PM / Domains: Add default power off governor function (v4)

Add a function deciding whether or not a given PM domain should
be powered off on the basis of the PM QoS constraints of devices
belonging to it and their PM QoS timing data.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

3 files changed, 129 insertions, 0 deletions

diff --git a/drivers/base/power/domain.c b/drivers/base/power/domain.c
index 3af9f5a71ad5..91896194e76b 100644
--- a/drivers/base/power/domain.c
+++ b/drivers/base/power/domain.c
@@ -398,6 +398,17 @@ static int pm_genpd_poweroff(struct generic_pm_domain *genpd)
         }
 
         genpd->status = GPD_STATE_POWER_OFF;
+        genpd->power_off_time = ktime_get();
+
+        /* Update PM QoS information for devices in the domain. */
+        list_for_each_entry_reverse(pdd, &genpd->dev_list, list_node) {
+                struct gpd_timing_data *td = &to_gpd_data(pdd)->td;
+
+                pm_runtime_update_max_time_suspended(pdd->dev,
+                                        td->start_latency_ns +
+                                        td->restore_state_latency_ns +
+                                        genpd->power_on_latency_ns);
+        }
 
         list_for_each_entry(link, &genpd->slave_links, slave_node) {
                 genpd_sd_counter_dec(link->master);
@@ -1487,6 +1498,7 @@ void pm_genpd_init(struct generic_pm_domain *genpd,
         genpd->resume_count = 0;
         genpd->device_count = 0;
         genpd->suspended_count = 0;
+        genpd->max_off_time_ns = -1;
         genpd->domain.ops.runtime_suspend = pm_genpd_runtime_suspend;
         genpd->domain.ops.runtime_resume = pm_genpd_runtime_resume;
         genpd->domain.ops.runtime_idle = pm_generic_runtime_idle;
diff --git a/drivers/base/power/domain_governor.c b/drivers/base/power/domain_governor.c
index 97b21c10c496..da78540e9b40 100644
--- a/drivers/base/power/domain_governor.c
+++ b/drivers/base/power/domain_governor.c
@@ -10,6 +10,7 @@
 #include <linux/kernel.h>
 #include <linux/pm_domain.h>
 #include <linux/pm_qos.h>
+#include <linux/hrtimer.h>
 
 /**
  * default_stop_ok - Default PM domain governor routine for stopping devices.
@@ -28,6 +29,115 @@ bool default_stop_ok(struct device *dev)
                 && td->break_even_ns < dev->power.max_time_suspended_ns;
 }
 
+/**
+ * default_power_down_ok - Default generic PM domain power off governor routine.
+ * @pd: PM domain to check.
+ *
+ * This routine must be executed under the PM domain's lock.
+ */
+static bool default_power_down_ok(struct dev_pm_domain *pd)
+{
+        struct generic_pm_domain *genpd = pd_to_genpd(pd);
+        struct gpd_link *link;
+        struct pm_domain_data *pdd;
+        s64 min_dev_off_time_ns;
+        s64 off_on_time_ns;
+        ktime_t time_now = ktime_get();
+
+        off_on_time_ns = genpd->power_off_latency_ns +
+                                genpd->power_on_latency_ns;
+        /*
+         * It doesn't make sense to remove power from the domain if saving
+         * the state of all devices in it and the power off/power on operations
+         * take too much time.
+         *
+         * All devices in this domain have been stopped already at this point.
+         */
+        list_for_each_entry(pdd, &genpd->dev_list, list_node) {
+                if (pdd->dev->driver)
+                        off_on_time_ns +=
+                                to_gpd_data(pdd)->td.save_state_latency_ns;
+        }
+
+        /*
+         * Check if subdomains can be off for enough time.
+         *
+         * All subdomains have been powered off already at this point.
+         */
+        list_for_each_entry(link, &genpd->master_links, master_node) {
+                struct generic_pm_domain *sd = link->slave;
+                s64 sd_max_off_ns = sd->max_off_time_ns;
+
+                if (sd_max_off_ns < 0)
+                        continue;
+
+                sd_max_off_ns -= ktime_to_ns(ktime_sub(time_now,
+                                                       sd->power_off_time));
+                /*
+                 * Check if the subdomain is allowed to be off long enough for
+                 * the current domain to turn off and on (that's how much time
+                 * it will have to wait worst case).
+                 */
+                if (sd_max_off_ns <= off_on_time_ns)
+                        return false;
+        }
+
+        /*
+         * Check if the devices in the domain can be off enough time.
+         */
+        min_dev_off_time_ns = -1;
+        list_for_each_entry(pdd, &genpd->dev_list, list_node) {
+                struct gpd_timing_data *td;
+                struct device *dev = pdd->dev;
+                s64 dev_off_time_ns;
+
+                if (!dev->driver || dev->power.max_time_suspended_ns < 0)
+                        continue;
+
+                td = &to_gpd_data(pdd)->td;
+                dev_off_time_ns = dev->power.max_time_suspended_ns -
+                        (td->start_latency_ns + td->restore_state_latency_ns +
+                                ktime_to_ns(ktime_sub(time_now,
+                                                dev->power.suspend_time)));
+                if (dev_off_time_ns <= off_on_time_ns)
+                        return false;
+
+                if (min_dev_off_time_ns > dev_off_time_ns
+                    || min_dev_off_time_ns < 0)
+                        min_dev_off_time_ns = dev_off_time_ns;
+        }
+
+        if (min_dev_off_time_ns < 0) {
+                /*
+                 * There are no latency constraints, so the domain can spend
+                 * arbitrary time in the "off" state.
+                 */
+                genpd->max_off_time_ns = -1;
+                return true;
+        }
+
+        /*
+         * The difference between the computed minimum delta and the time needed
+         * to turn the domain on is the maximum theoretical time this domain can
+         * spend in the "off" state.
+         */
+        min_dev_off_time_ns -= genpd->power_on_latency_ns;
+
+        /*
+         * If the difference between the computed minimum delta and the time
+         * needed to turn the domain off and back on on is smaller than the
+         * domain's power break even time, removing power from the domain is not
+         * worth it.
+         */
+        if (genpd->break_even_ns >
+            min_dev_off_time_ns - genpd->power_off_latency_ns)
+                return false;
+
+        genpd->max_off_time_ns = min_dev_off_time_ns;
+        return true;
+}
+
 struct dev_power_governor simple_qos_governor = {
         .stop_ok = default_stop_ok,
+        .power_down_ok = default_power_down_ok,
 };
diff --git a/include/linux/pm_domain.h b/include/linux/pm_domain.h
index f6745c213a57..cc1a2450ff7b 100644
--- a/include/linux/pm_domain.h
+++ b/include/linux/pm_domain.h
@@ -61,8 +61,13 @@ struct generic_pm_domain {
         bool suspend_power_off; /* Power status before system suspend */
         bool dev_irq_safe;      /* Device callbacks are IRQ-safe */
         int (*power_off)(struct generic_pm_domain *domain);
+        s64 power_off_latency_ns;
         int (*power_on)(struct generic_pm_domain *domain);
+        s64 power_on_latency_ns;
         struct gpd_dev_ops dev_ops;
+        s64 break_even_ns;      /* Power break even for the entire domain. */
+        s64 max_off_time_ns;    /* Maximum allowed "suspended" time. */
+        ktime_t power_off_time;
 };
 
 static inline struct generic_pm_domain *pd_to_genpd(struct dev_pm_domain *pd)
@@ -80,6 +85,8 @@ struct gpd_link {