[PATCH 4/6] ARM: reset: add reset functionality for jumping to a physical address

On Wed, 8 Jun 2011, Will Deacon wrote:

On Wed, Jun 08, 2011 at 04:55:11PM +0100, Frank Hofmann wrote:

quoted

On Tue, 7 Jun 2011, Frank Hofmann wrote:

quoted

On Tue, 7 Jun 2011, Dave Martin wrote:

[ ... ]

quoted

quoted

How about:

	typeof(cpu_reset) *phys_reset =
		(typeof(cpu_reset) *)virt_to_phys(cpu_reset);

Function pointers ;-)
Thanks.

Hmmm ...

Just found a problem with this.

If you have a MULTI_CPU config, this doesn't compile. For two reasons:

1. you cannot use cpu_reset as argument to virt_to_phys because you can't
    take the address
    That bit can be fixed by changing the MULTI_CPU #define in
    <asm/proc-fns.h> not to include the macro argument.
    (There is no code in the arm tree using cpu_reset_whatever names which
    would break from that change ... still, not that nice)

2. even when you do that, you loose the "typeof()" information and the
    above still doesn't compile.

Only a manual type override,

 	void (*phys_reset)(unsigned long) = (void (*)(unsigned long))cpu_reset;

is accepted then.

Damn, yes, I assumed the MULTI_CPU case would just pointer at the structure field,
but it takes the argument as parameter for the invocation. Oh well, I'll hardcode
the type after all then!

It's not just that - the worse bit is that as long as the #define looks 
like:

#define	cpu_reset(addr)		processor.reset(addr)

compile is being refused; one has to ditch the argument part of the macro 
to be able to take the address.

I'm unsure how desirable that change is; it's got the unwanted consequence 
that people who decide to use function names like "graphics_cpu_reset" or 
"cpu_reset_specialregisters" would fall flat on their face.

As said, not that anyone does so right now; just not nice to introduce 
such behaviour.

FrankH.

I'll send a v2 once I've finished cleaning up the code as I've tried to make it
more useful following on from your earlier feedback.

Cheers,

Will