amd_iommu fallback question

amd_iommu fallback question

[Christoph Hellwig]

Hi Joerg,

I've got a little question on the amd_iommu code, hope you (or someone
else on the list) can answer it.  Most of the dma_map_ops methods have
a fallback for the case when get_domain() return -EINVAL, which basically
duplicates the normal no-iommu implementation, with the map_sg method
being and exception that always returns and error (return value of 0
is an error for its calling convention).  Is that an oversight?  If yes
so I wonder why it hasn't been caught yet.  Or is the whole fallback
code dead and can be removed?

Re: amd_iommu fallback question

[Joerg Roedel]

Hi Christoph,

On Thu, Apr 05, 2018 at 09:36:12PM +0200, Christoph Hellwig wrote:
> I've got a little question on the amd_iommu code, hope you (or someone
> else on the list) can answer it.  Most of the dma_map_ops methods have
> a fallback for the case when get_domain() return -EINVAL, which basically
> duplicates the normal no-iommu implementation, with the map_sg method
> being and exception that always returns and error (return value of 0
> is an error for its calling convention).  Is that an oversight?  If yes
> so I wonder why it hasn't been caught yet.  Or is the whole fallback
> code dead and can be removed?

Yeah, the fall-back code is basically dead since the amd driver uses
per-device dma_ops. It was just an oversight that it is not removed yet.


	Joerg