libbpf-tools: add futexctn

Signed-off-by: Wenbo Zhang <[email protected]>

libbpf-tools/.gitignore

@@ -26,6 +26,7 @@
 /fsdist
 /fsslower
 /funclatency
+/futexctn
 /gethostlatency
 /hardirqs
 /javagc

libbpf-tools/Makefile

@@ -32,6 +32,7 @@ $(error Architecture $(ARCH) is not supported yet. Please open an issue)
 endif
 
 BZ_APPS = \
+ futexctn \
  memleak \
  opensnoop \
  #

libbpf-tools/futexctn.bpf.c

@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2023 Wenbo Zhang */
+#include <vmlinux.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_core_read.h>
+#include <bpf/bpf_tracing.h>
+
+#include "futexctn.h"
+#include "bits.bpf.h"
+#include "maps.bpf.h"
+
+#define MAX_ENTRIES 10240
+
+#define FUTEX_WAIT 0
+#define FUTEX_PRIVATE_FLAG 128
+#define FUTEX_CLOCK_REALTIME 256
+#define FUTEX_CMD_MASK ~(FUTEX_PRIVATE_FLAG | FUTEX_CLOCK_REALTIME)
+
+const volatile bool targ_summary = false;
+const volatile bool targ_ms = false;
+const volatile __u64 targ_lock = 0;
+const volatile pid_t targ_pid = 0;
+const volatile pid_t targ_tid = 0;
+
+struct val_t {
+ u64 ts;
+ u64 uaddr;
+};
+
+struct {
+ __uint(type, BPF_MAP_TYPE_HASH);
+ __uint(max_entries, MAX_ENTRIES);
+ __type(key, u64);
+ __type(value, struct val_t);
+} start SEC(".maps");
+
+struct {
+ __uint(type, BPF_MAP_TYPE_STACK_TRACE);
+ __uint(key_size, sizeof(u32));
+} stackmap SEC(".maps");
+
+struct {
+ __uint(type, BPF_MAP_TYPE_HASH);
+ __uint(max_entries, MAX_ENTRIES);
+ __type(key, struct hist_key);
+ __type(value, struct hist);
+} hists SEC(".maps");
+
+static struct hist initial_hist = {};
+
+SEC("tracepoint/syscalls/sys_enter_futex")
+int futex_enter(struct trace_event_raw_sys_enter *ctx)
+{
+ struct val_t v = {};
+ u64 pid_tgid;
+ u32 tid;
+
+ if (((int)ctx->args[1] & FUTEX_CMD_MASK) != FUTEX_WAIT)
+ return 0;
+
+ pid_tgid = bpf_get_current_pid_tgid();
+ tid = (__u32)pid_tgid;
+ if (targ_pid && targ_pid != pid_tgid >> 32)
+ return 0;
+ if (targ_tid && targ_tid != tid)
+ return 0;
+ v.uaddr = ctx->args[0];
+ if (targ_lock && targ_lock != v.uaddr)
+ return 0;
+ v.ts = bpf_ktime_get_ns();
+ bpf_map_update_elem(&start, &pid_tgid, &v, BPF_ANY);
+ return 0;
+}
+
+SEC("tracepoint/syscalls/sys_exit_futex")
+int futex_exit(struct trace_event_raw_sys_exit *ctx)
+{
+ u64 pid_tgid, slot, ts, min, max;
+ struct hist_key hkey = {};
+ struct hist *histp;
+ struct val_t *vp;
+ s64 delta;
+
+ ts = bpf_ktime_get_ns();
+ pid_tgid = bpf_get_current_pid_tgid();
+ vp = bpf_map_lookup_elem(&start, &pid_tgid);
+ if (!vp)
+ return 0;
+ if ((int)ctx->ret < 0)
+ goto cleanup;
+
+ delta = (s64)(ts - vp->ts);
+ if (delta < 0)
+ goto cleanup;
+
+ hkey.pid_tgid = pid_tgid;
+ hkey.uaddr = vp->uaddr;
+ if (!targ_summary)
+ hkey.user_stack_id = bpf_get_stackid(ctx, &stackmap, BPF_F_USER_STACK);
+ else
+ hkey.pid_tgid >>= 32;
+
+ histp = bpf_map_lookup_or_try_init(&hists, &hkey, &initial_hist);
+ if (!histp)
+ goto cleanup;
+
+ if (targ_ms)
+ delta /= 1000000U;
+ else
+ delta /= 1000U;
+
+ slot = log2l(delta);
+ if (slot >= MAX_SLOTS)
+ slot = MAX_SLOTS - 1;
+ __sync_fetch_and_add(&histp->slots[slot], 1);
+ __sync_fetch_and_add(&histp->contended, 1);
+ __sync_fetch_and_add(&histp->total_elapsed, delta);
+ min = __sync_fetch_and_add(&histp->min, 0);
+ if (!min || min > delta)
+ __sync_val_compare_and_swap(&histp->min, min, delta);
+ max = __sync_fetch_and_add(&histp->max, 0);
+ if (max < delta)
+ __sync_val_compare_and_swap(&histp->max, max, delta);
+ bpf_get_current_comm(&histp->comm, sizeof(histp->comm));
+
+cleanup:
+ bpf_map_delete_elem(&start, &pid_tgid);
+ return 0;
+}
+
+char LICENSE[] SEC("license") = "GPL";