Re: rt2x00: Ralink RT5572 very high peak current consumption

[Stanislaw Gruszka <sgruszka@redhat.com>]

On Mon, May 26, 2014 at 03:37:36PM +0000, Matthias Fend wrote:
> Hello,
> 
> Here are some updates concerning my current issues.
> 
> > -----Ursprüngliche Nachricht-----
> > Von: linux-wireless-owner@vger.kernel.org [mailto:linux-wireless-
> > owner@vger.kernel.org] Im Auftrag von Matthias Fend
> > Gesendet: Freitag, 23. Mai 2014 14:25
> > An: linux-wireless@vger.kernel.org
> > Betreff: rt2x00: Ralink RT5572 very high peak current consumption
> > 
> > Hello Ralink experts,
> > 
> > I recognized spurious problems with Ralink RT5572 USB-dongles when using
> > this dongles as access point in 5G band.
> > The problems may appear as short usb disconnect-connect events or failed
> > vendor request messages at different offsets with different error codes or
> > will not appear for days ;) During error research I noticed that, most likely
> > during the beacon send time (100ms interval), there are very high peak
> > currents in the supply of the USB-dongle.
> > 
> > I tried it with two different dongles, the base current for all three is in the
> > range of  ~200mA:
> > Buffalo WI-U2-300D: Peak ~1A
> > TP-Link TL-WDN3200: Peak ~1.2mA
> > 
> > We could reproduce this issue on two different machines (x86 and arm), did
> > test with kernel versions 3.10.19, 3.13.0 and 3.14.4 tried hostapd-2.0 and
> > hostapd-2.1 and loaded chip firmware version V0.29, V0.33 and V0.35.
> > 
> > As a comparison I also created an 5G access point setup with a windows7
> > machine - all with the latest drivers from the manufacturer webpage.
> > Buffalo WI-U2-300D: Peak 330mA
> > TP-Link TL-WDN3200: Peak 330mA
> > Which seems more legit than the values from the linux setups.
> > 
> > The access point functionality itself worked with every tested combination,
> > but there is definitively something wrong.
> > I also started to grab the usb communication of the windows driver to see if
> > there is an obvious difference in some of the written registers.
> > But this kind of reverse engineering is not as easy since the windows driver
> > does not write the same registers as the linux drivers but write some others -
> > and of course I have no datasheet/register description.
> > 
> > Until now the only interesting thing I found out is that the linux driver writes
> > 0x00000082 to the TXOP_HLDR_ET (0x1608) register whereas the windows
> > driver uses a value of 0x0000000a. Changing this register during runtime in a
> > linux system reduces the current peaks to 50 percent of their previous value.
> > But there is still something else wrong.
> > 
> > So, in my despair I try to get help in any form (hints, datasheets/manuals,
> > ideas, experience, ...).
> 
> I found out that the discovered high current peaks are somehow related to invalidated beacons.
> Usually I can see 1 block with a normal tx-condition current followed by 7 additional blocks with high current peaks.
> For a test I copied the beacon data from HW_BEACON_BASE0 to HW_BEACON_BASE2 and then the first and the third beacon block have normal current values while the others still generate high current peaks
> 
> It seems that setting all five words of the TXWI at the beacon base does not complete disable the generation of this beacon.
> If I modify the TXIW_W0 of the disabled beacons to 0x40000000 (set TXWI_W0_PHYMODE as in the really used beacon) then the current consumption does not exceed the usual value. But it seems that there is still a short activity for this beacon which, in my opinion, should not be the case.
> 
> So, currently I'm wondering about two things:
> -)How do I completely turn off unused beacons (from the measurements with the windows setup I know that this should be somehow possible)?

Perhaps this could be MULTI_BCN_NUM value in MAC_BSSID_DW1 register.
Try below patch. If it would help we will need to change way we
clear/setup beacons on multiple BSSes, because seems that HW
send MULTI_BCN_NUM beacons from first registers. 

Stanislaw

diff --git a/drivers/net/wireless/rt2x00/rt2800lib.c b/drivers/net/wireless/rt2x00/rt2800lib.c
index c17fcf2..2e799d5 100644
--- a/drivers/net/wireless/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/rt2x00/rt2800lib.c
@@ -1556,7 +1556,7 @@ void rt2800_config_intf(struct rt2x00_dev *rt2x00dev, struct rt2x00_intf *intf,
 		if (!is_zero_ether_addr((const u8 *)conf->bssid)) {
 			reg = le32_to_cpu(conf->bssid[1]);
 			rt2x00_set_field32(&reg, MAC_BSSID_DW1_BSS_ID_MASK, 3);
-			rt2x00_set_field32(&reg, MAC_BSSID_DW1_BSS_BCN_NUM, 7);
+			rt2x00_set_field32(&reg, MAC_BSSID_DW1_BSS_BCN_NUM, 0);
 			conf->bssid[1] = cpu_to_le32(reg);
 		}