Re: [PATCH v6 2/3] KVM: selftests: Add a test to get/set triple fault event

From: Chenyi Qiang <chenyi.qiang@intel.com>
To: Sean Christopherson <seanjc@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Wanpeng Li <wanpengli@tencent.com>,
	Jim Mattson <jmattson@google.com>, Joerg Roedel <joro@8bytes.org>,
	Xiaoyao Li <xiaoyao.li@intel.com>,
	kvm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v6 2/3] KVM: selftests: Add a test to get/set triple fault event
Date: Mon, 23 May 2022 14:46:11 +0800	[thread overview]
Message-ID: <5c5a7597-d6e3-7a05-ead8-659c45aea222@intel.com> (raw)
In-Reply-To: <YoVHAIGcFgJit1qp@google.com>

On 5/19/2022 3:20 AM, Sean Christopherson wrote:
> On Thu, Apr 21, 2022, Chenyi Qiang wrote:
>> +#ifndef __x86_64__
>> +# error This test is 64-bit only
> 
> No need, all of KVM selftests are 64-bit only.
> 
>> +#endif
>> +
>> +#define VCPU_ID			0
>> +#define ARBITRARY_IO_PORT	0x2000
>> +
>> +/* The virtual machine object. */
>> +static struct kvm_vm *vm;
>> +
>> +static void l2_guest_code(void)
>> +{
>> +	/*
>> +	 * Generate an exit to L0 userspace, i.e. main(), via I/O to an
>> +	 * arbitrary port.
>> +	 */
> 
> I think we can test a "real" triple fault without too much effort by abusing
> vcpu->run->request_interrupt_window.  E.g. map the run struct into L2, clear
> PIN_BASED_EXT_INTR_MASK in vmcs12, and then in l2_guest_code() do:
> 
> 	asm volatile("cli");
> 
> 	run->request_interrupt_window = true;
> 

Maybe, A GUEST_SYNC to main() to change the request_interrupt_window 
also works.

> 	asm volatile("sti; ud2");
> 

How does the "real" triple fault occur here? Through UD2?

> 	asm volatile("inb %%dx, %%al"
> 		     : : [port] "d" (ARBITRARY_IO_PORT) : "rax"); 	
> 
> The STI shadow will block the IRQ-window VM-Exit until after the ud2, i.e. until
> after the triple fault is pending.
> 
> And then main() can
> 
>    1. verify it got KVM_EXIT_IRQ_WINDOW_OPEN with a pending triple fault
>    2. clear the triple fault and re-enter L1+l2
>    3. continue with the existing code, e.g. verify it got KVM_EXIT_IO, stuff a
>       pending triple fault, etc...
> 
> That said, typing that out makes me think we should technically handle/prevent this
> in KVM since triple fault / shutdown is kinda sorta just a special exception.
> 
> Heh, and a potentially bad/crazy idea would be to use a reserved/magic vector in
> kvm_vcpu_events.exception to save/restore triple fault, e.g. we could steal
> NMI_VECTOR.
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index f2d0ee9296b9..d58c0cfd3cd3 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -4743,7 +4743,8 @@ static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
>          return (kvm_arch_interrupt_allowed(vcpu) &&
>                  kvm_cpu_accept_dm_intr(vcpu) &&
>                  !kvm_event_needs_reinjection(vcpu) &&
> -               !vcpu->arch.exception.pending);
> +               !vcpu->arch.exception.pending &&
> +               !kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu));
>   }
> 
>   static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
> 
> 
> 