Re: [PATCH v4 0/4] Add support for the Idle HLT intercept feature

[Manali Shukla <manali.shukla@amd.com>]

On 10/22/2024 11:18 AM, Manali Shukla wrote:
> The upcoming new Idle HLT Intercept feature allows for the HLT
> instruction execution by a vCPU to be intercepted by the hypervisor
> only if there are no pending V_INTR and V_NMI events for the vCPU.
> When the vCPU is expected to service the pending V_INTR and V_NMI
> events, the Idle HLT intercept won’t trigger. The feature allows the
> hypervisor to determine if the vCPU is actually idle and reduces
> wasteful VMEXITs.
> 
> The idle HLT intercept feature is used for enlightened guests who wish
> to securely handle the events. When an enlightened guest does a HLT
> while an interrupt is pending, hypervisor will not have a way to
> figure out whether the guest needs to be re-entered or not. The Idle
> HLT intercept feature allows the HLT execution only if there are no
> pending V_INTR and V_NMI events.
> 
> Presence of the Idle HLT Intercept feature is indicated via CPUID
> function Fn8000_000A_EDX[30].
> 
> Document for the Idle HLT intercept feature is available at [1].
> 
> This series is based on kvm-next/next (64dbb3a771a1) + [2].
> 
> Experiments done:
> ----------------
> 
> kvm_amd.avic is set to '0' for this experiment.
> 
> The below numbers represent the average of 10 runs.
> 
> Normal guest (L1)
> The below netperf command was run on the guest with smp = 1 (pinned).
> 
> netperf -H <host ip> -t TCP_RR -l 60
> ----------------------------------------------------------------
> |with Idle HLT(transactions/Sec)|w/o Idle HLT(transactions/Sec)|
> ----------------------------------------------------------------
> |         25645.7136            |        25773.2796            |
> ----------------------------------------------------------------
> 
> Number of transactions/sec with and without idle HLT intercept feature
> are almost same.
> 
> Nested guest (L2)
> The below netperf command was run on L2 guest with smp = 1 (pinned).
> 
> netperf -H <host ip> -t TCP_RR -l 60
> ----------------------------------------------------------------
> |with Idle HLT(transactions/Sec)|w/o Idle HLT(transactions/Sec)|
> ----------------------------------------------------------------
> |          5655.4468            |          5755.2189           |
> ----------------------------------------------------------------
> 
> Number of transactions/sec with and without idle HLT intercept feature
> are almost same.
> 
> Testing Done:
> - Tested the functionality for the Idle HLT intercept feature
>   using selftest svm_idle_hlt_test.
> - Tested SEV and SEV-ES guest for the Idle HLT intercept functionality.
> - Tested the Idle HLT intercept functionality on nested guest.
> 
> v3 -> v4
> - Drop the patches to add vcpu_get_stat() into a new series [2].
> - Added nested Idle HLT intercept support.
> 
> v2 -> v3
> - Incorporated Andrew's suggestion to structure vcpu_stat_types in
>   a way that each architecture can share the generic types and also
>   provide its own.
> 
> v1 -> v2
> - Done changes in svm_idle_hlt_test based on the review comments from Sean.
> - Added an enum based approach to get binary stats in vcpu_get_stat() which
>   doesn't use string to get stat data based on the comments from Sean.
> - Added self_halt() and cli() helpers based on the comments from Sean.
> 
> [1]: AMD64 Architecture Programmer's Manual Pub. 24593, April 2024,
>      Vol 2, 15.9 Instruction Intercepts (Table 15-7: IDLE_HLT).
>      https://bugzilla.kernel.org/attachment.cgi?id=306250
> 
> [2]: https://lore.kernel.org/kvm/20241021062226.108657-1-manali.shukla@amd.com/T/#t
> 
> Manali Shukla (4):
>   x86/cpufeatures: Add CPUID feature bit for Idle HLT intercept
>   KVM: SVM: Add Idle HLT intercept support
>   KVM: nSVM: implement the nested idle halt intercept
>   KVM: selftests: KVM: SVM: Add Idle HLT intercept test
> 
>  arch/x86/include/asm/cpufeatures.h            |  1 +
>  arch/x86/include/asm/svm.h                    |  1 +
>  arch/x86/include/uapi/asm/svm.h               |  2 +
>  arch/x86/kvm/governed_features.h              |  1 +
>  arch/x86/kvm/svm/nested.c                     |  7 ++
>  arch/x86/kvm/svm/svm.c                        | 15 +++-
>  tools/testing/selftests/kvm/Makefile          |  1 +
>  .../selftests/kvm/include/x86_64/processor.h  |  1 +
>  .../selftests/kvm/x86_64/svm_idle_hlt_test.c  | 89 +++++++++++++++++++
>  9 files changed, 115 insertions(+), 3 deletions(-)
>  create mode 100644 tools/testing/selftests/kvm/x86_64/svm_idle_hlt_test.c
> 
> 
> base-commit: c8d430db8eec7d4fd13a6bea27b7086a54eda6da
> prerequisite-patch-id: ca912571db5c004f77b70843b8dd35517ff1267f
> prerequisite-patch-id: 164ea3b4346f9e04bc69819278d20f5e1b5df5ed
> prerequisite-patch-id: 90d870f426ebc2cec43c0dd89b701ee998385455
> prerequisite-patch-id: 45812b799c517a4521782a1fdbcda881237e1eda

A gentle reminder.

-Manali