2 files changed, 60 insertions, 33 deletions

diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
index 23a06cba392c..5a2cf5f91ccb 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c
@@ -36,6 +36,7 @@
 #define MAX_PSTATE_SHIFT	32
 #define LPSTATE_SHIFT		48
 #define GPSTATE_SHIFT		56
+#define MAX_NR_CHIPS		32
 
 #define MAX_RAMP_DOWN_TIME				5120
 /*
@@ -1046,12 +1047,20 @@ static int init_chip_info(void)
 	unsigned int *chip;
 	unsigned int cpu, i;
 	unsigned int prev_chip_id = UINT_MAX;
+	cpumask_t *chip_cpu_mask;
 	int ret = 0;
 
 	chip = kcalloc(num_possible_cpus(), sizeof(*chip), GFP_KERNEL);
 	if (!chip)
 		return -ENOMEM;
 
+	/* Allocate a chip cpu mask large enough to fit mask for all chips */
+	chip_cpu_mask = kcalloc(MAX_NR_CHIPS, sizeof(cpumask_t), GFP_KERNEL);
+	if (!chip_cpu_mask) {
+		ret = -ENOMEM;
+		goto free_and_return;
+	}
+
 	for_each_possible_cpu(cpu) {
 		unsigned int id = cpu_to_chip_id(cpu);
 
@@ -1059,22 +1068,25 @@ static int init_chip_info(void)
 			prev_chip_id = id;
 			chip[nr_chips++] = id;
 		}
+		cpumask_set_cpu(cpu, &chip_cpu_mask[nr_chips-1]);
 	}
 
 	chips = kcalloc(nr_chips, sizeof(struct chip), GFP_KERNEL);
 	if (!chips) {
 		ret = -ENOMEM;
-		goto free_and_return;
+		goto out_free_chip_cpu_mask;
 	}
 
 	for (i = 0; i < nr_chips; i++) {
 		chips[i].id = chip[i];
-		cpumask_copy(&chips[i].mask, cpumask_of_node(chip[i]));
+		cpumask_copy(&chips[i].mask, &chip_cpu_mask[i]);
 		INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
 		for_each_cpu(cpu, &chips[i].mask)
 			per_cpu(chip_info, cpu) =  &chips[i];
 	}
 
+out_free_chip_cpu_mask:
+	kfree(chip_cpu_mask);
 free_and_return:
 	kfree(chip);
 	return ret;
diff --git a/drivers/cpuidle/cpuidle-pseries.c b/drivers/cpuidle/cpuidle-pseries.c
index a2b5c6f60cf0..7e7ab5597d7a 100644
--- a/drivers/cpuidle/cpuidle-pseries.c
+++ b/drivers/cpuidle/cpuidle-pseries.c
@@ -346,11 +346,9 @@ static int pseries_cpuidle_driver_init(void)
 static void __init fixup_cede0_latency(void)
 {
 	struct xcede_latency_payload *payload;
-	u64 min_latency_us;
+	u64 min_xcede_latency_us = UINT_MAX;
 	int i;
 
-	min_latency_us = dedicated_states[1].exit_latency; // CEDE latency
-
 	if (parse_cede_parameters())
 		return;
 
@@ -358,42 +356,45 @@ static void __init fixup_cede0_latency(void)
 		nr_xcede_records);
 
 	payload = &xcede_latency_parameter.payload;
+
+	/*
+	 * The CEDE idle state maps to CEDE(0). While the hypervisor
+	 * does not advertise CEDE(0) exit latency values, it does
+	 * advertise the latency values of the extended CEDE states.
+	 * We use the lowest advertised exit latency value as a proxy
+	 * for the exit latency of CEDE(0).
+	 */
 	for (i = 0; i < nr_xcede_records; i++) {
 		struct xcede_latency_record *record = &payload->records[i];
+		u8 hint = record->hint;
 		u64 latency_tb = be64_to_cpu(record->latency_ticks);
 		u64 latency_us = DIV_ROUND_UP_ULL(tb_to_ns(latency_tb), NSEC_PER_USEC);
 
-		if (latency_us == 0)
-			pr_warn("cpuidle: xcede record %d has an unrealistic latency of 0us.\n", i);
-
-		if (latency_us < min_latency_us)
-			min_latency_us = latency_us;
-	}
-
-	/*
-	 * By default, we assume that CEDE(0) has exit latency 10us,
-	 * since there is no way for us to query from the platform.
-	 *
-	 * However, if the wakeup latency of an Extended CEDE state is
-	 * smaller than 10us, then we can be sure that CEDE(0)
-	 * requires no more than that.
-	 *
-	 * Perform the fix-up.
-	 */
-	if (min_latency_us < dedicated_states[1].exit_latency) {
 		/*
-		 * We set a minimum of 1us wakeup latency for cede0 to
-		 * distinguish it from snooze
+		 * We expect the exit latency of an extended CEDE
+		 * state to be non-zero, it to since it takes at least
+		 * a few nanoseconds to wakeup the idle CPU and
+		 * dispatch the virtual processor into the Linux
+		 * Guest.
+		 *
+		 * So we consider only non-zero value for performing
+		 * the fixup of CEDE(0) latency.
 		 */
-		u64 cede0_latency = 1;
+		if (latency_us == 0) {
+			pr_warn("cpuidle: Skipping xcede record %d [hint=%d]. Exit latency = 0us\n",
+				i, hint);
+			continue;
+		}
 
-		if (min_latency_us > cede0_latency)
-			cede0_latency = min_latency_us - 1;
+		if (latency_us < min_xcede_latency_us)
+			min_xcede_latency_us = latency_us;
+	}
 
-		dedicated_states[1].exit_latency = cede0_latency;
-		dedicated_states[1].target_residency = 10 * (cede0_latency);
+	if (min_xcede_latency_us != UINT_MAX) {
+		dedicated_states[1].exit_latency = min_xcede_latency_us;
+		dedicated_states[1].target_residency = 10 * (min_xcede_latency_us);
 		pr_info("cpuidle: Fixed up CEDE exit latency to %llu us\n",
-			cede0_latency);
+			min_xcede_latency_us);
 	}
 
 }
@@ -402,7 +403,7 @@ static void __init fixup_cede0_latency(void)
  * pseries_idle_probe()
  * Choose state table for shared versus dedicated partition
  */
-static int pseries_idle_probe(void)
+static int __init pseries_idle_probe(void)
 {
 
 	if (cpuidle_disable != IDLE_NO_OVERRIDE)
@@ -419,7 +420,21 @@ static int pseries_idle_probe(void)
 			cpuidle_state_table = shared_states;
 			max_idle_state = ARRAY_SIZE(shared_states);
 		} else {
-			fixup_cede0_latency();
+			/*
+			 * Use firmware provided latency values
+			 * starting with POWER10 platforms. In the
+			 * case that we are running on a POWER10
+			 * platform but in an earlier compat mode, we
+			 * can still use the firmware provided values.
+			 *
+			 * However, on platforms prior to POWER10, we
+			 * cannot rely on the accuracy of the firmware
+			 * provided latency values. On such platforms,
+			 * go with the conservative default estimate
+			 * of 10us.
+			 */
+			if (cpu_has_feature(CPU_FTR_ARCH_31) || pvr_version_is(PVR_POWER10))
+				fixup_cede0_latency();
 			cpuidle_state_table = dedicated_states;
 			max_idle_state = NR_DEDICATED_STATES;
 		}