[U-Boot] [PATCH v3 2/8] SPL: Port SPL framework to powerpc

[Scott Wood]

On 10/05/2012 10:22:40 AM, Tom Rini wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 10/05/12 06:15, Stefan Roese wrote:
> > On 10/04/2012 07:14 PM, Scott Wood wrote:
> >> On 10/04/2012 02:36:34 AM, Stefan Roese wrote:
> >>> On 10/02/2012 10:08 PM, Scott Wood wrote:
> >>>>>>> +void __noreturn jump_to_image_linux(void *arg) +{ +
> >>>>>>> debug("Entering kernel arg pointer: 0x%p\n", arg); +
> >>>>>>> typedef void (*image_entry_arg_t)(void *, ulong r4,
> >>> ulong r5,
> >>>>>>> ulong r6, +					  ulong r7,  
> ulong r8,
> >>> ulong r9)
> >>>>>>> +		__attribute__ ((noreturn)); +	 
> image_entry_arg_t
> >>>>>>> image_entry = +
> >>>>>>> (image_entry_arg_t)spl_image.entry_point; + +
> >>>>>>> image_entry(arg, 0, 0, EPAPR_MAGIC,
> >>> CONFIG_SYS_BOOTMAPSZ, 0, 0);
> >>>>>>> +}
> >>>>>>
> >>>>>> At what point does the image get cache-flushed?
> >>>>>
> >>>>> Not at all right now. MPC5200 has dcache disabled, at least
> >>>>> in the SPL. Other PowerPC architectures might add a cache
> >>>>> flush here if needed
> >>> at
> >>>>> some time. Okay?
> >>>>
> >>>> Or they might forget to do so and have weird bugs.
> >>>>
> >>>> Why not just call flush_cache() on the image now?  Non-SPL
> >>>> does
> >>> this in
> >>>> common code, not even PPC-specific.
> >>>
> >>> Okay, probably better to add this code now. But shouldn't we
> >>> add this code to the common SPL framework code then? Right
> >>> before calling jump_to_image_linux()?
> >>
> >> Sure, I didn't mean it should go here (it needs to be somewhere
> >> that knows the image start/end, not just the entry point).  This
> >> is just the patch that prompted me to ask the question.
> >
> > I see.
> >
> > Tom, whats your input on this? Do you see any problems about
> > putting a flush_cache() into the common SPL framework code? Are any
> > of the ARM platforms currently using this framework already running
> > with d-cache enabled?
> 
> On ARM, jump_to_image_linux() calls cleanup_before_linux() to take
> care of flushing, etc.  So the PowerPC jump_to_image_linux should take
> care of what it needs to take care of.

It seems that most implementations of cleanup_before_linux() -- which  
is quite misnamed as we need to do this stuff no matter what we're  
loading -- disable and flush the entire cache.  Are there any  
implementations that just flush the range that the image was loaded  
into?  Is that information available at that point?

What if we're loading main U-Boot and not Linux?  Does  
cleanup_before_linux() still get called?

Normal U-Boot does this from generic code, why not SPL?  It should  
become a no-op on platforms that don't need it.

-Scott