H7 I2C driver appears to have an off-by-one in interrupt handling.

Using kernel trace, we've observed the H7 driver idling for periods of microseconds-milliseconds, and based on the syscall trace it appears to be handling interrupts slightly wrong.

There are two components.

First: it consistently waits for its interrupt (using a closed recv) before enabling the interrupt. This will only work if it's previously accumulated an unhandled notification, since interrupts are only delivered when enabled. I'm not totally sure how this is working at all, so I may have missed something.

Second: when it enables interrupts we often see it immediately take one, which suggests that something had already pended and the sleep-wait it's attempting isn't working.

More data needed, noting this so it's not in my brain.