On Wed, Aug 18, 2021 at 09:29:40PM +0000, Oliver Upton wrote:
> An error of ENOENT for the KVM_ARM_VCPU_INIT ioctl indicates that one of
> the requested feature flags is not supported by the kernel/hardware.
> Detect the case when KVM doesn't support the requested features and skip
> the test rather than failing it.
>
> Cc: Andrew Jones <drjones@xxxxxxxxxx>
> Signed-off-by: Oliver Upton <oupton@xxxxxxxxxx>
> ---
> Applies to 5.14-rc6. Tested by running all selftests on an Ampere Mt.
> Jade system.
>
> .../testing/selftests/kvm/lib/aarch64/processor.c | 15 ++++++++++++++-
> 1 file changed, 14 insertions(+), 1 deletion(-)
>
> diff --git a/tools/testing/selftests/kvm/lib/aarch64/processor.c b/tools/testing/selftests/kvm/lib/aarch64/processor.c
> index 632b74d6b3ca..b1064a0c5e62 100644
> --- a/tools/testing/selftests/kvm/lib/aarch64/processor.c
> +++ b/tools/testing/selftests/kvm/lib/aarch64/processor.c
> @@ -216,6 +216,7 @@ void aarch64_vcpu_setup(struct kvm_vm *vm, int vcpuid, struct kvm_vcpu_init *ini
> {
> struct kvm_vcpu_init default_init = { .target = -1, };
> uint64_t sctlr_el1, tcr_el1;
> + int ret;
>
> if (!init)
> init = &default_init;
> @@ -226,7 +227,19 @@ void aarch64_vcpu_setup(struct kvm_vm *vm, int vcpuid, struct kvm_vcpu_init *ini
> init->target = preferred.target;
> }
>
> - vcpu_ioctl(vm, vcpuid, KVM_ARM_VCPU_INIT, init);
> + ret = _vcpu_ioctl(vm, vcpuid, KVM_ARM_VCPU_INIT, init);
> +
> + /*
> + * Missing kernel feature support should result in skipping the test,
> + * not failing it.
> + */
> + if (ret && errno == ENOENT) {
> + print_skip("requested vCPU features not supported; skipping test.");
", skipping test" will already be appended by print_skip().
> + exit(KSFT_SKIP);
> + }
> +
> + TEST_ASSERT(!ret, "KVM_ARM_VCPU_INIT failed, rc: %i errno: %i (%s)",
> + ret, errno, strerror(errno));
>
> /*
> * Enable FP/ASIMD to avoid trapping when accessing Q0-Q15
> --
> 2.33.0.rc1.237.g0d66db33f3-goog
>
I think I'd rather try to keep exit(KSFT_SKIP)'s out of the lib code. It'd
be better if the test gets to decide whether to skip or not. How about
moving this check+skip into the test instead?
Thanks,
drew
