[Intel-gfx] [RFC i-g-t 1/2] intel-gpu-overlay: Use new class/instance based PMU API

Tvrtko Ursulin tursulin at ursulin.net
Wed Aug 2 12:33:11 UTC 2017


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

So it works with the i915 PMU RFC kernel patches.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
---
 overlay/gem-interrupts.c |  2 +-
 overlay/gpu-freq.c       |  4 +--
 overlay/gpu-top.c        | 51 +++++++++++++++++++++---------------
 overlay/perf.h           | 68 +++++++++++++++++++++++++++++-------------------
 overlay/power.c          |  2 +-
 overlay/rc6.c            |  6 ++---
 6 files changed, 78 insertions(+), 55 deletions(-)

diff --git a/overlay/gem-interrupts.c b/overlay/gem-interrupts.c
index 0150a1d03825..f7c88d4822a5 100644
--- a/overlay/gem-interrupts.c
+++ b/overlay/gem-interrupts.c
@@ -44,7 +44,7 @@ static int perf_open(void)
 	attr.type = i915_type_id();
 	if (attr.type == 0)
 		return -ENOENT;
-	attr.config = I915_PERF_INTERRUPTS;
+	attr.config = I915_PMU_INTERRUPTS;
 
 	return perf_event_open(&attr, -1, 0, -1, 0);
 }
diff --git a/overlay/gpu-freq.c b/overlay/gpu-freq.c
index 321c93882238..bfd79760a7c8 100644
--- a/overlay/gpu-freq.c
+++ b/overlay/gpu-freq.c
@@ -54,8 +54,8 @@ static int perf_open(void)
 {
 	int fd;
 
-	fd = perf_i915_open(I915_PERF_ACTUAL_FREQUENCY, -1);
-	if (perf_i915_open(I915_PERF_REQUESTED_FREQUENCY, fd) < 0) {
+	fd = perf_i915_open(I915_PMU_ACTUAL_FREQUENCY, -1);
+	if (perf_i915_open(I915_PMU_REQUESTED_FREQUENCY, fd) < 0) {
 		close(fd);
 		fd = -1;
 	}
diff --git a/overlay/gpu-top.c b/overlay/gpu-top.c
index 891a7ea7c0b1..fa3d8c5d4e3c 100644
--- a/overlay/gpu-top.c
+++ b/overlay/gpu-top.c
@@ -42,11 +42,6 @@
 #define   RING_WAIT		(1<<11)
 #define   RING_WAIT_SEMAPHORE	(1<<10)
 
-#define __I915_PERF_RING(n) (4*n)
-#define I915_PERF_RING_BUSY(n) (__I915_PERF_RING(n) + 0)
-#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_i915_open(int config, int group)
 {
 	struct perf_event_attr attr;
@@ -67,39 +62,53 @@ static int perf_i915_open(int config, int group)
 
 static int perf_init(struct gpu_top *gt)
 {
-	const char *names[] = {
-		"RCS",
-		"BCS",
-		"VCS0",
-		"VCS1",
-		NULL,
+	struct engine_desc {
+		unsigned class, instance;
+		const char *name;
+	} *d, engines[] = {
+		{ I915_ENGINE_CLASS_RENDER, 0, "rcs0" },
+		{ I915_ENGINE_CLASS_COPY, 0, "bcs0" },
+		{ I915_ENGINE_CLASS_VIDEO, 0, "vcs0" },
+		{ I915_ENGINE_CLASS_VIDEO, 1, "vcs1" },
+		{ I915_ENGINE_CLASS_VIDEO_ENHANCE, 0, "vecs0" },
+		{ 0, 0, NULL }
 	};
-	int n;
 
-	gt->fd = perf_i915_open(I915_PERF_RING_BUSY(0), -1);
+	d = &engines[0];
+
+	gt->fd = perf_i915_open(I915_PMU_ENGINE_BUSY(d->class, d->instance),
+				-1);
 	if (gt->fd < 0)
 		return -1;
 
-	if (perf_i915_open(I915_PERF_RING_WAIT(0), gt->fd) >= 0)
+	if (perf_i915_open(I915_PMU_ENGINE_WAIT(d->class, d->instance),
+			   gt->fd) >= 0)
 		gt->have_wait = 1;
 
-	if (perf_i915_open(I915_PERF_RING_SEMA(0), gt->fd) >= 0)
+	if (perf_i915_open(I915_PMU_ENGINE_SEMA(d->class, d->instance),
+			   gt->fd) >= 0)
 		gt->have_sema = 1;
 
-	gt->ring[0].name = names[0];
+	gt->ring[0].name = d->name;
 	gt->num_rings = 1;
 
-	for (n = 1; names[n]; n++) {
-		if (perf_i915_open(I915_PERF_RING_BUSY(n), gt->fd) >= 0) {
+	for (d++; d->name; d++) {
+		if (perf_i915_open(I915_PMU_ENGINE_BUSY(d->class, d->instance),
+				   gt->fd) >= 0) {
 			if (gt->have_wait &&
-			    perf_i915_open(I915_PERF_RING_WAIT(n), gt->fd) < 0)
+			    perf_i915_open(I915_PMU_ENGINE_WAIT(d->class,
+								d->instance),
+					   gt->fd) < 0)
 				return -1;
 
 			if (gt->have_sema &&
-			    perf_i915_open(I915_PERF_RING_SEMA(n), gt->fd) < 0)
+			    perf_i915_open(I915_PMU_ENGINE_SEMA(d->class,
+								d->instance),
+					   gt->fd) < 0)
 				return -1;
 
-			gt->ring[gt->num_rings++].name = names[n];
+			gt->ring[gt->num_rings++].name = d->name;
+printf("%s busy = %x\n", d->name, I915_PMU_ENGINE_BUSY(d->class, d->instance));
 		}
 	}
 
diff --git a/overlay/perf.h b/overlay/perf.h
index c44e65f9734c..2270f435fbc3 100644
--- a/overlay/perf.h
+++ b/overlay/perf.h
@@ -3,41 +3,55 @@
 
 #include <linux/perf_event.h>
 
-#define I915_SAMPLE_BUSY	0
-#define I915_SAMPLE_WAIT	1
-#define I915_SAMPLE_SEMA	2
+enum drm_i915_gem_engine_class {
+	I915_ENGINE_CLASS_OTHER = 0,
+	I915_ENGINE_CLASS_RENDER = 1,
+	I915_ENGINE_CLASS_COPY = 2,
+	I915_ENGINE_CLASS_VIDEO = 3,
+	I915_ENGINE_CLASS_VIDEO_ENHANCE = 4,
+	I915_ENGINE_CLASS_MAX /* non-ABI */
+};
 
-#define I915_SAMPLE_RCS		0
-#define I915_SAMPLE_VCS		1
-#define I915_SAMPLE_BCS		2
-#define I915_SAMPLE_VECS	3
+enum drm_i915_pmu_engine_sample {
+	I915_SAMPLE_QUEUED = 0,
+	I915_SAMPLE_BUSY = 1,
+	I915_SAMPLE_WAIT = 2,
+	I915_SAMPLE_SEMA = 3
+};
 
-#define __I915_PERF_COUNT(ring, id) ((ring) << 4 | (id))
+#define I915_PMU_SAMPLE_BITS (4)
+#define I915_PMU_SAMPLE_MASK (0xf)
+#define I915_PMU_SAMPLE_INSTANCE_BITS (8)
+#define I915_PMU_CLASS_SHIFT \
+	(I915_PMU_SAMPLE_BITS + I915_PMU_SAMPLE_INSTANCE_BITS)
 
-#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_PMU_ENGINE(class, instance, sample) \
+	((class) << I915_PMU_CLASS_SHIFT | \
+	(instance) << I915_PMU_SAMPLE_BITS | \
+	(sample))
 
-#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_PMU_ENGINE_QUEUED(class, instance) \
+	__I915_PMU_ENGINE(class, instance, I915_SAMPLE_QUEUED)
 
-#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_PMU_ENGINE_BUSY(class, instance) \
+	__I915_PMU_ENGINE(class, instance, I915_SAMPLE_BUSY)
 
-#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_PMU_ENGINE_WAIT(class, instance) \
+	__I915_PMU_ENGINE(class, instance, I915_SAMPLE_WAIT)
 
-#define I915_PERF_ACTUAL_FREQUENCY 32
-#define I915_PERF_REQUESTED_FREQUENCY 33
-#define I915_PERF_ENERGY 34
-#define I915_PERF_INTERRUPTS 35
+#define I915_PMU_ENGINE_SEMA(class, instance) \
+	__I915_PMU_ENGINE(class, instance, I915_SAMPLE_SEMA)
 
-#define I915_PERF_RC6_RESIDENCY		40
-#define I915_PERF_RC6p_RESIDENCY	41
-#define I915_PERF_RC6pp_RESIDENCY	42
+#define __I915_PMU_OTHER(x) (__I915_PMU_ENGINE(0xff, 0xff, 0xf) + 1 + (x))
+
+#define I915_PMU_ACTUAL_FREQUENCY 	__I915_PMU_OTHER(0)
+#define I915_PMU_REQUESTED_FREQUENCY	__I915_PMU_OTHER(1)
+#define I915_PMU_ENERGY			__I915_PMU_OTHER(2)
+#define I915_PMU_INTERRUPTS		__I915_PMU_OTHER(3)
+
+#define I915_PMU_RC6_RESIDENCY		__I915_PMU_OTHER(4)
+#define I915_PMU_RC6p_RESIDENCY		__I915_PMU_OTHER(5)
+#define I915_PMU_RC6pp_RESIDENCY	__I915_PMU_OTHER(6)
 
 static inline int
 perf_event_open(struct perf_event_attr *attr,
diff --git a/overlay/power.c b/overlay/power.c
index 2f1521b82cd6..b84082e8cef0 100644
--- a/overlay/power.c
+++ b/overlay/power.c
@@ -46,7 +46,7 @@ static int perf_open(void)
 	attr.type = i915_type_id();
 	if (attr.type == 0)
 		return -1;
-	attr.config = I915_PERF_ENERGY;
+	attr.config = I915_PMU_ENERGY;
 
 	attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED;
 	return perf_event_open(&attr, -1, 0, -1, 0);
diff --git a/overlay/rc6.c b/overlay/rc6.c
index d7047c2f4880..4809bbd8b0b7 100644
--- a/overlay/rc6.c
+++ b/overlay/rc6.c
@@ -60,15 +60,15 @@ static int perf_open(unsigned *flags)
 {
 	int fd;
 
-	fd = perf_i915_open(I915_PERF_RC6_RESIDENCY, -1);
+	fd = perf_i915_open(I915_PMU_RC6_RESIDENCY, -1);
 	if (fd < 0)
 		return -1;
 
 	*flags |= RC6;
-	if (perf_i915_open(I915_PERF_RC6p_RESIDENCY, fd) >= 0)
+	if (perf_i915_open(I915_PMU_RC6p_RESIDENCY, fd) >= 0)
 		*flags |= RC6p;
 
-	if (perf_i915_open(I915_PERF_RC6pp_RESIDENCY, fd) >= 0)
+	if (perf_i915_open(I915_PMU_RC6pp_RESIDENCY, fd) >= 0)
 		*flags |= RC6pp;
 
 	return fd;
-- 
2.9.4



More information about the Intel-gfx mailing list