Trouble shooting low rate MCS9 in 802.11ac

[Vu Hai NGUYEN]

>Strange that disabling checksum helped things.  Maybe your ethernet driver
>has issues?

I've no idea about this. Do you have any suggest how to check it? :D

>Looks like removing netfilter (and maybe ebtables?) would
>also help your performance...

I don't enable ebtables in my kernel config, so it shoud not be built in my kernel

>tx status is needed for bookkeping at least, even if you don't
>get useful status info in stock firmware.
>
>CT firmware will give you proper status if you apply some of the
>ath10k driver patches from my tree, but that is unlikely to help
>performance.

>This accounts for 76.58% of CPU usage. You've most likely trimmed down
>the remaining entries accounting for 24% before posting.

I saw there is option -a when running perf top but I did it and still, it doesn't 
print all processes :(

>By the looks of it you might be able to strip down netfilter
>(iptables) rules to none on all chains and tables if you just want to
>max out the performance at the cost of having no firewall whatsoever.

I'd cleaned all rules of iptables and removed all modules of iptables from kernel by command modprobe -r
But there's still some process of netfilter process running. I disabled netfilter on bridge interface by writing 0 to these files:
/proc/sys/net/bridge/bridge-nf-call-arptables
/proc/sys/net/bridge/bridge-nf-call-iptables
/proc/sys/net/bridge/bridge-nf-call-ip6tables

and the performance increases from 500 to 550 Mbps (iperf in tcp reports between 530 ~ 570Mbps). 
There is no more process like  "br_nf_pre_routing" and "br_nf_forward_ip".
Do you think I can do anything more to strip down netfilter?
Now the processor of my AP (about 20% idle)  is no more exhausted but not the same on STA (about 4.5 % idle)
Here is my perf top result on AP:

PerfTop:      99 irqs/sec  kernel:96.0%  exact:  0.0% [1000Hz cycles],  (all, 1 CPU)
-------------------------------------------------------------------------------

    24.40%  [kernel]       [k] armadaxp_enter_idle
     6.04%  [mac80211]     [k] ieee80211_tx_status
     5.07%  [kernel]       [k] memcpy
     3.38%  [kernel]       [k] mv_eth_poll
     3.38%  [mac80211]     [k] $a
     2.66%  [kernel]       [k] __memzero
     2.54%  [ath10k_pci]   [k] ath10k_ce_send_nolock
     2.17%  [ath10k_core]  [k] ath10k_txrx_tx_unref
     2.05%  [kernel]       [k] dev_queue_xmit
     2.05%  [kernel]       [k] br_handle_frame
     1.69%  [kernel]       [k] local_bh_enable
     1.69%  [kernel]       [k] eth_type_trans
     1.69%  [ath10k_core]  [k] ath10k_htt_tx
     1.69%  ld-2.13.so     [.] 0xf120  
     1.69%  [ath10k_pci]   [k] ath10k_ce_completed_send_next
     1.57%  [kernel]       [k] __netif_receive_skb
     1.45%  [kernel]       [k] dev_hard_start_xmit
     1.45%  [kernel]       [k] nf_iterate
     1.33%  [kernel]       [k] dev_gro_receive                                  
     1.21%  [kernel]       [k] kfree                                            
     1.21%  [kernel]       [k] __kfree_skb 

And on STA:

PerfTop:      51 irqs/sec  kernel:94.1%  exact:  0.0% [1000Hz cycles],  (all, 1 CPU)
-------------------------------------------------------------------------------

     7.51%  [kernel]       [k] memcpy
     6.65%  [ath10k_core]  [k] ath10k_htt_rx_amsdu_pop
     5.49%  [kernel]       [k] mv_eth_tx
     4.91%  [kernel]       [k] kfree
     4.91%  [kernel]       [k] __memzero
     4.05%  [kernel]       [k] armadaxp_enter_idle
     4.05%  [kernel]       [k] __kmalloc_track_caller
     4.05%  [kernel]       [k] dev_queue_xmit
     4.05%  [kernel]       [k] ip_rcv
     3.76%  [mac80211]     [k] ieee80211_rx
     3.76%  [kernel]       [k] __kfree_skb
     3.76%  [ath10k_core]  [k] $a
     2.89%  [mac80211]     [k] ieee80211_rx_handlers
     2.60%  [mac80211]     [k] $a
     2.60%  [kernel]       [k] ip_forward
     2.60%  [cfg80211]     [k] ieee80211_data_to_8023
     2.60%  [ath10k_pci]   [k] ath10k_ce_completed_send_next
     2.31%  ld-2.13.so     [.] 0x9d8c  
     2.31%  [cfg80211]     [k] ieee80211_hdrlen
     2.02%  [mac80211]     [k] $a

>You also might want to change qdisc (`tc` command) on all interfaces
>involved to pfifo in case you're using codel to further reduce CPU
>load at the cost of buffer bloating the system.

I installed tc from package iproute2 and did : tc qdisc but it doesn't improve the performance :D

NGUYEN Vu Hai
Acita-Sodielec
Route de Mayres - B.P. 9
12100 St GEORGES DE LUZENCON
FRANCE