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>
74 #include "ixgbe_logs.h"
75 #include "base/ixgbe_api.h"
76 #include "base/ixgbe_vf.h"
77 #include "ixgbe_ethdev.h"
78 #include "base/ixgbe_dcb.h"
79 #include "base/ixgbe_common.h"
80 #include "ixgbe_rxtx.h"
82 /* Bit Mask to indicate what bits required for building TX context */
83 #define IXGBE_TX_OFFLOAD_MASK ( \
88 PKT_TX_OUTER_IP_CKSUM)
91 #define RTE_PMD_USE_PREFETCH
94 #ifdef RTE_PMD_USE_PREFETCH
96 * Prefetch a cache line into all cache levels.
98 #define rte_ixgbe_prefetch(p) rte_prefetch0(p)
100 #define rte_ixgbe_prefetch(p) do {} while (0)
103 /*********************************************************************
107 **********************************************************************/
110 * Check for descriptors with their DD bit set and free mbufs.
111 * Return the total number of buffers freed.
113 static inline int __attribute__((always_inline))
114 ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
116 struct ixgbe_tx_entry *txep;
119 struct rte_mbuf *m, *free[RTE_IXGBE_TX_MAX_FREE_BUF_SZ];
121 /* check DD bit on threshold descriptor */
122 status = txq->tx_ring[txq->tx_next_dd].wb.status;
123 if (!(status & rte_cpu_to_le_32(IXGBE_ADVTXD_STAT_DD)))
127 * first buffer to free from S/W ring is at index
128 * tx_next_dd - (tx_rs_thresh-1)
130 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
132 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
133 /* free buffers one at a time */
134 m = __rte_pktmbuf_prefree_seg(txep->mbuf);
137 if (unlikely(m == NULL))
140 if (nb_free >= RTE_IXGBE_TX_MAX_FREE_BUF_SZ ||
141 (nb_free > 0 && m->pool != free[0]->pool)) {
142 rte_mempool_put_bulk(free[0]->pool,
143 (void **)free, nb_free);
151 rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
153 /* buffers were freed, update counters */
154 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
155 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
156 if (txq->tx_next_dd >= txq->nb_tx_desc)
157 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
159 return txq->tx_rs_thresh;
162 /* Populate 4 descriptors with data from 4 mbufs */
164 tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
166 uint64_t buf_dma_addr;
170 for (i = 0; i < 4; ++i, ++txdp, ++pkts) {
171 buf_dma_addr = rte_mbuf_data_dma_addr(*pkts);
172 pkt_len = (*pkts)->data_len;
174 /* write data to descriptor */
175 txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
177 txdp->read.cmd_type_len =
178 rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
180 txdp->read.olinfo_status =
181 rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
183 rte_prefetch0(&(*pkts)->pool);
187 /* Populate 1 descriptor with data from 1 mbuf */
189 tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
191 uint64_t buf_dma_addr;
194 buf_dma_addr = rte_mbuf_data_dma_addr(*pkts);
195 pkt_len = (*pkts)->data_len;
197 /* write data to descriptor */
198 txdp->read.buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
199 txdp->read.cmd_type_len =
200 rte_cpu_to_le_32((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
201 txdp->read.olinfo_status =
202 rte_cpu_to_le_32(pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
203 rte_prefetch0(&(*pkts)->pool);
207 * Fill H/W descriptor ring with mbuf data.
208 * Copy mbuf pointers to the S/W ring.
211 ixgbe_tx_fill_hw_ring(struct ixgbe_tx_queue *txq, struct rte_mbuf **pkts,
214 volatile union ixgbe_adv_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
215 struct ixgbe_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
216 const int N_PER_LOOP = 4;
217 const int N_PER_LOOP_MASK = N_PER_LOOP-1;
218 int mainpart, leftover;
222 * Process most of the packets in chunks of N pkts. Any
223 * leftover packets will get processed one at a time.
225 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
226 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
227 for (i = 0; i < mainpart; i += N_PER_LOOP) {
228 /* Copy N mbuf pointers to the S/W ring */
229 for (j = 0; j < N_PER_LOOP; ++j) {
230 (txep + i + j)->mbuf = *(pkts + i + j);
232 tx4(txdp + i, pkts + i);
235 if (unlikely(leftover > 0)) {
236 for (i = 0; i < leftover; ++i) {
237 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
238 tx1(txdp + mainpart + i, pkts + mainpart + i);
243 static inline uint16_t
244 tx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
247 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
248 volatile union ixgbe_adv_tx_desc *tx_r = txq->tx_ring;
252 * Begin scanning the H/W ring for done descriptors when the
253 * number of available descriptors drops below tx_free_thresh. For
254 * each done descriptor, free the associated buffer.
256 if (txq->nb_tx_free < txq->tx_free_thresh)
257 ixgbe_tx_free_bufs(txq);
259 /* Only use descriptors that are available */
260 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
261 if (unlikely(nb_pkts == 0))
264 /* Use exactly nb_pkts descriptors */
265 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
268 * At this point, we know there are enough descriptors in the
269 * ring to transmit all the packets. This assumes that each
270 * mbuf contains a single segment, and that no new offloads
271 * are expected, which would require a new context descriptor.
275 * See if we're going to wrap-around. If so, handle the top
276 * of the descriptor ring first, then do the bottom. If not,
277 * the processing looks just like the "bottom" part anyway...
279 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
280 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
281 ixgbe_tx_fill_hw_ring(txq, tx_pkts, n);
284 * We know that the last descriptor in the ring will need to
285 * have its RS bit set because tx_rs_thresh has to be
286 * a divisor of the ring size
288 tx_r[txq->tx_next_rs].read.cmd_type_len |=
289 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
290 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
295 /* Fill H/W descriptor ring with mbuf data */
296 ixgbe_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
297 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
300 * Determine if RS bit should be set
301 * This is what we actually want:
302 * if ((txq->tx_tail - 1) >= txq->tx_next_rs)
303 * but instead of subtracting 1 and doing >=, we can just do
304 * greater than without subtracting.
306 if (txq->tx_tail > txq->tx_next_rs) {
307 tx_r[txq->tx_next_rs].read.cmd_type_len |=
308 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
309 txq->tx_next_rs = (uint16_t)(txq->tx_next_rs +
311 if (txq->tx_next_rs >= txq->nb_tx_desc)
312 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
316 * Check for wrap-around. This would only happen if we used
317 * up to the last descriptor in the ring, no more, no less.
319 if (txq->tx_tail >= txq->nb_tx_desc)
322 /* update tail pointer */
324 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, txq->tx_tail);
330 ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
335 /* Try to transmit at least chunks of TX_MAX_BURST pkts */
336 if (likely(nb_pkts <= RTE_PMD_IXGBE_TX_MAX_BURST))
337 return tx_xmit_pkts(tx_queue, tx_pkts, nb_pkts);
339 /* transmit more than the max burst, in chunks of TX_MAX_BURST */
344 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_TX_MAX_BURST);
345 ret = tx_xmit_pkts(tx_queue, &(tx_pkts[nb_tx]), n);
346 nb_tx = (uint16_t)(nb_tx + ret);
347 nb_pkts = (uint16_t)(nb_pkts - ret);
356 ixgbe_set_xmit_ctx(struct ixgbe_tx_queue *txq,
357 volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
358 uint64_t ol_flags, union ixgbe_tx_offload tx_offload)
360 uint32_t type_tucmd_mlhl;
361 uint32_t mss_l4len_idx = 0;
363 uint32_t vlan_macip_lens;
364 union ixgbe_tx_offload tx_offload_mask;
365 uint32_t seqnum_seed = 0;
367 ctx_idx = txq->ctx_curr;
368 tx_offload_mask.data[0] = 0;
369 tx_offload_mask.data[1] = 0;
372 /* Specify which HW CTX to upload. */
373 mss_l4len_idx |= (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
375 if (ol_flags & PKT_TX_VLAN_PKT) {
376 tx_offload_mask.vlan_tci |= ~0;
379 /* check if TCP segmentation required for this packet */
380 if (ol_flags & PKT_TX_TCP_SEG) {
381 /* implies IP cksum in IPv4 */
382 if (ol_flags & PKT_TX_IP_CKSUM)
383 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4 |
384 IXGBE_ADVTXD_TUCMD_L4T_TCP |
385 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
387 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV6 |
388 IXGBE_ADVTXD_TUCMD_L4T_TCP |
389 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
391 tx_offload_mask.l2_len |= ~0;
392 tx_offload_mask.l3_len |= ~0;
393 tx_offload_mask.l4_len |= ~0;
394 tx_offload_mask.tso_segsz |= ~0;
395 mss_l4len_idx |= tx_offload.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT;
396 mss_l4len_idx |= tx_offload.l4_len << IXGBE_ADVTXD_L4LEN_SHIFT;
397 } else { /* no TSO, check if hardware checksum is needed */
398 if (ol_flags & PKT_TX_IP_CKSUM) {
399 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
400 tx_offload_mask.l2_len |= ~0;
401 tx_offload_mask.l3_len |= ~0;
404 switch (ol_flags & PKT_TX_L4_MASK) {
405 case PKT_TX_UDP_CKSUM:
406 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
407 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
408 mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
409 tx_offload_mask.l2_len |= ~0;
410 tx_offload_mask.l3_len |= ~0;
412 case PKT_TX_TCP_CKSUM:
413 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
414 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
415 mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
416 tx_offload_mask.l2_len |= ~0;
417 tx_offload_mask.l3_len |= ~0;
419 case PKT_TX_SCTP_CKSUM:
420 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
421 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
422 mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
423 tx_offload_mask.l2_len |= ~0;
424 tx_offload_mask.l3_len |= ~0;
427 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
428 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
433 if (ol_flags & PKT_TX_OUTER_IP_CKSUM) {
434 tx_offload_mask.outer_l2_len |= ~0;
435 tx_offload_mask.outer_l3_len |= ~0;
436 tx_offload_mask.l2_len |= ~0;
437 seqnum_seed |= tx_offload.outer_l3_len
438 << IXGBE_ADVTXD_OUTER_IPLEN;
439 seqnum_seed |= tx_offload.l2_len
440 << IXGBE_ADVTXD_TUNNEL_LEN;
443 txq->ctx_cache[ctx_idx].flags = ol_flags;
444 txq->ctx_cache[ctx_idx].tx_offload.data[0] =
445 tx_offload_mask.data[0] & tx_offload.data[0];
446 txq->ctx_cache[ctx_idx].tx_offload.data[1] =
447 tx_offload_mask.data[1] & tx_offload.data[1];
448 txq->ctx_cache[ctx_idx].tx_offload_mask = tx_offload_mask;
450 ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
451 vlan_macip_lens = tx_offload.l3_len;
452 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
453 vlan_macip_lens |= (tx_offload.outer_l2_len <<
454 IXGBE_ADVTXD_MACLEN_SHIFT);
456 vlan_macip_lens |= (tx_offload.l2_len <<
457 IXGBE_ADVTXD_MACLEN_SHIFT);
458 vlan_macip_lens |= ((uint32_t)tx_offload.vlan_tci << IXGBE_ADVTXD_VLAN_SHIFT);
459 ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
460 ctx_txd->mss_l4len_idx = rte_cpu_to_le_32(mss_l4len_idx);
461 ctx_txd->seqnum_seed = seqnum_seed;
465 * Check which hardware context can be used. Use the existing match
466 * or create a new context descriptor.
468 static inline uint32_t
469 what_advctx_update(struct ixgbe_tx_queue *txq, uint64_t flags,
470 union ixgbe_tx_offload tx_offload)
472 /* If match with the current used context */
473 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
474 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[0] ==
475 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[0]
476 & tx_offload.data[0])) &&
477 (txq->ctx_cache[txq->ctx_curr].tx_offload.data[1] ==
478 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data[1]
479 & tx_offload.data[1]))))
480 return txq->ctx_curr;
482 /* What if match with the next 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 /* Mismatch, use the previous context */
494 return IXGBE_CTX_NUM;
497 static inline uint32_t
498 tx_desc_cksum_flags_to_olinfo(uint64_t ol_flags)
502 if ((ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM)
503 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
504 if (ol_flags & PKT_TX_IP_CKSUM)
505 tmp |= IXGBE_ADVTXD_POPTS_IXSM;
506 if (ol_flags & PKT_TX_TCP_SEG)
507 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
511 static inline uint32_t
512 tx_desc_ol_flags_to_cmdtype(uint64_t ol_flags)
514 uint32_t cmdtype = 0;
516 if (ol_flags & PKT_TX_VLAN_PKT)
517 cmdtype |= IXGBE_ADVTXD_DCMD_VLE;
518 if (ol_flags & PKT_TX_TCP_SEG)
519 cmdtype |= IXGBE_ADVTXD_DCMD_TSE;
520 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
521 cmdtype |= (1 << IXGBE_ADVTXD_OUTERIPCS_SHIFT);
525 /* Default RS bit threshold values */
526 #ifndef DEFAULT_TX_RS_THRESH
527 #define DEFAULT_TX_RS_THRESH 32
529 #ifndef DEFAULT_TX_FREE_THRESH
530 #define DEFAULT_TX_FREE_THRESH 32
533 /* Reset transmit descriptors after they have been used */
535 ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
537 struct ixgbe_tx_entry *sw_ring = txq->sw_ring;
538 volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
539 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
540 uint16_t nb_tx_desc = txq->nb_tx_desc;
541 uint16_t desc_to_clean_to;
542 uint16_t nb_tx_to_clean;
545 /* Determine the last descriptor needing to be cleaned */
546 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
547 if (desc_to_clean_to >= nb_tx_desc)
548 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
550 /* Check to make sure the last descriptor to clean is done */
551 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
552 status = txr[desc_to_clean_to].wb.status;
553 if (!(status & rte_cpu_to_le_32(IXGBE_TXD_STAT_DD))) {
554 PMD_TX_FREE_LOG(DEBUG,
555 "TX descriptor %4u is not done"
556 "(port=%d queue=%d)",
558 txq->port_id, txq->queue_id);
559 /* Failed to clean any descriptors, better luck next time */
563 /* Figure out how many descriptors will be cleaned */
564 if (last_desc_cleaned > desc_to_clean_to)
565 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
568 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
571 PMD_TX_FREE_LOG(DEBUG,
572 "Cleaning %4u TX descriptors: %4u to %4u "
573 "(port=%d queue=%d)",
574 nb_tx_to_clean, last_desc_cleaned, desc_to_clean_to,
575 txq->port_id, txq->queue_id);
578 * The last descriptor to clean is done, so that means all the
579 * descriptors from the last descriptor that was cleaned
580 * up to the last descriptor with the RS bit set
581 * are done. Only reset the threshold descriptor.
583 txr[desc_to_clean_to].wb.status = 0;
585 /* Update the txq to reflect the last descriptor that was cleaned */
586 txq->last_desc_cleaned = desc_to_clean_to;
587 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
594 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
597 struct ixgbe_tx_queue *txq;
598 struct ixgbe_tx_entry *sw_ring;
599 struct ixgbe_tx_entry *txe, *txn;
600 volatile union ixgbe_adv_tx_desc *txr;
601 volatile union ixgbe_adv_tx_desc *txd, *txp;
602 struct rte_mbuf *tx_pkt;
603 struct rte_mbuf *m_seg;
604 uint64_t buf_dma_addr;
605 uint32_t olinfo_status;
606 uint32_t cmd_type_len;
617 union ixgbe_tx_offload tx_offload;
619 tx_offload.data[0] = 0;
620 tx_offload.data[1] = 0;
622 sw_ring = txq->sw_ring;
624 tx_id = txq->tx_tail;
625 txe = &sw_ring[tx_id];
628 /* Determine if the descriptor ring needs to be cleaned. */
629 if (txq->nb_tx_free < txq->tx_free_thresh)
630 ixgbe_xmit_cleanup(txq);
632 rte_prefetch0(&txe->mbuf->pool);
635 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
638 pkt_len = tx_pkt->pkt_len;
641 * Determine how many (if any) context descriptors
642 * are needed for offload functionality.
644 ol_flags = tx_pkt->ol_flags;
646 /* If hardware offload required */
647 tx_ol_req = ol_flags & IXGBE_TX_OFFLOAD_MASK;
649 tx_offload.l2_len = tx_pkt->l2_len;
650 tx_offload.l3_len = tx_pkt->l3_len;
651 tx_offload.l4_len = tx_pkt->l4_len;
652 tx_offload.vlan_tci = tx_pkt->vlan_tci;
653 tx_offload.tso_segsz = tx_pkt->tso_segsz;
654 tx_offload.outer_l2_len = tx_pkt->outer_l2_len;
655 tx_offload.outer_l3_len = tx_pkt->outer_l3_len;
657 /* If new context need be built or reuse the exist ctx. */
658 ctx = what_advctx_update(txq, tx_ol_req,
660 /* Only allocate context descriptor if required*/
661 new_ctx = (ctx == IXGBE_CTX_NUM);
666 * Keep track of how many descriptors are used this loop
667 * This will always be the number of segments + the number of
668 * Context descriptors required to transmit the packet
670 nb_used = (uint16_t)(tx_pkt->nb_segs + new_ctx);
673 nb_used + txq->nb_tx_used >= txq->tx_rs_thresh)
674 /* set RS on the previous packet in the burst */
675 txp->read.cmd_type_len |=
676 rte_cpu_to_le_32(IXGBE_TXD_CMD_RS);
679 * The number of descriptors that must be allocated for a
680 * packet is the number of segments of that packet, plus 1
681 * Context Descriptor for the hardware offload, if any.
682 * Determine the last TX descriptor to allocate in the TX ring
683 * for the packet, starting from the current position (tx_id)
686 tx_last = (uint16_t) (tx_id + nb_used - 1);
689 if (tx_last >= txq->nb_tx_desc)
690 tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
692 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
693 " tx_first=%u tx_last=%u",
694 (unsigned) txq->port_id,
695 (unsigned) txq->queue_id,
701 * Make sure there are enough TX descriptors available to
702 * transmit the entire packet.
703 * nb_used better be less than or equal to txq->tx_rs_thresh
705 if (nb_used > txq->nb_tx_free) {
706 PMD_TX_FREE_LOG(DEBUG,
707 "Not enough free TX descriptors "
708 "nb_used=%4u nb_free=%4u "
709 "(port=%d queue=%d)",
710 nb_used, txq->nb_tx_free,
711 txq->port_id, txq->queue_id);
713 if (ixgbe_xmit_cleanup(txq) != 0) {
714 /* Could not clean any descriptors */
720 /* nb_used better be <= txq->tx_rs_thresh */
721 if (unlikely(nb_used > txq->tx_rs_thresh)) {
722 PMD_TX_FREE_LOG(DEBUG,
723 "The number of descriptors needed to "
724 "transmit the packet exceeds the "
725 "RS bit threshold. This will impact "
727 "nb_used=%4u nb_free=%4u "
729 "(port=%d queue=%d)",
730 nb_used, txq->nb_tx_free,
732 txq->port_id, txq->queue_id);
734 * Loop here until there are enough TX
735 * descriptors or until the ring cannot be
738 while (nb_used > txq->nb_tx_free) {
739 if (ixgbe_xmit_cleanup(txq) != 0) {
741 * Could not clean any
753 * By now there are enough free TX descriptors to transmit
758 * Set common flags of all TX Data Descriptors.
760 * The following bits must be set in all Data Descriptors:
761 * - IXGBE_ADVTXD_DTYP_DATA
762 * - IXGBE_ADVTXD_DCMD_DEXT
764 * The following bits must be set in the first Data Descriptor
765 * and are ignored in the other ones:
766 * - IXGBE_ADVTXD_DCMD_IFCS
767 * - IXGBE_ADVTXD_MAC_1588
768 * - IXGBE_ADVTXD_DCMD_VLE
770 * The following bits must only be set in the last Data
772 * - IXGBE_TXD_CMD_EOP
774 * The following bits can be set in any Data Descriptor, but
775 * are only set in the last Data Descriptor:
778 cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
779 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
781 #ifdef RTE_LIBRTE_IEEE1588
782 if (ol_flags & PKT_TX_IEEE1588_TMST)
783 cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
789 if (ol_flags & PKT_TX_TCP_SEG) {
790 /* when TSO is on, paylen in descriptor is the
791 * not the packet len but the tcp payload len */
792 pkt_len -= (tx_offload.l2_len +
793 tx_offload.l3_len + tx_offload.l4_len);
797 * Setup the TX Advanced Context Descriptor if required
800 volatile struct ixgbe_adv_tx_context_desc *
803 ctx_txd = (volatile struct
804 ixgbe_adv_tx_context_desc *)
807 txn = &sw_ring[txe->next_id];
808 rte_prefetch0(&txn->mbuf->pool);
810 if (txe->mbuf != NULL) {
811 rte_pktmbuf_free_seg(txe->mbuf);
815 ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
818 txe->last_id = tx_last;
819 tx_id = txe->next_id;
824 * Setup the TX Advanced Data Descriptor,
825 * This path will go through
826 * whatever new/reuse the context descriptor
828 cmd_type_len |= tx_desc_ol_flags_to_cmdtype(ol_flags);
829 olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
830 olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
833 olinfo_status |= (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
838 txn = &sw_ring[txe->next_id];
839 rte_prefetch0(&txn->mbuf->pool);
841 if (txe->mbuf != NULL)
842 rte_pktmbuf_free_seg(txe->mbuf);
846 * Set up Transmit Data Descriptor.
848 slen = m_seg->data_len;
849 buf_dma_addr = rte_mbuf_data_dma_addr(m_seg);
850 txd->read.buffer_addr =
851 rte_cpu_to_le_64(buf_dma_addr);
852 txd->read.cmd_type_len =
853 rte_cpu_to_le_32(cmd_type_len | slen);
854 txd->read.olinfo_status =
855 rte_cpu_to_le_32(olinfo_status);
856 txe->last_id = tx_last;
857 tx_id = txe->next_id;
860 } while (m_seg != NULL);
863 * The last packet data descriptor needs End Of Packet (EOP)
865 cmd_type_len |= IXGBE_TXD_CMD_EOP;
866 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
867 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
869 /* Set RS bit only on threshold packets' last descriptor */
870 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
871 PMD_TX_FREE_LOG(DEBUG,
872 "Setting RS bit on TXD id="
873 "%4u (port=%d queue=%d)",
874 tx_last, txq->port_id, txq->queue_id);
876 cmd_type_len |= IXGBE_TXD_CMD_RS;
878 /* Update txq RS bit counters */
884 txd->read.cmd_type_len |= rte_cpu_to_le_32(cmd_type_len);
888 /* set RS on last packet in the burst */
890 txp->read.cmd_type_len |= rte_cpu_to_le_32(IXGBE_TXD_CMD_RS);
895 * Set the Transmit Descriptor Tail (TDT)
897 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
898 (unsigned) txq->port_id, (unsigned) txq->queue_id,
899 (unsigned) tx_id, (unsigned) nb_tx);
900 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, tx_id);
901 txq->tx_tail = tx_id;
906 /*********************************************************************
910 **********************************************************************/
912 #define IXGBE_PACKET_TYPE_ETHER 0X00
913 #define IXGBE_PACKET_TYPE_IPV4 0X01
914 #define IXGBE_PACKET_TYPE_IPV4_TCP 0X11
915 #define IXGBE_PACKET_TYPE_IPV4_UDP 0X21
916 #define IXGBE_PACKET_TYPE_IPV4_SCTP 0X41
917 #define IXGBE_PACKET_TYPE_IPV4_EXT 0X03
918 #define IXGBE_PACKET_TYPE_IPV4_EXT_TCP 0X13
919 #define IXGBE_PACKET_TYPE_IPV4_EXT_UDP 0X23
920 #define IXGBE_PACKET_TYPE_IPV4_EXT_SCTP 0X43
921 #define IXGBE_PACKET_TYPE_IPV6 0X04
922 #define IXGBE_PACKET_TYPE_IPV6_TCP 0X14
923 #define IXGBE_PACKET_TYPE_IPV6_UDP 0X24
924 #define IXGBE_PACKET_TYPE_IPV6_SCTP 0X44
925 #define IXGBE_PACKET_TYPE_IPV6_EXT 0X0C
926 #define IXGBE_PACKET_TYPE_IPV6_EXT_TCP 0X1C
927 #define IXGBE_PACKET_TYPE_IPV6_EXT_UDP 0X2C
928 #define IXGBE_PACKET_TYPE_IPV6_EXT_SCTP 0X4C
929 #define IXGBE_PACKET_TYPE_IPV4_IPV6 0X05
930 #define IXGBE_PACKET_TYPE_IPV4_IPV6_TCP 0X15
931 #define IXGBE_PACKET_TYPE_IPV4_IPV6_UDP 0X25
932 #define IXGBE_PACKET_TYPE_IPV4_IPV6_SCTP 0X45
933 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6 0X07
934 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_TCP 0X17
935 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_UDP 0X27
936 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_SCTP 0X47
937 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT 0X0D
938 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP 0X1D
939 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP 0X2D
940 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_SCTP 0X4D
941 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT 0X0F
942 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_TCP 0X1F
943 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_UDP 0X2F
944 #define IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_SCTP 0X4F
946 #define IXGBE_PACKET_TYPE_NVGRE 0X00
947 #define IXGBE_PACKET_TYPE_NVGRE_IPV4 0X01
948 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_TCP 0X11
949 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_UDP 0X21
950 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_SCTP 0X41
951 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT 0X03
952 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_TCP 0X13
953 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_UDP 0X23
954 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_SCTP 0X43
955 #define IXGBE_PACKET_TYPE_NVGRE_IPV6 0X04
956 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_TCP 0X14
957 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_UDP 0X24
958 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_SCTP 0X44
959 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT 0X0C
960 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_TCP 0X1C
961 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_UDP 0X2C
962 #define IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_SCTP 0X4C
963 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6 0X05
964 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_TCP 0X15
965 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_UDP 0X25
966 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT 0X0D
967 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_TCP 0X1D
968 #define IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_UDP 0X2D
970 #define IXGBE_PACKET_TYPE_VXLAN 0X80
971 #define IXGBE_PACKET_TYPE_VXLAN_IPV4 0X81
972 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_TCP 0x91
973 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_UDP 0xA1
974 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_SCTP 0xC1
975 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT 0x83
976 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_TCP 0X93
977 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_UDP 0XA3
978 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_SCTP 0XC3
979 #define IXGBE_PACKET_TYPE_VXLAN_IPV6 0X84
980 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_TCP 0X94
981 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_UDP 0XA4
982 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_SCTP 0XC4
983 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT 0X8C
984 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_TCP 0X9C
985 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_UDP 0XAC
986 #define IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_SCTP 0XCC
987 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6 0X85
988 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_TCP 0X95
989 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_UDP 0XA5
990 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT 0X8D
991 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_TCP 0X9D
992 #define IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_UDP 0XAD
994 #define IXGBE_PACKET_TYPE_MAX 0X80
995 #define IXGBE_PACKET_TYPE_TN_MAX 0X100
996 #define IXGBE_PACKET_TYPE_SHIFT 0X04
998 /* @note: fix ixgbe_dev_supported_ptypes_get() if any change here. */
999 static inline uint32_t
1000 ixgbe_rxd_pkt_info_to_pkt_type(uint32_t pkt_info, uint16_t ptype_mask)
1003 * Use 2 different table for normal packet and tunnel packet
1004 * to save the space.
1006 static const uint32_t
1007 ptype_table[IXGBE_PACKET_TYPE_MAX] __rte_cache_aligned = {
1008 [IXGBE_PACKET_TYPE_ETHER] = RTE_PTYPE_L2_ETHER,
1009 [IXGBE_PACKET_TYPE_IPV4] = RTE_PTYPE_L2_ETHER |
1011 [IXGBE_PACKET_TYPE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1012 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
1013 [IXGBE_PACKET_TYPE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1014 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
1015 [IXGBE_PACKET_TYPE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
1016 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP,
1017 [IXGBE_PACKET_TYPE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1018 RTE_PTYPE_L3_IPV4_EXT,
1019 [IXGBE_PACKET_TYPE_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1020 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP,
1021 [IXGBE_PACKET_TYPE_IPV4_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1022 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP,
1023 [IXGBE_PACKET_TYPE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1024 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_SCTP,
1025 [IXGBE_PACKET_TYPE_IPV6] = RTE_PTYPE_L2_ETHER |
1027 [IXGBE_PACKET_TYPE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1028 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
1029 [IXGBE_PACKET_TYPE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1030 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
1031 [IXGBE_PACKET_TYPE_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1032 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP,
1033 [IXGBE_PACKET_TYPE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1034 RTE_PTYPE_L3_IPV6_EXT,
1035 [IXGBE_PACKET_TYPE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1036 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP,
1037 [IXGBE_PACKET_TYPE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1038 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP,
1039 [IXGBE_PACKET_TYPE_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1040 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_SCTP,
1041 [IXGBE_PACKET_TYPE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1042 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1043 RTE_PTYPE_INNER_L3_IPV6,
1044 [IXGBE_PACKET_TYPE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1045 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1046 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1047 [IXGBE_PACKET_TYPE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1048 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1049 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1050 [IXGBE_PACKET_TYPE_IPV4_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1051 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1052 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1053 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6] = RTE_PTYPE_L2_ETHER |
1054 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1055 RTE_PTYPE_INNER_L3_IPV6,
1056 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1057 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1058 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1059 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1060 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1061 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1062 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1063 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1064 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1065 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1066 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1067 RTE_PTYPE_INNER_L3_IPV6_EXT,
1068 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1069 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1070 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1071 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1072 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1073 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1074 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1075 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
1076 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1077 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1078 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1079 RTE_PTYPE_INNER_L3_IPV6_EXT,
1080 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1081 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1082 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1083 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1084 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1085 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1086 [IXGBE_PACKET_TYPE_IPV4_EXT_IPV6_EXT_SCTP] =
1087 RTE_PTYPE_L2_ETHER |
1088 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_TUNNEL_IP |
1089 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1092 static const uint32_t
1093 ptype_table_tn[IXGBE_PACKET_TYPE_TN_MAX] __rte_cache_aligned = {
1094 [IXGBE_PACKET_TYPE_NVGRE] = RTE_PTYPE_L2_ETHER |
1095 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1096 RTE_PTYPE_INNER_L2_ETHER,
1097 [IXGBE_PACKET_TYPE_NVGRE_IPV4] = RTE_PTYPE_L2_ETHER |
1098 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1099 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1100 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1101 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1102 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT,
1103 [IXGBE_PACKET_TYPE_NVGRE_IPV6] = RTE_PTYPE_L2_ETHER |
1104 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1105 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6,
1106 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1107 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1108 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1109 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1110 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1111 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT,
1112 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1113 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1114 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1115 [IXGBE_PACKET_TYPE_NVGRE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1116 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1117 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1118 RTE_PTYPE_INNER_L4_TCP,
1119 [IXGBE_PACKET_TYPE_NVGRE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1120 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1121 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1122 RTE_PTYPE_INNER_L4_TCP,
1123 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1124 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1125 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1126 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1127 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1128 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1129 RTE_PTYPE_INNER_L4_TCP,
1130 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_TCP] =
1131 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1132 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_INNER_L2_ETHER |
1133 RTE_PTYPE_INNER_L3_IPV4,
1134 [IXGBE_PACKET_TYPE_NVGRE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1135 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1136 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4 |
1137 RTE_PTYPE_INNER_L4_UDP,
1138 [IXGBE_PACKET_TYPE_NVGRE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1139 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1140 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1141 RTE_PTYPE_INNER_L4_UDP,
1142 [IXGBE_PACKET_TYPE_NVGRE_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1143 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1144 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6 |
1145 RTE_PTYPE_INNER_L4_SCTP,
1146 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1147 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1148 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1149 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1150 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1151 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1152 RTE_PTYPE_INNER_L4_UDP,
1153 [IXGBE_PACKET_TYPE_NVGRE_IPV6_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1154 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1155 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV6_EXT |
1156 RTE_PTYPE_INNER_L4_SCTP,
1157 [IXGBE_PACKET_TYPE_NVGRE_IPV4_IPV6_EXT_UDP] =
1158 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1159 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_INNER_L2_ETHER |
1160 RTE_PTYPE_INNER_L3_IPV4,
1161 [IXGBE_PACKET_TYPE_NVGRE_IPV4_SCTP] = 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 RTE_PTYPE_INNER_L4_SCTP,
1165 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
1166 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1167 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1168 RTE_PTYPE_INNER_L4_SCTP,
1169 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1170 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_TUNNEL_GRE |
1171 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4_EXT |
1172 RTE_PTYPE_INNER_L4_TCP,
1173 [IXGBE_PACKET_TYPE_NVGRE_IPV4_EXT_UDP] = 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_IPV4_EXT |
1176 RTE_PTYPE_INNER_L4_UDP,
1178 [IXGBE_PACKET_TYPE_VXLAN] = RTE_PTYPE_L2_ETHER |
1179 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1180 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER,
1181 [IXGBE_PACKET_TYPE_VXLAN_IPV4] = RTE_PTYPE_L2_ETHER |
1182 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1183 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1184 RTE_PTYPE_INNER_L3_IPV4,
1185 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
1186 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1187 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1188 RTE_PTYPE_INNER_L3_IPV4_EXT,
1189 [IXGBE_PACKET_TYPE_VXLAN_IPV6] = RTE_PTYPE_L2_ETHER |
1190 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1191 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1192 RTE_PTYPE_INNER_L3_IPV6,
1193 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
1194 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1195 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1196 RTE_PTYPE_INNER_L3_IPV4,
1197 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1198 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1199 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1200 RTE_PTYPE_INNER_L3_IPV6_EXT,
1201 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
1202 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1203 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1204 RTE_PTYPE_INNER_L3_IPV4,
1205 [IXGBE_PACKET_TYPE_VXLAN_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
1206 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1207 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1208 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_TCP,
1209 [IXGBE_PACKET_TYPE_VXLAN_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1210 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1211 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1212 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
1213 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
1214 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1215 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1216 RTE_PTYPE_INNER_L3_IPV4,
1217 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
1218 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1219 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1220 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
1221 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_TCP] =
1222 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1223 RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN |
1224 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1225 [IXGBE_PACKET_TYPE_VXLAN_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
1226 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1227 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1228 RTE_PTYPE_INNER_L3_IPV4 | RTE_PTYPE_INNER_L4_UDP,
1229 [IXGBE_PACKET_TYPE_VXLAN_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1230 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1231 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1232 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
1233 [IXGBE_PACKET_TYPE_VXLAN_IPV6_SCTP] = RTE_PTYPE_L2_ETHER |
1234 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1235 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1236 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_SCTP,
1237 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
1238 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1239 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1240 RTE_PTYPE_INNER_L3_IPV4,
1241 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
1242 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP |
1243 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_INNER_L2_ETHER |
1244 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
1245 [IXGBE_PACKET_TYPE_VXLAN_IPV6_EXT_SCTP] = 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_IPV6_EXT | RTE_PTYPE_INNER_L4_SCTP,
1249 [IXGBE_PACKET_TYPE_VXLAN_IPV4_IPV6_EXT_UDP] =
1250 RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
1251 RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN |
1252 RTE_PTYPE_INNER_L2_ETHER | RTE_PTYPE_INNER_L3_IPV4,
1253 [IXGBE_PACKET_TYPE_VXLAN_IPV4_SCTP] = 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_IPV4 | RTE_PTYPE_INNER_L4_SCTP,
1257 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_SCTP] = 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_EXT | RTE_PTYPE_INNER_L4_SCTP,
1261 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_TCP] = 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_IPV4_EXT | RTE_PTYPE_INNER_L4_TCP,
1265 [IXGBE_PACKET_TYPE_VXLAN_IPV4_EXT_UDP] = 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_EXT | RTE_PTYPE_INNER_L4_UDP,
1271 if (unlikely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
1272 return RTE_PTYPE_UNKNOWN;
1274 pkt_info = (pkt_info >> IXGBE_PACKET_TYPE_SHIFT) & ptype_mask;
1276 /* For tunnel packet */
1277 if (pkt_info & IXGBE_PACKET_TYPE_TUNNEL_BIT) {
1278 /* Remove the tunnel bit to save the space. */
1279 pkt_info &= IXGBE_PACKET_TYPE_MASK_TUNNEL;
1280 return ptype_table_tn[pkt_info];
1284 * For x550, if it's not tunnel,
1285 * tunnel type bit should be set to 0.
1286 * Reuse 82599's mask.
1288 pkt_info &= IXGBE_PACKET_TYPE_MASK_82599;
1290 return ptype_table[pkt_info];
1293 static inline uint64_t
1294 ixgbe_rxd_pkt_info_to_pkt_flags(uint16_t pkt_info)
1296 static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
1297 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
1298 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
1299 PKT_RX_RSS_HASH, 0, 0, 0,
1300 0, 0, 0, PKT_RX_FDIR,
1302 #ifdef RTE_LIBRTE_IEEE1588
1303 static uint64_t ip_pkt_etqf_map[8] = {
1304 0, 0, 0, PKT_RX_IEEE1588_PTP,
1308 if (likely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
1309 return ip_pkt_etqf_map[(pkt_info >> 4) & 0X07] |
1310 ip_rss_types_map[pkt_info & 0XF];
1312 return ip_rss_types_map[pkt_info & 0XF];
1314 return ip_rss_types_map[pkt_info & 0XF];
1318 static inline uint64_t
1319 rx_desc_status_to_pkt_flags(uint32_t rx_status, uint64_t vlan_flags)
1324 * Check if VLAN present only.
1325 * Do not check whether L3/L4 rx checksum done by NIC or not,
1326 * That can be found from rte_eth_rxmode.hw_ip_checksum flag
1328 pkt_flags = (rx_status & IXGBE_RXD_STAT_VP) ? vlan_flags : 0;
1330 #ifdef RTE_LIBRTE_IEEE1588
1331 if (rx_status & IXGBE_RXD_STAT_TMST)
1332 pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
1337 static inline uint64_t
1338 rx_desc_error_to_pkt_flags(uint32_t rx_status)
1343 * Bit 31: IPE, IPv4 checksum error
1344 * Bit 30: L4I, L4I integrity error
1346 static uint64_t error_to_pkt_flags_map[4] = {
1347 PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD,
1348 PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_BAD,
1349 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_GOOD,
1350 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1352 pkt_flags = error_to_pkt_flags_map[(rx_status >>
1353 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1355 if ((rx_status & IXGBE_RXD_STAT_OUTERIPCS) &&
1356 (rx_status & IXGBE_RXDADV_ERR_OUTERIPER)) {
1357 pkt_flags |= PKT_RX_EIP_CKSUM_BAD;
1364 * LOOK_AHEAD defines how many desc statuses to check beyond the
1365 * current descriptor.
1366 * It must be a pound define for optimal performance.
1367 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1368 * function only works with LOOK_AHEAD=8.
1370 #define LOOK_AHEAD 8
1371 #if (LOOK_AHEAD != 8)
1372 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1375 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1377 volatile union ixgbe_adv_rx_desc *rxdp;
1378 struct ixgbe_rx_entry *rxep;
1379 struct rte_mbuf *mb;
1383 uint32_t s[LOOK_AHEAD];
1384 uint32_t pkt_info[LOOK_AHEAD];
1385 int i, j, nb_rx = 0;
1387 uint64_t vlan_flags = rxq->vlan_flags;
1389 /* get references to current descriptor and S/W ring entry */
1390 rxdp = &rxq->rx_ring[rxq->rx_tail];
1391 rxep = &rxq->sw_ring[rxq->rx_tail];
1393 status = rxdp->wb.upper.status_error;
1394 /* check to make sure there is at least 1 packet to receive */
1395 if (!(status & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1399 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1400 * reference packets that are ready to be received.
1402 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1403 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD) {
1404 /* Read desc statuses backwards to avoid race condition */
1405 for (j = LOOK_AHEAD-1; j >= 0; --j)
1406 s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);
1408 for (j = LOOK_AHEAD - 1; j >= 0; --j)
1409 pkt_info[j] = rte_le_to_cpu_32(rxdp[j].wb.lower.
1412 /* Compute how many status bits were set */
1414 for (j = 0; j < LOOK_AHEAD; ++j)
1415 nb_dd += s[j] & IXGBE_RXDADV_STAT_DD;
1419 /* Translate descriptor info to mbuf format */
1420 for (j = 0; j < nb_dd; ++j) {
1422 pkt_len = rte_le_to_cpu_16(rxdp[j].wb.upper.length) -
1424 mb->data_len = pkt_len;
1425 mb->pkt_len = pkt_len;
1426 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1428 /* convert descriptor fields to rte mbuf flags */
1429 pkt_flags = rx_desc_status_to_pkt_flags(s[j],
1431 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1432 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags
1433 ((uint16_t)pkt_info[j]);
1434 mb->ol_flags = pkt_flags;
1436 ixgbe_rxd_pkt_info_to_pkt_type
1437 (pkt_info[j], rxq->pkt_type_mask);
1439 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1440 mb->hash.rss = rte_le_to_cpu_32(
1441 rxdp[j].wb.lower.hi_dword.rss);
1442 else if (pkt_flags & PKT_RX_FDIR) {
1443 mb->hash.fdir.hash = rte_le_to_cpu_16(
1444 rxdp[j].wb.lower.hi_dword.csum_ip.csum) &
1445 IXGBE_ATR_HASH_MASK;
1446 mb->hash.fdir.id = rte_le_to_cpu_16(
1447 rxdp[j].wb.lower.hi_dword.csum_ip.ip_id);
1451 /* Move mbuf pointers from the S/W ring to the stage */
1452 for (j = 0; j < LOOK_AHEAD; ++j) {
1453 rxq->rx_stage[i + j] = rxep[j].mbuf;
1456 /* stop if all requested packets could not be received */
1457 if (nb_dd != LOOK_AHEAD)
1461 /* clear software ring entries so we can cleanup correctly */
1462 for (i = 0; i < nb_rx; ++i) {
1463 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1471 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1473 volatile union ixgbe_adv_rx_desc *rxdp;
1474 struct ixgbe_rx_entry *rxep;
1475 struct rte_mbuf *mb;
1480 /* allocate buffers in bulk directly into the S/W ring */
1481 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1482 rxep = &rxq->sw_ring[alloc_idx];
1483 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1484 rxq->rx_free_thresh);
1485 if (unlikely(diag != 0))
1488 rxdp = &rxq->rx_ring[alloc_idx];
1489 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1490 /* populate the static rte mbuf fields */
1495 mb->port = rxq->port_id;
1498 rte_mbuf_refcnt_set(mb, 1);
1499 mb->data_off = RTE_PKTMBUF_HEADROOM;
1501 /* populate the descriptors */
1502 dma_addr = rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(mb));
1503 rxdp[i].read.hdr_addr = 0;
1504 rxdp[i].read.pkt_addr = dma_addr;
1507 /* update state of internal queue structure */
1508 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1509 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1510 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1516 static inline uint16_t
1517 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1520 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1523 /* how many packets are ready to return? */
1524 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1526 /* copy mbuf pointers to the application's packet list */
1527 for (i = 0; i < nb_pkts; ++i)
1528 rx_pkts[i] = stage[i];
1530 /* update internal queue state */
1531 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1532 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1537 static inline uint16_t
1538 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1541 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1544 /* Any previously recv'd pkts will be returned from the Rx stage */
1545 if (rxq->rx_nb_avail)
1546 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1548 /* Scan the H/W ring for packets to receive */
1549 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1551 /* update internal queue state */
1552 rxq->rx_next_avail = 0;
1553 rxq->rx_nb_avail = nb_rx;
1554 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1556 /* if required, allocate new buffers to replenish descriptors */
1557 if (rxq->rx_tail > rxq->rx_free_trigger) {
1558 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1560 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1563 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1564 "queue_id=%u", (unsigned) rxq->port_id,
1565 (unsigned) rxq->queue_id);
1567 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1568 rxq->rx_free_thresh;
1571 * Need to rewind any previous receives if we cannot
1572 * allocate new buffers to replenish the old ones.
1574 rxq->rx_nb_avail = 0;
1575 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1576 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1577 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1582 /* update tail pointer */
1584 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, cur_free_trigger);
1587 if (rxq->rx_tail >= rxq->nb_rx_desc)
1590 /* received any packets this loop? */
1591 if (rxq->rx_nb_avail)
1592 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1597 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1599 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1604 if (unlikely(nb_pkts == 0))
1607 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1608 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1610 /* request is relatively large, chunk it up */
1615 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1616 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1617 nb_rx = (uint16_t)(nb_rx + ret);
1618 nb_pkts = (uint16_t)(nb_pkts - ret);
1627 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1630 struct ixgbe_rx_queue *rxq;
1631 volatile union ixgbe_adv_rx_desc *rx_ring;
1632 volatile union ixgbe_adv_rx_desc *rxdp;
1633 struct ixgbe_rx_entry *sw_ring;
1634 struct ixgbe_rx_entry *rxe;
1635 struct rte_mbuf *rxm;
1636 struct rte_mbuf *nmb;
1637 union ixgbe_adv_rx_desc rxd;
1646 uint64_t vlan_flags;
1651 rx_id = rxq->rx_tail;
1652 rx_ring = rxq->rx_ring;
1653 sw_ring = rxq->sw_ring;
1654 vlan_flags = rxq->vlan_flags;
1655 while (nb_rx < nb_pkts) {
1657 * The order of operations here is important as the DD status
1658 * bit must not be read after any other descriptor fields.
1659 * rx_ring and rxdp are pointing to volatile data so the order
1660 * of accesses cannot be reordered by the compiler. If they were
1661 * not volatile, they could be reordered which could lead to
1662 * using invalid descriptor fields when read from rxd.
1664 rxdp = &rx_ring[rx_id];
1665 staterr = rxdp->wb.upper.status_error;
1666 if (!(staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1673 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1674 * is likely to be invalid and to be dropped by the various
1675 * validation checks performed by the network stack.
1677 * Allocate a new mbuf to replenish the RX ring descriptor.
1678 * If the allocation fails:
1679 * - arrange for that RX descriptor to be the first one
1680 * being parsed the next time the receive function is
1681 * invoked [on the same queue].
1683 * - Stop parsing the RX ring and return immediately.
1685 * This policy do not drop the packet received in the RX
1686 * descriptor for which the allocation of a new mbuf failed.
1687 * Thus, it allows that packet to be later retrieved if
1688 * mbuf have been freed in the mean time.
1689 * As a side effect, holding RX descriptors instead of
1690 * systematically giving them back to the NIC may lead to
1691 * RX ring exhaustion situations.
1692 * However, the NIC can gracefully prevent such situations
1693 * to happen by sending specific "back-pressure" flow control
1694 * frames to its peer(s).
1696 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1697 "ext_err_stat=0x%08x pkt_len=%u",
1698 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1699 (unsigned) rx_id, (unsigned) staterr,
1700 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1702 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
1704 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1705 "queue_id=%u", (unsigned) rxq->port_id,
1706 (unsigned) rxq->queue_id);
1707 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1712 rxe = &sw_ring[rx_id];
1714 if (rx_id == rxq->nb_rx_desc)
1717 /* Prefetch next mbuf while processing current one. */
1718 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1721 * When next RX descriptor is on a cache-line boundary,
1722 * prefetch the next 4 RX descriptors and the next 8 pointers
1725 if ((rx_id & 0x3) == 0) {
1726 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1727 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1733 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(nmb));
1734 rxdp->read.hdr_addr = 0;
1735 rxdp->read.pkt_addr = dma_addr;
1738 * Initialize the returned mbuf.
1739 * 1) setup generic mbuf fields:
1740 * - number of segments,
1743 * - RX port identifier.
1744 * 2) integrate hardware offload data, if any:
1745 * - RSS flag & hash,
1746 * - IP checksum flag,
1747 * - VLAN TCI, if any,
1750 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1752 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1753 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1756 rxm->pkt_len = pkt_len;
1757 rxm->data_len = pkt_len;
1758 rxm->port = rxq->port_id;
1760 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1761 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1762 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1764 pkt_flags = rx_desc_status_to_pkt_flags(staterr, vlan_flags);
1765 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1766 pkt_flags = pkt_flags |
1767 ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1768 rxm->ol_flags = pkt_flags;
1770 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info,
1771 rxq->pkt_type_mask);
1773 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1774 rxm->hash.rss = rte_le_to_cpu_32(
1775 rxd.wb.lower.hi_dword.rss);
1776 else if (pkt_flags & PKT_RX_FDIR) {
1777 rxm->hash.fdir.hash = rte_le_to_cpu_16(
1778 rxd.wb.lower.hi_dword.csum_ip.csum) &
1779 IXGBE_ATR_HASH_MASK;
1780 rxm->hash.fdir.id = rte_le_to_cpu_16(
1781 rxd.wb.lower.hi_dword.csum_ip.ip_id);
1784 * Store the mbuf address into the next entry of the array
1785 * of returned packets.
1787 rx_pkts[nb_rx++] = rxm;
1789 rxq->rx_tail = rx_id;
1792 * If the number of free RX descriptors is greater than the RX free
1793 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1795 * Update the RDT with the value of the last processed RX descriptor
1796 * minus 1, to guarantee that the RDT register is never equal to the
1797 * RDH register, which creates a "full" ring situtation from the
1798 * hardware point of view...
1800 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1801 if (nb_hold > rxq->rx_free_thresh) {
1802 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1803 "nb_hold=%u nb_rx=%u",
1804 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1805 (unsigned) rx_id, (unsigned) nb_hold,
1807 rx_id = (uint16_t) ((rx_id == 0) ?
1808 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1809 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1812 rxq->nb_rx_hold = nb_hold;
1817 * Detect an RSC descriptor.
1819 static inline uint32_t
1820 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1822 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1823 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1827 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1829 * Fill the following info in the HEAD buffer of the Rx cluster:
1830 * - RX port identifier
1831 * - hardware offload data, if any:
1833 * - IP checksum flag
1834 * - VLAN TCI, if any
1836 * @head HEAD of the packet cluster
1837 * @desc HW descriptor to get data from
1838 * @rxq Pointer to the Rx queue
1841 ixgbe_fill_cluster_head_buf(
1842 struct rte_mbuf *head,
1843 union ixgbe_adv_rx_desc *desc,
1844 struct ixgbe_rx_queue *rxq,
1850 head->port = rxq->port_id;
1852 /* The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1853 * set in the pkt_flags field.
1855 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1856 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1857 pkt_flags = rx_desc_status_to_pkt_flags(staterr, rxq->vlan_flags);
1858 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1859 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1860 head->ol_flags = pkt_flags;
1862 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info, rxq->pkt_type_mask);
1864 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1865 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1866 else if (pkt_flags & PKT_RX_FDIR) {
1867 head->hash.fdir.hash =
1868 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1869 & IXGBE_ATR_HASH_MASK;
1870 head->hash.fdir.id =
1871 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1876 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1878 * @rx_queue Rx queue handle
1879 * @rx_pkts table of received packets
1880 * @nb_pkts size of rx_pkts table
1881 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1883 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1884 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1886 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1887 * 1) When non-EOP RSC completion arrives:
1888 * a) Update the HEAD of the current RSC aggregation cluster with the new
1889 * segment's data length.
1890 * b) Set the "next" pointer of the current segment to point to the segment
1891 * at the NEXTP index.
1892 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
1893 * in the sw_rsc_ring.
1894 * 2) When EOP arrives we just update the cluster's total length and offload
1895 * flags and deliver the cluster up to the upper layers. In our case - put it
1896 * in the rx_pkts table.
1898 * Returns the number of received packets/clusters (according to the "bulk
1899 * receive" interface).
1901 static inline uint16_t
1902 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
1905 struct ixgbe_rx_queue *rxq = rx_queue;
1906 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
1907 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
1908 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
1909 uint16_t rx_id = rxq->rx_tail;
1911 uint16_t nb_hold = rxq->nb_rx_hold;
1912 uint16_t prev_id = rxq->rx_tail;
1914 while (nb_rx < nb_pkts) {
1916 struct ixgbe_rx_entry *rxe;
1917 struct ixgbe_scattered_rx_entry *sc_entry;
1918 struct ixgbe_scattered_rx_entry *next_sc_entry;
1919 struct ixgbe_rx_entry *next_rxe = NULL;
1920 struct rte_mbuf *first_seg;
1921 struct rte_mbuf *rxm;
1922 struct rte_mbuf *nmb;
1923 union ixgbe_adv_rx_desc rxd;
1926 volatile union ixgbe_adv_rx_desc *rxdp;
1931 * The code in this whole file uses the volatile pointer to
1932 * ensure the read ordering of the status and the rest of the
1933 * descriptor fields (on the compiler level only!!!). This is so
1934 * UGLY - why not to just use the compiler barrier instead? DPDK
1935 * even has the rte_compiler_barrier() for that.
1937 * But most importantly this is just wrong because this doesn't
1938 * ensure memory ordering in a general case at all. For
1939 * instance, DPDK is supposed to work on Power CPUs where
1940 * compiler barrier may just not be enough!
1942 * I tried to write only this function properly to have a
1943 * starting point (as a part of an LRO/RSC series) but the
1944 * compiler cursed at me when I tried to cast away the
1945 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
1946 * keeping it the way it is for now.
1948 * The code in this file is broken in so many other places and
1949 * will just not work on a big endian CPU anyway therefore the
1950 * lines below will have to be revisited together with the rest
1954 * - Get rid of "volatile" crap and let the compiler do its
1956 * - Use the proper memory barrier (rte_rmb()) to ensure the
1957 * memory ordering below.
1959 rxdp = &rx_ring[rx_id];
1960 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
1962 if (!(staterr & IXGBE_RXDADV_STAT_DD))
1967 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1968 "staterr=0x%x data_len=%u",
1969 rxq->port_id, rxq->queue_id, rx_id, staterr,
1970 rte_le_to_cpu_16(rxd.wb.upper.length));
1973 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
1975 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
1976 "port_id=%u queue_id=%u",
1977 rxq->port_id, rxq->queue_id);
1979 rte_eth_devices[rxq->port_id].data->
1980 rx_mbuf_alloc_failed++;
1983 } else if (nb_hold > rxq->rx_free_thresh) {
1984 uint16_t next_rdt = rxq->rx_free_trigger;
1986 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
1988 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr,
1990 nb_hold -= rxq->rx_free_thresh;
1992 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
1993 "port_id=%u queue_id=%u",
1994 rxq->port_id, rxq->queue_id);
1996 rte_eth_devices[rxq->port_id].data->
1997 rx_mbuf_alloc_failed++;
2003 rxe = &sw_ring[rx_id];
2004 eop = staterr & IXGBE_RXDADV_STAT_EOP;
2006 next_id = rx_id + 1;
2007 if (next_id == rxq->nb_rx_desc)
2010 /* Prefetch next mbuf while processing current one. */
2011 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
2014 * When next RX descriptor is on a cache-line boundary,
2015 * prefetch the next 4 RX descriptors and the next 4 pointers
2018 if ((next_id & 0x3) == 0) {
2019 rte_ixgbe_prefetch(&rx_ring[next_id]);
2020 rte_ixgbe_prefetch(&sw_ring[next_id]);
2027 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(nmb));
2029 * Update RX descriptor with the physical address of the
2030 * new data buffer of the new allocated mbuf.
2034 rxm->data_off = RTE_PKTMBUF_HEADROOM;
2035 rxdp->read.hdr_addr = 0;
2036 rxdp->read.pkt_addr = dma;
2041 * Set data length & data buffer address of mbuf.
2043 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
2044 rxm->data_len = data_len;
2049 * Get next descriptor index:
2050 * - For RSC it's in the NEXTP field.
2051 * - For a scattered packet - it's just a following
2054 if (ixgbe_rsc_count(&rxd))
2056 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
2057 IXGBE_RXDADV_NEXTP_SHIFT;
2061 next_sc_entry = &sw_sc_ring[nextp_id];
2062 next_rxe = &sw_ring[nextp_id];
2063 rte_ixgbe_prefetch(next_rxe);
2066 sc_entry = &sw_sc_ring[rx_id];
2067 first_seg = sc_entry->fbuf;
2068 sc_entry->fbuf = NULL;
2071 * If this is the first buffer of the received packet,
2072 * set the pointer to the first mbuf of the packet and
2073 * initialize its context.
2074 * Otherwise, update the total length and the number of segments
2075 * of the current scattered packet, and update the pointer to
2076 * the last mbuf of the current packet.
2078 if (first_seg == NULL) {
2080 first_seg->pkt_len = data_len;
2081 first_seg->nb_segs = 1;
2083 first_seg->pkt_len += data_len;
2084 first_seg->nb_segs++;
2091 * If this is not the last buffer of the received packet, update
2092 * the pointer to the first mbuf at the NEXTP entry in the
2093 * sw_sc_ring and continue to parse the RX ring.
2095 if (!eop && next_rxe) {
2096 rxm->next = next_rxe->mbuf;
2097 next_sc_entry->fbuf = first_seg;
2102 * This is the last buffer of the received packet - return
2103 * the current cluster to the user.
2107 /* Initialize the first mbuf of the returned packet */
2108 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq, staterr);
2111 * Deal with the case, when HW CRC srip is disabled.
2112 * That can't happen when LRO is enabled, but still could
2113 * happen for scattered RX mode.
2115 first_seg->pkt_len -= rxq->crc_len;
2116 if (unlikely(rxm->data_len <= rxq->crc_len)) {
2117 struct rte_mbuf *lp;
2119 for (lp = first_seg; lp->next != rxm; lp = lp->next)
2122 first_seg->nb_segs--;
2123 lp->data_len -= rxq->crc_len - rxm->data_len;
2125 rte_pktmbuf_free_seg(rxm);
2127 rxm->data_len -= rxq->crc_len;
2129 /* Prefetch data of first segment, if configured to do so. */
2130 rte_packet_prefetch((char *)first_seg->buf_addr +
2131 first_seg->data_off);
2134 * Store the mbuf address into the next entry of the array
2135 * of returned packets.
2137 rx_pkts[nb_rx++] = first_seg;
2141 * Record index of the next RX descriptor to probe.
2143 rxq->rx_tail = rx_id;
2146 * If the number of free RX descriptors is greater than the RX free
2147 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
2149 * Update the RDT with the value of the last processed RX descriptor
2150 * minus 1, to guarantee that the RDT register is never equal to the
2151 * RDH register, which creates a "full" ring situtation from the
2152 * hardware point of view...
2154 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
2155 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
2156 "nb_hold=%u nb_rx=%u",
2157 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
2160 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, prev_id);
2164 rxq->nb_rx_hold = nb_hold;
2169 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2172 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
2176 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2179 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
2182 /*********************************************************************
2184 * Queue management functions
2186 **********************************************************************/
2188 static void __attribute__((cold))
2189 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
2193 if (txq->sw_ring != NULL) {
2194 for (i = 0; i < txq->nb_tx_desc; i++) {
2195 if (txq->sw_ring[i].mbuf != NULL) {
2196 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2197 txq->sw_ring[i].mbuf = NULL;
2203 static void __attribute__((cold))
2204 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
2207 txq->sw_ring != NULL)
2208 rte_free(txq->sw_ring);
2211 static void __attribute__((cold))
2212 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
2214 if (txq != NULL && txq->ops != NULL) {
2215 txq->ops->release_mbufs(txq);
2216 txq->ops->free_swring(txq);
2221 void __attribute__((cold))
2222 ixgbe_dev_tx_queue_release(void *txq)
2224 ixgbe_tx_queue_release(txq);
2227 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
2228 static void __attribute__((cold))
2229 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
2231 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
2232 struct ixgbe_tx_entry *txe = txq->sw_ring;
2235 /* Zero out HW ring memory */
2236 for (i = 0; i < txq->nb_tx_desc; i++) {
2237 txq->tx_ring[i] = zeroed_desc;
2240 /* Initialize SW ring entries */
2241 prev = (uint16_t) (txq->nb_tx_desc - 1);
2242 for (i = 0; i < txq->nb_tx_desc; i++) {
2243 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
2245 txd->wb.status = rte_cpu_to_le_32(IXGBE_TXD_STAT_DD);
2248 txe[prev].next_id = i;
2252 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2253 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2256 txq->nb_tx_used = 0;
2258 * Always allow 1 descriptor to be un-allocated to avoid
2259 * a H/W race condition
2261 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2262 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2264 memset((void *)&txq->ctx_cache, 0,
2265 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2268 static const struct ixgbe_txq_ops def_txq_ops = {
2269 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2270 .free_swring = ixgbe_tx_free_swring,
2271 .reset = ixgbe_reset_tx_queue,
2274 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2275 * the queue parameters. Used in tx_queue_setup by primary process and then
2276 * in dev_init by secondary process when attaching to an existing ethdev.
2278 void __attribute__((cold))
2279 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2281 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2282 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
2283 && (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2284 PMD_INIT_LOG(DEBUG, "Using simple tx code path");
2285 #ifdef RTE_IXGBE_INC_VECTOR
2286 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2287 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2288 ixgbe_txq_vec_setup(txq) == 0)) {
2289 PMD_INIT_LOG(DEBUG, "Vector tx enabled.");
2290 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2293 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2295 PMD_INIT_LOG(DEBUG, "Using full-featured tx code path");
2297 " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
2298 (unsigned long)txq->txq_flags,
2299 (unsigned long)IXGBE_SIMPLE_FLAGS);
2301 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2302 (unsigned long)txq->tx_rs_thresh,
2303 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2304 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2308 int __attribute__((cold))
2309 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2312 unsigned int socket_id,
2313 const struct rte_eth_txconf *tx_conf)
2315 const struct rte_memzone *tz;
2316 struct ixgbe_tx_queue *txq;
2317 struct ixgbe_hw *hw;
2318 uint16_t tx_rs_thresh, tx_free_thresh;
2320 PMD_INIT_FUNC_TRACE();
2321 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2324 * Validate number of transmit descriptors.
2325 * It must not exceed hardware maximum, and must be multiple
2328 if (nb_desc % IXGBE_TXD_ALIGN != 0 ||
2329 (nb_desc > IXGBE_MAX_RING_DESC) ||
2330 (nb_desc < IXGBE_MIN_RING_DESC)) {
2335 * The following two parameters control the setting of the RS bit on
2336 * transmit descriptors.
2337 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2338 * descriptors have been used.
2339 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2340 * descriptors are used or if the number of descriptors required
2341 * to transmit a packet is greater than the number of free TX
2343 * The following constraints must be satisfied:
2344 * tx_rs_thresh must be greater than 0.
2345 * tx_rs_thresh must be less than the size of the ring minus 2.
2346 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2347 * tx_rs_thresh must be a divisor of the ring size.
2348 * tx_free_thresh must be greater than 0.
2349 * tx_free_thresh must be less than the size of the ring minus 3.
2350 * One descriptor in the TX ring is used as a sentinel to avoid a
2351 * H/W race condition, hence the maximum threshold constraints.
2352 * When set to zero use default values.
2354 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2355 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2356 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2357 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2358 if (tx_rs_thresh >= (nb_desc - 2)) {
2359 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2360 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2361 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2362 (int)dev->data->port_id, (int)queue_idx);
2365 if (tx_rs_thresh > DEFAULT_TX_RS_THRESH) {
2366 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less or equal than %u. "
2367 "(tx_rs_thresh=%u port=%d queue=%d)",
2368 DEFAULT_TX_RS_THRESH, (unsigned int)tx_rs_thresh,
2369 (int)dev->data->port_id, (int)queue_idx);
2372 if (tx_free_thresh >= (nb_desc - 3)) {
2373 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2374 "tx_free_thresh must be less than the number of "
2375 "TX descriptors minus 3. (tx_free_thresh=%u "
2376 "port=%d queue=%d)",
2377 (unsigned int)tx_free_thresh,
2378 (int)dev->data->port_id, (int)queue_idx);
2381 if (tx_rs_thresh > tx_free_thresh) {
2382 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2383 "tx_free_thresh. (tx_free_thresh=%u "
2384 "tx_rs_thresh=%u port=%d queue=%d)",
2385 (unsigned int)tx_free_thresh,
2386 (unsigned int)tx_rs_thresh,
2387 (int)dev->data->port_id,
2391 if ((nb_desc % tx_rs_thresh) != 0) {
2392 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2393 "number of TX descriptors. (tx_rs_thresh=%u "
2394 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2395 (int)dev->data->port_id, (int)queue_idx);
2400 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2401 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2402 * by the NIC and all descriptors are written back after the NIC
2403 * accumulates WTHRESH descriptors.
2405 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2406 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2407 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2408 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2409 (int)dev->data->port_id, (int)queue_idx);
2413 /* Free memory prior to re-allocation if needed... */
2414 if (dev->data->tx_queues[queue_idx] != NULL) {
2415 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2416 dev->data->tx_queues[queue_idx] = NULL;
2419 /* First allocate the tx queue data structure */
2420 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2421 RTE_CACHE_LINE_SIZE, socket_id);
2426 * Allocate TX ring hardware descriptors. A memzone large enough to
2427 * handle the maximum ring size is allocated in order to allow for
2428 * resizing in later calls to the queue setup function.
2430 tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
2431 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2432 IXGBE_ALIGN, socket_id);
2434 ixgbe_tx_queue_release(txq);
2438 txq->nb_tx_desc = nb_desc;
2439 txq->tx_rs_thresh = tx_rs_thresh;
2440 txq->tx_free_thresh = tx_free_thresh;
2441 txq->pthresh = tx_conf->tx_thresh.pthresh;
2442 txq->hthresh = tx_conf->tx_thresh.hthresh;
2443 txq->wthresh = tx_conf->tx_thresh.wthresh;
2444 txq->queue_id = queue_idx;
2445 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2446 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2447 txq->port_id = dev->data->port_id;
2448 txq->txq_flags = tx_conf->txq_flags;
2449 txq->ops = &def_txq_ops;
2450 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2453 * Modification to set VFTDT for virtual function if vf is detected
2455 if (hw->mac.type == ixgbe_mac_82599_vf ||
2456 hw->mac.type == ixgbe_mac_X540_vf ||
2457 hw->mac.type == ixgbe_mac_X550_vf ||
2458 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2459 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2460 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2462 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2464 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2465 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2467 /* Allocate software ring */
2468 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2469 sizeof(struct ixgbe_tx_entry) * nb_desc,
2470 RTE_CACHE_LINE_SIZE, socket_id);
2471 if (txq->sw_ring == NULL) {
2472 ixgbe_tx_queue_release(txq);
2475 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2476 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2478 /* set up vector or scalar TX function as appropriate */
2479 ixgbe_set_tx_function(dev, txq);
2481 txq->ops->reset(txq);
2483 dev->data->tx_queues[queue_idx] = txq;
2490 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2492 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2493 * in the sw_rsc_ring is not set to NULL but rather points to the next
2494 * mbuf of this RSC aggregation (that has not been completed yet and still
2495 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2496 * will just free first "nb_segs" segments of the cluster explicitly by calling
2497 * an rte_pktmbuf_free_seg().
2499 * @m scattered cluster head
2501 static void __attribute__((cold))
2502 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2504 uint8_t i, nb_segs = m->nb_segs;
2505 struct rte_mbuf *next_seg;
2507 for (i = 0; i < nb_segs; i++) {
2509 rte_pktmbuf_free_seg(m);
2514 static void __attribute__((cold))
2515 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2519 #ifdef RTE_IXGBE_INC_VECTOR
2520 /* SSE Vector driver has a different way of releasing mbufs. */
2521 if (rxq->rx_using_sse) {
2522 ixgbe_rx_queue_release_mbufs_vec(rxq);
2527 if (rxq->sw_ring != NULL) {
2528 for (i = 0; i < rxq->nb_rx_desc; i++) {
2529 if (rxq->sw_ring[i].mbuf != NULL) {
2530 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2531 rxq->sw_ring[i].mbuf = NULL;
2534 if (rxq->rx_nb_avail) {
2535 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2536 struct rte_mbuf *mb;
2538 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2539 rte_pktmbuf_free_seg(mb);
2541 rxq->rx_nb_avail = 0;
2545 if (rxq->sw_sc_ring)
2546 for (i = 0; i < rxq->nb_rx_desc; i++)
2547 if (rxq->sw_sc_ring[i].fbuf) {
2548 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2549 rxq->sw_sc_ring[i].fbuf = NULL;
2553 static void __attribute__((cold))
2554 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2557 ixgbe_rx_queue_release_mbufs(rxq);
2558 rte_free(rxq->sw_ring);
2559 rte_free(rxq->sw_sc_ring);
2564 void __attribute__((cold))
2565 ixgbe_dev_rx_queue_release(void *rxq)
2567 ixgbe_rx_queue_release(rxq);
2571 * Check if Rx Burst Bulk Alloc function can be used.
2573 * 0: the preconditions are satisfied and the bulk allocation function
2575 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2576 * function must be used.
2578 static inline int __attribute__((cold))
2579 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2584 * Make sure the following pre-conditions are satisfied:
2585 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2586 * rxq->rx_free_thresh < rxq->nb_rx_desc
2587 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2588 * rxq->nb_rx_desc<(IXGBE_MAX_RING_DESC-RTE_PMD_IXGBE_RX_MAX_BURST)
2589 * Scattered packets are not supported. This should be checked
2590 * outside of this function.
2592 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2593 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2594 "rxq->rx_free_thresh=%d, "
2595 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2596 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2598 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2599 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2600 "rxq->rx_free_thresh=%d, "
2601 "rxq->nb_rx_desc=%d",
2602 rxq->rx_free_thresh, rxq->nb_rx_desc);
2604 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2605 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2606 "rxq->nb_rx_desc=%d, "
2607 "rxq->rx_free_thresh=%d",
2608 rxq->nb_rx_desc, rxq->rx_free_thresh);
2610 } else if (!(rxq->nb_rx_desc <
2611 (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST))) {
2612 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2613 "rxq->nb_rx_desc=%d, "
2614 "IXGBE_MAX_RING_DESC=%d, "
2615 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2616 rxq->nb_rx_desc, IXGBE_MAX_RING_DESC,
2617 RTE_PMD_IXGBE_RX_MAX_BURST);
2624 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2625 static void __attribute__((cold))
2626 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2628 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2630 uint16_t len = rxq->nb_rx_desc;
2633 * By default, the Rx queue setup function allocates enough memory for
2634 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2635 * extra memory at the end of the descriptor ring to be zero'd out. A
2636 * pre-condition for using the Rx burst bulk alloc function is that the
2637 * number of descriptors is less than or equal to
2638 * (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST). Check all the
2639 * constraints here to see if we need to zero out memory after the end
2640 * of the H/W descriptor ring.
2642 if (adapter->rx_bulk_alloc_allowed)
2643 /* zero out extra memory */
2644 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2647 * Zero out HW ring memory. Zero out extra memory at the end of
2648 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2649 * reads extra memory as zeros.
2651 for (i = 0; i < len; i++) {
2652 rxq->rx_ring[i] = zeroed_desc;
2656 * initialize extra software ring entries. Space for these extra
2657 * entries is always allocated
2659 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2660 for (i = rxq->nb_rx_desc; i < len; ++i) {
2661 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2664 rxq->rx_nb_avail = 0;
2665 rxq->rx_next_avail = 0;
2666 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2668 rxq->nb_rx_hold = 0;
2669 rxq->pkt_first_seg = NULL;
2670 rxq->pkt_last_seg = NULL;
2672 #ifdef RTE_IXGBE_INC_VECTOR
2673 rxq->rxrearm_start = 0;
2674 rxq->rxrearm_nb = 0;
2678 int __attribute__((cold))
2679 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2682 unsigned int socket_id,
2683 const struct rte_eth_rxconf *rx_conf,
2684 struct rte_mempool *mp)
2686 const struct rte_memzone *rz;
2687 struct ixgbe_rx_queue *rxq;
2688 struct ixgbe_hw *hw;
2690 struct ixgbe_adapter *adapter =
2691 (struct ixgbe_adapter *)dev->data->dev_private;
2693 PMD_INIT_FUNC_TRACE();
2694 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2697 * Validate number of receive descriptors.
2698 * It must not exceed hardware maximum, and must be multiple
2701 if (nb_desc % IXGBE_RXD_ALIGN != 0 ||
2702 (nb_desc > IXGBE_MAX_RING_DESC) ||
2703 (nb_desc < IXGBE_MIN_RING_DESC)) {
2707 /* Free memory prior to re-allocation if needed... */
2708 if (dev->data->rx_queues[queue_idx] != NULL) {
2709 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2710 dev->data->rx_queues[queue_idx] = NULL;
2713 /* First allocate the rx queue data structure */
2714 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2715 RTE_CACHE_LINE_SIZE, socket_id);
2719 rxq->nb_rx_desc = nb_desc;
2720 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2721 rxq->queue_id = queue_idx;
2722 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2723 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2724 rxq->port_id = dev->data->port_id;
2725 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2727 rxq->drop_en = rx_conf->rx_drop_en;
2728 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2731 * The packet type in RX descriptor is different for different NICs.
2732 * Some bits are used for x550 but reserved for other NICS.
2733 * So set different masks for different NICs.
2735 if (hw->mac.type == ixgbe_mac_X550 ||
2736 hw->mac.type == ixgbe_mac_X550EM_x ||
2737 hw->mac.type == ixgbe_mac_X550EM_a ||
2738 hw->mac.type == ixgbe_mac_X550_vf ||
2739 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2740 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2741 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_X550;
2743 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_82599;
2746 * Allocate RX ring hardware descriptors. A memzone large enough to
2747 * handle the maximum ring size is allocated in order to allow for
2748 * resizing in later calls to the queue setup function.
2750 rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
2751 RX_RING_SZ, IXGBE_ALIGN, socket_id);
2753 ixgbe_rx_queue_release(rxq);
2758 * Zero init all the descriptors in the ring.
2760 memset(rz->addr, 0, RX_RING_SZ);
2763 * Modified to setup VFRDT for Virtual Function
2765 if (hw->mac.type == ixgbe_mac_82599_vf ||
2766 hw->mac.type == ixgbe_mac_X540_vf ||
2767 hw->mac.type == ixgbe_mac_X550_vf ||
2768 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2769 hw->mac.type == ixgbe_mac_X550EM_a_vf) {
2771 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2773 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2776 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2778 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2781 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2782 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2785 * Certain constraints must be met in order to use the bulk buffer
2786 * allocation Rx burst function. If any of Rx queues doesn't meet them
2787 * the feature should be disabled for the whole port.
2789 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
2790 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
2791 "preconditions - canceling the feature for "
2792 "the whole port[%d]",
2793 rxq->queue_id, rxq->port_id);
2794 adapter->rx_bulk_alloc_allowed = false;
2798 * Allocate software ring. Allow for space at the end of the
2799 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2800 * function does not access an invalid memory region.
2803 if (adapter->rx_bulk_alloc_allowed)
2804 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2806 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2807 sizeof(struct ixgbe_rx_entry) * len,
2808 RTE_CACHE_LINE_SIZE, socket_id);
2809 if (!rxq->sw_ring) {
2810 ixgbe_rx_queue_release(rxq);
2815 * Always allocate even if it's not going to be needed in order to
2816 * simplify the code.
2818 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
2819 * be requested in ixgbe_dev_rx_init(), which is called later from
2823 rte_zmalloc_socket("rxq->sw_sc_ring",
2824 sizeof(struct ixgbe_scattered_rx_entry) * len,
2825 RTE_CACHE_LINE_SIZE, socket_id);
2826 if (!rxq->sw_sc_ring) {
2827 ixgbe_rx_queue_release(rxq);
2831 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
2832 "dma_addr=0x%"PRIx64,
2833 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
2834 rxq->rx_ring_phys_addr);
2836 if (!rte_is_power_of_2(nb_desc)) {
2837 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
2838 "preconditions - canceling the feature for "
2839 "the whole port[%d]",
2840 rxq->queue_id, rxq->port_id);
2841 adapter->rx_vec_allowed = false;
2843 ixgbe_rxq_vec_setup(rxq);
2845 dev->data->rx_queues[queue_idx] = rxq;
2847 ixgbe_reset_rx_queue(adapter, rxq);
2853 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2855 #define IXGBE_RXQ_SCAN_INTERVAL 4
2856 volatile union ixgbe_adv_rx_desc *rxdp;
2857 struct ixgbe_rx_queue *rxq;
2860 if (rx_queue_id >= dev->data->nb_rx_queues) {
2861 PMD_RX_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
2865 rxq = dev->data->rx_queues[rx_queue_id];
2866 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2868 while ((desc < rxq->nb_rx_desc) &&
2869 (rxdp->wb.upper.status_error &
2870 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD))) {
2871 desc += IXGBE_RXQ_SCAN_INTERVAL;
2872 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2873 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2874 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2875 desc - rxq->nb_rx_desc]);
2882 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2884 volatile union ixgbe_adv_rx_desc *rxdp;
2885 struct ixgbe_rx_queue *rxq = rx_queue;
2888 if (unlikely(offset >= rxq->nb_rx_desc))
2890 desc = rxq->rx_tail + offset;
2891 if (desc >= rxq->nb_rx_desc)
2892 desc -= rxq->nb_rx_desc;
2894 rxdp = &rxq->rx_ring[desc];
2895 return !!(rxdp->wb.upper.status_error &
2896 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD));
2899 void __attribute__((cold))
2900 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
2903 struct ixgbe_adapter *adapter =
2904 (struct ixgbe_adapter *)dev->data->dev_private;
2906 PMD_INIT_FUNC_TRACE();
2908 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2909 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
2912 txq->ops->release_mbufs(txq);
2913 txq->ops->reset(txq);
2917 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2918 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
2921 ixgbe_rx_queue_release_mbufs(rxq);
2922 ixgbe_reset_rx_queue(adapter, rxq);
2928 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
2932 PMD_INIT_FUNC_TRACE();
2934 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2935 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
2936 dev->data->rx_queues[i] = NULL;
2938 dev->data->nb_rx_queues = 0;
2940 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2941 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
2942 dev->data->tx_queues[i] = NULL;
2944 dev->data->nb_tx_queues = 0;
2947 /*********************************************************************
2949 * Device RX/TX init functions
2951 **********************************************************************/
2954 * Receive Side Scaling (RSS)
2955 * See section 7.1.2.8 in the following document:
2956 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
2959 * The source and destination IP addresses of the IP header and the source
2960 * and destination ports of TCP/UDP headers, if any, of received packets are
2961 * hashed against a configurable random key to compute a 32-bit RSS hash result.
2962 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
2963 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
2964 * RSS output index which is used as the RX queue index where to store the
2966 * The following output is supplied in the RX write-back descriptor:
2967 * - 32-bit result of the Microsoft RSS hash function,
2968 * - 4-bit RSS type field.
2972 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
2973 * Used as the default key.
2975 static uint8_t rss_intel_key[40] = {
2976 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2977 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2978 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2979 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2980 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2984 ixgbe_rss_disable(struct rte_eth_dev *dev)
2986 struct ixgbe_hw *hw;
2990 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2991 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
2992 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
2993 mrqc &= ~IXGBE_MRQC_RSSEN;
2994 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
2998 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
3008 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3009 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3011 hash_key = rss_conf->rss_key;
3012 if (hash_key != NULL) {
3013 /* Fill in RSS hash key */
3014 for (i = 0; i < 10; i++) {
3015 rss_key = hash_key[(i * 4)];
3016 rss_key |= hash_key[(i * 4) + 1] << 8;
3017 rss_key |= hash_key[(i * 4) + 2] << 16;
3018 rss_key |= hash_key[(i * 4) + 3] << 24;
3019 IXGBE_WRITE_REG_ARRAY(hw, rssrk_reg, i, rss_key);
3023 /* Set configured hashing protocols in MRQC register */
3024 rss_hf = rss_conf->rss_hf;
3025 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
3026 if (rss_hf & ETH_RSS_IPV4)
3027 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
3028 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
3029 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
3030 if (rss_hf & ETH_RSS_IPV6)
3031 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
3032 if (rss_hf & ETH_RSS_IPV6_EX)
3033 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
3034 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
3035 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3036 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
3037 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
3038 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
3039 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3040 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
3041 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3042 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
3043 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
3044 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
3048 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
3049 struct rte_eth_rss_conf *rss_conf)
3051 struct ixgbe_hw *hw;
3056 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3058 if (!ixgbe_rss_update_sp(hw->mac.type)) {
3059 PMD_DRV_LOG(ERR, "RSS hash update is not supported on this "
3063 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3066 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
3067 * "RSS enabling cannot be done dynamically while it must be
3068 * preceded by a software reset"
3069 * Before changing anything, first check that the update RSS operation
3070 * does not attempt to disable RSS, if RSS was enabled at
3071 * initialization time, or does not attempt to enable RSS, if RSS was
3072 * disabled at initialization time.
3074 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
3075 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3076 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
3077 if (rss_hf != 0) /* Enable RSS */
3079 return 0; /* Nothing to do */
3082 if (rss_hf == 0) /* Disable RSS */
3084 ixgbe_hw_rss_hash_set(hw, rss_conf);
3089 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
3090 struct rte_eth_rss_conf *rss_conf)
3092 struct ixgbe_hw *hw;
3101 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3102 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3103 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3104 hash_key = rss_conf->rss_key;
3105 if (hash_key != NULL) {
3106 /* Return RSS hash key */
3107 for (i = 0; i < 10; i++) {
3108 rss_key = IXGBE_READ_REG_ARRAY(hw, rssrk_reg, i);
3109 hash_key[(i * 4)] = rss_key & 0x000000FF;
3110 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
3111 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
3112 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
3116 /* Get RSS functions configured in MRQC register */
3117 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3118 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
3119 rss_conf->rss_hf = 0;
3123 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
3124 rss_hf |= ETH_RSS_IPV4;
3125 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
3126 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
3127 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
3128 rss_hf |= ETH_RSS_IPV6;
3129 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
3130 rss_hf |= ETH_RSS_IPV6_EX;
3131 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
3132 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
3133 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
3134 rss_hf |= ETH_RSS_IPV6_TCP_EX;
3135 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
3136 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
3137 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
3138 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
3139 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
3140 rss_hf |= ETH_RSS_IPV6_UDP_EX;
3141 rss_conf->rss_hf = rss_hf;
3146 ixgbe_rss_configure(struct rte_eth_dev *dev)
3148 struct rte_eth_rss_conf rss_conf;
3149 struct ixgbe_hw *hw;
3153 uint16_t sp_reta_size;
3156 PMD_INIT_FUNC_TRACE();
3157 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3159 sp_reta_size = ixgbe_reta_size_get(hw->mac.type);
3162 * Fill in redirection table
3163 * The byte-swap is needed because NIC registers are in
3164 * little-endian order.
3167 for (i = 0, j = 0; i < sp_reta_size; i++, j++) {
3168 reta_reg = ixgbe_reta_reg_get(hw->mac.type, i);
3170 if (j == dev->data->nb_rx_queues)
3172 reta = (reta << 8) | j;
3174 IXGBE_WRITE_REG(hw, reta_reg,
3179 * Configure the RSS key and the RSS protocols used to compute
3180 * the RSS hash of input packets.
3182 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
3183 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
3184 ixgbe_rss_disable(dev);
3187 if (rss_conf.rss_key == NULL)
3188 rss_conf.rss_key = rss_intel_key; /* Default hash key */
3189 ixgbe_hw_rss_hash_set(hw, &rss_conf);
3192 #define NUM_VFTA_REGISTERS 128
3193 #define NIC_RX_BUFFER_SIZE 0x200
3194 #define X550_RX_BUFFER_SIZE 0x180
3197 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
3199 struct rte_eth_vmdq_dcb_conf *cfg;
3200 struct ixgbe_hw *hw;
3201 enum rte_eth_nb_pools num_pools;
3202 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
3204 uint8_t nb_tcs; /* number of traffic classes */
3207 PMD_INIT_FUNC_TRACE();
3208 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3209 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3210 num_pools = cfg->nb_queue_pools;
3211 /* Check we have a valid number of pools */
3212 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
3213 ixgbe_rss_disable(dev);
3216 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
3217 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
3221 * split rx buffer up into sections, each for 1 traffic class
3223 switch (hw->mac.type) {
3224 case ixgbe_mac_X550:
3225 case ixgbe_mac_X550EM_x:
3226 case ixgbe_mac_X550EM_a:
3227 pbsize = (uint16_t)(X550_RX_BUFFER_SIZE / nb_tcs);
3230 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3233 for (i = 0; i < nb_tcs; i++) {
3234 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3236 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3237 /* clear 10 bits. */
3238 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
3239 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3241 /* zero alloc all unused TCs */
3242 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3243 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3245 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3246 /* clear 10 bits. */
3247 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3250 /* MRQC: enable vmdq and dcb */
3251 mrqc = (num_pools == ETH_16_POOLS) ?
3252 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN;
3253 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3255 /* PFVTCTL: turn on virtualisation and set the default pool */
3256 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3257 if (cfg->enable_default_pool) {
3258 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3260 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3263 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3265 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
3267 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
3269 * mapping is done with 3 bits per priority,
3270 * so shift by i*3 each time
3272 queue_mapping |= ((cfg->dcb_tc[i] & 0x07) << (i * 3));
3274 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
3276 /* RTRPCS: DCB related */
3277 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
3279 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3280 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3281 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3282 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3284 /* VFTA - enable all vlan filters */
3285 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3286 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3289 /* VFRE: pool enabling for receive - 16 or 32 */
3290 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0),
3291 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3294 * MPSAR - allow pools to read specific mac addresses
3295 * In this case, all pools should be able to read from mac addr 0
3297 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
3298 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
3300 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3301 for (i = 0; i < cfg->nb_pool_maps; i++) {
3302 /* set vlan id in VF register and set the valid bit */
3303 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3304 (cfg->pool_map[i].vlan_id & 0xFFF)));
3306 * Put the allowed pools in VFB reg. As we only have 16 or 32
3307 * pools, we only need to use the first half of the register
3310 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
3315 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
3316 * @dev: pointer to eth_dev structure
3317 * @dcb_config: pointer to ixgbe_dcb_config structure
3320 ixgbe_dcb_tx_hw_config(struct rte_eth_dev *dev,
3321 struct ixgbe_dcb_config *dcb_config)
3325 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3327 PMD_INIT_FUNC_TRACE();
3328 if (hw->mac.type != ixgbe_mac_82598EB) {
3329 /* Disable the Tx desc arbiter so that MTQC can be changed */
3330 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3331 reg |= IXGBE_RTTDCS_ARBDIS;
3332 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3334 /* Enable DCB for Tx with 8 TCs */
3335 if (dcb_config->num_tcs.pg_tcs == 8) {
3336 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3338 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3340 if (dcb_config->vt_mode)
3341 reg |= IXGBE_MTQC_VT_ENA;
3342 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3344 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3345 /* Disable drop for all queues in VMDQ mode*/
3346 for (q = 0; q < 128; q++)
3347 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3348 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3350 /* Enable drop for all queues in SRIOV mode */
3351 for (q = 0; q < 128; q++)
3352 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3353 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT) | IXGBE_QDE_ENABLE));
3356 /* Enable the Tx desc arbiter */
3357 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3358 reg &= ~IXGBE_RTTDCS_ARBDIS;
3359 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3361 /* Enable Security TX Buffer IFG for DCB */
3362 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3363 reg |= IXGBE_SECTX_DCB;
3364 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3369 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3370 * @dev: pointer to rte_eth_dev structure
3371 * @dcb_config: pointer to ixgbe_dcb_config structure
3374 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3375 struct ixgbe_dcb_config *dcb_config)
3377 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3378 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3379 struct ixgbe_hw *hw =
3380 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3382 PMD_INIT_FUNC_TRACE();
3383 if (hw->mac.type != ixgbe_mac_82598EB)
3384 /*PF VF Transmit Enable*/
3385 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3386 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3388 /*Configure general DCB TX parameters*/
3389 ixgbe_dcb_tx_hw_config(dev, dcb_config);
3393 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3394 struct ixgbe_dcb_config *dcb_config)
3396 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3397 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3398 struct ixgbe_dcb_tc_config *tc;
3401 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3402 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS) {
3403 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3404 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3406 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3407 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3409 /* User Priority to Traffic Class mapping */
3410 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3411 j = vmdq_rx_conf->dcb_tc[i];
3412 tc = &dcb_config->tc_config[j];
3413 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3419 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3420 struct ixgbe_dcb_config *dcb_config)
3422 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3423 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3424 struct ixgbe_dcb_tc_config *tc;
3427 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3428 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS) {
3429 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3430 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3432 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3433 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3436 /* User Priority to Traffic Class mapping */
3437 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3438 j = vmdq_tx_conf->dcb_tc[i];
3439 tc = &dcb_config->tc_config[j];
3440 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3446 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3447 struct ixgbe_dcb_config *dcb_config)
3449 struct rte_eth_dcb_rx_conf *rx_conf =
3450 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3451 struct ixgbe_dcb_tc_config *tc;
3454 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3455 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3457 /* User Priority to Traffic Class mapping */
3458 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3459 j = rx_conf->dcb_tc[i];
3460 tc = &dcb_config->tc_config[j];
3461 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3467 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3468 struct ixgbe_dcb_config *dcb_config)
3470 struct rte_eth_dcb_tx_conf *tx_conf =
3471 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3472 struct ixgbe_dcb_tc_config *tc;
3475 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3476 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3478 /* User Priority to Traffic Class mapping */
3479 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3480 j = tx_conf->dcb_tc[i];
3481 tc = &dcb_config->tc_config[j];
3482 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3488 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3489 * @hw: pointer to hardware structure
3490 * @dcb_config: pointer to ixgbe_dcb_config structure
3493 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
3494 struct ixgbe_dcb_config *dcb_config)
3500 PMD_INIT_FUNC_TRACE();
3502 * Disable the arbiter before changing parameters
3503 * (always enable recycle mode; WSP)
3505 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3506 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3508 if (hw->mac.type != ixgbe_mac_82598EB) {
3509 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3510 if (dcb_config->num_tcs.pg_tcs == 4) {
3511 if (dcb_config->vt_mode)
3512 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3513 IXGBE_MRQC_VMDQRT4TCEN;
3515 /* no matter the mode is DCB or DCB_RSS, just
3516 * set the MRQE to RSSXTCEN. RSS is controlled
3519 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3520 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3521 IXGBE_MRQC_RTRSS4TCEN;
3524 if (dcb_config->num_tcs.pg_tcs == 8) {
3525 if (dcb_config->vt_mode)
3526 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3527 IXGBE_MRQC_VMDQRT8TCEN;
3529 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3530 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3531 IXGBE_MRQC_RTRSS8TCEN;
3535 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3538 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3539 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3540 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3541 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3543 /* VFTA - enable all vlan filters */
3544 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3545 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3549 * Configure Rx packet plane (recycle mode; WSP) and
3552 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3553 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3557 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3558 uint16_t *max, uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3560 switch (hw->mac.type) {
3561 case ixgbe_mac_82598EB:
3562 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3564 case ixgbe_mac_82599EB:
3565 case ixgbe_mac_X540:
3566 case ixgbe_mac_X550:
3567 case ixgbe_mac_X550EM_x:
3568 case ixgbe_mac_X550EM_a:
3569 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3578 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3579 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3581 switch (hw->mac.type) {
3582 case ixgbe_mac_82598EB:
3583 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id, tsa);
3584 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id, tsa);
3586 case ixgbe_mac_82599EB:
3587 case ixgbe_mac_X540:
3588 case ixgbe_mac_X550:
3589 case ixgbe_mac_X550EM_x:
3590 case ixgbe_mac_X550EM_a:
3591 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id, tsa);
3592 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id, tsa, map);
3599 #define DCB_RX_CONFIG 1
3600 #define DCB_TX_CONFIG 1
3601 #define DCB_TX_PB 1024
3603 * ixgbe_dcb_hw_configure - Enable DCB and configure
3604 * general DCB in VT mode and non-VT mode parameters
3605 * @dev: pointer to rte_eth_dev structure
3606 * @dcb_config: pointer to ixgbe_dcb_config structure
3609 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3610 struct ixgbe_dcb_config *dcb_config)
3613 uint8_t i, pfc_en, nb_tcs;
3614 uint16_t pbsize, rx_buffer_size;
3615 uint8_t config_dcb_rx = 0;
3616 uint8_t config_dcb_tx = 0;
3617 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3618 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3619 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3620 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3621 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3622 struct ixgbe_dcb_tc_config *tc;
3623 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3624 struct ixgbe_hw *hw =
3625 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3627 switch (dev->data->dev_conf.rxmode.mq_mode) {
3628 case ETH_MQ_RX_VMDQ_DCB:
3629 dcb_config->vt_mode = true;
3630 if (hw->mac.type != ixgbe_mac_82598EB) {
3631 config_dcb_rx = DCB_RX_CONFIG;
3633 *get dcb and VT rx configuration parameters
3636 ixgbe_vmdq_dcb_rx_config(dev, dcb_config);
3637 /*Configure general VMDQ and DCB RX parameters*/
3638 ixgbe_vmdq_dcb_configure(dev);
3642 case ETH_MQ_RX_DCB_RSS:
3643 dcb_config->vt_mode = false;
3644 config_dcb_rx = DCB_RX_CONFIG;
3645 /* Get dcb TX configuration parameters from rte_eth_conf */
3646 ixgbe_dcb_rx_config(dev, dcb_config);
3647 /*Configure general DCB RX parameters*/
3648 ixgbe_dcb_rx_hw_config(hw, dcb_config);
3651 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3654 switch (dev->data->dev_conf.txmode.mq_mode) {
3655 case ETH_MQ_TX_VMDQ_DCB:
3656 dcb_config->vt_mode = true;
3657 config_dcb_tx = DCB_TX_CONFIG;
3658 /* get DCB and VT TX configuration parameters
3661 ixgbe_dcb_vt_tx_config(dev, dcb_config);
3662 /*Configure general VMDQ and DCB TX parameters*/
3663 ixgbe_vmdq_dcb_hw_tx_config(dev, dcb_config);
3667 dcb_config->vt_mode = false;
3668 config_dcb_tx = DCB_TX_CONFIG;
3669 /*get DCB TX configuration parameters from rte_eth_conf*/
3670 ixgbe_dcb_tx_config(dev, dcb_config);
3671 /*Configure general DCB TX parameters*/
3672 ixgbe_dcb_tx_hw_config(dev, dcb_config);
3675 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3679 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3681 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3682 if (nb_tcs == ETH_4_TCS) {
3683 /* Avoid un-configured priority mapping to TC0 */
3685 uint8_t mask = 0xFF;
3687 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3688 mask = (uint8_t)(mask & (~(1 << map[i])));
3689 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3690 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3694 /* Re-configure 4 TCs BW */
3695 for (i = 0; i < nb_tcs; i++) {
3696 tc = &dcb_config->tc_config[i];
3697 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
3698 (uint8_t)(100 / nb_tcs);
3699 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
3700 (uint8_t)(100 / nb_tcs);
3702 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
3703 tc = &dcb_config->tc_config[i];
3704 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
3705 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
3709 switch (hw->mac.type) {
3710 case ixgbe_mac_X550:
3711 case ixgbe_mac_X550EM_x:
3712 case ixgbe_mac_X550EM_a:
3713 rx_buffer_size = X550_RX_BUFFER_SIZE;
3716 rx_buffer_size = NIC_RX_BUFFER_SIZE;
3720 if (config_dcb_rx) {
3721 /* Set RX buffer size */
3722 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
3723 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
3725 for (i = 0; i < nb_tcs; i++) {
3726 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3728 /* zero alloc all unused TCs */
3729 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3730 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
3733 if (config_dcb_tx) {
3734 /* Only support an equally distributed
3735 * Tx packet buffer strategy.
3737 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
3738 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
3740 for (i = 0; i < nb_tcs; i++) {
3741 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
3742 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
3744 /* Clear unused TCs, if any, to zero buffer size*/
3745 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3746 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
3747 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
3751 /*Calculates traffic class credits*/
3752 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
3753 IXGBE_DCB_TX_CONFIG);
3754 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
3755 IXGBE_DCB_RX_CONFIG);
3757 if (config_dcb_rx) {
3758 /* Unpack CEE standard containers */
3759 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
3760 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3761 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
3762 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
3763 /* Configure PG(ETS) RX */
3764 ixgbe_dcb_hw_arbite_rx_config(hw, refill, max, bwgid, tsa, map);
3767 if (config_dcb_tx) {
3768 /* Unpack CEE standard containers */
3769 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
3770 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3771 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
3772 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
3773 /* Configure PG(ETS) TX */
3774 ixgbe_dcb_hw_arbite_tx_config(hw, refill, max, bwgid, tsa, map);
3777 /*Configure queue statistics registers*/
3778 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3780 /* Check if the PFC is supported */
3781 if (dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3782 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
3783 for (i = 0; i < nb_tcs; i++) {
3785 * If the TC count is 8,and the default high_water is 48,
3786 * the low_water is 16 as default.
3788 hw->fc.high_water[i] = (pbsize * 3) / 4;
3789 hw->fc.low_water[i] = pbsize / 4;
3790 /* Enable pfc for this TC */
3791 tc = &dcb_config->tc_config[i];
3792 tc->pfc = ixgbe_dcb_pfc_enabled;
3794 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3795 if (dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3797 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3804 * ixgbe_configure_dcb - Configure DCB Hardware
3805 * @dev: pointer to rte_eth_dev
3807 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3809 struct ixgbe_dcb_config *dcb_cfg =
3810 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3811 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3813 PMD_INIT_FUNC_TRACE();
3815 /* check support mq_mode for DCB */
3816 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
3817 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB) &&
3818 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB_RSS))
3821 if (dev->data->nb_rx_queues > ETH_DCB_NUM_QUEUES)
3824 /** Configure DCB hardware **/
3825 ixgbe_dcb_hw_configure(dev, dcb_cfg);
3829 * VMDq only support for 10 GbE NIC.
3832 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
3834 struct rte_eth_vmdq_rx_conf *cfg;
3835 struct ixgbe_hw *hw;
3836 enum rte_eth_nb_pools num_pools;
3837 uint32_t mrqc, vt_ctl, vlanctrl;
3841 PMD_INIT_FUNC_TRACE();
3842 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3843 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
3844 num_pools = cfg->nb_queue_pools;
3846 ixgbe_rss_disable(dev);
3848 /* MRQC: enable vmdq */
3849 mrqc = IXGBE_MRQC_VMDQEN;
3850 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3852 /* PFVTCTL: turn on virtualisation and set the default pool */
3853 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3854 if (cfg->enable_default_pool)
3855 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3857 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3859 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3861 for (i = 0; i < (int)num_pools; i++) {
3862 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
3863 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
3866 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3867 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3868 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3869 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3871 /* VFTA - enable all vlan filters */
3872 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
3873 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
3875 /* VFRE: pool enabling for receive - 64 */
3876 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
3877 if (num_pools == ETH_64_POOLS)
3878 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
3881 * MPSAR - allow pools to read specific mac addresses
3882 * In this case, all pools should be able to read from mac addr 0
3884 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
3885 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
3887 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3888 for (i = 0; i < cfg->nb_pool_maps; i++) {
3889 /* set vlan id in VF register and set the valid bit */
3890 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3891 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
3893 * Put the allowed pools in VFB reg. As we only have 16 or 64
3894 * pools, we only need to use the first half of the register
3897 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
3898 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i * 2),
3899 (cfg->pool_map[i].pools & UINT32_MAX));
3901 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i * 2 + 1)),
3902 ((cfg->pool_map[i].pools >> 32) & UINT32_MAX));
3906 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
3907 if (cfg->enable_loop_back) {
3908 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3909 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
3910 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
3913 IXGBE_WRITE_FLUSH(hw);
3917 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
3918 * @hw: pointer to hardware structure
3921 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
3926 PMD_INIT_FUNC_TRACE();
3927 /*PF VF Transmit Enable*/
3928 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
3929 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
3931 /* Disable the Tx desc arbiter so that MTQC can be changed */
3932 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3933 reg |= IXGBE_RTTDCS_ARBDIS;
3934 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3936 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3937 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3939 /* Disable drop for all queues */
3940 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3941 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3942 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3944 /* Enable the Tx desc arbiter */
3945 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3946 reg &= ~IXGBE_RTTDCS_ARBDIS;
3947 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3949 IXGBE_WRITE_FLUSH(hw);
3952 static int __attribute__((cold))
3953 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
3955 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
3959 /* Initialize software ring entries */
3960 for (i = 0; i < rxq->nb_rx_desc; i++) {
3961 volatile union ixgbe_adv_rx_desc *rxd;
3962 struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
3965 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
3966 (unsigned) rxq->queue_id);
3970 rte_mbuf_refcnt_set(mbuf, 1);
3972 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
3974 mbuf->port = rxq->port_id;
3977 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(mbuf));
3978 rxd = &rxq->rx_ring[i];
3979 rxd->read.hdr_addr = 0;
3980 rxd->read.pkt_addr = dma_addr;
3988 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
3990 struct ixgbe_hw *hw;
3993 ixgbe_rss_configure(dev);
3995 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3997 /* MRQC: enable VF RSS */
3998 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
3999 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
4000 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4002 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
4006 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
4010 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
4014 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4020 ixgbe_config_vf_default(struct rte_eth_dev *dev)
4022 struct ixgbe_hw *hw =
4023 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4025 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4027 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4032 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4033 IXGBE_MRQC_VMDQRT4TCEN);
4037 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4038 IXGBE_MRQC_VMDQRT8TCEN);
4042 "invalid pool number in IOV mode");
4049 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
4051 struct ixgbe_hw *hw =
4052 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4054 if (hw->mac.type == ixgbe_mac_82598EB)
4057 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4059 * SRIOV inactive scheme
4060 * any DCB/RSS w/o VMDq multi-queue setting
4062 switch (dev->data->dev_conf.rxmode.mq_mode) {
4064 case ETH_MQ_RX_DCB_RSS:
4065 case ETH_MQ_RX_VMDQ_RSS:
4066 ixgbe_rss_configure(dev);
4069 case ETH_MQ_RX_VMDQ_DCB:
4070 ixgbe_vmdq_dcb_configure(dev);
4073 case ETH_MQ_RX_VMDQ_ONLY:
4074 ixgbe_vmdq_rx_hw_configure(dev);
4077 case ETH_MQ_RX_NONE:
4079 /* if mq_mode is none, disable rss mode.*/
4080 ixgbe_rss_disable(dev);
4085 * SRIOV active scheme
4086 * Support RSS together with VMDq & SRIOV
4088 switch (dev->data->dev_conf.rxmode.mq_mode) {
4090 case ETH_MQ_RX_VMDQ_RSS:
4091 ixgbe_config_vf_rss(dev);
4093 case ETH_MQ_RX_VMDQ_DCB:
4094 ixgbe_vmdq_dcb_configure(dev);
4096 /* FIXME if support DCB/RSS together with VMDq & SRIOV */
4097 case ETH_MQ_RX_VMDQ_DCB_RSS:
4099 "Could not support DCB/RSS with VMDq & SRIOV");
4102 ixgbe_config_vf_default(dev);
4111 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
4113 struct ixgbe_hw *hw =
4114 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4118 if (hw->mac.type == ixgbe_mac_82598EB)
4121 /* disable arbiter before setting MTQC */
4122 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4123 rttdcs |= IXGBE_RTTDCS_ARBDIS;
4124 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4126 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4128 * SRIOV inactive scheme
4129 * any DCB w/o VMDq multi-queue setting
4131 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
4132 ixgbe_vmdq_tx_hw_configure(hw);
4134 mtqc = IXGBE_MTQC_64Q_1PB;
4135 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4138 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4141 * SRIOV active scheme
4142 * FIXME if support DCB together with VMDq & SRIOV
4145 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
4148 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
4151 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
4155 mtqc = IXGBE_MTQC_64Q_1PB;
4156 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
4158 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4161 /* re-enable arbiter */
4162 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
4163 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4169 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
4171 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
4172 * spec rev. 3.0 chapter 8.2.3.8.13.
4174 * @pool Memory pool of the Rx queue
4176 static inline uint32_t
4177 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
4179 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
4181 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
4184 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
4187 return IXGBE_RSCCTL_MAXDESC_16;
4188 else if (maxdesc >= 8)
4189 return IXGBE_RSCCTL_MAXDESC_8;
4190 else if (maxdesc >= 4)
4191 return IXGBE_RSCCTL_MAXDESC_4;
4193 return IXGBE_RSCCTL_MAXDESC_1;
4197 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
4200 * (Taken from FreeBSD tree)
4201 * (yes this is all very magic and confusing :)
4204 * @entry the register array entry
4205 * @vector the MSIX vector for this queue
4209 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
4211 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4214 vector |= IXGBE_IVAR_ALLOC_VAL;
4216 switch (hw->mac.type) {
4218 case ixgbe_mac_82598EB:
4220 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
4222 entry += (type * 64);
4223 index = (entry >> 2) & 0x1F;
4224 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4225 ivar &= ~(0xFF << (8 * (entry & 0x3)));
4226 ivar |= (vector << (8 * (entry & 0x3)));
4227 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4230 case ixgbe_mac_82599EB:
4231 case ixgbe_mac_X540:
4232 if (type == -1) { /* MISC IVAR */
4233 index = (entry & 1) * 8;
4234 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4235 ivar &= ~(0xFF << index);
4236 ivar |= (vector << index);
4237 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4238 } else { /* RX/TX IVARS */
4239 index = (16 * (entry & 1)) + (8 * type);
4240 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
4241 ivar &= ~(0xFF << index);
4242 ivar |= (vector << index);
4243 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
4253 void __attribute__((cold))
4254 ixgbe_set_rx_function(struct rte_eth_dev *dev)
4256 uint16_t i, rx_using_sse;
4257 struct ixgbe_adapter *adapter =
4258 (struct ixgbe_adapter *)dev->data->dev_private;
4261 * In order to allow Vector Rx there are a few configuration
4262 * conditions to be met and Rx Bulk Allocation should be allowed.
4264 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
4265 !adapter->rx_bulk_alloc_allowed) {
4266 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
4267 "preconditions or RTE_IXGBE_INC_VECTOR is "
4269 dev->data->port_id);
4271 adapter->rx_vec_allowed = false;
4275 * Initialize the appropriate LRO callback.
4277 * If all queues satisfy the bulk allocation preconditions
4278 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
4279 * Otherwise use a single allocation version.
4281 if (dev->data->lro) {
4282 if (adapter->rx_bulk_alloc_allowed) {
4283 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a bulk "
4284 "allocation version");
4285 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4287 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a single "
4288 "allocation version");
4289 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4291 } else if (dev->data->scattered_rx) {
4293 * Set the non-LRO scattered callback: there are Vector and
4294 * single allocation versions.
4296 if (adapter->rx_vec_allowed) {
4297 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
4298 "callback (port=%d).",
4299 dev->data->port_id);
4301 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
4302 } else if (adapter->rx_bulk_alloc_allowed) {
4303 PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
4304 "allocation callback (port=%d).",
4305 dev->data->port_id);
4306 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4308 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
4309 "single allocation) "
4310 "Scattered Rx callback "
4312 dev->data->port_id);
4314 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4317 * Below we set "simple" callbacks according to port/queues parameters.
4318 * If parameters allow we are going to choose between the following
4322 * - Single buffer allocation (the simplest one)
4324 } else if (adapter->rx_vec_allowed) {
4325 PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
4326 "burst size no less than %d (port=%d).",
4327 RTE_IXGBE_DESCS_PER_LOOP,
4328 dev->data->port_id);
4330 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
4331 } else if (adapter->rx_bulk_alloc_allowed) {
4332 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
4333 "satisfied. Rx Burst Bulk Alloc function "
4334 "will be used on port=%d.",
4335 dev->data->port_id);
4337 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
4339 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
4340 "satisfied, or Scattered Rx is requested "
4342 dev->data->port_id);
4344 dev->rx_pkt_burst = ixgbe_recv_pkts;
4347 /* Propagate information about RX function choice through all queues. */
4350 (dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec ||
4351 dev->rx_pkt_burst == ixgbe_recv_pkts_vec);
4353 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4354 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4356 rxq->rx_using_sse = rx_using_sse;
4361 * ixgbe_set_rsc - configure RSC related port HW registers
4363 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4364 * of 82599 Spec (x540 configuration is virtually the same).
4368 * Returns 0 in case of success or a non-zero error code
4371 ixgbe_set_rsc(struct rte_eth_dev *dev)
4373 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4374 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4375 struct rte_eth_dev_info dev_info = { 0 };
4376 bool rsc_capable = false;
4381 dev->dev_ops->dev_infos_get(dev, &dev_info);
4382 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4385 if (!rsc_capable && rx_conf->enable_lro) {
4386 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4391 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4393 if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
4395 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4396 * 3.0 RSC configuration requires HW CRC stripping being
4397 * enabled. If user requested both HW CRC stripping off
4398 * and RSC on - return an error.
4400 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4405 /* RFCTL configuration */
4407 uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4409 if (rx_conf->enable_lro)
4411 * Since NFS packets coalescing is not supported - clear
4412 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4415 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4416 IXGBE_RFCTL_NFSR_DIS);
4418 rfctl |= IXGBE_RFCTL_RSC_DIS;
4420 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4423 /* If LRO hasn't been requested - we are done here. */
4424 if (!rx_conf->enable_lro)
4427 /* Set RDRXCTL.RSCACKC bit */
4428 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4429 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4430 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4432 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4433 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4434 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4436 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4438 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4440 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4442 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4445 * ixgbe PMD doesn't support header-split at the moment.
4447 * Following the 4.6.7.2.1 chapter of the 82599/x540
4448 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4449 * should be configured even if header split is not
4450 * enabled. We will configure it 128 bytes following the
4451 * recommendation in the spec.
4453 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4454 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4455 IXGBE_SRRCTL_BSIZEHDR_MASK;
4458 * TODO: Consider setting the Receive Descriptor Minimum
4459 * Threshold Size for an RSC case. This is not an obviously
4460 * beneficiary option but the one worth considering...
4463 rscctl |= IXGBE_RSCCTL_RSCEN;
4464 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4465 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4468 * RSC: Set ITR interval corresponding to 2K ints/s.
4470 * Full-sized RSC aggregations for a 10Gb/s link will
4471 * arrive at about 20K aggregation/s rate.
4473 * 2K inst/s rate will make only 10% of the
4474 * aggregations to be closed due to the interrupt timer
4475 * expiration for a streaming at wire-speed case.
4477 * For a sparse streaming case this setting will yield
4478 * at most 500us latency for a single RSC aggregation.
4480 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4481 eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
4483 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4484 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4485 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4486 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4489 * RSC requires the mapping of the queue to the
4492 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4497 PMD_INIT_LOG(DEBUG, "enabling LRO mode");
4503 * Initializes Receive Unit.
4505 int __attribute__((cold))
4506 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4508 struct ixgbe_hw *hw;
4509 struct ixgbe_rx_queue *rxq;
4520 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4523 PMD_INIT_FUNC_TRACE();
4524 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4527 * Make sure receives are disabled while setting
4528 * up the RX context (registers, descriptor rings, etc.).
4530 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4531 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4533 /* Enable receipt of broadcasted frames */
4534 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4535 fctrl |= IXGBE_FCTRL_BAM;
4536 fctrl |= IXGBE_FCTRL_DPF;
4537 fctrl |= IXGBE_FCTRL_PMCF;
4538 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4541 * Configure CRC stripping, if any.
4543 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4544 if (rx_conf->hw_strip_crc)
4545 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4547 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4550 * Configure jumbo frame support, if any.
4552 if (rx_conf->jumbo_frame == 1) {
4553 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4554 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4555 maxfrs &= 0x0000FFFF;
4556 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4557 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4559 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4562 * If loopback mode is configured for 82599, set LPBK bit.
4564 if (hw->mac.type == ixgbe_mac_82599EB &&
4565 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4566 hlreg0 |= IXGBE_HLREG0_LPBK;
4568 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4570 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4572 /* Setup RX queues */
4573 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4574 rxq = dev->data->rx_queues[i];
4577 * Reset crc_len in case it was changed after queue setup by a
4578 * call to configure.
4580 rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
4582 /* Setup the Base and Length of the Rx Descriptor Rings */
4583 bus_addr = rxq->rx_ring_phys_addr;
4584 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4585 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4586 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4587 (uint32_t)(bus_addr >> 32));
4588 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4589 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4590 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4591 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4593 /* Configure the SRRCTL register */
4594 #ifdef RTE_HEADER_SPLIT_ENABLE
4596 * Configure Header Split
4598 if (rx_conf->header_split) {
4599 if (hw->mac.type == ixgbe_mac_82599EB) {
4600 /* Must setup the PSRTYPE register */
4603 psrtype = IXGBE_PSRTYPE_TCPHDR |
4604 IXGBE_PSRTYPE_UDPHDR |
4605 IXGBE_PSRTYPE_IPV4HDR |
4606 IXGBE_PSRTYPE_IPV6HDR;
4607 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4609 srrctl = ((rx_conf->split_hdr_size <<
4610 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4611 IXGBE_SRRCTL_BSIZEHDR_MASK);
4612 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4615 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4617 /* Set if packets are dropped when no descriptors available */
4619 srrctl |= IXGBE_SRRCTL_DROP_EN;
4622 * Configure the RX buffer size in the BSIZEPACKET field of
4623 * the SRRCTL register of the queue.
4624 * The value is in 1 KB resolution. Valid values can be from
4627 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4628 RTE_PKTMBUF_HEADROOM);
4629 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4630 IXGBE_SRRCTL_BSIZEPKT_MASK);
4632 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4634 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4635 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4637 /* It adds dual VLAN length for supporting dual VLAN */
4638 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4639 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4640 dev->data->scattered_rx = 1;
4643 if (rx_conf->enable_scatter)
4644 dev->data->scattered_rx = 1;
4647 * Device configured with multiple RX queues.
4649 ixgbe_dev_mq_rx_configure(dev);
4652 * Setup the Checksum Register.
4653 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4654 * Enable IP/L4 checkum computation by hardware if requested to do so.
4656 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4657 rxcsum |= IXGBE_RXCSUM_PCSD;
4658 if (rx_conf->hw_ip_checksum)
4659 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4661 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4663 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4665 if (hw->mac.type == ixgbe_mac_82599EB ||
4666 hw->mac.type == ixgbe_mac_X540) {
4667 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4668 if (rx_conf->hw_strip_crc)
4669 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4671 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4672 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4673 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4676 rc = ixgbe_set_rsc(dev);
4680 ixgbe_set_rx_function(dev);
4686 * Initializes Transmit Unit.
4688 void __attribute__((cold))
4689 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4691 struct ixgbe_hw *hw;
4692 struct ixgbe_tx_queue *txq;
4698 PMD_INIT_FUNC_TRACE();
4699 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4701 /* Enable TX CRC (checksum offload requirement) and hw padding
4704 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4705 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
4706 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4708 /* Setup the Base and Length of the Tx Descriptor Rings */
4709 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4710 txq = dev->data->tx_queues[i];
4712 bus_addr = txq->tx_ring_phys_addr;
4713 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
4714 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4715 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
4716 (uint32_t)(bus_addr >> 32));
4717 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
4718 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4719 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4720 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4721 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4724 * Disable Tx Head Writeback RO bit, since this hoses
4725 * bookkeeping if things aren't delivered in order.
4727 switch (hw->mac.type) {
4728 case ixgbe_mac_82598EB:
4729 txctrl = IXGBE_READ_REG(hw,
4730 IXGBE_DCA_TXCTRL(txq->reg_idx));
4731 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4732 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
4736 case ixgbe_mac_82599EB:
4737 case ixgbe_mac_X540:
4738 case ixgbe_mac_X550:
4739 case ixgbe_mac_X550EM_x:
4740 case ixgbe_mac_X550EM_a:
4742 txctrl = IXGBE_READ_REG(hw,
4743 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
4744 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4745 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
4751 /* Device configured with multiple TX queues. */
4752 ixgbe_dev_mq_tx_configure(dev);
4756 * Set up link for 82599 loopback mode Tx->Rx.
4758 static inline void __attribute__((cold))
4759 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
4761 PMD_INIT_FUNC_TRACE();
4763 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
4764 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
4766 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
4775 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
4776 ixgbe_reset_pipeline_82599(hw);
4778 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
4784 * Start Transmit and Receive Units.
4786 int __attribute__((cold))
4787 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
4789 struct ixgbe_hw *hw;
4790 struct ixgbe_tx_queue *txq;
4791 struct ixgbe_rx_queue *rxq;
4798 PMD_INIT_FUNC_TRACE();
4799 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4801 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4802 txq = dev->data->tx_queues[i];
4803 /* Setup Transmit Threshold Registers */
4804 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4805 txdctl |= txq->pthresh & 0x7F;
4806 txdctl |= ((txq->hthresh & 0x7F) << 8);
4807 txdctl |= ((txq->wthresh & 0x7F) << 16);
4808 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4811 if (hw->mac.type != ixgbe_mac_82598EB) {
4812 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
4813 dmatxctl |= IXGBE_DMATXCTL_TE;
4814 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
4817 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4818 txq = dev->data->tx_queues[i];
4819 if (!txq->tx_deferred_start) {
4820 ret = ixgbe_dev_tx_queue_start(dev, i);
4826 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4827 rxq = dev->data->rx_queues[i];
4828 if (!rxq->rx_deferred_start) {
4829 ret = ixgbe_dev_rx_queue_start(dev, i);
4835 /* Enable Receive engine */
4836 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4837 if (hw->mac.type == ixgbe_mac_82598EB)
4838 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4839 rxctrl |= IXGBE_RXCTRL_RXEN;
4840 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4842 /* If loopback mode is enabled for 82599, set up the link accordingly */
4843 if (hw->mac.type == ixgbe_mac_82599EB &&
4844 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4845 ixgbe_setup_loopback_link_82599(hw);
4851 * Start Receive Units for specified queue.
4853 int __attribute__((cold))
4854 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4856 struct ixgbe_hw *hw;
4857 struct ixgbe_rx_queue *rxq;
4861 PMD_INIT_FUNC_TRACE();
4862 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4864 if (rx_queue_id < dev->data->nb_rx_queues) {
4865 rxq = dev->data->rx_queues[rx_queue_id];
4867 /* Allocate buffers for descriptor rings */
4868 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
4869 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
4873 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4874 rxdctl |= IXGBE_RXDCTL_ENABLE;
4875 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4877 /* Wait until RX Enable ready */
4878 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4881 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4882 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4884 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
4887 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4888 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
4889 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
4897 * Stop Receive Units for specified queue.
4899 int __attribute__((cold))
4900 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4902 struct ixgbe_hw *hw;
4903 struct ixgbe_adapter *adapter =
4904 (struct ixgbe_adapter *)dev->data->dev_private;
4905 struct ixgbe_rx_queue *rxq;
4909 PMD_INIT_FUNC_TRACE();
4910 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4912 if (rx_queue_id < dev->data->nb_rx_queues) {
4913 rxq = dev->data->rx_queues[rx_queue_id];
4915 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4916 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4917 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4919 /* Wait until RX Enable bit clear */
4920 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4923 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4924 } while (--poll_ms && (rxdctl & IXGBE_RXDCTL_ENABLE));
4926 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
4929 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4931 ixgbe_rx_queue_release_mbufs(rxq);
4932 ixgbe_reset_rx_queue(adapter, rxq);
4933 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
4942 * Start Transmit Units for specified queue.
4944 int __attribute__((cold))
4945 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4947 struct ixgbe_hw *hw;
4948 struct ixgbe_tx_queue *txq;
4952 PMD_INIT_FUNC_TRACE();
4953 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4955 if (tx_queue_id < dev->data->nb_tx_queues) {
4956 txq = dev->data->tx_queues[tx_queue_id];
4957 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4958 txdctl |= IXGBE_TXDCTL_ENABLE;
4959 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4961 /* Wait until TX Enable ready */
4962 if (hw->mac.type == ixgbe_mac_82599EB) {
4963 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4966 txdctl = IXGBE_READ_REG(hw,
4967 IXGBE_TXDCTL(txq->reg_idx));
4968 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4970 PMD_INIT_LOG(ERR, "Could not enable "
4971 "Tx Queue %d", tx_queue_id);
4974 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4975 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4976 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
4984 * Stop Transmit Units for specified queue.
4986 int __attribute__((cold))
4987 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4989 struct ixgbe_hw *hw;
4990 struct ixgbe_tx_queue *txq;
4992 uint32_t txtdh, txtdt;
4995 PMD_INIT_FUNC_TRACE();
4996 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4998 if (tx_queue_id >= dev->data->nb_tx_queues)
5001 txq = dev->data->tx_queues[tx_queue_id];
5003 /* Wait until TX queue is empty */
5004 if (hw->mac.type == ixgbe_mac_82599EB) {
5005 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5007 rte_delay_us(RTE_IXGBE_WAIT_100_US);
5008 txtdh = IXGBE_READ_REG(hw,
5009 IXGBE_TDH(txq->reg_idx));
5010 txtdt = IXGBE_READ_REG(hw,
5011 IXGBE_TDT(txq->reg_idx));
5012 } while (--poll_ms && (txtdh != txtdt));
5014 PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
5015 "when stopping.", tx_queue_id);
5018 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5019 txdctl &= ~IXGBE_TXDCTL_ENABLE;
5020 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5022 /* Wait until TX Enable bit clear */
5023 if (hw->mac.type == ixgbe_mac_82599EB) {
5024 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5027 txdctl = IXGBE_READ_REG(hw,
5028 IXGBE_TXDCTL(txq->reg_idx));
5029 } while (--poll_ms && (txdctl & IXGBE_TXDCTL_ENABLE));
5031 PMD_INIT_LOG(ERR, "Could not disable "
5032 "Tx Queue %d", tx_queue_id);
5035 if (txq->ops != NULL) {
5036 txq->ops->release_mbufs(txq);
5037 txq->ops->reset(txq);
5039 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
5045 ixgbe_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5046 struct rte_eth_rxq_info *qinfo)
5048 struct ixgbe_rx_queue *rxq;
5050 rxq = dev->data->rx_queues[queue_id];
5052 qinfo->mp = rxq->mb_pool;
5053 qinfo->scattered_rx = dev->data->scattered_rx;
5054 qinfo->nb_desc = rxq->nb_rx_desc;
5056 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
5057 qinfo->conf.rx_drop_en = rxq->drop_en;
5058 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
5062 ixgbe_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5063 struct rte_eth_txq_info *qinfo)
5065 struct ixgbe_tx_queue *txq;
5067 txq = dev->data->tx_queues[queue_id];
5069 qinfo->nb_desc = txq->nb_tx_desc;
5071 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
5072 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
5073 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
5075 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
5076 qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
5077 qinfo->conf.txq_flags = txq->txq_flags;
5078 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
5082 * [VF] Initializes Receive Unit.
5084 int __attribute__((cold))
5085 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
5087 struct ixgbe_hw *hw;
5088 struct ixgbe_rx_queue *rxq;
5090 uint32_t srrctl, psrtype = 0;
5095 PMD_INIT_FUNC_TRACE();
5096 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5098 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
5099 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5100 "it should be power of 2");
5104 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
5105 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5106 "it should be equal to or less than %d",
5107 hw->mac.max_rx_queues);
5112 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
5113 * disables the VF receipt of packets if the PF MTU is > 1500.
5114 * This is done to deal with 82599 limitations that imposes
5115 * the PF and all VFs to share the same MTU.
5116 * Then, the PF driver enables again the VF receipt of packet when
5117 * the VF driver issues a IXGBE_VF_SET_LPE request.
5118 * In the meantime, the VF device cannot be used, even if the VF driver
5119 * and the Guest VM network stack are ready to accept packets with a
5120 * size up to the PF MTU.
5121 * As a work-around to this PF behaviour, force the call to
5122 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
5123 * VF packets received can work in all cases.
5125 ixgbevf_rlpml_set_vf(hw,
5126 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
5128 /* Setup RX queues */
5129 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5130 rxq = dev->data->rx_queues[i];
5132 /* Allocate buffers for descriptor rings */
5133 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
5137 /* Setup the Base and Length of the Rx Descriptor Rings */
5138 bus_addr = rxq->rx_ring_phys_addr;
5140 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
5141 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5142 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
5143 (uint32_t)(bus_addr >> 32));
5144 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
5145 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
5146 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
5147 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
5150 /* Configure the SRRCTL register */
5151 #ifdef RTE_HEADER_SPLIT_ENABLE
5153 * Configure Header Split
5155 if (dev->data->dev_conf.rxmode.header_split) {
5156 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
5157 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
5158 IXGBE_SRRCTL_BSIZEHDR_MASK);
5159 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
5162 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
5164 /* Set if packets are dropped when no descriptors available */
5166 srrctl |= IXGBE_SRRCTL_DROP_EN;
5169 * Configure the RX buffer size in the BSIZEPACKET field of
5170 * the SRRCTL register of the queue.
5171 * The value is in 1 KB resolution. Valid values can be from
5174 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
5175 RTE_PKTMBUF_HEADROOM);
5176 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
5177 IXGBE_SRRCTL_BSIZEPKT_MASK);
5180 * VF modification to write virtual function SRRCTL register
5182 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
5184 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
5185 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
5187 if (dev->data->dev_conf.rxmode.enable_scatter ||
5188 /* It adds dual VLAN length for supporting dual VLAN */
5189 (dev->data->dev_conf.rxmode.max_rx_pkt_len +
5190 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
5191 if (!dev->data->scattered_rx)
5192 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
5193 dev->data->scattered_rx = 1;
5197 #ifdef RTE_HEADER_SPLIT_ENABLE
5198 if (dev->data->dev_conf.rxmode.header_split)
5199 /* Must setup the PSRTYPE register */
5200 psrtype = IXGBE_PSRTYPE_TCPHDR |
5201 IXGBE_PSRTYPE_UDPHDR |
5202 IXGBE_PSRTYPE_IPV4HDR |
5203 IXGBE_PSRTYPE_IPV6HDR;
5206 /* Set RQPL for VF RSS according to max Rx queue */
5207 psrtype |= (dev->data->nb_rx_queues >> 1) <<
5208 IXGBE_PSRTYPE_RQPL_SHIFT;
5209 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
5211 ixgbe_set_rx_function(dev);
5217 * [VF] Initializes Transmit Unit.
5219 void __attribute__((cold))
5220 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
5222 struct ixgbe_hw *hw;
5223 struct ixgbe_tx_queue *txq;
5228 PMD_INIT_FUNC_TRACE();
5229 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5231 /* Setup the Base and Length of the Tx Descriptor Rings */
5232 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5233 txq = dev->data->tx_queues[i];
5234 bus_addr = txq->tx_ring_phys_addr;
5235 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
5236 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5237 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
5238 (uint32_t)(bus_addr >> 32));
5239 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
5240 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
5241 /* Setup the HW Tx Head and TX Tail descriptor pointers */
5242 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
5243 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
5246 * Disable Tx Head Writeback RO bit, since this hoses
5247 * bookkeeping if things aren't delivered in order.
5249 txctrl = IXGBE_READ_REG(hw,
5250 IXGBE_VFDCA_TXCTRL(i));
5251 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
5252 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
5258 * [VF] Start Transmit and Receive Units.
5260 void __attribute__((cold))
5261 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
5263 struct ixgbe_hw *hw;
5264 struct ixgbe_tx_queue *txq;
5265 struct ixgbe_rx_queue *rxq;
5271 PMD_INIT_FUNC_TRACE();
5272 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5274 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5275 txq = dev->data->tx_queues[i];
5276 /* Setup Transmit Threshold Registers */
5277 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5278 txdctl |= txq->pthresh & 0x7F;
5279 txdctl |= ((txq->hthresh & 0x7F) << 8);
5280 txdctl |= ((txq->wthresh & 0x7F) << 16);
5281 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5284 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5286 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5287 txdctl |= IXGBE_TXDCTL_ENABLE;
5288 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5291 /* Wait until TX Enable ready */
5294 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5295 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
5297 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
5299 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5301 rxq = dev->data->rx_queues[i];
5303 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5304 rxdctl |= IXGBE_RXDCTL_ENABLE;
5305 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
5307 /* Wait until RX Enable ready */
5311 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5312 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
5314 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
5316 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
5321 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
5322 int __attribute__((weak))
5323 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
5328 uint16_t __attribute__((weak))
5329 ixgbe_recv_pkts_vec(
5330 void __rte_unused *rx_queue,
5331 struct rte_mbuf __rte_unused **rx_pkts,
5332 uint16_t __rte_unused nb_pkts)
5337 uint16_t __attribute__((weak))
5338 ixgbe_recv_scattered_pkts_vec(
5339 void __rte_unused *rx_queue,
5340 struct rte_mbuf __rte_unused **rx_pkts,
5341 uint16_t __rte_unused nb_pkts)
5346 int __attribute__((weak))
5347 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)