Skip to contentSkip to conversation

perf tools: Get a perf cgroup more portably in BPF

PATCH Superseded Namhyung Kim 2 participants 3 messages Sep 22, 2022 – Sep 23, 2022

Subsystems (2)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

PATCH v3 1/1 perf tools: Get a perf cgroup more portably in BPF

The perf_event_cgrp_id can be different on other configurations. To be more portable as CO-RE, it needs to get the cgroup subsys id using the bpf_core_enum_value() helper.

---
v3 changes)
* check compiler features for enum value

v2 changes)
* fix off_cpu.bpf.c too
* get perf_subsys_id only once

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 11 ++++++++++-
 tools/perf/util/bpf_skel/off_cpu.bpf.c      | 12 ++++++++----
 2 files changed, 18 insertions(+), 5 deletions(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +10 -1
index 292c430768b5..8e7520e273db 100644
@@ -48,6 +48,7 @@ const volatile __u32 num_cpus = 1;
 
 int enabled = 0;
 int use_cgroup_v2 = 0;
+int perf_subsys_id = -1;
 
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
@@ -58,7 +59,15 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 	int level;
 	int cnt;
 
-	cgrp = BPF_CORE_READ(p, cgroups, subsys[perf_event_cgrp_id], cgroup);
+	if (perf_subsys_id == -1) {
+#if __has_builtin(__builtin_preserve_enum_value)
+		perf_subsys_id = bpf_core_enum_value(enum cgroup_subsys_id,
+						     perf_event_cgrp_id);
+#else
+		perf_subsys_id = perf_event_cgrp_id;
+#endif
+	}
+	cgrp = BPF_CORE_READ(p, cgroups, subsys[perf_subsys_id], cgroup);
 	level = BPF_CORE_READ(cgrp, level);
 
 	for (cnt = 0; i < MAX_LEVELS; i++) {
tools/perf/util/bpf_skel/off_cpu.bpf.c +8 -4
index c4ba2bcf179f..e917ef7b8875 100644
@@ -94,6 +94,8 @@ const volatile bool has_prev_state = false;
 const volatile bool needs_cgroup = false;
 const volatile bool uses_cgroup_v1 = false;
 
+int perf_subsys_id = -1;
+
 /*
  * Old kernel used to call it task_struct->state and now it's '__state'.
  * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
@@ -119,11 +121,13 @@ static inline __u64 get_cgroup_id(struct task_struct *t)
 {
 	struct cgroup *cgrp;
 
-	if (uses_cgroup_v1)
-		cgrp = BPF_CORE_READ(t, cgroups, subsys[perf_event_cgrp_id], cgroup);
-	else
-		cgrp = BPF_CORE_READ(t, cgroups, dfl_cgrp);
+	if (!uses_cgroup_v1)
+		return BPF_CORE_READ(t, cgroups, dfl_cgrp, kn, id);
+
+	if (perf_subsys_id == -1)
+		perf_subsys_id = bpf_core_enum_value(enum cgroup_subsys_id, perf_event_cgrp_id);
 
+	cgrp = BPF_CORE_READ(t, cgroups, subsys[perf_subsys_id], cgroup);
 	return BPF_CORE_READ(cgrp, kn, id);
 }
 
Signature · 1 line
2.37.3.998.g577e59143f-goog
reviewer
#1 lore raw
Quoted 64 lines · Namhyung Kim

The perf_event_cgrp_id can be different on other configurations. To be more portable as CO-RE, it needs to get the cgroup subsys id using the bpf_core_enum_value() helper.

---
v3 changes)
* check compiler features for enum value

v2 changes)
* fix off_cpu.bpf.c too
* get perf_subsys_id only once

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 11 ++++++++++-
 tools/perf/util/bpf_skel/off_cpu.bpf.c      | 12 ++++++++----
 2 files changed, 18 insertions(+), 5 deletions(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +10 -1
index 292c430768b5..8e7520e273db 100644
@@ -48,6 +48,7 @@ const volatile __u32 num_cpus = 1;
 
 int enabled = 0;
 int use_cgroup_v2 = 0;
+int perf_subsys_id = -1;
 
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
@@ -58,7 +59,15 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
        int level;
        int cnt;
 
-       cgrp = BPF_CORE_READ(p, cgroups, subsys[perf_event_cgrp_id], cgroup);
+       if (perf_subsys_id == -1) {
+#if __has_builtin(__builtin_preserve_enum_value)
+               perf_subsys_id = bpf_core_enum_value(enum cgroup_subsys_id,
+                                                    perf_event_cgrp_id);
+#else
+               perf_subsys_id = perf_event_cgrp_id;
+#endif
+       }
+       cgrp = BPF_CORE_READ(p, cgroups, subsys[perf_subsys_id], cgroup);
        level = BPF_CORE_READ(cgrp, level);
 
        for (cnt = 0; i < MAX_LEVELS; i++) {
tools/perf/util/bpf_skel/off_cpu.bpf.c +2 -0
index c4ba2bcf179f..e917ef7b8875 100644
@@ -94,6 +94,8 @@ const volatile bool has_prev_state = false;
 const volatile bool needs_cgroup = false;
 const volatile bool uses_cgroup_v1 = false;
 
+int perf_subsys_id = -1;
+
 /*
  * Old kernel used to call it task_struct->state and now it's '__state'.
  * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
@@ -119,11 +121,13 @@ static inline __u64 get_cgroup_id(struct task_struct *t)
 {
        struct cgroup *cgrp;

tools/perf/util/bpf_skel/off_cpu.bpf.c static inline __u64 get_cgroup_id(struct task_struct *t) · line 128

Quoted 10 lines
tools/perf/util/bpf_skel/off_cpu.bpf.c +5 -4
 
-       if (uses_cgroup_v1)
-               cgrp = BPF_CORE_READ(t, cgroups, subsys[perf_event_cgrp_id], cgroup);
-       else
-               cgrp = BPF_CORE_READ(t, cgroups, dfl_cgrp);
+       if (!uses_cgroup_v1)
+               return BPF_CORE_READ(t, cgroups, dfl_cgrp, kn, id);
+
+       if (perf_subsys_id == -1)
+               perf_subsys_id = bpf_core_enum_value(enum cgroup_subsys_id, perf_event_cgrp_id);

Should the "#if __has_builtin(__builtin_preserve_enum_value)" test also be here?

It feels a shame that bpf_core_enum_value isn't defined something like:

#if __has_builtin(__builtin_preserve_enum_value) #define bpf_core_enum_value(enum_type, enum_value) \
__builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE)
#else
#define bpf_core_enum_value(enum_type, enum_value) enum_value #endif

for backward clang compatibility, but I could see why an error would be preferable.

Thanks,
Ian

Quoted 8 lines
tools/perf/util/bpf_skel/off_cpu.bpf.c +1 -1
+       cgrp = BPF_CORE_READ(t, cgroups, subsys[perf_subsys_id], cgroup);
        return BPF_CORE_READ(cgrp, kn, id);
 }
 
--

2.37.3.998.g577e59143f-goog

author maintainer
#2 lore raw
Quoted 67 lines · Ian Rogers
Quoted 65 lines · Namhyung Kim

The perf_event_cgrp_id can be different on other configurations. To be more portable as CO-RE, it needs to get the cgroup subsys id using the bpf_core_enum_value() helper.

---
v3 changes)
* check compiler features for enum value

v2 changes)
* fix off_cpu.bpf.c too
* get perf_subsys_id only once

 tools/perf/util/bpf_skel/bperf_cgroup.bpf.c | 11 ++++++++++-
 tools/perf/util/bpf_skel/off_cpu.bpf.c      | 12 ++++++++----
 2 files changed, 18 insertions(+), 5 deletions(-)
tools/perf/util/bpf_skel/bperf_cgroup.bpf.c +10 -1
index 292c430768b5..8e7520e273db 100644
@@ -48,6 +48,7 @@ const volatile __u32 num_cpus = 1;
 
 int enabled = 0;
 int use_cgroup_v2 = 0;
+int perf_subsys_id = -1;
 
 static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
 {
@@ -58,7 +59,15 @@ static inline int get_cgroup_v1_idx(__u32 *cgrps, int size)
        int level;
        int cnt;
 
-       cgrp = BPF_CORE_READ(p, cgroups, subsys[perf_event_cgrp_id], cgroup);
+       if (perf_subsys_id == -1) {
+#if __has_builtin(__builtin_preserve_enum_value)
+               perf_subsys_id = bpf_core_enum_value(enum cgroup_subsys_id,
+                                                    perf_event_cgrp_id);
+#else
+               perf_subsys_id = perf_event_cgrp_id;
+#endif
+       }
+       cgrp = BPF_CORE_READ(p, cgroups, subsys[perf_subsys_id], cgroup);
        level = BPF_CORE_READ(cgrp, level);
 
        for (cnt = 0; i < MAX_LEVELS; i++) {
tools/perf/util/bpf_skel/off_cpu.bpf.c +2 -0
index c4ba2bcf179f..e917ef7b8875 100644
@@ -94,6 +94,8 @@ const volatile bool has_prev_state = false;
 const volatile bool needs_cgroup = false;
 const volatile bool uses_cgroup_v1 = false;
 
+int perf_subsys_id = -1;
+
 /*
  * Old kernel used to call it task_struct->state and now it's '__state'.
  * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
@@ -119,11 +121,13 @@ static inline __u64 get_cgroup_id(struct task_struct *t)
 {
        struct cgroup *cgrp;
 
Quoted 11 lines

tools/perf/util/bpf_skel/off_cpu.bpf.c static inline __u64 get_cgroup_id(struct task_struct *t) · line 128

Quoted 9 lines
tools/perf/util/bpf_skel/off_cpu.bpf.c +5 -4
-       if (uses_cgroup_v1)
-               cgrp = BPF_CORE_READ(t, cgroups, subsys[perf_event_cgrp_id], cgroup);
-       else
-               cgrp = BPF_CORE_READ(t, cgroups, dfl_cgrp);
+       if (!uses_cgroup_v1)
+               return BPF_CORE_READ(t, cgroups, dfl_cgrp, kn, id);
+
+       if (perf_subsys_id == -1)
+               perf_subsys_id = bpf_core_enum_value(enum cgroup_subsys_id, perf_event_cgrp_id);

Should the "#if __has_builtin(__builtin_preserve_enum_value)" test also be here?

Ouch! Will send v4.

Quoted 10 lines

It feels a shame that bpf_core_enum_value isn't defined something like:

#if __has_builtin(__builtin_preserve_enum_value) #define bpf_core_enum_value(enum_type, enum_value) \
__builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE)
#else
#define bpf_core_enum_value(enum_type, enum_value) enum_value #endif

Yeah, that would be much easier to use.

Thanks,
Namhyung

Quoted 15 lines

for backward clang compatibility, but I could see why an error would be preferable.

Thanks,
Ian

Quoted 8 lines
tools/perf/util/bpf_skel/off_cpu.bpf.c +1 -1
+       cgrp = BPF_CORE_READ(t, cgroups, subsys[perf_subsys_id], cgroup);
        return BPF_CORE_READ(cgrp, kn, id);
 }
 
--

2.37.3.998.g577e59143f-goog