Hi there,
hoping for some help or insight into an issue we're having...
RZ/N1D custom board (yocto poky 3.1.4 dunfell) based on the dev kit; A7 is running linux (4.9.0), M3 is running Goal (2.20.0) and we're using the CTC/RPC for communication between the A7/M3.
We're using the linux swupdate application to upgrade (dual copy) the board including: kernel, device tree, rootfs (squashfs), M3 app; each of these artifacts is encrypted.
When running swupdate on the A7 side, the M3 side starts having peripheral errors (overruns on the UART, I2C errors).
Has anyone seen an issue like this before, where A7 activity affects the M3?
Things I've checked:
- peripheral conflicts between the linux and M3; we're using a h/w semaphore to safely share some gpio but that's it, no other conflicts found.
- IRQ priorities on the M3 NVIC side; these were all the same and the lowest priority (RZ/N1 OS Library User's Manual 5.6/7.3, since goal doesn't support nested IRQs right?).
- running strace on swupdate; didn't see anything that looks like it could be an issue.
- the file swupdate is processing is in DDR (/tmp); I found stress-ng mem test results from when we brought up the board and they showed no difference in DDR performance between the dev kit and our board.
- monitored with mpstat/iostat during swupdate, doesn't look like anything is off the rails.
Things I've tried but the issue persists:
- making the M3 UART/I2C interrupts non-kernel IRQs and increasing their priorities.
- removing CTC/RPC support.
- not running our A7 application.
- swupdate dry-run option, so that all writes go to /dev/null.
Things that I'm still looking in to:
- profiling/tracing with perf.
- more stress-ng testing (ran some stress-ng cpu/io tests but couldn't reproduce).
- bypassing artifact decryption.
Any suggestions for identifying the root cause?