Jetson AGX Orin 64GB fails to boot after enabling full CPU lockstep on Jetson Linux R39.2.1

We are trying to enable full CPU lockstep on a NVIDIA Jetson AGX Orin 64GB, with a P3701-0005 module and P3737-0000 carrier board, using Jetson Linux R39.2.1 / JetPack 7.2.1.

The standard unmodified R39.2.1 image flashes and boots correctly. For the lockstep image, we set:
ccplex_platform_features = <0x0 0x00000000>;

num_clusters_2core_pair_lockstep = <0x3>;

num_clusters_1core_pair_lockstep = <0x0>;
in bootloader/tegra234-mb1-bct-misc-common.dtsi

We also decompiled the DTB selected for P3701-0005:
kernel/dtb/tegra234-p3737-0000+p3701-0005-nv.dtb

changed the C7 min-residency-us property to:

min-residency-us = <0xffffffff>;

and recompiled the DTB with dtc.

The modified image completes the flashing process, but the board does not boot correctly and shows no screen or serial output, while the clean R39.2.1 image works.

Is CPU lockstep officially supported on P3701-0005 with Jetson Linux R39.2.1? Are additional changes required? What is the recommended method to verify the active CEMODE after boot?
Thanks.

Hi,

This is not validated on Jetpack 6 and 7. It may not work properly.

Could it work on inferior versions of the software? If so in which ones? If not, there is any other comercial solution like this were we can execute programs in lockstep mode with Linux Kernel on Desktop?

Thanks in advance.

Hi,

Based on NVIDIA’s reply, full CPU lockstep has not been validated on JetPack 6 or 7, so the configuration changes may not be sufficient to boot successfully.

There is an earlier report of successful booting with full lockstep on JetPack 5.0.1. However, support for the AGX Orin 64GB module was introduced in JetPack 5.1.1, so that result may not apply to your P3701-0005.

NVIDIA also suggested an ATF/OP-TEE patch during the earlier investigation. I could not establish that it resolves this issue or applies to R39.2.1.

To investigate further, could you confirm whether the missing serial output was checked through J26, the micro-USB debug port, at 115200 baud, using the same connection that captures boot messages from the working image? If any output is available from power-on, it could help identify where boot stops. Debug UART instructions.

Best regards,
Alejandro Godínez
Embedded Software Engineer, RidgeRun

Contact us: support@ridgerun.com
Developers Wiki: https://developer.ridgerun.com
Website: https://www.ridgerun.com