Re: [PATCH v4] eliminate mutex in fast path of __register_frame

[Thomas Neumann via Gcc-patches]

At least when building LibreOffice with Clang (16 trunk) with ASan and 
UBsan enabled against libstdc++ (with --gcc-toolchain and 
LD_LIBRARY_PATH to pick up a libstdc++ trunk build including this change 
at build and run-time), at least one of the LibreOffice tests executed 
during the build started to fail with

I could not reproduce the issue when building LibreOffice with gcc, but 
after reading the compiler-rt version of crtbegin.c I think the problem 
is the destruction order in compiler-rt. It calls 
__deregister_frame_info_bases after our lookup structure has already 
been destroyed.

Can you try if the patch below fixes the problem? It keeps the data 
structures alive at shutdown, though, which will probably make some leak 
detectors unhappy.

Alternatively we could simply remove the gcc_assert (ob) in line 285 of 
that file. As far as I can see in crt-begin nothing bad happens if we 
return nullptr at shutdown.

Best

Thomas


diff --git a/libgcc/unwind-dw2-fde.c b/libgcc/unwind-dw2-fde.c
index 919abfe0664..d427318280c 100644
--- a/libgcc/unwind-dw2-fde.c
+++ b/libgcc/unwind-dw2-fde.c
@@ -49,16 +49,6 @@ typedef __UINTPTR_TYPE__ uintptr_type;

 static struct btree registered_frames;

-static void
-release_registered_frames (void) __attribute__ ((destructor (110)));
-static void
-release_registered_frames (void)
-{
-  /* Release the b-tree and all frames. Frame releases that happen 
later are
-   * silently ignored */
-  btree_destroy (&registered_frames);
-}
-
 static void
 get_pc_range (const struct object *ob, uintptr_type *range);
 static void