Dear Graeme Russ,
In message 4D8739F6.5040805@gmail.com you wrote:
I kind of like the idea of different reset sources (CPU exception, hardware failure, user initiated) but agree copying the linux architecture is over the top.
What's the difference as far as do_reset() is concenred? It shall just (hard) reset the system, nothing else.
Is there any reason reset() could not take a 'reason' parameter? It could be a bit-mask with CPU, SOC and arch reserved bits (unhandled exception, user initiated, panic etc) and board specific bits
What for? To perform the intended purpose, no parameter is needed.
Board or arch specific code could handle different reasons however they please (like logging it in NVRAM prior to restart, gracefully shutting down multiple CPU's, clearing DMA buffers etc)
That would be a layer higher than do_reset() (for example, in panic()).
All 'hang', 'panic', 'reset' etc code can be simplified into a single code path (although calling 'reset' to 'hang' is a bit odd)
hang() and reset() are intentionally very different things. A call to hang() is supposed to hang (surprise, surprise!) infinitely. It must not cause a reset.
panic() is a higher software layer. It probably results in calling reset() in the end.
Best regards,
Wolfgang Denk