From mboxrd@z Thu Jan  1 00:00:00 1970
From: u.kleine-koenig@pengutronix.de (Uwe =?iso-8859-1?Q?Kleine-K=F6nig?=)
Date: Tue, 9 Sep 2014 11:43:39 +0200
Subject: [PATCH] GICv3: Fixing 32 bit compatibility
In-Reply-To: <20140909092653.GC4588@rric.localhost>
References: <1410185480-5153-1-git-send-email-rric@kernel.org>
 <20140909080042.GO23009@pengutronix.de>
 <20140909091201.GA4527@rric.localhost>
 <20140909091912.GQ23009@pengutronix.de>
 <20140909092653.GC4588@rric.localhost>
Message-ID: <20140909094339.GR23009@pengutronix.de>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On Tue, Sep 09, 2014 at 11:26:53AM +0200, Robert Richter wrote:
> On 09.09.14 11:19:12, Uwe Kleine-K?nig wrote:
> > On Tue, Sep 09, 2014 at 11:12:01AM +0200, Robert Richter wrote:
> > > On 09.09.14 10:00:42, Uwe Kleine-K?nig wrote:
> > > > On Mon, Sep 08, 2014 at 04:11:19PM +0200, Robert Richter wrote:
> > > 
> > > > > @@ -479,7 +479,7 @@ static void gic_raise_softirq(const struct cpumask *mask, unsigned int irq)
> > > > >     smp_wmb();
> > > > >  
> > > > >     for_each_cpu_mask(cpu, *mask) {
> > > > > -           u64 cluster_id = cpu_logical_map(cpu) & ~0xffUL;
> > > > > +           u64 cluster_id = cpu_logical_map(cpu) & ~0xffULL;
> > > > This doesn't change anything, does it?
> > > 
> > > It does, not in 64 bit but in 32 bit there unsigned long is 32
> > > bit. So, bit masks are broken if you compile a 32 bit kernel.
> > Can you make an example where the result actually changes?
> 
> Upper 32 bits will be cleared above.
ah, I missed the ~ while thinking about the difference.

> > > > I wonder if it would be cleaner to use (u64)0xff here.
> > > 
> > > No, that's ULL for. This is commonly used in x86 too.
> > I don't care much here, but I'd say ULL is to force an unsigned long
> > long. If you want to make it obvious that you want a 64bit value, a cast
> > to u64 makes this more clear.
> 
> ULL is no cast at all.
That's irrelevant here if it's a cast or not. It makes the constant be
an unsigned long long one.

If you want an unsigned long long value, use ULL (or a cast to unsigned
long long if the value permits), if you want an u64 use a cast to u64.

Probably it doesn't make sense to continue to argue here. The type of
cpu_logical_map(cpu) varies per architecture but is always <= 64 bits
wide and IIRC unsigned long long is guaranteed to be >= 64 bits wide so
everything is fine with your patch today.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |