Re: [PATCH] ath9k: fix per-packet TX-power cap for TPC

[Jonas Jelonek <jelonek.jonas@gmail.com>]

> On 30. Mar 2023, at 11:31, Toke Høiland-Jørgensen <toke@kernel.org> wrote:
> 
> Jonas Jelonek <jelonek.jonas@gmail.com> writes:
> 
>> Fix incorrect usage of plain rate_idx as index into the max (power) per
>> rate lookup table.
>> 
>> For transmit power control (TPC), the ath9k driver maintains internal
>> tables (in struct ath_hw) to store the max allowed power level per rate.
>> They are used to limit a given TX-power according to regulatory and user
>> limits in the TX-path per packet. The tables are filled in a predefined
>> order, starting with values for CCK + OFDM rates and followed by the
>> values for MCS rates. Thus, the maximum power levels for MCS do not
>> start at index 0 in the table but are shifted by a fixed value.
>> 
>> The TX-power limiting in ath_get_rate_txpower did not apply this shift,
>> thus retrieved the incorrect maximum power level. For example, the
>> maximum power for OFDM rate 0 was used for MCS rate 0. If STBC was used,
>> the power was mostly limited to 0 because the STBC table is zeroed for
>> legacy CCK/OFDM rates. This patch fixes this table lookup.
>> 
>> Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
> 
> So what effect does this bug have in practice? Also, how did you test
> the patch? :)
> 

It actually may not have an effect in most of current practical applications. The
member ‘tpc_enabled' in struct ath_hw is by default set to false and thus, TPC
in ath9k is usually disabled. The code path will be skipped in that case. 
But it has an effect as I am still working on TPC per packet in ath9k as part of
research. Looking at my traces I saw that the TX-power is capped at 0 for
MCS rates 0-7 in case the STBC flag is set in struct ieee80211_tx_info. Thus,
transmission was significantly worse as usual, and I also was not able to 
manually set a proper TX-power via my testing API. With other rates it is working
fine. I also double-checked that it isn’t a problem of how TX-power is set and
reported with our upcoming API.

I tested this with my OpenWrt-based AP-STA desk setup using two different
ath9k wifi chips (AR9280, AR9580), Kernel 5.15.98. I guess that should be
sufficient as ath9k hasn’t changed that significantly since that and several
changes are already backported. I compile-tested it with latest 6.3 kernel
(+allyesconfig), the patch applied flawlessly and encountered no problems. 
After applying my patch, I could see that everything was working as expected.
Transmission performs equally as if TPC was disabled, and setting + reporting
TX-power was working again. Also verified this with some verbose debugging
in ath_get_rate_txpower to see which index + max power is used.

Jonas