perf_counter: Fix/complete ftrace event records sampling

This patch implements the kernel side support for ftrace event
record sampling.

A new counter sampling attribute is added:

PERF_SAMPLE_TP_RECORD

which requests ftrace events record sampling. In this case
if a PERF_TYPE_TRACEPOINT counter is active and a tracepoint
fires, we emit the tracepoint binary record to the
perfcounter event buffer, as a sample.

Result, after setting PERF_SAMPLE_TP_RECORD attribute from perf
record:

perf record -f -F 1 -a -e workqueue:workqueue_execution
perf report -D

0x21e18 [0x48]: event: 9
.
. ... raw event: size 72 bytes
. 0000: 09 00 00 00 01 00 48 00 d0 c7 00 81 ff ff ff ff ......H........
. 0010: 0a 00 00 00 0a 00 00 00 21 00 00 00 00 00 00 00 ........!......
. 0020: 2b 00 01 02 0a 00 00 00 0a 00 00 00 65 76 65 6e +...........eve
. 0030: 74 73 2f 31 00 00 00 00 00 00 00 00 0a 00 00 00 ts/1...........
. 0040: e0 b1 31 81 ff ff ff ff .......
.
0x21e18 [0x48]: PERF_EVENT_SAMPLE (IP, 1): 10: 0xffffffff8100c7d0 period: 33

The raw ftrace binary record starts at offset 0020.

Translation:

struct trace_entry {
type = 0x2b = 43;
flags = 1;
preempt_count = 2;
pid = 0xa = 10;
tgid = 0xa = 10;
}

thread_comm = "events/1"
thread_pid = 0xa = 10;
func = 0xffffffff8131b1e0 = flush_to_ldisc()

What will come next?

- Userspace support ('perf trace'), 'flight data recorder' mode
for perf trace, etc.

- The unconditional copy from the profiling callback brings
some costs however if someone wants no such sampling to
occur, and needs to be fixed in the future. For that we need
to have an instant access to the perf counter attribute.
This is a matter of a flag to add in the struct ftrace_event.

- Take care of the events recursivity! Don't ever try to record
a lock event for example, it seems some locking is used in
the profiling fast path and lead to a tracing recursivity.
That will be fixed using raw spinlock or recursivity
protection.

- [...]

- Profit! :-)

Signed-off-by: Frederic Weisbecker <[e-mail]>
Cc: Li Zefan <[e-mail]>
Cc: Tom Zanussi <[e-mail]>
Cc: Arnaldo Carvalho de Melo <[e-mail]>
Cc: Peter Zijlstra <[e-mail]>
Cc: Mike Galbraith <[e-mail]>
Cc: Steven Rostedt <[e-mail]>
Cc: Paul Mackerras <[e-mail]>
Cc: Pekka Enberg <[e-mail]>
Cc: Gabriel Munteanu <[e-mail]>
Cc: Lai Jiangshan <[e-mail]>
Signed-off-by: Ingo Molnar <[e-mail]>

include/linux/ftrace_event.h

···
 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
 };
 
-
+void tracing_generic_entry_update(struct trace_entry *entry,
+				  unsigned long flags,
+				  int pc);
 struct ring_buffer_event *
 trace_current_buffer_lock_reserve(int type, unsigned long len,
 				  unsigned long flags, int pc);

include/linux/perf_counter.h

···
 	PERF_SAMPLE_CPU				= 1U << 7,
 	PERF_SAMPLE_PERIOD			= 1U << 8,
 	PERF_SAMPLE_STREAM_ID			= 1U << 9,
+	PERF_SAMPLE_TP_RECORD			= 1U << 10,
 
-	PERF_SAMPLE_MAX = 1U << 10,		/* non-ABI */
+	PERF_SAMPLE_MAX = 1U << 11,		/* non-ABI */
 };
 
 /*
···
 	__u64				ip[PERF_MAX_STACK_DEPTH];
 };
 
+struct perf_tracepoint_record {
+	int				size;
+	char				*record;
+};
+
 struct task_struct;
 
 /**
···
 	struct pt_regs			*regs;
 	u64				addr;
 	u64				period;
+	void				*private;
 };
 
 extern int perf_counter_overflow(struct perf_counter *counter, int nmi,

include/trace/ftrace.h

···
 /*
  * Generate the functions needed for tracepoint perf_counter support.
  *
- * static void ftrace_profile_<call>(proto)
- * {
- * 	extern void perf_tpcounter_event(int, u64, u64);
- * 	u64 __addr = 0, __count = 1;
- *
- * 	<assign>   <-- here we expand the TP_perf_assign() macro
- *
- * 	perf_tpcounter_event(event_<call>.id, __addr, __count);
- * }
+ * NOTE: The insertion profile callback (ftrace_profile_<call>) is defined later
  *
  * static int ftrace_profile_enable_<call>(struct ftrace_event_call *event_call)
  * {
···
  *
  */
 
-#undef TP_fast_assign
-#define TP_fast_assign(args...)
-
-#undef TP_perf_assign
-#define TP_perf_assign(args...) args
-
-#undef __perf_addr
-#define __perf_addr(a) __addr = (a)
-
-#undef __perf_count
-#define __perf_count(c) __count = (c)
-
 #undef TRACE_EVENT
 #define TRACE_EVENT(call, proto, args, tstruct, assign, print)		\
 									\
-static void ftrace_profile_##call(proto)				\
-{									\
-	extern void perf_tpcounter_event(int, u64, u64);		\
-	u64 __addr = 0, __count = 1;					\
-	{ assign; }							\
-	perf_tpcounter_event(event_##call.id, __addr, __count);		\
-}									\
+static void ftrace_profile_##call(proto);				\
 									\
 static int ftrace_profile_enable_##call(struct ftrace_event_call *event_call) \
 {									\
···
 }
 
 #include TRACE_INCLUDE(TRACE_INCLUDE_FILE)
-
-#undef TP_fast_assign
-#define TP_fast_assign(args...) args
-
-#undef TP_perf_assign
-#define TP_perf_assign(args...)
 
 #endif
 
···
 }
 
 #include TRACE_INCLUDE(TRACE_INCLUDE_FILE)
+
+/*
+ * Define the insertion callback to profile events
+ *
+ * The job is very similar to ftrace_raw_event_<call> except that we don't
+ * insert in the ring buffer but in a perf counter.
+ *
+ * static void ftrace_profile_<call>(proto)
+ * {
+ *	struct ftrace_data_offsets_<call> __maybe_unused __data_offsets;
+ *	struct ftrace_event_call *event_call = &event_<call>;
+ *	extern void perf_tpcounter_event(int, u64, u64, void *, int);
+ *	struct ftrace_raw_##call *entry;
+ *	u64 __addr = 0, __count = 1;
+ *	unsigned long irq_flags;
+ *	int __entry_size;
+ *	int __data_size;
+ *	int pc;
+ *
+ *	local_save_flags(irq_flags);
+ *	pc = preempt_count();
+ *
+ *	__data_size = ftrace_get_offsets_<call>(&__data_offsets, args);
+ *	__entry_size = __data_size + sizeof(*entry);
+ *
+ *	do {
+ *		char raw_data[__entry_size]; <- allocate our sample in the stack
+ *		struct trace_entry *ent;
+ *
+ *		entry = (struct ftrace_raw_<call> *)raw_data;
+ *		ent = &entry->ent;
+ *		tracing_generic_entry_update(ent, irq_flags, pc);
+ *		ent->type = event_call->id;
+ *
+ *		<tstruct> <- do some jobs with dynamic arrays
+ *
+ *		<assign>  <- affect our values
+ *
+ *		perf_tpcounter_event(event_call->id, __addr, __count, entry,
+ *			     __entry_size);  <- submit them to perf counter
+ *	} while (0);
+ *
+ * }
+ */
+
+#ifdef CONFIG_EVENT_PROFILE
+
+#undef __perf_addr
+#define __perf_addr(a) __addr = (a)
+
+#undef __perf_count
+#define __perf_count(c) __count = (c)
+
+#undef TRACE_EVENT
+#define TRACE_EVENT(call, proto, args, tstruct, assign, print)		\
+static void ftrace_profile_##call(proto)				\
+{									\
+	struct ftrace_data_offsets_##call __maybe_unused __data_offsets;\
+	struct ftrace_event_call *event_call = &event_##call;		\
+	extern void perf_tpcounter_event(int, u64, u64, void *, int);	\
+	struct ftrace_raw_##call *entry;				\
+	u64 __addr = 0, __count = 1;					\
+	unsigned long irq_flags;					\
+	int __entry_size;						\
+	int __data_size;						\
+	int pc;								\
+									\
+	local_save_flags(irq_flags);					\
+	pc = preempt_count();						\
+									\
+	__data_size = ftrace_get_offsets_##call(&__data_offsets, args); \
+	__entry_size = ALIGN(__data_size + sizeof(*entry), sizeof(u64));\
+									\
+	do {								\
+		char raw_data[__entry_size];				\
+		struct trace_entry *ent;				\
+									\
+		entry = (struct ftrace_raw_##call *)raw_data;		\
+		ent = &entry->ent;					\
+		tracing_generic_entry_update(ent, irq_flags, pc);	\
+		ent->type = event_call->id;				\
+									\
+		tstruct							\
+									\
+		{ assign; }						\
+									\
+		perf_tpcounter_event(event_call->id, __addr, __count, entry,\
+			     __entry_size);				\
+	} while (0);							\
+									\
+}
+
+#include TRACE_INCLUDE(TRACE_INCLUDE_FILE)
+#endif /* CONFIG_EVENT_PROFILE */
 
 #undef _TRACE_PROFILE_INIT
 

kernel/perf_counter.c

···
 		u64 counter;
 	} group_entry;
 	struct perf_callchain_entry *callchain = NULL;
+	struct perf_tracepoint_record *tp;
 	int callchain_size = 0;
 	u64 time;
 	struct {
···
 			header.size += sizeof(u64);
 	}
 
+	if (sample_type & PERF_SAMPLE_TP_RECORD) {
+		tp = data->private;
+		header.size += tp->size;
+	}
+
 	ret = perf_output_begin(&handle, counter, header.size, nmi, 1);
 	if (ret)
 		return;
···
 			perf_output_put(&handle, nr);
 		}
 	}
+
+	if (sample_type & PERF_SAMPLE_TP_RECORD)
+		perf_output_copy(&handle, tp->record, tp->size);
 
 	perf_output_end(&handle);
 }
···
 };
 
 #ifdef CONFIG_EVENT_PROFILE
-void perf_tpcounter_event(int event_id, u64 addr, u64 count)
+void perf_tpcounter_event(int event_id, u64 addr, u64 count, void *record,
+			  int entry_size)
 {
+	struct perf_tracepoint_record tp = {
+		.size = entry_size,
+		.record = record,
+	};
+
 	struct perf_sample_data data = {
 		.regs = get_irq_regs(),
 		.addr = addr,
+		.private = &tp,
 	};
 
 	if (!data.regs)

kernel/trace/trace.c

···
 		((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
 		(need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
 }
+EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
 
 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
 						    int type,

kernel/trace/trace.h

···
 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
 					  int *ent_cpu, u64 *ent_ts);
 
-void tracing_generic_entry_update(struct trace_entry *entry,
-				  unsigned long flags,
-				  int pc);
-
 void default_wait_pipe(struct trace_iterator *iter);
 void poll_wait_pipe(struct trace_iterator *iter);
 

tools/perf/builtin-record.c

···
 	if (call_graph)
 		attr->sample_type	|= PERF_SAMPLE_CALLCHAIN;
 
+
 	attr->mmap		= track;
 	attr->comm		= track;
 	attr->inherit		= (cpu < 0) && inherit;