1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation.
3 * Copyright 2014 6WIND S.A.
17 #include <rte_byteorder.h>
18 #include <rte_common.h>
19 #include <rte_cycles.h>
21 #include <rte_debug.h>
22 #include <rte_interrupts.h>
24 #include <rte_memory.h>
25 #include <rte_memzone.h>
26 #include <rte_launch.h>
28 #include <rte_per_lcore.h>
29 #include <rte_lcore.h>
30 #include <rte_atomic.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_mempool.h>
33 #include <rte_malloc.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev_driver.h>
37 #include <rte_prefetch.h>
41 #include <rte_string_fns.h>
42 #include <rte_errno.h>
46 #include "ixgbe_logs.h"
47 #include "base/ixgbe_api.h"
48 #include "base/ixgbe_vf.h"
49 #include "ixgbe_ethdev.h"
50 #include "base/ixgbe_dcb.h"
51 #include "base/ixgbe_common.h"
52 #include "ixgbe_rxtx.h"
54 #ifdef RTE_LIBRTE_IEEE1588
55 #define IXGBE_TX_IEEE1588_TMST PKT_TX_IEEE1588_TMST
57 #define IXGBE_TX_IEEE1588_TMST 0
59 /* Bit Mask to indicate what bits required for building TX context */
60 #define IXGBE_TX_OFFLOAD_MASK ( \
70 PKT_TX_OUTER_IP_CKSUM | \
71 PKT_TX_SEC_OFFLOAD | \
72 IXGBE_TX_IEEE1588_TMST)
74 #define IXGBE_TX_OFFLOAD_NOTSUP_MASK \
75 (PKT_TX_OFFLOAD_MASK ^ IXGBE_TX_OFFLOAD_MASK)
78 #define RTE_PMD_USE_PREFETCH
81 #ifdef RTE_PMD_USE_PREFETCH
83 * Prefetch a cache line into all cache levels.
85 #define rte_ixgbe_prefetch(p) rte_prefetch0(p)
87 #define rte_ixgbe_prefetch(p) do {} while (0)
90 #ifdef RTE_IXGBE_INC_VECTOR
91 uint16_t ixgbe_xmit_fixed_burst_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
95 /*********************************************************************
99 **********************************************************************/
102 * Check for descriptors with their DD bit set and free mbufs.
103 * Return the total number of buffers freed.
105 static __rte_always_inline int
106 ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
108 struct ixgbe_tx_entry *txep;
111 struct rte_mbuf *m, *free[RTE_IXGBE_TX_MAX_FREE_BUF_SZ];
113 /* check DD bit on threshold descriptor */
114 status = txq->tx_ring[txq->tx_next_dd].wb.status;
115 if (!(status & rte_cpu_to_le_32(IXGBE_ADVTXD_STAT_DD)))
119 * first buffer to free from S/W ring is at index
120 * tx_next_dd - (tx_rs_thresh-1)
122 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
124 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
125 /* free buffers one at a time */
126 m = rte_pktmbuf_prefree_seg(txep->mbuf);
129 if (unlikely(m == NULL))
132 if (nb_free >= RTE_IXGBE_TX_MAX_FREE_BUF_SZ ||
133 (nb_free > 0 && m->pool != free[0]->pool)) {
134 rte_mempool_put_bulk(free[0]->pool,
135 (void **)free, nb_free);
143 rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
145 /* buffers were freed, update counters */
146 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
147 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
148 if (txq->tx_next_dd >= txq->nb_tx_desc)
149 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
151 return txq->tx_rs_thresh;
154 /* Populate 4 descriptors with data from 4 mbufs */
156 tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
158 uint64_t buf_dma_addr;
162 for (i = 0; i < 4; ++i, ++txdp, ++pkts) {
163 buf_dma_addr = rte_mbuf_data_iova(*pkts);
164 pkt_len = (*pkts)->data_len;
166 /* write data to descriptor */
167 txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
169 txdp->read.cmd_type_len =
170 rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
172 txdp->read.olinfo_status =
173 rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
175 rte_prefetch0(&(*pkts)->pool);
179 /* Populate 1 descriptor with data from 1 mbuf */
181 tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
183 uint64_t buf_dma_addr;
186 buf_dma_addr = rte_mbuf_data_iova(*pkts);
187 pkt_len = (*pkts)->data_len;
189 /* write data to descriptor */
190 txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
191 txdp->read.cmd_type_len =
192 rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
193 txdp->read.olinfo_status =
194 rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
195 rte_prefetch0(&(*pkts)->pool);
199 * Fill H/W descriptor ring with mbuf data.
200 * Copy mbuf pointers to the S/W ring.
203 ixgbe_tx_fill_hw_ring(struct ixgbe_tx_queue *txq, struct rte_mbuf **pkts,
206 volatile union ixgbe_adv_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
207 struct ixgbe_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
208 const int N_PER_LOOP = 4;
209 const int N_PER_LOOP_MASK = N_PER_LOOP-1;
210 int mainpart, leftover;
214 * Process most of the packets in chunks of N pkts. Any
215 * leftover packets will get processed one at a time.
217 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
218 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
219 for (i = 0; i < mainpart; i += N_PER_LOOP) {
220 /* Copy N mbuf pointers to the S/W ring */
221 for (j = 0; j < N_PER_LOOP; ++j) {
222 (txep + i + j)->mbuf = *(pkts + i + j);
224 tx4(txdp + i, pkts + i);
227 if (unlikely(leftover > 0)) {
228 for (i = 0; i < leftover; ++i) {
229 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
230 tx1(txdp + mainpart + i, pkts + mainpart + i);
235 static inline uint16_t
236 tx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
239 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
240 volatile union ixgbe_adv_tx_desc *tx_r = txq->tx_ring;
244 * Begin scanning the H/W ring for done descriptors when the
245 * number of available descriptors drops below tx_free_thresh. For
246 * each done descriptor, free the associated buffer.
248 if (txq->nb_tx_free < txq->tx_free_thresh)
249 ixgbe_tx_free_bufs(txq);
251 /* Only use descriptors that are available */
252 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
253 if (unlikely(nb_pkts == 0))
256 /* Use exactly nb_pkts descriptors */
257 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
260 * At this point, we know there are enough descriptors in the
261 * ring to transmit all the packets. This assumes that each
262 * mbuf contains a single segment, and that no new offloads
263 * are expected, which would require a new context descriptor.
267 * See if we're going to wrap-around. If so, handle the top
268 * of the descriptor ring first, then do the bottom. If not,
269 * the processing looks just like the "bottom" part anyway...
271 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
272 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
273 ixgbe_tx_fill_hw_ring(txq, tx_pkts, n);
276 * We know that the last descriptor in the ring will need to
277 * have its RS bit set because tx_rs_thresh has to be
278 * a divisor of the ring size
280 tx_r[txq->tx_next_rs].read.cmd_type_len |=
281 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
282 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
287 /* Fill H/W descriptor ring with mbuf data */
288 ixgbe_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
289 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
292 * Determine if RS bit should be set
293 * This is what we actually want:
294 * if ((txq->tx_tail - 1) >= txq->tx_next_rs)
295 * but instead of subtracting 1 and doing >=, we can just do
296 * greater than without subtracting.
298 if (txq->tx_tail > txq->tx_next_rs) {
299 tx_r[txq->tx_next_rs].read.cmd_type_len |=
300 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
301 txq->tx_next_rs = (uint16_t)(txq->tx_next_rs +
303 if (txq->tx_next_rs >= txq->nb_tx_desc)
304 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
308 * Check for wrap-around. This would only happen if we used
309 * up to the last descriptor in the ring, no more, no less.
311 if (txq->tx_tail >= txq->nb_tx_desc)
314 /* update tail pointer */
316 IXGBE_PCI_REG_WRITE_RELAXED(txq->tdt_reg_addr, txq->tx_tail);
322 ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
327 /* Try to transmit at least chunks of TX_MAX_BURST pkts */
328 if (likely(nb_pkts <= RTE_PMD_IXGBE_TX_MAX_BURST))
329 return tx_xmit_pkts(tx_queue, tx_pkts, nb_pkts);
331 /* transmit more than the max burst, in chunks of TX_MAX_BURST */
336 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_TX_MAX_BURST);
337 ret = tx_xmit_pkts(tx_queue, &(tx_pkts[nb_tx]), n);
338 nb_tx = (uint16_t)(nb_tx + ret);
339 nb_pkts = (uint16_t)(nb_pkts - ret);
347 #ifdef RTE_IXGBE_INC_VECTOR
349 ixgbe_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
353 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
358 num = (uint16_t)RTE_MIN(nb_pkts, txq->tx_rs_thresh);
359 ret = ixgbe_xmit_fixed_burst_vec(tx_queue, &tx_pkts[nb_tx],
372 ixgbe_set_xmit_ctx(struct ixgbe_tx_queue *txq,
373 volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
374 uint64_t ol_flags, union ixgbe_tx_offload tx_offload,
375 __rte_unused uint64_t *mdata)
377 uint32_t type_tucmd_mlhl;
378 uint32_t mss_l4len_idx = 0;
380 uint32_t vlan_macip_lens;
381 union ixgbe_tx_offload tx_offload_mask;
382 uint32_t seqnum_seed = 0;
384 ctx_idx = txq->ctx_curr;
385 tx_offload_mask.data[0] = 0;
386 tx_offload_mask.data[1] = 0;
389 /* Specify which HW CTX to upload. */
390 mss_l4len_idx |= (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
392 if (ol_flags & PKT_TX_VLAN_PKT) {
393 tx_offload_mask.vlan_tci |= ~0;
396 /* check if TCP segmentation required for this packet */
397 if (ol_flags & PKT_TX_TCP_SEG) {
398 /* implies IP cksum in IPv4 */
399 if (ol_flags & PKT_TX_IP_CKSUM)
400 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4 |
401 IXGBE_ADVTXD_TUCMD_L4T_TCP |
402 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
404 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV6 |
405 IXGBE_ADVTXD_TUCMD_L4T_TCP |
406 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
408 tx_offload_mask.l2_len |= ~0;
409 tx_offload_mask.l3_len |= ~0;
410 tx_offload_mask.l4_len |= ~0;
411 tx_offload_mask.tso_segsz |= ~0;
412 mss_l4len_idx |= tx_offload.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT;
413 mss_l4len_idx |= tx_offload.l4_len << IXGBE_ADVTXD_L4LEN_SHIFT;
414 } else { /* no TSO, check if hardware checksum is needed */
415 if (ol_flags & PKT_TX_IP_CKSUM) {
416 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
417 tx_offload_mask.l2_len |= ~0;
418 tx_offload_mask.l3_len |= ~0;
421 switch (ol_flags & PKT_TX_L4_MASK) {
422 case PKT_TX_UDP_CKSUM:
423 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
424 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
425 mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
426 tx_offload_mask.l2_len |= ~0;
427 tx_offload_mask.l3_len |= ~0;
429 case PKT_TX_TCP_CKSUM:
430 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
431 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
432 mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
433 tx_offload_mask.l2_len |= ~0;
434 tx_offload_mask.l3_len |= ~0;
436 case PKT_TX_SCTP_CKSUM:
437 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
438 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
439 mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
440 tx_offload_mask.l2_len |= ~0;
441 tx_offload_mask.l3_len |= ~0;
444 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
445 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
450 if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
451 tx_offload_mask.outer_l2_len |= ~0;
452 tx_offload_mask.outer_l3_len |= ~0;
453 tx_offload_mask.l2_len |= ~0;
454 seqnum_seed |= tx_offload.outer_l3_len
455 << IXGBE_ADVTXD_OUTER_IPLEN;
456 seqnum_seed |= tx_offload.l2_len
457 << IXGBE_ADVTXD_TUNNEL_LEN;
459 #ifdef RTE_LIBRTE_SECURITY
460 if (ol_flags & PKT_TX_SEC_OFFLOAD) {
461 union ixgbe_crypto_tx_desc_md *md =
462 (union ixgbe_crypto_tx_desc_md *)mdata;
464 (IXGBE_ADVTXD_IPSEC_SA_INDEX_MASK & md->sa_idx);
465 type_tucmd_mlhl |= md->enc ?
466 (IXGBE_ADVTXD_TUCMD_IPSEC_TYPE_ESP |
467 IXGBE_ADVTXD_TUCMD_IPSEC_ENCRYPT_EN) : 0;
469 (md->pad_len & IXGBE_ADVTXD_IPSEC_ESP_LEN_MASK);
470 tx_offload_mask.sa_idx |= ~0;
471 tx_offload_mask.sec_pad_len |= ~0;
475 txq->ctx_cache[ctx_idx].flags = ol_flags;
476 txq->ctx_cache[ctx_idx].tx_offload.data[0] =
477 tx_offload_mask.data[0] & tx_offload.data[0];
478 txq->ctx_cache[ctx_idx].tx_offload.data[1] =
479 tx_offload_mask.data[1] & tx_offload.data[1];
480 txq->ctx_cache[ctx_idx].tx_offload_mask = tx_offload_mask;
482 ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
483 vlan_macip_lens = tx_offload.l3_len;
484 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
485 vlan_macip_lens |= (tx_offload.outer_l2_len <<
486 IXGBE_ADVTXD_MACLEN_SHIFT);
488 vlan_macip_lens |= (tx_offload.l2_len <<
489 IXGBE_ADVTXD_MACLEN_SHIFT);
490 vlan_macip_lens |= ((uint32_t)tx_offload.vlan_tci << IXGBE_ADVTXD_VLAN_SHIFT);
491 ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
492 ctx_txd->mss_l4len_idx = rte_cpu_to_le_32(mss_l4len_idx);
493 ctx_txd->seqnum_seed = seqnum_seed;
497 * Check which hardware context can be used. Use the existing match
498 * or create a new context descriptor.
500 static inline uint32_t
501 what_advctx_update(struct ixgbe_tx_queue *txq, uint64_t flags,
502 union ixgbe_tx_offload tx_offload)
504 /* If match with the current used context */
505 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
506 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[0] ==
507 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[0]
508 & tx_offload.data[0])) &&
509 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[1] ==
510 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[1]
511 & tx_offload.data[1]))))
512 return txq->ctx_curr;
514 /* What if match with the next context */
516 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
517 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[0] ==
518 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[0]
519 & tx_offload.data[0])) &&
520 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[1] ==
521 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[1]
522 & tx_offload.data[1]))))
523 return txq->ctx_curr;
525 /* Mismatch, use the previous context */
526 return IXGBE_CTX_NUM;
529 static inline uint32_t
530 tx_desc_cksum_flags_to_olinfo(uint64_t ol_flags)
534 if ((ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM)
535 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
536 if (ol_flags & PKT_TX_IP_CKSUM)
537 tmp |= IXGBE_ADVTXD_POPTS_IXSM;
538 if (ol_flags & PKT_TX_TCP_SEG)
539 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
543 static inline uint32_t
544 tx_desc_ol_flags_to_cmdtype(uint64_t ol_flags)
546 uint32_t cmdtype = 0;
548 if (ol_flags & PKT_TX_VLAN_PKT)
549 cmdtype |= IXGBE_ADVTXD_DCMD_VLE;
550 if (ol_flags & PKT_TX_TCP_SEG)
551 cmdtype |= IXGBE_ADVTXD_DCMD_TSE;
552 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
553 cmdtype |= (1 << IXGBE_ADVTXD_OUTERIPCS_SHIFT);
554 if (ol_flags & PKT_TX_MACSEC)
555 cmdtype |= IXGBE_ADVTXD_MAC_LINKSEC;
559 /* Default RS bit threshold values */
560 #ifndef DEFAULT_TX_RS_THRESH
561 #define DEFAULT_TX_RS_THRESH 32
563 #ifndef DEFAULT_TX_FREE_THRESH
564 #define DEFAULT_TX_FREE_THRESH 32
567 /* Reset transmit descriptors after they have been used */
569 ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
571 struct ixgbe_tx_entry *sw_ring = txq->sw_ring;
572 volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
573 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
574 uint16_t nb_tx_desc = txq->nb_tx_desc;
575 uint16_t desc_to_clean_to;
576 uint16_t nb_tx_to_clean;
579 /* Determine the last descriptor needing to be cleaned */
580 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
581 if (desc_to_clean_to >= nb_tx_desc)
582 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
584 /* Check to make sure the last descriptor to clean is done */
585 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
586 status = txr[desc_to_clean_to].wb.status;
587 if (!(status & rte_cpu_to_le_32(IXGBE_TXD_STAT_DD))) {
588 PMD_TX_FREE_LOG(DEBUG,
589 "TX descriptor %4u is not done"
590 "(port=%d queue=%d)",
592 txq->port_id, txq->queue_id);
593 /* Failed to clean any descriptors, better luck next time */
597 /* Figure out how many descriptors will be cleaned */
598 if (last_desc_cleaned > desc_to_clean_to)
599 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
602 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
605 PMD_TX_FREE_LOG(DEBUG,
606 "Cleaning %4u TX descriptors: %4u to %4u "
607 "(port=%d queue=%d)",
608 nb_tx_to_clean, last_desc_cleaned, desc_to_clean_to,
609 txq->port_id, txq->queue_id);
612 * The last descriptor to clean is done, so that means all the
613 * descriptors from the last descriptor that was cleaned
614 * up to the last descriptor with the RS bit set
615 * are done. Only reset the threshold descriptor.
617 txr[desc_to_clean_to].wb.status = 0;
619 /* Update the txq to reflect the last descriptor that was cleaned */
620 txq->last_desc_cleaned = desc_to_clean_to;
621 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
628 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
631 struct ixgbe_tx_queue *txq;
632 struct ixgbe_tx_entry *sw_ring;
633 struct ixgbe_tx_entry *txe, *txn;
634 volatile union ixgbe_adv_tx_desc *txr;
635 volatile union ixgbe_adv_tx_desc *txd, *txp;
636 struct rte_mbuf *tx_pkt;
637 struct rte_mbuf *m_seg;
638 uint64_t buf_dma_addr;
639 uint32_t olinfo_status;
640 uint32_t cmd_type_len;
651 union ixgbe_tx_offload tx_offload;
652 #ifdef RTE_LIBRTE_SECURITY
656 tx_offload.data[0] = 0;
657 tx_offload.data[1] = 0;
659 sw_ring = txq->sw_ring;
661 tx_id = txq->tx_tail;
662 txe = &sw_ring[tx_id];
665 /* Determine if the descriptor ring needs to be cleaned. */
666 if (txq->nb_tx_free < txq->tx_free_thresh)
667 ixgbe_xmit_cleanup(txq);
669 rte_prefetch0(&txe->mbuf->pool);
672 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
675 pkt_len = tx_pkt->pkt_len;
678 * Determine how many (if any) context descriptors
679 * are needed for offload functionality.
681 ol_flags = tx_pkt->ol_flags;
682 #ifdef RTE_LIBRTE_SECURITY
683 use_ipsec = txq->using_ipsec && (ol_flags & PKT_TX_SEC_OFFLOAD);
686 /* If hardware offload required */
687 tx_ol_req = ol_flags & IXGBE_TX_OFFLOAD_MASK;
689 tx_offload.l2_len = tx_pkt->l2_len;
690 tx_offload.l3_len = tx_pkt->l3_len;
691 tx_offload.l4_len = tx_pkt->l4_len;
692 tx_offload.vlan_tci = tx_pkt->vlan_tci;
693 tx_offload.tso_segsz = tx_pkt->tso_segsz;
694 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
695 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
696 #ifdef RTE_LIBRTE_SECURITY
698 union ixgbe_crypto_tx_desc_md *ipsec_mdata =
699 (union ixgbe_crypto_tx_desc_md *)
701 tx_offload.sa_idx = ipsec_mdata->sa_idx;
702 tx_offload.sec_pad_len = ipsec_mdata->pad_len;
706 /* If new context need be built or reuse the exist ctx. */
707 ctx = what_advctx_update(txq, tx_ol_req,
709 /* Only allocate context descriptor if required*/
710 new_ctx = (ctx == IXGBE_CTX_NUM);
715 * Keep track of how many descriptors are used this loop
716 * This will always be the number of segments + the number of
717 * Context descriptors required to transmit the packet
719 nb_used = (uint16_t)(tx_pkt->nb_segs + new_ctx);
722 nb_used + txq->nb_tx_used >= txq->tx_rs_thresh)
723 /* set RS on the previous packet in the burst */
724 txp->read.cmd_type_len |=
725 rte_cpu_to_le_32(IXGBE_TXD_CMD_RS);
728 * The number of descriptors that must be allocated for a
729 * packet is the number of segments of that packet, plus 1
730 * Context Descriptor for the hardware offload, if any.
731 * Determine the last TX descriptor to allocate in the TX ring
732 * for the packet, starting from the current position (tx_id)
735 tx_last = (uint16_t) (tx_id + nb_used - 1);
738 if (tx_last >= txq->nb_tx_desc)
739 tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
741 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
742 " tx_first=%u tx_last=%u",
743 (unsigned) txq->port_id,
744 (unsigned) txq->queue_id,
750 * Make sure there are enough TX descriptors available to
751 * transmit the entire packet.
752 * nb_used better be less than or equal to txq->tx_rs_thresh
754 if (nb_used > txq->nb_tx_free) {
755 PMD_TX_FREE_LOG(DEBUG,
756 "Not enough free TX descriptors "
757 "nb_used=%4u nb_free=%4u "
758 "(port=%d queue=%d)",
759 nb_used, txq->nb_tx_free,
760 txq->port_id, txq->queue_id);
762 if (ixgbe_xmit_cleanup(txq) != 0) {
763 /* Could not clean any descriptors */
769 /* nb_used better be <= txq->tx_rs_thresh */
770 if (unlikely(nb_used > txq->tx_rs_thresh)) {
771 PMD_TX_FREE_LOG(DEBUG,
772 "The number of descriptors needed to "
773 "transmit the packet exceeds the "
774 "RS bit threshold. This will impact "
776 "nb_used=%4u nb_free=%4u "
778 "(port=%d queue=%d)",
779 nb_used, txq->nb_tx_free,
781 txq->port_id, txq->queue_id);
783 * Loop here until there are enough TX
784 * descriptors or until the ring cannot be
787 while (nb_used > txq->nb_tx_free) {
788 if (ixgbe_xmit_cleanup(txq) != 0) {
790 * Could not clean any
802 * By now there are enough free TX descriptors to transmit
807 * Set common flags of all TX Data Descriptors.
809 * The following bits must be set in all Data Descriptors:
810 * - IXGBE_ADVTXD_DTYP_DATA
811 * - IXGBE_ADVTXD_DCMD_DEXT
813 * The following bits must be set in the first Data Descriptor
814 * and are ignored in the other ones:
815 * - IXGBE_ADVTXD_DCMD_IFCS
816 * - IXGBE_ADVTXD_MAC_1588
817 * - IXGBE_ADVTXD_DCMD_VLE
819 * The following bits must only be set in the last Data
821 * - IXGBE_TXD_CMD_EOP
823 * The following bits can be set in any Data Descriptor, but
824 * are only set in the last Data Descriptor:
827 cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
828 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
830 #ifdef RTE_LIBRTE_IEEE1588
831 if (ol_flags & PKT_TX_IEEE1588_TMST)
832 cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
838 if (ol_flags & PKT_TX_TCP_SEG) {
839 /* when TSO is on, paylen in descriptor is the
840 * not the packet len but the tcp payload len */
841 pkt_len -= (tx_offload.l2_len +
842 tx_offload.l3_len + tx_offload.l4_len);
846 * Setup the TX Advanced Context Descriptor if required
849 volatile struct ixgbe_adv_tx_context_desc *
852 ctx_txd = (volatile struct
853 ixgbe_adv_tx_context_desc *)
856 txn = &sw_ring[txe->next_id];
857 rte_prefetch0(&txn->mbuf->pool);
859 if (txe->mbuf != NULL) {
860 rte_pktmbuf_free_seg(txe->mbuf);
864 ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
865 tx_offload, &tx_pkt->udata64);
867 txe->last_id = tx_last;
868 tx_id = txe->next_id;
873 * Setup the TX Advanced Data Descriptor,
874 * This path will go through
875 * whatever new/reuse the context descriptor
877 cmd_type_len |= tx_desc_ol_flags_to_cmdtype(ol_flags);
878 olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
879 olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
882 olinfo_status |= (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
883 #ifdef RTE_LIBRTE_SECURITY
885 olinfo_status |= IXGBE_ADVTXD_POPTS_IPSEC;
891 txn = &sw_ring[txe->next_id];
892 rte_prefetch0(&txn->mbuf->pool);
894 if (txe->mbuf != NULL)
895 rte_pktmbuf_free_seg(txe->mbuf);
899 * Set up Transmit Data Descriptor.
901 slen = m_seg->data_len;
902 buf_dma_addr = rte_mbuf_data_iova(m_seg);
903 txd->read.buffer_addr =
904 rte_cpu_to_le_64(buf_dma_addr);
905 txd->read.cmd_type_len =
906 rte_cpu_to_le_32(cmd_type_len | slen);
907 txd->read.olinfo_status =
908 rte_cpu_to_le_32(olinfo_status);
909 txe->last_id = tx_last;
910 tx_id = txe->next_id;
913 } while (m_seg != NULL);
916 * The last packet data descriptor needs End Of Packet (EOP)
918 cmd_type_len |= IXGBE_TXD_CMD_EOP;
919 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
920 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
922 /* Set RS bit only on threshold packets' last descriptor */
923 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
924 PMD_TX_FREE_LOG(DEBUG,
925 "Setting RS bit on TXD id="
926 "%4u (port=%d queue=%d)",
927 tx_last, txq->port_id, txq->queue_id);
929 cmd_type_len |= IXGBE_TXD_CMD_RS;
931 /* Update txq RS bit counters */
937 txd->read.cmd_type_len |= rte_cpu_to_le_32(cmd_type_len);
941 /* set RS on last packet in the burst */
943 txp->read.cmd_type_len |= rte_cpu_to_le_32(IXGBE_TXD_CMD_RS);
948 * Set the Transmit Descriptor Tail (TDT)
950 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
951 (unsigned) txq->port_id, (unsigned) txq->queue_id,
952 (unsigned) tx_id, (unsigned) nb_tx);
953 IXGBE_PCI_REG_WRITE_RELAXED(txq->tdt_reg_addr, tx_id);
954 txq->tx_tail = tx_id;
959 /*********************************************************************
963 **********************************************************************/
965 ixgbe_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
970 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
972 for (i = 0; i < nb_pkts; i++) {
974 ol_flags = m->ol_flags;
977 * Check if packet meets requirements for number of segments
979 * NOTE: for ixgbe it's always (40 - WTHRESH) for both TSO and
983 if (m->nb_segs > IXGBE_TX_MAX_SEG - txq->wthresh) {
988 if (ol_flags & IXGBE_TX_OFFLOAD_NOTSUP_MASK) {
989 rte_errno = -ENOTSUP;
993 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
994 ret = rte_validate_tx_offload(m);
1000 ret = rte_net_intel_cksum_prepare(m);
1010 /*********************************************************************
1014 **********************************************************************/
1016 #define IXGBE_PACKET_TYPE_ETHER 0X00
1017 #define IXGBE_PACKET_TYPE_IPV4 0X01
1018 #define IXGBE_PACKET_TYPE_IPV4_TCP 0X11
1019 #define IXGBE_PACKET_TYPE_IPV4_UDP 0X21
1020 #define IXGBE_PACKET_TYPE_IPV4_SCTP 0X41
1021 #define IXGBE_PACKET_TYPE_IPV4_EXT 0X03
1022 #define IXGBE_PACKET_TYPE_IPV4_EXT_TCP 0X13
1023 #define IXGBE_PACKET_TYPE_IPV4_EXT_UDP 0X23
1024 #define IXGBE_PACKET_TYPE_IPV4_EXT_SCTP 0X43
1025 #define IXGBE_PACKET_TYPE_IPV6 0X04
1026 #define IXGBE_PACKET_TYPE_IPV6_TCP 0X14
1027 #define IXGBE_PACKET_TYPE_IPV6_UDP 0X24
1028 #define IXGBE_PACKET_TYPE_IPV6_SCTP 0X44
1029 #define IXGBE_PACKET_TYPE_IPV6_EXT 0X0C
1030 #define IXGBE_PACKET_TYPE_IPV6_EXT_TCP 0X1C
1031 #define IXGBE_PACKET_TYPE_IPV6_EXT_UDP 0X2C
1032 #define IXGBE_PACKET_TYPE_IPV6_EXT_SCTP 0X4C
1033 #define IXGBE_PACKET_TYPE_IPV4_IPV6 0X05
1034 #define IXGBE_PACKET_TYPE_IPV4_IPV6_TCP 0X15
1035 #define IXGBE_PACKET_TYPE_IPV4_IPV6_UDP 0X25
1036 #define IXGBE_PACKET_TYPE_IPV4_IPV6_SCTP 0X45
1037 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6 0X07
1038 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_TCP 0X17
1039 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_UDP 0X27
1040 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_SCTP 0X47
1041 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT 0X0D
1042 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP 0X1D
1043 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP 0X2D
1044 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_SCTP 0X4D
1045 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT 0X0F
1046 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_TCP 0X1F
1047 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_UDP 0X2F
1048 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_SCTP 0X4F
1050 #define IXGBE_PACKET_TYPE_NVGRE 0X00
1051 #define IXGBE_PACKET_TYPE_NVGRE_IPV4 0X01
1052 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_TCP 0X11
1053 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_UDP 0X21
1054 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_SCTP 0X41
1055 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT 0X03
1056 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_TCP 0X13
1057 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_UDP 0X23
1058 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_SCTP 0X43
1059 #define IXGBE_PACKET_TYPE_NVGRE_IPV6 0X04
1060 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_TCP 0X14
1061 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_UDP 0X24
1062 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_SCTP 0X44
1063 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT 0X0C
1064 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_TCP 0X1C
1065 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_UDP 0X2C
1066 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_SCTP 0X4C
1067 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6 0X05
1068 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_TCP 0X15
1069 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_UDP 0X25
1070 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT 0X0D
1071 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_TCP 0X1D
1072 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_UDP 0X2D
1074 #define IXGBE_PACKET_TYPE_VXLAN 0X80
1075 #define IXGBE_PACKET_TYPE_VXLAN_IPV4 0X81
1076 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_TCP 0x91
1077 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_UDP 0xA1
1078 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_SCTP 0xC1
1079 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT 0x83
1080 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_TCP 0X93
1081 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_UDP 0XA3
1082 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_SCTP 0XC3
1083 #define IXGBE_PACKET_TYPE_VXLAN_IPV6 0X84
1084 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_TCP 0X94
1085 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_UDP 0XA4
1086 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_SCTP 0XC4
1087 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT 0X8C
1088 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_TCP 0X9C
1089 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_UDP 0XAC
1090 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_SCTP 0XCC
1091 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6 0X85
1092 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_TCP 0X95
1093 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_UDP 0XA5
1094 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT 0X8D
1095 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_TCP 0X9D
1096 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_UDP 0XAD
1099 * Use 2 different table for normal packet and tunnel packet
1100 * to save the space.
1103 ptype_table[IXGBE_PACKET_TYPE_MAX] __rte_cache_aligned = {
1104 [IXGBE_PACKET_TYPE_ETHER] = RTE_PTYPE_L2_ETHER,
1105 [IXGBE_PACKET_TYPE_IPV4] = RTE_PTYPE_L2_ETHER |
1107 [IXGBE_PACKET_TYPE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1108 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
1109 [IXGBE_PACKET_TYPE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1110 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
1111 [IXGBE_PACKET_TYPE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1112 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP,
1113 [IXGBE_PACKET_TYPE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1114 RTE_PTYPE_L3_IPV4_EXT,
1115 [IXGBE_PACKET_TYPE_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1116 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP,
1117 [IXGBE_PACKET_TYPE_IPV4_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1118 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP,
1119 [IXGBE_PACKET_TYPE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1120 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_SCTP,
1121 [IXGBE_PACKET_TYPE_IPV6] = RTE_PTYPE_L2_ETHER |
1123 [IXGBE_PACKET_TYPE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1124 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
1125 [IXGBE_PACKET_TYPE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1126 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
1127 [IXGBE_PACKET_TYPE_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1128 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP,
1129 [IXGBE_PACKET_TYPE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1130 RTE_PTYPE_L3_IPV6_EXT,
1131 [IXGBE_PACKET_TYPE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1132 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP,
1133 [IXGBE_PACKET_TYPE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1134 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP,
1135 [IXGBE_PACKET_TYPE_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1136 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_SCTP,
1137 [IXGBE_PACKET_TYPE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1138 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1139 RTE_PTYPE_INNER_L3_IPV6,
1140 [IXGBE_PACKET_TYPE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1141 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1142 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1143 [IXGBE_PACKET_TYPE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1144 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1145 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1146 [IXGBE_PACKET_TYPE_IPV4_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1147 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1148 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1149 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6] = RTE_PTYPE_L2_ETHER |
1150 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1151 RTE_PTYPE_INNER_L3_IPV6,
1152 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1153 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1154 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1155 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1156 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1157 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1158 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1159 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1160 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1161 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1162 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1163 RTE_PTYPE_INNER_L3_IPV6_EXT,
1164 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1165 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1166 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1167 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1168 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1169 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1170 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1171 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1172 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1173 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1174 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1175 RTE_PTYPE_INNER_L3_IPV6_EXT,
1176 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1177 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1178 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1179 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1180 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1181 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1182 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_SCTP] =
1183 RTE_PTYPE_L2_ETHER |
1184 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1185 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1189 ptype_table_tn[IXGBE_PACKET_TYPE_TN_MAX] __rte_cache_aligned = {
1190 [IXGBE_PACKET_TYPE_NVGRE] = RTE_PTYPE_L2_ETHER |
1191 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1192 RTE_PTYPE_INNER_L2_ETHER,
1193 [IXGBE_PACKET_TYPE_NVGRE_IPV4] = RTE_PTYPE_L2_ETHER |
1194 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1195 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1196 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1197 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1198 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT,
1199 [IXGBE_PACKET_TYPE_NVGRE_IPV6] = RTE_PTYPE_L2_ETHER |
1200 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1201 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6,
1202 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1203 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1204 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1205 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1206 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1207 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT,
1208 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1209 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1210 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1211 [IXGBE_PACKET_TYPE_NVGRE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1212 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1213 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1214 RTE_PTYPE_INNER_L4_TCP,
1215 [IXGBE_PACKET_TYPE_NVGRE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1216 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1217 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1218 RTE_PTYPE_INNER_L4_TCP,
1219 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1220 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1221 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1222 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1223 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1224 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1225 RTE_PTYPE_INNER_L4_TCP,
1226 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_TCP] =
1227 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1228 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_INNER_L2_ETHER |
1229 RTE_PTYPE_INNER_L3_IPV4,
1230 [IXGBE_PACKET_TYPE_NVGRE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1231 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1232 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1233 RTE_PTYPE_INNER_L4_UDP,
1234 [IXGBE_PACKET_TYPE_NVGRE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1235 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1236 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1237 RTE_PTYPE_INNER_L4_UDP,
1238 [IXGBE_PACKET_TYPE_NVGRE_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1239 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1240 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1241 RTE_PTYPE_INNER_L4_SCTP,
1242 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1243 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1244 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1245 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1246 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1247 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1248 RTE_PTYPE_INNER_L4_UDP,
1249 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1250 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1251 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1252 RTE_PTYPE_INNER_L4_SCTP,
1253 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_UDP] =
1254 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1255 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_INNER_L2_ETHER |
1256 RTE_PTYPE_INNER_L3_IPV4,
1257 [IXGBE_PACKET_TYPE_NVGRE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1258 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1259 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1260 RTE_PTYPE_INNER_L4_SCTP,
1261 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1262 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1263 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1264 RTE_PTYPE_INNER_L4_SCTP,
1265 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1266 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1267 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1268 RTE_PTYPE_INNER_L4_TCP,
1269 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1270 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1271 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1272 RTE_PTYPE_INNER_L4_UDP,
1274 [IXGBE_PACKET_TYPE_VXLAN] = RTE_PTYPE_L2_ETHER |
1275 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1276 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER,
1277 [IXGBE_PACKET_TYPE_VXLAN_IPV4] = RTE_PTYPE_L2_ETHER |
1278 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1279 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1280 RTE_PTYPE_INNER_L3_IPV4,
1281 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1282 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1283 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1284 RTE_PTYPE_INNER_L3_IPV4_EXT,
1285 [IXGBE_PACKET_TYPE_VXLAN_IPV6] = RTE_PTYPE_L2_ETHER |
1286 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1287 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1288 RTE_PTYPE_INNER_L3_IPV6,
1289 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1290 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1291 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1292 RTE_PTYPE_INNER_L3_IPV4,
1293 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1294 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1295 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1296 RTE_PTYPE_INNER_L3_IPV6_EXT,
1297 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1298 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1299 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1300 RTE_PTYPE_INNER_L3_IPV4,
1301 [IXGBE_PACKET_TYPE_VXLAN_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1302 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1303 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1304 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_TCP,
1305 [IXGBE_PACKET_TYPE_VXLAN_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1306 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1307 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1308 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1309 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1310 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1311 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1312 RTE_PTYPE_INNER_L3_IPV4,
1313 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1314 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1315 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1316 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1317 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_TCP] =
1318 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1319 RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN |
1320 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1321 [IXGBE_PACKET_TYPE_VXLAN_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1322 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1323 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1324 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_UDP,
1325 [IXGBE_PACKET_TYPE_VXLAN_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1326 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1327 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1328 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1329 [IXGBE_PACKET_TYPE_VXLAN_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1330 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1331 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1332 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1333 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1334 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1335 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1336 RTE_PTYPE_INNER_L3_IPV4,
1337 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1338 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1339 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1340 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1341 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1342 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1343 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1344 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1345 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_UDP] =
1346 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1347 RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN |
1348 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1349 [IXGBE_PACKET_TYPE_VXLAN_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1350 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1351 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1352 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_SCTP,
1353 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1354 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1355 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1356 RTE_PTYPE_INNER_L3_IPV4_EXT | RTE_PTYPE_INNER_L4_SCTP,
1357 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1358 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1359 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1360 RTE_PTYPE_INNER_L3_IPV4_EXT | RTE_PTYPE_INNER_L4_TCP,
1361 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1362 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1363 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1364 RTE_PTYPE_INNER_L3_IPV4_EXT | RTE_PTYPE_INNER_L4_UDP,
1367 /* @note: fix ixgbe_dev_supported_ptypes_get() if any change here. */
1368 static inline uint32_t
1369 ixgbe_rxd_pkt_info_to_pkt_type(uint32_t pkt_info, uint16_t ptype_mask)
1372 if (unlikely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
1373 return RTE_PTYPE_UNKNOWN;
1375 pkt_info = (pkt_info >> IXGBE_PACKET_TYPE_SHIFT) & ptype_mask;
1377 /* For tunnel packet */
1378 if (pkt_info & IXGBE_PACKET_TYPE_TUNNEL_BIT) {
1379 /* Remove the tunnel bit to save the space. */
1380 pkt_info &= IXGBE_PACKET_TYPE_MASK_TUNNEL;
1381 return ptype_table_tn[pkt_info];
1385 * For x550, if it's not tunnel,
1386 * tunnel type bit should be set to 0.
1387 * Reuse 82599's mask.
1389 pkt_info &= IXGBE_PACKET_TYPE_MASK_82599;
1391 return ptype_table[pkt_info];
1394 static inline uint64_t
1395 ixgbe_rxd_pkt_info_to_pkt_flags(uint16_t pkt_info)
1397 static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
1398 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
1399 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
1400 PKT_RX_RSS_HASH, 0, 0, 0,
1401 0, 0, 0, PKT_RX_FDIR,
1403 #ifdef RTE_LIBRTE_IEEE1588
1404 static uint64_t ip_pkt_etqf_map[8] = {
1405 0, 0, 0, PKT_RX_IEEE1588_PTP,
1409 if (likely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
1410 return ip_pkt_etqf_map[(pkt_info >> 4) & 0X07] |
1411 ip_rss_types_map[pkt_info & 0XF];
1413 return ip_rss_types_map[pkt_info & 0XF];
1415 return ip_rss_types_map[pkt_info & 0XF];
1419 static inline uint64_t
1420 rx_desc_status_to_pkt_flags(uint32_t rx_status, uint64_t vlan_flags)
1425 * Check if VLAN present only.
1426 * Do not check whether L3/L4 rx checksum done by NIC or not,
1427 * That can be found from rte_eth_rxmode.offloads flag
1429 pkt_flags = (rx_status & IXGBE_RXD_STAT_VP) ? vlan_flags : 0;
1431 #ifdef RTE_LIBRTE_IEEE1588
1432 if (rx_status & IXGBE_RXD_STAT_TMST)
1433 pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
1438 static inline uint64_t
1439 rx_desc_error_to_pkt_flags(uint32_t rx_status)
1444 * Bit 31: IPE, IPv4 checksum error
1445 * Bit 30: L4I, L4I integrity error
1447 static uint64_t error_to_pkt_flags_map[4] = {
1448 PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD,
1449 PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_BAD,
1450 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_GOOD,
1451 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1453 pkt_flags = error_to_pkt_flags_map[(rx_status >>
1454 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1456 if ((rx_status & IXGBE_RXD_STAT_OUTERIPCS) &&
1457 (rx_status & IXGBE_RXDADV_ERR_OUTERIPER)) {
1458 pkt_flags |= PKT_RX_EIP_CKSUM_BAD;
1461 #ifdef RTE_LIBRTE_SECURITY
1462 if (rx_status & IXGBE_RXD_STAT_SECP) {
1463 pkt_flags |= PKT_RX_SEC_OFFLOAD;
1464 if (rx_status & IXGBE_RXDADV_LNKSEC_ERROR_BAD_SIG)
1465 pkt_flags |= PKT_RX_SEC_OFFLOAD_FAILED;
1473 * LOOK_AHEAD defines how many desc statuses to check beyond the
1474 * current descriptor.
1475 * It must be a pound define for optimal performance.
1476 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1477 * function only works with LOOK_AHEAD=8.
1479 #define LOOK_AHEAD 8
1480 #if (LOOK_AHEAD != 8)
1481 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1484 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1486 volatile union ixgbe_adv_rx_desc *rxdp;
1487 struct ixgbe_rx_entry *rxep;
1488 struct rte_mbuf *mb;
1492 uint32_t s[LOOK_AHEAD];
1493 uint32_t pkt_info[LOOK_AHEAD];
1494 int i, j, nb_rx = 0;
1496 uint64_t vlan_flags = rxq->vlan_flags;
1498 /* get references to current descriptor and S/W ring entry */
1499 rxdp = &rxq->rx_ring[rxq->rx_tail];
1500 rxep = &rxq->sw_ring[rxq->rx_tail];
1502 status = rxdp->wb.upper.status_error;
1503 /* check to make sure there is at least 1 packet to receive */
1504 if (!(status & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1508 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1509 * reference packets that are ready to be received.
1511 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1512 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD) {
1513 /* Read desc statuses backwards to avoid race condition */
1514 for (j = 0; j < LOOK_AHEAD; j++)
1515 s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);
1519 /* Compute how many status bits were set */
1520 for (nb_dd = 0; nb_dd < LOOK_AHEAD &&
1521 (s[nb_dd] & IXGBE_RXDADV_STAT_DD); nb_dd++)
1524 for (j = 0; j < nb_dd; j++)
1525 pkt_info[j] = rte_le_to_cpu_32(rxdp[j].wb.lower.
1530 /* Translate descriptor info to mbuf format */
1531 for (j = 0; j < nb_dd; ++j) {
1533 pkt_len = rte_le_to_cpu_16(rxdp[j].wb.upper.length) -
1535 mb->data_len = pkt_len;
1536 mb->pkt_len = pkt_len;
1537 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1539 /* convert descriptor fields to rte mbuf flags */
1540 pkt_flags = rx_desc_status_to_pkt_flags(s[j],
1542 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1543 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags
1544 ((uint16_t)pkt_info[j]);
1545 mb->ol_flags = pkt_flags;
1547 ixgbe_rxd_pkt_info_to_pkt_type
1548 (pkt_info[j], rxq->pkt_type_mask);
1550 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1551 mb->hash.rss = rte_le_to_cpu_32(
1552 rxdp[j].wb.lower.hi_dword.rss);
1553 else if (pkt_flags & PKT_RX_FDIR) {
1554 mb->hash.fdir.hash = rte_le_to_cpu_16(
1555 rxdp[j].wb.lower.hi_dword.csum_ip.csum) &
1556 IXGBE_ATR_HASH_MASK;
1557 mb->hash.fdir.id = rte_le_to_cpu_16(
1558 rxdp[j].wb.lower.hi_dword.csum_ip.ip_id);
1562 /* Move mbuf pointers from the S/W ring to the stage */
1563 for (j = 0; j < LOOK_AHEAD; ++j) {
1564 rxq->rx_stage[i + j] = rxep[j].mbuf;
1567 /* stop if all requested packets could not be received */
1568 if (nb_dd != LOOK_AHEAD)
1572 /* clear software ring entries so we can cleanup correctly */
1573 for (i = 0; i < nb_rx; ++i) {
1574 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1582 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1584 volatile union ixgbe_adv_rx_desc *rxdp;
1585 struct ixgbe_rx_entry *rxep;
1586 struct rte_mbuf *mb;
1591 /* allocate buffers in bulk directly into the S/W ring */
1592 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1593 rxep = &rxq->sw_ring[alloc_idx];
1594 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1595 rxq->rx_free_thresh);
1596 if (unlikely(diag != 0))
1599 rxdp = &rxq->rx_ring[alloc_idx];
1600 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1601 /* populate the static rte mbuf fields */
1604 mb->port = rxq->port_id;
1607 rte_mbuf_refcnt_set(mb, 1);
1608 mb->data_off = RTE_PKTMBUF_HEADROOM;
1610 /* populate the descriptors */
1611 dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mb));
1612 rxdp[i].read.hdr_addr = 0;
1613 rxdp[i].read.pkt_addr = dma_addr;
1616 /* update state of internal queue structure */
1617 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1618 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1619 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1625 static inline uint16_t
1626 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1629 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1632 /* how many packets are ready to return? */
1633 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1635 /* copy mbuf pointers to the application's packet list */
1636 for (i = 0; i < nb_pkts; ++i)
1637 rx_pkts[i] = stage[i];
1639 /* update internal queue state */
1640 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1641 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1646 static inline uint16_t
1647 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1650 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1653 /* Any previously recv'd pkts will be returned from the Rx stage */
1654 if (rxq->rx_nb_avail)
1655 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1657 /* Scan the H/W ring for packets to receive */
1658 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1660 /* update internal queue state */
1661 rxq->rx_next_avail = 0;
1662 rxq->rx_nb_avail = nb_rx;
1663 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1665 /* if required, allocate new buffers to replenish descriptors */
1666 if (rxq->rx_tail > rxq->rx_free_trigger) {
1667 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1669 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1672 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1673 "queue_id=%u", (unsigned) rxq->port_id,
1674 (unsigned) rxq->queue_id);
1676 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1677 rxq->rx_free_thresh;
1680 * Need to rewind any previous receives if we cannot
1681 * allocate new buffers to replenish the old ones.
1683 rxq->rx_nb_avail = 0;
1684 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1685 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1686 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1691 /* update tail pointer */
1693 IXGBE_PCI_REG_WRITE_RELAXED(rxq->rdt_reg_addr,
1697 if (rxq->rx_tail >= rxq->nb_rx_desc)
1700 /* received any packets this loop? */
1701 if (rxq->rx_nb_avail)
1702 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1707 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1709 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1714 if (unlikely(nb_pkts == 0))
1717 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1718 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1720 /* request is relatively large, chunk it up */
1725 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1726 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1727 nb_rx = (uint16_t)(nb_rx + ret);
1728 nb_pkts = (uint16_t)(nb_pkts - ret);
1737 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1740 struct ixgbe_rx_queue *rxq;
1741 volatile union ixgbe_adv_rx_desc *rx_ring;
1742 volatile union ixgbe_adv_rx_desc *rxdp;
1743 struct ixgbe_rx_entry *sw_ring;
1744 struct ixgbe_rx_entry *rxe;
1745 struct rte_mbuf *rxm;
1746 struct rte_mbuf *nmb;
1747 union ixgbe_adv_rx_desc rxd;
1756 uint64_t vlan_flags;
1761 rx_id = rxq->rx_tail;
1762 rx_ring = rxq->rx_ring;
1763 sw_ring = rxq->sw_ring;
1764 vlan_flags = rxq->vlan_flags;
1765 while (nb_rx < nb_pkts) {
1767 * The order of operations here is important as the DD status
1768 * bit must not be read after any other descriptor fields.
1769 * rx_ring and rxdp are pointing to volatile data so the order
1770 * of accesses cannot be reordered by the compiler. If they were
1771 * not volatile, they could be reordered which could lead to
1772 * using invalid descriptor fields when read from rxd.
1774 rxdp = &rx_ring[rx_id];
1775 staterr = rxdp->wb.upper.status_error;
1776 if (!(staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1783 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1784 * is likely to be invalid and to be dropped by the various
1785 * validation checks performed by the network stack.
1787 * Allocate a new mbuf to replenish the RX ring descriptor.
1788 * If the allocation fails:
1789 * - arrange for that RX descriptor to be the first one
1790 * being parsed the next time the receive function is
1791 * invoked [on the same queue].
1793 * - Stop parsing the RX ring and return immediately.
1795 * This policy do not drop the packet received in the RX
1796 * descriptor for which the allocation of a new mbuf failed.
1797 * Thus, it allows that packet to be later retrieved if
1798 * mbuf have been freed in the mean time.
1799 * As a side effect, holding RX descriptors instead of
1800 * systematically giving them back to the NIC may lead to
1801 * RX ring exhaustion situations.
1802 * However, the NIC can gracefully prevent such situations
1803 * to happen by sending specific "back-pressure" flow control
1804 * frames to its peer(s).
1806 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1807 "ext_err_stat=0x%08x pkt_len=%u",
1808 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1809 (unsigned) rx_id, (unsigned) staterr,
1810 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1812 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
1814 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1815 "queue_id=%u", (unsigned) rxq->port_id,
1816 (unsigned) rxq->queue_id);
1817 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1822 rxe = &sw_ring[rx_id];
1824 if (rx_id == rxq->nb_rx_desc)
1827 /* Prefetch next mbuf while processing current one. */
1828 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1831 * When next RX descriptor is on a cache-line boundary,
1832 * prefetch the next 4 RX descriptors and the next 8 pointers
1835 if ((rx_id & 0x3) == 0) {
1836 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1837 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1843 rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
1844 rxdp->read.hdr_addr = 0;
1845 rxdp->read.pkt_addr = dma_addr;
1848 * Initialize the returned mbuf.
1849 * 1) setup generic mbuf fields:
1850 * - number of segments,
1853 * - RX port identifier.
1854 * 2) integrate hardware offload data, if any:
1855 * - RSS flag & hash,
1856 * - IP checksum flag,
1857 * - VLAN TCI, if any,
1860 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1862 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1863 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1866 rxm->pkt_len = pkt_len;
1867 rxm->data_len = pkt_len;
1868 rxm->port = rxq->port_id;
1870 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1871 /* Only valid if PKT_RX_VLAN set in pkt_flags */
1872 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1874 pkt_flags = rx_desc_status_to_pkt_flags(staterr, vlan_flags);
1875 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1876 pkt_flags = pkt_flags |
1877 ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1878 rxm->ol_flags = pkt_flags;
1880 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info,
1881 rxq->pkt_type_mask);
1883 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1884 rxm->hash.rss = rte_le_to_cpu_32(
1885 rxd.wb.lower.hi_dword.rss);
1886 else if (pkt_flags & PKT_RX_FDIR) {
1887 rxm->hash.fdir.hash = rte_le_to_cpu_16(
1888 rxd.wb.lower.hi_dword.csum_ip.csum) &
1889 IXGBE_ATR_HASH_MASK;
1890 rxm->hash.fdir.id = rte_le_to_cpu_16(
1891 rxd.wb.lower.hi_dword.csum_ip.ip_id);
1894 * Store the mbuf address into the next entry of the array
1895 * of returned packets.
1897 rx_pkts[nb_rx++] = rxm;
1899 rxq->rx_tail = rx_id;
1902 * If the number of free RX descriptors is greater than the RX free
1903 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1905 * Update the RDT with the value of the last processed RX descriptor
1906 * minus 1, to guarantee that the RDT register is never equal to the
1907 * RDH register, which creates a "full" ring situtation from the
1908 * hardware point of view...
1910 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1911 if (nb_hold > rxq->rx_free_thresh) {
1912 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1913 "nb_hold=%u nb_rx=%u",
1914 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1915 (unsigned) rx_id, (unsigned) nb_hold,
1917 rx_id = (uint16_t) ((rx_id == 0) ?
1918 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1919 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1922 rxq->nb_rx_hold = nb_hold;
1927 * Detect an RSC descriptor.
1929 static inline uint32_t
1930 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1932 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1933 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1937 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1939 * Fill the following info in the HEAD buffer of the Rx cluster:
1940 * - RX port identifier
1941 * - hardware offload data, if any:
1943 * - IP checksum flag
1944 * - VLAN TCI, if any
1946 * @head HEAD of the packet cluster
1947 * @desc HW descriptor to get data from
1948 * @rxq Pointer to the Rx queue
1951 ixgbe_fill_cluster_head_buf(
1952 struct rte_mbuf *head,
1953 union ixgbe_adv_rx_desc *desc,
1954 struct ixgbe_rx_queue *rxq,
1960 head->port = rxq->port_id;
1962 /* The vlan_tci field is only valid when PKT_RX_VLAN is
1963 * set in the pkt_flags field.
1965 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1966 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1967 pkt_flags = rx_desc_status_to_pkt_flags(staterr, rxq->vlan_flags);
1968 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1969 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1970 head->ol_flags = pkt_flags;
1972 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info, rxq->pkt_type_mask);
1974 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1975 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1976 else if (pkt_flags & PKT_RX_FDIR) {
1977 head->hash.fdir.hash =
1978 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1979 & IXGBE_ATR_HASH_MASK;
1980 head->hash.fdir.id =
1981 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1986 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1988 * @rx_queue Rx queue handle
1989 * @rx_pkts table of received packets
1990 * @nb_pkts size of rx_pkts table
1991 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1993 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1994 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1996 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1997 * 1) When non-EOP RSC completion arrives:
1998 * a) Update the HEAD of the current RSC aggregation cluster with the new
1999 * segment's data length.
2000 * b) Set the "next" pointer of the current segment to point to the segment
2001 * at the NEXTP index.
2002 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
2003 * in the sw_rsc_ring.
2004 * 2) When EOP arrives we just update the cluster's total length and offload
2005 * flags and deliver the cluster up to the upper layers. In our case - put it
2006 * in the rx_pkts table.
2008 * Returns the number of received packets/clusters (according to the "bulk
2009 * receive" interface).
2011 static inline uint16_t
2012 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
2015 struct ixgbe_rx_queue *rxq = rx_queue;
2016 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
2017 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
2018 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
2019 uint16_t rx_id = rxq->rx_tail;
2021 uint16_t nb_hold = rxq->nb_rx_hold;
2022 uint16_t prev_id = rxq->rx_tail;
2024 while (nb_rx < nb_pkts) {
2026 struct ixgbe_rx_entry *rxe;
2027 struct ixgbe_scattered_rx_entry *sc_entry;
2028 struct ixgbe_scattered_rx_entry *next_sc_entry;
2029 struct ixgbe_rx_entry *next_rxe = NULL;
2030 struct rte_mbuf *first_seg;
2031 struct rte_mbuf *rxm;
2032 struct rte_mbuf *nmb;
2033 union ixgbe_adv_rx_desc rxd;
2036 volatile union ixgbe_adv_rx_desc *rxdp;
2041 * The code in this whole file uses the volatile pointer to
2042 * ensure the read ordering of the status and the rest of the
2043 * descriptor fields (on the compiler level only!!!). This is so
2044 * UGLY - why not to just use the compiler barrier instead? DPDK
2045 * even has the rte_compiler_barrier() for that.
2047 * But most importantly this is just wrong because this doesn't
2048 * ensure memory ordering in a general case at all. For
2049 * instance, DPDK is supposed to work on Power CPUs where
2050 * compiler barrier may just not be enough!
2052 * I tried to write only this function properly to have a
2053 * starting point (as a part of an LRO/RSC series) but the
2054 * compiler cursed at me when I tried to cast away the
2055 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
2056 * keeping it the way it is for now.
2058 * The code in this file is broken in so many other places and
2059 * will just not work on a big endian CPU anyway therefore the
2060 * lines below will have to be revisited together with the rest
2064 * - Get rid of "volatile" and let the compiler do its job.
2065 * - Use the proper memory barrier (rte_rmb()) to ensure the
2066 * memory ordering below.
2068 rxdp = &rx_ring[rx_id];
2069 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
2071 if (!(staterr & IXGBE_RXDADV_STAT_DD))
2076 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
2077 "staterr=0x%x data_len=%u",
2078 rxq->port_id, rxq->queue_id, rx_id, staterr,
2079 rte_le_to_cpu_16(rxd.wb.upper.length));
2082 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
2084 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
2085 "port_id=%u queue_id=%u",
2086 rxq->port_id, rxq->queue_id);
2088 rte_eth_devices[rxq->port_id].data->
2089 rx_mbuf_alloc_failed++;
2092 } else if (nb_hold > rxq->rx_free_thresh) {
2093 uint16_t next_rdt = rxq->rx_free_trigger;
2095 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
2097 IXGBE_PCI_REG_WRITE_RELAXED(rxq->rdt_reg_addr,
2099 nb_hold -= rxq->rx_free_thresh;
2101 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
2102 "port_id=%u queue_id=%u",
2103 rxq->port_id, rxq->queue_id);
2105 rte_eth_devices[rxq->port_id].data->
2106 rx_mbuf_alloc_failed++;
2112 rxe = &sw_ring[rx_id];
2113 eop = staterr & IXGBE_RXDADV_STAT_EOP;
2115 next_id = rx_id + 1;
2116 if (next_id == rxq->nb_rx_desc)
2119 /* Prefetch next mbuf while processing current one. */
2120 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
2123 * When next RX descriptor is on a cache-line boundary,
2124 * prefetch the next 4 RX descriptors and the next 4 pointers
2127 if ((next_id & 0x3) == 0) {
2128 rte_ixgbe_prefetch(&rx_ring[next_id]);
2129 rte_ixgbe_prefetch(&sw_ring[next_id]);
2136 rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
2138 * Update RX descriptor with the physical address of the
2139 * new data buffer of the new allocated mbuf.
2143 rxm->data_off = RTE_PKTMBUF_HEADROOM;
2144 rxdp->read.hdr_addr = 0;
2145 rxdp->read.pkt_addr = dma;
2150 * Set data length & data buffer address of mbuf.
2152 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
2153 rxm->data_len = data_len;
2158 * Get next descriptor index:
2159 * - For RSC it's in the NEXTP field.
2160 * - For a scattered packet - it's just a following
2163 if (ixgbe_rsc_count(&rxd))
2165 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
2166 IXGBE_RXDADV_NEXTP_SHIFT;
2170 next_sc_entry = &sw_sc_ring[nextp_id];
2171 next_rxe = &sw_ring[nextp_id];
2172 rte_ixgbe_prefetch(next_rxe);
2175 sc_entry = &sw_sc_ring[rx_id];
2176 first_seg = sc_entry->fbuf;
2177 sc_entry->fbuf = NULL;
2180 * If this is the first buffer of the received packet,
2181 * set the pointer to the first mbuf of the packet and
2182 * initialize its context.
2183 * Otherwise, update the total length and the number of segments
2184 * of the current scattered packet, and update the pointer to
2185 * the last mbuf of the current packet.
2187 if (first_seg == NULL) {
2189 first_seg->pkt_len = data_len;
2190 first_seg->nb_segs = 1;
2192 first_seg->pkt_len += data_len;
2193 first_seg->nb_segs++;
2200 * If this is not the last buffer of the received packet, update
2201 * the pointer to the first mbuf at the NEXTP entry in the
2202 * sw_sc_ring and continue to parse the RX ring.
2204 if (!eop && next_rxe) {
2205 rxm->next = next_rxe->mbuf;
2206 next_sc_entry->fbuf = first_seg;
2210 /* Initialize the first mbuf of the returned packet */
2211 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq, staterr);
2214 * Deal with the case, when HW CRC srip is disabled.
2215 * That can't happen when LRO is enabled, but still could
2216 * happen for scattered RX mode.
2218 first_seg->pkt_len -= rxq->crc_len;
2219 if (unlikely(rxm->data_len <= rxq->crc_len)) {
2220 struct rte_mbuf *lp;
2222 for (lp = first_seg; lp->next != rxm; lp = lp->next)
2225 first_seg->nb_segs--;
2226 lp->data_len -= rxq->crc_len - rxm->data_len;
2228 rte_pktmbuf_free_seg(rxm);
2230 rxm->data_len -= rxq->crc_len;
2232 /* Prefetch data of first segment, if configured to do so. */
2233 rte_packet_prefetch((char *)first_seg->buf_addr +
2234 first_seg->data_off);
2237 * Store the mbuf address into the next entry of the array
2238 * of returned packets.
2240 rx_pkts[nb_rx++] = first_seg;
2244 * Record index of the next RX descriptor to probe.
2246 rxq->rx_tail = rx_id;
2249 * If the number of free RX descriptors is greater than the RX free
2250 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
2252 * Update the RDT with the value of the last processed RX descriptor
2253 * minus 1, to guarantee that the RDT register is never equal to the
2254 * RDH register, which creates a "full" ring situtation from the
2255 * hardware point of view...
2257 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
2258 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
2259 "nb_hold=%u nb_rx=%u",
2260 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
2263 IXGBE_PCI_REG_WRITE_RELAXED(rxq->rdt_reg_addr, prev_id);
2267 rxq->nb_rx_hold = nb_hold;
2272 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2275 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
2279 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2282 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
2285 /*********************************************************************
2287 * Queue management functions
2289 **********************************************************************/
2291 static void __attribute__((cold))
2292 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
2296 if (txq->sw_ring != NULL) {
2297 for (i = 0; i < txq->nb_tx_desc; i++) {
2298 if (txq->sw_ring[i].mbuf != NULL) {
2299 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2300 txq->sw_ring[i].mbuf = NULL;
2306 static void __attribute__((cold))
2307 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
2310 txq->sw_ring != NULL)
2311 rte_free(txq->sw_ring);
2314 static void __attribute__((cold))
2315 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
2317 if (txq != NULL && txq->ops != NULL) {
2318 txq->ops->release_mbufs(txq);
2319 txq->ops->free_swring(txq);
2324 void __attribute__((cold))
2325 ixgbe_dev_tx_queue_release(void *txq)
2327 ixgbe_tx_queue_release(txq);
2330 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
2331 static void __attribute__((cold))
2332 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
2334 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
2335 struct ixgbe_tx_entry *txe = txq->sw_ring;
2338 /* Zero out HW ring memory */
2339 for (i = 0; i < txq->nb_tx_desc; i++) {
2340 txq->tx_ring[i] = zeroed_desc;
2343 /* Initialize SW ring entries */
2344 prev = (uint16_t) (txq->nb_tx_desc - 1);
2345 for (i = 0; i < txq->nb_tx_desc; i++) {
2346 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
2348 txd->wb.status = rte_cpu_to_le_32(IXGBE_TXD_STAT_DD);
2351 txe[prev].next_id = i;
2355 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2356 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2359 txq->nb_tx_used = 0;
2361 * Always allow 1 descriptor to be un-allocated to avoid
2362 * a H/W race condition
2364 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2365 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2367 memset((void *)&txq->ctx_cache, 0,
2368 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2371 static const struct ixgbe_txq_ops def_txq_ops = {
2372 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2373 .free_swring = ixgbe_tx_free_swring,
2374 .reset = ixgbe_reset_tx_queue,
2377 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2378 * the queue parameters. Used in tx_queue_setup by primary process and then
2379 * in dev_init by secondary process when attaching to an existing ethdev.
2381 void __attribute__((cold))
2382 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2384 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2385 if ((txq->offloads == 0) &&
2386 #ifdef RTE_LIBRTE_SECURITY
2387 !(txq->using_ipsec) &&
2389 (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2390 PMD_INIT_LOG(DEBUG, "Using simple tx code path");
2391 dev->tx_pkt_prepare = NULL;
2392 #ifdef RTE_IXGBE_INC_VECTOR
2393 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2394 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2395 ixgbe_txq_vec_setup(txq) == 0)) {
2396 PMD_INIT_LOG(DEBUG, "Vector tx enabled.");
2397 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2400 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2402 PMD_INIT_LOG(DEBUG, "Using full-featured tx code path");
2404 " - offloads = 0x%" PRIx64,
2407 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2408 (unsigned long)txq->tx_rs_thresh,
2409 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2410 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2411 dev->tx_pkt_prepare = ixgbe_prep_pkts;
2416 ixgbe_get_tx_queue_offloads(struct rte_eth_dev *dev)
2424 ixgbe_get_tx_port_offloads(struct rte_eth_dev *dev)
2426 uint64_t tx_offload_capa;
2427 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2430 DEV_TX_OFFLOAD_VLAN_INSERT |
2431 DEV_TX_OFFLOAD_IPV4_CKSUM |
2432 DEV_TX_OFFLOAD_UDP_CKSUM |
2433 DEV_TX_OFFLOAD_TCP_CKSUM |
2434 DEV_TX_OFFLOAD_SCTP_CKSUM |
2435 DEV_TX_OFFLOAD_TCP_TSO |
2436 DEV_TX_OFFLOAD_MULTI_SEGS;
2438 if (hw->mac.type == ixgbe_mac_82599EB ||
2439 hw->mac.type == ixgbe_mac_X540)
2440 tx_offload_capa |= DEV_TX_OFFLOAD_MACSEC_INSERT;
2442 if (hw->mac.type == ixgbe_mac_X550 ||
2443 hw->mac.type == ixgbe_mac_X550EM_x ||
2444 hw->mac.type == ixgbe_mac_X550EM_a)
2445 tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
2447 #ifdef RTE_LIBRTE_SECURITY
2448 if (dev->security_ctx)
2449 tx_offload_capa |= DEV_TX_OFFLOAD_SECURITY;
2451 return tx_offload_capa;
2454 int __attribute__((cold))
2455 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2458 unsigned int socket_id,
2459 const struct rte_eth_txconf *tx_conf)
2461 const struct rte_memzone *tz;
2462 struct ixgbe_tx_queue *txq;
2463 struct ixgbe_hw *hw;
2464 uint16_t tx_rs_thresh, tx_free_thresh;
2467 PMD_INIT_FUNC_TRACE();
2468 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2470 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
2473 * Validate number of transmit descriptors.
2474 * It must not exceed hardware maximum, and must be multiple
2477 if (nb_desc % IXGBE_TXD_ALIGN != 0 ||
2478 (nb_desc > IXGBE_MAX_RING_DESC) ||
2479 (nb_desc < IXGBE_MIN_RING_DESC)) {
2484 * The following two parameters control the setting of the RS bit on
2485 * transmit descriptors.
2486 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2487 * descriptors have been used.
2488 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2489 * descriptors are used or if the number of descriptors required
2490 * to transmit a packet is greater than the number of free TX
2492 * The following constraints must be satisfied:
2493 * tx_rs_thresh must be greater than 0.
2494 * tx_rs_thresh must be less than the size of the ring minus 2.
2495 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2496 * tx_rs_thresh must be a divisor of the ring size.
2497 * tx_free_thresh must be greater than 0.
2498 * tx_free_thresh must be less than the size of the ring minus 3.
2499 * One descriptor in the TX ring is used as a sentinel to avoid a
2500 * H/W race condition, hence the maximum threshold constraints.
2501 * When set to zero use default values.
2503 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2504 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2505 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2506 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2507 if (tx_rs_thresh >= (nb_desc - 2)) {
2508 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2509 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2510 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2511 (int)dev->data->port_id, (int)queue_idx);
2514 if (tx_rs_thresh > DEFAULT_TX_RS_THRESH) {
2515 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less or equal than %u. "
2516 "(tx_rs_thresh=%u port=%d queue=%d)",
2517 DEFAULT_TX_RS_THRESH, (unsigned int)tx_rs_thresh,
2518 (int)dev->data->port_id, (int)queue_idx);
2521 if (tx_free_thresh >= (nb_desc - 3)) {
2522 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2523 "tx_free_thresh must be less than the number of "
2524 "TX descriptors minus 3. (tx_free_thresh=%u "
2525 "port=%d queue=%d)",
2526 (unsigned int)tx_free_thresh,
2527 (int)dev->data->port_id, (int)queue_idx);
2530 if (tx_rs_thresh > tx_free_thresh) {
2531 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2532 "tx_free_thresh. (tx_free_thresh=%u "
2533 "tx_rs_thresh=%u port=%d queue=%d)",
2534 (unsigned int)tx_free_thresh,
2535 (unsigned int)tx_rs_thresh,
2536 (int)dev->data->port_id,
2540 if ((nb_desc % tx_rs_thresh) != 0) {
2541 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2542 "number of TX descriptors. (tx_rs_thresh=%u "
2543 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2544 (int)dev->data->port_id, (int)queue_idx);
2549 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2550 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2551 * by the NIC and all descriptors are written back after the NIC
2552 * accumulates WTHRESH descriptors.
2554 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2555 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2556 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2557 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2558 (int)dev->data->port_id, (int)queue_idx);
2562 /* Free memory prior to re-allocation if needed... */
2563 if (dev->data->tx_queues[queue_idx] != NULL) {
2564 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2565 dev->data->tx_queues[queue_idx] = NULL;
2568 /* First allocate the tx queue data structure */
2569 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2570 RTE_CACHE_LINE_SIZE, socket_id);
2575 * Allocate TX ring hardware descriptors. A memzone large enough to
2576 * handle the maximum ring size is allocated in order to allow for
2577 * resizing in later calls to the queue setup function.
2579 tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
2580 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2581 IXGBE_ALIGN, socket_id);
2583 ixgbe_tx_queue_release(txq);
2587 txq->nb_tx_desc = nb_desc;
2588 txq->tx_rs_thresh = tx_rs_thresh;
2589 txq->tx_free_thresh = tx_free_thresh;
2590 txq->pthresh = tx_conf->tx_thresh.pthresh;
2591 txq->hthresh = tx_conf->tx_thresh.hthresh;
2592 txq->wthresh = tx_conf->tx_thresh.wthresh;
2593 txq->queue_id = queue_idx;
2594 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2595 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2596 txq->port_id = dev->data->port_id;
2597 txq->offloads = offloads;
2598 txq->ops = &def_txq_ops;
2599 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2600 #ifdef RTE_LIBRTE_SECURITY
2601 txq->using_ipsec = !!(dev->data->dev_conf.txmode.offloads &
2602 DEV_TX_OFFLOAD_SECURITY);
2606 * Modification to set VFTDT for virtual function if vf is detected
2608 if (hw->mac.type == ixgbe_mac_82599_vf ||
2609 hw->mac.type == ixgbe_mac_X540_vf ||
2610 hw->mac.type == ixgbe_mac_X550_vf ||
2611 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2612 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2613 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2615 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2617 txq->tx_ring_phys_addr = tz->iova;
2618 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2620 /* Allocate software ring */
2621 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2622 sizeof(struct ixgbe_tx_entry) * nb_desc,
2623 RTE_CACHE_LINE_SIZE, socket_id);
2624 if (txq->sw_ring == NULL) {
2625 ixgbe_tx_queue_release(txq);
2628 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2629 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2631 /* set up vector or scalar TX function as appropriate */
2632 ixgbe_set_tx_function(dev, txq);
2634 txq->ops->reset(txq);
2636 dev->data->tx_queues[queue_idx] = txq;
2643 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2645 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2646 * in the sw_rsc_ring is not set to NULL but rather points to the next
2647 * mbuf of this RSC aggregation (that has not been completed yet and still
2648 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2649 * will just free first "nb_segs" segments of the cluster explicitly by calling
2650 * an rte_pktmbuf_free_seg().
2652 * @m scattered cluster head
2654 static void __attribute__((cold))
2655 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2657 uint16_t i, nb_segs = m->nb_segs;
2658 struct rte_mbuf *next_seg;
2660 for (i = 0; i < nb_segs; i++) {
2662 rte_pktmbuf_free_seg(m);
2667 static void __attribute__((cold))
2668 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2672 #ifdef RTE_IXGBE_INC_VECTOR
2673 /* SSE Vector driver has a different way of releasing mbufs. */
2674 if (rxq->rx_using_sse) {
2675 ixgbe_rx_queue_release_mbufs_vec(rxq);
2680 if (rxq->sw_ring != NULL) {
2681 for (i = 0; i < rxq->nb_rx_desc; i++) {
2682 if (rxq->sw_ring[i].mbuf != NULL) {
2683 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2684 rxq->sw_ring[i].mbuf = NULL;
2687 if (rxq->rx_nb_avail) {
2688 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2689 struct rte_mbuf *mb;
2691 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2692 rte_pktmbuf_free_seg(mb);
2694 rxq->rx_nb_avail = 0;
2698 if (rxq->sw_sc_ring)
2699 for (i = 0; i < rxq->nb_rx_desc; i++)
2700 if (rxq->sw_sc_ring[i].fbuf) {
2701 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2702 rxq->sw_sc_ring[i].fbuf = NULL;
2706 static void __attribute__((cold))
2707 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2710 ixgbe_rx_queue_release_mbufs(rxq);
2711 rte_free(rxq->sw_ring);
2712 rte_free(rxq->sw_sc_ring);
2717 void __attribute__((cold))
2718 ixgbe_dev_rx_queue_release(void *rxq)
2720 ixgbe_rx_queue_release(rxq);
2724 * Check if Rx Burst Bulk Alloc function can be used.
2726 * 0: the preconditions are satisfied and the bulk allocation function
2728 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2729 * function must be used.
2731 static inline int __attribute__((cold))
2732 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2737 * Make sure the following pre-conditions are satisfied:
2738 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2739 * rxq->rx_free_thresh < rxq->nb_rx_desc
2740 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2741 * Scattered packets are not supported. This should be checked
2742 * outside of this function.
2744 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2745 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2746 "rxq->rx_free_thresh=%d, "
2747 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2748 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2750 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2751 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2752 "rxq->rx_free_thresh=%d, "
2753 "rxq->nb_rx_desc=%d",
2754 rxq->rx_free_thresh, rxq->nb_rx_desc);
2756 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2757 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2758 "rxq->nb_rx_desc=%d, "
2759 "rxq->rx_free_thresh=%d",
2760 rxq->nb_rx_desc, rxq->rx_free_thresh);
2767 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2768 static void __attribute__((cold))
2769 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2771 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2773 uint16_t len = rxq->nb_rx_desc;
2776 * By default, the Rx queue setup function allocates enough memory for
2777 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2778 * extra memory at the end of the descriptor ring to be zero'd out.
2780 if (adapter->rx_bulk_alloc_allowed)
2781 /* zero out extra memory */
2782 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2785 * Zero out HW ring memory. Zero out extra memory at the end of
2786 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2787 * reads extra memory as zeros.
2789 for (i = 0; i < len; i++) {
2790 rxq->rx_ring[i] = zeroed_desc;
2794 * initialize extra software ring entries. Space for these extra
2795 * entries is always allocated
2797 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2798 for (i = rxq->nb_rx_desc; i < len; ++i) {
2799 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2802 rxq->rx_nb_avail = 0;
2803 rxq->rx_next_avail = 0;
2804 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2806 rxq->nb_rx_hold = 0;
2807 rxq->pkt_first_seg = NULL;
2808 rxq->pkt_last_seg = NULL;
2810 #ifdef RTE_IXGBE_INC_VECTOR
2811 rxq->rxrearm_start = 0;
2812 rxq->rxrearm_nb = 0;
2817 ixgbe_is_vf(struct rte_eth_dev *dev)
2819 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2821 switch (hw->mac.type) {
2822 case ixgbe_mac_82599_vf:
2823 case ixgbe_mac_X540_vf:
2824 case ixgbe_mac_X550_vf:
2825 case ixgbe_mac_X550EM_x_vf:
2826 case ixgbe_mac_X550EM_a_vf:
2834 ixgbe_get_rx_queue_offloads(struct rte_eth_dev *dev)
2836 uint64_t offloads = 0;
2837 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2839 if (hw->mac.type != ixgbe_mac_82598EB)
2840 offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
2846 ixgbe_get_rx_port_offloads(struct rte_eth_dev *dev)
2849 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2851 offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
2852 DEV_RX_OFFLOAD_UDP_CKSUM |
2853 DEV_RX_OFFLOAD_TCP_CKSUM |
2854 DEV_RX_OFFLOAD_KEEP_CRC |
2855 DEV_RX_OFFLOAD_JUMBO_FRAME |
2856 DEV_RX_OFFLOAD_SCATTER;
2858 if (hw->mac.type == ixgbe_mac_82598EB)
2859 offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
2861 if (ixgbe_is_vf(dev) == 0)
2862 offloads |= (DEV_RX_OFFLOAD_VLAN_FILTER |
2863 DEV_RX_OFFLOAD_VLAN_EXTEND);
2866 * RSC is only supported by 82599 and x540 PF devices in a non-SR-IOV
2869 if ((hw->mac.type == ixgbe_mac_82599EB ||
2870 hw->mac.type == ixgbe_mac_X540 ||
2871 hw->mac.type == ixgbe_mac_X550) &&
2872 !RTE_ETH_DEV_SRIOV(dev).active)
2873 offloads |= DEV_RX_OFFLOAD_TCP_LRO;
2875 if (hw->mac.type == ixgbe_mac_82599EB ||
2876 hw->mac.type == ixgbe_mac_X540)
2877 offloads |= DEV_RX_OFFLOAD_MACSEC_STRIP;
2879 if (hw->mac.type == ixgbe_mac_X550 ||
2880 hw->mac.type == ixgbe_mac_X550EM_x ||
2881 hw->mac.type == ixgbe_mac_X550EM_a)
2882 offloads |= DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM;
2884 #ifdef RTE_LIBRTE_SECURITY
2885 if (dev->security_ctx)
2886 offloads |= DEV_RX_OFFLOAD_SECURITY;
2892 int __attribute__((cold))
2893 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2896 unsigned int socket_id,
2897 const struct rte_eth_rxconf *rx_conf,
2898 struct rte_mempool *mp)
2900 const struct rte_memzone *rz;
2901 struct ixgbe_rx_queue *rxq;
2902 struct ixgbe_hw *hw;
2904 struct ixgbe_adapter *adapter =
2905 (struct ixgbe_adapter *)dev->data->dev_private;
2908 PMD_INIT_FUNC_TRACE();
2909 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2911 offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
2914 * Validate number of receive descriptors.
2915 * It must not exceed hardware maximum, and must be multiple
2918 if (nb_desc % IXGBE_RXD_ALIGN != 0 ||
2919 (nb_desc > IXGBE_MAX_RING_DESC) ||
2920 (nb_desc < IXGBE_MIN_RING_DESC)) {
2924 /* Free memory prior to re-allocation if needed... */
2925 if (dev->data->rx_queues[queue_idx] != NULL) {
2926 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2927 dev->data->rx_queues[queue_idx] = NULL;
2930 /* First allocate the rx queue data structure */
2931 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2932 RTE_CACHE_LINE_SIZE, socket_id);
2936 rxq->nb_rx_desc = nb_desc;
2937 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2938 rxq->queue_id = queue_idx;
2939 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2940 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2941 rxq->port_id = dev->data->port_id;
2942 if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
2943 rxq->crc_len = ETHER_CRC_LEN;
2946 rxq->drop_en = rx_conf->rx_drop_en;
2947 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2948 rxq->offloads = offloads;
2951 * The packet type in RX descriptor is different for different NICs.
2952 * Some bits are used for x550 but reserved for other NICS.
2953 * So set different masks for different NICs.
2955 if (hw->mac.type == ixgbe_mac_X550 ||
2956 hw->mac.type == ixgbe_mac_X550EM_x ||
2957 hw->mac.type == ixgbe_mac_X550EM_a ||
2958 hw->mac.type == ixgbe_mac_X550_vf ||
2959 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2960 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2961 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_X550;
2963 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_82599;
2966 * Allocate RX ring hardware descriptors. A memzone large enough to
2967 * handle the maximum ring size is allocated in order to allow for
2968 * resizing in later calls to the queue setup function.
2970 rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
2971 RX_RING_SZ, IXGBE_ALIGN, socket_id);
2973 ixgbe_rx_queue_release(rxq);
2978 * Zero init all the descriptors in the ring.
2980 memset(rz->addr, 0, RX_RING_SZ);
2983 * Modified to setup VFRDT for Virtual Function
2985 if (hw->mac.type == ixgbe_mac_82599_vf ||
2986 hw->mac.type == ixgbe_mac_X540_vf ||
2987 hw->mac.type == ixgbe_mac_X550_vf ||
2988 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2989 hw->mac.type == ixgbe_mac_X550EM_a_vf) {
2991 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2993 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2996 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2998 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
3001 rxq->rx_ring_phys_addr = rz->iova;
3002 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
3005 * Certain constraints must be met in order to use the bulk buffer
3006 * allocation Rx burst function. If any of Rx queues doesn't meet them
3007 * the feature should be disabled for the whole port.
3009 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
3010 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
3011 "preconditions - canceling the feature for "
3012 "the whole port[%d]",
3013 rxq->queue_id, rxq->port_id);
3014 adapter->rx_bulk_alloc_allowed = false;
3018 * Allocate software ring. Allow for space at the end of the
3019 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
3020 * function does not access an invalid memory region.
3023 if (adapter->rx_bulk_alloc_allowed)
3024 len += RTE_PMD_IXGBE_RX_MAX_BURST;
3026 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
3027 sizeof(struct ixgbe_rx_entry) * len,
3028 RTE_CACHE_LINE_SIZE, socket_id);
3029 if (!rxq->sw_ring) {
3030 ixgbe_rx_queue_release(rxq);
3035 * Always allocate even if it's not going to be needed in order to
3036 * simplify the code.
3038 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
3039 * be requested in ixgbe_dev_rx_init(), which is called later from
3043 rte_zmalloc_socket("rxq->sw_sc_ring",
3044 sizeof(struct ixgbe_scattered_rx_entry) * len,
3045 RTE_CACHE_LINE_SIZE, socket_id);
3046 if (!rxq->sw_sc_ring) {
3047 ixgbe_rx_queue_release(rxq);
3051 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
3052 "dma_addr=0x%"PRIx64,
3053 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
3054 rxq->rx_ring_phys_addr);
3056 if (!rte_is_power_of_2(nb_desc)) {
3057 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
3058 "preconditions - canceling the feature for "
3059 "the whole port[%d]",
3060 rxq->queue_id, rxq->port_id);
3061 adapter->rx_vec_allowed = false;
3063 ixgbe_rxq_vec_setup(rxq);
3065 dev->data->rx_queues[queue_idx] = rxq;
3067 ixgbe_reset_rx_queue(adapter, rxq);
3073 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
3075 #define IXGBE_RXQ_SCAN_INTERVAL 4
3076 volatile union ixgbe_adv_rx_desc *rxdp;
3077 struct ixgbe_rx_queue *rxq;
3080 rxq = dev->data->rx_queues[rx_queue_id];
3081 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
3083 while ((desc < rxq->nb_rx_desc) &&
3084 (rxdp->wb.upper.status_error &
3085 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD))) {
3086 desc += IXGBE_RXQ_SCAN_INTERVAL;
3087 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
3088 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
3089 rxdp = &(rxq->rx_ring[rxq->rx_tail +
3090 desc - rxq->nb_rx_desc]);
3097 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
3099 volatile union ixgbe_adv_rx_desc *rxdp;
3100 struct ixgbe_rx_queue *rxq = rx_queue;
3103 if (unlikely(offset >= rxq->nb_rx_desc))
3105 desc = rxq->rx_tail + offset;
3106 if (desc >= rxq->nb_rx_desc)
3107 desc -= rxq->nb_rx_desc;
3109 rxdp = &rxq->rx_ring[desc];
3110 return !!(rxdp->wb.upper.status_error &
3111 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD));
3115 ixgbe_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
3117 struct ixgbe_rx_queue *rxq = rx_queue;
3118 volatile uint32_t *status;
3119 uint32_t nb_hold, desc;
3121 if (unlikely(offset >= rxq->nb_rx_desc))
3124 #ifdef RTE_IXGBE_INC_VECTOR
3125 if (rxq->rx_using_sse)
3126 nb_hold = rxq->rxrearm_nb;
3129 nb_hold = rxq->nb_rx_hold;
3130 if (offset >= rxq->nb_rx_desc - nb_hold)
3131 return RTE_ETH_RX_DESC_UNAVAIL;
3133 desc = rxq->rx_tail + offset;
3134 if (desc >= rxq->nb_rx_desc)
3135 desc -= rxq->nb_rx_desc;
3137 status = &rxq->rx_ring[desc].wb.upper.status_error;
3138 if (*status & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD))
3139 return RTE_ETH_RX_DESC_DONE;
3141 return RTE_ETH_RX_DESC_AVAIL;
3145 ixgbe_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
3147 struct ixgbe_tx_queue *txq = tx_queue;
3148 volatile uint32_t *status;
3151 if (unlikely(offset >= txq->nb_tx_desc))
3154 desc = txq->tx_tail + offset;
3155 /* go to next desc that has the RS bit */
3156 desc = ((desc + txq->tx_rs_thresh - 1) / txq->tx_rs_thresh) *
3158 if (desc >= txq->nb_tx_desc) {
3159 desc -= txq->nb_tx_desc;
3160 if (desc >= txq->nb_tx_desc)
3161 desc -= txq->nb_tx_desc;
3164 status = &txq->tx_ring[desc].wb.status;
3165 if (*status & rte_cpu_to_le_32(IXGBE_ADVTXD_STAT_DD))
3166 return RTE_ETH_TX_DESC_DONE;
3168 return RTE_ETH_TX_DESC_FULL;
3171 void __attribute__((cold))
3172 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
3175 struct ixgbe_adapter *adapter =
3176 (struct ixgbe_adapter *)dev->data->dev_private;
3178 PMD_INIT_FUNC_TRACE();
3180 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3181 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
3184 txq->ops->release_mbufs(txq);
3185 txq->ops->reset(txq);
3189 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3190 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
3193 ixgbe_rx_queue_release_mbufs(rxq);
3194 ixgbe_reset_rx_queue(adapter, rxq);
3200 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
3204 PMD_INIT_FUNC_TRACE();
3206 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3207 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
3208 dev->data->rx_queues[i] = NULL;
3210 dev->data->nb_rx_queues = 0;
3212 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3213 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
3214 dev->data->tx_queues[i] = NULL;
3216 dev->data->nb_tx_queues = 0;
3219 /*********************************************************************
3221 * Device RX/TX init functions
3223 **********************************************************************/
3226 * Receive Side Scaling (RSS)
3227 * See section 7.1.2.8 in the following document:
3228 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
3231 * The source and destination IP addresses of the IP header and the source
3232 * and destination ports of TCP/UDP headers, if any, of received packets are
3233 * hashed against a configurable random key to compute a 32-bit RSS hash result.
3234 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
3235 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
3236 * RSS output index which is used as the RX queue index where to store the
3238 * The following output is supplied in the RX write-back descriptor:
3239 * - 32-bit result of the Microsoft RSS hash function,
3240 * - 4-bit RSS type field.
3244 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
3245 * Used as the default key.
3247 static uint8_t rss_intel_key[40] = {
3248 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
3249 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
3250 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3251 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
3252 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
3256 ixgbe_rss_disable(struct rte_eth_dev *dev)
3258 struct ixgbe_hw *hw;
3262 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3263 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3264 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3265 mrqc &= ~IXGBE_MRQC_RSSEN;
3266 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
3270 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
3280 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3281 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3283 hash_key = rss_conf->rss_key;
3284 if (hash_key != NULL) {
3285 /* Fill in RSS hash key */
3286 for (i = 0; i < 10; i++) {
3287 rss_key = hash_key[(i * 4)];
3288 rss_key |= hash_key[(i * 4) + 1] << 8;
3289 rss_key |= hash_key[(i * 4) + 2] << 16;
3290 rss_key |= hash_key[(i * 4) + 3] << 24;
3291 IXGBE_WRITE_REG_ARRAY(hw, rssrk_reg, i, rss_key);
3295 /* Set configured hashing protocols in MRQC register */
3296 rss_hf = rss_conf->rss_hf;
3297 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
3298 if (rss_hf & ETH_RSS_IPV4)
3299 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
3300 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
3301 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
3302 if (rss_hf & ETH_RSS_IPV6)
3303 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
3304 if (rss_hf & ETH_RSS_IPV6_EX)
3305 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
3306 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
3307 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3308 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
3309 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
3310 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
3311 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3312 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
3313 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3314 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
3315 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
3316 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
3320 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
3321 struct rte_eth_rss_conf *rss_conf)
3323 struct ixgbe_hw *hw;
3328 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3330 if (!ixgbe_rss_update_sp(hw->mac.type)) {
3331 PMD_DRV_LOG(ERR, "RSS hash update is not supported on this "
3335 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3338 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
3339 * "RSS enabling cannot be done dynamically while it must be
3340 * preceded by a software reset"
3341 * Before changing anything, first check that the update RSS operation
3342 * does not attempt to disable RSS, if RSS was enabled at
3343 * initialization time, or does not attempt to enable RSS, if RSS was
3344 * disabled at initialization time.
3346 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
3347 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3348 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
3349 if (rss_hf != 0) /* Enable RSS */
3351 return 0; /* Nothing to do */
3354 if (rss_hf == 0) /* Disable RSS */
3356 ixgbe_hw_rss_hash_set(hw, rss_conf);
3361 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
3362 struct rte_eth_rss_conf *rss_conf)
3364 struct ixgbe_hw *hw;
3373 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3374 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3375 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3376 hash_key = rss_conf->rss_key;
3377 if (hash_key != NULL) {
3378 /* Return RSS hash key */
3379 for (i = 0; i < 10; i++) {
3380 rss_key = IXGBE_READ_REG_ARRAY(hw, rssrk_reg, i);
3381 hash_key[(i * 4)] = rss_key & 0x000000FF;
3382 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
3383 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
3384 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
3388 /* Get RSS functions configured in MRQC register */
3389 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3390 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
3391 rss_conf->rss_hf = 0;
3395 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
3396 rss_hf |= ETH_RSS_IPV4;
3397 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
3398 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
3399 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
3400 rss_hf |= ETH_RSS_IPV6;
3401 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
3402 rss_hf |= ETH_RSS_IPV6_EX;
3403 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
3404 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
3405 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
3406 rss_hf |= ETH_RSS_IPV6_TCP_EX;
3407 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
3408 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
3409 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
3410 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
3411 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
3412 rss_hf |= ETH_RSS_IPV6_UDP_EX;
3413 rss_conf->rss_hf = rss_hf;
3418 ixgbe_rss_configure(struct rte_eth_dev *dev)
3420 struct rte_eth_rss_conf rss_conf;
3421 struct ixgbe_hw *hw;
3425 uint16_t sp_reta_size;
3428 PMD_INIT_FUNC_TRACE();
3429 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3431 sp_reta_size = ixgbe_reta_size_get(hw->mac.type);
3434 * Fill in redirection table
3435 * The byte-swap is needed because NIC registers are in
3436 * little-endian order.
3439 for (i = 0, j = 0; i < sp_reta_size; i++, j++) {
3440 reta_reg = ixgbe_reta_reg_get(hw->mac.type, i);
3442 if (j == dev->data->nb_rx_queues)
3444 reta = (reta << 8) | j;
3446 IXGBE_WRITE_REG(hw, reta_reg,
3451 * Configure the RSS key and the RSS protocols used to compute
3452 * the RSS hash of input packets.
3454 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
3455 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
3456 ixgbe_rss_disable(dev);
3459 if (rss_conf.rss_key == NULL)
3460 rss_conf.rss_key = rss_intel_key; /* Default hash key */
3461 ixgbe_hw_rss_hash_set(hw, &rss_conf);
3464 #define NUM_VFTA_REGISTERS 128
3465 #define NIC_RX_BUFFER_SIZE 0x200
3466 #define X550_RX_BUFFER_SIZE 0x180
3469 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
3471 struct rte_eth_vmdq_dcb_conf *cfg;
3472 struct ixgbe_hw *hw;
3473 enum rte_eth_nb_pools num_pools;
3474 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
3476 uint8_t nb_tcs; /* number of traffic classes */
3479 PMD_INIT_FUNC_TRACE();
3480 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3481 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3482 num_pools = cfg->nb_queue_pools;
3483 /* Check we have a valid number of pools */
3484 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
3485 ixgbe_rss_disable(dev);
3488 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
3489 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
3493 * split rx buffer up into sections, each for 1 traffic class
3495 switch (hw->mac.type) {
3496 case ixgbe_mac_X550:
3497 case ixgbe_mac_X550EM_x:
3498 case ixgbe_mac_X550EM_a:
3499 pbsize = (uint16_t)(X550_RX_BUFFER_SIZE / nb_tcs);
3502 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3505 for (i = 0; i < nb_tcs; i++) {
3506 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3508 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3509 /* clear 10 bits. */
3510 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
3511 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3513 /* zero alloc all unused TCs */
3514 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3515 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3517 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3518 /* clear 10 bits. */
3519 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3522 /* MRQC: enable vmdq and dcb */
3523 mrqc = (num_pools == ETH_16_POOLS) ?
3524 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN;
3525 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3527 /* PFVTCTL: turn on virtualisation and set the default pool */
3528 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3529 if (cfg->enable_default_pool) {
3530 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3532 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3535 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3537 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
3539 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
3541 * mapping is done with 3 bits per priority,
3542 * so shift by i*3 each time
3544 queue_mapping |= ((cfg->dcb_tc[i] & 0x07) << (i * 3));
3546 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
3548 /* RTRPCS: DCB related */
3549 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
3551 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3552 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3553 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3554 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3556 /* VFTA - enable all vlan filters */
3557 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3558 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3561 /* VFRE: pool enabling for receive - 16 or 32 */
3562 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0),
3563 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3566 * MPSAR - allow pools to read specific mac addresses
3567 * In this case, all pools should be able to read from mac addr 0
3569 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
3570 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
3572 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3573 for (i = 0; i < cfg->nb_pool_maps; i++) {
3574 /* set vlan id in VF register and set the valid bit */
3575 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3576 (cfg->pool_map[i].vlan_id & 0xFFF)));
3578 * Put the allowed pools in VFB reg. As we only have 16 or 32
3579 * pools, we only need to use the first half of the register
3582 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
3587 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
3588 * @dev: pointer to eth_dev structure
3589 * @dcb_config: pointer to ixgbe_dcb_config structure
3592 ixgbe_dcb_tx_hw_config(struct rte_eth_dev *dev,
3593 struct ixgbe_dcb_config *dcb_config)
3596 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3598 PMD_INIT_FUNC_TRACE();
3599 if (hw->mac.type != ixgbe_mac_82598EB) {
3600 /* Disable the Tx desc arbiter so that MTQC can be changed */
3601 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3602 reg |= IXGBE_RTTDCS_ARBDIS;
3603 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3605 /* Enable DCB for Tx with 8 TCs */
3606 if (dcb_config->num_tcs.pg_tcs == 8) {
3607 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3609 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3611 if (dcb_config->vt_mode)
3612 reg |= IXGBE_MTQC_VT_ENA;
3613 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3615 /* Enable the Tx desc arbiter */
3616 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3617 reg &= ~IXGBE_RTTDCS_ARBDIS;
3618 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3620 /* Enable Security TX Buffer IFG for DCB */
3621 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3622 reg |= IXGBE_SECTX_DCB;
3623 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3628 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3629 * @dev: pointer to rte_eth_dev structure
3630 * @dcb_config: pointer to ixgbe_dcb_config structure
3633 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3634 struct ixgbe_dcb_config *dcb_config)
3636 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3637 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3638 struct ixgbe_hw *hw =
3639 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3641 PMD_INIT_FUNC_TRACE();
3642 if (hw->mac.type != ixgbe_mac_82598EB)
3643 /*PF VF Transmit Enable*/
3644 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3645 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3647 /*Configure general DCB TX parameters*/
3648 ixgbe_dcb_tx_hw_config(dev, dcb_config);
3652 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3653 struct ixgbe_dcb_config *dcb_config)
3655 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3656 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3657 struct ixgbe_dcb_tc_config *tc;
3660 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3661 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS) {
3662 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3663 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3665 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3666 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3669 /* Initialize User Priority to Traffic Class mapping */
3670 for (j = 0; j < IXGBE_DCB_MAX_TRAFFIC_CLASS; j++) {
3671 tc = &dcb_config->tc_config[j];
3672 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap = 0;
3675 /* User Priority to Traffic Class mapping */
3676 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3677 j = vmdq_rx_conf->dcb_tc[i];
3678 tc = &dcb_config->tc_config[j];
3679 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap |=
3685 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3686 struct ixgbe_dcb_config *dcb_config)
3688 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3689 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3690 struct ixgbe_dcb_tc_config *tc;
3693 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3694 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS) {
3695 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3696 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3698 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3699 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3702 /* Initialize User Priority to Traffic Class mapping */
3703 for (j = 0; j < IXGBE_DCB_MAX_TRAFFIC_CLASS; j++) {
3704 tc = &dcb_config->tc_config[j];
3705 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap = 0;
3708 /* User Priority to Traffic Class mapping */
3709 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3710 j = vmdq_tx_conf->dcb_tc[i];
3711 tc = &dcb_config->tc_config[j];
3712 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap |=
3718 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3719 struct ixgbe_dcb_config *dcb_config)
3721 struct rte_eth_dcb_rx_conf *rx_conf =
3722 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3723 struct ixgbe_dcb_tc_config *tc;
3726 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3727 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3729 /* Initialize User Priority to Traffic Class mapping */
3730 for (j = 0; j < IXGBE_DCB_MAX_TRAFFIC_CLASS; j++) {
3731 tc = &dcb_config->tc_config[j];
3732 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap = 0;
3735 /* User Priority to Traffic Class mapping */
3736 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3737 j = rx_conf->dcb_tc[i];
3738 tc = &dcb_config->tc_config[j];
3739 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap |=
3745 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3746 struct ixgbe_dcb_config *dcb_config)
3748 struct rte_eth_dcb_tx_conf *tx_conf =
3749 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3750 struct ixgbe_dcb_tc_config *tc;
3753 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3754 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3756 /* Initialize User Priority to Traffic Class mapping */
3757 for (j = 0; j < IXGBE_DCB_MAX_TRAFFIC_CLASS; j++) {
3758 tc = &dcb_config->tc_config[j];
3759 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap = 0;
3762 /* User Priority to Traffic Class mapping */
3763 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3764 j = tx_conf->dcb_tc[i];
3765 tc = &dcb_config->tc_config[j];
3766 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap |=
3772 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3773 * @dev: pointer to eth_dev structure
3774 * @dcb_config: pointer to ixgbe_dcb_config structure
3777 ixgbe_dcb_rx_hw_config(struct rte_eth_dev *dev,
3778 struct ixgbe_dcb_config *dcb_config)
3784 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3786 PMD_INIT_FUNC_TRACE();
3788 * Disable the arbiter before changing parameters
3789 * (always enable recycle mode; WSP)
3791 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3792 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3794 if (hw->mac.type != ixgbe_mac_82598EB) {
3795 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3796 if (dcb_config->num_tcs.pg_tcs == 4) {
3797 if (dcb_config->vt_mode)
3798 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3799 IXGBE_MRQC_VMDQRT4TCEN;
3801 /* no matter the mode is DCB or DCB_RSS, just
3802 * set the MRQE to RSSXTCEN. RSS is controlled
3805 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3806 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3807 IXGBE_MRQC_RTRSS4TCEN;
3810 if (dcb_config->num_tcs.pg_tcs == 8) {
3811 if (dcb_config->vt_mode)
3812 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3813 IXGBE_MRQC_VMDQRT8TCEN;
3815 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3816 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3817 IXGBE_MRQC_RTRSS8TCEN;
3821 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3823 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3824 /* Disable drop for all queues in VMDQ mode*/
3825 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3826 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3828 (q << IXGBE_QDE_IDX_SHIFT)));
3830 /* Enable drop for all queues in SRIOV mode */
3831 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3832 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3834 (q << IXGBE_QDE_IDX_SHIFT) |
3839 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3840 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3841 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3842 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3844 /* VFTA - enable all vlan filters */
3845 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3846 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3850 * Configure Rx packet plane (recycle mode; WSP) and
3853 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3854 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3858 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3859 uint16_t *max, uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3861 switch (hw->mac.type) {
3862 case ixgbe_mac_82598EB:
3863 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3865 case ixgbe_mac_82599EB:
3866 case ixgbe_mac_X540:
3867 case ixgbe_mac_X550:
3868 case ixgbe_mac_X550EM_x:
3869 case ixgbe_mac_X550EM_a:
3870 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3879 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3880 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3882 switch (hw->mac.type) {
3883 case ixgbe_mac_82598EB:
3884 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id, tsa);
3885 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id, tsa);
3887 case ixgbe_mac_82599EB:
3888 case ixgbe_mac_X540:
3889 case ixgbe_mac_X550:
3890 case ixgbe_mac_X550EM_x:
3891 case ixgbe_mac_X550EM_a:
3892 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id, tsa);
3893 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id, tsa, map);
3900 #define DCB_RX_CONFIG 1
3901 #define DCB_TX_CONFIG 1
3902 #define DCB_TX_PB 1024
3904 * ixgbe_dcb_hw_configure - Enable DCB and configure
3905 * general DCB in VT mode and non-VT mode parameters
3906 * @dev: pointer to rte_eth_dev structure
3907 * @dcb_config: pointer to ixgbe_dcb_config structure
3910 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3911 struct ixgbe_dcb_config *dcb_config)
3914 uint8_t i, pfc_en, nb_tcs;
3915 uint16_t pbsize, rx_buffer_size;
3916 uint8_t config_dcb_rx = 0;
3917 uint8_t config_dcb_tx = 0;
3918 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3919 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3920 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3921 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3922 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3923 struct ixgbe_dcb_tc_config *tc;
3924 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3925 struct ixgbe_hw *hw =
3926 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3927 struct ixgbe_bw_conf *bw_conf =
3928 IXGBE_DEV_PRIVATE_TO_BW_CONF(dev->data->dev_private);
3930 switch (dev->data->dev_conf.rxmode.mq_mode) {
3931 case ETH_MQ_RX_VMDQ_DCB:
3932 dcb_config->vt_mode = true;
3933 if (hw->mac.type != ixgbe_mac_82598EB) {
3934 config_dcb_rx = DCB_RX_CONFIG;
3936 *get dcb and VT rx configuration parameters
3939 ixgbe_vmdq_dcb_rx_config(dev, dcb_config);
3940 /*Configure general VMDQ and DCB RX parameters*/
3941 ixgbe_vmdq_dcb_configure(dev);
3945 case ETH_MQ_RX_DCB_RSS:
3946 dcb_config->vt_mode = false;
3947 config_dcb_rx = DCB_RX_CONFIG;
3948 /* Get dcb TX configuration parameters from rte_eth_conf */
3949 ixgbe_dcb_rx_config(dev, dcb_config);
3950 /*Configure general DCB RX parameters*/
3951 ixgbe_dcb_rx_hw_config(dev, dcb_config);
3954 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3957 switch (dev->data->dev_conf.txmode.mq_mode) {
3958 case ETH_MQ_TX_VMDQ_DCB:
3959 dcb_config->vt_mode = true;
3960 config_dcb_tx = DCB_TX_CONFIG;
3961 /* get DCB and VT TX configuration parameters
3964 ixgbe_dcb_vt_tx_config(dev, dcb_config);
3965 /*Configure general VMDQ and DCB TX parameters*/
3966 ixgbe_vmdq_dcb_hw_tx_config(dev, dcb_config);
3970 dcb_config->vt_mode = false;
3971 config_dcb_tx = DCB_TX_CONFIG;
3972 /*get DCB TX configuration parameters from rte_eth_conf*/
3973 ixgbe_dcb_tx_config(dev, dcb_config);
3974 /*Configure general DCB TX parameters*/
3975 ixgbe_dcb_tx_hw_config(dev, dcb_config);
3978 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3982 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3984 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3985 if (nb_tcs == ETH_4_TCS) {
3986 /* Avoid un-configured priority mapping to TC0 */
3988 uint8_t mask = 0xFF;
3990 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3991 mask = (uint8_t)(mask & (~(1 << map[i])));
3992 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3993 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3997 /* Re-configure 4 TCs BW */
3998 for (i = 0; i < nb_tcs; i++) {
3999 tc = &dcb_config->tc_config[i];
4000 if (bw_conf->tc_num != nb_tcs)
4001 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
4002 (uint8_t)(100 / nb_tcs);
4003 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
4004 (uint8_t)(100 / nb_tcs);
4006 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
4007 tc = &dcb_config->tc_config[i];
4008 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
4009 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
4012 /* Re-configure 8 TCs BW */
4013 for (i = 0; i < nb_tcs; i++) {
4014 tc = &dcb_config->tc_config[i];
4015 if (bw_conf->tc_num != nb_tcs)
4016 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
4017 (uint8_t)(100 / nb_tcs + (i & 1));
4018 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
4019 (uint8_t)(100 / nb_tcs + (i & 1));
4023 switch (hw->mac.type) {
4024 case ixgbe_mac_X550:
4025 case ixgbe_mac_X550EM_x:
4026 case ixgbe_mac_X550EM_a:
4027 rx_buffer_size = X550_RX_BUFFER_SIZE;
4030 rx_buffer_size = NIC_RX_BUFFER_SIZE;
4034 if (config_dcb_rx) {
4035 /* Set RX buffer size */
4036 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
4037 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
4039 for (i = 0; i < nb_tcs; i++) {
4040 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
4042 /* zero alloc all unused TCs */
4043 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
4044 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
4047 if (config_dcb_tx) {
4048 /* Only support an equally distributed
4049 * Tx packet buffer strategy.
4051 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
4052 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
4054 for (i = 0; i < nb_tcs; i++) {
4055 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
4056 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
4058 /* Clear unused TCs, if any, to zero buffer size*/
4059 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
4060 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
4061 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
4065 /*Calculates traffic class credits*/
4066 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
4067 IXGBE_DCB_TX_CONFIG);
4068 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
4069 IXGBE_DCB_RX_CONFIG);
4071 if (config_dcb_rx) {
4072 /* Unpack CEE standard containers */
4073 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
4074 ixgbe_dcb_unpack_max_cee(dcb_config, max);
4075 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
4076 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
4077 /* Configure PG(ETS) RX */
4078 ixgbe_dcb_hw_arbite_rx_config(hw, refill, max, bwgid, tsa, map);
4081 if (config_dcb_tx) {
4082 /* Unpack CEE standard containers */
4083 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
4084 ixgbe_dcb_unpack_max_cee(dcb_config, max);
4085 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
4086 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
4087 /* Configure PG(ETS) TX */
4088 ixgbe_dcb_hw_arbite_tx_config(hw, refill, max, bwgid, tsa, map);
4091 /*Configure queue statistics registers*/
4092 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
4094 /* Check if the PFC is supported */
4095 if (dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
4096 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
4097 for (i = 0; i < nb_tcs; i++) {
4099 * If the TC count is 8,and the default high_water is 48,
4100 * the low_water is 16 as default.
4102 hw->fc.high_water[i] = (pbsize * 3) / 4;
4103 hw->fc.low_water[i] = pbsize / 4;
4104 /* Enable pfc for this TC */
4105 tc = &dcb_config->tc_config[i];
4106 tc->pfc = ixgbe_dcb_pfc_enabled;
4108 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
4109 if (dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
4111 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
4118 * ixgbe_configure_dcb - Configure DCB Hardware
4119 * @dev: pointer to rte_eth_dev
4121 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
4123 struct ixgbe_dcb_config *dcb_cfg =
4124 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
4125 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
4127 PMD_INIT_FUNC_TRACE();
4129 /* check support mq_mode for DCB */
4130 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
4131 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB) &&
4132 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB_RSS))
4135 if (dev->data->nb_rx_queues > ETH_DCB_NUM_QUEUES)
4138 /** Configure DCB hardware **/
4139 ixgbe_dcb_hw_configure(dev, dcb_cfg);
4143 * VMDq only support for 10 GbE NIC.
4146 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
4148 struct rte_eth_vmdq_rx_conf *cfg;
4149 struct ixgbe_hw *hw;
4150 enum rte_eth_nb_pools num_pools;
4151 uint32_t mrqc, vt_ctl, vlanctrl;
4155 PMD_INIT_FUNC_TRACE();
4156 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4157 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
4158 num_pools = cfg->nb_queue_pools;
4160 ixgbe_rss_disable(dev);
4162 /* MRQC: enable vmdq */
4163 mrqc = IXGBE_MRQC_VMDQEN;
4164 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4166 /* PFVTCTL: turn on virtualisation and set the default pool */
4167 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
4168 if (cfg->enable_default_pool)
4169 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
4171 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
4173 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
4175 for (i = 0; i < (int)num_pools; i++) {
4176 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
4177 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
4180 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
4181 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4182 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
4183 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
4185 /* VFTA - enable all vlan filters */
4186 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
4187 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
4189 /* VFRE: pool enabling for receive - 64 */
4190 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
4191 if (num_pools == ETH_64_POOLS)
4192 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
4195 * MPSAR - allow pools to read specific mac addresses
4196 * In this case, all pools should be able to read from mac addr 0
4198 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
4199 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
4201 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
4202 for (i = 0; i < cfg->nb_pool_maps; i++) {
4203 /* set vlan id in VF register and set the valid bit */
4204 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
4205 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
4207 * Put the allowed pools in VFB reg. As we only have 16 or 64
4208 * pools, we only need to use the first half of the register
4211 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
4212 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i * 2),
4213 (cfg->pool_map[i].pools & UINT32_MAX));
4215 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i * 2 + 1)),
4216 ((cfg->pool_map[i].pools >> 32) & UINT32_MAX));
4220 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
4221 if (cfg->enable_loop_back) {
4222 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4223 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
4224 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
4227 IXGBE_WRITE_FLUSH(hw);
4231 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
4232 * @hw: pointer to hardware structure
4235 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
4240 PMD_INIT_FUNC_TRACE();
4241 /*PF VF Transmit Enable*/
4242 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
4243 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
4245 /* Disable the Tx desc arbiter so that MTQC can be changed */
4246 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4247 reg |= IXGBE_RTTDCS_ARBDIS;
4248 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
4250 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
4251 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
4253 /* Disable drop for all queues */
4254 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
4255 IXGBE_WRITE_REG(hw, IXGBE_QDE,
4256 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
4258 /* Enable the Tx desc arbiter */
4259 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4260 reg &= ~IXGBE_RTTDCS_ARBDIS;
4261 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
4263 IXGBE_WRITE_FLUSH(hw);
4266 static int __attribute__((cold))
4267 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
4269 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
4273 /* Initialize software ring entries */
4274 for (i = 0; i < rxq->nb_rx_desc; i++) {
4275 volatile union ixgbe_adv_rx_desc *rxd;
4276 struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
4279 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
4280 (unsigned) rxq->queue_id);
4284 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
4285 mbuf->port = rxq->port_id;
4288 rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
4289 rxd = &rxq->rx_ring[i];
4290 rxd->read.hdr_addr = 0;
4291 rxd->read.pkt_addr = dma_addr;
4299 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
4301 struct ixgbe_hw *hw;
4304 ixgbe_rss_configure(dev);
4306 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4308 /* MRQC: enable VF RSS */
4309 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
4310 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
4311 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4313 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
4317 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
4321 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
4325 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4331 ixgbe_config_vf_default(struct rte_eth_dev *dev)
4333 struct ixgbe_hw *hw =
4334 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4336 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4338 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4343 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4344 IXGBE_MRQC_VMDQRT4TCEN);
4348 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4349 IXGBE_MRQC_VMDQRT8TCEN);
4353 "invalid pool number in IOV mode");
4360 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
4362 struct ixgbe_hw *hw =
4363 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4365 if (hw->mac.type == ixgbe_mac_82598EB)
4368 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4370 * SRIOV inactive scheme
4371 * any DCB/RSS w/o VMDq multi-queue setting
4373 switch (dev->data->dev_conf.rxmode.mq_mode) {
4375 case ETH_MQ_RX_DCB_RSS:
4376 case ETH_MQ_RX_VMDQ_RSS:
4377 ixgbe_rss_configure(dev);
4380 case ETH_MQ_RX_VMDQ_DCB:
4381 ixgbe_vmdq_dcb_configure(dev);
4384 case ETH_MQ_RX_VMDQ_ONLY:
4385 ixgbe_vmdq_rx_hw_configure(dev);
4388 case ETH_MQ_RX_NONE:
4390 /* if mq_mode is none, disable rss mode.*/
4391 ixgbe_rss_disable(dev);
4395 /* SRIOV active scheme
4396 * Support RSS together with SRIOV.
4398 switch (dev->data->dev_conf.rxmode.mq_mode) {
4400 case ETH_MQ_RX_VMDQ_RSS:
4401 ixgbe_config_vf_rss(dev);
4403 case ETH_MQ_RX_VMDQ_DCB:
4405 /* In SRIOV, the configuration is the same as VMDq case */
4406 ixgbe_vmdq_dcb_configure(dev);
4408 /* DCB/RSS together with SRIOV is not supported */
4409 case ETH_MQ_RX_VMDQ_DCB_RSS:
4410 case ETH_MQ_RX_DCB_RSS:
4412 "Could not support DCB/RSS with VMDq & SRIOV");
4415 ixgbe_config_vf_default(dev);
4424 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
4426 struct ixgbe_hw *hw =
4427 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4431 if (hw->mac.type == ixgbe_mac_82598EB)
4434 /* disable arbiter before setting MTQC */
4435 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4436 rttdcs |= IXGBE_RTTDCS_ARBDIS;
4437 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4439 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4441 * SRIOV inactive scheme
4442 * any DCB w/o VMDq multi-queue setting
4444 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
4445 ixgbe_vmdq_tx_hw_configure(hw);
4447 mtqc = IXGBE_MTQC_64Q_1PB;
4448 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4451 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4454 * SRIOV active scheme
4455 * FIXME if support DCB together with VMDq & SRIOV
4458 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
4461 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
4464 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
4468 mtqc = IXGBE_MTQC_64Q_1PB;
4469 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
4471 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4474 /* re-enable arbiter */
4475 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
4476 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4482 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
4484 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
4485 * spec rev. 3.0 chapter 8.2.3.8.13.
4487 * @pool Memory pool of the Rx queue
4489 static inline uint32_t
4490 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
4492 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
4494 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
4497 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
4500 return IXGBE_RSCCTL_MAXDESC_16;
4501 else if (maxdesc >= 8)
4502 return IXGBE_RSCCTL_MAXDESC_8;
4503 else if (maxdesc >= 4)
4504 return IXGBE_RSCCTL_MAXDESC_4;
4506 return IXGBE_RSCCTL_MAXDESC_1;
4510 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
4513 * (Taken from FreeBSD tree)
4514 * (yes this is all very magic and confusing :)
4517 * @entry the register array entry
4518 * @vector the MSIX vector for this queue
4522 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
4524 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4527 vector |= IXGBE_IVAR_ALLOC_VAL;
4529 switch (hw->mac.type) {
4531 case ixgbe_mac_82598EB:
4533 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
4535 entry += (type * 64);
4536 index = (entry >> 2) & 0x1F;
4537 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4538 ivar &= ~(0xFF << (8 * (entry & 0x3)));
4539 ivar |= (vector << (8 * (entry & 0x3)));
4540 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4543 case ixgbe_mac_82599EB:
4544 case ixgbe_mac_X540:
4545 if (type == -1) { /* MISC IVAR */
4546 index = (entry & 1) * 8;
4547 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4548 ivar &= ~(0xFF << index);
4549 ivar |= (vector << index);
4550 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4551 } else { /* RX/TX IVARS */
4552 index = (16 * (entry & 1)) + (8 * type);
4553 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
4554 ivar &= ~(0xFF << index);
4555 ivar |= (vector << index);
4556 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
4566 void __attribute__((cold))
4567 ixgbe_set_rx_function(struct rte_eth_dev *dev)
4569 uint16_t i, rx_using_sse;
4570 struct ixgbe_adapter *adapter =
4571 (struct ixgbe_adapter *)dev->data->dev_private;
4574 * In order to allow Vector Rx there are a few configuration
4575 * conditions to be met and Rx Bulk Allocation should be allowed.
4577 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
4578 !adapter->rx_bulk_alloc_allowed) {
4579 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
4580 "preconditions or RTE_IXGBE_INC_VECTOR is "
4582 dev->data->port_id);
4584 adapter->rx_vec_allowed = false;
4588 * Initialize the appropriate LRO callback.
4590 * If all queues satisfy the bulk allocation preconditions
4591 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
4592 * Otherwise use a single allocation version.
4594 if (dev->data->lro) {
4595 if (adapter->rx_bulk_alloc_allowed) {
4596 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a bulk "
4597 "allocation version");
4598 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4600 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a single "
4601 "allocation version");
4602 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4604 } else if (dev->data->scattered_rx) {
4606 * Set the non-LRO scattered callback: there are Vector and
4607 * single allocation versions.
4609 if (adapter->rx_vec_allowed) {
4610 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
4611 "callback (port=%d).",
4612 dev->data->port_id);
4614 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
4615 } else if (adapter->rx_bulk_alloc_allowed) {
4616 PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
4617 "allocation callback (port=%d).",
4618 dev->data->port_id);
4619 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4621 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
4622 "single allocation) "
4623 "Scattered Rx callback "
4625 dev->data->port_id);
4627 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4630 * Below we set "simple" callbacks according to port/queues parameters.
4631 * If parameters allow we are going to choose between the following
4635 * - Single buffer allocation (the simplest one)
4637 } else if (adapter->rx_vec_allowed) {
4638 PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
4639 "burst size no less than %d (port=%d).",
4640 RTE_IXGBE_DESCS_PER_LOOP,
4641 dev->data->port_id);
4643 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
4644 } else if (adapter->rx_bulk_alloc_allowed) {
4645 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
4646 "satisfied. Rx Burst Bulk Alloc function "
4647 "will be used on port=%d.",
4648 dev->data->port_id);
4650 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
4652 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
4653 "satisfied, or Scattered Rx is requested "
4655 dev->data->port_id);
4657 dev->rx_pkt_burst = ixgbe_recv_pkts;
4660 /* Propagate information about RX function choice through all queues. */
4663 (dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec ||
4664 dev->rx_pkt_burst == ixgbe_recv_pkts_vec);
4666 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4667 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4669 rxq->rx_using_sse = rx_using_sse;
4670 #ifdef RTE_LIBRTE_SECURITY
4671 rxq->using_ipsec = !!(dev->data->dev_conf.rxmode.offloads &
4672 DEV_RX_OFFLOAD_SECURITY);
4678 * ixgbe_set_rsc - configure RSC related port HW registers
4680 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4681 * of 82599 Spec (x540 configuration is virtually the same).
4685 * Returns 0 in case of success or a non-zero error code
4688 ixgbe_set_rsc(struct rte_eth_dev *dev)
4690 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4691 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4692 struct rte_eth_dev_info dev_info = { 0 };
4693 bool rsc_capable = false;
4699 dev->dev_ops->dev_infos_get(dev, &dev_info);
4700 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4703 if (!rsc_capable && (rx_conf->offloads & DEV_RX_OFFLOAD_TCP_LRO)) {
4704 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4709 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4711 if ((rx_conf->offloads & DEV_RX_OFFLOAD_KEEP_CRC) &&
4712 (rx_conf->offloads & DEV_RX_OFFLOAD_TCP_LRO)) {
4714 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4715 * 3.0 RSC configuration requires HW CRC stripping being
4716 * enabled. If user requested both HW CRC stripping off
4717 * and RSC on - return an error.
4719 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4724 /* RFCTL configuration */
4725 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4726 if ((rsc_capable) && (rx_conf->offloads & DEV_RX_OFFLOAD_TCP_LRO))
4728 * Since NFS packets coalescing is not supported - clear
4729 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4732 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4733 IXGBE_RFCTL_NFSR_DIS);
4735 rfctl |= IXGBE_RFCTL_RSC_DIS;
4736 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4738 /* If LRO hasn't been requested - we are done here. */
4739 if (!(rx_conf->offloads & DEV_RX_OFFLOAD_TCP_LRO))
4742 /* Set RDRXCTL.RSCACKC bit */
4743 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4744 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4745 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4747 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4748 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4749 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4751 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4753 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4755 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4757 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4760 * ixgbe PMD doesn't support header-split at the moment.
4762 * Following the 4.6.7.2.1 chapter of the 82599/x540
4763 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4764 * should be configured even if header split is not
4765 * enabled. We will configure it 128 bytes following the
4766 * recommendation in the spec.
4768 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4769 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4770 IXGBE_SRRCTL_BSIZEHDR_MASK;
4773 * TODO: Consider setting the Receive Descriptor Minimum
4774 * Threshold Size for an RSC case. This is not an obviously
4775 * beneficiary option but the one worth considering...
4778 rscctl |= IXGBE_RSCCTL_RSCEN;
4779 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4780 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4783 * RSC: Set ITR interval corresponding to 2K ints/s.
4785 * Full-sized RSC aggregations for a 10Gb/s link will
4786 * arrive at about 20K aggregation/s rate.
4788 * 2K inst/s rate will make only 10% of the
4789 * aggregations to be closed due to the interrupt timer
4790 * expiration for a streaming at wire-speed case.
4792 * For a sparse streaming case this setting will yield
4793 * at most 500us latency for a single RSC aggregation.
4795 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4796 eitr |= IXGBE_EITR_INTERVAL_US(IXGBE_QUEUE_ITR_INTERVAL_DEFAULT);
4797 eitr |= IXGBE_EITR_CNT_WDIS;
4799 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4800 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4801 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4802 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4805 * RSC requires the mapping of the queue to the
4808 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4813 PMD_INIT_LOG(DEBUG, "enabling LRO mode");
4819 * Initializes Receive Unit.
4821 int __attribute__((cold))
4822 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4824 struct ixgbe_hw *hw;
4825 struct ixgbe_rx_queue *rxq;
4836 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4839 PMD_INIT_FUNC_TRACE();
4840 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4843 * Make sure receives are disabled while setting
4844 * up the RX context (registers, descriptor rings, etc.).
4846 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4847 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4849 /* Enable receipt of broadcasted frames */
4850 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4851 fctrl |= IXGBE_FCTRL_BAM;
4852 fctrl |= IXGBE_FCTRL_DPF;
4853 fctrl |= IXGBE_FCTRL_PMCF;
4854 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4857 * Configure CRC stripping, if any.
4859 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4860 if (rx_conf->offloads & DEV_RX_OFFLOAD_KEEP_CRC)
4861 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4863 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4866 * Configure jumbo frame support, if any.
4868 if (rx_conf->offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
4869 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4870 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4871 maxfrs &= 0x0000FFFF;
4872 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4873 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4875 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4878 * If loopback mode is configured for 82599, set LPBK bit.
4880 if (hw->mac.type == ixgbe_mac_82599EB &&
4881 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4882 hlreg0 |= IXGBE_HLREG0_LPBK;
4884 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4886 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4889 * Assume no header split and no VLAN strip support
4890 * on any Rx queue first .
4892 rx_conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
4893 /* Setup RX queues */
4894 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4895 rxq = dev->data->rx_queues[i];
4898 * Reset crc_len in case it was changed after queue setup by a
4899 * call to configure.
4901 if (rx_conf->offloads & DEV_RX_OFFLOAD_KEEP_CRC)
4902 rxq->crc_len = ETHER_CRC_LEN;
4906 /* Setup the Base and Length of the Rx Descriptor Rings */
4907 bus_addr = rxq->rx_ring_phys_addr;
4908 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4909 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4910 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4911 (uint32_t)(bus_addr >> 32));
4912 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4913 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4914 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4915 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4917 /* Configure the SRRCTL register */
4918 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4920 /* Set if packets are dropped when no descriptors available */
4922 srrctl |= IXGBE_SRRCTL_DROP_EN;
4925 * Configure the RX buffer size in the BSIZEPACKET field of
4926 * the SRRCTL register of the queue.
4927 * The value is in 1 KB resolution. Valid values can be from
4930 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4931 RTE_PKTMBUF_HEADROOM);
4932 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4933 IXGBE_SRRCTL_BSIZEPKT_MASK);
4935 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4937 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4938 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4940 /* It adds dual VLAN length for supporting dual VLAN */
4941 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4942 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4943 dev->data->scattered_rx = 1;
4944 if (rxq->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
4945 rx_conf->offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
4948 if (rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER)
4949 dev->data->scattered_rx = 1;
4952 * Device configured with multiple RX queues.
4954 ixgbe_dev_mq_rx_configure(dev);
4957 * Setup the Checksum Register.
4958 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4959 * Enable IP/L4 checkum computation by hardware if requested to do so.
4961 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4962 rxcsum |= IXGBE_RXCSUM_PCSD;
4963 if (rx_conf->offloads & DEV_RX_OFFLOAD_CHECKSUM)
4964 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4966 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4968 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4970 if (hw->mac.type == ixgbe_mac_82599EB ||
4971 hw->mac.type == ixgbe_mac_X540) {
4972 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4973 if (rx_conf->offloads & DEV_RX_OFFLOAD_KEEP_CRC)
4974 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4976 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4977 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4978 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4981 rc = ixgbe_set_rsc(dev);
4985 ixgbe_set_rx_function(dev);
4991 * Initializes Transmit Unit.
4993 void __attribute__((cold))
4994 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4996 struct ixgbe_hw *hw;
4997 struct ixgbe_tx_queue *txq;
5003 PMD_INIT_FUNC_TRACE();
5004 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5006 /* Enable TX CRC (checksum offload requirement) and hw padding
5009 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
5010 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
5011 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
5013 /* Setup the Base and Length of the Tx Descriptor Rings */
5014 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5015 txq = dev->data->tx_queues[i];
5017 bus_addr = txq->tx_ring_phys_addr;
5018 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
5019 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5020 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
5021 (uint32_t)(bus_addr >> 32));
5022 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
5023 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
5024 /* Setup the HW Tx Head and TX Tail descriptor pointers */
5025 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
5026 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
5029 * Disable Tx Head Writeback RO bit, since this hoses
5030 * bookkeeping if things aren't delivered in order.
5032 switch (hw->mac.type) {
5033 case ixgbe_mac_82598EB:
5034 txctrl = IXGBE_READ_REG(hw,
5035 IXGBE_DCA_TXCTRL(txq->reg_idx));
5036 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
5037 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
5041 case ixgbe_mac_82599EB:
5042 case ixgbe_mac_X540:
5043 case ixgbe_mac_X550:
5044 case ixgbe_mac_X550EM_x:
5045 case ixgbe_mac_X550EM_a:
5047 txctrl = IXGBE_READ_REG(hw,
5048 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
5049 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
5050 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
5056 /* Device configured with multiple TX queues. */
5057 ixgbe_dev_mq_tx_configure(dev);
5061 * Set up link for 82599 loopback mode Tx->Rx.
5063 static inline void __attribute__((cold))
5064 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
5066 PMD_INIT_FUNC_TRACE();
5068 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
5069 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
5071 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
5080 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
5081 ixgbe_reset_pipeline_82599(hw);
5083 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
5089 * Start Transmit and Receive Units.
5091 int __attribute__((cold))
5092 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
5094 struct ixgbe_hw *hw;
5095 struct ixgbe_tx_queue *txq;
5096 struct ixgbe_rx_queue *rxq;
5103 PMD_INIT_FUNC_TRACE();
5104 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5106 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5107 txq = dev->data->tx_queues[i];
5108 /* Setup Transmit Threshold Registers */
5109 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5110 txdctl |= txq->pthresh & 0x7F;
5111 txdctl |= ((txq->hthresh & 0x7F) << 8);
5112 txdctl |= ((txq->wthresh & 0x7F) << 16);
5113 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5116 if (hw->mac.type != ixgbe_mac_82598EB) {
5117 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
5118 dmatxctl |= IXGBE_DMATXCTL_TE;
5119 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
5122 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5123 txq = dev->data->tx_queues[i];
5124 if (!txq->tx_deferred_start) {
5125 ret = ixgbe_dev_tx_queue_start(dev, i);
5131 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5132 rxq = dev->data->rx_queues[i];
5133 if (!rxq->rx_deferred_start) {
5134 ret = ixgbe_dev_rx_queue_start(dev, i);
5140 /* Enable Receive engine */
5141 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
5142 if (hw->mac.type == ixgbe_mac_82598EB)
5143 rxctrl |= IXGBE_RXCTRL_DMBYPS;
5144 rxctrl |= IXGBE_RXCTRL_RXEN;
5145 hw->mac.ops.enable_rx_dma(hw, rxctrl);
5147 /* If loopback mode is enabled for 82599, set up the link accordingly */
5148 if (hw->mac.type == ixgbe_mac_82599EB &&
5149 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
5150 ixgbe_setup_loopback_link_82599(hw);
5152 #ifdef RTE_LIBRTE_SECURITY
5153 if ((dev->data->dev_conf.rxmode.offloads &
5154 DEV_RX_OFFLOAD_SECURITY) ||
5155 (dev->data->dev_conf.txmode.offloads &
5156 DEV_TX_OFFLOAD_SECURITY)) {
5157 ret = ixgbe_crypto_enable_ipsec(dev);
5160 "ixgbe_crypto_enable_ipsec fails with %d.",
5171 * Start Receive Units for specified queue.
5173 int __attribute__((cold))
5174 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
5176 struct ixgbe_hw *hw;
5177 struct ixgbe_rx_queue *rxq;
5181 PMD_INIT_FUNC_TRACE();
5182 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5184 rxq = dev->data->rx_queues[rx_queue_id];
5186 /* Allocate buffers for descriptor rings */
5187 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
5188 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
5192 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
5193 rxdctl |= IXGBE_RXDCTL_ENABLE;
5194 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
5196 /* Wait until RX Enable ready */
5197 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5200 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
5201 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
5203 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", rx_queue_id);
5205 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
5206 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
5207 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
5213 * Stop Receive Units for specified queue.
5215 int __attribute__((cold))
5216 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
5218 struct ixgbe_hw *hw;
5219 struct ixgbe_adapter *adapter =
5220 (struct ixgbe_adapter *)dev->data->dev_private;
5221 struct ixgbe_rx_queue *rxq;
5225 PMD_INIT_FUNC_TRACE();
5226 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5228 rxq = dev->data->rx_queues[rx_queue_id];
5230 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
5231 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5232 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
5234 /* Wait until RX Enable bit clear */
5235 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5238 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
5239 } while (--poll_ms && (rxdctl & IXGBE_RXDCTL_ENABLE));
5241 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d", rx_queue_id);
5243 rte_delay_us(RTE_IXGBE_WAIT_100_US);
5245 ixgbe_rx_queue_release_mbufs(rxq);
5246 ixgbe_reset_rx_queue(adapter, rxq);
5247 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
5254 * Start Transmit Units for specified queue.
5256 int __attribute__((cold))
5257 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
5259 struct ixgbe_hw *hw;
5260 struct ixgbe_tx_queue *txq;
5264 PMD_INIT_FUNC_TRACE();
5265 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5267 txq = dev->data->tx_queues[tx_queue_id];
5268 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
5269 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5270 txdctl |= IXGBE_TXDCTL_ENABLE;
5271 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5273 /* Wait until TX Enable ready */
5274 if (hw->mac.type == ixgbe_mac_82599EB) {
5275 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5278 txdctl = IXGBE_READ_REG(hw,
5279 IXGBE_TXDCTL(txq->reg_idx));
5280 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
5282 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d",
5286 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
5287 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
5293 * Stop Transmit Units for specified queue.
5295 int __attribute__((cold))
5296 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
5298 struct ixgbe_hw *hw;
5299 struct ixgbe_tx_queue *txq;
5301 uint32_t txtdh, txtdt;
5304 PMD_INIT_FUNC_TRACE();
5305 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5307 txq = dev->data->tx_queues[tx_queue_id];
5309 /* Wait until TX queue is empty */
5310 if (hw->mac.type == ixgbe_mac_82599EB) {
5311 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5313 rte_delay_us(RTE_IXGBE_WAIT_100_US);
5314 txtdh = IXGBE_READ_REG(hw,
5315 IXGBE_TDH(txq->reg_idx));
5316 txtdt = IXGBE_READ_REG(hw,
5317 IXGBE_TDT(txq->reg_idx));
5318 } while (--poll_ms && (txtdh != txtdt));
5321 "Tx Queue %d is not empty when stopping.",
5325 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5326 txdctl &= ~IXGBE_TXDCTL_ENABLE;
5327 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5329 /* Wait until TX Enable bit clear */
5330 if (hw->mac.type == ixgbe_mac_82599EB) {
5331 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5334 txdctl = IXGBE_READ_REG(hw,
5335 IXGBE_TXDCTL(txq->reg_idx));
5336 } while (--poll_ms && (txdctl & IXGBE_TXDCTL_ENABLE));
5338 PMD_INIT_LOG(ERR, "Could not disable Tx Queue %d",
5342 if (txq->ops != NULL) {
5343 txq->ops->release_mbufs(txq);
5344 txq->ops->reset(txq);
5346 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
5352 ixgbe_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5353 struct rte_eth_rxq_info *qinfo)
5355 struct ixgbe_rx_queue *rxq;
5357 rxq = dev->data->rx_queues[queue_id];
5359 qinfo->mp = rxq->mb_pool;
5360 qinfo->scattered_rx = dev->data->scattered_rx;
5361 qinfo->nb_desc = rxq->nb_rx_desc;
5363 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
5364 qinfo->conf.rx_drop_en = rxq->drop_en;
5365 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
5366 qinfo->conf.offloads = rxq->offloads;
5370 ixgbe_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5371 struct rte_eth_txq_info *qinfo)
5373 struct ixgbe_tx_queue *txq;
5375 txq = dev->data->tx_queues[queue_id];
5377 qinfo->nb_desc = txq->nb_tx_desc;
5379 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
5380 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
5381 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
5383 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
5384 qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
5385 qinfo->conf.offloads = txq->offloads;
5386 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
5390 * [VF] Initializes Receive Unit.
5392 int __attribute__((cold))
5393 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
5395 struct ixgbe_hw *hw;
5396 struct ixgbe_rx_queue *rxq;
5397 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
5399 uint32_t srrctl, psrtype = 0;
5404 PMD_INIT_FUNC_TRACE();
5405 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5407 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
5408 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5409 "it should be power of 2");
5413 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
5414 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5415 "it should be equal to or less than %d",
5416 hw->mac.max_rx_queues);
5421 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
5422 * disables the VF receipt of packets if the PF MTU is > 1500.
5423 * This is done to deal with 82599 limitations that imposes
5424 * the PF and all VFs to share the same MTU.
5425 * Then, the PF driver enables again the VF receipt of packet when
5426 * the VF driver issues a IXGBE_VF_SET_LPE request.
5427 * In the meantime, the VF device cannot be used, even if the VF driver
5428 * and the Guest VM network stack are ready to accept packets with a
5429 * size up to the PF MTU.
5430 * As a work-around to this PF behaviour, force the call to
5431 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
5432 * VF packets received can work in all cases.
5434 ixgbevf_rlpml_set_vf(hw,
5435 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
5438 * Assume no header split and no VLAN strip support
5439 * on any Rx queue first .
5441 rxmode->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
5442 /* Setup RX queues */
5443 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5444 rxq = dev->data->rx_queues[i];
5446 /* Allocate buffers for descriptor rings */
5447 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
5451 /* Setup the Base and Length of the Rx Descriptor Rings */
5452 bus_addr = rxq->rx_ring_phys_addr;
5454 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
5455 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5456 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
5457 (uint32_t)(bus_addr >> 32));
5458 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
5459 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
5460 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
5461 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
5464 /* Configure the SRRCTL register */
5465 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
5467 /* Set if packets are dropped when no descriptors available */
5469 srrctl |= IXGBE_SRRCTL_DROP_EN;
5472 * Configure the RX buffer size in the BSIZEPACKET field of
5473 * the SRRCTL register of the queue.
5474 * The value is in 1 KB resolution. Valid values can be from
5477 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
5478 RTE_PKTMBUF_HEADROOM);
5479 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
5480 IXGBE_SRRCTL_BSIZEPKT_MASK);
5483 * VF modification to write virtual function SRRCTL register
5485 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
5487 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
5488 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
5490 if (rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ||
5491 /* It adds dual VLAN length for supporting dual VLAN */
5492 (rxmode->max_rx_pkt_len +
5493 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
5494 if (!dev->data->scattered_rx)
5495 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
5496 dev->data->scattered_rx = 1;
5499 if (rxq->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
5500 rxmode->offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
5503 /* Set RQPL for VF RSS according to max Rx queue */
5504 psrtype |= (dev->data->nb_rx_queues >> 1) <<
5505 IXGBE_PSRTYPE_RQPL_SHIFT;
5506 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
5508 ixgbe_set_rx_function(dev);
5514 * [VF] Initializes Transmit Unit.
5516 void __attribute__((cold))
5517 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
5519 struct ixgbe_hw *hw;
5520 struct ixgbe_tx_queue *txq;
5525 PMD_INIT_FUNC_TRACE();
5526 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5528 /* Setup the Base and Length of the Tx Descriptor Rings */
5529 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5530 txq = dev->data->tx_queues[i];
5531 bus_addr = txq->tx_ring_phys_addr;
5532 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
5533 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5534 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
5535 (uint32_t)(bus_addr >> 32));
5536 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
5537 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
5538 /* Setup the HW Tx Head and TX Tail descriptor pointers */
5539 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
5540 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
5543 * Disable Tx Head Writeback RO bit, since this hoses
5544 * bookkeeping if things aren't delivered in order.
5546 txctrl = IXGBE_READ_REG(hw,
5547 IXGBE_VFDCA_TXCTRL(i));
5548 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
5549 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
5555 * [VF] Start Transmit and Receive Units.
5557 void __attribute__((cold))
5558 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
5560 struct ixgbe_hw *hw;
5561 struct ixgbe_tx_queue *txq;
5562 struct ixgbe_rx_queue *rxq;
5568 PMD_INIT_FUNC_TRACE();
5569 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5571 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5572 txq = dev->data->tx_queues[i];
5573 /* Setup Transmit Threshold Registers */
5574 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5575 txdctl |= txq->pthresh & 0x7F;
5576 txdctl |= ((txq->hthresh & 0x7F) << 8);
5577 txdctl |= ((txq->wthresh & 0x7F) << 16);
5578 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5581 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5583 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5584 txdctl |= IXGBE_TXDCTL_ENABLE;
5585 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5588 /* Wait until TX Enable ready */
5591 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5592 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
5594 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
5596 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5598 rxq = dev->data->rx_queues[i];
5600 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5601 rxdctl |= IXGBE_RXDCTL_ENABLE;
5602 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
5604 /* Wait until RX Enable ready */
5608 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5609 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
5611 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
5613 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
5619 ixgbe_rss_conf_init(struct ixgbe_rte_flow_rss_conf *out,
5620 const struct rte_flow_action_rss *in)
5622 if (in->key_len > RTE_DIM(out->key) ||
5623 in->queue_num > RTE_DIM(out->queue))
5625 out->conf = (struct rte_flow_action_rss){
5629 .key_len = in->key_len,
5630 .queue_num = in->queue_num,
5631 .key = memcpy(out->key, in->key, in->key_len),
5632 .queue = memcpy(out->queue, in->queue,
5633 sizeof(*in->queue) * in->queue_num),
5639 ixgbe_action_rss_same(const struct rte_flow_action_rss *comp,
5640 const struct rte_flow_action_rss *with)
5642 return (comp->func == with->func &&
5643 comp->level == with->level &&
5644 comp->types == with->types &&
5645 comp->key_len == with->key_len &&
5646 comp->queue_num == with->queue_num &&
5647 !memcmp(comp->key, with->key, with->key_len) &&
5648 !memcmp(comp->queue, with->queue,
5649 sizeof(*with->queue) * with->queue_num));
5653 ixgbe_config_rss_filter(struct rte_eth_dev *dev,
5654 struct ixgbe_rte_flow_rss_conf *conf, bool add)
5656 struct ixgbe_hw *hw;
5660 uint16_t sp_reta_size;
5662 struct rte_eth_rss_conf rss_conf = {
5663 .rss_key = conf->conf.key_len ?
5664 (void *)(uintptr_t)conf->conf.key : NULL,
5665 .rss_key_len = conf->conf.key_len,
5666 .rss_hf = conf->conf.types,
5668 struct ixgbe_filter_info *filter_info =
5669 IXGBE_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
5671 PMD_INIT_FUNC_TRACE();
5672 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5674 sp_reta_size = ixgbe_reta_size_get(hw->mac.type);
5677 if (ixgbe_action_rss_same(&filter_info->rss_info.conf,
5679 ixgbe_rss_disable(dev);
5680 memset(&filter_info->rss_info, 0,
5681 sizeof(struct ixgbe_rte_flow_rss_conf));
5687 if (filter_info->rss_info.conf.queue_num)
5689 /* Fill in redirection table
5690 * The byte-swap is needed because NIC registers are in
5691 * little-endian order.
5694 for (i = 0, j = 0; i < sp_reta_size; i++, j++) {
5695 reta_reg = ixgbe_reta_reg_get(hw->mac.type, i);
5697 if (j == conf->conf.queue_num)
5699 reta = (reta << 8) | conf->conf.queue[j];
5701 IXGBE_WRITE_REG(hw, reta_reg,
5705 /* Configure the RSS key and the RSS protocols used to compute
5706 * the RSS hash of input packets.
5708 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
5709 ixgbe_rss_disable(dev);
5712 if (rss_conf.rss_key == NULL)
5713 rss_conf.rss_key = rss_intel_key; /* Default hash key */
5714 ixgbe_hw_rss_hash_set(hw, &rss_conf);
5716 if (ixgbe_rss_conf_init(&filter_info->rss_info, &conf->conf))
5722 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
5724 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
5730 ixgbe_recv_pkts_vec(
5731 void __rte_unused *rx_queue,
5732 struct rte_mbuf __rte_unused **rx_pkts,
5733 uint16_t __rte_unused nb_pkts)
5739 ixgbe_recv_scattered_pkts_vec(
5740 void __rte_unused *rx_queue,
5741 struct rte_mbuf __rte_unused **rx_pkts,
5742 uint16_t __rte_unused nb_pkts)
5748 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)