Serial port symlinks broken in bookworm.

[metachip]

There appears to be a bug in the assignment of the primary and secondary serial ports in bookworm.

It looks like a problem in:

raspberrypi-sys-mods: /etc/udev/rules.d/99-com.rules

When bluetooth is disabled in config.txt as:

dtoverlay=disable-bt

I would expect that

/dev/serial0 -> ttyAMA0
/dev/serial1 -> ttyS0

and this is what bullseye does, correctly, but instead with bookworm we see:

/dev/serial1 -> ttyAMA0

and /dev/serial0 is not symlinked at all.

This is contrary to the advice here:

https://www.raspberrypi.com/documentation/computers/configuration.html#primary-and-secondary-uart

and looks like a bug in bookworm :)

[XECDesign]

@pelwell, is this something that has been or will be fixed with newer firmware/kernel or just expected behaviour?

[pelwell]

It's definitely not expected behaviour, and it may have been fixed with the latest firmware (coming to an apt upgrade soon) - knowing which Pi it was would help.

[pelwell]

Reproduced on a Pi 400.

[metachip]

Serial Device Alias Fault Diagnosis

The problem is the line:

readlink /sys/class/tty/ttyS[0-9]*/device/of_node | sed 's/base/:/' | cut -d: -f2

in

/etc/udev/rules.d/99-com.rules

On Pi5 it works.

On Pi4 it does not.

The problem is the format of the descriptor for ttyS0 is different.

On Pi5:

$ echo cat /proc/device-tree/model
Raspberry Pi 5 Model B Rev 1.0
$ tail -2 /boot/firmware/config.txt
enable_uart=1
dtoverlay=disable-bt
$ readlink /sys/class/tty/ttyS0/device/of_node
../../../../firmware/devicetree/base/soc/serial@7d50c000

On Pi4:

$ echo cat /proc/device-tree/model
Raspberry Pi 4 Model B Rev 1.4
$ tail -2 /boot/firmware/config.txt
enable_uart=1
dtoverlay=disable-bt
$ readlink /sys/class/tty/ttyS0/device/of_node
$ ls /sys/class/tty/ttyS0/device/
driver driver_override modalias power subsystem tty uevent

Bottom line:

/etc/udev/rules.d/99-com.rules

requires modification to work on models other than Pi5. The test on of_node fails on Pi4 because it does not exist.

[pelwell]

The real problem is that the use of the bluetooth alias to determine which UART is which should be conditional on the existence of the console alias, not vice-versa:

diff --git a/etc.armhf/udev/rules.d/99-com.rules b/etc.armhf/udev/rules.d/99-com.rules
index 1931d3d..7ad5f38 100644
--- a/etc.armhf/udev/rules.d/99-com.rules
+++ b/etc.armhf/udev/rules.d/99-com.rules
@@ -17,11 +17,11 @@ SUBSYSTEM=="pwm", ACTION!="remove", PROGRAM="/bin/sh -c 'chgrp -R gpio /sys%p &&
 KERNEL=="ttyAMA[0-9]*|ttyS[0-9]*", PROGRAM="/bin/sh -c '\
         ALIASES=/proc/device-tree/aliases; \
         TTYNODE=$$(readlink /sys/class/tty/%k/device/of_node | sed 's/base/:/' | cut -d: -f2); \
-        if [ -e $$ALIASES/bluetooth ] && [ $$TTYNODE/bluetooth = $$(strings $$ALIASES/bluetooth) ]; then \
-            echo 1; \
-        elif [ -e $$ALIASES/console ]; then \
+        if [ -e $$ALIASES/console ]; then \
             if [ $$TTYNODE = $$(strings $$ALIASES/console) ]; then \
-                echo 0;\
+                echo 0; \
+            elif [ -e $$ALIASES/bluetooth ] && [ $$TTYNODE/bluetooth = $$(strings $$ALIASES/bluetooth) ]; then \
+                echo 1; \
             else \
                 exit 1; \
             fi \

Doing so with dtoverlay=disable-bt results in:

/dev/serial0 -> ttyAMA0

which is arguably more correct than:

/dev/serial0 -> ttyAMA0
/dev/serial1 -> ttyS0

since /dev/ttyS0 doesn't actually work (try it - the device node is disabled, and yet somehow the driver is creating a /dev/ entry for it, perhaps because of 8250.nr_uarts=1 on the command line).

[pelwell]

There's now a PR with this change: #85