Re: [REGRESSION] Recent swiotlb DMA_FROM_DEVICE fixes break ath9k-based AP

[Johannes Berg]

On Mon, 2022-03-28 at 11:48 +0200, Johannes Berg wrote:
> 
> However, this basically means that the direction argument to the flush
> APIs are completely useless, and we do have to define something
> new/else...
> 

No I worded that badly - the direction isn't useless, but thinking of it
in terms of a buffer property rather than data movement is inaccurate.
So then if we need something else to indicate how data was expected to
be moved, the direction argument becomes useless, since it's not a
buffer property but rather a temporal thing on a specific place that
expected certain data movement.

johannes