On Sat, Mar 22, 2014 at 5:01 PM, Arnd Bergmann [off-list ref] wrote:

On Saturday 22 March 2014, Rob Herring wrote:

quoted

I think this series stands on its own. It is largely refactoring
existing code and supporting existing command line options (arm64 just
changes from earlyprintk= to earlycon=). There will be cases where
changing the kernel command line is the only way to setup the
earlycon. Also, I think we would still want the kernel command line to
control whether or not we enable the earlycon (i.e. earlycon=dt).
There's not a standard way for how bootloaders would decide to set
"linux,stdout-path" or not.

The DT support won't be all that quick to implement. The challenge
with the DT parsing is we have to do it on the flattened DT. I have
something functioning, but it doesn't do any address translation which
is the hard part.

Ah, I hadn't realized that we call parse_early_param() twice, both
in start_kernel() and before that in setup_arch(). You are right
that the first one of these can only look at the flat device tree
at the moment. I wonder if that's out of necessity or just coincidence
though.

There is relatively little going on between the first parse_early_param()
and the call to unflatten_device_tree(), so it may be possible to move
the former down, or the latter up.

        parse_early_param();

        sort(&meminfo.bank, meminfo.nr_banks, sizeof(meminfo.bank[0]), meminfo_cmp, NULL);

        early_paging_init(mdesc, lookup_processor_type(read_cpuid_id()));
        setup_dma_zone(mdesc);
        sanity_check_meminfo();
        arm_memblock_init(&meminfo, mdesc);

        paging_init(mdesc);
        request_standard_resources(mdesc);

        if (mdesc->restart)
                arm_pm_restart = mdesc->restart;

        unflatten_device_tree();

On powerpc, unflatten_device_tree is called way before paging_init, so I assume
it has to be possible, but it might end up being harder to do than what you
have in mind with parsing the flat device tree.

IIRC, the unflattening cannot be done before memblock is up which is
in arm_memblock_init. I believe that will get done earlier similar to
PPC with Laura's meminfo removal series. However, in interest of
getting output enabled as early as possible, the earliest point is
always going to be with a flat tree. So I think we do want to be able
to parse the flat tree for this. There's code in u-boot to do the
address translation on FDT we can steal. Really, i'd like to put that
code in libfdt, but there's a licensing problem with the u-boot code
being GPL and libfdt being BSD.

Rob