Re: Problem compiling sched_ule.c

On Saturday 16 July 2005 10:09 pm, Craig Rodrigues wrote:
> Hi,
>
> After the recent changes to the macros to atomic.h, I am getting
> compilation warnings in sched_ule.c.
>
> /usr/src/sys/kern/sched_ule.c: In function `kseq_assign':
> /usr/src/sys/kern/sched_ule.c:654: warning: passing arg 1 of
> `atomic_cmpset_int' from incompatible pointer type
> /usr/src/sys/kern/sched_ule.c:654: warning: passing arg 2 of
> `atomic_cmpset_int' makes integer from pointer without a cast
> /usr/src/sys/kern/sched_ule.c:654: warning: passing arg 3 of
> `atomic_cmpset_int' makes integer from pointer without a cast
> /usr/src/sys/kern/sched_ule.c: In function `kseq_notify':
> /usr/src/sys/kern/sched_ule.c:691: warning: passing arg 1 of
> `atomic_cmpset_int' from incompatible pointer type
> /usr/src/sys/kern/sched_ule.c:691: warning: passing arg 2 of
> `atomic_cmpset_int' makes integer from pointer without a cast
> /usr/src/sys/kern/sched_ule.c:691: warning: passing arg 3 of
> `atomic_cmpset_int' makes integer from pointer without a cast *** Error
> code 1
>
>
> Is this the correct way to fix this:
>
>
> --- /usr/src/sys/kern/sched_ule.c.orig	Sat Jul 16 21:42:07 2005
> +++ /usr/src/sys/kern/sched_ule.c	Sat Jul 16 22:09:00 2005
> _at__at_ -651,7 +651,8 _at__at_
>
>  	do {
>  		*(volatile struct kse **)&ke = kseq->ksq_assigned;
> -	} while(!atomic_cmpset_ptr(&kseq->ksq_assigned, ke, NULL));
> +	} while(!atomic_cmpset_ptr((uintptr_t *)&kseq->ksq_assigned,
> +	        (uintptr_t)ke, (uintptr_t)NULL));
>  	for (; ke != NULL; ke = nke) {
>  		nke = ke->ke_assign;
>  		kseq->ksq_group->ksg_load--;
> _at__at_ -688,7 +689,8 _at__at_
>  	 */
>  	do {
>  		*(volatile struct kse **)&ke->ke_assign = kseq->ksq_assigned;
> -	} while(!atomic_cmpset_ptr(&kseq->ksq_assigned, ke->ke_assign, ke));
> +	} while(!atomic_cmpset_ptr((uintptr_t *)&kseq->ksq_assigned,
> +	        (uintptr_t)ke->ke_assign, (uintptr_t)ke));
>  	/*
>  	 * Without sched_lock we could lose a race where we set NEEDRESCHED
>  	 * on a thread that is switched out before the IPI is delivered.  This

Yes.

-- 
John Baldwin <jhb_at_FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org