Re: [PATCH 0/3] Revert HAFDBS changes

["Pierre-Clément Tosi" <ptosi@google.com>]

Hi Chris,

On Fri, Oct 27, 2023 at 01:23:51PM +1300, Chris Packham wrote:
> As discussed this series reverts the HAFDBS changes that caused an issue
> on AC5/AC5X. I think there are some improvements that can be made to the
> initial memory map for the AC5/AC5X but so far nothing I've found makes
> it compatible with the HAFDBS changes.

Would you mind briefly explaining what those issues are and/or point me to the
discussion where it was decided to revert these patches?

The feature is quite useful for users of CONFIG_CMO_BY_VA_ONLY, to speed up the
boot sequence: instead of reverting these patches altogether, would a reasonable
alternative be to put them behind a build option?

Also, could you confirm that the "initial memory map" you are referring to above
only describes actual memory as, IIRC, some boards were using mappings **much**
larger than their DRAM address space?

> 
> 
> Chris Packham (3):
>   Revert "armv8: enable HAFDBS for other ELx when FEAT_HAFDBS is
>     present"
>   Revert "arm64: Use level-2 for largest block mappings when FEAT_HAFDBS
>     is present"
>   Revert "arm64: Use FEAT_HAFDBS to track dirty pages when available"
> 
>  arch/arm/cpu/armv8/cache_v8.c      | 30 +++---------------------------
>  arch/arm/include/asm/armv8/mmu.h   | 20 ++++----------------
>  arch/arm/include/asm/global_data.h |  2 --
>  3 files changed, 7 insertions(+), 45 deletions(-)
> 
> -- 
> 2.42.0
> 

-- 
Pierre