blob: 43cea94c787b9b4445fcacccb4deda969ea768ab [file] [log] [blame]
Zhimin2fbf7002020-05-29 06:01:04 +08001/******************************************************************************
2*
3* Copyright (c) 2019 Intel.
4*
5* Licensed under the Apache License, Version 2.0 (the "License");
6* you may not use this file except in compliance with the License.
7* You may obtain a copy of the License at
8*
9* http://www.apache.org/licenses/LICENSE-2.0
10*
11* Unless required by applicable law or agreed to in writing, software
12* distributed under the License is distributed on an "AS IS" BASIS,
13* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14* See the License for the specific language governing permissions and
15* limitations under the License.
16*
17*******************************************************************************/
18
19/**
20 * @brief This file has all definitions for the Ethernet Data Interface Layer
21 * @file ethernet.c
22 * @ingroup group_lte_source_auxlib
23 * @author Intel Corporation
24 **/
25
26
27#include <stdio.h>
28#include <string.h>
29#include <stdint.h>
30#include <unistd.h>
31#include <errno.h>
32#include <sys/queue.h>
33#include <err.h>
34#include <assert.h>
35
36#include <linux/limits.h>
37#include <sys/types.h>
38#include <stdlib.h>
39#include <math.h>
40
41#include <rte_config.h>
42#include <rte_common.h>
43#include <rte_log.h>
44#include <rte_memory.h>
45#include <rte_memcpy.h>
46#include <rte_memzone.h>
47#include <rte_eal.h>
48#include <rte_per_lcore.h>
49#include <rte_launch.h>
50#include <rte_atomic.h>
51#include <rte_cycles.h>
52#include <rte_prefetch.h>
53#include <rte_lcore.h>
54#include <rte_per_lcore.h>
55#include <rte_branch_prediction.h>
56#include <rte_interrupts.h>
57#include <rte_pci.h>
58#include <rte_debug.h>
59#include <rte_ether.h>
60#include <rte_ethdev.h>
61#include <rte_ring.h>
62#include <rte_mempool.h>
63#include <rte_mbuf.h>
64#include <rte_errno.h>
65
66#include "ethernet.h"
67#include "ethdi.h"
68
69/* Our mbuf pools. */
70struct rte_mempool *_eth_mbuf_pool = NULL;
71struct rte_mempool *_eth_mbuf_pool_inderect = NULL;
72struct rte_mempool *_eth_mbuf_pool_rx = NULL;
73struct rte_mempool *_eth_mbuf_pool_small = NULL;
74struct rte_mempool *_eth_mbuf_pool_big = NULL;
75
76struct rte_mempool *socket_direct_pool = NULL;
77struct rte_mempool *socket_indirect_pool = NULL;
78
79
80/*
81 * Make sure the ring indexes are big enough to cover buf space x2
82 * This ring-buffer maintains the property head - tail <= RINGSIZE.
83 * head == tail: ring buffer empty
84 * head - tail == RINGSIZE: ring buffer full
85 */
86typedef uint16_t ring_idx;
87static struct {
88 ring_idx head;
89 ring_idx read_head;
90 ring_idx tail;
91 char buf[1024]; /* needs power of 2! */
92} io_ring = { {0}, 0, 0};
93
94#define RINGSIZE sizeof(io_ring.buf)
95#define RINGMASK (RINGSIZE - 1)
96
97int __xran_delayed_msg(const char *fmt, ...)
98{
99#if 0
100 va_list ap;
101 int msg_len;
102 char localbuf[RINGSIZE];
103 ring_idx old_head, new_head;
104 ring_idx copy_len;
105
106 /* first prep a copy of the message on the local stack */
107 va_start(ap, fmt);
108 msg_len = vsnprintf(localbuf, RINGSIZE, fmt, ap);
109 va_end(ap);
110
111 /* atomically reserve space in the ring */
112 for (;;) {
113 old_head = io_ring.head; /* snapshot head */
114 /* free always within range of [0, RINGSIZE] - proof by induction */
115 const ring_idx free = RINGSIZE - (old_head - io_ring.tail);
116
117 copy_len = RTE_MIN(msg_len, free);
118 if (copy_len <= 0)
119 return 0; /* vsnprintf error or ringbuff full. Drop log. */
120
121 new_head = old_head + copy_len;
122 RTE_ASSERT((ring_idx)(new_head - io_ring.tail) <= RINGSIZE);
123
124 if (likely(__atomic_compare_exchange_n(&io_ring.head, &old_head,
125 new_head, 0, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)))
126 break;
127 }
128
129 /* Now copy data in at ease. */
130 const int copy_start = (old_head & RINGMASK);
131 if (copy_start < (new_head & RINGMASK)) /* no wrap */
132 memcpy(io_ring.buf + copy_start, localbuf, copy_len);
133 else { /* wrap-around */
134 const int chunk_len = RINGSIZE - copy_start;
135
136 memcpy(io_ring.buf + copy_start, localbuf, chunk_len);
137 memcpy(io_ring.buf, localbuf + chunk_len, copy_len - chunk_len);
138 }
139
140 /* wait for previous writes to complete before updating read_head. */
141 while (io_ring.read_head != old_head)
142 rte_pause();
143 io_ring.read_head = new_head;
144
145
146 return copy_len;
147 #endif
148 return 0;
149}
150
151/*
152 * Display part of the message stored in the ring buffer.
153 * Might require multiple calls to print the full message.
154 * Will return 0 when nothing left to print.
155 */
156#if 0
157int xran_show_delayed_message(void)
158{
159 ring_idx tail = io_ring.tail;
160 ring_idx wlen = io_ring.read_head - tail; /* always within [0, RINGSIZE] */
161
162 if (wlen <= 0)
163 return 0;
164
165 tail &= RINGMASK; /* modulo the range down now that we have wlen */
166
167 /* Make sure we're not going over buffer end. Next call will wrap. */
168 if (tail + wlen > RINGSIZE)
169 wlen = RINGSIZE - tail;
170
171 RTE_ASSERT(tail + wlen <= RINGSIZE);
172
173 /* We use write() here to avoid recaculating string length in fwrite(). */
174 const ssize_t written = write(STDOUT_FILENO, io_ring.buf + tail, wlen);
175 if (written <= 0)
176 return 0; /* To avoid moving tail the wrong way on error. */
177
178 /* Move tail up. Only we touch it. And we only print from one core. */
179 io_ring.tail += written;
180
181 return written; /* next invocation will print the rest if any */
182}
183#endif
184
185void xran_init_mbuf_pool(void)
186{
187 /* Init the buffer pool */
188 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
189 _eth_mbuf_pool = rte_pktmbuf_pool_create("mempool", NUM_MBUFS,
190 MBUF_CACHE, 0, MBUF_POOL_ELEMENT, rte_socket_id());
191#ifdef XRAN_ATTACH_MBUF
192 _eth_mbuf_pool_inderect = rte_pktmbuf_pool_create("mempool_indirect", NUM_MBUFS,
193 MBUF_CACHE, 0, MBUF_POOL_ELEMENT, rte_socket_id());*/
194#endif
195 _eth_mbuf_pool_rx = rte_pktmbuf_pool_create("mempool_rx", NUM_MBUFS,
196 MBUF_CACHE, 0, MBUF_POOL_ELEMENT, rte_socket_id());
197 _eth_mbuf_pool_small = rte_pktmbuf_pool_create("mempool_small",
198 NUM_MBUFS, MBUF_CACHE, 0, MBUF_POOL_ELM_SMALL, rte_socket_id());
199 _eth_mbuf_pool_big = rte_pktmbuf_pool_create("mempool_big",
200 NUM_MBUFS_BIG, 0, 0, MBUF_POOL_ELM_BIG, rte_socket_id());
201 } else {
202 _eth_mbuf_pool = rte_mempool_lookup("mempool");
203 _eth_mbuf_pool_inderect = rte_mempool_lookup("mempool_indirect");
204 _eth_mbuf_pool_rx = rte_mempool_lookup("mempool_rx");
205 _eth_mbuf_pool_small = rte_mempool_lookup("mempool_small");
206 _eth_mbuf_pool_big = rte_mempool_lookup("mempool_big");
207 }
208 if (_eth_mbuf_pool == NULL)
209 rte_panic("Cannot create mbuf pool: %s\n", rte_strerror(rte_errno));
210#ifdef XRAN_ATTACH_MBUF
211 if (_eth_mbuf_pool_inderect == NULL)
212 rte_panic("Cannot create mbuf pool: %s\n", rte_strerror(rte_errno));
213#endif
214 if (_eth_mbuf_pool_rx == NULL)
215 rte_panic("Cannot create mbuf pool: %s\n", rte_strerror(rte_errno));
216 if (_eth_mbuf_pool_small == NULL)
217 rte_panic("Cannot create small mbuf pool: %s\n", rte_strerror(rte_errno));
218 if (_eth_mbuf_pool_big == NULL)
219 rte_panic("Cannot create big mbuf pool: %s\n", rte_strerror(rte_errno));
220
221 if (socket_direct_pool == NULL)
222 socket_direct_pool = _eth_mbuf_pool;
223
224 if (socket_indirect_pool == NULL)
225 socket_indirect_pool = _eth_mbuf_pool_inderect;
226}
227
228/* Init NIC port, then start the port */
229void xran_init_port(int p_id, struct ether_addr *p_lls_cu_addr)
230{
231 static uint16_t nb_rxd = BURST_SIZE;
232 static uint16_t nb_txd = BURST_SIZE;
233 struct ether_addr addr;
234 struct rte_eth_rxmode rxmode =
235 { .split_hdr_size = 0,
236 .max_rx_pkt_len = MAX_RX_LEN,
237 .offloads=(DEV_RX_OFFLOAD_JUMBO_FRAME|DEV_RX_OFFLOAD_CRC_STRIP)
238 };
239 struct rte_eth_txmode txmode = {
240 .mq_mode = ETH_MQ_TX_NONE
241 };
242 struct rte_eth_conf port_conf = {
243 .rxmode = rxmode,
244 .txmode = txmode
245 };
246 struct rte_eth_rxconf rxq_conf;
247 struct rte_eth_txconf txq_conf;
248
249 int ret;
250 struct rte_eth_dev_info dev_info;
251 const char *drv_name = "";
252 int sock_id = rte_eth_dev_socket_id(p_id);
253
254 rte_eth_dev_info_get(p_id, &dev_info);
255 if (dev_info.driver_name)
256 drv_name = dev_info.driver_name;
257 printf("initializing port %d for TX, drv=%s\n", p_id, drv_name);
258
259 rte_eth_macaddr_get(p_id, &addr);
260
261 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
262 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
263 (unsigned)p_id,
264 addr.addr_bytes[0], addr.addr_bytes[1], addr.addr_bytes[2],
265 addr.addr_bytes[3], addr.addr_bytes[4], addr.addr_bytes[5]);
266
267 /* Init port */
268 ret = rte_eth_dev_configure(p_id, 1, 1, &port_conf);
269 if (ret < 0)
270 rte_panic("Cannot configure port %u (%d)\n", p_id, ret);
271
272 ret = rte_eth_dev_adjust_nb_rx_tx_desc(p_id, &nb_rxd,&nb_txd);
273
274 if (ret < 0) {
275 printf("\n");
276 rte_exit(EXIT_FAILURE, "Cannot adjust number of "
277 "descriptors: err=%d, port=%d\n", ret, p_id);
278 }
279 printf("Port %u: nb_rxd %d nb_txd %d\n", p_id, nb_rxd, nb_txd);
280
281 /* Init RX queues */
282 rxq_conf = dev_info.default_rxconf;
283 ret = rte_eth_rx_queue_setup(p_id, 0, nb_rxd,
284 sock_id, &rxq_conf, _eth_mbuf_pool_rx);
285 if (ret < 0)
286 rte_panic("Cannot init RX for port %u (%d)\n",
287 p_id, ret);
288
289 /* Init TX queues */
290 txq_conf = dev_info.default_txconf;
291 ret = rte_eth_tx_queue_setup(p_id, 0, nb_txd, sock_id, &txq_conf);
292 if (ret < 0)
293 rte_panic("Cannot init TX for port %u (%d)\n",
294 p_id, ret);
295
296 /* Start port */
297 ret = rte_eth_dev_start(p_id);
298 if (ret < 0)
299 rte_panic("Cannot start port %u (%d)\n", p_id, ret);
300
301// rte_eth_promiscuous_enable(p_id);
302}
303
304#if 0
305void xran_memdump(void *addr, int len)
306{
307 int i;
308 char tmp_buf[len * 2 + len / 16 + 1];
309 char *p = tmp_buf;
310
311 return;
312#if 0
313 for (i = 0; i < len; ++i) {
314 sprintf(p, "%.2X ", ((uint8_t *)addr)[i]);
315 if (i % 16 == 15)
316 *p++ = '\n';
317 }
318 *p = 0;
319 nlog("%s", tmp_buf);
320#endif
321}
322
323/* Prepend ethernet header, possibly vlan tag. */
324void xran_add_eth_hdr(struct ether_addr *dst, uint16_t ethertype, struct rte_mbuf *mb)
325{
326 /* add in the ethernet header */
327 struct ether_hdr *const h = (void *)rte_pktmbuf_prepend(mb, sizeof(*h));
328
329 PANIC_ON(h == NULL, "mbuf prepend of ether_hdr failed");
330
331 /* Fill in the ethernet header. */
332 rte_eth_macaddr_get(mb->port, &h->s_addr); /* set source addr */
333 h->d_addr = *dst; /* set dst addr */
334 h->ether_type = rte_cpu_to_be_16(ethertype); /* ethertype too */
335
336#if defined(DPDKIO_DEBUG) && DPDKIO_DEBUG > 1
337 {
338 char dst[ETHER_ADDR_FMT_SIZE] = "(empty)";
339 char src[ETHER_ADDR_FMT_SIZE] = "(empty)";
340
341 nlog("*** packet for TX below (len %d) ***", rte_pktmbuf_pkt_len(mb));
342 ether_format_addr(src, sizeof(src), &h->s_addr);
343 ether_format_addr(dst, sizeof(dst), &h->d_addr);
344 nlog("src: %s dst: %s ethertype: %.4X", src, dst, ethertype);
345 }
346#endif
347#ifdef VLAN_SUPPORT
348 mb->vlan_tci = FLEXRAN_UP_VLAN_TAG;
349 dlog("Inserting vlan tag of %d", FLEXRAN_UP_VLAN_TAG);
350 rte_vlan_insert(&mb);
351#endif
352}
353
354int xran_send_mbuf(struct ether_addr *dst, struct rte_mbuf *mb)
355{
356 xran_add_eth_hdr(dst, ETHER_TYPE_ETHDI, mb);
357
358 if (rte_eth_tx_burst(mb->port, 0, &mb, 1) == 1)
359 return 1;
360
361 elog("packet sending failed on port %d", mb->port);
362 rte_pktmbuf_free(mb);
363
364 return 0; /* fail */
365}
366
367int xran_send_message_burst(int dst_id, int pkt_type, void *body, int len)
368{
369 struct rte_mbuf *mbufs[BURST_SIZE];
370 int i;
371 uint8_t *src = body;
372 const struct xran_ethdi_ctx *const ctx = xran_ethdi_get_ctx();
373
374 /* We're limited by maximum mbuf size on the receive size.
375 * We can change this but this would be a bigger rework. */
376 RTE_ASSERT(len < MBUF_POOL_ELM_BIG);
377
378 /* Allocate the required number of mbufs. */
379 const uint8_t count = ceilf((float)len / MAX_DATA_SIZE);
380 if (rte_pktmbuf_alloc_bulk(_eth_mbuf_pool, mbufs, count) != 0)
381 rte_panic("Failed to allocate %d mbufs\n", count);
382
383 nlog("burst transfer with data size %lu", MAX_DATA_SIZE);
384 for (i = 0; len > 0; ++i) {
385 char *p;
386 struct burst_hdr *bhdr;
387 struct ethdi_hdr *edi_hdr;
388
389 /* Setup the ethdi_hdr. */
390 edi_hdr = (void *)rte_pktmbuf_append(mbufs[i], sizeof(*edi_hdr));
391 if (edi_hdr == NULL)
392 rte_panic("append of ethdi_hdr failed\n");
393 edi_hdr->pkt_type = PKT_BURST;
394 /* edi_hdr->source_id setup in tx_from_ring */
395 edi_hdr->dest_id = dst_id;
396
397 /* Setup the burst header */
398 bhdr = (void *)rte_pktmbuf_append(mbufs[i], sizeof(*bhdr));
399 if (bhdr == NULL) /* append failed. */
400 rte_panic("mbuf prepend of burst_hdr failed\n");
401 bhdr->original_type = pkt_type;
402 bhdr->pkt_idx = i; /* save the index of the burst chunk. */
403 bhdr->total_pkts = count;
404
405 /* now copy in the actual data */
406 const int curr_data_len = RTE_MIN(len, MAX_TX_LEN -
407 rte_pktmbuf_pkt_len(mbufs[i]) - sizeof(struct ether_hdr));
408 p = (void *)rte_pktmbuf_append(mbufs[i], curr_data_len);
409 if (p == NULL)
410 rte_panic("mbuf append of %d data bytes failed\n", curr_data_len);
411 /* This copy is unavoidable, as we're splitting one big buffer
412 * into multiple mbufs. */
413 rte_memcpy(p, src, curr_data_len);
414
415 dlog("curr_data_len[%d] = %d", i, curr_data_len);
416 dlog("packet %d size %d", i, rte_pktmbuf_pkt_len(mbufs[i]));
417
418 /* Update our source data pointer and remaining length. */
419 len -= curr_data_len;
420 src += curr_data_len;
421 }
422
423 /* Now enqueue the full prepared burst. */
424 i = rte_ring_enqueue_bulk(ctx->tx_ring[0], (void **)mbufs, count, NULL);
425 PANIC_ON(i != count, "failed to enqueue all mbufs: %d/%d", i, count);
426 dlog("%d packets enqueued on port %d.", count, ctx->io_cfg.port);
427
428 return 1;
429}
430
431#endif
432
433/* Prepend ethernet header, possibly vlan tag. */
434void xran_add_eth_hdr_vlan(struct ether_addr *dst, uint16_t ethertype, struct rte_mbuf *mb, uint16_t vlan_tci)
435{
436 /* add in the ethernet header */
437 struct ether_hdr *h = (struct ether_hdr *)rte_pktmbuf_mtod(mb, struct ether_hdr*);
438
439 PANIC_ON(h == NULL, "mbuf prepend of ether_hdr failed");
440
441 /* Fill in the ethernet header. */
442 rte_eth_macaddr_get(mb->port, &h->s_addr); /* set source addr */
443 h->d_addr = *dst; /* set dst addr */
444 h->ether_type = rte_cpu_to_be_16(ethertype); /* ethertype too */
445
446#if defined(DPDKIO_DEBUG) && DPDKIO_DEBUG > 1
447 {
448 char dst[ETHER_ADDR_FMT_SIZE] = "(empty)";
449 char src[ETHER_ADDR_FMT_SIZE] = "(empty)";
450
451 nlog("*** packet for TX below (len %d) ***", rte_pktmbuf_pkt_len(mb));
452 ether_format_addr(src, sizeof(src), &h->s_addr);
453 ether_format_addr(dst, sizeof(dst), &h->d_addr);
454 nlog("src: %s dst: %s ethertype: %.4X", src, dst, ethertype);
455 }
456#endif
457#ifdef VLAN_SUPPORT
458 mb->vlan_tci = vlan_tci;
459 dlog("Inserting vlan tag of %d", vlan_tci);
460 rte_vlan_insert(&mb);
461#endif
462}
463
464