Re: [PATCH 4/7] swiotlb: Allow arch override of address_needs_mapping

[Kumar Gala <galak@kernel.crashing.org>]

On Apr 9, 2009, at 1:50 PM, FUJITA Tomonori wrote:

> On Wed, 8 Apr 2009 23:59:18 -0500
> Kumar Gala <galak@kernel.crashing.org> wrote:
>
>>
>> On Apr 8, 2009, at 6:01 PM, FUJITA Tomonori wrote:
>>
>>>> Becky's patches of last week also added __weak annotations to
>>>> swiotlb_bus_to_virt, virt_to_bus and bus_to_phys; added the hwdev
>>>> parameter to swiotlb_bus_to_phys; and added a weak
>>>> swiotlb_arch_address_needs_mapping.  I assume that was needed  
>>>> because
>>>> powerpc needs non-trivial implementations for those functions.
>>>
>>> Hmm, what she added are wrappers of virt_to_bus and bus_to_virt. We
>>> can remove these and directly use virt_to_bus and bus_to_virt.
>>>
>>> About __weak address_needs_mapping function, as I said, removing it
>>> and using dma_map_ops is a proper solution.
>>
>> Is this something you are looking at doing in the .31 timeframe?
>>
>> I'm looking at the fact that we need to switch over to using struct
>> dma_map_ops on ppc.  (I'm guessing this might be the patches you
>> mentioned the other day).  If so did you add set_dma_mask() to the
>> generic dma_map_ops?
>
> Yeah, I'll send patches to convert ppc to use dma_map_ops. In .31
> timeframe, I plan to:
>
> - add a generic dma-mapping.h and convert ia64, x86, and ppc to use it
>
> - clean up swiotlb.
>
> - try to convert archs supporting multiple dma ops to use dma_map_ops
>
> - rewrite ia64 and x86 dma ops initialization
>
>
> BTW, the ppc specific swiotlb patchset is available?

Still cleaning it up, but I'll post a WIP patch so people have a sense  
of what the changes look like.

- k