4 * Copyright(c) 2010-2015 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 0, PKT_RX_L4_CKSUM_BAD, PKT_RX_IP_CKSUM_BAD,
1348 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1350 pkt_flags = error_to_pkt_flags_map[(rx_status >>
1351 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1353 if ((rx_status & IXGBE_RXD_STAT_OUTERIPCS) &&
1354 (rx_status & IXGBE_RXDADV_ERR_OUTERIPER)) {
1355 pkt_flags |= PKT_RX_EIP_CKSUM_BAD;
1362 * LOOK_AHEAD defines how many desc statuses to check beyond the
1363 * current descriptor.
1364 * It must be a pound define for optimal performance.
1365 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1366 * function only works with LOOK_AHEAD=8.
1368 #define LOOK_AHEAD 8
1369 #if (LOOK_AHEAD != 8)
1370 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1373 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1375 volatile union ixgbe_adv_rx_desc *rxdp;
1376 struct ixgbe_rx_entry *rxep;
1377 struct rte_mbuf *mb;
1381 uint32_t s[LOOK_AHEAD];
1382 uint32_t pkt_info[LOOK_AHEAD];
1383 int i, j, nb_rx = 0;
1385 uint64_t vlan_flags = rxq->vlan_flags;
1387 /* get references to current descriptor and S/W ring entry */
1388 rxdp = &rxq->rx_ring[rxq->rx_tail];
1389 rxep = &rxq->sw_ring[rxq->rx_tail];
1391 status = rxdp->wb.upper.status_error;
1392 /* check to make sure there is at least 1 packet to receive */
1393 if (!(status & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1397 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1398 * reference packets that are ready to be received.
1400 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1401 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD) {
1402 /* Read desc statuses backwards to avoid race condition */
1403 for (j = LOOK_AHEAD-1; j >= 0; --j)
1404 s[j] = rte_le_to_cpu_32(rxdp[j].wb.upper.status_error);
1406 for (j = LOOK_AHEAD - 1; j >= 0; --j)
1407 pkt_info[j] = rte_le_to_cpu_32(rxdp[j].wb.lower.
1410 /* Compute how many status bits were set */
1412 for (j = 0; j < LOOK_AHEAD; ++j)
1413 nb_dd += s[j] & IXGBE_RXDADV_STAT_DD;
1417 /* Translate descriptor info to mbuf format */
1418 for (j = 0; j < nb_dd; ++j) {
1420 pkt_len = rte_le_to_cpu_16(rxdp[j].wb.upper.length) -
1422 mb->data_len = pkt_len;
1423 mb->pkt_len = pkt_len;
1424 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1426 /* convert descriptor fields to rte mbuf flags */
1427 pkt_flags = rx_desc_status_to_pkt_flags(s[j],
1429 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1430 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags
1431 ((uint16_t)pkt_info[j]);
1432 mb->ol_flags = pkt_flags;
1434 ixgbe_rxd_pkt_info_to_pkt_type
1435 (pkt_info[j], rxq->pkt_type_mask);
1437 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1438 mb->hash.rss = rte_le_to_cpu_32(
1439 rxdp[j].wb.lower.hi_dword.rss);
1440 else if (pkt_flags & PKT_RX_FDIR) {
1441 mb->hash.fdir.hash = rte_le_to_cpu_16(
1442 rxdp[j].wb.lower.hi_dword.csum_ip.csum) &
1443 IXGBE_ATR_HASH_MASK;
1444 mb->hash.fdir.id = rte_le_to_cpu_16(
1445 rxdp[j].wb.lower.hi_dword.csum_ip.ip_id);
1449 /* Move mbuf pointers from the S/W ring to the stage */
1450 for (j = 0; j < LOOK_AHEAD; ++j) {
1451 rxq->rx_stage[i + j] = rxep[j].mbuf;
1454 /* stop if all requested packets could not be received */
1455 if (nb_dd != LOOK_AHEAD)
1459 /* clear software ring entries so we can cleanup correctly */
1460 for (i = 0; i < nb_rx; ++i) {
1461 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1469 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1471 volatile union ixgbe_adv_rx_desc *rxdp;
1472 struct ixgbe_rx_entry *rxep;
1473 struct rte_mbuf *mb;
1478 /* allocate buffers in bulk directly into the S/W ring */
1479 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1480 rxep = &rxq->sw_ring[alloc_idx];
1481 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1482 rxq->rx_free_thresh);
1483 if (unlikely(diag != 0))
1486 rxdp = &rxq->rx_ring[alloc_idx];
1487 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1488 /* populate the static rte mbuf fields */
1493 mb->port = rxq->port_id;
1496 rte_mbuf_refcnt_set(mb, 1);
1497 mb->data_off = RTE_PKTMBUF_HEADROOM;
1499 /* populate the descriptors */
1500 dma_addr = rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(mb));
1501 rxdp[i].read.hdr_addr = 0;
1502 rxdp[i].read.pkt_addr = dma_addr;
1505 /* update state of internal queue structure */
1506 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1507 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1508 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1514 static inline uint16_t
1515 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1518 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1521 /* how many packets are ready to return? */
1522 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1524 /* copy mbuf pointers to the application's packet list */
1525 for (i = 0; i < nb_pkts; ++i)
1526 rx_pkts[i] = stage[i];
1528 /* update internal queue state */
1529 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1530 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1535 static inline uint16_t
1536 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1539 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1542 /* Any previously recv'd pkts will be returned from the Rx stage */
1543 if (rxq->rx_nb_avail)
1544 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1546 /* Scan the H/W ring for packets to receive */
1547 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1549 /* update internal queue state */
1550 rxq->rx_next_avail = 0;
1551 rxq->rx_nb_avail = nb_rx;
1552 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1554 /* if required, allocate new buffers to replenish descriptors */
1555 if (rxq->rx_tail > rxq->rx_free_trigger) {
1556 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1558 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1561 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1562 "queue_id=%u", (unsigned) rxq->port_id,
1563 (unsigned) rxq->queue_id);
1565 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1566 rxq->rx_free_thresh;
1569 * Need to rewind any previous receives if we cannot
1570 * allocate new buffers to replenish the old ones.
1572 rxq->rx_nb_avail = 0;
1573 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1574 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1575 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1580 /* update tail pointer */
1582 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, cur_free_trigger);
1585 if (rxq->rx_tail >= rxq->nb_rx_desc)
1588 /* received any packets this loop? */
1589 if (rxq->rx_nb_avail)
1590 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1595 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1597 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1602 if (unlikely(nb_pkts == 0))
1605 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1606 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1608 /* request is relatively large, chunk it up */
1613 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1614 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1615 nb_rx = (uint16_t)(nb_rx + ret);
1616 nb_pkts = (uint16_t)(nb_pkts - ret);
1625 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1628 struct ixgbe_rx_queue *rxq;
1629 volatile union ixgbe_adv_rx_desc *rx_ring;
1630 volatile union ixgbe_adv_rx_desc *rxdp;
1631 struct ixgbe_rx_entry *sw_ring;
1632 struct ixgbe_rx_entry *rxe;
1633 struct rte_mbuf *rxm;
1634 struct rte_mbuf *nmb;
1635 union ixgbe_adv_rx_desc rxd;
1644 uint64_t vlan_flags;
1649 rx_id = rxq->rx_tail;
1650 rx_ring = rxq->rx_ring;
1651 sw_ring = rxq->sw_ring;
1652 vlan_flags = rxq->vlan_flags;
1653 while (nb_rx < nb_pkts) {
1655 * The order of operations here is important as the DD status
1656 * bit must not be read after any other descriptor fields.
1657 * rx_ring and rxdp are pointing to volatile data so the order
1658 * of accesses cannot be reordered by the compiler. If they were
1659 * not volatile, they could be reordered which could lead to
1660 * using invalid descriptor fields when read from rxd.
1662 rxdp = &rx_ring[rx_id];
1663 staterr = rxdp->wb.upper.status_error;
1664 if (!(staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1671 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1672 * is likely to be invalid and to be dropped by the various
1673 * validation checks performed by the network stack.
1675 * Allocate a new mbuf to replenish the RX ring descriptor.
1676 * If the allocation fails:
1677 * - arrange for that RX descriptor to be the first one
1678 * being parsed the next time the receive function is
1679 * invoked [on the same queue].
1681 * - Stop parsing the RX ring and return immediately.
1683 * This policy do not drop the packet received in the RX
1684 * descriptor for which the allocation of a new mbuf failed.
1685 * Thus, it allows that packet to be later retrieved if
1686 * mbuf have been freed in the mean time.
1687 * As a side effect, holding RX descriptors instead of
1688 * systematically giving them back to the NIC may lead to
1689 * RX ring exhaustion situations.
1690 * However, the NIC can gracefully prevent such situations
1691 * to happen by sending specific "back-pressure" flow control
1692 * frames to its peer(s).
1694 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1695 "ext_err_stat=0x%08x pkt_len=%u",
1696 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1697 (unsigned) rx_id, (unsigned) staterr,
1698 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1700 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
1702 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1703 "queue_id=%u", (unsigned) rxq->port_id,
1704 (unsigned) rxq->queue_id);
1705 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1710 rxe = &sw_ring[rx_id];
1712 if (rx_id == rxq->nb_rx_desc)
1715 /* Prefetch next mbuf while processing current one. */
1716 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1719 * When next RX descriptor is on a cache-line boundary,
1720 * prefetch the next 4 RX descriptors and the next 8 pointers
1723 if ((rx_id & 0x3) == 0) {
1724 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1725 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1731 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(nmb));
1732 rxdp->read.hdr_addr = 0;
1733 rxdp->read.pkt_addr = dma_addr;
1736 * Initialize the returned mbuf.
1737 * 1) setup generic mbuf fields:
1738 * - number of segments,
1741 * - RX port identifier.
1742 * 2) integrate hardware offload data, if any:
1743 * - RSS flag & hash,
1744 * - IP checksum flag,
1745 * - VLAN TCI, if any,
1748 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1750 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1751 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1754 rxm->pkt_len = pkt_len;
1755 rxm->data_len = pkt_len;
1756 rxm->port = rxq->port_id;
1758 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1759 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1760 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1762 pkt_flags = rx_desc_status_to_pkt_flags(staterr, vlan_flags);
1763 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1764 pkt_flags = pkt_flags |
1765 ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1766 rxm->ol_flags = pkt_flags;
1768 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info,
1769 rxq->pkt_type_mask);
1771 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1772 rxm->hash.rss = rte_le_to_cpu_32(
1773 rxd.wb.lower.hi_dword.rss);
1774 else if (pkt_flags & PKT_RX_FDIR) {
1775 rxm->hash.fdir.hash = rte_le_to_cpu_16(
1776 rxd.wb.lower.hi_dword.csum_ip.csum) &
1777 IXGBE_ATR_HASH_MASK;
1778 rxm->hash.fdir.id = rte_le_to_cpu_16(
1779 rxd.wb.lower.hi_dword.csum_ip.ip_id);
1782 * Store the mbuf address into the next entry of the array
1783 * of returned packets.
1785 rx_pkts[nb_rx++] = rxm;
1787 rxq->rx_tail = rx_id;
1790 * If the number of free RX descriptors is greater than the RX free
1791 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1793 * Update the RDT with the value of the last processed RX descriptor
1794 * minus 1, to guarantee that the RDT register is never equal to the
1795 * RDH register, which creates a "full" ring situtation from the
1796 * hardware point of view...
1798 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1799 if (nb_hold > rxq->rx_free_thresh) {
1800 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1801 "nb_hold=%u nb_rx=%u",
1802 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1803 (unsigned) rx_id, (unsigned) nb_hold,
1805 rx_id = (uint16_t) ((rx_id == 0) ?
1806 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1807 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1810 rxq->nb_rx_hold = nb_hold;
1815 * Detect an RSC descriptor.
1817 static inline uint32_t
1818 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1820 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1821 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1825 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1827 * Fill the following info in the HEAD buffer of the Rx cluster:
1828 * - RX port identifier
1829 * - hardware offload data, if any:
1831 * - IP checksum flag
1832 * - VLAN TCI, if any
1834 * @head HEAD of the packet cluster
1835 * @desc HW descriptor to get data from
1836 * @rxq Pointer to the Rx queue
1839 ixgbe_fill_cluster_head_buf(
1840 struct rte_mbuf *head,
1841 union ixgbe_adv_rx_desc *desc,
1842 struct ixgbe_rx_queue *rxq,
1848 head->port = rxq->port_id;
1850 /* The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1851 * set in the pkt_flags field.
1853 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1854 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1855 pkt_flags = rx_desc_status_to_pkt_flags(staterr, rxq->vlan_flags);
1856 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1857 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags((uint16_t)pkt_info);
1858 head->ol_flags = pkt_flags;
1860 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info, rxq->pkt_type_mask);
1862 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1863 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1864 else if (pkt_flags & PKT_RX_FDIR) {
1865 head->hash.fdir.hash =
1866 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1867 & IXGBE_ATR_HASH_MASK;
1868 head->hash.fdir.id =
1869 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1874 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1876 * @rx_queue Rx queue handle
1877 * @rx_pkts table of received packets
1878 * @nb_pkts size of rx_pkts table
1879 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1881 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1882 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1884 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1885 * 1) When non-EOP RSC completion arrives:
1886 * a) Update the HEAD of the current RSC aggregation cluster with the new
1887 * segment's data length.
1888 * b) Set the "next" pointer of the current segment to point to the segment
1889 * at the NEXTP index.
1890 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
1891 * in the sw_rsc_ring.
1892 * 2) When EOP arrives we just update the cluster's total length and offload
1893 * flags and deliver the cluster up to the upper layers. In our case - put it
1894 * in the rx_pkts table.
1896 * Returns the number of received packets/clusters (according to the "bulk
1897 * receive" interface).
1899 static inline uint16_t
1900 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
1903 struct ixgbe_rx_queue *rxq = rx_queue;
1904 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
1905 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
1906 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
1907 uint16_t rx_id = rxq->rx_tail;
1909 uint16_t nb_hold = rxq->nb_rx_hold;
1910 uint16_t prev_id = rxq->rx_tail;
1912 while (nb_rx < nb_pkts) {
1914 struct ixgbe_rx_entry *rxe;
1915 struct ixgbe_scattered_rx_entry *sc_entry;
1916 struct ixgbe_scattered_rx_entry *next_sc_entry;
1917 struct ixgbe_rx_entry *next_rxe = NULL;
1918 struct rte_mbuf *first_seg;
1919 struct rte_mbuf *rxm;
1920 struct rte_mbuf *nmb;
1921 union ixgbe_adv_rx_desc rxd;
1924 volatile union ixgbe_adv_rx_desc *rxdp;
1929 * The code in this whole file uses the volatile pointer to
1930 * ensure the read ordering of the status and the rest of the
1931 * descriptor fields (on the compiler level only!!!). This is so
1932 * UGLY - why not to just use the compiler barrier instead? DPDK
1933 * even has the rte_compiler_barrier() for that.
1935 * But most importantly this is just wrong because this doesn't
1936 * ensure memory ordering in a general case at all. For
1937 * instance, DPDK is supposed to work on Power CPUs where
1938 * compiler barrier may just not be enough!
1940 * I tried to write only this function properly to have a
1941 * starting point (as a part of an LRO/RSC series) but the
1942 * compiler cursed at me when I tried to cast away the
1943 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
1944 * keeping it the way it is for now.
1946 * The code in this file is broken in so many other places and
1947 * will just not work on a big endian CPU anyway therefore the
1948 * lines below will have to be revisited together with the rest
1952 * - Get rid of "volatile" crap and let the compiler do its
1954 * - Use the proper memory barrier (rte_rmb()) to ensure the
1955 * memory ordering below.
1957 rxdp = &rx_ring[rx_id];
1958 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
1960 if (!(staterr & IXGBE_RXDADV_STAT_DD))
1965 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1966 "staterr=0x%x data_len=%u",
1967 rxq->port_id, rxq->queue_id, rx_id, staterr,
1968 rte_le_to_cpu_16(rxd.wb.upper.length));
1971 nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
1973 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
1974 "port_id=%u queue_id=%u",
1975 rxq->port_id, rxq->queue_id);
1977 rte_eth_devices[rxq->port_id].data->
1978 rx_mbuf_alloc_failed++;
1981 } else if (nb_hold > rxq->rx_free_thresh) {
1982 uint16_t next_rdt = rxq->rx_free_trigger;
1984 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
1986 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr,
1988 nb_hold -= rxq->rx_free_thresh;
1990 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
1991 "port_id=%u queue_id=%u",
1992 rxq->port_id, rxq->queue_id);
1994 rte_eth_devices[rxq->port_id].data->
1995 rx_mbuf_alloc_failed++;
2001 rxe = &sw_ring[rx_id];
2002 eop = staterr & IXGBE_RXDADV_STAT_EOP;
2004 next_id = rx_id + 1;
2005 if (next_id == rxq->nb_rx_desc)
2008 /* Prefetch next mbuf while processing current one. */
2009 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
2012 * When next RX descriptor is on a cache-line boundary,
2013 * prefetch the next 4 RX descriptors and the next 4 pointers
2016 if ((next_id & 0x3) == 0) {
2017 rte_ixgbe_prefetch(&rx_ring[next_id]);
2018 rte_ixgbe_prefetch(&sw_ring[next_id]);
2025 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(nmb));
2027 * Update RX descriptor with the physical address of the
2028 * new data buffer of the new allocated mbuf.
2032 rxm->data_off = RTE_PKTMBUF_HEADROOM;
2033 rxdp->read.hdr_addr = 0;
2034 rxdp->read.pkt_addr = dma;
2039 * Set data length & data buffer address of mbuf.
2041 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
2042 rxm->data_len = data_len;
2047 * Get next descriptor index:
2048 * - For RSC it's in the NEXTP field.
2049 * - For a scattered packet - it's just a following
2052 if (ixgbe_rsc_count(&rxd))
2054 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
2055 IXGBE_RXDADV_NEXTP_SHIFT;
2059 next_sc_entry = &sw_sc_ring[nextp_id];
2060 next_rxe = &sw_ring[nextp_id];
2061 rte_ixgbe_prefetch(next_rxe);
2064 sc_entry = &sw_sc_ring[rx_id];
2065 first_seg = sc_entry->fbuf;
2066 sc_entry->fbuf = NULL;
2069 * If this is the first buffer of the received packet,
2070 * set the pointer to the first mbuf of the packet and
2071 * initialize its context.
2072 * Otherwise, update the total length and the number of segments
2073 * of the current scattered packet, and update the pointer to
2074 * the last mbuf of the current packet.
2076 if (first_seg == NULL) {
2078 first_seg->pkt_len = data_len;
2079 first_seg->nb_segs = 1;
2081 first_seg->pkt_len += data_len;
2082 first_seg->nb_segs++;
2089 * If this is not the last buffer of the received packet, update
2090 * the pointer to the first mbuf at the NEXTP entry in the
2091 * sw_sc_ring and continue to parse the RX ring.
2093 if (!eop && next_rxe) {
2094 rxm->next = next_rxe->mbuf;
2095 next_sc_entry->fbuf = first_seg;
2100 * This is the last buffer of the received packet - return
2101 * the current cluster to the user.
2105 /* Initialize the first mbuf of the returned packet */
2106 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq, staterr);
2109 * Deal with the case, when HW CRC srip is disabled.
2110 * That can't happen when LRO is enabled, but still could
2111 * happen for scattered RX mode.
2113 first_seg->pkt_len -= rxq->crc_len;
2114 if (unlikely(rxm->data_len <= rxq->crc_len)) {
2115 struct rte_mbuf *lp;
2117 for (lp = first_seg; lp->next != rxm; lp = lp->next)
2120 first_seg->nb_segs--;
2121 lp->data_len -= rxq->crc_len - rxm->data_len;
2123 rte_pktmbuf_free_seg(rxm);
2125 rxm->data_len -= rxq->crc_len;
2127 /* Prefetch data of first segment, if configured to do so. */
2128 rte_packet_prefetch((char *)first_seg->buf_addr +
2129 first_seg->data_off);
2132 * Store the mbuf address into the next entry of the array
2133 * of returned packets.
2135 rx_pkts[nb_rx++] = first_seg;
2139 * Record index of the next RX descriptor to probe.
2141 rxq->rx_tail = rx_id;
2144 * If the number of free RX descriptors is greater than the RX free
2145 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
2147 * Update the RDT with the value of the last processed RX descriptor
2148 * minus 1, to guarantee that the RDT register is never equal to the
2149 * RDH register, which creates a "full" ring situtation from the
2150 * hardware point of view...
2152 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
2153 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
2154 "nb_hold=%u nb_rx=%u",
2155 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
2158 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, prev_id);
2162 rxq->nb_rx_hold = nb_hold;
2167 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2170 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
2174 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
2177 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
2180 /*********************************************************************
2182 * Queue management functions
2184 **********************************************************************/
2186 static void __attribute__((cold))
2187 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
2191 if (txq->sw_ring != NULL) {
2192 for (i = 0; i < txq->nb_tx_desc; i++) {
2193 if (txq->sw_ring[i].mbuf != NULL) {
2194 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
2195 txq->sw_ring[i].mbuf = NULL;
2201 static void __attribute__((cold))
2202 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
2205 txq->sw_ring != NULL)
2206 rte_free(txq->sw_ring);
2209 static void __attribute__((cold))
2210 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
2212 if (txq != NULL && txq->ops != NULL) {
2213 txq->ops->release_mbufs(txq);
2214 txq->ops->free_swring(txq);
2219 void __attribute__((cold))
2220 ixgbe_dev_tx_queue_release(void *txq)
2222 ixgbe_tx_queue_release(txq);
2225 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
2226 static void __attribute__((cold))
2227 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
2229 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
2230 struct ixgbe_tx_entry *txe = txq->sw_ring;
2233 /* Zero out HW ring memory */
2234 for (i = 0; i < txq->nb_tx_desc; i++) {
2235 txq->tx_ring[i] = zeroed_desc;
2238 /* Initialize SW ring entries */
2239 prev = (uint16_t) (txq->nb_tx_desc - 1);
2240 for (i = 0; i < txq->nb_tx_desc; i++) {
2241 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
2243 txd->wb.status = rte_cpu_to_le_32(IXGBE_TXD_STAT_DD);
2246 txe[prev].next_id = i;
2250 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2251 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2254 txq->nb_tx_used = 0;
2256 * Always allow 1 descriptor to be un-allocated to avoid
2257 * a H/W race condition
2259 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2260 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2262 memset((void *)&txq->ctx_cache, 0,
2263 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2266 static const struct ixgbe_txq_ops def_txq_ops = {
2267 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2268 .free_swring = ixgbe_tx_free_swring,
2269 .reset = ixgbe_reset_tx_queue,
2272 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2273 * the queue parameters. Used in tx_queue_setup by primary process and then
2274 * in dev_init by secondary process when attaching to an existing ethdev.
2276 void __attribute__((cold))
2277 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2279 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2280 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
2281 && (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2282 PMD_INIT_LOG(DEBUG, "Using simple tx code path");
2283 #ifdef RTE_IXGBE_INC_VECTOR
2284 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2285 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2286 ixgbe_txq_vec_setup(txq) == 0)) {
2287 PMD_INIT_LOG(DEBUG, "Vector tx enabled.");
2288 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2291 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2293 PMD_INIT_LOG(DEBUG, "Using full-featured tx code path");
2295 " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
2296 (unsigned long)txq->txq_flags,
2297 (unsigned long)IXGBE_SIMPLE_FLAGS);
2299 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2300 (unsigned long)txq->tx_rs_thresh,
2301 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2302 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2306 int __attribute__((cold))
2307 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2310 unsigned int socket_id,
2311 const struct rte_eth_txconf *tx_conf)
2313 const struct rte_memzone *tz;
2314 struct ixgbe_tx_queue *txq;
2315 struct ixgbe_hw *hw;
2316 uint16_t tx_rs_thresh, tx_free_thresh;
2318 PMD_INIT_FUNC_TRACE();
2319 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2322 * Validate number of transmit descriptors.
2323 * It must not exceed hardware maximum, and must be multiple
2326 if (nb_desc % IXGBE_TXD_ALIGN != 0 ||
2327 (nb_desc > IXGBE_MAX_RING_DESC) ||
2328 (nb_desc < IXGBE_MIN_RING_DESC)) {
2333 * The following two parameters control the setting of the RS bit on
2334 * transmit descriptors.
2335 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2336 * descriptors have been used.
2337 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2338 * descriptors are used or if the number of descriptors required
2339 * to transmit a packet is greater than the number of free TX
2341 * The following constraints must be satisfied:
2342 * tx_rs_thresh must be greater than 0.
2343 * tx_rs_thresh must be less than the size of the ring minus 2.
2344 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2345 * tx_rs_thresh must be a divisor of the ring size.
2346 * tx_free_thresh must be greater than 0.
2347 * tx_free_thresh must be less than the size of the ring minus 3.
2348 * One descriptor in the TX ring is used as a sentinel to avoid a
2349 * H/W race condition, hence the maximum threshold constraints.
2350 * When set to zero use default values.
2352 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2353 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2354 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2355 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2356 if (tx_rs_thresh >= (nb_desc - 2)) {
2357 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2358 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2359 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2360 (int)dev->data->port_id, (int)queue_idx);
2363 if (tx_rs_thresh > DEFAULT_TX_RS_THRESH) {
2364 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less or equal than %u. "
2365 "(tx_rs_thresh=%u port=%d queue=%d)",
2366 DEFAULT_TX_RS_THRESH, (unsigned int)tx_rs_thresh,
2367 (int)dev->data->port_id, (int)queue_idx);
2370 if (tx_free_thresh >= (nb_desc - 3)) {
2371 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2372 "tx_free_thresh must be less than the number of "
2373 "TX descriptors minus 3. (tx_free_thresh=%u "
2374 "port=%d queue=%d)",
2375 (unsigned int)tx_free_thresh,
2376 (int)dev->data->port_id, (int)queue_idx);
2379 if (tx_rs_thresh > tx_free_thresh) {
2380 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2381 "tx_free_thresh. (tx_free_thresh=%u "
2382 "tx_rs_thresh=%u port=%d queue=%d)",
2383 (unsigned int)tx_free_thresh,
2384 (unsigned int)tx_rs_thresh,
2385 (int)dev->data->port_id,
2389 if ((nb_desc % tx_rs_thresh) != 0) {
2390 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2391 "number of TX descriptors. (tx_rs_thresh=%u "
2392 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2393 (int)dev->data->port_id, (int)queue_idx);
2398 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2399 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2400 * by the NIC and all descriptors are written back after the NIC
2401 * accumulates WTHRESH descriptors.
2403 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2404 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2405 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2406 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2407 (int)dev->data->port_id, (int)queue_idx);
2411 /* Free memory prior to re-allocation if needed... */
2412 if (dev->data->tx_queues[queue_idx] != NULL) {
2413 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2414 dev->data->tx_queues[queue_idx] = NULL;
2417 /* First allocate the tx queue data structure */
2418 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2419 RTE_CACHE_LINE_SIZE, socket_id);
2424 * Allocate TX ring hardware descriptors. A memzone large enough to
2425 * handle the maximum ring size is allocated in order to allow for
2426 * resizing in later calls to the queue setup function.
2428 tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
2429 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2430 IXGBE_ALIGN, socket_id);
2432 ixgbe_tx_queue_release(txq);
2436 txq->nb_tx_desc = nb_desc;
2437 txq->tx_rs_thresh = tx_rs_thresh;
2438 txq->tx_free_thresh = tx_free_thresh;
2439 txq->pthresh = tx_conf->tx_thresh.pthresh;
2440 txq->hthresh = tx_conf->tx_thresh.hthresh;
2441 txq->wthresh = tx_conf->tx_thresh.wthresh;
2442 txq->queue_id = queue_idx;
2443 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2444 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2445 txq->port_id = dev->data->port_id;
2446 txq->txq_flags = tx_conf->txq_flags;
2447 txq->ops = &def_txq_ops;
2448 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2451 * Modification to set VFTDT for virtual function if vf is detected
2453 if (hw->mac.type == ixgbe_mac_82599_vf ||
2454 hw->mac.type == ixgbe_mac_X540_vf ||
2455 hw->mac.type == ixgbe_mac_X550_vf ||
2456 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2457 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2458 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2460 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2462 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2463 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2465 /* Allocate software ring */
2466 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2467 sizeof(struct ixgbe_tx_entry) * nb_desc,
2468 RTE_CACHE_LINE_SIZE, socket_id);
2469 if (txq->sw_ring == NULL) {
2470 ixgbe_tx_queue_release(txq);
2473 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2474 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2476 /* set up vector or scalar TX function as appropriate */
2477 ixgbe_set_tx_function(dev, txq);
2479 txq->ops->reset(txq);
2481 dev->data->tx_queues[queue_idx] = txq;
2488 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2490 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2491 * in the sw_rsc_ring is not set to NULL but rather points to the next
2492 * mbuf of this RSC aggregation (that has not been completed yet and still
2493 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2494 * will just free first "nb_segs" segments of the cluster explicitly by calling
2495 * an rte_pktmbuf_free_seg().
2497 * @m scattered cluster head
2499 static void __attribute__((cold))
2500 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2502 uint8_t i, nb_segs = m->nb_segs;
2503 struct rte_mbuf *next_seg;
2505 for (i = 0; i < nb_segs; i++) {
2507 rte_pktmbuf_free_seg(m);
2512 static void __attribute__((cold))
2513 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2517 #ifdef RTE_IXGBE_INC_VECTOR
2518 /* SSE Vector driver has a different way of releasing mbufs. */
2519 if (rxq->rx_using_sse) {
2520 ixgbe_rx_queue_release_mbufs_vec(rxq);
2525 if (rxq->sw_ring != NULL) {
2526 for (i = 0; i < rxq->nb_rx_desc; i++) {
2527 if (rxq->sw_ring[i].mbuf != NULL) {
2528 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2529 rxq->sw_ring[i].mbuf = NULL;
2532 if (rxq->rx_nb_avail) {
2533 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2534 struct rte_mbuf *mb;
2536 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2537 rte_pktmbuf_free_seg(mb);
2539 rxq->rx_nb_avail = 0;
2543 if (rxq->sw_sc_ring)
2544 for (i = 0; i < rxq->nb_rx_desc; i++)
2545 if (rxq->sw_sc_ring[i].fbuf) {
2546 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2547 rxq->sw_sc_ring[i].fbuf = NULL;
2551 static void __attribute__((cold))
2552 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2555 ixgbe_rx_queue_release_mbufs(rxq);
2556 rte_free(rxq->sw_ring);
2557 rte_free(rxq->sw_sc_ring);
2562 void __attribute__((cold))
2563 ixgbe_dev_rx_queue_release(void *rxq)
2565 ixgbe_rx_queue_release(rxq);
2569 * Check if Rx Burst Bulk Alloc function can be used.
2571 * 0: the preconditions are satisfied and the bulk allocation function
2573 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2574 * function must be used.
2576 static inline int __attribute__((cold))
2577 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2582 * Make sure the following pre-conditions are satisfied:
2583 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2584 * rxq->rx_free_thresh < rxq->nb_rx_desc
2585 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2586 * rxq->nb_rx_desc<(IXGBE_MAX_RING_DESC-RTE_PMD_IXGBE_RX_MAX_BURST)
2587 * Scattered packets are not supported. This should be checked
2588 * outside of this function.
2590 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2591 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2592 "rxq->rx_free_thresh=%d, "
2593 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2594 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2596 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2597 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2598 "rxq->rx_free_thresh=%d, "
2599 "rxq->nb_rx_desc=%d",
2600 rxq->rx_free_thresh, rxq->nb_rx_desc);
2602 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2603 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2604 "rxq->nb_rx_desc=%d, "
2605 "rxq->rx_free_thresh=%d",
2606 rxq->nb_rx_desc, rxq->rx_free_thresh);
2608 } else if (!(rxq->nb_rx_desc <
2609 (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST))) {
2610 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2611 "rxq->nb_rx_desc=%d, "
2612 "IXGBE_MAX_RING_DESC=%d, "
2613 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2614 rxq->nb_rx_desc, IXGBE_MAX_RING_DESC,
2615 RTE_PMD_IXGBE_RX_MAX_BURST);
2622 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2623 static void __attribute__((cold))
2624 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2626 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2628 uint16_t len = rxq->nb_rx_desc;
2631 * By default, the Rx queue setup function allocates enough memory for
2632 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2633 * extra memory at the end of the descriptor ring to be zero'd out. A
2634 * pre-condition for using the Rx burst bulk alloc function is that the
2635 * number of descriptors is less than or equal to
2636 * (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST). Check all the
2637 * constraints here to see if we need to zero out memory after the end
2638 * of the H/W descriptor ring.
2640 if (adapter->rx_bulk_alloc_allowed)
2641 /* zero out extra memory */
2642 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2645 * Zero out HW ring memory. Zero out extra memory at the end of
2646 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2647 * reads extra memory as zeros.
2649 for (i = 0; i < len; i++) {
2650 rxq->rx_ring[i] = zeroed_desc;
2654 * initialize extra software ring entries. Space for these extra
2655 * entries is always allocated
2657 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2658 for (i = rxq->nb_rx_desc; i < len; ++i) {
2659 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2662 rxq->rx_nb_avail = 0;
2663 rxq->rx_next_avail = 0;
2664 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2666 rxq->nb_rx_hold = 0;
2667 rxq->pkt_first_seg = NULL;
2668 rxq->pkt_last_seg = NULL;
2670 #ifdef RTE_IXGBE_INC_VECTOR
2671 rxq->rxrearm_start = 0;
2672 rxq->rxrearm_nb = 0;
2676 int __attribute__((cold))
2677 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2680 unsigned int socket_id,
2681 const struct rte_eth_rxconf *rx_conf,
2682 struct rte_mempool *mp)
2684 const struct rte_memzone *rz;
2685 struct ixgbe_rx_queue *rxq;
2686 struct ixgbe_hw *hw;
2688 struct ixgbe_adapter *adapter =
2689 (struct ixgbe_adapter *)dev->data->dev_private;
2691 PMD_INIT_FUNC_TRACE();
2692 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2695 * Validate number of receive descriptors.
2696 * It must not exceed hardware maximum, and must be multiple
2699 if (nb_desc % IXGBE_RXD_ALIGN != 0 ||
2700 (nb_desc > IXGBE_MAX_RING_DESC) ||
2701 (nb_desc < IXGBE_MIN_RING_DESC)) {
2705 /* Free memory prior to re-allocation if needed... */
2706 if (dev->data->rx_queues[queue_idx] != NULL) {
2707 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2708 dev->data->rx_queues[queue_idx] = NULL;
2711 /* First allocate the rx queue data structure */
2712 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2713 RTE_CACHE_LINE_SIZE, socket_id);
2717 rxq->nb_rx_desc = nb_desc;
2718 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2719 rxq->queue_id = queue_idx;
2720 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2721 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2722 rxq->port_id = dev->data->port_id;
2723 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2725 rxq->drop_en = rx_conf->rx_drop_en;
2726 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2729 * The packet type in RX descriptor is different for different NICs.
2730 * Some bits are used for x550 but reserved for other NICS.
2731 * So set different masks for different NICs.
2733 if (hw->mac.type == ixgbe_mac_X550 ||
2734 hw->mac.type == ixgbe_mac_X550EM_x ||
2735 hw->mac.type == ixgbe_mac_X550EM_a ||
2736 hw->mac.type == ixgbe_mac_X550_vf ||
2737 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2738 hw->mac.type == ixgbe_mac_X550EM_a_vf)
2739 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_X550;
2741 rxq->pkt_type_mask = IXGBE_PACKET_TYPE_MASK_82599;
2744 * Allocate RX ring hardware descriptors. A memzone large enough to
2745 * handle the maximum ring size is allocated in order to allow for
2746 * resizing in later calls to the queue setup function.
2748 rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
2749 RX_RING_SZ, IXGBE_ALIGN, socket_id);
2751 ixgbe_rx_queue_release(rxq);
2756 * Zero init all the descriptors in the ring.
2758 memset(rz->addr, 0, RX_RING_SZ);
2761 * Modified to setup VFRDT for Virtual Function
2763 if (hw->mac.type == ixgbe_mac_82599_vf ||
2764 hw->mac.type == ixgbe_mac_X540_vf ||
2765 hw->mac.type == ixgbe_mac_X550_vf ||
2766 hw->mac.type == ixgbe_mac_X550EM_x_vf ||
2767 hw->mac.type == ixgbe_mac_X550EM_a_vf) {
2769 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2771 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2774 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2776 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2779 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2780 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2783 * Certain constraints must be met in order to use the bulk buffer
2784 * allocation Rx burst function. If any of Rx queues doesn't meet them
2785 * the feature should be disabled for the whole port.
2787 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
2788 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
2789 "preconditions - canceling the feature for "
2790 "the whole port[%d]",
2791 rxq->queue_id, rxq->port_id);
2792 adapter->rx_bulk_alloc_allowed = false;
2796 * Allocate software ring. Allow for space at the end of the
2797 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2798 * function does not access an invalid memory region.
2801 if (adapter->rx_bulk_alloc_allowed)
2802 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2804 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2805 sizeof(struct ixgbe_rx_entry) * len,
2806 RTE_CACHE_LINE_SIZE, socket_id);
2807 if (!rxq->sw_ring) {
2808 ixgbe_rx_queue_release(rxq);
2813 * Always allocate even if it's not going to be needed in order to
2814 * simplify the code.
2816 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
2817 * be requested in ixgbe_dev_rx_init(), which is called later from
2821 rte_zmalloc_socket("rxq->sw_sc_ring",
2822 sizeof(struct ixgbe_scattered_rx_entry) * len,
2823 RTE_CACHE_LINE_SIZE, socket_id);
2824 if (!rxq->sw_sc_ring) {
2825 ixgbe_rx_queue_release(rxq);
2829 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
2830 "dma_addr=0x%"PRIx64,
2831 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
2832 rxq->rx_ring_phys_addr);
2834 if (!rte_is_power_of_2(nb_desc)) {
2835 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
2836 "preconditions - canceling the feature for "
2837 "the whole port[%d]",
2838 rxq->queue_id, rxq->port_id);
2839 adapter->rx_vec_allowed = false;
2841 ixgbe_rxq_vec_setup(rxq);
2843 dev->data->rx_queues[queue_idx] = rxq;
2845 ixgbe_reset_rx_queue(adapter, rxq);
2851 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2853 #define IXGBE_RXQ_SCAN_INTERVAL 4
2854 volatile union ixgbe_adv_rx_desc *rxdp;
2855 struct ixgbe_rx_queue *rxq;
2858 if (rx_queue_id >= dev->data->nb_rx_queues) {
2859 PMD_RX_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
2863 rxq = dev->data->rx_queues[rx_queue_id];
2864 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2866 while ((desc < rxq->nb_rx_desc) &&
2867 (rxdp->wb.upper.status_error &
2868 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD))) {
2869 desc += IXGBE_RXQ_SCAN_INTERVAL;
2870 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2871 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2872 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2873 desc - rxq->nb_rx_desc]);
2880 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2882 volatile union ixgbe_adv_rx_desc *rxdp;
2883 struct ixgbe_rx_queue *rxq = rx_queue;
2886 if (unlikely(offset >= rxq->nb_rx_desc))
2888 desc = rxq->rx_tail + offset;
2889 if (desc >= rxq->nb_rx_desc)
2890 desc -= rxq->nb_rx_desc;
2892 rxdp = &rxq->rx_ring[desc];
2893 return !!(rxdp->wb.upper.status_error &
2894 rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD));
2897 void __attribute__((cold))
2898 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
2901 struct ixgbe_adapter *adapter =
2902 (struct ixgbe_adapter *)dev->data->dev_private;
2904 PMD_INIT_FUNC_TRACE();
2906 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2907 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
2910 txq->ops->release_mbufs(txq);
2911 txq->ops->reset(txq);
2915 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2916 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
2919 ixgbe_rx_queue_release_mbufs(rxq);
2920 ixgbe_reset_rx_queue(adapter, rxq);
2926 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
2930 PMD_INIT_FUNC_TRACE();
2932 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2933 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
2934 dev->data->rx_queues[i] = NULL;
2936 dev->data->nb_rx_queues = 0;
2938 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2939 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
2940 dev->data->tx_queues[i] = NULL;
2942 dev->data->nb_tx_queues = 0;
2945 /*********************************************************************
2947 * Device RX/TX init functions
2949 **********************************************************************/
2952 * Receive Side Scaling (RSS)
2953 * See section 7.1.2.8 in the following document:
2954 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
2957 * The source and destination IP addresses of the IP header and the source
2958 * and destination ports of TCP/UDP headers, if any, of received packets are
2959 * hashed against a configurable random key to compute a 32-bit RSS hash result.
2960 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
2961 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
2962 * RSS output index which is used as the RX queue index where to store the
2964 * The following output is supplied in the RX write-back descriptor:
2965 * - 32-bit result of the Microsoft RSS hash function,
2966 * - 4-bit RSS type field.
2970 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
2971 * Used as the default key.
2973 static uint8_t rss_intel_key[40] = {
2974 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2975 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2976 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2977 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2978 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2982 ixgbe_rss_disable(struct rte_eth_dev *dev)
2984 struct ixgbe_hw *hw;
2988 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2989 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
2990 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
2991 mrqc &= ~IXGBE_MRQC_RSSEN;
2992 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
2996 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
3006 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3007 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3009 hash_key = rss_conf->rss_key;
3010 if (hash_key != NULL) {
3011 /* Fill in RSS hash key */
3012 for (i = 0; i < 10; i++) {
3013 rss_key = hash_key[(i * 4)];
3014 rss_key |= hash_key[(i * 4) + 1] << 8;
3015 rss_key |= hash_key[(i * 4) + 2] << 16;
3016 rss_key |= hash_key[(i * 4) + 3] << 24;
3017 IXGBE_WRITE_REG_ARRAY(hw, rssrk_reg, i, rss_key);
3021 /* Set configured hashing protocols in MRQC register */
3022 rss_hf = rss_conf->rss_hf;
3023 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
3024 if (rss_hf & ETH_RSS_IPV4)
3025 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
3026 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
3027 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
3028 if (rss_hf & ETH_RSS_IPV6)
3029 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
3030 if (rss_hf & ETH_RSS_IPV6_EX)
3031 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
3032 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
3033 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3034 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
3035 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
3036 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
3037 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3038 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
3039 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3040 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
3041 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
3042 IXGBE_WRITE_REG(hw, mrqc_reg, mrqc);
3046 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
3047 struct rte_eth_rss_conf *rss_conf)
3049 struct ixgbe_hw *hw;
3054 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3056 if (!ixgbe_rss_update_sp(hw->mac.type)) {
3057 PMD_DRV_LOG(ERR, "RSS hash update is not supported on this "
3061 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3064 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
3065 * "RSS enabling cannot be done dynamically while it must be
3066 * preceded by a software reset"
3067 * Before changing anything, first check that the update RSS operation
3068 * does not attempt to disable RSS, if RSS was enabled at
3069 * initialization time, or does not attempt to enable RSS, if RSS was
3070 * disabled at initialization time.
3072 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
3073 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3074 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
3075 if (rss_hf != 0) /* Enable RSS */
3077 return 0; /* Nothing to do */
3080 if (rss_hf == 0) /* Disable RSS */
3082 ixgbe_hw_rss_hash_set(hw, rss_conf);
3087 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
3088 struct rte_eth_rss_conf *rss_conf)
3090 struct ixgbe_hw *hw;
3099 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3100 mrqc_reg = ixgbe_mrqc_reg_get(hw->mac.type);
3101 rssrk_reg = ixgbe_rssrk_reg_get(hw->mac.type, 0);
3102 hash_key = rss_conf->rss_key;
3103 if (hash_key != NULL) {
3104 /* Return RSS hash key */
3105 for (i = 0; i < 10; i++) {
3106 rss_key = IXGBE_READ_REG_ARRAY(hw, rssrk_reg, i);
3107 hash_key[(i * 4)] = rss_key & 0x000000FF;
3108 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
3109 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
3110 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
3114 /* Get RSS functions configured in MRQC register */
3115 mrqc = IXGBE_READ_REG(hw, mrqc_reg);
3116 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
3117 rss_conf->rss_hf = 0;
3121 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
3122 rss_hf |= ETH_RSS_IPV4;
3123 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
3124 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
3125 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
3126 rss_hf |= ETH_RSS_IPV6;
3127 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
3128 rss_hf |= ETH_RSS_IPV6_EX;
3129 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
3130 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
3131 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
3132 rss_hf |= ETH_RSS_IPV6_TCP_EX;
3133 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
3134 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
3135 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
3136 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
3137 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
3138 rss_hf |= ETH_RSS_IPV6_UDP_EX;
3139 rss_conf->rss_hf = rss_hf;
3144 ixgbe_rss_configure(struct rte_eth_dev *dev)
3146 struct rte_eth_rss_conf rss_conf;
3147 struct ixgbe_hw *hw;
3151 uint16_t sp_reta_size;
3154 PMD_INIT_FUNC_TRACE();
3155 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3157 sp_reta_size = ixgbe_reta_size_get(hw->mac.type);
3160 * Fill in redirection table
3161 * The byte-swap is needed because NIC registers are in
3162 * little-endian order.
3165 for (i = 0, j = 0; i < sp_reta_size; i++, j++) {
3166 reta_reg = ixgbe_reta_reg_get(hw->mac.type, i);
3168 if (j == dev->data->nb_rx_queues)
3170 reta = (reta << 8) | j;
3172 IXGBE_WRITE_REG(hw, reta_reg,
3177 * Configure the RSS key and the RSS protocols used to compute
3178 * the RSS hash of input packets.
3180 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
3181 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
3182 ixgbe_rss_disable(dev);
3185 if (rss_conf.rss_key == NULL)
3186 rss_conf.rss_key = rss_intel_key; /* Default hash key */
3187 ixgbe_hw_rss_hash_set(hw, &rss_conf);
3190 #define NUM_VFTA_REGISTERS 128
3191 #define NIC_RX_BUFFER_SIZE 0x200
3192 #define X550_RX_BUFFER_SIZE 0x180
3195 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
3197 struct rte_eth_vmdq_dcb_conf *cfg;
3198 struct ixgbe_hw *hw;
3199 enum rte_eth_nb_pools num_pools;
3200 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
3202 uint8_t nb_tcs; /* number of traffic classes */
3205 PMD_INIT_FUNC_TRACE();
3206 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3207 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3208 num_pools = cfg->nb_queue_pools;
3209 /* Check we have a valid number of pools */
3210 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
3211 ixgbe_rss_disable(dev);
3214 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
3215 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
3219 * split rx buffer up into sections, each for 1 traffic class
3221 switch (hw->mac.type) {
3222 case ixgbe_mac_X550:
3223 case ixgbe_mac_X550EM_x:
3224 case ixgbe_mac_X550EM_a:
3225 pbsize = (uint16_t)(X550_RX_BUFFER_SIZE / nb_tcs);
3228 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3231 for (i = 0; i < nb_tcs; i++) {
3232 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3234 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3235 /* clear 10 bits. */
3236 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
3237 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3239 /* zero alloc all unused TCs */
3240 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3241 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
3243 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
3244 /* clear 10 bits. */
3245 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3248 /* MRQC: enable vmdq and dcb */
3249 mrqc = (num_pools == ETH_16_POOLS) ?
3250 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN;
3251 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3253 /* PFVTCTL: turn on virtualisation and set the default pool */
3254 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3255 if (cfg->enable_default_pool) {
3256 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3258 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3261 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3263 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
3265 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
3267 * mapping is done with 3 bits per priority,
3268 * so shift by i*3 each time
3270 queue_mapping |= ((cfg->dcb_tc[i] & 0x07) << (i * 3));
3272 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
3274 /* RTRPCS: DCB related */
3275 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
3277 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3278 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3279 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3280 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3282 /* VFTA - enable all vlan filters */
3283 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3284 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3287 /* VFRE: pool enabling for receive - 16 or 32 */
3288 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0),
3289 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3292 * MPSAR - allow pools to read specific mac addresses
3293 * In this case, all pools should be able to read from mac addr 0
3295 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
3296 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
3298 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3299 for (i = 0; i < cfg->nb_pool_maps; i++) {
3300 /* set vlan id in VF register and set the valid bit */
3301 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3302 (cfg->pool_map[i].vlan_id & 0xFFF)));
3304 * Put the allowed pools in VFB reg. As we only have 16 or 32
3305 * pools, we only need to use the first half of the register
3308 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
3313 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
3314 * @hw: pointer to hardware structure
3315 * @dcb_config: pointer to ixgbe_dcb_config structure
3318 ixgbe_dcb_tx_hw_config(struct ixgbe_hw *hw,
3319 struct ixgbe_dcb_config *dcb_config)
3324 PMD_INIT_FUNC_TRACE();
3325 if (hw->mac.type != ixgbe_mac_82598EB) {
3326 /* Disable the Tx desc arbiter so that MTQC can be changed */
3327 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3328 reg |= IXGBE_RTTDCS_ARBDIS;
3329 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3331 /* Enable DCB for Tx with 8 TCs */
3332 if (dcb_config->num_tcs.pg_tcs == 8) {
3333 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3335 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3337 if (dcb_config->vt_mode)
3338 reg |= IXGBE_MTQC_VT_ENA;
3339 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3341 /* Disable drop for all queues */
3342 for (q = 0; q < 128; q++)
3343 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3344 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3346 /* Enable the Tx desc arbiter */
3347 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3348 reg &= ~IXGBE_RTTDCS_ARBDIS;
3349 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3351 /* Enable Security TX Buffer IFG for DCB */
3352 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3353 reg |= IXGBE_SECTX_DCB;
3354 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3359 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3360 * @dev: pointer to rte_eth_dev structure
3361 * @dcb_config: pointer to ixgbe_dcb_config structure
3364 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3365 struct ixgbe_dcb_config *dcb_config)
3367 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3368 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3369 struct ixgbe_hw *hw =
3370 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3372 PMD_INIT_FUNC_TRACE();
3373 if (hw->mac.type != ixgbe_mac_82598EB)
3374 /*PF VF Transmit Enable*/
3375 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3376 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3378 /*Configure general DCB TX parameters*/
3379 ixgbe_dcb_tx_hw_config(hw, dcb_config);
3383 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3384 struct ixgbe_dcb_config *dcb_config)
3386 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3387 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3388 struct ixgbe_dcb_tc_config *tc;
3391 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3392 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS) {
3393 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3394 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3396 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3397 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3399 /* User Priority to Traffic Class mapping */
3400 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3401 j = vmdq_rx_conf->dcb_tc[i];
3402 tc = &dcb_config->tc_config[j];
3403 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3409 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3410 struct ixgbe_dcb_config *dcb_config)
3412 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3413 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3414 struct ixgbe_dcb_tc_config *tc;
3417 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3418 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS) {
3419 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3420 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3422 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3423 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3426 /* User Priority to Traffic Class mapping */
3427 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3428 j = vmdq_tx_conf->dcb_tc[i];
3429 tc = &dcb_config->tc_config[j];
3430 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3436 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3437 struct ixgbe_dcb_config *dcb_config)
3439 struct rte_eth_dcb_rx_conf *rx_conf =
3440 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3441 struct ixgbe_dcb_tc_config *tc;
3444 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3445 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3447 /* User Priority to Traffic Class mapping */
3448 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3449 j = rx_conf->dcb_tc[i];
3450 tc = &dcb_config->tc_config[j];
3451 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3457 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3458 struct ixgbe_dcb_config *dcb_config)
3460 struct rte_eth_dcb_tx_conf *tx_conf =
3461 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3462 struct ixgbe_dcb_tc_config *tc;
3465 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3466 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3468 /* User Priority to Traffic Class mapping */
3469 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3470 j = tx_conf->dcb_tc[i];
3471 tc = &dcb_config->tc_config[j];
3472 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3478 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3479 * @hw: pointer to hardware structure
3480 * @dcb_config: pointer to ixgbe_dcb_config structure
3483 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
3484 struct ixgbe_dcb_config *dcb_config)
3490 PMD_INIT_FUNC_TRACE();
3492 * Disable the arbiter before changing parameters
3493 * (always enable recycle mode; WSP)
3495 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3496 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3498 if (hw->mac.type != ixgbe_mac_82598EB) {
3499 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3500 if (dcb_config->num_tcs.pg_tcs == 4) {
3501 if (dcb_config->vt_mode)
3502 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3503 IXGBE_MRQC_VMDQRT4TCEN;
3505 /* no matter the mode is DCB or DCB_RSS, just
3506 * set the MRQE to RSSXTCEN. RSS is controlled
3509 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3510 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3511 IXGBE_MRQC_RTRSS4TCEN;
3514 if (dcb_config->num_tcs.pg_tcs == 8) {
3515 if (dcb_config->vt_mode)
3516 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3517 IXGBE_MRQC_VMDQRT8TCEN;
3519 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3520 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3521 IXGBE_MRQC_RTRSS8TCEN;
3525 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3528 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3529 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3530 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3531 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3533 /* VFTA - enable all vlan filters */
3534 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3535 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3539 * Configure Rx packet plane (recycle mode; WSP) and
3542 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3543 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3547 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3548 uint16_t *max, uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3550 switch (hw->mac.type) {
3551 case ixgbe_mac_82598EB:
3552 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3554 case ixgbe_mac_82599EB:
3555 case ixgbe_mac_X540:
3556 case ixgbe_mac_X550:
3557 case ixgbe_mac_X550EM_x:
3558 case ixgbe_mac_X550EM_a:
3559 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3568 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3569 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3571 switch (hw->mac.type) {
3572 case ixgbe_mac_82598EB:
3573 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id, tsa);
3574 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id, tsa);
3576 case ixgbe_mac_82599EB:
3577 case ixgbe_mac_X540:
3578 case ixgbe_mac_X550:
3579 case ixgbe_mac_X550EM_x:
3580 case ixgbe_mac_X550EM_a:
3581 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id, tsa);
3582 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id, tsa, map);
3589 #define DCB_RX_CONFIG 1
3590 #define DCB_TX_CONFIG 1
3591 #define DCB_TX_PB 1024
3593 * ixgbe_dcb_hw_configure - Enable DCB and configure
3594 * general DCB in VT mode and non-VT mode parameters
3595 * @dev: pointer to rte_eth_dev structure
3596 * @dcb_config: pointer to ixgbe_dcb_config structure
3599 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3600 struct ixgbe_dcb_config *dcb_config)
3603 uint8_t i, pfc_en, nb_tcs;
3604 uint16_t pbsize, rx_buffer_size;
3605 uint8_t config_dcb_rx = 0;
3606 uint8_t config_dcb_tx = 0;
3607 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3608 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3609 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3610 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3611 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3612 struct ixgbe_dcb_tc_config *tc;
3613 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3614 struct ixgbe_hw *hw =
3615 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3617 switch (dev->data->dev_conf.rxmode.mq_mode) {
3618 case ETH_MQ_RX_VMDQ_DCB:
3619 dcb_config->vt_mode = true;
3620 if (hw->mac.type != ixgbe_mac_82598EB) {
3621 config_dcb_rx = DCB_RX_CONFIG;
3623 *get dcb and VT rx configuration parameters
3626 ixgbe_vmdq_dcb_rx_config(dev, dcb_config);
3627 /*Configure general VMDQ and DCB RX parameters*/
3628 ixgbe_vmdq_dcb_configure(dev);
3632 case ETH_MQ_RX_DCB_RSS:
3633 dcb_config->vt_mode = false;
3634 config_dcb_rx = DCB_RX_CONFIG;
3635 /* Get dcb TX configuration parameters from rte_eth_conf */
3636 ixgbe_dcb_rx_config(dev, dcb_config);
3637 /*Configure general DCB RX parameters*/
3638 ixgbe_dcb_rx_hw_config(hw, dcb_config);
3641 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3644 switch (dev->data->dev_conf.txmode.mq_mode) {
3645 case ETH_MQ_TX_VMDQ_DCB:
3646 dcb_config->vt_mode = true;
3647 config_dcb_tx = DCB_TX_CONFIG;
3648 /* get DCB and VT TX configuration parameters
3651 ixgbe_dcb_vt_tx_config(dev, dcb_config);
3652 /*Configure general VMDQ and DCB TX parameters*/
3653 ixgbe_vmdq_dcb_hw_tx_config(dev, dcb_config);
3657 dcb_config->vt_mode = false;
3658 config_dcb_tx = DCB_TX_CONFIG;
3659 /*get DCB TX configuration parameters from rte_eth_conf*/
3660 ixgbe_dcb_tx_config(dev, dcb_config);
3661 /*Configure general DCB TX parameters*/
3662 ixgbe_dcb_tx_hw_config(hw, dcb_config);
3665 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3669 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3671 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3672 if (nb_tcs == ETH_4_TCS) {
3673 /* Avoid un-configured priority mapping to TC0 */
3675 uint8_t mask = 0xFF;
3677 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3678 mask = (uint8_t)(mask & (~(1 << map[i])));
3679 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3680 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3684 /* Re-configure 4 TCs BW */
3685 for (i = 0; i < nb_tcs; i++) {
3686 tc = &dcb_config->tc_config[i];
3687 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
3688 (uint8_t)(100 / nb_tcs);
3689 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
3690 (uint8_t)(100 / nb_tcs);
3692 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
3693 tc = &dcb_config->tc_config[i];
3694 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
3695 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
3699 switch (hw->mac.type) {
3700 case ixgbe_mac_X550:
3701 case ixgbe_mac_X550EM_x:
3702 case ixgbe_mac_X550EM_a:
3703 rx_buffer_size = X550_RX_BUFFER_SIZE;
3706 rx_buffer_size = NIC_RX_BUFFER_SIZE;
3710 if (config_dcb_rx) {
3711 /* Set RX buffer size */
3712 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
3713 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
3715 for (i = 0; i < nb_tcs; i++) {
3716 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3718 /* zero alloc all unused TCs */
3719 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3720 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
3723 if (config_dcb_tx) {
3724 /* Only support an equally distributed
3725 * Tx packet buffer strategy.
3727 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
3728 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
3730 for (i = 0; i < nb_tcs; i++) {
3731 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
3732 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
3734 /* Clear unused TCs, if any, to zero buffer size*/
3735 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3736 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
3737 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
3741 /*Calculates traffic class credits*/
3742 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
3743 IXGBE_DCB_TX_CONFIG);
3744 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config, max_frame,
3745 IXGBE_DCB_RX_CONFIG);
3747 if (config_dcb_rx) {
3748 /* Unpack CEE standard containers */
3749 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
3750 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3751 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
3752 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
3753 /* Configure PG(ETS) RX */
3754 ixgbe_dcb_hw_arbite_rx_config(hw, refill, max, bwgid, tsa, map);
3757 if (config_dcb_tx) {
3758 /* Unpack CEE standard containers */
3759 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
3760 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3761 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
3762 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
3763 /* Configure PG(ETS) TX */
3764 ixgbe_dcb_hw_arbite_tx_config(hw, refill, max, bwgid, tsa, map);
3767 /*Configure queue statistics registers*/
3768 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3770 /* Check if the PFC is supported */
3771 if (dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3772 pbsize = (uint16_t)(rx_buffer_size / nb_tcs);
3773 for (i = 0; i < nb_tcs; i++) {
3775 * If the TC count is 8,and the default high_water is 48,
3776 * the low_water is 16 as default.
3778 hw->fc.high_water[i] = (pbsize * 3) / 4;
3779 hw->fc.low_water[i] = pbsize / 4;
3780 /* Enable pfc for this TC */
3781 tc = &dcb_config->tc_config[i];
3782 tc->pfc = ixgbe_dcb_pfc_enabled;
3784 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3785 if (dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3787 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3794 * ixgbe_configure_dcb - Configure DCB Hardware
3795 * @dev: pointer to rte_eth_dev
3797 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3799 struct ixgbe_dcb_config *dcb_cfg =
3800 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3801 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3803 PMD_INIT_FUNC_TRACE();
3805 /* check support mq_mode for DCB */
3806 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
3807 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB) &&
3808 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB_RSS))
3811 if (dev->data->nb_rx_queues != ETH_DCB_NUM_QUEUES)
3814 /** Configure DCB hardware **/
3815 ixgbe_dcb_hw_configure(dev, dcb_cfg);
3819 * VMDq only support for 10 GbE NIC.
3822 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
3824 struct rte_eth_vmdq_rx_conf *cfg;
3825 struct ixgbe_hw *hw;
3826 enum rte_eth_nb_pools num_pools;
3827 uint32_t mrqc, vt_ctl, vlanctrl;
3831 PMD_INIT_FUNC_TRACE();
3832 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3833 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
3834 num_pools = cfg->nb_queue_pools;
3836 ixgbe_rss_disable(dev);
3838 /* MRQC: enable vmdq */
3839 mrqc = IXGBE_MRQC_VMDQEN;
3840 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3842 /* PFVTCTL: turn on virtualisation and set the default pool */
3843 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3844 if (cfg->enable_default_pool)
3845 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3847 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3849 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3851 for (i = 0; i < (int)num_pools; i++) {
3852 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
3853 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
3856 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3857 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3858 vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
3859 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3861 /* VFTA - enable all vlan filters */
3862 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
3863 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
3865 /* VFRE: pool enabling for receive - 64 */
3866 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
3867 if (num_pools == ETH_64_POOLS)
3868 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
3871 * MPSAR - allow pools to read specific mac addresses
3872 * In this case, all pools should be able to read from mac addr 0
3874 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
3875 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
3877 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3878 for (i = 0; i < cfg->nb_pool_maps; i++) {
3879 /* set vlan id in VF register and set the valid bit */
3880 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN |
3881 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
3883 * Put the allowed pools in VFB reg. As we only have 16 or 64
3884 * pools, we only need to use the first half of the register
3887 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
3888 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i * 2),
3889 (cfg->pool_map[i].pools & UINT32_MAX));
3891 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i * 2 + 1)),
3892 ((cfg->pool_map[i].pools >> 32) & UINT32_MAX));
3896 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
3897 if (cfg->enable_loop_back) {
3898 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3899 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
3900 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
3903 IXGBE_WRITE_FLUSH(hw);
3907 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
3908 * @hw: pointer to hardware structure
3911 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
3916 PMD_INIT_FUNC_TRACE();
3917 /*PF VF Transmit Enable*/
3918 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
3919 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
3921 /* Disable the Tx desc arbiter so that MTQC can be changed */
3922 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3923 reg |= IXGBE_RTTDCS_ARBDIS;
3924 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3926 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3927 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3929 /* Disable drop for all queues */
3930 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3931 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3932 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3934 /* Enable the Tx desc arbiter */
3935 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3936 reg &= ~IXGBE_RTTDCS_ARBDIS;
3937 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3939 IXGBE_WRITE_FLUSH(hw);
3942 static int __attribute__((cold))
3943 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
3945 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
3949 /* Initialize software ring entries */
3950 for (i = 0; i < rxq->nb_rx_desc; i++) {
3951 volatile union ixgbe_adv_rx_desc *rxd;
3952 struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
3955 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
3956 (unsigned) rxq->queue_id);
3960 rte_mbuf_refcnt_set(mbuf, 1);
3962 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
3964 mbuf->port = rxq->port_id;
3967 rte_cpu_to_le_64(rte_mbuf_data_dma_addr_default(mbuf));
3968 rxd = &rxq->rx_ring[i];
3969 rxd->read.hdr_addr = 0;
3970 rxd->read.pkt_addr = dma_addr;
3978 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
3980 struct ixgbe_hw *hw;
3983 ixgbe_rss_configure(dev);
3985 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3987 /* MRQC: enable VF RSS */
3988 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
3989 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
3990 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3992 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
3996 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
4000 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
4004 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4010 ixgbe_config_vf_default(struct rte_eth_dev *dev)
4012 struct ixgbe_hw *hw =
4013 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4015 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4017 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4022 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4023 IXGBE_MRQC_VMDQRT4TCEN);
4027 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
4028 IXGBE_MRQC_VMDQRT8TCEN);
4032 "invalid pool number in IOV mode");
4039 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
4041 struct ixgbe_hw *hw =
4042 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4044 if (hw->mac.type == ixgbe_mac_82598EB)
4047 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4049 * SRIOV inactive scheme
4050 * any DCB/RSS w/o VMDq multi-queue setting
4052 switch (dev->data->dev_conf.rxmode.mq_mode) {
4054 case ETH_MQ_RX_DCB_RSS:
4055 case ETH_MQ_RX_VMDQ_RSS:
4056 ixgbe_rss_configure(dev);
4059 case ETH_MQ_RX_VMDQ_DCB:
4060 ixgbe_vmdq_dcb_configure(dev);
4063 case ETH_MQ_RX_VMDQ_ONLY:
4064 ixgbe_vmdq_rx_hw_configure(dev);
4067 case ETH_MQ_RX_NONE:
4069 /* if mq_mode is none, disable rss mode.*/
4070 ixgbe_rss_disable(dev);
4075 * SRIOV active scheme
4076 * Support RSS together with VMDq & SRIOV
4078 switch (dev->data->dev_conf.rxmode.mq_mode) {
4080 case ETH_MQ_RX_VMDQ_RSS:
4081 ixgbe_config_vf_rss(dev);
4084 /* FIXME if support DCB/RSS together with VMDq & SRIOV */
4085 case ETH_MQ_RX_VMDQ_DCB:
4086 case ETH_MQ_RX_VMDQ_DCB_RSS:
4088 "Could not support DCB with VMDq & SRIOV");
4091 ixgbe_config_vf_default(dev);
4100 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
4102 struct ixgbe_hw *hw =
4103 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4107 if (hw->mac.type == ixgbe_mac_82598EB)
4110 /* disable arbiter before setting MTQC */
4111 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
4112 rttdcs |= IXGBE_RTTDCS_ARBDIS;
4113 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4115 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
4117 * SRIOV inactive scheme
4118 * any DCB w/o VMDq multi-queue setting
4120 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
4121 ixgbe_vmdq_tx_hw_configure(hw);
4123 mtqc = IXGBE_MTQC_64Q_1PB;
4124 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4127 switch (RTE_ETH_DEV_SRIOV(dev).active) {
4130 * SRIOV active scheme
4131 * FIXME if support DCB together with VMDq & SRIOV
4134 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
4137 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
4140 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
4144 mtqc = IXGBE_MTQC_64Q_1PB;
4145 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
4147 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
4150 /* re-enable arbiter */
4151 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
4152 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
4158 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
4160 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
4161 * spec rev. 3.0 chapter 8.2.3.8.13.
4163 * @pool Memory pool of the Rx queue
4165 static inline uint32_t
4166 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
4168 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
4170 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
4173 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
4176 return IXGBE_RSCCTL_MAXDESC_16;
4177 else if (maxdesc >= 8)
4178 return IXGBE_RSCCTL_MAXDESC_8;
4179 else if (maxdesc >= 4)
4180 return IXGBE_RSCCTL_MAXDESC_4;
4182 return IXGBE_RSCCTL_MAXDESC_1;
4186 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
4189 * (Taken from FreeBSD tree)
4190 * (yes this is all very magic and confusing :)
4193 * @entry the register array entry
4194 * @vector the MSIX vector for this queue
4198 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
4200 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4203 vector |= IXGBE_IVAR_ALLOC_VAL;
4205 switch (hw->mac.type) {
4207 case ixgbe_mac_82598EB:
4209 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
4211 entry += (type * 64);
4212 index = (entry >> 2) & 0x1F;
4213 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
4214 ivar &= ~(0xFF << (8 * (entry & 0x3)));
4215 ivar |= (vector << (8 * (entry & 0x3)));
4216 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
4219 case ixgbe_mac_82599EB:
4220 case ixgbe_mac_X540:
4221 if (type == -1) { /* MISC IVAR */
4222 index = (entry & 1) * 8;
4223 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
4224 ivar &= ~(0xFF << index);
4225 ivar |= (vector << index);
4226 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
4227 } else { /* RX/TX IVARS */
4228 index = (16 * (entry & 1)) + (8 * type);
4229 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
4230 ivar &= ~(0xFF << index);
4231 ivar |= (vector << index);
4232 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
4242 void __attribute__((cold))
4243 ixgbe_set_rx_function(struct rte_eth_dev *dev)
4245 uint16_t i, rx_using_sse;
4246 struct ixgbe_adapter *adapter =
4247 (struct ixgbe_adapter *)dev->data->dev_private;
4250 * In order to allow Vector Rx there are a few configuration
4251 * conditions to be met and Rx Bulk Allocation should be allowed.
4253 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
4254 !adapter->rx_bulk_alloc_allowed) {
4255 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
4256 "preconditions or RTE_IXGBE_INC_VECTOR is "
4258 dev->data->port_id);
4260 adapter->rx_vec_allowed = false;
4264 * Initialize the appropriate LRO callback.
4266 * If all queues satisfy the bulk allocation preconditions
4267 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
4268 * Otherwise use a single allocation version.
4270 if (dev->data->lro) {
4271 if (adapter->rx_bulk_alloc_allowed) {
4272 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a bulk "
4273 "allocation version");
4274 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4276 PMD_INIT_LOG(DEBUG, "LRO is requested. Using a single "
4277 "allocation version");
4278 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4280 } else if (dev->data->scattered_rx) {
4282 * Set the non-LRO scattered callback: there are Vector and
4283 * single allocation versions.
4285 if (adapter->rx_vec_allowed) {
4286 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
4287 "callback (port=%d).",
4288 dev->data->port_id);
4290 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
4291 } else if (adapter->rx_bulk_alloc_allowed) {
4292 PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
4293 "allocation callback (port=%d).",
4294 dev->data->port_id);
4295 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
4297 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
4298 "single allocation) "
4299 "Scattered Rx callback "
4301 dev->data->port_id);
4303 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
4306 * Below we set "simple" callbacks according to port/queues parameters.
4307 * If parameters allow we are going to choose between the following
4311 * - Single buffer allocation (the simplest one)
4313 } else if (adapter->rx_vec_allowed) {
4314 PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
4315 "burst size no less than %d (port=%d).",
4316 RTE_IXGBE_DESCS_PER_LOOP,
4317 dev->data->port_id);
4319 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
4320 } else if (adapter->rx_bulk_alloc_allowed) {
4321 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
4322 "satisfied. Rx Burst Bulk Alloc function "
4323 "will be used on port=%d.",
4324 dev->data->port_id);
4326 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
4328 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
4329 "satisfied, or Scattered Rx is requested "
4331 dev->data->port_id);
4333 dev->rx_pkt_burst = ixgbe_recv_pkts;
4336 /* Propagate information about RX function choice through all queues. */
4339 (dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec ||
4340 dev->rx_pkt_burst == ixgbe_recv_pkts_vec);
4342 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4343 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4345 rxq->rx_using_sse = rx_using_sse;
4350 * ixgbe_set_rsc - configure RSC related port HW registers
4352 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4353 * of 82599 Spec (x540 configuration is virtually the same).
4357 * Returns 0 in case of success or a non-zero error code
4360 ixgbe_set_rsc(struct rte_eth_dev *dev)
4362 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4363 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4364 struct rte_eth_dev_info dev_info = { 0 };
4365 bool rsc_capable = false;
4370 dev->dev_ops->dev_infos_get(dev, &dev_info);
4371 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4374 if (!rsc_capable && rx_conf->enable_lro) {
4375 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4380 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4382 if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
4384 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4385 * 3.0 RSC configuration requires HW CRC stripping being
4386 * enabled. If user requested both HW CRC stripping off
4387 * and RSC on - return an error.
4389 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4394 /* RFCTL configuration */
4396 uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4398 if (rx_conf->enable_lro)
4400 * Since NFS packets coalescing is not supported - clear
4401 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4404 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4405 IXGBE_RFCTL_NFSR_DIS);
4407 rfctl |= IXGBE_RFCTL_RSC_DIS;
4409 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4412 /* If LRO hasn't been requested - we are done here. */
4413 if (!rx_conf->enable_lro)
4416 /* Set RDRXCTL.RSCACKC bit */
4417 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4418 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4419 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4421 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4422 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4423 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4425 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4427 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4429 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4431 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4434 * ixgbe PMD doesn't support header-split at the moment.
4436 * Following the 4.6.7.2.1 chapter of the 82599/x540
4437 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4438 * should be configured even if header split is not
4439 * enabled. We will configure it 128 bytes following the
4440 * recommendation in the spec.
4442 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4443 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4444 IXGBE_SRRCTL_BSIZEHDR_MASK;
4447 * TODO: Consider setting the Receive Descriptor Minimum
4448 * Threshold Size for an RSC case. This is not an obviously
4449 * beneficiary option but the one worth considering...
4452 rscctl |= IXGBE_RSCCTL_RSCEN;
4453 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4454 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4457 * RSC: Set ITR interval corresponding to 2K ints/s.
4459 * Full-sized RSC aggregations for a 10Gb/s link will
4460 * arrive at about 20K aggregation/s rate.
4462 * 2K inst/s rate will make only 10% of the
4463 * aggregations to be closed due to the interrupt timer
4464 * expiration for a streaming at wire-speed case.
4466 * For a sparse streaming case this setting will yield
4467 * at most 500us latency for a single RSC aggregation.
4469 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4470 eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
4472 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4473 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4474 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4475 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4478 * RSC requires the mapping of the queue to the
4481 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4486 PMD_INIT_LOG(DEBUG, "enabling LRO mode");
4492 * Initializes Receive Unit.
4494 int __attribute__((cold))
4495 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4497 struct ixgbe_hw *hw;
4498 struct ixgbe_rx_queue *rxq;
4509 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4512 PMD_INIT_FUNC_TRACE();
4513 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4516 * Make sure receives are disabled while setting
4517 * up the RX context (registers, descriptor rings, etc.).
4519 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4520 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4522 /* Enable receipt of broadcasted frames */
4523 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4524 fctrl |= IXGBE_FCTRL_BAM;
4525 fctrl |= IXGBE_FCTRL_DPF;
4526 fctrl |= IXGBE_FCTRL_PMCF;
4527 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4530 * Configure CRC stripping, if any.
4532 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4533 if (rx_conf->hw_strip_crc)
4534 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4536 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4539 * Configure jumbo frame support, if any.
4541 if (rx_conf->jumbo_frame == 1) {
4542 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4543 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4544 maxfrs &= 0x0000FFFF;
4545 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4546 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4548 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4551 * If loopback mode is configured for 82599, set LPBK bit.
4553 if (hw->mac.type == ixgbe_mac_82599EB &&
4554 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4555 hlreg0 |= IXGBE_HLREG0_LPBK;
4557 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4559 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4561 /* Setup RX queues */
4562 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4563 rxq = dev->data->rx_queues[i];
4566 * Reset crc_len in case it was changed after queue setup by a
4567 * call to configure.
4569 rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
4571 /* Setup the Base and Length of the Rx Descriptor Rings */
4572 bus_addr = rxq->rx_ring_phys_addr;
4573 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4574 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4575 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4576 (uint32_t)(bus_addr >> 32));
4577 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4578 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4579 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4580 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4582 /* Configure the SRRCTL register */
4583 #ifdef RTE_HEADER_SPLIT_ENABLE
4585 * Configure Header Split
4587 if (rx_conf->header_split) {
4588 if (hw->mac.type == ixgbe_mac_82599EB) {
4589 /* Must setup the PSRTYPE register */
4592 psrtype = IXGBE_PSRTYPE_TCPHDR |
4593 IXGBE_PSRTYPE_UDPHDR |
4594 IXGBE_PSRTYPE_IPV4HDR |
4595 IXGBE_PSRTYPE_IPV6HDR;
4596 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4598 srrctl = ((rx_conf->split_hdr_size <<
4599 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4600 IXGBE_SRRCTL_BSIZEHDR_MASK);
4601 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4604 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4606 /* Set if packets are dropped when no descriptors available */
4608 srrctl |= IXGBE_SRRCTL_DROP_EN;
4611 * Configure the RX buffer size in the BSIZEPACKET field of
4612 * the SRRCTL register of the queue.
4613 * The value is in 1 KB resolution. Valid values can be from
4616 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4617 RTE_PKTMBUF_HEADROOM);
4618 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4619 IXGBE_SRRCTL_BSIZEPKT_MASK);
4621 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4623 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4624 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4626 /* It adds dual VLAN length for supporting dual VLAN */
4627 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4628 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4629 dev->data->scattered_rx = 1;
4632 if (rx_conf->enable_scatter)
4633 dev->data->scattered_rx = 1;
4636 * Device configured with multiple RX queues.
4638 ixgbe_dev_mq_rx_configure(dev);
4641 * Setup the Checksum Register.
4642 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4643 * Enable IP/L4 checkum computation by hardware if requested to do so.
4645 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4646 rxcsum |= IXGBE_RXCSUM_PCSD;
4647 if (rx_conf->hw_ip_checksum)
4648 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4650 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4652 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4654 if (hw->mac.type == ixgbe_mac_82599EB ||
4655 hw->mac.type == ixgbe_mac_X540) {
4656 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4657 if (rx_conf->hw_strip_crc)
4658 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4660 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4661 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4662 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4665 rc = ixgbe_set_rsc(dev);
4669 ixgbe_set_rx_function(dev);
4675 * Initializes Transmit Unit.
4677 void __attribute__((cold))
4678 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4680 struct ixgbe_hw *hw;
4681 struct ixgbe_tx_queue *txq;
4687 PMD_INIT_FUNC_TRACE();
4688 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4690 /* Enable TX CRC (checksum offload requirement) and hw padding
4693 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4694 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
4695 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4697 /* Setup the Base and Length of the Tx Descriptor Rings */
4698 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4699 txq = dev->data->tx_queues[i];
4701 bus_addr = txq->tx_ring_phys_addr;
4702 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
4703 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4704 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
4705 (uint32_t)(bus_addr >> 32));
4706 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
4707 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4708 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4709 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4710 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4713 * Disable Tx Head Writeback RO bit, since this hoses
4714 * bookkeeping if things aren't delivered in order.
4716 switch (hw->mac.type) {
4717 case ixgbe_mac_82598EB:
4718 txctrl = IXGBE_READ_REG(hw,
4719 IXGBE_DCA_TXCTRL(txq->reg_idx));
4720 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4721 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
4725 case ixgbe_mac_82599EB:
4726 case ixgbe_mac_X540:
4727 case ixgbe_mac_X550:
4728 case ixgbe_mac_X550EM_x:
4729 case ixgbe_mac_X550EM_a:
4731 txctrl = IXGBE_READ_REG(hw,
4732 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
4733 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4734 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
4740 /* Device configured with multiple TX queues. */
4741 ixgbe_dev_mq_tx_configure(dev);
4745 * Set up link for 82599 loopback mode Tx->Rx.
4747 static inline void __attribute__((cold))
4748 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
4750 PMD_INIT_FUNC_TRACE();
4752 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
4753 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
4755 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
4764 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
4765 ixgbe_reset_pipeline_82599(hw);
4767 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
4773 * Start Transmit and Receive Units.
4775 int __attribute__((cold))
4776 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
4778 struct ixgbe_hw *hw;
4779 struct ixgbe_tx_queue *txq;
4780 struct ixgbe_rx_queue *rxq;
4787 PMD_INIT_FUNC_TRACE();
4788 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4790 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4791 txq = dev->data->tx_queues[i];
4792 /* Setup Transmit Threshold Registers */
4793 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4794 txdctl |= txq->pthresh & 0x7F;
4795 txdctl |= ((txq->hthresh & 0x7F) << 8);
4796 txdctl |= ((txq->wthresh & 0x7F) << 16);
4797 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4800 if (hw->mac.type != ixgbe_mac_82598EB) {
4801 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
4802 dmatxctl |= IXGBE_DMATXCTL_TE;
4803 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
4806 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4807 txq = dev->data->tx_queues[i];
4808 if (!txq->tx_deferred_start) {
4809 ret = ixgbe_dev_tx_queue_start(dev, i);
4815 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4816 rxq = dev->data->rx_queues[i];
4817 if (!rxq->rx_deferred_start) {
4818 ret = ixgbe_dev_rx_queue_start(dev, i);
4824 /* Enable Receive engine */
4825 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4826 if (hw->mac.type == ixgbe_mac_82598EB)
4827 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4828 rxctrl |= IXGBE_RXCTRL_RXEN;
4829 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4831 /* If loopback mode is enabled for 82599, set up the link accordingly */
4832 if (hw->mac.type == ixgbe_mac_82599EB &&
4833 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4834 ixgbe_setup_loopback_link_82599(hw);
4840 * Start Receive Units for specified queue.
4842 int __attribute__((cold))
4843 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4845 struct ixgbe_hw *hw;
4846 struct ixgbe_rx_queue *rxq;
4850 PMD_INIT_FUNC_TRACE();
4851 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4853 if (rx_queue_id < dev->data->nb_rx_queues) {
4854 rxq = dev->data->rx_queues[rx_queue_id];
4856 /* Allocate buffers for descriptor rings */
4857 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
4858 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
4862 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4863 rxdctl |= IXGBE_RXDCTL_ENABLE;
4864 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4866 /* Wait until RX Enable ready */
4867 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4870 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4871 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4873 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
4876 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4877 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
4878 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
4886 * Stop Receive Units for specified queue.
4888 int __attribute__((cold))
4889 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4891 struct ixgbe_hw *hw;
4892 struct ixgbe_adapter *adapter =
4893 (struct ixgbe_adapter *)dev->data->dev_private;
4894 struct ixgbe_rx_queue *rxq;
4898 PMD_INIT_FUNC_TRACE();
4899 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4901 if (rx_queue_id < dev->data->nb_rx_queues) {
4902 rxq = dev->data->rx_queues[rx_queue_id];
4904 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4905 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4906 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4908 /* Wait until RX Enable bit clear */
4909 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4912 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4913 } while (--poll_ms && (rxdctl & IXGBE_RXDCTL_ENABLE));
4915 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
4918 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4920 ixgbe_rx_queue_release_mbufs(rxq);
4921 ixgbe_reset_rx_queue(adapter, rxq);
4922 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
4931 * Start Transmit Units for specified queue.
4933 int __attribute__((cold))
4934 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4936 struct ixgbe_hw *hw;
4937 struct ixgbe_tx_queue *txq;
4941 PMD_INIT_FUNC_TRACE();
4942 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4944 if (tx_queue_id < dev->data->nb_tx_queues) {
4945 txq = dev->data->tx_queues[tx_queue_id];
4946 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4947 txdctl |= IXGBE_TXDCTL_ENABLE;
4948 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4950 /* Wait until TX Enable ready */
4951 if (hw->mac.type == ixgbe_mac_82599EB) {
4952 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4955 txdctl = IXGBE_READ_REG(hw,
4956 IXGBE_TXDCTL(txq->reg_idx));
4957 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4959 PMD_INIT_LOG(ERR, "Could not enable "
4960 "Tx Queue %d", tx_queue_id);
4963 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4964 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4965 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
4973 * Stop Transmit Units for specified queue.
4975 int __attribute__((cold))
4976 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4978 struct ixgbe_hw *hw;
4979 struct ixgbe_tx_queue *txq;
4981 uint32_t txtdh, txtdt;
4984 PMD_INIT_FUNC_TRACE();
4985 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4987 if (tx_queue_id >= dev->data->nb_tx_queues)
4990 txq = dev->data->tx_queues[tx_queue_id];
4992 /* Wait until TX queue is empty */
4993 if (hw->mac.type == ixgbe_mac_82599EB) {
4994 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4996 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4997 txtdh = IXGBE_READ_REG(hw,
4998 IXGBE_TDH(txq->reg_idx));
4999 txtdt = IXGBE_READ_REG(hw,
5000 IXGBE_TDT(txq->reg_idx));
5001 } while (--poll_ms && (txtdh != txtdt));
5003 PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
5004 "when stopping.", tx_queue_id);
5007 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
5008 txdctl &= ~IXGBE_TXDCTL_ENABLE;
5009 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
5011 /* Wait until TX Enable bit clear */
5012 if (hw->mac.type == ixgbe_mac_82599EB) {
5013 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
5016 txdctl = IXGBE_READ_REG(hw,
5017 IXGBE_TXDCTL(txq->reg_idx));
5018 } while (--poll_ms && (txdctl & IXGBE_TXDCTL_ENABLE));
5020 PMD_INIT_LOG(ERR, "Could not disable "
5021 "Tx Queue %d", tx_queue_id);
5024 if (txq->ops != NULL) {
5025 txq->ops->release_mbufs(txq);
5026 txq->ops->reset(txq);
5028 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
5034 ixgbe_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5035 struct rte_eth_rxq_info *qinfo)
5037 struct ixgbe_rx_queue *rxq;
5039 rxq = dev->data->rx_queues[queue_id];
5041 qinfo->mp = rxq->mb_pool;
5042 qinfo->scattered_rx = dev->data->scattered_rx;
5043 qinfo->nb_desc = rxq->nb_rx_desc;
5045 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
5046 qinfo->conf.rx_drop_en = rxq->drop_en;
5047 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
5051 ixgbe_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
5052 struct rte_eth_txq_info *qinfo)
5054 struct ixgbe_tx_queue *txq;
5056 txq = dev->data->tx_queues[queue_id];
5058 qinfo->nb_desc = txq->nb_tx_desc;
5060 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
5061 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
5062 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
5064 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
5065 qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
5066 qinfo->conf.txq_flags = txq->txq_flags;
5067 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
5071 * [VF] Initializes Receive Unit.
5073 int __attribute__((cold))
5074 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
5076 struct ixgbe_hw *hw;
5077 struct ixgbe_rx_queue *rxq;
5079 uint32_t srrctl, psrtype = 0;
5084 PMD_INIT_FUNC_TRACE();
5085 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5087 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
5088 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5089 "it should be power of 2");
5093 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
5094 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
5095 "it should be equal to or less than %d",
5096 hw->mac.max_rx_queues);
5101 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
5102 * disables the VF receipt of packets if the PF MTU is > 1500.
5103 * This is done to deal with 82599 limitations that imposes
5104 * the PF and all VFs to share the same MTU.
5105 * Then, the PF driver enables again the VF receipt of packet when
5106 * the VF driver issues a IXGBE_VF_SET_LPE request.
5107 * In the meantime, the VF device cannot be used, even if the VF driver
5108 * and the Guest VM network stack are ready to accept packets with a
5109 * size up to the PF MTU.
5110 * As a work-around to this PF behaviour, force the call to
5111 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
5112 * VF packets received can work in all cases.
5114 ixgbevf_rlpml_set_vf(hw,
5115 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
5117 /* Setup RX queues */
5118 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5119 rxq = dev->data->rx_queues[i];
5121 /* Allocate buffers for descriptor rings */
5122 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
5126 /* Setup the Base and Length of the Rx Descriptor Rings */
5127 bus_addr = rxq->rx_ring_phys_addr;
5129 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
5130 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5131 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
5132 (uint32_t)(bus_addr >> 32));
5133 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
5134 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
5135 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
5136 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
5139 /* Configure the SRRCTL register */
5140 #ifdef RTE_HEADER_SPLIT_ENABLE
5142 * Configure Header Split
5144 if (dev->data->dev_conf.rxmode.header_split) {
5145 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
5146 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
5147 IXGBE_SRRCTL_BSIZEHDR_MASK);
5148 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
5151 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
5153 /* Set if packets are dropped when no descriptors available */
5155 srrctl |= IXGBE_SRRCTL_DROP_EN;
5158 * Configure the RX buffer size in the BSIZEPACKET field of
5159 * the SRRCTL register of the queue.
5160 * The value is in 1 KB resolution. Valid values can be from
5163 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
5164 RTE_PKTMBUF_HEADROOM);
5165 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
5166 IXGBE_SRRCTL_BSIZEPKT_MASK);
5169 * VF modification to write virtual function SRRCTL register
5171 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
5173 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
5174 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
5176 if (dev->data->dev_conf.rxmode.enable_scatter ||
5177 /* It adds dual VLAN length for supporting dual VLAN */
5178 (dev->data->dev_conf.rxmode.max_rx_pkt_len +
5179 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
5180 if (!dev->data->scattered_rx)
5181 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
5182 dev->data->scattered_rx = 1;
5186 #ifdef RTE_HEADER_SPLIT_ENABLE
5187 if (dev->data->dev_conf.rxmode.header_split)
5188 /* Must setup the PSRTYPE register */
5189 psrtype = IXGBE_PSRTYPE_TCPHDR |
5190 IXGBE_PSRTYPE_UDPHDR |
5191 IXGBE_PSRTYPE_IPV4HDR |
5192 IXGBE_PSRTYPE_IPV6HDR;
5195 /* Set RQPL for VF RSS according to max Rx queue */
5196 psrtype |= (dev->data->nb_rx_queues >> 1) <<
5197 IXGBE_PSRTYPE_RQPL_SHIFT;
5198 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
5200 ixgbe_set_rx_function(dev);
5206 * [VF] Initializes Transmit Unit.
5208 void __attribute__((cold))
5209 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
5211 struct ixgbe_hw *hw;
5212 struct ixgbe_tx_queue *txq;
5217 PMD_INIT_FUNC_TRACE();
5218 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5220 /* Setup the Base and Length of the Tx Descriptor Rings */
5221 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5222 txq = dev->data->tx_queues[i];
5223 bus_addr = txq->tx_ring_phys_addr;
5224 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
5225 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
5226 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
5227 (uint32_t)(bus_addr >> 32));
5228 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
5229 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
5230 /* Setup the HW Tx Head and TX Tail descriptor pointers */
5231 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
5232 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
5235 * Disable Tx Head Writeback RO bit, since this hoses
5236 * bookkeeping if things aren't delivered in order.
5238 txctrl = IXGBE_READ_REG(hw,
5239 IXGBE_VFDCA_TXCTRL(i));
5240 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
5241 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
5247 * [VF] Start Transmit and Receive Units.
5249 void __attribute__((cold))
5250 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
5252 struct ixgbe_hw *hw;
5253 struct ixgbe_tx_queue *txq;
5254 struct ixgbe_rx_queue *rxq;
5260 PMD_INIT_FUNC_TRACE();
5261 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
5263 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5264 txq = dev->data->tx_queues[i];
5265 /* Setup Transmit Threshold Registers */
5266 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5267 txdctl |= txq->pthresh & 0x7F;
5268 txdctl |= ((txq->hthresh & 0x7F) << 8);
5269 txdctl |= ((txq->wthresh & 0x7F) << 16);
5270 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5273 for (i = 0; i < dev->data->nb_tx_queues; i++) {
5275 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5276 txdctl |= IXGBE_TXDCTL_ENABLE;
5277 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
5280 /* Wait until TX Enable ready */
5283 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
5284 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
5286 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
5288 for (i = 0; i < dev->data->nb_rx_queues; i++) {
5290 rxq = dev->data->rx_queues[i];
5292 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5293 rxdctl |= IXGBE_RXDCTL_ENABLE;
5294 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
5296 /* Wait until RX Enable ready */
5300 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
5301 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
5303 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
5305 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
5310 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
5311 int __attribute__((weak))
5312 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
5317 uint16_t __attribute__((weak))
5318 ixgbe_recv_pkts_vec(
5319 void __rte_unused *rx_queue,
5320 struct rte_mbuf __rte_unused **rx_pkts,
5321 uint16_t __rte_unused nb_pkts)
5326 uint16_t __attribute__((weak))
5327 ixgbe_recv_scattered_pkts_vec(
5328 void __rte_unused *rx_queue,
5329 struct rte_mbuf __rte_unused **rx_pkts,
5330 uint16_t __rte_unused nb_pkts)
5335 int __attribute__((weak))
5336 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)