On Dec 4 20:41, Takashi Yano wrote:
> Currently, the signal queue is touched by the thread sig as well as
> other threads that call sigaction_worker(). This potentially has
> a possibility to destroy the signal queue chain. A possible worst
> result may be a self-loop chain which causes infinite loop. With
> this patch, lock()/unlock() are introduce to avoid such a situation.
I was hoping for a lockfree solution, but never mind that for now.
>
> diff --git a/winsup/cygwin/sigproc.cc b/winsup/cygwin/sigproc.cc
> index 7e02e61f7..cc3113b88 100644
> --- a/winsup/cygwin/sigproc.cc
> +++ b/winsup/cygwin/sigproc.cc
> @@ -106,12 +106,27 @@ class pending_signals
> {
> sigpacket sigs[_NSIG + 1];
> sigpacket start;
> + volatile unsigned locked;
> bool retry;
> + void lock ()
> + {
> + while (InterlockedExchange (&locked, 1))
> + {
> +#ifdef __x86_64__
> + __asm__ ("pause");
> +#else
> +#error unimplemented for this target
> +#endif
> + yield ();
> + }
> + }
> + void unlock () { locked = 0; }
>
> public:
> + pending_signals (): locked(0) {}
> void add (sigpacket&);
> bool pending () {retry = !!start.next; return retry;}
> - void clear (int sig);
> + void clear (int sig, bool need_lock);
> void clear (_cygtls *tls);
> friend void sig_dispatch_pending (bool);
> friend void wait_sig (VOID *arg);
Given this is used in C-code only, what about just using SRWLocks
in Exclusive-only mode, rather than a spinlock?
Thanks,
Corinna
