[stable 9] panic on reboot: ipmi_wd_event()

Eugene Grosbein egrosbein at rdtc.ru
Tue Aug 7 09:37:31 UTC 2012


> Ok, can you try this instead of the other patch.  It makes the IPMI driver not 
> bother clearing the watchdog if it isn't running.

I suffered from exactly same problem (same kdb backtrace)
and this patch fixed it, thanks!

> Index: dev/ipmi/ipmi.c
> ===================================================================
> --- dev/ipmi/ipmi.c	(revision 238992)
> +++ dev/ipmi/ipmi.c	(working copy)
> @@ -653,11 +653,12 @@
>  		if (timeout == 0)
>  			timeout = 1;
>  		e = ipmi_set_watchdog(sc, timeout);
> -		if (e == 0)
> +		if (e == 0) {
>  			*error = 0;
> -		else
> +			sc->ipmi_watchdog_active = 1;
> +		} else
>  			(void)ipmi_set_watchdog(sc, 0);
> -	} else {
> +	} else if (atomic_readandclear_int(&sc->ipmi_watchdog_active) != 0) {
>  		e = ipmi_set_watchdog(sc, 0);
>  		if (e != 0 && cmd == 0)
>  			*error = EOPNOTSUPP;
> Index: dev/ipmi/ipmivars.h
> ===================================================================
> --- dev/ipmi/ipmivars.h	(revision 238992)
> +++ dev/ipmi/ipmivars.h	(working copy)
> @@ -105,6 +105,7 @@
>  	struct cdev		*ipmi_cdev;
>  	TAILQ_HEAD(,ipmi_request) ipmi_pending_requests;
>  	eventhandler_tag	ipmi_watchdog_tag;
> +	int			ipmi_watchdog_active;
>  	struct intr_config_hook	ipmi_ich;
>  	struct mtx		ipmi_lock;
>  	struct cv		ipmi_request_added;
> 



More information about the freebsd-stable mailing list