From mboxrd@z Thu Jan  1 00:00:00 1970
From: Gleb Natapov <gleb@redhat.com>
Subject: Re: [PATCH][RFC] Use return value from kvm_set_irq() to re-inject
 PIT interrupts.
Date: Wed, 26 Aug 2009 16:50:40 +0300
Message-ID: <20090826135040.GF30093@redhat.com>
References: <20090824120623.GC30093@redhat.com>
 <20090824163256.GA8653@amt.cnet>
 <20090824171646.GA11762@redhat.com>
 <20090824174427.GA10311@amt.cnet>
 <20090824181905.GB11762@redhat.com>
 <20090826124348.GA32235@amt.cnet>
 <20090826131917.GE30093@redhat.com>
 <20090826133842.GB676@amt.cnet>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: kvm@vger.kernel.org, avi@redhat.com, zamsden@redhat.com
To: Marcelo Tosatti <mtosatti@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx1.redhat.com ([209.132.183.28]:41943 "EHLO mx1.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1757403AbZHZNuk (ORCPT <rfc822;kvm@vger.kernel.org>);
	Wed, 26 Aug 2009 09:50:40 -0400
Received: from int-mx05.intmail.prod.int.phx2.redhat.com (int-mx05.intmail.prod.int.phx2.redhat.com [10.5.11.18])
	by mx1.redhat.com (8.13.8/8.13.8) with ESMTP id n7QDog5S008856
	for <kvm@vger.kernel.org>; Wed, 26 Aug 2009 09:50:42 -0400
Content-Disposition: inline
In-Reply-To: <20090826133842.GB676@amt.cnet>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On Wed, Aug 26, 2009 at 10:38:42AM -0300, Marcelo Tosatti wrote:
> > > > > If you don't have a host timer emulating the guest PIT, to periodically
> > > > > bang on kvm_set_irq, how do you know when to attempt reinjection?
> > > > > 
> > > > > You keep calling kvm_set_irq on every guest entry to figure out when 
> > > > > reinjection is possible?
> > > > If we have timer to inject then yes. It is relatively cheap. Most of the
> > > > time pending count will be zero.
> > > 
> > > Won't work with non-tick-based emulation on the host.
> > Why? This is the most important point, can you elaborate?
> 
> >From http://www.mail-archive.com/kvm@vger.kernel.org/msg18644.html.
> 
> An injectable timer interrupt is defined by:
> 
> - time(now) >= time(next_expiration)
> - Previous timer interrupt has been acked (thus we can inject).
> 
> The thing is, sure you can drop ack notifiers and check IRR
> on every guest entry, but why bother if you can receive an
> asynchronous notification?
> 
> Would you prefer to replace
> 
> +               if (!ktimer->can_inject)
> 
> With
> 		kvm_set_irq()
> 
> ?
> 
> Not relatively cheap.
Most of the times time(now) will be less then time(next_expiration) so
on most entries kvm_set_irq() will not be called at all. When interrupt
has to be injected I prefer to try to inject it ASAP. PIC and APIC
effectively have 2 element interrupt queue (irr/isr) so injection may
succeed even though ack was not yet received.

--
			Gleb.