4 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
17 * * Neither the name of Intel Corporation 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.
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.
36 #include <rte_ethdev.h>
37 #include <rte_common.h>
39 #include "base/fm10k_type.h"
41 #ifdef RTE_PMD_PACKET_PREFETCH
42 #define rte_packet_prefetch(p) rte_prefetch1(p)
44 #define rte_packet_prefetch(p) do {} while (0)
47 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
48 static inline void dump_rxd(union fm10k_rx_desc *rxd)
50 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
51 PMD_RX_LOG(DEBUG, "| GLORT | PKT HDR & TYPE |");
52 PMD_RX_LOG(DEBUG, "| 0x%08x | 0x%08x |", rxd->d.glort,
54 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
55 PMD_RX_LOG(DEBUG, "| VLAN & LEN | STATUS |");
56 PMD_RX_LOG(DEBUG, "| 0x%08x | 0x%08x |", rxd->d.vlan_len,
58 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
59 PMD_RX_LOG(DEBUG, "| RESERVED | RSS_HASH |");
60 PMD_RX_LOG(DEBUG, "| 0x%08x | 0x%08x |", 0, rxd->d.rss);
61 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
62 PMD_RX_LOG(DEBUG, "| TIME TAG |");
63 PMD_RX_LOG(DEBUG, "| 0x%016"PRIx64" |", rxd->q.timestamp);
64 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
68 /* @note: When this function is changed, make corresponding change to
69 * fm10k_dev_supported_ptypes_get()
72 rx_desc_to_ol_flags(struct rte_mbuf *m, const union fm10k_rx_desc *d)
75 ptype_table[FM10K_RXD_PKTTYPE_MASK >> FM10K_RXD_PKTTYPE_SHIFT]
76 __rte_cache_aligned = {
77 [FM10K_PKTTYPE_OTHER] = RTE_PTYPE_L2_ETHER,
78 [FM10K_PKTTYPE_IPV4] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4,
79 [FM10K_PKTTYPE_IPV4_EX] = RTE_PTYPE_L2_ETHER |
80 RTE_PTYPE_L3_IPV4_EXT,
81 [FM10K_PKTTYPE_IPV6] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6,
82 [FM10K_PKTTYPE_IPV6_EX] = RTE_PTYPE_L2_ETHER |
83 RTE_PTYPE_L3_IPV6_EXT,
84 [FM10K_PKTTYPE_IPV4 | FM10K_PKTTYPE_TCP] = RTE_PTYPE_L2_ETHER |
85 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
86 [FM10K_PKTTYPE_IPV6 | FM10K_PKTTYPE_TCP] = RTE_PTYPE_L2_ETHER |
87 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
88 [FM10K_PKTTYPE_IPV4 | FM10K_PKTTYPE_UDP] = RTE_PTYPE_L2_ETHER |
89 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
90 [FM10K_PKTTYPE_IPV6 | FM10K_PKTTYPE_UDP] = RTE_PTYPE_L2_ETHER |
91 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
94 m->packet_type = ptype_table[(d->w.pkt_info & FM10K_RXD_PKTTYPE_MASK)
95 >> FM10K_RXD_PKTTYPE_SHIFT];
97 if (d->w.pkt_info & FM10K_RXD_RSSTYPE_MASK)
98 m->ol_flags |= PKT_RX_RSS_HASH;
100 if (unlikely((d->d.staterr &
101 (FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)) ==
102 (FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)))
103 m->ol_flags |= PKT_RX_IP_CKSUM_BAD;
105 if (unlikely((d->d.staterr &
106 (FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)) ==
107 (FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)))
108 m->ol_flags |= PKT_RX_L4_CKSUM_BAD;
110 if (unlikely(d->d.staterr & FM10K_RXD_STATUS_HBO))
111 m->ol_flags |= PKT_RX_HBUF_OVERFLOW;
113 if (unlikely(d->d.staterr & FM10K_RXD_STATUS_RXE))
114 m->ol_flags |= PKT_RX_RECIP_ERR;
118 fm10k_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
121 struct rte_mbuf *mbuf;
122 union fm10k_rx_desc desc;
123 struct fm10k_rx_queue *q = rx_queue;
129 next_dd = q->next_dd;
131 nb_pkts = RTE_MIN(nb_pkts, q->alloc_thresh);
132 for (count = 0; count < nb_pkts; ++count) {
133 mbuf = q->sw_ring[next_dd];
134 desc = q->hw_ring[next_dd];
135 if (!(desc.d.staterr & FM10K_RXD_STATUS_DD))
137 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
140 rte_pktmbuf_pkt_len(mbuf) = desc.w.length;
141 rte_pktmbuf_data_len(mbuf) = desc.w.length;
144 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
145 rx_desc_to_ol_flags(mbuf, &desc);
148 mbuf->hash.rss = desc.d.rss;
150 * Packets in fm10k device always carry at least one VLAN tag.
151 * For those packets coming in without VLAN tag,
152 * the port default VLAN tag will be used.
153 * So, always PKT_RX_VLAN_PKT flag is set and vlan_tci
154 * is valid for each RX packet's mbuf.
156 mbuf->ol_flags |= PKT_RX_VLAN_PKT;
157 mbuf->vlan_tci = desc.w.vlan;
159 * mbuf->vlan_tci_outer is an idle field in fm10k driver,
160 * so it can be selected to store sglort value.
163 mbuf->vlan_tci_outer = rte_le_to_cpu_16(desc.w.sglort);
165 rx_pkts[count] = mbuf;
166 if (++next_dd == q->nb_desc) {
171 /* Prefetch next mbuf while processing current one. */
172 rte_prefetch0(q->sw_ring[next_dd]);
175 * When next RX descriptor is on a cache-line boundary,
176 * prefetch the next 4 RX descriptors and the next 8 pointers
179 if ((next_dd & 0x3) == 0) {
180 rte_prefetch0(&q->hw_ring[next_dd]);
181 rte_prefetch0(&q->sw_ring[next_dd]);
185 q->next_dd = next_dd;
187 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
188 ret = rte_mempool_get_bulk(q->mp,
189 (void **)&q->sw_ring[q->next_alloc],
192 if (unlikely(ret != 0)) {
193 uint8_t port = q->port_id;
194 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
196 * Need to restore next_dd if we cannot allocate new
197 * buffers to replenish the old ones.
199 q->next_dd = (q->next_dd + q->nb_desc - count) %
201 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
205 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
206 mbuf = q->sw_ring[q->next_alloc];
208 /* setup static mbuf fields */
209 fm10k_pktmbuf_reset(mbuf, q->port_id);
211 /* write descriptor */
212 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
213 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
214 q->hw_ring[q->next_alloc] = desc;
216 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
217 q->next_trigger += q->alloc_thresh;
218 if (q->next_trigger >= q->nb_desc) {
219 q->next_trigger = q->alloc_thresh - 1;
228 fm10k_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
231 struct rte_mbuf *mbuf;
232 union fm10k_rx_desc desc;
233 struct fm10k_rx_queue *q = rx_queue;
235 uint16_t nb_rcv, nb_seg;
238 struct rte_mbuf *first_seg = q->pkt_first_seg;
239 struct rte_mbuf *last_seg = q->pkt_last_seg;
242 next_dd = q->next_dd;
245 nb_seg = RTE_MIN(nb_pkts, q->alloc_thresh);
246 for (count = 0; count < nb_seg; count++) {
247 mbuf = q->sw_ring[next_dd];
248 desc = q->hw_ring[next_dd];
249 if (!(desc.d.staterr & FM10K_RXD_STATUS_DD))
251 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
255 if (++next_dd == q->nb_desc) {
260 /* Prefetch next mbuf while processing current one. */
261 rte_prefetch0(q->sw_ring[next_dd]);
264 * When next RX descriptor is on a cache-line boundary,
265 * prefetch the next 4 RX descriptors and the next 8 pointers
268 if ((next_dd & 0x3) == 0) {
269 rte_prefetch0(&q->hw_ring[next_dd]);
270 rte_prefetch0(&q->sw_ring[next_dd]);
273 /* Fill data length */
274 rte_pktmbuf_data_len(mbuf) = desc.w.length;
277 * If this is the first buffer of the received packet,
278 * set the pointer to the first mbuf of the packet and
279 * initialize its context.
280 * Otherwise, update the total length and the number of segments
281 * of the current scattered packet, and update the pointer to
282 * the last mbuf of the current packet.
286 first_seg->pkt_len = desc.w.length;
289 (uint16_t)(first_seg->pkt_len +
290 rte_pktmbuf_data_len(mbuf));
291 first_seg->nb_segs++;
292 last_seg->next = mbuf;
296 * If this is not the last buffer of the received packet,
297 * update the pointer to the last mbuf of the current scattered
298 * packet and continue to parse the RX ring.
300 if (!(desc.d.staterr & FM10K_RXD_STATUS_EOP)) {
305 first_seg->ol_flags = 0;
306 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
307 rx_desc_to_ol_flags(first_seg, &desc);
309 first_seg->hash.rss = desc.d.rss;
311 * Packets in fm10k device always carry at least one VLAN tag.
312 * For those packets coming in without VLAN tag,
313 * the port default VLAN tag will be used.
314 * So, always PKT_RX_VLAN_PKT flag is set and vlan_tci
315 * is valid for each RX packet's mbuf.
317 first_seg->ol_flags |= PKT_RX_VLAN_PKT;
318 first_seg->vlan_tci = desc.w.vlan;
320 * mbuf->vlan_tci_outer is an idle field in fm10k driver,
321 * so it can be selected to store sglort value.
324 first_seg->vlan_tci_outer =
325 rte_le_to_cpu_16(desc.w.sglort);
327 /* Prefetch data of first segment, if configured to do so. */
328 rte_packet_prefetch((char *)first_seg->buf_addr +
329 first_seg->data_off);
332 * Store the mbuf address into the next entry of the array
333 * of returned packets.
335 rx_pkts[nb_rcv++] = first_seg;
338 * Setup receipt context for a new packet.
343 q->next_dd = next_dd;
345 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
346 ret = rte_mempool_get_bulk(q->mp,
347 (void **)&q->sw_ring[q->next_alloc],
350 if (unlikely(ret != 0)) {
351 uint8_t port = q->port_id;
352 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
354 * Need to restore next_dd if we cannot allocate new
355 * buffers to replenish the old ones.
357 q->next_dd = (q->next_dd + q->nb_desc - count) %
359 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
363 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
364 mbuf = q->sw_ring[q->next_alloc];
366 /* setup static mbuf fields */
367 fm10k_pktmbuf_reset(mbuf, q->port_id);
369 /* write descriptor */
370 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
371 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
372 q->hw_ring[q->next_alloc] = desc;
374 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
375 q->next_trigger += q->alloc_thresh;
376 if (q->next_trigger >= q->nb_desc) {
377 q->next_trigger = q->alloc_thresh - 1;
382 q->pkt_first_seg = first_seg;
383 q->pkt_last_seg = last_seg;
389 fm10k_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
391 volatile union fm10k_rx_desc *rxdp;
392 struct fm10k_rx_queue *rxq = rx_queue;
396 if (unlikely(offset >= rxq->nb_desc)) {
397 PMD_DRV_LOG(ERR, "Invalid RX descriptor offset %u", offset);
401 desc = rxq->next_dd + offset;
402 if (desc >= rxq->nb_desc)
403 desc -= rxq->nb_desc;
405 rxdp = &rxq->hw_ring[desc];
407 ret = !!(rxdp->w.status &
408 rte_cpu_to_le_16(FM10K_RXD_STATUS_DD));
414 * Free multiple TX mbuf at a time if they are in the same pool
416 * @txep: software desc ring index that starts to free
417 * @num: number of descs to free
420 static inline void tx_free_bulk_mbuf(struct rte_mbuf **txep, int num)
422 struct rte_mbuf *m, *free[RTE_FM10K_TX_MAX_FREE_BUF_SZ];
426 if (unlikely(num == 0))
429 m = __rte_pktmbuf_prefree_seg(txep[0]);
430 if (likely(m != NULL)) {
433 for (i = 1; i < num; i++) {
434 m = __rte_pktmbuf_prefree_seg(txep[i]);
435 if (likely(m != NULL)) {
436 if (likely(m->pool == free[0]->pool))
439 rte_mempool_put_bulk(free[0]->pool,
440 (void *)free, nb_free);
447 rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
449 for (i = 1; i < num; i++) {
450 m = __rte_pktmbuf_prefree_seg(txep[i]);
452 rte_mempool_put(m->pool, m);
458 static inline void tx_free_descriptors(struct fm10k_tx_queue *q)
460 uint16_t next_rs, count = 0;
462 next_rs = fifo_peek(&q->rs_tracker);
463 if (!(q->hw_ring[next_rs].flags & FM10K_TXD_FLAG_DONE))
466 /* the DONE flag is set on this descriptor so remove the ID
467 * from the RS bit tracker and free the buffers */
468 fifo_remove(&q->rs_tracker);
470 /* wrap around? if so, free buffers from last_free up to but NOT
471 * including nb_desc */
472 if (q->last_free > next_rs) {
473 count = q->nb_desc - q->last_free;
474 tx_free_bulk_mbuf(&q->sw_ring[q->last_free], count);
478 /* adjust free descriptor count before the next loop */
479 q->nb_free += count + (next_rs + 1 - q->last_free);
481 /* free buffers from last_free, up to and including next_rs */
482 if (q->last_free <= next_rs) {
483 count = next_rs - q->last_free + 1;
484 tx_free_bulk_mbuf(&q->sw_ring[q->last_free], count);
485 q->last_free += count;
488 if (q->last_free == q->nb_desc)
492 static inline void tx_xmit_pkt(struct fm10k_tx_queue *q, struct rte_mbuf *mb)
495 uint8_t flags, hdrlen;
497 /* always set the LAST flag on the last descriptor used to
498 * transmit the packet */
499 flags = FM10K_TXD_FLAG_LAST;
500 last_id = q->next_free + mb->nb_segs - 1;
501 if (last_id >= q->nb_desc)
502 last_id = last_id - q->nb_desc;
504 /* but only set the RS flag on the last descriptor if rs_thresh
505 * descriptors will be used since the RS flag was last set */
506 if ((q->nb_used + mb->nb_segs) >= q->rs_thresh) {
507 flags |= FM10K_TXD_FLAG_RS;
508 fifo_insert(&q->rs_tracker, last_id);
511 q->nb_used = q->nb_used + mb->nb_segs;
514 q->nb_free -= mb->nb_segs;
516 q->hw_ring[q->next_free].flags = 0;
518 q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_FTAG;
519 /* set checksum flags on first descriptor of packet. SCTP checksum
520 * offload is not supported, but we do not explicitly check for this
521 * case in favor of greatly simplified processing. */
522 if (mb->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
523 q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_CSUM;
525 /* set vlan if requested */
526 if (mb->ol_flags & PKT_TX_VLAN_PKT)
527 q->hw_ring[q->next_free].vlan = mb->vlan_tci;
529 q->sw_ring[q->next_free] = mb;
530 q->hw_ring[q->next_free].buffer_addr =
531 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
532 q->hw_ring[q->next_free].buflen =
533 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
535 if (mb->ol_flags & PKT_TX_TCP_SEG) {
536 hdrlen = mb->outer_l2_len + mb->outer_l3_len + mb->l2_len +
537 mb->l3_len + mb->l4_len;
538 if (q->hw_ring[q->next_free].flags & FM10K_TXD_FLAG_FTAG)
539 hdrlen += sizeof(struct fm10k_ftag);
541 if (likely((hdrlen >= FM10K_TSO_MIN_HEADERLEN) &&
542 (hdrlen <= FM10K_TSO_MAX_HEADERLEN) &&
543 (mb->tso_segsz >= FM10K_TSO_MINMSS))) {
544 q->hw_ring[q->next_free].mss = mb->tso_segsz;
545 q->hw_ring[q->next_free].hdrlen = hdrlen;
549 if (++q->next_free == q->nb_desc)
552 /* fill up the rings */
553 for (mb = mb->next; mb != NULL; mb = mb->next) {
554 q->sw_ring[q->next_free] = mb;
555 q->hw_ring[q->next_free].buffer_addr =
556 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
557 q->hw_ring[q->next_free].buflen =
558 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
559 q->hw_ring[q->next_free].flags = 0;
560 if (++q->next_free == q->nb_desc)
564 q->hw_ring[last_id].flags |= flags;
568 fm10k_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
571 struct fm10k_tx_queue *q = tx_queue;
575 for (count = 0; count < nb_pkts; ++count) {
578 /* running low on descriptors? try to free some... */
579 if (q->nb_free < q->free_thresh)
580 tx_free_descriptors(q);
582 /* make sure there are enough free descriptors to transmit the
583 * entire packet before doing anything */
584 if (q->nb_free < mb->nb_segs)
587 /* sanity check to make sure the mbuf is valid */
588 if ((mb->nb_segs == 0) ||
589 ((mb->nb_segs > 1) && (mb->next == NULL)))
592 /* process the packet */
596 /* update the tail pointer if any packets were processed */
597 if (likely(count > 0))
598 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_free);