4 * Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <sys/queue.h>
47 #include <rte_byteorder.h>
48 #include <rte_common.h>
49 #include <rte_cycles.h>
51 #include <rte_debug.h>
52 #include <rte_interrupts.h>
54 #include <rte_memory.h>
55 #include <rte_memzone.h>
56 #include <rte_launch.h>
57 #include <rte_tailq.h>
59 #include <rte_per_lcore.h>
60 #include <rte_lcore.h>
61 #include <rte_atomic.h>
62 #include <rte_branch_prediction.h>
64 #include <rte_mempool.h>
65 #include <rte_malloc.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_prefetch.h>
73 #include <rte_string_fns.h>
74 #include <rte_errno.h>
76 #include "ixgbe_logs.h"
77 #include "ixgbe/ixgbe_api.h"
78 #include "ixgbe/ixgbe_vf.h"
79 #include "ixgbe_ethdev.h"
80 #include "ixgbe/ixgbe_dcb.h"
83 #define RTE_PMD_IXGBE_TX_MAX_BURST 32
85 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
86 #define RTE_PMD_IXGBE_RX_MAX_BURST 32
89 static inline struct rte_mbuf *
90 rte_rxmbuf_alloc(struct rte_mempool *mp)
94 m = __rte_mbuf_raw_alloc(mp);
95 __rte_mbuf_sanity_check_raw(m, RTE_MBUF_PKT, 0);
99 #define RTE_MBUF_DATA_DMA_ADDR(mb) \
100 (uint64_t) ((mb)->buf_physaddr + (uint64_t)((char *)((mb)->pkt.data) - \
101 (char *)(mb)->buf_addr))
103 #define RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb) \
104 (uint64_t) ((mb)->buf_physaddr + RTE_PKTMBUF_HEADROOM)
107 * Structure associated with each descriptor of the RX ring of a RX queue.
109 struct igb_rx_entry {
110 struct rte_mbuf *mbuf; /**< mbuf associated with RX descriptor. */
114 * Structure associated with each descriptor of the TX ring of a TX queue.
116 struct igb_tx_entry {
117 struct rte_mbuf *mbuf; /**< mbuf associated with TX desc, if any. */
118 uint16_t next_id; /**< Index of next descriptor in ring. */
119 uint16_t last_id; /**< Index of last scattered descriptor. */
123 * Structure associated with each RX queue.
125 struct igb_rx_queue {
126 struct rte_mempool *mb_pool; /**< mbuf pool to populate RX ring. */
127 volatile union ixgbe_adv_rx_desc *rx_ring; /**< RX ring virtual address. */
128 uint64_t rx_ring_phys_addr; /**< RX ring DMA address. */
129 volatile uint32_t *rdt_reg_addr; /**< RDT register address. */
130 volatile uint32_t *rdh_reg_addr; /**< RDH register address. */
131 struct igb_rx_entry *sw_ring; /**< address of RX software ring. */
132 struct rte_mbuf *pkt_first_seg; /**< First segment of current packet. */
133 struct rte_mbuf *pkt_last_seg; /**< Last segment of current packet. */
134 uint16_t nb_rx_desc; /**< number of RX descriptors. */
135 uint16_t rx_tail; /**< current value of RDT register. */
136 uint16_t nb_rx_hold; /**< number of held free RX desc. */
137 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
138 uint16_t rx_nb_avail; /**< nr of staged pkts ready to ret to app */
139 uint16_t rx_next_avail; /**< idx of next staged pkt to ret to app */
140 uint16_t rx_free_trigger; /**< triggers rx buffer allocation */
142 uint16_t rx_free_thresh; /**< max free RX desc to hold. */
143 uint16_t queue_id; /**< RX queue index. */
144 uint16_t reg_idx; /**< RX queue register index. */
145 uint8_t port_id; /**< Device port identifier. */
146 uint8_t crc_len; /**< 0 if CRC stripped, 4 otherwise. */
147 uint8_t drop_en; /**< If not 0, set SRRCTL.Drop_En. */
148 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
149 /** need to alloc dummy mbuf, for wraparound when scanning hw ring */
150 struct rte_mbuf fake_mbuf;
151 /** hold packets to return to application */
152 struct rte_mbuf *rx_stage[RTE_PMD_IXGBE_RX_MAX_BURST*2];
157 * IXGBE CTX Constants
159 enum ixgbe_advctx_num {
160 IXGBE_CTX_0 = 0, /**< CTX0 */
161 IXGBE_CTX_1 = 1, /**< CTX1 */
162 IXGBE_CTX_NUM = 2, /**< CTX NUMBER */
166 * Structure to check if new context need be built
169 struct ixgbe_advctx_info {
170 uint16_t flags; /**< ol_flags for context build. */
171 uint32_t cmp_mask; /**< compare mask for vlan_macip_lens */
172 union rte_vlan_macip vlan_macip_lens; /**< vlan, mac ip length. */
176 * Structure associated with each TX queue.
178 struct igb_tx_queue {
179 /** TX ring virtual address. */
180 volatile union ixgbe_adv_tx_desc *tx_ring;
181 uint64_t tx_ring_phys_addr; /**< TX ring DMA address. */
182 struct igb_tx_entry *sw_ring; /**< virtual address of SW ring. */
183 volatile uint32_t *tdt_reg_addr; /**< Address of TDT register. */
184 uint16_t nb_tx_desc; /**< number of TX descriptors. */
185 uint16_t tx_tail; /**< current value of TDT reg. */
186 uint16_t tx_free_thresh;/**< minimum TX before freeing. */
187 /** Number of TX descriptors to use before RS bit is set. */
188 uint16_t tx_rs_thresh;
189 /** Number of TX descriptors used since RS bit was set. */
191 /** Index to last TX descriptor to have been cleaned. */
192 uint16_t last_desc_cleaned;
193 /** Total number of TX descriptors ready to be allocated. */
195 uint16_t tx_next_dd; /**< next desc to scan for DD bit */
196 uint16_t tx_next_rs; /**< next desc to set RS bit */
197 uint16_t queue_id; /**< TX queue index. */
198 uint16_t reg_idx; /**< TX queue register index. */
199 uint8_t port_id; /**< Device port identifier. */
200 uint8_t pthresh; /**< Prefetch threshold register. */
201 uint8_t hthresh; /**< Host threshold register. */
202 uint8_t wthresh; /**< Write-back threshold reg. */
203 uint32_t txq_flags; /**< Holds flags for this TXq */
204 uint32_t ctx_curr; /**< Hardware context states. */
205 /** Hardware context0 history. */
206 struct ixgbe_advctx_info ctx_cache[IXGBE_CTX_NUM];
211 #define RTE_PMD_USE_PREFETCH
214 #ifdef RTE_PMD_USE_PREFETCH
216 * Prefetch a cache line into all cache levels.
218 #define rte_ixgbe_prefetch(p) rte_prefetch0(p)
220 #define rte_ixgbe_prefetch(p) do {} while(0)
223 #ifdef RTE_PMD_PACKET_PREFETCH
224 #define rte_packet_prefetch(p) rte_prefetch1(p)
226 #define rte_packet_prefetch(p) do {} while(0)
229 /*********************************************************************
233 **********************************************************************/
236 * The "simple" TX queue functions require that the following
237 * flags are set when the TX queue is configured:
238 * - ETH_TXQ_FLAGS_NOMULTSEGS
239 * - ETH_TXQ_FLAGS_NOVLANOFFL
240 * - ETH_TXQ_FLAGS_NOXSUMSCTP
241 * - ETH_TXQ_FLAGS_NOXSUMUDP
242 * - ETH_TXQ_FLAGS_NOXSUMTCP
243 * and that the RS bit threshold (tx_rs_thresh) is at least equal to
244 * RTE_PMD_IXGBE_TX_MAX_BURST.
246 #define IXGBE_SIMPLE_FLAGS ((uint32_t)ETH_TXQ_FLAGS_NOMULTSEGS | \
247 ETH_TXQ_FLAGS_NOOFFLOADS)
250 * Check for descriptors with their DD bit set and free mbufs.
251 * Return the total number of buffers freed.
253 static inline int __attribute__((always_inline))
254 ixgbe_tx_free_bufs(struct igb_tx_queue *txq)
256 struct igb_tx_entry *txep;
260 /* check DD bit on threshold descriptor */
261 status = txq->tx_ring[txq->tx_next_dd].wb.status;
262 if (! (status & IXGBE_ADVTXD_STAT_DD))
266 * first buffer to free from S/W ring is at index
267 * tx_next_dd - (tx_rs_thresh-1)
269 txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
271 /* prefetch the mbufs that are about to be freed */
272 for (i = 0; i < txq->tx_rs_thresh; ++i)
273 rte_prefetch0((txep + i)->mbuf);
275 /* free buffers one at a time */
276 if ((txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT) != 0) {
277 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
278 rte_mempool_put(txep->mbuf->pool, txep->mbuf);
282 for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
283 rte_pktmbuf_free_seg(txep->mbuf);
288 /* buffers were freed, update counters */
289 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
290 txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
291 if (txq->tx_next_dd >= txq->nb_tx_desc)
292 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
294 return txq->tx_rs_thresh;
298 * Populate descriptors with the following info:
299 * 1.) buffer_addr = phys_addr + headroom
300 * 2.) cmd_type_len = DCMD_DTYP_FLAGS | pkt_len
301 * 3.) olinfo_status = pkt_len << PAYLEN_SHIFT
304 /* Defines for Tx descriptor */
305 #define DCMD_DTYP_FLAGS (IXGBE_ADVTXD_DTYP_DATA |\
306 IXGBE_ADVTXD_DCMD_IFCS |\
307 IXGBE_ADVTXD_DCMD_DEXT |\
308 IXGBE_ADVTXD_DCMD_EOP)
310 /* Populate 4 descriptors with data from 4 mbufs */
312 tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
314 uint64_t buf_dma_addr;
318 for (i = 0; i < 4; ++i, ++txdp, ++pkts) {
319 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
320 pkt_len = (*pkts)->pkt.data_len;
322 /* write data to descriptor */
323 txdp->read.buffer_addr = buf_dma_addr;
324 txdp->read.cmd_type_len =
325 ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
326 txdp->read.olinfo_status =
327 (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
331 /* Populate 1 descriptor with data from 1 mbuf */
333 tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
335 uint64_t buf_dma_addr;
338 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(*pkts);
339 pkt_len = (*pkts)->pkt.data_len;
341 /* write data to descriptor */
342 txdp->read.buffer_addr = buf_dma_addr;
343 txdp->read.cmd_type_len =
344 ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
345 txdp->read.olinfo_status =
346 (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
350 * Fill H/W descriptor ring with mbuf data.
351 * Copy mbuf pointers to the S/W ring.
354 ixgbe_tx_fill_hw_ring(struct igb_tx_queue *txq, struct rte_mbuf **pkts,
357 volatile union ixgbe_adv_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
358 struct igb_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
359 const int N_PER_LOOP = 4;
360 const int N_PER_LOOP_MASK = N_PER_LOOP-1;
361 int mainpart, leftover;
365 * Process most of the packets in chunks of N pkts. Any
366 * leftover packets will get processed one at a time.
368 mainpart = (nb_pkts & ((uint32_t) ~N_PER_LOOP_MASK));
369 leftover = (nb_pkts & ((uint32_t) N_PER_LOOP_MASK));
370 for (i = 0; i < mainpart; i += N_PER_LOOP) {
371 /* Copy N mbuf pointers to the S/W ring */
372 for (j = 0; j < N_PER_LOOP; ++j) {
373 (txep + i + j)->mbuf = *(pkts + i + j);
375 tx4(txdp + i, pkts + i);
378 if (unlikely(leftover > 0)) {
379 for (i = 0; i < leftover; ++i) {
380 (txep + mainpart + i)->mbuf = *(pkts + mainpart + i);
381 tx1(txdp + mainpart + i, pkts + mainpart + i);
386 static inline uint16_t
387 tx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
390 struct igb_tx_queue *txq = (struct igb_tx_queue *)tx_queue;
391 volatile union ixgbe_adv_tx_desc *tx_r = txq->tx_ring;
395 * Begin scanning the H/W ring for done descriptors when the
396 * number of available descriptors drops below tx_free_thresh. For
397 * each done descriptor, free the associated buffer.
399 if (txq->nb_tx_free < txq->tx_free_thresh)
400 ixgbe_tx_free_bufs(txq);
402 /* Only use descriptors that are available */
403 nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
404 if (unlikely(nb_pkts == 0))
407 /* Use exactly nb_pkts descriptors */
408 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
411 * At this point, we know there are enough descriptors in the
412 * ring to transmit all the packets. This assumes that each
413 * mbuf contains a single segment, and that no new offloads
414 * are expected, which would require a new context descriptor.
418 * See if we're going to wrap-around. If so, handle the top
419 * of the descriptor ring first, then do the bottom. If not,
420 * the processing looks just like the "bottom" part anyway...
422 if ((txq->tx_tail + nb_pkts) > txq->nb_tx_desc) {
423 n = (uint16_t)(txq->nb_tx_desc - txq->tx_tail);
424 ixgbe_tx_fill_hw_ring(txq, tx_pkts, n);
427 * We know that the last descriptor in the ring will need to
428 * have its RS bit set because tx_rs_thresh has to be
429 * a divisor of the ring size
431 tx_r[txq->tx_next_rs].read.cmd_type_len |=
432 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
433 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
438 /* Fill H/W descriptor ring with mbuf data */
439 ixgbe_tx_fill_hw_ring(txq, tx_pkts + n, (uint16_t)(nb_pkts - n));
440 txq->tx_tail = (uint16_t)(txq->tx_tail + (nb_pkts - n));
443 * Determine if RS bit should be set
444 * This is what we actually want:
445 * if ((txq->tx_tail - 1) >= txq->tx_next_rs)
446 * but instead of subtracting 1 and doing >=, we can just do
447 * greater than without subtracting.
449 if (txq->tx_tail > txq->tx_next_rs) {
450 tx_r[txq->tx_next_rs].read.cmd_type_len |=
451 rte_cpu_to_le_32(IXGBE_ADVTXD_DCMD_RS);
452 txq->tx_next_rs = (uint16_t)(txq->tx_next_rs +
454 if (txq->tx_next_rs >= txq->nb_tx_desc)
455 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
459 * Check for wrap-around. This would only happen if we used
460 * up to the last descriptor in the ring, no more, no less.
462 if (txq->tx_tail >= txq->nb_tx_desc)
465 /* update tail pointer */
467 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, txq->tx_tail);
473 ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
478 /* Try to transmit at least chunks of TX_MAX_BURST pkts */
479 if (likely(nb_pkts <= RTE_PMD_IXGBE_TX_MAX_BURST))
480 return tx_xmit_pkts(tx_queue, tx_pkts, nb_pkts);
482 /* transmit more than the max burst, in chunks of TX_MAX_BURST */
486 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_TX_MAX_BURST);
487 ret = tx_xmit_pkts(tx_queue, &(tx_pkts[nb_tx]), n);
488 nb_tx = (uint16_t)(nb_tx + ret);
489 nb_pkts = (uint16_t)(nb_pkts - ret);
498 ixgbe_set_xmit_ctx(struct igb_tx_queue* txq,
499 volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
500 uint16_t ol_flags, uint32_t vlan_macip_lens)
502 uint32_t type_tucmd_mlhl;
503 uint32_t mss_l4len_idx;
507 ctx_idx = txq->ctx_curr;
511 if (ol_flags & PKT_TX_VLAN_PKT) {
512 cmp_mask |= TX_VLAN_CMP_MASK;
515 if (ol_flags & PKT_TX_IP_CKSUM) {
516 type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
517 cmp_mask |= TX_MAC_LEN_CMP_MASK;
520 /* Specify which HW CTX to upload. */
521 mss_l4len_idx = (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
522 switch (ol_flags & PKT_TX_L4_MASK) {
523 case PKT_TX_UDP_CKSUM:
524 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
525 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
526 mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
527 cmp_mask |= TX_MACIP_LEN_CMP_MASK;
529 case PKT_TX_TCP_CKSUM:
530 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
531 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
532 mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
533 cmp_mask |= TX_MACIP_LEN_CMP_MASK;
535 case PKT_TX_SCTP_CKSUM:
536 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
537 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
538 mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
539 cmp_mask |= TX_MACIP_LEN_CMP_MASK;
542 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
543 IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
547 txq->ctx_cache[ctx_idx].flags = ol_flags;
548 txq->ctx_cache[ctx_idx].cmp_mask = cmp_mask;
549 txq->ctx_cache[ctx_idx].vlan_macip_lens.data =
550 vlan_macip_lens & cmp_mask;
552 ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
553 ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
554 ctx_txd->mss_l4len_idx = rte_cpu_to_le_32(mss_l4len_idx);
555 ctx_txd->seqnum_seed = 0;
559 * Check which hardware context can be used. Use the existing match
560 * or create a new context descriptor.
562 static inline uint32_t
563 what_advctx_update(struct igb_tx_queue *txq, uint16_t flags,
564 uint32_t vlan_macip_lens)
566 /* If match with the current used context */
567 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
568 (txq->ctx_cache[txq->ctx_curr].vlan_macip_lens.data ==
569 (txq->ctx_cache[txq->ctx_curr].cmp_mask & vlan_macip_lens)))) {
570 return txq->ctx_curr;
573 /* What if match with the next context */
575 if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
576 (txq->ctx_cache[txq->ctx_curr].vlan_macip_lens.data ==
577 (txq->ctx_cache[txq->ctx_curr].cmp_mask & vlan_macip_lens)))) {
578 return txq->ctx_curr;
581 /* Mismatch, use the previous context */
582 return (IXGBE_CTX_NUM);
585 static inline uint32_t
586 tx_desc_cksum_flags_to_olinfo(uint16_t ol_flags)
588 static const uint32_t l4_olinfo[2] = {0, IXGBE_ADVTXD_POPTS_TXSM};
589 static const uint32_t l3_olinfo[2] = {0, IXGBE_ADVTXD_POPTS_IXSM};
592 tmp = l4_olinfo[(ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM];
593 tmp |= l3_olinfo[(ol_flags & PKT_TX_IP_CKSUM) != 0];
597 static inline uint32_t
598 tx_desc_vlan_flags_to_cmdtype(uint16_t ol_flags)
600 static const uint32_t vlan_cmd[2] = {0, IXGBE_ADVTXD_DCMD_VLE};
601 return vlan_cmd[(ol_flags & PKT_TX_VLAN_PKT) != 0];
604 /* Default RS bit threshold values */
605 #ifndef DEFAULT_TX_RS_THRESH
606 #define DEFAULT_TX_RS_THRESH 32
608 #ifndef DEFAULT_TX_FREE_THRESH
609 #define DEFAULT_TX_FREE_THRESH 32
612 /* Reset transmit descriptors after they have been used */
614 ixgbe_xmit_cleanup(struct igb_tx_queue *txq)
616 struct igb_tx_entry *sw_ring = txq->sw_ring;
617 volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
618 uint16_t last_desc_cleaned = txq->last_desc_cleaned;
619 uint16_t nb_tx_desc = txq->nb_tx_desc;
620 uint16_t desc_to_clean_to;
621 uint16_t nb_tx_to_clean;
623 /* Determine the last descriptor needing to be cleaned */
624 desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
625 if (desc_to_clean_to >= nb_tx_desc)
626 desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
628 /* Check to make sure the last descriptor to clean is done */
629 desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
630 if (! (txr[desc_to_clean_to].wb.status & IXGBE_TXD_STAT_DD))
632 PMD_TX_FREE_LOG(DEBUG,
633 "TX descriptor %4u is not done"
634 "(port=%d queue=%d)",
636 txq->port_id, txq->queue_id);
637 /* Failed to clean any descriptors, better luck next time */
641 /* Figure out how many descriptors will be cleaned */
642 if (last_desc_cleaned > desc_to_clean_to)
643 nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
646 nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
649 PMD_TX_FREE_LOG(DEBUG,
650 "Cleaning %4u TX descriptors: %4u to %4u "
651 "(port=%d queue=%d)",
652 nb_tx_to_clean, last_desc_cleaned, desc_to_clean_to,
653 txq->port_id, txq->queue_id);
656 * The last descriptor to clean is done, so that means all the
657 * descriptors from the last descriptor that was cleaned
658 * up to the last descriptor with the RS bit set
659 * are done. Only reset the threshold descriptor.
661 txr[desc_to_clean_to].wb.status = 0;
663 /* Update the txq to reflect the last descriptor that was cleaned */
664 txq->last_desc_cleaned = desc_to_clean_to;
665 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
672 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
675 struct igb_tx_queue *txq;
676 struct igb_tx_entry *sw_ring;
677 struct igb_tx_entry *txe, *txn;
678 volatile union ixgbe_adv_tx_desc *txr;
679 volatile union ixgbe_adv_tx_desc *txd;
680 struct rte_mbuf *tx_pkt;
681 struct rte_mbuf *m_seg;
682 uint64_t buf_dma_addr;
683 uint32_t olinfo_status;
684 uint32_t cmd_type_len;
693 uint32_t vlan_macip_lens;
698 sw_ring = txq->sw_ring;
700 tx_id = txq->tx_tail;
701 txe = &sw_ring[tx_id];
703 /* Determine if the descriptor ring needs to be cleaned. */
704 if ((txq->nb_tx_desc - txq->nb_tx_free) > txq->tx_free_thresh) {
705 ixgbe_xmit_cleanup(txq);
709 for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
712 pkt_len = tx_pkt->pkt.pkt_len;
714 RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
717 * Determine how many (if any) context descriptors
718 * are needed for offload functionality.
720 ol_flags = tx_pkt->ol_flags;
721 vlan_macip_lens = tx_pkt->pkt.vlan_macip.data;
723 /* If hardware offload required */
724 tx_ol_req = (uint16_t)(ol_flags & PKT_TX_OFFLOAD_MASK);
726 /* If new context need be built or reuse the exist ctx. */
727 ctx = what_advctx_update(txq, tx_ol_req,
729 /* Only allocate context descriptor if required*/
730 new_ctx = (ctx == IXGBE_CTX_NUM);
735 * Keep track of how many descriptors are used this loop
736 * This will always be the number of segments + the number of
737 * Context descriptors required to transmit the packet
739 nb_used = (uint16_t)(tx_pkt->pkt.nb_segs + new_ctx);
742 * The number of descriptors that must be allocated for a
743 * packet is the number of segments of that packet, plus 1
744 * Context Descriptor for the hardware offload, if any.
745 * Determine the last TX descriptor to allocate in the TX ring
746 * for the packet, starting from the current position (tx_id)
749 tx_last = (uint16_t) (tx_id + nb_used - 1);
752 if (tx_last >= txq->nb_tx_desc)
753 tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
755 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
756 " tx_first=%u tx_last=%u\n",
757 (unsigned) txq->port_id,
758 (unsigned) txq->queue_id,
764 * Make sure there are enough TX descriptors available to
765 * transmit the entire packet.
766 * nb_used better be less than or equal to txq->tx_rs_thresh
768 if (nb_used > txq->nb_tx_free) {
769 PMD_TX_FREE_LOG(DEBUG,
770 "Not enough free TX descriptors "
771 "nb_used=%4u nb_free=%4u "
772 "(port=%d queue=%d)",
773 nb_used, txq->nb_tx_free,
774 txq->port_id, txq->queue_id);
776 if (ixgbe_xmit_cleanup(txq) != 0) {
777 /* Could not clean any descriptors */
783 /* nb_used better be <= txq->tx_rs_thresh */
784 if (unlikely(nb_used > txq->tx_rs_thresh)) {
785 PMD_TX_FREE_LOG(DEBUG,
786 "The number of descriptors needed to "
787 "transmit the packet exceeds the "
788 "RS bit threshold. This will impact "
790 "nb_used=%4u nb_free=%4u "
792 "(port=%d queue=%d)",
793 nb_used, txq->nb_tx_free,
795 txq->port_id, txq->queue_id);
797 * Loop here until there are enough TX
798 * descriptors or until the ring cannot be
801 while (nb_used > txq->nb_tx_free) {
802 if (ixgbe_xmit_cleanup(txq) != 0) {
804 * Could not clean any
816 * By now there are enough free TX descriptors to transmit
821 * Set common flags of all TX Data Descriptors.
823 * The following bits must be set in all Data Descriptors:
824 * - IXGBE_ADVTXD_DTYP_DATA
825 * - IXGBE_ADVTXD_DCMD_DEXT
827 * The following bits must be set in the first Data Descriptor
828 * and are ignored in the other ones:
829 * - IXGBE_ADVTXD_DCMD_IFCS
830 * - IXGBE_ADVTXD_MAC_1588
831 * - IXGBE_ADVTXD_DCMD_VLE
833 * The following bits must only be set in the last Data
835 * - IXGBE_TXD_CMD_EOP
837 * The following bits can be set in any Data Descriptor, but
838 * are only set in the last Data Descriptor:
841 cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
842 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
843 olinfo_status = (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
844 #ifdef RTE_LIBRTE_IEEE1588
845 if (ol_flags & PKT_TX_IEEE1588_TMST)
846 cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
851 * Setup the TX Advanced Context Descriptor if required
854 volatile struct ixgbe_adv_tx_context_desc *
857 ctx_txd = (volatile struct
858 ixgbe_adv_tx_context_desc *)
861 txn = &sw_ring[txe->next_id];
862 RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
864 if (txe->mbuf != NULL) {
865 rte_pktmbuf_free_seg(txe->mbuf);
869 ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
872 txe->last_id = tx_last;
873 tx_id = txe->next_id;
878 * Setup the TX Advanced Data Descriptor,
879 * This path will go through
880 * whatever new/reuse the context descriptor
882 cmd_type_len |= tx_desc_vlan_flags_to_cmdtype(ol_flags);
883 olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
884 olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
890 txn = &sw_ring[txe->next_id];
892 if (txe->mbuf != NULL)
893 rte_pktmbuf_free_seg(txe->mbuf);
897 * Set up Transmit Data Descriptor.
899 slen = m_seg->pkt.data_len;
900 buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
901 txd->read.buffer_addr =
902 rte_cpu_to_le_64(buf_dma_addr);
903 txd->read.cmd_type_len =
904 rte_cpu_to_le_32(cmd_type_len | slen);
905 txd->read.olinfo_status =
906 rte_cpu_to_le_32(olinfo_status);
907 txe->last_id = tx_last;
908 tx_id = txe->next_id;
910 m_seg = m_seg->pkt.next;
911 } while (m_seg != NULL);
914 * The last packet data descriptor needs End Of Packet (EOP)
916 cmd_type_len |= IXGBE_TXD_CMD_EOP;
917 txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
918 txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
920 /* Set RS bit only on threshold packets' last descriptor */
921 if (txq->nb_tx_used >= txq->tx_rs_thresh) {
922 PMD_TX_FREE_LOG(DEBUG,
923 "Setting RS bit on TXD id="
924 "%4u (port=%d queue=%d)",
925 tx_last, txq->port_id, txq->queue_id);
927 cmd_type_len |= IXGBE_TXD_CMD_RS;
929 /* Update txq RS bit counters */
932 txd->read.cmd_type_len |= rte_cpu_to_le_32(cmd_type_len);
938 * Set the Transmit Descriptor Tail (TDT)
940 PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u tx_tail=%u nb_tx=%u",
941 (unsigned) txq->port_id, (unsigned) txq->queue_id,
942 (unsigned) tx_id, (unsigned) nb_tx);
943 IXGBE_PCI_REG_WRITE(txq->tdt_reg_addr, tx_id);
944 txq->tx_tail = tx_id;
949 /*********************************************************************
953 **********************************************************************/
954 static inline uint16_t
955 rx_desc_hlen_type_rss_to_pkt_flags(uint32_t hl_tp_rs)
959 static uint16_t ip_pkt_types_map[16] = {
960 0, PKT_RX_IPV4_HDR, PKT_RX_IPV4_HDR_EXT, PKT_RX_IPV4_HDR_EXT,
961 PKT_RX_IPV6_HDR, 0, 0, 0,
962 PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
963 PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
966 static uint16_t ip_rss_types_map[16] = {
967 0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
968 0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
969 PKT_RX_RSS_HASH, 0, 0, 0,
970 0, 0, 0, PKT_RX_FDIR,
973 #ifdef RTE_LIBRTE_IEEE1588
974 static uint32_t ip_pkt_etqf_map[8] = {
975 0, 0, 0, PKT_RX_IEEE1588_PTP,
979 pkt_flags = (uint16_t) ((hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ?
980 ip_pkt_etqf_map[(hl_tp_rs >> 4) & 0x07] :
981 ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F]);
983 pkt_flags = (uint16_t) ((hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ? 0 :
984 ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F]);
987 return (uint16_t)(pkt_flags | ip_rss_types_map[hl_tp_rs & 0xF]);
990 static inline uint16_t
991 rx_desc_status_to_pkt_flags(uint32_t rx_status)
996 * Check if VLAN present only.
997 * Do not check whether L3/L4 rx checksum done by NIC or not,
998 * That can be found from rte_eth_rxmode.hw_ip_checksum flag
1000 pkt_flags = (uint16_t)((rx_status & IXGBE_RXD_STAT_VP) ?
1001 PKT_RX_VLAN_PKT : 0);
1003 #ifdef RTE_LIBRTE_IEEE1588
1004 if (rx_status & IXGBE_RXD_STAT_TMST)
1005 pkt_flags = (uint16_t)(pkt_flags | PKT_RX_IEEE1588_TMST);
1010 static inline uint16_t
1011 rx_desc_error_to_pkt_flags(uint32_t rx_status)
1014 * Bit 31: IPE, IPv4 checksum error
1015 * Bit 30: L4I, L4I integrity error
1017 static uint16_t error_to_pkt_flags_map[4] = {
1018 0, PKT_RX_L4_CKSUM_BAD, PKT_RX_IP_CKSUM_BAD,
1019 PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
1021 return error_to_pkt_flags_map[(rx_status >>
1022 IXGBE_RXDADV_ERR_CKSUM_BIT) & IXGBE_RXDADV_ERR_CKSUM_MSK];
1025 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
1027 * LOOK_AHEAD defines how many desc statuses to check beyond the
1028 * current descriptor.
1029 * It must be a pound define for optimal performance.
1030 * Do not change the value of LOOK_AHEAD, as the ixgbe_rx_scan_hw_ring
1031 * function only works with LOOK_AHEAD=8.
1033 #define LOOK_AHEAD 8
1034 #if (LOOK_AHEAD != 8)
1035 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
1038 ixgbe_rx_scan_hw_ring(struct igb_rx_queue *rxq)
1040 volatile union ixgbe_adv_rx_desc *rxdp;
1041 struct igb_rx_entry *rxep;
1042 struct rte_mbuf *mb;
1044 int s[LOOK_AHEAD], nb_dd;
1045 int i, j, nb_rx = 0;
1048 /* get references to current descriptor and S/W ring entry */
1049 rxdp = &rxq->rx_ring[rxq->rx_tail];
1050 rxep = &rxq->sw_ring[rxq->rx_tail];
1052 /* check to make sure there is at least 1 packet to receive */
1053 if (! (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD))
1057 * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
1058 * reference packets that are ready to be received.
1060 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST;
1061 i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD)
1063 /* Read desc statuses backwards to avoid race condition */
1064 for (j = LOOK_AHEAD-1; j >= 0; --j)
1065 s[j] = rxdp[j].wb.upper.status_error;
1067 /* Clear everything but the status bits (LSB) */
1068 for (j = 0; j < LOOK_AHEAD; ++j)
1069 s[j] &= IXGBE_RXDADV_STAT_DD;
1071 /* Compute how many status bits were set */
1072 nb_dd = s[0]+s[1]+s[2]+s[3]+s[4]+s[5]+s[6]+s[7];
1075 /* Translate descriptor info to mbuf format */
1076 for (j = 0; j < nb_dd; ++j) {
1078 pkt_len = (uint16_t)(rxdp[j].wb.upper.length -
1080 mb->pkt.data_len = pkt_len;
1081 mb->pkt.pkt_len = pkt_len;
1082 mb->pkt.vlan_macip.f.vlan_tci = rxdp[j].wb.upper.vlan;
1083 mb->pkt.hash.rss = rxdp[j].wb.lower.hi_dword.rss;
1085 /* convert descriptor fields to rte mbuf flags */
1086 mb->ol_flags = rx_desc_hlen_type_rss_to_pkt_flags(
1087 rxdp[j].wb.lower.lo_dword.data);
1088 /* reuse status field from scan list */
1089 mb->ol_flags = (uint16_t)(mb->ol_flags |
1090 rx_desc_status_to_pkt_flags(s[j]));
1091 mb->ol_flags = (uint16_t)(mb->ol_flags |
1092 rx_desc_error_to_pkt_flags(s[j]));
1095 /* Move mbuf pointers from the S/W ring to the stage */
1096 for (j = 0; j < LOOK_AHEAD; ++j) {
1097 rxq->rx_stage[i + j] = rxep[j].mbuf;
1100 /* stop if all requested packets could not be received */
1101 if (nb_dd != LOOK_AHEAD)
1105 /* clear software ring entries so we can cleanup correctly */
1106 for (i = 0; i < nb_rx; ++i) {
1107 rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
1115 ixgbe_rx_alloc_bufs(struct igb_rx_queue *rxq)
1117 volatile union ixgbe_adv_rx_desc *rxdp;
1118 struct igb_rx_entry *rxep;
1119 struct rte_mbuf *mb;
1124 /* allocate buffers in bulk directly into the S/W ring */
1125 alloc_idx = (uint16_t)(rxq->rx_free_trigger -
1126 (rxq->rx_free_thresh - 1));
1127 rxep = &rxq->sw_ring[alloc_idx];
1128 diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
1129 rxq->rx_free_thresh);
1130 if (unlikely(diag != 0))
1133 rxdp = &rxq->rx_ring[alloc_idx];
1134 for (i = 0; i < rxq->rx_free_thresh; ++i) {
1135 /* populate the static rte mbuf fields */
1137 rte_mbuf_refcnt_set(mb, 1);
1138 mb->type = RTE_MBUF_PKT;
1139 mb->pkt.next = NULL;
1140 mb->pkt.data = (char *)mb->buf_addr + RTE_PKTMBUF_HEADROOM;
1141 mb->pkt.nb_segs = 1;
1142 mb->pkt.in_port = rxq->port_id;
1144 /* populate the descriptors */
1145 dma_addr = (uint64_t)mb->buf_physaddr + RTE_PKTMBUF_HEADROOM;
1146 rxdp[i].read.hdr_addr = dma_addr;
1147 rxdp[i].read.pkt_addr = dma_addr;
1150 /* update tail pointer */
1152 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rxq->rx_free_trigger);
1154 /* update state of internal queue structure */
1155 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_trigger +
1156 rxq->rx_free_thresh);
1157 if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
1158 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
1164 static inline uint16_t
1165 ixgbe_rx_fill_from_stage(struct igb_rx_queue *rxq, struct rte_mbuf **rx_pkts,
1168 struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
1171 /* how many packets are ready to return? */
1172 nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
1174 /* copy mbuf pointers to the application's packet list */
1175 for (i = 0; i < nb_pkts; ++i)
1176 rx_pkts[i] = stage[i];
1178 /* update internal queue state */
1179 rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
1180 rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
1185 static inline uint16_t
1186 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1189 struct igb_rx_queue *rxq = (struct igb_rx_queue *)rx_queue;
1192 /* Any previously recv'd pkts will be returned from the Rx stage */
1193 if (rxq->rx_nb_avail)
1194 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1196 /* Scan the H/W ring for packets to receive */
1197 nb_rx = (uint16_t)ixgbe_rx_scan_hw_ring(rxq);
1199 /* update internal queue state */
1200 rxq->rx_next_avail = 0;
1201 rxq->rx_nb_avail = nb_rx;
1202 rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
1204 /* if required, allocate new buffers to replenish descriptors */
1205 if (rxq->rx_tail > rxq->rx_free_trigger) {
1206 if (ixgbe_rx_alloc_bufs(rxq) != 0) {
1208 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1209 "queue_id=%u\n", (unsigned) rxq->port_id,
1210 (unsigned) rxq->queue_id);
1212 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
1213 rxq->rx_free_thresh;
1216 * Need to rewind any previous receives if we cannot
1217 * allocate new buffers to replenish the old ones.
1219 rxq->rx_nb_avail = 0;
1220 rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
1221 for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
1222 rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
1228 if (rxq->rx_tail >= rxq->nb_rx_desc)
1231 /* received any packets this loop? */
1232 if (rxq->rx_nb_avail)
1233 return ixgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
1238 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
1240 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
1245 if (unlikely(nb_pkts == 0))
1248 if (likely(nb_pkts <= RTE_PMD_IXGBE_RX_MAX_BURST))
1249 return rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
1251 /* request is relatively large, chunk it up */
1255 n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_IXGBE_RX_MAX_BURST);
1256 ret = rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
1257 nb_rx = (uint16_t)(nb_rx + ret);
1258 nb_pkts = (uint16_t)(nb_pkts - ret);
1265 #endif /* RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC */
1268 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1271 struct igb_rx_queue *rxq;
1272 volatile union ixgbe_adv_rx_desc *rx_ring;
1273 volatile union ixgbe_adv_rx_desc *rxdp;
1274 struct igb_rx_entry *sw_ring;
1275 struct igb_rx_entry *rxe;
1276 struct rte_mbuf *rxm;
1277 struct rte_mbuf *nmb;
1278 union ixgbe_adv_rx_desc rxd;
1281 uint32_t hlen_type_rss;
1291 rx_id = rxq->rx_tail;
1292 rx_ring = rxq->rx_ring;
1293 sw_ring = rxq->sw_ring;
1294 while (nb_rx < nb_pkts) {
1296 * The order of operations here is important as the DD status
1297 * bit must not be read after any other descriptor fields.
1298 * rx_ring and rxdp are pointing to volatile data so the order
1299 * of accesses cannot be reordered by the compiler. If they were
1300 * not volatile, they could be reordered which could lead to
1301 * using invalid descriptor fields when read from rxd.
1303 rxdp = &rx_ring[rx_id];
1304 staterr = rxdp->wb.upper.status_error;
1305 if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1312 * If the IXGBE_RXDADV_STAT_EOP flag is not set, the RX packet
1313 * is likely to be invalid and to be dropped by the various
1314 * validation checks performed by the network stack.
1316 * Allocate a new mbuf to replenish the RX ring descriptor.
1317 * If the allocation fails:
1318 * - arrange for that RX descriptor to be the first one
1319 * being parsed the next time the receive function is
1320 * invoked [on the same queue].
1322 * - Stop parsing the RX ring and return immediately.
1324 * This policy do not drop the packet received in the RX
1325 * descriptor for which the allocation of a new mbuf failed.
1326 * Thus, it allows that packet to be later retrieved if
1327 * mbuf have been freed in the mean time.
1328 * As a side effect, holding RX descriptors instead of
1329 * systematically giving them back to the NIC may lead to
1330 * RX ring exhaustion situations.
1331 * However, the NIC can gracefully prevent such situations
1332 * to happen by sending specific "back-pressure" flow control
1333 * frames to its peer(s).
1335 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
1336 "ext_err_stat=0x%08x pkt_len=%u\n",
1337 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1338 (unsigned) rx_id, (unsigned) staterr,
1339 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1341 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1343 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1344 "queue_id=%u\n", (unsigned) rxq->port_id,
1345 (unsigned) rxq->queue_id);
1346 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1351 rxe = &sw_ring[rx_id];
1353 if (rx_id == rxq->nb_rx_desc)
1356 /* Prefetch next mbuf while processing current one. */
1357 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1360 * When next RX descriptor is on a cache-line boundary,
1361 * prefetch the next 4 RX descriptors and the next 8 pointers
1364 if ((rx_id & 0x3) == 0) {
1365 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1366 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1372 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1373 rxdp->read.hdr_addr = dma_addr;
1374 rxdp->read.pkt_addr = dma_addr;
1377 * Initialize the returned mbuf.
1378 * 1) setup generic mbuf fields:
1379 * - number of segments,
1382 * - RX port identifier.
1383 * 2) integrate hardware offload data, if any:
1384 * - RSS flag & hash,
1385 * - IP checksum flag,
1386 * - VLAN TCI, if any,
1389 pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
1391 rxm->pkt.data = (char*) rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
1392 rte_packet_prefetch(rxm->pkt.data);
1393 rxm->pkt.nb_segs = 1;
1394 rxm->pkt.next = NULL;
1395 rxm->pkt.pkt_len = pkt_len;
1396 rxm->pkt.data_len = pkt_len;
1397 rxm->pkt.in_port = rxq->port_id;
1399 hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1400 /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
1401 rxm->pkt.vlan_macip.f.vlan_tci =
1402 rte_le_to_cpu_16(rxd.wb.upper.vlan);
1404 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1405 pkt_flags = (uint16_t)(pkt_flags |
1406 rx_desc_status_to_pkt_flags(staterr));
1407 pkt_flags = (uint16_t)(pkt_flags |
1408 rx_desc_error_to_pkt_flags(staterr));
1409 rxm->ol_flags = pkt_flags;
1411 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1412 rxm->pkt.hash.rss = rxd.wb.lower.hi_dword.rss;
1413 else if (pkt_flags & PKT_RX_FDIR) {
1414 rxm->pkt.hash.fdir.hash =
1415 (uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
1416 & IXGBE_ATR_HASH_MASK);
1417 rxm->pkt.hash.fdir.id = rxd.wb.lower.hi_dword.csum_ip.ip_id;
1420 * Store the mbuf address into the next entry of the array
1421 * of returned packets.
1423 rx_pkts[nb_rx++] = rxm;
1425 rxq->rx_tail = rx_id;
1428 * If the number of free RX descriptors is greater than the RX free
1429 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1431 * Update the RDT with the value of the last processed RX descriptor
1432 * minus 1, to guarantee that the RDT register is never equal to the
1433 * RDH register, which creates a "full" ring situtation from the
1434 * hardware point of view...
1436 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1437 if (nb_hold > rxq->rx_free_thresh) {
1438 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1439 "nb_hold=%u nb_rx=%u\n",
1440 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1441 (unsigned) rx_id, (unsigned) nb_hold,
1443 rx_id = (uint16_t) ((rx_id == 0) ?
1444 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1445 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1448 rxq->nb_rx_hold = nb_hold;
1453 ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1456 struct igb_rx_queue *rxq;
1457 volatile union ixgbe_adv_rx_desc *rx_ring;
1458 volatile union ixgbe_adv_rx_desc *rxdp;
1459 struct igb_rx_entry *sw_ring;
1460 struct igb_rx_entry *rxe;
1461 struct rte_mbuf *first_seg;
1462 struct rte_mbuf *last_seg;
1463 struct rte_mbuf *rxm;
1464 struct rte_mbuf *nmb;
1465 union ixgbe_adv_rx_desc rxd;
1466 uint64_t dma; /* Physical address of mbuf data buffer */
1468 uint32_t hlen_type_rss;
1478 rx_id = rxq->rx_tail;
1479 rx_ring = rxq->rx_ring;
1480 sw_ring = rxq->sw_ring;
1483 * Retrieve RX context of current packet, if any.
1485 first_seg = rxq->pkt_first_seg;
1486 last_seg = rxq->pkt_last_seg;
1488 while (nb_rx < nb_pkts) {
1491 * The order of operations here is important as the DD status
1492 * bit must not be read after any other descriptor fields.
1493 * rx_ring and rxdp are pointing to volatile data so the order
1494 * of accesses cannot be reordered by the compiler. If they were
1495 * not volatile, they could be reordered which could lead to
1496 * using invalid descriptor fields when read from rxd.
1498 rxdp = &rx_ring[rx_id];
1499 staterr = rxdp->wb.upper.status_error;
1500 if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
1507 * Allocate a new mbuf to replenish the RX ring descriptor.
1508 * If the allocation fails:
1509 * - arrange for that RX descriptor to be the first one
1510 * being parsed the next time the receive function is
1511 * invoked [on the same queue].
1513 * - Stop parsing the RX ring and return immediately.
1515 * This policy does not drop the packet received in the RX
1516 * descriptor for which the allocation of a new mbuf failed.
1517 * Thus, it allows that packet to be later retrieved if
1518 * mbuf have been freed in the mean time.
1519 * As a side effect, holding RX descriptors instead of
1520 * systematically giving them back to the NIC may lead to
1521 * RX ring exhaustion situations.
1522 * However, the NIC can gracefully prevent such situations
1523 * to happen by sending specific "back-pressure" flow control
1524 * frames to its peer(s).
1526 PMD_RX_LOG(DEBUG, "\nport_id=%u queue_id=%u rx_id=%u "
1527 "staterr=0x%x data_len=%u\n",
1528 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1529 (unsigned) rx_id, (unsigned) staterr,
1530 (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
1532 nmb = rte_rxmbuf_alloc(rxq->mb_pool);
1534 PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
1535 "queue_id=%u\n", (unsigned) rxq->port_id,
1536 (unsigned) rxq->queue_id);
1537 rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1542 rxe = &sw_ring[rx_id];
1544 if (rx_id == rxq->nb_rx_desc)
1547 /* Prefetch next mbuf while processing current one. */
1548 rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
1551 * When next RX descriptor is on a cache-line boundary,
1552 * prefetch the next 4 RX descriptors and the next 8 pointers
1555 if ((rx_id & 0x3) == 0) {
1556 rte_ixgbe_prefetch(&rx_ring[rx_id]);
1557 rte_ixgbe_prefetch(&sw_ring[rx_id]);
1561 * Update RX descriptor with the physical address of the new
1562 * data buffer of the new allocated mbuf.
1566 dma = rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
1567 rxdp->read.hdr_addr = dma;
1568 rxdp->read.pkt_addr = dma;
1571 * Set data length & data buffer address of mbuf.
1573 data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
1574 rxm->pkt.data_len = data_len;
1575 rxm->pkt.data = (char*) rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
1578 * If this is the first buffer of the received packet,
1579 * set the pointer to the first mbuf of the packet and
1580 * initialize its context.
1581 * Otherwise, update the total length and the number of segments
1582 * of the current scattered packet, and update the pointer to
1583 * the last mbuf of the current packet.
1585 if (first_seg == NULL) {
1587 first_seg->pkt.pkt_len = data_len;
1588 first_seg->pkt.nb_segs = 1;
1590 first_seg->pkt.pkt_len = (uint16_t)(first_seg->pkt.pkt_len
1592 first_seg->pkt.nb_segs++;
1593 last_seg->pkt.next = rxm;
1597 * If this is not the last buffer of the received packet,
1598 * update the pointer to the last mbuf of the current scattered
1599 * packet and continue to parse the RX ring.
1601 if (! (staterr & IXGBE_RXDADV_STAT_EOP)) {
1607 * This is the last buffer of the received packet.
1608 * If the CRC is not stripped by the hardware:
1609 * - Subtract the CRC length from the total packet length.
1610 * - If the last buffer only contains the whole CRC or a part
1611 * of it, free the mbuf associated to the last buffer.
1612 * If part of the CRC is also contained in the previous
1613 * mbuf, subtract the length of that CRC part from the
1614 * data length of the previous mbuf.
1616 rxm->pkt.next = NULL;
1617 if (unlikely(rxq->crc_len > 0)) {
1618 first_seg->pkt.pkt_len -= ETHER_CRC_LEN;
1619 if (data_len <= ETHER_CRC_LEN) {
1620 rte_pktmbuf_free_seg(rxm);
1621 first_seg->pkt.nb_segs--;
1622 last_seg->pkt.data_len = (uint16_t)
1623 (last_seg->pkt.data_len -
1624 (ETHER_CRC_LEN - data_len));
1625 last_seg->pkt.next = NULL;
1628 (uint16_t) (data_len - ETHER_CRC_LEN);
1632 * Initialize the first mbuf of the returned packet:
1633 * - RX port identifier,
1634 * - hardware offload data, if any:
1635 * - RSS flag & hash,
1636 * - IP checksum flag,
1637 * - VLAN TCI, if any,
1640 first_seg->pkt.in_port = rxq->port_id;
1643 * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
1644 * set in the pkt_flags field.
1646 first_seg->pkt.vlan_macip.f.vlan_tci =
1647 rte_le_to_cpu_16(rxd.wb.upper.vlan);
1648 hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
1649 pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
1650 pkt_flags = (uint16_t)(pkt_flags |
1651 rx_desc_status_to_pkt_flags(staterr));
1652 pkt_flags = (uint16_t)(pkt_flags |
1653 rx_desc_error_to_pkt_flags(staterr));
1654 first_seg->ol_flags = pkt_flags;
1656 if (likely(pkt_flags & PKT_RX_RSS_HASH))
1657 first_seg->pkt.hash.rss = rxd.wb.lower.hi_dword.rss;
1658 else if (pkt_flags & PKT_RX_FDIR) {
1659 first_seg->pkt.hash.fdir.hash =
1660 (uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
1661 & IXGBE_ATR_HASH_MASK);
1662 first_seg->pkt.hash.fdir.id =
1663 rxd.wb.lower.hi_dword.csum_ip.ip_id;
1666 /* Prefetch data of first segment, if configured to do so. */
1667 rte_packet_prefetch(first_seg->pkt.data);
1670 * Store the mbuf address into the next entry of the array
1671 * of returned packets.
1673 rx_pkts[nb_rx++] = first_seg;
1676 * Setup receipt context for a new packet.
1682 * Record index of the next RX descriptor to probe.
1684 rxq->rx_tail = rx_id;
1687 * Save receive context.
1689 rxq->pkt_first_seg = first_seg;
1690 rxq->pkt_last_seg = last_seg;
1693 * If the number of free RX descriptors is greater than the RX free
1694 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
1696 * Update the RDT with the value of the last processed RX descriptor
1697 * minus 1, to guarantee that the RDT register is never equal to the
1698 * RDH register, which creates a "full" ring situtation from the
1699 * hardware point of view...
1701 nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
1702 if (nb_hold > rxq->rx_free_thresh) {
1703 PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
1704 "nb_hold=%u nb_rx=%u\n",
1705 (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
1706 (unsigned) rx_id, (unsigned) nb_hold,
1708 rx_id = (uint16_t) ((rx_id == 0) ?
1709 (rxq->nb_rx_desc - 1) : (rx_id - 1));
1710 IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
1713 rxq->nb_rx_hold = nb_hold;
1717 /*********************************************************************
1719 * Queue management functions
1721 **********************************************************************/
1724 * Rings setup and release.
1726 * TDBA/RDBA should be aligned on 16 byte boundary. But TDLEN/RDLEN should be
1727 * multiple of 128 bytes. So we align TDBA/RDBA on 128 byte boundary. This will
1728 * also optimize cache line size effect. H/W supports up to cache line size 128.
1730 #define IXGBE_ALIGN 128
1733 * Maximum number of Ring Descriptors.
1735 * Since RDLEN/TDLEN should be multiple of 128 bytes, the number of ring
1736 * descriptors should meet the following condition:
1737 * (num_ring_desc * sizeof(rx/tx descriptor)) % 128 == 0
1739 #define IXGBE_MIN_RING_DESC 64
1740 #define IXGBE_MAX_RING_DESC 4096
1743 * Create memzone for HW rings. malloc can't be used as the physical address is
1744 * needed. If the memzone is already created, then this function returns a ptr
1747 static const struct rte_memzone *
1748 ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
1749 uint16_t queue_id, uint32_t ring_size, int socket_id)
1751 char z_name[RTE_MEMZONE_NAMESIZE];
1752 const struct rte_memzone *mz;
1754 rte_snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
1755 dev->driver->pci_drv.name, ring_name,
1756 dev->data->port_id, queue_id);
1758 mz = rte_memzone_lookup(z_name);
1762 return rte_memzone_reserve_aligned(z_name, ring_size,
1763 socket_id, 0, IXGBE_ALIGN);
1767 ixgbe_tx_queue_release_mbufs(struct igb_tx_queue *txq)
1771 if (txq->sw_ring != NULL) {
1772 for (i = 0; i < txq->nb_tx_desc; i++) {
1773 if (txq->sw_ring[i].mbuf != NULL) {
1774 rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
1775 txq->sw_ring[i].mbuf = NULL;
1782 ixgbe_tx_queue_release(struct igb_tx_queue *txq)
1785 ixgbe_tx_queue_release_mbufs(txq);
1786 rte_free(txq->sw_ring);
1792 ixgbe_dev_tx_queue_release(void *txq)
1794 ixgbe_tx_queue_release(txq);
1797 /* (Re)set dynamic igb_tx_queue fields to defaults */
1799 ixgbe_reset_tx_queue(struct igb_tx_queue *txq)
1801 struct igb_tx_entry *txe = txq->sw_ring;
1804 /* Zero out HW ring memory */
1805 for (i = 0; i < sizeof(union ixgbe_adv_tx_desc) * txq->nb_tx_desc; i++) {
1806 ((volatile char *)txq->tx_ring)[i] = 0;
1809 /* Initialize SW ring entries */
1810 prev = (uint16_t) (txq->nb_tx_desc - 1);
1811 for (i = 0; i < txq->nb_tx_desc; i++) {
1812 volatile union ixgbe_adv_tx_desc *txd = &txq->tx_ring[i];
1813 txd->wb.status = IXGBE_TXD_STAT_DD;
1816 txe[prev].next_id = i;
1820 txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
1821 txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
1824 txq->nb_tx_used = 0;
1826 * Always allow 1 descriptor to be un-allocated to avoid
1827 * a H/W race condition
1829 txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
1830 txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
1832 memset((void*)&txq->ctx_cache, 0,
1833 IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
1837 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
1840 unsigned int socket_id,
1841 const struct rte_eth_txconf *tx_conf)
1843 const struct rte_memzone *tz;
1844 struct igb_tx_queue *txq;
1845 struct ixgbe_hw *hw;
1846 uint16_t tx_rs_thresh, tx_free_thresh;
1848 PMD_INIT_FUNC_TRACE();
1849 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1852 * Validate number of transmit descriptors.
1853 * It must not exceed hardware maximum, and must be multiple
1856 if (((nb_desc * sizeof(union ixgbe_adv_tx_desc)) % IXGBE_ALIGN) != 0 ||
1857 (nb_desc > IXGBE_MAX_RING_DESC) ||
1858 (nb_desc < IXGBE_MIN_RING_DESC)) {
1863 * The following two parameters control the setting of the RS bit on
1864 * transmit descriptors.
1865 * TX descriptors will have their RS bit set after txq->tx_rs_thresh
1866 * descriptors have been used.
1867 * The TX descriptor ring will be cleaned after txq->tx_free_thresh
1868 * descriptors are used or if the number of descriptors required
1869 * to transmit a packet is greater than the number of free TX
1871 * The following constraints must be satisfied:
1872 * tx_rs_thresh must be greater than 0.
1873 * tx_rs_thresh must be less than the size of the ring minus 2.
1874 * tx_rs_thresh must be less than or equal to tx_free_thresh.
1875 * tx_rs_thresh must be a divisor of the ring size.
1876 * tx_free_thresh must be greater than 0.
1877 * tx_free_thresh must be less than the size of the ring minus 3.
1878 * One descriptor in the TX ring is used as a sentinel to avoid a
1879 * H/W race condition, hence the maximum threshold constraints.
1880 * When set to zero use default values.
1882 tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
1883 tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
1884 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
1885 tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
1886 if (tx_rs_thresh >= (nb_desc - 2)) {
1887 RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than the number "
1888 "of TX descriptors minus 2. (tx_rs_thresh=%u port=%d "
1889 "queue=%d)\n", (unsigned int)tx_rs_thresh,
1890 (int)dev->data->port_id, (int)queue_idx);
1893 if (tx_free_thresh >= (nb_desc - 3)) {
1894 RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than the "
1895 "tx_free_thresh must be less than the number of TX "
1896 "descriptors minus 3. (tx_free_thresh=%u port=%d "
1897 "queue=%d)\n", (unsigned int)tx_free_thresh,
1898 (int)dev->data->port_id, (int)queue_idx);
1901 if (tx_rs_thresh > tx_free_thresh) {
1902 RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than or equal to "
1903 "tx_free_thresh. (tx_free_thresh=%u tx_rs_thresh=%u "
1904 "port=%d queue=%d)\n", (unsigned int)tx_free_thresh,
1905 (unsigned int)tx_rs_thresh, (int)dev->data->port_id,
1909 if ((nb_desc % tx_rs_thresh) != 0) {
1910 RTE_LOG(ERR, PMD, "tx_rs_thresh must be a divisor of the "
1911 "number of TX descriptors. (tx_rs_thresh=%u port=%d "
1912 "queue=%d)\n", (unsigned int)tx_rs_thresh,
1913 (int)dev->data->port_id, (int)queue_idx);
1918 * If rs_bit_thresh is greater than 1, then TX WTHRESH should be
1919 * set to 0. If WTHRESH is greater than zero, the RS bit is ignored
1920 * by the NIC and all descriptors are written back after the NIC
1921 * accumulates WTHRESH descriptors.
1923 if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
1924 RTE_LOG(ERR, PMD, "TX WTHRESH must be set to 0 if "
1925 "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
1926 "port=%d queue=%d)\n", (unsigned int)tx_rs_thresh,
1927 (int)dev->data->port_id, (int)queue_idx);
1931 /* Free memory prior to re-allocation if needed... */
1932 if (dev->data->tx_queues[queue_idx] != NULL)
1933 ixgbe_tx_queue_release(dev->data->tx_queues[queue_idx]);
1935 /* First allocate the tx queue data structure */
1936 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct igb_tx_queue),
1937 CACHE_LINE_SIZE, socket_id);
1942 * Allocate TX ring hardware descriptors. A memzone large enough to
1943 * handle the maximum ring size is allocated in order to allow for
1944 * resizing in later calls to the queue setup function.
1946 tz = ring_dma_zone_reserve(dev, "tx_ring", queue_idx,
1947 sizeof(union ixgbe_adv_tx_desc) * IXGBE_MAX_RING_DESC,
1950 ixgbe_tx_queue_release(txq);
1954 txq->nb_tx_desc = nb_desc;
1955 txq->tx_rs_thresh = tx_rs_thresh;
1956 txq->tx_free_thresh = tx_free_thresh;
1957 txq->pthresh = tx_conf->tx_thresh.pthresh;
1958 txq->hthresh = tx_conf->tx_thresh.hthresh;
1959 txq->wthresh = tx_conf->tx_thresh.wthresh;
1960 txq->queue_id = queue_idx;
1961 txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
1962 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
1963 txq->port_id = dev->data->port_id;
1964 txq->txq_flags = tx_conf->txq_flags;
1967 * Modification to set VFTDT for virtual function if vf is detected
1969 if (hw->mac.type == ixgbe_mac_82599_vf)
1970 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
1972 txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
1974 txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
1975 txq->tx_ring = (union ixgbe_adv_tx_desc *) tz->addr;
1977 /* Allocate software ring */
1978 txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
1979 sizeof(struct igb_tx_entry) * nb_desc,
1980 CACHE_LINE_SIZE, socket_id);
1981 if (txq->sw_ring == NULL) {
1982 ixgbe_tx_queue_release(txq);
1985 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64"\n",
1986 txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
1988 ixgbe_reset_tx_queue(txq);
1990 dev->data->tx_queues[queue_idx] = txq;
1992 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
1993 if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS) &&
1994 (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
1995 PMD_INIT_LOG(INFO, "Using simple tx code path\n");
1996 dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
1998 PMD_INIT_LOG(INFO, "Using full-featured tx code path\n");
1999 PMD_INIT_LOG(INFO, " - txq_flags = %lx [IXGBE_SIMPLE_FLAGS=%lx]\n", (long unsigned)txq->txq_flags, (long unsigned)IXGBE_SIMPLE_FLAGS);
2000 PMD_INIT_LOG(INFO, " - tx_rs_thresh = %lu [RTE_PMD_IXGBE_TX_MAX_BURST=%lu]\n", (long unsigned)txq->tx_rs_thresh, (long unsigned)RTE_PMD_IXGBE_TX_MAX_BURST);
2001 dev->tx_pkt_burst = ixgbe_xmit_pkts;
2008 ixgbe_rx_queue_release_mbufs(struct igb_rx_queue *rxq)
2012 if (rxq->sw_ring != NULL) {
2013 for (i = 0; i < rxq->nb_rx_desc; i++) {
2014 if (rxq->sw_ring[i].mbuf != NULL) {
2015 rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
2016 rxq->sw_ring[i].mbuf = NULL;
2019 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2020 if (rxq->rx_nb_avail) {
2021 for (i = 0; i < rxq->rx_nb_avail; ++i) {
2022 struct rte_mbuf *mb;
2023 mb = rxq->rx_stage[rxq->rx_next_avail + i];
2024 rte_pktmbuf_free_seg(mb);
2026 rxq->rx_nb_avail = 0;
2033 ixgbe_rx_queue_release(struct igb_rx_queue *rxq)
2036 ixgbe_rx_queue_release_mbufs(rxq);
2037 rte_free(rxq->sw_ring);
2043 ixgbe_dev_rx_queue_release(void *rxq)
2045 ixgbe_rx_queue_release(rxq);
2049 * Check if Rx Burst Bulk Alloc function can be used.
2051 * 0: the preconditions are satisfied and the bulk allocation function
2053 * -EINVAL: the preconditions are NOT satisfied and the default Rx burst
2054 * function must be used.
2057 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2058 check_rx_burst_bulk_alloc_preconditions(struct igb_rx_queue *rxq)
2060 check_rx_burst_bulk_alloc_preconditions(__rte_unused struct igb_rx_queue *rxq)
2066 * Make sure the following pre-conditions are satisfied:
2067 * rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST
2068 * rxq->rx_free_thresh < rxq->nb_rx_desc
2069 * (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
2070 * rxq->nb_rx_desc<(IXGBE_MAX_RING_DESC-RTE_PMD_IXGBE_RX_MAX_BURST)
2071 * Scattered packets are not supported. This should be checked
2072 * outside of this function.
2074 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2075 if (! (rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST))
2077 else if (! (rxq->rx_free_thresh < rxq->nb_rx_desc))
2079 else if (! ((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0))
2081 else if (! (rxq->nb_rx_desc <
2082 (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST)))
2091 /* Reset dynamic igb_rx_queue fields back to defaults */
2093 ixgbe_reset_rx_queue(struct igb_rx_queue *rxq)
2099 * By default, the Rx queue setup function allocates enough memory for
2100 * IXGBE_MAX_RING_DESC. The Rx Burst bulk allocation function requires
2101 * extra memory at the end of the descriptor ring to be zero'd out. A
2102 * pre-condition for using the Rx burst bulk alloc function is that the
2103 * number of descriptors is less than or equal to
2104 * (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST). Check all the
2105 * constraints here to see if we need to zero out memory after the end
2106 * of the H/W descriptor ring.
2108 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2109 if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
2110 /* zero out extra memory */
2111 len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_IXGBE_RX_MAX_BURST);
2114 /* do not zero out extra memory */
2115 len = rxq->nb_rx_desc;
2118 * Zero out HW ring memory. Zero out extra memory at the end of
2119 * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
2120 * reads extra memory as zeros.
2122 for (i = 0; i < len * sizeof(union ixgbe_adv_rx_desc); i++) {
2123 ((volatile char *)rxq->rx_ring)[i] = 0;
2126 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2128 * initialize extra software ring entries. Space for these extra
2129 * entries is always allocated
2131 memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
2132 for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST; ++i) {
2133 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
2136 rxq->rx_nb_avail = 0;
2137 rxq->rx_next_avail = 0;
2138 rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
2139 #endif /* RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC */
2141 rxq->nb_rx_hold = 0;
2142 rxq->pkt_first_seg = NULL;
2143 rxq->pkt_last_seg = NULL;
2147 ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
2150 unsigned int socket_id,
2151 const struct rte_eth_rxconf *rx_conf,
2152 struct rte_mempool *mp)
2154 const struct rte_memzone *rz;
2155 struct igb_rx_queue *rxq;
2156 struct ixgbe_hw *hw;
2157 int use_def_burst_func = 1;
2160 PMD_INIT_FUNC_TRACE();
2161 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2164 * Validate number of receive descriptors.
2165 * It must not exceed hardware maximum, and must be multiple
2168 if (((nb_desc * sizeof(union ixgbe_adv_rx_desc)) % IXGBE_ALIGN) != 0 ||
2169 (nb_desc > IXGBE_MAX_RING_DESC) ||
2170 (nb_desc < IXGBE_MIN_RING_DESC)) {
2174 /* Free memory prior to re-allocation if needed... */
2175 if (dev->data->rx_queues[queue_idx] != NULL)
2176 ixgbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
2178 /* First allocate the rx queue data structure */
2179 rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct igb_rx_queue),
2180 CACHE_LINE_SIZE, socket_id);
2184 rxq->nb_rx_desc = nb_desc;
2185 rxq->rx_free_thresh = rx_conf->rx_free_thresh;
2186 rxq->queue_id = queue_idx;
2187 rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
2188 queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
2189 rxq->port_id = dev->data->port_id;
2190 rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
2192 rxq->drop_en = rx_conf->rx_drop_en;
2195 * Allocate RX ring hardware descriptors. A memzone large enough to
2196 * handle the maximum ring size is allocated in order to allow for
2197 * resizing in later calls to the queue setup function.
2199 rz = ring_dma_zone_reserve(dev, "rx_ring", queue_idx,
2200 IXGBE_MAX_RING_DESC * sizeof(union ixgbe_adv_rx_desc),
2203 ixgbe_rx_queue_release(rxq);
2207 * Modified to setup VFRDT for Virtual Function
2209 if (hw->mac.type == ixgbe_mac_82599_vf) {
2211 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
2213 IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDH(queue_idx));
2217 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDT(rxq->reg_idx));
2219 IXGBE_PCI_REG_ADDR(hw, IXGBE_RDH(rxq->reg_idx));
2222 rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
2223 rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
2226 * Allocate software ring. Allow for space at the end of the
2227 * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
2228 * function does not access an invalid memory region.
2230 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2231 len = (uint16_t)(nb_desc + RTE_PMD_IXGBE_RX_MAX_BURST);
2235 rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
2236 sizeof(struct igb_rx_entry) * len,
2237 CACHE_LINE_SIZE, socket_id);
2238 if (rxq->sw_ring == NULL) {
2239 ixgbe_rx_queue_release(rxq);
2242 PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64"\n",
2243 rxq->sw_ring, rxq->rx_ring, rxq->rx_ring_phys_addr);
2246 * Certain constaints must be met in order to use the bulk buffer
2247 * allocation Rx burst function.
2249 use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
2251 /* Check if pre-conditions are satisfied, and no Scattered Rx */
2252 if (!use_def_burst_func && !dev->data->scattered_rx) {
2253 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
2254 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
2255 "satisfied. Rx Burst Bulk Alloc function will be "
2256 "used on port=%d, queue=%d.\n",
2257 rxq->port_id, rxq->queue_id);
2258 dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
2261 PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions "
2262 "are not satisfied, Scattered Rx is requested, "
2263 "or RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC is not "
2264 "enabled (port=%d, queue=%d).\n",
2265 rxq->port_id, rxq->queue_id);
2267 dev->data->rx_queues[queue_idx] = rxq;
2269 ixgbe_reset_rx_queue(rxq);
2275 ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
2277 #define IXGBE_RXQ_SCAN_INTERVAL 4
2278 volatile union ixgbe_adv_rx_desc *rxdp;
2279 struct igb_rx_queue *rxq;
2282 if (rx_queue_id >= dev->data->nb_rx_queues) {
2283 PMD_RX_LOG(ERR, "Invalid RX queue id=%d\n", rx_queue_id);
2287 rxq = dev->data->rx_queues[rx_queue_id];
2288 rxdp = &(rxq->rx_ring[rxq->rx_tail]);
2290 while ((desc < rxq->nb_rx_desc) &&
2291 (rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD)) {
2292 desc += IXGBE_RXQ_SCAN_INTERVAL;
2293 rxdp += IXGBE_RXQ_SCAN_INTERVAL;
2294 if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
2295 rxdp = &(rxq->rx_ring[rxq->rx_tail +
2296 desc - rxq->nb_rx_desc]);
2303 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
2305 volatile union ixgbe_adv_rx_desc *rxdp;
2306 struct igb_rx_queue *rxq = rx_queue;
2309 if (unlikely(offset >= rxq->nb_rx_desc))
2311 desc = rxq->rx_tail + offset;
2312 if (desc >= rxq->nb_rx_desc)
2313 desc -= rxq->nb_rx_desc;
2315 rxdp = &rxq->rx_ring[desc];
2316 return !!(rxdp->wb.upper.status_error & IXGBE_RXDADV_STAT_DD);
2320 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
2324 PMD_INIT_FUNC_TRACE();
2326 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2327 struct igb_tx_queue *txq = dev->data->tx_queues[i];
2329 ixgbe_tx_queue_release_mbufs(txq);
2330 ixgbe_reset_tx_queue(txq);
2334 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2335 struct igb_rx_queue *rxq = dev->data->rx_queues[i];
2337 ixgbe_rx_queue_release_mbufs(rxq);
2338 ixgbe_reset_rx_queue(rxq);
2343 /*********************************************************************
2345 * Device RX/TX init functions
2347 **********************************************************************/
2350 * Receive Side Scaling (RSS)
2351 * See section 7.1.2.8 in the following document:
2352 * "Intel 82599 10 GbE Controller Datasheet" - Revision 2.1 October 2009
2355 * The source and destination IP addresses of the IP header and the source
2356 * and destination ports of TCP/UDP headers, if any, of received packets are
2357 * hashed against a configurable random key to compute a 32-bit RSS hash result.
2358 * The seven (7) LSBs of the 32-bit hash result are used as an index into a
2359 * 128-entry redirection table (RETA). Each entry of the RETA provides a 3-bit
2360 * RSS output index which is used as the RX queue index where to store the
2362 * The following output is supplied in the RX write-back descriptor:
2363 * - 32-bit result of the Microsoft RSS hash function,
2364 * - 4-bit RSS type field.
2368 * RSS random key supplied in section 7.1.2.8.3 of the Intel 82599 datasheet.
2369 * Used as the default key.
2371 static uint8_t rss_intel_key[40] = {
2372 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
2373 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
2374 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2375 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
2376 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
2380 ixgbe_rss_disable(struct rte_eth_dev *dev)
2382 struct ixgbe_hw *hw;
2385 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2386 mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
2387 mrqc &= ~IXGBE_MRQC_RSSEN;
2388 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2392 ixgbe_rss_configure(struct rte_eth_dev *dev)
2394 struct ixgbe_hw *hw;
2403 PMD_INIT_FUNC_TRACE();
2404 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2406 rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
2407 if (rss_hf == 0) { /* Disable RSS */
2408 ixgbe_rss_disable(dev);
2411 hash_key = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
2412 if (hash_key == NULL)
2413 hash_key = rss_intel_key; /* Default hash key */
2415 /* Fill in RSS hash key */
2416 for (i = 0; i < 10; i++) {
2417 rss_key = hash_key[(i * 4)];
2418 rss_key |= hash_key[(i * 4) + 1] << 8;
2419 rss_key |= hash_key[(i * 4) + 2] << 16;
2420 rss_key |= hash_key[(i * 4) + 3] << 24;
2421 IXGBE_WRITE_REG_ARRAY(hw, IXGBE_RSSRK(0), i, rss_key);
2424 /* Fill in redirection table */
2426 for (i = 0, j = 0; i < 128; i++, j++) {
2427 if (j == dev->data->nb_rx_queues) j = 0;
2428 reta = (reta << 8) | j;
2430 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), rte_bswap32(reta));
2433 /* Set configured hashing functions in MRQC register */
2434 mrqc = IXGBE_MRQC_RSSEN; /* RSS enable */
2435 if (rss_hf & ETH_RSS_IPV4)
2436 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
2437 if (rss_hf & ETH_RSS_IPV4_TCP)
2438 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
2439 if (rss_hf & ETH_RSS_IPV6)
2440 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
2441 if (rss_hf & ETH_RSS_IPV6_EX)
2442 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
2443 if (rss_hf & ETH_RSS_IPV6_TCP)
2444 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2445 if (rss_hf & ETH_RSS_IPV6_TCP_EX)
2446 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
2447 if (rss_hf & ETH_RSS_IPV4_UDP)
2448 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2449 if (rss_hf & ETH_RSS_IPV6_UDP)
2450 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2451 if (rss_hf & ETH_RSS_IPV6_UDP_EX)
2452 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
2453 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2456 #define NUM_VFTA_REGISTERS 128
2457 #define NIC_RX_BUFFER_SIZE 0x200
2460 ixgbe_vmdq_dcb_configure(struct rte_eth_dev *dev)
2462 struct rte_eth_vmdq_dcb_conf *cfg;
2463 struct ixgbe_hw *hw;
2464 enum rte_eth_nb_pools num_pools;
2465 uint32_t mrqc, vt_ctl, queue_mapping, vlanctrl;
2467 uint8_t nb_tcs; /* number of traffic classes */
2470 PMD_INIT_FUNC_TRACE();
2471 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2472 cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
2473 num_pools = cfg->nb_queue_pools;
2474 /* Check we have a valid number of pools */
2475 if (num_pools != ETH_16_POOLS && num_pools != ETH_32_POOLS) {
2476 ixgbe_rss_disable(dev);
2479 /* 16 pools -> 8 traffic classes, 32 pools -> 4 traffic classes */
2480 nb_tcs = (uint8_t)(ETH_VMDQ_DCB_NUM_QUEUES / (int)num_pools);
2484 * split rx buffer up into sections, each for 1 traffic class
2486 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
2487 for (i = 0 ; i < nb_tcs; i++) {
2488 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2489 rxpbsize &= (~(0x3FF << IXGBE_RXPBSIZE_SHIFT));
2490 /* clear 10 bits. */
2491 rxpbsize |= (pbsize << IXGBE_RXPBSIZE_SHIFT); /* set value */
2492 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2494 /* zero alloc all unused TCs */
2495 for (i = nb_tcs; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2496 uint32_t rxpbsize = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i));
2497 rxpbsize &= (~( 0x3FF << IXGBE_RXPBSIZE_SHIFT ));
2498 /* clear 10 bits. */
2499 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2502 /* MRQC: enable vmdq and dcb */
2503 mrqc = ((num_pools == ETH_16_POOLS) ? \
2504 IXGBE_MRQC_VMDQRT8TCEN : IXGBE_MRQC_VMDQRT4TCEN );
2505 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2507 /* PFVTCTL: turn on virtualisation and set the default pool */
2508 vt_ctl = IXGBE_VT_CTL_VT_ENABLE | IXGBE_VT_CTL_REPLEN;
2509 if (cfg->enable_default_pool) {
2510 vt_ctl |= (cfg->default_pool << IXGBE_VT_CTL_POOL_SHIFT);
2512 vt_ctl |= IXGBE_VT_CTL_DIS_DEFPL;
2515 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
2517 /* RTRUP2TC: mapping user priorities to traffic classes (TCs) */
2519 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++)
2521 * mapping is done with 3 bits per priority,
2522 * so shift by i*3 each time
2524 queue_mapping |= ((cfg->dcb_queue[i] & 0x07) << (i * 3));
2526 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, queue_mapping);
2528 /* RTRPCS: DCB related */
2529 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, IXGBE_RMCS_RRM);
2531 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
2532 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2533 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
2534 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
2536 /* VFTA - enable all vlan filters */
2537 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
2538 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
2541 /* VFRE: pool enabling for receive - 16 or 32 */
2542 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), \
2543 num_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
2546 * MPSAR - allow pools to read specific mac addresses
2547 * In this case, all pools should be able to read from mac addr 0
2549 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(0), 0xFFFFFFFF);
2550 IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(0), 0xFFFFFFFF);
2552 /* PFVLVF, PFVLVFB: set up filters for vlan tags as configured */
2553 for (i = 0; i < cfg->nb_pool_maps; i++) {
2554 /* set vlan id in VF register and set the valid bit */
2555 IXGBE_WRITE_REG(hw, IXGBE_VLVF(i), (IXGBE_VLVF_VIEN | \
2556 (cfg->pool_map[i].vlan_id & 0xFFF)));
2558 * Put the allowed pools in VFB reg. As we only have 16 or 32
2559 * pools, we only need to use the first half of the register
2562 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(i*2), cfg->pool_map[i].pools);
2567 * ixgbe_dcb_config_tx_hw_config - Configure general DCB TX parameters
2568 * @hw: pointer to hardware structure
2569 * @dcb_config: pointer to ixgbe_dcb_config structure
2572 ixgbe_dcb_tx_hw_config(struct ixgbe_hw *hw,
2573 struct ixgbe_dcb_config *dcb_config)
2578 PMD_INIT_FUNC_TRACE();
2579 if (hw->mac.type != ixgbe_mac_82598EB) {
2580 /* Disable the Tx desc arbiter so that MTQC can be changed */
2581 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2582 reg |= IXGBE_RTTDCS_ARBDIS;
2583 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
2585 /* Enable DCB for Tx with 8 TCs */
2586 if (dcb_config->num_tcs.pg_tcs == 8) {
2587 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2590 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2592 if (dcb_config->vt_mode)
2593 reg |= IXGBE_MTQC_VT_ENA;
2594 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2596 /* Disable drop for all queues */
2597 for (q = 0; q < 128; q++)
2598 IXGBE_WRITE_REG(hw, IXGBE_QDE,
2599 (IXGBE_QDE_WRITE | (q << IXGBE_QDE_IDX_SHIFT)));
2601 /* Enable the Tx desc arbiter */
2602 reg = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2603 reg &= ~IXGBE_RTTDCS_ARBDIS;
2604 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, reg);
2606 /* Enable Security TX Buffer IFG for DCB */
2607 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2608 reg |= IXGBE_SECTX_DCB;
2609 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2615 * ixgbe_vmdq_dcb_hw_tx_config - Configure general VMDQ+DCB TX parameters
2616 * @dev: pointer to rte_eth_dev structure
2617 * @dcb_config: pointer to ixgbe_dcb_config structure
2620 ixgbe_vmdq_dcb_hw_tx_config(struct rte_eth_dev *dev,
2621 struct ixgbe_dcb_config *dcb_config)
2623 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
2624 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
2625 struct ixgbe_hw *hw =
2626 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2628 PMD_INIT_FUNC_TRACE();
2629 if (hw->mac.type != ixgbe_mac_82598EB)
2630 /*PF VF Transmit Enable*/
2631 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0),
2632 vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ? 0xFFFF : 0xFFFFFFFF);
2634 /*Configure general DCB TX parameters*/
2635 ixgbe_dcb_tx_hw_config(hw,dcb_config);
2640 ixgbe_vmdq_dcb_rx_config(struct rte_eth_dev *dev,
2641 struct ixgbe_dcb_config *dcb_config)
2643 struct rte_eth_vmdq_dcb_conf *vmdq_rx_conf =
2644 &dev->data->dev_conf.rx_adv_conf.vmdq_dcb_conf;
2645 struct ixgbe_dcb_tc_config *tc;
2648 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
2649 if (vmdq_rx_conf->nb_queue_pools == ETH_16_POOLS ) {
2650 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
2651 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
2654 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
2655 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
2657 /* User Priority to Traffic Class mapping */
2658 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2659 j = vmdq_rx_conf->dcb_queue[i];
2660 tc = &dcb_config->tc_config[j];
2661 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
2667 ixgbe_dcb_vt_tx_config(struct rte_eth_dev *dev,
2668 struct ixgbe_dcb_config *dcb_config)
2670 struct rte_eth_vmdq_dcb_tx_conf *vmdq_tx_conf =
2671 &dev->data->dev_conf.tx_adv_conf.vmdq_dcb_tx_conf;
2672 struct ixgbe_dcb_tc_config *tc;
2675 /* convert rte_eth_conf.rx_adv_conf to struct ixgbe_dcb_config */
2676 if (vmdq_tx_conf->nb_queue_pools == ETH_16_POOLS ) {
2677 dcb_config->num_tcs.pg_tcs = ETH_8_TCS;
2678 dcb_config->num_tcs.pfc_tcs = ETH_8_TCS;
2681 dcb_config->num_tcs.pg_tcs = ETH_4_TCS;
2682 dcb_config->num_tcs.pfc_tcs = ETH_4_TCS;
2685 /* User Priority to Traffic Class mapping */
2686 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2687 j = vmdq_tx_conf->dcb_queue[i];
2688 tc = &dcb_config->tc_config[j];
2689 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
2696 ixgbe_dcb_rx_config(struct rte_eth_dev *dev,
2697 struct ixgbe_dcb_config *dcb_config)
2699 struct rte_eth_dcb_rx_conf *rx_conf =
2700 &dev->data->dev_conf.rx_adv_conf.dcb_rx_conf;
2701 struct ixgbe_dcb_tc_config *tc;
2704 dcb_config->num_tcs.pg_tcs = (uint8_t)rx_conf->nb_tcs;
2705 dcb_config->num_tcs.pfc_tcs = (uint8_t)rx_conf->nb_tcs;
2707 /* User Priority to Traffic Class mapping */
2708 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2709 j = rx_conf->dcb_queue[i];
2710 tc = &dcb_config->tc_config[j];
2711 tc->path[IXGBE_DCB_RX_CONFIG].up_to_tc_bitmap =
2717 ixgbe_dcb_tx_config(struct rte_eth_dev *dev,
2718 struct ixgbe_dcb_config *dcb_config)
2720 struct rte_eth_dcb_tx_conf *tx_conf =
2721 &dev->data->dev_conf.tx_adv_conf.dcb_tx_conf;
2722 struct ixgbe_dcb_tc_config *tc;
2725 dcb_config->num_tcs.pg_tcs = (uint8_t)tx_conf->nb_tcs;
2726 dcb_config->num_tcs.pfc_tcs = (uint8_t)tx_conf->nb_tcs;
2728 /* User Priority to Traffic Class mapping */
2729 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2730 j = tx_conf->dcb_queue[i];
2731 tc = &dcb_config->tc_config[j];
2732 tc->path[IXGBE_DCB_TX_CONFIG].up_to_tc_bitmap =
2738 * ixgbe_dcb_rx_hw_config - Configure general DCB RX HW parameters
2739 * @hw: pointer to hardware structure
2740 * @dcb_config: pointer to ixgbe_dcb_config structure
2743 ixgbe_dcb_rx_hw_config(struct ixgbe_hw *hw,
2744 struct ixgbe_dcb_config *dcb_config)
2750 PMD_INIT_FUNC_TRACE();
2752 * Disable the arbiter before changing parameters
2753 * (always enable recycle mode; WSP)
2755 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC | IXGBE_RTRPCS_ARBDIS;
2756 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
2758 if (hw->mac.type != ixgbe_mac_82598EB) {
2759 reg = IXGBE_READ_REG(hw, IXGBE_MRQC);
2760 if (dcb_config->num_tcs.pg_tcs == 4) {
2761 if (dcb_config->vt_mode)
2762 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
2763 IXGBE_MRQC_VMDQRT4TCEN;
2765 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
2766 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
2770 if (dcb_config->num_tcs.pg_tcs == 8) {
2771 if (dcb_config->vt_mode)
2772 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
2773 IXGBE_MRQC_VMDQRT8TCEN;
2775 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, 0);
2776 reg = (reg & ~IXGBE_MRQC_MRQE_MASK) |
2781 IXGBE_WRITE_REG(hw, IXGBE_MRQC, reg);
2784 /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
2785 vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2786 vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
2787 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
2789 /* VFTA - enable all vlan filters */
2790 for (i = 0; i < NUM_VFTA_REGISTERS; i++) {
2791 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
2795 * Configure Rx packet plane (recycle mode; WSP) and
2798 reg = IXGBE_RTRPCS_RRM | IXGBE_RTRPCS_RAC;
2799 IXGBE_WRITE_REG(hw, IXGBE_RTRPCS, reg);
2805 ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
2806 uint16_t *max,uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
2808 switch (hw->mac.type) {
2809 case ixgbe_mac_82598EB:
2810 ixgbe_dcb_config_rx_arbiter_82598(hw, refill, max, tsa);
2812 case ixgbe_mac_82599EB:
2813 case ixgbe_mac_X540:
2814 ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
2823 ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *max,
2824 uint8_t *bwg_id, uint8_t *tsa, uint8_t *map)
2826 switch (hw->mac.type) {
2827 case ixgbe_mac_82598EB:
2828 ixgbe_dcb_config_tx_desc_arbiter_82598(hw, refill, max, bwg_id,tsa);
2829 ixgbe_dcb_config_tx_data_arbiter_82598(hw, refill, max, bwg_id,tsa);
2831 case ixgbe_mac_82599EB:
2832 case ixgbe_mac_X540:
2833 ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id,tsa);
2834 ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id,tsa, map);
2841 #define DCB_RX_CONFIG 1
2842 #define DCB_TX_CONFIG 1
2843 #define DCB_TX_PB 1024
2845 * ixgbe_dcb_hw_configure - Enable DCB and configure
2846 * general DCB in VT mode and non-VT mode parameters
2847 * @dev: pointer to rte_eth_dev structure
2848 * @dcb_config: pointer to ixgbe_dcb_config structure
2851 ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
2852 struct ixgbe_dcb_config *dcb_config)
2855 uint8_t i,pfc_en,nb_tcs;
2857 uint8_t config_dcb_rx = 0;
2858 uint8_t config_dcb_tx = 0;
2859 uint8_t tsa[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
2860 uint8_t bwgid[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
2861 uint16_t refill[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
2862 uint16_t max[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
2863 uint8_t map[IXGBE_DCB_MAX_TRAFFIC_CLASS] = {0};
2864 struct ixgbe_dcb_tc_config *tc;
2865 uint32_t max_frame = dev->data->max_frame_size;
2866 struct ixgbe_hw *hw =
2867 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2869 switch(dev->data->dev_conf.rxmode.mq_mode){
2870 case ETH_MQ_RX_VMDQ_DCB:
2871 dcb_config->vt_mode = true;
2872 if (hw->mac.type != ixgbe_mac_82598EB) {
2873 config_dcb_rx = DCB_RX_CONFIG;
2875 *get dcb and VT rx configuration parameters
2878 ixgbe_vmdq_dcb_rx_config(dev,dcb_config);
2879 /*Configure general VMDQ and DCB RX parameters*/
2880 ixgbe_vmdq_dcb_configure(dev);
2884 dcb_config->vt_mode = false;
2885 config_dcb_rx = DCB_RX_CONFIG;
2886 /* Get dcb TX configuration parameters from rte_eth_conf */
2887 ixgbe_dcb_rx_config(dev,dcb_config);
2888 /*Configure general DCB RX parameters*/
2889 ixgbe_dcb_rx_hw_config(hw, dcb_config);
2892 PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration\n");
2895 switch (dev->data->dev_conf.txmode.mq_mode) {
2896 case ETH_MQ_TX_VMDQ_DCB:
2897 dcb_config->vt_mode = true;
2898 config_dcb_tx = DCB_TX_CONFIG;
2899 /* get DCB and VT TX configuration parameters from rte_eth_conf */
2900 ixgbe_dcb_vt_tx_config(dev,dcb_config);
2901 /*Configure general VMDQ and DCB TX parameters*/
2902 ixgbe_vmdq_dcb_hw_tx_config(dev,dcb_config);
2906 dcb_config->vt_mode = false;
2907 config_dcb_tx = DCB_TX_CONFIG;
2908 /*get DCB TX configuration parameters from rte_eth_conf*/
2909 ixgbe_dcb_tx_config(dev,dcb_config);
2910 /*Configure general DCB TX parameters*/
2911 ixgbe_dcb_tx_hw_config(hw, dcb_config);
2914 PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration\n");
2918 nb_tcs = dcb_config->num_tcs.pfc_tcs;
2920 ixgbe_dcb_unpack_map_cee(dcb_config, IXGBE_DCB_RX_CONFIG, map);
2921 if(nb_tcs == ETH_4_TCS) {
2922 /* Avoid un-configured priority mapping to TC0 */
2924 uint8_t mask = 0xFF;
2925 for (i = 0; i < ETH_DCB_NUM_USER_PRIORITIES - 4; i++)
2926 mask = (uint8_t)(mask & (~ (1 << map[i])));
2927 for (i = 0; mask && (i < IXGBE_DCB_MAX_TRAFFIC_CLASS); i++) {
2928 if ((mask & 0x1) && (j < ETH_DCB_NUM_USER_PRIORITIES))
2932 /* Re-configure 4 TCs BW */
2933 for (i = 0; i < nb_tcs; i++) {
2934 tc = &dcb_config->tc_config[i];
2935 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent =
2936 (uint8_t)(100 / nb_tcs);
2937 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent =
2938 (uint8_t)(100 / nb_tcs);
2940 for (; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
2941 tc = &dcb_config->tc_config[i];
2942 tc->path[IXGBE_DCB_TX_CONFIG].bwg_percent = 0;
2943 tc->path[IXGBE_DCB_RX_CONFIG].bwg_percent = 0;
2948 /* Set RX buffer size */
2949 pbsize = (uint16_t)(NIC_RX_BUFFER_SIZE / nb_tcs);
2950 uint32_t rxpbsize = pbsize << IXGBE_RXPBSIZE_SHIFT;
2951 for (i = 0 ; i < nb_tcs; i++) {
2952 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), rxpbsize);
2954 /* zero alloc all unused TCs */
2955 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2956 IXGBE_WRITE_REG(hw, IXGBE_RXPBSIZE(i), 0);
2960 /* Only support an equally distributed Tx packet buffer strategy. */
2961 uint32_t txpktsize = IXGBE_TXPBSIZE_MAX / nb_tcs;
2962 uint32_t txpbthresh = (txpktsize / DCB_TX_PB) - IXGBE_TXPKT_SIZE_MAX;
2963 for (i = 0; i < nb_tcs; i++) {
2964 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), txpktsize);
2965 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), txpbthresh);
2967 /* Clear unused TCs, if any, to zero buffer size*/
2968 for (; i < ETH_DCB_NUM_USER_PRIORITIES; i++) {
2969 IXGBE_WRITE_REG(hw, IXGBE_TXPBSIZE(i), 0);
2970 IXGBE_WRITE_REG(hw, IXGBE_TXPBTHRESH(i), 0);
2974 /*Calculates traffic class credits*/
2975 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
2976 IXGBE_DCB_TX_CONFIG);
2977 ixgbe_dcb_calculate_tc_credits_cee(hw, dcb_config,max_frame,
2978 IXGBE_DCB_RX_CONFIG);
2981 /* Unpack CEE standard containers */
2982 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_RX_CONFIG, refill);
2983 ixgbe_dcb_unpack_max_cee(dcb_config, max);
2984 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_RX_CONFIG, bwgid);
2985 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_RX_CONFIG, tsa);
2986 /* Configure PG(ETS) RX */
2987 ixgbe_dcb_hw_arbite_rx_config(hw,refill,max,bwgid,tsa,map);
2991 /* Unpack CEE standard containers */
2992 ixgbe_dcb_unpack_refill_cee(dcb_config, IXGBE_DCB_TX_CONFIG, refill);
2993 ixgbe_dcb_unpack_max_cee(dcb_config, max);
2994 ixgbe_dcb_unpack_bwgid_cee(dcb_config, IXGBE_DCB_TX_CONFIG, bwgid);
2995 ixgbe_dcb_unpack_tsa_cee(dcb_config, IXGBE_DCB_TX_CONFIG, tsa);
2996 /* Configure PG(ETS) TX */
2997 ixgbe_dcb_hw_arbite_tx_config(hw,refill,max,bwgid,tsa,map);
3000 /*Configure queue statistics registers*/
3001 ixgbe_dcb_config_tc_stats_82599(hw, dcb_config);
3003 /* Check if the PFC is supported */
3004 if(dev->data->dev_conf.dcb_capability_en & ETH_DCB_PFC_SUPPORT) {
3005 pbsize = (uint16_t) (NIC_RX_BUFFER_SIZE / nb_tcs);
3006 for (i = 0; i < nb_tcs; i++) {
3008 * If the TC count is 8,and the default high_water is 48,
3009 * the low_water is 16 as default.
3011 hw->fc.high_water[i] = (pbsize * 3 ) / 4;
3012 hw->fc.low_water[i] = pbsize / 4;
3013 /* Enable pfc for this TC */
3014 tc = &dcb_config->tc_config[i];
3015 tc->pfc = ixgbe_dcb_pfc_enabled;
3017 ixgbe_dcb_unpack_pfc_cee(dcb_config, map, &pfc_en);
3018 if(dcb_config->num_tcs.pfc_tcs == ETH_4_TCS)
3020 ret = ixgbe_dcb_config_pfc(hw, pfc_en, map);
3027 * ixgbe_configure_dcb - Configure DCB Hardware
3028 * @dev: pointer to rte_eth_dev
3030 void ixgbe_configure_dcb(struct rte_eth_dev *dev)
3032 struct ixgbe_dcb_config *dcb_cfg =
3033 IXGBE_DEV_PRIVATE_TO_DCB_CFG(dev->data->dev_private);
3034 struct rte_eth_conf *dev_conf = &(dev->data->dev_conf);
3036 PMD_INIT_FUNC_TRACE();
3038 /* check support mq_mode for DCB */
3039 if ((dev_conf->rxmode.mq_mode != ETH_MQ_RX_VMDQ_DCB) ||
3040 (dev_conf->rxmode.mq_mode != ETH_MQ_RX_DCB))
3043 if ((dev_conf->txmode.mq_mode != ETH_MQ_TX_VMDQ_DCB) ||
3044 (dev_conf->txmode.mq_mode != ETH_MQ_TX_DCB))
3047 if ((dev->data->nb_rx_queues != ETH_DCB_NUM_QUEUES) ||
3048 (dev->data->nb_tx_queues != ETH_DCB_NUM_QUEUES))
3051 /** Configure DCB hardware **/
3052 ixgbe_dcb_hw_configure(dev,dcb_cfg);
3058 ixgbe_alloc_rx_queue_mbufs(struct igb_rx_queue *rxq)
3060 struct igb_rx_entry *rxe = rxq->sw_ring;
3064 /* Initialize software ring entries */
3065 for (i = 0; i < rxq->nb_rx_desc; i++) {
3066 volatile union ixgbe_adv_rx_desc *rxd;
3067 struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mb_pool);
3069 PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u\n",
3070 (unsigned) rxq->queue_id);
3074 rte_mbuf_refcnt_set(mbuf, 1);
3075 mbuf->type = RTE_MBUF_PKT;
3076 mbuf->pkt.next = NULL;
3077 mbuf->pkt.data = (char *)mbuf->buf_addr + RTE_PKTMBUF_HEADROOM;
3078 mbuf->pkt.nb_segs = 1;
3079 mbuf->pkt.in_port = rxq->port_id;
3082 rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mbuf));
3083 rxd = &rxq->rx_ring[i];
3084 rxd->read.hdr_addr = dma_addr;
3085 rxd->read.pkt_addr = dma_addr;
3093 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
3095 struct ixgbe_hw *hw =
3096 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3098 if (hw->mac.type == ixgbe_mac_82598EB)
3101 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3103 * SRIOV inactive scheme
3104 * any DCB/RSS w/o VMDq multi-queue setting
3106 if (dev->data->nb_rx_queues > 1)
3107 switch (dev->data->dev_conf.rxmode.mq_mode) {
3108 case ETH_MQ_RX_NONE:
3109 /* if mq_mode not assign, we use rss mode.*/
3111 ixgbe_rss_configure(dev);
3114 case ETH_MQ_RX_VMDQ_DCB:
3115 ixgbe_vmdq_dcb_configure(dev);
3118 default: ixgbe_rss_disable(dev);
3121 ixgbe_rss_disable(dev);
3123 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3125 * SRIOV active scheme
3126 * FIXME if support DCB/RSS together with VMDq & SRIOV
3129 IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQEN);
3133 IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQRT4TCEN);
3137 IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQRT8TCEN);
3140 RTE_LOG(ERR, PMD, "invalid pool number in IOV mode\n");
3148 ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
3150 struct ixgbe_hw *hw =
3151 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3155 if (hw->mac.type == ixgbe_mac_82598EB)
3158 /* disable arbiter before setting MTQC */
3159 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3160 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3161 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3163 if (RTE_ETH_DEV_SRIOV(dev).active == 0) {
3165 * SRIOV inactive scheme
3166 * any DCB w/o VMDq multi-queue setting
3168 mtqc = IXGBE_MTQC_64Q_1PB;
3170 switch (RTE_ETH_DEV_SRIOV(dev).active) {
3173 * SRIOV active scheme
3174 * FIXME if support DCB together with VMDq & SRIOV
3177 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF;
3180 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_32VF;
3183 mtqc = IXGBE_MTQC_VT_ENA | IXGBE_MTQC_RT_ENA |
3187 mtqc = IXGBE_MTQC_64Q_1PB;
3188 RTE_LOG(ERR, PMD, "invalid pool number in IOV mode\n");
3192 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3194 /* re-enable arbiter */
3195 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3196 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3202 * Initializes Receive Unit.
3205 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
3207 struct ixgbe_hw *hw;
3208 struct igb_rx_queue *rxq;
3209 struct rte_pktmbuf_pool_private *mbp_priv;
3222 PMD_INIT_FUNC_TRACE();
3223 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3226 * Make sure receives are disabled while setting
3227 * up the RX context (registers, descriptor rings, etc.).
3229 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3230 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3232 /* Enable receipt of broadcasted frames */
3233 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3234 fctrl |= IXGBE_FCTRL_BAM;
3235 fctrl |= IXGBE_FCTRL_DPF;
3236 fctrl |= IXGBE_FCTRL_PMCF;
3237 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3240 * Configure CRC stripping, if any.
3242 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3243 if (dev->data->dev_conf.rxmode.hw_strip_crc)
3244 hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
3246 hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
3249 * Configure jumbo frame support, if any.
3251 if (dev->data->dev_conf.rxmode.jumbo_frame == 1) {
3252 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3253 maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
3254 maxfrs &= 0x0000FFFF;
3255 maxfrs |= (dev->data->dev_conf.rxmode.max_rx_pkt_len << 16);
3256 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
3258 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
3260 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3262 /* Setup RX queues */
3263 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3264 rxq = dev->data->rx_queues[i];
3266 /* Allocate buffers for descriptor rings */
3267 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
3272 * Reset crc_len in case it was changed after queue setup by a
3273 * call to configure.
3275 rxq->crc_len = (uint8_t)
3276 ((dev->data->dev_conf.rxmode.hw_strip_crc) ? 0 :
3279 /* Setup the Base and Length of the Rx Descriptor Rings */
3280 bus_addr = rxq->rx_ring_phys_addr;
3281 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(rxq->reg_idx),
3282 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
3283 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(rxq->reg_idx),
3284 (uint32_t)(bus_addr >> 32));
3285 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(rxq->reg_idx),
3286 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
3287 IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
3288 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), 0);
3290 /* Configure the SRRCTL register */
3291 #ifdef RTE_HEADER_SPLIT_ENABLE
3293 * Configure Header Split
3295 if (dev->data->dev_conf.rxmode.header_split) {
3296 if (hw->mac.type == ixgbe_mac_82599EB) {
3297 /* Must setup the PSRTYPE register */
3299 psrtype = IXGBE_PSRTYPE_TCPHDR |
3300 IXGBE_PSRTYPE_UDPHDR |
3301 IXGBE_PSRTYPE_IPV4HDR |
3302 IXGBE_PSRTYPE_IPV6HDR;
3303 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
3305 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
3306 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
3307 IXGBE_SRRCTL_BSIZEHDR_MASK);
3308 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
3311 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3313 /* Set if packets are dropped when no descriptors available */
3315 srrctl |= IXGBE_SRRCTL_DROP_EN;
3318 * Configure the RX buffer size in the BSIZEPACKET field of
3319 * the SRRCTL register of the queue.
3320 * The value is in 1 KB resolution. Valid values can be from
3323 mbp_priv = (struct rte_pktmbuf_pool_private *)
3324 ((char *)rxq->mb_pool + sizeof(struct rte_mempool));
3325 buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
3326 RTE_PKTMBUF_HEADROOM);
3327 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
3328 IXGBE_SRRCTL_BSIZEPKT_MASK);
3329 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
3331 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
3332 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
3334 /* It adds dual VLAN length for supporting dual VLAN */
3335 if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
3336 2 * IXGBE_VLAN_TAG_SIZE) > buf_size){
3337 dev->data->scattered_rx = 1;
3338 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
3343 * Device configured with multiple RX queues.
3345 ixgbe_dev_mq_rx_configure(dev);
3348 * Setup the Checksum Register.
3349 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
3350 * Enable IP/L4 checkum computation by hardware if requested to do so.
3352 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3353 rxcsum |= IXGBE_RXCSUM_PCSD;
3354 if (dev->data->dev_conf.rxmode.hw_ip_checksum)
3355 rxcsum |= IXGBE_RXCSUM_IPPCSE;
3357 rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
3359 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3361 if (hw->mac.type == ixgbe_mac_82599EB) {
3362 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3363 if (dev->data->dev_conf.rxmode.hw_strip_crc)
3364 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3366 rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
3367 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3368 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3375 * Initializes Transmit Unit.
3378 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
3380 struct ixgbe_hw *hw;
3381 struct igb_tx_queue *txq;
3387 PMD_INIT_FUNC_TRACE();
3388 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3390 /* Enable TX CRC (checksum offload requirement) */
3391 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3392 hlreg0 |= IXGBE_HLREG0_TXCRCEN;
3393 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3395 /* Setup the Base and Length of the Tx Descriptor Rings */
3396 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3397 txq = dev->data->tx_queues[i];
3399 bus_addr = txq->tx_ring_phys_addr;
3400 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(txq->reg_idx),
3401 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
3402 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(txq->reg_idx),
3403 (uint32_t)(bus_addr >> 32));
3404 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(txq->reg_idx),
3405 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
3406 /* Setup the HW Tx Head and TX Tail descriptor pointers */
3407 IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
3408 IXGBE_WRITE_REG(hw, IXGBE_TDT(txq->reg_idx), 0);
3411 * Disable Tx Head Writeback RO bit, since this hoses
3412 * bookkeeping if things aren't delivered in order.
3414 switch (hw->mac.type) {
3415 case ixgbe_mac_82598EB:
3416 txctrl = IXGBE_READ_REG(hw,
3417 IXGBE_DCA_TXCTRL(txq->reg_idx));
3418 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
3419 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(txq->reg_idx),
3423 case ixgbe_mac_82599EB:
3424 case ixgbe_mac_X540:
3426 txctrl = IXGBE_READ_REG(hw,
3427 IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
3428 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
3429 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(txq->reg_idx),
3435 /* Device configured with multiple TX queues. */
3436 ixgbe_dev_mq_tx_configure(dev);
3440 * Start Transmit and Receive Units.
3443 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
3445 struct ixgbe_hw *hw;
3446 struct igb_tx_queue *txq;
3447 struct igb_rx_queue *rxq;
3455 PMD_INIT_FUNC_TRACE();
3456 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3458 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3459 txq = dev->data->tx_queues[i];
3460 /* Setup Transmit Threshold Registers */
3461 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
3462 txdctl |= txq->pthresh & 0x7F;
3463 txdctl |= ((txq->hthresh & 0x7F) << 8);
3464 txdctl |= ((txq->wthresh & 0x7F) << 16);
3465 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
3468 if (hw->mac.type != ixgbe_mac_82598EB) {
3469 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3470 dmatxctl |= IXGBE_DMATXCTL_TE;
3471 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3474 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3475 txq = dev->data->tx_queues[i];
3476 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
3477 txdctl |= IXGBE_TXDCTL_ENABLE;
3478 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txq->reg_idx), txdctl);
3480 /* Wait until TX Enable ready */
3481 if (hw->mac.type == ixgbe_mac_82599EB) {
3485 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
3486 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
3488 PMD_INIT_LOG(ERR, "Could not enable "
3489 "Tx Queue %d\n", i);
3492 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3493 rxq = dev->data->rx_queues[i];
3494 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
3495 rxdctl |= IXGBE_RXDCTL_ENABLE;
3496 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxq->reg_idx), rxdctl);
3498 /* Wait until RX Enable ready */
3502 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
3503 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3505 PMD_INIT_LOG(ERR, "Could not enable "
3506 "Rx Queue %d\n", i);
3508 IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
3511 /* Enable Receive engine */
3512 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3513 if (hw->mac.type == ixgbe_mac_82598EB)
3514 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3515 rxctrl |= IXGBE_RXCTRL_RXEN;
3516 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3521 * [VF] Initializes Receive Unit.
3524 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
3526 struct ixgbe_hw *hw;
3527 struct igb_rx_queue *rxq;
3528 struct rte_pktmbuf_pool_private *mbp_priv;
3535 PMD_INIT_FUNC_TRACE();
3536 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3538 /* Setup RX queues */
3539 dev->rx_pkt_burst = ixgbe_recv_pkts;
3540 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3541 rxq = dev->data->rx_queues[i];
3543 /* Allocate buffers for descriptor rings */
3544 ret = ixgbe_alloc_rx_queue_mbufs(rxq);
3548 /* Setup the Base and Length of the Rx Descriptor Rings */
3549 bus_addr = rxq->rx_ring_phys_addr;
3551 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(i),
3552 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
3553 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(i),
3554 (uint32_t)(bus_addr >> 32));
3555 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(i),
3556 rxq->nb_rx_desc * sizeof(union ixgbe_adv_rx_desc));
3557 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(i), 0);
3558 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), 0);
3561 /* Configure the SRRCTL register */
3562 #ifdef RTE_HEADER_SPLIT_ENABLE
3564 * Configure Header Split
3566 if (dev->data->dev_conf.rxmode.header_split) {
3568 /* Must setup the PSRTYPE register */
3570 psrtype = IXGBE_PSRTYPE_TCPHDR |
3571 IXGBE_PSRTYPE_UDPHDR |
3572 IXGBE_PSRTYPE_IPV4HDR |
3573 IXGBE_PSRTYPE_IPV6HDR;
3575 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE(i), psrtype);
3577 srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
3578 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
3579 IXGBE_SRRCTL_BSIZEHDR_MASK);
3580 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
3583 srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3585 /* Set if packets are dropped when no descriptors available */
3587 srrctl |= IXGBE_SRRCTL_DROP_EN;
3590 * Configure the RX buffer size in the BSIZEPACKET field of
3591 * the SRRCTL register of the queue.
3592 * The value is in 1 KB resolution. Valid values can be from
3595 mbp_priv = (struct rte_pktmbuf_pool_private *)
3596 ((char *)rxq->mb_pool + sizeof(struct rte_mempool));
3597 buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
3598 RTE_PKTMBUF_HEADROOM);
3599 srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
3600 IXGBE_SRRCTL_BSIZEPKT_MASK);
3603 * VF modification to write virtual function SRRCTL register
3605 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), srrctl);
3607 buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
3608 IXGBE_SRRCTL_BSIZEPKT_SHIFT);
3610 /* It adds dual VLAN length for supporting dual VLAN */
3611 if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
3612 2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
3613 dev->data->scattered_rx = 1;
3614 dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
3622 * [VF] Initializes Transmit Unit.
3625 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
3627 struct ixgbe_hw *hw;
3628 struct igb_tx_queue *txq;
3633 PMD_INIT_FUNC_TRACE();
3634 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3636 /* Setup the Base and Length of the Tx Descriptor Rings */
3637 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3638 txq = dev->data->tx_queues[i];
3639 bus_addr = txq->tx_ring_phys_addr;
3640 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(i),
3641 (uint32_t)(bus_addr & 0x00000000ffffffffULL));
3642 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(i),
3643 (uint32_t)(bus_addr >> 32));
3644 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(i),
3645 txq->nb_tx_desc * sizeof(union ixgbe_adv_tx_desc));
3646 /* Setup the HW Tx Head and TX Tail descriptor pointers */
3647 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(i), 0);
3648 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(i), 0);
3651 * Disable Tx Head Writeback RO bit, since this hoses
3652 * bookkeeping if things aren't delivered in order.
3654 txctrl = IXGBE_READ_REG(hw,
3655 IXGBE_VFDCA_TXCTRL(i));
3656 txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN;
3657 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(i),
3663 * [VF] Start Transmit and Receive Units.
3666 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
3668 struct ixgbe_hw *hw;
3669 struct igb_tx_queue *txq;
3670 struct igb_rx_queue *rxq;
3676 PMD_INIT_FUNC_TRACE();
3677 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3679 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3680 txq = dev->data->tx_queues[i];
3681 /* Setup Transmit Threshold Registers */
3682 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
3683 txdctl |= txq->pthresh & 0x7F;
3684 txdctl |= ((txq->hthresh & 0x7F) << 8);
3685 txdctl |= ((txq->wthresh & 0x7F) << 16);
3686 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
3689 for (i = 0; i < dev->data->nb_tx_queues; i++) {
3691 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
3692 txdctl |= IXGBE_TXDCTL_ENABLE;
3693 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(i), txdctl);
3696 /* Wait until TX Enable ready */
3699 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
3700 } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
3702 PMD_INIT_LOG(ERR, "Could not enable "
3703 "Tx Queue %d\n", i);
3705 for (i = 0; i < dev->data->nb_rx_queues; i++) {
3707 rxq = dev->data->rx_queues[i];
3709 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
3710 rxdctl |= IXGBE_RXDCTL_ENABLE;
3711 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(i), rxdctl);
3713 /* Wait until RX Enable ready */
3717 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
3718 } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3720 PMD_INIT_LOG(ERR, "Could not enable "
3721 "Rx Queue %d\n", i);
3723 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);