drm/v3d: Expose performance counters to userspace

The V3D engine has several hardware performance counters that can of
interest for userspace performance analysis tools.

This exposes new ioctls to create and destroy performance monitor
objects, as well as to query the counter values.

Each created performance monitor object has an ID that can be attached
to CL/CSD submissions, so the driver enables the requested counters when
the job is submitted, and updates the performance monitor values when
the job is done.

It is up to the user to ensure all the jobs have been finished before
getting the performance monitor values. It is also up to the user to
properly synchronize BCL jobs when submitting jobs with different
performance monitors attached.

Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: David Airlie <airlied@linux.ie>
Cc: Emma Anholt <emma@anholt.net>
To: dri-devel@lists.freedesktop.org
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Acked-by: Melissa Wen <mwen@igalia.com>
Signed-off-by: Melissa Wen <melissa.srw@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210608111541.461991-1-jasuarez@igalia.com

8 files changed, 470 insertions, 0 deletions

diff --git a/drivers/gpu/drm/v3d/Makefile b/drivers/gpu/drm/v3d/Makefile
index db4cfc155821..e8b314137020 100644
--- a/drivers/gpu/drm/v3d/Makefile
+++ b/drivers/gpu/drm/v3d/Makefile
@@ -9,6 +9,7 @@ v3d-y := \
 	v3d_gem.o \
 	v3d_irq.o \
 	v3d_mmu.o \
+	v3d_perfmon.o \
 	v3d_trace_points.o \
 	v3d_sched.o
 
diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 99e22beea90b..9403c3b36aca 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -94,6 +94,9 @@ static int v3d_get_param_ioctl(struct drm_device *dev, void *data,
 	case DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH:
 		args->value = 1;
 		return 0;
+	case DRM_V3D_PARAM_SUPPORTS_PERFMON:
+		args->value = (v3d->ver >= 40);
+		return 0;
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", args->param);
 		return -EINVAL;
@@ -121,6 +124,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 				      1, NULL);
 	}
 
+	v3d_perfmon_open_file(v3d_priv);
 	file->driver_priv = v3d_priv;
 
 	return 0;
@@ -136,6 +140,7 @@ v3d_postclose(struct drm_device *dev, struct drm_file *file)
 		drm_sched_entity_destroy(&v3d_priv->sched_entity[q]);
 	}
 
+	v3d_perfmon_close_file(v3d_priv);
 	kfree(v3d_priv);
 }
 
@@ -156,6 +161,9 @@ static const struct drm_ioctl_desc v3d_drm_ioctls[] = {
 	DRM_IOCTL_DEF_DRV(V3D_GET_BO_OFFSET, v3d_get_bo_offset_ioctl, DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(V3D_SUBMIT_TFU, v3d_submit_tfu_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
 	DRM_IOCTL_DEF_DRV(V3D_SUBMIT_CSD, v3d_submit_csd_ioctl, DRM_RENDER_ALLOW | DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(V3D_PERFMON_CREATE, v3d_perfmon_create_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(V3D_PERFMON_DESTROY, v3d_perfmon_destroy_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(V3D_PERFMON_GET_VALUES, v3d_perfmon_get_values_ioctl, DRM_RENDER_ALLOW),
 };
 
 static const struct drm_driver v3d_drm_driver = {
diff --git a/drivers/gpu/drm/v3d/v3d_drv.h b/drivers/gpu/drm/v3d/v3d_drv.h
index 8a390738d65b..270134779073 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.h
+++ b/drivers/gpu/drm/v3d/v3d_drv.h
@@ -37,6 +37,40 @@ struct v3d_queue_state {
 	u64 emit_seqno;
 };
 
+/* Performance monitor object. The perform lifetime is controlled by userspace
+ * using perfmon related ioctls. A perfmon can be attached to a submit_cl
+ * request, and when this is the case, HW perf counters will be activated just
+ * before the submit_cl is submitted to the GPU and disabled when the job is
+ * done. This way, only events related to a specific job will be counted.
+ */
+struct v3d_perfmon {
+	/* Tracks the number of users of the perfmon, when this counter reaches
+	 * zero the perfmon is destroyed.
+	 */
+	refcount_t refcnt;
+
+	/* Protects perfmon stop, as it can be invoked from multiple places. */
+	struct mutex lock;
+
+	/* Number of counters activated in this perfmon instance
+	 * (should be less than DRM_V3D_MAX_PERF_COUNTERS).
+	 */
+	u8 ncounters;
+
+	/* Events counted by the HW perf counters. */
+	u8 counters[DRM_V3D_MAX_PERF_COUNTERS];
+
+	/* Storage for counter values. Counters are incremented by the
+	 * HW perf counter values every time the perfmon is attached
+	 * to a GPU job.  This way, perfmon users don't have to
+	 * retrieve the results after each job if they want to track
+	 * events covering several submissions.  Note that counter
+	 * values can't be reset, but you can fake a reset by
+	 * destroying the perfmon and creating a new one.
+	 */
+	u64 values[];
+};
+
 struct v3d_dev {
 	struct drm_device drm;
 
@@ -89,6 +123,9 @@ struct v3d_dev {
 	 */
 	spinlock_t job_lock;
 
+	/* Used to track the active perfmon if any. */
+	struct v3d_perfmon *active_perfmon;
+
 	/* Protects bo_stats */
 	struct mutex bo_lock;
 
@@ -133,6 +170,11 @@ v3d_has_csd(struct v3d_dev *v3d)
 struct v3d_file_priv {
 	struct v3d_dev *v3d;
 
+	struct {
+		struct idr idr;
+		struct mutex lock;
+	} perfmon;
+
 	struct drm_sched_entity sched_entity[V3D_MAX_QUEUES];
 };
 
@@ -205,6 +247,11 @@ struct v3d_job {
 	 */
 	struct dma_fence *done_fence;
 
+	/* Pointer to a performance monitor object if the user requested it,
+	 * NULL otherwise.
+	 */
+	struct v3d_perfmon *perfmon;
+
 	/* Callback for the freeing of the job on refcount going to 0. */
 	void (*free)(struct kref *ref);
 };
@@ -353,3 +400,19 @@ void v3d_mmu_remove_ptes(struct v3d_bo *bo);
 /* v3d_sched.c */
 int v3d_sched_init(struct v3d_dev *v3d);
 void v3d_sched_fini(struct v3d_dev *v3d);
+
+/* v3d_perfmon.c */
+void v3d_perfmon_get(struct v3d_perfmon *perfmon);
+void v3d_perfmon_put(struct v3d_perfmon *perfmon);
+void v3d_perfmon_start(struct v3d_dev *v3d, struct v3d_perfmon *perfmon);
+void v3d_perfmon_stop(struct v3d_dev *v3d, struct v3d_perfmon *perfmon,
+		      bool capture);
+struct v3d_perfmon *v3d_perfmon_find(struct v3d_file_priv *v3d_priv, int id);
+void v3d_perfmon_open_file(struct v3d_file_priv *v3d_priv);
+void v3d_perfmon_close_file(struct v3d_file_priv *v3d_priv);
+int v3d_perfmon_create_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv);
+int v3d_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *file_priv);
+int v3d_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
+				 struct drm_file *file_priv);
diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
index 4eb354226972..5689da118197 100644
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -126,6 +126,8 @@ v3d_reset(struct v3d_dev *v3d)
 	v3d_mmu_set_page_table(v3d);
 	v3d_irq_reset(v3d);
 
+	v3d_perfmon_stop(v3d, v3d->active_perfmon, false);
+
 	trace_v3d_reset_end(dev);
 }
 
@@ -375,6 +377,9 @@ v3d_job_free(struct kref *ref)
 	pm_runtime_mark_last_busy(job->v3d->drm.dev);
 	pm_runtime_put_autosuspend(job->v3d->drm.dev);
 
+	if (job->perfmon)
+		v3d_perfmon_put(job->perfmon);
+
 	kfree(job);
 }
 
@@ -539,6 +544,9 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 
 	trace_v3d_submit_cl_ioctl(&v3d->drm, args->rcl_start, args->rcl_end);
 
+	if (args->pad != 0)
+		return -EINVAL;
+
 	if (args->flags != 0 &&
 	    args->flags != DRM_V3D_SUBMIT_CL_FLUSH_CACHE) {
 		DRM_INFO("invalid flags: %d\n", args->flags);
@@ -611,8 +619,20 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 	if (ret)
 		goto fail;
 
+	if (args->perfmon_id) {
+		render->base.perfmon = v3d_perfmon_find(v3d_priv,
+							args->perfmon_id);
+
+		if (!render->base.perfmon) {
+			ret = -ENOENT;
+			goto fail;
+		}
+	}
+
 	mutex_lock(&v3d->sched_lock);
 	if (bin) {
+		bin->base.perfmon = render->base.perfmon;
+		v3d_perfmon_get(bin->base.perfmon);
 		ret = v3d_push_job(v3d_priv, &bin->base, V3D_BIN);
 		if (ret)
 			goto fail_unreserve;
@@ -633,6 +653,8 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 		ret = drm_gem_fence_array_add(&clean_job->deps, render_fence);
 		if (ret)
 			goto fail_unreserve;
+		clean_job->perfmon = render->base.perfmon;
+		v3d_perfmon_get(clean_job->perfmon);
 		ret = v3d_push_job(v3d_priv, clean_job, V3D_CACHE_CLEAN);
 		if (ret)
 			goto fail_unreserve;
@@ -827,6 +849,15 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 	if (ret)
 		goto fail;
 
+	if (args->perfmon_id) {
+		job->base.perfmon = v3d_perfmon_find(v3d_priv,
+						     args->perfmon_id);
+		if (!job->base.perfmon) {
+			ret = -ENOENT;
+			goto fail;
+		}
+	}
+
 	mutex_lock(&v3d->sched_lock);
 	ret = v3d_push_job(v3d_priv, &job->base, V3D_CSD);
 	if (ret)
diff --git a/drivers/gpu/drm/v3d/v3d_perfmon.c b/drivers/gpu/drm/v3d/v3d_perfmon.c
new file mode 100644
index 000000000000..0288ef063513
--- /dev/null
+++ b/drivers/gpu/drm/v3d/v3d_perfmon.c
@@ -0,0 +1,213 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2021 Raspberry Pi
+ */
+
+#include "v3d_drv.h"
+#include "v3d_regs.h"
+
+#define V3D_PERFMONID_MIN	1
+#define V3D_PERFMONID_MAX	U32_MAX
+
+void v3d_perfmon_get(struct v3d_perfmon *perfmon)
+{
+	if (perfmon)
+		refcount_inc(&perfmon->refcnt);
+}
+
+void v3d_perfmon_put(struct v3d_perfmon *perfmon)
+{
+	if (perfmon && refcount_dec_and_test(&perfmon->refcnt))
+		kfree(perfmon);
+}
+
+void v3d_perfmon_start(struct v3d_dev *v3d, struct v3d_perfmon *perfmon)
+{
+	unsigned int i;
+	u32 mask;
+	u8 ncounters = perfmon->ncounters;
+
+	if (WARN_ON_ONCE(!perfmon || v3d->active_perfmon))
+		return;
+
+	mask = GENMASK(ncounters - 1, 0);
+
+	for (i = 0; i < ncounters; i++) {
+		u32 source = i / 4;
+		u32 channel = V3D_SET_FIELD(perfmon->counters[i], V3D_PCTR_S0);
+
+		i++;
+		channel |= V3D_SET_FIELD(i < ncounters ? perfmon->counters[i] : 0,
+					 V3D_PCTR_S1);
+		i++;
+		channel |= V3D_SET_FIELD(i < ncounters ? perfmon->counters[i] : 0,
+					 V3D_PCTR_S2);
+		i++;
+		channel |= V3D_SET_FIELD(i < ncounters ? perfmon->counters[i] : 0,
+					 V3D_PCTR_S3);
+		V3D_CORE_WRITE(0, V3D_V4_PCTR_0_SRC_X(source), channel);
+	}
+
+	V3D_CORE_WRITE(0, V3D_V4_PCTR_0_CLR, mask);
+	V3D_CORE_WRITE(0, V3D_PCTR_0_OVERFLOW, mask);
+	V3D_CORE_WRITE(0, V3D_V4_PCTR_0_EN, mask);
+
+	v3d->active_perfmon = perfmon;
+}
+
+void v3d_perfmon_stop(struct v3d_dev *v3d, struct v3d_perfmon *perfmon,
+		      bool capture)
+{
+	unsigned int i;
+
+	if (!perfmon || !v3d->active_perfmon)
+		return;
+
+	mutex_lock(&perfmon->lock);
+	if (perfmon != v3d->active_perfmon) {
+		mutex_unlock(&perfmon->lock);
+		return;
+	}
+
+	if (capture)
+		for (i = 0; i < perfmon->ncounters; i++)
+			perfmon->values[i] += V3D_CORE_READ(0, V3D_PCTR_0_PCTRX(i));
+
+	V3D_CORE_WRITE(0, V3D_V4_PCTR_0_EN, 0);
+
+	v3d->active_perfmon = NULL;
+	mutex_unlock(&perfmon->lock);
+}
+
+struct v3d_perfmon *v3d_perfmon_find(struct v3d_file_priv *v3d_priv, int id)
+{
+	struct v3d_perfmon *perfmon;
+
+	mutex_lock(&v3d_priv->perfmon.lock);
+	perfmon = idr_find(&v3d_priv->perfmon.idr, id);
+	v3d_perfmon_get(perfmon);
+	mutex_unlock(&v3d_priv->perfmon.lock);
+
+	return perfmon;
+}
+
+void v3d_perfmon_open_file(struct v3d_file_priv *v3d_priv)
+{
+	mutex_init(&v3d_priv->perfmon.lock);
+	idr_init(&v3d_priv->perfmon.idr);
+}
+
+static int v3d_perfmon_idr_del(int id, void *elem, void *data)
+{
+	struct v3d_perfmon *perfmon = elem;
+
+	v3d_perfmon_put(perfmon);
+
+	return 0;
+}
+
+void v3d_perfmon_close_file(struct v3d_file_priv *v3d_priv)
+{
+	mutex_lock(&v3d_priv->perfmon.lock);
+	idr_for_each(&v3d_priv->perfmon.idr, v3d_perfmon_idr_del, NULL);
+	idr_destroy(&v3d_priv->perfmon.idr);
+	mutex_unlock(&v3d_priv->perfmon.lock);
+}
+
+int v3d_perfmon_create_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv)
+{
+	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
+	struct drm_v3d_perfmon_create *req = data;
+	struct v3d_perfmon *perfmon;
+	unsigned int i;
+	int ret;
+
+	/* Number of monitored counters cannot exceed HW limits. */
+	if (req->ncounters > DRM_V3D_MAX_PERF_COUNTERS ||
+	    !req->ncounters)
+		return -EINVAL;
+
+	/* Make sure all counters are valid. */
+	for (i = 0; i < req->ncounters; i++) {
+		if (req->counters[i] >= V3D_PERFCNT_NUM)
+			return -EINVAL;
+	}
+
+	perfmon = kzalloc(struct_size(perfmon, values, req->ncounters),
+			  GFP_KERNEL);
+	if (!perfmon)
+		return -ENOMEM;
+
+	for (i = 0; i < req->ncounters; i++)
+		perfmon->counters[i] = req->counters[i];
+
+	perfmon->ncounters = req->ncounters;
+
+	refcount_set(&perfmon->refcnt, 1);
+	mutex_init(&perfmon->lock);
+
+	mutex_lock(&v3d_priv->perfmon.lock);
+	ret = idr_alloc(&v3d_priv->perfmon.idr, perfmon, V3D_PERFMONID_MIN,
+			V3D_PERFMONID_MAX, GFP_KERNEL);
+	mutex_unlock(&v3d_priv->perfmon.lock);
+
+	if (ret < 0) {
+		kfree(perfmon);
+		return ret;
+	}
+
+	req->id = ret;
+
+	return 0;
+}
+
+int v3d_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *file_priv)
+{
+	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
+	struct drm_v3d_perfmon_destroy *req = data;
+	struct v3d_perfmon *perfmon;
+
+	mutex_lock(&v3d_priv->perfmon.lock);
+	perfmon = idr_remove(&v3d_priv->perfmon.idr, req->id);
+	mutex_unlock(&v3d_priv->perfmon.lock);
+
+	if (!perfmon)
+		return -EINVAL;
+
+	v3d_perfmon_put(perfmon);
+
+	return 0;
+}
+
+int v3d_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
+				 struct drm_file *file_priv)
+{
+	struct v3d_dev *v3d = to_v3d_dev(dev);
+	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
+	struct drm_v3d_perfmon_get_values *req = data;
+	struct v3d_perfmon *perfmon;
+	int ret = 0;
+
+	if (req->pad != 0)
+		return -EINVAL;
+
+	mutex_lock(&v3d_priv->perfmon.lock);
+	perfmon = idr_find(&v3d_priv->perfmon.idr, req->id);
+	v3d_perfmon_get(perfmon);
+	mutex_unlock(&v3d_priv->perfmon.lock);
+
+	if (!perfmon)
+		return -EINVAL;
+
+	v3d_perfmon_stop(v3d, perfmon, true);
+
+	if (copy_to_user(u64_to_user_ptr(req->values_ptr), perfmon->values,
+			 perfmon->ncounters * sizeof(u64)))
+		ret = -EFAULT;
+
+	v3d_perfmon_put(perfmon);
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/v3d/v3d_regs.h b/drivers/gpu/drm/v3d/v3d_regs.h
index 9bcb57781d31..3663e0d6bf76 100644
--- a/drivers/gpu/drm/v3d/v3d_regs.h
+++ b/drivers/gpu/drm/v3d/v3d_regs.h
@@ -347,6 +347,8 @@
 /* Each src reg muxes four counters each. */
 #define V3D_V4_PCTR_0_SRC_0_3                          0x00660
 #define V3D_V4_PCTR_0_SRC_28_31                        0x0067c
+#define V3D_V4_PCTR_0_SRC_X(x)                         (V3D_V4_PCTR_0_SRC_0_3 + \
+							4 * (x))
 # define V3D_PCTR_S0_MASK                              V3D_MASK(6, 0)
 # define V3D_PCTR_S0_SHIFT                             0
 # define V3D_PCTR_S1_MASK                              V3D_MASK(14, 8)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c
index a39bdd5cfc4f..dd7fcc36d726 100644
--- a/drivers/gpu/drm/v3d/v3d_sched.c
+++ b/drivers/gpu/drm/v3d/v3d_sched.c
@@ -63,6 +63,16 @@ v3d_job_free(struct drm_sched_job *sched_job)
 	v3d_job_put(job);
 }
 
+static void
+v3d_switch_perfmon(struct v3d_dev *v3d, struct v3d_job *job)
+{
+	if (job->perfmon != v3d->active_perfmon)
+		v3d_perfmon_stop(v3d, v3d->active_perfmon, true);
+
+	if (job->perfmon && v3d->active_perfmon != job->perfmon)
+		v3d_perfmon_start(v3d, job->perfmon);
+}
+
 /*
  * Returns the fences that the job depends on, one by one.
  *
@@ -120,6 +130,8 @@ static struct dma_fence *v3d_bin_job_run(struct drm_sched_job *sched_job)
 	trace_v3d_submit_cl(dev, false, to_v3d_fence(fence)->seqno,
 			    job->start, job->end);
 
+	v3d_switch_perfmon(v3d, &job->base);
+
 	/* Set the current and end address of the control list.
 	 * Writing the end register is what starts the job.
 	 */
@@ -169,6 +181,8 @@ static struct dma_fence *v3d_render_job_run(struct drm_sched_job *sched_job)
 	trace_v3d_submit_cl(dev, true, to_v3d_fence(fence)->seqno,
 			    job->start, job->end);
 
+	v3d_switch_perfmon(v3d, &job->base);
+
 	/* XXX: Set the QCFG */
 
 	/* Set the current and end address of the control list.
@@ -240,6 +254,8 @@ v3d_csd_job_run(struct drm_sched_job *sched_job)
 
 	trace_v3d_submit_csd(dev, to_v3d_fence(fence)->seqno);
 
+	v3d_switch_perfmon(v3d, &job->base);
+
 	for (i = 1; i <= 6; i++)
 		V3D_CORE_WRITE(0, V3D_CSD_QUEUED_CFG0 + 4 * i, job->args.cfg[i]);
 	/* CFG0 write kicks off the job. */
diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
index 1ce746e228d9..4104f22fb3d3 100644
--- a/include/uapi/drm/v3d_drm.h
+++ b/include/uapi/drm/v3d_drm.h
@@ -38,6 +38,9 @@ extern "C" {
 #define DRM_V3D_GET_BO_OFFSET                     0x05
 #define DRM_V3D_SUBMIT_TFU                        0x06
 #define DRM_V3D_SUBMIT_CSD                        0x07
+#define DRM_V3D_PERFMON_CREATE                    0x08
+#define DRM_V3D_PERFMON_DESTROY                   0x09
+#define DRM_V3D_PERFMON_GET_VALUES                0x0a
 
 #define DRM_IOCTL_V3D_SUBMIT_CL           DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CL, struct drm_v3d_submit_cl)
 #define DRM_IOCTL_V3D_WAIT_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_WAIT_BO, struct drm_v3d_wait_bo)
@@ -47,6 +50,12 @@ extern "C" {
 #define DRM_IOCTL_V3D_GET_BO_OFFSET       DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_GET_BO_OFFSET, struct drm_v3d_get_bo_offset)
 #define DRM_IOCTL_V3D_SUBMIT_TFU          DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_TFU, struct drm_v3d_submit_tfu)
 #define DRM_IOCTL_V3D_SUBMIT_CSD          DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CSD, struct drm_v3d_submit_csd)
+#define DRM_IOCTL_V3D_PERFMON_CREATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_CREATE, \
+						   struct drm_v3d_perfmon_create)
+#define DRM_IOCTL_V3D_PERFMON_DESTROY     DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_DESTROY, \
+						   struct drm_v3d_perfmon_destroy)
+#define DRM_IOCTL_V3D_PERFMON_GET_VALUES  DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_GET_VALUES, \
+						   struct drm_v3d_perfmon_get_values)
 
 #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
 
@@ -127,6 +136,11 @@ struct drm_v3d_submit_cl {
 	__u32 bo_handle_count;
 
 	__u32 flags;
+
+	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
+	__u32 perfmon_id;
+
+	__u32 pad;
 };
 
 /**
@@ -195,6 +209,7 @@ enum drm_v3d_param {
 	DRM_V3D_PARAM_SUPPORTS_TFU,
 	DRM_V3D_PARAM_SUPPORTS_CSD,
 	DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH,
+	DRM_V3D_PARAM_SUPPORTS_PERFMON,
 };
 
 struct drm_v3d_get_param {
@@ -258,6 +273,127 @@ struct drm_v3d_submit_csd {
 	__u32 in_sync;
 	/* Sync object to signal when the CSD job is done. */
 	__u32 out_sync;
+
+	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
+	__u32 perfmon_id;
+};
+
+enum {
+	V3D_PERFCNT_FEP_VALID_PRIMTS_NO_PIXELS,
+	V3D_PERFCNT_FEP_VALID_PRIMS,
+	V3D_PERFCNT_FEP_EZ_NFCLIP_QUADS,
+	V3D_PERFCNT_FEP_VALID_QUADS,
+	V3D_PERFCNT_TLB_QUADS_STENCIL_FAIL,
+	V3D_PERFCNT_TLB_QUADS_STENCILZ_FAIL,
+	V3D_PERFCNT_TLB_QUADS_STENCILZ_PASS,
+	V3D_PERFCNT_TLB_QUADS_ZERO_COV,
+	V3D_PERFCNT_TLB_QUADS_NONZERO_COV,
+	V3D_PERFCNT_TLB_QUADS_WRITTEN,
+	V3D_PERFCNT_PTB_PRIM_VIEWPOINT_DISCARD,
+	V3D_PERFCNT_PTB_PRIM_CLIP,
+	V3D_PERFCNT_PTB_PRIM_REV,
+	V3D_PERFCNT_QPU_IDLE_CYCLES,
+	V3D_PERFCNT_QPU_ACTIVE_CYCLES_VERTEX_COORD_USER,
+	V3D_PERFCNT_QPU_ACTIVE_CYCLES_FRAG,
+	V3D_PERFCNT_QPU_CYCLES_VALID_INSTR,
+	V3D_PERFCNT_QPU_CYCLES_TMU_STALL,
+	V3D_PERFCNT_QPU_CYCLES_SCOREBOARD_STALL,
+	V3D_PERFCNT_QPU_CYCLES_VARYINGS_STALL,
+	V3D_PERFCNT_QPU_IC_HIT,
+	V3D_PERFCNT_QPU_IC_MISS,
+	V3D_PERFCNT_QPU_UC_HIT,
+	V3D_PERFCNT_QPU_UC_MISS,
+	V3D_PERFCNT_TMU_TCACHE_ACCESS,
+	V3D_PERFCNT_TMU_TCACHE_MISS,
+	V3D_PERFCNT_VPM_VDW_STALL,
+	V3D_PERFCNT_VPM_VCD_STALL,
+	V3D_PERFCNT_BIN_ACTIVE,
+	V3D_PERFCNT_RDR_ACTIVE,
+	V3D_PERFCNT_L2T_HITS,
+	V3D_PERFCNT_L2T_MISSES,
+	V3D_PERFCNT_CYCLE_COUNT,
+	V3D_PERFCNT_QPU_CYCLES_STALLED_VERTEX_COORD_USER,
+	V3D_PERFCNT_QPU_CYCLES_STALLED_FRAGMENT,
+	V3D_PERFCNT_PTB_PRIMS_BINNED,
+	V3D_PERFCNT_AXI_WRITES_WATCH_0,
+	V3D_PERFCNT_AXI_READS_WATCH_0,
+	V3D_PERFCNT_AXI_WRITE_STALLS_WATCH_0,
+	V3D_PERFCNT_AXI_READ_STALLS_WATCH_0,
+	V3D_PERFCNT_AXI_WRITE_BYTES_WATCH_0,
+	V3D_PERFCNT_AXI_READ_BYTES_WATCH_0,
+	V3D_PERFCNT_AXI_WRITES_WATCH_1,
+	V3D_PERFCNT_AXI_READS_WATCH_1,
+	V3D_PERFCNT_AXI_WRITE_STALLS_WATCH_1,
+	V3D_PERFCNT_AXI_READ_STALLS_WATCH_1,
+	V3D_PERFCNT_AXI_WRITE_BYTES_WATCH_1,
+	V3D_PERFCNT_AXI_READ_BYTES_WATCH_1,
+	V3D_PERFCNT_TLB_PARTIAL_QUADS,
+	V3D_PERFCNT_TMU_CONFIG_ACCESSES,
+	V3D_PERFCNT_L2T_NO_ID_STALL,
+	V3D_PERFCNT_L2T_COM_QUE_STALL,
+	V3D_PERFCNT_L2T_TMU_WRITES,
+	V3D_PERFCNT_TMU_ACTIVE_CYCLES,
+	V3D_PERFCNT_TMU_STALLED_CYCLES,
+	V3D_PERFCNT_CLE_ACTIVE,
+	V3D_PERFCNT_L2T_TMU_READS,
+	V3D_PERFCNT_L2T_CLE_READS,
+	V3D_PERFCNT_L2T_VCD_READS,
+	V3D_PERFCNT_L2T_TMUCFG_READS,
+	V3D_PERFCNT_L2T_SLC0_READS,
+	V3D_PERFCNT_L2T_SLC1_READS,
+	V3D_PERFCNT_L2T_SLC2_READS,
+	V3D_PERFCNT_L2T_TMU_W_MISSES,
+	V3D_PERFCNT_L2T_TMU_R_MISSES,
+	V3D_PERFCNT_L2T_CLE_MISSES,
+	V3D_PERFCNT_L2T_VCD_MISSES,
+	V3D_PERFCNT_L2T_TMUCFG_MISSES,
+	V3D_PERFCNT_L2T_SLC0_MISSES,
+	V3D_PERFCNT_L2T_SLC1_MISSES,
+	V3D_PERFCNT_L2T_SLC2_MISSES,
+	V3D_PERFCNT_CORE_MEM_WRITES,
+	V3D_PERFCNT_L2T_MEM_WRITES,
+	V3D_PERFCNT_PTB_MEM_WRITES,
+	V3D_PERFCNT_TLB_MEM_WRITES,
+	V3D_PERFCNT_CORE_MEM_READS,
+	V3D_PERFCNT_L2T_MEM_READS,
+	V3D_PERFCNT_PTB_MEM_READS,
+	V3D_PERFCNT_PSE_MEM_READS,
+	V3D_PERFCNT_TLB_MEM_READS,
+	V3D_PERFCNT_GMP_MEM_READS,
+	V3D_PERFCNT_PTB_W_MEM_WORDS,
+	V3D_PERFCNT_TLB_W_MEM_WORDS,
+	V3D_PERFCNT_PSE_R_MEM_WORDS,
+	V3D_PERFCNT_TLB_R_MEM_WORDS,
+	V3D_PERFCNT_TMU_MRU_HITS,
+	V3D_PERFCNT_COMPUTE_ACTIVE,
+	V3D_PERFCNT_NUM,
+};
+
+#define DRM_V3D_MAX_PERF_COUNTERS                 32
+
+struct drm_v3d_perfmon_create {
+	__u32 id;
+	__u32 ncounters;
+	__u8 counters[DRM_V3D_MAX_PERF_COUNTERS];
+};
+
+struct drm_v3d_perfmon_destroy {
+	__u32 id;
+};
+
+/*
+ * Returns the values of the performance counters tracked by this
+ * perfmon (as an array of ncounters u64 values).
+ *
+ * No implicit synchronization is performed, so the user has to
+ * guarantee that any jobs using this perfmon have already been
+ * completed  (probably by blocking on the seqno returned by the
+ * last exec that used the perfmon).
+ */
+struct drm_v3d_perfmon_get_values {
+	__u32 id;
+	__u32 pad;
+	__u64 values_ptr;
 };
 
 #if defined(__cplusplus)