Re: [PATCH] Define wc_wmb, a write barrier for PCI write combining

[Benjamin LaHaise <bcrl@kvack.org>]

On Fri, Feb 24, 2006 at 08:20:50PM -0800, Bryan O'Sullivan wrote:
> On some platforms, a regular wmb() is not sufficient to guarantee that
> PCI writes have been flushed to the bus if write combining is in effect.
> 
> This change introduces a new macro, wc_wmb(), that makes this guarantee.
> 
> It does so by way of a new header file, <linux/system.h>.  This header
> can be a site for oft-replicated code from <asm-*/system.h>, but isn't
> just yet.
> 
> We also define a version of wc_wmb() with the required semantics
> on x86_64.

This does not do what you're trying to accomplish.  sfence has no impact 
on the flushing of the write combining buffers -- they're normally flushed 
after a few cycles of no subsequent merges.  If it is really necessary to 
flush such a write, you will have to do a read from the target PCI device 
(well, bus would do).

		-ben
-- 
"Ladies and gentlemen, I'm sorry to interrupt, but the police are here 
and they've asked us to stop the party."  Don't Email: <dont@kvack.org>.