So this means that we will leak the memory we have allocated for the alternate stack when ASAN is enabled. That would happen for every thread that terminates. I also wonder if ASAN allocates large enough alternate stack. And what does it do with the alternate stack we have allocated? Does it just overwrite it with its own?

Asan sets alternate stack before coreclr tries to do it. So when EnsureSignalAlternateStack is called it checks that alternate stack already exists and returns:

BOOL EnsureSignalAlternateStack()
{
    stack_t oss;

    // Query the current alternate signal stack
    int st = sigaltstack(NULL, &oss);

    if ((st == 0) && (oss.ss_flags == SS_DISABLE))
    {
        ...
    }

    return (st == 0);
}

As EnsureSignalAlternateStack doesn't allocate memory FreeSignalAlternateStack shouldn't free it.

Size of Asan's alternate stack is 4 * SIGSTKSZ (sanitizer_posix_libcdep.cc), coreclr sets alternate with the following size:

int altStackSize = SIGSTKSZ + ALIGN_UP(sizeof(SignalHandlerWorkerReturnPoint), 16) + GetVirtualPageSize();

Using option use_sigaltstack we can disable setting alternate stack in Asan, but as coreclr's size of the alternate stack could be less than Asan's one it can lead to some problems. For example Android also sets alternate stack but it's too small for Asan:

  // If the alternate stack is already in place, do nothing.
  // Android always sets an alternate stack, but it's too small for us.
  if (!SANITIZER_ANDROID && !(oldstack.ss_flags & SS_DISABLE)) return;

@kbaladurin I am sorry for a late response, I was on a paternity leave for the last month. Looking at the code in the sanitizer, their alt stack allocation doesn't create a guard page that enables catching stack overflow. So I'd prefer disabling their creation of the alt stack using the option you have mentioned and rather increase our alt stack size if the HAS_ASAN is defined to the value they use (resp, to the max of what we use and they use, to stay on the safe size).

@janvorli thank you for suggestion! I've updated PR.