2024-03-08 12:20:14

by Johannes Berg

[permalink] [raw]
Subject: Re: regression: [PATCH] wifi: mac80211: set station RX-NSS on reconfig

On Fri, 2024-03-08 at 12:58 +0100, Andreas Tobler wrote:
> Hello Johannes, all.
>
> with this commit[0] we encounter a performance regression in one of our
> AP/STA tests. The performance drop is about 100Mbps+
>
> Our setup is a hostap (2.10) configured AP on 802.11ac, channel 46.
> (40MHz). The client connects via wpa (2.10) and then we produce traffic.
> Before this commit, the throughput was around 250 Mbps, with this commit
> the performance is now down to 135 Mbps.
>
> The wifi cards we use are WLE600VX from compex, a ath10k based one. The
> kernel we're on is 6.1.80. We've seen it on 6.6 too but didn't investigate.
>
> Now, the ieee80211_sta_set_rx_nss() function has two FIXME comments. One
> in where it handles the ht capability and one in the vht part. Both
> FIXME's are: 'consider rx_highest?'
>

I think you'll find that all that is unrelated, and this fixes it:

https://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next.git/commit/?id=a8bca3e9371dc5e276af4168be099b2a05554c2a

johannes


2024-03-08 14:22:04

by Andreas Tobler

[permalink] [raw]
Subject: Re: regression: [PATCH] wifi: mac80211: set station RX-NSS on reconfig



On 08.03.2024 13:19, Johannes Berg wrote:
> On Fri, 2024-03-08 at 12:58 +0100, Andreas Tobler wrote:
>> Hello Johannes, all.
>>
>> with this commit[0] we encounter a performance regression in one of our
>> AP/STA tests. The performance drop is about 100Mbps+
>>
>> Our setup is a hostap (2.10) configured AP on 802.11ac, channel 46.
>> (40MHz). The client connects via wpa (2.10) and then we produce traffic.
>> Before this commit, the throughput was around 250 Mbps, with this commit
>> the performance is now down to 135 Mbps.
>>
>> The wifi cards we use are WLE600VX from compex, a ath10k based one. The
>> kernel we're on is 6.1.80. We've seen it on 6.6 too but didn't investigate.
>>
>> Now, the ieee80211_sta_set_rx_nss() function has two FIXME comments. One
>> in where it handles the ht capability and one in the vht part. Both
>> FIXME's are: 'consider rx_highest?'
>>
>
> I think you'll find that all that is unrelated, and this fixes it:

[link dropped since outlook garbles it completely]

Amazing! I can confirm that my performance drop has gone.
Thank you very much!
Andreas


Attachments:
smime.p7s (4.65 kB)
S/MIME Cryptographic Signature