| // SPDX-License-Identifier: GPL-2.0 |
| /* |
| * This file is based on code from OCTEON SDK by Cavium Networks. |
| * |
| * Copyright (c) 2003-2007 Cavium Networks |
| */ |
| |
| #include <linux/kernel.h> |
| #include <linux/netdevice.h> |
| #include <linux/interrupt.h> |
| #include <net/dst.h> |
| |
| #include "octeon-ethernet.h" |
| #include "ethernet-defines.h" |
| #include "ethernet-util.h" |
| |
| static int number_spi_ports; |
| static int need_retrain[2] = { 0, 0 }; |
| |
| static void cvm_oct_spxx_int_pr(union cvmx_spxx_int_reg spx_int_reg, int index) |
| { |
| if (spx_int_reg.s.spf) |
| pr_err("SPI%d: SRX Spi4 interface down\n", index); |
| if (spx_int_reg.s.calerr) |
| pr_err("SPI%d: SRX Spi4 Calendar table parity error\n", index); |
| if (spx_int_reg.s.syncerr) |
| pr_err("SPI%d: SRX Consecutive Spi4 DIP4 errors have exceeded SPX_ERR_CTL[ERRCNT]\n", |
| index); |
| if (spx_int_reg.s.diperr) |
| pr_err("SPI%d: SRX Spi4 DIP4 error\n", index); |
| if (spx_int_reg.s.tpaovr) |
| pr_err("SPI%d: SRX Selected port has hit TPA overflow\n", |
| index); |
| if (spx_int_reg.s.rsverr) |
| pr_err("SPI%d: SRX Spi4 reserved control word detected\n", |
| index); |
| if (spx_int_reg.s.drwnng) |
| pr_err("SPI%d: SRX Spi4 receive FIFO drowning/overflow\n", |
| index); |
| if (spx_int_reg.s.clserr) |
| pr_err("SPI%d: SRX Spi4 packet closed on non-16B alignment without EOP\n", |
| index); |
| if (spx_int_reg.s.spiovr) |
| pr_err("SPI%d: SRX Spi4 async FIFO overflow\n", index); |
| if (spx_int_reg.s.abnorm) |
| pr_err("SPI%d: SRX Abnormal packet termination (ERR bit)\n", |
| index); |
| if (spx_int_reg.s.prtnxa) |
| pr_err("SPI%d: SRX Port out of range\n", index); |
| } |
| |
| static void cvm_oct_stxx_int_pr(union cvmx_stxx_int_reg stx_int_reg, int index) |
| { |
| if (stx_int_reg.s.syncerr) |
| pr_err("SPI%d: STX Interface encountered a fatal error\n", |
| index); |
| if (stx_int_reg.s.frmerr) |
| pr_err("SPI%d: STX FRMCNT has exceeded STX_DIP_CNT[MAXFRM]\n", |
| index); |
| if (stx_int_reg.s.unxfrm) |
| pr_err("SPI%d: STX Unexpected framing sequence\n", index); |
| if (stx_int_reg.s.nosync) |
| pr_err("SPI%d: STX ERRCNT has exceeded STX_DIP_CNT[MAXDIP]\n", |
| index); |
| if (stx_int_reg.s.diperr) |
| pr_err("SPI%d: STX DIP2 error on the Spi4 Status channel\n", |
| index); |
| if (stx_int_reg.s.datovr) |
| pr_err("SPI%d: STX Spi4 FIFO overflow error\n", index); |
| if (stx_int_reg.s.ovrbst) |
| pr_err("SPI%d: STX Transmit packet burst too big\n", index); |
| if (stx_int_reg.s.calpar1) |
| pr_err("SPI%d: STX Calendar Table Parity Error Bank%d\n", |
| index, 1); |
| if (stx_int_reg.s.calpar0) |
| pr_err("SPI%d: STX Calendar Table Parity Error Bank%d\n", |
| index, 0); |
| } |
| |
| static irqreturn_t cvm_oct_spi_spx_int(int index) |
| { |
| union cvmx_spxx_int_reg spx_int_reg; |
| union cvmx_stxx_int_reg stx_int_reg; |
| |
| spx_int_reg.u64 = cvmx_read_csr(CVMX_SPXX_INT_REG(index)); |
| cvmx_write_csr(CVMX_SPXX_INT_REG(index), spx_int_reg.u64); |
| if (!need_retrain[index]) { |
| spx_int_reg.u64 &= cvmx_read_csr(CVMX_SPXX_INT_MSK(index)); |
| cvm_oct_spxx_int_pr(spx_int_reg, index); |
| } |
| |
| stx_int_reg.u64 = cvmx_read_csr(CVMX_STXX_INT_REG(index)); |
| cvmx_write_csr(CVMX_STXX_INT_REG(index), stx_int_reg.u64); |
| if (!need_retrain[index]) { |
| stx_int_reg.u64 &= cvmx_read_csr(CVMX_STXX_INT_MSK(index)); |
| cvm_oct_stxx_int_pr(stx_int_reg, index); |
| } |
| |
| cvmx_write_csr(CVMX_SPXX_INT_MSK(index), 0); |
| cvmx_write_csr(CVMX_STXX_INT_MSK(index), 0); |
| need_retrain[index] = 1; |
| |
| return IRQ_HANDLED; |
| } |
| |
| static irqreturn_t cvm_oct_spi_rml_interrupt(int cpl, void *dev_id) |
| { |
| irqreturn_t return_status = IRQ_NONE; |
| union cvmx_npi_rsl_int_blocks rsl_int_blocks; |
| |
| /* Check and see if this interrupt was caused by the GMX block */ |
| rsl_int_blocks.u64 = cvmx_read_csr(CVMX_NPI_RSL_INT_BLOCKS); |
| if (rsl_int_blocks.s.spx1) /* 19 - SPX1_INT_REG & STX1_INT_REG */ |
| return_status = cvm_oct_spi_spx_int(1); |
| |
| if (rsl_int_blocks.s.spx0) /* 18 - SPX0_INT_REG & STX0_INT_REG */ |
| return_status = cvm_oct_spi_spx_int(0); |
| |
| return return_status; |
| } |
| |
| static void cvm_oct_spi_enable_error_reporting(int interface) |
| { |
| union cvmx_spxx_int_msk spxx_int_msk; |
| union cvmx_stxx_int_msk stxx_int_msk; |
| |
| spxx_int_msk.u64 = cvmx_read_csr(CVMX_SPXX_INT_MSK(interface)); |
| spxx_int_msk.s.calerr = 1; |
| spxx_int_msk.s.syncerr = 1; |
| spxx_int_msk.s.diperr = 1; |
| spxx_int_msk.s.tpaovr = 1; |
| spxx_int_msk.s.rsverr = 1; |
| spxx_int_msk.s.drwnng = 1; |
| spxx_int_msk.s.clserr = 1; |
| spxx_int_msk.s.spiovr = 1; |
| spxx_int_msk.s.abnorm = 1; |
| spxx_int_msk.s.prtnxa = 1; |
| cvmx_write_csr(CVMX_SPXX_INT_MSK(interface), spxx_int_msk.u64); |
| |
| stxx_int_msk.u64 = cvmx_read_csr(CVMX_STXX_INT_MSK(interface)); |
| stxx_int_msk.s.frmerr = 1; |
| stxx_int_msk.s.unxfrm = 1; |
| stxx_int_msk.s.nosync = 1; |
| stxx_int_msk.s.diperr = 1; |
| stxx_int_msk.s.datovr = 1; |
| stxx_int_msk.s.ovrbst = 1; |
| stxx_int_msk.s.calpar1 = 1; |
| stxx_int_msk.s.calpar0 = 1; |
| cvmx_write_csr(CVMX_STXX_INT_MSK(interface), stxx_int_msk.u64); |
| } |
| |
| static void cvm_oct_spi_poll(struct net_device *dev) |
| { |
| static int spi4000_port; |
| struct octeon_ethernet *priv = netdev_priv(dev); |
| int interface; |
| |
| for (interface = 0; interface < 2; interface++) { |
| if ((priv->port == interface * 16) && need_retrain[interface]) { |
| if (cvmx_spi_restart_interface |
| (interface, CVMX_SPI_MODE_DUPLEX, 10) == 0) { |
| need_retrain[interface] = 0; |
| cvm_oct_spi_enable_error_reporting(interface); |
| } |
| } |
| |
| /* |
| * The SPI4000 TWSI interface is very slow. In order |
| * not to bring the system to a crawl, we only poll a |
| * single port every second. This means negotiation |
| * speed changes take up to 10 seconds, but at least |
| * we don't waste absurd amounts of time waiting for |
| * TWSI. |
| */ |
| if (priv->port == spi4000_port) { |
| /* |
| * This function does nothing if it is called on an |
| * interface without a SPI4000. |
| */ |
| cvmx_spi4000_check_speed(interface, priv->port); |
| /* |
| * Normal ordering increments. By decrementing |
| * we only match once per iteration. |
| */ |
| spi4000_port--; |
| if (spi4000_port < 0) |
| spi4000_port = 10; |
| } |
| } |
| } |
| |
| int cvm_oct_spi_init(struct net_device *dev) |
| { |
| int r; |
| struct octeon_ethernet *priv = netdev_priv(dev); |
| |
| if (number_spi_ports == 0) { |
| r = request_irq(OCTEON_IRQ_RML, cvm_oct_spi_rml_interrupt, |
| IRQF_SHARED, "SPI", &number_spi_ports); |
| if (r) |
| return r; |
| } |
| number_spi_ports++; |
| |
| if (priv->port == 0 || priv->port == 16) { |
| cvm_oct_spi_enable_error_reporting(INTERFACE(priv->port)); |
| priv->poll = cvm_oct_spi_poll; |
| } |
| cvm_oct_common_init(dev); |
| return 0; |
| } |
| |
| void cvm_oct_spi_uninit(struct net_device *dev) |
| { |
| int interface; |
| |
| cvm_oct_common_uninit(dev); |
| number_spi_ports--; |
| if (number_spi_ports == 0) { |
| for (interface = 0; interface < 2; interface++) { |
| cvmx_write_csr(CVMX_SPXX_INT_MSK(interface), 0); |
| cvmx_write_csr(CVMX_STXX_INT_MSK(interface), 0); |
| } |
| free_irq(OCTEON_IRQ_RML, &number_spi_ports); |
| } |
| } |