[PATCH v4 7/7] powerpc/powernv: nest pmu cpumask and cpu hotplug support

[Madhavan Srinivasan]

Adds cpumask attribute to be used by each nest pmu since nest
units are per-chip. Only one cpu (first online cpu) from each node/chip
is designated to read counters.

On cpu hotplug, dying cpu is checked to see whether it is one of the
designated cpus, if yes, next online cpu from the same node/chip is
designated as new cpu to read counters.

Cc: Michael Ellerman <mpe at ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh at kernel.crashing.org>
Cc: Paul Mackerras <paulus at samba.org>
Cc: Anton Blanchard <anton at samba.org>
Cc: Sukadev Bhattiprolu <sukadev at linux.vnet.ibm.com>
Cc: Anshuman Khandual <khandual at linux.vnet.ibm.com>
Cc: Stephane Eranian <eranian at google.com>
Cc: Preeti U Murthy <preetium at andrew.cmu.edu>
Cc: Ingo Molnar <mingo at kernel.org>
Cc: Peter Zijlstra <peterz at infradead.org>
Signed-off-by: Madhavan Srinivasan <maddy at linux.vnet.ibm.com>
---
 arch/powerpc/perf/nest-pmu.c | 146 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 146 insertions(+)

diff --git a/arch/powerpc/perf/nest-pmu.c b/arch/powerpc/perf/nest-pmu.c
index c2ada13..31943c5 100644
--- a/arch/powerpc/perf/nest-pmu.c
+++ b/arch/powerpc/perf/nest-pmu.c
@@ -12,6 +12,7 @@
 
 static struct perchip_nest_info p8_nest_perchip_info[P8_NEST_MAX_CHIPS];
 static struct nest_pmu *per_nest_pmu_arr[P8_NEST_MAX_PMUS];
+static cpumask_t nest_pmu_cpu_mask;
 
 PMU_FORMAT_ATTR(event, "config:0-20");
 struct attribute *p8_nest_format_attrs[] = {
@@ -24,6 +25,147 @@ struct attribute_group p8_nest_format_group = {
 	.attrs = p8_nest_format_attrs,
 };
 
+static ssize_t nest_pmu_cpumask_get_attr(struct device *dev,
+				struct device_attribute *attr, char *buf)
+{
+	return cpumap_print_to_pagebuf(true, buf, &nest_pmu_cpu_mask);
+}
+
+static DEVICE_ATTR(cpumask, S_IRUGO, nest_pmu_cpumask_get_attr, NULL);
+
+static struct attribute *nest_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group nest_pmu_cpumask_attr_group = {
+	.attrs = nest_pmu_cpumask_attrs,
+};
+
+static void nest_init(void *dummy)
+{
+	opal_nest_ima_control(P8_NEST_ENGINE_START);
+}
+
+static void nest_change_cpu_context(int old_cpu, int new_cpu)
+{
+	int i;
+
+	for (i = 0; per_nest_pmu_arr[i] != NULL; i++)
+		perf_pmu_migrate_context(&per_nest_pmu_arr[i]->pmu,
+						old_cpu, new_cpu);
+}
+
+static void nest_exit_cpu(int cpu)
+{
+	int nid, target = -1;
+	struct cpumask *l_cpumask;
+
+	/*
+	 * Check in the designated list for this cpu. Dont bother
+	 * if not one of them.
+	 */
+	if (!cpumask_test_and_clear_cpu(cpu, &nest_pmu_cpu_mask))
+		return;
+
+	/*
+	 * Now that this cpu is one of the designated,
+	 * find a next cpu a) which is online and b) in same chip.
+	 */
+	nid = cpu_to_node(cpu);
+	l_cpumask = cpumask_of_node(nid);
+	target = cpumask_next(cpu, l_cpumask);
+
+	/*
+	 * Update the cpumask with the target cpu and
+	 * migrate the context if needed
+	 */
+	if (target >= 0 && target <= nr_cpu_ids) {
+		cpumask_set_cpu(target, &nest_pmu_cpu_mask);
+		nest_change_cpu_context(cpu, target);
+	}
+}
+
+static void nest_init_cpu(int cpu)
+{
+	int nid, fcpu, ncpu;
+	struct cpumask *l_cpumask, tmp_mask;
+
+	nid = cpu_to_node(cpu);
+	l_cpumask = cpumask_of_node(nid);
+
+	/*
+	 * if empty cpumask, just add incoming cpu and move on.
+	 */
+	if (!cpumask_and(&tmp_mask, l_cpumask, &nest_pmu_cpu_mask)) {
+		cpumask_set_cpu(cpu, &nest_pmu_cpu_mask);
+		return;
+	}
+
+	/*
+	 * Alway have the first online cpu of a chip as designated one.
+	 */
+	fcpu = cpumask_first(l_cpumask);
+	ncpu = cpumask_next(cpu, l_cpumask);
+	if (cpu == fcpu) {
+		if (cpumask_test_and_clear_cpu(ncpu, &nest_pmu_cpu_mask)) {
+			cpumask_set_cpu(cpu, &nest_pmu_cpu_mask);
+			nest_change_cpu_context(ncpu, cpu);
+		}
+	}
+}
+
+static int nest_pmu_cpu_notifier(struct notifier_block *self,
+				unsigned long action, void *hcpu)
+{
+	long cpu = (long)hcpu;
+
+	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_ONLINE:
+		nest_init_cpu(cpu);
+		break;
+	case CPU_DOWN_PREPARE:
+	       nest_exit_cpu(cpu);
+	       break;
+	default:
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block nest_pmu_cpu_nb = {
+	.notifier_call  = nest_pmu_cpu_notifier,
+	.priority       = CPU_PRI_PERF + 1,
+};
+
+void nest_pmu_cpumask_init(void)
+{
+	const struct cpumask *l_cpumask;
+	int cpu, nid;
+
+	cpu_notifier_register_begin();
+
+	/*
+	 * Nest PMUs are per-chip counters. So designate a cpu
+	 * from each chip for counter collection.
+	 */
+	for_each_online_node(nid) {
+		l_cpumask = cpumask_of_node(nid);
+
+		/* designate first online cpu in this node */
+		cpu = cpumask_first(l_cpumask);
+		cpumask_set_cpu(cpu, &nest_pmu_cpu_mask);
+	}
+
+	/* Initialize Nest PMUs in each node using designated cpus */
+	on_each_cpu_mask(&nest_pmu_cpu_mask, (smp_call_func_t)nest_init, NULL, 1);
+
+	__register_cpu_notifier(&nest_pmu_cpu_nb);
+
+	cpu_notifier_register_done();
+}
+
 static int p8_nest_event_init(struct perf_event *event)
 {
 	int chip_id;
@@ -239,6 +381,7 @@ static int nest_pmu_create(struct device_node *dev, int pmu_index)
 			sprintf(buf, "Nest_%s", (char *)pp->value);
 			pmu_ptr->pmu.name = (char *)buf;
 			pmu_ptr->attr_groups[1] = &p8_nest_format_group;
+			pmu_ptr->attr_groups[2] = &nest_pmu_cpumask_attr_group;
 			continue;
 		}
 
@@ -363,6 +506,9 @@ static int __init nest_pmu_init(void)
 	if (ret)
 		return ret;
 
+	/* Add cpumask and register for hotplug notification */
+	nest_pmu_cpumask_init();
+
 	return 0;
 }
 device_initcall(nest_pmu_init);
-- 
1.9.1