[PATCH 0/3] KVM: arm64: selftests: Basic nested guest support

Wei-Lin Chang posted 3 patches 1 week, 2 days ago
tools/testing/selftests/kvm/Makefile.kvm      |   3 +
.../selftests/kvm/arm64/hello_nested.c        |  65 ++++++++
.../selftests/kvm/include/arm64/nested.h      |  25 +++
.../selftests/kvm/include/arm64/processor.h   |   9 +
.../testing/selftests/kvm/lib/arm64/nested.c  | 154 ++++++++++++++++++
.../selftests/kvm/lib/arm64/nested_asm.S      |  35 ++++
6 files changed, 291 insertions(+)
create mode 100644 tools/testing/selftests/kvm/arm64/hello_nested.c
create mode 100644 tools/testing/selftests/kvm/include/arm64/nested.h
create mode 100644 tools/testing/selftests/kvm/lib/arm64/nested.c
create mode 100644 tools/testing/selftests/kvm/lib/arm64/nested_asm.S
[PATCH 0/3] KVM: arm64: selftests: Basic nested guest support
Posted by Wei-Lin Chang 1 week, 2 days ago
Hi,

This series adds basic support for running nested guests (L2) in
kselftest. The first patch adds library functions. While designing the
APIs for userspace, I referenced Joey's approach for kvm-unit-tests [1].
In summary, four preparatory functions are provided for userspace to
set up state to run an L2 in EL1:

 - prepare_l2_stack()            <- sets up stack for L2
 - prepare_hyp_state()           <- sets up vEL2 registers
 - prepare_eret_destination()    <- userspace passes a function pointer
                                    for L2 to run
 - prepare_nested_sync_handler() <- sets up hvc handler in order to
                                    regain control after L2's hvc

After calling those functions, userspace can vcpu_run(), and when
run_l2() is called within the guest, the supplied function will be run
in L2, with the control flow managed by the library code in nested.c and
nested_asm.S. After running the L2 function, run_l2() will automatically
return. Note that the L2 function supplied by the user does not have to
call hvc.

Patch 2 demonstrates usage of the APIs introduced above, with a simple
L1 -> L2 -> L1 sequence, with an empty L2 function.

Patch 3 enhances the library functions by setting up L2 -> L1 stage-2
translation. Currently the translation is simple, with start level 0, 4
levels, 4KB granules, normal cachable, 48-bit IA, 40-bit OA.

[1]: https://lore.kernel.org/kvmarm/20260306142656.2775185-1-joey.gouly@arm.com/

Wei-Lin Chang (3):
  KVM: arm64: selftests: Add library functions for NV
  KVM: arm64: sefltests: Add basic NV selftest
  KVM: arm64: selftests: Enable stage-2 in NV preparation functions

 tools/testing/selftests/kvm/Makefile.kvm      |   3 +
 .../selftests/kvm/arm64/hello_nested.c        |  65 ++++++++
 .../selftests/kvm/include/arm64/nested.h      |  25 +++
 .../selftests/kvm/include/arm64/processor.h   |   9 +
 .../testing/selftests/kvm/lib/arm64/nested.c  | 154 ++++++++++++++++++
 .../selftests/kvm/lib/arm64/nested_asm.S      |  35 ++++
 6 files changed, 291 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/arm64/hello_nested.c
 create mode 100644 tools/testing/selftests/kvm/include/arm64/nested.h
 create mode 100644 tools/testing/selftests/kvm/lib/arm64/nested.c
 create mode 100644 tools/testing/selftests/kvm/lib/arm64/nested_asm.S

-- 
2.43.0
Re: [PATCH 0/3] KVM: arm64: selftests: Basic nested guest support
Posted by Itaru Kitayama 1 week, 1 day ago
Hi Wei Lin,

On Wed, Mar 25, 2026 at 12:36:17AM +0000, Wei-Lin Chang wrote:
> Hi,
> 
> This series adds basic support for running nested guests (L2) in
> kselftest. The first patch adds library functions. While designing the
> APIs for userspace, I referenced Joey's approach for kvm-unit-tests [1].
> In summary, four preparatory functions are provided for userspace to
> set up state to run an L2 in EL1:
> 
>  - prepare_l2_stack()            <- sets up stack for L2
>  - prepare_hyp_state()           <- sets up vEL2 registers
>  - prepare_eret_destination()    <- userspace passes a function pointer
>                                     for L2 to run
>  - prepare_nested_sync_handler() <- sets up hvc handler in order to
>                                     regain control after L2's hvc
> 
> After calling those functions, userspace can vcpu_run(), and when
> run_l2() is called within the guest, the supplied function will be run
> in L2, with the control flow managed by the library code in nested.c and
> nested_asm.S. After running the L2 function, run_l2() will automatically
> return. Note that the L2 function supplied by the user does not have to
> call hvc.
> 
> Patch 2 demonstrates usage of the APIs introduced above, with a simple
> L1 -> L2 -> L1 sequence, with an empty L2 function.
> 
> Patch 3 enhances the library functions by setting up L2 -> L1 stage-2
> translation. Currently the translation is simple, with start level 0, 4
> levels, 4KB granules, normal cachable, 48-bit IA, 40-bit OA.
> 
> [1]: https://lore.kernel.org/kvmarm/20260306142656.2775185-1-joey.gouly@arm.com/

Look like this selftest assumes nested guest's MMU is disabled (L2 IPA
to L1 IPA to PA), but I couldn's find the explicit SCTLR.M bit operation in this
series, how do you make sure it is always off?

Thanks,
Itaru.

> 
> Wei-Lin Chang (3):
>   KVM: arm64: selftests: Add library functions for NV
>   KVM: arm64: sefltests: Add basic NV selftest
>   KVM: arm64: selftests: Enable stage-2 in NV preparation functions
> 
>  tools/testing/selftests/kvm/Makefile.kvm      |   3 +
>  .../selftests/kvm/arm64/hello_nested.c        |  65 ++++++++
>  .../selftests/kvm/include/arm64/nested.h      |  25 +++
>  .../selftests/kvm/include/arm64/processor.h   |   9 +
>  .../testing/selftests/kvm/lib/arm64/nested.c  | 154 ++++++++++++++++++
>  .../selftests/kvm/lib/arm64/nested_asm.S      |  35 ++++
>  6 files changed, 291 insertions(+)
>  create mode 100644 tools/testing/selftests/kvm/arm64/hello_nested.c
>  create mode 100644 tools/testing/selftests/kvm/include/arm64/nested.h
>  create mode 100644 tools/testing/selftests/kvm/lib/arm64/nested.c
>  create mode 100644 tools/testing/selftests/kvm/lib/arm64/nested_asm.S
> 
> -- 
> 2.43.0
>