On Thu, Nov 29, 2018 at 9:25 AM Rob Clark <robdclark@xxxxxxxxx> wrote:
>
> On Thu, Nov 29, 2018 at 9:14 AM Christoph Hellwig <hch@xxxxxx> wrote:
> >
> > On Thu, Nov 29, 2018 at 07:33:15PM +0530, Vivek Gautam wrote:
> > > dma_map_sg() expects a DMA domain. However, the drm devices
> > > have been traditionally using unmanaged iommu domain which
> > > is non-dma type. Using dma mapping APIs with that domain is bad.
> > >
> > > Replace dma_map_sg() calls with dma_sync_sg_for_device{|cpu}()
> > > to do the cache maintenance.
> >
> > As I told you before: hell no. If you spent the slightest amount of
> > actually trying to understand what you are doing here you'd know this
> > can't work. Just turn on dma debugging and this will blow up in your
> > face.
>
> you can tone it down.. we weren't the ones who created the dma/iommu
> mess, we are just trying to find a way to work around it
>
> > Either you use the DMA API properly, that is you use it to map and
> > to sync, or you don't use it at all. Mix and match between iommu
> > APIs and DMA APIs is simply not possible.
>
> I'd *love* nothing more to not use the dma api.. but on arm there is
> no other way to do cache maint.
btw, one of us will try this w/ dma debugging enabled..
Maybe the thing we need to do is just implement a blacklist of
compatible strings for devices which should skip the automatic
iommu/dma hookup. Maybe a bit ugly, but it would also solve a problem
preventing us from enabling per-process pagetables for a5xx (where we
need to control the domain/context-bank that is allocated by the dma
api).
Like I've said before, I'm open to other suggestions. But the
automagic behind-the-scenes dma+iommu hookup really screwed us, and we
need to find some sort of solution instead of downstream hacks.
BR,
-R
