[LLVMdev] Multi-threading and mutexes in LLVM

[David Chisnall]

On 7 Jun 2014, at 07:50, Chandler Carruth <chandlerc at google.com> wrote:

> I don't see any reason not to reserve a word in the mutex so that in (an ABI-compatible) debug build the mutex can support deadlock detection. Some people are super concerned about having an extra word in a mutex, but I'm not at all.

Making a mutex span multiple cache lines can have very serious and unpredictable performance impacts in the contended case (if you're not contended, a mutex is probably the wrong synchronisation primitive for you) on modern CPUs.

> For libc++, it would probably need to be behind an ABI-breaking macro on Mac and FreeBSD, but we haven't committed to any ABI stability on Linux, so we could probably enable it by default there, and get into build bots.
> 
> Maybe bring this up on the cfe-dev list to discuss with Marshall and other folks interested in libc++?

On FreeBSD and OS X, the underlying pthread_mutex can already do deadlock detection, so I don't see why you'd need to add another word.  The PTHREAD_MUTEX_ERRORCHECK attribute has been part of POSIX since 1997, so I'd expect it to be supported everywhere.  

For FreeBSD, we also had a GSoC student a couple of years ago who ported the WITNESS lock order reversal checking framework from the kernel to userspace, for more detailed checking.

David