drivers/net/e1000/em_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <sys/queue.h>
  35 #include <stdio.h>
  36 #include <errno.h>
  37 #include <stdint.h>
  38 #include <stdarg.h>
  39
  40 #include <rte_common.h>
  41 #include <rte_interrupts.h>
  42 #include <rte_byteorder.h>
  43 #include <rte_log.h>
  44 #include <rte_debug.h>
  45 #include <rte_pci.h>
  46 #include <rte_ether.h>
  47 #include <rte_ethdev.h>
  48 #include <rte_memory.h>
  49 #include <rte_memzone.h>
  50 #include <rte_eal.h>
  51 #include <rte_atomic.h>
  52 #include <rte_malloc.h>
  53 #include <rte_dev.h>
  54
  55 #include "e1000_logs.h"
  56 #include "base/e1000_api.h"
  57 #include "e1000_ethdev.h"
  58
  59 #define EM_EIAC                 0x000DC
  60
  61 #define PMD_ROUNDUP(x,y)        (((x) + (y) - 1)/(y) * (y))
  62
  63
  64 static int eth_em_configure(struct rte_eth_dev *dev);
  65 static int eth_em_start(struct rte_eth_dev *dev);
  66 static void eth_em_stop(struct rte_eth_dev *dev);
  67 static void eth_em_close(struct rte_eth_dev *dev);
  68 static void eth_em_promiscuous_enable(struct rte_eth_dev *dev);
  69 static void eth_em_promiscuous_disable(struct rte_eth_dev *dev);
  70 static void eth_em_allmulticast_enable(struct rte_eth_dev *dev);
  71 static void eth_em_allmulticast_disable(struct rte_eth_dev *dev);
  72 static int eth_em_link_update(struct rte_eth_dev *dev,
  73                                 int wait_to_complete);
  74 static void eth_em_stats_get(struct rte_eth_dev *dev,
  75                                 struct rte_eth_stats *rte_stats);
  76 static void eth_em_stats_reset(struct rte_eth_dev *dev);
  77 static void eth_em_infos_get(struct rte_eth_dev *dev,
  78                                 struct rte_eth_dev_info *dev_info);
  79 static int eth_em_flow_ctrl_get(struct rte_eth_dev *dev,
  80                                 struct rte_eth_fc_conf *fc_conf);
  81 static int eth_em_flow_ctrl_set(struct rte_eth_dev *dev,
  82                                 struct rte_eth_fc_conf *fc_conf);
  83 static int eth_em_interrupt_setup(struct rte_eth_dev *dev);
  84 static int eth_em_rxq_interrupt_setup(struct rte_eth_dev *dev);
  85 static int eth_em_interrupt_get_status(struct rte_eth_dev *dev);
  86 static int eth_em_interrupt_action(struct rte_eth_dev *dev);
  87 static void eth_em_interrupt_handler(struct rte_intr_handle *handle,
  88                                                         void *param);
  89
  90 static int em_hw_init(struct e1000_hw *hw);
  91 static int em_hardware_init(struct e1000_hw *hw);
  92 static void em_hw_control_acquire(struct e1000_hw *hw);
  93 static void em_hw_control_release(struct e1000_hw *hw);
  94 static void em_init_manageability(struct e1000_hw *hw);
  95 static void em_release_manageability(struct e1000_hw *hw);
  96
  97 static int eth_em_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
  98
  99 static int eth_em_vlan_filter_set(struct rte_eth_dev *dev,
 100                 uint16_t vlan_id, int on);
 101 static void eth_em_vlan_offload_set(struct rte_eth_dev *dev, int mask);
 102 static void em_vlan_hw_filter_enable(struct rte_eth_dev *dev);
 103 static void em_vlan_hw_filter_disable(struct rte_eth_dev *dev);
 104 static void em_vlan_hw_strip_enable(struct rte_eth_dev *dev);
 105 static void em_vlan_hw_strip_disable(struct rte_eth_dev *dev);
 106
 107 /*
 108 static void eth_em_vlan_filter_set(struct rte_eth_dev *dev,
 109                                         uint16_t vlan_id, int on);
 110 */
 111
 112 static int eth_em_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
 113 static int eth_em_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
 114 static void em_lsc_intr_disable(struct e1000_hw *hw);
 115 static void em_rxq_intr_enable(struct e1000_hw *hw);
 116 static void em_rxq_intr_disable(struct e1000_hw *hw);
 117
 118 static int eth_em_led_on(struct rte_eth_dev *dev);
 119 static int eth_em_led_off(struct rte_eth_dev *dev);
 120
 121 static int em_get_rx_buffer_size(struct e1000_hw *hw);
 122 static void eth_em_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 123                 uint32_t index, uint32_t pool);
 124 static void eth_em_rar_clear(struct rte_eth_dev *dev, uint32_t index);
 125
 126 static int eth_em_set_mc_addr_list(struct rte_eth_dev *dev,
 127                                    struct ether_addr *mc_addr_set,
 128                                    uint32_t nb_mc_addr);
 129
 130 #define EM_FC_PAUSE_TIME 0x0680
 131 #define EM_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
 132 #define EM_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
 133
 134 static enum e1000_fc_mode em_fc_setting = e1000_fc_full;
 135
 136 /*
 137  * The set of PCI devices this driver supports
 138  */
 139 static const struct rte_pci_id pci_id_em_map[] = {
 140         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82540EM) },
 141         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82545EM_COPPER) },
 142         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82545EM_FIBER) },
 143         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82546EB_COPPER) },
 144         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82546EB_FIBER) },
 145         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82546EB_QUAD_COPPER) },
 146         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_COPPER) },
 147         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_FIBER) },
 148         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_SERDES) },
 149         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_SERDES_DUAL) },
 150         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_SERDES_QUAD) },
 151         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_QUAD_COPPER) },
 152         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571PT_QUAD_COPPER) },
 153         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_QUAD_FIBER) },
 154         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82571EB_QUAD_COPPER_LP) },
 155         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82572EI_COPPER) },
 156         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82572EI_FIBER) },
 157         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82572EI_SERDES) },
 158         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82572EI) },
 159         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82573L) },
 160         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82574L) },
 161         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82574LA) },
 162         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_82583V) },
 163         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_LPT_I217_LM) },
 164         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_LPT_I217_V) },
 165         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_LPTLP_I218_LM) },
 166         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_LPTLP_I218_V) },
 167         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_I218_LM2) },
 168         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_I218_V2) },
 169         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_I218_LM3) },
 170         { RTE_PCI_DEVICE(E1000_INTEL_VENDOR_ID, E1000_DEV_ID_PCH_I218_V3) },
 171         { .vendor_id = 0, /* sentinel */ },
 172 };
 173
 174 static const struct eth_dev_ops eth_em_ops = {
 175         .dev_configure        = eth_em_configure,
 176         .dev_start            = eth_em_start,
 177         .dev_stop             = eth_em_stop,
 178         .dev_close            = eth_em_close,
 179         .promiscuous_enable   = eth_em_promiscuous_enable,
 180         .promiscuous_disable  = eth_em_promiscuous_disable,
 181         .allmulticast_enable  = eth_em_allmulticast_enable,
 182         .allmulticast_disable = eth_em_allmulticast_disable,
 183         .link_update          = eth_em_link_update,
 184         .stats_get            = eth_em_stats_get,
 185         .stats_reset          = eth_em_stats_reset,
 186         .dev_infos_get        = eth_em_infos_get,
 187         .mtu_set              = eth_em_mtu_set,
 188         .vlan_filter_set      = eth_em_vlan_filter_set,
 189         .vlan_offload_set     = eth_em_vlan_offload_set,
 190         .rx_queue_setup       = eth_em_rx_queue_setup,
 191         .rx_queue_release     = eth_em_rx_queue_release,
 192         .rx_queue_count       = eth_em_rx_queue_count,
 193         .rx_descriptor_done   = eth_em_rx_descriptor_done,
 194         .tx_queue_setup       = eth_em_tx_queue_setup,
 195         .tx_queue_release     = eth_em_tx_queue_release,
 196         .rx_queue_intr_enable = eth_em_rx_queue_intr_enable,
 197         .rx_queue_intr_disable = eth_em_rx_queue_intr_disable,
 198         .dev_led_on           = eth_em_led_on,
 199         .dev_led_off          = eth_em_led_off,
 200         .flow_ctrl_get        = eth_em_flow_ctrl_get,
 201         .flow_ctrl_set        = eth_em_flow_ctrl_set,
 202         .mac_addr_add         = eth_em_rar_set,
 203         .mac_addr_remove      = eth_em_rar_clear,
 204         .set_mc_addr_list     = eth_em_set_mc_addr_list,
 205         .rxq_info_get         = em_rxq_info_get,
 206         .txq_info_get         = em_txq_info_get,
 207 };
 208
 209 /**
 210  * Atomically reads the link status information from global
 211  * structure rte_eth_dev.
 212  *
 213  * @param dev
 214  *   - Pointer to the structure rte_eth_dev to read from.
 215  *   - Pointer to the buffer to be saved with the link status.
 216  *
 217  * @return
 218  *   - On success, zero.
 219  *   - On failure, negative value.
 220  */
 221 static inline int
 222 rte_em_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 223                                 struct rte_eth_link *link)
 224 {
 225         struct rte_eth_link *dst = link;
 226         struct rte_eth_link *src = &(dev->data->dev_link);
 227
 228         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 229                                         *(uint64_t *)src) == 0)
 230                 return -1;
 231
 232         return 0;
 233 }
 234
 235 /**
 236  * Atomically writes the link status information into global
 237  * structure rte_eth_dev.
 238  *
 239  * @param dev
 240  *   - Pointer to the structure rte_eth_dev to read from.
 241  *   - Pointer to the buffer to be saved with the link status.
 242  *
 243  * @return
 244  *   - On success, zero.
 245  *   - On failure, negative value.
 246  */
 247 static inline int
 248 rte_em_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 249                                 struct rte_eth_link *link)
 250 {
 251         struct rte_eth_link *dst = &(dev->data->dev_link);
 252         struct rte_eth_link *src = link;
 253
 254         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 255                                         *(uint64_t *)src) == 0)
 256                 return -1;
 257
 258         return 0;
 259 }
 260
 261 /**
 262  *  eth_em_dev_is_ich8 - Check for ICH8 device
 263  *  @hw: pointer to the HW structure
 264  *
 265  *  return TRUE for ICH8, otherwise FALSE
 266  **/
 267 static bool
 268 eth_em_dev_is_ich8(struct e1000_hw *hw)
 269 {
 270         DEBUGFUNC("eth_em_dev_is_ich8");
 271
 272         switch (hw->device_id) {
 273         case E1000_DEV_ID_PCH_LPT_I217_LM:
 274         case E1000_DEV_ID_PCH_LPT_I217_V:
 275         case E1000_DEV_ID_PCH_LPTLP_I218_LM:
 276         case E1000_DEV_ID_PCH_LPTLP_I218_V:
 277         case E1000_DEV_ID_PCH_I218_V2:
 278         case E1000_DEV_ID_PCH_I218_LM2:
 279         case E1000_DEV_ID_PCH_I218_V3:
 280         case E1000_DEV_ID_PCH_I218_LM3:
 281                 return 1;
 282         default:
 283                 return 0;
 284         }
 285 }
 286
 287 static int
 288 eth_em_dev_init(struct rte_eth_dev *eth_dev)
 289 {
 290         struct rte_pci_device *pci_dev;
 291         struct e1000_adapter *adapter =
 292                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
 293         struct e1000_hw *hw =
 294                 E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 295         struct e1000_vfta * shadow_vfta =
 296                 E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
 297
 298         pci_dev = eth_dev->pci_dev;
 299
 300         eth_dev->dev_ops = &eth_em_ops;
 301         eth_dev->rx_pkt_burst = (eth_rx_burst_t)&eth_em_recv_pkts;
 302         eth_dev->tx_pkt_burst = (eth_tx_burst_t)&eth_em_xmit_pkts;
 303
 304         /* for secondary processes, we don't initialise any further as primary
 305          * has already done this work. Only check we don't need a different
 306          * RX function */
 307         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
 308                 if (eth_dev->data->scattered_rx)
 309                         eth_dev->rx_pkt_burst =
 310                                 (eth_rx_burst_t)&eth_em_recv_scattered_pkts;
 311                 return 0;
 312         }
 313
 314         rte_eth_copy_pci_info(eth_dev, pci_dev);
 315
 316         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
 317         hw->device_id = pci_dev->id.device_id;
 318         adapter->stopped = 0;
 319
 320         /* For ICH8 support we'll need to map the flash memory BAR */
 321         if (eth_em_dev_is_ich8(hw))
 322                 hw->flash_address = (void *)pci_dev->mem_resource[1].addr;
 323
 324         if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS ||
 325                         em_hw_init(hw) != 0) {
 326                 PMD_INIT_LOG(ERR, "port_id %d vendorID=0x%x deviceID=0x%x: "
 327                         "failed to init HW",
 328                         eth_dev->data->port_id, pci_dev->id.vendor_id,
 329                         pci_dev->id.device_id);
 330                 return -ENODEV;
 331         }
 332
 333         /* Allocate memory for storing MAC addresses */
 334         eth_dev->data->mac_addrs = rte_zmalloc("e1000", ETHER_ADDR_LEN *
 335                         hw->mac.rar_entry_count, 0);
 336         if (eth_dev->data->mac_addrs == NULL) {
 337                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
 338                         "store MAC addresses",
 339                         ETHER_ADDR_LEN * hw->mac.rar_entry_count);
 340                 return -ENOMEM;
 341         }
 342
 343         /* Copy the permanent MAC address */
 344         ether_addr_copy((struct ether_addr *) hw->mac.addr,
 345                 eth_dev->data->mac_addrs);
 346
 347         /* initialize the vfta */
 348         memset(shadow_vfta, 0, sizeof(*shadow_vfta));
 349
 350         PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
 351                      eth_dev->data->port_id, pci_dev->id.vendor_id,
 352                      pci_dev->id.device_id);
 353
 354         rte_intr_callback_register(&(pci_dev->intr_handle),
 355                 eth_em_interrupt_handler, (void *)eth_dev);
 356
 357         return 0;
 358 }
 359
 360 static int
 361 eth_em_dev_uninit(struct rte_eth_dev *eth_dev)
 362 {
 363         struct rte_pci_device *pci_dev;
 364         struct e1000_adapter *adapter =
 365                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
 366
 367         PMD_INIT_FUNC_TRACE();
 368
 369         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 370                 return -EPERM;
 371
 372         pci_dev = eth_dev->pci_dev;
 373
 374         if (adapter->stopped == 0)
 375                 eth_em_close(eth_dev);
 376
 377         eth_dev->dev_ops = NULL;
 378         eth_dev->rx_pkt_burst = NULL;
 379         eth_dev->tx_pkt_burst = NULL;
 380
 381         rte_free(eth_dev->data->mac_addrs);
 382         eth_dev->data->mac_addrs = NULL;
 383
 384         /* disable uio intr before callback unregister */
 385         rte_intr_disable(&(pci_dev->intr_handle));
 386         rte_intr_callback_unregister(&(pci_dev->intr_handle),
 387                 eth_em_interrupt_handler, (void *)eth_dev);
 388
 389         return 0;
 390 }
 391
 392 static struct eth_driver rte_em_pmd = {
 393         .pci_drv = {
 394                 .id_table = pci_id_em_map,
 395                 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
 396                         RTE_PCI_DRV_DETACHABLE,
 397                 .probe = rte_eth_dev_pci_probe,
 398                 .remove = rte_eth_dev_pci_remove,
 399         },
 400         .eth_dev_init = eth_em_dev_init,
 401         .eth_dev_uninit = eth_em_dev_uninit,
 402         .dev_private_size = sizeof(struct e1000_adapter),
 403 };
 404
 405 static int
 406 em_hw_init(struct e1000_hw *hw)
 407 {
 408         int diag;
 409
 410         diag = hw->mac.ops.init_params(hw);
 411         if (diag != 0) {
 412                 PMD_INIT_LOG(ERR, "MAC Initialization Error");
 413                 return diag;
 414         }
 415         diag = hw->nvm.ops.init_params(hw);
 416         if (diag != 0) {
 417                 PMD_INIT_LOG(ERR, "NVM Initialization Error");
 418                 return diag;
 419         }
 420         diag = hw->phy.ops.init_params(hw);
 421         if (diag != 0) {
 422                 PMD_INIT_LOG(ERR, "PHY Initialization Error");
 423                 return diag;
 424         }
 425         (void) e1000_get_bus_info(hw);
 426
 427         hw->mac.autoneg = 1;
 428         hw->phy.autoneg_wait_to_complete = 0;
 429         hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
 430
 431         e1000_init_script_state_82541(hw, TRUE);
 432         e1000_set_tbi_compatibility_82543(hw, TRUE);
 433
 434         /* Copper options */
 435         if (hw->phy.media_type == e1000_media_type_copper) {
 436                 hw->phy.mdix = 0; /* AUTO_ALL_MODES */
 437                 hw->phy.disable_polarity_correction = 0;
 438                 hw->phy.ms_type = e1000_ms_hw_default;
 439         }
 440
 441         /*
 442          * Start from a known state, this is important in reading the nvm
 443          * and mac from that.
 444          */
 445         e1000_reset_hw(hw);
 446
 447         /* Make sure we have a good EEPROM before we read from it */
 448         if (e1000_validate_nvm_checksum(hw) < 0) {
 449                 /*
 450                  * Some PCI-E parts fail the first check due to
 451                  * the link being in sleep state, call it again,
 452                  * if it fails a second time its a real issue.
 453                  */
 454                 diag = e1000_validate_nvm_checksum(hw);
 455                 if (diag < 0) {
 456                         PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
 457                         goto error;
 458                 }
 459         }
 460
 461         /* Read the permanent MAC address out of the EEPROM */
 462         diag = e1000_read_mac_addr(hw);
 463         if (diag != 0) {
 464                 PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
 465                 goto error;
 466         }
 467
 468         /* Now initialize the hardware */
 469         diag = em_hardware_init(hw);
 470         if (diag != 0) {
 471                 PMD_INIT_LOG(ERR, "Hardware initialization failed");
 472                 goto error;
 473         }
 474
 475         hw->mac.get_link_status = 1;
 476
 477         /* Indicate SOL/IDER usage */
 478         diag = e1000_check_reset_block(hw);
 479         if (diag < 0) {
 480                 PMD_INIT_LOG(ERR, "PHY reset is blocked due to "
 481                         "SOL/IDER session");
 482         }
 483         return 0;
 484
 485 error:
 486         em_hw_control_release(hw);
 487         return diag;
 488 }
 489
 490 static int
 491 eth_em_configure(struct rte_eth_dev *dev)
 492 {
 493         struct e1000_interrupt *intr =
 494                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
 495
 496         PMD_INIT_FUNC_TRACE();
 497         intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
 498         PMD_INIT_FUNC_TRACE();
 499
 500         return 0;
 501 }
 502
 503 static void
 504 em_set_pba(struct e1000_hw *hw)
 505 {
 506         uint32_t pba;
 507
 508         /*
 509          * Packet Buffer Allocation (PBA)
 510          * Writing PBA sets the receive portion of the buffer
 511          * the remainder is used for the transmit buffer.
 512          * Devices before the 82547 had a Packet Buffer of 64K.
 513          * After the 82547 the buffer was reduced to 40K.
 514          */
 515         switch (hw->mac.type) {
 516                 case e1000_82547:
 517                 case e1000_82547_rev_2:
 518                 /* 82547: Total Packet Buffer is 40K */
 519                         pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
 520                         break;
 521                 case e1000_82571:
 522                 case e1000_82572:
 523                 case e1000_80003es2lan:
 524                         pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
 525                         break;
 526                 case e1000_82573: /* 82573: Total Packet Buffer is 32K */
 527                         pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
 528                         break;
 529                 case e1000_82574:
 530                 case e1000_82583:
 531                         pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
 532                         break;
 533                 case e1000_ich8lan:
 534                         pba = E1000_PBA_8K;
 535                         break;
 536                 case e1000_ich9lan:
 537                 case e1000_ich10lan:
 538                         pba = E1000_PBA_10K;
 539                         break;
 540                 case e1000_pchlan:
 541                 case e1000_pch2lan:
 542                 case e1000_pch_lpt:
 543                         pba = E1000_PBA_26K;
 544                         break;
 545                 default:
 546                         pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
 547         }
 548
 549         E1000_WRITE_REG(hw, E1000_PBA, pba);
 550 }
 551
 552 static int
 553 eth_em_start(struct rte_eth_dev *dev)
 554 {
 555         struct e1000_adapter *adapter =
 556                 E1000_DEV_PRIVATE(dev->data->dev_private);
 557         struct e1000_hw *hw =
 558                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 559         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
 560         int ret, mask;
 561         uint32_t intr_vector = 0;
 562         uint32_t *speeds;
 563         int num_speeds;
 564         bool autoneg;
 565
 566         PMD_INIT_FUNC_TRACE();
 567
 568         eth_em_stop(dev);
 569
 570         e1000_power_up_phy(hw);
 571
 572         /* Set default PBA value */
 573         em_set_pba(hw);
 574
 575         /* Put the address into the Receive Address Array */
 576         e1000_rar_set(hw, hw->mac.addr, 0);
 577
 578         /*
 579          * With the 82571 adapter, RAR[0] may be overwritten
 580          * when the other port is reset, we make a duplicate
 581          * in RAR[14] for that eventuality, this assures
 582          * the interface continues to function.
 583          */
 584         if (hw->mac.type == e1000_82571) {
 585                 e1000_set_laa_state_82571(hw, TRUE);
 586                 e1000_rar_set(hw, hw->mac.addr, E1000_RAR_ENTRIES - 1);
 587         }
 588
 589         /* Initialize the hardware */
 590         if (em_hardware_init(hw)) {
 591                 PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
 592                 return -EIO;
 593         }
 594
 595         E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN);
 596
 597         /* Configure for OS presence */
 598         em_init_manageability(hw);
 599
 600         if (dev->data->dev_conf.intr_conf.rxq != 0) {
 601                 intr_vector = dev->data->nb_rx_queues;
 602                 if (rte_intr_efd_enable(intr_handle, intr_vector))
 603                         return -1;
 604         }
 605
 606         if (rte_intr_dp_is_en(intr_handle)) {
 607                 intr_handle->intr_vec =
 608                         rte_zmalloc("intr_vec",
 609                                         dev->data->nb_rx_queues * sizeof(int), 0);
 610                 if (intr_handle->intr_vec == NULL) {
 611                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
 612                                                 " intr_vec\n", dev->data->nb_rx_queues);
 613                         return -ENOMEM;
 614                 }
 615
 616                 /* enable rx interrupt */
 617                 em_rxq_intr_enable(hw);
 618         }
 619
 620         eth_em_tx_init(dev);
 621
 622         ret = eth_em_rx_init(dev);
 623         if (ret) {
 624                 PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
 625                 em_dev_clear_queues(dev);
 626                 return ret;
 627         }
 628
 629         e1000_clear_hw_cntrs_base_generic(hw);
 630
 631         mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK | \
 632                         ETH_VLAN_EXTEND_MASK;
 633         eth_em_vlan_offload_set(dev, mask);
 634
 635         /* Set Interrupt Throttling Rate to maximum allowed value. */
 636         E1000_WRITE_REG(hw, E1000_ITR, UINT16_MAX);
 637
 638         /* Setup link speed and duplex */
 639         speeds = &dev->data->dev_conf.link_speeds;
 640         if (*speeds == ETH_LINK_SPEED_AUTONEG) {
 641                 hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
 642                 hw->mac.autoneg = 1;
 643         } else {
 644                 num_speeds = 0;
 645                 autoneg = (*speeds & ETH_LINK_SPEED_FIXED) == 0;
 646
 647                 /* Reset */
 648                 hw->phy.autoneg_advertised = 0;
 649
 650                 if (*speeds & ~(ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
 651                                 ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
 652                                 ETH_LINK_SPEED_1G | ETH_LINK_SPEED_FIXED)) {
 653                         num_speeds = -1;
 654                         goto error_invalid_config;
 655                 }
 656                 if (*speeds & ETH_LINK_SPEED_10M_HD) {
 657                         hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
 658                         num_speeds++;
 659                 }
 660                 if (*speeds & ETH_LINK_SPEED_10M) {
 661                         hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
 662                         num_speeds++;
 663                 }
 664                 if (*speeds & ETH_LINK_SPEED_100M_HD) {
 665                         hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
 666                         num_speeds++;
 667                 }
 668                 if (*speeds & ETH_LINK_SPEED_100M) {
 669                         hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
 670                         num_speeds++;
 671                 }
 672                 if (*speeds & ETH_LINK_SPEED_1G) {
 673                         hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
 674                         num_speeds++;
 675                 }
 676                 if (num_speeds == 0 || (!autoneg && (num_speeds > 1)))
 677                         goto error_invalid_config;
 678
 679                 /* Set/reset the mac.autoneg based on the link speed,
 680                  * fixed or not
 681                  */
 682                 if (!autoneg) {
 683                         hw->mac.autoneg = 0;
 684                         hw->mac.forced_speed_duplex =
 685                                         hw->phy.autoneg_advertised;
 686                 } else {
 687                         hw->mac.autoneg = 1;
 688                 }
 689         }
 690
 691         e1000_setup_link(hw);
 692
 693         if (rte_intr_allow_others(intr_handle)) {
 694                 /* check if lsc interrupt is enabled */
 695                 if (dev->data->dev_conf.intr_conf.lsc != 0) {
 696                         ret = eth_em_interrupt_setup(dev);
 697                         if (ret) {
 698                                 PMD_INIT_LOG(ERR, "Unable to setup interrupts");
 699                                 em_dev_clear_queues(dev);
 700                                 return ret;
 701                         }
 702                 }
 703         } else {
 704                 rte_intr_callback_unregister(intr_handle,
 705                                                 eth_em_interrupt_handler,
 706                                                 (void *)dev);
 707                 if (dev->data->dev_conf.intr_conf.lsc != 0)
 708                         PMD_INIT_LOG(INFO, "lsc won't enable because of"
 709                                      " no intr multiplex\n");
 710         }
 711         /* check if rxq interrupt is enabled */
 712         if (dev->data->dev_conf.intr_conf.rxq != 0)
 713                 eth_em_rxq_interrupt_setup(dev);
 714
 715         rte_intr_enable(intr_handle);
 716
 717         adapter->stopped = 0;
 718
 719         PMD_INIT_LOG(DEBUG, "<<");
 720
 721         return 0;
 722
 723 error_invalid_config:
 724         PMD_INIT_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
 725                      dev->data->dev_conf.link_speeds, dev->data->port_id);
 726         em_dev_clear_queues(dev);
 727         return -EINVAL;
 728 }
 729
 730 /*********************************************************************
 731  *
 732  *  This routine disables all traffic on the adapter by issuing a
 733  *  global reset on the MAC.
 734  *
 735  **********************************************************************/
 736 static void
 737 eth_em_stop(struct rte_eth_dev *dev)
 738 {
 739         struct rte_eth_link link;
 740         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 741         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
 742
 743         em_rxq_intr_disable(hw);
 744         em_lsc_intr_disable(hw);
 745
 746         e1000_reset_hw(hw);
 747         if (hw->mac.type >= e1000_82544)
 748                 E1000_WRITE_REG(hw, E1000_WUC, 0);
 749
 750         /* Power down the phy. Needed to make the link go down */
 751         e1000_power_down_phy(hw);
 752
 753         em_dev_clear_queues(dev);
 754
 755         /* clear the recorded link status */
 756         memset(&link, 0, sizeof(link));
 757         rte_em_dev_atomic_write_link_status(dev, &link);
 758
 759         if (!rte_intr_allow_others(intr_handle))
 760                 /* resume to the default handler */
 761                 rte_intr_callback_register(intr_handle,
 762                                            eth_em_interrupt_handler,
 763                                            (void *)dev);
 764
 765         /* Clean datapath event and queue/vec mapping */
 766         rte_intr_efd_disable(intr_handle);
 767         if (intr_handle->intr_vec != NULL) {
 768                 rte_free(intr_handle->intr_vec);
 769                 intr_handle->intr_vec = NULL;
 770         }
 771 }
 772
 773 static void
 774 eth_em_close(struct rte_eth_dev *dev)
 775 {
 776         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 777         struct e1000_adapter *adapter =
 778                 E1000_DEV_PRIVATE(dev->data->dev_private);
 779
 780         eth_em_stop(dev);
 781         adapter->stopped = 1;
 782         em_dev_free_queues(dev);
 783         e1000_phy_hw_reset(hw);
 784         em_release_manageability(hw);
 785         em_hw_control_release(hw);
 786 }
 787
 788 static int
 789 em_get_rx_buffer_size(struct e1000_hw *hw)
 790 {
 791         uint32_t rx_buf_size;
 792
 793         rx_buf_size = ((E1000_READ_REG(hw, E1000_PBA) & UINT16_MAX) << 10);
 794         return rx_buf_size;
 795 }
 796
 797 /*********************************************************************
 798  *
 799  *  Initialize the hardware
 800  *
 801  **********************************************************************/
 802 static int
 803 em_hardware_init(struct e1000_hw *hw)
 804 {
 805         uint32_t rx_buf_size;
 806         int diag;
 807
 808         /* Issue a global reset */
 809         e1000_reset_hw(hw);
 810
 811         /* Let the firmware know the OS is in control */
 812         em_hw_control_acquire(hw);
 813
 814         /*
 815          * These parameters control the automatic generation (Tx) and
 816          * response (Rx) to Ethernet PAUSE frames.
 817          * - High water mark should allow for at least two standard size (1518)
 818          *   frames to be received after sending an XOFF.
 819          * - Low water mark works best when it is very near the high water mark.
 820          *   This allows the receiver to restart by sending XON when it has
 821          *   drained a bit. Here we use an arbitrary value of 1500 which will
 822          *   restart after one full frame is pulled from the buffer. There
 823          *   could be several smaller frames in the buffer and if so they will
 824          *   not trigger the XON until their total number reduces the buffer
 825          *   by 1500.
 826          * - The pause time is fairly large at 1000 x 512ns = 512 usec.
 827          */
 828         rx_buf_size = em_get_rx_buffer_size(hw);
 829
 830         hw->fc.high_water = rx_buf_size - PMD_ROUNDUP(ETHER_MAX_LEN * 2, 1024);
 831         hw->fc.low_water = hw->fc.high_water - 1500;
 832
 833         if (hw->mac.type == e1000_80003es2lan)
 834                 hw->fc.pause_time = UINT16_MAX;
 835         else
 836                 hw->fc.pause_time = EM_FC_PAUSE_TIME;
 837
 838         hw->fc.send_xon = 1;
 839
 840         /* Set Flow control, use the tunable location if sane */
 841         if (em_fc_setting <= e1000_fc_full)
 842                 hw->fc.requested_mode = em_fc_setting;
 843         else
 844                 hw->fc.requested_mode = e1000_fc_none;
 845
 846         /* Workaround: no TX flow ctrl for PCH */
 847         if (hw->mac.type == e1000_pchlan)
 848                 hw->fc.requested_mode = e1000_fc_rx_pause;
 849
 850         /* Override - settings for PCH2LAN, ya its magic :) */
 851         if (hw->mac.type == e1000_pch2lan) {
 852                 hw->fc.high_water = 0x5C20;
 853                 hw->fc.low_water = 0x5048;
 854                 hw->fc.pause_time = 0x0650;
 855                 hw->fc.refresh_time = 0x0400;
 856         } else if (hw->mac.type == e1000_pch_lpt) {
 857                 hw->fc.requested_mode = e1000_fc_full;
 858         }
 859
 860         diag = e1000_init_hw(hw);
 861         if (diag < 0)
 862                 return diag;
 863         e1000_check_for_link(hw);
 864         return 0;
 865 }
 866
 867 /* This function is based on em_update_stats_counters() in e1000/if_em.c */
 868 static void
 869 eth_em_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
 870 {
 871         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 872         struct e1000_hw_stats *stats =
 873                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
 874         int pause_frames;
 875
 876         if(hw->phy.media_type == e1000_media_type_copper ||
 877                         (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
 878                 stats->symerrs += E1000_READ_REG(hw,E1000_SYMERRS);
 879                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
 880         }
 881
 882         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
 883         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
 884         stats->scc += E1000_READ_REG(hw, E1000_SCC);
 885         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
 886
 887         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
 888         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
 889         stats->colc += E1000_READ_REG(hw, E1000_COLC);
 890         stats->dc += E1000_READ_REG(hw, E1000_DC);
 891         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
 892         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
 893         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
 894
 895         /*
 896          * For watchdog management we need to know if we have been
 897          * paused during the last interval, so capture that here.
 898          */
 899         pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
 900         stats->xoffrxc += pause_frames;
 901         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
 902         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
 903         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
 904         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
 905         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
 906         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
 907         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
 908         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
 909         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
 910         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
 911         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
 912         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
 913
 914         /*
 915          * For the 64-bit byte counters the low dword must be read first.
 916          * Both registers clear on the read of the high dword.
 917          */
 918
 919         stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
 920         stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
 921         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
 922         stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
 923
 924         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
 925         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
 926         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
 927         stats->roc += E1000_READ_REG(hw, E1000_ROC);
 928         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
 929
 930         stats->tor += E1000_READ_REG(hw, E1000_TORH);
 931         stats->tot += E1000_READ_REG(hw, E1000_TOTH);
 932
 933         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
 934         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
 935         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
 936         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
 937         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
 938         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
 939         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
 940         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
 941         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
 942         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
 943
 944         /* Interrupt Counts */
 945
 946         if (hw->mac.type >= e1000_82571) {
 947                 stats->iac += E1000_READ_REG(hw, E1000_IAC);
 948                 stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
 949                 stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
 950                 stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
 951                 stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
 952                 stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
 953                 stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
 954                 stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
 955                 stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
 956         }
 957
 958         if (hw->mac.type >= e1000_82543) {
 959                 stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
 960                 stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
 961                 stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
 962                 stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
 963                 stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
 964                 stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
 965         }
 966
 967         if (rte_stats == NULL)
 968                 return;
 969
 970         /* Rx Errors */
 971         rte_stats->imissed = stats->mpc;
 972         rte_stats->ierrors = stats->crcerrs +
 973                              stats->rlec + stats->ruc + stats->roc +
 974                              stats->rxerrc + stats->algnerrc + stats->cexterr;
 975
 976         /* Tx Errors */
 977         rte_stats->oerrors = stats->ecol + stats->latecol;
 978
 979         rte_stats->ipackets = stats->gprc;
 980         rte_stats->opackets = stats->gptc;
 981         rte_stats->ibytes   = stats->gorc;
 982         rte_stats->obytes   = stats->gotc;
 983 }
 984
 985 static void
 986 eth_em_stats_reset(struct rte_eth_dev *dev)
 987 {
 988         struct e1000_hw_stats *hw_stats =
 989                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
 990
 991         /* HW registers are cleared on read */
 992         eth_em_stats_get(dev, NULL);
 993
 994         /* Reset software totals */
 995         memset(hw_stats, 0, sizeof(*hw_stats));
 996 }
 997
 998 static int
 999 eth_em_rx_queue_intr_enable(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id)
1000 {
1001         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1002
1003         em_rxq_intr_enable(hw);
1004         rte_intr_enable(&dev->pci_dev->intr_handle);
1005
1006         return 0;
1007 }
1008
1009 static int
1010 eth_em_rx_queue_intr_disable(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id)
1011 {
1012         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1013
1014         em_rxq_intr_disable(hw);
1015
1016         return 0;
1017 }
1018
1019 static uint32_t
1020 em_get_max_pktlen(const struct e1000_hw *hw)
1021 {
1022         switch (hw->mac.type) {
1023         case e1000_82571:
1024         case e1000_82572:
1025         case e1000_ich9lan:
1026         case e1000_ich10lan:
1027         case e1000_pch2lan:
1028         case e1000_pch_lpt:
1029         case e1000_82574:
1030         case e1000_80003es2lan: /* 9K Jumbo Frame size */
1031         case e1000_82583:
1032                 return 0x2412;
1033         case e1000_pchlan:
1034                 return 0x1000;
1035         /* Adapters that do not support jumbo frames */
1036         case e1000_ich8lan:
1037                 return ETHER_MAX_LEN;
1038         default:
1039                 return MAX_JUMBO_FRAME_SIZE;
1040         }
1041 }
1042
1043 static void
1044 eth_em_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1045 {
1046         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1047
1048         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
1049         dev_info->max_rx_pktlen = em_get_max_pktlen(hw);
1050         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
1051
1052         /*
1053          * Starting with 631xESB hw supports 2 TX/RX queues per port.
1054          * Unfortunatelly, all these nics have just one TX context.
1055          * So we have few choises for TX:
1056          * - Use just one TX queue.
1057          * - Allow cksum offload only for one TX queue.
1058          * - Don't allow TX cksum offload at all.
1059          * For now, option #1 was chosen.
1060          * To use second RX queue we have to use extended RX descriptor
1061          * (Multiple Receive Queues are mutually exclusive with UDP
1062          * fragmentation and are not supported when a legacy receive
1063          * descriptor format is used).
1064          * Which means separate RX routinies - as legacy nics (82540, 82545)
1065          * don't support extended RXD.
1066          * To avoid it we support just one RX queue for now (no RSS).
1067          */
1068
1069         dev_info->max_rx_queues = 1;
1070         dev_info->max_tx_queues = 1;
1071
1072         dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1073                 .nb_max = E1000_MAX_RING_DESC,
1074                 .nb_min = E1000_MIN_RING_DESC,
1075                 .nb_align = EM_RXD_ALIGN,
1076         };
1077
1078         dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1079                 .nb_max = E1000_MAX_RING_DESC,
1080                 .nb_min = E1000_MIN_RING_DESC,
1081                 .nb_align = EM_TXD_ALIGN,
1082         };
1083
1084         dev_info->speed_capa = ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
1085                         ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
1086                         ETH_LINK_SPEED_1G;
1087 }
1088
1089 /* return 0 means link status changed, -1 means not changed */
1090 static int
1091 eth_em_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1092 {
1093         struct e1000_hw *hw =
1094                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1095         struct rte_eth_link link, old;
1096         int link_check, count;
1097
1098         link_check = 0;
1099         hw->mac.get_link_status = 1;
1100
1101         /* possible wait-to-complete in up to 9 seconds */
1102         for (count = 0; count < EM_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
1103                 /* Read the real link status */
1104                 switch (hw->phy.media_type) {
1105                 case e1000_media_type_copper:
1106                         /* Do the work to read phy */
1107                         e1000_check_for_link(hw);
1108                         link_check = !hw->mac.get_link_status;
1109                         break;
1110
1111                 case e1000_media_type_fiber:
1112                         e1000_check_for_link(hw);
1113                         link_check = (E1000_READ_REG(hw, E1000_STATUS) &
1114                                         E1000_STATUS_LU);
1115                         break;
1116
1117                 case e1000_media_type_internal_serdes:
1118                         e1000_check_for_link(hw);
1119                         link_check = hw->mac.serdes_has_link;
1120                         break;
1121
1122                 default:
1123                         break;
1124                 }
1125                 if (link_check || wait_to_complete == 0)
1126                         break;
1127                 rte_delay_ms(EM_LINK_UPDATE_CHECK_INTERVAL);
1128         }
1129         memset(&link, 0, sizeof(link));
1130         rte_em_dev_atomic_read_link_status(dev, &link);
1131         old = link;
1132
1133         /* Now we check if a transition has happened */
1134         if (link_check && (link.link_status == ETH_LINK_DOWN)) {
1135                 uint16_t duplex, speed;
1136                 hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
1137                 link.link_duplex = (duplex == FULL_DUPLEX) ?
1138                                 ETH_LINK_FULL_DUPLEX :
1139                                 ETH_LINK_HALF_DUPLEX;
1140                 link.link_speed = speed;
1141                 link.link_status = ETH_LINK_UP;
1142                 link.link_autoneg = !(dev->data->dev_conf.link_speeds &
1143                                 ETH_LINK_SPEED_FIXED);
1144         } else if (!link_check && (link.link_status == ETH_LINK_UP)) {
1145                 link.link_speed = 0;
1146                 link.link_duplex = ETH_LINK_HALF_DUPLEX;
1147                 link.link_status = ETH_LINK_DOWN;
1148                 link.link_autoneg = ETH_LINK_SPEED_FIXED;
1149         }
1150         rte_em_dev_atomic_write_link_status(dev, &link);
1151
1152         /* not changed */
1153         if (old.link_status == link.link_status)
1154                 return -1;
1155
1156         /* changed */
1157         return 0;
1158 }
1159
1160 /*
1161  * em_hw_control_acquire sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1162  * For ASF and Pass Through versions of f/w this means
1163  * that the driver is loaded. For AMT version type f/w
1164  * this means that the network i/f is open.
1165  */
1166 static void
1167 em_hw_control_acquire(struct e1000_hw *hw)
1168 {
1169         uint32_t ctrl_ext, swsm;
1170
1171         /* Let firmware know the driver has taken over */
1172         if (hw->mac.type == e1000_82573) {
1173                 swsm = E1000_READ_REG(hw, E1000_SWSM);
1174                 E1000_WRITE_REG(hw, E1000_SWSM, swsm | E1000_SWSM_DRV_LOAD);
1175
1176         } else {
1177                 ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1178                 E1000_WRITE_REG(hw, E1000_CTRL_EXT,
1179                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1180         }
1181 }
1182
1183 /*
1184  * em_hw_control_release resets {CTRL_EXTT|FWSM}:DRV_LOAD bit.
1185  * For ASF and Pass Through versions of f/w this means that the
1186  * driver is no longer loaded. For AMT versions of the
1187  * f/w this means that the network i/f is closed.
1188  */
1189 static void
1190 em_hw_control_release(struct e1000_hw *hw)
1191 {
1192         uint32_t ctrl_ext, swsm;
1193
1194         /* Let firmware taken over control of h/w */
1195         if (hw->mac.type == e1000_82573) {
1196                 swsm = E1000_READ_REG(hw, E1000_SWSM);
1197                 E1000_WRITE_REG(hw, E1000_SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1198         } else {
1199                 ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1200                 E1000_WRITE_REG(hw, E1000_CTRL_EXT,
1201                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1202         }
1203 }
1204
1205 /*
1206  * Bit of a misnomer, what this really means is
1207  * to enable OS management of the system... aka
1208  * to disable special hardware management features.
1209  */
1210 static void
1211 em_init_manageability(struct e1000_hw *hw)
1212 {
1213         if (e1000_enable_mng_pass_thru(hw)) {
1214                 uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
1215                 uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
1216
1217                 /* disable hardware interception of ARP */
1218                 manc &= ~(E1000_MANC_ARP_EN);
1219
1220                 /* enable receiving management packets to the host */
1221                 manc |= E1000_MANC_EN_MNG2HOST;
1222                 manc2h |= 1 << 5;  /* Mng Port 623 */
1223                 manc2h |= 1 << 6;  /* Mng Port 664 */
1224                 E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
1225                 E1000_WRITE_REG(hw, E1000_MANC, manc);
1226         }
1227 }
1228
1229 /*
1230  * Give control back to hardware management
1231  * controller if there is one.
1232  */
1233 static void
1234 em_release_manageability(struct e1000_hw *hw)
1235 {
1236         uint32_t manc;
1237
1238         if (e1000_enable_mng_pass_thru(hw)) {
1239                 manc = E1000_READ_REG(hw, E1000_MANC);
1240
1241                 /* re-enable hardware interception of ARP */
1242                 manc |= E1000_MANC_ARP_EN;
1243                 manc &= ~E1000_MANC_EN_MNG2HOST;
1244
1245                 E1000_WRITE_REG(hw, E1000_MANC, manc);
1246         }
1247 }
1248
1249 static void
1250 eth_em_promiscuous_enable(struct rte_eth_dev *dev)
1251 {
1252         struct e1000_hw *hw =
1253                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1254         uint32_t rctl;
1255
1256         rctl = E1000_READ_REG(hw, E1000_RCTL);
1257         rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1258         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1259 }
1260
1261 static void
1262 eth_em_promiscuous_disable(struct rte_eth_dev *dev)
1263 {
1264         struct e1000_hw *hw =
1265                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1266         uint32_t rctl;
1267
1268         rctl = E1000_READ_REG(hw, E1000_RCTL);
1269         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_SBP);
1270         if (dev->data->all_multicast == 1)
1271                 rctl |= E1000_RCTL_MPE;
1272         else
1273                 rctl &= (~E1000_RCTL_MPE);
1274         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1275 }
1276
1277 static void
1278 eth_em_allmulticast_enable(struct rte_eth_dev *dev)
1279 {
1280         struct e1000_hw *hw =
1281                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1282         uint32_t rctl;
1283
1284         rctl = E1000_READ_REG(hw, E1000_RCTL);
1285         rctl |= E1000_RCTL_MPE;
1286         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1287 }
1288
1289 static void
1290 eth_em_allmulticast_disable(struct rte_eth_dev *dev)
1291 {
1292         struct e1000_hw *hw =
1293                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1294         uint32_t rctl;
1295
1296         if (dev->data->promiscuous == 1)
1297                 return; /* must remain in all_multicast mode */
1298         rctl = E1000_READ_REG(hw, E1000_RCTL);
1299         rctl &= (~E1000_RCTL_MPE);
1300         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1301 }
1302
1303 static int
1304 eth_em_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1305 {
1306         struct e1000_hw *hw =
1307                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1308         struct e1000_vfta * shadow_vfta =
1309                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
1310         uint32_t vfta;
1311         uint32_t vid_idx;
1312         uint32_t vid_bit;
1313
1314         vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
1315                               E1000_VFTA_ENTRY_MASK);
1316         vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
1317         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
1318         if (on)
1319                 vfta |= vid_bit;
1320         else
1321                 vfta &= ~vid_bit;
1322         E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
1323
1324         /* update local VFTA copy */
1325         shadow_vfta->vfta[vid_idx] = vfta;
1326
1327         return 0;
1328 }
1329
1330 static void
1331 em_vlan_hw_filter_disable(struct rte_eth_dev *dev)
1332 {
1333         struct e1000_hw *hw =
1334                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1335         uint32_t reg;
1336
1337         /* Filter Table Disable */
1338         reg = E1000_READ_REG(hw, E1000_RCTL);
1339         reg &= ~E1000_RCTL_CFIEN;
1340         reg &= ~E1000_RCTL_VFE;
1341         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1342 }
1343
1344 static void
1345 em_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1346 {
1347         struct e1000_hw *hw =
1348                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1349         struct e1000_vfta * shadow_vfta =
1350                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
1351         uint32_t reg;
1352         int i;
1353
1354         /* Filter Table Enable, CFI not used for packet acceptance */
1355         reg = E1000_READ_REG(hw, E1000_RCTL);
1356         reg &= ~E1000_RCTL_CFIEN;
1357         reg |= E1000_RCTL_VFE;
1358         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1359
1360         /* restore vfta from local copy */
1361         for (i = 0; i < IGB_VFTA_SIZE; i++)
1362                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
1363 }
1364
1365 static void
1366 em_vlan_hw_strip_disable(struct rte_eth_dev *dev)
1367 {
1368         struct e1000_hw *hw =
1369                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1370         uint32_t reg;
1371
1372         /* VLAN Mode Disable */
1373         reg = E1000_READ_REG(hw, E1000_CTRL);
1374         reg &= ~E1000_CTRL_VME;
1375         E1000_WRITE_REG(hw, E1000_CTRL, reg);
1376
1377 }
1378
1379 static void
1380 em_vlan_hw_strip_enable(struct rte_eth_dev *dev)
1381 {
1382         struct e1000_hw *hw =
1383                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1384         uint32_t reg;
1385
1386         /* VLAN Mode Enable */
1387         reg = E1000_READ_REG(hw, E1000_CTRL);
1388         reg |= E1000_CTRL_VME;
1389         E1000_WRITE_REG(hw, E1000_CTRL, reg);
1390 }
1391
1392 static void
1393 eth_em_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1394 {
1395         if(mask & ETH_VLAN_STRIP_MASK){
1396                 if (dev->data->dev_conf.rxmode.hw_vlan_strip)
1397                         em_vlan_hw_strip_enable(dev);
1398                 else
1399                         em_vlan_hw_strip_disable(dev);
1400         }
1401
1402         if(mask & ETH_VLAN_FILTER_MASK){
1403                 if (dev->data->dev_conf.rxmode.hw_vlan_filter)
1404                         em_vlan_hw_filter_enable(dev);
1405                 else
1406                         em_vlan_hw_filter_disable(dev);
1407         }
1408 }
1409
1410 /*
1411  * It enables the interrupt mask and then enable the interrupt.
1412  *
1413  * @param dev
1414  *  Pointer to struct rte_eth_dev.
1415  *
1416  * @return
1417  *  - On success, zero.
1418  *  - On failure, a negative value.
1419  */
1420 static int
1421 eth_em_interrupt_setup(struct rte_eth_dev *dev)
1422 {
1423         uint32_t regval;
1424         struct e1000_hw *hw =
1425                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1426
1427         /* clear interrupt */
1428         E1000_READ_REG(hw, E1000_ICR);
1429         regval = E1000_READ_REG(hw, E1000_IMS);
1430         E1000_WRITE_REG(hw, E1000_IMS, regval | E1000_ICR_LSC);
1431         return 0;
1432 }
1433
1434 /*
1435  * It clears the interrupt causes and enables the interrupt.
1436  * It will be called once only during nic initialized.
1437  *
1438  * @param dev
1439  *  Pointer to struct rte_eth_dev.
1440  *
1441  * @return
1442  *  - On success, zero.
1443  *  - On failure, a negative value.
1444  */
1445 static int
1446 eth_em_rxq_interrupt_setup(struct rte_eth_dev *dev)
1447 {
1448         struct e1000_hw *hw =
1449         E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1450
1451         E1000_READ_REG(hw, E1000_ICR);
1452         em_rxq_intr_enable(hw);
1453         return 0;
1454 }
1455
1456 /*
1457  * It enable receive packet interrupt.
1458  * @param hw
1459  * Pointer to struct e1000_hw
1460  *
1461  * @return
1462  */
1463 static void
1464 em_rxq_intr_enable(struct e1000_hw *hw)
1465 {
1466         E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_RXT0);
1467         E1000_WRITE_FLUSH(hw);
1468 }
1469
1470 /*
1471  * It disabled lsc interrupt.
1472  * @param hw
1473  * Pointer to struct e1000_hw
1474  *
1475  * @return
1476  */
1477 static void
1478 em_lsc_intr_disable(struct e1000_hw *hw)
1479 {
1480         E1000_WRITE_REG(hw, E1000_IMC, E1000_IMS_LSC);
1481         E1000_WRITE_FLUSH(hw);
1482 }
1483
1484 /*
1485  * It disabled receive packet interrupt.
1486  * @param hw
1487  * Pointer to struct e1000_hw
1488  *
1489  * @return
1490  */
1491 static void
1492 em_rxq_intr_disable(struct e1000_hw *hw)
1493 {
1494         E1000_READ_REG(hw, E1000_ICR);
1495         E1000_WRITE_REG(hw, E1000_IMC, E1000_IMS_RXT0);
1496         E1000_WRITE_FLUSH(hw);
1497 }
1498
1499 /*
1500  * It reads ICR and gets interrupt causes, check it and set a bit flag
1501  * to update link status.
1502  *
1503  * @param dev
1504  *  Pointer to struct rte_eth_dev.
1505  *
1506  * @return
1507  *  - On success, zero.
1508  *  - On failure, a negative value.
1509  */
1510 static int
1511 eth_em_interrupt_get_status(struct rte_eth_dev *dev)
1512 {
1513         uint32_t icr;
1514         struct e1000_hw *hw =
1515                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1516         struct e1000_interrupt *intr =
1517                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1518
1519         /* read-on-clear nic registers here */
1520         icr = E1000_READ_REG(hw, E1000_ICR);
1521         if (icr & E1000_ICR_LSC) {
1522                 intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
1523         }
1524
1525         return 0;
1526 }
1527
1528 /*
1529  * It executes link_update after knowing an interrupt is prsent.
1530  *
1531  * @param dev
1532  *  Pointer to struct rte_eth_dev.
1533  *
1534  * @return
1535  *  - On success, zero.
1536  *  - On failure, a negative value.
1537  */
1538 static int
1539 eth_em_interrupt_action(struct rte_eth_dev *dev)
1540 {
1541         struct e1000_hw *hw =
1542                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1543         struct e1000_interrupt *intr =
1544                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1545         uint32_t tctl, rctl;
1546         struct rte_eth_link link;
1547         int ret;
1548
1549         if (!(intr->flags & E1000_FLAG_NEED_LINK_UPDATE))
1550                 return -1;
1551
1552         intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
1553         rte_intr_enable(&(dev->pci_dev->intr_handle));
1554
1555         /* set get_link_status to check register later */
1556         hw->mac.get_link_status = 1;
1557         ret = eth_em_link_update(dev, 0);
1558
1559         /* check if link has changed */
1560         if (ret < 0)
1561                 return 0;
1562
1563         memset(&link, 0, sizeof(link));
1564         rte_em_dev_atomic_read_link_status(dev, &link);
1565         if (link.link_status) {
1566                 PMD_INIT_LOG(INFO, " Port %d: Link Up - speed %u Mbps - %s",
1567                              dev->data->port_id, (unsigned)link.link_speed,
1568                              link.link_duplex == ETH_LINK_FULL_DUPLEX ?
1569                              "full-duplex" : "half-duplex");
1570         } else {
1571                 PMD_INIT_LOG(INFO, " Port %d: Link Down", dev->data->port_id);
1572         }
1573         PMD_INIT_LOG(DEBUG, "PCI Address: %04d:%02d:%02d:%d",
1574                      dev->pci_dev->addr.domain, dev->pci_dev->addr.bus,
1575                      dev->pci_dev->addr.devid, dev->pci_dev->addr.function);
1576
1577         tctl = E1000_READ_REG(hw, E1000_TCTL);
1578         rctl = E1000_READ_REG(hw, E1000_RCTL);
1579         if (link.link_status) {
1580                 /* enable Tx/Rx */
1581                 tctl |= E1000_TCTL_EN;
1582                 rctl |= E1000_RCTL_EN;
1583         } else {
1584                 /* disable Tx/Rx */
1585                 tctl &= ~E1000_TCTL_EN;
1586                 rctl &= ~E1000_RCTL_EN;
1587         }
1588         E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1589         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1590         E1000_WRITE_FLUSH(hw);
1591
1592         return 0;
1593 }
1594
1595 /**
1596  * Interrupt handler which shall be registered at first.
1597  *
1598  * @param handle
1599  *  Pointer to interrupt handle.
1600  * @param param
1601  *  The address of parameter (struct rte_eth_dev *) regsitered before.
1602  *
1603  * @return
1604  *  void
1605  */
1606 static void
1607 eth_em_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1608                                                         void *param)
1609 {
1610         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1611
1612         eth_em_interrupt_get_status(dev);
1613         eth_em_interrupt_action(dev);
1614         _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
1615 }
1616
1617 static int
1618 eth_em_led_on(struct rte_eth_dev *dev)
1619 {
1620         struct e1000_hw *hw;
1621
1622         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1623         return e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
1624 }
1625
1626 static int
1627 eth_em_led_off(struct rte_eth_dev *dev)
1628 {
1629         struct e1000_hw *hw;
1630
1631         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1632         return e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP;
1633 }
1634
1635 static int
1636 eth_em_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1637 {
1638         struct e1000_hw *hw;
1639         uint32_t ctrl;
1640         int tx_pause;
1641         int rx_pause;
1642
1643         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1644         fc_conf->pause_time = hw->fc.pause_time;
1645         fc_conf->high_water = hw->fc.high_water;
1646         fc_conf->low_water = hw->fc.low_water;
1647         fc_conf->send_xon = hw->fc.send_xon;
1648         fc_conf->autoneg = hw->mac.autoneg;
1649
1650         /*
1651          * Return rx_pause and tx_pause status according to actual setting of
1652          * the TFCE and RFCE bits in the CTRL register.
1653          */
1654         ctrl = E1000_READ_REG(hw, E1000_CTRL);
1655         if (ctrl & E1000_CTRL_TFCE)
1656                 tx_pause = 1;
1657         else
1658                 tx_pause = 0;
1659
1660         if (ctrl & E1000_CTRL_RFCE)
1661                 rx_pause = 1;
1662         else
1663                 rx_pause = 0;
1664
1665         if (rx_pause && tx_pause)
1666                 fc_conf->mode = RTE_FC_FULL;
1667         else if (rx_pause)
1668                 fc_conf->mode = RTE_FC_RX_PAUSE;
1669         else if (tx_pause)
1670                 fc_conf->mode = RTE_FC_TX_PAUSE;
1671         else
1672                 fc_conf->mode = RTE_FC_NONE;
1673
1674         return 0;
1675 }
1676
1677 static int
1678 eth_em_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1679 {
1680         struct e1000_hw *hw;
1681         int err;
1682         enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
1683                 e1000_fc_none,
1684                 e1000_fc_rx_pause,
1685                 e1000_fc_tx_pause,
1686                 e1000_fc_full
1687         };
1688         uint32_t rx_buf_size;
1689         uint32_t max_high_water;
1690         uint32_t rctl;
1691
1692         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1693         if (fc_conf->autoneg != hw->mac.autoneg)
1694                 return -ENOTSUP;
1695         rx_buf_size = em_get_rx_buffer_size(hw);
1696         PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
1697
1698         /* At least reserve one Ethernet frame for watermark */
1699         max_high_water = rx_buf_size - ETHER_MAX_LEN;
1700         if ((fc_conf->high_water > max_high_water) ||
1701             (fc_conf->high_water < fc_conf->low_water)) {
1702                 PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
1703                 PMD_INIT_LOG(ERR, "high water must <= 0x%x", max_high_water);
1704                 return -EINVAL;
1705         }
1706
1707         hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
1708         hw->fc.pause_time     = fc_conf->pause_time;
1709         hw->fc.high_water     = fc_conf->high_water;
1710         hw->fc.low_water      = fc_conf->low_water;
1711         hw->fc.send_xon       = fc_conf->send_xon;
1712
1713         err = e1000_setup_link_generic(hw);
1714         if (err == E1000_SUCCESS) {
1715
1716                 /* check if we want to forward MAC frames - driver doesn't have native
1717                  * capability to do that, so we'll write the registers ourselves */
1718
1719                 rctl = E1000_READ_REG(hw, E1000_RCTL);
1720
1721                 /* set or clear MFLCN.PMCF bit depending on configuration */
1722                 if (fc_conf->mac_ctrl_frame_fwd != 0)
1723                         rctl |= E1000_RCTL_PMCF;
1724                 else
1725                         rctl &= ~E1000_RCTL_PMCF;
1726
1727                 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1728                 E1000_WRITE_FLUSH(hw);
1729
1730                 return 0;
1731         }
1732
1733         PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
1734         return -EIO;
1735 }
1736
1737 static void
1738 eth_em_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1739                 uint32_t index, __rte_unused uint32_t pool)
1740 {
1741         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1742
1743         e1000_rar_set(hw, mac_addr->addr_bytes, index);
1744 }
1745
1746 static void
1747 eth_em_rar_clear(struct rte_eth_dev *dev, uint32_t index)
1748 {
1749         uint8_t addr[ETHER_ADDR_LEN];
1750         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1751
1752         memset(addr, 0, sizeof(addr));
1753
1754         e1000_rar_set(hw, addr, index);
1755 }
1756
1757 static int
1758 eth_em_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1759 {
1760         struct rte_eth_dev_info dev_info;
1761         struct e1000_hw *hw;
1762         uint32_t frame_size;
1763         uint32_t rctl;
1764
1765         eth_em_infos_get(dev, &dev_info);
1766         frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + VLAN_TAG_SIZE;
1767
1768         /* check that mtu is within the allowed range */
1769         if ((mtu < ETHER_MIN_MTU) || (frame_size > dev_info.max_rx_pktlen))
1770                 return -EINVAL;
1771
1772         /* refuse mtu that requires the support of scattered packets when this
1773          * feature has not been enabled before. */
1774         if (!dev->data->scattered_rx &&
1775             frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)
1776                 return -EINVAL;
1777
1778         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1779         rctl = E1000_READ_REG(hw, E1000_RCTL);
1780
1781         /* switch to jumbo mode if needed */
1782         if (frame_size > ETHER_MAX_LEN) {
1783                 dev->data->dev_conf.rxmode.jumbo_frame = 1;
1784                 rctl |= E1000_RCTL_LPE;
1785         } else {
1786                 dev->data->dev_conf.rxmode.jumbo_frame = 0;
1787                 rctl &= ~E1000_RCTL_LPE;
1788         }
1789         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1790
1791         /* update max frame size */
1792         dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1793         return 0;
1794 }
1795
1796 static int
1797 eth_em_set_mc_addr_list(struct rte_eth_dev *dev,
1798                         struct ether_addr *mc_addr_set,
1799                         uint32_t nb_mc_addr)
1800 {
1801         struct e1000_hw *hw;
1802
1803         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1804         e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
1805         return 0;
1806 }
1807
1808 RTE_PMD_REGISTER_PCI(net_e1000_em, rte_em_pmd.pci_drv);
1809 RTE_PMD_REGISTER_PCI_TABLE(net_e1000_em, pci_id_em_map);
1810 RTE_PMD_REGISTER_KMOD_DEP(net_e1000_em, "* igb_uio | uio_pci_generic | vfio");