Re: [PATCH v3 3/6] KVM: SVM: implement force_intercept_exceptions_mask

[Maxim Levitsky <mlevitsk@redhat.com>]

On Thu, 2021-09-02 at 17:34 +0000, Sean Christopherson wrote:
> On Wed, Aug 11, 2021, Maxim Levitsky wrote:
> > On Wed, 2021-08-11 at 15:29 +0300, Maxim Levitsky wrote:
> > > diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> > > index e45259177009..19f54b07161a 100644
> > > --- a/arch/x86/kvm/svm/svm.c
> > > +++ b/arch/x86/kvm/svm/svm.c
> > > @@ -233,6 +233,8 @@ static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
> > >  #define MSRS_RANGE_SIZE 2048
> > >  #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
> > >  
> > > +static int svm_handle_invalid_exit(struct kvm_vcpu *vcpu, u64 exit_code);
> > > +
> > >  u32 svm_msrpm_offset(u32 msr)
> > >  {
> > >  	u32 offset;
> > > @@ -1153,6 +1155,22 @@ static void svm_recalc_instruction_intercepts(struct kvm_vcpu *vcpu,
> > >  	}
> > >  }
> > >  
> > > +static void svm_init_force_exceptions_intercepts(struct vcpu_svm *svm)
> > > +{
> > > +	int exc;
> > > +
> > > +	svm->force_intercept_exceptions_mask = force_intercept_exceptions_mask;
> 
> Ah, the param is being snapshotted on vCPU creation, hence the writable module
> param.  That works, though it'd be better to snapshot it on a per-VM basic, not
> per-vCPU, and do so in common x86 code so that the param doesn't need to be
> exported.

I have nothing against that.

> 
> > > +	for (exc = 0 ; exc < 32 ; exc++) {
> 
> for_each_set_bit()
I used a helper function instead, IMHO a bit cleaner.

> 
> > > +		if (!(svm->force_intercept_exceptions_mask & (1 << exc)))
> > > +			continue;
> > > +
> > > +		/* Those are defined to have undefined behavior in the SVM spec */
> > > +		if (exc != 2 && exc != 9)
> 
> Maybe add a pr_warn_once() to let the user know they done messed up?
> 
> And given that there are already intercepts with undefined behavior, it's probably
> best to disallow intercepting anything we aren't 100% postive will be handled
> correctly, e.g. intercepting vector 31 is nonsensical at this time.

Or I think I'll just drop this check altogether - this is a debug feature anyway.

> 
> > > +			continue;
> > > +		set_exception_intercept(svm, exc);
> 
> ...
> 
> > > +static int gen_exc_interception(struct kvm_vcpu *vcpu)
> > > +{
> > > +	/*
> > > +	 * Generic exception intercept handler which forwards a guest exception
> > > +	 * as-is to the guest.
> > > +	 * For exceptions that don't have a special intercept handler.
> > > +	 *
> > > +	 * Used only for 'force_intercept_exceptions_mask' KVM debug feature.
> > > +	 */
> > > +	struct vcpu_svm *svm = to_svm(vcpu);
> > > +	int exc = svm->vmcb->control.exit_code - SVM_EXIT_EXCP_BASE;
> > > +
> > > +	/* SVM doesn't provide us with an error code for the #DF */
> > > +	u32 err_code = exc == DF_VECTOR ? 0 : svm->vmcb->control.exit_info_1;
> 
> Might be better to handle this in the x86_exception_has_error_code() path to
> avoid confusing readers with respect to exceptions that don't have an error code,
> e.g.
> 
> 	else if (x86_exception_has_error_code(exc)) {
> 		/* SVM doesn't provide the error code on #DF :-( */
> 		if (exc == DF_VECTOR)
> 			kvm_queue_exception_e(vcpu, exc, 0);
> 		else
> 			kvm_queue_exception_e(vcpu, exc, svm->vmcb->control.exit_info_1);
> 	} else {
> 		...
> 	}
> 
> Alternatively, can we zero svm->vmcb->control.exit_info_1 on #DF to make it more
> obvious that SVM leaves stale data in exit_info_1 (assuming that's true)?  E.g.
> 
> 	...
> 
> 	if (exc == TS_VECTOR) {
> 		...
> 	} else if (x86_exception_has_error_code(exc)) {
> 		/* SVM doesn't provide the error code on #DF :-( */
> 		if (exc == DF_VECTOR)
> 			svm->vmcb->control.exit_info_1 = 0;
> 
> 		kvm_queue_exception_e(vcpu, exc, svm->vmcb->control.exit_info_1);
> 	} else {
> 		...
> 	}

Makes sense.

> 
> 		
> > > +
> > > +	if (!(svm->force_intercept_exceptions_mask & (1 << exc)))
> 
> BIT(exc)
I added a helper function in common x86 code for this.

> 
> > > +		return svm_handle_invalid_exit(vcpu, svm->vmcb->control.exit_code);
> > > +
> > > +	if (exc == TS_VECTOR) {
> > > +		/*
> > > +		 * SVM doesn't provide us with an error code to be able to
> > > +		 * re-inject the #TS exception, so just disable its
> > > +		 * intercept, and let the guest re-execute the instruction.
> > > +		 */
> > > +		vmcb_clr_intercept(&svm->vmcb01.ptr->control,
> > > +				   INTERCEPT_EXCEPTION_OFFSET + TS_VECTOR);
> 
> Maybe just disallow intercepting #TS altogether?  Or does this fall into your
> Win98 use case? :-)

Win98 does indeed generate few #TS exceptions but so far I haven't noticed
any issues related to task switches. Anyway I would like to intercept
as much as possible since this is a debug feature. A single interception
is still better that nothing.


> 
> > > +		recalc_intercepts(svm);
> > > +	} else if (x86_exception_has_error_code(exc))
> > > +		kvm_queue_exception_e(vcpu, exc, err_code);
> > > +	else
> > > +		kvm_queue_exception(vcpu, exc);
> > > +	return 1;
> > > +}
> > > +
> > >  static bool is_erratum_383(void)
> > >  {
> > >  	int err, i;
> > > @@ -3065,6 +3131,10 @@ static int (*const svm_exit_handlers[])(struct kvm_vcpu *vcpu) = {
> > >  	[SVM_EXIT_WRITE_DR5]			= dr_interception,
> > >  	[SVM_EXIT_WRITE_DR6]			= dr_interception,
> > >  	[SVM_EXIT_WRITE_DR7]			= dr_interception,
> > > +
> > > +	[SVM_EXIT_EXCP_BASE ...
> > > +	SVM_EXIT_EXCP_BASE + 31]		= gen_exc_interception,
> 
> This generates a Sparse warning due to the duplicate initializer.  IMO that's a
> very good warning as I have zero idea how the compiler actually handles this
> particular scenario, e.g. do later entries take priority, is it technically
> "undefined" behavior, etc...
> 
> arch/x86/kvm/svm/svm.c:3065:10: warning: Initializer entry defined twice
> arch/x86/kvm/svm/svm.c:3067:29:   also defined here
> 
> I don't have a clever solution though :-('

Good catch. I thought that this would make sense but standards never make sense.
I'll do this manually.

> 
> > > +
> > >  	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
> > >  	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
> > >  	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
> > > diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
> > > index 524d943f3efc..187ada7c5b03 100644
> > > --- a/arch/x86/kvm/svm/svm.h
> > > +++ b/arch/x86/kvm/svm/svm.h
> > > @@ -196,6 +196,7 @@ struct vcpu_svm {
> > >  	bool ghcb_sa_free;
> > >  
> > >  	bool guest_state_loaded;
> > > +	u32 force_intercept_exceptions_mask;
> > >  };
> > >  
> > >  struct svm_cpu_data {
> > > @@ -351,8 +352,11 @@ static inline void clr_exception_intercept(struct vcpu_svm *svm, u32 bit)
> > >  	struct vmcb *vmcb = svm->vmcb01.ptr;
> > >  
> > >  	WARN_ON_ONCE(bit >= 32);
> > > -	vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
> > >  
> > > +	if ((1 << bit) & svm->force_intercept_exceptions_mask)
> 
> BIT(bit)
Fixed with helper function as well.

> 
> > > +		return;
> > > +
> > > +	vmcb_clr_intercept(&vmcb->control, INTERCEPT_EXCEPTION_OFFSET + bit);
> > >  	recalc_intercepts(svm);
> > >  }


Thanks for the review!
Best regards,
	Maxim Levitsky