ARM/dma-mappіng: don't override ->dma_coherent when set from a bus no…

…tifier

Commit ae626eb ("ARM/dma-mapping: use dma-direct unconditionally")
caused a regression on the mvebu platform, wherein devices that are
dma-coherent are marked as dma-noncoherent, because although
mvebu_hwcc_notifier() after that commit still marks then as coherent,
the arm_coherent_dma_ops() function, which is called later, overwrites
this setting, since it is being called from drivers/of/device.c with
coherency parameter determined by of_dma_is_coherent(), and the
device-trees do not declare the 'dma-coherent' property.

Fix this by defaulting never clearing the dma_coherent flag in
arm_coherent_dma_ops().

Fixes: ae626eb ("ARM/dma-mapping: use dma-direct unconditionally")
Reported-by: Marek Behún <kabel@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Tested-by: Marek Behún <kabel@kernel.org>

arch/arm/mm/dma-mapping.c

@@ -1769,8 +1769,16 @@ static void arm_teardown_iommu_dma_ops(struct device *dev) { }
 void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
 			const struct iommu_ops *iommu, bool coherent)
 {
-	dev->archdata.dma_coherent = coherent;
-	dev->dma_coherent = coherent;
+	/*
+	 * Due to legacy code that sets the ->dma_coherent flag from a bus
+	 * notifier we can't just assign coherent to the ->dma_coherent flag
+	 * here, but instead have to make sure we only set but never clear it
+	 * for now.
+	 */
+	if (coherent) {
+		dev->archdata.dma_coherent = true;
+		dev->dma_coherent = true;
+	}
 
 	/*
 	 * Don't override the dma_ops if they have already been set. Ideally