Public bug reported:
Package: linux-nvidia (Jammy)
Series: jammy
Instance: testflinger:nvidia-h100:hidon
Test suite: ubuntu_kvm_unit_tests
Sub-tests affected: vmx, la57
== Summary ==
vmx and la57 sub-tests of ubuntu_kvm_unit_tests fail consistently on the
nvidia-h100:hidon testflinger instance (NVIDIA DGX H100, Sapphire Rapids CPU)
regardless of the kernel version under test. Manual verification on the host
confirms the failures reproduce identically on kernels that previously passed
in Jenkins, indicating this is a host-side (hardware/firmware/qemu/test-suite)
issue rather than a kernel regression.
This blocks SRU verification for every linux-nvidia cycle and forces
unstable/failed results that need manual review each cycle.
== Symptoms ==
vmx sub-test:
VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
(Invalid VMCS pointer)
FAIL vmx
la57 sub-test:
FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
FAIL la57
== Test matrix (all on nvidia-h100:hidon) ==
-------------------Kernel under test------------------|---vmx--|--la57--| Source
------------------------------------------------------|--------|--------|-------
5.15.0-1105.106---------------------------------------|--PASS--|--PASS--|
Jenkins run 2670499 (2026-05-18, Hinted)
5.15.0-1105.106---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 2655441 (2026-06-11, first hard fail)
5.15.0-1107.107 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
5.15.0-1112.113---------------------------------------|--PASS--|--PASS--|
Jenkins run 3247468 (2026-09-25)
5.15.0-1112.113 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
5.15.0-1113.114---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3339441 (2026-10-07)
5.15.0-1114.115---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3391783 (2026-10-19)
5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a-|--FAIL--|--FAIL--|
Manual run
5.15.0-1114.115 + revert 2c2e241715e09 as well--------|--FAIL--|--FAIL--|
Manual run
== Analysis ==
1. The same kernel (5.15.0-1112.113) PASSED in Jenkins on 2026-09-25
(run 3247468) but FAILS when run manually on the same host today.
Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
when run manually today.
-> Not a kernel regression. Every kernel tested manually on the host
today fails identically, regardless of Jenkins history.
2. Reverting the three suspect KVM/x86 commits from the 1112->1113 delta
does NOT fix the failures:
- e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested VM-Enter
- f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
- 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
-> None of these commits are the cause.
3. The failure is intermittent in history: 1105.106 flipped between
PASS/Hinted/Failed across May-June 2026 before becoming consistently
failing from October 2026 onward.
-> Suggests a host state change (firmware, qemu, kvm-unit-tests
build, or hardware degradation) rather than a kernel change.
== Host details ==
- Instance: nvidia-h100:hidon (NVIDIA DGX H100, Sapphire Rapids)
- BIOS: 1.1.3 (10/30/2023)
- Microcode: revision 0x2b000661, date 2025-08-25
- QEMU: 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.31)
- kvm-unit-tests HEAD: d7386d28 (canonical-kernel-team sru branch)
== Proposed hint ==
Add a hint for ubuntu_kvm_unit_tests test-cases vmx and la57 on
instance-type nvidia-h100:hidon referencing this bug, so SRU cycles are
not blocked by this pre-existing host-side failure while it is being
investigated by the HW/SUT team.
== Next steps ==
- HW/SUT team to investigate hidon host state (BIOS/microcode update,
qemu version change, hardware degradation).
- Re-run vmx/la57 on hidon after host maintenance to confirm whether
the failures clear.
** Affects: linux-nvidia (Ubuntu)
Importance: High
Status: New
** Affects: linux-nvidia (Ubuntu Jammy)
Importance: High
Status: New
** Changed in: linux-nvidia (Ubuntu)
Importance: Undecided => High
** Also affects: linux-nvidia (Ubuntu Jammy)
Importance: Undecided
Status: New
** Changed in: linux-nvidia (Ubuntu Jammy)
Importance: Undecided => High
** Description changed:
Package: linux-nvidia (Jammy)
Series: jammy
Instance: testflinger:nvidia-h100:hidon
Test suite: ubuntu_kvm_unit_tests
Sub-tests affected: vmx, la57
== Summary ==
vmx and la57 sub-tests of ubuntu_kvm_unit_tests fail consistently on the
nvidia-h100:hidon testflinger instance (NVIDIA DGX H100, Sapphire Rapids CPU)
regardless of the kernel version under test. Manual verification on the host
confirms the failures reproduce identically on kernels that previously passed
in Jenkins, indicating this is a host-side (hardware/firmware/qemu/test-suite)
issue rather than a kernel regression.
This blocks SRU verification for every linux-nvidia cycle and forces
unstable/failed results that need manual review each cycle.
== Symptoms ==
vmx sub-test:
- VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
- (Invalid VMCS pointer)
- FAIL vmx
+ VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
+ (Invalid VMCS pointer)
+ FAIL vmx
la57 sub-test:
- FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
- FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
- SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
- FAIL la57
+ FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
+ FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
+ SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
+ FAIL la57
== Test matrix (all on nvidia-h100:hidon) ==
- Kernel under test | vmx | la57 | Source
+ Kernel under test--------| vmx----| la57---| Source
-------------------------|--------|--------|--------------------------
5.15.0-1105.106 | PASS | PASS | Jenkins run 2670499 (2026-05-18,
Hinted)
5.15.0-1105.106 | FAIL | FAIL | Jenkins run 2655441 (2026-06-11,
first hard fail)
5.15.0-1107.107 (manual) | FAIL | FAIL | Manual run on hidon (2026-10-22)
5.15.0-1112.113 | PASS | PASS | Jenkins run 3247468 (2026-09-25)
5.15.0-1112.113 (manual) | FAIL | FAIL | Manual run on hidon (2026-10-22)
5.15.0-1113.114 | FAIL | FAIL | Jenkins run 3339441 (2026-10-07)
5.15.0-1114.115 | FAIL | FAIL | Jenkins run 3391783 (2026-10-19)
5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a | FAIL | FAIL | Manual
run
5.15.0-1114.115 + revert 2c2e241715e09 as well | FAIL | FAIL | Manual
run
== Analysis ==
1. The same kernel (5.15.0-1112.113) PASSED in Jenkins on 2026-09-25
- (run 3247468) but FAILS when run manually on the same host today.
- Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
- when run manually today.
- -> Not a kernel regression. Every kernel tested manually on the host
- today fails identically, regardless of Jenkins history.
+ (run 3247468) but FAILS when run manually on the same host today.
+ Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
+ when run manually today.
+ -> Not a kernel regression. Every kernel tested manually on the host
+ today fails identically, regardless of Jenkins history.
2. Reverting the three suspect KVM/x86 commits from the 1112->1113 delta
- does NOT fix the failures:
- - e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested
VM-Enter
- - f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
- - 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
- -> None of these commits are the cause.
+ does NOT fix the failures:
+ - e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested
VM-Enter
+ - f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
+ - 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
+ -> None of these commits are the cause.
3. The failure is intermittent in history: 1105.106 flipped between
- PASS/Hinted/Failed across May-June 2026 before becoming consistently
- failing from October 2026 onward.
- -> Suggests a host state change (firmware, qemu, kvm-unit-tests
- build, or hardware degradation) rather than a kernel change.
+ PASS/Hinted/Failed across May-June 2026 before becoming consistently
+ failing from October 2026 onward.
+ -> Suggests a host state change (firmware, qemu, kvm-unit-tests
+ build, or hardware degradation) rather than a kernel change.
== Host details ==
- Instance: nvidia-h100:hidon (NVIDIA DGX H100, Sapphire Rapids)
- BIOS: 1.1.3 (10/30/2023)
- Microcode: revision 0x2b000661, date 2025-08-25
- QEMU: 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.31)
- kvm-unit-tests HEAD: d7386d28 (canonical-kernel-team sru branch)
== Proposed hint ==
Add a hint for ubuntu_kvm_unit_tests test-cases vmx and la57 on
instance-type nvidia-h100:hidon referencing this bug, so SRU cycles are
not blocked by this pre-existing host-side failure while it is being
investigated by the HW/SUT team.
== Next steps ==
- HW/SUT team to investigate hidon host state (BIOS/microcode update,
- qemu version change, hardware degradation).
+ qemu version change, hardware degradation).
- Re-run vmx/la57 on hidon after host maintenance to confirm whether
- the failures clear.
+ the failures clear.
** Description changed:
Package: linux-nvidia (Jammy)
Series: jammy
Instance: testflinger:nvidia-h100:hidon
Test suite: ubuntu_kvm_unit_tests
Sub-tests affected: vmx, la57
== Summary ==
vmx and la57 sub-tests of ubuntu_kvm_unit_tests fail consistently on the
nvidia-h100:hidon testflinger instance (NVIDIA DGX H100, Sapphire Rapids CPU)
regardless of the kernel version under test. Manual verification on the host
confirms the failures reproduce identically on kernels that previously passed
in Jenkins, indicating this is a host-side (hardware/firmware/qemu/test-suite)
issue rather than a kernel regression.
This blocks SRU verification for every linux-nvidia cycle and forces
unstable/failed results that need manual review each cycle.
== Symptoms ==
vmx sub-test:
VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
(Invalid VMCS pointer)
FAIL vmx
la57 sub-test:
FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
FAIL la57
== Test matrix (all on nvidia-h100:hidon) ==
- Kernel under test--------| vmx----| la57---| Source
+ Kernel under test--------|---vmx--|--la57--| Source
-------------------------|--------|--------|--------------------------
- 5.15.0-1105.106 | PASS | PASS | Jenkins run 2670499 (2026-05-18,
Hinted)
- 5.15.0-1105.106 | FAIL | FAIL | Jenkins run 2655441 (2026-06-11,
first hard fail)
- 5.15.0-1107.107 (manual) | FAIL | FAIL | Manual run on hidon (2026-10-22)
- 5.15.0-1112.113 | PASS | PASS | Jenkins run 3247468 (2026-09-25)
- 5.15.0-1112.113 (manual) | FAIL | FAIL | Manual run on hidon (2026-10-22)
- 5.15.0-1113.114 | FAIL | FAIL | Jenkins run 3339441 (2026-10-07)
- 5.15.0-1114.115 | FAIL | FAIL | Jenkins run 3391783 (2026-10-19)
+ 5.15.0-1105.106 |--PASS--|--PASS--| Jenkins run 2670499 (2026-05-18,
Hinted)
+ 5.15.0-1105.106 |--FAIL--|--FAIL--| Jenkins run 2655441 (2026-06-11,
first hard fail)
+ 5.15.0-1107.107 (manual) |--FAIL--|--FAIL--| Manual run on hidon (2026-10-22)
+ 5.15.0-1112.113 |--PASS--|--PASS--| Jenkins run 3247468 (2026-09-25)
+ 5.15.0-1112.113 (manual) |--FAIL--|--FAIL--| Manual run on hidon (2026-10-22)
+ 5.15.0-1113.114 |--FAIL--|--FAIL--| Jenkins run 3339441 (2026-10-07)
+ 5.15.0-1114.115 |--FAIL--|--FAIL--| Jenkins run 3391783 (2026-10-19)
5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a | FAIL | FAIL | Manual
run
5.15.0-1114.115 + revert 2c2e241715e09 as well | FAIL | FAIL | Manual
run
== Analysis ==
1. The same kernel (5.15.0-1112.113) PASSED in Jenkins on 2026-09-25
(run 3247468) but FAILS when run manually on the same host today.
Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
when run manually today.
-> Not a kernel regression. Every kernel tested manually on the host
today fails identically, regardless of Jenkins history.
2. Reverting the three suspect KVM/x86 commits from the 1112->1113 delta
does NOT fix the failures:
- e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested
VM-Enter
- f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
- 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
-> None of these commits are the cause.
3. The failure is intermittent in history: 1105.106 flipped between
PASS/Hinted/Failed across May-June 2026 before becoming consistently
failing from October 2026 onward.
-> Suggests a host state change (firmware, qemu, kvm-unit-tests
build, or hardware degradation) rather than a kernel change.
== Host details ==
- Instance: nvidia-h100:hidon (NVIDIA DGX H100, Sapphire Rapids)
- BIOS: 1.1.3 (10/30/2023)
- Microcode: revision 0x2b000661, date 2025-08-25
- QEMU: 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.31)
- kvm-unit-tests HEAD: d7386d28 (canonical-kernel-team sru branch)
== Proposed hint ==
Add a hint for ubuntu_kvm_unit_tests test-cases vmx and la57 on
instance-type nvidia-h100:hidon referencing this bug, so SRU cycles are
not blocked by this pre-existing host-side failure while it is being
investigated by the HW/SUT team.
== Next steps ==
- HW/SUT team to investigate hidon host state (BIOS/microcode update,
qemu version change, hardware degradation).
- Re-run vmx/la57 on hidon after host maintenance to confirm whether
the failures clear.
** Description changed:
Package: linux-nvidia (Jammy)
Series: jammy
Instance: testflinger:nvidia-h100:hidon
Test suite: ubuntu_kvm_unit_tests
Sub-tests affected: vmx, la57
== Summary ==
vmx and la57 sub-tests of ubuntu_kvm_unit_tests fail consistently on the
nvidia-h100:hidon testflinger instance (NVIDIA DGX H100, Sapphire Rapids CPU)
regardless of the kernel version under test. Manual verification on the host
confirms the failures reproduce identically on kernels that previously passed
in Jenkins, indicating this is a host-side (hardware/firmware/qemu/test-suite)
issue rather than a kernel regression.
This blocks SRU verification for every linux-nvidia cycle and forces
unstable/failed results that need manual review each cycle.
== Symptoms ==
vmx sub-test:
VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
(Invalid VMCS pointer)
FAIL vmx
la57 sub-test:
FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
FAIL la57
== Test matrix (all on nvidia-h100:hidon) ==
- Kernel under test--------|---vmx--|--la57--| Source
- -------------------------|--------|--------|--------------------------
- 5.15.0-1105.106 |--PASS--|--PASS--| Jenkins run 2670499 (2026-05-18,
Hinted)
- 5.15.0-1105.106 |--FAIL--|--FAIL--| Jenkins run 2655441 (2026-06-11,
first hard fail)
- 5.15.0-1107.107 (manual) |--FAIL--|--FAIL--| Manual run on hidon (2026-10-22)
- 5.15.0-1112.113 |--PASS--|--PASS--| Jenkins run 3247468 (2026-09-25)
- 5.15.0-1112.113 (manual) |--FAIL--|--FAIL--| Manual run on hidon (2026-10-22)
- 5.15.0-1113.114 |--FAIL--|--FAIL--| Jenkins run 3339441 (2026-10-07)
- 5.15.0-1114.115 |--FAIL--|--FAIL--| Jenkins run 3391783 (2026-10-19)
- 5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a | FAIL | FAIL | Manual
run
- 5.15.0-1114.115 + revert 2c2e241715e09 as well | FAIL | FAIL | Manual
run
+ Kernel under test-------------------------------------|---vmx--|--la57--|
Source
+
------------------------------------------------------|--------|--------|-------
+ 5.15.0-1105.106---------------------------------------|--PASS--|--PASS--|
Jenkins run 2670499 (2026-05-18, Hinted)
+ 5.15.0-1105.106---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 2655441 (2026-06-11, first hard fail)
+ 5.15.0-1107.107 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
+ 5.15.0-1112.113---------------------------------------|--PASS--|--PASS--|
Jenkins run 3247468 (2026-09-25)
+ 5.15.0-1112.113 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
+ 5.15.0-1113.114---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3339441 (2026-10-07)
+ 5.15.0-1114.115---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3391783 (2026-10-19)
+ 5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a-| FAIL | FAIL | Manual
run
+ 5.15.0-1114.115 + revert 2c2e241715e09 as well--------| FAIL | FAIL | Manual
run
== Analysis ==
1. The same kernel (5.15.0-1112.113) PASSED in Jenkins on 2026-09-25
(run 3247468) but FAILS when run manually on the same host today.
Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
when run manually today.
-> Not a kernel regression. Every kernel tested manually on the host
today fails identically, regardless of Jenkins history.
2. Reverting the three suspect KVM/x86 commits from the 1112->1113 delta
does NOT fix the failures:
- e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested
VM-Enter
- f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
- 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
-> None of these commits are the cause.
3. The failure is intermittent in history: 1105.106 flipped between
PASS/Hinted/Failed across May-June 2026 before becoming consistently
failing from October 2026 onward.
-> Suggests a host state change (firmware, qemu, kvm-unit-tests
build, or hardware degradation) rather than a kernel change.
== Host details ==
- Instance: nvidia-h100:hidon (NVIDIA DGX H100, Sapphire Rapids)
- BIOS: 1.1.3 (10/30/2023)
- Microcode: revision 0x2b000661, date 2025-08-25
- QEMU: 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.31)
- kvm-unit-tests HEAD: d7386d28 (canonical-kernel-team sru branch)
== Proposed hint ==
Add a hint for ubuntu_kvm_unit_tests test-cases vmx and la57 on
instance-type nvidia-h100:hidon referencing this bug, so SRU cycles are
not blocked by this pre-existing host-side failure while it is being
investigated by the HW/SUT team.
== Next steps ==
- HW/SUT team to investigate hidon host state (BIOS/microcode update,
qemu version change, hardware degradation).
- Re-run vmx/la57 on hidon after host maintenance to confirm whether
the failures clear.
** Description changed:
Package: linux-nvidia (Jammy)
Series: jammy
Instance: testflinger:nvidia-h100:hidon
Test suite: ubuntu_kvm_unit_tests
Sub-tests affected: vmx, la57
== Summary ==
vmx and la57 sub-tests of ubuntu_kvm_unit_tests fail consistently on the
nvidia-h100:hidon testflinger instance (NVIDIA DGX H100, Sapphire Rapids CPU)
regardless of the kernel version under test. Manual verification on the host
confirms the failures reproduce identically on kernels that previously passed
in Jenkins, indicating this is a host-side (hardware/firmware/qemu/test-suite)
issue rather than a kernel regression.
This blocks SRU verification for every linux-nvidia cycle and forces
unstable/failed results that need manual review each cycle.
== Symptoms ==
vmx sub-test:
VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
(Invalid VMCS pointer)
FAIL vmx
la57 sub-test:
FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
FAIL la57
== Test matrix (all on nvidia-h100:hidon) ==
Kernel under test-------------------------------------|---vmx--|--la57--|
Source
------------------------------------------------------|--------|--------|-------
5.15.0-1105.106---------------------------------------|--PASS--|--PASS--|
Jenkins run 2670499 (2026-05-18, Hinted)
5.15.0-1105.106---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 2655441 (2026-06-11, first hard fail)
5.15.0-1107.107 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
5.15.0-1112.113---------------------------------------|--PASS--|--PASS--|
Jenkins run 3247468 (2026-09-25)
5.15.0-1112.113 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
5.15.0-1113.114---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3339441 (2026-10-07)
5.15.0-1114.115---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3391783 (2026-10-19)
- 5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a-| FAIL | FAIL | Manual
run
- 5.15.0-1114.115 + revert 2c2e241715e09 as well--------| FAIL | FAIL | Manual
run
+ 5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a-|--FAIL--|--FAIL--|
Manual run
+ 5.15.0-1114.115 + revert 2c2e241715e09 as well--------|--FAIL--|--FAIL--|
Manual run
== Analysis ==
1. The same kernel (5.15.0-1112.113) PASSED in Jenkins on 2026-09-25
(run 3247468) but FAILS when run manually on the same host today.
Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
when run manually today.
-> Not a kernel regression. Every kernel tested manually on the host
today fails identically, regardless of Jenkins history.
2. Reverting the three suspect KVM/x86 commits from the 1112->1113 delta
does NOT fix the failures:
- e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested
VM-Enter
- f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
- 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
-> None of these commits are the cause.
3. The failure is intermittent in history: 1105.106 flipped between
PASS/Hinted/Failed across May-June 2026 before becoming consistently
failing from October 2026 onward.
-> Suggests a host state change (firmware, qemu, kvm-unit-tests
build, or hardware degradation) rather than a kernel change.
== Host details ==
- Instance: nvidia-h100:hidon (NVIDIA DGX H100, Sapphire Rapids)
- BIOS: 1.1.3 (10/30/2023)
- Microcode: revision 0x2b000661, date 2025-08-25
- QEMU: 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.31)
- kvm-unit-tests HEAD: d7386d28 (canonical-kernel-team sru branch)
== Proposed hint ==
Add a hint for ubuntu_kvm_unit_tests test-cases vmx and la57 on
instance-type nvidia-h100:hidon referencing this bug, so SRU cycles are
not blocked by this pre-existing host-side failure while it is being
investigated by the HW/SUT team.
== Next steps ==
- HW/SUT team to investigate hidon host state (BIOS/microcode update,
qemu version change, hardware degradation).
- Re-run vmx/la57 on hidon after host maintenance to confirm whether
the failures clear.
** Description changed:
Package: linux-nvidia (Jammy)
Series: jammy
Instance: testflinger:nvidia-h100:hidon
Test suite: ubuntu_kvm_unit_tests
Sub-tests affected: vmx, la57
== Summary ==
vmx and la57 sub-tests of ubuntu_kvm_unit_tests fail consistently on the
nvidia-h100:hidon testflinger instance (NVIDIA DGX H100, Sapphire Rapids CPU)
regardless of the kernel version under test. Manual verification on the host
confirms the failures reproduce identically on kernels that previously passed
in Jenkins, indicating this is a host-side (hardware/firmware/qemu/test-suite)
issue rather than a kernel regression.
This blocks SRU verification for every linux-nvidia cycle and forces
unstable/failed results that need manual review each cycle.
== Symptoms ==
vmx sub-test:
VM-Fail on vmlaunch: error number is 8. See Intel 30.4.
(Invalid VMCS pointer)
FAIL vmx
la57 sub-test:
FAIL: Write to MSR_CSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
FAIL: Write to MSR_LSTAR with value ffaaaaaaaaaaaaaa did NOT succeed as
expected
SUMMARY: 54 tests, 2 unexpected failures, 4 skipped
FAIL la57
== Test matrix (all on nvidia-h100:hidon) ==
- Kernel under test-------------------------------------|---vmx--|--la57--|
Source
+ -------------------Kernel under test------------------|---vmx--|--la57--|
Source
------------------------------------------------------|--------|--------|-------
5.15.0-1105.106---------------------------------------|--PASS--|--PASS--|
Jenkins run 2670499 (2026-05-18, Hinted)
5.15.0-1105.106---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 2655441 (2026-06-11, first hard fail)
5.15.0-1107.107 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
5.15.0-1112.113---------------------------------------|--PASS--|--PASS--|
Jenkins run 3247468 (2026-09-25)
5.15.0-1112.113 (manual)------------------------------|--FAIL--|--FAIL--|
Manual run on hidon (2026-10-22)
5.15.0-1113.114---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3339441 (2026-10-07)
5.15.0-1114.115---------------------------------------|--FAIL--|--FAIL--|
Jenkins run 3391783 (2026-10-19)
5.15.0-1114.115 + revert e72b7e34f5cc + f07b146eff14a-|--FAIL--|--FAIL--|
Manual run
5.15.0-1114.115 + revert 2c2e241715e09 as well--------|--FAIL--|--FAIL--|
Manual run
== Analysis ==
1. The same kernel (5.15.0-1112.113) PASSED in Jenkins on 2026-09-25
(run 3247468) but FAILS when run manually on the same host today.
Additionally, 5.15.0-1107.107 (older than the Jenkins PASS) also FAILS
when run manually today.
-> Not a kernel regression. Every kernel tested manually on the host
today fails identically, regardless of Jenkins history.
2. Reverting the three suspect KVM/x86 commits from the 1112->1113 delta
does NOT fix the failures:
- e72b7e34f5cc KVM: nVMX: Service local TLB flushes on failed nested
VM-Enter
- f07b146eff14a KVM: nVMX: Always flush vpid02 on first use
- 2c2e241715e09 x86/mm: Fix check/use ordering in switch_mm_irqs_off()
-> None of these commits are the cause.
3. The failure is intermittent in history: 1105.106 flipped between
PASS/Hinted/Failed across May-June 2026 before becoming consistently
failing from October 2026 onward.
-> Suggests a host state change (firmware, qemu, kvm-unit-tests
build, or hardware degradation) rather than a kernel change.
== Host details ==
- Instance: nvidia-h100:hidon (NVIDIA DGX H100, Sapphire Rapids)
- BIOS: 1.1.3 (10/30/2023)
- Microcode: revision 0x2b000661, date 2025-08-25
- QEMU: 6.2.0 (Debian 1:6.2+dfsg-2ubuntu6.31)
- kvm-unit-tests HEAD: d7386d28 (canonical-kernel-team sru branch)
== Proposed hint ==
Add a hint for ubuntu_kvm_unit_tests test-cases vmx and la57 on
instance-type nvidia-h100:hidon referencing this bug, so SRU cycles are
not blocked by this pre-existing host-side failure while it is being
investigated by the HW/SUT team.
== Next steps ==
- HW/SUT team to investigate hidon host state (BIOS/microcode update,
qemu version change, hardware degradation).
- Re-run vmx/la57 on hidon after host maintenance to confirm whether
the failures clear.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2171961
Title:
ubuntu_kvm_unit_tests: vmx and la57 fail on nvidia-h100:hidon
(Sapphire Rapids) regardless of kernel version
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-nvidia/+bug/2171961/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
