1 // SPDX-License-Identifier: GPL-2.0-only OR BSD-3-Clause 2 3 /* Gigabit Ethernet driver for Mellanox BlueField SoC 4 * 5 * Copyright (C) 2020-2021 NVIDIA CORPORATION & AFFILIATES 6 */ 7 8 #include <linux/acpi.h> 9 #include <linux/device.h> 10 #include <linux/dma-mapping.h> 11 #include <linux/etherdevice.h> 12 #include <linux/interrupt.h> 13 #include <linux/iopoll.h> 14 #include <linux/module.h> 15 #include <linux/phy.h> 16 #include <linux/platform_device.h> 17 #include <linux/rtnetlink.h> 18 #include <linux/skbuff.h> 19 20 #include "mlxbf_gige.h" 21 #include "mlxbf_gige_regs.h" 22 23 /* Allocate SKB whose payload pointer aligns with the Bluefield 24 * hardware DMA limitation, i.e. DMA operation can't cross 25 * a 4KB boundary. A maximum packet size of 2KB is assumed in the 26 * alignment formula. The alignment logic overallocates an SKB, 27 * and then adjusts the headroom so that the SKB data pointer is 28 * naturally aligned to a 2KB boundary. 29 */ 30 struct sk_buff *mlxbf_gige_alloc_skb(struct mlxbf_gige *priv, 31 unsigned int map_len, 32 dma_addr_t *buf_dma, 33 enum dma_data_direction dir) 34 { 35 struct sk_buff *skb; 36 u64 addr, offset; 37 38 /* Overallocate the SKB so that any headroom adjustment (to 39 * provide 2KB natural alignment) does not exceed payload area 40 */ 41 skb = netdev_alloc_skb(priv->netdev, MLXBF_GIGE_DEFAULT_BUF_SZ * 2); 42 if (!skb) 43 return NULL; 44 45 /* Adjust the headroom so that skb->data is naturally aligned to 46 * a 2KB boundary, which is the maximum packet size supported. 47 */ 48 addr = (long)skb->data; 49 offset = (addr + MLXBF_GIGE_DEFAULT_BUF_SZ - 1) & 50 ~(MLXBF_GIGE_DEFAULT_BUF_SZ - 1); 51 offset -= addr; 52 if (offset) 53 skb_reserve(skb, offset); 54 55 /* Return streaming DMA mapping to caller */ 56 *buf_dma = dma_map_single(priv->dev, skb->data, map_len, dir); 57 if (dma_mapping_error(priv->dev, *buf_dma)) { 58 dev_kfree_skb(skb); 59 *buf_dma = (dma_addr_t)0; 60 return NULL; 61 } 62 63 return skb; 64 } 65 66 static void mlxbf_gige_initial_mac(struct mlxbf_gige *priv) 67 { 68 u8 mac[ETH_ALEN]; 69 u64 local_mac; 70 71 eth_zero_addr(mac); 72 mlxbf_gige_get_mac_rx_filter(priv, MLXBF_GIGE_LOCAL_MAC_FILTER_IDX, 73 &local_mac); 74 u64_to_ether_addr(local_mac, mac); 75 76 if (is_valid_ether_addr(mac)) { 77 eth_hw_addr_set(priv->netdev, mac); 78 } else { 79 /* Provide a random MAC if for some reason the device has 80 * not been configured with a valid MAC address already. 81 */ 82 eth_hw_addr_random(priv->netdev); 83 } 84 85 local_mac = ether_addr_to_u64(priv->netdev->dev_addr); 86 mlxbf_gige_set_mac_rx_filter(priv, MLXBF_GIGE_LOCAL_MAC_FILTER_IDX, 87 local_mac); 88 } 89 90 static void mlxbf_gige_cache_stats(struct mlxbf_gige *priv) 91 { 92 struct mlxbf_gige_stats *p; 93 94 /* Cache stats that will be cleared by clean port operation */ 95 p = &priv->stats; 96 p->rx_din_dropped_pkts += readq(priv->base + 97 MLXBF_GIGE_RX_DIN_DROP_COUNTER); 98 p->rx_filter_passed_pkts += readq(priv->base + 99 MLXBF_GIGE_RX_PASS_COUNTER_ALL); 100 p->rx_filter_discard_pkts += readq(priv->base + 101 MLXBF_GIGE_RX_DISC_COUNTER_ALL); 102 } 103 104 static int mlxbf_gige_clean_port(struct mlxbf_gige *priv) 105 { 106 u64 control; 107 u64 temp; 108 int err; 109 110 /* Set the CLEAN_PORT_EN bit to trigger SW reset */ 111 control = readq(priv->base + MLXBF_GIGE_CONTROL); 112 control |= MLXBF_GIGE_CONTROL_CLEAN_PORT_EN; 113 writeq(control, priv->base + MLXBF_GIGE_CONTROL); 114 115 /* Ensure completion of "clean port" write before polling status */ 116 mb(); 117 118 err = readq_poll_timeout_atomic(priv->base + MLXBF_GIGE_STATUS, temp, 119 (temp & MLXBF_GIGE_STATUS_READY), 120 100, 100000); 121 122 /* Clear the CLEAN_PORT_EN bit at end of this loop */ 123 control = readq(priv->base + MLXBF_GIGE_CONTROL); 124 control &= ~MLXBF_GIGE_CONTROL_CLEAN_PORT_EN; 125 writeq(control, priv->base + MLXBF_GIGE_CONTROL); 126 127 return err; 128 } 129 130 static int mlxbf_gige_open(struct net_device *netdev) 131 { 132 struct mlxbf_gige *priv = netdev_priv(netdev); 133 struct phy_device *phydev = netdev->phydev; 134 u64 control; 135 u64 int_en; 136 int err; 137 138 /* Perform general init of GigE block */ 139 control = readq(priv->base + MLXBF_GIGE_CONTROL); 140 control |= MLXBF_GIGE_CONTROL_PORT_EN; 141 writeq(control, priv->base + MLXBF_GIGE_CONTROL); 142 143 mlxbf_gige_cache_stats(priv); 144 err = mlxbf_gige_clean_port(priv); 145 if (err) 146 return err; 147 148 /* Clear driver's valid_polarity to match hardware, 149 * since the above call to clean_port() resets the 150 * receive polarity used by hardware. 151 */ 152 priv->valid_polarity = 0; 153 154 phy_start(phydev); 155 156 err = mlxbf_gige_tx_init(priv); 157 if (err) 158 goto phy_deinit; 159 err = mlxbf_gige_rx_init(priv); 160 if (err) 161 goto tx_deinit; 162 163 netif_napi_add(netdev, &priv->napi, mlxbf_gige_poll); 164 napi_enable(&priv->napi); 165 netif_start_queue(netdev); 166 167 err = mlxbf_gige_request_irqs(priv); 168 if (err) 169 goto napi_deinit; 170 171 mlxbf_gige_enable_mac_rx_filter(priv, MLXBF_GIGE_BCAST_MAC_FILTER_IDX); 172 mlxbf_gige_enable_mac_rx_filter(priv, MLXBF_GIGE_LOCAL_MAC_FILTER_IDX); 173 mlxbf_gige_enable_multicast_rx(priv); 174 175 /* Set bits in INT_EN that we care about */ 176 int_en = MLXBF_GIGE_INT_EN_HW_ACCESS_ERROR | 177 MLXBF_GIGE_INT_EN_TX_CHECKSUM_INPUTS | 178 MLXBF_GIGE_INT_EN_TX_SMALL_FRAME_SIZE | 179 MLXBF_GIGE_INT_EN_TX_PI_CI_EXCEED_WQ_SIZE | 180 MLXBF_GIGE_INT_EN_SW_CONFIG_ERROR | 181 MLXBF_GIGE_INT_EN_SW_ACCESS_ERROR | 182 MLXBF_GIGE_INT_EN_RX_RECEIVE_PACKET; 183 184 /* Ensure completion of all initialization before enabling interrupts */ 185 mb(); 186 187 writeq(int_en, priv->base + MLXBF_GIGE_INT_EN); 188 189 return 0; 190 191 napi_deinit: 192 netif_stop_queue(netdev); 193 napi_disable(&priv->napi); 194 netif_napi_del(&priv->napi); 195 mlxbf_gige_rx_deinit(priv); 196 197 tx_deinit: 198 mlxbf_gige_tx_deinit(priv); 199 200 phy_deinit: 201 phy_stop(phydev); 202 return err; 203 } 204 205 static int mlxbf_gige_stop(struct net_device *netdev) 206 { 207 struct mlxbf_gige *priv = netdev_priv(netdev); 208 209 writeq(0, priv->base + MLXBF_GIGE_INT_EN); 210 netif_stop_queue(netdev); 211 napi_disable(&priv->napi); 212 netif_napi_del(&priv->napi); 213 mlxbf_gige_free_irqs(priv); 214 215 phy_stop(netdev->phydev); 216 217 mlxbf_gige_rx_deinit(priv); 218 mlxbf_gige_tx_deinit(priv); 219 mlxbf_gige_cache_stats(priv); 220 mlxbf_gige_clean_port(priv); 221 222 return 0; 223 } 224 225 static int mlxbf_gige_eth_ioctl(struct net_device *netdev, 226 struct ifreq *ifr, int cmd) 227 { 228 if (!(netif_running(netdev))) 229 return -EINVAL; 230 231 return phy_mii_ioctl(netdev->phydev, ifr, cmd); 232 } 233 234 static void mlxbf_gige_set_rx_mode(struct net_device *netdev) 235 { 236 struct mlxbf_gige *priv = netdev_priv(netdev); 237 bool new_promisc_enabled; 238 239 new_promisc_enabled = netdev->flags & IFF_PROMISC; 240 241 /* Only write to the hardware registers if the new setting 242 * of promiscuous mode is different from the current one. 243 */ 244 if (new_promisc_enabled != priv->promisc_enabled) { 245 priv->promisc_enabled = new_promisc_enabled; 246 247 if (new_promisc_enabled) 248 mlxbf_gige_enable_promisc(priv); 249 else 250 mlxbf_gige_disable_promisc(priv); 251 } 252 } 253 254 static void mlxbf_gige_get_stats64(struct net_device *netdev, 255 struct rtnl_link_stats64 *stats) 256 { 257 struct mlxbf_gige *priv = netdev_priv(netdev); 258 259 netdev_stats_to_stats64(stats, &netdev->stats); 260 261 stats->rx_length_errors = priv->stats.rx_truncate_errors; 262 stats->rx_fifo_errors = priv->stats.rx_din_dropped_pkts + 263 readq(priv->base + MLXBF_GIGE_RX_DIN_DROP_COUNTER); 264 stats->rx_crc_errors = priv->stats.rx_mac_errors; 265 stats->rx_errors = stats->rx_length_errors + 266 stats->rx_fifo_errors + 267 stats->rx_crc_errors; 268 269 stats->tx_fifo_errors = priv->stats.tx_fifo_full; 270 stats->tx_errors = stats->tx_fifo_errors; 271 } 272 273 static const struct net_device_ops mlxbf_gige_netdev_ops = { 274 .ndo_open = mlxbf_gige_open, 275 .ndo_stop = mlxbf_gige_stop, 276 .ndo_start_xmit = mlxbf_gige_start_xmit, 277 .ndo_set_mac_address = eth_mac_addr, 278 .ndo_validate_addr = eth_validate_addr, 279 .ndo_eth_ioctl = mlxbf_gige_eth_ioctl, 280 .ndo_set_rx_mode = mlxbf_gige_set_rx_mode, 281 .ndo_get_stats64 = mlxbf_gige_get_stats64, 282 }; 283 284 static void mlxbf_gige_bf2_adjust_link(struct net_device *netdev) 285 { 286 struct phy_device *phydev = netdev->phydev; 287 288 phy_print_status(phydev); 289 } 290 291 static void mlxbf_gige_bf3_adjust_link(struct net_device *netdev) 292 { 293 struct mlxbf_gige *priv = netdev_priv(netdev); 294 struct phy_device *phydev = netdev->phydev; 295 u8 sgmii_mode; 296 u16 ipg_size; 297 u32 val; 298 299 if (phydev->link && phydev->speed != priv->prev_speed) { 300 switch (phydev->speed) { 301 case 1000: 302 ipg_size = MLXBF_GIGE_1G_IPG_SIZE; 303 sgmii_mode = MLXBF_GIGE_1G_SGMII_MODE; 304 break; 305 case 100: 306 ipg_size = MLXBF_GIGE_100M_IPG_SIZE; 307 sgmii_mode = MLXBF_GIGE_100M_SGMII_MODE; 308 break; 309 case 10: 310 ipg_size = MLXBF_GIGE_10M_IPG_SIZE; 311 sgmii_mode = MLXBF_GIGE_10M_SGMII_MODE; 312 break; 313 default: 314 return; 315 } 316 317 val = readl(priv->plu_base + MLXBF_GIGE_PLU_TX_REG0); 318 val &= ~(MLXBF_GIGE_PLU_TX_IPG_SIZE_MASK | MLXBF_GIGE_PLU_TX_SGMII_MODE_MASK); 319 val |= FIELD_PREP(MLXBF_GIGE_PLU_TX_IPG_SIZE_MASK, ipg_size); 320 val |= FIELD_PREP(MLXBF_GIGE_PLU_TX_SGMII_MODE_MASK, sgmii_mode); 321 writel(val, priv->plu_base + MLXBF_GIGE_PLU_TX_REG0); 322 323 val = readl(priv->plu_base + MLXBF_GIGE_PLU_RX_REG0); 324 val &= ~MLXBF_GIGE_PLU_RX_SGMII_MODE_MASK; 325 val |= FIELD_PREP(MLXBF_GIGE_PLU_RX_SGMII_MODE_MASK, sgmii_mode); 326 writel(val, priv->plu_base + MLXBF_GIGE_PLU_RX_REG0); 327 328 priv->prev_speed = phydev->speed; 329 } 330 331 phy_print_status(phydev); 332 } 333 334 static void mlxbf_gige_bf2_set_phy_link_mode(struct phy_device *phydev) 335 { 336 /* MAC only supports 1000T full duplex mode */ 337 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_1000baseT_Half_BIT); 338 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_100baseT_Full_BIT); 339 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_100baseT_Half_BIT); 340 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Full_BIT); 341 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Half_BIT); 342 343 /* Only symmetric pause with flow control enabled is supported so no 344 * need to negotiate pause. 345 */ 346 linkmode_clear_bit(ETHTOOL_LINK_MODE_Pause_BIT, phydev->advertising); 347 linkmode_clear_bit(ETHTOOL_LINK_MODE_Asym_Pause_BIT, phydev->advertising); 348 } 349 350 static void mlxbf_gige_bf3_set_phy_link_mode(struct phy_device *phydev) 351 { 352 /* MAC only supports full duplex mode */ 353 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_1000baseT_Half_BIT); 354 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_100baseT_Half_BIT); 355 phy_remove_link_mode(phydev, ETHTOOL_LINK_MODE_10baseT_Half_BIT); 356 357 /* Only symmetric pause with flow control enabled is supported so no 358 * need to negotiate pause. 359 */ 360 linkmode_clear_bit(ETHTOOL_LINK_MODE_Pause_BIT, phydev->advertising); 361 linkmode_clear_bit(ETHTOOL_LINK_MODE_Asym_Pause_BIT, phydev->advertising); 362 } 363 364 static struct mlxbf_gige_link_cfg mlxbf_gige_link_cfgs[] = { 365 [MLXBF_GIGE_VERSION_BF2] = { 366 .set_phy_link_mode = mlxbf_gige_bf2_set_phy_link_mode, 367 .adjust_link = mlxbf_gige_bf2_adjust_link, 368 .phy_mode = PHY_INTERFACE_MODE_GMII 369 }, 370 [MLXBF_GIGE_VERSION_BF3] = { 371 .set_phy_link_mode = mlxbf_gige_bf3_set_phy_link_mode, 372 .adjust_link = mlxbf_gige_bf3_adjust_link, 373 .phy_mode = PHY_INTERFACE_MODE_SGMII 374 } 375 }; 376 377 static int mlxbf_gige_probe(struct platform_device *pdev) 378 { 379 struct phy_device *phydev; 380 struct net_device *netdev; 381 struct mlxbf_gige *priv; 382 void __iomem *llu_base; 383 void __iomem *plu_base; 384 void __iomem *base; 385 int addr, phy_irq; 386 unsigned int i; 387 int err; 388 389 base = devm_platform_ioremap_resource(pdev, MLXBF_GIGE_RES_MAC); 390 if (IS_ERR(base)) 391 return PTR_ERR(base); 392 393 llu_base = devm_platform_ioremap_resource(pdev, MLXBF_GIGE_RES_LLU); 394 if (IS_ERR(llu_base)) 395 return PTR_ERR(llu_base); 396 397 plu_base = devm_platform_ioremap_resource(pdev, MLXBF_GIGE_RES_PLU); 398 if (IS_ERR(plu_base)) 399 return PTR_ERR(plu_base); 400 401 netdev = devm_alloc_etherdev(&pdev->dev, sizeof(*priv)); 402 if (!netdev) 403 return -ENOMEM; 404 405 SET_NETDEV_DEV(netdev, &pdev->dev); 406 netdev->netdev_ops = &mlxbf_gige_netdev_ops; 407 netdev->ethtool_ops = &mlxbf_gige_ethtool_ops; 408 priv = netdev_priv(netdev); 409 priv->netdev = netdev; 410 411 platform_set_drvdata(pdev, priv); 412 priv->dev = &pdev->dev; 413 priv->pdev = pdev; 414 415 spin_lock_init(&priv->lock); 416 417 priv->hw_version = readq(base + MLXBF_GIGE_VERSION); 418 419 /* Attach MDIO device */ 420 err = mlxbf_gige_mdio_probe(pdev, priv); 421 if (err) 422 return err; 423 424 priv->base = base; 425 priv->llu_base = llu_base; 426 priv->plu_base = plu_base; 427 428 priv->rx_q_entries = MLXBF_GIGE_DEFAULT_RXQ_SZ; 429 priv->tx_q_entries = MLXBF_GIGE_DEFAULT_TXQ_SZ; 430 431 for (i = 0; i <= MLXBF_GIGE_MAX_FILTER_IDX; i++) 432 mlxbf_gige_disable_mac_rx_filter(priv, i); 433 mlxbf_gige_disable_multicast_rx(priv); 434 mlxbf_gige_disable_promisc(priv); 435 436 /* Write initial MAC address to hardware */ 437 mlxbf_gige_initial_mac(priv); 438 439 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 440 if (err) { 441 dev_err(&pdev->dev, "DMA configuration failed: 0x%x\n", err); 442 goto out; 443 } 444 445 priv->error_irq = platform_get_irq(pdev, MLXBF_GIGE_ERROR_INTR_IDX); 446 priv->rx_irq = platform_get_irq(pdev, MLXBF_GIGE_RECEIVE_PKT_INTR_IDX); 447 priv->llu_plu_irq = platform_get_irq(pdev, MLXBF_GIGE_LLU_PLU_INTR_IDX); 448 449 phy_irq = acpi_dev_gpio_irq_get_by(ACPI_COMPANION(&pdev->dev), "phy-gpios", 0); 450 if (phy_irq < 0) { 451 dev_err(&pdev->dev, "Error getting PHY irq. Use polling instead"); 452 phy_irq = PHY_POLL; 453 } 454 455 phydev = phy_find_first(priv->mdiobus); 456 if (!phydev) { 457 err = -ENODEV; 458 goto out; 459 } 460 461 addr = phydev->mdio.addr; 462 priv->mdiobus->irq[addr] = phy_irq; 463 phydev->irq = phy_irq; 464 465 err = phy_connect_direct(netdev, phydev, 466 mlxbf_gige_link_cfgs[priv->hw_version].adjust_link, 467 mlxbf_gige_link_cfgs[priv->hw_version].phy_mode); 468 if (err) { 469 dev_err(&pdev->dev, "Could not attach to PHY\n"); 470 goto out; 471 } 472 473 mlxbf_gige_link_cfgs[priv->hw_version].set_phy_link_mode(phydev); 474 475 /* Display information about attached PHY device */ 476 phy_attached_info(phydev); 477 478 err = register_netdev(netdev); 479 if (err) { 480 dev_err(&pdev->dev, "Failed to register netdev\n"); 481 phy_disconnect(phydev); 482 goto out; 483 } 484 485 return 0; 486 487 out: 488 mlxbf_gige_mdio_remove(priv); 489 return err; 490 } 491 492 static int mlxbf_gige_remove(struct platform_device *pdev) 493 { 494 struct mlxbf_gige *priv = platform_get_drvdata(pdev); 495 496 unregister_netdev(priv->netdev); 497 phy_disconnect(priv->netdev->phydev); 498 mlxbf_gige_mdio_remove(priv); 499 platform_set_drvdata(pdev, NULL); 500 501 return 0; 502 } 503 504 static void mlxbf_gige_shutdown(struct platform_device *pdev) 505 { 506 struct mlxbf_gige *priv = platform_get_drvdata(pdev); 507 508 rtnl_lock(); 509 netif_device_detach(priv->netdev); 510 511 if (netif_running(priv->netdev)) 512 dev_close(priv->netdev); 513 514 rtnl_unlock(); 515 } 516 517 static const struct acpi_device_id __maybe_unused mlxbf_gige_acpi_match[] = { 518 { "MLNXBF17", 0 }, 519 {}, 520 }; 521 MODULE_DEVICE_TABLE(acpi, mlxbf_gige_acpi_match); 522 523 static struct platform_driver mlxbf_gige_driver = { 524 .probe = mlxbf_gige_probe, 525 .remove = mlxbf_gige_remove, 526 .shutdown = mlxbf_gige_shutdown, 527 .driver = { 528 .name = KBUILD_MODNAME, 529 .acpi_match_table = ACPI_PTR(mlxbf_gige_acpi_match), 530 }, 531 }; 532 533 module_platform_driver(mlxbf_gige_driver); 534 535 MODULE_DESCRIPTION("Mellanox BlueField SoC Gigabit Ethernet Driver"); 536 MODULE_AUTHOR("David Thompson <davthompson@nvidia.com>"); 537 MODULE_AUTHOR("Asmaa Mnebhi <asmaa@nvidia.com>"); 538 MODULE_LICENSE("Dual BSD/GPL"); 539