Patch ping

[Paolo Carlini]

Jonathan Wakely wrote:

>The main reason I thought of the CAS solution is that it wouldn't change
>the ABI, but hopefully you'll be able to squeeze a mutex in somewhere
>without too much trouble.
>  
>
You are right, that would make everything much simpler from the ABI
point of view, but really,  if we want to fix this bug in the 4.1 time
frame (for concreteness) either we use a mutex or we use only
__atomic_add and __exchange_and_add. This is our toolbox.

I'm also thinking that the memory leak is demonstrably bounded and if we
could find a very simple way to make it much smaller whereas not zero
yet, would be still a satisfactory improvement, for now.

Paolo.