KVM: arm64: Add support for FEAT_NV2p1 and FEAT_NV3
Marc Zyngier
maintainerv4 cover KVM: arm64: Add support for FEAT_NV2p1 and FEAT_NV3
This is the fourth (and hopefully last) version of a series adding support for two extensions targeting Nested Virt on arm64:
- FEAT_NV2p1 is effectively a bug fix for two registers (CNTHCTL_EL2 and CPTR_EL2) that are missing stateful bits when accessed from EL1 in a NV configuration. When this is present, the hypervisor can avoid a bunch of traps.
- FEAT_NV3 is much more ambitious, and changes the way ERET behaves in a NV environment. By moving EL1 accesses to HCR_EL2 from memory (via VNCR) to a dedicated register (NVHCR_EL2), the HW can detect whether the guest is performing an ERET for itself (NVHCR_EL2.TGE==1) or to its own guest (NVHCR_EL2.TGE==0). In the former case, ERET is done directly, and no trap occurs. Similar optimisations are available for a class of TLBI instructions.
The whole thing has been tested on an FVP model, and shown measurable improvements for an L1 guest (about 1.5% fewer instructions).
Given that this isn't very convincing on its own, I have built an approximate emulation of FEAT_NV3 that L1 (and deeper levels) can use on actual production hardware. For these deeper levels, the numbers are in the double digit of percentage point reduction, see below for details.
Does it make NV better? Yes!
Does it make NV good? Get real!
Anyway, patches on top of -rc4.
* Side notes about performance evaluation with NV3 emulation:
If you are interested in the NV3 emulation, I have a branch [0] that contains it all. Note that the improvement only kicks while running an L2 guest running at EL2 -- L1 will not see any improvement. But I get close to 40% improvements on L2, and 60% on L3 for the only workload that really matters (Debian Installer).
If all your nested guests are running this hack, then you can pass "kvm-arm.mode=nested,nvtge" on the host command line, and all the nested guests will opt in the accelerated behaviour.
If you like to live dangerously (or have a machine that does not have FEAT_HCX), you can instead pass 'kvm-arm.mode=nested,nvtge=force', and ERET will get the acceleration without even guest buy-in.
* From v3 [3]:
- Fixed comment in the Mapping/Fields generator
- Dropped double semi-colon in HCRX configuration
- Simplified the kvm_has_nv3() predicate
- Added NVHCR_EL2 to the get-reg-list selftest
- Collected RBs from Yao Yuan, with thanks
* From v2 [2]:
- Improved comments on the subtleties of trapping ERET in nested contexts
- Rebased on top of v7.2-rc4
- Collected RBs from Yao Yuan, with thanks
* From v1 [1]:
- Expanded generation of RESx and UNKN symbols to both Mapping and Fields qualifiers
- Dropped HFGIRT_EL2.ERET checks on ERET fast-path
- Fixed the NVHCR_EL2 trap handling
- Added FORCE_RESx() to HCRX_EL2 constraints
- Fixed ID_AA64MMFR4_EL1 sanitisation
- Added NV3 acceleration for TLBIOS
- Collected RBs from Joey, with thanks
[0] https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/log/?h=kvm-arm64/nv3
[1] https://lore.kernel.org/r/[email protected]
[2] https://lore.kernel.org/r/[email protected]
[3] https://lore.kernel.org/r/[email protected]
Marc Zyngier (29):
arm64: sysreg: Emit RESx/UNKN values for Mapping/Fields definitions
arm64: Update ID_AA64MMFR4_EL1 description to 2026-03 JSON release
KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
KVM: arm64: Drop __HCRX_EL2_* masks
KVM: arm64: Plumb HCRX_EL2.SRMASKEn in HCRX_EL2 sanitisation
KVM: arm64: Classify CPTR_EL2 as a SR_LOC_SPECIAL register
KVM: arm64: Don't evaluate HCR_EL2.NV nor HFGITR_EL2.ERET on ERET fast
path
arm64: Add ARM64_HAS_NV2P1 capability
KVM: arm64: Relax CPTR_EL2 handling when FEAT_NV2p1 is present
KVM: arm64: Relax CNTHCTL_EL2 handling when FEAT_NV2p1 is present
KVM: arm64: Expose FEAT_NV2p1 to NV guests
arm64: Add FEAT_NV2p1 detection
arm64: sysreg: Add NVHCR_EL2 description as a mirror of HCR_EL2
arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3
arm64: Add ARM64_HAS_NV3 capability
KVM: arm64: Split NV-specific exit fixups from the non-NV handling
KVM: arm64: Add NV3 control bits to HCRX_EL2 sanitisation
KVM: arm64: Add kvm_has_nv{2,3}() predicates
KVM: arm64: Make HCR_EL2 a non-VNCR register
KVM: arm64: Add sanitisation for NVHCR_EL2
KVM: arm64: Add NVHCR_EL2 handling to the sysreg array
KVM: arm64: Add routing for NVHCR_EL2 trap
KVM: arm64: Add NVHCR_EL2 context switching
KVM: arm64: Engage NV3 ERET trap elision
KVM: arm64: Engage NV3 TLBI trap elision
KVM: arm64: Add FEAT_NV3 detection
KVM: arm64: Expose FEAT_NV3 to guests
KVM: arm64: selftest: Add NVHCR_EL2 to get-reg-list
arm64: Add override for ID_AA64MMFR4_EL1.NV_frac
arch/arm64/include/asm/cpufeature.h | 1 + arch/arm64/include/asm/kvm_arm.h | 15 ----- arch/arm64/include/asm/kvm_emulate.h | 44 +++++++++++++- arch/arm64/include/asm/kvm_host.h | 3 +- arch/arm64/include/asm/vncr_mapping.h | 2 +- arch/arm64/kernel/cpufeature.c | 18 +++++- arch/arm64/kernel/image-vars.h | 1 + arch/arm64/kernel/pi/idreg-override.c | 10 ++++ arch/arm64/kvm/arch_timer.c | 10 +++- arch/arm64/kvm/config.c | 27 ++++++++- arch/arm64/kvm/emulate-nested.c | 18 ++++-- arch/arm64/kvm/hyp/include/hyp/switch.h | 27 +++++++-- arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h | 11 ++++ arch/arm64/kvm/hyp/vhe/switch.c | 52 +++++++++++++---- arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 21 +++++-- arch/arm64/kvm/nested.c | 14 ++++- arch/arm64/kvm/sys_regs.c | 57 +++++++++++++++++-- arch/arm64/tools/cpucaps | 2 + arch/arm64/tools/gen-sysreg.awk | 8 +-- arch/arm64/tools/sysreg | 44 ++++++++++++-- .../selftests/kvm/arm64/get-reg-list.c | 2 + 21 files changed, 322 insertions(+), 65 deletions(-)
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 1/29 arm64: sysreg: Emit RESx/UNKN values for Mapping/Fields definitions
The sysreg file is using the Mapping or Fields qualifiers to indicate that a given encoding is only a mapping to a particular register (or an instance of a more generic register definition).
As a result, we don't output any definition, and instead expect the canonical definitions to be used.
This works rather well for individual fields, but creates problems for macros that refer to more generic classes of bits such as RESx.
Relax the above rule by emitting the RESx and UNKN values for Mapping and Fields qualifiers as well.
--- arch/arm64/tools/gen-sysreg.awk | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
arch/arm64/tools/gen-sysreg.awk +4 -4
@@ -228,7 +228,7 @@ $1 == "EndSysreg" && block_current() == "Sysreg" { } # Currently this is effectivey a comment, in future we may want to emit -# defines for the fields. +# defines for the fields. We do emit RESx and UNKN values in any case. ($1 == "Fields" || $1 == "Mapping") && block_current() == "Sysreg" { expect_fields(2)
@@ -239,9 +239,9 @@ $1 == "EndSysreg" && block_current() == "Sysreg" { print "" next_bit = -1 - res0 = null - res1 = null - unkn = null + res0 = $2 "_RES0" + res1 = $2 "_RES1" + unkn = $2 "_UNKN" next }
Signature · 1 line
2.47.3
Re: [PATCH v4 01/29] arm64: sysreg: Emit RESx/UNKN values for Mapping/Fields definitions
Quoted 4 lines · Marc Zyngier
The sysreg file is using the Mapping or Fields qualifiers to indicate that a given encoding is only a mapping to a particular register (or an instance of a more generic register definition).
Quoted 8 lines
As a result, we don't output any definition, and instead expect the canonical definitions to be used.
This works rather well for individual fields, but creates problems for macros that refer to more generic classes of bits such as RESx.
Relax the above rule by emitting the RESx and UNKN values for Mapping and Fields qualifiers as well.
It might be worth stating a bit more explicitly, something like:
Emit RESx and UNKN values for Mapping and Fields qualifiers as well, defining these in terms of the RESx and UNKN values for the canonical Sysreg or SysregFields definition.
For example, the following defintion:
| Sysreg FOO ... | Fields BAR | EndSysreg
... will generate:
| #define FOO_RES0 (BAR_RES0) | #define FOO_RES1 (BAR_RES1) | #define FOO_RES0 (BAR_UNKN)
Regardless of the above, the patch looks reasonable to me, so:
Mark.
Quoted 35 lines
Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/tools/gen-sysreg.awk | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)arch/arm64/tools/gen-sysreg.awk +4
-4@@ -228,7 +228,7 @@ $1 == "EndSysreg" && block_current() == "Sysreg" { } # Currently this is effectivey a comment, in future we may want to emit -# defines for the fields. +# defines for the fields. We do emit RESx and UNKN values in any case. ($1 == "Fields" || $1 == "Mapping") && block_current() == "Sysreg" { expect_fields(2)@@ -239,9 +239,9 @@ $1 == "EndSysreg" && block_current() == "Sysreg" { print "" next_bit = -1 - res0 = null - res1 = null - unkn = null + res0 = $2 "_RES0" + res1 = $2 "_RES1" + unkn = $2 "_UNKN" next }Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 2/29 arm64: Update ID_AA64MMFR4_EL1 description to 2026-03 JSON release
In reply to Marc ZyngierID_AA64MMFR4_EL1 has gained a few fields and enum values in the past few months, so resync its definition with the 2026-03 JSON release.
--- arch/arm64/tools/sysreg | 32 ++++++++++++++++++++++++++++---- 1 file changed, 28 insertions(+), 4 deletions(-)
arch/arm64/tools/sysreg +28 -4
@@ -2386,17 +2386,40 @@ EndEnum EndSysreg Sysreg ID_AA64MMFR4_EL1 3 0 0 7 4 -Res0 63:48 -UnsignedEnum 47:44 SRMASK +UnsignedEnum 63:60 MTEFGT + 0b0000 NI + 0b0001 IMP +EndEnum +UnsignedEnum 59:56 SCRX + 0b0000 NI + 0b0001 IMP +EndEnum +UnsignedEnum 55:52 TEV + 0b0000 NI + 0b0001 IMP +EndEnum +UnsignedEnum 51:48 TPS + 0b0000 VAL_0000 + 0b0001 VAL_0001 + 0b0010 VAL_0010 +EndEnum +UnsignedEnum 47:44 SRMASK + 0b0000 NI + 0b0001 IMP + 0b0010 SRMASK2 +EndEnum +UnsignedEnum 43:40 TLBID 0b0000 NI 0b0001 IMP EndEnum -Res0 43:40 UnsignedEnum 39:36 E3DSE 0b0000 NI 0b0001 IMP EndEnum -Res0 35:32 +UnsignedEnum 35:32 EAESR + 0b0000 NI + 0b0001 IMP +EndEnum UnsignedEnum 31:28 RMEGDI 0b0000 NI 0b0001 IMP
@@ -2410,6 +2433,7 @@ UnsignedEnum 23:20 NV_frac 0b0000 NV_NV2 0b0001 NV2_ONLY 0b0010 NV2P1 + 0b0011 NV3 EndEnum UnsignedEnum 19:16 FGWTE3 0b0000 NI
Signature · 1 line
2.47.3
Re: [PATCH v4 02/29] arm64: Update ID_AA64MMFR4_EL1 description to 2026-03 JSON release
Quoted 4 lines · Marc Zyngier
ID_AA64MMFR4_EL1has gained a few fields and enum values in the past few months, so resync its definition with the 2026-03 JSON release.Signed-off-by Marc Zyngier <[email protected]>
Checked the values.
Quoted 64 lines
--- arch/arm64/tools/sysreg | 32 ++++++++++++++++++++++++++++---- 1 file changed, 28 insertions(+), 4 deletions(-)arch/arm64/tools/sysreg +28
-4@@ -2386,17 +2386,40 @@ EndEnum EndSysreg Sysreg ID_AA64MMFR4_EL1 3 0 0 7 4 -Res0 63:48 -UnsignedEnum 47:44 SRMASK +UnsignedEnum 63:60 MTEFGT + 0b0000 NI + 0b0001 IMP +EndEnum +UnsignedEnum 59:56 SCRX + 0b0000 NI + 0b0001 IMP +EndEnum +UnsignedEnum 55:52 TEV + 0b0000 NI + 0b0001 IMP +EndEnum +UnsignedEnum 51:48 TPS + 0b0000 VAL_0000 + 0b0001 VAL_0001 + 0b0010 VAL_0010 +EndEnum +UnsignedEnum 47:44 SRMASK + 0b0000 NI + 0b0001 IMP + 0b0010 SRMASK2 +EndEnum +UnsignedEnum 43:40 TLBID 0b0000 NI 0b0001 IMP EndEnum -Res0 43:40 UnsignedEnum 39:36 E3DSE 0b0000 NI 0b0001 IMP EndEnum -Res0 35:32 +UnsignedEnum 35:32 EAESR + 0b0000 NI + 0b0001 IMP +EndEnum UnsignedEnum 31:28 RMEGDI 0b0000 NI 0b0001 IMP@@ -2410,6 +2433,7 @@ UnsignedEnum 23:20 NV_frac 0b0000 NV_NV2 0b0001 NV2_ONLY 0b0010 NV2P1 + 0b0011 NV3 EndEnum UnsignedEnum 19:16 FGWTE3 0b0000 NISignature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 3/29 KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
In reply to Marc ZyngierThe way we merge the guest-provided HCRX_EL2 value with the host's is bonkers. We try to make it look like the FGT registers by using positive and negative polarities for traps, but most of these bits are not strictly about trapping, as they actively change the way some architectural state is managed.
It would be far better to deal with these bits like we do for HCR_EL2, by enumerating the list of bits we don't allow the guest to override. This is simplified by the fact that HCRX_EL2 only affects EL1, and not EL2.
Re-jig the HCRX_EL2 handling with a macro that list the bits excluded from the merge (TMEA, PTTWI, EnIDCP128).
--- arch/arm64/kvm/hyp/include/hyp/switch.h | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-)
arch/arm64/kvm/hyp/include/hyp/switch.h +20 -2
@@ -325,6 +325,24 @@ static inline void __deactivate_traps_mpam(void)write_sysreg_s (MPAMHCR_HOST_FLAGS ,SYS_MPAMHCR_EL2 ); } +/* +* Just like for HCR_EL2, we can't let the guest mess with some of the +* basics we rely on in HCRX_EL2. However, the major difference is that +* HCRX_EL2 only affects EL1, and never EL2 (sudden outburst of sanity, I +* guess). So it is always the guest inflicting it on its own guestx. +* +* Things we don't want to let the guest control are: +* +* - TMEA: That's for us to decide how an SEA is routed, not the guest. +* +* - PTTWI: Similarly, it is for us to decide whether Reduced Coherency for +* the PTW is a thing. It really isn't. +* +* - EnIDCP128: We don't allow IMPDEF sysregs -- full stop. +*/ +#define NV_HCRX_GUEST_EXCLUDE (HCRX_EL2_TMEA | HCRX_EL2_PTTWI | \ + HCRX_EL2_EnIDCP128) +static inline void __activate_traps_common (struct kvm_vcpu *vcpu) {struct kvm_cpu_context *hctxt =host_data_ptr (host_ctxt);
@@ -350,8 +368,8 @@ static inline void __activate_traps_common(struct kvm_vcpu *vcpu)u64 hcrx = vcpu->arch.hcrx_el2;if (is_nested_ctxt (vcpu)) {u64 val =__vcpu_sys_reg (vcpu,HCRX_EL2 ); - hcrx |= val & __HCRX_EL2_MASK; - hcrx &= ~(~val & __HCRX_EL2_nMASK); + hcrx |= (val & ~NV_HCRX_GUEST_EXCLUDE ); + hcrx &= ~(~val & ~NV_HCRX_GUEST_EXCLUDE ); }ctxt_sys_reg (hctxt,HCRX_EL2 ) =read_sysreg_s (SYS_HCRX_EL2 );
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 03/29] KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Pre-existing issues:
- [High] The NV_HCRX_GUEST_EXCLUDE macro fails to exclude HCRX_EL2_MCE2, allowing a nested hypervisor to break KVM's transparent MOPS migration handling. --
--- Patch [3]: [PATCH v4 03/29] KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE --- commit 32a9437b5c31bcb57be7ba071a094800d7980587
Author: Marc Zyngier <[email protected]>
KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
This commit refactors HCRX_EL2 handling by introducing a macro that explicitly lists the bits excluded from the merge, preventing the nested guest from overriding them.
Quoted 4 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +0
-0
[ ... ]
Quoted 11 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +7
-0@@ -325,6 +325,24 @@ static inline void __deactivate_traps_mpam(void)write_sysreg_s (MPAMHCR_HOST_FLAGS ,SYS_MPAMHCR_EL2 ); } +/* +* Just like for HCR_EL2, we can't let the guest mess with some of the +* basics we rely on in HCRX_EL2. However, the major difference is that +* HCRX_EL2 only affects EL1, and never EL2 (sudden outburst of sanity, I +* guess). So it is always the guest inflicting it on its own guestx. +* +* Things we don't want to let the guest control are:
arch/arm64/kvm/hyp/include/hyp/switch.h static inline void __deactivate_traps_mpam(void) · line 344
Quoted 10 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +10
-0+* +* - TMEA: That's for us to decide how an SEA is routed, not the guest. +* +* - PTTWI: Similarly, it is for us to decide whether Reduced Coherency for +* the PTW is a thing. It really isn't. +* +* - EnIDCP128: We don't allow IMPDEF sysregs -- full stop. +*/ +#define NV_HCRX_GUEST_EXCLUDE (HCRX_EL2_TMEA | HCRX_EL2_PTTWI | \ + HCRX_EL2_EnIDCP128)
[Severity: High]
This isn't a bug introduced by this patch, but since NV_HCRX_GUEST_EXCLUDE enumerates the bits the nested guest is not allowed to override, should HCRX_EL2_MCE2 be excluded here as well?
KVM (L0) relies on intercepting MOPS exceptions (via HCRX_EL2.MCE2=1) to handle vCPU physical migrations transparently. If L0 merges a value from the L1 hypervisor that sets MCE2 to 0 (which happens naturally if L1 does not support MOPS), and L0 then migrates the physical CPU while L2 is executing a MOPS instruction, the MOPS exception could trap to EL1 (L2) instead of EL2 (L0).
Would this cause the nested guest kernel to crash with an unexpected exception, potentially breaking stability during host migrations?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
Marc Zyngier
maintainerRe: [PATCH v4 03/29] KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
Quoted 51 lines · On Thu, 30 Jul 2026 08:29:29 +0100, [email protected] wrote:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Pre-existing issues:
- [High] TheNV_HCRX_GUEST_EXCLUDEmacro fails to excludeHCRX_EL2_MCE2, allowing a nested hypervisor to break KVM's transparent MOPS migration handling. ----- Patch [3]: [PATCH v4 03/29] KVM: arm64: Merge guest's
HCRX_EL2usingNV_HCRX_GUEST_EXCLUDE--- commit32a9437b5c31bcb57be7ba071a094800d7980587
Author: Marc Zyngier <[email protected]>KVM: arm64: Merge guest's
HCRX_EL2usingNV_HCRX_GUEST_EXCLUDEThis commit refactors
HCRX_EL2handling by introducing a macro that explicitly lists the bits excluded from the merge, preventing the nested guest from overriding them.Quoted 4 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +0
-0[ ... ]
Quoted 11 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +7
-0@@ -325,6 +325,24 @@ static inline void __deactivate_traps_mpam(void)write_sysreg_s (MPAMHCR_HOST_FLAGS ,SYS_MPAMHCR_EL2 ); } +/* +* Just like for HCR_EL2, we can't let the guest mess with some of the +* basics we rely on in HCRX_EL2. However, the major difference is that +* HCRX_EL2 only affects EL1, and never EL2 (sudden outburst of sanity, I +* guess). So it is always the guest inflicting it on its own guestx. +* +* Things we don't want to let the guest control are: arch/arm64/kvm/hyp/include/hyp/switch.h static inline void __deactivate_traps_mpam(void) · line 344
Quoted 10 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +10
-0+* +* - TMEA: That's for us to decide how an SEA is routed, not the guest. +* +* - PTTWI: Similarly, it is for us to decide whether Reduced Coherency for +* the PTW is a thing. It really isn't. +* +* - EnIDCP128: We don't allow IMPDEF sysregs -- full stop. +*/ +#define NV_HCRX_GUEST_EXCLUDE (HCRX_EL2_TMEA | HCRX_EL2_PTTWI | \ + HCRX_EL2_EnIDCP128)[Severity: High]
This isn't a bug introduced by this patch, but sinceNV_HCRX_GUEST_EXCLUDEenumerates the bits the nested guest is not allowed to override, shouldHCRX_EL2_MCE2be excluded here as well?
Quoted 4 lines
KVM (L0) relies on intercepting MOPS exceptions (via HCRX_EL2.MCE2=1) to handle vCPU physical migrations transparently. If L0 merges a value from the L1 hypervisor that sets MCE2 to 0 (which happens naturally if L1 does not support MOPS),
If L1 doesn't support MOPS, then HCRX_EL2.MCEn==0, and the instructions UNDEF.
Quoted 3 lines
and L0 then migrates the physical CPU while L2 is executing a MOPS instruction, the MOPS exception could trap to EL1 (L2) instead of EL2 (L0).
No, the instruction will have UNDEF in L2.
Furthermore, trapping non-matching CPUs is a guest decision, and L0 is not in charge of deciding this. However, what is missing is the redirection of the exception to L1 when it fires.
Something else to fix at some point.
M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
Marc Zyngier
maintainerRe: [PATCH v4 03/29] KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
Quoted 5 lines · Marc Zyngier
Furthermore, trapping non-matching CPUs is a guest decision, and L0 is not in charge of deciding this. However, what is missing is the redirection of the exception to L1 when it fires.
I've changed my tune on this.
Flipping between L0 and L1 to fixup the brokenness of the architecture is not great, because the vcpu could have migrated in the interval and, and be subjected to another similar exception. This results in forward-progress issues on top of the obvious performance overhead.
I'll add something like this, either as a separate patch or as a fixup if I have to respin the series.
M.
arch/arm64/kvm/hyp/include/hyp/switch.h +6 -2
@@ -330,7 +330,7 @@ static inline void __deactivate_traps_mpam(void)* Just like for HCR_EL2, we can't let the guest mess with some of the * basics we rely on in HCRX_EL2. However, the major difference is that * HCRX_EL2 only affects EL1, and never EL2 (sudden outburst of sanity, I -* guess). So it is always the guest inflicting it on its own guestx. +* guess). So it is always the guest inflicting it on its own guest. * * Things we don't want to let the guest control are: *
@@ -339,10 +339,14 @@ static inline void __deactivate_traps_mpam(void)* - PTTWI: Similarly, it is for us to decide whether Reduced Coherency for * the PTW is a thing. It really isn't. * +* - MCE2: FEAT_MOPS generated exceptions for Option-{A,B} interoperability +* are always handled on the host and not forwarded to a guest hypervisor +* in order to give the (poor) illusion of a homogeneous system. +* * - EnIDCP128: We don't allow IMPDEF sysregs -- full stop. */ #define NV_HCRX_GUEST_EXCLUDE (HCRX_EL2_TMEA | HCRX_EL2_PTTWI | \ - HCRX_EL2_EnIDCP128) + HCRX_EL2_MCE2 | HCRX_EL2_EnIDCP128)static inline void __activate_traps_common (struct kvm_vcpu *vcpu) {
Signature · 1 line
Without deviation from the norm, progress is not possible.
Re: [PATCH v4 03/29] KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE
In reply to Marc ZyngierQuoted 25 lines · Marc Zyngier
The way we merge the guest-provided
HCRX_EL2value with the host's is bonkers. We try to make it look like the FGT registers by using positive and negative polarities for traps, but most of these bits are not strictly about trapping, as they actively change the way some architectural state is managed.It would be far better to deal with these bits like we do for
HCR_EL2, by enumerating the list of bits we don't allow the guest to override. This is simplified by the fact thatHCRX_EL2only affects EL1, and not EL2.Re-jig the
HCRX_EL2handling with a macro that list the bits excluded from the merge (TMEA, PTTWI, EnIDCP128).Reviewed-by Yuan Yao <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/kvm/hyp/include/hyp/switch.h | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-)arch/arm64/kvm/hyp/include/hyp/switch.h +0
-0
arch/arm64/kvm/hyp/include/hyp/switch.h static inline void __deactivate_traps_mpam(void) · line 332
Quoted 8 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +5
-0@@ -325,6 +325,24 @@ static inline void __deactivate_traps_mpam(void)write_sysreg_s (MPAMHCR_HOST_FLAGS ,SYS_MPAMHCR_EL2 ); } +/* +* Just like for HCR_EL2, we can't let the guest mess with some of the +* basics we rely on in HCRX_EL2. However, the major difference is that +* HCRX_EL2 only affects EL1, and never EL2 (sudden outburst of sanity, I +* guess). So it is always the guest inflicting it on its own guestx.
Unsure if "guestx" is a typo or a funny way of saying guest's HCRX_EL2!
Quoted 17 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +13
-0+* +* Things we don't want to let the guest control are: +* +* - TMEA: That's for us to decide how an SEA is routed, not the guest. +* +* - PTTWI: Similarly, it is for us to decide whether Reduced Coherency for +* the PTW is a thing. It really isn't. +* +* - EnIDCP128: We don't allow IMPDEF sysregs -- full stop. +*/ +#define NV_HCRX_GUEST_EXCLUDE (HCRX_EL2_TMEA | HCRX_EL2_PTTWI | \ + HCRX_EL2_EnIDCP128) +static inline void __activate_traps_common (struct kvm_vcpu *vcpu) {struct kvm_cpu_context *hctxt =host_data_ptr (host_ctxt);
arch/arm64/kvm/hyp/include/hyp/switch.h static inline void __activate_traps_common(struct kvm_vcpu *vcpu) · line 375
Quoted 10 lines
arch/arm64/kvm/hyp/include/hyp/switch.h +2
-2@@ -350,8 +368,8 @@ static inline void __activate_traps_common(struct kvm_vcpu *vcpu)u64 hcrx = vcpu->arch.hcrx_el2;if (is_nested_ctxt (vcpu)) {u64 val =__vcpu_sys_reg (vcpu,HCRX_EL2 ); - hcrx |= val & __HCRX_EL2_MASK; - hcrx &= ~(~val & __HCRX_EL2_nMASK); + hcrx |= (val & ~NV_HCRX_GUEST_EXCLUDE ); + hcrx &= ~(~val & ~NV_HCRX_GUEST_EXCLUDE ); }ctxt_sys_reg (hctxt,HCRX_EL2 ) =read_sysreg_s (SYS_HCRX_EL2 );
Marc Zyngier
maintainerPATCH v4 4/29 KVM: arm64: Drop __HCRX_EL2_* masks
In reply to Marc ZyngierThe __HCRX_EL2_* masks are a leftover from a time where we didn't have much sanitisation for the system registers. Since we are now in a better place, rely on the existing checks to detect unhandled bits in HCRX_EL2.
--- arch/arm64/include/asm/kvm_arm.h | 15 --------------- arch/arm64/kvm/config.c | 5 +++-- arch/arm64/kvm/emulate-nested.c | 5 ----- 3 files changed, 3 insertions(+), 22 deletions(-)
arch/arm64/include/asm/kvm_arm.h +0 -15
@@ -287,21 +287,6 @@ GENMASK(19, 18) | \ GENMASK(15, 0)) -/* -* Polarity masks for HCRX_EL2, limited to the bits that we know about -* at this point in time. It doesn't mean that we actually *handle* -* them, but that at least those that are not advertised to a guest -* will be RES0 for that guest. -*/ -#define __HCRX_EL2_MASK (BIT_ULL(6)) -#define __HCRX_EL2_nMASK (GENMASK_ULL(24, 14) | \ - GENMASK_ULL(11, 7) | \ - GENMASK_ULL(5, 0)) -#define __HCRX_EL2_RES0 ~(__HCRX_EL2_nMASK | __HCRX_EL2_MASK) -#define __HCRX_EL2_RES1 ~(__HCRX_EL2_nMASK | \ - __HCRX_EL2_MASK | \ - __HCRX_EL2_RES0) -/* Hyp Prefetch Fault Address Register (HPFAR/HDFAR) */ #define HPFAR_MASK (~UL(0xf))/*
arch/arm64/kvm/config.c +3 -2
@@ -930,10 +930,12 @@ static const struct reg_bits_to_feat_map hcrx_feat_map[] = {NEEDS_FEAT (HCRX_EL2_EnASR,FEAT_LS64_V ),NEEDS_FEAT (HCRX_EL2_EnALS,FEAT_LS64 ),NEEDS_FEAT (HCRX_EL2_EnAS0,FEAT_LS64_ACCDATA ), +FORCE_RES0 (HCRX_EL2_RES0 ), + FORCE_RES1(HCRX_EL2_RES1), }; -static const DECLARE_FEAT_MAP(hcrx_desc, __HCRX_EL2, +static const DECLARE_FEAT_MAP(hcrx_desc, HCRX_EL2, hcrx_feat_map, FEAT_HCX); static const struct reg_bits_to_feat_map hcr_feat_map[] = {
@@ -1579,7 +1581,6 @@ struct resx get_reg_fixed_bits(struct kvm *kvm, enum vcpu_sysreg reg)break ;case HCRX_EL2 : resx =compute_reg_resx_bits (kvm, &hcrx_desc,0 ,0 ); - resx.res1 |= __HCRX_EL2_RES1;break ;case HCR_EL2 : resx =compute_reg_resx_bits (kvm, &hcr_desc,0 ,0 );
arch/arm64/kvm/emulate-nested.c +0 -5
@@ -2320,7 +2320,6 @@ int __init populate_nv_trap_config(void)BUILD_BUG_ON (__NR_CGT_GROUP_IDS__ >BIT (TC_CGT_BITS ));BUILD_BUG_ON (__NR_FGT_GROUP_IDS__ >BIT (TC_FGT_BITS ));BUILD_BUG_ON (__NR_FG_FILTER_IDS__ >BIT (TC_FGF_BITS )); -BUILD_BUG_ON (__HCRX_EL2_MASK & __HCRX_EL2_nMASK);for (int i =0 ; i <ARRAY_SIZE (encoding_to_cgt); i++) {const struct encoding_to_trap_config *cgt = &encoding_to_cgt[i];
@@ -2346,10 +2345,6 @@ int __init populate_nv_trap_config(void) } } -if (__HCRX_EL2_RES0 !=HCRX_EL2_RES0 ) -kvm_info ("Sanitised HCR_EL2_RES0 = %016llx, expecting %016llx\n" , - __HCRX_EL2_RES0,HCRX_EL2_RES0 ); -kvm_info ("nv: %ld coarse grained trap handlers\n" ,ARRAY_SIZE (encoding_to_cgt));
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 5/29 KVM: arm64: Plumb HCRX_EL2.SRMASKEn in HCRX_EL2 sanitisation
In reply to Marc ZyngierHCRX_EL2.SRMASKEn is a new bit enabling FEAT_SRMASK for a guest. We don't plan to support it any time soon, but it doesn't hurt to actively document it, specially as we are going to add more bits we actually care about.
--- arch/arm64/kvm/config.c | 1 + arch/arm64/tools/sysreg | 4 +++- 2 files changed, 4 insertions(+), 1 deletion(-)
arch/arm64/kvm/config.c +1 -0
@@ -904,6 +904,7 @@ static const DECLARE_FEAT_MAP_FGT(hdfgwtr2_desc, hdfgwtr2_masks,static const struct reg_bits_to_feat_map hcrx_feat_map[] = { +NEEDS_FEAT (HCRX_EL2_SRMASKEn,FEAT_SRMASK ),NEEDS_FEAT (HCRX_EL2_PACMEn, feat_pauth_lr),NEEDS_FEAT (HCRX_EL2_EnFPM,FEAT_FPMR ),NEEDS_FEAT (HCRX_EL2_GCSEn,FEAT_GCS ),
arch/arm64/tools/sysreg +3 -1
@@ -4545,7 +4545,9 @@ Fields ZCR_ELx EndSysreg Sysreg HCRX_EL2 3 4 1 2 2 -Res0 63:25 +Res0 63:27 +Field 26 SRMASKEn +Res0 25 Field 24 PACMEn Field 23 EnFPM Field 22 GCSEn
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 6/29 KVM: arm64: Classify CPTR_EL2 as a SR_LOC_SPECIAL register
In reply to Marc ZyngierIt may not be obvious unless you look at it closely, but CPTR_EL2 is treated very differently from other registers. It is one the registers that, despite looking very similar between EL1 and EL2 when E2H==1, have RES0 bits that get in the way.
Make it clear that CPTR_EL2 is odd by classifying it as SR_LOC_SPECIAL, just like CNTHCTL_EL2 (and for the same reasons). This makes it possible to use vcpu_read_sys_reg() with it, and will be necessary once we support FEAT_NV2P1.
--- arch/arm64/include/asm/kvm_emulate.h | 2 +- arch/arm64/kvm/sys_regs.c | 20 ++++++++++++++++++-- 2 files changed, 19 insertions(+), 3 deletions(-)
arch/arm64/include/asm/kvm_emulate.h +1 -1
@@ -617,7 +617,7 @@ static __always_inline void kvm_incr_pc(struct kvm_vcpu *vcpu)*/ static inline u64 vcpu_sanitised_cptr_el2 (const struct kvm_vcpu *vcpu) { -u64 cptr =__vcpu_sys_reg (vcpu,CPTR_EL2 ); +u64 cptr =vcpu_read_sys_reg (vcpu,CPTR_EL2 );if (!vcpu_el2_e2h_is_set (vcpu)) cptr =translate_cptr_el2_to_cpacr_el1 (cptr);
arch/arm64/kvm/sys_regs.c +18 -2
@@ -183,8 +183,6 @@ static void locate_register(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg,switch (reg) {MAPPED_EL2_SYSREG (SCTLR_EL2 ,SCTLR_EL1 , translate_sctlr_el2_to_sctlr_el1 ); -MAPPED_EL2_SYSREG (CPTR_EL2 ,CPACR_EL1 , - translate_cptr_el2_to_cpacr_el1 );MAPPED_EL2_SYSREG (TTBR0_EL2 ,TTBR0_EL1 , translate_ttbr0_el2_to_ttbr0_el1 );MAPPED_EL2_SYSREG (TTBR1_EL2 ,TTBR1_EL1 ,NULL );
@@ -210,6 +208,19 @@ static void locate_register(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg, loc->loc = ((is_hyp_ctxt (vcpu) &&vcpu_el2_e2h_is_set (vcpu)) ?SR_LOC_SPECIAL :SR_LOC_MEMORY );break ; +case CPTR_EL2 : +/* +* CPTR_EL2 is just as special, and needs a certain amount +* of handholding. It always lives in memory, due to being +* heavily trapped thanks to CPACR_EL1.TCPAC being RES0. +* FEAT_NV2p1 fixes this. +*/ +locate_mapped_el2_register (vcpu,CPTR_EL2 ,CPACR_EL1 , + translate_cptr_el2_to_cpacr_el1, + loc); +if (is_hyp_ctxt (vcpu) &&vcpu_el2_e2h_is_set (vcpu)) + loc->loc =SR_LOC_SPECIAL ; +break ;default : loc->loc =locate_direct_register (vcpu, reg); }
@@ -314,6 +325,8 @@ u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg) val &=CNTKCTL_VALID_BITS ; val |=__vcpu_sys_reg (vcpu, reg) & ~CNTKCTL_VALID_BITS ;return val; +case CPTR_EL2 : +return __vcpu_sys_reg (vcpu, reg);default :WARN_ON_ONCE (1 ); }
@@ -359,6 +372,9 @@ void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, enum vcpu_sysreg reg)*/ write_sysreg_el1 (val,SYS_CNTKCTL );break ; +case CPTR_EL2 : +write_sysreg_el1 (val,SYS_CPACR ); +break ;default :WARN_ON_ONCE (1 ); }
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 7/29 KVM: arm64: Don't evaluate HCR_EL2.NV nor HFGITR_EL2.ERET on ERET fast path
In reply to Marc ZyngierWe currently avoid using the ERET fast path if the guest has HCR_EL2.NV set. This is an odd check, as NV doesn't mean much if HCR_EL2.TGE==1.
Similarly, evaluating HFGITR_EL2.ERET makes little sense, as this only applies to the nested context, while the ERET fast-path is purely for the benefit of L1.
Replace these bizarre checks with is_nested_ctxt() which makes a lot more sense: if we are running an L2, the ERET trap must go to L1.
--- arch/arm64/kvm/hyp/vhe/switch.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-)
arch/arm64/kvm/hyp/vhe/switch.c +7 -5
@@ -344,13 +344,15 @@ static bool kvm_hyp_handle_eret(struct kvm_vcpu *vcpu, u64 *exit_code)* if this is a VHE guest hypervisor returning to its own * userspace, or the hypervisor performing a local exception * return. No need to save/restore registers, no need to -* switch S2 MMU. Just do the canonical ERET. +* switch S2 MMU. Just do the canonical ERET unless we are in +* nested context. * -* Unless the trap has to be forwarded further down the line, -* of course... +* Note that this is made possible because KVM itself never traps +* ERET when running an L2. The consequence is that any ERET trap is +* the result of HCR_EL2 or HFGITR_EL2 programming by L1 for its own +* guest, and the exception must be forwarded to L1. */ -if ((__vcpu_sys_reg (vcpu,HCR_EL2 ) &HCR_NV ) || - (__vcpu_sys_reg (vcpu,HFGITR_EL2 ) &HFGITR_EL2_ERET )) +if (is_nested_ctxt (vcpu))return false; spsr =read_sysreg_el1 (SYS_SPSR );
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 8/29 arm64: Add ARM64_HAS_NV2P1 capability
In reply to Marc ZyngierAs we're about to deal with FEAT_NV2P1, add a new capability that will be used to key any support for it.
--- arch/arm64/tools/cpucaps | 1 + 1 file changed, 1 insertion(+)
arch/arm64/tools/cpucaps +1 -0
@@ -51,6 +51,7 @@ HAS_LS64_V HAS_LSUI HAS_MOPS HAS_NESTED_VIRT +HAS_NV2P1 HAS_BBML2_NOABORT HAS_PAN HAS_PMUV3
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 9/29 KVM: arm64: Relax CPTR_EL2 handling when FEAT_NV2p1 is present
In reply to Marc ZyngierWith FEAT_NV2P1, it is no longer necessary to trap CPTR_EL2 accesses via CPACR_EL1, as CPACR_EL1.TCPAC is guaranteed to be stateful.
Prevent such trapping and context switch CPACTR_EL1 in NV contexts when NV2P1 is present.
--- arch/arm64/kvm/hyp/include/hyp/switch.h | 5 +++-- arch/arm64/kvm/hyp/vhe/switch.c | 3 +++ arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 8 +++++--- arch/arm64/kvm/sys_regs.c | 5 ++++- 4 files changed, 15 insertions(+), 6 deletions(-)
arch/arm64/kvm/hyp/include/hyp/switch.h +3 -2
@@ -108,9 +108,10 @@ static inline void __activate_cptr_traps_vhe(struct kvm_vcpu *vcpu)* The architecture is a bit crap (what a surprise): an EL2 guest * writing to CPTR_EL2 via CPACR_EL1 can't set any of TCPAC or TTA, * as they are RES0 in the guest's view. To work around it, trap the -* sucker using the very same bit it can't set... +* sucker using the very same bit it can't set. FEAT_NV2p1 fixes it. */ -if (vcpu_el2_e2h_is_set (vcpu) &&is_hyp_ctxt (vcpu)) +if (!cpus_have_final_cap (ARM64_HAS_NV2P1 ) && +vcpu_el2_e2h_is_set (vcpu) &&is_hyp_ctxt (vcpu)) val |=CPTR_EL2_TCPAC ;/*
arch/arm64/kvm/hyp/vhe/switch.c +3 -0
@@ -443,6 +443,9 @@ static bool kvm_hyp_handle_cpacr_el1(struct kvm_vcpu *vcpu, u64 *exit_code)u64 esr =kvm_vcpu_get_esr (vcpu);int rt; +if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) +return false; +if (!is_hyp_ctxt (vcpu) ||esr_sys64_to_sysreg (esr) !=SYS_CPACR_EL1 )return false;
arch/arm64/kvm/hyp/vhe/sysreg-sr.c +5 -3
@@ -42,10 +42,12 @@ static void __sysreg_save_vel2_state(struct kvm_vcpu *vcpu)u64 val;/* -* We don't save CPTR_EL2, as accesses to CPACR_EL1 -* are always trapped, ensuring that the in-memory -* copy is always up-to-date. A small blessing... +* Without FEAT_NV2p1, we don't save CPTR_EL2, as accesses +* to CPACR_EL1 are always trapped, ensuring that the +* in-memory copy is always up-to-date. A small blessing... */ +if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) +__vcpu_assign_sys_reg (vcpu,CPTR_EL2 ,read_sysreg_el1 (SYS_CPACR ));__vcpu_assign_sys_reg (vcpu,SCTLR_EL2 ,read_sysreg_el1 (SYS_SCTLR ));__vcpu_assign_sys_reg (vcpu,TTBR0_EL2 ,read_sysreg_el1 (SYS_TTBR0 ));__vcpu_assign_sys_reg (vcpu,TTBR1_EL2 ,read_sysreg_el1 (SYS_TTBR1 ));
arch/arm64/kvm/sys_regs.c +4 -1
@@ -326,7 +326,10 @@ u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg) val |=__vcpu_sys_reg (vcpu, reg) & ~CNTKCTL_VALID_BITS ;return val;case CPTR_EL2 : -return __vcpu_sys_reg (vcpu, reg); +if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) +return read_sysreg_el1 (SYS_CPACR ); +else +return __vcpu_sys_reg (vcpu, reg);default :WARN_ON_ONCE (1 ); }
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 09/29] KVM: arm64: Relax CPTR_EL2 handling when FEAT_NV2p1 is present
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] Userspace modifications to CPTR_EL2 via KVM_SET_ONE_REG are silently lost on systems with…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Userspace modifications to CPTR_EL2 via KVM_SET_ONE_REG are silently lost on systems with FEAT_NV2p1. --
--- Patch [9]: [PATCH v4 09/29] KVM: arm64: Relax CPTR_EL2 handling when FEAT_NV2p1 is present --- commit 0ba038cc87ef366635c1e40d5069280819355ee7
Author: Marc Zyngier <[email protected]>
KVM: arm64: Relax CPTR_EL2 handling when FEAT_NV2p1 is present
This patch updates the handling of CPTR_EL2 accesses when FEAT_NV2P1 is present, preventing trapping via CPACR_EL1 and directly context switching CPACR_EL1 in NV contexts instead.
Quoted 7 lines
arch/arm64/kvm/hyp/vhe/sysreg-sr.c +0
-0@@ -42,10 +42,12 @@ static void __sysreg_save_vel2_state(struct kvm_vcpu *vcpu)u64 val;
arch/arm64/kvm/hyp/vhe/sysreg-sr.c static void __sysreg_save_vel2_state(struct kvm_vcpu *vcpu) · line 50
Quoted 10 lines
arch/arm64/kvm/hyp/vhe/sysreg-sr.c +5
-3/* -* We don't save CPTR_EL2, as accesses to CPACR_EL1 -* are always trapped, ensuring that the in-memory -* copy is always up-to-date. A small blessing... +* Without FEAT_NV2p1, we don't save CPTR_EL2, as accesses +* to CPACR_EL1 are always trapped, ensuring that the +* in-memory copy is always up-to-date. A small blessing... */ +if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) +__vcpu_assign_sys_reg (vcpu,CPTR_EL2 ,read_sysreg_el1 (SYS_CPACR ));
[Severity: High]
Will this silently overwrite userspace modifications to CPTR_EL2 during vcpu_put()?
When userspace calls KVM_SET_ONE_REG for CPTR_EL2, the ioctl is wrapped in vcpu_load() and vcpu_put(). vcpu_load() loads the old memory state into the physical CPU.
Then, because CPTR_EL2 lacks a custom .set_user handler, kvm_sys_reg_set_user() falls back to updating the in-memory array:
if (r->set_user) {
ret = (r->set_user)(vcpu, r, val);
} else {
__vcpu_assign_sys_reg(vcpu, r->reg, val);
ret = 0;
}This updates the in-memory array, but not the physical hardware register.
When vcpu_put() is called, this newly added code in __sysreg_save_vel2_state() reads the stale old value from the physical hardware register and overwrites the newly set in-memory value, silently destroying the userspace modification.
Could a custom .set_user handler be needed for CPTR_EL2 to ensure the physical CPU register is updated when modified by userspace?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=9
Marc Zyngier
maintainerPATCH v4 10/29 KVM: arm64: Relax CNTHCTL_EL2 handling when FEAT_NV2p1 is present
In reply to Marc ZyngierWith NV2p1, it is no longer necessary to use the split approach where bits of CNTHCTL_EL2 cannot be accessed via CNTKCTL_EL1, and we can treat the CNTKCTL_EL1 accessor as if it was "normal".
Key the special casing on FEAT_NV2P1 not being implemented.
--- arch/arm64/kvm/arch_timer.c | 10 ++++++++-- arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 13 ++++++++++--- arch/arm64/kvm/sys_regs.c | 6 ++++-- 3 files changed, 22 insertions(+), 7 deletions(-)
arch/arm64/kvm/arch_timer.c +8 -2
@@ -876,8 +876,14 @@ static void timer_set_traps(struct kvm_vcpu *vcpu, struct timer_map *map)assign_clear_set_bit (tvt02,CNTHCTL_EL1NVVCT , clr, set);assign_clear_set_bit (tpt02,CNTHCTL_EL1NVPCT , clr, set); -/* This only happens on VHE, so use the CNTHCTL_EL2 accessor. */ -sysreg_clear_set (cnthctl_el2, clr, set); +/* +* This only happens on VHE, so use the CNTHCTL_EL2 accessor, unless +* we are sure CNTKCTL_EL1 is completely stateful with FEAT_NV2p1. +*/ +if (!cpus_have_final_cap (ARM64_HAS_NV2P1 )) +sysreg_clear_set (cnthctl_el2, clr, set); +else +sysreg_clear_set (cntkctl_el1, clr, set); }void kvm_timer_vcpu_load (struct kvm_vcpu *vcpu)
arch/arm64/kvm/hyp/vhe/sysreg-sr.c +10 -3
@@ -69,11 +69,18 @@ static void __sysreg_save_vel2_state(struct kvm_vcpu *vcpu)* The EL1 view of CNTKCTL_EL1 has a bunch of RES0 bits where * the interesting CNTHCTL_EL2 bits live. So preserve these * bits when reading back the guest-visible value. +* +* While NV2p1 fixes some of that, it makes CNTHCTL_EL2.ECV +* even more broken than it already was with NV2. */ val =read_sysreg_el1 (SYS_CNTKCTL ); - val &=CNTKCTL_VALID_BITS ; - __vcpu_rmw_sys_reg(vcpu, CNTHCTL_EL2, &=, ~CNTKCTL_VALID_BITS); - __vcpu_rmw_sys_reg(vcpu, CNTHCTL_EL2, |=, val); +if (!cpus_have_final_cap (ARM64_HAS_NV2P1 )) { + val &=CNTKCTL_VALID_BITS ; + __vcpu_rmw_sys_reg(vcpu, CNTHCTL_EL2, &=, ~CNTKCTL_VALID_BITS); + __vcpu_rmw_sys_reg(vcpu, CNTHCTL_EL2, |=, val); + }else { +__vcpu_assign_sys_reg (vcpu,CNTHCTL_EL2 , val); + } }__vcpu_assign_sys_reg (vcpu,SP_EL2 ,read_sysreg (sp_el1));
arch/arm64/kvm/sys_regs.c +4 -2
@@ -322,8 +322,10 @@ u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg)switch (reg) {case CNTHCTL_EL2 : val =read_sysreg_el1 (SYS_CNTKCTL ); - val &=CNTKCTL_VALID_BITS ; - val |=__vcpu_sys_reg (vcpu, reg) & ~CNTKCTL_VALID_BITS ; +if (!cpus_have_final_cap (ARM64_HAS_NV2P1 )) { + val &=CNTKCTL_VALID_BITS ; + val |=__vcpu_sys_reg (vcpu, reg) & ~CNTKCTL_VALID_BITS ; + }return val;case CPTR_EL2 :if (cpus_have_final_cap (ARM64_HAS_NV2P1 ))
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 11/29 KVM: arm64: Expose FEAT_NV2p1 to NV guests
In reply to Marc ZyngierSince NV2p1 is reducing the number of traps, it is valuable to expose it to NV guests. Do so.
--- arch/arm64/kvm/nested.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
arch/arm64/kvm/nested.c +6 -2
@@ -1728,7 +1728,7 @@ u64 limit_nv_id_reg(struct kvm *kvm, u32 reg, u64 val)* You get EITHER * * - FEAT_VHE without FEAT_E2H0 -* - FEAT_NV limited to FEAT_NV2 +* - FEAT_NV limited to FEAT_NV2(p1) * - HCR_EL2.NV1 being RES0 * * OR
@@ -1740,7 +1740,11 @@ u64 limit_nv_id_reg(struct kvm *kvm, u32 reg, u64 val)if (test_bit (KVM_ARM_VCPU_HAS_EL2_E2H0 , kvm->arch.vcpu_features)) { val =0 ; }else { - val =SYS_FIELD_PREP_ENUM (ID_AA64MMFR4_EL1 , NV_frac,NV2_ONLY ); + val &= ID_AA64MMFR4_EL1_NV_frac; +if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) + val =ID_REG_LIMIT_FIELD_ENUM (val,ID_AA64MMFR4_EL1 , NV_frac,NV2P1 ); +else + val =SYS_FIELD_PREP_ENUM (ID_AA64MMFR4_EL1 , NV_frac,NV2_ONLY ); val |=SYS_FIELD_PREP_ENUM (ID_AA64MMFR4_EL1 ,E2H0 ,NI_NV1 ); }break ;
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 11/29] KVM: arm64: Expose FEAT_NV2p1 to NV guests
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] Missing context synchronization event (`isb()`) after writing to a control-plane system register…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Missing context synchronization event (isb()) after writing to a control-plane system register (CNTKCTL_EL1 / CNTHCTL_EL2). --
--- Patch [11]: [PATCH v4 11/29] KVM: arm64: Expose FEAT_NV2p1 to NV guests --- commit 5b82c2dde5c66c45b12e75de220aab8518c99896
Author: Marc Zyngier <[email protected]>
KVM: arm64: Expose FEAT_NV2p1 to NV guests
This commit exposes FEAT_NV2p1 to nested virtualization guests. By reducing the number of traps and updating timer configurations, it brings valuable performance improvements to NV guests.
Quoted 9 lines
arch/arm64/kvm/arch_timer.c +0
-1@@ -876,8 +876,14 @@ static void timer_set_traps(struct kvm_vcpu *vcpu, struct timer_map *map)assign_clear_set_bit (tvt02,CNTHCTL_EL1NVVCT , clr, set);assign_clear_set_bit (tpt02,CNTHCTL_EL1NVPCT , clr, set); -/* This only happens on VHE, so use the CNTHCTL_EL2 accessor. */
arch/arm64/kvm/arch_timer.c static void timer_set_traps(struct kvm_vcpu *vcpu, struct timer_map *map) · line 887
Quoted 10 lines
arch/arm64/kvm/arch_timer.c +8
-1-sysreg_clear_set (cnthctl_el2, clr, set); +/* +* This only happens on VHE, so use the CNTHCTL_EL2 accessor, unless +* we are sure CNTKCTL_EL1 is completely stateful with FEAT_NV2p1. +*/ +if (!cpus_have_final_cap (ARM64_HAS_NV2P1 )) +sysreg_clear_set (cnthctl_el2, clr, set); +else +sysreg_clear_set (cntkctl_el1, clr, set); }
[Severity: High]
Does this write to a control-plane system register require an isb() before returning?
The ARM64 architecture generally requires an isb() as the very next instruction after writing to control-plane registers like cnthctl_el2 and cntkctl_el1 to ensure subsequent instructions do not execute with a stale configuration.
Because sysreg_clear_set() expands to a hardware write without an isb(), timer_set_traps() returns immediately after the write. Could the CPU execute subsequent instructions in the caller with stale pipeline state before a context synchronization event occurs?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=11
Marc Zyngier
maintainerPATCH v4 12/29 arm64: Add FEAT_NV2p1 detection
In reply to Marc ZyngierAdd the necessary NV2p1 probing to the cpufeature infrastructure.
--- arch/arm64/kernel/cpufeature.c | 7 +++++++ 1 file changed, 7 insertions(+)
arch/arm64/kernel/cpufeature.c +7 -0
@@ -2620,6 +2620,13 @@ static const struct arm64_cpu_capabilities arm64_features[] = { {/* Sentinel */ } }, }, + { + .desc = "FEAT_NV2p1", + .capability = ARM64_HAS_NV2P1, + .type = ARM64_CPUCAP_SYSTEM_FEATURE, + .matches = has_cpuid_feature, +ARM64_CPUID_FIELDS (ID_AA64MMFR4_EL1 , NV_frac,NV2P1 ) + }, { .capability =ARM64_HAS_32BIT_EL0_DO_NOT_USE , .type = ARM64_CPUCAP_SYSTEM_FEATURE,
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 13/29 arm64: sysreg: Add NVHCR_EL2 description as a mirror of HCR_EL2
In reply to Marc ZyngierFEAT_NV3 introduces a new register that contains the HCR_EL2 value exposed to a NV guest. As such, it has the exact same layout as HCR_EL2.
Describe NVHCR_EL2 as a mapping to HCR_EL2.
--- arch/arm64/tools/sysreg | 3 +++ 1 file changed, 3 insertions(+)
arch/arm64/tools/sysreg +3 -0
@@ -4266,6 +4266,9 @@ Field 1 E2TRE Field 0 E0HTRE EndSysreg +Sysreg NVHCR_EL2 3 4 1 5 0 +Mapping HCR_EL2 +EndSysreg Sysreg HDFGRTR2_EL2 3 4 3 1 0 Res0 63:25
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 14/29 arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3
In reply to Marc ZyngierFEAT_NV3 introduces 4 new HCRX_EL2 control bits. Describe them in the sysreg file.
--- arch/arm64/tools/sysreg | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/tools/sysreg +6 -1
@@ -4548,7 +4548,12 @@ Fields ZCR_ELx EndSysreg Sysreg HCRX_EL2 3 4 1 2 2 -Res0 63:27 +Res0 63:35 +Field 34 NVnTTLBOS +Field 33 NVnTTLBIS +Field 32 NVnTTLB +Res0 31:28 +Field 27 NVTGE Field 26 SRMASKEn Res0 25 Field 24 PACMEn
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 14/29] arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] Adding new HCRX_EL2 fields (NVnTTLBOS, NVnTTLBIS, NVnTTLB, NVTGE) to the sysreg file silently…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Adding new HCRX_EL2 fields (NVnTTLBOS, NVnTTLBIS, NVnTTLB, NVTGE) to the sysreg file silently removes them from the auto-generated HCRX_EL2_RES0 mask, but KVM's feature map was not updated to handle them. This semantic drift breaks KVM validation and allows guests to inject unvirtualized CPU features into the host. --
--- Patch [14]: [PATCH v4 14/29] arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3 --- commit 98082153874cf6a7ea1e64f1536725fa89404dc8
Author: Marc Zyngier <[email protected]>
arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3
FEAT_NV3 introduces 4 new HCRX_EL2 control bits. Describe them in the sysreg file.
Quoted 5 lines
arch/arm64/tools/sysreg +0
-0
arch/arm64/tools/sysreg Fields ZCR_ELx · line 4556
Quoted 10 lines
arch/arm64/tools/sysreg +6
-1@@ -4548,7 +4548,12 @@ Fields ZCR_ELx EndSysreg Sysreg HCRX_EL2 3 4 1 2 2 -Res0 63:27 +Res0 63:35 +Field 34 NVnTTLBOS +Field 33 NVnTTLBIS +Field 32 NVnTTLB +Res0 31:28 +Field 27 NVTGE
[Severity: High]
By changing these bits from Res0 to Field, they are silently removed from the auto-generated HCRX_EL2_RES0 mask. Does this cause KVM's boot-time validation loop to fail?
In arch/arm64/kvm/config.c, check_feat_map() asserts that all non-RES0 bits are accounted for. Since hcrx_feat_map is not updated with NEEDS_FEAT() mappings for these new fields, it seems this will unconditionally trigger an error on boot:
arch/arm64/kvm/config.c:check_feat_map() {
...
if (mask != ~resx)
kvm_err("Undefined %s behaviour, bits %016llx\n",
str, mask ^ ~resx);
...
}Furthermore, without these bits being tracked in KVM's feature map, can a nested guest set them in its virtual HCRX_EL2 register?
If the guest writes these bits, it appears they could bypass filtering during guest entry and be written directly into the physical CPU register:
arch/arm64/kvm/hyp/include/hyp/switch.h:__activate_traps_common() {
...
hcrx |= (val & ~NV_HCRX_GUEST_EXCLUDE);
hcrx &= ~(~val & ~NV_HCRX_GUEST_EXCLUDE);
...
}Could this allow an untrusted guest to misconfigure unvirtualized hardware state?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=14
Marc Zyngier
maintainerRe: [PATCH v4 14/29] arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3
Quoted 3 lines · On Thu, 30 Jul 2026 08:40:05 +0100, [email protected] wrote:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Adding newHCRX_EL2fields (NVnTTLBOS, NVnTTLBIS, NVnTTLB, NVTGE) to the sysreg file silently removes them from the auto-generatedHCRX_EL2_RES0mask, but KVM's feature map was not updated to handle them. This semantic drift breaks KVM validation and allows guests to inject unvirtualized CPU features into the host.
"semantic drift" my arse. You're a stuck record.
M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
Marc Zyngier
maintainerPATCH v4 15/29 arm64: Add ARM64_HAS_NV3 capability
In reply to Marc ZyngierAs a bunch of KVM code is going to depend on FEAT_NV3 being detected on the host, add a new capability that will describe it.
--- arch/arm64/tools/cpucaps | 1 + 1 file changed, 1 insertion(+)
arch/arm64/tools/cpucaps +1 -0
@@ -52,6 +52,7 @@ HAS_LSUI HAS_MOPS HAS_NESTED_VIRT HAS_NV2P1 +HAS_NV3 HAS_BBML2_NOABORT HAS_PAN HAS_PMUV3
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 16/29 KVM: arm64: Split NV-specific exit fixups from the non-NV handling
In reply to Marc ZyngierIn order to facilitate further changes, move the NV handling of early fixups in its own helper. This also makes the code slightly simpler to parse.
--- arch/arm64/kvm/hyp/vhe/switch.c | 18 +++++++++++------- 1 file changed, 11 insertions(+), 7 deletions(-)
arch/arm64/kvm/hyp/vhe/switch.c +11 -7
@@ -539,18 +539,15 @@ static const exit_handler_fn hyp_exit_handlers[] = { [0x3F] = kvm_hyp_handle_impdef, }; -static inline bool fixup_guest_exit (struct kvm_vcpu *vcpu,u64 *exit_code) +static void fixup_nv_guest_exit (struct kvm_vcpu *vcpu) { -synchronize_vcpu_pstate (vcpu); -/* * If we were in HYP context on entry, adjust the PSTATE view * so that the usual helpers work correctly. This enforces our * invariant that the guest's HYP context status is preserved * across a run. */ -if (vcpu_has_nv (vcpu) && -unlikely (host_data_test_flag (VCPU_IN_HYP_CONTEXT ))) { +if (unlikely (host_data_test_flag (VCPU_IN_HYP_CONTEXT ))) {u64 mode = *vcpu_cpsr (vcpu) & (PSR_MODE_MASK |PSR_MODE32_BIT );switch (mode) {
@@ -567,8 +564,15 @@ static inline bool fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code) }/* Apply extreme paranoia! */ -BUG_ON (vcpu_has_nv (vcpu) && - !!host_data_test_flag (VCPU_IN_HYP_CONTEXT ) !=is_hyp_ctxt (vcpu)); +BUG_ON (!!host_data_test_flag (VCPU_IN_HYP_CONTEXT ) !=is_hyp_ctxt (vcpu)); +} + +static bool fixup_guest_exit (struct kvm_vcpu *vcpu,u64 *exit_code) +{ +synchronize_vcpu_pstate (vcpu); + +if (vcpu_has_nv (vcpu)) +fixup_nv_guest_exit (vcpu);return __fixup_guest_exit (vcpu, exit_code, hyp_exit_handlers); }
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 17/29 KVM: arm64: Add NV3 control bits to HCRX_EL2 sanitisation
In reply to Marc ZyngierExpose the FEAT_NV3 control bits to the sanitisation code so that KVM stops moaning about the unattributed bits.
--- arch/arm64/kvm/config.c | 6 ++++++ 1 file changed, 6 insertions(+)
arch/arm64/kvm/config.c +6 -0
@@ -225,6 +225,7 @@ struct reg_feat_map_desc {#define FEAT_HCX ID_AA64MMFR1_EL1, HCX, IMP#define FEAT_S2PIE ID_AA64MMFR3_EL1, S2PIE, IMP#define FEAT_GCIE ID_AA64PFR2_EL1, GCIE, IMP +#define FEAT_NV3 ID_AA64MMFR4_EL1, NV_frac, NV3static bool not_feat_aa64el3 (struct kvm *kvm) {
@@ -904,6 +905,11 @@ static const DECLARE_FEAT_MAP_FGT(hdfgwtr2_desc, hdfgwtr2_masks,static const struct reg_bits_to_feat_map hcrx_feat_map[] = { +NEEDS_FEAT (HCRX_EL2_NVTGE | + HCRX_EL2_NVnTTLB | + HCRX_EL2_NVnTTLBIS | + HCRX_EL2_NVnTTLBOS, +FEAT_NV3 ),NEEDS_FEAT (HCRX_EL2_SRMASKEn,FEAT_SRMASK ),NEEDS_FEAT (HCRX_EL2_PACMEn, feat_pauth_lr),NEEDS_FEAT (HCRX_EL2_EnFPM,FEAT_FPMR ),
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 18/29 KVM: arm64: Add kvm_has_nv{2,3}() predicates
In reply to Marc ZyngierAdd a new set of predicates indicating whether VM is capable of NV2, NV3, and is in a nested NV3 context.
This is going to become useful as we start dealing with a mix of behaviours (NV2, NV3, NV2 on NV3...).
--- arch/arm64/include/asm/kvm_emulate.h | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)
arch/arm64/include/asm/kvm_emulate.h +19 -0
@@ -266,6 +266,25 @@ static inline bool vserror_state_is_nested(struct kvm_vcpu *vcpu) (__vcpu_sys_reg (vcpu,HCRX_EL2 ) &HCRX_EL2_TMEA ); } +static inline bool kvm_has_nv2 (struct kvm *kvm) +{ +return (cpus_have_final_cap (ARM64_HAS_NESTED_VIRT ) && +kvm_has_feat (kvm,ID_AA64MMFR4_EL1 , NV_frac,NV2_ONLY )); +} + +static inline bool kvm_has_nv3 (struct kvm *kvm) +{ +return (cpus_have_final_cap (ARM64_HAS_NV3 ) && +kvm_has_feat (kvm,ID_AA64MMFR4_EL1 , NV_frac,NV3 )); +} + +static inline bool is_nested_nv3_ctxt (struct kvm_vcpu *vcpu) +{ +return (has_vhe () &&kvm_has_nv3 (vcpu->kvm) &&is_nested_ctxt (vcpu) && + (__vcpu_sys_reg (vcpu,HCR_EL2 ) &HCR_EL2_NV ) && + (__vcpu_sys_reg (vcpu,HCRX_EL2 ) &HCRX_EL2_NVTGE )); +} +/* * The layout of SPSR for an AArch32 state is different when observed from an * AArch64 SPSR_ELx or an AArch32 SPSR_*. This function generates the AArch32
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 19/29 KVM: arm64: Make HCR_EL2 a non-VNCR register
In reply to Marc ZyngierFEAT_NV3 makes a fundamental change to the architecture, by moving guest-initiated HCR_EL2 accesses to the NVHCR_EL2 register. As the names suggests, this is HCR_EL2 for a NV guest.
But where do NVHCR_EL2 accesses from a guest go? The are redirected to the VNCR page, right where HCR_EL2 is stored in the NV2 case. Does it hurt? Good. There's more coming.
The challenge here is to make KVM work seamlessly, without rewriting everything. Which implies that things such as __vcpu_sys_reg(HCR_EL2) must work, no matter the underlying NV implementation.
A simple way to deal with it is to move HCR_EL2's canonical storage outside of VNCR for the vast majority of the KVM code, and only have a copy at entry/exit times. Given that we don't really support NV3 yet, this is pretty simple.
In the process, advertise NVHCR_EL2 as the register that now holds offset 0x78 in the VNCR page.
--- arch/arm64/include/asm/kvm_host.h | 3 ++- arch/arm64/include/asm/vncr_mapping.h | 2 +- arch/arm64/kvm/hyp/vhe/switch.c | 9 +++++++++ 3 files changed, 12 insertions(+), 2 deletions(-)
arch/arm64/include/asm/kvm_host.h +2 -1
@@ -543,6 +543,7 @@ enum vcpu_sysreg {MDCR_EL2 ,/* Monitor Debug Configuration Register (EL2) */ CNTHCTL_EL2 ,/* Counter-timer Hypervisor Control register */ ZCR_EL2 ,/* SVE Control Register (EL2) */ +HCR_EL2 ,/* Hypervisor Control Register */ /* Any VNCR-capable reg goes after this point */ MARKER(__VNCR_START__),
@@ -571,7 +572,7 @@ enum vcpu_sysreg {VNCR (TFSR_EL1 ),/* Tag Fault Status Register (EL1) */ VNCR (VPIDR_EL2 ),/* Virtualization Processor ID Register */ VNCR (VMPIDR_EL2 ),/* Virtualization Multiprocessor ID Register */ -VNCR (HCR_EL2 ),/* Hypervisor Configuration Register */ +VNCR (NVHCR_EL2 ),/* NV Hypervisor Configuration Register */ VNCR (HSTR_EL2 ),/* Hypervisor System Trap Register */ VNCR (VTTBR_EL2 ),/* Virtualization Translation Table Base Register */ VNCR(VTCR_EL2), /* Virtualization Translation Control Register */
arch/arm64/include/asm/vncr_mapping.h +1 -1
@@ -11,7 +11,7 @@#define VNCR_VTCR_EL2 0x040#define VNCR_VMPIDR_EL2 0x050#define VNCR_CNTVOFF_EL2 0x060 -#define VNCR_HCR_EL2 0x078 +#define VNCR_NVHCR_EL2 0x078#define VNCR_HSTR_EL2 0x080#define VNCR_VPIDR_EL2 0x088#define VNCR_TPIDR_EL2 0x090
arch/arm64/kvm/hyp/vhe/switch.c +9 -0
@@ -70,6 +70,9 @@ static u64 __compute_hcr(struct kvm_vcpu *vcpu)if (!vcpu_el2_e2h_is_set (vcpu)) hcr |=HCR_NV1 ; +/* Publish the guest's view of HCR_EL2 to the HW */ +__vcpu_assign_sys_reg (vcpu,NVHCR_EL2 ,__vcpu_sys_reg (vcpu,HCR_EL2 )); +/* * Nothing in HCR_EL2 should impact running in hypervisor * context, apart from bits we have defined as RESx (E2H,
@@ -549,6 +552,7 @@ static void fixup_nv_guest_exit(struct kvm_vcpu *vcpu)*/ if (unlikely (host_data_test_flag (VCPU_IN_HYP_CONTEXT ))) {u64 mode = *vcpu_cpsr (vcpu) & (PSR_MODE_MASK |PSR_MODE32_BIT ); +u64 hcr;switch (mode) {case PSR_MODE_EL1t:
@@ -561,6 +565,11 @@ static void fixup_nv_guest_exit(struct kvm_vcpu *vcpu) *vcpu_cpsr (vcpu) &= ~(PSR_MODE_MASK |PSR_MODE32_BIT ); *vcpu_cpsr (vcpu) |= mode; + +/* Publish the latest HCR_EL2 to the emulation */ + hcr =__vcpu_sys_reg (vcpu,NVHCR_EL2 ); + +__vcpu_assign_sys_reg (vcpu,HCR_EL2 , hcr); }/* Apply extreme paranoia! */
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 20/29 KVM: arm64: Add sanitisation for NVHCR_EL2
In reply to Marc ZyngierJust like any other VNCR-based register, NVHCR_EL2 requires some level of sanitisation. Being specified as a live copy of HCR_EL2, it adopts the exact same format, but depends on FEAT_NV3 instead.
A subtle aspect is that we only want to apply the sanitisation if FEAT_NV3 is actually present, as the VNCR location is otherwise used to back accesses to HCR_EL2.
--- arch/arm64/kvm/config.c | 15 +++++++++++++++ arch/arm64/kvm/nested.c | 4 ++++ 2 files changed, 19 insertions(+)
arch/arm64/kvm/config.c +15 -0
@@ -1019,6 +1019,9 @@ static const struct reg_bits_to_feat_map hcr_feat_map[] = { static const DECLARE_FEAT_MAP(hcr_desc, HCR_EL2, hcr_feat_map, FEAT_AA64EL2); +static const DECLARE_FEAT_MAP(nvhcr_desc, NVHCR_EL2, + hcr_feat_map, FEAT_NV3); +static const struct reg_bits_to_feat_map sctlr2_feat_map[] = { NEEDS_FEAT(SCTLR2_EL1_NMEA | SCTLR2_EL1_EASE,
@@ -1393,6 +1396,7 @@ void __init check_feature_map(void)check_reg_desc (&hdfgwtr2_desc);check_reg_desc (&hcrx_desc);check_reg_desc (&hcr_desc); +check_reg_desc (&nvhcr_desc);check_reg_desc (&sctlr2_desc);check_reg_desc (&tcr2_el2_desc);check_reg_desc (&sctlr_el1_desc);
@@ -1592,6 +1596,17 @@ struct resx get_reg_fixed_bits(struct kvm *kvm, enum vcpu_sysreg reg)case HCR_EL2 : resx =compute_reg_resx_bits (kvm, &hcr_desc,0 ,0 );break ; +case NVHCR_EL2 : +/* +* Only apply sanitisation if we do have FEAT_NV3. +* Otherwise, the register aliases with HCR_EL2 in VNCR, +* and we're better off relying on data transfers between +* NVHCR_EL2 and HCR_EL2 to sanitise things. +*/ + resx = (kvm_has_nv3 (kvm) ? +compute_reg_resx_bits (kvm, &nvhcr_desc,0 ,0 ) : + (typeof (resx )){}); +break ;case SCTLR2_EL1 :case SCTLR2_EL2 : resx =compute_reg_resx_bits (kvm, &sctlr2_desc,0 ,0 );
arch/arm64/kvm/nested.c +4 -0
@@ -1830,6 +1830,10 @@ int kvm_init_nv_sysregs(struct kvm_vcpu *vcpu) resx =get_reg_fixed_bits (kvm,HCR_EL2 );set_sysreg_masks (kvm,HCR_EL2 , resx); +/* NVHCR_EL2 */ + resx =get_reg_fixed_bits (kvm,NVHCR_EL2 ); +set_sysreg_masks (kvm,NVHCR_EL2 , resx); +/* HCRX_EL2 */ resx =get_reg_fixed_bits (kvm,HCRX_EL2 );set_sysreg_masks (kvm,HCRX_EL2 , resx);
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 21/29 KVM: arm64: Add NVHCR_EL2 handling to the sysreg array
In reply to Marc ZyngierExpose NVHCR_EL2 to userspace, and treat the direct access as UNDEF, as that would only outline a bug in our exception routing.
The generic accessors are also updated to deal with the relatively uncommon location of that register.
--- arch/arm64/kvm/sys_regs.c | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+)
arch/arm64/kvm/sys_regs.c +28 -0
@@ -221,6 +221,20 @@ static void locate_register(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg,if (is_hyp_ctxt (vcpu) &&vcpu_el2_e2h_is_set (vcpu)) loc->loc =SR_LOC_SPECIAL ;break ; +case NVHCR_EL2 : +/* +* Yes, NVHCR_EL2 maps to itself when loaded in nested +* context. If you feel like the architecture is double +* backing on itself upside down, you're not alone. +*/ +WARN_ON_ONCE (!kvm_has_nv3 (vcpu->kvm)); +if (is_hyp_ctxt (vcpu)) { + loc->loc =SR_LOC_MEMORY ; + }else { + loc->loc =SR_LOC_LOADED |SR_LOC_MAPPED ; + loc->map_reg =NVHCR_EL2 ; + } +break ;default : loc->loc =locate_direct_register (vcpu, reg); }
@@ -260,6 +274,7 @@ static u64 read_sr_from_cpu(enum vcpu_sysreg reg)case DACR32_EL2 : val =read_sysreg_s (SYS_DACR32_EL2 );break ;case IFSR32_EL2 : val =read_sysreg_s (SYS_IFSR32_EL2 );break ;case DBGVCR32_EL2 : val =read_sysreg_s (SYS_DBGVCR32_EL2 );break ; +case NVHCR_EL2 : val =read_sysreg_s (SYS_NVHCR_EL2 );break ;default :WARN_ON_ONCE (1 ); }
@@ -298,6 +313,7 @@ static void write_sr_to_cpu(enum vcpu_sysreg reg, u64 val)case DACR32_EL2 :write_sysreg_s (val,SYS_DACR32_EL2 );break ;case IFSR32_EL2 :write_sysreg_s (val,SYS_IFSR32_EL2 );break ;case DBGVCR32_EL2 :write_sysreg_s (val,SYS_DBGVCR32_EL2 );break ; +case NVHCR_EL2 :write_sysreg_s (val,SYS_NVHCR_EL2 );break ;default :WARN_ON_ONCE (1 ); } }
@@ -2861,6 +2877,16 @@ static unsigned int vncr_el2_visibility(const struct kvm_vcpu *vcpu,return REG_HIDDEN ; } +static unsigned int nvhcr_el2_visibility (const struct kvm_vcpu *vcpu, +const struct sys_reg_desc *rd) +{ +if (el2_visibility (vcpu, rd) ==0 && +kvm_has_feat (vcpu->kvm,ID_AA64MMFR4_EL1 , NV_frac,NV3 )) +return 0 ; + +return REG_HIDDEN ; +} +static unsigned int sctlr2_visibility (const struct kvm_vcpu *vcpu,const struct sys_reg_desc *rd) {
@@ -3774,6 +3800,8 @@ static const struct sys_reg_desc sys_reg_descs[] = { sve_el2_visibility),EL2_REG_VNCR (HCRX_EL2 , reset_val,0 ), +EL2_REG_FILTERED (NVHCR_EL2 , undef_access, reset_val,0 , + nvhcr_el2_visibility),EL2_REG (TTBR0_EL2 , access_rw, reset_val,0 ), EL2_REG(TTBR1_EL2, access_rw, reset_val, 0),
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 21/29] KVM: arm64: Add NVHCR_EL2 handling to the sysreg array
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [High] Inverted context logic for NVHCR_EL2 in locate_register() incorrectly maps the register to an…
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Inverted context logic for NVHCR_EL2 in locate_register() incorrectly maps the register to an unmanaged hardware CPU register.
- [High] Missing explicit Context Synchronization Event (ISB) after writing to the architectural control-plane system register NVHCR_EL2. --
--- Patch [21]: [PATCH v4 21/29] KVM: arm64: Add NVHCR_EL2 handling to the sysreg array --- commit aea2461c165a3a2d9e5915b4946e1ec44773006d
Author: Marc Zyngier <[email protected]>
Subject: KVM: arm64: Add NVHCR_EL2 handling to the sysreg array
This commit exposes NVHCR_EL2 to userspace and treats direct access as UNDEF to catch exception routing bugs. It also updates generic accessors to handle the register's location.
Quoted 12 lines
arch/arm64/kvm/sys_regs.c +4
-0@@ -221,6 +221,20 @@ static void locate_register(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg,if (is_hyp_ctxt (vcpu) &&vcpu_el2_e2h_is_set (vcpu)) loc->loc =SR_LOC_SPECIAL ;break ; +case NVHCR_EL2 : +/* +* Yes, NVHCR_EL2 maps to itself when loaded in nested +* context. If you feel like the architecture is double
arch/arm64/kvm/sys_regs.c static void locate_register(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg, · line 237
Quoted 10 lines
arch/arm64/kvm/sys_regs.c +10
-0+* backing on itself upside down, you're not alone. +*/ +WARN_ON_ONCE (!kvm_has_nv3 (vcpu->kvm)); +if (is_hyp_ctxt (vcpu)) { + loc->loc =SR_LOC_MEMORY ; + }else { + loc->loc =SR_LOC_LOADED |SR_LOC_MAPPED ; + loc->map_reg =NVHCR_EL2 ; + } +break ;
[Severity: High]
Does this invert the context logic for NVHCR_EL2?
If !is_hyp_ctxt(vcpu) (meaning the vCPU is running a nested guest), this maps the register to an unmanaged hardware CPU register rather than the guest's VNCR memory array.
Since KVM doesn't save or restore the physical SYS_NVHCR_EL2 register on context switches, could this cause KVM_SET_ONE_REG to lose the guest's NVHCR_EL2 state, and KVM_GET_ONE_REG to leak uninitialized host physical register state?
[ ... ]
arch/arm64/kvm/sys_regs.c static void write_sr_to_cpu(enum vcpu_sysreg reg, u64 val) · line 316
Quoted 5 lines
arch/arm64/kvm/sys_regs.c +1
-0@@ -298,6 +313,7 @@ static void write_sr_to_cpu(enum vcpu_sysreg reg, u64 val)case DACR32_EL2 :write_sysreg_s (val,SYS_DACR32_EL2 );break ;case IFSR32_EL2 :write_sysreg_s (val,SYS_IFSR32_EL2 );break ;case DBGVCR32_EL2 :write_sysreg_s (val,SYS_DBGVCR32_EL2 );break ; +case NVHCR_EL2 :write_sysreg_s (val,SYS_NVHCR_EL2 );break ;
[Severity: High]
Is an explicit context synchronization event (isb) required here after writing to NVHCR_EL2?
The write_sysreg_s macro modifies a control-plane system register but is followed immediately by a branch (break), skipping the architectural requirement of having an isb as the very next instruction.
Could this leave the CPU pipeline operating in an architecturally undefined state when userspace calls KVM_SET_ONE_REG?
Quoted 3 lines
arch/arm64/kvm/sys_regs.c +0
-0default :WARN_ON_ONCE (1 ); } }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=21
Marc Zyngier
maintainerPATCH v4 22/29 KVM: arm64: Add routing for NVHCR_EL2 trap
In reply to Marc ZyngierNVHCR_EL2 accesses from EL1 are taken to EL2 when HCRX_EL2.NVTGE==0 and HCR_EL2.NV==1. Describe this in the exception routing tables.
--- arch/arm64/kvm/emulate-nested.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
arch/arm64/kvm/emulate-nested.c +13 -0
@@ -136,6 +136,8 @@ enum cgt_group_id {CGT_CPTR_TTA ,CGT_MDCR_HPMN , + CGT_HCR_NV_HCRX_nNVTGE, +/* Must be last */ __NR_CGT_GROUP_IDS__ };
@@ -588,6 +590,15 @@ static enum trap_behaviour check_mdcr_hpmn(struct kvm_vcpu *vcpu)return BEHAVE_HANDLE_LOCALLY ; } +static enum trap_behaviour check_hcr_nv_hcrx_nnvtge (struct kvm_vcpu *vcpu) +{ +if ((__vcpu_sys_reg (vcpu,HCR_EL2 ) &HCR_EL2_NV ) && + !(__vcpu_sys_reg (vcpu,HCRX_EL2 ) &HCRX_EL2_NVTGE )) +return BEHAVE_FORWARD_RW ; + +return BEHAVE_HANDLE_LOCALLY ; +} +#define CCC (id, fn) \ [id - __COMPLEX_CONDITIONS__] = fn
@@ -598,6 +609,7 @@ static const complex_condition_check ccc[] = {CCC (CGT_CNTHCTL_EL1NVVCT , check_cnthctl_el1nvvct),CCC (CGT_CPTR_TTA , check_cptr_tta),CCC (CGT_MDCR_HPMN , check_mdcr_hpmn), + CCC(CGT_HCR_NV_HCRX_nNVTGE, check_hcr_nv_hcrx_nnvtge), };/*
@@ -853,6 +865,7 @@ static const struct encoding_to_trap_config encoding_to_cgt[] __initconst = {SR_TRAP (SYS_SCTLR2_EL2 ,CGT_HCR_NV ),SR_RANGE_TRAP (SYS_HCR_EL2 ,SYS_HCRX_EL2 ,CGT_HCR_NV ), +SR_TRAP (SYS_NVHCR_EL2 , CGT_HCR_NV_HCRX_nNVTGE),SR_TRAP (SYS_SMPRIMAP_EL2 ,CGT_HCR_NV ),SR_TRAP (SYS_SMCR_EL2 ,CGT_HCR_NV ), SR_RANGE_TRAP(SYS_TTBR0_EL2,
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 22/29] KVM: arm64: Add routing for NVHCR_EL2 trap
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] The evaluation of HCRX_EL2.NVTGE ignores its architectural dependency on HCR_EL2.E2H, leading to…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The evaluation of HCRX_EL2.NVTGE ignores its architectural dependency on HCR_EL2.E2H, leading to incorrect trap routing for non-VHE guest hypervisors. --
--- Patch [22]: [PATCH v4 22/29] KVM: arm64: Add routing for NVHCR_EL2 trap --- commit 007c598e0e2ba89316901f92f901362aec3f3a5b
Author: Marc Zyngier <[email protected]>
KVM: arm64: Add routing for NVHCR_EL2 trap
This commit describes the exception routing tables for NVHCR_EL2 accesses from EL1, taking them to EL2 when HCRX_EL2.NVTGE==0 and HCR_EL2.NV==1.
Quoted 4 lines
arch/arm64/kvm/emulate-nested.c +0
-0
[ ... ]
arch/arm64/kvm/emulate-nested.c static enum trap_behaviour check_mdcr_hpmn(struct kvm_vcpu *vcpu) · line 596
Quoted 8 lines
arch/arm64/kvm/emulate-nested.c +4
-0@@ -588,6 +590,15 @@ static enum trap_behaviour check_mdcr_hpmn(struct kvm_vcpu *vcpu)return BEHAVE_HANDLE_LOCALLY ; } +static enum trap_behaviour check_hcr_nv_hcrx_nnvtge (struct kvm_vcpu *vcpu) +{ +if ((__vcpu_sys_reg (vcpu,HCR_EL2 ) &HCR_EL2_NV ) && + !(__vcpu_sys_reg (vcpu,HCRX_EL2 ) &HCRX_EL2_NVTGE ))
[Severity: High]
According to the ARM Architecture Reference Manual, the HCRX_EL2.NVTGE bit behaves as 0 for all purposes other than reading the value of the bit when HCR_EL2.E2H is 0.
Does this raw check of HCRX_EL2_NVTGE need to account for the HCR_EL2.E2H dependency?
If a non-VHE guest hypervisor sets HCR_EL2.E2H = 0, HCR_EL2.NV = 1, and HCRX_EL2.NVTGE = 1 (which is architecturally valid), it appears KVM might falsely conclude the trap shouldn't be forwarded. It would then handle it locally in L0, which injects an UNDEF exception into the L2 guest.
There appears to be a similar pattern in is_nested_nv3_ctxt() in arch/arm64/include/asm/kvm_emulate.h where the raw NVTGE bit is evaluated without factoring in the E2H dependency.
Quoted 5 lines
arch/arm64/kvm/emulate-nested.c +5
-0+return BEHAVE_FORWARD_RW ; + +return BEHAVE_HANDLE_LOCALLY ; +} +
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=22
Marc Zyngier
maintainerPATCH v4 23/29 KVM: arm64: Add NVHCR_EL2 context switching
In reply to Marc ZyngierSince NVHCR_EL2 represents the HCR_EL2 state of the EL1 guest, it must be dealt with in some particular way:
- for a guest in hyp context (an L1 by definition), NVHCR_EL2 directly reflects HCR_EL2 as read and written by the guest itself. It must therefore be eagerly synced back with the emulation code which only knows about HCR_EL2. This is unconditional if NV3 is available on the host.
- For an L2 guest, NVHCR_EL2 is controlled by the L1 guest, and we just context switch it like any other EL1 register. Yes, EL1, as that's where this thing runs from the PoV of L1. This is conditioned on the guest using NV3.
--- arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h | 11 +++++++++++ arch/arm64/kvm/hyp/vhe/switch.c | 10 ++++++++-- 2 files changed, 19 insertions(+), 2 deletions(-)
arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h +11 -0
@@ -172,6 +172,10 @@ static inline void __sysreg_save_el1_state(struct kvm_cpu_context *ctxt)if (ctxt_has_sctlr2 (ctxt))ctxt_sys_reg (ctxt,SCTLR2_EL1 ) =read_sysreg_el1 (SYS_SCTLR2 ); + +/* Retrieve L2's HCR_EL2, and save it for future use */ +if (is_nested_nv3_ctxt (ctxt_to_vcpu (ctxt))) +ctxt_sys_reg (ctxt,NVHCR_EL2 ) =read_sysreg_s (SYS_NVHCR_EL2 ); }static inline void __sysreg_save_el2_return_state (struct kvm_cpu_context *ctxt)
@@ -285,6 +289,13 @@ static inline void __sysreg_restore_el1_state(struct kvm_cpu_context *ctxt,if (ctxt_has_sctlr2 (ctxt))write_sysreg_el1 (ctxt_sys_reg (ctxt,SCTLR2_EL1 ),SYS_SCTLR2 ); + +/* +* Publish the L2 view of HCR_EL2 to the HW if L1 is using NV3. +* Otherwise, the data is already in place in the L1's own VNCR. +*/ +if (is_nested_nv3_ctxt (ctxt_to_vcpu (ctxt))) +write_sysreg_s (ctxt_sys_reg (ctxt,NVHCR_EL2 ),SYS_NVHCR_EL2 ); }/* Read the VCPU state's PSTATE, but translate (v)EL2 to EL1. */
arch/arm64/kvm/hyp/vhe/switch.c +8 -2
@@ -71,7 +71,10 @@ static u64 __compute_hcr(struct kvm_vcpu *vcpu) hcr |=HCR_NV1 ;/* Publish the guest's view of HCR_EL2 to the HW */ -__vcpu_assign_sys_reg (vcpu,NVHCR_EL2 ,__vcpu_sys_reg (vcpu,HCR_EL2 )); +if (cpus_have_final_cap (ARM64_HAS_NV3 ) &&vcpu_el2_e2h_is_set (vcpu)) +write_sysreg_s (__vcpu_sys_reg (vcpu,HCR_EL2 ),SYS_NVHCR_EL2 ); +else +__vcpu_assign_sys_reg (vcpu,NVHCR_EL2 ,__vcpu_sys_reg (vcpu,HCR_EL2 ));/* * Nothing in HCR_EL2 should impact running in hypervisor
@@ -567,7 +570,10 @@ static void fixup_nv_guest_exit(struct kvm_vcpu *vcpu) *vcpu_cpsr (vcpu) |= mode;/* Publish the latest HCR_EL2 to the emulation */ - hcr =__vcpu_sys_reg (vcpu,NVHCR_EL2 ); + hcr = (cpus_have_final_cap (ARM64_HAS_NV3 ) && +vcpu_el2_e2h_is_set (vcpu)) ? +read_sysreg_s (SYS_NVHCR_EL2 ) : +__vcpu_sys_reg (vcpu,NVHCR_EL2 );__vcpu_assign_sys_reg (vcpu,HCR_EL2 , hcr); }
Signature · 1 line
2.47.3
[email protected]
AI reviewRe: [PATCH v4 23/29] KVM: arm64: Add NVHCR_EL2 context switching
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] TOCTOU vulnerability in NV3 context switching leading to cross-VM information leak --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] TOCTOU vulnerability in NV3 context switching leading to cross-VM information leak --
--- Patch [23]: [PATCH v4 23/29] KVM: arm64: Add NVHCR_EL2 context switching --- commit f6089755f833008f36c0718e0727dbb85a4f24ae
Author: Marc Zyngier <[email protected]>
KVM: arm64: Add NVHCR_EL2 context switching
This commit introduces NVHCR_EL2 context switching. For a guest in hyp context, NVHCR_EL2 reflects HCR_EL2 and is eagerly synced. For an L2 guest, it is context switched like any EL1 register, conditioned on the guest using NV3.
arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h static inline void __sysreg_save_el1_state(struct kvm_cpu_context *ctxt) · line 178
Quoted 12 lines
arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h +4
-0@@ -172,6 +172,10 @@ static inline void __sysreg_save_el1_state(struct kvm_cpu_context *ctxt)if (ctxt_has_sctlr2 (ctxt))ctxt_sys_reg (ctxt,SCTLR2_EL1 ) =read_sysreg_el1 (SYS_SCTLR2 ); + +/* Retrieve L2's HCR_EL2, and save it for future use */ +if (is_nested_nv3_ctxt (ctxt_to_vcpu (ctxt))) +ctxt_sys_reg (ctxt,NVHCR_EL2 ) =read_sysreg_s (SYS_NVHCR_EL2 );
[Severity: High]
Does this introduce a TOCTOU regression leading to a cross-VM information leak?
KVM uses the guest-modifiable VNCR state to decide whether to read the hardware register, allowing a TOCTOU that reads uninitialized or stale hardware state.
Looking at is_nested_nv3_ctxt(), it evaluates HCRX_EL2.NVTGE directly from the guest-modifiable VNCR page:
arch/arm64/include/asm/kvm_emulate.h:is_nested_nv3_ctxt() {
...
return (... && (__vcpu_sys_reg(vcpu, HCRX_EL2) & HCRX_EL2_NVTGE));
}If an L1 guest concurrently modifies its VNCR page from another vCPU to change NVTGE from 0 to 1 during an L2 guest's execution, KVM might read from SYS_NVHCR_EL2 here on vcpu_put() or during sysreg trap handling. This could copy the leaked hardware state into the L2 guest's context, where it can be read.
[ ... ]
arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h static inline void __sysreg_restore_el1_state(struct kvm_cpu_context *ctxt, · line 298
Quoted 11 lines
arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h +7
-0@@ -285,6 +289,13 @@ static inline void __sysreg_restore_el1_state(struct kvm_cpu_context *ctxt,if (ctxt_has_sctlr2 (ctxt))write_sysreg_el1 (ctxt_sys_reg (ctxt,SCTLR2_EL1 ),SYS_SCTLR2 ); + +/* +* Publish the L2 view of HCR_EL2 to the HW if L1 is using NV3. +* Otherwise, the data is already in place in the L1's own VNCR. +*/ +if (is_nested_nv3_ctxt (ctxt_to_vcpu (ctxt))) +write_sysreg_s (ctxt_sys_reg (ctxt,NVHCR_EL2 ),SYS_NVHCR_EL2 );
[Severity: High]
Can the same concurrent VNCR page modification cause KVM to skip initializing the hardware register?
If NVTGE is 0 during vcpu_load(), KVM skips initializing the hardware register, leaving the previous VM's state in hardware. A malicious L1 guest can trigger this by configuring an L2 guest, spawning a concurrent vCPU to modify the VNCR page, and executing a trapped instruction in L2.
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=23
Marc Zyngier
maintainerPATCH v4 24/29 KVM: arm64: Engage NV3 ERET trap elision
In reply to Marc ZyngierWhen running on NV3 HW, always engage ERET trap elision when running the L1 context, as there is no benefit in not doing so.
An L1 can itself engage trap elision by setting its own view of HCRX_EL2.NVTGE==1, which will subsequently be honored.
--- arch/arm64/include/asm/kvm_emulate.h | 10 ++++++++++ arch/arm64/kvm/hyp/vhe/switch.c | 4 ++++ 2 files changed, 14 insertions(+)
arch/arm64/include/asm/kvm_emulate.h +10 -0
@@ -705,6 +705,16 @@ static inline void vcpu_set_hcrx(struct kvm_vcpu *vcpu)if (kvm_has_feat (kvm,ID_AA64ISAR1_EL1 ,LS64 ,LS64_V )) vcpu->arch.hcrx_el2 |= HCRX_EL2_EnASR; + +/* +* NV3 is a host-specific extension, and we always use +* it when present and that the guest uses NV. It may +* be hidden from the guest though. +*/ +if (cpus_have_final_cap (ARM64_HAS_NV3 ) && +vcpu_has_nv (vcpu) &&vcpu_el2_e2h_is_set (vcpu)) { + vcpu->arch.hcrx_el2 |=HCRX_EL2_NVTGE ; + } } }#endif /* __ARM64_KVM_EMULATE_H__ */
arch/arm64/kvm/hyp/vhe/switch.c +4 -0
@@ -345,6 +345,10 @@ static bool kvm_hyp_handle_eret(struct kvm_vcpu *vcpu, u64 *exit_code)u64 esr =kvm_vcpu_get_esr (vcpu);u64 spsr, elr, mode; +/* With NV3, the fast path is handled in HW */ +if (cpus_have_final_cap (ARM64_HAS_NV3 ) &&vcpu_el2_e2h_is_set (vcpu)) +return false; +/* * Going through the whole put/load motions is a waste of time * if this is a VHE guest hypervisor returning to its own
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 25/29 KVM: arm64: Engage NV3 TLBI trap elision
In reply to Marc ZyngierSimilarly to the ERET elision mechanism, FEAT_NV3 can elide TLBIs that only affects the guest's S1 translation. Enable this, with the express condition that the guest isn't NV2 aware, as we otherwise need to trap these TLBIs to deal with VNCR mappings.
--- arch/arm64/include/asm/kvm_emulate.h | 13 +++++++++++++ 1 file changed, 13 insertions(+)
arch/arm64/include/asm/kvm_emulate.h +13 -0
@@ -714,6 +714,19 @@ static inline void vcpu_set_hcrx(struct kvm_vcpu *vcpu)if (cpus_have_final_cap (ARM64_HAS_NV3 ) &&vcpu_has_nv (vcpu) &&vcpu_el2_e2h_is_set (vcpu)) { vcpu->arch.hcrx_el2 |=HCRX_EL2_NVTGE ; + +/* +* If the guest is NV2-capable, then we need to see +* all the TLBIs, as configured in HCR_EL2. +* Otherwise, relax the TLBI traps to only TGE=0. +*/ +if (!kvm_has_nv2 (vcpu->kvm)) { + vcpu->arch.hcrx_el2 |= (HCRX_EL2_NVnTTLB | + HCRX_EL2_NVnTTLBIS); + +if (kvm_has_feat (kvm,ID_AA64ISAR0_EL1 ,TLB ,OS )) + vcpu->arch.hcrx_el2 |= HCRX_EL2_NVnTTLBOS; + } } } }
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 26/29 KVM: arm64: Add FEAT_NV3 detection
In reply to Marc ZyngierNow that everything is in place to engage the FEAT_NV3 fast-path, add the detection code to cpufeature.c.
--- arch/arm64/kernel/cpufeature.c | 7 +++++++ 1 file changed, 7 insertions(+)
arch/arm64/kernel/cpufeature.c +7 -0
@@ -2627,6 +2627,13 @@ static const struct arm64_cpu_capabilities arm64_features[] = { .matches = has_cpuid_feature,ARM64_CPUID_FIELDS (ID_AA64MMFR4_EL1 , NV_frac,NV2P1 ) }, + { + .desc ="FEAT_NV3" , + .capability = ARM64_HAS_NV3, + .type = ARM64_CPUCAP_SYSTEM_FEATURE, + .matches = has_cpuid_feature, +ARM64_CPUID_FIELDS (ID_AA64MMFR4_EL1 , NV_frac,NV3 ) + }, { .capability =ARM64_HAS_32BIT_EL0_DO_NOT_USE , .type = ARM64_CPUCAP_SYSTEM_FEATURE,
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 27/29 KVM: arm64: Expose FEAT_NV3 to guests
In reply to Marc ZyngierFurther enable FEAT_NV3 by making it visible to NV guests.
--- arch/arm64/kvm/nested.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/arm64/kvm/nested.c +4 -2
@@ -1728,7 +1728,7 @@ u64 limit_nv_id_reg(struct kvm *kvm, u32 reg, u64 val)* You get EITHER * * - FEAT_VHE without FEAT_E2H0 -* - FEAT_NV limited to FEAT_NV2(p1) +* - FEAT_NV limited to FEAT_NV2(p1)/NV3 * - HCR_EL2.NV1 being RES0 * * OR
@@ -1741,7 +1741,9 @@ u64 limit_nv_id_reg(struct kvm *kvm, u32 reg, u64 val) val =0 ; }else { val &= ID_AA64MMFR4_EL1_NV_frac; -if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) +if (cpus_have_final_cap (ARM64_HAS_NV3 )) + val =ID_REG_LIMIT_FIELD_ENUM (val,ID_AA64MMFR4_EL1 , NV_frac,NV3 ); +else if (cpus_have_final_cap (ARM64_HAS_NV2P1 )) val =ID_REG_LIMIT_FIELD_ENUM (val,ID_AA64MMFR4_EL1 , NV_frac,NV2P1 );else val =SYS_FIELD_PREP_ENUM (ID_AA64MMFR4_EL1 , NV_frac,NV2_ONLY );
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 28/29 KVM: arm64: selftest: Add NVHCR_EL2 to get-reg-list
In reply to Marc ZyngierSince we now expose a new system register on NV3-capable HW when the guest is exposed to it, add it to the existing checker. FWIW.
--- tools/testing/selftests/kvm/arm64/get-reg-list.c | 2 ++ 1 file changed, 2 insertions(+)
tools/testing/selftests/kvm/arm64/get-reg-list.c +2 -0
@@ -67,6 +67,7 @@ static struct feature_id_reg feat_id_regs[] = {REG_FEAT (VDISR_EL2 ,ID_AA64PFR0_EL1 ,RAS ,IMP ),REG_FEAT (VSESR_EL2 ,ID_AA64PFR0_EL1 ,RAS ,IMP ),REG_FEAT (VNCR_EL2 ,ID_AA64MMFR4_EL1 , NV_frac,NV2_ONLY ), +REG_FEAT (NVHCR_EL2 ,ID_AA64MMFR4_EL1 , NV_frac,NV3 ),REG_FEAT (CNTHV_CTL_EL2 ,ID_AA64MMFR1_EL1 ,VH ,IMP ),REG_FEAT (CNTHV_CVAL_EL2 ,ID_AA64MMFR1_EL1 ,VH ,IMP ), REG_FEAT(ZCR_EL2, ID_AA64PFR0_EL1, SVE, IMP),
@@ -770,6 +771,7 @@ static __u64 el2_regs[] = {SYS_REG (SP_EL2 ),SYS_REG (VDISR_EL2 ),SYS_REG (VSESR_EL2 ), + SYS_REG(NVHCR_EL2), };static __u64 el2_e2h0_regs[] = {
Signature · 1 line
2.47.3
Marc Zyngier
maintainerPATCH v4 29/29 arm64: Add override for ID_AA64MMFR4_EL1.NV_frac
In reply to Marc ZyngierIn a very unsurprising turn of events, there is a large class of firmware that is totally unable to deal with FEAT_NV3, and doesn't set the required SCR2_EL3.NV3En bit, leading to an UNDEF exception or an unhandled trap to EL3, depending on the implementation.
Allow the unfortunate user to override ID_AA64MMFR4_EL1.NV_frac and get a working system. Hopefully firmware will be fixed before actually HW ships, but I have been there before... :-/
--- arch/arm64/include/asm/cpufeature.h | 1 + arch/arm64/kernel/cpufeature.c | 4 +++- arch/arm64/kernel/image-vars.h | 1 + arch/arm64/kernel/pi/idreg-override.c | 10 ++++++++++ 4 files changed, 15 insertions(+), 1 deletion(-)
arch/arm64/include/asm/cpufeature.h +1 -0
@@ -968,6 +968,7 @@ struct arm64_ftr_reg *get_arm64_ftr_reg(u32 sys_id);extern struct arm64_ftr_override id_aa64mmfr0_override;extern struct arm64_ftr_override id_aa64mmfr1_override;extern struct arm64_ftr_override id_aa64mmfr2_override; +extern struct arm64_ftr_override id_aa64mmfr4_override;extern struct arm64_ftr_override id_aa64pfr0_override;extern struct arm64_ftr_override id_aa64pfr1_override;extern struct arm64_ftr_override id_aa64zfr0_override;
arch/arm64/kernel/cpufeature.c +3 -1
@@ -785,6 +785,7 @@ static const struct arm64_ftr_bits ftr_raz[] = {struct arm64_ftr_override __read_mostly id_aa64mmfr0_override;struct arm64_ftr_override __read_mostly id_aa64mmfr1_override;struct arm64_ftr_override __read_mostly id_aa64mmfr2_override; +struct arm64_ftr_override __read_mostly id_aa64mmfr4_override;struct arm64_ftr_override __read_mostly id_aa64pfr0_override;struct arm64_ftr_override __read_mostly id_aa64pfr1_override;struct arm64_ftr_override __read_mostly id_aa64zfr0_override;
@@ -858,7 +859,8 @@ static const struct __ftr_reg_entry {ARM64_FTR_REG_OVERRIDE (SYS_ID_AA64MMFR2_EL1 , ftr_id_aa64mmfr2, &id_aa64mmfr2_override),ARM64_FTR_REG (SYS_ID_AA64MMFR3_EL1 , ftr_id_aa64mmfr3), -ARM64_FTR_REG (SYS_ID_AA64MMFR4_EL1 , ftr_id_aa64mmfr4), +ARM64_FTR_REG_OVERRIDE (SYS_ID_AA64MMFR4_EL1 , ftr_id_aa64mmfr4, + &id_aa64mmfr4_override),/* Op1 = 0, CRn = 10, CRm = 4 */ ARM64_FTR_REG(SYS_MPAMIDR_EL1, ftr_mpamidr),
arch/arm64/kernel/image-vars.h +1 -0
@@ -51,6 +51,7 @@ PI_EXPORT_SYM(id_aa64isar2_override);PI_EXPORT_SYM (id_aa64mmfr0_override);PI_EXPORT_SYM (id_aa64mmfr1_override);PI_EXPORT_SYM (id_aa64mmfr2_override); +PI_EXPORT_SYM (id_aa64mmfr4_override);PI_EXPORT_SYM (id_aa64pfr0_override);PI_EXPORT_SYM (id_aa64pfr1_override);PI_EXPORT_SYM (id_aa64smfr0_override);
arch/arm64/kernel/pi/idreg-override.c +10 -0
@@ -106,6 +106,15 @@ static const struct ftr_set_desc mmfr2 __prel64_initconst = { }, }; +static const struct ftr_set_desc mmfr4 __prel64_initconst = { + .name ="id_aa64mmfr4" , + .override = &id_aa64mmfr4_override, + .fields = { +FIELD ("nv_frac" , ID_AA64MMFR4_EL1_NV_frac_SHIFT,NULL ), + {} + }, +}; +static bool __init pfr0_sve_filter (u64 val) {/*
@@ -220,6 +229,7 @@ PREL64(const struct ftr_set_desc, reg) regs[] __prel64_initconst = { { &mmfr0 }, { &mmfr1 }, { &mmfr2 }, + { &mmfr4 }, { &pfr0 }, { &pfr1 }, { &isar1 },
Signature · 1 line
2.47.3
Quoted 9 lines · Marc Zyngier
This is the fourth (and hopefully last) version of a series adding support for two extensions targeting Nested Virt on arm64:
-
FEAT_NV2p1is effectively a bug fix for two registers (CNTHCTL_EL2andCPTR_EL2) that are missing stateful bits when accessed from EL1 in a NV configuration. When this is present, the hypervisor can avoid a bunch of traps.[...]
Applied to next, thanks!
[01/29] arm64: sysreg: Emit RESx/UNKN values for Mapping/Fields definitions
https://git.kernel.org/kvmarm/kvmarm/c/6ae7fab7d9d3 [02/29] arm64: Update ID_AA64MMFR4_EL1 description to 2026-03 JSON release https://git.kernel.org/kvmarm/kvmarm/c/1441f717ab5c [03/29] KVM: arm64: Merge guest's HCRX_EL2 using NV_HCRX_GUEST_EXCLUDE https://git.kernel.org/kvmarm/kvmarm/c/4422d8b7502d [04/29] KVM: arm64: Drop __HCRX_EL2_* masks https://git.kernel.org/kvmarm/kvmarm/c/34a7514b87ac [05/29] KVM: arm64: Plumb HCRX_EL2.SRMASKEn in HCRX_EL2 sanitisation https://git.kernel.org/kvmarm/kvmarm/c/fed0a1e87c28 [06/29] KVM: arm64: Classify CPTR_EL2 as a SR_LOC_SPECIAL register https://git.kernel.org/kvmarm/kvmarm/c/0b72c4060057 [07/29] KVM: arm64: Don't evaluate HCR_EL2.NV nor HFGITR_EL2.ERET on ERET fast path https://git.kernel.org/kvmarm/kvmarm/c/c938c826c1ef [08/29] arm64: Add ARM64_HAS_NV2P1 capability https://git.kernel.org/kvmarm/kvmarm/c/0dbb49dcdbbe [09/29] KVM: arm64: Relax CPTR_EL2 handling when FEAT_NV2p1 is present https://git.kernel.org/kvmarm/kvmarm/c/649ba043262c [10/29] KVM: arm64: Relax CNTHCTL_EL2 handling when FEAT_NV2p1 is present https://git.kernel.org/kvmarm/kvmarm/c/7c614e169826 [11/29] KVM: arm64: Expose FEAT_NV2p1 to NV guests https://git.kernel.org/kvmarm/kvmarm/c/4eb776d0b984 [12/29] arm64: Add FEAT_NV2p1 detection https://git.kernel.org/kvmarm/kvmarm/c/1f4177ea49b5 [13/29] arm64: sysreg: Add NVHCR_EL2 description as a mirror of HCR_EL2 https://git.kernel.org/kvmarm/kvmarm/c/7c81d7c4ef6e [14/29] arm64: sysreg: Add HCRX_EL2 bits related to FEAT_NV3 https://git.kernel.org/kvmarm/kvmarm/c/9809bdf151e2 [15/29] arm64: Add ARM64_HAS_NV3 capability https://git.kernel.org/kvmarm/kvmarm/c/5be1b8de27dc [16/29] KVM: arm64: Split NV-specific exit fixups from the non-NV handling https://git.kernel.org/kvmarm/kvmarm/c/dbb45adfcdef [17/29] KVM: arm64: Add NV3 control bits to HCRX_EL2 sanitisation https://git.kernel.org/kvmarm/kvmarm/c/6ec15c963e53 [18/29] KVM: arm64: Add kvm_has_nv{2,3}() predicates https://git.kernel.org/kvmarm/kvmarm/c/3908e3027235 [19/29] KVM: arm64: Make HCR_EL2 a non-VNCR register https://git.kernel.org/kvmarm/kvmarm/c/3fc7867a95c0 [20/29] KVM: arm64: Add sanitisation for NVHCR_EL2 https://git.kernel.org/kvmarm/kvmarm/c/3b3ae6bfaca0 [21/29] KVM: arm64: Add NVHCR_EL2 handling to the sysreg array https://git.kernel.org/kvmarm/kvmarm/c/0c0b2ad26b36 [22/29] KVM: arm64: Add routing for NVHCR_EL2 trap https://git.kernel.org/kvmarm/kvmarm/c/d6debf95f1aa [23/29] KVM: arm64: Add NVHCR_EL2 context switching https://git.kernel.org/kvmarm/kvmarm/c/62ed180231f4 [24/29] KVM: arm64: Engage NV3 ERET trap elision https://git.kernel.org/kvmarm/kvmarm/c/c962853bf5e1 [25/29] KVM: arm64: Engage NV3 TLBI trap elision https://git.kernel.org/kvmarm/kvmarm/c/328ab9719ee0 [26/29] KVM: arm64: Add FEAT_NV3 detection https://git.kernel.org/kvmarm/kvmarm/c/a64183418a80 [27/29] KVM: arm64: Expose FEAT_NV3 to guests https://git.kernel.org/kvmarm/kvmarm/c/89cdbeb18382 [28/29] KVM: arm64: selftest: Add NVHCR_EL2 to get-reg-list https://git.kernel.org/kvmarm/kvmarm/c/1f4a10ff9090 [29/29] arm64: Add override for ID_AA64MMFR4_EL1.NV_frac https://git.kernel.org/kvmarm/kvmarm/c/13c164acef36
--
Best,
Oliver