Re: [PATCH 2/6] KVM: SVM: Emulate nRIP feature when reinjecting INT3

[Gleb Natapov <gleb@redhat.com>]

On Mon, Feb 22, 2010 at 06:51:19PM +0100, Jan Kiszka wrote:
> When in guest debugging mode, we have to reinject those #BP software
> exceptions that are caused by guest-injected INT3. As older AMD
> processors to not support the required nRIP VMCB field, try to emulate
> it by moving RIP by one on injection. Fix it up again in case the
> injection failed and we were able to catch this. This does not work for
> unintercepted faults, but it is better than doing nothing.
> 
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
>  arch/x86/kvm/svm.c |   72 +++++++++++++++++++++++++++++++++++++--------------
>  1 files changed, 52 insertions(+), 20 deletions(-)
> 
> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
> index 1d76899..754e2f7 100644
> --- a/arch/x86/kvm/svm.c
> +++ b/arch/x86/kvm/svm.c
> @@ -46,6 +46,7 @@ MODULE_LICENSE("GPL");
>  #define SVM_FEATURE_NPT  (1 << 0)
>  #define SVM_FEATURE_LBRV (1 << 1)
>  #define SVM_FEATURE_SVML (1 << 2)
> +#define SVM_FEATURE_NRIP (1 << 3)
>  #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
>  
>  #define NESTED_EXIT_HOST	0	/* Exit handled on host level */
> @@ -109,6 +110,10 @@ struct vcpu_svm {
>  	struct nested_state nested;
>  
>  	bool nmi_singlestep;
> +
> +	unsigned int3_injected;
> +	u16 int3_cs;
> +	u64 int3_rip;
>  };
>  
>  /* enable NPT for AMD64 and X86 with PAE */
> @@ -235,23 +240,6 @@ static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
>  	vcpu->arch.efer = efer;
>  }
>  
> -static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
> -				bool has_error_code, u32 error_code)
> -{
> -	struct vcpu_svm *svm = to_svm(vcpu);
> -
> -	/* If we are within a nested VM we'd better #VMEXIT and let the
> -	   guest handle the exception */
> -	if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
> -		return;
> -
> -	svm->vmcb->control.event_inj = nr
> -		| SVM_EVTINJ_VALID
> -		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
> -		| SVM_EVTINJ_TYPE_EXEPT;
> -	svm->vmcb->control.event_inj_err = error_code;
> -}
> -
Why have you moved svm_queue_exception() function?

>  static int is_external_interrupt(u32 info)
>  {
>  	info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
> @@ -297,6 +285,39 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
>  	svm_set_interrupt_shadow(vcpu, 0);
>  }
>  
> +static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
> +				bool has_error_code, u32 error_code)
> +{
> +	struct vcpu_svm *svm = to_svm(vcpu);
> +
> +	/* If we are within a nested VM we'd better #VMEXIT and let the
> +	   guest handle the exception */
> +	if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
> +		return;
> +
> +	if (nr == BP_VECTOR && !svm_has(SVM_FEATURE_NRIP)) {
> +		u64 old_rip = kvm_rip_read(&svm->vcpu);
> +
> +		/*
> +		 * For guest debugging where we have to reinject #BP if some
> +		 * INT3 is guest-owned:
> +		 * Emulate nRIP by moving RIP forward. Will fail if injection
> +		 * raises a fault that is not intercepted. Still better than
> +		 * failing in all cases.
> +		 */
> +		skip_emulated_instruction(&svm->vcpu);
> +		svm->int3_cs = svm->vmcb->save.cs.selector;
> +		svm->int3_rip = kvm_rip_read(&svm->vcpu);
> +		svm->int3_injected = svm->int3_rip - old_rip;
> +	}
> +
> +	svm->vmcb->control.event_inj = nr
> +		| SVM_EVTINJ_VALID
> +		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
> +		| SVM_EVTINJ_TYPE_EXEPT;
> +	svm->vmcb->control.event_inj_err = error_code;
> +}
> +
>  static int has_svm(void)
>  {
>  	const char *msg;
> @@ -2703,8 +2724,10 @@ static void svm_complete_interrupts(struct vcpu_svm *svm)
>  	kvm_clear_exception_queue(&svm->vcpu);
>  	kvm_clear_interrupt_queue(&svm->vcpu);
> 
int int3_injected = svm->int3_injected;
svm->int3_injected = 0;

Will save us from zeroing svm->int3_injected in two places.

> -	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
> +	if (!(exitintinfo & SVM_EXITINTINFO_VALID)) {
> +		svm->int3_injected = 0;
>  		return;
> +	}
>  
>  	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
>  	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
> @@ -2714,12 +2737,20 @@ static void svm_complete_interrupts(struct vcpu_svm *svm)
>  		svm->vcpu.arch.nmi_injected = true;
>  		break;
>  	case SVM_EXITINTINFO_TYPE_EXEPT:
> -		/* In case of software exception do not reinject an exception
> -		   vector, but re-execute and instruction instead */
>  		if (is_nested(svm))
>  			break;
>  		if (kvm_exception_is_soft(vector))
> +			if (vector == BP_VECTOR && svm->int3_injected &&
> +			    svm->vmcb->save.cs.selector == svm->int3_cs &&
> +			    kvm_rip_read(&svm->vcpu) == svm->int3_rip)
> +				kvm_rip_write(&svm->vcpu,
> +					      kvm_rip_read(&svm->vcpu) -
> +					      svm->int3_injected);
>  			break;
Something is very wrong here. break does not belong to any of above ifs,
so code below is unreachable.

> +		/*
> +		 * In case of other software exceptions, do not reinject the
> +		 * vector, but re-execute the instruction instead.
> +		 */
>  		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
>  			u32 err = svm->vmcb->control.exit_int_info_err;
>  			kvm_queue_exception_e(&svm->vcpu, vector, err);
> @@ -2733,6 +2764,7 @@ static void svm_complete_interrupts(struct vcpu_svm *svm)
>  	default:
>  		break;
>  	}
> +	svm->int3_injected = 0;
>  }
>  
>  #ifdef CONFIG_X86_64
> -- 
> 1.6.0.2

--
			Gleb.