Wireless networking without CONFIG_PM..

I wonder if anybody has really ever tested that? Because I think it's broken..

In particular, I made the mistake of not enabling CONFIG_PM on a new
laptop, and it caused some *seriously* nasty-to-debug problems. The
mac80211 code goes crazy, that upsets the wireless driver, and then
the wrieless driver in question had a nasty bug where it would
double-release its firmware, which then caused vmalloc corruption and
all kinds of really odd recursive panics etc.

And as far as I can tell, the root cause of the problem is a bad
choice in net/mac80211/main.c:

  int ieee80211_register_hw(struct ieee80211_hw *hw)
  {
  ...
        if ((hw->wiphy->wowlan.flags || hw->wiphy->wowlan.n_patterns)
  #ifdef CONFIG_PM
            && (!local->ops->suspend || !local->ops->resume)
  #endif
            )
                return -EINVAL;

which means that if the wiphy says it supports wake-on-wireless lan,
and CONFIG_PM isn't enabled, we'll return -EINVAL and will refuse to
do any wireless at all.

It's that a bit extreme? Or outright stupid? What is the advantage of
saying that "if you don't have CONFIG_PM enabled, we'll just refuse to
register any hardware that talks about it's wake-on-wireless
patterns"?

Maybe there is some reason for that return -EINVAL, but I don't think
it makes sense with that particular placement of #ifdef CONFIG_PM.
Maybe if the #ifdef was around the whole test? Or maybe it should just
be removed.

Or am I missing some big reason for why it makes sense to do that? Hmm?

I'll make a separate bug-report email to the intel iwlwifi people
about their absolutely horribly broken error handling which then made
it such a disaster, but I thought I'd bring the generic mac80211 issue
up. I don't think there are a lot of drivers that do the whole wowlan
thing, and obviously most people use wireless on laptops where you
want CONFIG_PM anyway, so it probably hasn't triggered very much.

                  Linus