On 11/2/07, Nick Kossifidis <mickflemm@xxxxxxxxx> wrote: > 2007/11/2, Luis R. Rodriguez <mcgrof@xxxxxxxxx>: > > On 11/2/07, Nick Kossifidis <mickflemm@xxxxxxxxx> wrote: > > > 2007/11/2, Nick Kossifidis <mickflemm@xxxxxxxxx>: > > > > 2007/11/2, Luis R. Rodriguez <mcgrof@xxxxxxxxx>: > > > > > Sorry, forgot this last hunk in my patch series... This applies on top > > > > > of my series. > > > > > > > > > > This sets the ACK bitrate to the lower rates. Without this I get > > > > > about 70% packet loss when using the 11M rate. Not sure exactly what rates > > > > > this is setting the HW to send the ACKs in but it sure does help. > > > > > > > > > > > > > 6Mbits for OFDM (have checked it out with various sniffers), if not > > > > set i usualy get 24Mbits for acks when transmiting on 54Mbit. > > > > > > > > Haven't checked it for CCK but i guess it's 1 or 2 Mbits, have a look > > > > at my comments inside reg.h... > > > > > > > > #define AR5K_STA_ID1_ACKCTS_6MB 0x01000000 /* Use 6Mbit/s > > > > for ACK/CTS (?) */ > > > > #define AR5K_STA_ID1_BASE_RATE_11B 0x02000000 /* Use 11b > > > > base rate (for ACK/CTS ?) [5211+] */ > > > > > > > > > > Also note that setting these bits is safe and cause a more stable > > > behaviour. If these bits are not set then control_rate is being used > > > > Why do you say that? How do you know? > > > > Using sniffers/analyzers with madwifi i can see that either > control_rate is being used or base rate (i guess HAL chooses if it > will use control_rate or base rate). I know for sure that if i > explicitly set these bits to use lower rates with madwifi (which also > has this function that writes directly on the register) i only get > 6Mbits for ack frames and if i don't i usualy get 24Mbits > (control_rate) but nothing else (that's because i'm transmitting @ 24 > - 54Mbits). Excellent, we just need to confirm this behavior with ath5k, but of course we're still having problems with higher rates though.. so this may take a bit. > > > (check out rate tables -remember these tables are hardcoded in hal, > > > that's why they have put control_rate there, to let us know what rate > > > is being used for ack/cts frames by hw-). > > > > > > Here it is... > > > (transmision rate) -> (ack/cts rate) > > > > > > OFDM > > > 6 - 12 -> 6 > > > 12 - 24 -> 12 > > > 24 - 54 -> 24 > > > > > > CCK > > > 1 - 2 -> 1 > > > 2 - 11 -> 2 > > > > This is not a concoction by the HAL or driver, this is what the spec > > yields. Let me quote the relevant section, which is in tx.c on > > mac80211: > > > > IEEE 802.11, 9.6: > > - control response frame (CTS or ACK) shall be transmitted using the > > same rate as the immediately previous frame in the frame exchange > > sequence, if this rate belongs to the PHY mandatory rates, or else > > at the highest possible rate belonging to the PHY rates in the > > BSSBasicRateSet > > > > I don't see how this is relevant, as i told you i can only see ack > frames on the above rates (6,12,24) with madwifi regardless of the > previously transmited frame's rate, Its relevant as this is the rules followed to determine what rate to TX back ACKs at. Essentially what you pasted are what the ACK rates for each respective TX rate. > also BSSBasicRateSet is nothing > standard, i checked out 802.11-1999 and here is what it says... > > "The set of data rates (in units of 500 kbit/s) that must > be supported by all STAs to join this BSS. The STA > that is creating the BSS must be able to receive and > transmit at each of the data rates listed in the set." > > so what we get (6,12,24 for acks) is not defined anywhere, so i guess > is hw-related. Notice how it mentions mandatory rates though. Mandatory rates for IEEE 802.11g PHY are 1, 2, 5.5, 11, 6, 12, 24 Mbps. I believe that's where the 1, 2, 6, 12 and 24 come from. > > Also I think it might be possible to change the ACK rate per used TX > > rate. I am not sure where this can be set though. In my latest > > approach I just removed the control_rate all together as we should be > > following the spec. I left the control rate though just to gradually > > shift in the right direction and as I'm not yet 100% sure this is just > > for the ACK timeout. > > > > control_rate isn't used anywhere in ath5k (like the rest of rate > tables), Well it is -- its used to compute the value of the rate duration registers. I'll do some more tests and confirm the value of the ACKs on ath5k currently by setting this low bit rate for ACKs or not. Depending on these results we should change the current computation. > i'm not sure if rate control algorithms on madwifi use this. Bleh. > As i've said before we should remove that led blinking stuff and get > rid of them Sure, I'll be working on that next as well as removing the rate macros. > (btw shouldn't we register gpio led with led class ?) Haven't looked at the led stuff yet, but feel free! > > > So it still doesn't make sense that you had 70% loss i mean any rate > > > between 2 and 11Mbits would cause the same ack rate (2Mbits), even if > > > you didn't set that bit. > > > > Yeah, it seemed strange to me too. BTW I was a bit far away from the access point, I just moved and get 0% packet loss now without this patch. It may have been environmental noise.. not sure what the problem was.. Luis - To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
