Hi Jing,
On Thu, Mar 17, 2022 at 12:56:30AM +0000, Jing Zhang wrote:
> This tracepoint only provides a hook for poking vcpu exits information,
> not exported to tracefs.
> A timestamp is added for the last vcpu exit, which would be useful for
> analysis for vcpu exits.
>
> Signed-off-by: Jing Zhang <jingzhangos@xxxxxxxxxx>
> ---
> arch/arm64/include/asm/kvm_host.h | 3 +++
> arch/arm64/kvm/arm.c | 2 ++
> arch/arm64/kvm/trace_arm.h | 8 ++++++++
> 3 files changed, 13 insertions(+)
>
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index daa68b053bdc..576f2c18d008 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -415,6 +415,9 @@ struct kvm_vcpu_arch {
>
> /* Arch specific exit reason */
> enum arm_exit_reason exit_reason;
> +
> + /* Timestamp for the last vcpu exit */
> + u64 last_exit_time;
> };
>
> /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index f49ebdd9c990..98631f79c182 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -783,6 +783,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
> ret = 1;
> run->exit_reason = KVM_EXIT_UNKNOWN;
> while (ret > 0) {
> + trace_kvm_vcpu_exits(vcpu);
> /*
> * Check conditions before entering the guest
> */
> @@ -898,6 +899,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
> local_irq_enable();
>
> trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
> + vcpu->arch.last_exit_time = ktime_to_ns(ktime_get());
>
> /* Exit types that need handling before we can be preempted */
> handle_exit_early(vcpu, ret);
> diff --git a/arch/arm64/kvm/trace_arm.h b/arch/arm64/kvm/trace_arm.h
> index 33e4e7dd2719..3e7dfd640e23 100644
> --- a/arch/arm64/kvm/trace_arm.h
> +++ b/arch/arm64/kvm/trace_arm.h
> @@ -301,6 +301,14 @@ TRACE_EVENT(kvm_timer_emulate,
> __entry->timer_idx, __entry->should_fire)
> );
>
> +/*
> + * Following tracepoints are not exported in tracefs and provide hooking
> + * mechanisms only for testing and debugging purposes.
> + */
> +DECLARE_TRACE(kvm_vcpu_exits,
> + TP_PROTO(struct kvm_vcpu *vcpu),
> + TP_ARGS(vcpu));
> +
When we were discussing this earlier, I wasn't aware of the kvm_exit
tracepoint which I think encapsulates what you're looking for.
ESR_EL2.EC is the critical piece to determine what caused the exit.
It is probably also important to call out that this trace point only
will fire for a 'full' KVM exit (i.e. out of hyp and back to the
kernel). There are several instances where the exit is handled in hyp
and we immediately resume the guest.
Now -- I am bordering on clueless with tracepoints, but it isn't
immediately obvious how the attached program can determine the vCPU that
triggered the TP. If we are going to propose modularizing certain KVM
metrics with tracepoints then that would be a rather critical piece of
information.
Apologies for any confusion I added to the whole situation, but
hopefully we can still engage in a broader conversation regarding
how to package up optional KVM metrics.
--
Thanks,
Oliver
