[PATCH] KVM: ioapic: add check for RTC_GSI

[PATCH] KVM: ioapic: add check for RTC_GSI

[Chris J Arges]

cppcheck found the following error:
[ioapic.c:114]: (error) Array index -1 is out of bounds.

If CONFIG_X86 is not defined, RTC_GSI == -1U which means that an out of bounds
error could occur when accessing &ioapic->redirtbl[RTC_GSI].

This patch adds a check to kvm_rtc_eoi_tracking_restore_one that is similar to
how kvm_rtc_eoi_tracking_restore_all checks this condition.

Signed-off-by: Chris J Arges <chris.j.arges@canonical.com>
---
 virt/kvm/ioapic.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/virt/kvm/ioapic.c b/virt/kvm/ioapic.c
index e8ce34c..859f5b8 100644
--- a/virt/kvm/ioapic.c
+++ b/virt/kvm/ioapic.c
@@ -136,6 +136,9 @@ void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
 {
 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
 
+	if (RTC_GSI >= IOAPIC_NUM_PINS)
+		return;
+
 	spin_lock(&ioapic->lock);
 	__rtc_irq_eoi_tracking_restore_one(vcpu);
 	spin_unlock(&ioapic->lock);
-- 
1.9.1

Re: [PATCH] KVM: ioapic: add check for RTC_GSI

[Jan Kiszka]

On 2014-09-10 18:53, Chris J Arges wrote:
> cppcheck found the following error:
> [ioapic.c:114]: (error) Array index -1 is out of bounds.
> 
> If CONFIG_X86 is not defined, RTC_GSI == -1U which means that an out of bounds

That issue is limited to the dead ia64 architecture, no other builds
ioapic.c (besides x86, of course). Maybe finally remove KVM support for
that arch instead of bloating otherwise correct code?

Jan

> error could occur when accessing &ioapic->redirtbl[RTC_GSI].
> 
> This patch adds a check to kvm_rtc_eoi_tracking_restore_one that is similar to
> how kvm_rtc_eoi_tracking_restore_all checks this condition.
> 
> Signed-off-by: Chris J Arges <chris.j.arges@canonical.com>
> ---
>  virt/kvm/ioapic.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/virt/kvm/ioapic.c b/virt/kvm/ioapic.c
> index e8ce34c..859f5b8 100644
> --- a/virt/kvm/ioapic.c
> +++ b/virt/kvm/ioapic.c
> @@ -136,6 +136,9 @@ void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
>  {
>  	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
>  
> +	if (RTC_GSI >= IOAPIC_NUM_PINS)
> +		return;
> +
>  	spin_lock(&ioapic->lock);
>  	__rtc_irq_eoi_tracking_restore_one(vcpu);
>  	spin_unlock(&ioapic->lock);
> 

-- 
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux

Re: [PATCH] KVM: ioapic: add check for RTC_GSI

[Paolo Bonzini]

Il 11/09/2014 08:01, Jan Kiszka ha scritto:
> That issue is limited to the dead ia64 architecture, no other builds
> ioapic.c (besides x86, of course). Maybe finally remove KVM support for
> that arch instead of bloating otherwise correct code?

Yes, we should.  I've just been too busy to do it.

Paolo

Re: [PATCH] KVM: ioapic: add check for RTC_GSI

[Jan Kiszka]

On 2014-09-11 11:08, Paolo Bonzini wrote:
> Il 11/09/2014 08:01, Jan Kiszka ha scritto:
>> That issue is limited to the dead ia64 architecture, no other builds
>> ioapic.c (besides x86, of course). Maybe finally remove KVM support for
>> that arch instead of bloating otherwise correct code?
> 
> Yes, we should.  I've just been too busy to do it.

As I enjoy creating large negative diffs :), I would almost do it - if I
had a test environment for IA64...

Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux

Re: [PATCH] KVM: ioapic: add check for RTC_GSI

[Paolo Bonzini]

Il 11/09/2014 11:19, Jan Kiszka ha scritto:
> On 2014-09-11 11:08, Paolo Bonzini wrote:
>> Il 11/09/2014 08:01, Jan Kiszka ha scritto:
>>> That issue is limited to the dead ia64 architecture, no other builds
>>> ioapic.c (besides x86, of course). Maybe finally remove KVM support for
>>> that arch instead of bloating otherwise correct code?
>>
>> Yes, we should.  I've just been too busy to do it.
> 
> As I enjoy creating large negative diffs :), I would almost do it - if I
> had a test environment for IA64...

I have one. :)

Paolo