ARM: OMAP4: Fix EMU clock domain always on

Commit d043d87 (ARM: OMAP2+: clockdomain: bypass clockdomain handling
when disabling unused clks) skips the decrementing of a clock-domains
use count if the clocks use count is zero. However, for OMAP4 devices
this is causing the EMU clock-domain to be stuck ON as the use count is
not getting decremented correctly.

The scenario that leads to this problem is described below ...

omap_hwmod_setup_all
--> _setup
    --> clkdm_hwmod_enable
    	--> EMU clock domain usecount = 1
    --> _enable_clocks
        --> clk_enable
	    --> trace_clk_div_div usecount = 1
            --> clkdm_hwmod_enable
    	        --> EMU clock domain usecount = 2
--> _idle
    --> _disable_clocks
	--> clk_disable
            --> trace_clk_div_div usecount = 0
            --> clkdm_hwmod_disable
                --> skips decrement of EMU clock domain usecount
		    because trace_clk_div_div is 0!
    	        --> EMU clock domain usecount = 2
    --> clkdm_hwmod_disable
    	--> EMU clock domain usecount = 1

Hence, due to the order that a clocks use count is decremented and the
clock domain is disabled, it is possible that the clock domain can have
a non-zero use count when the actual clock has a use count of 0.
Therefore, we should only bypass the clock-domain handling when both the
clock-domain and clock in the clock-domain have a use count of 0 and
warn when the clock-domain has a zero use count and the clock has a
non-zero use count.

Signed-off-by: Jon Hunter <jon-hunter@ti.com>
[paul@pwsan.com: fixed checkpatch warning]
Signed-off-by: Paul Walmsley <paul@pwsan.com>

arch/arm/mach-omap2/clockdomain.c

@@ -998,7 +998,8 @@ int clkdm_clk_disable(struct clockdomain *clkdm, struct clk *clk)
 	spin_lock_irqsave(&clkdm->lock, flags);
 
 	/* corner case: disabling unused clocks */
-	if (__clk_get_enable_count(clk) == 0)
+	if ((__clk_get_enable_count(clk) == 0) &&
+	    (atomic_read(&clkdm->usecount) == 0))
 		goto ccd_exit;
 
 	if (atomic_read(&clkdm->usecount) == 0) {