Hi, for our system we are evaluating if we can achieve better performances using a preempt-rt kernel.
I'm using Real-time Edge Yocto as explained in RTEDGEYOCTOUG rev 2.8
On the imx8mp-evk target, there are no stability issues. On our custom board, there is a Marvell mv88e6390 dsa switch and, if I enable its DT node, the system becomes unstable. At random times we get kernel panic, sometimes in the first ten minutes, sometimes after one and a half hours. The error trace is not helpful, it looks like a stack overflow or a stack corruption. Sometimes it's a NULL (or random unallocated address) pointer dereference, but sometimes the error is different, I've seen an error about killing the idle task a couple of times. What seems constant is that the error occurs during the scheduler context after an interrupt "low-level" service.
I can't tell if the problem originates in the Marvell driver code, but in the current state of the device tree the marvel node is the main difference between the evk and the custom board
