vmx: Fix EPT accessed and dirty flag test
v3 cover vmx: Fix EPT accessed and dirty flag test
Fixes to EPT test plus improvements to test_vmx_vmlaunch reporting.
* Changes in v3:
- Split fixes into 3 patches.
- Provide more detailed patch descriptions.
- Improve change log entries.
* Changes in v2:
- Use the saved EPTP to restore the EPTP after each sub-test instead of manually unwinding what was done by the sub-test, which is error prone and hard to follow
- Explicitly setup a dummy EPTP, as calling the test in isolation will cause test failures due to lack a good starting EPTP
- Cleanup test_vmx_vmlaunch to generate clearer and
more consolidated test reports.
New format suggested by [email protected]
Cathy Avery (3):
vmx: Cleanup test_vmx_vmlaunch to generate clearer and more
consolidated test reports
vmx: Explicitly setup a dummy EPTP in EPT accessed and dirty flag test
vmx: Correctly refresh EPTP value in EPT accessed and dirty flag test
x86/vmx_tests.c | 39 +++++++++++++++++++++++---------------- 1 file changed, 23 insertions(+), 16 deletions(-)
Signature · 1 line
2.31.1
PATCH v3 1/3 vmx: Cleanup test_vmx_vmlaunch to generate clearer and more consolidated test reports
In cases when xerror is 0 ( the test is not expected to error ) and the test does error we get a confusing test result as the vmlaunch status is based on !xerror:
FAIL: Enable-EPT enabled; EPT page walk length 24: vmlaunch succeeds
This patch also eliminates the double call to report per launch and clarifies the failure messages. New format suggested by [email protected]
--- x86/vmx_tests.c | 21 ++++++++++++++------- 1 file changed, 14 insertions(+), 7 deletions(-)
x86/vmx_tests.c +14 -7
@@ -3392,14 +3392,21 @@ static void test_vmx_vmlaunch(u32 xerror)bool success =vmlaunch_succeeds ();u32 vmx_inst_err; -report (success == !xerror,"vmlaunch %s" , - !xerror ?"succeeds" :"fails" ); -if (!success && xerror) { - vmx_inst_err =vmcs_read (VMX_INST_ERROR ); +if (!success) + vmx_inst_err =vmcs_read (VMX_INST_ERROR ); + +if (success && !xerror) +report_pass ("VMLAUNCH succeeded as expected" ); +else if (success && xerror) +report_fail ("VMLAUNCH succeeded unexpectedly, wanted VM-Fail with error code = %d" , + xerror); +else if (!success && !xerror) +report_fail ("VMLAUNCH hit unexpected VM-Fail with error code = %d" , + vmx_inst_err); +else report (vmx_inst_err == xerror, -"VMX inst error is %d (actual %d)" , xerror, - vmx_inst_err); - } +"VMLAUNCH hit VM-Fail as expected, wanted error code %d, got %d" , + xerror, vmx_inst_err); }/*
Signature · 1 line
2.31.1
Re: [kvm-unit-tests v3 PATCH 1/3] vmx: Cleanup test_vmx_vmlaunch to generate clearer and more consolidated test reports
Quoted 19 lines · Cathy Avery
In cases when xerror is 0 ( the test is not expected to error ) and the test does error we get a confusing test result as the vmlaunch status is based on !xerror:
FAIL: Enable-EPT enabled; EPT page walk length 24: vmlaunch succeeds
This patch also eliminates the double call to report per launch and clarifies the failure messages. New format suggested by [email protected]
Signed-off-by Cathy Avery <[email protected]>--- x86/vmx_tests.c | 21 ++++++++++++++------- 1 file changed, 14 insertions(+), 7 deletions(-)x86/vmx_tests.c +0
-0
x86/vmx_tests.c static void test_vmx_vmlaunch(u32 xerror) · line 3396
Quoted 9 lines
x86/vmx_tests.c +2
-4@@ -3392,14 +3392,21 @@ static void test_vmx_vmlaunch(u32 xerror)bool success =vmlaunch_succeeds ();u32 vmx_inst_err; -report (success == !xerror,"vmlaunch %s" , - !xerror ?"succeeds" :"fails" ); -if (!success && xerror) { - vmx_inst_err =vmcs_read (VMX_INST_ERROR ); +if (!success) + vmx_inst_err =vmcs_read (VMX_INST_ERROR );
Needs to be indended. With that fixed,
PATCH v3 2/3 vmx: Explicitly setup a dummy EPTP in EPT accessed and dirty flag test
In reply to Cathy Averytest_ept_eptp is not explicitly calling setup_dummy_ept() to initialize EPTP to a good starting value resulting in test failures when it is called in isolation or when EPTP has been changed by some previous test.
--- x86/vmx_tests.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
x86/vmx_tests.c +4 -2
@@ -4714,12 +4714,11 @@ 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 i, maxphysaddr;u64 j, resv_bits_mask =0 ; +u64 eptp_saved, eptp;if (!((ctrl_cpu_rev[0 ].clr &CPU_SECONDARY ) && (ctrl_cpu_rev[1 ].clr &CPU_EPT ))) {
@@ -4727,6 +4726,9 @@ static void test_ept_eptp(void)return ; } +setup_dummy_ept (); + eptp = eptp_saved =vmcs_read (EPTP ); +/* Support for 4-level EPT is mandatory. */ report (is_4_level_ept_supported (),"4-level EPT support check" );
Signature · 1 line
2.31.1
Re: [kvm-unit-tests v3 PATCH 2/3] vmx: Explicitly setup a dummy EPTP in EPT accessed and dirty flag test
Quoted 6 lines · Cathy Avery
test_ept_eptpis not explicitly callingsetup_dummy_ept()to initialize EPTP to a good starting value resulting in test failures when it is called in isolation or when EPTP has been changed by some previous test.Signed-off-by Cathy Avery <[email protected]>---
PATCH v3 3/3 vmx: Correctly refresh EPTP value in EPT accessed and dirty flag test
In reply to Cathy AveryIf ept_ad is not supported by the processor or has been turned off via kvm module param, test_ept_eptp() will incorrectly leave EPTP_AD_FLAG set in variable eptp causing the following failures of subsequent test_vmx_valid_controls calls:
FAIL: Enable-EPT enabled; reserved bits [11:7] 0: vmlaunch succeeds
FAIL: Enable-EPT enabled; reserved bits [63:N] 0: vmlaunch succeeds
Use the saved EPTP to restore the EPTP after each sub-test instead of manually unwinding what was done by the sub-test, which is error prone and hard to follow.
--- x86/vmx_tests.c | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-)
x86/vmx_tests.c +5 -7
@@ -4751,8 +4751,7 @@ static void test_ept_eptp(void)test_vmx_invalid_controls ();report_prefix_pop (); } - - eptp = (eptp & ~EPT_MEM_TYPE_MASK ) |6ul ; + eptp = eptp_saved;/* * Page walk length (bits 5:3). Note, the value in VMCS.EPTP "is 1
@@ -4771,9 +4770,7 @@ static void test_ept_eptp(void)test_vmx_invalid_controls ();report_prefix_pop (); } - - eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK ) | -3ul <<EPTP_PG_WALK_LEN_SHIFT ; + eptp = eptp_saved;/* * Accessed and dirty flag (bit 6)
@@ -4793,6 +4790,7 @@ static void test_ept_eptp(void) eptp |=EPTP_AD_FLAG ;test_eptp_ad_bit (eptp, false); } + eptp = eptp_saved;/* * Reserved bits [11:7] and [63:N]
@@ -4811,8 +4809,7 @@ static void test_ept_eptp(void)test_vmx_invalid_controls ();report_prefix_pop (); } - - eptp = (eptp & ~(EPTP_RESERV_BITS_MASK <<EPTP_RESERV_BITS_SHIFT )); + eptp = eptp_saved; maxphysaddr =cpuid_maxphyaddr ();for (i =0 ; i < (63 - maxphysaddr +1 ); i++) {
@@ -4831,6 +4828,7 @@ static void test_ept_eptp(void)test_vmx_invalid_controls ();report_prefix_pop (); } + eptp = eptp_saved; secondary &= ~(CPU_EPT |CPU_URG );vmcs_write (CPU_EXEC_CTRL1 , secondary);
Signature · 1 line
2.31.1
Re: [kvm-unit-tests v3 PATCH 3/3] vmx: Correctly refresh EPTP value in EPT accessed and dirty flag test
Quoted 9 lines · Cathy Avery
If
ept_adis not supported by the processor or has been turned off via kvm module param,test_ept_eptp()will incorrectly leaveEPTP_AD_FLAGset in variable eptp causing the following failures of subsequenttest_vmx_valid_controlscalls:FAIL: Enable-EPT enabled; reserved bits [11:7] 0: vmlaunch succeeds
FAIL: Enable-EPT enabled; reserved bits [63:N] 0: vmlaunch succeeds
Quoted 6 lines
Use the saved EPTP to restore the EPTP after each sub-test instead of manually unwinding what was done by the sub-test, which is error prone and hard to follow.
Signed-off-by Cathy Avery <[email protected]>---
Thanks much for seeing this through!