From mboxrd@z Thu Jan  1 00:00:00 1970
From: Avi Kivity <avi@redhat.com>
Subject: Re: [PATCH] kvm/ia64: Ensure SIGINT delivered to main thread (vcpu
 0).
Date: Mon, 20 Oct 2008 10:43:33 +0200
Message-ID: <48FC44B5.4010206@redhat.com>
References: <42DFA526FC41B1429CE7279EF83C6BDC01B489F8@pdsmsx415.ccr.corp.intel.com> <48FC3204.4030902@redhat.com> <42DFA526FC41B1429CE7279EF83C6BDC01B48ADD@pdsmsx415.ccr.corp.intel.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
Cc: kvm-ia64@vger.kernel.org, kvm@vger.kernel.org
To: "Zhang, Xiantao" <xiantao.zhang@intel.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx2.redhat.com ([66.187.237.31]:58796 "EHLO mx2.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1750724AbYJTIn1 (ORCPT <rfc822;kvm@vger.kernel.org>);
	Mon, 20 Oct 2008 04:43:27 -0400
In-Reply-To: <42DFA526FC41B1429CE7279EF83C6BDC01B48ADD@pdsmsx415.ccr.corp.intel.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

Zhang, Xiantao wrote:
> Avi Kivity wrote:
>   
>> Zhang, Xiantao wrote:
>>     
>>> Before Aps going to block status, it should make sure SIGINT is
>>> masked, otherwise, it may eat SIGINT from user killing the guest, and
>>> results in Qemu hanging there, becasue main thread can't get it to
>>> free guest's resource. 
>>>
>>>       
>>>  	vcpu_load(vcpu);
>>>
>>> +	if (vcpu->sigset_active)
>>> +		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); +
>>>  	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED))
>>>       
> {
>   
>>>  		kvm_vcpu_block(vcpu);
>>>  		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
>>> @@ -680,9 +683,6 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
>>>  		*vcpu, struct kvm_run *kvm_run) return -EAGAIN;
>>>
>>>       
>> The 'return' here means you are returning with the wrong signal mask.
>>     
>
> Yes, but the signal is eaten by the thread, and can't be delivered to
> main thread when returing to usespace. So in order to ensure the main
> thread receive the signal SIGINT we should mask this signal on Aps going
> into halt, and only main thread can receive such signals. 
>
>   

Then you should block SIGINT in the vcpu thread in userspace.

This patch is correct, except that it leaks the signal mask. You need to 
restore it when you return -EAGAIN.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.