[PATCH] hostapd: Add wowlan_triggers config param
Sat Sep 13 13:40:17 PDT 2014
On Sat, Sep 13, 2014 at 7:17 AM, Jouni Malinen <j at w1.fi> wrote:
> On Wed, Sep 03, 2014 at 02:58:37PM -0700, Dmitry Shmidt wrote:
> > New kernels in wiphy_suspend() will call cfg80211_leave_all()
> > that will eventually end up in cfg80211_stop_ap() unless
> > wowlan_triggers were set.
> Could you please clarify why this is needed? If kernel behavior changed
> at some point to do something that is not desired, wouldn't the correct
> way of fixing this be by reverting parts of the kernel behavior rather
> than by trying to make hostapd do something.
On some stage (from kernel 3.8 I think) wireless stack calls
on suspend if no wowlan triggers were registered - I think logic here was to
disconnect if system can not wake from packet anyway.
This logic was expanded to AP mode as well (in case of AP cfg80211_stop_ap()
will be called). It can also be explained - if you can not wake main cpu on
you can not be "good" AP.
So in order to be able to work as AP we need to honor wireless stack logic.
And I think that it is not a kernel bug.
> I don't understand what WoWLAN triggers have to do with AP mode
> operations. Or would there really be a case where WoWLAN can be used
> from a station device to wake up the AP?
I suspect nobody expected AP's cpu can go to suspend mode. But on embedded
system where most of simple AP operations are done in wlan FW, it makes
to enable main cpu suspend.
> I'm assuming this is trying to undo some parts of kernel commit
> 8125696991194aacb1173b6e8196d19098b44e17 ('cfg80211/mac80211: disconnect
> on suspend'). Is that the case or did I miss what the main point of the
> hostapd changes here would be?
> Jouni Malinen PGP id EFC895FA
> HostAP mailing list
> HostAP at lists.shmoo.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Hostap