Tags: jimbertools/BPI-Router-Linux

net: ethernet: mtk_eth_soc: apply XGMAC force-mode handling for GMAC1

Align mtk_mac_prepare/link_down with MTK SDK xgmac-force-mode logic.
For XGMII on netsys v3, keep FORCE_MODE programming active for GMAC1
(internal switch WAN path), while still avoiding TRX disable on GMAC1.
Also stop clearing XGMAC force-link bits from non-XGMII link_down path.

This addresses missing GMAC1 force-mode state programming that can
cause unstable TX behavior on WAN (USXGMII).

net: pcs: mtk_usxgmii: replace inline PCS reset with delayed link_poll

The current pcs_get_state() calls pcs_config() (full PCS reset + PHY reset)
every time the link status reads as momentarily down. Since phylink polls
this frequently, it causes a reset storm that disrupts active TX transfers,
resulting in extremely slow upload/transmit speeds on GMAC1 (WAN).

Replace the inline reset with a 1-second delayed work (link_poll) that:
- Only triggers recovery if link is genuinely down (not transient)
- Avoids disrupting active data transfers
- Properly re-schedules itself until link recovers

Based on MTK SDK patch 999-2610 by Bo-Cun Chen.

net: ethernet: mtk_eth_soc: change NAPI poll weight to 256

Increase NAPI poll weight from default 64 to 256 as per MTK SDK patch.
This parameter strongly affects peak throughput of the SW path.

Backported from MTK SDK patch 999-2719.

firmware: update aeonsemi AN8831X firmware from 1.8.5 to 1.9.1

Fixes WAN link flapping (carrier changes every ~15s) on BPI-R4 Pro.
Firmware sourced from stock OpenWrt build (matching MTK SDK 1.9.1).

Revert "uncomment aeon_dpc_ra_enable for performance issues"

This reverts commit 0949602.

uncomment aeon_dpc_ra_enable for performance issues

Merge tag 'v6.12.80' into 6.12-main

This is the 6.12.80 stable release

# -----BEGIN PGP SIGNATURE-----
#
# iQIzBAABCgAdFiEEZH8oZUiU471FcZm+ONu9yGCSaT4FAmnOT38ACgkQONu9yGCS
# aT5hkRAAxKAlYcj9XCzPHRhH0g4cMOGs6BxPUAdYYjeGhc77Vj7Tlnm0a6nvbF+y
# XOIcifC30UI8RnwVna0h1sEfQEekATdRBoMojstft6Wfv4c3LU4mgOtmn/95G1Yc
# D+Ebqfq0n3iIPtadXtPPtOoaVl6Bx/16EJ3tNAo05lb9RLsCSWtZDC88Pwk7UCHu
# r0E1c3bWwLhf7hqd+XzDOnLEv9ChMz6mlN611wSFE0yjCfcoT0Z12rKZ9mMlOvtA
# 5r+MlH1Bde64Kqsut6VlUbmQRWqOkf6wGSwwyUs+p2a9Zh7rZtotZrpLRvdMA1DE
# i00/Vv1hz2rWoao1JEZjmNdFBNS6kBNgakwBqgI8Pn+8rVYYD+dplVNO+JZlv4gR
# +MiIahUhUsF4Cadg0Q4LZQSaxrg34sdGXPH9wbhiswDW62nyIdZS0IJ5soakRt2F
# uavak0Om6lHzEPlwK6CC+hEBsaBWTTDjncX7+y35RexEFh0CoHR44jZ3+XInHs8h
# FGzWYXZt6XrTd4Ntri6C+3gjXiKAzN+ZDbpiiq85O0Fdwhov6W2ondP1kCfmmC/4
# Q2kwHdBD5Q+F5d/0jcXIN6kasvlbS5NLJy1kiklo4qvEfnwRDD/vtrY4IQuFCQBw
# 3b4bKlhEf95MGMynxMaKQOmXCEk/BjXvHjXBw9XeG5zlXVu2tRQ=
# =lFMr
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu Apr  2 13:14:07 2026 CEST
# gpg:                using RSA key 647F28654894E3BD457199BE38DBBDC86092693E
# gpg: Can't check signature: No public key

net: dsa/phy: remove temporary bring-up debug instrumentation

Drop temporary debug messages that were added while debugging DSA mux/phylink
lifecycle and AS21xxx bring-up.

- remove verbose `pr_info`/`dev_info`/`pr_debug` progress traces in DSA setup
  and mxl862xx paths
- remove AS21xxx debug chatter (`aeon_dbg` callsites and noisy init/exit prints)
- keep actionable warning/error logs for real failures

This is a log-noise cleanup only, with no functional behavior changes.

Signed-off-by: Rudy Andram <rmandrad@gmail.com>

net: phy: as21xxx: make PHY matching side-effect free and non-fatal

The AS21xxx match callbacks were mutating device state while the core was still
deciding which driver to bind.

Specifically, match paths were writing `phydev->phy_id` (and for gen1 also
touching vendor registers). On C45-discovered devices, `phydev->phy_id` may
start as 0, so this could leave sysfs showing `0x00000000` or stale values.
Also, returning negative errors from `match_phy_device()` can abort driver
matching and leave the PHY unbound, which breaks management LED setup.

Fix this by:
- making gen1/gen2 match callbacks return `0` on read failure
- removing `phydev->phy_id` updates from match callbacks
- removing side-effect register writes from match callbacks
- updating `phydev->phy_id` in gen1/gen2 probe after successful init using a
  best-effort PID re-read

This keeps matching pure, avoids premature bind aborts, and ensures sysfs
`phy_id` is updated only after the PHY is actually initialized.

Signed-off-by: Rudy Andram <rmandrad@gmail.com>

net: phy: as21: fix phy-id overwrite for all phys (breaking leds on m…

…t7988 internal 1G phy)