xref: /dpdk/drivers/net/dpaa/dpaa_rxtx.c (revision 5d944582d0282f0d133619cff7515d89bc4ed216)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  *   Copyright 2016 Freescale Semiconductor, Inc. All rights reserved.
4  *   Copyright 2017 NXP
5  *
6  */
7 
8 /* System headers */
9 #include <inttypes.h>
10 #include <unistd.h>
11 #include <stdio.h>
12 #include <limits.h>
13 #include <sched.h>
14 #include <pthread.h>
15 
16 #include <rte_byteorder.h>
17 #include <rte_common.h>
18 #include <rte_interrupts.h>
19 #include <rte_log.h>
20 #include <rte_debug.h>
21 #include <rte_pci.h>
22 #include <rte_atomic.h>
23 #include <rte_branch_prediction.h>
24 #include <rte_memory.h>
25 #include <rte_tailq.h>
26 #include <rte_eal.h>
27 #include <rte_alarm.h>
28 #include <rte_ether.h>
29 #include <rte_ethdev_driver.h>
30 #include <rte_malloc.h>
31 #include <rte_ring.h>
32 #include <rte_ip.h>
33 #include <rte_tcp.h>
34 #include <rte_udp.h>
35 #include <rte_net.h>
36 #include <rte_eventdev.h>
37 
38 #include "dpaa_ethdev.h"
39 #include "dpaa_rxtx.h"
40 #include <rte_dpaa_bus.h>
41 #include <dpaa_mempool.h>
42 
43 #include <qman.h>
44 #include <fsl_usd.h>
45 #include <fsl_qman.h>
46 #include <fsl_bman.h>
47 #include <of.h>
48 #include <netcfg.h>
49 
50 #define DPAA_MBUF_TO_CONTIG_FD(_mbuf, _fd, _bpid) \
51 	do { \
52 		(_fd)->cmd = 0; \
53 		(_fd)->opaque_addr = 0; \
54 		(_fd)->opaque = QM_FD_CONTIG << DPAA_FD_FORMAT_SHIFT; \
55 		(_fd)->opaque |= ((_mbuf)->data_off) << DPAA_FD_OFFSET_SHIFT; \
56 		(_fd)->opaque |= (_mbuf)->pkt_len; \
57 		(_fd)->addr = (_mbuf)->buf_iova; \
58 		(_fd)->bpid = _bpid; \
59 	} while (0)
60 
61 #if (defined RTE_LIBRTE_DPAA_DEBUG_DRIVER)
62 void dpaa_display_frame(const struct qm_fd *fd)
63 {
64 	int ii;
65 	char *ptr;
66 
67 	printf("%s::bpid %x addr %08x%08x, format %d off %d, len %d stat %x\n",
68 	       __func__, fd->bpid, fd->addr_hi, fd->addr_lo, fd->format,
69 		fd->offset, fd->length20, fd->status);
70 
71 	ptr = (char *)rte_dpaa_mem_ptov(fd->addr);
72 	ptr += fd->offset;
73 	printf("%02x ", *ptr);
74 	for (ii = 1; ii < fd->length20; ii++) {
75 		printf("%02x ", *ptr);
76 		if ((ii % 16) == 0)
77 			printf("\n");
78 		ptr++;
79 	}
80 	printf("\n");
81 }
82 #else
83 #define dpaa_display_frame(a)
84 #endif
85 
86 static inline void dpaa_slow_parsing(struct rte_mbuf *m __rte_unused,
87 				     uint64_t prs __rte_unused)
88 {
89 	DPAA_DP_LOG(DEBUG, "Slow parsing");
90 	/*TBD:XXX: to be implemented*/
91 }
92 
93 static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
94 					uint64_t fd_virt_addr)
95 {
96 	struct annotations_t *annot = GET_ANNOTATIONS(fd_virt_addr);
97 	uint64_t prs = *((uint64_t *)(&annot->parse)) & DPAA_PARSE_MASK;
98 
99 	DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
100 
101 	switch (prs) {
102 	case DPAA_PKT_TYPE_IPV4:
103 		m->packet_type = RTE_PTYPE_L2_ETHER |
104 			RTE_PTYPE_L3_IPV4;
105 		break;
106 	case DPAA_PKT_TYPE_IPV6:
107 		m->packet_type = RTE_PTYPE_L2_ETHER |
108 			RTE_PTYPE_L3_IPV6;
109 		break;
110 	case DPAA_PKT_TYPE_ETHER:
111 		m->packet_type = RTE_PTYPE_L2_ETHER;
112 		break;
113 	case DPAA_PKT_TYPE_IPV4_FRAG:
114 	case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
115 	case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
116 	case DPAA_PKT_TYPE_IPV4_FRAG_SCTP:
117 		m->packet_type = RTE_PTYPE_L2_ETHER |
118 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_FRAG;
119 		break;
120 	case DPAA_PKT_TYPE_IPV6_FRAG:
121 	case DPAA_PKT_TYPE_IPV6_FRAG_UDP:
122 	case DPAA_PKT_TYPE_IPV6_FRAG_TCP:
123 	case DPAA_PKT_TYPE_IPV6_FRAG_SCTP:
124 		m->packet_type = RTE_PTYPE_L2_ETHER |
125 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_FRAG;
126 		break;
127 	case DPAA_PKT_TYPE_IPV4_EXT:
128 		m->packet_type = RTE_PTYPE_L2_ETHER |
129 			RTE_PTYPE_L3_IPV4_EXT;
130 		break;
131 	case DPAA_PKT_TYPE_IPV6_EXT:
132 		m->packet_type = RTE_PTYPE_L2_ETHER |
133 			RTE_PTYPE_L3_IPV6_EXT;
134 		break;
135 	case DPAA_PKT_TYPE_IPV4_TCP:
136 		m->packet_type = RTE_PTYPE_L2_ETHER |
137 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP;
138 		break;
139 	case DPAA_PKT_TYPE_IPV6_TCP:
140 		m->packet_type = RTE_PTYPE_L2_ETHER |
141 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
142 		break;
143 	case DPAA_PKT_TYPE_IPV4_UDP:
144 		m->packet_type = RTE_PTYPE_L2_ETHER |
145 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP;
146 		break;
147 	case DPAA_PKT_TYPE_IPV6_UDP:
148 		m->packet_type = RTE_PTYPE_L2_ETHER |
149 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
150 		break;
151 	case DPAA_PKT_TYPE_IPV4_EXT_UDP:
152 		m->packet_type = RTE_PTYPE_L2_ETHER |
153 			RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP;
154 		break;
155 	case DPAA_PKT_TYPE_IPV6_EXT_UDP:
156 		m->packet_type = RTE_PTYPE_L2_ETHER |
157 			RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP;
158 		break;
159 	case DPAA_PKT_TYPE_IPV4_EXT_TCP:
160 		m->packet_type = RTE_PTYPE_L2_ETHER |
161 			RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP;
162 		break;
163 	case DPAA_PKT_TYPE_IPV6_EXT_TCP:
164 		m->packet_type = RTE_PTYPE_L2_ETHER |
165 			RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP;
166 		break;
167 	case DPAA_PKT_TYPE_IPV4_SCTP:
168 		m->packet_type = RTE_PTYPE_L2_ETHER |
169 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP;
170 		break;
171 	case DPAA_PKT_TYPE_IPV6_SCTP:
172 		m->packet_type = RTE_PTYPE_L2_ETHER |
173 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
174 		break;
175 	case DPAA_PKT_TYPE_NONE:
176 		m->packet_type = 0;
177 		break;
178 	/* More switch cases can be added */
179 	default:
180 		dpaa_slow_parsing(m, prs);
181 	}
182 
183 	m->tx_offload = annot->parse.ip_off[0];
184 	m->tx_offload |= (annot->parse.l4_off - annot->parse.ip_off[0])
185 					<< DPAA_PKT_L3_LEN_SHIFT;
186 
187 	/* Set the hash values */
188 	m->hash.rss = (uint32_t)(annot->hash);
189 	/* All packets with Bad checksum are dropped by interface (and
190 	 * corresponding notification issued to RX error queues).
191 	 */
192 	m->ol_flags = PKT_RX_RSS_HASH | PKT_RX_IP_CKSUM_GOOD;
193 
194 	/* Check if Vlan is present */
195 	if (prs & DPAA_PARSE_VLAN_MASK)
196 		m->ol_flags |= PKT_RX_VLAN;
197 	/* Packet received without stripping the vlan */
198 }
199 
200 static inline void dpaa_checksum(struct rte_mbuf *mbuf)
201 {
202 	struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
203 	char *l3_hdr = (char *)eth_hdr + mbuf->l2_len;
204 	struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
205 	struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
206 
207 	DPAA_DP_LOG(DEBUG, "Calculating checksum for mbuf: %p", mbuf);
208 
209 	if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
210 	    ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
211 	    RTE_PTYPE_L3_IPV4_EXT)) {
212 		ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
213 		ipv4_hdr->hdr_checksum = 0;
214 		ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
215 	} else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
216 		   RTE_PTYPE_L3_IPV6) ||
217 		   ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
218 		   RTE_PTYPE_L3_IPV6_EXT))
219 		ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
220 
221 	if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP) {
222 		struct tcp_hdr *tcp_hdr = (struct tcp_hdr *)(l3_hdr +
223 					  mbuf->l3_len);
224 		tcp_hdr->cksum = 0;
225 		if (eth_hdr->ether_type == htons(ETHER_TYPE_IPv4))
226 			tcp_hdr->cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
227 							       tcp_hdr);
228 		else /* assume ethertype == ETHER_TYPE_IPv6 */
229 			tcp_hdr->cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
230 							       tcp_hdr);
231 	} else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) ==
232 		   RTE_PTYPE_L4_UDP) {
233 		struct udp_hdr *udp_hdr = (struct udp_hdr *)(l3_hdr +
234 							     mbuf->l3_len);
235 		udp_hdr->dgram_cksum = 0;
236 		if (eth_hdr->ether_type == htons(ETHER_TYPE_IPv4))
237 			udp_hdr->dgram_cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
238 								     udp_hdr);
239 		else /* assume ethertype == ETHER_TYPE_IPv6 */
240 			udp_hdr->dgram_cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
241 								     udp_hdr);
242 	}
243 }
244 
245 static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
246 					 struct qm_fd *fd, char *prs_buf)
247 {
248 	struct dpaa_eth_parse_results_t *prs;
249 
250 	DPAA_DP_LOG(DEBUG, " Offloading checksum for mbuf: %p", mbuf);
251 
252 	prs = GET_TX_PRS(prs_buf);
253 	prs->l3r = 0;
254 	prs->l4r = 0;
255 	if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
256 	   ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
257 	   RTE_PTYPE_L3_IPV4_EXT))
258 		prs->l3r = DPAA_L3_PARSE_RESULT_IPV4;
259 	else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
260 		   RTE_PTYPE_L3_IPV6) ||
261 		 ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
262 		RTE_PTYPE_L3_IPV6_EXT))
263 		prs->l3r = DPAA_L3_PARSE_RESULT_IPV6;
264 
265 	if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP)
266 		prs->l4r = DPAA_L4_PARSE_RESULT_TCP;
267 	else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP)
268 		prs->l4r = DPAA_L4_PARSE_RESULT_UDP;
269 
270 	prs->ip_off[0] = mbuf->l2_len;
271 	prs->l4_off = mbuf->l3_len + mbuf->l2_len;
272 	/* Enable L3 (and L4, if TCP or UDP) HW checksum*/
273 	fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
274 }
275 
276 static inline void
277 dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
278 {
279 	if (!mbuf->packet_type) {
280 		struct rte_net_hdr_lens hdr_lens;
281 
282 		mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
283 				RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
284 				| RTE_PTYPE_L4_MASK);
285 		mbuf->l2_len = hdr_lens.l2_len;
286 		mbuf->l3_len = hdr_lens.l3_len;
287 	}
288 	if (mbuf->data_off < (DEFAULT_TX_ICEOF +
289 	    sizeof(struct dpaa_eth_parse_results_t))) {
290 		DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
291 			"Not enough Headroom "
292 			"space for correct Checksum offload."
293 			"So Calculating checksum in Software.");
294 		dpaa_checksum(mbuf);
295 	} else {
296 		dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
297 	}
298 }
299 
300 struct rte_mbuf *
301 dpaa_eth_sg_to_mbuf(const 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 = DPAA_MEMPOOL_PTOV(bp_info, 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 = DPAA_MEMPOOL_PTOV(bp_info, 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 = DPAA_MEMPOOL_PTOV(bp_info,
338 					     qm_sg_entry_get64(sg_temp));
339 		cur_seg = (struct rte_mbuf *)((char *)sg_vaddr -
340 						      bp_info->meta_data_size);
341 		cur_seg->data_off = sg_temp->offset;
342 		cur_seg->data_len = sg_temp->length;
343 		first_seg->pkt_len += sg_temp->length;
344 		first_seg->nb_segs += 1;
345 		rte_mbuf_refcnt_set(cur_seg, 1);
346 		prev_seg->next = cur_seg;
347 		if (sg_temp->final) {
348 			cur_seg->next = NULL;
349 			break;
350 		}
351 		prev_seg = cur_seg;
352 	}
353 
354 	dpaa_eth_packet_info(first_seg, (uint64_t)vaddr);
355 	rte_pktmbuf_free_seg(temp);
356 
357 	return first_seg;
358 }
359 
360 static inline struct rte_mbuf *
361 dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
362 {
363 	struct rte_mbuf *mbuf;
364 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
365 	void *ptr;
366 	uint8_t format =
367 		(fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
368 	uint16_t offset;
369 	uint32_t length;
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 	ptr = DPAA_MEMPOOL_PTOV(bp_info, qm_fd_addr(fd));
377 
378 	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
379 
380 	offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
381 	length = fd->opaque & DPAA_FD_LENGTH_MASK;
382 
383 	/* Ignoring case when format != qm_fd_contig */
384 	dpaa_display_frame(fd);
385 
386 	mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
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 enum qman_cb_dqrr_result dpaa_rx_cb(void *event __always_unused,
403 				    struct qman_portal *qm __always_unused,
404 				    struct qman_fq *fq,
405 				    const struct qm_dqrr_entry *dqrr,
406 				    void **bufs)
407 {
408 	const struct qm_fd *fd = &dqrr->fd;
409 
410 	*bufs = dpaa_eth_fd_to_mbuf(fd,
411 			((struct dpaa_if *)fq->dpaa_intf)->ifid);
412 	return qman_cb_dqrr_consume;
413 }
414 
415 static uint16_t
416 dpaa_eth_queue_portal_rx(struct qman_fq *fq,
417 			 struct rte_mbuf **bufs,
418 			 uint16_t nb_bufs)
419 {
420 	int ret;
421 
422 	if (unlikely(fq->qp == NULL)) {
423 		ret = rte_dpaa_portal_fq_init((void *)0, fq);
424 		if (ret) {
425 			DPAA_PMD_ERR("Failure in affining portal %d", ret);
426 			return 0;
427 		}
428 	}
429 
430 	return qman_portal_poll_rx(nb_bufs, (void **)bufs, fq->qp);
431 }
432 
433 enum qman_cb_dqrr_result
434 dpaa_rx_cb_parallel(void *event,
435 		    struct qman_portal *qm __always_unused,
436 		    struct qman_fq *fq,
437 		    const struct qm_dqrr_entry *dqrr,
438 		    void **bufs)
439 {
440 	u32 ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
441 	struct rte_mbuf *mbuf;
442 	struct rte_event *ev = (struct rte_event *)event;
443 
444 	mbuf = dpaa_eth_fd_to_mbuf(&dqrr->fd, ifid);
445 	ev->event_ptr = (void *)mbuf;
446 	ev->flow_id = fq->ev.flow_id;
447 	ev->sub_event_type = fq->ev.sub_event_type;
448 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
449 	ev->op = RTE_EVENT_OP_NEW;
450 	ev->sched_type = fq->ev.sched_type;
451 	ev->queue_id = fq->ev.queue_id;
452 	ev->priority = fq->ev.priority;
453 	ev->impl_opaque = (uint8_t)DPAA_INVALID_MBUF_SEQN;
454 	mbuf->seqn = DPAA_INVALID_MBUF_SEQN;
455 	*bufs = mbuf;
456 
457 	return qman_cb_dqrr_consume;
458 }
459 
460 enum qman_cb_dqrr_result
461 dpaa_rx_cb_atomic(void *event,
462 		  struct qman_portal *qm __always_unused,
463 		  struct qman_fq *fq,
464 		  const struct qm_dqrr_entry *dqrr,
465 		  void **bufs)
466 {
467 	u8 index;
468 	u32 ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
469 	struct rte_mbuf *mbuf;
470 	struct rte_event *ev = (struct rte_event *)event;
471 
472 	mbuf = dpaa_eth_fd_to_mbuf(&dqrr->fd, ifid);
473 	ev->event_ptr = (void *)mbuf;
474 	ev->flow_id = fq->ev.flow_id;
475 	ev->sub_event_type = fq->ev.sub_event_type;
476 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
477 	ev->op = RTE_EVENT_OP_NEW;
478 	ev->sched_type = fq->ev.sched_type;
479 	ev->queue_id = fq->ev.queue_id;
480 	ev->priority = fq->ev.priority;
481 
482 	/* Save active dqrr entries */
483 	index = DQRR_PTR2IDX(dqrr);
484 	DPAA_PER_LCORE_DQRR_SIZE++;
485 	DPAA_PER_LCORE_DQRR_HELD |= 1 << index;
486 	DPAA_PER_LCORE_DQRR_MBUF(index) = mbuf;
487 	ev->impl_opaque = index + 1;
488 	mbuf->seqn = (uint32_t)index + 1;
489 	*bufs = mbuf;
490 
491 	return qman_cb_dqrr_defer;
492 }
493 
494 uint16_t dpaa_eth_queue_rx(void *q,
495 			   struct rte_mbuf **bufs,
496 			   uint16_t nb_bufs)
497 {
498 	struct qman_fq *fq = q;
499 	struct qm_dqrr_entry *dq;
500 	uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
501 	int ret;
502 
503 	if (likely(fq->is_static))
504 		return dpaa_eth_queue_portal_rx(fq, bufs, nb_bufs);
505 
506 	if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
507 		ret = rte_dpaa_portal_init((void *)0);
508 		if (ret) {
509 			DPAA_PMD_ERR("Failure in affining portal");
510 			return 0;
511 		}
512 	}
513 
514 	ret = qman_set_vdq(fq, (nb_bufs > DPAA_MAX_DEQUEUE_NUM_FRAMES) ?
515 				DPAA_MAX_DEQUEUE_NUM_FRAMES : nb_bufs);
516 	if (ret)
517 		return 0;
518 
519 	do {
520 		dq = qman_dequeue(fq);
521 		if (!dq)
522 			continue;
523 		bufs[num_rx++] = dpaa_eth_fd_to_mbuf(&dq->fd, ifid);
524 		qman_dqrr_consume(fq, dq);
525 	} while (fq->flags & QMAN_FQ_STATE_VDQCR);
526 
527 	return num_rx;
528 }
529 
530 static void *dpaa_get_pktbuf(struct dpaa_bp_info *bp_info)
531 {
532 	int ret;
533 	uint64_t buf = 0;
534 	struct bm_buffer bufs;
535 
536 	ret = bman_acquire(bp_info->bp, &bufs, 1, 0);
537 	if (ret <= 0) {
538 		DPAA_PMD_WARN("Failed to allocate buffers %d", ret);
539 		return (void *)buf;
540 	}
541 
542 	DPAA_DP_LOG(DEBUG, "got buffer 0x%lx from pool %d",
543 		    (uint64_t)bufs.addr, bufs.bpid);
544 
545 	buf = (uint64_t)DPAA_MEMPOOL_PTOV(bp_info, bufs.addr)
546 				- bp_info->meta_data_size;
547 	if (!buf)
548 		goto out;
549 
550 out:
551 	return (void *)buf;
552 }
553 
554 static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
555 					     struct dpaa_if *dpaa_intf)
556 {
557 	struct rte_mbuf *dpaa_mbuf;
558 
559 	/* allocate pktbuffer on bpid for dpaa port */
560 	dpaa_mbuf = dpaa_get_pktbuf(dpaa_intf->bp_info);
561 	if (!dpaa_mbuf)
562 		return NULL;
563 
564 	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + RTE_PKTMBUF_HEADROOM, (void *)
565 		((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
566 
567 	/* Copy only the required fields */
568 	dpaa_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
569 	dpaa_mbuf->pkt_len = mbuf->pkt_len;
570 	dpaa_mbuf->ol_flags = mbuf->ol_flags;
571 	dpaa_mbuf->packet_type = mbuf->packet_type;
572 	dpaa_mbuf->tx_offload = mbuf->tx_offload;
573 	rte_pktmbuf_free(mbuf);
574 	return dpaa_mbuf;
575 }
576 
577 int
578 dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
579 		struct qm_fd *fd,
580 		uint32_t bpid)
581 {
582 	struct rte_mbuf *cur_seg = mbuf, *prev_seg = NULL;
583 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(bpid);
584 	struct rte_mbuf *temp, *mi;
585 	struct qm_sg_entry *sg_temp, *sgt;
586 	int i = 0;
587 
588 	DPAA_DP_LOG(DEBUG, "Creating SG FD to transmit");
589 
590 	temp = rte_pktmbuf_alloc(bp_info->mp);
591 	if (!temp) {
592 		DPAA_PMD_ERR("Failure in allocation of mbuf");
593 		return -1;
594 	}
595 	if (temp->buf_len < ((mbuf->nb_segs * sizeof(struct qm_sg_entry))
596 				+ temp->data_off)) {
597 		DPAA_PMD_ERR("Insufficient space in mbuf for SG entries");
598 		return -1;
599 	}
600 
601 	fd->cmd = 0;
602 	fd->opaque_addr = 0;
603 
604 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
605 		if (!mbuf->packet_type) {
606 			struct rte_net_hdr_lens hdr_lens;
607 
608 			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
609 					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
610 					| RTE_PTYPE_L4_MASK);
611 			mbuf->l2_len = hdr_lens.l2_len;
612 			mbuf->l3_len = hdr_lens.l3_len;
613 		}
614 		if (temp->data_off < DEFAULT_TX_ICEOF
615 			+ sizeof(struct dpaa_eth_parse_results_t))
616 			temp->data_off = DEFAULT_TX_ICEOF
617 				+ sizeof(struct dpaa_eth_parse_results_t);
618 		dcbz_64(temp->buf_addr);
619 		dpaa_checksum_offload(mbuf, fd, temp->buf_addr);
620 	}
621 
622 	sgt = temp->buf_addr + temp->data_off;
623 	fd->format = QM_FD_SG;
624 	fd->addr = temp->buf_iova;
625 	fd->offset = temp->data_off;
626 	fd->bpid = bpid;
627 	fd->length20 = mbuf->pkt_len;
628 
629 	while (i < DPAA_SGT_MAX_ENTRIES) {
630 		sg_temp = &sgt[i++];
631 		sg_temp->opaque = 0;
632 		sg_temp->val = 0;
633 		sg_temp->addr = cur_seg->buf_iova;
634 		sg_temp->offset = cur_seg->data_off;
635 		sg_temp->length = cur_seg->data_len;
636 		if (RTE_MBUF_DIRECT(cur_seg)) {
637 			if (rte_mbuf_refcnt_read(cur_seg) > 1) {
638 				/*If refcnt > 1, invalid bpid is set to ensure
639 				 * buffer is not freed by HW.
640 				 */
641 				sg_temp->bpid = 0xff;
642 				rte_mbuf_refcnt_update(cur_seg, -1);
643 			} else {
644 				sg_temp->bpid =
645 					DPAA_MEMPOOL_TO_BPID(cur_seg->pool);
646 			}
647 			cur_seg = cur_seg->next;
648 		} else {
649 			/* Get owner MBUF from indirect buffer */
650 			mi = rte_mbuf_from_indirect(cur_seg);
651 			if (rte_mbuf_refcnt_read(mi) > 1) {
652 				/*If refcnt > 1, invalid bpid is set to ensure
653 				 * owner buffer is not freed by HW.
654 				 */
655 				sg_temp->bpid = 0xff;
656 			} else {
657 				sg_temp->bpid = DPAA_MEMPOOL_TO_BPID(mi->pool);
658 				rte_mbuf_refcnt_update(mi, 1);
659 			}
660 			prev_seg = cur_seg;
661 			cur_seg = cur_seg->next;
662 			prev_seg->next = NULL;
663 			rte_pktmbuf_free(prev_seg);
664 		}
665 		if (cur_seg == NULL) {
666 			sg_temp->final = 1;
667 			cpu_to_hw_sg(sg_temp);
668 			break;
669 		}
670 		cpu_to_hw_sg(sg_temp);
671 	}
672 	return 0;
673 }
674 
675 /* Handle mbufs which are not segmented (non SG) */
676 static inline void
677 tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
678 			    struct dpaa_bp_info *bp_info,
679 			    struct qm_fd *fd_arr)
680 {
681 	struct rte_mbuf *mi = NULL;
682 
683 	if (RTE_MBUF_DIRECT(mbuf)) {
684 		if (rte_mbuf_refcnt_read(mbuf) > 1) {
685 			/* In case of direct mbuf and mbuf being cloned,
686 			 * BMAN should _not_ release buffer.
687 			 */
688 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
689 			/* Buffer should be releasd by EAL */
690 			rte_mbuf_refcnt_update(mbuf, -1);
691 		} else {
692 			/* In case of direct mbuf and no cloning, mbuf can be
693 			 * released by BMAN.
694 			 */
695 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
696 		}
697 	} else {
698 		/* This is data-containing core mbuf: 'mi' */
699 		mi = rte_mbuf_from_indirect(mbuf);
700 		if (rte_mbuf_refcnt_read(mi) > 1) {
701 			/* In case of indirect mbuf, and mbuf being cloned,
702 			 * BMAN should _not_ release it and let EAL release
703 			 * it through pktmbuf_free below.
704 			 */
705 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
706 		} else {
707 			/* In case of indirect mbuf, and no cloning, core mbuf
708 			 * should be released by BMAN.
709 			 * Increate refcnt of core mbuf so that when
710 			 * pktmbuf_free is called and mbuf is released, EAL
711 			 * doesn't try to release core mbuf which would have
712 			 * been released by BMAN.
713 			 */
714 			rte_mbuf_refcnt_update(mi, 1);
715 			DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
716 		}
717 		rte_pktmbuf_free(mbuf);
718 	}
719 
720 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
721 		dpaa_unsegmented_checksum(mbuf, fd_arr);
722 }
723 
724 /* Handle all mbufs on dpaa BMAN managed pool */
725 static inline uint16_t
726 tx_on_dpaa_pool(struct rte_mbuf *mbuf,
727 		struct dpaa_bp_info *bp_info,
728 		struct qm_fd *fd_arr)
729 {
730 	DPAA_DP_LOG(DEBUG, "BMAN offloaded buffer, mbuf: %p", mbuf);
731 
732 	if (mbuf->nb_segs == 1) {
733 		/* Case for non-segmented buffers */
734 		tx_on_dpaa_pool_unsegmented(mbuf, bp_info, fd_arr);
735 	} else if (mbuf->nb_segs > 1 &&
736 		   mbuf->nb_segs <= DPAA_SGT_MAX_ENTRIES) {
737 		if (dpaa_eth_mbuf_to_sg_fd(mbuf, fd_arr, bp_info->bpid)) {
738 			DPAA_PMD_DEBUG("Unable to create Scatter Gather FD");
739 			return 1;
740 		}
741 	} else {
742 		DPAA_PMD_DEBUG("Number of Segments not supported");
743 		return 1;
744 	}
745 
746 	return 0;
747 }
748 
749 /* Handle all mbufs on an external pool (non-dpaa) */
750 static inline uint16_t
751 tx_on_external_pool(struct qman_fq *txq, struct rte_mbuf *mbuf,
752 		    struct qm_fd *fd_arr)
753 {
754 	struct dpaa_if *dpaa_intf = txq->dpaa_intf;
755 	struct rte_mbuf *dmable_mbuf;
756 
757 	DPAA_DP_LOG(DEBUG, "Non-BMAN offloaded buffer."
758 		    "Allocating an offloaded buffer");
759 	dmable_mbuf = dpaa_get_dmable_mbuf(mbuf, dpaa_intf);
760 	if (!dmable_mbuf) {
761 		DPAA_DP_LOG(DEBUG, "no dpaa buffers.");
762 		return 1;
763 	}
764 
765 	DPAA_MBUF_TO_CONTIG_FD(dmable_mbuf, fd_arr, dpaa_intf->bp_info->bpid);
766 
767 	return 0;
768 }
769 
770 uint16_t
771 dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
772 {
773 	struct rte_mbuf *mbuf, *mi = NULL;
774 	struct rte_mempool *mp;
775 	struct dpaa_bp_info *bp_info;
776 	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
777 	uint32_t frames_to_send, loop, sent = 0;
778 	uint16_t state;
779 	int ret;
780 	uint32_t seqn, index, flags[DPAA_TX_BURST_SIZE] = {0};
781 
782 	if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
783 		ret = rte_dpaa_portal_init((void *)0);
784 		if (ret) {
785 			DPAA_PMD_ERR("Failure in affining portal");
786 			return 0;
787 		}
788 	}
789 
790 	DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
791 
792 	while (nb_bufs) {
793 		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
794 				DPAA_TX_BURST_SIZE : nb_bufs;
795 		for (loop = 0; loop < frames_to_send; loop++) {
796 			mbuf = *(bufs++);
797 			if (likely(RTE_MBUF_DIRECT(mbuf))) {
798 				mp = mbuf->pool;
799 				bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
800 				if (likely(mp->ops_index ==
801 						bp_info->dpaa_ops_index &&
802 					mbuf->nb_segs == 1 &&
803 					rte_mbuf_refcnt_read(mbuf) == 1)) {
804 					DPAA_MBUF_TO_CONTIG_FD(mbuf,
805 						&fd_arr[loop], bp_info->bpid);
806 					if (mbuf->ol_flags &
807 						DPAA_TX_CKSUM_OFFLOAD_MASK)
808 						dpaa_unsegmented_checksum(mbuf,
809 							&fd_arr[loop]);
810 					continue;
811 				}
812 			} else {
813 				mi = rte_mbuf_from_indirect(mbuf);
814 				mp = mi->pool;
815 			}
816 
817 			bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
818 			if (likely(mp->ops_index == bp_info->dpaa_ops_index)) {
819 				state = tx_on_dpaa_pool(mbuf, bp_info,
820 							&fd_arr[loop]);
821 				if (unlikely(state)) {
822 					/* Set frames_to_send & nb_bufs so
823 					 * that packets are transmitted till
824 					 * previous frame.
825 					 */
826 					frames_to_send = loop;
827 					nb_bufs = loop;
828 					goto send_pkts;
829 				}
830 			} else {
831 				state = tx_on_external_pool(q, mbuf,
832 							    &fd_arr[loop]);
833 				if (unlikely(state)) {
834 					/* Set frames_to_send & nb_bufs so
835 					 * that packets are transmitted till
836 					 * previous frame.
837 					 */
838 					frames_to_send = loop;
839 					nb_bufs = loop;
840 					goto send_pkts;
841 				}
842 			}
843 			seqn = mbuf->seqn;
844 			if (seqn != DPAA_INVALID_MBUF_SEQN) {
845 				index = seqn - 1;
846 				if (DPAA_PER_LCORE_DQRR_HELD & (1 << index)) {
847 					flags[loop] =
848 					   ((index & QM_EQCR_DCA_IDXMASK) << 8);
849 					flags[loop] |= QMAN_ENQUEUE_FLAG_DCA;
850 					DPAA_PER_LCORE_DQRR_SIZE--;
851 					DPAA_PER_LCORE_DQRR_HELD &=
852 								~(1 << index);
853 				}
854 			}
855 		}
856 
857 send_pkts:
858 		loop = 0;
859 		while (loop < frames_to_send) {
860 			loop += qman_enqueue_multi(q, &fd_arr[loop],
861 						   &flags[loop],
862 						   frames_to_send - loop);
863 		}
864 		nb_bufs -= frames_to_send;
865 		sent += frames_to_send;
866 	}
867 
868 	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", sent, q);
869 
870 	return sent;
871 }
872 
873 uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
874 			      struct rte_mbuf **bufs __rte_unused,
875 		uint16_t nb_bufs __rte_unused)
876 {
877 	DPAA_DP_LOG(DEBUG, "Drop all packets");
878 
879 	/* Drop all incoming packets. No need to free packets here
880 	 * because the rte_eth f/w frees up the packets through tx_buffer
881 	 * callback in case this functions returns count less than nb_bufs
882 	 */
883 	return 0;
884 }
885