On Thu, 02 Apr 2015, Jassi Brar wrote:

On Wed, Apr 1, 2015 at 7:12 AM, Michael Turquette [off-list ref] wrote:

quoted

Quoting Jassi Brar (2015-03-02 02:28:44)

quoted

On 2 March 2015 at 15:48, Lee Jones [off-list ref] wrote:

quoted

On Mon, 02 Mar 2015, Jassi Brar wrote:

quoted

On Mon, Mar 2, 2015 at 2:06 PM, Lee Jones [off-list ref] wrote:

quoted

On Sat, 28 Feb 2015, Jassi Brar wrote:

quoted

On 28 February 2015 at 02:44, Lee Jones [off-list ref] wrote:

quoted

Lots of platforms contain clocks which if turned off would prove fatal.
The only way to recover from these catastrophic failures is to restart
the board(s).  Now, when a clock is registered with the framework it is
compared against a list of provided always-on clock names which must be
kept ungated.  If it matches, we enable the existing CLK_IGNORE_UNUSED
flag, which will prevent the common clk framework from attempting to
gate it during the clk_disable_unused() procedure.

If a clock is critical on a certain board, it could be got+enabled
during early boot so there is always a user.

I tried this.  There was push-back from the DT maintainers.

  http://lists.infradead.org/pipermail/linux-arm-kernel/2015-February/324417.html

Thanks, I wasn't aware of the history.

quoted

quoted

To be able to do that from DT, maybe add a new, say, CLK_ALWAYS_ON
flag could be made to initialize the clock with one phantom user
already. Or just reuse the CLK_IGNORE_UNUSED?

How is that different to what this set is doing?

The phantom user - that's there but none can see it.

How about?

+       of_property_for_each_string(np, "clock-always-on", prop, clkname) {
+               clk = __clk_lookup(clkname);
+               if (!clk)
+                       continue;
+
+               clk->core->enable_count = 1;
+               clk->core->prepare_count = 1;
+       }

This is only fractionally different from the current implementation.

I believe the current way it slightly nicer, as we don't have to fake
the user count.

Well... the user is indeed there, isn't it? It's just not known to
Linux. So 'fake' isn't most applicable here.
Otherwise you might have to stub out some existing and future
functions for CLK_IGNORE_UNUSED. And how do we explain to userspace
which would see power drawn but no user of the clock?

Jassi,

This is broken. What if the parent of this clock has
{enable,prepare}_count of zero? The way we propagate these refcounts up
the tree would fall over.

Yeah it needs to be done at higher level,
- clk->core->enable_count = 1;
- clk->core->prepare_count = 1;
+ clk_prepare_enable(clk);

FYI: https://lkml.org/lkml/2015/4/1/267

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org ? Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog