Re: [PATCH] KVM: x86: check ISR and TMR to construct eoi exit bitmap

[Alex Williamson <alex.williamson@redhat.com>]

On Fri, 2014-08-08 at 14:02 +0800, Yang Zhang wrote:
> From: Yang Zhang <yang.z.zhang@Intel.com>
> 
> Guest may mask the IOAPIC entry before issue EOI. In such case,
> EOI will not be intercepted by hypervisor due to the corrensponding
> bit in eoi exit bitmap is not setting.
> 
> The solution is to check ISR + TMR to construct the EOI exit bitmap.
> 
> This patch is a better fixing for the issue that commit "0f6c0a740b"
> tries to solve.
> 
> Signed-off-by: Yang Zhang <yang.z.zhang@Intel.com>
> ---
>  arch/x86/kvm/lapic.c |   22 ++++++++++++++++++++++
>  arch/x86/kvm/lapic.h |    2 ++
>  arch/x86/kvm/x86.c   |    9 +++++++++
>  virt/kvm/ioapic.c    |    7 ++++---
>  4 files changed, 37 insertions(+), 3 deletions(-)
> 
> hi, alex
> 
> This patch is not tested since i don't have the environment to do it. Can you
> help to test it? thanks.

Yes, this also fixes the problem.

> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index 08e8a89..d2f9a6e 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
...
> @@ -515,6 +520,23 @@ static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
>  	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
>  }
>  
> +void kvm_apic_zap_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap,
> +		u32 *tmr)
> +{
> +	u32 i, reg_off, intr_in_service[8];

There's no reason for intr_in_service to be an array here.  Thanks,

Alex

> +	struct kvm_lapic *apic = vcpu->arch.apic;
> +
> +	for (i = 0; i < 8; i++) {
> +		reg_off = 0x10 * i;
> +		intr_in_service[i] = apic_read_reg(apic, APIC_ISR + reg_off) &
> +			apic_read_reg(apic, APIC_TMR + reg_off);
> +		if (intr_in_service[i]) {
> +			*((u32 *)eoi_exit_bitmap + i) |= intr_in_service[i];
> +			tmr[i] |= intr_in_service[i];
> +		}
> +	}
> +}
> +