xref: /dpdk/drivers/net/dpaa/dpaa_rxtx.c (revision 3e0ceb9f17fff027fc6c8f18de35e11719ffa61e)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright 2016 Freescale Semiconductor, Inc. All rights reserved.
5  *   Copyright 2017 NXP.
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  Freescale Semiconductor, Inc 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 /* System headers */
35 #include <inttypes.h>
36 #include <unistd.h>
37 #include <stdio.h>
38 #include <limits.h>
39 #include <sched.h>
40 #include <pthread.h>
41 
42 #include <rte_byteorder.h>
43 #include <rte_common.h>
44 #include <rte_interrupts.h>
45 #include <rte_log.h>
46 #include <rte_debug.h>
47 #include <rte_pci.h>
48 #include <rte_atomic.h>
49 #include <rte_branch_prediction.h>
50 #include <rte_memory.h>
51 #include <rte_tailq.h>
52 #include <rte_eal.h>
53 #include <rte_alarm.h>
54 #include <rte_ether.h>
55 #include <rte_ethdev.h>
56 #include <rte_malloc.h>
57 #include <rte_ring.h>
58 #include <rte_ip.h>
59 #include <rte_tcp.h>
60 #include <rte_udp.h>
61 
62 #include "dpaa_ethdev.h"
63 #include "dpaa_rxtx.h"
64 #include <rte_dpaa_bus.h>
65 #include <dpaa_mempool.h>
66 
67 #include <fsl_usd.h>
68 #include <fsl_qman.h>
69 #include <fsl_bman.h>
70 #include <of.h>
71 #include <netcfg.h>
72 
73 #define DPAA_MBUF_TO_CONTIG_FD(_mbuf, _fd, _bpid) \
74 	do { \
75 		(_fd)->cmd = 0; \
76 		(_fd)->opaque_addr = 0; \
77 		(_fd)->opaque = QM_FD_CONTIG << DPAA_FD_FORMAT_SHIFT; \
78 		(_fd)->opaque |= ((_mbuf)->data_off) << DPAA_FD_OFFSET_SHIFT; \
79 		(_fd)->opaque |= (_mbuf)->pkt_len; \
80 		(_fd)->addr = (_mbuf)->buf_iova; \
81 		(_fd)->bpid = _bpid; \
82 	} while (0)
83 
84 #if (defined RTE_LIBRTE_DPAA_DEBUG_DRIVER)
85 void dpaa_display_frame(const struct qm_fd *fd)
86 {
87 	int ii;
88 	char *ptr;
89 
90 	printf("%s::bpid %x addr %08x%08x, format %d off %d, len %d stat %x\n",
91 	       __func__, fd->bpid, fd->addr_hi, fd->addr_lo, fd->format,
92 		fd->offset, fd->length20, fd->status);
93 
94 	ptr = (char *)rte_dpaa_mem_ptov(fd->addr);
95 	ptr += fd->offset;
96 	printf("%02x ", *ptr);
97 	for (ii = 1; ii < fd->length20; ii++) {
98 		printf("%02x ", *ptr);
99 		if ((ii % 16) == 0)
100 			printf("\n");
101 		ptr++;
102 	}
103 	printf("\n");
104 }
105 #else
106 #define dpaa_display_frame(a)
107 #endif
108 
109 static inline void dpaa_slow_parsing(struct rte_mbuf *m __rte_unused,
110 				     uint64_t prs __rte_unused)
111 {
112 	DPAA_DP_LOG(DEBUG, "Slow parsing");
113 	/*TBD:XXX: to be implemented*/
114 }
115 
116 static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
117 					uint64_t fd_virt_addr)
118 {
119 	struct annotations_t *annot = GET_ANNOTATIONS(fd_virt_addr);
120 	uint64_t prs = *((uint64_t *)(&annot->parse)) & DPAA_PARSE_MASK;
121 
122 	DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
123 
124 	switch (prs) {
125 	case DPAA_PKT_TYPE_NONE:
126 		m->packet_type = 0;
127 		break;
128 	case DPAA_PKT_TYPE_ETHER:
129 		m->packet_type = RTE_PTYPE_L2_ETHER;
130 		break;
131 	case DPAA_PKT_TYPE_IPV4:
132 		m->packet_type = RTE_PTYPE_L2_ETHER |
133 			RTE_PTYPE_L3_IPV4;
134 		break;
135 	case DPAA_PKT_TYPE_IPV6:
136 		m->packet_type = RTE_PTYPE_L2_ETHER |
137 			RTE_PTYPE_L3_IPV6;
138 		break;
139 	case DPAA_PKT_TYPE_IPV4_FRAG:
140 	case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
141 	case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
142 	case DPAA_PKT_TYPE_IPV4_FRAG_SCTP:
143 		m->packet_type = RTE_PTYPE_L2_ETHER |
144 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_FRAG;
145 		break;
146 	case DPAA_PKT_TYPE_IPV6_FRAG:
147 	case DPAA_PKT_TYPE_IPV6_FRAG_UDP:
148 	case DPAA_PKT_TYPE_IPV6_FRAG_TCP:
149 	case DPAA_PKT_TYPE_IPV6_FRAG_SCTP:
150 		m->packet_type = RTE_PTYPE_L2_ETHER |
151 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_FRAG;
152 		break;
153 	case DPAA_PKT_TYPE_IPV4_EXT:
154 		m->packet_type = RTE_PTYPE_L2_ETHER |
155 			RTE_PTYPE_L3_IPV4_EXT;
156 		break;
157 	case DPAA_PKT_TYPE_IPV6_EXT:
158 		m->packet_type = RTE_PTYPE_L2_ETHER |
159 			RTE_PTYPE_L3_IPV6_EXT;
160 		break;
161 	case DPAA_PKT_TYPE_IPV4_TCP:
162 		m->packet_type = RTE_PTYPE_L2_ETHER |
163 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP;
164 		break;
165 	case DPAA_PKT_TYPE_IPV6_TCP:
166 		m->packet_type = RTE_PTYPE_L2_ETHER |
167 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
168 		break;
169 	case DPAA_PKT_TYPE_IPV4_UDP:
170 		m->packet_type = RTE_PTYPE_L2_ETHER |
171 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP;
172 		break;
173 	case DPAA_PKT_TYPE_IPV6_UDP:
174 		m->packet_type = RTE_PTYPE_L2_ETHER |
175 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
176 		break;
177 	case DPAA_PKT_TYPE_IPV4_EXT_UDP:
178 		m->packet_type = RTE_PTYPE_L2_ETHER |
179 			RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP;
180 		break;
181 	case DPAA_PKT_TYPE_IPV6_EXT_UDP:
182 		m->packet_type = RTE_PTYPE_L2_ETHER |
183 			RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP;
184 		break;
185 	case DPAA_PKT_TYPE_IPV4_EXT_TCP:
186 		m->packet_type = RTE_PTYPE_L2_ETHER |
187 			RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP;
188 		break;
189 	case DPAA_PKT_TYPE_IPV6_EXT_TCP:
190 		m->packet_type = RTE_PTYPE_L2_ETHER |
191 			RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP;
192 		break;
193 	case DPAA_PKT_TYPE_IPV4_SCTP:
194 		m->packet_type = RTE_PTYPE_L2_ETHER |
195 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP;
196 		break;
197 	case DPAA_PKT_TYPE_IPV6_SCTP:
198 		m->packet_type = RTE_PTYPE_L2_ETHER |
199 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
200 		break;
201 	/* More switch cases can be added */
202 	default:
203 		dpaa_slow_parsing(m, prs);
204 	}
205 
206 	m->tx_offload = annot->parse.ip_off[0];
207 	m->tx_offload |= (annot->parse.l4_off - annot->parse.ip_off[0])
208 					<< DPAA_PKT_L3_LEN_SHIFT;
209 
210 	/* Set the hash values */
211 	m->hash.rss = (uint32_t)(rte_be_to_cpu_64(annot->hash));
212 	m->ol_flags = PKT_RX_RSS_HASH;
213 	/* All packets with Bad checksum are dropped by interface (and
214 	 * corresponding notification issued to RX error queues).
215 	 */
216 	m->ol_flags |= PKT_RX_IP_CKSUM_GOOD;
217 
218 	/* Check if Vlan is present */
219 	if (prs & DPAA_PARSE_VLAN_MASK)
220 		m->ol_flags |= PKT_RX_VLAN;
221 	/* Packet received without stripping the vlan */
222 }
223 
224 static inline void dpaa_checksum(struct rte_mbuf *mbuf)
225 {
226 	struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
227 	char *l3_hdr = (char *)eth_hdr + mbuf->l2_len;
228 	struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
229 	struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
230 
231 	DPAA_DP_LOG(DEBUG, "Calculating checksum for mbuf: %p", mbuf);
232 
233 	if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
234 	    ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
235 	    RTE_PTYPE_L3_IPV4_EXT)) {
236 		ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
237 		ipv4_hdr->hdr_checksum = 0;
238 		ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
239 	} else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
240 		   RTE_PTYPE_L3_IPV6) ||
241 		   ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
242 		   RTE_PTYPE_L3_IPV6_EXT))
243 		ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
244 
245 	if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP) {
246 		struct tcp_hdr *tcp_hdr = (struct tcp_hdr *)(l3_hdr +
247 					  mbuf->l3_len);
248 		tcp_hdr->cksum = 0;
249 		if (eth_hdr->ether_type == htons(ETHER_TYPE_IPv4))
250 			tcp_hdr->cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
251 							       tcp_hdr);
252 		else /* assume ethertype == ETHER_TYPE_IPv6 */
253 			tcp_hdr->cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
254 							       tcp_hdr);
255 	} else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) ==
256 		   RTE_PTYPE_L4_UDP) {
257 		struct udp_hdr *udp_hdr = (struct udp_hdr *)(l3_hdr +
258 							     mbuf->l3_len);
259 		udp_hdr->dgram_cksum = 0;
260 		if (eth_hdr->ether_type == htons(ETHER_TYPE_IPv4))
261 			udp_hdr->dgram_cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
262 								     udp_hdr);
263 		else /* assume ethertype == ETHER_TYPE_IPv6 */
264 			udp_hdr->dgram_cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
265 								     udp_hdr);
266 	}
267 }
268 
269 static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
270 					 struct qm_fd *fd, char *prs_buf)
271 {
272 	struct dpaa_eth_parse_results_t *prs;
273 
274 	DPAA_DP_LOG(DEBUG, " Offloading checksum for mbuf: %p", mbuf);
275 
276 	prs = GET_TX_PRS(prs_buf);
277 	prs->l3r = 0;
278 	prs->l4r = 0;
279 	if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
280 	   ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
281 	   RTE_PTYPE_L3_IPV4_EXT))
282 		prs->l3r = DPAA_L3_PARSE_RESULT_IPV4;
283 	else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
284 		   RTE_PTYPE_L3_IPV6) ||
285 		 ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
286 		RTE_PTYPE_L3_IPV6_EXT))
287 		prs->l3r = DPAA_L3_PARSE_RESULT_IPV6;
288 
289 	if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP)
290 		prs->l4r = DPAA_L4_PARSE_RESULT_TCP;
291 	else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP)
292 		prs->l4r = DPAA_L4_PARSE_RESULT_UDP;
293 
294 	prs->ip_off[0] = mbuf->l2_len;
295 	prs->l4_off = mbuf->l3_len + mbuf->l2_len;
296 	/* Enable L3 (and L4, if TCP or UDP) HW checksum*/
297 	fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
298 }
299 
300 struct rte_mbuf *
301 dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
302 {
303 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
304 	struct rte_mbuf *first_seg, *prev_seg, *cur_seg, *temp;
305 	struct qm_sg_entry *sgt, *sg_temp;
306 	void *vaddr, *sg_vaddr;
307 	int i = 0;
308 	uint8_t fd_offset = fd->offset;
309 
310 	DPAA_DP_LOG(DEBUG, "Received an SG frame");
311 
312 	vaddr = rte_dpaa_mem_ptov(qm_fd_addr(fd));
313 	if (!vaddr) {
314 		DPAA_PMD_ERR("unable to convert physical address");
315 		return NULL;
316 	}
317 	sgt = vaddr + fd_offset;
318 	sg_temp = &sgt[i++];
319 	hw_sg_to_cpu(sg_temp);
320 	temp = (struct rte_mbuf *)((char *)vaddr - bp_info->meta_data_size);
321 	sg_vaddr = rte_dpaa_mem_ptov(qm_sg_entry_get64(sg_temp));
322 
323 	first_seg = (struct rte_mbuf *)((char *)sg_vaddr -
324 						bp_info->meta_data_size);
325 	first_seg->data_off = sg_temp->offset;
326 	first_seg->data_len = sg_temp->length;
327 	first_seg->pkt_len = sg_temp->length;
328 	rte_mbuf_refcnt_set(first_seg, 1);
329 
330 	first_seg->port = ifid;
331 	first_seg->nb_segs = 1;
332 	first_seg->ol_flags = 0;
333 	prev_seg = first_seg;
334 	while (i < DPAA_SGT_MAX_ENTRIES) {
335 		sg_temp = &sgt[i++];
336 		hw_sg_to_cpu(sg_temp);
337 		sg_vaddr = rte_dpaa_mem_ptov(qm_sg_entry_get64(sg_temp));
338 		cur_seg = (struct rte_mbuf *)((char *)sg_vaddr -
339 						      bp_info->meta_data_size);
340 		cur_seg->data_off = sg_temp->offset;
341 		cur_seg->data_len = sg_temp->length;
342 		first_seg->pkt_len += sg_temp->length;
343 		first_seg->nb_segs += 1;
344 		rte_mbuf_refcnt_set(cur_seg, 1);
345 		prev_seg->next = cur_seg;
346 		if (sg_temp->final) {
347 			cur_seg->next = NULL;
348 			break;
349 		}
350 		prev_seg = cur_seg;
351 	}
352 
353 	dpaa_eth_packet_info(first_seg, (uint64_t)vaddr);
354 	rte_pktmbuf_free_seg(temp);
355 
356 	return first_seg;
357 }
358 
359 static inline struct rte_mbuf *dpaa_eth_fd_to_mbuf(struct qm_fd *fd,
360 							uint32_t ifid)
361 {
362 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
363 	struct rte_mbuf *mbuf;
364 	void *ptr;
365 	uint8_t format =
366 		(fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
367 	uint16_t offset =
368 		(fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
369 	uint32_t length = fd->opaque & DPAA_FD_LENGTH_MASK;
370 
371 	DPAA_DP_LOG(DEBUG, " FD--->MBUF");
372 
373 	if (unlikely(format == qm_fd_sg))
374 		return dpaa_eth_sg_to_mbuf(fd, ifid);
375 
376 	/* Ignoring case when format != qm_fd_contig */
377 	dpaa_display_frame(fd);
378 	ptr = rte_dpaa_mem_ptov(fd->addr);
379 	/* Ignoring case when ptr would be NULL. That is only possible incase
380 	 * of a corrupted packet
381 	 */
382 
383 	mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
384 	/* Prefetch the Parse results and packet data to L1 */
385 	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
386 	rte_prefetch0((void *)((uint8_t *)ptr + offset));
387 
388 	mbuf->data_off = offset;
389 	mbuf->data_len = length;
390 	mbuf->pkt_len = length;
391 
392 	mbuf->port = ifid;
393 	mbuf->nb_segs = 1;
394 	mbuf->ol_flags = 0;
395 	mbuf->next = NULL;
396 	rte_mbuf_refcnt_set(mbuf, 1);
397 	dpaa_eth_packet_info(mbuf, (uint64_t)mbuf->buf_addr);
398 
399 	return mbuf;
400 }
401 
402 uint16_t dpaa_eth_queue_rx(void *q,
403 			   struct rte_mbuf **bufs,
404 			   uint16_t nb_bufs)
405 {
406 	struct qman_fq *fq = q;
407 	struct qm_dqrr_entry *dq;
408 	uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
409 	int ret;
410 
411 	ret = rte_dpaa_portal_init((void *)0);
412 	if (ret) {
413 		DPAA_PMD_ERR("Failure in affining portal");
414 		return 0;
415 	}
416 
417 	ret = qman_set_vdq(fq, (nb_bufs > DPAA_MAX_DEQUEUE_NUM_FRAMES) ?
418 				DPAA_MAX_DEQUEUE_NUM_FRAMES : nb_bufs);
419 	if (ret)
420 		return 0;
421 
422 	do {
423 		dq = qman_dequeue(fq);
424 		if (!dq)
425 			continue;
426 		bufs[num_rx++] = dpaa_eth_fd_to_mbuf(&dq->fd, ifid);
427 		qman_dqrr_consume(fq, dq);
428 	} while (fq->flags & QMAN_FQ_STATE_VDQCR);
429 
430 	return num_rx;
431 }
432 
433 static void *dpaa_get_pktbuf(struct dpaa_bp_info *bp_info)
434 {
435 	int ret;
436 	uint64_t buf = 0;
437 	struct bm_buffer bufs;
438 
439 	ret = bman_acquire(bp_info->bp, &bufs, 1, 0);
440 	if (ret <= 0) {
441 		DPAA_PMD_WARN("Failed to allocate buffers %d", ret);
442 		return (void *)buf;
443 	}
444 
445 	DPAA_DP_LOG(DEBUG, "got buffer 0x%lx from pool %d",
446 		    (uint64_t)bufs.addr, bufs.bpid);
447 
448 	buf = (uint64_t)rte_dpaa_mem_ptov(bufs.addr) - bp_info->meta_data_size;
449 	if (!buf)
450 		goto out;
451 
452 out:
453 	return (void *)buf;
454 }
455 
456 static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
457 					     struct dpaa_if *dpaa_intf)
458 {
459 	struct rte_mbuf *dpaa_mbuf;
460 
461 	/* allocate pktbuffer on bpid for dpaa port */
462 	dpaa_mbuf = dpaa_get_pktbuf(dpaa_intf->bp_info);
463 	if (!dpaa_mbuf)
464 		return NULL;
465 
466 	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + mbuf->data_off, (void *)
467 		((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
468 
469 	/* Copy only the required fields */
470 	dpaa_mbuf->data_off = mbuf->data_off;
471 	dpaa_mbuf->pkt_len = mbuf->pkt_len;
472 	dpaa_mbuf->ol_flags = mbuf->ol_flags;
473 	dpaa_mbuf->packet_type = mbuf->packet_type;
474 	dpaa_mbuf->tx_offload = mbuf->tx_offload;
475 	rte_pktmbuf_free(mbuf);
476 	return dpaa_mbuf;
477 }
478 
479 int
480 dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
481 		struct qm_fd *fd,
482 		uint32_t bpid)
483 {
484 	struct rte_mbuf *cur_seg = mbuf, *prev_seg = NULL;
485 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(bpid);
486 	struct rte_mbuf *temp, *mi;
487 	struct qm_sg_entry *sg_temp, *sgt;
488 	int i = 0;
489 
490 	DPAA_DP_LOG(DEBUG, "Creating SG FD to transmit");
491 
492 	temp = rte_pktmbuf_alloc(bp_info->mp);
493 	if (!temp) {
494 		DPAA_PMD_ERR("Failure in allocation of mbuf");
495 		return -1;
496 	}
497 	if (temp->buf_len < ((mbuf->nb_segs * sizeof(struct qm_sg_entry))
498 				+ temp->data_off)) {
499 		DPAA_PMD_ERR("Insufficient space in mbuf for SG entries");
500 		return -1;
501 	}
502 
503 	fd->cmd = 0;
504 	fd->opaque_addr = 0;
505 
506 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
507 		if (temp->data_off < DEFAULT_TX_ICEOF
508 			+ sizeof(struct dpaa_eth_parse_results_t))
509 			temp->data_off = DEFAULT_TX_ICEOF
510 				+ sizeof(struct dpaa_eth_parse_results_t);
511 		dcbz_64(temp->buf_addr);
512 		dpaa_checksum_offload(mbuf, fd, temp->buf_addr);
513 	}
514 
515 	sgt = temp->buf_addr + temp->data_off;
516 	fd->format = QM_FD_SG;
517 	fd->addr = temp->buf_iova;
518 	fd->offset = temp->data_off;
519 	fd->bpid = bpid;
520 	fd->length20 = mbuf->pkt_len;
521 
522 	while (i < DPAA_SGT_MAX_ENTRIES) {
523 		sg_temp = &sgt[i++];
524 		sg_temp->opaque = 0;
525 		sg_temp->val = 0;
526 		sg_temp->addr = cur_seg->buf_iova;
527 		sg_temp->offset = cur_seg->data_off;
528 		sg_temp->length = cur_seg->data_len;
529 		if (RTE_MBUF_DIRECT(cur_seg)) {
530 			if (rte_mbuf_refcnt_read(cur_seg) > 1) {
531 				/*If refcnt > 1, invalid bpid is set to ensure
532 				 * buffer is not freed by HW.
533 				 */
534 				sg_temp->bpid = 0xff;
535 				rte_mbuf_refcnt_update(cur_seg, -1);
536 			} else {
537 				sg_temp->bpid =
538 					DPAA_MEMPOOL_TO_BPID(cur_seg->pool);
539 			}
540 			cur_seg = cur_seg->next;
541 		} else {
542 			/* Get owner MBUF from indirect buffer */
543 			mi = rte_mbuf_from_indirect(cur_seg);
544 			if (rte_mbuf_refcnt_read(mi) > 1) {
545 				/*If refcnt > 1, invalid bpid is set to ensure
546 				 * owner buffer is not freed by HW.
547 				 */
548 				sg_temp->bpid = 0xff;
549 			} else {
550 				sg_temp->bpid = DPAA_MEMPOOL_TO_BPID(mi->pool);
551 				rte_mbuf_refcnt_update(mi, 1);
552 			}
553 			prev_seg = cur_seg;
554 			cur_seg = cur_seg->next;
555 			prev_seg->next = NULL;
556 			rte_pktmbuf_free(prev_seg);
557 		}
558 		if (cur_seg == NULL) {
559 			sg_temp->final = 1;
560 			cpu_to_hw_sg(sg_temp);
561 			break;
562 		}
563 		cpu_to_hw_sg(sg_temp);
564 	}
565 	return 0;
566 }
567 
568 /* Handle mbufs which are not segmented (non SG) */
569 static inline void
570 tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
571 			    struct dpaa_bp_info *bp_info,
572 			    struct qm_fd *fd_arr)
573 {
574 	struct rte_mbuf *mi = NULL;
575 
576 	if (RTE_MBUF_DIRECT(mbuf)) {
577 		if (rte_mbuf_refcnt_read(mbuf) > 1) {
578 			/* In case of direct mbuf and mbuf being cloned,
579 			 * BMAN should _not_ release buffer.
580 			 */
581 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
582 			/* Buffer should be releasd by EAL */
583 			rte_mbuf_refcnt_update(mbuf, -1);
584 		} else {
585 			/* In case of direct mbuf and no cloning, mbuf can be
586 			 * released by BMAN.
587 			 */
588 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
589 		}
590 	} else {
591 		/* This is data-containing core mbuf: 'mi' */
592 		mi = rte_mbuf_from_indirect(mbuf);
593 		if (rte_mbuf_refcnt_read(mi) > 1) {
594 			/* In case of indirect mbuf, and mbuf being cloned,
595 			 * BMAN should _not_ release it and let EAL release
596 			 * it through pktmbuf_free below.
597 			 */
598 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
599 		} else {
600 			/* In case of indirect mbuf, and no cloning, core mbuf
601 			 * should be released by BMAN.
602 			 * Increate refcnt of core mbuf so that when
603 			 * pktmbuf_free is called and mbuf is released, EAL
604 			 * doesn't try to release core mbuf which would have
605 			 * been released by BMAN.
606 			 */
607 			rte_mbuf_refcnt_update(mi, 1);
608 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
609 		}
610 		rte_pktmbuf_free(mbuf);
611 	}
612 
613 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
614 		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
615 		    sizeof(struct dpaa_eth_parse_results_t))) {
616 			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
617 				"Not enough Headroom "
618 				"space for correct Checksum offload."
619 				"So Calculating checksum in Software.");
620 			dpaa_checksum(mbuf);
621 		} else {
622 			dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
623 		}
624 	}
625 }
626 
627 /* Handle all mbufs on dpaa BMAN managed pool */
628 static inline uint16_t
629 tx_on_dpaa_pool(struct rte_mbuf *mbuf,
630 		struct dpaa_bp_info *bp_info,
631 		struct qm_fd *fd_arr)
632 {
633 	DPAA_DP_LOG(DEBUG, "BMAN offloaded buffer, mbuf: %p", mbuf);
634 
635 	if (mbuf->nb_segs == 1) {
636 		/* Case for non-segmented buffers */
637 		tx_on_dpaa_pool_unsegmented(mbuf, bp_info, fd_arr);
638 	} else if (mbuf->nb_segs > 1 &&
639 		   mbuf->nb_segs <= DPAA_SGT_MAX_ENTRIES) {
640 		if (dpaa_eth_mbuf_to_sg_fd(mbuf, fd_arr, bp_info->bpid)) {
641 			DPAA_PMD_DEBUG("Unable to create Scatter Gather FD");
642 			return 1;
643 		}
644 	} else {
645 		DPAA_PMD_DEBUG("Number of Segments not supported");
646 		return 1;
647 	}
648 
649 	return 0;
650 }
651 
652 /* Handle all mbufs on an external pool (non-dpaa) */
653 static inline uint16_t
654 tx_on_external_pool(struct qman_fq *txq, struct rte_mbuf *mbuf,
655 		    struct qm_fd *fd_arr)
656 {
657 	struct dpaa_if *dpaa_intf = txq->dpaa_intf;
658 	struct rte_mbuf *dmable_mbuf;
659 
660 	DPAA_DP_LOG(DEBUG, "Non-BMAN offloaded buffer."
661 		    "Allocating an offloaded buffer");
662 	dmable_mbuf = dpaa_get_dmable_mbuf(mbuf, dpaa_intf);
663 	if (!dmable_mbuf) {
664 		DPAA_DP_LOG(DEBUG, "no dpaa buffers.");
665 		return 1;
666 	}
667 
668 	DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, dpaa_intf->bp_info->bpid);
669 
670 	return 0;
671 }
672 
673 uint16_t
674 dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
675 {
676 	struct rte_mbuf *mbuf, *mi = NULL;
677 	struct rte_mempool *mp;
678 	struct dpaa_bp_info *bp_info;
679 	struct qm_fd fd_arr[MAX_TX_RING_SLOTS];
680 	uint32_t frames_to_send, loop, i = 0;
681 	uint16_t state;
682 	int ret;
683 
684 	ret = rte_dpaa_portal_init((void *)0);
685 	if (ret) {
686 		DPAA_PMD_ERR("Failure in affining portal");
687 		return 0;
688 	}
689 
690 	DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
691 
692 	while (nb_bufs) {
693 		frames_to_send = (nb_bufs >> 3) ? MAX_TX_RING_SLOTS : nb_bufs;
694 		for (loop = 0; loop < frames_to_send; loop++, i++) {
695 			mbuf = bufs[i];
696 			if (RTE_MBUF_DIRECT(mbuf)) {
697 				mp = mbuf->pool;
698 			} else {
699 				mi = rte_mbuf_from_indirect(mbuf);
700 				mp = mi->pool;
701 			}
702 
703 			bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
704 			if (likely(mp->ops_index == bp_info->dpaa_ops_index)) {
705 				state = tx_on_dpaa_pool(mbuf, bp_info,
706 							&fd_arr[loop]);
707 				if (unlikely(state)) {
708 					/* Set frames_to_send & nb_bufs so
709 					 * that packets are transmitted till
710 					 * previous frame.
711 					 */
712 					frames_to_send = loop;
713 					nb_bufs = loop;
714 					goto send_pkts;
715 				}
716 			} else {
717 				state = tx_on_external_pool(q, mbuf,
718 							    &fd_arr[loop]);
719 				if (unlikely(state)) {
720 					/* Set frames_to_send & nb_bufs so
721 					 * that packets are transmitted till
722 					 * previous frame.
723 					 */
724 					frames_to_send = loop;
725 					nb_bufs = loop;
726 					goto send_pkts;
727 				}
728 			}
729 		}
730 
731 send_pkts:
732 		loop = 0;
733 		while (loop < frames_to_send) {
734 			loop += qman_enqueue_multi(q, &fd_arr[loop],
735 					frames_to_send - loop);
736 		}
737 		nb_bufs -= frames_to_send;
738 	}
739 
740 	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", i, q);
741 
742 	return i;
743 }
744 
745 uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
746 			      struct rte_mbuf **bufs __rte_unused,
747 		uint16_t nb_bufs __rte_unused)
748 {
749 	DPAA_DP_LOG(DEBUG, "Drop all packets");
750 
751 	/* Drop all incoming packets. No need to free packets here
752 	 * because the rte_eth f/w frees up the packets through tx_buffer
753 	 * callback in case this functions returns count less than nb_bufs
754 	 */
755 	return 0;
756 }
757