Re: [PATCH v4 3/7] [RFC] arm/arm64: introduce is_dma_coherent

[Stefano Stabellini <stefano.stabellini@eu.citrix.com>]

On Fri, 24 Oct 2014, Stefano Stabellini wrote:
> On Fri, 24 Oct 2014, Catalin Marinas wrote:
> > On Fri, Oct 24, 2014 at 12:39:59PM +0100, Stefano Stabellini wrote:
> > > On Fri, 24 Oct 2014, Catalin Marinas wrote:
> > > > I think a better way would be some Xen hook around
> > > > set_arch_dma_coherent_ops(). Does Xen have its own device tracking
> > > > structures? If not, you may be able to add another bitfield to the
> > > > kernel one.
> > > 
> > > We don't have an additional device tracking struct on Xen.
> > > I agree that a new bit somewhere would be the best solution, but I am
> > > not sure where. Maybe in dev_archdata under arm and arm64? After all it
> > > is already used to keep pointers to dma and coherency related
> > > structures.
> > 
> > I was thinking about something like below (maybe with some additional
> > ARCH_HAS_NONCOHERENT_DMA config for architectures that are always
> > coherent):

I don't think that introducing ARCH_HAS_NONCOHERENT_DMA is necessary here.


> > diff --git a/drivers/of/platform.c b/drivers/of/platform.c
> > index 3b64d0bf5bba..ae399ccbd569 100644
> > --- a/drivers/of/platform.c
> > +++ b/drivers/of/platform.c
> > @@ -183,6 +183,7 @@ static void of_dma_configure(struct device *dev)
> >  	 * dma coherent operations.
> >  	 */
> >  	if (of_dma_is_coherent(dev->of_node)) {
> > +		dev->dma_coherent = 1;
> >  		set_arch_dma_coherent_ops(dev);
> >  		dev_dbg(dev, "device is dma coherent\n");
> >  	}
> > diff --git a/include/linux/device.h b/include/linux/device.h
> > index ce1f21608b16..e00ca876db01 100644
> > --- a/include/linux/device.h
> > +++ b/include/linux/device.h
> > @@ -796,6 +796,7 @@ struct device {
> >  
> >  	bool			offline_disabled:1;
> >  	bool			offline:1;
> > +	bool			dma_coherent:1;
> >  };
> >  
> >  static inline struct device *kobj_to_dev(struct kobject *kobj)
> > diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> > index d5d388160f42..9c9ba5a5428e 100644
> > --- a/include/linux/dma-mapping.h
> > +++ b/include/linux/dma-mapping.h
> > @@ -78,6 +78,11 @@ static inline int is_device_dma_capable(struct device *dev)
> >  	return dev->dma_mask != NULL && *dev->dma_mask != DMA_MASK_NONE;
> >  }
> >  
> > +static inline int is_device_dma_coherent(struct device *dev)
> > +{
> > +	return dev->dma_coherent;
> > +}
> > +
> >  #ifdef CONFIG_HAS_DMA
> >  #include <asm/dma-mapping.h>
> >  #else
> 
> This is probably the cleanest option. I am going to send it out and see
> what the comments are.
> 
> I might still be able to request a backport if it doesn't make 3.18.
> 
> 
> > > However given the timing constraints I hope you would be OK with the
> > > suboptimal solution for now and create a common is_dma_coherent function
> > > in 3.19?
> > 
> > If you want to push something for 3.18, you could have a temporary
> > solution but I would prefer a bool or something in the dev_archdata
> > structure. Another untested patch:
> > 
> > diff --git a/arch/arm64/include/asm/device.h b/arch/arm64/include/asm/device.h
> > index cf98b362094b..243ef256b8c9 100644
> > --- a/arch/arm64/include/asm/device.h
> > +++ b/arch/arm64/include/asm/device.h
> > @@ -21,6 +21,7 @@ struct dev_archdata {
> >  #ifdef CONFIG_IOMMU_API
> >  	void *iommu;			/* private IOMMU data */
> >  #endif
> > +	bool dma_coherent;
> >  };
> >  
> >  struct pdev_archdata {
> > diff --git a/arch/arm64/include/asm/dma-mapping.h b/arch/arm64/include/asm/dma-mapping.h
> > index adeae3f6f0fc..b6bc4c268878 100644
> > --- a/arch/arm64/include/asm/dma-mapping.h
> > +++ b/arch/arm64/include/asm/dma-mapping.h
> > @@ -54,11 +54,17 @@ static inline void set_dma_ops(struct device *dev, struct dma_map_ops *ops)
> >  
> >  static inline int set_arch_dma_coherent_ops(struct device *dev)
> >  {
> > +	dev->dev_archdata.dma_coherent = true;
> >  	set_dma_ops(dev, &coherent_swiotlb_dma_ops);
> >  	return 0;
> >  }
> >  #define set_arch_dma_coherent_ops	set_arch_dma_coherent_ops
> >  
> > +static inline int is_device_dma_coherent(struct device *dev)
> > +{
> > +	return dev->dev_archdata.dma_coherent;
> > +}
> > +
> >  #include <asm-generic/dma-mapping-common.h>
> >  
> >  static inline dma_addr_t phys_to_dma(struct device *dev, phys_addr_t paddr)
> > 
> > 
> > This way you don't have to test for swiotlb vs iommu ops (we don't have
> > the latter yet on arm64 but they are coming).
> > 
> > -- 
> > Catalin
> > 
>