1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell RVU Ethernet driver
4 * Copyright (C) 2020 Marvell.
8 #include <linux/interrupt.h>
13 #include "otx2_common.h"
14 #include "otx2_struct.h"
17 static void otx2_nix_rq_op_stats(struct queue_stats *stats,
18 struct otx2_nic *pfvf, int qidx)
20 u64 incr = (u64)qidx << 32;
23 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_OCTS);
24 stats->bytes = otx2_atomic64_add(incr, ptr);
26 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_PKTS);
27 stats->pkts = otx2_atomic64_add(incr, ptr);
30 static void otx2_nix_sq_op_stats(struct queue_stats *stats,
31 struct otx2_nic *pfvf, int qidx)
33 u64 incr = (u64)qidx << 32;
36 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_OCTS);
37 stats->bytes = otx2_atomic64_add(incr, ptr);
39 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_PKTS);
40 stats->pkts = otx2_atomic64_add(incr, ptr);
43 void otx2_update_lmac_stats(struct otx2_nic *pfvf)
47 if (!netif_running(pfvf->netdev))
50 mutex_lock(&pfvf->mbox.lock);
51 req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
53 mutex_unlock(&pfvf->mbox.lock);
57 otx2_sync_mbox_msg(&pfvf->mbox);
58 mutex_unlock(&pfvf->mbox.lock);
61 void otx2_update_lmac_fec_stats(struct otx2_nic *pfvf)
65 if (!netif_running(pfvf->netdev))
67 mutex_lock(&pfvf->mbox.lock);
68 req = otx2_mbox_alloc_msg_cgx_fec_stats(&pfvf->mbox);
70 otx2_sync_mbox_msg(&pfvf->mbox);
71 mutex_unlock(&pfvf->mbox.lock);
74 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
76 struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
81 otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
85 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
87 struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
92 otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
96 void otx2_get_dev_stats(struct otx2_nic *pfvf)
98 struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
100 #define OTX2_GET_RX_STATS(reg) \
101 otx2_read64(pfvf, NIX_LF_RX_STATX(reg))
102 #define OTX2_GET_TX_STATS(reg) \
103 otx2_read64(pfvf, NIX_LF_TX_STATX(reg))
105 dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
106 dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
107 dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
108 dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
109 dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
110 dev_stats->rx_frames = dev_stats->rx_bcast_frames +
111 dev_stats->rx_mcast_frames +
112 dev_stats->rx_ucast_frames;
114 dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
115 dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
116 dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
117 dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
118 dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
119 dev_stats->tx_frames = dev_stats->tx_bcast_frames +
120 dev_stats->tx_mcast_frames +
121 dev_stats->tx_ucast_frames;
124 void otx2_get_stats64(struct net_device *netdev,
125 struct rtnl_link_stats64 *stats)
127 struct otx2_nic *pfvf = netdev_priv(netdev);
128 struct otx2_dev_stats *dev_stats;
130 otx2_get_dev_stats(pfvf);
132 dev_stats = &pfvf->hw.dev_stats;
133 stats->rx_bytes = dev_stats->rx_bytes;
134 stats->rx_packets = dev_stats->rx_frames;
135 stats->rx_dropped = dev_stats->rx_drops;
136 stats->multicast = dev_stats->rx_mcast_frames;
138 stats->tx_bytes = dev_stats->tx_bytes;
139 stats->tx_packets = dev_stats->tx_frames;
140 stats->tx_dropped = dev_stats->tx_drops;
142 EXPORT_SYMBOL(otx2_get_stats64);
144 /* Sync MAC address with RVU AF */
145 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
147 struct nix_set_mac_addr *req;
150 mutex_lock(&pfvf->mbox.lock);
151 req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
153 mutex_unlock(&pfvf->mbox.lock);
157 ether_addr_copy(req->mac_addr, mac);
159 err = otx2_sync_mbox_msg(&pfvf->mbox);
160 mutex_unlock(&pfvf->mbox.lock);
164 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
165 struct net_device *netdev)
167 struct nix_get_mac_addr_rsp *rsp;
168 struct mbox_msghdr *msghdr;
172 mutex_lock(&pfvf->mbox.lock);
173 req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
175 mutex_unlock(&pfvf->mbox.lock);
179 err = otx2_sync_mbox_msg(&pfvf->mbox);
181 mutex_unlock(&pfvf->mbox.lock);
185 msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
186 if (IS_ERR(msghdr)) {
187 mutex_unlock(&pfvf->mbox.lock);
188 return PTR_ERR(msghdr);
190 rsp = (struct nix_get_mac_addr_rsp *)msghdr;
191 ether_addr_copy(netdev->dev_addr, rsp->mac_addr);
192 mutex_unlock(&pfvf->mbox.lock);
197 int otx2_set_mac_address(struct net_device *netdev, void *p)
199 struct otx2_nic *pfvf = netdev_priv(netdev);
200 struct sockaddr *addr = p;
202 if (!is_valid_ether_addr(addr->sa_data))
203 return -EADDRNOTAVAIL;
205 if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data)) {
206 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
207 /* update dmac field in vlan offload rule */
208 if (netif_running(netdev) &&
209 pfvf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
210 otx2_install_rxvlan_offload_flow(pfvf);
211 /* update dmac address in ntuple and DMAC filter list */
212 if (pfvf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
213 otx2_dmacflt_update_pfmac_flow(pfvf);
220 EXPORT_SYMBOL(otx2_set_mac_address);
222 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
224 struct nix_frs_cfg *req;
227 mutex_lock(&pfvf->mbox.lock);
228 req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
230 mutex_unlock(&pfvf->mbox.lock);
234 req->maxlen = pfvf->max_frs;
236 err = otx2_sync_mbox_msg(&pfvf->mbox);
237 mutex_unlock(&pfvf->mbox.lock);
241 int otx2_config_pause_frm(struct otx2_nic *pfvf)
243 struct cgx_pause_frm_cfg *req;
246 if (is_otx2_lbkvf(pfvf->pdev))
249 mutex_lock(&pfvf->mbox.lock);
250 req = otx2_mbox_alloc_msg_cgx_cfg_pause_frm(&pfvf->mbox);
256 req->rx_pause = !!(pfvf->flags & OTX2_FLAG_RX_PAUSE_ENABLED);
257 req->tx_pause = !!(pfvf->flags & OTX2_FLAG_TX_PAUSE_ENABLED);
260 err = otx2_sync_mbox_msg(&pfvf->mbox);
262 mutex_unlock(&pfvf->mbox.lock);
266 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
268 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
269 struct nix_rss_flowkey_cfg_rsp *rsp;
270 struct nix_rss_flowkey_cfg *req;
273 mutex_lock(&pfvf->mbox.lock);
274 req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
276 mutex_unlock(&pfvf->mbox.lock);
279 req->mcam_index = -1; /* Default or reserved index */
280 req->flowkey_cfg = rss->flowkey_cfg;
281 req->group = DEFAULT_RSS_CONTEXT_GROUP;
283 err = otx2_sync_mbox_msg(&pfvf->mbox);
287 rsp = (struct nix_rss_flowkey_cfg_rsp *)
288 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
294 pfvf->hw.flowkey_alg_idx = rsp->alg_idx;
296 mutex_unlock(&pfvf->mbox.lock);
300 int otx2_set_rss_table(struct otx2_nic *pfvf, int ctx_id)
302 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
303 const int index = rss->rss_size * ctx_id;
304 struct mbox *mbox = &pfvf->mbox;
305 struct otx2_rss_ctx *rss_ctx;
306 struct nix_aq_enq_req *aq;
309 mutex_lock(&mbox->lock);
310 rss_ctx = rss->rss_ctx[ctx_id];
311 /* Get memory to put this msg */
312 for (idx = 0; idx < rss->rss_size; idx++) {
313 aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
315 /* The shared memory buffer can be full.
318 err = otx2_sync_mbox_msg(mbox);
320 mutex_unlock(&mbox->lock);
323 aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
325 mutex_unlock(&mbox->lock);
330 aq->rss.rq = rss_ctx->ind_tbl[idx];
333 aq->qidx = index + idx;
334 aq->ctype = NIX_AQ_CTYPE_RSS;
335 aq->op = NIX_AQ_INSTOP_INIT;
337 err = otx2_sync_mbox_msg(mbox);
338 mutex_unlock(&mbox->lock);
342 void otx2_set_rss_key(struct otx2_nic *pfvf)
344 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
345 u64 *key = (u64 *)&rss->key[4];
348 /* 352bit or 44byte key needs to be configured as below
349 * NIX_LF_RX_SECRETX0 = key<351:288>
350 * NIX_LF_RX_SECRETX1 = key<287:224>
351 * NIX_LF_RX_SECRETX2 = key<223:160>
352 * NIX_LF_RX_SECRETX3 = key<159:96>
353 * NIX_LF_RX_SECRETX4 = key<95:32>
354 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
356 otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
357 (u64)(*((u32 *)&rss->key)) << 32);
358 idx = sizeof(rss->key) / sizeof(u64);
361 otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
365 int otx2_rss_init(struct otx2_nic *pfvf)
367 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
368 struct otx2_rss_ctx *rss_ctx;
371 rss->rss_size = sizeof(*rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
373 /* Init RSS key if it is not setup already */
375 netdev_rss_key_fill(rss->key, sizeof(rss->key));
376 otx2_set_rss_key(pfvf);
378 if (!netif_is_rxfh_configured(pfvf->netdev)) {
379 /* Set RSS group 0 as default indirection table */
380 rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP] = kzalloc(rss->rss_size,
382 if (!rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP])
385 rss_ctx = rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP];
386 for (idx = 0; idx < rss->rss_size; idx++)
387 rss_ctx->ind_tbl[idx] =
388 ethtool_rxfh_indir_default(idx,
391 ret = otx2_set_rss_table(pfvf, DEFAULT_RSS_CONTEXT_GROUP);
395 /* Flowkey or hash config to be used for generating flow tag */
396 rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
397 NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
398 NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
399 NIX_FLOW_KEY_TYPE_SCTP | NIX_FLOW_KEY_TYPE_VLAN |
400 NIX_FLOW_KEY_TYPE_IPV4_PROTO;
402 ret = otx2_set_flowkey_cfg(pfvf);
410 /* Setup UDP segmentation algorithm in HW */
411 static void otx2_setup_udp_segmentation(struct nix_lso_format_cfg *lso, bool v4)
413 struct nix_lso_format *field;
415 field = (struct nix_lso_format *)&lso->fields[0];
416 lso->field_mask = GENMASK(18, 0);
418 /* IP's Length field */
419 field->layer = NIX_TXLAYER_OL3;
420 /* In ipv4, length field is at offset 2 bytes, for ipv6 it's 4 */
421 field->offset = v4 ? 2 : 4;
422 field->sizem1 = 1; /* i.e 2 bytes */
423 field->alg = NIX_LSOALG_ADD_PAYLEN;
426 /* No ID field in IPv6 header */
429 field->layer = NIX_TXLAYER_OL3;
431 field->sizem1 = 1; /* i.e 2 bytes */
432 field->alg = NIX_LSOALG_ADD_SEGNUM;
436 /* Update length in UDP header */
437 field->layer = NIX_TXLAYER_OL4;
440 field->alg = NIX_LSOALG_ADD_PAYLEN;
443 /* Setup segmentation algorithms in HW and retrieve algorithm index */
444 void otx2_setup_segmentation(struct otx2_nic *pfvf)
446 struct nix_lso_format_cfg_rsp *rsp;
447 struct nix_lso_format_cfg *lso;
448 struct otx2_hw *hw = &pfvf->hw;
451 mutex_lock(&pfvf->mbox.lock);
453 /* UDPv4 segmentation */
454 lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
458 /* Setup UDP/IP header fields that HW should update per segment */
459 otx2_setup_udp_segmentation(lso, true);
461 err = otx2_sync_mbox_msg(&pfvf->mbox);
465 rsp = (struct nix_lso_format_cfg_rsp *)
466 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
470 hw->lso_udpv4_idx = rsp->lso_format_idx;
472 /* UDPv6 segmentation */
473 lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
477 /* Setup UDP/IP header fields that HW should update per segment */
478 otx2_setup_udp_segmentation(lso, false);
480 err = otx2_sync_mbox_msg(&pfvf->mbox);
484 rsp = (struct nix_lso_format_cfg_rsp *)
485 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
489 hw->lso_udpv6_idx = rsp->lso_format_idx;
490 mutex_unlock(&pfvf->mbox.lock);
493 mutex_unlock(&pfvf->mbox.lock);
494 netdev_info(pfvf->netdev,
495 "Failed to get LSO index for UDP GSO offload, disabling\n");
496 pfvf->netdev->hw_features &= ~NETIF_F_GSO_UDP_L4;
499 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
501 /* Configure CQE interrupt coalescing parameters
503 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
504 * set 1 less than cq_ecount_wait. And cq_time_wait is in
505 * usecs, convert that to 100ns count.
507 otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
508 ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
509 ((u64)pfvf->hw.cq_qcount_wait << 32) |
510 (pfvf->hw.cq_ecount_wait - 1));
513 int __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
518 buf = napi_alloc_frag_align(pool->rbsize, OTX2_ALIGN);
522 *dma = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
523 DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
524 if (unlikely(dma_mapping_error(pfvf->dev, *dma))) {
532 static int otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
538 ret = __otx2_alloc_rbuf(pfvf, pool, dma);
543 int otx2_alloc_buffer(struct otx2_nic *pfvf, struct otx2_cq_queue *cq,
546 if (unlikely(__otx2_alloc_rbuf(pfvf, cq->rbpool, dma))) {
547 struct refill_work *work;
548 struct delayed_work *dwork;
550 work = &pfvf->refill_wrk[cq->cq_idx];
551 dwork = &work->pool_refill_work;
552 /* Schedule a task if no other task is running */
553 if (!cq->refill_task_sched) {
554 cq->refill_task_sched = true;
555 schedule_delayed_work(dwork,
556 msecs_to_jiffies(100));
563 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
565 struct otx2_nic *pfvf = netdev_priv(netdev);
567 schedule_work(&pfvf->reset_task);
569 EXPORT_SYMBOL(otx2_tx_timeout);
571 void otx2_get_mac_from_af(struct net_device *netdev)
573 struct otx2_nic *pfvf = netdev_priv(netdev);
576 err = otx2_hw_get_mac_addr(pfvf, netdev);
578 dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
580 /* If AF doesn't provide a valid MAC, generate a random one */
581 if (!is_valid_ether_addr(netdev->dev_addr))
582 eth_hw_addr_random(netdev);
584 EXPORT_SYMBOL(otx2_get_mac_from_af);
586 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl)
588 struct otx2_hw *hw = &pfvf->hw;
589 struct nix_txschq_config *req;
593 dwrr_val = mtu_to_dwrr_weight(pfvf, pfvf->max_frs);
595 req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
602 schq = hw->txschq_list[lvl][0];
603 /* Set topology e.t.c configuration */
604 if (lvl == NIX_TXSCH_LVL_SMQ) {
605 req->reg[0] = NIX_AF_SMQX_CFG(schq);
606 req->regval[0] = ((pfvf->netdev->max_mtu + OTX2_ETH_HLEN) << 8)
609 req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
613 parent = hw->txschq_list[NIX_TXSCH_LVL_TL4][0];
614 req->reg[1] = NIX_AF_MDQX_PARENT(schq);
615 req->regval[1] = parent << 16;
617 /* Set DWRR quantum */
618 req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
619 req->regval[2] = dwrr_val;
620 } else if (lvl == NIX_TXSCH_LVL_TL4) {
621 parent = hw->txschq_list[NIX_TXSCH_LVL_TL3][0];
622 req->reg[0] = NIX_AF_TL4X_PARENT(schq);
623 req->regval[0] = parent << 16;
625 req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
626 req->regval[1] = dwrr_val;
627 } else if (lvl == NIX_TXSCH_LVL_TL3) {
628 parent = hw->txschq_list[NIX_TXSCH_LVL_TL2][0];
629 req->reg[0] = NIX_AF_TL3X_PARENT(schq);
630 req->regval[0] = parent << 16;
632 req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
633 req->regval[1] = dwrr_val;
634 if (lvl == hw->txschq_link_cfg_lvl) {
636 req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
637 /* Enable this queue and backpressure */
638 req->regval[2] = BIT_ULL(13) | BIT_ULL(12);
640 } else if (lvl == NIX_TXSCH_LVL_TL2) {
641 parent = hw->txschq_list[NIX_TXSCH_LVL_TL1][0];
642 req->reg[0] = NIX_AF_TL2X_PARENT(schq);
643 req->regval[0] = parent << 16;
646 req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
647 req->regval[1] = TXSCH_TL1_DFLT_RR_PRIO << 24 | dwrr_val;
649 if (lvl == hw->txschq_link_cfg_lvl) {
651 req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
652 /* Enable this queue and backpressure */
653 req->regval[2] = BIT_ULL(13) | BIT_ULL(12);
655 } else if (lvl == NIX_TXSCH_LVL_TL1) {
656 /* Default config for TL1.
657 * For VF this is always ignored.
660 /* On CN10K, if RR_WEIGHT is greater than 16384, HW will
661 * clip it to 16384, so configuring a 24bit max value
662 * will work on both OTx2 and CN10K.
664 req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
665 req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
668 req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
669 req->regval[1] = (TXSCH_TL1_DFLT_RR_PRIO << 1);
672 req->reg[2] = NIX_AF_TL1X_CIR(schq);
676 return otx2_sync_mbox_msg(&pfvf->mbox);
679 int otx2_txsch_alloc(struct otx2_nic *pfvf)
681 struct nix_txsch_alloc_req *req;
684 /* Get memory to put this msg */
685 req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
689 /* Request one schq per level */
690 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
693 return otx2_sync_mbox_msg(&pfvf->mbox);
696 int otx2_txschq_stop(struct otx2_nic *pfvf)
698 struct nix_txsch_free_req *free_req;
701 mutex_lock(&pfvf->mbox.lock);
702 /* Free the transmit schedulers */
703 free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
705 mutex_unlock(&pfvf->mbox.lock);
709 free_req->flags = TXSCHQ_FREE_ALL;
710 err = otx2_sync_mbox_msg(&pfvf->mbox);
711 mutex_unlock(&pfvf->mbox.lock);
713 /* Clear the txschq list */
714 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
715 for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
716 pfvf->hw.txschq_list[lvl][schq] = 0;
721 void otx2_sqb_flush(struct otx2_nic *pfvf)
723 int qidx, sqe_tail, sqe_head;
727 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
728 for (qidx = 0; qidx < pfvf->hw.tx_queues; qidx++) {
729 incr = (u64)qidx << 32;
731 val = otx2_atomic64_add(incr, ptr);
732 sqe_head = (val >> 20) & 0x3F;
733 sqe_tail = (val >> 28) & 0x3F;
734 if (sqe_head == sqe_tail)
742 /* RED and drop levels of CQ on packet reception.
743 * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
745 #define RQ_PASS_LVL_CQ(skid, qsize) ((((skid) + 16) * 256) / (qsize))
746 #define RQ_DROP_LVL_CQ(skid, qsize) (((skid) * 256) / (qsize))
748 /* RED and drop levels of AURA for packet reception.
749 * For AURA level is measure of fullness (0x0 = empty, 255 = full).
750 * Eg: For RQ length 1K, for pass/drop level 204/230.
751 * RED accepts pkts if free pointers > 102 & <= 205.
752 * Drops pkts if free pointers < 102.
754 #define RQ_BP_LVL_AURA (255 - ((85 * 256) / 100)) /* BP when 85% is full */
755 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
756 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
758 static int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
760 struct otx2_qset *qset = &pfvf->qset;
761 struct nix_aq_enq_req *aq;
763 /* Get memory to put this msg */
764 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
770 aq->rq.pb_caching = 1;
771 aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
772 aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
773 aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
774 aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
776 aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
777 aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
778 aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
779 aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
780 aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
781 aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
785 aq->ctype = NIX_AQ_CTYPE_RQ;
786 aq->op = NIX_AQ_INSTOP_INIT;
788 return otx2_sync_mbox_msg(&pfvf->mbox);
791 int otx2_sq_aq_init(void *dev, u16 qidx, u16 sqb_aura)
793 struct otx2_nic *pfvf = dev;
794 struct otx2_snd_queue *sq;
795 struct nix_aq_enq_req *aq;
797 sq = &pfvf->qset.sq[qidx];
798 sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
799 /* Get memory to put this msg */
800 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
804 aq->sq.cq = pfvf->hw.rx_queues + qidx;
805 aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
808 /* Only one SMQ is allocated, map all SQ's to that SMQ */
809 aq->sq.smq = pfvf->hw.txschq_list[NIX_TXSCH_LVL_SMQ][0];
810 aq->sq.smq_rr_quantum = mtu_to_dwrr_weight(pfvf, pfvf->max_frs);
811 aq->sq.default_chan = pfvf->hw.tx_chan_base;
812 aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
813 aq->sq.sqb_aura = sqb_aura;
814 aq->sq.sq_int_ena = NIX_SQINT_BITS;
816 /* Due pipelining impact minimum 2000 unused SQ CQE's
817 * need to maintain to avoid CQ overflow.
819 aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (pfvf->qset.sqe_cnt));
823 aq->ctype = NIX_AQ_CTYPE_SQ;
824 aq->op = NIX_AQ_INSTOP_INIT;
826 return otx2_sync_mbox_msg(&pfvf->mbox);
829 static int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
831 struct otx2_qset *qset = &pfvf->qset;
832 struct otx2_snd_queue *sq;
833 struct otx2_pool *pool;
836 pool = &pfvf->qset.pool[sqb_aura];
837 sq = &qset->sq[qidx];
838 sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
839 sq->sqe_cnt = qset->sqe_cnt;
841 err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
845 err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
850 sq->sqe_base = sq->sqe->base;
851 sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
856 err = qmem_alloc(pfvf->dev, &sq->timestamps, qset->sqe_cnt,
857 sizeof(*sq->timestamps));
863 sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
864 sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
865 /* Set SQE threshold to 10% of total SQEs */
866 sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
867 sq->aura_id = sqb_aura;
868 sq->aura_fc_addr = pool->fc_addr->base;
869 sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
874 return pfvf->hw_ops->sq_aq_init(pfvf, qidx, sqb_aura);
878 static int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
880 struct otx2_qset *qset = &pfvf->qset;
881 struct nix_aq_enq_req *aq;
882 struct otx2_cq_queue *cq;
885 cq = &qset->cq[qidx];
887 if (qidx < pfvf->hw.rx_queues) {
890 cq->cqe_cnt = qset->rqe_cnt;
893 cq->cint_idx = qidx - pfvf->hw.rx_queues;
894 cq->cqe_cnt = qset->sqe_cnt;
896 cq->cqe_size = pfvf->qset.xqe_size;
898 /* Allocate memory for CQEs */
899 err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
903 /* Save CQE CPU base for faster reference */
904 cq->cqe_base = cq->cqe->base;
905 /* In case where all RQs auras point to single pool,
906 * all CQs receive buffer pool also point to same pool.
908 pool_id = ((cq->cq_type == CQ_RX) &&
909 (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
910 cq->rbpool = &qset->pool[pool_id];
911 cq->refill_task_sched = false;
913 /* Get memory to put this msg */
914 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
919 aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
921 aq->cq.base = cq->cqe->iova;
922 aq->cq.cint_idx = cq->cint_idx;
923 aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
925 aq->cq.avg_level = 255;
927 if (qidx < pfvf->hw.rx_queues) {
928 aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
931 if (!is_otx2_lbkvf(pfvf->pdev)) {
932 /* Enable receive CQ backpressure */
934 aq->cq.bpid = pfvf->bpid[0];
936 /* Set backpressure level is same as cq pass level */
937 aq->cq.bp = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
943 aq->ctype = NIX_AQ_CTYPE_CQ;
944 aq->op = NIX_AQ_INSTOP_INIT;
946 return otx2_sync_mbox_msg(&pfvf->mbox);
949 static void otx2_pool_refill_task(struct work_struct *work)
951 struct otx2_cq_queue *cq;
952 struct otx2_pool *rbpool;
953 struct refill_work *wrk;
954 int qidx, free_ptrs = 0;
955 struct otx2_nic *pfvf;
958 wrk = container_of(work, struct refill_work, pool_refill_work.work);
960 qidx = wrk - pfvf->refill_wrk;
961 cq = &pfvf->qset.cq[qidx];
963 free_ptrs = cq->pool_ptrs;
966 while (cq->pool_ptrs) {
967 if (otx2_alloc_rbuf(pfvf, rbpool, &bufptr)) {
968 /* Schedule a WQ if we fails to free atleast half of the
969 * pointers else enable napi for this RQ.
971 if (!((free_ptrs - cq->pool_ptrs) > free_ptrs / 2)) {
972 struct delayed_work *dwork;
974 dwork = &wrk->pool_refill_work;
975 schedule_delayed_work(dwork,
976 msecs_to_jiffies(100));
978 cq->refill_task_sched = false;
982 pfvf->hw_ops->aura_freeptr(pfvf, qidx, bufptr + OTX2_HEAD_ROOM);
986 cq->refill_task_sched = false;
989 int otx2_config_nix_queues(struct otx2_nic *pfvf)
993 /* Initialize RX queues */
994 for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
995 u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
997 err = otx2_rq_init(pfvf, qidx, lpb_aura);
1002 /* Initialize TX queues */
1003 for (qidx = 0; qidx < pfvf->hw.tx_queues; qidx++) {
1004 u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1006 err = otx2_sq_init(pfvf, qidx, sqb_aura);
1011 /* Initialize completion queues */
1012 for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1013 err = otx2_cq_init(pfvf, qidx);
1018 pfvf->cq_op_addr = (__force u64 *)otx2_get_regaddr(pfvf,
1019 NIX_LF_CQ_OP_STATUS);
1021 /* Initialize work queue for receive buffer refill */
1022 pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
1023 sizeof(struct refill_work), GFP_KERNEL);
1024 if (!pfvf->refill_wrk)
1027 for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1028 pfvf->refill_wrk[qidx].pf = pfvf;
1029 INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
1030 otx2_pool_refill_task);
1035 int otx2_config_nix(struct otx2_nic *pfvf)
1037 struct nix_lf_alloc_req *nixlf;
1038 struct nix_lf_alloc_rsp *rsp;
1041 pfvf->qset.xqe_size = NIX_XQESZ_W16 ? 128 : 512;
1043 /* Get memory to put this msg */
1044 nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
1048 /* Set RQ/SQ/CQ counts */
1049 nixlf->rq_cnt = pfvf->hw.rx_queues;
1050 nixlf->sq_cnt = pfvf->hw.tx_queues;
1051 nixlf->cq_cnt = pfvf->qset.cq_cnt;
1052 nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
1053 nixlf->rss_grps = MAX_RSS_GROUPS;
1054 nixlf->xqe_sz = NIX_XQESZ_W16;
1055 /* We don't know absolute NPA LF idx attached.
1056 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
1057 * NPA LF attached to this RVU PF/VF.
1059 nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
1060 /* Disable alignment pad, enable L2 length check,
1061 * enable L4 TCP/UDP checksum verification.
1063 nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
1065 err = otx2_sync_mbox_msg(&pfvf->mbox);
1069 rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
1072 return PTR_ERR(rsp);
1080 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
1082 struct otx2_qset *qset = &pfvf->qset;
1083 struct otx2_hw *hw = &pfvf->hw;
1084 struct otx2_snd_queue *sq;
1088 for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1089 sq = &qset->sq[qidx];
1092 for (sqb = 0; sqb < sq->sqb_count; sqb++) {
1093 if (!sq->sqb_ptrs[sqb])
1095 iova = sq->sqb_ptrs[sqb];
1096 pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1097 dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
1099 DMA_ATTR_SKIP_CPU_SYNC);
1100 put_page(virt_to_page(phys_to_virt(pa)));
1106 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
1108 int pool_id, pool_start = 0, pool_end = 0, size = 0;
1111 if (type == AURA_NIX_SQ) {
1112 pool_start = otx2_get_pool_idx(pfvf, type, 0);
1113 pool_end = pool_start + pfvf->hw.sqpool_cnt;
1114 size = pfvf->hw.sqb_size;
1116 if (type == AURA_NIX_RQ) {
1117 pool_start = otx2_get_pool_idx(pfvf, type, 0);
1118 pool_end = pfvf->hw.rqpool_cnt;
1119 size = pfvf->rbsize;
1122 /* Free SQB and RQB pointers from the aura pool */
1123 for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
1124 iova = otx2_aura_allocptr(pfvf, pool_id);
1126 if (type == AURA_NIX_RQ)
1127 iova -= OTX2_HEAD_ROOM;
1129 pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1130 dma_unmap_page_attrs(pfvf->dev, iova, size,
1132 DMA_ATTR_SKIP_CPU_SYNC);
1133 put_page(virt_to_page(phys_to_virt(pa)));
1134 iova = otx2_aura_allocptr(pfvf, pool_id);
1139 void otx2_aura_pool_free(struct otx2_nic *pfvf)
1141 struct otx2_pool *pool;
1144 if (!pfvf->qset.pool)
1147 for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
1148 pool = &pfvf->qset.pool[pool_id];
1149 qmem_free(pfvf->dev, pool->stack);
1150 qmem_free(pfvf->dev, pool->fc_addr);
1152 devm_kfree(pfvf->dev, pfvf->qset.pool);
1153 pfvf->qset.pool = NULL;
1156 static int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
1157 int pool_id, int numptrs)
1159 struct npa_aq_enq_req *aq;
1160 struct otx2_pool *pool;
1163 pool = &pfvf->qset.pool[pool_id];
1165 /* Allocate memory for HW to update Aura count.
1166 * Alloc one cache line, so that it fits all FC_STYPE modes.
1168 if (!pool->fc_addr) {
1169 err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
1174 /* Initialize this aura's context via AF */
1175 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1177 /* Shared mbox memory buffer is full, flush it and retry */
1178 err = otx2_sync_mbox_msg(&pfvf->mbox);
1181 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1186 aq->aura_id = aura_id;
1187 /* Will be filled by AF with correct pool context address */
1188 aq->aura.pool_addr = pool_id;
1189 aq->aura.pool_caching = 1;
1190 aq->aura.shift = ilog2(numptrs) - 8;
1191 aq->aura.count = numptrs;
1192 aq->aura.limit = numptrs;
1193 aq->aura.avg_level = 255;
1195 aq->aura.fc_ena = 1;
1196 aq->aura.fc_addr = pool->fc_addr->iova;
1197 aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1199 /* Enable backpressure for RQ aura */
1200 if (aura_id < pfvf->hw.rqpool_cnt && !is_otx2_lbkvf(pfvf->pdev)) {
1201 aq->aura.bp_ena = 0;
1202 /* If NIX1 LF is attached then specify NIX1_RX.
1204 * Below NPA_AURA_S[BP_ENA] is set according to the
1205 * NPA_BPINTF_E enumeration given as:
1206 * 0x0 + a*0x1 where 'a' is 0 for NIX0_RX and 1 for NIX1_RX so
1207 * NIX0_RX is 0x0 + 0*0x1 = 0
1208 * NIX1_RX is 0x0 + 1*0x1 = 1
1209 * But in HRM it is given that
1210 * "NPA_AURA_S[BP_ENA](w1[33:32]) - Enable aura backpressure to
1211 * NIX-RX based on [BP] level. One bit per NIX-RX; index
1212 * enumerated by NPA_BPINTF_E."
1214 if (pfvf->nix_blkaddr == BLKADDR_NIX1)
1215 aq->aura.bp_ena = 1;
1216 aq->aura.nix0_bpid = pfvf->bpid[0];
1218 /* Set backpressure level for RQ's Aura */
1219 aq->aura.bp = RQ_BP_LVL_AURA;
1223 aq->ctype = NPA_AQ_CTYPE_AURA;
1224 aq->op = NPA_AQ_INSTOP_INIT;
1229 static int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1230 int stack_pages, int numptrs, int buf_size)
1232 struct npa_aq_enq_req *aq;
1233 struct otx2_pool *pool;
1236 pool = &pfvf->qset.pool[pool_id];
1237 /* Alloc memory for stack which is used to store buffer pointers */
1238 err = qmem_alloc(pfvf->dev, &pool->stack,
1239 stack_pages, pfvf->hw.stack_pg_bytes);
1243 pool->rbsize = buf_size;
1245 /* Initialize this pool's context via AF */
1246 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1248 /* Shared mbox memory buffer is full, flush it and retry */
1249 err = otx2_sync_mbox_msg(&pfvf->mbox);
1251 qmem_free(pfvf->dev, pool->stack);
1254 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1256 qmem_free(pfvf->dev, pool->stack);
1261 aq->aura_id = pool_id;
1262 aq->pool.stack_base = pool->stack->iova;
1263 aq->pool.stack_caching = 1;
1265 aq->pool.buf_size = buf_size / 128;
1266 aq->pool.stack_max_pages = stack_pages;
1267 aq->pool.shift = ilog2(numptrs) - 8;
1268 aq->pool.ptr_start = 0;
1269 aq->pool.ptr_end = ~0ULL;
1272 aq->ctype = NPA_AQ_CTYPE_POOL;
1273 aq->op = NPA_AQ_INSTOP_INIT;
1278 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1280 int qidx, pool_id, stack_pages, num_sqbs;
1281 struct otx2_qset *qset = &pfvf->qset;
1282 struct otx2_hw *hw = &pfvf->hw;
1283 struct otx2_snd_queue *sq;
1284 struct otx2_pool *pool;
1288 /* Calculate number of SQBs needed.
1290 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1291 * Last SQE is used for pointing to next SQB.
1293 num_sqbs = (hw->sqb_size / 128) - 1;
1294 num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1296 /* Get no of stack pages needed */
1298 (num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1300 for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1301 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1302 /* Initialize aura context */
1303 err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1307 /* Initialize pool context */
1308 err = otx2_pool_init(pfvf, pool_id, stack_pages,
1309 num_sqbs, hw->sqb_size);
1314 /* Flush accumulated messages */
1315 err = otx2_sync_mbox_msg(&pfvf->mbox);
1319 /* Allocate pointers and free them to aura/pool */
1320 for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1321 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1322 pool = &pfvf->qset.pool[pool_id];
1324 sq = &qset->sq[qidx];
1326 sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(*sq->sqb_ptrs), GFP_KERNEL);
1327 if (!sq->sqb_ptrs) {
1332 for (ptr = 0; ptr < num_sqbs; ptr++) {
1333 err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1337 pfvf->hw_ops->aura_freeptr(pfvf, pool_id, bufptr);
1339 sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1344 return err ? -ENOMEM : 0;
1347 otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1348 otx2_aura_pool_free(pfvf);
1352 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1354 struct otx2_hw *hw = &pfvf->hw;
1355 int stack_pages, pool_id, rq;
1356 struct otx2_pool *pool;
1357 int err, ptr, num_ptrs;
1360 num_ptrs = pfvf->qset.rqe_cnt;
1363 (num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1365 for (rq = 0; rq < hw->rx_queues; rq++) {
1366 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1367 /* Initialize aura context */
1368 err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1372 for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1373 err = otx2_pool_init(pfvf, pool_id, stack_pages,
1374 num_ptrs, pfvf->rbsize);
1379 /* Flush accumulated messages */
1380 err = otx2_sync_mbox_msg(&pfvf->mbox);
1385 /* Allocate pointers and free them to aura/pool */
1386 for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1387 pool = &pfvf->qset.pool[pool_id];
1388 for (ptr = 0; ptr < num_ptrs; ptr++) {
1389 err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1392 pfvf->hw_ops->aura_freeptr(pfvf, pool_id,
1393 bufptr + OTX2_HEAD_ROOM);
1398 return err ? -ENOMEM : 0;
1400 otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1401 otx2_aura_pool_free(pfvf);
1405 int otx2_config_npa(struct otx2_nic *pfvf)
1407 struct otx2_qset *qset = &pfvf->qset;
1408 struct npa_lf_alloc_req *npalf;
1409 struct otx2_hw *hw = &pfvf->hw;
1412 /* Pool - Stack of free buffer pointers
1413 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1419 qset->pool = devm_kcalloc(pfvf->dev, hw->pool_cnt,
1420 sizeof(struct otx2_pool), GFP_KERNEL);
1424 /* Get memory to put this msg */
1425 npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1429 /* Set aura and pool counts */
1430 npalf->nr_pools = hw->pool_cnt;
1431 aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1432 npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1434 return otx2_sync_mbox_msg(&pfvf->mbox);
1437 int otx2_detach_resources(struct mbox *mbox)
1439 struct rsrc_detach *detach;
1441 mutex_lock(&mbox->lock);
1442 detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1444 mutex_unlock(&mbox->lock);
1449 detach->partial = false;
1451 /* Send detach request to AF */
1452 otx2_mbox_msg_send(&mbox->mbox, 0);
1453 mutex_unlock(&mbox->lock);
1456 EXPORT_SYMBOL(otx2_detach_resources);
1458 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1460 struct rsrc_attach *attach;
1461 struct msg_req *msix;
1464 mutex_lock(&pfvf->mbox.lock);
1465 /* Get memory to put this msg */
1466 attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1468 mutex_unlock(&pfvf->mbox.lock);
1472 attach->npalf = true;
1473 attach->nixlf = true;
1475 /* Send attach request to AF */
1476 err = otx2_sync_mbox_msg(&pfvf->mbox);
1478 mutex_unlock(&pfvf->mbox.lock);
1482 pfvf->nix_blkaddr = BLKADDR_NIX0;
1484 /* If the platform has two NIX blocks then LF may be
1485 * allocated from NIX1.
1487 if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1488 pfvf->nix_blkaddr = BLKADDR_NIX1;
1490 /* Get NPA and NIX MSIX vector offsets */
1491 msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1493 mutex_unlock(&pfvf->mbox.lock);
1497 err = otx2_sync_mbox_msg(&pfvf->mbox);
1499 mutex_unlock(&pfvf->mbox.lock);
1502 mutex_unlock(&pfvf->mbox.lock);
1504 if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1505 pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1507 "RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1513 EXPORT_SYMBOL(otx2_attach_npa_nix);
1515 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1517 struct hwctx_disable_req *req;
1519 mutex_lock(&mbox->lock);
1520 /* Request AQ to disable this context */
1522 req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1524 req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1527 mutex_unlock(&mbox->lock);
1533 if (otx2_sync_mbox_msg(mbox))
1534 dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1537 mutex_unlock(&mbox->lock);
1540 int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
1542 struct nix_bp_cfg_req *req;
1545 req = otx2_mbox_alloc_msg_nix_bp_enable(&pfvf->mbox);
1547 req = otx2_mbox_alloc_msg_nix_bp_disable(&pfvf->mbox);
1554 req->bpid_per_chan = 0;
1556 return otx2_sync_mbox_msg(&pfvf->mbox);
1559 /* Mbox message handlers */
1560 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1561 struct cgx_stats_rsp *rsp)
1565 for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1566 pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1567 for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1568 pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1571 void mbox_handler_cgx_fec_stats(struct otx2_nic *pfvf,
1572 struct cgx_fec_stats_rsp *rsp)
1574 pfvf->hw.cgx_fec_corr_blks += rsp->fec_corr_blks;
1575 pfvf->hw.cgx_fec_uncorr_blks += rsp->fec_uncorr_blks;
1578 void mbox_handler_nix_txsch_alloc(struct otx2_nic *pf,
1579 struct nix_txsch_alloc_rsp *rsp)
1583 /* Setup transmit scheduler list */
1584 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
1585 for (schq = 0; schq < rsp->schq[lvl]; schq++)
1586 pf->hw.txschq_list[lvl][schq] =
1587 rsp->schq_list[lvl][schq];
1589 pf->hw.txschq_link_cfg_lvl = rsp->link_cfg_lvl;
1591 EXPORT_SYMBOL(mbox_handler_nix_txsch_alloc);
1593 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1594 struct npa_lf_alloc_rsp *rsp)
1596 pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1597 pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1599 EXPORT_SYMBOL(mbox_handler_npa_lf_alloc);
1601 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1602 struct nix_lf_alloc_rsp *rsp)
1604 pfvf->hw.sqb_size = rsp->sqb_size;
1605 pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1606 pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1607 pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1608 pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1609 pfvf->hw.cgx_links = rsp->cgx_links;
1610 pfvf->hw.lbk_links = rsp->lbk_links;
1611 pfvf->hw.tx_link = rsp->tx_link;
1613 EXPORT_SYMBOL(mbox_handler_nix_lf_alloc);
1615 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1616 struct msix_offset_rsp *rsp)
1618 pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1619 pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1621 EXPORT_SYMBOL(mbox_handler_msix_offset);
1623 void mbox_handler_nix_bp_enable(struct otx2_nic *pfvf,
1624 struct nix_bp_cfg_rsp *rsp)
1628 for (chan = 0; chan < rsp->chan_cnt; chan++) {
1629 chan_id = ((rsp->chan_bpid[chan] >> 10) & 0x7F);
1630 pfvf->bpid[chan_id] = rsp->chan_bpid[chan] & 0x3FF;
1633 EXPORT_SYMBOL(mbox_handler_nix_bp_enable);
1635 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1637 struct otx2_qset *qset = &pfvf->qset;
1638 struct otx2_hw *hw = &pfvf->hw;
1641 for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1644 int vector = pci_irq_vector(pfvf->pdev, irq);
1646 irq_set_affinity_hint(vector, NULL);
1647 free_cpumask_var(hw->affinity_mask[irq]);
1648 free_irq(vector, &qset->napi[qidx]);
1652 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1654 struct otx2_hw *hw = &pfvf->hw;
1655 int vec, cpu, irq, cint;
1657 vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1658 cpu = cpumask_first(cpu_online_mask);
1661 for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1662 if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1665 cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1667 irq = pci_irq_vector(pfvf->pdev, vec);
1668 irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1670 cpu = cpumask_next(cpu, cpu_online_mask);
1671 if (unlikely(cpu >= nr_cpu_ids))
1676 u16 otx2_get_max_mtu(struct otx2_nic *pfvf)
1678 struct nix_hw_info *rsp;
1679 struct msg_req *req;
1683 mutex_lock(&pfvf->mbox.lock);
1685 req = otx2_mbox_alloc_msg_nix_get_hw_info(&pfvf->mbox);
1691 rc = otx2_sync_mbox_msg(&pfvf->mbox);
1693 rsp = (struct nix_hw_info *)
1694 otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
1696 /* HW counts VLAN insertion bytes (8 for double tag)
1697 * irrespective of whether SQE is requesting to insert VLAN
1698 * in the packet or not. Hence these 8 bytes have to be
1699 * discounted from max packet size otherwise HW will throw
1702 max_mtu = rsp->max_mtu - 8 - OTX2_ETH_HLEN;
1704 /* Also save DWRR MTU, needed for DWRR weight calculation */
1705 pfvf->hw.dwrr_mtu = rsp->rpm_dwrr_mtu;
1706 if (!pfvf->hw.dwrr_mtu)
1707 pfvf->hw.dwrr_mtu = 1;
1711 mutex_unlock(&pfvf->mbox.lock);
1714 "Failed to get MTU from hardware setting default value(1500)\n");
1719 EXPORT_SYMBOL(otx2_get_max_mtu);
1721 #define M(_name, _id, _fn_name, _req_type, _rsp_type) \
1723 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf, \
1724 struct _req_type *req, \
1725 struct _rsp_type *rsp) \
1727 /* Nothing to do here */ \
1730 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
1731 MBOX_UP_CGX_MESSAGES