@@ -39,6 +39,10 @@
* event: rapl_energy_psys
* perf code: 0x5
*
+ * per_core counter: consumption of a single physical core
+ * event: rapl_energy_per_core
+ * perf code: 0x6
+ *
* We manage those counters as free running (read-only). They may be
* use simultaneously by other tools, such as turbostat.
*
@@ -76,6 +80,7 @@ enum perf_rapl_events {
PERF_RAPL_RAM, /* DRAM */
PERF_RAPL_PP1, /* gpu */
PERF_RAPL_PSYS, /* psys */
+ PERF_RAPL_PERCORE, /* per-core */
PERF_RAPL_MAX,
NR_RAPL_DOMAINS = PERF_RAPL_MAX,
@@ -87,6 +92,7 @@ static const char *const rapl_domain_names[NR_RAPL_DOMAINS] __initconst = {
"dram",
"pp1-gpu",
"psys",
+ "per-core",
};
/*
@@ -135,11 +141,13 @@ struct rapl_model {
unsigned long events;
unsigned int msr_power_unit;
enum rapl_unit_quirk unit_quirk;
+ bool per_core;
};
/* 1/2^hw_unit Joule */
static int rapl_hw_unit[NR_RAPL_DOMAINS] __read_mostly;
static struct rapl_pmus *rapl_pmus;
+static struct rapl_pmus *rapl_pmus_per_core;
static unsigned int rapl_cntr_mask;
static u64 rapl_timer_ms;
static struct perf_msr *rapl_msrs;
@@ -345,9 +353,14 @@ static int rapl_pmu_event_init(struct perf_event *event)
u64 cfg = event->attr.config & RAPL_EVENT_MASK;
int bit, ret = 0;
struct rapl_pmu *rapl_pmu;
+ struct rapl_pmus *curr_rapl_pmus;
/* only look at RAPL events */
- if (event->attr.type != rapl_pmus->pmu.type)
+ if (event->attr.type == rapl_pmus->pmu.type)
+ curr_rapl_pmus = rapl_pmus;
+ else if (rapl_pmus_per_core && event->attr.type == rapl_pmus_per_core->pmu.type)
+ curr_rapl_pmus = rapl_pmus_per_core;
+ else
return -ENOENT;
/* check only supported bits are set */
@@ -374,9 +387,14 @@ static int rapl_pmu_event_init(struct perf_event *event)
return -EINVAL;
/* must be done before validate_group */
- rapl_pmu = cpu_to_rapl_pmu(event->cpu);
+ if (curr_rapl_pmus == rapl_pmus_per_core)
+ rapl_pmu = curr_rapl_pmus->rapl_pmu[topology_core_id(event->cpu)];
+ else
+ rapl_pmu = curr_rapl_pmus->rapl_pmu[get_rapl_pmu_idx(event->cpu)];
+
if (!rapl_pmu)
return -EINVAL;
+
event->cpu = rapl_pmu->cpu;
event->pmu_private = rapl_pmu;
event->hw.event_base = rapl_msrs[bit].msr;
@@ -408,17 +426,38 @@ static struct attribute_group rapl_pmu_attr_group = {
.attrs = rapl_pmu_attrs,
};
+static ssize_t rapl_get_attr_per_core_cpumask(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ return cpumap_print_to_pagebuf(true, buf, &rapl_pmus_per_core->cpumask);
+}
+
+static struct device_attribute dev_attr_per_core_cpumask = __ATTR(cpumask, 0444,
+ rapl_get_attr_per_core_cpumask,
+ NULL);
+
+static struct attribute *rapl_pmu_per_core_attrs[] = {
+ &dev_attr_per_core_cpumask.attr,
+ NULL,
+};
+
+static struct attribute_group rapl_pmu_per_core_attr_group = {
+ .attrs = rapl_pmu_per_core_attrs,
+};
+
RAPL_EVENT_ATTR_STR(energy-cores, rapl_cores, "event=0x01");
RAPL_EVENT_ATTR_STR(energy-pkg , rapl_pkg, "event=0x02");
RAPL_EVENT_ATTR_STR(energy-ram , rapl_ram, "event=0x03");
RAPL_EVENT_ATTR_STR(energy-gpu , rapl_gpu, "event=0x04");
RAPL_EVENT_ATTR_STR(energy-psys, rapl_psys, "event=0x05");
+RAPL_EVENT_ATTR_STR(energy-per-core, rapl_per_core, "event=0x06");
RAPL_EVENT_ATTR_STR(energy-cores.unit, rapl_cores_unit, "Joules");
RAPL_EVENT_ATTR_STR(energy-pkg.unit , rapl_pkg_unit, "Joules");
RAPL_EVENT_ATTR_STR(energy-ram.unit , rapl_ram_unit, "Joules");
RAPL_EVENT_ATTR_STR(energy-gpu.unit , rapl_gpu_unit, "Joules");
RAPL_EVENT_ATTR_STR(energy-psys.unit, rapl_psys_unit, "Joules");
+RAPL_EVENT_ATTR_STR(energy-per-core.unit, rapl_per_core_unit, "Joules");
/*
* we compute in 0.23 nJ increments regardless of MSR
@@ -428,6 +467,7 @@ RAPL_EVENT_ATTR_STR(energy-pkg.scale, rapl_pkg_scale, "2.3283064365386962890
RAPL_EVENT_ATTR_STR(energy-ram.scale, rapl_ram_scale, "2.3283064365386962890625e-10");
RAPL_EVENT_ATTR_STR(energy-gpu.scale, rapl_gpu_scale, "2.3283064365386962890625e-10");
RAPL_EVENT_ATTR_STR(energy-psys.scale, rapl_psys_scale, "2.3283064365386962890625e-10");
+RAPL_EVENT_ATTR_STR(energy-per-core.scale, rapl_per_core_scale, "2.3283064365386962890625e-10");
/*
* There are no default events, but we need to create
@@ -461,6 +501,13 @@ static const struct attribute_group *rapl_attr_groups[] = {
NULL,
};
+static const struct attribute_group *rapl_per_core_attr_groups[] = {
+ &rapl_pmu_per_core_attr_group,
+ &rapl_pmu_format_group,
+ &rapl_pmu_events_group,
+ NULL,
+};
+
static struct attribute *rapl_events_cores[] = {
EVENT_PTR(rapl_cores),
EVENT_PTR(rapl_cores_unit),
@@ -521,6 +568,18 @@ static struct attribute_group rapl_events_psys_group = {
.attrs = rapl_events_psys,
};
+static struct attribute *rapl_events_per_core[] = {
+ EVENT_PTR(rapl_per_core),
+ EVENT_PTR(rapl_per_core_unit),
+ EVENT_PTR(rapl_per_core_scale),
+ NULL,
+};
+
+static struct attribute_group rapl_events_per_core_group = {
+ .name = "events",
+ .attrs = rapl_events_per_core,
+};
+
static bool test_msr(int idx, void *data)
{
return test_bit(idx, (unsigned long *) data);
@@ -535,6 +594,7 @@ static struct perf_msr intel_rapl_msrs[] = {
[PERF_RAPL_RAM] = { MSR_DRAM_ENERGY_STATUS, &rapl_events_ram_group, test_msr, false, RAPL_MSR_MASK },
[PERF_RAPL_PP1] = { MSR_PP1_ENERGY_STATUS, &rapl_events_gpu_group, test_msr, false, RAPL_MSR_MASK },
[PERF_RAPL_PSYS] = { MSR_PLATFORM_ENERGY_STATUS, &rapl_events_psys_group, test_msr, false, RAPL_MSR_MASK },
+ [PERF_RAPL_PERCORE] = { 0, &rapl_events_per_core_group, NULL, false, 0 },
};
static struct perf_msr intel_rapl_spr_msrs[] = {
@@ -543,6 +603,7 @@ static struct perf_msr intel_rapl_spr_msrs[] = {
[PERF_RAPL_RAM] = { MSR_DRAM_ENERGY_STATUS, &rapl_events_ram_group, test_msr, false, RAPL_MSR_MASK },
[PERF_RAPL_PP1] = { MSR_PP1_ENERGY_STATUS, &rapl_events_gpu_group, test_msr, false, RAPL_MSR_MASK },
[PERF_RAPL_PSYS] = { MSR_PLATFORM_ENERGY_STATUS, &rapl_events_psys_group, test_msr, true, RAPL_MSR_MASK },
+ [PERF_RAPL_PERCORE] = { 0, &rapl_events_per_core_group, NULL, false, 0 },
};
/*
@@ -556,6 +617,7 @@ static struct perf_msr amd_rapl_msrs[] = {
[PERF_RAPL_RAM] = { 0, &rapl_events_ram_group, NULL, false, 0 },
[PERF_RAPL_PP1] = { 0, &rapl_events_gpu_group, NULL, false, 0 },
[PERF_RAPL_PSYS] = { 0, &rapl_events_psys_group, NULL, false, 0 },
+ [PERF_RAPL_PERCORE] = { MSR_AMD_CORE_ENERGY_STATUS, &rapl_events_per_core_group, test_msr, false, RAPL_MSR_MASK },
};
static int __rapl_cpu_offline(struct rapl_pmus *rapl_pmus, unsigned int rapl_pmu_idx,
@@ -583,8 +645,16 @@ static int __rapl_cpu_offline(struct rapl_pmus *rapl_pmus, unsigned int rapl_pmu
static int rapl_cpu_offline(unsigned int cpu)
{
- return __rapl_cpu_offline(rapl_pmus, get_rapl_pmu_idx(cpu),
- get_rapl_pmu_cpumask(cpu), cpu);
+ int ret;
+
+ ret = __rapl_cpu_offline(rapl_pmus, get_rapl_pmu_idx(cpu),
+ get_rapl_pmu_cpumask(cpu), cpu);
+
+ if (ret == 0 && rapl_model->per_core)
+ ret = __rapl_cpu_offline(rapl_pmus_per_core, topology_core_id(cpu),
+ topology_sibling_cpumask(cpu), cpu);
+
+ return ret;
}
static int __rapl_cpu_online(struct rapl_pmus *rapl_pmus, unsigned int rapl_pmu_idx,
@@ -622,10 +692,17 @@ static int __rapl_cpu_online(struct rapl_pmus *rapl_pmus, unsigned int rapl_pmu_
static int rapl_cpu_online(unsigned int cpu)
{
- return __rapl_cpu_online(rapl_pmus, get_rapl_pmu_idx(cpu),
- get_rapl_pmu_cpumask(cpu), cpu);
-}
+ int ret;
+
+ ret = __rapl_cpu_online(rapl_pmus, get_rapl_pmu_idx(cpu),
+ get_rapl_pmu_cpumask(cpu), cpu);
+ if (ret == 0 && rapl_model->per_core)
+ ret = __rapl_cpu_online(rapl_pmus_per_core, topology_core_id(cpu),
+ topology_sibling_cpumask(cpu), cpu);
+
+ return ret;
+}
static int rapl_check_hw_unit(void)
{
@@ -687,7 +764,7 @@ static void __init rapl_advertise(void)
}
}
-static void cleanup_rapl_pmus(void)
+static void cleanup_rapl_pmus(struct rapl_pmus *rapl_pmus)
{
int i;
@@ -705,12 +782,15 @@ static const struct attribute_group *rapl_attr_update[] = {
NULL,
};
-static int __init init_rapl_pmus(void)
-{
- int nr_rapl_pmu = topology_max_packages() * topology_max_dies_per_package();
+static const struct attribute_group *rapl_per_core_attr_update[] = {
+ &rapl_events_per_core_group,
+};
- if (rapl_pmu_is_pkg_scope())
- nr_rapl_pmu = topology_max_packages();
+static int __init init_rapl_pmus(struct rapl_pmus **rapl_pmus_ptr, int nr_rapl_pmu,
+ const struct attribute_group **rapl_attr_groups,
+ const struct attribute_group **rapl_attr_update)
+{
+ struct rapl_pmus *rapl_pmus;
rapl_pmus = kzalloc(struct_size(rapl_pmus, rapl_pmu, nr_rapl_pmu), GFP_KERNEL);
if (!rapl_pmus)
@@ -728,6 +808,9 @@ static int __init init_rapl_pmus(void)
rapl_pmus->pmu.read = rapl_pmu_event_read;
rapl_pmus->pmu.module = THIS_MODULE;
rapl_pmus->pmu.capabilities = PERF_PMU_CAP_NO_EXCLUDE;
+
+ *rapl_pmus_ptr = rapl_pmus;
+
return 0;
}
@@ -794,9 +877,11 @@ static struct rapl_model model_spr = {
};
static struct rapl_model model_amd_hygon = {
- .events = BIT(PERF_RAPL_PKG),
+ .events = BIT(PERF_RAPL_PKG) |
+ BIT(PERF_RAPL_PERCORE),
.msr_power_unit = MSR_AMD_RAPL_POWER_UNIT,
.rapl_msrs = amd_rapl_msrs,
+ .per_core = true,
};
static const struct x86_cpu_id rapl_model_match[] __initconst = {
@@ -853,6 +938,11 @@ static int __init rapl_pmu_init(void)
{
const struct x86_cpu_id *id;
int ret;
+ int nr_rapl_pmu = topology_max_packages() * topology_max_dies_per_package();
+ int nr_cores = topology_max_packages() * topology_num_cores_per_package();
+
+ if (rapl_pmu_is_pkg_scope())
+ nr_rapl_pmu = topology_max_packages();
id = x86_match_cpu(rapl_model_match);
if (!id)
@@ -869,10 +959,23 @@ static int __init rapl_pmu_init(void)
if (ret)
return ret;
- ret = init_rapl_pmus();
+ ret = init_rapl_pmus(&rapl_pmus, nr_rapl_pmu, rapl_attr_groups, rapl_attr_update);
if (ret)
return ret;
+ if (rapl_model->per_core) {
+ ret = init_rapl_pmus(&rapl_pmus_per_core, nr_cores,
+ rapl_per_core_attr_groups, rapl_per_core_attr_update);
+ if (ret) {
+ /*
+ * If initialization of per_core PMU fails, reset per_core
+ * flag, and continue with power PMU initialization.
+ */
+ pr_warn("Per-core PMU initialization failed (%d)\n", ret);
+ rapl_model->per_core = false;
+ }
+ }
+
/*
* Install callbacks. Core will call them for each online cpu.
*/
@@ -886,14 +989,28 @@ static int __init rapl_pmu_init(void)
if (ret)
goto out1;
+ if (rapl_model->per_core) {
+ ret = perf_pmu_register(&rapl_pmus_per_core->pmu, "power_per_core", -1);
+ if (ret) {
+ /*
+ * If registration of per_core PMU fails, cleanup per_core PMU
+ * variables, reset the per_core flag and keep the
+ * power PMU untouched.
+ */
+ pr_warn("Per-core PMU registration failed (%d)\n", ret);
+ cleanup_rapl_pmus(rapl_pmus_per_core);
+ rapl_model->per_core = false;
+ }
+ }
rapl_advertise();
+
return 0;
out1:
cpuhp_remove_state(CPUHP_AP_PERF_X86_RAPL_ONLINE);
out:
pr_warn("Initialization failed (%d), disabled\n", ret);
- cleanup_rapl_pmus();
+ cleanup_rapl_pmus(rapl_pmus);
return ret;
}
module_init(rapl_pmu_init);
@@ -902,6 +1019,10 @@ static void __exit intel_rapl_exit(void)
{
cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_RAPL_ONLINE);
perf_pmu_unregister(&rapl_pmus->pmu);
- cleanup_rapl_pmus();
+ cleanup_rapl_pmus(rapl_pmus);
+ if (rapl_model->per_core) {
+ perf_pmu_unregister(&rapl_pmus_per_core->pmu);
+ cleanup_rapl_pmus(rapl_pmus_per_core);
+ }
}
module_exit(intel_rapl_exit);
Add a new "power_per_core" PMU and "energy-per-core" event for monitoring energy consumption by each core. The existing energy-cores event aggregates the energy consumption at the package level. This new event aligns with the AMD's per_core energy counters. Tested the package level and core level PMU counters with workloads pinned to different CPUs. Results with workload pinned to CPU 1 in core 1 on a AMD Zen4 Genoa machine: $ perf stat -a --per-core -e power_per_core/energy-per-core/ sleep 1 Performance counter stats for 'system wide': S0-D0-C0 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C1 1 5.72 Joules power_per_core/energy-per-core/ S0-D0-C2 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C3 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C4 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C5 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C6 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C7 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C8 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C9 1 0.02 Joules power_per_core/energy-per-core/ S0-D0-C10 1 0.02 Joules power_per_core/energy-per-core/ Signed-off-by: Dhananjay Ugwekar <Dhananjay.Ugwekar@amd.com> --- arch/x86/events/rapl.c | 155 ++++++++++++++++++++++++++++++++++++----- 1 file changed, 138 insertions(+), 17 deletions(-)