[PATCH v8 0/6] lx2160a: change the pad muxes through the RCW override
From: Vincent Jardin via B4 Relay
Date: Wed Nov 18 2026 - 06:00:36 EST
The LX2160A muxes its IIC, XSPI, SDHC, UART, IRQ and EVT pads from
fields of the RCW. They are read-only in RCWSR12..14 of DCFG, and are
changed at runtime through the override registers in the DCSR view of
DCFG, which read 0 until written.
Since commit 8a1365c7bbc1 ("arm64: dts: lx2160a: add pinmux and i2c
gpio to support bus recovery"), fsl-lx216x.dtsi drives the override
registers with pinctrl-single for the I2C bus recovery. Its first pin
state clears every other field of the word: a board whose RCW programs
a non-zero pad mux loses those pads at the first I2C probe.
This series reads the fields in effect from RCWSR and writes RCWCR, as
fsl_guts_rcw_rmw() already does for the SerDes protocols:
- patches 1 and 2 describe the pad mux as a child of the DCFG node
- patch 3 exports the RCW override of fsl-guts for the pad-mux fields
only
- patch 4 adds the pinctrl driver: one group per RCW field, one
function per encoding
- patch 5 moves the pad muxes of fsl-lx216x.dtsi under DCFG, which
gains its DCSR view. Labels and pin node names stay, so no board
file changes, and the pinctrl-single node is removed
- patch 6 notes above i2c6 and i2c7 that their pin states, and so
the bus recovery, work only while the other bus is disabled: IIC7
and IIC8 share one field
Patch 4 uses the exports of patch 3, so it cannot be applied alone.
As confirmed with Linus Walleij[1], patches 1 to 4 should go through
the pinctrl tree now that patch 3 is Ack'd by the soc/fsl maintainer, and
then patches 5 and 6 through the i.MX tree once the driver is in,
since the I2C pin states of fsl-lx216x.dtsi should follow this path.
[1] https://lore.kernel.org/all/CAD++jLnWcVDbh-vf20w4nfvOSZik619XqoEjXTBZ29sJkX-bCQ@xxxxxxxxxxxxxx/
DTBs built before patch 5 keep their pinctrl-single node and behave as
before. There is no Fixes: tag, since patch 5 depends on a new driver.
I did not believe it should be backported then.
Tested on a Nodebox v3 (LX2160A) whose RCW muxes XSPI1_A_DATA[7:4],
SDHC1 and IIC5 are not set with the defaults, so I could confirm:
- with pinctrl-single the first I2C probe cleared those fields
- with this series RCWSR12..14 read after boot exactly what the RCW and
the board's pin states select
- all I2C controllers come up with their recovery states and a
forced bus recovery on IIC1 (multi-master enabled for the test)
changes only that field
- RCWSR27 reads what the RCW selects for EC1, EC2, USB_EXT, GTX_CLK
and SDHC1_DS. Selecting SDHC1_DS again from debugfs is the first
override of that word, and keeps its other fields
Thanks to Vladimir Oltean for many comments and for pointing at
fsl_guts_rcw_rmw() when I tried to prime the override registers
from the RCW instead:
https://github.com/nxp-qoriq/rcw/pull/8
Signed-off-by: Vincent Jardin <vjardin@xxxxxxx>
---
Changes in v8:
- Patch 6 keeps the pin states of i2c6 and i2c7 in fsl-lx216x.dtsi
and comments each node that they, and so the bus recovery, work
only while the other bus is disabled, instead of deleting them: a
board that uses one of the two buses keeps its bus recovery
(Vladimir Oltean)
- dt-bindings: the function/group constraints are listed per group, in
the order of the RM and of the driver tables, instead of per
function: one rule per RCW field. (Vladimir Oltean)
- pinctrl: set .suppress_bind_attrs, as the other drivers of
drivers/pinctrl/freescale do: unbinding with a hog on the node read
freed group descriptors (Vincent J, code reviews)
- Patches 2, 3 and 5 are not changed, apart from the tags: Reviewed-by of
Vladimir Oltean on patches 2 to 5, Acked-by of Christophe Leroy on
patch 3
- Link to v7: https://lore.kernel.org/r/20261117-for-upstream-lx2160a-rcw-pinmux-v1-v7-0-5d7201970629@xxxxxxx
Changes in v7:
- dt-bindings, pinctrl: add the clk-out function, encoding 0 of
CLK_OUT_PMUX, so debugfs shows clk-out instead of 0x0. Rev. 1 of
the Reference Manual marks 0b000 as reserved in the field: its
revision history says "Updated 0b000 as reserved" and removes the
CLK_OUT configuration section, while table 19 still lists
000 (CLK_OUT) (Vladimir Oltean)
- dt-bindings: the description says that a group is one RCW field and
that some functions carry two controllers at once, such as i2c on
sdhc2-data74 for IIC7 and IIC8 (Vladimir Oltean)
- New patch 6: drop the I2C bus recovery of i2c6 and i2c7, which share
the SDHC2_DAT74 field (Vladimir Oltean)
- Tested on the Nodebox v3: debugfs shows clk-out, i2c6 and i2c7 have
no recovery state while the other I2C buses keep theirs, and the
eMMC on the SDHC2_DAT[7:4] pads still runs 8-bit HS400
- Link to v6: https://lore.kernel.org/r/20261115-for-upstream-lx2160a-rcw-pinmux-v1-v6-0-fbdce4565a5c@xxxxxxx
Changes in v6:
- pinctrl: the driver is no longer specific to the LX2160A, so that
other Layerscape SoCs can be added to it: pinctrl-layerscape-rcw.c
(suggested by Vladimir Oltean)
- Patches 1, 2 and 5 are not changed
- Link to v5: https://lore.kernel.org/r/20261115-for-upstream-lx2160a-rcw-pinmux-v1-v5-0-d246166ea469@xxxxxxx
Changes in v5:
- soc: fsl: guts: the exports become fsl_guts_pmux_read(),
fsl_guts_pmux_override() and fsl_guts_pmux_override_supported(). guts
lists the pad-mux fields of the LX2160A and refuses any other field
or register, so the serdes fields cannot be changed (Vladimir Oltean)
- dt-bindings, pinctrl, dts: add the EC1, EC2, USB_EXT, GTX_CLK and
SDHC1_DS pad muxes of RCWSR27, the second reg of the node (Vladimir Oltean)
- Patch 3 is reworked: drop the review tags, new round of review is
needed (TBC, from Christophe Leroy and Frank Li)
- Patch 1 gets a reg, 5 groups and 4 functions:
same drop the the Reviewed-by of Rob Herring and Frank Li
- Patch 2: its description names RCWSR27 too, I assume the
Reviewed-by of Rob Herring can remain
- Link to v4: https://lore.kernel.org/r/20261114-for-upstream-lx2160a-rcw-pinmux-v1-v4-0-eefcea066a55@xxxxxxx
Changes in v4:
- pinctrl: MAINTAINERS with imx@xxxxxxxxxxxxxxx
FUNC and GROUP macros with LX2160A_PMUX_ prefix (Frank Li)
- Add the Reviewed and Ack tags of Rob Herring and Frank Li
- Patches 1, 2, 3 and 5 are not changed
- Link to v3: https://lore.kernel.org/r/20261108-for-upstream-lx2160a-rcw-pinmux-v1-v3-0-5a0063f1da9e@xxxxxxx
Changes in v3:
- dt-bindings, dts: the pad-mux child is bus-addressed like the
clock-controller child of the LS1028A dcfg: its reg is the RCWSR12
to RCWSR14 window, under #address-cells, #size-cells and ranges on
the dcfg node, the override registers stay the DCSR view of the
parent (Krzysztof Kozlowski)
- cover letter: patch 4 cannot be applied without patch 3 (Linus Walleij)
- Patches 3 and 4 are unchanged
- Link to v2: https://lore.kernel.org/r/20261103-for-upstream-lx2160a-rcw-pinmux-v1-v2-0-3aad0860b984@xxxxxxx
Changes in v2:
- dt-bindings: list, for each function, the groups, so the
schema refuses the associations the driver refuses (Frank Li)
- Patches 2 to 5 are unchanged
- Link to v1: https://lore.kernel.org/r/20261103-for-upstream-lx2160a-rcw-pinmux-v1-v1-0-0a61a9837744@xxxxxxx
---
Vincent Jardin (6):
dt-bindings: pinctrl: add the NXP LX2160A RCW pad mux
dt-bindings: soc: fsl: layerscape-dcfg: add the LX2160A pad-mux child
soc: fsl: guts: add the RCW override of the pad muxes
pinctrl: freescale: add the Layerscape RCW pad-mux driver
arm64: dts: lx2160a: describe the pad muxes as the DCFG RCW override
arm64: dts: lx2160a: note that i2c6 and i2c7 share their pad mux
.../bindings/pinctrl/fsl,lx2160a-rcw-pinmux.yaml | 417 ++++++++++++++++++++
.../bindings/soc/fsl/fsl,layerscape-dcfg.yaml | 18 +
MAINTAINERS | 8 +
arch/arm64/boot/dts/freescale/fsl-lx216x.dtsi | 387 ++++++++++---------
drivers/pinctrl/freescale/Kconfig | 15 +
drivers/pinctrl/freescale/Makefile | 1 +
drivers/pinctrl/freescale/pinctrl-layerscape-rcw.c | 429 +++++++++++++++++++++
drivers/soc/fsl/guts.c | 125 +++++-
include/linux/fsl/guts.h | 3 +
9 files changed, 1227 insertions(+), 176 deletions(-)
---
base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4
change-id: 20261103-for-upstream-lx2160a-rcw-pinmux-v1-31f9ad97dc86
Best regards,
--
Vincent Jardin <vjardin@xxxxxxx>