[PATCH] ARM: at91: Fix: Change internal SRAM memory type to "MT_MEMORY_SO"

[Russell King - ARM Linux]

On Fri, May 24, 2013 at 06:52:54PM +0200, Jean-Christophe PLAGNIOL-VILLARD wrote:
> On 17:40 Fri 24 May     , Russell King - ARM Linux wrote:
> > On Fri, May 24, 2013 at 04:03:22PM +0200, Jean-Christophe PLAGNIOL-VILLARD wrote:
> > > On 12:20 Fri 24 May     , Russell King - ARM Linux wrote:
> > > > On Fri, May 24, 2013 at 07:11:04AM +0000, Yang, Wenyou wrote:
> > > > > The story is: for sama5d3x with Cortex-A5 core, if not so, when copying
> > > > > code snippet to the internal SRAM, then jump to run this code, but fail
> > > > > to run.
> > > > 
> > > > And that is where your mistake is - you forgot that you're working with
> > > > a CPU with harvard caches which will require some cache maintanence
> > > > between copying the code and executing it.
> > > > 
> > > > You want to look at flush_icache_range() rather than making this memory
> > > > strongly ordered.
> > > 
> > > I understand your point but today we map a SRAM as MT_DEVICE
> > 
> > If you map SRAM as MT_DEVICE then you won't be able to execute code from
> > it.  It needs to be a normal memory mapping.
> 
> Yeah that a bug on AT91, by luck it work on armv3/v4 with MT_DEVICE, I should
> have spot it earlier when cleanning the at91 but did not
> 
> That's why Yang change the SRAM mapping as MT_MEMORY_SO

I said "normal memory".  Strongly ordered is not "normal memory".