/** * @file kernel/net/e1000.c * @brief Intel Gigabit Ethernet device driver * * @copyright * This file is part of ToaruOS and is released under the terms * of the NCSA / University of Illinois License - see LICENSE.md * Copyright (C) 2017-2021 K. Lange * * @ref https://www.intel.com/content/dam/www/public/us/en/documents/manuals/pcie-gbe-controllers-open-source-manual.pdf */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define INTS (ICR_LSC | ICR_RXO | ICR_RXT0 | ICR_TXQE | ICR_TXDW | ICR_ACK | ICR_RXDMT0 | ICR_SRPD) struct e1000_nic { struct EthernetDevice eth; uint32_t pci_device; uint16_t deviceid; uintptr_t mmio_addr; int irq_number; int has_eeprom; int rx_index; int tx_index; int link_status; spin_lock_t tx_lock; uint8_t * rx_virt[E1000_NUM_RX_DESC]; uint8_t * tx_virt[E1000_NUM_TX_DESC]; struct e1000_rx_desc * rx; struct e1000_tx_desc * tx; uintptr_t rx_phys; uintptr_t tx_phys; int configured; process_t * queuer; process_t * processor; netif_counters_t counts; }; static int device_count = 0; static struct e1000_nic * devices[32] = {NULL}; static uint32_t mmio_read32(uintptr_t addr) { return *((volatile uint32_t*)(addr)); } static void mmio_write32(uintptr_t addr, uint32_t val) { (*((volatile uint32_t*)(addr))) = val; } static void write_command(struct e1000_nic * device, uint16_t addr, uint32_t val) { mmio_write32(device->mmio_addr + addr, val); } static uint32_t read_command(struct e1000_nic * device, uint16_t addr) { return mmio_read32(device->mmio_addr + addr); } static void delay_yield(size_t subticks) { unsigned long s, ss; relative_time(0, subticks, &s, &ss); sleep_until((process_t *)this_core->current_process, s, ss); switch_task(0); } static int eeprom_detect(struct e1000_nic * device) { /* Definitely not */ if (device->deviceid == 0x10d3) return 0; write_command(device, E1000_REG_EEPROM, 1); for (int i = 0; i < 10000 && !device->has_eeprom; ++i) { uint32_t val = read_command(device, E1000_REG_EEPROM); if (val & 0x10) device->has_eeprom = 1; } return 0; } static uint16_t eeprom_read(struct e1000_nic * device, uint8_t addr) { uint32_t temp = 0; write_command(device, E1000_REG_EEPROM, 1 | ((uint32_t)(addr) << 8)); while (!((temp = read_command(device, E1000_REG_EEPROM)) & (1 << 4))); return (uint16_t)((temp >> 16) & 0xFFFF); } static void write_mac(struct e1000_nic * device) { uint32_t low, high; memcpy(&low, &device->eth.mac[0], 4); memcpy(&high,&device->eth.mac[4], 2); memset((uint8_t *)&high + 2, 0, 2); high |= 0x80000000; write_command(device, E1000_REG_RXADDR + 0, low); write_command(device, E1000_REG_RXADDR + 4, high); } static void read_mac(struct e1000_nic * device) { if (device->has_eeprom) { uint32_t t; t = eeprom_read(device, 0); device->eth.mac[0] = t & 0xFF; device->eth.mac[1] = t >> 8; t = eeprom_read(device, 1); device->eth.mac[2] = t & 0xFF; device->eth.mac[3] = t >> 8; t = eeprom_read(device, 2); device->eth.mac[4] = t & 0xFF; device->eth.mac[5] = t >> 8; } else { uint32_t mac_addr_low = *(uint32_t *)(device->mmio_addr + E1000_REG_RXADDR); uint32_t mac_addr_high = *(uint32_t *)(device->mmio_addr + E1000_REG_RXADDR + 4); device->eth.mac[0] = (mac_addr_low >> 0 ) & 0xFF; device->eth.mac[1] = (mac_addr_low >> 8 ) & 0xFF; device->eth.mac[2] = (mac_addr_low >> 16) & 0xFF; device->eth.mac[3] = (mac_addr_low >> 24) & 0xFF; device->eth.mac[4] = (mac_addr_high>> 0 ) & 0xFF; device->eth.mac[5] = (mac_addr_high>> 8 ) & 0xFF; } } static void e1000_handle(struct e1000_nic * nic, uint32_t status) { write_command(nic, E1000_REG_ICR, status); if (!nic->configured) { return; } if (status & ICR_LSC) { nic->link_status= (read_command(nic, E1000_REG_STATUS) & (1 << 1)); } make_process_ready(nic->queuer); } static void e1000_queuer(void * data) { struct e1000_nic * nic = data; int head = read_command(nic, E1000_REG_RXDESCHEAD); int budget = 8; while (1) { int processed = 0; if (head == nic->rx_index) { head = read_command(nic, E1000_REG_RXDESCHEAD); } if (head != nic->rx_index) { while ((nic->rx[nic->rx_index].status & 0x01) && (processed < budget)) { int i = nic->rx_index; if (!(nic->rx[i].errors & (0x97))) { nic->counts.rx_count++; nic->counts.rx_bytes += nic->rx[i].length; net_eth_handle((void*)nic->rx_virt[i], nic->eth.device_node, nic->rx[i].length); } else { printf("error bits set in packet: %x\n", nic->rx[i].errors); } processed++; nic->rx[i].status = 0; if (++nic->rx_index == E1000_NUM_RX_DESC) { nic->rx_index = 0; } if (nic->rx_index == head) { head = read_command(nic, E1000_REG_RXDESCHEAD); if (nic->rx_index == head) break; } write_command(nic, E1000_REG_RXDESCTAIL, nic->rx_index); read_command(nic, E1000_REG_STATUS); } } if (processed == 0) { switch_task(0); } else { if (this_core->cpu_id == 0) switch_task(0); } } } static int irq_handler(struct regs *r) { int irq = r->int_no - 32; int handled = 0; for (int i = 0; i < device_count; ++i) { if (devices[i]->irq_number == irq) { uint32_t status = read_command(devices[i], E1000_REG_ICR); if (status) { e1000_handle(devices[i], status); if (!handled) { handled = 1; irq_ack(irq); } } } } return handled; } static int tx_full(struct e1000_nic * device, int tx_tail, int tx_head) { if (tx_tail == tx_head) return 0; if (device->tx_index == tx_head) return 1; if (((device->tx_index + 1) & E1000_NUM_TX_DESC) == tx_head) return 1; return 0; } static void send_packet(struct e1000_nic * device, uint8_t* payload, size_t payload_size) { spin_lock(device->tx_lock); int tx_tail = read_command(device, E1000_REG_TXDESCTAIL); int tx_head = read_command(device, E1000_REG_TXDESCHEAD); if (tx_full(device, tx_tail, tx_head)) { int timeout = 1000; do { spin_unlock(device->tx_lock); delay_yield(10000); timeout--; if (timeout == 0) { printf("e1000: wait for tx timed out, giving up\n"); return; } spin_lock(device->tx_lock); tx_tail = read_command(device, E1000_REG_TXDESCTAIL); tx_head = read_command(device, E1000_REG_TXDESCHEAD); } while (tx_full(device, tx_tail, tx_head)); } memcpy(device->tx_virt[device->tx_index], payload, payload_size); device->tx[device->tx_index].length = payload_size; device->tx[device->tx_index].cmd = CMD_EOP | CMD_IFCS | CMD_RS; //| CMD_RPS; device->tx[device->tx_index].status = 0; device->counts.tx_count++; device->counts.tx_bytes += payload_size; if (++device->tx_index == E1000_NUM_TX_DESC) { device->tx_index = 0; } write_command(device, E1000_REG_TXDESCTAIL, device->tx_index); read_command(device, E1000_REG_STATUS); spin_unlock(device->tx_lock); } static void init_rx(struct e1000_nic * device) { write_command(device, E1000_REG_RXDESCLO, device->rx_phys); write_command(device, E1000_REG_RXDESCHI, 0); write_command(device, E1000_REG_RXDESCLEN, E1000_NUM_RX_DESC * sizeof(struct e1000_rx_desc)); write_command(device, E1000_REG_RXDESCHEAD, 0); write_command(device, E1000_REG_RXDESCTAIL, E1000_NUM_RX_DESC - 1); device->rx_index = 0; write_command(device, E1000_REG_RCTRL, RCTL_EN | (1 << 2) | /* store bad packets */ (1 << 4) | /* multicast promiscuous */ (1 << 15) | /* broadcast accept */ (1 << 25) | /* Extended size... */ (3 << 16) | /* 4096 */ (1 << 26) /* strip CRC */ ); } static void init_tx(struct e1000_nic * device) { write_command(device, E1000_REG_TXDESCLO, device->tx_phys); write_command(device, E1000_REG_TXDESCHI, 0); write_command(device, E1000_REG_TXDESCLEN, E1000_NUM_TX_DESC * sizeof(struct e1000_tx_desc)); write_command(device, E1000_REG_TXDESCHEAD, 0); write_command(device, E1000_REG_TXDESCTAIL, 0); device->tx_index = 0; uint32_t tctl = read_command(device, E1000_REG_TCTRL); /* Collision threshold */ tctl &= ~(0xFF << 4); tctl |= (15 << 4); /* Turn it on */ tctl |= TCTL_EN; tctl |= TCTL_PSP; tctl |= (1 << 24); /* retransmit on late collision */ write_command(device, E1000_REG_TCTRL, tctl); } static int ioctl_e1000(fs_node_t * node, unsigned long request, void * argp) { struct e1000_nic * nic = node->device; switch (request) { case SIOCGIFHWADDR: /* fill argp with mac */ memcpy(argp, nic->eth.mac, 6); return 0; case SIOCGIFADDR: if (nic->eth.ipv4_addr == 0) return -ENOENT; memcpy(argp, &nic->eth.ipv4_addr, sizeof(nic->eth.ipv4_addr)); return 0; case SIOCSIFADDR: memcpy(&nic->eth.ipv4_addr, argp, sizeof(nic->eth.ipv4_addr)); return 0; case SIOCGIFNETMASK: if (nic->eth.ipv4_subnet == 0) return -ENOENT; memcpy(argp, &nic->eth.ipv4_subnet, sizeof(nic->eth.ipv4_subnet)); return 0; case SIOCSIFNETMASK: memcpy(&nic->eth.ipv4_subnet, argp, sizeof(nic->eth.ipv4_subnet)); return 0; case SIOCGIFGATEWAY: if (nic->eth.ipv4_subnet == 0) return -ENOENT; memcpy(argp, &nic->eth.ipv4_gateway, sizeof(nic->eth.ipv4_gateway)); return 0; case SIOCSIFGATEWAY: memcpy(&nic->eth.ipv4_gateway, argp, sizeof(nic->eth.ipv4_gateway)); net_arp_ask(nic->eth.ipv4_gateway, node); return 0; case SIOCGIFADDR6: return -ENOENT; case SIOCSIFADDR6: memcpy(&nic->eth.ipv6_addr, argp, sizeof(nic->eth.ipv6_addr)); return 0; case SIOCGIFFLAGS: { uint32_t * flags = argp; *flags = IFF_RUNNING; if (nic->link_status) *flags |= IFF_UP; /* We turn these on in our init_tx */ *flags |= IFF_BROADCAST; *flags |= IFF_MULTICAST; return 0; } case SIOCGIFMTU: { uint32_t * mtu = argp; *mtu = nic->eth.mtu; return 0; } case SIOCGIFCOUNTS: { memcpy(argp, &nic->counts, sizeof(netif_counters_t)); return 0; } default: return -EINVAL; } } static ssize_t write_e1000(fs_node_t *node, off_t offset, size_t size, uint8_t *buffer) { struct e1000_nic * nic = node->device; /* write packet */ send_packet(nic, buffer, size); return size; } static void ints_off(struct e1000_nic * nic) { write_command(nic, E1000_REG_IMC, 0xFFFFFFFF); write_command(nic, E1000_REG_ICR, 0xFFFFFFFF); read_command(nic, E1000_REG_STATUS); } static void e1000_init(struct e1000_nic * nic) { uint32_t e1000_device_pci = nic->pci_device; nic->rx_phys = mmu_allocate_n_frames(2) << 12; nic->rx = mmu_map_mmio_region(nic->rx_phys, 8192); nic->tx_phys = mmu_allocate_n_frames(2) << 12; nic->tx = mmu_map_mmio_region(nic->tx_phys, 8192); memset(nic->rx, 0, sizeof(struct e1000_rx_desc) * E1000_NUM_RX_DESC); memset(nic->tx, 0, sizeof(struct e1000_tx_desc) * E1000_NUM_TX_DESC); /* Allocate buffers */ for (int i = 0; i < E1000_NUM_RX_DESC; ++i) { nic->rx[i].addr = mmu_allocate_a_frame() << 12; nic->rx_virt[i] = mmu_map_mmio_region(nic->rx[i].addr, 4096); mmu_frame_map_address(mmu_get_page((uintptr_t)nic->rx_virt[i],0),MMU_FLAG_WRITABLE|MMU_FLAG_WC,nic->rx[i].addr); nic->rx[i].status = 0; } for (int i = 0; i < E1000_NUM_TX_DESC; ++i) { nic->tx[i].addr = mmu_allocate_a_frame() << 12; nic->tx_virt[i] = mmu_map_mmio_region(nic->tx[i].addr, 4096); mmu_frame_allocate(mmu_get_page((uintptr_t)nic->tx_virt[i],0),MMU_FLAG_WRITABLE|MMU_FLAG_WC); memset(nic->tx_virt[i], 0, 4096); nic->tx[i].status = 0; nic->tx[i].cmd = (1 << 0); } uint16_t command_reg = pci_read_field(e1000_device_pci, PCI_COMMAND, 2); command_reg |= (1 << 2); command_reg |= (1 << 0); pci_write_field(e1000_device_pci, PCI_COMMAND, 2, command_reg); delay_yield(10000); /* Is this size enough? */ uint32_t initial_bar = pci_read_field(e1000_device_pci, PCI_BAR0, 4); nic->mmio_addr = (uintptr_t)mmu_map_mmio_region(initial_bar, 0x8000); eeprom_detect(nic); read_mac(nic); write_mac(nic); #define CTRL_PHY_RST (1UL << 31UL) #define CTRL_RST (1UL << 26UL) #define CTRL_SLU (1UL << 6UL) #define CTRL_LRST (1UL << 3UL) nic->irq_number = pci_get_interrupt(e1000_device_pci); irq_install_handler(nic->irq_number, irq_handler, nic->eth.if_name); /* Disable interrupts */ ints_off(nic); /* Turn off receive + transmit */ write_command(nic, E1000_REG_RCTRL, 0); write_command(nic, E1000_REG_TCTRL, TCTL_PSP); read_command(nic, E1000_REG_STATUS); delay_yield(10000); /* Reset everything */ uint32_t ctrl = read_command(nic, E1000_REG_CTRL); ctrl |= CTRL_RST; write_command(nic, E1000_REG_CTRL, ctrl); delay_yield(20000); /* Turn off interrupts _again_ */ ints_off(nic); /* Recommended flow control settings? */ write_command(nic, 0x0028, 0x002C8001); write_command(nic, 0x002c, 0x0100); write_command(nic, 0x0030, 0x8808); write_command(nic, 0x0170, 0xFFFF); /* Link up */ uint32_t status = read_command(nic, E1000_REG_CTRL); status |= CTRL_SLU; status |= (2 << 8); /* Speed to gigabit... */ status &= ~CTRL_LRST; status &= ~CTRL_PHY_RST; write_command(nic, E1000_REG_CTRL, status); /* Clear statistical counters */ for (int i = 0; i < 128; ++i) { write_command(nic, 0x5200 + i * 4, 0); } for (int i = 0; i < 64; ++i) { read_command(nic, 0x4000 + i * 4); } init_rx(nic); init_tx(nic); write_command(nic, E1000_REG_RDTR, 0); write_command(nic, E1000_REG_ITR, 500); read_command(nic, E1000_REG_STATUS); nic->link_status = (read_command(nic, E1000_REG_STATUS) & (1 << 1)); nic->eth.device_node = calloc(sizeof(fs_node_t),1); snprintf(nic->eth.device_node->name, 100, "%s", nic->eth.if_name); nic->eth.device_node->flags = FS_BLOCKDEVICE; /* NETDEVICE? */ nic->eth.device_node->mask = 0666; /* temporary; shouldn't be doing this with these device files */ nic->eth.device_node->ioctl = ioctl_e1000; nic->eth.device_node->write = write_e1000; nic->eth.device_node->device = nic; nic->eth.mtu = 1500; /* guess */ net_add_interface(nic->eth.if_name, nic->eth.device_node); char worker_name[34]; snprintf(worker_name, 33, "[%s]", nic->eth.if_name); nic->queuer = spawn_worker_thread(e1000_queuer, worker_name, nic); nic->configured = 1; /* Twiddle interrupts */ write_command(nic, E1000_REG_IMS, INTS); delay_yield(10000); } static void find_e1000(uint32_t device, uint16_t vendorid, uint16_t deviceid, void * found) { if ((vendorid == 0x8086) && (deviceid == 0x100e || deviceid == 0x1004 || deviceid == 0x100f || deviceid == 0x10ea || deviceid == 0x10d3)) { /* Allocate a device */ struct e1000_nic * nic = calloc(1,sizeof(struct e1000_nic)); nic->pci_device = device; nic->deviceid = deviceid; devices[device_count++] = nic; snprintf(nic->eth.if_name, 31, "enp%ds%d", (int)pci_extract_bus(device), (int)pci_extract_slot(device)); e1000_init(nic); *(int*)found = 1; } } static int e1000_install(int argc, char * argv[]) { uint32_t found = 0; pci_scan(&find_e1000, -1, &found); if (!found) { /* TODO: Clean up? Remove ourselves? */ return -ENODEV; } return 0; } static int fini(void) { /* TODO: Uninstall device */ return 0; } struct Module metadata = { .name = "e1000", .init = e1000_install, .fini = fini, };