x86/dma/amd_gart: Use dma_direct_{alloc,free}()

This gains support for CMA allocations for the force_iommu case, and
cleans up the code a bit.
Tested-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Jon Mason <jdmason@kudzu.us>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Muli Ben-Yehuda <mulix@mulix.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: iommu@lists.linux-foundation.org
Link: http://lkml.kernel.org/r/20180319103826.12853-7-hch@lst.deSigned-off-by: Ingo Molnar <mingo@kernel.org>

arch/x86/kernel/amd_gart_64.c

@@ -480,29 +480,21 @@ static void *
 gart_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_addr,
 		    gfp_t flag, unsigned long attrs)
 {
-	dma_addr_t paddr;
-	unsigned long align_mask;
-	struct page *page;
-
-	if (force_iommu && dev->coherent_dma_mask > DMA_BIT_MASK(24)) {
-		flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
-		page = alloc_pages(flag | __GFP_ZERO, get_order(size));
-		if (!page)
-			return NULL;
+	void *vaddr;
 
-		align_mask = (1UL << get_order(size)) - 1;
-		paddr = dma_map_area(dev, page_to_phys(page), size,
-				     DMA_BIDIRECTIONAL, align_mask);
+	vaddr = dma_direct_alloc(dev, size, dma_addr, flag, attrs);
+	if (!vaddr ||
+	    !force_iommu || dev->coherent_dma_mask <= DMA_BIT_MASK(24))
+		return vaddr;
 
+	*dma_addr = dma_map_area(dev, virt_to_phys(vaddr), size,
+			DMA_BIDIRECTIONAL, (1UL << get_order(size)) - 1);
 	flush_gart();
-		if (paddr != bad_dma_addr) {
-			*dma_addr = paddr;
-			return page_address(page);
-		}
-		__free_pages(page, get_order(size));
-	} else
-		return dma_direct_alloc(dev, size, dma_addr, flag, attrs);
-
+	if (unlikely(*dma_addr == bad_dma_addr))
+		goto out_free;
+	return vaddr;
+out_free:
+	dma_direct_free(dev, size, vaddr, *dma_addr, attrs);
 	return NULL;
 }