Thomas Graf [off-list ref] writes:

I tracked down an issue with using the Kernel 4.9.277 and 5.4.131 back
to this patch.

In 802.11bg mode the effective transmit rate drops to 3MBit/s.
With that patch and perfect SNR (> 45dB) I get and rc_stats:

*********************
best   __________rate_________    ________statistics________
________last_______    ______sum-of________
rate  [name idx airtime max_tp]  [avg(tp) avg(prob) sd(prob)]
[prob.|retry|suc|att]  [#success | #attempts]
        1     0    9738    0.9       0.9      93.7      0.5      75.0
1     3 4           562   572
        2     1    4922    1.6       1.6      97.0      1.5     100.0
1     2 2           559   576
        5.5   2    1858    4.7       4.7      99.9      0.0     100.0
2     2 2           542   565
       11     3     982    9.1       9.1      94.9      1.2     100.0
4     3 3           560   578
        6     4    1648    5.3       5.3      95.3      1.3     100.0
3     3 3           538   595
        9     5    1112    8.0       8.0      97.5      1.8     100.0
4     2 2           560   596
       12     6     844   10.5      10.5      92.4      2.0      75.0
5     3 4           572   615
       18     7     576   15.5      15.4      89.0      1.4      66.6
5     2 3           559   609
   D   24     8     440   20.4      20.4      99.4      1.1     100.0
6     2 2           560   604
  C    36     9     308   29.1      29.1      98.7      1.6     100.0
6     2 2           565   600
 B     48    10     240   37.3      36.3      87.3      1.4      66.6
6     2 3           608   671
A   P  54    11     216   41.6      41.6      97.3      1.7     100.0
6     3 3           565   620

Total packet count::    ideal 6111      lookaround 652
*********************

After reverting that patch the transmit rate is back to 20MBit/s:
*********************
best   __________rate_________    ________statistics________
________last_______    ______sum-of________
rate  [name idx airtime max_tp]  [avg(tp) avg(prob) sd(prob)]
[prob.|retry|suc|att]  [#success | #attempts]
        1     0    9738    0.9       0.9      99.8      0.6     100.0
1     0 0            41   44
        2     1    4922    1.6       1.6     100.0      0.0     100.0
1     0 0            45   45
        5.5   2    1858    4.7       4.7      99.9      0.0     100.0
2     0 0            43   44
       11     3     982    9.1       9.1      97.0      1.4     100.0
4     0 0            44   45
        6     4    1648    5.3       5.3      99.9      0.0     100.0
3     0 0            43   44
        9     5    1112    8.0       8.0      99.9      0.0     100.0
4     0 0            45   47
       12     6     844   10.5       8.8      74.9      0.7       0.0
5     0 0            43   47
       18     7     576   15.5      15.5     100.0      0.0     100.0
5     0 0            42   42
   D   24     8     440   20.4      20.4      99.9      0.0     100.0
6     0 0            60   62
  C P  36     9     308   29.1      29.1      99.9      0.0     100.0
6     0 0            44   46
A      48    10     240   37.3      37.3      93.8      1.2     100.0
6     1 1           587   618
 B     54    11     216   41.6      32.3      69.8      2.1      33.3
6     0 0         24506   25849

Total packet count::    ideal 4962      lookaround 552
*********************

I'm using the ath9k driver with a sparklan WPEA-121N card.

For my case reverting that patch is enough.
What I see is that IEEE80211_TX_CTL_RATE_CTRL_PROBE is set but not
used anywhere in the code. Maybe there is an easy and better fix than
just reverting.

I lost track, is this regression resolved now or is it still open?

-- 
https://patchwork.kernel.org/project/linux-wireless/list/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches