) },
    { ON,  "GL_EXT_texture3D",                  F(EXT_texture3D) },
-   { OFF, "GL_EXT_texture_compression_s3tc",   F(EXT_texture_compression_s3tc) },
    { OFF, "GL_EXT_texture_cube_map",           F(ARB_texture_cube_map) },
    { ON,  "GL_EXT_texture_edge_clamp",         F(SGIS_texture_edge_clamp) },
-   { OFF, "GL_EXT_texture_env_add",            F(EXT_texture_env_add) },
    { OFF, "GL_EXT_texture_env_combine",        F(EXT_texture_env_combine) },
    { OFF, "GL_EXT_texture_env_dot3",           F(EXT_texture_env_dot3) },
    { OFF, "GL_EXT_texture_filter_anisotropic", F(EXT_texture_filter_anisotropic) },
diff --git a/src/mesa/main/renderbuffer.c b/src/mesa/main/renderbuffer.c
index 3bb062bae2a..1fff668531a 100644
--- a/src/mesa/main/renderbuffer.c
+++ b/src/mesa/main/renderbuffer.c
@@ -1142,7 +1142,8 @@ _mesa_soft_renderbuffer_storage(GLcontext *ctx, struct gl_renderbuffer *rb,
 
    rb->Width = width;
    rb->Height = height;
-   rb->_BaseFormat = _mesa_base_fbo_format(ctx, rb->InternalFormat);
+   rb->_BaseFormat = _mesa_base_fbo_format(ctx, internalFormat);
+   ASSERT(rb->_BaseFormat);
 
    return GL_TRUE;
 }
diff --git a/src/mesa/main/texrender.c b/src/mesa/main/texrender.c
index e2432be6ca4..ac5fdf8d17a 100644
--- a/src/mesa/main/texrender.c
+++ b/src/mesa/main/texrender.c
@@ -499,21 +499,24 @@ update_wrapper(GLcontext *ctx, const struct gl_renderbuffer_attachment *att)
    if (trb->TexImage->TexFormat == MESA_FORMAT_Z24_S8) {
       trb->Base.Format = MESA_FORMAT_Z24_S8;
       trb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT;
+      trb->Base._BaseFormat = GL_DEPTH_STENCIL;
    }
    else if (trb->TexImage->TexFormat == MESA_FORMAT_Z16) {
       trb->Base.Format = MESA_FORMAT_Z16;
       trb->Base.DataType = GL_UNSIGNED_SHORT;
+      trb->Base._BaseFormat = GL_DEPTH_COMPONENT;
    }
    else if (trb->TexImage->TexFormat == MESA_FORMAT_Z32) {
       trb->Base.Format = MESA_FORMAT_Z32;
       trb->Base.DataType = GL_UNSIGNED_INT;
+      trb->Base._BaseFormat = GL_DEPTH_COMPONENT;
    }
    else {
       trb->Base.Format = trb->TexImage->TexFormat;
       trb->Base.DataType = CHAN_TYPE;
+      trb->Base._BaseFormat = GL_RGBA;
    }
    trb->Base.Data = trb->TexImage->Data;
-   trb->Base._BaseFormat = _mesa_base_fbo_format(ctx, trb->Base.InternalFormat);
 }
 
 
diff --git a/src/mesa/main/version.h b/src/mesa/main/version.h
index 1c5749fbcaf..34415923e8a 100644
--- a/src/mesa/main/version.h
+++ b/src/mesa/main/version.h
@@ -32,7 +32,7 @@
 #define MESA_MAJOR 7
 #define MESA_MINOR 7
 #define MESA_PATCH 1
-#define MESA_VERSION_STRING "7.7.1-DEVEL"
+#define MESA_VERSION_STRING "7.7.1-rc1"
 
 /* To make version comparison easy */
 #define MESA_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c
index 3b162d1dad2..786495d3d39 100644
--- a/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -434,7 +434,7 @@ make_texture(struct st_context *st,
 
 /**
  * Draw quad with texcoords and optional color.
- * Coords are window coords with y=0=bottom.
+ * Coords are gallium window coords with y=0=top.
  * \param color  may be null
  * \param invertTex  if true, flip texcoords vertically
  */
@@ -621,10 +621,15 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
       pipe->set_sampler_textures(pipe, 1, &pt);
    }
 
-   /* Compute window coords (y=0=bottom) with pixel zoom.
+   /* Compute Gallium window coords (y=0=top) with pixel zoom.
     * Recall that these coords are transformed by the current
     * vertex shader and viewport transformation.
     */
+   if (st_fb_orientation(ctx->DrawBuffer) == Y_0_BOTTOM) {
+      y = ctx->DrawBuffer->Height - (int) (y + height * ctx->Pixel.ZoomY);
+      invertTex = !invertTex;
+   }
+
    x0 = (GLfloat) x;
    x1 = x + width * ctx->Pixel.ZoomX;
    y0 = (GLfloat) y;
@@ -963,6 +968,7 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy,
    GLfloat *color;
    enum pipe_format srcFormat, texFormat;
    int ptw, pth;
+   GLboolean invertTex = GL_FALSE;
 
    pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 
@@ -1038,8 +1044,8 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy,
       }
    }
 
-   if (st_fb_orientation(ctx->DrawBuffer) == Y_0_TOP) {
-      srcy = ctx->DrawBuffer->Height - srcy - height;
+   if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) {
+      srcy = ctx->ReadBuffer->Height - srcy - height;
 
       if (srcy < 0) {
          height -= -srcy;
@@ -1048,6 +1054,8 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy,
 
       if (height < 0)
          return;
+
+      invertTex = !invertTex;
    }
 
    /* Need to use POT texture? */
@@ -1077,7 +1085,9 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy,
    if (!pt)
       return;
 
-
+   /* Make temporary texture which is a copy of the src region.
+    * We'll draw a quad with this texture to draw the dest image.
+    */
    if (srcFormat == texFormat) {
       /* copy source framebuffer surface into mipmap/texture */
       struct pipe_surface *psRead = screen->get_tex_surface(screen,
@@ -1098,6 +1108,13 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy,
                            psRead,
                            srcx, srcy, width, height);
       }
+
+      if (0) {
+         /* debug */
+         debug_dump_surface("copypixsrcsurf", psRead);
+         debug_dump_surface("copypixtemptex", psTex);
+      }
+
       pipe_surface_reference(&psRead, NULL); 
       pipe_surface_reference(&psTex, NULL);
    }
@@ -1145,7 +1162,7 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy,
    /* draw textured quad */
    draw_textured_quad(ctx, dstx, dsty, ctx->Current.RasterPos[2],
                       width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY,
-                      pt, stvp, stfp, color, GL_TRUE);
+                      pt, stvp, stfp, color, invertTex);
 
    pipe_texture_reference(&pt, NULL);
 }
diff --git a/src/mesa/state_tracker/st_cb_fbo.c b/src/mesa/state_tracker/st_cb_fbo.c
index 9cdff5354b4..d2242575db6 100644
--- a/src/mesa/state_tracker/st_cb_fbo.c
+++ b/src/mesa/state_tracker/st_cb_fbo.c
@@ -420,17 +420,12 @@ st_finish_render_texture(GLcontext *ctx,
 
    st_flush( ctx->st, PIPE_FLUSH_RENDER_CACHE, NULL );
 
-   if (strb->surface)
-      pipe_surface_reference( &strb->surface, NULL );
-
    strb->rtt = NULL;
 
    /*
    printf("FINISH RENDER TO TEXTURE surf=%p\n", strb->surface);
    */
 
-   _mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
-
    /* restore previous framebuffer state */
    st_invalidate_state(ctx, _NEW_BUFFERS);
 }
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index 0858a367a17..49c18a02f03 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -863,6 +863,8 @@ decompress_with_blit(GLcontext * ctx, GLenum target, GLint level,
 
    _mesa_unmap_pbo_dest(ctx, &ctx->Pack);
 
+   screen->tex_transfer_destroy(tex_xfer);
+
    /* destroy the temp / dest surface */
    util_destroy_rgba_surface(dst_texture, dst_surface);
 }