* [PATCH 1/3] KVM: arm64: nv: Forward PIR_EL2, PIRE0_EL2 and POR_EL2 traps
2026-11-18 15:29 [PATCH 0/3] KVM: arm64: nv: trap forwarding bugs Dan Carpenter
@ 2026-11-18 15:29 ` Dan Carpenter
2026-11-18 15:30 ` [PATCH 2/3] KVM: arm64: nv: Forward CNTP_TVAL_EL02 and CNTV_TVAL_EL02 traps Dan Carpenter
2026-11-18 15:30 ` [PATCH 3/3] KVM: arm64: nv: Report EL2 registers with no trap forwarding Dan Carpenter
2 siblings, 0 replies; 4+ messages in thread
From: Dan Carpenter @ 2026-11-18 15:29 UTC (permalink / raw)
To: Marc Zyngier
Cc: Oliver Upton, Fuad Tabba, Joey Gouly, Steffen Eiden,
Suzuki K Poulose, Zenghui Yu, Catalin Marinas, Will Deacon,
Mark Rutland, linux-arm-kernel, kvmarm, Karthik Bhargavan,
Clément Blaudeau
These registers were added to KVM, but not to the encoding_to_cgt[]
table which decides when a trap is forwarded to a nested hypervisor.
The result is that if an L1 hypervisor runs its L2 with HCR_EL2.NV set
(meaning the L2 is itself a hypervisor), and the L2 accesses these
registers, KVM handles the access itself and it goes to the L1's own
registers instead of being emulated on L1. This is quite a serious bug
because it lets the L2 change the permissions of the L1's own stage 1
translations. It could mark everything as no-access and cause a crash
in L1 or mark everything as read-write-execute which weakens the
protections in L1.
Add the three registers with CGT_HCR_NV, MAIR_EL2 and AMAIR_EL2.
Fixes: 874ae1d48e60 ("KVM: arm64: Handle PIR{,E0}_EL2 traps")
Fixes: 5970e9903f03 ("KVM: arm64: Add basic support for POR_EL2")
Cc: stable@vger.kernel.org
Signed-off-by: Dan Carpenter <dan@pereq.eu>
---
arch/arm64/kvm/emulate-nested.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/arch/arm64/kvm/emulate-nested.c b/arch/arm64/kvm/emulate-nested.c
index 2bec81668361..bf4b8a44b9c4 100644
--- a/arch/arm64/kvm/emulate-nested.c
+++ b/arch/arm64/kvm/emulate-nested.c
@@ -889,6 +889,9 @@ static const struct encoding_to_trap_config encoding_to_cgt[] __initconst = {
SR_TRAP(SYS_PMSCR_EL2, CGT_HCR_NV),
SR_TRAP(SYS_MAIR_EL2, CGT_HCR_NV),
SR_TRAP(SYS_AMAIR_EL2, CGT_HCR_NV),
+ SR_TRAP(SYS_PIR_EL2, CGT_HCR_NV),
+ SR_TRAP(SYS_PIRE0_EL2, CGT_HCR_NV),
+ SR_TRAP(SYS_POR_EL2, CGT_HCR_NV),
SR_TRAP(SYS_MPAMHCR_EL2, CGT_HCR_NV),
SR_TRAP(SYS_MPAMVPMV_EL2, CGT_HCR_NV),
SR_TRAP(SYS_MPAM2_EL2, CGT_HCR_NV),
--
2.53.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* [PATCH 2/3] KVM: arm64: nv: Forward CNTP_TVAL_EL02 and CNTV_TVAL_EL02 traps
2026-11-18 15:29 [PATCH 0/3] KVM: arm64: nv: trap forwarding bugs Dan Carpenter
2026-11-18 15:29 ` [PATCH 1/3] KVM: arm64: nv: Forward PIR_EL2, PIRE0_EL2 and POR_EL2 traps Dan Carpenter
@ 2026-11-18 15:30 ` Dan Carpenter
2026-11-18 15:30 ` [PATCH 3/3] KVM: arm64: nv: Report EL2 registers with no trap forwarding Dan Carpenter
2 siblings, 0 replies; 4+ messages in thread
From: Dan Carpenter @ 2026-11-18 15:30 UTC (permalink / raw)
To: Marc Zyngier
Cc: Oliver Upton, Fuad Tabba, Joey Gouly, Steffen Eiden,
Suzuki K Poulose, Zenghui Yu, Catalin Marinas, Will Deacon,
Mark Rutland, linux-arm-kernel, kvmarm, Karthik Bhargavan,
Clément Blaudeau
There are a couple timer related entries messing in the nested
virtualization traps.
With nested virtualization you can run a hypervisor in L1 and then you
can run another hypervisor under that one in L2 and a guest under that
in L3. Without these then if L2 tries to read or write L3's timers it
instead reads and writes to its own timers.
Fixes: c271269e3570 ("KVM: arm64: nv: Add trap routing for CNTHCTL_EL2.EL1{NVPCT,NVVCT,TVT,TVCT}")
Cc: stable@vger.kernel.org
Signed-off-by: Dan Carpenter <dan@pereq.eu>
---
arch/arm64/kvm/emulate-nested.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/arch/arm64/kvm/emulate-nested.c b/arch/arm64/kvm/emulate-nested.c
index bf4b8a44b9c4..9e0c15e6be47 100644
--- a/arch/arm64/kvm/emulate-nested.c
+++ b/arch/arm64/kvm/emulate-nested.c
@@ -942,8 +942,10 @@ static const struct encoding_to_trap_config encoding_to_cgt[] __initconst = {
sys_reg(3, 5, 14, 1, 0), CGT_HCR_NV),
SR_TRAP(SYS_CNTP_CTL_EL02, CGT_CNTHCTL_EL1NVPCT),
SR_TRAP(SYS_CNTP_CVAL_EL02, CGT_CNTHCTL_EL1NVPCT),
+ SR_TRAP(SYS_CNTP_TVAL_EL02, CGT_HCR_NV),
SR_TRAP(SYS_CNTV_CTL_EL02, CGT_CNTHCTL_EL1NVVCT),
SR_TRAP(SYS_CNTV_CVAL_EL02, CGT_CNTHCTL_EL1NVVCT),
+ SR_TRAP(SYS_CNTV_TVAL_EL02, CGT_HCR_NV),
SR_TRAP(OP_AT_S1E2R, CGT_HCR_NV),
SR_TRAP(OP_AT_S1E2W, CGT_HCR_NV),
SR_TRAP(OP_AT_S12E1R, CGT_HCR_NV),
--
2.53.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* [PATCH 3/3] KVM: arm64: nv: Report EL2 registers with no trap forwarding
2026-11-18 15:29 [PATCH 0/3] KVM: arm64: nv: trap forwarding bugs Dan Carpenter
2026-11-18 15:29 ` [PATCH 1/3] KVM: arm64: nv: Forward PIR_EL2, PIRE0_EL2 and POR_EL2 traps Dan Carpenter
2026-11-18 15:30 ` [PATCH 2/3] KVM: arm64: nv: Forward CNTP_TVAL_EL02 and CNTV_TVAL_EL02 traps Dan Carpenter
@ 2026-11-18 15:30 ` Dan Carpenter
2 siblings, 0 replies; 4+ messages in thread
From: Dan Carpenter @ 2026-11-18 15:30 UTC (permalink / raw)
To: Marc Zyngier
Cc: Oliver Upton, Fuad Tabba, Joey Gouly, Steffen Eiden,
Suzuki K Poulose, Zenghui Yu, Catalin Marinas, Will Deacon,
Mark Rutland, linux-arm-kernel, kvmarm, Karthik Bhargavan,
Clément Blaudeau
One of the recurring issues we run into is that people add a trap to
KVM but they forget to add a matching trap for nested virtualization.
Add a boot time warning to print a warning for missing traps.
Signed-off-by: Dan Carpenter <dan@pereq.eu>
---
arch/arm64/include/asm/kvm_host.h | 2 +-
arch/arm64/kvm/emulate-nested.c | 5 ++++-
arch/arm64/kvm/sys_regs.c | 23 +++++++++++++++++++++--
3 files changed, 26 insertions(+), 4 deletions(-)
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index 8efb21b2d729..0614c3b3a40d 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -1379,7 +1379,7 @@ void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
int __init kvm_sys_reg_table_init(void);
struct sys_reg_desc;
int __init populate_sysreg_config(const struct sys_reg_desc *sr,
- unsigned int idx);
+ unsigned int idx, bool needs_nv_trap);
int __init populate_nv_trap_config(void);
void kvm_calculate_traps(struct kvm_vcpu *vcpu);
diff --git a/arch/arm64/kvm/emulate-nested.c b/arch/arm64/kvm/emulate-nested.c
index 9e0c15e6be47..39136a89d74b 100644
--- a/arch/arm64/kvm/emulate-nested.c
+++ b/arch/arm64/kvm/emulate-nested.c
@@ -2434,7 +2434,7 @@ int __init populate_nv_trap_config(void)
}
int __init populate_sysreg_config(const struct sys_reg_desc *sr,
- unsigned int idx)
+ unsigned int idx, bool needs_nv_trap)
{
union trap_config tc;
u32 encoding;
@@ -2452,6 +2452,9 @@ int __init populate_sysreg_config(const struct sys_reg_desc *sr,
encoding = sys_reg(sr->Op0, sr->Op1, sr->CRn, sr->CRm, sr->Op2);
tc = get_trap_config(encoding);
+ if (needs_nv_trap && !tc.val)
+ kvm_err("sysreg %s has no NV trap configuration\n", sr->name);
+
if (tc.sri) {
kvm_err("sysreg %s (%d) duplicate entry (%d)\n",
sr->name, idx - 1, tc.sri);
diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
index ef4554656ddf..afd7e76bedff 100644
--- a/arch/arm64/kvm/sys_regs.c
+++ b/arch/arm64/kvm/sys_regs.c
@@ -5978,6 +5978,23 @@ int kvm_vcpu_finalize_sys_regs(struct kvm_vcpu *vcpu)
return 0;
}
+/* Keep the nested virtualization traps in sync with the KVM traps. */
+static bool __init needs_nv_trap(const struct sys_reg_desc *r)
+{
+ /* op1 4 is the EL2 encodings, op1 5 the _EL12 and _EL02 aliases */
+ if (r->Op1 != 4 && r->Op1 != 5)
+ return false;
+
+ if (!r->access ||
+ r->access == bad_vncr_trap ||
+ r->access == trap_raz_wi ||
+ r->access == bad_redir_trap ||
+ r->access == undef_access)
+ return false;
+
+ return true;
+}
+
int __init kvm_sys_reg_table_init(void)
{
const struct sys_reg_desc *gicv3_regs, *gicv5_regs;
@@ -6010,10 +6027,12 @@ int __init kvm_sys_reg_table_init(void)
check_feature_map();
for (i = 0; !ret && i < ARRAY_SIZE(sys_reg_descs); i++)
- ret = populate_sysreg_config(sys_reg_descs + i, i);
+ ret = populate_sysreg_config(sys_reg_descs + i, i,
+ needs_nv_trap(sys_reg_descs + i));
for (i = 0; !ret && i < ARRAY_SIZE(sys_insn_descs); i++)
- ret = populate_sysreg_config(sys_insn_descs + i, i);
+ ret = populate_sysreg_config(sys_insn_descs + i, i,
+ needs_nv_trap(sys_insn_descs + i));
return ret;
}
--
2.53.0
^ permalink raw reply related [flat|nested] 4+ messages in thread