On Mon, Jul 21, 2014 at 03:43:08PM -0600, Stephen Warren wrote:

On 07/16/2014 02:27 AM, Peter De Schrijver wrote:

quoted

On Wed, Jul 16, 2014 at 09:19:33AM +0200, Thierry Reding wrote:

quoted

* PGP Signed by an unknown key

On Tue, Jul 15, 2014 at 06:24:32PM +0300, Peter De Schrijver wrote:
[...]

quoted

diff --git a/drivers/clk/tegra/clk.c b/drivers/clk/tegra/clk.c
index d081732..65cde4e 100644
--- a/drivers/clk/tegra/clk.c
+++ b/drivers/clk/tegra/clk.c

@@ -290,10 +290,13 @@ struct clk ** __init tegra_lookup_dt_id(int clk_id,
 
 tegra_clk_apply_init_table_func tegra_clk_apply_init_table;
 
-void __init tegra_clocks_apply_init_table(void)
+static int __init tegra_clocks_apply_init_table(void)
 {
 	if (!tegra_clk_apply_init_table)
-		return;
+		return 0;

Shouldn't this be an error? Or perhaps WARN()? To make sure this gets

An arch_initcall will be called for every ARM platform I think? In case
this gets called on a non-Tegra platform, tegra_clk_apply_init_table will not
be set and therefore a silent return 0; seems the most appropriate thing to do
to me?

This is one reason that doing all the initialization from separate
initcalls sucks. Much better to have a single top-level initialization
function that calls exactly what is needed, only what is needed, and
only runs on the correct SoCs.

But failing that, I guess you need to say something like
of_is_compatible(root node, "nvidia Tegra"), but of course the
definition of "nvidia Tegra" is an ever-growing list of possible values
that needs to be used from each separate initcall...

FWIW, we have soc_is_tegra() now in include/soc/tegra/common.h which is
meant to be used for exactly this purpose. I agree that it isn't optimal
but it's pretty good. It should be easy to refactor this to make it
callable from a top-level initialization function when a decision
regarding that has been made.

Thierry