glthread, tc: Fix buffer release with glthread and tc

_mesa_glthread_release_upload_buffer will call _mesa_reference_buffer_object
which will then end up in tc_resource_release.

Doing so from glthread is bad because we now have 2 threads concurrently
accessing TC batch: glthread and the unmarshalling/tc thread.

Fix this by queuing the release call through an "Internal" glthread function.

Fixes: b3133e250e ("gallium: add pipe_context::resource_release to eliminate buffer refcounting")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37763>

src/mesa/glapi/glapi/gen/gl_API.xml

@@ -11097,6 +11097,11 @@
     <!-- Internal function for glthread to implement ancillary buffer invalidation. -->
     <function name="InternalInvalidateFramebufferAncillaryMESA" es2="3.0">
     </function>
+
+    <!-- Internal function for glthread to defer buffer release. -->
+    <function name="InternalReleaseBufferMESA" es2="3.0" marshal="async">
+        <param name="buffer" type="GLvoid *"/>
+    </function>
 </category>
 
 <xi:include href="OES_EGL_image.xml" xmlns:xi="http://www.w3.org/2001/XInclude"/>

src/mesa/glapi/glapi/gen/static_data.py

@@ -1612,6 +1612,7 @@ all_functions = [
     "DrawArraysInstancedBaseInstanceDrawID",
     "DrawElementsInstancedBaseVertexBaseInstanceDrawID",
     "InternalInvalidateFramebufferAncillaryMESA",
+    "InternalReleaseBufferMESA",
     "DrawElementsPacked",
     "DrawElementsUserBufPacked",
     "TexStorageAttribs2DEXT",

src/mesa/main/glthread.c

@@ -284,7 +284,7 @@ _mesa_glthread_destroy(struct gl_context *ctx)
          util_queue_fence_destroy(&glthread->batches[i].fence);
 
       _mesa_DeinitHashTable(&glthread->VAOs, free_vao, NULL);
-      _mesa_glthread_release_upload_buffer(ctx);
+      _mesa_glthread_release_upload_buffer(ctx, false);
    }
 }
 

src/mesa/main/glthread.h

@@ -331,7 +331,7 @@ void _mesa_glthread_flush_batch(struct gl_context *ctx);
 void _mesa_glthread_finish(struct gl_context *ctx);
 void _mesa_glthread_finish_before(struct gl_context *ctx, const char *func);
 bool _mesa_glthread_invalidate_zsbuf(struct gl_context *ctx);
-void _mesa_glthread_release_upload_buffer(struct gl_context *ctx);
+void _mesa_glthread_release_upload_buffer(struct gl_context *ctx, bool async_release);
 void _mesa_glthread_upload(struct gl_context *ctx, const void *data,
                            GLsizeiptr size, unsigned *out_offset,
                            struct gl_buffer_object **out_buffer,
@@ -405,6 +405,8 @@ void _mesa_glthread_unbind_uploaded_vbos(struct gl_context *ctx);
 void _mesa_glthread_PixelStorei(struct gl_context *ctx, GLenum pname,
                                 GLint param);
 
+void _mesa_InternalReleaseBufferMESA(void *buffer);
+
 #ifdef __cplusplus
 }
 #endif

src/mesa/main/glthread_bufferobj.c

@@ -64,8 +64,15 @@ new_upload_buffer(struct gl_context *ctx, GLsizeiptr size, uint8_t **ptr)
    return obj;
 }
 
+void _mesa_InternalReleaseBufferMESA(void *buffer)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct gl_buffer_object *buf = buffer;
+   _mesa_reference_buffer_object(ctx, &buf, NULL);
+}
+
 void
-_mesa_glthread_release_upload_buffer(struct gl_context *ctx)
+_mesa_glthread_release_upload_buffer(struct gl_context *ctx, bool async_release)
 {
    struct glthread_state *glthread = &ctx->GLThread;
 
@@ -79,7 +86,14 @@ _mesa_glthread_release_upload_buffer(struct gl_context *ctx)
       p_atomic_add(&glthread->upload_buffer->RefCount,
                    -(PRIVATE_REFCOUNT - glthread->upload_buffer->CtxRefCount));
    }
+
+   if (async_release) {
+      /* Defer to avoid calling tc_resource_release from this thread. */
+      _mesa_marshal_InternalReleaseBufferMESA(glthread->upload_buffer);
+      glthread->upload_buffer = NULL;
+   } else {
       _mesa_reference_buffer_object(ctx, &glthread->upload_buffer, NULL);
+   }
 }
 
 void
@@ -120,7 +134,7 @@ _mesa_glthread_upload(struct gl_context *ctx, const void *data,
          return;
       }
 
-      _mesa_glthread_release_upload_buffer(ctx);
+      _mesa_glthread_release_upload_buffer(ctx, true);
 
       glthread->upload_buffer =
          new_upload_buffer(ctx, default_size, &glthread->upload_ptr);