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;
102 fm10k_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
105 struct rte_mbuf *mbuf;
106 union fm10k_rx_desc desc;
107 struct fm10k_rx_queue *q = rx_queue;
113 next_dd = q->next_dd;
115 nb_pkts = RTE_MIN(nb_pkts, q->alloc_thresh);
116 for (count = 0; count < nb_pkts; ++count) {
117 if (!(q->hw_ring[next_dd].d.staterr & FM10K_RXD_STATUS_DD))
119 mbuf = q->sw_ring[next_dd];
120 desc = q->hw_ring[next_dd];
121 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
124 rte_pktmbuf_pkt_len(mbuf) = desc.w.length;
125 rte_pktmbuf_data_len(mbuf) = desc.w.length;
128 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
129 rx_desc_to_ol_flags(mbuf, &desc);
132 mbuf->hash.rss = desc.d.rss;
134 * Packets in fm10k device always carry at least one VLAN tag.
135 * For those packets coming in without VLAN tag,
136 * the port default VLAN tag will be used.
137 * So, always PKT_RX_VLAN_PKT flag is set and vlan_tci
138 * is valid for each RX packet's mbuf.
140 mbuf->ol_flags |= PKT_RX_VLAN_PKT;
141 mbuf->vlan_tci = desc.w.vlan;
143 * mbuf->vlan_tci_outer is an idle field in fm10k driver,
144 * so it can be selected to store sglort value.
147 mbuf->vlan_tci_outer = rte_le_to_cpu_16(desc.w.sglort);
149 rx_pkts[count] = mbuf;
150 if (++next_dd == q->nb_desc) {
155 /* Prefetch next mbuf while processing current one. */
156 rte_prefetch0(q->sw_ring[next_dd]);
159 * When next RX descriptor is on a cache-line boundary,
160 * prefetch the next 4 RX descriptors and the next 8 pointers
163 if ((next_dd & 0x3) == 0) {
164 rte_prefetch0(&q->hw_ring[next_dd]);
165 rte_prefetch0(&q->sw_ring[next_dd]);
169 q->next_dd = next_dd;
171 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
172 ret = rte_mempool_get_bulk(q->mp,
173 (void **)&q->sw_ring[q->next_alloc],
176 if (unlikely(ret != 0)) {
177 uint8_t port = q->port_id;
178 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
180 * Need to restore next_dd if we cannot allocate new
181 * buffers to replenish the old ones.
183 q->next_dd = (q->next_dd + q->nb_desc - count) %
185 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
189 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
190 mbuf = q->sw_ring[q->next_alloc];
192 /* setup static mbuf fields */
193 fm10k_pktmbuf_reset(mbuf, q->port_id);
195 /* write descriptor */
196 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
197 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
198 q->hw_ring[q->next_alloc] = desc;
200 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
201 q->next_trigger += q->alloc_thresh;
202 if (q->next_trigger >= q->nb_desc) {
203 q->next_trigger = q->alloc_thresh - 1;
212 fm10k_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
215 struct rte_mbuf *mbuf;
216 union fm10k_rx_desc desc;
217 struct fm10k_rx_queue *q = rx_queue;
219 uint16_t nb_rcv, nb_seg;
222 struct rte_mbuf *first_seg = q->pkt_first_seg;
223 struct rte_mbuf *last_seg = q->pkt_last_seg;
226 next_dd = q->next_dd;
229 nb_seg = RTE_MIN(nb_pkts, q->alloc_thresh);
230 for (count = 0; count < nb_seg; count++) {
231 if (!(q->hw_ring[next_dd].d.staterr & FM10K_RXD_STATUS_DD))
233 mbuf = q->sw_ring[next_dd];
234 desc = q->hw_ring[next_dd];
235 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
239 if (++next_dd == q->nb_desc) {
244 /* Prefetch next mbuf while processing current one. */
245 rte_prefetch0(q->sw_ring[next_dd]);
248 * When next RX descriptor is on a cache-line boundary,
249 * prefetch the next 4 RX descriptors and the next 8 pointers
252 if ((next_dd & 0x3) == 0) {
253 rte_prefetch0(&q->hw_ring[next_dd]);
254 rte_prefetch0(&q->sw_ring[next_dd]);
257 /* Fill data length */
258 rte_pktmbuf_data_len(mbuf) = desc.w.length;
261 * If this is the first buffer of the received packet,
262 * set the pointer to the first mbuf of the packet and
263 * initialize its context.
264 * Otherwise, update the total length and the number of segments
265 * of the current scattered packet, and update the pointer to
266 * the last mbuf of the current packet.
270 first_seg->pkt_len = desc.w.length;
273 (uint16_t)(first_seg->pkt_len +
274 rte_pktmbuf_data_len(mbuf));
275 first_seg->nb_segs++;
276 last_seg->next = mbuf;
280 * If this is not the last buffer of the received packet,
281 * update the pointer to the last mbuf of the current scattered
282 * packet and continue to parse the RX ring.
284 if (!(desc.d.staterr & FM10K_RXD_STATUS_EOP)) {
289 first_seg->ol_flags = 0;
290 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
291 rx_desc_to_ol_flags(first_seg, &desc);
293 first_seg->hash.rss = desc.d.rss;
295 * Packets in fm10k device always carry at least one VLAN tag.
296 * For those packets coming in without VLAN tag,
297 * the port default VLAN tag will be used.
298 * So, always PKT_RX_VLAN_PKT flag is set and vlan_tci
299 * is valid for each RX packet's mbuf.
301 first_seg->ol_flags |= PKT_RX_VLAN_PKT;
302 first_seg->vlan_tci = desc.w.vlan;
304 * mbuf->vlan_tci_outer is an idle field in fm10k driver,
305 * so it can be selected to store sglort value.
308 first_seg->vlan_tci_outer =
309 rte_le_to_cpu_16(desc.w.sglort);
311 /* Prefetch data of first segment, if configured to do so. */
312 rte_packet_prefetch((char *)first_seg->buf_addr +
313 first_seg->data_off);
316 * Store the mbuf address into the next entry of the array
317 * of returned packets.
319 rx_pkts[nb_rcv++] = first_seg;
322 * Setup receipt context for a new packet.
327 q->next_dd = next_dd;
329 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
330 ret = rte_mempool_get_bulk(q->mp,
331 (void **)&q->sw_ring[q->next_alloc],
334 if (unlikely(ret != 0)) {
335 uint8_t port = q->port_id;
336 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
338 * Need to restore next_dd if we cannot allocate new
339 * buffers to replenish the old ones.
341 q->next_dd = (q->next_dd + q->nb_desc - count) %
343 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
347 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
348 mbuf = q->sw_ring[q->next_alloc];
350 /* setup static mbuf fields */
351 fm10k_pktmbuf_reset(mbuf, q->port_id);
353 /* write descriptor */
354 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
355 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
356 q->hw_ring[q->next_alloc] = desc;
358 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
359 q->next_trigger += q->alloc_thresh;
360 if (q->next_trigger >= q->nb_desc) {
361 q->next_trigger = q->alloc_thresh - 1;
366 q->pkt_first_seg = first_seg;
367 q->pkt_last_seg = last_seg;
373 fm10k_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
375 volatile union fm10k_rx_desc *rxdp;
376 struct fm10k_rx_queue *rxq = rx_queue;
380 if (unlikely(offset >= rxq->nb_desc)) {
381 PMD_DRV_LOG(ERR, "Invalid RX descriptor offset %u", offset);
385 desc = rxq->next_dd + offset;
386 if (desc >= rxq->nb_desc)
387 desc -= rxq->nb_desc;
389 rxdp = &rxq->hw_ring[desc];
391 ret = !!(rxdp->w.status &
392 rte_cpu_to_le_16(FM10K_RXD_STATUS_DD));
398 * Free multiple TX mbuf at a time if they are in the same pool
400 * @txep: software desc ring index that starts to free
401 * @num: number of descs to free
404 static inline void tx_free_bulk_mbuf(struct rte_mbuf **txep, int num)
406 struct rte_mbuf *m, *free[RTE_FM10K_TX_MAX_FREE_BUF_SZ];
410 if (unlikely(num == 0))
413 m = __rte_pktmbuf_prefree_seg(txep[0]);
414 if (likely(m != NULL)) {
417 for (i = 1; i < num; i++) {
418 m = __rte_pktmbuf_prefree_seg(txep[i]);
419 if (likely(m != NULL)) {
420 if (likely(m->pool == free[0]->pool))
423 rte_mempool_put_bulk(free[0]->pool,
424 (void *)free, nb_free);
431 rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
433 for (i = 1; i < num; i++) {
434 m = __rte_pktmbuf_prefree_seg(txep[i]);
436 rte_mempool_put(m->pool, m);
442 static inline void tx_free_descriptors(struct fm10k_tx_queue *q)
444 uint16_t next_rs, count = 0;
446 next_rs = fifo_peek(&q->rs_tracker);
447 if (!(q->hw_ring[next_rs].flags & FM10K_TXD_FLAG_DONE))
450 /* the DONE flag is set on this descriptor so remove the ID
451 * from the RS bit tracker and free the buffers */
452 fifo_remove(&q->rs_tracker);
454 /* wrap around? if so, free buffers from last_free up to but NOT
455 * including nb_desc */
456 if (q->last_free > next_rs) {
457 count = q->nb_desc - q->last_free;
458 tx_free_bulk_mbuf(&q->sw_ring[q->last_free], count);
462 /* adjust free descriptor count before the next loop */
463 q->nb_free += count + (next_rs + 1 - q->last_free);
465 /* free buffers from last_free, up to and including next_rs */
466 if (q->last_free <= next_rs) {
467 count = next_rs - q->last_free + 1;
468 tx_free_bulk_mbuf(&q->sw_ring[q->last_free], count);
469 q->last_free += count;
472 if (q->last_free == q->nb_desc)
476 static inline void tx_xmit_pkt(struct fm10k_tx_queue *q, struct rte_mbuf *mb)
479 uint8_t flags, hdrlen;
481 /* always set the LAST flag on the last descriptor used to
482 * transmit the packet */
483 flags = FM10K_TXD_FLAG_LAST;
484 last_id = q->next_free + mb->nb_segs - 1;
485 if (last_id >= q->nb_desc)
486 last_id = last_id - q->nb_desc;
488 /* but only set the RS flag on the last descriptor if rs_thresh
489 * descriptors will be used since the RS flag was last set */
490 if ((q->nb_used + mb->nb_segs) >= q->rs_thresh) {
491 flags |= FM10K_TXD_FLAG_RS;
492 fifo_insert(&q->rs_tracker, last_id);
495 q->nb_used = q->nb_used + mb->nb_segs;
498 q->nb_free -= mb->nb_segs;
500 q->hw_ring[q->next_free].flags = 0;
502 q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_FTAG;
503 /* set checksum flags on first descriptor of packet. SCTP checksum
504 * offload is not supported, but we do not explicitly check for this
505 * case in favor of greatly simplified processing. */
506 if (mb->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
507 q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_CSUM;
509 /* set vlan if requested */
510 if (mb->ol_flags & PKT_TX_VLAN_PKT)
511 q->hw_ring[q->next_free].vlan = mb->vlan_tci;
513 q->sw_ring[q->next_free] = mb;
514 q->hw_ring[q->next_free].buffer_addr =
515 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
516 q->hw_ring[q->next_free].buflen =
517 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
519 if (mb->ol_flags & PKT_TX_TCP_SEG) {
520 hdrlen = mb->outer_l2_len + mb->outer_l3_len + mb->l2_len +
521 mb->l3_len + mb->l4_len;
522 if (q->hw_ring[q->next_free].flags & FM10K_TXD_FLAG_FTAG)
523 hdrlen += sizeof(struct fm10k_ftag);
525 if (likely((hdrlen >= FM10K_TSO_MIN_HEADERLEN) &&
526 (hdrlen <= FM10K_TSO_MAX_HEADERLEN) &&
527 (mb->tso_segsz >= FM10K_TSO_MINMSS))) {
528 q->hw_ring[q->next_free].mss = mb->tso_segsz;
529 q->hw_ring[q->next_free].hdrlen = hdrlen;
533 if (++q->next_free == q->nb_desc)
536 /* fill up the rings */
537 for (mb = mb->next; mb != NULL; mb = mb->next) {
538 q->sw_ring[q->next_free] = mb;
539 q->hw_ring[q->next_free].buffer_addr =
540 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
541 q->hw_ring[q->next_free].buflen =
542 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
543 q->hw_ring[q->next_free].flags = 0;
544 if (++q->next_free == q->nb_desc)
548 q->hw_ring[last_id].flags |= flags;
552 fm10k_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
555 struct fm10k_tx_queue *q = tx_queue;
559 for (count = 0; count < nb_pkts; ++count) {
562 /* running low on descriptors? try to free some... */
563 if (q->nb_free < q->free_thresh)
564 tx_free_descriptors(q);
566 /* make sure there are enough free descriptors to transmit the
567 * entire packet before doing anything */
568 if (q->nb_free < mb->nb_segs)
571 /* sanity check to make sure the mbuf is valid */
572 if ((mb->nb_segs == 0) ||
573 ((mb->nb_segs > 1) && (mb->next == NULL)))
576 /* process the packet */
580 /* update the tail pointer if any packets were processed */
581 if (likely(count > 0))
582 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_free);