overlay: Read power from perf_events

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

8 files changed, 131 insertions, 73 deletions

diff --git a/overlay/Makefile.am b/overlay/Makefile.am
index 41f42cc8..8f829fe4 100644
--- a/overlay/Makefile.am
+++ b/overlay/Makefile.am
@@ -29,6 +29,8 @@ intel_gpu_overlay_SOURCES = \
 	igfx.c \
 	overlay.h \
 	overlay.c \
+	perf.h \
+	perf.c \
 	power.h \
 	power.c \
 	rc6.h \
diff --git a/overlay/gpu-perf.c b/overlay/gpu-perf.c
index 142357cb..fc215634 100644
--- a/overlay/gpu-perf.c
+++ b/overlay/gpu-perf.c
@@ -24,7 +24,6 @@
 
 #include <stdint.h>
 #include <stdbool.h>
-#include <linux/perf_event.h>
 #include <sys/types.h>
 #include <sys/mman.h>
 #include <sys/ioctl.h>
@@ -35,6 +34,7 @@
 #include <fcntl.h>
 #include <errno.h>
 
+#include "perf.h"
 #include "gpu-perf.h"
 #include "debugfs.h"
 
@@ -59,27 +59,6 @@ struct sample_event {
 	uint32_t raw[0];
 };
 
-static int
-perf_event_open(struct perf_event_attr *attr,
-		pid_t pid,
-		int cpu,
-		int group_fd,
-		unsigned long flags)
-{
-#ifndef __NR_perf_event_open
-#if defined(__i386__)
-#define __NR_perf_event_open 336
-#elif defined(__x86_64__)
-#define __NR_perf_event_open 298
-#else
-#define __NR_perf_event_open 0
-#endif
-#endif
-
-    attr->size = sizeof(*attr);
-    return syscall(__NR_perf_event_open, attr, pid, cpu, group_fd, flags);
-}
-
 static uint64_t tracepoint_id(const char *sys, const char *name)
 {
 	char buf[1024];
diff --git a/overlay/gpu-top.c b/overlay/gpu-top.c
index 751d5fcb..1efbbfd3 100644
--- a/overlay/gpu-top.c
+++ b/overlay/gpu-top.c
@@ -22,7 +22,6 @@
  *
  */
 
-#include <linux/perf_event.h>
 #include <stdint.h>
 #include <stdlib.h>
 #include <stdio.h>
@@ -31,6 +30,7 @@
 #include <fcntl.h>
 #include <errno.h>
 
+#include "perf.h"
 #include "igfx.h"
 #include "gpu-top.h"
 
@@ -46,45 +46,6 @@
 #define I915_PERF_RING_WAIT(n) (__I915_PERF_RING(n) + 1)
 #define I915_PERF_RING_SEMA(n) (__I915_PERF_RING(n) + 2)
 
-static int
-perf_event_open(struct perf_event_attr *attr,
-		pid_t pid,
-		int cpu,
-		int group_fd,
-		unsigned long flags)
-{
-#ifndef __NR_perf_event_open
-#if defined(__i386__)
-#define __NR_perf_event_open 336
-#elif defined(__x86_64__)
-#define __NR_perf_event_open 298
-#else
-#define __NR_perf_event_open 0
-#endif
-#endif
-    attr->size = sizeof(*attr);
-    return syscall(__NR_perf_event_open, attr, pid, cpu, group_fd, flags);
-}
-
-static uint64_t i915_type_id(void)
-{
-	char buf[1024];
-	int fd, n;
-
-	fd = open("/sys/bus/event_source/devices/i915/type", 0);
-	if (fd < 0) {
-		n = -1;
-	} else {
-		n = read(fd, buf, sizeof(buf)-1);
-		close(fd);
-	}
-	if (n < 0)
-		return 0;
-
-	buf[n] = '\0';
-	return strtoull(buf, 0, 0);
-}
-
 static int perf_i915_open(int config, int group)
 {
 	struct perf_event_attr attr;
@@ -106,9 +67,9 @@ static int perf_i915_open(int config, int group)
 static int perf_init(struct gpu_top *gt)
 {
 	const char *names[] = {
-		"render",
-		"bitstream",
-		"bliter",
+		"RCS",
+		"VCS",
+		"BCS",
 		NULL,
 	};
 	int n;
diff --git a/overlay/overlay.c b/overlay/overlay.c
index 27281fd6..0c013cff 100644
--- a/overlay/overlay.c
+++ b/overlay/overlay.c
@@ -742,7 +742,7 @@ int main(int argc, char **argv)
 	config_init(&config);
 
 	opterr = 0;
-	while ((i = getopt_long(argc, argv, "c:f:", long_options, &index)) != -1) {
+	while ((i = getopt_long(argc, argv, "c:f", long_options, &index)) != -1) {
 		switch (i) {
 		case 'c':
 			config_parse_string(&config, optarg);
diff --git a/overlay/perf.c b/overlay/perf.c
new file mode 100644
index 00000000..b8fdc675
--- /dev/null
+++ b/overlay/perf.c
@@ -0,0 +1,26 @@
+#include <stdint.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <stdlib.h>
+
+#include "perf.h"
+
+uint64_t i915_type_id(void)
+{
+	char buf[1024];
+	int fd, n;
+
+	fd = open("/sys/bus/event_source/devices/i915/type", 0);
+	if (fd < 0) {
+		n = -1;
+	} else {
+		n = read(fd, buf, sizeof(buf)-1);
+		close(fd);
+	}
+	if (n < 0)
+		return 0;
+
+	buf[n] = '\0';
+	return strtoull(buf, 0, 0);
+}
+
diff --git a/overlay/perf.h b/overlay/perf.h
new file mode 100644
index 00000000..e2be0f7e
--- /dev/null
+++ b/overlay/perf.h
@@ -0,0 +1,60 @@
+#ifndef I915_PERF_H
+#define I915_PERF_H
+
+#include <linux/perf_event.h>
+
+#define I915_SAMPLE_BUSY	0
+#define I915_SAMPLE_WAIT	1
+#define I915_SAMPLE_SEMA	2
+
+#define I915_SAMPLE_RCS		0
+#define I915_SAMPLE_VCS		1
+#define I915_SAMPLE_BCS		2
+#define I915_SAMPLE_VECS	3
+
+#define __I915_PERF_COUNT(ring, id) ((ring) << 4 | (id))
+
+#define I915_PERF_COUNT_RCS_BUSY __I915_PERF_COUNT(I915_SAMPLE_RCS, I915_SAMPLE_BUSY)
+#define I915_PERF_COUNT_RCS_WAIT __I915_PERF_COUNT(I915_SAMPLE_RCS, I915_SAMPLE_WAIT)
+#define I915_PERF_COUNT_RCS_SEMA __I915_PERF_COUNT(I915_SAMPLE_RCS, I915_SAMPLE_SEMA)
+
+#define I915_PERF_COUNT_VCS_BUSY __I915_PERF_COUNT(I915_SAMPLE_VCS, I915_SAMPLE_BUSY)
+#define I915_PERF_COUNT_VCS_WAIT __I915_PERF_COUNT(I915_SAMPLE_VCS, I915_SAMPLE_WAIT)
+#define I915_PERF_COUNT_VCS_SEMA __I915_PERF_COUNT(I915_SAMPLE_VCS, I915_SAMPLE_SEMA)
+
+#define I915_PERF_COUNT_BCS_BUSY __I915_PERF_COUNT(I915_SAMPLE_BCS, I915_SAMPLE_BUSY)
+#define I915_PERF_COUNT_BCS_WAIT __I915_PERF_COUNT(I915_SAMPLE_BCS, I915_SAMPLE_WAIT)
+#define I915_PERF_COUNT_BCS_SEMA __I915_PERF_COUNT(I915_SAMPLE_BCS, I915_SAMPLE_SEMA)
+
+#define I915_PERF_COUNT_VECS_BUSY __I915_PERF_COUNT(I915_SAMPLE_VECS, I915_SAMPLE_BUSY)
+#define I915_PERF_COUNT_VECS_WAIT __I915_PERF_COUNT(I915_SAMPLE_VECS, I915_SAMPLE_WAIT)
+#define I915_PERF_COUNT_VECS_SEMA __I915_PERF_COUNT(I915_SAMPLE_VECS, I915_SAMPLE_SEMA)
+
+#define I915_PERF_ACTUAL_FREQUENCY 32
+#define I915_PERF_REQUESTED_FREQUENCY 33
+#define I915_PERF_ENERGY 34
+#define I915_PERF_INTERRUPTS 35
+
+static inline int
+perf_event_open(struct perf_event_attr *attr,
+		pid_t pid,
+		int cpu,
+		int group_fd,
+		unsigned long flags)
+{
+#ifndef __NR_perf_event_open
+#if defined(__i386__)
+#define __NR_perf_event_open 336
+#elif defined(__x86_64__)
+#define __NR_perf_event_open 298
+#else
+#define __NR_perf_event_open 0
+#endif
+#endif
+    attr->size = sizeof(*attr);
+    return syscall(__NR_perf_event_open, attr, pid, cpu, group_fd, flags);
+}
+
+uint64_t i915_type_id(void);
+
+#endif /* I915_PERF_H */
diff --git a/overlay/power.c b/overlay/power.c
index 520a1099..6c5c3749 100644
--- a/overlay/power.c
+++ b/overlay/power.c
@@ -22,6 +22,7 @@
  *
  */
 
+#include <stdint.h>
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
@@ -30,11 +31,27 @@
 #include <time.h>
 #include <errno.h>
 
+#include "perf.h"
 #include "power.h"
 #include "debugfs.h"
 
 /* XXX Is this exposed through RAPL? */
 
+static int perf_open(void)
+{
+	struct perf_event_attr attr;
+
+	memset(&attr, 0, sizeof (attr));
+
+	attr.type = i915_type_id();
+	if (attr.type == 0)
+		return -ENOENT;
+	attr.config = I915_PERF_ENERGY;
+
+	attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED;
+	return perf_event_open(&attr, -1, 0, -1, 0);
+}
+
 int power_init(struct power *power)
 {
 	char buf[4096];
@@ -42,6 +59,10 @@ int power_init(struct power *power)
 
 	memset(power, 0, sizeof(*power));
 
+	power->fd = perf_open();
+	if (power->fd != -1)
+		return 0;
+
 	sprintf(buf, "%s/i915_energy_uJ", debugfs_dri_path);
 	fd = open(buf, 0);
 	if (fd < 0)
@@ -100,17 +121,25 @@ int power_update(struct power *power)
 	if (power->error)
 		return power->error;
 
-	s->energy = file_to_u64("i915_energy_uJ");
-	s->timestamp = clock_ms_to_u64();
+	if (power->fd != -1) {
+		uint64_t data[2];
+		int len;
+
+		len = read(power->fd, data, sizeof(data));
+		if (len < 0)
+			return power->error = errno;
+
+		s->energy = data[0];
+		s->timestamp = data[1] / (1000*1000);
+	} else {
+		s->energy = file_to_u64("i915_energy_uJ");
+		s->timestamp = clock_ms_to_u64();
+	}
+
 	if (power->count == 1)
 		return EAGAIN;
 
 	d_time = s->timestamp - d->timestamp;
-	if (d_time < 900) { /* HW sample rate seems to be stable ~1Hz */
-		power->count--;
-		return power->count <= 1 ? EAGAIN : 0;
-	}
-
 	power->power_mW = (s->energy - d->energy) / d_time;
 	power->new_sample = 1;
 	return 0;
diff --git a/overlay/power.h b/overlay/power.h
index a046bd5a..bf8346ce 100644
--- a/overlay/power.h
+++ b/overlay/power.h
@@ -33,6 +33,7 @@ struct power {
 		uint64_t timestamp;
 	} stat[2];
 
+	int fd;
 	int error;
 	int count;
 	int new_sample;