97f4a452da12e045c18b32c0cddc15d422cd14ea
[dpdk.git] / app / test-pmd / txonly.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4
5 #include <stdarg.h>
6 #include <string.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <stdint.h>
10 #include <unistd.h>
11 #include <inttypes.h>
12
13 #include <sys/queue.h>
14 #include <sys/stat.h>
15
16 #include <rte_common.h>
17 #include <rte_byteorder.h>
18 #include <rte_log.h>
19 #include <rte_debug.h>
20 #include <rte_cycles.h>
21 #include <rte_memory.h>
22 #include <rte_memcpy.h>
23 #include <rte_launch.h>
24 #include <rte_eal.h>
25 #include <rte_per_lcore.h>
26 #include <rte_lcore.h>
27 #include <rte_atomic.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_mempool.h>
30 #include <rte_mbuf.h>
31 #include <rte_interrupts.h>
32 #include <rte_pci.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_ip.h>
36 #include <rte_tcp.h>
37 #include <rte_udp.h>
38 #include <rte_string_fns.h>
39 #include <rte_flow.h>
40
41 #include "testpmd.h"
42
43 /* use RFC863 Discard Protocol */
44 uint16_t tx_udp_src_port = 9;
45 uint16_t tx_udp_dst_port = 9;
46
47 /* use RFC5735 / RFC2544 reserved network test addresses */
48 uint32_t tx_ip_src_addr = (198U << 24) | (18 << 16) | (0 << 8) | 1;
49 uint32_t tx_ip_dst_addr = (198U << 24) | (18 << 16) | (0 << 8) | 2;
50
51 #define IP_DEFTTL  64   /* from RFC 1340. */
52
53 static struct rte_ipv4_hdr pkt_ip_hdr; /**< IP header of transmitted packets. */
54 RTE_DEFINE_PER_LCORE(uint8_t, _ip_var); /**< IP address variation */
55 static struct rte_udp_hdr pkt_udp_hdr; /**< UDP header of tx packets. */
56 RTE_DEFINE_PER_LCORE(uint64_t, timestamp_qskew);
57                                         /**< Timestamp offset per queue */
58 static uint64_t timestamp_mask; /**< Timestamp dynamic flag mask */
59 static int32_t timestamp_off; /**< Timestamp dynamic field offset */
60 static bool timestamp_enable; /**< Timestamp enable */
61 static uint64_t timestamp_initial[RTE_MAX_ETHPORTS];
62
63 static void
64 copy_buf_to_pkt_segs(void* buf, unsigned len, struct rte_mbuf *pkt,
65                      unsigned offset)
66 {
67         struct rte_mbuf *seg;
68         void *seg_buf;
69         unsigned copy_len;
70
71         seg = pkt;
72         while (offset >= seg->data_len) {
73                 offset -= seg->data_len;
74                 seg = seg->next;
75         }
76         copy_len = seg->data_len - offset;
77         seg_buf = rte_pktmbuf_mtod_offset(seg, char *, offset);
78         while (len > copy_len) {
79                 rte_memcpy(seg_buf, buf, (size_t) copy_len);
80                 len -= copy_len;
81                 buf = ((char*) buf + copy_len);
82                 seg = seg->next;
83                 seg_buf = rte_pktmbuf_mtod(seg, char *);
84                 copy_len = seg->data_len;
85         }
86         rte_memcpy(seg_buf, buf, (size_t) len);
87 }
88
89 static inline void
90 copy_buf_to_pkt(void* buf, unsigned len, struct rte_mbuf *pkt, unsigned offset)
91 {
92         if (offset + len <= pkt->data_len) {
93                 rte_memcpy(rte_pktmbuf_mtod_offset(pkt, char *, offset),
94                         buf, (size_t) len);
95                 return;
96         }
97         copy_buf_to_pkt_segs(buf, len, pkt, offset);
98 }
99
100 static void
101 setup_pkt_udp_ip_headers(struct rte_ipv4_hdr *ip_hdr,
102                          struct rte_udp_hdr *udp_hdr,
103                          uint16_t pkt_data_len)
104 {
105         uint16_t *ptr16;
106         uint32_t ip_cksum;
107         uint16_t pkt_len;
108
109         /*
110          * Initialize UDP header.
111          */
112         pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr));
113         udp_hdr->src_port = rte_cpu_to_be_16(tx_udp_src_port);
114         udp_hdr->dst_port = rte_cpu_to_be_16(tx_udp_dst_port);
115         udp_hdr->dgram_len      = RTE_CPU_TO_BE_16(pkt_len);
116         udp_hdr->dgram_cksum    = 0; /* No UDP checksum. */
117
118         /*
119          * Initialize IP header.
120          */
121         pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr));
122         ip_hdr->version_ihl   = RTE_IPV4_VHL_DEF;
123         ip_hdr->type_of_service   = 0;
124         ip_hdr->fragment_offset = 0;
125         ip_hdr->time_to_live   = IP_DEFTTL;
126         ip_hdr->next_proto_id = IPPROTO_UDP;
127         ip_hdr->packet_id = 0;
128         ip_hdr->total_length   = RTE_CPU_TO_BE_16(pkt_len);
129         ip_hdr->src_addr = rte_cpu_to_be_32(tx_ip_src_addr);
130         ip_hdr->dst_addr = rte_cpu_to_be_32(tx_ip_dst_addr);
131
132         /*
133          * Compute IP header checksum.
134          */
135         ptr16 = (unaligned_uint16_t*) ip_hdr;
136         ip_cksum = 0;
137         ip_cksum += ptr16[0]; ip_cksum += ptr16[1];
138         ip_cksum += ptr16[2]; ip_cksum += ptr16[3];
139         ip_cksum += ptr16[4];
140         ip_cksum += ptr16[6]; ip_cksum += ptr16[7];
141         ip_cksum += ptr16[8]; ip_cksum += ptr16[9];
142
143         /*
144          * Reduce 32 bit checksum to 16 bits and complement it.
145          */
146         ip_cksum = ((ip_cksum & 0xFFFF0000) >> 16) +
147                 (ip_cksum & 0x0000FFFF);
148         if (ip_cksum > 65535)
149                 ip_cksum -= 65535;
150         ip_cksum = (~ip_cksum) & 0x0000FFFF;
151         if (ip_cksum == 0)
152                 ip_cksum = 0xFFFF;
153         ip_hdr->hdr_checksum = (uint16_t) ip_cksum;
154 }
155
156 static inline bool
157 pkt_burst_prepare(struct rte_mbuf *pkt, struct rte_mempool *mbp,
158                 struct rte_ether_hdr *eth_hdr, const uint16_t vlan_tci,
159                 const uint16_t vlan_tci_outer, const uint64_t ol_flags,
160                 const uint16_t idx, const struct fwd_stream *fs)
161 {
162         struct rte_mbuf *pkt_segs[RTE_MAX_SEGS_PER_PKT];
163         struct rte_mbuf *pkt_seg;
164         uint32_t nb_segs, pkt_len;
165         uint8_t i;
166
167         if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND))
168                 nb_segs = rte_rand() % tx_pkt_nb_segs + 1;
169         else
170                 nb_segs = tx_pkt_nb_segs;
171
172         if (nb_segs > 1) {
173                 if (rte_mempool_get_bulk(mbp, (void **)pkt_segs, nb_segs - 1))
174                         return false;
175         }
176
177         rte_pktmbuf_reset_headroom(pkt);
178         pkt->data_len = tx_pkt_seg_lengths[0];
179         pkt->ol_flags &= EXT_ATTACHED_MBUF;
180         pkt->ol_flags |= ol_flags;
181         pkt->vlan_tci = vlan_tci;
182         pkt->vlan_tci_outer = vlan_tci_outer;
183         pkt->l2_len = sizeof(struct rte_ether_hdr);
184         pkt->l3_len = sizeof(struct rte_ipv4_hdr);
185
186         pkt_len = pkt->data_len;
187         pkt_seg = pkt;
188         for (i = 1; i < nb_segs; i++) {
189                 pkt_seg->next = pkt_segs[i - 1];
190                 pkt_seg = pkt_seg->next;
191                 pkt_seg->data_len = tx_pkt_seg_lengths[i];
192                 pkt_len += pkt_seg->data_len;
193         }
194         pkt_seg->next = NULL; /* Last segment of packet. */
195         /*
196          * Copy headers in first packet segment(s).
197          */
198         copy_buf_to_pkt(eth_hdr, sizeof(*eth_hdr), pkt, 0);
199         copy_buf_to_pkt(&pkt_ip_hdr, sizeof(pkt_ip_hdr), pkt,
200                         sizeof(struct rte_ether_hdr));
201         if (txonly_multi_flow) {
202                 uint8_t  ip_var = RTE_PER_LCORE(_ip_var);
203                 struct rte_ipv4_hdr *ip_hdr;
204                 uint32_t addr;
205
206                 ip_hdr = rte_pktmbuf_mtod_offset(pkt,
207                                 struct rte_ipv4_hdr *,
208                                 sizeof(struct rte_ether_hdr));
209                 /*
210                  * Generate multiple flows by varying IP src addr. This
211                  * enables packets are well distributed by RSS in
212                  * receiver side if any and txonly mode can be a decent
213                  * packet generator for developer's quick performance
214                  * regression test.
215                  */
216                 addr = (tx_ip_dst_addr | (ip_var++ << 8)) + rte_lcore_id();
217                 ip_hdr->src_addr = rte_cpu_to_be_32(addr);
218                 RTE_PER_LCORE(_ip_var) = ip_var;
219         }
220         copy_buf_to_pkt(&pkt_udp_hdr, sizeof(pkt_udp_hdr), pkt,
221                         sizeof(struct rte_ether_hdr) +
222                         sizeof(struct rte_ipv4_hdr));
223         if (unlikely(timestamp_enable)) {
224                 uint64_t skew = RTE_PER_LCORE(timestamp_qskew);
225                 struct {
226                         rte_be32_t signature;
227                         rte_be16_t pkt_idx;
228                         rte_be16_t queue_idx;
229                         rte_be64_t ts;
230                 } timestamp_mark;
231
232                 if (unlikely(!skew)) {
233                         struct rte_eth_dev *dev = &rte_eth_devices[fs->tx_port];
234                         unsigned int txqs_n = dev->data->nb_tx_queues;
235                         uint64_t phase = tx_pkt_times_inter * fs->tx_queue /
236                                          (txqs_n ? txqs_n : 1);
237                         /*
238                          * Initialize the scheduling time phase shift
239                          * depending on queue index.
240                          */
241                         skew = timestamp_initial[fs->tx_port] +
242                                tx_pkt_times_inter + phase;
243                         RTE_PER_LCORE(timestamp_qskew) = skew;
244                 }
245                 timestamp_mark.pkt_idx = rte_cpu_to_be_16(idx);
246                 timestamp_mark.queue_idx = rte_cpu_to_be_16(fs->tx_queue);
247                 timestamp_mark.signature = rte_cpu_to_be_32(0xBEEFC0DE);
248                 if (unlikely(!idx)) {
249                         skew += tx_pkt_times_inter;
250                         pkt->ol_flags |= timestamp_mask;
251                         *RTE_MBUF_DYNFIELD
252                                 (pkt, timestamp_off, uint64_t *) = skew;
253                         RTE_PER_LCORE(timestamp_qskew) = skew;
254                         timestamp_mark.ts = rte_cpu_to_be_64(skew);
255                 } else if (tx_pkt_times_intra) {
256                         skew += tx_pkt_times_intra;
257                         pkt->ol_flags |= timestamp_mask;
258                         *RTE_MBUF_DYNFIELD
259                                 (pkt, timestamp_off, uint64_t *) = skew;
260                         RTE_PER_LCORE(timestamp_qskew) = skew;
261                         timestamp_mark.ts = rte_cpu_to_be_64(skew);
262                 } else {
263                         timestamp_mark.ts = RTE_BE64(0);
264                 }
265                 copy_buf_to_pkt(&timestamp_mark, sizeof(timestamp_mark), pkt,
266                         sizeof(struct rte_ether_hdr) +
267                         sizeof(struct rte_ipv4_hdr) +
268                         sizeof(pkt_udp_hdr));
269         }
270         /*
271          * Complete first mbuf of packet and append it to the
272          * burst of packets to be transmitted.
273          */
274         pkt->nb_segs = nb_segs;
275         pkt->pkt_len = pkt_len;
276
277         return true;
278 }
279
280 /*
281  * Transmit a burst of multi-segments packets.
282  */
283 static void
284 pkt_burst_transmit(struct fwd_stream *fs)
285 {
286         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
287         struct rte_port *txp;
288         struct rte_mbuf *pkt;
289         struct rte_mempool *mbp;
290         struct rte_ether_hdr eth_hdr;
291         uint16_t nb_tx;
292         uint16_t nb_pkt;
293         uint16_t vlan_tci, vlan_tci_outer;
294         uint32_t retry;
295         uint64_t ol_flags = 0;
296         uint64_t tx_offloads;
297 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
298         uint64_t start_tsc;
299         uint64_t end_tsc;
300         uint64_t core_cycles;
301 #endif
302
303 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
304         start_tsc = rte_rdtsc();
305 #endif
306
307         mbp = current_fwd_lcore()->mbp;
308         txp = &ports[fs->tx_port];
309         tx_offloads = txp->dev_conf.txmode.offloads;
310         vlan_tci = txp->tx_vlan_id;
311         vlan_tci_outer = txp->tx_vlan_id_outer;
312         if (tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT)
313                 ol_flags = PKT_TX_VLAN_PKT;
314         if (tx_offloads & DEV_TX_OFFLOAD_QINQ_INSERT)
315                 ol_flags |= PKT_TX_QINQ_PKT;
316         if (tx_offloads & DEV_TX_OFFLOAD_MACSEC_INSERT)
317                 ol_flags |= PKT_TX_MACSEC;
318
319         /*
320          * Initialize Ethernet header.
321          */
322         rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr], &eth_hdr.d_addr);
323         rte_ether_addr_copy(&ports[fs->tx_port].eth_addr, &eth_hdr.s_addr);
324         eth_hdr.ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
325
326         if (rte_mempool_get_bulk(mbp, (void **)pkts_burst,
327                                 nb_pkt_per_burst) == 0) {
328                 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) {
329                         if (unlikely(!pkt_burst_prepare(pkts_burst[nb_pkt], mbp,
330                                                         &eth_hdr, vlan_tci,
331                                                         vlan_tci_outer,
332                                                         ol_flags,
333                                                         nb_pkt, fs))) {
334                                 rte_mempool_put_bulk(mbp,
335                                                 (void **)&pkts_burst[nb_pkt],
336                                                 nb_pkt_per_burst - nb_pkt);
337                                 break;
338                         }
339                 }
340         } else {
341                 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) {
342                         pkt = rte_mbuf_raw_alloc(mbp);
343                         if (pkt == NULL)
344                                 break;
345                         if (unlikely(!pkt_burst_prepare(pkt, mbp, &eth_hdr,
346                                                         vlan_tci,
347                                                         vlan_tci_outer,
348                                                         ol_flags,
349                                                         nb_pkt, fs))) {
350                                 rte_pktmbuf_free(pkt);
351                                 break;
352                         }
353                         pkts_burst[nb_pkt] = pkt;
354                 }
355         }
356
357         if (nb_pkt == 0)
358                 return;
359
360         nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_pkt);
361
362         /*
363          * Retry if necessary
364          */
365         if (unlikely(nb_tx < nb_pkt) && fs->retry_enabled) {
366                 retry = 0;
367                 while (nb_tx < nb_pkt && retry++ < burst_tx_retry_num) {
368                         rte_delay_us(burst_tx_delay_time);
369                         nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
370                                         &pkts_burst[nb_tx], nb_pkt - nb_tx);
371                 }
372         }
373         fs->tx_packets += nb_tx;
374
375         if (txonly_multi_flow)
376                 RTE_PER_LCORE(_ip_var) -= nb_pkt - nb_tx;
377
378 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
379         fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
380 #endif
381         if (unlikely(nb_tx < nb_pkt)) {
382                 if (verbose_level > 0 && fs->fwd_dropped == 0)
383                         printf("port %d tx_queue %d - drop "
384                                "(nb_pkt:%u - nb_tx:%u)=%u packets\n",
385                                fs->tx_port, fs->tx_queue,
386                                (unsigned) nb_pkt, (unsigned) nb_tx,
387                                (unsigned) (nb_pkt - nb_tx));
388                 fs->fwd_dropped += (nb_pkt - nb_tx);
389                 do {
390                         rte_pktmbuf_free(pkts_burst[nb_tx]);
391                 } while (++nb_tx < nb_pkt);
392         }
393
394 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
395         end_tsc = rte_rdtsc();
396         core_cycles = (end_tsc - start_tsc);
397         fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
398 #endif
399 }
400
401 static void
402 tx_only_begin(portid_t pi)
403 {
404         uint16_t pkt_data_len;
405         int dynf;
406
407         pkt_data_len = (uint16_t) (tx_pkt_length - (
408                                         sizeof(struct rte_ether_hdr) +
409                                         sizeof(struct rte_ipv4_hdr) +
410                                         sizeof(struct rte_udp_hdr)));
411         setup_pkt_udp_ip_headers(&pkt_ip_hdr, &pkt_udp_hdr, pkt_data_len);
412
413         timestamp_enable = false;
414         timestamp_mask = 0;
415         timestamp_off = -1;
416         RTE_PER_LCORE(timestamp_qskew) = 0;
417         dynf = rte_mbuf_dynflag_lookup
418                                 (RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME, NULL);
419         if (dynf >= 0)
420                 timestamp_mask = 1ULL << dynf;
421         dynf = rte_mbuf_dynfield_lookup
422                                 (RTE_MBUF_DYNFIELD_TIMESTAMP_NAME, NULL);
423         if (dynf >= 0)
424                 timestamp_off = dynf;
425         timestamp_enable = tx_pkt_times_inter &&
426                            timestamp_mask &&
427                            timestamp_off >= 0 &&
428                            !rte_eth_read_clock(pi, &timestamp_initial[pi]);
429 }
430
431 struct fwd_engine tx_only_engine = {
432         .fwd_mode_name  = "txonly",
433         .port_fwd_begin = tx_only_begin,
434         .port_fwd_end   = NULL,
435         .packet_fwd     = pkt_burst_transmit,
436 };