Re: [RFC PATCH 5/5] KVM: arm64: Enable HAFDBS for guests not on migration

[Leonardo Bras <leo.bras@arm.com>]

On Thu, Oct 29, 2026 at 10:54:24AM +0800, Tian Zheng wrote:
> Hi Leo,
> 
> I've read your discussion with Oliver on DBM RFC v2. As a suggestion,
> maybe we should go with Oliver's approach, i.e. revert to the v1 scheme
> and continue fixing from there.

Hi Tian,
I have been reading a bit more about permission indirection to try figuring 
out a way that the change may work for both ways.

The problem with the previous approach is that it assume DBM=write at all 
times, which is not architecturally correct. Having DBM=write only on 
HAFDBS=on as suggested would require us to transverse the whole pagetable 
to convert encodings when we enable/disable it, which can be quite 
expensive.

One option would be to always enable HAFDBS when system supports HDBSS, but 
I still have to figure out if that causes any performance regression before 
suggesting.

The option I am pursuing is thinking on an unified code path for direct 
permissison (both HACDBS=on and HACDBS=off) so we don't have to convert 
between them, neither worry about the impact of HAFDBS=on on performance. 

I will reply on the thread once I figure one of those options out.

Thanks!
Leo