[PATCH 4/7] arm: initial support for Marvell Armada 370/XP SoCs

[Jean-Christophe PLAGNIOL-VILLARD]

On 16:09 Mon 06 May     , Thomas Petazzoni wrote:
> Sascha, Jean-Christophe,
> 
> I have one question below regarding the SoC code.
> 
> On Fri,  3 May 2013 18:51:08 +0200, Thomas Petazzoni wrote:
> > +static inline void mvebu_memory_find(unsigned long *phys_base,
> > +				     unsigned long *phys_size)
> > +{
> > +	void __iomem *sdram_win = IOMEM(MVEBU_SDRAM_WIN_BASE);
> > +	int cs;
> 
> 
> Here...
> 
> > +	*phys_base = ~0;
> > +	*phys_size = 0;
> > +
> > +	for (cs = 0; cs < 4; cs++) {
> > +		uint32_t base = readl(sdram_win + DDR_BASE_CS_OFF(cs));
> > +		uint32_t ctrl = readl(sdram_win + DDR_SIZE_CS_OFF(cs));
> 
> ... here ...
> 
> 
> > +#if defined(CONFIG_ARCH_ARMADA_370)
> > +static int mvebu_init_clocks(void)
> > +{
> > +	uint32_t val;
> > +	unsigned int rate;
> > +	void __iomem *sar = IOMEM(MVEBU_SAR_BASE) + SAR_LOW_REG_OFF;
> > +
> > +	val = readl(sar);
> 
> ... and here, I'm directly poking at physical addresses, but it seems
> like Barebox can run with the MMU enabled. Should I be mapping those
> registers before accessing them? I see that drivers use
> dev_request_mem_region(), but this clock and memory code isn't (yet?) a
> driver per-se.

no as only the ddr is cached but the dev_request_mem_region track who request
what yes please create a device/drivers as you will be able to probe a dt
(even builtin) in this case