Re: [PATCH] ARM64: errata: Add workaround for HIP10/HIP10C erratum 162200803

[Marc Zyngier <maz@kernel.org>]

On Tue, 08 Jul 2025 13:05:24 +0100,
Zhou Wang <wangzhou1@hisilicon.com> wrote:
> 
> On 2025/7/3 18:44, Marc Zyngier wrote:
> > On Wed, 02 Jul 2025 10:57:13 +0100,
> > Zhou Wang <wangzhou1@hisilicon.com> wrote:
> >>
> >> However，migration between different KVMs will be broken :(
> >> I am not sure that should we consider this case as well?
> > 
> > This isn't optional. You cannot break migration on existing systems,
> > and the only case that *must* break is to restore a VM that hasn't
> > seen this limitation on a HW that enforces it.
> 
> So the whole story is that we should let GICD_TYPE.num_LPIs as a
> writable field, and set this field from VMM(e.g. QEMU). And it will
> be failed if doing migration between VMs with different num_LPI
> configurations.

Note that on a non-broken system, setting num_LPIs to any value should
always work.

> After above done, this errata could be added.

Which would be the only case where num_LPIs wouldn't be writable, and
set to a fixed value.

Note that an alternative would be to not have a kernel erratum, but to
force num_LPIs from userspace when running on your platform. But that
implies that: you have your own, non-upstream VMM, and that there is
no opportunity for userspace to adversely impact the system as a whole.

> Not sure my thought above is right? I will try to prepare a num_LPIs
> writable patch if the direction is right.

I think you got the gist of it.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.