[PATCH] arm: dma-mapping: move consistent_init to early_initcall

[skannan@codeaurora.org (Saravana Kannan)]

> On Fri, Dec 17, 2010 at 02:26:44AM -0800, Saravana Kannan wrote:
>> After Catalin's response to clarify, if we still end up not treating
>> secure domain as a "DMA device", then what's the alternative? Can we get
>> an explicit "cache invalidate API" that's outside of the DMA APIs? Or a
>> general uncached pages alloc/free APIs?
>
> The answer _always_ is that if the existing APIs don't give you want
> you need, then either the APIs need extending or a new API needs to
> be created.

I will wait to hear Catalin's clarification before I proceed further with
this point.

>
> In any case, have you looked at the latest set of SCM patches?  They
> don't make use of DMA coherent memory anymore, so the problem would
> seem to be resolved.
>

Yes, I saw them and even pointed out a few issues during internal-prelim
review. The new patches get around it with what we hope is a temporary
solution -- implement cache invalidate in assembly and call it over a
kmalloc'ed and page aligned memory.

-Saravana
-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.