Re: [PATCH] panic.c: export panic_on_oops

[David Woodhouse <dwmw2@infradead.org>]

On Mon, 2009-10-12 at 14:36 +0200, Ingo Molnar wrote:
> Well other than not using sleeping locks in that codepath it should be 
> properly serializable. If the serial driver, netconsole, fbcon and all 
> the other non-trivial console drivers can do it then MTD should be able 
> to do it too.

It's distinctly non-trivial. A 'real' user of the hardware may have just
triggered a block erase, which could take hundreds (or even thousands)
of milliseconds to complete. We can't always suspend that erase; do we
really want to _wait_ for it?

If we're actually in a _panic_ situation, then yes -- we want to wait,
and that's what the device driver's panic_write() method should do.

But for less critical output? I'm not so sure.

Of course, I'm currently looking at revamping the MTD APIs so they're
not entirely synchronous, and are queue-based like the block device API.
Then the console ->write() method could just queue the write, and we
need some other logic to ensure that it really does _happen_ when the
system crashes. A kind of 'purge queue of console writes' function in a
panic handler... except that as you point out, sometimes we actually
want that even when we haven't actually panicked.

-- 
dwmw2