Re: arch/x86/Kconfig selects invalid HAVE_READQ, HAVE_WRITEQ vars

From: "H. Peter Anvin" <hpa@zytor.com>
To: Roland Dreier <rdreier@cisco.com>
Cc: Hitoshi Mitake <h.mitake@gmail.com>, Ingo Molnar <mingo@elte.hu>,
	Thomas Gleixner <tglx@linutronix.de>,
	"Robert P. J. Day" <rpjday@crashcourse.ca>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: arch/x86/Kconfig selects invalid HAVE_READQ, HAVE_WRITEQ vars
Date: Tue, 21 Apr 2009 12:09:20 -0700	[thread overview]
Message-ID: <49EE19E0.8040405@zytor.com> (raw)
In-Reply-To: <adaab69x6nd.fsf@cisco.com>

Roland Dreier wrote:
>  > > However I worry that this just leaves driver authors too much rope.
>  > > Choosing readq_atomic() vs. readq() is just one more thing to get wrong.
> 
>  > ... as is having each driver implementing their own substitutes.
> 
> Yes, I agree with that.  However at least status quo ante (readq/writeq
> 64-bit only) means that driver authors who use readq/writeq are forced
> (by a compile error) to spend a little thought on what 32-bit fallback
> they should use.
> 
> I guess one possibility is to make readq/writeq the atomic version, and
> add readq_nonatomic()/writeq_nonatomic() for 32-bit architectures.  Then
> it's much more opt-in -- but then that makes the (perhaps) more common
> case look a bit uglier.
> 

Do you really expect driver authors to type writeq_nonatomic() for every 
register reference?

I think an #include at the top is one thing, but something that 
heavyweight for each call site really isn't going to fly.

	-hpa