Re: [PATCH] ARM: mach-shmobile: sh7372 generic suspend/resume

[Russell King - ARM Linux <linux@arm.linux.org.uk>]

On Wed, Jul 13, 2011 at 08:05:04PM +0900, Magnus Damm wrote:
> On Thu, Jul 7, 2011 at 10:50 PM, Russell King - ARM Linux
> <linux@arm.linux.org.uk> wrote:
> > On Thu, Jul 07, 2011 at 10:48:13PM +0900, Magnus Damm wrote:
> >> From: Magnus Damm <damm@opensource.se>
> >>
> >> Convert the sh7372 Core Standby code to make use
> >> of the new generic ARM cpu suspend/resume code.
> >
> > Thanks.  Only thing which stands out is:
> >
> >> +     .global sh7372_resume_core_standby
> >> +sh7372_resume_core_standby:
> >> +     ldr     pc, 1f
> >> +1:   .long   cpu_resume - PAGE_OFFSET + PLAT_PHYS_OFFSET
> >
> > Is this because sh7372_resume_core_standby must be ARM stuff, but
> > cpu_resume may be Thumb?
> 
> That may have something to do with it, but I suspect not. Perhaps you
> wonder why I'm not pointing the reset vector directly to cpu_resume()
> - that's because we need the reset vector to be aligned to 4k.

Wouldn't a simple branch instruction do here?  Branches are relative
(within 24-bit) so you wouldn't need any games with v:p stuff.  You
may have to put that in .data so that it can reach cpu_resume if
you're using a large initramfs image.

> I simply nicked the PA->VA conversion code from the SMP entry code in headsmp.S:
> 
> 	.align  12
> ENTRY(shmobile_secondary_vector)
> 	ldr     pc, 1f
> 1:	.long   secondary_startup - PAGE_OFFSET + PLAT_PHYS_OFFSET
> 
> Any suggestions?

Again, a simple branch instruction should avoid the need to reference
PLAT_PHYS_OFFSET or virtual addresses.