Re: softirq parameters

[Matthew Wilcox]

On Wed, Aug 07, 2002 at 03:24:23PM +1000, Rusty Russell wrote:
> Partially agree.  Removing all args might be worthwhile.  But all these
> softirqs use the "cpu" arg to access per-cpu data, which should be
> changed to use the per_cpu_data mechanism anyway, which removes the
> point of the arg.

I see.  That makes a lot of sense.

> Things haven't been changed over because I haven't pushed the per-cpu
> interface changes (required for some archs 8() to Linus yet.  But you'll
> want them so we can save space (you only need allocate per-cpu data for
> cpus where cpu_possible(i) is true).

So what we want is something more like:

struct softnet_data softnet_data __per_cpu_data = { NULL };

static void void net_tx_action(void *arg)
{
	struct softnet_data *data = arg;
	if (arg->completion_queue) {
	...
}

	open_softirq(NET_TX_SOFTIRQ, net_tx_action, softnet_data);

and have kernel/softirq.c do:

                do {
                        if (pending & 1)
                                h->action(this_cpu(h->data));
                        h++;
                        pending >>= 1;
                } while (pending);

right?

-- 
Revolutions do not require corporate support.