x86: fix pci-gart failure handling

blk_rq_map_sg doesn't initialize sg->dma_address/length to zero
anymore. Some low level drivers reuse sg lists without initializing so
IOMMUs might get non-zero dma_address/length. If map_sg fails, we need
pass the number of the mapped entries to gart_unmap_sg.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

arch/x86/kernel/pci-gart_64.c

@@ -435,7 +435,7 @@ static int gart_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 
 error:
 	flush_gart();
-	gart_unmap_sg(dev, sg, nents, dir);
+	gart_unmap_sg(dev, sg, out, dir);
 	/* When it was forced or merged try again in a dumb way */
 	if (force_iommu || iommu_merge) {
 		out = dma_map_sg_nonforce(dev, sg, nents, dir);