Skip to contentSkip to conversation

Squelch warnings emitted by 'perf test'

PATCH Superseded Ramkumar Ramachandra 5 participants 16 messages Mar 17, 2014 – Mar 20, 2014 3 messages not in the archive

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover Squelch warnings emitted by 'perf test'

Hi,

The motivation for the series is mainly [2/5] and [5/5] -- to squelch the warnings emitted by 'perf test'. I made some effort to understand what's going on, but I didn't get very far. [5/5] feels especially like a band-aid fix: please let me know how to do it right.

Thanks.

Ramkumar Ramachandra (5):
MAINTAINERS: add tools/lib/traceevent/ to perf subsystem tools lib traceevent: handle the '->' operator tools lib traceevent: use else-if cascade, not separate ifs tools lib traceevent: field_is_long() includes unsigned long perf test: squelch warnings about undefined sizeof

 MAINTAINERS                        |  1 +
 tools/lib/traceevent/event-parse.c | 23 +++++++++++++----------
 2 files changed, 14 insertions(+), 10 deletions(-)
Signature · 1 line
1.9.0.431.g014438b
#1 lore raw

PATCH 1/5 MAINTAINERS: add tools/lib/traceevent/ to perf subsystem

Cc Steven Rostedt <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
---
 MAINTAINERS | 1 +
 1 file changed, 1 insertion(+)
MAINTAINERS +1 -0
index b3fdb0f..788effd 100644
@@ -6693,6 +6693,7 @@ F:	arch/*/kernel/*/*/perf_event*.c
 F:	arch/*/include/asm/perf_event.h
 F:	arch/*/kernel/perf_callchain.c
 F:	tools/perf/
+F:	tools/lib/traceevent/
 
 PERSONALITY HANDLING
 M:	Christoph Hellwig <[email protected]>
Signature · 1 line
1.9.0.431.g014438b
#2 lore raw

Re: [PATCH 1/5] MAINTAINERS: add tools/lib/traceevent/ to perf subsystem

Quoted 12 lines · On Mon, 17 Mar 2014 18:26:34 -0400 Ramkumar Ramachandra <[email protected]>…
Cc Steven Rostedt <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
---
 MAINTAINERS | 1 +
 1 file changed, 1 insertion(+)
MAINTAINERS +0 -0
index b3fdb0f..788effd 100644

MAINTAINERS F: arch/*/kernel/*/*/perf_event*.c · line 6696

Quoted 4 lines
MAINTAINERS +1 -0
@@ -6693,6 +6693,7 @@ F:	arch/*/kernel/*/*/perf_event*.c
 F:	arch/*/include/asm/perf_event.h
 F:	arch/*/kernel/perf_callchain.c
 F:	tools/perf/
+F:	tools/lib/traceevent/

If this goes in, I probably should get my name added as well here. Note, I'm currently working on some clean ups to the perf system on the kernel side.

-- Steve

Quoted 3 lines
MAINTAINERS +0 -0
 PERSONALITY HANDLING
 M:	Christoph Hellwig <[email protected]>
#3 lore raw

PATCH 2/5 tools lib traceevent: handle the '->' operator

In reply to Ramkumar Ramachandra

perf test emits the following warning:

  $ perf test
   5: parse events tests
    ...
    Warning: unknown op '->'
    ...

Add the operator to the checks.

Cc Steven Rostedt <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
---
 tools/lib/traceevent/event-parse.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/lib/traceevent/event-parse.c +4 -1
index 1587ea39..42bc571 100644
@@ -1759,6 +1759,8 @@ static int get_op_prio(char *op)
 			return 14;
 		} else if (strcmp(op, "||") == 0) {
 			return 15;
+		} else if (strcmp(op, "->") == 0) {
+			return 17;
 		} else {
 			do_warning("unknown op '%s'", op);
 			return -1;
@@ -1858,7 +1860,8 @@ process_op(struct event_format *event, struct print_arg *arg, char **tok)
 		   strcmp(token, "<=") == 0 ||
 		   strcmp(token, ">=") == 0 ||
 		   strcmp(token, "==") == 0 ||
-		   strcmp(token, "!=") == 0) {
+		   strcmp(token, "!=") == 0 ||
+		   strcmp(token, "->") == 0) {
 
 		left = alloc_arg();
 		if (!left)
Signature · 1 line
1.9.0.431.g014438b

A message that is not in the archive (sent off-list or not yet received): [email protected]

#5 lore raw

Re: [PATCH 2/5] tools lib traceevent: handle the '->' operator

Hi Namhyung,

Quoted 6 lines · Namhyung Kim

Hmm.. AFAIK we don't support the '->' operator so I think it's an error in the callsite. Where do you see the message though? I couldn't find it on my setup.

I'd also like to add following patch to see the location of parse failure easily.

I applied your patch and ran 'perf test'. I get:

Warning: [i915:i915_gem_evict_vm] unknown op '->'

So, where should we dig to find this callsite?

Ram

#6 lore raw

Re: [PATCH 2/5] tools lib traceevent: handle the '->' operator

Hi Ramkumar,

Quoted 7 lines · Ramkumar Ramachandra

Hi Namhyung,

Quoted 4 lines · Namhyung Kim

Hmm.. AFAIK we don't support the '->' operator so I think it's an error in the callsite. Where do you see the message though? I couldn't find it on my setup.

Quoted 8 lines
Quoted 2 lines

I'd also like to add following patch to see the location of parse failure easily.

I applied your patch and ran 'perf test'. I get:

Warning: [i915:i915_gem_evict_vm] unknown op '->'

So, where should we dig to find this callsite?

/sys/kernel/debug/tracing/events/i915/i915_gen_evict_vm/format :)

Thanks,
Namhyung

#7 lore raw

Re: [PATCH 2/5] tools lib traceevent: handle the '->' operator

Quoted 14 lines · On Tue, 18 Mar 2014 23:18:24 +0900 Namhyung Kim <[email protected]> wrote:

Hi Ramkumar,

Quoted 10 lines · Ramkumar Ramachandra

Hi Namhyung,

Quoted 6 lines · Namhyung Kim

Hmm.. AFAIK we don't support the '->' operator so I think it's an error in the callsite. Where do you see the message though? I couldn't find it on my setup.

I'd also like to add following patch to see the location of parse failure easily.

Quoted 8 lines
Quoted 5 lines

I applied your patch and ran 'perf test'. I get:

Warning: [i915:i915_gem_evict_vm] unknown op '->'

So, where should we dig to find this callsite?

/sys/kernel/debug/tracing/events/i915/i915_gen_evict_vm/format :)

Actually, that's a bug in the kernel. You should *NEVER* redirect from the ring buffer. There's no guarantee that the data still exists and this can cause a kernel oops.

-- Steve

#8 lore raw

PATCH 3/5 tools lib traceevent: use else-if cascade, not separate ifs

In reply to Ramkumar Ramachandra

When token cannot be more than one value, it seems wasteful to go through all the strcmp() calls. Use an else-if cascade instead.

Cc Steven Rostedt <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
---
 tools/lib/traceevent/event-parse.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)
tools/lib/traceevent/event-parse.c +4 -8
index 42bc571..7a8d9ae 100644
@@ -2770,21 +2770,17 @@ process_function(struct event_format *event, struct print_arg *arg,
 		free_token(token);
 		is_flag_field = 1;
 		return process_flags(event, arg, tok);
-	}
-	if (strcmp(token, "__print_symbolic") == 0) {
+	} else if (strcmp(token, "__print_symbolic") == 0) {
 		free_token(token);
 		is_symbolic_field = 1;
 		return process_symbols(event, arg, tok);
-	}
-	if (strcmp(token, "__print_hex") == 0) {
+	} else if (strcmp(token, "__print_hex") == 0) {
 		free_token(token);
 		return process_hex(event, arg, tok);
-	}
-	if (strcmp(token, "__get_str") == 0) {
+	} else if (strcmp(token, "__get_str") == 0) {
 		free_token(token);
 		return process_str(event, arg, tok);
-	}
-	if (strcmp(token, "__get_dynamic_array") == 0) {
+	} else if (strcmp(token, "__get_dynamic_array") == 0) {
 		free_token(token);
 		return process_dynamic_array(event, arg, tok);
 	}
Signature · 1 line
1.9.0.431.g014438b

A message that is not in the archive (sent off-list or not yet received): [email protected]

#10 lore raw

Re: [PATCH 3/5] tools lib traceevent: use else-if cascade, not separate ifs

Em Tue, Mar 18, 2014 at 05:02:09PM +0900, Namhyung Kim escreveu:

Quoted 6 lines · Ramkumar Ramachandra
Quoted 2 lines

When token cannot be more than one value, it seems wasteful to go through all the strcmp() calls. Use an else-if cascade instead.

I think the end result will be same since it returns from inside the block, right?

Exactly :-)

We could have it changed to something like:

	if (strcmp()) {
		ret = bla();
		goto out_free_token;
	}

.
.
.
out_free_token:

	free_token(token);
	return ret;
Quoted 43 lines

Thanks,
Namhyung

Quoted 40 lines
Cc Steven Rostedt <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
---
 tools/lib/traceevent/event-parse.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)
tools/lib/traceevent/event-parse.c +4 -8
index 42bc571..7a8d9ae 100644
@@ -2770,21 +2770,17 @@ process_function(struct event_format *event, struct print_arg *arg,
 		free_token(token);
 		is_flag_field = 1;
 		return process_flags(event, arg, tok);
-	}
-	if (strcmp(token, "__print_symbolic") == 0) {
+	} else if (strcmp(token, "__print_symbolic") == 0) {
 		free_token(token);
 		is_symbolic_field = 1;
 		return process_symbols(event, arg, tok);
-	}
-	if (strcmp(token, "__print_hex") == 0) {
+	} else if (strcmp(token, "__print_hex") == 0) {
 		free_token(token);
 		return process_hex(event, arg, tok);
-	}
-	if (strcmp(token, "__get_str") == 0) {
+	} else if (strcmp(token, "__get_str") == 0) {
 		free_token(token);
 		return process_str(event, arg, tok);
-	}
-	if (strcmp(token, "__get_dynamic_array") == 0) {
+	} else if (strcmp(token, "__get_dynamic_array") == 0) {
 		free_token(token);
 		return process_dynamic_array(event, arg, tok);
 	}
#11 lore raw

Re: [PATCH 3/5] tools lib traceevent: use else-if cascade, not separate ifs

Quoted 16 lines · On Tue, 18 Mar 2014 10:10:48 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Tue, Mar 18, 2014 at 05:02:09PM +0900, Namhyung Kim escreveu:

Quoted 6 lines · Ramkumar Ramachandra
Quoted 2 lines

When token cannot be more than one value, it seems wasteful to go through all the strcmp() calls. Use an else-if cascade instead.

I think the end result will be same since it returns from inside the block, right?

Exactly :-)

We could have it changed to something like:

	if (strcmp()) {
		ret = bla();
		goto out_free_token;
	}
Quoted 8 lines

.
.
.
out_free_token:

	free_token(token);
	return ret;

Honestly, I prefer the original, thus I would add a NAK to this patch.

-- Steve

#12 lore raw

Re: [PATCH 3/5] tools lib traceevent: use else-if cascade, not separate ifs

Em Tue, Mar 18, 2014 at 09:30:06AM -0400, Steven Rostedt escreveu:

Quoted 24 lines · On Tue, 18 Mar 2014 10:10:48 -0300 Arnaldo Carvalho de Melo <[email protected]>…
Quoted 16 lines

Em Tue, Mar 18, 2014 at 05:02:09PM +0900, Namhyung Kim escreveu:

Quoted 6 lines · Ramkumar Ramachandra
Quoted 2 lines

When token cannot be more than one value, it seems wasteful to go through all the strcmp() calls. Use an else-if cascade instead.

I think the end result will be same since it returns from inside the block, right?

Exactly :-)

We could have it changed to something like:

	if (strcmp()) {
		ret = bla();
		goto out_free_token;
	}
Quoted 6 lines

.
.
.
out_free_token:

	free_token(token);
	return ret;
Quoted 1 line

Honestly, I prefer the original, thus I would add a NAK to this patch.

No problem with that, if I had written that code, I'd prefer the gotos, but that is kinda personal preference.

Yeah, I'd nack that patch as well, changes nothing.

- Arnaldo

#13 lore raw

PATCH 4/5 tools lib traceevent: field_is_long() includes unsigned long

In reply to Ramkumar Ramachandra

Note in the comment that field_is_long() also checks for "unsigned long" and "unsigned long long" in addition to "long long".

Cc Steven Rostedt <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
---
 tools/lib/traceevent/event-parse.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/lib/traceevent/event-parse.c +1 -1
index 7a8d9ae..567d9ba 100644
@@ -1233,7 +1233,7 @@ static int field_is_dynamic(struct format_field *field)
 
 static int field_is_long(struct format_field *field)
 {
-	/* includes long long */
+	/* includes long long, unsigned long, and unsigned long long */
 	if (strstr(field->type, "long"))
 		return 1;
 
Signature · 1 line
1.9.0.431.g014438b
#14 lore raw

PATCH 5/5 perf test: squelch warnings about undefined sizeof

In reply to Ramkumar Ramachandra

perf test emits the following warnings on the parse events test:

  $ perf test
   5: parse events tests
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    ...

Squelch the warnings by explicitly ignoring the sizeof function.

Cc Steven Rostedt <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>

---
Probably incorrect.

 tools/lib/traceevent/event-parse.c | 4 ++++
 1 file changed, 4 insertions(+)
tools/lib/traceevent/event-parse.c +4 -0
index 567d9ba..06c269a 100644
@@ -2783,6 +2783,10 @@ process_function(struct event_format *event, struct print_arg *arg,
 	} else if (strcmp(token, "__get_dynamic_array") == 0) {
 		free_token(token);
 		return process_dynamic_array(event, arg, tok);
+	} else if (strcmp(token, "sizeof") == 0) {
+		/* ignore sizeof function */
+		free_token(token);
+		return 0;
 	}
 
 	func = find_func_handler(event->pevent, token);
Signature · 1 line
1.9.0.431.g014438b

A message that is not in the archive (sent off-list or not yet received): [email protected]

#16 lore raw

Re: [PATCH 5/5] perf test: squelch warnings about undefined sizeof

Quoted 15 lines · Namhyung Kim
Quoted 14 lines · Ramkumar Ramachandra

perf test emits the following warnings on the parse events test:

  $ perf test
   5: parse events tests
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    ...
Quoted 5 lines
Quoted 1 line

Squelch the warnings by explicitly ignoring the sizeof function.

It just hides the warning leaving the real problem untouched. If you really don't want to see those, I guess installing proper plugin for the failing events will help you (in case you didn't).

it wont' help, sizeof is special.. AFAIK we did not figure out a way to handle that so far

but I think it was used only in one trace subsystem.. maybe we could replace it and forbid to use it in future ;-)

jirka

#17 lore raw

Re: [PATCH 5/5] perf test: squelch warnings about undefined sizeof

Hi Jiri,

Quoted 22 lines · Jiri Olsa
Quoted 15 lines · Namhyung Kim
Quoted 14 lines · Ramkumar Ramachandra

perf test emits the following warnings on the parse events test:

  $ perf test
   5: parse events tests
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    Warning: function sizeof not defined
    ...
Quoted 5 lines
Quoted 1 line

Squelch the warnings by explicitly ignoring the sizeof function.

It just hides the warning leaving the real problem untouched. If you really don't want to see those, I guess installing proper plugin for the failing events will help you (in case you didn't).

Quoted 5 lines

it wont' help, sizeof is special.. AFAIK we did not figure out a way to handle that so far

but I think it was used only in one trace subsystem.. maybe we could replace it and forbid to use it in future ;-)

Yep, I know. But I meant that if we have a handler for the failing event, it might print the event without even using the sizeof somehow. ;-)

Thanks,
Namhyung

#18 lore raw

Re: [PATCH 5/5] perf test: squelch warnings about undefined sizeof

In reply to Jiri Olsa
Quoted 11 lines · Jiri Olsa
Quoted 5 lines
Quoted 1 line

Squelch the warnings by explicitly ignoring the sizeof function.

It just hides the warning leaving the real problem untouched. If you really don't want to see those, I guess installing proper plugin for the failing events will help you (in case you didn't).

it wont' help, sizeof is special.. AFAIK we did not figure out a way to handle that so far

but I think it was used only in one trace subsystem.. maybe we could replace it and forbid to use it in future ;-)

It only seems to be used by xen:

Warning: [xen:xen_mmu_set_pte] function sizeof not defined
Warning: [xen:xen_mmu_set_pte_atomic] function sizeof not defined
Warning: [xen:xen_mmu_set_domain_pte] function sizeof not defined
Warning: [xen:xen_mmu_set_pte_at] function sizeof not defined
Warning: [xen:xen_mmu_set_pmd] function sizeof not defined
Warning: [xen:xen_mmu_set_pud] function sizeof not defined
Warning: [xen:xen_mmu_set_pgd] function sizeof not defined
Warning: [xen:xen_mmu_ptep_modify_prot_start] function sizeof not defined
Warning: [xen:xen_mmu_ptep_modify_prot_commit] function sizeof not defined

These events are defined in include/trace/events/xen.h. Any hints on how to proceed?