Re: [PATCH v2 12/15] kvm: i8254: Check LAPIC EOI pending when injecting irq on SVM AVIC

[Alexander Graf <graf@amazon.com>]

On 15.08.19 18:25, Suthikulpanit, Suravee wrote:
> ACK notifiers don't work with AMD SVM w/ AVIC when the PIT interrupt
> is delivered as edge-triggered fixed interrupt since AMD processors
> cannot exit on EOI for these interrupts.
> 
> Add code to check LAPIC pending EOI before injecting any pending PIT
> interrupt on AMD SVM when AVIC is activated.
> 
> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> ---
>   arch/x86/kvm/i8254.c | 31 +++++++++++++++++++++++++------
>   1 file changed, 25 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/x86/kvm/i8254.c b/arch/x86/kvm/i8254.c
> index 4a6dc54..31c4a9b 100644
> --- a/arch/x86/kvm/i8254.c
> +++ b/arch/x86/kvm/i8254.c
> @@ -34,10 +34,12 @@
>   
>   #include <linux/kvm_host.h>
>   #include <linux/slab.h>
> +#include <asm/virtext.h>
>   
>   #include "ioapic.h"
>   #include "irq.h"
>   #include "i8254.h"
> +#include "lapic.h"
>   #include "x86.h"
>   
>   #ifndef CONFIG_X86_64
> @@ -236,6 +238,12 @@ static void destroy_pit_timer(struct kvm_pit *pit)
>   	kthread_flush_work(&pit->expired);
>   }
>   
> +static inline void kvm_pit_reset_reinject(struct kvm_pit *pit)
> +{
> +	atomic_set(&pit->pit_state.pending, 0);
> +	atomic_set(&pit->pit_state.irq_ack, 1);
> +}
> +
>   static void pit_do_work(struct kthread_work *work)
>   {
>   	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
> @@ -244,6 +252,23 @@ static void pit_do_work(struct kthread_work *work)
>   	int i;
>   	struct kvm_kpit_state *ps = &pit->pit_state;
>   
> +	/*
> +	 * Since, AMD SVM AVIC accelerates write access to APIC EOI
> +	 * register for edge-trigger interrupts. PIT will not be able
> +	 * to receive the IRQ ACK notifier and will always be zero.
> +	 * Therefore, we check if any LAPIC EOI pending for vector 0
> +	 * and reset irq_ack if no pending.
> +	 */
> +	if (cpu_has_svm(NULL) && kvm->arch.apicv_state == APICV_ACTIVATED) {
> +		int eoi = 0;
> +
> +		kvm_for_each_vcpu(i, vcpu, kvm)
> +			if (kvm_apic_pending_eoi(vcpu, 0))
> +				eoi++;
> +		if (!eoi)
> +			kvm_pit_reset_reinject(pit);

In which case would eoi be != 0 when APIC-V is active?


Alex