[PATCH v2] ARM: realview: basic device tree implementation

[Arnd Bergmann]

On Thursday 22 May 2014 14:39:20 Nicolas Pitre wrote:
> On Thu, 22 May 2014, Arnd Bergmann wrote:
> 
> > On Thursday 22 May 2014 14:09:52 Nicolas Pitre wrote:
> > > On Thu, 22 May 2014, Arnd Bergmann wrote:
> > > 
> > > > On Thursday 22 May 2014 15:45:03 Catalin Marinas wrote:
> > > > > By making the DMA memory Strongly Ordered, we avoid the
> > > > > L220 write buffer. In addition to the above commit, we also had
> > > > > 2503a5ecd86c (ARM: 6201/1: RealView: Do not use outer_sync() on
> > > > > ARM11MPCore boards with L220).
> > > > 
> > > > Ah, so this also stands in the way of multiplatform realview 
> > > 
> > > Isn't the real value of a multiplatform realview in the ability to use 
> > > it with Qemu?  Did Qemu go to the extreme of emulating all those cache 
> > > issues?
> > 
> > The qemu developers have in the past always argued strongly that they
> > want to run the same kernel that runs on real hardware. Having a kernel
> > that runs only on the emulated version but not on actual hardware is
> > clearly against that goal, and  we should use CONFIG_ARCH_VIRT for
> > that purpose.
> 
> But the kernel that actually runs on the real hardware also runs on Qemu 
> already.  It is just that, by some coincidence, a multi-platform kernel 
> that pretends to have the same peripherals as a Realview board could 
> also be used on Qemu.
> 
> Otherwise I really don't see the point in making the code more complex 
> just to make a multi-platform kernel run on actual Realview hardware 
> that is quite obsolete now and that very few people have access to.

The main reason is that out of principle I want to get to the point where
every ARMv6 and ARMv7 machine is part of CONFIG_ARCH_MULTIPLATFORM.

Then again, once we introduce CONFIG_BROKEN_MULTIPLATFORM, we could
have an option that depends on that, and turns on both
ARM_DMA_MEM_BUFFERABLE as well as the hack to remove the outer_sync
from mb() and wmb() for CONFIG_SMP.

	Arnd