perf/x86/intel/rapl: Convert it to a per package facility
authorThomas Gleixner <tglx@linutronix.de>
Mon, 22 Feb 2016 22:19:26 +0000 (22:19 +0000)
committerIngo Molnar <mingo@kernel.org>
Mon, 29 Feb 2016 08:35:26 +0000 (09:35 +0100)
RAPL is a per package facility and we already have a mechanism for a dedicated
per package reader. So there is no point to have multiple CPUs doing the
same. The current implementation actually starts two timers on two CPUs if one
does:

perf stat -C1,2 -e -e power/energy-pkg ....

which makes the whole concept of 1 reader per package moot.

What's worse is that the above returns the double of the actual energy
consumption, but that's a different problem to address and cannot be solved by
removing the pointless per cpuness of that mechanism.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.845369524@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/events/intel/rapl.c

index 753d90c..019e541 100644 (file)
@@ -129,15 +129,23 @@ struct rapl_pmu {
        struct hrtimer          hrtimer;
 };
 
+struct rapl_pmus {
+       struct pmu              pmu;
+       unsigned int            maxpkg;
+       struct rapl_pmu         *pmus[];
+};
+
  /* 1/2^hw_unit Joule */
 static int rapl_hw_unit[NR_RAPL_DOMAINS] __read_mostly;
-static struct pmu rapl_pmu_class;
+static struct rapl_pmus *rapl_pmus;
 static cpumask_t rapl_cpu_mask;
-static int rapl_cntr_mask;
+static unsigned int rapl_cntr_mask;
 static u64 rapl_timer_ms;
 
-static DEFINE_PER_CPU(struct rapl_pmu *, rapl_pmu);
-static DEFINE_PER_CPU(struct rapl_pmu *, rapl_pmu_to_free);
+static inline struct rapl_pmu *cpu_to_rapl_pmu(unsigned int cpu)
+{
+       return rapl_pmus->pmus[topology_logical_package_id(cpu)];
+}
 
 static inline u64 rapl_read_counter(struct perf_event *event)
 {
@@ -317,12 +325,12 @@ static void rapl_pmu_event_del(struct perf_event *event, int flags)
 
 static int rapl_pmu_event_init(struct perf_event *event)
 {
-       struct rapl_pmu *pmu = __this_cpu_read(rapl_pmu);
        u64 cfg = event->attr.config & RAPL_EVENT_MASK;
        int bit, msr, ret = 0;
+       struct rapl_pmu *pmu;
 
        /* only look at RAPL events */
-       if (event->attr.type != rapl_pmu_class.type)
+       if (event->attr.type != rapl_pmus->pmu.type)
                return -ENOENT;
 
        /* check only supported bits are set */
@@ -370,6 +378,7 @@ static int rapl_pmu_event_init(struct perf_event *event)
                return -EINVAL;
 
        /* must be done before validate_group */
+       pmu = cpu_to_rapl_pmu(event->cpu);
        event->cpu = pmu->cpu;
        event->pmu_private = pmu;
        event->hw.event_base = msr;
@@ -502,116 +511,62 @@ const struct attribute_group *rapl_attr_groups[] = {
        NULL,
 };
 
-static struct pmu rapl_pmu_class = {
-       .attr_groups    = rapl_attr_groups,
-       .task_ctx_nr    = perf_invalid_context, /* system-wide only */
-       .event_init     = rapl_pmu_event_init,
-       .add            = rapl_pmu_event_add, /* must have */
-       .del            = rapl_pmu_event_del, /* must have */
-       .start          = rapl_pmu_event_start,
-       .stop           = rapl_pmu_event_stop,
-       .read           = rapl_pmu_event_read,
-};
-
 static void rapl_cpu_exit(int cpu)
 {
-       struct rapl_pmu *pmu = per_cpu(rapl_pmu, cpu);
-       int i, phys_id = topology_physical_package_id(cpu);
-       int target = -1;
+       struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu);
+       int target;
 
-       /* find a new cpu on same package */
-       for_each_online_cpu(i) {
-               if (i == cpu)
-                       continue;
-               if (phys_id == topology_physical_package_id(i)) {
-                       target = i;
-                       break;
-               }
-       }
-       /*
-        * clear cpu from cpumask
-        * if was set in cpumask and still some cpu on package,
-        * then move to new cpu
-        */
-       if (cpumask_test_and_clear_cpu(cpu, &rapl_cpu_mask) && target >= 0)
-               cpumask_set_cpu(target, &rapl_cpu_mask);
+       /* Check if exiting cpu is used for collecting rapl events */
+       if (!cpumask_test_and_clear_cpu(cpu, &rapl_cpu_mask))
+               return;
 
-       WARN_ON(cpumask_empty(&rapl_cpu_mask));
-       /*
-        * migrate events and context to new cpu
-        */
-       if (target >= 0)
-               perf_pmu_migrate_context(pmu->pmu, cpu, target);
+       pmu->cpu = -1;
+       /* Find a new cpu to collect rapl events */
+       target = cpumask_any_but(topology_core_cpumask(cpu), cpu);
 
-       /* cancel overflow polling timer for CPU */
-       hrtimer_cancel(&pmu->hrtimer);
+       /* Migrate rapl events to the new target */
+       if (target < nr_cpu_ids) {
+               cpumask_set_cpu(target, &rapl_cpu_mask);
+               pmu->cpu = target;
+               perf_pmu_migrate_context(pmu->pmu, cpu, target);
+       }
 }
 
 static void rapl_cpu_init(int cpu)
 {
-       int i, phys_id = topology_physical_package_id(cpu);
+       struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu);
+       int target;
+
+       /*
+        * Check if there is an online cpu in the package which collects rapl
+        * events already.
+        */
+       target = cpumask_any_and(&rapl_cpu_mask, topology_core_cpumask(cpu));
+       if (target < nr_cpu_ids)
+               return;
 
-       /* check if phys_is is already covered */
-       for_each_cpu(i, &rapl_cpu_mask) {
-               if (phys_id == topology_physical_package_id(i))
-                       return;
-       }
-       /* was not found, so add it */
        cpumask_set_cpu(cpu, &rapl_cpu_mask);
+       pmu->cpu = cpu;
 }
 
 static int rapl_cpu_prepare(int cpu)
 {
-       struct rapl_pmu *pmu = per_cpu(rapl_pmu, cpu);
-       int phys_id = topology_physical_package_id(cpu);
+       struct rapl_pmu *pmu = cpu_to_rapl_pmu(cpu);
 
        if (pmu)
                return 0;
 
-       if (phys_id < 0)
-               return -1;
-
        pmu = kzalloc_node(sizeof(*pmu), GFP_KERNEL, cpu_to_node(cpu));
        if (!pmu)
-               return -1;
-       raw_spin_lock_init(&pmu->lock);
+               return -ENOMEM;
 
+       raw_spin_lock_init(&pmu->lock);
        INIT_LIST_HEAD(&pmu->active_list);
-
-       pmu->pmu = &rapl_pmu_class;
-       pmu->cpu = cpu;
-
+       pmu->pmu = &rapl_pmus->pmu;
        pmu->timer_interval = ms_to_ktime(rapl_timer_ms);
-
+       pmu->cpu = -1;
        rapl_hrtimer_init(pmu);
-
-       /* set RAPL pmu for this cpu for now */
-       per_cpu(rapl_pmu, cpu) = pmu;
-       per_cpu(rapl_pmu_to_free, cpu) = NULL;
-
-       return 0;
-}
-
-static void rapl_cpu_kfree(int cpu)
-{
-       struct rapl_pmu *pmu = per_cpu(rapl_pmu_to_free, cpu);
-
-       kfree(pmu);
-
-       per_cpu(rapl_pmu_to_free, cpu) = NULL;
-}
-
-static int rapl_cpu_dying(int cpu)
-{
-       struct rapl_pmu *pmu = per_cpu(rapl_pmu, cpu);
-
-       if (!pmu)
-               return 0;
-
-       per_cpu(rapl_pmu, cpu) = NULL;
-
-       per_cpu(rapl_pmu_to_free, cpu) = pmu;
-
+       rapl_pmus->pmus[topology_logical_package_id(cpu)] = pmu;
        return 0;
 }
 
@@ -624,24 +579,16 @@ static int rapl_cpu_notifier(struct notifier_block *self,
        case CPU_UP_PREPARE:
                rapl_cpu_prepare(cpu);
                break;
-       case CPU_STARTING:
-               rapl_cpu_init(cpu);
-               break;
-       case CPU_UP_CANCELED:
-       case CPU_DYING:
-               rapl_cpu_dying(cpu);
-               break;
+
+       case CPU_DOWN_FAILED:
        case CPU_ONLINE:
-       case CPU_DEAD:
-               rapl_cpu_kfree(cpu);
+               rapl_cpu_init(cpu);
                break;
+
        case CPU_DOWN_PREPARE:
                rapl_cpu_exit(cpu);
                break;
-       default:
-               break;
        }
-
        return NOTIFY_OK;
 }
 
@@ -703,10 +650,14 @@ static void __init rapl_advertise(void)
 
 static int __init rapl_prepare_cpus(void)
 {
-       unsigned int cpu;
+       unsigned int cpu, pkg;
        int ret;
 
        for_each_online_cpu(cpu) {
+               pkg = topology_logical_package_id(cpu);
+               if (rapl_pmus->pmus[pkg])
+                       continue;
+
                ret = rapl_cpu_prepare(cpu);
                if (ret)
                        return ret;
@@ -717,10 +668,33 @@ static int __init rapl_prepare_cpus(void)
 
 static void __init cleanup_rapl_pmus(void)
 {
-       int cpu;
+       int i;
+
+       for (i = 0; i < rapl_pmus->maxpkg; i++)
+               kfree(rapl_pmus->pmus + i);
+       kfree(rapl_pmus);
+}
 
-       for_each_online_cpu(cpu)
-               kfree(per_cpu(rapl_pmu, cpu));
+static int __init init_rapl_pmus(void)
+{
+       int maxpkg = topology_max_packages();
+       size_t size;
+
+       size = sizeof(*rapl_pmus) + maxpkg * sizeof(struct rapl_pmu *);
+       rapl_pmus = kzalloc(size, GFP_KERNEL);
+       if (!rapl_pmus)
+               return -ENOMEM;
+
+       rapl_pmus->maxpkg               = maxpkg;
+       rapl_pmus->pmu.attr_groups      = rapl_attr_groups;
+       rapl_pmus->pmu.task_ctx_nr      = perf_invalid_context;
+       rapl_pmus->pmu.event_init       = rapl_pmu_event_init;
+       rapl_pmus->pmu.add              = rapl_pmu_event_add;
+       rapl_pmus->pmu.del              = rapl_pmu_event_del;
+       rapl_pmus->pmu.start            = rapl_pmu_event_start;
+       rapl_pmus->pmu.stop             = rapl_pmu_event_stop;
+       rapl_pmus->pmu.read             = rapl_pmu_event_read;
+       return 0;
 }
 
 static const struct x86_cpu_id rapl_cpu_match[] __initconst = {
@@ -771,13 +745,17 @@ static int __init rapl_pmu_init(void)
        if (ret)
                return ret;
 
+       ret = init_rapl_pmus();
+       if (ret)
+               return ret;
+
        cpu_notifier_register_begin();
 
        ret = rapl_prepare_cpus();
        if (ret)
                goto out;
 
-       ret = perf_pmu_register(&rapl_pmu_class, "power", -1);
+       ret = perf_pmu_register(&rapl_pmus->pmu, "power", -1);
        if (ret)
                goto out;