Format of funcgraph_{entry, exit} ftrace event

Format of funcgraph_{entry, exit} ftrace event

[Matteo Bertolino]

Dear community, 
I wish to ask you some clarifications about the format of two ftrace events: `funcgraph_entry` and `funcgraph_exit`, basically those in `$ANY_DEBUG_FS/tracing/events/ftrace`.
Let me copy and paste them here:


# cat funcgraph_entry/format
name: funcgraph_entry
ID: 11
format:
	field:unsigned short common_type;	offset:0;	size:2;	signed:0;
	field:unsigned char common_flags;	offset:2;	size:1;	signed:0;
	field:unsigned char common_preempt_count;	offset:3;	size:1;	signed:0;
	field:int common_pid;	offset:4;	size:4;	signed:1;

	field:unsigned long func;	offset:8;	size:8;	signed:0;
	field:int depth;	offset:16;	size:4;	signed:1;

print fmt: "--> %ps (%d)", (void *)REC->func, REC->depth


AND

# cat funcgraph_exit/format
name: funcgraph_exit
ID: 10
format:
	field:unsigned short common_type;	offset:0;	size:2;	signed:0;
	field:unsigned char common_flags;	offset:2;	size:1;	signed:0;
	field:unsigned char common_preempt_count;	offset:3;	size:1;	signed:0;
	field:int common_pid;	offset:4;	size:4;	signed:1;

	field:unsigned long func;	offset:8;	size:8;	signed:0;
	field:int depth;	offset:16;	size:4;	signed:1;
	field:unsigned int overrun;	offset:20;	size:4;	signed:0;
	field:unsigned long long calltime;	offset:24;	size:8;	signed:0;
	field:unsigned long long rettime;	offset:32;	size:8;	signed:0;

print fmt: "<-- %ps (%d) (start: %llx  end: %llx) over: %d", (void *)REC->func, REC->depth, REC->calltime, REC->rettime, REC->depth


While writing a `trace.dat`, I know that all the entries have an header, described by the Linux ringbuffer format (file `header_event`):
# compressed entry header
	type_len    :    5 bits
	time_delta  :   27 bits
	array       :   32 bits

	padding     : type == 29
	time_extend : type == 30
	time_stamp : type == 31
	data max type_len  == 28

Such header includes a timestamp (that can be a delta of the timestamp present in the `header_page`).

I have one question for the ftrace's community:
Why are the following fields of a `funcgraph_exit` entry useful? ` calltime` and ` rettime`. 
They look like redundant for me, because the timing information is already present on the `header_event`. 

Please can you clarify me whether I am wrong? 
Best Regards && have lovely holidays,
Matteo

Re: Format of funcgraph_{entry, exit} ftrace event

[Steven Rostedt]

On Tue, 9 Aug 2022 13:32:04 +0000
Matteo Bertolino <matteo.bertolino@huawei.com> wrote:

> 
> 
> While writing a `trace.dat`, I know that all the entries have an header, described by the Linux ringbuffer format (file `header_event`):
> # compressed entry header
> 	type_len    :    5 bits
> 	time_delta  :   27 bits
> 	array       :   32 bits
> 
> 	padding     : type == 29
> 	time_extend : type == 30
> 	time_stamp : type == 31
> 	data max type_len  == 28
> 
> Such header includes a timestamp (that can be a delta of the timestamp present in the `header_page`).
> 
> I have one question for the ftrace's community:
> Why are the following fields of a `funcgraph_exit` entry useful? ` calltime` and ` rettime`. 
> They look like redundant for me, because the timing information is already present on the `header_event`. 

The function graph infrastructure records the calltime and rettime and
that can be used by other users than the tracer. That is, the users are
not guaranteed to have timestamps of their own. Yes it is redundant for
the tracing, but it is also a way to test to make sure the timings
being recorded are accurate for the other use cases.

-- Steve