Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 8 Jun 2021 16:35:43 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Wanpeng Li <kernellwp@gmail.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        Paolo Bonzini <pbonzini@redhat.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>
Subject: Re: [PATCH v2 3/3] KVM: X86: Let's harden the ipi fastpath condition
 edge-trigger mode
Message-ID: <YL+cX8K3r7EWrk33@google.com>
References: <1623050385-100988-1-git-send-email-wanpengli@tencent.com>
 <1623050385-100988-3-git-send-email-wanpengli@tencent.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1623050385-100988-3-git-send-email-wanpengli@tencent.com>
Precedence: bulk

On Mon, Jun 07, 2021, Wanpeng Li wrote:
> From: Wanpeng Li <wanpengli@tencent.com>
> 
> Let's harden the ipi fastpath condition edge-trigger mode.

Can you elaborate on the motivation for this patch?

Intel's SDM states that the trigger mode is ignored for all IPIs except INIT,
and even clarifies that the local xAPIC will override the bit and send the IPI
as edge-triggered.

AMD's APM on the other hand explicitly lists level-triggered Fixed IPIs as a
valid ICR combination.

Regardless of which of the two conflicting specs we want KVM to emulate (which
is currently AMD), I don't see why the fastpath code should care, as I can't
find anything in the kvm_apic_send_ipi() path that would go awry if it's called
from the fastpath for a level-triggered IPI.

Related side topic, anyone happen to know if KVM (and Qemu's) emulation of IPIs
intentionally follows AMD instead of Intel?  I suspect it's unintentional,
especially since KVM's initial xAPIC emulation came from Intel.  Not that it's
likely to matter, but allowing level-triggered IPIs is bizarre, e.g. getting an
EOI sent to the right I/O APIC at the right time via a level-triggered IPI seems
extremely convoluted.

> Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> ---
>  arch/x86/kvm/x86.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index b594275..dbd3e9d 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -1922,6 +1922,7 @@ static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data
>  		return 1;
>  
>  	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
> +		((data & APIC_INT_LEVELTRIG) == 0) &&
>  		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
>  		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
>  		((u32)(data >> 32) != X2APIC_BROADCAST)) {
> -- 
> 2.7.4
>