Skip to contentSkip to conversation

samples: bpf: split xdpsock stats into new struct

PATCH Applied Ciara Loftus 3 participants 6 messages Oct 2, 2020 – Oct 6, 2020

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/3 samples: bpf: split xdpsock stats into new struct

New statistics will be added in future commits. In preparation for this, let's split out the existing statistics into their own struct.

---
 samples/bpf/xdpsock_user.c | 123 +++++++++++++++++++++----------------
 1 file changed, 69 insertions(+), 54 deletions(-)
samples/bpf/xdpsock_user.c +69 -54
index b220173dbe1e..4c5022e6479e 100644
@@ -91,18 +91,7 @@ static bool opt_need_wakeup = true;
 static u32 opt_num_xsks = 1;
 static u32 prog_id;
 
-struct xsk_umem_info {
-	struct xsk_ring_prod fq;
-	struct xsk_ring_cons cq;
-	struct xsk_umem *umem;
-	void *buffer;
-};
-
-struct xsk_socket_info {
-	struct xsk_ring_cons rx;
-	struct xsk_ring_prod tx;
-	struct xsk_umem_info *umem;
-	struct xsk_socket *xsk;
+struct xsk_ring_stats {
 	unsigned long rx_npkts;
 	unsigned long tx_npkts;
 	unsigned long rx_dropped_npkts;
@@ -119,6 +108,21 @@ struct xsk_socket_info {
 	unsigned long prev_rx_full_npkts;
 	unsigned long prev_rx_fill_empty_npkts;
 	unsigned long prev_tx_empty_npkts;
+};
+
+struct xsk_umem_info {
+	struct xsk_ring_prod fq;
+	struct xsk_ring_cons cq;
+	struct xsk_umem *umem;
+	void *buffer;
+};
+
+struct xsk_socket_info {
+	struct xsk_ring_cons rx;
+	struct xsk_ring_prod tx;
+	struct xsk_umem_info *umem;
+	struct xsk_socket *xsk;
+	struct xsk_ring_stats ring_stats;
 	u32 outstanding_tx;
 };
 
@@ -173,12 +177,12 @@ static int xsk_get_xdp_stats(int fd, struct xsk_socket_info *xsk)
 		return err;
 
 	if (optlen == sizeof(struct xdp_statistics)) {
-		xsk->rx_dropped_npkts = stats.rx_dropped;
-		xsk->rx_invalid_npkts = stats.rx_invalid_descs;
-		xsk->tx_invalid_npkts = stats.tx_invalid_descs;
-		xsk->rx_full_npkts = stats.rx_ring_full;
-		xsk->rx_fill_empty_npkts = stats.rx_fill_ring_empty_descs;
-		xsk->tx_empty_npkts = stats.tx_ring_empty_descs;
+		xsk->ring_stats.rx_dropped_npkts = stats.rx_dropped;
+		xsk->ring_stats.rx_invalid_npkts = stats.rx_invalid_descs;
+		xsk->ring_stats.tx_invalid_npkts = stats.tx_invalid_descs;
+		xsk->ring_stats.rx_full_npkts = stats.rx_ring_full;
+		xsk->ring_stats.rx_fill_empty_npkts = stats.rx_fill_ring_empty_descs;
+		xsk->ring_stats.tx_empty_npkts = stats.tx_ring_empty_descs;
 		return 0;
 	}
 
@@ -198,9 +202,9 @@ static void dump_stats(void)
 		double rx_pps, tx_pps, dropped_pps, rx_invalid_pps, full_pps, fill_empty_pps,
 			tx_invalid_pps, tx_empty_pps;
 
-		rx_pps = (xsks[i]->rx_npkts - xsks[i]->prev_rx_npkts) *
+		rx_pps = (xsks[i]->ring_stats.rx_npkts - xsks[i]->ring_stats.prev_rx_npkts) *
 			 1000000000. / dt;
-		tx_pps = (xsks[i]->tx_npkts - xsks[i]->prev_tx_npkts) *
+		tx_pps = (xsks[i]->ring_stats.tx_npkts - xsks[i]->ring_stats.prev_tx_npkts) *
 			 1000000000. / dt;
 
 		printf("\n sock%d@", i);
@@ -209,47 +213,58 @@ static void dump_stats(void)
 
 		printf("%-15s %-11s %-11s %-11.2f\n", "", "pps", "pkts",
 		       dt / 1000000000.);
-		printf(fmt, "rx", rx_pps, xsks[i]->rx_npkts);
-		printf(fmt, "tx", tx_pps, xsks[i]->tx_npkts);
+		printf(fmt, "rx", rx_pps, xsks[i]->ring_stats.rx_npkts);
+		printf(fmt, "tx", tx_pps, xsks[i]->ring_stats.tx_npkts);
 
-		xsks[i]->prev_rx_npkts = xsks[i]->rx_npkts;
-		xsks[i]->prev_tx_npkts = xsks[i]->tx_npkts;
+		xsks[i]->ring_stats.prev_rx_npkts = xsks[i]->ring_stats.rx_npkts;
+		xsks[i]->ring_stats.prev_tx_npkts = xsks[i]->ring_stats.tx_npkts;
 
 		if (opt_extra_stats) {
 			if (!xsk_get_xdp_stats(xsk_socket__fd(xsks[i]->xsk), xsks[i])) {
-				dropped_pps = (xsks[i]->rx_dropped_npkts -
-						xsks[i]->prev_rx_dropped_npkts) * 1000000000. / dt;
-				rx_invalid_pps = (xsks[i]->rx_invalid_npkts -
-						xsks[i]->prev_rx_invalid_npkts) * 1000000000. / dt;
-				tx_invalid_pps = (xsks[i]->tx_invalid_npkts -
-						xsks[i]->prev_tx_invalid_npkts) * 1000000000. / dt;
-				full_pps = (xsks[i]->rx_full_npkts -
-						xsks[i]->prev_rx_full_npkts) * 1000000000. / dt;
-				fill_empty_pps = (xsks[i]->rx_fill_empty_npkts -
-						xsks[i]->prev_rx_fill_empty_npkts)
-						* 1000000000. / dt;
-				tx_empty_pps = (xsks[i]->tx_empty_npkts -
-						xsks[i]->prev_tx_empty_npkts) * 1000000000. / dt;
+				dropped_pps = (xsks[i]->ring_stats.rx_dropped_npkts -
+						xsks[i]->ring_stats.prev_rx_dropped_npkts) *
+							1000000000. / dt;
+				rx_invalid_pps = (xsks[i]->ring_stats.rx_invalid_npkts -
+						xsks[i]->ring_stats.prev_rx_invalid_npkts) *
+							1000000000. / dt;
+				tx_invalid_pps = (xsks[i]->ring_stats.tx_invalid_npkts -
+						xsks[i]->ring_stats.prev_tx_invalid_npkts) *
+							1000000000. / dt;
+				full_pps = (xsks[i]->ring_stats.rx_full_npkts -
+						xsks[i]->ring_stats.prev_rx_full_npkts) *
+							1000000000. / dt;
+				fill_empty_pps = (xsks[i]->ring_stats.rx_fill_empty_npkts -
+						xsks[i]->ring_stats.prev_rx_fill_empty_npkts) *
+							1000000000. / dt;
+				tx_empty_pps = (xsks[i]->ring_stats.tx_empty_npkts -
+						xsks[i]->ring_stats.prev_tx_empty_npkts) *
+							1000000000. / dt;
 
 				printf(fmt, "rx dropped", dropped_pps,
-				       xsks[i]->rx_dropped_npkts);
+				       xsks[i]->ring_stats.rx_dropped_npkts);
 				printf(fmt, "rx invalid", rx_invalid_pps,
-				       xsks[i]->rx_invalid_npkts);
+				       xsks[i]->ring_stats.rx_invalid_npkts);
 				printf(fmt, "tx invalid", tx_invalid_pps,
-				       xsks[i]->tx_invalid_npkts);
+				       xsks[i]->ring_stats.tx_invalid_npkts);
 				printf(fmt, "rx queue full", full_pps,
-				       xsks[i]->rx_full_npkts);
+				       xsks[i]->ring_stats.rx_full_npkts);
 				printf(fmt, "fill ring empty", fill_empty_pps,
-				       xsks[i]->rx_fill_empty_npkts);
+				       xsks[i]->ring_stats.rx_fill_empty_npkts);
 				printf(fmt, "tx ring empty", tx_empty_pps,
-				       xsks[i]->tx_empty_npkts);
-
-				xsks[i]->prev_rx_dropped_npkts = xsks[i]->rx_dropped_npkts;
-				xsks[i]->prev_rx_invalid_npkts = xsks[i]->rx_invalid_npkts;
-				xsks[i]->prev_tx_invalid_npkts = xsks[i]->tx_invalid_npkts;
-				xsks[i]->prev_rx_full_npkts = xsks[i]->rx_full_npkts;
-				xsks[i]->prev_rx_fill_empty_npkts = xsks[i]->rx_fill_empty_npkts;
-				xsks[i]->prev_tx_empty_npkts = xsks[i]->tx_empty_npkts;
+				       xsks[i]->ring_stats.tx_empty_npkts);
+
+				xsks[i]->ring_stats.prev_rx_dropped_npkts =
+					xsks[i]->ring_stats.rx_dropped_npkts;
+				xsks[i]->ring_stats.prev_rx_invalid_npkts =
+					xsks[i]->ring_stats.rx_invalid_npkts;
+				xsks[i]->ring_stats.prev_tx_invalid_npkts =
+					xsks[i]->ring_stats.tx_invalid_npkts;
+				xsks[i]->ring_stats.prev_rx_full_npkts =
+					xsks[i]->ring_stats.rx_full_npkts;
+				xsks[i]->ring_stats.prev_rx_fill_empty_npkts =
+					xsks[i]->ring_stats.rx_fill_empty_npkts;
+				xsks[i]->ring_stats.prev_tx_empty_npkts =
+					xsks[i]->ring_stats.tx_empty_npkts;
 			} else {
 				printf("%-15s\n", "Error retrieving extra stats");
 			}
@@ -936,7 +951,7 @@ static inline void complete_tx_l2fwd(struct xsk_socket_info *xsk,
 		xsk_ring_prod__submit(&xsk->umem->fq, rcvd);
 		xsk_ring_cons__release(&xsk->umem->cq, rcvd);
 		xsk->outstanding_tx -= rcvd;
-		xsk->tx_npkts += rcvd;
+		xsk->ring_stats.tx_npkts += rcvd;
 	}
 }
 
@@ -956,7 +971,7 @@ static inline void complete_tx_only(struct xsk_socket_info *xsk,
 	if (rcvd > 0) {
 		xsk_ring_cons__release(&xsk->umem->cq, rcvd);
 		xsk->outstanding_tx -= rcvd;
-		xsk->tx_npkts += rcvd;
+		xsk->ring_stats.tx_npkts += rcvd;
 	}
 }
 
@@ -996,7 +1011,7 @@ static void rx_drop(struct xsk_socket_info *xsk, struct pollfd *fds)
 
 	xsk_ring_prod__submit(&xsk->umem->fq, rcvd);
 	xsk_ring_cons__release(&xsk->rx, rcvd);
-	xsk->rx_npkts += rcvd;
+	xsk->ring_stats.rx_npkts += rcvd;
 }
 
 static void rx_drop_all(void)
@@ -1155,7 +1170,7 @@ static void l2fwd(struct xsk_socket_info *xsk, struct pollfd *fds)
 	xsk_ring_prod__submit(&xsk->tx, rcvd);
 	xsk_ring_cons__release(&xsk->rx, rcvd);
 
-	xsk->rx_npkts += rcvd;
+	xsk->ring_stats.rx_npkts += rcvd;
 	xsk->outstanding_tx += rcvd;
 }
 
Signature · 1 line
2.17.1
#1 lore raw

PATCH 2/3 samples: bpf: count syscalls in xdpsock

Categorise and record syscalls issued in the xdpsock sample app. The categories recorded are:

  rx_empty_polls:    polls when the rx ring is empty
  fill_fail_polls:   polls when failed to get addr from fill ring
  copy_tx_sendtos:   sendtos issued for tx when copy mode enabled

tx_wakeup_sendtos: sendtos issued when tx ring needs waking up
opt_polls: polls issued since the '-p' flag is set

Print the stats using '-a' on the xdpsock command line.

---
 samples/bpf/xdpsock_user.c | 113 +++++++++++++++++++++++++++++++++----
 1 file changed, 103 insertions(+), 10 deletions(-)
samples/bpf/xdpsock_user.c +103 -10
index 4c5022e6479e..ff119ede4ab1 100644
@@ -79,6 +79,7 @@ static u16 opt_pkt_size = MIN_PKT_SIZE;
 static u32 opt_pkt_fill_pattern = 0x12345678;
 static bool opt_extra_stats;
 static bool opt_quiet;
+static bool opt_app_stats;
 static int opt_poll;
 static int opt_interval = 1;
 static u32 opt_xdp_bind_flags = XDP_USE_NEED_WAKEUP;
@@ -110,6 +111,19 @@ struct xsk_ring_stats {
 	unsigned long prev_tx_empty_npkts;
 };
 
+struct xsk_app_stats {
+	unsigned long rx_empty_polls;
+	unsigned long fill_fail_polls;
+	unsigned long copy_tx_sendtos;
+	unsigned long tx_wakeup_sendtos;
+	unsigned long opt_polls;
+	unsigned long prev_rx_empty_polls;
+	unsigned long prev_fill_fail_polls;
+	unsigned long prev_copy_tx_sendtos;
+	unsigned long prev_tx_wakeup_sendtos;
+	unsigned long prev_opt_polls;
+};
+
 struct xsk_umem_info {
 	struct xsk_ring_prod fq;
 	struct xsk_ring_cons cq;
@@ -123,6 +137,7 @@ struct xsk_socket_info {
 	struct xsk_umem_info *umem;
 	struct xsk_socket *xsk;
 	struct xsk_ring_stats ring_stats;
+	struct xsk_app_stats app_stats;
 	u32 outstanding_tx;
 };
 
@@ -189,6 +204,45 @@ static int xsk_get_xdp_stats(int fd, struct xsk_socket_info *xsk)
 	return -EINVAL;
 }
 
+static void dump_app_stats(long dt)
+{
+	int i;
+
+	for (i = 0; i < num_socks && xsks[i]; i++) {
+		char *fmt = "%-18s %'-14.0f %'-14lu\n";
+		double rx_empty_polls_ps, fill_fail_polls_ps, copy_tx_sendtos_ps,
+				tx_wakeup_sendtos_ps, opt_polls_ps;
+
+		rx_empty_polls_ps = (xsks[i]->app_stats.rx_empty_polls -
+					xsks[i]->app_stats.prev_rx_empty_polls) * 1000000000. / dt;
+		fill_fail_polls_ps = (xsks[i]->app_stats.fill_fail_polls -
+					xsks[i]->app_stats.prev_fill_fail_polls) * 1000000000. / dt;
+		copy_tx_sendtos_ps = (xsks[i]->app_stats.copy_tx_sendtos -
+					xsks[i]->app_stats.prev_copy_tx_sendtos) * 1000000000. / dt;
+		tx_wakeup_sendtos_ps = (xsks[i]->app_stats.tx_wakeup_sendtos -
+					xsks[i]->app_stats.prev_tx_wakeup_sendtos)
+										* 1000000000. / dt;
+		opt_polls_ps = (xsks[i]->app_stats.opt_polls -
+					xsks[i]->app_stats.prev_opt_polls) * 1000000000. / dt;
+
+		printf("\n%-18s %-14s %-14s\n", "", "calls/s", "count");
+		printf(fmt, "rx empty polls", rx_empty_polls_ps, xsks[i]->app_stats.rx_empty_polls);
+		printf(fmt, "fill fail polls", fill_fail_polls_ps,
+							xsks[i]->app_stats.fill_fail_polls);
+		printf(fmt, "copy tx sendtos", copy_tx_sendtos_ps,
+							xsks[i]->app_stats.copy_tx_sendtos);
+		printf(fmt, "tx wakeup sendtos", tx_wakeup_sendtos_ps,
+							xsks[i]->app_stats.tx_wakeup_sendtos);
+		printf(fmt, "opt polls", opt_polls_ps, xsks[i]->app_stats.opt_polls);
+
+		xsks[i]->app_stats.prev_rx_empty_polls = xsks[i]->app_stats.rx_empty_polls;
+		xsks[i]->app_stats.prev_fill_fail_polls = xsks[i]->app_stats.fill_fail_polls;
+		xsks[i]->app_stats.prev_copy_tx_sendtos = xsks[i]->app_stats.copy_tx_sendtos;
+		xsks[i]->app_stats.prev_tx_wakeup_sendtos = xsks[i]->app_stats.tx_wakeup_sendtos;
+		xsks[i]->app_stats.prev_opt_polls = xsks[i]->app_stats.opt_polls;
+	}
+}
+
 static void dump_stats(void)
 {
 	unsigned long now = get_nsecs();
@@ -198,7 +252,7 @@ static void dump_stats(void)
 	prev_time = now;
 
 	for (i = 0; i < num_socks && xsks[i]; i++) {
-		char *fmt = "%-15s %'-11.0f %'-11lu\n";
+		char *fmt = "%-18s %'-14.0f %'-14lu\n";
 		double rx_pps, tx_pps, dropped_pps, rx_invalid_pps, full_pps, fill_empty_pps,
 			tx_invalid_pps, tx_empty_pps;
 
@@ -211,7 +265,7 @@ static void dump_stats(void)
 		print_benchmark(false);
 		printf("\n");
 
-		printf("%-15s %-11s %-11s %-11.2f\n", "", "pps", "pkts",
+		printf("%-18s %-14s %-14s %-14.2f\n", "", "pps", "pkts",
 		       dt / 1000000000.);
 		printf(fmt, "rx", rx_pps, xsks[i]->ring_stats.rx_npkts);
 		printf(fmt, "tx", tx_pps, xsks[i]->ring_stats.tx_npkts);
@@ -270,6 +324,9 @@ static void dump_stats(void)
 			}
 		}
 	}
+
+	if (opt_app_stats)
+		dump_app_stats(dt);
 }
 
 static bool is_benchmark_done(void)
@@ -708,6 +765,17 @@ static struct xsk_socket_info *xsk_configure_socket(struct xsk_umem_info *umem,
 	if (ret)
 		exit_with_error(-ret);
 
+	xsk->app_stats.rx_empty_polls = 0;
+	xsk->app_stats.fill_fail_polls = 0;
+	xsk->app_stats.copy_tx_sendtos = 0;
+	xsk->app_stats.tx_wakeup_sendtos = 0;
+	xsk->app_stats.opt_polls = 0;
+	xsk->app_stats.prev_rx_empty_polls = 0;
+	xsk->app_stats.prev_fill_fail_polls = 0;
+	xsk->app_stats.prev_copy_tx_sendtos = 0;
+	xsk->app_stats.prev_tx_wakeup_sendtos = 0;
+	xsk->app_stats.prev_opt_polls = 0;
+
 	return xsk;
 }
 
@@ -735,6 +803,7 @@ static struct option long_options[] = {
 	{"tx-pkt-pattern", required_argument, 0, 'P'},
 	{"extra-stats", no_argument, 0, 'x'},
 	{"quiet", no_argument, 0, 'Q'},
+	{"app-stats", no_argument, 0, 'a'},
 	{0, 0, 0, 0}
 };
 
@@ -771,6 +840,7 @@ static void usage(const char *prog)
 		"  -P, --tx-pkt-pattern=nPacket fill pattern. Default: 0x%x\n"
 		"  -x, --extra-stats	Display extra statistics.\n"
 		"  -Q, --quiet          Do not display any stats.\n"
+		"  -a, --app-stats	Display application (syscall) statistics.\n"
 		"\n";
 	fprintf(stderr, str, prog, XSK_UMEM__DEFAULT_FRAME_SIZE,
 		opt_batch_size, MIN_PKT_SIZE, MIN_PKT_SIZE,
@@ -786,7 +856,7 @@ static void parse_command_line(int argc, char **argv)
 	opterr = 0;
 
 	for (;;) {
-		c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:xQ",
+		c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:xQa",
 				long_options, &option_index);
 		if (c == -1)
 			break;
@@ -873,6 +943,9 @@ static void parse_command_line(int argc, char **argv)
 		case 'Q':
 			opt_quiet = 1;
 			break;
+		case 'a':
+			opt_app_stats = 1;
+			break;
 		default:
 			usage(basename(argv[0]));
 		}
@@ -923,8 +996,10 @@ static inline void complete_tx_l2fwd(struct xsk_socket_info *xsk,
 	 * is driven by the NAPI loop. So as an optimization, we do not have to call
 	 * sendto() all the time in zero-copy mode for l2fwd.
 	 */
-	if (opt_xdp_bind_flags & XDP_COPY)
+	if (opt_xdp_bind_flags & XDP_COPY) {
+		xsk->app_stats.copy_tx_sendtos++;
 		kick_tx(xsk);
+	}
 
 	ndescs = (xsk->outstanding_tx > opt_batch_size) ? opt_batch_size :
 		xsk->outstanding_tx;
@@ -939,8 +1014,10 @@ static inline void complete_tx_l2fwd(struct xsk_socket_info *xsk,
 		while (ret != rcvd) {
 			if (ret < 0)
 				exit_with_error(-ret);
-			if (xsk_ring_prod__needs_wakeup(&umem->fq))
+			if (xsk_ring_prod__needs_wakeup(&umem->fq)) {
+				xsk->app_stats.fill_fail_polls++;
 				ret = poll(fds, num_socks, opt_timeout);
+			}
 			ret = xsk_ring_prod__reserve(&umem->fq, rcvd, &idx_fq);
 		}
 
@@ -964,8 +1041,10 @@ static inline void complete_tx_only(struct xsk_socket_info *xsk,
 	if (!xsk->outstanding_tx)
 		return;
 
-	if (!opt_need_wakeup || xsk_ring_prod__needs_wakeup(&xsk->tx))
+	if (!opt_need_wakeup || xsk_ring_prod__needs_wakeup(&xsk->tx)) {
+		xsk->app_stats.tx_wakeup_sendtos++;
 		kick_tx(xsk);
+	}
 
 	rcvd = xsk_ring_cons__peek(&xsk->umem->cq, batch_size, &idx);
 	if (rcvd > 0) {
@@ -983,8 +1062,10 @@ static void rx_drop(struct xsk_socket_info *xsk, struct pollfd *fds)
 
 	rcvd = xsk_ring_cons__peek(&xsk->rx, opt_batch_size, &idx_rx);
 	if (!rcvd) {
-		if (xsk_ring_prod__needs_wakeup(&xsk->umem->fq))
+		if (xsk_ring_prod__needs_wakeup(&xsk->umem->fq)) {
+			xsk->app_stats.rx_empty_polls++;
 			ret = poll(fds, num_socks, opt_timeout);
+		}
 		return;
 	}
 
@@ -992,8 +1073,10 @@ static void rx_drop(struct xsk_socket_info *xsk, struct pollfd *fds)
 	while (ret != rcvd) {
 		if (ret < 0)
 			exit_with_error(-ret);
-		if (xsk_ring_prod__needs_wakeup(&xsk->umem->fq))
+		if (xsk_ring_prod__needs_wakeup(&xsk->umem->fq)) {
+			xsk->app_stats.fill_fail_polls++;
 			ret = poll(fds, num_socks, opt_timeout);
+		}
 		ret = xsk_ring_prod__reserve(&xsk->umem->fq, rcvd, &idx_fq);
 	}
 
@@ -1026,6 +1109,8 @@ static void rx_drop_all(void)
 
 	for (;;) {
 		if (opt_poll) {
+			for (i = 0; i < num_socks; i++)
+				xsks[i]->app_stats.opt_polls++;
 			ret = poll(fds, num_socks, opt_timeout);
 			if (ret <= 0)
 				continue;
@@ -1106,6 +1191,8 @@ static void tx_only_all(void)
 		int batch_size = get_batch_size(pkt_cnt);
 
 		if (opt_poll) {
+			for (i = 0; i < num_socks; i++)
+				xsks[i]->app_stats.opt_polls++;
 			ret = poll(fds, num_socks, opt_timeout);
 			if (ret <= 0)
 				continue;
@@ -1137,8 +1224,10 @@ static void l2fwd(struct xsk_socket_info *xsk, struct pollfd *fds)
 
 	rcvd = xsk_ring_cons__peek(&xsk->rx, opt_batch_size, &idx_rx);
 	if (!rcvd) {
-		if (xsk_ring_prod__needs_wakeup(&xsk->umem->fq))
+		if (xsk_ring_prod__needs_wakeup(&xsk->umem->fq)) {
+			xsk->app_stats.rx_empty_polls++;
 			ret = poll(fds, num_socks, opt_timeout);
+		}
 		return;
 	}
 
@@ -1147,8 +1236,10 @@ static void l2fwd(struct xsk_socket_info *xsk, struct pollfd *fds)
 		if (ret < 0)
 			exit_with_error(-ret);
 		complete_tx_l2fwd(xsk, fds);
-		if (xsk_ring_prod__needs_wakeup(&xsk->tx))
+		if (xsk_ring_prod__needs_wakeup(&xsk->tx)) {
+			xsk->app_stats.tx_wakeup_sendtos++;
 			kick_tx(xsk);
+		}
 		ret = xsk_ring_prod__reserve(&xsk->tx, rcvd, &idx_tx);
 	}
 
@@ -1186,6 +1277,8 @@ static void l2fwd_all(void)
 
 	for (;;) {
 		if (opt_poll) {
+			for (i = 0; i < num_socks; i++)
+				xsks[i]->app_stats.opt_polls++;
 			ret = poll(fds, num_socks, opt_timeout);
 			if (ret <= 0)
 				continue;
Signature · 1 line
2.17.1
Acked-by on v1 2/3
#2 lore raw

Re: [PATCH bpf-next 2/3] samples: bpf: count syscalls in xdpsock

Quoted 6 lines · Ciara Loftus

Categorise and record syscalls issued in the xdpsock sample app. The categories recorded are:

   rx_empty_polls:    polls when the rx ring is empty
   fill_fail_polls:   polls when failed to get addr from fill ring
   copy_tx_sendtos:   sendtos issued for tx when copy mode enabled
Quoted 6 lines

tx_wakeup_sendtos: sendtos issued when tx ring needs waking up
opt_polls: polls issued since the '-p' flag is set

Print the stats using '-a' on the xdpsock command line.

I am not a xdp expert and cannot really judge whether such additional stats will be useful or not. Beyond that the patch looks good to me.

#3 lore raw

PATCH 3/3 samples: bpf: driver interrupt statistics in xdpsock

In reply to Ciara Loftus

Add an option to count the number of interrupts generated per second and total number of interrupts during the lifetime of the application for a given interface. This information is extracted from /proc/interrupts. Since there is no naming convention across drivers, the user must provide the string which is specific to their interface in the /proc/interrupts file on the command line.

Usage:

./xdpsock ... -I <irq_str>

eg. for queue 0 of i40e device eth0:

./xdpsock ... -I i40e-eth0-TxRx-0

---
 samples/bpf/xdpsock_user.c | 120 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 119 insertions(+), 1 deletion(-)
samples/bpf/xdpsock_user.c +119 -1
index ff119ede4ab1..1149e94ca32f 100644
@@ -11,6 +11,7 @@
 #include <linux/if_xdp.h>
 #include <linux/if_ether.h>
 #include <linux/ip.h>
+#include <linux/limits.h>
 #include <linux/udp.h>
 #include <arpa/inet.h>
 #include <locale.h>
@@ -80,6 +81,9 @@ static u32 opt_pkt_fill_pattern = 0x12345678;
 static bool opt_extra_stats;
 static bool opt_quiet;
 static bool opt_app_stats;
+static const char *opt_irq_str = "";
+static u32 irq_no;
+static int irqs_at_init = -1;
 static int opt_poll;
 static int opt_interval = 1;
 static u32 opt_xdp_bind_flags = XDP_USE_NEED_WAKEUP;
@@ -111,6 +115,11 @@ struct xsk_ring_stats {
 	unsigned long prev_tx_empty_npkts;
 };
 
+struct xsk_driver_stats {
+	unsigned long intrs;
+	unsigned long prev_intrs;
+};
+
 struct xsk_app_stats {
 	unsigned long rx_empty_polls;
 	unsigned long fill_fail_polls;
@@ -138,6 +147,7 @@ struct xsk_socket_info {
 	struct xsk_socket *xsk;
 	struct xsk_ring_stats ring_stats;
 	struct xsk_app_stats app_stats;
+	struct xsk_driver_stats drv_stats;
 	u32 outstanding_tx;
 };
 
@@ -243,6 +253,100 @@ static void dump_app_stats(long dt)
 	}
 }
 
+static bool get_interrupt_number(void)
+{
+	FILE *f_int_proc;
+	char line[4096];
+	bool found = false;
+
+	f_int_proc = fopen("/proc/interrupts", "r");
+	if (f_int_proc == NULL) {
+		printf("Failed to open /proc/interrupts.\n");
+		return found;
+	}
+
+	while (!feof(f_int_proc) && !found) {
+		/* Make sure to read a full line at a time */
+		if (fgets(line, sizeof(line), f_int_proc) == NULL ||
+				line[strlen(line) - 1] != '\n') {
+			printf("Error reading from interrupts file\n");
+			break;
+		}
+
+		/* Extract interrupt number from line */
+		if (strstr(line, opt_irq_str) != NULL) {
+			irq_no = atoi(line);
+			found = true;
+			break;
+		}
+	}
+
+	fclose(f_int_proc);
+
+	return found;
+}
+
+static int get_irqs(void)
+{
+	char count_path[PATH_MAX];
+	int total_intrs = -1;
+	FILE *f_count_proc;
+	char line[4096];
+
+	snprintf(count_path, sizeof(count_path),
+		"/sys/kernel/irq/%i/per_cpu_count", irq_no);
+	f_count_proc = fopen(count_path, "r");
+	if (f_count_proc == NULL) {
+		printf("Failed to open %s\n", count_path);
+		return total_intrs;
+	}
+
+	if (fgets(line, sizeof(line), f_count_proc) == NULL ||
+			line[strlen(line) - 1] != '\n') {
+		printf("Error reading from %s\n", count_path);
+	} else {
+		static const char com[2] = ",";
+		char *token;
+
+		total_intrs = 0;
+		token = strtok(line, com);
+		while (token != NULL) {
+			/* sum up interrupts across all cores */
+			total_intrs += atoi(token);
+			token = strtok(NULL, com);
+		}
+	}
+
+	fclose(f_count_proc);
+
+	return total_intrs;
+}
+
+static void dump_driver_stats(long dt)
+{
+	int i;
+
+	for (i = 0; i < num_socks && xsks[i]; i++) {
+		char *fmt = "%-18s %'-14.0f %'-14lu\n";
+		double intrs_ps;
+		int n_ints = get_irqs();
+
+		if (n_ints < 0) {
+			printf("error getting intr info for intr %i\n", irq_no);
+			return;
+		}
+		xsks[i]->drv_stats.intrs = n_ints - irqs_at_init;
+
+		intrs_ps = (xsks[i]->drv_stats.intrs - xsks[i]->drv_stats.prev_intrs) *
+			 1000000000. / dt;
+
+		printf("\n%-18s %-14s %-14s\n", "", "intrs/s", "count");
+		printf(fmt, "irqs", intrs_ps, xsks[i]->drv_stats.intrs);
+
+		xsks[i]->drv_stats.prev_intrs = xsks[i]->drv_stats.intrs;
+	}
+}
+
 static void dump_stats(void)
 {
 	unsigned long now = get_nsecs();
@@ -327,6 +431,8 @@ static void dump_stats(void)
 
 	if (opt_app_stats)
 		dump_app_stats(dt);
+	if (irq_no)
+		dump_driver_stats(dt);
 }
 
 static bool is_benchmark_done(void)
@@ -804,6 +910,7 @@ static struct option long_options[] = {
 	{"extra-stats", no_argument, 0, 'x'},
 	{"quiet", no_argument, 0, 'Q'},
 	{"app-stats", no_argument, 0, 'a'},
+	{"irq-string", no_argument, 0, 'I'},
 	{0, 0, 0, 0}
 };
 
@@ -841,6 +948,7 @@ static void usage(const char *prog)
 		"  -x, --extra-stats	Display extra statistics.\n"
 		"  -Q, --quiet          Do not display any stats.\n"
 		"  -a, --app-stats	Display application (syscall) statistics.\n"
+		"  -I, --irq-string	Display driver interrupt statistics for interface associated with irq-string.\n"
 		"\n";
 	fprintf(stderr, str, prog, XSK_UMEM__DEFAULT_FRAME_SIZE,
 		opt_batch_size, MIN_PKT_SIZE, MIN_PKT_SIZE,
@@ -856,7 +964,7 @@ static void parse_command_line(int argc, char **argv)
 	opterr = 0;
 
 	for (;;) {
-		c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:xQa",
+		c = getopt_long(argc, argv, "Frtli:q:pSNn:czf:muMd:b:C:s:P:xQaI:",
 				long_options, &option_index);
 		if (c == -1)
 			break;
@@ -945,6 +1053,16 @@ static void parse_command_line(int argc, char **argv)
 			break;
 		case 'a':
 			opt_app_stats = 1;
+			break;
+		case 'I':
+			opt_irq_str = optarg;
+			if (get_interrupt_number())
+				irqs_at_init = get_irqs();
+			if (irqs_at_init < 0) {
+				fprintf(stderr, "ERROR: Failed to get irqs for %s\n", opt_irq_str);
+				usage(basename(argv[0]));
+			}
+
 			break;
 		default:
 			usage(basename(argv[0]));
Signature · 1 line
2.17.1
#5 lore raw
In reply to Ciara Loftus
This series was applied to bpf/bpf-next.git (refs/heads/master):

Hello:

This series was applied to bpf/bpf-next.git (refs/heads/master):

Quoted 7 lines · On Fri, 2 Oct 2020 13:36:10 +0000 you wrote:

New statistics will be added in future commits. In preparation for this, let's split out the existing statistics into their own struct.

---
 samples/bpf/xdpsock_user.c | 123 +++++++++++++++++++++----------------
 1 file changed, 69 insertions(+), 54 deletions(-)

Here is the summary with links:
- [bpf-next,1/3] samples: bpf: split xdpsock stats into new struct

    https://git.kernel.org/bpf/bpf-next/c/2e8806f032f5
  - [bpf-next,2/3] samples: bpf: count syscalls in xdpsock
    https://git.kernel.org/bpf/bpf-next/c/60dc609dbd54
  - [bpf-next,3/3] samples: bpf: driver interrupt statistics in xdpsock
    https://git.kernel.org/bpf/bpf-next/c/67ed375530e2

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html