samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0
|
|
|
|
* Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
|
|
|
|
*/
|
|
|
|
static const char *__doc__ = " XDP RX-queue info extract example\n\n"
|
|
|
|
"Monitor how many packets per sec (pps) are received\n"
|
|
|
|
"per NIC RX queue index and which CPU processed the packet\n"
|
|
|
|
;
|
|
|
|
|
|
|
|
#include <errno.h>
|
|
|
|
#include <signal.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <stdbool.h>
|
|
|
|
#include <string.h>
|
|
|
|
#include <unistd.h>
|
|
|
|
#include <locale.h>
|
|
|
|
#include <sys/resource.h>
|
|
|
|
#include <getopt.h>
|
|
|
|
#include <net/if.h>
|
|
|
|
#include <time.h>
|
|
|
|
|
|
|
|
#include <arpa/inet.h>
|
|
|
|
#include <linux/if_link.h>
|
|
|
|
|
2019-06-15 15:14:47 +00:00
|
|
|
#include "bpf.h"
|
|
|
|
#include "libbpf.h"
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
#include "bpf_util.h"
|
|
|
|
|
|
|
|
static int ifindex = -1;
|
|
|
|
static char ifname_buf[IF_NAMESIZE];
|
|
|
|
static char *ifname;
|
2019-02-01 21:42:30 +00:00
|
|
|
static __u32 prog_id;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
2019-02-01 21:42:28 +00:00
|
|
|
static __u32 xdp_flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
2018-05-10 17:24:43 +00:00
|
|
|
static struct bpf_map *stats_global_map;
|
|
|
|
static struct bpf_map *rx_queue_index_map;
|
|
|
|
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
/* Exit return codes */
|
|
|
|
#define EXIT_OK 0
|
|
|
|
#define EXIT_FAIL 1
|
|
|
|
#define EXIT_FAIL_OPTION 2
|
|
|
|
#define EXIT_FAIL_XDP 3
|
|
|
|
#define EXIT_FAIL_BPF 4
|
|
|
|
#define EXIT_FAIL_MEM 5
|
|
|
|
|
|
|
|
static const struct option long_options[] = {
|
|
|
|
{"help", no_argument, NULL, 'h' },
|
|
|
|
{"dev", required_argument, NULL, 'd' },
|
|
|
|
{"skb-mode", no_argument, NULL, 'S' },
|
|
|
|
{"sec", required_argument, NULL, 's' },
|
|
|
|
{"no-separators", no_argument, NULL, 'z' },
|
|
|
|
{"action", required_argument, NULL, 'a' },
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
{"readmem", no_argument, NULL, 'r' },
|
2018-06-25 14:27:48 +00:00
|
|
|
{"swapmac", no_argument, NULL, 'm' },
|
2019-02-01 21:42:28 +00:00
|
|
|
{"force", no_argument, NULL, 'F' },
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
{0, 0, NULL, 0 }
|
|
|
|
};
|
|
|
|
|
|
|
|
static void int_exit(int sig)
|
|
|
|
{
|
2019-02-01 21:42:30 +00:00
|
|
|
__u32 curr_prog_id = 0;
|
|
|
|
|
|
|
|
if (ifindex > -1) {
|
|
|
|
if (bpf_get_link_xdp_id(ifindex, &curr_prog_id, xdp_flags)) {
|
|
|
|
printf("bpf_get_link_xdp_id failed\n");
|
|
|
|
exit(EXIT_FAIL);
|
|
|
|
}
|
|
|
|
if (prog_id == curr_prog_id) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"Interrupted: Removing XDP program on ifindex:%d device:%s\n",
|
|
|
|
ifindex, ifname);
|
|
|
|
bpf_set_link_xdp_fd(ifindex, -1, xdp_flags);
|
|
|
|
} else if (!curr_prog_id) {
|
|
|
|
printf("couldn't find a prog id on a given iface\n");
|
|
|
|
} else {
|
|
|
|
printf("program on interface changed, not removing\n");
|
|
|
|
}
|
|
|
|
}
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
exit(EXIT_OK);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct config {
|
|
|
|
__u32 action;
|
|
|
|
int ifindex;
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
__u32 options;
|
|
|
|
};
|
|
|
|
enum cfg_options_flags {
|
|
|
|
NO_TOUCH = 0x0U,
|
|
|
|
READ_MEM = 0x1U,
|
2018-06-25 14:27:48 +00:00
|
|
|
SWAP_MAC = 0x2U,
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
};
|
|
|
|
#define XDP_ACTION_MAX (XDP_TX + 1)
|
|
|
|
#define XDP_ACTION_MAX_STRLEN 11
|
|
|
|
static const char *xdp_action_names[XDP_ACTION_MAX] = {
|
|
|
|
[XDP_ABORTED] = "XDP_ABORTED",
|
|
|
|
[XDP_DROP] = "XDP_DROP",
|
|
|
|
[XDP_PASS] = "XDP_PASS",
|
|
|
|
[XDP_TX] = "XDP_TX",
|
|
|
|
};
|
|
|
|
|
|
|
|
static const char *action2str(int action)
|
|
|
|
{
|
|
|
|
if (action < XDP_ACTION_MAX)
|
|
|
|
return xdp_action_names[action];
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int parse_xdp_action(char *action_str)
|
|
|
|
{
|
|
|
|
size_t maxlen;
|
|
|
|
__u64 action = -1;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < XDP_ACTION_MAX; i++) {
|
|
|
|
maxlen = XDP_ACTION_MAX_STRLEN;
|
|
|
|
if (strncmp(xdp_action_names[i], action_str, maxlen) == 0) {
|
|
|
|
action = i;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return action;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void list_xdp_actions(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
printf("Available XDP --action <options>\n");
|
|
|
|
for (i = 0; i < XDP_ACTION_MAX; i++)
|
|
|
|
printf("\t%s\n", xdp_action_names[i]);
|
|
|
|
printf("\n");
|
|
|
|
}
|
|
|
|
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
static char* options2str(enum cfg_options_flags flag)
|
|
|
|
{
|
|
|
|
if (flag == NO_TOUCH)
|
|
|
|
return "no_touch";
|
2018-06-25 14:27:48 +00:00
|
|
|
if (flag & SWAP_MAC)
|
|
|
|
return "swapmac";
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
if (flag & READ_MEM)
|
|
|
|
return "read";
|
|
|
|
fprintf(stderr, "ERR: Unknown config option flags");
|
|
|
|
exit(EXIT_FAIL);
|
|
|
|
}
|
|
|
|
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
static void usage(char *argv[])
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
printf("\nDOCUMENTATION:\n%s\n", __doc__);
|
|
|
|
printf(" Usage: %s (options-see-below)\n", argv[0]);
|
|
|
|
printf(" Listing options:\n");
|
|
|
|
for (i = 0; long_options[i].name != 0; i++) {
|
|
|
|
printf(" --%-12s", long_options[i].name);
|
|
|
|
if (long_options[i].flag != NULL)
|
|
|
|
printf(" flag (internal value:%d)",
|
|
|
|
*long_options[i].flag);
|
|
|
|
else
|
|
|
|
printf(" short-option: -%c",
|
|
|
|
long_options[i].val);
|
|
|
|
printf("\n");
|
|
|
|
}
|
|
|
|
printf("\n");
|
|
|
|
list_xdp_actions();
|
|
|
|
}
|
|
|
|
|
|
|
|
#define NANOSEC_PER_SEC 1000000000 /* 10^9 */
|
|
|
|
static __u64 gettime(void)
|
|
|
|
{
|
|
|
|
struct timespec t;
|
|
|
|
int res;
|
|
|
|
|
|
|
|
res = clock_gettime(CLOCK_MONOTONIC, &t);
|
|
|
|
if (res < 0) {
|
|
|
|
fprintf(stderr, "Error with gettimeofday! (%i)\n", res);
|
|
|
|
exit(EXIT_FAIL);
|
|
|
|
}
|
|
|
|
return (__u64) t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Common stats data record shared with _kern.c */
|
|
|
|
struct datarec {
|
|
|
|
__u64 processed;
|
|
|
|
__u64 issue;
|
|
|
|
};
|
|
|
|
struct record {
|
|
|
|
__u64 timestamp;
|
|
|
|
struct datarec total;
|
|
|
|
struct datarec *cpu;
|
|
|
|
};
|
|
|
|
struct stats_record {
|
|
|
|
struct record stats;
|
|
|
|
struct record *rxq;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct datarec *alloc_record_per_cpu(void)
|
|
|
|
{
|
|
|
|
unsigned int nr_cpus = bpf_num_possible_cpus();
|
|
|
|
struct datarec *array;
|
|
|
|
size_t size;
|
|
|
|
|
|
|
|
size = sizeof(struct datarec) * nr_cpus;
|
|
|
|
array = malloc(size);
|
|
|
|
memset(array, 0, size);
|
|
|
|
if (!array) {
|
|
|
|
fprintf(stderr, "Mem alloc error (nr_cpus:%u)\n", nr_cpus);
|
|
|
|
exit(EXIT_FAIL_MEM);
|
|
|
|
}
|
|
|
|
return array;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct record *alloc_record_per_rxq(void)
|
|
|
|
{
|
2018-05-10 17:24:43 +00:00
|
|
|
unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
struct record *array;
|
|
|
|
size_t size;
|
|
|
|
|
|
|
|
size = sizeof(struct record) * nr_rxqs;
|
|
|
|
array = malloc(size);
|
|
|
|
memset(array, 0, size);
|
|
|
|
if (!array) {
|
|
|
|
fprintf(stderr, "Mem alloc error (nr_rxqs:%u)\n", nr_rxqs);
|
|
|
|
exit(EXIT_FAIL_MEM);
|
|
|
|
}
|
|
|
|
return array;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct stats_record *alloc_stats_record(void)
|
|
|
|
{
|
2018-05-10 17:24:43 +00:00
|
|
|
unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
struct stats_record *rec;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
rec = malloc(sizeof(*rec));
|
|
|
|
memset(rec, 0, sizeof(*rec));
|
|
|
|
if (!rec) {
|
|
|
|
fprintf(stderr, "Mem alloc error\n");
|
|
|
|
exit(EXIT_FAIL_MEM);
|
|
|
|
}
|
|
|
|
rec->rxq = alloc_record_per_rxq();
|
|
|
|
for (i = 0; i < nr_rxqs; i++)
|
|
|
|
rec->rxq[i].cpu = alloc_record_per_cpu();
|
|
|
|
|
|
|
|
rec->stats.cpu = alloc_record_per_cpu();
|
|
|
|
return rec;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void free_stats_record(struct stats_record *r)
|
|
|
|
{
|
2018-05-10 17:24:43 +00:00
|
|
|
unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < nr_rxqs; i++)
|
|
|
|
free(r->rxq[i].cpu);
|
|
|
|
|
|
|
|
free(r->rxq);
|
|
|
|
free(r->stats.cpu);
|
|
|
|
free(r);
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool map_collect_percpu(int fd, __u32 key, struct record *rec)
|
|
|
|
{
|
|
|
|
/* For percpu maps, userspace gets a value per possible CPU */
|
|
|
|
unsigned int nr_cpus = bpf_num_possible_cpus();
|
|
|
|
struct datarec values[nr_cpus];
|
|
|
|
__u64 sum_processed = 0;
|
|
|
|
__u64 sum_issue = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if ((bpf_map_lookup_elem(fd, &key, values)) != 0) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"ERR: bpf_map_lookup_elem failed key:0x%X\n", key);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
/* Get time as close as possible to reading map contents */
|
|
|
|
rec->timestamp = gettime();
|
|
|
|
|
|
|
|
/* Record and sum values from each CPU */
|
|
|
|
for (i = 0; i < nr_cpus; i++) {
|
|
|
|
rec->cpu[i].processed = values[i].processed;
|
|
|
|
sum_processed += values[i].processed;
|
|
|
|
rec->cpu[i].issue = values[i].issue;
|
|
|
|
sum_issue += values[i].issue;
|
|
|
|
}
|
|
|
|
rec->total.processed = sum_processed;
|
|
|
|
rec->total.issue = sum_issue;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void stats_collect(struct stats_record *rec)
|
|
|
|
{
|
|
|
|
int fd, i, max_rxqs;
|
|
|
|
|
2018-05-10 17:24:43 +00:00
|
|
|
fd = bpf_map__fd(stats_global_map);
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
map_collect_percpu(fd, 0, &rec->stats);
|
|
|
|
|
2018-05-10 17:24:43 +00:00
|
|
|
fd = bpf_map__fd(rx_queue_index_map);
|
|
|
|
max_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
for (i = 0; i < max_rxqs; i++)
|
|
|
|
map_collect_percpu(fd, i, &rec->rxq[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
static double calc_period(struct record *r, struct record *p)
|
|
|
|
{
|
|
|
|
double period_ = 0;
|
|
|
|
__u64 period = 0;
|
|
|
|
|
|
|
|
period = r->timestamp - p->timestamp;
|
|
|
|
if (period > 0)
|
|
|
|
period_ = ((double) period / NANOSEC_PER_SEC);
|
|
|
|
|
|
|
|
return period_;
|
|
|
|
}
|
|
|
|
|
|
|
|
static __u64 calc_pps(struct datarec *r, struct datarec *p, double period_)
|
|
|
|
{
|
|
|
|
__u64 packets = 0;
|
|
|
|
__u64 pps = 0;
|
|
|
|
|
|
|
|
if (period_ > 0) {
|
|
|
|
packets = r->processed - p->processed;
|
|
|
|
pps = packets / period_;
|
|
|
|
}
|
|
|
|
return pps;
|
|
|
|
}
|
|
|
|
|
|
|
|
static __u64 calc_errs_pps(struct datarec *r,
|
|
|
|
struct datarec *p, double period_)
|
|
|
|
{
|
|
|
|
__u64 packets = 0;
|
|
|
|
__u64 pps = 0;
|
|
|
|
|
|
|
|
if (period_ > 0) {
|
|
|
|
packets = r->issue - p->issue;
|
|
|
|
pps = packets / period_;
|
|
|
|
}
|
|
|
|
return pps;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void stats_print(struct stats_record *stats_rec,
|
|
|
|
struct stats_record *stats_prev,
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
int action, __u32 cfg_opt)
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
{
|
2018-05-10 17:24:43 +00:00
|
|
|
unsigned int nr_rxqs = bpf_map__def(rx_queue_index_map)->max_entries;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
unsigned int nr_cpus = bpf_num_possible_cpus();
|
|
|
|
double pps = 0, err = 0;
|
|
|
|
struct record *rec, *prev;
|
|
|
|
double t;
|
|
|
|
int rxq;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
/* Header */
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
printf("\nRunning XDP on dev:%s (ifindex:%d) action:%s options:%s\n",
|
|
|
|
ifname, ifindex, action2str(action), options2str(cfg_opt));
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
|
|
|
/* stats_global_map */
|
|
|
|
{
|
|
|
|
char *fmt_rx = "%-15s %-7d %'-11.0f %'-10.0f %s\n";
|
|
|
|
char *fm2_rx = "%-15s %-7s %'-11.0f\n";
|
|
|
|
char *errstr = "";
|
|
|
|
|
|
|
|
printf("%-15s %-7s %-11s %-11s\n",
|
|
|
|
"XDP stats", "CPU", "pps", "issue-pps");
|
|
|
|
|
|
|
|
rec = &stats_rec->stats;
|
|
|
|
prev = &stats_prev->stats;
|
|
|
|
t = calc_period(rec, prev);
|
|
|
|
for (i = 0; i < nr_cpus; i++) {
|
|
|
|
struct datarec *r = &rec->cpu[i];
|
|
|
|
struct datarec *p = &prev->cpu[i];
|
|
|
|
|
|
|
|
pps = calc_pps (r, p, t);
|
|
|
|
err = calc_errs_pps(r, p, t);
|
|
|
|
if (err > 0)
|
|
|
|
errstr = "invalid-ifindex";
|
|
|
|
if (pps > 0)
|
|
|
|
printf(fmt_rx, "XDP-RX CPU",
|
|
|
|
i, pps, err, errstr);
|
|
|
|
}
|
|
|
|
pps = calc_pps (&rec->total, &prev->total, t);
|
|
|
|
err = calc_errs_pps(&rec->total, &prev->total, t);
|
|
|
|
printf(fm2_rx, "XDP-RX CPU", "total", pps, err);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* rx_queue_index_map */
|
|
|
|
printf("\n%-15s %-7s %-11s %-11s\n",
|
|
|
|
"RXQ stats", "RXQ:CPU", "pps", "issue-pps");
|
|
|
|
|
|
|
|
for (rxq = 0; rxq < nr_rxqs; rxq++) {
|
|
|
|
char *fmt_rx = "%-15s %3d:%-3d %'-11.0f %'-10.0f %s\n";
|
|
|
|
char *fm2_rx = "%-15s %3d:%-3s %'-11.0f\n";
|
|
|
|
char *errstr = "";
|
|
|
|
int rxq_ = rxq;
|
|
|
|
|
|
|
|
/* Last RXQ in map catch overflows */
|
|
|
|
if (rxq_ == nr_rxqs - 1)
|
|
|
|
rxq_ = -1;
|
|
|
|
|
|
|
|
rec = &stats_rec->rxq[rxq];
|
|
|
|
prev = &stats_prev->rxq[rxq];
|
|
|
|
t = calc_period(rec, prev);
|
|
|
|
for (i = 0; i < nr_cpus; i++) {
|
|
|
|
struct datarec *r = &rec->cpu[i];
|
|
|
|
struct datarec *p = &prev->cpu[i];
|
|
|
|
|
|
|
|
pps = calc_pps (r, p, t);
|
|
|
|
err = calc_errs_pps(r, p, t);
|
|
|
|
if (err > 0) {
|
|
|
|
if (rxq_ == -1)
|
|
|
|
errstr = "map-overflow-RXQ";
|
|
|
|
else
|
|
|
|
errstr = "err";
|
|
|
|
}
|
|
|
|
if (pps > 0)
|
|
|
|
printf(fmt_rx, "rx_queue_index",
|
|
|
|
rxq_, i, pps, err, errstr);
|
|
|
|
}
|
|
|
|
pps = calc_pps (&rec->total, &prev->total, t);
|
|
|
|
err = calc_errs_pps(&rec->total, &prev->total, t);
|
|
|
|
if (pps || err)
|
|
|
|
printf(fm2_rx, "rx_queue_index", rxq_, "sum", pps, err);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* Pointer swap trick */
|
|
|
|
static inline void swap(struct stats_record **a, struct stats_record **b)
|
|
|
|
{
|
|
|
|
struct stats_record *tmp;
|
|
|
|
|
|
|
|
tmp = *a;
|
|
|
|
*a = *b;
|
|
|
|
*b = tmp;
|
|
|
|
}
|
|
|
|
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
static void stats_poll(int interval, int action, __u32 cfg_opt)
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
{
|
|
|
|
struct stats_record *record, *prev;
|
|
|
|
|
|
|
|
record = alloc_stats_record();
|
|
|
|
prev = alloc_stats_record();
|
|
|
|
stats_collect(record);
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
swap(&prev, &record);
|
|
|
|
stats_collect(record);
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
stats_print(record, prev, action, cfg_opt);
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
sleep(interval);
|
|
|
|
}
|
|
|
|
|
|
|
|
free_stats_record(record);
|
|
|
|
free_stats_record(prev);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int main(int argc, char **argv)
|
|
|
|
{
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
__u32 cfg_options= NO_TOUCH ; /* Default: Don't touch packet memory */
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
struct rlimit r = {10 * 1024 * 1024, RLIM_INFINITY};
|
2018-05-10 17:24:43 +00:00
|
|
|
struct bpf_prog_load_attr prog_load_attr = {
|
|
|
|
.prog_type = BPF_PROG_TYPE_XDP,
|
|
|
|
};
|
2019-02-01 21:42:30 +00:00
|
|
|
struct bpf_prog_info info = {};
|
|
|
|
__u32 info_len = sizeof(info);
|
2018-05-10 17:24:43 +00:00
|
|
|
int prog_fd, map_fd, opt, err;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
bool use_separators = true;
|
|
|
|
struct config cfg = { 0 };
|
2018-05-10 17:24:43 +00:00
|
|
|
struct bpf_object *obj;
|
|
|
|
struct bpf_map *map;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
char filename[256];
|
|
|
|
int longindex = 0;
|
|
|
|
int interval = 2;
|
|
|
|
__u32 key = 0;
|
|
|
|
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
char action_str_buf[XDP_ACTION_MAX_STRLEN + 1 /* for \0 */] = { 0 };
|
|
|
|
int action = XDP_PASS; /* Default action */
|
|
|
|
char *action_str = NULL;
|
|
|
|
|
|
|
|
snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
|
2018-05-10 17:24:43 +00:00
|
|
|
prog_load_attr.file = filename;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
|
|
|
if (setrlimit(RLIMIT_MEMLOCK, &r)) {
|
|
|
|
perror("setrlimit(RLIMIT_MEMLOCK)");
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2018-05-10 17:24:43 +00:00
|
|
|
if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
|
|
|
|
return EXIT_FAIL;
|
|
|
|
|
|
|
|
map = bpf_map__next(NULL, obj);
|
|
|
|
stats_global_map = bpf_map__next(map, obj);
|
|
|
|
rx_queue_index_map = bpf_map__next(stats_global_map, obj);
|
|
|
|
if (!map || !stats_global_map || !rx_queue_index_map) {
|
|
|
|
printf("finding a map in obj file failed\n");
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
return EXIT_FAIL;
|
|
|
|
}
|
2018-05-10 17:24:43 +00:00
|
|
|
map_fd = bpf_map__fd(map);
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
2018-05-10 17:24:43 +00:00
|
|
|
if (!prog_fd) {
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
fprintf(stderr, "ERR: load_bpf_file: %s\n", strerror(errno));
|
|
|
|
return EXIT_FAIL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Parse commands line args */
|
2019-02-01 21:42:28 +00:00
|
|
|
while ((opt = getopt_long(argc, argv, "FhSrmzd:s:a:",
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
long_options, &longindex)) != -1) {
|
|
|
|
switch (opt) {
|
|
|
|
case 'd':
|
|
|
|
if (strlen(optarg) >= IF_NAMESIZE) {
|
|
|
|
fprintf(stderr, "ERR: --dev name too long\n");
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
ifname = (char *)&ifname_buf;
|
|
|
|
strncpy(ifname, optarg, IF_NAMESIZE);
|
|
|
|
ifindex = if_nametoindex(ifname);
|
|
|
|
if (ifindex == 0) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"ERR: --dev name unknown err(%d):%s\n",
|
|
|
|
errno, strerror(errno));
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case 's':
|
|
|
|
interval = atoi(optarg);
|
|
|
|
break;
|
|
|
|
case 'S':
|
|
|
|
xdp_flags |= XDP_FLAGS_SKB_MODE;
|
|
|
|
break;
|
|
|
|
case 'z':
|
|
|
|
use_separators = false;
|
|
|
|
break;
|
|
|
|
case 'a':
|
|
|
|
action_str = (char *)&action_str_buf;
|
|
|
|
strncpy(action_str, optarg, XDP_ACTION_MAX_STRLEN);
|
|
|
|
break;
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
case 'r':
|
|
|
|
cfg_options |= READ_MEM;
|
|
|
|
break;
|
2018-06-25 14:27:48 +00:00
|
|
|
case 'm':
|
|
|
|
cfg_options |= SWAP_MAC;
|
|
|
|
break;
|
2019-02-01 21:42:28 +00:00
|
|
|
case 'F':
|
|
|
|
xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
|
|
|
|
break;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
case 'h':
|
|
|
|
error:
|
|
|
|
default:
|
|
|
|
usage(argv);
|
|
|
|
return EXIT_FAIL_OPTION;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/* Required option */
|
|
|
|
if (ifindex == -1) {
|
|
|
|
fprintf(stderr, "ERR: required option --dev missing\n");
|
|
|
|
usage(argv);
|
|
|
|
return EXIT_FAIL_OPTION;
|
|
|
|
}
|
|
|
|
cfg.ifindex = ifindex;
|
|
|
|
|
|
|
|
/* Parse action string */
|
|
|
|
if (action_str) {
|
|
|
|
action = parse_xdp_action(action_str);
|
|
|
|
if (action < 0) {
|
|
|
|
fprintf(stderr, "ERR: Invalid XDP --action: %s\n",
|
|
|
|
action_str);
|
|
|
|
list_xdp_actions();
|
|
|
|
return EXIT_FAIL_OPTION;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
cfg.action = action;
|
2018-06-25 14:27:48 +00:00
|
|
|
|
|
|
|
/* XDP_TX requires changing MAC-addrs, else HW may drop */
|
|
|
|
if (action == XDP_TX)
|
|
|
|
cfg_options |= SWAP_MAC;
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
cfg.options = cfg_options;
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
|
|
|
/* Trick to pretty printf with thousands separators use %' */
|
|
|
|
if (use_separators)
|
|
|
|
setlocale(LC_NUMERIC, "en_US");
|
|
|
|
|
|
|
|
/* User-side setup ifindex in config_map */
|
2018-05-10 17:24:43 +00:00
|
|
|
err = bpf_map_update_elem(map_fd, &key, &cfg, 0);
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
if (err) {
|
|
|
|
fprintf(stderr, "Store config failed (err:%d)\n", err);
|
|
|
|
exit(EXIT_FAIL_BPF);
|
|
|
|
}
|
|
|
|
|
2018-08-15 14:57:14 +00:00
|
|
|
/* Remove XDP program when program is interrupted or killed */
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
signal(SIGINT, int_exit);
|
2018-08-15 14:57:14 +00:00
|
|
|
signal(SIGTERM, int_exit);
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
|
2018-05-10 17:24:43 +00:00
|
|
|
if (bpf_set_link_xdp_fd(ifindex, prog_fd, xdp_flags) < 0) {
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
fprintf(stderr, "link set xdp fd failed\n");
|
|
|
|
return EXIT_FAIL_XDP;
|
|
|
|
}
|
|
|
|
|
2019-02-01 21:42:30 +00:00
|
|
|
err = bpf_obj_get_info_by_fd(prog_fd, &info, &info_len);
|
|
|
|
if (err) {
|
|
|
|
printf("can't get prog info - %s\n", strerror(errno));
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
prog_id = info.id;
|
|
|
|
|
samples/bpf: extend xdp_rxq_info to read packet payload
There is a cost associated with reading the packet data payload
that this test ignored. Add option --read to allow enabling
reading part of the payload.
This sample/tool helps us analyse an issue observed with a NIC
mlx5 (ConnectX-5 Ex) and an Intel(R) Xeon(R) CPU E5-1650 v4.
With no_touch of data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:no_touch
XDP stats CPU pps issue-pps
XDP-RX CPU 0 14,465,157 0
XDP-RX CPU 1 14,464,728 0
XDP-RX CPU 2 14,465,283 0
XDP-RX CPU 3 14,465,282 0
XDP-RX CPU 4 14,464,159 0
XDP-RX CPU 5 14,465,379 0
XDP-RX CPU total 86,789,992
When not touching data, we observe that the CPUs have idle cycles.
When reading data the CPUs are 100% busy in softirq.
With reading data:
Running XDP on dev:mlx5p1 (ifindex:8) action:XDP_DROP options:read
XDP stats CPU pps issue-pps
XDP-RX CPU 0 9,620,639 0
XDP-RX CPU 1 9,489,843 0
XDP-RX CPU 2 9,407,854 0
XDP-RX CPU 3 9,422,289 0
XDP-RX CPU 4 9,321,959 0
XDP-RX CPU 5 9,395,242 0
XDP-RX CPU total 56,657,828
The effect seen above is a result of cache-misses occuring when
more RXQs are being used. Based on perf-event observations, our
conclusion is that the CPUs DDIO (Direct Data I/O) choose to
deliver packet into main memory, instead of L3-cache. We also
found, that this can be mitigated by either using less RXQs or by
reducing NICs the RX-ring size.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-06-25 14:27:43 +00:00
|
|
|
stats_poll(interval, action, cfg_options);
|
samples/bpf: program demonstrating access to xdp_rxq_info
This sample program can be used for monitoring and reporting how many
packets per sec (pps) are received per NIC RX queue index and which
CPU processed the packet. In itself it is a useful tool for quickly
identifying RSS imbalance issues, see below.
The default XDP action is XDP_PASS in-order to provide a monitor
mode. For benchmarking purposes it is possible to specify other XDP
actions on the cmdline --action.
Output below shows an imbalance RSS case where most RXQ's deliver to
CPU-0 while CPU-2 only get packets from a single RXQ. Looking at
things from a CPU level the two CPUs are processing approx the same
amount, BUT looking at the rx_queue_index levels it is clear that
RXQ-2 receive much better service, than other RXQs which all share CPU-0.
Running XDP on dev:i40e1 (ifindex:3) action:XDP_PASS
XDP stats CPU pps issue-pps
XDP-RX CPU 0 900,473 0
XDP-RX CPU 2 906,921 0
XDP-RX CPU total 1,807,395
RXQ stats RXQ:CPU pps issue-pps
rx_queue_index 0:0 180,098 0
rx_queue_index 0:sum 180,098
rx_queue_index 1:0 180,098 0
rx_queue_index 1:sum 180,098
rx_queue_index 2:2 906,921 0
rx_queue_index 2:sum 906,921
rx_queue_index 3:0 180,098 0
rx_queue_index 3:sum 180,098
rx_queue_index 4:0 180,082 0
rx_queue_index 4:sum 180,082
rx_queue_index 5:0 180,093 0
rx_queue_index 5:sum 180,093
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2018-01-03 10:26:19 +00:00
|
|
|
return EXIT_OK;
|
|
|
|
}
|