[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Z9Q3-FM-QwiIn9gh@google.com>
Date: Fri, 14 Mar 2025 14:06:48 +0000
From: Quentin Perret <qperret@...gle.com>
To: Marc Zyngier <maz@...nel.org>
Cc: Oliver Upton <oliver.upton@...ux.dev>, Joey Gouly <joey.gouly@....com>,
Suzuki K Poulose <suzuki.poulose@....com>,
Zenghui Yu <yuzenghui@...wei.com>,
Catalin Marinas <catalin.marinas@....com>,
Will Deacon <will@...nel.org>,
Vincent Donnefort <vdonnefort@...gle.com>,
linux-arm-kernel@...ts.infradead.org, kvmarm@...ts.linux.dev,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 4/6] KVM: arm64: Move hyp state to hyp_vmemmap
On Friday 14 Mar 2025 at 11:31:36 (+0000), Marc Zyngier wrote:
> On Thu, 27 Feb 2025 00:33:08 +0000,
> Quentin Perret <qperret@...gle.com> wrote:
> > @@ -698,13 +697,13 @@ int __pkvm_host_share_hyp(u64 pfn)
> > if (ret)
> > goto unlock;
> > if (IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
> > - ret = __hyp_check_page_state_range((u64)virt, size, PKVM_NOPAGE);
> > + ret = __hyp_check_page_state_range(phys, size, PKVM_NOPAGE);
>
> OK, I think I finally clicked here. Does it mean that all the tracking
> is now done in terms of PAs instead of VAs?
Yep, that's exactly that. The hyp_vmemmap is indexed by pfn, so I felt
that the conversion to a PA-based tracking made sense. That also make it
clear that the 'hyp state' is not a property of a mapping, but really of
the underlying physical page.
> > if (ret)
> > goto unlock;
> > }
> >
> > - prot = pkvm_mkstate(PAGE_HYP, PKVM_PAGE_SHARED_BORROWED);
> > - WARN_ON(pkvm_create_mappings_locked(virt, virt + size, prot));
> > + __hyp_set_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
> > + WARN_ON(pkvm_create_mappings_locked(virt, virt + size, PAGE_HYP));
>
> And this is the split between the state now being kept in the on a PA
> base and the actual mapping that is now only takes the page attributes
> and no SW bits?
Precisely, and the next patch in this series takes advantage of the
fact that we're now de-correlating the hyp state from the presence of a
hyp s1 mapping in the linear map range. In the future there'll be more
use-cases for this I think (e.g. the hyp allocator where we'll have
pages owned by the hypervisor but only mapped in the 'private' range,
things like that).
Thanks,
Quentin
Powered by blists - more mailing lists