ns16550a reg-shift incorrect for qemu-system-riscv64

Description of problem

Missing reg-shift 0 on the ns16550n in qemu-system-riscv64 creates an impossible assumption case.

Steps to reproduce

1. qemu-system-riscv64 -M virt,dumpdtb=dtb

2. dtc dtb | less

            serial@10000000 {
                    interrupts = <0x0a>;
                    interrupt-parent = <0x03>;
                    clock-frequency = "\08@";
                    reg = <0x00 0x10000000 0x00 0x100>;
                    compatible = "ns16550a";
            };

Generally, ns16550a has a default reg-shift of 0 on x86,x86_64 for compatibility reasons. All other architectures have an assumed reg-shift of 2 (or having the reg-shift assumption overridden by fdt providing a reg-shift property)

Beyond the above, anything non-standard is assumed to be specified by the "reg-shift" property fdt.

qemu-system-riscv64 seems to "assume" a reg-shift of 0. Other riscv64 devices don't supply "reg-shift" (SiFive Unmatched) and "assume" 2. The above means driver writers don't actually know what to "assume" on riscv64 ns16550a when no reg-shift is present.

Essentially, qemu-system-riscv64 needs to do one of the following:

• If serial ns16550a with a uart reg-shift of 0 is intentional, qemu needs to advertise the deviance via "reg-shift 0"
• If serial ns16550a with a uart reg-shift of 0 is unintentional, it needs updated to 2 so drivers can assume 2 on riscv64.