[S390] Vertical cpu management.

If vertical cpu polarization is active then the hypervisor will
dispatch certain cpus for a longer time than other cpus for maximum
performance. For example if a guest would have three virtual cpus,
each of them with a share of 33 percent, then in case of vertical
cpu polarization all of the processing time would be combined to a
single cpu which would run all the time, while the other two cpus
would get nearly no cpu time.

There are three different types of vertical cpus: high, medium and
low. Low cpus hardly get any real cpu time, while high cpus get a
full real cpu. Medium cpus get something in between.

In order to switch between the two possible modes (default is
horizontal) a 0 for horizontal polarization or a 1 for vertical
polarization must be written to the dispatching sysfs attribute:

/sys/devices/system/cpu/dispatching

The polarization of each single cpu can be figured out by the
polarization sysfs attribute of each cpu:

/sys/devices/system/cpu/cpuX/polarization

horizontal, vertical:high, vertical:medium, vertical:low or unknown.

When switching polarization the polarization attribute may contain
the value unknown until the configuration change is done and the
kernel has figured out the new polarization of each cpu.

Note that running a system with different types of vertical cpus may
result in significant performance regressions. If possible only one
type of vertical cpus should be used. All other cpus should be
offlined.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>

2 files changed, 138 insertions, 17 deletions

diff --git a/arch/s390/kernel/smp.c b/arch/s390/kernel/smp.c
index 5448aa8..d1e8e8a 100644
--- a/arch/s390/kernel/smp.c
+++ b/arch/s390/kernel/smp.c
@@ -68,7 +68,9 @@ enum s390_cpu_state {
 };
 
 DEFINE_MUTEX(smp_cpu_state_mutex);
+int smp_cpu_polarization[NR_CPUS];
 static int smp_cpu_state[NR_CPUS];
+static int cpu_management;
 
 static DEFINE_PER_CPU(struct cpu, cpu_devices);
 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
@@ -454,6 +456,7 @@ static int smp_rescan_cpus_sigp(cpumask_t avail)
 		if (cpu_known(cpu_id))
 			continue;
 		__cpu_logical_map[logical_cpu] = cpu_id;
+		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 		if (!cpu_stopped(logical_cpu))
 			continue;
 		cpu_set(logical_cpu, cpu_present_map);
@@ -487,6 +490,7 @@ static int smp_rescan_cpus_sclp(cpumask_t avail)
 		if (cpu_known(cpu_id))
 			continue;
 		__cpu_logical_map[logical_cpu] = cpu_id;
+		smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 		cpu_set(logical_cpu, cpu_present_map);
 		if (cpu >= info->configured)
 			smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
@@ -844,6 +848,7 @@ void __init smp_prepare_boot_cpu(void)
 	S390_lowcore.percpu_offset = __per_cpu_offset[0];
 	current_set[0] = current;
 	smp_cpu_state[0] = CPU_STATE_CONFIGURED;
+	smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 	spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 }
 
@@ -895,15 +900,19 @@ static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
 	case 0:
 		if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 			rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
-			if (!rc)
+			if (!rc) {
 				smp_cpu_state[cpu] = CPU_STATE_STANDBY;
+				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
+			}
 		}
 		break;
 	case 1:
 		if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 			rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
-			if (!rc)
+			if (!rc) {
 				smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
+				smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
+			}
 		}
 		break;
 	default:
@@ -917,6 +926,34 @@ out:
 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 #endif /* CONFIG_HOTPLUG_CPU */
 
+static ssize_t cpu_polarization_show(struct sys_device *dev, char *buf)
+{
+	int cpu = dev->id;
+	ssize_t count;
+
+	mutex_lock(&smp_cpu_state_mutex);
+	switch (smp_cpu_polarization[cpu]) {
+	case POLARIZATION_HRZ:
+		count = sprintf(buf, "horizontal\n");
+		break;
+	case POLARIZATION_VL:
+		count = sprintf(buf, "vertical:low\n");
+		break;
+	case POLARIZATION_VM:
+		count = sprintf(buf, "vertical:medium\n");
+		break;
+	case POLARIZATION_VH:
+		count = sprintf(buf, "vertical:high\n");
+		break;
+	default:
+		count = sprintf(buf, "unknown\n");
+		break;
+	}
+	mutex_unlock(&smp_cpu_state_mutex);
+	return count;
+}
+static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
+
 static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
 {
 	return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
@@ -929,6 +966,7 @@ static struct attribute *cpu_common_attrs[] = {
 	&attr_configure.attr,
 #endif
 	&attr_address.attr,
+	&attr_polarization.attr,
 	NULL,
 };
 
@@ -1073,11 +1111,48 @@ static ssize_t __ref rescan_store(struct sys_device *dev,
 out:
 	put_online_cpus();
 	mutex_unlock(&smp_cpu_state_mutex);
+	if (!cpus_empty(newcpus))
+		topology_schedule_update();
 	return rc ? rc : count;
 }
 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
 #endif /* CONFIG_HOTPLUG_CPU */
 
+static ssize_t dispatching_show(struct sys_device *dev, char *buf)
+{
+	ssize_t count;
+
+	mutex_lock(&smp_cpu_state_mutex);
+	count = sprintf(buf, "%d\n", cpu_management);
+	mutex_unlock(&smp_cpu_state_mutex);
+	return count;
+}
+
+static ssize_t dispatching_store(struct sys_device *dev, const char *buf,
+				 size_t count)
+{
+	int val, rc;
+	char delim;
+
+	if (sscanf(buf, "%d %c", &val, &delim) != 1)
+		return -EINVAL;
+	if (val != 0 && val != 1)
+		return -EINVAL;
+	rc = 0;
+	mutex_lock(&smp_cpu_state_mutex);
+	get_online_cpus();
+	if (cpu_management == val)
+		goto out;
+	rc = topology_set_cpu_management(val);
+	if (!rc)
+		cpu_management = val;
+out:
+	put_online_cpus();
+	mutex_unlock(&smp_cpu_state_mutex);
+	return rc ? rc : count;
+}
+static SYSDEV_ATTR(dispatching, 0644, dispatching_show, dispatching_store);
+
 static int __init topology_init(void)
 {
 	int cpu;
@@ -1091,6 +1166,10 @@ static int __init topology_init(void)
 	if (rc)
 		return rc;
 #endif
+	rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
+			       &attr_dispatching.attr);
+	if (rc)
+		return rc;
 	for_each_present_cpu(cpu) {
 		rc = smp_add_present_cpu(cpu);
 		if (rc)
diff --git a/arch/s390/kernel/topology.c b/arch/s390/kernel/topology.c
index 369dc1c..12b39b3 100644
--- a/arch/s390/kernel/topology.c
+++ b/arch/s390/kernel/topology.c
@@ -1,6 +1,4 @@
 /*
- *  arch/s390/kernel/topology.c
- *
  *    Copyright IBM Corp. 2007
  *    Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
  */
@@ -19,9 +17,17 @@
 #include <asm/sysinfo.h>
 
 #define CPU_BITS 64
+#define NR_MAG 6
+
+#define PTF_HORIZONTAL	(0UL)
+#define PTF_VERTICAL	(1UL)
+#define PTF_CHECK	(2UL)
 
 struct tl_cpu {
-	unsigned char reserved[6];
+	unsigned char reserved0[4];
+	unsigned char :6;
+	unsigned char pp:2;
+	unsigned char reserved1;
 	unsigned short origin;
 	unsigned long mask[CPU_BITS / BITS_PER_LONG];
 };
@@ -36,8 +42,6 @@ union tl_entry {
 	struct tl_container container;
 };
 
-#define NR_MAG 6
-
 struct tl_info {
 	unsigned char reserved0[2];
 	unsigned short length;
@@ -96,8 +100,10 @@ static void add_cpus_to_core(struct tl_cpu *tl_cpu, struct core_info *core)
 
 		rcpu = CPU_BITS - 1 - cpu + tl_cpu->origin;
 		for_each_present_cpu(lcpu) {
-			if (__cpu_logical_map[lcpu] == rcpu)
+			if (__cpu_logical_map[lcpu] == rcpu) {
 				cpu_set(lcpu, core->mask);
+				smp_cpu_polarization[lcpu] = tl_cpu->pp;
+			}
 		}
 	}
 }
@@ -127,7 +133,7 @@ static void tl_to_cores(struct tl_info *info)
 
 	mutex_lock(&smp_cpu_state_mutex);
 	clear_cores();
-	tle = (union tl_entry *)&info->tle;
+	tle = info->tle;
 	end = (union tl_entry *)((unsigned long)info + info->length);
 	while (tle < end) {
 		switch (tle->nl) {
@@ -152,7 +158,17 @@ static void tl_to_cores(struct tl_info *info)
 	mutex_unlock(&smp_cpu_state_mutex);
 }
 
-static int ptf(void)
+static void topology_update_polarization_simple(void)
+{
+	int cpu;
+
+	mutex_lock(&smp_cpu_state_mutex);
+	for_each_present_cpu(cpu)
+		smp_cpu_polarization[cpu] = POLARIZATION_HRZ;
+	mutex_unlock(&smp_cpu_state_mutex);
+}
+
+static int ptf(unsigned long fc)
 {
 	int rc;
 
@@ -161,7 +177,25 @@ static int ptf(void)
 		"	ipm	%0\n"
 		"	srl	%0,28\n"
 		: "=d" (rc)
-		: "d" (2UL)  : "cc");
+		: "d" (fc)  : "cc");
+	return rc;
+}
+
+int topology_set_cpu_management(int fc)
+{
+	int cpu;
+	int rc;
+
+	if (!machine_has_topology)
+		return -EOPNOTSUPP;
+	if (fc)
+		rc = ptf(PTF_VERTICAL);
+	else
+		rc = ptf(PTF_HORIZONTAL);
+	if (rc)
+		return -EBUSY;
+	for_each_present_cpu(cpu)
+		smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 	return rc;
 }
 
@@ -171,9 +205,10 @@ void arch_update_cpu_topology(void)
 	struct sys_device *sysdev;
 	int cpu;
 
-	if (!machine_has_topology)
+	if (!machine_has_topology) {
+		topology_update_polarization_simple();
 		return;
-	ptf();
+	}
 	stsi(info, 15, 1, 2);
 	tl_to_cores(info);
 	for_each_online_cpu(cpu) {
@@ -187,10 +222,15 @@ static void topology_work_fn(struct work_struct *work)
 	arch_reinit_sched_domains();
 }
 
+void topology_schedule_update(void)
+{
+	schedule_work(&topology_work);
+}
+
 static void topology_timer_fn(unsigned long ignored)
 {
-	if (ptf())
-		schedule_work(&topology_work);
+	if (ptf(PTF_CHECK))
+		topology_schedule_update();
 	set_topology_timer();
 }
 
@@ -211,9 +251,11 @@ static int __init init_topology_update(void)
 {
 	int rc;
 
-	if (!machine_has_topology)
+	if (!machine_has_topology) {
+		topology_update_polarization_simple();
 		return 0;
-	init_timer(&topology_timer);
+	}
+	init_timer_deferrable(&topology_timer);
 	if (machine_has_topology_irq) {
 		rc = register_external_interrupt(0x2005, topology_interrupt);
 		if (rc)