Re: [PATCH v2 00/20] KVM: Use atomic SRCU for gfn-to-pfn cache, reinstate guest mode for x86 nesting

[David Woodhouse <dwmw2@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 8358 bytes --]

On Thu, 2026-11-05 at 02:00 +0000, David Woodhouse wrote:
> From: David Woodhouse <dwmw@amazon.co.uk>
> 
> Based on v7.4-rc1 (kvm/next doesn't have the atomic SRCU flavour yet),
> this series:
> 
>  • Converts the GPC to use atomic SRCU, ditching the rwlock which was a
>    scaling and latency problem, and which PREEMPT_RT hated because it
>    turned the rwlock into a sleeping lock¹.
> 
>  • Fixes the pre-existing bugs which torture-testing that conversion
>    shook out: a guest-reachable dirty ring overflow from the hypercall
>    page write (128 pushes of the same gfn against 64 reserved entries,
>    without PML), and a lost TLB flush request in record_steal_time().
> 
>  • Adds KVM_XEN_VCPU_ATTR_TYPE_WRITE_HYPERCALL_PAGE, for VMMs which
>    intercept the guest's MSR write and need to replay it from host
>    context².
> 
>  • Cleans up the intentional lack of dirty-tracking of Xen shinfo and
>    vcpu_info pages (Sean).
> 
>  • Converts steal-time / preempted status to use gfn_to_pfn_cache.
>    (Carsten)
> 

I love it when sashiko-bot finds things which my own local runs don't.
I'm still iterating on the latter part of the series; I think it
probably makes sense to merge the first half (patches 1-9 above), and
the nested parts (below) can come later.

Sean, what do you think?

>  • Reinstates guest-mode pinning: replacing the old GUEST_USES_PFN
>    mode, letting nested state pages be cached and pinned while in
>    guest context.
> 
>  • nVMX: Uses GPC for the L1 MSR bitmap, and guest mode for APIC
>    access and vAPIC pages. (Fred)
> 
>  • nSVM: Uses GPC for vmcb12 / MSR-permissions / IO-permissions pages.
>    This cuts the vmcb12 transition cost measurably (selftest included).
> 
>  • Re-posts KVM_REQ_GET_NESTED_STATE_PAGES on memslot updates so the
>    pinned caches are revalidated lazily, riding out the transient
>    KVM_MEMSLOT_INVALID window of an in-flight update via the pending
>    request rather than ever entering L2 with a stale latched address.
> 
> I had briefly split out some of the preliminary fixes and cleanups
> into a separate series which didn't depend on atomic SRCU. But I've
> joined it all back together now that atomic SRCU is merged. And I've
> ditched half of those 'fixes' as a bad idea anyway³.
> 
> 
> Tested: per-commit builds across the series; KVM selftests including
> the new invalidation, transition and race tests on 192-way AMD, and
> the applicable subset on Intel under PROVE_LOCKING + PROVE_RCU +
> DEBUG_ATOMIC_SLEEP (which catches the attribute-ioctl srcu fix on the
> very first xen_shinfo_test run when it's reverted; that config is now
> a standing part of the testing here). The new mmu_notifier race test
> asserts that every one of 5000 event channel deliveries into HVA-based
> caches survives continuous MADV_DONTNEED invalidation of the backing
> page, and soaks at 10/10.
> 
> ¹ https://syzkaller.appspot.com/bug?extid=208f7f3e5f59c11aeb90
> ² https://lore.kernel.org/kvm/c855535b4262ecd41f67734d19e8f48a7f014c2a.camel@infradead.org/
> ³ https://lore.kernel.org/kvm/ac6b1891d6bdf3a855a5b840c462a6e7d464a146.camel@infradead.org/
> 
> v2: https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/gpc-srcu-v2
> 
>  • Absorb the fixes/cleanups which had been split out (and drop the
>    delivery-retry half of them³, plus the SCHEDOP_poll fix which Sean
>    is picking up separately)
>  • Fix the mmu_notifier walk to wait out a grace period still owed by
>    a concurrent mutator
>  • Keep kvm->srcu in the Xen attribute ioctls for the memslot lookups
>  • Pair smp_mb__after_atomic()/__after_spinlock() for the publish/veto
>    store->load ordering
>  • Nested: treat a refresh failure against a KVM_MEMSLOT_INVALID
>    placeholder as transient; re-set vmcs02 controls after a lazy
>    degradation heals; initialize caches once at vCPU creation
>  • New: dirty ring overflow fix, WRITE_HYPERCALL_PAGE attribute,
>    the two Xen race selftests
> 
> v1: https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/gpc-srcu
>     https://lore.kernel.org/kvm/20260920211920.928306-1-dwmw2@infradead.org/
> 
> Carsten Stollmaier (1):
>   KVM: x86: Use gfn_to_pfn_cache for steal time / preempted status
> 
> David Woodhouse (13):
>   KVM: x86/xen: Fill the hypercall page with a single guest write
>   KVM: x86/xen: Add KVM_XEN_VCPU_ATTR_TYPE_WRITE_HYPERCALL_PAGE
>   KVM: x86: Request the guest TLB flush from record_steal_time()
>   KVM: selftests: Actually deactivate the HVA-based shared_info cache in
>     xen_shinfo_test
>   KVM: pfncache: Use atomic SRCU for readers instead of a rwlock
>   KVM: pfncache: Add guest-mode pinning (GUEST_USES_PFN successor)
>   KVM: x86: Post KVM_REQ_GET_NESTED_STATE_PAGES on memslot updates
>   KVM: nSVM: Use a gfn_to_pfn_cache for the vmcb12 page
>   KVM: nSVM: Cache L1's MSR permissions map pages
>   KVM: nSVM: Cache L1's IO permissions map pages
>   KVM: selftests: Add nested transition benchmark
>   KVM: selftests: Add Xen evtchn vs memslot-update race test
>   KVM: selftests: Add Xen evtchn vs mmu_notifier race test for HVA
>     caches
> 
> Fred Griffoul (3):
>   KVM: nVMX: Implement cache for L1 MSR bitmap
>   KVM: nVMX: Use pinned pfncache for L1 APIC virtualization pages
>   KVM: selftests: Add nested VMX APIC cache invalidation test
> 
> Sean Christopherson (3):
>   KVM: x86/xen: Extract delivery of event to vCPU into a separate helper
>   KVM: x86/xen: Explicitly tag "shared info" page as never being dirty
>     tracked
>   KVM: x86/xen: Don't dirty track "vCPU info" page
> 
>  Documentation/virt/kvm/api.rst                |  27 +-
>  arch/x86/include/asm/kvm_host.h               |   2 +-
>  arch/x86/include/uapi/asm/kvm.h               |   3 +
>  arch/x86/kvm/msrs.c                           |   7 +-
>  arch/x86/kvm/svm/nested.c                     | 174 ++++-
>  arch/x86/kvm/svm/svm.c                        |  17 +
>  arch/x86/kvm/svm/svm.h                        |  29 +
>  arch/x86/kvm/vmx/nested.c                     | 517 +++++++++++--
>  arch/x86/kvm/vmx/nested.h                     |   1 +
>  arch/x86/kvm/vmx/vmx.c                        |  13 +-
>  arch/x86/kvm/vmx/vmx.h                        |  21 +-
>  arch/x86/kvm/x86.c                            | 209 ++++--
>  arch/x86/kvm/x86.h                            |  33 +
>  arch/x86/kvm/xen.c                            | 380 ++++++----
>  include/linux/kvm_host.h                      | 121 +++-
>  include/linux/kvm_types.h                     |  64 +-
>  include/linux/srcu.h                          |   7 +
>  tools/testing/selftests/kvm/Makefile.kvm      |   4 +
>  .../kvm/x86/nested_transition_bench.c         | 210 ++++++
>  .../selftests/kvm/x86/vmx_apic_update_test.c  | 323 +++++++++
>  .../kvm/x86/xen_evtchn_hva_notify_race_test.c | 188 +++++
>  .../kvm/x86/xen_evtchn_memslot_race_test.c    | 232 ++++++
>  .../selftests/kvm/x86/xen_shinfo_test.c       |   2 +-
>  .../selftests/kvm/x86/xen_vmcall_test.c       | 118 +++
>  virt/kvm/kvm_main.c                           |   9 +
>  virt/kvm/pfncache.c                           | 677 +++++++++++++++---
>  26 files changed, 2962 insertions(+), 426 deletions(-)
>  create mode 100644 tools/testing/selftests/kvm/x86/nested_transition_bench.c
>  create mode 100644 tools/testing/selftests/kvm/x86/vmx_apic_update_test.c
>  create mode 100644 tools/testing/selftests/kvm/x86/xen_evtchn_hva_notify_race_test.c
>  create mode 100644 tools/testing/selftests/kvm/x86/xen_evtchn_memslot_race_test.c
> 
> base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 6179 bytes --]