KVM: Fix deadlock for invalid memslots

[PATCH v2 0/3] KVM: Fix deadlock for invalid memslots

Posted by Yan Zhao 1 month, 1 week ago

Hi,

This series addresses the deadlock issue encountered with invalid memslots
during prefaulting or TDX private EPT violations.

Patches 1-2 are the new fixes from Sean.
            Patch 1 is for the prefaulting case,
            patch 2 for the TDX private EPT violation case.

Patch 3 updates the selftest for prefaulting.
        The ioctl KVM_PRE_FAULT_MEMORY is now expected to return EAGAIN
	instead of ENOENT when prefaulting GFNs in an invalid memslot.

The TDX-specific selftest is not included in this series, though it's
passed locally.

v2:
- Use Sean suggested fixes (patches 1-2).
- Updated selftest for the prefault case accordingly.
- code base: kvm-x86-next-2025.08.21

v1:
https://lore.kernel.org/all/20250519023613.30329-1-yan.y.zhao@intel.com

Sean Christopherson (2):
  KVM: x86/mmu: Return -EAGAIN if userspace deletes/moves memslot during
    prefault
  KVM: TDX: Do not retry locally when the retry is caused by invalid
    memslot

Yan Zhao (1):
  KVM: selftests: Test prefault memory during concurrent memslot removal

 arch/x86/kvm/mmu/mmu.c                        | 10 +-
 arch/x86/kvm/vmx/tdx.c                        | 11 +++
 .../selftests/kvm/pre_fault_memory_test.c     | 94 +++++++++++++++----
 virt/kvm/kvm_main.c                           |  1 +
 4 files changed, 98 insertions(+), 18 deletions(-)

-- 
2.43.2

Re: [PATCH v2 0/3] KVM: Fix deadlock for invalid memslots

Posted by Sean Christopherson 2 weeks, 3 days ago

On Fri, 22 Aug 2025 15:03:04 +0800, Yan Zhao wrote:
> This series addresses the deadlock issue encountered with invalid memslots
> during prefaulting or TDX private EPT violations.
> 
> Patches 1-2 are the new fixes from Sean.
>             Patch 1 is for the prefaulting case,
>             patch 2 for the TDX private EPT violation case.
> 
> [...]

Applied 1 and 2 to kvm-x86 mmu.  I'll post a proper patch for the selftest once
I figure out what I intended with respect to range.size.

Thanks!

[1/3] KVM: x86/mmu: Return -EAGAIN if userspace deletes/moves memslot during prefault
      https://github.com/kvm-x86/linux/commit/3ccbf6f47098
[2/3] KVM: TDX: Do not retry locally when the retry is caused by invalid memslot
      https://github.com/kvm-x86/linux/commit/2bc2694fe20b

--
https://github.com/kvm-x86/linux/tree/next