provider: add 'profile' provider

The profile provider supports profiling by allowing the user to specify
how many times it will fire pre-second. Values of 1-1000 are supported,
and the profile provider supports two probe formats as it's original
implementation commit 12b8ca2("profile provider support").

  * profile:[N]hz: Profile on all CPUs N times per second
  * profile:[C]:[N]hz: Profile on CPU C N times per second

Test this changes with 4 cores x86-64 PC as below.

$ sudo ./ply 'profile:2:100hz { @[cpu]=count(); } i:1s { print(@); }'
@:
{  2 }: 100

@:
{  2 }: 200

@:
{  2 }: 300

$ sudo ./ply 'profile:100hz { @[cpu]=count(); } i:1s { print(@); }'
@:
{  0 }: 100
{  1 }: 100
{  2 }: 100
{  3 }: 100

@:
{  0 }: 200
{  1 }: 200
{  2 }: 200
{  3 }: 200

@:
{  0 }: 300
{  1 }: 300
{  2 }: 300
{  3 }: 300

Change-Id: Ie669db20ed52947f5c75bf91c34f385a64029b19
Signed-off-by: Ism Hong <[email protected]>

include/ply/perf_event.h

@@ -15,6 +15,8 @@ int perf_event_attach(struct ply_probe *pb, const char *name,
 		      int task_mode);
 int perf_event_attach_raw(struct ply_probe *pb, int type, unsigned long config,
 			  unsigned long long period, int task_mode);
+int perf_event_attach_profile(struct ply_probe *pb, int cpu,
+			  unsigned long long freq);
 
 int perf_event_enable (int group_fd);
 int perf_event_disable(int group_fd);

man/ply.1.ronn

@@ -360,6 +360,15 @@ Examples:
   * _interval:1_: Called for every second
   * _interval:500ms_: Called for every 500 milli-second
 
+### profile
+
+The profile provider supports profiling by allowing the user to specify
+how many times it will fire per-second.  Values of 1-1000 are supported,
+and the profile provider supports two probe formats:
+
+  * profile:[N]hz: Profile on all CPUs N times per second
+  * profile:[C]:[N]hz: Profile on CPU C N times per second
+
 
 ## EXAMPLE
 

src/libply/Makefile.am

@@ -41,6 +41,7 @@ libply_la_SOURCES     = 	\
 	provider/kprobe.c	\
 	provider/kprobe.h	\
 	provider/kretprobe.c	\
+	provider/profile.c	\
 	provider/special.c	\
 	provider/tracepoint.c	\
 	provider/xprobe.c	\

src/libply/aux/perf_event.c

@@ -99,6 +99,29 @@ int perf_event_attach_raw(struct ply_probe *pb, int type, unsigned long config,
 	return fd;
 }
 
+int perf_event_attach_profile(struct ply_probe *pb, int cpu,
+			  unsigned long long freq)
+{
+	struct perf_event_attr attr = {};
+	int fd;
+
+	attr.type = PERF_TYPE_SOFTWARE;
+	attr.config = PERF_COUNT_SW_CPU_CLOCK;
+	attr.sample_freq = freq;
+	attr.freq = 1;
+
+	fd = perf_event_open(&attr, -1, cpu, -1, 0);
+	if (fd < 0)
+		return -errno;
+
+	if (ioctl(fd, PERF_EVENT_IOC_SET_BPF, pb->bpf_fd)) {
+		close(fd);
+		return -errno;
+	}
+
+	return fd;
+}
+
 int perf_event_enable(int group_fd)
 {
 	if (ioctl(group_fd, PERF_EVENT_IOC_ENABLE, PERF_IOC_FLAG_GROUP))

src/libply/provider.c

@@ -22,6 +22,7 @@ extern struct provider built_in;
 extern struct provider begin_provider;
 extern struct provider end_provider;
 extern struct provider interval;
+extern struct provider profile;
 
 struct provider *provider_get(const char *name)
 {
@@ -45,6 +46,7 @@ void provider_init(void)
 	SLIST_INSERT_HEAD(&heads, &begin_provider, entry);
 	SLIST_INSERT_HEAD(&heads, &built_in, entry);
 	SLIST_INSERT_HEAD(&heads, &interval, entry);
+	SLIST_INSERT_HEAD(&heads, &profile, entry);
 	SLIST_INSERT_HEAD(&heads, &tracepoint, entry);
 	SLIST_INSERT_HEAD(&heads, &kretprobe, entry);
 	/* place kprobe at head so that 'k' can match first. */

src/libply/provider/profile.c

@@ -0,0 +1,123 @@
+/*
+ * Copyright  Ism Hong <[email protected]>
+ *
+ * SPDX-License-Identifier: GPL-2.0
+ */
+
+#include <assert.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <ply/ply.h>
+#include <ply/internal.h>
+
+
+struct profile_data {
+	int cpu;
+	int ncpus;
+	unsigned long long freq;
+	int *evfds;
+};
+
+static int profile_sym_alloc(struct ply_probe *pb, struct node *n)
+{
+	return -ENOENT;
+}
+
+static int profile_probe(struct ply_probe *pb)
+{
+	int cpu = -1, ncpus = 0;
+	struct profile_data *data;
+	int freq = -1;
+
+	/*
+	 * Expected format is either profile:[n]hz where n is a number between
+	 * 1 and 1000, or profile:[c]:[n]hz where c is the CPU to profile.
+	 */
+	if (sscanf(pb->probe, "profile:%d:%dhz", &cpu, &freq) != 2) {
+		cpu = -1;
+		if (sscanf(pb->probe, "profile:%dhz", &freq) != 1)
+			return -EINVAL;
+	}
+
+	if (freq < 0 || freq > 1000)
+		return -EINVAL;
+
+	ncpus = sysconf(_SC_NPROCESSORS_ONLN);
+
+	if (cpu < -1 || cpu > ncpus)
+		return -EINVAL;
+
+	if (cpu >= 0)
+		ncpus = 1;
+
+	data = calloc(1, sizeof(*data));
+	if (!data)
+		return -ENOMEM;
+
+	data->evfds = calloc(ncpus, sizeof (int));
+	if (!data->evfds) {
+		free(data);
+		return -ENOMEM;
+	}
+
+	data->freq = (unsigned long long)freq;
+	data->cpu = cpu;
+	data->ncpus = ncpus ;
+
+	pb->provider_data = data;
+	return 0;
+}
+
+static int profile_attach(struct ply_probe *pb)
+{
+	struct profile_data *data = pb->provider_data;
+	int cpu;
+
+	if (data->cpu != -1) {
+		data->evfds[0] = perf_event_attach_profile(pb, data->cpu,
+						 data->freq);
+		if (data->evfds[0] < 0) {
+			_e("%s: Unable to attach profile probe: %s\n",
+			   pb->probe, strerror(errno));
+			return data->evfds[0];
+		}
+	} else {
+		for (cpu = 0; cpu < data->ncpus; cpu++) {
+			data->evfds[cpu] = perf_event_attach_profile(pb, cpu, data->freq);
+			if (data->evfds[cpu] < 0) {
+				_e("%s: Unable to attach profile probe: %s\n",
+						pb->probe, strerror(errno));
+				return data->evfds[cpu];
+			}
+		}
+	}
+
+	return 0;
+}
+
+static int profile_detach(struct ply_probe *pb)
+{
+	struct profile_data *data = pb->provider_data;
+
+	for (int i = 0; i < data->ncpus; i++) {
+		if (data->evfds[i] > 0)
+			close(data->evfds[i]);
+	}
+	free(data->evfds);
+	free(data);
+
+	return 0;
+}
+
+struct provider profile = {
+	.name = "profile",
+	.prog_type = BPF_PROG_TYPE_PERF_EVENT,
+
+	.sym_alloc = profile_sym_alloc,
+	.probe 	   = profile_probe,
+
+	.attach = profile_attach,
+	.detach = profile_detach,
+};

test/rootfs/lib/ply/test.sh

@@ -84,4 +84,13 @@ cat /tmp/tracepoint-dyn | awk '
     END      { exit(!(unames >= 10)); }' \
 || fail "at least 10 unames" "$(cat /tmp/tracepoint-dyn)"
 
+case=profile
+ply 'profile:0:100hz { @["profile_test"]=count(); }
+     i:1s { exit(0);}' >/tmp/profile \
+&& \
+cat /tmp/profile | awk -F': ' '
+    /profile_test/  { count = $2; }
+    END             { exit(count!=100); }' \
+|| fail "count should be 100 for 100hz sampling in 1 second" "$(cat /tmp/profile)"
+
 exit $total_fails