Re: [PATCH] Remove More Unneccessary CPU Notifiers

[Rusty Russell]

In message <20040124101039.296c34fd.akpm@xxxxxxxx> you write:
> Or are you saying that we should just leave the per-cpu accounting in a
> non-zero state when its CPU has gone away, and rely upon the stats
> gathering code iterating across all cpu_possible cpus?

In general, yes!  In general, if you cared about performance you
wouldn't be doing such iteration.

> That's a bit lame in the case of __get_page_state() at least.  We've had
> problems with excess CPU consumption in there at times and it would be good
> to be able to change that function to iterate across all online CPUs, not
> all possible ones.  We can do that if we have a notifier which spills the
> numbers from the gone-away CPU into the local CPU's slot.

Well, that's what's happening at the moment if you look at the code:

	while (cpu < NR_CPUS) {
		unsigned long *in, *out, off;

		if (!cpu_possible(cpu)) {
			cpu++;
			continue;
		}

Spilling the stats is a fine optimization, sure, but that can come
later.

Especially since it need only be done at CPU_DEAD time: the hotplug
CPU patch adds a convenient macro for such things.
"hotcpu_notifier()" compiles out when !CONFIG_HOTPLUG_CPU.

What's there at the moment really is just wasted code.
RUsty.
--
  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/