Re: sysctl -a is slow

Jaakko Heinonen wrote:
> On 2010-09-20, David Xu wrote:
>> I redirect all output to a disk file, and it still needs 1 second to
>> complete, this machine is dual-core pentium E5500, faster than previous
>> one which is a dual-core AMD 5000+ machine, the 5000+ needs 2
>> seconds to complete.
>>
>> $/usr/bin/time sysctl -b kern.geom.confdot > sysctl_geom_confdot.txt
>>         1.00 real         0.00 user         0.00 sys
> 
> I couldn't reproduce the problem myself but I bet that it's a lost
> wakeup in g_waitfor_event(). Can you try this patch?
> 
> %%%
> Index: sys/geom/geom_event.c
> ===================================================================
> --- sys/geom/geom_event.c	(revision 214048)
> +++ sys/geom/geom_event.c	(working copy)
> _at__at_ -220,11 +220,12 _at__at_ one_event(void)
>  	mtx_lock(&g_eventlock);
>  	TAILQ_REMOVE(&g_events, ep, events);
>  	ep->flag &= ~EV_INPROGRESS;
> -	mtx_unlock(&g_eventlock);
>  	if (ep->flag & EV_WAKEUP) {
>  		ep->flag |= EV_DONE;
>  		wakeup(ep);
> +		mtx_unlock(&g_eventlock);
>  	} else {
> +		mtx_unlock(&g_eventlock);
>  		g_free(ep);
>  	}
>  	g_topology_unlock();
> _at__at_ -365,11 +366,14 _at__at_ g_waitfor_event(g_event_t *func, void *a
>  	va_end(ap);
>  	if (error)
>  		return (error);
> -	do 
> -		tsleep(ep, PRIBIO, "g_waitfor_event", hz);
> -	while (!(ep->flag & EV_DONE));
> +
> +	mtx_lock(&g_eventlock);
> +	while (!(ep->flag & EV_DONE))
> +		msleep(ep, &g_eventlock, PRIBIO, "g_waitfor_event", hz);
>  	if (ep->flag & EV_CANCELED)
>  		error = EAGAIN;
> +	mtx_unlock(&g_eventlock);
> +
>  	g_free(ep);
>  	return (error);
>  }
> %%%
> 
Yes, the patch seems fixed the problem, I can not reproduce it.

Thanks!