synchronous external abort: kernel: Workqueue: events check_temperature_work_handler [apex] #41
Description
I'm using a pcie Coral in an ARM Pinebook Pro to run several services including Frigate NVR. Frigate can use CPU, GPU, or Coral for object detection, but focuses on Coral.
When I use the CPU detector, the server can run for days without a problem, but it's slow, the CPU runs hot, and it doesn't detect much. When I use the Coral detector, the Frigate interface is more responsive, CPU is reasonable around 30-60%, and it detects more objects. However, after some time, between about 15 minutes and 12 hours, the computer will become unresponsive, with the power LED flashing between green and orange. Here are the logs at that time:
Internal error: synchronous external abort: 0000000096000210 [#1] PREEMPT SMP
....
hostname kernel: Workqueue: events check_temperature_work_handler [apex]
I have searched the internet and don't find the combination of synchronous external abort and apex. Pinebook Pro is a low powered device that isn't very popular. It's possible that no one else is using this combination of computer/Coral/Frigate. Could it be related to the CPU (rk3399) being relatively slow compared to the devices other people are using for Frigate (Intel N100 is recommended) or other Coral projects? I.e. if I'm planning to switch to a new ARMv9 server, do I need to solve this or will that switch solve it? Conversely, some people are using Raspberries and the Pinebook Pro is more powerful than that.