Skip to content

KVM: arm64: Fixes for the exposed debug architecture

Oliver Upton kvm kvmarm linux-arm-kernel 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest3 patches 5 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/3KVM: arm64: Stash OSLSR_EL1 in the cpu contextarch/arm64/include/asm/kvm_host.h arch/arm64/kvm/sys_regs.c+7 -71 reply
2/3KVM: arm64: Allow the guest to change the OS Lock statusarch/arm64/kvm/hyp/include/hyp/sysreg-sr.h arch/arm64/kvm/sys_regs.c+24 -31 reply
3/3KVM: arm64: Raise KVM's reported debug architecture to v8.2arch/arm64/kvm/sys_regs.c+7 -23 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [PATCH 0/3] KVM: arm64: Fixes for the exposed debug architecture

I had a conversation with Marc about some of the quirks around the debug architecture on KVM and incorporated some of his suggestions into a series here. Of course, any glaring mistakes/choices made in this series is on me :-)

Anyhow:

KVM's implementation of the debug architecture is a bit deviant as it stands. For one, KVM handles the OS Lock as RAZ/WI, even though the architecture mandates it. Additionally, KVM advertises more than it can actually support: FEAT_DoubleLock is exposed as implemented to the guest, though OSDLR_EL1 is handled as RAZ/WI too.

Only v8.2+ revisions of the debug architecture permit implementations to omit DoubleLock. Fortunately, the delta between v8.0 and v8.2 is entirely focused on external debug, a feature that KVM does not support and likely never will. So, there isn't much of a hurdle to bump KVM's reported DebugVer to v8.2, thereby allowing KVM to omit DoubleLock from ID_AA64DFR0_EL1. Of the remaining bits of external debug visible to the guest, the only additional thing to address is the OSLAR_EL1 issue by simply context switching the host/guest values.

Patch 1 changes the way KVM backs OSLSR_EL1 in the sys reg table. Instead of returning a static value from its handler, stash a copy of it in kvm_cpu_context and return that when read.

Patch 2 makes the material change of allowing a guest to actually toggle the OSLK bit by redirecting writes to OSLAR_EL1.OSLK to OSLSR_EL1.OSLK. When saving context, simply stash the value of OSLSR_EL1. On resume, apply OSLSR_EL1.OSLK to OSLAR_EL1.OSLK.

Finally, Patch 3 raises the KVM debug architecture to v8.2 and exposes FEAT_DoubleLock as NI to the guest. With the changes to OSLAR_EL1 in this series, KVM now does what it says on the tin.

This series applies cleanly to 5.15-rc4, and was (lightly) tested by booting 5.15-rc4 as a kvmtool guest on this kernel.

Oliver Upton (3):
KVM: arm64: Stash OSLSR_EL1 in the cpu context
KVM: arm64: Allow the guest to change the OS Lock status
KVM: arm64: Raise KVM's reported debug architecture to v8.2

 arch/arm64/include/asm/kvm_host.h          |  1 +
 arch/arm64/kvm/hyp/include/hyp/sysreg-sr.h |  5 +++
 arch/arm64/kvm/sys_regs.c                  | 42 ++++++++++++++++------
 3 files changed, 37 insertions(+), 11 deletions(-)
Signature · 1 line
2.33.0.1079.g6e70778dc9-goog