Skip to contentSkip to conversation

perf tools: Fix is_bpf_image function logic

PATCH In review Jiri Olsa 3 participants 5 messages May 12, 2020 perf events

lore
Share & tools
lore

Jiri Olsa

author reviewer
#0 lore raw

PATCH 1/1 perf tools: Fix is_bpf_image function logic

Adrian reported that is_bpf_image is not working the way it was intended - passing on trampolines and dispatcher names. Instead it returned true for all the bpf names.

The reason even this logic worked properly is that all bpf objects, even trampolines and dispatcher, were assigned DSO_BINARY_TYPE__BPF_IMAGE binary_type.

The later for bpf_prog objects, the binary_type was fixed in bpf load event processing, which is executed after the ksymbol code.

Fixing the is_bpf_image logic, so it properly recognizes trampoline and dispatcher objects.

---
 tools/perf/util/machine.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/machine.c +2 -2
index 8ed2135893bb..d5384807372b 100644
@@ -738,8 +738,8 @@ int machine__process_switch_event(struct machine *machine __maybe_unused,
 
 static int is_bpf_image(const char *name)
 {
-	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) ||
-	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1);
+	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) == 0 ||
+	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1) == 0;
 }
 
 static int machine__process_ksymbol_register(struct machine *machine,
Signature · 1 line
2.25.4
#1 lore raw

Em Tue, May 12, 2020 at 02:23:10PM +0200, Jiri Olsa escreveu:

Quoted 8 lines

Adrian reported that is_bpf_image is not working the way it was intended - passing on trampolines and dispatcher names. Instead it returned true for all the bpf names.

The reason even this logic worked properly is that all bpf objects, even trampolines and dispatcher, were assigned DSO_BINARY_TYPE__BPF_IMAGE binary_type.

Quoted 5 lines

The later for bpf_prog objects, the binary_type was fixed in bpf load event processing, which is executed after the ksymbol code.

Fixing the is_bpf_image logic, so it properly recognizes trampoline and dispatcher objects.

This is not applying on top of torvalds/master, not tip/perf/urgent, and you forgot to add the Fixes: line, lemme try to find this...

- Arnaldo

Quoted 24 lines
---
 tools/perf/util/machine.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/machine.c +2 -2
index 8ed2135893bb..d5384807372b 100644
@@ -738,8 +738,8 @@ int machine__process_switch_event(struct machine *machine __maybe_unused,
 
 static int is_bpf_image(const char *name)
 {
-	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) ||
-	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1);
+	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) == 0 ||
+	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1) == 0;
 }
 
 static int machine__process_ksymbol_register(struct machine *machine,
Signature · 1 line
2.25.4
Signature · 1 line
- Arnaldo
#2 lore raw
Quoted 9 lines · Arnaldo Carvalho de Melo

Em Tue, May 12, 2020 at 02:23:10PM +0200, Jiri Olsa escreveu:

Quoted 8 lines

Adrian reported that is_bpf_image is not working the way it was intended - passing on trampolines and dispatcher names. Instead it returned true for all the bpf names.

The reason even this logic worked properly is that all bpf objects, even trampolines and dispatcher, were assigned DSO_BINARY_TYPE__BPF_IMAGE binary_type.

Quoted 7 lines
Quoted 5 lines

The later for bpf_prog objects, the binary_type was fixed in bpf load event processing, which is executed after the ksymbol code.

Fixing the is_bpf_image logic, so it properly recognizes trampoline and dispatcher objects.

This is not applying on top of torvalds/master, not tip/perf/urgent, and

right.. it's on top of your's perf/core.. I can rebase on perf/urgent

Quoted 1 line

you forgot to add the Fixes: line, lemme try to find this...

oops, sorry

Fixes 3c29d4483e85 ("perf annotate: Add basic support for bpf_image")

jirka

Quoted 32 lines

- Arnaldo

Quoted 24 lines
---
 tools/perf/util/machine.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/machine.c +2 -2
index 8ed2135893bb..d5384807372b 100644
@@ -738,8 +738,8 @@ int machine__process_switch_event(struct machine *machine __maybe_unused,
 
 static int is_bpf_image(const char *name)
 {
-	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) ||
-	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1);
+	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) == 0 ||
+	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1) == 0;
 }
 
 static int machine__process_ksymbol_register(struct machine *machine,
Signature · 1 line
2.25.4
Signature · 1 line
- Arnaldo
#3 lore raw

Em Tue, May 12, 2020 at 03:36:09PM +0200, Jiri Olsa escreveu:

Quoted 13 lines · Arnaldo Carvalho de Melo
Quoted 12 lines

Em Tue, May 12, 2020 at 02:23:10PM +0200, Jiri Olsa escreveu:

Quoted 11 lines

Adrian reported that is_bpf_image is not working the way it was intended - passing on trampolines and dispatcher names. Instead it returned true for all the bpf names.

The reason even this logic worked properly is that all bpf objects, even trampolines and dispatcher, were assigned DSO_BINARY_TYPE__BPF_IMAGE binary_type.

The later for bpf_prog objects, the binary_type was fixed in bpf load event processing, which is executed after the ksymbol code.

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Fixing the is_bpf_image logic, so it properly recognizes trampoline and dispatcher objects.

This is not applying on top of torvalds/master, not tip/perf/urgent, and

right.. it's on top of your's perf/core.. I can rebase on perf/urgent

You don't need to, this hasn't hit torvalds/master, it'll be in the next merge window, the one for 5.8.

Quoted 5 lines
Quoted 1 line

you forgot to add the Fixes: line, lemme try to find this...

oops, sorry

Fixes 3c29d4483e85 ("perf annotate: Add basic support for bpf_image")

I did it already, and:

[acme@five perf]$ git tag --contains 3c29d4483e85
perf-core-for-mingo-5.8-20200420
perf-core-for-mingo-5.8-20200506
perf-for-bpf-2020-05-06
[acme@five perf]$

So can't go to perf/urgent at the moment.

- Arnaldo

Quoted 35 lines

jirka

Quoted 32 lines

- Arnaldo

Quoted 24 lines
---
 tools/perf/util/machine.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/machine.c +2 -2
index 8ed2135893bb..d5384807372b 100644
@@ -738,8 +738,8 @@ int machine__process_switch_event(struct machine *machine __maybe_unused,
 
 static int is_bpf_image(const char *name)
 {
-	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) ||
-	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1);
+	return strncmp(name, "bpf_trampoline_", sizeof("bpf_trampoline_") - 1) == 0 ||
+	       strncmp(name, "bpf_dispatcher_", sizeof("bpf_dispatcher_") - 1) == 0;
 }
 
 static int machine__process_ksymbol_register(struct machine *machine,
Signature · 1 line
2.25.4
Signature · 1 line
- Arnaldo
Signature · 1 line
- Arnaldo
#4 lore raw
Quoted 17 lines · Arnaldo Carvalho de Melo

Em Tue, May 12, 2020 at 03:36:09PM +0200, Jiri Olsa escreveu:

Quoted 16 lines · Arnaldo Carvalho de Melo
Quoted 15 lines

Em Tue, May 12, 2020 at 02:23:10PM +0200, Jiri Olsa escreveu:

Quoted 13 lines

Adrian reported that is_bpf_image is not working the way it was intended - passing on trampolines and dispatcher names. Instead it returned true for all the bpf names.

The reason even this logic worked properly is that all bpf objects, even trampolines and dispatcher, were assigned DSO_BINARY_TYPE__BPF_IMAGE binary_type.

The later for bpf_prog objects, the binary_type was fixed in bpf load event processing, which is executed after the ksymbol code.

Fixing the is_bpf_image logic, so it properly recognizes trampoline and dispatcher objects.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

This is not applying on top of torvalds/master, not tip/perf/urgent, and

right.. it's on top of your's perf/core.. I can rebase on perf/urgent

You don't need to, this hasn't hit torvalds/master, it'll be in the next merge window, the one for 5.8.

perfect, thanks

jirka