A couple hist trigger patches

PATCH In review Tom Zanussi 4 participants 11 messages Feb 4, 2019 – Mar 5, 2019 Tracing

lore

Share & tools

lore

Tom Zanussi

author

2019-02-04 21:07 UTC #0 lore raw

cover A couple hist trigger patches

From: Tom Zanussi <[email protected]>

Hi,

Here are a couple miscellaneous hist trigger patches which can be applied independently of the onchange/snapshot patches.

The first is just an additional use of str_has_prefix() that was apparently missed in the original str_has_prefix() conversion, with an update suggested by Joe Perches.

The second fixes a bug I noticed when testing the onchange/snapshot fixes.

I pulled these out into this separate patchset because they should be applied regardless of what happens with the onchange/snapshot patchset.

Thanks,

Tom

The following changes since commit 67748dbeaf2b1240c0b87588df4bb0fb9471a751:

sh: ftrace: Fix missing parenthesis in WARN_ON() (2019-01-08 10:19:02 -0600)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/zanussi/linux-trace.git ftrace/hist-fixes-v1

Tom Zanussi (2):
tracing: Use str_has_prefix() in synth_event_create()
tracing: Use strncpy instead of memcpy for string keys in hist
triggers

 kernel/trace/trace_events_hist.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Signature · 1 line

2.14.1

Tom Zanussi

author

2019-02-04 21:07 UTC #1 lore raw

PATCH 1/2 tracing: Use str_has_prefix() in synth_event_create()

From: Tom Zanussi <[email protected]>

Since we now have a str_has_prefix() that returns the length, we can use that instead of explicitly calculating it.

Signed-off-by Tom Zanussi <[email protected]>

Cc Joe Perches <[email protected]>

---
 kernel/trace/trace_events_hist.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

kernel/trace/trace_events_hist.c +2 -2

index 449d90cfa151..c4a667503bf0 100644
@@ -1200,8 +1200,8 @@ static int synth_event_create(int argc, const char **argv)
 
 	/* This interface accepts group name prefix */
 	if (strchr(name, '/')) {
-		len = sizeof(SYNTH_SYSTEM "/") - 1;
-		if (strncmp(name, SYNTH_SYSTEM "/", len))
+		len = str_has_prefix(name, SYNTH_SYSTEM "/");
+		if (len == 0)
 			return -EINVAL;
 		name += len;
 	}

Signature · 1 line

2.14.1

Tom Zanussi

author

2019-02-04 21:07 UTC #2 lore raw

PATCH 2/2 tracing: Use strncpy instead of memcpy for string keys in hist triggers

In reply to Tom Zanussi

From: Tom Zanussi <[email protected]>

Because there may be random garbage beyond a string's null terminator, it's not correct to copy the the complete character array for use as a hist trigger key. This results in multiple histogram entries for the 'same' string key.

So, in the case of a string key, use strncpy instead of memcpy to avoid copying in the extra bytes.

Before, using the gdbus entries in the following hist trigger as an example:

  # echo 'hist:key=comm' > /sys/kernel/debug/tracing/events/sched/sched_waking/trigger
  # cat /sys/kernel/debug/tracing/events/sched/sched_waking/hist

...

  { comm: ImgDecoder #4                      } hitcount:        203
  { comm: gmain                              } hitcount:        213
  { comm: gmain                              } hitcount:        216
  { comm: StreamTrans #73                    } hitcount:        221
  { comm: mozStorage #3                      } hitcount:        230
  { comm: gdbus                              } hitcount:        233
  { comm: StyleThread#5                      } hitcount:        253
  { comm: gdbus                              } hitcount:        256
  { comm: gdbus                              } hitcount:        260
  { comm: StyleThread#4                      } hitcount:        271

...

  # cat /sys/kernel/debug/tracing/events/sched/sched_waking/hist | egrep gdbus | wc -l
  51

After:

  # cat /sys/kernel/debug/tracing/events/sched/sched_waking/hist | egrep gdbus | wc -l
  1

Signed-off-by Tom Zanussi <[email protected]>

Cc Namhyung Kim <[email protected]>

---
 kernel/trace/trace_events_hist.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

kernel/trace/trace_events_hist.c +3 -2

index c4a667503bf0..1b349689b897 100644
@@ -4695,9 +4695,10 @@ static inline void add_to_key(char *compound_key, void *key,
 		/* ensure NULL-termination */
 		if (size > key_field->size - 1)
 			size = key_field->size - 1;
-	}
 
-	memcpy(compound_key + key_field->offset, key, size);
+		strncpy(compound_key + key_field->offset, (char *)key, size);
+	} else
+		memcpy(compound_key + key_field->offset, key, size);
 }
 
 static void

Signature · 1 line

2.14.1

Steven Rostedt

maintainer

2019-03-04 21:50 UTC #3 lore raw

Re: [PATCH 2/2] tracing: Use strncpy instead of memcpy for string keys in hist triggers

Quoted 51 lines · On Mon, 4 Feb 2019 15:07:24 -0600 Tom Zanussi <[email protected]> wrote:

From: Tom Zanussi <[email protected]>
Because there may be random garbage beyond a string's null terminator, it's not correct to copy the the complete character array for use as a hist trigger key. This results in multiple histogram entries for the 'same' string key.
So, in the case of a string key, use strncpy instead of memcpy to avoid copying in the extra bytes.
Before, using the gdbus entries in the following hist trigger as an example:
  # echo 'hist:key=comm' > /sys/kernel/debug/tracing/events/sched/sched_waking/trigger
  # cat /sys/kernel/debug/tracing/events/sched/sched_waking/hist
...
  { comm: ImgDecoder #4                      } hitcount:        203
  { comm: gmain                              } hitcount:        213
  { comm: gmain                              } hitcount:        216
  { comm: StreamTrans #73                    } hitcount:        221
  { comm: mozStorage #3                      } hitcount:        230
  { comm: gdbus                              } hitcount:        233
  { comm: StyleThread#5                      } hitcount:        253
  { comm: gdbus                              } hitcount:        256
  { comm: gdbus                              } hitcount:        260
  { comm: StyleThread#4                      } hitcount:        271
...
  # cat /sys/kernel/debug/tracing/events/sched/sched_waking/hist | egrep gdbus | wc -l
  51
After:
  # cat /sys/kernel/debug/tracing/events/sched/sched_waking/hist | egrep gdbus | wc -l
  1
Signed-off-by Tom Zanussi <[email protected]>
Cc Namhyung Kim <[email protected]>
---
 kernel/trace/trace_events_hist.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/trace/trace_events_hist.c +0 -0
index c4a667503bf0..1b349689b897 100644
@@ -4695,9 +4695,10 @@ static inline void add_to_key(char *compound_key, void *key,
 		/* ensure NULL-termination */

kernel/trace/trace_events_hist.c static inline void add_to_key(char *compound_key, void *key, · line 4703

Quoted 10 lines

kernel/trace/trace_events_hist.c +3 -2

 		if (size > key_field->size - 1)
 			size = key_field->size - 1;
-	}
 
-	memcpy(compound_key + key_field->offset, key, size);
+		strncpy(compound_key + key_field->offset, (char *)key, size);
+	} else
+		memcpy(compound_key + key_field->offset, key, size);
 }

Shouldn't we use strncpy() in save_comm() too. Feels safer.

-- Steve

Steven Rostedt

maintainer

2019-03-04 21:56 UTC #4 lore raw

Re: [PATCH 2/2] tracing: Use strncpy instead of memcpy for string keys in hist triggers

Quoted 7 lines · On Mon, 4 Mar 2019 16:50:00 -0500 Steven Rostedt <[email protected]> wrote:

Quoted 7 lines

+++ b/kernel/trace/trace_events_hist.c

kernel/trace/trace_events_hist.c +0 -1

@@ -4695,9 +4695,10 @@ static inline void add_to_key(char *compound_key, void *key,
 		/* ensure NULL-termination */
 		if (size > key_field->size - 1)
 			size = key_field->size - 1;
-	}

Quoted 8 lines

kernel/trace/trace_events_hist.c static inline void add_to_key(char *compound_key, void *key, · line 4703
Quoted 6 lines
kernel/trace/trace_events_hist.c +3 -1
-	memcpy(compound_key + key_field->offset, key, size);
+		strncpy(compound_key + key_field->offset, (char *)key, size);
+	} else
+		memcpy(compound_key + key_field->offset, key, size);
 }
 
Shouldn't we use strncpy() in save_comm() too. Feels safer.

Note, if that is changed, it can be another patch. This one is fine as is. I just was looking at other use cases of memcpy() in that file.

-- Steve

Tom Zanussi

2019-03-04 22:22 UTC #5 lore raw

Re: [PATCH 2/2] tracing: Use strncpy instead of memcpy for string keys in hist triggers

Hi Steve,

Quoted 21 lines · Steven Rostedt

Quoted 8 lines · On Mon, 4 Mar 2019 16:50:00 -0500 Steven Rostedt <[email protected]> wrote:
Quoted 8 lines
+++ b/kernel/trace/trace_events_hist.c
kernel/trace/trace_events_hist.c +0 -1
@@ -4695,9 +4695,10 @@ static inline void add_to_key(char
 *compound_key, void *key,
 		/* ensure NULL-termination */
 		if (size > key_field->size - 1)
 			size = key_field->size - 1;
-	}
 
Quoted 9 lines
kernel/trace/trace_events_hist.c static inline void add_to_key(char *compound_key, void *key, · line 4702
Quoted 7 lines
kernel/trace/trace_events_hist.c +3 -1
-	memcpy(compound_key + key_field->offset, key, size);
+		strncpy(compound_key + key_field->offset, (char *)key,
 size);
+	} else
+		memcpy(compound_key + key_field->offset, key, size);
 }
Shouldn't we use strncpy() in save_comm() too. Feels safer.

Quoted 4 lines

Note, if that is changed, it can be another patch. This one is fine as
is. I just was looking at other use cases of memcpy() in that file.

Hmm, I don't think it's really necessary - it's not used in a key so don't care about anything after the null, and TASK_COMM_LEN is used in the memcpy.

Tom

Quoted 1 line

-- Steve

Tom Zanussi

2019-03-04 22:31 UTC #6 lore raw

Re: [PATCH 2/2] tracing: Use strncpy instead of memcpy for string keys in hist triggers

Hi Steve,

Quoted 31 lines · Tom Zanussi

Hi Steve,
Quoted 23 lines · Steven Rostedt
Quoted 8 lines · On Mon, 4 Mar 2019 16:50:00 -0500 Steven Rostedt <[email protected]> wrote:
Quoted 8 lines
+++ b/kernel/trace/trace_events_hist.c
kernel/trace/trace_events_hist.c +0 -1
@@ -4695,9 +4695,10 @@ static inline void add_to_key(char
 *compound_key, void *key,
 		/* ensure NULL-termination */
 		if (size > key_field->size - 1)
 			size = key_field->size - 1;
-	}
 
Quoted 11 lines
Quoted 9 lines
(quoted diff) +3 -1
-	memcpy(compound_key + key_field->offset, key, size);
+		strncpy(compound_key + key_field->offset, (char
 *)key,
 size);
+	} else
+		memcpy(compound_key + key_field->offset, key,
size);
}
Shouldn't we use strncpy() in save_comm() too. Feels safer.
Quoted 4 lines
Note, if that is changed, it can be another patch. This one is fine as
is. I just was looking at other use cases of memcpy() in that file.

Quoted 4 lines

Hmm, I don't think it's really necessary - it's not used in a key so don't care about anything after the null, and TASK_COMM_LEN is used in
the memcpy.

Never mind, yeah, it would make sense to do this, will create another patch...

Tom

Quoted 4 lines

Tom
Quoted 1 line
-- Steve

Steven Rostedt

maintainer

2019-03-04 23:45 UTC #7 lore raw

Re: [PATCH 2/2] tracing: Use strncpy instead of memcpy for string keys in hist triggers

Quoted 7 lines · On Mon, 04 Mar 2019 16:31:40 -0600 Tom Zanussi <[email protected]>…

Quoted 4 lines
Hmm, I don't think it's really necessary - it's not used in a key so don't care about anything after the null, and TASK_COMM_LEN is used in
the memcpy.
Never mind, yeah, it would make sense to do this, will create another patch...

And probably should change the memcpy() of comm in kernel/trace/trace.c too. It could be that memcpy() is a little bit faster than strncpy(), and this is done on scheduling switches when tracing is active, but still, I'm starting to think that isn't a good choice.

-- Steve

Tom Zanussi

author

2019-03-05 00:02 UTC #8 lore raw

Re: [PATCH 2/2] tracing: Use strncpy instead of memcpy for string keys in hist triggers

Quoted 14 lines · Steven Rostedt

Quoted 10 lines · On Mon, 04 Mar 2019 16:31:40 -0600 Tom Zanussi <[email protected]>…
Quoted 6 lines
Hmm, I don't think it's really necessary - it's not used in a key so
don't care about anything after the null, and TASK_COMM_LEN is used
in
the memcpy.
Never mind, yeah, it would make sense to do this, will create another
patch...

Quoted 6 lines

And probably should change the memcpy() of comm in kernel/trace/trace.c
too. It could be that memcpy() is a little bit faster than strncpy(), and this is done on scheduling switches when tracing is active, but still, I'm starting to think that isn't a good choice.

OK, will add that too.

Tom

Quoted 1 line

-- Steve

Tom Zanussi

2019-03-04 20:11 UTC #9 lore raw

In reply to Tom Zanussi

Ping...

Quoted 43 lines · Tom Zanussi

From: Tom Zanussi <[email protected]>
Hi,
Here are a couple miscellaneous hist trigger patches which can be applied independently of the onchange/snapshot patches.
The first is just an additional use of str_has_prefix() that was apparently missed in the original str_has_prefix() conversion, with an
update suggested by Joe Perches.
The second fixes a bug I noticed when testing the onchange/snapshot fixes.
I pulled these out into this separate patchset because they should be applied regardless of what happens with the onchange/snapshot patchset.
Thanks,
Tom
The following changes since commit
67748dbeaf2b1240c0b87588df4bb0fb9471a751:
sh: ftrace: Fix missing parenthesis in WARN_ON() (2019-01-08
10:19:02 -0600)
are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/zanussi/linux-
trace.git ftrace/hist-fixes-v1
Tom Zanussi (2):
tracing: Use str_has_prefix() in synth_event_create()
tracing: Use strncpy instead of memcpy for string keys in hist
triggers
 kernel/trace/trace_events_hist.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Steven Rostedt

maintainer

2019-03-04 21:26 UTC #10 lore raw

Quoted 3 lines · On Mon, 04 Mar 2019 14:11:03 -0600 Tom Zanussi <[email protected]> wrote:

Ping...

Thanks for the ping. It got buried.

-- Steve