Re: [RFC PATCH v5 16/29] KVM: selftests: TDX: Add TDX HLT exit test

[Yan Zhao]

On Sat, Mar 02, 2024 at 03:31:07PM +0800, Binbin Wu wrote:
> On 12/13/2023 4:46 AM, Sagi Shahar wrote:
> > The test verifies that the guest runs TDVMCALL<INSTRUCTION.HLT> and the
> > guest vCPU enters to the halted state.
> > 
> > Signed-off-by: Erdem Aktas <erdemaktas@xxxxxxxxxx>
> > Signed-off-by: Sagi Shahar <sagis@xxxxxxxxxx>
> > Signed-off-by: Ackerley Tng <ackerleytng@xxxxxxxxxx>
> > Signed-off-by: Ryan Afranji <afranji@xxxxxxxxxx>
> > ---
> >   .../selftests/kvm/include/x86_64/tdx/tdx.h    |  2 +
> >   .../selftests/kvm/lib/x86_64/tdx/tdx.c        | 10 +++
> >   .../selftests/kvm/x86_64/tdx_vm_tests.c       | 78 +++++++++++++++++++
> >   3 files changed, 90 insertions(+)
> > 
> > diff --git a/tools/testing/selftests/kvm/include/x86_64/tdx/tdx.h b/tools/testing/selftests/kvm/include/x86_64/tdx/tdx.h
> > index 85ba6aab79a7..b18e39d20498 100644
> > --- a/tools/testing/selftests/kvm/include/x86_64/tdx/tdx.h
> > +++ b/tools/testing/selftests/kvm/include/x86_64/tdx/tdx.h
> > @@ -8,6 +8,7 @@
> >   #define TDG_VP_VMCALL_GET_TD_VM_CALL_INFO 0x10000
> >   #define TDG_VP_VMCALL_REPORT_FATAL_ERROR 0x10003
> > +#define TDG_VP_VMCALL_INSTRUCTION_HLT 12
> >   #define TDG_VP_VMCALL_INSTRUCTION_IO 30
> >   #define TDG_VP_VMCALL_INSTRUCTION_RDMSR 31
> >   #define TDG_VP_VMCALL_INSTRUCTION_WRMSR 32
> > @@ -20,5 +21,6 @@ uint64_t tdg_vp_vmcall_get_td_vmcall_info(uint64_t *r11, uint64_t *r12,
> >   					uint64_t *r13, uint64_t *r14);
> >   uint64_t tdg_vp_vmcall_instruction_rdmsr(uint64_t index, uint64_t *ret_value);
> >   uint64_t tdg_vp_vmcall_instruction_wrmsr(uint64_t index, uint64_t value);
> > +uint64_t tdg_vp_vmcall_instruction_hlt(uint64_t interrupt_blocked_flag);
> >   #endif // SELFTEST_TDX_TDX_H
> > diff --git a/tools/testing/selftests/kvm/lib/x86_64/tdx/tdx.c b/tools/testing/selftests/kvm/lib/x86_64/tdx/tdx.c
> > index 88ea6f2a6469..9485bafedc38 100644
> > --- a/tools/testing/selftests/kvm/lib/x86_64/tdx/tdx.c
> > +++ b/tools/testing/selftests/kvm/lib/x86_64/tdx/tdx.c
> > @@ -114,3 +114,13 @@ uint64_t tdg_vp_vmcall_instruction_wrmsr(uint64_t index, uint64_t value)
> >   	return __tdx_hypercall(&args, 0);
> >   }
> > +
> > +uint64_t tdg_vp_vmcall_instruction_hlt(uint64_t interrupt_blocked_flag)
> > +{
> > +	struct tdx_hypercall_args args = {
> > +		.r11 = TDG_VP_VMCALL_INSTRUCTION_HLT,
> > +		.r12 = interrupt_blocked_flag,
> > +	};
> > +
> > +	return __tdx_hypercall(&args, 0);
> > +}
> > diff --git a/tools/testing/selftests/kvm/x86_64/tdx_vm_tests.c b/tools/testing/selftests/kvm/x86_64/tdx_vm_tests.c
> > index 5db3701cc6d9..5fae4c6e5f95 100644
> > --- a/tools/testing/selftests/kvm/x86_64/tdx_vm_tests.c
> > +++ b/tools/testing/selftests/kvm/x86_64/tdx_vm_tests.c
> > @@ -721,6 +721,83 @@ void verify_guest_msr_writes(void)
> >   	printf("\t ... PASSED\n");
> >   }
> > +/*
> > + * Verifies HLT functionality.
> > + */
> > +void guest_hlt(void)
> > +{
> > +	uint64_t ret;
> > +	uint64_t interrupt_blocked_flag;
> > +
> > +	interrupt_blocked_flag = 0;
> > +	ret = tdg_vp_vmcall_instruction_hlt(interrupt_blocked_flag);
> > +	if (ret)
> > +		tdx_test_fatal(ret);
> > +
> > +	tdx_test_success();
> > +}
> > +
> > +void _verify_guest_hlt(int signum);
> > +
> > +void wake_me(int interval)
> > +{
> > +	struct sigaction action;
> > +
> > +	action.sa_handler = _verify_guest_hlt;
> > +	sigemptyset(&action.sa_mask);
> > +	action.sa_flags = 0;
> > +
> > +	TEST_ASSERT(sigaction(SIGALRM, &action, NULL) == 0,
> > +		    "Could not set the alarm handler!");
> > +
> > +	alarm(interval);
> > +}
> > +
> > +void _verify_guest_hlt(int signum)
> > +{
> > +	struct kvm_vm *vm;
> > +	static struct kvm_vcpu *vcpu;
> > +
> > +	/*
> > +	 * This function will also be called by SIGALRM handler to check the
> > +	 * vCPU MP State. If vm has been initialized, then we are in the signal
> > +	 * handler. Check the MP state and let the guest run again.
> > +	 */
> > +	if (vcpu != NULL) {
> 
> What if the following case if there is a bug in KVM so that:
> 
> In guest, execution of tdg_vp_vmcall_instruction_hlt() return 0, but the
> vcpu is not actually halted. Then guest will call tdx_test_success().
> 
> And the first call of _verify_guest_hlt() will call kvm_vm_free(vm) to free
> the vm, which also frees the vcpu, and 1 second later, in this path vcpu
> will
> be accessed after free.
>
Right. Another possibility is that if buggy KVM returns success to guest
without putting guest to halted state, the selftest will still print
"PASSED" because the second _verify_guest_hlt() (after waiting for 1s)
has no chance to get executed before the process exits.

> > +		struct kvm_mp_state mp_state;
> > +
> > +		vcpu_mp_state_get(vcpu, &mp_state);
> > +		TEST_ASSERT_EQ(mp_state.mp_state, KVM_MP_STATE_HALTED);
> > +
> > +		/* Let the guest to run and finish the test.*/
> > +		mp_state.mp_state = KVM_MP_STATE_RUNNABLE;
> > +		vcpu_mp_state_set(vcpu, &mp_state);
> > +		return;
> > +	}
> > +
> > +	vm = td_create();
> > +	td_initialize(vm, VM_MEM_SRC_ANONYMOUS, 0);
> > +	vcpu = td_vcpu_add(vm, 0, guest_hlt);
> > +	td_finalize(vm);
> > +
> > +	printf("Verifying HLT:\n");
> > +
> > +	printf("\t ... Running guest\n");
> > +
> > +	/* Wait 1 second for guest to execute HLT */
> > +	wake_me(1);
> > +	td_vcpu_run(vcpu);
> > +
> > +	TDX_TEST_ASSERT_SUCCESS(vcpu);
> > +
> > +	kvm_vm_free(vm);
> > +	printf("\t ... PASSED\n");
> > +}
> > +
> > +void verify_guest_hlt(void)
> > +{
> > +	_verify_guest_hlt(0);
> > +}
> >   int main(int argc, char **argv)
> >   {
> > @@ -740,6 +817,7 @@ int main(int argc, char **argv)
> >   	run_in_new_process(&verify_guest_reads);
> >   	run_in_new_process(&verify_guest_msr_writes);
> >   	run_in_new_process(&verify_guest_msr_reads);
> > +	run_in_new_process(&verify_guest_hlt);
> >   	return 0;
> >   }
> 
>