Re: [PATCH RFC wireless-next] wifi: mac80211: correct rx freq handling for S1G

[Johannes Berg <johannes@sipsolutions.net>]

On Fri, 2026-10-02 at 17:29 +1000, Lachlan Hodges wrote:
> 
> +++ b/include/net/mac80211.h
> @@ -1741,6 +1741,8 @@ enum mac80211_rx_encoding {
>   * @freq: frequency the radio was tuned to when receiving this frame, in MHz
>   *	This field must be set for management frames, but isn't strictly needed
>   *	for data (other) frames - for those it only affects radiotap reporting.
> + *	For S1G, when operating on a 2MHz primary channel, this may be the
> + *	center frequency of the 2MHz primary rather than the 1MHz primary.

Should it really say "may be"? Vs. something more specific about being
1/2 MHz center depending on how it was transmitted/received?

> -static bool ieee80211_rx_valid_freq(int freq, struct ieee80211_link_data *link)
> +bool __ieee80211_rx_valid_freq(struct wiphy *wiphy,
> +			       struct ieee80211_rx_status *status,
> +			       const struct cfg80211_chan_def *chandef)
> +{
> +	u32 pri_khz = ieee80211_channel_to_khz(chandef->chan);
> +	u32 rx_khz = ieee80211_rx_status_to_khz(status);
> +	struct ieee80211_channel *sibling;
> +
> +	if (rx_khz == pri_khz)
> +		return true;

This is also in the original, but now it's called more I think, might
make sense to check them before conversion to kHz? But not sure what the
compiler would do here for the *1000.

> +	/* Any non-S1G case from here is not a valid freq */
> +	if (!cfg80211_chandef_is_s1g(chandef))
> +		return false;
> +
> +	if (!chandef->s1g_primary_2mhz)
> +		return false;
> +
> +	/*
> +	 * Find the sibling 1MHz channel of the 2MHz primary to calculate
> +	 * the 2MHz primary center frequency.
> +	 */
> +	sibling = cfg80211_s1g_get_primary_sibling(wiphy, chandef);
> +	if (!sibling)
> +		return false;
> +
> +	return rx_khz == (pri_khz + ieee80211_channel_to_khz(sibling)) / 2;


All of this gets really complex, IMHO.

Maybe here's another thought: we have

	u16 freq: 13, freq_offset: 1;

What if we make that

	u16 freq: 13, freq_offset: 1,
	    s1g_width_2mhz: 1;

(or something, handwaving about the name) and ask that the driver puts
the 1 MHz center frequency into freq/_offset (so mac80211's comparison
on RX is just ==), but we can still get back the right frequency to
report further out (scan, radiotap) again? Or use two bits (above/below)
then we don't even need to have the sibling channel lookup (we'd just
believe the driver).

I don't know, I'm just thinking out loud, but I feel like maybe that'd
make it more likely we don't break it (again) when we have HT/VHT/etc.
in mind?

johannes