Skip to contentSkip to conversation

ci/cirrus-ci-fedora.yml: Disable the "memory" test in the KVM job

PATCH In review Thomas Huth 3 participants 4 messages Apr 3, 2023 KVM

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ci/cirrus-ci-fedora.yml: Disable the "memory" test in the KVM job

Two of the sub-tests are currently failing in the Fedora KVM job on Cirrus-CI:

FAIL: clflushopt (ABSENT)
FAIL: clwb (ABSENT)

Looks like the features have been marked as disabled in the L0 host, while the hardware supports them. Since neither VMX nor SVM have intercept controls for the instructions, KVM has no way to enforce the guest's CPUID model, so this test is failing now in this environment. There's not much we can do here except for disabling the test here.

---
 ci/cirrus-ci-fedora.yml | 1 -
 1 file changed, 1 deletion(-)
ci/cirrus-ci-fedora.yml +0 -1
index 918c9a36..c83aff46 100644
@@ -35,7 +35,6 @@ fedora_task:
         ioapic
         ioapic-split
         kvmclock_test
-        memory
         pcid-asymmetric
         pcid-disabled
         pcid-enabled
Signature · 1 line
2.31.1
#1 lore raw
Quoted 22 lines · Thomas Huth

Two of the sub-tests are currently failing in the Fedora KVM job on Cirrus-CI:

FAIL: clflushopt (ABSENT)
FAIL: clwb (ABSENT)

Looks like the features have been marked as disabled in the L0 host, while the hardware supports them. Since neither VMX nor SVM have intercept controls for the instructions, KVM has no way to enforce the guest's CPUID model, so this test is failing now in this environment. There's not much we can do here except for disabling the test here.

---
 ci/cirrus-ci-fedora.yml | 1 -
 1 file changed, 1 deletion(-)
ci/cirrus-ci-fedora.yml +0 -0
index 918c9a36..c83aff46 100644

ci/cirrus-ci-fedora.yml fedora_task: · line 38

Quoted 4 lines
ci/cirrus-ci-fedora.yml +0 -1
@@ -35,7 +35,6 @@ fedora_task:
         ioapic
         ioapic-split
         kvmclock_test
-        memory

What if we fix the test by making it TCG-only? None of the other instructions in the test have intercepts, i.e. will also fail if the instructions are supported in bare metal but not the test VM.

An alternative would be to force emulation when using KVM, but KVM doesn't currently emulating pcommit (deprecated by Intel), clwb, or any of the fence instructions (at least, not afaict; I'm somewhat surprised *fence isn't "required").

x86/unittests.cfg +1 -0
index f324e32d..5afb5dad 100644
@@ -185,6 +185,7 @@ arch = x86_64
 
 [memory]
 file = memory.flat
+accel = tcg
 extra_params = -cpu max
 arch = x86_64
maintainer
#2 lore raw
Quoted 9 lines · Sean Christopherson

An alternative would be to force emulation when using KVM, but KVM doesn't currently emulating pcommit (deprecated by Intel), clwb, or any of the fence instructions (at least, not afaict; I'm somewhat surprised *fence isn't "required").

x86/unittests.cfg +0 -0
index f324e32d..5afb5dad 100644

x86/unittests.cfg arch = x86_64 · line 191

Quoted 7 lines
x86/unittests.cfg +1 -0
@@ -185,6 +185,7 @@ arch = x86_64
 
  [memory]
  file = memory.flat
+accel = tcg
  extra_params = -cpu max
  arch = x86_64
 

I think we should just drop the "CPUID bit non-present" case.

Paolo

#3 lore raw
Quoted 5 lines · Paolo Bonzini
Quoted 4 lines · Sean Christopherson

An alternative would be to force emulation when using KVM, but KVM doesn't currently emulating pcommit (deprecated by Intel), clwb, or any of the fence instructions (at least, not afaict; I'm somewhat surprised *fence isn't "required").

Quoted 13 lines

x86/unittests.cfg arch = x86_64 · line 190

Quoted 11 lines
x86/unittests.cfg +1 -0
index f324e32d..5afb5dad 100644
@@ -185,6 +185,7 @@ arch = x86_64
  [memory]
  file = memory.flat
+accel = tcg
  extra_params = -cpu max
  arch = x86_64
 

I think we should just drop the "CPUID bit non-present" case.

Oh, yeah, that's even better.