[PATCH wireless] wifi: b43: back off the N-PHY TX power per rate

[Rosen Penev]

On Thu, Oct 15, 2026 at 1:11 PM Alessio Ferri <alessio.ferri at mythread.it> wrote:
>
> Il giorno Thu, 15 Oct 2026 10:59:28 -0700
> Rosen Penev <rosenp at gmail.com> ha scritto:
>
> > On N-PHY rev 3+ the hardware TX power control aims at the highest
> > per-rate target, and an adjusted power table backs each rate group off
> > from it. b43 fills that table from nphy->tx_power_offset[], but never
> > sets the offsets, so they stay zero and every rate is sent at the
> > power allowed for CCK.
> >
> > The SPROM power offsets exist to avoid this: on boards with external
> > PAs, OFDM and MCS rates have to be sent a few dB lower than CCK to
> > keep the PA linear. Sent at full power, 16-QAM and 64-QAM frames are
> > too distorted to decode. On an ASUS RT-N16 (BCM4718, N-PHY rev 5,
> > radio 2056, ofdm2gpo 0x66666666) the router could only send at 6-18
> > Mbit/s and MCS 0-2, even at -50 dBm, and 24 Mbit/s and up or MCS 3 and
> > up delivered nothing.
> >
> > Set the offsets from the per-rate targets b43 already computes from
> > the SPROM, as brcmsmac does, before the table is built.
> >
> > With this, the RT-N16 sends at 54 Mbit/s, and at MCS 7 with HT
> > enabled, and TCP from the router to a client goes from 9 to 17-18
> > Mbit/s.
> >
> > Fixes: 8615eb2870f1 ("b43: N-PHY: support setting custom TX power")
> > Assisted-by: LLM
> > Signed-off-by: Rosen Penev <rosenp at gmail.com>
> > ---
> >  drivers/net/wireless/broadcom/b43/phy_n.c | 11 +++++++++++
> >  1 file changed, 11 insertions(+)
> >
> > diff --git a/drivers/net/wireless/broadcom/b43/phy_n.c
> > b/drivers/net/wireless/broadcom/b43/phy_n.c index
> > 22359bd9db47..c9b2b7496983 100644 ---
> > a/drivers/net/wireless/broadcom/b43/phy_n.c +++
> > b/drivers/net/wireless/broadcom/b43/phy_n.c @@ -5834,6 +5834,7 @@
> > static enum b43_txpwr_result b43_nphy_op_recalc_txpower(struct
> > b43_wldev *dev, struct ieee80211_channel *channel =
> > dev->wl->hw->conf.chandef.chan; struct b43_ppr *ppr =
> > &nphy->tx_pwr_max_ppr; u8 max; /* qdBm */
> > +     int i;
> >
> >       if (nphy->tx_pwr_last_recalc_freq == channel->center_freq &&
> >           nphy->tx_pwr_last_recalc_limit == phy->desired_txpower)
> > @@ -5868,6 +5869,16 @@ static enum b43_txpwr_result
> > b43_nphy_op_recalc_txpower(struct b43_wldev *dev, /* Make sure we
> > didn't go too low */ b43_ppr_apply_min(dev, ppr, INT_TO_Q52(8));
> >
> > +     /*
> > +      * The hardware power control aims at the highest per-rate
> > target.
> > +      * The adjusted power table backs other rates off from it,
> > so that
> > +      * e.g. 64-QAM isn't sent at the power allowed for CCK.
> > +      */
> > +     max = b43_ppr_get_max(dev, ppr);
> > +     memset(nphy->tx_power_offset, 0,
> > sizeof(nphy->tx_power_offset));
> > +     for (i = 0; i < B43_PPR_RATES_NUM; i++)
> > +             nphy->tx_power_offset[i] = max - ppr->__all_rates[i];
> > +
> >       /* Apply */
> >       b43_mac_suspend(dev);
> >       b43_nphy_tx_power_ctl_setup(dev);
>
> I'm not the maintainer, and i also have the same exact patch in my repo
> https://github.com/aleferri/b43-6362-wip/tree/main/patches/mainline, so
> i'm all for it, but isn't the ppr currently populated with 2Ghz sprom
> params for all bands?
This patch comes from work on an RT-N16, which is 2ghz only, hence why
5ghz is not touched. I submitted this since there's a Fixes: tag
that's appropriate here. The actual fixes I'll submit in the next
merge window. It's quite late in the cycle.

There's no maintainer here. It's an orphaned driver. I wanted to focus
on brcmsmac but that driver is really slow.
>
> Alessio