[linux-pm] [PATCH v4 2/4] mfd: omap: control: core system control driver

[Tony Lindgren]

* Tony Lindgren <tony at atomide.com> [120808 07:11]:
> * Tony Lindgren <tony at atomide.com> [120808 07:05]:
> > * Konstantin Baydarov <kbaidarov at dev.rtsoft.ru> [120725 04:10]:
> > > +
> > > +u32 omap_control_status_read(void)
> > > +{
> > > +	return __raw_readl(omap_control_base);
> > > +}
> > 
> > Ah OK it's changed here.. Sorry I was looking at the older
> > version.
> > 
> > > +void __init of_omap_control_init(const struct of_device_id *matches)
> > > +{
> > > +	struct device_node *np;
> > > +	struct property *pp = 0;
> > > +
> > > +	for_each_matching_node(np, matches) {
> > > +		pp = of_find_property(np, "reg", NULL);
> > > +		if(pp) {
> > > +			omap_control_phys_base = (unsigned long)be32_to_cpup(pp->value);
> > > +			omap_control_mapsize = (size_t)be32_to_cpup( (void*)((char*)pp->value + 4) );
> > > +			/*
> > > +			 * Map control module register CONTROL_STATUS register:
> > > +			 * omap24xx - OMAP24XX_CONTROL_STATUS
> > > +			 * am33xx   - AM33XX_CONTROL_STATUS
> > > +			 * omap34xx - OMAP343X_CONTROL_STATUS
> > > +			 * omap44xx - OMAP4_CTRL_MODULE_CORE_STATUS
> > > +			 * omap54xx - OMAP5XXX_CONTROL_STATUS
> > > +			 */
> > > +			omap_control_base = ioremap(omap_control_phys_base, omap_control_mapsize);
> > > +		}
> > > +	}
> > > +}
> > 
> > You should probably add a function for setting omap_control_base
> > separately from *set_globals* in arch/arm/mach-omap2/common.c.
> > That way it's initialized early for id.c, and you can initialize
> > everything else later as regular device drivers.
> > 
> > FYI, we want to initialize as much as possible late so we have
> > proper debugging console set up in case things go wrong.
> 
> Hmm it seems that omap_control_base is now only initialized for DT boot
> case? This will break booting on almost all systems..

Looking at things more, looks like omap_type() is getting called
early from timer.c. So maybe just let set_globals initialize
omap_control_base and let omap_type() read that directly. That way
the SCM core driver can behave like a regular device driver and it
does not need to be initialized early.

Regards,

Tony