From mboxrd@z Thu Jan  1 00:00:00 1970
From: Paul Mundt <lethal@linux-sh.org>
Subject: Re: [PATCH] IRQ: simplify OMAP2 mask_irq/unmask_irq code
Date: Wed, 21 May 2008 09:55:14 +0900
Message-ID: <20080521005514.GA24455@linux-sh.org>
References: <alpine.DEB.1.00.0805201220090.9463@utopia.booyaka.com> <9c9fda240805201618y43d65160h246ea69a43785515@mail.gmail.com> <alpine.DEB.1.00.0805201802040.9463@utopia.booyaka.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Return-path: <linux-omap-owner@vger.kernel.org>
Received: from mta23.gyao.ne.jp ([125.63.38.249]:18963 "EHLO mx.gate01.com"
	rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP
	id S932714AbYEUA4r (ORCPT <rfc822;linux-omap@vger.kernel.org>);
	Tue, 20 May 2008 20:56:47 -0400
Content-Disposition: inline
In-Reply-To: <alpine.DEB.1.00.0805201802040.9463@utopia.booyaka.com>
Sender: linux-omap-owner@vger.kernel.org
List-Id: linux-omap@vger.kernel.org
To: Paul Walmsley <paul@pwsan.com>
Cc: Kyungmin Park <kmpark@infradead.org>, linux-omap@vger.kernel.org

On Tue, May 20, 2008 at 06:12:04PM -0600, Paul Walmsley wrote:
> On Wed, 21 May 2008, Kyungmin Park wrote:
> > On Wed, May 21, 2008 at 3:21 AM, Paul Walmsley <paul@pwsan.com> wrote:
> > >
> > >  static void omap_mask_irq(unsigned int irq)
> > >  {
> > > -       int offset = (irq >> 5) << 5;
> > > +       int offset = irq & (~(IRQ_BITS_PER_REG - 1));
> > >
> > > -       if (irq >= 64)
> > > -               irq %= 64;
> > > -       else if (irq >= 32)
> > > -               irq %= 32;
> > > +       irq %= IRQ_BITS_PER_REG;
> > 
> > Is it the right conversion?
> > If the irq is greater then 32 and less then or equal to  64 it's
> > result is different.
> > E.g, If irq is 63 then original irq is 63, but new code is 31
> 
> Hmm, in that condition, the result looks the same to me: irq % 32, either 
> way?
> 
> More practically, if you look at what it does with that irq variable 
> afterwards, it seems to be a bug if irq is ever greater than 31:
> 
>         intc_bank_write_reg(1 << irq, &irq_banks[0], INTC_MIR_CLEAR0 + 
>                             offset);
> 
> I think the only case where the new code would work differently than the 
> previous code is if irq > 95.  But that would be a bug, since the shift 
> value would then be > 32, for a 32-bit register.
> 
> > And if this code is right, how about to use mask instead of modulo op?
> > irq &= (IRQ_BITS_PER_REG - 1);
> 
> Hehe, very good point, that would probably save even more cycles!  If you 
> agree with the above, perhaps I can convert the code to use that also, 
> and add your Signed-off-by also?
> 
Kyungmin's idea looks good to me. If you roll the two together, feel free
to add my Acked-by also.