From mboxrd@z Thu Jan  1 00:00:00 1970
From: Gleb Natapov <gleb@redhat.com>
Subject: Re: [regression] KVM: hangs and "irq timeout" booting HURD unless
 -no-kvm-irqchip passed
Date: Tue, 8 Feb 2011 16:47:25 +0200
Message-ID: <20110208144725.GN14984@redhat.com>
References: <1283171743-10766-1-git-send-email-avi@redhat.com>
 <1283171743-10766-2-git-send-email-avi@redhat.com>
 <20110207060030.GA1896@elie>
 <4D4FE7E5.2040308@redhat.com>
 <20110207124513.GB14984@redhat.com>
 <20110207132729.GC14984@redhat.com>
 <20110208014032.GB1858@elie>
 <20110208120037.GJ14984@redhat.com>
 <20110208142253.GA8243@amt.cnet>
 <4D515695.5080600@redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Marcelo Tosatti <mtosatti@redhat.com>,
	Jonathan Nieder <jrnieder@gmail.com>, kvm@vger.kernel.org,
	Michael Tokarev <mjt@tls.msk.ru>,
	Guillem Jover <guillem@debian.org>
To: Avi Kivity <avi@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx1.redhat.com ([209.132.183.28]:1031 "EHLO mx1.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1752645Ab1BHOre (ORCPT <rfc822;kvm@vger.kernel.org>);
	Tue, 8 Feb 2011 09:47:34 -0500
Content-Disposition: inline
In-Reply-To: <4D515695.5080600@redhat.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On Tue, Feb 08, 2011 at 04:43:33PM +0200, Avi Kivity wrote:
> On 02/08/2011 04:22 PM, Marcelo Tosatti wrote:
> >I don't think the isr_ack logic is overly complex that it should be
> >removed. For some cases it is still beneficial, see example case on
> >commit e48258009d941, which is not handled by kick coalescing of
> >kvm_vcpu_kick.
> 
> On the other hand, I think it can be done differently.  For example
> LVT0 is probably programmed to mask interrupts; we can simply look
> at it and not kick if that's the case.  We can use notifiers from
> the lapic to the pic to avoid looking at lapic data.
> 
I believe this is what my patch is doing. Look at pic_unlock(). The code
search for vcpu to kick by calling kvm_apic_accept_pic_intr() function
(which checks that LVT is masked). If no vcpu is found we kicks bsp.
Why? I removed that.

> The advantage in this way is that we avoid introducing state,
> instead relying on existing state.  New state is always bad since it
> has to be kept in sync with guest visible state.
> 
> -- 
> error compiling committee.c: too many arguments to function

--
			Gleb.