Re: [PATCH] hw/arm/virt: Second uart for normal-world

[Daniel Thompson]

On Mon, Dec 09, 2019 at 05:10:30PM +0000, Peter Maydell wrote:
> On Mon, 9 Dec 2019 at 17:08, Daniel Thompson <address@hidden> wrote:
> > I don't object to making it command line dependant (it is certainly
> > lower risk) but, out of interest, has using /aliases to force the
> > kernel to enumerate the serial nodes in the existing order been ruled
> > out for any reason.
> 
> No, I don't think anybody's investigated that (I wasn't aware
> that you could do something like that). Bear in mind that the
> kernel is not the only consumer of the DT, though -- you need
> to use a mechanism that all DT consumers will handle correctly.

The syntax for /aliases is standardized (in the DT documentation) but
AFAIK the exact semantic meaning of an alias relies somewhat on idiom.
It is true that the DT binding documentation for some serial drivers
does include details of /aliases but sadly PL011 is not amoung them.

I took a fairly detailed look at FreeBSD. I don't think /aliases is
used to control enumeration order but that appears to be because
aliases are handled in a different way to Linux. For example 
FreeBSD allows a custom console to be selected using FDT syntax
(hw.fdt.console=serial0 or hw.fdt.console=/path/to/fdt-uart ) which
means the Linux-like approach (such as console=ttyAMA0) need not be
used.

In summary I think that support for /aliases can and should be added
since it the best way to help DT systems figure out how to match qemu
uart numbering to its own naming.

However I agree we still need a way to create systems with only a
single UART even if I have not yet been able to come up with a test
case that proves /aliases is insufficient ;-)


Daniel.