Re: [PATCH] x86: Remove readq()/writeq() on 32-bit

From: Jeff Garzik <jeff@garzik.org>
To: "H. Peter Anvin" <hpa@zytor.com>
Cc: Hitoshi Mitake <h.mitake@gmail.com>,
	Roland Dreier <rdreier@cisco.com>, Ingo Molnar <mingo@elte.hu>,
	David Miller <davem@davemloft.net>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	tglx@linutronix.de, rpjday@crashcourse.ca,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] x86: Remove readq()/writeq() on 32-bit
Date: Wed, 13 May 2009 18:39:24 -0400	[thread overview]
Message-ID: <4A0B4C1C.9000706@garzik.org> (raw)
In-Reply-To: <4A0B2B54.5090803@zytor.com>

H. Peter Anvin wrote:
> #include <linux/io64_lh.h>
> 
> linux/io64_lh.h would then look like:
> 
> #ifndef _LINUX_IO64_LH_H
> #define _LINUX_IO64_LH_H
> 
> #include <linux/io.h>
> 
> #ifndef HAVE_READQ
> 
> /* Low-High nonatomic readq() */
> 
> #endif
> 
> #ifndef HAVE_WRITEQ
> 
> /* Low-High nonatomic writeq() */

Judging from this thread and past, I think people will continue to 
complain and get confused, even with the above.

How about

1) tree-wide rename: readq -> readq_na, writeq -> writeq_na

2) make all PCI writel-enabled arches provide readq_na and writeq_na

3) 64-bit PCI writel-enabled arches provide readq and HAVE_READQ, ditto 
for writeq.  32-bit provides neither readq nor HAVE_READQ.

4) If your hardware has non-standard ordering, handle it the obvious way 
in the driver, as you would do with any other special case.

	Jeff

P.S. I use "writel-enabled" to note that some platforms do not provide 
PCI read*/write* functions at all.