Date: Fri, 12 Sep 2014 11:40:38 -0700 From: Adrian Chadd <adrian@freebsd.org> To: Nathan Whitehorn <nwhitehorn@freebsd.org> Cc: "freebsd-wireless@freebsd.org" <freebsd-wireless@freebsd.org> Subject: Re: Issues with urtwn Message-ID: <CAJ-Vmo=Ox_eAstz9tAcDQ7sBLfaAsJZvn7j6z7Nx3BtSjbhyAw@mail.gmail.com> In-Reply-To: <541333ED.8020005@freebsd.org> References: <540C751F.6050202@freebsd.org> <CAJ-VmokyPcS077wHiP4Mdetms=meqk47v29fKA1edidhorVQpg@mail.gmail.com> <540C92D6.4030106@freebsd.org> <CAJ-VmomMwJOSz7hyAfeEgPE=qBfYm7fTOo5km8JJk4g62JxTkg@mail.gmail.com> <540CC53A.90600@freebsd.org> <CAJ-Vmokt_kgxW3aPEDcNwg_ZVrCotqF_tOP1YjZCtO=nCZ8z5Q@mail.gmail.com> <540E2A2D.4090301@freebsd.org> <CAJ-VmongioOZQS561Qq5S1T0UVnBifxrQf8P0rr8jEWzk=dumQ@mail.gmail.com> <541333ED.8020005@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
What the hell is calling sta power save? Can you go put in a stack trace (maybe use dtrace :) whenever the sta powersave routines get called? That's odd. So you see TIM=1 from the AP, but since powersave isn't enabled, it doesn't transition the NIC back to normal. So ieee80211_sta_pwrsave() is set in vap->iv_sta_ps. That's only called in: * the scan task for bgscan; * transitioning in/out of IEEE80211_S_SLEEP state. So, I don't know why you're seeing the above. In fact, if you hacked up urtwn to at least do powersave, the tim_notify will transition the VAP from SLEEP to RUN. In fact, hm. Can you file a bug for the above? There's the bug you're seeing, then there's the "we see a TIM flag for us, maybe we need to force transition out of sleep even if we think we're not in powersave". Thanks, -a On 12 September 2014 10:57, Nathan Whitehorn <nwhitehorn@freebsd.org> wrote: > This is what the debug output looks like when things go wrong: > wlan0: [54:78:1a:a0:91:22] sta power save mode on > wlan0: wlan0: [54:78:1a:a0:91:22] save frame with age 41, 1 now queued > [54:78:1a:a0:91:22] sta power save mode off > wlan0: [54:78:1a:a0:91:22] flush ps queue, 1 packets queued > wlan0: [54:78:1a:a0:91:22] sta power save mode on > wlan0: [54:78:1a:a0:91:22] save frame with age 41, 1 now queued > wlan0: [54:78:1a:a0:91:22] save frame with age 0, 2 now queued > wlan0: [54:78:1a:a0:91:22] save frame with age 0, 3 now queued > wlan0: [54:78:1a:a0:91:22] save frame with age 0, 4 now queued > wlan0: ieee80211_sta_tim_notify: TIM=1 > wlan0: ieee80211_sta_tim_notify: TIM=1 > wlan0: ieee80211_sta_tim_notify: TIM=1 > wlan0: ieee80211_sta_tim_notify: TIM=1 > wlan0: ieee80211_sta_tim_notify: TIM=1 > wlan0: [54:78:1a:a0:91:22] save frame with age 0, 5 now queued > wlan0: [54:78:1a:a0:91:22] save frame with age 0, 6 now queued > > Let me know if I can test anything else. > -Nathan > > > On 09/08/14 15:17, Adrian Chadd wrote: >> >> Please compile your kernel with IEEE80211_DEBUG, then enable debugging >> - wlandebug +state +power >> >> You can disable powersave with 'ifconfig wlan0 -powersave', but it >> shouldn't be enabled by default. >> >> >> >> -a >> >> >> On 8 September 2014 15:14, Nathan Whitehorn <nwhitehorn@freebsd.org> >> wrote: >>> >>> So it's definitely to do with powersave. Here's a bunch of iterations of >>> ifconfig list sta on my laptop: >>> ADDR AID CHAN RATE RSSI IDLE TXSEQ RXSEQ CAPS FLAG >>> 54:78:1a:a0:91:22 149 1 54M 37.0 0 4385 37104 EPS A >>> HTCAP >>> RSN WME >>> ADDR AID CHAN RATE RSSI IDLE TXSEQ RXSEQ CAPS FLAG >>> 54:78:1a:a0:91:22 149 1 54M 37.5 0 4412 39360 EPS A >>> HTCAP >>> RSN WME >>> ADDR AID CHAN RATE RSSI IDLE TXSEQ RXSEQ CAPS FLAG >>> 54:78:1a:a0:91:22 149 1 54M 37.5 0 4417 39360 EPS AP >>> HTCAP >>> RSN WME >>> ADDR AID CHAN RATE RSSI IDLE TXSEQ RXSEQ CAPS FLAG >>> 54:78:1a:a0:91:22 149 1 54M 37.5 0 4417 39360 EPS AP >>> HTCAP >>> RSN WME >>> ADDR AID CHAN RATE RSSI IDLE TXSEQ RXSEQ CAPS FLAG >>> 54:78:1a:a0:91:22 149 1 54M 37.5 0 4417 39360 EPS AP >>> HTCAP >>> RSN WME >>> >>> You can see the connection die on the third line, when the txseq and >>> rxseq >>> counters stop incrementing and 'P' gets added to the FLAG field. Does >>> this >>> mean the AP has turned on powersave on its end? >>> -Nathan >>> >>> >>> On 09/07/14 14:07, Adrian Chadd wrote: >>>> >>>> Hi, >>>> >>>> The way it's supposed to work in the legacy 802.11 powersave world is >>>> that you send a/any data frame with the powermgt bit in the 802.11 >>>> header set to 0 and the AP goes "oh they're awake!" and sends you your >>>> buffered frames. >>>> >>>> By default powersave isn't enabled, so we should never be _telling_ >>>> the AP that we're going to sleep and the stack always sends data >>>> frames with pwrmgt=0. >>>> >>>> You can ensure it's disabled by ifconfig wlan0 -powersave >>>> >>>> The code in -HEAD that manages that is in ieee80211_power.c. I added >>>> an explicit powersave support mode for NICs that need it done for them >>>> - and the only one it's enabled for right now is ath(4). >>>> >>>> The only reason net80211 sends pwrmgt changes outside of having >>>> net80211 power save enabled is the background scan code. >>>> >>>> I'd compile in IEEE80211_DEBUG in your kernel, then I'd use wlandebug >>>> +scan to see if somehow there's some scanning going on; and wlandebug >>>> +power to see if any power save transitions occur. >>>> >>>> Are you absolutely sure it's a receive side buffering problem, rather >>>> than a send side problem? >>>> >>>> It's also possible that the NIC stops receiving and the AP treats that >>>> as "oh ok, they've gone to sleep for a while." ath(4) now does this in >>>> hostap mode. >>>> >>>> >>>> -a >>>> >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJ-Vmo=Ox_eAstz9tAcDQ7sBLfaAsJZvn7j6z7Nx3BtSjbhyAw>