** Description changed:
SRU Justification:
[Impact]
A series of upstream arm64 and KVM architecture patches are missing from
the Jammy and Noble kernels. These commits correct the early
initialization and handling of HCR_EL2.E2H regarding feature detection
for FEAT_E2H0 (systems omitting support for running a host at EL1
without Virtualization Host Extensions). The lack of these fixes on
Google Cloud Platform (-gcp) kernel running on ARM64 cloud instances can
result in unstable KVM state initialization, improper architectural
register detection, or boot-time dependency breakages. Because
individual patches in the series introduced tightly coupled edge-case
bugs that were sequentially resolved by the later commits, the entire
sequence must be backported as an atomic block to Noble release.
[Fix]
- Backport the upstream 4-commit sequence that revamps the general
- HCR_EL2.E2H RES1 detection safely and initializes it early. These
- patches (already present in the v7.0 base and partially upstreamed to
- stable branches such as 6.12.64) must be evaluated against the 6.8
- (Noble) codebase to cleanly resolve code drift in the early ARM64 CPU
- boot/setup phases.
+ Target patches:
+
+ 3944382fa6f2 ("arm64: Treat HCR_EL2.E2H as RES1 when ID_AA64MMFR4_EL1.E2H0 is
negative")
+ b3320142f3db ("arm64: Fix early handling of FEAT_E2H0 not being implemented")
+ 7a68b55ff39b ("KVM: arm64: Initialize HCR_EL2.E2H early")
+ ca88ecdce5f5 ("arm64: Revamp HCR_EL2.E2H RES1 detection")
+
+ Some additional patches were needed for these to apply/build cleanly:
[Test Plan]
Provision an ARM64-based Instance on Google Cloud Platform running the
target kernel version (Noble). Validate hardware virtualization
capability flags and boot log diagnostic buffers via dmesg | grep -i kvm
or dmesg | grep -i e2h. Verify that the KVM subsystem initializes
correctly and hypervisor execution states do not throw system traps or
register state corruptions under immediate host-guest context switching.
Execute the standard Ubuntu Kernel Regression Test (KRT) suite for
virtualization topologies on ARM64 architectures.
[Regression potential]
The regression potential is isolated strictly to ARM64 (aarch64)
platforms, with zero risk to x86 or other architectures. Because these
patches alter register initialization logic (HCR_EL2) at very early boot
and hypervisor init stages, regressions could manifest as boot failures,
kernel panics during early CPU feature detection, or broken nested
virtualization on specific ARM hardware implementations. However, this
risk is deeply mitigated by pulling the entire 4-commit sequence as a
comprehensive lifecycle block, directly mirroring stable upstream
patterns (such as those fed by the 6.12 stable branches).
[Other Info]
SF:00442966
** Description changed:
SRU Justification:
[Impact]
A series of upstream arm64 and KVM architecture patches are missing from
the Jammy and Noble kernels. These commits correct the early
initialization and handling of HCR_EL2.E2H regarding feature detection
for FEAT_E2H0 (systems omitting support for running a host at EL1
without Virtualization Host Extensions). The lack of these fixes on
Google Cloud Platform (-gcp) kernel running on ARM64 cloud instances can
result in unstable KVM state initialization, improper architectural
register detection, or boot-time dependency breakages. Because
individual patches in the series introduced tightly coupled edge-case
bugs that were sequentially resolved by the later commits, the entire
sequence must be backported as an atomic block to Noble release.
[Fix]
Target patches:
3944382fa6f2 ("arm64: Treat HCR_EL2.E2H as RES1 when ID_AA64MMFR4_EL1.E2H0 is
negative")
b3320142f3db ("arm64: Fix early handling of FEAT_E2H0 not being implemented")
7a68b55ff39b ("KVM: arm64: Initialize HCR_EL2.E2H early")
ca88ecdce5f5 ("arm64: Revamp HCR_EL2.E2H RES1 detection")
- Some additional patches were needed for these to apply/build cleanly:
+ Some additional patches may be needed for these to apply/build cleanly:
[Test Plan]
Provision an ARM64-based Instance on Google Cloud Platform running the
target kernel version (Noble). Validate hardware virtualization
capability flags and boot log diagnostic buffers via dmesg | grep -i kvm
or dmesg | grep -i e2h. Verify that the KVM subsystem initializes
correctly and hypervisor execution states do not throw system traps or
register state corruptions under immediate host-guest context switching.
Execute the standard Ubuntu Kernel Regression Test (KRT) suite for
virtualization topologies on ARM64 architectures.
[Regression potential]
The regression potential is isolated strictly to ARM64 (aarch64)
platforms, with zero risk to x86 or other architectures. Because these
patches alter register initialization logic (HCR_EL2) at very early boot
and hypervisor init stages, regressions could manifest as boot failures,
kernel panics during early CPU feature detection, or broken nested
virtualization on specific ARM hardware implementations. However, this
risk is deeply mitigated by pulling the entire 4-commit sequence as a
comprehensive lifecycle block, directly mirroring stable upstream
patterns (such as those fed by the 6.12 stable branches).
[Other Info]
SF:00442966
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2160745
Title:
KVM backports to 6.8 related to HCR_EL2.E2H
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-gcp/+bug/2160745/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
