Real life sort of benchmark

src/messages.h

@@ -50,7 +50,7 @@ enum limits {
 	MAX_TIMER_HANDSHAKES = 90 / REKEY_TIMEOUT,
 	MAX_QUEUED_INCOMING_HANDSHAKES = 4096, /* TODO: replace this with DQL */
 	MAX_STAGED_PACKETS = 128,
-	MAX_QUEUED_PACKETS = 1024 /* TODO: replace this with DQL */
+	MAX_QUEUED_PACKETS = 102400 /* TODO: replace this with DQL */
 };
 
 enum message_type {

src/receive.c

@@ -580,6 +580,8 @@ void packet_receive(struct wireguard_device *wg, struct sk_buff *skb)
 		break;
 	}
 	case cpu_to_le32(MESSAGE_DATA):
+		if (skb->len > MESSAGE_MINIMUM_LENGTH)
+			goto err;
 		PACKET_CB(skb)->ds = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
 		packet_consume_data(wg, skb);
 		break;

src/send.c

@@ -291,6 +291,12 @@ void packet_tx_worker(struct work_struct *work)
 	}
 }
 
+static inline void report_bench(unsigned int bytes, cycles_t duration)
+{
+	cycles_t cCPB = duration * 100;
+	pr_err("%u bytes in %u cycles (%u cCPB)\n", bytes, duration, cCPB / bytes);
+}
+
 void packet_encrypt_worker(struct work_struct *work)
 {
 	struct crypt_queue *queue =
@@ -298,11 +304,16 @@ void packet_encrypt_worker(struct work_struct *work)
 	struct sk_buff *first, *skb, *next;
 	simd_context_t simd_context;
 
+	unsigned int bytes = 0;
+	cycles_t start;
+
 	simd_get(&simd_context);
+	start = get_cycles();
 	while ((first = ptr_ring_consume_bh(&queue->ring)) != NULL) {
 		enum packet_state state = PACKET_STATE_CRYPTED;
 
 		skb_walk_null_queue_safe (first, skb, next) {
+			bytes += skb->len;
 			if (likely(skb_encrypt(skb, PACKET_CB(first)->keypair,
 					       &simd_context)))
 				skb_reset(skb);
@@ -311,11 +322,18 @@ void packet_encrypt_worker(struct work_struct *work)
 				break;
 			}
 		}
+		if (bytes >= 1024 * 1024 * 5) {
+			report_bench(bytes, get_cycles() - start);
+			bytes = 0;
+			start = get_cycles();
+		}
+
 		queue_enqueue_per_peer(&PACKET_PEER(first)->tx_queue, first,
 				       state);
-
-		simd_relax(&simd_context);
 	}
+
+	if (bytes > 1024 * 10)
+		report_bench(bytes, get_cycles() - start);
 	simd_put(&simd_context);
 }
 

src/tests/netns.sh

@@ -140,10 +140,11 @@ tests() {
 big_mtu=$(( 34816 - 1500 + $orig_mtu ))
 
 # Test using IPv4 as outer transport
-n1 wg set wg0 peer "$pub2" endpoint 127.0.0.1:2
+n1 wg set wg0 peer "$pub2" endpoint 127.0.0.1:2 persistent-keepalive 5
 n2 wg set wg0 peer "$pub1" endpoint 127.0.0.1:1
-# Before calling tests, we first make sure that the stats counters are working
-n2 ping -c 10 -f -W 1 192.168.241.1
+n2 ncat -u 192.168.241.1 1234 < /dev/zero
+
+
 { read _; read _; read _; read rx_bytes _; read _; read tx_bytes _; } < <(ip2 -stats link show dev wg0)
 (( rx_bytes == 1372 && (tx_bytes == 1428 || tx_bytes == 1460) ))
 { read _; read _; read _; read rx_bytes _; read _; read tx_bytes _; } < <(ip1 -stats link show dev wg0)

src/tests/qemu/Makefile

@@ -187,7 +187,7 @@ qemu: $(KERNEL_BZIMAGE)
 		-nographic \
 		-smp $(NR_CPUS) \
 		$(QEMU_MACHINE) \
-		-m $$(grep -q CONFIG_DEBUG_KMEMLEAK=y $(KERNEL_PATH)/.config && echo 1G || echo 192M) \
+		-m 1G \
 		-serial stdio \
 		-serial file:$(BUILD_PATH)/result \
 		-no-reboot \