gpu/trace: add a gpu total memory usage tracepoint

This change adds the below gpu memory tracepoint:
gpu_mem/gpu_mem_total: track global or proc gpu memory total usages

Per process tracking of total gpu memory usage in the gem layer is not
appropriate and hard to implement with trivial overhead. So for the gfx
device driver layer to track total gpu memory usage both globally and
per process in an easy and uniform way is to integrate the tracepoint in
this patch to the underlying varied implementations of gpu memory
tracking system from vendors.

Putting this tracepoint in the common trace events can not only help
wean the gfx drivers off of debugfs but also greatly help the downstream
Android gpu vendors because debugfs is to be deprecated in the upcoming
Android release. Then the gpu memory tracking of both Android kernel and
the upstream linux kernel can stay closely, which can benefit the whole
kernel eco-system in the long term.

Link: http://lkml.kernel.org/r/20200302235044.59163-1-zzyiwei@google.com

Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Yiwei Zhang <zzyiwei@google.com>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

1 files changed, 57 insertions, 0 deletions

diff --git a/include/trace/events/gpu_mem.h b/include/trace/events/gpu_mem.h
new file mode 100644
index 000000000000..1897822a9150
--- /dev/null
+++ b/include/trace/events/gpu_mem.h
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * GPU memory trace points
+ *
+ * Copyright (C) 2020 Google, Inc.
+ */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM gpu_mem
+
+#if !defined(_TRACE_GPU_MEM_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_GPU_MEM_H
+
+#include <linux/tracepoint.h>
+
+/*
+ * The gpu_memory_total event indicates that there's an update to either the
+ * global or process total gpu memory counters.
+ *
+ * This event should be emitted whenever the kernel device driver allocates,
+ * frees, imports, unimports memory in the GPU addressable space.
+ *
+ * @gpu_id: This is the gpu id.
+ *
+ * @pid: Put 0 for global total, while positive pid for process total.
+ *
+ * @size: Virtual size of the allocation in bytes.
+ *
+ */
+TRACE_EVENT(gpu_mem_total,
+
+	TP_PROTO(uint32_t gpu_id, uint32_t pid, uint64_t size),
+
+	TP_ARGS(gpu_id, pid, size),
+
+	TP_STRUCT__entry(
+		__field(uint32_t, gpu_id)
+		__field(uint32_t, pid)
+		__field(uint64_t, size)
+	),
+
+	TP_fast_assign(
+		__entry->gpu_id = gpu_id;
+		__entry->pid = pid;
+		__entry->size = size;
+	),
+
+	TP_printk("gpu_id=%u pid=%u size=%llu",
+		__entry->gpu_id,
+		__entry->pid,
+		__entry->size)
+);
+
+#endif /* _TRACE_GPU_MEM_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>