Merge branch 'master' of git+ssh://[email protected]/git/mesa/mesa into glsl-compiler-1

20 files changed, 791 insertions, 327 deletions

diff --git a/src/mesa/drivers/dri/i915tex/Makefile b/src/mesa/drivers/dri/i915tex/Makefile
index 3b3f3f5a3f7..b218929dce7 100644
--- a/src/mesa/drivers/dri/i915tex/Makefile
+++ b/src/mesa/drivers/dri/i915tex/Makefile
@@ -60,7 +60,8 @@ C_SOURCES = \
 
 ASM_SOURCES = 
 
-DRIVER_DEFINES = -I../intel
+DRIVER_DEFINES = -I../intel $(shell pkg-config libdrm --atleast-version=2.3.1 \
+				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
 
 include ../Makefile.template
 
diff --git a/src/mesa/drivers/dri/i915tex/intel_blit.c b/src/mesa/drivers/dri/i915tex/intel_blit.c
index 550669ab0c8..dbe4ba2ac5e 100644
--- a/src/mesa/drivers/dri/i915tex/intel_blit.c
+++ b/src/mesa/drivers/dri/i915tex/intel_blit.c
@@ -55,8 +55,6 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
 
    struct intel_context *intel;
    const intelScreenPrivate *intelScreen;
-   GLboolean missed_target;
-   int64_t ust;
 
    DBG("%s\n", __FUNCTION__);
 
@@ -68,41 +66,6 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
 
    intelScreen = intel->intelScreen;
 
-   if (!rect && !intel->swap_scheduled && intelScreen->drmMinor >= 6 &&
-	!(intel->vblank_flags & VBLANK_FLAG_NO_IRQ) &&
-	intelScreen->current_rotation == 0) {
-      unsigned int interval = driGetVBlankInterval(dPriv, intel->vblank_flags);
-      unsigned int target;
-      drm_i915_vblank_swap_t swap;
- 
-      swap.drawable = dPriv->hHWDrawable;
-      swap.seqtype = DRM_VBLANK_ABSOLUTE;
-      target = swap.sequence = intel->vbl_seq + interval;
-
-      if (intel->vblank_flags & VBLANK_FLAG_SYNC) {
-	 swap.seqtype |= DRM_VBLANK_NEXTONMISS;
-      } else if (interval == 0) {
-	 goto noschedule;
-      }
-
-      if ( intel->vblank_flags & VBLANK_FLAG_SECONDARY ) {
-	 swap.seqtype |= DRM_VBLANK_SECONDARY;
-      }
-
-      intel_batchbuffer_flush(intel->batch);
-
-      if (!drmCommandWriteRead(intel->driFd, DRM_I915_VBLANK_SWAP, &swap,
-			       sizeof(swap))) {
-	 intel->swap_scheduled = 1;
-	 intel->vbl_seq = swap.sequence;
-	 swap.sequence -= target;
-	 missed_target = swap.sequence > 0 && swap.sequence <= (1 << 23);
-      }
-   } else {
-      intel->swap_scheduled = 0;
-   }
-noschedule:
-  
    if (intel->last_swap_fence) {
       driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE);
       driFenceUnReference(intel->last_swap_fence);
@@ -111,122 +74,88 @@ noschedule:
    intel->last_swap_fence = intel->first_swap_fence;
    intel->first_swap_fence = NULL;
 
-   if (!intel->swap_scheduled) {
-      if (!rect) {
-	 driWaitForVBlank(dPriv, &intel->vbl_seq, intel->vblank_flags,
-			  &missed_target);
+   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
+    * should work regardless.
+    */
+   LOCK_HARDWARE(intel);
+
+   if (dPriv && dPriv->numClipRects) {
+      struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+      const struct intel_region *frontRegion
+	 = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT);
+      const struct intel_region *backRegion
+	 = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT);
+      const int nbox = dPriv->numClipRects;
+      const drm_clip_rect_t *pbox = dPriv->pClipRects;
+      const int pitch = frontRegion->pitch;
+      const int cpp = frontRegion->cpp;
+      int BR13, CMD;
+      int i;
+
+      ASSERT(intel_fb);
+      ASSERT(intel_fb->Base.Name == 0);    /* Not a user-created FBO */
+      ASSERT(frontRegion);
+      ASSERT(backRegion);
+      ASSERT(frontRegion->pitch == backRegion->pitch);
+      ASSERT(frontRegion->cpp == backRegion->cpp);
+
+      if (cpp == 2) {
+	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24);
+	 CMD = XY_SRC_COPY_BLT_CMD;
+      }
+      else {
+	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
+	 CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
+		XY_SRC_COPY_BLT_WRITE_RGB);
       }
 
+      for (i = 0; i < nbox; i++, pbox++) {
+	 drm_clip_rect_t box;
 
-      /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
-       * should work regardless.
-       */
-      LOCK_HARDWARE(intel);
-
-      if (intel->driDrawable && intel->driDrawable->numClipRects) {
-	 const intelScreenPrivate *intelScreen = intel->intelScreen;
-	 struct gl_framebuffer *fb
-	    = (struct gl_framebuffer *) dPriv->driverPrivate;
-	 const struct intel_region *frontRegion
-	    = intel_get_rb_region(fb, BUFFER_FRONT_LEFT);
-	 const struct intel_region *backRegion
-	    = intel_get_rb_region(fb, BUFFER_BACK_LEFT);
-	 const int nbox = dPriv->numClipRects;
-	 const drm_clip_rect_t *pbox = dPriv->pClipRects;
-	 const int pitch = frontRegion->pitch;
-	 const int cpp = frontRegion->cpp;
-	 int BR13, CMD;
-	 int i;
-
-	 ASSERT(fb);
-	 ASSERT(fb->Name == 0);    /* Not a user-created FBO */
-	 ASSERT(frontRegion);
-	 ASSERT(backRegion);
-	 ASSERT(frontRegion->pitch == backRegion->pitch);
-	 ASSERT(frontRegion->cpp == backRegion->cpp);
-
-	 if (cpp == 2) {
-	    BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24);
-	    CMD = XY_SRC_COPY_BLT_CMD;
-	 }
-	 else {
-	    BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
-	    CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
-		   XY_SRC_COPY_BLT_WRITE_RGB);
-	 }
+	 if (pbox->x1 > pbox->x2 ||
+	     pbox->y1 > pbox->y2 ||
+	     pbox->x2 > intelScreen->width || pbox->y2 > intelScreen->height)
+	    continue;
 
-	 for (i = 0; i < nbox; i++, pbox++) {
-	    drm_clip_rect_t box;
+	 box = *pbox;
 
-	    if (pbox->x1 > pbox->x2 ||
-		pbox->y1 > pbox->y2 ||
-		pbox->x2 > intelScreen->width || pbox->y2 > intelScreen->height)
-	       continue;
+	 if (rect) {
+	    if (rect->x1 > box.x1)
+	       box.x1 = rect->x1;
+	    if (rect->y1 > box.y1)
+	       box.y1 = rect->y1;
+	    if (rect->x2 < box.x2)
+	       box.x2 = rect->x2;
+	    if (rect->y2 < box.y2)
+	       box.y2 = rect->y2;
 
-	    box = *pbox;
-
-	    if (rect) {
-	       if (rect->x1 > box.x1)
-		  box.x1 = rect->x1;
-	       if (rect->y1 > box.y1)
-		  box.y1 = rect->y1;
-	       if (rect->x2 < box.x2)
-		  box.x2 = rect->x2;
-	       if (rect->y2 < box.y2)
-		  box.y2 = rect->y2;
-
-	       if (box.x1 > box.x2 || box.y1 > box.y2)
-		  continue;
-	    }
-
-	    BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
-	    OUT_BATCH(CMD);
-	    OUT_BATCH(BR13);
-	    OUT_BATCH((pbox->y1 << 16) | pbox->x1);
-	    OUT_BATCH((pbox->y2 << 16) | pbox->x2);
-
-	    if (intel->sarea->pf_current_page == 0)
-	       OUT_RELOC(frontRegion->buffer,
-			 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
-			 DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
-	    else
-	       OUT_RELOC(backRegion->buffer,
-			 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
-			 DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
-	    OUT_BATCH((pbox->y1 << 16) | pbox->x1);
-	    OUT_BATCH(BR13 & 0xffff);
-
-	    if (intel->sarea->pf_current_page == 0)
-	       OUT_RELOC(backRegion->buffer,
-			 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
-			 DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
-	    else
-	       OUT_RELOC(frontRegion->buffer,
-			 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
-			 DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
-
-	    ADVANCE_BATCH();
+	    if (box.x1 > box.x2 || box.y1 > box.y2)
+	       continue;
 	 }
 
-	 if (intel->first_swap_fence)
-	    driFenceUnReference(intel->first_swap_fence);
-	 intel->first_swap_fence = intel_batchbuffer_flush(intel->batch);
-	 driFenceReference(intel->first_swap_fence);
-      }
+	 BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+	 OUT_BATCH(CMD);
+	 OUT_BATCH(BR13);
+	 OUT_BATCH((pbox->y1 << 16) | pbox->x1);
+	 OUT_BATCH((pbox->y2 << 16) | pbox->x2);
 
-      UNLOCK_HARDWARE(intel);
-   }
+	 OUT_RELOC(frontRegion->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+		   DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
+	 OUT_BATCH((pbox->y1 << 16) | pbox->x1);
+	 OUT_BATCH(BR13 & 0xffff);
+	 OUT_RELOC(backRegion->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		   DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
 
-   if (!rect) {
-      intel->swap_count++;
-      (*dri_interface->getUST) (&ust);
-      if (missed_target) {
-         intel->swap_missed_count++;
-         intel->swap_missed_ust = ust - intel->swap_ust;
+	 ADVANCE_BATCH();
       }
 
-      intel->swap_ust = ust;
+      if (intel->first_swap_fence)
+	 driFenceUnReference(intel->first_swap_fence);
+      intel->first_swap_fence = intel_batchbuffer_flush(intel->batch);
+      driFenceReference(intel->first_swap_fence);
    }
+
+   UNLOCK_HARDWARE(intel);
 }
 
 
@@ -406,6 +335,7 @@ void
 intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
 {
    struct intel_context *intel = intel_context(ctx);
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
    GLuint clear_depth;
    GLbitfield skipBuffers = 0;
    BATCH_LOCALS;
@@ -417,7 +347,7 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
     */
    clear_depth = 0;
    if (mask & BUFFER_BIT_DEPTH) {
-      clear_depth = (GLuint) (ctx->DrawBuffer->_DepthMax * ctx->Depth.Clear);
+      clear_depth = (GLuint) (fb->_DepthMax * ctx->Depth.Clear);
    }
    if (mask & BUFFER_BIT_STENCIL) {
       clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
@@ -440,12 +370,12 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
       int i;
 
       /* Get clear bounds after locking */
-      cx = ctx->DrawBuffer->_Xmin;
-      cy = ctx->DrawBuffer->_Ymin;
-      cw = ctx->DrawBuffer->_Xmax - ctx->DrawBuffer->_Xmin;
-      ch = ctx->DrawBuffer->_Ymax - ctx->DrawBuffer->_Ymin;
+      cx = fb->_Xmin;
+      cy = fb->_Ymin;
+      cw = fb->_Xmax - cx;
+      ch = fb->_Ymax - cy;
 
-      if (intel->ctx.DrawBuffer->Name == 0) {
+      if (fb->Name == 0) {
          /* clearing a window */
 
          /* flip top to bottom */
@@ -453,16 +383,6 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
          clear.y1 = intel->driDrawable->y + intel->driDrawable->h - cy - ch;
          clear.x2 = clear.x1 + cw;
          clear.y2 = clear.y1 + ch;
-
-         /* adjust for page flipping */
-         if (intel->sarea->pf_current_page == 1) {
-            const GLuint tmp = mask;
-            mask &= ~(BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_BACK_LEFT);
-            if (tmp & BUFFER_BIT_FRONT_LEFT)
-               mask |= BUFFER_BIT_BACK_LEFT;
-            if (tmp & BUFFER_BIT_BACK_LEFT)
-               mask |= BUFFER_BIT_FRONT_LEFT;
-         }
       }
       else {
          /* clearing FBO */
@@ -480,8 +400,7 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
          drm_clip_rect_t b;
          GLuint buf;
          GLuint clearMask = mask;      /* use copy, since we modify it below */
-         GLboolean all = (cw == ctx->DrawBuffer->Width &&
-                          ch == ctx->DrawBuffer->Height);
+         GLboolean all = (cw == fb->Width && ch == fb->Height);
 
          if (!all) {
             intel_intersect_cliprects(&b, &clear, box);
@@ -499,11 +418,10 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
             const GLbitfield bufBit = 1 << buf;
             if ((clearMask & bufBit) && !(bufBit & skipBuffers)) {
                /* OK, clear this renderbuffer */
-               const struct intel_renderbuffer *irb
-                  = intel_renderbuffer(ctx->DrawBuffer->
-                                       Attachment[buf].Renderbuffer);
+               struct intel_region *irb_region =
+		  intel_get_rb_region(fb, buf);
                struct _DriBufferObject *write_buffer =
-                  intel_region_buffer(intel->intelScreen, irb->region,
+                  intel_region_buffer(intel->intelScreen, irb_region,
                                       all ? INTEL_WRITE_FULL :
                                       INTEL_WRITE_PART);
 
@@ -511,16 +429,15 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
                GLint pitch, cpp;
                GLuint BR13, CMD;
 
-               ASSERT(irb);
-               ASSERT(irb->region);
+               ASSERT(irb_region);
 
-               pitch = irb->region->pitch;
-               cpp = irb->region->cpp;
+               pitch = irb_region->pitch;
+               cpp = irb_region->cpp;
 
                DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
                    __FUNCTION__,
-                   irb->region->buffer, (pitch * cpp),
-                   irb->region->draw_offset,
+                   irb_region->buffer, (pitch * cpp),
+                   irb_region->draw_offset,
                    b.x1, b.y1, b.x2 - b.x1, b.y2 - b.y1);
 
 
@@ -558,6 +475,8 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
                   _mesa_debug(ctx, "hardware blit clear buf %d rb id %d\n",
                   buf, irb->Base.Name);
                 */
+	       intel_wait_flips(intel, INTEL_BATCH_NO_CLIPRECTS);
+
                BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
                OUT_BATCH(CMD);
                OUT_BATCH(BR13);
@@ -565,7 +484,7 @@ intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
                OUT_BATCH((b.y2 << 16) | b.x2);
                OUT_RELOC(write_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
                          DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
-                         irb->region->draw_offset);
+                         irb_region->draw_offset);
                OUT_BATCH(clearVal);
                ADVANCE_BATCH();
                clearMask &= ~bufBit;    /* turn off bit, for faster loop exit */
diff --git a/src/mesa/drivers/dri/i915tex/intel_buffers.c b/src/mesa/drivers/dri/i915tex/intel_buffers.c
index 1ded0b5417f..45fd2fa8de0 100644
--- a/src/mesa/drivers/dri/i915tex/intel_buffers.c
+++ b/src/mesa/drivers/dri/i915tex/intel_buffers.c
@@ -34,6 +34,7 @@
 #include "intel_tris.h"
 #include "intel_regions.h"
 #include "intel_batchbuffer.h"
+#include "intel_reg.h"
 #include "context.h"
 #include "utils.h"
 #include "drirenderbuffer.h"
@@ -42,6 +43,23 @@
 #include "vblank.h"
 
 
+/* This block can be removed when libdrm >= 2.3.1 is required */
+
+#ifndef DRM_VBLANK_FLIP
+
+#define DRM_VBLANK_FLIP 0x8000000
+
+typedef struct drm_i915_flip {
+   int pipes;
+} drm_i915_flip_t;
+
+#undef DRM_IOCTL_I915_FLIP
+#define DRM_IOCTL_I915_FLIP DRM_IOW(DRM_COMMAND_BASE + DRM_I915_FLIP, \
+				    drm_i915_flip_t)
+
+#endif
+
+
 /**
  * XXX move this into a new dri/common/cliprects.c file.
  */
@@ -155,11 +173,14 @@ static void
 intelSetBackClipRects(struct intel_context *intel)
 {
    __DRIdrawablePrivate *dPriv = intel->driDrawable;
+   struct intel_framebuffer *intel_fb;
 
    if (!dPriv)
       return;
 
-   if (intel->sarea->pf_enabled == 0 && dPriv->numBackClipRects == 0) {
+   intel_fb = dPriv->driverPrivate;
+
+   if (intel_fb->pf_active || dPriv->numBackClipRects == 0) {
       /* use the front clip rects */
       intel->numClipRects = dPriv->numClipRects;
       intel->pClipRects = dPriv->pClipRects;
@@ -185,7 +206,7 @@ intelWindowMoved(struct intel_context *intel)
 {
    GLcontext *ctx = &intel->ctx;
    __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   GLframebuffer *drawFb = (GLframebuffer *) dPriv->driverPrivate;
+   struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
 
    if (!intel->ctx.DrawBuffer) {
       /* when would this happen? -BP */
@@ -197,7 +218,7 @@ intelWindowMoved(struct intel_context *intel)
    }
    else {
       /* drawing to a window */
-      switch (drawFb->_ColorDrawBufferMask[0]) {
+      switch (intel_fb->Base._ColorDrawBufferMask[0]) {
       case BUFFER_BIT_FRONT_LEFT:
          intelSetFrontClipRects(intel);
          break;
@@ -210,10 +231,6 @@ intelWindowMoved(struct intel_context *intel)
       }
    }
 
-   /* Update Mesa's notion of window size */
-   driUpdateFramebufferSize(ctx, dPriv);
-   drawFb->Initialized = GL_TRUE; /* XXX remove someday */
-
    if (intel->intelScreen->driScrnPriv->ddxMinor >= 7) {
       drmI830Sarea *sarea = intel->sarea;
       drm_clip_rect_t drw_rect = { .x1 = dPriv->x, .x2 = dPriv->x + dPriv->w,
@@ -226,22 +243,110 @@ intelWindowMoved(struct intel_context *intel)
 				     .y2 = sarea->pipeB_y + sarea->pipeB_h };
       GLint areaA = driIntersectArea( drw_rect, pipeA_rect );
       GLint areaB = driIntersectArea( drw_rect, pipeB_rect );
-      GLuint flags = intel->vblank_flags;
+      GLuint flags = intel_fb->vblank_flags;
+      GLboolean pf_active;
+      GLint pf_pipes;
 
+      /* Update page flipping info
+       */
+      pf_pipes = 0;
+
+      if (areaA > 0)
+	 pf_pipes |= 1;
+
+      if (areaB > 0)
+	 pf_pipes |= 2;
+
+      intel_fb->pf_current_page = (intel->sarea->pf_current_page >>
+				   (intel_fb->pf_pipes & 0x2)) & 0x3;
+
+      intel_fb->pf_num_pages = intel->intelScreen->third.handle ? 3 : 2;
+
+      pf_active = pf_pipes && (pf_pipes & intel->sarea->pf_active) == pf_pipes;
+
+      if (INTEL_DEBUG & DEBUG_LOCK)
+	 if (pf_active != intel_fb->pf_active)
+	    _mesa_printf("%s - Page flipping %sactive\n", __progname,
+			 pf_active ? "" : "in");
+
+      if (pf_active) {
+	 /* Sync pages between pipes if we're flipping on both at the same time */
+	 if (pf_pipes == 0x3 &&	pf_pipes != intel_fb->pf_pipes &&
+	     (intel->sarea->pf_current_page & 0x3) !=
+	     (((intel->sarea->pf_current_page) >> 2) & 0x3)) {
+	    drm_i915_flip_t flip;
+
+	    if (intel_fb->pf_current_page ==
+		(intel->sarea->pf_current_page & 0x3)) {
+	       /* XXX: This is ugly, but emitting two flips 'in a row' can cause
+		* lockups for unknown reasons.
+		*/
+               intel->sarea->pf_current_page =
+		  intel->sarea->pf_current_page & 0x3;
+	       intel->sarea->pf_current_page |=
+		  ((intel_fb->pf_current_page + intel_fb->pf_num_pages - 1) %
+		   intel_fb->pf_num_pages) << 2;
+
+	       flip.pipes = 0x2;
+	    } else {
+               intel->sarea->pf_current_page =
+		  intel->sarea->pf_current_page & (0x3 << 2);
+	       intel->sarea->pf_current_page |=
+		  (intel_fb->pf_current_page + intel_fb->pf_num_pages - 1) %
+		  intel_fb->pf_num_pages;
+
+	       flip.pipes = 0x1;
+	    }
+
+	    drmCommandWrite(intel->driFd, DRM_I915_FLIP, &flip, sizeof(flip));
+	 }
+
+	 intel_fb->pf_pipes = pf_pipes;
+      }
+
+      intel_fb->pf_active = pf_active;
+      intel_flip_renderbuffers(intel_fb);
+      intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer);
+
+      /* Update vblank info
+       */
       if (areaB > areaA || (areaA == areaB && areaB > 0)) {
-	 flags = intel->vblank_flags | VBLANK_FLAG_SECONDARY;
+	 flags = intel_fb->vblank_flags | VBLANK_FLAG_SECONDARY;
       } else {
-	 flags = intel->vblank_flags & ~VBLANK_FLAG_SECONDARY;
+	 flags = intel_fb->vblank_flags & ~VBLANK_FLAG_SECONDARY;
       }
 
-      if (flags != intel->vblank_flags) {
-	 intel->vblank_flags = flags;
-	 driGetCurrentVBlank(dPriv, intel->vblank_flags, &intel->vbl_seq);
+      if (flags != intel_fb->vblank_flags) {
+	 drmVBlank vbl;
+	 int i;
+
+	 vbl.request.type = DRM_VBLANK_ABSOLUTE;
+
+	 if ( intel_fb->vblank_flags & VBLANK_FLAG_SECONDARY ) {
+	    vbl.request.type |= DRM_VBLANK_SECONDARY;
+	 }
+
+	 for (i = 0; i < intel_fb->pf_num_pages; i++) {
+	    vbl.request.sequence = intel_fb->color_rb[i]->vbl_pending;
+	    drmWaitVBlank(intel->driFd, &vbl);
+	 }
+
+	 intel_fb->vblank_flags = flags;
+	 driGetCurrentVBlank(dPriv, intel_fb->vblank_flags, &intel_fb->vbl_seq);
+	 intel_fb->vbl_waited = intel_fb->vbl_seq;
+
+	 for (i = 0; i < intel_fb->pf_num_pages; i++) {
+	    intel_fb->color_rb[i]->vbl_pending = intel_fb->vbl_waited;
+	 }
       }
    } else {
-      intel->vblank_flags &= ~VBLANK_FLAG_SECONDARY;
+      intel_fb->vblank_flags &= ~VBLANK_FLAG_SECONDARY;
    }
 
+   /* Update Mesa's notion of window size */
+   driUpdateFramebufferSize(ctx, dPriv);
+   intel_fb->Base.Initialized = GL_TRUE; /* XXX remove someday */
+
    /* Update hardware scissor */
    ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
                        ctx->Scissor.Width, ctx->Scissor.Height);
@@ -259,6 +364,7 @@ static void
 intelClearWithTris(struct intel_context *intel, GLbitfield mask)
 {
    GLcontext *ctx = &intel->ctx;
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
    drm_clip_rect_t clear;
 
    if (INTEL_DEBUG & DEBUG_BLIT)
@@ -274,10 +380,10 @@ intelClearWithTris(struct intel_context *intel, GLbitfield mask)
       intel->vtbl.install_meta_state(intel);
 
       /* Get clear bounds after locking */
-      cx = ctx->DrawBuffer->_Xmin;
-      cy = ctx->DrawBuffer->_Ymin;
-      ch = ctx->DrawBuffer->_Ymax - ctx->DrawBuffer->_Ymin;
-      cw = ctx->DrawBuffer->_Xmax - ctx->DrawBuffer->_Xmin;
+      cx = fb->_Xmin;
+      cy = fb->_Ymin;
+      ch = fb->_Ymax - cx;
+      cw = fb->_Xmax - cy;
 
       /* note: regardless of 'all', cx, cy, cw, ch are now correct */
       clear.x1 = cx;
@@ -291,9 +397,9 @@ intelClearWithTris(struct intel_context *intel, GLbitfield mask)
       if (mask &
           (BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH)) {
          struct intel_region *backRegion =
-            intel_get_rb_region(ctx->DrawBuffer, BUFFER_BACK_LEFT);
+            intel_get_rb_region(fb, BUFFER_BACK_LEFT);
          struct intel_region *depthRegion =
-            intel_get_rb_region(ctx->DrawBuffer, BUFFER_DEPTH);
+            intel_get_rb_region(fb, BUFFER_DEPTH);
          const GLuint clearColor = (backRegion && backRegion->cpp == 4)
             ? intel->ClearColor8888 : intel->ClearColor565;
 
@@ -330,8 +436,7 @@ intelClearWithTris(struct intel_context *intel, GLbitfield mask)
          const GLuint bufBit = 1 << buf;
          if (mask & bufBit) {
             struct intel_renderbuffer *irbColor =
-               intel_renderbuffer(ctx->DrawBuffer->
-                                  Attachment[buf].Renderbuffer);
+               intel_renderbuffer(fb->Attachment[buf].Renderbuffer);
             GLuint color = (irbColor->region->cpp == 4)
                ? intel->ClearColor8888 : intel->ClearColor565;
 
@@ -372,6 +477,7 @@ intelRotateWindow(struct intel_context *intel,
 {
    intelScreenPrivate *screen = intel->intelScreen;
    drm_clip_rect_t fullRect;
+   struct intel_framebuffer *intel_fb;
    struct intel_region *src;
    const drm_clip_rect_t *clipRects;
    int numClipRects;
@@ -421,16 +527,18 @@ intelRotateWindow(struct intel_context *intel,
 
    intel->vtbl.meta_draw_region(intel, screen->rotated_region, NULL);    /* ? */
 
-   if (srcBuf == BUFFER_BIT_FRONT_LEFT) {
-      src = intel->intelScreen->front_region;
-      clipRects = dPriv->pClipRects;
-      numClipRects = dPriv->numClipRects;
-   }
-   else {
-      src = intel->intelScreen->back_region;
+   intel_fb = dPriv->driverPrivate;
+
+   if ((srcBuf == BUFFER_BIT_BACK_LEFT && !intel_fb->pf_active)) {
+      src = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT);
       clipRects = dPriv->pBackClipRects;
       numClipRects = dPriv->numBackClipRects;
    }
+   else {
+      src = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT);
+      clipRects = dPriv->pClipRects;
+      numClipRects = dPriv->numClipRects;
+   }
 
    if (src->cpp == 4) {
       format = GL_BGRA;
@@ -516,6 +624,7 @@ intelClear(GLcontext *ctx, GLbitfield mask)
    GLbitfield tri_mask = 0;
    GLbitfield blit_mask = 0;
    GLbitfield swrast_mask = 0;
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
    GLuint i;
 
    if (0)
@@ -535,7 +644,7 @@ intelClear(GLcontext *ctx, GLbitfield mask)
    /* HW stencil */
    if (mask & BUFFER_BIT_STENCIL) {
       const struct intel_region *stencilRegion
-         = intel_get_rb_region(ctx->DrawBuffer, BUFFER_STENCIL);
+         = intel_get_rb_region(fb, BUFFER_STENCIL);
       if (stencilRegion) {
          /* have hw stencil */
          if ((ctx->Stencil.WriteMask[0] & 0xff) != 0xff) {
@@ -564,7 +673,7 @@ intelClear(GLcontext *ctx, GLbitfield mask)
    for (i = 0; i < BUFFER_COUNT; i++) {
       GLuint bufBit = 1 << i;
       if ((blit_mask | tri_mask) & bufBit) {
-         if (!ctx->DrawBuffer->Attachment[i].Renderbuffer->ClassID) {
+         if (!fb->Attachment[i].Renderbuffer->ClassID) {
             blit_mask &= ~bufBit;
             tri_mask &= ~bufBit;
             swrast_mask |= bufBit;
@@ -586,15 +695,43 @@ intelClear(GLcontext *ctx, GLbitfield mask)
 }
 
 
+/* Emit wait for pending flips */
+void
+intel_wait_flips(struct intel_context *intel, GLuint batch_flags)
+{
+   struct intel_framebuffer *intel_fb =
+      (struct intel_framebuffer *) intel->ctx.DrawBuffer;
+   struct intel_renderbuffer *intel_rb =
+      intel_get_renderbuffer(&intel_fb->Base,
+			     intel_fb->Base._ColorDrawBufferMask[0] ==
+			     BUFFER_BIT_FRONT_LEFT ? BUFFER_FRONT_LEFT :
+			     BUFFER_BACK_LEFT);
+
+   if (intel_fb->Base.Name == 0 && intel_rb->pf_pending == intel_fb->pf_seq) {
+      GLint pf_pipes = intel_fb->pf_pipes;
+      BATCH_LOCALS;
+
+      /* Wait for pending flips to take effect */
+      BEGIN_BATCH(2, batch_flags);
+      OUT_BATCH(pf_pipes & 0x1 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_A_FLIP)
+		: 0);
+      OUT_BATCH(pf_pipes & 0x2 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_B_FLIP)
+		: 0);
+      ADVANCE_BATCH();
+
+      intel_rb->pf_pending--;
+   }
+}
+
 
 /* Flip the front & back buffers
  */
-static void
+static GLboolean
 intelPageFlip(const __DRIdrawablePrivate * dPriv)
 {
-#if 0
    struct intel_context *intel;
-   int tmp, ret;
+   int ret;
+   struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
 
    if (INTEL_DEBUG & DEBUG_IOCTL)
       fprintf(stderr, "%s\n", __FUNCTION__);
@@ -605,28 +742,45 @@ intelPageFlip(const __DRIdrawablePrivate * dPriv)
 
    intel = (struct intel_context *) dPriv->driContextPriv->driverPrivate;
 
+   if (intel->intelScreen->drmMinor < 9)
+      return GL_FALSE;
+
    intelFlush(&intel->ctx);
+
+   ret = 0;
+
    LOCK_HARDWARE(intel);
 
-   if (dPriv->pClipRects) {
-      *(drm_clip_rect_t *) intel->sarea->boxes = dPriv->pClipRects[0];
-      intel->sarea->nbox = 1;
-   }
+   if (dPriv->numClipRects && intel_fb->pf_active) {
+      drm_i915_flip_t flip;
 
-   ret = drmCommandNone(intel->driFd, DRM_I830_FLIP);
-   if (ret) {
-      fprintf(stderr, "%s: %d\n", __FUNCTION__, ret);
-      UNLOCK_HARDWARE(intel);
-      exit(1);
+      flip.pipes = intel_fb->pf_pipes;
+
+      ret = drmCommandWrite(intel->driFd, DRM_I915_FLIP, &flip, sizeof(flip));
    }
 
-   tmp = intel->sarea->last_enqueue;
-   intelRefillBatchLocked(intel);
    UNLOCK_HARDWARE(intel);
 
+   if (ret || !intel_fb->pf_active)
+      return GL_FALSE;
 
-   intelSetDrawBuffer(&intel->ctx, intel->ctx.Color.DriverDrawBuffer);
-#endif
+   if (!dPriv->numClipRects) {
+      usleep(10000);	/* throttle invisible client 10ms */
+   }
+
+   intel_fb->pf_current_page = (intel->sarea->pf_current_page >>
+				(intel_fb->pf_pipes & 0x2)) & 0x3;
+
+   if (dPriv->numClipRects != 0) {
+      intel_get_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT)->pf_pending =
+      intel_get_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT)->pf_pending =
+	 ++intel_fb->pf_seq;
+   }
+
+   intel_flip_renderbuffers(intel_fb);
+   intel_draw_buffer(&intel->ctx, &intel_fb->Base);
+
+   return GL_TRUE;
 }
 
 #if 0
@@ -641,7 +795,7 @@ intelSwapBuffers(__DRIdrawablePrivate * dPriv)
          if (ctx && ctx->DrawBuffer == fb) {
             _mesa_notifySwapBuffers(ctx);       /* flush pending rendering */
          }
-         if (0 /*intel->doPageFlip */ ) {       /* doPageFlip is never set !!! */
+         if (intel->doPageFlip) {
             intelPageFlip(dPriv);
          }
          else {
@@ -657,6 +811,83 @@ intelSwapBuffers(__DRIdrawablePrivate * dPriv)
 #else
 /* Trunk version:
  */
+
+static GLboolean
+intelScheduleSwap(const __DRIdrawablePrivate * dPriv, GLboolean *missed_target)
+{
+   struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+   unsigned int interval = driGetVBlankInterval(dPriv, intel_fb->vblank_flags);
+   struct intel_context *intel =
+      intelScreenContext(dPriv->driScreenPriv->private);
+   const intelScreenPrivate *intelScreen = intel->intelScreen;
+   unsigned int target;
+   drm_i915_vblank_swap_t swap;
+   GLboolean ret;
+
+   if ((intel_fb->vblank_flags & VBLANK_FLAG_NO_IRQ) ||
+       intelScreen->current_rotation != 0 ||
+       intelScreen->drmMinor < (intel_fb->pf_active ? 9 : 6))
+      return GL_FALSE;
+
+   swap.seqtype = DRM_VBLANK_ABSOLUTE;
+
+   if (intel_fb->vblank_flags & VBLANK_FLAG_SYNC) {
+      swap.seqtype |= DRM_VBLANK_NEXTONMISS;
+   } else if (interval == 0) {
+      return GL_FALSE;
+   }
+
+   swap.drawable = dPriv->hHWDrawable;
+   target = swap.sequence = intel_fb->vbl_seq + interval;
+
+   if ( intel_fb->vblank_flags & VBLANK_FLAG_SECONDARY ) {
+      swap.seqtype |= DRM_VBLANK_SECONDARY;
+   }
+
+   LOCK_HARDWARE(intel);
+
+   intel_batchbuffer_flush(intel->batch);
+
+   if ( intel_fb->pf_active ) {
+      swap.seqtype |= DRM_VBLANK_FLIP;
+
+      intel_fb->pf_current_page = (((intel->sarea->pf_current_page >>
+				     (intel_fb->pf_pipes & 0x2)) & 0x3) + 1) %
+				  intel_fb->pf_num_pages;
+   }
+
+   if (!drmCommandWriteRead(intel->driFd, DRM_I915_VBLANK_SWAP, &swap,
+			    sizeof(swap))) {
+      intel_fb->vbl_seq = swap.sequence;
+      swap.sequence -= target;
+      *missed_target = swap.sequence > 0 && swap.sequence <= (1 << 23);
+
+      intel_get_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT)->vbl_pending =
+	 intel_get_renderbuffer(&intel_fb->Base,
+				BUFFER_FRONT_LEFT)->vbl_pending =
+	 intel_fb->vbl_seq;
+
+      if (swap.seqtype & DRM_VBLANK_FLIP) {
+	 intel_flip_renderbuffers(intel_fb);
+	 intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer);
+      }
+
+      ret = GL_TRUE;
+   } else {
+      if (swap.seqtype & DRM_VBLANK_FLIP) {
+	 intel_fb->pf_current_page = ((intel->sarea->pf_current_page >>
+					(intel_fb->pf_pipes & 0x2)) & 0x3) %
+				     intel_fb->pf_num_pages;
+      }
+
+      ret = GL_FALSE;
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+   return ret;
+}
+  
 void
 intelSwapBuffers(__DRIdrawablePrivate * dPriv)
 {
@@ -671,16 +902,34 @@ intelSwapBuffers(__DRIdrawablePrivate * dPriv)
 
       if (ctx->Visual.doubleBufferMode) {
          intelScreenPrivate *screen = intel->intelScreen;
-         _mesa_notifySwapBuffers(ctx);  /* flush pending rendering comands */
-         if (0 /*intel->doPageFlip */ ) {       /* doPageFlip is never set !!! */
-            intelPageFlip(dPriv);
-         }
-         else {
-            intelCopyBuffer(dPriv, NULL);
-         }
-         if (screen->current_rotation != 0) {
-            intelRotateWindow(intel, dPriv, BUFFER_BIT_FRONT_LEFT);
-         }
+	 GLboolean missed_target;
+	 struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+	 int64_t ust;
+         
+	 _mesa_notifySwapBuffers(ctx);  /* flush pending rendering comands */
+
+         if (screen->current_rotation != 0 ||
+	     !intelScheduleSwap(dPriv, &missed_target)) {
+	    driWaitForVBlank(dPriv, &intel_fb->vbl_seq, intel_fb->vblank_flags,
+			     &missed_target);
+
+	    if (screen->current_rotation != 0 || !intelPageFlip(dPriv)) {
+	       intelCopyBuffer(dPriv, NULL);
+	    }
+
+	    if (screen->current_rotation != 0) {
+	       intelRotateWindow(intel, dPriv, BUFFER_BIT_FRONT_LEFT);
+	    }
+	 }
+
+	 intel_fb->swap_count++;
+	 (*dri_interface->getUST) (&ust);
+	 if (missed_target) {
+	    intel_fb->swap_missed_count++;
+	    intel_fb->swap_missed_ust = ust - intel_fb->swap_ust;
+	 }
+
+	 intel_fb->swap_ust = ust;
       }
    }
    else {
@@ -788,10 +1037,6 @@ intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb)
     */
    if (fb->Name == 0) {
       /* drawing to window system buffer */
-      if (intel->sarea->pf_current_page == 1) {
-         /* page flipped back/front */
-         front ^= 1;
-      }
       if (front) {
          intelSetFrontClipRects(intel);
          colorRegion = intel_get_rb_region(fb, BUFFER_FRONT_LEFT);
diff --git a/src/mesa/drivers/dri/i915tex/intel_buffers.h b/src/mesa/drivers/dri/i915tex/intel_buffers.h
index 0faf0553474..3b686cb5c18 100644
--- a/src/mesa/drivers/dri/i915tex/intel_buffers.h
+++ b/src/mesa/drivers/dri/i915tex/intel_buffers.h
@@ -30,6 +30,7 @@
 
 
 struct intel_context;
+struct intel_framebuffer;
 
 
 extern GLboolean
@@ -41,6 +42,8 @@ extern struct intel_region *intel_readbuf_region(struct intel_context *intel);
 
 extern struct intel_region *intel_drawbuf_region(struct intel_context *intel);
 
+extern void intel_wait_flips(struct intel_context *intel, GLuint batch_flags);
+
 extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
 
 extern void intelWindowMoved(struct intel_context *intel);
diff --git a/src/mesa/drivers/dri/i915tex/intel_context.c b/src/mesa/drivers/dri/i915tex/intel_context.c
index a5ce08b1701..5c2cdf0c7d6 100644
--- a/src/mesa/drivers/dri/i915tex/intel_context.c
+++ b/src/mesa/drivers/dri/i915tex/intel_context.c
@@ -59,6 +59,7 @@
 #include "intel_buffer_objects.h"
 #include "intel_fbo.h"
 
+#include "drirenderbuffer.h"
 #include "vblank.h"
 #include "utils.h"
 #include "xmlpool.h"            /* for symbolic values of enum-type options */
@@ -442,10 +443,6 @@ intelInitContext(struct intel_context *intel,
 
    intel->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS);
 
-   intel->vblank_flags = (intel->intelScreen->irq_active != 0)
-      ? driGetDefaultVBlankFlags(&intel->optionCache) : VBLANK_FLAG_NO_IRQ;
-
-   (*dri_interface->getUST) (&intel->swap_ust);
    _math_matrix_ctr(&intel->ViewportMatrix);
 
    /* Disable imaging extension until convolution is working in
@@ -550,27 +547,30 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
    if (driContextPriv) {
       struct intel_context *intel =
          (struct intel_context *) driContextPriv->driverPrivate;
-      GLframebuffer *drawFb = (GLframebuffer *) driDrawPriv->driverPrivate;
+      struct intel_framebuffer *intel_fb =
+	 (struct intel_framebuffer *) driDrawPriv->driverPrivate;
       GLframebuffer *readFb = (GLframebuffer *) driReadPriv->driverPrivate;
 
 
       /* XXX FBO temporary fix-ups! */
       /* if the renderbuffers don't have regions, init them from the context */
       {
-         struct intel_renderbuffer *irbFront
-            = intel_get_renderbuffer(drawFb, BUFFER_FRONT_LEFT);
-         struct intel_renderbuffer *irbBack
-            = intel_get_renderbuffer(drawFb, BUFFER_BACK_LEFT);
          struct intel_renderbuffer *irbDepth
-            = intel_get_renderbuffer(drawFb, BUFFER_DEPTH);
+            = intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH);
          struct intel_renderbuffer *irbStencil
-            = intel_get_renderbuffer(drawFb, BUFFER_STENCIL);
+            = intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL);
 
-         if (irbFront && !irbFront->region) {
-            intel_region_reference(&irbFront->region, intel->intelScreen->front_region);
+         if (intel_fb->color_rb[0] && !intel_fb->color_rb[0]->region) {
+            intel_region_reference(&intel_fb->color_rb[0]->region,
+				   intel->intelScreen->front_region);
+         }
+         if (intel_fb->color_rb[1] && !intel_fb->color_rb[1]->region) {
+            intel_region_reference(&intel_fb->color_rb[1]->region,
+				   intel->intelScreen->back_region);
          }
-         if (irbBack && !irbBack->region) {
-            intel_region_reference(&irbBack->region, intel->intelScreen->back_region);
+         if (intel_fb->color_rb[2] && !intel_fb->color_rb[2]->region) {
+            intel_region_reference(&intel_fb->color_rb[2]->region,
+				   intel->intelScreen->third_region);
          }
          if (irbDepth && !irbDepth->region) {
             intel_region_reference(&irbDepth->region, intel->intelScreen->depth_region);
@@ -580,29 +580,34 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
          }
       }
 
-      /* set initial GLframebuffer size to match window, if needed */
-      if (drawFb->Width == 0 && driDrawPriv->w) {
-         _mesa_resize_framebuffer(&intel->ctx, drawFb,
+      /* set GLframebuffer size to match window, if needed */
+      if (intel_fb->Base.Width != driDrawPriv->w) {
+         _mesa_resize_framebuffer(&intel->ctx, &intel_fb->Base,
                                   driDrawPriv->w, driDrawPriv->h);
       }         
-      if (readFb->Width == 0 && driReadPriv->w) {
+      if (readFb->Width != driReadPriv->w) {
          _mesa_resize_framebuffer(&intel->ctx, readFb,
                                   driReadPriv->w, driReadPriv->h);
       }         
 
-      _mesa_make_current(&intel->ctx, drawFb, readFb);
+      _mesa_make_current(&intel->ctx, &intel_fb->Base, readFb);
 
       /* The drawbuffer won't always be updated by _mesa_make_current: 
        */
-      if (intel->ctx.DrawBuffer == drawFb) {
+      if (intel->ctx.DrawBuffer == &intel_fb->Base) {
 
 	 if (intel->driDrawable != driDrawPriv) {
-	    driDrawableInitVBlank(driDrawPriv, intel->vblank_flags, &intel->vbl_seq);	    
+	    intel_fb->vblank_flags = (intel->intelScreen->irq_active != 0)
+	       ? driGetDefaultVBlankFlags(&intel->optionCache)
+	       : VBLANK_FLAG_NO_IRQ;
+	    (*dri_interface->getUST) (&intel_fb->swap_ust);
+	    driDrawableInitVBlank(driDrawPriv, intel_fb->vblank_flags,
+				  &intel_fb->vbl_seq);
 	    intel->driDrawable = driDrawPriv;
 	    intelWindowMoved(intel);
 	 }
 
-	 intel_draw_buffer(&intel->ctx, drawFb);
+	 intel_draw_buffer(&intel->ctx, &intel_fb->Base);
       }
    }
    else {
@@ -683,7 +688,6 @@ intelContendedLock(struct intel_context *intel, GLuint flags)
       intel->current_rotation = sarea->rotation;
    }
 
-
    /* Drawable changed?
     */
    if (dPriv && intel->lastStamp != dPriv->lastStamp) {
@@ -693,24 +697,40 @@ intelContendedLock(struct intel_context *intel, GLuint flags)
 }
 
 
+
 /* Lock the hardware and validate our state.  
  */
 void LOCK_HARDWARE( struct intel_context *intel )
 {
     char __ret=0;
-
+    struct intel_framebuffer *intel_fb = NULL;
+    struct intel_renderbuffer *intel_rb = NULL;
     _glthread_LOCK_MUTEX(lockMutex);
     assert(!intel->locked);
 
-    if (intel->swap_scheduled) {
+    if (intel->driDrawable) {
+       intel_fb = intel->driDrawable->driverPrivate;
+
+       if (intel_fb)
+	  intel_rb =
+	     intel_get_renderbuffer(&intel_fb->Base,
+				    intel_fb->Base._ColorDrawBufferMask[0] ==
+				    BUFFER_BIT_FRONT_LEFT ? BUFFER_FRONT_LEFT :
+				    BUFFER_BACK_LEFT);
+    }
+
+    if (intel_rb && (intel_fb->vbl_waited - intel_rb->vbl_pending) > (1<<23)) {
 	drmVBlank vbl;
+
 	vbl.request.type = DRM_VBLANK_ABSOLUTE;
-	if ( intel->vblank_flags & VBLANK_FLAG_SECONDARY ) {
+
+	if ( intel_fb->vblank_flags & VBLANK_FLAG_SECONDARY ) {
 	    vbl.request.type |= DRM_VBLANK_SECONDARY;
 	}
-	vbl.request.sequence = intel->vbl_seq;
+
+	vbl.request.sequence = intel_rb->vbl_pending;
 	drmWaitVBlank(intel->driFd, &vbl);
-	intel->swap_scheduled = 0;
+	intel_fb->vbl_waited = vbl.reply.sequence;
     }
 
     DRM_CAS(intel->driHwLock, intel->hHWContext,
diff --git a/src/mesa/drivers/dri/i915tex/intel_context.h b/src/mesa/drivers/dri/i915tex/intel_context.h
index 96b911501f3..44c20af7f80 100644
--- a/src/mesa/drivers/dri/i915tex/intel_context.h
+++ b/src/mesa/drivers/dri/i915tex/intel_context.h
@@ -274,19 +274,6 @@ struct intel_context
     */
    driOptionCache optionCache;
 
-   /* VBI
-    */
-   GLuint vbl_seq;
-   GLuint vblank_flags;
-
-   int64_t swap_ust;
-   int64_t swap_missed_ust;
-
-   GLuint swap_count;
-   GLuint swap_missed_count;
-
-   GLuint swap_scheduled;
-
   /* Rotation. Need to match that of the
    * current screen.
    */
@@ -294,7 +281,6 @@ struct intel_context
   int width;
   int height;
   int current_rotation;
-
 };
 
 /* These are functions now:
diff --git a/src/mesa/drivers/dri/i915tex/intel_fbo.c b/src/mesa/drivers/dri/i915tex/intel_fbo.c
index ab0e569bd94..8d430553822 100644
--- a/src/mesa/drivers/dri/i915tex/intel_fbo.c
+++ b/src/mesa/drivers/dri/i915tex/intel_fbo.c
@@ -75,11 +75,35 @@ intel_get_renderbuffer(struct gl_framebuffer *fb, GLuint attIndex)
 }
 
 
+void
+intel_flip_renderbuffers(struct intel_framebuffer *intel_fb)
+{
+   int current_page = intel_fb->pf_current_page;
+   int next_page = (current_page + 1) % intel_fb->pf_num_pages;
+
+   if (intel_fb->color_rb[current_page] &&
+       intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer !=
+       &intel_fb->color_rb[current_page]->Base) {
+      _mesa_remove_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT);
+      _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT,
+			     &intel_fb->color_rb[current_page]->Base);
+   }
+
+   if (intel_fb->color_rb[next_page] &&
+       intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer !=
+       &intel_fb->color_rb[next_page]->Base) {
+      _mesa_remove_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT);
+      _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT,
+			     &intel_fb->color_rb[next_page]->Base);
+   }
+}
+
+
 struct intel_region *
 intel_get_rb_region(struct gl_framebuffer *fb, GLuint attIndex)
 {
-   struct intel_renderbuffer *irb
-      = intel_renderbuffer(fb->Attachment[attIndex].Renderbuffer);
+   struct intel_renderbuffer *irb = intel_get_renderbuffer(fb, attIndex);
+
    if (irb)
       return irb->region;
    else
@@ -94,7 +118,9 @@ intel_get_rb_region(struct gl_framebuffer *fb, GLuint attIndex)
 static struct gl_framebuffer *
 intel_new_framebuffer(GLcontext * ctx, GLuint name)
 {
-   /* there's no intel_framebuffer at this time, just use Mesa's class */
+   /* Only drawable state in intel_framebuffer at this time, just use Mesa's
+    * class
+    */
    return _mesa_new_framebuffer(ctx, name);
 }
 
@@ -271,10 +297,24 @@ static GLboolean
 intel_alloc_window_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
                            GLenum internalFormat, GLuint width, GLuint height)
 {
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_framebuffer *intel_fb;
+
    ASSERT(rb->Name == 0);
    rb->Width = width;
    rb->Height = height;
    rb->_ActualFormat = internalFormat;
+
+   if (intel && intel->driDrawable &&
+       (intel_fb = intel->driDrawable->driverPrivate) &&
+       intel_fb->pf_num_pages == 3 &&
+       rb == &intel_fb->color_rb[intel_fb->pf_current_page]->Base &&
+       (rb = &intel_fb->color_rb[(intel_fb->pf_current_page + 2) % 3]->Base)) {
+      rb->Width = width;
+      rb->Height = height;
+      rb->_ActualFormat = internalFormat;
+   }
+
    return GL_TRUE;
 }
 
diff --git a/src/mesa/drivers/dri/i915tex/intel_fbo.h b/src/mesa/drivers/dri/i915tex/intel_fbo.h
index 221f09b39ab..963f5e706f4 100644
--- a/src/mesa/drivers/dri/i915tex/intel_fbo.h
+++ b/src/mesa/drivers/dri/i915tex/intel_fbo.h
@@ -32,6 +32,35 @@
 struct intel_context;
 struct intel_region;
 
+/**
+ * Intel framebuffer, derived from gl_framebuffer.
+ */
+struct intel_framebuffer
+{
+   struct gl_framebuffer Base;
+
+   struct intel_renderbuffer *color_rb[3];
+
+   /* Drawable page flipping state */
+   GLboolean pf_active;
+   GLuint pf_seq;
+   GLint pf_pipes;
+   GLint pf_current_page;
+   GLint pf_num_pages;
+
+   /* VBI
+    */
+   GLuint vbl_seq;
+   GLuint vblank_flags;
+   GLuint vbl_waited;
+
+   int64_t swap_ust;
+   int64_t swap_missed_ust;
+
+   GLuint swap_count;
+   GLuint swap_missed_count;
+};
+
 
 /**
  * Intel renderbuffer, derived from gl_renderbuffer.
@@ -49,6 +78,10 @@ struct intel_renderbuffer
 
    GLuint PairedDepth;   /**< only used if this is a depth renderbuffer */
    GLuint PairedStencil; /**< only used if this is a stencil renderbuffer */
+
+   GLuint pf_pending;  /**< sequence number of pending flip */
+
+   GLuint vbl_pending;   /**< vblank sequence number of pending flip */
 };
 
 
@@ -69,6 +102,8 @@ extern struct intel_renderbuffer *intel_get_renderbuffer(struct gl_framebuffer
                                                          *fb,
                                                          GLuint attIndex);
 
+extern void intel_flip_renderbuffers(struct intel_framebuffer *intel_fb);
+
 
 /* XXX make inline or macro */
 extern struct intel_region *intel_get_rb_region(struct gl_framebuffer *fb,
diff --git a/src/mesa/drivers/dri/i915tex/intel_reg.h b/src/mesa/drivers/dri/i915tex/intel_reg.h
index 1ec153266c7..7828ba6ad39 100644
--- a/src/mesa/drivers/dri/i915tex/intel_reg.h
+++ b/src/mesa/drivers/dri/i915tex/intel_reg.h
@@ -81,4 +81,8 @@
 #define XY_SRC_COPY_BLT_WRITE_ALPHA     (1<<21)
 #define XY_SRC_COPY_BLT_WRITE_RGB       (1<<20)
 
+#define MI_WAIT_FOR_EVENT               ((0x3<<23))
+#define MI_WAIT_FOR_PLANE_B_FLIP        (1<<6)
+#define MI_WAIT_FOR_PLANE_A_FLIP        (1<<2)
+
 #endif
diff --git a/src/mesa/drivers/dri/i915tex/intel_screen.c b/src/mesa/drivers/dri/i915tex/intel_screen.c
index ac83254593b..a6342046b5f 100644
--- a/src/mesa/drivers/dri/i915tex/intel_screen.c
+++ b/src/mesa/drivers/dri/i915tex/intel_screen.c
@@ -98,6 +98,18 @@ intelMapScreenRegions(__DRIscreenPrivate * sPriv)
       return GL_FALSE;
    }
 
+   if (intelScreen->third.handle) {
+      if (0)
+	 _mesa_printf("Third 0x%08x ", intelScreen->third.handle);
+      if (drmMap(sPriv->fd,
+		 intelScreen->third.handle,
+		 intelScreen->third.size,
+		 (drmAddress *) & intelScreen->third.map) != 0) {
+	 intelUnmapScreenRegions(intelScreen);
+	 return GL_FALSE;
+      }
+   }
+
    if (0)
       _mesa_printf("Depth 0x%08x ", intelScreen->depth.handle);
    if (drmMap(sPriv->fd,
@@ -119,9 +131,9 @@ intelMapScreenRegions(__DRIscreenPrivate * sPriv)
    }
 #endif
    if (0)
-      printf("Mappings:  front: %p  back: %p  depth: %p  tex: %p\n",
+      printf("Mappings:  front: %p  back: %p  third: %p  depth: %p  tex: %p\n",
              intelScreen->front.map,
-             intelScreen->back.map,
+             intelScreen->back.map, intelScreen->third.map,
              intelScreen->depth.map, intelScreen->tex.map);
    return GL_TRUE;
 }
@@ -191,6 +203,18 @@ intel_recreate_static_regions(intelScreenPrivate *intelScreen)
 			    intelScreen->back.pitch / intelScreen->cpp,
 			    intelScreen->height);
 
+   if (intelScreen->third.handle) {
+      intelScreen->third_region =
+	 intel_recreate_static(intelScreen,
+			       intelScreen->third_region,
+			       DRM_BO_FLAG_MEM_TT,
+			       intelScreen->third.offset,
+			       intelScreen->third.map,
+			       intelScreen->cpp,
+			       intelScreen->third.pitch / intelScreen->cpp,
+			       intelScreen->height);
+   }
+
    /* Still assuming front.cpp == depth.cpp
     */
    intelScreen->depth_region =
@@ -240,6 +264,13 @@ intelUnmapScreenRegions(intelScreenPrivate * intelScreen)
 #endif
       intelScreen->back.map = NULL;
    }
+   if (intelScreen->third.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->third.map, intelScreen->third.size) != 0)
+         printf("drmUnmap third failed!\n");
+#endif
+      intelScreen->third.map = NULL;
+   }
    if (intelScreen->depth.map) {
 #if REALLY_UNMAP
       drmUnmap(intelScreen->depth.map, intelScreen->depth.size);
@@ -325,6 +356,13 @@ intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen,
    intelScreen->back.handle = sarea->back_handle;
    intelScreen->back.size = sarea->back_size;
 
+   if (intelScreen->driScrnPriv->ddxMinor >= 8) {
+      intelScreen->third.offset = sarea->third_offset;
+      intelScreen->third.pitch = sarea->pitch * intelScreen->cpp;
+      intelScreen->third.handle = sarea->third_handle;
+      intelScreen->third.size = sarea->third_size;
+   }
+
    intelScreen->depth.offset = sarea->depth_offset;
    intelScreen->depth.pitch = sarea->pitch * intelScreen->cpp;
    intelScreen->depth.handle = sarea->depth_handle;
@@ -541,31 +579,49 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
                              mesaVis->depthBits != 24);
       GLenum rgbFormat = (mesaVis->redBits == 5 ? GL_RGB5 : GL_RGBA8);
 
-      struct gl_framebuffer *fb = _mesa_create_framebuffer(mesaVis);
+      struct intel_framebuffer *intel_fb = CALLOC_STRUCT(intel_framebuffer);
+
+      if (!intel_fb)
+	 return GL_FALSE;
+
+      _mesa_initialize_framebuffer(&intel_fb->Base, mesaVis);
 
       /* setup the hardware-based renderbuffers */
       {
-         struct intel_renderbuffer *frontRb
+         intel_fb->color_rb[0]
             = intel_create_renderbuffer(rgbFormat,
                                         screen->width, screen->height,
                                         screen->front.offset,
                                         screen->front.pitch,
                                         screen->cpp,
                                         screen->front.map);
-         intel_set_span_functions(&frontRb->Base);
-         _mesa_add_renderbuffer(fb, BUFFER_FRONT_LEFT, &frontRb->Base);
+         intel_set_span_functions(&intel_fb->color_rb[0]->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT,
+				&intel_fb->color_rb[0]->Base);
       }
 
       if (mesaVis->doubleBufferMode) {
-         struct intel_renderbuffer *backRb
+         intel_fb->color_rb[1]
             = intel_create_renderbuffer(rgbFormat,
                                         screen->width, screen->height,
                                         screen->back.offset,
                                         screen->back.pitch,
                                         screen->cpp,
                                         screen->back.map);
-         intel_set_span_functions(&backRb->Base);
-         _mesa_add_renderbuffer(fb, BUFFER_BACK_LEFT, &backRb->Base);
+         intel_set_span_functions(&intel_fb->color_rb[1]->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT,
+				&intel_fb->color_rb[1]->Base);
+
+	 if (screen->third.handle) {
+	    intel_fb->color_rb[2]
+	       = intel_create_renderbuffer(rgbFormat,
+					   screen->width, screen->height,
+					   screen->third.offset,
+					   screen->third.pitch,
+					   screen->cpp,
+					   screen->third.map);
+	    intel_set_span_functions(&intel_fb->color_rb[2]->Base);
+	 }
       }
 
       if (mesaVis->depthBits == 24 && mesaVis->stencilBits == 8) {
@@ -579,8 +635,10 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
                                         screen->depth.map);
          intel_set_span_functions(&depthStencilRb->Base);
          /* note: bind RB to two attachment points */
-         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthStencilRb->Base);
-         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &depthStencilRb->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH,
+				&depthStencilRb->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_STENCIL,
+				&depthStencilRb->Base);
       }
       else if (mesaVis->depthBits == 16) {
          /* just 16-bit depth buffer, no hw stencil */
@@ -592,17 +650,19 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
                                         screen->cpp,    /* 2! */
                                         screen->depth.map);
          intel_set_span_functions(&depthRb->Base);
-         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, &depthRb->Base);
       }
 
       /* now add any/all software-based renderbuffers we may need */
-      _mesa_add_soft_renderbuffers(fb, GL_FALSE,        /* never sw color */
-                                   GL_FALSE,    /* never sw depth */
-                                   swStencil, mesaVis->accumRedBits > 0, GL_FALSE,      /* never sw alpha */
-                                   GL_FALSE /* never sw aux */ );
-      driDrawPriv->driverPrivate = (void *) fb;
-
-      return (driDrawPriv->driverPrivate != NULL);
+      _mesa_add_soft_renderbuffers(&intel_fb->Base,
+                                   GL_FALSE, /* never sw color */
+                                   GL_FALSE, /* never sw depth */
+                                   swStencil, mesaVis->accumRedBits > 0,
+                                   GL_FALSE, /* never sw alpha */
+                                   GL_FALSE  /* never sw aux */ );
+      driDrawPriv->driverPrivate = (void *) intel_fb;
+
+      return GL_TRUE;
    }
 }
 
@@ -619,21 +679,20 @@ intelDestroyBuffer(__DRIdrawablePrivate * driDrawPriv)
 static int
 intelGetSwapInfo(__DRIdrawablePrivate * dPriv, __DRIswapInfo * sInfo)
 {
-   struct intel_context *intel;
+   struct intel_framebuffer *intel_fb;
 
-   if ((dPriv == NULL) || (dPriv->driContextPriv == NULL)
-       || (dPriv->driContextPriv->driverPrivate == NULL)
+   if ((dPriv == NULL) || (dPriv->driverPrivate == NULL)
        || (sInfo == NULL)) {
       return -1;
    }
 
-   intel = dPriv->driContextPriv->driverPrivate;
-   sInfo->swap_count = intel->swap_count;
-   sInfo->swap_ust = intel->swap_ust;
-   sInfo->swap_missed_count = intel->swap_missed_count;
+   intel_fb = dPriv->driverPrivate;
+   sInfo->swap_count = intel_fb->swap_count;
+   sInfo->swap_ust = intel_fb->swap_ust;
+   sInfo->swap_missed_count = intel_fb->swap_missed_count;
 
    sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0)
-      ? driCalculateSwapUsage(dPriv, 0, intel->swap_missed_ust)
+      ? driCalculateSwapUsage(dPriv, 0, intel_fb->swap_missed_ust)
       : 0.0;
 
    return 0;
diff --git a/src/mesa/drivers/dri/i915tex/intel_screen.h b/src/mesa/drivers/dri/i915tex/intel_screen.h
index 17698773f3d..05e2f1f2eac 100644
--- a/src/mesa/drivers/dri/i915tex/intel_screen.h
+++ b/src/mesa/drivers/dri/i915tex/intel_screen.h
@@ -51,12 +51,14 @@ typedef struct
 {
    intelRegion front;
    intelRegion back;
+   intelRegion third;
    intelRegion rotated;
    intelRegion depth;
    intelRegion tex;
 
    struct intel_region *front_region;
    struct intel_region *back_region;
+   struct intel_region *third_region;
    struct intel_region *depth_region;
    struct intel_region *rotated_region;
 
diff --git a/src/mesa/drivers/dri/i915tex/intel_tris.c b/src/mesa/drivers/dri/i915tex/intel_tris.c
index 1ba49d8f6ec..5fe3d4561fc 100644
--- a/src/mesa/drivers/dri/i915tex/intel_tris.c
+++ b/src/mesa/drivers/dri/i915tex/intel_tris.c
@@ -43,6 +43,7 @@
 #include "intel_context.h"
 #include "intel_tris.h"
 #include "intel_batchbuffer.h"
+#include "intel_buffers.h"
 #include "intel_reg.h"
 #include "intel_span.h"
 #include "intel_tex.h"
@@ -102,6 +103,8 @@ intelStartInlinePrimitive(struct intel_context *intel,
 
 /*    _mesa_printf("%s *", __progname); */
 
+   intel_wait_flips(intel, batch_flags);
+
    /* Emit a slot which will be filled with the inline primitive
     * command later.
     */
diff --git a/src/mesa/drivers/dri/i915tex/server/i830_common.h b/src/mesa/drivers/dri/i915tex/server/i830_common.h
index 06f28ed19a4..d4d58886ce0 100644
--- a/src/mesa/drivers/dri/i915tex/server/i830_common.h
+++ b/src/mesa/drivers/dri/i915tex/server/i830_common.h
@@ -129,6 +129,12 @@ typedef struct {
 	int pipeB_y;
 	int pipeB_w;
 	int pipeB_h;
+
+	/* Triple buffering */
+	drm_handle_t third_handle;
+	int third_offset;
+	int third_size;
+	unsigned int third_tiled;
 } drmI830Sarea;
 
 /* Flags for perf_boxes
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_fifo.c b/src/mesa/drivers/dri/nouveau/nouveau_fifo.c
index 7af9f1e3c2f..67b5aa4f8a8 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_fifo.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_fifo.c
@@ -38,6 +38,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "nouveau_object.h"
 #include "nouveau_sync.h"
 
+#ifdef NOUVEAU_RING_DEBUG
+int nouveau_fifo_remaining=0;
+#endif
+
 
 #define RING_SKIPS 8
 
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_fifo.h b/src/mesa/drivers/dri/nouveau/nouveau_fifo.h
index 490089f71a4..23325dcea53 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_fifo.h
+++ b/src/mesa/drivers/dri/nouveau/nouveau_fifo.h
@@ -74,17 +74,33 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
 #ifdef NOUVEAU_RING_DEBUG
 
+extern int nouveau_fifo_remaining;
+
 #define OUT_RINGp(ptr,sz) do {                                                  \
 uint32_t* p=(uint32_t*)(ptr);							\
 int i; printf("OUT_RINGp: (size 0x%x dwords)\n",sz); for(i=0;i<sz;i++) printf(" 0x%08x   %f\n", *(p+i), *((float*)(p+i))); 	\
+nouveau_fifo_remaining-=sz;							\
 }while(0)
 
 #define OUT_RING(n) do {                                                        \
     printf("OUT_RINGn: 0x%08x (%s)\n", n, __func__);                            \
+    nouveau_fifo_remaining--;							\
 }while(0)
 
 #define OUT_RINGf(n) do {                                                       \
     printf("OUT_RINGf: %.04f (%s)\n", n, __func__);                             \
+    nouveau_fifo_remaining--;							\
+}while(0)
+
+#define BEGIN_RING_SIZE(subchannel,tag,size) do {					\
+	if (nouveau_fifo_remaining!=0)							\
+		printf("RING ERROR : remaining %d\n",nouveau_fifo_remaining);		\
+	nouveau_state_cache_flush(nmesa);						\
+	if (nmesa->fifo.free <= (size))							\
+		WAIT_RING(nmesa,(size));						\
+	OUT_RING( ((size)<<18) | ((subchannel) << 13) | (tag));				\
+	nmesa->fifo.free -= ((size) + 1);                                               \
+	nouveau_fifo_remaining=size;							\
 }while(0)
 
 #else
@@ -110,8 +126,6 @@ if (NOUVEAU_RING_TRACE) \
 *((float*)(nmesa->fifo.buffer+nmesa->fifo.current++))=(n);			\
 }while(0)
 
-#endif
-
 #define BEGIN_RING_SIZE(subchannel,tag,size) do {					\
 	nouveau_state_cache_flush(nmesa);						\
 	if (nmesa->fifo.free <= (size))							\
@@ -120,6 +134,8 @@ if (NOUVEAU_RING_TRACE) \
 	nmesa->fifo.free -= ((size) + 1);                                               \
 }while(0)
 
+#endif
+
 extern void WAIT_RING(nouveauContextPtr nmesa,u_int32_t size);
 extern void nouveau_state_cache_flush(nouveauContextPtr nmesa);
 extern void nouveau_state_cache_init(nouveauContextPtr nmesa);
diff --git a/src/mesa/drivers/dri/nouveau/nv04_state.c b/src/mesa/drivers/dri/nouveau/nv04_state.c
index 4129ecc50ea..25df3d2a624 100644
--- a/src/mesa/drivers/dri/nouveau/nv04_state.c
+++ b/src/mesa/drivers/dri/nouveau/nv04_state.c
@@ -455,35 +455,37 @@ static GLboolean nv04BindBuffers(nouveauContextPtr nmesa, int num_color,
 		nouveau_renderbuffer *depth)
 {
 	GLuint x, y, w, h;
+	uint32_t depth_pitch=(depth?depth->pitch:0+15)&~15+16;
+	if (depth_pitch<256) depth_pitch=256;
 
 	w = color[0]->mesa.Width;
 	h = color[0]->mesa.Height;
 	x = nmesa->drawX;
 	y = nmesa->drawY;
 
+	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_FORMAT, 1);
+	if (color[0]->mesa._ActualFormat == GL_RGBA8)
+		OUT_RING(0x108/*A8R8G8B8*/);
+	else
+		OUT_RING(0x103/*R5G6B5*/);
+
 	/* FIXME pitches have to be aligned ! */
 	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
-	OUT_RING(color[0]->pitch|(depth?(depth->pitch<<16):0));
+	OUT_RING(color[0]->pitch|(depth_pitch<<16));
 	OUT_RING(color[0]->offset);
-
 	if (depth) {
 		BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
 		OUT_RING(depth->offset);
 	}
 
-	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_CLIP_HORIZONTAL, 2);
-	OUT_RING((w<<16)|x);
-	OUT_RING((h<<16)|y);
+//	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_CLIP_HORIZONTAL, 2);
+//	OUT_RING((w<<16)|x);
+//	OUT_RING((h<<16)|y);
 
-	/* FIXME not sure... */
-	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_CLIP_SIZE, 1);
-	OUT_RING((h<<16)|w);
 
-	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_FORMAT, 1);
-	if (color[0]->mesa._ActualFormat == GL_RGBA8)
-		OUT_RING(108/*A8R8G8B8*/);
-	else
-		OUT_RING(103/*R5G6B5*/);
+	/* FIXME not sure... */
+/*	BEGIN_RING_SIZE(NvSubCtxSurf3D, NV04_CONTEXT_SURFACES_3D_CLIP_SIZE, 1);
+	OUT_RING((h<<16)|w);*/
 
 	return GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/nouveau/nv04_swtcl.c b/src/mesa/drivers/dri/nouveau/nv04_swtcl.c
index 9b5332b77a7..cb072e0bdbc 100644
--- a/src/mesa/drivers/dri/nouveau/nv04_swtcl.c
+++ b/src/mesa/drivers/dri/nouveau/nv04_swtcl.c
@@ -473,6 +473,7 @@ static inline void nv04OutputVertexFormat(struct nouveau_context* nmesa)
 	/*
 	 * Tell t_vertex about the vertex format
 	 */
+	nmesa->vertex_attr_count = 0;
 	RENDERINPUTS_COPY(index, nmesa->render_inputs_bitset);
 
 	// SX SY SZ INVW
@@ -503,7 +504,7 @@ static inline void nv04OutputVertexFormat(struct nouveau_context* nmesa)
 	nmesa->vertex_size=_tnl_install_attrs( ctx,
 			nmesa->vertex_attrs, 
 			nmesa->vertex_attr_count,
-			ctx->Viewport._WindowMap.m, 0 );
+			nmesa->viewport.m, 0 );
 }
 
 
diff --git a/src/mesa/main/getstring.c b/src/mesa/main/getstring.c
index e41e9899ef6..973649da0dd 100644
--- a/src/mesa/main/getstring.c
+++ b/src/mesa/main/getstring.c
@@ -114,7 +114,8 @@ _mesa_GetString( GLenum name )
                       ctx->Extensions.ARB_shader_objects &&
                       ctx->Extensions.ARB_vertex_shader &&
                       ctx->Extensions.ARB_fragment_shader &&
-                      ctx->Extensions.ARB_texture_non_power_of_two) {
+                      ctx->Extensions.ARB_texture_non_power_of_two &&
+                      ctx->Extensions.EXT_blend_equation_separate) {
                      if (ctx->Extensions.ARB_shading_language_120 &&
                          ctx->Extensions.EXT_pixel_buffer_object &&
                          ctx->Extensions.EXT_texture_sRGB) {
diff --git a/src/mesa/x86/mmx_blend.S b/src/mesa/x86/mmx_blend.S
index 397d635a621..20ac5a20adf 100644
--- a/src/mesa/x86/mmx_blend.S
+++ b/src/mesa/x86/mmx_blend.S
@@ -1,4 +1,4 @@
-
+	;
 /*
  * Written by Jos� Fonseca <[email protected]>
  */
@@ -240,7 +240,9 @@ TWO(PUNPCKHDQ  ( MA2, MA2 ))                    /*    pa2    |    pa2    |    pa
 ONE(MOVD       ( MSS, REGIND(rgba) ))		/*     |     |     |     | sa1 | sb1 | sg1 | sr1 */	;\
 TWO(MOVQ       ( MSS, REGIND(rgba) ))		/* sa2 | sb2 | sg2 | sr2 | sa1 | sb1 | sg1 | sr1 */
 
-
+/* Kevin F. Quinn <[email protected]> 2 July 2006
+ * Replace data segment constants with text-segment
+ * constants (via pushl/movq)
     SEG_DATA
 
 ALIGNDATA8
@@ -249,6 +251,11 @@ const_0080:
 
 const_80:
     D_LONG 0x80808080, 0x80808080
+*/
+#define const_0080_l 0x00800080
+#define const_0080_h 0x00800080
+#define const_80_l 0x80808080
+#define const_80_h 0x80808080
 
     SEG_TEXT
 
@@ -302,8 +309,16 @@ TWO(MOVQ       ( MM1, REGIND(rgba) ))
 #define TAG(x) CONCAT(x,_min)
 #define LLTAG(x) LLBL2(x,_min)
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
+#define INIT \
+    MOVQ       ( CONTENT(const_80), MM7 )
+ */
 #define INIT \
-    MOVQ       ( CONTENT(const_80), MM7 )	/* 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80*/
+    PUSH_L     ( CONST(const_80_h) ) 		/* 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80*/	;\
+    PUSH_L     ( CONST(const_80_l) ) 									;\
+    MOVQ       ( REGIND(ESP), MM7 ) 									;\
+    ADD_L      ( CONST(8), ESP)
 
 #define MAIN( rgba, dest ) \
     GMB_LOAD( rgba, dest, MM1, MM2 )									;\
@@ -326,8 +341,16 @@ TWO(MOVQ       ( MM1, REGIND(rgba) ))
 #define TAG(x) CONCAT(x,_max)
 #define LLTAG(x) LLBL2(x,_max)
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 #define INIT \
-    MOVQ       ( CONTENT(const_80), MM7 )	/* 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80*/
+    MOVQ       ( CONTENT(const_80), MM7 )
+ */
+#define INIT \
+    PUSH_L     ( CONST(const_80_l) ) 		/* 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80| 0x80*/	;\
+    PUSH_L     ( CONST(const_80_h) ) 									;\
+    MOVQ       ( REGIND(ESP), MM7 ) 									;\
+    ADD_L      ( CONST(8), ESP)
 
 #define MAIN( rgba, dest ) \
     GMB_LOAD( rgba, dest, MM1, MM2 )									;\
@@ -350,9 +373,17 @@ TWO(MOVQ       ( MM1, REGIND(rgba) ))
 #define TAG(x) CONCAT(x,_modulate)
 #define LLTAG(x) LLBL2(x,_modulate)
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
+#define INIT \
+    MOVQ       ( CONTENT(const_0080), MM7 )
+ */
 #define INIT \
     PXOR       ( MM0, MM0 )			/*   0x0000  |   0x0000  |   0x0000  |   0x0000  */	;\
-    MOVQ       ( CONTENT(const_0080), MM7 )	/*   0x0080  |   0x0080  |   0x0080  |   0x0080  */
+    PUSH_L     ( CONST(const_0080_l) ) 	/*   0x0080  |   0x0080  |   0x0080  |   0x0080  */	;\
+    PUSH_L     ( CONST(const_0080_h) ) 								;\
+    MOVQ       ( REGIND(ESP), MM7 ) 									;\
+    ADD_L      ( CONST(8), ESP)
 
 #define MAIN( rgba, dest ) \
     GMB_LOAD( rgba, dest, MM1, MM2 )									;\
diff --git a/src/mesa/x86/read_rgba_span_x86.S b/src/mesa/x86/read_rgba_span_x86.S
index 960cffa50bc..a6902838164 100644
--- a/src/mesa/x86/read_rgba_span_x86.S
+++ b/src/mesa/x86/read_rgba_span_x86.S
@@ -32,6 +32,8 @@
 
 	.file	"read_rgba_span_x86.S"
 #if !defined(__DJGPP__) && !defined(__MINGW32__) /* this one cries for assyntax.h */
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	.section	.rodata
 	.align 16
 	.type	mask, @object
@@ -45,6 +47,19 @@ mask:
 	.long	0x00ff0000
 	.long	0x00ff0000
 	.long	0x00ff0000
+ */
+#define	LOAD_MASK(mvins,m1,m2) \
+   	pushl	$0xff00ff00 ;\
+   	pushl	$0xff00ff00 ;\
+   	pushl	$0xff00ff00 ;\
+   	pushl	$0xff00ff00 ;\
+	mvins	(%esp), m1	;\
+   	pushl	$0x00ff0000 ;\
+   	pushl	$0x00ff0000 ;\
+   	pushl	$0x00ff0000 ;\
+   	pushl	$0x00ff0000 ;\
+	mvins	(%esp), m2	;\
+	addl	$32, %esp
 
 
 /* I implemented these as macros because the appear in quite a few places,
@@ -84,8 +99,12 @@ _generic_read_RGBA_span_BGRA8888_REV_MMX:
 #ifdef USE_INNER_EMMS
 	emms
 #endif
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	movq	mask, %mm1
 	movq	mask+16, %mm2
+ */
+	LOAD_MASK(movq,%mm1,%mm2)
 
 	movl	8(%esp), %ebx	/* source pointer */
 	movl	16(%esp), %edx	/* number of pixels to copy */
@@ -182,8 +201,12 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE:
 #ifdef USE_INNER_EMMS
 	emms
 #endif
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	movq	mask, %mm1
 	movq	mask+16, %mm2
+ */
+	LOAD_MASK(movq,%mm1,%mm2)
 
 	movl	16(%esp), %ebx	/* source pointer */
 	movl	24(%esp), %edx	/* number of pixels to copy */
@@ -341,8 +364,12 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE2:
 	pushl	%esi
 	pushl	%ebx
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	movdqa	mask, %xmm1
 	movdqa	mask+16, %xmm2
+ */
+	LOAD_MASK(movdqa,%xmm1,%xmm2)
 
 	movl	12(%esp), %ebx	/* source pointer */
 	movl	20(%esp), %edx	/* number of pixels to copy */
@@ -464,6 +491,10 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE2:
 
 
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
+ */
+#if 0
 	.section	.rodata
 
 	.align	16
@@ -510,6 +541,26 @@ scale:
 
 alpha:	.long	0x00000000
 	.long	0x00ff0000
+#endif
+
+#define MASK_565_L	0x07e0f800
+#define MASK_565_H	0x0000001f
+#define SCALE_ADJUST	5
+#if SCALE_ADJUST == 5
+#define PRESCALE_L 0x00100001
+#define PRESCALE_H 0x00000200
+#define SCALE_L 0x40C620E8
+#define SCALE_H 0x0000839d
+#elif SCALE_ADJUST == 0
+#define PRESCALE_L 0x00200001
+#define PRESCALE_H 0x00000800
+#define SCALE_L 0x01040108
+#define SCALE_H 0x00000108
+#else
+#error SCALE_ADJUST must either be 5 or 0.
+#endif
+#define ALPHA_L 0x00000000
+#define ALPHA_H 0x00ff0000
 
 /**
  * MMX optimized version of the RGB565 to RGBA copy routine.
@@ -530,9 +581,25 @@ _generic_read_RGBA_span_RGB565_MMX:
 	movl	8(%esp), %edx	/* destination pointer */
 	movl	12(%esp), %ecx	/* number of pixels to copy */
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	movq	mask_565, %mm5
 	movq	prescale, %mm6
 	movq	scale, %mm7
+ */
+ 	pushl	MASK_565_H
+ 	pushl	MASK_565_L
+	movq	(%esp), %mm5
+ 	pushl	PRESCALE_H
+ 	pushl	PRESCALE_L
+	movq	(%esp), %mm6
+ 	pushl	SCALE_H
+ 	pushl	SCALE_L
+	movq	(%esp), %mm7
+ 	pushl	ALPHA_H
+ 	pushl	ALPHA_L
+	movq	(%esp), %mm3
+	addl	$32,%esp
 
 	sarl	$2, %ecx
 	jle	.L01		/* Bail early if the count is negative. */
@@ -581,8 +648,13 @@ _generic_read_RGBA_span_RGB565_MMX:
 	/* Always set the alpha value to 0xff.
 	 */
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	por	alpha, %mm0
 	por	alpha, %mm2
+ */
+ 	por %mm3, %mm0
+ 	por %mm3, %mm2
 
 
 	/* Pack the 16-bit values to 8-bit values and store the converted
@@ -609,8 +681,13 @@ _generic_read_RGBA_span_RGB565_MMX:
 	pmulhuw	%mm7, %mm0
 	pmulhuw	%mm7, %mm2
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	por	alpha, %mm0
 	por	alpha, %mm2
+ */
+ 	por %mm3, %mm0
+ 	por %mm3, %mm2
 
 	packuswb	%mm2, %mm0
 
@@ -647,8 +724,13 @@ _generic_read_RGBA_span_RGB565_MMX:
 	pmulhuw	%mm7, %mm0
 	pmulhuw	%mm7, %mm2
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	por	alpha, %mm0
 	por	alpha, %mm2
+ */
+ 	por %mm3, %mm0
+ 	por %mm3, %mm2
 
 	packuswb	%mm2, %mm0
 
@@ -675,7 +757,11 @@ _generic_read_RGBA_span_RGB565_MMX:
 #endif
 	pmulhuw	%mm7, %mm0
 
+/* Kevin F. Quinn 2nd July 2006
+ * Replace data segment constants with text-segment instructions
 	por	alpha, %mm0
+ */
+ 	por %mm3, %mm0
 
 	packuswb	%mm0, %mm0