On 31.07.2015 08:16, Jisheng Zhang wrote:

On Thu, 30 Jul 2015 14:35:51 +0200
Antoine Tenart [off-list ref] wrote:

quoted

The BG2Q SoC uses cpufreq-dt for cpufreq. Register a platform device for
this.

Signed-off-by: Antoine Tenart <redacted>
---

[...]

quoted

diff --git a/arch/arm/mach-berlin/berlin.c b/arch/arm/mach-berlin/berlin.c
index ac181c6797ee..0de906374994 100644
--- a/arch/arm/mach-berlin/berlin.c
+++ b/arch/arm/mach-berlin/berlin.c

@@ -15,9 +15,21 @@

[...]

quoted

+static void __init berlin_init_late(void)
+{
+	if (of_find_matching_node(NULL, berlin2q_match))
+		platform_device_register_simple("cpufreq-dt", -1, NULL, 0);

I'm not sure is it acceptable to always register cpufreq-dt platform device.
In cpufreq-dt driver, the probe will exit if there's no cpuclk node.

Jisheng,

registering the _device_ without checking for any requirements of
the _driver_ is usual practice. It it the driver's job to check
all of them and report an error or warning. So, I am fine with
adding the device unconditionally.

Using the .init_late() callback seems reasonable - but can you
evaluate if we can have cpufreq-dt also for BG2/BG2CD? I'd also
accept a single, default operating-point.

BTW, there is one true point about Jisheng's comment: try to order
your patches logically correct to avoid unnecessary warnings, i.e.

1/3 add cpuclk
2/3 add cpufreq-dt properties
3/3 register cpufreq-dt device

If you consider a git-bisect run, you can land in between any of the
three patches but they are always in the above order - so ordering is
important.

Sebastian

quoted

+}
+
  static const char * const berlin_dt_compat[] = {
  	"marvell,berlin",
  	NULL,

@@ -25,6 +37,7 @@ static const char * const berlin_dt_compat[] = {

  DT_MACHINE_START(BERLIN_DT, "Marvell Berlin")
  	.dt_compat	= berlin_dt_compat,
+	.init_late	= berlin_init_late,
  	/*
  	 * with DT probing for L2CCs, berlin_init_machine can be removed.
  	 * Note: 88DE3005 (Armada 1500-mini) uses pl310 l2cc