Re: [PATCH 1/2] mm: move force_dma_unencrypted() to mem_encrypt.h

From: Christian Borntraeger
Date: Thu Feb 20 2020 - 11:23:32 EST

On 20.02.20 17:11, Christoph Hellwig wrote:
> On Thu, Feb 20, 2020 at 05:06:05PM +0100, Halil Pasic wrote:
>> Currently force_dma_unencrypted() is only used by the direct
>> implementation of the DMA API, and thus resides in dma-direct.h. But
>> there is nothing dma-direct specific about it: if one was -- for
>> whatever reason -- to implement custom DMA ops that have to in the
>> encrypted/protected scenarios dma-direct currently deals with, one would
>> need exactly this kind of information.
> I really don't think it has business being anywhre else, and your completely
> bogus second patch just proves the point.

>From a users perspective it makes absolutely perfect sense to use the
bounce buffers when they are NEEDED.
Forcing the user to specify iommu_platform just because you need bounce buffers
really feels wrong. And obviously we have a severe performance issue
because of the indirections.

Now: I understand that you want to get this fixes differently, but maybe you
could help to outline how this could be fixed proper.