cpuidle: powernv/pseries: Auto-promotion of snooze to deeper idle state

The idle cpus which stay in snooze for a long period can degrade the
perfomance of the sibling cpus. If the cpu stays in snooze for more
than target residency of the next available idle state, then exit from
snooze. This gives a chance to the cpuidle governor to re-evaluate the
last idle state of the cpu to promote it to deeper idle states.
Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

drivers/cpuidle/cpuidle-powernv.c

@@ -29,18 +29,25 @@ struct cpuidle_driver powernv_idle_driver = {
 
 static int max_idle_state;
 static struct cpuidle_state *cpuidle_state_table;
+static u64 snooze_timeout;
+static bool snooze_timeout_en;
 
 static int snooze_loop(struct cpuidle_device *dev,
 			struct cpuidle_driver *drv,
 			int index)
 {
+	u64 snooze_exit_time;
+
 	local_irq_enable();
 	set_thread_flag(TIF_POLLING_NRFLAG);
 
+	snooze_exit_time = get_tb() + snooze_timeout;
 	ppc64_runlatch_off();
 	while (!need_resched()) {
 		HMT_low();
 		HMT_very_low();
+		if (snooze_timeout_en && get_tb() > snooze_exit_time)
+			break;
 	}
 
 	HMT_medium();
@@ -252,6 +259,11 @@ static int powernv_idle_probe(void)
 		cpuidle_state_table = powernv_states;
 		/* Device tree can indicate more idle states */
 		max_idle_state = powernv_add_idle_states();
+		if (max_idle_state > 1) {
+			snooze_timeout_en = true;
+			snooze_timeout = powernv_states[1].target_residency *
+					 tb_ticks_per_usec;
+		}
  	} else
  		return -ENODEV;
 

drivers/cpuidle/cpuidle-pseries.c

@@ -27,6 +27,8 @@ struct cpuidle_driver pseries_idle_driver = {
 
 static int max_idle_state;
 static struct cpuidle_state *cpuidle_state_table;
+static u64 snooze_timeout;
+static bool snooze_timeout_en;
 
 static inline void idle_loop_prolog(unsigned long *in_purr)
 {
@@ -58,14 +60,18 @@ static int snooze_loop(struct cpuidle_device *dev,
 			int index)
 {
 	unsigned long in_purr;
+	u64 snooze_exit_time;
 
 	idle_loop_prolog(&in_purr);
 	local_irq_enable();
 	set_thread_flag(TIF_POLLING_NRFLAG);
+	snooze_exit_time = get_tb() + snooze_timeout;
 
 	while (!need_resched()) {
 		HMT_low();
 		HMT_very_low();
+		if (snooze_timeout_en && get_tb() > snooze_exit_time)
+			break;
 	}
 
 	HMT_medium();
@@ -244,6 +250,11 @@ static int pseries_idle_probe(void)
 	} else
 		return -ENODEV;
 
+	if (max_idle_state > 1) {
+		snooze_timeout_en = true;
+		snooze_timeout = cpuidle_state_table[1].target_residency *
+				 tb_ticks_per_usec;
+	}
 	return 0;
 }