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>
62 #include <rte_mempool.h>
63 #include <rte_malloc.h>
65 #include <rte_ether.h>
66 #include <rte_ethdev.h>
67 #include <rte_prefetch.h>
71 #include <rte_string_fns.h>
72 #include <rte_errno.h>
75 #include "ixgbe_logs.h"
76 #include "base/ixgbe_api.h"
77 #include "base/ixgbe_vf.h"
78 #include "ixgbe_ethdev.h"
79 #include "base/ixgbe_dcb.h"
80 #include "base/ixgbe_common.h"
81 #include "ixgbe_rxtx.h"
83 /* Bit Mask to indicate what bits required for building TX context */
84 #define IXGBE_TX_OFFLOAD_MASK ( \
90 static inline struct rte_mbuf *
91 rte_rxmbuf_alloc(struct rte_mempool *mp)
95 m = __rte_mbuf_raw_alloc(mp);
96 __rte_mbuf_sanity_check_raw(m, 0);
102 #define RTE_PMD_USE_PREFETCH
105 #ifdef RTE_PMD_USE_PREFETCH
107 * Prefetch a cache line into all cache levels.
109 #define rte_ixgbe_prefetch(p) rte_prefetch0(p)
111 #define rte_ixgbe_prefetch(p) do {} while(0)
114 /*********************************************************************
118 **********************************************************************/
121 * Check for descriptors with their DD bit set and free mbufs.
122 * Return the total number of buffers freed.
124 static inline int __attribute__((always_inline))
125 ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
127 struct ixgbe_tx_entry *txep;
131 /* check DD bit on threshold descriptor */
132 status = txq->tx_ring[txq->tx_next_dd].wb.status;
133 if (! (status & IXGBE_ADVTXD_STAT_DD))
137 * first buffer to free from S/W ring is at index
138 * tx_next_dd - (tx_rs_thresh-1)
140 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
142 /* free buffers one at a time */
143 if ((txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT) != 0) {
144 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
145 txep->mbuf->next = NULL;
146 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
150 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
151 rte_pktmbuf_free_seg(txep->mbuf);
156 /* buffers were freed, update counters */
157 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
158 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
159 if (txq->tx_next_dd >= txq->nb_tx_desc)
160 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
162 return txq->tx_rs_thresh;
165 /* Populate 4 descriptors with data from 4 mbufs */
167 tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
169 uint64_t buf_dma_addr;
173 for (i = 0; i < 4; ++i, ++txdp, ++pkts) {
174 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
175 pkt_len = (*pkts)->data_len;
177 /* write data to descriptor */
178 txdp->read.buffer_addr = buf_dma_addr;
179 txdp->read.cmd_type_len =
180 ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
181 txdp->read.olinfo_status =
182 (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 = buf_dma_addr;
199 txdp->read.cmd_type_len =
200 ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
201 txdp->read.olinfo_status =
202 (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 */
343 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_TX_MAX_BURST);
344 ret = tx_xmit_pkts(tx_queue, &(tx_pkts[nb_tx]), n);
345 nb_tx = (uint16_t)(nb_tx + ret);
346 nb_pkts = (uint16_t)(nb_pkts - ret);
355 ixgbe_set_xmit_ctx(struct ixgbe_tx_queue *txq,
356 volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
357 uint64_t ol_flags, union ixgbe_tx_offload tx_offload)
359 uint32_t type_tucmd_mlhl;
360 uint32_t mss_l4len_idx = 0;
362 uint32_t vlan_macip_lens;
363 union ixgbe_tx_offload tx_offload_mask;
365 ctx_idx = txq->ctx_curr;
366 tx_offload_mask.data = 0;
369 /* Specify which HW CTX to upload. */
370 mss_l4len_idx |= (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
372 if (ol_flags & PKT_TX_VLAN_PKT) {
373 tx_offload_mask.vlan_tci |= ~0;
376 /* check if TCP segmentation required for this packet */
377 if (ol_flags & PKT_TX_TCP_SEG) {
378 /* implies IP cksum in IPv4 */
379 if (ol_flags & PKT_TX_IP_CKSUM)
380 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4 |
381 IXGBE_ADVTXD_TUCMD_L4T_TCP |
382 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
384 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV6 |
385 IXGBE_ADVTXD_TUCMD_L4T_TCP |
386 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
388 tx_offload_mask.l2_len |= ~0;
389 tx_offload_mask.l3_len |= ~0;
390 tx_offload_mask.l4_len |= ~0;
391 tx_offload_mask.tso_segsz |= ~0;
392 mss_l4len_idx |= tx_offload.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT;
393 mss_l4len_idx |= tx_offload.l4_len << IXGBE_ADVTXD_L4LEN_SHIFT;
394 } else { /* no TSO, check if hardware checksum is needed */
395 if (ol_flags & PKT_TX_IP_CKSUM) {
396 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
397 tx_offload_mask.l2_len |= ~0;
398 tx_offload_mask.l3_len |= ~0;
401 switch (ol_flags & PKT_TX_L4_MASK) {
402 case PKT_TX_UDP_CKSUM:
403 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
404 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
405 mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
406 tx_offload_mask.l2_len |= ~0;
407 tx_offload_mask.l3_len |= ~0;
409 case PKT_TX_TCP_CKSUM:
410 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
411 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
412 mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
413 tx_offload_mask.l2_len |= ~0;
414 tx_offload_mask.l3_len |= ~0;
415 tx_offload_mask.l4_len |= ~0;
417 case PKT_TX_SCTP_CKSUM:
418 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
419 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
420 mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
421 tx_offload_mask.l2_len |= ~0;
422 tx_offload_mask.l3_len |= ~0;
425 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
426 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
431 txq->ctx_cache[ctx_idx].flags = ol_flags;
432 txq->ctx_cache[ctx_idx].tx_offload.data =
433 tx_offload_mask.data & tx_offload.data;
434 txq->ctx_cache[ctx_idx].tx_offload_mask = tx_offload_mask;
436 ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
437 vlan_macip_lens = tx_offload.l3_len;
438 vlan_macip_lens |= (tx_offload.l2_len << IXGBE_ADVTXD_MACLEN_SHIFT);
439 vlan_macip_lens |= ((uint32_t)tx_offload.vlan_tci << IXGBE_ADVTXD_VLAN_SHIFT);
440 ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
441 ctx_txd->mss_l4len_idx = rte_cpu_to_le_32(mss_l4len_idx);
442 ctx_txd->seqnum_seed = 0;
446 * Check which hardware context can be used. Use the existing match
447 * or create a new context descriptor.
449 static inline uint32_t
450 what_advctx_update(struct ixgbe_tx_queue *txq, uint64_t flags,
451 union ixgbe_tx_offload tx_offload)
453 /* If match with the current used context */
454 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
455 (txq->ctx_cache[txq->ctx_curr].tx_offload.data ==
456 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data & tx_offload.data)))) {
457 return txq->ctx_curr;
460 /* What if match with the next context */
462 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
463 (txq->ctx_cache[txq->ctx_curr].tx_offload.data ==
464 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data & tx_offload.data)))) {
465 return txq->ctx_curr;
468 /* Mismatch, use the previous context */
469 return (IXGBE_CTX_NUM);
472 static inline uint32_t
473 tx_desc_cksum_flags_to_olinfo(uint64_t ol_flags)
476 if ((ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM)
477 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
478 if (ol_flags & PKT_TX_IP_CKSUM)
479 tmp |= IXGBE_ADVTXD_POPTS_IXSM;
480 if (ol_flags & PKT_TX_TCP_SEG)
481 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
485 static inline uint32_t
486 tx_desc_ol_flags_to_cmdtype(uint64_t ol_flags)
488 uint32_t cmdtype = 0;
489 if (ol_flags & PKT_TX_VLAN_PKT)
490 cmdtype |= IXGBE_ADVTXD_DCMD_VLE;
491 if (ol_flags & PKT_TX_TCP_SEG)
492 cmdtype |= IXGBE_ADVTXD_DCMD_TSE;
496 /* Default RS bit threshold values */
497 #ifndef DEFAULT_TX_RS_THRESH
498 #define DEFAULT_TX_RS_THRESH 32
500 #ifndef DEFAULT_TX_FREE_THRESH
501 #define DEFAULT_TX_FREE_THRESH 32
504 /* Reset transmit descriptors after they have been used */
506 ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
508 struct ixgbe_tx_entry *sw_ring = txq->sw_ring;
509 volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
510 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
511 uint16_t nb_tx_desc = txq->nb_tx_desc;
512 uint16_t desc_to_clean_to;
513 uint16_t nb_tx_to_clean;
515 /* Determine the last descriptor needing to be cleaned */
516 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
517 if (desc_to_clean_to >= nb_tx_desc)
518 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
520 /* Check to make sure the last descriptor to clean is done */
521 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
522 if (! (txr[desc_to_clean_to].wb.status & IXGBE_TXD_STAT_DD))
524 PMD_TX_FREE_LOG(DEBUG,
525 "TX descriptor %4u is not done"
526 "(port=%d queue=%d)",
528 txq->port_id, txq->queue_id);
529 /* Failed to clean any descriptors, better luck next time */
533 /* Figure out how many descriptors will be cleaned */
534 if (last_desc_cleaned > desc_to_clean_to)
535 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
538 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
541 PMD_TX_FREE_LOG(DEBUG,
542 "Cleaning %4u TX descriptors: %4u to %4u "
543 "(port=%d queue=%d)",
544 nb_tx_to_clean, last_desc_cleaned, desc_to_clean_to,
545 txq->port_id, txq->queue_id);
548 * The last descriptor to clean is done, so that means all the
549 * descriptors from the last descriptor that was cleaned
550 * up to the last descriptor with the RS bit set
551 * are done. Only reset the threshold descriptor.
553 txr[desc_to_clean_to].wb.status = 0;
555 /* Update the txq to reflect the last descriptor that was cleaned */
556 txq->last_desc_cleaned = desc_to_clean_to;
557 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
564 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
567 struct ixgbe_tx_queue *txq;
568 struct ixgbe_tx_entry *sw_ring;
569 struct ixgbe_tx_entry *txe, *txn;
570 volatile union ixgbe_adv_tx_desc *txr;
571 volatile union ixgbe_adv_tx_desc *txd;
572 struct rte_mbuf *tx_pkt;
573 struct rte_mbuf *m_seg;
574 uint64_t buf_dma_addr;
575 uint32_t olinfo_status;
576 uint32_t cmd_type_len;
587 union ixgbe_tx_offload tx_offload = {0};
590 sw_ring = txq->sw_ring;
592 tx_id = txq->tx_tail;
593 txe = &sw_ring[tx_id];
595 /* Determine if the descriptor ring needs to be cleaned. */
596 if (txq->nb_tx_free < txq->tx_free_thresh)
597 ixgbe_xmit_cleanup(txq);
599 rte_prefetch0(&txe->mbuf->pool);
602 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
605 pkt_len = tx_pkt->pkt_len;
608 * Determine how many (if any) context descriptors
609 * are needed for offload functionality.
611 ol_flags = tx_pkt->ol_flags;
613 /* If hardware offload required */
614 tx_ol_req = ol_flags & IXGBE_TX_OFFLOAD_MASK;
616 tx_offload.l2_len = tx_pkt->l2_len;
617 tx_offload.l3_len = tx_pkt->l3_len;
618 tx_offload.l4_len = tx_pkt->l4_len;
619 tx_offload.vlan_tci = tx_pkt->vlan_tci;
620 tx_offload.tso_segsz = tx_pkt->tso_segsz;
622 /* If new context need be built or reuse the exist ctx. */
623 ctx = what_advctx_update(txq, tx_ol_req,
625 /* Only allocate context descriptor if required*/
626 new_ctx = (ctx == IXGBE_CTX_NUM);
631 * Keep track of how many descriptors are used this loop
632 * This will always be the number of segments + the number of
633 * Context descriptors required to transmit the packet
635 nb_used = (uint16_t)(tx_pkt->nb_segs + new_ctx);
638 * The number of descriptors that must be allocated for a
639 * packet is the number of segments of that packet, plus 1
640 * Context Descriptor for the hardware offload, if any.
641 * Determine the last TX descriptor to allocate in the TX ring
642 * for the packet, starting from the current position (tx_id)
645 tx_last = (uint16_t) (tx_id + nb_used - 1);
648 if (tx_last >= txq->nb_tx_desc)
649 tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
651 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
652 " tx_first=%u tx_last=%u",
653 (unsigned) txq->port_id,
654 (unsigned) txq->queue_id,
660 * Make sure there are enough TX descriptors available to
661 * transmit the entire packet.
662 * nb_used better be less than or equal to txq->tx_rs_thresh
664 if (nb_used > txq->nb_tx_free) {
665 PMD_TX_FREE_LOG(DEBUG,
666 "Not enough free TX descriptors "
667 "nb_used=%4u nb_free=%4u "
668 "(port=%d queue=%d)",
669 nb_used, txq->nb_tx_free,
670 txq->port_id, txq->queue_id);
672 if (ixgbe_xmit_cleanup(txq) != 0) {
673 /* Could not clean any descriptors */
679 /* nb_used better be <= txq->tx_rs_thresh */
680 if (unlikely(nb_used > txq->tx_rs_thresh)) {
681 PMD_TX_FREE_LOG(DEBUG,
682 "The number of descriptors needed to "
683 "transmit the packet exceeds the "
684 "RS bit threshold. This will impact "
686 "nb_used=%4u nb_free=%4u "
688 "(port=%d queue=%d)",
689 nb_used, txq->nb_tx_free,
691 txq->port_id, txq->queue_id);
693 * Loop here until there are enough TX
694 * descriptors or until the ring cannot be
697 while (nb_used > txq->nb_tx_free) {
698 if (ixgbe_xmit_cleanup(txq) != 0) {
700 * Could not clean any
712 * By now there are enough free TX descriptors to transmit
717 * Set common flags of all TX Data Descriptors.
719 * The following bits must be set in all Data Descriptors:
720 * - IXGBE_ADVTXD_DTYP_DATA
721 * - IXGBE_ADVTXD_DCMD_DEXT
723 * The following bits must be set in the first Data Descriptor
724 * and are ignored in the other ones:
725 * - IXGBE_ADVTXD_DCMD_IFCS
726 * - IXGBE_ADVTXD_MAC_1588
727 * - IXGBE_ADVTXD_DCMD_VLE
729 * The following bits must only be set in the last Data
731 * - IXGBE_TXD_CMD_EOP
733 * The following bits can be set in any Data Descriptor, but
734 * are only set in the last Data Descriptor:
737 cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
738 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
740 #ifdef RTE_LIBRTE_IEEE1588
741 if (ol_flags & PKT_TX_IEEE1588_TMST)
742 cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
748 if (ol_flags & PKT_TX_TCP_SEG) {
749 /* when TSO is on, paylen in descriptor is the
750 * not the packet len but the tcp payload len */
751 pkt_len -= (tx_offload.l2_len +
752 tx_offload.l3_len + tx_offload.l4_len);
756 * Setup the TX Advanced Context Descriptor if required
759 volatile struct ixgbe_adv_tx_context_desc *
762 ctx_txd = (volatile struct
763 ixgbe_adv_tx_context_desc *)
766 txn = &sw_ring[txe->next_id];
767 rte_prefetch0(&txn->mbuf->pool);
769 if (txe->mbuf != NULL) {
770 rte_pktmbuf_free_seg(txe->mbuf);
774 ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
777 txe->last_id = tx_last;
778 tx_id = txe->next_id;
783 * Setup the TX Advanced Data Descriptor,
784 * This path will go through
785 * whatever new/reuse the context descriptor
787 cmd_type_len |= tx_desc_ol_flags_to_cmdtype(ol_flags);
788 olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
789 olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
792 olinfo_status |= (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
797 txn = &sw_ring[txe->next_id];
798 rte_prefetch0(&txn->mbuf->pool);
800 if (txe->mbuf != NULL)
801 rte_pktmbuf_free_seg(txe->mbuf);
805 * Set up Transmit Data Descriptor.
807 slen = m_seg->data_len;
808 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
809 txd->read.buffer_addr =
810 rte_cpu_to_le_64(buf_dma_addr);
811 txd->read.cmd_type_len =
812 rte_cpu_to_le_32(cmd_type_len | slen);
813 txd->read.olinfo_status =
814 rte_cpu_to_le_32(olinfo_status);
815 txe->last_id = tx_last;
816 tx_id = txe->next_id;
819 } while (m_seg != NULL);
822 * The last packet data descriptor needs End Of Packet (EOP)
824 cmd_type_len |= IXGBE_TXD_CMD_EOP;
825 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
826 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
828 /* Set RS bit only on threshold packets' last descriptor */
829 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
830 PMD_TX_FREE_LOG(DEBUG,
831 "Setting RS bit on TXD id="
832 "%4u (port=%d queue=%d)",
833 tx_last, txq->port_id, txq->queue_id);
835 cmd_type_len |= IXGBE_TXD_CMD_RS;
837 /* Update txq RS bit counters */
840 txd->read.cmd_type_len |= rte_cpu_to_le_32(cmd_type_len);
846 * Set the Transmit Descriptor Tail (TDT)
848 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
849 (unsigned) txq->port_id, (unsigned) txq->queue_id,
850 (unsigned) tx_id, (unsigned) nb_tx);
851 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, tx_id);
852 txq->tx_tail = tx_id;
857 /*********************************************************************
861 **********************************************************************/
863 #define IXGBE_PACKET_TYPE_IPV4 0X01
864 #define IXGBE_PACKET_TYPE_IPV4_TCP 0X11
865 #define IXGBE_PACKET_TYPE_IPV4_UDP 0X21
866 #define IXGBE_PACKET_TYPE_IPV4_SCTP 0X41
867 #define IXGBE_PACKET_TYPE_IPV4_EXT 0X03
868 #define IXGBE_PACKET_TYPE_IPV4_EXT_SCTP 0X43
869 #define IXGBE_PACKET_TYPE_IPV6 0X04
870 #define IXGBE_PACKET_TYPE_IPV6_TCP 0X14
871 #define IXGBE_PACKET_TYPE_IPV6_UDP 0X24
872 #define IXGBE_PACKET_TYPE_IPV6_EXT 0X0C
873 #define IXGBE_PACKET_TYPE_IPV6_EXT_TCP 0X1C
874 #define IXGBE_PACKET_TYPE_IPV6_EXT_UDP 0X2C
875 #define IXGBE_PACKET_TYPE_IPV4_IPV6 0X05
876 #define IXGBE_PACKET_TYPE_IPV4_IPV6_TCP 0X15
877 #define IXGBE_PACKET_TYPE_IPV4_IPV6_UDP 0X25
878 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT 0X0D
879 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP 0X1D
880 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP 0X2D
881 #define IXGBE_PACKET_TYPE_MAX 0X80
882 #define IXGBE_PACKET_TYPE_MASK 0X7F
883 #define IXGBE_PACKET_TYPE_SHIFT 0X04
884 static inline uint32_t
885 ixgbe_rxd_pkt_info_to_pkt_type(uint16_t pkt_info)
887 static const uint32_t
888 ptype_table[IXGBE_PACKET_TYPE_MAX] __rte_cache_aligned = {
889 [IXGBE_PACKET_TYPE_IPV4] = RTE_PTYPE_L2_ETHER |
891 [IXGBE_PACKET_TYPE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
892 RTE_PTYPE_L3_IPV4_EXT,
893 [IXGBE_PACKET_TYPE_IPV6] = RTE_PTYPE_L2_ETHER |
895 [IXGBE_PACKET_TYPE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
896 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
897 RTE_PTYPE_INNER_L3_IPV6,
898 [IXGBE_PACKET_TYPE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
899 RTE_PTYPE_L3_IPV6_EXT,
900 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
901 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
902 RTE_PTYPE_INNER_L3_IPV6_EXT,
903 [IXGBE_PACKET_TYPE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
904 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
905 [IXGBE_PACKET_TYPE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
906 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
907 [IXGBE_PACKET_TYPE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
908 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
909 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
910 [IXGBE_PACKET_TYPE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
911 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP,
912 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
913 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
914 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
915 [IXGBE_PACKET_TYPE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
916 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
917 [IXGBE_PACKET_TYPE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
918 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
919 [IXGBE_PACKET_TYPE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
920 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
921 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
922 [IXGBE_PACKET_TYPE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
923 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP,
924 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
925 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
926 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
927 [IXGBE_PACKET_TYPE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
928 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP,
929 [IXGBE_PACKET_TYPE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
930 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_SCTP,
932 if (unlikely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
933 return RTE_PTYPE_UNKNOWN;
935 pkt_info = (pkt_info >> IXGBE_PACKET_TYPE_SHIFT) &
936 IXGBE_PACKET_TYPE_MASK;
938 return ptype_table[pkt_info];
941 static inline uint64_t
942 ixgbe_rxd_pkt_info_to_pkt_flags(uint16_t pkt_info)
944 static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
945 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
946 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
947 PKT_RX_RSS_HASH, 0, 0, 0,
948 0, 0, 0, PKT_RX_FDIR,
950 #ifdef RTE_LIBRTE_IEEE1588
951 static uint64_t ip_pkt_etqf_map[8] = {
952 0, 0, 0, PKT_RX_IEEE1588_PTP,
956 if (likely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
957 return ip_pkt_etqf_map[(pkt_info >> 4) & 0X07] |
958 ip_rss_types_map[pkt_info & 0XF];
960 return ip_rss_types_map[pkt_info & 0XF];
962 return ip_rss_types_map[pkt_info & 0XF];
965 #else /* RTE_NEXT_ABI */
966 static inline uint64_t
967 rx_desc_hlen_type_rss_to_pkt_flags(uint32_t hl_tp_rs)
971 static const uint64_t ip_pkt_types_map[16] = {
972 0, PKT_RX_IPV4_HDR, PKT_RX_IPV4_HDR_EXT, PKT_RX_IPV4_HDR_EXT,
973 PKT_RX_IPV6_HDR, 0, 0, 0,
974 PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
975 PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
978 static const uint64_t ip_rss_types_map[16] = {
979 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
980 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
981 PKT_RX_RSS_HASH, 0, 0, 0,
982 0, 0, 0, PKT_RX_FDIR,
985 #ifdef RTE_LIBRTE_IEEE1588
986 static uint64_t ip_pkt_etqf_map[8] = {
987 0, 0, 0, PKT_RX_IEEE1588_PTP,
991 pkt_flags = (hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ?
992 ip_pkt_etqf_map[(hl_tp_rs >> 4) & 0x07] :
993 ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
995 pkt_flags = (hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ? 0 :
996 ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
999 return pkt_flags | ip_rss_types_map[hl_tp_rs & 0xF];
1001 #endif /* RTE_NEXT_ABI */
1003 static inline uint64_t
1004 rx_desc_status_to_pkt_flags(uint32_t rx_status)
1009 * Check if VLAN present only.
1010 * Do not check whether L3/L4 rx checksum done by NIC or not,
1011 * That can be found from rte_eth_rxmode.hw_ip_checksum flag
1013 pkt_flags = (rx_status & IXGBE_RXD_STAT_VP) ? PKT_RX_VLAN_PKT : 0;
1015 #ifdef RTE_LIBRTE_IEEE1588
1016 if (rx_status & IXGBE_RXD_STAT_TMST)
1017 pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
1022 static inline uint64_t
1023 rx_desc_error_to_pkt_flags(uint32_t rx_status)
1026 * Bit 31: IPE, IPv4 checksum error
1027 * Bit 30: L4I, L4I integrity error
1029 static uint64_t error_to_pkt_flags_map[4] = {
1030 0, PKT_RX_L4_CKSUM_BAD, PKT_RX_IP_CKSUM_BAD,
1031 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1033 return error_to_pkt_flags_map[(rx_status >>
1034 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1038 * LOOK_AHEAD defines how many desc statuses to check beyond the
1039 * current descriptor.
1040 * It must be a pound define for optimal performance.
1041 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1042 * function only works with LOOK_AHEAD=8.
1044 #define LOOK_AHEAD 8
1045 #if (LOOK_AHEAD != 8)
1046 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1049 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1051 volatile union ixgbe_adv_rx_desc *rxdp;
1052 struct ixgbe_rx_entry *rxep;
1053 struct rte_mbuf *mb;
1058 uint32_t s[LOOK_AHEAD];
1059 uint16_t pkt_info[LOOK_AHEAD];
1061 int s[LOOK_AHEAD], nb_dd;
1062 #endif /* RTE_NEXT_ABI */
1063 int i, j, nb_rx = 0;
1066 /* get references to current descriptor and S/W ring entry */
1067 rxdp = &rxq->rx_ring[rxq->rx_tail];
1068 rxep = &rxq->sw_ring[rxq->rx_tail];
1070 /* check to make sure there is at least 1 packet to receive */
1071 if (! (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD))
1075 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1076 * reference packets that are ready to be received.
1078 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1079 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD)
1081 /* Read desc statuses backwards to avoid race condition */
1082 for (j = LOOK_AHEAD-1; j >= 0; --j)
1083 s[j] = rxdp[j].wb.upper.status_error;
1086 for (j = LOOK_AHEAD - 1; j >= 0; --j)
1087 pkt_info[j] = rxdp[j].wb.lower.lo_dword.
1089 #endif /* RTE_NEXT_ABI */
1091 /* Compute how many status bits were set */
1093 for (j = 0; j < LOOK_AHEAD; ++j)
1094 nb_dd += s[j] & IXGBE_RXDADV_STAT_DD;
1098 /* Translate descriptor info to mbuf format */
1099 for (j = 0; j < nb_dd; ++j) {
1101 pkt_len = (uint16_t)(rxdp[j].wb.upper.length - rxq->crc_len);
1102 mb->data_len = pkt_len;
1103 mb->pkt_len = pkt_len;
1104 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1106 /* convert descriptor fields to rte mbuf flags */
1108 pkt_flags = rx_desc_status_to_pkt_flags(s[j]);
1109 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1111 ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info[j]);
1112 mb->ol_flags = pkt_flags;
1114 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info[j]);
1115 #else /* RTE_NEXT_ABI */
1116 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(
1117 rxdp[j].wb.lower.lo_dword.data);
1118 /* reuse status field from scan list */
1119 pkt_flags |= rx_desc_status_to_pkt_flags(s[j]);
1120 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1121 mb->ol_flags = pkt_flags;
1122 #endif /* RTE_NEXT_ABI */
1124 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1125 mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
1126 else if (pkt_flags & PKT_RX_FDIR) {
1127 mb->hash.fdir.hash =
1128 (uint16_t)((rxdp[j].wb.lower.hi_dword.csum_ip.csum)
1129 & IXGBE_ATR_HASH_MASK);
1130 mb->hash.fdir.id = rxdp[j].wb.lower.hi_dword.csum_ip.ip_id;
1134 /* Move mbuf pointers from the S/W ring to the stage */
1135 for (j = 0; j < LOOK_AHEAD; ++j) {
1136 rxq->rx_stage[i + j] = rxep[j].mbuf;
1139 /* stop if all requested packets could not be received */
1140 if (nb_dd != LOOK_AHEAD)
1144 /* clear software ring entries so we can cleanup correctly */
1145 for (i = 0; i < nb_rx; ++i) {
1146 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1154 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1156 volatile union ixgbe_adv_rx_desc *rxdp;
1157 struct ixgbe_rx_entry *rxep;
1158 struct rte_mbuf *mb;
1163 /* allocate buffers in bulk directly into the S/W ring */
1164 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1165 rxep = &rxq->sw_ring[alloc_idx];
1166 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1167 rxq->rx_free_thresh);
1168 if (unlikely(diag != 0))
1171 rxdp = &rxq->rx_ring[alloc_idx];
1172 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1173 /* populate the static rte mbuf fields */
1178 mb->port = rxq->port_id;
1181 rte_mbuf_refcnt_set(mb, 1);
1182 mb->data_off = RTE_PKTMBUF_HEADROOM;
1184 /* populate the descriptors */
1185 dma_addr = rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
1186 rxdp[i].read.hdr_addr = dma_addr;
1187 rxdp[i].read.pkt_addr = dma_addr;
1190 /* update state of internal queue structure */
1191 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1192 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1193 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1199 static inline uint16_t
1200 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1203 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1206 /* how many packets are ready to return? */
1207 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1209 /* copy mbuf pointers to the application's packet list */
1210 for (i = 0; i < nb_pkts; ++i)
1211 rx_pkts[i] = stage[i];
1213 /* update internal queue state */
1214 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1215 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1220 static inline uint16_t
1221 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1224 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1227 /* Any previously recv'd pkts will be returned from the Rx stage */
1228 if (rxq->rx_nb_avail)
1229 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1231 /* Scan the H/W ring for packets to receive */
1232 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1234 /* update internal queue state */
1235 rxq->rx_next_avail = 0;
1236 rxq->rx_nb_avail = nb_rx;
1237 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1239 /* if required, allocate new buffers to replenish descriptors */
1240 if (rxq->rx_tail > rxq->rx_free_trigger) {
1241 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1243 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1245 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1246 "queue_id=%u", (unsigned) rxq->port_id,
1247 (unsigned) rxq->queue_id);
1249 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1250 rxq->rx_free_thresh;
1253 * Need to rewind any previous receives if we cannot
1254 * allocate new buffers to replenish the old ones.
1256 rxq->rx_nb_avail = 0;
1257 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1258 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1259 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1264 /* update tail pointer */
1266 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, cur_free_trigger);
1269 if (rxq->rx_tail >= rxq->nb_rx_desc)
1272 /* received any packets this loop? */
1273 if (rxq->rx_nb_avail)
1274 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1279 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1281 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1286 if (unlikely(nb_pkts == 0))
1289 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1290 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1292 /* request is relatively large, chunk it up */
1296 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1297 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1298 nb_rx = (uint16_t)(nb_rx + ret);
1299 nb_pkts = (uint16_t)(nb_pkts - ret);
1308 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1311 struct ixgbe_rx_queue *rxq;
1312 volatile union ixgbe_adv_rx_desc *rx_ring;
1313 volatile union ixgbe_adv_rx_desc *rxdp;
1314 struct ixgbe_rx_entry *sw_ring;
1315 struct ixgbe_rx_entry *rxe;
1316 struct rte_mbuf *rxm;
1317 struct rte_mbuf *nmb;
1318 union ixgbe_adv_rx_desc rxd;
1324 uint32_t hlen_type_rss;
1335 rx_id = rxq->rx_tail;
1336 rx_ring = rxq->rx_ring;
1337 sw_ring = rxq->sw_ring;
1338 while (nb_rx < nb_pkts) {
1340 * The order of operations here is important as the DD status
1341 * bit must not be read after any other descriptor fields.
1342 * rx_ring and rxdp are pointing to volatile data so the order
1343 * of accesses cannot be reordered by the compiler. If they were
1344 * not volatile, they could be reordered which could lead to
1345 * using invalid descriptor fields when read from rxd.
1347 rxdp = &rx_ring[rx_id];
1348 staterr = rxdp->wb.upper.status_error;
1349 if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1356 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1357 * is likely to be invalid and to be dropped by the various
1358 * validation checks performed by the network stack.
1360 * Allocate a new mbuf to replenish the RX ring descriptor.
1361 * If the allocation fails:
1362 * - arrange for that RX descriptor to be the first one
1363 * being parsed the next time the receive function is
1364 * invoked [on the same queue].
1366 * - Stop parsing the RX ring and return immediately.
1368 * This policy do not drop the packet received in the RX
1369 * descriptor for which the allocation of a new mbuf failed.
1370 * Thus, it allows that packet to be later retrieved if
1371 * mbuf have been freed in the mean time.
1372 * As a side effect, holding RX descriptors instead of
1373 * systematically giving them back to the NIC may lead to
1374 * RX ring exhaustion situations.
1375 * However, the NIC can gracefully prevent such situations
1376 * to happen by sending specific "back-pressure" flow control
1377 * frames to its peer(s).
1379 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1380 "ext_err_stat=0x%08x pkt_len=%u",
1381 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1382 (unsigned) rx_id, (unsigned) staterr,
1383 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1385 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1387 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1388 "queue_id=%u", (unsigned) rxq->port_id,
1389 (unsigned) rxq->queue_id);
1390 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1395 rxe = &sw_ring[rx_id];
1397 if (rx_id == rxq->nb_rx_desc)
1400 /* Prefetch next mbuf while processing current one. */
1401 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1404 * When next RX descriptor is on a cache-line boundary,
1405 * prefetch the next 4 RX descriptors and the next 8 pointers
1408 if ((rx_id & 0x3) == 0) {
1409 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1410 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1416 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1417 rxdp->read.hdr_addr = dma_addr;
1418 rxdp->read.pkt_addr = dma_addr;
1421 * Initialize the returned mbuf.
1422 * 1) setup generic mbuf fields:
1423 * - number of segments,
1426 * - RX port identifier.
1427 * 2) integrate hardware offload data, if any:
1428 * - RSS flag & hash,
1429 * - IP checksum flag,
1430 * - VLAN TCI, if any,
1433 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1435 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1436 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1439 rxm->pkt_len = pkt_len;
1440 rxm->data_len = pkt_len;
1441 rxm->port = rxq->port_id;
1444 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.hs_rss.
1446 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1447 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1449 pkt_flags = rx_desc_status_to_pkt_flags(staterr);
1450 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1451 pkt_flags = pkt_flags |
1452 ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
1453 rxm->ol_flags = pkt_flags;
1454 rxm->packet_type = ixgbe_rxd_pkt_info_to_pkt_type(pkt_info);
1455 #else /* RTE_NEXT_ABI */
1456 hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1457 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1458 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1460 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1461 pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
1462 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1463 rxm->ol_flags = pkt_flags;
1464 #endif /* RTE_NEXT_ABI */
1466 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1467 rxm->hash.rss = rxd.wb.lower.hi_dword.rss;
1468 else if (pkt_flags & PKT_RX_FDIR) {
1469 rxm->hash.fdir.hash =
1470 (uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
1471 & IXGBE_ATR_HASH_MASK);
1472 rxm->hash.fdir.id = rxd.wb.lower.hi_dword.csum_ip.ip_id;
1475 * Store the mbuf address into the next entry of the array
1476 * of returned packets.
1478 rx_pkts[nb_rx++] = rxm;
1480 rxq->rx_tail = rx_id;
1483 * If the number of free RX descriptors is greater than the RX free
1484 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1486 * Update the RDT with the value of the last processed RX descriptor
1487 * minus 1, to guarantee that the RDT register is never equal to the
1488 * RDH register, which creates a "full" ring situtation from the
1489 * hardware point of view...
1491 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1492 if (nb_hold > rxq->rx_free_thresh) {
1493 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1494 "nb_hold=%u nb_rx=%u",
1495 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1496 (unsigned) rx_id, (unsigned) nb_hold,
1498 rx_id = (uint16_t) ((rx_id == 0) ?
1499 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1500 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1503 rxq->nb_rx_hold = nb_hold;
1508 * Detect an RSC descriptor.
1510 static inline uint32_t
1511 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1513 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1514 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1518 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1520 * Fill the following info in the HEAD buffer of the Rx cluster:
1521 * - RX port identifier
1522 * - hardware offload data, if any:
1524 * - IP checksum flag
1525 * - VLAN TCI, if any
1527 * @head HEAD of the packet cluster
1528 * @desc HW descriptor to get data from
1529 * @port_id Port ID of the Rx queue
1532 ixgbe_fill_cluster_head_buf(
1533 struct rte_mbuf *head,
1534 union ixgbe_adv_rx_desc *desc,
1542 head->port = port_id;
1544 /* The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1545 * set in the pkt_flags field.
1547 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1548 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.hs_rss.pkt_info);
1549 pkt_flags = rx_desc_status_to_pkt_flags(staterr);
1550 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1551 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
1552 head->ol_flags = pkt_flags;
1553 head->packet_type = ixgbe_rxd_pkt_info_to_pkt_type(pkt_info);
1554 #else /* RTE_NEXT_ABI */
1555 uint32_t hlen_type_rss;
1558 head->port = port_id;
1561 * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1562 * set in the pkt_flags field.
1564 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1565 hlen_type_rss = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1566 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1567 pkt_flags |= rx_desc_status_to_pkt_flags(staterr);
1568 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1569 head->ol_flags = pkt_flags;
1570 #endif /* RTE_NEXT_ABI */
1572 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1573 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1574 else if (pkt_flags & PKT_RX_FDIR) {
1575 head->hash.fdir.hash =
1576 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1577 & IXGBE_ATR_HASH_MASK;
1578 head->hash.fdir.id =
1579 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1584 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1586 * @rx_queue Rx queue handle
1587 * @rx_pkts table of received packets
1588 * @nb_pkts size of rx_pkts table
1589 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1591 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1592 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1594 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1595 * 1) When non-EOP RSC completion arrives:
1596 * a) Update the HEAD of the current RSC aggregation cluster with the new
1597 * segment's data length.
1598 * b) Set the "next" pointer of the current segment to point to the segment
1599 * at the NEXTP index.
1600 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
1601 * in the sw_rsc_ring.
1602 * 2) When EOP arrives we just update the cluster's total length and offload
1603 * flags and deliver the cluster up to the upper layers. In our case - put it
1604 * in the rx_pkts table.
1606 * Returns the number of received packets/clusters (according to the "bulk
1607 * receive" interface).
1609 static inline uint16_t
1610 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
1613 struct ixgbe_rx_queue *rxq = rx_queue;
1614 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
1615 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
1616 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
1617 uint16_t rx_id = rxq->rx_tail;
1619 uint16_t nb_hold = rxq->nb_rx_hold;
1620 uint16_t prev_id = rxq->rx_tail;
1622 while (nb_rx < nb_pkts) {
1624 struct ixgbe_rx_entry *rxe;
1625 struct ixgbe_scattered_rx_entry *sc_entry;
1626 struct ixgbe_scattered_rx_entry *next_sc_entry;
1627 struct ixgbe_rx_entry *next_rxe;
1628 struct rte_mbuf *first_seg;
1629 struct rte_mbuf *rxm;
1630 struct rte_mbuf *nmb;
1631 union ixgbe_adv_rx_desc rxd;
1634 volatile union ixgbe_adv_rx_desc *rxdp;
1639 * The code in this whole file uses the volatile pointer to
1640 * ensure the read ordering of the status and the rest of the
1641 * descriptor fields (on the compiler level only!!!). This is so
1642 * UGLY - why not to just use the compiler barrier instead? DPDK
1643 * even has the rte_compiler_barrier() for that.
1645 * But most importantly this is just wrong because this doesn't
1646 * ensure memory ordering in a general case at all. For
1647 * instance, DPDK is supposed to work on Power CPUs where
1648 * compiler barrier may just not be enough!
1650 * I tried to write only this function properly to have a
1651 * starting point (as a part of an LRO/RSC series) but the
1652 * compiler cursed at me when I tried to cast away the
1653 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
1654 * keeping it the way it is for now.
1656 * The code in this file is broken in so many other places and
1657 * will just not work on a big endian CPU anyway therefore the
1658 * lines below will have to be revisited together with the rest
1662 * - Get rid of "volatile" crap and let the compiler do its
1664 * - Use the proper memory barrier (rte_rmb()) to ensure the
1665 * memory ordering below.
1667 rxdp = &rx_ring[rx_id];
1668 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
1670 if (!(staterr & IXGBE_RXDADV_STAT_DD))
1675 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1676 "staterr=0x%x data_len=%u",
1677 rxq->port_id, rxq->queue_id, rx_id, staterr,
1678 rte_le_to_cpu_16(rxd.wb.upper.length));
1681 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1683 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
1684 "port_id=%u queue_id=%u",
1685 rxq->port_id, rxq->queue_id);
1687 rte_eth_devices[rxq->port_id].data->
1688 rx_mbuf_alloc_failed++;
1692 else if (nb_hold > rxq->rx_free_thresh) {
1693 uint16_t next_rdt = rxq->rx_free_trigger;
1695 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
1697 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr,
1699 nb_hold -= rxq->rx_free_thresh;
1701 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
1702 "port_id=%u queue_id=%u",
1703 rxq->port_id, rxq->queue_id);
1705 rte_eth_devices[rxq->port_id].data->
1706 rx_mbuf_alloc_failed++;
1712 rxe = &sw_ring[rx_id];
1713 eop = staterr & IXGBE_RXDADV_STAT_EOP;
1715 next_id = rx_id + 1;
1716 if (next_id == rxq->nb_rx_desc)
1719 /* Prefetch next mbuf while processing current one. */
1720 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
1723 * When next RX descriptor is on a cache-line boundary,
1724 * prefetch the next 4 RX descriptors and the next 4 pointers
1727 if ((next_id & 0x3) == 0) {
1728 rte_ixgbe_prefetch(&rx_ring[next_id]);
1729 rte_ixgbe_prefetch(&sw_ring[next_id]);
1736 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1738 * Update RX descriptor with the physical address of the
1739 * new data buffer of the new allocated mbuf.
1743 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1744 rxdp->read.hdr_addr = dma;
1745 rxdp->read.pkt_addr = dma;
1750 * Set data length & data buffer address of mbuf.
1752 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
1753 rxm->data_len = data_len;
1758 * Get next descriptor index:
1759 * - For RSC it's in the NEXTP field.
1760 * - For a scattered packet - it's just a following
1763 if (ixgbe_rsc_count(&rxd))
1765 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1766 IXGBE_RXDADV_NEXTP_SHIFT;
1770 next_sc_entry = &sw_sc_ring[nextp_id];
1771 next_rxe = &sw_ring[nextp_id];
1772 rte_ixgbe_prefetch(next_rxe);
1775 sc_entry = &sw_sc_ring[rx_id];
1776 first_seg = sc_entry->fbuf;
1777 sc_entry->fbuf = NULL;
1780 * If this is the first buffer of the received packet,
1781 * set the pointer to the first mbuf of the packet and
1782 * initialize its context.
1783 * Otherwise, update the total length and the number of segments
1784 * of the current scattered packet, and update the pointer to
1785 * the last mbuf of the current packet.
1787 if (first_seg == NULL) {
1789 first_seg->pkt_len = data_len;
1790 first_seg->nb_segs = 1;
1792 first_seg->pkt_len += data_len;
1793 first_seg->nb_segs++;
1800 * If this is not the last buffer of the received packet, update
1801 * the pointer to the first mbuf at the NEXTP entry in the
1802 * sw_sc_ring and continue to parse the RX ring.
1805 rxm->next = next_rxe->mbuf;
1806 next_sc_entry->fbuf = first_seg;
1811 * This is the last buffer of the received packet - return
1812 * the current cluster to the user.
1816 /* Initialize the first mbuf of the returned packet */
1817 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq->port_id,
1820 /* Prefetch data of first segment, if configured to do so. */
1821 rte_packet_prefetch((char *)first_seg->buf_addr +
1822 first_seg->data_off);
1825 * Store the mbuf address into the next entry of the array
1826 * of returned packets.
1828 rx_pkts[nb_rx++] = first_seg;
1832 * Record index of the next RX descriptor to probe.
1834 rxq->rx_tail = rx_id;
1837 * If the number of free RX descriptors is greater than the RX free
1838 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1840 * Update the RDT with the value of the last processed RX descriptor
1841 * minus 1, to guarantee that the RDT register is never equal to the
1842 * RDH register, which creates a "full" ring situtation from the
1843 * hardware point of view...
1845 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
1846 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1847 "nb_hold=%u nb_rx=%u",
1848 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
1851 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, prev_id);
1855 rxq->nb_rx_hold = nb_hold;
1860 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1863 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
1867 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1870 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
1873 /*********************************************************************
1875 * Queue management functions
1877 **********************************************************************/
1880 * Rings setup and release.
1882 * TDBA/RDBA should be aligned on 16 byte boundary. But TDLEN/RDLEN should be
1883 * multiple of 128 bytes. So we align TDBA/RDBA on 128 byte boundary. This will
1884 * also optimize cache line size effect. H/W supports up to cache line size 128.
1886 #define IXGBE_ALIGN 128
1889 * Maximum number of Ring Descriptors.
1891 * Since RDLEN/TDLEN should be multiple of 128 bytes, the number of ring
1892 * descriptors should meet the following condition:
1893 * (num_ring_desc * sizeof(rx/tx descriptor)) % 128 == 0
1895 #define IXGBE_MIN_RING_DESC 32
1896 #define IXGBE_MAX_RING_DESC 4096
1899 * Create memzone for HW rings. malloc can't be used as the physical address is
1900 * needed. If the memzone is already created, then this function returns a ptr
1903 static const struct rte_memzone * __attribute__((cold))
1904 ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
1905 uint16_t queue_id, uint32_t ring_size, int socket_id)
1907 char z_name[RTE_MEMZONE_NAMESIZE];
1908 const struct rte_memzone *mz;
1910 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
1911 dev->driver->pci_drv.name, ring_name,
1912 dev->data->port_id, queue_id);
1914 mz = rte_memzone_lookup(z_name);
1918 #ifdef RTE_LIBRTE_XEN_DOM0
1919 return rte_memzone_reserve_bounded(z_name, ring_size,
1920 socket_id, 0, IXGBE_ALIGN, RTE_PGSIZE_2M);
1922 return rte_memzone_reserve_aligned(z_name, ring_size,
1923 socket_id, 0, IXGBE_ALIGN);
1927 static void __attribute__((cold))
1928 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
1932 if (txq->sw_ring != NULL) {
1933 for (i = 0; i < txq->nb_tx_desc; i++) {
1934 if (txq->sw_ring[i].mbuf != NULL) {
1935 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
1936 txq->sw_ring[i].mbuf = NULL;
1942 static void __attribute__((cold))
1943 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
1946 txq->sw_ring != NULL)
1947 rte_free(txq->sw_ring);
1950 static void __attribute__((cold))
1951 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
1953 if (txq != NULL && txq->ops != NULL) {
1954 txq->ops->release_mbufs(txq);
1955 txq->ops->free_swring(txq);
1960 void __attribute__((cold))
1961 ixgbe_dev_tx_queue_release(void *txq)
1963 ixgbe_tx_queue_release(txq);
1966 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
1967 static void __attribute__((cold))
1968 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
1970 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
1971 struct ixgbe_tx_entry *txe = txq->sw_ring;
1974 /* Zero out HW ring memory */
1975 for (i = 0; i < txq->nb_tx_desc; i++) {
1976 txq->tx_ring[i] = zeroed_desc;
1979 /* Initialize SW ring entries */
1980 prev = (uint16_t) (txq->nb_tx_desc - 1);
1981 for (i = 0; i < txq->nb_tx_desc; i++) {
1982 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
1983 txd->wb.status = IXGBE_TXD_STAT_DD;
1986 txe[prev].next_id = i;
1990 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1991 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1994 txq->nb_tx_used = 0;
1996 * Always allow 1 descriptor to be un-allocated to avoid
1997 * a H/W race condition
1999 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2000 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2002 memset((void*)&txq->ctx_cache, 0,
2003 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2006 static const struct ixgbe_txq_ops def_txq_ops = {
2007 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2008 .free_swring = ixgbe_tx_free_swring,
2009 .reset = ixgbe_reset_tx_queue,
2012 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2013 * the queue parameters. Used in tx_queue_setup by primary process and then
2014 * in dev_init by secondary process when attaching to an existing ethdev.
2016 void __attribute__((cold))
2017 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2019 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2020 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
2021 && (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2022 PMD_INIT_LOG(INFO, "Using simple tx code path");
2023 #ifdef RTE_IXGBE_INC_VECTOR
2024 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2025 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2026 ixgbe_txq_vec_setup(txq) == 0)) {
2027 PMD_INIT_LOG(INFO, "Vector tx enabled.");
2028 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2031 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2033 PMD_INIT_LOG(INFO, "Using full-featured tx code path");
2035 " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
2036 (unsigned long)txq->txq_flags,
2037 (unsigned long)IXGBE_SIMPLE_FLAGS);
2039 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2040 (unsigned long)txq->tx_rs_thresh,
2041 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2042 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2046 int __attribute__((cold))
2047 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2050 unsigned int socket_id,
2051 const struct rte_eth_txconf *tx_conf)
2053 const struct rte_memzone *tz;
2054 struct ixgbe_tx_queue *txq;
2055 struct ixgbe_hw *hw;
2056 uint16_t tx_rs_thresh, tx_free_thresh;
2058 PMD_INIT_FUNC_TRACE();
2059 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2062 * Validate number of transmit descriptors.
2063 * It must not exceed hardware maximum, and must be multiple
2066 if (((nb_desc * sizeof(union ixgbe_adv_tx_desc)) % IXGBE_ALIGN) != 0 ||
2067 (nb_desc > IXGBE_MAX_RING_DESC) ||
2068 (nb_desc < IXGBE_MIN_RING_DESC)) {
2073 * The following two parameters control the setting of the RS bit on
2074 * transmit descriptors.
2075 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2076 * descriptors have been used.
2077 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2078 * descriptors are used or if the number of descriptors required
2079 * to transmit a packet is greater than the number of free TX
2081 * The following constraints must be satisfied:
2082 * tx_rs_thresh must be greater than 0.
2083 * tx_rs_thresh must be less than the size of the ring minus 2.
2084 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2085 * tx_rs_thresh must be a divisor of the ring size.
2086 * tx_free_thresh must be greater than 0.
2087 * tx_free_thresh must be less than the size of the ring minus 3.
2088 * One descriptor in the TX ring is used as a sentinel to avoid a
2089 * H/W race condition, hence the maximum threshold constraints.
2090 * When set to zero use default values.
2092 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2093 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2094 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2095 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2096 if (tx_rs_thresh >= (nb_desc - 2)) {
2097 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2098 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2099 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2100 (int)dev->data->port_id, (int)queue_idx);
2103 if (tx_free_thresh >= (nb_desc - 3)) {
2104 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2105 "tx_free_thresh must be less than the number of "
2106 "TX descriptors minus 3. (tx_free_thresh=%u "
2107 "port=%d queue=%d)",
2108 (unsigned int)tx_free_thresh,
2109 (int)dev->data->port_id, (int)queue_idx);
2112 if (tx_rs_thresh > tx_free_thresh) {
2113 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2114 "tx_free_thresh. (tx_free_thresh=%u "
2115 "tx_rs_thresh=%u port=%d queue=%d)",
2116 (unsigned int)tx_free_thresh,
2117 (unsigned int)tx_rs_thresh,
2118 (int)dev->data->port_id,
2122 if ((nb_desc % tx_rs_thresh) != 0) {
2123 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2124 "number of TX descriptors. (tx_rs_thresh=%u "
2125 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2126 (int)dev->data->port_id, (int)queue_idx);
2131 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2132 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2133 * by the NIC and all descriptors are written back after the NIC
2134 * accumulates WTHRESH descriptors.
2136 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2137 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2138 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2139 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2140 (int)dev->data->port_id, (int)queue_idx);
2144 /* Free memory prior to re-allocation if needed... */
2145 if (dev->data->tx_queues[queue_idx] != NULL) {
2146 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2147 dev->data->tx_queues[queue_idx] = NULL;
2150 /* First allocate the tx queue data structure */
2151 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2152 RTE_CACHE_LINE_SIZE, socket_id);
2157 * Allocate TX ring hardware descriptors. A memzone large enough to
2158 * handle the maximum ring size is allocated in order to allow for
2159 * resizing in later calls to the queue setup function.
2161 tz = ring_dma_zone_reserve(dev, "tx_ring", queue_idx,
2162 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2165 ixgbe_tx_queue_release(txq);
2169 txq->nb_tx_desc = nb_desc;
2170 txq->tx_rs_thresh = tx_rs_thresh;
2171 txq->tx_free_thresh = tx_free_thresh;
2172 txq->pthresh = tx_conf->tx_thresh.pthresh;
2173 txq->hthresh = tx_conf->tx_thresh.hthresh;
2174 txq->wthresh = tx_conf->tx_thresh.wthresh;
2175 txq->queue_id = queue_idx;
2176 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2177 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2178 txq->port_id = dev->data->port_id;
2179 txq->txq_flags = tx_conf->txq_flags;
2180 txq->ops = &def_txq_ops;
2181 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2184 * Modification to set VFTDT for virtual function if vf is detected
2186 if (hw->mac.type == ixgbe_mac_82599_vf ||
2187 hw->mac.type == ixgbe_mac_X540_vf ||
2188 hw->mac.type == ixgbe_mac_X550_vf ||
2189 hw->mac.type == ixgbe_mac_X550EM_x_vf)
2190 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2192 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2193 #ifndef RTE_LIBRTE_XEN_DOM0
2194 txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
2196 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2198 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2200 /* Allocate software ring */
2201 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2202 sizeof(struct ixgbe_tx_entry) * nb_desc,
2203 RTE_CACHE_LINE_SIZE, socket_id);
2204 if (txq->sw_ring == NULL) {
2205 ixgbe_tx_queue_release(txq);
2208 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2209 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2211 /* set up vector or scalar TX function as appropriate */
2212 ixgbe_set_tx_function(dev, txq);
2214 txq->ops->reset(txq);
2216 dev->data->tx_queues[queue_idx] = txq;
2223 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2225 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2226 * in the sw_rsc_ring is not set to NULL but rather points to the next
2227 * mbuf of this RSC aggregation (that has not been completed yet and still
2228 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2229 * will just free first "nb_segs" segments of the cluster explicitly by calling
2230 * an rte_pktmbuf_free_seg().
2232 * @m scattered cluster head
2234 static void __attribute__((cold))
2235 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2237 uint8_t i, nb_segs = m->nb_segs;
2238 struct rte_mbuf *next_seg;
2240 for (i = 0; i < nb_segs; i++) {
2242 rte_pktmbuf_free_seg(m);
2247 static void __attribute__((cold))
2248 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2252 #ifdef RTE_IXGBE_INC_VECTOR
2253 /* SSE Vector driver has a different way of releasing mbufs. */
2254 if (rxq->rx_using_sse) {
2255 ixgbe_rx_queue_release_mbufs_vec(rxq);
2260 if (rxq->sw_ring != NULL) {
2261 for (i = 0; i < rxq->nb_rx_desc; i++) {
2262 if (rxq->sw_ring[i].mbuf != NULL) {
2263 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2264 rxq->sw_ring[i].mbuf = NULL;
2267 if (rxq->rx_nb_avail) {
2268 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2269 struct rte_mbuf *mb;
2270 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2271 rte_pktmbuf_free_seg(mb);
2273 rxq->rx_nb_avail = 0;
2277 if (rxq->sw_sc_ring)
2278 for (i = 0; i < rxq->nb_rx_desc; i++)
2279 if (rxq->sw_sc_ring[i].fbuf) {
2280 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2281 rxq->sw_sc_ring[i].fbuf = NULL;
2285 static void __attribute__((cold))
2286 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2289 ixgbe_rx_queue_release_mbufs(rxq);
2290 rte_free(rxq->sw_ring);
2291 rte_free(rxq->sw_sc_ring);
2296 void __attribute__((cold))
2297 ixgbe_dev_rx_queue_release(void *rxq)
2299 ixgbe_rx_queue_release(rxq);
2303 * Check if Rx Burst Bulk Alloc function can be used.
2305 * 0: the preconditions are satisfied and the bulk allocation function
2307 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2308 * function must be used.
2310 static inline int __attribute__((cold))
2311 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2316 * Make sure the following pre-conditions are satisfied:
2317 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2318 * rxq->rx_free_thresh < rxq->nb_rx_desc
2319 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2320 * rxq->nb_rx_desc<(IXGBE_MAX_RING_DESC-RTE_PMD_IXGBE_RX_MAX_BURST)
2321 * Scattered packets are not supported. This should be checked
2322 * outside of this function.
2324 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2325 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2326 "rxq->rx_free_thresh=%d, "
2327 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2328 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2330 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2331 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2332 "rxq->rx_free_thresh=%d, "
2333 "rxq->nb_rx_desc=%d",
2334 rxq->rx_free_thresh, rxq->nb_rx_desc);
2336 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2337 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2338 "rxq->nb_rx_desc=%d, "
2339 "rxq->rx_free_thresh=%d",
2340 rxq->nb_rx_desc, rxq->rx_free_thresh);
2342 } else if (!(rxq->nb_rx_desc <
2343 (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST))) {
2344 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2345 "rxq->nb_rx_desc=%d, "
2346 "IXGBE_MAX_RING_DESC=%d, "
2347 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2348 rxq->nb_rx_desc, IXGBE_MAX_RING_DESC,
2349 RTE_PMD_IXGBE_RX_MAX_BURST);
2356 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2357 static void __attribute__((cold))
2358 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2360 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2362 uint16_t len = rxq->nb_rx_desc;
2365 * By default, the Rx queue setup function allocates enough memory for
2366 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2367 * extra memory at the end of the descriptor ring to be zero'd out. A
2368 * pre-condition for using the Rx burst bulk alloc function is that the
2369 * number of descriptors is less than or equal to
2370 * (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST). Check all the
2371 * constraints here to see if we need to zero out memory after the end
2372 * of the H/W descriptor ring.
2374 if (adapter->rx_bulk_alloc_allowed)
2375 /* zero out extra memory */
2376 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2379 * Zero out HW ring memory. Zero out extra memory at the end of
2380 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2381 * reads extra memory as zeros.
2383 for (i = 0; i < len; i++) {
2384 rxq->rx_ring[i] = zeroed_desc;
2388 * initialize extra software ring entries. Space for these extra
2389 * entries is always allocated
2391 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2392 for (i = rxq->nb_rx_desc; i < len; ++i) {
2393 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2396 rxq->rx_nb_avail = 0;
2397 rxq->rx_next_avail = 0;
2398 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2400 rxq->nb_rx_hold = 0;
2401 rxq->pkt_first_seg = NULL;
2402 rxq->pkt_last_seg = NULL;
2404 #ifdef RTE_IXGBE_INC_VECTOR
2405 rxq->rxrearm_start = 0;
2406 rxq->rxrearm_nb = 0;
2410 int __attribute__((cold))
2411 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2414 unsigned int socket_id,
2415 const struct rte_eth_rxconf *rx_conf,
2416 struct rte_mempool *mp)
2418 const struct rte_memzone *rz;
2419 struct ixgbe_rx_queue *rxq;
2420 struct ixgbe_hw *hw;
2422 struct ixgbe_adapter *adapter =
2423 (struct ixgbe_adapter *)dev->data->dev_private;
2425 PMD_INIT_FUNC_TRACE();
2426 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2429 * Validate number of receive descriptors.
2430 * It must not exceed hardware maximum, and must be multiple
2433 if (((nb_desc * sizeof(union ixgbe_adv_rx_desc)) % IXGBE_ALIGN) != 0 ||
2434 (nb_desc > IXGBE_MAX_RING_DESC) ||
2435 (nb_desc < IXGBE_MIN_RING_DESC)) {
2439 /* Free memory prior to re-allocation if needed... */
2440 if (dev->data->rx_queues[queue_idx] != NULL) {
2441 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2442 dev->data->rx_queues[queue_idx] = NULL;
2445 /* First allocate the rx queue data structure */
2446 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2447 RTE_CACHE_LINE_SIZE, socket_id);
2451 rxq->nb_rx_desc = nb_desc;
2452 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2453 rxq->queue_id = queue_idx;
2454 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2455 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2456 rxq->port_id = dev->data->port_id;
2457 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2459 rxq->drop_en = rx_conf->rx_drop_en;
2460 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2463 * Allocate RX ring hardware descriptors. A memzone large enough to
2464 * handle the maximum ring size is allocated in order to allow for
2465 * resizing in later calls to the queue setup function.
2467 rz = ring_dma_zone_reserve(dev, "rx_ring", queue_idx,
2468 RX_RING_SZ, socket_id);
2470 ixgbe_rx_queue_release(rxq);
2475 * Zero init all the descriptors in the ring.
2477 memset (rz->addr, 0, RX_RING_SZ);
2480 * Modified to setup VFRDT for Virtual Function
2482 if (hw->mac.type == ixgbe_mac_82599_vf ||
2483 hw->mac.type == ixgbe_mac_X540_vf ||
2484 hw->mac.type == ixgbe_mac_X550_vf ||
2485 hw->mac.type == ixgbe_mac_X550EM_x_vf) {
2487 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2489 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2493 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2495 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2497 #ifndef RTE_LIBRTE_XEN_DOM0
2498 rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
2500 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2502 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2505 * Certain constraints must be met in order to use the bulk buffer
2506 * allocation Rx burst function. If any of Rx queues doesn't meet them
2507 * the feature should be disabled for the whole port.
2509 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
2510 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
2511 "preconditions - canceling the feature for "
2512 "the whole port[%d]",
2513 rxq->queue_id, rxq->port_id);
2514 adapter->rx_bulk_alloc_allowed = false;
2518 * Allocate software ring. Allow for space at the end of the
2519 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2520 * function does not access an invalid memory region.
2523 if (adapter->rx_bulk_alloc_allowed)
2524 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2526 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2527 sizeof(struct ixgbe_rx_entry) * len,
2528 RTE_CACHE_LINE_SIZE, socket_id);
2529 if (!rxq->sw_ring) {
2530 ixgbe_rx_queue_release(rxq);
2535 * Always allocate even if it's not going to be needed in order to
2536 * simplify the code.
2538 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
2539 * be requested in ixgbe_dev_rx_init(), which is called later from
2543 rte_zmalloc_socket("rxq->sw_sc_ring",
2544 sizeof(struct ixgbe_scattered_rx_entry) * len,
2545 RTE_CACHE_LINE_SIZE, socket_id);
2546 if (!rxq->sw_sc_ring) {
2547 ixgbe_rx_queue_release(rxq);
2551 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
2552 "dma_addr=0x%"PRIx64,
2553 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
2554 rxq->rx_ring_phys_addr);
2556 if (!rte_is_power_of_2(nb_desc)) {
2557 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
2558 "preconditions - canceling the feature for "
2559 "the whole port[%d]",
2560 rxq->queue_id, rxq->port_id);
2561 adapter->rx_vec_allowed = false;
2563 ixgbe_rxq_vec_setup(rxq);
2565 dev->data->rx_queues[queue_idx] = rxq;
2567 ixgbe_reset_rx_queue(adapter, rxq);
2573 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2575 #define IXGBE_RXQ_SCAN_INTERVAL 4
2576 volatile union ixgbe_adv_rx_desc *rxdp;
2577 struct ixgbe_rx_queue *rxq;
2580 if (rx_queue_id >= dev->data->nb_rx_queues) {
2581 PMD_RX_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
2585 rxq = dev->data->rx_queues[rx_queue_id];
2586 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2588 while ((desc < rxq->nb_rx_desc) &&
2589 (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD)) {
2590 desc += IXGBE_RXQ_SCAN_INTERVAL;
2591 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2592 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2593 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2594 desc - rxq->nb_rx_desc]);
2601 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2603 volatile union ixgbe_adv_rx_desc *rxdp;
2604 struct ixgbe_rx_queue *rxq = rx_queue;
2607 if (unlikely(offset >= rxq->nb_rx_desc))
2609 desc = rxq->rx_tail + offset;
2610 if (desc >= rxq->nb_rx_desc)
2611 desc -= rxq->nb_rx_desc;
2613 rxdp = &rxq->rx_ring[desc];
2614 return !!(rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD);
2617 void __attribute__((cold))
2618 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
2621 struct ixgbe_adapter *adapter =
2622 (struct ixgbe_adapter *)dev->data->dev_private;
2624 PMD_INIT_FUNC_TRACE();
2626 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2627 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
2629 txq->ops->release_mbufs(txq);
2630 txq->ops->reset(txq);
2634 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2635 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
2637 ixgbe_rx_queue_release_mbufs(rxq);
2638 ixgbe_reset_rx_queue(adapter, rxq);
2644 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
2648 PMD_INIT_FUNC_TRACE();
2650 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2651 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
2652 dev->data->rx_queues[i] = NULL;
2654 dev->data->nb_rx_queues = 0;
2656 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2657 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
2658 dev->data->tx_queues[i] = NULL;
2660 dev->data->nb_tx_queues = 0;
2663 /*********************************************************************
2665 * Device RX/TX init functions
2667 **********************************************************************/
2670 * Receive Side Scaling (RSS)
2671 * See section 7.1.2.8 in the following document:
2672 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
2675 * The source and destination IP addresses of the IP header and the source
2676 * and destination ports of TCP/UDP headers, if any, of received packets are
2677 * hashed against a configurable random key to compute a 32-bit RSS hash result.
2678 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
2679 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
2680 * RSS output index which is used as the RX queue index where to store the
2682 * The following output is supplied in the RX write-back descriptor:
2683 * - 32-bit result of the Microsoft RSS hash function,
2684 * - 4-bit RSS type field.
2688 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
2689 * Used as the default key.
2691 static uint8_t rss_intel_key[40] = {
2692 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2693 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2694 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2695 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2696 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2700 ixgbe_rss_disable(struct rte_eth_dev *dev)
2702 struct ixgbe_hw *hw;
2705 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2706 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2707 mrqc &= ~IXGBE_MRQC_RSSEN;
2708 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2712 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
2720 hash_key = rss_conf->rss_key;
2721 if (hash_key != NULL) {
2722 /* Fill in RSS hash key */
2723 for (i = 0; i < 10; i++) {
2724 rss_key = hash_key[(i * 4)];
2725 rss_key |= hash_key[(i * 4) + 1] << 8;
2726 rss_key |= hash_key[(i * 4) + 2] << 16;
2727 rss_key |= hash_key[(i * 4) + 3] << 24;
2728 IXGBE_WRITE_REG_ARRAY(hw, IXGBE_RSSRK(0), i, rss_key);
2732 /* Set configured hashing protocols in MRQC register */
2733 rss_hf = rss_conf->rss_hf;
2734 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
2735 if (rss_hf & ETH_RSS_IPV4)
2736 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
2737 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
2738 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
2739 if (rss_hf & ETH_RSS_IPV6)
2740 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
2741 if (rss_hf & ETH_RSS_IPV6_EX)
2742 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
2743 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
2744 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2745 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
2746 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
2747 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
2748 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2749 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
2750 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2751 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
2752 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2753 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2757 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
2758 struct rte_eth_rss_conf *rss_conf)
2760 struct ixgbe_hw *hw;
2764 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2767 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
2768 * "RSS enabling cannot be done dynamically while it must be
2769 * preceded by a software reset"
2770 * Before changing anything, first check that the update RSS operation
2771 * does not attempt to disable RSS, if RSS was enabled at
2772 * initialization time, or does not attempt to enable RSS, if RSS was
2773 * disabled at initialization time.
2775 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
2776 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2777 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
2778 if (rss_hf != 0) /* Enable RSS */
2780 return 0; /* Nothing to do */
2783 if (rss_hf == 0) /* Disable RSS */
2785 ixgbe_hw_rss_hash_set(hw, rss_conf);
2790 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2791 struct rte_eth_rss_conf *rss_conf)
2793 struct ixgbe_hw *hw;
2800 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2801 hash_key = rss_conf->rss_key;
2802 if (hash_key != NULL) {
2803 /* Return RSS hash key */
2804 for (i = 0; i < 10; i++) {
2805 rss_key = IXGBE_READ_REG_ARRAY(hw, IXGBE_RSSRK(0), i);
2806 hash_key[(i * 4)] = rss_key & 0x000000FF;
2807 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
2808 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
2809 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
2813 /* Get RSS functions configured in MRQC register */
2814 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2815 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
2816 rss_conf->rss_hf = 0;
2820 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
2821 rss_hf |= ETH_RSS_IPV4;
2822 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
2823 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2824 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
2825 rss_hf |= ETH_RSS_IPV6;
2826 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
2827 rss_hf |= ETH_RSS_IPV6_EX;
2828 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
2829 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2830 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
2831 rss_hf |= ETH_RSS_IPV6_TCP_EX;
2832 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
2833 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2834 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
2835 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2836 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
2837 rss_hf |= ETH_RSS_IPV6_UDP_EX;
2838 rss_conf->rss_hf = rss_hf;
2843 ixgbe_rss_configure(struct rte_eth_dev *dev)
2845 struct rte_eth_rss_conf rss_conf;
2846 struct ixgbe_hw *hw;
2851 PMD_INIT_FUNC_TRACE();
2852 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2855 * Fill in redirection table
2856 * The byte-swap is needed because NIC registers are in
2857 * little-endian order.
2860 for (i = 0, j = 0; i < 128; i++, j++) {
2861 if (j == dev->data->nb_rx_queues)
2863 reta = (reta << 8) | j;
2865 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2),
2870 * Configure the RSS key and the RSS protocols used to compute
2871 * the RSS hash of input packets.
2873 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
2874 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
2875 ixgbe_rss_disable(dev);
2878 if (rss_conf.rss_key == NULL)
2879 rss_conf.rss_key = rss_intel_key; /* Default hash key */
2880 ixgbe_hw_rss_hash_set(hw, &rss_conf);
2883 #define NUM_VFTA_REGISTERS 128
2884 #define NIC_RX_BUFFER_SIZE 0x200
2887 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
2889 struct rte_eth_vmdq_dcb_conf *cfg;
2890 struct ixgbe_hw *hw;
2891 enum rte_eth_nb_pools num_pools;
2892 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
2894 uint8_t nb_tcs; /* number of traffic classes */
2897 PMD_INIT_FUNC_TRACE();
2898 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2899 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
2900 num_pools = cfg->nb_queue_pools;
2901 /* Check we have a valid number of pools */
2902 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
2903 ixgbe_rss_disable(dev);
2906 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
2907 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
2911 * split rx buffer up into sections, each for 1 traffic class
2913 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
2914 for (i = 0 ; i < nb_tcs; i++) {
2915 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2916 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
2917 /* clear 10 bits. */
2918 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
2919 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2921 /* zero alloc all unused TCs */
2922 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2923 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2924 rxpbsize &= (~( 0x3FF << IXGBE_RXPBSIZE_SHIFT ));
2925 /* clear 10 bits. */
2926 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2929 /* MRQC: enable vmdq and dcb */
2930 mrqc = ((num_pools == ETH_16_POOLS) ? \
2931 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN );
2932 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2934 /* PFVTCTL: turn on virtualisation and set the default pool */
2935 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2936 if (cfg->enable_default_pool) {
2937 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
2939 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
2942 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
2944 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
2946 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
2948 * mapping is done with 3 bits per priority,
2949 * so shift by i*3 each time
2951 queue_mapping |= ((cfg->dcb_queue[i] & 0x07) << (i * 3));
2953 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
2955 /* RTRPCS: DCB related */
2956 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
2958 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
2959 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2960 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
2961 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
2963 /* VFTA - enable all vlan filters */
2964 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
2965 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
2968 /* VFRE: pool enabling for receive - 16 or 32 */
2969 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), \
2970 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
2973 * MPSAR - allow pools to read specific mac addresses
2974 * In this case, all pools should be able to read from mac addr 0
2976 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
2977 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
2979 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
2980 for (i = 0; i < cfg->nb_pool_maps; i++) {
2981 /* set vlan id in VF register and set the valid bit */
2982 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
2983 (cfg->pool_map[i].vlan_id & 0xFFF)));
2985 * Put the allowed pools in VFB reg. As we only have 16 or 32
2986 * pools, we only need to use the first half of the register
2989 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
2994 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
2995 * @hw: pointer to hardware structure
2996 * @dcb_config: pointer to ixgbe_dcb_config structure
2999 ixgbe_dcb_tx_hw_config(struct ixgbe_hw *hw,
3000 struct ixgbe_dcb_config *dcb_config)
3005 PMD_INIT_FUNC_TRACE();
3006 if (hw->mac.type != ixgbe_mac_82598EB) {
3007 /* Disable the Tx desc arbiter so that MTQC can be changed */
3008 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3009 reg |= IXGBE_RTTDCS_ARBDIS;
3010 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3012 /* Enable DCB for Tx with 8 TCs */
3013 if (dcb_config->num_tcs.pg_tcs == 8) {
3014 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3017 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3019 if (dcb_config->vt_mode)
3020 reg |= IXGBE_MTQC_VT_ENA;
3021 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3023 /* Disable drop for all queues */
3024 for (q = 0; q < 128; q++)
3025 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3026 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3028 /* Enable the Tx desc arbiter */
3029 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3030 reg &= ~IXGBE_RTTDCS_ARBDIS;
3031 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3033 /* Enable Security TX Buffer IFG for DCB */
3034 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3035 reg |= IXGBE_SECTX_DCB;
3036 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3042 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3043 * @dev: pointer to rte_eth_dev structure
3044 * @dcb_config: pointer to ixgbe_dcb_config structure
3047 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3048 struct ixgbe_dcb_config *dcb_config)
3050 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3051 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3052 struct ixgbe_hw *hw =
3053 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3055 PMD_INIT_FUNC_TRACE();
3056 if (hw->mac.type != ixgbe_mac_82598EB)
3057 /*PF VF Transmit Enable*/
3058 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3059 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3061 /*Configure general DCB TX parameters*/
3062 ixgbe_dcb_tx_hw_config(hw,dcb_config);
3067 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3068 struct ixgbe_dcb_config *dcb_config)
3070 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3071 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3072 struct ixgbe_dcb_tc_config *tc;
3075 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3076 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS ) {
3077 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3078 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3081 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3082 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3084 /* User Priority to Traffic Class mapping */
3085 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3086 j = vmdq_rx_conf->dcb_queue[i];
3087 tc = &dcb_config->tc_config[j];
3088 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3094 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3095 struct ixgbe_dcb_config *dcb_config)
3097 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3098 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3099 struct ixgbe_dcb_tc_config *tc;
3102 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3103 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ) {
3104 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3105 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3108 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3109 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3112 /* User Priority to Traffic Class mapping */
3113 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3114 j = vmdq_tx_conf->dcb_queue[i];
3115 tc = &dcb_config->tc_config[j];
3116 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3123 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3124 struct ixgbe_dcb_config *dcb_config)
3126 struct rte_eth_dcb_rx_conf *rx_conf =
3127 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3128 struct ixgbe_dcb_tc_config *tc;
3131 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3132 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3134 /* User Priority to Traffic Class mapping */
3135 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3136 j = rx_conf->dcb_queue[i];
3137 tc = &dcb_config->tc_config[j];
3138 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3144 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3145 struct ixgbe_dcb_config *dcb_config)
3147 struct rte_eth_dcb_tx_conf *tx_conf =
3148 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3149 struct ixgbe_dcb_tc_config *tc;
3152 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3153 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3155 /* User Priority to Traffic Class mapping */
3156 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3157 j = tx_conf->dcb_queue[i];
3158 tc = &dcb_config->tc_config[j];
3159 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3165 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3166 * @hw: pointer to hardware structure
3167 * @dcb_config: pointer to ixgbe_dcb_config structure
3170 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
3171 struct ixgbe_dcb_config *dcb_config)
3177 PMD_INIT_FUNC_TRACE();
3179 * Disable the arbiter before changing parameters
3180 * (always enable recycle mode; WSP)
3182 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3183 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3185 if (hw->mac.type != ixgbe_mac_82598EB) {
3186 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3187 if (dcb_config->num_tcs.pg_tcs == 4) {
3188 if (dcb_config->vt_mode)
3189 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3190 IXGBE_MRQC_VMDQRT4TCEN;
3192 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3193 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3197 if (dcb_config->num_tcs.pg_tcs == 8) {
3198 if (dcb_config->vt_mode)
3199 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3200 IXGBE_MRQC_VMDQRT8TCEN;
3202 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3203 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3208 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3211 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3212 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3213 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
3214 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3216 /* VFTA - enable all vlan filters */
3217 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3218 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3222 * Configure Rx packet plane (recycle mode; WSP) and
3225 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3226 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3232 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3233 uint16_t *max,uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3235 switch (hw->mac.type) {
3236 case ixgbe_mac_82598EB:
3237 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3239 case ixgbe_mac_82599EB:
3240 case ixgbe_mac_X540:
3241 case ixgbe_mac_X550:
3242 case ixgbe_mac_X550EM_x:
3243 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3252 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3253 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3255 switch (hw->mac.type) {
3256 case ixgbe_mac_82598EB:
3257 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id,tsa);
3258 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id,tsa);
3260 case ixgbe_mac_82599EB:
3261 case ixgbe_mac_X540:
3262 case ixgbe_mac_X550:
3263 case ixgbe_mac_X550EM_x:
3264 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id,tsa);
3265 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id,tsa, map);
3272 #define DCB_RX_CONFIG 1
3273 #define DCB_TX_CONFIG 1
3274 #define DCB_TX_PB 1024
3276 * ixgbe_dcb_hw_configure - Enable DCB and configure
3277 * general DCB in VT mode and non-VT mode parameters
3278 * @dev: pointer to rte_eth_dev structure
3279 * @dcb_config: pointer to ixgbe_dcb_config structure
3282 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3283 struct ixgbe_dcb_config *dcb_config)
3286 uint8_t i,pfc_en,nb_tcs;
3288 uint8_t config_dcb_rx = 0;
3289 uint8_t config_dcb_tx = 0;
3290 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3291 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3292 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3293 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3294 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3295 struct ixgbe_dcb_tc_config *tc;
3296 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3297 struct ixgbe_hw *hw =
3298 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3300 switch(dev->data->dev_conf.rxmode.mq_mode){
3301 case ETH_MQ_RX_VMDQ_DCB:
3302 dcb_config->vt_mode = true;
3303 if (hw->mac.type != ixgbe_mac_82598EB) {
3304 config_dcb_rx = DCB_RX_CONFIG;
3306 *get dcb and VT rx configuration parameters
3309 ixgbe_vmdq_dcb_rx_config(dev,dcb_config);
3310 /*Configure general VMDQ and DCB RX parameters*/
3311 ixgbe_vmdq_dcb_configure(dev);
3315 dcb_config->vt_mode = false;
3316 config_dcb_rx = DCB_RX_CONFIG;
3317 /* Get dcb TX configuration parameters from rte_eth_conf */
3318 ixgbe_dcb_rx_config(dev,dcb_config);
3319 /*Configure general DCB RX parameters*/
3320 ixgbe_dcb_rx_hw_config(hw, dcb_config);
3323 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3326 switch (dev->data->dev_conf.txmode.mq_mode) {
3327 case ETH_MQ_TX_VMDQ_DCB:
3328 dcb_config->vt_mode = true;
3329 config_dcb_tx = DCB_TX_CONFIG;
3330 /* get DCB and VT TX configuration parameters from rte_eth_conf */
3331 ixgbe_dcb_vt_tx_config(dev,dcb_config);
3332 /*Configure general VMDQ and DCB TX parameters*/
3333 ixgbe_vmdq_dcb_hw_tx_config(dev,dcb_config);
3337 dcb_config->vt_mode = false;
3338 config_dcb_tx = DCB_TX_CONFIG;
3339 /*get DCB TX configuration parameters from rte_eth_conf*/
3340 ixgbe_dcb_tx_config(dev,dcb_config);
3341 /*Configure general DCB TX parameters*/
3342 ixgbe_dcb_tx_hw_config(hw, dcb_config);
3345 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3349 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3351 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3352 if(nb_tcs == ETH_4_TCS) {
3353 /* Avoid un-configured priority mapping to TC0 */
3355 uint8_t mask = 0xFF;
3356 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3357 mask = (uint8_t)(mask & (~ (1 << map[i])));
3358 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3359 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3363 /* Re-configure 4 TCs BW */
3364 for (i = 0; i < nb_tcs; i++) {
3365 tc = &dcb_config->tc_config[i];
3366 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
3367 (uint8_t)(100 / nb_tcs);
3368 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
3369 (uint8_t)(100 / nb_tcs);
3371 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
3372 tc = &dcb_config->tc_config[i];
3373 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
3374 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
3379 /* Set RX buffer size */
3380 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3381 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
3382 for (i = 0 ; i < nb_tcs; i++) {
3383 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3385 /* zero alloc all unused TCs */
3386 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3387 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
3391 /* Only support an equally distributed Tx packet buffer strategy. */
3392 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
3393 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
3394 for (i = 0; i < nb_tcs; i++) {
3395 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
3396 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
3398 /* Clear unused TCs, if any, to zero buffer size*/
3399 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3400 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
3401 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
3405 /*Calculates traffic class credits*/
3406 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
3407 IXGBE_DCB_TX_CONFIG);
3408 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
3409 IXGBE_DCB_RX_CONFIG);
3412 /* Unpack CEE standard containers */
3413 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
3414 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3415 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
3416 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
3417 /* Configure PG(ETS) RX */
3418 ixgbe_dcb_hw_arbite_rx_config(hw,refill,max,bwgid,tsa,map);
3422 /* Unpack CEE standard containers */
3423 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
3424 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3425 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
3426 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
3427 /* Configure PG(ETS) TX */
3428 ixgbe_dcb_hw_arbite_tx_config(hw,refill,max,bwgid,tsa,map);
3431 /*Configure queue statistics registers*/
3432 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3434 /* Check if the PFC is supported */
3435 if(dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3436 pbsize = (uint16_t) (NIC_RX_BUFFER_SIZE / nb_tcs);
3437 for (i = 0; i < nb_tcs; i++) {
3439 * If the TC count is 8,and the default high_water is 48,
3440 * the low_water is 16 as default.
3442 hw->fc.high_water[i] = (pbsize * 3 ) / 4;
3443 hw->fc.low_water[i] = pbsize / 4;
3444 /* Enable pfc for this TC */
3445 tc = &dcb_config->tc_config[i];
3446 tc->pfc = ixgbe_dcb_pfc_enabled;
3448 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3449 if(dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3451 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3458 * ixgbe_configure_dcb - Configure DCB Hardware
3459 * @dev: pointer to rte_eth_dev
3461 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3463 struct ixgbe_dcb_config *dcb_cfg =
3464 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3465 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3467 PMD_INIT_FUNC_TRACE();
3469 /* check support mq_mode for DCB */
3470 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
3471 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB))
3474 if (dev->data->nb_rx_queues != ETH_DCB_NUM_QUEUES)
3477 /** Configure DCB hardware **/
3478 ixgbe_dcb_hw_configure(dev,dcb_cfg);
3484 * VMDq only support for 10 GbE NIC.
3487 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
3489 struct rte_eth_vmdq_rx_conf *cfg;
3490 struct ixgbe_hw *hw;
3491 enum rte_eth_nb_pools num_pools;
3492 uint32_t mrqc, vt_ctl, vlanctrl;
3496 PMD_INIT_FUNC_TRACE();
3497 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3498 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
3499 num_pools = cfg->nb_queue_pools;
3501 ixgbe_rss_disable(dev);
3503 /* MRQC: enable vmdq */
3504 mrqc = IXGBE_MRQC_VMDQEN;
3505 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3507 /* PFVTCTL: turn on virtualisation and set the default pool */
3508 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3509 if (cfg->enable_default_pool)
3510 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3512 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3514 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3516 for (i = 0; i < (int)num_pools; i++) {
3517 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
3518 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
3521 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3522 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3523 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
3524 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3526 /* VFTA - enable all vlan filters */
3527 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
3528 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
3530 /* VFRE: pool enabling for receive - 64 */
3531 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
3532 if (num_pools == ETH_64_POOLS)
3533 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
3536 * MPSAR - allow pools to read specific mac addresses
3537 * In this case, all pools should be able to read from mac addr 0
3539 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
3540 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
3542 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3543 for (i = 0; i < cfg->nb_pool_maps; i++) {
3544 /* set vlan id in VF register and set the valid bit */
3545 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
3546 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
3548 * Put the allowed pools in VFB reg. As we only have 16 or 64
3549 * pools, we only need to use the first half of the register
3552 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
3553 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), \
3554 (cfg->pool_map[i].pools & UINT32_MAX));
3556 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i*2+1)), \
3557 ((cfg->pool_map[i].pools >> 32) \
3562 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
3563 if (cfg->enable_loop_back) {
3564 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3565 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
3566 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
3569 IXGBE_WRITE_FLUSH(hw);
3573 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
3574 * @hw: pointer to hardware structure
3577 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
3582 PMD_INIT_FUNC_TRACE();
3583 /*PF VF Transmit Enable*/
3584 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
3585 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
3587 /* Disable the Tx desc arbiter so that MTQC can be changed */
3588 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3589 reg |= IXGBE_RTTDCS_ARBDIS;
3590 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3592 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3593 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3595 /* Disable drop for all queues */
3596 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3597 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3598 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3600 /* Enable the Tx desc arbiter */
3601 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3602 reg &= ~IXGBE_RTTDCS_ARBDIS;
3603 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3605 IXGBE_WRITE_FLUSH(hw);
3610 static int __attribute__((cold))
3611 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
3613 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
3617 /* Initialize software ring entries */
3618 for (i = 0; i < rxq->nb_rx_desc; i++) {
3619 volatile union ixgbe_adv_rx_desc *rxd;
3620 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mb_pool);
3622 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
3623 (unsigned) rxq->queue_id);
3627 rte_mbuf_refcnt_set(mbuf, 1);
3629 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
3631 mbuf->port = rxq->port_id;
3634 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
3635 rxd = &rxq->rx_ring[i];
3636 rxd->read.hdr_addr = dma_addr;
3637 rxd->read.pkt_addr = dma_addr;
3645 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
3647 struct ixgbe_hw *hw;
3650 ixgbe_rss_configure(dev);
3652 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3654 /* MRQC: enable VF RSS */
3655 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
3656 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
3657 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3659 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
3663 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
3667 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
3671 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3677 ixgbe_config_vf_default(struct rte_eth_dev *dev)
3679 struct ixgbe_hw *hw =
3680 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3682 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3684 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3689 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3690 IXGBE_MRQC_VMDQRT4TCEN);
3694 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3695 IXGBE_MRQC_VMDQRT8TCEN);
3699 "invalid pool number in IOV mode");
3706 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
3708 struct ixgbe_hw *hw =
3709 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3711 if (hw->mac.type == ixgbe_mac_82598EB)
3714 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3716 * SRIOV inactive scheme
3717 * any DCB/RSS w/o VMDq multi-queue setting
3719 switch (dev->data->dev_conf.rxmode.mq_mode) {
3721 ixgbe_rss_configure(dev);
3724 case ETH_MQ_RX_VMDQ_DCB:
3725 ixgbe_vmdq_dcb_configure(dev);
3728 case ETH_MQ_RX_VMDQ_ONLY:
3729 ixgbe_vmdq_rx_hw_configure(dev);
3732 case ETH_MQ_RX_NONE:
3733 /* if mq_mode is none, disable rss mode.*/
3734 default: ixgbe_rss_disable(dev);
3738 * SRIOV active scheme
3739 * Support RSS together with VMDq & SRIOV
3741 switch (dev->data->dev_conf.rxmode.mq_mode) {
3743 case ETH_MQ_RX_VMDQ_RSS:
3744 ixgbe_config_vf_rss(dev);
3747 /* FIXME if support DCB/RSS together with VMDq & SRIOV */
3748 case ETH_MQ_RX_VMDQ_DCB:
3749 case ETH_MQ_RX_VMDQ_DCB_RSS:
3751 "Could not support DCB with VMDq & SRIOV");
3754 ixgbe_config_vf_default(dev);
3763 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
3765 struct ixgbe_hw *hw =
3766 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3770 if (hw->mac.type == ixgbe_mac_82598EB)
3773 /* disable arbiter before setting MTQC */
3774 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3775 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3776 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3778 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3780 * SRIOV inactive scheme
3781 * any DCB w/o VMDq multi-queue setting
3783 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
3784 ixgbe_vmdq_tx_hw_configure(hw);
3786 mtqc = IXGBE_MTQC_64Q_1PB;
3787 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3790 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3793 * SRIOV active scheme
3794 * FIXME if support DCB together with VMDq & SRIOV
3797 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3800 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
3803 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
3807 mtqc = IXGBE_MTQC_64Q_1PB;
3808 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
3810 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3813 /* re-enable arbiter */
3814 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3815 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3821 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
3823 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
3824 * spec rev. 3.0 chapter 8.2.3.8.13.
3826 * @pool Memory pool of the Rx queue
3828 static inline uint32_t
3829 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
3831 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
3833 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
3836 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
3839 return IXGBE_RSCCTL_MAXDESC_16;
3840 else if (maxdesc >= 8)
3841 return IXGBE_RSCCTL_MAXDESC_8;
3842 else if (maxdesc >= 4)
3843 return IXGBE_RSCCTL_MAXDESC_4;
3845 return IXGBE_RSCCTL_MAXDESC_1;
3849 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
3852 * (Taken from FreeBSD tree)
3853 * (yes this is all very magic and confusing :)
3856 * @entry the register array entry
3857 * @vector the MSIX vector for this queue
3861 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
3863 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3866 vector |= IXGBE_IVAR_ALLOC_VAL;
3868 switch (hw->mac.type) {
3870 case ixgbe_mac_82598EB:
3872 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
3874 entry += (type * 64);
3875 index = (entry >> 2) & 0x1F;
3876 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
3877 ivar &= ~(0xFF << (8 * (entry & 0x3)));
3878 ivar |= (vector << (8 * (entry & 0x3)));
3879 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
3882 case ixgbe_mac_82599EB:
3883 case ixgbe_mac_X540:
3884 if (type == -1) { /* MISC IVAR */
3885 index = (entry & 1) * 8;
3886 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
3887 ivar &= ~(0xFF << index);
3888 ivar |= (vector << index);
3889 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
3890 } else { /* RX/TX IVARS */
3891 index = (16 * (entry & 1)) + (8 * type);
3892 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
3893 ivar &= ~(0xFF << index);
3894 ivar |= (vector << index);
3895 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
3905 void __attribute__((cold))
3906 ixgbe_set_rx_function(struct rte_eth_dev *dev)
3908 uint16_t i, rx_using_sse;
3909 struct ixgbe_adapter *adapter =
3910 (struct ixgbe_adapter *)dev->data->dev_private;
3913 * In order to allow Vector Rx there are a few configuration
3914 * conditions to be met and Rx Bulk Allocation should be allowed.
3916 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
3917 !adapter->rx_bulk_alloc_allowed) {
3918 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
3919 "preconditions or RTE_IXGBE_INC_VECTOR is "
3921 dev->data->port_id);
3923 adapter->rx_vec_allowed = false;
3927 * Initialize the appropriate LRO callback.
3929 * If all queues satisfy the bulk allocation preconditions
3930 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
3931 * Otherwise use a single allocation version.
3933 if (dev->data->lro) {
3934 if (adapter->rx_bulk_alloc_allowed) {
3935 PMD_INIT_LOG(INFO, "LRO is requested. Using a bulk "
3936 "allocation version");
3937 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
3939 PMD_INIT_LOG(INFO, "LRO is requested. Using a single "
3940 "allocation version");
3941 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
3943 } else if (dev->data->scattered_rx) {
3945 * Set the non-LRO scattered callback: there are Vector and
3946 * single allocation versions.
3948 if (adapter->rx_vec_allowed) {
3949 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
3950 "callback (port=%d).",
3951 dev->data->port_id);
3953 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
3954 } else if (adapter->rx_bulk_alloc_allowed) {
3955 PMD_INIT_LOG(INFO, "Using a Scattered with bulk "
3956 "allocation callback (port=%d).",
3957 dev->data->port_id);
3958 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
3960 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
3961 "single allocation) "
3962 "Scattered Rx callback "
3964 dev->data->port_id);
3966 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
3969 * Below we set "simple" callbacks according to port/queues parameters.
3970 * If parameters allow we are going to choose between the following
3974 * - Single buffer allocation (the simplest one)
3976 } else if (adapter->rx_vec_allowed) {
3977 PMD_INIT_LOG(INFO, "Vector rx enabled, please make sure RX "
3978 "burst size no less than 32.");
3980 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
3981 } else if (adapter->rx_bulk_alloc_allowed) {
3982 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
3983 "satisfied. Rx Burst Bulk Alloc function "
3984 "will be used on port=%d.",
3985 dev->data->port_id);
3987 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
3989 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
3990 "satisfied, or Scattered Rx is requested "
3992 dev->data->port_id);
3994 dev->rx_pkt_burst = ixgbe_recv_pkts;
3997 /* Propagate information about RX function choice through all queues. */
4000 (dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec ||
4001 dev->rx_pkt_burst == ixgbe_recv_pkts_vec);
4003 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4004 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4005 rxq->rx_using_sse = rx_using_sse;
4010 * ixgbe_set_rsc - configure RSC related port HW registers
4012 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4013 * of 82599 Spec (x540 configuration is virtually the same).
4017 * Returns 0 in case of success or a non-zero error code
4020 ixgbe_set_rsc(struct rte_eth_dev *dev)
4022 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4023 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4024 struct rte_eth_dev_info dev_info = { 0 };
4025 bool rsc_capable = false;
4030 dev->dev_ops->dev_infos_get(dev, &dev_info);
4031 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4034 if (!rsc_capable && rx_conf->enable_lro) {
4035 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4040 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4042 if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
4044 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4045 * 3.0 RSC configuration requires HW CRC stripping being
4046 * enabled. If user requested both HW CRC stripping off
4047 * and RSC on - return an error.
4049 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4054 /* RFCTL configuration */
4056 uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4057 if (rx_conf->enable_lro)
4059 * Since NFS packets coalescing is not supported - clear
4060 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4063 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4064 IXGBE_RFCTL_NFSR_DIS);
4066 rfctl |= IXGBE_RFCTL_RSC_DIS;
4068 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4071 /* If LRO hasn't been requested - we are done here. */
4072 if (!rx_conf->enable_lro)
4075 /* Set RDRXCTL.RSCACKC bit */
4076 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4077 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4078 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4080 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4081 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4082 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4084 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4086 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4088 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4090 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4093 * ixgbe PMD doesn't support header-split at the moment.
4095 * Following the 4.6.7.2.1 chapter of the 82599/x540
4096 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4097 * should be configured even if header split is not
4098 * enabled. We will configure it 128 bytes following the
4099 * recommendation in the spec.
4101 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4102 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4103 IXGBE_SRRCTL_BSIZEHDR_MASK;
4106 * TODO: Consider setting the Receive Descriptor Minimum
4107 * Threshold Size for an RSC case. This is not an obviously
4108 * beneficiary option but the one worth considering...
4111 rscctl |= IXGBE_RSCCTL_RSCEN;
4112 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4113 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4116 * RSC: Set ITR interval corresponding to 2K ints/s.
4118 * Full-sized RSC aggregations for a 10Gb/s link will
4119 * arrive at about 20K aggregation/s rate.
4121 * 2K inst/s rate will make only 10% of the
4122 * aggregations to be closed due to the interrupt timer
4123 * expiration for a streaming at wire-speed case.
4125 * For a sparse streaming case this setting will yield
4126 * at most 500us latency for a single RSC aggregation.
4128 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4129 eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
4131 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4132 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4133 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4134 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4137 * RSC requires the mapping of the queue to the
4140 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4145 PMD_INIT_LOG(INFO, "enabling LRO mode");
4151 * Initializes Receive Unit.
4153 int __attribute__((cold))
4154 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4156 struct ixgbe_hw *hw;
4157 struct ixgbe_rx_queue *rxq;
4168 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4171 PMD_INIT_FUNC_TRACE();
4172 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4175 * Make sure receives are disabled while setting
4176 * up the RX context (registers, descriptor rings, etc.).
4178 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4179 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4181 /* Enable receipt of broadcasted frames */
4182 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4183 fctrl |= IXGBE_FCTRL_BAM;
4184 fctrl |= IXGBE_FCTRL_DPF;
4185 fctrl |= IXGBE_FCTRL_PMCF;
4186 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4189 * Configure CRC stripping, if any.
4191 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4192 if (rx_conf->hw_strip_crc)
4193 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4195 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4198 * Configure jumbo frame support, if any.
4200 if (rx_conf->jumbo_frame == 1) {
4201 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4202 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4203 maxfrs &= 0x0000FFFF;
4204 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4205 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4207 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4210 * If loopback mode is configured for 82599, set LPBK bit.
4212 if (hw->mac.type == ixgbe_mac_82599EB &&
4213 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4214 hlreg0 |= IXGBE_HLREG0_LPBK;
4216 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4218 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4220 /* Setup RX queues */
4221 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4222 rxq = dev->data->rx_queues[i];
4225 * Reset crc_len in case it was changed after queue setup by a
4226 * call to configure.
4228 rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
4230 /* Setup the Base and Length of the Rx Descriptor Rings */
4231 bus_addr = rxq->rx_ring_phys_addr;
4232 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4233 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4234 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4235 (uint32_t)(bus_addr >> 32));
4236 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4237 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4238 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4239 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4241 /* Configure the SRRCTL register */
4242 #ifdef RTE_HEADER_SPLIT_ENABLE
4244 * Configure Header Split
4246 if (rx_conf->header_split) {
4247 if (hw->mac.type == ixgbe_mac_82599EB) {
4248 /* Must setup the PSRTYPE register */
4250 psrtype = IXGBE_PSRTYPE_TCPHDR |
4251 IXGBE_PSRTYPE_UDPHDR |
4252 IXGBE_PSRTYPE_IPV4HDR |
4253 IXGBE_PSRTYPE_IPV6HDR;
4254 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4256 srrctl = ((rx_conf->split_hdr_size <<
4257 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4258 IXGBE_SRRCTL_BSIZEHDR_MASK);
4259 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4262 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4264 /* Set if packets are dropped when no descriptors available */
4266 srrctl |= IXGBE_SRRCTL_DROP_EN;
4269 * Configure the RX buffer size in the BSIZEPACKET field of
4270 * the SRRCTL register of the queue.
4271 * The value is in 1 KB resolution. Valid values can be from
4274 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4275 RTE_PKTMBUF_HEADROOM);
4276 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4277 IXGBE_SRRCTL_BSIZEPKT_MASK);
4279 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4281 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4282 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4284 /* It adds dual VLAN length for supporting dual VLAN */
4285 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4286 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4287 dev->data->scattered_rx = 1;
4290 if (rx_conf->enable_scatter)
4291 dev->data->scattered_rx = 1;
4294 * Device configured with multiple RX queues.
4296 ixgbe_dev_mq_rx_configure(dev);
4299 * Setup the Checksum Register.
4300 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4301 * Enable IP/L4 checkum computation by hardware if requested to do so.
4303 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4304 rxcsum |= IXGBE_RXCSUM_PCSD;
4305 if (rx_conf->hw_ip_checksum)
4306 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4308 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4310 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4312 if (hw->mac.type == ixgbe_mac_82599EB ||
4313 hw->mac.type == ixgbe_mac_X540) {
4314 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4315 if (rx_conf->hw_strip_crc)
4316 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4318 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4319 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4320 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4323 rc = ixgbe_set_rsc(dev);
4327 ixgbe_set_rx_function(dev);
4333 * Initializes Transmit Unit.
4335 void __attribute__((cold))
4336 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4338 struct ixgbe_hw *hw;
4339 struct ixgbe_tx_queue *txq;
4345 PMD_INIT_FUNC_TRACE();
4346 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4348 /* Enable TX CRC (checksum offload requirement) and hw padding
4349 * (TSO requirement) */
4350 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4351 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
4352 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4354 /* Setup the Base and Length of the Tx Descriptor Rings */
4355 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4356 txq = dev->data->tx_queues[i];
4358 bus_addr = txq->tx_ring_phys_addr;
4359 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
4360 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4361 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
4362 (uint32_t)(bus_addr >> 32));
4363 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
4364 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4365 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4366 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4367 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4370 * Disable Tx Head Writeback RO bit, since this hoses
4371 * bookkeeping if things aren't delivered in order.
4373 switch (hw->mac.type) {
4374 case ixgbe_mac_82598EB:
4375 txctrl = IXGBE_READ_REG(hw,
4376 IXGBE_DCA_TXCTRL(txq->reg_idx));
4377 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4378 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
4382 case ixgbe_mac_82599EB:
4383 case ixgbe_mac_X540:
4384 case ixgbe_mac_X550:
4385 case ixgbe_mac_X550EM_x:
4387 txctrl = IXGBE_READ_REG(hw,
4388 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
4389 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4390 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
4396 /* Device configured with multiple TX queues. */
4397 ixgbe_dev_mq_tx_configure(dev);
4401 * Set up link for 82599 loopback mode Tx->Rx.
4403 static inline void __attribute__((cold))
4404 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
4406 PMD_INIT_FUNC_TRACE();
4408 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
4409 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
4411 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
4420 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
4421 ixgbe_reset_pipeline_82599(hw);
4423 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
4429 * Start Transmit and Receive Units.
4431 int __attribute__((cold))
4432 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
4434 struct ixgbe_hw *hw;
4435 struct ixgbe_tx_queue *txq;
4436 struct ixgbe_rx_queue *rxq;
4443 PMD_INIT_FUNC_TRACE();
4444 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4446 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4447 txq = dev->data->tx_queues[i];
4448 /* Setup Transmit Threshold Registers */
4449 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4450 txdctl |= txq->pthresh & 0x7F;
4451 txdctl |= ((txq->hthresh & 0x7F) << 8);
4452 txdctl |= ((txq->wthresh & 0x7F) << 16);
4453 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4456 if (hw->mac.type != ixgbe_mac_82598EB) {
4457 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
4458 dmatxctl |= IXGBE_DMATXCTL_TE;
4459 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
4462 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4463 txq = dev->data->tx_queues[i];
4464 if (!txq->tx_deferred_start) {
4465 ret = ixgbe_dev_tx_queue_start(dev, i);
4471 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4472 rxq = dev->data->rx_queues[i];
4473 if (!rxq->rx_deferred_start) {
4474 ret = ixgbe_dev_rx_queue_start(dev, i);
4480 /* Enable Receive engine */
4481 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4482 if (hw->mac.type == ixgbe_mac_82598EB)
4483 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4484 rxctrl |= IXGBE_RXCTRL_RXEN;
4485 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4487 /* If loopback mode is enabled for 82599, set up the link accordingly */
4488 if (hw->mac.type == ixgbe_mac_82599EB &&
4489 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4490 ixgbe_setup_loopback_link_82599(hw);
4496 * Start Receive Units for specified queue.
4498 int __attribute__((cold))
4499 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4501 struct ixgbe_hw *hw;
4502 struct ixgbe_rx_queue *rxq;
4506 PMD_INIT_FUNC_TRACE();
4507 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4509 if (rx_queue_id < dev->data->nb_rx_queues) {
4510 rxq = dev->data->rx_queues[rx_queue_id];
4512 /* Allocate buffers for descriptor rings */
4513 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
4514 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
4518 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4519 rxdctl |= IXGBE_RXDCTL_ENABLE;
4520 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4522 /* Wait until RX Enable ready */
4523 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4526 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4527 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4529 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
4532 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4533 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
4541 * Stop Receive Units for specified queue.
4543 int __attribute__((cold))
4544 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4546 struct ixgbe_hw *hw;
4547 struct ixgbe_adapter *adapter =
4548 (struct ixgbe_adapter *)dev->data->dev_private;
4549 struct ixgbe_rx_queue *rxq;
4553 PMD_INIT_FUNC_TRACE();
4554 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4556 if (rx_queue_id < dev->data->nb_rx_queues) {
4557 rxq = dev->data->rx_queues[rx_queue_id];
4559 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4560 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4561 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4563 /* Wait until RX Enable ready */
4564 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4567 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4568 } while (--poll_ms && (rxdctl | IXGBE_RXDCTL_ENABLE));
4570 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
4573 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4575 ixgbe_rx_queue_release_mbufs(rxq);
4576 ixgbe_reset_rx_queue(adapter, rxq);
4585 * Start Transmit Units for specified queue.
4587 int __attribute__((cold))
4588 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4590 struct ixgbe_hw *hw;
4591 struct ixgbe_tx_queue *txq;
4595 PMD_INIT_FUNC_TRACE();
4596 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4598 if (tx_queue_id < dev->data->nb_tx_queues) {
4599 txq = dev->data->tx_queues[tx_queue_id];
4600 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4601 txdctl |= IXGBE_TXDCTL_ENABLE;
4602 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4604 /* Wait until TX Enable ready */
4605 if (hw->mac.type == ixgbe_mac_82599EB) {
4606 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4609 txdctl = IXGBE_READ_REG(hw,
4610 IXGBE_TXDCTL(txq->reg_idx));
4611 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4613 PMD_INIT_LOG(ERR, "Could not enable "
4614 "Tx Queue %d", tx_queue_id);
4617 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4618 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4626 * Stop Transmit Units for specified queue.
4628 int __attribute__((cold))
4629 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4631 struct ixgbe_hw *hw;
4632 struct ixgbe_tx_queue *txq;
4634 uint32_t txtdh, txtdt;
4637 PMD_INIT_FUNC_TRACE();
4638 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4640 if (tx_queue_id < dev->data->nb_tx_queues) {
4641 txq = dev->data->tx_queues[tx_queue_id];
4643 /* Wait until TX queue is empty */
4644 if (hw->mac.type == ixgbe_mac_82599EB) {
4645 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4647 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4648 txtdh = IXGBE_READ_REG(hw,
4649 IXGBE_TDH(txq->reg_idx));
4650 txtdt = IXGBE_READ_REG(hw,
4651 IXGBE_TDT(txq->reg_idx));
4652 } while (--poll_ms && (txtdh != txtdt));
4654 PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
4655 "when stopping.", tx_queue_id);
4658 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4659 txdctl &= ~IXGBE_TXDCTL_ENABLE;
4660 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4662 /* Wait until TX Enable ready */
4663 if (hw->mac.type == ixgbe_mac_82599EB) {
4664 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4667 txdctl = IXGBE_READ_REG(hw,
4668 IXGBE_TXDCTL(txq->reg_idx));
4669 } while (--poll_ms && (txdctl | IXGBE_TXDCTL_ENABLE));
4671 PMD_INIT_LOG(ERR, "Could not disable "
4672 "Tx Queue %d", tx_queue_id);
4675 if (txq->ops != NULL) {
4676 txq->ops->release_mbufs(txq);
4677 txq->ops->reset(txq);
4686 * [VF] Initializes Receive Unit.
4688 int __attribute__((cold))
4689 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
4691 struct ixgbe_hw *hw;
4692 struct ixgbe_rx_queue *rxq;
4694 uint32_t srrctl, psrtype = 0;
4699 PMD_INIT_FUNC_TRACE();
4700 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4702 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
4703 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
4704 "it should be power of 2");
4708 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
4709 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
4710 "it should be equal to or less than %d",
4711 hw->mac.max_rx_queues);
4716 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
4717 * disables the VF receipt of packets if the PF MTU is > 1500.
4718 * This is done to deal with 82599 limitations that imposes
4719 * the PF and all VFs to share the same MTU.
4720 * Then, the PF driver enables again the VF receipt of packet when
4721 * the VF driver issues a IXGBE_VF_SET_LPE request.
4722 * In the meantime, the VF device cannot be used, even if the VF driver
4723 * and the Guest VM network stack are ready to accept packets with a
4724 * size up to the PF MTU.
4725 * As a work-around to this PF behaviour, force the call to
4726 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
4727 * VF packets received can work in all cases.
4729 ixgbevf_rlpml_set_vf(hw,
4730 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
4732 /* Setup RX queues */
4733 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4734 rxq = dev->data->rx_queues[i];
4736 /* Allocate buffers for descriptor rings */
4737 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
4741 /* Setup the Base and Length of the Rx Descriptor Rings */
4742 bus_addr = rxq->rx_ring_phys_addr;
4744 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
4745 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4746 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
4747 (uint32_t)(bus_addr >> 32));
4748 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
4749 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4750 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
4751 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
4754 /* Configure the SRRCTL register */
4755 #ifdef RTE_HEADER_SPLIT_ENABLE
4757 * Configure Header Split
4759 if (dev->data->dev_conf.rxmode.header_split) {
4760 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
4761 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4762 IXGBE_SRRCTL_BSIZEHDR_MASK);
4763 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4766 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4768 /* Set if packets are dropped when no descriptors available */
4770 srrctl |= IXGBE_SRRCTL_DROP_EN;
4773 * Configure the RX buffer size in the BSIZEPACKET field of
4774 * the SRRCTL register of the queue.
4775 * The value is in 1 KB resolution. Valid values can be from
4778 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4779 RTE_PKTMBUF_HEADROOM);
4780 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4781 IXGBE_SRRCTL_BSIZEPKT_MASK);
4784 * VF modification to write virtual function SRRCTL register
4786 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
4788 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4789 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4791 if (dev->data->dev_conf.rxmode.enable_scatter ||
4792 /* It adds dual VLAN length for supporting dual VLAN */
4793 (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4794 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
4795 if (!dev->data->scattered_rx)
4796 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
4797 dev->data->scattered_rx = 1;
4801 #ifdef RTE_HEADER_SPLIT_ENABLE
4802 if (dev->data->dev_conf.rxmode.header_split)
4803 /* Must setup the PSRTYPE register */
4804 psrtype = IXGBE_PSRTYPE_TCPHDR |
4805 IXGBE_PSRTYPE_UDPHDR |
4806 IXGBE_PSRTYPE_IPV4HDR |
4807 IXGBE_PSRTYPE_IPV6HDR;
4810 /* Set RQPL for VF RSS according to max Rx queue */
4811 psrtype |= (dev->data->nb_rx_queues >> 1) <<
4812 IXGBE_PSRTYPE_RQPL_SHIFT;
4813 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
4815 ixgbe_set_rx_function(dev);
4821 * [VF] Initializes Transmit Unit.
4823 void __attribute__((cold))
4824 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
4826 struct ixgbe_hw *hw;
4827 struct ixgbe_tx_queue *txq;
4832 PMD_INIT_FUNC_TRACE();
4833 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4835 /* Setup the Base and Length of the Tx Descriptor Rings */
4836 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4837 txq = dev->data->tx_queues[i];
4838 bus_addr = txq->tx_ring_phys_addr;
4839 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
4840 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4841 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
4842 (uint32_t)(bus_addr >> 32));
4843 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
4844 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4845 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4846 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
4847 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
4850 * Disable Tx Head Writeback RO bit, since this hoses
4851 * bookkeeping if things aren't delivered in order.
4853 txctrl = IXGBE_READ_REG(hw,
4854 IXGBE_VFDCA_TXCTRL(i));
4855 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4856 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
4862 * [VF] Start Transmit and Receive Units.
4864 void __attribute__((cold))
4865 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
4867 struct ixgbe_hw *hw;
4868 struct ixgbe_tx_queue *txq;
4869 struct ixgbe_rx_queue *rxq;
4875 PMD_INIT_FUNC_TRACE();
4876 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4878 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4879 txq = dev->data->tx_queues[i];
4880 /* Setup Transmit Threshold Registers */
4881 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4882 txdctl |= txq->pthresh & 0x7F;
4883 txdctl |= ((txq->hthresh & 0x7F) << 8);
4884 txdctl |= ((txq->wthresh & 0x7F) << 16);
4885 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
4888 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4890 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4891 txdctl |= IXGBE_TXDCTL_ENABLE;
4892 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
4895 /* Wait until TX Enable ready */
4898 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4899 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4901 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
4903 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4905 rxq = dev->data->rx_queues[i];
4907 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
4908 rxdctl |= IXGBE_RXDCTL_ENABLE;
4909 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
4911 /* Wait until RX Enable ready */
4915 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
4916 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4918 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
4920 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
4925 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
4926 int __attribute__((weak))
4927 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
4932 uint16_t __attribute__((weak))
4933 ixgbe_recv_pkts_vec(
4934 void __rte_unused *rx_queue,
4935 struct rte_mbuf __rte_unused **rx_pkts,
4936 uint16_t __rte_unused nb_pkts)
4941 uint16_t __attribute__((weak))
4942 ixgbe_recv_scattered_pkts_vec(
4943 void __rte_unused *rx_queue,
4944 struct rte_mbuf __rte_unused **rx_pkts,
4945 uint16_t __rte_unused nb_pkts)
4950 int __attribute__((weak))
4951 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)