Re: [PATCH 2/3] clk: WARN_ON about to disable a critical clock

From: Dirk Behme
Date: Mon Jul 03 2017 - 11:25:45 EST

On 03.07.2017 16:25, Lee Jones wrote:
On Mon, 03 Jul 2017, Dirk Behme wrote:

On 03.07.2017 13:53, Lee Jones wrote:
On Tue, 27 Jun 2017, Dirk Behme wrote:

On 11.02.2016 01:43, Michael Turquette wrote:
Quoting Lee Jones (2016-01-18 06:28:50)
Signed-off-by: Lee Jones <lee.jones@xxxxxxxxxx>

Looks good to me.


drivers/clk/clk.c | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index 835cb85..178b364 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -575,6 +575,9 @@ static void clk_core_unprepare(struct clk_core *core)
if (WARN_ON(core->prepare_count == 0))
+ if (WARN_ON(core->prepare_count == 1 && core->flags & CLK_IS_CRITICAL))
+ return;
if (--core->prepare_count > 0)
@@ -680,6 +683,9 @@ static void clk_core_disable(struct clk_core *core)
if (WARN_ON(core->enable_count == 0))
+ if (WARN_ON(core->enable_count == 1 && core->flags & CLK_IS_CRITICAL))
+ return;
if (--core->enable_count > 0)

I have a question regarding this patch, which is mainline meanwhile [1]:

Having the following clock configuration:

|--> child clk '1' (crit)
clk source --> parent clk 'A' (crit) -->|
|--> child clk '2'

Clock '2' might be used, or not. It might be disabled or not. It doesn't
matter. Clock '1' is not allowed to be disabled. Therefore its marked as

Parent clock 'A' is marked as critical because its not allowed to be
disabled, even if the enable_count of all child clocks is 0. To avoid that
by disabling parent clock 'A' the child clock '1' is disabled, too, whats
not allowed as its marked as critical.

Now, child clock '2' is used and enabled & disabled continuously by a (SPI)
driver. What is ok. But:

Disabling child clock '2' results in the attempt to disable parent clock
'A', too, which has correct enable_count 1 (from enabling the child '2').
What results

a) in the WARN_ON output


b) enable_count of 'A' never decreases to 0. Being off by one after the

It sounds like both is wrong for a configuration like above.

Clock A still has one user, Clock 1.

Why is that wrong?

Because clock 1 is not a (Linux kernel clock framework) used clock. Its
enable count is 0. So from Linux kernel (clock framework) point of view
clock 1 is unused.

All critical clocks are 'used'. That's the point of critical clocks.

Could you translate 'used' to enable_count? Whats valid enable_count numbers for critical clocks?

Best regards


The increase/decrease of enable count of parent clock A is only driven by
the Linux kernel usage of clock 2.

Opinions or proposal how to fix/change this?

Best regards