All the mail mirrored from lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] perf documentation: Add desription forHEADER_TRACING_DATA
@ 2020-03-03  7:08 Thomas Richter
  2020-03-11 13:58 ` Arnaldo Carvalho de Melo
  0 siblings, 1 reply; 3+ messages in thread
From: Thomas Richter @ 2020-03-03  7:08 UTC (permalink / raw
  To: linux-kernel, linux-perf-users, acme
  Cc: gor, sumanthk, heiko.carstens, Thomas Richter

Add description and layout in the perf.data file for
the header part describing trace data used in commands
perf record -e XXX
where XXX is a probe or tracepoint.

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
---
 .../Documentation/perf.data-file-format.txt   | 106 +++++++++++++++++-
 1 file changed, 105 insertions(+), 1 deletion(-)

diff --git a/tools/perf/Documentation/perf.data-file-format.txt b/tools/perf/Documentation/perf.data-file-format.txt
index b0152e1095c5..e3f78269e417 100644
--- a/tools/perf/Documentation/perf.data-file-format.txt
+++ b/tools/perf/Documentation/perf.data-file-format.txt
@@ -67,9 +67,113 @@ flags. These define the valid bits:
 
 	HEADER_RESERVED		= 0,	/* always cleared */
 	HEADER_FIRST_FEATURE	= 1,
+
 	HEADER_TRACING_DATA	= 1,
 
-Describe me.
+The header consists of several entries of variable length such as file
+contents of several trace files like 'header_page', 'header_event'.
+Also complete probe/<event-name>/format files are included.
+The numbers in parenthesis are structure members refered to later in
+the description and example.
+
+struct tracing_data {
+	char birthday[3] = { 23, 8, 68 };	Eyecatcher
+	char tracing[12] = "tracing0.6";	Type and version
+	uint8_t bigendian;			Big or little endian platform
+	uint8_t sizelong;			Size of long in bytes
+	uint32_t pagesize:			Pagesize in bytes
+	char hdr_page[12] = "header_page"	Name of file             (1)
+	uint64_t hdr_page_size;			File size in bytes       (1a)
+	char hdr_page_contents[hdr_page_size];	Contents of file         (1b)
+	char hdr_event[13] = "header_event"	Name of file             (2)
+	uint64_t hdr_event_size;		File size in bytes
+	char hdr_page_contents[hdr_event_size];	Contents of file
+	uint32_t ftrace_count;			Number of ftrace entries (3)
+	struct {				Size and content of file
+	    uint64_t file_size;
+	    char file_content[file_size]
+	} ftrace[ftrace_count]			ftrace_count array elements
+	uint32_t event_types;			Number of event types    (4)
+	struct {
+	    char name[];			Name, null terminatedg   (5)
+	    uint32_t count;			Number of probes         (6)
+	    struct {
+	       uint64_t file_size;					 (7)
+	       char file_content[file_size];				 (8)
+	    }[count];				Count always > 0 or omitted
+	}[event_types];
+	uint32_t proc_kallsyms_size;		Always zero
+	uint64_t printk_file_size;		File size in bytes       (9)
+	char printk_format[printk_file_size];	Contents of file
+	uint64_t saved_cmdlines_size;		File size in bytes       (10)
+	char saved_cmdlines[saved_cmdlines_size];	Contents of file
+};
+
+The structure description is straight forward. It contains file name, size
+and contents from
+- /sys/kernel/debug/tracing/events/header_page (1, 1a, 1b)
+and
+- /sys/kernel/debug/tracing/events/header_event (2).
+
+This is followed by a variable length part of format files found in directory
+/sys/kernel/debug/tracing//events/ftrace/*/format for events selected
+by the perf record -e XXX commands. Note member 'ftrace_count' can be zero.
+
+Next are the number of probe types specified by the perf record -e XXXX
+command (4). Each type has stored its name (5), which is a null terminated
+string, and the number of probes (6) belonging to this type. For each
+active probe follows the file size (7) and file contents (8) of that probe.
+
+Finally the file size and contents of two more files is saved:
+- Size of /proc/kallsyms (always hard coded to zero and no file contents)
+- /sys/kernel/debug/tracing/printk_formats (9)
+- /sys/kernel/debug/tracing/saved_cmdlines (10)
+
+Example for the dynamic probes and trace points:
+Consider these probes have been set up:
+ # ./perf probe -l
+  probe:vfs_getname    (on getname_flags:72@fs/namei.c with pathname)
+  probe:xxx            (on do_file_open_root:11@fs/namei.c)
+ #
+and this command used for recording:
+ # ./perf record -e probe:vfs_getname -e probe:xxx \
+	-e tcp:tcp_retransmit_skb -- touch /tmp/xxx
+
+This command line leads to two events in directory
+   /sys/kernel/debug/tracing/events/probe/vfs_getname
+   /sys/kernel/debug/tracing/events/probe/xxx
+and a tracepoint in directory
+   /sys/kernel/debug/tracing/events/tcp/tcp_retransmit_skb.
+
+This results in this event_count (5) layout in the perf.data:
+using command:
+ # od -Ax -t x4z -v perf.data
+
+0011a0 00000000 00000270 726f6265 00000000  >.......probe....<
+             ^^ (4)^^^'' ''(5)''' ''--(6)-
+	     4: # of types --> 2
+	     5: first one named 'probe'
+	     6: 2 probes defined
+0011b0 02000000 00000001 766e616d 653a2078  >........vname: x<
+       --###### (7)##### ##
+             7: first one 'named' xxx with format file size 0x176
+0011c0 78780a49 443a2031 3138360a 666f726d  >xx.ID: 1186.form<
+...
+001320 45432d3e 5f5f7072 6f62655f 69700a00  >EC->__probe_ip..<
+                                        ''
+001330 00000000 0001e36e 616d653a 20766673  >.......name: vfs<
+       ''(7)''' ''''''
+              7: second one named 'vfs_getname' with format file size 0x1e3
+001340 5f676574 6e616d65 0a49443a 20313138  >_getname.ID: 118<
+...
+001510 70617468 6e616d65 290a7463 70000000  >pathname).tcp...<
+                             ''(5 )'--(6)-
+001520 00010000 00000000 05696e61 6d653a20  >.........iname: <
+       ----#### (7)##### ####
+       5: second event type named 'tcp'
+       6: 1 probe defined
+       7: size of format file 0x568
+001530 7463705f 72657472 616e736d 69745f73  >tcp_retransmit_s<
 
 	HEADER_BUILD_ID = 2,
 
-- 
2.23.0


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH] perf documentation: Add desription forHEADER_TRACING_DATA
  2020-03-03  7:08 [PATCH] perf documentation: Add desription forHEADER_TRACING_DATA Thomas Richter
@ 2020-03-11 13:58 ` Arnaldo Carvalho de Melo
  2020-03-12 11:48   ` Thomas Richter
  0 siblings, 1 reply; 3+ messages in thread
From: Arnaldo Carvalho de Melo @ 2020-03-11 13:58 UTC (permalink / raw
  To: Thomas Richter
  Cc: linux-kernel, linux-perf-users, gor, sumanthk, heiko.carstens

Em Tue, Mar 03, 2020 at 08:08:46AM +0100, Thomas Richter escreveu:
> Add description and layout in the perf.data file for
> the header part describing trace data used in commands
> perf record -e XXX
> where XXX is a probe or tracepoint.

Did you write this from scratch?  I was going thru it and couldn't find
what that "Note member 'ftrace_count' can be zero." refers to, couldn't
find this ftrace_counter, is this outdated?

[root@five ~]# find /sys/kernel/debug/tracing/events/ -type f | xargs grep ftrace_count
[root@five ~]# find /sys/kernel/debug/tracing/events/ -name ftrace_count
[root@five ~]# uname -a
Linux five 5.5.5-200.fc31.x86_64 #1 SMP Wed Feb 19 23:28:07 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
[root@five ~]#

The part about using a probe and then go on dissecting it is really
nice, I'll try reproducing it soon,

Thanks,

- Arnaldo
 
> Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
> ---
>  .../Documentation/perf.data-file-format.txt   | 106 +++++++++++++++++-
>  1 file changed, 105 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/perf/Documentation/perf.data-file-format.txt b/tools/perf/Documentation/perf.data-file-format.txt
> index b0152e1095c5..e3f78269e417 100644
> --- a/tools/perf/Documentation/perf.data-file-format.txt
> +++ b/tools/perf/Documentation/perf.data-file-format.txt
> @@ -67,9 +67,113 @@ flags. These define the valid bits:
>  
>  	HEADER_RESERVED		= 0,	/* always cleared */
>  	HEADER_FIRST_FEATURE	= 1,
> +
>  	HEADER_TRACING_DATA	= 1,
>  
> -Describe me.
> +The header consists of several entries of variable length such as file
> +contents of several trace files like 'header_page', 'header_event'.
> +Also complete probe/<event-name>/format files are included.
> +The numbers in parenthesis are structure members refered to later in
> +the description and example.
> +
> +struct tracing_data {
> +	char birthday[3] = { 23, 8, 68 };	Eyecatcher
> +	char tracing[12] = "tracing0.6";	Type and version
> +	uint8_t bigendian;			Big or little endian platform
> +	uint8_t sizelong;			Size of long in bytes
> +	uint32_t pagesize:			Pagesize in bytes
> +	char hdr_page[12] = "header_page"	Name of file             (1)
> +	uint64_t hdr_page_size;			File size in bytes       (1a)
> +	char hdr_page_contents[hdr_page_size];	Contents of file         (1b)
> +	char hdr_event[13] = "header_event"	Name of file             (2)
> +	uint64_t hdr_event_size;		File size in bytes
> +	char hdr_page_contents[hdr_event_size];	Contents of file
> +	uint32_t ftrace_count;			Number of ftrace entries (3)
> +	struct {				Size and content of file
> +	    uint64_t file_size;
> +	    char file_content[file_size]
> +	} ftrace[ftrace_count]			ftrace_count array elements
> +	uint32_t event_types;			Number of event types    (4)
> +	struct {
> +	    char name[];			Name, null terminatedg   (5)
> +	    uint32_t count;			Number of probes         (6)
> +	    struct {
> +	       uint64_t file_size;					 (7)
> +	       char file_content[file_size];				 (8)
> +	    }[count];				Count always > 0 or omitted
> +	}[event_types];
> +	uint32_t proc_kallsyms_size;		Always zero
> +	uint64_t printk_file_size;		File size in bytes       (9)
> +	char printk_format[printk_file_size];	Contents of file
> +	uint64_t saved_cmdlines_size;		File size in bytes       (10)
> +	char saved_cmdlines[saved_cmdlines_size];	Contents of file
> +};
> +
> +The structure description is straight forward. It contains file name, size
> +and contents from
> +- /sys/kernel/debug/tracing/events/header_page (1, 1a, 1b)
> +and
> +- /sys/kernel/debug/tracing/events/header_event (2).
> +
> +This is followed by a variable length part of format files found in directory
> +/sys/kernel/debug/tracing//events/ftrace/*/format for events selected
> +by the perf record -e XXX commands. Note member 'ftrace_count' can be zero.
> +
> +Next are the number of probe types specified by the perf record -e XXXX
> +command (4). Each type has stored its name (5), which is a null terminated
> +string, and the number of probes (6) belonging to this type. For each
> +active probe follows the file size (7) and file contents (8) of that probe.
> +
> +Finally the file size and contents of two more files is saved:
> +- Size of /proc/kallsyms (always hard coded to zero and no file contents)
> +- /sys/kernel/debug/tracing/printk_formats (9)
> +- /sys/kernel/debug/tracing/saved_cmdlines (10)
> +
> +Example for the dynamic probes and trace points:
> +Consider these probes have been set up:
> + # ./perf probe -l
> +  probe:vfs_getname    (on getname_flags:72@fs/namei.c with pathname)
> +  probe:xxx            (on do_file_open_root:11@fs/namei.c)
> + #
> +and this command used for recording:
> + # ./perf record -e probe:vfs_getname -e probe:xxx \
> +	-e tcp:tcp_retransmit_skb -- touch /tmp/xxx
> +
> +This command line leads to two events in directory
> +   /sys/kernel/debug/tracing/events/probe/vfs_getname
> +   /sys/kernel/debug/tracing/events/probe/xxx
> +and a tracepoint in directory
> +   /sys/kernel/debug/tracing/events/tcp/tcp_retransmit_skb.
> +
> +This results in this event_count (5) layout in the perf.data:
> +using command:
> + # od -Ax -t x4z -v perf.data
> +
> +0011a0 00000000 00000270 726f6265 00000000  >.......probe....<
> +             ^^ (4)^^^'' ''(5)''' ''--(6)-
> +	     4: # of types --> 2
> +	     5: first one named 'probe'
> +	     6: 2 probes defined
> +0011b0 02000000 00000001 766e616d 653a2078  >........vname: x<
> +       --###### (7)##### ##
> +             7: first one 'named' xxx with format file size 0x176
> +0011c0 78780a49 443a2031 3138360a 666f726d  >xx.ID: 1186.form<
> +...
> +001320 45432d3e 5f5f7072 6f62655f 69700a00  >EC->__probe_ip..<
> +                                        ''
> +001330 00000000 0001e36e 616d653a 20766673  >.......name: vfs<
> +       ''(7)''' ''''''
> +              7: second one named 'vfs_getname' with format file size 0x1e3
> +001340 5f676574 6e616d65 0a49443a 20313138  >_getname.ID: 118<
> +...
> +001510 70617468 6e616d65 290a7463 70000000  >pathname).tcp...<
> +                             ''(5 )'--(6)-
> +001520 00010000 00000000 05696e61 6d653a20  >.........iname: <
> +       ----#### (7)##### ####
> +       5: second event type named 'tcp'
> +       6: 1 probe defined
> +       7: size of format file 0x568
> +001530 7463705f 72657472 616e736d 69745f73  >tcp_retransmit_s<
>  
>  	HEADER_BUILD_ID = 2,
>  
> -- 
> 2.23.0
> 

-- 

- Arnaldo

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH] perf documentation: Add desription forHEADER_TRACING_DATA
  2020-03-11 13:58 ` Arnaldo Carvalho de Melo
@ 2020-03-12 11:48   ` Thomas Richter
  0 siblings, 0 replies; 3+ messages in thread
From: Thomas Richter @ 2020-03-12 11:48 UTC (permalink / raw
  To: Arnaldo Carvalho de Melo
  Cc: linux-kernel, linux-perf-users, gor, sumanthk, heiko.carstens

On 3/11/20 2:58 PM, Arnaldo Carvalho de Melo wrote:
> Em Tue, Mar 03, 2020 at 08:08:46AM +0100, Thomas Richter escreveu:
>> Add description and layout in the perf.data file for
>> the header part describing trace data used in commands
>> perf record -e XXX
>> where XXX is a probe or tracepoint.
> 
> Did you write this from scratch?  I was going thru it and couldn't find
> what that "Note member 'ftrace_count' can be zero." refers to, couldn't
> find this ftrace_counter, is this outdated?
> 
> [root@five ~]# find /sys/kernel/debug/tracing/events/ -type f | xargs grep ftrace_count
> [root@five ~]# find /sys/kernel/debug/tracing/events/ -name ftrace_count
> [root@five ~]# uname -a
> Linux five 5.5.5-200.fc31.x86_64 #1 SMP Wed Feb 19 23:28:07 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
> [root@five ~]#
> 
> The part about using a probe and then go on dissecting it is really
> nice, I'll try reproducing it soon,
> 
> Thanks,
> 

Arnaldo,
        
yes, I wrote this from scratch. I decided to invest some time into the
documentation after I have spent nearly 3 week in debugging why commands

  perf record -e probe:vfs_getname

followed by

  perf report 
  
did not print the file name on s390. Remember we talked about failing
test case 65: Check open filename arg using perf trace + vfs_getname.
(Steven Rostedt and Masami Hiramatsu helped me on this).

On ftrace_count: This is not a file available in /sys/kernel/debug/tracing.
It is a value saved in function

tracing_data_get
  record_ftrace_files
     copy_event_system --> Check for some format file in directories
                           /sys/kernel/debug/tracing/events/ftrace/*/
Depending on what probe was defined, there is no hit and zero is written
to the output.

Hope this helps.
-- 
Thomas Richter, Dept 3252, IBM s390 Linux Development, Boeblingen, Germany
--
Vorsitzender des Aufsichtsrats: Matthias Hartmann
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2020-03-12 11:48 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2020-03-03  7:08 [PATCH] perf documentation: Add desription forHEADER_TRACING_DATA Thomas Richter
2020-03-11 13:58 ` Arnaldo Carvalho de Melo
2020-03-12 11:48   ` Thomas Richter

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.