[PATCH 1/5] soc: qcom: Add suspend to idle support

[Ulf Hansson]

On 9 June 2016 at 20:09, Andy Gross <andy.gross at linaro.org> wrote:
> On Thu, Jun 09, 2016 at 09:39:34AM +0200, Ulf Hansson wrote:
>> + Daniel
>>
>> On 19 May 2016 at 07:00, Andy Gross <andy.gross at linaro.org> wrote:
>> > This patch adds suspend to idle support for Qualcomm processors.  While
>> > suspend to memory will be a valid state, there won't be any special
>> > handling or power savings over the suspend to idle.
>> >
>> > Signed-off-by: Andy Gross <andy.gross at linaro.org>
>> > ---
>> >  drivers/soc/qcom/Makefile  |  1 +
>> >  drivers/soc/qcom/suspend.c | 77 ++++++++++++++++++++++++++++++++++++++++++++++
>> >  2 files changed, 78 insertions(+)
>> >  create mode 100644 drivers/soc/qcom/suspend.c
>> >
>> > diff --git a/drivers/soc/qcom/Makefile b/drivers/soc/qcom/Makefile
>> > index fdd664e..7c479d3 100644
>> > --- a/drivers/soc/qcom/Makefile
>> > +++ b/drivers/soc/qcom/Makefile
>> > @@ -7,3 +7,4 @@ obj-$(CONFIG_QCOM_SMEM_STATE) += smem_state.o
>> >  obj-$(CONFIG_QCOM_SMP2P)       += smp2p.o
>> >  obj-$(CONFIG_QCOM_SMSM)        += smsm.o
>> >  obj-$(CONFIG_QCOM_WCNSS_CTRL) += wcnss_ctrl.o
>> > +obj-$(CONFIG_SUSPEND)  +=      suspend.o
>> > diff --git a/drivers/soc/qcom/suspend.c b/drivers/soc/qcom/suspend.c
>> > new file mode 100644
>> > index 0000000..7d3f2dd
>> > --- /dev/null
>> > +++ b/drivers/soc/qcom/suspend.c
>> > @@ -0,0 +1,77 @@
>> > +/*
>> > + * (C) Copyright 2016 Linaro Ltd.
>> > + *
>> > + * This program is free software; you can redistribute it and/or
>> > + * modify it under the terms of the GNU General Public License
>> > + * as published by the Free Software Foundation; version 2
>> > + * of the License.
>> > + */
>> > +#include <linux/module.h>
>> > +#include <linux/of.h>
>> > +#include <linux/cpuidle.h>
>> > +#include <linux/suspend.h>
>> > +
>> > +
>> > +static void qcom_pm_enter_freeze(struct cpuidle_device *dev,
>> > +       struct cpuidle_driver *drv,
>> > +       int index)
>> > +{
>> > +       drv->states[index].enter(dev, drv, index);
>> > +}
>> > +
>> > +static const struct of_device_id qcom_idle_state_match[] = {
>> > +       { .compatible = "qcom,idle-state-spc", },
>> > +       { },
>> > +};
>> > +
>> > +static const struct platform_suspend_ops qcom_suspend_ops = {
>> > +       .valid          = suspend_valid_only_mem,
>> > +};
>> > +
>> > +static int __init qcom_pm_init(void)
>> > +{
>> > +       struct cpuidle_device *cpu_dev;
>> > +       struct cpuidle_driver *cpu_drv;
>> > +       int state_count;
>> > +       struct device_node *state_np, *cpu_np;
>> > +       const struct of_device_id *match;
>> > +       int i;
>> > +
>> > +       /* configure CPU enter_freeze if applicable */
>> > +       for_each_present_cpu(i) {
>> > +               cpu_np = of_get_cpu_node(i, NULL);
>> > +               cpu_dev = per_cpu_ptr(cpuidle_devices, i);
>> > +               cpu_drv = cpuidle_get_cpu_driver(cpu_dev);
>> > +
>> > +               if (!cpu_dev || !cpu_drv) {
>> > +                       of_node_put(cpu_np);
>> > +                       return -EPROBE_DEFER;
>>
>> This isn't a driver, so returning -EPROBE_DEFER doesn't make sense.
>
> Right.  It's not like it'll get retried.
>
>> > +               }
>> > +
>> > +               state_count = 0;
>> > +               state_np = of_parse_phandle(cpu_np, "cpu-idle-states",
>> > +                                                 state_count);
>> > +
>> > +               while (state_np) {
>> > +                       match = of_match_node(qcom_idle_state_match,
>> > +                                             state_np);
>> > +
>> > +                       state_count++;
>> > +                       if (match)
>> > +                               cpu_drv->states[state_count].enter_freeze =
>> > +                                               &qcom_pm_enter_freeze;
>> > +                       of_node_put(state_np);
>> > +
>> > +                       state_np = of_parse_phandle(cpu_np, "cpu-idle-states",
>> > +                                                   state_count);
>> > +               }
>> > +
>> > +               of_node_put(cpu_np);
>> > +       }
>> > +
>> > +       suspend_set_ops(&qcom_suspend_ops);
>>
>> I don't think this will work!
>>
>> When building a multi defconfig for ARM, you might overwrite the
>> suspend_ops (there's only one set) as here you don't know that it's
>> actually the QCOM platform that is running, right!?
>>
>> Perhaps this code actually belongs closer to the cpuidle driver?
>
> Hmmmm, I might have to get creative.  I originally had a DT entry for the pm,
> but that doesn't make sense as this is purely a software construct.  The db410c
> uses the arm cpuidle driver so I can't really hook it in there.  I'll have to
> come up with something else.

I did a little research.

I think you should be able to use the struct cpuidle_ops->init()
callback. This is being invoked when the arm cpuidle driver is
initialized.

arm_idle_init()
  ->arm_cpuidle_init()
     ->cpuidle_ops->init()

In the QCOM case, it's the spm driver that registers these cpuidle_ops
(drivers/soc/qcom/spm.c). If you fold in the code from $subject patch
in qcom_cpuidle_init(), that should work I think.

Kind regards
Uffe