[RFC i-g-t 0/3] New intel-gpu-top

[RFC i-g-t 0/3] New intel-gpu-top

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

I've found a new potential lead vehicle to sell my per-client engine busyness
patches and that is a rewrite of intel-gpu-top.

Old one is neglected and dangerous - so how about a new one which looks like
this?

intel-gpu-top - load avg  9.47,  3.34,  0.13; 1000/ 950 MHz;    0% RC6;   5202mW;     1206 irqs/s

    rcs0 100.00% ( 0.56/ 4.78/ 3.86) |██████████████████████████████████████████████████████████████████|  0.00% wait,   0.00% sema
    bcs0  17.72% ( 0.81/ 0.00/ 0.17) |███████████▋                                                      |  0.00% wait,   0.00% sema
    vcs0  61.08% ( 0.81/ 0.00/ 0.63) |████████████████████████████████████████▎                         |  0.00% wait,   0.00% sema
   vecs0   0.00% ( 0.00/ 0.00/ 0.00) |                                                                  |  0.00% wait,   0.00% sema

  PID            NAME           rcs0                       bcs0                       vcs0                       vecs0
 9654       neverball |████████████             ||                         ||                         ||                         |
 7792            Xorg |████████▊                ||                         ||                         ||                         |
 9659        gem_wsim |████                     ||████▍                    ||███████████████▎         ||                         |
 7846           xfwm4 |                         ||                         ||                         ||                         |

So this series renames the old one to intel_legacy_top and replaces it with a
new one which uses PMU and some extra patches as well.

Code is not production quality but an attempt to spark some interest on the
topic.

Tvrtko Ursulin (3):
  intel-gpu-top: Rename to intel-legacy-top
  intel-gpu-top: New version using PMU
  intel-gpu-top: Support for client stats

 tools/.gitignore         |    1 +
 tools/Makefile.am        |    2 +
 tools/Makefile.sources   |    1 +
 tools/intel_gpu_top.c    | 1311 +++++++++++++++++++++++++---------------------
 tools/intel_legacy_top.c |  718 +++++++++++++++++++++++++
 tools/meson.build        |    3 +-
 6 files changed, 1446 insertions(+), 590 deletions(-)
 create mode 100644 tools/intel_legacy_top.c

-- 
2.14.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[RFC i-g-t 1/3] intel-gpu-top: Rename to intel-legacy-top

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/.gitignore                              | 2 +-
 tools/Makefile.sources                        | 2 +-
 tools/{intel_gpu_top.c => intel_legacy_top.c} | 0
 tools/meson.build                             | 2 +-
 4 files changed, 3 insertions(+), 3 deletions(-)
 rename tools/{intel_gpu_top.c => intel_legacy_top.c} (100%)

diff --git a/tools/.gitignore b/tools/.gitignore
index 2de665a36688..19a1f7cb8e50 100644
--- a/tools/.gitignore
+++ b/tools/.gitignore
@@ -17,7 +17,7 @@ intel_framebuffer_dump
 intel_gem_info
 intel_gpu_frequency
 intel_gpu_time
-intel_gpu_top
+intel_legacy_top
 intel_gtt
 intel_guc_logger
 intel_gvtg_test
diff --git a/tools/Makefile.sources b/tools/Makefile.sources
index abd23a0f4628..9699b7d2f737 100644
--- a/tools/Makefile.sources
+++ b/tools/Makefile.sources
@@ -17,7 +17,7 @@ tools_prog_lists =		\
 	intel_gpu_frequency	\
 	intel_firmware_decode	\
 	intel_gpu_time		\
-	intel_gpu_top		\
+	intel_legacy_top	\
 	intel_gtt		\
 	intel_guc_logger        \
 	intel_infoframes	\
diff --git a/tools/intel_gpu_top.c b/tools/intel_legacy_top.c
similarity index 100%
rename from tools/intel_gpu_top.c
rename to tools/intel_legacy_top.c
diff --git a/tools/meson.build b/tools/meson.build
index a96735c2244d..ebce4e305d00 100644
--- a/tools/meson.build
+++ b/tools/meson.build
@@ -23,7 +23,7 @@ tools_progs = [
 	'intel_gpu_frequency',
 	'intel_firmware_decode',
 	'intel_gpu_time',
-	'intel_gpu_top',
+	'intel_legacy_top',
 	'intel_gtt',
 	'intel_guc_logger',
 	'intel_infoframes',
-- 
2.14.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[RFC i-g-t 2/3] intel-gpu-top: New version using PMU

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/.gitignore       |   1 +
 tools/Makefile.am      |   2 +
 tools/Makefile.sources |   1 +
 tools/intel_gpu_top.c  | 593 +++++++++++++++++++++++++++++++++++++++++++++++++
 tools/meson.build      |   1 +
 5 files changed, 598 insertions(+)
 create mode 100644 tools/intel_gpu_top.c

diff --git a/tools/.gitignore b/tools/.gitignore
index 19a1f7cb8e50..6e3042810176 100644
--- a/tools/.gitignore
+++ b/tools/.gitignore
@@ -17,6 +17,7 @@ intel_framebuffer_dump
 intel_gem_info
 intel_gpu_frequency
 intel_gpu_time
+intel_gpu_top
 intel_legacy_top
 intel_gtt
 intel_guc_logger
diff --git a/tools/Makefile.am b/tools/Makefile.am
index dcf282eaff4e..8f6c15791a3b 100644
--- a/tools/Makefile.am
+++ b/tools/Makefile.am
@@ -30,6 +30,8 @@ intel_aubdump_la_LDFLAGS = -module -avoid-version -no-undefined
 intel_aubdump_la_SOURCES = aubdump.c
 intel_aubdump_la_LIBADD = $(top_builddir)/lib/libintel_tools.la -ldl
 
+intel_gpu_top_LDADD = $(top_builddir)/lib/libigt_perf.la -lm
+
 bin_SCRIPTS = intel_aubdump
 CLEANFILES = $(bin_SCRIPTS)
 
diff --git a/tools/Makefile.sources b/tools/Makefile.sources
index 9699b7d2f737..6f8668bd4d56 100644
--- a/tools/Makefile.sources
+++ b/tools/Makefile.sources
@@ -17,6 +17,7 @@ tools_prog_lists =		\
 	intel_gpu_frequency	\
 	intel_firmware_decode	\
 	intel_gpu_time		\
+	intel_gpu_top		\
 	intel_legacy_top	\
 	intel_gtt		\
 	intel_guc_logger        \
diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
new file mode 100644
index 000000000000..59a112240092
--- /dev/null
+++ b/tools/intel_gpu_top.c
@@ -0,0 +1,593 @@
+#include <stdio.h>
+#include <sys/types.h>
+#include <dirent.h>
+#include <stdint.h>
+#include <assert.h>
+#include <string.h>
+#include <ctype.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <sys/ioctl.h>
+#include <errno.h>
+#include <math.h>
+
+#include "igt_perf.h"
+
+struct pmu_pair {
+	uint64_t cur;
+	uint64_t prev;
+};
+
+struct pmu_counter {
+	uint64_t config;
+	unsigned int idx;
+	struct pmu_pair val;
+};
+
+#define NUM_LOADS (3)
+
+struct engine {
+	const char *name;
+	double qd[3];
+	double load_avg[NUM_LOADS];
+	struct pmu_counter busy;
+	struct pmu_counter wait;
+	struct pmu_counter sema;
+	struct pmu_counter queued;
+	struct pmu_counter runnable;
+	struct pmu_counter running;
+};
+
+struct engines {
+	unsigned int num_engines;
+	unsigned int num_counters;
+	DIR *root;
+	int fd;
+	struct pmu_pair ts;
+
+	int rapl_fd;
+	double rapl_scale;
+
+	struct pmu_counter freq_req;
+	struct pmu_counter freq_act;
+	struct pmu_counter irq;
+	struct pmu_counter rc6;
+	struct pmu_counter rapl;
+
+	double qd_scale;
+
+	double load_exp[NUM_LOADS];
+	double load_avg[NUM_LOADS];
+
+	struct engine engine;
+};
+
+static uint64_t
+get_pmu_config(int dirfd, const char *name, const char *counter)
+{
+	char buf[128], *p;
+	int fd, ret;
+
+	ret = snprintf(buf, sizeof(buf), "%s-%s", name, counter);
+	if (ret < 0 || ret == sizeof(buf))
+		return -1;
+
+	fd = openat(dirfd, buf, O_RDONLY);
+	if (fd < 0)
+		return -1;
+
+	ret = read(fd, buf, sizeof(buf));
+	close(fd);
+	if (ret <= 0)
+		return -1;
+
+	p = index(buf, '0');
+	if (!p)
+		return -1;
+
+	return strtoul(p, NULL, 0);
+}
+
+#define engine_ptr(engines, n) \
+	((struct engine *)((unsigned char *)(&engines->engine) + (n) * sizeof(struct engine)))
+
+static struct engines *discover_engines(void)
+{
+	const char *sysfs_root = "/sys/devices/i915/events";
+	struct engines *engines;
+	struct dirent *dent;
+	int ret = 0;
+	DIR *d;
+
+	engines = malloc(sizeof(struct engines));
+	if (!engines)
+		return NULL;
+	memset(engines, 0, sizeof(*engines));
+
+	engines->num_engines = 0;
+
+	d = opendir(sysfs_root);
+	if (!d)
+		return NULL;
+
+	while ((dent = readdir(d)) != NULL) {
+		const char *endswith = "-busy";
+		const unsigned int endlen = strlen(endswith);
+		struct engine *engine =
+				engine_ptr(engines, engines->num_engines);
+		char buf[256];
+
+		if (dent->d_type != DT_REG)
+			continue;
+
+		if (strlen(dent->d_name) >= sizeof(buf)) {
+			ret = -1;
+			break;
+		}
+
+		strcpy(buf, dent->d_name);
+
+		/* xxxN-busy */
+		if (strlen(buf) < (endlen + 4))
+			continue;
+		if (strcmp(&buf[strlen(buf) - endlen], endswith))
+			continue;
+
+		memset(engine, 0, sizeof(*engine));
+
+		buf[strlen(buf) - endlen] = 0;
+		engine->name = strdup(buf);
+		if (!engine->name) {
+			ret = -1;
+			break;
+		}
+
+		engine->busy.config = get_pmu_config(dirfd(d),
+						     engine->name,
+						     "busy");
+		if (engine->busy.config == -1) {
+			ret = -1;
+			break;
+		}
+
+		engines->num_engines++;
+		engines = realloc(engines, sizeof(struct engines) +
+				  engines->num_engines * sizeof(struct engine));
+		if (!engines) {
+			ret = -ENOMEM;
+			break;
+		}
+	}
+
+	if (ret)
+		free(engines);
+	else
+		engines->root = d;
+
+	return ret == 0 ? engines : NULL;
+}
+
+static int
+filename_to_buf(const char *filename, char *buf, unsigned int bufsize)
+{
+	int fd;
+	ssize_t ret;
+
+	fd = open(filename, O_RDONLY);
+	if (fd < 0)
+		return -1;
+
+	ret = read(fd, buf, bufsize - 1);
+	close(fd);
+	if (ret < 1)
+		return -1;
+
+	buf[ret] = '\0';
+
+	return 0;
+}
+
+static uint64_t filename_to_u64(const char *filename, int base)
+{
+	char buf[64], *b;
+
+	if (filename_to_buf(filename, buf, sizeof(buf)))
+		return 0;
+
+	/*
+	 * Handle both single integer and key=value formats by skipping
+	 * leading non-digits.
+	 */
+	b = buf;
+	while (*b && !isdigit(*b))
+		b++;
+
+	return strtoull(b, NULL, base);
+}
+
+static uint64_t rapl_type_id(void)
+{
+	return filename_to_u64("/sys/devices/power/type", 10);
+}
+
+static uint64_t rapl_gpu_power(void)
+{
+	return filename_to_u64("/sys/devices/power/events/energy-gpu", 0);
+}
+
+static double filename_to_double(const char *filename)
+{
+	char buf[64];
+
+	if (filename_to_buf(filename, buf, sizeof(buf)))
+		return 0;
+
+	return strtod(buf, NULL);
+}
+
+static double rapl_gpu_power_scale(void)
+{
+	return filename_to_double("/sys/devices/power/events/energy-gpu.scale");
+}
+
+static double i915_qd_scale(void)
+{
+	return filename_to_double("/sys/devices/i915/events/rcs0-queued.scale");
+}
+
+#define __open_pmu(engines, pmu, idx) \
+({ \
+	int fd__; \
+\
+	fd__ = perf_i915_open_group((pmu)->config, (engines)->fd); \
+	if (fd__ >= 0) { \
+		if ((engines)->fd == -1) \
+			(engines)->fd = fd__; \
+		(pmu)->idx = (idx)++; \
+		(engines)->num_counters++; \
+	} \
+\
+	fd__; \
+})
+
+static int pmu_init(struct engines *engines)
+{
+	unsigned int idx = 0;
+	unsigned int i;
+	int fd;
+
+	engines->fd = -1;
+	engines->num_counters = 0;
+
+	engines->freq_req.config = I915_PMU_REQUESTED_FREQUENCY;
+	fd = __open_pmu(engines, &engines->freq_req, idx);
+	if (fd < 0)
+		return -1;
+
+	engines->freq_act.config = I915_PMU_ACTUAL_FREQUENCY;
+	fd = __open_pmu(engines, &engines->freq_act, idx);
+	if (fd < 0)
+		return -1;
+
+	engines->irq.config = I915_PMU_INTERRUPTS;
+	fd = __open_pmu(engines, &engines->irq, idx);
+	if (fd < 0)
+		return -1;
+
+	engines->rc6.config = I915_PMU_RC6_RESIDENCY;
+	fd = __open_pmu(engines, &engines->rc6, idx);
+	if (fd < 0)
+		return -1;
+
+	engines->qd_scale = i915_qd_scale();
+
+	for (i = 0; i < engines->num_engines; i++) {
+		struct engine *engine = engine_ptr(engines, i);
+		struct {
+			struct pmu_counter *pmu;
+			const char *counter;
+		} *cnt, counters[] = {
+			{ .pmu = &engine->busy, .counter = "busy" },
+			{ .pmu = &engine->wait, .counter = "wait" },
+			{ .pmu = &engine->sema, .counter = "sema" },
+			{ .pmu = &engine->queued, .counter = "queued" },
+			{ .pmu = &engine->runnable, .counter = "runnable" },
+			{ .pmu = &engine->running, .counter = "running" },
+			{ .pmu = NULL, .counter = NULL },
+		};
+
+		for (cnt = counters; cnt->pmu; cnt++) {
+			if (!cnt->pmu->config)
+				cnt->pmu->config =
+					get_pmu_config(dirfd(engines->root),
+						       engine->name,
+						       cnt->counter);
+			fd = __open_pmu(engines, cnt->pmu, idx);
+			if (fd < 0)
+				return -1;
+		}
+	}
+
+	engines->rapl_scale = rapl_gpu_power_scale();
+	if (engines->rapl_scale != NAN)
+		engines->rapl_scale *= 1e3; /* from nano to micro */
+	engines->rapl.config = rapl_gpu_power();
+	engines->rapl_fd = igt_perf_open(rapl_type_id(), engines->rapl.config);
+	if (engines->rapl_fd < 0)
+		return -1;
+
+	return 0;
+}
+
+static uint64_t pmu_read_multi(int fd, unsigned int num, uint64_t *val)
+{
+	uint64_t buf[2 + num];
+	unsigned int i;
+
+	assert(read(fd, buf, sizeof(buf)) == sizeof(buf));
+
+	for (i = 0; i < num; i++)
+		val[i] = buf[2 + i];
+
+	return buf[1];
+}
+
+
+static double pmu_calc(struct pmu_pair *p, double d, double t, double s)
+{
+	double pct;
+
+	pct = p->cur - p->prev;
+	pct /= d;
+	pct /= t;
+	pct *= s;
+
+	if (s == 100.0 && pct > 100.0)
+		pct = 100.0;
+
+	return pct;
+}
+
+static uint64_t __pmu_read_single(int fd, uint64_t *ts)
+{
+	uint64_t data[2];
+
+	assert(read(fd, data, sizeof(data)) == sizeof(data));
+
+	if (ts)
+		*ts = data[1];
+
+	return data[0];
+}
+
+static uint64_t pmu_read_single(int fd)
+{
+	return __pmu_read_single(fd, NULL);
+}
+
+static void __update_sample(struct pmu_counter *counter, uint64_t val)
+{
+	counter->val.prev = counter->val.cur;
+	counter->val.cur = val;
+}
+
+static void update_sample(struct pmu_counter *counter, uint64_t *val)
+{
+	__update_sample(counter, val[counter->idx]);
+}
+
+static void pmu_sample(struct engines *engines)
+{
+	const int num_val = engines->num_counters;
+	uint64_t val[num_val];
+	unsigned int i;
+
+	engines->ts.prev = engines->ts.cur;
+	engines->ts.cur = pmu_read_multi(engines->fd, num_val, val);
+
+	__update_sample(&engines->rapl, pmu_read_single(engines->rapl_fd));
+
+	update_sample(&engines->freq_req, val);
+	update_sample(&engines->freq_act, val);
+	update_sample(&engines->irq, val);
+	update_sample(&engines->rc6, val);
+
+	for (i = 0; i < engines->num_engines; i++) {
+		struct engine *engine = engine_ptr(engines, i);
+
+		update_sample(&engine->busy, val);
+		update_sample(&engine->sema, val);
+		update_sample(&engine->wait, val);
+		update_sample(&engine->queued, val);
+		update_sample(&engine->runnable, val);
+		update_sample(&engine->running, val);
+	}
+}
+
+static const char *bars[] = { " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
+
+static void
+print_percentage_bar(double percent, int max_len)
+{
+	int bar_len = percent * (8 * (max_len - 2)) / 100.0;
+	int i;
+
+	putchar('|');
+
+	for (i = bar_len; i >= 8; i -= 8)
+		printf("%s", bars[8]);
+	if (i)
+		printf("%s", bars[i]);
+
+	for (i = 0; i < (max_len - 2 - (bar_len + 7) / 8); i++)
+		putchar(' ');
+
+	putchar('|');
+}
+
+#define DEFAULT_PERIOD_MS (1000)
+
+static void
+usage(const char *appname)
+{
+	printf("intel_gpu_top - Display a top-like summary of Intel GPU usage\n"
+		"\n"
+		"Usage: %s [parameters]\n"
+		"\n"
+		"\tThe following parameters are optional:\n"
+		"\t[-s <samples>]       refresh period in ms (default %ums)\n"
+		"\t[-h]                 show this help text\n"
+		"\n",
+		appname, DEFAULT_PERIOD_MS);
+}
+
+static double update_load(double load, double exp, double val)
+{
+	return val + exp * (load - val);
+}
+
+int main(int argc, char **argv)
+{
+	unsigned int period_us = DEFAULT_PERIOD_MS * 1000;
+	const double load_period[NUM_LOADS] = { 1.0, 30.0, 900.0 };
+	struct engines *engines;
+	int con_w = -1, con_h = -1;
+	struct winsize ws;
+	unsigned int i;
+	double period;
+	int ret, ch;
+
+	/* Parse options */
+	while ((ch = getopt(argc, argv, "s:h")) != -1) {
+		switch (ch) {
+		case 's':
+			period_us = atoi(optarg) * 1000;
+			break;
+		case 'h':
+			usage(argv[0]);
+			exit(0);
+		default:
+			fprintf(stderr, "Invalid option %c!\n", (char)optopt);
+			usage(argv[0]);
+			exit(1);
+		}
+	}
+
+	/* Get terminal size. */
+	if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
+		con_w = ws.ws_col;
+		con_h = ws.ws_row;
+	}
+
+	engines = discover_engines();
+	if (!engines) {
+		fprintf(stderr, "Failed to detect engines!\n");
+		return 1;
+	}
+
+	ret = pmu_init(engines);
+	if (ret) {
+		fprintf(stderr, "Failed to initialize PMU!\n");
+		return 1;
+	}
+
+	/* Load average setup. */
+	period = (double)period_us / 1e6;
+	for (i = 0; i < NUM_LOADS; i++)
+		engines->load_exp[i] = exp(-period / load_period[i]);
+
+	pmu_sample(engines);
+
+	for (;;) {
+		double t, freq[2], irq, rc6, power;
+		double qd = 0;
+		int lines = 0;
+		unsigned int j;
+
+		usleep(period_us);
+
+		pmu_sample(engines);
+		t = (double)(engines->ts.cur - engines->ts.prev) / 1e9;
+
+		printf("\033[H\033[J");
+
+		freq[0] = pmu_calc(&engines->freq_req.val, 1.0, t, 1);
+		freq[1] = pmu_calc(&engines->freq_act.val, 1.0, t, 1);
+		irq = pmu_calc(&engines->irq.val, 1.0, t, 1);
+		rc6 = pmu_calc(&engines->rc6.val, 1e9, t, 100);
+		power = pmu_calc(&engines->rapl.val, 1.0, t, engines->rapl_scale);
+
+		for (i = 0; i < engines->num_engines; i++) {
+			struct engine *engine = engine_ptr(engines, i);
+
+			engine->qd[0] = pmu_calc(&engine->queued.val, 1, t,
+						 engines->qd_scale);
+			engine->qd[1] = pmu_calc(&engine->runnable.val, 1, t,
+						 engines->qd_scale);
+			engine->qd[2] = pmu_calc(&engine->running.val, 1, t,
+						 engines->qd_scale);
+
+			qd += engine->qd[1] + engine->qd[2];
+
+			for (j = 0; j < NUM_LOADS; j++) {
+				engine->load_avg[j] =
+					update_load(engine->load_avg[j],
+						    engines->load_exp[j],
+						    engine->qd[1] +
+						    engine->qd[2]);
+			}
+		}
+
+		for (j = 0; j < NUM_LOADS; j++) {
+			engines->load_avg[j] =
+				update_load(engines->load_avg[j],
+					    engines->load_exp[j],
+					    qd);
+		}
+
+		printf("intel-gpu-top - load avg %5.2f, %5.2f, %5.2f; %4.0f/%4.0f MHz;  %3.0f%% RC6; %6.0fmW; %8.0f irqs/s\n",
+		       engines->load_avg[0],
+		       engines->load_avg[1],
+		       engines->load_avg[2],
+		       freq[0], freq[1],
+		       rc6, power, irq);
+		lines++;
+
+		printf("\n");
+		lines++;
+
+		for (i = 0; i < engines->num_engines && lines < con_h; i++) {
+			struct engine *engine = engine_ptr(engines, i);
+			unsigned int max_w = con_w - 1;
+			unsigned int len;
+			double val[2];
+			char buf[128];
+
+			val[0] = pmu_calc(&engine->wait.val, 1e9, t, 100);
+			val[1] = pmu_calc(&engine->sema.val, 1e9, t, 100);
+			len = snprintf(buf, sizeof(buf),
+				       "%6.2f%% wait, %6.2f%% sema",
+				       val[0], val[1]);
+
+			val[0] = pmu_calc(&engine->busy.val, 1e9, t, 100);
+			len += printf("%8s %6.2f%% (%5.2f/%5.2f/%5.2f) ",
+				      engine->name,
+				      val[0],
+				      engine->qd[0],
+				      engine->qd[1],
+				      engine->qd[2]);
+			print_percentage_bar(val[0], max_w - len);
+
+			printf("%s\n", buf);
+
+			lines++;
+		}
+
+		printf("\n");
+	}
+
+	return 0;
+}
diff --git a/tools/meson.build b/tools/meson.build
index ebce4e305d00..36038f7a9d22 100644
--- a/tools/meson.build
+++ b/tools/meson.build
@@ -23,6 +23,7 @@ tools_progs = [
 	'intel_gpu_frequency',
 	'intel_firmware_decode',
 	'intel_gpu_time',
+	'intel_gpu-top',
 	'intel_legacy_top',
 	'intel_gtt',
 	'intel_guc_logger',
-- 
2.14.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[RFC i-g-t 3/3] intel-gpu-top: Support for client stats

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 tools/intel_gpu_top.c | 268 +++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 263 insertions(+), 5 deletions(-)

diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
index 59a112240092..63f0c1941258 100644
--- a/tools/intel_gpu_top.c
+++ b/tools/intel_gpu_top.c
@@ -407,8 +407,222 @@ static void pmu_sample(struct engines *engines)
 	}
 }
 
+enum client_status {
+	FREE = 0, /* mbz */
+	ALIVE,
+	PROBE
+};
+
+struct client {
+	enum client_status status;
+	unsigned int id;
+	unsigned int pid;
+	char name[128];
+	unsigned int samples;
+	unsigned long total;
+	struct engines *engines;
+	unsigned long *val;
+	uint64_t *last;
+};
+
+#define SYSFS_ENABLE "/sys/class/drm/card0/clients/enable_stats"
+#define SYSFS_CLIENTS "/sys/class/drm/card0/clients/"
+
+#define PERIOD_US (1000e3)
+
+static struct client *clients;
+static unsigned int num_clients;
+
+#define for_each_client(c, tmp) \
+	for (tmp = num_clients, c = clients; tmp > 0; tmp--, c++)
+
+static uint64_t read_client_busy(unsigned int id, const char *engine)
+{
+	char buf[256];
+	ssize_t ret;
+
+	ret = snprintf(buf, sizeof(buf), SYSFS_CLIENTS "/%u/busy/%s",
+		       id, engine);
+	assert(ret > 0);
+
+	return filename_to_u64(buf, 10);
+}
+
+static struct client *find_client(enum client_status status, unsigned int id)
+{
+	struct client *c;
+	unsigned int tmp;
+
+	for_each_client(c, tmp) {
+		if ((status == FREE && c->status == FREE) ||
+		    (status == c->status && c->id == id)) {
+			return c;
+		}
+	}
+
+	return NULL;
+}
+
+static void update_client(struct client *c, unsigned int pid, char *name)
+{
+	uint64_t val[c->engines->num_engines];
+	unsigned int i;
+
+	if (c->pid != pid)
+		c->pid = pid;
+
+	if (strncmp(c->name, name, sizeof(c->name)))
+		strncpy(c->name, name, sizeof(c->name));
+
+	for (i = 0; i < c->engines->num_engines; i++) {
+		struct engine *engine = engine_ptr(c->engines, i);
+
+		val[i] = read_client_busy(c->id, engine->name);
+	}
+
+	c->total = 0;
+
+	for (i = 0; i < c->engines->num_engines; i++) {
+		assert(val[i] >= c->last[i]);
+		c->val[i] = val[i] - c->last[i];
+		c->total += c->val[i];
+		c->last[i] = val[i];
+	}
+
+	c->samples++;
+	c->status = ALIVE;
+}
+
+static void
+add_client(unsigned int id, unsigned int pid, char *name,
+	   struct engines *engines)
+{
+	struct client *c;
+
+	assert(!find_client(ALIVE, id));
+
+	c = find_client(FREE, 0);
+	if (!c) {
+		unsigned int idx = num_clients;
+
+		num_clients += (num_clients + 2) / 2;
+		clients = realloc(clients, num_clients * sizeof(*c));
+		assert(clients);
+		c = &clients[idx];
+		memset(c, 0, (num_clients - idx) * sizeof(*c));
+	}
+
+	c->id = id;
+	c->engines = engines;
+	c->val = calloc(engines->num_engines, sizeof(c->val));
+	c->last = calloc(engines->num_engines, sizeof(c->last));
+	assert(c->val && c->last);
+
+	update_client(c, pid, name);
+
+// 	c = find_client(c->status, c->id);
+}
+
+static void free_client(struct client *c)
+{
+	free(c->val);
+	free(c->last);
+	memset(c, 0, sizeof(*c));
+}
+
+static char *read_client_sysfs(unsigned int id, const char *field)
+{
+	char buf[256];
+	ssize_t ret;
+
+	ret = snprintf(buf, sizeof(buf), SYSFS_CLIENTS "/%u/%s", id, field);
+	assert(ret > 0);
+
+	ret = filename_to_buf(buf, buf, sizeof(buf));
+	assert(ret == 0);
+
+	return strdup(buf);
+}
+
+static void scan(struct engines *engines)
+{
+	struct dirent *dent;
+	struct client *c;
+	char *pid, *name;
+	unsigned int tmp;
+	unsigned int id;
+	DIR *d;
+
+	for_each_client(c, tmp) {
+		if (c->status == ALIVE)
+			c->status = PROBE;
+	}
+
+	d = opendir(SYSFS_CLIENTS);
+	assert(!NULL);
+
+	while ((dent = readdir(d)) != NULL) {
+		if (dent->d_type != DT_DIR)
+			continue;
+		if (!isdigit(dent->d_name[0]))
+			continue;
+
+		id = atoi(dent->d_name);
+
+		name = read_client_sysfs(id, "name");
+		assert(name);
+
+		pid = read_client_sysfs(id, "pid");
+		assert(pid);
+
+		c = find_client(PROBE, id);
+		if (c) {
+			update_client(c, atoi(pid), name);
+			continue;
+		}
+
+		add_client(id, atoi(pid), name, engines);
+
+		free(name);
+		free(pid);
+	}
+
+	closedir(d);
+
+	for_each_client(c, tmp) {
+		if (c->status == PROBE)
+			free_client(c);
+	}
+}
+
+static int cmp(const void *_a, const void *_b)
+{
+	const struct client *a = _a;
+	const struct client *b = _b;
+	long tot_a = a->total;
+	long tot_b = b->total;
+
+	if (a->status != ALIVE || a->samples < 2)
+		tot_a = -1;
+	if (b->status != ALIVE || b->samples < 2)
+		tot_b = -1;
+
+	if (tot_a == tot_b)
+		return a->pid - b->pid;
+	else
+		return -(tot_a - tot_b);
+}
+
 static const char *bars[] = { " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" };
 
+static void n_spaces(const unsigned int n)
+{
+	unsigned int i;
+
+	for (i = 0; i < n; i++)
+		putchar(' ');
+}
+
 static void
 print_percentage_bar(double percent, int max_len)
 {
@@ -422,8 +636,7 @@ print_percentage_bar(double percent, int max_len)
 	if (i)
 		printf("%s", bars[i]);
 
-	for (i = 0; i < (max_len - 2 - (bar_len + 7) / 8); i++)
-		putchar(' ');
+	n_spaces(max_len - 2 - (bar_len + 7) / 8);
 
 	putchar('|');
 }
@@ -500,16 +713,21 @@ int main(int argc, char **argv)
 		engines->load_exp[i] = exp(-period / load_period[i]);
 
 	pmu_sample(engines);
+	scan(engines);
 
 	for (;;) {
 		double t, freq[2], irq, rc6, power;
 		double qd = 0;
 		int lines = 0;
+		struct client *c;
+		unsigned int len, engine_w;
 		unsigned int j;
 
 		usleep(period_us);
 
 		pmu_sample(engines);
+		scan(engines);
+
 		t = (double)(engines->ts.cur - engines->ts.prev) / 1e9;
 
 		printf("\033[H\033[J");
@@ -561,8 +779,6 @@ int main(int argc, char **argv)
 
 		for (i = 0; i < engines->num_engines && lines < con_h; i++) {
 			struct engine *engine = engine_ptr(engines, i);
-			unsigned int max_w = con_w - 1;
-			unsigned int len;
 			double val[2];
 			char buf[128];
 
@@ -579,7 +795,7 @@ int main(int argc, char **argv)
 				      engine->qd[0],
 				      engine->qd[1],
 				      engine->qd[2]);
-			print_percentage_bar(val[0], max_w - len);
+			print_percentage_bar(val[0], con_w - 1 - len);
 
 			printf("%s\n", buf);
 
@@ -587,6 +803,48 @@ int main(int argc, char **argv)
 		}
 
 		printf("\n");
+
+		printf("\033[7m");
+		len = printf("%5s%16s", "PID", "NAME");
+		engine_w = (con_w - len) / engines->num_engines;
+		for (i = 0; i < engines->num_engines && lines < con_h; i++) {
+			struct engine *engine = engine_ptr(engines, i);
+			unsigned int pad = (engine_w - strlen(engine->name)) / 2;
+
+			n_spaces(pad);
+			printf("%s", engine->name);
+			n_spaces(engine_w - pad - strlen(engine->name));
+			len += pad + strlen(engine->name) +
+			       (engine_w - pad - strlen(engine->name));
+		}
+		n_spaces(con_w - len);
+		printf("\033[0m\n");
+		lines++;
+
+		qsort(clients, num_clients, sizeof(*c), cmp);
+
+		for_each_client(c, i) {
+			if (lines > con_h)
+				break;
+
+			assert(c->status != PROBE);
+			if (c->status != ALIVE || c->samples < 2)
+				break;
+
+			printf("%5u%16s ", c->pid, c->name);
+
+			for (i = 0; i < engines->num_engines; i++) {
+				double pct;
+
+				pct = (double)c->val[i] / period_us / 1e3 * 100;
+
+				print_percentage_bar(pct, engine_w);
+			}
+
+			putchar('\n');
+			lines++;
+		}
+
 	}
 
 	return 0;
-- 
2.14.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✓ Fi.CI.BAT: success for New intel-gpu-top

[Patchwork]

== Series Details ==

Series: New intel-gpu-top
URL   : https://patchwork.freedesktop.org/series/38282/
State : success

== Summary ==

IGT patchset tested on top of latest successful build
1d227a47223d2ff5c6745c0d82e96e70d456d5de lib/kms: Clear unused fields for getproperty ioctl

with latest DRM-Tip kernel build CI_DRM_3774
5ad7866768dc drm-tip: 2018y-02m-14d-15h-14m-50s UTC integration manifest

No testlist changes.

fi-bdw-5557u     total:288  pass:267  dwarn:0   dfail:0   fail:0   skip:21  time:421s
fi-bdw-gvtdvm    total:288  pass:264  dwarn:0   dfail:0   fail:0   skip:24  time:426s
fi-blb-e6850     total:288  pass:223  dwarn:1   dfail:0   fail:0   skip:64  time:377s
fi-bsw-n3050     total:288  pass:242  dwarn:0   dfail:0   fail:0   skip:46  time:489s
fi-bwr-2160      total:288  pass:183  dwarn:0   dfail:0   fail:0   skip:105 time:290s
fi-bxt-dsi       total:288  pass:258  dwarn:0   dfail:0   fail:0   skip:30  time:483s
fi-bxt-j4205     total:288  pass:259  dwarn:0   dfail:0   fail:0   skip:29  time:489s
fi-byt-j1900     total:288  pass:253  dwarn:0   dfail:0   fail:0   skip:35  time:473s
fi-byt-n2820     total:288  pass:249  dwarn:0   dfail:0   fail:0   skip:39  time:459s
fi-cfl-s2        total:288  pass:262  dwarn:0   dfail:0   fail:0   skip:26  time:567s
fi-cnl-y3        total:288  pass:262  dwarn:0   dfail:0   fail:0   skip:26  time:578s
fi-elk-e7500     total:288  pass:229  dwarn:0   dfail:0   fail:0   skip:59  time:426s
fi-gdg-551       total:288  pass:179  dwarn:0   dfail:0   fail:1   skip:108 time:284s
fi-glk-1         total:288  pass:260  dwarn:0   dfail:0   fail:0   skip:28  time:512s
fi-hsw-4770      total:288  pass:261  dwarn:0   dfail:0   fail:0   skip:27  time:391s
fi-ilk-650       total:288  pass:228  dwarn:0   dfail:0   fail:0   skip:60  time:413s
fi-ivb-3520m     total:288  pass:259  dwarn:0   dfail:0   fail:0   skip:29  time:463s
fi-ivb-3770      total:288  pass:255  dwarn:0   dfail:0   fail:0   skip:33  time:420s
fi-kbl-7500u     total:288  pass:263  dwarn:1   dfail:0   fail:0   skip:24  time:467s
fi-kbl-7560u     total:288  pass:269  dwarn:0   dfail:0   fail:0   skip:19  time:498s
fi-kbl-r         total:288  pass:261  dwarn:0   dfail:0   fail:0   skip:27  time:501s
fi-pnv-d510      total:288  pass:222  dwarn:1   dfail:0   fail:0   skip:65  time:591s
fi-skl-6260u     total:288  pass:268  dwarn:0   dfail:0   fail:0   skip:20  time:424s
fi-skl-6600u     total:288  pass:261  dwarn:0   dfail:0   fail:0   skip:27  time:510s
fi-skl-6700hq    total:288  pass:262  dwarn:0   dfail:0   fail:0   skip:26  time:525s
fi-skl-6700k2    total:288  pass:264  dwarn:0   dfail:0   fail:0   skip:24  time:497s
fi-skl-6770hq    total:288  pass:268  dwarn:0   dfail:0   fail:0   skip:20  time:475s
fi-skl-guc       total:288  pass:260  dwarn:0   dfail:0   fail:0   skip:28  time:414s
fi-skl-gvtdvm    total:288  pass:265  dwarn:0   dfail:0   fail:0   skip:23  time:433s
fi-snb-2520m     total:288  pass:248  dwarn:0   dfail:0   fail:0   skip:40  time:541s
fi-snb-2600      total:288  pass:248  dwarn:0   dfail:0   fail:0   skip:40  time:398s
Blacklisted hosts:
fi-glk-dsi       total:288  pass:258  dwarn:0   dfail:0   fail:0   skip:30  time:470s
fi-kbl-7567u     total:288  pass:268  dwarn:0   dfail:0   fail:0   skip:20  time:457s

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_915/issues.html
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [RFC i-g-t 1/3] intel-gpu-top: Rename to intel-legacy-top

[Petri Latvala]

On Wed, Feb 14, 2018 at 06:52:05PM +0000, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> ---
>  tools/.gitignore                              | 2 +-
>  tools/Makefile.sources                        | 2 +-
>  tools/{intel_gpu_top.c => intel_legacy_top.c} | 0
>  tools/meson.build                             | 2 +-
>  4 files changed, 3 insertions(+), 3 deletions(-)
>  rename tools/{intel_gpu_top.c => intel_legacy_top.c} (100%)


If the old tool is to be kept, the name intel_legacy_top doesn't quite
capture what it does. A more important topic though is whether it
should be kept at all. What are the features in the old tool that your
rewrite doesn't have?


-- 
Petri Latvala
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [RFC i-g-t 1/3] intel-gpu-top: Rename to intel-legacy-top

[Chris Wilson]

Quoting Petri Latvala (2018-02-15 09:10:45)
> On Wed, Feb 14, 2018 at 06:52:05PM +0000, Tvrtko Ursulin wrote:
> > From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> > 
> > Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> > ---
> >  tools/.gitignore                              | 2 +-
> >  tools/Makefile.sources                        | 2 +-
> >  tools/{intel_gpu_top.c => intel_legacy_top.c} | 0
> >  tools/meson.build                             | 2 +-
> >  4 files changed, 3 insertions(+), 3 deletions(-)
> >  rename tools/{intel_gpu_top.c => intel_legacy_top.c} (100%)
> 
> 
> If the old tool is to be kept, 

No. The old tool should be removed as it is a machine killer.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [RFC i-g-t 1/3] intel-gpu-top: Rename to intel-legacy-top

[Tvrtko Ursulin]

On 15/02/2018 09:10, Petri Latvala wrote:
> On Wed, Feb 14, 2018 at 06:52:05PM +0000, Tvrtko Ursulin wrote:
>> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
>>
>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
>> ---
>>   tools/.gitignore                              | 2 +-
>>   tools/Makefile.sources                        | 2 +-
>>   tools/{intel_gpu_top.c => intel_legacy_top.c} | 0
>>   tools/meson.build                             | 2 +-
>>   4 files changed, 3 insertions(+), 3 deletions(-)
>>   rename tools/{intel_gpu_top.c => intel_legacy_top.c} (100%)
> 
> 
> If the old tool is to be kept, the name intel_legacy_top doesn't quite
> capture what it does. A more important topic though is whether it
> should be kept at all. What are the features in the old tool that your
> rewrite doesn't have?

It's a bit different in target audience and capabilities I think.

The current one exposes what OA is, I assume at least, able to provide 
today in a safe way. The new tool doesn't do any of that but just 
provides basic, more end-user friendly, engine busyness and related stats.

It may be that when considering gpu-top work, intel-gpu-top rewrite 
hasn't even got a place. Or it might have as a minimal, easy to use and 
simple tool. TBD.

On 15/02/2018 09:17, Chris Wilson wrote:
 > Quoting Petri Latvala (2018-02-15 09:10:45)
 >>
 >> If the old tool is to be kept,
 >
 > No. The old tool should be removed as it is a machine killer.

And it is not up to date with current hardware so no complaints from me.

Regards,

Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [RFC i-g-t 1/3] intel-gpu-top: Rename to intel-legacy-top

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-02-15 09:57:45)
> 
> On 15/02/2018 09:10, Petri Latvala wrote:
> > On Wed, Feb 14, 2018 at 06:52:05PM +0000, Tvrtko Ursulin wrote:
> >> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> >>
> >> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> >> ---
> >>   tools/.gitignore                              | 2 +-
> >>   tools/Makefile.sources                        | 2 +-
> >>   tools/{intel_gpu_top.c => intel_legacy_top.c} | 0
> >>   tools/meson.build                             | 2 +-
> >>   4 files changed, 3 insertions(+), 3 deletions(-)
> >>   rename tools/{intel_gpu_top.c => intel_legacy_top.c} (100%)
> > 
> > 
> > If the old tool is to be kept, the name intel_legacy_top doesn't quite
> > capture what it does. A more important topic though is whether it
> > should be kept at all. What are the features in the old tool that your
> > rewrite doesn't have?
> 
> It's a bit different in target audience and capabilities I think.
> 
> The current one exposes what OA is, I assume at least, able to provide 
> today in a safe way. The new tool doesn't do any of that but just 
> provides basic, more end-user friendly, engine busyness and related stats.
> 
> It may be that when considering gpu-top work, intel-gpu-top rewrite 
> hasn't even got a place. Or it might have as a minimal, easy to use and 
> simple tool. TBD.

I liken it to classic "top", even when we have graphical system
monitors, for a quick assay of what's running, top is unbeatable.
I think each tool has its use. There may be future refactoring to a
single tool with different interfaces, but for now it looks to be a nice
little demonstration vehicle with some practical value.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx