4 * Copyright(c) 2010-2015 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.
34 #include <sys/queue.h>
45 #include <rte_byteorder.h>
46 #include <rte_common.h>
47 #include <rte_cycles.h>
49 #include <rte_debug.h>
50 #include <rte_interrupts.h>
52 #include <rte_memory.h>
53 #include <rte_memzone.h>
54 #include <rte_launch.h>
56 #include <rte_per_lcore.h>
57 #include <rte_lcore.h>
58 #include <rte_atomic.h>
59 #include <rte_branch_prediction.h>
61 #include <rte_mempool.h>
62 #include <rte_malloc.h>
64 #include <rte_ether.h>
65 #include <rte_ethdev.h>
66 #include <rte_prefetch.h>
71 #include <rte_string_fns.h>
72 #include <rte_errno.h>
74 #include "base/vmxnet3_defs.h"
75 #include "vmxnet3_ring.h"
77 #include "vmxnet3_logs.h"
78 #include "vmxnet3_ethdev.h"
80 static const uint32_t rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
82 static int vmxnet3_post_rx_bufs(vmxnet3_rx_queue_t*, uint8_t);
83 static void vmxnet3_tq_tx_complete(vmxnet3_tx_queue_t *);
84 #ifdef RTE_LIBRTE_VMXNET3_DEBUG_DRIVER_NOT_USED
85 static void vmxnet3_rxq_dump(struct vmxnet3_rx_queue *);
86 static void vmxnet3_txq_dump(struct vmxnet3_tx_queue *);
89 static 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, 0);
99 #ifdef RTE_LIBRTE_VMXNET3_DEBUG_DRIVER_NOT_USED
101 vmxnet3_rxq_dump(struct vmxnet3_rx_queue *rxq)
109 "RXQ: cmd0 base : 0x%p cmd1 base : 0x%p comp ring base : 0x%p.",
110 rxq->cmd_ring[0].base, rxq->cmd_ring[1].base, rxq->comp_ring.base);
112 "RXQ: cmd0 basePA : 0x%lx cmd1 basePA : 0x%lx comp ring basePA : 0x%lx.",
113 (unsigned long)rxq->cmd_ring[0].basePA,
114 (unsigned long)rxq->cmd_ring[1].basePA,
115 (unsigned long)rxq->comp_ring.basePA);
117 avail = vmxnet3_cmd_ring_desc_avail(&rxq->cmd_ring[0]);
119 "RXQ:cmd0: size=%u; free=%u; next2proc=%u; queued=%u",
120 (uint32_t)rxq->cmd_ring[0].size, avail,
121 rxq->comp_ring.next2proc,
122 rxq->cmd_ring[0].size - avail);
124 avail = vmxnet3_cmd_ring_desc_avail(&rxq->cmd_ring[1]);
125 PMD_RX_LOG(DEBUG, "RXQ:cmd1 size=%u; free=%u; next2proc=%u; queued=%u",
126 (uint32_t)rxq->cmd_ring[1].size, avail, rxq->comp_ring.next2proc,
127 rxq->cmd_ring[1].size - avail);
132 vmxnet3_txq_dump(struct vmxnet3_tx_queue *txq)
139 PMD_TX_LOG(DEBUG, "TXQ: cmd base : 0x%p comp ring base : 0x%p data ring base : 0x%p.",
140 txq->cmd_ring.base, txq->comp_ring.base, txq->data_ring.base);
141 PMD_TX_LOG(DEBUG, "TXQ: cmd basePA : 0x%lx comp ring basePA : 0x%lx data ring basePA : 0x%lx.",
142 (unsigned long)txq->cmd_ring.basePA,
143 (unsigned long)txq->comp_ring.basePA,
144 (unsigned long)txq->data_ring.basePA);
146 avail = vmxnet3_cmd_ring_desc_avail(&txq->cmd_ring);
147 PMD_TX_LOG(DEBUG, "TXQ: size=%u; free=%u; next2proc=%u; queued=%u",
148 (uint32_t)txq->cmd_ring.size, avail,
149 txq->comp_ring.next2proc, txq->cmd_ring.size - avail);
154 vmxnet3_cmd_ring_release_mbufs(vmxnet3_cmd_ring_t *ring)
156 while (ring->next2comp != ring->next2fill) {
157 /* No need to worry about tx desc ownership, device is quiesced by now. */
158 vmxnet3_buf_info_t *buf_info = ring->buf_info + ring->next2comp;
161 rte_pktmbuf_free(buf_info->m);
166 vmxnet3_cmd_ring_adv_next2comp(ring);
171 vmxnet3_cmd_ring_release(vmxnet3_cmd_ring_t *ring)
173 vmxnet3_cmd_ring_release_mbufs(ring);
174 rte_free(ring->buf_info);
175 ring->buf_info = NULL;
180 vmxnet3_dev_tx_queue_release(void *txq)
182 vmxnet3_tx_queue_t *tq = txq;
185 /* Release the cmd_ring */
186 vmxnet3_cmd_ring_release(&tq->cmd_ring);
191 vmxnet3_dev_rx_queue_release(void *rxq)
194 vmxnet3_rx_queue_t *rq = rxq;
197 /* Release both the cmd_rings */
198 for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++)
199 vmxnet3_cmd_ring_release(&rq->cmd_ring[i]);
204 vmxnet3_dev_tx_queue_reset(void *txq)
206 vmxnet3_tx_queue_t *tq = txq;
207 struct vmxnet3_cmd_ring *ring = &tq->cmd_ring;
208 struct vmxnet3_comp_ring *comp_ring = &tq->comp_ring;
209 struct vmxnet3_data_ring *data_ring = &tq->data_ring;
213 /* Release the cmd_ring mbufs */
214 vmxnet3_cmd_ring_release_mbufs(&tq->cmd_ring);
217 /* Tx vmxnet rings structure initialization*/
220 ring->gen = VMXNET3_INIT_GEN;
221 comp_ring->next2proc = 0;
222 comp_ring->gen = VMXNET3_INIT_GEN;
224 size = sizeof(struct Vmxnet3_TxDesc) * ring->size;
225 size += sizeof(struct Vmxnet3_TxCompDesc) * comp_ring->size;
226 size += sizeof(struct Vmxnet3_TxDataDesc) * data_ring->size;
228 memset(ring->base, 0, size);
232 vmxnet3_dev_rx_queue_reset(void *rxq)
235 vmxnet3_rx_queue_t *rq = rxq;
236 struct vmxnet3_cmd_ring *ring0, *ring1;
237 struct vmxnet3_comp_ring *comp_ring;
241 /* Release both the cmd_rings mbufs */
242 for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++)
243 vmxnet3_cmd_ring_release_mbufs(&rq->cmd_ring[i]);
246 ring0 = &rq->cmd_ring[0];
247 ring1 = &rq->cmd_ring[1];
248 comp_ring = &rq->comp_ring;
250 /* Rx vmxnet rings structure initialization */
251 ring0->next2fill = 0;
252 ring1->next2fill = 0;
253 ring0->next2comp = 0;
254 ring1->next2comp = 0;
255 ring0->gen = VMXNET3_INIT_GEN;
256 ring1->gen = VMXNET3_INIT_GEN;
257 comp_ring->next2proc = 0;
258 comp_ring->gen = VMXNET3_INIT_GEN;
260 size = sizeof(struct Vmxnet3_RxDesc) * (ring0->size + ring1->size);
261 size += sizeof(struct Vmxnet3_RxCompDesc) * comp_ring->size;
263 memset(ring0->base, 0, size);
267 vmxnet3_dev_clear_queues(struct rte_eth_dev *dev)
271 PMD_INIT_FUNC_TRACE();
273 for (i = 0; i < dev->data->nb_tx_queues; i++) {
274 struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
278 vmxnet3_dev_tx_queue_reset(txq);
282 for (i = 0; i < dev->data->nb_rx_queues; i++) {
283 struct vmxnet3_rx_queue *rxq = dev->data->rx_queues[i];
287 vmxnet3_dev_rx_queue_reset(rxq);
293 vmxnet3_unmap_pkt(uint16_t eop_idx, vmxnet3_tx_queue_t *txq)
296 struct rte_mbuf *mbuf;
298 /* Release cmd_ring descriptor and free mbuf */
299 RTE_ASSERT(txq->cmd_ring.base[eop_idx].txd.eop == 1);
301 mbuf = txq->cmd_ring.buf_info[eop_idx].m;
303 rte_panic("EOP desc does not point to a valid mbuf");
304 rte_pktmbuf_free(mbuf);
306 txq->cmd_ring.buf_info[eop_idx].m = NULL;
308 while (txq->cmd_ring.next2comp != eop_idx) {
309 /* no out-of-order completion */
310 RTE_ASSERT(txq->cmd_ring.base[txq->cmd_ring.next2comp].txd.cq == 0);
311 vmxnet3_cmd_ring_adv_next2comp(&txq->cmd_ring);
315 /* Mark the txd for which tcd was generated as completed */
316 vmxnet3_cmd_ring_adv_next2comp(&txq->cmd_ring);
318 return completed + 1;
322 vmxnet3_tq_tx_complete(vmxnet3_tx_queue_t *txq)
325 vmxnet3_comp_ring_t *comp_ring = &txq->comp_ring;
326 struct Vmxnet3_TxCompDesc *tcd = (struct Vmxnet3_TxCompDesc *)
327 (comp_ring->base + comp_ring->next2proc);
329 while (tcd->gen == comp_ring->gen) {
330 completed += vmxnet3_unmap_pkt(tcd->txdIdx, txq);
332 vmxnet3_comp_ring_adv_next2proc(comp_ring);
333 tcd = (struct Vmxnet3_TxCompDesc *)(comp_ring->base +
334 comp_ring->next2proc);
337 PMD_TX_LOG(DEBUG, "Processed %d tx comps & command descs.", completed);
341 vmxnet3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
345 vmxnet3_tx_queue_t *txq = tx_queue;
346 struct vmxnet3_hw *hw = txq->hw;
347 Vmxnet3_TxQueueCtrl *txq_ctrl = &txq->shared->ctrl;
348 uint32_t deferred = rte_le_to_cpu_32(txq_ctrl->txNumDeferred);
350 if (unlikely(txq->stopped)) {
351 PMD_TX_LOG(DEBUG, "Tx queue is stopped.");
355 /* Free up the comp_descriptors aggressively */
356 vmxnet3_tq_tx_complete(txq);
359 while (nb_tx < nb_pkts) {
360 Vmxnet3_GenericDesc *gdesc;
361 vmxnet3_buf_info_t *tbi;
362 uint32_t first2fill, avail, dw2;
363 struct rte_mbuf *txm = tx_pkts[nb_tx];
364 struct rte_mbuf *m_seg = txm;
366 bool tso = (txm->ol_flags & PKT_TX_TCP_SEG) != 0;
367 /* # of descriptors needed for a packet. */
368 unsigned count = txm->nb_segs;
370 avail = vmxnet3_cmd_ring_desc_avail(&txq->cmd_ring);
372 /* Is command ring full? */
373 if (unlikely(avail == 0)) {
374 PMD_TX_LOG(DEBUG, "No free ring descriptors");
375 txq->stats.tx_ring_full++;
376 txq->stats.drop_total += (nb_pkts - nb_tx);
380 /* Command ring is not full but cannot handle the
381 * multi-segmented packet. Let's try the next packet
384 PMD_TX_LOG(DEBUG, "Running out of ring descriptors "
385 "(avail %d needed %d)", avail, count);
386 txq->stats.drop_total++;
388 txq->stats.drop_tso++;
389 rte_pktmbuf_free(txm);
394 /* Drop non-TSO packet that is excessively fragmented */
395 if (unlikely(!tso && count > VMXNET3_MAX_TXD_PER_PKT)) {
396 PMD_TX_LOG(ERR, "Non-TSO packet cannot occupy more than %d tx "
397 "descriptors. Packet dropped.", VMXNET3_MAX_TXD_PER_PKT);
398 txq->stats.drop_too_many_segs++;
399 txq->stats.drop_total++;
400 rte_pktmbuf_free(txm);
405 if (txm->nb_segs == 1 && rte_pktmbuf_pkt_len(txm) <= VMXNET3_HDR_COPY_SIZE) {
406 struct Vmxnet3_TxDataDesc *tdd;
408 tdd = txq->data_ring.base + txq->cmd_ring.next2fill;
409 copy_size = rte_pktmbuf_pkt_len(txm);
410 rte_memcpy(tdd->data, rte_pktmbuf_mtod(txm, char *), copy_size);
413 /* use the previous gen bit for the SOP desc */
414 dw2 = (txq->cmd_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
415 first2fill = txq->cmd_ring.next2fill;
417 /* Remember the transmit buffer for cleanup */
418 tbi = txq->cmd_ring.buf_info + txq->cmd_ring.next2fill;
420 /* NB: the following assumes that VMXNET3 maximum
421 * transmit buffer size (16K) is greater than
422 * maximum size of mbuf segment size.
424 gdesc = txq->cmd_ring.base + txq->cmd_ring.next2fill;
426 gdesc->txd.addr = rte_cpu_to_le_64(txq->data_ring.basePA +
427 txq->cmd_ring.next2fill *
428 sizeof(struct Vmxnet3_TxDataDesc));
430 gdesc->txd.addr = rte_mbuf_data_dma_addr(m_seg);
432 gdesc->dword[2] = dw2 | m_seg->data_len;
435 /* move to the next2fill descriptor */
436 vmxnet3_cmd_ring_adv_next2fill(&txq->cmd_ring);
438 /* use the right gen for non-SOP desc */
439 dw2 = txq->cmd_ring.gen << VMXNET3_TXD_GEN_SHIFT;
440 } while ((m_seg = m_seg->next) != NULL);
442 /* set the last buf_info for the pkt */
444 /* Update the EOP descriptor */
445 gdesc->dword[3] |= VMXNET3_TXD_EOP | VMXNET3_TXD_CQ;
447 /* Add VLAN tag if present */
448 gdesc = txq->cmd_ring.base + first2fill;
449 if (txm->ol_flags & PKT_TX_VLAN_PKT) {
451 gdesc->txd.tci = txm->vlan_tci;
455 uint16_t mss = txm->tso_segsz;
459 gdesc->txd.hlen = txm->l2_len + txm->l3_len + txm->l4_len;
460 gdesc->txd.om = VMXNET3_OM_TSO;
461 gdesc->txd.msscof = mss;
463 deferred += (rte_pktmbuf_pkt_len(txm) - gdesc->txd.hlen + mss - 1) / mss;
464 } else if (txm->ol_flags & PKT_TX_L4_MASK) {
465 gdesc->txd.om = VMXNET3_OM_CSUM;
466 gdesc->txd.hlen = txm->l2_len + txm->l3_len;
468 switch (txm->ol_flags & PKT_TX_L4_MASK) {
469 case PKT_TX_TCP_CKSUM:
470 gdesc->txd.msscof = gdesc->txd.hlen + offsetof(struct tcp_hdr, cksum);
472 case PKT_TX_UDP_CKSUM:
473 gdesc->txd.msscof = gdesc->txd.hlen + offsetof(struct udp_hdr, dgram_cksum);
476 PMD_TX_LOG(WARNING, "requested cksum offload not supported %#llx",
477 txm->ol_flags & PKT_TX_L4_MASK);
483 gdesc->txd.om = VMXNET3_OM_NONE;
484 gdesc->txd.msscof = 0;
488 /* flip the GEN bit on the SOP */
489 rte_compiler_barrier();
490 gdesc->dword[2] ^= VMXNET3_TXD_GEN;
492 txq_ctrl->txNumDeferred = rte_cpu_to_le_32(deferred);
496 PMD_TX_LOG(DEBUG, "vmxnet3 txThreshold: %u", rte_le_to_cpu_32(txq_ctrl->txThreshold));
498 if (deferred >= rte_le_to_cpu_32(txq_ctrl->txThreshold)) {
499 txq_ctrl->txNumDeferred = 0;
500 /* Notify vSwitch that packets are available. */
501 VMXNET3_WRITE_BAR0_REG(hw, (VMXNET3_REG_TXPROD + txq->queue_id * VMXNET3_REG_ALIGN),
502 txq->cmd_ring.next2fill);
509 * Allocates mbufs and clusters. Post rx descriptors with buffer details
510 * so that device can receive packets in those buffers.
512 * Among the two rings, 1st ring contains buffers of type 0 and type1.
513 * bufs_per_pkt is set such that for non-LRO cases all the buffers required
514 * by a frame will fit in 1st ring (1st buf of type0 and rest of type1).
515 * 2nd ring contains buffers of type 1 alone. Second ring mostly be used
520 vmxnet3_post_rx_bufs(vmxnet3_rx_queue_t *rxq, uint8_t ring_id)
523 uint32_t i = 0, val = 0;
524 struct vmxnet3_cmd_ring *ring = &rxq->cmd_ring[ring_id];
527 /* Usually: One HEAD type buf per packet
528 * val = (ring->next2fill % rxq->hw->bufs_per_pkt) ?
529 * VMXNET3_RXD_BTYPE_BODY : VMXNET3_RXD_BTYPE_HEAD;
532 /* We use single packet buffer so all heads here */
533 val = VMXNET3_RXD_BTYPE_HEAD;
535 /* All BODY type buffers for 2nd ring */
536 val = VMXNET3_RXD_BTYPE_BODY;
539 while (vmxnet3_cmd_ring_desc_avail(ring) > 0) {
540 struct Vmxnet3_RxDesc *rxd;
541 struct rte_mbuf *mbuf;
542 vmxnet3_buf_info_t *buf_info = &ring->buf_info[ring->next2fill];
544 rxd = (struct Vmxnet3_RxDesc *)(ring->base + ring->next2fill);
546 /* Allocate blank mbuf for the current Rx Descriptor */
547 mbuf = rte_rxmbuf_alloc(rxq->mp);
548 if (unlikely(mbuf == NULL)) {
549 PMD_RX_LOG(ERR, "Error allocating mbuf");
550 rxq->stats.rx_buf_alloc_failure++;
556 * Load mbuf pointer into buf_info[ring_size]
557 * buf_info structure is equivalent to cookie for virtio-virtqueue
560 buf_info->len = (uint16_t)(mbuf->buf_len -
561 RTE_PKTMBUF_HEADROOM);
563 rte_mbuf_data_dma_addr_default(mbuf);
565 /* Load Rx Descriptor with the buffer's GPA */
566 rxd->addr = buf_info->bufPA;
568 /* After this point rxd->addr MUST not be NULL */
570 rxd->len = buf_info->len;
571 /* Flip gen bit at the end to change ownership */
572 rxd->gen = ring->gen;
574 vmxnet3_cmd_ring_adv_next2fill(ring);
578 /* Return error only if no buffers are posted at present */
579 if (vmxnet3_cmd_ring_desc_avail(ring) >= (ring->size - 1))
586 /* Receive side checksum and other offloads */
588 vmxnet3_rx_offload(const Vmxnet3_RxCompDesc *rcd, struct rte_mbuf *rxm)
590 /* Check for hardware stripped VLAN tag */
592 rxm->ol_flags |= PKT_RX_VLAN_PKT;
593 rxm->vlan_tci = rte_le_to_cpu_16((uint16_t)rcd->tci);
597 if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE) {
598 rxm->ol_flags |= PKT_RX_RSS_HASH;
599 rxm->hash.rss = rcd->rssHash;
602 /* Check packet type, checksum errors, etc. Only support IPv4 for now. */
604 struct ether_hdr *eth = rte_pktmbuf_mtod(rxm, struct ether_hdr *);
605 struct ipv4_hdr *ip = (struct ipv4_hdr *)(eth + 1);
607 if (((ip->version_ihl & 0xf) << 2) > (int)sizeof(struct ipv4_hdr))
608 rxm->packet_type = RTE_PTYPE_L3_IPV4_EXT;
610 rxm->packet_type = RTE_PTYPE_L3_IPV4;
614 rxm->ol_flags |= PKT_RX_IP_CKSUM_BAD;
616 if ((rcd->tcp || rcd->udp) && !rcd->tuc)
617 rxm->ol_flags |= PKT_RX_L4_CKSUM_BAD;
623 * Process the Rx Completion Ring of given vmxnet3_rx_queue
624 * for nb_pkts burst and return the number of packets received
627 vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
630 uint32_t nb_rxd, idx;
632 vmxnet3_rx_queue_t *rxq;
633 Vmxnet3_RxCompDesc *rcd;
634 vmxnet3_buf_info_t *rbi;
636 struct rte_mbuf *rxm = NULL;
637 struct vmxnet3_hw *hw;
647 rcd = &rxq->comp_ring.base[rxq->comp_ring.next2proc].rcd;
649 if (unlikely(rxq->stopped)) {
650 PMD_RX_LOG(DEBUG, "Rx queue is stopped.");
654 while (rcd->gen == rxq->comp_ring.gen) {
655 if (nb_rx >= nb_pkts)
659 ring_idx = (uint8_t)((rcd->rqID == rxq->qid1) ? 0 : 1);
660 rxd = (Vmxnet3_RxDesc *)rxq->cmd_ring[ring_idx].base + idx;
661 RTE_SET_USED(rxd); /* used only for assert when enabled */
662 rbi = rxq->cmd_ring[ring_idx].buf_info + idx;
664 PMD_RX_LOG(DEBUG, "rxd idx: %d ring idx: %d.", idx, ring_idx);
666 RTE_ASSERT(rcd->len <= rxd->len);
669 /* Get the packet buffer pointer from buf_info */
672 /* Clear descriptor associated buf_info to be reused */
676 /* Update the index that we received a packet */
677 rxq->cmd_ring[ring_idx].next2comp = idx;
679 /* For RCD with EOP set, check if there is frame error */
680 if (unlikely(rcd->eop && rcd->err)) {
681 rxq->stats.drop_total++;
682 rxq->stats.drop_err++;
685 rxq->stats.drop_fcs++;
686 PMD_RX_LOG(ERR, "Recv packet dropped due to frame err.");
688 PMD_RX_LOG(ERR, "Error in received packet rcd#:%d rxd:%d",
689 (int)(rcd - (struct Vmxnet3_RxCompDesc *)
690 rxq->comp_ring.base), rcd->rxdIdx);
691 rte_pktmbuf_free_seg(rxm);
696 /* Initialize newly received packet buffer */
697 rxm->port = rxq->port_id;
700 rxm->pkt_len = (uint16_t)rcd->len;
701 rxm->data_len = (uint16_t)rcd->len;
702 rxm->data_off = RTE_PKTMBUF_HEADROOM;
707 * If this is the first buffer of the received packet,
708 * set the pointer to the first mbuf of the packet
709 * Otherwise, update the total length and the number of segments
710 * of the current scattered packet, and update the pointer to
711 * the last mbuf of the current packet.
714 RTE_ASSERT(rxd->btype == VMXNET3_RXD_BTYPE_HEAD);
716 if (unlikely(rcd->len == 0)) {
717 RTE_ASSERT(rcd->eop);
720 "Rx buf was skipped. rxring[%d][%d])",
722 rte_pktmbuf_free_seg(rxm);
726 rxq->start_seg = rxm;
727 vmxnet3_rx_offload(rcd, rxm);
729 struct rte_mbuf *start = rxq->start_seg;
731 RTE_ASSERT(rxd->btype == VMXNET3_RXD_BTYPE_BODY);
733 start->pkt_len += rxm->data_len;
736 rxq->last_seg->next = rxm;
741 rx_pkts[nb_rx++] = rxq->start_seg;
742 rxq->start_seg = NULL;
746 rxq->cmd_ring[ring_idx].next2comp = idx;
747 VMXNET3_INC_RING_IDX_ONLY(rxq->cmd_ring[ring_idx].next2comp, rxq->cmd_ring[ring_idx].size);
749 /* It's time to allocate some new buf and renew descriptors */
750 vmxnet3_post_rx_bufs(rxq, ring_idx);
751 if (unlikely(rxq->shared->ctrl.updateRxProd)) {
752 VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[ring_idx] + (rxq->queue_id * VMXNET3_REG_ALIGN),
753 rxq->cmd_ring[ring_idx].next2fill);
756 /* Advance to the next descriptor in comp_ring */
757 vmxnet3_comp_ring_adv_next2proc(&rxq->comp_ring);
759 rcd = &rxq->comp_ring.base[rxq->comp_ring.next2proc].rcd;
761 if (nb_rxd > rxq->cmd_ring[0].size) {
763 "Used up quota of receiving packets,"
764 " relinquish control.");
773 * Create memzone for device rings. malloc can't be used as the physical address is
774 * needed. If the memzone is already created, then this function returns a ptr
777 static const struct rte_memzone *
778 ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
779 uint16_t queue_id, uint32_t ring_size, int socket_id)
781 char z_name[RTE_MEMZONE_NAMESIZE];
782 const struct rte_memzone *mz;
784 snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
785 dev->driver->pci_drv.name, ring_name,
786 dev->data->port_id, queue_id);
788 mz = rte_memzone_lookup(z_name);
792 return rte_memzone_reserve_aligned(z_name, ring_size,
793 socket_id, 0, VMXNET3_RING_BA_ALIGN);
797 vmxnet3_dev_tx_queue_setup(struct rte_eth_dev *dev,
800 unsigned int socket_id,
801 __attribute__((unused)) const struct rte_eth_txconf *tx_conf)
803 struct vmxnet3_hw *hw = dev->data->dev_private;
804 const struct rte_memzone *mz;
805 struct vmxnet3_tx_queue *txq;
806 struct vmxnet3_cmd_ring *ring;
807 struct vmxnet3_comp_ring *comp_ring;
808 struct vmxnet3_data_ring *data_ring;
811 PMD_INIT_FUNC_TRACE();
813 if ((tx_conf->txq_flags & ETH_TXQ_FLAGS_NOXSUMSCTP) !=
814 ETH_TXQ_FLAGS_NOXSUMSCTP) {
815 PMD_INIT_LOG(ERR, "SCTP checksum offload not supported");
819 txq = rte_zmalloc("ethdev_tx_queue", sizeof(struct vmxnet3_tx_queue), RTE_CACHE_LINE_SIZE);
821 PMD_INIT_LOG(ERR, "Can not allocate tx queue structure");
825 txq->queue_id = queue_idx;
826 txq->port_id = dev->data->port_id;
827 txq->shared = &hw->tqd_start[queue_idx];
829 txq->qid = queue_idx;
832 ring = &txq->cmd_ring;
833 comp_ring = &txq->comp_ring;
834 data_ring = &txq->data_ring;
836 /* Tx vmxnet ring length should be between 512-4096 */
837 if (nb_desc < VMXNET3_DEF_TX_RING_SIZE) {
838 PMD_INIT_LOG(ERR, "VMXNET3 Tx Ring Size Min: %u",
839 VMXNET3_DEF_TX_RING_SIZE);
841 } else if (nb_desc > VMXNET3_TX_RING_MAX_SIZE) {
842 PMD_INIT_LOG(ERR, "VMXNET3 Tx Ring Size Max: %u",
843 VMXNET3_TX_RING_MAX_SIZE);
846 ring->size = nb_desc;
847 ring->size &= ~VMXNET3_RING_SIZE_MASK;
849 comp_ring->size = data_ring->size = ring->size;
851 /* Tx vmxnet rings structure initialization*/
854 ring->gen = VMXNET3_INIT_GEN;
855 comp_ring->next2proc = 0;
856 comp_ring->gen = VMXNET3_INIT_GEN;
858 size = sizeof(struct Vmxnet3_TxDesc) * ring->size;
859 size += sizeof(struct Vmxnet3_TxCompDesc) * comp_ring->size;
860 size += sizeof(struct Vmxnet3_TxDataDesc) * data_ring->size;
862 mz = ring_dma_zone_reserve(dev, "txdesc", queue_idx, size, socket_id);
864 PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
867 memset(mz->addr, 0, mz->len);
869 /* cmd_ring initialization */
870 ring->base = mz->addr;
871 ring->basePA = mz->phys_addr;
873 /* comp_ring initialization */
874 comp_ring->base = ring->base + ring->size;
875 comp_ring->basePA = ring->basePA +
876 (sizeof(struct Vmxnet3_TxDesc) * ring->size);
878 /* data_ring initialization */
879 data_ring->base = (Vmxnet3_TxDataDesc *)(comp_ring->base + comp_ring->size);
880 data_ring->basePA = comp_ring->basePA +
881 (sizeof(struct Vmxnet3_TxCompDesc) * comp_ring->size);
883 /* cmd_ring0 buf_info allocation */
884 ring->buf_info = rte_zmalloc("tx_ring_buf_info",
885 ring->size * sizeof(vmxnet3_buf_info_t), RTE_CACHE_LINE_SIZE);
886 if (ring->buf_info == NULL) {
887 PMD_INIT_LOG(ERR, "ERROR: Creating tx_buf_info structure");
891 /* Update the data portion with txq */
892 dev->data->tx_queues[queue_idx] = txq;
898 vmxnet3_dev_rx_queue_setup(struct rte_eth_dev *dev,
901 unsigned int socket_id,
902 __attribute__((unused)) const struct rte_eth_rxconf *rx_conf,
903 struct rte_mempool *mp)
905 const struct rte_memzone *mz;
906 struct vmxnet3_rx_queue *rxq;
907 struct vmxnet3_hw *hw = dev->data->dev_private;
908 struct vmxnet3_cmd_ring *ring0, *ring1, *ring;
909 struct vmxnet3_comp_ring *comp_ring;
914 PMD_INIT_FUNC_TRACE();
916 rxq = rte_zmalloc("ethdev_rx_queue", sizeof(struct vmxnet3_rx_queue), RTE_CACHE_LINE_SIZE);
918 PMD_INIT_LOG(ERR, "Can not allocate rx queue structure");
923 rxq->queue_id = queue_idx;
924 rxq->port_id = dev->data->port_id;
925 rxq->shared = &hw->rqd_start[queue_idx];
927 rxq->qid1 = queue_idx;
928 rxq->qid2 = queue_idx + hw->num_rx_queues;
931 ring0 = &rxq->cmd_ring[0];
932 ring1 = &rxq->cmd_ring[1];
933 comp_ring = &rxq->comp_ring;
935 /* Rx vmxnet rings length should be between 256-4096 */
936 if (nb_desc < VMXNET3_DEF_RX_RING_SIZE) {
937 PMD_INIT_LOG(ERR, "VMXNET3 Rx Ring Size Min: 256");
939 } else if (nb_desc > VMXNET3_RX_RING_MAX_SIZE) {
940 PMD_INIT_LOG(ERR, "VMXNET3 Rx Ring Size Max: 4096");
943 ring0->size = nb_desc;
944 ring0->size &= ~VMXNET3_RING_SIZE_MASK;
945 ring1->size = ring0->size;
948 comp_ring->size = ring0->size + ring1->size;
950 /* Rx vmxnet rings structure initialization */
951 ring0->next2fill = 0;
952 ring1->next2fill = 0;
953 ring0->next2comp = 0;
954 ring1->next2comp = 0;
955 ring0->gen = VMXNET3_INIT_GEN;
956 ring1->gen = VMXNET3_INIT_GEN;
957 comp_ring->next2proc = 0;
958 comp_ring->gen = VMXNET3_INIT_GEN;
960 size = sizeof(struct Vmxnet3_RxDesc) * (ring0->size + ring1->size);
961 size += sizeof(struct Vmxnet3_RxCompDesc) * comp_ring->size;
963 mz = ring_dma_zone_reserve(dev, "rxdesc", queue_idx, size, socket_id);
965 PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
968 memset(mz->addr, 0, mz->len);
970 /* cmd_ring0 initialization */
971 ring0->base = mz->addr;
972 ring0->basePA = mz->phys_addr;
974 /* cmd_ring1 initialization */
975 ring1->base = ring0->base + ring0->size;
976 ring1->basePA = ring0->basePA + sizeof(struct Vmxnet3_RxDesc) * ring0->size;
978 /* comp_ring initialization */
979 comp_ring->base = ring1->base + ring1->size;
980 comp_ring->basePA = ring1->basePA + sizeof(struct Vmxnet3_RxDesc) *
983 /* cmd_ring0-cmd_ring1 buf_info allocation */
984 for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++) {
986 ring = &rxq->cmd_ring[i];
988 snprintf(mem_name, sizeof(mem_name), "rx_ring_%d_buf_info", i);
990 ring->buf_info = rte_zmalloc(mem_name, ring->size * sizeof(vmxnet3_buf_info_t), RTE_CACHE_LINE_SIZE);
991 if (ring->buf_info == NULL) {
992 PMD_INIT_LOG(ERR, "ERROR: Creating rx_buf_info structure");
997 /* Update the data portion with rxq */
998 dev->data->rx_queues[queue_idx] = rxq;
1004 * Initializes Receive Unit
1005 * Load mbufs in rx queue in advance
1008 vmxnet3_dev_rxtx_init(struct rte_eth_dev *dev)
1010 struct vmxnet3_hw *hw = dev->data->dev_private;
1015 PMD_INIT_FUNC_TRACE();
1017 for (i = 0; i < hw->num_rx_queues; i++) {
1018 vmxnet3_rx_queue_t *rxq = dev->data->rx_queues[i];
1020 for (j = 0; j < VMXNET3_RX_CMDRING_SIZE; j++) {
1021 /* Passing 0 as alloc_num will allocate full ring */
1022 ret = vmxnet3_post_rx_bufs(rxq, j);
1024 PMD_INIT_LOG(ERR, "ERROR: Posting Rxq: %d buffers ring: %d", i, j);
1027 /* Updating device with the index:next2fill to fill the mbufs for coming packets */
1028 if (unlikely(rxq->shared->ctrl.updateRxProd)) {
1029 VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[j] + (rxq->queue_id * VMXNET3_REG_ALIGN),
1030 rxq->cmd_ring[j].next2fill);
1033 rxq->stopped = FALSE;
1034 rxq->start_seg = NULL;
1037 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1038 struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
1040 txq->stopped = FALSE;
1046 static uint8_t rss_intel_key[40] = {
1047 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
1048 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
1049 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
1050 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
1051 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
1055 * Configure RSS feature
1058 vmxnet3_rss_configure(struct rte_eth_dev *dev)
1060 struct vmxnet3_hw *hw = dev->data->dev_private;
1061 struct VMXNET3_RSSConf *dev_rss_conf;
1062 struct rte_eth_rss_conf *port_rss_conf;
1066 PMD_INIT_FUNC_TRACE();
1068 dev_rss_conf = hw->rss_conf;
1069 port_rss_conf = &dev->data->dev_conf.rx_adv_conf.rss_conf;
1071 /* loading hashFunc */
1072 dev_rss_conf->hashFunc = VMXNET3_RSS_HASH_FUNC_TOEPLITZ;
1073 /* loading hashKeySize */
1074 dev_rss_conf->hashKeySize = VMXNET3_RSS_MAX_KEY_SIZE;
1075 /* loading indTableSize : Must not exceed VMXNET3_RSS_MAX_IND_TABLE_SIZE (128)*/
1076 dev_rss_conf->indTableSize = (uint16_t)(hw->num_rx_queues * 4);
1078 if (port_rss_conf->rss_key == NULL) {
1079 /* Default hash key */
1080 port_rss_conf->rss_key = rss_intel_key;
1083 /* loading hashKey */
1084 memcpy(&dev_rss_conf->hashKey[0], port_rss_conf->rss_key, dev_rss_conf->hashKeySize);
1086 /* loading indTable */
1087 for (i = 0, j = 0; i < dev_rss_conf->indTableSize; i++, j++) {
1088 if (j == dev->data->nb_rx_queues)
1090 dev_rss_conf->indTable[i] = j;
1093 /* loading hashType */
1094 dev_rss_conf->hashType = 0;
1095 rss_hf = port_rss_conf->rss_hf & VMXNET3_RSS_OFFLOAD_ALL;
1096 if (rss_hf & ETH_RSS_IPV4)
1097 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_IPV4;
1098 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
1099 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_TCP_IPV4;
1100 if (rss_hf & ETH_RSS_IPV6)
1101 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_IPV6;
1102 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
1103 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_TCP_IPV6;
1105 return VMXNET3_SUCCESS;