From: Nithin Dabilpuram Date: Sun, 26 May 2019 16:49:40 +0000 (+0530) Subject: net/octeontx2: configure TM HW resources X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=2f6d5220fbea6db4ff635e923e16c0cf40223d5f;p=dpdk.git net/octeontx2: configure TM HW resources This patch sets up and configure hierarchy in hw nodes. Since all the registers are with RVU AF, register configuration is also done using mbox communication. Signed-off-by: Nithin Dabilpuram Signed-off-by: Krzysztof Kanas --- diff --git a/drivers/net/octeontx2/otx2_tm.c b/drivers/net/octeontx2/otx2_tm.c index 91f31df059..c6154e4d44 100644 --- a/drivers/net/octeontx2/otx2_tm.c +++ b/drivers/net/octeontx2/otx2_tm.c @@ -20,6 +20,41 @@ enum otx2_tm_node_level { OTX2_TM_LVL_MAX, }; +static inline +uint64_t shaper2regval(struct shaper_params *shaper) +{ + return (shaper->burst_exponent << 37) | (shaper->burst_mantissa << 29) | + (shaper->div_exp << 13) | (shaper->exponent << 9) | + (shaper->mantissa << 1); +} + +static int +nix_get_link(struct otx2_eth_dev *dev) +{ + int link = 13 /* SDP */; + uint16_t lmac_chan; + uint16_t map; + + lmac_chan = dev->tx_chan_base; + + /* CGX lmac link */ + if (lmac_chan >= 0x800) { + map = lmac_chan & 0x7FF; + link = 4 * ((map >> 8) & 0xF) + ((map >> 4) & 0xF); + } else if (lmac_chan < 0x700) { + /* LBK channel */ + link = 12; + } + + return link; +} + +static uint8_t +nix_get_relchan(struct otx2_eth_dev *dev) +{ + return dev->tx_chan_base & 0xff; +} + static bool nix_tm_have_tl1_access(struct otx2_eth_dev *dev) { @@ -28,6 +63,24 @@ nix_tm_have_tl1_access(struct otx2_eth_dev *dev) !is_lbk && !dev->maxvf; } +static int +find_prio_anchor(struct otx2_eth_dev *dev, uint32_t node_id) +{ + struct otx2_nix_tm_node *child_node; + + TAILQ_FOREACH(child_node, &dev->node_list, node) { + if (!child_node->parent) + continue; + if (!(child_node->parent->id == node_id)) + continue; + if (child_node->priority == child_node->parent->rr_prio) + continue; + return child_node->hw_id - child_node->priority; + } + return 0; +} + + static struct otx2_nix_tm_shaper_profile * nix_tm_shaper_profile_search(struct otx2_eth_dev *dev, uint32_t shaper_id) { @@ -40,6 +93,451 @@ nix_tm_shaper_profile_search(struct otx2_eth_dev *dev, uint32_t shaper_id) return NULL; } +static inline uint64_t +shaper_rate_to_nix(uint64_t cclk_hz, uint64_t cclk_ticks, + uint64_t value, uint64_t *exponent_p, + uint64_t *mantissa_p, uint64_t *div_exp_p) +{ + uint64_t div_exp, exponent, mantissa; + + /* Boundary checks */ + if (value < MIN_SHAPER_RATE(cclk_hz, cclk_ticks) || + value > MAX_SHAPER_RATE(cclk_hz, cclk_ticks)) + return 0; + + if (value <= SHAPER_RATE(cclk_hz, cclk_ticks, 0, 0, 0)) { + /* Calculate rate div_exp and mantissa using + * the following formula: + * + * value = (cclk_hz * (256 + mantissa) + * / ((cclk_ticks << div_exp) * 256) + */ + div_exp = 0; + exponent = 0; + mantissa = MAX_RATE_MANTISSA; + + while (value < (cclk_hz / (cclk_ticks << div_exp))) + div_exp += 1; + + while (value < + ((cclk_hz * (256 + mantissa)) / + ((cclk_ticks << div_exp) * 256))) + mantissa -= 1; + } else { + /* Calculate rate exponent and mantissa using + * the following formula: + * + * value = (cclk_hz * ((256 + mantissa) << exponent) + * / (cclk_ticks * 256) + * + */ + div_exp = 0; + exponent = MAX_RATE_EXPONENT; + mantissa = MAX_RATE_MANTISSA; + + while (value < (cclk_hz * (1 << exponent)) / cclk_ticks) + exponent -= 1; + + while (value < (cclk_hz * ((256 + mantissa) << exponent)) / + (cclk_ticks * 256)) + mantissa -= 1; + } + + if (div_exp > MAX_RATE_DIV_EXP || + exponent > MAX_RATE_EXPONENT || mantissa > MAX_RATE_MANTISSA) + return 0; + + if (div_exp_p) + *div_exp_p = div_exp; + if (exponent_p) + *exponent_p = exponent; + if (mantissa_p) + *mantissa_p = mantissa; + + /* Calculate real rate value */ + return SHAPER_RATE(cclk_hz, cclk_ticks, exponent, mantissa, div_exp); +} + +static inline uint64_t +lx_shaper_rate_to_nix(uint64_t cclk_hz, uint32_t hw_lvl, + uint64_t value, uint64_t *exponent, + uint64_t *mantissa, uint64_t *div_exp) +{ + if (hw_lvl == NIX_TXSCH_LVL_TL1) + return shaper_rate_to_nix(cclk_hz, L1_TIME_WHEEL_CCLK_TICKS, + value, exponent, mantissa, div_exp); + else + return shaper_rate_to_nix(cclk_hz, LX_TIME_WHEEL_CCLK_TICKS, + value, exponent, mantissa, div_exp); +} + +static inline uint64_t +shaper_burst_to_nix(uint64_t value, uint64_t *exponent_p, + uint64_t *mantissa_p) +{ + uint64_t exponent, mantissa; + + if (value < MIN_SHAPER_BURST || value > MAX_SHAPER_BURST) + return 0; + + /* Calculate burst exponent and mantissa using + * the following formula: + * + * value = (((256 + mantissa) << (exponent + 1) + / 256) + * + */ + exponent = MAX_BURST_EXPONENT; + mantissa = MAX_BURST_MANTISSA; + + while (value < (1ull << (exponent + 1))) + exponent -= 1; + + while (value < ((256 + mantissa) << (exponent + 1)) / 256) + mantissa -= 1; + + if (exponent > MAX_BURST_EXPONENT || mantissa > MAX_BURST_MANTISSA) + return 0; + + if (exponent_p) + *exponent_p = exponent; + if (mantissa_p) + *mantissa_p = mantissa; + + return SHAPER_BURST(exponent, mantissa); +} + +static int +configure_shaper_cir_pir_reg(struct otx2_eth_dev *dev, + struct otx2_nix_tm_node *tm_node, + struct shaper_params *cir, + struct shaper_params *pir) +{ + uint32_t shaper_profile_id = RTE_TM_SHAPER_PROFILE_ID_NONE; + struct otx2_nix_tm_shaper_profile *shaper_profile = NULL; + struct rte_tm_shaper_params *param; + + shaper_profile_id = tm_node->params.shaper_profile_id; + + shaper_profile = nix_tm_shaper_profile_search(dev, shaper_profile_id); + if (shaper_profile) { + param = &shaper_profile->profile; + /* Calculate CIR exponent and mantissa */ + if (param->committed.rate) + cir->rate = lx_shaper_rate_to_nix(CCLK_HZ, + tm_node->hw_lvl_id, + param->committed.rate, + &cir->exponent, + &cir->mantissa, + &cir->div_exp); + + /* Calculate PIR exponent and mantissa */ + if (param->peak.rate) + pir->rate = lx_shaper_rate_to_nix(CCLK_HZ, + tm_node->hw_lvl_id, + param->peak.rate, + &pir->exponent, + &pir->mantissa, + &pir->div_exp); + + /* Calculate CIR burst exponent and mantissa */ + if (param->committed.size) + cir->burst = shaper_burst_to_nix(param->committed.size, + &cir->burst_exponent, + &cir->burst_mantissa); + + /* Calculate PIR burst exponent and mantissa */ + if (param->peak.size) + pir->burst = shaper_burst_to_nix(param->peak.size, + &pir->burst_exponent, + &pir->burst_mantissa); + } + + return 0; +} + +static int +send_tm_reqval(struct otx2_mbox *mbox, struct nix_txschq_config *req) +{ + int rc; + + if (req->num_regs > MAX_REGS_PER_MBOX_MSG) + return -ERANGE; + + rc = otx2_mbox_process(mbox); + if (rc) + return rc; + + req->num_regs = 0; + return 0; +} + +static int +populate_tm_registers(struct otx2_eth_dev *dev, + struct otx2_nix_tm_node *tm_node) +{ + uint64_t strict_schedul_prio, rr_prio; + struct otx2_mbox *mbox = dev->mbox; + volatile uint64_t *reg, *regval; + uint64_t parent = 0, child = 0; + struct shaper_params cir, pir; + struct nix_txschq_config *req; + uint64_t rr_quantum; + uint32_t hw_lvl; + uint32_t schq; + int rc; + + memset(&cir, 0, sizeof(cir)); + memset(&pir, 0, sizeof(pir)); + + /* Skip leaf nodes */ + if (tm_node->hw_lvl_id == NIX_TXSCH_LVL_CNT) + return 0; + + /* Root node will not have a parent node */ + if (tm_node->hw_lvl_id == dev->otx2_tm_root_lvl) + parent = tm_node->parent_hw_id; + else + parent = tm_node->parent->hw_id; + + /* Do we need this trigger to configure TL1 */ + if (dev->otx2_tm_root_lvl == NIX_TXSCH_LVL_TL2 && + tm_node->hw_lvl_id == dev->otx2_tm_root_lvl) { + schq = parent; + /* + * Default config for TL1. + * For VF this is always ignored. + */ + + req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); + req->lvl = NIX_TXSCH_LVL_TL1; + + /* Set DWRR quantum */ + req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq); + req->regval[0] = TXSCH_TL1_DFLT_RR_QTM; + req->num_regs++; + + req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq); + req->regval[1] = (TXSCH_TL1_DFLT_RR_PRIO << 1); + req->num_regs++; + + req->reg[2] = NIX_AF_TL1X_CIR(schq); + req->regval[2] = 0; + req->num_regs++; + + rc = send_tm_reqval(mbox, req); + if (rc) + goto error; + } + + if (tm_node->hw_lvl_id != NIX_TXSCH_LVL_SMQ) + child = find_prio_anchor(dev, tm_node->id); + + rr_prio = tm_node->rr_prio; + hw_lvl = tm_node->hw_lvl_id; + strict_schedul_prio = tm_node->priority; + schq = tm_node->hw_id; + rr_quantum = (tm_node->weight * NIX_TM_RR_QUANTUM_MAX) / + MAX_SCHED_WEIGHT; + + configure_shaper_cir_pir_reg(dev, tm_node, &cir, &pir); + + otx2_tm_dbg("Configure node %p, lvl %u hw_lvl %u, id %u, hw_id %u," + "parent_hw_id %" PRIx64 ", pir %" PRIx64 ", cir %" PRIx64, + tm_node, tm_node->level_id, hw_lvl, + tm_node->id, schq, parent, pir.rate, cir.rate); + + rc = -EFAULT; + + switch (hw_lvl) { + case NIX_TXSCH_LVL_SMQ: + req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); + req->lvl = hw_lvl; + reg = req->reg; + regval = req->regval; + req->num_regs = 0; + + /* Set xoff which will be cleared later */ + *reg++ = NIX_AF_SMQX_CFG(schq); + *regval++ = BIT_ULL(50) | ((uint64_t)NIX_MAX_VTAG_INS << 36) | + (NIX_MAX_HW_FRS << 8) | NIX_MIN_HW_FRS; + req->num_regs++; + *reg++ = NIX_AF_MDQX_PARENT(schq); + *regval++ = parent << 16; + req->num_regs++; + *reg++ = NIX_AF_MDQX_SCHEDULE(schq); + *regval++ = (strict_schedul_prio << 24) | rr_quantum; + req->num_regs++; + if (pir.rate && pir.burst) { + *reg++ = NIX_AF_MDQX_PIR(schq); + *regval++ = shaper2regval(&pir) | 1; + req->num_regs++; + } + + if (cir.rate && cir.burst) { + *reg++ = NIX_AF_MDQX_CIR(schq); + *regval++ = shaper2regval(&cir) | 1; + req->num_regs++; + } + + rc = send_tm_reqval(mbox, req); + if (rc) + goto error; + break; + case NIX_TXSCH_LVL_TL4: + req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); + req->lvl = hw_lvl; + req->num_regs = 0; + reg = req->reg; + regval = req->regval; + + *reg++ = NIX_AF_TL4X_PARENT(schq); + *regval++ = parent << 16; + req->num_regs++; + *reg++ = NIX_AF_TL4X_TOPOLOGY(schq); + *regval++ = (child << 32) | (rr_prio << 1); + req->num_regs++; + *reg++ = NIX_AF_TL4X_SCHEDULE(schq); + *regval++ = (strict_schedul_prio << 24) | rr_quantum; + req->num_regs++; + if (pir.rate && pir.burst) { + *reg++ = NIX_AF_TL4X_PIR(schq); + *regval++ = shaper2regval(&pir) | 1; + req->num_regs++; + } + if (cir.rate && cir.burst) { + *reg++ = NIX_AF_TL4X_CIR(schq); + *regval++ = shaper2regval(&cir) | 1; + req->num_regs++; + } + + rc = send_tm_reqval(mbox, req); + if (rc) + goto error; + break; + case NIX_TXSCH_LVL_TL3: + req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); + req->lvl = hw_lvl; + req->num_regs = 0; + reg = req->reg; + regval = req->regval; + + *reg++ = NIX_AF_TL3X_PARENT(schq); + *regval++ = parent << 16; + req->num_regs++; + *reg++ = NIX_AF_TL3X_TOPOLOGY(schq); + *regval++ = (child << 32) | (rr_prio << 1); + req->num_regs++; + *reg++ = NIX_AF_TL3X_SCHEDULE(schq); + *regval++ = (strict_schedul_prio << 24) | rr_quantum; + req->num_regs++; + if (pir.rate && pir.burst) { + *reg++ = NIX_AF_TL3X_PIR(schq); + *regval++ = shaper2regval(&pir) | 1; + req->num_regs++; + } + if (cir.rate && cir.burst) { + *reg++ = NIX_AF_TL3X_CIR(schq); + *regval++ = shaper2regval(&cir) | 1; + req->num_regs++; + } + + rc = send_tm_reqval(mbox, req); + if (rc) + goto error; + break; + case NIX_TXSCH_LVL_TL2: + req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); + req->lvl = hw_lvl; + req->num_regs = 0; + reg = req->reg; + regval = req->regval; + + *reg++ = NIX_AF_TL2X_PARENT(schq); + *regval++ = parent << 16; + req->num_regs++; + *reg++ = NIX_AF_TL2X_TOPOLOGY(schq); + *regval++ = (child << 32) | (rr_prio << 1); + req->num_regs++; + *reg++ = NIX_AF_TL2X_SCHEDULE(schq); + if (dev->otx2_tm_root_lvl == NIX_TXSCH_LVL_TL2) + *regval++ = (1 << 24) | rr_quantum; + else + *regval++ = (strict_schedul_prio << 24) | rr_quantum; + req->num_regs++; + *reg++ = NIX_AF_TL3_TL2X_LINKX_CFG(schq, nix_get_link(dev)); + *regval++ = BIT_ULL(12) | nix_get_relchan(dev); + req->num_regs++; + if (pir.rate && pir.burst) { + *reg++ = NIX_AF_TL2X_PIR(schq); + *regval++ = shaper2regval(&pir) | 1; + req->num_regs++; + } + if (cir.rate && cir.burst) { + *reg++ = NIX_AF_TL2X_CIR(schq); + *regval++ = shaper2regval(&cir) | 1; + req->num_regs++; + } + + rc = send_tm_reqval(mbox, req); + if (rc) + goto error; + break; + case NIX_TXSCH_LVL_TL1: + req = otx2_mbox_alloc_msg_nix_txschq_cfg(mbox); + req->lvl = hw_lvl; + req->num_regs = 0; + reg = req->reg; + regval = req->regval; + + *reg++ = NIX_AF_TL1X_SCHEDULE(schq); + *regval++ = rr_quantum; + req->num_regs++; + *reg++ = NIX_AF_TL1X_TOPOLOGY(schq); + *regval++ = (child << 32) | (rr_prio << 1 /*RR_PRIO*/); + req->num_regs++; + if (cir.rate && cir.burst) { + *reg++ = NIX_AF_TL1X_CIR(schq); + *regval++ = shaper2regval(&cir) | 1; + req->num_regs++; + } + + rc = send_tm_reqval(mbox, req); + if (rc) + goto error; + break; + } + + return 0; +error: + otx2_err("Txschq cfg request failed for node %p, rc=%d", tm_node, rc); + return rc; +} + + +static int +nix_tm_txsch_reg_config(struct otx2_eth_dev *dev) +{ + struct otx2_nix_tm_node *tm_node; + uint32_t lvl; + int rc = 0; + + if (nix_get_link(dev) == 13) + return -EPERM; + + for (lvl = 0; lvl < (uint32_t)dev->otx2_tm_root_lvl + 1; lvl++) { + TAILQ_FOREACH(tm_node, &dev->node_list, node) { + if (tm_node->hw_lvl_id == lvl) { + rc = populate_tm_registers(dev, tm_node); + if (rc) + goto exit; + } + } + } +exit: + return rc; +} + static struct otx2_nix_tm_node * nix_tm_node_search(struct otx2_eth_dev *dev, uint32_t node_id, bool user) @@ -443,6 +941,12 @@ nix_tm_alloc_resources(struct rte_eth_dev *eth_dev, bool xmit_enable) return rc; } + rc = nix_tm_txsch_reg_config(dev); + if (rc) { + otx2_err("TM failed to configure sched registers=%d", rc); + return rc; + } + return 0; } diff --git a/drivers/net/octeontx2/otx2_tm.h b/drivers/net/octeontx2/otx2_tm.h index 94023fa999..af1bb18621 100644 --- a/drivers/net/octeontx2/otx2_tm.h +++ b/drivers/net/octeontx2/otx2_tm.h @@ -64,4 +64,86 @@ TAILQ_HEAD(otx2_nix_tm_shaper_profile_list, otx2_nix_tm_shaper_profile); /* = NIX_MAX_HW_MTU */ #define DEFAULT_RR_WEIGHT 71 +/** NIX rate limits */ +#define MAX_RATE_DIV_EXP 12 +#define MAX_RATE_EXPONENT 0xf +#define MAX_RATE_MANTISSA 0xff + +/** NIX rate limiter time-wheel resolution */ +#define L1_TIME_WHEEL_CCLK_TICKS 240 +#define LX_TIME_WHEEL_CCLK_TICKS 860 + +#define CCLK_HZ 1000000000 + +/* NIX rate calculation + * CCLK = coprocessor-clock frequency in MHz + * CCLK_TICKS = rate limiter time-wheel resolution + * + * PIR_ADD = ((256 + NIX_*_PIR[RATE_MANTISSA]) + * << NIX_*_PIR[RATE_EXPONENT]) / 256 + * PIR = (CCLK / (CCLK_TICKS << NIX_*_PIR[RATE_DIVIDER_EXPONENT])) + * * PIR_ADD + * + * CIR_ADD = ((256 + NIX_*_CIR[RATE_MANTISSA]) + * << NIX_*_CIR[RATE_EXPONENT]) / 256 + * CIR = (CCLK / (CCLK_TICKS << NIX_*_CIR[RATE_DIVIDER_EXPONENT])) + * * CIR_ADD + */ +#define SHAPER_RATE(cclk_hz, cclk_ticks, \ + exponent, mantissa, div_exp) \ + (((uint64_t)(cclk_hz) * ((256 + (mantissa)) << (exponent))) \ + / (((cclk_ticks) << (div_exp)) * 256)) + +#define L1_SHAPER_RATE(cclk_hz, exponent, mantissa, div_exp) \ + SHAPER_RATE(cclk_hz, L1_TIME_WHEEL_CCLK_TICKS, \ + exponent, mantissa, div_exp) + +#define LX_SHAPER_RATE(cclk_hz, exponent, mantissa, div_exp) \ + SHAPER_RATE(cclk_hz, LX_TIME_WHEEL_CCLK_TICKS, \ + exponent, mantissa, div_exp) + +/* Shaper rate limits */ +#define MIN_SHAPER_RATE(cclk_hz, cclk_ticks) \ + SHAPER_RATE(cclk_hz, cclk_ticks, 0, 0, MAX_RATE_DIV_EXP) + +#define MAX_SHAPER_RATE(cclk_hz, cclk_ticks) \ + SHAPER_RATE(cclk_hz, cclk_ticks, MAX_RATE_EXPONENT, \ + MAX_RATE_MANTISSA, 0) + +#define MIN_L1_SHAPER_RATE(cclk_hz) \ + MIN_SHAPER_RATE(cclk_hz, L1_TIME_WHEEL_CCLK_TICKS) + +#define MAX_L1_SHAPER_RATE(cclk_hz) \ + MAX_SHAPER_RATE(cclk_hz, L1_TIME_WHEEL_CCLK_TICKS) + +/** TM Shaper - low level operations */ + +/** NIX burst limits */ +#define MAX_BURST_EXPONENT 0xf +#define MAX_BURST_MANTISSA 0xff + +/* NIX burst calculation + * PIR_BURST = ((256 + NIX_*_PIR[BURST_MANTISSA]) + * << (NIX_*_PIR[BURST_EXPONENT] + 1)) + * / 256 + * + * CIR_BURST = ((256 + NIX_*_CIR[BURST_MANTISSA]) + * << (NIX_*_CIR[BURST_EXPONENT] + 1)) + * / 256 + */ +#define SHAPER_BURST(exponent, mantissa) \ + (((256 + (mantissa)) << ((exponent) + 1)) / 256) + +/** Shaper burst limits */ +#define MIN_SHAPER_BURST \ + SHAPER_BURST(0, 0) + +#define MAX_SHAPER_BURST \ + SHAPER_BURST(MAX_BURST_EXPONENT,\ + MAX_BURST_MANTISSA) + +/* Default TL1 priority and Quantum from AF */ +#define TXSCH_TL1_DFLT_RR_QTM ((1 << 24) - 1) +#define TXSCH_TL1_DFLT_RR_PRIO 1 + #endif /* __OTX2_TM_H__ */