4 * Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
5 * Copyright 2014 6WIND S.A.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * * Neither the name of Intel Corporation nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <sys/queue.h>
46 #include <rte_byteorder.h>
47 #include <rte_common.h>
48 #include <rte_cycles.h>
50 #include <rte_debug.h>
51 #include <rte_interrupts.h>
53 #include <rte_memory.h>
54 #include <rte_memzone.h>
55 #include <rte_launch.h>
57 #include <rte_per_lcore.h>
58 #include <rte_lcore.h>
59 #include <rte_atomic.h>
60 #include <rte_branch_prediction.h>
62 #include <rte_mempool.h>
63 #include <rte_malloc.h>
65 #include <rte_ether.h>
66 #include <rte_ethdev.h>
67 #include <rte_prefetch.h>
71 #include <rte_string_fns.h>
72 #include <rte_errno.h>
75 #include "ixgbe_logs.h"
76 #include "base/ixgbe_api.h"
77 #include "base/ixgbe_vf.h"
78 #include "ixgbe_ethdev.h"
79 #include "base/ixgbe_dcb.h"
80 #include "base/ixgbe_common.h"
81 #include "ixgbe_rxtx.h"
83 /* Bit Mask to indicate what bits required for building TX context */
84 #define IXGBE_TX_OFFLOAD_MASK ( \
90 static inline struct rte_mbuf *
91 rte_rxmbuf_alloc(struct rte_mempool *mp)
95 m = __rte_mbuf_raw_alloc(mp);
96 __rte_mbuf_sanity_check_raw(m, 0);
102 #define RTE_PMD_USE_PREFETCH
105 #ifdef RTE_PMD_USE_PREFETCH
107 * Prefetch a cache line into all cache levels.
109 #define rte_ixgbe_prefetch(p) rte_prefetch0(p)
111 #define rte_ixgbe_prefetch(p) do {} while(0)
114 /*********************************************************************
118 **********************************************************************/
121 * Check for descriptors with their DD bit set and free mbufs.
122 * Return the total number of buffers freed.
124 static inline int __attribute__((always_inline))
125 ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
127 struct ixgbe_tx_entry *txep;
131 /* check DD bit on threshold descriptor */
132 status = txq->tx_ring[txq->tx_next_dd].wb.status;
133 if (! (status & IXGBE_ADVTXD_STAT_DD))
137 * first buffer to free from S/W ring is at index
138 * tx_next_dd - (tx_rs_thresh-1)
140 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
142 /* free buffers one at a time */
143 if ((txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT) != 0) {
144 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
145 txep->mbuf->next = NULL;
146 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
150 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
151 rte_pktmbuf_free_seg(txep->mbuf);
156 /* buffers were freed, update counters */
157 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
158 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
159 if (txq->tx_next_dd >= txq->nb_tx_desc)
160 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
162 return txq->tx_rs_thresh;
165 /* Populate 4 descriptors with data from 4 mbufs */
167 tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
169 uint64_t buf_dma_addr;
173 for (i = 0; i < 4; ++i, ++txdp, ++pkts) {
174 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
175 pkt_len = (*pkts)->data_len;
177 /* write data to descriptor */
178 txdp->read.buffer_addr = buf_dma_addr;
179 txdp->read.cmd_type_len =
180 ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
181 txdp->read.olinfo_status =
182 (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
183 rte_prefetch0(&(*pkts)->pool);
187 /* Populate 1 descriptor with data from 1 mbuf */
189 tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
191 uint64_t buf_dma_addr;
194 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
195 pkt_len = (*pkts)->data_len;
197 /* write data to descriptor */
198 txdp->read.buffer_addr = buf_dma_addr;
199 txdp->read.cmd_type_len =
200 ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
201 txdp->read.olinfo_status =
202 (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
203 rte_prefetch0(&(*pkts)->pool);
207 * Fill H/W descriptor ring with mbuf data.
208 * Copy mbuf pointers to the S/W ring.
211 ixgbe_tx_fill_hw_ring(struct ixgbe_tx_queue *txq, struct rte_mbuf **pkts,
214 volatile union ixgbe_adv_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
215 struct ixgbe_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
216 const int N_PER_LOOP = 4;
217 const int N_PER_LOOP_MASK = N_PER_LOOP-1;
218 int mainpart, leftover;
222 * Process most of the packets in chunks of N pkts. Any
223 * leftover packets will get processed one at a time.
225 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
226 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
227 for (i = 0; i < mainpart; i += N_PER_LOOP) {
228 /* Copy N mbuf pointers to the S/W ring */
229 for (j = 0; j < N_PER_LOOP; ++j) {
230 (txep + i + j)->mbuf = *(pkts + i + j);
232 tx4(txdp + i, pkts + i);
235 if (unlikely(leftover > 0)) {
236 for (i = 0; i < leftover; ++i) {
237 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
238 tx1(txdp + mainpart + i, pkts + mainpart + i);
243 static inline uint16_t
244 tx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
247 struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
248 volatile union ixgbe_adv_tx_desc *tx_r = txq->tx_ring;
252 * Begin scanning the H/W ring for done descriptors when the
253 * number of available descriptors drops below tx_free_thresh. For
254 * each done descriptor, free the associated buffer.
256 if (txq->nb_tx_free < txq->tx_free_thresh)
257 ixgbe_tx_free_bufs(txq);
259 /* Only use descriptors that are available */
260 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
261 if (unlikely(nb_pkts == 0))
264 /* Use exactly nb_pkts descriptors */
265 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
268 * At this point, we know there are enough descriptors in the
269 * ring to transmit all the packets. This assumes that each
270 * mbuf contains a single segment, and that no new offloads
271 * are expected, which would require a new context descriptor.
275 * See if we're going to wrap-around. If so, handle the top
276 * of the descriptor ring first, then do the bottom. If not,
277 * the processing looks just like the "bottom" part anyway...
279 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
280 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
281 ixgbe_tx_fill_hw_ring(txq, tx_pkts, n);
284 * We know that the last descriptor in the ring will need to
285 * have its RS bit set because tx_rs_thresh has to be
286 * a divisor of the ring size
288 tx_r[txq->tx_next_rs].read.cmd_type_len |=
289 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
290 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
295 /* Fill H/W descriptor ring with mbuf data */
296 ixgbe_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
297 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
300 * Determine if RS bit should be set
301 * This is what we actually want:
302 * if ((txq->tx_tail - 1) >= txq->tx_next_rs)
303 * but instead of subtracting 1 and doing >=, we can just do
304 * greater than without subtracting.
306 if (txq->tx_tail > txq->tx_next_rs) {
307 tx_r[txq->tx_next_rs].read.cmd_type_len |=
308 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
309 txq->tx_next_rs = (uint16_t)(txq->tx_next_rs +
311 if (txq->tx_next_rs >= txq->nb_tx_desc)
312 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
316 * Check for wrap-around. This would only happen if we used
317 * up to the last descriptor in the ring, no more, no less.
319 if (txq->tx_tail >= txq->nb_tx_desc)
322 /* update tail pointer */
324 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, txq->tx_tail);
330 ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
335 /* Try to transmit at least chunks of TX_MAX_BURST pkts */
336 if (likely(nb_pkts <= RTE_PMD_IXGBE_TX_MAX_BURST))
337 return tx_xmit_pkts(tx_queue, tx_pkts, nb_pkts);
339 /* transmit more than the max burst, in chunks of TX_MAX_BURST */
343 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_TX_MAX_BURST);
344 ret = tx_xmit_pkts(tx_queue, &(tx_pkts[nb_tx]), n);
345 nb_tx = (uint16_t)(nb_tx + ret);
346 nb_pkts = (uint16_t)(nb_pkts - ret);
355 ixgbe_set_xmit_ctx(struct ixgbe_tx_queue *txq,
356 volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
357 uint64_t ol_flags, union ixgbe_tx_offload tx_offload)
359 uint32_t type_tucmd_mlhl;
360 uint32_t mss_l4len_idx = 0;
362 uint32_t vlan_macip_lens;
363 union ixgbe_tx_offload tx_offload_mask;
365 ctx_idx = txq->ctx_curr;
366 tx_offload_mask.data = 0;
369 /* Specify which HW CTX to upload. */
370 mss_l4len_idx |= (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
372 if (ol_flags & PKT_TX_VLAN_PKT) {
373 tx_offload_mask.vlan_tci |= ~0;
376 /* check if TCP segmentation required for this packet */
377 if (ol_flags & PKT_TX_TCP_SEG) {
378 /* implies IP cksum in IPv4 */
379 if (ol_flags & PKT_TX_IP_CKSUM)
380 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4 |
381 IXGBE_ADVTXD_TUCMD_L4T_TCP |
382 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
384 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV6 |
385 IXGBE_ADVTXD_TUCMD_L4T_TCP |
386 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
388 tx_offload_mask.l2_len |= ~0;
389 tx_offload_mask.l3_len |= ~0;
390 tx_offload_mask.l4_len |= ~0;
391 tx_offload_mask.tso_segsz |= ~0;
392 mss_l4len_idx |= tx_offload.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT;
393 mss_l4len_idx |= tx_offload.l4_len << IXGBE_ADVTXD_L4LEN_SHIFT;
394 } else { /* no TSO, check if hardware checksum is needed */
395 if (ol_flags & PKT_TX_IP_CKSUM) {
396 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
397 tx_offload_mask.l2_len |= ~0;
398 tx_offload_mask.l3_len |= ~0;
401 switch (ol_flags & PKT_TX_L4_MASK) {
402 case PKT_TX_UDP_CKSUM:
403 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
404 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
405 mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
406 tx_offload_mask.l2_len |= ~0;
407 tx_offload_mask.l3_len |= ~0;
409 case PKT_TX_TCP_CKSUM:
410 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
411 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
412 mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
413 tx_offload_mask.l2_len |= ~0;
414 tx_offload_mask.l3_len |= ~0;
415 tx_offload_mask.l4_len |= ~0;
417 case PKT_TX_SCTP_CKSUM:
418 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
419 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
420 mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
421 tx_offload_mask.l2_len |= ~0;
422 tx_offload_mask.l3_len |= ~0;
425 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
426 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
431 txq->ctx_cache[ctx_idx].flags = ol_flags;
432 txq->ctx_cache[ctx_idx].tx_offload.data =
433 tx_offload_mask.data & tx_offload.data;
434 txq->ctx_cache[ctx_idx].tx_offload_mask = tx_offload_mask;
436 ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
437 vlan_macip_lens = tx_offload.l3_len;
438 vlan_macip_lens |= (tx_offload.l2_len << IXGBE_ADVTXD_MACLEN_SHIFT);
439 vlan_macip_lens |= ((uint32_t)tx_offload.vlan_tci << IXGBE_ADVTXD_VLAN_SHIFT);
440 ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
441 ctx_txd->mss_l4len_idx = rte_cpu_to_le_32(mss_l4len_idx);
442 ctx_txd->seqnum_seed = 0;
446 * Check which hardware context can be used. Use the existing match
447 * or create a new context descriptor.
449 static inline uint32_t
450 what_advctx_update(struct ixgbe_tx_queue *txq, uint64_t flags,
451 union ixgbe_tx_offload tx_offload)
453 /* If match with the current used context */
454 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
455 (txq->ctx_cache[txq->ctx_curr].tx_offload.data ==
456 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data & tx_offload.data)))) {
457 return txq->ctx_curr;
460 /* What if match with the next context */
462 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
463 (txq->ctx_cache[txq->ctx_curr].tx_offload.data ==
464 (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data & tx_offload.data)))) {
465 return txq->ctx_curr;
468 /* Mismatch, use the previous context */
469 return (IXGBE_CTX_NUM);
472 static inline uint32_t
473 tx_desc_cksum_flags_to_olinfo(uint64_t ol_flags)
476 if ((ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM)
477 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
478 if (ol_flags & PKT_TX_IP_CKSUM)
479 tmp |= IXGBE_ADVTXD_POPTS_IXSM;
480 if (ol_flags & PKT_TX_TCP_SEG)
481 tmp |= IXGBE_ADVTXD_POPTS_TXSM;
485 static inline uint32_t
486 tx_desc_ol_flags_to_cmdtype(uint64_t ol_flags)
488 uint32_t cmdtype = 0;
489 if (ol_flags & PKT_TX_VLAN_PKT)
490 cmdtype |= IXGBE_ADVTXD_DCMD_VLE;
491 if (ol_flags & PKT_TX_TCP_SEG)
492 cmdtype |= IXGBE_ADVTXD_DCMD_TSE;
496 /* Default RS bit threshold values */
497 #ifndef DEFAULT_TX_RS_THRESH
498 #define DEFAULT_TX_RS_THRESH 32
500 #ifndef DEFAULT_TX_FREE_THRESH
501 #define DEFAULT_TX_FREE_THRESH 32
504 /* Reset transmit descriptors after they have been used */
506 ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
508 struct ixgbe_tx_entry *sw_ring = txq->sw_ring;
509 volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
510 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
511 uint16_t nb_tx_desc = txq->nb_tx_desc;
512 uint16_t desc_to_clean_to;
513 uint16_t nb_tx_to_clean;
515 /* Determine the last descriptor needing to be cleaned */
516 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
517 if (desc_to_clean_to >= nb_tx_desc)
518 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
520 /* Check to make sure the last descriptor to clean is done */
521 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
522 if (! (txr[desc_to_clean_to].wb.status & IXGBE_TXD_STAT_DD))
524 PMD_TX_FREE_LOG(DEBUG,
525 "TX descriptor %4u is not done"
526 "(port=%d queue=%d)",
528 txq->port_id, txq->queue_id);
529 /* Failed to clean any descriptors, better luck next time */
533 /* Figure out how many descriptors will be cleaned */
534 if (last_desc_cleaned > desc_to_clean_to)
535 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
538 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
541 PMD_TX_FREE_LOG(DEBUG,
542 "Cleaning %4u TX descriptors: %4u to %4u "
543 "(port=%d queue=%d)",
544 nb_tx_to_clean, last_desc_cleaned, desc_to_clean_to,
545 txq->port_id, txq->queue_id);
548 * The last descriptor to clean is done, so that means all the
549 * descriptors from the last descriptor that was cleaned
550 * up to the last descriptor with the RS bit set
551 * are done. Only reset the threshold descriptor.
553 txr[desc_to_clean_to].wb.status = 0;
555 /* Update the txq to reflect the last descriptor that was cleaned */
556 txq->last_desc_cleaned = desc_to_clean_to;
557 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
564 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
567 struct ixgbe_tx_queue *txq;
568 struct ixgbe_tx_entry *sw_ring;
569 struct ixgbe_tx_entry *txe, *txn;
570 volatile union ixgbe_adv_tx_desc *txr;
571 volatile union ixgbe_adv_tx_desc *txd;
572 struct rte_mbuf *tx_pkt;
573 struct rte_mbuf *m_seg;
574 uint64_t buf_dma_addr;
575 uint32_t olinfo_status;
576 uint32_t cmd_type_len;
587 union ixgbe_tx_offload tx_offload = {0};
590 sw_ring = txq->sw_ring;
592 tx_id = txq->tx_tail;
593 txe = &sw_ring[tx_id];
595 /* Determine if the descriptor ring needs to be cleaned. */
596 if (txq->nb_tx_free < txq->tx_free_thresh)
597 ixgbe_xmit_cleanup(txq);
599 rte_prefetch0(&txe->mbuf->pool);
602 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
605 pkt_len = tx_pkt->pkt_len;
608 * Determine how many (if any) context descriptors
609 * are needed for offload functionality.
611 ol_flags = tx_pkt->ol_flags;
613 /* If hardware offload required */
614 tx_ol_req = ol_flags & IXGBE_TX_OFFLOAD_MASK;
616 tx_offload.l2_len = tx_pkt->l2_len;
617 tx_offload.l3_len = tx_pkt->l3_len;
618 tx_offload.l4_len = tx_pkt->l4_len;
619 tx_offload.vlan_tci = tx_pkt->vlan_tci;
620 tx_offload.tso_segsz = tx_pkt->tso_segsz;
622 /* If new context need be built or reuse the exist ctx. */
623 ctx = what_advctx_update(txq, tx_ol_req,
625 /* Only allocate context descriptor if required*/
626 new_ctx = (ctx == IXGBE_CTX_NUM);
631 * Keep track of how many descriptors are used this loop
632 * This will always be the number of segments + the number of
633 * Context descriptors required to transmit the packet
635 nb_used = (uint16_t)(tx_pkt->nb_segs + new_ctx);
638 * The number of descriptors that must be allocated for a
639 * packet is the number of segments of that packet, plus 1
640 * Context Descriptor for the hardware offload, if any.
641 * Determine the last TX descriptor to allocate in the TX ring
642 * for the packet, starting from the current position (tx_id)
645 tx_last = (uint16_t) (tx_id + nb_used - 1);
648 if (tx_last >= txq->nb_tx_desc)
649 tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
651 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
652 " tx_first=%u tx_last=%u",
653 (unsigned) txq->port_id,
654 (unsigned) txq->queue_id,
660 * Make sure there are enough TX descriptors available to
661 * transmit the entire packet.
662 * nb_used better be less than or equal to txq->tx_rs_thresh
664 if (nb_used > txq->nb_tx_free) {
665 PMD_TX_FREE_LOG(DEBUG,
666 "Not enough free TX descriptors "
667 "nb_used=%4u nb_free=%4u "
668 "(port=%d queue=%d)",
669 nb_used, txq->nb_tx_free,
670 txq->port_id, txq->queue_id);
672 if (ixgbe_xmit_cleanup(txq) != 0) {
673 /* Could not clean any descriptors */
679 /* nb_used better be <= txq->tx_rs_thresh */
680 if (unlikely(nb_used > txq->tx_rs_thresh)) {
681 PMD_TX_FREE_LOG(DEBUG,
682 "The number of descriptors needed to "
683 "transmit the packet exceeds the "
684 "RS bit threshold. This will impact "
686 "nb_used=%4u nb_free=%4u "
688 "(port=%d queue=%d)",
689 nb_used, txq->nb_tx_free,
691 txq->port_id, txq->queue_id);
693 * Loop here until there are enough TX
694 * descriptors or until the ring cannot be
697 while (nb_used > txq->nb_tx_free) {
698 if (ixgbe_xmit_cleanup(txq) != 0) {
700 * Could not clean any
712 * By now there are enough free TX descriptors to transmit
717 * Set common flags of all TX Data Descriptors.
719 * The following bits must be set in all Data Descriptors:
720 * - IXGBE_ADVTXD_DTYP_DATA
721 * - IXGBE_ADVTXD_DCMD_DEXT
723 * The following bits must be set in the first Data Descriptor
724 * and are ignored in the other ones:
725 * - IXGBE_ADVTXD_DCMD_IFCS
726 * - IXGBE_ADVTXD_MAC_1588
727 * - IXGBE_ADVTXD_DCMD_VLE
729 * The following bits must only be set in the last Data
731 * - IXGBE_TXD_CMD_EOP
733 * The following bits can be set in any Data Descriptor, but
734 * are only set in the last Data Descriptor:
737 cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
738 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
740 #ifdef RTE_LIBRTE_IEEE1588
741 if (ol_flags & PKT_TX_IEEE1588_TMST)
742 cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
748 if (ol_flags & PKT_TX_TCP_SEG) {
749 /* when TSO is on, paylen in descriptor is the
750 * not the packet len but the tcp payload len */
751 pkt_len -= (tx_offload.l2_len +
752 tx_offload.l3_len + tx_offload.l4_len);
756 * Setup the TX Advanced Context Descriptor if required
759 volatile struct ixgbe_adv_tx_context_desc *
762 ctx_txd = (volatile struct
763 ixgbe_adv_tx_context_desc *)
766 txn = &sw_ring[txe->next_id];
767 rte_prefetch0(&txn->mbuf->pool);
769 if (txe->mbuf != NULL) {
770 rte_pktmbuf_free_seg(txe->mbuf);
774 ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
777 txe->last_id = tx_last;
778 tx_id = txe->next_id;
783 * Setup the TX Advanced Data Descriptor,
784 * This path will go through
785 * whatever new/reuse the context descriptor
787 cmd_type_len |= tx_desc_ol_flags_to_cmdtype(ol_flags);
788 olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
789 olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
792 olinfo_status |= (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
797 txn = &sw_ring[txe->next_id];
798 rte_prefetch0(&txn->mbuf->pool);
800 if (txe->mbuf != NULL)
801 rte_pktmbuf_free_seg(txe->mbuf);
805 * Set up Transmit Data Descriptor.
807 slen = m_seg->data_len;
808 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
809 txd->read.buffer_addr =
810 rte_cpu_to_le_64(buf_dma_addr);
811 txd->read.cmd_type_len =
812 rte_cpu_to_le_32(cmd_type_len | slen);
813 txd->read.olinfo_status =
814 rte_cpu_to_le_32(olinfo_status);
815 txe->last_id = tx_last;
816 tx_id = txe->next_id;
819 } while (m_seg != NULL);
822 * The last packet data descriptor needs End Of Packet (EOP)
824 cmd_type_len |= IXGBE_TXD_CMD_EOP;
825 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
826 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
828 /* Set RS bit only on threshold packets' last descriptor */
829 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
830 PMD_TX_FREE_LOG(DEBUG,
831 "Setting RS bit on TXD id="
832 "%4u (port=%d queue=%d)",
833 tx_last, txq->port_id, txq->queue_id);
835 cmd_type_len |= IXGBE_TXD_CMD_RS;
837 /* Update txq RS bit counters */
840 txd->read.cmd_type_len |= rte_cpu_to_le_32(cmd_type_len);
846 * Set the Transmit Descriptor Tail (TDT)
848 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
849 (unsigned) txq->port_id, (unsigned) txq->queue_id,
850 (unsigned) tx_id, (unsigned) nb_tx);
851 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, tx_id);
852 txq->tx_tail = tx_id;
857 /*********************************************************************
861 **********************************************************************/
863 #define IXGBE_PACKET_TYPE_IPV4 0X01
864 #define IXGBE_PACKET_TYPE_IPV4_TCP 0X11
865 #define IXGBE_PACKET_TYPE_IPV4_UDP 0X21
866 #define IXGBE_PACKET_TYPE_IPV4_SCTP 0X41
867 #define IXGBE_PACKET_TYPE_IPV4_EXT 0X03
868 #define IXGBE_PACKET_TYPE_IPV4_EXT_SCTP 0X43
869 #define IXGBE_PACKET_TYPE_IPV6 0X04
870 #define IXGBE_PACKET_TYPE_IPV6_TCP 0X14
871 #define IXGBE_PACKET_TYPE_IPV6_UDP 0X24
872 #define IXGBE_PACKET_TYPE_IPV6_EXT 0X0C
873 #define IXGBE_PACKET_TYPE_IPV6_EXT_TCP 0X1C
874 #define IXGBE_PACKET_TYPE_IPV6_EXT_UDP 0X2C
875 #define IXGBE_PACKET_TYPE_IPV4_IPV6 0X05
876 #define IXGBE_PACKET_TYPE_IPV4_IPV6_TCP 0X15
877 #define IXGBE_PACKET_TYPE_IPV4_IPV6_UDP 0X25
878 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT 0X0D
879 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP 0X1D
880 #define IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP 0X2D
881 #define IXGBE_PACKET_TYPE_MAX 0X80
882 #define IXGBE_PACKET_TYPE_MASK 0X7F
883 #define IXGBE_PACKET_TYPE_SHIFT 0X04
884 static inline uint32_t
885 ixgbe_rxd_pkt_info_to_pkt_type(uint16_t pkt_info)
887 static const uint32_t
888 ptype_table[IXGBE_PACKET_TYPE_MAX] __rte_cache_aligned = {
889 [IXGBE_PACKET_TYPE_IPV4] = RTE_PTYPE_L2_ETHER |
891 [IXGBE_PACKET_TYPE_IPV4_EXT] = RTE_PTYPE_L2_ETHER |
892 RTE_PTYPE_L3_IPV4_EXT,
893 [IXGBE_PACKET_TYPE_IPV6] = RTE_PTYPE_L2_ETHER |
895 [IXGBE_PACKET_TYPE_IPV4_IPV6] = RTE_PTYPE_L2_ETHER |
896 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
897 RTE_PTYPE_INNER_L3_IPV6,
898 [IXGBE_PACKET_TYPE_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
899 RTE_PTYPE_L3_IPV6_EXT,
900 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT] = RTE_PTYPE_L2_ETHER |
901 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
902 RTE_PTYPE_INNER_L3_IPV6_EXT,
903 [IXGBE_PACKET_TYPE_IPV4_TCP] = RTE_PTYPE_L2_ETHER |
904 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
905 [IXGBE_PACKET_TYPE_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
906 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
907 [IXGBE_PACKET_TYPE_IPV4_IPV6_TCP] = RTE_PTYPE_L2_ETHER |
908 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
909 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_TCP,
910 [IXGBE_PACKET_TYPE_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
911 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP,
912 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_TCP] = RTE_PTYPE_L2_ETHER |
913 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
914 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_TCP,
915 [IXGBE_PACKET_TYPE_IPV4_UDP] = RTE_PTYPE_L2_ETHER |
916 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
917 [IXGBE_PACKET_TYPE_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
918 RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
919 [IXGBE_PACKET_TYPE_IPV4_IPV6_UDP] = RTE_PTYPE_L2_ETHER |
920 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
921 RTE_PTYPE_INNER_L3_IPV6 | RTE_PTYPE_INNER_L4_UDP,
922 [IXGBE_PACKET_TYPE_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
923 RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP,
924 [IXGBE_PACKET_TYPE_IPV4_IPV6_EXT_UDP] = RTE_PTYPE_L2_ETHER |
925 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_TUNNEL_IP |
926 RTE_PTYPE_INNER_L3_IPV6_EXT | RTE_PTYPE_INNER_L4_UDP,
927 [IXGBE_PACKET_TYPE_IPV4_SCTP] = RTE_PTYPE_L2_ETHER |
928 RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP,
929 [IXGBE_PACKET_TYPE_IPV4_EXT_SCTP] = RTE_PTYPE_L2_ETHER |
930 RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_SCTP,
932 if (unlikely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
933 return RTE_PTYPE_UNKNOWN;
935 pkt_info = (pkt_info >> IXGBE_PACKET_TYPE_SHIFT) &
936 IXGBE_PACKET_TYPE_MASK;
938 return ptype_table[pkt_info];
941 static inline uint64_t
942 ixgbe_rxd_pkt_info_to_pkt_flags(uint16_t pkt_info)
944 static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
945 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
946 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
947 PKT_RX_RSS_HASH, 0, 0, 0,
948 0, 0, 0, PKT_RX_FDIR,
950 #ifdef RTE_LIBRTE_IEEE1588
951 static uint64_t ip_pkt_etqf_map[8] = {
952 0, 0, 0, PKT_RX_IEEE1588_PTP,
956 if (likely(pkt_info & IXGBE_RXDADV_PKTTYPE_ETQF))
957 return ip_pkt_etqf_map[(pkt_info >> 4) & 0X07] |
958 ip_rss_types_map[pkt_info & 0XF];
960 return ip_rss_types_map[pkt_info & 0XF];
962 return ip_rss_types_map[pkt_info & 0XF];
965 #else /* RTE_NEXT_ABI */
966 static inline uint64_t
967 rx_desc_hlen_type_rss_to_pkt_flags(uint32_t hl_tp_rs)
971 static const uint64_t ip_pkt_types_map[16] = {
972 0, PKT_RX_IPV4_HDR, PKT_RX_IPV4_HDR_EXT, PKT_RX_IPV4_HDR_EXT,
973 PKT_RX_IPV6_HDR, 0, 0, 0,
974 PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
975 PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
978 static const uint64_t ip_rss_types_map[16] = {
979 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
980 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
981 PKT_RX_RSS_HASH, 0, 0, 0,
982 0, 0, 0, PKT_RX_FDIR,
985 #ifdef RTE_LIBRTE_IEEE1588
986 static uint64_t ip_pkt_etqf_map[8] = {
987 0, 0, 0, PKT_RX_IEEE1588_PTP,
991 pkt_flags = (hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ?
992 ip_pkt_etqf_map[(hl_tp_rs >> 4) & 0x07] :
993 ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
995 pkt_flags = (hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ? 0 :
996 ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
999 return pkt_flags | ip_rss_types_map[hl_tp_rs & 0xF];
1001 #endif /* RTE_NEXT_ABI */
1003 static inline uint64_t
1004 rx_desc_status_to_pkt_flags(uint32_t rx_status)
1009 * Check if VLAN present only.
1010 * Do not check whether L3/L4 rx checksum done by NIC or not,
1011 * That can be found from rte_eth_rxmode.hw_ip_checksum flag
1013 pkt_flags = (rx_status & IXGBE_RXD_STAT_VP) ? PKT_RX_VLAN_PKT : 0;
1015 #ifdef RTE_LIBRTE_IEEE1588
1016 if (rx_status & IXGBE_RXD_STAT_TMST)
1017 pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
1022 static inline uint64_t
1023 rx_desc_error_to_pkt_flags(uint32_t rx_status)
1026 * Bit 31: IPE, IPv4 checksum error
1027 * Bit 30: L4I, L4I integrity error
1029 static uint64_t error_to_pkt_flags_map[4] = {
1030 0, PKT_RX_L4_CKSUM_BAD, PKT_RX_IP_CKSUM_BAD,
1031 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1033 return error_to_pkt_flags_map[(rx_status >>
1034 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1038 * LOOK_AHEAD defines how many desc statuses to check beyond the
1039 * current descriptor.
1040 * It must be a pound define for optimal performance.
1041 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1042 * function only works with LOOK_AHEAD=8.
1044 #define LOOK_AHEAD 8
1045 #if (LOOK_AHEAD != 8)
1046 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1049 ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
1051 volatile union ixgbe_adv_rx_desc *rxdp;
1052 struct ixgbe_rx_entry *rxep;
1053 struct rte_mbuf *mb;
1058 uint32_t s[LOOK_AHEAD];
1059 uint16_t pkt_info[LOOK_AHEAD];
1061 int s[LOOK_AHEAD], nb_dd;
1062 #endif /* RTE_NEXT_ABI */
1063 int i, j, nb_rx = 0;
1066 /* get references to current descriptor and S/W ring entry */
1067 rxdp = &rxq->rx_ring[rxq->rx_tail];
1068 rxep = &rxq->sw_ring[rxq->rx_tail];
1070 /* check to make sure there is at least 1 packet to receive */
1071 if (! (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD))
1075 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1076 * reference packets that are ready to be received.
1078 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1079 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD)
1081 /* Read desc statuses backwards to avoid race condition */
1082 for (j = LOOK_AHEAD-1; j >= 0; --j)
1083 s[j] = rxdp[j].wb.upper.status_error;
1086 for (j = LOOK_AHEAD - 1; j >= 0; --j)
1087 pkt_info[j] = rxdp[j].wb.lower.lo_dword.
1089 #endif /* RTE_NEXT_ABI */
1091 /* Compute how many status bits were set */
1093 for (j = 0; j < LOOK_AHEAD; ++j)
1094 nb_dd += s[j] & IXGBE_RXDADV_STAT_DD;
1098 /* Translate descriptor info to mbuf format */
1099 for (j = 0; j < nb_dd; ++j) {
1101 pkt_len = (uint16_t)(rxdp[j].wb.upper.length - rxq->crc_len);
1102 mb->data_len = pkt_len;
1103 mb->pkt_len = pkt_len;
1104 mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
1106 /* convert descriptor fields to rte mbuf flags */
1108 pkt_flags = rx_desc_status_to_pkt_flags(s[j]);
1109 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1111 ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info[j]);
1112 mb->ol_flags = pkt_flags;
1114 ixgbe_rxd_pkt_info_to_pkt_type(pkt_info[j]);
1115 #else /* RTE_NEXT_ABI */
1116 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(
1117 rxdp[j].wb.lower.lo_dword.data);
1118 /* reuse status field from scan list */
1119 pkt_flags |= rx_desc_status_to_pkt_flags(s[j]);
1120 pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
1121 mb->ol_flags = pkt_flags;
1122 #endif /* RTE_NEXT_ABI */
1124 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1125 mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
1126 else if (pkt_flags & PKT_RX_FDIR) {
1127 mb->hash.fdir.hash =
1128 (uint16_t)((rxdp[j].wb.lower.hi_dword.csum_ip.csum)
1129 & IXGBE_ATR_HASH_MASK);
1130 mb->hash.fdir.id = rxdp[j].wb.lower.hi_dword.csum_ip.ip_id;
1134 /* Move mbuf pointers from the S/W ring to the stage */
1135 for (j = 0; j < LOOK_AHEAD; ++j) {
1136 rxq->rx_stage[i + j] = rxep[j].mbuf;
1139 /* stop if all requested packets could not be received */
1140 if (nb_dd != LOOK_AHEAD)
1144 /* clear software ring entries so we can cleanup correctly */
1145 for (i = 0; i < nb_rx; ++i) {
1146 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1154 ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
1156 volatile union ixgbe_adv_rx_desc *rxdp;
1157 struct ixgbe_rx_entry *rxep;
1158 struct rte_mbuf *mb;
1163 /* allocate buffers in bulk directly into the S/W ring */
1164 alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
1165 rxep = &rxq->sw_ring[alloc_idx];
1166 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1167 rxq->rx_free_thresh);
1168 if (unlikely(diag != 0))
1171 rxdp = &rxq->rx_ring[alloc_idx];
1172 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1173 /* populate the static rte mbuf fields */
1178 mb->port = rxq->port_id;
1181 rte_mbuf_refcnt_set(mb, 1);
1182 mb->data_off = RTE_PKTMBUF_HEADROOM;
1184 /* populate the descriptors */
1185 dma_addr = rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
1186 rxdp[i].read.hdr_addr = dma_addr;
1187 rxdp[i].read.pkt_addr = dma_addr;
1190 /* update state of internal queue structure */
1191 rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
1192 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1193 rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
1199 static inline uint16_t
1200 ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1203 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1206 /* how many packets are ready to return? */
1207 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1209 /* copy mbuf pointers to the application's packet list */
1210 for (i = 0; i < nb_pkts; ++i)
1211 rx_pkts[i] = stage[i];
1213 /* update internal queue state */
1214 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1215 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1220 static inline uint16_t
1221 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1224 struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
1227 /* Any previously recv'd pkts will be returned from the Rx stage */
1228 if (rxq->rx_nb_avail)
1229 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1231 /* Scan the H/W ring for packets to receive */
1232 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1234 /* update internal queue state */
1235 rxq->rx_next_avail = 0;
1236 rxq->rx_nb_avail = nb_rx;
1237 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1239 /* if required, allocate new buffers to replenish descriptors */
1240 if (rxq->rx_tail > rxq->rx_free_trigger) {
1241 uint16_t cur_free_trigger = rxq->rx_free_trigger;
1243 if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
1245 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1246 "queue_id=%u", (unsigned) rxq->port_id,
1247 (unsigned) rxq->queue_id);
1249 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1250 rxq->rx_free_thresh;
1253 * Need to rewind any previous receives if we cannot
1254 * allocate new buffers to replenish the old ones.
1256 rxq->rx_nb_avail = 0;
1257 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1258 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1259 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1264 /* update tail pointer */
1266 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, cur_free_trigger);
1269 if (rxq->rx_tail >= rxq->nb_rx_desc)
1272 /* received any packets this loop? */
1273 if (rxq->rx_nb_avail)
1274 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1279 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1281 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1286 if (unlikely(nb_pkts == 0))
1289 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1290 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1292 /* request is relatively large, chunk it up */
1296 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1297 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1298 nb_rx = (uint16_t)(nb_rx + ret);
1299 nb_pkts = (uint16_t)(nb_pkts - ret);
1308 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1311 struct ixgbe_rx_queue *rxq;
1312 volatile union ixgbe_adv_rx_desc *rx_ring;
1313 volatile union ixgbe_adv_rx_desc *rxdp;
1314 struct ixgbe_rx_entry *sw_ring;
1315 struct ixgbe_rx_entry *rxe;
1316 struct rte_mbuf *rxm;
1317 struct rte_mbuf *nmb;
1318 union ixgbe_adv_rx_desc rxd;
1324 uint32_t hlen_type_rss;
1335 rx_id = rxq->rx_tail;
1336 rx_ring = rxq->rx_ring;
1337 sw_ring = rxq->sw_ring;
1338 while (nb_rx < nb_pkts) {
1340 * The order of operations here is important as the DD status
1341 * bit must not be read after any other descriptor fields.
1342 * rx_ring and rxdp are pointing to volatile data so the order
1343 * of accesses cannot be reordered by the compiler. If they were
1344 * not volatile, they could be reordered which could lead to
1345 * using invalid descriptor fields when read from rxd.
1347 rxdp = &rx_ring[rx_id];
1348 staterr = rxdp->wb.upper.status_error;
1349 if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1356 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1357 * is likely to be invalid and to be dropped by the various
1358 * validation checks performed by the network stack.
1360 * Allocate a new mbuf to replenish the RX ring descriptor.
1361 * If the allocation fails:
1362 * - arrange for that RX descriptor to be the first one
1363 * being parsed the next time the receive function is
1364 * invoked [on the same queue].
1366 * - Stop parsing the RX ring and return immediately.
1368 * This policy do not drop the packet received in the RX
1369 * descriptor for which the allocation of a new mbuf failed.
1370 * Thus, it allows that packet to be later retrieved if
1371 * mbuf have been freed in the mean time.
1372 * As a side effect, holding RX descriptors instead of
1373 * systematically giving them back to the NIC may lead to
1374 * RX ring exhaustion situations.
1375 * However, the NIC can gracefully prevent such situations
1376 * to happen by sending specific "back-pressure" flow control
1377 * frames to its peer(s).
1379 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1380 "ext_err_stat=0x%08x pkt_len=%u",
1381 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1382 (unsigned) rx_id, (unsigned) staterr,
1383 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1385 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1387 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1388 "queue_id=%u", (unsigned) rxq->port_id,
1389 (unsigned) rxq->queue_id);
1390 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1395 rxe = &sw_ring[rx_id];
1397 if (rx_id == rxq->nb_rx_desc)
1400 /* Prefetch next mbuf while processing current one. */
1401 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1404 * When next RX descriptor is on a cache-line boundary,
1405 * prefetch the next 4 RX descriptors and the next 8 pointers
1408 if ((rx_id & 0x3) == 0) {
1409 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1410 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1416 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1417 rxdp->read.hdr_addr = dma_addr;
1418 rxdp->read.pkt_addr = dma_addr;
1421 * Initialize the returned mbuf.
1422 * 1) setup generic mbuf fields:
1423 * - number of segments,
1426 * - RX port identifier.
1427 * 2) integrate hardware offload data, if any:
1428 * - RSS flag & hash,
1429 * - IP checksum flag,
1430 * - VLAN TCI, if any,
1433 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1435 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1436 rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
1439 rxm->pkt_len = pkt_len;
1440 rxm->data_len = pkt_len;
1441 rxm->port = rxq->port_id;
1444 pkt_info = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.hs_rss.
1446 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1447 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1449 pkt_flags = rx_desc_status_to_pkt_flags(staterr);
1450 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1451 pkt_flags = pkt_flags |
1452 ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
1453 rxm->ol_flags = pkt_flags;
1454 rxm->packet_type = ixgbe_rxd_pkt_info_to_pkt_type(pkt_info);
1455 #else /* RTE_NEXT_ABI */
1456 hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1457 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1458 rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
1460 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1461 pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
1462 pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
1463 rxm->ol_flags = pkt_flags;
1464 #endif /* RTE_NEXT_ABI */
1466 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1467 rxm->hash.rss = rxd.wb.lower.hi_dword.rss;
1468 else if (pkt_flags & PKT_RX_FDIR) {
1469 rxm->hash.fdir.hash =
1470 (uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
1471 & IXGBE_ATR_HASH_MASK);
1472 rxm->hash.fdir.id = rxd.wb.lower.hi_dword.csum_ip.ip_id;
1475 * Store the mbuf address into the next entry of the array
1476 * of returned packets.
1478 rx_pkts[nb_rx++] = rxm;
1480 rxq->rx_tail = rx_id;
1483 * If the number of free RX descriptors is greater than the RX free
1484 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1486 * Update the RDT with the value of the last processed RX descriptor
1487 * minus 1, to guarantee that the RDT register is never equal to the
1488 * RDH register, which creates a "full" ring situtation from the
1489 * hardware point of view...
1491 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1492 if (nb_hold > rxq->rx_free_thresh) {
1493 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1494 "nb_hold=%u nb_rx=%u",
1495 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1496 (unsigned) rx_id, (unsigned) nb_hold,
1498 rx_id = (uint16_t) ((rx_id == 0) ?
1499 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1500 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1503 rxq->nb_rx_hold = nb_hold;
1508 * Detect an RSC descriptor.
1510 static inline uint32_t
1511 ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
1513 return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
1514 IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
1518 * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
1520 * Fill the following info in the HEAD buffer of the Rx cluster:
1521 * - RX port identifier
1522 * - hardware offload data, if any:
1524 * - IP checksum flag
1525 * - VLAN TCI, if any
1527 * @head HEAD of the packet cluster
1528 * @desc HW descriptor to get data from
1529 * @port_id Port ID of the Rx queue
1532 ixgbe_fill_cluster_head_buf(
1533 struct rte_mbuf *head,
1534 union ixgbe_adv_rx_desc *desc,
1542 head->port = port_id;
1544 /* The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1545 * set in the pkt_flags field.
1547 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1548 pkt_info = rte_le_to_cpu_32(desc->wb.lower.lo_dword.hs_rss.pkt_info);
1549 pkt_flags = rx_desc_status_to_pkt_flags(staterr);
1550 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1551 pkt_flags |= ixgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
1552 head->ol_flags = pkt_flags;
1553 head->packet_type = ixgbe_rxd_pkt_info_to_pkt_type(pkt_info);
1554 #else /* RTE_NEXT_ABI */
1555 uint32_t hlen_type_rss;
1558 head->port = port_id;
1561 * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1562 * set in the pkt_flags field.
1564 head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
1565 hlen_type_rss = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
1566 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1567 pkt_flags |= rx_desc_status_to_pkt_flags(staterr);
1568 pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
1569 head->ol_flags = pkt_flags;
1570 #endif /* RTE_NEXT_ABI */
1572 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1573 head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
1574 else if (pkt_flags & PKT_RX_FDIR) {
1575 head->hash.fdir.hash =
1576 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
1577 & IXGBE_ATR_HASH_MASK;
1578 head->hash.fdir.id =
1579 rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
1584 * ixgbe_recv_pkts_lro - receive handler for and LRO case.
1586 * @rx_queue Rx queue handle
1587 * @rx_pkts table of received packets
1588 * @nb_pkts size of rx_pkts table
1589 * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
1591 * Handles the Rx HW ring completions when RSC feature is configured. Uses an
1592 * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
1594 * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
1595 * 1) When non-EOP RSC completion arrives:
1596 * a) Update the HEAD of the current RSC aggregation cluster with the new
1597 * segment's data length.
1598 * b) Set the "next" pointer of the current segment to point to the segment
1599 * at the NEXTP index.
1600 * c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
1601 * in the sw_rsc_ring.
1602 * 2) When EOP arrives we just update the cluster's total length and offload
1603 * flags and deliver the cluster up to the upper layers. In our case - put it
1604 * in the rx_pkts table.
1606 * Returns the number of received packets/clusters (according to the "bulk
1607 * receive" interface).
1609 static inline uint16_t
1610 ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
1613 struct ixgbe_rx_queue *rxq = rx_queue;
1614 volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
1615 struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
1616 struct ixgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_ring;
1617 uint16_t rx_id = rxq->rx_tail;
1619 uint16_t nb_hold = rxq->nb_rx_hold;
1620 uint16_t prev_id = rxq->rx_tail;
1622 while (nb_rx < nb_pkts) {
1624 struct ixgbe_rx_entry *rxe;
1625 struct ixgbe_scattered_rx_entry *sc_entry;
1626 struct ixgbe_scattered_rx_entry *next_sc_entry;
1627 struct ixgbe_rx_entry *next_rxe;
1628 struct rte_mbuf *first_seg;
1629 struct rte_mbuf *rxm;
1630 struct rte_mbuf *nmb;
1631 union ixgbe_adv_rx_desc rxd;
1634 volatile union ixgbe_adv_rx_desc *rxdp;
1639 * The code in this whole file uses the volatile pointer to
1640 * ensure the read ordering of the status and the rest of the
1641 * descriptor fields (on the compiler level only!!!). This is so
1642 * UGLY - why not to just use the compiler barrier instead? DPDK
1643 * even has the rte_compiler_barrier() for that.
1645 * But most importantly this is just wrong because this doesn't
1646 * ensure memory ordering in a general case at all. For
1647 * instance, DPDK is supposed to work on Power CPUs where
1648 * compiler barrier may just not be enough!
1650 * I tried to write only this function properly to have a
1651 * starting point (as a part of an LRO/RSC series) but the
1652 * compiler cursed at me when I tried to cast away the
1653 * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
1654 * keeping it the way it is for now.
1656 * The code in this file is broken in so many other places and
1657 * will just not work on a big endian CPU anyway therefore the
1658 * lines below will have to be revisited together with the rest
1662 * - Get rid of "volatile" crap and let the compiler do its
1664 * - Use the proper memory barrier (rte_rmb()) to ensure the
1665 * memory ordering below.
1667 rxdp = &rx_ring[rx_id];
1668 staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
1670 if (!(staterr & IXGBE_RXDADV_STAT_DD))
1675 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1676 "staterr=0x%x data_len=%u",
1677 rxq->port_id, rxq->queue_id, rx_id, staterr,
1678 rte_le_to_cpu_16(rxd.wb.upper.length));
1681 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1683 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
1684 "port_id=%u queue_id=%u",
1685 rxq->port_id, rxq->queue_id);
1687 rte_eth_devices[rxq->port_id].data->
1688 rx_mbuf_alloc_failed++;
1692 else if (nb_hold > rxq->rx_free_thresh) {
1693 uint16_t next_rdt = rxq->rx_free_trigger;
1695 if (!ixgbe_rx_alloc_bufs(rxq, false)) {
1697 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr,
1699 nb_hold -= rxq->rx_free_thresh;
1701 PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
1702 "port_id=%u queue_id=%u",
1703 rxq->port_id, rxq->queue_id);
1705 rte_eth_devices[rxq->port_id].data->
1706 rx_mbuf_alloc_failed++;
1712 rxe = &sw_ring[rx_id];
1713 eop = staterr & IXGBE_RXDADV_STAT_EOP;
1715 next_id = rx_id + 1;
1716 if (next_id == rxq->nb_rx_desc)
1719 /* Prefetch next mbuf while processing current one. */
1720 rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
1723 * When next RX descriptor is on a cache-line boundary,
1724 * prefetch the next 4 RX descriptors and the next 4 pointers
1727 if ((next_id & 0x3) == 0) {
1728 rte_ixgbe_prefetch(&rx_ring[next_id]);
1729 rte_ixgbe_prefetch(&sw_ring[next_id]);
1736 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1738 * Update RX descriptor with the physical address of the
1739 * new data buffer of the new allocated mbuf.
1743 rxm->data_off = RTE_PKTMBUF_HEADROOM;
1744 rxdp->read.hdr_addr = dma;
1745 rxdp->read.pkt_addr = dma;
1750 * Set data length & data buffer address of mbuf.
1752 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
1753 rxm->data_len = data_len;
1758 * Get next descriptor index:
1759 * - For RSC it's in the NEXTP field.
1760 * - For a scattered packet - it's just a following
1763 if (ixgbe_rsc_count(&rxd))
1765 (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1766 IXGBE_RXDADV_NEXTP_SHIFT;
1770 next_sc_entry = &sw_sc_ring[nextp_id];
1771 next_rxe = &sw_ring[nextp_id];
1772 rte_ixgbe_prefetch(next_rxe);
1775 sc_entry = &sw_sc_ring[rx_id];
1776 first_seg = sc_entry->fbuf;
1777 sc_entry->fbuf = NULL;
1780 * If this is the first buffer of the received packet,
1781 * set the pointer to the first mbuf of the packet and
1782 * initialize its context.
1783 * Otherwise, update the total length and the number of segments
1784 * of the current scattered packet, and update the pointer to
1785 * the last mbuf of the current packet.
1787 if (first_seg == NULL) {
1789 first_seg->pkt_len = data_len;
1790 first_seg->nb_segs = 1;
1792 first_seg->pkt_len += data_len;
1793 first_seg->nb_segs++;
1800 * If this is not the last buffer of the received packet, update
1801 * the pointer to the first mbuf at the NEXTP entry in the
1802 * sw_sc_ring and continue to parse the RX ring.
1805 rxm->next = next_rxe->mbuf;
1806 next_sc_entry->fbuf = first_seg;
1811 * This is the last buffer of the received packet - return
1812 * the current cluster to the user.
1816 /* Initialize the first mbuf of the returned packet */
1817 ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq->port_id,
1820 /* Prefetch data of first segment, if configured to do so. */
1821 rte_packet_prefetch((char *)first_seg->buf_addr +
1822 first_seg->data_off);
1825 * Store the mbuf address into the next entry of the array
1826 * of returned packets.
1828 rx_pkts[nb_rx++] = first_seg;
1832 * Record index of the next RX descriptor to probe.
1834 rxq->rx_tail = rx_id;
1837 * If the number of free RX descriptors is greater than the RX free
1838 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1840 * Update the RDT with the value of the last processed RX descriptor
1841 * minus 1, to guarantee that the RDT register is never equal to the
1842 * RDH register, which creates a "full" ring situtation from the
1843 * hardware point of view...
1845 if (!bulk_alloc && nb_hold > rxq->rx_free_thresh) {
1846 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1847 "nb_hold=%u nb_rx=%u",
1848 rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
1851 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, prev_id);
1855 rxq->nb_rx_hold = nb_hold;
1860 ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1863 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
1867 ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1870 return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
1873 /*********************************************************************
1875 * Queue management functions
1877 **********************************************************************/
1880 * Rings setup and release.
1882 * TDBA/RDBA should be aligned on 16 byte boundary. But TDLEN/RDLEN should be
1883 * multiple of 128 bytes. So we align TDBA/RDBA on 128 byte boundary. This will
1884 * also optimize cache line size effect. H/W supports up to cache line size 128.
1886 #define IXGBE_ALIGN 128
1889 * Maximum number of Ring Descriptors.
1891 * Since RDLEN/TDLEN should be multiple of 128 bytes, the number of ring
1892 * descriptors should meet the following condition:
1893 * (num_ring_desc * sizeof(rx/tx descriptor)) % 128 == 0
1895 #define IXGBE_MIN_RING_DESC 32
1896 #define IXGBE_MAX_RING_DESC 4096
1899 * Create memzone for HW rings. malloc can't be used as the physical address is
1900 * needed. If the memzone is already created, then this function returns a ptr
1903 static const struct rte_memzone * __attribute__((cold))
1904 ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
1905 uint16_t queue_id, uint32_t ring_size, int socket_id)
1907 char z_name[RTE_MEMZONE_NAMESIZE];
1908 const struct rte_memzone *mz;
1910 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
1911 dev->driver->pci_drv.name, ring_name,
1912 dev->data->port_id, queue_id);
1914 mz = rte_memzone_lookup(z_name);
1918 #ifdef RTE_LIBRTE_XEN_DOM0
1919 return rte_memzone_reserve_bounded(z_name, ring_size,
1920 socket_id, 0, IXGBE_ALIGN, RTE_PGSIZE_2M);
1922 return rte_memzone_reserve_aligned(z_name, ring_size,
1923 socket_id, 0, IXGBE_ALIGN);
1927 static void __attribute__((cold))
1928 ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
1932 if (txq->sw_ring != NULL) {
1933 for (i = 0; i < txq->nb_tx_desc; i++) {
1934 if (txq->sw_ring[i].mbuf != NULL) {
1935 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
1936 txq->sw_ring[i].mbuf = NULL;
1942 static void __attribute__((cold))
1943 ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
1946 txq->sw_ring != NULL)
1947 rte_free(txq->sw_ring);
1950 static void __attribute__((cold))
1951 ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
1953 if (txq != NULL && txq->ops != NULL) {
1954 txq->ops->release_mbufs(txq);
1955 txq->ops->free_swring(txq);
1960 void __attribute__((cold))
1961 ixgbe_dev_tx_queue_release(void *txq)
1963 ixgbe_tx_queue_release(txq);
1966 /* (Re)set dynamic ixgbe_tx_queue fields to defaults */
1967 static void __attribute__((cold))
1968 ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
1970 static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
1971 struct ixgbe_tx_entry *txe = txq->sw_ring;
1974 /* Zero out HW ring memory */
1975 for (i = 0; i < txq->nb_tx_desc; i++) {
1976 txq->tx_ring[i] = zeroed_desc;
1979 /* Initialize SW ring entries */
1980 prev = (uint16_t) (txq->nb_tx_desc - 1);
1981 for (i = 0; i < txq->nb_tx_desc; i++) {
1982 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
1983 txd->wb.status = IXGBE_TXD_STAT_DD;
1986 txe[prev].next_id = i;
1990 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1991 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1994 txq->nb_tx_used = 0;
1996 * Always allow 1 descriptor to be un-allocated to avoid
1997 * a H/W race condition
1999 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
2000 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
2002 memset((void*)&txq->ctx_cache, 0,
2003 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
2006 static const struct ixgbe_txq_ops def_txq_ops = {
2007 .release_mbufs = ixgbe_tx_queue_release_mbufs,
2008 .free_swring = ixgbe_tx_free_swring,
2009 .reset = ixgbe_reset_tx_queue,
2012 /* Takes an ethdev and a queue and sets up the tx function to be used based on
2013 * the queue parameters. Used in tx_queue_setup by primary process and then
2014 * in dev_init by secondary process when attaching to an existing ethdev.
2016 void __attribute__((cold))
2017 ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_tx_queue *txq)
2019 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
2020 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
2021 && (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
2022 PMD_INIT_LOG(INFO, "Using simple tx code path");
2023 #ifdef RTE_IXGBE_INC_VECTOR
2024 if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
2025 (rte_eal_process_type() != RTE_PROC_PRIMARY ||
2026 ixgbe_txq_vec_setup(txq) == 0)) {
2027 PMD_INIT_LOG(INFO, "Vector tx enabled.");
2028 dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
2031 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
2033 PMD_INIT_LOG(INFO, "Using full-featured tx code path");
2035 " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
2036 (unsigned long)txq->txq_flags,
2037 (unsigned long)IXGBE_SIMPLE_FLAGS);
2039 " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
2040 (unsigned long)txq->tx_rs_thresh,
2041 (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
2042 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2046 int __attribute__((cold))
2047 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
2050 unsigned int socket_id,
2051 const struct rte_eth_txconf *tx_conf)
2053 const struct rte_memzone *tz;
2054 struct ixgbe_tx_queue *txq;
2055 struct ixgbe_hw *hw;
2056 uint16_t tx_rs_thresh, tx_free_thresh;
2058 PMD_INIT_FUNC_TRACE();
2059 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2062 * Validate number of transmit descriptors.
2063 * It must not exceed hardware maximum, and must be multiple
2066 if (((nb_desc * sizeof(union ixgbe_adv_tx_desc)) % IXGBE_ALIGN) != 0 ||
2067 (nb_desc > IXGBE_MAX_RING_DESC) ||
2068 (nb_desc < IXGBE_MIN_RING_DESC)) {
2073 * The following two parameters control the setting of the RS bit on
2074 * transmit descriptors.
2075 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
2076 * descriptors have been used.
2077 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
2078 * descriptors are used or if the number of descriptors required
2079 * to transmit a packet is greater than the number of free TX
2081 * The following constraints must be satisfied:
2082 * tx_rs_thresh must be greater than 0.
2083 * tx_rs_thresh must be less than the size of the ring minus 2.
2084 * tx_rs_thresh must be less than or equal to tx_free_thresh.
2085 * tx_rs_thresh must be a divisor of the ring size.
2086 * tx_free_thresh must be greater than 0.
2087 * tx_free_thresh must be less than the size of the ring minus 3.
2088 * One descriptor in the TX ring is used as a sentinel to avoid a
2089 * H/W race condition, hence the maximum threshold constraints.
2090 * When set to zero use default values.
2092 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
2093 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
2094 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
2095 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
2096 if (tx_rs_thresh >= (nb_desc - 2)) {
2097 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
2098 "of TX descriptors minus 2. (tx_rs_thresh=%u "
2099 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2100 (int)dev->data->port_id, (int)queue_idx);
2103 if (tx_free_thresh >= (nb_desc - 3)) {
2104 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
2105 "tx_free_thresh must be less than the number of "
2106 "TX descriptors minus 3. (tx_free_thresh=%u "
2107 "port=%d queue=%d)",
2108 (unsigned int)tx_free_thresh,
2109 (int)dev->data->port_id, (int)queue_idx);
2112 if (tx_rs_thresh > tx_free_thresh) {
2113 PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
2114 "tx_free_thresh. (tx_free_thresh=%u "
2115 "tx_rs_thresh=%u port=%d queue=%d)",
2116 (unsigned int)tx_free_thresh,
2117 (unsigned int)tx_rs_thresh,
2118 (int)dev->data->port_id,
2122 if ((nb_desc % tx_rs_thresh) != 0) {
2123 PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
2124 "number of TX descriptors. (tx_rs_thresh=%u "
2125 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2126 (int)dev->data->port_id, (int)queue_idx);
2131 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
2132 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
2133 * by the NIC and all descriptors are written back after the NIC
2134 * accumulates WTHRESH descriptors.
2136 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
2137 PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
2138 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
2139 "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
2140 (int)dev->data->port_id, (int)queue_idx);
2144 /* Free memory prior to re-allocation if needed... */
2145 if (dev->data->tx_queues[queue_idx] != NULL) {
2146 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
2147 dev->data->tx_queues[queue_idx] = NULL;
2150 /* First allocate the tx queue data structure */
2151 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
2152 RTE_CACHE_LINE_SIZE, socket_id);
2157 * Allocate TX ring hardware descriptors. A memzone large enough to
2158 * handle the maximum ring size is allocated in order to allow for
2159 * resizing in later calls to the queue setup function.
2161 tz = ring_dma_zone_reserve(dev, "tx_ring", queue_idx,
2162 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
2165 ixgbe_tx_queue_release(txq);
2169 txq->nb_tx_desc = nb_desc;
2170 txq->tx_rs_thresh = tx_rs_thresh;
2171 txq->tx_free_thresh = tx_free_thresh;
2172 txq->pthresh = tx_conf->tx_thresh.pthresh;
2173 txq->hthresh = tx_conf->tx_thresh.hthresh;
2174 txq->wthresh = tx_conf->tx_thresh.wthresh;
2175 txq->queue_id = queue_idx;
2176 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2177 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2178 txq->port_id = dev->data->port_id;
2179 txq->txq_flags = tx_conf->txq_flags;
2180 txq->ops = &def_txq_ops;
2181 txq->tx_deferred_start = tx_conf->tx_deferred_start;
2184 * Modification to set VFTDT for virtual function if vf is detected
2186 if (hw->mac.type == ixgbe_mac_82599_vf ||
2187 hw->mac.type == ixgbe_mac_X540_vf ||
2188 hw->mac.type == ixgbe_mac_X550_vf ||
2189 hw->mac.type == ixgbe_mac_X550EM_x_vf)
2190 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
2192 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
2193 #ifndef RTE_LIBRTE_XEN_DOM0
2194 txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
2196 txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
2198 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
2200 /* Allocate software ring */
2201 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
2202 sizeof(struct ixgbe_tx_entry) * nb_desc,
2203 RTE_CACHE_LINE_SIZE, socket_id);
2204 if (txq->sw_ring == NULL) {
2205 ixgbe_tx_queue_release(txq);
2208 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
2209 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
2211 /* set up vector or scalar TX function as appropriate */
2212 ixgbe_set_tx_function(dev, txq);
2214 txq->ops->reset(txq);
2216 dev->data->tx_queues[queue_idx] = txq;
2223 * ixgbe_free_sc_cluster - free the not-yet-completed scattered cluster
2225 * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
2226 * in the sw_rsc_ring is not set to NULL but rather points to the next
2227 * mbuf of this RSC aggregation (that has not been completed yet and still
2228 * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
2229 * will just free first "nb_segs" segments of the cluster explicitly by calling
2230 * an rte_pktmbuf_free_seg().
2232 * @m scattered cluster head
2234 static void __attribute__((cold))
2235 ixgbe_free_sc_cluster(struct rte_mbuf *m)
2237 uint8_t i, nb_segs = m->nb_segs;
2238 struct rte_mbuf *next_seg;
2240 for (i = 0; i < nb_segs; i++) {
2242 rte_pktmbuf_free_seg(m);
2247 static void __attribute__((cold))
2248 ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
2252 #ifdef RTE_IXGBE_INC_VECTOR
2253 /* SSE Vector driver has a different way of releasing mbufs. */
2254 if (rxq->rx_using_sse) {
2255 ixgbe_rx_queue_release_mbufs_vec(rxq);
2260 if (rxq->sw_ring != NULL) {
2261 for (i = 0; i < rxq->nb_rx_desc; i++) {
2262 if (rxq->sw_ring[i].mbuf != NULL) {
2263 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2264 rxq->sw_ring[i].mbuf = NULL;
2267 if (rxq->rx_nb_avail) {
2268 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2269 struct rte_mbuf *mb;
2270 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2271 rte_pktmbuf_free_seg(mb);
2273 rxq->rx_nb_avail = 0;
2277 if (rxq->sw_sc_ring)
2278 for (i = 0; i < rxq->nb_rx_desc; i++)
2279 if (rxq->sw_sc_ring[i].fbuf) {
2280 ixgbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
2281 rxq->sw_sc_ring[i].fbuf = NULL;
2285 static void __attribute__((cold))
2286 ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
2289 ixgbe_rx_queue_release_mbufs(rxq);
2290 rte_free(rxq->sw_ring);
2291 rte_free(rxq->sw_sc_ring);
2296 void __attribute__((cold))
2297 ixgbe_dev_rx_queue_release(void *rxq)
2299 ixgbe_rx_queue_release(rxq);
2303 * Check if Rx Burst Bulk Alloc function can be used.
2305 * 0: the preconditions are satisfied and the bulk allocation function
2307 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2308 * function must be used.
2310 static inline int __attribute__((cold))
2311 check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
2316 * Make sure the following pre-conditions are satisfied:
2317 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2318 * rxq->rx_free_thresh < rxq->nb_rx_desc
2319 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2320 * rxq->nb_rx_desc<(IXGBE_MAX_RING_DESC-RTE_PMD_IXGBE_RX_MAX_BURST)
2321 * Scattered packets are not supported. This should be checked
2322 * outside of this function.
2324 if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
2325 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2326 "rxq->rx_free_thresh=%d, "
2327 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2328 rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
2330 } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
2331 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2332 "rxq->rx_free_thresh=%d, "
2333 "rxq->nb_rx_desc=%d",
2334 rxq->rx_free_thresh, rxq->nb_rx_desc);
2336 } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
2337 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2338 "rxq->nb_rx_desc=%d, "
2339 "rxq->rx_free_thresh=%d",
2340 rxq->nb_rx_desc, rxq->rx_free_thresh);
2342 } else if (!(rxq->nb_rx_desc <
2343 (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST))) {
2344 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
2345 "rxq->nb_rx_desc=%d, "
2346 "IXGBE_MAX_RING_DESC=%d, "
2347 "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
2348 rxq->nb_rx_desc, IXGBE_MAX_RING_DESC,
2349 RTE_PMD_IXGBE_RX_MAX_BURST);
2356 /* Reset dynamic ixgbe_rx_queue fields back to defaults */
2357 static void __attribute__((cold))
2358 ixgbe_reset_rx_queue(struct ixgbe_adapter *adapter, struct ixgbe_rx_queue *rxq)
2360 static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
2362 uint16_t len = rxq->nb_rx_desc;
2365 * By default, the Rx queue setup function allocates enough memory for
2366 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2367 * extra memory at the end of the descriptor ring to be zero'd out. A
2368 * pre-condition for using the Rx burst bulk alloc function is that the
2369 * number of descriptors is less than or equal to
2370 * (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST). Check all the
2371 * constraints here to see if we need to zero out memory after the end
2372 * of the H/W descriptor ring.
2374 if (adapter->rx_bulk_alloc_allowed)
2375 /* zero out extra memory */
2376 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2379 * Zero out HW ring memory. Zero out extra memory at the end of
2380 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2381 * reads extra memory as zeros.
2383 for (i = 0; i < len; i++) {
2384 rxq->rx_ring[i] = zeroed_desc;
2388 * initialize extra software ring entries. Space for these extra
2389 * entries is always allocated
2391 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2392 for (i = rxq->nb_rx_desc; i < len; ++i) {
2393 rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
2396 rxq->rx_nb_avail = 0;
2397 rxq->rx_next_avail = 0;
2398 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2400 rxq->nb_rx_hold = 0;
2401 rxq->pkt_first_seg = NULL;
2402 rxq->pkt_last_seg = NULL;
2405 int __attribute__((cold))
2406 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2409 unsigned int socket_id,
2410 const struct rte_eth_rxconf *rx_conf,
2411 struct rte_mempool *mp)
2413 const struct rte_memzone *rz;
2414 struct ixgbe_rx_queue *rxq;
2415 struct ixgbe_hw *hw;
2417 struct ixgbe_adapter *adapter =
2418 (struct ixgbe_adapter *)dev->data->dev_private;
2420 PMD_INIT_FUNC_TRACE();
2421 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2424 * Validate number of receive descriptors.
2425 * It must not exceed hardware maximum, and must be multiple
2428 if (((nb_desc * sizeof(union ixgbe_adv_rx_desc)) % IXGBE_ALIGN) != 0 ||
2429 (nb_desc > IXGBE_MAX_RING_DESC) ||
2430 (nb_desc < IXGBE_MIN_RING_DESC)) {
2434 /* Free memory prior to re-allocation if needed... */
2435 if (dev->data->rx_queues[queue_idx] != NULL) {
2436 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2437 dev->data->rx_queues[queue_idx] = NULL;
2440 /* First allocate the rx queue data structure */
2441 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
2442 RTE_CACHE_LINE_SIZE, socket_id);
2446 rxq->nb_rx_desc = nb_desc;
2447 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2448 rxq->queue_id = queue_idx;
2449 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2450 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2451 rxq->port_id = dev->data->port_id;
2452 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2454 rxq->drop_en = rx_conf->rx_drop_en;
2455 rxq->rx_deferred_start = rx_conf->rx_deferred_start;
2458 * Allocate RX ring hardware descriptors. A memzone large enough to
2459 * handle the maximum ring size is allocated in order to allow for
2460 * resizing in later calls to the queue setup function.
2462 rz = ring_dma_zone_reserve(dev, "rx_ring", queue_idx,
2463 RX_RING_SZ, socket_id);
2465 ixgbe_rx_queue_release(rxq);
2470 * Zero init all the descriptors in the ring.
2472 memset (rz->addr, 0, RX_RING_SZ);
2475 * Modified to setup VFRDT for Virtual Function
2477 if (hw->mac.type == ixgbe_mac_82599_vf ||
2478 hw->mac.type == ixgbe_mac_X540_vf ||
2479 hw->mac.type == ixgbe_mac_X550_vf ||
2480 hw->mac.type == ixgbe_mac_X550EM_x_vf) {
2482 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2484 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2488 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2490 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2492 #ifndef RTE_LIBRTE_XEN_DOM0
2493 rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
2495 rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
2497 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2500 * Certain constraints must be met in order to use the bulk buffer
2501 * allocation Rx burst function. If any of Rx queues doesn't meet them
2502 * the feature should be disabled for the whole port.
2504 if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
2505 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
2506 "preconditions - canceling the feature for "
2507 "the whole port[%d]",
2508 rxq->queue_id, rxq->port_id);
2509 adapter->rx_bulk_alloc_allowed = false;
2513 * Allocate software ring. Allow for space at the end of the
2514 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2515 * function does not access an invalid memory region.
2518 if (adapter->rx_bulk_alloc_allowed)
2519 len += RTE_PMD_IXGBE_RX_MAX_BURST;
2521 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2522 sizeof(struct ixgbe_rx_entry) * len,
2523 RTE_CACHE_LINE_SIZE, socket_id);
2524 if (!rxq->sw_ring) {
2525 ixgbe_rx_queue_release(rxq);
2530 * Always allocate even if it's not going to be needed in order to
2531 * simplify the code.
2533 * This ring is used in LRO and Scattered Rx cases and Scattered Rx may
2534 * be requested in ixgbe_dev_rx_init(), which is called later from
2538 rte_zmalloc_socket("rxq->sw_sc_ring",
2539 sizeof(struct ixgbe_scattered_rx_entry) * len,
2540 RTE_CACHE_LINE_SIZE, socket_id);
2541 if (!rxq->sw_sc_ring) {
2542 ixgbe_rx_queue_release(rxq);
2546 PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_sc_ring=%p hw_ring=%p "
2547 "dma_addr=0x%"PRIx64,
2548 rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
2549 rxq->rx_ring_phys_addr);
2551 if (!rte_is_power_of_2(nb_desc)) {
2552 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
2553 "preconditions - canceling the feature for "
2554 "the whole port[%d]",
2555 rxq->queue_id, rxq->port_id);
2556 adapter->rx_vec_allowed = false;
2558 ixgbe_rxq_vec_setup(rxq);
2560 dev->data->rx_queues[queue_idx] = rxq;
2562 ixgbe_reset_rx_queue(adapter, rxq);
2568 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2570 #define IXGBE_RXQ_SCAN_INTERVAL 4
2571 volatile union ixgbe_adv_rx_desc *rxdp;
2572 struct ixgbe_rx_queue *rxq;
2575 if (rx_queue_id >= dev->data->nb_rx_queues) {
2576 PMD_RX_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
2580 rxq = dev->data->rx_queues[rx_queue_id];
2581 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2583 while ((desc < rxq->nb_rx_desc) &&
2584 (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD)) {
2585 desc += IXGBE_RXQ_SCAN_INTERVAL;
2586 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2587 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2588 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2589 desc - rxq->nb_rx_desc]);
2596 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2598 volatile union ixgbe_adv_rx_desc *rxdp;
2599 struct ixgbe_rx_queue *rxq = rx_queue;
2602 if (unlikely(offset >= rxq->nb_rx_desc))
2604 desc = rxq->rx_tail + offset;
2605 if (desc >= rxq->nb_rx_desc)
2606 desc -= rxq->nb_rx_desc;
2608 rxdp = &rxq->rx_ring[desc];
2609 return !!(rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD);
2612 void __attribute__((cold))
2613 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
2616 struct ixgbe_adapter *adapter =
2617 (struct ixgbe_adapter *)dev->data->dev_private;
2619 PMD_INIT_FUNC_TRACE();
2621 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2622 struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
2624 txq->ops->release_mbufs(txq);
2625 txq->ops->reset(txq);
2629 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2630 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
2632 ixgbe_rx_queue_release_mbufs(rxq);
2633 ixgbe_reset_rx_queue(adapter, rxq);
2639 ixgbe_dev_free_queues(struct rte_eth_dev *dev)
2643 PMD_INIT_FUNC_TRACE();
2645 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2646 ixgbe_dev_rx_queue_release(dev->data->rx_queues[i]);
2647 dev->data->rx_queues[i] = NULL;
2649 dev->data->nb_rx_queues = 0;
2651 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2652 ixgbe_dev_tx_queue_release(dev->data->tx_queues[i]);
2653 dev->data->tx_queues[i] = NULL;
2655 dev->data->nb_tx_queues = 0;
2658 /*********************************************************************
2660 * Device RX/TX init functions
2662 **********************************************************************/
2665 * Receive Side Scaling (RSS)
2666 * See section 7.1.2.8 in the following document:
2667 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
2670 * The source and destination IP addresses of the IP header and the source
2671 * and destination ports of TCP/UDP headers, if any, of received packets are
2672 * hashed against a configurable random key to compute a 32-bit RSS hash result.
2673 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
2674 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
2675 * RSS output index which is used as the RX queue index where to store the
2677 * The following output is supplied in the RX write-back descriptor:
2678 * - 32-bit result of the Microsoft RSS hash function,
2679 * - 4-bit RSS type field.
2683 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
2684 * Used as the default key.
2686 static uint8_t rss_intel_key[40] = {
2687 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2688 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2689 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2690 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2691 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2695 ixgbe_rss_disable(struct rte_eth_dev *dev)
2697 struct ixgbe_hw *hw;
2700 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2701 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2702 mrqc &= ~IXGBE_MRQC_RSSEN;
2703 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2707 ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
2715 hash_key = rss_conf->rss_key;
2716 if (hash_key != NULL) {
2717 /* Fill in RSS hash key */
2718 for (i = 0; i < 10; i++) {
2719 rss_key = hash_key[(i * 4)];
2720 rss_key |= hash_key[(i * 4) + 1] << 8;
2721 rss_key |= hash_key[(i * 4) + 2] << 16;
2722 rss_key |= hash_key[(i * 4) + 3] << 24;
2723 IXGBE_WRITE_REG_ARRAY(hw, IXGBE_RSSRK(0), i, rss_key);
2727 /* Set configured hashing protocols in MRQC register */
2728 rss_hf = rss_conf->rss_hf;
2729 mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
2730 if (rss_hf & ETH_RSS_IPV4)
2731 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
2732 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
2733 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
2734 if (rss_hf & ETH_RSS_IPV6)
2735 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
2736 if (rss_hf & ETH_RSS_IPV6_EX)
2737 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
2738 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
2739 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2740 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
2741 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
2742 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
2743 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2744 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
2745 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2746 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
2747 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2748 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2752 ixgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
2753 struct rte_eth_rss_conf *rss_conf)
2755 struct ixgbe_hw *hw;
2759 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2762 * Excerpt from section 7.1.2.8 Receive-Side Scaling (RSS):
2763 * "RSS enabling cannot be done dynamically while it must be
2764 * preceded by a software reset"
2765 * Before changing anything, first check that the update RSS operation
2766 * does not attempt to disable RSS, if RSS was enabled at
2767 * initialization time, or does not attempt to enable RSS, if RSS was
2768 * disabled at initialization time.
2770 rss_hf = rss_conf->rss_hf & IXGBE_RSS_OFFLOAD_ALL;
2771 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2772 if (!(mrqc & IXGBE_MRQC_RSSEN)) { /* RSS disabled */
2773 if (rss_hf != 0) /* Enable RSS */
2775 return 0; /* Nothing to do */
2778 if (rss_hf == 0) /* Disable RSS */
2780 ixgbe_hw_rss_hash_set(hw, rss_conf);
2785 ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
2786 struct rte_eth_rss_conf *rss_conf)
2788 struct ixgbe_hw *hw;
2795 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2796 hash_key = rss_conf->rss_key;
2797 if (hash_key != NULL) {
2798 /* Return RSS hash key */
2799 for (i = 0; i < 10; i++) {
2800 rss_key = IXGBE_READ_REG_ARRAY(hw, IXGBE_RSSRK(0), i);
2801 hash_key[(i * 4)] = rss_key & 0x000000FF;
2802 hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
2803 hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
2804 hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
2808 /* Get RSS functions configured in MRQC register */
2809 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2810 if ((mrqc & IXGBE_MRQC_RSSEN) == 0) { /* RSS is disabled */
2811 rss_conf->rss_hf = 0;
2815 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
2816 rss_hf |= ETH_RSS_IPV4;
2817 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
2818 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2819 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
2820 rss_hf |= ETH_RSS_IPV6;
2821 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
2822 rss_hf |= ETH_RSS_IPV6_EX;
2823 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
2824 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2825 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
2826 rss_hf |= ETH_RSS_IPV6_TCP_EX;
2827 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
2828 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2829 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
2830 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2831 if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
2832 rss_hf |= ETH_RSS_IPV6_UDP_EX;
2833 rss_conf->rss_hf = rss_hf;
2838 ixgbe_rss_configure(struct rte_eth_dev *dev)
2840 struct rte_eth_rss_conf rss_conf;
2841 struct ixgbe_hw *hw;
2846 PMD_INIT_FUNC_TRACE();
2847 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2850 * Fill in redirection table
2851 * The byte-swap is needed because NIC registers are in
2852 * little-endian order.
2855 for (i = 0, j = 0; i < 128; i++, j++) {
2856 if (j == dev->data->nb_rx_queues)
2858 reta = (reta << 8) | j;
2860 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2),
2865 * Configure the RSS key and the RSS protocols used to compute
2866 * the RSS hash of input packets.
2868 rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
2869 if ((rss_conf.rss_hf & IXGBE_RSS_OFFLOAD_ALL) == 0) {
2870 ixgbe_rss_disable(dev);
2873 if (rss_conf.rss_key == NULL)
2874 rss_conf.rss_key = rss_intel_key; /* Default hash key */
2875 ixgbe_hw_rss_hash_set(hw, &rss_conf);
2878 #define NUM_VFTA_REGISTERS 128
2879 #define NIC_RX_BUFFER_SIZE 0x200
2882 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
2884 struct rte_eth_vmdq_dcb_conf *cfg;
2885 struct ixgbe_hw *hw;
2886 enum rte_eth_nb_pools num_pools;
2887 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
2889 uint8_t nb_tcs; /* number of traffic classes */
2892 PMD_INIT_FUNC_TRACE();
2893 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2894 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
2895 num_pools = cfg->nb_queue_pools;
2896 /* Check we have a valid number of pools */
2897 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
2898 ixgbe_rss_disable(dev);
2901 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
2902 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
2906 * split rx buffer up into sections, each for 1 traffic class
2908 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
2909 for (i = 0 ; i < nb_tcs; i++) {
2910 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2911 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
2912 /* clear 10 bits. */
2913 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
2914 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2916 /* zero alloc all unused TCs */
2917 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2918 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2919 rxpbsize &= (~( 0x3FF << IXGBE_RXPBSIZE_SHIFT ));
2920 /* clear 10 bits. */
2921 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2924 /* MRQC: enable vmdq and dcb */
2925 mrqc = ((num_pools == ETH_16_POOLS) ? \
2926 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN );
2927 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2929 /* PFVTCTL: turn on virtualisation and set the default pool */
2930 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2931 if (cfg->enable_default_pool) {
2932 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
2934 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
2937 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
2939 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
2941 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
2943 * mapping is done with 3 bits per priority,
2944 * so shift by i*3 each time
2946 queue_mapping |= ((cfg->dcb_queue[i] & 0x07) << (i * 3));
2948 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
2950 /* RTRPCS: DCB related */
2951 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
2953 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
2954 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2955 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
2956 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
2958 /* VFTA - enable all vlan filters */
2959 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
2960 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
2963 /* VFRE: pool enabling for receive - 16 or 32 */
2964 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), \
2965 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
2968 * MPSAR - allow pools to read specific mac addresses
2969 * In this case, all pools should be able to read from mac addr 0
2971 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
2972 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
2974 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
2975 for (i = 0; i < cfg->nb_pool_maps; i++) {
2976 /* set vlan id in VF register and set the valid bit */
2977 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
2978 (cfg->pool_map[i].vlan_id & 0xFFF)));
2980 * Put the allowed pools in VFB reg. As we only have 16 or 32
2981 * pools, we only need to use the first half of the register
2984 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
2989 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
2990 * @hw: pointer to hardware structure
2991 * @dcb_config: pointer to ixgbe_dcb_config structure
2994 ixgbe_dcb_tx_hw_config(struct ixgbe_hw *hw,
2995 struct ixgbe_dcb_config *dcb_config)
3000 PMD_INIT_FUNC_TRACE();
3001 if (hw->mac.type != ixgbe_mac_82598EB) {
3002 /* Disable the Tx desc arbiter so that MTQC can be changed */
3003 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3004 reg |= IXGBE_RTTDCS_ARBDIS;
3005 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3007 /* Enable DCB for Tx with 8 TCs */
3008 if (dcb_config->num_tcs.pg_tcs == 8) {
3009 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3012 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3014 if (dcb_config->vt_mode)
3015 reg |= IXGBE_MTQC_VT_ENA;
3016 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3018 /* Disable drop for all queues */
3019 for (q = 0; q < 128; q++)
3020 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3021 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3023 /* Enable the Tx desc arbiter */
3024 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3025 reg &= ~IXGBE_RTTDCS_ARBDIS;
3026 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3028 /* Enable Security TX Buffer IFG for DCB */
3029 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3030 reg |= IXGBE_SECTX_DCB;
3031 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
3037 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
3038 * @dev: pointer to rte_eth_dev structure
3039 * @dcb_config: pointer to ixgbe_dcb_config structure
3042 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
3043 struct ixgbe_dcb_config *dcb_config)
3045 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3046 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3047 struct ixgbe_hw *hw =
3048 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3050 PMD_INIT_FUNC_TRACE();
3051 if (hw->mac.type != ixgbe_mac_82598EB)
3052 /*PF VF Transmit Enable*/
3053 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
3054 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
3056 /*Configure general DCB TX parameters*/
3057 ixgbe_dcb_tx_hw_config(hw,dcb_config);
3062 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
3063 struct ixgbe_dcb_config *dcb_config)
3065 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
3066 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
3067 struct ixgbe_dcb_tc_config *tc;
3070 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3071 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS ) {
3072 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3073 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3076 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3077 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3079 /* User Priority to Traffic Class mapping */
3080 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3081 j = vmdq_rx_conf->dcb_queue[i];
3082 tc = &dcb_config->tc_config[j];
3083 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3089 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
3090 struct ixgbe_dcb_config *dcb_config)
3092 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
3093 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
3094 struct ixgbe_dcb_tc_config *tc;
3097 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
3098 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ) {
3099 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
3100 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
3103 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
3104 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
3107 /* User Priority to Traffic Class mapping */
3108 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3109 j = vmdq_tx_conf->dcb_queue[i];
3110 tc = &dcb_config->tc_config[j];
3111 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3118 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
3119 struct ixgbe_dcb_config *dcb_config)
3121 struct rte_eth_dcb_rx_conf *rx_conf =
3122 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
3123 struct ixgbe_dcb_tc_config *tc;
3126 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
3127 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
3129 /* User Priority to Traffic Class mapping */
3130 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3131 j = rx_conf->dcb_queue[i];
3132 tc = &dcb_config->tc_config[j];
3133 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
3139 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
3140 struct ixgbe_dcb_config *dcb_config)
3142 struct rte_eth_dcb_tx_conf *tx_conf =
3143 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
3144 struct ixgbe_dcb_tc_config *tc;
3147 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
3148 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
3150 /* User Priority to Traffic Class mapping */
3151 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3152 j = tx_conf->dcb_queue[i];
3153 tc = &dcb_config->tc_config[j];
3154 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
3160 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
3161 * @hw: pointer to hardware structure
3162 * @dcb_config: pointer to ixgbe_dcb_config structure
3165 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
3166 struct ixgbe_dcb_config *dcb_config)
3172 PMD_INIT_FUNC_TRACE();
3174 * Disable the arbiter before changing parameters
3175 * (always enable recycle mode; WSP)
3177 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
3178 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3180 if (hw->mac.type != ixgbe_mac_82598EB) {
3181 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
3182 if (dcb_config->num_tcs.pg_tcs == 4) {
3183 if (dcb_config->vt_mode)
3184 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3185 IXGBE_MRQC_VMDQRT4TCEN;
3187 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3188 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3192 if (dcb_config->num_tcs.pg_tcs == 8) {
3193 if (dcb_config->vt_mode)
3194 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3195 IXGBE_MRQC_VMDQRT8TCEN;
3197 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
3198 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
3203 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
3206 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3207 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3208 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
3209 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3211 /* VFTA - enable all vlan filters */
3212 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
3213 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
3217 * Configure Rx packet plane (recycle mode; WSP) and
3220 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
3221 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
3227 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
3228 uint16_t *max,uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3230 switch (hw->mac.type) {
3231 case ixgbe_mac_82598EB:
3232 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
3234 case ixgbe_mac_82599EB:
3235 case ixgbe_mac_X540:
3236 case ixgbe_mac_X550:
3237 case ixgbe_mac_X550EM_x:
3238 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
3247 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
3248 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
3250 switch (hw->mac.type) {
3251 case ixgbe_mac_82598EB:
3252 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id,tsa);
3253 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id,tsa);
3255 case ixgbe_mac_82599EB:
3256 case ixgbe_mac_X540:
3257 case ixgbe_mac_X550:
3258 case ixgbe_mac_X550EM_x:
3259 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id,tsa);
3260 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id,tsa, map);
3267 #define DCB_RX_CONFIG 1
3268 #define DCB_TX_CONFIG 1
3269 #define DCB_TX_PB 1024
3271 * ixgbe_dcb_hw_configure - Enable DCB and configure
3272 * general DCB in VT mode and non-VT mode parameters
3273 * @dev: pointer to rte_eth_dev structure
3274 * @dcb_config: pointer to ixgbe_dcb_config structure
3277 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
3278 struct ixgbe_dcb_config *dcb_config)
3281 uint8_t i,pfc_en,nb_tcs;
3283 uint8_t config_dcb_rx = 0;
3284 uint8_t config_dcb_tx = 0;
3285 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3286 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3287 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3288 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3289 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
3290 struct ixgbe_dcb_tc_config *tc;
3291 uint32_t max_frame = dev->data->mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
3292 struct ixgbe_hw *hw =
3293 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3295 switch(dev->data->dev_conf.rxmode.mq_mode){
3296 case ETH_MQ_RX_VMDQ_DCB:
3297 dcb_config->vt_mode = true;
3298 if (hw->mac.type != ixgbe_mac_82598EB) {
3299 config_dcb_rx = DCB_RX_CONFIG;
3301 *get dcb and VT rx configuration parameters
3304 ixgbe_vmdq_dcb_rx_config(dev,dcb_config);
3305 /*Configure general VMDQ and DCB RX parameters*/
3306 ixgbe_vmdq_dcb_configure(dev);
3310 dcb_config->vt_mode = false;
3311 config_dcb_rx = DCB_RX_CONFIG;
3312 /* Get dcb TX configuration parameters from rte_eth_conf */
3313 ixgbe_dcb_rx_config(dev,dcb_config);
3314 /*Configure general DCB RX parameters*/
3315 ixgbe_dcb_rx_hw_config(hw, dcb_config);
3318 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
3321 switch (dev->data->dev_conf.txmode.mq_mode) {
3322 case ETH_MQ_TX_VMDQ_DCB:
3323 dcb_config->vt_mode = true;
3324 config_dcb_tx = DCB_TX_CONFIG;
3325 /* get DCB and VT TX configuration parameters from rte_eth_conf */
3326 ixgbe_dcb_vt_tx_config(dev,dcb_config);
3327 /*Configure general VMDQ and DCB TX parameters*/
3328 ixgbe_vmdq_dcb_hw_tx_config(dev,dcb_config);
3332 dcb_config->vt_mode = false;
3333 config_dcb_tx = DCB_TX_CONFIG;
3334 /*get DCB TX configuration parameters from rte_eth_conf*/
3335 ixgbe_dcb_tx_config(dev,dcb_config);
3336 /*Configure general DCB TX parameters*/
3337 ixgbe_dcb_tx_hw_config(hw, dcb_config);
3340 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
3344 nb_tcs = dcb_config->num_tcs.pfc_tcs;
3346 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
3347 if(nb_tcs == ETH_4_TCS) {
3348 /* Avoid un-configured priority mapping to TC0 */
3350 uint8_t mask = 0xFF;
3351 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
3352 mask = (uint8_t)(mask & (~ (1 << map[i])));
3353 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
3354 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
3358 /* Re-configure 4 TCs BW */
3359 for (i = 0; i < nb_tcs; i++) {
3360 tc = &dcb_config->tc_config[i];
3361 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
3362 (uint8_t)(100 / nb_tcs);
3363 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
3364 (uint8_t)(100 / nb_tcs);
3366 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
3367 tc = &dcb_config->tc_config[i];
3368 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
3369 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
3374 /* Set RX buffer size */
3375 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
3376 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
3377 for (i = 0 ; i < nb_tcs; i++) {
3378 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
3380 /* zero alloc all unused TCs */
3381 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3382 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
3386 /* Only support an equally distributed Tx packet buffer strategy. */
3387 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
3388 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
3389 for (i = 0; i < nb_tcs; i++) {
3390 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
3391 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
3393 /* Clear unused TCs, if any, to zero buffer size*/
3394 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
3395 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
3396 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
3400 /*Calculates traffic class credits*/
3401 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
3402 IXGBE_DCB_TX_CONFIG);
3403 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
3404 IXGBE_DCB_RX_CONFIG);
3407 /* Unpack CEE standard containers */
3408 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
3409 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3410 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
3411 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
3412 /* Configure PG(ETS) RX */
3413 ixgbe_dcb_hw_arbite_rx_config(hw,refill,max,bwgid,tsa,map);
3417 /* Unpack CEE standard containers */
3418 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
3419 ixgbe_dcb_unpack_max_cee(dcb_config, max);
3420 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
3421 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
3422 /* Configure PG(ETS) TX */
3423 ixgbe_dcb_hw_arbite_tx_config(hw,refill,max,bwgid,tsa,map);
3426 /*Configure queue statistics registers*/
3427 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3429 /* Check if the PFC is supported */
3430 if(dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3431 pbsize = (uint16_t) (NIC_RX_BUFFER_SIZE / nb_tcs);
3432 for (i = 0; i < nb_tcs; i++) {
3434 * If the TC count is 8,and the default high_water is 48,
3435 * the low_water is 16 as default.
3437 hw->fc.high_water[i] = (pbsize * 3 ) / 4;
3438 hw->fc.low_water[i] = pbsize / 4;
3439 /* Enable pfc for this TC */
3440 tc = &dcb_config->tc_config[i];
3441 tc->pfc = ixgbe_dcb_pfc_enabled;
3443 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3444 if(dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3446 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3453 * ixgbe_configure_dcb - Configure DCB Hardware
3454 * @dev: pointer to rte_eth_dev
3456 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3458 struct ixgbe_dcb_config *dcb_cfg =
3459 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3460 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3462 PMD_INIT_FUNC_TRACE();
3464 /* check support mq_mode for DCB */
3465 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) &&
3466 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB))
3469 if (dev->data->nb_rx_queues != ETH_DCB_NUM_QUEUES)
3472 /** Configure DCB hardware **/
3473 ixgbe_dcb_hw_configure(dev,dcb_cfg);
3479 * VMDq only support for 10 GbE NIC.
3482 ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
3484 struct rte_eth_vmdq_rx_conf *cfg;
3485 struct ixgbe_hw *hw;
3486 enum rte_eth_nb_pools num_pools;
3487 uint32_t mrqc, vt_ctl, vlanctrl;
3491 PMD_INIT_FUNC_TRACE();
3492 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3493 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
3494 num_pools = cfg->nb_queue_pools;
3496 ixgbe_rss_disable(dev);
3498 /* MRQC: enable vmdq */
3499 mrqc = IXGBE_MRQC_VMDQEN;
3500 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3502 /* PFVTCTL: turn on virtualisation and set the default pool */
3503 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
3504 if (cfg->enable_default_pool)
3505 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
3507 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
3509 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
3511 for (i = 0; i < (int)num_pools; i++) {
3512 vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
3513 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
3516 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
3517 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3518 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
3519 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
3521 /* VFTA - enable all vlan filters */
3522 for (i = 0; i < NUM_VFTA_REGISTERS; i++)
3523 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), UINT32_MAX);
3525 /* VFRE: pool enabling for receive - 64 */
3526 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), UINT32_MAX);
3527 if (num_pools == ETH_64_POOLS)
3528 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), UINT32_MAX);
3531 * MPSAR - allow pools to read specific mac addresses
3532 * In this case, all pools should be able to read from mac addr 0
3534 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), UINT32_MAX);
3535 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), UINT32_MAX);
3537 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
3538 for (i = 0; i < cfg->nb_pool_maps; i++) {
3539 /* set vlan id in VF register and set the valid bit */
3540 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
3541 (cfg->pool_map[i].vlan_id & IXGBE_RXD_VLAN_ID_MASK)));
3543 * Put the allowed pools in VFB reg. As we only have 16 or 64
3544 * pools, we only need to use the first half of the register
3547 if (((cfg->pool_map[i].pools >> 32) & UINT32_MAX) == 0)
3548 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), \
3549 (cfg->pool_map[i].pools & UINT32_MAX));
3551 IXGBE_WRITE_REG(hw, IXGBE_VLVFB((i*2+1)), \
3552 ((cfg->pool_map[i].pools >> 32) \
3557 /* PFDMA Tx General Switch Control Enables VMDQ loopback */
3558 if (cfg->enable_loop_back) {
3559 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3560 for (i = 0; i < RTE_IXGBE_VMTXSW_REGISTER_COUNT; i++)
3561 IXGBE_WRITE_REG(hw, IXGBE_VMTXSW(i), UINT32_MAX);
3564 IXGBE_WRITE_FLUSH(hw);
3568 * ixgbe_dcb_config_tx_hw_config - Configure general VMDq TX parameters
3569 * @hw: pointer to hardware structure
3572 ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
3577 PMD_INIT_FUNC_TRACE();
3578 /*PF VF Transmit Enable*/
3579 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), UINT32_MAX);
3580 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), UINT32_MAX);
3582 /* Disable the Tx desc arbiter so that MTQC can be changed */
3583 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3584 reg |= IXGBE_RTTDCS_ARBDIS;
3585 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3587 reg = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3588 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
3590 /* Disable drop for all queues */
3591 for (q = 0; q < IXGBE_MAX_RX_QUEUE_NUM; q++)
3592 IXGBE_WRITE_REG(hw, IXGBE_QDE,
3593 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
3595 /* Enable the Tx desc arbiter */
3596 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3597 reg &= ~IXGBE_RTTDCS_ARBDIS;
3598 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
3600 IXGBE_WRITE_FLUSH(hw);
3605 static int __attribute__((cold))
3606 ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
3608 struct ixgbe_rx_entry *rxe = rxq->sw_ring;
3612 /* Initialize software ring entries */
3613 for (i = 0; i < rxq->nb_rx_desc; i++) {
3614 volatile union ixgbe_adv_rx_desc *rxd;
3615 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mb_pool);
3617 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
3618 (unsigned) rxq->queue_id);
3622 rte_mbuf_refcnt_set(mbuf, 1);
3624 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
3626 mbuf->port = rxq->port_id;
3629 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
3630 rxd = &rxq->rx_ring[i];
3631 rxd->read.hdr_addr = dma_addr;
3632 rxd->read.pkt_addr = dma_addr;
3640 ixgbe_config_vf_rss(struct rte_eth_dev *dev)
3642 struct ixgbe_hw *hw;
3645 ixgbe_rss_configure(dev);
3647 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3649 /* MRQC: enable VF RSS */
3650 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
3651 mrqc &= ~IXGBE_MRQC_MRQE_MASK;
3652 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3654 mrqc |= IXGBE_MRQC_VMDQRSS64EN;
3658 mrqc |= IXGBE_MRQC_VMDQRSS32EN;
3662 PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
3666 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3672 ixgbe_config_vf_default(struct rte_eth_dev *dev)
3674 struct ixgbe_hw *hw =
3675 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3677 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3679 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3684 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3685 IXGBE_MRQC_VMDQRT4TCEN);
3689 IXGBE_WRITE_REG(hw, IXGBE_MRQC,
3690 IXGBE_MRQC_VMDQRT8TCEN);
3694 "invalid pool number in IOV mode");
3701 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
3703 struct ixgbe_hw *hw =
3704 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3706 if (hw->mac.type == ixgbe_mac_82598EB)
3709 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3711 * SRIOV inactive scheme
3712 * any DCB/RSS w/o VMDq multi-queue setting
3714 switch (dev->data->dev_conf.rxmode.mq_mode) {
3716 ixgbe_rss_configure(dev);
3719 case ETH_MQ_RX_VMDQ_DCB:
3720 ixgbe_vmdq_dcb_configure(dev);
3723 case ETH_MQ_RX_VMDQ_ONLY:
3724 ixgbe_vmdq_rx_hw_configure(dev);
3727 case ETH_MQ_RX_NONE:
3728 /* if mq_mode is none, disable rss mode.*/
3729 default: ixgbe_rss_disable(dev);
3733 * SRIOV active scheme
3734 * Support RSS together with VMDq & SRIOV
3736 switch (dev->data->dev_conf.rxmode.mq_mode) {
3738 case ETH_MQ_RX_VMDQ_RSS:
3739 ixgbe_config_vf_rss(dev);
3742 /* FIXME if support DCB/RSS together with VMDq & SRIOV */
3743 case ETH_MQ_RX_VMDQ_DCB:
3744 case ETH_MQ_RX_VMDQ_DCB_RSS:
3746 "Could not support DCB with VMDq & SRIOV");
3749 ixgbe_config_vf_default(dev);
3758 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
3760 struct ixgbe_hw *hw =
3761 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3765 if (hw->mac.type == ixgbe_mac_82598EB)
3768 /* disable arbiter before setting MTQC */
3769 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3770 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3771 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3773 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3775 * SRIOV inactive scheme
3776 * any DCB w/o VMDq multi-queue setting
3778 if (dev->data->dev_conf.txmode.mq_mode == ETH_MQ_TX_VMDQ_ONLY)
3779 ixgbe_vmdq_tx_hw_configure(hw);
3781 mtqc = IXGBE_MTQC_64Q_1PB;
3782 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3785 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3788 * SRIOV active scheme
3789 * FIXME if support DCB together with VMDq & SRIOV
3792 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3795 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
3798 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
3802 mtqc = IXGBE_MTQC_64Q_1PB;
3803 PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
3805 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3808 /* re-enable arbiter */
3809 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3810 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3816 * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
3818 * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
3819 * spec rev. 3.0 chapter 8.2.3.8.13.
3821 * @pool Memory pool of the Rx queue
3823 static inline uint32_t
3824 ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
3826 struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
3828 /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
3831 (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
3834 return IXGBE_RSCCTL_MAXDESC_16;
3835 else if (maxdesc >= 8)
3836 return IXGBE_RSCCTL_MAXDESC_8;
3837 else if (maxdesc >= 4)
3838 return IXGBE_RSCCTL_MAXDESC_4;
3840 return IXGBE_RSCCTL_MAXDESC_1;
3844 * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
3847 * (Taken from FreeBSD tree)
3848 * (yes this is all very magic and confusing :)
3851 * @entry the register array entry
3852 * @vector the MSIX vector for this queue
3856 ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
3858 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3861 vector |= IXGBE_IVAR_ALLOC_VAL;
3863 switch (hw->mac.type) {
3865 case ixgbe_mac_82598EB:
3867 entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
3869 entry += (type * 64);
3870 index = (entry >> 2) & 0x1F;
3871 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
3872 ivar &= ~(0xFF << (8 * (entry & 0x3)));
3873 ivar |= (vector << (8 * (entry & 0x3)));
3874 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
3877 case ixgbe_mac_82599EB:
3878 case ixgbe_mac_X540:
3879 if (type == -1) { /* MISC IVAR */
3880 index = (entry & 1) * 8;
3881 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
3882 ivar &= ~(0xFF << index);
3883 ivar |= (vector << index);
3884 IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
3885 } else { /* RX/TX IVARS */
3886 index = (16 * (entry & 1)) + (8 * type);
3887 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
3888 ivar &= ~(0xFF << index);
3889 ivar |= (vector << index);
3890 IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
3900 void __attribute__((cold))
3901 ixgbe_set_rx_function(struct rte_eth_dev *dev)
3903 uint16_t i, rx_using_sse;
3904 struct ixgbe_adapter *adapter =
3905 (struct ixgbe_adapter *)dev->data->dev_private;
3908 * In order to allow Vector Rx there are a few configuration
3909 * conditions to be met and Rx Bulk Allocation should be allowed.
3911 if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
3912 !adapter->rx_bulk_alloc_allowed) {
3913 PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
3914 "preconditions or RTE_IXGBE_INC_VECTOR is "
3916 dev->data->port_id);
3918 adapter->rx_vec_allowed = false;
3922 * Initialize the appropriate LRO callback.
3924 * If all queues satisfy the bulk allocation preconditions
3925 * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
3926 * Otherwise use a single allocation version.
3928 if (dev->data->lro) {
3929 if (adapter->rx_bulk_alloc_allowed) {
3930 PMD_INIT_LOG(INFO, "LRO is requested. Using a bulk "
3931 "allocation version");
3932 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
3934 PMD_INIT_LOG(INFO, "LRO is requested. Using a single "
3935 "allocation version");
3936 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
3938 } else if (dev->data->scattered_rx) {
3940 * Set the non-LRO scattered callback: there are Vector and
3941 * single allocation versions.
3943 if (adapter->rx_vec_allowed) {
3944 PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
3945 "callback (port=%d).",
3946 dev->data->port_id);
3948 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
3949 } else if (adapter->rx_bulk_alloc_allowed) {
3950 PMD_INIT_LOG(INFO, "Using a Scattered with bulk "
3951 "allocation callback (port=%d).",
3952 dev->data->port_id);
3953 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
3955 PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
3956 "single allocation) "
3957 "Scattered Rx callback "
3959 dev->data->port_id);
3961 dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
3964 * Below we set "simple" callbacks according to port/queues parameters.
3965 * If parameters allow we are going to choose between the following
3969 * - Single buffer allocation (the simplest one)
3971 } else if (adapter->rx_vec_allowed) {
3972 PMD_INIT_LOG(INFO, "Vector rx enabled, please make sure RX "
3973 "burst size no less than 32.");
3975 dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
3976 } else if (adapter->rx_bulk_alloc_allowed) {
3977 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
3978 "satisfied. Rx Burst Bulk Alloc function "
3979 "will be used on port=%d.",
3980 dev->data->port_id);
3982 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
3984 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
3985 "satisfied, or Scattered Rx is requested "
3987 dev->data->port_id);
3989 dev->rx_pkt_burst = ixgbe_recv_pkts;
3992 /* Propagate information about RX function choice through all queues. */
3995 (dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec ||
3996 dev->rx_pkt_burst == ixgbe_recv_pkts_vec);
3998 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3999 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4000 rxq->rx_using_sse = rx_using_sse;
4005 * ixgbe_set_rsc - configure RSC related port HW registers
4007 * Configures the port's RSC related registers according to the 4.6.7.2 chapter
4008 * of 82599 Spec (x540 configuration is virtually the same).
4012 * Returns 0 in case of success or a non-zero error code
4015 ixgbe_set_rsc(struct rte_eth_dev *dev)
4017 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4018 struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4019 struct rte_eth_dev_info dev_info = { 0 };
4020 bool rsc_capable = false;
4025 dev->dev_ops->dev_infos_get(dev, &dev_info);
4026 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
4029 if (!rsc_capable && rx_conf->enable_lro) {
4030 PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
4035 /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
4037 if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
4039 * According to chapter of 4.6.7.2.1 of the Spec Rev.
4040 * 3.0 RSC configuration requires HW CRC stripping being
4041 * enabled. If user requested both HW CRC stripping off
4042 * and RSC on - return an error.
4044 PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
4049 /* RFCTL configuration */
4051 uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4052 if (rx_conf->enable_lro)
4054 * Since NFS packets coalescing is not supported - clear
4055 * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
4058 rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
4059 IXGBE_RFCTL_NFSR_DIS);
4061 rfctl |= IXGBE_RFCTL_RSC_DIS;
4063 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4066 /* If LRO hasn't been requested - we are done here. */
4067 if (!rx_conf->enable_lro)
4070 /* Set RDRXCTL.RSCACKC bit */
4071 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4072 rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
4073 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4075 /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
4076 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4077 struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
4079 IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
4081 IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
4083 IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
4085 IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
4088 * ixgbe PMD doesn't support header-split at the moment.
4090 * Following the 4.6.7.2.1 chapter of the 82599/x540
4091 * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
4092 * should be configured even if header split is not
4093 * enabled. We will configure it 128 bytes following the
4094 * recommendation in the spec.
4096 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
4097 srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4098 IXGBE_SRRCTL_BSIZEHDR_MASK;
4101 * TODO: Consider setting the Receive Descriptor Minimum
4102 * Threshold Size for an RSC case. This is not an obviously
4103 * beneficiary option but the one worth considering...
4106 rscctl |= IXGBE_RSCCTL_RSCEN;
4107 rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
4108 psrtype |= IXGBE_PSRTYPE_TCPHDR;
4111 * RSC: Set ITR interval corresponding to 2K ints/s.
4113 * Full-sized RSC aggregations for a 10Gb/s link will
4114 * arrive at about 20K aggregation/s rate.
4116 * 2K inst/s rate will make only 10% of the
4117 * aggregations to be closed due to the interrupt timer
4118 * expiration for a streaming at wire-speed case.
4120 * For a sparse streaming case this setting will yield
4121 * at most 500us latency for a single RSC aggregation.
4123 eitr &= ~IXGBE_EITR_ITR_INT_MASK;
4124 eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
4126 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4127 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
4128 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4129 IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
4132 * RSC requires the mapping of the queue to the
4135 ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
4140 PMD_INIT_LOG(INFO, "enabling LRO mode");
4146 * Initializes Receive Unit.
4148 int __attribute__((cold))
4149 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
4151 struct ixgbe_hw *hw;
4152 struct ixgbe_rx_queue *rxq;
4163 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
4166 PMD_INIT_FUNC_TRACE();
4167 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4170 * Make sure receives are disabled while setting
4171 * up the RX context (registers, descriptor rings, etc.).
4173 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4174 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4176 /* Enable receipt of broadcasted frames */
4177 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4178 fctrl |= IXGBE_FCTRL_BAM;
4179 fctrl |= IXGBE_FCTRL_DPF;
4180 fctrl |= IXGBE_FCTRL_PMCF;
4181 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4184 * Configure CRC stripping, if any.
4186 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4187 if (rx_conf->hw_strip_crc)
4188 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
4190 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
4193 * Configure jumbo frame support, if any.
4195 if (rx_conf->jumbo_frame == 1) {
4196 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4197 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
4198 maxfrs &= 0x0000FFFF;
4199 maxfrs |= (rx_conf->max_rx_pkt_len << 16);
4200 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
4202 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
4205 * If loopback mode is configured for 82599, set LPBK bit.
4207 if (hw->mac.type == ixgbe_mac_82599EB &&
4208 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4209 hlreg0 |= IXGBE_HLREG0_LPBK;
4211 hlreg0 &= ~IXGBE_HLREG0_LPBK;
4213 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4215 /* Setup RX queues */
4216 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4217 rxq = dev->data->rx_queues[i];
4220 * Reset crc_len in case it was changed after queue setup by a
4221 * call to configure.
4223 rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
4225 /* Setup the Base and Length of the Rx Descriptor Rings */
4226 bus_addr = rxq->rx_ring_phys_addr;
4227 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
4228 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4229 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
4230 (uint32_t)(bus_addr >> 32));
4231 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
4232 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4233 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4234 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
4236 /* Configure the SRRCTL register */
4237 #ifdef RTE_HEADER_SPLIT_ENABLE
4239 * Configure Header Split
4241 if (rx_conf->header_split) {
4242 if (hw->mac.type == ixgbe_mac_82599EB) {
4243 /* Must setup the PSRTYPE register */
4245 psrtype = IXGBE_PSRTYPE_TCPHDR |
4246 IXGBE_PSRTYPE_UDPHDR |
4247 IXGBE_PSRTYPE_IPV4HDR |
4248 IXGBE_PSRTYPE_IPV6HDR;
4249 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
4251 srrctl = ((rx_conf->split_hdr_size <<
4252 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4253 IXGBE_SRRCTL_BSIZEHDR_MASK);
4254 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4257 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4259 /* Set if packets are dropped when no descriptors available */
4261 srrctl |= IXGBE_SRRCTL_DROP_EN;
4264 * Configure the RX buffer size in the BSIZEPACKET field of
4265 * the SRRCTL register of the queue.
4266 * The value is in 1 KB resolution. Valid values can be from
4269 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4270 RTE_PKTMBUF_HEADROOM);
4271 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4272 IXGBE_SRRCTL_BSIZEPKT_MASK);
4274 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
4276 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4277 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4279 /* It adds dual VLAN length for supporting dual VLAN */
4280 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4281 2 * IXGBE_VLAN_TAG_SIZE > buf_size)
4282 dev->data->scattered_rx = 1;
4285 if (rx_conf->enable_scatter)
4286 dev->data->scattered_rx = 1;
4289 * Device configured with multiple RX queues.
4291 ixgbe_dev_mq_rx_configure(dev);
4294 * Setup the Checksum Register.
4295 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
4296 * Enable IP/L4 checkum computation by hardware if requested to do so.
4298 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
4299 rxcsum |= IXGBE_RXCSUM_PCSD;
4300 if (rx_conf->hw_ip_checksum)
4301 rxcsum |= IXGBE_RXCSUM_IPPCSE;
4303 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
4305 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
4307 if (hw->mac.type == ixgbe_mac_82599EB ||
4308 hw->mac.type == ixgbe_mac_X540) {
4309 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4310 if (rx_conf->hw_strip_crc)
4311 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4313 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
4314 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4315 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4318 rc = ixgbe_set_rsc(dev);
4322 ixgbe_set_rx_function(dev);
4328 * Initializes Transmit Unit.
4330 void __attribute__((cold))
4331 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
4333 struct ixgbe_hw *hw;
4334 struct ixgbe_tx_queue *txq;
4340 PMD_INIT_FUNC_TRACE();
4341 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4343 /* Enable TX CRC (checksum offload requirement) and hw padding
4344 * (TSO requirement) */
4345 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4346 hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
4347 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4349 /* Setup the Base and Length of the Tx Descriptor Rings */
4350 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4351 txq = dev->data->tx_queues[i];
4353 bus_addr = txq->tx_ring_phys_addr;
4354 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
4355 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4356 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
4357 (uint32_t)(bus_addr >> 32));
4358 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
4359 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4360 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4361 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4362 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4365 * Disable Tx Head Writeback RO bit, since this hoses
4366 * bookkeeping if things aren't delivered in order.
4368 switch (hw->mac.type) {
4369 case ixgbe_mac_82598EB:
4370 txctrl = IXGBE_READ_REG(hw,
4371 IXGBE_DCA_TXCTRL(txq->reg_idx));
4372 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4373 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
4377 case ixgbe_mac_82599EB:
4378 case ixgbe_mac_X540:
4379 case ixgbe_mac_X550:
4380 case ixgbe_mac_X550EM_x:
4382 txctrl = IXGBE_READ_REG(hw,
4383 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
4384 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4385 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
4391 /* Device configured with multiple TX queues. */
4392 ixgbe_dev_mq_tx_configure(dev);
4396 * Set up link for 82599 loopback mode Tx->Rx.
4398 static inline void __attribute__((cold))
4399 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
4401 PMD_INIT_FUNC_TRACE();
4403 if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
4404 if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
4406 PMD_INIT_LOG(ERR, "Could not enable loopback mode");
4415 IXGBE_AUTOC_LMS_10G_LINK_NO_AN | IXGBE_AUTOC_FLU);
4416 ixgbe_reset_pipeline_82599(hw);
4418 hw->mac.ops.release_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM);
4424 * Start Transmit and Receive Units.
4426 int __attribute__((cold))
4427 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
4429 struct ixgbe_hw *hw;
4430 struct ixgbe_tx_queue *txq;
4431 struct ixgbe_rx_queue *rxq;
4438 PMD_INIT_FUNC_TRACE();
4439 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4441 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4442 txq = dev->data->tx_queues[i];
4443 /* Setup Transmit Threshold Registers */
4444 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4445 txdctl |= txq->pthresh & 0x7F;
4446 txdctl |= ((txq->hthresh & 0x7F) << 8);
4447 txdctl |= ((txq->wthresh & 0x7F) << 16);
4448 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4451 if (hw->mac.type != ixgbe_mac_82598EB) {
4452 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
4453 dmatxctl |= IXGBE_DMATXCTL_TE;
4454 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
4457 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4458 txq = dev->data->tx_queues[i];
4459 if (!txq->tx_deferred_start) {
4460 ret = ixgbe_dev_tx_queue_start(dev, i);
4466 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4467 rxq = dev->data->rx_queues[i];
4468 if (!rxq->rx_deferred_start) {
4469 ret = ixgbe_dev_rx_queue_start(dev, i);
4475 /* Enable Receive engine */
4476 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4477 if (hw->mac.type == ixgbe_mac_82598EB)
4478 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4479 rxctrl |= IXGBE_RXCTRL_RXEN;
4480 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4482 /* If loopback mode is enabled for 82599, set up the link accordingly */
4483 if (hw->mac.type == ixgbe_mac_82599EB &&
4484 dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
4485 ixgbe_setup_loopback_link_82599(hw);
4491 * Start Receive Units for specified queue.
4493 int __attribute__((cold))
4494 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4496 struct ixgbe_hw *hw;
4497 struct ixgbe_rx_queue *rxq;
4501 PMD_INIT_FUNC_TRACE();
4502 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4504 if (rx_queue_id < dev->data->nb_rx_queues) {
4505 rxq = dev->data->rx_queues[rx_queue_id];
4507 /* Allocate buffers for descriptor rings */
4508 if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
4509 PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
4513 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4514 rxdctl |= IXGBE_RXDCTL_ENABLE;
4515 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4517 /* Wait until RX Enable ready */
4518 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4521 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4522 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4524 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
4527 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
4528 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
4536 * Stop Receive Units for specified queue.
4538 int __attribute__((cold))
4539 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
4541 struct ixgbe_hw *hw;
4542 struct ixgbe_adapter *adapter =
4543 (struct ixgbe_adapter *)dev->data->dev_private;
4544 struct ixgbe_rx_queue *rxq;
4548 PMD_INIT_FUNC_TRACE();
4549 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4551 if (rx_queue_id < dev->data->nb_rx_queues) {
4552 rxq = dev->data->rx_queues[rx_queue_id];
4554 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4555 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4556 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
4558 /* Wait until RX Enable ready */
4559 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4562 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
4563 } while (--poll_ms && (rxdctl | IXGBE_RXDCTL_ENABLE));
4565 PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
4568 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4570 ixgbe_rx_queue_release_mbufs(rxq);
4571 ixgbe_reset_rx_queue(adapter, rxq);
4580 * Start Transmit Units for specified queue.
4582 int __attribute__((cold))
4583 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4585 struct ixgbe_hw *hw;
4586 struct ixgbe_tx_queue *txq;
4590 PMD_INIT_FUNC_TRACE();
4591 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4593 if (tx_queue_id < dev->data->nb_tx_queues) {
4594 txq = dev->data->tx_queues[tx_queue_id];
4595 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4596 txdctl |= IXGBE_TXDCTL_ENABLE;
4597 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4599 /* Wait until TX Enable ready */
4600 if (hw->mac.type == ixgbe_mac_82599EB) {
4601 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4604 txdctl = IXGBE_READ_REG(hw,
4605 IXGBE_TXDCTL(txq->reg_idx));
4606 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4608 PMD_INIT_LOG(ERR, "Could not enable "
4609 "Tx Queue %d", tx_queue_id);
4612 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
4613 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
4621 * Stop Transmit Units for specified queue.
4623 int __attribute__((cold))
4624 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
4626 struct ixgbe_hw *hw;
4627 struct ixgbe_tx_queue *txq;
4629 uint32_t txtdh, txtdt;
4632 PMD_INIT_FUNC_TRACE();
4633 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4635 if (tx_queue_id < dev->data->nb_tx_queues) {
4636 txq = dev->data->tx_queues[tx_queue_id];
4638 /* Wait until TX queue is empty */
4639 if (hw->mac.type == ixgbe_mac_82599EB) {
4640 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4642 rte_delay_us(RTE_IXGBE_WAIT_100_US);
4643 txtdh = IXGBE_READ_REG(hw,
4644 IXGBE_TDH(txq->reg_idx));
4645 txtdt = IXGBE_READ_REG(hw,
4646 IXGBE_TDT(txq->reg_idx));
4647 } while (--poll_ms && (txtdh != txtdt));
4649 PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
4650 "when stopping.", tx_queue_id);
4653 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
4654 txdctl &= ~IXGBE_TXDCTL_ENABLE;
4655 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
4657 /* Wait until TX Enable ready */
4658 if (hw->mac.type == ixgbe_mac_82599EB) {
4659 poll_ms = RTE_IXGBE_REGISTER_POLL_WAIT_10_MS;
4662 txdctl = IXGBE_READ_REG(hw,
4663 IXGBE_TXDCTL(txq->reg_idx));
4664 } while (--poll_ms && (txdctl | IXGBE_TXDCTL_ENABLE));
4666 PMD_INIT_LOG(ERR, "Could not disable "
4667 "Tx Queue %d", tx_queue_id);
4670 if (txq->ops != NULL) {
4671 txq->ops->release_mbufs(txq);
4672 txq->ops->reset(txq);
4681 * [VF] Initializes Receive Unit.
4683 int __attribute__((cold))
4684 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
4686 struct ixgbe_hw *hw;
4687 struct ixgbe_rx_queue *rxq;
4689 uint32_t srrctl, psrtype = 0;
4694 PMD_INIT_FUNC_TRACE();
4695 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4697 if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
4698 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
4699 "it should be power of 2");
4703 if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
4704 PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
4705 "it should be equal to or less than %d",
4706 hw->mac.max_rx_queues);
4711 * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
4712 * disables the VF receipt of packets if the PF MTU is > 1500.
4713 * This is done to deal with 82599 limitations that imposes
4714 * the PF and all VFs to share the same MTU.
4715 * Then, the PF driver enables again the VF receipt of packet when
4716 * the VF driver issues a IXGBE_VF_SET_LPE request.
4717 * In the meantime, the VF device cannot be used, even if the VF driver
4718 * and the Guest VM network stack are ready to accept packets with a
4719 * size up to the PF MTU.
4720 * As a work-around to this PF behaviour, force the call to
4721 * ixgbevf_rlpml_set_vf even if jumbo frames are not used. This way,
4722 * VF packets received can work in all cases.
4724 ixgbevf_rlpml_set_vf(hw,
4725 (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len);
4727 /* Setup RX queues */
4728 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4729 rxq = dev->data->rx_queues[i];
4731 /* Allocate buffers for descriptor rings */
4732 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
4736 /* Setup the Base and Length of the Rx Descriptor Rings */
4737 bus_addr = rxq->rx_ring_phys_addr;
4739 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
4740 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4741 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
4742 (uint32_t)(bus_addr >> 32));
4743 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
4744 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
4745 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
4746 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
4749 /* Configure the SRRCTL register */
4750 #ifdef RTE_HEADER_SPLIT_ENABLE
4752 * Configure Header Split
4754 if (dev->data->dev_conf.rxmode.header_split) {
4755 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
4756 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
4757 IXGBE_SRRCTL_BSIZEHDR_MASK);
4758 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
4761 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
4763 /* Set if packets are dropped when no descriptors available */
4765 srrctl |= IXGBE_SRRCTL_DROP_EN;
4768 * Configure the RX buffer size in the BSIZEPACKET field of
4769 * the SRRCTL register of the queue.
4770 * The value is in 1 KB resolution. Valid values can be from
4773 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
4774 RTE_PKTMBUF_HEADROOM);
4775 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
4776 IXGBE_SRRCTL_BSIZEPKT_MASK);
4779 * VF modification to write virtual function SRRCTL register
4781 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
4783 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
4784 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
4786 if (dev->data->dev_conf.rxmode.enable_scatter ||
4787 /* It adds dual VLAN length for supporting dual VLAN */
4788 (dev->data->dev_conf.rxmode.max_rx_pkt_len +
4789 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
4790 if (!dev->data->scattered_rx)
4791 PMD_INIT_LOG(DEBUG, "forcing scatter mode");
4792 dev->data->scattered_rx = 1;
4796 #ifdef RTE_HEADER_SPLIT_ENABLE
4797 if (dev->data->dev_conf.rxmode.header_split)
4798 /* Must setup the PSRTYPE register */
4799 psrtype = IXGBE_PSRTYPE_TCPHDR |
4800 IXGBE_PSRTYPE_UDPHDR |
4801 IXGBE_PSRTYPE_IPV4HDR |
4802 IXGBE_PSRTYPE_IPV6HDR;
4805 /* Set RQPL for VF RSS according to max Rx queue */
4806 psrtype |= (dev->data->nb_rx_queues >> 1) <<
4807 IXGBE_PSRTYPE_RQPL_SHIFT;
4808 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
4810 ixgbe_set_rx_function(dev);
4816 * [VF] Initializes Transmit Unit.
4818 void __attribute__((cold))
4819 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
4821 struct ixgbe_hw *hw;
4822 struct ixgbe_tx_queue *txq;
4827 PMD_INIT_FUNC_TRACE();
4828 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4830 /* Setup the Base and Length of the Tx Descriptor Rings */
4831 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4832 txq = dev->data->tx_queues[i];
4833 bus_addr = txq->tx_ring_phys_addr;
4834 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
4835 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
4836 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
4837 (uint32_t)(bus_addr >> 32));
4838 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
4839 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
4840 /* Setup the HW Tx Head and TX Tail descriptor pointers */
4841 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
4842 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
4845 * Disable Tx Head Writeback RO bit, since this hoses
4846 * bookkeeping if things aren't delivered in order.
4848 txctrl = IXGBE_READ_REG(hw,
4849 IXGBE_VFDCA_TXCTRL(i));
4850 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
4851 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
4857 * [VF] Start Transmit and Receive Units.
4859 void __attribute__((cold))
4860 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
4862 struct ixgbe_hw *hw;
4863 struct ixgbe_tx_queue *txq;
4864 struct ixgbe_rx_queue *rxq;
4870 PMD_INIT_FUNC_TRACE();
4871 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4873 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4874 txq = dev->data->tx_queues[i];
4875 /* Setup Transmit Threshold Registers */
4876 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4877 txdctl |= txq->pthresh & 0x7F;
4878 txdctl |= ((txq->hthresh & 0x7F) << 8);
4879 txdctl |= ((txq->wthresh & 0x7F) << 16);
4880 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
4883 for (i = 0; i < dev->data->nb_tx_queues; i++) {
4885 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4886 txdctl |= IXGBE_TXDCTL_ENABLE;
4887 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
4890 /* Wait until TX Enable ready */
4893 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
4894 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
4896 PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
4898 for (i = 0; i < dev->data->nb_rx_queues; i++) {
4900 rxq = dev->data->rx_queues[i];
4902 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
4903 rxdctl |= IXGBE_RXDCTL_ENABLE;
4904 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
4906 /* Wait until RX Enable ready */
4910 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
4911 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4913 PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
4915 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);
4920 /* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
4921 int __attribute__((weak))
4922 ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
4927 uint16_t __attribute__((weak))
4928 ixgbe_recv_pkts_vec(
4929 void __rte_unused *rx_queue,
4930 struct rte_mbuf __rte_unused **rx_pkts,
4931 uint16_t __rte_unused nb_pkts)
4936 uint16_t __attribute__((weak))
4937 ixgbe_recv_scattered_pkts_vec(
4938 void __rte_unused *rx_queue,
4939 struct rte_mbuf __rte_unused **rx_pkts,
4940 uint16_t __rte_unused nb_pkts)
4945 int __attribute__((weak))
4946 ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)