Re: [PATCH 2/2] swiotlb-xen: override common mmap and get_sgtable dma ops

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 6/16/21 7:42 AM, Roman Skakun wrote:
> This commit is dedicated to fix incorrect conversion from
> cpu_addr to page address in cases when we get virtual
> address which allocated through xen_swiotlb_alloc_coherent()
> and can be mapped in the vmalloc range.
> As the result, virt_to_page() cannot convert this address
> properly and return incorrect page address.
>
> Need to detect such cases and obtains the page address using
> vmalloc_to_page() instead.
>
> The reference code for mmap() and get_sgtable() was copied
> from kernel/dma/ops_helpers.c and modified to provide
> additional detections as described above.
>
> In order to simplify code there was added a new
> dma_cpu_addr_to_page() helper.
>
> Signed-off-by: Roman Skakun <roman_skakun@epam.com>
> Reviewed-by: Andrii Anisov <andrii_anisov@epam.com>
> ---
>  drivers/xen/swiotlb-xen.c | 42 +++++++++++++++++++++++++++++++--------
>  1 file changed, 34 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
> index 90bc5fc321bc..9331a8500547 100644
> --- a/drivers/xen/swiotlb-xen.c
> +++ b/drivers/xen/swiotlb-xen.c
> @@ -118,6 +118,14 @@ static int is_xen_swiotlb_buffer(struct device *dev, dma_addr_t dma_addr)
>  	return 0;
>  }
>  
> +static struct page *cpu_addr_to_page(void *cpu_addr)
> +{
> +	if (is_vmalloc_addr(cpu_addr))
> +		return vmalloc_to_page(cpu_addr);
> +	else
> +		return virt_to_page(cpu_addr);
> +}
> +
>  static int
>  xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs)
>  {
> @@ -337,7 +345,7 @@ xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
>  	int order = get_order(size);
>  	phys_addr_t phys;
>  	u64 dma_mask = DMA_BIT_MASK(32);
> -	struct page *page;
> +	struct page *page = cpu_addr_to_page(vaddr);
>  
>  	if (hwdev && hwdev->coherent_dma_mask)
>  		dma_mask = hwdev->coherent_dma_mask;
> @@ -349,11 +357,6 @@ xen_swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
>  	/* Convert the size to actually allocated. */
>  	size = 1UL << (order + XEN_PAGE_SHIFT);
>  
> -	if (is_vmalloc_addr(vaddr))
> -		page = vmalloc_to_page(vaddr);
> -	else
> -		page = virt_to_page(vaddr);
> -
>  	if (!WARN_ON((dev_addr + size - 1 > dma_mask) ||
>  		     range_straddles_page_boundary(phys, size)) &&
>  	    TestClearPageXenRemapped(page))
> @@ -573,7 +576,23 @@ xen_swiotlb_dma_mmap(struct device *dev, struct vm_area_struct *vma,
>  		     void *cpu_addr, dma_addr_t dma_addr, size_t size,
>  		     unsigned long attrs)
>  {
> -	return dma_common_mmap(dev, vma, cpu_addr, dma_addr, size, attrs);
> +	unsigned long user_count = vma_pages(vma);
> +	unsigned long count = PAGE_ALIGN(size) >> PAGE_SHIFT;
> +	unsigned long off = vma->vm_pgoff;
> +	struct page *page = cpu_addr_to_page(cpu_addr);
> +	int ret;
> +
> +	vma->vm_page_prot = dma_pgprot(dev, vma->vm_page_prot, attrs);
> +
> +	if (dma_mmap_from_dev_coherent(dev, vma, cpu_addr, size, &ret))
> +		return ret;
> +
> +	if (off >= count || user_count > count - off)
> +		return -ENXIO;
> +
> +	return remap_pfn_range(vma, vma->vm_start,
> +			page_to_pfn(page) + vma->vm_pgoff,
> +			user_count << PAGE_SHIFT, vma->vm_page_prot);
>  }


I wonder now whether we could avoid code duplication between here and dma_common_mmap()/dma_common_get_sgtable() and use your helper there.


Christoph, would that work?  I.e. something like


diff --git a/kernel/dma/ops_helpers.c b/kernel/dma/ops_helpers.c
index 910ae69cae77..43411c2fa47b 100644
--- a/kernel/dma/ops_helpers.c
+++ b/kernel/dma/ops_helpers.c
@@ -12,7 +12,7 @@ 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)
 {
-       struct page *page = virt_to_page(cpu_addr);
+       struct page *page = cpu_addr_to_page(cpu_addr);
        int ret;
 
        ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
@@ -43,7 +43,7 @@ int dma_common_mmap(struct device *dev, struct vm_area_struct *vma,
                return -ENXIO;
 
        return remap_pfn_range(vma, vma->vm_start,
-                       page_to_pfn(virt_to_page(cpu_addr)) + vma->vm_pgoff,
+                       page_to_pfn(cpu_addr_to_page(cpu_addr)) + vma->vm_pgoff,
                        user_count << PAGE_SHIFT, vma->vm_page_prot);
 #else
        return -ENXIO;


-boris


>  
>  /*
> @@ -585,7 +604,14 @@ xen_swiotlb_get_sgtable(struct device *dev, struct sg_table *sgt,
>  			void *cpu_addr, dma_addr_t handle, size_t size,
>  			unsigned long attrs)
>  {
> -	return dma_common_get_sgtable(dev, sgt, cpu_addr, handle, size, attrs);
> +	struct page *page = cpu_addr_to_page(cpu_addr);
> +	int ret;
> +
> +	ret = sg_alloc_table(sgt, 1, GFP_KERNEL);
> +	if (!ret)
> +		sg_set_page(sgt->sgl, page, PAGE_ALIGN(size), 0);
> +
> +	return ret;
>  }
>  
>  const struct dma_map_ops xen_swiotlb_dma_ops = {