1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell OcteonTx2 RVU Ethernet driver
4 * Copyright (C) 2020 Marvell International Ltd.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
11 #include <linux/interrupt.h>
12 #include <linux/pci.h>
16 #include "otx2_common.h"
17 #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 otx2_mbox_lock(&pfvf->mbox);
53 req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
55 otx2_mbox_unlock(&pfvf->mbox);
59 otx2_sync_mbox_msg(&pfvf->mbox);
60 otx2_mbox_unlock(&pfvf->mbox);
63 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
65 struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
70 otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
74 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
76 struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
81 otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
85 void otx2_get_dev_stats(struct otx2_nic *pfvf)
87 struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
89 #define OTX2_GET_RX_STATS(reg) \
90 otx2_read64(pfvf, NIX_LF_RX_STATX(reg))
91 #define OTX2_GET_TX_STATS(reg) \
92 otx2_read64(pfvf, NIX_LF_TX_STATX(reg))
94 dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
95 dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
96 dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
97 dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
98 dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
99 dev_stats->rx_frames = dev_stats->rx_bcast_frames +
100 dev_stats->rx_mcast_frames +
101 dev_stats->rx_ucast_frames;
103 dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
104 dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
105 dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
106 dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
107 dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
108 dev_stats->tx_frames = dev_stats->tx_bcast_frames +
109 dev_stats->tx_mcast_frames +
110 dev_stats->tx_ucast_frames;
113 void otx2_get_stats64(struct net_device *netdev,
114 struct rtnl_link_stats64 *stats)
116 struct otx2_nic *pfvf = netdev_priv(netdev);
117 struct otx2_dev_stats *dev_stats;
119 otx2_get_dev_stats(pfvf);
121 dev_stats = &pfvf->hw.dev_stats;
122 stats->rx_bytes = dev_stats->rx_bytes;
123 stats->rx_packets = dev_stats->rx_frames;
124 stats->rx_dropped = dev_stats->rx_drops;
125 stats->multicast = dev_stats->rx_mcast_frames;
127 stats->tx_bytes = dev_stats->tx_bytes;
128 stats->tx_packets = dev_stats->tx_frames;
129 stats->tx_dropped = dev_stats->tx_drops;
132 /* Sync MAC address with RVU AF */
133 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
135 struct nix_set_mac_addr *req;
138 otx2_mbox_lock(&pfvf->mbox);
139 req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
141 otx2_mbox_unlock(&pfvf->mbox);
145 ether_addr_copy(req->mac_addr, mac);
147 err = otx2_sync_mbox_msg(&pfvf->mbox);
148 otx2_mbox_unlock(&pfvf->mbox);
152 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
153 struct net_device *netdev)
155 struct nix_get_mac_addr_rsp *rsp;
156 struct mbox_msghdr *msghdr;
160 otx2_mbox_lock(&pfvf->mbox);
161 req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
163 otx2_mbox_unlock(&pfvf->mbox);
167 err = otx2_sync_mbox_msg(&pfvf->mbox);
169 otx2_mbox_unlock(&pfvf->mbox);
173 msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
174 if (IS_ERR(msghdr)) {
175 otx2_mbox_unlock(&pfvf->mbox);
176 return PTR_ERR(msghdr);
178 rsp = (struct nix_get_mac_addr_rsp *)msghdr;
179 ether_addr_copy(netdev->dev_addr, rsp->mac_addr);
180 otx2_mbox_unlock(&pfvf->mbox);
185 int otx2_set_mac_address(struct net_device *netdev, void *p)
187 struct otx2_nic *pfvf = netdev_priv(netdev);
188 struct sockaddr *addr = p;
190 if (!is_valid_ether_addr(addr->sa_data))
191 return -EADDRNOTAVAIL;
193 if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data))
194 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
201 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
203 struct nix_frs_cfg *req;
206 otx2_mbox_lock(&pfvf->mbox);
207 req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
209 otx2_mbox_unlock(&pfvf->mbox);
213 /* SMQ config limits maximum pkt size that can be transmitted */
214 req->update_smq = true;
215 pfvf->max_frs = mtu + OTX2_ETH_HLEN;
216 req->maxlen = pfvf->max_frs;
218 err = otx2_sync_mbox_msg(&pfvf->mbox);
219 otx2_mbox_unlock(&pfvf->mbox);
223 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
225 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
226 struct nix_rss_flowkey_cfg *req;
229 otx2_mbox_lock(&pfvf->mbox);
230 req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
232 otx2_mbox_unlock(&pfvf->mbox);
235 req->mcam_index = -1; /* Default or reserved index */
236 req->flowkey_cfg = rss->flowkey_cfg;
237 req->group = DEFAULT_RSS_CONTEXT_GROUP;
239 err = otx2_sync_mbox_msg(&pfvf->mbox);
240 otx2_mbox_unlock(&pfvf->mbox);
244 int otx2_set_rss_table(struct otx2_nic *pfvf)
246 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
247 struct mbox *mbox = &pfvf->mbox;
248 struct nix_aq_enq_req *aq;
251 otx2_mbox_lock(mbox);
252 /* Get memory to put this msg */
253 for (idx = 0; idx < rss->rss_size; idx++) {
254 aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
256 /* The shared memory buffer can be full.
259 err = otx2_sync_mbox_msg(mbox);
261 otx2_mbox_unlock(mbox);
264 aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
266 otx2_mbox_unlock(mbox);
271 aq->rss.rq = rss->ind_tbl[idx];
275 aq->ctype = NIX_AQ_CTYPE_RSS;
276 aq->op = NIX_AQ_INSTOP_INIT;
278 err = otx2_sync_mbox_msg(mbox);
279 otx2_mbox_unlock(mbox);
283 void otx2_set_rss_key(struct otx2_nic *pfvf)
285 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
286 u64 *key = (u64 *)&rss->key[4];
289 /* 352bit or 44byte key needs to be configured as below
290 * NIX_LF_RX_SECRETX0 = key<351:288>
291 * NIX_LF_RX_SECRETX1 = key<287:224>
292 * NIX_LF_RX_SECRETX2 = key<223:160>
293 * NIX_LF_RX_SECRETX3 = key<159:96>
294 * NIX_LF_RX_SECRETX4 = key<95:32>
295 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
297 otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
298 (u64)(*((u32 *)&rss->key)) << 32);
299 idx = sizeof(rss->key) / sizeof(u64);
302 otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
306 int otx2_rss_init(struct otx2_nic *pfvf)
308 struct otx2_rss_info *rss = &pfvf->hw.rss_info;
311 rss->rss_size = sizeof(rss->ind_tbl);
313 /* Init RSS key if it is not setup already */
315 netdev_rss_key_fill(rss->key, sizeof(rss->key));
316 otx2_set_rss_key(pfvf);
318 if (!netif_is_rxfh_configured(pfvf->netdev)) {
319 /* Default indirection table */
320 for (idx = 0; idx < rss->rss_size; idx++)
322 ethtool_rxfh_indir_default(idx,
325 ret = otx2_set_rss_table(pfvf);
329 /* Flowkey or hash config to be used for generating flow tag */
330 rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
331 NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
332 NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
333 NIX_FLOW_KEY_TYPE_SCTP;
335 ret = otx2_set_flowkey_cfg(pfvf);
343 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
345 /* Configure CQE interrupt coalescing parameters
347 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
348 * set 1 less than cq_ecount_wait. And cq_time_wait is in
349 * usecs, convert that to 100ns count.
351 otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
352 ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
353 ((u64)pfvf->hw.cq_qcount_wait << 32) |
354 (pfvf->hw.cq_ecount_wait - 1));
357 dma_addr_t otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
362 /* Check if request can be accommodated in previous allocated page */
363 if (pool->page && ((pool->page_offset + pool->rbsize) <=
364 (PAGE_SIZE << pool->rbpage_order))) {
371 /* Allocate a new page */
372 pool->page = alloc_pages(gfp | __GFP_COMP | __GFP_NOWARN,
374 if (unlikely(!pool->page))
377 pool->page_offset = 0;
379 iova = (u64)otx2_dma_map_page(pfvf, pool->page, pool->page_offset,
380 pool->rbsize, DMA_FROM_DEVICE);
382 if (!pool->page_offset)
383 __free_pages(pool->page, pool->rbpage_order);
387 pool->page_offset += pool->rbsize;
391 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
393 struct otx2_nic *pfvf = netdev_priv(netdev);
395 schedule_work(&pfvf->reset_task);
398 void otx2_get_mac_from_af(struct net_device *netdev)
400 struct otx2_nic *pfvf = netdev_priv(netdev);
403 err = otx2_hw_get_mac_addr(pfvf, netdev);
405 dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
407 /* If AF doesn't provide a valid MAC, generate a random one */
408 if (!is_valid_ether_addr(netdev->dev_addr))
409 eth_hw_addr_random(netdev);
412 static int otx2_get_link(struct otx2_nic *pfvf)
418 if (pfvf->hw.tx_chan_base >= CGX_CHAN_BASE) {
419 map = pfvf->hw.tx_chan_base & 0x7FF;
420 link = 4 * ((map >> 8) & 0xF) + ((map >> 4) & 0xF);
423 if (pfvf->hw.tx_chan_base < SDP_CHAN_BASE)
429 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl)
431 struct otx2_hw *hw = &pfvf->hw;
432 struct nix_txschq_config *req;
435 req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
442 schq = hw->txschq_list[lvl][0];
443 /* Set topology e.t.c configuration */
444 if (lvl == NIX_TXSCH_LVL_SMQ) {
445 req->reg[0] = NIX_AF_SMQX_CFG(schq);
446 req->regval[0] = ((pfvf->netdev->mtu + OTX2_ETH_HLEN) << 8) |
449 req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
453 parent = hw->txschq_list[NIX_TXSCH_LVL_TL4][0];
454 req->reg[1] = NIX_AF_MDQX_PARENT(schq);
455 req->regval[1] = parent << 16;
457 /* Set DWRR quantum */
458 req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
459 req->regval[2] = DFLT_RR_QTM;
460 } else if (lvl == NIX_TXSCH_LVL_TL4) {
461 parent = hw->txschq_list[NIX_TXSCH_LVL_TL3][0];
462 req->reg[0] = NIX_AF_TL4X_PARENT(schq);
463 req->regval[0] = parent << 16;
465 req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
466 req->regval[1] = DFLT_RR_QTM;
467 } else if (lvl == NIX_TXSCH_LVL_TL3) {
468 parent = hw->txschq_list[NIX_TXSCH_LVL_TL2][0];
469 req->reg[0] = NIX_AF_TL3X_PARENT(schq);
470 req->regval[0] = parent << 16;
472 req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
473 req->regval[1] = DFLT_RR_QTM;
474 } else if (lvl == NIX_TXSCH_LVL_TL2) {
475 parent = hw->txschq_list[NIX_TXSCH_LVL_TL1][0];
476 req->reg[0] = NIX_AF_TL2X_PARENT(schq);
477 req->regval[0] = parent << 16;
480 req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
481 req->regval[1] = TXSCH_TL1_DFLT_RR_PRIO << 24 | DFLT_RR_QTM;
484 req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq,
485 otx2_get_link(pfvf));
486 /* Enable this queue and backpressure */
487 req->regval[2] = BIT_ULL(13) | BIT_ULL(12);
489 } else if (lvl == NIX_TXSCH_LVL_TL1) {
490 /* Default config for TL1.
491 * For VF this is always ignored.
494 /* Set DWRR quantum */
495 req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
496 req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
499 req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
500 req->regval[1] = (TXSCH_TL1_DFLT_RR_PRIO << 1);
503 req->reg[2] = NIX_AF_TL1X_CIR(schq);
507 return otx2_sync_mbox_msg(&pfvf->mbox);
510 int otx2_txsch_alloc(struct otx2_nic *pfvf)
512 struct nix_txsch_alloc_req *req;
515 /* Get memory to put this msg */
516 req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
520 /* Request one schq per level */
521 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
524 return otx2_sync_mbox_msg(&pfvf->mbox);
527 int otx2_txschq_stop(struct otx2_nic *pfvf)
529 struct nix_txsch_free_req *free_req;
532 otx2_mbox_lock(&pfvf->mbox);
533 /* Free the transmit schedulers */
534 free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
536 otx2_mbox_unlock(&pfvf->mbox);
540 free_req->flags = TXSCHQ_FREE_ALL;
541 err = otx2_sync_mbox_msg(&pfvf->mbox);
542 otx2_mbox_unlock(&pfvf->mbox);
544 /* Clear the txschq list */
545 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
546 for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
547 pfvf->hw.txschq_list[lvl][schq] = 0;
552 void otx2_sqb_flush(struct otx2_nic *pfvf)
554 int qidx, sqe_tail, sqe_head;
557 ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
558 for (qidx = 0; qidx < pfvf->hw.tx_queues; qidx++) {
559 incr = (u64)qidx << 32;
561 val = otx2_atomic64_add(incr, ptr);
562 sqe_head = (val >> 20) & 0x3F;
563 sqe_tail = (val >> 28) & 0x3F;
564 if (sqe_head == sqe_tail)
571 /* RED and drop levels of CQ on packet reception.
572 * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
574 #define RQ_PASS_LVL_CQ(skid, qsize) ((((skid) + 16) * 256) / (qsize))
575 #define RQ_DROP_LVL_CQ(skid, qsize) (((skid) * 256) / (qsize))
577 /* RED and drop levels of AURA for packet reception.
578 * For AURA level is measure of fullness (0x0 = empty, 255 = full).
579 * Eg: For RQ length 1K, for pass/drop level 204/230.
580 * RED accepts pkts if free pointers > 102 & <= 205.
581 * Drops pkts if free pointers < 102.
583 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
584 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
586 /* Send skid of 2000 packets required for CQ size of 4K CQEs. */
587 #define SEND_CQ_SKID 2000
589 static int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
591 struct otx2_qset *qset = &pfvf->qset;
592 struct nix_aq_enq_req *aq;
594 /* Get memory to put this msg */
595 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
601 aq->rq.pb_caching = 1;
602 aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
603 aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
604 aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
605 aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
607 aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
608 aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
609 aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
610 aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
611 aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
612 aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
616 aq->ctype = NIX_AQ_CTYPE_RQ;
617 aq->op = NIX_AQ_INSTOP_INIT;
619 return otx2_sync_mbox_msg(&pfvf->mbox);
622 static int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
624 struct otx2_qset *qset = &pfvf->qset;
625 struct otx2_snd_queue *sq;
626 struct nix_aq_enq_req *aq;
627 struct otx2_pool *pool;
630 pool = &pfvf->qset.pool[sqb_aura];
631 sq = &qset->sq[qidx];
632 sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
633 sq->sqe_cnt = qset->sqe_cnt;
635 err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
639 err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
644 sq->sqe_base = sq->sqe->base;
645 sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
650 sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
651 sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
652 /* Set SQE threshold to 10% of total SQEs */
653 sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
654 sq->aura_id = sqb_aura;
655 sq->aura_fc_addr = pool->fc_addr->base;
656 sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
657 sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
662 /* Get memory to put this msg */
663 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
667 aq->sq.cq = pfvf->hw.rx_queues + qidx;
668 aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
671 /* Only one SMQ is allocated, map all SQ's to that SMQ */
672 aq->sq.smq = pfvf->hw.txschq_list[NIX_TXSCH_LVL_SMQ][0];
673 aq->sq.smq_rr_quantum = DFLT_RR_QTM;
674 aq->sq.default_chan = pfvf->hw.tx_chan_base;
675 aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
676 aq->sq.sqb_aura = sqb_aura;
677 aq->sq.sq_int_ena = NIX_SQINT_BITS;
679 /* Due pipelining impact minimum 2000 unused SQ CQE's
680 * need to maintain to avoid CQ overflow.
682 aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (sq->sqe_cnt));
686 aq->ctype = NIX_AQ_CTYPE_SQ;
687 aq->op = NIX_AQ_INSTOP_INIT;
689 return otx2_sync_mbox_msg(&pfvf->mbox);
692 static int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
694 struct otx2_qset *qset = &pfvf->qset;
695 struct nix_aq_enq_req *aq;
696 struct otx2_cq_queue *cq;
699 cq = &qset->cq[qidx];
701 if (qidx < pfvf->hw.rx_queues) {
704 cq->cqe_cnt = qset->rqe_cnt;
707 cq->cint_idx = qidx - pfvf->hw.rx_queues;
708 cq->cqe_cnt = qset->sqe_cnt;
710 cq->cqe_size = pfvf->qset.xqe_size;
712 /* Allocate memory for CQEs */
713 err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
717 /* Save CQE CPU base for faster reference */
718 cq->cqe_base = cq->cqe->base;
719 /* In case where all RQs auras point to single pool,
720 * all CQs receive buffer pool also point to same pool.
722 pool_id = ((cq->cq_type == CQ_RX) &&
723 (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
724 cq->rbpool = &qset->pool[pool_id];
725 cq->refill_task_sched = false;
727 /* Get memory to put this msg */
728 aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
733 aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
735 aq->cq.base = cq->cqe->iova;
736 aq->cq.cint_idx = cq->cint_idx;
737 aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
739 aq->cq.avg_level = 255;
741 if (qidx < pfvf->hw.rx_queues) {
742 aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
748 aq->ctype = NIX_AQ_CTYPE_CQ;
749 aq->op = NIX_AQ_INSTOP_INIT;
751 return otx2_sync_mbox_msg(&pfvf->mbox);
754 static void otx2_pool_refill_task(struct work_struct *work)
756 struct otx2_cq_queue *cq;
757 struct otx2_pool *rbpool;
758 struct refill_work *wrk;
759 int qidx, free_ptrs = 0;
760 struct otx2_nic *pfvf;
763 wrk = container_of(work, struct refill_work, pool_refill_work.work);
765 qidx = wrk - pfvf->refill_wrk;
766 cq = &pfvf->qset.cq[qidx];
768 free_ptrs = cq->pool_ptrs;
770 while (cq->pool_ptrs) {
771 bufptr = otx2_alloc_rbuf(pfvf, rbpool, GFP_KERNEL);
773 /* Schedule a WQ if we fails to free atleast half of the
774 * pointers else enable napi for this RQ.
776 if (!((free_ptrs - cq->pool_ptrs) > free_ptrs / 2)) {
777 struct delayed_work *dwork;
779 dwork = &wrk->pool_refill_work;
780 schedule_delayed_work(dwork,
781 msecs_to_jiffies(100));
783 cq->refill_task_sched = false;
787 otx2_aura_freeptr(pfvf, qidx, bufptr + OTX2_HEAD_ROOM);
790 cq->refill_task_sched = false;
793 int otx2_config_nix_queues(struct otx2_nic *pfvf)
797 /* Initialize RX queues */
798 for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
799 u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
801 err = otx2_rq_init(pfvf, qidx, lpb_aura);
806 /* Initialize TX queues */
807 for (qidx = 0; qidx < pfvf->hw.tx_queues; qidx++) {
808 u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
810 err = otx2_sq_init(pfvf, qidx, sqb_aura);
815 /* Initialize completion queues */
816 for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
817 err = otx2_cq_init(pfvf, qidx);
822 /* Initialize work queue for receive buffer refill */
823 pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
824 sizeof(struct refill_work), GFP_KERNEL);
825 if (!pfvf->refill_wrk)
828 for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
829 pfvf->refill_wrk[qidx].pf = pfvf;
830 INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
831 otx2_pool_refill_task);
836 int otx2_config_nix(struct otx2_nic *pfvf)
838 struct nix_lf_alloc_req *nixlf;
839 struct nix_lf_alloc_rsp *rsp;
842 pfvf->qset.xqe_size = NIX_XQESZ_W16 ? 128 : 512;
844 /* Get memory to put this msg */
845 nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
849 /* Set RQ/SQ/CQ counts */
850 nixlf->rq_cnt = pfvf->hw.rx_queues;
851 nixlf->sq_cnt = pfvf->hw.tx_queues;
852 nixlf->cq_cnt = pfvf->qset.cq_cnt;
853 nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
854 nixlf->rss_grps = 1; /* Single RSS indir table supported, for now */
855 nixlf->xqe_sz = NIX_XQESZ_W16;
856 /* We don't know absolute NPA LF idx attached.
857 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
858 * NPA LF attached to this RVU PF/VF.
860 nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
861 /* Disable alignment pad, enable L2 length check,
862 * enable L4 TCP/UDP checksum verification.
864 nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
866 err = otx2_sync_mbox_msg(&pfvf->mbox);
870 rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
881 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
883 struct otx2_qset *qset = &pfvf->qset;
884 struct otx2_hw *hw = &pfvf->hw;
885 struct otx2_snd_queue *sq;
889 for (qidx = 0; qidx < hw->tx_queues; qidx++) {
890 sq = &qset->sq[qidx];
893 for (sqb = 0; sqb < sq->sqb_count; sqb++) {
894 if (!sq->sqb_ptrs[sqb])
896 iova = sq->sqb_ptrs[sqb];
897 pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
898 dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
900 DMA_ATTR_SKIP_CPU_SYNC);
901 put_page(virt_to_page(phys_to_virt(pa)));
907 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
909 int pool_id, pool_start = 0, pool_end = 0, size = 0;
912 if (type == AURA_NIX_SQ) {
913 pool_start = otx2_get_pool_idx(pfvf, type, 0);
914 pool_end = pool_start + pfvf->hw.sqpool_cnt;
915 size = pfvf->hw.sqb_size;
917 if (type == AURA_NIX_RQ) {
918 pool_start = otx2_get_pool_idx(pfvf, type, 0);
919 pool_end = pfvf->hw.rqpool_cnt;
923 /* Free SQB and RQB pointers from the aura pool */
924 for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
925 iova = otx2_aura_allocptr(pfvf, pool_id);
927 if (type == AURA_NIX_RQ)
928 iova -= OTX2_HEAD_ROOM;
930 pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
931 dma_unmap_page_attrs(pfvf->dev, iova, size,
933 DMA_ATTR_SKIP_CPU_SYNC);
934 put_page(virt_to_page(phys_to_virt(pa)));
935 iova = otx2_aura_allocptr(pfvf, pool_id);
940 void otx2_aura_pool_free(struct otx2_nic *pfvf)
942 struct otx2_pool *pool;
945 if (!pfvf->qset.pool)
948 for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
949 pool = &pfvf->qset.pool[pool_id];
950 qmem_free(pfvf->dev, pool->stack);
951 qmem_free(pfvf->dev, pool->fc_addr);
953 devm_kfree(pfvf->dev, pfvf->qset.pool);
956 static int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
957 int pool_id, int numptrs)
959 struct npa_aq_enq_req *aq;
960 struct otx2_pool *pool;
963 pool = &pfvf->qset.pool[pool_id];
965 /* Allocate memory for HW to update Aura count.
966 * Alloc one cache line, so that it fits all FC_STYPE modes.
968 if (!pool->fc_addr) {
969 err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
974 /* Initialize this aura's context via AF */
975 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
977 /* Shared mbox memory buffer is full, flush it and retry */
978 err = otx2_sync_mbox_msg(&pfvf->mbox);
981 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
986 aq->aura_id = aura_id;
987 /* Will be filled by AF with correct pool context address */
988 aq->aura.pool_addr = pool_id;
989 aq->aura.pool_caching = 1;
990 aq->aura.shift = ilog2(numptrs) - 8;
991 aq->aura.count = numptrs;
992 aq->aura.limit = numptrs;
993 aq->aura.avg_level = 255;
996 aq->aura.fc_addr = pool->fc_addr->iova;
997 aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1000 aq->ctype = NPA_AQ_CTYPE_AURA;
1001 aq->op = NPA_AQ_INSTOP_INIT;
1006 static int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1007 int stack_pages, int numptrs, int buf_size)
1009 struct npa_aq_enq_req *aq;
1010 struct otx2_pool *pool;
1013 pool = &pfvf->qset.pool[pool_id];
1014 /* Alloc memory for stack which is used to store buffer pointers */
1015 err = qmem_alloc(pfvf->dev, &pool->stack,
1016 stack_pages, pfvf->hw.stack_pg_bytes);
1020 pool->rbsize = buf_size;
1021 pool->rbpage_order = get_order(buf_size);
1023 /* Initialize this pool's context via AF */
1024 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1026 /* Shared mbox memory buffer is full, flush it and retry */
1027 err = otx2_sync_mbox_msg(&pfvf->mbox);
1029 qmem_free(pfvf->dev, pool->stack);
1032 aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1034 qmem_free(pfvf->dev, pool->stack);
1039 aq->aura_id = pool_id;
1040 aq->pool.stack_base = pool->stack->iova;
1041 aq->pool.stack_caching = 1;
1043 aq->pool.buf_size = buf_size / 128;
1044 aq->pool.stack_max_pages = stack_pages;
1045 aq->pool.shift = ilog2(numptrs) - 8;
1046 aq->pool.ptr_start = 0;
1047 aq->pool.ptr_end = ~0ULL;
1050 aq->ctype = NPA_AQ_CTYPE_POOL;
1051 aq->op = NPA_AQ_INSTOP_INIT;
1056 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1058 int qidx, pool_id, stack_pages, num_sqbs;
1059 struct otx2_qset *qset = &pfvf->qset;
1060 struct otx2_hw *hw = &pfvf->hw;
1061 struct otx2_snd_queue *sq;
1062 struct otx2_pool *pool;
1066 /* Calculate number of SQBs needed.
1068 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1069 * Last SQE is used for pointing to next SQB.
1071 num_sqbs = (hw->sqb_size / 128) - 1;
1072 num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1074 /* Get no of stack pages needed */
1076 (num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1078 for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1079 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1080 /* Initialize aura context */
1081 err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1085 /* Initialize pool context */
1086 err = otx2_pool_init(pfvf, pool_id, stack_pages,
1087 num_sqbs, hw->sqb_size);
1092 /* Flush accumulated messages */
1093 err = otx2_sync_mbox_msg(&pfvf->mbox);
1097 /* Allocate pointers and free them to aura/pool */
1098 for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1099 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1100 pool = &pfvf->qset.pool[pool_id];
1102 sq = &qset->sq[qidx];
1104 sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(u64 *), GFP_KERNEL);
1108 for (ptr = 0; ptr < num_sqbs; ptr++) {
1109 bufptr = otx2_alloc_rbuf(pfvf, pool, GFP_KERNEL);
1112 otx2_aura_freeptr(pfvf, pool_id, bufptr);
1113 sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1115 otx2_get_page(pool);
1120 otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1121 otx2_aura_pool_free(pfvf);
1125 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1127 struct otx2_hw *hw = &pfvf->hw;
1128 int stack_pages, pool_id, rq;
1129 struct otx2_pool *pool;
1130 int err, ptr, num_ptrs;
1133 num_ptrs = pfvf->qset.rqe_cnt;
1136 (num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1138 for (rq = 0; rq < hw->rx_queues; rq++) {
1139 pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1140 /* Initialize aura context */
1141 err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1145 for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1146 err = otx2_pool_init(pfvf, pool_id, stack_pages,
1147 num_ptrs, pfvf->rbsize);
1152 /* Flush accumulated messages */
1153 err = otx2_sync_mbox_msg(&pfvf->mbox);
1157 /* Allocate pointers and free them to aura/pool */
1158 for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1159 pool = &pfvf->qset.pool[pool_id];
1160 for (ptr = 0; ptr < num_ptrs; ptr++) {
1161 bufptr = otx2_alloc_rbuf(pfvf, pool, GFP_KERNEL);
1164 otx2_aura_freeptr(pfvf, pool_id,
1165 bufptr + OTX2_HEAD_ROOM);
1167 otx2_get_page(pool);
1172 otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1173 otx2_aura_pool_free(pfvf);
1177 int otx2_config_npa(struct otx2_nic *pfvf)
1179 struct otx2_qset *qset = &pfvf->qset;
1180 struct npa_lf_alloc_req *npalf;
1181 struct otx2_hw *hw = &pfvf->hw;
1184 /* Pool - Stack of free buffer pointers
1185 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1191 qset->pool = devm_kzalloc(pfvf->dev, sizeof(struct otx2_pool) *
1192 hw->pool_cnt, GFP_KERNEL);
1196 /* Get memory to put this msg */
1197 npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1201 /* Set aura and pool counts */
1202 npalf->nr_pools = hw->pool_cnt;
1203 aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1204 npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1206 return otx2_sync_mbox_msg(&pfvf->mbox);
1209 int otx2_detach_resources(struct mbox *mbox)
1211 struct rsrc_detach *detach;
1213 otx2_mbox_lock(mbox);
1214 detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1216 otx2_mbox_unlock(mbox);
1221 detach->partial = false;
1223 /* Send detach request to AF */
1224 otx2_mbox_msg_send(&mbox->mbox, 0);
1225 otx2_mbox_unlock(mbox);
1229 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1231 struct rsrc_attach *attach;
1232 struct msg_req *msix;
1235 otx2_mbox_lock(&pfvf->mbox);
1236 /* Get memory to put this msg */
1237 attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1239 otx2_mbox_unlock(&pfvf->mbox);
1243 attach->npalf = true;
1244 attach->nixlf = true;
1246 /* Send attach request to AF */
1247 err = otx2_sync_mbox_msg(&pfvf->mbox);
1249 otx2_mbox_unlock(&pfvf->mbox);
1253 pfvf->nix_blkaddr = BLKADDR_NIX0;
1255 /* If the platform has two NIX blocks then LF may be
1256 * allocated from NIX1.
1258 if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1259 pfvf->nix_blkaddr = BLKADDR_NIX1;
1261 /* Get NPA and NIX MSIX vector offsets */
1262 msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1264 otx2_mbox_unlock(&pfvf->mbox);
1268 err = otx2_sync_mbox_msg(&pfvf->mbox);
1270 otx2_mbox_unlock(&pfvf->mbox);
1273 otx2_mbox_unlock(&pfvf->mbox);
1275 if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1276 pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1278 "RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1285 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1287 struct hwctx_disable_req *req;
1289 otx2_mbox_lock(mbox);
1290 /* Request AQ to disable this context */
1292 req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1294 req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1297 otx2_mbox_unlock(mbox);
1303 if (otx2_sync_mbox_msg(mbox))
1304 dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1307 otx2_mbox_unlock(mbox);
1310 /* Mbox message handlers */
1311 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1312 struct cgx_stats_rsp *rsp)
1316 for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1317 pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1318 for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1319 pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1322 void mbox_handler_nix_txsch_alloc(struct otx2_nic *pf,
1323 struct nix_txsch_alloc_rsp *rsp)
1327 /* Setup transmit scheduler list */
1328 for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
1329 for (schq = 0; schq < rsp->schq[lvl]; schq++)
1330 pf->hw.txschq_list[lvl][schq] =
1331 rsp->schq_list[lvl][schq];
1334 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1335 struct npa_lf_alloc_rsp *rsp)
1337 pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1338 pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1341 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1342 struct nix_lf_alloc_rsp *rsp)
1344 pfvf->hw.sqb_size = rsp->sqb_size;
1345 pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1346 pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1347 pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1348 pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1351 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1352 struct msix_offset_rsp *rsp)
1354 pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1355 pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1358 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1360 struct otx2_qset *qset = &pfvf->qset;
1361 struct otx2_hw *hw = &pfvf->hw;
1364 for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1367 int vector = pci_irq_vector(pfvf->pdev, irq);
1369 irq_set_affinity_hint(vector, NULL);
1370 free_cpumask_var(hw->affinity_mask[irq]);
1371 free_irq(vector, &qset->napi[qidx]);
1375 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1377 struct otx2_hw *hw = &pfvf->hw;
1378 int vec, cpu, irq, cint;
1380 vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1381 cpu = cpumask_first(cpu_online_mask);
1384 for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1385 if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1388 cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1390 irq = pci_irq_vector(pfvf->pdev, vec);
1391 irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1393 cpu = cpumask_next(cpu, cpu_online_mask);
1394 if (unlikely(cpu >= nr_cpu_ids))
1399 #define M(_name, _id, _fn_name, _req_type, _rsp_type) \
1401 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf, \
1402 struct _req_type *req, \
1403 struct _rsp_type *rsp) \
1405 /* Nothing to do here */ \
1408 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
1409 MBOX_UP_CGX_MESSAGES