Handling __cxa_atexit(a,b,NULL)

[Kamil Rytarowski <n54%gmx.com@localhost>]

I've narrowed down a bug in TSan/NetBSD with __cxa_atexit(a,b,NULL)
handling. I'm not sure whether the problem is in NetBSD or TSan.

extern "C" int __cxa_atexit ( void (*f)(void *), void *p, void *d );

TSan intercepts all atexit() and __cxa_atexit() calls and wraps them in
setup_at_exit_wrapper(() into a new struct that preserves the original
arguments "a" and "b" from __cxa_atexit() / "a" from atexit(). In case
of intercepting atexit() calls we end up with __cxa_atexit(x,y,NULL). In
libc we handle this variation in the same way as atexit(x) and drop the
"y" argument.

static int setup_at_exit_wrapper(ThreadState *thr, uptr pc, void(*f)(),
      void *arg, void *dso) {
  AtExitCtx *ctx = (AtExitCtx*)InternalAlloc(sizeof(AtExitCtx));
  ctx->f = f;
  ctx->arg = arg;
  Release(thr, pc, (uptr)ctx);
  // Memory allocation in __cxa_atexit will race with free during exit,
  // because we do not see synchronization around atexit callback list.
  ThreadIgnoreBegin(thr, pc);


  int res = REAL(__cxa_atexit)(at_exit_wrapper, ctx, dso);
  ThreadIgnoreEnd(thr, pc);
  return res;
}


Itanium C++ ABI:

https://itanium-cxx-abi.github.io/cxx-abi/abi.html#dso-dtor-motivation

Assuming that the problem is in NetBSD, I've verified that this patch works:

http://netbsd.org/~kamil/patch-00038-__cxa_atexit.txt

Attachment: signature.asc
Description: OpenPGP digital signature