PM: domains: Reverse the order of performance and enabling ops
authorAbel Vesa <abel.vesa@linaro.org>
Tue, 15 Nov 2022 21:25:43 +0000 (23:25 +0200)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Fri, 25 Nov 2022 18:31:54 +0000 (19:31 +0100)
The ->set_performance_state() needs to be called before ->power_on()
when a genpd is powered on, and after ->power_off() when a genpd is
powered off. Do this in order to let the provider know to which
performance state to power on the genpd, on the power on sequence, and
also to maintain the performance for that genpd until after powering off,
on power off sequence.

There is no scenario where a consumer would need its genpd enabled and
then its performance state increased. Instead, in every scenario, the
consumer needs the genpd to be enabled from the start at a specific
performance state.

And same logic applies to the powering down. No consumer would need its
genpd performance state dropped right before powering down.

Now, there are currently two vendors which use ->set_performance_state()
in their genpd providers. One of them is Tegra, but the only genpd provider
(PMC) that makes use of ->set_performance_state() doesn't implement the
->power_on() or ->power_off(), and so it will not be affected by the ops
reversal.

The other vendor that uses it is Qualcomm, in multiple genpd providers
actually (RPM, RPMh and CPR). But all Qualcomm genpd providers that make
use of ->set_performance_state() need the order between enabling ops and
the performance setting op to be reversed. And the reason for that is that
it currently translates into two different voltages in order to power on
a genpd to a specific performance state. Basically, ->power_on() switches
to the minimum (enabling) voltage for that genpd, and then
->set_performance_state() sets it to the voltage level required by the
consumer.

By reversing the call order, we rely on the provider to know what to do
on each call, but most popular usecase is to cache the performance state
and postpone the voltage setting until the ->power_on() gets called.

As for the reason of still needing the ->power_on() and ->power_off() for a
provider which could get away with just having ->set_performance_state()
implemented, there are consumers that do not (nor should) provide an
opp-table. For those consumers, ->set_performance_state() will not be
called, and so they will enable the genpd to its minimum performance state
by a ->power_on() call. Same logic goes for the disabling.

Signed-off-by: Abel Vesa <abel.vesa@linaro.org>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/base/power/domain.c

index 2f0787d..b46aa49 100644 (file)
@@ -939,8 +939,8 @@ static int genpd_runtime_suspend(struct device *dev)
                return 0;
 
        genpd_lock(genpd);
-       gpd_data->rpm_pstate = genpd_drop_performance_state(dev);
        genpd_power_off(genpd, true, 0);
+       gpd_data->rpm_pstate = genpd_drop_performance_state(dev);
        genpd_unlock(genpd);
 
        return 0;
@@ -978,9 +978,8 @@ static int genpd_runtime_resume(struct device *dev)
                goto out;
 
        genpd_lock(genpd);
+       genpd_restore_performance_state(dev, gpd_data->rpm_pstate);
        ret = genpd_power_on(genpd, 0);
-       if (!ret)
-               genpd_restore_performance_state(dev, gpd_data->rpm_pstate);
        genpd_unlock(genpd);
 
        if (ret)
@@ -1018,8 +1017,8 @@ err_stop:
 err_poweroff:
        if (!pm_runtime_is_irq_safe(dev) || genpd_is_irq_safe(genpd)) {
                genpd_lock(genpd);
-               gpd_data->rpm_pstate = genpd_drop_performance_state(dev);
                genpd_power_off(genpd, true, 0);
+               gpd_data->rpm_pstate = genpd_drop_performance_state(dev);
                genpd_unlock(genpd);
        }
 
@@ -2707,17 +2706,6 @@ static int __genpd_dev_pm_attach(struct device *dev, struct device *base_dev,
        dev->pm_domain->detach = genpd_dev_pm_detach;
        dev->pm_domain->sync = genpd_dev_pm_sync;
 
-       if (power_on) {
-               genpd_lock(pd);
-               ret = genpd_power_on(pd, 0);
-               genpd_unlock(pd);
-       }
-
-       if (ret) {
-               genpd_remove_device(pd, dev);
-               return -EPROBE_DEFER;
-       }
-
        /* Set the default performance state */
        pstate = of_get_required_opp_performance_state(dev->of_node, index);
        if (pstate < 0 && pstate != -ENODEV && pstate != -EOPNOTSUPP) {
@@ -2729,6 +2717,24 @@ static int __genpd_dev_pm_attach(struct device *dev, struct device *base_dev,
                        goto err;
                dev_gpd_data(dev)->default_pstate = pstate;
        }
+
+       if (power_on) {
+               genpd_lock(pd);
+               ret = genpd_power_on(pd, 0);
+               genpd_unlock(pd);
+       }
+
+       if (ret) {
+               /* Drop the default performance state */
+               if (dev_gpd_data(dev)->default_pstate) {
+                       dev_pm_genpd_set_performance_state(dev, 0);
+                       dev_gpd_data(dev)->default_pstate = 0;
+               }
+
+               genpd_remove_device(pd, dev);
+               return -EPROBE_DEFER;
+       }
+
        return 1;
 
 err: