Re: [KVM PATCH v6 3/4] KVM: Fix races in irqfd using new eventfd_kref_get interface

From: Avi Kivity <avi@redhat.com>
To: Gregory Haskins <ghaskins@novell.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>,
	kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
	davidel@xmailserver.org
Subject: Re: [KVM PATCH v6 3/4] KVM: Fix races in irqfd using new	eventfd_kref_get interface
Date: Wed, 01 Jul 2009 12:44:39 +0300	[thread overview]
Message-ID: <4A4B3007.7080303@redhat.com> (raw)
In-Reply-To: <4A48F148.5030501@novell.com>

On 06/29/2009 07:52 PM, Gregory Haskins wrote:
>
>>    One idea I had to make it even clearer was to have a shutdown list
>> of irqfds per-kvm, together with the items list, and make work_struct for
>> shutdown global, not per-irqfd.  We can then unconditionally do
>> list_move + schedule_work to shut down an irqfd, and it's safe to do
>> even if it is already on the shutdown list - it just gets moved to tail.
>>
>>      
>
> Hmm..I'm not sure that churn really buys us anything, tho.  Technically
> the "active" bit is redundant with list_del_init()+list_empty() that I
> employed in previous versions.  However, I made it explicit with the
> active bit to be more self-documenting.  IMO, the latest code is pretty
> clear, and the change you are proposing is moving towards a slightly
> trickier variant like I originally had.  I'd say "lets leave this as is".
>    

Could retain self documentation by introducing a helper irqfd_active() 
which does the list_blah() magic.

-- 
error compiling committee.c: too many arguments to function