x86/amd-iommu: Fix rounding-bug in __unmap_single

In the __unmap_single function the dma_addr is rounded down
to a page boundary before the dma pages are unmapped. The
address is later also used to flush the TLB entries for that
mapping. But without the offset into the dma page the amount
of pages to flush might be miscalculated in the TLB flushing
path. This patch fixes this bug by using the original
address to flush the TLB.

Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>

arch/x86/kernel/amd_iommu.c

@@ -1953,6 +1953,7 @@ static void __unmap_single(struct dma_ops_domain *dma_dom,
 			   size_t size,
 			   int dir)
 {
+	dma_addr_t flush_addr;
 	dma_addr_t i, start;
 	unsigned int pages;
 
@@ -1960,6 +1961,7 @@ static void __unmap_single(struct dma_ops_domain *dma_dom,
 	    (dma_addr + size > dma_dom->aperture_size))
 		return;
 
+	flush_addr = dma_addr;
 	pages = iommu_num_pages(dma_addr, size, PAGE_SIZE);
 	dma_addr &= PAGE_MASK;
 	start = dma_addr;
@@ -1974,7 +1976,7 @@ static void __unmap_single(struct dma_ops_domain *dma_dom,
 	dma_ops_free_addresses(dma_dom, dma_addr, pages);
 
 	if (amd_iommu_unmap_flush || dma_dom->need_flush) {
-		iommu_flush_pages(&dma_dom->domain, dma_addr, size);
+		iommu_flush_pages(&dma_dom->domain, flush_addr, size);
 		dma_dom->need_flush = false;
 	}
 }