[Mesa-dev] [PATCH] i965: Fix up a failed CPU/WC mmaping with a GTT mapping

[Chris Wilson]

Not all objects will be mappable for direct access by the CPU (either
using WC/CPU or WC paths), for example, a dmabuf wrapping an object on a
foreign device or an object wrapping access to stolen memory. Since
either the physical pages are not known or even do not exist, we need to
use the mediated, indirect access via the GTT. (If one day, the kernel
does suddenly start providing mediated access via a regular WB/WC
mmapping, we no longer need the fallback.)

Cc: Kenneth Graunke <kenneth at whitecape.org>
Cc: Matt Turner <mattst88 at gmail.com>
---
Note that until we actually have WC mmaps, we can legitmately hit the
fallback case for !llc writing into a linear buffer. So please just
squash this into the WC mmap patch.
---
 src/mesa/drivers/dri/i965/brw_bufmgr.c | 34 +++++++++++++++++++++++++++++++---
 1 file changed, 31 insertions(+), 3 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_bufmgr.c b/src/mesa/drivers/dri/i965/brw_bufmgr.c
index 98a75dd4a6..055ac59a4a 100644
--- a/src/mesa/drivers/dri/i965/brw_bufmgr.c
+++ b/src/mesa/drivers/dri/i965/brw_bufmgr.c
@@ -740,6 +740,13 @@ brw_bo_map_cpu(struct brw_context *brw, struct brw_bo *bo, unsigned flags)
 }
 
 static void *
+brw_bo_map_wc(struct brw_context *brw, struct brw_bo *bo, unsigned flags)
+{
+   /* Intentionally left blank */
+   return NULL;
+}
+
+static void *
 brw_bo_map_gtt(struct brw_context *brw, struct brw_bo *bo, unsigned flags)
 {
    struct brw_bufmgr *bufmgr = bo->bufmgr;
@@ -823,12 +830,33 @@ can_map_cpu(struct brw_bo *bo, unsigned flags)
 void *
 brw_bo_map(struct brw_context *brw, struct brw_bo *bo, unsigned flags)
 {
+
    if (bo->tiling_mode != I915_TILING_NONE && !(flags & MAP_RAW))
       return brw_bo_map_gtt(brw, bo, flags);
-   else if (can_map_cpu(bo, flags))
-      return brw_bo_map_cpu(brw, bo, flags);
+
+   void *map;
+
+   if (can_map_cpu(bo, flags))
+      map = brw_bo_map_cpu(brw, bo, flags);
    else
-      return brw_bo_map_gtt(brw, bo, flags);
+      map = brw_bo_map_wc(brw, bo, flags);
+
+   /* Allow the attempt to fail by falling back to the GTT where necessary.
+    *
+    * Not every buffer can be mmaped directly using the CPU (or WC), for
+    * example buffers that wrap stolen memory or are imported from other
+    * devices. For those, we have little choice but to use a GTT mmapping.
+    * However, if we use a slow GTT mmapping for reads where we expected fast
+    * access, that order of magnitude difference in throughput will be clearly
+    * expressed by angry users.
+    */
+   if (!map) {
+      perf_debug("Fallback GTT mapping for %s with access flags %x\n",
+                 bo->name, flags);
+      map = brw_bo_map_gtt(brw, bo, flags);
+   }
+
+   return map;
 }
 
 int
-- 
2.13.2