1/*-
2 *   BSD LICENSE
3 *
4 *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 *   All rights reserved.
6 *
7 *   Redistribution and use in source and binary forms, with or without
8 *   modification, are permitted provided that the following conditions
9 *   are met:
10 *
11 *     * Redistributions of source code must retain the above copyright
12 *       notice, this list of conditions and the following disclaimer.
13 *     * Redistributions in binary form must reproduce the above copyright
14 *       notice, this list of conditions and the following disclaimer in
15 *       the documentation and/or other materials provided with the
16 *       distribution.
17 *     * Neither the name of Intel Corporation nor the names of its
18 *       contributors may be used to endorse or promote products derived
19 *       from this software without specific prior written permission.
20 *
21 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34#include "test.h"
35
36#include <unistd.h>
37#include <string.h>
38#include <rte_mempool.h>
39#include <rte_cycles.h>
40#include <rte_common.h>
41#include <rte_mbuf.h>
42#include <rte_distributor.h>
43
44#define ITER_POWER 20 /* log 2 of how many iterations we do when timing. */
45#define BURST 32
46#define BIG_BATCH 1024
47
48/* static vars - zero initialized by default */
49static volatile int quit;
50static volatile unsigned worker_idx;
51
52struct worker_stats {
53	volatile unsigned handled_packets;
54} __rte_cache_aligned;
55struct worker_stats worker_stats[RTE_MAX_LCORE];
56
57/* worker thread used for testing the time to do a round-trip of a cache
58 * line between two cores and back again
59 */
60static int
61flip_bit(volatile uint64_t *arg)
62{
63	uint64_t old_val = 0;
64	while (old_val != 2) {
65		while (!*arg)
66			rte_pause();
67		old_val = *arg;
68		*arg = 0;
69	}
70	return 0;
71}
72
73/* test case to time the number of cycles to round-trip a cache line between
74 * two cores and back again.
75 */
76static void
77time_cache_line_switch(void)
78{
79	/* allocate a full cache line for data, we use only first byte of it */
80	uint64_t data[RTE_CACHE_LINE_SIZE*3 / sizeof(uint64_t)];
81
82	unsigned i, slaveid = rte_get_next_lcore(rte_lcore_id(), 0, 0);
83	volatile uint64_t *pdata = &data[0];
84	*pdata = 1;
85	rte_eal_remote_launch((lcore_function_t *)flip_bit, &data[0], slaveid);
86	while (*pdata)
87		rte_pause();
88
89	const uint64_t start_time = rte_rdtsc();
90	for (i = 0; i < (1 << ITER_POWER); i++) {
91		while (*pdata)
92			rte_pause();
93		*pdata = 1;
94	}
95	const uint64_t end_time = rte_rdtsc();
96
97	while (*pdata)
98		rte_pause();
99	*pdata = 2;
100	rte_eal_wait_lcore(slaveid);
101	printf("==== Cache line switch test ===\n");
102	printf("Time for %u iterations = %"PRIu64" ticks\n", (1<<ITER_POWER),
103			end_time-start_time);
104	printf("Ticks per iteration = %"PRIu64"\n\n",
105			(end_time-start_time) >> ITER_POWER);
106}
107
108/* returns the total count of the number of packets handled by the worker
109 * functions given below.
110 */
111static unsigned
112total_packet_count(void)
113{
114	unsigned i, count = 0;
115	for (i = 0; i < worker_idx; i++)
116		count += worker_stats[i].handled_packets;
117	return count;
118}
119
120/* resets the packet counts for a new test */
121static void
122clear_packet_count(void)
123{
124	memset(&worker_stats, 0, sizeof(worker_stats));
125}
126
127/* this is the basic worker function for performance tests.
128 * it does nothing but return packets and count them.
129 */
130static int
131handle_work(void *arg)
132{
133	struct rte_mbuf *pkt = NULL;
134	struct rte_distributor *d = arg;
135	unsigned count = 0;
136	unsigned id = __sync_fetch_and_add(&worker_idx, 1);
137
138	pkt = rte_distributor_get_pkt(d, id, NULL);
139	while (!quit) {
140		worker_stats[id].handled_packets++, count++;
141		pkt = rte_distributor_get_pkt(d, id, pkt);
142	}
143	worker_stats[id].handled_packets++, count++;
144	rte_distributor_return_pkt(d, id, pkt);
145	return 0;
146}
147
148/* this basic performance test just repeatedly sends in 32 packets at a time
149 * to the distributor and verifies at the end that we got them all in the worker
150 * threads and finally how long per packet the processing took.
151 */
152static inline int
153perf_test(struct rte_distributor *d, struct rte_mempool *p)
154{
155	unsigned i;
156	uint64_t start, end;
157	struct rte_mbuf *bufs[BURST];
158
159	clear_packet_count();
160	if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
161		printf("Error getting mbufs from pool\n");
162		return -1;
163	}
164	/* ensure we have different hash value for each pkt */
165	for (i = 0; i < BURST; i++)
166		bufs[i]->hash.usr = i;
167
168	start = rte_rdtsc();
169	for (i = 0; i < (1<<ITER_POWER); i++)
170		rte_distributor_process(d, bufs, BURST);
171	end = rte_rdtsc();
172
173	do {
174		usleep(100);
175		rte_distributor_process(d, NULL, 0);
176	} while (total_packet_count() < (BURST << ITER_POWER));
177
178	printf("=== Performance test of distributor ===\n");
179	printf("Time per burst:  %"PRIu64"\n", (end - start) >> ITER_POWER);
180	printf("Time per packet: %"PRIu64"\n\n",
181			((end - start) >> ITER_POWER)/BURST);
182	rte_mempool_put_bulk(p, (void *)bufs, BURST);
183
184	for (i = 0; i < rte_lcore_count() - 1; i++)
185		printf("Worker %u handled %u packets\n", i,
186				worker_stats[i].handled_packets);
187	printf("Total packets: %u (%x)\n", total_packet_count(),
188			total_packet_count());
189	printf("=== Perf test done ===\n\n");
190
191	return 0;
192}
193
194/* Useful function which ensures that all worker functions terminate */
195static void
196quit_workers(struct rte_distributor *d, struct rte_mempool *p)
197{
198	const unsigned num_workers = rte_lcore_count() - 1;
199	unsigned i;
200	struct rte_mbuf *bufs[RTE_MAX_LCORE];
201	rte_mempool_get_bulk(p, (void *)bufs, num_workers);
202
203	quit = 1;
204	for (i = 0; i < num_workers; i++)
205		bufs[i]->hash.usr = i << 1;
206	rte_distributor_process(d, bufs, num_workers);
207
208	rte_mempool_put_bulk(p, (void *)bufs, num_workers);
209
210	rte_distributor_process(d, NULL, 0);
211	rte_eal_mp_wait_lcore();
212	quit = 0;
213	worker_idx = 0;
214}
215
216static int
217test_distributor_perf(void)
218{
219	static struct rte_distributor *d;
220	static struct rte_mempool *p;
221
222	if (rte_lcore_count() < 2) {
223		printf("ERROR: not enough cores to test distributor\n");
224		return -1;
225	}
226
227	/* first time how long it takes to round-trip a cache line */
228	time_cache_line_switch();
229
230	if (d == NULL) {
231		d = rte_distributor_create("Test_perf", rte_socket_id(),
232				rte_lcore_count() - 1);
233		if (d == NULL) {
234			printf("Error creating distributor\n");
235			return -1;
236		}
237	} else {
238		rte_distributor_flush(d);
239		rte_distributor_clear_returns(d);
240	}
241
242	const unsigned nb_bufs = (511 * rte_lcore_count()) < BIG_BATCH ?
243			(BIG_BATCH * 2) - 1 : (511 * rte_lcore_count());
244	if (p == NULL) {
245		p = rte_pktmbuf_pool_create("DPT_MBUF_POOL", nb_bufs, BURST,
246			0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
247		if (p == NULL) {
248			printf("Error creating mempool\n");
249			return -1;
250		}
251	}
252
253	rte_eal_mp_remote_launch(handle_work, d, SKIP_MASTER);
254	if (perf_test(d, p) < 0)
255		return -1;
256	quit_workers(d, p);
257
258	return 0;
259}
260
261REGISTER_TEST_COMMAND(distributor_perf_autotest, test_distributor_perf);
262