[PATCH URGENT] arm: dma-mapping: Set DMA IOMMU ops in arm_iommu_attach_device()

[Will Deacon]

On Fri, Jan 23, 2015 at 05:00:45PM +0000, Arnd Bergmann wrote:
> On Friday 23 January 2015 17:55:25 Laurent Pinchart wrote:
> > 
> > On Friday 23 January 2015 16:27:24 Arnd Bergmann wrote:
> > > On Friday 23 January 2015 16:21:49 Laurent Pinchart wrote:
> > > > +/**
> > > > + * arm_iommu_detach_device
> > > > + * @dev: valid struct device pointer
> > > > + *
> > > > + * Detaches the provided device from a previously attached map.
> > > > + * This voids the dma operations (dma_map_ops pointer)
> > > > + */
> > > > +void arm_iommu_detach_device(struct device *dev)
> > > > +{
> > > > +       __arm_iommu_detach_device(dev);
> > > > +       set_dma_ops(dev, NULL);
> > > > +}
> > > > 
> > > >  EXPORT_SYMBOL_GPL(arm_iommu_detach_device);
> > > 
> > > Would this introduce a regression in the case where the device is
> > > cache coherent and needs arm_coherent_dma_ops set?
> > 
> > I think we need to handle that case (as well as the coherent IOMMU case in 
> > arm_iommu_attach_device), but this patch only tries to restore the previous 
> > behaviour to fix the bug detailed in the commit message. Would you prefer 
> > fixing both issues in one go ?
> > 
> 
> No, I was specifically trying to find out whether the new behavior would
> be worse than what we had in 3.18. If it's a preexisting bug that nobody
> has run into, there is no hurry.

Agreed, I think Laurent's proposal here is the smallest thing that fixes
the issue without breaking the new of_xlate-based configuration.

  Acked-by: Will Deacon <will.deacon at arm.com>

> It's quite likely that to date, all IOMMU users on ARM32 are not
> cache coherent.

Highbank has an ARM SMMU, but I don't see the iommu_coherent_ops getting
used, so your assertion is probably right.

Will