Commit e4789198 authored Jun 13, 2014 by Vikram Mulukutla

qcom: clock-cpu-8994: Don't set CPU clocks to HANDOFF_ENABLED_CLK



Setting the cluster clock's refcount to 1 during handoff will cause
the clock framework's late_initcall_sync logic to decrement
the refcount to zero - if all the CPUs in the cluster are either
power collapsed or hotplugged out before late_initcall_sync.

Since the cluster clock muxes cannot be switched when the cluster
is in power collapse, it is wrong for a context outside of the
hotplug notifiers or the low power mode driver to decrement
the refcount on the cluster clocks to zero; this will cause a
mux switch without any guarantee that the cluster is not power
collapsed at that point.

Therefore, always return HANDOFF_DISABLED_CLK for the cluster
clocks. The probe() function will increment the cluster clock
refcount for each online CPU either way.

Change-Id: I615e44a006d249eb8f0f548fc7ebca9f293f159c
Signed-off-by: Vikram Mulukutla <markivx@codeaurora.org>

parent b5a909ab

drivers/clk/qcom/clock-cpu-8994.c

+0 −12

Original line number	Diff line number	Diff line
		@@ -591,22 +591,10 @@ static inline struct cpu_clk_8994 to_cpu_clk_8994(struct clk c)
		return container_of(c, struct cpu_clk_8994, c);
		}

		static struct clk *logical_cpu_to_clk(int cpu);

		static enum handoff cpu_clk_8994_handoff(struct clk *c)
		{
		int cpu;

		c->rate = clk_get_rate(c->parent);

		/*
		* Don't unnecessarily turn on the parents for an offline CPU and
		* then have them turned off at late init.
		*/
		for_each_online_cpu(cpu) {
		if (logical_cpu_to_clk(cpu) == c)
		return HANDOFF_ENABLED_CLK;
		}
		return HANDOFF_DISABLED_CLK;
		}