Re: [PATCH v3 3/4] ARM: dma-mapping: Switch to physical address mapping callbacks

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Sep 16, 2025 at 10:32:06AM +0300, Leon Romanovsky wrote:
> +	if (!dev->dma_coherent &&
> +	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO)))
> +		__dma_page_cpu_to_dev(phys_to_page(phys), offset, size, dir);

I'd keep going and get rid of the page here too, maybe as a second
patch in this series:

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 88c2d68a69c9ee..a84d12cd0ba4a9 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -624,16 +624,14 @@ static void __arm_dma_free(struct device *dev, size_t size, void *cpu_addr,
 	kfree(buf);
 }
 
-static void dma_cache_maint_page(struct page *page, unsigned long offset,
+static void dma_cache_maint_page(phys_addr_t paddr,
 	size_t size, enum dma_data_direction dir,
 	void (*op)(const void *, size_t, int))
 {
-	unsigned long pfn;
+	unsigned long pfn = paddr / PAGE_SIZE;
+	unsigned int offset = paddr % PAGE_SIZE;
 	size_t left = size;
 
-	pfn = page_to_pfn(page) + offset / PAGE_SIZE;
-	offset %= PAGE_SIZE;
-
 	/*
 	 * A single sg entry may refer to multiple physically contiguous
 	 * pages.  But we still need to process highmem pages individually.
@@ -644,17 +642,17 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
 		size_t len = left;
 		void *vaddr;
 
-		page = pfn_to_page(pfn);
-
-		if (PageHighMem(page)) {
+		if (PhysHighMem(pfn << PAGE_SHIFT)) {
 			if (len + offset > PAGE_SIZE)
 				len = PAGE_SIZE - offset;
 
 			if (cache_is_vipt_nonaliasing()) {
-				vaddr = kmap_atomic(page);
+				vaddr = kmap_atomic_pfn(pfn);
 				op(vaddr + offset, len, dir);
 				kunmap_atomic(vaddr);
 			} else {
+				struct page *page = pfn_to_page(pfn);
+
 				vaddr = kmap_high_get(page);
 				if (vaddr) {
 					op(vaddr + offset, len, dir);
@@ -662,7 +660,7 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
 				}
 			}
 		} else {
-			vaddr = page_address(page) + offset;
+			vaddr = phys_to_virt(pfn) + offset;
 			op(vaddr, len, dir);
 		}
 		offset = 0;
@@ -676,14 +674,11 @@ static void dma_cache_maint_page(struct page *page, unsigned long offset,
  * Note: Drivers should NOT use this function directly.
  * Use the driver DMA support - see dma-mapping.h (dma_sync_*)
  */
-static void __dma_page_cpu_to_dev(struct page *page, unsigned long off,
+static void __dma_page_cpu_to_dev(phys_addr_t paddr,
 	size_t size, enum dma_data_direction dir)
 {
-	phys_addr_t paddr;
+	dma_cache_maint_page(paddr, size, dir, dmac_map_area);
 
-	dma_cache_maint_page(page, off, size, dir, dmac_map_area);
-
-	paddr = page_to_phys(page) + off;
 	if (dir == DMA_FROM_DEVICE) {
 		outer_inv_range(paddr, paddr + size);
 	} else {

> +	if (!dev->dma_coherent &&
> +	    !(attrs & (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_MMIO))) {
>  		page = phys_to_page(iommu_iova_to_phys(mapping->domain, iova));
>  		__dma_page_dev_to_cpu(page, offset, size, dir);

Same treatment here..

Looks Ok though, I didn't notice any pitfalls

Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>

Jason