Re: [GIT PULL] First batch of KVM changes for 4.1

From: Peter Zijlstra <peterz@infradead.org>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: torvalds@linux-foundation.org, linux-kernel@vger.kernel.org,
	gleb@kernel.org, kvm@vger.kernel.org,
	Ralf Baechle <ralf@linux-mips.org>,
	mtosatti@redhat.com, luto@kernel.org
Subject: Re: [GIT PULL] First batch of KVM changes for 4.1
Date: Fri, 17 Apr 2015 12:55:06 +0200	[thread overview]
Message-ID: <20150417105506.GF5029@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <5530E28F.2030401@redhat.com>

On Fri, Apr 17, 2015 at 12:38:07PM +0200, Paolo Bonzini wrote:
> 
> 
> On 17/04/2015 12:36, Peter Zijlstra wrote:
> > Now you make everybody pay for your crap, x86-64 paravirt or not. Keep
> > the cost by those who need it.
> > 
> > Please take it out, ASAP.
> 
> I'll just implement the static key.

Can you first show that:

preempt_out:
	int cpu = smp_processor_id();
	if (vcpu->cpu != cpu)
		vcpu->cpu = cpu;

preempt_in:
	int cpu = smp_processor_id();
	if (unlikely(vcpu->cpu != cpu))
		do_vcpu_migration_callback(cpu);

Is actually a measurable performance hit and we actually _need_ the
migration callback?

Also, it looks like you already do exactly this for other things, look
at:

	kvm_sched_in()
	  kvm_arch_vcpu_load()
	    if (unlikely(vcpu->cpu != cpu) ... )

So no, I don't believe for one second you need this.