PM: domains: Add a flag to defer power-off until all consumers probe

In some cases, power domains are active on boot and must remain turned
on until all their dependent drivers probe. Examples are:

- Boot-time framebuffers
- Devices that run coprocessors which are handed off already running
- Parent power domains with children that are also on at boot

The genpd core currently powers off the genpd as soon as a single
consumer device probes and goes into runtime suspend or when general
probing is complete, whichever comes first. That breaks any devices
which haven't probed yet.

To fix this, add a GENPD_FLAG_DEFER_OFF which requests that the genpd
core refuse to power down a domain if there are any consumer devices
that either haven't probed yet, or whose device nodes do not exist yet
(but fwlinks do). Genpd providers can set this if they expect to be
critical for devices (e.g. if they are powered on at boot).

It is possible for a device to be runtime suspended from its probe
callback. If this is the last device to probe, this is allowable. To
account for this, check whether the device whose callbacks are being
invoked in the probing state, and in that case, allow 1 instead of 0
pending devices.

Signed-off-by: Hector Martin <marcan@marcan.st>

Author: marcan

drivers/pmdomain/core.c

@@ -7,6 +7,7 @@
 #define pr_fmt(fmt) "PM: " fmt
 
 #include <linux/delay.h>
+#include <linux/fwnode.h>
 #include <linux/kernel.h>
 #include <linux/io.h>
 #include <linux/platform_device.h>
@@ -129,6 +130,7 @@ static const struct genpd_lock_ops genpd_spin_ops = {
 #define genpd_is_cpu_domain(genpd)	(genpd->flags & GENPD_FLAG_CPU_DOMAIN)
 #define genpd_is_rpm_always_on(genpd)	(genpd->flags & GENPD_FLAG_RPM_ALWAYS_ON)
 #define genpd_is_opp_table_fw(genpd)	(genpd->flags & GENPD_FLAG_OPP_TABLE_FW)
+#define genpd_is_defer_off(genpd)	(genpd->flags & GENPD_FLAG_DEFER_OFF)
 
 static inline bool irq_safe_dev_in_sleep_domain(struct device *dev,
 		const struct generic_pm_domain *genpd)
@@ -691,6 +693,27 @@ static void genpd_queue_power_off_work(struct generic_pm_domain *genpd)
 	queue_work(pm_wq, &genpd->power_off_work);
 }
 
+/**
+ * genpd_must_defer - Check whether the genpd cannot be safely powered off.
+ * @genpd: PM domain about to be powered down.
+ * @one_dev_probing: True if we are being called from RPM callbacks on a device that
+ * is probing, to allow poweroff if that device is the sole remaining consumer probing.
+ *
+ * Returns true if the @genpd has the GENPD_FLAG_DEFER_OFF flag and there
+ * are any consumer devices which either do not exist yet (only represented
+ * by fwlinks) or whose drivers have not probed yet.
+ */
+static bool genpd_must_defer(struct generic_pm_domain *genpd, bool one_dev_probing)
+{
+	if (genpd_is_defer_off(genpd) && genpd->has_provider) {
+		int absent = fw_devlink_count_absent_consumers(genpd->provider);
+
+		if (absent > (one_dev_probing ? 1 : 0))
+			return true;
+	}
+	return false;
+}
+
 /**
  * genpd_power_off - Remove power from a given PM domain.
  * @genpd: PM domain to power down.
@@ -704,7 +727,7 @@ static void genpd_queue_power_off_work(struct generic_pm_domain *genpd)
  * have been powered down, remove power from @genpd.
  */
 static int genpd_power_off(struct generic_pm_domain *genpd, bool one_dev_on,
-			   unsigned int depth)
+			   bool one_dev_probing, unsigned int depth)
 {
 	struct pm_domain_data *pdd;
 	struct gpd_link *link;
@@ -754,6 +777,14 @@ static int genpd_power_off(struct generic_pm_domain *genpd, bool one_dev_on,
 	if (not_suspended > 1 || (not_suspended == 1 && !one_dev_on))
 		return -EBUSY;
 
+	/*
+	 * Do not allow PM domain to be powered off if it is marked
+	 * as GENPD_FLAG_DEFER_OFF and there are consumer devices
+	 * which have not probed yet.
+	 */
+	if (genpd_must_defer(genpd, one_dev_probing))
+		return -EBUSY;
+
 	if (genpd->gov && genpd->gov->power_down_ok) {
 		if (!genpd->gov->power_down_ok(&genpd->domain))
 			return -EAGAIN;
@@ -780,7 +811,7 @@ static int genpd_power_off(struct generic_pm_domain *genpd, bool one_dev_on,
 	list_for_each_entry(link, &genpd->child_links, child_node) {
 		genpd_sd_counter_dec(link->parent);
 		genpd_lock_nested(link->parent, depth + 1);
-		genpd_power_off(link->parent, false, depth + 1);
+		genpd_power_off(link->parent, false, false, depth + 1);
 		genpd_unlock(link->parent);
 	}
 
@@ -838,7 +869,7 @@ static int genpd_power_on(struct generic_pm_domain *genpd, unsigned int depth)
 					child_node) {
 		genpd_sd_counter_dec(link->parent);
 		genpd_lock_nested(link->parent, depth + 1);
-		genpd_power_off(link->parent, false, depth + 1);
+		genpd_power_off(link->parent, false, false, depth + 1);
 		genpd_unlock(link->parent);
 	}
 
@@ -905,7 +936,7 @@ static void genpd_power_off_work_fn(struct work_struct *work)
 	genpd = container_of(work, struct generic_pm_domain, power_off_work);
 
 	genpd_lock(genpd);
-	genpd_power_off(genpd, false, 0);
+	genpd_power_off(genpd, false, false, 0);
 	genpd_unlock(genpd);
 }
 
@@ -970,6 +1001,7 @@ static int genpd_runtime_suspend(struct device *dev)
 	struct generic_pm_domain_data *gpd_data = dev_gpd_data(dev);
 	struct gpd_timing_data *td = gpd_data->td;
 	bool runtime_pm = pm_runtime_enabled(dev);
+	bool probing = dev->links.status != DL_DEV_DRIVER_BOUND;
 	ktime_t time_start = 0;
 	s64 elapsed_ns;
 	int ret;
@@ -1024,7 +1056,7 @@ static int genpd_runtime_suspend(struct device *dev)
 		return 0;
 
 	genpd_lock(genpd);
-	genpd_power_off(genpd, true, 0);
+	genpd_power_off(genpd, true, probing, 0);
 	gpd_data->rpm_pstate = genpd_drop_performance_state(dev);
 	genpd_unlock(genpd);
 
@@ -1045,6 +1077,7 @@ static int genpd_runtime_resume(struct device *dev)
 	struct generic_pm_domain_data *gpd_data = dev_gpd_data(dev);
 	struct gpd_timing_data *td = gpd_data->td;
 	bool timed = td && pm_runtime_enabled(dev);
+	bool probing = dev->links.status != DL_DEV_DRIVER_BOUND;
 	ktime_t time_start = 0;
 	s64 elapsed_ns;
 	int ret;
@@ -1102,7 +1135,7 @@ static int genpd_runtime_resume(struct device *dev)
 err_poweroff:
 	if (!pm_runtime_is_irq_safe(dev) || genpd_is_irq_safe(genpd)) {
 		genpd_lock(genpd);
-		genpd_power_off(genpd, true, 0);
+		genpd_power_off(genpd, true, probing, 0);
 		gpd_data->rpm_pstate = genpd_drop_performance_state(dev);
 		genpd_unlock(genpd);
 	}
@@ -1169,6 +1202,9 @@ static void genpd_sync_power_off(struct generic_pm_domain *genpd, bool use_lock,
 	    || atomic_read(&genpd->sd_count) > 0)
 		return;
 
+	if (genpd_must_defer(genpd, false))
+		return;
+
 	/* Check that the children are in their deepest (powered-off) state. */
 	list_for_each_entry(link, &genpd->parent_links, parent_node) {
 		struct generic_pm_domain *child = link->child;
@@ -2135,6 +2171,12 @@ int pm_genpd_init(struct generic_pm_domain *genpd,
 		return -EINVAL;
 	}
 
+	/* Deferred-off power domains should be powered on at initialization. */
+	if (genpd_is_defer_off(genpd) && !genpd_status_on(genpd)) {
+		pr_warn("deferred-off PM domain %s is not on at init\n", genpd->name);
+		genpd->flags &= ~GENPD_FLAG_DEFER_OFF;
+	}
+
 	/* Multiple states but no governor doesn't make sense. */
 	if (!gov && genpd->state_count > 1)
 		pr_warn("%s: no governor for states\n", genpd->name);

include/linux/pm_domain.h

@@ -92,6 +92,13 @@ struct dev_pm_domain_list {
  * GENPD_FLAG_OPP_TABLE_FW:	The genpd provider supports performance states,
  *				but its corresponding OPP tables are not
  *				described in DT, but are given directly by FW.
+ *
+ * GENPD_FLAG_DEFER_OFF:	Defer powerdown if there are any consumer
+ *				device fwlinks indicating that some consumer
+ *				devices have not yet probed. This is useful
+ *				for power domains which are active at boot and
+ *				must not be shut down until all consumers
+ *				complete their probe sequence.
  */
 #define GENPD_FLAG_PM_CLK	 (1U << 0)
 #define GENPD_FLAG_IRQ_SAFE	 (1U << 1)
@@ -101,6 +108,7 @@ struct dev_pm_domain_list {
 #define GENPD_FLAG_RPM_ALWAYS_ON (1U << 5)
 #define GENPD_FLAG_MIN_RESIDENCY (1U << 6)
 #define GENPD_FLAG_OPP_TABLE_FW	 (1U << 7)
+#define GENPD_FLAG_DEFER_OFF	 (1U << 8)
 
 enum gpd_status {
 	GENPD_STATE_ON = 0,	/* PM domain is on */