[PATCH] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type

[PATCH] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type

[Marc Zyngier]

The G.a revision of the ARM ARM had it pretty clear that HCR_EL2.FWB
had no influence on "The way that stage 1 memory types and attributes
are combined with stage 2 Device type and attributes." (D5.5.5).

However, this wording was lost in further revisions of the architecture.

Restore the intended behaviour, which is to take the strongest memory
type of S1 and S2 in this case, as if FWB was 0. The specification is
being fixed accordingly.

Fixes: be04cebf3e788 ("KVM: arm64: nv: Add emulation of AT S12E{0,1}{R,W}")
Signed-off-by: Marc Zyngier <maz@kernel.org>
---
 arch/arm64/kvm/at.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/kvm/at.c b/arch/arm64/kvm/at.c
index 8c5d7990e5b31..98cb499fa4b11 100644
--- a/arch/arm64/kvm/at.c
+++ b/arch/arm64/kvm/at.c
@@ -739,8 +739,15 @@ static u64 compute_par_s12(struct kvm_vcpu *vcpu, u64 s1_par,
 			final_attr = s1_parattr;
 			break;
 		default:
-			/* MemAttr[2]=0, Device from S2 */
-			final_attr = s2_memattr & GENMASK(1,0) << 2;
+			/*
+			 * MemAttr[2]=0, Device from S2.
+			 *
+			 * FWB does not influence the way that stage 1
+ 			 * memory types and attributes are combined
+ 			 * with stage 2 Device type and attributes.
+			 */
+			final_attr = min(s2_memattr_to_attr(s2_memattr),
+					 s1_parattr);
 		}
 	} else {
 		/* Combination of R_HMNDG, R_TNHFM and R_GQFSF */
-- 
2.39.2

Re: [PATCH] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type

[Oliver Upton]

On Mon, Nov 25, 2024 at 09:47:56AM +0000, Marc Zyngier wrote:
> The G.a revision of the ARM ARM had it pretty clear that HCR_EL2.FWB
> had no influence on "The way that stage 1 memory types and attributes
> are combined with stage 2 Device type and attributes." (D5.5.5).
> 
> However, this wording was lost in further revisions of the architecture.
> 
> Restore the intended behaviour, which is to take the strongest memory
> type of S1 and S2 in this case, as if FWB was 0. The specification is
> being fixed accordingly.

Since you're already asking for a spec fix, could you mention that the
column headers in DDI0487K.a Table D8-95 are incorrect? MemAttr[1:0] is
used twice, although I believe the first column is actually MemAttr[3:2].

> Fixes: be04cebf3e788 ("KVM: arm64: nv: Add emulation of AT S12E{0,1}{R,W}")
> Signed-off-by: Marc Zyngier <maz@kernel.org>
> ---
>  arch/arm64/kvm/at.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/kvm/at.c b/arch/arm64/kvm/at.c
> index 8c5d7990e5b31..98cb499fa4b11 100644
> --- a/arch/arm64/kvm/at.c
> +++ b/arch/arm64/kvm/at.c
> @@ -739,8 +739,15 @@ static u64 compute_par_s12(struct kvm_vcpu *vcpu, u64 s1_par,
>  			final_attr = s1_parattr;
>  			break;
>  		default:
> -			/* MemAttr[2]=0, Device from S2 */
> -			final_attr = s2_memattr & GENMASK(1,0) << 2;
> +			/*
> +			 * MemAttr[2]=0, Device from S2.
> +			 *
> +			 * FWB does not influence the way that stage 1
> + 			 * memory types and attributes are combined
> + 			 * with stage 2 Device type and attributes.
> +			 */
> +			final_attr = min(s2_memattr_to_attr(s2_memattr),
> +					 s1_parattr);

Otherwise, LGTM.

-- 
Thanks,
Oliver

Re: [PATCH] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type

[Marc Zyngier]

On Tue, 26 Nov 2024 15:27:00 +0000,
Oliver Upton <oliver.upton@linux.dev> wrote:
> 
> On Mon, Nov 25, 2024 at 09:47:56AM +0000, Marc Zyngier wrote:
> > The G.a revision of the ARM ARM had it pretty clear that HCR_EL2.FWB
> > had no influence on "The way that stage 1 memory types and attributes
> > are combined with stage 2 Device type and attributes." (D5.5.5).
> > 
> > However, this wording was lost in further revisions of the architecture.
> > 
> > Restore the intended behaviour, which is to take the strongest memory
> > type of S1 and S2 in this case, as if FWB was 0. The specification is
> > being fixed accordingly.
> 
> Since you're already asking for a spec fix, could you mention that the
> column headers in DDI0487K.a Table D8-95 are incorrect? MemAttr[1:0] is
> used twice, although I believe the first column is actually MemAttr[3:2].

That one has already been fixed as D22366, as described in the Known
Issues document for version K.a (issue 07) [1].

Thanks,

	M.

[1] https://developer.arm.com/documentation/102105/latest/

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

Re: [PATCH] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type

[Oliver Upton]

On Tue, Nov 26, 2024 at 04:30:16PM +0000, Marc Zyngier wrote:
> On Tue, 26 Nov 2024 15:27:00 +0000,
> Oliver Upton <oliver.upton@linux.dev> wrote:
> > 
> > On Mon, Nov 25, 2024 at 09:47:56AM +0000, Marc Zyngier wrote:
> > > The G.a revision of the ARM ARM had it pretty clear that HCR_EL2.FWB
> > > had no influence on "The way that stage 1 memory types and attributes
> > > are combined with stage 2 Device type and attributes." (D5.5.5).
> > > 
> > > However, this wording was lost in further revisions of the architecture.
> > > 
> > > Restore the intended behaviour, which is to take the strongest memory
> > > type of S1 and S2 in this case, as if FWB was 0. The specification is
> > > being fixed accordingly.
> > 
> > Since you're already asking for a spec fix, could you mention that the
> > column headers in DDI0487K.a Table D8-95 are incorrect? MemAttr[1:0] is
> > used twice, although I believe the first column is actually MemAttr[3:2].
> 
> That one has already been fixed as D22366, as described in the Known
> Issues document for version K.a (issue 07) [1].

Ah, right on. Thanks!

-- 
Best,
Oliver

Re: [PATCH] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type

[Oliver Upton]

On Mon, 25 Nov 2024 09:47:56 +0000, Marc Zyngier wrote:
> The G.a revision of the ARM ARM had it pretty clear that HCR_EL2.FWB
> had no influence on "The way that stage 1 memory types and attributes
> are combined with stage 2 Device type and attributes." (D5.5.5).
> 
> However, this wording was lost in further revisions of the architecture.
> 
> Restore the intended behaviour, which is to take the strongest memory
> type of S1 and S2 in this case, as if FWB was 0. The specification is
> being fixed accordingly.
> 
> [...]

Applied to fixes, thanks!

[1/1] KVM: arm64: Fix S1/S2 combination when FWB==1 and S2 has Device memory type
      https://git.kernel.org/kvmarm/kvmarm/c/6fc3a49f2385

--
Best,
Oliver