Skip to contentSkip to conversation

arm64/sme: Fix SMPRI_EL1 traps for KVM guests

PATCH Applied Mark Brown 2 participants 7 messages Nov 1, 2022

Subsystems (4)
lore
Share & tools
lore

Mark Brown

author
#0 lore raw

v2 cover arm64/sme: Fix SMPRI_EL1 traps for KVM guests

When adding support for SME I missed management of traps for SMPRI_EL1 in KVM guests, and also the need to manage TPIDR2_EL0 fine grained traps for EL1. This series adds that management, and also updates the documentation to help avoid such issues in future.

This was noticed due to Vincent Donnefort noticing that qemu has implemented support for SME virtualisation but not fine grained traps which was causing problems for qemu in nVHE mode since that already used fine grained traps with SME, the FGT registers are not implemented and UNDEF. As a result of this change VHE will also be affected so a separate series "KVM: arm: Refuse to enable KVM on systems with SME but not FGT" has been sent which will ensure we don't run on such systems though initial review suggests that this will likely be treated as a qemu documentation issue.

v2:
- Rebase onto v6.1-rc2
- Also manage TPIDR2 fine grained traps for VHE mode.

Mark Brown (2):
arm64: booting: Document our requirements for fine grained traps with
SME
KVM: arm64: Trap access to SMPRI_EL1 and TPIDR2 in VHE mode

 Documentation/arm64/booting.rst |  8 ++++++++
 arch/arm64/kvm/hyp/vhe/switch.c | 26 ++++++++++++++++++++++++--
 2 files changed, 32 insertions(+), 2 deletions(-)

base-commit: 30a0b95b1335e12efef89dd78518ed3e4a71a763

Signature · 1 line
2.30.2
author
#1 lore raw

PATCH v2 1/2 arm64: booting: Document our requirements for fine grained traps with SME

With SME we require that fine grained traps on access to TPIDR2_EL0 and SMPRI_EL1 are disabled but did not document that fact. Add the relevant register bits.

---
 Documentation/arm64/booting.rst | 8 ++++++++
 1 file changed, 8 insertions(+)
Documentation/arm64/booting.rst +8 -0
index 8aefa1001ae5..8c324ad638de 100644
@@ -340,6 +340,14 @@ Before jumping into the kernel, the following conditions must be met:
     - SMCR_EL2.LEN must be initialised to the same value for all CPUs the
       kernel will execute on.
 
+    - HWFGRTR_EL2.nTPIDR2_EL0 (bit 55) must be initialised to 0b01.
+
+    - HWFGWTR_EL2.nTPIDR2_EL0 (bit 55) must be initialised to 0b01.
+
+    - HWFGRTR_EL2.nSMPRI_EL1 (bit 54) must be initialised to 0b01.
+
+    - HWFGWTR_EL2.nSMPRI_EL1 (bit 54) must be initialised to 0b01.
+
   For CPUs with the Scalable Matrix Extension FA64 feature (FEAT_SME_FA64)
 
   - If EL3 is present:
Signature · 1 line
2.30.2
author
#2 lore raw

PATCH v2 2/2 KVM: arm64: Trap access to SMPRI_EL1 and TPIDR2 in VHE mode

In reply to Mark Brown

On systems with SME access to the SMPRI_EL1 priority management register is controlled by the nSMPRI_EL1 fine grained trap and TPIDR2_EL0 is controlled by nTPIDR2_EL0. We manage these traps in nVHE mode but do not do so when in VHE mode, add the required management.

Without this these registers could be used as side channels where implemented.

Fixes 861262ab8627 ("KVM: arm64: Handle SME host state when running guests")
---
 arch/arm64/kvm/hyp/vhe/switch.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)
arch/arm64/kvm/hyp/vhe/switch.c +24 -2
index 7acb87eaa092..9dac3a1a85f7 100644
@@ -63,10 +63,20 @@ static void __activate_traps(struct kvm_vcpu *vcpu)
 		__activate_traps_fpsimd32(vcpu);
 	}
 
-	if (cpus_have_final_cap(ARM64_SME))
+	if (cpus_have_final_cap(ARM64_SME)) {
 		write_sysreg(read_sysreg(sctlr_el2) & ~SCTLR_ELx_ENTP2,
 			     sctlr_el2);
 
+		sysreg_clear_set_s(SYS_HFGRTR_EL2,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK,
+				   0);
+		sysreg_clear_set_s(SYS_HFGWTR_EL2,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK,
+				   0);
+	}
+
 	write_sysreg(val, cpacr_el1);
 
 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el1);
@@ -88,9 +98,21 @@ static void __deactivate_traps(struct kvm_vcpu *vcpu)
 	 */
 	asm(ALTERNATIVE("nop", "isb", ARM64_WORKAROUND_SPECULATIVE_AT));
 
-	if (cpus_have_final_cap(ARM64_SME))
+	if (cpus_have_final_cap(ARM64_SME)) {
+		/*
+		 * Enable access to SMPRI_EL1 - we don't need to
+		 * control nTPIDR2_EL0 in VHE mode.
+		 */
+		sysreg_clear_set_s(SYS_HFGRTR_EL2, 0,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK);
+		sysreg_clear_set_s(SYS_HFGWTR_EL2, 0,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK);
+
 		write_sysreg(read_sysreg(sctlr_el2) | SCTLR_ELx_ENTP2,
 			     sctlr_el2);
+	}
 
 	write_sysreg(CPACR_EL1_DEFAULT, cpacr_el1);
 
Signature · 1 line
2.30.2
maintainer
#3 lore raw

Re: [PATCH v2 2/2] KVM: arm64: Trap access to SMPRI_EL1 and TPIDR2 in VHE mode

Quoted 20 lines · Mark Brown

On systems with SME access to the SMPRI_EL1 priority management register is controlled by the nSMPRI_EL1 fine grained trap and TPIDR2_EL0 is controlled by nTPIDR2_EL0. We manage these traps in nVHE mode but do not do so when in VHE mode, add the required management.

Without this these registers could be used as side channels where implemented.

Fixes 861262ab8627 ("KVM: arm64: Handle SME host state when running guests")
---
 arch/arm64/kvm/hyp/vhe/switch.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)
arch/arm64/kvm/hyp/vhe/switch.c +0 -0
index 7acb87eaa092..9dac3a1a85f7 100644

arch/arm64/kvm/hyp/vhe/switch.c static void __activate_traps(struct kvm_vcpu *vcpu) · line 68

Quoted 7 lines
arch/arm64/kvm/hyp/vhe/switch.c +1 -1
@@ -63,10 +63,20 @@ static void __activate_traps(struct kvm_vcpu *vcpu)
 		__activate_traps_fpsimd32(vcpu);
 	}
 
-	if (cpus_have_final_cap(ARM64_SME))
+	if (cpus_have_final_cap(ARM64_SME)) {
 		write_sysreg(read_sysreg(sctlr_el2) & ~SCTLR_ELx_ENTP2,
 			     sctlr_el2);

I still question this. As far as I can tell, it only affects the host context (HCR_EL2.{E2H,TGE}={1,1}).

This is outlined in the description of the HFGWTR_EL2.nTPIDR2_EL0 bit:

<quote>

	If EL2 is implemented and enabled in the current Security
	state, HCR_EL2.{E2H, TGE} != {1, 1}, and either EL3 is
	not implemented or SCR_EL3.FGTEn == 1, then MSR writes
	of TPIDR2_EL0 at EL1 and EL0 using AArch64 are
	trapped to EL2 and reported with EC syndrome value 0x18,
	unless the write generates a higher priority exception.

</quote>

Similar language is there for the read counterpart.

So I can only conclude that messing with SCTLR_EL2 is superfluous and doesn't affect the execution in a guest context.

Quoted 15 lines
arch/arm64/kvm/hyp/vhe/switch.c +10 -0
+		sysreg_clear_set_s(SYS_HFGRTR_EL2,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK,
+				   0);
+		sysreg_clear_set_s(SYS_HFGWTR_EL2,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK,
+				   0);
+	}
+
 	write_sysreg(val, cpacr_el1);
 
 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el1);

arch/arm64/kvm/hyp/vhe/switch.c static void __deactivate_traps(struct kvm_vcpu *vcpu) · line 105

Quoted 9 lines
arch/arm64/kvm/hyp/vhe/switch.c +5 -1
@@ -88,9 +98,21 @@ static void __deactivate_traps(struct kvm_vcpu *vcpu)
 	 */
 	asm(ALTERNATIVE("nop", "isb", ARM64_WORKAROUND_SPECULATIVE_AT));
 
-	if (cpus_have_final_cap(ARM64_SME))
+	if (cpus_have_final_cap(ARM64_SME)) {
+		/*
+		 * Enable access to SMPRI_EL1 - we don't need to
+		 * control nTPIDR2_EL0 in VHE mode.
+		 */

This comment is factually wrong.

arch/arm64/kvm/hyp/vhe/switch.c static void __deactivate_traps(struct kvm_vcpu *vcpu) · line 114

Quoted 9 lines
arch/arm64/kvm/hyp/vhe/switch.c +7 -0
+		sysreg_clear_set_s(SYS_HFGRTR_EL2, 0,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK);
+		sysreg_clear_set_s(SYS_HFGWTR_EL2, 0,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK);
+
 		write_sysreg(read_sysreg(sctlr_el2) | SCTLR_ELx_ENTP2,
 			     sctlr_el2);

Same as above. It serves no purpose AFAICT.

arch/arm64/kvm/hyp/vhe/switch.c static void __deactivate_traps(struct kvm_vcpu *vcpu) · line 118

Quoted 4 lines
arch/arm64/kvm/hyp/vhe/switch.c +1 -0
+	}
 
 	write_sysreg(CPACR_EL1_DEFAULT, cpacr_el1);
 

And then you realise that the VHE and nVHE behaviours are strictly identical and that they should simply be merged. Or rather, that we should make use of the existing infrastructure to enable/disable the trapping of the SME ops when it actually matters: at load/put time for VHE, at enter/exit for nVHE.

I came up with the following patch, compile-tested only.

Thanks,

	M.

From 5c0913a6de47e32b071ea5534319015e0a61db71 Mon Sep 17 00:00:00 2001
From: Marc Zyngier <[email protected]>
Date: Tue, 1 Nov 2022 12:19:51 +0000
Subject: [PATCH] KVM: arm64: Fix SMPRI_EL1/TPIDR2_EL0 trapping on VHE

The trapping of SMPRI_EL1 and TPIDR2_EL0 currently only really work on nVHE, as only this mode uses the fine-grained trapping that controls these two registers.

Move the trapping enable/disable code into __{de,}activate_traps_common(), allowing it to be called when it actually matters on VHE, and remove the flipping of EL2 control for TPIDR2_EL0, which only affects the host access of this register.

Fixes 861262ab8627 ("KVM: arm64: Handle SME host state when running guests")
---
 arch/arm64/kvm/hyp/include/hyp/switch.h | 20 +++++++++++++++++++
 arch/arm64/kvm/hyp/nvhe/switch.c        | 26 -------------------------
 arch/arm64/kvm/hyp/vhe/switch.c         |  8 --------
 3 files changed, 20 insertions(+), 34 deletions(-)
arch/arm64/kvm/hyp/include/hyp/switch.h +20 -0
index 6cbbb6c02f66..3330d1b76bdd 100644
@@ -87,6 +87,17 @@ static inline void __activate_traps_common(struct kvm_vcpu *vcpu)
 
 	vcpu->arch.mdcr_el2_host = read_sysreg(mdcr_el2);
 	write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
+
+	if (cpus_have_final_cap(ARM64_SME)) {
+		sysreg_clear_set_s(SYS_HFGRTR_EL2,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK,
+				   0);
+		sysreg_clear_set_s(SYS_HFGWTR_EL2,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK,
+				   0);
+	}
 }
 
 static inline void __deactivate_traps_common(struct kvm_vcpu *vcpu)
@@ -96,6 +107,15 @@ static inline void __deactivate_traps_common(struct kvm_vcpu *vcpu)
 	write_sysreg(0, hstr_el2);
 	if (kvm_arm_support_pmu_v3())
 		write_sysreg(0, pmuserenr_el0);
+
+	if (cpus_have_final_cap(ARM64_SME)) {
+		sysreg_clear_set_s(SYS_HFGRTR_EL2, 0,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK);
+		sysreg_clear_set_s(SYS_HFGWTR_EL2, 0,
+				   HFGxTR_EL2_nSMPRI_EL1_MASK |
+				   HFGxTR_EL2_nTPIDR2_EL0_MASK);
+	}
 }
 
 static inline void ___activate_traps(struct kvm_vcpu *vcpu)
arch/arm64/kvm/hyp/nvhe/switch.c +0 -26
index 8e9d49a964be..c2cb46ca4fb6 100644
@@ -55,18 +55,6 @@ static void __activate_traps(struct kvm_vcpu *vcpu)
 	write_sysreg(val, cptr_el2);
 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el2);
 
-	if (cpus_have_final_cap(ARM64_SME)) {
-		val = read_sysreg_s(SYS_HFGRTR_EL2);
-		val &= ~(HFGxTR_EL2_nTPIDR2_EL0_MASK |
-			 HFGxTR_EL2_nSMPRI_EL1_MASK);
-		write_sysreg_s(val, SYS_HFGRTR_EL2);
-
-		val = read_sysreg_s(SYS_HFGWTR_EL2);
-		val &= ~(HFGxTR_EL2_nTPIDR2_EL0_MASK |
-			 HFGxTR_EL2_nSMPRI_EL1_MASK);
-		write_sysreg_s(val, SYS_HFGWTR_EL2);
-	}
-
 	if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) {
 		struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt;
 
@@ -110,20 +98,6 @@ static void __deactivate_traps(struct kvm_vcpu *vcpu)
 
 	write_sysreg(this_cpu_ptr(&kvm_init_params)->hcr_el2, hcr_el2);
 
-	if (cpus_have_final_cap(ARM64_SME)) {
-		u64 val;
-
-		val = read_sysreg_s(SYS_HFGRTR_EL2);
-		val |= HFGxTR_EL2_nTPIDR2_EL0_MASK |
-			HFGxTR_EL2_nSMPRI_EL1_MASK;
-		write_sysreg_s(val, SYS_HFGRTR_EL2);
-
-		val = read_sysreg_s(SYS_HFGWTR_EL2);
-		val |= HFGxTR_EL2_nTPIDR2_EL0_MASK |
-			HFGxTR_EL2_nSMPRI_EL1_MASK;
-		write_sysreg_s(val, SYS_HFGWTR_EL2);
-	}
-
 	cptr = CPTR_EL2_DEFAULT;
 	if (vcpu_has_sve(vcpu) && (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED))
 		cptr |= CPTR_EL2_TZ;
arch/arm64/kvm/hyp/vhe/switch.c +0 -8
index 7acb87eaa092..1a97391fedd2 100644
@@ -63,10 +63,6 @@ static void __activate_traps(struct kvm_vcpu *vcpu)
 		__activate_traps_fpsimd32(vcpu);
 	}
 
-	if (cpus_have_final_cap(ARM64_SME))
-		write_sysreg(read_sysreg(sctlr_el2) & ~SCTLR_ELx_ENTP2,
-			     sctlr_el2);
-
 	write_sysreg(val, cpacr_el1);
 
 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el1);
@@ -88,10 +84,6 @@ static void __deactivate_traps(struct kvm_vcpu *vcpu)
 	 */
 	asm(ALTERNATIVE("nop", "isb", ARM64_WORKAROUND_SPECULATIVE_AT));
 
-	if (cpus_have_final_cap(ARM64_SME))
-		write_sysreg(read_sysreg(sctlr_el2) | SCTLR_ELx_ENTP2,
-			     sctlr_el2);
-
 	write_sysreg(CPACR_EL1_DEFAULT, cpacr_el1);
 
 	if (!arm64_kernel_unmapped_at_el0())
Signature · 5 lines
2.34.1


-- 
Without deviation from the norm, progress is not possible.
author Reviewed-by
#4 lore raw

Re: [PATCH v2 2/2] KVM: arm64: Trap access to SMPRI_EL1 and TPIDR2 in VHE mode

Quoted 11 lines · Marc Zyngier

Review

Quoted 4 lines · Mark Brown
(quoted diff) +1 -1
-	if (cpus_have_final_cap(ARM64_SME))
+	if (cpus_have_final_cap(ARM64_SME)) {
 		write_sysreg(read_sysreg(sctlr_el2) & ~SCTLR_ELx_ENTP2,
 			     sctlr_el2);

I still question this. As far as I can tell, it only affects the host context (HCR_EL2.{E2H,TGE}={1,1}).

This is outlined in the description of the HFGWTR_EL2.nTPIDR2_EL0 bit:

Oh, I see what you meant there - I was purely focusing on the new code with the fine grained traps, not the existing code.

Quoted 2 lines

So I can only conclude that messing with SCTLR_EL2 is superfluous and doesn't affect the execution in a guest context.

Yes, if you look at the pseudocode for TPIDR2_EL0 that's the case. It's either EnTP2 for HCR_EL2.<E2H,TGE> == '11' or the fine grained trap otherwise.

Quoted 6 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+		/*
+		 * Enable access to SMPRI_EL1 - we don't need to
+		 * control nTPIDR2_EL0 in VHE mode.
+		 */

This comment is factually wrong.

Bah, I fixed one but not the other sorry :/

Quoted 13 lines

Date: Tue, 1 Nov 2022 12:19:51 +0000
Subject: [PATCH] KVM: arm64: Fix SMPRI_EL1/TPIDR2_EL0 trapping on VHE

The trapping of SMPRI_EL1 and TPIDR2_EL0 currently only really work on nVHE, as only this mode uses the fine-grained trapping that controls these two registers.

Move the trapping enable/disable code into __{de,}activate_traps_common(), allowing it to be called when it actually matters on VHE, and remove the flipping of EL2 control for TPIDR2_EL0, which only affects the host access of this register.

Quoted 3 lines
Fixes 861262ab8627 ("KVM: arm64: Handle SME host state when running guests")
maintainer
#5 lore raw

Re: [PATCH v2 2/2] KVM: arm64: Trap access to SMPRI_EL1 and TPIDR2 in VHE mode

Quoted 15 lines · Mark Brown
Quoted 13 lines · Marc Zyngier

Date: Tue, 1 Nov 2022 12:19:51 +0000
Subject: [PATCH] KVM: arm64: Fix SMPRI_EL1/TPIDR2_EL0 trapping on VHE

The trapping of SMPRI_EL1 and TPIDR2_EL0 currently only really work on nVHE, as only this mode uses the fine-grained trapping that controls these two registers.

Move the trapping enable/disable code into __{de,}activate_traps_common(), allowing it to be called when it actually matters on VHE, and remove the flipping of EL2 control for TPIDR2_EL0, which only affects the host access of this register.

Quoted 6 lines
Quoted 3 lines
Fixes 861262ab8627 ("KVM: arm64: Handle SME host state when running guests")

Thanks. I've now applied the first patch and this one to kvmarm/fixes.

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
maintainer Applied
#6 lore raw

Re: (subset) [PATCH v2 0/2] arm64/sme: Fix SMPRI_EL1 traps for KVM guests

In reply to Mark Brown
Quoted 15 lines · Mark Brown

When adding support for SME I missed management of traps for SMPRI_EL1 in KVM guests, and also the need to manage TPIDR2_EL0 fine grained traps for EL1. This series adds that management, and also updates the documentation to help avoid such issues in future.

This was noticed due to Vincent Donnefort noticing that qemu has implemented support for SME virtualisation but not fine grained traps which was causing problems for qemu in nVHE mode since that already used fine grained traps with SME, the FGT registers are not implemented and UNDEF. As a result of this change VHE will also be affected so a separate series "KVM: arm: Refuse to enable KVM on systems with SME but not FGT" has been sent which will ensure we don't run on such systems though initial review suggests that this will likely be treated as a qemu documentation issue.

Quoted 1 line

[...]

Applied to fixes, thanks!

[1/2] arm64: booting: Document our requirements for fine grained traps with SME

      commit: be0ddf5293a7895a8c9096e1a8560930c6a0ab3f

Cheers,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.