Architectural updates retroactively made FEAT_E2H0 optional, meaning
hardware can implement FEAT_VHE without FEAT_E2H0. On such CPUs,
HCR_EL2.E2H can reset to an unknown state and must be initialized early
so later code can reliably detect whether E2H mode is active.

Without these fixes, booting 6.6.y as a guest under KVM nested
virtualization will hang at boot.

This series targets 6.6.y specifically because these patches are all
present in 6.12.y and applying these patches to 6.1.y and presumably
older kernels has more conflicts and results in other issues booting
as a guest under nested virtualization. More work is needed to enable
that.

Changes in v2:
- Updated commit messages to reference correct upstream SHA1s.
- Preserved inline EL2 state initialization in Patch 4 to avoid
  unrelated code churn.
- Added upstream commit 3855a7b91d42 ("KVM: arm64: Initialize SCTLR_EL1 in
  __kvm_hyp_init_cpu()") as Patch 5.
- Verified boot and KVM initialization across all KVM execution modes
  (nVHE, hVHE/nested, VHE, and protected).

v1:
https://lore.kernel.org/kvmarm/20260701204342.2654385-1-coltonlewis@google.com/ (local)

Ahmed Genidi (1):
  KVM: arm64: Initialize SCTLR_EL1 in __kvm_hyp_init_cpu()

Marc Zyngier (4):
  arm64: sysreg: Add layout for ID_AA64MMFR4_EL1
  arm64: Treat HCR_EL2.E2H as RES1 when ID_AA64MMFR4_EL1.E2H0 is
    negative
  arm64: Fix early handling of FEAT_E2H0 not being implemented
  arm64: Revamp HCR_EL2.E2H RES1 detection

Mark Rutland (1):
  KVM: arm64: Initialize HCR_EL2.E2H early

 arch/arm64/include/asm/el2_setup.h   | 56 ++++++++++++++++++++++++++--
 arch/arm64/kernel/head.S             | 20 ++++------
 arch/arm64/kvm/hyp/nvhe/hyp-init.S   |  3 +-
 arch/arm64/kvm/hyp/nvhe/psci-relay.c |  3 ++
 arch/arm64/tools/sysreg              | 37 ++++++++++++++++++
 5 files changed, 101 insertions(+), 18 deletions(-)


base-commit: da47cbc254661aa66d61ef061485a7080305c4be
--
2.55.0.795.g602f6c329a-goog