[PATCH v12 00/12] KVM: guest_memfd: Add NUMA mempolicy support

Sean Christopherson posted 12 patches 4 months ago
There is a newer version of this series
include/uapi/linux/magic.h                    |   1 +
tools/testing/selftests/kvm/arm64/vgic_irq.c  |   2 +-
.../testing/selftests/kvm/guest_memfd_test.c  |  98 +++++
.../selftests/kvm/include/kvm_syscalls.h      |  81 +++++
.../testing/selftests/kvm/include/kvm_util.h  |  29 +-
tools/testing/selftests/kvm/include/numaif.h  | 110 +++---
.../selftests/kvm/kvm_binary_stats_test.c     |   4 +-
tools/testing/selftests/kvm/lib/kvm_util.c    |  55 +--
.../kvm/x86/private_mem_conversions_test.c    |   9 +-
.../selftests/kvm/x86/xapic_ipi_test.c        |   5 +-
virt/kvm/guest_memfd.c                        | 344 ++++++++++++++----
virt/kvm/kvm_main.c                           |   7 +-
virt/kvm/kvm_mm.h                             |   9 +-
13 files changed, 576 insertions(+), 178 deletions(-)
create mode 100644 tools/testing/selftests/kvm/include/kvm_syscalls.h
[PATCH v12 00/12] KVM: guest_memfd: Add NUMA mempolicy support
Posted by Sean Christopherson 4 months ago
Shivank's series to add support for NUMA-aware memory placement in
guest_memfd.  This is based on:

  https://github.com/kvm-x86/linux.git gmem

which is an unstable topic branch that contains the guest_memfd MMAP fixes
destined for 6.18 (to avoid conflicts), and three non-KVM changes related to
mempolicy that were in previous versions of this series (I want to keep this
version KVM focused, and AFAICT there is nothing left to discuss in the prep
paches).

Once 6.18-rc1 is cut I'll turn "gmem" into a proper topic branch, rebase it,
and freeze the hashes.

v12:
 - Add missing functionality to KVM selftests' existing numaif.h instead of
   linking to libnuma (which appears to have caveats with -static).
 - Add KVM_SYSCALL_DEFINE() infrastructure to reduce the boilerplate needed
   to wrap syscalls and/or to assert that a syscall succeeds.
 - Rename kvm_gmem to gmem_file, and use gmem_inode for the inode structure.
 - Track flags in a gmem_inode field instead of using i_private.
 - Add comments to call out subtleties in the mempolicy code (e.g. that
   returning NULL for vm_operations_struct.get_policy() is important for ABI
   reasons).
 - Improve debugability of guest_memfd_test (I kept generating SIGBUS when
   tweaking the tests).
 - Test mbind() with private memory (sadly, verifying placement with
   move_pages() doesn't work due to the dependency on valid page tables).

- V11: Rebase on kvm-next, remove RFC tag, use Ackerley's latest patch
       and fix a rcu race bug during kvm module unload.
- V10: Rebase on top of Fuad's V17. Use latest guest_memfd inode patch
       from Ackerley (with David's review comments). Use newer kmem_cache_create()
       API variant with arg parameter (Vlastimil)
- v9: Rebase on top of Fuad's V13 and incorporate review comments
- v8: Rebase on top of Fuad's V12: Host mmaping for guest_memfd memory.
- v7: Use inodes to store NUMA policy instead of file [0].
- v4-v6: Current approach using shared_policy support and vm_ops (based on
         suggestions from David [1] and guest_memfd bi-weekly upstream
         call discussion [2]).
- v3: Introduced fbind() syscall for VMM memory-placement configuration.
- v1,v2: Extended the KVM_CREATE_GUEST_MEMFD IOCTL to pass mempolicy.

[0] https://lore.kernel.org/all/diqzbjumm167.fsf@ackerleytng-ctop.c.googlers.com
[1] https://lore.kernel.org/all/6fbef654-36e2-4be5-906e-2a648a845278@redhat.com
[2] https://lore.kernel.org/all/2b77e055-98ac-43a1-a7ad-9f9065d7f38f@amd.com

Ackerley Tng (1):
  KVM: guest_memfd: Use guest mem inodes instead of anonymous inodes

Sean Christopherson (7):
  KVM: guest_memfd: Rename "struct kvm_gmem" to "struct gmem_file"
  KVM: guest_memfd: Add macro to iterate over gmem_files for a
    mapping/inode
  KVM: selftests: Define wrappers for common syscalls to assert success
  KVM: selftests: Report stacktraces SIGBUS, SIGSEGV, SIGILL, and SIGFPE
    by default
  KVM: selftests: Add additional equivalents to libnuma APIs in KVM's
    numaif.h
  KVM: selftests: Use proper uAPI headers to pick up mempolicy.h
    definitions
  KVM: guest_memfd: Add gmem_inode.flags field instead of using
    i_private

Shivank Garg (4):
  KVM: guest_memfd: Add slab-allocated inode cache
  KVM: guest_memfd: Enforce NUMA mempolicy using shared policy
  KVM: selftests: Add helpers to probe for NUMA support, and multi-node
    systems
  KVM: selftests: Add guest_memfd tests for mmap and NUMA policy support

 include/uapi/linux/magic.h                    |   1 +
 tools/testing/selftests/kvm/arm64/vgic_irq.c  |   2 +-
 .../testing/selftests/kvm/guest_memfd_test.c  |  98 +++++
 .../selftests/kvm/include/kvm_syscalls.h      |  81 +++++
 .../testing/selftests/kvm/include/kvm_util.h  |  29 +-
 tools/testing/selftests/kvm/include/numaif.h  | 110 +++---
 .../selftests/kvm/kvm_binary_stats_test.c     |   4 +-
 tools/testing/selftests/kvm/lib/kvm_util.c    |  55 +--
 .../kvm/x86/private_mem_conversions_test.c    |   9 +-
 .../selftests/kvm/x86/xapic_ipi_test.c        |   5 +-
 virt/kvm/guest_memfd.c                        | 344 ++++++++++++++----
 virt/kvm/kvm_main.c                           |   7 +-
 virt/kvm/kvm_mm.h                             |   9 +-
 13 files changed, 576 insertions(+), 178 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/include/kvm_syscalls.h


base-commit: 67033eaa5ea2cb67b6cdaa91d7f5c42bfafb36f7
-- 
2.51.0.710.ga91ca5db03-goog
Re: [PATCH v12 00/12] KVM: guest_memfd: Add NUMA mempolicy support
Posted by Ackerley Tng 4 months ago
Sean Christopherson <seanjc@google.com> writes:

> Shivank's series to add support for NUMA-aware memory placement in
> guest_memfd.  This is based on:
>
>   https://github.com/kvm-x86/linux.git gmem
>
> which is an unstable topic branch that contains the guest_memfd MMAP fixes
> destined for 6.18 (to avoid conflicts), and three non-KVM changes related to
> mempolicy that were in previous versions of this series

For future reference, these are the three specific patches:

[1] https://lore.kernel.org/all/20250827175247.83322-4-shivankg@amd.com/
[2] https://lore.kernel.org/all/20250827175247.83322-5-shivankg@amd.com/
[3] https://lore.kernel.org/all/20250827175247.83322-6-shivankg@amd.com/

Might have missed this, did we discuss how these 3 would get merged? I
noticed this patch was withdrawn, not sure what that means: [4]

[4] https://lore.kernel.org/all/20250625000155.62D08C4CEE3@smtp.kernel.org/

> (I want to keep this
> version KVM focused, and AFAICT there is nothing left to discuss in the prep
> paches).
>
> Once 6.18-rc1 is cut I'll turn "gmem" into a proper topic branch, rebase it,
> and freeze the hashes.
>
> v12:
>  - Add missing functionality to KVM selftests' existing numaif.h instead of
>    linking to libnuma (which appears to have caveats with -static).
>  - Add KVM_SYSCALL_DEFINE() infrastructure to reduce the boilerplate needed
>    to wrap syscalls and/or to assert that a syscall succeeds.
>  - Rename kvm_gmem to gmem_file, and use gmem_inode for the inode structure.
>  - Track flags in a gmem_inode field instead of using i_private.
>  - Add comments to call out subtleties in the mempolicy code (e.g. that
>    returning NULL for vm_operations_struct.get_policy() is important for ABI
>    reasons).
>  - Improve debugability of guest_memfd_test (I kept generating SIGBUS when
>    tweaking the tests).
>  - Test mbind() with private memory (sadly, verifying placement with
>    move_pages() doesn't work due to the dependency on valid page tables).
>
> - V11: Rebase on kvm-next, remove RFC tag, use Ackerley's latest patch
>        and fix a rcu race bug during kvm module unload.
> - V10: Rebase on top of Fuad's V17. Use latest guest_memfd inode patch
>        from Ackerley (with David's review comments). Use newer kmem_cache_create()
>        API variant with arg parameter (Vlastimil)
> - v9: Rebase on top of Fuad's V13 and incorporate review comments
> - v8: Rebase on top of Fuad's V12: Host mmaping for guest_memfd memory.
> - v7: Use inodes to store NUMA policy instead of file [0].
> - v4-v6: Current approach using shared_policy support and vm_ops (based on
>          suggestions from David [1] and guest_memfd bi-weekly upstream
>          call discussion [2]).
> - v3: Introduced fbind() syscall for VMM memory-placement configuration.
> - v1,v2: Extended the KVM_CREATE_GUEST_MEMFD IOCTL to pass mempolicy.
>
> [0] https://lore.kernel.org/all/diqzbjumm167.fsf@ackerleytng-ctop.c.googlers.com
> [1] https://lore.kernel.org/all/6fbef654-36e2-4be5-906e-2a648a845278@redhat.com
> [2] https://lore.kernel.org/all/2b77e055-98ac-43a1-a7ad-9f9065d7f38f@amd.com
>
> Ackerley Tng (1):
>   KVM: guest_memfd: Use guest mem inodes instead of anonymous inodes
>
> Sean Christopherson (7):
>   KVM: guest_memfd: Rename "struct kvm_gmem" to "struct gmem_file"
>   KVM: guest_memfd: Add macro to iterate over gmem_files for a
>     mapping/inode
>   KVM: selftests: Define wrappers for common syscalls to assert success
>   KVM: selftests: Report stacktraces SIGBUS, SIGSEGV, SIGILL, and SIGFPE
>     by default
>   KVM: selftests: Add additional equivalents to libnuma APIs in KVM's
>     numaif.h
>   KVM: selftests: Use proper uAPI headers to pick up mempolicy.h
>     definitions
>   KVM: guest_memfd: Add gmem_inode.flags field instead of using
>     i_private
>
> Shivank Garg (4):
>   KVM: guest_memfd: Add slab-allocated inode cache
>   KVM: guest_memfd: Enforce NUMA mempolicy using shared policy
>   KVM: selftests: Add helpers to probe for NUMA support, and multi-node
>     systems
>   KVM: selftests: Add guest_memfd tests for mmap and NUMA policy support
>
>  include/uapi/linux/magic.h                    |   1 +
>  tools/testing/selftests/kvm/arm64/vgic_irq.c  |   2 +-
>  .../testing/selftests/kvm/guest_memfd_test.c  |  98 +++++
>  .../selftests/kvm/include/kvm_syscalls.h      |  81 +++++
>  .../testing/selftests/kvm/include/kvm_util.h  |  29 +-
>  tools/testing/selftests/kvm/include/numaif.h  | 110 +++---
>  .../selftests/kvm/kvm_binary_stats_test.c     |   4 +-
>  tools/testing/selftests/kvm/lib/kvm_util.c    |  55 +--
>  .../kvm/x86/private_mem_conversions_test.c    |   9 +-
>  .../selftests/kvm/x86/xapic_ipi_test.c        |   5 +-
>  virt/kvm/guest_memfd.c                        | 344 ++++++++++++++----
>  virt/kvm/kvm_main.c                           |   7 +-
>  virt/kvm/kvm_mm.h                             |   9 +-
>  13 files changed, 576 insertions(+), 178 deletions(-)
>  create mode 100644 tools/testing/selftests/kvm/include/kvm_syscalls.h
>
>
> base-commit: 67033eaa5ea2cb67b6cdaa91d7f5c42bfafb36f7
> -- 
> 2.51.0.710.ga91ca5db03-goog
Re: [PATCH v12 00/12] KVM: guest_memfd: Add NUMA mempolicy support
Posted by Garg, Shivank 4 months ago

On 10/10/2025 2:28 AM, Ackerley Tng wrote:
> For future reference, these are the three specific patches:
> 
> [1] https://lore.kernel.org/all/20250827175247.83322-4-shivankg@amd.com/
> [2] https://lore.kernel.org/all/20250827175247.83322-5-shivankg@amd.com/
> [3] https://lore.kernel.org/all/20250827175247.83322-6-shivankg@amd.com/
> 
> Might have missed this, did we discuss how these 3 would get merged? I
> noticed this patch was withdrawn, not sure what that means: [4]
> 

Andrew confirmed he's fine with these MM changes going through the KVM tree.

> [4] https://lore.kernel.org/all/20250625000155.62D08C4CEE3@smtp.kernel.org/

Regarding [4]:
https://lore.kernel.org/linux-mm/aFlHIjLBwn3LQFMC@casper.infradead.org/
Re: [PATCH v12 00/12] KVM: guest_memfd: Add NUMA mempolicy support
Posted by Ackerley Tng 4 months ago
"Garg, Shivank" <shivankg@amd.com> writes:

> On 10/10/2025 2:28 AM, Ackerley Tng wrote:
>> For future reference, these are the three specific patches:
>> 
>> [1] https://lore.kernel.org/all/20250827175247.83322-4-shivankg@amd.com/
>> [2] https://lore.kernel.org/all/20250827175247.83322-5-shivankg@amd.com/
>> [3] https://lore.kernel.org/all/20250827175247.83322-6-shivankg@amd.com/
>> 
>> Might have missed this, did we discuss how these 3 would get merged? I
>> noticed this patch was withdrawn, not sure what that means: [4]
>> 
>
> Andrew confirmed he's fine with these MM changes going through the KVM tree.
>
>> [4] https://lore.kernel.org/all/20250625000155.62D08C4CEE3@smtp.kernel.org/
>

Thanks Shivank and Andrew!

> Regarding [4]:
> https://lore.kernel.org/linux-mm/aFlHIjLBwn3LQFMC@casper.infradead.org/