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];
1037 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
1039 * LOOK_AHEAD defines how many desc statuses to check beyond the
1040 * current descriptor.
1041 * It must be a pound define for optimal performance.
1042 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1043 * function only works with LOOK_AHEAD=8.
1045 #define LOOK_AHEAD 8
1046 #if (LOOK_AHEAD != 8)
1047 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1050 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1052 volatile union ixgbe_adv_rx_desc *rxdp;
1053 struct ixgbe_rx_entry *rxep;
1054 struct rte_mbuf *mb;
1059 uint32_t s[LOOK_AHEAD];
1060 uint16_t pkt_info[LOOK_AHEAD];
1062 int s[LOOK_AHEAD], nb_dd;
1063 #endif /* RTE_NEXT_ABI */
1064 int i, j, nb_rx = 0;
1067 /* get references to current descriptor and S/W ring entry */
1068 rxdp = &rxq->rx_ring[rxq->rx_tail];
1069 rxep = &rxq->sw_ring[rxq->rx_tail];
1071 /* check to make sure there is at least 1 packet to receive */
1072 if (! (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD))
1076 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1077 * reference packets that are ready to be received.
1079 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1080 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD)
1082 /* Read desc statuses backwards to avoid race condition */
1083 for (j = LOOK_AHEAD-1; j >= 0; --j)
1084 s[j] = rxdp[j].wb.upper.status_error;
1087 for (j = LOOK_AHEAD - 1; j >= 0; --j)
1088 pkt_info[j] = rxdp[j].wb.lower.lo_dword.
1090 #endif /* RTE_NEXT_ABI */
1092 /* Compute how many status bits were set */
1094 for (j = 0; j < LOOK_AHEAD; ++j)
1095 nb_dd += s[j] & IXGBE_RXDADV_STAT_DD;
1099 /* Translate descriptor info to mbuf format */
1100 for (j = 0; j < nb_dd; ++j) {
1102 pkt_len = (uint16_t)(rxdp[j].wb.upper.length - rxq->crc_len);
1103 mb->data_len = pkt_len;
1104 mb->pkt_len = pkt_len;
1105 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1107 /* convert descriptor fields to rte mbuf flags */
1109 pkt_flags = rx_desc_status_to_pkt_flags(s[j]);
1110 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1112 ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info[j]);
1113 mb->ol_flags = pkt_flags;
1115 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info[j]);
1116 #else /* RTE_NEXT_ABI */
1117 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(
1118 rxdp[j].wb.lower.lo_dword.data);
1119 /* reuse status field from scan list */
1120 pkt_flags |= rx_desc_status_to_pkt_flags(s[j]);
1121 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1122 mb->ol_flags = pkt_flags;
1123 #endif /* RTE_NEXT_ABI */
1125 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1126 mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
1127 else if (pkt_flags & PKT_RX_FDIR) {
1128 mb->hash.fdir.hash =
1129 (uint16_t)((rxdp[j].wb.lower.hi_dword.csum_ip.csum)
1130 & IXGBE_ATR_HASH_MASK);
1131 mb->hash.fdir.id = rxdp[j].wb.lower.hi_dword.csum_ip.ip_id;
1135 /* Move mbuf pointers from the S/W ring to the stage */
1136 for (j = 0; j < LOOK_AHEAD; ++j) {
1137 rxq->rx_stage[i + j] = rxep[j].mbuf;
1140 /* stop if all requested packets could not be received */
1141 if (nb_dd != LOOK_AHEAD)
1145 /* clear software ring entries so we can cleanup correctly */
1146 for (i = 0; i < nb_rx; ++i) {
1147 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1155 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1157 volatile union ixgbe_adv_rx_desc *rxdp;
1158 struct ixgbe_rx_entry *rxep;
1159 struct rte_mbuf *mb;
1164 /* allocate buffers in bulk directly into the S/W ring */
1165 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1166 rxep = &rxq->sw_ring[alloc_idx];
1167 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1168 rxq->rx_free_thresh);
1169 if (unlikely(diag != 0))
1172 rxdp = &rxq->rx_ring[alloc_idx];
1173 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1174 /* populate the static rte mbuf fields */
1179 mb->port = rxq->port_id;
1182 rte_mbuf_refcnt_set(mb, 1);
1183 mb->data_off = RTE_PKTMBUF_HEADROOM;
1185 /* populate the descriptors */
1186 dma_addr = rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
1187 rxdp[i].read.hdr_addr = dma_addr;
1188 rxdp[i].read.pkt_addr = dma_addr;
1191 /* update state of internal queue structure */
1192 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1193 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1194 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1200 static inline uint16_t
1201 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1204 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1207 /* how many packets are ready to return? */
1208 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1210 /* copy mbuf pointers to the application's packet list */
1211 for (i = 0; i < nb_pkts; ++i)
1212 rx_pkts[i] = stage[i];
1214 /* update internal queue state */
1215 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1216 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1221 static inline uint16_t
1222 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1225 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1228 /* Any previously recv'd pkts will be returned from the Rx stage */
1229 if (rxq->rx_nb_avail)
1230 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1232 /* Scan the H/W ring for packets to receive */
1233 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1235 /* update internal queue state */
1236 rxq->rx_next_avail = 0;
1237 rxq->rx_nb_avail = nb_rx;
1238 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1240 /* if required, allocate new buffers to replenish descriptors */
1241 if (rxq->rx_tail > rxq->rx_free_trigger) {
1242 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1244 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1246 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1247 "queue_id=%u", (unsigned) rxq->port_id,
1248 (unsigned) rxq->queue_id);
1250 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1251 rxq->rx_free_thresh;
1254 * Need to rewind any previous receives if we cannot
1255 * allocate new buffers to replenish the old ones.
1257 rxq->rx_nb_avail = 0;
1258 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1259 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1260 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1265 /* update tail pointer */
1267 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, cur_free_trigger);
1270 if (rxq->rx_tail >= rxq->nb_rx_desc)
1273 /* received any packets this loop? */
1274 if (rxq->rx_nb_avail)
1275 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1280 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1282 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1287 if (unlikely(nb_pkts == 0))
1290 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1291 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1293 /* request is relatively large, chunk it up */
1297 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1298 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1299 nb_rx = (uint16_t)(nb_rx + ret);
1300 nb_pkts = (uint16_t)(nb_pkts - ret);
1310 /* Stub to avoid extra ifdefs */
1312 ixgbe_recv_pkts_bulk_alloc(__rte_unused void *rx_queue,
1313 __rte_unused struct rte_mbuf **rx_pkts, __rte_unused uint16_t nb_pkts)
1319 ixgbe_rx_alloc_bufs(__rte_unused struct ixgbe_rx_queue *rxq,
1320 __rte_unused bool reset_mbuf)
1324 #endif /* RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC */
1327 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1330 struct ixgbe_rx_queue *rxq;
1331 volatile union ixgbe_adv_rx_desc *rx_ring;
1332 volatile union ixgbe_adv_rx_desc *rxdp;
1333 struct ixgbe_rx_entry *sw_ring;
1334 struct ixgbe_rx_entry *rxe;
1335 struct rte_mbuf *rxm;
1336 struct rte_mbuf *nmb;
1337 union ixgbe_adv_rx_desc rxd;
1343 uint32_t hlen_type_rss;
1354 rx_id = rxq->rx_tail;
1355 rx_ring = rxq->rx_ring;
1356 sw_ring = rxq->sw_ring;
1357 while (nb_rx < nb_pkts) {
1359 * The order of operations here is important as the DD status
1360 * bit must not be read after any other descriptor fields.
1361 * rx_ring and rxdp are pointing to volatile data so the order
1362 * of accesses cannot be reordered by the compiler. If they were
1363 * not volatile, they could be reordered which could lead to
1364 * using invalid descriptor fields when read from rxd.
1366 rxdp = &rx_ring[rx_id];
1367 staterr = rxdp->wb.upper.status_error;
1368 if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1375 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1376 * is likely to be invalid and to be dropped by the various
1377 * validation checks performed by the network stack.
1379 * Allocate a new mbuf to replenish the RX ring descriptor.
1380 * If the allocation fails:
1381 * - arrange for that RX descriptor to be the first one
1382 * being parsed the next time the receive function is
1383 * invoked [on the same queue].
1385 * - Stop parsing the RX ring and return immediately.
1387 * This policy do not drop the packet received in the RX
1388 * descriptor for which the allocation of a new mbuf failed.
1389 * Thus, it allows that packet to be later retrieved if
1390 * mbuf have been freed in the mean time.
1391 * As a side effect, holding RX descriptors instead of
1392 * systematically giving them back to the NIC may lead to
1393 * RX ring exhaustion situations.
1394 * However, the NIC can gracefully prevent such situations
1395 * to happen by sending specific "back-pressure" flow control
1396 * frames to its peer(s).
1398 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1399 "ext_err_stat=0x%08x pkt_len=%u",
1400 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1401 (unsigned) rx_id, (unsigned) staterr,
1402 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1404 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1406 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1407 "queue_id=%u", (unsigned) rxq->port_id,
1408 (unsigned) rxq->queue_id);
1409 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1414 rxe = &sw_ring[rx_id];
1416 if (rx_id == rxq->nb_rx_desc)
1419 /* Prefetch next mbuf while processing current one. */
1420 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1423 * When next RX descriptor is on a cache-line boundary,
1424 * prefetch the next 4 RX descriptors and the next 8 pointers
1427 if ((rx_id & 0x3) == 0) {
1428 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1429 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1435 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1436 rxdp->read.hdr_addr = dma_addr;
1437 rxdp->read.pkt_addr = dma_addr;
1440 * Initialize the returned mbuf.
1441 * 1) setup generic mbuf fields:
1442 * - number of segments,
1445 * - RX port identifier.
1446 * 2) integrate hardware offload data, if any:
1447 * - RSS flag & hash,
1448 * - IP checksum flag,
1449 * - VLAN TCI, if any,
1452 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1454 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1455 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1458 rxm->pkt_len = pkt_len;
1459 rxm->data_len = pkt_len;
1460 rxm->port = rxq->port_id;
1463 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.hs_rss.
1465 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1466 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1468 pkt_flags = rx_desc_status_to_pkt_flags(staterr);
1469 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1470 pkt_flags = pkt_flags |
1471 ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
1472 rxm->ol_flags = pkt_flags;
1473 rxm->packet_type = ixgbe_rxd_pkt_info_to_pkt_type(pkt_info);
1474 #else /* RTE_NEXT_ABI */
1475 hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1476 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1477 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1479 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1480 pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
1481 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1482 rxm->ol_flags = pkt_flags;
1483 #endif /* RTE_NEXT_ABI */
1485 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1486 rxm->hash.rss = rxd.wb.lower.hi_dword.rss;
1487 else if (pkt_flags & PKT_RX_FDIR) {
1488 rxm->hash.fdir.hash =
1489 (uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
1490 & IXGBE_ATR_HASH_MASK);
1491 rxm->hash.fdir.id = rxd.wb.lower.hi_dword.csum_ip.ip_id;
1494 * Store the mbuf address into the next entry of the array
1495 * of returned packets.
1497 rx_pkts[nb_rx++] = rxm;
1499 rxq->rx_tail = rx_id;
1502 * If the number of free RX descriptors is greater than the RX free
1503 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1505 * Update the RDT with the value of the last processed RX descriptor
1506 * minus 1, to guarantee that the RDT register is never equal to the
1507 * RDH register, which creates a "full" ring situtation from the
1508 * hardware point of view...
1510 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1511 if (nb_hold > rxq->rx_free_thresh) {
1512 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1513 "nb_hold=%u nb_rx=%u",
1514 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1515 (unsigned) rx_id, (unsigned) nb_hold,
1517 rx_id = (uint16_t) ((rx_id == 0) ?
1518 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1519 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1522 rxq->nb_rx_hold = nb_hold;
1527 * Detect an RSC descriptor.
1529 static inline uint32_t
1530 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1532 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1533 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1537 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1539 * Fill the following info in the HEAD buffer of the Rx cluster:
1540 * - RX port identifier
1541 * - hardware offload data, if any:
1543 * - IP checksum flag
1544 * - VLAN TCI, if any
1546 * @head HEAD of the packet cluster
1547 * @desc HW descriptor to get data from
1548 * @port_id Port ID of the Rx queue
1551 ixgbe_fill_cluster_head_buf(
1552 struct rte_mbuf *head,
1553 union ixgbe_adv_rx_desc *desc,
1561 head->port = port_id;
1563 /* The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1564 * set in the pkt_flags field.
1566 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1567 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.hs_rss.pkt_info);
1568 pkt_flags = rx_desc_status_to_pkt_flags(staterr);
1569 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1570 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
1571 head->ol_flags = pkt_flags;
1572 head->packet_type = ixgbe_rxd_pkt_info_to_pkt_type(pkt_info);
1573 #else /* RTE_NEXT_ABI */
1574 uint32_t hlen_type_rss;
1577 head->port = port_id;
1580 * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1581 * set in the pkt_flags field.
1583 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1584 hlen_type_rss = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1585 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1586 pkt_flags |= rx_desc_status_to_pkt_flags(staterr);
1587 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1588 head->ol_flags = pkt_flags;
1589 #endif /* RTE_NEXT_ABI */
1591 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1592 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1593 else if (pkt_flags & PKT_RX_FDIR) {
1594 head->hash.fdir.hash =
1595 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1596 & IXGBE_ATR_HASH_MASK;
1597 head->hash.fdir.id =
1598 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1603 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1605 * @rx_queue Rx queue handle
1606 * @rx_pkts table of received packets
1607 * @nb_pkts size of rx_pkts table
1608 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1610 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1611 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1613 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1614 * 1) When non-EOP RSC completion arrives:
1615 * a) Update the HEAD of the current RSC aggregation cluster with the new
1616 * segment's data length.
1617 * b) Set the "next" pointer of the current segment to point to the segment
1618 * at the NEXTP index.
1619 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
1620 * in the sw_rsc_ring.
1621 * 2) When EOP arrives we just update the cluster's total length and offload
1622 * flags and deliver the cluster up to the upper layers. In our case - put it
1623 * in the rx_pkts table.
1625 * Returns the number of received packets/clusters (according to the "bulk
1626 * receive" interface).
1628 static inline uint16_t
1629 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
1632 struct ixgbe_rx_queue *rxq = rx_queue;
1633 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
1634 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
1635 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
1636 uint16_t rx_id = rxq->rx_tail;
1638 uint16_t nb_hold = rxq->nb_rx_hold;
1639 uint16_t prev_id = rxq->rx_tail;
1641 while (nb_rx < nb_pkts) {
1643 struct ixgbe_rx_entry *rxe;
1644 struct ixgbe_scattered_rx_entry *sc_entry;
1645 struct ixgbe_scattered_rx_entry *next_sc_entry;
1646 struct ixgbe_rx_entry *next_rxe;
1647 struct rte_mbuf *first_seg;
1648 struct rte_mbuf *rxm;
1649 struct rte_mbuf *nmb;
1650 union ixgbe_adv_rx_desc rxd;
1653 volatile union ixgbe_adv_rx_desc *rxdp;
1658 * The code in this whole file uses the volatile pointer to
1659 * ensure the read ordering of the status and the rest of the
1660 * descriptor fields (on the compiler level only!!!). This is so
1661 * UGLY - why not to just use the compiler barrier instead? DPDK
1662 * even has the rte_compiler_barrier() for that.
1664 * But most importantly this is just wrong because this doesn't
1665 * ensure memory ordering in a general case at all. For
1666 * instance, DPDK is supposed to work on Power CPUs where
1667 * compiler barrier may just not be enough!
1669 * I tried to write only this function properly to have a
1670 * starting point (as a part of an LRO/RSC series) but the
1671 * compiler cursed at me when I tried to cast away the
1672 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
1673 * keeping it the way it is for now.
1675 * The code in this file is broken in so many other places and
1676 * will just not work on a big endian CPU anyway therefore the
1677 * lines below will have to be revisited together with the rest
1681 * - Get rid of "volatile" crap and let the compiler do its
1683 * - Use the proper memory barrier (rte_rmb()) to ensure the
1684 * memory ordering below.
1686 rxdp = &rx_ring[rx_id];
1687 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
1689 if (!(staterr & IXGBE_RXDADV_STAT_DD))
1694 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1695 "staterr=0x%x data_len=%u",
1696 rxq->port_id, rxq->queue_id, rx_id, staterr,
1697 rte_le_to_cpu_16(rxd.wb.upper.length));
1700 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1702 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
1703 "port_id=%u queue_id=%u",
1704 rxq->port_id, rxq->queue_id);
1706 rte_eth_devices[rxq->port_id].data->
1707 rx_mbuf_alloc_failed++;
1711 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
1712 else if (nb_hold > rxq->rx_free_thresh) {
1713 uint16_t next_rdt = rxq->rx_free_trigger;
1715 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
1717 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr,
1719 nb_hold -= rxq->rx_free_thresh;
1721 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
1722 "port_id=%u queue_id=%u",
1723 rxq->port_id, rxq->queue_id);
1725 rte_eth_devices[rxq->port_id].data->
1726 rx_mbuf_alloc_failed++;
1733 rxe = &sw_ring[rx_id];
1734 eop = staterr & IXGBE_RXDADV_STAT_EOP;
1736 next_id = rx_id + 1;
1737 if (next_id == rxq->nb_rx_desc)
1740 /* Prefetch next mbuf while processing current one. */
1741 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
1744 * When next RX descriptor is on a cache-line boundary,
1745 * prefetch the next 4 RX descriptors and the next 4 pointers
1748 if ((next_id & 0x3) == 0) {
1749 rte_ixgbe_prefetch(&rx_ring[next_id]);
1750 rte_ixgbe_prefetch(&sw_ring[next_id]);
1757 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1759 * Update RX descriptor with the physical address of the
1760 * new data buffer of the new allocated mbuf.
1764 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1765 rxdp->read.hdr_addr = dma;
1766 rxdp->read.pkt_addr = dma;
1771 * Set data length & data buffer address of mbuf.
1773 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
1774 rxm->data_len = data_len;
1779 * Get next descriptor index:
1780 * - For RSC it's in the NEXTP field.
1781 * - For a scattered packet - it's just a following
1784 if (ixgbe_rsc_count(&rxd))
1786 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1787 IXGBE_RXDADV_NEXTP_SHIFT;
1791 next_sc_entry = &sw_sc_ring[nextp_id];
1792 next_rxe = &sw_ring[nextp_id];
1793 rte_ixgbe_prefetch(next_rxe);
1796 sc_entry = &sw_sc_ring[rx_id];
1797 first_seg = sc_entry->fbuf;
1798 sc_entry->fbuf = NULL;
1801 * If this is the first buffer of the received packet,
1802 * set the pointer to the first mbuf of the packet and
1803 * initialize its context.
1804 * Otherwise, update the total length and the number of segments
1805 * of the current scattered packet, and update the pointer to
1806 * the last mbuf of the current packet.
1808 if (first_seg == NULL) {
1810 first_seg->pkt_len = data_len;
1811 first_seg->nb_segs = 1;
1813 first_seg->pkt_len += data_len;
1814 first_seg->nb_segs++;
1821 * If this is not the last buffer of the received packet, update
1822 * the pointer to the first mbuf at the NEXTP entry in the
1823 * sw_sc_ring and continue to parse the RX ring.
1826 rxm->next = next_rxe->mbuf;
1827 next_sc_entry->fbuf = first_seg;
1832 * This is the last buffer of the received packet - return
1833 * the current cluster to the user.
1837 /* Initialize the first mbuf of the returned packet */
1838 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq->port_id,
1841 /* Prefetch data of first segment, if configured to do so. */
1842 rte_packet_prefetch((char *)first_seg->buf_addr +
1843 first_seg->data_off);
1846 * Store the mbuf address into the next entry of the array
1847 * of returned packets.
1849 rx_pkts[nb_rx++] = first_seg;
1853 * Record index of the next RX descriptor to probe.
1855 rxq->rx_tail = rx_id;
1858 * If the number of free RX descriptors is greater than the RX free
1859 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1861 * Update the RDT with the value of the last processed RX descriptor
1862 * minus 1, to guarantee that the RDT register is never equal to the
1863 * RDH register, which creates a "full" ring situtation from the
1864 * hardware point of view...
1866 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
1867 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1868 "nb_hold=%u nb_rx=%u",
1869 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
1872 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, prev_id);
1876 rxq->nb_rx_hold = nb_hold;
1881 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1884 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
1888 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1891 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
1894 /*********************************************************************
1896 * Queue management functions
1898 **********************************************************************/
1901 * Rings setup and release.
1903 * TDBA/RDBA should be aligned on 16 byte boundary. But TDLEN/RDLEN should be
1904 * multiple of 128 bytes. So we align TDBA/RDBA on 128 byte boundary. This will
1905 * also optimize cache line size effect. H/W supports up to cache line size 128.
1907 #define IXGBE_ALIGN 128
1910 * Maximum number of Ring Descriptors.
1912 * Since RDLEN/TDLEN should be multiple of 128 bytes, the number of ring
1913 * descriptors should meet the following condition:
1914 * (num_ring_desc * sizeof(rx/tx descriptor)) % 128 == 0
1916 #define IXGBE_MIN_RING_DESC 32
1917 #define IXGBE_MAX_RING_DESC 4096
1920 * Create memzone for HW rings. malloc can't be used as the physical address is
1921 * needed. If the memzone is already created, then this function returns a ptr
1924 static const struct rte_memzone * __attribute__((cold))
1925 ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
1926 uint16_t queue_id, uint32_t ring_size, int socket_id)
1928 char z_name[RTE_MEMZONE_NAMESIZE];
1929 const struct rte_memzone *mz;
1931 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
1932 dev->driver->pci_drv.name, ring_name,
1933 dev->data->port_id, queue_id);
1935 mz = rte_memzone_lookup(z_name);
1939 #ifdef RTE_LIBRTE_XEN_DOM0
1940 return rte_memzone_reserve_bounded(z_name, ring_size,
1941 socket_id, 0, IXGBE_ALIGN, RTE_PGSIZE_2M);
1943 return rte_memzone_reserve_aligned(z_name, ring_size,
1944 socket_id, 0, IXGBE_ALIGN);
1948 static void __attribute__((cold))
1949 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
1953 if (txq->sw_ring != NULL) {
1954 for (i = 0; i < txq->nb_tx_desc; i++) {
1955 if (txq->sw_ring[i].mbuf != NULL) {
1956 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
1957 txq->sw_ring[i].mbuf = NULL;
1963 static void __attribute__((cold))
1964 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
1967 txq->sw_ring != NULL)
1968 rte_free(txq->sw_ring);
1971 static void __attribute__((cold))
1972 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
1974 if (txq != NULL && txq->ops != NULL) {
1975 txq->ops->release_mbufs(txq);
1976 txq->ops->free_swring(txq);
1981 void __attribute__((cold))
1982 ixgbe_dev_tx_queue_release(void *txq)
1984 ixgbe_tx_queue_release(txq);
1987 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
1988 static void __attribute__((cold))
1989 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
1991 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
1992 struct ixgbe_tx_entry *txe = txq->sw_ring;
1995 /* Zero out HW ring memory */
1996 for (i = 0; i < txq->nb_tx_desc; i++) {
1997 txq->tx_ring[i] = zeroed_desc;
2000 /* Initialize SW ring entries */
2001 prev = (uint16_t) (txq->nb_tx_desc - 1);
2002 for (i = 0; i < txq->nb_tx_desc; i++) {
2003 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
2004 txd->wb.status = IXGBE_TXD_STAT_DD;
2007 txe[prev].next_id = i;
2011 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
2012 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
2015 txq->nb_tx_used = 0;
2017 * Always allow 1 descriptor to be un-allocated to avoid
2018 * a H/W race condition
2020 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2021 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2023 memset((void*)&txq->ctx_cache, 0,
2024 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2027 static const struct ixgbe_txq_ops def_txq_ops = {
2028 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2029 .free_swring = ixgbe_tx_free_swring,
2030 .reset = ixgbe_reset_tx_queue,
2033 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2034 * the queue parameters. Used in tx_queue_setup by primary process and then
2035 * in dev_init by secondary process when attaching to an existing ethdev.
2037 void __attribute__((cold))
2038 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2040 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2041 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
2042 && (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2043 PMD_INIT_LOG(INFO, "Using simple tx code path");
2044 #ifdef RTE_IXGBE_INC_VECTOR
2045 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2046 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2047 ixgbe_txq_vec_setup(txq) == 0)) {
2048 PMD_INIT_LOG(INFO, "Vector tx enabled.");
2049 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2052 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2054 PMD_INIT_LOG(INFO, "Using full-featured tx code path");
2056 " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
2057 (unsigned long)txq->txq_flags,
2058 (unsigned long)IXGBE_SIMPLE_FLAGS);
2060 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2061 (unsigned long)txq->tx_rs_thresh,
2062 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2063 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2067 int __attribute__((cold))
2068 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2071 unsigned int socket_id,
2072 const struct rte_eth_txconf *tx_conf)
2074 const struct rte_memzone *tz;
2075 struct ixgbe_tx_queue *txq;
2076 struct ixgbe_hw *hw;
2077 uint16_t tx_rs_thresh, tx_free_thresh;
2079 PMD_INIT_FUNC_TRACE();
2080 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2083 * Validate number of transmit descriptors.
2084 * It must not exceed hardware maximum, and must be multiple
2087 if (((nb_desc * sizeof(union ixgbe_adv_tx_desc)) % IXGBE_ALIGN) != 0 ||
2088 (nb_desc > IXGBE_MAX_RING_DESC) ||
2089 (nb_desc < IXGBE_MIN_RING_DESC)) {
2094 * The following two parameters control the setting of the RS bit on
2095 * transmit descriptors.
2096 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2097 * descriptors have been used.
2098 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2099 * descriptors are used or if the number of descriptors required
2100 * to transmit a packet is greater than the number of free TX
2102 * The following constraints must be satisfied:
2103 * tx_rs_thresh must be greater than 0.
2104 * tx_rs_thresh must be less than the size of the ring minus 2.
2105 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2106 * tx_rs_thresh must be a divisor of the ring size.
2107 * tx_free_thresh must be greater than 0.
2108 * tx_free_thresh must be less than the size of the ring minus 3.
2109 * One descriptor in the TX ring is used as a sentinel to avoid a
2110 * H/W race condition, hence the maximum threshold constraints.
2111 * When set to zero use default values.
2113 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2114 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2115 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2116 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2117 if (tx_rs_thresh >= (nb_desc - 2)) {
2118 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2119 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2120 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2121 (int)dev->data->port_id, (int)queue_idx);
2124 if (tx_free_thresh >= (nb_desc - 3)) {
2125 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2126 "tx_free_thresh must be less than the number of "
2127 "TX descriptors minus 3. (tx_free_thresh=%u "
2128 "port=%d queue=%d)",
2129 (unsigned int)tx_free_thresh,
2130 (int)dev->data->port_id, (int)queue_idx);
2133 if (tx_rs_thresh > tx_free_thresh) {
2134 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2135 "tx_free_thresh. (tx_free_thresh=%u "
2136 "tx_rs_thresh=%u port=%d queue=%d)",
2137 (unsigned int)tx_free_thresh,
2138 (unsigned int)tx_rs_thresh,
2139 (int)dev->data->port_id,
2143 if ((nb_desc % tx_rs_thresh) != 0) {
2144 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2145 "number of TX descriptors. (tx_rs_thresh=%u "
2146 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2147 (int)dev->data->port_id, (int)queue_idx);
2152 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2153 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2154 * by the NIC and all descriptors are written back after the NIC
2155 * accumulates WTHRESH descriptors.
2157 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2158 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2159 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2160 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2161 (int)dev->data->port_id, (int)queue_idx);
2165 /* Free memory prior to re-allocation if needed... */
2166 if (dev->data->tx_queues[queue_idx] != NULL) {
2167 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2168 dev->data->tx_queues[queue_idx] = NULL;
2171 /* First allocate the tx queue data structure */
2172 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2173 RTE_CACHE_LINE_SIZE, socket_id);
2178 * Allocate TX ring hardware descriptors. A memzone large enough to
2179 * handle the maximum ring size is allocated in order to allow for
2180 * resizing in later calls to the queue setup function.
2182 tz = ring_dma_zone_reserve(dev, "tx_ring", queue_idx,
2183 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2186 ixgbe_tx_queue_release(txq);
2190 txq->nb_tx_desc = nb_desc;
2191 txq->tx_rs_thresh = tx_rs_thresh;
2192 txq->tx_free_thresh = tx_free_thresh;
2193 txq->pthresh = tx_conf->tx_thresh.pthresh;
2194 txq->hthresh = tx_conf->tx_thresh.hthresh;
2195 txq->wthresh = tx_conf->tx_thresh.wthresh;
2196 txq->queue_id = queue_idx;
2197 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2198 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2199 txq->port_id = dev->data->port_id;
2200 txq->txq_flags = tx_conf->txq_flags;
2201 txq->ops = &def_txq_ops;
2202 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2205 * Modification to set VFTDT for virtual function if vf is detected
2207 if (hw->mac.type == ixgbe_mac_82599_vf ||
2208 hw->mac.type == ixgbe_mac_X540_vf ||
2209 hw->mac.type == ixgbe_mac_X550_vf ||
2210 hw->mac.type == ixgbe_mac_X550EM_x_vf)
2211 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2213 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2214 #ifndef RTE_LIBRTE_XEN_DOM0
2215 txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
2217 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2219 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2221 /* Allocate software ring */
2222 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2223 sizeof(struct ixgbe_tx_entry) * nb_desc,
2224 RTE_CACHE_LINE_SIZE, socket_id);
2225 if (txq->sw_ring == NULL) {
2226 ixgbe_tx_queue_release(txq);
2229 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2230 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2232 /* set up vector or scalar TX function as appropriate */
2233 ixgbe_set_tx_function(dev, txq);
2235 txq->ops->reset(txq);
2237 dev->data->tx_queues[queue_idx] = txq;
2244 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2246 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2247 * in the sw_rsc_ring is not set to NULL but rather points to the next
2248 * mbuf of this RSC aggregation (that has not been completed yet and still
2249 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2250 * will just free first "nb_segs" segments of the cluster explicitly by calling
2251 * an rte_pktmbuf_free_seg().
2253 * @m scattered cluster head
2255 static void __attribute__((cold))
2256 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2258 uint8_t i, nb_segs = m->nb_segs;
2259 struct rte_mbuf *next_seg;
2261 for (i = 0; i < nb_segs; i++) {
2263 rte_pktmbuf_free_seg(m);
2268 static void __attribute__((cold))
2269 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2273 if (rxq->sw_ring != NULL) {
2274 for (i = 0; i < rxq->nb_rx_desc; i++) {
2275 if (rxq->sw_ring[i].mbuf != NULL) {
2276 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2277 rxq->sw_ring[i].mbuf = NULL;
2280 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2281 if (rxq->rx_nb_avail) {
2282 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2283 struct rte_mbuf *mb;
2284 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2285 rte_pktmbuf_free_seg(mb);
2287 rxq->rx_nb_avail = 0;
2292 if (rxq->sw_sc_ring)
2293 for (i = 0; i < rxq->nb_rx_desc; i++)
2294 if (rxq->sw_sc_ring[i].fbuf) {
2295 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2296 rxq->sw_sc_ring[i].fbuf = NULL;
2300 static void __attribute__((cold))
2301 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2304 ixgbe_rx_queue_release_mbufs(rxq);
2305 rte_free(rxq->sw_ring);
2306 rte_free(rxq->sw_sc_ring);
2311 void __attribute__((cold))
2312 ixgbe_dev_rx_queue_release(void *rxq)
2314 ixgbe_rx_queue_release(rxq);
2318 * Check if Rx Burst Bulk Alloc function can be used.
2320 * 0: the preconditions are satisfied and the bulk allocation function
2322 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2323 * function must be used.
2325 static inline int __attribute__((cold))
2326 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2327 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2329 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct ixgbe_rx_queue *rxq)
2335 * Make sure the following pre-conditions are satisfied:
2336 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2337 * rxq->rx_free_thresh < rxq->nb_rx_desc
2338 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2339 * rxq->nb_rx_desc<(IXGBE_MAX_RING_DESC-RTE_PMD_IXGBE_RX_MAX_BURST)
2340 * Scattered packets are not supported. This should be checked
2341 * outside of this function.
2343 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2344 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2345 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2346 "rxq->rx_free_thresh=%d, "
2347 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2348 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2350 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2351 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2352 "rxq->rx_free_thresh=%d, "
2353 "rxq->nb_rx_desc=%d",
2354 rxq->rx_free_thresh, rxq->nb_rx_desc);
2356 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2357 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2358 "rxq->nb_rx_desc=%d, "
2359 "rxq->rx_free_thresh=%d",
2360 rxq->nb_rx_desc, rxq->rx_free_thresh);
2362 } else if (!(rxq->nb_rx_desc <
2363 (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST))) {
2364 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2365 "rxq->nb_rx_desc=%d, "
2366 "IXGBE_MAX_RING_DESC=%d, "
2367 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2368 rxq->nb_rx_desc, IXGBE_MAX_RING_DESC,
2369 RTE_PMD_IXGBE_RX_MAX_BURST);
2379 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2380 static void __attribute__((cold))
2381 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2383 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2385 uint16_t len = rxq->nb_rx_desc;
2388 * By default, the Rx queue setup function allocates enough memory for
2389 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2390 * extra memory at the end of the descriptor ring to be zero'd out. A
2391 * pre-condition for using the Rx burst bulk alloc function is that the
2392 * number of descriptors is less than or equal to
2393 * (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST). Check all the
2394 * constraints here to see if we need to zero out memory after the end
2395 * of the H/W descriptor ring.
2397 if (adapter->rx_bulk_alloc_allowed)
2398 /* zero out extra memory */
2399 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2402 * Zero out HW ring memory. Zero out extra memory at the end of
2403 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2404 * reads extra memory as zeros.
2406 for (i = 0; i < len; i++) {
2407 rxq->rx_ring[i] = zeroed_desc;
2410 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2412 * initialize extra software ring entries. Space for these extra
2413 * entries is always allocated
2415 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2416 for (i = rxq->nb_rx_desc; i < len; ++i) {
2417 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2420 rxq->rx_nb_avail = 0;
2421 rxq->rx_next_avail = 0;
2422 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2423 #endif /* RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC */
2425 rxq->nb_rx_hold = 0;
2426 rxq->pkt_first_seg = NULL;
2427 rxq->pkt_last_seg = NULL;
2430 int __attribute__((cold))
2431 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2434 unsigned int socket_id,
2435 const struct rte_eth_rxconf *rx_conf,
2436 struct rte_mempool *mp)
2438 const struct rte_memzone *rz;
2439 struct ixgbe_rx_queue *rxq;
2440 struct ixgbe_hw *hw;
2442 struct ixgbe_adapter *adapter =
2443 (struct ixgbe_adapter *)dev->data->dev_private;
2445 PMD_INIT_FUNC_TRACE();
2446 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2449 * Validate number of receive descriptors.
2450 * It must not exceed hardware maximum, and must be multiple
2453 if (((nb_desc * sizeof(union ixgbe_adv_rx_desc)) % IXGBE_ALIGN) != 0 ||
2454 (nb_desc > IXGBE_MAX_RING_DESC) ||
2455 (nb_desc < IXGBE_MIN_RING_DESC)) {
2459 /* Free memory prior to re-allocation if needed... */
2460 if (dev->data->rx_queues[queue_idx] != NULL) {
2461 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2462 dev->data->rx_queues[queue_idx] = NULL;
2465 /* First allocate the rx queue data structure */
2466 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2467 RTE_CACHE_LINE_SIZE, socket_id);
2471 rxq->nb_rx_desc = nb_desc;
2472 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2473 rxq->queue_id = queue_idx;
2474 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2475 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2476 rxq->port_id = dev->data->port_id;
2477 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2479 rxq->drop_en = rx_conf->rx_drop_en;
2480 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2483 * Allocate RX ring hardware descriptors. A memzone large enough to
2484 * handle the maximum ring size is allocated in order to allow for
2485 * resizing in later calls to the queue setup function.
2487 rz = ring_dma_zone_reserve(dev, "rx_ring", queue_idx,
2488 RX_RING_SZ, socket_id);
2490 ixgbe_rx_queue_release(rxq);
2495 * Zero init all the descriptors in the ring.
2497 memset (rz->addr, 0, RX_RING_SZ);
2500 * Modified to setup VFRDT for Virtual Function
2502 if (hw->mac.type == ixgbe_mac_82599_vf ||
2503 hw->mac.type == ixgbe_mac_X540_vf ||
2504 hw->mac.type == ixgbe_mac_X550_vf ||
2505 hw->mac.type == ixgbe_mac_X550EM_x_vf) {
2507 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2509 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2513 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2515 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2517 #ifndef RTE_LIBRTE_XEN_DOM0
2518 rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
2520 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2522 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2525 * Certain constraints must be met in order to use the bulk buffer
2526 * allocation Rx burst function. If any of Rx queues doesn't meet them
2527 * the feature should be disabled for the whole port.
2529 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
2530 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
2531 "preconditions - canceling the feature for "
2532 "the whole port[%d]",
2533 rxq->queue_id, rxq->port_id);
2534 adapter->rx_bulk_alloc_allowed = false;
2538 * Allocate software ring. Allow for space at the end of the
2539 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2540 * function does not access an invalid memory region.
2543 if (adapter->rx_bulk_alloc_allowed)
2544 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2546 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2547 sizeof(struct ixgbe_rx_entry) * len,
2548 RTE_CACHE_LINE_SIZE, socket_id);
2549 if (!rxq->sw_ring) {
2550 ixgbe_rx_queue_release(rxq);
2555 * Always allocate even if it's not going to be needed in order to
2556 * simplify the code.
2558 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
2559 * be requested in ixgbe_dev_rx_init(), which is called later from
2563 rte_zmalloc_socket("rxq->sw_sc_ring",
2564 sizeof(struct ixgbe_scattered_rx_entry) * len,
2565 RTE_CACHE_LINE_SIZE, socket_id);
2566 if (!rxq->sw_sc_ring) {
2567 ixgbe_rx_queue_release(rxq);
2571 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
2572 "dma_addr=0x%"PRIx64,
2573 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
2574 rxq->rx_ring_phys_addr);
2576 if (!rte_is_power_of_2(nb_desc)) {
2577 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
2578 "preconditions - canceling the feature for "
2579 "the whole port[%d]",
2580 rxq->queue_id, rxq->port_id);
2581 adapter->rx_vec_allowed = false;
2583 ixgbe_rxq_vec_setup(rxq);
2585 dev->data->rx_queues[queue_idx] = rxq;
2587 ixgbe_reset_rx_queue(adapter, rxq);
2593 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2595 #define IXGBE_RXQ_SCAN_INTERVAL 4
2596 volatile union ixgbe_adv_rx_desc *rxdp;
2597 struct ixgbe_rx_queue *rxq;
2600 if (rx_queue_id >= dev->data->nb_rx_queues) {
2601 PMD_RX_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
2605 rxq = dev->data->rx_queues[rx_queue_id];
2606 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2608 while ((desc < rxq->nb_rx_desc) &&
2609 (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD)) {
2610 desc += IXGBE_RXQ_SCAN_INTERVAL;
2611 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2612 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2613 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2614 desc - rxq->nb_rx_desc]);
2621 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2623 volatile union ixgbe_adv_rx_desc *rxdp;
2624 struct ixgbe_rx_queue *rxq = rx_queue;
2627 if (unlikely(offset >= rxq->nb_rx_desc))
2629 desc = rxq->rx_tail + offset;
2630 if (desc >= rxq->nb_rx_desc)
2631 desc -= rxq->nb_rx_desc;
2633 rxdp = &rxq->rx_ring[desc];
2634 return !!(rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD);
2637 void __attribute__((cold))
2638 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
2641 struct ixgbe_adapter *adapter =
2642 (struct ixgbe_adapter *)dev->data->dev_private;
2644 PMD_INIT_FUNC_TRACE();
2646 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2647 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
2649 txq->ops->release_mbufs(txq);
2650 txq->ops->reset(txq);
2654 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2655 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
2657 ixgbe_rx_queue_release_mbufs(rxq);
2658 ixgbe_reset_rx_queue(adapter, rxq);
2664 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
2668 PMD_INIT_FUNC_TRACE();
2670 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2671 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
2672 dev->data->rx_queues[i] = NULL;
2674 dev->data->nb_rx_queues = 0;
2676 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2677 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
2678 dev->data->tx_queues[i] = NULL;
2680 dev->data->nb_tx_queues = 0;
2683 /*********************************************************************
2685 * Device RX/TX init functions
2687 **********************************************************************/
2690 * Receive Side Scaling (RSS)
2691 * See section 7.1.2.8 in the following document:
2692 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
2695 * The source and destination IP addresses of the IP header and the source
2696 * and destination ports of TCP/UDP headers, if any, of received packets are
2697 * hashed against a configurable random key to compute a 32-bit RSS hash result.
2698 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
2699 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
2700 * RSS output index which is used as the RX queue index where to store the
2702 * The following output is supplied in the RX write-back descriptor:
2703 * - 32-bit result of the Microsoft RSS hash function,
2704 * - 4-bit RSS type field.
2708 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
2709 * Used as the default key.
2711 static uint8_t rss_intel_key[40] = {
2712 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2713 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2714 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2715 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2716 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2720 ixgbe_rss_disable(struct rte_eth_dev *dev)
2722 struct ixgbe_hw *hw;
2725 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2726 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2727 mrqc &= ~IXGBE_MRQC_RSSEN;
2728 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2732 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
2740 hash_key = rss_conf->rss_key;
2741 if (hash_key != NULL) {
2742 /* Fill in RSS hash key */
2743 for (i = 0; i < 10; i++) {
2744 rss_key = hash_key[(i * 4)];
2745 rss_key |= hash_key[(i * 4) + 1] << 8;
2746 rss_key |= hash_key[(i * 4) + 2] << 16;
2747 rss_key |= hash_key[(i * 4) + 3] << 24;
2748 IXGBE_WRITE_REG_ARRAY(hw, IXGBE_RSSRK(0), i, rss_key);
2752 /* Set configured hashing protocols in MRQC register */
2753 rss_hf = rss_conf->rss_hf;
2754 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
2755 if (rss_hf & ETH_RSS_IPV4)
2756 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
2757 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
2758 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
2759 if (rss_hf & ETH_RSS_IPV6)
2760 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
2761 if (rss_hf & ETH_RSS_IPV6_EX)
2762 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
2763 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
2764 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2765 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
2766 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
2767 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
2768 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2769 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
2770 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2771 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
2772 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2773 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2777 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
2778 struct rte_eth_rss_conf *rss_conf)
2780 struct ixgbe_hw *hw;
2784 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2787 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
2788 * "RSS enabling cannot be done dynamically while it must be
2789 * preceded by a software reset"
2790 * Before changing anything, first check that the update RSS operation
2791 * does not attempt to disable RSS, if RSS was enabled at
2792 * initialization time, or does not attempt to enable RSS, if RSS was
2793 * disabled at initialization time.
2795 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
2796 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2797 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
2798 if (rss_hf != 0) /* Enable RSS */
2800 return 0; /* Nothing to do */
2803 if (rss_hf == 0) /* Disable RSS */
2805 ixgbe_hw_rss_hash_set(hw, rss_conf);
2810 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2811 struct rte_eth_rss_conf *rss_conf)
2813 struct ixgbe_hw *hw;
2820 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2821 hash_key = rss_conf->rss_key;
2822 if (hash_key != NULL) {
2823 /* Return RSS hash key */
2824 for (i = 0; i < 10; i++) {
2825 rss_key = IXGBE_READ_REG_ARRAY(hw, IXGBE_RSSRK(0), i);
2826 hash_key[(i * 4)] = rss_key & 0x000000FF;
2827 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
2828 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
2829 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
2833 /* Get RSS functions configured in MRQC register */
2834 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2835 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
2836 rss_conf->rss_hf = 0;
2840 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
2841 rss_hf |= ETH_RSS_IPV4;
2842 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
2843 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2844 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
2845 rss_hf |= ETH_RSS_IPV6;
2846 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
2847 rss_hf |= ETH_RSS_IPV6_EX;
2848 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
2849 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2850 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
2851 rss_hf |= ETH_RSS_IPV6_TCP_EX;
2852 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
2853 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2854 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
2855 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2856 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
2857 rss_hf |= ETH_RSS_IPV6_UDP_EX;
2858 rss_conf->rss_hf = rss_hf;
2863 ixgbe_rss_configure(struct rte_eth_dev *dev)
2865 struct rte_eth_rss_conf rss_conf;
2866 struct ixgbe_hw *hw;
2871 PMD_INIT_FUNC_TRACE();
2872 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2875 * Fill in redirection table
2876 * The byte-swap is needed because NIC registers are in
2877 * little-endian order.
2880 for (i = 0, j = 0; i < 128; i++, j++) {
2881 if (j == dev->data->nb_rx_queues)
2883 reta = (reta << 8) | j;
2885 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2),
2890 * Configure the RSS key and the RSS protocols used to compute
2891 * the RSS hash of input packets.
2893 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
2894 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
2895 ixgbe_rss_disable(dev);
2898 if (rss_conf.rss_key == NULL)
2899 rss_conf.rss_key = rss_intel_key; /* Default hash key */
2900 ixgbe_hw_rss_hash_set(hw, &rss_conf);
2903 #define NUM_VFTA_REGISTERS 128
2904 #define NIC_RX_BUFFER_SIZE 0x200
2907 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
2909 struct rte_eth_vmdq_dcb_conf *cfg;
2910 struct ixgbe_hw *hw;
2911 enum rte_eth_nb_pools num_pools;
2912 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
2914 uint8_t nb_tcs; /* number of traffic classes */
2917 PMD_INIT_FUNC_TRACE();
2918 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2919 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
2920 num_pools = cfg->nb_queue_pools;
2921 /* Check we have a valid number of pools */
2922 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
2923 ixgbe_rss_disable(dev);
2926 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
2927 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
2931 * split rx buffer up into sections, each for 1 traffic class
2933 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
2934 for (i = 0 ; i < nb_tcs; i++) {
2935 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2936 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
2937 /* clear 10 bits. */
2938 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
2939 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2941 /* zero alloc all unused TCs */
2942 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2943 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2944 rxpbsize &= (~( 0x3FF << IXGBE_RXPBSIZE_SHIFT ));
2945 /* clear 10 bits. */
2946 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2949 /* MRQC: enable vmdq and dcb */
2950 mrqc = ((num_pools == ETH_16_POOLS) ? \
2951 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN );
2952 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2954 /* PFVTCTL: turn on virtualisation and set the default pool */
2955 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2956 if (cfg->enable_default_pool) {
2957 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
2959 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
2962 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
2964 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
2966 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
2968 * mapping is done with 3 bits per priority,
2969 * so shift by i*3 each time
2971 queue_mapping |= ((cfg->dcb_queue[i] & 0x07) << (i * 3));
2973 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
2975 /* RTRPCS: DCB related */
2976 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
2978 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
2979 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2980 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
2981 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
2983 /* VFTA - enable all vlan filters */
2984 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
2985 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
2988 /* VFRE: pool enabling for receive - 16 or 32 */
2989 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), \
2990 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
2993 * MPSAR - allow pools to read specific mac addresses
2994 * In this case, all pools should be able to read from mac addr 0
2996 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
2997 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
2999 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3000 for (i = 0; i < cfg->nb_pool_maps; i++) {
3001 /* set vlan id in VF register and set the valid bit */
3002 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
3003 (cfg->pool_map[i].vlan_id & 0xFFF)));
3005 * Put the allowed pools in VFB reg. As we only have 16 or 32
3006 * pools, we only need to use the first half of the register
3009 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
3014 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
3015 * @hw: pointer to hardware structure
3016 * @dcb_config: pointer to ixgbe_dcb_config structure
3019 ixgbe_dcb_tx_hw_config(struct ixgbe_hw *hw,
3020 struct ixgbe_dcb_config *dcb_config)
3025 PMD_INIT_FUNC_TRACE();
3026 if (hw->mac.type != ixgbe_mac_82598EB) {
3027 /* Disable the Tx desc arbiter so that MTQC can be changed */
3028 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3029 reg |= IXGBE_RTTDCS_ARBDIS;
3030 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3032 /* Enable DCB for Tx with 8 TCs */
3033 if (dcb_config->num_tcs.pg_tcs == 8) {
3034 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3037 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3039 if (dcb_config->vt_mode)
3040 reg |= IXGBE_MTQC_VT_ENA;
3041 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3043 /* Disable drop for all queues */
3044 for (q = 0; q < 128; q++)
3045 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3046 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3048 /* Enable the Tx desc arbiter */
3049 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3050 reg &= ~IXGBE_RTTDCS_ARBDIS;
3051 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3053 /* Enable Security TX Buffer IFG for DCB */
3054 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3055 reg |= IXGBE_SECTX_DCB;
3056 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3062 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3063 * @dev: pointer to rte_eth_dev structure
3064 * @dcb_config: pointer to ixgbe_dcb_config structure
3067 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3068 struct ixgbe_dcb_config *dcb_config)
3070 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3071 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3072 struct ixgbe_hw *hw =
3073 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3075 PMD_INIT_FUNC_TRACE();
3076 if (hw->mac.type != ixgbe_mac_82598EB)
3077 /*PF VF Transmit Enable*/
3078 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3079 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3081 /*Configure general DCB TX parameters*/
3082 ixgbe_dcb_tx_hw_config(hw,dcb_config);
3087 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3088 struct ixgbe_dcb_config *dcb_config)
3090 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3091 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3092 struct ixgbe_dcb_tc_config *tc;
3095 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3096 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS ) {
3097 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3098 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3101 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3102 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3104 /* User Priority to Traffic Class mapping */
3105 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3106 j = vmdq_rx_conf->dcb_queue[i];
3107 tc = &dcb_config->tc_config[j];
3108 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3114 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3115 struct ixgbe_dcb_config *dcb_config)
3117 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3118 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3119 struct ixgbe_dcb_tc_config *tc;
3122 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3123 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ) {
3124 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3125 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3128 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3129 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3132 /* User Priority to Traffic Class mapping */
3133 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3134 j = vmdq_tx_conf->dcb_queue[i];
3135 tc = &dcb_config->tc_config[j];
3136 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3143 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3144 struct ixgbe_dcb_config *dcb_config)
3146 struct rte_eth_dcb_rx_conf *rx_conf =
3147 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3148 struct ixgbe_dcb_tc_config *tc;
3151 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3152 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3154 /* User Priority to Traffic Class mapping */
3155 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3156 j = rx_conf->dcb_queue[i];
3157 tc = &dcb_config->tc_config[j];
3158 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3164 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3165 struct ixgbe_dcb_config *dcb_config)
3167 struct rte_eth_dcb_tx_conf *tx_conf =
3168 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3169 struct ixgbe_dcb_tc_config *tc;
3172 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3173 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3175 /* User Priority to Traffic Class mapping */
3176 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3177 j = tx_conf->dcb_queue[i];
3178 tc = &dcb_config->tc_config[j];
3179 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3185 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3186 * @hw: pointer to hardware structure
3187 * @dcb_config: pointer to ixgbe_dcb_config structure
3190 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
3191 struct ixgbe_dcb_config *dcb_config)
3197 PMD_INIT_FUNC_TRACE();
3199 * Disable the arbiter before changing parameters
3200 * (always enable recycle mode; WSP)
3202 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3203 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3205 if (hw->mac.type != ixgbe_mac_82598EB) {
3206 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3207 if (dcb_config->num_tcs.pg_tcs == 4) {
3208 if (dcb_config->vt_mode)
3209 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3210 IXGBE_MRQC_VMDQRT4TCEN;
3212 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3213 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3217 if (dcb_config->num_tcs.pg_tcs == 8) {
3218 if (dcb_config->vt_mode)
3219 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3220 IXGBE_MRQC_VMDQRT8TCEN;
3222 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3223 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3228 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3231 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3232 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3233 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
3234 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3236 /* VFTA - enable all vlan filters */
3237 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3238 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3242 * Configure Rx packet plane (recycle mode; WSP) and
3245 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3246 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3252 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3253 uint16_t *max,uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3255 switch (hw->mac.type) {
3256 case ixgbe_mac_82598EB:
3257 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3259 case ixgbe_mac_82599EB:
3260 case ixgbe_mac_X540:
3261 case ixgbe_mac_X550:
3262 case ixgbe_mac_X550EM_x:
3263 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3272 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3273 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3275 switch (hw->mac.type) {
3276 case ixgbe_mac_82598EB:
3277 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id,tsa);
3278 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id,tsa);
3280 case ixgbe_mac_82599EB:
3281 case ixgbe_mac_X540:
3282 case ixgbe_mac_X550:
3283 case ixgbe_mac_X550EM_x:
3284 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id,tsa);
3285 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id,tsa, map);
3292 #define DCB_RX_CONFIG 1
3293 #define DCB_TX_CONFIG 1
3294 #define DCB_TX_PB 1024
3296 * ixgbe_dcb_hw_configure - Enable DCB and configure
3297 * general DCB in VT mode and non-VT mode parameters
3298 * @dev: pointer to rte_eth_dev structure
3299 * @dcb_config: pointer to ixgbe_dcb_config structure
3302 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3303 struct ixgbe_dcb_config *dcb_config)
3306 uint8_t i,pfc_en,nb_tcs;
3308 uint8_t config_dcb_rx = 0;
3309 uint8_t config_dcb_tx = 0;
3310 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3311 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3312 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3313 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3314 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3315 struct ixgbe_dcb_tc_config *tc;
3316 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3317 struct ixgbe_hw *hw =
3318 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3320 switch(dev->data->dev_conf.rxmode.mq_mode){
3321 case ETH_MQ_RX_VMDQ_DCB:
3322 dcb_config->vt_mode = true;
3323 if (hw->mac.type != ixgbe_mac_82598EB) {
3324 config_dcb_rx = DCB_RX_CONFIG;
3326 *get dcb and VT rx configuration parameters
3329 ixgbe_vmdq_dcb_rx_config(dev,dcb_config);
3330 /*Configure general VMDQ and DCB RX parameters*/
3331 ixgbe_vmdq_dcb_configure(dev);
3335 dcb_config->vt_mode = false;
3336 config_dcb_rx = DCB_RX_CONFIG;
3337 /* Get dcb TX configuration parameters from rte_eth_conf */
3338 ixgbe_dcb_rx_config(dev,dcb_config);
3339 /*Configure general DCB RX parameters*/
3340 ixgbe_dcb_rx_hw_config(hw, dcb_config);
3343 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3346 switch (dev->data->dev_conf.txmode.mq_mode) {
3347 case ETH_MQ_TX_VMDQ_DCB:
3348 dcb_config->vt_mode = true;
3349 config_dcb_tx = DCB_TX_CONFIG;
3350 /* get DCB and VT TX configuration parameters from rte_eth_conf */
3351 ixgbe_dcb_vt_tx_config(dev,dcb_config);
3352 /*Configure general VMDQ and DCB TX parameters*/
3353 ixgbe_vmdq_dcb_hw_tx_config(dev,dcb_config);
3357 dcb_config->vt_mode = false;
3358 config_dcb_tx = DCB_TX_CONFIG;
3359 /*get DCB TX configuration parameters from rte_eth_conf*/
3360 ixgbe_dcb_tx_config(dev,dcb_config);
3361 /*Configure general DCB TX parameters*/
3362 ixgbe_dcb_tx_hw_config(hw, dcb_config);
3365 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3369 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3371 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3372 if(nb_tcs == ETH_4_TCS) {
3373 /* Avoid un-configured priority mapping to TC0 */
3375 uint8_t mask = 0xFF;
3376 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3377 mask = (uint8_t)(mask & (~ (1 << map[i])));
3378 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3379 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3383 /* Re-configure 4 TCs BW */
3384 for (i = 0; i < nb_tcs; i++) {
3385 tc = &dcb_config->tc_config[i];
3386 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
3387 (uint8_t)(100 / nb_tcs);
3388 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
3389 (uint8_t)(100 / nb_tcs);
3391 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
3392 tc = &dcb_config->tc_config[i];
3393 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
3394 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
3399 /* Set RX buffer size */
3400 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3401 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
3402 for (i = 0 ; i < nb_tcs; i++) {
3403 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3405 /* zero alloc all unused TCs */
3406 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3407 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
3411 /* Only support an equally distributed Tx packet buffer strategy. */
3412 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
3413 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
3414 for (i = 0; i < nb_tcs; i++) {
3415 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
3416 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
3418 /* Clear unused TCs, if any, to zero buffer size*/
3419 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3420 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
3421 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
3425 /*Calculates traffic class credits*/
3426 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
3427 IXGBE_DCB_TX_CONFIG);
3428 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
3429 IXGBE_DCB_RX_CONFIG);
3432 /* Unpack CEE standard containers */
3433 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
3434 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3435 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
3436 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
3437 /* Configure PG(ETS) RX */
3438 ixgbe_dcb_hw_arbite_rx_config(hw,refill,max,bwgid,tsa,map);
3442 /* Unpack CEE standard containers */
3443 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
3444 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3445 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
3446 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
3447 /* Configure PG(ETS) TX */
3448 ixgbe_dcb_hw_arbite_tx_config(hw,refill,max,bwgid,tsa,map);
3451 /*Configure queue statistics registers*/
3452 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3454 /* Check if the PFC is supported */
3455 if(dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3456 pbsize = (uint16_t) (NIC_RX_BUFFER_SIZE / nb_tcs);
3457 for (i = 0; i < nb_tcs; i++) {
3459 * If the TC count is 8,and the default high_water is 48,
3460 * the low_water is 16 as default.
3462 hw->fc.high_water[i] = (pbsize * 3 ) / 4;
3463 hw->fc.low_water[i] = pbsize / 4;
3464 /* Enable pfc for this TC */
3465 tc = &dcb_config->tc_config[i];
3466 tc->pfc = ixgbe_dcb_pfc_enabled;
3468 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3469 if(dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3471 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3478 * ixgbe_configure_dcb - Configure DCB Hardware
3479 * @dev: pointer to rte_eth_dev
3481 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3483 struct ixgbe_dcb_config *dcb_cfg =
3484 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3485 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3487 PMD_INIT_FUNC_TRACE();
3489 /* check support mq_mode for DCB */
3490 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
3491 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB))
3494 if (dev->data->nb_rx_queues != ETH_DCB_NUM_QUEUES)
3497 /** Configure DCB hardware **/
3498 ixgbe_dcb_hw_configure(dev,dcb_cfg);
3504 * VMDq only support for 10 GbE NIC.
3507 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
3509 struct rte_eth_vmdq_rx_conf *cfg;
3510 struct ixgbe_hw *hw;
3511 enum rte_eth_nb_pools num_pools;
3512 uint32_t mrqc, vt_ctl, vlanctrl;
3516 PMD_INIT_FUNC_TRACE();
3517 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3518 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
3519 num_pools = cfg->nb_queue_pools;
3521 ixgbe_rss_disable(dev);
3523 /* MRQC: enable vmdq */
3524 mrqc = IXGBE_MRQC_VMDQEN;
3525 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3527 /* PFVTCTL: turn on virtualisation and set the default pool */
3528 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3529 if (cfg->enable_default_pool)
3530 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3532 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3534 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3536 for (i = 0; i < (int)num_pools; i++) {
3537 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
3538 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
3541 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3542 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3543 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
3544 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3546 /* VFTA - enable all vlan filters */
3547 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
3548 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
3550 /* VFRE: pool enabling for receive - 64 */
3551 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
3552 if (num_pools == ETH_64_POOLS)
3553 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
3556 * MPSAR - allow pools to read specific mac addresses
3557 * In this case, all pools should be able to read from mac addr 0
3559 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
3560 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
3562 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3563 for (i = 0; i < cfg->nb_pool_maps; i++) {
3564 /* set vlan id in VF register and set the valid bit */
3565 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
3566 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
3568 * Put the allowed pools in VFB reg. As we only have 16 or 64
3569 * pools, we only need to use the first half of the register
3572 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
3573 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), \
3574 (cfg->pool_map[i].pools & UINT32_MAX));
3576 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i*2+1)), \
3577 ((cfg->pool_map[i].pools >> 32) \
3582 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
3583 if (cfg->enable_loop_back) {
3584 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3585 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
3586 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
3589 IXGBE_WRITE_FLUSH(hw);
3593 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
3594 * @hw: pointer to hardware structure
3597 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
3602 PMD_INIT_FUNC_TRACE();
3603 /*PF VF Transmit Enable*/
3604 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
3605 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
3607 /* Disable the Tx desc arbiter so that MTQC can be changed */
3608 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3609 reg |= IXGBE_RTTDCS_ARBDIS;
3610 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3612 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3613 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3615 /* Disable drop for all queues */
3616 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3617 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3618 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3620 /* Enable the Tx desc arbiter */
3621 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3622 reg &= ~IXGBE_RTTDCS_ARBDIS;
3623 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3625 IXGBE_WRITE_FLUSH(hw);
3630 static int __attribute__((cold))
3631 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
3633 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
3637 /* Initialize software ring entries */
3638 for (i = 0; i < rxq->nb_rx_desc; i++) {
3639 volatile union ixgbe_adv_rx_desc *rxd;
3640 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mb_pool);
3642 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
3643 (unsigned) rxq->queue_id);
3647 rte_mbuf_refcnt_set(mbuf, 1);
3649 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
3651 mbuf->port = rxq->port_id;
3654 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
3655 rxd = &rxq->rx_ring[i];
3656 rxd->read.hdr_addr = dma_addr;
3657 rxd->read.pkt_addr = dma_addr;
3665 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
3667 struct ixgbe_hw *hw;
3670 ixgbe_rss_configure(dev);
3672 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3674 /* MRQC: enable VF RSS */
3675 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
3676 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
3677 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3679 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
3683 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
3687 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
3691 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3697 ixgbe_config_vf_default(struct rte_eth_dev *dev)
3699 struct ixgbe_hw *hw =
3700 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3702 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3704 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3709 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3710 IXGBE_MRQC_VMDQRT4TCEN);
3714 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3715 IXGBE_MRQC_VMDQRT8TCEN);
3719 "invalid pool number in IOV mode");
3726 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
3728 struct ixgbe_hw *hw =
3729 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3731 if (hw->mac.type == ixgbe_mac_82598EB)
3734 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3736 * SRIOV inactive scheme
3737 * any DCB/RSS w/o VMDq multi-queue setting
3739 switch (dev->data->dev_conf.rxmode.mq_mode) {
3741 ixgbe_rss_configure(dev);
3744 case ETH_MQ_RX_VMDQ_DCB:
3745 ixgbe_vmdq_dcb_configure(dev);
3748 case ETH_MQ_RX_VMDQ_ONLY:
3749 ixgbe_vmdq_rx_hw_configure(dev);
3752 case ETH_MQ_RX_NONE:
3753 /* if mq_mode is none, disable rss mode.*/
3754 default: ixgbe_rss_disable(dev);
3758 * SRIOV active scheme
3759 * Support RSS together with VMDq & SRIOV
3761 switch (dev->data->dev_conf.rxmode.mq_mode) {
3763 case ETH_MQ_RX_VMDQ_RSS:
3764 ixgbe_config_vf_rss(dev);
3767 /* FIXME if support DCB/RSS together with VMDq & SRIOV */
3768 case ETH_MQ_RX_VMDQ_DCB:
3769 case ETH_MQ_RX_VMDQ_DCB_RSS:
3771 "Could not support DCB with VMDq & SRIOV");
3774 ixgbe_config_vf_default(dev);
3783 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
3785 struct ixgbe_hw *hw =
3786 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3790 if (hw->mac.type == ixgbe_mac_82598EB)
3793 /* disable arbiter before setting MTQC */
3794 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3795 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3796 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3798 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3800 * SRIOV inactive scheme
3801 * any DCB w/o VMDq multi-queue setting
3803 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
3804 ixgbe_vmdq_tx_hw_configure(hw);
3806 mtqc = IXGBE_MTQC_64Q_1PB;
3807 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3810 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3813 * SRIOV active scheme
3814 * FIXME if support DCB together with VMDq & SRIOV
3817 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3820 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
3823 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
3827 mtqc = IXGBE_MTQC_64Q_1PB;
3828 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
3830 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3833 /* re-enable arbiter */
3834 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3835 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3841 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
3843 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
3844 * spec rev. 3.0 chapter 8.2.3.8.13.
3846 * @pool Memory pool of the Rx queue
3848 static inline uint32_t
3849 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
3851 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
3853 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
3856 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
3859 return IXGBE_RSCCTL_MAXDESC_16;
3860 else if (maxdesc >= 8)
3861 return IXGBE_RSCCTL_MAXDESC_8;
3862 else if (maxdesc >= 4)
3863 return IXGBE_RSCCTL_MAXDESC_4;
3865 return IXGBE_RSCCTL_MAXDESC_1;
3869 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
3872 * (Taken from FreeBSD tree)
3873 * (yes this is all very magic and confusing :)
3876 * @entry the register array entry
3877 * @vector the MSIX vector for this queue
3881 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
3883 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3886 vector |= IXGBE_IVAR_ALLOC_VAL;
3888 switch (hw->mac.type) {
3890 case ixgbe_mac_82598EB:
3892 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
3894 entry += (type * 64);
3895 index = (entry >> 2) & 0x1F;
3896 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
3897 ivar &= ~(0xFF << (8 * (entry & 0x3)));
3898 ivar |= (vector << (8 * (entry & 0x3)));
3899 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
3902 case ixgbe_mac_82599EB:
3903 case ixgbe_mac_X540:
3904 if (type == -1) { /* MISC IVAR */
3905 index = (entry & 1) * 8;
3906 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
3907 ivar &= ~(0xFF << index);
3908 ivar |= (vector << index);
3909 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
3910 } else { /* RX/TX IVARS */
3911 index = (16 * (entry & 1)) + (8 * type);
3912 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
3913 ivar &= ~(0xFF << index);
3914 ivar |= (vector << index);
3915 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
3925 void __attribute__((cold))
3926 ixgbe_set_rx_function(struct rte_eth_dev *dev)
3928 struct ixgbe_adapter *adapter =
3929 (struct ixgbe_adapter *)dev->data->dev_private;
3932 * In order to allow Vector Rx there are a few configuration
3933 * conditions to be met and Rx Bulk Allocation should be allowed.
3935 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
3936 !adapter->rx_bulk_alloc_allowed) {
3937 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
3938 "preconditions or RTE_IXGBE_INC_VECTOR is "
3940 dev->data->port_id);
3942 adapter->rx_vec_allowed = false;
3946 * Initialize the appropriate LRO callback.
3948 * If all queues satisfy the bulk allocation preconditions
3949 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
3950 * Otherwise use a single allocation version.
3952 if (dev->data->lro) {
3953 if (adapter->rx_bulk_alloc_allowed) {
3954 PMD_INIT_LOG(INFO, "LRO is requested. Using a bulk "
3955 "allocation version");
3956 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
3958 PMD_INIT_LOG(INFO, "LRO is requested. Using a single "
3959 "allocation version");
3960 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
3962 } else if (dev->data->scattered_rx) {
3964 * Set the non-LRO scattered callback: there are Vector and
3965 * single allocation versions.
3967 if (adapter->rx_vec_allowed) {
3968 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
3969 "callback (port=%d).",
3970 dev->data->port_id);
3972 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
3973 } else if (adapter->rx_bulk_alloc_allowed) {
3974 PMD_INIT_LOG(INFO, "Using a Scattered with bulk "
3975 "allocation callback (port=%d).",
3976 dev->data->port_id);
3977 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
3979 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
3980 "single allocation) "
3981 "Scattered Rx callback "
3983 dev->data->port_id);
3985 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
3988 * Below we set "simple" callbacks according to port/queues parameters.
3989 * If parameters allow we are going to choose between the following
3993 * - Single buffer allocation (the simplest one)
3995 } else if (adapter->rx_vec_allowed) {
3996 PMD_INIT_LOG(INFO, "Vector rx enabled, please make sure RX "
3997 "burst size no less than 32.");
3999 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
4000 } else if (adapter->rx_bulk_alloc_allowed) {
4001 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
4002 "satisfied. Rx Burst Bulk Alloc function "
4003 "will be used on port=%d.",
4004 dev->data->port_id);
4006 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
4008 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
4009 "satisfied, or Scattered Rx is requested, "
4010 "or RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC "
4011 "is not enabled (port=%d).",
4012 dev->data->port_id);
4014 dev->rx_pkt_burst = ixgbe_recv_pkts;
4019 * ixgbe_set_rsc - configure RSC related port HW registers
4021 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4022 * of 82599 Spec (x540 configuration is virtually the same).
4026 * Returns 0 in case of success or a non-zero error code
4029 ixgbe_set_rsc(struct rte_eth_dev *dev)
4031 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4032 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4033 struct rte_eth_dev_info dev_info = { 0 };
4034 bool rsc_capable = false;
4039 dev->dev_ops->dev_infos_get(dev, &dev_info);
4040 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4043 if (!rsc_capable && rx_conf->enable_lro) {
4044 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4049 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4051 if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
4053 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4054 * 3.0 RSC configuration requires HW CRC stripping being
4055 * enabled. If user requested both HW CRC stripping off
4056 * and RSC on - return an error.
4058 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4063 /* RFCTL configuration */
4065 uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4066 if (rx_conf->enable_lro)
4068 * Since NFS packets coalescing is not supported - clear
4069 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4072 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4073 IXGBE_RFCTL_NFSR_DIS);
4075 rfctl |= IXGBE_RFCTL_RSC_DIS;
4077 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4080 /* If LRO hasn't been requested - we are done here. */
4081 if (!rx_conf->enable_lro)
4084 /* Set RDRXCTL.RSCACKC bit */
4085 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4086 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4087 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4089 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4090 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4091 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4093 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4095 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4097 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4099 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4102 * ixgbe PMD doesn't support header-split at the moment.
4104 * Following the 4.6.7.2.1 chapter of the 82599/x540
4105 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4106 * should be configured even if header split is not
4107 * enabled. We will configure it 128 bytes following the
4108 * recommendation in the spec.
4110 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4111 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4112 IXGBE_SRRCTL_BSIZEHDR_MASK;
4115 * TODO: Consider setting the Receive Descriptor Minimum
4116 * Threshold Size for an RSC case. This is not an obviously
4117 * beneficiary option but the one worth considering...
4120 rscctl |= IXGBE_RSCCTL_RSCEN;
4121 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4122 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4125 * RSC: Set ITR interval corresponding to 2K ints/s.
4127 * Full-sized RSC aggregations for a 10Gb/s link will
4128 * arrive at about 20K aggregation/s rate.
4130 * 2K inst/s rate will make only 10% of the
4131 * aggregations to be closed due to the interrupt timer
4132 * expiration for a streaming at wire-speed case.
4134 * For a sparse streaming case this setting will yield
4135 * at most 500us latency for a single RSC aggregation.
4137 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4138 eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
4140 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4141 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4142 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4143 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4146 * RSC requires the mapping of the queue to the
4149 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4154 PMD_INIT_LOG(INFO, "enabling LRO mode");
4160 * Initializes Receive Unit.
4162 int __attribute__((cold))
4163 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4165 struct ixgbe_hw *hw;
4166 struct ixgbe_rx_queue *rxq;
4177 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4180 PMD_INIT_FUNC_TRACE();
4181 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4184 * Make sure receives are disabled while setting
4185 * up the RX context (registers, descriptor rings, etc.).
4187 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4188 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4190 /* Enable receipt of broadcasted frames */
4191 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4192 fctrl |= IXGBE_FCTRL_BAM;
4193 fctrl |= IXGBE_FCTRL_DPF;
4194 fctrl |= IXGBE_FCTRL_PMCF;
4195 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4198 * Configure CRC stripping, if any.
4200 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4201 if (rx_conf->hw_strip_crc)
4202 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4204 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4207 * Configure jumbo frame support, if any.
4209 if (rx_conf->jumbo_frame == 1) {
4210 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4211 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4212 maxfrs &= 0x0000FFFF;
4213 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4214 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4216 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4219 * If loopback mode is configured for 82599, set LPBK bit.
4221 if (hw->mac.type == ixgbe_mac_82599EB &&
4222 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4223 hlreg0 |= IXGBE_HLREG0_LPBK;
4225 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4227 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4229 /* Setup RX queues */
4230 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4231 rxq = dev->data->rx_queues[i];
4234 * Reset crc_len in case it was changed after queue setup by a
4235 * call to configure.
4237 rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
4239 /* Setup the Base and Length of the Rx Descriptor Rings */
4240 bus_addr = rxq->rx_ring_phys_addr;
4241 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4242 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4243 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4244 (uint32_t)(bus_addr >> 32));
4245 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4246 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4247 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4248 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4250 /* Configure the SRRCTL register */
4251 #ifdef RTE_HEADER_SPLIT_ENABLE
4253 * Configure Header Split
4255 if (rx_conf->header_split) {
4256 if (hw->mac.type == ixgbe_mac_82599EB) {
4257 /* Must setup the PSRTYPE register */
4259 psrtype = IXGBE_PSRTYPE_TCPHDR |
4260 IXGBE_PSRTYPE_UDPHDR |
4261 IXGBE_PSRTYPE_IPV4HDR |
4262 IXGBE_PSRTYPE_IPV6HDR;
4263 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4265 srrctl = ((rx_conf->split_hdr_size <<
4266 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4267 IXGBE_SRRCTL_BSIZEHDR_MASK);
4268 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4271 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4273 /* Set if packets are dropped when no descriptors available */
4275 srrctl |= IXGBE_SRRCTL_DROP_EN;
4278 * Configure the RX buffer size in the BSIZEPACKET field of
4279 * the SRRCTL register of the queue.
4280 * The value is in 1 KB resolution. Valid values can be from
4283 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4284 RTE_PKTMBUF_HEADROOM);
4285 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4286 IXGBE_SRRCTL_BSIZEPKT_MASK);
4288 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4290 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4291 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4293 /* It adds dual VLAN length for supporting dual VLAN */
4294 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4295 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4296 dev->data->scattered_rx = 1;
4299 if (rx_conf->enable_scatter)
4300 dev->data->scattered_rx = 1;
4303 * Device configured with multiple RX queues.
4305 ixgbe_dev_mq_rx_configure(dev);
4308 * Setup the Checksum Register.
4309 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4310 * Enable IP/L4 checkum computation by hardware if requested to do so.
4312 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4313 rxcsum |= IXGBE_RXCSUM_PCSD;
4314 if (rx_conf->hw_ip_checksum)
4315 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4317 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4319 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4321 if (hw->mac.type == ixgbe_mac_82599EB ||
4322 hw->mac.type == ixgbe_mac_X540) {
4323 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4324 if (rx_conf->hw_strip_crc)
4325 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4327 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4328 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4329 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4332 rc = ixgbe_set_rsc(dev);
4336 ixgbe_set_rx_function(dev);
4342 * Initializes Transmit Unit.
4344 void __attribute__((cold))
4345 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4347 struct ixgbe_hw *hw;
4348 struct ixgbe_tx_queue *txq;
4354 PMD_INIT_FUNC_TRACE();
4355 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4357 /* Enable TX CRC (checksum offload requirement) and hw padding
4358 * (TSO requirement) */
4359 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4360 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
4361 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4363 /* Setup the Base and Length of the Tx Descriptor Rings */
4364 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4365 txq = dev->data->tx_queues[i];
4367 bus_addr = txq->tx_ring_phys_addr;
4368 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
4369 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4370 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
4371 (uint32_t)(bus_addr >> 32));
4372 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
4373 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4374 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4375 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4376 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4379 * Disable Tx Head Writeback RO bit, since this hoses
4380 * bookkeeping if things aren't delivered in order.
4382 switch (hw->mac.type) {
4383 case ixgbe_mac_82598EB:
4384 txctrl = IXGBE_READ_REG(hw,
4385 IXGBE_DCA_TXCTRL(txq->reg_idx));
4386 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4387 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
4391 case ixgbe_mac_82599EB:
4392 case ixgbe_mac_X540:
4393 case ixgbe_mac_X550:
4394 case ixgbe_mac_X550EM_x:
4396 txctrl = IXGBE_READ_REG(hw,
4397 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
4398 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4399 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
4405 /* Device configured with multiple TX queues. */
4406 ixgbe_dev_mq_tx_configure(dev);
4410 * Set up link for 82599 loopback mode Tx->Rx.
4412 static inline void __attribute__((cold))
4413 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
4415 PMD_INIT_FUNC_TRACE();
4417 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
4418 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
4420 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
4429 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
4430 ixgbe_reset_pipeline_82599(hw);
4432 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
4438 * Start Transmit and Receive Units.
4440 int __attribute__((cold))
4441 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
4443 struct ixgbe_hw *hw;
4444 struct ixgbe_tx_queue *txq;
4445 struct ixgbe_rx_queue *rxq;
4452 PMD_INIT_FUNC_TRACE();
4453 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4455 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4456 txq = dev->data->tx_queues[i];
4457 /* Setup Transmit Threshold Registers */
4458 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4459 txdctl |= txq->pthresh & 0x7F;
4460 txdctl |= ((txq->hthresh & 0x7F) << 8);
4461 txdctl |= ((txq->wthresh & 0x7F) << 16);
4462 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4465 if (hw->mac.type != ixgbe_mac_82598EB) {
4466 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
4467 dmatxctl |= IXGBE_DMATXCTL_TE;
4468 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
4471 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4472 txq = dev->data->tx_queues[i];
4473 if (!txq->tx_deferred_start) {
4474 ret = ixgbe_dev_tx_queue_start(dev, i);
4480 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4481 rxq = dev->data->rx_queues[i];
4482 if (!rxq->rx_deferred_start) {
4483 ret = ixgbe_dev_rx_queue_start(dev, i);
4489 /* Enable Receive engine */
4490 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4491 if (hw->mac.type == ixgbe_mac_82598EB)
4492 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4493 rxctrl |= IXGBE_RXCTRL_RXEN;
4494 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4496 /* If loopback mode is enabled for 82599, set up the link accordingly */
4497 if (hw->mac.type == ixgbe_mac_82599EB &&
4498 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4499 ixgbe_setup_loopback_link_82599(hw);
4505 * Start Receive Units for specified queue.
4507 int __attribute__((cold))
4508 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4510 struct ixgbe_hw *hw;
4511 struct ixgbe_rx_queue *rxq;
4515 PMD_INIT_FUNC_TRACE();
4516 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4518 if (rx_queue_id < dev->data->nb_rx_queues) {
4519 rxq = dev->data->rx_queues[rx_queue_id];
4521 /* Allocate buffers for descriptor rings */
4522 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
4523 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
4527 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4528 rxdctl |= IXGBE_RXDCTL_ENABLE;
4529 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4531 /* Wait until RX Enable ready */
4532 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4535 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4536 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4538 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
4541 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4542 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
4550 * Stop Receive Units for specified queue.
4552 int __attribute__((cold))
4553 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4555 struct ixgbe_hw *hw;
4556 struct ixgbe_adapter *adapter =
4557 (struct ixgbe_adapter *)dev->data->dev_private;
4558 struct ixgbe_rx_queue *rxq;
4562 PMD_INIT_FUNC_TRACE();
4563 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4565 if (rx_queue_id < dev->data->nb_rx_queues) {
4566 rxq = dev->data->rx_queues[rx_queue_id];
4568 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4569 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4570 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4572 /* Wait until RX Enable ready */
4573 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4576 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4577 } while (--poll_ms && (rxdctl | IXGBE_RXDCTL_ENABLE));
4579 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
4582 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4584 ixgbe_rx_queue_release_mbufs(rxq);
4585 ixgbe_reset_rx_queue(adapter, rxq);
4594 * Start Transmit Units for specified queue.
4596 int __attribute__((cold))
4597 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4599 struct ixgbe_hw *hw;
4600 struct ixgbe_tx_queue *txq;
4604 PMD_INIT_FUNC_TRACE();
4605 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4607 if (tx_queue_id < dev->data->nb_tx_queues) {
4608 txq = dev->data->tx_queues[tx_queue_id];
4609 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4610 txdctl |= IXGBE_TXDCTL_ENABLE;
4611 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4613 /* Wait until TX Enable ready */
4614 if (hw->mac.type == ixgbe_mac_82599EB) {
4615 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4618 txdctl = IXGBE_READ_REG(hw,
4619 IXGBE_TXDCTL(txq->reg_idx));
4620 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4622 PMD_INIT_LOG(ERR, "Could not enable "
4623 "Tx Queue %d", tx_queue_id);
4626 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4627 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4635 * Stop Transmit Units for specified queue.
4637 int __attribute__((cold))
4638 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4640 struct ixgbe_hw *hw;
4641 struct ixgbe_tx_queue *txq;
4643 uint32_t txtdh, txtdt;
4646 PMD_INIT_FUNC_TRACE();
4647 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4649 if (tx_queue_id < dev->data->nb_tx_queues) {
4650 txq = dev->data->tx_queues[tx_queue_id];
4652 /* Wait until TX queue is empty */
4653 if (hw->mac.type == ixgbe_mac_82599EB) {
4654 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4656 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4657 txtdh = IXGBE_READ_REG(hw,
4658 IXGBE_TDH(txq->reg_idx));
4659 txtdt = IXGBE_READ_REG(hw,
4660 IXGBE_TDT(txq->reg_idx));
4661 } while (--poll_ms && (txtdh != txtdt));
4663 PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
4664 "when stopping.", tx_queue_id);
4667 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4668 txdctl &= ~IXGBE_TXDCTL_ENABLE;
4669 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4671 /* Wait until TX Enable ready */
4672 if (hw->mac.type == ixgbe_mac_82599EB) {
4673 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4676 txdctl = IXGBE_READ_REG(hw,
4677 IXGBE_TXDCTL(txq->reg_idx));
4678 } while (--poll_ms && (txdctl | IXGBE_TXDCTL_ENABLE));
4680 PMD_INIT_LOG(ERR, "Could not disable "
4681 "Tx Queue %d", tx_queue_id);
4684 if (txq->ops != NULL) {
4685 txq->ops->release_mbufs(txq);
4686 txq->ops->reset(txq);
4695 * [VF] Initializes Receive Unit.
4697 int __attribute__((cold))
4698 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
4700 struct ixgbe_hw *hw;
4701 struct ixgbe_rx_queue *rxq;
4703 uint32_t srrctl, psrtype = 0;
4708 PMD_INIT_FUNC_TRACE();
4709 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4711 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
4712 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
4713 "it should be power of 2");
4717 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
4718 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
4719 "it should be equal to or less than %d",
4720 hw->mac.max_rx_queues);
4725 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
4726 * disables the VF receipt of packets if the PF MTU is > 1500.
4727 * This is done to deal with 82599 limitations that imposes
4728 * the PF and all VFs to share the same MTU.
4729 * Then, the PF driver enables again the VF receipt of packet when
4730 * the VF driver issues a IXGBE_VF_SET_LPE request.
4731 * In the meantime, the VF device cannot be used, even if the VF driver
4732 * and the Guest VM network stack are ready to accept packets with a
4733 * size up to the PF MTU.
4734 * As a work-around to this PF behaviour, force the call to
4735 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
4736 * VF packets received can work in all cases.
4738 ixgbevf_rlpml_set_vf(hw,
4739 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
4741 /* Setup RX queues */
4742 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4743 rxq = dev->data->rx_queues[i];
4745 /* Allocate buffers for descriptor rings */
4746 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
4750 /* Setup the Base and Length of the Rx Descriptor Rings */
4751 bus_addr = rxq->rx_ring_phys_addr;
4753 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
4754 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4755 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
4756 (uint32_t)(bus_addr >> 32));
4757 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
4758 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4759 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
4760 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
4763 /* Configure the SRRCTL register */
4764 #ifdef RTE_HEADER_SPLIT_ENABLE
4766 * Configure Header Split
4768 if (dev->data->dev_conf.rxmode.header_split) {
4769 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
4770 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4771 IXGBE_SRRCTL_BSIZEHDR_MASK);
4772 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4775 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4777 /* Set if packets are dropped when no descriptors available */
4779 srrctl |= IXGBE_SRRCTL_DROP_EN;
4782 * Configure the RX buffer size in the BSIZEPACKET field of
4783 * the SRRCTL register of the queue.
4784 * The value is in 1 KB resolution. Valid values can be from
4787 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4788 RTE_PKTMBUF_HEADROOM);
4789 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4790 IXGBE_SRRCTL_BSIZEPKT_MASK);
4793 * VF modification to write virtual function SRRCTL register
4795 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
4797 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4798 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4800 if (dev->data->dev_conf.rxmode.enable_scatter ||
4801 /* It adds dual VLAN length for supporting dual VLAN */
4802 (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4803 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
4804 if (!dev->data->scattered_rx)
4805 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
4806 dev->data->scattered_rx = 1;
4810 #ifdef RTE_HEADER_SPLIT_ENABLE
4811 if (dev->data->dev_conf.rxmode.header_split)
4812 /* Must setup the PSRTYPE register */
4813 psrtype = IXGBE_PSRTYPE_TCPHDR |
4814 IXGBE_PSRTYPE_UDPHDR |
4815 IXGBE_PSRTYPE_IPV4HDR |
4816 IXGBE_PSRTYPE_IPV6HDR;
4819 /* Set RQPL for VF RSS according to max Rx queue */
4820 psrtype |= (dev->data->nb_rx_queues >> 1) <<
4821 IXGBE_PSRTYPE_RQPL_SHIFT;
4822 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
4824 ixgbe_set_rx_function(dev);
4830 * [VF] Initializes Transmit Unit.
4832 void __attribute__((cold))
4833 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
4835 struct ixgbe_hw *hw;
4836 struct ixgbe_tx_queue *txq;
4841 PMD_INIT_FUNC_TRACE();
4842 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4844 /* Setup the Base and Length of the Tx Descriptor Rings */
4845 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4846 txq = dev->data->tx_queues[i];
4847 bus_addr = txq->tx_ring_phys_addr;
4848 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
4849 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4850 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
4851 (uint32_t)(bus_addr >> 32));
4852 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
4853 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4854 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4855 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
4856 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
4859 * Disable Tx Head Writeback RO bit, since this hoses
4860 * bookkeeping if things aren't delivered in order.
4862 txctrl = IXGBE_READ_REG(hw,
4863 IXGBE_VFDCA_TXCTRL(i));
4864 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4865 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
4871 * [VF] Start Transmit and Receive Units.
4873 void __attribute__((cold))
4874 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
4876 struct ixgbe_hw *hw;
4877 struct ixgbe_tx_queue *txq;
4878 struct ixgbe_rx_queue *rxq;
4884 PMD_INIT_FUNC_TRACE();
4885 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4887 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4888 txq = dev->data->tx_queues[i];
4889 /* Setup Transmit Threshold Registers */
4890 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4891 txdctl |= txq->pthresh & 0x7F;
4892 txdctl |= ((txq->hthresh & 0x7F) << 8);
4893 txdctl |= ((txq->wthresh & 0x7F) << 16);
4894 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
4897 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4899 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4900 txdctl |= IXGBE_TXDCTL_ENABLE;
4901 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
4904 /* Wait until TX Enable ready */
4907 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4908 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4910 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
4912 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4914 rxq = dev->data->rx_queues[i];
4916 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
4917 rxdctl |= IXGBE_RXDCTL_ENABLE;
4918 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
4920 /* Wait until RX Enable ready */
4924 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
4925 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4927 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
4929 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
4934 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
4935 int __attribute__((weak))
4936 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
4941 uint16_t __attribute__((weak))
4942 ixgbe_recv_pkts_vec(
4943 void __rte_unused *rx_queue,
4944 struct rte_mbuf __rte_unused **rx_pkts,
4945 uint16_t __rte_unused nb_pkts)
4950 uint16_t __attribute__((weak))
4951 ixgbe_recv_scattered_pkts_vec(
4952 void __rte_unused *rx_queue,
4953 struct rte_mbuf __rte_unused **rx_pkts,
4954 uint16_t __rte_unused nb_pkts)
4959 int __attribute__((weak))
4960 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)