[PATCH 1/3] ARM: Introduce atomic MMIO clear/set

[ezequiel.garcia@free-electrons.com (Ezequiel Garcia)]

On Sat, Aug 10, 2013 at 07:43:08PM +0400, Alexander Shiyan wrote:
> > On Sat, Aug 10, 2013 at 11:02:38AM -0300, Ezequiel Garcia wrote:
> > > On Sat, Aug 10, 2013 at 04:49:28PM +0400, Alexander Shiyan wrote:
> > > > > Some SoC have MMIO regions that are shared across orthogonal
> > > > > subsystems. This commit implements a possible solution for the
> > > > > thread-safe access of such regions through a spinlock-protected API
> > > > > with clear-set semantics.
> > > > > 
> > > > > Concurrent access is protected with a single spinlock for the
> > > > > entire MMIO address space. While this protects shared-registers,
> > > > > it also serializes access to unrelated/unshared registers.
> [...]
> > > > > +void atomic_io_clear_set(void __iomem *reg, u32 clear, u32 set)
> > > > > +{
> > > > > +	spin_lock(&__io_lock);
> > > > > +	writel((readl(reg) & ~clear) | set, reg);
> > > > > +	spin_unlock(&__io_lock);
> > > > > +}
> > > > > +EXPORT_SYMBOL(atomic_io_clear_set);
> > > > 
> > > > So, one lock is used to all possible registers?
> > > > Seems a regmap-mmio can be used for such access.
> > > > 
> > > 
> > > Thanks for the hint! Quite frankly, I wasn't familiar with regmap-mmio.
> > > 
> > > However, after reading some code, I fail to see how that helps in this case.
> > > 
> > > Note that we need to access the *same* MMIO address from completely
> > > different (and unrelated) drivers, such as watchdog and clocksource.
> > > 
> > > So I wonder who would "own" the regmap descriptor, and how does the other
> > > one gets aware of that descriptor?
> > > 
> > > In addition given we can use orion_wdt (originally meant for mach-kirkwood)
> > > to support mvebu SoC watchdog, we need to sort this out in a completely
> > > multiplatform capable way.
> > > 
> > > Ideas?
> > 
> > Answering myself...
> > 
> > How about using drivers/mfd/syscon.c to create the regmap owner for the shared
> > register (TIMER_CTRL in this case, but others might appear) ?
> > 
> > Or adding a new mfd implementation if syscon does not fit ?
> > 
> > Does this sound like an overkill ?
> 
> Yes, syscon is designed especially for such cases.
> 

Indeed, syscon looks like a nice match for this use case.
(although it still looks like an overkill to me).

I've been trying to implement a working solution based in syscon but I'm
unable to overcome an issue.

The problem is that we need the register/regmap to initialize the clocksource
driver for this machine (aka the timer). Of course, this happens at a
*very* early point, way before the syscon driver is available... :-(

Maybe someone has an idea?
-- 
Ezequiel Garc?a, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com