openocd: targeting wrong serial port / device

[leocencetti]

Describe the bug

We have a Linux machine with two connected targets for our CI validation pipeline:

• A Nucleo H743ZI board.
• An OpenMV H7 board, connected via STLink V3 programmer.

We use custom udev rules to ensure that the serial ports are always enumerated as follows:

• Nucleo H743ZI --> /dev/nucleo_h743zi
• OpenMV H7 (STLink) --> /dev/openmv_h7

These rules rely on the serial number of the connected device and are therefore bijective.

Now, to the issue: when the two devices are connected to the same USB port through an external USB hub, twister seems unable to select the correct device.

For example, running tests for OpenMV H7:

zephyr/scripts/twister --ninja -p openmv_h7 -T zephyr/tests/kernel/common/ --device-testing --device-serial /dev/openmv_h7

always results in twister flashing the Nucleo board instead (as if the /dev/openmv_h7 was discarded). Targeting the Nucleo board will (correctly) flash the device at /dev/nucleo_h743zi.
The devices are correctly enumerated according to the udev rules.

This issue is not present if the two devices are connected to different USB ports (we are able to flash either one correctly).

To Reproduce
Steps to reproduce the behavior:

1. Connect two devices to the same USB port via an external USB hub
2. Build unit tests with twister and target one of the two devices.
3. twister will default to either of the two devices.

Expected behavior

Targeting a specific serial port will result in twister flashing said port.

Impact

Major, disrupted CI workflow

Environment (please complete the following information):

• OS: Linux (Ubuntu 22.04 LTS)
• Toolchain: Zephyr SDK 0.14.1
• Zephyr Version: 3.0.0

[henrikbrixandersen]

Can you reproduce this by using e.g. west flash without using twister?

[leocencetti]

@henrikbrixandersen yes, I have tried to run:

west flash -r openocd --serial <serial-number>

The same device is selected no matter was is passed as <serial-number>, even with completely random values.

[henrikbrixandersen]

When using --serial with the openocd west runner, the serial number is passed on to the openocd via the _ZEPHYR_BOARD_SERIAL variable. The board openocd.cfg file needs to support this.

At least boards/arm/nucleo_h743zi/support/openocd.cfg does not support this variable. I don't see the OpenMV H7 board in-tree.

[leocencetti]

When using --serial with the openocd west runner, the serial number is passed on to the openocd via the _ZEPHYR_BOARD_SERIAL variable. The board openocd.cfg file needs to support this.

At least boards/arm/nucleo_h743zi/support/openocd.cfg does not support this variable.

@henrikbrixandersen Ok this explains it. Then selection is done based on order? Do you have a reference for a board configuration that DOES support the --serial option?

I don't see the OpenMV H7 board in-tree.

It's not there indeed, we created our own configuration based on the nucleo_h743zi (they share the same MCU).
Nevertheless, from what I understand from the above, this issue would be reproducible for all boards that do not support individual targeting via --serial option.

[henrikbrixandersen]

@henrikbrixandersen Ok this explains it. Then selection is done based on order? Do you have a reference for a board configuration that DOES support the --serial option?

One example is the NEORV32 board definition: https://github.com/zephyrproject-rtos/zephyr/blob/main/boards/riscv/neorv32/support/openocd.cfg

It's not there indeed, we created our own configuration based on the nucleo_h743zi (they share the same MCU). Nevertheless, from what I understand from the above, this issue would be reproducible for all boards that do not support individual targeting via --serial option.

Correct.