Re: [RFC][PATCH] misc: Introduce reboot_reason driver

On Thursday 10 December 2015 13:43:16 John Stultz wrote:

On Thu, Dec 10, 2015 at 12:24 PM, Rob Herring [off-list ref] wrote:

quoted

The fact that we are using notifiers for reset reason and triggering
is probably some indication that some infrastructure is needed. But I
don't think you need to do that here as long as it is all kernel
internals. We'll make the 2nd guy do it. 

Though, just so I understand better, what is problematic w/ the reset
notifiers? They provide the reboot command argument, which is the core
of what is needed. It actually seemed like it was almost designed with
this problem in mind.

Notifiers in general are a bit of a kludge. We often use them in places
that have not been abstracted well enough yet, and they make it
less obvious what is actually going on when something happens, or
in what order things are called.

I'm actually less worried about the notifier side here than about
the general problem of the communication channel. The reboot reason
is only one of a number of things that the kernel needs to communicate
to the boot loader. Other things may include:

- boot device
- location of the kernel
- command line
- properties of the /chosen DT node in general
- boot scripts
- ethernet MAC addresses
- bootloader console configuration

Every bootloader is different here regarding what can be configured
and how we do it. Often the configuration is done entirely from user
space, but some platforms have kernel support. So picking one particular
aspect and trying to unify that one with a common kernel interface
but ignoring all the others may cause problems if we later want to
add a more general abstraction.

It also looks like at least some of the interfaces require a checksum
to be updated, or are based on variable-length entries, both of which
require a proper driver.

	Arnd