Re: [PATCH] Fixes for dw_dmac and atmel-mci for AP700x

From: Arnd Bergmann <arnd.bergmann@linaro.org>
To: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Hein Tibosch <hein_tibosch@yahoo.es>,
	"Hans-Christian Egtvedt" <hans-christian.egtvedt@atmel.com>,
	Nicolas Ferre <nicolas.ferre@atmel.com>,
	Havard Skinnemoen <havard@skinnemoen.net>,
	"ludovic.desroches" <ludovic.desroches@atmel.com>,
	linux-kernel@vger.kernel.org,
	"spear-devel" <spear-devel@list.st.com>
Subject: Re: [PATCH] Fixes for dw_dmac and atmel-mci for AP700x
Date: Tue, 21 Aug 2012 08:34:53 +0000	[thread overview]
Message-ID: <201208210834.53544.arnd.bergmann@linaro.org> (raw)
In-Reply-To: <CAKohpomMPeewDBVxVR=g-op7E53rqt-AZgOdyoib6W+5QsLOOA@mail.gmail.com>

On Tuesday 21 August 2012, Viresh Kumar wrote:
> On 21 August 2012 13:02, Hein Tibosch <hein_tibosch@yahoo.es> wrote:
> 
> > On 8/21/2012 2:35 PM, Viresh Kumar wrote:
> >
> > It got swapped as 0xAABB.CCDD => 0xCCDD.AABB
> >
> 
> @Arnd: How do we explain this? shouldn't it be DD CC BB AA??

Yes, this is very strange. Maybe the compiler already splits the
access into two 16-byte loads and that confuses the device?

> > Memory barriers: within the AVR32 code, one often sees explicit ways to
> > introduce memory barriers, e.g.:
> >
> >     hsmc_readl(hsmc, MODE0); /* I/O barrier */
> >
> 
> For ARM it has become a bit complex about using barriers. That's why they
> are added in readl/writel to remove any confusion.

To be more exact: the reason why readl/writel have the barriers is that
device drivers written for x86 expect the barrier semantics to be implied.
ARM also has readl_relaxed/writel_relaxed, which don't have a barrier
against DMA but still enforce ordering between the readl calls (implied
by the CPU architecture).

	Arnd