Skip to contentSkip to conversation

KVM: SVM: GIF and EFER.SVME are independent

PATCH Applied Yosry Ahmed 2 participants 9 messages Nov 21, 2025 – Jan 12

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v3 cover KVM: SVM: GIF and EFER.SVME are independent

Clearing EFER.SVME is not architected to set GIF, so GIF may be clear even when EFER.SVME is clear.

This is covered in the discussion at [1].

v2 -> v3:
- Keep setting GIF when force-leaving nested (Sean).
- Moved the relevant selftests patches from the series at [2] here (Sean).

v2: https://lore.kernel.org/kvm/[email protected]/

[1]https://lore.kernel.org/all/[email protected]/ [2]https://lore.kernel.org/kvm/[email protected]/

Jim Mattson (2):
KVM: SVM: Allow KVM_SET_NESTED_STATE to clear GIF when SVME==0
KVM: SVM: Don't set GIF when clearing EFER.SVME

Yosry Ahmed (2):
KVM: selftests: Use TEST_ASSERT_EQ() in test_vmx_nested_state()
KVM: selftests: Extend vmx_set_nested_state_test to cover SVM

 arch/x86/kvm/svm/nested.c                     |   6 +-
 arch/x86/kvm/svm/svm.c                        |   1 -
 tools/testing/selftests/kvm/Makefile.kvm      |   2 +-
 ...d_state_test.c => nested_set_state_test.c} | 128 ++++++++++++++++--
 4 files changed, 120 insertions(+), 17 deletions(-)
 rename tools/testing/selftests/kvm/x86/{vmx_set_nested_state_test.c => nested_set_state_test.c} (70%)

base-commit: 115d5de2eef32ac5cd488404b44b38789362dbe6

Signature · 1 line
2.52.0.rc2.455.g230fcf2819-goog
#1 lore raw

PATCH v3 1/4 KVM: SVM: Allow KVM_SET_NESTED_STATE to clear GIF when SVME==0

From: Jim Mattson <[email protected]>

GIF==0 together with EFER.SVME==0 is a valid architectural state. Don't return -EINVAL for KVM_SET_NESTED_STATE when this combination is specified.

Fixes cc440cdad5b7 ("KVM: nSVM: implement KVM_GET_NESTED_STATE and KVM_SET_NESTED_STATE")
---
 arch/x86/kvm/svm/nested.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kvm/svm/nested.c +2 -2
index c81005b24522..3e4bd8d69788 100644
@@ -1784,8 +1784,8 @@ static int svm_set_nested_state(struct kvm_vcpu *vcpu,
 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
 	 */
 	if (!(vcpu->arch.efer & EFER_SVME)) {
-		/* GIF=1 and no guest mode are required if SVME=0.  */
-		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
+		/* GUEST_MODE must be clear when SVME==0 */
+		if (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)
 			return -EINVAL;
 	}
 
Signature · 1 line
2.52.0.rc2.455.g230fcf2819-goog
#2 lore raw

Re: [PATCH v3 1/4] KVM: SVM: Allow KVM_SET_NESTED_STATE to clear GIF when SVME==0

Quoted 19 lines · Yosry Ahmed

From: Jim Mattson <[email protected]>

GIF==0 together with EFER.SVME==0 is a valid architectural state. Don't return -EINVAL for KVM_SET_NESTED_STATE when this combination is specified.

Fixes cc440cdad5b7 ("KVM: nSVM: implement KVM_GET_NESTED_STATE and KVM_SET_NESTED_STATE")
---
 arch/x86/kvm/svm/nested.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kvm/svm/nested.c +0 -0
index c81005b24522..3e4bd8d69788 100644

arch/x86/kvm/svm/nested.c static int svm_set_nested_state(struct kvm_vcpu *vcpu, · line 1788

Quoted 7 lines
arch/x86/kvm/svm/nested.c +2 -2
@@ -1784,8 +1784,8 @@ static int svm_set_nested_state(struct kvm_vcpu *vcpu,
 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
 	 */
 	if (!(vcpu->arch.efer & EFER_SVME)) {
-		/* GIF=1 and no guest mode are required if SVME=0.  */
-		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
+		/* GUEST_MODE must be clear when SVME==0 */
+		if (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)

Hmm, this is technically wrong, as it will allow KVM_STATE_NESTED_RUN_PENDING. Now, arguably KVM already has a flaw there as KVM allows KVM_STATE_NESTED_RUN_PENDING without KVM_STATE_NESTED_GUEST_MODE for SVME=1, but I'd prefer not to make the hole bigger.

The nested if-statement is also unnecessary.

How about this instead? (not yet tested)

	/*
	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
	 * If SVME is disabled, the only valid states are "none" and GIF=1
	 * (clearing SVME does NOT set GIF, i.e. GIF=0 is allowed).
	 */
	if (!(vcpu->arch.efer & EFER_SVME) && kvm_state->flags &&
	    kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
		return -EINVAL;
#3 lore raw

Re: [PATCH v3 1/4] KVM: SVM: Allow KVM_SET_NESTED_STATE to clear GIF when SVME==0

Quoted 37 lines · Sean Christopherson
Quoted 19 lines · Yosry Ahmed

From: Jim Mattson <[email protected]>

GIF==0 together with EFER.SVME==0 is a valid architectural state. Don't return -EINVAL for KVM_SET_NESTED_STATE when this combination is specified.

Fixes cc440cdad5b7 ("KVM: nSVM: implement KVM_GET_NESTED_STATE and KVM_SET_NESTED_STATE")
---
 arch/x86/kvm/svm/nested.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kvm/svm/nested.c +0 -0
index c81005b24522..3e4bd8d69788 100644

arch/x86/kvm/svm/nested.c static int svm_set_nested_state(struct kvm_vcpu *vcpu, · line 1788

Quoted 7 lines
arch/x86/kvm/svm/nested.c +2 -2
@@ -1784,8 +1784,8 @@ static int svm_set_nested_state(struct kvm_vcpu *vcpu,
 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
 	 */
 	if (!(vcpu->arch.efer & EFER_SVME)) {
-		/* GIF=1 and no guest mode are required if SVME=0.  */
-		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
+		/* GUEST_MODE must be clear when SVME==0 */
+		if (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)

Hmm, this is technically wrong, as it will allow KVM_STATE_NESTED_RUN_PENDING. Now, arguably KVM already has a flaw there as KVM allows KVM_STATE_NESTED_RUN_PENDING without KVM_STATE_NESTED_GUEST_MODE for SVME=1, but I'd prefer not to make the hole bigger.

The nested if-statement is also unnecessary.

How about this instead? (not yet tested)

Quoted 9 lines
	/*
	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
	 * If SVME is disabled, the only valid states are "none" and GIF=1
	 * (clearing SVME does NOT set GIF, i.e. GIF=0 is allowed).
	 */
	if (!(vcpu->arch.efer & EFER_SVME) && kvm_state->flags &&
	    kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
		return -EINVAL;

Looks good to me, with the tiny exception that at this point clearing SVME does set GIF. Maybe re-order the patches?

Let me know if you want me to send a new version or if you'll fix it up while applying.

#4 lore raw

Re: [PATCH v3 1/4] KVM: SVM: Allow KVM_SET_NESTED_STATE to clear GIF when SVME==0

Quoted 48 lines · Yosry Ahmed
Quoted 37 lines · Sean Christopherson
Quoted 19 lines · Yosry Ahmed

From: Jim Mattson <[email protected]>

GIF==0 together with EFER.SVME==0 is a valid architectural state. Don't return -EINVAL for KVM_SET_NESTED_STATE when this combination is specified.

Fixes cc440cdad5b7 ("KVM: nSVM: implement KVM_GET_NESTED_STATE and KVM_SET_NESTED_STATE")
---
 arch/x86/kvm/svm/nested.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kvm/svm/nested.c +0 -0
index c81005b24522..3e4bd8d69788 100644

arch/x86/kvm/svm/nested.c static int svm_set_nested_state(struct kvm_vcpu *vcpu, · line 1788

Quoted 7 lines
arch/x86/kvm/svm/nested.c +2 -2
@@ -1784,8 +1784,8 @@ static int svm_set_nested_state(struct kvm_vcpu *vcpu,
 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
 	 */
 	if (!(vcpu->arch.efer & EFER_SVME)) {
-		/* GIF=1 and no guest mode are required if SVME=0.  */
-		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
+		/* GUEST_MODE must be clear when SVME==0 */
+		if (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)

Hmm, this is technically wrong, as it will allow KVM_STATE_NESTED_RUN_PENDING. Now, arguably KVM already has a flaw there as KVM allows KVM_STATE_NESTED_RUN_PENDING without KVM_STATE_NESTED_GUEST_MODE for SVME=1, but I'd prefer not to make the hole bigger.

The nested if-statement is also unnecessary.

How about this instead? (not yet tested)

Quoted 9 lines
	/*
	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
	 * If SVME is disabled, the only valid states are "none" and GIF=1
	 * (clearing SVME does NOT set GIF, i.e. GIF=0 is allowed).
	 */
	if (!(vcpu->arch.efer & EFER_SVME) && kvm_state->flags &&
	    kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
		return -EINVAL;
Quoted 5 lines

Looks good to me, with the tiny exception that at this point clearing SVME does set GIF. Maybe re-order the patches?

Let me know if you want me to send a new version or if you'll fix it up while applying.

No need for a new version.

#5 lore raw

PATCH v3 2/4 KVM: SVM: Don't set GIF when clearing EFER.SVME

In reply to Yosry Ahmed

From: Jim Mattson <[email protected]>

Clearing EFER.SVME is not architected to set GIF. Don't set GIF when emulating a change to EFER that clears EFER.SVME.

However, keep setting GIF if clearing EFER.SVME causes force-leaving the nested guest through svm_leave_nested(), to maintain a sane behavior of not leaving GIF cleared after exiting the guest. In every other path, setting GIF is either correct/desirable, or irrelevant because the caller immediately and unconditionally sets/clears GIF.

This is more-or-less KVM defining HW behavior, but leaving GIF cleared would also be defining HW behavior anyway.

Note that if force-leaving the nested guest is considered a SHUTDOWN, then this could violate the APM-specified behavior:

If the processor enters the shutdown state (due to a triple fault for instance) while GIF is clear, it can only be restarted by means of a RESET.

However, a SHUTDOWN leaves the VMCB undefined, so there's not a lot that KVM can do in this case. Also, if vGIF is enabled on SHUTDOWN, KVM has no way of finding out of GIF was cleared.

The only way for KVM to handle this without making up HW behavior is to completely terminate the VM, so settle for doing the relatively "sane" thing of setting GIF when force-leaving nested.

Fixes c513f484c558 ("KVM: nSVM: leave guest mode when clearing EFER.SVME")
[yosry.ahmed: stitched together jmattson's patch and seanjc's diff]
---
 arch/x86/kvm/svm/nested.c | 2 ++
 arch/x86/kvm/svm/svm.c    | 1 -
 2 files changed, 2 insertions(+), 1 deletion(-)
arch/x86/kvm/svm/nested.c +2 -0
index 3e4bd8d69788..9f91272fb735 100644
@@ -1363,6 +1363,8 @@ void svm_leave_nested(struct kvm_vcpu *vcpu)
 		nested_svm_uninit_mmu_context(vcpu);
 		vmcb_mark_all_dirty(svm->vmcb);
 
+		svm_set_gif(svm, true);
+
 		if (kvm_apicv_activated(vcpu->kvm))
 			kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
 	}
arch/x86/kvm/svm/svm.c +0 -1
index f56c2d895011..f62e94b404f4 100644
@@ -215,7 +215,6 @@ int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 	if ((old_efer & EFER_SVME) != (efer & EFER_SVME)) {
 		if (!(efer & EFER_SVME)) {
 			svm_leave_nested(vcpu);
-			svm_set_gif(svm, true);
 			/* #GP intercept is still needed for vmware backdoor */
 			if (!enable_vmware_backdoor)
 				clr_exception_intercept(svm, GP_VECTOR);
Signature · 1 line
2.52.0.rc2.455.g230fcf2819-goog
#6 lore raw

PATCH v3 3/4 KVM: selftests: Use TEST_ASSERT_EQ() in test_vmx_nested_state()

In reply to Yosry Ahmed

The assert messages do not add much value, so use TEST_ASSERT_EQ(), which also nicely displays the addresses in hex. While at it, also assert the values of state->flags.

---
 tools/testing/selftests/kvm/x86/vmx_set_nested_state_test.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
tools/testing/selftests/kvm/x86/vmx_set_nested_state_test.c +4 -2
index 67a62a5a8895..b59a8a17084d 100644
@@ -241,8 +241,10 @@ void test_vmx_nested_state(struct kvm_vcpu *vcpu)
 	TEST_ASSERT(state->size >= sizeof(*state) && state->size <= state_sz,
 		    "Size must be between %ld and %d.  The size returned was %d.",
 		    sizeof(*state), state_sz, state->size);
-	TEST_ASSERT(state->hdr.vmx.vmxon_pa == -1ull, "vmxon_pa must be -1ull.");
-	TEST_ASSERT(state->hdr.vmx.vmcs12_pa == -1ull, "vmcs_pa must be -1ull.");
+
+	TEST_ASSERT_EQ(state->hdr.vmx.vmxon_pa, -1ull);
+	TEST_ASSERT_EQ(state->hdr.vmx.vmcs12_pa, -1ull);
+	TEST_ASSERT_EQ(state->flags, 0);
 
 	free(state);
 }
Signature · 1 line
2.52.0.rc2.455.g230fcf2819-goog
#7 lore raw

PATCH v3 4/4 KVM: selftests: Extend vmx_set_nested_state_test to cover SVM

In reply to Yosry Ahmed

Add test cases for the validation checks in svm_set_nested_state(), and allow the test to run with SVM as well as VMX. The SVM test also makes sure that KVM_SET_NESTED_STATE accepts GIF being set or cleared if EFER.SVME is cleared, verifying a recently fixed bug where GIF was incorrectly expected to always be set when EFER.SVME is cleared.

---
 tools/testing/selftests/kvm/Makefile.kvm      |   2 +-
 ...d_state_test.c => nested_set_state_test.c} | 122 ++++++++++++++++--
 2 files changed, 112 insertions(+), 12 deletions(-)
 rename tools/testing/selftests/kvm/x86/{vmx_set_nested_state_test.c => nested_set_state_test.c} (71%)
tools/testing/selftests/kvm/Makefile.kvm +1 -1
index 7ebf30a87a2b..7567c936202f 100644
@@ -92,6 +92,7 @@ TEST_GEN_PROGS_x86 += x86/nested_close_kvm_test
 TEST_GEN_PROGS_x86 += x86/nested_emulation_test
 TEST_GEN_PROGS_x86 += x86/nested_exceptions_test
 TEST_GEN_PROGS_x86 += x86/nested_invalid_cr3_test
+TEST_GEN_PROGS_x86 += x86/nested_set_state_test
 TEST_GEN_PROGS_x86 += x86/nested_tsc_adjust_test
 TEST_GEN_PROGS_x86 += x86/nested_tsc_scaling_test
 TEST_GEN_PROGS_x86 += x86/platform_info_test
@@ -120,7 +121,6 @@ TEST_GEN_PROGS_x86 += x86/vmx_exception_with_invalid_guest_state
 TEST_GEN_PROGS_x86 += x86/vmx_msrs_test
 TEST_GEN_PROGS_x86 += x86/vmx_invalid_nested_guest_state
 TEST_GEN_PROGS_x86 += x86/vmx_nested_la57_state_test
-TEST_GEN_PROGS_x86 += x86/vmx_set_nested_state_test
 TEST_GEN_PROGS_x86 += x86/apic_bus_clock_test
 TEST_GEN_PROGS_x86 += x86/xapic_ipi_test
 TEST_GEN_PROGS_x86 += x86/xapic_state_test
tools/testing/selftests/kvm/x86/nested_set_state_test.c renamed from tools/testing/selftests/kvm/x86/vmx_set_nested_state_test.c +111 -11

157 lines not on this page · raw message

Signature · 1 line
2.52.0.rc2.455.g230fcf2819-goog
maintainer Applied
#8 lore raw
In reply to Yosry Ahmed
Quoted 11 lines · Yosry Ahmed

Clearing EFER.SVME is not architected to set GIF, so GIF may be clear even when EFER.SVME is clear.

This is covered in the discussion at [1].

v2 -> v3:
- Keep setting GIF when force-leaving nested (Sean).
- Moved the relevant selftests patches from the series at [2] here (Sean).

[...]

Applied to kvm-x86 svm, with my suggested fixup and the order of 1 and 2 swapped. Thanks!

[1/4] KVM: SVM: Don't set GIF when clearing EFER.SVME

      https://github.com/kvm-x86/linux/commit/8312f1b9dd71
[2/4] KVM: SVM: Allow KVM_SET_NESTED_STATE to clear GIF when SVME==0
      https://github.com/kvm-x86/linux/commit/6f4d3ebc24c6
[3/4] KVM: selftests: Use TEST_ASSERT_EQ() in test_vmx_nested_state()
      https://github.com/kvm-x86/linux/commit/bda6ae6f2966
[4/4] KVM: selftests: Extend vmx_set_nested_state_test to cover SVM
      https://github.com/kvm-x86/linux/commit/ca2eccb953fd

--
https://github.com/kvm-x86/linux/tree/next