[PATCH net v2 0/2] net: phy: qcom: at803x: IPQ5018 analog initialization fixes

[Yongzhao Chen <yongzhao.derek@gmail.com>]

When the IPQ5018 internal GE PHY is connected to another PHY without a
cable, its analog setup matters for the 1000BASE-T link. Two problems
in that setup are fixed here.

Patch 1 fixes the short-cable DAC values, which were written to the
wrong bits. It is the patch sent as v1 [1], unchanged except for
Andrew's Reviewed-by.

Patch 2 applies the analog settings in probe, right after the PHY reset,
instead of only when the MAC attaches the PHY. The PHY starts
autonegotiating when it leaves reset, so until attach it negotiated
with its reset defaults. On a Redmi AX5400, where it connects to PHY4
of a QCA8337, 1000BASE-T never came up in that state and SmartSpeed on
the switch PHY dropped its 1000BASE-T advertisement for good. With
patch 2 the link came up at 1 Gb/s with SmartSpeed left enabled, and
the SmartSpeed workaround discussed in [2] is no longer needed.

On that board, with both patches backported to OpenWrt's Linux 6.18.52,
the link was verified at 1 Gb/s after a first boot, three reboots, a
power-off cold boot, interface down/up cycles, renegotiations and a
network restart. During a separate 10-minute observation, sampled link
status remained at 1 Gb/s and no new switch-side CPU PHY link-down
events were logged. After the cold boot the switch side first reported
1 Gb/s at 4.4 s; it went down at MAC attach and recovered at 25.2 s.

Patch 2 accesses the PHY right after reset_control_reset(), which
pulses GCC_GEPHY_MISC_ARES for about 1 us. The vendor SDK waits 200 ms
after deasserting each Ethernet reset, but it does so for every block
alike, so that does not establish a GE PHY-specific minimum delay.
This patch adds no post-reset delay. Diagnostic warm-boot tests on
this board read back the values correctly after writing them in probe.
I have no specification for the required post-reset interval. George,
does the GE PHY require a minimum delay or a readiness check after
ARES is deasserted, before its analog settings are written and
autonegotiation is restarted?

Thanks to Ziyang Huang for asking whether the DAC settings had been
corrected [3], which is how the first problem was found, and to Andrew
Lunn for his reviews in the v3 thread, which kept the investigation
going until the cause was found.

Changes since v1:
- Added patch 2.
- Patch 1: added Andrew's Reviewed-by and Ziyang's Suggested-by.

[1] https://lore.kernel.org/netdev/20260927155136.2489-1-yongzhao.derek@gmail.com/
[2] https://lore.kernel.org/netdev/20260923215858.1653-1-yongzhao.derek@gmail.com/
[3] https://lore.kernel.org/netdev/SEYPR01MB58827E0D18ACC93AF98A4109C98E2@SEYPR01MB5882.apcprd01.prod.exchangelabs.com/

Yongzhao Chen (2):
  net: phy: qcom: at803x: Fix IPQ5018 short-cable DAC values
  net: phy: qcom: at803x: Apply IPQ5018 analog settings at probe

 drivers/net/phy/qcom/at803x.c | 94 ++++++++++++++++++++++++++---------
 1 file changed, 70 insertions(+), 24 deletions(-)


base-commit: a7bfaba4823e3c165bb2004c74eff7c096672bc7
-- 
2.43.0