[ruby-core:89909] Re: Thread::Light patch against r65832

Koichi Sasada <[e-mail]> wrote:
> On 2018/11/20 17:44, Eric Wong wrote:
> > Mutex and ConditionVariable are NOT scheduling points for
> > Thread::Light switching; however they may process signal
> > handling and handle I/O dispatch for other native threads.
> 
> Why not?
> How to synchronize multiple Thread::Light instances (lthreads here)?

I'm not sure how deadlock detection would work, and I don't
think there is data race there.

Main synchronization should be Queue/SizedQueue (like "mailbox"
in actor model).

But I think this is a good change to maintain compatibility
and avoid inadvertant switching:

diff --git a/vm_core.h b/vm_core.h
index 9e10b321da..2244afb524 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -1842,24 +1842,27 @@ static inline int
 rb_tl_switchable(const rb_execution_context_t *ec)
 {
     const rb_thread_t *th = rb_ec_thread_ptr(ec);
 
     /* dangerous, don't allow switching inside trap handler: */
     if (ec->interrupt_mask & TRAP_INTERRUPT_MASK) return 0;
 
+    /* don't switch if a Mutex is held */
+    if (th->keeping_mutexes) return 0;
+
     /* auto-fibers can switch away to root fiber */
     if (rb_tl_sched_p(ec)) return 1;
 
     /* no auto-fibers, yet, but we can create and switch to them */
     if (!th->root_fiber) return 1;
 
     /* root fiber can switch to auto-fibers, because ensure works */
     if (th->root_fiber == ec->fiber_ptr) return 1;
 
     /*
      * no auto-switching away from regular Fibers because they lack
      * ensure support: https://bugs.ruby-lang.org/issues/595
      */
     return 0;
 }
 
 /* tracer */

(Gotta run, back in 16 hours maybe?)

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>