[PATCH] Clear temporary interrupt blocking on halt in real mode

[PATCH] Clear temporary interrupt blocking on halt in real mode

[Laurent Vivier]

When "halt" is emulated, skip_emulated_instruction() is called and interruptibility state is cleared.
But when halt is emulated in real mode, skip_emulated_instruction() is not called and the interruptiblity state is not cleared.

The following code, from gPXE, never exits from loop because interrupts are not delivered to increase %fs:(0x6c):

        movl    %fs:(0x6c), %eax
1:      pushf
        sti
        hlt
        popf
        cmpl    %fs:(0x6c), %eax
        je      1b

This patch clears the interruptibility state when halt is emulated in real mode.

Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
---
 arch/x86/kvm/vmx.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index c4510fe..82c4324 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -2329,7 +2329,17 @@ static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 	    handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
 								error_code)) {
 		if (vcpu->arch.halt_request) {
+			u32 interruptibility;
 			vcpu->arch.halt_request = 0;
+			/*
+			 * We emulated an instruction, so temporary interrupt blocking
+			 * should be removed, if set.
+                         */
+			interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
+			if (interruptibility & 3)
+				vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
+					     interruptibility & ~3);
+			vcpu->arch.interrupt_window_open = 1;
 			return kvm_emulate_halt(vcpu);
 		}
 		return 1;
-- 
1.5.2.4

Re: [PATCH] Clear temporary interrupt blocking on halt in real mode

[Avi Kivity]

Laurent Vivier wrote:
> When "halt" is emulated, skip_emulated_instruction() is called and interruptibility state is cleared.
> But when halt is emulated in real mode, skip_emulated_instruction() is not called and the interruptiblity state is not cleared.
>
> The following code, from gPXE, never exits from loop because interrupts are not delivered to increase %fs:(0x6c):
>
>         movl    %fs:(0x6c), %eax
> 1:      pushf
>         sti
>         hlt
>         popf
>         cmpl    %fs:(0x6c), %eax
>         je      1b
>
> This patch clears the interruptibility state when halt is emulated in real mode.
>
> Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
> ---
>  arch/x86/kvm/vmx.c |   10 ++++++++++
>  1 files changed, 10 insertions(+), 0 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index c4510fe..82c4324 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2329,7 +2329,17 @@ static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
>  	    handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
>  								error_code)) {
>  		if (vcpu->arch.halt_request) {
> +			u32 interruptibility;
>  			vcpu->arch.halt_request = 0;
> +			/*
> +			 * We emulated an instruction, so temporary interrupt blocking
> +			 * should be removed, if set.
> +                         */
> +			interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
> +			if (interruptibility & 3)
> +				vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
> +					     interruptibility & ~3);
> +			vcpu->arch.interrupt_window_open = 1;
>  			return kvm_emulate_halt(vcpu);
>  		}
>  		return 1;
>   


Shouldn't interruptibility state be cleared if *any* instruction is 
emulated?

(of course, an emulated instruction may update interruptibility state 
itself, say sti).

-- 
error compiling committee.c: too many arguments to function

Re: [PATCH] Clear temporary interrupt blocking on halt in real mode

[Laurent Vivier]

Le dimanche 10 août 2008 à 11:36 +0300, Avi Kivity a écrit :
> Laurent Vivier wrote:
> > When "halt" is emulated, skip_emulated_instruction() is called and interruptibility state is cleared.
> > But when halt is emulated in real mode, skip_emulated_instruction() is not called and the interruptiblity state is not cleared.
> >
> > The following code, from gPXE, never exits from loop because interrupts are not delivered to increase %fs:(0x6c):
> >
> >         movl    %fs:(0x6c), %eax
> > 1:      pushf
> >         sti
> >         hlt
> >         popf
> >         cmpl    %fs:(0x6c), %eax
> >         je      1b
> >
> > This patch clears the interruptibility state when halt is emulated in real mode.
> >
> > Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
> > ---
> >  arch/x86/kvm/vmx.c |   10 ++++++++++
> >  1 files changed, 10 insertions(+), 0 deletions(-)
> >
> > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> > index c4510fe..82c4324 100644
> > --- a/arch/x86/kvm/vmx.c
> > +++ b/arch/x86/kvm/vmx.c
> > @@ -2329,7 +2329,17 @@ static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
> >  	    handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
> >  								error_code)) {
> >  		if (vcpu->arch.halt_request) {
> > +			u32 interruptibility;
> >  			vcpu->arch.halt_request = 0;
> > +			/*
> > +			 * We emulated an instruction, so temporary interrupt blocking
> > +			 * should be removed, if set.
> > +                         */
> > +			interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
> > +			if (interruptibility & 3)
> > +				vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
> > +					     interruptibility & ~3);
> > +			vcpu->arch.interrupt_window_open = 1;
> >  			return kvm_emulate_halt(vcpu);
> >  		}
> >  		return 1;
> >   
> 
> 
> Shouldn't interruptibility state be cleared if *any* instruction is 
> emulated?
> 
> (of course, an emulated instruction may update interruptibility state 
> itself, say sti).

I don't know what to do with that...

I think emulated instructions already clear interruptibility state via
skip_emulated_instruction().

The case of kvm_emulate_halt() and real mode is a special case.

Laurent
-- 
----------------- Laurent.Vivier@bull.net  ------------------
  "La perfection est atteinte non quand il ne reste rien à
ajouter mais quand il ne reste rien à enlever." Saint Exupéry