Re: [PATCH v6 4/5] pinctrl: freescale: add the Layerscape RCW pad-mux driver

From: Vladimir Oltean

Date: Tue Nov 17 2026 - 16:02:19 EST


On Tue, Nov 17, 2026 at 05:44:00PM +0100, Vincent Jardin wrote:
> Hi Vladimir,
>
> > The "i2c" and "gpio" functions of lx2160a_sdhc2_data74_funcs[] here
> > hides some real unpleasantness. These 4 pins either get all assigned to
> > I2C7_SCL, I2C7_SDA, I2C8_SCL, I2C8_SDA, or they get all assigned to GPIO.
> > But they cannot be assigned in pairs of two (or individually, i.o.w. not
> > in the granularity we need), which makes independent I2C bus recovery
> > practically impossible (i2c6 as I2C while i2c7 is as GPIO).
>
> yep, the doc is clear about it: SDHC2_DAT74_PMUX is a single field for
> the 4 pads, 0b001 gives GPIO_2[18:15] and 0b010 gives IIC7 and IIC8
> together (RM rev 1, 4.9.8.9). IIC7 and IIC8 are not available on any
> other pad.
>
> Checking from other angles, the A-010650 workaround uses one write for both
> buses: 0x700100130 = RegA OR 0x031. That mask looks wrong to me though: bits 5:4
> belong to SDHC2_BASE_PMUX, and OR-ing 0x001 onto 0b010 gives the reserved
> 0b011. Could someone at NXP have a look ?
>
> > The way the code is currently written, the pin_request() call will fail here:
>
> Yes, in fact, it is not new with this series: both nodes have shared the
> same setting since
> 8a1365c7bbc1 ("arm64: dts: lx2160a: add pinmux and i2c gpio to support bus recovery")
> from Carlos Song (NXP) where the two pinctrl-single states covered the same
> bits, so the second controller would have failed the same way. Wihtin the
> kernel, there is no board that enables i2c6 or i2c7, so nobody did hit it!
>
> > Due to the fundamental issue of this limitation, I don't think it makes
> > sense to have the pinctrl states described in fsl-lx216x.dtsi for i2c6
> > and i2c7. Perhaps you could remove them in another change (either as a
> > follow-up, or another patch in v7 if you need to resend).
>
> Done for v7 as an additional patch: the DTS drops pinctrl-names, pinctrl-0/1
> and scl/sda-gpios from i2c6 and i2c7, and the two states nothing else uses.
> A board that needs one of these buses can add its own states, or set the
> field once with a hog on the pad-mux node.

I'm not entirely happy with my own suggestion either, I'll comment more
on v7.

>
> I tested on my board for regressions, since there those 4 pads carry
> the eMMC DAT[7:4]: the other I2C buses still set up their GPIO recovery,
> and the eMMC still runs in 8-bit HS400.
>
> I also went through the other fields. 7 more settings mix two
> controllers the same way:
> - flextimer on iic2, irq3-0, irq7-4, evt4-3 and evt2-0 (FTM1 and FTM2 channels)
> - sdhc-spi on sdhc1-base (SDHC1 with SPI3_PCS0) and usb on usb-ext (USB1 and USB2)
> No dts claims any of them.
>
> About your previous comment on clk-out: v7 names encoding 0 "clk-out".
> On my board, with debugfs, it shows now:
>
> pin 20 (evt2-0) 0:? evt
> pin 21 (clk-out) 0:? clk-out
> pin 22 (asleep) 0:? asleep
>
> and the U-Boot port of this driver shows "clk-out : clk-out" in
> "pinmux status".
> My take: the reserved marking comes from RM rev 1: its revision history
> says "Updated 0b000 as reserved" and the CLK_OUT configuration section was
> removed, while table 19 still lists 000 (CLK_OUT).
>
> There is a similar gap on EC1: the signal tables list CLK_OUT2 on
> EC1_TXD0/GPIO4_DAT03, but EC1_PMUX only has 0b00 RGMII and 0b01 GPIO,
> 0b10 and 0b11 being reserved. Would you know which value selects CLK_OUT2 ?

These are good questions. I've opened a ticket internally and I'm hoping
to get some answers regarding the correctness of the documented A-010650
workaround procedure and the CLK_OUT/CLK_OUT2 signals, which are in a weird
superposition of documented and not documented at the same time.