diff --git a/arch/arm/common/mcpm_entry.c b/arch/arm/common/mcpm_entry.c
index 3c165fc..5f8a52a 100644
--- a/arch/arm/common/mcpm_entry.c
+++ b/arch/arm/common/mcpm_entry.c
@@ -55,22 +55,81 @@
 	return (platform_ops) ? true : false;
 }
 
+/*
+ * We can't use regular spinlocks. In the switcher case, it is possible
+ * for an outbound CPU to call power_down() after its inbound counterpart
+ * is already live using the same logical CPU number which trips lockdep
+ * debugging.
+ */
+static arch_spinlock_t mcpm_lock = __ARCH_SPIN_LOCK_UNLOCKED;
+
+static int mcpm_cpu_use_count[MAX_NR_CLUSTERS][MAX_CPUS_PER_CLUSTER];
+
+static inline bool mcpm_cluster_unused(unsigned int cluster)
+{
+	int i, cnt;
+	for (i = 0, cnt = 0; i < MAX_CPUS_PER_CLUSTER; i++)
+		cnt |= mcpm_cpu_use_count[cluster][i];
+	return !cnt;
+}
+
 int mcpm_cpu_power_up(unsigned int cpu, unsigned int cluster)
 {
+	bool cpu_is_down, cluster_is_down;
+	int ret = 0;
+
 	if (!platform_ops)
 		return -EUNATCH; /* try not to shadow power_up errors */
 	might_sleep();
-	return platform_ops->power_up(cpu, cluster);
+
+	/* backward compatibility callback */
+	if (platform_ops->power_up)
+		return platform_ops->power_up(cpu, cluster);
+
+	pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
+
+	/*
+	 * Since this is called with IRQs enabled, and no arch_spin_lock_irq
+	 * variant exists, we need to disable IRQs manually here.
+	 */
+	local_irq_disable();
+	arch_spin_lock(&mcpm_lock);
+
+	cpu_is_down = !mcpm_cpu_use_count[cluster][cpu];
+	cluster_is_down = mcpm_cluster_unused(cluster);
+
+	mcpm_cpu_use_count[cluster][cpu]++;
+	/*
+	 * The only possible values are:
+	 * 0 = CPU down
+	 * 1 = CPU (still) up
+	 * 2 = CPU requested to be up before it had a chance
+	 *     to actually make itself down.
+	 * Any other value is a bug.
+	 */
+	BUG_ON(mcpm_cpu_use_count[cluster][cpu] != 1 &&
+	       mcpm_cpu_use_count[cluster][cpu] != 2);
+
+	if (cluster_is_down)
+		ret = platform_ops->cluster_powerup(cluster);
+	if (cpu_is_down && !ret)
+		ret = platform_ops->cpu_powerup(cpu, cluster);
+
+	arch_spin_unlock(&mcpm_lock);
+	local_irq_enable();
+	return ret;
 }
 
 typedef void (*phys_reset_t)(unsigned long);
 
 void mcpm_cpu_power_down(void)
 {
+	unsigned int mpidr, cpu, cluster;
+	bool cpu_going_down, last_man;
 	phys_reset_t phys_reset;
 
-	if (WARN_ON_ONCE(!platform_ops || !platform_ops->power_down))
-		return;
+	if (WARN_ON_ONCE(!platform_ops))
+	       return;
 	BUG_ON(!irqs_disabled());
 
 	/*
@@ -79,28 +138,65 @@
 	 */
 	setup_mm_for_reboot();
 
-	platform_ops->power_down();
+	/* backward compatibility callback */
+	if (platform_ops->power_down) {
+		platform_ops->power_down();
+		goto not_dead;
+	}
 
+	mpidr = read_cpuid_mpidr();
+	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
+	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
+	pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
+
+	__mcpm_cpu_going_down(cpu, cluster);
+
+	arch_spin_lock(&mcpm_lock);
+	BUG_ON(__mcpm_cluster_state(cluster) != CLUSTER_UP);
+
+	mcpm_cpu_use_count[cluster][cpu]--;
+	BUG_ON(mcpm_cpu_use_count[cluster][cpu] != 0 &&
+	       mcpm_cpu_use_count[cluster][cpu] != 1);
+	cpu_going_down = !mcpm_cpu_use_count[cluster][cpu];
+	last_man = mcpm_cluster_unused(cluster);
+
+	if (last_man && __mcpm_outbound_enter_critical(cpu, cluster)) {
+		platform_ops->cpu_powerdown_prepare(cpu, cluster);
+		platform_ops->cluster_powerdown_prepare(cluster);
+		arch_spin_unlock(&mcpm_lock);
+		platform_ops->cluster_cache_disable();
+		__mcpm_outbound_leave_critical(cluster, CLUSTER_DOWN);
+	} else {
+		if (cpu_going_down)
+			platform_ops->cpu_powerdown_prepare(cpu, cluster);
+		arch_spin_unlock(&mcpm_lock);
+		/*
+		 * If cpu_going_down is false here, that means a power_up
+		 * request raced ahead of us.  Even if we do not want to
+		 * shut this CPU down, the caller still expects execution
+		 * to return through the system resume entry path, like
+		 * when the WFI is aborted due to a new IRQ or the like..
+		 * So let's continue with cache cleaning in all cases.
+		 */
+		platform_ops->cpu_cache_disable();
+	}
+
+	__mcpm_cpu_down(cpu, cluster);
+
+	/* Now we are prepared for power-down, do it: */
+	if (cpu_going_down)
+		wfi();
+
+not_dead:
 	/*
 	 * It is possible for a power_up request to happen concurrently
 	 * with a power_down request for the same CPU. In this case the
-	 * power_down method might not be able to actually enter a
-	 * powered down state with the WFI instruction if the power_up
-	 * method has removed the required reset condition.  The
-	 * power_down method is then allowed to return. We must perform
-	 * a re-entry in the kernel as if the power_up method just had
-	 * deasserted reset on the CPU.
-	 *
-	 * To simplify race issues, the platform specific implementation
-	 * must accommodate for the possibility of unordered calls to
-	 * power_down and power_up with a usage count. Therefore, if a
-	 * call to power_up is issued for a CPU that is not down, then
-	 * the next call to power_down must not attempt a full shutdown
-	 * but only do the minimum (normally disabling L1 cache and CPU
-	 * coherency) and return just as if a concurrent power_up request
-	 * had happened as described above.
+	 * CPU might not be able to actually enter a powered down state
+	 * with the WFI instruction if the power_up request has removed
+	 * the required reset condition.  We must perform a re-entry in
+	 * the kernel as if the power_up method just had deasserted reset
+	 * on the CPU.
 	 */
-
 	phys_reset = (phys_reset_t)(unsigned long)virt_to_phys(cpu_reset);
 	phys_reset(virt_to_phys(mcpm_entry_point));
 
@@ -125,26 +221,66 @@
 
 void mcpm_cpu_suspend(u64 expected_residency)
 {
-	phys_reset_t phys_reset;
-
-	if (WARN_ON_ONCE(!platform_ops || !platform_ops->suspend))
+	if (WARN_ON_ONCE(!platform_ops))
 		return;
-	BUG_ON(!irqs_disabled());
 
-	/* Very similar to mcpm_cpu_power_down() */
-	setup_mm_for_reboot();
-	platform_ops->suspend(expected_residency);
-	phys_reset = (phys_reset_t)(unsigned long)virt_to_phys(cpu_reset);
-	phys_reset(virt_to_phys(mcpm_entry_point));
-	BUG();
+	/* backward compatibility callback */
+	if (platform_ops->suspend) {
+		phys_reset_t phys_reset;
+		BUG_ON(!irqs_disabled());
+		setup_mm_for_reboot();
+		platform_ops->suspend(expected_residency);
+		phys_reset = (phys_reset_t)(unsigned long)virt_to_phys(cpu_reset);
+		phys_reset(virt_to_phys(mcpm_entry_point));
+		BUG();
+	}
+
+	/* Some platforms might have to enable special resume modes, etc. */
+	if (platform_ops->cpu_suspend_prepare) {
+		unsigned int mpidr = read_cpuid_mpidr();
+		unsigned int cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
+		unsigned int cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1); 
+		arch_spin_lock(&mcpm_lock);
+		platform_ops->cpu_suspend_prepare(cpu, cluster);
+		arch_spin_unlock(&mcpm_lock);
+	}
+	mcpm_cpu_power_down();
 }
 
 int mcpm_cpu_powered_up(void)
 {
+	unsigned int mpidr, cpu, cluster;
+	bool cpu_was_down, first_man;
+	unsigned long flags;
+
 	if (!platform_ops)
 		return -EUNATCH;
-	if (platform_ops->powered_up)
+
+	/* backward compatibility callback */
+	if (platform_ops->powered_up) {
 		platform_ops->powered_up();
+		return 0;
+	}
+
+	mpidr = read_cpuid_mpidr();
+	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
+	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
+	local_irq_save(flags);
+	arch_spin_lock(&mcpm_lock);
+
+	cpu_was_down = !mcpm_cpu_use_count[cluster][cpu];
+	first_man = mcpm_cluster_unused(cluster);
+
+	if (first_man && platform_ops->cluster_is_up)
+		platform_ops->cluster_is_up(cluster);
+	if (cpu_was_down)
+		mcpm_cpu_use_count[cluster][cpu] = 1;
+	if (platform_ops->cpu_is_up)
+		platform_ops->cpu_is_up(cpu, cluster);
+
+	arch_spin_unlock(&mcpm_lock);
+	local_irq_restore(flags);
+
 	return 0;
 }
 
@@ -334,8 +470,10 @@
 	}
 	mpidr = read_cpuid_mpidr();
 	this_cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
-	for_each_online_cpu(i)
+	for_each_online_cpu(i) {
+		mcpm_cpu_use_count[this_cluster][i] = 1;
 		mcpm_sync.clusters[this_cluster].cpus[i].cpu = CPU_UP;
+	}
 	mcpm_sync.clusters[this_cluster].cluster = CLUSTER_UP;
 	sync_cache_w(&mcpm_sync);
 
diff --git a/arch/arm/include/asm/mcpm.h b/arch/arm/include/asm/mcpm.h
index 3446f6a..50b378f 100644
--- a/arch/arm/include/asm/mcpm.h
+++ b/arch/arm/include/asm/mcpm.h
@@ -171,12 +171,73 @@
 int mcpm_cpu_powered_up(void);
 
 /*
- * Platform specific methods used in the implementation of the above API.
+ * Platform specific callbacks used in the implementation of the above API.
+ *
+ * cpu_powerup:
+ * Make given CPU runable. Called with MCPM lock held and IRQs disabled.
+ * The given cluster is assumed to be set up (cluster_powerup would have
+ * been called beforehand). Must return 0 for success or negative error code.
+ *
+ * cluster_powerup:
+ * Set up power for given cluster. Called with MCPM lock held and IRQs
+ * disabled. Called before first cpu_powerup when cluster is down. Must
+ * return 0 for success or negative error code.
+ *
+ * cpu_suspend_prepare:
+ * Special suspend configuration. Called on target CPU with MCPM lock held
+ * and IRQs disabled. This callback is optional. If provided, it is called
+ * before cpu_powerdown_prepare.
+ *
+ * cpu_powerdown_prepare:
+ * Configure given CPU for power down. Called on target CPU with MCPM lock
+ * held and IRQs disabled. Power down must be effective only at the next WFI instruction.
+ *
+ * cluster_powerdown_prepare:
+ * Configure given cluster for power down. Called on one CPU from target
+ * cluster with MCPM lock held and IRQs disabled. A cpu_powerdown_prepare
+ * for each CPU in the cluster has happened when this occurs.
+ *
+ * cpu_cache_disable:
+ * Clean and disable CPU level cache for the calling CPU. Called on with IRQs
+ * disabled only. The CPU is no longer cache coherent with the rest of the
+ * system when this returns.
+ *
+ * cluster_cache_disable:
+ * Clean and disable the cluster wide cache as well as the CPU level cache
+ * for the calling CPU. No call to cpu_cache_disable will happen for this
+ * CPU. Called with IRQs disabled and only when all the other CPUs are done
+ * with their own cpu_cache_disable. The cluster is no longer cache coherent
+ * with the rest of the system when this returns.
+ *
+ * cpu_is_up:
+ * Called on given CPU after it has been powered up or resumed. The MCPM lock
+ * is held and IRQs disabled. This callback is optional.
+ *
+ * cluster_is_up:
+ * Called by the first CPU to be powered up or resumed in given cluster.
+ * The MCPM lock is held and IRQs disabled. This callback is optional. If
+ * provided, it is called before cpu_is_up for that CPU.
+ *
+ * wait_for_powerdown:
+ * Wait until given CPU is powered down. This is called in sleeping context.
+ * Some reasonable timeout must be considered. Must return 0 for success or
+ * negative error code.
  */
 struct mcpm_platform_ops {
+	int (*cpu_powerup)(unsigned int cpu, unsigned int cluster);
+	int (*cluster_powerup)(unsigned int cluster);
+	void (*cpu_suspend_prepare)(unsigned int cpu, unsigned int cluster);
+	void (*cpu_powerdown_prepare)(unsigned int cpu, unsigned int cluster);
+	void (*cluster_powerdown_prepare)(unsigned int cluster);
+	void (*cpu_cache_disable)(void);
+	void (*cluster_cache_disable)(void);
+	void (*cpu_is_up)(unsigned int cpu, unsigned int cluster);
+	void (*cluster_is_up)(unsigned int cluster);
+	int (*wait_for_powerdown)(unsigned int cpu, unsigned int cluster);
+
+	/* deprecated callbacks */
 	int (*power_up)(unsigned int cpu, unsigned int cluster);
 	void (*power_down)(void);
-	int (*wait_for_powerdown)(unsigned int cpu, unsigned int cluster);
 	void (*suspend)(u64);
 	void (*powered_up)(void);
 };