DMA write fails when dma_alloc_coherent is replaced with dma_map_single

Question

I am experimenting the DMA flow with and without SMMU on ARM. Our custom application requires kernel address, physical address and bus address.

With SMMU enabled, I managed to get the kernel address and bus address using dma_alloc_coherent(), but I could not get the physical address via virt_to_phys(). That made me replace dma_alloc_coherent() with dma_map_single as follows:

// Get kernel address
res->KernelAddress = (u64)kzalloc( size , GFP_KERNEL | GFP_DMA);

// Get bus address
res->BusAddress = (u64)dma_map_single(&DevExt->pdev->dev, &res->KernelAddress, size, PCI_DMA_BIDIRECTIONAL);

// Get physical address
res->PhysicalAddress = (u64)virt_to_phys( (void*)res->KernelAddress );

dma_sync_single_for_cpu(&DevExt->pdev->dev, res->BusAddress, size, PCI_DMA_BIDIRECTIONAL);

// Map physical address to virtual address in user space
if (remap_pfn_range(vma, vma->kern_addr, vma->phys_addr,size,vma->vm_page_prot));

There was no issue with the above flow. I managed to get 3 addresses but DMA write to the returned location failed. Are other DMA APIs required for the above flow to work?

It is better if any snippet is shared!