1 /* Copyright 2008-2016 Cisco Systems, Inc. All rights reserved.
2 * Copyright 2007 Nuova Systems, Inc. All rights reserved.
4 * Copyright (c) 2014, Cisco Systems, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
23 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
25 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
27 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
29 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 #include <rte_ethdev.h>
35 #include <rte_prefetch.h>
37 #include "enic_compat.h"
38 #include "rq_enet_desc.h"
40 #include <rte_ether.h>
44 #define RTE_PMD_USE_PREFETCH
46 #ifdef RTE_PMD_USE_PREFETCH
47 /*Prefetch a cache line into all cache levels. */
48 #define rte_enic_prefetch(p) rte_prefetch0(p)
50 #define rte_enic_prefetch(p) do {} while (0)
53 #ifdef RTE_PMD_PACKET_PREFETCH
54 #define rte_packet_prefetch(p) rte_prefetch1(p)
56 #define rte_packet_prefetch(p) do {} while (0)
59 static inline uint16_t
60 enic_cq_rx_desc_ciflags(struct cq_enet_rq_desc *crd)
62 return le16_to_cpu(crd->completed_index_flags) & ~CQ_DESC_COMP_NDX_MASK;
65 static inline uint16_t
66 enic_cq_rx_desc_bwflags(struct cq_enet_rq_desc *crd)
68 return le16_to_cpu(crd->bytes_written_flags) &
69 ~CQ_ENET_RQ_DESC_BYTES_WRITTEN_MASK;
73 enic_cq_rx_desc_packet_error(uint16_t bwflags)
75 return (bwflags & CQ_ENET_RQ_DESC_FLAGS_TRUNCATED) ==
76 CQ_ENET_RQ_DESC_FLAGS_TRUNCATED;
80 enic_cq_rx_desc_eop(uint16_t ciflags)
82 return (ciflags & CQ_ENET_RQ_DESC_FLAGS_EOP)
83 == CQ_ENET_RQ_DESC_FLAGS_EOP;
87 enic_cq_rx_desc_csum_not_calc(struct cq_enet_rq_desc *cqrd)
89 return (le16_to_cpu(cqrd->q_number_rss_type_flags) &
90 CQ_ENET_RQ_DESC_FLAGS_CSUM_NOT_CALC) ==
91 CQ_ENET_RQ_DESC_FLAGS_CSUM_NOT_CALC;
95 enic_cq_rx_desc_ipv4_csum_ok(struct cq_enet_rq_desc *cqrd)
97 return (cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_IPV4_CSUM_OK) ==
98 CQ_ENET_RQ_DESC_FLAGS_IPV4_CSUM_OK;
101 static inline uint8_t
102 enic_cq_rx_desc_tcp_udp_csum_ok(struct cq_enet_rq_desc *cqrd)
104 return (cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_TCP_UDP_CSUM_OK) ==
105 CQ_ENET_RQ_DESC_FLAGS_TCP_UDP_CSUM_OK;
108 static inline uint8_t
109 enic_cq_rx_desc_rss_type(struct cq_enet_rq_desc *cqrd)
111 return (uint8_t)((le16_to_cpu(cqrd->q_number_rss_type_flags) >>
112 CQ_DESC_Q_NUM_BITS) & CQ_ENET_RQ_DESC_RSS_TYPE_MASK);
115 static inline uint32_t
116 enic_cq_rx_desc_rss_hash(struct cq_enet_rq_desc *cqrd)
118 return le32_to_cpu(cqrd->rss_hash);
121 static inline uint16_t
122 enic_cq_rx_desc_vlan(struct cq_enet_rq_desc *cqrd)
124 return le16_to_cpu(cqrd->vlan);
127 static inline uint16_t
128 enic_cq_rx_desc_n_bytes(struct cq_desc *cqd)
130 struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
131 return le16_to_cpu(cqrd->bytes_written_flags) &
132 CQ_ENET_RQ_DESC_BYTES_WRITTEN_MASK;
135 /* Find the offset to L5. This is needed by enic TSO implementation.
136 * Return 0 if not a TCP packet or can't figure out the length.
138 static inline uint8_t tso_header_len(struct rte_mbuf *mbuf)
140 struct ether_hdr *eh;
142 struct ipv4_hdr *ip4;
143 struct ipv6_hdr *ip6;
148 /* offset past Ethernet header */
149 eh = rte_pktmbuf_mtod(mbuf, struct ether_hdr *);
150 ether_type = eh->ether_type;
151 hdr_len = sizeof(struct ether_hdr);
152 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN)) {
153 vh = rte_pktmbuf_mtod_offset(mbuf, struct vlan_hdr *, hdr_len);
154 ether_type = vh->eth_proto;
155 hdr_len += sizeof(struct vlan_hdr);
158 /* offset past IP header */
159 switch (rte_be_to_cpu_16(ether_type)) {
160 case ETHER_TYPE_IPv4:
161 ip4 = rte_pktmbuf_mtod_offset(mbuf, struct ipv4_hdr *, hdr_len);
162 if (ip4->next_proto_id != IPPROTO_TCP)
164 hdr_len += (ip4->version_ihl & 0xf) * 4;
166 case ETHER_TYPE_IPv6:
167 ip6 = rte_pktmbuf_mtod_offset(mbuf, struct ipv6_hdr *, hdr_len);
168 if (ip6->proto != IPPROTO_TCP)
170 hdr_len += sizeof(struct ipv6_hdr);
176 if ((hdr_len + sizeof(struct tcp_hdr)) > mbuf->pkt_len)
179 /* offset past TCP header */
180 th = rte_pktmbuf_mtod_offset(mbuf, struct tcp_hdr *, hdr_len);
181 hdr_len += (th->data_off >> 4) * 4;
183 if (hdr_len > mbuf->pkt_len)
189 static inline uint8_t
190 enic_cq_rx_check_err(struct cq_desc *cqd)
192 struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
195 bwflags = enic_cq_rx_desc_bwflags(cqrd);
196 if (unlikely(enic_cq_rx_desc_packet_error(bwflags)))
201 /* Lookup table to translate RX CQ flags to mbuf flags. */
202 static inline uint32_t
203 enic_cq_rx_flags_to_pkt_type(struct cq_desc *cqd)
205 struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
206 uint8_t cqrd_flags = cqrd->flags;
207 static const uint32_t cq_type_table[128] __rte_cache_aligned = {
208 [0x00] = RTE_PTYPE_UNKNOWN,
209 [0x20] = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_NONFRAG,
210 [0x22] = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP,
211 [0x24] = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_TCP,
212 [0x60] = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_FRAG,
213 [0x62] = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP,
214 [0x64] = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_TCP,
215 [0x10] = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_NONFRAG,
216 [0x12] = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_UDP,
217 [0x14] = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_TCP,
218 [0x50] = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_FRAG,
219 [0x52] = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_UDP,
220 [0x54] = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_TCP,
221 /* All others reserved */
223 cqrd_flags &= CQ_ENET_RQ_DESC_FLAGS_IPV4_FRAGMENT
224 | CQ_ENET_RQ_DESC_FLAGS_IPV4 | CQ_ENET_RQ_DESC_FLAGS_IPV6
225 | CQ_ENET_RQ_DESC_FLAGS_TCP | CQ_ENET_RQ_DESC_FLAGS_UDP;
226 return cq_type_table[cqrd_flags];
230 enic_cq_rx_to_pkt_flags(struct cq_desc *cqd, struct rte_mbuf *mbuf)
232 struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
233 uint16_t ciflags, bwflags, pkt_flags = 0, vlan_tci;
234 ciflags = enic_cq_rx_desc_ciflags(cqrd);
235 bwflags = enic_cq_rx_desc_bwflags(cqrd);
236 vlan_tci = enic_cq_rx_desc_vlan(cqrd);
240 /* flags are meaningless if !EOP */
241 if (unlikely(!enic_cq_rx_desc_eop(ciflags)))
242 goto mbuf_flags_done;
244 /* VLAN STRIPPED flag. The L2 packet type updated here also */
245 if (bwflags & CQ_ENET_RQ_DESC_FLAGS_VLAN_STRIPPED) {
246 pkt_flags |= PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED;
247 mbuf->packet_type |= RTE_PTYPE_L2_ETHER;
250 mbuf->packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
252 mbuf->packet_type |= RTE_PTYPE_L2_ETHER;
254 mbuf->vlan_tci = vlan_tci;
257 if (enic_cq_rx_desc_rss_type(cqrd)) {
258 pkt_flags |= PKT_RX_RSS_HASH;
259 mbuf->hash.rss = enic_cq_rx_desc_rss_hash(cqrd);
263 if (mbuf->packet_type & RTE_PTYPE_L3_IPV4) {
264 if (enic_cq_rx_desc_csum_not_calc(cqrd))
265 pkt_flags |= (PKT_RX_IP_CKSUM_UNKNOWN &
266 PKT_RX_L4_CKSUM_UNKNOWN);
269 l4_flags = mbuf->packet_type & RTE_PTYPE_L4_MASK;
271 if (enic_cq_rx_desc_ipv4_csum_ok(cqrd))
272 pkt_flags |= PKT_RX_IP_CKSUM_GOOD;
274 pkt_flags |= PKT_RX_IP_CKSUM_BAD;
276 if (l4_flags & (RTE_PTYPE_L4_UDP | RTE_PTYPE_L4_TCP)) {
277 if (enic_cq_rx_desc_tcp_udp_csum_ok(cqrd))
278 pkt_flags |= PKT_RX_L4_CKSUM_GOOD;
280 pkt_flags |= PKT_RX_L4_CKSUM_BAD;
286 mbuf->ol_flags = pkt_flags;
289 /* dummy receive function to replace actual function in
290 * order to do safe reconfiguration operations.
293 enic_dummy_recv_pkts(__rte_unused void *rx_queue,
294 __rte_unused struct rte_mbuf **rx_pkts,
295 __rte_unused uint16_t nb_pkts)
301 enic_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
304 struct vnic_rq *sop_rq = rx_queue;
305 struct vnic_rq *data_rq;
307 struct enic *enic = vnic_dev_priv(sop_rq->vdev);
311 struct rte_mbuf *nmb, *rxmb;
314 volatile struct cq_desc *cqd_ptr;
317 struct rte_mbuf *first_seg = sop_rq->pkt_first_seg;
318 struct rte_mbuf *last_seg = sop_rq->pkt_last_seg;
320 cq = &enic->cq[enic_cq_rq(enic, sop_rq->index)];
321 cq_idx = cq->to_clean; /* index of cqd, rqd, mbuf_table */
322 cqd_ptr = (struct cq_desc *)(cq->ring.descs) + cq_idx;
324 data_rq = &enic->rq[sop_rq->data_queue_idx];
326 while (nb_rx < nb_pkts) {
327 volatile struct rq_enet_desc *rqd_ptr;
330 uint8_t packet_error;
333 /* Check for pkts available */
334 color = (cqd_ptr->type_color >> CQ_DESC_COLOR_SHIFT)
335 & CQ_DESC_COLOR_MASK;
336 if (color == cq->last_color)
339 /* Get the cq descriptor and extract rq info from it */
341 rq_num = cqd.q_number & CQ_DESC_Q_NUM_MASK;
342 rq_idx = cqd.completed_index & CQ_DESC_COMP_NDX_MASK;
344 rq = &enic->rq[rq_num];
345 rqd_ptr = ((struct rq_enet_desc *)rq->ring.descs) + rq_idx;
347 /* allocate a new mbuf */
348 nmb = rte_mbuf_raw_alloc(rq->mp);
350 rte_atomic64_inc(&enic->soft_stats.rx_nombuf);
354 /* A packet error means descriptor and data are untrusted */
355 packet_error = enic_cq_rx_check_err(&cqd);
357 /* Get the mbuf to return and replace with one just allocated */
358 rxmb = rq->mbuf_ring[rq_idx];
359 rq->mbuf_ring[rq_idx] = nmb;
361 /* Increment cqd, rqd, mbuf_table index */
363 if (unlikely(cq_idx == cq->ring.desc_count)) {
365 cq->last_color = cq->last_color ? 0 : 1;
368 /* Prefetch next mbuf & desc while processing current one */
369 cqd_ptr = (struct cq_desc *)(cq->ring.descs) + cq_idx;
370 rte_enic_prefetch(cqd_ptr);
372 ciflags = enic_cq_rx_desc_ciflags(
373 (struct cq_enet_rq_desc *)&cqd);
375 /* Push descriptor for newly allocated mbuf */
376 nmb->data_off = RTE_PKTMBUF_HEADROOM;
377 dma_addr = (dma_addr_t)(nmb->buf_physaddr +
378 RTE_PKTMBUF_HEADROOM);
379 rq_enet_desc_enc(rqd_ptr, dma_addr,
380 (rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
381 : RQ_ENET_TYPE_NOT_SOP),
382 nmb->buf_len - RTE_PKTMBUF_HEADROOM);
384 /* Fill in the rest of the mbuf */
385 seg_length = enic_cq_rx_desc_n_bytes(&cqd);
389 first_seg->nb_segs = 1;
390 first_seg->pkt_len = seg_length;
392 first_seg->pkt_len = (uint16_t)(first_seg->pkt_len
394 first_seg->nb_segs++;
395 last_seg->next = rxmb;
399 rxmb->port = enic->port_id;
400 rxmb->data_len = seg_length;
404 if (!(enic_cq_rx_desc_eop(ciflags))) {
409 /* cq rx flags are only valid if eop bit is set */
410 first_seg->packet_type = enic_cq_rx_flags_to_pkt_type(&cqd);
411 enic_cq_rx_to_pkt_flags(&cqd, first_seg);
413 if (unlikely(packet_error)) {
414 rte_pktmbuf_free(first_seg);
415 rte_atomic64_inc(&enic->soft_stats.rx_packet_errors);
420 /* prefetch mbuf data for caller */
421 rte_packet_prefetch(RTE_PTR_ADD(first_seg->buf_addr,
422 RTE_PKTMBUF_HEADROOM));
424 /* store the mbuf address into the next entry of the array */
425 rx_pkts[nb_rx++] = first_seg;
428 sop_rq->pkt_first_seg = first_seg;
429 sop_rq->pkt_last_seg = last_seg;
431 cq->to_clean = cq_idx;
433 if ((sop_rq->rx_nb_hold + data_rq->rx_nb_hold) >
434 sop_rq->rx_free_thresh) {
435 if (data_rq->in_use) {
436 data_rq->posted_index =
437 enic_ring_add(data_rq->ring.desc_count,
438 data_rq->posted_index,
439 data_rq->rx_nb_hold);
440 data_rq->rx_nb_hold = 0;
442 sop_rq->posted_index = enic_ring_add(sop_rq->ring.desc_count,
443 sop_rq->posted_index,
445 sop_rq->rx_nb_hold = 0;
449 iowrite32_relaxed(data_rq->posted_index,
450 &data_rq->ctrl->posted_index);
451 rte_compiler_barrier();
452 iowrite32_relaxed(sop_rq->posted_index,
453 &sop_rq->ctrl->posted_index);
460 static inline void enic_free_wq_bufs(struct vnic_wq *wq, u16 completed_index)
462 struct vnic_wq_buf *buf;
463 struct rte_mbuf *m, *free[ENIC_MAX_WQ_DESCS];
464 unsigned int nb_to_free, nb_free = 0, i;
465 struct rte_mempool *pool;
466 unsigned int tail_idx;
467 unsigned int desc_count = wq->ring.desc_count;
469 nb_to_free = enic_ring_sub(desc_count, wq->tail_idx, completed_index)
471 tail_idx = wq->tail_idx;
472 buf = &wq->bufs[tail_idx];
473 pool = ((struct rte_mbuf *)buf->mb)->pool;
474 for (i = 0; i < nb_to_free; i++) {
475 buf = &wq->bufs[tail_idx];
476 m = __rte_pktmbuf_prefree_seg((struct rte_mbuf *)(buf->mb));
479 if (unlikely(m == NULL)) {
480 tail_idx = enic_ring_incr(desc_count, tail_idx);
484 if (likely(m->pool == pool)) {
485 RTE_ASSERT(nb_free < ENIC_MAX_WQ_DESCS);
488 rte_mempool_put_bulk(pool, (void *)free, nb_free);
493 tail_idx = enic_ring_incr(desc_count, tail_idx);
496 rte_mempool_put_bulk(pool, (void **)free, nb_free);
498 wq->tail_idx = tail_idx;
499 wq->ring.desc_avail += nb_to_free;
502 unsigned int enic_cleanup_wq(__rte_unused struct enic *enic, struct vnic_wq *wq)
506 completed_index = *((uint32_t *)wq->cqmsg_rz->addr) & 0xffff;
508 if (wq->last_completed_index != completed_index) {
509 enic_free_wq_bufs(wq, completed_index);
510 wq->last_completed_index = completed_index;
515 uint16_t enic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
519 unsigned int pkt_len, data_len;
520 unsigned int nb_segs;
521 struct rte_mbuf *tx_pkt;
522 struct vnic_wq *wq = (struct vnic_wq *)tx_queue;
523 struct enic *enic = vnic_dev_priv(wq->vdev);
524 unsigned short vlan_id;
526 uint64_t ol_flags_mask;
527 unsigned int wq_desc_avail;
529 struct vnic_wq_buf *buf;
530 unsigned int desc_count;
531 struct wq_enet_desc *descs, *desc_p, desc_tmp;
533 uint8_t vlan_tag_insert;
536 uint8_t offload_mode;
539 enic_cleanup_wq(enic, wq);
540 wq_desc_avail = vnic_wq_desc_avail(wq);
541 head_idx = wq->head_idx;
542 desc_count = wq->ring.desc_count;
543 ol_flags_mask = PKT_TX_VLAN_PKT | PKT_TX_IP_CKSUM | PKT_TX_L4_MASK;
545 nb_pkts = RTE_MIN(nb_pkts, ENIC_TX_XMIT_MAX);
547 for (index = 0; index < nb_pkts; index++) {
549 pkt_len = tx_pkt->pkt_len;
550 data_len = tx_pkt->data_len;
551 ol_flags = tx_pkt->ol_flags;
552 nb_segs = tx_pkt->nb_segs;
554 if (pkt_len > ENIC_TX_MAX_PKT_SIZE) {
555 rte_pktmbuf_free(tx_pkt);
556 rte_atomic64_inc(&enic->soft_stats.tx_oversized);
560 if (nb_segs > wq_desc_avail) {
569 bus_addr = (dma_addr_t)
570 (tx_pkt->buf_physaddr + tx_pkt->data_off);
572 descs = (struct wq_enet_desc *)wq->ring.descs;
573 desc_p = descs + head_idx;
575 eop = (data_len == pkt_len);
576 offload_mode = WQ_ENET_OFFLOAD_MODE_CSUM;
579 if (tx_pkt->tso_segsz) {
580 header_len = tso_header_len(tx_pkt);
582 offload_mode = WQ_ENET_OFFLOAD_MODE_TSO;
583 mss = tx_pkt->tso_segsz;
586 if ((ol_flags & ol_flags_mask) && (header_len == 0)) {
587 if (ol_flags & PKT_TX_IP_CKSUM)
588 mss |= ENIC_CALC_IP_CKSUM;
590 /* Nic uses just 1 bit for UDP and TCP */
591 switch (ol_flags & PKT_TX_L4_MASK) {
592 case PKT_TX_TCP_CKSUM:
593 case PKT_TX_UDP_CKSUM:
594 mss |= ENIC_CALC_TCP_UDP_CKSUM;
599 if (ol_flags & PKT_TX_VLAN_PKT) {
601 vlan_id = tx_pkt->vlan_tci;
604 wq_enet_desc_enc(&desc_tmp, bus_addr, data_len, mss, header_len,
605 offload_mode, eop, eop, 0, vlan_tag_insert,
609 buf = &wq->bufs[head_idx];
610 buf->mb = (void *)tx_pkt;
611 head_idx = enic_ring_incr(desc_count, head_idx);
615 for (tx_pkt = tx_pkt->next; tx_pkt; tx_pkt =
617 data_len = tx_pkt->data_len;
619 if (tx_pkt->next == NULL)
621 desc_p = descs + head_idx;
622 bus_addr = (dma_addr_t)(tx_pkt->buf_physaddr
624 wq_enet_desc_enc((struct wq_enet_desc *)
625 &desc_tmp, bus_addr, data_len,
626 mss, 0, offload_mode, eop, eop,
627 0, vlan_tag_insert, vlan_id,
631 buf = &wq->bufs[head_idx];
632 buf->mb = (void *)tx_pkt;
633 head_idx = enic_ring_incr(desc_count, head_idx);
640 iowrite32_relaxed(head_idx, &wq->ctrl->posted_index);
642 wq->ring.desc_avail = wq_desc_avail;
643 wq->head_idx = head_idx;