Hello,

I started making the core of if_bah.c (the SMC chipset ARCnet driver
for NetBSD/Amiga) MI, and to write an ISA frontend for it. I noticed
that it would be VERY NICE to have a MI version of softint callbacks. 

I'd propose just to adopt the Amiga interface, but cgd convinced me
that it should be modified... I summarize a proposal for an interface:

/*
 * interface to interested parties, e.g. drivers
 */

typedef void (*sicb_func_t)(void *p);

struct sicb {
	sicb_func_t f;
	void *p;
	TAILQ_ENTRY(sicb) tq;
};



int add_sicallback __P((struct sicb *));
/*
 * add_sicallback enters its parameter into si_queue.
 * after that, a soft int is generated (e.g., using the force bit for
 * IPL 1 on Amiga or DraCo), which will be executed when no other
 * interupts are pending anymore.
 * this interupt will call call_sicallback(), which will call all
 * entries on the list and remove them.
 * Note that be might get away with a LIST_ENTRY, at the cost of
 * calling the functions LIFO.
 */


/*
 * some internal stuff
 */

TAILQ_HEAD(sicb) si_queue;


/*
 * Is called at lowest spl when no other interrupts are pending.
 * Dequeues and executes each entry it finds in turn as follows:
 * p->si_func(p->si_par);
 */

call_sicallback __P((void));

	---

We can actually implement most of this MI, given a MD initialization
callback, which will register call_sicallback with the interupt
machinery, and a setsoftcback() function (macro?), which will do the
right thing to make call_sicallback be called eventually.

Hm, low effort implementation would just call call_sicallback directly
on setsoftcback(); you wouldn't get the benefits of calling the stuff
at low priority without blocking ohter interrupts, but code  using it
would at least run.

One last remark: I've the impression, that we could either implement
this through the soft network queue mechanism, or use it to
reimplement the latter in a machine-independent way.

Regards,
	Ignatios Souvatzis