diff --git a/docs/gallium/debugging.rst b/docs/gallium/debugging.rst
index 7252357a945..34dedff542c 100644
--- a/docs/gallium/debugging.rst
+++ b/docs/gallium/debugging.rst
@@ -32,6 +32,12 @@ specified file. Paths may be relative or absolute; relative paths are relative
 to the working directory.  For example, setting it to "trace.xml" will cause
 the trace to be written to a file of the same name in the working directory.
 
+.. envvar:: GALLIUM_TRACE_TC <bool> (false)
+
+If enabled while :ref:`trace` is active, this variable specifies that the threaded context
+should be traced for drivers which implement it. By default, the driver thread is traced,
+which will include any reordering of the command stream from threaded context.
+
 .. envvar:: GALLIUM_TRACE_TRIGGER <string> ("")
 
 If set while :ref:`trace` is active, this variable specifies a filename to monitor.
diff --git a/src/gallium/auxiliary/driver_trace/tr_context.c b/src/gallium/auxiliary/driver_trace/tr_context.c
index 787db7b5f0b..e64e01a5868 100644
--- a/src/gallium/auxiliary/driver_trace/tr_context.c
+++ b/src/gallium/auxiliary/driver_trace/tr_context.c
@@ -1028,8 +1028,6 @@ trace_context_sampler_view_destroy(struct pipe_context *_pipe,
    struct pipe_context *pipe = tr_ctx->pipe;
    struct pipe_sampler_view *view = tr_view->sampler_view;
 
-   assert(_view->context == _pipe);
-
    trace_dump_call_begin("pipe_context", "sampler_view_destroy");
 
    trace_dump_arg(ptr, pipe);
diff --git a/src/gallium/auxiliary/driver_trace/tr_context.h b/src/gallium/auxiliary/driver_trace/tr_context.h
index 95469e87530..a9d75ce921c 100644
--- a/src/gallium/auxiliary/driver_trace/tr_context.h
+++ b/src/gallium/auxiliary/driver_trace/tr_context.h
@@ -55,6 +55,8 @@ struct trace_context
 
    struct pipe_framebuffer_state unwrapped_state;
    bool seen_fb_state;
+
+   bool threaded;
 };
 
 
@@ -77,7 +79,8 @@ struct pipe_context *
 trace_context_create(struct trace_screen *tr_scr,
                      struct pipe_context *pipe);
 
-
+struct pipe_context *
+trace_context_create_threaded(struct pipe_screen *_screen, struct pipe_context *pipe);
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/auxiliary/driver_trace/tr_screen.c b/src/gallium/auxiliary/driver_trace/tr_screen.c
index 2adda8be291..76f947b9386 100644
--- a/src/gallium/auxiliary/driver_trace/tr_screen.c
+++ b/src/gallium/auxiliary/driver_trace/tr_screen.c
@@ -27,7 +27,9 @@
 
 #include "util/format/u_format.h"
 #include "util/u_memory.h"
+#include "util/hash_table.h"
 #include "util/simple_list.h"
+#include "util/u_threaded_context.h"
 
 #include "tr_dump.h"
 #include "tr_dump_defines.h"
@@ -39,6 +41,7 @@
 
 
 static bool trace = false;
+static struct hash_table *trace_screens;
 
 static const char *
 trace_screen_get_name(struct pipe_screen *_screen)
@@ -263,6 +266,28 @@ trace_screen_is_format_supported(struct pipe_screen *_screen,
 }
 
 
+struct pipe_context *
+trace_context_create_threaded(struct pipe_screen *screen, struct pipe_context *pipe)
+{
+   if (!trace_screens)
+      return pipe;
+
+   struct hash_entry *he = _mesa_hash_table_search(trace_screens, screen);
+   if (!he)
+      return pipe;
+   struct trace_screen *tr_scr = trace_screen(he->data);
+
+   if (tr_scr->trace_tc)
+      return pipe;
+
+   struct pipe_context *ctx = trace_context_create(tr_scr, pipe);
+   if (ctx) {
+      struct trace_context *tr_ctx = trace_context(ctx);
+      tr_ctx->threaded = true;
+   }
+   return ctx;
+}
+
 static struct pipe_context *
 trace_screen_context_create(struct pipe_screen *_screen, void *priv,
                             unsigned flags)
@@ -271,19 +296,20 @@ trace_screen_context_create(struct pipe_screen *_screen, void *priv,
    struct pipe_screen *screen = tr_scr->screen;
    struct pipe_context *result;
 
+   result = screen->context_create(screen, priv, flags);
+
    trace_dump_call_begin("pipe_screen", "context_create");
 
    trace_dump_arg(ptr, screen);
    trace_dump_arg(ptr, priv);
    trace_dump_arg(uint, flags);
 
-   result = screen->context_create(screen, priv, flags);
-
    trace_dump_ret(ptr, result);
 
    trace_dump_call_end();
 
-   result = trace_context_create(tr_scr, result);
+   if (result && (tr_scr->trace_tc || result->draw_vbo != tc_draw_vbo))
+      result = trace_context_create(tr_scr, result);
 
    return result;
 }
@@ -707,6 +733,9 @@ trace_screen_fence_finish(struct pipe_screen *_screen,
    struct pipe_context *ctx = _ctx ? trace_context(_ctx)->pipe : NULL;
    int result;
 
+   result = screen->fence_finish(screen, ctx, fence, timeout);
+
+
    trace_dump_call_begin("pipe_screen", "fence_finish");
 
    trace_dump_arg(ptr, screen);
@@ -714,8 +743,6 @@ trace_screen_fence_finish(struct pipe_screen *_screen,
    trace_dump_arg(ptr, fence);
    trace_dump_arg(uint, timeout);
 
-   result = screen->fence_finish(screen, ctx, fence, timeout);
-
    trace_dump_ret(bool, result);
 
    trace_dump_call_end();
@@ -804,6 +831,17 @@ trace_screen_destroy(struct pipe_screen *_screen)
    trace_dump_arg(ptr, screen);
    trace_dump_call_end();
 
+   if (trace_screens) {
+      struct hash_entry *he = _mesa_hash_table_search(trace_screens, screen);
+      if (he) {
+         _mesa_hash_table_remove(trace_screens, he);
+         if (!_mesa_hash_table_num_entries(trace_screens)) {
+            _mesa_hash_table_destroy(trace_screens, NULL);
+            trace_screens = NULL;
+         }
+      }
+   }
+
    screen->destroy(screen);
 
    FREE(tr_scr);
@@ -904,6 +942,12 @@ trace_screen_create(struct pipe_screen *screen)
    trace_dump_ret(ptr, screen);
    trace_dump_call_end();
 
+   if (!trace_screens)
+      trace_screens = _mesa_hash_table_create(NULL, _mesa_hash_pointer, _mesa_key_pointer_equal);
+   _mesa_hash_table_insert(trace_screens, screen, tr_scr);
+
+   tr_scr->trace_tc = debug_get_bool_option("GALLIUM_TRACE_TC", false);
+
    return &tr_scr->base;
 
 error2:
diff --git a/src/gallium/auxiliary/driver_trace/tr_screen.h b/src/gallium/auxiliary/driver_trace/tr_screen.h
index 65ea4fb6a43..295574d1883 100644
--- a/src/gallium/auxiliary/driver_trace/tr_screen.h
+++ b/src/gallium/auxiliary/driver_trace/tr_screen.h
@@ -51,6 +51,7 @@ struct trace_screen
    struct pipe_screen base;
 
    struct pipe_screen *screen;
+   bool trace_tc;
 };
 
 
diff --git a/src/gallium/auxiliary/driver_trace/tr_texture.c b/src/gallium/auxiliary/driver_trace/tr_texture.c
index d644e1b08ba..fd9b3d41884 100644
--- a/src/gallium/auxiliary/driver_trace/tr_texture.c
+++ b/src/gallium/auxiliary/driver_trace/tr_texture.c
@@ -70,7 +70,6 @@ error:
 void
 trace_surf_destroy(struct trace_surface *tr_surf)
 {
-   trace_context_check(tr_surf->base.context);
    pipe_resource_reference(&tr_surf->base.texture, NULL);
    pipe_surface_reference(&tr_surf->surface, NULL);
    FREE(tr_surf);
@@ -91,15 +90,15 @@ trace_transfer_create(struct trace_context *tr_ctx,
    if (!tr_trans)
       goto error;
 
-   memcpy(&tr_trans->base, transfer, sizeof(struct pipe_transfer));
+   memcpy(&tr_trans->base, transfer, tr_ctx->threaded ? sizeof(struct threaded_transfer) : sizeof(struct pipe_transfer));
 
-   tr_trans->base.resource = NULL;
+   tr_trans->base.b.resource = NULL;
    tr_trans->transfer = transfer;
 
-   pipe_resource_reference(&tr_trans->base.resource, res);
-   assert(tr_trans->base.resource == res);
+   pipe_resource_reference(&tr_trans->base.b.resource, res);
+   assert(tr_trans->base.b.resource == res);
 
-   return &tr_trans->base;
+   return &tr_trans->base.b;
 
 error:
    tr_ctx->pipe->transfer_unmap(tr_ctx->pipe, transfer);
@@ -111,7 +110,7 @@ void
 trace_transfer_destroy(struct trace_context *tr_context,
                        struct trace_transfer *tr_trans)
 {
-   pipe_resource_reference(&tr_trans->base.resource, NULL);
+   pipe_resource_reference(&tr_trans->base.b.resource, NULL);
    FREE(tr_trans);
 }
 
diff --git a/src/gallium/auxiliary/driver_trace/tr_texture.h b/src/gallium/auxiliary/driver_trace/tr_texture.h
index e5dfc53fdb0..567842a5b22 100644
--- a/src/gallium/auxiliary/driver_trace/tr_texture.h
+++ b/src/gallium/auxiliary/driver_trace/tr_texture.h
@@ -33,6 +33,7 @@
 #include "pipe/p_state.h"
 
 #include "tr_screen.h"
+#include "util/u_threaded_context.h"
 
 struct trace_context;
 
@@ -63,7 +64,7 @@ struct trace_sampler_view
 
 struct trace_transfer
 {
-   struct pipe_transfer base;
+   struct threaded_transfer base;
 
    struct pipe_transfer *transfer;
    struct pipe_context *pipe;
diff --git a/src/gallium/auxiliary/util/u_threaded_context.c b/src/gallium/auxiliary/util/u_threaded_context.c
index 56cb4f0de2e..75f75312613 100644
--- a/src/gallium/auxiliary/util/u_threaded_context.c
+++ b/src/gallium/auxiliary/util/u_threaded_context.c
@@ -30,6 +30,7 @@
 #include "util/u_inlines.h"
 #include "util/u_memory.h"
 #include "util/u_upload_mgr.h"
+#include "driver_trace/tr_context.h"
 #include "util/log.h"
 #include "compiler/shader_info.h"
 
@@ -3091,6 +3092,8 @@ threaded_context_create(struct pipe_context *pipe,
    }
    memset(tc, 0, sizeof(*tc));
 
+   pipe = trace_context_create_threaded(pipe->screen, pipe);
+
    assert((uintptr_t)tc % 16 == 0);
    /* These should be static asserts, but they don't work with MSVC */
    assert(offsetof(struct threaded_context, batch_slots) % 16 == 0);