[PATCH] KVM: i8259: initialize IMR to 0xff on reset

[PATCH] KVM: i8259: initialize IMR to 0xff on reset

[Avi Kivity]

Otherwise, a wily interrupt can slip through while the guest isn't prepared
for it (and while the irq base is zero).

Signed-off-by: Avi Kivity <avi@redhat.com>
---
 arch/x86/kvm/i8259.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
index 8d10c06..5de9ee0 100644
--- a/arch/x86/kvm/i8259.c
+++ b/arch/x86/kvm/i8259.c
@@ -275,7 +275,7 @@ void kvm_pic_reset(struct kvm_kpic_state *s)
 
 	s->last_irr = 0;
 	s->irr = 0;
-	s->imr = 0;
+	s->imr = 0xff;
 	s->isr = 0;
 	s->isr_ack = 0xff;
 	s->priority_add = 0;
-- 
1.7.1

Re: [PATCH] KVM: i8259: initialize IMR to 0xff on reset

[Avi Kivity]

  On 08/30/2010 12:20 PM, Avi Kivity wrote:
> Otherwise, a wily interrupt can slip through while the guest isn't prepared
> for it (and while the irq base is zero).
>
>
>   	s->last_irr = 0;
>   	s->irr = 0;
> -	s->imr = 0;
> +	s->imr = 0xff;
>   	s->isr = 0;
>   	s->isr_ack = 0xff;
>   	s->priority_add = 0;

For some reason, this bug only shows on the non-atomic-injection branch 
(specifically, the KVM_REQ_EVENT part).

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

Re: [PATCH] KVM: i8259: initialize IMR to 0xff on reset

[Paolo Bonzini]

On 08/30/2010 11:20 AM, Avi Kivity wrote:
> Otherwise, a wily interrupt can slip through while the guest isn't prepared
> for it (and while the irq base is zero).
>
> Signed-off-by: Avi Kivity<avi@redhat.com>
> ---
>   arch/x86/kvm/i8259.c |    2 +-
>   1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
> index 8d10c06..5de9ee0 100644
> --- a/arch/x86/kvm/i8259.c
> +++ b/arch/x86/kvm/i8259.c
> @@ -275,7 +275,7 @@ void kvm_pic_reset(struct kvm_kpic_state *s)
>
>   	s->last_irr = 0;
>   	s->irr = 0;
> -	s->imr = 0;
> +	s->imr = 0xff;
>   	s->isr = 0;
>   	s->isr_ack = 0xff;
>   	s->priority_add = 0;

Sounds sane, but the datasheet says explicitly that upon reset "The 
Interrupt Mask Register is cleared"...  (FWIW, I checked because it 
looked like QEMU and Xen also had the same behavior of setting IMR to zero).

Paolo

Re: [PATCH] KVM: i8259: initialize IMR to 0xff on reset

[Avi Kivity]

  On 08/30/2010 04:36 PM, Paolo Bonzini wrote:
> On 08/30/2010 11:20 AM, Avi Kivity wrote:
>> Otherwise, a wily interrupt can slip through while the guest isn't 
>> prepared
>> for it (and while the irq base is zero).
>>
>> Signed-off-by: Avi Kivity<avi@redhat.com>
>> ---
>>   arch/x86/kvm/i8259.c |    2 +-
>>   1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
>> index 8d10c06..5de9ee0 100644
>> --- a/arch/x86/kvm/i8259.c
>> +++ b/arch/x86/kvm/i8259.c
>> @@ -275,7 +275,7 @@ void kvm_pic_reset(struct kvm_kpic_state *s)
>>
>>       s->last_irr = 0;
>>       s->irr = 0;
>> -    s->imr = 0;
>> +    s->imr = 0xff;
>>       s->isr = 0;
>>       s->isr_ack = 0xff;
>>       s->priority_add = 0;
>
> Sounds sane, but the datasheet says explicitly that upon reset "The 
> Interrupt Mask Register is cleared"...  (FWIW, I checked because it 
> looked like QEMU and Xen also had the same behavior of setting IMR to 
> zero).

Hm, right.

Reading the spec, the correct approach is not to do a full reset on 
ICW1(D4=1), instead just do what the spec says (which is different from 
reset).  I'll try it out.

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