Re: CVS commit: src/sys/netinet6

[Ryota Ozaki <ozaki-r%netbsd.org@localhost>]

On Tue, Dec 26, 2017 at 11:35 AM, Paul Goyette <paul%whooppee.com@localhost> wrote:
>
>> To generate a diff of this commit:
>
>
> # cvs rdiff -u -r1.139 -r1.140 src/sys/netinet6/nd6_nbr.c
> @@ -1097,7 +1097,11 @@ nd6_dad_stoptimer(struct dadq *dp)
>  #ifdef NET_MPSAFE
>         callout_halt(&dp->dad_timer_ch, NULL);
>  #else
> -       callout_halt(&dp->dad_timer_ch, softnet_lock);
> +       /* XXX still need the trick for softnet_lock */
> +       if (mutex_owned(softnet_lock))
> +               callout_halt(&dp->dad_timer_ch, softnet_lock);
> +       else
> +               callout_halt(&dp->dad_timer_ch, NULL);
>  #endif
>  }
>
> This goes against the restriction noted in the mutex(9) man page:
>
>    [mutex_owned()] should not be used to make locking decisions at run
>    time.  ...
>
> Please find a different way to make this run-time decision.

I know the restriction well, but for softnet_lock, following the restriction
is sometimes hard. I don't have an option to statically decide if
softnet_lock is held or not without messing up many codes.

An option we can have here is to give up using callout_halt and use
callout_stop that may be unsafe.

Which is better for us?

  ozaki-r