This patch adds a new uncore PMU to expose the Intel
RAPL (Running Average Power Limit) energy consumption counters.
Up to 3 counters, each counting a particular RAPL event are exposed.
The RAPL counters are available on Intel SandyBridge,
IvyBridge, Haswell. The server skus add a 3rd counter to measure
DRAM power consumption.
The following events are available nd exposed in sysfs:
- power/energy-cores: power consumption of all cores on socket
- power/energy-pkg: power consumption of all cores + LLc cache
- power/energy-dram: power consumption of DRAM (server skus only)
The RAPL PMU is uncore by nature and is implemented such
that it only works in system-wide mode. Measuring only
one CPU per socket is sufficient. The /sys/devices/rapl/cpumask
is exported and can be used by tools to figure out which CPU
to monitor by default. For instance, on a 2-socket system, 2 CPUs
(one on each socket) will be shown.
The counters all count in the same unit. The perf_events API
exposes all RAPL counters as 64-bit integers counting in unit
of 1/2^32 Joules (or 0.23 nJ). User level tools must convert
the counts by multiplying them by 0.23 and divide 10^9 to
obtain Joules. The reason for this is that the kernel avoids
doing floating point math whenever possible because it is
expensive (user floating-point state must be saved). The method
used avoids kernel floating-point and minimizes the loss of
precision (bits). Thanks to PeterZ for suggesting this approach.
To convert the raw count in Watt: W = C * 0.23 / (1e9 * time)
The kernel exposes both the scaling factor (0.23 nJ) and the
unit (Joules) in sysfs:
$ ls -1 /sys/devices/power/events/energy-*
/sys/devices/power/events/energy-cores
/sys/devices/power/events/energy-cores.scale
/sys/devices/power/events/energy-cores.unit
/sys/devices/power/events/energy-pkg
/sys/devices/power/events/energy-pkg.scale
/sys/devices/power/events/energy-pkg.unit
$ cat /sys/devices/power/events/energy-cores.scale
2.3e-10
$ cat cat /sys/devices/power/events/energy-cores.unit
Joules
RAPL PMU is a new standalone PMU which registers with the
perf_event core subsystem. The PMU type (attr->type) is
dynamically allocated and is available from /sys/device/rapl/type.
Sampling is not supported by the RAPL PMU. There is no
privilege level filtering either.
The PMU exports a cpumask in /sys/devices/power/cpumask. It
is used by perf to ensure only one instance of each RAPL event
is measured per processor socket. Hotplug CPU is also supported.
The perf stat infrasrtructure is modified to now show event
unit. It also applies the scaling factor. As such it will print
RAPL events in Joules (and not increments on 0.23 nJ):
# perf stat -a -e power/energy-pkg/,power/energy-cores/,cycles -I 1000 sleep 1000
# time counts unit events
1.000282860 2.51 Joules power/energy-pkg/ [100.00%]
1.000282860 0.31 Joules power/energy-cores/
1.000282860 37765378 ? cycles [100.00%]
The patch adds a hrtimer to poll the counters given that
they do no interrupt on overflow. Hardware counters are 32-bit
wide.
In v2, we add the locking necesarry to protect the rapl_pmu
struct. We also add a description at the top of the file.
We check for Intel only processor. We improved the data
layout of the rapl_pmu struct. We also lifted the restriction
of the number of instances of RAPL counters that can be active
at the same time. RAPL is free running counters, so ought to be
able to measure events as many times as necessary in parallel
via multiple tools. There is never multiplexing among RAPL events.
In v3, we have renamed the event to be more generic power/* instead
of rapl/*. We have modified perf stat to print the event with the
unit and scaling factors.
Supported CPUs: SandyBridge, IvyBridge, Haswell.
Signed-off-by: Stephane Eranian <[email protected]>
Stephane Eranian (4):
perf: add active_entry list head to struct perf_event
perf stat: add event unit and scale support
perf,x86: add Intel RAPL PMU support
perf,x86: add RAPL hrtimer support
arch/x86/kernel/cpu/Makefile | 2 +-
arch/x86/kernel/cpu/perf_event_intel_rapl.c | 717 +++++++++++++++++++++++++++
include/linux/perf_event.h | 1 +
kernel/events/core.c | 1 +
tools/perf/builtin-stat.c | 72 ++-
tools/perf/util/evsel.c | 2 +
tools/perf/util/evsel.h | 3 +
tools/perf/util/parse-events.c | 1 +
tools/perf/util/pmu.c | 170 ++++++-
tools/perf/util/pmu.h | 3 +
10 files changed, 950 insertions(+), 22 deletions(-)
create mode 100644 arch/x86/kernel/cpu/perf_event_intel_rapl.c
--
1.7.9.5
This patch adds a new fields to the struct perf_event.
It is intended to be used to chain events which are
active (enabled). It helps in the hardware layer
for PMU which do not have actual counter restrictions, i.e.,
free running read-only counters. Active events are chained
as opposed to being tracked via the counter they use.
Signed-off-by: Stephane Eranian <[email protected]>
---
include/linux/perf_event.h | 1 +
kernel/events/core.c | 1 +
2 files changed, 2 insertions(+)
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 2e069d1..a376384 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -435,6 +435,7 @@ struct perf_event {
struct perf_cgroup *cgrp; /* cgroup event is attach to */
int cgrp_defer_enabled;
#endif
+ struct list_head active_entry;
#endif /* CONFIG_PERF_EVENTS */
};
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 5bd7fe4..6ef9d19 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -6629,6 +6629,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
INIT_LIST_HEAD(&event->event_entry);
INIT_LIST_HEAD(&event->sibling_list);
INIT_LIST_HEAD(&event->rb_entry);
+ INIT_LIST_HEAD(&event->active_entry);
init_waitqueue_head(&event->waitq);
init_irq_work(&event->pending, perf_pending_event);
--
1.7.9.5
This patch adds perf stat support fo rhandling event units and
scales as exported by the kernel.
The kernel can export PMU events actual unit and scaling factor
via sysfs:
$ ls -1 /sys/devices/power/events/energy-*
/sys/devices/power/events/energy-cores
/sys/devices/power/events/energy-cores.scale
/sys/devices/power/events/energy-cores.unit
/sys/devices/power/events/energy-pkg
/sys/devices/power/events/energy-pkg.scale
/sys/devices/power/events/energy-pkg.unit
$ cat /sys/devices/power/events/energy-cores.scale
2.3e-10
$ cat cat /sys/devices/power/events/energy-cores.unit
Joules
This patch modifies the pmu event alias code to check
for the presence of the .unit and .scale files to load
the corresponding values. They are then used by perf stat
transparentely:
# perf stat -a -e power/energy-pkg/,power/energy-cores/,cycles -I 1000 sleep 1000
# time counts unit events
1.000214717 3.07 Joules power/energy-pkg/ [100.00%]
1.000214717 0.53 Joules power/energy-cores/
1.000214717 12965028 ? cycles [100.00%]
2.000749289 3.01 Joules power/energy-pkg/
2.000749289 0.52 Joules power/energy-cores/
2.000749289 15817043 ? cycles
Signed-off-by: Stephane Eranian <[email protected]>
---
tools/perf/builtin-stat.c | 72 ++++++++++++-----
tools/perf/util/evsel.c | 2 +
tools/perf/util/evsel.h | 3 +
tools/perf/util/parse-events.c | 1 +
tools/perf/util/pmu.c | 170 +++++++++++++++++++++++++++++++++++++++-
tools/perf/util/pmu.h | 3 +
6 files changed, 230 insertions(+), 21 deletions(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 1a9c95d..43dea3b 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -138,6 +138,7 @@ static const char *post_cmd = NULL;
static bool sync_run = false;
static unsigned int interval = 0;
static unsigned int initial_delay = 0;
+static unsigned int unit_width = 4; /* strlen("unit") */
static bool forever = false;
static struct timespec ref_time;
static struct cpu_map *aggr_map;
@@ -462,17 +463,17 @@ static void print_interval(void)
if (num_print_interval == 0 && !csv_output) {
switch (aggr_mode) {
case AGGR_SOCKET:
- fprintf(output, "# time socket cpus counts events\n");
+ fprintf(output, "# time socket cpus counts %*s events\n", unit_width, "unit");
break;
case AGGR_CORE:
- fprintf(output, "# time core cpus counts events\n");
+ fprintf(output, "# time core cpus counts %*s events\n", unit_width, "unit");
break;
case AGGR_NONE:
- fprintf(output, "# time CPU counts events\n");
+ fprintf(output, "# time CPU counts %*s events\n", unit_width, "unit");
break;
case AGGR_GLOBAL:
default:
- fprintf(output, "# time counts events\n");
+ fprintf(output, "# time counts %*s events\n", unit_width, "unit");
}
}
@@ -517,6 +518,7 @@ static int __run_perf_stat(int argc, const char **argv)
unsigned long long t0, t1;
struct perf_evsel *counter;
struct timespec ts;
+ size_t l;
int status = 0;
const bool forks = (argc > 0);
@@ -566,6 +568,10 @@ static int __run_perf_stat(int argc, const char **argv)
return -1;
}
counter->supported = true;
+
+ l = strlen(counter->unit);
+ if (l > unit_width)
+ unit_width = l;
}
if (perf_evlist__apply_filters(evsel_list)) {
@@ -911,19 +917,32 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
double total, ratio = 0.0, total2;
const char *fmt;
- if (csv_output)
- fmt = "%.0f%s%s";
- else if (big_num)
- fmt = "%'18.0f%s%-25s";
- else
- fmt = "%18.0f%s%-25s";
+ if (csv_output) {
+ if (evsel->scale != 1.0)
+ fmt = "%.2f%s%s%s%s";
+ else
+ fmt = "%.0f%s%s%s%s";
+ } else if (big_num)
+ if (evsel->scale != 1.0)
+ fmt = "%'18.2f%s%-*s%s%-25s";
+ else
+ fmt = "%'18.0f%s%-*s%s%-25s";
+ else {
+ if (evsel->scale != 1.0)
+ fmt = "%18.2f%s%-*s%s%-25s";
+ else
+ fmt = "%18.0f%s%-*s%s%-25s";
+ }
aggr_printout(evsel, cpu, nr);
if (aggr_mode == AGGR_GLOBAL)
cpu = 0;
- fprintf(output, fmt, avg, csv_sep, perf_evsel__name(evsel));
+ if (csv_output)
+ fprintf(output, fmt, avg, csv_sep, evsel->unit, csv_sep, perf_evsel__name(evsel));
+ else
+ fprintf(output, fmt, avg, csv_sep, unit_width, evsel->unit, csv_sep, perf_evsel__name(evsel));
if (evsel->cgrp)
fprintf(output, "%s%s", csv_sep, evsel->cgrp->name);
@@ -1062,6 +1081,7 @@ static void print_aggr(char *prefix)
{
struct perf_evsel *counter;
int cpu, cpu2, s, s2, id, nr;
+ double uval;
u64 ena, run, val;
if (!(aggr_map || aggr_get_id))
@@ -1088,9 +1108,13 @@ static void print_aggr(char *prefix)
if (run == 0 || ena == 0) {
aggr_printout(counter, id, nr);
- fprintf(output, "%*s%s%*s",
+ fprintf(output, "%*s%s%*s%s%*s",
csv_output ? 0 : 18,
counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED,
+
+ csv_sep,
+ csv_output ? 0 : -10,
+ counter->unit,
csv_sep,
csv_output ? 0 : -24,
perf_evsel__name(counter));
@@ -1102,11 +1126,12 @@ static void print_aggr(char *prefix)
fputc('\n', output);
continue;
}
+ uval = val * counter->scale;
if (nsec_counter(counter))
- nsec_printout(id, nr, counter, val);
+ nsec_printout(id, nr, counter, uval);
else
- abs_printout(id, nr, counter, val);
+ abs_printout(id, nr, counter, uval);
if (!csv_output) {
print_noise(counter, 1.0);
@@ -1129,6 +1154,7 @@ static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
struct perf_stat *ps = counter->priv;
double avg = avg_stats(&ps->res_stats[0]);
int scaled = counter->counts->scaled;
+ double uval;
if (prefix)
fprintf(output, "%s", prefix);
@@ -1148,10 +1174,12 @@ static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
return;
}
+ uval = avg * counter->scale;
+
if (nsec_counter(counter))
- nsec_printout(-1, 0, counter, avg);
+ nsec_printout(-1, 0, counter, uval);
else
- abs_printout(-1, 0, counter, avg);
+ abs_printout(-1, 0, counter, uval);
print_noise(counter, avg);
@@ -1178,6 +1206,7 @@ static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
static void print_counter(struct perf_evsel *counter, char *prefix)
{
u64 ena, run, val;
+ double uval;
int cpu;
for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
@@ -1189,12 +1218,15 @@ static void print_counter(struct perf_evsel *counter, char *prefix)
fprintf(output, "%s", prefix);
if (run == 0 || ena == 0) {
- fprintf(output, "CPU%*d%s%*s%s%*s",
+ fprintf(output, "CPU%*d%s%*s%s%*s%s%*s",
csv_output ? 0 : -4,
perf_evsel__cpus(counter)->map[cpu], csv_sep,
csv_output ? 0 : 18,
counter->supported ? CNTR_NOT_COUNTED : CNTR_NOT_SUPPORTED,
csv_sep,
+ csv_output ? 0 : -10,
+ counter->unit,
+ csv_sep,
csv_output ? 0 : -24,
perf_evsel__name(counter));
@@ -1206,10 +1238,12 @@ static void print_counter(struct perf_evsel *counter, char *prefix)
continue;
}
+ uval = val * counter->scale;
+
if (nsec_counter(counter))
- nsec_printout(cpu, 0, counter, val);
+ nsec_printout(cpu, 0, counter, uval);
else
- abs_printout(cpu, 0, counter, val);
+ abs_printout(cpu, 0, counter, uval);
if (!csv_output) {
print_noise(counter, 1.0);
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 3a334f0..867971b 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -162,6 +162,8 @@ void perf_evsel__init(struct perf_evsel *evsel,
evsel->idx = idx;
evsel->attr = *attr;
evsel->leader = evsel;
+ evsel->unit = "?";
+ evsel->scale = 1.0;
INIT_LIST_HEAD(&evsel->node);
hists__init(&evsel->hists);
evsel->sample_size = __perf_evsel__sample_size(attr->sample_type);
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 5aa68cd..d5c6606 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -68,6 +68,8 @@ struct perf_evsel {
u32 ids;
struct hists hists;
char *name;
+ double scale;
+ const char *unit;
struct event_format *tp_format;
union {
void *priv;
@@ -130,6 +132,7 @@ extern const char *perf_evsel__sw_names[PERF_COUNT_SW_MAX];
int __perf_evsel__hw_cache_type_op_res_name(u8 type, u8 op, u8 result,
char *bf, size_t size);
const char *perf_evsel__name(struct perf_evsel *evsel);
+
const char *perf_evsel__group_name(struct perf_evsel *evsel);
int perf_evsel__group_desc(struct perf_evsel *evsel, char *buf, size_t size);
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index c90e55c..be7eba8 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -838,6 +838,7 @@ int parse_events_name(struct list_head *list, char *name)
list_for_each_entry(evsel, list, node) {
if (!evsel->name)
evsel->name = strdup(name);
+ pmu_get_event_unit_scale(evsel);
}
return 0;
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 64362fe..ae17132 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -4,6 +4,7 @@
#include <unistd.h>
#include <stdio.h>
#include <dirent.h>
+#include <locale.h>
#include "sysfs.h"
#include "util.h"
#include "pmu.h"
@@ -14,6 +15,8 @@ struct perf_pmu_alias {
char *name;
struct list_head terms;
struct list_head list;
+ char *unit;
+ double scale;
};
struct perf_pmu_format {
@@ -95,7 +98,89 @@ static int pmu_format(const char *name, struct list_head *format)
return 0;
}
-static int perf_pmu__new_alias(struct list_head *list, char *name, FILE *file)
+static int perf_pmu__parse_scale(struct perf_pmu_alias *alias, char *dir, char *name)
+{
+ struct stat st;
+ ssize_t sret;
+ char scale[128];
+ int fd, ret = -1;
+ char path[PATH_MAX];
+ char *lc;
+
+ snprintf(path, PATH_MAX, "%s/%s.scale", dir, name);
+
+ fd = open(path, O_RDONLY);
+ if (fd == -1)
+ return -1;
+
+ if (fstat(fd, &st) < 0)
+ goto error;
+
+ sret = read(fd, scale, sizeof(scale)-1);
+ if (sret < 0)
+ goto error;
+
+ scale[sret] = '\0';
+ /*
+ * save current locale
+ */
+ lc = setlocale(LC_NUMERIC, NULL);
+
+ /*
+ * force to C locale to ensure kernel
+ * scale string is converted correctly.
+ * kernel uses default C locale.
+ */
+ setlocale(LC_NUMERIC, "C");
+
+ alias->scale = strtod(scale, NULL);
+
+ /* restore locale */
+ setlocale(LC_NUMERIC, lc);
+
+ ret = 0;
+error:
+ close(fd);
+ return ret;
+}
+
+static int perf_pmu__parse_unit(struct perf_pmu_alias *alias, char *dir, char *name)
+{
+ struct stat st;
+ ssize_t sret;
+ int fd;
+ char path[PATH_MAX];
+
+ snprintf(path, PATH_MAX, "%s/%s.unit", dir, name);
+
+ fd = open(path, O_RDONLY);
+ if (fd == -1)
+ return -1;
+
+ if (fstat(fd, &st) < 0)
+ goto error;
+
+ alias->unit = malloc(st.st_size + 1);
+ if (!alias->unit)
+ goto error;
+
+ sret = read(fd, alias->unit, st.st_size);
+ if (sret < 0)
+ goto error;
+
+ close(fd);
+
+ alias->unit[sret] = '\0';
+
+ return 0;
+error:
+ close(fd);
+ free(alias->unit);
+ alias->unit = NULL;
+ return -1;
+}
+
+static int perf_pmu__new_alias(struct list_head *list, char *dir, char *name, FILE *file)
{
struct perf_pmu_alias *alias;
char buf[256];
@@ -111,6 +196,9 @@ static int perf_pmu__new_alias(struct list_head *list, char *name, FILE *file)
return -ENOMEM;
INIT_LIST_HEAD(&alias->terms);
+ alias->scale = 1.0;
+ alias->unit = NULL;
+
ret = parse_events_terms(&alias->terms, buf);
if (ret) {
free(alias);
@@ -118,7 +206,14 @@ static int perf_pmu__new_alias(struct list_head *list, char *name, FILE *file)
}
alias->name = strdup(name);
+ /*
+ * load unit name and scale if available
+ */
+ perf_pmu__parse_unit(alias, dir, name);
+ perf_pmu__parse_scale(alias, dir, name);
+
list_add_tail(&alias->list, list);
+
return 0;
}
@@ -130,6 +225,7 @@ static int pmu_aliases_parse(char *dir, struct list_head *head)
{
struct dirent *evt_ent;
DIR *event_dir;
+ size_t len;
int ret = 0;
event_dir = opendir(dir);
@@ -144,13 +240,24 @@ static int pmu_aliases_parse(char *dir, struct list_head *head)
if (!strcmp(name, ".") || !strcmp(name, ".."))
continue;
+ /*
+ * skip .unit and .scale info files
+ * parsed in perf_pmu__new_alias()
+ */
+ len = strlen(name);
+ if (len > 5 && !strcmp(name + len - 5, ".unit"))
+ continue;
+ if (len > 6 && !strcmp(name + len - 6, ".scale"))
+ continue;
+
snprintf(path, PATH_MAX, "%s/%s", dir, name);
ret = -EINVAL;
file = fopen(path, "r");
if (!file)
break;
- ret = perf_pmu__new_alias(head, name, file);
+
+ ret = perf_pmu__new_alias(head, dir, name, file);
fclose(file);
}
@@ -653,3 +760,62 @@ bool pmu_have_event(const char *pname, const char *name)
}
return false;
}
+
+static const char *pmu_event_unit(struct perf_pmu *pmu, const char *name)
+{
+ struct perf_pmu_alias *alias;
+ char buf[1024];
+ char *fname;
+ const char *unit = "";
+
+ if (!name)
+ return unit;
+
+ list_for_each_entry(alias, &pmu->aliases, list) {
+ fname = format_alias(buf, sizeof(buf), pmu, alias);
+ if (!strcmp(fname, name)) {
+ unit = alias->unit;
+ break;
+ }
+ }
+ return unit;
+}
+
+static double pmu_event_scale(struct perf_pmu *pmu, const char *name)
+{
+ struct perf_pmu_alias *alias;
+ char buf[1024];
+ char *fname;
+ double scale = 1.0;
+
+ if (!name)
+ return 1.0;
+
+ list_for_each_entry(alias, &pmu->aliases, list) {
+ fname = format_alias(buf, sizeof(buf), pmu, alias);
+ if (!strcmp(fname, name)) {
+ scale = alias->scale;
+ break;
+ }
+ }
+ return scale;
+}
+
+int pmu_get_event_unit_scale(struct perf_evsel *evsel)
+{
+ __u32 type = evsel->attr.type;
+ struct perf_pmu *pmu;
+
+ if (!evsel->name)
+ return -1;
+
+ list_for_each_entry(pmu, &pmus, list) {
+ if (pmu->type == type)
+ goto found;
+ }
+ return -1;
+found:
+ evsel->unit = pmu_event_unit(pmu, evsel->name);
+ evsel->scale = pmu_event_scale(pmu, evsel->name);
+ return 0;
+}
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 1179b26..6bf23b2 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -4,6 +4,7 @@
#include <linux/bitops.h>
#include <linux/perf_event.h>
#include <stdbool.h>
+#include "util/evsel.h"
enum {
PERF_PMU_FORMAT_VALUE_CONFIG,
@@ -45,4 +46,6 @@ void print_pmu_events(const char *event_glob, bool name_only);
bool pmu_have_event(const char *pname, const char *name);
int perf_pmu__test(void);
+
+int pmu_get_event_unit_scale(struct perf_evsel *evsel);
#endif /* __PMU_H */
--
1.7.9.5
The RAPL PMU counters do not interrupt on overflow.
Therefore, the kernel needs to poll the counters
to avoid missing an overflow. This patch adds
the hrtimer code to do this.
The timer internval is calculated at boot time
based on the power unit used by the HW.
Signed-off-by: Stephane Eranian <[email protected]>
---
arch/x86/kernel/cpu/perf_event_intel_rapl.c | 75 +++++++++++++++++++++++++--
1 file changed, 70 insertions(+), 5 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event_intel_rapl.c b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
index 3d71d39..ed0566a 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_rapl.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
@@ -92,11 +92,13 @@ static struct kobj_attribute format_attr_##_var = \
struct rapl_pmu {
spinlock_t lock;
- atomic_t refcnt;
int hw_unit; /* 1/2^hw_unit Joule */
- int phys_id;
- int n_active; /* number of active events */
+ struct hrtimer hrtimer;
struct list_head active_list;
+ ktime_t timer_interval; /* in ktime_t unit */
+ int n_active; /* number of active events */
+ int phys_id;
+ atomic_t refcnt;
};
static struct pmu rapl_pmu_class;
@@ -161,6 +163,47 @@ static u64 rapl_event_update(struct perf_event *event)
return new_raw_count;
}
+static void rapl_start_hrtimer(struct rapl_pmu *pmu)
+{
+ __hrtimer_start_range_ns(&pmu->hrtimer,
+ pmu->timer_interval, 0,
+ HRTIMER_MODE_REL_PINNED, 0);
+}
+
+static void rapl_stop_hrtimer(struct rapl_pmu *pmu)
+{
+ hrtimer_cancel(&pmu->hrtimer);
+}
+
+static enum hrtimer_restart rapl_hrtimer_handle(struct hrtimer *hrtimer)
+{
+ struct rapl_pmu *pmu = container_of(hrtimer, struct rapl_pmu, hrtimer);
+ struct perf_event *event;
+ unsigned long flags;
+
+ if (!pmu->n_active)
+ return HRTIMER_NORESTART;
+
+ spin_lock_irqsave(&pmu->lock, flags);
+
+ list_for_each_entry(event, &pmu->active_list, active_entry) {
+ rapl_event_update(event);
+ }
+
+ spin_unlock_irqrestore(&pmu->lock, flags);
+
+ hrtimer_forward_now(&pmu->hrtimer, pmu->timer_interval);
+
+ return HRTIMER_RESTART;
+}
+
+static void rapl_hrtimer_init(struct rapl_pmu *pmu)
+{
+ hrtimer_init(&pmu->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+ pmu->hrtimer.function = rapl_hrtimer_handle;
+}
+
+
static void __rapl_pmu_event_start(struct rapl_pmu *pmu,
struct perf_event *event)
{
@@ -174,6 +217,8 @@ static void __rapl_pmu_event_start(struct rapl_pmu *pmu,
local64_set(&event->hw.prev_count, rapl_read_counter(event));
pmu->n_active++;
+ if (pmu->n_active == 1)
+ rapl_start_hrtimer(pmu);
}
static void rapl_pmu_event_start(struct perf_event *event, int mode)
@@ -198,6 +243,8 @@ static void rapl_pmu_event_stop(struct perf_event *event, int mode)
if (!(hwc->state & PERF_HES_STOPPED)) {
WARN_ON_ONCE(pmu->n_active <= 0);
pmu->n_active--;
+ if (pmu->n_active == 0)
+ rapl_stop_hrtimer(pmu);
list_del(&event->active_entry);
@@ -439,6 +486,7 @@ static int rapl_cpu_prepare(int cpu)
{
struct rapl_pmu *pmu = per_cpu(rapl_pmu, cpu);
int phys_id = topology_physical_package_id(cpu);
+ u64 ms;
if (pmu)
return 0;
@@ -464,6 +512,20 @@ static int rapl_cpu_prepare(int cpu)
rdmsrl(MSR_RAPL_POWER_UNIT, pmu->hw_unit);
pmu->hw_unit = (pmu->hw_unit >> 8) & 0x1FULL;
+ /*
+ * use reference of 200W for scaling the timeout
+ * to avoid missing counter overflows.
+ * 200W = 200 Joules/sec
+ * divide interval by 2 to avoid lockstep (2 * 100)
+ * if hw unit is 32, then we use 2 ms 1/200/2
+ */
+ if (pmu->hw_unit < 32)
+ ms = 1000 * (1ULL << (32 - pmu->hw_unit - 1)) / (2 * 100);
+ else
+ ms = 2;
+
+ pmu->timer_interval = ms_to_ktime(ms);
+
/* set RAPL pmu for this cpu for now */
per_cpu(rapl_pmu_kfree, cpu) = NULL;
per_cpu(rapl_pmu, cpu) = pmu;
@@ -625,6 +687,7 @@ static int __init rapl_pmu_init(void)
}
rapl_cpu_prepare(cpu);
cpumask_set_cpu(cpu, &rapl_cpu_mask);
+ rapl_hrtimer_init(per_cpu(rapl_pmu, cpu));
}
perf_cpu_notifier(rapl_cpu_notifier);
@@ -641,9 +704,11 @@ static int __init rapl_pmu_init(void)
pr_info("RAPL PMU detected, hw unit 2^-%d Joules,"
" API unit is 2^-32 Joules,"
- " %d fixed counters\n",
+ " %d fixed counters"
+ " %llu ms ovfl timer\n",
pmu->hw_unit,
- hweight32(rapl_cntr_mask));
+ hweight32(rapl_cntr_mask),
+ ktime_to_ms(pmu->timer_interval));
put_online_cpus();
--
1.7.9.5
This patch adds a new uncore PMU to expose the Intel
RAPL energy consumption counters. Up to 3 counters,
each counting a particular RAPL event are exposed.
The RAPL counters are available on Intel SandyBridge,
IvyBridge, Haswell. The server skus add a 3rd counter.
The following events are available and exposed in sysfs:
- power/energy-cores: power consumption of all cores on socket
- power/energy-pkg: power consumption of all cores + LLc cache
- power/energy-dram: power consumption of DRAM (servers only)
For each event both the unit (Joules) and scale (0.23 nJ)
is exposed in sysfs for use by perf stat and other tools.
Files are:
/sys/devices/power/events/energy-*.unit
/sys/devices/power/events/energy-*.scale
The RAPL PMU is uncore by nature and is implemented such
that it only works in system-wide mode. Measuring only
one CPU per socket is sufficient. The /sys/devices/power/cpumask
file can be used by tools to figure out which CPUs
to monitor by default. For instance, on a 2-socket system, 2 CPUs
(one on each socket) will be shown.
The counters all count in the same unit (exposed via sysfs).
The perf_events API exposes all RAPL counters as 64-bit integers
counting in unit of 1/2^32 Joules (or 0.23 nJ). User level tools
must convert the counts by multiplying them by 0.23 and divide 10^9
to obtain Joules. The reason for this is that the kernel avoids
doing floating point math whenever possible because it is
expensive (user floating-point state must be saved). The method
used avoids kernel floating-point and minimizes the loss of
precision (bits). Thanks to PeterZ for suggesting this approach.
To convert the raw count in Watt:
W = C * 0.23 / (1e9 * time)
or ldexp(C, -32).
RAPL PMU is a new standalone PMU which registers with the
perf_event core subsystem. The PMU type (attr->type) is
dynamically allocated and is available from /sys/device/power/type.
Sampling is not supported by the RAPL PMU. There is no
privilege level filtering either.
Signed-off-by: Stephane Eranian <[email protected]>
---
arch/x86/kernel/cpu/Makefile | 2 +-
arch/x86/kernel/cpu/perf_event_intel_rapl.c | 652 +++++++++++++++++++++++++++
2 files changed, 653 insertions(+), 1 deletion(-)
create mode 100644 arch/x86/kernel/cpu/perf_event_intel_rapl.c
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index 47b56a7..6359506 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -36,7 +36,7 @@ obj-$(CONFIG_CPU_SUP_AMD) += perf_event_amd_iommu.o
endif
obj-$(CONFIG_CPU_SUP_INTEL) += perf_event_p6.o perf_event_knc.o perf_event_p4.o
obj-$(CONFIG_CPU_SUP_INTEL) += perf_event_intel_lbr.o perf_event_intel_ds.o perf_event_intel.o
-obj-$(CONFIG_CPU_SUP_INTEL) += perf_event_intel_uncore.o
+obj-$(CONFIG_CPU_SUP_INTEL) += perf_event_intel_uncore.o perf_event_intel_rapl.o
endif
diff --git a/arch/x86/kernel/cpu/perf_event_intel_rapl.c b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
new file mode 100644
index 0000000..c61b411
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
@@ -0,0 +1,652 @@
+/*
+ * perf_event_intel_rapl.c: support Intel RAPL energy consumption counters
+ * Copyright (C) 2013 Google, Inc., Stephane Eranian
+ *
+ * Intel RAPL interface is specified in the IA-32 Manual Vol3b
+ * section 14.7.1 (September 2013)
+ *
+ * RAPL provides more controls than just reporting energy consumption
+ * however here we only expose the 3 energy consumption free running
+ * counters (pp0, pkg, dram).
+ *
+ * Each of those counters increments in a power unit defined by the
+ * RAPL_POWER_UNIT MSR. On SandyBridge, this unit is 1/(2^16) Joules
+ * but it can vary.
+ *
+ * Counter to rapl events mappings:
+ *
+ * pp0 counter: consumption of all physical cores (power plane 0)
+ * event: power/energy_cores
+ * perf code: 0x1
+ *
+ * pkg counter: consumption of the whole processor package
+ * event: power/energy_pkg
+ * perf code: 0x2
+ *
+ * dram counter: consumption of the dram domain (servers only)
+ * event: power/energy_dram
+ * perf code: 0x3
+ *
+ * We manage those counters as free running (read-only). They may be
+ * use simultaneously by other tools, such as turbostat.
+ *
+ * The events only support system-wide mode counting. There is no
+ * sampling support because it does not make sense and is not
+ * supported by the RAPL hardware.
+ *
+ * Because we want to avoid floating-point operations in the kernel,
+ * the events are all reported in fixed point arithmetic (32.32).
+ * Tools must adjust the counts to convert them to Watts using
+ * the duration of the measurement. Tools may use a function such as
+ * ldexp(raw_count, -32);
+ */
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/perf_event.h>
+#include <asm/cpu_device_id.h>
+#include "perf_event.h"
+
+/*
+ * RAPL energy status counters
+ */
+#define RAPL_IDX_PP0_NRG_STAT 0 /* all cores */
+#define INTEL_RAPL_PP0 0x1 /* pseudo-encoding */
+#define RAPL_IDX_PKG_NRG_STAT 1 /* entire package */
+#define INTEL_RAPL_PKG 0x2 /* pseudo-encoding */
+#define RAPL_IDX_RAM_NRG_STAT 2 /* DRAM */
+#define INTEL_RAPL_RAM 0x3 /* pseudo-encoding */
+
+/* Clients have PP0, PKG */
+#define RAPL_IDX_CLN (1<<RAPL_IDX_PP0_NRG_STAT|\
+ 1<<RAPL_IDX_PKG_NRG_STAT)
+
+/* Servers have PP0, PKG, RAM */
+#define RAPL_IDX_SRV (1<<RAPL_IDX_PP0_NRG_STAT|\
+ 1<<RAPL_IDX_PKG_NRG_STAT|\
+ 1<<RAPL_IDX_RAM_NRG_STAT)
+
+/*
+ * event code: LSB 8 bits, passed in attr->config
+ * any other bit is reserved
+ */
+#define RAPL_EVENT_MASK 0xFFULL
+
+#define DEFINE_RAPL_FORMAT_ATTR(_var, _name, _format) \
+static ssize_t __rapl_##_var##_show(struct kobject *kobj, \
+ struct kobj_attribute *attr, \
+ char *page) \
+{ \
+ BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \
+ return sprintf(page, _format "\n"); \
+} \
+static struct kobj_attribute format_attr_##_var = \
+ __ATTR(_name, 0444, __rapl_##_var##_show, NULL)
+
+#define RAPL_EVENT_DESC(_name, _config) \
+{ \
+ .attr = __ATTR(_name, 0444, rapl_event_show, NULL), \
+ .config = _config, \
+}
+
+#define RAPL_CNTR_WIDTH 32 /* 32-bit rapl counters */
+
+struct rapl_pmu {
+ spinlock_t lock;
+ atomic_t refcnt;
+ int hw_unit; /* 1/2^hw_unit Joule */
+ int phys_id;
+ int n_active; /* number of active events */
+ struct list_head active_list;
+};
+
+static struct pmu rapl_pmu_class;
+static cpumask_t rapl_cpu_mask;
+static int rapl_cntr_mask;
+
+static DEFINE_PER_CPU(struct rapl_pmu *, rapl_pmu);
+static DEFINE_PER_CPU(struct rapl_pmu *, rapl_pmu_kfree);
+
+static DEFINE_SPINLOCK(rapl_hotplug_lock);
+
+static inline u64 rapl_read_counter(struct perf_event *event)
+{
+ u64 raw;
+ rdmsrl(event->hw.event_base, raw);
+ return raw;
+}
+
+static inline u64 rapl_scale(u64 v)
+{
+ /*
+ * scale delta to smallest unit (1/2^32)
+ * users must then scale back: count * 1/(1e9*2^32) to get Joules
+ * or use ldexp(count, -32).
+ * Watts = Joules/Time delta
+ */
+ return v << (32 - __get_cpu_var(rapl_pmu)->hw_unit);
+}
+
+static u64 rapl_event_update(struct perf_event *event)
+{
+ struct hw_perf_event *hwc = &event->hw;
+ u64 prev_raw_count, new_raw_count;
+ s64 delta, sdelta;
+ int shift = RAPL_CNTR_WIDTH;
+
+again:
+ prev_raw_count = local64_read(&hwc->prev_count);
+ rdmsrl(event->hw.event_base, new_raw_count);
+
+ if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+ new_raw_count) != prev_raw_count) {
+ cpu_relax();
+ goto again;
+ }
+
+ /*
+ * Now we have the new raw value and have updated the prev
+ * timestamp already. We can now calculate the elapsed delta
+ * (event-)time and add that to the generic event.
+ *
+ * Careful, not all hw sign-extends above the physical width
+ * of the count.
+ */
+ delta = (new_raw_count << shift) - (prev_raw_count << shift);
+ delta >>= shift;
+
+ sdelta = rapl_scale(delta);
+
+ local64_add(sdelta, &event->count);
+
+ return new_raw_count;
+}
+
+static void __rapl_pmu_event_start(struct rapl_pmu *pmu,
+ struct perf_event *event)
+{
+ if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
+ return;
+
+ event->hw.state = 0;
+
+ list_add_tail(&event->active_entry, &pmu->active_list);
+
+ local64_set(&event->hw.prev_count, rapl_read_counter(event));
+
+ pmu->n_active++;
+}
+
+static void rapl_pmu_event_start(struct perf_event *event, int mode)
+{
+ struct rapl_pmu *pmu = __get_cpu_var(rapl_pmu);
+ unsigned long flags;
+
+ spin_lock_irqsave(&pmu->lock, flags);
+ __rapl_pmu_event_start(pmu, event);
+ spin_unlock_irqrestore(&pmu->lock, flags);
+}
+
+static void rapl_pmu_event_stop(struct perf_event *event, int mode)
+{
+ struct rapl_pmu *pmu = __get_cpu_var(rapl_pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ unsigned long flags;
+
+ spin_lock_irqsave(&pmu->lock, flags);
+
+ /* mark event as deactivated and stopped */
+ if (!(hwc->state & PERF_HES_STOPPED)) {
+ WARN_ON_ONCE(pmu->n_active <= 0);
+ pmu->n_active--;
+
+ list_del(&event->active_entry);
+
+ WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+ hwc->state |= PERF_HES_STOPPED;
+ }
+
+ /* check if update of sw counter is necessary */
+ if ((mode & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
+ /*
+ * Drain the remaining delta count out of a event
+ * that we are disabling:
+ */
+ rapl_event_update(event);
+ hwc->state |= PERF_HES_UPTODATE;
+ }
+
+ spin_unlock_irqrestore(&pmu->lock, flags);
+}
+
+static int rapl_pmu_event_add(struct perf_event *event, int mode)
+{
+ struct rapl_pmu *pmu = __get_cpu_var(rapl_pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ unsigned long flags;
+
+ spin_lock_irqsave(&pmu->lock, flags);
+
+ hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+ if (mode & PERF_EF_START)
+ __rapl_pmu_event_start(pmu, event);
+
+ spin_unlock_irqrestore(&pmu->lock, flags);
+
+ return 0;
+}
+
+static void rapl_pmu_event_del(struct perf_event *event, int flags)
+{
+ rapl_pmu_event_stop(event, PERF_EF_UPDATE);
+}
+
+static int rapl_pmu_event_init(struct perf_event *event)
+{
+ u64 cfg = event->attr.config & RAPL_EVENT_MASK;
+ int bit, msr, ret = 0;
+
+ /* only look at RAPL events */
+ if (event->attr.type != rapl_pmu_class.type)
+ return -ENOENT;
+
+ /* check only supported bits are set */
+ if (event->attr.config & ~RAPL_EVENT_MASK)
+ return -EINVAL;
+
+ /*
+ * check event is known (determines counter)
+ */
+ switch (cfg) {
+ case INTEL_RAPL_PP0:
+ bit = RAPL_IDX_PP0_NRG_STAT;
+ msr = MSR_PP0_ENERGY_STATUS;
+ break;
+ case INTEL_RAPL_PKG:
+ bit = RAPL_IDX_PKG_NRG_STAT;
+ msr = MSR_PKG_ENERGY_STATUS;
+ break;
+ case INTEL_RAPL_RAM:
+ bit = RAPL_IDX_RAM_NRG_STAT;
+ msr = MSR_DRAM_ENERGY_STATUS;
+ break;
+ default:
+ return -EINVAL;
+ }
+ /* check event supported */
+ if (!(rapl_cntr_mask & (1 << bit)))
+ return -EINVAL;
+
+ /* unsupported modes and filters */
+ if (event->attr.exclude_user ||
+ event->attr.exclude_kernel ||
+ event->attr.exclude_hv ||
+ event->attr.exclude_idle ||
+ event->attr.exclude_host ||
+ event->attr.exclude_guest ||
+ event->attr.sample_period) /* no sampling */
+ return -EINVAL;
+
+ /* must be done before validate_group */
+ event->hw.event_base = msr;
+ event->hw.config = cfg;
+ event->hw.idx = bit;
+
+ return ret;
+}
+
+static void rapl_pmu_event_read(struct perf_event *event)
+{
+ rapl_event_update(event);
+}
+
+static ssize_t rapl_get_attr_cpumask(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ int n = cpulist_scnprintf(buf, PAGE_SIZE - 2, &rapl_cpu_mask);
+
+ buf[n++] = '\n';
+ buf[n] = '\0';
+ return n;
+}
+
+static DEVICE_ATTR(cpumask, S_IRUGO, rapl_get_attr_cpumask, NULL);
+
+static struct attribute *rapl_pmu_attrs[] = {
+ &dev_attr_cpumask.attr,
+ NULL,
+};
+
+static struct attribute_group rapl_pmu_attr_group = {
+ .attrs = rapl_pmu_attrs,
+};
+
+EVENT_ATTR_STR(energy-cores, rapl_cores, "event=0x01");
+EVENT_ATTR_STR(energy-pkg , rapl_pkg, "event=0x02");
+EVENT_ATTR_STR(energy-ram , rapl_ram, "event=0x03");
+
+EVENT_ATTR_STR(energy-cores.unit, rapl_cores_unit, "Joules");
+EVENT_ATTR_STR(energy-pkg.unit , rapl_pkg_unit, "Joules");
+EVENT_ATTR_STR(energy-ram.unit , rapl_ram_unit, "Joules");
+
+/*
+ * we compute in 0.23 nJ increments regardless of MSR
+ */
+EVENT_ATTR_STR(energy-cores.scale, rapl_cores_scale, "2.3e-10");
+EVENT_ATTR_STR(energy-pkg.scale, rapl_pkg_scale, "2.3e-10");
+EVENT_ATTR_STR(energy-ram.scale, rapl_ram_scale, "2.3e-10");
+
+static struct attribute *rapl_events_srv_attr[] = {
+ EVENT_PTR(rapl_cores),
+ EVENT_PTR(rapl_pkg),
+ EVENT_PTR(rapl_ram),
+
+ EVENT_PTR(rapl_cores_unit),
+ EVENT_PTR(rapl_pkg_unit),
+ EVENT_PTR(rapl_ram_unit),
+
+ EVENT_PTR(rapl_cores_scale),
+ EVENT_PTR(rapl_pkg_scale),
+ EVENT_PTR(rapl_ram_scale),
+ NULL,
+};
+
+static struct attribute *rapl_events_cln_attr[] = {
+ EVENT_PTR(rapl_cores),
+ EVENT_PTR(rapl_pkg),
+
+ EVENT_PTR(rapl_cores_unit),
+ EVENT_PTR(rapl_pkg_unit),
+
+ EVENT_PTR(rapl_cores_scale),
+ EVENT_PTR(rapl_pkg_scale),
+ NULL,
+};
+
+static struct attribute_group rapl_pmu_events_group = {
+ .name = "events",
+ .attrs = NULL, /* patched at runtime */
+};
+
+DEFINE_RAPL_FORMAT_ATTR(event, event, "config:0-7");
+static struct attribute *rapl_formats_attr[] = {
+ &format_attr_event.attr,
+ NULL,
+};
+
+static struct attribute_group rapl_pmu_format_group = {
+ .name = "format",
+ .attrs = rapl_formats_attr,
+};
+
+const struct attribute_group *rapl_attr_groups[] = {
+ &rapl_pmu_attr_group,
+ &rapl_pmu_format_group,
+ &rapl_pmu_events_group,
+ 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_exit_cpu(int cpu)
+{
+ int i, phys_id = topology_physical_package_id(cpu);
+
+ /* if CPU not in RAPL mask, nothing to do */
+ if (!cpumask_test_and_clear_cpu(cpu, &rapl_cpu_mask))
+ return;
+
+ /* find a new cpu on same package */
+ for_each_online_cpu(i) {
+ if (i == cpu || i == 0)
+ continue;
+ if (phys_id == topology_physical_package_id(i)) {
+ cpumask_set_cpu(i, &rapl_cpu_mask);
+ break;
+ }
+ }
+
+ WARN_ON(cpumask_empty(&rapl_cpu_mask));
+}
+
+static void rapl_init_cpu(int cpu)
+{
+ int i, phys_id = topology_physical_package_id(cpu);
+
+ spin_lock(&rapl_hotplug_lock);
+
+ /* 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);
+
+ spin_unlock(&rapl_hotplug_lock);
+}
+
+static int rapl_cpu_prepare(int cpu)
+{
+ struct rapl_pmu *pmu = per_cpu(rapl_pmu, cpu);
+ int phys_id = topology_physical_package_id(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;
+
+ spin_lock_init(&pmu->lock);
+ atomic_set(&pmu->refcnt, 1);
+
+ INIT_LIST_HEAD(&pmu->active_list);
+
+ pmu->phys_id = phys_id;
+ /*
+ * grab power unit as: 1/2^unit Joules
+ *
+ * we cache in local PMU instance
+ */
+ rdmsrl(MSR_RAPL_POWER_UNIT, pmu->hw_unit);
+ pmu->hw_unit = (pmu->hw_unit >> 8) & 0x1FULL;
+
+ /* set RAPL pmu for this cpu for now */
+ per_cpu(rapl_pmu_kfree, cpu) = NULL;
+ per_cpu(rapl_pmu, cpu) = pmu;
+
+ return 0;
+}
+
+static int rapl_cpu_starting(int cpu)
+{
+ struct rapl_pmu *pmu2;
+ struct rapl_pmu *pmu1 = per_cpu(rapl_pmu, cpu);
+ int i, phys_id = topology_physical_package_id(cpu);
+
+ if (pmu1)
+ return 0;
+
+ spin_lock(&rapl_hotplug_lock);
+
+ for_each_online_cpu(i) {
+ pmu2 = per_cpu(rapl_pmu, i);
+
+ if (!pmu2 || i == cpu)
+ continue;
+
+ if (pmu2->phys_id == phys_id) {
+ per_cpu(rapl_pmu, cpu) = pmu2;
+ per_cpu(rapl_pmu_kfree, cpu) = pmu1;
+ atomic_inc(&pmu2->refcnt);
+ break;
+ }
+ }
+ spin_unlock(&rapl_hotplug_lock);
+ return 0;
+}
+
+static int rapl_cpu_dying(int cpu)
+{
+ struct rapl_pmu *pmu = per_cpu(rapl_pmu, cpu);
+ struct perf_event *event, *tmp;
+
+ if (!pmu)
+ return 0;
+
+ spin_lock(&rapl_hotplug_lock);
+
+ /*
+ * stop all syswide RAPL events on that CPU
+ * as a consequence also stops the hrtimer
+ */
+ list_for_each_entry_safe(event, tmp, &pmu->active_list, active_entry) {
+ rapl_pmu_event_stop(event, PERF_EF_UPDATE);
+ }
+
+ per_cpu(rapl_pmu, cpu) = NULL;
+
+ if (atomic_dec_and_test(&pmu->refcnt))
+ kfree(pmu);
+
+ spin_unlock(&rapl_hotplug_lock);
+ return 0;
+}
+
+static int rapl_cpu_notifier(struct notifier_block *self,
+ unsigned long action, void *hcpu)
+{
+ unsigned int cpu = (long)hcpu;
+
+ /* allocate/free data structure for uncore box */
+ switch (action & ~CPU_TASKS_FROZEN) {
+ case CPU_UP_PREPARE:
+ rapl_cpu_prepare(cpu);
+ break;
+ case CPU_STARTING:
+ rapl_cpu_starting(cpu);
+ break;
+ case CPU_UP_CANCELED:
+ case CPU_DYING:
+ rapl_cpu_dying(cpu);
+ break;
+ case CPU_ONLINE:
+ kfree(per_cpu(rapl_pmu_kfree, cpu));
+ per_cpu(rapl_pmu_kfree, cpu) = NULL;
+ break;
+ case CPU_DEAD:
+ per_cpu(rapl_pmu, cpu) = NULL;
+ break;
+ default:
+ break;
+ }
+
+ /* select the cpu that collects uncore events */
+ switch (action & ~CPU_TASKS_FROZEN) {
+ case CPU_DOWN_FAILED:
+ case CPU_STARTING:
+ rapl_init_cpu(cpu);
+ break;
+ case CPU_DOWN_PREPARE:
+ rapl_exit_cpu(cpu);
+ break;
+ default:
+ break;
+ }
+
+ return NOTIFY_OK;
+}
+
+static const struct x86_cpu_id rapl_cpu_match[] = {
+ [0] = { .vendor = X86_VENDOR_INTEL, .family = 6 },
+ [1] = {},
+};
+static int __init rapl_pmu_init(void)
+{
+ struct rapl_pmu *pmu;
+ int i, cpu, ret;
+
+ /*
+ * check for Intel processor family 6
+ */
+ if (!x86_match_cpu(rapl_cpu_match))
+ return 0;
+
+ /* check supported CPU */
+ switch (boot_cpu_data.x86_model) {
+ case 42: /* Sandy Bridge */
+ case 58: /* Ivy Bridge */
+ case 60: /* Haswell */
+ rapl_cntr_mask = RAPL_IDX_CLN;
+ rapl_pmu_events_group.attrs = rapl_events_cln_attr;
+ break;
+ case 45: /* Sandy Bridge-EP */
+ case 62: /* IvyTown */
+ rapl_cntr_mask = RAPL_IDX_SRV;
+ rapl_pmu_events_group.attrs = rapl_events_srv_attr;
+ break;
+
+ default:
+ /* unsupported */
+ return 0;
+ }
+ get_online_cpus();
+
+ for_each_online_cpu(cpu) {
+ int phys_id = topology_physical_package_id(cpu);
+
+ /* save on prepare by only calling prepare for new phys_id */
+ for_each_cpu(i, &rapl_cpu_mask) {
+ if (phys_id == topology_physical_package_id(i)) {
+ phys_id = -1;
+ break;
+ }
+ }
+ if (phys_id < 0) {
+ pmu = per_cpu(rapl_pmu, i);
+ if (pmu) {
+ per_cpu(rapl_pmu, cpu) = pmu;
+ atomic_inc(&pmu->refcnt);
+ }
+ continue;
+ }
+ rapl_cpu_prepare(cpu);
+ cpumask_set_cpu(cpu, &rapl_cpu_mask);
+ }
+
+ perf_cpu_notifier(rapl_cpu_notifier);
+
+ ret = perf_pmu_register(&rapl_pmu_class, "power", -1);
+ WARN_ON(ret);
+ if (!ret) {
+ pr_info("RAPL PMU detected, registration failed, RAPL PMU disabled\n");
+ put_online_cpus();
+ return -1;
+ }
+
+ pmu = __get_cpu_var(rapl_pmu);
+
+ pr_info("RAPL PMU detected, hw unit 2^-%d Joules,"
+ " API unit is 2^-32 Joules,"
+ " %d fixed counters\n",
+ pmu->hw_unit,
+ hweight32(rapl_cntr_mask));
+
+ put_online_cpus();
+
+ return 0;
+}
+device_initcall(rapl_pmu_init);
--
1.7.9.5
On Wed, Oct 23, 2013 at 02:58:04PM +0200, Stephane Eranian wrote:
SNIP
> +
> + perf_cpu_notifier(rapl_cpu_notifier);
> +
> + ret = perf_pmu_register(&rapl_pmu_class, "power", -1);
> + WARN_ON(ret);
> + if (!ret) {
> + pr_info("RAPL PMU detected, registration failed, RAPL PMU disabled\n");
> + put_online_cpus();
> + return -1;
> + }
should above rather be:
if (WARN_ON(ret)) {
pr_info("RAPL PMU detected, registration failed, RAPL PMU disabled\n");
jirka
On Wed, Oct 23, 2013 at 02:58:04PM +0200, Stephane Eranian wrote:
SNIP
> + pmu = per_cpu(rapl_pmu, i);
> + if (pmu) {
> + per_cpu(rapl_pmu, cpu) = pmu;
> + atomic_inc(&pmu->refcnt);
> + }
> + continue;
> + }
> + rapl_cpu_prepare(cpu);
> + cpumask_set_cpu(cpu, &rapl_cpu_mask);
> + }
> +
> + perf_cpu_notifier(rapl_cpu_notifier);
hum, this should be rather called below only if we succeed
with the perf_pmu_register
> +
> + ret = perf_pmu_register(&rapl_pmu_class, "power", -1);
> + WARN_ON(ret);
> + if (!ret) {
> + pr_info("RAPL PMU detected, registration failed, RAPL PMU disabled\n");
> + put_online_cpus();
> + return -1;
> + }
> +
> + pmu = __get_cpu_var(rapl_pmu);
> +
> + pr_info("RAPL PMU detected, hw unit 2^-%d Joules,"
> + " API unit is 2^-32 Joules,"
> + " %d fixed counters\n",
> + pmu->hw_unit,
> + hweight32(rapl_cntr_mask));
> +
> + put_online_cpus();
> +
> + return 0;
> +}
> +device_initcall(rapl_pmu_init);
> --
> 1.7.9.5
>
On Wed, Oct 23, 2013 at 02:58:04PM +0200, Stephane Eranian wrote:
SNIP
> +
> +static void rapl_init_cpu(int cpu)
> +{
> + int i, phys_id = topology_physical_package_id(cpu);
> +
> + spin_lock(&rapl_hotplug_lock);
> +
> + /* check if phys_is is already covered */
> + for_each_cpu(i, &rapl_cpu_mask) {
> + if (phys_id == topology_physical_package_id(i))
> + return;
missing 'spin_unlock(&rapl_hotplug_lock)' above
> + }
> + /* was not found, so add it */
> + cpumask_set_cpu(cpu, &rapl_cpu_mask);
> +
> + spin_unlock(&rapl_hotplug_lock);
> +}
> +
On Wed, Oct 23, 2013 at 02:58:02PM +0200, Stephane Eranian wrote:
> This patch adds a new fields to the struct perf_event.
> It is intended to be used to chain events which are
> active (enabled). It helps in the hardware layer
> for PMU which do not have actual counter restrictions, i.e.,
> free running read-only counters. Active events are chained
> as opposed to being tracked via the counter they use.
>
> Signed-off-by: Stephane Eranian <[email protected]>
> ---
> include/linux/perf_event.h | 1 +
> kernel/events/core.c | 1 +
> 2 files changed, 2 insertions(+)
>
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index 2e069d1..a376384 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -435,6 +435,7 @@ struct perf_event {
> struct perf_cgroup *cgrp; /* cgroup event is attach to */
> int cgrp_defer_enabled;
> #endif
> + struct list_head active_entry;
Could this be in union with 'hlist_entry' ? It looks
as 'same purpose' and 'mutualy exclusive stuff.
jirka
On Wed, Oct 23, 2013 at 02:58:05PM +0200, Stephane Eranian wrote:
> The RAPL PMU counters do not interrupt on overflow.
> Therefore, the kernel needs to poll the counters
> to avoid missing an overflow. This patch adds
> the hrtimer code to do this.
>
> The timer internval is calculated at boot time
> based on the power unit used by the HW.
>
> Signed-off-by: Stephane Eranian <[email protected]>
> ---
> arch/x86/kernel/cpu/perf_event_intel_rapl.c | 75 +++++++++++++++++++++++++--
> 1 file changed, 70 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/kernel/cpu/perf_event_intel_rapl.c b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
> index 3d71d39..ed0566a 100644
> --- a/arch/x86/kernel/cpu/perf_event_intel_rapl.c
> +++ b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
> @@ -92,11 +92,13 @@ static struct kobj_attribute format_attr_##_var = \
>
> struct rapl_pmu {
> spinlock_t lock;
> - atomic_t refcnt;
> int hw_unit; /* 1/2^hw_unit Joule */
> - int phys_id;
> - int n_active; /* number of active events */
> + struct hrtimer hrtimer;
> struct list_head active_list;
> + ktime_t timer_interval; /* in ktime_t unit */
> + int n_active; /* number of active events */
> + int phys_id;
> + atomic_t refcnt;
> };
>
> static struct pmu rapl_pmu_class;
> @@ -161,6 +163,47 @@ static u64 rapl_event_update(struct perf_event *event)
> return new_raw_count;
> }
>
> +static void rapl_start_hrtimer(struct rapl_pmu *pmu)
> +{
> + __hrtimer_start_range_ns(&pmu->hrtimer,
> + pmu->timer_interval, 0,
> + HRTIMER_MODE_REL_PINNED, 0);
> +}
> +
> +static void rapl_stop_hrtimer(struct rapl_pmu *pmu)
> +{
> + hrtimer_cancel(&pmu->hrtimer);
> +}
> +
> +static enum hrtimer_restart rapl_hrtimer_handle(struct hrtimer *hrtimer)
> +{
> + struct rapl_pmu *pmu = container_of(hrtimer, struct rapl_pmu, hrtimer);
> + struct perf_event *event;
> + unsigned long flags;
> +
> + if (!pmu->n_active)
> + return HRTIMER_NORESTART;
> +
> + spin_lock_irqsave(&pmu->lock, flags);
> +
> + list_for_each_entry(event, &pmu->active_list, active_entry) {
> + rapl_event_update(event);
> + }
hi,
I dont fully understand the reason for the timer,
I'm probably missing something..
- the timer calls rapl_event_update for all defined events
- but rapl_pmu_event_read calls rapl_event_update any time the
event is read (sys_read)
The rapl_event_update only read msr and updates
event->count|hw,prev_count.
What's the timer purpose then?
thanks for info,
jirka
On Fri, Oct 25, 2013 at 4:56 PM, Jiri Olsa <[email protected]> wrote:
>
> On Wed, Oct 23, 2013 at 02:58:02PM +0200, Stephane Eranian wrote:
> > This patch adds a new fields to the struct perf_event.
> > It is intended to be used to chain events which are
> > active (enabled). It helps in the hardware layer
> > for PMU which do not have actual counter restrictions, i.e.,
> > free running read-only counters. Active events are chained
> > as opposed to being tracked via the counter they use.
> >
> > Signed-off-by: Stephane Eranian <[email protected]>
> > ---
> > include/linux/perf_event.h | 1 +
> > kernel/events/core.c | 1 +
> > 2 files changed, 2 insertions(+)
> >
> > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> > index 2e069d1..a376384 100644
> > --- a/include/linux/perf_event.h
> > +++ b/include/linux/perf_event.h
> > @@ -435,6 +435,7 @@ struct perf_event {
> > struct perf_cgroup *cgrp; /* cgroup event is attach to */
> > int cgrp_defer_enabled;
> > #endif
> > + struct list_head active_entry;
>
> Could this be in union with 'hlist_entry' ? It looks
> as 'same purpose' and 'mutualy exclusive stuff.
>
You're saying that I could use the hlist_entry field because
it is currently only used by the sw events in the generic layer.
But it seems to be a complicated rcu list for the purpose here.
On Fri, Oct 25, 2013 at 1:14 PM, Jiri Olsa <[email protected]> wrote:
> On Wed, Oct 23, 2013 at 02:58:04PM +0200, Stephane Eranian wrote:
>
> SNIP
>
>> +
>> +static void rapl_init_cpu(int cpu)
>> +{
>> + int i, phys_id = topology_physical_package_id(cpu);
>> +
>> + spin_lock(&rapl_hotplug_lock);
>> +
>> + /* check if phys_is is already covered */
>> + for_each_cpu(i, &rapl_cpu_mask) {
>> + if (phys_id == topology_physical_package_id(i))
>> + return;
>
> missing 'spin_unlock(&rapl_hotplug_lock)' above
>
Good catch. I fixed that now.
>> + }
>> + /* was not found, so add it */
>> + cpumask_set_cpu(cpu, &rapl_cpu_mask);
>> +
>> + spin_unlock(&rapl_hotplug_lock);
>> +}
>> +
On Fri, Oct 25, 2013 at 7:44 PM, Jiri Olsa <[email protected]> wrote:
> On Wed, Oct 23, 2013 at 02:58:05PM +0200, Stephane Eranian wrote:
>> The RAPL PMU counters do not interrupt on overflow.
>> Therefore, the kernel needs to poll the counters
>> to avoid missing an overflow. This patch adds
>> the hrtimer code to do this.
>>
>> The timer internval is calculated at boot time
>> based on the power unit used by the HW.
>>
>> Signed-off-by: Stephane Eranian <[email protected]>
>> ---
>> arch/x86/kernel/cpu/perf_event_intel_rapl.c | 75 +++++++++++++++++++++++++--
>> 1 file changed, 70 insertions(+), 5 deletions(-)
>>
>> diff --git a/arch/x86/kernel/cpu/perf_event_intel_rapl.c b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
>> index 3d71d39..ed0566a 100644
>> --- a/arch/x86/kernel/cpu/perf_event_intel_rapl.c
>> +++ b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
>> @@ -92,11 +92,13 @@ static struct kobj_attribute format_attr_##_var = \
>>
>> struct rapl_pmu {
>> spinlock_t lock;
>> - atomic_t refcnt;
>> int hw_unit; /* 1/2^hw_unit Joule */
>> - int phys_id;
>> - int n_active; /* number of active events */
>> + struct hrtimer hrtimer;
>> struct list_head active_list;
>> + ktime_t timer_interval; /* in ktime_t unit */
>> + int n_active; /* number of active events */
>> + int phys_id;
>> + atomic_t refcnt;
>> };
>>
>> static struct pmu rapl_pmu_class;
>> @@ -161,6 +163,47 @@ static u64 rapl_event_update(struct perf_event *event)
>> return new_raw_count;
>> }
>>
>> +static void rapl_start_hrtimer(struct rapl_pmu *pmu)
>> +{
>> + __hrtimer_start_range_ns(&pmu->hrtimer,
>> + pmu->timer_interval, 0,
>> + HRTIMER_MODE_REL_PINNED, 0);
>> +}
>> +
>> +static void rapl_stop_hrtimer(struct rapl_pmu *pmu)
>> +{
>> + hrtimer_cancel(&pmu->hrtimer);
>> +}
>> +
>> +static enum hrtimer_restart rapl_hrtimer_handle(struct hrtimer *hrtimer)
>> +{
>> + struct rapl_pmu *pmu = container_of(hrtimer, struct rapl_pmu, hrtimer);
>> + struct perf_event *event;
>> + unsigned long flags;
>> +
>> + if (!pmu->n_active)
>> + return HRTIMER_NORESTART;
>> +
>> + spin_lock_irqsave(&pmu->lock, flags);
>> +
>> + list_for_each_entry(event, &pmu->active_list, active_entry) {
>> + rapl_event_update(event);
>> + }
>
> hi,
> I dont fully understand the reason for the timer,
> I'm probably missing something..
>
The reason is rather simple and is similar to what happens with uncore.
The counter are narrow, 32-bit and there is no interrupt capability. We
need to poll the counters and accumulate in the sw counter to avoid missing
an overflow.
> - the timer calls rapl_event_update for all defined events
No, only for the defined RAPL events which is what we want.
> - but rapl_pmu_event_read calls rapl_event_update any time the
> event is read (sys_read)
>
Yes, but we want to prevent missing a counter overflow. It may happen
if the counter counts in a unit which increments fast.
> The rapl_event_update only read msr and updates
> event->count|hw,prev_count.
No, it does update the count:
local64_add(sdelta, &event->count);
On Sat, Oct 26, 2013 at 06:57:37PM +0200, Stephane Eranian wrote:
> On Fri, Oct 25, 2013 at 4:56 PM, Jiri Olsa <[email protected]> wrote:
> >
> > On Wed, Oct 23, 2013 at 02:58:02PM +0200, Stephane Eranian wrote:
> > > This patch adds a new fields to the struct perf_event.
> > > It is intended to be used to chain events which are
> > > active (enabled). It helps in the hardware layer
> > > for PMU which do not have actual counter restrictions, i.e.,
> > > free running read-only counters. Active events are chained
> > > as opposed to being tracked via the counter they use.
> > >
> > > Signed-off-by: Stephane Eranian <[email protected]>
> > > ---
> > > include/linux/perf_event.h | 1 +
> > > kernel/events/core.c | 1 +
> > > 2 files changed, 2 insertions(+)
> > >
> > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> > > index 2e069d1..a376384 100644
> > > --- a/include/linux/perf_event.h
> > > +++ b/include/linux/perf_event.h
> > > @@ -435,6 +435,7 @@ struct perf_event {
> > > struct perf_cgroup *cgrp; /* cgroup event is attach to */
> > > int cgrp_defer_enabled;
> > > #endif
> > > + struct list_head active_entry;
> >
> > Could this be in union with 'hlist_entry' ? It looks
> > as 'same purpose' and 'mutualy exclusive stuff.
> >
> You're saying that I could use the hlist_entry field because
> it is currently only used by the sw events in the generic layer.
> But it seems to be a complicated rcu list for the purpose here.
nope, I just meant saving little space like:
union {
struct list_head active_entry;
struct hlist_node hlist_entry;
}
just a nitpick
jirka
On Sat, Oct 26, 2013 at 07:07:06PM +0200, Stephane Eranian wrote:
> On Fri, Oct 25, 2013 at 7:44 PM, Jiri Olsa <[email protected]> wrote:
> > On Wed, Oct 23, 2013 at 02:58:05PM +0200, Stephane Eranian wrote:
SNIP
> >> + list_for_each_entry(event, &pmu->active_list, active_entry) {
> >> + rapl_event_update(event);
> >> + }
> >
> > hi,
> > I dont fully understand the reason for the timer,
> > I'm probably missing something..
> >
> The reason is rather simple and is similar to what happens with uncore.
> The counter are narrow, 32-bit and there is no interrupt capability. We
> need to poll the counters and accumulate in the sw counter to avoid missing
> an overflow.
>
> > - the timer calls rapl_event_update for all defined events
>
> No, only for the defined RAPL events which is what we want.
ok, that's what I meant
>
> > - but rapl_pmu_event_read calls rapl_event_update any time the
> > event is read (sys_read)
> >
> Yes, but we want to prevent missing a counter overflow. It may happen
> if the counter counts in a unit which increments fast.
>
> > The rapl_event_update only read msr and updates
> > event->count|hw,prev_count.
> No, it does update the count:
> local64_add(sdelta, &event->count);
ah, there's the shift that takes care of the
overflowed msr value.. ok
thanks,
jirka
On Sat, Oct 26, 2013 at 7:53 PM, Jiri Olsa <[email protected]> wrote:
> On Sat, Oct 26, 2013 at 07:07:06PM +0200, Stephane Eranian wrote:
>> On Fri, Oct 25, 2013 at 7:44 PM, Jiri Olsa <[email protected]> wrote:
>> > On Wed, Oct 23, 2013 at 02:58:05PM +0200, Stephane Eranian wrote:
>
> SNIP
>
>> >> + list_for_each_entry(event, &pmu->active_list, active_entry) {
>> >> + rapl_event_update(event);
>> >> + }
>> >
>> > hi,
>> > I dont fully understand the reason for the timer,
>> > I'm probably missing something..
>> >
>> The reason is rather simple and is similar to what happens with uncore.
>> The counter are narrow, 32-bit and there is no interrupt capability. We
>> need to poll the counters and accumulate in the sw counter to avoid missing
>> an overflow.
>>
>> > - the timer calls rapl_event_update for all defined events
>>
>> No, only for the defined RAPL events which is what we want.
>
> ok, that's what I meant
>
>>
>> > - but rapl_pmu_event_read calls rapl_event_update any time the
>> > event is read (sys_read)
>> >
>> Yes, but we want to prevent missing a counter overflow. It may happen
>> if the counter counts in a unit which increments fast.
>>
>> > The rapl_event_update only read msr and updates
>> > event->count|hw,prev_count.
>> No, it does update the count:
>> local64_add(sdelta, &event->count);
>
> ah, there's the shift that takes care of the
> overflowed msr value.. ok
>
No, we try to poll the counter faster than it can possibly overflow.
On Sat, Oct 26, 2013 at 7:44 PM, Jiri Olsa <[email protected]> wrote:
> On Sat, Oct 26, 2013 at 06:57:37PM +0200, Stephane Eranian wrote:
>> On Fri, Oct 25, 2013 at 4:56 PM, Jiri Olsa <[email protected]> wrote:
>> >
>> > On Wed, Oct 23, 2013 at 02:58:02PM +0200, Stephane Eranian wrote:
>> > > This patch adds a new fields to the struct perf_event.
>> > > It is intended to be used to chain events which are
>> > > active (enabled). It helps in the hardware layer
>> > > for PMU which do not have actual counter restrictions, i.e.,
>> > > free running read-only counters. Active events are chained
>> > > as opposed to being tracked via the counter they use.
>> > >
>> > > Signed-off-by: Stephane Eranian <[email protected]>
>> > > ---
>> > > include/linux/perf_event.h | 1 +
>> > > kernel/events/core.c | 1 +
>> > > 2 files changed, 2 insertions(+)
>> > >
>> > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
>> > > index 2e069d1..a376384 100644
>> > > --- a/include/linux/perf_event.h
>> > > +++ b/include/linux/perf_event.h
>> > > @@ -435,6 +435,7 @@ struct perf_event {
>> > > struct perf_cgroup *cgrp; /* cgroup event is attach to */
>> > > int cgrp_defer_enabled;
>> > > #endif
>> > > + struct list_head active_entry;
>> >
>> > Could this be in union with 'hlist_entry' ? It looks
>> > as 'same purpose' and 'mutualy exclusive stuff.
>> >
>> You're saying that I could use the hlist_entry field because
>> it is currently only used by the sw events in the generic layer.
>> But it seems to be a complicated rcu list for the purpose here.
>
> nope, I just meant saving little space like:
>
> union {
> struct list_head active_entry;
> struct hlist_node hlist_entry;
> }
>
> just a nitpick
>
But you are relying on the fact that active_entry and hlist_entry can never be
used at the same time. You're saying this because *so far* hlist_entry is
only used for SW events and what I added is only used by RAPL.
If the goal is the same, then we should add a better description for the field:
"chain the event in the list of active events for a PMU instance on a CPU".
Handled by PMU specific code only (not generic code).
Hi,
I was thinking about the scaling issue over the week-end.
We agreed that it was necessary to export the
scaling via sysfs per event. The RAPL v3 has
an implementation of that including in the perf tool.
If we have that, then it may not be necessary anymore
to express the raw count in the 1/2^32 J unit like we
are currently doing. This loses a bit of precision. We
could as well expose the actual raw count and export
the actual unit via sysfs. For instance, on SNB/IVB the
unit is 1/2^16, but on Haswell it is 1/2^14.
I see two issues with that approach though:
- the interpretation of the raw count changes from machine to
machine and needs to ALWAYS be used with the scaling.
It is not always possible to just compare raw counts directly.
- we would need a way to express that ratio without actually
calculating it in the kernel. There are 6 ratios possible. So
we either have a lookup table with the floating point values
already computed and encoded as strings. Or we need to
add a calculator style parsing in the perf tool (or any other)
tool to parse a ratio: 1 / 65536 or any basic mathematical
expression. After all, the scaling support has to be generic.
Other events may use a different form of scaling ratios. But
that seems overkill to me.
So in the end, it may be the case that what we have not in RAPLv3
is the simplest approach.
Any opinion?
On Sat, Oct 26, 2013 at 7:00 PM, Stephane Eranian <[email protected]> wrote:
> On Fri, Oct 25, 2013 at 1:14 PM, Jiri Olsa <[email protected]> wrote:
>> On Wed, Oct 23, 2013 at 02:58:04PM +0200, Stephane Eranian wrote:
>>
>> SNIP
>>
>>> +
>>> +static void rapl_init_cpu(int cpu)
>>> +{
>>> + int i, phys_id = topology_physical_package_id(cpu);
>>> +
>>> + spin_lock(&rapl_hotplug_lock);
>>> +
>>> + /* check if phys_is is already covered */
>>> + for_each_cpu(i, &rapl_cpu_mask) {
>>> + if (phys_id == topology_physical_package_id(i))
>>> + return;
>>
>> missing 'spin_unlock(&rapl_hotplug_lock)' above
>>
> Good catch. I fixed that now.
>
>>> + }
>>> + /* was not found, so add it */
>>> + cpumask_set_cpu(cpu, &rapl_cpu_mask);
>>> +
>>> + spin_unlock(&rapl_hotplug_lock);
>>> +}
>>> +
On Mon, Oct 28, 2013 at 11:33:50AM +0100, Stephane Eranian wrote:
> If we have that, then it may not be necessary anymore
> to express the raw count in the 1/2^32 J unit like we
> are currently doing. This loses a bit of precision. We
> could as well expose the actual raw count and export
> the actual unit via sysfs. For instance, on SNB/IVB the
> unit is 1/2^16, but on Haswell it is 1/2^14.
2^-32 can losslessly express both 2^-16 and 2^-14.
Notably: 2^18/2^32 = 2^(18-32) = 2^-14.
So no, 2^-32 does not loose precision.
The only side effect of always using 2^-32 is that we can only maximally
represent 2^32 (from 64-32), whereas when using 2^-14 we could maximally
represent 2^50.
That said; 2^32 Joule ~ 4.2 GJ which is a rather large quantity of
energy; one I would hope is out there when measuring package energy
costs over any reasonable amount of time.
So the only reason to switch away from using the 32.32 fixed point would
be if someone can make a reasonable argument for why 4.2 GJ is not
sufficient and they need 1 PJ (yes, peta-joule, as in we need a private
nuclear reactor to power this CPU).
Peter,
On Mon, Oct 28, 2013 at 1:17 PM, Peter Zijlstra <[email protected]> wrote:
> On Mon, Oct 28, 2013 at 11:33:50AM +0100, Stephane Eranian wrote:
>> If we have that, then it may not be necessary anymore
>> to express the raw count in the 1/2^32 J unit like we
>> are currently doing. This loses a bit of precision. We
>> could as well expose the actual raw count and export
>> the actual unit via sysfs. For instance, on SNB/IVB the
>> unit is 1/2^16, but on Haswell it is 1/2^14.
>
> 2^-32 can losslessly express both 2^-16 and 2^-14.
>
> Notably: 2^18/2^32 = 2^(18-32) = 2^-14.
>
> So no, 2^-32 does not loose precision.
>
You are correct. No bits are lost.
> The only side effect of always using 2^-32 is that we can only maximally
> represent 2^32 (from 64-32), whereas when using 2^-14 we could maximally
> represent 2^50.
>
> That said; 2^32 Joule ~ 4.2 GJ which is a rather large quantity of
> energy; one I would hope is out there when measuring package energy
> costs over any reasonable amount of time.
>
> So the only reason to switch away from using the 32.32 fixed point would
> be if someone can make a reasonable argument for why 4.2 GJ is not
> sufficient and they need 1 PJ (yes, peta-joule, as in we need a private
> nuclear reactor to power this CPU).
I think we are fine with what we have. Simple, no precision lost, constant
user-visible scaling factor easy to export as a string to user tools. Comparison
of raw count possible directly.
I will post v4 very soon.
Thanks.