1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell RVU Ethernet driver
4 * Copyright (C) 2020 Marvell.
8 #include <linux/interrupt.h>
10 #include <net/page_pool/helpers.h>
12 #include <linux/bitfield.h>
15 #include "otx2_common.h"
16 #include "otx2_struct.h"
19 static void otx2_nix_rq_op_stats(struct queue_stats *stats,
20 struct otx2_nic *pfvf, int qidx)
22 u64 incr = (u64)qidx << 32;
25 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_OCTS);
26 stats->bytes = otx2_atomic64_add(incr, ptr);
28 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_PKTS);
29 stats->pkts = otx2_atomic64_add(incr, ptr);
32 static void otx2_nix_sq_op_stats(struct queue_stats *stats,
33 struct otx2_nic *pfvf, int qidx)
35 u64 incr = (u64)qidx << 32;
38 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_OCTS);
39 stats->bytes = otx2_atomic64_add(incr, ptr);
41 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_PKTS);
42 stats->pkts = otx2_atomic64_add(incr, ptr);
45 void otx2_update_lmac_stats(struct otx2_nic *pfvf)
49 if (!netif_running(pfvf->netdev))
52 mutex_lock(&pfvf->mbox.lock);
53 req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
55 mutex_unlock(&pfvf->mbox.lock);
59 otx2_sync_mbox_msg(&pfvf->mbox);
60 mutex_unlock(&pfvf->mbox.lock);
63 void otx2_update_lmac_fec_stats(struct otx2_nic *pfvf)
67 if (!netif_running(pfvf->netdev))
69 mutex_lock(&pfvf->mbox.lock);
70 req = otx2_mbox_alloc_msg_cgx_fec_stats(&pfvf->mbox);
72 otx2_sync_mbox_msg(&pfvf->mbox);
73 mutex_unlock(&pfvf->mbox.lock);
76 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
78 struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
83 otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
87 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
89 struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
94 if (qidx >= pfvf->hw.non_qos_queues) {
95 if (!test_bit(qidx - pfvf->hw.non_qos_queues, pfvf->qos.qos_sq_bmap))
99 otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
103 void otx2_get_dev_stats(struct otx2_nic *pfvf)
105 struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
107 dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
108 dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
109 dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
110 dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
111 dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
112 dev_stats->rx_frames = dev_stats->rx_bcast_frames +
113 dev_stats->rx_mcast_frames +
114 dev_stats->rx_ucast_frames;
116 dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
117 dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
118 dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
119 dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
120 dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
121 dev_stats->tx_frames = dev_stats->tx_bcast_frames +
122 dev_stats->tx_mcast_frames +
123 dev_stats->tx_ucast_frames;
126 void otx2_get_stats64(struct net_device *netdev,
127 struct rtnl_link_stats64 *stats)
129 struct otx2_nic *pfvf = netdev_priv(netdev);
130 struct otx2_dev_stats *dev_stats;
132 otx2_get_dev_stats(pfvf);
134 dev_stats = &pfvf->hw.dev_stats;
135 stats->rx_bytes = dev_stats->rx_bytes;
136 stats->rx_packets = dev_stats->rx_frames;
137 stats->rx_dropped = dev_stats->rx_drops;
138 stats->multicast = dev_stats->rx_mcast_frames;
140 stats->tx_bytes = dev_stats->tx_bytes;
141 stats->tx_packets = dev_stats->tx_frames;
142 stats->tx_dropped = dev_stats->tx_drops;
144 EXPORT_SYMBOL(otx2_get_stats64);
146 /* Sync MAC address with RVU AF */
147 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
149 struct nix_set_mac_addr *req;
152 mutex_lock(&pfvf->mbox.lock);
153 req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
155 mutex_unlock(&pfvf->mbox.lock);
159 ether_addr_copy(req->mac_addr, mac);
161 err = otx2_sync_mbox_msg(&pfvf->mbox);
162 mutex_unlock(&pfvf->mbox.lock);
166 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
167 struct net_device *netdev)
169 struct nix_get_mac_addr_rsp *rsp;
170 struct mbox_msghdr *msghdr;
174 mutex_lock(&pfvf->mbox.lock);
175 req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
177 mutex_unlock(&pfvf->mbox.lock);
181 err = otx2_sync_mbox_msg(&pfvf->mbox);
183 mutex_unlock(&pfvf->mbox.lock);
187 msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
188 if (IS_ERR(msghdr)) {
189 mutex_unlock(&pfvf->mbox.lock);
190 return PTR_ERR(msghdr);
192 rsp = (struct nix_get_mac_addr_rsp *)msghdr;
193 eth_hw_addr_set(netdev, rsp->mac_addr);
194 mutex_unlock(&pfvf->mbox.lock);
199 int otx2_set_mac_address(struct net_device *netdev, void *p)
201 struct otx2_nic *pfvf = netdev_priv(netdev);
202 struct sockaddr *addr = p;
204 if (!is_valid_ether_addr(addr->sa_data))
205 return -EADDRNOTAVAIL;
207 if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data)) {
208 eth_hw_addr_set(netdev, addr->sa_data);
209 /* update dmac field in vlan offload rule */
210 if (netif_running(netdev) &&
211 pfvf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
212 otx2_install_rxvlan_offload_flow(pfvf);
213 /* update dmac address in ntuple and DMAC filter list */
214 if (pfvf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
215 otx2_dmacflt_update_pfmac_flow(pfvf);
222 EXPORT_SYMBOL(otx2_set_mac_address);
224 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
226 struct nix_frs_cfg *req;
230 maxlen = otx2_get_max_mtu(pfvf) + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
232 mutex_lock(&pfvf->mbox.lock);
233 req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
235 mutex_unlock(&pfvf->mbox.lock);
239 req->maxlen = pfvf->netdev->mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
241 /* Use max receive length supported by hardware for loopback devices */
242 if (is_otx2_lbkvf(pfvf->pdev))
243 req->maxlen = maxlen;
245 err = otx2_sync_mbox_msg(&pfvf->mbox);
246 mutex_unlock(&pfvf->mbox.lock);
250 int otx2_config_pause_frm(struct otx2_nic *pfvf)
252 struct cgx_pause_frm_cfg *req;
255 if (is_otx2_lbkvf(pfvf->pdev))
258 mutex_lock(&pfvf->mbox.lock);
259 req = otx2_mbox_alloc_msg_cgx_cfg_pause_frm(&pfvf->mbox);
265 req->rx_pause = !!(pfvf->flags & OTX2_FLAG_RX_PAUSE_ENABLED);
266 req->tx_pause = !!(pfvf->flags & OTX2_FLAG_TX_PAUSE_ENABLED);
269 err = otx2_sync_mbox_msg(&pfvf->mbox);
271 mutex_unlock(&pfvf->mbox.lock);
274 EXPORT_SYMBOL(otx2_config_pause_frm);
276 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
278 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
279 struct nix_rss_flowkey_cfg_rsp *rsp;
280 struct nix_rss_flowkey_cfg *req;
283 mutex_lock(&pfvf->mbox.lock);
284 req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
286 mutex_unlock(&pfvf->mbox.lock);
289 req->mcam_index = -1; /* Default or reserved index */
290 req->flowkey_cfg = rss->flowkey_cfg;
291 req->group = DEFAULT_RSS_CONTEXT_GROUP;
293 err = otx2_sync_mbox_msg(&pfvf->mbox);
297 rsp = (struct nix_rss_flowkey_cfg_rsp *)
298 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
304 pfvf->hw.flowkey_alg_idx = rsp->alg_idx;
306 mutex_unlock(&pfvf->mbox.lock);
310 int otx2_set_rss_table(struct otx2_nic *pfvf, int ctx_id)
312 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
313 const int index = rss->rss_size * ctx_id;
314 struct mbox *mbox = &pfvf->mbox;
315 struct otx2_rss_ctx *rss_ctx;
316 struct nix_aq_enq_req *aq;
319 mutex_lock(&mbox->lock);
320 rss_ctx = rss->rss_ctx[ctx_id];
321 /* Get memory to put this msg */
322 for (idx = 0; idx < rss->rss_size; idx++) {
323 aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
325 /* The shared memory buffer can be full.
328 err = otx2_sync_mbox_msg(mbox);
330 mutex_unlock(&mbox->lock);
333 aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
335 mutex_unlock(&mbox->lock);
340 aq->rss.rq = rss_ctx->ind_tbl[idx];
343 aq->qidx = index + idx;
344 aq->ctype = NIX_AQ_CTYPE_RSS;
345 aq->op = NIX_AQ_INSTOP_INIT;
347 err = otx2_sync_mbox_msg(mbox);
348 mutex_unlock(&mbox->lock);
352 void otx2_set_rss_key(struct otx2_nic *pfvf)
354 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
355 u64 *key = (u64 *)&rss->key[4];
358 /* 352bit or 44byte key needs to be configured as below
359 * NIX_LF_RX_SECRETX0 = key<351:288>
360 * NIX_LF_RX_SECRETX1 = key<287:224>
361 * NIX_LF_RX_SECRETX2 = key<223:160>
362 * NIX_LF_RX_SECRETX3 = key<159:96>
363 * NIX_LF_RX_SECRETX4 = key<95:32>
364 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
366 otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
367 (u64)(*((u32 *)&rss->key)) << 32);
368 idx = sizeof(rss->key) / sizeof(u64);
371 otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
375 int otx2_rss_init(struct otx2_nic *pfvf)
377 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
378 struct otx2_rss_ctx *rss_ctx;
381 rss->rss_size = sizeof(*rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
383 /* Init RSS key if it is not setup already */
385 netdev_rss_key_fill(rss->key, sizeof(rss->key));
386 otx2_set_rss_key(pfvf);
388 if (!netif_is_rxfh_configured(pfvf->netdev)) {
389 /* Set RSS group 0 as default indirection table */
390 rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP] = kzalloc(rss->rss_size,
392 if (!rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP])
395 rss_ctx = rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP];
396 for (idx = 0; idx < rss->rss_size; idx++)
397 rss_ctx->ind_tbl[idx] =
398 ethtool_rxfh_indir_default(idx,
401 ret = otx2_set_rss_table(pfvf, DEFAULT_RSS_CONTEXT_GROUP);
405 /* Flowkey or hash config to be used for generating flow tag */
406 rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
407 NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
408 NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
409 NIX_FLOW_KEY_TYPE_SCTP | NIX_FLOW_KEY_TYPE_VLAN |
410 NIX_FLOW_KEY_TYPE_IPV4_PROTO;
412 ret = otx2_set_flowkey_cfg(pfvf);
420 /* Setup UDP segmentation algorithm in HW */
421 static void otx2_setup_udp_segmentation(struct nix_lso_format_cfg *lso, bool v4)
423 struct nix_lso_format *field;
425 field = (struct nix_lso_format *)&lso->fields[0];
426 lso->field_mask = GENMASK(18, 0);
428 /* IP's Length field */
429 field->layer = NIX_TXLAYER_OL3;
430 /* In ipv4, length field is at offset 2 bytes, for ipv6 it's 4 */
431 field->offset = v4 ? 2 : 4;
432 field->sizem1 = 1; /* i.e 2 bytes */
433 field->alg = NIX_LSOALG_ADD_PAYLEN;
436 /* No ID field in IPv6 header */
439 field->layer = NIX_TXLAYER_OL3;
441 field->sizem1 = 1; /* i.e 2 bytes */
442 field->alg = NIX_LSOALG_ADD_SEGNUM;
446 /* Update length in UDP header */
447 field->layer = NIX_TXLAYER_OL4;
450 field->alg = NIX_LSOALG_ADD_PAYLEN;
453 /* Setup segmentation algorithms in HW and retrieve algorithm index */
454 void otx2_setup_segmentation(struct otx2_nic *pfvf)
456 struct nix_lso_format_cfg_rsp *rsp;
457 struct nix_lso_format_cfg *lso;
458 struct otx2_hw *hw = &pfvf->hw;
461 mutex_lock(&pfvf->mbox.lock);
463 /* UDPv4 segmentation */
464 lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
468 /* Setup UDP/IP header fields that HW should update per segment */
469 otx2_setup_udp_segmentation(lso, true);
471 err = otx2_sync_mbox_msg(&pfvf->mbox);
475 rsp = (struct nix_lso_format_cfg_rsp *)
476 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
480 hw->lso_udpv4_idx = rsp->lso_format_idx;
482 /* UDPv6 segmentation */
483 lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
487 /* Setup UDP/IP header fields that HW should update per segment */
488 otx2_setup_udp_segmentation(lso, false);
490 err = otx2_sync_mbox_msg(&pfvf->mbox);
494 rsp = (struct nix_lso_format_cfg_rsp *)
495 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
499 hw->lso_udpv6_idx = rsp->lso_format_idx;
500 mutex_unlock(&pfvf->mbox.lock);
503 mutex_unlock(&pfvf->mbox.lock);
504 netdev_info(pfvf->netdev,
505 "Failed to get LSO index for UDP GSO offload, disabling\n");
506 pfvf->netdev->hw_features &= ~NETIF_F_GSO_UDP_L4;
509 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
511 /* Configure CQE interrupt coalescing parameters
513 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
514 * set 1 less than cq_ecount_wait. And cq_time_wait is in
515 * usecs, convert that to 100ns count.
517 otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
518 ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
519 ((u64)pfvf->hw.cq_qcount_wait << 32) |
520 (pfvf->hw.cq_ecount_wait - 1));
523 static int otx2_alloc_pool_buf(struct otx2_nic *pfvf, struct otx2_pool *pool,
526 unsigned int offset = 0;
530 sz = SKB_DATA_ALIGN(pool->rbsize);
531 sz = ALIGN(sz, OTX2_ALIGN);
533 page = page_pool_alloc_frag(pool->page_pool, &offset, sz, GFP_ATOMIC);
537 *dma = page_pool_get_dma_addr(page) + offset;
541 static int __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
547 return otx2_alloc_pool_buf(pfvf, pool, dma);
549 buf = napi_alloc_frag_align(pool->rbsize, OTX2_ALIGN);
553 *dma = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
554 DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
555 if (unlikely(dma_mapping_error(pfvf->dev, *dma))) {
563 int otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
569 ret = __otx2_alloc_rbuf(pfvf, pool, dma);
574 int otx2_alloc_buffer(struct otx2_nic *pfvf, struct otx2_cq_queue *cq,
577 if (unlikely(__otx2_alloc_rbuf(pfvf, cq->rbpool, dma)))
582 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
584 struct otx2_nic *pfvf = netdev_priv(netdev);
586 schedule_work(&pfvf->reset_task);
588 EXPORT_SYMBOL(otx2_tx_timeout);
590 void otx2_get_mac_from_af(struct net_device *netdev)
592 struct otx2_nic *pfvf = netdev_priv(netdev);
595 err = otx2_hw_get_mac_addr(pfvf, netdev);
597 dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
599 /* If AF doesn't provide a valid MAC, generate a random one */
600 if (!is_valid_ether_addr(netdev->dev_addr))
601 eth_hw_addr_random(netdev);
603 EXPORT_SYMBOL(otx2_get_mac_from_af);
605 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl, int prio, bool txschq_for_pfc)
607 u16 (*schq_list)[MAX_TXSCHQ_PER_FUNC];
608 struct otx2_hw *hw = &pfvf->hw;
609 struct nix_txschq_config *req;
613 dwrr_val = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
615 req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
622 schq_list = hw->txschq_list;
625 schq_list = pfvf->pfc_schq_list;
628 schq = schq_list[lvl][prio];
629 /* Set topology e.t.c configuration */
630 if (lvl == NIX_TXSCH_LVL_SMQ) {
631 req->reg[0] = NIX_AF_SMQX_CFG(schq);
632 req->regval[0] = ((u64)pfvf->tx_max_pktlen << 8) | OTX2_MIN_MTU;
633 req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
635 /* Set link type for DWRR MTU selection on CN10K silicons */
636 if (!is_dev_otx2(pfvf->pdev))
637 req->regval[0] |= FIELD_PREP(GENMASK_ULL(58, 57),
638 (u64)hw->smq_link_type);
641 parent = schq_list[NIX_TXSCH_LVL_TL4][prio];
642 req->reg[1] = NIX_AF_MDQX_PARENT(schq);
643 req->regval[1] = parent << 16;
645 /* Set DWRR quantum */
646 req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
647 req->regval[2] = dwrr_val;
648 } else if (lvl == NIX_TXSCH_LVL_TL4) {
649 parent = schq_list[NIX_TXSCH_LVL_TL3][prio];
650 req->reg[0] = NIX_AF_TL4X_PARENT(schq);
651 req->regval[0] = parent << 16;
653 req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
654 req->regval[1] = dwrr_val;
655 } else if (lvl == NIX_TXSCH_LVL_TL3) {
656 parent = schq_list[NIX_TXSCH_LVL_TL2][prio];
657 req->reg[0] = NIX_AF_TL3X_PARENT(schq);
658 req->regval[0] = parent << 16;
660 req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
661 req->regval[1] = dwrr_val;
662 if (lvl == hw->txschq_link_cfg_lvl) {
664 req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
665 /* Enable this queue and backpressure
666 * and set relative channel
668 req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
670 } else if (lvl == NIX_TXSCH_LVL_TL2) {
671 parent = schq_list[NIX_TXSCH_LVL_TL1][prio];
672 req->reg[0] = NIX_AF_TL2X_PARENT(schq);
673 req->regval[0] = parent << 16;
676 req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
677 req->regval[1] = TXSCH_TL1_DFLT_RR_PRIO << 24 | dwrr_val;
679 if (lvl == hw->txschq_link_cfg_lvl) {
681 req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
682 /* Enable this queue and backpressure
683 * and set relative channel
685 req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
687 } else if (lvl == NIX_TXSCH_LVL_TL1) {
688 /* Default config for TL1.
689 * For VF this is always ignored.
692 /* On CN10K, if RR_WEIGHT is greater than 16384, HW will
693 * clip it to 16384, so configuring a 24bit max value
694 * will work on both OTx2 and CN10K.
696 req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
697 req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
700 req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
701 req->regval[1] = (TXSCH_TL1_DFLT_RR_PRIO << 1);
704 req->reg[2] = NIX_AF_TL1X_CIR(schq);
708 return otx2_sync_mbox_msg(&pfvf->mbox);
710 EXPORT_SYMBOL(otx2_txschq_config);
712 int otx2_smq_flush(struct otx2_nic *pfvf, int smq)
714 struct nix_txschq_config *req;
717 mutex_lock(&pfvf->mbox.lock);
719 req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
721 mutex_unlock(&pfvf->mbox.lock);
725 req->lvl = NIX_TXSCH_LVL_SMQ;
726 req->reg[0] = NIX_AF_SMQX_CFG(smq);
727 req->regval[0] |= BIT_ULL(49);
730 rc = otx2_sync_mbox_msg(&pfvf->mbox);
731 mutex_unlock(&pfvf->mbox.lock);
734 EXPORT_SYMBOL(otx2_smq_flush);
736 int otx2_txsch_alloc(struct otx2_nic *pfvf)
738 struct nix_txsch_alloc_req *req;
739 struct nix_txsch_alloc_rsp *rsp;
742 /* Get memory to put this msg */
743 req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
747 /* Request one schq per level */
748 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
750 rc = otx2_sync_mbox_msg(&pfvf->mbox);
754 rsp = (struct nix_txsch_alloc_rsp *)
755 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
759 /* Setup transmit scheduler list */
760 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
761 for (schq = 0; schq < rsp->schq[lvl]; schq++)
762 pfvf->hw.txschq_list[lvl][schq] =
763 rsp->schq_list[lvl][schq];
765 pfvf->hw.txschq_link_cfg_lvl = rsp->link_cfg_lvl;
766 pfvf->hw.txschq_aggr_lvl_rr_prio = rsp->aggr_lvl_rr_prio;
771 void otx2_txschq_free_one(struct otx2_nic *pfvf, u16 lvl, u16 schq)
773 struct nix_txsch_free_req *free_req;
776 mutex_lock(&pfvf->mbox.lock);
778 free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
780 mutex_unlock(&pfvf->mbox.lock);
781 netdev_err(pfvf->netdev,
782 "Failed alloc txschq free req\n");
786 free_req->schq_lvl = lvl;
787 free_req->schq = schq;
789 err = otx2_sync_mbox_msg(&pfvf->mbox);
791 netdev_err(pfvf->netdev,
792 "Failed stop txschq %d at level %d\n", schq, lvl);
795 mutex_unlock(&pfvf->mbox.lock);
797 EXPORT_SYMBOL(otx2_txschq_free_one);
799 void otx2_txschq_stop(struct otx2_nic *pfvf)
803 /* free non QOS TLx nodes */
804 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
805 otx2_txschq_free_one(pfvf, lvl,
806 pfvf->hw.txschq_list[lvl][0]);
808 /* Clear the txschq list */
809 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
810 for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
811 pfvf->hw.txschq_list[lvl][schq] = 0;
816 void otx2_sqb_flush(struct otx2_nic *pfvf)
818 int qidx, sqe_tail, sqe_head;
819 struct otx2_snd_queue *sq;
822 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
823 for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
824 sq = &pfvf->qset.sq[qidx];
828 incr = (u64)qidx << 32;
829 val = otx2_atomic64_add(incr, ptr);
830 sqe_head = (val >> 20) & 0x3F;
831 sqe_tail = (val >> 28) & 0x3F;
832 if (sqe_head != sqe_tail)
833 usleep_range(50, 60);
837 /* RED and drop levels of CQ on packet reception.
838 * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
840 #define RQ_PASS_LVL_CQ(skid, qsize) ((((skid) + 16) * 256) / (qsize))
841 #define RQ_DROP_LVL_CQ(skid, qsize) (((skid) * 256) / (qsize))
843 /* RED and drop levels of AURA for packet reception.
844 * For AURA level is measure of fullness (0x0 = empty, 255 = full).
845 * Eg: For RQ length 1K, for pass/drop level 204/230.
846 * RED accepts pkts if free pointers > 102 & <= 205.
847 * Drops pkts if free pointers < 102.
849 #define RQ_BP_LVL_AURA (255 - ((85 * 256) / 100)) /* BP when 85% is full */
850 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
851 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
853 static int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
855 struct otx2_qset *qset = &pfvf->qset;
856 struct nix_aq_enq_req *aq;
858 /* Get memory to put this msg */
859 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
865 aq->rq.pb_caching = 1;
866 aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
867 aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
868 aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
869 aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
871 aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
872 aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
873 aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
874 aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
875 aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
876 aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
880 aq->ctype = NIX_AQ_CTYPE_RQ;
881 aq->op = NIX_AQ_INSTOP_INIT;
883 return otx2_sync_mbox_msg(&pfvf->mbox);
886 int otx2_sq_aq_init(void *dev, u16 qidx, u16 sqb_aura)
888 struct otx2_nic *pfvf = dev;
889 struct otx2_snd_queue *sq;
890 struct nix_aq_enq_req *aq;
892 sq = &pfvf->qset.sq[qidx];
893 sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
894 /* Get memory to put this msg */
895 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
899 aq->sq.cq = pfvf->hw.rx_queues + qidx;
900 aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
903 aq->sq.smq = otx2_get_smq_idx(pfvf, qidx);
904 aq->sq.smq_rr_quantum = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
905 aq->sq.default_chan = pfvf->hw.tx_chan_base;
906 aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
907 aq->sq.sqb_aura = sqb_aura;
908 aq->sq.sq_int_ena = NIX_SQINT_BITS;
910 /* Due pipelining impact minimum 2000 unused SQ CQE's
911 * need to maintain to avoid CQ overflow.
913 aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (pfvf->qset.sqe_cnt));
917 aq->ctype = NIX_AQ_CTYPE_SQ;
918 aq->op = NIX_AQ_INSTOP_INIT;
920 return otx2_sync_mbox_msg(&pfvf->mbox);
923 int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
925 struct otx2_qset *qset = &pfvf->qset;
926 struct otx2_snd_queue *sq;
927 struct otx2_pool *pool;
930 pool = &pfvf->qset.pool[sqb_aura];
931 sq = &qset->sq[qidx];
932 sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
933 sq->sqe_cnt = qset->sqe_cnt;
935 err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
939 if (qidx < pfvf->hw.tx_queues) {
940 err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
946 sq->sqe_base = sq->sqe->base;
947 sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
951 if (pfvf->ptp && qidx < pfvf->hw.tx_queues) {
952 err = qmem_alloc(pfvf->dev, &sq->timestamps, qset->sqe_cnt,
953 sizeof(*sq->timestamps));
960 sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
961 sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
962 /* Set SQE threshold to 10% of total SQEs */
963 sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
964 sq->aura_id = sqb_aura;
965 sq->aura_fc_addr = pool->fc_addr->base;
966 sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
971 return pfvf->hw_ops->sq_aq_init(pfvf, qidx, sqb_aura);
975 static int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
977 struct otx2_qset *qset = &pfvf->qset;
978 int err, pool_id, non_xdp_queues;
979 struct nix_aq_enq_req *aq;
980 struct otx2_cq_queue *cq;
982 cq = &qset->cq[qidx];
984 non_xdp_queues = pfvf->hw.rx_queues + pfvf->hw.tx_queues;
985 if (qidx < pfvf->hw.rx_queues) {
988 cq->cqe_cnt = qset->rqe_cnt;
990 xdp_rxq_info_reg(&cq->xdp_rxq, pfvf->netdev, qidx, 0);
991 } else if (qidx < non_xdp_queues) {
993 cq->cint_idx = qidx - pfvf->hw.rx_queues;
994 cq->cqe_cnt = qset->sqe_cnt;
996 if (pfvf->hw.xdp_queues &&
997 qidx < non_xdp_queues + pfvf->hw.xdp_queues) {
998 cq->cq_type = CQ_XDP;
999 cq->cint_idx = qidx - non_xdp_queues;
1000 cq->cqe_cnt = qset->sqe_cnt;
1002 cq->cq_type = CQ_QOS;
1003 cq->cint_idx = qidx - non_xdp_queues -
1004 pfvf->hw.xdp_queues;
1005 cq->cqe_cnt = qset->sqe_cnt;
1008 cq->cqe_size = pfvf->qset.xqe_size;
1010 /* Allocate memory for CQEs */
1011 err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
1015 /* Save CQE CPU base for faster reference */
1016 cq->cqe_base = cq->cqe->base;
1017 /* In case where all RQs auras point to single pool,
1018 * all CQs receive buffer pool also point to same pool.
1020 pool_id = ((cq->cq_type == CQ_RX) &&
1021 (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
1022 cq->rbpool = &qset->pool[pool_id];
1023 cq->refill_task_sched = false;
1025 /* Get memory to put this msg */
1026 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
1031 aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
1033 aq->cq.base = cq->cqe->iova;
1034 aq->cq.cint_idx = cq->cint_idx;
1035 aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
1036 aq->cq.qint_idx = 0;
1037 aq->cq.avg_level = 255;
1039 if (qidx < pfvf->hw.rx_queues) {
1040 aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
1041 aq->cq.drop_ena = 1;
1043 if (!is_otx2_lbkvf(pfvf->pdev)) {
1044 /* Enable receive CQ backpressure */
1047 aq->cq.bpid = pfvf->bpid[pfvf->queue_to_pfc_map[qidx]];
1049 aq->cq.bpid = pfvf->bpid[0];
1052 /* Set backpressure level is same as cq pass level */
1053 aq->cq.bp = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
1059 aq->ctype = NIX_AQ_CTYPE_CQ;
1060 aq->op = NIX_AQ_INSTOP_INIT;
1062 return otx2_sync_mbox_msg(&pfvf->mbox);
1065 static void otx2_pool_refill_task(struct work_struct *work)
1067 struct otx2_cq_queue *cq;
1068 struct refill_work *wrk;
1069 struct otx2_nic *pfvf;
1072 wrk = container_of(work, struct refill_work, pool_refill_work.work);
1074 qidx = wrk - pfvf->refill_wrk;
1075 cq = &pfvf->qset.cq[qidx];
1077 cq->refill_task_sched = false;
1080 napi_schedule(wrk->napi);
1084 int otx2_config_nix_queues(struct otx2_nic *pfvf)
1088 /* Initialize RX queues */
1089 for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
1090 u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
1092 err = otx2_rq_init(pfvf, qidx, lpb_aura);
1097 /* Initialize TX queues */
1098 for (qidx = 0; qidx < pfvf->hw.non_qos_queues; qidx++) {
1099 u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1101 err = otx2_sq_init(pfvf, qidx, sqb_aura);
1106 /* Initialize completion queues */
1107 for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1108 err = otx2_cq_init(pfvf, qidx);
1113 pfvf->cq_op_addr = (__force u64 *)otx2_get_regaddr(pfvf,
1114 NIX_LF_CQ_OP_STATUS);
1116 /* Initialize work queue for receive buffer refill */
1117 pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
1118 sizeof(struct refill_work), GFP_KERNEL);
1119 if (!pfvf->refill_wrk)
1122 for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1123 pfvf->refill_wrk[qidx].pf = pfvf;
1124 INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
1125 otx2_pool_refill_task);
1130 int otx2_config_nix(struct otx2_nic *pfvf)
1132 struct nix_lf_alloc_req *nixlf;
1133 struct nix_lf_alloc_rsp *rsp;
1136 pfvf->qset.xqe_size = pfvf->hw.xqe_size;
1138 /* Get memory to put this msg */
1139 nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
1143 /* Set RQ/SQ/CQ counts */
1144 nixlf->rq_cnt = pfvf->hw.rx_queues;
1145 nixlf->sq_cnt = otx2_get_total_tx_queues(pfvf);
1146 nixlf->cq_cnt = pfvf->qset.cq_cnt;
1147 nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
1148 nixlf->rss_grps = MAX_RSS_GROUPS;
1149 nixlf->xqe_sz = pfvf->hw.xqe_size == 128 ? NIX_XQESZ_W16 : NIX_XQESZ_W64;
1150 /* We don't know absolute NPA LF idx attached.
1151 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
1152 * NPA LF attached to this RVU PF/VF.
1154 nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
1155 /* Disable alignment pad, enable L2 length check,
1156 * enable L4 TCP/UDP checksum verification.
1158 nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
1160 err = otx2_sync_mbox_msg(&pfvf->mbox);
1164 rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
1167 return PTR_ERR(rsp);
1175 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
1177 struct otx2_qset *qset = &pfvf->qset;
1178 struct otx2_hw *hw = &pfvf->hw;
1179 struct otx2_snd_queue *sq;
1183 for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
1184 sq = &qset->sq[qidx];
1187 for (sqb = 0; sqb < sq->sqb_count; sqb++) {
1188 if (!sq->sqb_ptrs[sqb])
1190 iova = sq->sqb_ptrs[sqb];
1191 pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1192 dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
1194 DMA_ATTR_SKIP_CPU_SYNC);
1195 put_page(virt_to_page(phys_to_virt(pa)));
1201 void otx2_free_bufs(struct otx2_nic *pfvf, struct otx2_pool *pool,
1207 pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1208 page = virt_to_head_page(phys_to_virt(pa));
1210 if (pool->page_pool) {
1211 page_pool_put_full_page(pool->page_pool, page, true);
1213 dma_unmap_page_attrs(pfvf->dev, iova, size,
1215 DMA_ATTR_SKIP_CPU_SYNC);
1221 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
1223 int pool_id, pool_start = 0, pool_end = 0, size = 0;
1224 struct otx2_pool *pool;
1227 if (type == AURA_NIX_SQ) {
1228 pool_start = otx2_get_pool_idx(pfvf, type, 0);
1229 pool_end = pool_start + pfvf->hw.sqpool_cnt;
1230 size = pfvf->hw.sqb_size;
1232 if (type == AURA_NIX_RQ) {
1233 pool_start = otx2_get_pool_idx(pfvf, type, 0);
1234 pool_end = pfvf->hw.rqpool_cnt;
1235 size = pfvf->rbsize;
1238 /* Free SQB and RQB pointers from the aura pool */
1239 for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
1240 iova = otx2_aura_allocptr(pfvf, pool_id);
1241 pool = &pfvf->qset.pool[pool_id];
1243 if (type == AURA_NIX_RQ)
1244 iova -= OTX2_HEAD_ROOM;
1246 otx2_free_bufs(pfvf, pool, iova, size);
1248 iova = otx2_aura_allocptr(pfvf, pool_id);
1253 void otx2_aura_pool_free(struct otx2_nic *pfvf)
1255 struct otx2_pool *pool;
1258 if (!pfvf->qset.pool)
1261 for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
1262 pool = &pfvf->qset.pool[pool_id];
1263 qmem_free(pfvf->dev, pool->stack);
1264 qmem_free(pfvf->dev, pool->fc_addr);
1265 page_pool_destroy(pool->page_pool);
1266 pool->page_pool = NULL;
1268 devm_kfree(pfvf->dev, pfvf->qset.pool);
1269 pfvf->qset.pool = NULL;
1272 int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
1273 int pool_id, int numptrs)
1275 struct npa_aq_enq_req *aq;
1276 struct otx2_pool *pool;
1279 pool = &pfvf->qset.pool[pool_id];
1281 /* Allocate memory for HW to update Aura count.
1282 * Alloc one cache line, so that it fits all FC_STYPE modes.
1284 if (!pool->fc_addr) {
1285 err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
1290 /* Initialize this aura's context via AF */
1291 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1293 /* Shared mbox memory buffer is full, flush it and retry */
1294 err = otx2_sync_mbox_msg(&pfvf->mbox);
1297 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1302 aq->aura_id = aura_id;
1303 /* Will be filled by AF with correct pool context address */
1304 aq->aura.pool_addr = pool_id;
1305 aq->aura.pool_caching = 1;
1306 aq->aura.shift = ilog2(numptrs) - 8;
1307 aq->aura.count = numptrs;
1308 aq->aura.limit = numptrs;
1309 aq->aura.avg_level = 255;
1311 aq->aura.fc_ena = 1;
1312 aq->aura.fc_addr = pool->fc_addr->iova;
1313 aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1315 /* Enable backpressure for RQ aura */
1316 if (aura_id < pfvf->hw.rqpool_cnt && !is_otx2_lbkvf(pfvf->pdev)) {
1317 aq->aura.bp_ena = 0;
1318 /* If NIX1 LF is attached then specify NIX1_RX.
1320 * Below NPA_AURA_S[BP_ENA] is set according to the
1321 * NPA_BPINTF_E enumeration given as:
1322 * 0x0 + a*0x1 where 'a' is 0 for NIX0_RX and 1 for NIX1_RX so
1323 * NIX0_RX is 0x0 + 0*0x1 = 0
1324 * NIX1_RX is 0x0 + 1*0x1 = 1
1325 * But in HRM it is given that
1326 * "NPA_AURA_S[BP_ENA](w1[33:32]) - Enable aura backpressure to
1327 * NIX-RX based on [BP] level. One bit per NIX-RX; index
1328 * enumerated by NPA_BPINTF_E."
1330 if (pfvf->nix_blkaddr == BLKADDR_NIX1)
1331 aq->aura.bp_ena = 1;
1333 aq->aura.nix0_bpid = pfvf->bpid[pfvf->queue_to_pfc_map[aura_id]];
1335 aq->aura.nix0_bpid = pfvf->bpid[0];
1338 /* Set backpressure level for RQ's Aura */
1339 aq->aura.bp = RQ_BP_LVL_AURA;
1343 aq->ctype = NPA_AQ_CTYPE_AURA;
1344 aq->op = NPA_AQ_INSTOP_INIT;
1349 int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1350 int stack_pages, int numptrs, int buf_size, int type)
1352 struct page_pool_params pp_params = { 0 };
1353 struct npa_aq_enq_req *aq;
1354 struct otx2_pool *pool;
1357 pool = &pfvf->qset.pool[pool_id];
1358 /* Alloc memory for stack which is used to store buffer pointers */
1359 err = qmem_alloc(pfvf->dev, &pool->stack,
1360 stack_pages, pfvf->hw.stack_pg_bytes);
1364 pool->rbsize = buf_size;
1366 /* Initialize this pool's context via AF */
1367 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1369 /* Shared mbox memory buffer is full, flush it and retry */
1370 err = otx2_sync_mbox_msg(&pfvf->mbox);
1372 qmem_free(pfvf->dev, pool->stack);
1375 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1377 qmem_free(pfvf->dev, pool->stack);
1382 aq->aura_id = pool_id;
1383 aq->pool.stack_base = pool->stack->iova;
1384 aq->pool.stack_caching = 1;
1386 aq->pool.buf_size = buf_size / 128;
1387 aq->pool.stack_max_pages = stack_pages;
1388 aq->pool.shift = ilog2(numptrs) - 8;
1389 aq->pool.ptr_start = 0;
1390 aq->pool.ptr_end = ~0ULL;
1393 aq->ctype = NPA_AQ_CTYPE_POOL;
1394 aq->op = NPA_AQ_INSTOP_INIT;
1396 if (type != AURA_NIX_RQ) {
1397 pool->page_pool = NULL;
1401 pp_params.order = get_order(buf_size);
1402 pp_params.flags = PP_FLAG_DMA_MAP;
1403 pp_params.pool_size = min(OTX2_PAGE_POOL_SZ, numptrs);
1404 pp_params.nid = NUMA_NO_NODE;
1405 pp_params.dev = pfvf->dev;
1406 pp_params.dma_dir = DMA_FROM_DEVICE;
1407 pool->page_pool = page_pool_create(&pp_params);
1408 if (IS_ERR(pool->page_pool)) {
1409 netdev_err(pfvf->netdev, "Creation of page pool failed\n");
1410 return PTR_ERR(pool->page_pool);
1416 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1418 int qidx, pool_id, stack_pages, num_sqbs;
1419 struct otx2_qset *qset = &pfvf->qset;
1420 struct otx2_hw *hw = &pfvf->hw;
1421 struct otx2_snd_queue *sq;
1422 struct otx2_pool *pool;
1426 /* Calculate number of SQBs needed.
1428 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1429 * Last SQE is used for pointing to next SQB.
1431 num_sqbs = (hw->sqb_size / 128) - 1;
1432 num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1434 /* Get no of stack pages needed */
1436 (num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1438 for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1439 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1440 /* Initialize aura context */
1441 err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1445 /* Initialize pool context */
1446 err = otx2_pool_init(pfvf, pool_id, stack_pages,
1447 num_sqbs, hw->sqb_size, AURA_NIX_SQ);
1452 /* Flush accumulated messages */
1453 err = otx2_sync_mbox_msg(&pfvf->mbox);
1457 /* Allocate pointers and free them to aura/pool */
1458 for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1459 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1460 pool = &pfvf->qset.pool[pool_id];
1462 sq = &qset->sq[qidx];
1464 sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(*sq->sqb_ptrs), GFP_KERNEL);
1465 if (!sq->sqb_ptrs) {
1470 for (ptr = 0; ptr < num_sqbs; ptr++) {
1471 err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1474 pfvf->hw_ops->aura_freeptr(pfvf, pool_id, bufptr);
1475 sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1480 return err ? -ENOMEM : 0;
1483 otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1484 otx2_aura_pool_free(pfvf);
1488 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1490 struct otx2_hw *hw = &pfvf->hw;
1491 int stack_pages, pool_id, rq;
1492 struct otx2_pool *pool;
1493 int err, ptr, num_ptrs;
1496 num_ptrs = pfvf->qset.rqe_cnt;
1499 (num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1501 for (rq = 0; rq < hw->rx_queues; rq++) {
1502 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1503 /* Initialize aura context */
1504 err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1508 for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1509 err = otx2_pool_init(pfvf, pool_id, stack_pages,
1510 num_ptrs, pfvf->rbsize, AURA_NIX_RQ);
1515 /* Flush accumulated messages */
1516 err = otx2_sync_mbox_msg(&pfvf->mbox);
1520 /* Allocate pointers and free them to aura/pool */
1521 for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1522 pool = &pfvf->qset.pool[pool_id];
1523 for (ptr = 0; ptr < num_ptrs; ptr++) {
1524 err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1527 pfvf->hw_ops->aura_freeptr(pfvf, pool_id,
1528 bufptr + OTX2_HEAD_ROOM);
1533 otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1534 otx2_aura_pool_free(pfvf);
1538 int otx2_config_npa(struct otx2_nic *pfvf)
1540 struct otx2_qset *qset = &pfvf->qset;
1541 struct npa_lf_alloc_req *npalf;
1542 struct otx2_hw *hw = &pfvf->hw;
1545 /* Pool - Stack of free buffer pointers
1546 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1552 qset->pool = devm_kcalloc(pfvf->dev, hw->pool_cnt,
1553 sizeof(struct otx2_pool), GFP_KERNEL);
1557 /* Get memory to put this msg */
1558 npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1562 /* Set aura and pool counts */
1563 npalf->nr_pools = hw->pool_cnt;
1564 aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1565 npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1567 return otx2_sync_mbox_msg(&pfvf->mbox);
1570 int otx2_detach_resources(struct mbox *mbox)
1572 struct rsrc_detach *detach;
1574 mutex_lock(&mbox->lock);
1575 detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1577 mutex_unlock(&mbox->lock);
1582 detach->partial = false;
1584 /* Send detach request to AF */
1585 otx2_mbox_msg_send(&mbox->mbox, 0);
1586 mutex_unlock(&mbox->lock);
1589 EXPORT_SYMBOL(otx2_detach_resources);
1591 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1593 struct rsrc_attach *attach;
1594 struct msg_req *msix;
1597 mutex_lock(&pfvf->mbox.lock);
1598 /* Get memory to put this msg */
1599 attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1601 mutex_unlock(&pfvf->mbox.lock);
1605 attach->npalf = true;
1606 attach->nixlf = true;
1608 /* Send attach request to AF */
1609 err = otx2_sync_mbox_msg(&pfvf->mbox);
1611 mutex_unlock(&pfvf->mbox.lock);
1615 pfvf->nix_blkaddr = BLKADDR_NIX0;
1617 /* If the platform has two NIX blocks then LF may be
1618 * allocated from NIX1.
1620 if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1621 pfvf->nix_blkaddr = BLKADDR_NIX1;
1623 /* Get NPA and NIX MSIX vector offsets */
1624 msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1626 mutex_unlock(&pfvf->mbox.lock);
1630 err = otx2_sync_mbox_msg(&pfvf->mbox);
1632 mutex_unlock(&pfvf->mbox.lock);
1635 mutex_unlock(&pfvf->mbox.lock);
1637 if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1638 pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1640 "RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1646 EXPORT_SYMBOL(otx2_attach_npa_nix);
1648 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1650 struct hwctx_disable_req *req;
1652 mutex_lock(&mbox->lock);
1653 /* Request AQ to disable this context */
1655 req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1657 req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1660 mutex_unlock(&mbox->lock);
1666 if (otx2_sync_mbox_msg(mbox))
1667 dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1670 mutex_unlock(&mbox->lock);
1673 int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
1675 struct nix_bp_cfg_req *req;
1678 req = otx2_mbox_alloc_msg_nix_bp_enable(&pfvf->mbox);
1680 req = otx2_mbox_alloc_msg_nix_bp_disable(&pfvf->mbox);
1687 req->chan_cnt = pfvf->pfc_en ? IEEE_8021QAZ_MAX_TCS : 1;
1688 req->bpid_per_chan = pfvf->pfc_en ? 1 : 0;
1691 req->bpid_per_chan = 0;
1694 return otx2_sync_mbox_msg(&pfvf->mbox);
1696 EXPORT_SYMBOL(otx2_nix_config_bp);
1698 /* Mbox message handlers */
1699 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1700 struct cgx_stats_rsp *rsp)
1704 for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1705 pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1706 for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1707 pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1710 void mbox_handler_cgx_fec_stats(struct otx2_nic *pfvf,
1711 struct cgx_fec_stats_rsp *rsp)
1713 pfvf->hw.cgx_fec_corr_blks += rsp->fec_corr_blks;
1714 pfvf->hw.cgx_fec_uncorr_blks += rsp->fec_uncorr_blks;
1717 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1718 struct npa_lf_alloc_rsp *rsp)
1720 pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1721 pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1723 EXPORT_SYMBOL(mbox_handler_npa_lf_alloc);
1725 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1726 struct nix_lf_alloc_rsp *rsp)
1728 pfvf->hw.sqb_size = rsp->sqb_size;
1729 pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1730 pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1731 pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1732 pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1733 pfvf->hw.cgx_links = rsp->cgx_links;
1734 pfvf->hw.lbk_links = rsp->lbk_links;
1735 pfvf->hw.tx_link = rsp->tx_link;
1737 EXPORT_SYMBOL(mbox_handler_nix_lf_alloc);
1739 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1740 struct msix_offset_rsp *rsp)
1742 pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1743 pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1745 EXPORT_SYMBOL(mbox_handler_msix_offset);
1747 void mbox_handler_nix_bp_enable(struct otx2_nic *pfvf,
1748 struct nix_bp_cfg_rsp *rsp)
1752 for (chan = 0; chan < rsp->chan_cnt; chan++) {
1753 chan_id = ((rsp->chan_bpid[chan] >> 10) & 0x7F);
1754 pfvf->bpid[chan_id] = rsp->chan_bpid[chan] & 0x3FF;
1757 EXPORT_SYMBOL(mbox_handler_nix_bp_enable);
1759 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1761 struct otx2_qset *qset = &pfvf->qset;
1762 struct otx2_hw *hw = &pfvf->hw;
1765 for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1768 int vector = pci_irq_vector(pfvf->pdev, irq);
1770 irq_set_affinity_hint(vector, NULL);
1771 free_cpumask_var(hw->affinity_mask[irq]);
1772 free_irq(vector, &qset->napi[qidx]);
1776 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1778 struct otx2_hw *hw = &pfvf->hw;
1779 int vec, cpu, irq, cint;
1781 vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1782 cpu = cpumask_first(cpu_online_mask);
1785 for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1786 if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1789 cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1791 irq = pci_irq_vector(pfvf->pdev, vec);
1792 irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1794 cpu = cpumask_next(cpu, cpu_online_mask);
1795 if (unlikely(cpu >= nr_cpu_ids))
1800 static u32 get_dwrr_mtu(struct otx2_nic *pfvf, struct nix_hw_info *hw)
1802 if (is_otx2_lbkvf(pfvf->pdev)) {
1803 pfvf->hw.smq_link_type = SMQ_LINK_TYPE_LBK;
1804 return hw->lbk_dwrr_mtu;
1807 pfvf->hw.smq_link_type = SMQ_LINK_TYPE_RPM;
1808 return hw->rpm_dwrr_mtu;
1811 u16 otx2_get_max_mtu(struct otx2_nic *pfvf)
1813 struct nix_hw_info *rsp;
1814 struct msg_req *req;
1818 mutex_lock(&pfvf->mbox.lock);
1820 req = otx2_mbox_alloc_msg_nix_get_hw_info(&pfvf->mbox);
1826 rc = otx2_sync_mbox_msg(&pfvf->mbox);
1828 rsp = (struct nix_hw_info *)
1829 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
1831 /* HW counts VLAN insertion bytes (8 for double tag)
1832 * irrespective of whether SQE is requesting to insert VLAN
1833 * in the packet or not. Hence these 8 bytes have to be
1834 * discounted from max packet size otherwise HW will throw
1837 max_mtu = rsp->max_mtu - 8 - OTX2_ETH_HLEN;
1839 /* Also save DWRR MTU, needed for DWRR weight calculation */
1840 pfvf->hw.dwrr_mtu = get_dwrr_mtu(pfvf, rsp);
1841 if (!pfvf->hw.dwrr_mtu)
1842 pfvf->hw.dwrr_mtu = 1;
1846 mutex_unlock(&pfvf->mbox.lock);
1849 "Failed to get MTU from hardware setting default value(1500)\n");
1854 EXPORT_SYMBOL(otx2_get_max_mtu);
1856 int otx2_handle_ntuple_tc_features(struct net_device *netdev, netdev_features_t features)
1858 netdev_features_t changed = features ^ netdev->features;
1859 struct otx2_nic *pfvf = netdev_priv(netdev);
1860 bool ntuple = !!(features & NETIF_F_NTUPLE);
1861 bool tc = !!(features & NETIF_F_HW_TC);
1863 if ((changed & NETIF_F_NTUPLE) && !ntuple)
1864 otx2_destroy_ntuple_flows(pfvf);
1866 if ((changed & NETIF_F_NTUPLE) && ntuple) {
1867 if (!pfvf->flow_cfg->max_flows) {
1869 "Can't enable NTUPLE, MCAM entries not allocated\n");
1874 if ((changed & NETIF_F_HW_TC) && !tc &&
1875 otx2_tc_flower_rule_cnt(pfvf)) {
1876 netdev_err(netdev, "Can't disable TC hardware offload while flows are active\n");
1880 if ((changed & NETIF_F_NTUPLE) && ntuple &&
1881 otx2_tc_flower_rule_cnt(pfvf) && !(changed & NETIF_F_HW_TC)) {
1883 "Can't enable NTUPLE when TC flower offload is active, disable TC rules and retry\n");
1889 EXPORT_SYMBOL(otx2_handle_ntuple_tc_features);
1891 #define M(_name, _id, _fn_name, _req_type, _rsp_type) \
1893 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf, \
1894 struct _req_type *req, \
1895 struct _rsp_type *rsp) \
1897 /* Nothing to do here */ \
1900 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
1901 MBOX_UP_CGX_MESSAGES
1902 MBOX_UP_MCS_MESSAGES