From 43f3f05fb66a0e5616a70e5ddbbd77d0576dacbc Mon Sep 17 00:00:00 2001 From: Nithin Dabilpuram Date: Fri, 3 Apr 2020 14:22:08 +0530 Subject: [PATCH] net/octeontx2: support dynamic topology update Modify resource allocation and freeing logic to support dynamic topology commit while to traffic is flowing. This patch also modifies SQ flush to timeout based on minimum shaper rate configured. SQ flush is further split to pre/post functions to adhere to HW spec of 96XX C0. Signed-off-by: Nithin Dabilpuram Signed-off-by: Krzysztof Kanas --- drivers/common/octeontx2/otx2_dev.h | 9 + drivers/net/octeontx2/otx2_ethdev.c | 3 +- drivers/net/octeontx2/otx2_ethdev.h | 1 + drivers/net/octeontx2/otx2_tm.c | 542 +++++++++++++++++++++------- drivers/net/octeontx2/otx2_tm.h | 7 +- 5 files changed, 422 insertions(+), 140 deletions(-) diff --git a/drivers/common/octeontx2/otx2_dev.h b/drivers/common/octeontx2/otx2_dev.h index 0b0a949b24..13b75e1188 100644 --- a/drivers/common/octeontx2/otx2_dev.h +++ b/drivers/common/octeontx2/otx2_dev.h @@ -46,6 +46,15 @@ ((RVU_PCI_REV_MAJOR(otx2_dev_revid(dev)) == 0x0) && \ (RVU_PCI_REV_MIDR_ID(otx2_dev_revid(dev)) == 0x0)) +#define otx2_dev_is_96xx_Cx(dev) \ + ((RVU_PCI_REV_MAJOR(otx2_dev_revid(dev)) == 0x2) && \ + (RVU_PCI_REV_MIDR_ID(otx2_dev_revid(dev)) == 0x0)) + +#define otx2_dev_is_96xx_C0(dev) \ + ((RVU_PCI_REV_MAJOR(otx2_dev_revid(dev)) == 0x2) && \ + (RVU_PCI_REV_MINOR(otx2_dev_revid(dev)) == 0x0) && \ + (RVU_PCI_REV_MIDR_ID(otx2_dev_revid(dev)) == 0x0)) + struct otx2_dev; /* Link status callback */ diff --git a/drivers/net/octeontx2/otx2_ethdev.c b/drivers/net/octeontx2/otx2_ethdev.c index 21e318d589..f26d42b1f3 100644 --- a/drivers/net/octeontx2/otx2_ethdev.c +++ b/drivers/net/octeontx2/otx2_ethdev.c @@ -998,7 +998,7 @@ otx2_nix_tx_queue_release(void *_txq) otx2_nix_dbg("Releasing txq %u", txq->sq); /* Flush and disable tm */ - otx2_nix_tm_sw_xoff(txq, eth_dev->data->dev_started); + otx2_nix_sq_flush_pre(txq, eth_dev->data->dev_started); /* Free sqb's and disable sq */ nix_sq_uninit(txq); @@ -1007,6 +1007,7 @@ otx2_nix_tx_queue_release(void *_txq) rte_mempool_free(txq->sqb_pool); txq->sqb_pool = NULL; } + otx2_nix_sq_flush_post(txq); rte_free(txq); } diff --git a/drivers/net/octeontx2/otx2_ethdev.h b/drivers/net/octeontx2/otx2_ethdev.h index 26690185c6..a7faef705c 100644 --- a/drivers/net/octeontx2/otx2_ethdev.h +++ b/drivers/net/octeontx2/otx2_ethdev.h @@ -307,6 +307,7 @@ struct otx2_eth_dev { uint16_t link_cfg_lvl; uint16_t tm_flags; uint16_t tm_leaf_cnt; + uint64_t tm_rate_min; struct otx2_nix_tm_node_list node_list; struct otx2_nix_tm_shaper_profile_list shaper_profile_list; struct otx2_rss_info rss_info; diff --git a/drivers/net/octeontx2/otx2_tm.c b/drivers/net/octeontx2/otx2_tm.c index 108f44ca70..8dd29962cd 100644 --- a/drivers/net/octeontx2/otx2_tm.c +++ b/drivers/net/octeontx2/otx2_tm.c @@ -59,8 +59,16 @@ static bool nix_tm_have_tl1_access(struct otx2_eth_dev *dev) { bool is_lbk = otx2_dev_is_lbk(dev); - return otx2_dev_is_pf(dev) && !otx2_dev_is_Ax(dev) && - !is_lbk && !dev->maxvf; + return otx2_dev_is_pf(dev) && !otx2_dev_is_Ax(dev) && !is_lbk; +} + +static bool +nix_tm_is_leaf(struct otx2_eth_dev *dev, int lvl) +{ + if (nix_tm_have_tl1_access(dev)) + return (lvl == OTX2_TM_LVL_QUEUE); + + return (lvl == OTX2_TM_LVL_SCH4); } static int @@ -424,6 +432,48 @@ prepare_tm_shaper_reg(struct otx2_nix_tm_node *tm_node, return k; } +static uint8_t +prepare_tm_sw_xoff(struct otx2_nix_tm_node *tm_node, bool enable, + volatile uint64_t *reg, volatile uint64_t *regval) +{ + uint32_t hw_lvl = tm_node->hw_lvl; + uint32_t schq = tm_node->hw_id; + uint8_t k = 0; + + otx2_tm_dbg("sw xoff config node %s(%u) lvl %u id %u, enable %u (%p)", + nix_hwlvl2str(hw_lvl), schq, tm_node->lvl, + tm_node->id, enable, tm_node); + + regval[k] = enable; + + switch (hw_lvl) { + case NIX_TXSCH_LVL_MDQ: + reg[k] = NIX_AF_MDQX_SW_XOFF(schq); + k++; + break; + case NIX_TXSCH_LVL_TL4: + reg[k] = NIX_AF_TL4X_SW_XOFF(schq); + k++; + break; + case NIX_TXSCH_LVL_TL3: + reg[k] = NIX_AF_TL3X_SW_XOFF(schq); + k++; + break; + case NIX_TXSCH_LVL_TL2: + reg[k] = NIX_AF_TL2X_SW_XOFF(schq); + k++; + break; + case NIX_TXSCH_LVL_TL1: + reg[k] = NIX_AF_TL1X_SW_XOFF(schq); + k++; + break; + default: + break; + } + + return k; +} + static int populate_tm_reg(struct otx2_eth_dev *dev, struct otx2_nix_tm_node *tm_node) @@ -692,12 +742,13 @@ nix_tm_node_add_to_list(struct otx2_eth_dev *dev, uint32_t node_id, uint16_t lvl, bool user, struct rte_tm_node_params *params) { - struct otx2_nix_tm_shaper_profile *shaper_profile; + struct otx2_nix_tm_shaper_profile *profile; struct otx2_nix_tm_node *tm_node, *parent_node; - uint32_t shaper_profile_id; + struct shaper_params cir, pir; + uint32_t profile_id; - shaper_profile_id = params->shaper_profile_id; - shaper_profile = nix_tm_shaper_profile_search(dev, shaper_profile_id); + profile_id = params->shaper_profile_id; + profile = nix_tm_shaper_profile_search(dev, profile_id); parent_node = nix_tm_node_search(dev, parent_node_id, user); @@ -709,6 +760,10 @@ nix_tm_node_add_to_list(struct otx2_eth_dev *dev, uint32_t node_id, tm_node->lvl = lvl; tm_node->hw_lvl = hw_lvl; + /* Maintain minimum weight */ + if (!weight) + weight = 1; + tm_node->id = node_id; tm_node->priority = priority; tm_node->weight = weight; @@ -720,10 +775,22 @@ nix_tm_node_add_to_list(struct otx2_eth_dev *dev, uint32_t node_id, tm_node->flags = NIX_TM_NODE_USER; rte_memcpy(&tm_node->params, params, sizeof(struct rte_tm_node_params)); - if (shaper_profile) - shaper_profile->reference_count++; + if (profile) + profile->reference_count++; + + memset(&cir, 0, sizeof(cir)); + memset(&pir, 0, sizeof(pir)); + shaper_config_to_nix(profile, &cir, &pir); + tm_node->parent = parent_node; tm_node->parent_hw_id = UINT32_MAX; + /* C0 doesn't support STALL when both PIR & CIR are enabled */ + if (lvl < OTX2_TM_LVL_QUEUE && + otx2_dev_is_96xx_Cx(dev) && + pir.rate && cir.rate) + tm_node->red_algo = NIX_REDALG_DISCARD; + else + tm_node->red_algo = NIX_REDALG_STD; TAILQ_INSERT_TAIL(&dev->node_list, tm_node, node); @@ -747,24 +814,67 @@ nix_tm_clear_shaper_profiles(struct otx2_eth_dev *dev) } static int -nix_smq_xoff(struct otx2_eth_dev *dev, uint16_t smq, bool enable) +nix_clear_path_xoff(struct otx2_eth_dev *dev, + struct otx2_nix_tm_node *tm_node) +{ + struct nix_txschq_config *req; + struct otx2_nix_tm_node *p; + int rc; + + /* Manipulating SW_XOFF not supported on Ax */ + if (otx2_dev_is_Ax(dev)) + return 0; + + /* Enable nodes in path for flush to succeed */ + if (!nix_tm_is_leaf(dev, tm_node->lvl)) + p = tm_node; + else + p = tm_node->parent; + while (p) { + if (!(p->flags & NIX_TM_NODE_ENABLED) && + (p->flags & NIX_TM_NODE_HWRES)) { + req = otx2_mbox_alloc_msg_nix_txschq_cfg(dev->mbox); + req->lvl = p->hw_lvl; + req->num_regs = prepare_tm_sw_xoff(p, false, req->reg, + req->regval); + rc = otx2_mbox_process(dev->mbox); + if (rc) + return rc; + + p->flags |= NIX_TM_NODE_ENABLED; + } + p = p->parent; + } + + return 0; +} + +static int +nix_smq_xoff(struct otx2_eth_dev *dev, + struct otx2_nix_tm_node *tm_node, + bool enable) { struct otx2_mbox *mbox = dev->mbox; struct nix_txschq_config *req; + uint16_t smq; + int rc; + + smq = tm_node->hw_id; + otx2_tm_dbg("Setting SMQ %u XOFF/FLUSH to %s", smq, + enable ? "enable" : "disable"); + + rc = nix_clear_path_xoff(dev, tm_node); + if (rc) + return rc; req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); req->lvl = NIX_TXSCH_LVL_SMQ; req->num_regs = 1; req->reg[0] = NIX_AF_SMQX_CFG(smq); - /* Unmodified fields */ - req->regval[0] = ((uint64_t)NIX_MAX_VTAG_INS << 36) | - (NIX_MAX_HW_FRS << 8) | NIX_MIN_HW_FRS; - - if (enable) - req->regval[0] |= BIT_ULL(50) | BIT_ULL(49); - else - req->regval[0] |= 0; + req->regval[0] = enable ? (BIT_ULL(50) | BIT_ULL(49)) : 0; + req->regval_mask[0] = enable ? + ~(BIT_ULL(50) | BIT_ULL(49)) : ~BIT_ULL(50); return otx2_mbox_process(mbox); } @@ -780,6 +890,9 @@ otx2_nix_sq_sqb_aura_fc(void *__txq, bool enable) uint64_t aura_handle; int rc; + otx2_tm_dbg("Setting SQ %u SQB aura FC to %s", txq->sq, + enable ? "enable" : "disable"); + lf = otx2_npa_lf_obj_get(); if (!lf) return -EFAULT; @@ -824,22 +937,41 @@ otx2_nix_sq_sqb_aura_fc(void *__txq, bool enable) return 0; } -static void +static int nix_txq_flush_sq_spin(struct otx2_eth_txq *txq) { uint16_t sqb_cnt, head_off, tail_off; struct otx2_eth_dev *dev = txq->dev; + uint64_t wdata, val, prev; uint16_t sq = txq->sq; - uint64_t reg, val; int64_t *regaddr; + uint64_t timeout;/* 10's of usec */ + + /* Wait for enough time based on shaper min rate */ + timeout = (txq->qconf.nb_desc * NIX_MAX_HW_FRS * 8 * 1E5); + timeout = timeout / dev->tm_rate_min; + if (!timeout) + timeout = 10000; + + wdata = ((uint64_t)sq << 32); + regaddr = (int64_t *)(dev->base + NIX_LF_SQ_OP_STATUS); + val = otx2_atomic64_add_nosync(wdata, regaddr); + + /* Spin multiple iterations as "txq->fc_cache_pkts" can still + * have space to send pkts even though fc_mem is disabled + */ while (true) { - reg = ((uint64_t)sq << 32); - regaddr = (int64_t *)(dev->base + NIX_LF_SQ_OP_PKTS); - val = otx2_atomic64_add_nosync(reg, regaddr); + prev = val; + rte_delay_us(10); + val = otx2_atomic64_add_nosync(wdata, regaddr); + /* Continue on error */ + if (val & BIT_ULL(63)) + continue; + + if (prev != val) + continue; - regaddr = (int64_t *)(dev->base + NIX_LF_SQ_OP_STATUS); - val = otx2_atomic64_add_nosync(reg, regaddr); sqb_cnt = val & 0xFFFF; head_off = (val >> 20) & 0x3F; tail_off = (val >> 28) & 0x3F; @@ -850,114 +982,219 @@ nix_txq_flush_sq_spin(struct otx2_eth_txq *txq) break; } - rte_pause(); + /* Timeout */ + if (!timeout) + goto exit; + timeout--; } + + return 0; +exit: + return -EFAULT; } -int -otx2_nix_tm_sw_xoff(void *__txq, bool dev_started) +/* Flush and disable tx queue and its parent SMQ */ +int otx2_nix_sq_flush_pre(void *_txq, bool dev_started) { - struct otx2_eth_txq *txq = __txq; - struct otx2_eth_dev *dev = txq->dev; - struct otx2_mbox *mbox = dev->mbox; - struct nix_aq_enq_req *req; - struct nix_aq_enq_rsp *rsp; - uint16_t smq; + struct otx2_nix_tm_node *tm_node, *sibling; + struct otx2_eth_txq *txq; + struct otx2_eth_dev *dev; + uint16_t sq; + bool user; int rc; - /* Get smq from sq */ - req = otx2_mbox_alloc_msg_nix_aq_enq(mbox); - req->qidx = txq->sq; - req->ctype = NIX_AQ_CTYPE_SQ; - req->op = NIX_AQ_INSTOP_READ; - rc = otx2_mbox_process_msg(mbox, (void *)&rsp); - if (rc) { - otx2_err("Failed to get smq, rc=%d", rc); - return -EIO; - } + txq = _txq; + dev = txq->dev; + sq = txq->sq; - /* Check if sq is enabled */ - if (!rsp->sq.ena) - return 0; + user = !!(dev->tm_flags & NIX_TM_COMMITTED); - smq = rsp->sq.smq; + /* Find the node for this SQ */ + tm_node = nix_tm_node_search(dev, sq, user); + if (!tm_node || !(tm_node->flags & NIX_TM_NODE_ENABLED)) { + otx2_err("Invalid node/state for sq %u", sq); + return -EFAULT; + } /* Enable CGX RXTX to drain pkts */ if (!dev_started) { - rc = otx2_cgx_rxtx_start(dev); - if (rc) + /* Though it enables both RX MCAM Entries and CGX Link + * we assume all the rx queues are stopped way back. + */ + otx2_mbox_alloc_msg_nix_lf_start_rx(dev->mbox); + rc = otx2_mbox_process(dev->mbox); + if (rc) { + otx2_err("cgx start failed, rc=%d", rc); return rc; - } - - rc = otx2_nix_sq_sqb_aura_fc(txq, false); - if (rc < 0) { - otx2_err("Failed to disable sqb aura fc, rc=%d", rc); - goto cleanup; + } } /* Disable smq xoff for case it was enabled earlier */ - rc = nix_smq_xoff(dev, smq, false); + rc = nix_smq_xoff(dev, tm_node->parent, false); if (rc) { - otx2_err("Failed to enable smq for sq %u, rc=%d", txq->sq, rc); - goto cleanup; + otx2_err("Failed to enable smq %u, rc=%d", + tm_node->parent->hw_id, rc); + return rc; } - /* Wait for sq entries to be flushed */ - nix_txq_flush_sq_spin(txq); + /* As per HRM, to disable an SQ, all other SQ's + * that feed to same SMQ must be paused before SMQ flush. + */ + TAILQ_FOREACH(sibling, &dev->node_list, node) { + if (sibling->parent != tm_node->parent) + continue; + if (!(sibling->flags & NIX_TM_NODE_ENABLED)) + continue; - /* Flush and enable smq xoff */ - rc = nix_smq_xoff(dev, smq, true); - if (rc) { - otx2_err("Failed to disable smq for sq %u, rc=%d", txq->sq, rc); - return rc; + sq = sibling->id; + txq = dev->eth_dev->data->tx_queues[sq]; + if (!txq) + continue; + + rc = otx2_nix_sq_sqb_aura_fc(txq, false); + if (rc) { + otx2_err("Failed to disable sqb aura fc, rc=%d", rc); + goto cleanup; + } + + /* Wait for sq entries to be flushed */ + rc = nix_txq_flush_sq_spin(txq); + if (rc) { + otx2_err("Failed to drain sq %u, rc=%d\n", txq->sq, rc); + return rc; + } } + tm_node->flags &= ~NIX_TM_NODE_ENABLED; + + /* Disable and flush */ + rc = nix_smq_xoff(dev, tm_node->parent, true); + if (rc) { + otx2_err("Failed to disable smq %u, rc=%d", + tm_node->parent->hw_id, rc); + goto cleanup; + } cleanup: /* Restore cgx state */ - if (!dev_started) - rc |= otx2_cgx_rxtx_stop(dev); + if (!dev_started) { + otx2_mbox_alloc_msg_nix_lf_stop_rx(dev->mbox); + rc |= otx2_mbox_process(dev->mbox); + } return rc; } +int otx2_nix_sq_flush_post(void *_txq) +{ + struct otx2_nix_tm_node *tm_node, *sibling; + struct otx2_eth_txq *txq = _txq; + struct otx2_eth_txq *s_txq; + struct otx2_eth_dev *dev; + bool once = false; + uint16_t sq, s_sq; + bool user; + int rc; + + dev = txq->dev; + sq = txq->sq; + user = !!(dev->tm_flags & NIX_TM_COMMITTED); + + /* Find the node for this SQ */ + tm_node = nix_tm_node_search(dev, sq, user); + if (!tm_node) { + otx2_err("Invalid node for sq %u", sq); + return -EFAULT; + } + + /* Enable all the siblings back */ + TAILQ_FOREACH(sibling, &dev->node_list, node) { + if (sibling->parent != tm_node->parent) + continue; + + if (sibling->id == sq) + continue; + + if (!(sibling->flags & NIX_TM_NODE_ENABLED)) + continue; + + s_sq = sibling->id; + s_txq = dev->eth_dev->data->tx_queues[s_sq]; + if (!s_txq) + continue; + + if (!once) { + /* Enable back if any SQ is still present */ + rc = nix_smq_xoff(dev, tm_node->parent, false); + if (rc) { + otx2_err("Failed to enable smq %u, rc=%d", + tm_node->parent->hw_id, rc); + return rc; + } + once = true; + } + + rc = otx2_nix_sq_sqb_aura_fc(s_txq, true); + if (rc) { + otx2_err("Failed to enable sqb aura fc, rc=%d", rc); + return rc; + } + } + + return 0; +} + static int -nix_tm_sw_xon(struct otx2_eth_txq *txq, - uint16_t smq, uint32_t rr_quantum) +nix_sq_sched_data(struct otx2_eth_dev *dev, + struct otx2_nix_tm_node *tm_node, + bool rr_quantum_only) { - struct otx2_eth_dev *dev = txq->dev; + struct rte_eth_dev *eth_dev = dev->eth_dev; struct otx2_mbox *mbox = dev->mbox; + uint16_t sq = tm_node->id, smq; struct nix_aq_enq_req *req; + uint64_t rr_quantum; int rc; - otx2_tm_dbg("Enabling sq(%u)->smq(%u), rr_quantum %u", - txq->sq, txq->sq, rr_quantum); - /* Set smq from sq */ + smq = tm_node->parent->hw_id; + rr_quantum = NIX_TM_WEIGHT_TO_RR_QUANTUM(tm_node->weight); + + if (rr_quantum_only) + otx2_tm_dbg("Update sq(%u) rr_quantum 0x%"PRIx64, sq, rr_quantum); + else + otx2_tm_dbg("Enabling sq(%u)->smq(%u), rr_quantum 0x%"PRIx64, + sq, smq, rr_quantum); + + if (sq > eth_dev->data->nb_tx_queues) + return -EFAULT; + req = otx2_mbox_alloc_msg_nix_aq_enq(mbox); - req->qidx = txq->sq; + req->qidx = sq; req->ctype = NIX_AQ_CTYPE_SQ; req->op = NIX_AQ_INSTOP_WRITE; - req->sq.smq = smq; + + /* smq update only when needed */ + if (!rr_quantum_only) { + req->sq.smq = smq; + req->sq_mask.smq = ~req->sq_mask.smq; + } req->sq.smq_rr_quantum = rr_quantum; - req->sq_mask.smq = ~req->sq_mask.smq; req->sq_mask.smq_rr_quantum = ~req->sq_mask.smq_rr_quantum; rc = otx2_mbox_process(mbox); - if (rc) { + if (rc) otx2_err("Failed to set smq, rc=%d", rc); - return -EIO; - } + return rc; +} + +int otx2_nix_sq_enable(void *_txq) +{ + struct otx2_eth_txq *txq = _txq; + int rc; /* Enable sqb_aura fc */ rc = otx2_nix_sq_sqb_aura_fc(txq, true); - if (rc < 0) { - otx2_err("Failed to enable sqb aura fc, rc=%d", rc); - return rc; - } - - /* Disable smq xoff */ - rc = nix_smq_xoff(dev, smq, false); if (rc) { - otx2_err("Failed to enable smq for sq %u", txq->sq); + otx2_err("Failed to enable sqb aura fc, rc=%d", rc); return rc; } @@ -968,12 +1205,11 @@ static int nix_tm_free_resources(struct otx2_eth_dev *dev, uint32_t flags_mask, uint32_t flags, bool hw_only) { - struct otx2_nix_tm_shaper_profile *shaper_profile; + struct otx2_nix_tm_shaper_profile *profile; struct otx2_nix_tm_node *tm_node, *next_node; struct otx2_mbox *mbox = dev->mbox; struct nix_txsch_free_req *req; - uint32_t shaper_profile_id; - bool skip_node = false; + uint32_t profile_id; int rc = 0; next_node = TAILQ_FIRST(&dev->node_list); @@ -985,37 +1221,40 @@ nix_tm_free_resources(struct otx2_eth_dev *dev, uint32_t flags_mask, if ((tm_node->flags & flags_mask) != flags) continue; - if (nix_tm_have_tl1_access(dev) && - tm_node->hw_lvl == NIX_TXSCH_LVL_TL1) - skip_node = true; - - otx2_tm_dbg("Free hwres for node %u, hwlvl %u, hw_id %u (%p)", - tm_node->id, tm_node->hw_lvl, - tm_node->hw_id, tm_node); - /* Free specific HW resource if requested */ - if (!skip_node && flags_mask && + if (!nix_tm_is_leaf(dev, tm_node->lvl) && + tm_node->hw_lvl != NIX_TXSCH_LVL_TL1 && tm_node->flags & NIX_TM_NODE_HWRES) { + /* Free specific HW resource */ + otx2_tm_dbg("Free hwres %s(%u) lvl %u id %u (%p)", + nix_hwlvl2str(tm_node->hw_lvl), + tm_node->hw_id, tm_node->lvl, + tm_node->id, tm_node); + + rc = nix_clear_path_xoff(dev, tm_node); + if (rc) + return rc; + req = otx2_mbox_alloc_msg_nix_txsch_free(mbox); req->flags = 0; req->schq_lvl = tm_node->hw_lvl; req->schq = tm_node->hw_id; rc = otx2_mbox_process(mbox); if (rc) - break; - } else { - skip_node = false; + return rc; + tm_node->flags &= ~NIX_TM_NODE_HWRES; } - tm_node->flags &= ~NIX_TM_NODE_HWRES; /* Leave software elements if needed */ if (hw_only) continue; - shaper_profile_id = tm_node->params.shaper_profile_id; - shaper_profile = - nix_tm_shaper_profile_search(dev, shaper_profile_id); - if (shaper_profile) - shaper_profile->reference_count--; + otx2_tm_dbg("Free node lvl %u id %u (%p)", + tm_node->lvl, tm_node->id, tm_node); + + profile_id = tm_node->params.shaper_profile_id; + profile = nix_tm_shaper_profile_search(dev, profile_id); + if (profile) + profile->reference_count--; TAILQ_REMOVE(&dev->node_list, tm_node, node); rte_free(tm_node); @@ -1060,8 +1299,8 @@ nix_tm_assign_id_to_node(struct otx2_eth_dev *dev, uint32_t hw_id, schq_con_index, prio_offset; uint32_t l_id, schq_index; - otx2_tm_dbg("Assign hw id for child node %u, lvl %u, hw_lvl %u (%p)", - child->id, child->lvl, child->hw_lvl, child); + otx2_tm_dbg("Assign hw id for child node %s lvl %u id %u (%p)", + nix_hwlvl2str(child->hw_lvl), child->lvl, child->id, child); child->flags |= NIX_TM_NODE_HWRES; @@ -1219,8 +1458,8 @@ nix_tm_alloc_resources(struct rte_eth_dev *eth_dev, bool xmit_enable) { struct otx2_eth_dev *dev = otx2_eth_pmd_priv(eth_dev); struct otx2_nix_tm_node *tm_node; - uint16_t sq, smq, rr_quantum; struct otx2_eth_txq *txq; + uint16_t sq; int rc; nix_tm_update_parent_info(dev); @@ -1237,42 +1476,68 @@ nix_tm_alloc_resources(struct rte_eth_dev *eth_dev, bool xmit_enable) return rc; } - /* Enable xmit as all the topology is ready */ - TAILQ_FOREACH(tm_node, &dev->node_list, node) { - if (tm_node->flags & NIX_TM_NODE_ENABLED) - continue; + /* Trigger MTU recalculate as SMQ needs MTU conf */ + if (eth_dev->data->dev_started && eth_dev->data->nb_rx_queues) { + rc = otx2_nix_recalc_mtu(eth_dev); + if (rc) { + otx2_err("TM MTU update failed, rc=%d", rc); + return rc; + } + } - /* Enable xmit on sq */ - if (tm_node->lvl != OTX2_TM_LVL_QUEUE) { + /* Mark all non-leaf's as enabled */ + TAILQ_FOREACH(tm_node, &dev->node_list, node) { + if (!nix_tm_is_leaf(dev, tm_node->lvl)) tm_node->flags |= NIX_TM_NODE_ENABLED; + } + + if (!xmit_enable) + return 0; + + /* Update SQ Sched Data while SQ is idle */ + TAILQ_FOREACH(tm_node, &dev->node_list, node) { + if (!nix_tm_is_leaf(dev, tm_node->lvl)) continue; + + rc = nix_sq_sched_data(dev, tm_node, false); + if (rc) { + otx2_err("SQ %u sched update failed, rc=%d", + tm_node->id, rc); + return rc; } + } - /* Don't enable SMQ or mark as enable */ - if (!xmit_enable) + /* Finally XON all SMQ's */ + TAILQ_FOREACH(tm_node, &dev->node_list, node) { + if (tm_node->hw_lvl != NIX_TXSCH_LVL_SMQ) continue; - sq = tm_node->id; - if (sq > eth_dev->data->nb_tx_queues) { - rc = -EFAULT; - break; + rc = nix_smq_xoff(dev, tm_node, false); + if (rc) { + otx2_err("Failed to enable smq %u, rc=%d", + tm_node->hw_id, rc); + return rc; } + } - txq = eth_dev->data->tx_queues[sq]; + /* Enable xmit as all the topology is ready */ + TAILQ_FOREACH(tm_node, &dev->node_list, node) { + if (!nix_tm_is_leaf(dev, tm_node->lvl)) + continue; - smq = tm_node->parent->hw_id; - rr_quantum = NIX_TM_WEIGHT_TO_RR_QUANTUM(tm_node->weight); + sq = tm_node->id; + txq = eth_dev->data->tx_queues[sq]; - rc = nix_tm_sw_xon(txq, smq, rr_quantum); - if (rc) - break; + rc = otx2_nix_sq_enable(txq); + if (rc) { + otx2_err("TM sw xon failed on SQ %u, rc=%d", + tm_node->id, rc); + return rc; + } tm_node->flags |= NIX_TM_NODE_ENABLED; } - if (rc) - otx2_err("TM failed to enable xmit on sq %u, rc=%d", sq, rc); - - return rc; + return 0; } static int @@ -1282,7 +1547,7 @@ nix_tm_prepare_default_tree(struct rte_eth_dev *eth_dev) uint32_t def = eth_dev->data->nb_tx_queues; struct rte_tm_node_params params; uint32_t leaf_parent, i; - int rc = 0; + int rc = 0, leaf_level; /* Default params */ memset(¶ms, 0, sizeof(params)); @@ -1325,6 +1590,7 @@ nix_tm_prepare_default_tree(struct rte_eth_dev *eth_dev) goto exit; leaf_parent = def + 4; + leaf_level = OTX2_TM_LVL_QUEUE; } else { dev->otx2_tm_root_lvl = NIX_TXSCH_LVL_TL2; rc = nix_tm_node_add_to_list(dev, def, RTE_TM_NODE_ID_NULL, 0, @@ -1356,6 +1622,7 @@ nix_tm_prepare_default_tree(struct rte_eth_dev *eth_dev) goto exit; leaf_parent = def + 3; + leaf_level = OTX2_TM_LVL_SCH4; } /* Add leaf nodes */ @@ -1363,7 +1630,7 @@ nix_tm_prepare_default_tree(struct rte_eth_dev *eth_dev) rc = nix_tm_node_add_to_list(dev, i, leaf_parent, 0, DEFAULT_RR_WEIGHT, NIX_TXSCH_LVL_CNT, - OTX2_TM_LVL_QUEUE, false, ¶ms); + leaf_level, false, ¶ms); if (rc) break; } @@ -1378,6 +1645,7 @@ void otx2_nix_tm_conf_init(struct rte_eth_dev *eth_dev) TAILQ_INIT(&dev->node_list); TAILQ_INIT(&dev->shaper_profile_list); + dev->tm_rate_min = 1E9; /* 1Gbps */ } int otx2_nix_tm_init_default(struct rte_eth_dev *eth_dev) @@ -1455,7 +1723,7 @@ otx2_nix_tm_get_leaf_data(struct otx2_eth_dev *dev, uint16_t sq, tm_node = nix_tm_node_search(dev, sq, true); /* Check if we found a valid leaf node */ - if (!tm_node || tm_node->lvl != OTX2_TM_LVL_QUEUE || + if (!tm_node || !nix_tm_is_leaf(dev, tm_node->lvl) || !tm_node->parent || tm_node->parent->hw_id == UINT32_MAX) { return -EIO; } @@ -1464,7 +1732,7 @@ otx2_nix_tm_get_leaf_data(struct otx2_eth_dev *dev, uint16_t sq, *smq = tm_node->parent->hw_id; *rr_quantum = NIX_TM_WEIGHT_TO_RR_QUANTUM(tm_node->weight); - rc = nix_smq_xoff(dev, *smq, false); + rc = nix_smq_xoff(dev, tm_node->parent, false); if (rc) return rc; tm_node->flags |= NIX_TM_NODE_ENABLED; diff --git a/drivers/net/octeontx2/otx2_tm.h b/drivers/net/octeontx2/otx2_tm.h index ad7727ee94..413120ae81 100644 --- a/drivers/net/octeontx2/otx2_tm.h +++ b/drivers/net/octeontx2/otx2_tm.h @@ -10,6 +10,7 @@ #include #define NIX_TM_DEFAULT_TREE BIT_ULL(0) +#define NIX_TM_COMMITTED BIT_ULL(1) #define NIX_TM_TL1_NO_SP BIT_ULL(3) struct otx2_eth_dev; @@ -19,7 +20,9 @@ int otx2_nix_tm_init_default(struct rte_eth_dev *eth_dev); int otx2_nix_tm_fini(struct rte_eth_dev *eth_dev); int otx2_nix_tm_get_leaf_data(struct otx2_eth_dev *dev, uint16_t sq, uint32_t *rr_quantum, uint16_t *smq); -int otx2_nix_tm_sw_xoff(void *_txq, bool dev_started); +int otx2_nix_sq_flush_pre(void *_txq, bool dev_started); +int otx2_nix_sq_flush_post(void *_txq); +int otx2_nix_sq_enable(void *_txq); int otx2_nix_sq_sqb_aura_fc(void *_txq, bool enable); struct otx2_nix_tm_node { @@ -40,6 +43,7 @@ struct otx2_nix_tm_node { #define NIX_TM_NODE_USER BIT_ULL(2) /* Shaper algorithm for RED state @NIX_REDALG_E */ uint32_t red_algo:2; + struct otx2_nix_tm_node *parent; struct rte_tm_node_params params; }; @@ -70,7 +74,6 @@ TAILQ_HEAD(otx2_nix_tm_shaper_profile_list, otx2_nix_tm_shaper_profile); ((((__weight) & MAX_SCHED_WEIGHT) * \ NIX_TM_RR_QUANTUM_MAX) / MAX_SCHED_WEIGHT) - /* DEFAULT_RR_WEIGHT * NIX_TM_RR_QUANTUM_MAX / MAX_SCHED_WEIGHT */ /* = NIX_MAX_HW_MTU */ #define DEFAULT_RR_WEIGHT 71 -- 2.20.1