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, "+----------------|----------------+");
69 rx_desc_to_ol_flags(struct rte_mbuf *m, const union fm10k_rx_desc *d)
72 ptype_table[FM10K_RXD_PKTTYPE_MASK >> FM10K_RXD_PKTTYPE_SHIFT]
73 __rte_cache_aligned = {
74 [FM10K_PKTTYPE_OTHER] = RTE_PTYPE_L2_ETHER,
75 [FM10K_PKTTYPE_IPV4] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4,
76 [FM10K_PKTTYPE_IPV4_EX] = RTE_PTYPE_L2_ETHER |
77 RTE_PTYPE_L3_IPV4_EXT,
78 [FM10K_PKTTYPE_IPV6] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6,
79 [FM10K_PKTTYPE_IPV6_EX] = RTE_PTYPE_L2_ETHER |
80 RTE_PTYPE_L3_IPV6_EXT,
81 [FM10K_PKTTYPE_IPV4 | FM10K_PKTTYPE_TCP] = RTE_PTYPE_L2_ETHER |
82 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
83 [FM10K_PKTTYPE_IPV6 | FM10K_PKTTYPE_TCP] = RTE_PTYPE_L2_ETHER |
84 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
85 [FM10K_PKTTYPE_IPV4 | FM10K_PKTTYPE_UDP] = RTE_PTYPE_L2_ETHER |
86 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
87 [FM10K_PKTTYPE_IPV6 | FM10K_PKTTYPE_UDP] = RTE_PTYPE_L2_ETHER |
88 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
91 m->packet_type = ptype_table[(d->w.pkt_info & FM10K_RXD_PKTTYPE_MASK)
92 >> FM10K_RXD_PKTTYPE_SHIFT];
94 if (d->w.pkt_info & FM10K_RXD_RSSTYPE_MASK)
95 m->ol_flags |= PKT_RX_RSS_HASH;
97 if (unlikely((d->d.staterr &
98 (FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)) ==
99 (FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)))
100 m->ol_flags |= PKT_RX_IP_CKSUM_BAD;
102 if (unlikely((d->d.staterr &
103 (FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)) ==
104 (FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)))
105 m->ol_flags |= PKT_RX_L4_CKSUM_BAD;
107 if (d->d.staterr & FM10K_RXD_STATUS_VEXT)
108 m->ol_flags |= PKT_RX_VLAN_PKT;
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 rx_pkts[count] = mbuf;
151 if (++next_dd == q->nb_desc) {
156 /* Prefetch next mbuf while processing current one. */
157 rte_prefetch0(q->sw_ring[next_dd]);
160 * When next RX descriptor is on a cache-line boundary,
161 * prefetch the next 4 RX descriptors and the next 8 pointers
164 if ((next_dd & 0x3) == 0) {
165 rte_prefetch0(&q->hw_ring[next_dd]);
166 rte_prefetch0(&q->sw_ring[next_dd]);
170 q->next_dd = next_dd;
172 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
173 ret = rte_mempool_get_bulk(q->mp,
174 (void **)&q->sw_ring[q->next_alloc],
177 if (unlikely(ret != 0)) {
178 uint8_t port = q->port_id;
179 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
181 * Need to restore next_dd if we cannot allocate new
182 * buffers to replenish the old ones.
184 q->next_dd = (q->next_dd + q->nb_desc - count) %
186 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
190 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
191 mbuf = q->sw_ring[q->next_alloc];
193 /* setup static mbuf fields */
194 fm10k_pktmbuf_reset(mbuf, q->port_id);
196 /* write descriptor */
197 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
198 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
199 q->hw_ring[q->next_alloc] = desc;
201 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
202 q->next_trigger += q->alloc_thresh;
203 if (q->next_trigger >= q->nb_desc) {
204 q->next_trigger = q->alloc_thresh - 1;
213 fm10k_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
216 struct rte_mbuf *mbuf;
217 union fm10k_rx_desc desc;
218 struct fm10k_rx_queue *q = rx_queue;
220 uint16_t nb_rcv, nb_seg;
223 struct rte_mbuf *first_seg = q->pkt_first_seg;
224 struct rte_mbuf *last_seg = q->pkt_last_seg;
227 next_dd = q->next_dd;
230 nb_seg = RTE_MIN(nb_pkts, q->alloc_thresh);
231 for (count = 0; count < nb_seg; count++) {
232 mbuf = q->sw_ring[next_dd];
233 desc = q->hw_ring[next_dd];
234 if (!(desc.d.staterr & FM10K_RXD_STATUS_DD))
236 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
240 if (++next_dd == q->nb_desc) {
245 /* Prefetch next mbuf while processing current one. */
246 rte_prefetch0(q->sw_ring[next_dd]);
249 * When next RX descriptor is on a cache-line boundary,
250 * prefetch the next 4 RX descriptors and the next 8 pointers
253 if ((next_dd & 0x3) == 0) {
254 rte_prefetch0(&q->hw_ring[next_dd]);
255 rte_prefetch0(&q->sw_ring[next_dd]);
258 /* Fill data length */
259 rte_pktmbuf_data_len(mbuf) = desc.w.length;
262 * If this is the first buffer of the received packet,
263 * set the pointer to the first mbuf of the packet and
264 * initialize its context.
265 * Otherwise, update the total length and the number of segments
266 * of the current scattered packet, and update the pointer to
267 * the last mbuf of the current packet.
271 first_seg->pkt_len = desc.w.length;
274 (uint16_t)(first_seg->pkt_len +
275 rte_pktmbuf_data_len(mbuf));
276 first_seg->nb_segs++;
277 last_seg->next = mbuf;
281 * If this is not the last buffer of the received packet,
282 * update the pointer to the last mbuf of the current scattered
283 * packet and continue to parse the RX ring.
285 if (!(desc.d.staterr & FM10K_RXD_STATUS_EOP)) {
290 first_seg->ol_flags = 0;
291 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
292 rx_desc_to_ol_flags(first_seg, &desc);
294 first_seg->hash.rss = desc.d.rss;
296 /* Prefetch data of first segment, if configured to do so. */
297 rte_packet_prefetch((char *)first_seg->buf_addr +
298 first_seg->data_off);
301 * Store the mbuf address into the next entry of the array
302 * of returned packets.
304 rx_pkts[nb_rcv++] = first_seg;
307 * Setup receipt context for a new packet.
312 q->next_dd = next_dd;
314 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
315 ret = rte_mempool_get_bulk(q->mp,
316 (void **)&q->sw_ring[q->next_alloc],
319 if (unlikely(ret != 0)) {
320 uint8_t port = q->port_id;
321 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
323 * Need to restore next_dd if we cannot allocate new
324 * buffers to replenish the old ones.
326 q->next_dd = (q->next_dd + q->nb_desc - count) %
328 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
332 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
333 mbuf = q->sw_ring[q->next_alloc];
335 /* setup static mbuf fields */
336 fm10k_pktmbuf_reset(mbuf, q->port_id);
338 /* write descriptor */
339 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
340 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
341 q->hw_ring[q->next_alloc] = desc;
343 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
344 q->next_trigger += q->alloc_thresh;
345 if (q->next_trigger >= q->nb_desc) {
346 q->next_trigger = q->alloc_thresh - 1;
351 q->pkt_first_seg = first_seg;
352 q->pkt_last_seg = last_seg;
357 static inline void tx_free_descriptors(struct fm10k_tx_queue *q)
359 uint16_t next_rs, count = 0;
361 next_rs = fifo_peek(&q->rs_tracker);
362 if (!(q->hw_ring[next_rs].flags & FM10K_TXD_FLAG_DONE))
365 /* the DONE flag is set on this descriptor so remove the ID
366 * from the RS bit tracker and free the buffers */
367 fifo_remove(&q->rs_tracker);
369 /* wrap around? if so, free buffers from last_free up to but NOT
370 * including nb_desc */
371 if (q->last_free > next_rs) {
372 count = q->nb_desc - q->last_free;
373 while (q->last_free < q->nb_desc) {
374 rte_pktmbuf_free_seg(q->sw_ring[q->last_free]);
375 q->sw_ring[q->last_free] = NULL;
381 /* adjust free descriptor count before the next loop */
382 q->nb_free += count + (next_rs + 1 - q->last_free);
384 /* free buffers from last_free, up to and including next_rs */
385 while (q->last_free <= next_rs) {
386 rte_pktmbuf_free_seg(q->sw_ring[q->last_free]);
387 q->sw_ring[q->last_free] = NULL;
391 if (q->last_free == q->nb_desc)
395 static inline void tx_xmit_pkt(struct fm10k_tx_queue *q, struct rte_mbuf *mb)
398 uint8_t flags, hdrlen;
400 /* always set the LAST flag on the last descriptor used to
401 * transmit the packet */
402 flags = FM10K_TXD_FLAG_LAST;
403 last_id = q->next_free + mb->nb_segs - 1;
404 if (last_id >= q->nb_desc)
405 last_id = last_id - q->nb_desc;
407 /* but only set the RS flag on the last descriptor if rs_thresh
408 * descriptors will be used since the RS flag was last set */
409 if ((q->nb_used + mb->nb_segs) >= q->rs_thresh) {
410 flags |= FM10K_TXD_FLAG_RS;
411 fifo_insert(&q->rs_tracker, last_id);
414 q->nb_used = q->nb_used + mb->nb_segs;
417 q->nb_free -= mb->nb_segs;
419 q->hw_ring[q->next_free].flags = 0;
420 /* set checksum flags on first descriptor of packet. SCTP checksum
421 * offload is not supported, but we do not explicitly check for this
422 * case in favor of greatly simplified processing. */
423 if (mb->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
424 q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_CSUM;
426 /* set vlan if requested */
427 if (mb->ol_flags & PKT_TX_VLAN_PKT)
428 q->hw_ring[q->next_free].vlan = mb->vlan_tci;
430 q->sw_ring[q->next_free] = mb;
431 q->hw_ring[q->next_free].buffer_addr =
432 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
433 q->hw_ring[q->next_free].buflen =
434 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
436 if (mb->ol_flags & PKT_TX_TCP_SEG) {
437 hdrlen = mb->outer_l2_len + mb->outer_l3_len + mb->l2_len +
438 mb->l3_len + mb->l4_len;
439 if (q->hw_ring[q->next_free].flags & FM10K_TXD_FLAG_FTAG)
440 hdrlen += sizeof(struct fm10k_ftag);
442 if (likely((hdrlen >= FM10K_TSO_MIN_HEADERLEN) &&
443 (hdrlen <= FM10K_TSO_MAX_HEADERLEN) &&
444 (mb->tso_segsz >= FM10K_TSO_MINMSS))) {
445 q->hw_ring[q->next_free].mss = mb->tso_segsz;
446 q->hw_ring[q->next_free].hdrlen = hdrlen;
450 if (++q->next_free == q->nb_desc)
453 /* fill up the rings */
454 for (mb = mb->next; mb != NULL; mb = mb->next) {
455 q->sw_ring[q->next_free] = mb;
456 q->hw_ring[q->next_free].buffer_addr =
457 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
458 q->hw_ring[q->next_free].buflen =
459 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
460 q->hw_ring[q->next_free].flags = 0;
461 if (++q->next_free == q->nb_desc)
465 q->hw_ring[last_id].flags |= flags;
469 fm10k_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
472 struct fm10k_tx_queue *q = tx_queue;
476 for (count = 0; count < nb_pkts; ++count) {
479 /* running low on descriptors? try to free some... */
480 if (q->nb_free < q->free_thresh)
481 tx_free_descriptors(q);
483 /* make sure there are enough free descriptors to transmit the
484 * entire packet before doing anything */
485 if (q->nb_free < mb->nb_segs)
488 /* sanity check to make sure the mbuf is valid */
489 if ((mb->nb_segs == 0) ||
490 ((mb->nb_segs > 1) && (mb->next == NULL)))
493 /* process the packet */
497 /* update the tail pointer if any packets were processed */
498 if (likely(count > 0))
499 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_free);