Re: [parisc-linux] Re: [PATCH] Fix the cpumask rewrite

From: Jeff Garzik <jgarzik@pobox.com>
To: Linus Torvalds <torvalds@osdl.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>, Pavel Machek <pavel@suse.cz>,
	vojtech@suse.cz, James Bottomley <James.Bottomley@SteelEye.com>,
	Andrew Morton <akpm@osdl.org>, Paul Jackson <pj@sgi.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	PARISC list <parisc-linux@lists.parisc-linux.org>
Subject: Re: [parisc-linux] Re: [PATCH] Fix the cpumask rewrite
Date: Thu, 01 Jul 2004 12:52:11 -0400	[thread overview]
Message-ID: <40E4413B.5050000@pobox.com> (raw)
In-Reply-To: <Pine.LNX.4.58.0407010908260.11212@ppc970.osdl.org>

Linus Torvalds wrote:
> It _happens_ to be one on x86, so sadly it works on 99% of the machines 
> out there. And to avoid the extra (suprefluous) mb(), there are
> 
> 	smp_mb__before_clear_bit()
> 	smp_mb__after_clear_bit()
> 
> that only works with "clear_bit()", on the assumption that the way you'd 
> do locking is:
> 
> 	lock:
> 		while (test_and_set_bit(..)) /* This is a memory barrier */
> 			while (test_bit(..))
> 				cpu_relax();
> 
> 	.. protected region ..
> 
> 	unlock:
> 		smp_mb__before_clear_bit();
> 		clear_bit(..);

FWIW one of the major uses of bitops currently is e.g. in 
include/linux/netdevice.h, where bitops are used for atomic selection of 
code paths, but not spinning:

static inline void netif_carrier_off(struct net_device *dev)
{
         if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state))
                 linkwatch_fire_event(dev);
}