On 06/26/14 03:52, Viresh Kumar wrote:

On 26 June 2014 00:32, Stephen Boyd [off-list ref] wrote:

quoted

I don't think this driver should be using regulator_get_optional() (Mark
B. please correct me if I'm wrong). I doubt a supply is actually
optional for CPUs, just some DTs aren't specifying them. In those cases,
the regulator core will insert a dummy supply and the code will work
without having to check for probe defer and error pointers.

Hi Stephen,

Thanks for your comments.

Leaving the above one, I have tried to fix all you mentioned. And it surely
looks much better now.

I would like to wait for a day or two before sending V2, as people might
be reviewing it and the above issue is still wide open..

But in case you wanna test it (completely changed I must say, but
for good), its here:

git://git.linaro.org/people/viresh.kumar/linux.git cpufreq/cpu0-krait-v2

I gave it a spin. It looks mostly good except for the infinite loop:

diff --git a/drivers/cpufreq/cpufreq-cpu0.c b/drivers/cpufreq/cpufreq-cpu0.c
index b7ee67c4d1c0..6744321ae33d 100644
--- a/drivers/cpufreq/cpufreq-cpu0.c
+++ b/drivers/cpufreq/cpufreq-cpu0.c

@@ -138,8 +138,10 @@ try_again:
                }
 
                /* Try with "cpu-supply" */
-               if (reg == reg_cpu0)
+               if (reg == reg_cpu0) {
+                       reg = reg_cpu;
                        goto try_again;
+               }
 
                dev_warn(cpu_dev, "failed to get cpu%d regulator: %ld\n",
                         cpu, PTR_ERR(cpu_reg));

and I think we just want reg_cpu to be "cpu", not "cpu-supply" because I
think the regulator core adds in the "-supply" part already.

After fixing that I can get cpufreq going. I'm currently working on
populating the OPPs at runtime without relying on DT. So eventually I'll
need this patch:

diff --git a/drivers/cpufreq/cpufreq-cpu0.c b/drivers/cpufreq/cpufreq-cpu0.c
index b7ee67c4d1c0..6744321ae33d 100644
--- a/drivers/cpufreq/cpufreq-cpu0.c
+++ b/drivers/cpufreq/cpufreq-cpu0.c

@@ -239,11 +241,6 @@ static int cpu0_cpufreq_init(struct cpufreq_policy *policy)
        }
 
        ret = of_init_opp_table(cpu_dev);
-       if (ret) {
-               dev_err(cpu_dev, "failed to init OPP table: %d\n", ret);
-               goto out_put_node;
-       }
-
        ret = dev_pm_opp_init_cpufreq_table(cpu_dev, &freq_table);
        if (ret) {
                dev_err(cpu_dev, "failed to init cpufreq table: %d\n", ret);

which I hope is ok.

Finally, checking for equivalent pointers from clk_get() will work now,
but it isn't future-proof if/when the clock framework starts returning
dynamically allocated clock pointers for each clk_get() invocation.
Maybe we need a function in the common clock framework that tells us if
the clocks are the same either via DT or by taking two clock pointers?

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation