2018-03-28 19:05:39 +00:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
|
/* Copyright (c) 2018 Facebook */
|
2023-01-15 07:16:12 +00:00
|
|
|
#include "vmlinux.h"
|
2020-01-20 13:06:49 +00:00
|
|
|
#include <bpf/bpf_helpers.h>
|
2018-03-28 19:05:39 +00:00
|
|
|
|
|
|
|
|
SEC("raw_tracepoint/task_rename")
|
|
|
|
|
int prog(struct bpf_raw_tracepoint_args *ctx)
|
|
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
samples/bpf: replace broken overhead microbenchmark with fib_table_lookup
The test_overhead bpf program is designed to compare performance
between tracepoint and kprobe. Initially it used task_rename and
urandom_read tracepoint.
However, commit 14c174633f34 ("random: remove unused tracepoints")
removed urandom_read tracepoint, and for this reason the test_overhead
got broken.
This commit introduces new microbenchmark using fib_table_lookup.
This microbenchmark sends UDP packets to localhost in order to invoke
fib_table_lookup.
In a nutshell:
fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
addr.sin_addr.s_addr = inet_addr(DUMMY_IP);
addr.sin_port = htons(DUMMY_PORT);
for() {
sendto(fd, buf, strlen(buf), 0,
(struct sockaddr *)&addr, sizeof(addr));
}
on 4 cpus in parallel:
lookup per sec
base (no tracepoints, no kprobes) 381k
with kprobe at fib_table_lookup() 325k
with tracepoint at fib:fib_table_lookup 330k
with raw_tracepoint at fib:fib_table_lookup 365k
Fixes: 14c174633f34 ("random: remove unused tracepoints")
Signed-off-by: Daniel T. Lee <danieltimlee@gmail.com>
Link: https://lore.kernel.org/r/20230115071613.125791-6-danieltimlee@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2023-01-15 07:16:08 +00:00
|
|
|
SEC("raw_tracepoint/fib_table_lookup")
|
2018-03-28 19:05:39 +00:00
|
|
|
int prog2(struct bpf_raw_tracepoint_args *ctx)
|
|
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
char _license[] SEC("license") = "GPL";
|
