Re: [PATCH] RFC: kills consistent_dma_mask

[Jes Sorensen <jes@wildopensource.com>]

>>>>> "Krzysztof" == Krzysztof Halasa <khc@pm.waw.pl> writes:

Krzysztof> "David S. Miller" <davem@redhat.com> writes:
>> See, to show something is broken, you have to show a device that
>> will break currently.

Krzysztof> SBE wanXL sync serial adapter. 32 bits for buffers but 28
Krzysztof> bits for consistent data.

Well if the buffers are dynamic, why would they want to be allocated
using the consistent interface?

Krzysztof> I can't imagine all devices work properly on all platforms
Krzysztof> wrt consistent allocs. Say, sound drivers setting only
Krzysztof> dma_mask to < 32 bits and expecting consistent alloc will
Krzysztof> use that and not consistent_dma_mask.

If sound drivers set the dma_mask to something and expect that to
apply to the consistent allocations, then they aren't complying with
the current API and needs to be fixed.

Krzysztof> Of course, there is a question if we want to support such
Krzysztof> sound cards on Itaniums and Opterons? Of course they work
Krzysztof> on i386 as i386 pci_alloc_consistent() ignores
Krzysztof> consistent_dma_mask.

So fix it on ia32 to respect that mask for consistent allocations,
problem solved.

Jes