Date: Mon, 06 Mar 2006 11:32:53 +0900 (JST)
Message-Id: <20060306.113253.88700204.nemoto@toshiba-tops.co.jp>
To: paulus@samba.org
Cc: akpm@osdl.org, ak@muc.de, clameter@engr.sgi.com,
       linux-kernel@vger.kernel.org, ralf@linux-mips.org, johnstul@us.ibm.com,
       rth@twiddle.net
Subject: Re: [PATCH] simplify update_times (avoid jiffies/jiffies_64
 aliasing problem)
From: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
In-Reply-To: <17418.7988.409791.419494@cargo.ozlabs.ibm.com>
References: <20060304112010.GA94875@muc.de>
	<20060304034050.40f29251.akpm@osdl.org>
	<17418.7988.409791.419494@cargo.ozlabs.ibm.com>
Mime-Version: 1.0
Content-Type: Text/Plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1215
Lines: 32

>>>>> On Sun, 5 Mar 2006 10:13:56 +1100, Paul Mackerras <paulus@samba.org> said:
paul> I think the thing that makes most sense is:

paul> #define jiffies ((unsigned long) jiffies_64)

paul> and fix the few drivers that use `jiffies' as a local variable.
paul> No-one should be trying to write to jiffies, and the compiler
paul> will do the right thing for reads of jiffies on 32-bit platforms
paul> (it does on ppc32 at least).

I think making jiffies a non l-value is good, but this drops volatile
attribute from jiffies (since jiffies_64 is not volatile).  It might
break some codes which do polling on jiffies.

Something like this would be better?

#if defined(__BIG_ENDIAN) && BITS_PER_LONG == 32
#define jiffies (*((const volatile unsigned long *)&jiffies_64 + 1))
#else
#define jiffies (*(const volatile unsigned long *)&jiffies_64)
#endif

And if we do not want to depend on -fno-strict-aliasing, the union
would be The Only Neat Thing to Do.

---
Atsushi Nemoto
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/