static struct pipe_surface *
rbug_context_create_surface(struct pipe_context *_pipe,
                            struct pipe_resource *_resource,
                            const struct pipe_surface *surf_tmpl)
{
   struct rbug_context *rb_pipe = rbug_context(_pipe);
   struct rbug_resource *rb_resource = rbug_resource(_resource);
   struct pipe_context *pipe = rb_pipe->pipe;
   struct pipe_resource *resource = rb_resource->resource;
   struct pipe_surface *result;

   pipe_mutex_lock(rb_pipe->call_mutex);
   result = pipe->create_surface(pipe,
                                 resource,
                                 surf_tmpl);
   pipe_mutex_unlock(rb_pipe->call_mutex);

   if (result)
      return rbug_surface_create(rb_pipe, rb_resource, result);
   return NULL;
}

/**
 * Copy image data from a VdpOutputSurface to application memory in the
 * surface's native format.
 */
VdpStatus
vlVdpOutputSurfaceGetBitsNative(VdpOutputSurface surface,
                                VdpRect const *source_rect,
                                void *const *destination_data,
                                uint32_t const *destination_pitches)
{
   vlVdpOutputSurface *vlsurface;
   struct pipe_context *pipe;
   struct pipe_resource *res;
   struct pipe_box box;
   struct pipe_transfer *transfer;
   uint8_t *map;

   vlsurface = vlGetDataHTAB(surface);
   if (!vlsurface)
      return VDP_STATUS_INVALID_HANDLE;

   pipe = vlsurface->device->context;
   if (!pipe)
      return VDP_STATUS_INVALID_HANDLE;

   pipe_mutex_lock(vlsurface->device->mutex);
   vlVdpResolveDelayedRendering(vlsurface->device, NULL, NULL);

   res = vlsurface->sampler_view->texture;
   box = RectToPipeBox(source_rect, res);
   map = pipe->transfer_map(pipe, res, 0, PIPE_TRANSFER_READ, &box, &transfer);
   if (!map) {
      pipe_mutex_unlock(vlsurface->device->mutex);
      return VDP_STATUS_RESOURCES;
   }

   util_copy_rect(*destination_data, res->format, *destination_pitches, 0, 0,
                  box.width, box.height, map, transfer->stride, 0, 0);

   pipe_transfer_unmap(pipe, transfer);
   pipe_mutex_unlock(vlsurface->device->mutex);

   return VDP_STATUS_OK;
}

VAStatus
vlVaDestroyImage(VADriverContextP ctx, VAImageID image)
{
   vlVaDriver *drv;
   VAImage  *vaimage;

   if (!ctx)
      return VA_STATUS_ERROR_INVALID_CONTEXT;

   drv = VL_VA_DRIVER(ctx);
   pipe_mutex_lock(drv->mutex);
   vaimage = handle_table_get(drv->htab, image);
   if (!vaimage) {
      pipe_mutex_unlock(drv->mutex);
      return VA_STATUS_ERROR_INVALID_IMAGE;
   }

   handle_table_remove(VL_VA_DRIVER(ctx)->htab, image);
   pipe_mutex_unlock(drv->mutex);
   FREE(vaimage);
   return vlVaDestroyBuffer(ctx, vaimage->buf);
}

static void
pb_cache_manager_flush(struct pb_manager *_mgr)
{
   struct pb_cache_manager *mgr = pb_cache_manager(_mgr);
   struct list_head *curr, *next;
   struct pb_cache_buffer *buf;

   pipe_mutex_lock(mgr->mutex);
   curr = mgr->delayed.next;
   next = curr->next;
   while(curr != &mgr->delayed) {
      buf = LIST_ENTRY(struct pb_cache_buffer, curr, head);
      _pb_cache_buffer_destroy(buf);
      curr = next; 
      next = curr->next;
   }
   pipe_mutex_unlock(mgr->mutex);
   
   assert(mgr->provider->flush);
   if(mgr->provider->flush)
      mgr->provider->flush(mgr->provider);
}

static void virgl_drm_resource_reference(struct virgl_drm_winsys *qdws,
                                       struct virgl_hw_res **dres,
                                       struct virgl_hw_res *sres)
{
   struct virgl_hw_res *old = *dres;
   if (pipe_reference(&(*dres)->reference, &sres->reference)) {

      if (!can_cache_resource(old)) {
         virgl_hw_res_destroy(qdws, old);
      } else {
         pipe_mutex_lock(qdws->mutex);
         virgl_cache_list_check_free(qdws);

         old->start = os_time_get();
         old->end = old->start + qdws->usecs;
         LIST_ADDTAIL(&old->head, &qdws->delayed);
         qdws->num_delayed++;
         pipe_mutex_unlock(qdws->mutex);
      }
   }
   *dres = sres;
}

static void
rbug_clear_render_target(struct pipe_context *_pipe,
                         struct pipe_surface *_dst,
                         const union pipe_color_union *color,
                         unsigned dstx, unsigned dsty,
                         unsigned width, unsigned height)
{
   struct rbug_context *rb_pipe = rbug_context(_pipe);
   struct rbug_surface *rb_surface_dst = rbug_surface(_dst);
   struct pipe_context *pipe = rb_pipe->pipe;
   struct pipe_surface *dst = rb_surface_dst->surface;

   pipe_mutex_lock(rb_pipe->call_mutex);
   pipe->clear_render_target(pipe,
                             dst,
                             color,
                             dstx,
                             dsty,
                             width,
                             height);
   pipe_mutex_unlock(rb_pipe->call_mutex);
}

static void *
pb_debug_buffer_map(struct pb_buffer *_buf, 
                    unsigned flags)
{
   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);
   void *map;
   
   pb_debug_buffer_check(buf);

   map = pb_map(buf->buffer, flags);
   if(!map)
      return NULL;
   
   if(map) {
      pipe_mutex_lock(buf->mutex);
      ++buf->map_count;
      debug_backtrace_capture(buf->map_backtrace, 1, PB_DEBUG_MAP_BACKTRACE);
      pipe_mutex_unlock(buf->mutex);
   }
   
   return (uint8_t *)map + buf->underflow_size;
}

static boolean
stw_st_framebuffer_flush_front(struct st_context_iface *stctx,
                               struct st_framebuffer_iface *stfb,
                               enum st_attachment_type statt)
{
   struct stw_st_framebuffer *stwfb = stw_st_framebuffer(stfb);
   boolean ret;
   HDC hDC;

   pipe_mutex_lock(stwfb->fb->mutex);

   /* We must not cache HDCs anywhere, as they can be invalidated by the
    * application, or screen resolution changes. */

   hDC = GetDC(stwfb->fb->hWnd);

   ret = stw_st_framebuffer_present_locked(hDC, &stwfb->base, statt);

   ReleaseDC(stwfb->fb->hWnd, hDC);

   return ret;
}

VAStatus
vlVaMapBuffer(VADriverContextP ctx, VABufferID buf_id, void **pbuff)
{
   vlVaDriver *drv;
   vlVaBuffer *buf;

   if (!ctx)
      return VA_STATUS_ERROR_INVALID_CONTEXT;

   drv = VL_VA_DRIVER(ctx);
   if (!drv)
      return VA_STATUS_ERROR_INVALID_CONTEXT;

   if (!pbuff)
      return VA_STATUS_ERROR_INVALID_PARAMETER;

   pipe_mutex_lock(drv->mutex);
   buf = handle_table_get(drv->htab, buf_id);
   if (!buf || buf->export_refcount > 0) {
      pipe_mutex_unlock(drv->mutex);
      return VA_STATUS_ERROR_INVALID_BUFFER;
   }

   if (buf->derived_surface.resource) {
      *pbuff = pipe_buffer_map(drv->pipe, buf->derived_surface.resource,
                               PIPE_TRANSFER_WRITE,
                               &buf->derived_surface.transfer);
      pipe_mutex_unlock(drv->mutex);

      if (!buf->derived_surface.transfer || !*pbuff)
         return VA_STATUS_ERROR_INVALID_BUFFER;

   } else {
      pipe_mutex_unlock(drv->mutex);
      *pbuff = buf->data;
   }

   return VA_STATUS_SUCCESS;
}

void *
debug_malloc(const char *file, unsigned line, const char *function,
             size_t size) 
{
   struct debug_memory_header *hdr;
   struct debug_memory_footer *ftr;
   
   hdr = os_malloc(sizeof(*hdr) + size + sizeof(*ftr));
   if(!hdr) {
      debug_printf("%s:%u:%s: out of memory when trying to allocate %lu bytes\n",
                   file, line, function,
                   (long unsigned)size);
      return NULL;
   }
 
   hdr->no = last_no++;
   hdr->file = file;
   hdr->line = line;
   hdr->function = function;
   hdr->size = size;
   hdr->magic = DEBUG_MEMORY_MAGIC;
   hdr->tag = 0;
#if DEBUG_FREED_MEMORY
   hdr->freed = FALSE;
#endif

#if DEBUG_MEMORY_STACK
   debug_backtrace_capture(hdr->backtrace, 0, DEBUG_MEMORY_STACK);
#endif

   ftr = footer_from_header(hdr);
   ftr->magic = DEBUG_MEMORY_MAGIC;
   
   pipe_mutex_lock(list_mutex);
   LIST_ADDTAIL(&hdr->head, &list);
   pipe_mutex_unlock(list_mutex);
   
   return data_from_header(hdr);
}

/**
 * Empty the cache. Useful when there is not enough memory.
 */
void
pb_cache_release_all_buffers(struct pb_cache *mgr)
{
   struct list_head *curr, *next;
   struct pb_cache_entry *buf;
   unsigned i;

   pipe_mutex_lock(mgr->mutex);
   for (i = 0; i < ARRAY_SIZE(mgr->buckets); i++) {
      struct list_head *cache = &mgr->buckets[i];

      curr = cache->next;
      next = curr->next;
      while (curr != cache) {
         buf = LIST_ENTRY(struct pb_cache_entry, curr, head);
         destroy_buffer_locked(buf);
         curr = next;
         next = curr->next;
      }
   }
   pipe_mutex_unlock(mgr->mutex);
}

void
intel_winsys_decode_bo(struct intel_winsys *winsys,
                       struct intel_bo *bo, int used)
{
   void *ptr;

   ptr = intel_bo_map(bo, false);
   if (!ptr) {
      debug_printf("failed to map buffer for decoding\n");
      return;
   }

   pipe_mutex_lock(winsys->mutex);

   if (!winsys->decode) {
      winsys->decode = drm_intel_decode_context_alloc(winsys->info.devid);
      if (!winsys->decode) {
         pipe_mutex_unlock(winsys->mutex);
         intel_bo_unmap(bo);
         return;
      }

      /* debug_printf()/debug_error() uses stderr by default */
      drm_intel_decode_set_output_file(winsys->decode, stderr);
   }

   /* in dwords */
   used /= 4;

   drm_intel_decode_set_batch_pointer(winsys->decode,
         ptr, gem_bo(bo)->offset64, used);

   drm_intel_decode(winsys->decode);

   pipe_mutex_unlock(winsys->mutex);

   intel_bo_unmap(bo);
}

static void
rbug_set_framebuffer_state(struct pipe_context *_pipe,
                           const struct pipe_framebuffer_state *_state)
{
   struct rbug_context *rb_pipe = rbug_context(_pipe);
   struct pipe_context *pipe = rb_pipe->pipe;
   struct pipe_framebuffer_state unwrapped_state;
   struct pipe_framebuffer_state *state = NULL;
   unsigned i;

   /* must protect curr status */
   pipe_mutex_lock(rb_pipe->call_mutex);

   rb_pipe->curr.nr_cbufs = 0;
   memset(rb_pipe->curr.cbufs, 0, sizeof(rb_pipe->curr.cbufs));
   rb_pipe->curr.zsbuf = NULL;

   /* unwrap the input state */
   if (_state) {
      memcpy(&unwrapped_state, _state, sizeof(unwrapped_state));

      rb_pipe->curr.nr_cbufs = _state->nr_cbufs;
      for(i = 0; i < _state->nr_cbufs; i++) {
         unwrapped_state.cbufs[i] = rbug_surface_unwrap(_state->cbufs[i]);
         if (_state->cbufs[i])
            rb_pipe->curr.cbufs[i] = rbug_resource(_state->cbufs[i]->texture);
      }
      unwrapped_state.zsbuf = rbug_surface_unwrap(_state->zsbuf);
      if (_state->zsbuf)
         rb_pipe->curr.zsbuf = rbug_resource(_state->zsbuf->texture);
      state = &unwrapped_state;
   }

   pipe->set_framebuffer_state(pipe,
                               state);

   pipe_mutex_unlock(rb_pipe->call_mutex);
}

VAStatus
vlVaQueryConfigAttributes(VADriverContextP ctx, VAConfigID config_id, VAProfile *profile,
                          VAEntrypoint *entrypoint, VAConfigAttrib *attrib_list, int *num_attribs)
{
   vlVaDriver *drv;
   vlVaConfig *config;

   if (!ctx)
      return VA_STATUS_ERROR_INVALID_CONTEXT;

   drv = VL_VA_DRIVER(ctx);

   if (!drv)
      return VA_STATUS_ERROR_INVALID_CONTEXT;

   pipe_mutex_lock(drv->mutex);
   config = handle_table_get(drv->htab, config_id);
   pipe_mutex_unlock(drv->mutex);

   if (!config)
      return VA_STATUS_ERROR_INVALID_CONFIG;

   *profile = PipeToProfile(config->profile);

   if (config->profile == PIPE_VIDEO_PROFILE_UNKNOWN) {
      *entrypoint = VAEntrypointVideoProc;
      *num_attribs = 0;
      return VA_STATUS_SUCCESS;
   }

   *entrypoint = config->entrypoint;

   *num_attribs = 1;
   attrib_list[0].type = VAConfigAttribRTFormat;
   attrib_list[0].value = config->rt_format;

   return VA_STATUS_SUCCESS;
}

/** Rasterize all scene's bins */
static void
lp_setup_rasterize_scene( struct lp_setup_context *setup )
{
   struct lp_scene *scene = setup->scene;
   struct llvmpipe_screen *screen = llvmpipe_screen(scene->pipe->screen);

   scene->num_active_queries = setup->active_binned_queries;
   memcpy(scene->active_queries, setup->active_queries,
          scene->num_active_queries * sizeof(scene->active_queries[0]));

   lp_scene_end_binning(scene);

   lp_fence_reference(&setup->last_fence, scene->fence);

   if (setup->last_fence)
      setup->last_fence->issued = TRUE;

   pipe_mutex_lock(screen->rast_mutex);

   /* FIXME: We enqueue the scene then wait on the rasterizer to finish.
    * This means we never actually run any vertex stuff in parallel to
    * rasterization (not in the same context at least) which is what the
    * multiple scenes per setup is about - when we get a new empty scene
    * any old one is already empty again because we waited here for
    * raster tasks to be finished. Ideally, we shouldn't need to wait here
    * and rely on fences elsewhere when waiting is necessary.
    * Certainly, lp_scene_end_rasterization() would need to be deferred too
    * and there's probably other bits why this doesn't actually work.
    */
   lp_rast_queue_scene(screen->rast, scene);
   lp_rast_finish(screen->rast);
   pipe_mutex_unlock(screen->rast_mutex);

   lp_scene_end_rasterization(setup->scene);
   lp_setup_reset( setup );

   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
}

static void
svga_buffer_transfer_unmap( struct pipe_context *pipe,
                            struct pipe_transfer *transfer )
{
   struct svga_screen *ss = svga_screen(pipe->screen);
   struct svga_context *svga = svga_context(pipe);
   struct svga_buffer *sbuf = svga_buffer(transfer->resource);

   pipe_mutex_lock(ss->swc_mutex);

   assert(sbuf->map.count);
   if (sbuf->map.count) {
      --sbuf->map.count;
   }

   if (svga_buffer_has_hw_storage(sbuf)) {
      svga_buffer_hw_storage_unmap(svga, sbuf);
   }

   if (transfer->usage & PIPE_TRANSFER_WRITE) {
      if (!(transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT)) {
         /*
          * Mapped range not flushed explicitly, so flush the whole buffer,
          * and tell the host to discard the contents when processing the DMA
          * command.
          */

         SVGA_DBG(DEBUG_DMA, "flushing the whole buffer\n");

         sbuf->dma.flags.discard = TRUE;

         svga_buffer_add_range(sbuf, 0, sbuf->b.b.width0);
      }
   }

   pipe_mutex_unlock(ss->swc_mutex);
   FREE(transfer);
}

struct pipe_screen *
virgl_drm_screen_create(int fd)
{
   struct pipe_screen *pscreen = NULL;

   pipe_mutex_lock(virgl_screen_mutex);
   if (!fd_tab) {
      fd_tab = util_hash_table_create(hash_fd, compare_fd);
      if (!fd_tab)
         goto unlock;
   }

   pscreen = util_hash_table_get(fd_tab, intptr_to_pointer(fd));
   if (pscreen) {
      virgl_screen(pscreen)->refcnt++;
   } else {
      struct virgl_winsys *vws;
      int dup_fd = dup(fd);

      vws = virgl_drm_winsys_create(dup_fd);

      pscreen = virgl_create_screen(vws);
      if (pscreen) {
         util_hash_table_set(fd_tab, intptr_to_pointer(dup_fd), pscreen);

         /* Bit of a hack, to avoid circular linkage dependency,
          * ie. pipe driver having to call in to winsys, we
          * override the pipe drivers screen->destroy():
          */
         virgl_screen(pscreen)->winsys_priv = pscreen->destroy;
         pscreen->destroy = virgl_drm_screen_destroy;
      }
   }

unlock:
   pipe_mutex_unlock(virgl_screen_mutex);
   return pscreen;
}

/**
 * Return a small integer serial number for the given pointer.
 */
static boolean
debug_serial(void *p, unsigned *pserial)
{
   unsigned serial;
   boolean found = TRUE;
#ifdef PIPE_SUBSYSTEM_WINDOWS_USER
   static boolean first = TRUE;

   if (first) {
      pipe_mutex_init(serials_mutex);
      first = FALSE;
   }
#endif

   pipe_mutex_lock(serials_mutex);
   if (!serials_hash)
      serials_hash = util_hash_table_create(hash_ptr, compare_ptr);

   serial = (unsigned) (uintptr_t) util_hash_table_get(serials_hash, p);
   if (!serial) {
      /* time to stop logging... (you'll have a 100 GB logfile at least at
       * this point)  TODO: avoid this
       */
      serial = ++serials_last;
      if (!serial) {
         debug_error("More than 2^32 objects detected, aborting.\n");
         os_abort();
      }

      util_hash_table_set(serials_hash, p, (void *) (uintptr_t) serial);
      found = FALSE;
   }
   pipe_mutex_unlock(serials_mutex);

   *pserial = serial;

   return found;
}

/**
 * Destroy a VdpVideoMixer.
 */
VdpStatus
vlVdpVideoMixerDestroy(VdpVideoMixer mixer)
{
   vlVdpVideoMixer *vmixer;

   vmixer = vlGetDataHTAB(mixer);
   if (!vmixer)
      return VDP_STATUS_INVALID_HANDLE;

   pipe_mutex_lock(vmixer->device->mutex);

   vlVdpResolveDelayedRendering(vmixer->device, NULL, NULL);

   vlRemoveDataHTAB(mixer);

   vl_compositor_cleanup_state(&vmixer->cstate);

   if (vmixer->deint.filter) {
      vl_deint_filter_cleanup(vmixer->deint.filter);
      FREE(vmixer->deint.filter);
   }

   if (vmixer->noise_reduction.filter) {
      vl_median_filter_cleanup(vmixer->noise_reduction.filter);
      FREE(vmixer->noise_reduction.filter);
   }

   if (vmixer->sharpness.filter) {
      vl_matrix_filter_cleanup(vmixer->sharpness.filter);
      FREE(vmixer->sharpness.filter);
   }
   pipe_mutex_unlock(vmixer->device->mutex);
   DeviceReference(&vmixer->device, NULL);

   FREE(vmixer);

   return VDP_STATUS_OK;
}

/** Rasterize all scene's bins */
static void
lp_setup_rasterize_scene( struct lp_setup_context *setup )
{
   struct lp_scene *scene = setup->scene;
   struct llvmpipe_screen *screen = llvmpipe_screen(scene->pipe->screen);

   lp_scene_end_binning(scene);

   lp_fence_reference(&setup->last_fence, scene->fence);

   if (setup->last_fence)
      setup->last_fence->issued = TRUE;

   pipe_mutex_lock(screen->rast_mutex);
   lp_rast_queue_scene(screen->rast, scene);
   lp_rast_finish(screen->rast);
   pipe_mutex_unlock(screen->rast_mutex);

   lp_scene_end_rasterization(setup->scene);
   lp_setup_reset( setup );

   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
}