[PATCH V2] staging: vchiq_2835_arm: Make cache-line-size a required DT property

[Michael Zoran]

On Sun, 2017-02-26 at 20:36 +0100, Noralf Trønnes wrote:
> Den 26.02.2017 19.11, skrev Stefan Wahren:
> > > Eric Anholt <eric at anholt.net> hat am 26. Februar 2017 um 18:16
> > > geschrieben:
> > > 
> > > ...
> > > 
> > > For staging, Greg has been taking patches without platform
> > > maintainer
> > > ack.  I think this is great -- the staging code needs *lots* of
> > > work,
> > > and it generally doesn't need any platform knowledge.
> > > 
> > > As far as this patch, I don't know why we wouldn't just use
> > > cache_line_size() instead of asking DT.
> > 
> > AFAIK this isn't implemented for arm32 yet only for arm64
> 
> cache_line_size() is available on arm32 as well.
> 
> Here's the logic:
> 
> arch/arm/mm/Kconfig:
> 
> config ARM_L1_CACHE_SHIFT_6
>      bool
>      default y if CPU_V7
>      help
>        Setting ARM L1 cache line size to 64 Bytes.
> 
> config ARM_L1_CACHE_SHIFT
>      int
>      default 7 if ARM_L1_CACHE_SHIFT_7
>      default 6 if ARM_L1_CACHE_SHIFT_6
>      default 5
> 
> 
> arch/arm/include/asm/cache.h:
> 
> #define L1_CACHE_SHIFT          CONFIG_ARM_L1_CACHE_SHIFT
> #define L1_CACHE_BYTES          (1 << L1_CACHE_SHIFT)
> 
> 
> include/linux/cache.h:
> 
> #ifndef CONFIG_ARCH_HAS_CACHE_LINE_SIZE
> #define cache_line_size()       L1_CACHE_BYTES
> #endif
> 
> 
> The only reason for using a DT property is if one should build a
> kernel
> that can boot on both Pi1 and Pi2/3. In theory it should be possible,
> but when I did the downstream "move to mainline arch code" work, I
> couldn't get such a unified kernel image to work. All kinds of memory
> corruption problems that seemed to be dma related.
> 
> 
> Noralf.
> 

The firmware also had a bug/feature that if the "cache-line-size"
property is missing, it assuming the kernel is very old and forces a
small cache line even on RPI 3 and even with arm64.  So we would still
need to require the property even if it wasn't used unless the firmware
can be changed.

I hit this the hard way when experimenting with converting vchiq to a
non platform driver and it why I submitted this patch in the first
place.  Things will work without the property, just they will be subtly
broken.