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>
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) {