See the email thread at https://lkml.org/lkml/2011/5/17/54

Michael has a system with several UARTs at [io 0x1900-0x1947].  Even when these are correctly described as PNP0500 devices in the ACPI namespace, Linux assigns the [io 0x1000-0x1fff] range to a P2P bridge window.  This should not happen, because the ACPI description should be enough to tell us that the 0x1900-0x1947 range is already in use.

This causes a hang on Michael's system.  The workaround for now is for Michael to describe the [io 0x1900-0x1947] range as a PNP0C02 device, because we handle those motherboard devices specially -- we do reserve their resources before we try to assign PCI resources.  But this is a hack and prevents the 8250 driver from claiming the UARTs as it should.