2008/9/23 Brian Prodoehl <[EMAIL PROTECTED]>: > On Mon, Sep 22, 2008 at 5:04 PM, Nick Kossifidis <[EMAIL PROTECTED]> wrote: >> 2008/9/22 Brian Prodoehl <[EMAIL PROTECTED]>: >>>> Ath5k is much slower then madwifi. Best speed I could get with iwconfig >>>> was >>>> 24Mb, Higher speeds did not respond. Same location, same computer, same >>>> AP with >>>> madwifi I get the full 54Mb with no drops. This will be a blocker for >>>> eeepc >>>> users and distros with the 2.6.27 kernels.... >>> >>> I'm chasing down this bug as well. I am almost certain that in my >>> case it is because the txpower is not set properly. I am seeing >>> between +28dBm and +31dBm out of my cards, which is well above where >>> they perform well. The signal is definitely getting clipped, and I'm >>> actually surprised that the rates up to 18Mbps work as well as they do >>> given how badly its clipping. >>> >>> I have a patch which allows iwconfig to actually dictate the txpower >>> setting used by ath5k (currently there's a disconnect, so what >>> iwconfig says never actually gets picked up and used by the ath5k >>> txpower routines), and it sets the AR5K_PHY_TXPOWER_RATEn registers >>> the same as madwifi (comparing register dumps at runtime), but I still >>> have the 1W ugly output regardless of what level I tell it. >>> AR5K_PHY_TXPOWER_RATE_MAX is the same as madwifi (0x003f) and all the >>> AR5K_PHY_PCDAC_TXPOWER registers are the same as madwifi (all 0's). >>> But, still the ugly output, so clearly I'm missing something. I'm >>> getting back to basics with madwifi-trace, so hopefully I find the >>> missing piece. >>> >>> Does anyone have any progress towards working transmit power control? >>> If I actually get anywhere I'll post the patch. >> >> To implement tx power support correctly we need to have information on >> the calibration data from the various EEPROM versions, check out >> ath_info for some work on this (more stuff on the way). Have in mind >> that both AR5K_PHY_TXPOWER_RATE, TPC and tx descriptor use an index >> value, not a dbm value. As far as i know they contain an index on the >> baseband PCDAC_TXPOWER table. Table has 64 entries so the mask is 3f >> for the index value, also 3f is the mask for half-dbm value (64/2 = >> 31.5dbm max) used in PCDAC_TXPOWER (that's in case you think it's the >> same thing). The tricky part is to have an implementation that works >> on all hw, using EEPROM information from various eeprom versions (and >> some information from registers on newer chips) and calibrating >> PCDAC_TXPOWER propertly. >> >> >> -- >> GPG ID: 0xD21DB2DB >> As you read this post global entropy rises. Have Fun ;-) >> Nick >> > > For verification's sake, the attached patch loads a hard-coded > PCDAC_TXPOWER table for some AR5414 card (taken from one of Bruno's > madwifi-trace captures). This gives me clean linear output power > between 10dBm and 24dBm (set with iwconfig) on my Ubiquiti XR2s, and > it lets me operate at bitrates above 24Mbps (all the way up to 54Mbps) > for the first time with ath5k. TCP throughput is consistently above > 20Mbps. > > Hopefully in the next couple days I can post a patch that loads the > PCDAC_TXPOWER table from the EEPROM, but almost all of the hardware I > have uses the AR5414. > > The patch is against the compat-wireless-old 20080912 snapshot. >
I have tried this solution before and it's not a nice one. We need to rev. engineer EEPROM and use calibration data etc. A static pcdac table is only a temporary solution for some chips, not all of them. We 'll work this out soon i hope, let's focus on ANI for now so we are ok on the rx path... -- GPG ID: 0xD21DB2DB As you read this post global entropy rises. Have Fun ;-) Nick _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
