4 * Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
5 * Copyright(c) 2016, Linaro Limited
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * 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
18 * * Neither the name of Intel Corporation nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 #include <rte_ethdev.h>
37 #include <rte_malloc.h>
39 #include "base/i40e_prototype.h"
40 #include "base/i40e_type.h"
41 #include "i40e_ethdev.h"
42 #include "i40e_rxtx.h"
43 #include "i40e_rxtx_vec_common.h"
47 #pragma GCC diagnostic ignored "-Wcast-qual"
50 i40e_rxq_rearm(struct i40e_rx_queue *rxq)
54 volatile union i40e_rx_desc *rxdp;
55 struct i40e_rx_entry *rxep = &rxq->sw_ring[rxq->rxrearm_start];
56 struct rte_mbuf *mb0, *mb1;
57 uint64x2_t dma_addr0, dma_addr1;
58 uint64x2_t zero = vdupq_n_u64(0);
62 rxdp = rxq->rx_ring + rxq->rxrearm_start;
64 /* Pull 'n' more MBUFs into the software ring */
65 if (unlikely(rte_mempool_get_bulk(rxq->mp,
67 RTE_I40E_RXQ_REARM_THRESH) < 0)) {
68 if (rxq->rxrearm_nb + RTE_I40E_RXQ_REARM_THRESH >=
70 for (i = 0; i < RTE_I40E_DESCS_PER_LOOP; i++) {
71 rxep[i].mbuf = &rxq->fake_mbuf;
72 vst1q_u64((uint64_t *)&rxdp[i].read, zero);
75 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
76 RTE_I40E_RXQ_REARM_THRESH;
80 p = vld1_u8((uint8_t *)&rxq->mbuf_initializer);
82 /* Initialize the mbufs in vector, process 2 mbufs in one loop */
83 for (i = 0; i < RTE_I40E_RXQ_REARM_THRESH; i += 2, rxep += 2) {
87 /* Flush mbuf with pkt template.
88 * Data to be rearmed is 6 bytes long.
89 * Though, RX will overwrite ol_flags that are coming next
90 * anyway. So overwrite whole 8 bytes with one load:
91 * 6 bytes of rearm_data plus first 2 bytes of ol_flags.
93 vst1_u8((uint8_t *)&mb0->rearm_data, p);
94 paddr = mb0->buf_physaddr + RTE_PKTMBUF_HEADROOM;
95 dma_addr0 = vdupq_n_u64(paddr);
97 /* flush desc with pa dma_addr */
98 vst1q_u64((uint64_t *)&rxdp++->read, dma_addr0);
100 vst1_u8((uint8_t *)&mb1->rearm_data, p);
101 paddr = mb1->buf_physaddr + RTE_PKTMBUF_HEADROOM;
102 dma_addr1 = vdupq_n_u64(paddr);
103 vst1q_u64((uint64_t *)&rxdp++->read, dma_addr1);
106 rxq->rxrearm_start += RTE_I40E_RXQ_REARM_THRESH;
107 if (rxq->rxrearm_start >= rxq->nb_rx_desc)
108 rxq->rxrearm_start = 0;
110 rxq->rxrearm_nb -= RTE_I40E_RXQ_REARM_THRESH;
112 rx_id = (uint16_t)((rxq->rxrearm_start == 0) ?
113 (rxq->nb_rx_desc - 1) : (rxq->rxrearm_start - 1));
115 /* Update the tail pointer on the NIC */
116 I40E_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
119 /* Handling the offload flags (olflags) field takes computation
120 * time when receiving packets. Therefore we provide a flag to disable
121 * the processing of the olflags field when they are not needed. This
122 * gives improved performance, at the cost of losing the offload info
123 * in the received packet
125 #ifdef RTE_LIBRTE_I40E_RX_OLFLAGS_ENABLE
128 desc_to_olflags_v(uint64x2_t descs[4], struct rte_mbuf **rx_pkts)
130 uint32x4_t vlan0, vlan1, rss, l3_l4e;
132 /* mask everything except RSS, flow director and VLAN flags
133 * bit2 is for VLAN tag, bit11 for flow director indication
134 * bit13:12 for RSS indication.
136 const uint32x4_t rss_vlan_msk = {
137 0x1c03804, 0x1c03804, 0x1c03804, 0x1c03804};
139 /* map rss and vlan type to rss hash and vlan flag */
140 const uint8x16_t vlan_flags = {
142 PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED, 0, 0, 0,
146 const uint8x16_t rss_flags = {
147 0, PKT_RX_FDIR, 0, 0,
148 0, 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH | PKT_RX_FDIR,
152 const uint8x16_t l3_l4e_flags = {
156 PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD,
157 PKT_RX_EIP_CKSUM_BAD,
158 PKT_RX_EIP_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD,
159 PKT_RX_EIP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD,
160 PKT_RX_EIP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD,
161 0, 0, 0, 0, 0, 0, 0, 0};
163 vlan0 = vzipq_u32(vreinterpretq_u32_u64(descs[0]),
164 vreinterpretq_u32_u64(descs[2])).val[1];
165 vlan1 = vzipq_u32(vreinterpretq_u32_u64(descs[1]),
166 vreinterpretq_u32_u64(descs[3])).val[1];
167 vlan0 = vzipq_u32(vlan0, vlan1).val[0];
169 vlan1 = vandq_u32(vlan0, rss_vlan_msk);
170 vlan0 = vreinterpretq_u32_u8(vqtbl1q_u8(vlan_flags,
171 vreinterpretq_u8_u32(vlan1)));
173 rss = vshrq_n_u32(vlan1, 11);
174 rss = vreinterpretq_u32_u8(vqtbl1q_u8(rss_flags,
175 vreinterpretq_u8_u32(rss)));
177 l3_l4e = vshrq_n_u32(vlan1, 22);
178 l3_l4e = vreinterpretq_u32_u8(vqtbl1q_u8(l3_l4e_flags,
179 vreinterpretq_u8_u32(l3_l4e)));
182 vlan0 = vorrq_u32(vlan0, rss);
183 vlan0 = vorrq_u32(vlan0, l3_l4e);
185 rx_pkts[0]->ol_flags = vgetq_lane_u32(vlan0, 0);
186 rx_pkts[1]->ol_flags = vgetq_lane_u32(vlan0, 1);
187 rx_pkts[2]->ol_flags = vgetq_lane_u32(vlan0, 2);
188 rx_pkts[3]->ol_flags = vgetq_lane_u32(vlan0, 3);
191 #define desc_to_olflags_v(descs, rx_pkts) do {} while (0)
194 #define PKTLEN_SHIFT 10
196 #define I40E_VPMD_DESC_DD_MASK 0x0001000100010001ULL
199 desc_to_ptype_v(uint64x2_t descs[4], struct rte_mbuf **rx_pkts)
205 for (i = 0; i < 4; i++) {
206 tmp = vreinterpretq_u8_u64(vshrq_n_u64(descs[i], 30));
207 ptype = vgetq_lane_u8(tmp, 8);
208 rx_pkts[i]->packet_type = i40e_rxd_pkt_type_mapping(ptype);
215 * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet
216 * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST
219 static inline uint16_t
220 _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
221 uint16_t nb_pkts, uint8_t *split_packet)
223 volatile union i40e_rx_desc *rxdp;
224 struct i40e_rx_entry *sw_ring;
225 uint16_t nb_pkts_recd;
229 /* mask to shuffle from desc. to mbuf */
230 uint8x16_t shuf_msk = {
231 0xFF, 0xFF, /* pkt_type set as unknown */
232 0xFF, 0xFF, /* pkt_type set as unknown */
233 14, 15, /* octet 15~14, low 16 bits pkt_len */
234 0xFF, 0xFF, /* skip high 16 bits pkt_len, zero out */
235 14, 15, /* octet 15~14, 16 bits data_len */
236 2, 3, /* octet 2~3, low 16 bits vlan_macip */
237 4, 5, 6, 7 /* octet 4~7, 32bits rss */
240 uint8x16_t eop_check = {
241 0x02, 0x00, 0x02, 0x00,
242 0x02, 0x00, 0x02, 0x00,
243 0x00, 0x00, 0x00, 0x00,
244 0x00, 0x00, 0x00, 0x00
247 uint16x8_t crc_adjust = {
248 0, 0, /* ignore pkt_type field */
249 rxq->crc_len, /* sub crc on pkt_len */
250 0, /* ignore high-16bits of pkt_len */
251 rxq->crc_len, /* sub crc on data_len */
252 0, 0, 0 /* ignore non-length fields */
255 /* nb_pkts shall be less equal than RTE_I40E_MAX_RX_BURST */
256 nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_MAX_RX_BURST);
258 /* nb_pkts has to be floor-aligned to RTE_I40E_DESCS_PER_LOOP */
259 nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_I40E_DESCS_PER_LOOP);
261 /* Just the act of getting into the function from the application is
262 * going to cost about 7 cycles
264 rxdp = rxq->rx_ring + rxq->rx_tail;
266 rte_prefetch_non_temporal(rxdp);
268 /* See if we need to rearm the RX queue - gives the prefetch a bit
271 if (rxq->rxrearm_nb > RTE_I40E_RXQ_REARM_THRESH)
274 /* Before we start moving massive data around, check to see if
275 * there is actually a packet available
277 if (!(rxdp->wb.qword1.status_error_len &
278 rte_cpu_to_le_32(1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
281 /* Cache is empty -> need to scan the buffer rings, but first move
282 * the next 'n' mbufs into the cache
284 sw_ring = &rxq->sw_ring[rxq->rx_tail];
286 /* A. load 4 packet in one loop
287 * [A*. mask out 4 unused dirty field in desc]
288 * B. copy 4 mbuf point from swring to rx_pkts
289 * C. calc the number of DD bits among the 4 packets
290 * [C*. extract the end-of-packet bit, if requested]
291 * D. fill info. from desc to mbuf
294 for (pos = 0, nb_pkts_recd = 0; pos < nb_pkts;
295 pos += RTE_I40E_DESCS_PER_LOOP,
296 rxdp += RTE_I40E_DESCS_PER_LOOP) {
297 uint64x2_t descs[RTE_I40E_DESCS_PER_LOOP];
298 uint8x16_t pkt_mb1, pkt_mb2, pkt_mb3, pkt_mb4;
299 uint16x8x2_t sterr_tmp1, sterr_tmp2;
300 uint64x2_t mbp1, mbp2;
305 int32x4_t len_shl = {0, 0, 0, PKTLEN_SHIFT};
307 /* B.1 load 1 mbuf point */
308 mbp1 = vld1q_u64((uint64_t *)&sw_ring[pos]);
309 /* Read desc statuses backwards to avoid race condition */
310 /* A.1 load 4 pkts desc */
311 descs[3] = vld1q_u64((uint64_t *)(rxdp + 3));
314 /* B.2 copy 2 mbuf point into rx_pkts */
315 vst1q_u64((uint64_t *)&rx_pkts[pos], mbp1);
317 /* B.1 load 1 mbuf point */
318 mbp2 = vld1q_u64((uint64_t *)&sw_ring[pos + 2]);
320 descs[2] = vld1q_u64((uint64_t *)(rxdp + 2));
321 /* B.1 load 2 mbuf point */
322 descs[1] = vld1q_u64((uint64_t *)(rxdp + 1));
323 descs[0] = vld1q_u64((uint64_t *)(rxdp));
325 /* B.2 copy 2 mbuf point into rx_pkts */
326 vst1q_u64((uint64_t *)&rx_pkts[pos + 2], mbp2);
329 rte_mbuf_prefetch_part2(rx_pkts[pos]);
330 rte_mbuf_prefetch_part2(rx_pkts[pos + 1]);
331 rte_mbuf_prefetch_part2(rx_pkts[pos + 2]);
332 rte_mbuf_prefetch_part2(rx_pkts[pos + 3]);
335 /* avoid compiler reorder optimization */
336 rte_compiler_barrier();
338 /* pkt 3,4 shift the pktlen field to be 16-bit aligned*/
339 uint32x4_t len3 = vshlq_u32(vreinterpretq_u32_u64(descs[3]),
341 descs[3] = vreinterpretq_u64_u32(len3);
342 uint32x4_t len2 = vshlq_u32(vreinterpretq_u32_u64(descs[2]),
344 descs[2] = vreinterpretq_u64_u32(len2);
346 /* D.1 pkt 3,4 convert format from desc to pktmbuf */
347 pkt_mb4 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[3]), shuf_msk);
348 pkt_mb3 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[2]), shuf_msk);
350 /* C.1 4=>2 filter staterr info only */
351 sterr_tmp2 = vzipq_u16(vreinterpretq_u16_u64(descs[1]),
352 vreinterpretq_u16_u64(descs[3]));
353 /* C.1 4=>2 filter staterr info only */
354 sterr_tmp1 = vzipq_u16(vreinterpretq_u16_u64(descs[0]),
355 vreinterpretq_u16_u64(descs[2]));
357 /* C.2 get 4 pkts staterr value */
358 staterr = vzipq_u16(sterr_tmp1.val[1],
359 sterr_tmp2.val[1]).val[0];
360 stat = vgetq_lane_u64(vreinterpretq_u64_u16(staterr), 0);
362 desc_to_olflags_v(descs, &rx_pkts[pos]);
364 /* D.2 pkt 3,4 set in_port/nb_seg and remove crc */
365 tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb4), crc_adjust);
366 pkt_mb4 = vreinterpretq_u8_u16(tmp);
367 tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb3), crc_adjust);
368 pkt_mb3 = vreinterpretq_u8_u16(tmp);
370 /* pkt 1,2 shift the pktlen field to be 16-bit aligned*/
371 uint32x4_t len1 = vshlq_u32(vreinterpretq_u32_u64(descs[1]),
373 descs[1] = vreinterpretq_u64_u32(len1);
374 uint32x4_t len0 = vshlq_u32(vreinterpretq_u32_u64(descs[0]),
376 descs[0] = vreinterpretq_u64_u32(len0);
378 /* D.1 pkt 1,2 convert format from desc to pktmbuf */
379 pkt_mb2 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[1]), shuf_msk);
380 pkt_mb1 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[0]), shuf_msk);
382 /* D.3 copy final 3,4 data to rx_pkts */
383 vst1q_u8((void *)&rx_pkts[pos + 3]->rx_descriptor_fields1,
385 vst1q_u8((void *)&rx_pkts[pos + 2]->rx_descriptor_fields1,
388 /* D.2 pkt 1,2 set in_port/nb_seg and remove crc */
389 tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb2), crc_adjust);
390 pkt_mb2 = vreinterpretq_u8_u16(tmp);
391 tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb1), crc_adjust);
392 pkt_mb1 = vreinterpretq_u8_u16(tmp);
394 /* C* extract and record EOP bit */
396 uint8x16_t eop_shuf_mask = {
397 0x00, 0x02, 0x04, 0x06,
398 0xFF, 0xFF, 0xFF, 0xFF,
399 0xFF, 0xFF, 0xFF, 0xFF,
400 0xFF, 0xFF, 0xFF, 0xFF};
403 /* and with mask to extract bits, flipping 1-0 */
404 eop_bits = vmvnq_u8(vreinterpretq_u8_u16(staterr));
405 eop_bits = vandq_u8(eop_bits, eop_check);
406 /* the staterr values are not in order, as the count
407 * count of dd bits doesn't care. However, for end of
408 * packet tracking, we do care, so shuffle. This also
409 * compresses the 32-bit values to 8-bit
411 eop_bits = vqtbl1q_u8(eop_bits, eop_shuf_mask);
413 /* store the resulting 32-bit value */
414 vst1q_lane_u32((uint32_t *)split_packet,
415 vreinterpretq_u32_u8(eop_bits), 0);
416 split_packet += RTE_I40E_DESCS_PER_LOOP;
418 /* zero-out next pointers */
419 rx_pkts[pos]->next = NULL;
420 rx_pkts[pos + 1]->next = NULL;
421 rx_pkts[pos + 2]->next = NULL;
422 rx_pkts[pos + 3]->next = NULL;
425 rte_prefetch_non_temporal(rxdp + RTE_I40E_DESCS_PER_LOOP);
427 /* D.3 copy final 1,2 data to rx_pkts */
428 vst1q_u8((void *)&rx_pkts[pos + 1]->rx_descriptor_fields1,
430 vst1q_u8((void *)&rx_pkts[pos]->rx_descriptor_fields1,
432 desc_to_ptype_v(descs, &rx_pkts[pos]);
433 /* C.4 calc avaialbe number of desc */
434 var = __builtin_popcountll(stat & I40E_VPMD_DESC_DD_MASK);
436 if (likely(var != RTE_I40E_DESCS_PER_LOOP))
440 /* Update our internal tail pointer */
441 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_pkts_recd);
442 rxq->rx_tail = (uint16_t)(rxq->rx_tail & (rxq->nb_rx_desc - 1));
443 rxq->rxrearm_nb = (uint16_t)(rxq->rxrearm_nb + nb_pkts_recd);
450 * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet
451 * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST
455 i40e_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
458 return _recv_raw_pkts_vec(rx_queue, rx_pkts, nb_pkts, NULL);
461 /* vPMD receive routine that reassembles scattered packets
463 * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet
464 * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST
468 i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
472 struct i40e_rx_queue *rxq = rx_queue;
473 uint8_t split_flags[RTE_I40E_VPMD_RX_BURST] = {0};
475 /* get some new buffers */
476 uint16_t nb_bufs = _recv_raw_pkts_vec(rxq, rx_pkts, nb_pkts,
481 /* happy day case, full burst + no packets to be joined */
482 const uint64_t *split_fl64 = (uint64_t *)split_flags;
484 if (rxq->pkt_first_seg == NULL &&
485 split_fl64[0] == 0 && split_fl64[1] == 0 &&
486 split_fl64[2] == 0 && split_fl64[3] == 0)
489 /* reassemble any packets that need reassembly*/
492 if (rxq->pkt_first_seg == NULL) {
493 /* find the first split flag, and only reassemble then*/
494 while (i < nb_bufs && !split_flags[i])
499 return i + reassemble_packets(rxq, &rx_pkts[i], nb_bufs - i,
504 vtx1(volatile struct i40e_tx_desc *txdp,
505 struct rte_mbuf *pkt, uint64_t flags)
507 uint64_t high_qw = (I40E_TX_DESC_DTYPE_DATA |
508 ((uint64_t)flags << I40E_TXD_QW1_CMD_SHIFT) |
509 ((uint64_t)pkt->data_len << I40E_TXD_QW1_TX_BUF_SZ_SHIFT));
511 uint64x2_t descriptor = {pkt->buf_physaddr + pkt->data_off, high_qw};
512 vst1q_u64((uint64_t *)txdp, descriptor);
516 vtx(volatile struct i40e_tx_desc *txdp,
517 struct rte_mbuf **pkt, uint16_t nb_pkts, uint64_t flags)
521 for (i = 0; i < nb_pkts; ++i, ++txdp, ++pkt)
522 vtx1(txdp, *pkt, flags);
526 i40e_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
529 struct i40e_tx_queue *txq = (struct i40e_tx_queue *)tx_queue;
530 volatile struct i40e_tx_desc *txdp;
531 struct i40e_tx_entry *txep;
532 uint16_t n, nb_commit, tx_id;
533 uint64_t flags = I40E_TD_CMD;
534 uint64_t rs = I40E_TX_DESC_CMD_RS | I40E_TD_CMD;
537 /* cross rx_thresh boundary is not allowed */
538 nb_pkts = RTE_MIN(nb_pkts, txq->tx_rs_thresh);
540 if (txq->nb_tx_free < txq->tx_free_thresh)
541 i40e_tx_free_bufs(txq);
543 nb_commit = nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
544 if (unlikely(nb_pkts == 0))
547 tx_id = txq->tx_tail;
548 txdp = &txq->tx_ring[tx_id];
549 txep = &txq->sw_ring[tx_id];
551 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
553 n = (uint16_t)(txq->nb_tx_desc - tx_id);
554 if (nb_commit >= n) {
555 tx_backlog_entry(txep, tx_pkts, n);
557 for (i = 0; i < n - 1; ++i, ++tx_pkts, ++txdp)
558 vtx1(txdp, *tx_pkts, flags);
560 vtx1(txdp, *tx_pkts++, rs);
562 nb_commit = (uint16_t)(nb_commit - n);
565 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
567 /* avoid reach the end of ring */
568 txdp = &txq->tx_ring[tx_id];
569 txep = &txq->sw_ring[tx_id];
572 tx_backlog_entry(txep, tx_pkts, nb_commit);
574 vtx(txdp, tx_pkts, nb_commit, flags);
576 tx_id = (uint16_t)(tx_id + nb_commit);
577 if (tx_id > txq->tx_next_rs) {
578 txq->tx_ring[txq->tx_next_rs].cmd_type_offset_bsz |=
579 rte_cpu_to_le_64(((uint64_t)I40E_TX_DESC_CMD_RS) <<
580 I40E_TXD_QW1_CMD_SHIFT);
582 (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh);
585 txq->tx_tail = tx_id;
587 I40E_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
592 void __attribute__((cold))
593 i40e_rx_queue_release_mbufs_vec(struct i40e_rx_queue *rxq)
595 _i40e_rx_queue_release_mbufs_vec(rxq);
598 int __attribute__((cold))
599 i40e_rxq_vec_setup(struct i40e_rx_queue *rxq)
601 return i40e_rxq_vec_setup_default(rxq);
604 int __attribute__((cold))
605 i40e_txq_vec_setup(struct i40e_tx_queue __rte_unused *txq)
610 int __attribute__((cold))
611 i40e_rx_vec_dev_conf_condition_check(struct rte_eth_dev *dev)
613 return i40e_rx_vec_dev_conf_condition_check_default(dev);