Skip to contentSkip to conversation

nVMX: Test for EP4TA pointing at MMIO

PATCH Applied Sean Christopherson 1 participant 5 messages Sep 11, 2023 – Jun 5, 2024 KVM

lore
Share & tools
lore

Sean Christopherson

author reviewer
#0 lore raw

cover nVMX: Test for EP4TA pointing at MMIO

Finally posting the testcase for the fix to back !visible roots with a dummy root[*]. The testcase is very simple: point EP4TA at MMIO (TPM base), do VM-Enter, and expect the VM to not die.

[*] https://lore.kernel.org/all/[email protected]

Sean Christopherson (3):
nVMX: Use helpers to check for WB memtype and 4-level EPT support
nVMX: Use setup_dummy_ept() to configure EPT for test_ept_eptp() test
nVMX: Add a testcase for running L2 with EP4TA that points at MMIO

 x86/vmx_tests.c | 31 ++++++++++++++-----------------
 1 file changed, 14 insertions(+), 17 deletions(-)

base-commit: d4fba74a42d222d2cfdde65351fac3531a1d6f5c

Signature · 1 line
2.42.0.283.g2d96d420d3-goog
author reviewer
#1 lore raw

PATCH 1/3 nVMX: Use helpers to check for WB memtype and 4-level EPT support

Use is_ept_memtype_supported() and is_4_level_ept_supported() to check for basic EPT support instead of open coding checks on ept_vpid.val.

Opportunstically add a report() failure if 4-level EPT isn't supported, as support for 4-level paging is mandatory in any sane configuration.

---
 x86/vmx_tests.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
x86/vmx_tests.c +5 -2
index c1540d39..907cbe1c 100644
@@ -1039,11 +1039,14 @@ static int __setup_ept(u64 hpa, bool enable_ad)
 		printf("\tEPT is not supported\n");
 		return 1;
 	}
-	if (!(ept_vpid.val & EPT_CAP_WB)) {
+	if (!is_ept_memtype_supported(EPT_MEM_TYPE_WB)) {
 		printf("\tWB memtype for EPT walks not supported\n");
 		return 1;
 	}
-	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
+
+	if (!is_4_level_ept_supported()) {
+		/* Support for 4-level EPT is mandatory. */
+		report(false, "4-level EPT support check");
 		printf("\tPWL4 is not supported\n");
 		return 1;
 	}
Signature · 1 line
2.42.0.283.g2d96d420d3-goog
author reviewer
#2 lore raw

PATCH 2/3 nVMX: Use setup_dummy_ept() to configure EPT for test_ept_eptp() test

In reply to Sean Christopherson

Use setup_dummy_ept() instead of open coding a rough equivalent in test_ept_eptp().

---
 x86/vmx_tests.c | 23 ++++++-----------------
 1 file changed, 6 insertions(+), 17 deletions(-)
x86/vmx_tests.c +6 -17
index 907cbe1c..9afca475 100644
@@ -4658,28 +4658,17 @@ static void test_ept_eptp(void)
 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
 	u64 eptp_saved = vmcs_read(EPTP);
-	u32 primary = primary_saved;
-	u32 secondary = secondary_saved;
-	u64 eptp = eptp_saved;
+	u32 secondary;
+	u64 eptp;
 	u32 i, maxphysaddr;
 	u64 j, resv_bits_mask = 0;
 
-	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
-	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
-		report_skip("%s : \"CPU secondary\" and/or \"enable EPT\" exec control not supported", __func__);
-		return;
-	}
-
-	/* Support for 4-level EPT is mandatory. */
 	report(is_4_level_ept_supported(), "4-level EPT support check");
 
-	primary |= CPU_SECONDARY;
-	vmcs_write(CPU_EXEC_CTRL0, primary);
-	secondary |= CPU_EPT;
-	vmcs_write(CPU_EXEC_CTRL1, secondary);
-	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
-	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
-	vmcs_write(EPTP, eptp);
+	setup_dummy_ept();
+
+	secondary = vmcs_read(CPU_EXEC_CTRL1);
+	eptp = vmcs_read(EPTP);
 
 	for (i = 0; i < 8; i++) {
 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
Signature · 1 line
2.42.0.283.g2d96d420d3-goog
author reviewer
#3 lore raw

PATCH 3/3 nVMX: Add a testcase for running L2 with EP4TA that points at MMIO

In reply to Sean Christopherson

Add a testcase in test_ept_eptp() to verify that KVM doesn't inject a triple fault (or any other unexpected "error") if L1 runs L2 with an EP4TA that points at MMIO memory. For a very, very long time, KVM synthesized a triple fault in response to a "legal-but-garbage" EP4TA _before_ completing emulation of nested VM-Enter, which is architectural wrong and triggered various warnings in KVM.

Use the TPM base address for the MMIO backing, as KVM doesn't emulate a TPM (in-kernel) and practically the address is guaranteed to be MMIO.

Drop the manual test for 4-level EPT support, as __setup_ept() performs said check/test.

---
 x86/vmx_tests.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
x86/vmx_tests.c +6 -1
index 9afca475..69a7c4a2 100644
@@ -4663,7 +4663,12 @@ static void test_ept_eptp(void)
 	u32 i, maxphysaddr;
 	u64 j, resv_bits_mask = 0;
 
-	report(is_4_level_ept_supported(), "4-level EPT support check");
+	if (__setup_ept(0xfed40000, false)) {
+		report_skip("%s : EPT not supported", __func__);
+		return;
+	}
+
+	test_vmx_valid_controls();
 
 	setup_dummy_ept();
 
Signature · 1 line
2.42.0.283.g2d96d420d3-goog
author reviewer Applied
#4 lore raw
In reply to Sean Christopherson
Quoted 4 lines · Sean Christopherson

Finally posting the testcase for the fix to back !visible roots with a dummy root[*]. The testcase is very simple: point EP4TA at MMIO (TPM base), do VM-Enter, and expect the VM to not die.

Quoted 8 lines

[*] https://lore.kernel.org/all/[email protected]

Sean Christopherson (3):
nVMX: Use helpers to check for WB memtype and 4-level EPT support
nVMX: Use setup_dummy_ept() to configure EPT for test_ept_eptp() test
nVMX: Add a testcase for running L2 with EP4TA that points at MMIO

[...]

Applied to kvm-x86 next, thanks!

[1/3] nVMX: Use helpers to check for WB memtype and 4-level EPT support

      https://github.com/kvm-x86/kvm-unit-tests/commit/ee7c8d4e630a
[2/3] nVMX: Use setup_dummy_ept() to configure EPT for test_ept_eptp() test
      https://github.com/kvm-x86/kvm-unit-tests/commit/5676a8651066
[3/3] nVMX: Add a testcase for running L2 with EP4TA that points at MMIO
      https://github.com/kvm-x86/kvm-unit-tests/commit/e87078ea65b0

--
https://github.com/kvm-x86/kvm-unit-tests/tree/next