diff --git a/drivers/base/power/domain.c b/drivers/base/power/domain.c
index 8e554e6a82a26..611c0ccbad5fa 100644
--- a/drivers/base/power/domain.c
+++ b/drivers/base/power/domain.c
@@ -239,6 +239,127 @@ static void genpd_update_accounting(struct generic_pm_domain *genpd)
 static inline void genpd_update_accounting(struct generic_pm_domain *genpd) {}
 #endif
 
+static int _genpd_reeval_performance_state(struct generic_pm_domain *genpd,
+					   unsigned int state)
+{
+	struct generic_pm_domain_data *pd_data;
+	struct pm_domain_data *pdd;
+	struct gpd_link *link;
+
+	/* New requested state is same as Max requested state */
+	if (state == genpd->performance_state)
+		return state;
+
+	/* New requested state is higher than Max requested state */
+	if (state > genpd->performance_state)
+		return state;
+
+	/* Traverse all devices within the domain */
+	list_for_each_entry(pdd, &genpd->dev_list, list_node) {
+		pd_data = to_gpd_data(pdd);
+
+		if (pd_data->performance_state > state)
+			state = pd_data->performance_state;
+	}
+
+	/*
+	 * Traverse all sub-domains within the domain. This can be
+	 * done without any additional locking as the link->performance_state
+	 * field is protected by the master genpd->lock, which is already taken.
+	 *
+	 * Also note that link->performance_state (subdomain's performance state
+	 * requirement to master domain) is different from
+	 * link->slave->performance_state (current performance state requirement
+	 * of the devices/sub-domains of the subdomain) and so can have a
+	 * different value.
+	 *
+	 * Note that we also take vote from powered-off sub-domains into account
+	 * as the same is done for devices right now.
+	 */
+	list_for_each_entry(link, &genpd->master_links, master_node) {
+		if (link->performance_state > state)
+			state = link->performance_state;
+	}
+
+	return state;
+}
+
+static int _genpd_set_performance_state(struct generic_pm_domain *genpd,
+					unsigned int state, int depth)
+{
+	struct generic_pm_domain *master;
+	struct gpd_link *link;
+	int master_state, ret;
+
+	if (state == genpd->performance_state)
+		return 0;
+
+	/* Propagate to masters of genpd */
+	list_for_each_entry(link, &genpd->slave_links, slave_node) {
+		master = link->master;
+
+		if (!master->set_performance_state)
+			continue;
+
+		/* Find master's performance state */
+		ret = dev_pm_opp_xlate_performance_state(genpd->opp_table,
+							 master->opp_table,
+							 state);
+		if (unlikely(ret < 0))
+			goto err;
+
+		master_state = ret;
+
+		genpd_lock_nested(master, depth + 1);
+
+		link->prev_performance_state = link->performance_state;
+		link->performance_state = master_state;
+		master_state = _genpd_reeval_performance_state(master,
+						master_state);
+		ret = _genpd_set_performance_state(master, master_state, depth + 1);
+		if (ret)
+			link->performance_state = link->prev_performance_state;
+
+		genpd_unlock(master);
+
+		if (ret)
+			goto err;
+	}
+
+	ret = genpd->set_performance_state(genpd, state);
+	if (ret)
+		goto err;
+
+	genpd->performance_state = state;
+	return 0;
+
+err:
+	/* Encountered an error, lets rollback */
+	list_for_each_entry_continue_reverse(link, &genpd->slave_links,
+					     slave_node) {
+		master = link->master;
+
+		if (!master->set_performance_state)
+			continue;
+
+		genpd_lock_nested(master, depth + 1);
+
+		master_state = link->prev_performance_state;
+		link->performance_state = master_state;
+
+		master_state = _genpd_reeval_performance_state(master,
+						master_state);
+		if (_genpd_set_performance_state(master, master_state, depth + 1)) {
+			pr_err("%s: Failed to roll back to %d performance state\n",
+			       master->name, master_state);
+		}
+
+		genpd_unlock(master);
+	}
+
+	return ret;
+}
+
 /**
  * dev_pm_genpd_set_performance_state- Set performance state of device's power
  * domain.
@@ -257,10 +378,9 @@ static inline void genpd_update_accounting(struct generic_pm_domain *genpd) {}
 int dev_pm_genpd_set_performance_state(struct device *dev, unsigned int state)
 {
 	struct generic_pm_domain *genpd;
-	struct generic_pm_domain_data *gpd_data, *pd_data;
-	struct pm_domain_data *pdd;
+	struct generic_pm_domain_data *gpd_data;
 	unsigned int prev;
-	int ret = 0;
+	int ret;
 
 	genpd = dev_to_genpd(dev);
 	if (IS_ERR(genpd))
@@ -281,47 +401,11 @@ int dev_pm_genpd_set_performance_state(struct device *dev, unsigned int state)
 	prev = gpd_data->performance_state;
 	gpd_data->performance_state = state;
 
-	/* New requested state is same as Max requested state */
-	if (state == genpd->performance_state)
-		goto unlock;
-
-	/* New requested state is higher than Max requested state */
-	if (state > genpd->performance_state)
-		goto update_state;
-
-	/* Traverse all devices within the domain */
-	list_for_each_entry(pdd, &genpd->dev_list, list_node) {
-		pd_data = to_gpd_data(pdd);
-
-		if (pd_data->performance_state > state)
-			state = pd_data->performance_state;
-	}
-
-	if (state == genpd->performance_state)
-		goto unlock;
-
-	/*
-	 * We aren't propagating performance state changes of a subdomain to its
-	 * masters as we don't have hardware that needs it. Over that, the
-	 * performance states of subdomain and its masters may not have
-	 * one-to-one mapping and would require additional information. We can
-	 * get back to this once we have hardware that needs it. For that
-	 * reason, we don't have to consider performance state of the subdomains
-	 * of genpd here.
-	 */
-
-update_state:
-	if (genpd_status_on(genpd)) {
-		ret = genpd->set_performance_state(genpd, state);
-		if (ret) {
-			gpd_data->performance_state = prev;
-			goto unlock;
-		}
-	}
-
-	genpd->performance_state = state;
+	state = _genpd_reeval_performance_state(genpd, state);
+	ret = _genpd_set_performance_state(genpd, state, 0);
+	if (ret)
+		gpd_data->performance_state = prev;
 
-unlock:
 	genpd_unlock(genpd);
 
 	return ret;
@@ -347,15 +431,6 @@ static int _genpd_power_on(struct generic_pm_domain *genpd, bool timed)
 		return ret;
 
 	elapsed_ns = ktime_to_ns(ktime_sub(ktime_get(), time_start));
-
-	if (unlikely(genpd->set_performance_state)) {
-		ret = genpd->set_performance_state(genpd, genpd->performance_state);
-		if (ret) {
-			pr_warn("%s: Failed to set performance state %d (%d)\n",
-				genpd->name, genpd->performance_state, ret);
-		}
-	}
-
 	if (elapsed_ns <= genpd->states[state_idx].power_on_latency_ns)
 		return ret;
 
@@ -1907,12 +1982,21 @@ int of_genpd_add_provider_simple(struct device_node *np,
 				ret);
 			goto unlock;
 		}
+
+		/*
+		 * Save table for faster processing while setting performance
+		 * state.
+		 */
+		genpd->opp_table = dev_pm_opp_get_opp_table(&genpd->dev);
+		WARN_ON(!genpd->opp_table);
 	}
 
 	ret = genpd_add_provider(np, genpd_xlate_simple, genpd);
 	if (ret) {
-		if (genpd->set_performance_state)
+		if (genpd->set_performance_state) {
+			dev_pm_opp_put_opp_table(genpd->opp_table);
 			dev_pm_opp_of_remove_table(&genpd->dev);
+		}
 
 		goto unlock;
 	}
@@ -1965,6 +2049,13 @@ int of_genpd_add_provider_onecell(struct device_node *np,
 					i, ret);
 				goto error;
 			}
+
+			/*
+			 * Save table for faster processing while setting
+			 * performance state.
+			 */
+			genpd->opp_table = dev_pm_opp_get_opp_table_indexed(&genpd->dev, i);
+			WARN_ON(!genpd->opp_table);
 		}
 
 		genpd->provider = &np->fwnode;
@@ -1989,8 +2080,10 @@ int of_genpd_add_provider_onecell(struct device_node *np,
 		genpd->provider = NULL;
 		genpd->has_provider = false;
 
-		if (genpd->set_performance_state)
+		if (genpd->set_performance_state) {
+			dev_pm_opp_put_opp_table(genpd->opp_table);
 			dev_pm_opp_of_remove_table(&genpd->dev);
+		}
 	}
 
 	mutex_unlock(&gpd_list_lock);
@@ -2024,6 +2117,7 @@ void of_genpd_del_provider(struct device_node *np)
 					if (!gpd->set_performance_state)
 						continue;
 
+					dev_pm_opp_put_opp_table(gpd->opp_table);
 					dev_pm_opp_of_remove_table(&gpd->dev);
 				}
 			}
diff --git a/drivers/opp/core.c b/drivers/opp/core.c
index 0eaa954b3f6c6..e5507add8f04f 100644
--- a/drivers/opp/core.c
+++ b/drivers/opp/core.c
@@ -196,12 +196,12 @@ unsigned long dev_pm_opp_get_max_volt_latency(struct device *dev)
 	if (IS_ERR(opp_table))
 		return 0;
 
-	count = opp_table->regulator_count;
-
 	/* Regulator may not be required for the device */
-	if (!count)
+	if (!opp_table->regulators)
 		goto put_opp_table;
 
+	count = opp_table->regulator_count;
+
 	uV = kmalloc_array(count, sizeof(*uV), GFP_KERNEL);
 	if (!uV)
 		goto put_opp_table;
@@ -843,6 +843,9 @@ static struct opp_table *_allocate_opp_table(struct device *dev, int index)
 	mutex_init(&opp_table->genpd_virt_dev_lock);
 	INIT_LIST_HEAD(&opp_table->dev_list);
 
+	/* Mark regulator count uninitialized */
+	opp_table->regulator_count = -1;
+
 	opp_dev = _add_opp_dev(dev, opp_table);
 	if (!opp_dev) {
 		kfree(opp_table);
@@ -1063,7 +1066,7 @@ struct dev_pm_opp *_opp_allocate(struct opp_table *table)
 	int count, supply_size;
 
 	/* Allocate space for at least one supply */
-	count = table->regulator_count ? table->regulator_count : 1;
+	count = table->regulator_count > 0 ? table->regulator_count : 1;
 	supply_size = sizeof(*opp->supplies) * count;
 
 	/* allocate new OPP node and supplies structures */
@@ -1084,6 +1087,9 @@ static bool _opp_supported_by_regulators(struct dev_pm_opp *opp,
 	struct regulator *reg;
 	int i;
 
+	if (!opp_table->regulators)
+		return true;
+
 	for (i = 0; i < opp_table->regulator_count; i++) {
 		reg = opp_table->regulators[i];
 
@@ -1368,7 +1374,7 @@ static int _allocate_set_opp_data(struct opp_table *opp_table)
 	struct dev_pm_set_opp_data *data;
 	int len, count = opp_table->regulator_count;
 
-	if (WARN_ON(!count))
+	if (WARN_ON(!opp_table->regulators))
 		return -EINVAL;
 
 	/* space for set_opp_data */
@@ -1465,7 +1471,7 @@ struct opp_table *dev_pm_opp_set_regulators(struct device *dev,
 
 	kfree(opp_table->regulators);
 	opp_table->regulators = NULL;
-	opp_table->regulator_count = 0;
+	opp_table->regulator_count = -1;
 err:
 	dev_pm_opp_put_opp_table(opp_table);
 
@@ -1494,7 +1500,7 @@ void dev_pm_opp_put_regulators(struct opp_table *opp_table)
 
 	kfree(opp_table->regulators);
 	opp_table->regulators = NULL;
-	opp_table->regulator_count = 0;
+	opp_table->regulator_count = -1;
 
 put_opp_table:
 	dev_pm_opp_put_opp_table(opp_table);
@@ -1707,6 +1713,69 @@ void dev_pm_opp_put_genpd_virt_dev(struct opp_table *opp_table,
 		dev_err(virt_dev, "Failed to find required device entry\n");
 }
 
+/**
+ * dev_pm_opp_xlate_performance_state() - Find required OPP's pstate for src_table.
+ * @src_table: OPP table which has dst_table as one of its required OPP table.
+ * @dst_table: Required OPP table of the src_table.
+ * @pstate: Current performance state of the src_table.
+ *
+ * This Returns pstate of the OPP (present in @dst_table) pointed out by the
+ * "required-opps" property of the OPP (present in @src_table) which has
+ * performance state set to @pstate.
+ *
+ * Return: Zero or positive performance state on success, otherwise negative
+ * value on errors.
+ */
+int dev_pm_opp_xlate_performance_state(struct opp_table *src_table,
+				       struct opp_table *dst_table,
+				       unsigned int pstate)
+{
+	struct dev_pm_opp *opp;
+	int dest_pstate = -EINVAL;
+	int i;
+
+	if (!pstate)
+		return 0;
+
+	/*
+	 * Normally the src_table will have the "required_opps" property set to
+	 * point to one of the OPPs in the dst_table, but in some cases the
+	 * genpd and its master have one to one mapping of performance states
+	 * and so none of them have the "required-opps" property set. Return the
+	 * pstate of the src_table as it is in such cases.
+	 */
+	if (!src_table->required_opp_count)
+		return pstate;
+
+	for (i = 0; i < src_table->required_opp_count; i++) {
+		if (src_table->required_opp_tables[i]->np == dst_table->np)
+			break;
+	}
+
+	if (unlikely(i == src_table->required_opp_count)) {
+		pr_err("%s: Couldn't find matching OPP table (%p: %p)\n",
+		       __func__, src_table, dst_table);
+		return -EINVAL;
+	}
+
+	mutex_lock(&src_table->lock);
+
+	list_for_each_entry(opp, &src_table->opp_list, node) {
+		if (opp->pstate == pstate) {
+			dest_pstate = opp->required_opps[i]->pstate;
+			goto unlock;
+		}
+	}
+
+	pr_err("%s: Couldn't find matching OPP (%p: %p)\n", __func__, src_table,
+	       dst_table);
+
+unlock:
+	mutex_unlock(&src_table->lock);
+
+	return dest_pstate;
+}
+
 /**
  * dev_pm_opp_add()  - Add an OPP table from a table definitions
  * @dev:	device for which we do this operation
@@ -1733,6 +1802,9 @@ int dev_pm_opp_add(struct device *dev, unsigned long freq, unsigned long u_volt)
 	if (!opp_table)
 		return -ENOMEM;
 
+	/* Fix regulator count for dynamic OPPs */
+	opp_table->regulator_count = 1;
+
 	ret = _opp_add_v1(opp_table, dev, freq, u_volt, true);
 	if (ret)
 		dev_pm_opp_put_opp_table(opp_table);
diff --git a/drivers/opp/of.c b/drivers/opp/of.c
index 840f85181a373..06f0f632ec477 100644
--- a/drivers/opp/of.c
+++ b/drivers/opp/of.c
@@ -114,19 +114,25 @@ static struct device_node *of_parse_required_opp(struct device_node *np,
 static struct opp_table *_find_table_of_opp_np(struct device_node *opp_np)
 {
 	struct opp_table *opp_table;
-	struct dev_pm_opp *opp;
+	struct device_node *opp_table_np;
 
 	lockdep_assert_held(&opp_table_lock);
 
+	opp_table_np = of_get_parent(opp_np);
+	if (!opp_table_np)
+		goto err;
+
+	/* It is safe to put the node now as all we need now is its address */
+	of_node_put(opp_table_np);
+
 	list_for_each_entry(opp_table, &opp_tables, node) {
-		opp = _find_opp_of_np(opp_table, opp_np);
-		if (opp) {
-			dev_pm_opp_put(opp);
+		if (opp_table_np == opp_table->np) {
 			_get_opp_table_kref(opp_table);
 			return opp_table;
 		}
 	}
 
+err:
 	return ERR_PTR(-ENODEV);
 }
 
@@ -385,12 +391,10 @@ static int opp_parse_supplies(struct dev_pm_opp *opp, struct device *dev,
 			      struct opp_table *opp_table)
 {
 	u32 *microvolt, *microamp = NULL;
-	int supplies, vcount, icount, ret, i, j;
+	int supplies = opp_table->regulator_count, vcount, icount, ret, i, j;
 	struct property *prop = NULL;
 	char name[NAME_MAX];
 
-	supplies = opp_table->regulator_count ? opp_table->regulator_count : 1;
-
 	/* Search for "opp-microvolt-<name>" */
 	if (opp_table->prop_name) {
 		snprintf(name, sizeof(name), "opp-microvolt-%s",
@@ -405,7 +409,13 @@ static int opp_parse_supplies(struct dev_pm_opp *opp, struct device *dev,
 
 		/* Missing property isn't a problem, but an invalid entry is */
 		if (!prop) {
-			if (!opp_table->regulator_count)
+			if (unlikely(supplies == -1)) {
+				/* Initialize regulator_count */
+				opp_table->regulator_count = 0;
+				return 0;
+			}
+
+			if (!supplies)
 				return 0;
 
 			dev_err(dev, "%s: opp-microvolt missing although OPP managing regulators\n",
@@ -414,6 +424,14 @@ static int opp_parse_supplies(struct dev_pm_opp *opp, struct device *dev,
 		}
 	}
 
+	if (unlikely(supplies == -1)) {
+		/* Initialize regulator_count */
+		supplies = opp_table->regulator_count = 1;
+	} else if (unlikely(!supplies)) {
+		dev_err(dev, "%s: opp-microvolt wasn't expected\n", __func__);
+		return -EINVAL;
+	}
+
 	vcount = of_property_count_u32_elems(opp->np, name);
 	if (vcount < 0) {
 		dev_err(dev, "%s: Invalid %s property (%d)\n",
@@ -975,19 +993,19 @@ EXPORT_SYMBOL_GPL(dev_pm_opp_of_get_sharing_cpus);
  * Returns the performance state of the OPP pointed out by the "required-opps"
  * property at @index in @np.
  *
- * Return: Positive performance state on success, otherwise 0 on errors.
+ * Return: Zero or positive performance state on success, otherwise negative
+ * value on errors.
  */
-unsigned int of_get_required_opp_performance_state(struct device_node *np,
-						   int index)
+int of_get_required_opp_performance_state(struct device_node *np, int index)
 {
 	struct dev_pm_opp *opp;
 	struct device_node *required_np;
 	struct opp_table *opp_table;
-	unsigned int pstate = 0;
+	int pstate = -EINVAL;
 
 	required_np = of_parse_required_opp(np, index);
 	if (!required_np)
-		return 0;
+		return -EINVAL;
 
 	opp_table = _find_table_of_opp_np(required_np);
 	if (IS_ERR(opp_table)) {
diff --git a/drivers/opp/opp.h b/drivers/opp/opp.h
index 8aec38792cae7..e24d81497375d 100644
--- a/drivers/opp/opp.h
+++ b/drivers/opp/opp.h
@@ -145,7 +145,9 @@ enum opp_table_access {
  * @prop_name: A name to postfix to many DT properties, while parsing them.
  * @clk: Device's clock handle
  * @regulators: Supply regulators
- * @regulator_count: Number of power supply regulators
+ * @regulator_count: Number of power supply regulators. Its value can be -1
+ * (uninitialized), 0 (no opp-microvolt property) or > 0 (has opp-microvolt
+ * property).
  * @genpd_performance_state: Device's power domain support performance state.
  * @is_genpd: Marks if the OPP table belongs to a genpd.
  * @set_opp: Platform specific set_opp callback
@@ -189,7 +191,7 @@ struct opp_table {
 	const char *prop_name;
 	struct clk *clk;
 	struct regulator **regulators;
-	unsigned int regulator_count;
+	int regulator_count;
 	bool genpd_performance_state;
 	bool is_genpd;
 
diff --git a/include/linux/pm_domain.h b/include/linux/pm_domain.h
index 642036952553c..dd364abb649a4 100644
--- a/include/linux/pm_domain.h
+++ b/include/linux/pm_domain.h
@@ -73,6 +73,7 @@ struct genpd_power_state {
 
 struct genpd_lock_ops;
 struct dev_pm_opp;
+struct opp_table;
 
 struct generic_pm_domain {
 	struct device dev;
@@ -94,6 +95,7 @@ struct generic_pm_domain {
 	unsigned int performance_state;	/* Aggregated max performance state */
 	int (*power_off)(struct generic_pm_domain *domain);
 	int (*power_on)(struct generic_pm_domain *domain);
+	struct opp_table *opp_table;	/* OPP table of the genpd */
 	unsigned int (*opp_to_performance_state)(struct generic_pm_domain *genpd,
 						 struct dev_pm_opp *opp);
 	int (*set_performance_state)(struct generic_pm_domain *genpd,
@@ -134,6 +136,10 @@ struct gpd_link {
 	struct list_head master_node;
 	struct generic_pm_domain *slave;
 	struct list_head slave_node;
+
+	/* Sub-domain's per-master domain performance state */
+	unsigned int performance_state;
+	unsigned int prev_performance_state;
 };
 
 struct gpd_timing_data {
diff --git a/include/linux/pm_opp.h b/include/linux/pm_opp.h
index 2b2c3fd985abf..0a2a88e5a383f 100644
--- a/include/linux/pm_opp.h
+++ b/include/linux/pm_opp.h
@@ -128,6 +128,7 @@ struct opp_table *dev_pm_opp_register_set_opp_helper(struct device *dev, int (*s
 void dev_pm_opp_unregister_set_opp_helper(struct opp_table *opp_table);
 struct opp_table *dev_pm_opp_set_genpd_virt_dev(struct device *dev, struct device *virt_dev, int index);
 void dev_pm_opp_put_genpd_virt_dev(struct opp_table *opp_table, struct device *virt_dev);
+int dev_pm_opp_xlate_performance_state(struct opp_table *src_table, struct opp_table *dst_table, unsigned int pstate);
 int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq);
 int dev_pm_opp_set_sharing_cpus(struct device *cpu_dev, const struct cpumask *cpumask);
 int dev_pm_opp_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpumask);
@@ -280,6 +281,12 @@ static inline struct opp_table *dev_pm_opp_set_genpd_virt_dev(struct device *dev
 }
 
 static inline void dev_pm_opp_put_genpd_virt_dev(struct opp_table *opp_table, struct device *virt_dev) {}
+
+static inline int dev_pm_opp_xlate_performance_state(struct opp_table *src_table, struct opp_table *dst_table, unsigned int pstate)
+{
+	return -ENOTSUPP;
+}
+
 static inline int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)
 {
 	return -ENOTSUPP;
@@ -314,7 +321,7 @@ void dev_pm_opp_of_cpumask_remove_table(const struct cpumask *cpumask);
 int dev_pm_opp_of_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpumask);
 struct device_node *dev_pm_opp_of_get_opp_desc_node(struct device *dev);
 struct device_node *dev_pm_opp_get_of_node(struct dev_pm_opp *opp);
-unsigned int of_get_required_opp_performance_state(struct device_node *np, int index);
+int of_get_required_opp_performance_state(struct device_node *np, int index);
 #else
 static inline int dev_pm_opp_of_add_table(struct device *dev)
 {
@@ -353,9 +360,9 @@ static inline struct device_node *dev_pm_opp_get_of_node(struct dev_pm_opp *opp)
 {
 	return NULL;
 }
-static inline unsigned int of_get_required_opp_performance_state(struct device_node *np, int index)
+static inline int of_get_required_opp_performance_state(struct device_node *np, int index)
 {
-	return 0;
+	return -ENOTSUPP;
 }
 #endif