4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5 * Copyright 2014 6WIND S.A.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * * Neither the name of Intel Corporation nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <sys/queue.h>
46 #include <rte_byteorder.h>
47 #include <rte_common.h>
48 #include <rte_cycles.h>
50 #include <rte_debug.h>
51 #include <rte_interrupts.h>
53 #include <rte_memory.h>
54 #include <rte_memzone.h>
55 #include <rte_launch.h>
57 #include <rte_per_lcore.h>
58 #include <rte_lcore.h>
59 #include <rte_atomic.h>
60 #include <rte_branch_prediction.h>
61 #include <rte_mempool.h>
62 #include <rte_malloc.h>
64 #include <rte_ether.h>
65 #include <rte_ethdev.h>
66 #include <rte_prefetch.h>
70 #include <rte_string_fns.h>
71 #include <rte_errno.h>
75 #include "ixgbe_logs.h"
76 #include "base/ixgbe_api.h"
77 #include "base/ixgbe_vf.h"
78 #include "ixgbe_ethdev.h"
79 #include "base/ixgbe_dcb.h"
80 #include "base/ixgbe_common.h"
81 #include "ixgbe_rxtx.h"
83 #ifdef RTE_LIBRTE_IEEE1588
84 #define IXGBE_TX_IEEE1588_TMST PKT_TX_IEEE1588_TMST
86 #define IXGBE_TX_IEEE1588_TMST 0
88 /* Bit Mask to indicate what bits required for building TX context */
89 #define IXGBE_TX_OFFLOAD_MASK ( \
95 PKT_TX_OUTER_IP_CKSUM | \
96 IXGBE_TX_IEEE1588_TMST)
98 #define IXGBE_TX_OFFLOAD_NOTSUP_MASK \
99 (PKT_TX_OFFLOAD_MASK ^ IXGBE_TX_OFFLOAD_MASK)
102 #define RTE_PMD_USE_PREFETCH
105 #ifdef RTE_PMD_USE_PREFETCH
107 * Prefetch a cache line into all cache levels.
109 #define rte_ixgbe_prefetch(p) rte_prefetch0(p)
111 #define rte_ixgbe_prefetch(p) do {} while (0)
114 /*********************************************************************
118 **********************************************************************/
121 * Check for descriptors with their DD bit set and free mbufs.
122 * Return the total number of buffers freed.
124 static inline int __attribute__((always_inline))
125 ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
127 struct ixgbe_tx_entry *txep;
130 struct rte_mbuf *m, *free[RTE_IXGBE_TX_MAX_FREE_BUF_SZ];
132 /* check DD bit on threshold descriptor */
133 status = txq->tx_ring[txq->tx_next_dd].wb.status;
134 if (!(status & rte_cpu_to_le_32(IXGBE_ADVTXD_STAT_DD)))
138 * first buffer to free from S/W ring is at index
139 * tx_next_dd - (tx_rs_thresh-1)
141 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
143 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
144 /* free buffers one at a time */
145 m = __rte_pktmbuf_prefree_seg(txep->mbuf);
148 if (unlikely(m == NULL))
151 if (nb_free >= RTE_IXGBE_TX_MAX_FREE_BUF_SZ ||
152 (nb_free > 0 && m->pool != free[0]->pool)) {
153 rte_mempool_put_bulk(free[0]->pool,
154 (void **)free, nb_free);
162 rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
164 /* buffers were freed, update counters */
165 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
166 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
167 if (txq->tx_next_dd >= txq->nb_tx_desc)
168 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
170 return txq->tx_rs_thresh;
173 /* Populate 4 descriptors with data from 4 mbufs */
175 tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
177 uint64_t buf_dma_addr;
181 for (i = 0; i < 4; ++i, ++txdp, ++pkts) {
182 buf_dma_addr = rte_mbuf_data_dma_addr(*pkts);
183 pkt_len = (*pkts)->data_len;
185 /* write data to descriptor */
186 txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
188 txdp->read.cmd_type_len =
189 rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
191 txdp->read.olinfo_status =
192 rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
194 rte_prefetch0(&(*pkts)->pool);
198 /* Populate 1 descriptor with data from 1 mbuf */
200 tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
202 uint64_t buf_dma_addr;
205 buf_dma_addr = rte_mbuf_data_dma_addr(*pkts);
206 pkt_len = (*pkts)->data_len;
208 /* write data to descriptor */
209 txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
210 txdp->read.cmd_type_len =
211 rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
212 txdp->read.olinfo_status =
213 rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
214 rte_prefetch0(&(*pkts)->pool);
218 * Fill H/W descriptor ring with mbuf data.
219 * Copy mbuf pointers to the S/W ring.
222 ixgbe_tx_fill_hw_ring(struct ixgbe_tx_queue *txq, struct rte_mbuf **pkts,
225 volatile union ixgbe_adv_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
226 struct ixgbe_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
227 const int N_PER_LOOP = 4;
228 const int N_PER_LOOP_MASK = N_PER_LOOP-1;
229 int mainpart, leftover;
233 * Process most of the packets in chunks of N pkts. Any
234 * leftover packets will get processed one at a time.
236 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
237 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
238 for (i = 0; i < mainpart; i += N_PER_LOOP) {
239 /* Copy N mbuf pointers to the S/W ring */
240 for (j = 0; j < N_PER_LOOP; ++j) {
241 (txep + i + j)->mbuf = *(pkts + i + j);
243 tx4(txdp + i, pkts + i);
246 if (unlikely(leftover > 0)) {
247 for (i = 0; i < leftover; ++i) {
248 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
249 tx1(txdp + mainpart + i, pkts + mainpart + i);
254 static inline uint16_t
255 tx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
258 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
259 volatile union ixgbe_adv_tx_desc *tx_r = txq->tx_ring;
263 * Begin scanning the H/W ring for done descriptors when the
264 * number of available descriptors drops below tx_free_thresh. For
265 * each done descriptor, free the associated buffer.
267 if (txq->nb_tx_free < txq->tx_free_thresh)
268 ixgbe_tx_free_bufs(txq);
270 /* Only use descriptors that are available */
271 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
272 if (unlikely(nb_pkts == 0))
275 /* Use exactly nb_pkts descriptors */
276 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
279 * At this point, we know there are enough descriptors in the
280 * ring to transmit all the packets. This assumes that each
281 * mbuf contains a single segment, and that no new offloads
282 * are expected, which would require a new context descriptor.
286 * See if we're going to wrap-around. If so, handle the top
287 * of the descriptor ring first, then do the bottom. If not,
288 * the processing looks just like the "bottom" part anyway...
290 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
291 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
292 ixgbe_tx_fill_hw_ring(txq, tx_pkts, n);
295 * We know that the last descriptor in the ring will need to
296 * have its RS bit set because tx_rs_thresh has to be
297 * a divisor of the ring size
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_rs_thresh - 1);
306 /* Fill H/W descriptor ring with mbuf data */
307 ixgbe_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
308 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
311 * Determine if RS bit should be set
312 * This is what we actually want:
313 * if ((txq->tx_tail - 1) >= txq->tx_next_rs)
314 * but instead of subtracting 1 and doing >=, we can just do
315 * greater than without subtracting.
317 if (txq->tx_tail > txq->tx_next_rs) {
318 tx_r[txq->tx_next_rs].read.cmd_type_len |=
319 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
320 txq->tx_next_rs = (uint16_t)(txq->tx_next_rs +
322 if (txq->tx_next_rs >= txq->nb_tx_desc)
323 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
327 * Check for wrap-around. This would only happen if we used
328 * up to the last descriptor in the ring, no more, no less.
330 if (txq->tx_tail >= txq->nb_tx_desc)
333 /* update tail pointer */
335 IXGBE_PCI_REG_WRITE_RELAXED(txq->tdt_reg_addr, txq->tx_tail);
341 ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
346 /* Try to transmit at least chunks of TX_MAX_BURST pkts */
347 if (likely(nb_pkts <= RTE_PMD_IXGBE_TX_MAX_BURST))
348 return tx_xmit_pkts(tx_queue, tx_pkts, nb_pkts);
350 /* transmit more than the max burst, in chunks of TX_MAX_BURST */
355 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_TX_MAX_BURST);
356 ret = tx_xmit_pkts(tx_queue, &(tx_pkts[nb_tx]), n);
357 nb_tx = (uint16_t)(nb_tx + ret);
358 nb_pkts = (uint16_t)(nb_pkts - ret);
367 ixgbe_set_xmit_ctx(struct ixgbe_tx_queue *txq,
368 volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
369 uint64_t ol_flags, union ixgbe_tx_offload tx_offload)
371 uint32_t type_tucmd_mlhl;
372 uint32_t mss_l4len_idx = 0;
374 uint32_t vlan_macip_lens;
375 union ixgbe_tx_offload tx_offload_mask;
376 uint32_t seqnum_seed = 0;
378 ctx_idx = txq->ctx_curr;
379 tx_offload_mask.data[0] = 0;
380 tx_offload_mask.data[1] = 0;
383 /* Specify which HW CTX to upload. */
384 mss_l4len_idx |= (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
386 if (ol_flags & PKT_TX_VLAN_PKT) {
387 tx_offload_mask.vlan_tci |= ~0;
390 /* check if TCP segmentation required for this packet */
391 if (ol_flags & PKT_TX_TCP_SEG) {
392 /* implies IP cksum in IPv4 */
393 if (ol_flags & PKT_TX_IP_CKSUM)
394 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4 |
395 IXGBE_ADVTXD_TUCMD_L4T_TCP |
396 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
398 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV6 |
399 IXGBE_ADVTXD_TUCMD_L4T_TCP |
400 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
402 tx_offload_mask.l2_len |= ~0;
403 tx_offload_mask.l3_len |= ~0;
404 tx_offload_mask.l4_len |= ~0;
405 tx_offload_mask.tso_segsz |= ~0;
406 mss_l4len_idx |= tx_offload.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT;
407 mss_l4len_idx |= tx_offload.l4_len << IXGBE_ADVTXD_L4LEN_SHIFT;
408 } else { /* no TSO, check if hardware checksum is needed */
409 if (ol_flags & PKT_TX_IP_CKSUM) {
410 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
411 tx_offload_mask.l2_len |= ~0;
412 tx_offload_mask.l3_len |= ~0;
415 switch (ol_flags & PKT_TX_L4_MASK) {
416 case PKT_TX_UDP_CKSUM:
417 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
418 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
419 mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
420 tx_offload_mask.l2_len |= ~0;
421 tx_offload_mask.l3_len |= ~0;
423 case PKT_TX_TCP_CKSUM:
424 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
425 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
426 mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
427 tx_offload_mask.l2_len |= ~0;
428 tx_offload_mask.l3_len |= ~0;
430 case PKT_TX_SCTP_CKSUM:
431 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
432 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
433 mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
434 tx_offload_mask.l2_len |= ~0;
435 tx_offload_mask.l3_len |= ~0;
438 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
439 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
444 if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
445 tx_offload_mask.outer_l2_len |= ~0;
446 tx_offload_mask.outer_l3_len |= ~0;
447 tx_offload_mask.l2_len |= ~0;
448 seqnum_seed |= tx_offload.outer_l3_len
449 << IXGBE_ADVTXD_OUTER_IPLEN;
450 seqnum_seed |= tx_offload.l2_len
451 << IXGBE_ADVTXD_TUNNEL_LEN;
454 txq->ctx_cache[ctx_idx].flags = ol_flags;
455 txq->ctx_cache[ctx_idx].tx_offload.data[0] =
456 tx_offload_mask.data[0] & tx_offload.data[0];
457 txq->ctx_cache[ctx_idx].tx_offload.data[1] =
458 tx_offload_mask.data[1] & tx_offload.data[1];
459 txq->ctx_cache[ctx_idx].tx_offload_mask = tx_offload_mask;
461 ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
462 vlan_macip_lens = tx_offload.l3_len;
463 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
464 vlan_macip_lens |= (tx_offload.outer_l2_len <<
465 IXGBE_ADVTXD_MACLEN_SHIFT);
467 vlan_macip_lens |= (tx_offload.l2_len <<
468 IXGBE_ADVTXD_MACLEN_SHIFT);
469 vlan_macip_lens |= ((uint32_t)tx_offload.vlan_tci << IXGBE_ADVTXD_VLAN_SHIFT);
470 ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
471 ctx_txd->mss_l4len_idx = rte_cpu_to_le_32(mss_l4len_idx);
472 ctx_txd->seqnum_seed = seqnum_seed;
476 * Check which hardware context can be used. Use the existing match
477 * or create a new context descriptor.
479 static inline uint32_t
480 what_advctx_update(struct ixgbe_tx_queue *txq, uint64_t flags,
481 union ixgbe_tx_offload tx_offload)
483 /* If match with the current used context */
484 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
485 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[0] ==
486 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[0]
487 & tx_offload.data[0])) &&
488 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[1] ==
489 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[1]
490 & tx_offload.data[1]))))
491 return txq->ctx_curr;
493 /* What if match with the next context */
495 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
496 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[0] ==
497 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[0]
498 & tx_offload.data[0])) &&
499 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[1] ==
500 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[1]
501 & tx_offload.data[1]))))
502 return txq->ctx_curr;
504 /* Mismatch, use the previous context */
505 return IXGBE_CTX_NUM;
508 static inline uint32_t
509 tx_desc_cksum_flags_to_olinfo(uint64_t ol_flags)
513 if ((ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM)
514 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
515 if (ol_flags & PKT_TX_IP_CKSUM)
516 tmp |= IXGBE_ADVTXD_POPTS_IXSM;
517 if (ol_flags & PKT_TX_TCP_SEG)
518 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
522 static inline uint32_t
523 tx_desc_ol_flags_to_cmdtype(uint64_t ol_flags)
525 uint32_t cmdtype = 0;
527 if (ol_flags & PKT_TX_VLAN_PKT)
528 cmdtype |= IXGBE_ADVTXD_DCMD_VLE;
529 if (ol_flags & PKT_TX_TCP_SEG)
530 cmdtype |= IXGBE_ADVTXD_DCMD_TSE;
531 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
532 cmdtype |= (1 << IXGBE_ADVTXD_OUTERIPCS_SHIFT);
533 if (ol_flags & PKT_TX_MACSEC)
534 cmdtype |= IXGBE_ADVTXD_MAC_LINKSEC;
538 /* Default RS bit threshold values */
539 #ifndef DEFAULT_TX_RS_THRESH
540 #define DEFAULT_TX_RS_THRESH 32
542 #ifndef DEFAULT_TX_FREE_THRESH
543 #define DEFAULT_TX_FREE_THRESH 32
546 /* Reset transmit descriptors after they have been used */
548 ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
550 struct ixgbe_tx_entry *sw_ring = txq->sw_ring;
551 volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
552 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
553 uint16_t nb_tx_desc = txq->nb_tx_desc;
554 uint16_t desc_to_clean_to;
555 uint16_t nb_tx_to_clean;
558 /* Determine the last descriptor needing to be cleaned */
559 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
560 if (desc_to_clean_to >= nb_tx_desc)
561 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
563 /* Check to make sure the last descriptor to clean is done */
564 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
565 status = txr[desc_to_clean_to].wb.status;
566 if (!(status & rte_cpu_to_le_32(IXGBE_TXD_STAT_DD))) {
567 PMD_TX_FREE_LOG(DEBUG,
568 "TX descriptor %4u is not done"
569 "(port=%d queue=%d)",
571 txq->port_id, txq->queue_id);
572 /* Failed to clean any descriptors, better luck next time */
576 /* Figure out how many descriptors will be cleaned */
577 if (last_desc_cleaned > desc_to_clean_to)
578 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
581 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
584 PMD_TX_FREE_LOG(DEBUG,
585 "Cleaning %4u TX descriptors: %4u to %4u "
586 "(port=%d queue=%d)",
587 nb_tx_to_clean, last_desc_cleaned, desc_to_clean_to,
588 txq->port_id, txq->queue_id);
591 * The last descriptor to clean is done, so that means all the
592 * descriptors from the last descriptor that was cleaned
593 * up to the last descriptor with the RS bit set
594 * are done. Only reset the threshold descriptor.
596 txr[desc_to_clean_to].wb.status = 0;
598 /* Update the txq to reflect the last descriptor that was cleaned */
599 txq->last_desc_cleaned = desc_to_clean_to;
600 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
607 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
610 struct ixgbe_tx_queue *txq;
611 struct ixgbe_tx_entry *sw_ring;
612 struct ixgbe_tx_entry *txe, *txn;
613 volatile union ixgbe_adv_tx_desc *txr;
614 volatile union ixgbe_adv_tx_desc *txd, *txp;
615 struct rte_mbuf *tx_pkt;
616 struct rte_mbuf *m_seg;
617 uint64_t buf_dma_addr;
618 uint32_t olinfo_status;
619 uint32_t cmd_type_len;
630 union ixgbe_tx_offload tx_offload;
632 tx_offload.data[0] = 0;
633 tx_offload.data[1] = 0;
635 sw_ring = txq->sw_ring;
637 tx_id = txq->tx_tail;
638 txe = &sw_ring[tx_id];
641 /* Determine if the descriptor ring needs to be cleaned. */
642 if (txq->nb_tx_free < txq->tx_free_thresh)
643 ixgbe_xmit_cleanup(txq);
645 rte_prefetch0(&txe->mbuf->pool);
648 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
651 pkt_len = tx_pkt->pkt_len;
654 * Determine how many (if any) context descriptors
655 * are needed for offload functionality.
657 ol_flags = tx_pkt->ol_flags;
659 /* If hardware offload required */
660 tx_ol_req = ol_flags & IXGBE_TX_OFFLOAD_MASK;
662 tx_offload.l2_len = tx_pkt->l2_len;
663 tx_offload.l3_len = tx_pkt->l3_len;
664 tx_offload.l4_len = tx_pkt->l4_len;
665 tx_offload.vlan_tci = tx_pkt->vlan_tci;
666 tx_offload.tso_segsz = tx_pkt->tso_segsz;
667 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
668 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
670 /* If new context need be built or reuse the exist ctx. */
671 ctx = what_advctx_update(txq, tx_ol_req,
673 /* Only allocate context descriptor if required*/
674 new_ctx = (ctx == IXGBE_CTX_NUM);
679 * Keep track of how many descriptors are used this loop
680 * This will always be the number of segments + the number of
681 * Context descriptors required to transmit the packet
683 nb_used = (uint16_t)(tx_pkt->nb_segs + new_ctx);
686 nb_used + txq->nb_tx_used >= txq->tx_rs_thresh)
687 /* set RS on the previous packet in the burst */
688 txp->read.cmd_type_len |=
689 rte_cpu_to_le_32(IXGBE_TXD_CMD_RS);
692 * The number of descriptors that must be allocated for a
693 * packet is the number of segments of that packet, plus 1
694 * Context Descriptor for the hardware offload, if any.
695 * Determine the last TX descriptor to allocate in the TX ring
696 * for the packet, starting from the current position (tx_id)
699 tx_last = (uint16_t) (tx_id + nb_used - 1);
702 if (tx_last >= txq->nb_tx_desc)
703 tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
705 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
706 " tx_first=%u tx_last=%u",
707 (unsigned) txq->port_id,
708 (unsigned) txq->queue_id,
714 * Make sure there are enough TX descriptors available to
715 * transmit the entire packet.
716 * nb_used better be less than or equal to txq->tx_rs_thresh
718 if (nb_used > txq->nb_tx_free) {
719 PMD_TX_FREE_LOG(DEBUG,
720 "Not enough free TX descriptors "
721 "nb_used=%4u nb_free=%4u "
722 "(port=%d queue=%d)",
723 nb_used, txq->nb_tx_free,
724 txq->port_id, txq->queue_id);
726 if (ixgbe_xmit_cleanup(txq) != 0) {
727 /* Could not clean any descriptors */
733 /* nb_used better be <= txq->tx_rs_thresh */
734 if (unlikely(nb_used > txq->tx_rs_thresh)) {
735 PMD_TX_FREE_LOG(DEBUG,
736 "The number of descriptors needed to "
737 "transmit the packet exceeds the "
738 "RS bit threshold. This will impact "
740 "nb_used=%4u nb_free=%4u "
742 "(port=%d queue=%d)",
743 nb_used, txq->nb_tx_free,
745 txq->port_id, txq->queue_id);
747 * Loop here until there are enough TX
748 * descriptors or until the ring cannot be
751 while (nb_used > txq->nb_tx_free) {
752 if (ixgbe_xmit_cleanup(txq) != 0) {
754 * Could not clean any
766 * By now there are enough free TX descriptors to transmit
771 * Set common flags of all TX Data Descriptors.
773 * The following bits must be set in all Data Descriptors:
774 * - IXGBE_ADVTXD_DTYP_DATA
775 * - IXGBE_ADVTXD_DCMD_DEXT
777 * The following bits must be set in the first Data Descriptor
778 * and are ignored in the other ones:
779 * - IXGBE_ADVTXD_DCMD_IFCS
780 * - IXGBE_ADVTXD_MAC_1588
781 * - IXGBE_ADVTXD_DCMD_VLE
783 * The following bits must only be set in the last Data
785 * - IXGBE_TXD_CMD_EOP
787 * The following bits can be set in any Data Descriptor, but
788 * are only set in the last Data Descriptor:
791 cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
792 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
794 #ifdef RTE_LIBRTE_IEEE1588
795 if (ol_flags & PKT_TX_IEEE1588_TMST)
796 cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
802 if (ol_flags & PKT_TX_TCP_SEG) {
803 /* when TSO is on, paylen in descriptor is the
804 * not the packet len but the tcp payload len */
805 pkt_len -= (tx_offload.l2_len +
806 tx_offload.l3_len + tx_offload.l4_len);
810 * Setup the TX Advanced Context Descriptor if required
813 volatile struct ixgbe_adv_tx_context_desc *
816 ctx_txd = (volatile struct
817 ixgbe_adv_tx_context_desc *)
820 txn = &sw_ring[txe->next_id];
821 rte_prefetch0(&txn->mbuf->pool);
823 if (txe->mbuf != NULL) {
824 rte_pktmbuf_free_seg(txe->mbuf);
828 ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
831 txe->last_id = tx_last;
832 tx_id = txe->next_id;
837 * Setup the TX Advanced Data Descriptor,
838 * This path will go through
839 * whatever new/reuse the context descriptor
841 cmd_type_len |= tx_desc_ol_flags_to_cmdtype(ol_flags);
842 olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
843 olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
846 olinfo_status |= (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
851 txn = &sw_ring[txe->next_id];
852 rte_prefetch0(&txn->mbuf->pool);
854 if (txe->mbuf != NULL)
855 rte_pktmbuf_free_seg(txe->mbuf);
859 * Set up Transmit Data Descriptor.
861 slen = m_seg->data_len;
862 buf_dma_addr = rte_mbuf_data_dma_addr(m_seg);
863 txd->read.buffer_addr =
864 rte_cpu_to_le_64(buf_dma_addr);
865 txd->read.cmd_type_len =
866 rte_cpu_to_le_32(cmd_type_len | slen);
867 txd->read.olinfo_status =
868 rte_cpu_to_le_32(olinfo_status);
869 txe->last_id = tx_last;
870 tx_id = txe->next_id;
873 } while (m_seg != NULL);
876 * The last packet data descriptor needs End Of Packet (EOP)
878 cmd_type_len |= IXGBE_TXD_CMD_EOP;
879 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
880 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
882 /* Set RS bit only on threshold packets' last descriptor */
883 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
884 PMD_TX_FREE_LOG(DEBUG,
885 "Setting RS bit on TXD id="
886 "%4u (port=%d queue=%d)",
887 tx_last, txq->port_id, txq->queue_id);
889 cmd_type_len |= IXGBE_TXD_CMD_RS;
891 /* Update txq RS bit counters */
897 txd->read.cmd_type_len |= rte_cpu_to_le_32(cmd_type_len);
901 /* set RS on last packet in the burst */
903 txp->read.cmd_type_len |= rte_cpu_to_le_32(IXGBE_TXD_CMD_RS);
908 * Set the Transmit Descriptor Tail (TDT)
910 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
911 (unsigned) txq->port_id, (unsigned) txq->queue_id,
912 (unsigned) tx_id, (unsigned) nb_tx);
913 IXGBE_PCI_REG_WRITE_RELAXED(txq->tdt_reg_addr, tx_id);
914 txq->tx_tail = tx_id;
919 /*********************************************************************
923 **********************************************************************/
925 ixgbe_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
930 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
932 for (i = 0; i < nb_pkts; i++) {
934 ol_flags = m->ol_flags;
937 * Check if packet meets requirements for number of segments
939 * NOTE: for ixgbe it's always (40 - WTHRESH) for both TSO and
943 if (m->nb_segs > IXGBE_TX_MAX_SEG - txq->wthresh) {
948 if (ol_flags & IXGBE_TX_OFFLOAD_NOTSUP_MASK) {
949 rte_errno = -ENOTSUP;
953 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
954 ret = rte_validate_tx_offload(m);
960 ret = rte_net_intel_cksum_prepare(m);
970 /*********************************************************************
974 **********************************************************************/
976 #define IXGBE_PACKET_TYPE_ETHER 0X00
977 #define IXGBE_PACKET_TYPE_IPV4 0X01
978 #define IXGBE_PACKET_TYPE_IPV4_TCP 0X11
979 #define IXGBE_PACKET_TYPE_IPV4_UDP 0X21
980 #define IXGBE_PACKET_TYPE_IPV4_SCTP 0X41
981 #define IXGBE_PACKET_TYPE_IPV4_EXT 0X03
982 #define IXGBE_PACKET_TYPE_IPV4_EXT_TCP 0X13
983 #define IXGBE_PACKET_TYPE_IPV4_EXT_UDP 0X23
984 #define IXGBE_PACKET_TYPE_IPV4_EXT_SCTP 0X43
985 #define IXGBE_PACKET_TYPE_IPV6 0X04
986 #define IXGBE_PACKET_TYPE_IPV6_TCP 0X14
987 #define IXGBE_PACKET_TYPE_IPV6_UDP 0X24
988 #define IXGBE_PACKET_TYPE_IPV6_SCTP 0X44
989 #define IXGBE_PACKET_TYPE_IPV6_EXT 0X0C
990 #define IXGBE_PACKET_TYPE_IPV6_EXT_TCP 0X1C
991 #define IXGBE_PACKET_TYPE_IPV6_EXT_UDP 0X2C
992 #define IXGBE_PACKET_TYPE_IPV6_EXT_SCTP 0X4C
993 #define IXGBE_PACKET_TYPE_IPV4_IPV6 0X05
994 #define IXGBE_PACKET_TYPE_IPV4_IPV6_TCP 0X15
995 #define IXGBE_PACKET_TYPE_IPV4_IPV6_UDP 0X25
996 #define IXGBE_PACKET_TYPE_IPV4_IPV6_SCTP 0X45
997 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6 0X07
998 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_TCP 0X17
999 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_UDP 0X27
1000 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_SCTP 0X47
1001 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT 0X0D
1002 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP 0X1D
1003 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP 0X2D
1004 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_SCTP 0X4D
1005 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT 0X0F
1006 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_TCP 0X1F
1007 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_UDP 0X2F
1008 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_SCTP 0X4F
1010 #define IXGBE_PACKET_TYPE_NVGRE 0X00
1011 #define IXGBE_PACKET_TYPE_NVGRE_IPV4 0X01
1012 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_TCP 0X11
1013 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_UDP 0X21
1014 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_SCTP 0X41
1015 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT 0X03
1016 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_TCP 0X13
1017 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_UDP 0X23
1018 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_SCTP 0X43
1019 #define IXGBE_PACKET_TYPE_NVGRE_IPV6 0X04
1020 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_TCP 0X14
1021 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_UDP 0X24
1022 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_SCTP 0X44
1023 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT 0X0C
1024 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_TCP 0X1C
1025 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_UDP 0X2C
1026 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_SCTP 0X4C
1027 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6 0X05
1028 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_TCP 0X15
1029 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_UDP 0X25
1030 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT 0X0D
1031 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_TCP 0X1D
1032 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_UDP 0X2D
1034 #define IXGBE_PACKET_TYPE_VXLAN 0X80
1035 #define IXGBE_PACKET_TYPE_VXLAN_IPV4 0X81
1036 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_TCP 0x91
1037 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_UDP 0xA1
1038 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_SCTP 0xC1
1039 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT 0x83
1040 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_TCP 0X93
1041 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_UDP 0XA3
1042 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_SCTP 0XC3
1043 #define IXGBE_PACKET_TYPE_VXLAN_IPV6 0X84
1044 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_TCP 0X94
1045 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_UDP 0XA4
1046 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_SCTP 0XC4
1047 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT 0X8C
1048 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_TCP 0X9C
1049 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_UDP 0XAC
1050 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_SCTP 0XCC
1051 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6 0X85
1052 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_TCP 0X95
1053 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_UDP 0XA5
1054 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT 0X8D
1055 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_TCP 0X9D
1056 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_UDP 0XAD
1058 #define IXGBE_PACKET_TYPE_MAX 0X80
1059 #define IXGBE_PACKET_TYPE_TN_MAX 0X100
1060 #define IXGBE_PACKET_TYPE_SHIFT 0X04
1062 /* @note: fix ixgbe_dev_supported_ptypes_get() if any change here. */
1063 static inline uint32_t
1064 ixgbe_rxd_pkt_info_to_pkt_type(uint32_t pkt_info, uint16_t ptype_mask)
1067 * Use 2 different table for normal packet and tunnel packet
1068 * to save the space.
1070 static const uint32_t
1071 ptype_table[IXGBE_PACKET_TYPE_MAX] __rte_cache_aligned = {
1072 [IXGBE_PACKET_TYPE_ETHER] = RTE_PTYPE_L2_ETHER,
1073 [IXGBE_PACKET_TYPE_IPV4] = RTE_PTYPE_L2_ETHER |
1075 [IXGBE_PACKET_TYPE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1076 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
1077 [IXGBE_PACKET_TYPE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1078 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
1079 [IXGBE_PACKET_TYPE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1080 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP,
1081 [IXGBE_PACKET_TYPE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1082 RTE_PTYPE_L3_IPV4_EXT,
1083 [IXGBE_PACKET_TYPE_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1084 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP,
1085 [IXGBE_PACKET_TYPE_IPV4_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1086 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP,
1087 [IXGBE_PACKET_TYPE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1088 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_SCTP,
1089 [IXGBE_PACKET_TYPE_IPV6] = RTE_PTYPE_L2_ETHER |
1091 [IXGBE_PACKET_TYPE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1092 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
1093 [IXGBE_PACKET_TYPE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1094 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
1095 [IXGBE_PACKET_TYPE_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1096 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP,
1097 [IXGBE_PACKET_TYPE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1098 RTE_PTYPE_L3_IPV6_EXT,
1099 [IXGBE_PACKET_TYPE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1100 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP,
1101 [IXGBE_PACKET_TYPE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1102 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP,
1103 [IXGBE_PACKET_TYPE_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1104 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_SCTP,
1105 [IXGBE_PACKET_TYPE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1106 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1107 RTE_PTYPE_INNER_L3_IPV6,
1108 [IXGBE_PACKET_TYPE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1109 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1110 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1111 [IXGBE_PACKET_TYPE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1112 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1113 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1114 [IXGBE_PACKET_TYPE_IPV4_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1115 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1116 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1117 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6] = RTE_PTYPE_L2_ETHER |
1118 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1119 RTE_PTYPE_INNER_L3_IPV6,
1120 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1121 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1122 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1123 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1124 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1125 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1126 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1127 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1128 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1129 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1130 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1131 RTE_PTYPE_INNER_L3_IPV6_EXT,
1132 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1133 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1134 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1135 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1136 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1137 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1138 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1139 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1140 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1141 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1142 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1143 RTE_PTYPE_INNER_L3_IPV6_EXT,
1144 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1145 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1146 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1147 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1148 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1149 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1150 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_SCTP] =
1151 RTE_PTYPE_L2_ETHER |
1152 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1153 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1156 static const uint32_t
1157 ptype_table_tn[IXGBE_PACKET_TYPE_TN_MAX] __rte_cache_aligned = {
1158 [IXGBE_PACKET_TYPE_NVGRE] = RTE_PTYPE_L2_ETHER |
1159 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1160 RTE_PTYPE_INNER_L2_ETHER,
1161 [IXGBE_PACKET_TYPE_NVGRE_IPV4] = RTE_PTYPE_L2_ETHER |
1162 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1163 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1164 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1165 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1166 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT,
1167 [IXGBE_PACKET_TYPE_NVGRE_IPV6] = RTE_PTYPE_L2_ETHER |
1168 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1169 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6,
1170 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1171 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1172 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1173 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1174 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1175 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT,
1176 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1177 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1178 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1179 [IXGBE_PACKET_TYPE_NVGRE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1180 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1181 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1182 RTE_PTYPE_INNER_L4_TCP,
1183 [IXGBE_PACKET_TYPE_NVGRE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1184 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1185 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1186 RTE_PTYPE_INNER_L4_TCP,
1187 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1188 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1189 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1190 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1191 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1192 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1193 RTE_PTYPE_INNER_L4_TCP,
1194 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_TCP] =
1195 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1196 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_INNER_L2_ETHER |
1197 RTE_PTYPE_INNER_L3_IPV4,
1198 [IXGBE_PACKET_TYPE_NVGRE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1199 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1200 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1201 RTE_PTYPE_INNER_L4_UDP,
1202 [IXGBE_PACKET_TYPE_NVGRE_IPV6_UDP] = 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_IPV6 |
1205 RTE_PTYPE_INNER_L4_UDP,
1206 [IXGBE_PACKET_TYPE_NVGRE_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1207 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1208 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1209 RTE_PTYPE_INNER_L4_SCTP,
1210 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1211 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1212 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1213 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1214 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1215 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1216 RTE_PTYPE_INNER_L4_UDP,
1217 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1218 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1219 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1220 RTE_PTYPE_INNER_L4_SCTP,
1221 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_UDP] =
1222 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1223 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_INNER_L2_ETHER |
1224 RTE_PTYPE_INNER_L3_IPV4,
1225 [IXGBE_PACKET_TYPE_NVGRE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1226 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1227 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1228 RTE_PTYPE_INNER_L4_SCTP,
1229 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1230 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1231 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1232 RTE_PTYPE_INNER_L4_SCTP,
1233 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1234 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1235 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1236 RTE_PTYPE_INNER_L4_TCP,
1237 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1238 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1239 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1240 RTE_PTYPE_INNER_L4_UDP,
1242 [IXGBE_PACKET_TYPE_VXLAN] = RTE_PTYPE_L2_ETHER |
1243 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1244 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER,
1245 [IXGBE_PACKET_TYPE_VXLAN_IPV4] = RTE_PTYPE_L2_ETHER |
1246 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1247 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1248 RTE_PTYPE_INNER_L3_IPV4,
1249 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1250 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1251 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1252 RTE_PTYPE_INNER_L3_IPV4_EXT,
1253 [IXGBE_PACKET_TYPE_VXLAN_IPV6] = RTE_PTYPE_L2_ETHER |
1254 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1255 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1256 RTE_PTYPE_INNER_L3_IPV6,
1257 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1258 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1259 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1260 RTE_PTYPE_INNER_L3_IPV4,
1261 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1262 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1263 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1264 RTE_PTYPE_INNER_L3_IPV6_EXT,
1265 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1266 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1267 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1268 RTE_PTYPE_INNER_L3_IPV4,
1269 [IXGBE_PACKET_TYPE_VXLAN_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1270 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1271 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1272 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_TCP,
1273 [IXGBE_PACKET_TYPE_VXLAN_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1274 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1275 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1276 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1277 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_TCP] = 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_IPV6_EXT_TCP] = 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_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1285 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_TCP] =
1286 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1287 RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN |
1288 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1289 [IXGBE_PACKET_TYPE_VXLAN_IPV4_UDP] = 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 | RTE_PTYPE_INNER_L4_UDP,
1293 [IXGBE_PACKET_TYPE_VXLAN_IPV6_UDP] = 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 | RTE_PTYPE_INNER_L4_UDP,
1297 [IXGBE_PACKET_TYPE_VXLAN_IPV6_SCTP] = 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_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1301 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_UDP] = 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,
1305 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_UDP] = 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_EXT | RTE_PTYPE_INNER_L4_UDP,
1309 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_SCTP] = 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_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1313 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_UDP] =
1314 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1315 RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN |
1316 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1317 [IXGBE_PACKET_TYPE_VXLAN_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1318 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1319 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1320 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_SCTP,
1321 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_SCTP] = 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_EXT | RTE_PTYPE_INNER_L4_SCTP,
1325 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_TCP] = 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_IPV4_EXT | RTE_PTYPE_INNER_L4_TCP,
1329 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_UDP] = 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_IPV4_EXT | RTE_PTYPE_INNER_L4_UDP,
1335 if (unlikely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
1336 return RTE_PTYPE_UNKNOWN;
1338 pkt_info = (pkt_info >> IXGBE_PACKET_TYPE_SHIFT) & ptype_mask;
1340 /* For tunnel packet */
1341 if (pkt_info & IXGBE_PACKET_TYPE_TUNNEL_BIT) {
1342 /* Remove the tunnel bit to save the space. */
1343 pkt_info &= IXGBE_PACKET_TYPE_MASK_TUNNEL;
1344 return ptype_table_tn[pkt_info];
1348 * For x550, if it's not tunnel,
1349 * tunnel type bit should be set to 0.
1350 * Reuse 82599's mask.
1352 pkt_info &= IXGBE_PACKET_TYPE_MASK_82599;
1354 return ptype_table[pkt_info];
1357 static inline uint64_t
1358 ixgbe_rxd_pkt_info_to_pkt_flags(uint16_t pkt_info)
1360 static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
1361 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
1362 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
1363 PKT_RX_RSS_HASH, 0, 0, 0,
1364 0, 0, 0, PKT_RX_FDIR,
1366 #ifdef RTE_LIBRTE_IEEE1588
1367 static uint64_t ip_pkt_etqf_map[8] = {
1368 0, 0, 0, PKT_RX_IEEE1588_PTP,
1372 if (likely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
1373 return ip_pkt_etqf_map[(pkt_info >> 4) & 0X07] |
1374 ip_rss_types_map[pkt_info & 0XF];
1376 return ip_rss_types_map[pkt_info & 0XF];
1378 return ip_rss_types_map[pkt_info & 0XF];
1382 static inline uint64_t
1383 rx_desc_status_to_pkt_flags(uint32_t rx_status, uint64_t vlan_flags)
1388 * Check if VLAN present only.
1389 * Do not check whether L3/L4 rx checksum done by NIC or not,
1390 * That can be found from rte_eth_rxmode.hw_ip_checksum flag
1392 pkt_flags = (rx_status & IXGBE_RXD_STAT_VP) ? vlan_flags : 0;
1394 #ifdef RTE_LIBRTE_IEEE1588
1395 if (rx_status & IXGBE_RXD_STAT_TMST)
1396 pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
1401 static inline uint64_t
1402 rx_desc_error_to_pkt_flags(uint32_t rx_status)
1407 * Bit 31: IPE, IPv4 checksum error
1408 * Bit 30: L4I, L4I integrity error
1410 static uint64_t error_to_pkt_flags_map[4] = {
1411 PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD,
1412 PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_BAD,
1413 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_GOOD,
1414 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1416 pkt_flags = error_to_pkt_flags_map[(rx_status >>
1417 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1419 if ((rx_status & IXGBE_RXD_STAT_OUTERIPCS) &&
1420 (rx_status & IXGBE_RXDADV_ERR_OUTERIPER)) {
1421 pkt_flags |= PKT_RX_EIP_CKSUM_BAD;
1428 * LOOK_AHEAD defines how many desc statuses to check beyond the
1429 * current descriptor.
1430 * It must be a pound define for optimal performance.
1431 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1432 * function only works with LOOK_AHEAD=8.
1434 #define LOOK_AHEAD 8
1435 #if (LOOK_AHEAD != 8)
1436 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1439 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1441 volatile union ixgbe_adv_rx_desc *rxdp;
1442 struct ixgbe_rx_entry *rxep;
1443 struct rte_mbuf *mb;
1447 uint32_t s[LOOK_AHEAD];
1448 uint32_t pkt_info[LOOK_AHEAD];
1449 int i, j, nb_rx = 0;
1451 uint64_t vlan_flags = rxq->vlan_flags;
1453 /* get references to current descriptor and S/W ring entry */
1454 rxdp = &rxq->rx_ring[rxq->rx_tail];
1455 rxep = &rxq->sw_ring[rxq->rx_tail];
1457 status = rxdp->wb.upper.status_error;
1458 /* check to make sure there is at least 1 packet to receive */
1459 if (!(status & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1463 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1464 * reference packets that are ready to be received.
1466 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1467 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD) {
1468 /* Read desc statuses backwards to avoid race condition */
1469 for (j = 0; j < LOOK_AHEAD; j++)
1470 s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);
1474 /* Compute how many status bits were set */
1475 for (nb_dd = 0; nb_dd < LOOK_AHEAD &&
1476 (s[nb_dd] & IXGBE_RXDADV_STAT_DD); nb_dd++)
1479 for (j = 0; j < nb_dd; j++)
1480 pkt_info[j] = rte_le_to_cpu_32(rxdp[j].wb.lower.
1485 /* Translate descriptor info to mbuf format */
1486 for (j = 0; j < nb_dd; ++j) {
1488 pkt_len = rte_le_to_cpu_16(rxdp[j].wb.upper.length) -
1490 mb->data_len = pkt_len;
1491 mb->pkt_len = pkt_len;
1492 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1494 /* convert descriptor fields to rte mbuf flags */
1495 pkt_flags = rx_desc_status_to_pkt_flags(s[j],
1497 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1498 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags
1499 ((uint16_t)pkt_info[j]);
1500 mb->ol_flags = pkt_flags;
1502 ixgbe_rxd_pkt_info_to_pkt_type
1503 (pkt_info[j], rxq->pkt_type_mask);
1505 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1506 mb->hash.rss = rte_le_to_cpu_32(
1507 rxdp[j].wb.lower.hi_dword.rss);
1508 else if (pkt_flags & PKT_RX_FDIR) {
1509 mb->hash.fdir.hash = rte_le_to_cpu_16(
1510 rxdp[j].wb.lower.hi_dword.csum_ip.csum) &
1511 IXGBE_ATR_HASH_MASK;
1512 mb->hash.fdir.id = rte_le_to_cpu_16(
1513 rxdp[j].wb.lower.hi_dword.csum_ip.ip_id);
1517 /* Move mbuf pointers from the S/W ring to the stage */
1518 for (j = 0; j < LOOK_AHEAD; ++j) {
1519 rxq->rx_stage[i + j] = rxep[j].mbuf;
1522 /* stop if all requested packets could not be received */
1523 if (nb_dd != LOOK_AHEAD)
1527 /* clear software ring entries so we can cleanup correctly */
1528 for (i = 0; i < nb_rx; ++i) {
1529 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1537 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1539 volatile union ixgbe_adv_rx_desc *rxdp;
1540 struct ixgbe_rx_entry *rxep;
1541 struct rte_mbuf *mb;
1546 /* allocate buffers in bulk directly into the S/W ring */
1547 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1548 rxep = &rxq->sw_ring[alloc_idx];
1549 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1550 rxq->rx_free_thresh);
1551 if (unlikely(diag != 0))
1554 rxdp = &rxq->rx_ring[alloc_idx];
1555 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1556 /* populate the static rte mbuf fields */
1561 mb->port = rxq->port_id;
1564 rte_mbuf_refcnt_set(mb, 1);
1565 mb->data_off = RTE_PKTMBUF_HEADROOM;
1567 /* populate the descriptors */
1568 dma_addr = rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(mb));
1569 rxdp[i].read.hdr_addr = 0;
1570 rxdp[i].read.pkt_addr = dma_addr;
1573 /* update state of internal queue structure */
1574 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1575 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1576 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1582 static inline uint16_t
1583 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1586 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1589 /* how many packets are ready to return? */
1590 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1592 /* copy mbuf pointers to the application's packet list */
1593 for (i = 0; i < nb_pkts; ++i)
1594 rx_pkts[i] = stage[i];
1596 /* update internal queue state */
1597 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1598 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1603 static inline uint16_t
1604 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1607 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1610 /* Any previously recv'd pkts will be returned from the Rx stage */
1611 if (rxq->rx_nb_avail)
1612 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1614 /* Scan the H/W ring for packets to receive */
1615 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1617 /* update internal queue state */
1618 rxq->rx_next_avail = 0;
1619 rxq->rx_nb_avail = nb_rx;
1620 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1622 /* if required, allocate new buffers to replenish descriptors */
1623 if (rxq->rx_tail > rxq->rx_free_trigger) {
1624 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1626 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1629 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1630 "queue_id=%u", (unsigned) rxq->port_id,
1631 (unsigned) rxq->queue_id);
1633 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1634 rxq->rx_free_thresh;
1637 * Need to rewind any previous receives if we cannot
1638 * allocate new buffers to replenish the old ones.
1640 rxq->rx_nb_avail = 0;
1641 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1642 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1643 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1648 /* update tail pointer */
1650 IXGBE_PCI_REG_WRITE_RELAXED(rxq->rdt_reg_addr,
1654 if (rxq->rx_tail >= rxq->nb_rx_desc)
1657 /* received any packets this loop? */
1658 if (rxq->rx_nb_avail)
1659 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1664 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1666 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1671 if (unlikely(nb_pkts == 0))
1674 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1675 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1677 /* request is relatively large, chunk it up */
1682 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1683 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1684 nb_rx = (uint16_t)(nb_rx + ret);
1685 nb_pkts = (uint16_t)(nb_pkts - ret);
1694 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1697 struct ixgbe_rx_queue *rxq;
1698 volatile union ixgbe_adv_rx_desc *rx_ring;
1699 volatile union ixgbe_adv_rx_desc *rxdp;
1700 struct ixgbe_rx_entry *sw_ring;
1701 struct ixgbe_rx_entry *rxe;
1702 struct rte_mbuf *rxm;
1703 struct rte_mbuf *nmb;
1704 union ixgbe_adv_rx_desc rxd;
1713 uint64_t vlan_flags;
1718 rx_id = rxq->rx_tail;
1719 rx_ring = rxq->rx_ring;
1720 sw_ring = rxq->sw_ring;
1721 vlan_flags = rxq->vlan_flags;
1722 while (nb_rx < nb_pkts) {
1724 * The order of operations here is important as the DD status
1725 * bit must not be read after any other descriptor fields.
1726 * rx_ring and rxdp are pointing to volatile data so the order
1727 * of accesses cannot be reordered by the compiler. If they were
1728 * not volatile, they could be reordered which could lead to
1729 * using invalid descriptor fields when read from rxd.
1731 rxdp = &rx_ring[rx_id];
1732 staterr = rxdp->wb.upper.status_error;
1733 if (!(staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1740 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1741 * is likely to be invalid and to be dropped by the various
1742 * validation checks performed by the network stack.
1744 * Allocate a new mbuf to replenish the RX ring descriptor.
1745 * If the allocation fails:
1746 * - arrange for that RX descriptor to be the first one
1747 * being parsed the next time the receive function is
1748 * invoked [on the same queue].
1750 * - Stop parsing the RX ring and return immediately.
1752 * This policy do not drop the packet received in the RX
1753 * descriptor for which the allocation of a new mbuf failed.
1754 * Thus, it allows that packet to be later retrieved if
1755 * mbuf have been freed in the mean time.
1756 * As a side effect, holding RX descriptors instead of
1757 * systematically giving them back to the NIC may lead to
1758 * RX ring exhaustion situations.
1759 * However, the NIC can gracefully prevent such situations
1760 * to happen by sending specific "back-pressure" flow control
1761 * frames to its peer(s).
1763 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1764 "ext_err_stat=0x%08x pkt_len=%u",
1765 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1766 (unsigned) rx_id, (unsigned) staterr,
1767 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1769 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
1771 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1772 "queue_id=%u", (unsigned) rxq->port_id,
1773 (unsigned) rxq->queue_id);
1774 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1779 rxe = &sw_ring[rx_id];
1781 if (rx_id == rxq->nb_rx_desc)
1784 /* Prefetch next mbuf while processing current one. */
1785 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1788 * When next RX descriptor is on a cache-line boundary,
1789 * prefetch the next 4 RX descriptors and the next 8 pointers
1792 if ((rx_id & 0x3) == 0) {
1793 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1794 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1800 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(nmb));
1801 rxdp->read.hdr_addr = 0;
1802 rxdp->read.pkt_addr = dma_addr;
1805 * Initialize the returned mbuf.
1806 * 1) setup generic mbuf fields:
1807 * - number of segments,
1810 * - RX port identifier.
1811 * 2) integrate hardware offload data, if any:
1812 * - RSS flag & hash,
1813 * - IP checksum flag,
1814 * - VLAN TCI, if any,
1817 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1819 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1820 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1823 rxm->pkt_len = pkt_len;
1824 rxm->data_len = pkt_len;
1825 rxm->port = rxq->port_id;
1827 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1828 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1829 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1831 pkt_flags = rx_desc_status_to_pkt_flags(staterr, vlan_flags);
1832 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1833 pkt_flags = pkt_flags |
1834 ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1835 rxm->ol_flags = pkt_flags;
1837 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info,
1838 rxq->pkt_type_mask);
1840 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1841 rxm->hash.rss = rte_le_to_cpu_32(
1842 rxd.wb.lower.hi_dword.rss);
1843 else if (pkt_flags & PKT_RX_FDIR) {
1844 rxm->hash.fdir.hash = rte_le_to_cpu_16(
1845 rxd.wb.lower.hi_dword.csum_ip.csum) &
1846 IXGBE_ATR_HASH_MASK;
1847 rxm->hash.fdir.id = rte_le_to_cpu_16(
1848 rxd.wb.lower.hi_dword.csum_ip.ip_id);
1851 * Store the mbuf address into the next entry of the array
1852 * of returned packets.
1854 rx_pkts[nb_rx++] = rxm;
1856 rxq->rx_tail = rx_id;
1859 * If the number of free RX descriptors is greater than the RX free
1860 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1862 * Update the RDT with the value of the last processed RX descriptor
1863 * minus 1, to guarantee that the RDT register is never equal to the
1864 * RDH register, which creates a "full" ring situtation from the
1865 * hardware point of view...
1867 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1868 if (nb_hold > rxq->rx_free_thresh) {
1869 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1870 "nb_hold=%u nb_rx=%u",
1871 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1872 (unsigned) rx_id, (unsigned) nb_hold,
1874 rx_id = (uint16_t) ((rx_id == 0) ?
1875 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1876 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1879 rxq->nb_rx_hold = nb_hold;
1884 * Detect an RSC descriptor.
1886 static inline uint32_t
1887 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1889 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1890 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1894 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1896 * Fill the following info in the HEAD buffer of the Rx cluster:
1897 * - RX port identifier
1898 * - hardware offload data, if any:
1900 * - IP checksum flag
1901 * - VLAN TCI, if any
1903 * @head HEAD of the packet cluster
1904 * @desc HW descriptor to get data from
1905 * @rxq Pointer to the Rx queue
1908 ixgbe_fill_cluster_head_buf(
1909 struct rte_mbuf *head,
1910 union ixgbe_adv_rx_desc *desc,
1911 struct ixgbe_rx_queue *rxq,
1917 head->port = rxq->port_id;
1919 /* The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1920 * set in the pkt_flags field.
1922 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1923 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1924 pkt_flags = rx_desc_status_to_pkt_flags(staterr, rxq->vlan_flags);
1925 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1926 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1927 head->ol_flags = pkt_flags;
1929 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info, rxq->pkt_type_mask);
1931 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1932 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1933 else if (pkt_flags & PKT_RX_FDIR) {
1934 head->hash.fdir.hash =
1935 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1936 & IXGBE_ATR_HASH_MASK;
1937 head->hash.fdir.id =
1938 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1943 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1945 * @rx_queue Rx queue handle
1946 * @rx_pkts table of received packets
1947 * @nb_pkts size of rx_pkts table
1948 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1950 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1951 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1953 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1954 * 1) When non-EOP RSC completion arrives:
1955 * a) Update the HEAD of the current RSC aggregation cluster with the new
1956 * segment's data length.
1957 * b) Set the "next" pointer of the current segment to point to the segment
1958 * at the NEXTP index.
1959 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
1960 * in the sw_rsc_ring.
1961 * 2) When EOP arrives we just update the cluster's total length and offload
1962 * flags and deliver the cluster up to the upper layers. In our case - put it
1963 * in the rx_pkts table.
1965 * Returns the number of received packets/clusters (according to the "bulk
1966 * receive" interface).
1968 static inline uint16_t
1969 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
1972 struct ixgbe_rx_queue *rxq = rx_queue;
1973 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
1974 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
1975 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
1976 uint16_t rx_id = rxq->rx_tail;
1978 uint16_t nb_hold = rxq->nb_rx_hold;
1979 uint16_t prev_id = rxq->rx_tail;
1981 while (nb_rx < nb_pkts) {
1983 struct ixgbe_rx_entry *rxe;
1984 struct ixgbe_scattered_rx_entry *sc_entry;
1985 struct ixgbe_scattered_rx_entry *next_sc_entry;
1986 struct ixgbe_rx_entry *next_rxe = NULL;
1987 struct rte_mbuf *first_seg;
1988 struct rte_mbuf *rxm;
1989 struct rte_mbuf *nmb;
1990 union ixgbe_adv_rx_desc rxd;
1993 volatile union ixgbe_adv_rx_desc *rxdp;
1998 * The code in this whole file uses the volatile pointer to
1999 * ensure the read ordering of the status and the rest of the
2000 * descriptor fields (on the compiler level only!!!). This is so
2001 * UGLY - why not to just use the compiler barrier instead? DPDK
2002 * even has the rte_compiler_barrier() for that.
2004 * But most importantly this is just wrong because this doesn't
2005 * ensure memory ordering in a general case at all. For
2006 * instance, DPDK is supposed to work on Power CPUs where
2007 * compiler barrier may just not be enough!
2009 * I tried to write only this function properly to have a
2010 * starting point (as a part of an LRO/RSC series) but the
2011 * compiler cursed at me when I tried to cast away the
2012 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
2013 * keeping it the way it is for now.
2015 * The code in this file is broken in so many other places and
2016 * will just not work on a big endian CPU anyway therefore the
2017 * lines below will have to be revisited together with the rest
2021 * - Get rid of "volatile" crap and let the compiler do its
2023 * - Use the proper memory barrier (rte_rmb()) to ensure the
2024 * memory ordering below.
2026 rxdp = &rx_ring[rx_id];
2027 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
2029 if (!(staterr & IXGBE_RXDADV_STAT_DD))
2034 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
2035 "staterr=0x%x data_len=%u",
2036 rxq->port_id, rxq->queue_id, rx_id, staterr,
2037 rte_le_to_cpu_16(rxd.wb.upper.length));
2040 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
2042 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
2043 "port_id=%u queue_id=%u",
2044 rxq->port_id, rxq->queue_id);
2046 rte_eth_devices[rxq->port_id].data->
2047 rx_mbuf_alloc_failed++;
2050 } else if (nb_hold > rxq->rx_free_thresh) {
2051 uint16_t next_rdt = rxq->rx_free_trigger;
2053 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
2055 IXGBE_PCI_REG_WRITE_RELAXED(rxq->rdt_reg_addr,
2057 nb_hold -= rxq->rx_free_thresh;
2059 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
2060 "port_id=%u queue_id=%u",
2061 rxq->port_id, rxq->queue_id);
2063 rte_eth_devices[rxq->port_id].data->
2064 rx_mbuf_alloc_failed++;
2070 rxe = &sw_ring[rx_id];
2071 eop = staterr & IXGBE_RXDADV_STAT_EOP;
2073 next_id = rx_id + 1;
2074 if (next_id == rxq->nb_rx_desc)
2077 /* Prefetch next mbuf while processing current one. */
2078 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
2081 * When next RX descriptor is on a cache-line boundary,
2082 * prefetch the next 4 RX descriptors and the next 4 pointers
2085 if ((next_id & 0x3) == 0) {
2086 rte_ixgbe_prefetch(&rx_ring[next_id]);
2087 rte_ixgbe_prefetch(&sw_ring[next_id]);
2094 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(nmb));
2096 * Update RX descriptor with the physical address of the
2097 * new data buffer of the new allocated mbuf.
2101 rxm->data_off = RTE_PKTMBUF_HEADROOM;
2102 rxdp->read.hdr_addr = 0;
2103 rxdp->read.pkt_addr = dma;
2108 * Set data length & data buffer address of mbuf.
2110 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
2111 rxm->data_len = data_len;
2116 * Get next descriptor index:
2117 * - For RSC it's in the NEXTP field.
2118 * - For a scattered packet - it's just a following
2121 if (ixgbe_rsc_count(&rxd))
2123 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
2124 IXGBE_RXDADV_NEXTP_SHIFT;
2128 next_sc_entry = &sw_sc_ring[nextp_id];
2129 next_rxe = &sw_ring[nextp_id];
2130 rte_ixgbe_prefetch(next_rxe);
2133 sc_entry = &sw_sc_ring[rx_id];
2134 first_seg = sc_entry->fbuf;
2135 sc_entry->fbuf = NULL;
2138 * If this is the first buffer of the received packet,
2139 * set the pointer to the first mbuf of the packet and
2140 * initialize its context.
2141 * Otherwise, update the total length and the number of segments
2142 * of the current scattered packet, and update the pointer to
2143 * the last mbuf of the current packet.
2145 if (first_seg == NULL) {
2147 first_seg->pkt_len = data_len;
2148 first_seg->nb_segs = 1;
2150 first_seg->pkt_len += data_len;
2151 first_seg->nb_segs++;
2158 * If this is not the last buffer of the received packet, update
2159 * the pointer to the first mbuf at the NEXTP entry in the
2160 * sw_sc_ring and continue to parse the RX ring.
2162 if (!eop && next_rxe) {
2163 rxm->next = next_rxe->mbuf;
2164 next_sc_entry->fbuf = first_seg;
2169 * This is the last buffer of the received packet - return
2170 * the current cluster to the user.
2174 /* Initialize the first mbuf of the returned packet */
2175 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq, staterr);
2178 * Deal with the case, when HW CRC srip is disabled.
2179 * That can't happen when LRO is enabled, but still could
2180 * happen for scattered RX mode.
2182 first_seg->pkt_len -= rxq->crc_len;
2183 if (unlikely(rxm->data_len <= rxq->crc_len)) {
2184 struct rte_mbuf *lp;
2186 for (lp = first_seg; lp->next != rxm; lp = lp->next)
2189 first_seg->nb_segs--;
2190 lp->data_len -= rxq->crc_len - rxm->data_len;
2192 rte_pktmbuf_free_seg(rxm);
2194 rxm->data_len -= rxq->crc_len;
2196 /* Prefetch data of first segment, if configured to do so. */
2197 rte_packet_prefetch((char *)first_seg->buf_addr +
2198 first_seg->data_off);
2201 * Store the mbuf address into the next entry of the array
2202 * of returned packets.
2204 rx_pkts[nb_rx++] = first_seg;
2208 * Record index of the next RX descriptor to probe.
2210 rxq->rx_tail = rx_id;
2213 * If the number of free RX descriptors is greater than the RX free
2214 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
2216 * Update the RDT with the value of the last processed RX descriptor
2217 * minus 1, to guarantee that the RDT register is never equal to the
2218 * RDH register, which creates a "full" ring situtation from the
2219 * hardware point of view...
2221 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
2222 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
2223 "nb_hold=%u nb_rx=%u",
2224 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
2227 IXGBE_PCI_REG_WRITE_RELAXED(rxq->rdt_reg_addr, prev_id);
2231 rxq->nb_rx_hold = nb_hold;
2236 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2239 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
2243 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2246 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
2249 /*********************************************************************
2251 * Queue management functions
2253 **********************************************************************/
2255 static void __attribute__((cold))
2256 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
2260 if (txq->sw_ring != NULL) {
2261 for (i = 0; i < txq->nb_tx_desc; i++) {
2262 if (txq->sw_ring[i].mbuf != NULL) {
2263 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2264 txq->sw_ring[i].mbuf = NULL;
2270 static void __attribute__((cold))
2271 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
2274 txq->sw_ring != NULL)
2275 rte_free(txq->sw_ring);
2278 static void __attribute__((cold))
2279 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
2281 if (txq != NULL && txq->ops != NULL) {
2282 txq->ops->release_mbufs(txq);
2283 txq->ops->free_swring(txq);
2288 void __attribute__((cold))
2289 ixgbe_dev_tx_queue_release(void *txq)
2291 ixgbe_tx_queue_release(txq);
2294 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
2295 static void __attribute__((cold))
2296 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
2298 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
2299 struct ixgbe_tx_entry *txe = txq->sw_ring;
2302 /* Zero out HW ring memory */
2303 for (i = 0; i < txq->nb_tx_desc; i++) {
2304 txq->tx_ring[i] = zeroed_desc;
2307 /* Initialize SW ring entries */
2308 prev = (uint16_t) (txq->nb_tx_desc - 1);
2309 for (i = 0; i < txq->nb_tx_desc; i++) {
2310 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
2312 txd->wb.status = rte_cpu_to_le_32(IXGBE_TXD_STAT_DD);
2315 txe[prev].next_id = i;
2319 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2320 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2323 txq->nb_tx_used = 0;
2325 * Always allow 1 descriptor to be un-allocated to avoid
2326 * a H/W race condition
2328 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2329 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2331 memset((void *)&txq->ctx_cache, 0,
2332 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2335 static const struct ixgbe_txq_ops def_txq_ops = {
2336 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2337 .free_swring = ixgbe_tx_free_swring,
2338 .reset = ixgbe_reset_tx_queue,
2341 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2342 * the queue parameters. Used in tx_queue_setup by primary process and then
2343 * in dev_init by secondary process when attaching to an existing ethdev.
2345 void __attribute__((cold))
2346 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2348 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2349 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
2350 && (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2351 PMD_INIT_LOG(DEBUG, "Using simple tx code path");
2352 dev->tx_pkt_prepare = NULL;
2353 #ifdef RTE_IXGBE_INC_VECTOR
2354 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2355 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2356 ixgbe_txq_vec_setup(txq) == 0)) {
2357 PMD_INIT_LOG(DEBUG, "Vector tx enabled.");
2358 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2361 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2363 PMD_INIT_LOG(DEBUG, "Using full-featured tx code path");
2365 " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
2366 (unsigned long)txq->txq_flags,
2367 (unsigned long)IXGBE_SIMPLE_FLAGS);
2369 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2370 (unsigned long)txq->tx_rs_thresh,
2371 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2372 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2373 dev->tx_pkt_prepare = ixgbe_prep_pkts;
2377 int __attribute__((cold))
2378 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2381 unsigned int socket_id,
2382 const struct rte_eth_txconf *tx_conf)
2384 const struct rte_memzone *tz;
2385 struct ixgbe_tx_queue *txq;
2386 struct ixgbe_hw *hw;
2387 uint16_t tx_rs_thresh, tx_free_thresh;
2389 PMD_INIT_FUNC_TRACE();
2390 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2393 * Validate number of transmit descriptors.
2394 * It must not exceed hardware maximum, and must be multiple
2397 if (nb_desc % IXGBE_TXD_ALIGN != 0 ||
2398 (nb_desc > IXGBE_MAX_RING_DESC) ||
2399 (nb_desc < IXGBE_MIN_RING_DESC)) {
2404 * The following two parameters control the setting of the RS bit on
2405 * transmit descriptors.
2406 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2407 * descriptors have been used.
2408 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2409 * descriptors are used or if the number of descriptors required
2410 * to transmit a packet is greater than the number of free TX
2412 * The following constraints must be satisfied:
2413 * tx_rs_thresh must be greater than 0.
2414 * tx_rs_thresh must be less than the size of the ring minus 2.
2415 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2416 * tx_rs_thresh must be a divisor of the ring size.
2417 * tx_free_thresh must be greater than 0.
2418 * tx_free_thresh must be less than the size of the ring minus 3.
2419 * One descriptor in the TX ring is used as a sentinel to avoid a
2420 * H/W race condition, hence the maximum threshold constraints.
2421 * When set to zero use default values.
2423 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2424 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2425 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2426 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2427 if (tx_rs_thresh >= (nb_desc - 2)) {
2428 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2429 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2430 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2431 (int)dev->data->port_id, (int)queue_idx);
2434 if (tx_rs_thresh > DEFAULT_TX_RS_THRESH) {
2435 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less or equal than %u. "
2436 "(tx_rs_thresh=%u port=%d queue=%d)",
2437 DEFAULT_TX_RS_THRESH, (unsigned int)tx_rs_thresh,
2438 (int)dev->data->port_id, (int)queue_idx);
2441 if (tx_free_thresh >= (nb_desc - 3)) {
2442 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2443 "tx_free_thresh must be less than the number of "
2444 "TX descriptors minus 3. (tx_free_thresh=%u "
2445 "port=%d queue=%d)",
2446 (unsigned int)tx_free_thresh,
2447 (int)dev->data->port_id, (int)queue_idx);
2450 if (tx_rs_thresh > tx_free_thresh) {
2451 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2452 "tx_free_thresh. (tx_free_thresh=%u "
2453 "tx_rs_thresh=%u port=%d queue=%d)",
2454 (unsigned int)tx_free_thresh,
2455 (unsigned int)tx_rs_thresh,
2456 (int)dev->data->port_id,
2460 if ((nb_desc % tx_rs_thresh) != 0) {
2461 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2462 "number of TX descriptors. (tx_rs_thresh=%u "
2463 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2464 (int)dev->data->port_id, (int)queue_idx);
2469 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2470 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2471 * by the NIC and all descriptors are written back after the NIC
2472 * accumulates WTHRESH descriptors.
2474 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2475 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2476 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2477 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2478 (int)dev->data->port_id, (int)queue_idx);
2482 /* Free memory prior to re-allocation if needed... */
2483 if (dev->data->tx_queues[queue_idx] != NULL) {
2484 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2485 dev->data->tx_queues[queue_idx] = NULL;
2488 /* First allocate the tx queue data structure */
2489 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2490 RTE_CACHE_LINE_SIZE, socket_id);
2495 * Allocate TX ring hardware descriptors. A memzone large enough to
2496 * handle the maximum ring size is allocated in order to allow for
2497 * resizing in later calls to the queue setup function.
2499 tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
2500 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2501 IXGBE_ALIGN, socket_id);
2503 ixgbe_tx_queue_release(txq);
2507 txq->nb_tx_desc = nb_desc;
2508 txq->tx_rs_thresh = tx_rs_thresh;
2509 txq->tx_free_thresh = tx_free_thresh;
2510 txq->pthresh = tx_conf->tx_thresh.pthresh;
2511 txq->hthresh = tx_conf->tx_thresh.hthresh;
2512 txq->wthresh = tx_conf->tx_thresh.wthresh;
2513 txq->queue_id = queue_idx;
2514 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2515 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2516 txq->port_id = dev->data->port_id;
2517 txq->txq_flags = tx_conf->txq_flags;
2518 txq->ops = &def_txq_ops;
2519 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2522 * Modification to set VFTDT for virtual function if vf is detected
2524 if (hw->mac.type == ixgbe_mac_82599_vf ||
2525 hw->mac.type == ixgbe_mac_X540_vf ||
2526 hw->mac.type == ixgbe_mac_X550_vf ||
2527 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2528 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2529 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2531 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2533 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2534 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2536 /* Allocate software ring */
2537 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2538 sizeof(struct ixgbe_tx_entry) * nb_desc,
2539 RTE_CACHE_LINE_SIZE, socket_id);
2540 if (txq->sw_ring == NULL) {
2541 ixgbe_tx_queue_release(txq);
2544 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2545 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2547 /* set up vector or scalar TX function as appropriate */
2548 ixgbe_set_tx_function(dev, txq);
2550 txq->ops->reset(txq);
2552 dev->data->tx_queues[queue_idx] = txq;
2559 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2561 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2562 * in the sw_rsc_ring is not set to NULL but rather points to the next
2563 * mbuf of this RSC aggregation (that has not been completed yet and still
2564 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2565 * will just free first "nb_segs" segments of the cluster explicitly by calling
2566 * an rte_pktmbuf_free_seg().
2568 * @m scattered cluster head
2570 static void __attribute__((cold))
2571 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2573 uint8_t i, nb_segs = m->nb_segs;
2574 struct rte_mbuf *next_seg;
2576 for (i = 0; i < nb_segs; i++) {
2578 rte_pktmbuf_free_seg(m);
2583 static void __attribute__((cold))
2584 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2588 #ifdef RTE_IXGBE_INC_VECTOR
2589 /* SSE Vector driver has a different way of releasing mbufs. */
2590 if (rxq->rx_using_sse) {
2591 ixgbe_rx_queue_release_mbufs_vec(rxq);
2596 if (rxq->sw_ring != NULL) {
2597 for (i = 0; i < rxq->nb_rx_desc; i++) {
2598 if (rxq->sw_ring[i].mbuf != NULL) {
2599 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2600 rxq->sw_ring[i].mbuf = NULL;
2603 if (rxq->rx_nb_avail) {
2604 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2605 struct rte_mbuf *mb;
2607 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2608 rte_pktmbuf_free_seg(mb);
2610 rxq->rx_nb_avail = 0;
2614 if (rxq->sw_sc_ring)
2615 for (i = 0; i < rxq->nb_rx_desc; i++)
2616 if (rxq->sw_sc_ring[i].fbuf) {
2617 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2618 rxq->sw_sc_ring[i].fbuf = NULL;
2622 static void __attribute__((cold))
2623 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2626 ixgbe_rx_queue_release_mbufs(rxq);
2627 rte_free(rxq->sw_ring);
2628 rte_free(rxq->sw_sc_ring);
2633 void __attribute__((cold))
2634 ixgbe_dev_rx_queue_release(void *rxq)
2636 ixgbe_rx_queue_release(rxq);
2640 * Check if Rx Burst Bulk Alloc function can be used.
2642 * 0: the preconditions are satisfied and the bulk allocation function
2644 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2645 * function must be used.
2647 static inline int __attribute__((cold))
2648 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2653 * Make sure the following pre-conditions are satisfied:
2654 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2655 * rxq->rx_free_thresh < rxq->nb_rx_desc
2656 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2657 * Scattered packets are not supported. This should be checked
2658 * outside of this function.
2660 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2661 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2662 "rxq->rx_free_thresh=%d, "
2663 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2664 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2666 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2667 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2668 "rxq->rx_free_thresh=%d, "
2669 "rxq->nb_rx_desc=%d",
2670 rxq->rx_free_thresh, rxq->nb_rx_desc);
2672 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2673 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2674 "rxq->nb_rx_desc=%d, "
2675 "rxq->rx_free_thresh=%d",
2676 rxq->nb_rx_desc, rxq->rx_free_thresh);
2683 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2684 static void __attribute__((cold))
2685 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2687 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2689 uint16_t len = rxq->nb_rx_desc;
2692 * By default, the Rx queue setup function allocates enough memory for
2693 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2694 * extra memory at the end of the descriptor ring to be zero'd out.
2696 if (adapter->rx_bulk_alloc_allowed)
2697 /* zero out extra memory */
2698 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2701 * Zero out HW ring memory. Zero out extra memory at the end of
2702 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2703 * reads extra memory as zeros.
2705 for (i = 0; i < len; i++) {
2706 rxq->rx_ring[i] = zeroed_desc;
2710 * initialize extra software ring entries. Space for these extra
2711 * entries is always allocated
2713 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2714 for (i = rxq->nb_rx_desc; i < len; ++i) {
2715 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2718 rxq->rx_nb_avail = 0;
2719 rxq->rx_next_avail = 0;
2720 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2722 rxq->nb_rx_hold = 0;
2723 rxq->pkt_first_seg = NULL;
2724 rxq->pkt_last_seg = NULL;
2726 #ifdef RTE_IXGBE_INC_VECTOR
2727 rxq->rxrearm_start = 0;
2728 rxq->rxrearm_nb = 0;
2732 int __attribute__((cold))
2733 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2736 unsigned int socket_id,
2737 const struct rte_eth_rxconf *rx_conf,
2738 struct rte_mempool *mp)
2740 const struct rte_memzone *rz;
2741 struct ixgbe_rx_queue *rxq;
2742 struct ixgbe_hw *hw;
2744 struct ixgbe_adapter *adapter =
2745 (struct ixgbe_adapter *)dev->data->dev_private;
2747 PMD_INIT_FUNC_TRACE();
2748 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2751 * Validate number of receive descriptors.
2752 * It must not exceed hardware maximum, and must be multiple
2755 if (nb_desc % IXGBE_RXD_ALIGN != 0 ||
2756 (nb_desc > IXGBE_MAX_RING_DESC) ||
2757 (nb_desc < IXGBE_MIN_RING_DESC)) {
2761 /* Free memory prior to re-allocation if needed... */
2762 if (dev->data->rx_queues[queue_idx] != NULL) {
2763 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2764 dev->data->rx_queues[queue_idx] = NULL;
2767 /* First allocate the rx queue data structure */
2768 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2769 RTE_CACHE_LINE_SIZE, socket_id);
2773 rxq->nb_rx_desc = nb_desc;
2774 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2775 rxq->queue_id = queue_idx;
2776 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2777 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2778 rxq->port_id = dev->data->port_id;
2779 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2781 rxq->drop_en = rx_conf->rx_drop_en;
2782 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2785 * The packet type in RX descriptor is different for different NICs.
2786 * Some bits are used for x550 but reserved for other NICS.
2787 * So set different masks for different NICs.
2789 if (hw->mac.type == ixgbe_mac_X550 ||
2790 hw->mac.type == ixgbe_mac_X550EM_x ||
2791 hw->mac.type == ixgbe_mac_X550EM_a ||
2792 hw->mac.type == ixgbe_mac_X550_vf ||
2793 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2794 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2795 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_X550;
2797 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_82599;
2800 * Allocate RX ring hardware descriptors. A memzone large enough to
2801 * handle the maximum ring size is allocated in order to allow for
2802 * resizing in later calls to the queue setup function.
2804 rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
2805 RX_RING_SZ, IXGBE_ALIGN, socket_id);
2807 ixgbe_rx_queue_release(rxq);
2812 * Zero init all the descriptors in the ring.
2814 memset(rz->addr, 0, RX_RING_SZ);
2817 * Modified to setup VFRDT for Virtual Function
2819 if (hw->mac.type == ixgbe_mac_82599_vf ||
2820 hw->mac.type == ixgbe_mac_X540_vf ||
2821 hw->mac.type == ixgbe_mac_X550_vf ||
2822 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2823 hw->mac.type == ixgbe_mac_X550EM_a_vf) {
2825 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2827 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2830 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2832 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2835 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2836 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2839 * Certain constraints must be met in order to use the bulk buffer
2840 * allocation Rx burst function. If any of Rx queues doesn't meet them
2841 * the feature should be disabled for the whole port.
2843 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
2844 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
2845 "preconditions - canceling the feature for "
2846 "the whole port[%d]",
2847 rxq->queue_id, rxq->port_id);
2848 adapter->rx_bulk_alloc_allowed = false;
2852 * Allocate software ring. Allow for space at the end of the
2853 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2854 * function does not access an invalid memory region.
2857 if (adapter->rx_bulk_alloc_allowed)
2858 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2860 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2861 sizeof(struct ixgbe_rx_entry) * len,
2862 RTE_CACHE_LINE_SIZE, socket_id);
2863 if (!rxq->sw_ring) {
2864 ixgbe_rx_queue_release(rxq);
2869 * Always allocate even if it's not going to be needed in order to
2870 * simplify the code.
2872 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
2873 * be requested in ixgbe_dev_rx_init(), which is called later from
2877 rte_zmalloc_socket("rxq->sw_sc_ring",
2878 sizeof(struct ixgbe_scattered_rx_entry) * len,
2879 RTE_CACHE_LINE_SIZE, socket_id);
2880 if (!rxq->sw_sc_ring) {
2881 ixgbe_rx_queue_release(rxq);
2885 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
2886 "dma_addr=0x%"PRIx64,
2887 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
2888 rxq->rx_ring_phys_addr);
2890 if (!rte_is_power_of_2(nb_desc)) {
2891 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
2892 "preconditions - canceling the feature for "
2893 "the whole port[%d]",
2894 rxq->queue_id, rxq->port_id);
2895 adapter->rx_vec_allowed = false;
2897 ixgbe_rxq_vec_setup(rxq);
2899 dev->data->rx_queues[queue_idx] = rxq;
2901 ixgbe_reset_rx_queue(adapter, rxq);
2907 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2909 #define IXGBE_RXQ_SCAN_INTERVAL 4
2910 volatile union ixgbe_adv_rx_desc *rxdp;
2911 struct ixgbe_rx_queue *rxq;
2914 rxq = dev->data->rx_queues[rx_queue_id];
2915 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2917 while ((desc < rxq->nb_rx_desc) &&
2918 (rxdp->wb.upper.status_error &
2919 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD))) {
2920 desc += IXGBE_RXQ_SCAN_INTERVAL;
2921 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2922 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2923 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2924 desc - rxq->nb_rx_desc]);
2931 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2933 volatile union ixgbe_adv_rx_desc *rxdp;
2934 struct ixgbe_rx_queue *rxq = rx_queue;
2937 if (unlikely(offset >= rxq->nb_rx_desc))
2939 desc = rxq->rx_tail + offset;
2940 if (desc >= rxq->nb_rx_desc)
2941 desc -= rxq->nb_rx_desc;
2943 rxdp = &rxq->rx_ring[desc];
2944 return !!(rxdp->wb.upper.status_error &
2945 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD));
2949 ixgbe_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
2951 struct ixgbe_rx_queue *rxq = rx_queue;
2952 volatile uint32_t *status;
2953 uint32_t nb_hold, desc;
2955 if (unlikely(offset >= rxq->nb_rx_desc))
2958 #ifdef RTE_IXGBE_INC_VECTOR
2959 if (rxq->rx_using_sse)
2960 nb_hold = rxq->rxrearm_nb;
2963 nb_hold = rxq->nb_rx_hold;
2964 if (offset >= rxq->nb_rx_desc - nb_hold)
2965 return RTE_ETH_RX_DESC_UNAVAIL;
2967 desc = rxq->rx_tail + offset;
2968 if (desc >= rxq->nb_rx_desc)
2969 desc -= rxq->nb_rx_desc;
2971 status = &rxq->rx_ring[desc].wb.upper.status_error;
2972 if (*status & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD))
2973 return RTE_ETH_RX_DESC_DONE;
2975 return RTE_ETH_RX_DESC_AVAIL;
2979 ixgbe_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
2981 struct ixgbe_tx_queue *txq = tx_queue;
2982 volatile uint32_t *status;
2985 if (unlikely(offset >= txq->nb_tx_desc))
2988 desc = txq->tx_tail + offset;
2989 /* go to next desc that has the RS bit */
2990 desc = ((desc + txq->tx_rs_thresh - 1) / txq->tx_rs_thresh) *
2992 if (desc >= txq->nb_tx_desc) {
2993 desc -= txq->nb_tx_desc;
2994 if (desc >= txq->nb_tx_desc)
2995 desc -= txq->nb_tx_desc;
2998 status = &txq->tx_ring[desc].wb.status;
2999 if (*status & rte_cpu_to_le_32(IXGBE_ADVTXD_STAT_DD))
3000 return RTE_ETH_TX_DESC_DONE;
3002 return RTE_ETH_TX_DESC_FULL;
3005 void __attribute__((cold))
3006 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
3009 struct ixgbe_adapter *adapter =
3010 (struct ixgbe_adapter *)dev->data->dev_private;
3012 PMD_INIT_FUNC_TRACE();
3014 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3015 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
3018 txq->ops->release_mbufs(txq);
3019 txq->ops->reset(txq);
3023 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3024 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
3027 ixgbe_rx_queue_release_mbufs(rxq);
3028 ixgbe_reset_rx_queue(adapter, rxq);
3034 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
3038 PMD_INIT_FUNC_TRACE();
3040 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3041 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
3042 dev->data->rx_queues[i] = NULL;
3044 dev->data->nb_rx_queues = 0;
3046 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3047 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
3048 dev->data->tx_queues[i] = NULL;
3050 dev->data->nb_tx_queues = 0;
3053 /*********************************************************************
3055 * Device RX/TX init functions
3057 **********************************************************************/
3060 * Receive Side Scaling (RSS)
3061 * See section 7.1.2.8 in the following document:
3062 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
3065 * The source and destination IP addresses of the IP header and the source
3066 * and destination ports of TCP/UDP headers, if any, of received packets are
3067 * hashed against a configurable random key to compute a 32-bit RSS hash result.
3068 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
3069 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
3070 * RSS output index which is used as the RX queue index where to store the
3072 * The following output is supplied in the RX write-back descriptor:
3073 * - 32-bit result of the Microsoft RSS hash function,
3074 * - 4-bit RSS type field.
3078 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
3079 * Used as the default key.
3081 static uint8_t rss_intel_key[40] = {
3082 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
3083 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
3084 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3085 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
3086 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
3090 ixgbe_rss_disable(struct rte_eth_dev *dev)
3092 struct ixgbe_hw *hw;
3096 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3097 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3098 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3099 mrqc &= ~IXGBE_MRQC_RSSEN;
3100 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
3104 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
3114 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3115 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3117 hash_key = rss_conf->rss_key;
3118 if (hash_key != NULL) {
3119 /* Fill in RSS hash key */
3120 for (i = 0; i < 10; i++) {
3121 rss_key = hash_key[(i * 4)];
3122 rss_key |= hash_key[(i * 4) + 1] << 8;
3123 rss_key |= hash_key[(i * 4) + 2] << 16;
3124 rss_key |= hash_key[(i * 4) + 3] << 24;
3125 IXGBE_WRITE_REG_ARRAY(hw, rssrk_reg, i, rss_key);
3129 /* Set configured hashing protocols in MRQC register */
3130 rss_hf = rss_conf->rss_hf;
3131 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
3132 if (rss_hf & ETH_RSS_IPV4)
3133 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
3134 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
3135 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
3136 if (rss_hf & ETH_RSS_IPV6)
3137 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
3138 if (rss_hf & ETH_RSS_IPV6_EX)
3139 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
3140 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
3141 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3142 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
3143 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
3144 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
3145 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3146 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
3147 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3148 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
3149 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
3150 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
3154 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
3155 struct rte_eth_rss_conf *rss_conf)
3157 struct ixgbe_hw *hw;
3162 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3164 if (!ixgbe_rss_update_sp(hw->mac.type)) {
3165 PMD_DRV_LOG(ERR, "RSS hash update is not supported on this "
3169 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3172 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
3173 * "RSS enabling cannot be done dynamically while it must be
3174 * preceded by a software reset"
3175 * Before changing anything, first check that the update RSS operation
3176 * does not attempt to disable RSS, if RSS was enabled at
3177 * initialization time, or does not attempt to enable RSS, if RSS was
3178 * disabled at initialization time.
3180 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
3181 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3182 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
3183 if (rss_hf != 0) /* Enable RSS */
3185 return 0; /* Nothing to do */
3188 if (rss_hf == 0) /* Disable RSS */
3190 ixgbe_hw_rss_hash_set(hw, rss_conf);
3195 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
3196 struct rte_eth_rss_conf *rss_conf)
3198 struct ixgbe_hw *hw;
3207 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3208 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3209 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3210 hash_key = rss_conf->rss_key;
3211 if (hash_key != NULL) {
3212 /* Return RSS hash key */
3213 for (i = 0; i < 10; i++) {
3214 rss_key = IXGBE_READ_REG_ARRAY(hw, rssrk_reg, i);
3215 hash_key[(i * 4)] = rss_key & 0x000000FF;
3216 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
3217 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
3218 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
3222 /* Get RSS functions configured in MRQC register */
3223 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3224 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
3225 rss_conf->rss_hf = 0;
3229 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
3230 rss_hf |= ETH_RSS_IPV4;
3231 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
3232 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
3233 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
3234 rss_hf |= ETH_RSS_IPV6;
3235 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
3236 rss_hf |= ETH_RSS_IPV6_EX;
3237 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
3238 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
3239 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
3240 rss_hf |= ETH_RSS_IPV6_TCP_EX;
3241 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
3242 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
3243 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
3244 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
3245 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
3246 rss_hf |= ETH_RSS_IPV6_UDP_EX;
3247 rss_conf->rss_hf = rss_hf;
3252 ixgbe_rss_configure(struct rte_eth_dev *dev)
3254 struct rte_eth_rss_conf rss_conf;
3255 struct ixgbe_hw *hw;
3259 uint16_t sp_reta_size;
3262 PMD_INIT_FUNC_TRACE();
3263 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3265 sp_reta_size = ixgbe_reta_size_get(hw->mac.type);
3268 * Fill in redirection table
3269 * The byte-swap is needed because NIC registers are in
3270 * little-endian order.
3273 for (i = 0, j = 0; i < sp_reta_size; i++, j++) {
3274 reta_reg = ixgbe_reta_reg_get(hw->mac.type, i);
3276 if (j == dev->data->nb_rx_queues)
3278 reta = (reta << 8) | j;
3280 IXGBE_WRITE_REG(hw, reta_reg,
3285 * Configure the RSS key and the RSS protocols used to compute
3286 * the RSS hash of input packets.
3288 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
3289 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
3290 ixgbe_rss_disable(dev);
3293 if (rss_conf.rss_key == NULL)
3294 rss_conf.rss_key = rss_intel_key; /* Default hash key */
3295 ixgbe_hw_rss_hash_set(hw, &rss_conf);
3298 #define NUM_VFTA_REGISTERS 128
3299 #define NIC_RX_BUFFER_SIZE 0x200
3300 #define X550_RX_BUFFER_SIZE 0x180
3303 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
3305 struct rte_eth_vmdq_dcb_conf *cfg;
3306 struct ixgbe_hw *hw;
3307 enum rte_eth_nb_pools num_pools;
3308 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
3310 uint8_t nb_tcs; /* number of traffic classes */
3313 PMD_INIT_FUNC_TRACE();
3314 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3315 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3316 num_pools = cfg->nb_queue_pools;
3317 /* Check we have a valid number of pools */
3318 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
3319 ixgbe_rss_disable(dev);
3322 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
3323 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
3327 * split rx buffer up into sections, each for 1 traffic class
3329 switch (hw->mac.type) {
3330 case ixgbe_mac_X550:
3331 case ixgbe_mac_X550EM_x:
3332 case ixgbe_mac_X550EM_a:
3333 pbsize = (uint16_t)(X550_RX_BUFFER_SIZE / nb_tcs);
3336 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3339 for (i = 0; i < nb_tcs; i++) {
3340 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3342 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3343 /* clear 10 bits. */
3344 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
3345 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3347 /* zero alloc all unused TCs */
3348 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3349 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3351 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3352 /* clear 10 bits. */
3353 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3356 /* MRQC: enable vmdq and dcb */
3357 mrqc = (num_pools == ETH_16_POOLS) ?
3358 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN;
3359 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3361 /* PFVTCTL: turn on virtualisation and set the default pool */
3362 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3363 if (cfg->enable_default_pool) {
3364 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3366 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3369 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3371 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
3373 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
3375 * mapping is done with 3 bits per priority,
3376 * so shift by i*3 each time
3378 queue_mapping |= ((cfg->dcb_tc[i] & 0x07) << (i * 3));
3380 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
3382 /* RTRPCS: DCB related */
3383 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
3385 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3386 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3387 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3388 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3390 /* VFTA - enable all vlan filters */
3391 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3392 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3395 /* VFRE: pool enabling for receive - 16 or 32 */
3396 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0),
3397 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3400 * MPSAR - allow pools to read specific mac addresses
3401 * In this case, all pools should be able to read from mac addr 0
3403 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
3404 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
3406 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3407 for (i = 0; i < cfg->nb_pool_maps; i++) {
3408 /* set vlan id in VF register and set the valid bit */
3409 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3410 (cfg->pool_map[i].vlan_id & 0xFFF)));
3412 * Put the allowed pools in VFB reg. As we only have 16 or 32
3413 * pools, we only need to use the first half of the register
3416 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
3421 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
3422 * @dev: pointer to eth_dev structure
3423 * @dcb_config: pointer to ixgbe_dcb_config structure
3426 ixgbe_dcb_tx_hw_config(struct rte_eth_dev *dev,
3427 struct ixgbe_dcb_config *dcb_config)
3431 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3433 PMD_INIT_FUNC_TRACE();
3434 if (hw->mac.type != ixgbe_mac_82598EB) {
3435 /* Disable the Tx desc arbiter so that MTQC can be changed */
3436 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3437 reg |= IXGBE_RTTDCS_ARBDIS;
3438 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3440 /* Enable DCB for Tx with 8 TCs */
3441 if (dcb_config->num_tcs.pg_tcs == 8) {
3442 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3444 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3446 if (dcb_config->vt_mode)
3447 reg |= IXGBE_MTQC_VT_ENA;
3448 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3450 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3451 /* Disable drop for all queues in VMDQ mode*/
3452 for (q = 0; q < 128; q++)
3453 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3454 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3456 /* Enable drop for all queues in SRIOV mode */
3457 for (q = 0; q < 128; q++)
3458 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3459 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT) | IXGBE_QDE_ENABLE));
3462 /* Enable the Tx desc arbiter */
3463 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3464 reg &= ~IXGBE_RTTDCS_ARBDIS;
3465 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3467 /* Enable Security TX Buffer IFG for DCB */
3468 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3469 reg |= IXGBE_SECTX_DCB;
3470 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3475 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3476 * @dev: pointer to rte_eth_dev structure
3477 * @dcb_config: pointer to ixgbe_dcb_config structure
3480 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3481 struct ixgbe_dcb_config *dcb_config)
3483 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3484 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3485 struct ixgbe_hw *hw =
3486 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3488 PMD_INIT_FUNC_TRACE();
3489 if (hw->mac.type != ixgbe_mac_82598EB)
3490 /*PF VF Transmit Enable*/
3491 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3492 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3494 /*Configure general DCB TX parameters*/
3495 ixgbe_dcb_tx_hw_config(dev, dcb_config);
3499 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3500 struct ixgbe_dcb_config *dcb_config)
3502 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3503 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3504 struct ixgbe_dcb_tc_config *tc;
3507 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3508 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS) {
3509 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3510 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3512 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3513 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3515 /* User Priority to Traffic Class mapping */
3516 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3517 j = vmdq_rx_conf->dcb_tc[i];
3518 tc = &dcb_config->tc_config[j];
3519 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3525 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3526 struct ixgbe_dcb_config *dcb_config)
3528 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3529 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3530 struct ixgbe_dcb_tc_config *tc;
3533 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3534 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS) {
3535 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3536 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3538 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3539 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3542 /* User Priority to Traffic Class mapping */
3543 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3544 j = vmdq_tx_conf->dcb_tc[i];
3545 tc = &dcb_config->tc_config[j];
3546 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3552 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3553 struct ixgbe_dcb_config *dcb_config)
3555 struct rte_eth_dcb_rx_conf *rx_conf =
3556 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3557 struct ixgbe_dcb_tc_config *tc;
3560 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3561 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3563 /* User Priority to Traffic Class mapping */
3564 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3565 j = rx_conf->dcb_tc[i];
3566 tc = &dcb_config->tc_config[j];
3567 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3573 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3574 struct ixgbe_dcb_config *dcb_config)
3576 struct rte_eth_dcb_tx_conf *tx_conf =
3577 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3578 struct ixgbe_dcb_tc_config *tc;
3581 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3582 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3584 /* User Priority to Traffic Class mapping */
3585 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3586 j = tx_conf->dcb_tc[i];
3587 tc = &dcb_config->tc_config[j];
3588 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3594 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3595 * @hw: pointer to hardware structure
3596 * @dcb_config: pointer to ixgbe_dcb_config structure
3599 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
3600 struct ixgbe_dcb_config *dcb_config)
3606 PMD_INIT_FUNC_TRACE();
3608 * Disable the arbiter before changing parameters
3609 * (always enable recycle mode; WSP)
3611 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3612 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3614 if (hw->mac.type != ixgbe_mac_82598EB) {
3615 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3616 if (dcb_config->num_tcs.pg_tcs == 4) {
3617 if (dcb_config->vt_mode)
3618 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3619 IXGBE_MRQC_VMDQRT4TCEN;
3621 /* no matter the mode is DCB or DCB_RSS, just
3622 * set the MRQE to RSSXTCEN. RSS is controlled
3625 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3626 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3627 IXGBE_MRQC_RTRSS4TCEN;
3630 if (dcb_config->num_tcs.pg_tcs == 8) {
3631 if (dcb_config->vt_mode)
3632 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3633 IXGBE_MRQC_VMDQRT8TCEN;
3635 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3636 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3637 IXGBE_MRQC_RTRSS8TCEN;
3641 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3644 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3645 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3646 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3647 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3649 /* VFTA - enable all vlan filters */
3650 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3651 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3655 * Configure Rx packet plane (recycle mode; WSP) and
3658 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3659 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3663 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3664 uint16_t *max, uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3666 switch (hw->mac.type) {
3667 case ixgbe_mac_82598EB:
3668 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3670 case ixgbe_mac_82599EB:
3671 case ixgbe_mac_X540:
3672 case ixgbe_mac_X550:
3673 case ixgbe_mac_X550EM_x:
3674 case ixgbe_mac_X550EM_a:
3675 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3684 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3685 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3687 switch (hw->mac.type) {
3688 case ixgbe_mac_82598EB:
3689 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id, tsa);
3690 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id, tsa);
3692 case ixgbe_mac_82599EB:
3693 case ixgbe_mac_X540:
3694 case ixgbe_mac_X550:
3695 case ixgbe_mac_X550EM_x:
3696 case ixgbe_mac_X550EM_a:
3697 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id, tsa);
3698 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id, tsa, map);
3705 #define DCB_RX_CONFIG 1
3706 #define DCB_TX_CONFIG 1
3707 #define DCB_TX_PB 1024
3709 * ixgbe_dcb_hw_configure - Enable DCB and configure
3710 * general DCB in VT mode and non-VT mode parameters
3711 * @dev: pointer to rte_eth_dev structure
3712 * @dcb_config: pointer to ixgbe_dcb_config structure
3715 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3716 struct ixgbe_dcb_config *dcb_config)
3719 uint8_t i, pfc_en, nb_tcs;
3720 uint16_t pbsize, rx_buffer_size;
3721 uint8_t config_dcb_rx = 0;
3722 uint8_t config_dcb_tx = 0;
3723 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3724 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3725 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3726 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3727 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3728 struct ixgbe_dcb_tc_config *tc;
3729 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3730 struct ixgbe_hw *hw =
3731 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3733 switch (dev->data->dev_conf.rxmode.mq_mode) {
3734 case ETH_MQ_RX_VMDQ_DCB:
3735 dcb_config->vt_mode = true;
3736 if (hw->mac.type != ixgbe_mac_82598EB) {
3737 config_dcb_rx = DCB_RX_CONFIG;
3739 *get dcb and VT rx configuration parameters
3742 ixgbe_vmdq_dcb_rx_config(dev, dcb_config);
3743 /*Configure general VMDQ and DCB RX parameters*/
3744 ixgbe_vmdq_dcb_configure(dev);
3748 case ETH_MQ_RX_DCB_RSS:
3749 dcb_config->vt_mode = false;
3750 config_dcb_rx = DCB_RX_CONFIG;
3751 /* Get dcb TX configuration parameters from rte_eth_conf */
3752 ixgbe_dcb_rx_config(dev, dcb_config);
3753 /*Configure general DCB RX parameters*/
3754 ixgbe_dcb_rx_hw_config(hw, dcb_config);
3757 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3760 switch (dev->data->dev_conf.txmode.mq_mode) {
3761 case ETH_MQ_TX_VMDQ_DCB:
3762 dcb_config->vt_mode = true;
3763 config_dcb_tx = DCB_TX_CONFIG;
3764 /* get DCB and VT TX configuration parameters
3767 ixgbe_dcb_vt_tx_config(dev, dcb_config);
3768 /*Configure general VMDQ and DCB TX parameters*/
3769 ixgbe_vmdq_dcb_hw_tx_config(dev, dcb_config);
3773 dcb_config->vt_mode = false;
3774 config_dcb_tx = DCB_TX_CONFIG;
3775 /*get DCB TX configuration parameters from rte_eth_conf*/
3776 ixgbe_dcb_tx_config(dev, dcb_config);
3777 /*Configure general DCB TX parameters*/
3778 ixgbe_dcb_tx_hw_config(dev, dcb_config);
3781 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3785 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3787 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3788 if (nb_tcs == ETH_4_TCS) {
3789 /* Avoid un-configured priority mapping to TC0 */
3791 uint8_t mask = 0xFF;
3793 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3794 mask = (uint8_t)(mask & (~(1 << map[i])));
3795 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3796 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3800 /* Re-configure 4 TCs BW */
3801 for (i = 0; i < nb_tcs; i++) {
3802 tc = &dcb_config->tc_config[i];
3803 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
3804 (uint8_t)(100 / nb_tcs);
3805 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
3806 (uint8_t)(100 / nb_tcs);
3808 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
3809 tc = &dcb_config->tc_config[i];
3810 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
3811 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
3815 switch (hw->mac.type) {
3816 case ixgbe_mac_X550:
3817 case ixgbe_mac_X550EM_x:
3818 case ixgbe_mac_X550EM_a:
3819 rx_buffer_size = X550_RX_BUFFER_SIZE;
3822 rx_buffer_size = NIC_RX_BUFFER_SIZE;
3826 if (config_dcb_rx) {
3827 /* Set RX buffer size */
3828 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
3829 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
3831 for (i = 0; i < nb_tcs; i++) {
3832 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3834 /* zero alloc all unused TCs */
3835 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3836 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
3839 if (config_dcb_tx) {
3840 /* Only support an equally distributed
3841 * Tx packet buffer strategy.
3843 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
3844 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
3846 for (i = 0; i < nb_tcs; i++) {
3847 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
3848 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
3850 /* Clear unused TCs, if any, to zero buffer size*/
3851 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3852 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
3853 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
3857 /*Calculates traffic class credits*/
3858 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
3859 IXGBE_DCB_TX_CONFIG);
3860 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
3861 IXGBE_DCB_RX_CONFIG);
3863 if (config_dcb_rx) {
3864 /* Unpack CEE standard containers */
3865 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
3866 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3867 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
3868 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
3869 /* Configure PG(ETS) RX */
3870 ixgbe_dcb_hw_arbite_rx_config(hw, refill, max, bwgid, tsa, map);
3873 if (config_dcb_tx) {
3874 /* Unpack CEE standard containers */
3875 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
3876 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3877 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
3878 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
3879 /* Configure PG(ETS) TX */
3880 ixgbe_dcb_hw_arbite_tx_config(hw, refill, max, bwgid, tsa, map);
3883 /*Configure queue statistics registers*/
3884 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3886 /* Check if the PFC is supported */
3887 if (dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3888 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
3889 for (i = 0; i < nb_tcs; i++) {
3891 * If the TC count is 8,and the default high_water is 48,
3892 * the low_water is 16 as default.
3894 hw->fc.high_water[i] = (pbsize * 3) / 4;
3895 hw->fc.low_water[i] = pbsize / 4;
3896 /* Enable pfc for this TC */
3897 tc = &dcb_config->tc_config[i];
3898 tc->pfc = ixgbe_dcb_pfc_enabled;
3900 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3901 if (dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3903 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3910 * ixgbe_configure_dcb - Configure DCB Hardware
3911 * @dev: pointer to rte_eth_dev
3913 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3915 struct ixgbe_dcb_config *dcb_cfg =
3916 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3917 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3919 PMD_INIT_FUNC_TRACE();
3921 /* check support mq_mode for DCB */
3922 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
3923 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB) &&
3924 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB_RSS))
3927 if (dev->data->nb_rx_queues > ETH_DCB_NUM_QUEUES)
3930 /** Configure DCB hardware **/
3931 ixgbe_dcb_hw_configure(dev, dcb_cfg);
3935 * VMDq only support for 10 GbE NIC.
3938 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
3940 struct rte_eth_vmdq_rx_conf *cfg;
3941 struct ixgbe_hw *hw;
3942 enum rte_eth_nb_pools num_pools;
3943 uint32_t mrqc, vt_ctl, vlanctrl;
3947 PMD_INIT_FUNC_TRACE();
3948 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3949 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
3950 num_pools = cfg->nb_queue_pools;
3952 ixgbe_rss_disable(dev);
3954 /* MRQC: enable vmdq */
3955 mrqc = IXGBE_MRQC_VMDQEN;
3956 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3958 /* PFVTCTL: turn on virtualisation and set the default pool */
3959 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3960 if (cfg->enable_default_pool)
3961 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3963 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3965 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3967 for (i = 0; i < (int)num_pools; i++) {
3968 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
3969 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
3972 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3973 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3974 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3975 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3977 /* VFTA - enable all vlan filters */
3978 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
3979 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
3981 /* VFRE: pool enabling for receive - 64 */
3982 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
3983 if (num_pools == ETH_64_POOLS)
3984 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
3987 * MPSAR - allow pools to read specific mac addresses
3988 * In this case, all pools should be able to read from mac addr 0
3990 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
3991 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
3993 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3994 for (i = 0; i < cfg->nb_pool_maps; i++) {
3995 /* set vlan id in VF register and set the valid bit */
3996 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3997 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
3999 * Put the allowed pools in VFB reg. As we only have 16 or 64
4000 * pools, we only need to use the first half of the register
4003 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
4004 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i * 2),
4005 (cfg->pool_map[i].pools & UINT32_MAX));
4007 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i * 2 + 1)),
4008 ((cfg->pool_map[i].pools >> 32) & UINT32_MAX));
4012 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
4013 if (cfg->enable_loop_back) {
4014 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4015 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
4016 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
4019 IXGBE_WRITE_FLUSH(hw);
4023 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
4024 * @hw: pointer to hardware structure
4027 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
4032 PMD_INIT_FUNC_TRACE();
4033 /*PF VF Transmit Enable*/
4034 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
4035 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
4037 /* Disable the Tx desc arbiter so that MTQC can be changed */
4038 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4039 reg |= IXGBE_RTTDCS_ARBDIS;
4040 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
4042 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
4043 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
4045 /* Disable drop for all queues */
4046 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
4047 IXGBE_WRITE_REG(hw, IXGBE_QDE,
4048 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
4050 /* Enable the Tx desc arbiter */
4051 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4052 reg &= ~IXGBE_RTTDCS_ARBDIS;
4053 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
4055 IXGBE_WRITE_FLUSH(hw);
4058 static int __attribute__((cold))
4059 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
4061 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
4065 /* Initialize software ring entries */
4066 for (i = 0; i < rxq->nb_rx_desc; i++) {
4067 volatile union ixgbe_adv_rx_desc *rxd;
4068 struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
4071 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
4072 (unsigned) rxq->queue_id);
4076 rte_mbuf_refcnt_set(mbuf, 1);
4078 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
4080 mbuf->port = rxq->port_id;
4083 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(mbuf));
4084 rxd = &rxq->rx_ring[i];
4085 rxd->read.hdr_addr = 0;
4086 rxd->read.pkt_addr = dma_addr;
4094 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
4096 struct ixgbe_hw *hw;
4099 ixgbe_rss_configure(dev);
4101 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4103 /* MRQC: enable VF RSS */
4104 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
4105 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
4106 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4108 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
4112 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
4116 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
4120 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4126 ixgbe_config_vf_default(struct rte_eth_dev *dev)
4128 struct ixgbe_hw *hw =
4129 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4131 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4133 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4138 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4139 IXGBE_MRQC_VMDQRT4TCEN);
4143 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4144 IXGBE_MRQC_VMDQRT8TCEN);
4148 "invalid pool number in IOV mode");
4155 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
4157 struct ixgbe_hw *hw =
4158 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4160 if (hw->mac.type == ixgbe_mac_82598EB)
4163 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4165 * SRIOV inactive scheme
4166 * any DCB/RSS w/o VMDq multi-queue setting
4168 switch (dev->data->dev_conf.rxmode.mq_mode) {
4170 case ETH_MQ_RX_DCB_RSS:
4171 case ETH_MQ_RX_VMDQ_RSS:
4172 ixgbe_rss_configure(dev);
4175 case ETH_MQ_RX_VMDQ_DCB:
4176 ixgbe_vmdq_dcb_configure(dev);
4179 case ETH_MQ_RX_VMDQ_ONLY:
4180 ixgbe_vmdq_rx_hw_configure(dev);
4183 case ETH_MQ_RX_NONE:
4185 /* if mq_mode is none, disable rss mode.*/
4186 ixgbe_rss_disable(dev);
4191 * SRIOV active scheme
4192 * Support RSS together with VMDq & SRIOV
4194 switch (dev->data->dev_conf.rxmode.mq_mode) {
4196 case ETH_MQ_RX_VMDQ_RSS:
4197 ixgbe_config_vf_rss(dev);
4199 case ETH_MQ_RX_VMDQ_DCB:
4200 ixgbe_vmdq_dcb_configure(dev);
4202 /* FIXME if support DCB/RSS together with VMDq & SRIOV */
4203 case ETH_MQ_RX_VMDQ_DCB_RSS:
4205 "Could not support DCB/RSS with VMDq & SRIOV");
4208 ixgbe_config_vf_default(dev);
4217 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
4219 struct ixgbe_hw *hw =
4220 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4224 if (hw->mac.type == ixgbe_mac_82598EB)
4227 /* disable arbiter before setting MTQC */
4228 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4229 rttdcs |= IXGBE_RTTDCS_ARBDIS;
4230 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4232 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4234 * SRIOV inactive scheme
4235 * any DCB w/o VMDq multi-queue setting
4237 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
4238 ixgbe_vmdq_tx_hw_configure(hw);
4240 mtqc = IXGBE_MTQC_64Q_1PB;
4241 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4244 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4247 * SRIOV active scheme
4248 * FIXME if support DCB together with VMDq & SRIOV
4251 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
4254 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
4257 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
4261 mtqc = IXGBE_MTQC_64Q_1PB;
4262 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
4264 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4267 /* re-enable arbiter */
4268 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
4269 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4275 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
4277 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
4278 * spec rev. 3.0 chapter 8.2.3.8.13.
4280 * @pool Memory pool of the Rx queue
4282 static inline uint32_t
4283 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
4285 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
4287 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
4290 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
4293 return IXGBE_RSCCTL_MAXDESC_16;
4294 else if (maxdesc >= 8)
4295 return IXGBE_RSCCTL_MAXDESC_8;
4296 else if (maxdesc >= 4)
4297 return IXGBE_RSCCTL_MAXDESC_4;
4299 return IXGBE_RSCCTL_MAXDESC_1;
4303 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
4306 * (Taken from FreeBSD tree)
4307 * (yes this is all very magic and confusing :)
4310 * @entry the register array entry
4311 * @vector the MSIX vector for this queue
4315 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
4317 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4320 vector |= IXGBE_IVAR_ALLOC_VAL;
4322 switch (hw->mac.type) {
4324 case ixgbe_mac_82598EB:
4326 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
4328 entry += (type * 64);
4329 index = (entry >> 2) & 0x1F;
4330 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4331 ivar &= ~(0xFF << (8 * (entry & 0x3)));
4332 ivar |= (vector << (8 * (entry & 0x3)));
4333 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4336 case ixgbe_mac_82599EB:
4337 case ixgbe_mac_X540:
4338 if (type == -1) { /* MISC IVAR */
4339 index = (entry & 1) * 8;
4340 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4341 ivar &= ~(0xFF << index);
4342 ivar |= (vector << index);
4343 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4344 } else { /* RX/TX IVARS */
4345 index = (16 * (entry & 1)) + (8 * type);
4346 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
4347 ivar &= ~(0xFF << index);
4348 ivar |= (vector << index);
4349 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
4359 void __attribute__((cold))
4360 ixgbe_set_rx_function(struct rte_eth_dev *dev)
4362 uint16_t i, rx_using_sse;
4363 struct ixgbe_adapter *adapter =
4364 (struct ixgbe_adapter *)dev->data->dev_private;
4367 * In order to allow Vector Rx there are a few configuration
4368 * conditions to be met and Rx Bulk Allocation should be allowed.
4370 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
4371 !adapter->rx_bulk_alloc_allowed) {
4372 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
4373 "preconditions or RTE_IXGBE_INC_VECTOR is "
4375 dev->data->port_id);
4377 adapter->rx_vec_allowed = false;
4381 * Initialize the appropriate LRO callback.
4383 * If all queues satisfy the bulk allocation preconditions
4384 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
4385 * Otherwise use a single allocation version.
4387 if (dev->data->lro) {
4388 if (adapter->rx_bulk_alloc_allowed) {
4389 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a bulk "
4390 "allocation version");
4391 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4393 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a single "
4394 "allocation version");
4395 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4397 } else if (dev->data->scattered_rx) {
4399 * Set the non-LRO scattered callback: there are Vector and
4400 * single allocation versions.
4402 if (adapter->rx_vec_allowed) {
4403 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
4404 "callback (port=%d).",
4405 dev->data->port_id);
4407 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
4408 } else if (adapter->rx_bulk_alloc_allowed) {
4409 PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
4410 "allocation callback (port=%d).",
4411 dev->data->port_id);
4412 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4414 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
4415 "single allocation) "
4416 "Scattered Rx callback "
4418 dev->data->port_id);
4420 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4423 * Below we set "simple" callbacks according to port/queues parameters.
4424 * If parameters allow we are going to choose between the following
4428 * - Single buffer allocation (the simplest one)
4430 } else if (adapter->rx_vec_allowed) {
4431 PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
4432 "burst size no less than %d (port=%d).",
4433 RTE_IXGBE_DESCS_PER_LOOP,
4434 dev->data->port_id);
4436 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
4437 } else if (adapter->rx_bulk_alloc_allowed) {
4438 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
4439 "satisfied. Rx Burst Bulk Alloc function "
4440 "will be used on port=%d.",
4441 dev->data->port_id);
4443 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
4445 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
4446 "satisfied, or Scattered Rx is requested "
4448 dev->data->port_id);
4450 dev->rx_pkt_burst = ixgbe_recv_pkts;
4453 /* Propagate information about RX function choice through all queues. */
4456 (dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec ||
4457 dev->rx_pkt_burst == ixgbe_recv_pkts_vec);
4459 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4460 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4462 rxq->rx_using_sse = rx_using_sse;
4467 * ixgbe_set_rsc - configure RSC related port HW registers
4469 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4470 * of 82599 Spec (x540 configuration is virtually the same).
4474 * Returns 0 in case of success or a non-zero error code
4477 ixgbe_set_rsc(struct rte_eth_dev *dev)
4479 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4480 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4481 struct rte_eth_dev_info dev_info = { 0 };
4482 bool rsc_capable = false;
4487 dev->dev_ops->dev_infos_get(dev, &dev_info);
4488 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4491 if (!rsc_capable && rx_conf->enable_lro) {
4492 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4497 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4499 if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
4501 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4502 * 3.0 RSC configuration requires HW CRC stripping being
4503 * enabled. If user requested both HW CRC stripping off
4504 * and RSC on - return an error.
4506 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4511 /* RFCTL configuration */
4513 uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4515 if (rx_conf->enable_lro)
4517 * Since NFS packets coalescing is not supported - clear
4518 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4521 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4522 IXGBE_RFCTL_NFSR_DIS);
4524 rfctl |= IXGBE_RFCTL_RSC_DIS;
4526 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4529 /* If LRO hasn't been requested - we are done here. */
4530 if (!rx_conf->enable_lro)
4533 /* Set RDRXCTL.RSCACKC bit */
4534 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4535 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4536 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4538 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4539 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4540 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4542 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4544 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4546 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4548 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4551 * ixgbe PMD doesn't support header-split at the moment.
4553 * Following the 4.6.7.2.1 chapter of the 82599/x540
4554 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4555 * should be configured even if header split is not
4556 * enabled. We will configure it 128 bytes following the
4557 * recommendation in the spec.
4559 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4560 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4561 IXGBE_SRRCTL_BSIZEHDR_MASK;
4564 * TODO: Consider setting the Receive Descriptor Minimum
4565 * Threshold Size for an RSC case. This is not an obviously
4566 * beneficiary option but the one worth considering...
4569 rscctl |= IXGBE_RSCCTL_RSCEN;
4570 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4571 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4574 * RSC: Set ITR interval corresponding to 2K ints/s.
4576 * Full-sized RSC aggregations for a 10Gb/s link will
4577 * arrive at about 20K aggregation/s rate.
4579 * 2K inst/s rate will make only 10% of the
4580 * aggregations to be closed due to the interrupt timer
4581 * expiration for a streaming at wire-speed case.
4583 * For a sparse streaming case this setting will yield
4584 * at most 500us latency for a single RSC aggregation.
4586 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4587 eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
4589 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4590 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4591 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4592 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4595 * RSC requires the mapping of the queue to the
4598 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4603 PMD_INIT_LOG(DEBUG, "enabling LRO mode");
4609 * Initializes Receive Unit.
4611 int __attribute__((cold))
4612 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4614 struct ixgbe_hw *hw;
4615 struct ixgbe_rx_queue *rxq;
4626 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4629 PMD_INIT_FUNC_TRACE();
4630 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4633 * Make sure receives are disabled while setting
4634 * up the RX context (registers, descriptor rings, etc.).
4636 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4637 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4639 /* Enable receipt of broadcasted frames */
4640 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4641 fctrl |= IXGBE_FCTRL_BAM;
4642 fctrl |= IXGBE_FCTRL_DPF;
4643 fctrl |= IXGBE_FCTRL_PMCF;
4644 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4647 * Configure CRC stripping, if any.
4649 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4650 if (rx_conf->hw_strip_crc)
4651 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4653 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4656 * Configure jumbo frame support, if any.
4658 if (rx_conf->jumbo_frame == 1) {
4659 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4660 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4661 maxfrs &= 0x0000FFFF;
4662 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4663 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4665 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4668 * If loopback mode is configured for 82599, set LPBK bit.
4670 if (hw->mac.type == ixgbe_mac_82599EB &&
4671 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4672 hlreg0 |= IXGBE_HLREG0_LPBK;
4674 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4676 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4678 /* Setup RX queues */
4679 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4680 rxq = dev->data->rx_queues[i];
4683 * Reset crc_len in case it was changed after queue setup by a
4684 * call to configure.
4686 rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
4688 /* Setup the Base and Length of the Rx Descriptor Rings */
4689 bus_addr = rxq->rx_ring_phys_addr;
4690 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4691 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4692 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4693 (uint32_t)(bus_addr >> 32));
4694 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4695 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4696 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4697 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4699 /* Configure the SRRCTL register */
4700 #ifdef RTE_HEADER_SPLIT_ENABLE
4702 * Configure Header Split
4704 if (rx_conf->header_split) {
4705 if (hw->mac.type == ixgbe_mac_82599EB) {
4706 /* Must setup the PSRTYPE register */
4709 psrtype = IXGBE_PSRTYPE_TCPHDR |
4710 IXGBE_PSRTYPE_UDPHDR |
4711 IXGBE_PSRTYPE_IPV4HDR |
4712 IXGBE_PSRTYPE_IPV6HDR;
4713 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4715 srrctl = ((rx_conf->split_hdr_size <<
4716 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4717 IXGBE_SRRCTL_BSIZEHDR_MASK);
4718 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4721 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4723 /* Set if packets are dropped when no descriptors available */
4725 srrctl |= IXGBE_SRRCTL_DROP_EN;
4728 * Configure the RX buffer size in the BSIZEPACKET field of
4729 * the SRRCTL register of the queue.
4730 * The value is in 1 KB resolution. Valid values can be from
4733 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4734 RTE_PKTMBUF_HEADROOM);
4735 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4736 IXGBE_SRRCTL_BSIZEPKT_MASK);
4738 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4740 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4741 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4743 /* It adds dual VLAN length for supporting dual VLAN */
4744 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4745 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4746 dev->data->scattered_rx = 1;
4749 if (rx_conf->enable_scatter)
4750 dev->data->scattered_rx = 1;
4753 * Device configured with multiple RX queues.
4755 ixgbe_dev_mq_rx_configure(dev);
4758 * Setup the Checksum Register.
4759 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4760 * Enable IP/L4 checkum computation by hardware if requested to do so.
4762 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4763 rxcsum |= IXGBE_RXCSUM_PCSD;
4764 if (rx_conf->hw_ip_checksum)
4765 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4767 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4769 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4771 if (hw->mac.type == ixgbe_mac_82599EB ||
4772 hw->mac.type == ixgbe_mac_X540) {
4773 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4774 if (rx_conf->hw_strip_crc)
4775 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4777 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4778 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4779 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4782 rc = ixgbe_set_rsc(dev);
4786 ixgbe_set_rx_function(dev);
4792 * Initializes Transmit Unit.
4794 void __attribute__((cold))
4795 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4797 struct ixgbe_hw *hw;
4798 struct ixgbe_tx_queue *txq;
4804 PMD_INIT_FUNC_TRACE();
4805 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4807 /* Enable TX CRC (checksum offload requirement) and hw padding
4810 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4811 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
4812 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4814 /* Setup the Base and Length of the Tx Descriptor Rings */
4815 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4816 txq = dev->data->tx_queues[i];
4818 bus_addr = txq->tx_ring_phys_addr;
4819 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
4820 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4821 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
4822 (uint32_t)(bus_addr >> 32));
4823 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
4824 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4825 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4826 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4827 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4830 * Disable Tx Head Writeback RO bit, since this hoses
4831 * bookkeeping if things aren't delivered in order.
4833 switch (hw->mac.type) {
4834 case ixgbe_mac_82598EB:
4835 txctrl = IXGBE_READ_REG(hw,
4836 IXGBE_DCA_TXCTRL(txq->reg_idx));
4837 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4838 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
4842 case ixgbe_mac_82599EB:
4843 case ixgbe_mac_X540:
4844 case ixgbe_mac_X550:
4845 case ixgbe_mac_X550EM_x:
4846 case ixgbe_mac_X550EM_a:
4848 txctrl = IXGBE_READ_REG(hw,
4849 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
4850 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4851 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
4857 /* Device configured with multiple TX queues. */
4858 ixgbe_dev_mq_tx_configure(dev);
4862 * Set up link for 82599 loopback mode Tx->Rx.
4864 static inline void __attribute__((cold))
4865 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
4867 PMD_INIT_FUNC_TRACE();
4869 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
4870 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
4872 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
4881 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
4882 ixgbe_reset_pipeline_82599(hw);
4884 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
4890 * Start Transmit and Receive Units.
4892 int __attribute__((cold))
4893 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
4895 struct ixgbe_hw *hw;
4896 struct ixgbe_tx_queue *txq;
4897 struct ixgbe_rx_queue *rxq;
4904 PMD_INIT_FUNC_TRACE();
4905 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4907 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4908 txq = dev->data->tx_queues[i];
4909 /* Setup Transmit Threshold Registers */
4910 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4911 txdctl |= txq->pthresh & 0x7F;
4912 txdctl |= ((txq->hthresh & 0x7F) << 8);
4913 txdctl |= ((txq->wthresh & 0x7F) << 16);
4914 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4917 if (hw->mac.type != ixgbe_mac_82598EB) {
4918 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
4919 dmatxctl |= IXGBE_DMATXCTL_TE;
4920 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
4923 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4924 txq = dev->data->tx_queues[i];
4925 if (!txq->tx_deferred_start) {
4926 ret = ixgbe_dev_tx_queue_start(dev, i);
4932 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4933 rxq = dev->data->rx_queues[i];
4934 if (!rxq->rx_deferred_start) {
4935 ret = ixgbe_dev_rx_queue_start(dev, i);
4941 /* Enable Receive engine */
4942 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4943 if (hw->mac.type == ixgbe_mac_82598EB)
4944 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4945 rxctrl |= IXGBE_RXCTRL_RXEN;
4946 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4948 /* If loopback mode is enabled for 82599, set up the link accordingly */
4949 if (hw->mac.type == ixgbe_mac_82599EB &&
4950 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4951 ixgbe_setup_loopback_link_82599(hw);
4957 * Start Receive Units for specified queue.
4959 int __attribute__((cold))
4960 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4962 struct ixgbe_hw *hw;
4963 struct ixgbe_rx_queue *rxq;
4967 PMD_INIT_FUNC_TRACE();
4968 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4970 if (rx_queue_id < dev->data->nb_rx_queues) {
4971 rxq = dev->data->rx_queues[rx_queue_id];
4973 /* Allocate buffers for descriptor rings */
4974 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
4975 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
4979 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4980 rxdctl |= IXGBE_RXDCTL_ENABLE;
4981 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4983 /* Wait until RX Enable ready */
4984 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4987 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4988 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4990 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
4993 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4994 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
4995 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
5003 * Stop Receive Units for specified queue.
5005 int __attribute__((cold))
5006 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
5008 struct ixgbe_hw *hw;
5009 struct ixgbe_adapter *adapter =
5010 (struct ixgbe_adapter *)dev->data->dev_private;
5011 struct ixgbe_rx_queue *rxq;
5015 PMD_INIT_FUNC_TRACE();
5016 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5018 if (rx_queue_id < dev->data->nb_rx_queues) {
5019 rxq = dev->data->rx_queues[rx_queue_id];
5021 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
5022 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5023 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
5025 /* Wait until RX Enable bit clear */
5026 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5029 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
5030 } while (--poll_ms && (rxdctl & IXGBE_RXDCTL_ENABLE));
5032 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
5035 rte_delay_us(RTE_IXGBE_WAIT_100_US);
5037 ixgbe_rx_queue_release_mbufs(rxq);
5038 ixgbe_reset_rx_queue(adapter, rxq);
5039 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
5048 * Start Transmit Units for specified queue.
5050 int __attribute__((cold))
5051 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
5053 struct ixgbe_hw *hw;
5054 struct ixgbe_tx_queue *txq;
5058 PMD_INIT_FUNC_TRACE();
5059 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5061 if (tx_queue_id < dev->data->nb_tx_queues) {
5062 txq = dev->data->tx_queues[tx_queue_id];
5063 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5064 txdctl |= IXGBE_TXDCTL_ENABLE;
5065 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5067 /* Wait until TX Enable ready */
5068 if (hw->mac.type == ixgbe_mac_82599EB) {
5069 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5072 txdctl = IXGBE_READ_REG(hw,
5073 IXGBE_TXDCTL(txq->reg_idx));
5074 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
5076 PMD_INIT_LOG(ERR, "Could not enable "
5077 "Tx Queue %d", tx_queue_id);
5080 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
5081 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
5082 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
5090 * Stop Transmit Units for specified queue.
5092 int __attribute__((cold))
5093 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
5095 struct ixgbe_hw *hw;
5096 struct ixgbe_tx_queue *txq;
5098 uint32_t txtdh, txtdt;
5101 PMD_INIT_FUNC_TRACE();
5102 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5104 if (tx_queue_id >= dev->data->nb_tx_queues)
5107 txq = dev->data->tx_queues[tx_queue_id];
5109 /* Wait until TX queue is empty */
5110 if (hw->mac.type == ixgbe_mac_82599EB) {
5111 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5113 rte_delay_us(RTE_IXGBE_WAIT_100_US);
5114 txtdh = IXGBE_READ_REG(hw,
5115 IXGBE_TDH(txq->reg_idx));
5116 txtdt = IXGBE_READ_REG(hw,
5117 IXGBE_TDT(txq->reg_idx));
5118 } while (--poll_ms && (txtdh != txtdt));
5120 PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
5121 "when stopping.", tx_queue_id);
5124 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5125 txdctl &= ~IXGBE_TXDCTL_ENABLE;
5126 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5128 /* Wait until TX Enable bit clear */
5129 if (hw->mac.type == ixgbe_mac_82599EB) {
5130 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5133 txdctl = IXGBE_READ_REG(hw,
5134 IXGBE_TXDCTL(txq->reg_idx));
5135 } while (--poll_ms && (txdctl & IXGBE_TXDCTL_ENABLE));
5137 PMD_INIT_LOG(ERR, "Could not disable "
5138 "Tx Queue %d", tx_queue_id);
5141 if (txq->ops != NULL) {
5142 txq->ops->release_mbufs(txq);
5143 txq->ops->reset(txq);
5145 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
5151 ixgbe_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5152 struct rte_eth_rxq_info *qinfo)
5154 struct ixgbe_rx_queue *rxq;
5156 rxq = dev->data->rx_queues[queue_id];
5158 qinfo->mp = rxq->mb_pool;
5159 qinfo->scattered_rx = dev->data->scattered_rx;
5160 qinfo->nb_desc = rxq->nb_rx_desc;
5162 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
5163 qinfo->conf.rx_drop_en = rxq->drop_en;
5164 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
5168 ixgbe_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5169 struct rte_eth_txq_info *qinfo)
5171 struct ixgbe_tx_queue *txq;
5173 txq = dev->data->tx_queues[queue_id];
5175 qinfo->nb_desc = txq->nb_tx_desc;
5177 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
5178 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
5179 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
5181 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
5182 qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
5183 qinfo->conf.txq_flags = txq->txq_flags;
5184 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
5188 * [VF] Initializes Receive Unit.
5190 int __attribute__((cold))
5191 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
5193 struct ixgbe_hw *hw;
5194 struct ixgbe_rx_queue *rxq;
5196 uint32_t srrctl, psrtype = 0;
5201 PMD_INIT_FUNC_TRACE();
5202 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5204 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
5205 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5206 "it should be power of 2");
5210 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
5211 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5212 "it should be equal to or less than %d",
5213 hw->mac.max_rx_queues);
5218 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
5219 * disables the VF receipt of packets if the PF MTU is > 1500.
5220 * This is done to deal with 82599 limitations that imposes
5221 * the PF and all VFs to share the same MTU.
5222 * Then, the PF driver enables again the VF receipt of packet when
5223 * the VF driver issues a IXGBE_VF_SET_LPE request.
5224 * In the meantime, the VF device cannot be used, even if the VF driver
5225 * and the Guest VM network stack are ready to accept packets with a
5226 * size up to the PF MTU.
5227 * As a work-around to this PF behaviour, force the call to
5228 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
5229 * VF packets received can work in all cases.
5231 ixgbevf_rlpml_set_vf(hw,
5232 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
5234 /* Setup RX queues */
5235 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5236 rxq = dev->data->rx_queues[i];
5238 /* Allocate buffers for descriptor rings */
5239 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
5243 /* Setup the Base and Length of the Rx Descriptor Rings */
5244 bus_addr = rxq->rx_ring_phys_addr;
5246 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
5247 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5248 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
5249 (uint32_t)(bus_addr >> 32));
5250 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
5251 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
5252 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
5253 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
5256 /* Configure the SRRCTL register */
5257 #ifdef RTE_HEADER_SPLIT_ENABLE
5259 * Configure Header Split
5261 if (dev->data->dev_conf.rxmode.header_split) {
5262 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
5263 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
5264 IXGBE_SRRCTL_BSIZEHDR_MASK);
5265 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
5268 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
5270 /* Set if packets are dropped when no descriptors available */
5272 srrctl |= IXGBE_SRRCTL_DROP_EN;
5275 * Configure the RX buffer size in the BSIZEPACKET field of
5276 * the SRRCTL register of the queue.
5277 * The value is in 1 KB resolution. Valid values can be from
5280 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
5281 RTE_PKTMBUF_HEADROOM);
5282 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
5283 IXGBE_SRRCTL_BSIZEPKT_MASK);
5286 * VF modification to write virtual function SRRCTL register
5288 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
5290 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
5291 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
5293 if (dev->data->dev_conf.rxmode.enable_scatter ||
5294 /* It adds dual VLAN length for supporting dual VLAN */
5295 (dev->data->dev_conf.rxmode.max_rx_pkt_len +
5296 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
5297 if (!dev->data->scattered_rx)
5298 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
5299 dev->data->scattered_rx = 1;
5303 #ifdef RTE_HEADER_SPLIT_ENABLE
5304 if (dev->data->dev_conf.rxmode.header_split)
5305 /* Must setup the PSRTYPE register */
5306 psrtype = IXGBE_PSRTYPE_TCPHDR |
5307 IXGBE_PSRTYPE_UDPHDR |
5308 IXGBE_PSRTYPE_IPV4HDR |
5309 IXGBE_PSRTYPE_IPV6HDR;
5312 /* Set RQPL for VF RSS according to max Rx queue */
5313 psrtype |= (dev->data->nb_rx_queues >> 1) <<
5314 IXGBE_PSRTYPE_RQPL_SHIFT;
5315 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
5317 ixgbe_set_rx_function(dev);
5323 * [VF] Initializes Transmit Unit.
5325 void __attribute__((cold))
5326 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
5328 struct ixgbe_hw *hw;
5329 struct ixgbe_tx_queue *txq;
5334 PMD_INIT_FUNC_TRACE();
5335 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5337 /* Setup the Base and Length of the Tx Descriptor Rings */
5338 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5339 txq = dev->data->tx_queues[i];
5340 bus_addr = txq->tx_ring_phys_addr;
5341 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
5342 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5343 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
5344 (uint32_t)(bus_addr >> 32));
5345 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
5346 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
5347 /* Setup the HW Tx Head and TX Tail descriptor pointers */
5348 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
5349 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
5352 * Disable Tx Head Writeback RO bit, since this hoses
5353 * bookkeeping if things aren't delivered in order.
5355 txctrl = IXGBE_READ_REG(hw,
5356 IXGBE_VFDCA_TXCTRL(i));
5357 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
5358 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
5364 * [VF] Start Transmit and Receive Units.
5366 void __attribute__((cold))
5367 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
5369 struct ixgbe_hw *hw;
5370 struct ixgbe_tx_queue *txq;
5371 struct ixgbe_rx_queue *rxq;
5377 PMD_INIT_FUNC_TRACE();
5378 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5380 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5381 txq = dev->data->tx_queues[i];
5382 /* Setup Transmit Threshold Registers */
5383 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5384 txdctl |= txq->pthresh & 0x7F;
5385 txdctl |= ((txq->hthresh & 0x7F) << 8);
5386 txdctl |= ((txq->wthresh & 0x7F) << 16);
5387 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5390 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5392 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5393 txdctl |= IXGBE_TXDCTL_ENABLE;
5394 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5397 /* Wait until TX Enable ready */
5400 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5401 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
5403 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
5405 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5407 rxq = dev->data->rx_queues[i];
5409 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5410 rxdctl |= IXGBE_RXDCTL_ENABLE;
5411 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
5413 /* Wait until RX Enable ready */
5417 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5418 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
5420 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
5422 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
5427 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
5428 int __attribute__((weak))
5429 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
5434 uint16_t __attribute__((weak))
5435 ixgbe_recv_pkts_vec(
5436 void __rte_unused *rx_queue,
5437 struct rte_mbuf __rte_unused **rx_pkts,
5438 uint16_t __rte_unused nb_pkts)
5443 uint16_t __attribute__((weak))
5444 ixgbe_recv_scattered_pkts_vec(
5445 void __rte_unused *rx_queue,
5446 struct rte_mbuf __rte_unused **rx_pkts,
5447 uint16_t __rte_unused nb_pkts)
5452 int __attribute__((weak))
5453 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)