Subject: Re: Incorrect signal levels reported using wpa_supplicant's
 driver_nl80211
From: Maxim Levitsky <maximlevitsky@gmail.com>
To: Dan Williams <dcbw@redhat.com>
Cc: "hostap@lists.shmoo.com" <hostap@lists.shmoo.com>,
	linux-wireless <linux-wireless@vger.kernel.org>,
	networkmanager-list <networkmanager-list@gnome.org>
In-Reply-To: <1257190833.1027.4.camel@localhost.localdomain>
References: <1256953837.9681.13.camel@maxim-laptop>
	 <1257190833.1027.4.camel@localhost.localdomain>
Content-Type: text/plain; charset="UTF-8"
Date: Tue, 03 Nov 2009 01:07:22 +0200
Message-ID: <1257203242.6717.14.camel@maxim-laptop>
Mime-Version: 1.0
Sender: linux-wireless-owner@vger.kernel.org

On Mon, 2009-11-02 at 11:40 -0800, Dan Williams wrote: 
> On Sat, 2009-10-31 at 03:50 +0200, Maxim Levitsky wrote:
> > Sorry for cross-posting, but this issue really spans all three systems.
> > 
> > I anylized why I get 100% quality on all access points except currently
> > connected, when I used driver_nl80211 of wpa_supplcant.
> > 
> > First, when NetworkManager plans to switch to this driver?
> 
> Soon.  We've got some patches for this, but we'll also need tons of
> testing.  The WEXT stuff is pretty baked while nl80211 is still under
> some flux.  But of course the only way we bake nl80211 is by switching
> to it...
Glad to hear that!

I use driver_nl80211 and I have 3 problems.

1 - noise levels, this what I reported here.
2 - problems with wpa_supplicant that does no deauth on disconnect,
this can be easily worked around in supplicant code, but I feel some
disagreement between kernel and wpa_supplicant developers.

3 - seems that ibss is broken (even with all recently posted patches on
linux-wireless, wpa_supplicant still hangs while it attempts to create
the network. In fact I only tested creation of the network, not if
anybody could join it.


Speaking of joing the IBSS, NM did forever support the WPA encryption,
however, its isn't set in the beacon that is broadcast (in fact I only
see probe requests, not real beacons...)
Thus any attempt to join such network results in NM asking for wep keys.

It would be very nice to have WPA/2 IBSS, as this is the only real
encryption possible.


> 
> > For me it gives me faster association speeds, especially when I toggle
> > wireless with rfkill button.
> > 
> > 
> > this is what happens on the kernel side:
> > --> n80211 encodes only dBm data. It does so in, nl80211_send_bss.
> > 	(or it can encode unspecified data, which has little use...)
> > 
> > --> kernel also gives maximum ranges in, cfg80211_wext_giwrange, which is used by NM:
> > 	range->max_qual.updated = IW_QUAL_NOISE_INVALID | IW_QUAL_DBM | IW_QUAL_QUAL_UPDATED | IW_QUAL_LEVEL_UPDATED ;
> > 	range->max_qual.level = -110;
> > 	range->max_qual.qual = 70;
> > Thus it reports that it can't report noise.
> > 
> > 
> > driver_nl80211 side:
> > 
> > --> sends data as is, done in bss_info_handler:
> > 	r->level = nla_get_u32(bss[NL80211_BSS_SIGNAL_MBM]);
> > 	r->level /= 100; /* mBm to dBm */
> > 	r->flags |= WPA_SCAN_LEVEL_DBM | WPA_SCAN_QUAL_INVALID;
> > 
> > Again explicitly says that has no quality data, sends only dBm or unspecified data;
> > 
> > 
> > 
> > NM side:
> > 
> > --> three strategies used (in wireless_qual_to_percent)
> > 	--> quality: (used with driver_wext), not reported by nl80211
> > 
> > 	--> dbm: used only if:
> > 		* valid and zero max_qual->level (but now set to -110....) 
> 
> IIRC non-zero max_qual->level was the indication that the driver wanted
> to use RSSI, not dBm.  Since the real "max" dBm is around 0 (ie, that's
> the highest signal strength you'll ever really see), max_qual->level
> meant the driver was reporting signal strength in dBm.  max_qual->level
> == -110 is kinda wrong, because that's the _minimum_ level, not the max.
This is very good point, a question for kernel developers, why it was
set as such? How about setting it to ?


> The noise floor is almost always around -100 dBm so setting
> max_qual->level is pretty useless.
> 
> This is *exactly* why 'qual' is there: so that the driver itself can
> figure out what the hell it's signal level is, and so that NM doesn't
> have to go around assuming stuff.
It appears as kernel developers want to drop quality calculations
altogether, and report everything in dBms (right?) 
This is also a consistent I think.

> 
> For WEXT reporting, mac80211 should really be constructing a 'qual'
> instead of leaving it 0.  Then we don't have ambiguities with dBm, RSSI,
> unspec, etc.
It does, but driver_nl80211 doesn't use wext that is.


Also a recent conversation on linux-wireless indicated that noise level
is a property of the wireless card, so I think we won't see them in scan
results any time soon.


Best regards,
Maxim Levitsky