summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/vc4/vc4_perfmon.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/vc4/vc4_perfmon.c')
-rw-r--r--drivers/gpu/drm/vc4/vc4_perfmon.c188
1 files changed, 188 insertions, 0 deletions
diff --git a/drivers/gpu/drm/vc4/vc4_perfmon.c b/drivers/gpu/drm/vc4/vc4_perfmon.c
new file mode 100644
index 000000000000..437e7a27f21d
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_perfmon.c
@@ -0,0 +1,188 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2018 Broadcom
+ */
+
+/**
+ * DOC: VC4 V3D performance monitor module
+ *
+ * The V3D block provides 16 hardware counters which can count various events.
+ */
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define VC4_PERFMONID_MIN 1
+#define VC4_PERFMONID_MAX U32_MAX
+
+void vc4_perfmon_get(struct vc4_perfmon *perfmon)
+{
+ if (perfmon)
+ refcount_inc(&perfmon->refcnt);
+}
+
+void vc4_perfmon_put(struct vc4_perfmon *perfmon)
+{
+ if (perfmon && refcount_dec_and_test(&perfmon->refcnt))
+ kfree(perfmon);
+}
+
+void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon)
+{
+ unsigned int i;
+ u32 mask;
+
+ if (WARN_ON_ONCE(!perfmon || vc4->active_perfmon))
+ return;
+
+ for (i = 0; i < perfmon->ncounters; i++)
+ V3D_WRITE(V3D_PCTRS(i), perfmon->events[i]);
+
+ mask = GENMASK(perfmon->ncounters - 1, 0);
+ V3D_WRITE(V3D_PCTRC, mask);
+ V3D_WRITE(V3D_PCTRE, V3D_PCTRE_EN | mask);
+ vc4->active_perfmon = perfmon;
+}
+
+void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
+ bool capture)
+{
+ unsigned int i;
+
+ if (WARN_ON_ONCE(!vc4->active_perfmon ||
+ perfmon != vc4->active_perfmon))
+ return;
+
+ if (capture) {
+ for (i = 0; i < perfmon->ncounters; i++)
+ perfmon->counters[i] += V3D_READ(V3D_PCTR(i));
+ }
+
+ V3D_WRITE(V3D_PCTRE, 0);
+ vc4->active_perfmon = NULL;
+}
+
+struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id)
+{
+ struct vc4_perfmon *perfmon;
+
+ mutex_lock(&vc4file->perfmon.lock);
+ perfmon = idr_find(&vc4file->perfmon.idr, id);
+ vc4_perfmon_get(perfmon);
+ mutex_unlock(&vc4file->perfmon.lock);
+
+ return perfmon;
+}
+
+void vc4_perfmon_open_file(struct vc4_file *vc4file)
+{
+ mutex_init(&vc4file->perfmon.lock);
+ idr_init(&vc4file->perfmon.idr);
+}
+
+static int vc4_perfmon_idr_del(int id, void *elem, void *data)
+{
+ struct vc4_perfmon *perfmon = elem;
+
+ vc4_perfmon_put(perfmon);
+
+ return 0;
+}
+
+void vc4_perfmon_close_file(struct vc4_file *vc4file)
+{
+ mutex_lock(&vc4file->perfmon.lock);
+ idr_for_each(&vc4file->perfmon.idr, vc4_perfmon_idr_del, NULL);
+ idr_destroy(&vc4file->perfmon.idr);
+ mutex_unlock(&vc4file->perfmon.lock);
+}
+
+int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file_priv)
+{
+ struct vc4_file *vc4file = file_priv->driver_priv;
+ struct drm_vc4_perfmon_create *req = data;
+ struct vc4_perfmon *perfmon;
+ unsigned int i;
+ int ret;
+
+ /* Number of monitored counters cannot exceed HW limits. */
+ if (req->ncounters > DRM_VC4_MAX_PERF_COUNTERS ||
+ !req->ncounters)
+ return -EINVAL;
+
+ /* Make sure all events are valid. */
+ for (i = 0; i < req->ncounters; i++) {
+ if (req->events[i] >= VC4_PERFCNT_NUM_EVENTS)
+ return -EINVAL;
+ }
+
+ perfmon = kzalloc(sizeof(*perfmon) + (req->ncounters * sizeof(u64)),
+ GFP_KERNEL);
+ if (!perfmon)
+ return -ENOMEM;
+
+ for (i = 0; i < req->ncounters; i++)
+ perfmon->events[i] = req->events[i];
+
+ perfmon->ncounters = req->ncounters;
+
+ refcount_set(&perfmon->refcnt, 1);
+
+ mutex_lock(&vc4file->perfmon.lock);
+ ret = idr_alloc(&vc4file->perfmon.idr, perfmon, VC4_PERFMONID_MIN,
+ VC4_PERFMONID_MAX, GFP_KERNEL);
+ mutex_unlock(&vc4file->perfmon.lock);
+
+ if (ret < 0) {
+ kfree(perfmon);
+ return ret;
+ }
+
+ req->id = ret;
+ return 0;
+}
+
+int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file_priv)
+{
+ struct vc4_file *vc4file = file_priv->driver_priv;
+ struct drm_vc4_perfmon_destroy *req = data;
+ struct vc4_perfmon *perfmon;
+
+ mutex_lock(&vc4file->perfmon.lock);
+ perfmon = idr_remove(&vc4file->perfmon.idr, req->id);
+ mutex_unlock(&vc4file->perfmon.lock);
+
+ if (!perfmon)
+ return -EINVAL;
+
+ vc4_perfmon_put(perfmon);
+ return 0;
+}
+
+int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file_priv)
+{
+ struct vc4_file *vc4file = file_priv->driver_priv;
+ struct drm_vc4_perfmon_get_values *req = data;
+ struct vc4_perfmon *perfmon;
+ int ret;
+
+ mutex_lock(&vc4file->perfmon.lock);
+ perfmon = idr_find(&vc4file->perfmon.idr, req->id);
+ vc4_perfmon_get(perfmon);
+ mutex_unlock(&vc4file->perfmon.lock);
+
+ if (!perfmon)
+ return -EINVAL;
+
+ if (copy_to_user(u64_to_user_ptr(req->values_ptr), perfmon->counters,
+ perfmon->ncounters * sizeof(u64)))
+ ret = -EFAULT;
+ else
+ ret = 0;
+
+ vc4_perfmon_put(perfmon);
+ return ret;
+}