From: Alan Cox <alan@lxorguk.ukuu.org.uk>

Seems nobody else is checking/testing this case as it keeps getting horked.

If we have no BAR4 mapping on an SFF controller this is *NOT* an error, it
just means it isn't doing BMDMA.

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 drivers/ata/libata-sff.c |    4 ++++
 1 file changed, 4 insertions(+)

diff -puN drivers/ata/libata-sff.c~libata-sff-unbreak-non-dma-capable-controllers-again drivers/ata/libata-sff.c
--- a/drivers/ata/libata-sff.c~libata-sff-unbreak-non-dma-capable-controllers-again
+++ a/drivers/ata/libata-sff.c
@@ -573,6 +573,10 @@ int ata_pci_init_bmdma(struct ata_host *
 	struct pci_dev *pdev = to_pci_dev(gdev);
 	int i, rc;
 
+	/* No BAR4 allocation: No DMA */
+	if (pci_resource_start(pdev, 4) == 0)
+		return 0;
+
 	/* TODO: If we get no DMA mask we should fall back to PIO */
 	rc = pci_set_dma_mask(pdev, ATA_DMA_MASK);
 	if (rc)
_