Several problems when running the fusa_coe_player.py example on AGX Thor connected to imx274 dual-camera via the Lattice HSB:
The CPU usage for this accelerated case is much higher than expected: 70% of a CPU instead of <10%
The number of macsec_ns interrupts received by the CPUs is very high (~80K per second) when running the FUSA Coe example.
Once the FUSA Coe example has been run once this high number of macsec_ns interrupts returns when running any of the other examples (for example linux_im274_player.py or linux_imx274_latency.py) and the latency is nearly three times higher.
Rebooting the machine returns interrupts and latency back to their previous values.