MT7921 TX Power Reporting Fix

Kernel patch fixing incorrect txpower reporting in the Linux mt76 driver for MT7921-based WiFi adapters (MT7921U, MT7922, and related chipsets).

Status

v4 was superseded on 2026-03-21 when MediaTek kernel engineer Sean Wang took over the rework, consolidating the fix into shared mt76_connac code so MT7921, MT7922, and MT7925 all benefit from a single patch rather than per-chip duplication.

Sean Wang's v2 series (3 patches) was posted to linux-wireless on 2026-04-01 (1/3 title: wifi: mt76: connac: use a helper to cache txpower_cur). Trailers carried forward from this repo's work:

• All 3 patches: Reported-by: Devin Wittmayer <lucid_duck@justthetip.ca> and Tested-by: Devin Wittmayer <lucid_duck@justthetip.ca>
• Patches 1 and 2: additionally Co-developed-by: Devin Wittmayer <lucid_duck@justthetip.ca> and Signed-off-by: Devin Wittmayer <lucid_duck@justthetip.ca>

Series is awaiting merge by Felix Fietkau into the mt76 tree, then forward to wireless-next and mainline.

Everything below documents the v1-v4 local history that led to Sean's rework. The technical analysis of the root bug (rate power loop discarding txpower_cur, and mt7921's add_chanctx not updating phy->chandef) is what Sean's v2 builds on.

---

The Problem

The mt7921 driver never updates phy->txpower_cur from the rate power configuration sent to firmware. mt76_get_txpower() reports bogus values via nl80211 -- typically 3 dBm regardless of actual regulatory or SAR limits:

$ iw dev wlan0 info | grep txpower
        txpower 3.00 dBm    # Wrong

This affects all MT7921/MT7922 devices on all bands. hcxdumptool and other tools that read txpower may crash or behave incorrectly.

Root Cause

Two issues compound:

1. The rate power loop in mt76_connac_mcu_rate_txpower_band() computes the correct bounded TX power per channel but discards the return value. The equivalent mt7915 code stores it to phy->txpower_cur -- the mt7921 connac path does not.

2. mt7921 uses the chanctx model but its add_chanctx callback does not update phy->chandef, leaving it stale after association. The rate power loop's channel comparison then fails silently.

The Fix (v4, superseded)

Two changes in one patch:

• Capture the rate power loop result in mt76_connac_mcu_rate_txpower_band() and store to txpower_cur when processing the current channel. Subtract the multi-chain path delta before storing, since mt76_get_txpower() adds it back when reporting -- matching how mt7915 handles this via mt76_get_power_bound().

• Sync phy->chandef from chanctx in add_chanctx and change_chanctx, with a lightweight mt7921_update_txpower_cur() helper that recomputes the bounded power for the current channel without reprogramming firmware rate tables.

After applying:

$ iw dev wlan0 info | grep txpower
        txpower 36.00 dBm   # Correct (matches regulatory limit)

Test Results

Tested on Alfa AWUS036AXML (MT7921AU), kernel 6.19.8, Canada (ISED):

Band	Regulatory Limit	v4 Result	Stock	Status
2.4 GHz	36 dBm	36 dBm	3 dBm	PASS
5 GHz	23 dBm	23 dBm	3 dBm	PASS
6 GHz	12 dBm	12 dBm	3 dBm	PASS

Independent testing by sam8641 (MT7921U USB + MT7922 PCIe, US regulatory, Debian 6.12.74):

Device	Band	Result	Status
MT7921U USB	2.4/5 GHz	30 dBm	PASS
MT7922 PCIe	2.4/5/6 GHz	30/30/18 dBm	PASS

Version History

Version	Date	Approach	Outcome
v1	2026-01-25	Store hw->conf.power_level in set_rate_txpower	Rejected: wrong source (Felix Fietkau)
v2	2026-01-30	Read bss_conf.txpower in bss_info_changed	Rejected: should come from rate power loop (Sean Wang)
v3	2026-03-17	Rate loop store + chanctx calls + BSS_CHANGED_TXPOWER	Rejected: chanctx calls too heavy, BSS_CHANGED breaks multi-vif (Sean Wang)
v4	2026-03-19	Rate loop store + lightweight helper + path delta fix	Superseded: Sean Wang took over the rework on 2026-03-21, consolidating into shared connac code
Sean v2	2026-04-01	Shared mt76_connac path with txpower_cur cached via helper	Posted to linux-wireless. Reported-by + Tested-by on all 3 patches; Co-developed-by + Signed-off-by on patches 1 and 2. Awaiting Felix Fietkau merge.

Upstream Status

• Current (Sean Wang v2): Posted to linux-wireless 2026-04-01, awaiting Felix Fietkau merge. Lore search: mt76 txpower_cur
• Lore (all submissions): lucid_duck@justthetip.ca
• v1 patchwork (historical): Link
• Community thread: morrownr/USB-WiFi#700

Files

• 0001-wifi-mt76-mt7921-fix-txpower-reporting-from-rate-pow.patch -- The v4 kernel patch (2 files changed, 41 insertions, 3 deletions)

Applying the Patch

cd /path/to/linux
git apply 0001-wifi-mt76-mt7921-fix-txpower-reporting-from-rate-pow.patch

Or to build just the mt76 modules:

make -C /lib/modules/$(uname -r)/build M=drivers/net/wireless/mediatek/mt76 modules

Stable Backport Note

v4 depends on commit 56e3867 which renamed mt76_tx_power_nss_delta() to mt76_tx_power_path_delta(). For kernels before ~6.18, replace mt76_tx_power_path_delta with mt76_tx_power_nss_delta -- they return identical values for 1-4 chain devices.

Known Limitations (follow-up work)

• iw set txpower fixed not reflected in managed/station mode (BSS_CHANGED_TXPOWER dropped per reviewer feedback -- needs a multi-vif-safe approach). Works in AP mode via IEEE80211_CONF_CHANGE_POWER.
• MT7925 needs analogous changes in its own main.c (the connac common code fix benefits it, but without the chandef sync the channel comparison never matches).

License

GPL-2.0 (same as the Linux kernel).