2021-03-08 08:48:19

by 邱名碩

[permalink] [raw]
Subject: Re: Questions for TX power get/set implementation of realtek WiFi

Pkshih <[email protected]> 於 2021年3月6日 週六 下午1:09寫道:
>
> On Tue, 2021-03-02 at 10:21 +0000, 邱名碩 wrote:
> > Hi guys,
> > I'm planning to implement the get_txpower hook function of the
> > struct ieee80211_ops and handle the IEEE80211_CONF_CHANGE_POWER for
> > rtl8xxxu driver. So that users can get the information of the current
> > tx power limit and set a new limit value with iw commands.
> >
> > I found there's phy_txpwr_idx_to_dbm() of each driver in
> > realtek/rtlwifi to translate the tx power index parsed from efuse to
> > tx power level in dBm. Does the power level translated represent the
> > current tx power limit in dBm? In my RTL8188CUS module, the power
> > index of CCK/OFDM from efuse for each channel is basically 36/38, thus
> > the translated txpower level is 11dBm. Is this the expected value for
> > the default txpower limit value? Or do I need some other formula to
> > compute?
>
> Basically, you can use a linear equation to translate power level in dBm to
> tx power index written to hardware registers.
> y = ax + b
> where,
> y: tx power index written to registers
> x: tx power level in dBm
> a: a constant, 2
> b: a constant induced by the value programmed in efuse that is used to
> adapt modules differences of RF characters.
>
Thanks so much. So it means that the txpwr_idx_to_dbm and dbm_to_idx
functions in rtlwifi family driver are basically correct. Except they
use the constant
b -7 for CCK and -8 for OFDM instead of the value in efuse. I should have
this value from efuse.

> >
> > And for the handler of IEEE80211_CONF_CHANGE_POWER, can I
> > translate the hw->conf.power_level with the phy_dbm_to_txpwr_idx
> > functions I found in some rtlwifi family drivers, then directly set
> > the txpower index with phy_set_txpower_index functions which write
> > power index to bbreg for different rates (CCK and OFDM)? Are there
> > some invalid values I should take care of because the TCP RX
> > performance seems to be extremely low when I write the power index
> > value translated from 20dBm to bbreg. Any suggestions for what I
> > should handle before writing the bbreg?
>
> The power level you're going to set must be less or equal to original
> one, because it's the hard limit of hardware capability. If the power
> level is too large, the receiver can't understand the signal. Then,
> they can't exchange data that leads low performance as you see.
> I think the performance would be okay if you connect to a near AP and
> power level 10dBm is adapted.
>

Thanks for clarification. So the power index from efuse is basically the
hard limit of the tx power for each channel (and rate)? That explains
my performance issue while I override it with higher value. Thanks

Chris
>
> ---
> Ping-Ke
>


2021-03-08 13:24:43

by Pkshih

[permalink] [raw]
Subject: Re: Questions for TX power get/set implementation of realtek WiFi

On Mon, 2021-03-08 at 08:46 +0000, 邱名碩 wrote:
> Pkshih <[email protected]> 於 2021年3月6日 週六 下午1:09寫道:
> >
> > On Tue, 2021-03-02 at 10:21 +0000, 邱名碩 wrote:
> > > Hi guys,
> > >     I'm planning to implement the get_txpower hook function of the
> > > struct ieee80211_ops and handle the IEEE80211_CONF_CHANGE_POWER for
> > > rtl8xxxu driver. So that users can get the information of the current
> > > tx power limit and set a new limit value with iw commands.
> > >
> > >     I found there's phy_txpwr_idx_to_dbm() of each driver in
> > > realtek/rtlwifi to translate the tx power index parsed from efuse to
> > > tx power level in dBm. Does the power level translated represent the
> > > current tx power limit in dBm? In my RTL8188CUS module, the power
> > > index of CCK/OFDM from efuse for each channel is basically 36/38, thus
> > > the translated txpower level is 11dBm. Is this the expected value for
> > > the default txpower limit value? Or do I need some other formula to
> > > compute?
> >
> > Basically, you can use a linear equation to translate power level in dBm to
> > tx power index written to hardware registers.
> >   y = ax + b
> > where,
> >   y: tx power index written to registers
> >   x: tx power level in dBm
> >   a: a constant, 2
> >   b: a constant induced by the value programmed in efuse that is used to
> >      adapt modules differences of RF characters.
> >
> Thanks so much. So it means that the txpwr_idx_to_dbm and dbm_to_idx
> functions in rtlwifi family driver are basically correct. Except they
> use the constant
> b -7 for CCK and -8 for OFDM instead of the value in efuse. I should have
> this value from efuse.

As I mentioned, the calibration value in efuse is used to output proper
power that every hardware module must be the same (with little tolerance).

I'm not sure why constant b isn't related to the value in efuse.
Maybe, you can assume the functions are correct, and configure the power
you desire.

>
> > >
> > >     And for the handler of IEEE80211_CONF_CHANGE_POWER, can I
> > > translate the hw->conf.power_level with the phy_dbm_to_txpwr_idx
> > > functions I found in some rtlwifi family drivers, then directly set
> > > the txpower index with phy_set_txpower_index functions which write
> > > power index to bbreg for different rates (CCK and OFDM)? Are there
> > > some invalid values I should take care of because the TCP RX
> > > performance seems to be extremely low when I write the power index
> > > value translated from 20dBm to bbreg. Any suggestions for what I
> > > should handle before writing the bbreg?
> >
> > The power level you're going to set must be less or equal to original
> > one, because it's the hard limit of hardware capability. If the power
> > level is too large, the receiver can't understand the signal. Then,
> > they can't exchange data that leads low performance as you see.
> > I think the performance would be okay if you connect to a near AP and
> > power level 10dBm is adapted.
> >
>
> Thanks for clarification. So the power index from efuse is basically the
> hard limit of the tx power for each channel (and rate)? That explains
> my performance issue while I override it with higher value. Thanks
>

The value in efuse isn't limit but calibration value of target power.
But, yes, you can seen it as hard limit as well.

--
Ping-Ke