bug#30322: Emacs 26 seems to have entered an infinite loop during GC on macOS

[John Wiegley]

Hi Eli,

I was writing an e-mail to a colleague today when my Emacs window completely 
froze up, taking 100% CPU constantly.  Since it's still in that state now, I 
decided to fire up the stochastic profiler to see what it's doing.  I've 
attached the screenshot.

It claims to be spending all its time in `lisp_align_free', behaving as if 
there's a CPU busy loop now taking place inside that function. Since very 
little time is being assigned to its callees, I believe an infinite loop has 
occurred here:

      int i = 0;
      bool aligned = busy;
      struct ablock **tem = &free_ablock;
      struct ablock *atop = &abase->blocks[aligned ? ABLOCKS_SIZE : 
ABLOCKS_SIZE - 1];

      while (*tem)
        {
          if (*tem >= (struct ablock *) abase && *tem < atop)
            {
              i++;
              *tem = (*tem)->x.next_free;
            }
          else
            tem = &(*tem)->x.next_free;
        }

I would like to suggest that before entering this loop, we record the initial 
value of *tem, and if we encounter again (or if we encounter it again X number 
of times), we report a non-fatal error and exit this function.  That way I 
could save my work and restart Emacs.

I've noticed such a "lock up until I force quit" happening before, but it's 
very rare. Maybe a few times a year?

John

Screen Shot 2018-02-01 at 12.03.14 PM.png
Description: PNG image