New dma-noncoherent code, looking for comment and people to test

[Kumar Gala]

On Sep 29, 2008, at 12:26 PM, Remi Machet wrote:

> Hi,
>
> I rewrote dma-noncoherent.c and I am looking for people to review  
> and test it
> on various platforms that use it to make sure I did not introduce  
> any bug.
> The platforms affected by this change are those that define
> CONFIG_NOT_COHERENT_CACHE:
> ppc44x
> walnut
> makalu
> kilauea
> ep405
> adder875
> ep88xc
> taishan
> warp
> sam440ep
> sequoia
> bamboo
> canyonlands
> rainier
> katmai
> ebony
> mpc885_ads
> mpc866_ads
> ppc40x
> c2k (already tested)
> prpmc2800
>
> The old code in dma-noncoherent.c uses a memory pool at a hard coded
> virtual address set by CONFIG_CONSISTENT_START (typically  
> 0xFF100000). If not
> set carefully this address can conflict with early ioremap in systems
> that enable HIGHMEM, on top of that the code is overly complex  
> because it
> needs to have its own memory manager.
> This is why I tried to re-implement the code using standard memory
> management APIs. The new code does not require the user to set
> CONFIG_CONSISTENT_START or CONFIG_CONSISTENT_SIZE, is much smaller and
> simplier. It also can allocate as much memory as available in ZONE_DMA
> (instead of being limited by CONFIG_CONSISTENT_SIZE).
>
> I also removed the HIGHMEM support in dma_sync since memory  
> allocated for
> DMA transfer should always be in ZONE_DMA (ie not in ZONE_HIGHMEM).
>
> Looking forward to any comment about why this code may not work or  
> is not
> as good as the original. If you do test this code on your platform,  
> let me
> know how it goes ... if no-one object and no bug is found I will  
> submit
> this patch in a month or so.
>
> Thanks !
>
> Remi

We really should change this code over to the new dma changes Becky's  
introduced so we just have a non-coherent set of DMA ops (thus we can  
do both non-coherent and coherent in the same system).

- k