On Mon, Mar 09, 2026 at 02:55:30PM +0100, Stefan Sperling wrote:
> 
> iwx inherited code from iwm which handles reordering of received A-MPDU
> subframes at the driver level. In Linux iwlwifi this code was added for
> 9k devices and multi-queue Rx operation.
> Our drivers only use a single Rx queue, but keeping the code close enough
> to Linux seemed reasonable the time when 9k devices were new, and it didn't
> hurt.
> 
> This code made its way into iwx because iwx started out as a copy of iwm.
> But it turns out AX200 and later devices perform Rx reordering in firmware
> already and set flags on Rx descriptors which the driver can use to check
> for duplicate frames which can be discarded.
> 
> I have recently become aware of two related changes the iwlwifi developers
> made in 2023:
> 
> 1) Multi-queue Rx has been disabled in Linux on iwm 9000 devices.
> (linux commit 29fa9a984b6d1075020f12071a89897fd62ed27f)
> This was done to allow for change number 2:
> 
> 2) The Rx-reordering code has been removed from iwiwifi, because it was
> only needed for 9k devices which do not provide the needed information in
> Rx descriptors. And the code was found to occasionally drop legit frames.
> (linux commit ff8e3a40d78bc414213b2724ad775adf98780a5a)
> 
> I am not sure to which degree the dropped legit frames are affecting
> our implementation of this in iwm. It seems to be working fine for us?
> Maybe we should look at potential packet loss on iwm 9k devices at some
> point, even though I do not recall any complaints about such an issue.
> 
> In any case, this means the Rx reordering code in iwx is pointless.
> We can rely on Rx descriptor flags instead.
> 
> Testing on any iwx devices would be appreciated. I have tested this diff
> on AX200 and BZ with no visible change in behaviour.

No regression on iwx ax210, thanks.

> ok?

ok kevlo@