On Mon, Oct 05, 2015 at 09:49:38PM -0500, Rob Herring wrote:

On Mon, Oct 5, 2015 at 6:52 PM, Frank Rowand [off-list ref] wrote:

quoted

The purpose of the *_sync initcall levels is to allow the corresponding
initcall level to use multiple threads of execution instead of a single
thread.  The *_sync level provides a location for a wait for all of the
threads at the corresponding init level to complete.  This is better
explained in the git log:

The things I was worried about like clk and regulator disabling are
actually late_initcall_sync, so maybe late_initcall is fine here after
all.

However, looking at other *_initcall_sync users, I'm not so sure they
are doing the same abuse.

They're trying to run at the point where we've completed deferred
probe processing - in other words, at the sync point.  What they really
want is an additional callback at the point where the syncs have
actually happened.