From mboxrd@z Thu Jan  1 00:00:00 1970
From: Sasha Levin <levinsasha928@gmail.com>
Subject: Re: [PATCH 2/2] kvm tools: Modify ioport to use interval rbtree
Date: Sat, 21 May 2011 13:55:38 +0300
Message-ID: <1305975338.3205.23.camel@lappy>
References: <1305967911-30045-1-git-send-email-levinsasha928@gmail.com>
	 <1305967911-30045-2-git-send-email-levinsasha928@gmail.com>
	 <4DD79478.6040602@gmail.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Cc: penberg@kernel.org, mingo@elte.hu, asias.hejun@gmail.com,
	prasadjoshi124@gmail.com, kvm@vger.kernel.org
To: Cyrill Gorcunov <gorcunov@gmail.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mail-ww0-f44.google.com ([74.125.82.44]:34636 "EHLO
	mail-ww0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1754634Ab1EUKzv (ORCPT <rfc822;kvm@vger.kernel.org>);
	Sat, 21 May 2011 06:55:51 -0400
Received: by wwa36 with SMTP id 36so4766003wwa.1
        for <kvm@vger.kernel.org>; Sat, 21 May 2011 03:55:49 -0700 (PDT)
In-Reply-To: <4DD79478.6040602@gmail.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On Sat, 2011-05-21 at 14:31 +0400, Cyrill Gorcunov wrote:
> On 05/21/2011 12:51 PM, Sasha Levin wrote:
> > Currently the ioport implementation is based on a USHRT_MAX length
> > array of ptrs to ioport_operations.
> > 
> > Instead, use an interval rbtree to map the ioports to
> > ioport_operations.
> > 
> > Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
> > ---
> ...
> > -static struct ioport_operations *ioport_ops[USHRT_MAX];
> > -
> >  void ioport__register(u16 port, struct ioport_operations *ops, int count)
> >  {
> > -	int i;
> > +	struct ioport_entry *entry;
> >  
> > -	for (i = 0; i < count; i++)
> > -		ioport_ops[port + i]	= ops;
> > +	entry = ioport_search(&ioport_tree, port);
> > +	if (entry)
> > +		rb_int_erase(&ioport_tree, &entry->node);
> > +
> 
>   Hi Sasha, if I understand this correct we're simply drop old registartion, right? I think
> it should not be like that, if one port get used for several drivers/purposes we need a
> ref-counting, but at moment I think we simply should not allow to re-register port without
> previously unregister it. Or I miss something?

Currently we register some ports as dummy ports in the ioport
initialization, and re-register them once they get someone who can use
them (for example, serial device).

Not allowing ports to re-register would mean we can't reassign ports to
serial console when the serial console module gets loaded.

-- 

Sasha.