Re: [PATCH] ARM: supplementing IO accessors with 64 bit capability

[Catalin Marinas <catalin.marinas@arm.com>]

On Wed, Oct 22, 2014 at 08:10:27PM +0100, Mathieu Poirier wrote:
> On 22 October 2014 18:44, Catalin Marinas <catalin.marinas@arm.com> wrote:
> > On Wed, Oct 22, 2014 at 05:06:23PM +0100, mathieu.poirier@linaro.org wrote:
> >> +static inline void __raw_writeq(u64 val, volatile void __iomem *addr)
> >> +{
> >> +     asm volatile("strd %1, %0"
> >> +                  : "+Qo" (*(volatile u64 __force *)addr)
> >> +                  : "r" (val));
> >> +}
> >> +
> >> +static inline u64 __raw_readq(const volatile void __iomem *addr)
> >> +{
> >> +     u64 val;
> >> +     asm volatile("ldrd %1, %0"
> >> +                  : "+Qo" (*(volatile u64 __force *)addr),
> >> +                    "=r" (val));
> >> +     return val;
> >> +}
> >> +#endif
> >
> > I'm curious why you need these. Do you have a device that needs a 64-bit
> > single access or you are trying to read two consecutive registers?
> 
> The fundamental data size of Coresight STM32 for ARMv7 is
> implementation defined and can be 32 or 64bit.  As such stimulus ports
> can support transaction sizes of up to 64 bit.

The STM programmer's model spec recommends something else (though I find
the "3.6 Data sizes" chapter a bit confusing):

  To ensure that code is portable between processor micro-architectures
  and system implementations, ARM recommends that only the native data
  size of the machine is used, and smaller sizes. For the 32-bit ARMv7
  architecture, only 8, 16, and 32-bit transfers are recommended. For an
  ARMv8 processor that supports the AArch64 Execution state, it is
  recommended that the fundamental data size of 64-bits is implemented.

Which means that you should not use readq/writeq on a 32-bit system.

-- 
Catalin