When we exit from a SVE guest we leave the SVE configuration in EL2 as it
was for the guest, only switching back to the host configuration on next
use by the host. This is perhaps a little surprising when encountered in
the code by itself, add comments explaining what is going on both in the
trap handler and when we configure the traps.
Signed-off-by: Mark Brown <[email protected]>
---
Changes in v2:
- Rebase onto v6.9-rc1.
- Link to v1: https://lore.kernel.org/r/[email protected]
---
arch/arm64/include/asm/kvm_emulate.h | 2 ++
arch/arm64/kvm/hyp/nvhe/hyp-main.c | 1 +
2 files changed, 3 insertions(+)
diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
index 975af30af31f..68a5c7b380a0 100644
--- a/arch/arm64/include/asm/kvm_emulate.h
+++ b/arch/arm64/include/asm/kvm_emulate.h
@@ -587,6 +587,7 @@ static __always_inline u64 kvm_get_reset_cptr_el2(struct kvm_vcpu *vcpu)
} else if (has_hvhe()) {
val = (CPACR_EL1_FPEN_EL0EN | CPACR_EL1_FPEN_EL1EN);
+ /* Leave traps enabled, we will restore EL2 lazily */
if (!vcpu_has_sve(vcpu) ||
(vcpu->arch.fp_state != FP_STATE_GUEST_OWNED))
val |= CPACR_EL1_ZEN_EL1EN | CPACR_EL1_ZEN_EL0EN;
@@ -595,6 +596,7 @@ static __always_inline u64 kvm_get_reset_cptr_el2(struct kvm_vcpu *vcpu)
} else {
val = CPTR_NVHE_EL2_RES1;
+ /* Leave traps enabled, we will restore EL2 lazily */
if (vcpu_has_sve(vcpu) &&
(vcpu->arch.fp_state == FP_STATE_GUEST_OWNED))
val |= CPTR_EL2_TZ;
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 2385fd03ed87..84deed83e580 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -420,6 +420,7 @@ void handle_trap(struct kvm_cpu_context *host_ctxt)
handle_host_smc(host_ctxt);
break;
case ESR_ELx_EC_SVE:
+ /* Handle lazy restore of the host VL */
if (has_hvhe())
sysreg_clear_set(cpacr_el1, 0, (CPACR_EL1_ZEN_EL1EN |
CPACR_EL1_ZEN_EL0EN));
---
base-commit: 4cece764965020c22cff7665b18a012006359095
change-id: 20231218-kvm-arm64-nvhe-sve-trap-f65a3dd46a6a
Best regards,
--
Mark Brown <[email protected]>