Re: [GIT PULL] dma-mapping updates for Linux 6.12

[Xi Ruoyao <xry111@xry111.site>]

On Sun, 2024-09-22 at 00:19 -0700, Christoph Hellwig wrote:
> On Sun, Sep 22, 2024 at 02:56:57PM +0800, Xi Ruoyao wrote:
> > With SND disabled, b5c58b2fdc427e7958412ecb2de2804a1f7c1572 boots fine
> > (no oops), but the camera does not work. 
> > f69e342eec008e1bab772d3963c3dd9979293e13 boots fine and the camera works
> > fine.
> > 
> > So the first bad commit is b5c58b2fdc427e7958412ecb2de2804a1f7c1572.
> 
> Thanks a lot for the bisection!
> 
> I took a quitck look what is special in the intersection of uvcvideo
> and that commit, and what springs to meind is that
> uvcvideo is one of the very few users of dma_alloc_noncontiguous, and
> while that and the free call for it got converted correctly, the
> conversion was missed for the vmap and mmap callbacks.
> 
> Can you test the fix below?

Yes with the fix the camera is working fine now.  Thanks!

> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 3672d619bcb691..2a9fa0c8cc00fe 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -1038,6 +1038,21 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
>  	return NULL;
>  }
>  
> +/*
> + * This is the actual return value from the iommu_dma_alloc_noncontiguous.
> + *
> + * The users of the DMA API should only care about the sg_table, but to make
> + * the DMA-API internal vmaping and freeing easier we stash away the page
> + * array as well (except for the fallback case).  This can go away any time,
> + * e.g. when a vmap-variant that takes a scatterlist comes along.
> + */
> +struct dma_sgt_handle {
> +	struct sg_table sgt;
> +	struct page **pages;
> +};
> +#define sgt_handle(sgt) \
> +	container_of((sgt), struct dma_sgt_handle, sgt)
> +
>  struct sg_table *iommu_dma_alloc_noncontiguous(struct device *dev, size_t size,
>  	       enum dma_data_direction dir, gfp_t gfp, unsigned long attrs)
>  {
> @@ -1066,6 +1081,24 @@ void iommu_dma_free_noncontiguous(struct device *dev, size_t size,
>  	kfree(sh);
>  }
>  
> +void *iommu_dma_vmap_noncontiguous(struct device *dev, size_t size,
> +		struct sg_table *sgt)
> +{
> +	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
> +
> +	return vmap(sgt_handle(sgt)->pages, count, VM_MAP, PAGE_KERNEL);
> +}
> +
> +int iommu_dma_mmap_noncontiguous(struct device *dev, struct vm_area_struct *vma,
> +		size_t size, struct sg_table *sgt)
> +{
> +	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
> +
> +	if (vma->vm_pgoff >= count || vma_pages(vma) > count - vma->vm_pgoff)
> +		return -ENXIO;
> +	return vm_map_pages(vma, sgt_handle(sgt)->pages, count);
> +}
> +
>  void iommu_dma_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle,
>  		size_t size, enum dma_data_direction dir)
>  {
> diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h
> index 9668ddf3696e71..b7773201414c27 100644
> --- a/include/linux/dma-map-ops.h
> +++ b/include/linux/dma-map-ops.h
> @@ -24,11 +24,6 @@ struct dma_map_ops {
>  			gfp_t gfp);
>  	void (*free_pages)(struct device *dev, size_t size, struct page *vaddr,
>  			dma_addr_t dma_handle, enum dma_data_direction dir);
> -	struct sg_table *(*alloc_noncontiguous)(struct device *dev, size_t size,
> -			enum dma_data_direction dir, gfp_t gfp,
> -			unsigned long attrs);
> -	void (*free_noncontiguous)(struct device *dev, size_t size,
> -			struct sg_table *sgt, enum dma_data_direction dir);
>  	int (*mmap)(struct device *, struct vm_area_struct *,
>  			void *, dma_addr_t, size_t, unsigned long attrs);
>  
> @@ -206,20 +201,6 @@ static inline int dma_mmap_from_global_coherent(struct vm_area_struct *vma,
>  }
>  #endif /* CONFIG_DMA_GLOBAL_POOL */
>  
> -/*
> - * This is the actual return value from the ->alloc_noncontiguous method.
> - * The users of the DMA API should only care about the sg_table, but to make
> - * the DMA-API internal vmaping and freeing easier we stash away the page
> - * array as well (except for the fallback case).  This can go away any time,
> - * e.g. when a vmap-variant that takes a scatterlist comes along.
> - */
> -struct dma_sgt_handle {
> -	struct sg_table sgt;
> -	struct page **pages;
> -};
> -#define sgt_handle(sgt) \
> -	container_of((sgt), struct dma_sgt_handle, sgt)
> -
>  int dma_common_get_sgtable(struct device *dev, struct sg_table *sgt,
>  		void *cpu_addr, dma_addr_t dma_addr, size_t size,
>  		unsigned long attrs);
> diff --git a/include/linux/iommu-dma.h b/include/linux/iommu-dma.h
> index 1bb55ca1ab79d7..7bf145a52d6a1a 100644
> --- a/include/linux/iommu-dma.h
> +++ b/include/linux/iommu-dma.h
> @@ -44,6 +44,12 @@ struct sg_table *iommu_dma_alloc_noncontiguous(struct device *dev, size_t size,
>  		enum dma_data_direction dir, gfp_t gfp, unsigned long attrs);
>  void iommu_dma_free_noncontiguous(struct device *dev, size_t size,
>  		struct sg_table *sgt, enum dma_data_direction dir);
> +void *iommu_dma_vmap_noncontiguous(struct device *dev, size_t size,
> +		struct sg_table *sgt);
> +#define iommu_dma_vunmap_noncontiguous(dev, vaddr) \
> +	vunmap(vaddr);
> +int iommu_dma_mmap_noncontiguous(struct device *dev, struct vm_area_struct *vma,
> +		size_t size, struct sg_table *sgt);
>  void iommu_dma_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle,
>  		size_t size, enum dma_data_direction dir);
>  void iommu_dma_sync_single_for_device(struct device *dev, dma_addr_t dma_handle,
> diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
> index b839683da0baf0..7911c754d9f42a 100644
> --- a/kernel/dma/mapping.c
> +++ b/kernel/dma/mapping.c
> @@ -750,7 +750,6 @@ static struct sg_table *alloc_single_sgt(struct device *dev, size_t size,
>  struct sg_table *dma_alloc_noncontiguous(struct device *dev, size_t size,
>  		enum dma_data_direction dir, gfp_t gfp, unsigned long attrs)
>  {
> -	const struct dma_map_ops *ops = get_dma_ops(dev);
>  	struct sg_table *sgt;
>  
>  	if (WARN_ON_ONCE(attrs & ~DMA_ATTR_ALLOC_SINGLE_PAGES))
> @@ -758,9 +757,7 @@ struct sg_table *dma_alloc_noncontiguous(struct device *dev, size_t size,
>  	if (WARN_ON_ONCE(gfp & __GFP_COMP))
>  		return NULL;
>  
> -	if (ops && ops->alloc_noncontiguous)
> -		sgt = ops->alloc_noncontiguous(dev, size, dir, gfp, attrs);
> -	else if (use_dma_iommu(dev))
> +	if (use_dma_iommu(dev))
>  		sgt = iommu_dma_alloc_noncontiguous(dev, size, dir, gfp, attrs);
>  	else
>  		sgt = alloc_single_sgt(dev, size, dir, gfp);
> @@ -786,13 +783,10 @@ static void free_single_sgt(struct device *dev, size_t size,
>  void dma_free_noncontiguous(struct device *dev, size_t size,
>  		struct sg_table *sgt, enum dma_data_direction dir)
>  {
> -	const struct dma_map_ops *ops = get_dma_ops(dev);
> -
>  	trace_dma_unmap_sg(dev, sgt->sgl, sgt->orig_nents, dir, 0);
>  	debug_dma_unmap_sg(dev, sgt->sgl, sgt->orig_nents, dir);
> -	if (ops && ops->free_noncontiguous)
> -		ops->free_noncontiguous(dev, size, sgt, dir);
> -	else if (use_dma_iommu(dev))
> +
> +	if (use_dma_iommu(dev))
>  		iommu_dma_free_noncontiguous(dev, size, sgt, dir);
>  	else
>  		free_single_sgt(dev, size, sgt, dir);
> @@ -802,37 +796,26 @@ EXPORT_SYMBOL_GPL(dma_free_noncontiguous);
>  void *dma_vmap_noncontiguous(struct device *dev, size_t size,
>  		struct sg_table *sgt)
>  {
> -	const struct dma_map_ops *ops = get_dma_ops(dev);
> -	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
>  
> -	if (ops && ops->alloc_noncontiguous)
> -		return vmap(sgt_handle(sgt)->pages, count, VM_MAP, PAGE_KERNEL);
> +	if (use_dma_iommu(dev))
> +		return iommu_dma_vmap_noncontiguous(dev, size, sgt);
> +
>  	return page_address(sg_page(sgt->sgl));
>  }
>  EXPORT_SYMBOL_GPL(dma_vmap_noncontiguous);
>  
>  void dma_vunmap_noncontiguous(struct device *dev, void *vaddr)
>  {
> -	const struct dma_map_ops *ops = get_dma_ops(dev);
> -
> -	if (ops && ops->alloc_noncontiguous)
> -		vunmap(vaddr);
> +	if (use_dma_iommu(dev))
> +		iommu_dma_vunmap_noncontiguous(dev, vaddr);
>  }
>  EXPORT_SYMBOL_GPL(dma_vunmap_noncontiguous);
>  
>  int dma_mmap_noncontiguous(struct device *dev, struct vm_area_struct *vma,
>  		size_t size, struct sg_table *sgt)
>  {
> -	const struct dma_map_ops *ops = get_dma_ops(dev);
> -
> -	if (ops && ops->alloc_noncontiguous) {
> -		unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
> -
> -		if (vma->vm_pgoff >= count ||
> -		    vma_pages(vma) > count - vma->vm_pgoff)
> -			return -ENXIO;
> -		return vm_map_pages(vma, sgt_handle(sgt)->pages, count);
> -	}
> +	if (use_dma_iommu(dev))
> +		return iommu_dma_mmap_noncontiguous(dev, vma, size, sgt);
>  	return dma_mmap_pages(dev, vma, size, sg_page(sgt->sgl));
>  }
>  EXPORT_SYMBOL_GPL(dma_mmap_noncontiguous);

-- 
Xi Ruoyao <xry111@xry111.site>
School of Aerospace Science and Technology, Xidian University