/***********************license start*********************************** * Copyright (c) 2003-2017 Cavium Inc. (support@cavium.com). All rights * reserved. * * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * * Neither the name of Cavium Inc. nor the names of * its contributors may be used to endorse or promote products * derived from this software without specific prior written * permission. * * This Software, including technical data, may be subject to U.S. export * control laws, including the U.S. Export Administration Act and its * associated regulations, and may be subject to export or import * regulations in other countries. * * TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE SOFTWARE IS PROVIDED "AS IS" * AND WITH ALL FAULTS AND CAVIUM INC. MAKES NO PROMISES, REPRESENTATIONS OR * WARRANTIES, EITHER EXPRESS, IMPLIED, STATUTORY, OR OTHERWISE, WITH RESPECT * TO THE SOFTWARE, INCLUDING ITS CONDITION, ITS CONFORMITY TO ANY * REPRESENTATION OR DESCRIPTION, OR THE EXISTENCE OF ANY LATENT OR PATENT * DEFECTS, AND CAVIUM SPECIFICALLY DISCLAIMS ALL IMPLIED (IF ANY) WARRANTIES * OF TITLE, MERCHANTABILITY, NONINFRINGEMENT, FITNESS FOR A PARTICULAR * PURPOSE, LACK OF VIRUSES, ACCURACY OR COMPLETENESS, QUIET ENJOYMENT, * QUIET POSSESSION OR CORRESPONDENCE TO DESCRIPTION. THE ENTIRE RISK * ARISING OUT OF USE OR PERFORMANCE OF THE SOFTWARE LIES WITH YOU. ***********************license end**************************************/ #include #include "libbdk-hal/if/bdk-if.h" #include "libbdk-hal/bdk-qlm.h" #include "libbdk-hal/qlm/bdk-qlm-common.h" #include "libbdk-arch/bdk-csrs-gser.h" #include "libbdk-arch/bdk-csrs-pem.h" #include "libbdk-hal/bdk-config.h" #include "libbdk-hal/bdk-utils.h" #include "libbdk-hal/bdk-twsi.h" /* Indexed by QLM number and lane */ static uint64_t prbs_errors[14][4]; /** * Figure out which lane mode to use for a given reference clock and GBaud * * @param mode_name String name for error messages * @param qlm QlM being configured * @param ref_clk Reference clock in hertz * @param baud_mhz Baud rate in Mhz * * @return Lane mode or -1 on failure */ int __bdk_qlm_get_lane_mode_for_speed_and_ref_clk(const char *mode_name, int qlm, int ref_clk, int baud_mhz) { if (baud_mhz <= 1250) { if ((ref_clk == REF_156MHZ) || (ref_clk == REF_100MHZ)) return BDK_GSER_LMODE_E_R_125G_REFCLK15625_SGMII; else { bdk_error("Invalid reference clock for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } } else if (baud_mhz <= 2500) { if (ref_clk == REF_100MHZ) return BDK_GSER_LMODE_E_R_25G_REFCLK100; else if (ref_clk == REF_125MHZ) return BDK_GSER_LMODE_E_R_25G_REFCLK125; else { bdk_error("Invalid reference clock for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } } else if (baud_mhz <= 3125) { if (ref_clk == REF_156MHZ) return BDK_GSER_LMODE_E_R_3125G_REFCLK15625_XAUI; else { bdk_error("Invalid reference clock for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } } else if (baud_mhz <= 5000) { if (ref_clk == REF_100MHZ) return BDK_GSER_LMODE_E_R_5G_REFCLK100; else if (ref_clk == REF_125MHZ) return BDK_GSER_LMODE_E_R_5G_REFCLK125; else return BDK_GSER_LMODE_E_R_5G_REFCLK15625_QSGMII; } else if (baud_mhz <= 6250) { if (ref_clk == REF_156MHZ) return BDK_GSER_LMODE_E_R_625G_REFCLK15625_RXAUI; else { bdk_error("Invalid reference clock for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } } else if (baud_mhz <= 8000) { if (ref_clk == REF_100MHZ) return BDK_GSER_LMODE_E_R_8G_REFCLK100; else if (ref_clk == REF_125MHZ) return BDK_GSER_LMODE_E_R_8G_REFCLK125; else { bdk_error("Invalid reference clock for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } } else /* Baud 10312.5 */ { if (ref_clk == REF_156MHZ) return BDK_GSER_LMODE_E_R_103125G_REFCLK15625_KR; else { bdk_error("Invalid reference clock for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } } bdk_error("Invalid speed for %s on QLM%d with speed %d, ref %d Mhz\n", mode_name, qlm, baud_mhz, ref_clk / 1000000); return -1; } /** * Setup the PEM to either driver or receive reset from PRST based on RC or EP * * @param node Node to use in a Numa setup * @param pem Which PEM to setuo * @param is_endpoint * Non zero if PEM is a EP */ void __bdk_qlm_setup_pem_reset(bdk_node_t node, int pem, int is_endpoint) { /* Make sure is_endpoint is either 0 or 1 */ is_endpoint = (is_endpoint != 0); BDK_CSR_MODIFY(c, node, BDK_RST_CTLX(pem), c.s.prst_link = 0; /* Link down doesn't automatically assert PERST */ c.s.rst_link = is_endpoint; /* Link down automatically assert soft reset for EP */ c.s.rst_drv = !is_endpoint; /* PERST is output for RC, input for EP */ c.s.rst_rcv = is_endpoint; /* Only read PERST in EP mode */ c.s.rst_chip = 0); /* PERST doesn't pull CHIP_RESET */ if (is_endpoint) { /* If we're configuring an endpoint manually the PEM will not be turned on by default by the hardware. Turn it on now */ BDK_CSR_INIT(pemx_on, node, BDK_PEMX_ON(pem)); if (!pemx_on.s.pemon) { BDK_CSR_MODIFY(c, node, BDK_PEMX_CLK_EN(pem), c.cn83xx.pceclk_gate = 0; c.cn83xx.csclk_gate = 0); BDK_CSR_MODIFY(c, node, BDK_PEMX_ON(pem), c.s.pemon = 1); } } } /** * Measure the reference clock of a QLM * * @param qlm QLM to measure * * @return Clock rate in Hz */ int __bdk_qlm_measure_refclock(bdk_node_t node, int qlm) { /* Clear the counter */ BDK_CSR_MODIFY(c, node, BDK_GSERX_REFCLK_EVT_CTRL(qlm), c.s.enb = 0; c.s.clr = 1); bdk_wait_usec(1); /* Give counter a chance to clear */ if (BDK_CSR_READ(node, BDK_GSERX_REFCLK_EVT_CNTR(qlm))) bdk_error("GSER%d: Ref clock counter not zero\n", qlm); /* Start counting */ uint64_t start = bdk_clock_get_count(BDK_CLOCK_TIME); BDK_CSR_MODIFY(c, node, BDK_GSERX_REFCLK_EVT_CTRL(qlm), c.s.enb = 1; c.s.clr = 0); /* Wait for a short time to get a number of counts */ bdk_wait_usec(20000); /* 20ms */ /* Stop counting */ BDK_CSR_MODIFY(c, node, BDK_GSERX_REFCLK_EVT_CTRL(qlm), c.s.enb = 0); uint64_t stop = bdk_clock_get_count(BDK_CLOCK_TIME); bdk_wait_usec(1); /* Give counter a chance to stabalize */ /* Calculate the rate */ uint64_t count = BDK_CSR_READ(node, BDK_GSERX_REFCLK_EVT_CNTR(qlm)); count *= bdk_clock_get_rate(bdk_numa_local(), BDK_CLOCK_TIME); count /= stop - start; return count; } /** * Put a QLM into hardware reset * * @param node Node to use in a numa setup * @param qlm QLM to use * * @return Zero on success, negative on failure */ int __bdk_qlm_reset(bdk_node_t node, int qlm) { BDK_CSR_MODIFY(c, node, BDK_GSERX_PHY_CTL(qlm), c.s.phy_reset = 1); return 0; } /** * Enable PRBS on a QLM * * @param node Node to use in a numa setup * @param qlm QLM to use * @param prbs PRBS mode (31, etc) * @param dir Directions to enable. This is so you can enable TX and later * enable RX after TX has run for a time * * @return Zero on success, negative on failure */ int __bdk_qlm_enable_prbs(bdk_node_t node, int qlm, int prbs, bdk_qlm_direction_t dir) { const int NUM_LANES = bdk_qlm_get_lanes(node, qlm); int mode; switch (prbs) { case 31: mode = 1; break; case 23: mode = 2; /* Or 3? */ break; case 16: mode = 4; break; case 15: mode = 5; break; case 11: mode = 6; break; case 7: mode = 7; break; default: mode = prbs & 0xff; for (int lane = 0; lane < NUM_LANES; lane++) BDK_CSR_WRITE(node, BDK_GSERX_LANEX_LBERT_PAT_CFG(qlm, lane), prbs >> 8); BDK_TRACE(QLM, "Using mode 0x%x with custom pattern 0x%x\n", mode, prbs >> 8); break; } /* For some reason PRBS doesn't work if GSER is configured for PCIe. Disconnect PCIe when we start PRBS */ BDK_CSR_INIT(gserx_cfg, node, BDK_GSERX_CFG(qlm)); if (gserx_cfg.s.pcie) { gserx_cfg.s.pcie = 0; BDK_CSR_WRITE(node, BDK_GSERX_CFG(qlm), gserx_cfg.u); bdk_warn("N%d.QLM%d: Disabling PCIe for PRBS/pattern generation\n", node, qlm); } /* For some reason PRBS doesn't work if GSER is configured for SATA. Disconnect SATA when we start PRBS */ if (gserx_cfg.s.sata) { gserx_cfg.s.sata = 0; BDK_CSR_WRITE(node, BDK_GSERX_CFG(qlm), gserx_cfg.u); bdk_warn("N%d.QLM%d: Disabling SATA for PRBS/pattern generation\n", node, qlm); bdk_warn("N%d.QLM%d: SATA PRBS/patterns always run at 6G\n", node, qlm); } BDK_CSR_MODIFY(c, node, BDK_GSERX_PHY_CTL(qlm), c.s.phy_reset = 0); if (dir & BDK_QLM_DIRECTION_TX) { /* Disable first in case already running */ for (int lane = 0; lane < NUM_LANES; lane++) BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pg_en = 0); for (int lane = 0; lane < NUM_LANES; lane++) BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pg_en = 1; /* Enable generator */ c.s.lbert_pg_width = 3; /* 20 bit */ c.s.lbert_pg_mode = mode); } if (dir & BDK_QLM_DIRECTION_RX) { /* Clear the error counter and Disable the matcher */ for (int lane = 0; lane < NUM_LANES; lane++) { prbs_errors[qlm][lane] = 0; BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pm_en = 0); } for (int lane = 0; lane < NUM_LANES; lane++) { BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pm_en = 1; /* Enable matcher */ c.s.lbert_pm_width = 3; /* 20 bit */ c.s.lbert_pm_mode = mode); } /* Tell the matcher to start sync */ for (int retry=0; retry < 4; retry++) { for (int lane = 0; lane < NUM_LANES; lane++) { BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pm_sync_start = 1); } /* Wait 10ms */ bdk_wait_usec(10000); } } return 0; } /** * Disable PRBS on a QLM * * @param node Node to use in a numa setup * @param qlm QLM to use * * @return Zero on success, negative on failure */ int __bdk_qlm_disable_prbs(bdk_node_t node, int qlm) { const int NUM_LANES = bdk_qlm_get_lanes(node, qlm); BDK_CSR_INIT(phy_ctl, node, BDK_GSERX_PHY_CTL(qlm)); if (phy_ctl.s.phy_reset) return -1; for (int lane = 0; lane < NUM_LANES; lane++) { prbs_errors[qlm][lane] = 0; BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pg_en = 0; c.s.lbert_pm_en = 0); } return 0; } /** * Return the number of PRBS errors since PRBS started running * * @param node Node to use in numa setup * @param qlm QLM to use * @param lane Which lane * @param clear Clear counter after return the current value * * @return Number of errors */ uint64_t __bdk_qlm_get_prbs_errors(bdk_node_t node, int qlm, int lane, int clear) { /* Restart synchronization */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pm_sync_start = 1); /* This CSR is self clearing per the CSR description, but it doesn't seem to do that. Instead it clears when we trigger sync again */ BDK_CSR_INIT(rx, node, BDK_GSERX_LANEX_LBERT_ECNT(qlm, lane)); uint64_t errors = rx.s.lbert_err_cnt; if (rx.s.lbert_err_ovbit14) errors <<= 7; prbs_errors[qlm][lane] += errors; uint64_t result = prbs_errors[qlm][lane]; if (clear) prbs_errors[qlm][lane] = 0; return result; } /** * Inject an error into PRBS * * @param node Node to use in numa setup * @param qlm QLM to use * @param lane Which lane */ void __bdk_qlm_inject_prbs_error(bdk_node_t node, int qlm, int lane) { BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_LBERT_CFG(qlm, lane), c.s.lbert_pg_err_insert = 1); } /** * Enable shallow loopback on a QLM * * @param node Node to use in a numa setup * @param qlm QLM to use * @param loop Type of loopback. Not all QLMs support all modes * * @return Zero on success, negative on failure */ int __bdk_qlm_enable_loop(bdk_node_t node, int qlm, bdk_qlm_loop_t loop) { bdk_error("Chip doesn't support shallow QLM loopback\n"); return -1; } /** * Initialize the QLM mode table * * @param node Node to initialize * @param qlm Which QLM * @param ref_clk Reference clock of the QLM in Hz */ void __bdk_qlm_init_mode_table(bdk_node_t node, int qlm, int ref_clk) { /* The QLM PLLs are controlled by an array of parameters indexed by the QLM mode for each QLM. We need to fill in these tables. Also each lane has some mode parameters, again in a array index by the lane_mode */ for (int lane_mode = 0; lane_mode < 12; lane_mode++) { /* The values used below are all from http://mawiki.caveonetworks.com/wiki/78xx/GSER_WEST */ BDK_CSR_INIT(pll_mode_0 , node, BDK_GSERX_PLL_PX_MODE_0(qlm, lane_mode)); BDK_CSR_INIT(pll_mode_1 , node, BDK_GSERX_PLL_PX_MODE_1(qlm, lane_mode)); BDK_CSR_INIT(lane_mode_0, node, BDK_GSERX_LANE_PX_MODE_0(qlm, lane_mode)); BDK_CSR_INIT(lane_mode_1, node, BDK_GSERX_LANE_PX_MODE_1(qlm, lane_mode)); switch (lane_mode) { case BDK_GSER_LMODE_E_R_25G_REFCLK100: case BDK_GSER_LMODE_E_R_5G_REFCLK100: case BDK_GSER_LMODE_E_R_8G_REFCLK100: /* These modes are used for PCIe where the defaults are correct. Skip programming these */ continue; case BDK_GSER_LMODE_E_R_125G_REFCLK15625_KX: pll_mode_0.s.pll_icp = 0x1; pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0x28; pll_mode_1.s.pll_16p5en = 0x1; pll_mode_1.s.pll_cpadj = 0x3; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; pll_mode_1.s.pll_div = 0x10; lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x2; lane_mode_0.s.rx_ldiv = 0x2; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x1; lane_mode_1.s.cdr_fgain = 0xc; lane_mode_1.s.ph_acc_adj = 0x1e; break; case BDK_GSER_LMODE_E_R_3125G_REFCLK15625_XAUI: pll_mode_0.s.pll_icp = 0x1; pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0x14; pll_mode_1.s.pll_16p5en = 0x1; pll_mode_1.s.pll_cpadj = 0x2; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; pll_mode_1.s.pll_div = 0x14; lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x1; lane_mode_0.s.rx_ldiv = 0x1; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x1; lane_mode_1.s.cdr_fgain = 0xc; lane_mode_1.s.ph_acc_adj = 0x1e; break; case BDK_GSER_LMODE_E_R_103125G_REFCLK15625_KR: pll_mode_0.s.pll_icp = 0x1; pll_mode_0.s.pll_rloop = 0x5; pll_mode_0.s.pll_pcs_div = 0xa; pll_mode_1.s.pll_16p5en = 0x1; pll_mode_1.s.pll_cpadj = 0x2; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x1; pll_mode_1.s.pll_div = 0x21; lane_mode_0.s.ctle = 0x3; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x0; lane_mode_0.s.rx_ldiv = 0x0; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x1; lane_mode_1.s.vma_mm = 0x0; lane_mode_1.s.cdr_fgain = 0xa; lane_mode_1.s.ph_acc_adj = 0xf; break; case BDK_GSER_LMODE_E_R_125G_REFCLK15625_SGMII: pll_mode_0.s.pll_icp = 0x1; pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0x28; pll_mode_1.s.pll_16p5en = 0x1; pll_mode_1.s.pll_cpadj = 0x3; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; pll_mode_1.s.pll_div = 0x10; lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x2; lane_mode_0.s.rx_ldiv = 0x2; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x1; lane_mode_1.s.cdr_fgain = 0xc; lane_mode_1.s.ph_acc_adj = 0x1e; if(ref_clk == REF_100MHZ) { pll_mode_0.s.pll_pcs_div = 0x28; pll_mode_1.s.pll_div = 0x19; pll_mode_1.s.pll_cpadj = 0x2; } break; case BDK_GSER_LMODE_E_R_5G_REFCLK15625_QSGMII: pll_mode_0.s.pll_icp = 0x1; /* Per Scott McIlhenny 5/17/2016 (t81) */ pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0xa; pll_mode_1.s.pll_16p5en = 0x0; pll_mode_1.s.pll_cpadj = 0x2; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; /* QSGMII is a special case. We use the same table entry for 100Mhz and 125Mhz clocks as the normal 156Mhz */ switch (ref_clk) { case REF_100MHZ: pll_mode_1.s.pll_div = 0x19; break; case REF_125MHZ: pll_mode_1.s.pll_div = 0x14; break; default: /* REF_156MHZ */ pll_mode_1.s.pll_div = 0x10; break; } lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x0; lane_mode_0.s.rx_ldiv = 0x0; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x1; /* Per Scott McIlhenny 5/17/2016 (t81) */ lane_mode_1.s.cdr_fgain = 0xc; lane_mode_1.s.ph_acc_adj = 0x1e; break; case BDK_GSER_LMODE_E_R_625G_REFCLK15625_RXAUI: pll_mode_0.s.pll_icp = 0x1; pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0xa; pll_mode_1.s.pll_16p5en = 0x0; pll_mode_1.s.pll_cpadj = 0x2; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; pll_mode_1.s.pll_div = 0x14; lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x0; lane_mode_0.s.rx_ldiv = 0x0; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x0; lane_mode_1.s.cdr_fgain = 0xa; lane_mode_1.s.ph_acc_adj = 0x14; break; case BDK_GSER_LMODE_E_R_25G_REFCLK125: pll_mode_0.s.pll_icp = 0x3; pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0x5; pll_mode_1.s.pll_16p5en = 0x0; pll_mode_1.s.pll_cpadj = 0x1; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; pll_mode_1.s.pll_div = 0x14; lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x1; lane_mode_0.s.tx_ldiv = 0x1; lane_mode_0.s.rx_ldiv = 0x1; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x1; lane_mode_1.s.cdr_fgain = 0xa; lane_mode_1.s.ph_acc_adj = 0x14; break; case BDK_GSER_LMODE_E_R_5G_REFCLK125: pll_mode_0.s.pll_icp = 0x3; pll_mode_0.s.pll_rloop = 0x3; pll_mode_0.s.pll_pcs_div = 0xa; pll_mode_1.s.pll_16p5en = 0x0; pll_mode_1.s.pll_cpadj = 0x1; pll_mode_1.s.pll_pcie3en = 0x0; pll_mode_1.s.pll_opr = 0x0; pll_mode_1.s.pll_div = 0x14; lane_mode_0.s.ctle = 0x0; lane_mode_0.s.pcie = 0x1; lane_mode_0.s.tx_ldiv = 0x0; lane_mode_0.s.rx_ldiv = 0x0; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x3; lane_mode_0.s.rx_mode = 0x3; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x0; lane_mode_1.s.cdr_fgain = 0xa; lane_mode_1.s.ph_acc_adj = 0x14; break; case BDK_GSER_LMODE_E_R_8G_REFCLK125: pll_mode_0.s.pll_icp = 0x2; pll_mode_0.s.pll_rloop = 0x5; pll_mode_0.s.pll_pcs_div = 0xa; pll_mode_1.s.pll_16p5en = 0x0; pll_mode_1.s.pll_cpadj = 0x1; pll_mode_1.s.pll_pcie3en = 0x1; pll_mode_1.s.pll_opr = 0x1; pll_mode_1.s.pll_div = 0x20; lane_mode_0.s.ctle = 0x3; lane_mode_0.s.pcie = 0x0; lane_mode_0.s.tx_ldiv = 0x0; lane_mode_0.s.rx_ldiv = 0x0; lane_mode_0.s.srate = 0x0; lane_mode_0.s.tx_mode = 0x2; lane_mode_0.s.rx_mode = 0x2; lane_mode_1.s.vma_fine_cfg_sel = 0x0; lane_mode_1.s.vma_mm = 0x0; lane_mode_1.s.cdr_fgain = 0xb; lane_mode_1.s.ph_acc_adj = 0x23; break; } BDK_CSR_WRITE(node, BDK_GSERX_PLL_PX_MODE_0(qlm, lane_mode), pll_mode_0.u); BDK_CSR_WRITE(node, BDK_GSERX_PLL_PX_MODE_1(qlm, lane_mode), pll_mode_1.u); BDK_CSR_WRITE(node, BDK_GSERX_LANE_PX_MODE_0(qlm, lane_mode), lane_mode_0.u); BDK_CSR_WRITE(node, BDK_GSERX_LANE_PX_MODE_1(qlm, lane_mode), lane_mode_1.u); } } /** * Given a valid PEM number, return its speed in Gbaud * * @param node Node to use in numa setup * @param pem PEM to get speed of * * @return Speed in Gbaud. Zero if disabled */ int __bdk_qlm_get_gbaud_mhz_pem(bdk_node_t node, int pem) { BDK_CSR_INIT(pem_cfg, node, BDK_PEMX_CFG(pem)); switch (pem_cfg.cn83xx.md) { case 0: /* Gen 1 */ return 2500; case 1: /* Gen 2 */ return 5000; case 2: /* Gen 3 */ return 8000; default: return 0; } } /** * Get the speed of a QLM using its LMODE. This can't be used on PCIe QLMs. * * @param node Node to use in numa setup * @param qlm Which QLM * * @return QLM speed on Gbaud */ int __bdk_qlm_get_gbaud_mhz_lmode(bdk_node_t node, int qlm) { /* QLM is not in PCIe, assume LMODE is good enough for determining the speed */ BDK_CSR_INIT(lane_mode, node, BDK_GSERX_LANE_MODE(qlm)); switch (lane_mode.s.lmode) { case BDK_GSER_LMODE_E_R_25G_REFCLK100: return 2500; case BDK_GSER_LMODE_E_R_5G_REFCLK100: return 5000; case BDK_GSER_LMODE_E_R_8G_REFCLK100: return 8000; case BDK_GSER_LMODE_E_R_125G_REFCLK15625_KX: return 1250; case BDK_GSER_LMODE_E_R_3125G_REFCLK15625_XAUI: return 3125; case BDK_GSER_LMODE_E_R_103125G_REFCLK15625_KR: return 10312; case BDK_GSER_LMODE_E_R_125G_REFCLK15625_SGMII: return 1250; case BDK_GSER_LMODE_E_R_5G_REFCLK15625_QSGMII: return 5000; case BDK_GSER_LMODE_E_R_625G_REFCLK15625_RXAUI: return 6250; case BDK_GSER_LMODE_E_R_25G_REFCLK125: return 2500; case BDK_GSER_LMODE_E_R_5G_REFCLK125: return 5000; case BDK_GSER_LMODE_E_R_8G_REFCLK125: return 8000; default: return 0; } } /** * Converts a measured reference clock to a likely ideal value. Rounds * clock speed to the nearest REF_*Mhz define. * * @param node Node to use in numa setup * @param qlm Which QLM * @param measured_hz * Measured value * * @return Value exactly matching a define */ int __bdk_qlm_round_refclock(bdk_node_t node, int qlm, int measured_hz) { int ref_clk; if ((measured_hz > REF_100MHZ - REF_100MHZ / 10) && (measured_hz < REF_100MHZ + REF_100MHZ / 10)) { ref_clk = REF_100MHZ; } else if ((measured_hz > REF_125MHZ - REF_125MHZ / 10) && (measured_hz < REF_125MHZ + REF_125MHZ / 10)) { ref_clk = REF_125MHZ; } else if ((measured_hz > REF_156MHZ - REF_156MHZ / 10) && (measured_hz < REF_156MHZ + REF_156MHZ / 10)) { ref_clk = REF_156MHZ; } else if (measured_hz < 1000000) { ref_clk = 0; /* Used for disabled QLMs */ } else { ref_clk = measured_hz; bdk_error("N%d.QLM%d: Unexpected reference clock speed of %d Mhz\n", node, qlm, measured_hz / 1000000); } return ref_clk; } /** * TWSI reads from the MCU randomly timeout. Retry a few times on * failure to try and recover * * @param node Node to use in a Numa setup. Can be an exact ID or a special * value. * @param twsi_id which TWSI bus to use * @param dev_addr Device address (7 bit) * @param internal_addr * Internal address. Can be 0, 1 or 2 bytes in width * @param num_bytes Number of data bytes to read (1-4) * @param ia_width_bytes * Internal address size in bytes (0, 1, or 2) * * @return Read data, or -1 on failure */ static int64_t mcu_read(bdk_node_t node, int twsi_id, uint8_t dev_addr, uint16_t internal_addr, int num_bytes, int ia_width_bytes) { int read_tries = 0; int64_t result; do { result = bdk_twsix_read_ia(node, twsi_id, dev_addr, internal_addr, num_bytes, ia_width_bytes); read_tries++; if (result < 0) { BDK_TRACE(QLM, "Timeout %d reading from MCU\n", read_tries); bdk_wait_usec(100000); } } while ((result < 0) && (read_tries < 3)); return result; } static void __bdk_qlm_set_reference(bdk_node_t node, int qlm, int ref_clk) { int use_clock; if (CAVIUM_IS_MODEL(CAVIUM_CN88XX) || CAVIUM_IS_MODEL(CAVIUM_CN83XX) || CAVIUM_IS_MODEL(CAVIUM_CN81XX)) { switch (ref_clk) { case REF_156MHZ: use_clock = 0; /* Common clock 0 */ BDK_TRACE(QLM, "Setting N%d.QLM%d to use common clock 0\n", node, qlm); break; case REF_100MHZ: use_clock = 1; /* Common clock 1 */ BDK_TRACE(QLM, "Setting N%d.QLM%d to use common clock 1\n", node, qlm); break; default: use_clock = 2; /* External clock */ BDK_TRACE(QLM, "Setting N%d.QLM%d to use external clock\n", node, qlm); break; } } else { bdk_error("Update __bdk_qlm_set_reference() for qlm auto config of this chip\n"); return; } BDK_CSR_MODIFY(c, node, BDK_GSERX_REFCLK_SEL(qlm), c.s.com_clk_sel = (use_clock != 2); c.s.use_com1 = (use_clock == 1)); } /** * For Cavium EVB and EBB board, query the MCU to determine the QLM setup. Applying * any configuration found. * * @param node Node to configure * * @return Zero on success, negative on failure */ int bdk_qlm_mcu_auto_config(bdk_node_t node) { const int MCU_TWSI_BUS = 0; const int MCU_TWSI_ADDRESS = 0x60; int64_t data; /* Check the two magic number bytes the MCU should return */ data = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x00, 1, 1); if (data != 0xa5) { printf("QLM Config: MCU not found, skipping auto configuration\n"); return -1; } data = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x01, 1, 1); if (data != 0x5a) { bdk_error("QLM Config: MCU magic number incorrect\n"); return -1; } /* Read the MCU version */ int mcu_major = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x02, 1, 1); int mcu_minor = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x03, 1, 1); BDK_TRACE(QLM, "MCU version %d.%d\n", mcu_major, mcu_minor); if ((mcu_major < 2) || ((mcu_major == 2) && (mcu_minor < 30))) { bdk_error("QLM Config: Unexpected MCU version %d.%d\n", mcu_major, mcu_minor); return -1; } /* Find out how many lanes the MCU thinks are available */ int lanes = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x16, 1, 1); BDK_TRACE(QLM, "MCU says board has %d lanes\n", lanes); int correct_lanes = 0; if (cavium_is_altpkg(CAVIUM_CN88XX)) correct_lanes = 22; else if (CAVIUM_IS_MODEL(CAVIUM_CN88XX)) correct_lanes = 32; else if (CAVIUM_IS_MODEL(CAVIUM_CN83XX)) correct_lanes = 22; else if (CAVIUM_IS_MODEL(CAVIUM_CN81XX)) correct_lanes = 8; if (lanes != correct_lanes) { bdk_error("QLM Config: Unexpected number of lanes (%d) from MCU\n", lanes); return -1; } int lane = 0; int qlm = 0; while (lane < lanes) { int write_status; int width; int mode; int speed; int refclk; /* TWSI reads from the MCU randomly timeout. Retry a few times on failure to try and recover */ int read_tries = 0; do { read_tries++; if (read_tries > 3) { bdk_error("QLM Config: Timeouts reading from MCU\n"); return -1; } /* Space request out 20ms */ bdk_wait_usec(20000); /* Select the lane we are interested in */ write_status = bdk_twsix_write_ia(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x16, 1, 1, lane); /* Space request out 20ms */ bdk_wait_usec(20000); /* Get the mode */ width = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x17, 1, 1); mode = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x18, 2, 1); speed = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x19, 2, 1); refclk = mcu_read(node, MCU_TWSI_BUS, MCU_TWSI_ADDRESS, 0x1a, 1, 1); } while ((write_status < 0) || (width < 0) || (mode < 0) || (speed < 0) || (refclk < 0)); BDK_TRACE(QLM, "MCU lane %d, width %d, mode 0x%x, speed 0x%x, ref 0x%x\n", lane, width, mode, speed, refclk); if ((width != 0) && (width != 1) && (width != 2) && (width != 4) && (width != 8)) { bdk_error("QLM Config: Unexpected interface width (%d) from MCU\n", width); return -1; } /* MCU reports a width of 0 for unconfigured QLMs. It reports a width of 1 for some combinations on CN80XX, and two on others. Convert either 0 or 1 to the actual width, or 2 for CN80XX. Yuck */ if ((width == 0) || (width == 1)) { if (cavium_is_altpkg(CAVIUM_CN81XX) && (qlm < 2)) width = 2; else width = bdk_qlm_get_lanes(node, qlm); } bdk_qlm_modes_t qlm_mode; int qlm_speed = (speed >> 8) * 1000 + (speed & 0xff) * 1000 / 256; int use_ref = 0; bdk_qlm_mode_flags_t qlm_flags = 0; if (mode < 0x4000) { switch (mode) { case 0x0000: /* No Configuration */ qlm_mode = BDK_QLM_MODE_DISABLED; break; case 0x0101: /* PCIe Host */ qlm_mode = (width == 8) ? BDK_QLM_MODE_PCIE_1X8 : (width == 4) ? BDK_QLM_MODE_PCIE_1X4 : BDK_QLM_MODE_PCIE_1X2; use_ref = REF_100MHZ; break; case 0x0102: /* PCIe Endpoint */ qlm_mode = (width == 8) ? BDK_QLM_MODE_PCIE_1X8 : (width == 4) ? BDK_QLM_MODE_PCIE_1X4 : BDK_QLM_MODE_PCIE_1X2; qlm_flags = BDK_QLM_MODE_FLAG_ENDPOINT; use_ref = 0; /* Use the external reference for EP mode */ break; case 0x1000: /* SGMII */ qlm_mode = (width == 4) ? BDK_QLM_MODE_SGMII_4X1 : (width == 2) ? BDK_QLM_MODE_SGMII_2X1 : BDK_QLM_MODE_SGMII_1X1; use_ref = REF_156MHZ; /* CN80XX parts on EBBs use phy port 2 for SGMII, while QSGMII uses the correct port. Fix this for DLM1 and DLM3 */ if (cavium_is_altpkg(CAVIUM_CN81XX)) { int bgx = (qlm == 3) ? 1 : 0; uint64_t phy = bdk_config_get_int(BDK_CONFIG_PHY_ADDRESS, 0, bgx, 2); bdk_config_set_int(phy, BDK_CONFIG_PHY_ADDRESS, 0, bgx, 1); } break; case 0x1100: /* QSGMII */ qlm_mode = BDK_QLM_MODE_QSGMII_4X1; use_ref = REF_100MHZ; break; case 0x2000: /* XAUI */ qlm_mode = BDK_QLM_MODE_XAUI_1X4; use_ref = REF_156MHZ; break; case 0x2100: /* RXAUI */ qlm_mode = (width == 2) ? BDK_QLM_MODE_RXAUI_1X2 : BDK_QLM_MODE_RXAUI_2X2; use_ref = REF_156MHZ; break; case 0x2200: /* DXAUI */ qlm_mode = BDK_QLM_MODE_XAUI_1X4; use_ref = REF_156MHZ; break; case 0x3001: /* Interlaken */ qlm_mode = BDK_QLM_MODE_ILK; use_ref = REF_156MHZ; break; default: bdk_error("QLM Config: Unexpected interface mode (0x%x) from MCU\n", mode); qlm_mode = BDK_QLM_MODE_DISABLED; break; } } else { switch (mode) { case 0x4000: /* SATA */ qlm_mode = (width == 2) ? BDK_QLM_MODE_SATA_2X1 : BDK_QLM_MODE_SATA_4X1; use_ref = REF_100MHZ; break; case 0x5001: /* XFI */ qlm_mode = (width == 4) ? BDK_QLM_MODE_XFI_4X1 : (width == 2) ? BDK_QLM_MODE_XFI_2X1 : BDK_QLM_MODE_XFI_1X1; use_ref = REF_156MHZ; break; case 0x5002: /* 10G-KR */ qlm_mode = (width == 4) ? BDK_QLM_MODE_10G_KR_4X1 : (width == 2) ? BDK_QLM_MODE_10G_KR_2X1 : BDK_QLM_MODE_10G_KR_1X1; use_ref = REF_156MHZ; break; case 0x6001: /* XLAUI */ qlm_mode = BDK_QLM_MODE_XLAUI_1X4; use_ref = REF_156MHZ; break; case 0x6002: /* 40G-KR4 */ qlm_mode = BDK_QLM_MODE_40G_KR4_1X4; use_ref = REF_156MHZ; break; default: bdk_error("QLM Config: Unexpected interface mode (0x%x) from MCU\n", mode); qlm_mode = BDK_QLM_MODE_DISABLED; break; } } lane += width; do { int internal_qlm = qlm; /* Alternate package parts have different QLM numbers for internal versus external. The MCU uses the external numbers */ if (cavium_is_altpkg(CAVIUM_CN88XX)) { switch (qlm) { case 0: /* QLM0 -> QLM4 */ internal_qlm = 4; break; case 1: /* QLM1 -> QLM5 */ internal_qlm = 5; break; case 2: /* QLM2 -> QLM0 */ internal_qlm = 0; break; case 3: /* QLM3 -> QLM1 */ internal_qlm = 1; break; case 4: /* DLM4 -> QLM2 */ internal_qlm = 2; break; case 5: /* DLM5 -> QLM6 */ internal_qlm = 6; break; case 6: /* DLM6 -> QLM7 */ internal_qlm = 7; break; default: bdk_error("Invalid external QLM%d from MCU\n", qlm); return -1; } } if (qlm_flags & BDK_QLM_MODE_FLAG_ENDPOINT) { BDK_TRACE(QLM, "Skipping N%d.QLM%d mode %s(%d), speed %d, flags 0x%x (EP should already be setup)\n", node, internal_qlm, bdk_qlm_mode_tostring(qlm_mode), qlm_mode, qlm_speed, qlm_flags); } else { BDK_TRACE(QLM, "Setting N%d.QLM%d mode %s(%d), speed %d, flags 0x%x\n", node, internal_qlm, bdk_qlm_mode_tostring(qlm_mode), qlm_mode, qlm_speed, qlm_flags); /* Set the reference clock for this QLM */ __bdk_qlm_set_reference(node, internal_qlm, use_ref); if (bdk_qlm_set_mode(node, internal_qlm, qlm_mode, qlm_speed, qlm_flags)) return -1; } int num_lanes = bdk_qlm_get_lanes(node, internal_qlm); /* CN86XX looks like two lanes each for DLM4-7 */ if (cavium_is_altpkg(CAVIUM_CN88XX) && (qlm >= 4)) num_lanes = 2; if (qlm_mode == BDK_QLM_MODE_PCIE_1X8) { /* PCIe x8 is a special case as the QLM config function actually configures both QLMs in one go */ qlm++; width -= 8; } else if ((qlm_mode == BDK_QLM_MODE_PCIE_1X4) && (width > num_lanes)) { /* PCIe x4 is a special case as the QLM config function actually configures both QLMs in one go */ qlm++; width -= 4; } else if (width >= num_lanes) { if (num_lanes == 1) width -= 2; /* Special case for CN80XX */ else width -= num_lanes; } else width = 0; qlm++; } while (width > 0); } return 0; } /** * Display the current settings of a QLM lane * * @param node Node the QLM is on * @param qlm QLM to display * @param qlm_lane Lane to use * @param show_tx Display TX parameters * @param show_rx Display RX parameters */ void bdk_qlm_display_settings(bdk_node_t node, int qlm, int qlm_lane, bool show_tx, bool show_rx) { const char *dir_label[] = {"Hold", "Inc", "Dec", "Hold"}; uint64_t rx_aeq_out_0 = BDK_CSR_READ(node, BDK_GSERX_LANEX_RX_AEQ_OUT_0(qlm, qlm_lane)); uint64_t rx_aeq_out_1 = BDK_CSR_READ(node, BDK_GSERX_LANEX_RX_AEQ_OUT_1(qlm, qlm_lane)); uint64_t rx_aeq_out_2 = BDK_CSR_READ(node, BDK_GSERX_LANEX_RX_AEQ_OUT_2(qlm, qlm_lane)); uint64_t rx_vma_status_0 = BDK_CSR_READ(node, BDK_GSERX_LANEX_RX_VMA_STATUS_0(qlm, qlm_lane)); uint64_t rx_vma_status_1 = BDK_CSR_READ(node, BDK_GSERX_LANEX_RX_VMA_STATUS_1(qlm, qlm_lane)); uint64_t sds_pin_mon_1 = BDK_CSR_READ(node, BDK_GSERX_LANEX_SDS_PIN_MON_1(qlm, qlm_lane)); uint64_t sds_pin_mon_2 = BDK_CSR_READ(node, BDK_GSERX_LANEX_SDS_PIN_MON_2(qlm, qlm_lane)); uint64_t br_rxx_eer = BDK_CSR_READ(node, BDK_GSERX_BR_RXX_EER(qlm, qlm_lane)); printf("N%d.QLM%d Lane %d:\n", node, qlm, qlm_lane); if (show_rx) { printf(" DFE Tap 1: %llu, Tap 2: %lld, Tap 3: %lld, Tap 4: %lld, Tap 5: %lld\n", bdk_extract(rx_aeq_out_1, 0, 5), bdk_extract_smag(rx_aeq_out_1, 5, 9), bdk_extract_smag(rx_aeq_out_1, 10, 14), bdk_extract_smag(rx_aeq_out_0, 0, 4), bdk_extract_smag(rx_aeq_out_0, 5, 9)); printf(" Pre-CTLE Gain: %llu, Post-CTLE Gain: %llu, CTLE Peak: %llu, CTLE Pole: %llu\n", bdk_extract(rx_aeq_out_2, 4, 4), bdk_extract(rx_aeq_out_2, 0, 4), bdk_extract(rx_vma_status_0, 2, 4), bdk_extract(rx_vma_status_0, 0, 2)); printf(" RX Equalization Tx Directions Hints TXPRE: %s, TXMAIN: %s, TXPOST: %s, Figure of Merit: %llu\n", dir_label[bdk_extract(br_rxx_eer, 0, 2)], dir_label[bdk_extract(br_rxx_eer, 2, 2)], dir_label[bdk_extract(br_rxx_eer, 4, 2)], bdk_extract(br_rxx_eer, 6, 8)); } if (show_tx) { printf(" TX Swing: %llu, Pre-emphasis Pre-cursor: %llu, Post-cursor: %llu\n", bdk_extract(sds_pin_mon_1, 1, 5), bdk_extract(sds_pin_mon_2, 0, 4), bdk_extract(sds_pin_mon_2, 4, 5)); printf(" TX Boost Enable: %llu, TX Turbo Mode: %llu\n", bdk_extract(sds_pin_mon_2, 10, 1), bdk_extract(sds_pin_mon_2, 9, 1)); } printf(" Training-done: %llu\n", bdk_extract(rx_vma_status_1, 7, 1)); } /** * Perform RX equalization on a QLM * * @param node Node the QLM is on * @param qlm QLM to perform RX equalization on * @param qlm_lane Lane to use, or -1 for all lanes * * @return Zero on success, negative if any lane failed RX equalization */ int __bdk_qlm_rx_equalization(bdk_node_t node, int qlm, int qlm_lane) { /* Don't touch QLMs is reset or powered down */ BDK_CSR_INIT(phy_ctl, node, BDK_GSERX_PHY_CTL(qlm)); if (phy_ctl.s.phy_pd || phy_ctl.s.phy_reset) return -1; /* Don't run on PCIe links */ if (bdk_qlm_get_mode(node, qlm) <= BDK_QLM_MODE_PCIE_1X8) return -1; int fail = 0; /* Bitmask of lanes that failed CDR Lock or Eltrical Idle check */ int pending = 0; /* Bitmask of lanes that we're waiting for */ int MAX_LANES = bdk_qlm_get_lanes(node, qlm); BDK_TRACE(QLM, "N%d.QLM%d: Starting RX equalization on lane %d\n", node, qlm, qlm_lane); for (int lane = 0; lane < MAX_LANES; lane++) { /* Skip lanes we don't care about */ if ((qlm_lane != -1) && (qlm_lane != lane)) continue; /* Check that the lane has completed CDR lock */ BDK_CSR_INIT(eie_detsts, node, BDK_GSERX_RX_EIE_DETSTS(qlm)); if (((1 << lane) & eie_detsts.s.cdrlock) == 0) { /* Mark bad so we skip this lane below */ fail |= 1 << lane; continue; } /* Enable software control */ BDK_CSR_MODIFY(c, node, BDK_GSERX_BR_RXX_CTL(qlm, lane), c.s.rxt_swm = 1); /* Clear the completion flag and initiate a new request */ BDK_CSR_MODIFY(c, node, BDK_GSERX_BR_RXX_EER(qlm, lane), c.s.rxt_esv = 0; c.s.rxt_eer = 1); /* Remember that we have to wait for this lane */ pending |= 1 << lane; } /* Timing a few of these over XFI on CN73XX, each takes 21-23ms. XLAUI was about the same time. DXAUI and RXAUI both took 2-3ms. Put the timeout at 250ms, which is roughly 10x my measurements. */ uint64_t timeout = bdk_clock_get_count(BDK_CLOCK_TIME) + bdk_clock_get_rate(node, BDK_CLOCK_TIME) / 4; while (pending) { for (int lane = 0; lane < MAX_LANES; lane++) { int lane_mask = 1 << lane; /* Only check lanes that are pending */ if (!(pending & lane_mask)) continue; /* Read the registers for checking Electrical Idle / CDR lock and the status of the RX equalization */ BDK_CSR_INIT(eie_detsts, node, BDK_GSERX_RX_EIE_DETSTS(qlm)); BDK_CSR_INIT(gserx_br_rxx_eer, node, BDK_GSERX_BR_RXX_EER(qlm, lane)); /* Mark failure if lane entered Electrical Idle or lost CDR Lock. The bit for the lane will have cleared in either EIESTS or CDRLOCK */ if (!(eie_detsts.s.eiests & eie_detsts.s.cdrlock & lane_mask)) { fail |= lane_mask; pending &= ~lane_mask; } else if (gserx_br_rxx_eer.s.rxt_esv) { /* Clear pending if RX equalization finished */ pending &= ~lane_mask; } } /* Break out of the loop on timeout */ if (bdk_clock_get_count(BDK_CLOCK_TIME) > timeout) break; } /* Cleanup and report status */ for (int lane = 0; lane < MAX_LANES; lane++) { /* Skip lanes we don't care about */ if ((qlm_lane != -1) && (qlm_lane != lane)) continue; int lane_mask = 1 << lane; /* Get the final RX equalization status */ BDK_CSR_INIT(gserx_br_rxx_eer, node, BDK_GSERX_BR_RXX_EER(qlm, lane)); /* Disable software control */ BDK_CSR_MODIFY(c, node, BDK_GSERX_BR_RXX_CTL(qlm, lane), c.s.rxt_swm = 0); /* Report status */ if (fail & lane_mask) { BDK_TRACE(QLM, "N%d.QLM%d: Lane %d RX equalization lost CDR Lock or entered Electrical Idle\n", node, qlm, lane); } else if ((pending & lane_mask) || !gserx_br_rxx_eer.s.rxt_esv) { BDK_TRACE(QLM, "N%d.QLM%d: Lane %d RX equalization timeout\n", node, qlm, lane); fail |= 1 << lane; } else { bdk_qlm_display_settings(node, qlm, lane, false, true); } } return (fail) ? -1 : 0; } /** * Configure the TX tuning parameters for a QLM lane. The tuning parameters can * be specified as -1 to maintain their current value * * @param node Node to configure * @param qlm QLM to configure * @param lane Lane to configure * @param tx_swing Transmit swing (coef 0) Range 0-31 * @param tx_pre Pre cursor emphasis (Coef -1). Range 0-15 * @param tx_post Post cursor emphasis (Coef +1). Range 0-31 * @param tx_gain Transmit gain. Range 0-7 * @param tx_vboost Transmit voltage boost. Range 0-1 * * @return Zero on success, negative on failure */ int __bdk_qlm_tune_lane_tx(bdk_node_t node, int qlm, int lane, int tx_swing, int tx_pre, int tx_post, int tx_gain, int tx_vboost) { /* Check tuning constraints */ if ((tx_swing < -1) || (tx_swing > 25)) { bdk_error("N%d.QLM%d: Lane %d: Invalid TX_SWING(%d)\n", node, qlm, lane, tx_swing); return -1; } if ((tx_pre < -1) || (tx_pre > 10)) { bdk_error("N%d.QLM%d: Lane %d: Invalid TX_PRE(%d)\n", node, qlm, lane, tx_pre); return -1; } if ((tx_post < -1) || (tx_post > 15)) { bdk_error("N%d.QLM%d: Lane %d: Invalid TX_POST(%d)\n", node, qlm, lane, tx_post); return -1; } if ((tx_pre >= 0) && (tx_post >= 0) && (tx_swing >= 0) && (tx_pre + tx_post - tx_swing > 2)) { bdk_error("N%d.QLM%d: Lane %d: TX_PRE(%d) + TX_POST(%d) - TX_SWING(%d) must be less than or equal to 2\n", node, qlm, lane, tx_pre, tx_post, tx_swing); return -1; } if ((tx_pre >= 0) && (tx_post >= 0) && (tx_swing >= 0) && (tx_pre + tx_post + tx_swing > 35)) { bdk_error("N%d.QLM%d: Lane %d: TX_PRE(%d) + TX_POST(%d) + TX_SWING(%d) must be less than or equal to 35\n", node, qlm, lane, tx_pre, tx_post, tx_swing); return -1; } if ((tx_gain < -1) || (tx_gain > 7)) { bdk_error("N%d.QLM%d: Lane %d: Invalid TX_GAIN(%d). TX_GAIN must be between 0 and 7\n", node, qlm, lane, tx_gain); return -1; } if ((tx_vboost < -1) || (tx_vboost > 1)) { bdk_error("N%d.QLM%d: Lane %d: Invalid TX_VBOOST(%d). TX_VBOOST must be 0 or 1.\n", node, qlm, lane, tx_vboost); return -1; } if ((tx_pre != -1) && (tx_post == -1)) { BDK_CSR_INIT(emphasis, node, BDK_GSERX_LANEX_TX_PRE_EMPHASIS(qlm, lane)); tx_post = emphasis.s.cfg_tx_premptap >> 4; } if ((tx_post != -1) && (tx_pre == -1)) { BDK_CSR_INIT(emphasis, node, BDK_GSERX_LANEX_TX_PRE_EMPHASIS(qlm, lane)); tx_pre = emphasis.s.cfg_tx_premptap & 0xf; } BDK_TRACE(QLM, "N%d.QLM%d: Lane %d: TX_SWING=%d, TX_PRE=%d, TX_POST=%d, TX_GAIN=%d, TX_VBOOST=%d\n", node, qlm, lane, tx_swing, tx_pre, tx_post, tx_gain, tx_vboost); /* Manual Tx Swing and Tx Equalization Programming Steps */ /* 1) Enable Tx swing and Tx emphasis overrides */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_TX_CFG_1(qlm, lane), c.s.tx_swing_ovrrd_en = (tx_swing != -1); c.s.tx_premptap_ovrrd_val = (tx_pre != -1) && (tx_post != -1); c.s.tx_vboost_en_ovrrd_en = (tx_vboost != -1)); /* Vboost override */ /* 2) Program the Tx swing and Tx emphasis Pre-cursor and Post-cursor values */ if (tx_swing != -1) BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_TX_CFG_0(qlm, lane), c.s.cfg_tx_swing = tx_swing); if ((tx_pre != -1) && (tx_post != -1)) BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_TX_PRE_EMPHASIS(qlm, lane), c.s.cfg_tx_premptap = (tx_post << 4) | tx_pre); /* Apply TX gain settings */ if (tx_gain != -1) BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_TX_CFG_3(qlm, lane), c.s.pcs_sds_tx_gain = tx_gain); /* Apply TX vboost settings */ if (tx_vboost != -1) BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_TX_CFG_3(qlm, lane), c.s.cfg_tx_vboost_en = tx_vboost); /* 3) Program override for the Tx coefficient request */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_PCS_CTLIFC_0(qlm, lane), if (((tx_pre != -1) && (tx_post != -1)) || (tx_swing != -1)) c.s.cfg_tx_coeff_req_ovrrd_val = 1; if (tx_vboost != -1) c.s.cfg_tx_vboost_en_ovrrd_val = 1; ); /* 4) Enable the Tx coefficient request override enable */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_PCS_CTLIFC_2(qlm, lane), if (((tx_pre != -1) && (tx_post != -1)) || (tx_swing != -1)) c.s.cfg_tx_coeff_req_ovrrd_en = 1; if (tx_vboost != -1) c.s.cfg_tx_vboost_en_ovrrd_en = 1 ); /* 5) Issue a Control Interface Configuration Override request to start the Tx equalizer Optimization cycle which applies the new Tx swing and equalization settings */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_PCS_CTLIFC_2(qlm, lane), c.s.ctlifc_ovrrd_req = 1); /* 6) Prepare for a subsequent Tx swing and Tx equalization adjustment: a) Disable the Tx coefficient request override enable */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_PCS_CTLIFC_2(qlm, lane), c.s.cfg_tx_coeff_req_ovrrd_en = 0); /* b) Issue a Control Interface Configuration Override request */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_PCS_CTLIFC_2(qlm, lane), c.s.ctlifc_ovrrd_req = 1); /* The new Tx swing and Pre-cursor and Post-cursor settings will now take effect. */ return 0; } /** * Some QLM speeds need to override the default tuning parameters * * @param node Node to use in a Numa setup * @param qlm QLM to configure * @param mode Desired mode * @param baud_mhz Desired speed */ void __bdk_qlm_tune(bdk_node_t node, int qlm, bdk_qlm_modes_t mode, int baud_mhz) { /* Note: This function is not called for CCPI. For CCPI tuning, see bdk-init-nz-node.c */ /* Tuning parameters override the KR training. Don't apply them for KR links */ switch (mode) { case BDK_QLM_MODE_10G_KR_1X1: case BDK_QLM_MODE_10G_KR_2X1: case BDK_QLM_MODE_10G_KR_4X1: case BDK_QLM_MODE_40G_KR4_1X4: return; case BDK_QLM_MODE_PCIE_1X1: case BDK_QLM_MODE_PCIE_2X1: case BDK_QLM_MODE_PCIE_1X2: case BDK_QLM_MODE_PCIE_1X4: case BDK_QLM_MODE_PCIE_1X8: /* Don't tune PCIe Gen3 as it has its own builtin, similar to KR */ if (baud_mhz > 5000) return; break; default: break; } /* We're apply tuning for all lanes on this QLM */ int num_lanes = bdk_qlm_get_lanes(node, qlm); for (int lane = 0; lane < num_lanes; lane++) { /* TX Swing: First read any board specific setting from the environment */ int swing = bdk_config_get_int(BDK_CONFIG_QLM_TUNING_TX_SWING, node, qlm, lane); /* If no setting, use hard coded generic defaults */ if (swing == -1) { if (baud_mhz == 6250) { /* Email from Brendan Metzner about RXAUI around 2/7/2016 */ swing = 0x12; } else if (baud_mhz == 10312) { /* From lab measurements of EBB8800 at 10.3125G */ swing = 0xd; } } /* TX Premptap: First read any board specific setting from the environment */ int premptap = bdk_config_get_int(BDK_CONFIG_QLM_TUNING_TX_PREMPTAP, node, qlm, lane); /* If no setting, use hard coded generic defaults */ if (premptap == -1) { if (baud_mhz == 6250) { /* From lab measurements of EBB8800 at 6.25G */ premptap = 0xa0; } else if (baud_mhz == 10312) { /* From lab measurements of EBB8800 at 10.3125G */ premptap = 0xd0; } } int tx_pre = (premptap == -1) ? -1 : premptap & 0xf; int tx_post = (premptap == -1) ? -1 : premptap >> 4; int gain = bdk_config_get_int(BDK_CONFIG_QLM_TUNING_TX_GAIN, node, qlm, lane); int vboost = bdk_config_get_int(BDK_CONFIG_QLM_TUNING_TX_VBOOST, node, qlm, lane); __bdk_qlm_tune_lane_tx(node, qlm, lane, swing, tx_pre, tx_post, gain, vboost); /* Email from Brendan Metzner about RXAUI around 2/7/2016 suggested the following setting for RXAUI at 6.25G with both PHY or cable. I'm applying it to all lanes running at 6.25G */ if (baud_mhz == 6250) { /* This is changing the Q/QB error sampler 0 threshold from 0xD to 0xF */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_RX_CFG_4(qlm, lane), c.s.cfg_rx_errdet_ctrl = 0xcf6f); } } } /** * Disables DFE for the specified QLM lane(s). * This function should only be called for low-loss channels. * * @param node Node to configure * @param qlm QLM to configure * @param lane Lane to configure, or -1 for all lanes */ void __bdk_qlm_dfe_disable(int node, int qlm, int lane) { int num_lanes = bdk_qlm_get_lanes(node, qlm); int l; for (l = 0; l < num_lanes; l++) { if ((lane != -1) && (lane != l)) continue; /* 1. Write GSERX_LANEx_RX_LOOP_CTRL = 0x0270 (var "loop_ctrl" with bits 8 & 1 cleared). * bit<1> dfe_en_byp = 1'b0 */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_RX_LOOP_CTRL(qlm, l), c.s.cfg_rx_lctrl = c.s.cfg_rx_lctrl & 0x3fd); /* 2. Write GSERX_LANEx_RX_VALBBD_CTRL_1 = 0x0000 (var "ctrl1" with all bits cleared) * bits<14:11> CFG_RX_DFE_C3_MVAL = 4'b0000 * bit<10> CFG_RX_DFE_C3_MSGN = 1'b0 * bits<9:6> CFG_RX_DFE_C2_MVAL = 4'b0000 * bit<5> CFG_RX_DFE_C2_MSGN = 1'b0 * bits<4:1> CFG_RX_DFE_C1_MVAL = 5'b0000 * bits<0> CFG_RX_DFE_C1_MSGN = 1'b0 */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_RX_VALBBD_CTRL_1(qlm, l), c.s.dfe_c3_mval = 0; c.s.dfe_c3_msgn = 0; c.s.dfe_c2_mval = 0; c.s.dfe_c2_msgn = 0; c.s.dfe_c1_mval = 0; c.s.dfe_c1_msgn = 0); /* 3. Write GSERX_LANEx_RX_VALBBD_CTRL_0 = 0x2400 (var "ctrl0" with following bits set/cleared) * bits<11:10> CFG_RX_DFE_GAIN = 0x1 * bits<9:6> CFG_RX_DFE_C5_MVAL = 4'b0000 * bit<5> CFG_RX_DFE_C5_MSGN = 1'b0 * bits<4:1> CFG_RX_DFE_C4_MVAL = 4'b0000 * bit<0> CFG_RX_DFE_C4_MSGN = 1'b0 */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_RX_VALBBD_CTRL_0(qlm, l), c.s.dfe_gain = 0x1; c.s.dfe_c5_mval = 0; c.s.dfe_c5_msgn = 0; c.s.dfe_c4_mval = 0; c.s.dfe_c4_msgn = 0); /* 4. Write GSER(0..13)_LANE(0..3)_RX_VALBBD_CTRL_2 = 0x003F //enable DFE tap overrides * bit<5> dfe_ovrd_en = 1 * bit<4> dfe_c5_ovrd_val = 1 * bit<3> dfe_c4_ovrd_val = 1 * bit<2> dfe_c3_ovrd_val = 1 * bit<1> dfe_c2_ovrd_val = 1 * bit<0> dfe_c1_ovrd_val = 1 */ BDK_CSR_MODIFY(c, node, BDK_GSERX_LANEX_RX_VALBBD_CTRL_2(qlm, l), c.s.dfe_ovrd_en = 0x1; c.s.dfe_c5_ovrd_val = 0x1; c.s.dfe_c4_ovrd_val = 0x1; c.s.dfe_c3_ovrd_val = 0x1; c.s.dfe_c2_ovrd_val = 0x1; c.s.dfe_c1_ovrd_val = 0x1); } } /** * Check if a specific lane is using KR training. This is used by low level GSER * code to remember which QLMs and lanes need to support KR training for BGX. The * hardware doesn't have a bit set aside to record this, so we repurpose the * register GSERX_SCRATCH. * * @param node Node to check * @param qlm QLM to check * @param lane Lane to check * * @return True if this lane uses KR with BGX, false otherwise */ bool __bdk_qlm_is_lane_kr(bdk_node_t node, int qlm, int lane) { uint64_t mask = BDK_CSR_READ(node, BDK_GSERX_SCRATCH(qlm)); return 1 & (mask >> lane); } /** * Set if a specific lane is using KR training. This is used by low level GSER * code to remember which QLMs and lanes need to support KR training for BGX. The * hardware doesn't have a bit set aside to record this, so we repurpose the * register GSERX_SCRATCH. * * @param node Node to set * @param qlm QLM to set * @param lane Lane to set * @param is_kr KR (true) or XFI/XLAUI (false) */ void __bdk_qlm_set_lane_kr(bdk_node_t node, int qlm, int lane, bool is_kr) { uint64_t mask = BDK_CSR_READ(node, BDK_GSERX_SCRATCH(qlm)); if (is_kr) mask |= 1 << lane; else mask &= ~(1 << lane); BDK_CSR_WRITE(node, BDK_GSERX_SCRATCH(qlm), mask); }