1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright 2016 Freescale Semiconductor, Inc. All rights reserved.
16 #include <rte_byteorder.h>
17 #include <rte_common.h>
18 #include <rte_interrupts.h>
20 #include <rte_debug.h>
22 #include <rte_atomic.h>
23 #include <rte_branch_prediction.h>
24 #include <rte_memory.h>
25 #include <rte_tailq.h>
27 #include <rte_alarm.h>
28 #include <rte_ether.h>
29 #include <rte_ethdev_driver.h>
30 #include <rte_malloc.h>
36 #include <rte_eventdev.h>
38 #include "dpaa_ethdev.h"
39 #include "dpaa_rxtx.h"
40 #include <rte_dpaa_bus.h>
41 #include <dpaa_mempool.h>
50 #define DPAA_MBUF_TO_CONTIG_FD(_mbuf, _fd, _bpid) \
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; \
61 #if (defined RTE_LIBRTE_DPAA_DEBUG_DRIVER)
62 static void dpaa_display_frame(const struct qm_fd *fd)
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);
71 ptr = (char *)rte_dpaa_mem_ptov(fd->addr);
73 printf("%02x ", *ptr);
74 for (ii = 1; ii < fd->length20; ii++) {
75 printf("%02x ", *ptr);
83 #define dpaa_display_frame(a)
86 static inline void dpaa_slow_parsing(struct rte_mbuf *m __rte_unused,
87 uint64_t prs __rte_unused)
89 DPAA_DP_LOG(DEBUG, "Slow parsing");
90 /*TBD:XXX: to be implemented*/
93 static inline void dpaa_eth_packet_info(struct rte_mbuf *m, void *fd_virt_addr)
95 struct annotations_t *annot = GET_ANNOTATIONS(fd_virt_addr);
96 uint64_t prs = *((uintptr_t *)(&annot->parse)) & DPAA_PARSE_MASK;
98 DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
101 case DPAA_PKT_TYPE_IPV4:
102 m->packet_type = RTE_PTYPE_L2_ETHER |
105 case DPAA_PKT_TYPE_IPV6:
106 m->packet_type = RTE_PTYPE_L2_ETHER |
109 case DPAA_PKT_TYPE_ETHER:
110 m->packet_type = RTE_PTYPE_L2_ETHER;
112 case DPAA_PKT_TYPE_IPV4_FRAG:
113 case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
114 case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
115 case DPAA_PKT_TYPE_IPV4_FRAG_SCTP:
116 m->packet_type = RTE_PTYPE_L2_ETHER |
117 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_FRAG;
119 case DPAA_PKT_TYPE_IPV6_FRAG:
120 case DPAA_PKT_TYPE_IPV6_FRAG_UDP:
121 case DPAA_PKT_TYPE_IPV6_FRAG_TCP:
122 case DPAA_PKT_TYPE_IPV6_FRAG_SCTP:
123 m->packet_type = RTE_PTYPE_L2_ETHER |
124 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_FRAG;
126 case DPAA_PKT_TYPE_IPV4_EXT:
127 m->packet_type = RTE_PTYPE_L2_ETHER |
128 RTE_PTYPE_L3_IPV4_EXT;
130 case DPAA_PKT_TYPE_IPV6_EXT:
131 m->packet_type = RTE_PTYPE_L2_ETHER |
132 RTE_PTYPE_L3_IPV6_EXT;
134 case DPAA_PKT_TYPE_IPV4_TCP:
135 m->packet_type = RTE_PTYPE_L2_ETHER |
136 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP;
138 case DPAA_PKT_TYPE_IPV6_TCP:
139 m->packet_type = RTE_PTYPE_L2_ETHER |
140 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
142 case DPAA_PKT_TYPE_IPV4_UDP:
143 m->packet_type = RTE_PTYPE_L2_ETHER |
144 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP;
146 case DPAA_PKT_TYPE_IPV6_UDP:
147 m->packet_type = RTE_PTYPE_L2_ETHER |
148 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
150 case DPAA_PKT_TYPE_IPV4_EXT_UDP:
151 m->packet_type = RTE_PTYPE_L2_ETHER |
152 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP;
154 case DPAA_PKT_TYPE_IPV6_EXT_UDP:
155 m->packet_type = RTE_PTYPE_L2_ETHER |
156 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP;
158 case DPAA_PKT_TYPE_IPV4_EXT_TCP:
159 m->packet_type = RTE_PTYPE_L2_ETHER |
160 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP;
162 case DPAA_PKT_TYPE_IPV6_EXT_TCP:
163 m->packet_type = RTE_PTYPE_L2_ETHER |
164 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP;
166 case DPAA_PKT_TYPE_IPV4_SCTP:
167 m->packet_type = RTE_PTYPE_L2_ETHER |
168 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP;
170 case DPAA_PKT_TYPE_IPV6_SCTP:
171 m->packet_type = RTE_PTYPE_L2_ETHER |
172 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
174 case DPAA_PKT_TYPE_NONE:
177 /* More switch cases can be added */
179 dpaa_slow_parsing(m, prs);
182 m->tx_offload = annot->parse.ip_off[0];
183 m->tx_offload |= (annot->parse.l4_off - annot->parse.ip_off[0])
184 << DPAA_PKT_L3_LEN_SHIFT;
186 /* Set the hash values */
187 m->hash.rss = (uint32_t)(annot->hash);
188 /* All packets with Bad checksum are dropped by interface (and
189 * corresponding notification issued to RX error queues).
191 m->ol_flags = PKT_RX_RSS_HASH | PKT_RX_IP_CKSUM_GOOD;
193 /* Check if Vlan is present */
194 if (prs & DPAA_PARSE_VLAN_MASK)
195 m->ol_flags |= PKT_RX_VLAN;
196 /* Packet received without stripping the vlan */
199 static inline void dpaa_checksum(struct rte_mbuf *mbuf)
201 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
202 char *l3_hdr = (char *)eth_hdr + mbuf->l2_len;
203 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
204 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
206 DPAA_DP_LOG(DEBUG, "Calculating checksum for mbuf: %p", mbuf);
208 if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
209 ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
210 RTE_PTYPE_L3_IPV4_EXT)) {
211 ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
212 ipv4_hdr->hdr_checksum = 0;
213 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
214 } else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
215 RTE_PTYPE_L3_IPV6) ||
216 ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
217 RTE_PTYPE_L3_IPV6_EXT))
218 ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
220 if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP) {
221 struct tcp_hdr *tcp_hdr = (struct tcp_hdr *)(l3_hdr +
224 if (eth_hdr->ether_type == htons(ETHER_TYPE_IPv4))
225 tcp_hdr->cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
227 else /* assume ethertype == ETHER_TYPE_IPv6 */
228 tcp_hdr->cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
230 } else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) ==
232 struct udp_hdr *udp_hdr = (struct udp_hdr *)(l3_hdr +
234 udp_hdr->dgram_cksum = 0;
235 if (eth_hdr->ether_type == htons(ETHER_TYPE_IPv4))
236 udp_hdr->dgram_cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
238 else /* assume ethertype == ETHER_TYPE_IPv6 */
239 udp_hdr->dgram_cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
244 static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
245 struct qm_fd *fd, char *prs_buf)
247 struct dpaa_eth_parse_results_t *prs;
249 DPAA_DP_LOG(DEBUG, " Offloading checksum for mbuf: %p", mbuf);
251 prs = GET_TX_PRS(prs_buf);
254 if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
255 ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
256 RTE_PTYPE_L3_IPV4_EXT))
257 prs->l3r = DPAA_L3_PARSE_RESULT_IPV4;
258 else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
259 RTE_PTYPE_L3_IPV6) ||
260 ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
261 RTE_PTYPE_L3_IPV6_EXT))
262 prs->l3r = DPAA_L3_PARSE_RESULT_IPV6;
264 if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP)
265 prs->l4r = DPAA_L4_PARSE_RESULT_TCP;
266 else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP)
267 prs->l4r = DPAA_L4_PARSE_RESULT_UDP;
269 prs->ip_off[0] = mbuf->l2_len;
270 prs->l4_off = mbuf->l3_len + mbuf->l2_len;
271 /* Enable L3 (and L4, if TCP or UDP) HW checksum*/
272 fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
276 dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
278 if (!mbuf->packet_type) {
279 struct rte_net_hdr_lens hdr_lens;
281 mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
282 RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
283 | RTE_PTYPE_L4_MASK);
284 mbuf->l2_len = hdr_lens.l2_len;
285 mbuf->l3_len = hdr_lens.l3_len;
287 if (mbuf->data_off < (DEFAULT_TX_ICEOF +
288 sizeof(struct dpaa_eth_parse_results_t))) {
289 DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
290 "Not enough Headroom "
291 "space for correct Checksum offload."
292 "So Calculating checksum in Software.");
295 dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
300 dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
302 struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
303 struct rte_mbuf *first_seg, *prev_seg, *cur_seg, *temp;
304 struct qm_sg_entry *sgt, *sg_temp;
305 void *vaddr, *sg_vaddr;
307 uint8_t fd_offset = fd->offset;
309 vaddr = DPAA_MEMPOOL_PTOV(bp_info, qm_fd_addr(fd));
311 DPAA_PMD_ERR("unable to convert physical address");
314 sgt = vaddr + fd_offset;
316 hw_sg_to_cpu(sg_temp);
317 temp = (struct rte_mbuf *)((char *)vaddr - bp_info->meta_data_size);
318 sg_vaddr = DPAA_MEMPOOL_PTOV(bp_info, qm_sg_entry_get64(sg_temp));
320 first_seg = (struct rte_mbuf *)((char *)sg_vaddr -
321 bp_info->meta_data_size);
322 first_seg->data_off = sg_temp->offset;
323 first_seg->data_len = sg_temp->length;
324 first_seg->pkt_len = sg_temp->length;
325 rte_mbuf_refcnt_set(first_seg, 1);
327 first_seg->port = ifid;
328 first_seg->nb_segs = 1;
329 first_seg->ol_flags = 0;
330 prev_seg = first_seg;
331 while (i < DPAA_SGT_MAX_ENTRIES) {
333 hw_sg_to_cpu(sg_temp);
334 sg_vaddr = DPAA_MEMPOOL_PTOV(bp_info,
335 qm_sg_entry_get64(sg_temp));
336 cur_seg = (struct rte_mbuf *)((char *)sg_vaddr -
337 bp_info->meta_data_size);
338 cur_seg->data_off = sg_temp->offset;
339 cur_seg->data_len = sg_temp->length;
340 first_seg->pkt_len += sg_temp->length;
341 first_seg->nb_segs += 1;
342 rte_mbuf_refcnt_set(cur_seg, 1);
343 prev_seg->next = cur_seg;
344 if (sg_temp->final) {
345 cur_seg->next = NULL;
350 DPAA_DP_LOG(DEBUG, "Received an SG frame len =%d, num_sg =%d",
351 first_seg->pkt_len, first_seg->nb_segs);
353 dpaa_eth_packet_info(first_seg, vaddr);
354 rte_pktmbuf_free_seg(temp);
359 static inline struct rte_mbuf *
360 dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
362 struct rte_mbuf *mbuf;
363 struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
366 (fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
370 if (unlikely(format == qm_fd_sg))
371 return dpaa_eth_sg_to_mbuf(fd, ifid);
373 offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
374 length = fd->opaque & DPAA_FD_LENGTH_MASK;
376 DPAA_DP_LOG(DEBUG, " FD--->MBUF off %d len = %d", offset, length);
378 /* Ignoring case when format != qm_fd_contig */
379 dpaa_display_frame(fd);
380 ptr = DPAA_MEMPOOL_PTOV(bp_info, qm_fd_addr(fd));
382 mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
383 /* Prefetch the Parse results and packet data to L1 */
384 rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
386 mbuf->data_off = offset;
387 mbuf->data_len = length;
388 mbuf->pkt_len = length;
394 rte_mbuf_refcnt_set(mbuf, 1);
395 dpaa_eth_packet_info(mbuf, mbuf->buf_addr);
400 /* Specific for LS1043 */
402 dpaa_rx_cb_no_prefetch(struct qman_fq **fq, struct qm_dqrr_entry **dqrr,
403 void **bufs, int num_bufs)
405 struct rte_mbuf *mbuf;
406 struct dpaa_bp_info *bp_info;
407 const struct qm_fd *fd;
409 struct dpaa_if *dpaa_intf;
414 bp_info = DPAA_BPID_TO_POOL_INFO(dqrr[0]->fd.bpid);
415 ptr = rte_dpaa_mem_ptov(qm_fd_addr(&dqrr[0]->fd));
416 rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
417 bufs[0] = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
419 for (i = 0; i < num_bufs; i++) {
420 if (i < num_bufs - 1) {
421 bp_info = DPAA_BPID_TO_POOL_INFO(dqrr[i + 1]->fd.bpid);
422 ptr = rte_dpaa_mem_ptov(qm_fd_addr(&dqrr[i + 1]->fd));
423 rte_prefetch0((void *)((uint8_t *)ptr +
425 bufs[i + 1] = (struct rte_mbuf *)((char *)ptr -
426 bp_info->meta_data_size);
430 dpaa_intf = fq[0]->dpaa_intf;
432 format = (fd->opaque & DPAA_FD_FORMAT_MASK) >>
433 DPAA_FD_FORMAT_SHIFT;
434 if (unlikely(format == qm_fd_sg)) {
435 bufs[i] = dpaa_eth_sg_to_mbuf(fd, dpaa_intf->ifid);
439 offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >>
440 DPAA_FD_OFFSET_SHIFT;
441 length = fd->opaque & DPAA_FD_LENGTH_MASK;
444 mbuf->data_off = offset;
445 mbuf->data_len = length;
446 mbuf->pkt_len = length;
447 mbuf->port = dpaa_intf->ifid;
452 rte_mbuf_refcnt_set(mbuf, 1);
453 dpaa_eth_packet_info(mbuf, mbuf->buf_addr);
458 dpaa_rx_cb(struct qman_fq **fq, struct qm_dqrr_entry **dqrr,
459 void **bufs, int num_bufs)
461 struct rte_mbuf *mbuf;
462 const struct qm_fd *fd;
463 struct dpaa_if *dpaa_intf;
468 for (i = 0; i < num_bufs; i++) {
470 dpaa_intf = fq[0]->dpaa_intf;
472 format = (fd->opaque & DPAA_FD_FORMAT_MASK) >>
473 DPAA_FD_FORMAT_SHIFT;
474 if (unlikely(format == qm_fd_sg)) {
475 bufs[i] = dpaa_eth_sg_to_mbuf(fd, dpaa_intf->ifid);
479 offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >>
480 DPAA_FD_OFFSET_SHIFT;
481 length = fd->opaque & DPAA_FD_LENGTH_MASK;
484 mbuf->data_off = offset;
485 mbuf->data_len = length;
486 mbuf->pkt_len = length;
487 mbuf->port = dpaa_intf->ifid;
492 rte_mbuf_refcnt_set(mbuf, 1);
493 dpaa_eth_packet_info(mbuf, mbuf->buf_addr);
497 void dpaa_rx_cb_prepare(struct qm_dqrr_entry *dq, void **bufs)
499 struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(dq->fd.bpid);
500 void *ptr = rte_dpaa_mem_ptov(qm_fd_addr(&dq->fd));
502 /* In case of LS1046, annotation stashing is disabled due to L2 cache
503 * being bottleneck in case of multicore scanario for this platform.
504 * So we prefetch the annoation beforehand, so that it is available
505 * in cache when accessed.
507 rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
509 *bufs = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
513 dpaa_eth_queue_portal_rx(struct qman_fq *fq,
514 struct rte_mbuf **bufs,
519 if (unlikely(fq->qp == NULL)) {
520 ret = rte_dpaa_portal_fq_init((void *)0, fq);
522 DPAA_PMD_ERR("Failure in affining portal %d", ret);
527 return qman_portal_poll_rx(nb_bufs, (void **)bufs, fq->qp);
530 enum qman_cb_dqrr_result
531 dpaa_rx_cb_parallel(void *event,
532 struct qman_portal *qm __always_unused,
534 const struct qm_dqrr_entry *dqrr,
537 u32 ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
538 struct rte_mbuf *mbuf;
539 struct rte_event *ev = (struct rte_event *)event;
541 mbuf = dpaa_eth_fd_to_mbuf(&dqrr->fd, ifid);
542 ev->event_ptr = (void *)mbuf;
543 ev->flow_id = fq->ev.flow_id;
544 ev->sub_event_type = fq->ev.sub_event_type;
545 ev->event_type = RTE_EVENT_TYPE_ETHDEV;
546 ev->op = RTE_EVENT_OP_NEW;
547 ev->sched_type = fq->ev.sched_type;
548 ev->queue_id = fq->ev.queue_id;
549 ev->priority = fq->ev.priority;
550 ev->impl_opaque = (uint8_t)DPAA_INVALID_MBUF_SEQN;
551 mbuf->seqn = DPAA_INVALID_MBUF_SEQN;
554 return qman_cb_dqrr_consume;
557 enum qman_cb_dqrr_result
558 dpaa_rx_cb_atomic(void *event,
559 struct qman_portal *qm __always_unused,
561 const struct qm_dqrr_entry *dqrr,
565 u32 ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
566 struct rte_mbuf *mbuf;
567 struct rte_event *ev = (struct rte_event *)event;
569 mbuf = dpaa_eth_fd_to_mbuf(&dqrr->fd, ifid);
570 ev->event_ptr = (void *)mbuf;
571 ev->flow_id = fq->ev.flow_id;
572 ev->sub_event_type = fq->ev.sub_event_type;
573 ev->event_type = RTE_EVENT_TYPE_ETHDEV;
574 ev->op = RTE_EVENT_OP_NEW;
575 ev->sched_type = fq->ev.sched_type;
576 ev->queue_id = fq->ev.queue_id;
577 ev->priority = fq->ev.priority;
579 /* Save active dqrr entries */
580 index = DQRR_PTR2IDX(dqrr);
581 DPAA_PER_LCORE_DQRR_SIZE++;
582 DPAA_PER_LCORE_DQRR_HELD |= 1 << index;
583 DPAA_PER_LCORE_DQRR_MBUF(index) = mbuf;
584 ev->impl_opaque = index + 1;
585 mbuf->seqn = (uint32_t)index + 1;
588 return qman_cb_dqrr_defer;
591 uint16_t dpaa_eth_queue_rx(void *q,
592 struct rte_mbuf **bufs,
595 struct qman_fq *fq = q;
596 struct qm_dqrr_entry *dq;
597 uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
598 int num_rx_bufs, ret;
599 uint32_t vdqcr_flags = 0;
601 if (likely(fq->is_static))
602 return dpaa_eth_queue_portal_rx(fq, bufs, nb_bufs);
604 if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
605 ret = rte_dpaa_portal_init((void *)0);
607 DPAA_PMD_ERR("Failure in affining portal");
612 /* Until request for four buffers, we provide exact number of buffers.
613 * Otherwise we do not set the QM_VDQCR_EXACT flag.
614 * Not setting QM_VDQCR_EXACT flag can provide two more buffers than
615 * requested, so we request two less in this case.
618 vdqcr_flags = QM_VDQCR_EXACT;
619 num_rx_bufs = nb_bufs;
621 num_rx_bufs = nb_bufs > DPAA_MAX_DEQUEUE_NUM_FRAMES ?
622 (DPAA_MAX_DEQUEUE_NUM_FRAMES - 2) : (nb_bufs - 2);
624 ret = qman_set_vdq(fq, num_rx_bufs, vdqcr_flags);
629 dq = qman_dequeue(fq);
632 bufs[num_rx++] = dpaa_eth_fd_to_mbuf(&dq->fd, ifid);
633 qman_dqrr_consume(fq, dq);
634 } while (fq->flags & QMAN_FQ_STATE_VDQCR);
639 static void *dpaa_get_pktbuf(struct dpaa_bp_info *bp_info)
643 struct bm_buffer bufs;
645 ret = bman_acquire(bp_info->bp, &bufs, 1, 0);
647 DPAA_PMD_WARN("Failed to allocate buffers %d", ret);
651 DPAA_DP_LOG(DEBUG, "got buffer 0x%" PRIx64 " from pool %d",
652 (uint64_t)bufs.addr, bufs.bpid);
654 buf = (size_t)DPAA_MEMPOOL_PTOV(bp_info, bufs.addr)
655 - bp_info->meta_data_size;
663 static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
664 struct dpaa_if *dpaa_intf)
666 struct rte_mbuf *dpaa_mbuf;
668 /* allocate pktbuffer on bpid for dpaa port */
669 dpaa_mbuf = dpaa_get_pktbuf(dpaa_intf->bp_info);
673 memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + RTE_PKTMBUF_HEADROOM, (void *)
674 ((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
676 /* Copy only the required fields */
677 dpaa_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
678 dpaa_mbuf->pkt_len = mbuf->pkt_len;
679 dpaa_mbuf->ol_flags = mbuf->ol_flags;
680 dpaa_mbuf->packet_type = mbuf->packet_type;
681 dpaa_mbuf->tx_offload = mbuf->tx_offload;
682 rte_pktmbuf_free(mbuf);
687 dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
691 struct rte_mbuf *cur_seg = mbuf, *prev_seg = NULL;
692 struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(bpid);
693 struct rte_mbuf *temp, *mi;
694 struct qm_sg_entry *sg_temp, *sgt;
697 DPAA_DP_LOG(DEBUG, "Creating SG FD to transmit");
699 temp = rte_pktmbuf_alloc(bp_info->mp);
701 DPAA_PMD_ERR("Failure in allocation of mbuf");
704 if (temp->buf_len < ((mbuf->nb_segs * sizeof(struct qm_sg_entry))
706 DPAA_PMD_ERR("Insufficient space in mbuf for SG entries");
713 if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
714 if (!mbuf->packet_type) {
715 struct rte_net_hdr_lens hdr_lens;
717 mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
718 RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
719 | RTE_PTYPE_L4_MASK);
720 mbuf->l2_len = hdr_lens.l2_len;
721 mbuf->l3_len = hdr_lens.l3_len;
723 if (temp->data_off < DEFAULT_TX_ICEOF
724 + sizeof(struct dpaa_eth_parse_results_t))
725 temp->data_off = DEFAULT_TX_ICEOF
726 + sizeof(struct dpaa_eth_parse_results_t);
727 dcbz_64(temp->buf_addr);
728 dpaa_checksum_offload(mbuf, fd, temp->buf_addr);
731 sgt = temp->buf_addr + temp->data_off;
732 fd->format = QM_FD_SG;
733 fd->addr = temp->buf_iova;
734 fd->offset = temp->data_off;
736 fd->length20 = mbuf->pkt_len;
738 while (i < DPAA_SGT_MAX_ENTRIES) {
742 sg_temp->addr = cur_seg->buf_iova;
743 sg_temp->offset = cur_seg->data_off;
744 sg_temp->length = cur_seg->data_len;
745 if (RTE_MBUF_DIRECT(cur_seg)) {
746 if (rte_mbuf_refcnt_read(cur_seg) > 1) {
747 /*If refcnt > 1, invalid bpid is set to ensure
748 * buffer is not freed by HW.
750 sg_temp->bpid = 0xff;
751 rte_mbuf_refcnt_update(cur_seg, -1);
754 DPAA_MEMPOOL_TO_BPID(cur_seg->pool);
756 cur_seg = cur_seg->next;
758 /* Get owner MBUF from indirect buffer */
759 mi = rte_mbuf_from_indirect(cur_seg);
760 if (rte_mbuf_refcnt_read(mi) > 1) {
761 /*If refcnt > 1, invalid bpid is set to ensure
762 * owner buffer is not freed by HW.
764 sg_temp->bpid = 0xff;
766 sg_temp->bpid = DPAA_MEMPOOL_TO_BPID(mi->pool);
767 rte_mbuf_refcnt_update(mi, 1);
770 cur_seg = cur_seg->next;
771 prev_seg->next = NULL;
772 rte_pktmbuf_free(prev_seg);
774 if (cur_seg == NULL) {
776 cpu_to_hw_sg(sg_temp);
779 cpu_to_hw_sg(sg_temp);
784 /* Handle mbufs which are not segmented (non SG) */
786 tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
787 struct dpaa_bp_info *bp_info,
788 struct qm_fd *fd_arr)
790 struct rte_mbuf *mi = NULL;
792 if (RTE_MBUF_DIRECT(mbuf)) {
793 if (rte_mbuf_refcnt_read(mbuf) > 1) {
794 /* In case of direct mbuf and mbuf being cloned,
795 * BMAN should _not_ release buffer.
797 DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
798 /* Buffer should be releasd by EAL */
799 rte_mbuf_refcnt_update(mbuf, -1);
801 /* In case of direct mbuf and no cloning, mbuf can be
804 DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
807 /* This is data-containing core mbuf: 'mi' */
808 mi = rte_mbuf_from_indirect(mbuf);
809 if (rte_mbuf_refcnt_read(mi) > 1) {
810 /* In case of indirect mbuf, and mbuf being cloned,
811 * BMAN should _not_ release it and let EAL release
812 * it through pktmbuf_free below.
814 DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
816 /* In case of indirect mbuf, and no cloning, core mbuf
817 * should be released by BMAN.
818 * Increate refcnt of core mbuf so that when
819 * pktmbuf_free is called and mbuf is released, EAL
820 * doesn't try to release core mbuf which would have
821 * been released by BMAN.
823 rte_mbuf_refcnt_update(mi, 1);
824 DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
826 rte_pktmbuf_free(mbuf);
829 if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
830 dpaa_unsegmented_checksum(mbuf, fd_arr);
833 /* Handle all mbufs on dpaa BMAN managed pool */
834 static inline uint16_t
835 tx_on_dpaa_pool(struct rte_mbuf *mbuf,
836 struct dpaa_bp_info *bp_info,
837 struct qm_fd *fd_arr)
839 DPAA_DP_LOG(DEBUG, "BMAN offloaded buffer, mbuf: %p", mbuf);
841 if (mbuf->nb_segs == 1) {
842 /* Case for non-segmented buffers */
843 tx_on_dpaa_pool_unsegmented(mbuf, bp_info, fd_arr);
844 } else if (mbuf->nb_segs > 1 &&
845 mbuf->nb_segs <= DPAA_SGT_MAX_ENTRIES) {
846 if (dpaa_eth_mbuf_to_sg_fd(mbuf, fd_arr, bp_info->bpid)) {
847 DPAA_PMD_DEBUG("Unable to create Scatter Gather FD");
851 DPAA_PMD_DEBUG("Number of Segments not supported");
858 /* Handle all mbufs on an external pool (non-dpaa) */
859 static inline uint16_t
860 tx_on_external_pool(struct qman_fq *txq, struct rte_mbuf *mbuf,
861 struct qm_fd *fd_arr)
863 struct dpaa_if *dpaa_intf = txq->dpaa_intf;
864 struct rte_mbuf *dmable_mbuf;
866 DPAA_DP_LOG(DEBUG, "Non-BMAN offloaded buffer."
867 "Allocating an offloaded buffer");
868 dmable_mbuf = dpaa_get_dmable_mbuf(mbuf, dpaa_intf);
870 DPAA_DP_LOG(DEBUG, "no dpaa buffers.");
874 DPAA_MBUF_TO_CONTIG_FD(dmable_mbuf, fd_arr, dpaa_intf->bp_info->bpid);
875 if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
876 dpaa_unsegmented_checksum(mbuf, fd_arr);
882 dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
884 struct rte_mbuf *mbuf, *mi = NULL;
885 struct rte_mempool *mp;
886 struct dpaa_bp_info *bp_info;
887 struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
888 uint32_t frames_to_send, loop, sent = 0;
891 uint32_t seqn, index, flags[DPAA_TX_BURST_SIZE] = {0};
893 if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
894 ret = rte_dpaa_portal_init((void *)0);
896 DPAA_PMD_ERR("Failure in affining portal");
901 DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
904 frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
905 DPAA_TX_BURST_SIZE : nb_bufs;
906 for (loop = 0; loop < frames_to_send; loop++) {
909 if (seqn != DPAA_INVALID_MBUF_SEQN) {
911 if (DPAA_PER_LCORE_DQRR_HELD & (1 << index)) {
913 ((index & QM_EQCR_DCA_IDXMASK) << 8);
914 flags[loop] |= QMAN_ENQUEUE_FLAG_DCA;
915 DPAA_PER_LCORE_DQRR_SIZE--;
916 DPAA_PER_LCORE_DQRR_HELD &=
921 if (likely(RTE_MBUF_DIRECT(mbuf))) {
923 bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
924 if (likely(mp->ops_index ==
925 bp_info->dpaa_ops_index &&
926 mbuf->nb_segs == 1 &&
927 rte_mbuf_refcnt_read(mbuf) == 1)) {
928 DPAA_MBUF_TO_CONTIG_FD(mbuf,
929 &fd_arr[loop], bp_info->bpid);
931 DPAA_TX_CKSUM_OFFLOAD_MASK)
932 dpaa_unsegmented_checksum(mbuf,
937 mi = rte_mbuf_from_indirect(mbuf);
941 bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
942 if (likely(mp->ops_index == bp_info->dpaa_ops_index)) {
943 state = tx_on_dpaa_pool(mbuf, bp_info,
945 if (unlikely(state)) {
946 /* Set frames_to_send & nb_bufs so
947 * that packets are transmitted till
950 frames_to_send = loop;
955 state = tx_on_external_pool(q, mbuf,
957 if (unlikely(state)) {
958 /* Set frames_to_send & nb_bufs so
959 * that packets are transmitted till
962 frames_to_send = loop;
971 while (loop < frames_to_send) {
972 loop += qman_enqueue_multi(q, &fd_arr[loop],
974 frames_to_send - loop);
976 nb_bufs -= frames_to_send;
977 sent += frames_to_send;
980 DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", sent, q);
985 uint16_t dpaa_eth_tx_drop_all(void *q __rte_unused,
986 struct rte_mbuf **bufs __rte_unused,
987 uint16_t nb_bufs __rte_unused)
989 DPAA_DP_LOG(DEBUG, "Drop all packets");
991 /* Drop all incoming packets. No need to free packets here
992 * because the rte_eth f/w frees up the packets through tx_buffer
993 * callback in case this functions returns count less than nb_bufs