PCI: dm: Ignore 64-bit memory regions if CONFIG_SYS_PCI_64BIT not set

Currently, qemu_arm_defconfig and qemu_arm64_defconfig only work with
the 'highmem=off' parameter passed to QEMU's virt machine. The reason is
that when 'highmem' is not disabled, QEMU appends 64-bit a memory
resource to the PCI controller's regions property in DT in addition to
the 32-bit PCI memory window in low memory. And the current DT parsing
code picks the last (thus the 64-bit one) memory resource, whose address
eventually gets silently truncated to 32 bits because
CONFIG_SYS_PCI_64BIT is not set, which obviously causes PCI to break.

Avoid this problem by ignoring memory regions whose addresses are above
the 32-bit boundary when CONFIG_SYS_PCI_64BIT is not set.

Signed-off-by: Tuomas Tynkkynen <tuomas.tynkkynen@iki.fi>
Reviewed-by: Simon Glass <sjg@chromium.org>

drivers/pci/pci-uclass.c

@@ -860,6 +860,13 @@ static int decode_regions(struct pci_controller *hose, ofnode parent_node,
 		} else {
 			continue;
 		}
+
+		if (!IS_ENABLED(CONFIG_SYS_PCI_64BIT) &&
+		    type == PCI_REGION_MEM && upper_32_bits(pci_addr)) {
+			debug(" - beyond the 32-bit boundary, ignoring\n");
+			continue;
+		}
+
 		pos = -1;
 		for (i = 0; i < hose->region_count; i++) {
 			if (hose->regions[i].flags == type)