anv: Use separate MOCS settings for external BOs

On Broadwell and above, we have to use different MOCS settings to allow
the kernel to take over and disable caching when needed for external
buffers.  On Broadwell, this is especially important because the kernel
can't disable eLLC so we have to do it in userspace.  We very badly
don't want to do that on everything so we need separate MOCS for
external and internal BOs.

In order to do this, we add an anv-specific BO flag for "external" and
use that to distinguish between buffers which may be shared with other
processes and/or display and those which are entirely internal.  That,
together with an anv_mocs_for_bo helper lets us choose the right MOCS
settings for each BO use.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99507
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
(cherry picked from commit 7a89a0d9edae638e68e4b4ee8e0cbb34baa9c080)
[Juan A. Suarez: resolve trivial conflicts]
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>

Conflicts:
	src/intel/vulkan/anv_image.c
	src/intel/vulkan/genX_gpu_memcpy.c

12 files changed, 80 insertions, 38 deletions

diff --git a/src/intel/vulkan/anv_allocator.c b/src/intel/vulkan/anv_allocator.c
index ab01d46cbeb..f62d48ae3fe 100644
--- a/src/intel/vulkan/anv_allocator.c
+++ b/src/intel/vulkan/anv_allocator.c
@@ -1253,7 +1253,8 @@ anv_bo_cache_lookup(struct anv_bo_cache *cache, uint32_t gem_handle)
    (EXEC_OBJECT_WRITE | \
     EXEC_OBJECT_ASYNC | \
     EXEC_OBJECT_SUPPORTS_48B_ADDRESS | \
-    EXEC_OBJECT_PINNED)
+    EXEC_OBJECT_PINNED | \
+    ANV_BO_EXTERNAL)
 
 VkResult
 anv_bo_cache_alloc(struct anv_device *device,
@@ -1311,6 +1312,7 @@ anv_bo_cache_import(struct anv_device *device,
                     struct anv_bo **bo_out)
 {
    assert(bo_flags == (bo_flags & ANV_BO_CACHE_SUPPORTED_FLAGS));
+   assert(bo_flags & ANV_BO_EXTERNAL);
 
    pthread_mutex_lock(&cache->mutex);
 
@@ -1327,7 +1329,7 @@ anv_bo_cache_import(struct anv_device *device,
        * client has imported a BO twice in different ways and they get what
        * they have coming.
        */
-      uint64_t new_flags = 0;
+      uint64_t new_flags = ANV_BO_EXTERNAL;
       new_flags |= (bo->bo.flags | bo_flags) & EXEC_OBJECT_WRITE;
       new_flags |= (bo->bo.flags & bo_flags) & EXEC_OBJECT_ASYNC;
       new_flags |= (bo->bo.flags & bo_flags) & EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
@@ -1411,6 +1413,12 @@ anv_bo_cache_export(struct anv_device *device,
    assert(anv_bo_cache_lookup(cache, bo_in->gem_handle) == bo_in);
    struct anv_cached_bo *bo = (struct anv_cached_bo *)bo_in;
 
+   /* This BO must have been flagged external in order for us to be able
+    * to export it.  This is done based on external options passed into
+    * anv_AllocateMemory.
+    */
+   assert(bo->bo.flags & ANV_BO_EXTERNAL);
+
    int fd = anv_gem_handle_to_fd(device, bo->bo.gem_handle);
    if (fd < 0)
       return vk_error(VK_ERROR_TOO_MANY_OBJECTS);
diff --git a/src/intel/vulkan/anv_batch_chain.c b/src/intel/vulkan/anv_batch_chain.c
index 0f7c8325ea4..3e13553ac18 100644
--- a/src/intel/vulkan/anv_batch_chain.c
+++ b/src/intel/vulkan/anv_batch_chain.c
@@ -1088,7 +1088,7 @@ anv_execbuf_add_bo(struct anv_execbuf *exec,
       obj->relocs_ptr = 0;
       obj->alignment = 0;
       obj->offset = bo->offset;
-      obj->flags = bo->flags | extra_flags;
+      obj->flags = (bo->flags & ~ANV_BO_FLAG_MASK) | extra_flags;
       obj->rsvd1 = 0;
       obj->rsvd2 = 0;
    }
diff --git a/src/intel/vulkan/anv_blorp.c b/src/intel/vulkan/anv_blorp.c
index 04bca4d261f..1cc632bdb0d 100644
--- a/src/intel/vulkan/anv_blorp.c
+++ b/src/intel/vulkan/anv_blorp.c
@@ -155,7 +155,7 @@ get_blorp_surf_for_anv_buffer(struct anv_device *device,
       .addr = {
          .buffer = buffer->address.bo,
          .offset = buffer->address.offset + offset,
-         .mocs = device->default_mocs,
+         .mocs = anv_mocs_for_bo(device, buffer->address.bo),
       },
    };
 
@@ -208,7 +208,7 @@ get_blorp_surf_for_anv_image(const struct anv_device *device,
       .addr = {
          .buffer = image->planes[plane].address.bo,
          .offset = image->planes[plane].address.offset + surface->offset,
-         .mocs = device->default_mocs,
+         .mocs = anv_mocs_for_bo(device, image->planes[plane].address.bo),
       },
    };
 
@@ -218,7 +218,7 @@ get_blorp_surf_for_anv_image(const struct anv_device *device,
       blorp_surf->aux_addr = (struct blorp_address) {
          .buffer = image->planes[plane].address.bo,
          .offset = image->planes[plane].address.offset + aux_surface->offset,
-         .mocs = device->default_mocs,
+         .mocs = anv_mocs_for_bo(device, image->planes[plane].address.bo),
       };
       blorp_surf->aux_usage = aux_usage;
 
@@ -668,12 +668,12 @@ void anv_CmdCopyBuffer(
       struct blorp_address src = {
          .buffer = src_buffer->address.bo,
          .offset = src_buffer->address.offset + pRegions[r].srcOffset,
-         .mocs = cmd_buffer->device->default_mocs,
+         .mocs = anv_mocs_for_bo(cmd_buffer->device, src_buffer->address.bo),
       };
       struct blorp_address dst = {
          .buffer = dst_buffer->address.bo,
          .offset = dst_buffer->address.offset + pRegions[r].dstOffset,
-         .mocs = cmd_buffer->device->default_mocs,
+         .mocs = anv_mocs_for_bo(cmd_buffer->device, dst_buffer->address.bo),
       };
 
       blorp_buffer_copy(&batch, src, dst, pRegions[r].size);
@@ -726,7 +726,7 @@ void anv_CmdUpdateBuffer(
       struct blorp_address dst = {
          .buffer = dst_buffer->address.bo,
          .offset = dst_buffer->address.offset + dstOffset,
-         .mocs = cmd_buffer->device->default_mocs,
+         .mocs = anv_mocs_for_bo(cmd_buffer->device, dst_buffer->address.bo),
       };
 
       blorp_buffer_copy(&batch, src, dst, copy_size);
@@ -1436,7 +1436,8 @@ anv_image_copy_to_shadow(struct anv_cmd_buffer *cmd_buffer,
          .buffer = image->planes[0].address.bo,
          .offset = image->planes[0].address.offset +
                    image->planes[0].shadow_surface.offset,
-         .mocs = cmd_buffer->device->default_mocs,
+         .mocs = anv_mocs_for_bo(cmd_buffer->device,
+                                 image->planes[0].address.bo),
       },
    };
 
diff --git a/src/intel/vulkan/anv_device.c b/src/intel/vulkan/anv_device.c
index 113aba70a32..50d16b8204b 100644
--- a/src/intel/vulkan/anv_device.c
+++ b/src/intel/vulkan/anv_device.c
@@ -2216,8 +2216,8 @@ VkResult anv_AllocateMemory(
              fd_info->handleType ==
                VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
 
-      result = anv_bo_cache_import(device, &device->bo_cache,
-                                   fd_info->fd, bo_flags, &mem->bo);
+      result = anv_bo_cache_import(device, &device->bo_cache, fd_info->fd,
+                                   bo_flags | ANV_BO_EXTERNAL, &mem->bo);
       if (result != VK_SUCCESS)
          goto fail;
 
@@ -2254,6 +2254,11 @@ VkResult anv_AllocateMemory(
        */
       close(fd_info->fd);
    } else {
+      const VkExportMemoryAllocateInfoKHR *fd_info =
+         vk_find_struct_const(pAllocateInfo->pNext, EXPORT_MEMORY_ALLOCATE_INFO_KHR);
+      if (fd_info && fd_info->handleTypes)
+         bo_flags |= ANV_BO_EXTERNAL;
+
       result = anv_bo_cache_alloc(device, &device->bo_cache,
                                   pAllocateInfo->allocationSize, bo_flags,
                                   &mem->bo);
diff --git a/src/intel/vulkan/anv_image.c b/src/intel/vulkan/anv_image.c
index 36d4ac13c75..28d70967b05 100644
--- a/src/intel/vulkan/anv_image.c
+++ b/src/intel/vulkan/anv_image.c
@@ -1095,7 +1095,7 @@ anv_image_fill_surface_state(struct anv_device *device,
                             .size = surface->isl.size,
                             .format = ISL_FORMAT_RAW,
                             .stride = 1,
-                            .mocs = device->default_mocs);
+                            .mocs = anv_mocs_for_bo(device, address.bo));
       state_inout->address = address,
       state_inout->aux_address = ANV_NULL_ADDRESS;
       state_inout->clear_address = ANV_NULL_ADDRESS;
@@ -1196,7 +1196,8 @@ anv_image_fill_surface_state(struct anv_device *device,
                           .aux_address = anv_address_physical(aux_address),
                           .clear_address = anv_address_physical(clear_address),
                           .use_clear_address = !anv_address_is_null(clear_address),
-                          .mocs = device->default_mocs,
+                          .mocs = anv_mocs_for_bo(device,
+                                                  state_inout->address.bo),
                           .x_offset_sa = tile_x_sa,
                           .y_offset_sa = tile_y_sa);
 
diff --git a/src/intel/vulkan/anv_intel.c b/src/intel/vulkan/anv_intel.c
index 06db5787a9c..ed1bc096c66 100644
--- a/src/intel/vulkan/anv_intel.c
+++ b/src/intel/vulkan/anv_intel.c
@@ -73,7 +73,7 @@ VkResult anv_CreateDmaBufImageINTEL(
 
    image = anv_image_from_handle(image_h);
 
-   uint64_t bo_flags = 0;
+   uint64_t bo_flags = ANV_BO_EXTERNAL;
    if (device->instance->physicalDevice.supports_48bit_addresses)
       bo_flags |= EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
    if (device->instance->physicalDevice.use_softpin)
diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index 4e50ab27898..1e7c87c70d0 100644
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -555,6 +555,10 @@ anv_multialloc_alloc2(struct anv_multialloc *ma,
    return anv_multialloc_alloc(ma, alloc ? alloc : parent_alloc, scope);
 }
 
+/* Extra ANV-defined BO flags which won't be passed to the kernel */
+#define ANV_BO_EXTERNAL    (1ull << 31)
+#define ANV_BO_FLAG_MASK   (1ull << 31)
+
 struct anv_bo {
    uint32_t gem_handle;
 
@@ -1012,6 +1016,7 @@ struct anv_device {
     struct anv_scratch_pool                     scratch_pool;
 
     uint32_t                                    default_mocs;
+    uint32_t                                    external_mocs;
 
     pthread_mutex_t                             mutex;
     pthread_cond_t                              queue_submit;
@@ -1041,6 +1046,15 @@ anv_binding_table_pool_free(struct anv_device *device, struct anv_state state) {
    anv_state_pool_free(anv_binding_table_pool(device), state);
 }
 
+static inline uint32_t
+anv_mocs_for_bo(const struct anv_device *device, const struct anv_bo *bo)
+{
+   if (bo->flags & ANV_BO_EXTERNAL)
+      return device->external_mocs;
+   else
+      return device->default_mocs;
+}
+
 static void inline
 anv_state_flush(struct anv_device *device, struct anv_state state)
 {
@@ -1322,6 +1336,12 @@ _anv_combine_address(struct anv_batch *batch, void *location,
       .AgeforQUADLRU = 0                                       \
    }
 
+#define GEN8_EXTERNAL_MOCS (struct GEN8_MEMORY_OBJECT_CONTROL_STATE) {     \
+      .MemoryTypeLLCeLLCCacheabilityControl = UCwithFenceifcoherentcycle,  \
+      .TargetCache = L3DefertoPATforLLCeLLCselection,                      \
+      .AgeforQUADLRU = 0                                                   \
+   }
+
 /* Skylake: MOCS is now an index into an array of 62 different caching
  * configurations programmed by the kernel.
  */
@@ -1331,9 +1351,9 @@ _anv_combine_address(struct anv_batch *batch, void *location,
       .IndextoMOCSTables                           = 2         \
    }
 
-#define GEN9_MOCS_PTE {                                 \
-      /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */       \
-      .IndextoMOCSTables                           = 1  \
+#define GEN9_EXTERNAL_MOCS (struct GEN9_MEMORY_OBJECT_CONTROL_STATE) {  \
+      /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */                       \
+      .IndextoMOCSTables                           = 1                  \
    }
 
 /* Cannonlake MOCS defines are duplicates of Skylake MOCS defines. */
@@ -1342,9 +1362,9 @@ _anv_combine_address(struct anv_batch *batch, void *location,
       .IndextoMOCSTables                           = 2         \
    }
 
-#define GEN10_MOCS_PTE {                                 \
-      /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */       \
-      .IndextoMOCSTables                           = 1  \
+#define GEN10_EXTERNAL_MOCS (struct GEN10_MEMORY_OBJECT_CONTROL_STATE) {   \
+      /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */                          \
+      .IndextoMOCSTables                           = 1                     \
    }
 
 /* Ice Lake MOCS defines are duplicates of Skylake MOCS defines. */
@@ -1353,9 +1373,9 @@ _anv_combine_address(struct anv_batch *batch, void *location,
       .IndextoMOCSTables                           = 2         \
    }
 
-#define GEN11_MOCS_PTE {                                 \
-      /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */       \
-      .IndextoMOCSTables                           = 1  \
+#define GEN11_EXTERNAL_MOCS (struct GEN11_MEMORY_OBJECT_CONTROL_STATE) {   \
+      /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */                          \
+      .IndextoMOCSTables                           = 1                     \
    }
 
 struct anv_device_memory {
diff --git a/src/intel/vulkan/gen7_cmd_buffer.c b/src/intel/vulkan/gen7_cmd_buffer.c
index cf1f8ee2829..da51cb9781c 100644
--- a/src/intel/vulkan/gen7_cmd_buffer.c
+++ b/src/intel/vulkan/gen7_cmd_buffer.c
@@ -246,7 +246,8 @@ genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer)
          ib.CutIndexEnable             = pipeline->primitive_restart;
 #endif
          ib.IndexFormat                = cmd_buffer->state.gfx.gen7.index_type;
-         ib.MemoryObjectControlState   = GENX(MOCS);
+         ib.IndexBufferMOCS            = anv_mocs_for_bo(cmd_buffer->device,
+                                                         buffer->address.bo);
 
          ib.BufferStartingAddress      = anv_address_add(buffer->address,
                                                          offset);
diff --git a/src/intel/vulkan/gen8_cmd_buffer.c b/src/intel/vulkan/gen8_cmd_buffer.c
index ca2baf84a19..752d04f3013 100644
--- a/src/intel/vulkan/gen8_cmd_buffer.c
+++ b/src/intel/vulkan/gen8_cmd_buffer.c
@@ -565,7 +565,8 @@ void genX(CmdBindIndexBuffer)(
 
    anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), ib) {
       ib.IndexFormat                = vk_to_gen_index_type[indexType];
-      ib.MemoryObjectControlState   = GENX(MOCS);
+      ib.IndexBufferMOCS            = anv_mocs_for_bo(cmd_buffer->device,
+                                                      buffer->address.bo);
       ib.BufferStartingAddress      = anv_address_add(buffer->address, offset);
       ib.BufferSize                 = buffer->size - offset;
    }
diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c
index b984ba62bc8..7cef4c55cf1 100644
--- a/src/intel/vulkan/genX_cmd_buffer.c
+++ b/src/intel/vulkan/genX_cmd_buffer.c
@@ -2516,12 +2516,11 @@ genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer)
          struct GENX(VERTEX_BUFFER_STATE) state = {
             .VertexBufferIndex = vb,
 
-#if GEN_GEN >= 8
-            .MemoryObjectControlState = GENX(MOCS),
-#else
+            .VertexBufferMOCS = anv_mocs_for_bo(cmd_buffer->device,
+                                                buffer->address.bo),
+#if GEN_GEN <= 7
             .BufferAccessType = pipeline->vb[vb].instanced ? INSTANCEDATA : VERTEXDATA,
             .InstanceDataStepRate = pipeline->vb[vb].instance_divisor,
-            .VertexBufferMemoryObjectControlState = GENX(MOCS),
 #endif
 
             .AddressModifyEnable = true,
@@ -2636,12 +2635,11 @@ emit_vertex_bo(struct anv_cmd_buffer *cmd_buffer,
          .VertexBufferIndex = index,
          .AddressModifyEnable = true,
          .BufferPitch = 0,
+         .VertexBufferMOCS = anv_mocs_for_bo(cmd_buffer->device, addr.bo),
 #if (GEN_GEN >= 8)
-         .MemoryObjectControlState = GENX(MOCS),
          .BufferStartingAddress = addr,
          .BufferSize = size
 #else
-         .VertexBufferMemoryObjectControlState = GENX(MOCS),
          .BufferStartingAddress = addr,
          .EndAddress = anv_address_add(addr, size),
 #endif
@@ -3393,9 +3391,7 @@ cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer)
    if (dw == NULL)
       return;
 
-   struct isl_depth_stencil_hiz_emit_info info = {
-      .mocs = device->default_mocs,
-   };
+   struct isl_depth_stencil_hiz_emit_info info = { };
 
    if (iview)
       info.view = &iview->planes[0].isl;
@@ -3413,6 +3409,8 @@ cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer)
                               image->planes[depth_plane].address.bo,
                               image->planes[depth_plane].address.offset +
                               surface->offset);
+      info.mocs =
+         anv_mocs_for_bo(device, image->planes[depth_plane].address.bo);
 
       const uint32_t ds =
          cmd_buffer->state.subpass->depth_stencil_attachment->attachment;
@@ -3444,6 +3442,8 @@ cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer)
                               image->planes[stencil_plane].address.bo,
                               image->planes[stencil_plane].address.offset +
                               surface->offset);
+      info.mocs =
+         anv_mocs_for_bo(device, image->planes[stencil_plane].address.bo);
    }
 
    isl_emit_depth_stencil_hiz_s(&device->isl_dev, dw, &info);
diff --git a/src/intel/vulkan/genX_gpu_memcpy.c b/src/intel/vulkan/genX_gpu_memcpy.c
index eaafcfa3b22..b51c1804659 100644
--- a/src/intel/vulkan/genX_gpu_memcpy.c
+++ b/src/intel/vulkan/genX_gpu_memcpy.c
@@ -158,11 +158,10 @@ genX(cmd_buffer_so_memcpy)(struct anv_cmd_buffer *cmd_buffer,
          .AddressModifyEnable = true,
          .BufferStartingAddress = { src, src_offset },
          .BufferPitch = bs,
+         .VertexBufferMOCS = anv_mocs_for_bo(cmd_buffer->device, src),
 #if (GEN_GEN >= 8)
-         .MemoryObjectControlState = GENX(MOCS),
          .BufferSize = size,
 #else
-         .VertexBufferMemoryObjectControlState = GENX(MOCS),
          .EndAddress = { src, src_offset + size - 1 },
 #endif
       });
@@ -219,7 +218,7 @@ genX(cmd_buffer_so_memcpy)(struct anv_cmd_buffer *cmd_buffer,
 
    anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_SO_BUFFER), sob) {
       sob.SOBufferIndex = 0;
-      sob.SOBufferObjectControlState = GENX(MOCS);
+      sob.SOBufferMOCS = anv_mocs_for_bo(cmd_buffer->device, dst),
       sob.SurfaceBaseAddress = (struct anv_address) { dst, dst_offset };
 
 #if GEN_GEN >= 8
diff --git a/src/intel/vulkan/genX_state.c b/src/intel/vulkan/genX_state.c
index b1014d9e797..d5d5e43924d 100644
--- a/src/intel/vulkan/genX_state.c
+++ b/src/intel/vulkan/genX_state.c
@@ -93,6 +93,12 @@ genX(init_device_state)(struct anv_device *device)
 {
    GENX(MEMORY_OBJECT_CONTROL_STATE_pack)(NULL, &device->default_mocs,
                                           &GENX(MOCS));
+#if GEN_GEN >= 8
+   GENX(MEMORY_OBJECT_CONTROL_STATE_pack)(NULL, &device->external_mocs,
+                                          &GENX(EXTERNAL_MOCS));
+#else
+   device->external_mocs = device->default_mocs;
+#endif
 
    struct anv_batch batch;