Hi Stefan,
* Stefan Sperling wrote:
> This patch adds A-MSDU rx offloading support for both iwm(4) and iwx(4)
> and re-enables net80211's software A-MSDU Rx support for all 11n drivers.
>
> Meaning iwn(4) and athn(4) will also be receiving A-MSDUs again.
> This feature has been turned off since July 2019:
> https://cvsweb.openbsd.org/cgi-bin/cvsweb/src/sys/net80211/ieee80211_input.c#rev1.207
> The root cause of the problem I saw at that time was related to dlg's Rx
> queue back-pressure mechanism. Once we figured this out, all wireless drivers
> were fixed to call if_input() only once per interrupt so this is no longer
> an issue.
>
> For a very brief period we tried to enable A-MSDUs again in -current but
> we found out that iwm/iwx needed additional work for the new devices which
> received support while A-MSDU was disabled in-tree:
> https://cvsweb.openbsd.org/cgi-bin/cvsweb/src/sys/net80211/ieee80211_input.c#rev1.226
> The patch below completes the missing bits to make A-MSDUs work on these
> new devices.
I have a kernel with your patch running since several hours and
noticed a regression. My usual "test case" is copying several large
files from my file server via NFSv3 to my laptop. In the beginning the
transfer rate was about 2-3M/s and after some time it dropped to around
50-300K/s and never recovered (transfer is now running for 2.5h).
I have the following device in a Thinkpad T450s connect to a Fritzbox
AP.
iwm0 at pci2 dev 0 function 0 "Intel Dual Band Wireless-AC 8265" rev 0x78, msi
iwm0: hw rev 0x230, fw ver 34.0.1, address
Cheers
Matthias
