drivers/net/e1000/igb_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 #include "igb_regs.h"
  59
  60 /*
  61  * Default values for port configuration
  62  */
  63 #define IGB_DEFAULT_RX_FREE_THRESH  32
  64 #define IGB_DEFAULT_RX_PTHRESH      8
  65 #define IGB_DEFAULT_RX_HTHRESH      8
  66 #define IGB_DEFAULT_RX_WTHRESH      0
  67
  68 #define IGB_DEFAULT_TX_PTHRESH      32
  69 #define IGB_DEFAULT_TX_HTHRESH      0
  70 #define IGB_DEFAULT_TX_WTHRESH      0
  71
  72 #define IGB_HKEY_MAX_INDEX 10
  73
  74 /* Bit shift and mask */
  75 #define IGB_4_BIT_WIDTH  (CHAR_BIT / 2)
  76 #define IGB_4_BIT_MASK   RTE_LEN2MASK(IGB_4_BIT_WIDTH, uint8_t)
  77 #define IGB_8_BIT_WIDTH  CHAR_BIT
  78 #define IGB_8_BIT_MASK   UINT8_MAX
  79
  80 /* Additional timesync values. */
  81 #define E1000_ETQF_FILTER_1588 3
  82 #define E1000_TIMINCA_INCVALUE 16000000
  83 #define E1000_TIMINCA_INIT     ((0x02 << E1000_TIMINCA_16NS_SHIFT) \
  84                                 | E1000_TIMINCA_INCVALUE)
  85
  86 static int  eth_igb_configure(struct rte_eth_dev *dev);
  87 static int  eth_igb_start(struct rte_eth_dev *dev);
  88 static void eth_igb_stop(struct rte_eth_dev *dev);
  89 static void eth_igb_close(struct rte_eth_dev *dev);
  90 static void eth_igb_promiscuous_enable(struct rte_eth_dev *dev);
  91 static void eth_igb_promiscuous_disable(struct rte_eth_dev *dev);
  92 static void eth_igb_allmulticast_enable(struct rte_eth_dev *dev);
  93 static void eth_igb_allmulticast_disable(struct rte_eth_dev *dev);
  94 static int  eth_igb_link_update(struct rte_eth_dev *dev,
  95                                 int wait_to_complete);
  96 static void eth_igb_stats_get(struct rte_eth_dev *dev,
  97                                 struct rte_eth_stats *rte_stats);
  98 static void eth_igb_stats_reset(struct rte_eth_dev *dev);
  99 static void eth_igb_infos_get(struct rte_eth_dev *dev,
 100                               struct rte_eth_dev_info *dev_info);
 101 static void eth_igbvf_infos_get(struct rte_eth_dev *dev,
 102                                 struct rte_eth_dev_info *dev_info);
 103 static int  eth_igb_flow_ctrl_get(struct rte_eth_dev *dev,
 104                                 struct rte_eth_fc_conf *fc_conf);
 105 static int  eth_igb_flow_ctrl_set(struct rte_eth_dev *dev,
 106                                 struct rte_eth_fc_conf *fc_conf);
 107 static int eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev);
 108 #ifdef RTE_NEXT_ABI
 109 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev);
 110 #endif
 111 static int eth_igb_interrupt_get_status(struct rte_eth_dev *dev);
 112 static int eth_igb_interrupt_action(struct rte_eth_dev *dev);
 113 static void eth_igb_interrupt_handler(struct rte_intr_handle *handle,
 114                                                         void *param);
 115 static int  igb_hardware_init(struct e1000_hw *hw);
 116 static void igb_hw_control_acquire(struct e1000_hw *hw);
 117 static void igb_hw_control_release(struct e1000_hw *hw);
 118 static void igb_init_manageability(struct e1000_hw *hw);
 119 static void igb_release_manageability(struct e1000_hw *hw);
 120
 121 static int  eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
 122
 123 static int eth_igb_vlan_filter_set(struct rte_eth_dev *dev,
 124                 uint16_t vlan_id, int on);
 125 static void eth_igb_vlan_tpid_set(struct rte_eth_dev *dev, uint16_t tpid_id);
 126 static void eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask);
 127
 128 static void igb_vlan_hw_filter_enable(struct rte_eth_dev *dev);
 129 static void igb_vlan_hw_filter_disable(struct rte_eth_dev *dev);
 130 static void igb_vlan_hw_strip_enable(struct rte_eth_dev *dev);
 131 static void igb_vlan_hw_strip_disable(struct rte_eth_dev *dev);
 132 static void igb_vlan_hw_extend_enable(struct rte_eth_dev *dev);
 133 static void igb_vlan_hw_extend_disable(struct rte_eth_dev *dev);
 134
 135 static int eth_igb_led_on(struct rte_eth_dev *dev);
 136 static int eth_igb_led_off(struct rte_eth_dev *dev);
 137
 138 static void igb_intr_disable(struct e1000_hw *hw);
 139 static int  igb_get_rx_buffer_size(struct e1000_hw *hw);
 140 static void eth_igb_rar_set(struct rte_eth_dev *dev,
 141                 struct ether_addr *mac_addr,
 142                 uint32_t index, uint32_t pool);
 143 static void eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index);
 144 static void eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
 145                 struct ether_addr *addr);
 146
 147 static void igbvf_intr_disable(struct e1000_hw *hw);
 148 static int igbvf_dev_configure(struct rte_eth_dev *dev);
 149 static int igbvf_dev_start(struct rte_eth_dev *dev);
 150 static void igbvf_dev_stop(struct rte_eth_dev *dev);
 151 static void igbvf_dev_close(struct rte_eth_dev *dev);
 152 static int eth_igbvf_link_update(struct e1000_hw *hw);
 153 static void eth_igbvf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats);
 154 static void eth_igbvf_stats_reset(struct rte_eth_dev *dev);
 155 static int igbvf_vlan_filter_set(struct rte_eth_dev *dev,
 156                 uint16_t vlan_id, int on);
 157 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on);
 158 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on);
 159 static void igbvf_default_mac_addr_set(struct rte_eth_dev *dev,
 160                 struct ether_addr *addr);
 161 static int igbvf_get_reg_length(struct rte_eth_dev *dev);
 162 static int igbvf_get_regs(struct rte_eth_dev *dev,
 163                 struct rte_dev_reg_info *regs);
 164
 165 static int eth_igb_rss_reta_update(struct rte_eth_dev *dev,
 166                                    struct rte_eth_rss_reta_entry64 *reta_conf,
 167                                    uint16_t reta_size);
 168 static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
 169                                   struct rte_eth_rss_reta_entry64 *reta_conf,
 170                                   uint16_t reta_size);
 171
 172 static int eth_igb_syn_filter_set(struct rte_eth_dev *dev,
 173                         struct rte_eth_syn_filter *filter,
 174                         bool add);
 175 static int eth_igb_syn_filter_get(struct rte_eth_dev *dev,
 176                         struct rte_eth_syn_filter *filter);
 177 static int eth_igb_syn_filter_handle(struct rte_eth_dev *dev,
 178                         enum rte_filter_op filter_op,
 179                         void *arg);
 180 static int igb_add_2tuple_filter(struct rte_eth_dev *dev,
 181                         struct rte_eth_ntuple_filter *ntuple_filter);
 182 static int igb_remove_2tuple_filter(struct rte_eth_dev *dev,
 183                         struct rte_eth_ntuple_filter *ntuple_filter);
 184 static int eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
 185                         struct rte_eth_flex_filter *filter,
 186                         bool add);
 187 static int eth_igb_get_flex_filter(struct rte_eth_dev *dev,
 188                         struct rte_eth_flex_filter *filter);
 189 static int eth_igb_flex_filter_handle(struct rte_eth_dev *dev,
 190                         enum rte_filter_op filter_op,
 191                         void *arg);
 192 static int igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
 193                         struct rte_eth_ntuple_filter *ntuple_filter);
 194 static int igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
 195                         struct rte_eth_ntuple_filter *ntuple_filter);
 196 static int igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
 197                         struct rte_eth_ntuple_filter *filter,
 198                         bool add);
 199 static int igb_get_ntuple_filter(struct rte_eth_dev *dev,
 200                         struct rte_eth_ntuple_filter *filter);
 201 static int igb_ntuple_filter_handle(struct rte_eth_dev *dev,
 202                                 enum rte_filter_op filter_op,
 203                                 void *arg);
 204 static int igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
 205                         struct rte_eth_ethertype_filter *filter,
 206                         bool add);
 207 static int igb_ethertype_filter_handle(struct rte_eth_dev *dev,
 208                                 enum rte_filter_op filter_op,
 209                                 void *arg);
 210 static int igb_get_ethertype_filter(struct rte_eth_dev *dev,
 211                         struct rte_eth_ethertype_filter *filter);
 212 static int eth_igb_filter_ctrl(struct rte_eth_dev *dev,
 213                      enum rte_filter_type filter_type,
 214                      enum rte_filter_op filter_op,
 215                      void *arg);
 216 static int eth_igb_get_reg_length(struct rte_eth_dev *dev);
 217 static int eth_igb_get_regs(struct rte_eth_dev *dev,
 218                 struct rte_dev_reg_info *regs);
 219 static int eth_igb_get_eeprom_length(struct rte_eth_dev *dev);
 220 static int eth_igb_get_eeprom(struct rte_eth_dev *dev,
 221                 struct rte_dev_eeprom_info *eeprom);
 222 static int eth_igb_set_eeprom(struct rte_eth_dev *dev,
 223                 struct rte_dev_eeprom_info *eeprom);
 224 static int eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
 225                                     struct ether_addr *mc_addr_set,
 226                                     uint32_t nb_mc_addr);
 227 static int igb_timesync_enable(struct rte_eth_dev *dev);
 228 static int igb_timesync_disable(struct rte_eth_dev *dev);
 229 static int igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
 230                                           struct timespec *timestamp,
 231                                           uint32_t flags);
 232 static int igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
 233                                           struct timespec *timestamp);
 234 #ifdef RTE_NEXT_ABI
 235 static int eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev,
 236                                         uint16_t queue_id);
 237 static int eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev,
 238                                          uint16_t queue_id);
 239 static void eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
 240                                        uint8_t queue, uint8_t msix_vector);
 241 static void eth_igb_write_ivar(struct e1000_hw *hw, uint8_t msix_vector,
 242                                uint8_t index, uint8_t offset);
 243 #endif
 244 static void eth_igb_configure_msix_intr(struct rte_eth_dev *dev);
 245
 246 /*
 247  * Define VF Stats MACRO for Non "cleared on read" register
 248  */
 249 #define UPDATE_VF_STAT(reg, last, cur)            \
 250 {                                                 \
 251         u32 latest = E1000_READ_REG(hw, reg);     \
 252         cur += latest - last;                     \
 253         last = latest;                            \
 254 }
 255
 256
 257 #define IGB_FC_PAUSE_TIME 0x0680
 258 #define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
 259 #define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
 260
 261 #define IGBVF_PMD_NAME "rte_igbvf_pmd"     /* PMD name */
 262
 263 static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
 264
 265 /*
 266  * The set of PCI devices this driver supports
 267  */
 268 static const struct rte_pci_id pci_id_igb_map[] = {
 269
 270 #define RTE_PCI_DEV_ID_DECL_IGB(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 271 #include "rte_pci_dev_ids.h"
 272
 273 {0},
 274 };
 275
 276 /*
 277  * The set of PCI devices this driver supports (for 82576&I350 VF)
 278  */
 279 static const struct rte_pci_id pci_id_igbvf_map[] = {
 280
 281 #define RTE_PCI_DEV_ID_DECL_IGBVF(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 282 #include "rte_pci_dev_ids.h"
 283
 284 {0},
 285 };
 286
 287 static const struct eth_dev_ops eth_igb_ops = {
 288         .dev_configure        = eth_igb_configure,
 289         .dev_start            = eth_igb_start,
 290         .dev_stop             = eth_igb_stop,
 291         .dev_close            = eth_igb_close,
 292         .promiscuous_enable   = eth_igb_promiscuous_enable,
 293         .promiscuous_disable  = eth_igb_promiscuous_disable,
 294         .allmulticast_enable  = eth_igb_allmulticast_enable,
 295         .allmulticast_disable = eth_igb_allmulticast_disable,
 296         .link_update          = eth_igb_link_update,
 297         .stats_get            = eth_igb_stats_get,
 298         .stats_reset          = eth_igb_stats_reset,
 299         .dev_infos_get        = eth_igb_infos_get,
 300         .mtu_set              = eth_igb_mtu_set,
 301         .vlan_filter_set      = eth_igb_vlan_filter_set,
 302         .vlan_tpid_set        = eth_igb_vlan_tpid_set,
 303         .vlan_offload_set     = eth_igb_vlan_offload_set,
 304         .rx_queue_setup       = eth_igb_rx_queue_setup,
 305 #ifdef RTE_NEXT_ABI
 306         .rx_queue_intr_enable = eth_igb_rx_queue_intr_enable,
 307         .rx_queue_intr_disable = eth_igb_rx_queue_intr_disable,
 308 #endif
 309         .rx_queue_release     = eth_igb_rx_queue_release,
 310         .rx_queue_count       = eth_igb_rx_queue_count,
 311         .rx_descriptor_done   = eth_igb_rx_descriptor_done,
 312         .tx_queue_setup       = eth_igb_tx_queue_setup,
 313         .tx_queue_release     = eth_igb_tx_queue_release,
 314         .dev_led_on           = eth_igb_led_on,
 315         .dev_led_off          = eth_igb_led_off,
 316         .flow_ctrl_get        = eth_igb_flow_ctrl_get,
 317         .flow_ctrl_set        = eth_igb_flow_ctrl_set,
 318         .mac_addr_add         = eth_igb_rar_set,
 319         .mac_addr_remove      = eth_igb_rar_clear,
 320         .mac_addr_set         = eth_igb_default_mac_addr_set,
 321         .reta_update          = eth_igb_rss_reta_update,
 322         .reta_query           = eth_igb_rss_reta_query,
 323         .rss_hash_update      = eth_igb_rss_hash_update,
 324         .rss_hash_conf_get    = eth_igb_rss_hash_conf_get,
 325         .filter_ctrl          = eth_igb_filter_ctrl,
 326         .set_mc_addr_list     = eth_igb_set_mc_addr_list,
 327         .timesync_enable      = igb_timesync_enable,
 328         .timesync_disable     = igb_timesync_disable,
 329         .timesync_read_rx_timestamp = igb_timesync_read_rx_timestamp,
 330         .timesync_read_tx_timestamp = igb_timesync_read_tx_timestamp,
 331         .get_reg_length       = eth_igb_get_reg_length,
 332         .get_reg              = eth_igb_get_regs,
 333         .get_eeprom_length    = eth_igb_get_eeprom_length,
 334         .get_eeprom           = eth_igb_get_eeprom,
 335         .set_eeprom           = eth_igb_set_eeprom,
 336 };
 337
 338 /*
 339  * dev_ops for virtual function, bare necessities for basic vf
 340  * operation have been implemented
 341  */
 342 static const struct eth_dev_ops igbvf_eth_dev_ops = {
 343         .dev_configure        = igbvf_dev_configure,
 344         .dev_start            = igbvf_dev_start,
 345         .dev_stop             = igbvf_dev_stop,
 346         .dev_close            = igbvf_dev_close,
 347         .link_update          = eth_igb_link_update,
 348         .stats_get            = eth_igbvf_stats_get,
 349         .stats_reset          = eth_igbvf_stats_reset,
 350         .vlan_filter_set      = igbvf_vlan_filter_set,
 351         .dev_infos_get        = eth_igbvf_infos_get,
 352         .rx_queue_setup       = eth_igb_rx_queue_setup,
 353         .rx_queue_release     = eth_igb_rx_queue_release,
 354         .tx_queue_setup       = eth_igb_tx_queue_setup,
 355         .tx_queue_release     = eth_igb_tx_queue_release,
 356         .set_mc_addr_list     = eth_igb_set_mc_addr_list,
 357         .mac_addr_set         = igbvf_default_mac_addr_set,
 358         .get_reg_length       = igbvf_get_reg_length,
 359         .get_reg              = igbvf_get_regs,
 360 };
 361
 362 /**
 363  * Atomically reads the link status information from global
 364  * structure rte_eth_dev.
 365  *
 366  * @param dev
 367  *   - Pointer to the structure rte_eth_dev to read from.
 368  *   - Pointer to the buffer to be saved with the link status.
 369  *
 370  * @return
 371  *   - On success, zero.
 372  *   - On failure, negative value.
 373  */
 374 static inline int
 375 rte_igb_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 376                                 struct rte_eth_link *link)
 377 {
 378         struct rte_eth_link *dst = link;
 379         struct rte_eth_link *src = &(dev->data->dev_link);
 380
 381         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 382                                         *(uint64_t *)src) == 0)
 383                 return -1;
 384
 385         return 0;
 386 }
 387
 388 /**
 389  * Atomically writes the link status information into global
 390  * structure rte_eth_dev.
 391  *
 392  * @param dev
 393  *   - Pointer to the structure rte_eth_dev to read from.
 394  *   - Pointer to the buffer to be saved with the link status.
 395  *
 396  * @return
 397  *   - On success, zero.
 398  *   - On failure, negative value.
 399  */
 400 static inline int
 401 rte_igb_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 402                                 struct rte_eth_link *link)
 403 {
 404         struct rte_eth_link *dst = &(dev->data->dev_link);
 405         struct rte_eth_link *src = link;
 406
 407         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 408                                         *(uint64_t *)src) == 0)
 409                 return -1;
 410
 411         return 0;
 412 }
 413
 414 static inline void
 415 igb_intr_enable(struct rte_eth_dev *dev)
 416 {
 417         struct e1000_interrupt *intr =
 418                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
 419         struct e1000_hw *hw =
 420                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 421
 422         E1000_WRITE_REG(hw, E1000_IMS, intr->mask);
 423         E1000_WRITE_FLUSH(hw);
 424 }
 425
 426 static void
 427 igb_intr_disable(struct e1000_hw *hw)
 428 {
 429         E1000_WRITE_REG(hw, E1000_IMC, ~0);
 430         E1000_WRITE_FLUSH(hw);
 431 }
 432
 433 static inline int32_t
 434 igb_pf_reset_hw(struct e1000_hw *hw)
 435 {
 436         uint32_t ctrl_ext;
 437         int32_t status;
 438
 439         status = e1000_reset_hw(hw);
 440
 441         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
 442         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
 443         ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
 444         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
 445         E1000_WRITE_FLUSH(hw);
 446
 447         return status;
 448 }
 449
 450 static void
 451 igb_identify_hardware(struct rte_eth_dev *dev)
 452 {
 453         struct e1000_hw *hw =
 454                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 455
 456         hw->vendor_id = dev->pci_dev->id.vendor_id;
 457         hw->device_id = dev->pci_dev->id.device_id;
 458         hw->subsystem_vendor_id = dev->pci_dev->id.subsystem_vendor_id;
 459         hw->subsystem_device_id = dev->pci_dev->id.subsystem_device_id;
 460
 461         e1000_set_mac_type(hw);
 462
 463         /* need to check if it is a vf device below */
 464 }
 465
 466 static int
 467 igb_reset_swfw_lock(struct e1000_hw *hw)
 468 {
 469         int ret_val;
 470
 471         /*
 472          * Do mac ops initialization manually here, since we will need
 473          * some function pointers set by this call.
 474          */
 475         ret_val = e1000_init_mac_params(hw);
 476         if (ret_val)
 477                 return ret_val;
 478
 479         /*
 480          * SMBI lock should not fail in this early stage. If this is the case,
 481          * it is due to an improper exit of the application.
 482          * So force the release of the faulty lock.
 483          */
 484         if (e1000_get_hw_semaphore_generic(hw) < 0) {
 485                 PMD_DRV_LOG(DEBUG, "SMBI lock released");
 486         }
 487         e1000_put_hw_semaphore_generic(hw);
 488
 489         if (hw->mac.ops.acquire_swfw_sync != NULL) {
 490                 uint16_t mask;
 491
 492                 /*
 493                  * Phy lock should not fail in this early stage. If this is the case,
 494                  * it is due to an improper exit of the application.
 495                  * So force the release of the faulty lock.
 496                  */
 497                 mask = E1000_SWFW_PHY0_SM << hw->bus.func;
 498                 if (hw->bus.func > E1000_FUNC_1)
 499                         mask <<= 2;
 500                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
 501                         PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
 502                                     hw->bus.func);
 503                 }
 504                 hw->mac.ops.release_swfw_sync(hw, mask);
 505
 506                 /*
 507                  * This one is more tricky since it is common to all ports; but
 508                  * swfw_sync retries last long enough (1s) to be almost sure that if
 509                  * lock can not be taken it is due to an improper lock of the
 510                  * semaphore.
 511                  */
 512                 mask = E1000_SWFW_EEP_SM;
 513                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
 514                         PMD_DRV_LOG(DEBUG, "SWFW common locks released");
 515                 }
 516                 hw->mac.ops.release_swfw_sync(hw, mask);
 517         }
 518
 519         return E1000_SUCCESS;
 520 }
 521
 522 static int
 523 eth_igb_dev_init(struct rte_eth_dev *eth_dev)
 524 {
 525         int error = 0;
 526         struct rte_pci_device *pci_dev;
 527         struct e1000_hw *hw =
 528                 E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 529         struct e1000_vfta * shadow_vfta =
 530                 E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
 531         struct e1000_filter_info *filter_info =
 532                 E1000_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
 533         struct e1000_adapter *adapter =
 534                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
 535
 536         uint32_t ctrl_ext;
 537
 538         pci_dev = eth_dev->pci_dev;
 539         eth_dev->dev_ops = &eth_igb_ops;
 540         eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
 541         eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
 542
 543         /* for secondary processes, we don't initialise any further as primary
 544          * has already done this work. Only check we don't need a different
 545          * RX function */
 546         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
 547                 if (eth_dev->data->scattered_rx)
 548                         eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
 549                 return 0;
 550         }
 551
 552         hw->hw_addr= (void *)pci_dev->mem_resource[0].addr;
 553
 554         igb_identify_hardware(eth_dev);
 555         if (e1000_setup_init_funcs(hw, FALSE) != E1000_SUCCESS) {
 556                 error = -EIO;
 557                 goto err_late;
 558         }
 559
 560         e1000_get_bus_info(hw);
 561
 562         /* Reset any pending lock */
 563         if (igb_reset_swfw_lock(hw) != E1000_SUCCESS) {
 564                 error = -EIO;
 565                 goto err_late;
 566         }
 567
 568         /* Finish initialization */
 569         if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS) {
 570                 error = -EIO;
 571                 goto err_late;
 572         }
 573
 574         hw->mac.autoneg = 1;
 575         hw->phy.autoneg_wait_to_complete = 0;
 576         hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
 577
 578         /* Copper options */
 579         if (hw->phy.media_type == e1000_media_type_copper) {
 580                 hw->phy.mdix = 0; /* AUTO_ALL_MODES */
 581                 hw->phy.disable_polarity_correction = 0;
 582                 hw->phy.ms_type = e1000_ms_hw_default;
 583         }
 584
 585         /*
 586          * Start from a known state, this is important in reading the nvm
 587          * and mac from that.
 588          */
 589         igb_pf_reset_hw(hw);
 590
 591         /* Make sure we have a good EEPROM before we read from it */
 592         if (e1000_validate_nvm_checksum(hw) < 0) {
 593                 /*
 594                  * Some PCI-E parts fail the first check due to
 595                  * the link being in sleep state, call it again,
 596                  * if it fails a second time its a real issue.
 597                  */
 598                 if (e1000_validate_nvm_checksum(hw) < 0) {
 599                         PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
 600                         error = -EIO;
 601                         goto err_late;
 602                 }
 603         }
 604
 605         /* Read the permanent MAC address out of the EEPROM */
 606         if (e1000_read_mac_addr(hw) != 0) {
 607                 PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
 608                 error = -EIO;
 609                 goto err_late;
 610         }
 611
 612         /* Allocate memory for storing MAC addresses */
 613         eth_dev->data->mac_addrs = rte_zmalloc("e1000",
 614                 ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
 615         if (eth_dev->data->mac_addrs == NULL) {
 616                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
 617                                                 "store MAC addresses",
 618                                 ETHER_ADDR_LEN * hw->mac.rar_entry_count);
 619                 error = -ENOMEM;
 620                 goto err_late;
 621         }
 622
 623         /* Copy the permanent MAC address */
 624         ether_addr_copy((struct ether_addr *)hw->mac.addr, &eth_dev->data->mac_addrs[0]);
 625
 626         /* initialize the vfta */
 627         memset(shadow_vfta, 0, sizeof(*shadow_vfta));
 628
 629         /* Now initialize the hardware */
 630         if (igb_hardware_init(hw) != 0) {
 631                 PMD_INIT_LOG(ERR, "Hardware initialization failed");
 632                 rte_free(eth_dev->data->mac_addrs);
 633                 eth_dev->data->mac_addrs = NULL;
 634                 error = -ENODEV;
 635                 goto err_late;
 636         }
 637         hw->mac.get_link_status = 1;
 638         adapter->stopped = 0;
 639
 640         /* Indicate SOL/IDER usage */
 641         if (e1000_check_reset_block(hw) < 0) {
 642                 PMD_INIT_LOG(ERR, "PHY reset is blocked due to"
 643                                         "SOL/IDER session");
 644         }
 645
 646         /* initialize PF if max_vfs not zero */
 647         igb_pf_host_init(eth_dev);
 648
 649         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
 650         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
 651         ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
 652         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
 653         E1000_WRITE_FLUSH(hw);
 654
 655         PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
 656                      eth_dev->data->port_id, pci_dev->id.vendor_id,
 657                      pci_dev->id.device_id);
 658
 659         /* enable support intr */
 660         igb_intr_enable(eth_dev);
 661
 662         TAILQ_INIT(&filter_info->flex_list);
 663         filter_info->flex_mask = 0;
 664         TAILQ_INIT(&filter_info->twotuple_list);
 665         filter_info->twotuple_mask = 0;
 666         TAILQ_INIT(&filter_info->fivetuple_list);
 667         filter_info->fivetuple_mask = 0;
 668
 669         return 0;
 670
 671 err_late:
 672         igb_hw_control_release(hw);
 673
 674         return (error);
 675 }
 676
 677 static int
 678 eth_igb_dev_uninit(struct rte_eth_dev *eth_dev)
 679 {
 680         struct rte_pci_device *pci_dev;
 681         struct e1000_hw *hw;
 682         struct e1000_adapter *adapter =
 683                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
 684
 685         PMD_INIT_FUNC_TRACE();
 686
 687         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 688                 return -EPERM;
 689
 690         hw = E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 691         pci_dev = eth_dev->pci_dev;
 692
 693         if (adapter->stopped == 0)
 694                 eth_igb_close(eth_dev);
 695
 696         eth_dev->dev_ops = NULL;
 697         eth_dev->rx_pkt_burst = NULL;
 698         eth_dev->tx_pkt_burst = NULL;
 699
 700         /* Reset any pending lock */
 701         igb_reset_swfw_lock(hw);
 702
 703         rte_free(eth_dev->data->mac_addrs);
 704         eth_dev->data->mac_addrs = NULL;
 705
 706         /* uninitialize PF if max_vfs not zero */
 707         igb_pf_host_uninit(eth_dev);
 708
 709         /* disable uio intr before callback unregister */
 710         rte_intr_disable(&(pci_dev->intr_handle));
 711         rte_intr_callback_unregister(&(pci_dev->intr_handle),
 712                 eth_igb_interrupt_handler, (void *)eth_dev);
 713
 714         return 0;
 715 }
 716
 717 /*
 718  * Virtual Function device init
 719  */
 720 static int
 721 eth_igbvf_dev_init(struct rte_eth_dev *eth_dev)
 722 {
 723         struct rte_pci_device *pci_dev;
 724         struct e1000_adapter *adapter =
 725                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
 726         struct e1000_hw *hw =
 727                 E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 728         int diag;
 729
 730         PMD_INIT_FUNC_TRACE();
 731
 732         eth_dev->dev_ops = &igbvf_eth_dev_ops;
 733         eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
 734         eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
 735
 736         /* for secondary processes, we don't initialise any further as primary
 737          * has already done this work. Only check we don't need a different
 738          * RX function */
 739         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
 740                 if (eth_dev->data->scattered_rx)
 741                         eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
 742                 return 0;
 743         }
 744
 745         pci_dev = eth_dev->pci_dev;
 746
 747         hw->device_id = pci_dev->id.device_id;
 748         hw->vendor_id = pci_dev->id.vendor_id;
 749         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
 750         adapter->stopped = 0;
 751
 752         /* Initialize the shared code (base driver) */
 753         diag = e1000_setup_init_funcs(hw, TRUE);
 754         if (diag != 0) {
 755                 PMD_INIT_LOG(ERR, "Shared code init failed for igbvf: %d",
 756                         diag);
 757                 return -EIO;
 758         }
 759
 760         /* init_mailbox_params */
 761         hw->mbx.ops.init_params(hw);
 762
 763         /* Disable the interrupts for VF */
 764         igbvf_intr_disable(hw);
 765
 766         diag = hw->mac.ops.reset_hw(hw);
 767
 768         /* Allocate memory for storing MAC addresses */
 769         eth_dev->data->mac_addrs = rte_zmalloc("igbvf", ETHER_ADDR_LEN *
 770                 hw->mac.rar_entry_count, 0);
 771         if (eth_dev->data->mac_addrs == NULL) {
 772                 PMD_INIT_LOG(ERR,
 773                         "Failed to allocate %d bytes needed to store MAC "
 774                         "addresses",
 775                         ETHER_ADDR_LEN * hw->mac.rar_entry_count);
 776                 return -ENOMEM;
 777         }
 778
 779         /* Copy the permanent MAC address */
 780         ether_addr_copy((struct ether_addr *) hw->mac.perm_addr,
 781                         &eth_dev->data->mac_addrs[0]);
 782
 783         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x "
 784                      "mac.type=%s",
 785                      eth_dev->data->port_id, pci_dev->id.vendor_id,
 786                      pci_dev->id.device_id, "igb_mac_82576_vf");
 787
 788         return 0;
 789 }
 790
 791 static int
 792 eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
 793 {
 794         struct e1000_adapter *adapter =
 795                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
 796
 797         PMD_INIT_FUNC_TRACE();
 798
 799         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 800                 return -EPERM;
 801
 802         if (adapter->stopped == 0)
 803                 igbvf_dev_close(eth_dev);
 804
 805         eth_dev->dev_ops = NULL;
 806         eth_dev->rx_pkt_burst = NULL;
 807         eth_dev->tx_pkt_burst = NULL;
 808
 809         rte_free(eth_dev->data->mac_addrs);
 810         eth_dev->data->mac_addrs = NULL;
 811
 812         return 0;
 813 }
 814
 815 static struct eth_driver rte_igb_pmd = {
 816         .pci_drv = {
 817                 .name = "rte_igb_pmd",
 818                 .id_table = pci_id_igb_map,
 819                 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
 820                         RTE_PCI_DRV_DETACHABLE,
 821         },
 822         .eth_dev_init = eth_igb_dev_init,
 823         .eth_dev_uninit = eth_igb_dev_uninit,
 824         .dev_private_size = sizeof(struct e1000_adapter),
 825 };
 826
 827 /*
 828  * virtual function driver struct
 829  */
 830 static struct eth_driver rte_igbvf_pmd = {
 831         .pci_drv = {
 832                 .name = "rte_igbvf_pmd",
 833                 .id_table = pci_id_igbvf_map,
 834                 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_DETACHABLE,
 835         },
 836         .eth_dev_init = eth_igbvf_dev_init,
 837         .eth_dev_uninit = eth_igbvf_dev_uninit,
 838         .dev_private_size = sizeof(struct e1000_adapter),
 839 };
 840
 841 static int
 842 rte_igb_pmd_init(const char *name __rte_unused, const char *params __rte_unused)
 843 {
 844         rte_eth_driver_register(&rte_igb_pmd);
 845         return 0;
 846 }
 847
 848 static void
 849 igb_vmdq_vlan_hw_filter_enable(struct rte_eth_dev *dev)
 850 {
 851         struct e1000_hw *hw =
 852                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 853         /* RCTL: enable VLAN filter since VMDq always use VLAN filter */
 854         uint32_t rctl = E1000_READ_REG(hw, E1000_RCTL);
 855         rctl |= E1000_RCTL_VFE;
 856         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
 857 }
 858
 859 /*
 860  * VF Driver initialization routine.
 861  * Invoked one at EAL init time.
 862  * Register itself as the [Virtual Poll Mode] Driver of PCI IGB devices.
 863  */
 864 static int
 865 rte_igbvf_pmd_init(const char *name __rte_unused, const char *params __rte_unused)
 866 {
 867         PMD_INIT_FUNC_TRACE();
 868
 869         rte_eth_driver_register(&rte_igbvf_pmd);
 870         return (0);
 871 }
 872
 873 static int
 874 eth_igb_configure(struct rte_eth_dev *dev)
 875 {
 876         struct e1000_interrupt *intr =
 877                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
 878
 879         PMD_INIT_FUNC_TRACE();
 880         intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
 881         PMD_INIT_FUNC_TRACE();
 882
 883         return (0);
 884 }
 885
 886 static int
 887 eth_igb_start(struct rte_eth_dev *dev)
 888 {
 889         struct e1000_hw *hw =
 890                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 891         struct e1000_adapter *adapter =
 892                 E1000_DEV_PRIVATE(dev->data->dev_private);
 893         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
 894         int ret, mask;
 895 #ifdef RTE_NEXT_ABI
 896         uint32_t intr_vector = 0;
 897 #endif
 898         uint32_t ctrl_ext;
 899
 900         PMD_INIT_FUNC_TRACE();
 901
 902         /* Power up the phy. Needed to make the link go Up */
 903         e1000_power_up_phy(hw);
 904
 905         /*
 906          * Packet Buffer Allocation (PBA)
 907          * Writing PBA sets the receive portion of the buffer
 908          * the remainder is used for the transmit buffer.
 909          */
 910         if (hw->mac.type == e1000_82575) {
 911                 uint32_t pba;
 912
 913                 pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
 914                 E1000_WRITE_REG(hw, E1000_PBA, pba);
 915         }
 916
 917         /* Put the address into the Receive Address Array */
 918         e1000_rar_set(hw, hw->mac.addr, 0);
 919
 920         /* Initialize the hardware */
 921         if (igb_hardware_init(hw)) {
 922                 PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
 923                 return (-EIO);
 924         }
 925         adapter->stopped = 0;
 926
 927         E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN << 16 | ETHER_TYPE_VLAN);
 928
 929         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
 930         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
 931         ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
 932         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
 933         E1000_WRITE_FLUSH(hw);
 934
 935         /* configure PF module if SRIOV enabled */
 936         igb_pf_host_configure(dev);
 937
 938 #ifdef RTE_NEXT_ABI
 939         /* check and configure queue intr-vector mapping */
 940         if (dev->data->dev_conf.intr_conf.rxq != 0)
 941                 intr_vector = dev->data->nb_rx_queues;
 942
 943         if (rte_intr_efd_enable(intr_handle, intr_vector))
 944                 return -1;
 945
 946         if (rte_intr_dp_is_en(intr_handle)) {
 947                 intr_handle->intr_vec =
 948                         rte_zmalloc("intr_vec",
 949                                     dev->data->nb_rx_queues * sizeof(int), 0);
 950                 if (intr_handle->intr_vec == NULL) {
 951                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
 952                                      " intr_vec\n", dev->data->nb_rx_queues);
 953                         return -ENOMEM;
 954                 }
 955         }
 956 #endif
 957
 958         /* confiugre msix for rx interrupt */
 959         eth_igb_configure_msix_intr(dev);
 960
 961         /* Configure for OS presence */
 962         igb_init_manageability(hw);
 963
 964         eth_igb_tx_init(dev);
 965
 966         /* This can fail when allocating mbufs for descriptor rings */
 967         ret = eth_igb_rx_init(dev);
 968         if (ret) {
 969                 PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
 970                 igb_dev_clear_queues(dev);
 971                 return ret;
 972         }
 973
 974         e1000_clear_hw_cntrs_base_generic(hw);
 975
 976         /*
 977          * VLAN Offload Settings
 978          */
 979         mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK | \
 980                         ETH_VLAN_EXTEND_MASK;
 981         eth_igb_vlan_offload_set(dev, mask);
 982
 983         if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_VMDQ_ONLY) {
 984                 /* Enable VLAN filter since VMDq always use VLAN filter */
 985                 igb_vmdq_vlan_hw_filter_enable(dev);
 986         }
 987
 988         if ((hw->mac.type == e1000_82576) || (hw->mac.type == e1000_82580) ||
 989                 (hw->mac.type == e1000_i350) || (hw->mac.type == e1000_i210) ||
 990                 (hw->mac.type == e1000_i211)) {
 991                 /* Configure EITR with the maximum possible value (0xFFFF) */
 992                 E1000_WRITE_REG(hw, E1000_EITR(0), 0xFFFF);
 993         }
 994
 995         /* Setup link speed and duplex */
 996         switch (dev->data->dev_conf.link_speed) {
 997         case ETH_LINK_SPEED_AUTONEG:
 998                 if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
 999                         hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
1000                 else if (dev->data->dev_conf.link_duplex == ETH_LINK_HALF_DUPLEX)
1001                         hw->phy.autoneg_advertised = E1000_ALL_HALF_DUPLEX;
1002                 else if (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX)
1003                         hw->phy.autoneg_advertised = E1000_ALL_FULL_DUPLEX;
1004                 else
1005                         goto error_invalid_config;
1006                 break;
1007         case ETH_LINK_SPEED_10:
1008                 if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
1009                         hw->phy.autoneg_advertised = E1000_ALL_10_SPEED;
1010                 else if (dev->data->dev_conf.link_duplex == ETH_LINK_HALF_DUPLEX)
1011                         hw->phy.autoneg_advertised = ADVERTISE_10_HALF;
1012                 else if (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX)
1013                         hw->phy.autoneg_advertised = ADVERTISE_10_FULL;
1014                 else
1015                         goto error_invalid_config;
1016                 break;
1017         case ETH_LINK_SPEED_100:
1018                 if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
1019                         hw->phy.autoneg_advertised = E1000_ALL_100_SPEED;
1020                 else if (dev->data->dev_conf.link_duplex == ETH_LINK_HALF_DUPLEX)
1021                         hw->phy.autoneg_advertised = ADVERTISE_100_HALF;
1022                 else if (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX)
1023                         hw->phy.autoneg_advertised = ADVERTISE_100_FULL;
1024                 else
1025                         goto error_invalid_config;
1026                 break;
1027         case ETH_LINK_SPEED_1000:
1028                 if ((dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX) ||
1029                                 (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX))
1030                         hw->phy.autoneg_advertised = ADVERTISE_1000_FULL;
1031                 else
1032                         goto error_invalid_config;
1033                 break;
1034         case ETH_LINK_SPEED_10000:
1035         default:
1036                 goto error_invalid_config;
1037         }
1038         e1000_setup_link(hw);
1039
1040         /* check if lsc interrupt feature is enabled */
1041         if (dev->data->dev_conf.intr_conf.lsc != 0) {
1042                 if (rte_intr_allow_others(intr_handle)) {
1043                         rte_intr_callback_register(intr_handle,
1044                                                    eth_igb_interrupt_handler,
1045                                                    (void *)dev);
1046                         eth_igb_lsc_interrupt_setup(dev);
1047                 } else
1048                         PMD_INIT_LOG(INFO, "lsc won't enable because of"
1049                                      " no intr multiplex\n");
1050         }
1051
1052 #ifdef RTE_NEXT_ABI
1053         /* check if rxq interrupt is enabled */
1054         if (dev->data->dev_conf.intr_conf.rxq != 0)
1055                 eth_igb_rxq_interrupt_setup(dev);
1056 #endif
1057
1058         /* enable uio/vfio intr/eventfd mapping */
1059         rte_intr_enable(intr_handle);
1060
1061         /* resume enabled intr since hw reset */
1062         igb_intr_enable(dev);
1063
1064         PMD_INIT_LOG(DEBUG, "<<");
1065
1066         return (0);
1067
1068 error_invalid_config:
1069         PMD_INIT_LOG(ERR, "Invalid link_speed/link_duplex (%u/%u) for port %u",
1070                      dev->data->dev_conf.link_speed,
1071                      dev->data->dev_conf.link_duplex, dev->data->port_id);
1072         igb_dev_clear_queues(dev);
1073         return (-EINVAL);
1074 }
1075
1076 /*********************************************************************
1077  *
1078  *  This routine disables all traffic on the adapter by issuing a
1079  *  global reset on the MAC.
1080  *
1081  **********************************************************************/
1082 static void
1083 eth_igb_stop(struct rte_eth_dev *dev)
1084 {
1085         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1086         struct e1000_filter_info *filter_info =
1087                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
1088         struct rte_eth_link link;
1089         struct e1000_flex_filter *p_flex;
1090         struct e1000_5tuple_filter *p_5tuple, *p_5tuple_next;
1091         struct e1000_2tuple_filter *p_2tuple, *p_2tuple_next;
1092         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
1093
1094         igb_intr_disable(hw);
1095
1096         /* disable intr eventfd mapping */
1097         rte_intr_disable(intr_handle);
1098
1099         igb_pf_reset_hw(hw);
1100         E1000_WRITE_REG(hw, E1000_WUC, 0);
1101
1102         /* Set bit for Go Link disconnect */
1103         if (hw->mac.type >= e1000_82580) {
1104                 uint32_t phpm_reg;
1105
1106                 phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1107                 phpm_reg |= E1000_82580_PM_GO_LINKD;
1108                 E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1109         }
1110
1111         /* Power down the phy. Needed to make the link go Down */
1112         if (hw->phy.media_type == e1000_media_type_copper)
1113                 e1000_power_down_phy(hw);
1114         else
1115                 e1000_shutdown_fiber_serdes_link(hw);
1116
1117         igb_dev_clear_queues(dev);
1118
1119         /* clear the recorded link status */
1120         memset(&link, 0, sizeof(link));
1121         rte_igb_dev_atomic_write_link_status(dev, &link);
1122
1123         /* Remove all flex filters of the device */
1124         while ((p_flex = TAILQ_FIRST(&filter_info->flex_list))) {
1125                 TAILQ_REMOVE(&filter_info->flex_list, p_flex, entries);
1126                 rte_free(p_flex);
1127         }
1128         filter_info->flex_mask = 0;
1129
1130         /* Remove all ntuple filters of the device */
1131         for (p_5tuple = TAILQ_FIRST(&filter_info->fivetuple_list);
1132              p_5tuple != NULL; p_5tuple = p_5tuple_next) {
1133                 p_5tuple_next = TAILQ_NEXT(p_5tuple, entries);
1134                 TAILQ_REMOVE(&filter_info->fivetuple_list,
1135                              p_5tuple, entries);
1136                 rte_free(p_5tuple);
1137         }
1138         filter_info->fivetuple_mask = 0;
1139         for (p_2tuple = TAILQ_FIRST(&filter_info->twotuple_list);
1140              p_2tuple != NULL; p_2tuple = p_2tuple_next) {
1141                 p_2tuple_next = TAILQ_NEXT(p_2tuple, entries);
1142                 TAILQ_REMOVE(&filter_info->twotuple_list,
1143                              p_2tuple, entries);
1144                 rte_free(p_2tuple);
1145         }
1146         filter_info->twotuple_mask = 0;
1147
1148 #ifdef RTE_NEXT_ABI
1149         /* Clean datapath event and queue/vec mapping */
1150         rte_intr_efd_disable(intr_handle);
1151         if (intr_handle->intr_vec != NULL) {
1152                 rte_free(intr_handle->intr_vec);
1153                 intr_handle->intr_vec = NULL;
1154         }
1155 #endif
1156 }
1157
1158 static void
1159 eth_igb_close(struct rte_eth_dev *dev)
1160 {
1161         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1162         struct e1000_adapter *adapter =
1163                 E1000_DEV_PRIVATE(dev->data->dev_private);
1164         struct rte_eth_link link;
1165 #ifdef RTE_NEXT_ABI
1166         struct rte_pci_device *pci_dev;
1167 #endif
1168
1169         eth_igb_stop(dev);
1170         adapter->stopped = 1;
1171
1172         e1000_phy_hw_reset(hw);
1173         igb_release_manageability(hw);
1174         igb_hw_control_release(hw);
1175
1176         /* Clear bit for Go Link disconnect */
1177         if (hw->mac.type >= e1000_82580) {
1178                 uint32_t phpm_reg;
1179
1180                 phpm_reg = E1000_READ_REG(hw, E1000_82580_PHY_POWER_MGMT);
1181                 phpm_reg &= ~E1000_82580_PM_GO_LINKD;
1182                 E1000_WRITE_REG(hw, E1000_82580_PHY_POWER_MGMT, phpm_reg);
1183         }
1184
1185         igb_dev_free_queues(dev);
1186
1187 #ifdef RTE_NEXT_ABI
1188         pci_dev = dev->pci_dev;
1189         if (pci_dev->intr_handle.intr_vec) {
1190                 rte_free(pci_dev->intr_handle.intr_vec);
1191                 pci_dev->intr_handle.intr_vec = NULL;
1192         }
1193 #endif
1194
1195         memset(&link, 0, sizeof(link));
1196         rte_igb_dev_atomic_write_link_status(dev, &link);
1197 }
1198
1199 static int
1200 igb_get_rx_buffer_size(struct e1000_hw *hw)
1201 {
1202         uint32_t rx_buf_size;
1203         if (hw->mac.type == e1000_82576) {
1204                 rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xffff) << 10;
1205         } else if (hw->mac.type == e1000_82580 || hw->mac.type == e1000_i350) {
1206                 /* PBS needs to be translated according to a lookup table */
1207                 rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xf);
1208                 rx_buf_size = (uint32_t) e1000_rxpbs_adjust_82580(rx_buf_size);
1209                 rx_buf_size = (rx_buf_size << 10);
1210         } else if (hw->mac.type == e1000_i210 || hw->mac.type == e1000_i211) {
1211                 rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0x3f) << 10;
1212         } else {
1213                 rx_buf_size = (E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10;
1214         }
1215
1216         return rx_buf_size;
1217 }
1218
1219 /*********************************************************************
1220  *
1221  *  Initialize the hardware
1222  *
1223  **********************************************************************/
1224 static int
1225 igb_hardware_init(struct e1000_hw *hw)
1226 {
1227         uint32_t rx_buf_size;
1228         int diag;
1229
1230         /* Let the firmware know the OS is in control */
1231         igb_hw_control_acquire(hw);
1232
1233         /*
1234          * These parameters control the automatic generation (Tx) and
1235          * response (Rx) to Ethernet PAUSE frames.
1236          * - High water mark should allow for at least two standard size (1518)
1237          *   frames to be received after sending an XOFF.
1238          * - Low water mark works best when it is very near the high water mark.
1239          *   This allows the receiver to restart by sending XON when it has
1240          *   drained a bit. Here we use an arbitrary value of 1500 which will
1241          *   restart after one full frame is pulled from the buffer. There
1242          *   could be several smaller frames in the buffer and if so they will
1243          *   not trigger the XON until their total number reduces the buffer
1244          *   by 1500.
1245          * - The pause time is fairly large at 1000 x 512ns = 512 usec.
1246          */
1247         rx_buf_size = igb_get_rx_buffer_size(hw);
1248
1249         hw->fc.high_water = rx_buf_size - (ETHER_MAX_LEN * 2);
1250         hw->fc.low_water = hw->fc.high_water - 1500;
1251         hw->fc.pause_time = IGB_FC_PAUSE_TIME;
1252         hw->fc.send_xon = 1;
1253
1254         /* Set Flow control, use the tunable location if sane */
1255         if ((igb_fc_setting != e1000_fc_none) && (igb_fc_setting < 4))
1256                 hw->fc.requested_mode = igb_fc_setting;
1257         else
1258                 hw->fc.requested_mode = e1000_fc_none;
1259
1260         /* Issue a global reset */
1261         igb_pf_reset_hw(hw);
1262         E1000_WRITE_REG(hw, E1000_WUC, 0);
1263
1264         diag = e1000_init_hw(hw);
1265         if (diag < 0)
1266                 return (diag);
1267
1268         E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN << 16 | ETHER_TYPE_VLAN);
1269         e1000_get_phy_info(hw);
1270         e1000_check_for_link(hw);
1271
1272         return (0);
1273 }
1274
1275 /* This function is based on igb_update_stats_counters() in igb/if_igb.c */
1276 static void
1277 eth_igb_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1278 {
1279         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1280         struct e1000_hw_stats *stats =
1281                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1282         int pause_frames;
1283
1284         if(hw->phy.media_type == e1000_media_type_copper ||
1285             (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
1286                 stats->symerrs +=
1287                     E1000_READ_REG(hw,E1000_SYMERRS);
1288                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
1289         }
1290
1291         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
1292         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
1293         stats->scc += E1000_READ_REG(hw, E1000_SCC);
1294         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
1295
1296         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
1297         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
1298         stats->colc += E1000_READ_REG(hw, E1000_COLC);
1299         stats->dc += E1000_READ_REG(hw, E1000_DC);
1300         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
1301         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
1302         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
1303         /*
1304         ** For watchdog management we need to know if we have been
1305         ** paused during the last interval, so capture that here.
1306         */
1307         pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
1308         stats->xoffrxc += pause_frames;
1309         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
1310         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
1311         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
1312         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
1313         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
1314         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
1315         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
1316         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
1317         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
1318         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
1319         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
1320         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
1321
1322         /* For the 64-bit byte counters the low dword must be read first. */
1323         /* Both registers clear on the read of the high dword */
1324
1325         stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
1326         stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
1327         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
1328         stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
1329
1330         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
1331         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
1332         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
1333         stats->roc += E1000_READ_REG(hw, E1000_ROC);
1334         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
1335
1336         stats->tor += E1000_READ_REG(hw, E1000_TORH);
1337         stats->tot += E1000_READ_REG(hw, E1000_TOTH);
1338
1339         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
1340         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
1341         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
1342         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
1343         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
1344         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
1345         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
1346         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
1347         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
1348         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
1349
1350         /* Interrupt Counts */
1351
1352         stats->iac += E1000_READ_REG(hw, E1000_IAC);
1353         stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
1354         stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
1355         stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
1356         stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
1357         stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
1358         stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
1359         stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
1360         stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
1361
1362         /* Host to Card Statistics */
1363
1364         stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
1365         stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
1366         stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
1367         stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
1368         stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
1369         stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
1370         stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
1371         stats->hgorc += E1000_READ_REG(hw, E1000_HGORCL);
1372         stats->hgorc += ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32);
1373         stats->hgotc += E1000_READ_REG(hw, E1000_HGOTCL);
1374         stats->hgotc += ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32);
1375         stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
1376         stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
1377         stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
1378
1379         stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
1380         stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
1381         stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
1382         stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
1383         stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
1384         stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
1385
1386         if (rte_stats == NULL)
1387                 return;
1388
1389         /* Rx Errors */
1390         rte_stats->ibadcrc = stats->crcerrs;
1391         rte_stats->ibadlen = stats->rlec + stats->ruc + stats->roc;
1392         rte_stats->imissed = stats->mpc;
1393         rte_stats->ierrors = rte_stats->ibadcrc +
1394                              rte_stats->ibadlen +
1395                              rte_stats->imissed +
1396                              stats->rxerrc + stats->algnerrc + stats->cexterr;
1397
1398         /* Tx Errors */
1399         rte_stats->oerrors = stats->ecol + stats->latecol;
1400
1401         /* XON/XOFF pause frames */
1402         rte_stats->tx_pause_xon  = stats->xontxc;
1403         rte_stats->rx_pause_xon  = stats->xonrxc;
1404         rte_stats->tx_pause_xoff = stats->xofftxc;
1405         rte_stats->rx_pause_xoff = stats->xoffrxc;
1406
1407         rte_stats->ipackets = stats->gprc;
1408         rte_stats->opackets = stats->gptc;
1409         rte_stats->ibytes   = stats->gorc;
1410         rte_stats->obytes   = stats->gotc;
1411 }
1412
1413 static void
1414 eth_igb_stats_reset(struct rte_eth_dev *dev)
1415 {
1416         struct e1000_hw_stats *hw_stats =
1417                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1418
1419         /* HW registers are cleared on read */
1420         eth_igb_stats_get(dev, NULL);
1421
1422         /* Reset software totals */
1423         memset(hw_stats, 0, sizeof(*hw_stats));
1424 }
1425
1426 static void
1427 eth_igbvf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1428 {
1429         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1430         struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats*)
1431                           E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1432
1433         /* Good Rx packets, include VF loopback */
1434         UPDATE_VF_STAT(E1000_VFGPRC,
1435             hw_stats->last_gprc, hw_stats->gprc);
1436
1437         /* Good Rx octets, include VF loopback */
1438         UPDATE_VF_STAT(E1000_VFGORC,
1439             hw_stats->last_gorc, hw_stats->gorc);
1440
1441         /* Good Tx packets, include VF loopback */
1442         UPDATE_VF_STAT(E1000_VFGPTC,
1443             hw_stats->last_gptc, hw_stats->gptc);
1444
1445         /* Good Tx octets, include VF loopback */
1446         UPDATE_VF_STAT(E1000_VFGOTC,
1447             hw_stats->last_gotc, hw_stats->gotc);
1448
1449         /* Rx Multicst packets */
1450         UPDATE_VF_STAT(E1000_VFMPRC,
1451             hw_stats->last_mprc, hw_stats->mprc);
1452
1453         /* Good Rx loopback packets */
1454         UPDATE_VF_STAT(E1000_VFGPRLBC,
1455             hw_stats->last_gprlbc, hw_stats->gprlbc);
1456
1457         /* Good Rx loopback octets */
1458         UPDATE_VF_STAT(E1000_VFGORLBC,
1459             hw_stats->last_gorlbc, hw_stats->gorlbc);
1460
1461         /* Good Tx loopback packets */
1462         UPDATE_VF_STAT(E1000_VFGPTLBC,
1463             hw_stats->last_gptlbc, hw_stats->gptlbc);
1464
1465         /* Good Tx loopback octets */
1466         UPDATE_VF_STAT(E1000_VFGOTLBC,
1467             hw_stats->last_gotlbc, hw_stats->gotlbc);
1468
1469         if (rte_stats == NULL)
1470                 return;
1471
1472         rte_stats->ipackets = hw_stats->gprc;
1473         rte_stats->ibytes = hw_stats->gorc;
1474         rte_stats->opackets = hw_stats->gptc;
1475         rte_stats->obytes = hw_stats->gotc;
1476         rte_stats->imcasts = hw_stats->mprc;
1477         rte_stats->ilbpackets = hw_stats->gprlbc;
1478         rte_stats->ilbbytes = hw_stats->gorlbc;
1479         rte_stats->olbpackets = hw_stats->gptlbc;
1480         rte_stats->olbbytes = hw_stats->gotlbc;
1481
1482 }
1483
1484 static void
1485 eth_igbvf_stats_reset(struct rte_eth_dev *dev)
1486 {
1487         struct e1000_vf_stats *hw_stats = (struct e1000_vf_stats*)
1488                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1489
1490         /* Sync HW register to the last stats */
1491         eth_igbvf_stats_get(dev, NULL);
1492
1493         /* reset HW current stats*/
1494         memset(&hw_stats->gprc, 0, sizeof(*hw_stats) -
1495                offsetof(struct e1000_vf_stats, gprc));
1496
1497 }
1498
1499 static void
1500 eth_igb_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1501 {
1502         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1503
1504         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
1505         dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
1506         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
1507         dev_info->rx_offload_capa =
1508                 DEV_RX_OFFLOAD_VLAN_STRIP |
1509                 DEV_RX_OFFLOAD_IPV4_CKSUM |
1510                 DEV_RX_OFFLOAD_UDP_CKSUM  |
1511                 DEV_RX_OFFLOAD_TCP_CKSUM;
1512         dev_info->tx_offload_capa =
1513                 DEV_TX_OFFLOAD_VLAN_INSERT |
1514                 DEV_TX_OFFLOAD_IPV4_CKSUM  |
1515                 DEV_TX_OFFLOAD_UDP_CKSUM   |
1516                 DEV_TX_OFFLOAD_TCP_CKSUM   |
1517                 DEV_TX_OFFLOAD_SCTP_CKSUM;
1518
1519         switch (hw->mac.type) {
1520         case e1000_82575:
1521                 dev_info->max_rx_queues = 4;
1522                 dev_info->max_tx_queues = 4;
1523                 dev_info->max_vmdq_pools = 0;
1524                 break;
1525
1526         case e1000_82576:
1527                 dev_info->max_rx_queues = 16;
1528                 dev_info->max_tx_queues = 16;
1529                 dev_info->max_vmdq_pools = ETH_8_POOLS;
1530                 dev_info->vmdq_queue_num = 16;
1531                 break;
1532
1533         case e1000_82580:
1534                 dev_info->max_rx_queues = 8;
1535                 dev_info->max_tx_queues = 8;
1536                 dev_info->max_vmdq_pools = ETH_8_POOLS;
1537                 dev_info->vmdq_queue_num = 8;
1538                 break;
1539
1540         case e1000_i350:
1541                 dev_info->max_rx_queues = 8;
1542                 dev_info->max_tx_queues = 8;
1543                 dev_info->max_vmdq_pools = ETH_8_POOLS;
1544                 dev_info->vmdq_queue_num = 8;
1545                 break;
1546
1547         case e1000_i354:
1548                 dev_info->max_rx_queues = 8;
1549                 dev_info->max_tx_queues = 8;
1550                 break;
1551
1552         case e1000_i210:
1553                 dev_info->max_rx_queues = 4;
1554                 dev_info->max_tx_queues = 4;
1555                 dev_info->max_vmdq_pools = 0;
1556                 break;
1557
1558         case e1000_i211:
1559                 dev_info->max_rx_queues = 2;
1560                 dev_info->max_tx_queues = 2;
1561                 dev_info->max_vmdq_pools = 0;
1562                 break;
1563
1564         default:
1565                 /* Should not happen */
1566                 break;
1567         }
1568         dev_info->hash_key_size = IGB_HKEY_MAX_INDEX * sizeof(uint32_t);
1569         dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
1570         dev_info->flow_type_rss_offloads = IGB_RSS_OFFLOAD_ALL;
1571
1572         dev_info->default_rxconf = (struct rte_eth_rxconf) {
1573                 .rx_thresh = {
1574                         .pthresh = IGB_DEFAULT_RX_PTHRESH,
1575                         .hthresh = IGB_DEFAULT_RX_HTHRESH,
1576                         .wthresh = IGB_DEFAULT_RX_WTHRESH,
1577                 },
1578                 .rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
1579                 .rx_drop_en = 0,
1580         };
1581
1582         dev_info->default_txconf = (struct rte_eth_txconf) {
1583                 .tx_thresh = {
1584                         .pthresh = IGB_DEFAULT_TX_PTHRESH,
1585                         .hthresh = IGB_DEFAULT_TX_HTHRESH,
1586                         .wthresh = IGB_DEFAULT_TX_WTHRESH,
1587                 },
1588                 .txq_flags = 0,
1589         };
1590 }
1591
1592 static void
1593 eth_igbvf_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1594 {
1595         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1596
1597         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
1598         dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
1599         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
1600         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP |
1601                                 DEV_RX_OFFLOAD_IPV4_CKSUM |
1602                                 DEV_RX_OFFLOAD_UDP_CKSUM  |
1603                                 DEV_RX_OFFLOAD_TCP_CKSUM;
1604         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
1605                                 DEV_TX_OFFLOAD_IPV4_CKSUM  |
1606                                 DEV_TX_OFFLOAD_UDP_CKSUM   |
1607                                 DEV_TX_OFFLOAD_TCP_CKSUM   |
1608                                 DEV_TX_OFFLOAD_SCTP_CKSUM;
1609         switch (hw->mac.type) {
1610         case e1000_vfadapt:
1611                 dev_info->max_rx_queues = 2;
1612                 dev_info->max_tx_queues = 2;
1613                 break;
1614         case e1000_vfadapt_i350:
1615                 dev_info->max_rx_queues = 1;
1616                 dev_info->max_tx_queues = 1;
1617                 break;
1618         default:
1619                 /* Should not happen */
1620                 break;
1621         }
1622
1623         dev_info->default_rxconf = (struct rte_eth_rxconf) {
1624                 .rx_thresh = {
1625                         .pthresh = IGB_DEFAULT_RX_PTHRESH,
1626                         .hthresh = IGB_DEFAULT_RX_HTHRESH,
1627                         .wthresh = IGB_DEFAULT_RX_WTHRESH,
1628                 },
1629                 .rx_free_thresh = IGB_DEFAULT_RX_FREE_THRESH,
1630                 .rx_drop_en = 0,
1631         };
1632
1633         dev_info->default_txconf = (struct rte_eth_txconf) {
1634                 .tx_thresh = {
1635                         .pthresh = IGB_DEFAULT_TX_PTHRESH,
1636                         .hthresh = IGB_DEFAULT_TX_HTHRESH,
1637                         .wthresh = IGB_DEFAULT_TX_WTHRESH,
1638                 },
1639                 .txq_flags = 0,
1640         };
1641 }
1642
1643 /* return 0 means link status changed, -1 means not changed */
1644 static int
1645 eth_igb_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1646 {
1647         struct e1000_hw *hw =
1648                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1649         struct rte_eth_link link, old;
1650         int link_check, count;
1651
1652         link_check = 0;
1653         hw->mac.get_link_status = 1;
1654
1655         /* possible wait-to-complete in up to 9 seconds */
1656         for (count = 0; count < IGB_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
1657                 /* Read the real link status */
1658                 switch (hw->phy.media_type) {
1659                 case e1000_media_type_copper:
1660                         /* Do the work to read phy */
1661                         e1000_check_for_link(hw);
1662                         link_check = !hw->mac.get_link_status;
1663                         break;
1664
1665                 case e1000_media_type_fiber:
1666                         e1000_check_for_link(hw);
1667                         link_check = (E1000_READ_REG(hw, E1000_STATUS) &
1668                                       E1000_STATUS_LU);
1669                         break;
1670
1671                 case e1000_media_type_internal_serdes:
1672                         e1000_check_for_link(hw);
1673                         link_check = hw->mac.serdes_has_link;
1674                         break;
1675
1676                 /* VF device is type_unknown */
1677                 case e1000_media_type_unknown:
1678                         eth_igbvf_link_update(hw);
1679                         link_check = !hw->mac.get_link_status;
1680                         break;
1681
1682                 default:
1683                         break;
1684                 }
1685                 if (link_check || wait_to_complete == 0)
1686                         break;
1687                 rte_delay_ms(IGB_LINK_UPDATE_CHECK_INTERVAL);
1688         }
1689         memset(&link, 0, sizeof(link));
1690         rte_igb_dev_atomic_read_link_status(dev, &link);
1691         old = link;
1692
1693         /* Now we check if a transition has happened */
1694         if (link_check) {
1695                 hw->mac.ops.get_link_up_info(hw, &link.link_speed,
1696                                           &link.link_duplex);
1697                 link.link_status = 1;
1698         } else if (!link_check) {
1699                 link.link_speed = 0;
1700                 link.link_duplex = 0;
1701                 link.link_status = 0;
1702         }
1703         rte_igb_dev_atomic_write_link_status(dev, &link);
1704
1705         /* not changed */
1706         if (old.link_status == link.link_status)
1707                 return -1;
1708
1709         /* changed */
1710         return 0;
1711 }
1712
1713 /*
1714  * igb_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
1715  * For ASF and Pass Through versions of f/w this means
1716  * that the driver is loaded.
1717  */
1718 static void
1719 igb_hw_control_acquire(struct e1000_hw *hw)
1720 {
1721         uint32_t ctrl_ext;
1722
1723         /* Let firmware know the driver has taken over */
1724         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1725         E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1726 }
1727
1728 /*
1729  * igb_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
1730  * For ASF and Pass Through versions of f/w this means that the
1731  * driver is no longer loaded.
1732  */
1733 static void
1734 igb_hw_control_release(struct e1000_hw *hw)
1735 {
1736         uint32_t ctrl_ext;
1737
1738         /* Let firmware taken over control of h/w */
1739         ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1740         E1000_WRITE_REG(hw, E1000_CTRL_EXT,
1741                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1742 }
1743
1744 /*
1745  * Bit of a misnomer, what this really means is
1746  * to enable OS management of the system... aka
1747  * to disable special hardware management features.
1748  */
1749 static void
1750 igb_init_manageability(struct e1000_hw *hw)
1751 {
1752         if (e1000_enable_mng_pass_thru(hw)) {
1753                 uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
1754                 uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
1755
1756                 /* disable hardware interception of ARP */
1757                 manc &= ~(E1000_MANC_ARP_EN);
1758
1759                 /* enable receiving management packets to the host */
1760                 manc |= E1000_MANC_EN_MNG2HOST;
1761                 manc2h |= 1 << 5;  /* Mng Port 623 */
1762                 manc2h |= 1 << 6;  /* Mng Port 664 */
1763                 E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
1764                 E1000_WRITE_REG(hw, E1000_MANC, manc);
1765         }
1766 }
1767
1768 static void
1769 igb_release_manageability(struct e1000_hw *hw)
1770 {
1771         if (e1000_enable_mng_pass_thru(hw)) {
1772                 uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
1773
1774                 manc |= E1000_MANC_ARP_EN;
1775                 manc &= ~E1000_MANC_EN_MNG2HOST;
1776
1777                 E1000_WRITE_REG(hw, E1000_MANC, manc);
1778         }
1779 }
1780
1781 static void
1782 eth_igb_promiscuous_enable(struct rte_eth_dev *dev)
1783 {
1784         struct e1000_hw *hw =
1785                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1786         uint32_t rctl;
1787
1788         rctl = E1000_READ_REG(hw, E1000_RCTL);
1789         rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1790         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1791 }
1792
1793 static void
1794 eth_igb_promiscuous_disable(struct rte_eth_dev *dev)
1795 {
1796         struct e1000_hw *hw =
1797                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1798         uint32_t rctl;
1799
1800         rctl = E1000_READ_REG(hw, E1000_RCTL);
1801         rctl &= (~E1000_RCTL_UPE);
1802         if (dev->data->all_multicast == 1)
1803                 rctl |= E1000_RCTL_MPE;
1804         else
1805                 rctl &= (~E1000_RCTL_MPE);
1806         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1807 }
1808
1809 static void
1810 eth_igb_allmulticast_enable(struct rte_eth_dev *dev)
1811 {
1812         struct e1000_hw *hw =
1813                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1814         uint32_t rctl;
1815
1816         rctl = E1000_READ_REG(hw, E1000_RCTL);
1817         rctl |= E1000_RCTL_MPE;
1818         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1819 }
1820
1821 static void
1822 eth_igb_allmulticast_disable(struct rte_eth_dev *dev)
1823 {
1824         struct e1000_hw *hw =
1825                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1826         uint32_t rctl;
1827
1828         if (dev->data->promiscuous == 1)
1829                 return; /* must remain in all_multicast mode */
1830         rctl = E1000_READ_REG(hw, E1000_RCTL);
1831         rctl &= (~E1000_RCTL_MPE);
1832         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1833 }
1834
1835 static int
1836 eth_igb_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1837 {
1838         struct e1000_hw *hw =
1839                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1840         struct e1000_vfta * shadow_vfta =
1841                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
1842         uint32_t vfta;
1843         uint32_t vid_idx;
1844         uint32_t vid_bit;
1845
1846         vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
1847                               E1000_VFTA_ENTRY_MASK);
1848         vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
1849         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
1850         if (on)
1851                 vfta |= vid_bit;
1852         else
1853                 vfta &= ~vid_bit;
1854         E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
1855
1856         /* update local VFTA copy */
1857         shadow_vfta->vfta[vid_idx] = vfta;
1858
1859         return 0;
1860 }
1861
1862 static void
1863 eth_igb_vlan_tpid_set(struct rte_eth_dev *dev, uint16_t tpid)
1864 {
1865         struct e1000_hw *hw =
1866                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1867         uint32_t reg = ETHER_TYPE_VLAN ;
1868
1869         reg |= (tpid << 16);
1870         E1000_WRITE_REG(hw, E1000_VET, reg);
1871 }
1872
1873 static void
1874 igb_vlan_hw_filter_disable(struct rte_eth_dev *dev)
1875 {
1876         struct e1000_hw *hw =
1877                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1878         uint32_t reg;
1879
1880         /* Filter Table Disable */
1881         reg = E1000_READ_REG(hw, E1000_RCTL);
1882         reg &= ~E1000_RCTL_CFIEN;
1883         reg &= ~E1000_RCTL_VFE;
1884         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1885 }
1886
1887 static void
1888 igb_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1889 {
1890         struct e1000_hw *hw =
1891                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1892         struct e1000_vfta * shadow_vfta =
1893                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
1894         uint32_t reg;
1895         int i;
1896
1897         /* Filter Table Enable, CFI not used for packet acceptance */
1898         reg = E1000_READ_REG(hw, E1000_RCTL);
1899         reg &= ~E1000_RCTL_CFIEN;
1900         reg |= E1000_RCTL_VFE;
1901         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1902
1903         /* restore VFTA table */
1904         for (i = 0; i < IGB_VFTA_SIZE; i++)
1905                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
1906 }
1907
1908 static void
1909 igb_vlan_hw_strip_disable(struct rte_eth_dev *dev)
1910 {
1911         struct e1000_hw *hw =
1912                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1913         uint32_t reg;
1914
1915         /* VLAN Mode Disable */
1916         reg = E1000_READ_REG(hw, E1000_CTRL);
1917         reg &= ~E1000_CTRL_VME;
1918         E1000_WRITE_REG(hw, E1000_CTRL, reg);
1919 }
1920
1921 static void
1922 igb_vlan_hw_strip_enable(struct rte_eth_dev *dev)
1923 {
1924         struct e1000_hw *hw =
1925                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1926         uint32_t reg;
1927
1928         /* VLAN Mode Enable */
1929         reg = E1000_READ_REG(hw, E1000_CTRL);
1930         reg |= E1000_CTRL_VME;
1931         E1000_WRITE_REG(hw, E1000_CTRL, reg);
1932 }
1933
1934 static void
1935 igb_vlan_hw_extend_disable(struct rte_eth_dev *dev)
1936 {
1937         struct e1000_hw *hw =
1938                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1939         uint32_t reg;
1940
1941         /* CTRL_EXT: Extended VLAN */
1942         reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
1943         reg &= ~E1000_CTRL_EXT_EXTEND_VLAN;
1944         E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
1945
1946         /* Update maximum packet length */
1947         if (dev->data->dev_conf.rxmode.jumbo_frame == 1)
1948                 E1000_WRITE_REG(hw, E1000_RLPML,
1949                         dev->data->dev_conf.rxmode.max_rx_pkt_len +
1950                                                 VLAN_TAG_SIZE);
1951 }
1952
1953 static void
1954 igb_vlan_hw_extend_enable(struct rte_eth_dev *dev)
1955 {
1956         struct e1000_hw *hw =
1957                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1958         uint32_t reg;
1959
1960         /* CTRL_EXT: Extended VLAN */
1961         reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
1962         reg |= E1000_CTRL_EXT_EXTEND_VLAN;
1963         E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
1964
1965         /* Update maximum packet length */
1966         if (dev->data->dev_conf.rxmode.jumbo_frame == 1)
1967                 E1000_WRITE_REG(hw, E1000_RLPML,
1968                         dev->data->dev_conf.rxmode.max_rx_pkt_len +
1969                                                 2 * VLAN_TAG_SIZE);
1970 }
1971
1972 static void
1973 eth_igb_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1974 {
1975         if(mask & ETH_VLAN_STRIP_MASK){
1976                 if (dev->data->dev_conf.rxmode.hw_vlan_strip)
1977                         igb_vlan_hw_strip_enable(dev);
1978                 else
1979                         igb_vlan_hw_strip_disable(dev);
1980         }
1981
1982         if(mask & ETH_VLAN_FILTER_MASK){
1983                 if (dev->data->dev_conf.rxmode.hw_vlan_filter)
1984                         igb_vlan_hw_filter_enable(dev);
1985                 else
1986                         igb_vlan_hw_filter_disable(dev);
1987         }
1988
1989         if(mask & ETH_VLAN_EXTEND_MASK){
1990                 if (dev->data->dev_conf.rxmode.hw_vlan_extend)
1991                         igb_vlan_hw_extend_enable(dev);
1992                 else
1993                         igb_vlan_hw_extend_disable(dev);
1994         }
1995 }
1996
1997
1998 /**
1999  * It enables the interrupt mask and then enable the interrupt.
2000  *
2001  * @param dev
2002  *  Pointer to struct rte_eth_dev.
2003  *
2004  * @return
2005  *  - On success, zero.
2006  *  - On failure, a negative value.
2007  */
2008 static int
2009 eth_igb_lsc_interrupt_setup(struct rte_eth_dev *dev)
2010 {
2011         struct e1000_interrupt *intr =
2012                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2013
2014         intr->mask |= E1000_ICR_LSC;
2015
2016         return 0;
2017 }
2018
2019 #ifdef RTE_NEXT_ABI
2020 /* It clears the interrupt causes and enables the interrupt.
2021  * It will be called once only during nic initialized.
2022  *
2023  * @param dev
2024  *  Pointer to struct rte_eth_dev.
2025  *
2026  * @return
2027  *  - On success, zero.
2028  *  - On failure, a negative value.
2029  */
2030 static int eth_igb_rxq_interrupt_setup(struct rte_eth_dev *dev)
2031 {
2032         uint32_t mask, regval;
2033         struct e1000_hw *hw =
2034                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2035         struct rte_eth_dev_info dev_info;
2036
2037         memset(&dev_info, 0, sizeof(dev_info));
2038         eth_igb_infos_get(dev, &dev_info);
2039
2040         mask = 0xFFFFFFFF >> (32 - dev_info.max_rx_queues);
2041         regval = E1000_READ_REG(hw, E1000_EIMS);
2042         E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
2043
2044         return 0;
2045 }
2046 #endif
2047
2048 /*
2049  * It reads ICR and gets interrupt causes, check it and set a bit flag
2050  * to update link status.
2051  *
2052  * @param dev
2053  *  Pointer to struct rte_eth_dev.
2054  *
2055  * @return
2056  *  - On success, zero.
2057  *  - On failure, a negative value.
2058  */
2059 static int
2060 eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
2061 {
2062         uint32_t icr;
2063         struct e1000_hw *hw =
2064                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2065         struct e1000_interrupt *intr =
2066                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2067
2068         igb_intr_disable(hw);
2069
2070         /* read-on-clear nic registers here */
2071         icr = E1000_READ_REG(hw, E1000_ICR);
2072
2073         intr->flags = 0;
2074         if (icr & E1000_ICR_LSC) {
2075                 intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
2076         }
2077
2078         if (icr & E1000_ICR_VMMB)
2079                 intr->flags |= E1000_FLAG_MAILBOX;
2080
2081         return 0;
2082 }
2083
2084 /*
2085  * It executes link_update after knowing an interrupt is prsent.
2086  *
2087  * @param dev
2088  *  Pointer to struct rte_eth_dev.
2089  *
2090  * @return
2091  *  - On success, zero.
2092  *  - On failure, a negative value.
2093  */
2094 static int
2095 eth_igb_interrupt_action(struct rte_eth_dev *dev)
2096 {
2097         struct e1000_hw *hw =
2098                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2099         struct e1000_interrupt *intr =
2100                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
2101         uint32_t tctl, rctl;
2102         struct rte_eth_link link;
2103         int ret;
2104
2105         if (intr->flags & E1000_FLAG_MAILBOX) {
2106                 igb_pf_mbx_process(dev);
2107                 intr->flags &= ~E1000_FLAG_MAILBOX;
2108         }
2109
2110         igb_intr_enable(dev);
2111         rte_intr_enable(&(dev->pci_dev->intr_handle));
2112
2113         if (intr->flags & E1000_FLAG_NEED_LINK_UPDATE) {
2114                 intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
2115
2116                 /* set get_link_status to check register later */
2117                 hw->mac.get_link_status = 1;
2118                 ret = eth_igb_link_update(dev, 0);
2119
2120                 /* check if link has changed */
2121                 if (ret < 0)
2122                         return 0;
2123
2124                 memset(&link, 0, sizeof(link));
2125                 rte_igb_dev_atomic_read_link_status(dev, &link);
2126                 if (link.link_status) {
2127                         PMD_INIT_LOG(INFO,
2128                                      " Port %d: Link Up - speed %u Mbps - %s",
2129                                      dev->data->port_id,
2130                                      (unsigned)link.link_speed,
2131                                      link.link_duplex == ETH_LINK_FULL_DUPLEX ?
2132                                      "full-duplex" : "half-duplex");
2133                 } else {
2134                         PMD_INIT_LOG(INFO, " Port %d: Link Down",
2135                                      dev->data->port_id);
2136                 }
2137
2138                 PMD_INIT_LOG(DEBUG, "PCI Address: %04d:%02d:%02d:%d",
2139                              dev->pci_dev->addr.domain,
2140                              dev->pci_dev->addr.bus,
2141                              dev->pci_dev->addr.devid,
2142                              dev->pci_dev->addr.function);
2143                 tctl = E1000_READ_REG(hw, E1000_TCTL);
2144                 rctl = E1000_READ_REG(hw, E1000_RCTL);
2145                 if (link.link_status) {
2146                         /* enable Tx/Rx */
2147                         tctl |= E1000_TCTL_EN;
2148                         rctl |= E1000_RCTL_EN;
2149                 } else {
2150                         /* disable Tx/Rx */
2151                         tctl &= ~E1000_TCTL_EN;
2152                         rctl &= ~E1000_RCTL_EN;
2153                 }
2154                 E1000_WRITE_REG(hw, E1000_TCTL, tctl);
2155                 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2156                 E1000_WRITE_FLUSH(hw);
2157                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC);
2158         }
2159
2160         return 0;
2161 }
2162
2163 /**
2164  * Interrupt handler which shall be registered at first.
2165  *
2166  * @param handle
2167  *  Pointer to interrupt handle.
2168  * @param param
2169  *  The address of parameter (struct rte_eth_dev *) regsitered before.
2170  *
2171  * @return
2172  *  void
2173  */
2174 static void
2175 eth_igb_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
2176                                                         void *param)
2177 {
2178         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2179
2180         eth_igb_interrupt_get_status(dev);
2181         eth_igb_interrupt_action(dev);
2182 }
2183
2184 static int
2185 eth_igb_led_on(struct rte_eth_dev *dev)
2186 {
2187         struct e1000_hw *hw;
2188
2189         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2190         return (e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP);
2191 }
2192
2193 static int
2194 eth_igb_led_off(struct rte_eth_dev *dev)
2195 {
2196         struct e1000_hw *hw;
2197
2198         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2199         return (e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP);
2200 }
2201
2202 static int
2203 eth_igb_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2204 {
2205         struct e1000_hw *hw;
2206         uint32_t ctrl;
2207         int tx_pause;
2208         int rx_pause;
2209
2210         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2211         fc_conf->pause_time = hw->fc.pause_time;
2212         fc_conf->high_water = hw->fc.high_water;
2213         fc_conf->low_water = hw->fc.low_water;
2214         fc_conf->send_xon = hw->fc.send_xon;
2215         fc_conf->autoneg = hw->mac.autoneg;
2216
2217         /*
2218          * Return rx_pause and tx_pause status according to actual setting of
2219          * the TFCE and RFCE bits in the CTRL register.
2220          */
2221         ctrl = E1000_READ_REG(hw, E1000_CTRL);
2222         if (ctrl & E1000_CTRL_TFCE)
2223                 tx_pause = 1;
2224         else
2225                 tx_pause = 0;
2226
2227         if (ctrl & E1000_CTRL_RFCE)
2228                 rx_pause = 1;
2229         else
2230                 rx_pause = 0;
2231
2232         if (rx_pause && tx_pause)
2233                 fc_conf->mode = RTE_FC_FULL;
2234         else if (rx_pause)
2235                 fc_conf->mode = RTE_FC_RX_PAUSE;
2236         else if (tx_pause)
2237                 fc_conf->mode = RTE_FC_TX_PAUSE;
2238         else
2239                 fc_conf->mode = RTE_FC_NONE;
2240
2241         return 0;
2242 }
2243
2244 static int
2245 eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2246 {
2247         struct e1000_hw *hw;
2248         int err;
2249         enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
2250                 e1000_fc_none,
2251                 e1000_fc_rx_pause,
2252                 e1000_fc_tx_pause,
2253                 e1000_fc_full
2254         };
2255         uint32_t rx_buf_size;
2256         uint32_t max_high_water;
2257         uint32_t rctl;
2258
2259         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2260         if (fc_conf->autoneg != hw->mac.autoneg)
2261                 return -ENOTSUP;
2262         rx_buf_size = igb_get_rx_buffer_size(hw);
2263         PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
2264
2265         /* At least reserve one Ethernet frame for watermark */
2266         max_high_water = rx_buf_size - ETHER_MAX_LEN;
2267         if ((fc_conf->high_water > max_high_water) ||
2268             (fc_conf->high_water < fc_conf->low_water)) {
2269                 PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
2270                 PMD_INIT_LOG(ERR, "high water must <=  0x%x", max_high_water);
2271                 return (-EINVAL);
2272         }
2273
2274         hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
2275         hw->fc.pause_time     = fc_conf->pause_time;
2276         hw->fc.high_water     = fc_conf->high_water;
2277         hw->fc.low_water      = fc_conf->low_water;
2278         hw->fc.send_xon       = fc_conf->send_xon;
2279
2280         err = e1000_setup_link_generic(hw);
2281         if (err == E1000_SUCCESS) {
2282
2283                 /* check if we want to forward MAC frames - driver doesn't have native
2284                  * capability to do that, so we'll write the registers ourselves */
2285
2286                 rctl = E1000_READ_REG(hw, E1000_RCTL);
2287
2288                 /* set or clear MFLCN.PMCF bit depending on configuration */
2289                 if (fc_conf->mac_ctrl_frame_fwd != 0)
2290                         rctl |= E1000_RCTL_PMCF;
2291                 else
2292                         rctl &= ~E1000_RCTL_PMCF;
2293
2294                 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2295                 E1000_WRITE_FLUSH(hw);
2296
2297                 return 0;
2298         }
2299
2300         PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
2301         return (-EIO);
2302 }
2303
2304 #define E1000_RAH_POOLSEL_SHIFT      (18)
2305 static void
2306 eth_igb_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
2307                 uint32_t index, __rte_unused uint32_t pool)
2308 {
2309         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2310         uint32_t rah;
2311
2312         e1000_rar_set(hw, mac_addr->addr_bytes, index);
2313         rah = E1000_READ_REG(hw, E1000_RAH(index));
2314         rah |= (0x1 << (E1000_RAH_POOLSEL_SHIFT + pool));
2315         E1000_WRITE_REG(hw, E1000_RAH(index), rah);
2316 }
2317
2318 static void
2319 eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index)
2320 {
2321         uint8_t addr[ETHER_ADDR_LEN];
2322         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2323
2324         memset(addr, 0, sizeof(addr));
2325
2326         e1000_rar_set(hw, addr, index);
2327 }
2328
2329 static void
2330 eth_igb_default_mac_addr_set(struct rte_eth_dev *dev,
2331                                 struct ether_addr *addr)
2332 {
2333         eth_igb_rar_clear(dev, 0);
2334
2335         eth_igb_rar_set(dev, (void *)addr, 0, 0);
2336 }
2337 /*
2338  * Virtual Function operations
2339  */
2340 static void
2341 igbvf_intr_disable(struct e1000_hw *hw)
2342 {
2343         PMD_INIT_FUNC_TRACE();
2344
2345         /* Clear interrupt mask to stop from interrupts being generated */
2346         E1000_WRITE_REG(hw, E1000_EIMC, 0xFFFF);
2347
2348         E1000_WRITE_FLUSH(hw);
2349 }
2350
2351 static void
2352 igbvf_stop_adapter(struct rte_eth_dev *dev)
2353 {
2354         u32 reg_val;
2355         u16 i;
2356         struct rte_eth_dev_info dev_info;
2357         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2358
2359         memset(&dev_info, 0, sizeof(dev_info));
2360         eth_igbvf_infos_get(dev, &dev_info);
2361
2362         /* Clear interrupt mask to stop from interrupts being generated */
2363         igbvf_intr_disable(hw);
2364
2365         /* Clear any pending interrupts, flush previous writes */
2366         E1000_READ_REG(hw, E1000_EICR);
2367
2368         /* Disable the transmit unit.  Each queue must be disabled. */
2369         for (i = 0; i < dev_info.max_tx_queues; i++)
2370                 E1000_WRITE_REG(hw, E1000_TXDCTL(i), E1000_TXDCTL_SWFLSH);
2371
2372         /* Disable the receive unit by stopping each queue */
2373         for (i = 0; i < dev_info.max_rx_queues; i++) {
2374                 reg_val = E1000_READ_REG(hw, E1000_RXDCTL(i));
2375                 reg_val &= ~E1000_RXDCTL_QUEUE_ENABLE;
2376                 E1000_WRITE_REG(hw, E1000_RXDCTL(i), reg_val);
2377                 while (E1000_READ_REG(hw, E1000_RXDCTL(i)) & E1000_RXDCTL_QUEUE_ENABLE)
2378                         ;
2379         }
2380
2381         /* flush all queues disables */
2382         E1000_WRITE_FLUSH(hw);
2383         msec_delay(2);
2384 }
2385
2386 static int eth_igbvf_link_update(struct e1000_hw *hw)
2387 {
2388         struct e1000_mbx_info *mbx = &hw->mbx;
2389         struct e1000_mac_info *mac = &hw->mac;
2390         int ret_val = E1000_SUCCESS;
2391
2392         PMD_INIT_LOG(DEBUG, "e1000_check_for_link_vf");
2393
2394         /*
2395          * We only want to run this if there has been a rst asserted.
2396          * in this case that could mean a link change, device reset,
2397          * or a virtual function reset
2398          */
2399
2400         /* If we were hit with a reset or timeout drop the link */
2401         if (!e1000_check_for_rst(hw, 0) || !mbx->timeout)
2402                 mac->get_link_status = TRUE;
2403
2404         if (!mac->get_link_status)
2405                 goto out;
2406
2407         /* if link status is down no point in checking to see if pf is up */
2408         if (!(E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU))
2409                 goto out;
2410
2411         /* if we passed all the tests above then the link is up and we no
2412          * longer need to check for link */
2413         mac->get_link_status = FALSE;
2414
2415 out:
2416         return ret_val;
2417 }
2418
2419
2420 static int
2421 igbvf_dev_configure(struct rte_eth_dev *dev)
2422 {
2423         struct rte_eth_conf* conf = &dev->data->dev_conf;
2424
2425         PMD_INIT_LOG(DEBUG, "Configured Virtual Function port id: %d",
2426                      dev->data->port_id);
2427
2428         /*
2429          * VF has no ability to enable/disable HW CRC
2430          * Keep the persistent behavior the same as Host PF
2431          */
2432 #ifndef RTE_LIBRTE_E1000_PF_DISABLE_STRIP_CRC
2433         if (!conf->rxmode.hw_strip_crc) {
2434                 PMD_INIT_LOG(INFO, "VF can't disable HW CRC Strip");
2435                 conf->rxmode.hw_strip_crc = 1;
2436         }
2437 #else
2438         if (conf->rxmode.hw_strip_crc) {
2439                 PMD_INIT_LOG(INFO, "VF can't enable HW CRC Strip");
2440                 conf->rxmode.hw_strip_crc = 0;
2441         }
2442 #endif
2443
2444         return 0;
2445 }
2446
2447 static int
2448 igbvf_dev_start(struct rte_eth_dev *dev)
2449 {
2450         struct e1000_hw *hw =
2451                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2452         struct e1000_adapter *adapter =
2453                 E1000_DEV_PRIVATE(dev->data->dev_private);
2454         int ret;
2455
2456         PMD_INIT_FUNC_TRACE();
2457
2458         hw->mac.ops.reset_hw(hw);
2459         adapter->stopped = 0;
2460
2461         /* Set all vfta */
2462         igbvf_set_vfta_all(dev,1);
2463
2464         eth_igbvf_tx_init(dev);
2465
2466         /* This can fail when allocating mbufs for descriptor rings */
2467         ret = eth_igbvf_rx_init(dev);
2468         if (ret) {
2469                 PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
2470                 igb_dev_clear_queues(dev);
2471                 return ret;
2472         }
2473
2474         return 0;
2475 }
2476
2477 static void
2478 igbvf_dev_stop(struct rte_eth_dev *dev)
2479 {
2480         PMD_INIT_FUNC_TRACE();
2481
2482         igbvf_stop_adapter(dev);
2483
2484         /*
2485           * Clear what we set, but we still keep shadow_vfta to
2486           * restore after device starts
2487           */
2488         igbvf_set_vfta_all(dev,0);
2489
2490         igb_dev_clear_queues(dev);
2491 }
2492
2493 static void
2494 igbvf_dev_close(struct rte_eth_dev *dev)
2495 {
2496         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2497         struct e1000_adapter *adapter =
2498                 E1000_DEV_PRIVATE(dev->data->dev_private);
2499
2500         PMD_INIT_FUNC_TRACE();
2501
2502         e1000_reset_hw(hw);
2503
2504         igbvf_dev_stop(dev);
2505         adapter->stopped = 1;
2506         igb_dev_free_queues(dev);
2507 }
2508
2509 static int igbvf_set_vfta(struct e1000_hw *hw, uint16_t vid, bool on)
2510 {
2511         struct e1000_mbx_info *mbx = &hw->mbx;
2512         uint32_t msgbuf[2];
2513
2514         /* After set vlan, vlan strip will also be enabled in igb driver*/
2515         msgbuf[0] = E1000_VF_SET_VLAN;
2516         msgbuf[1] = vid;
2517         /* Setting the 8 bit field MSG INFO to TRUE indicates "add" */
2518         if (on)
2519                 msgbuf[0] |= E1000_VF_SET_VLAN_ADD;
2520
2521         return (mbx->ops.write_posted(hw, msgbuf, 2, 0));
2522 }
2523
2524 static void igbvf_set_vfta_all(struct rte_eth_dev *dev, bool on)
2525 {
2526         struct e1000_hw *hw =
2527                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2528         struct e1000_vfta * shadow_vfta =
2529                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2530         int i = 0, j = 0, vfta = 0, mask = 1;
2531
2532         for (i = 0; i < IGB_VFTA_SIZE; i++){
2533                 vfta = shadow_vfta->vfta[i];
2534                 if(vfta){
2535                         mask = 1;
2536                         for (j = 0; j < 32; j++){
2537                                 if(vfta & mask)
2538                                         igbvf_set_vfta(hw,
2539                                                 (uint16_t)((i<<5)+j), on);
2540                                 mask<<=1;
2541                         }
2542                 }
2543         }
2544
2545 }
2546
2547 static int
2548 igbvf_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2549 {
2550         struct e1000_hw *hw =
2551                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2552         struct e1000_vfta * shadow_vfta =
2553                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
2554         uint32_t vid_idx = 0;
2555         uint32_t vid_bit = 0;
2556         int ret = 0;
2557
2558         PMD_INIT_FUNC_TRACE();
2559
2560         /*vind is not used in VF driver, set to 0, check ixgbe_set_vfta_vf*/
2561         ret = igbvf_set_vfta(hw, vlan_id, !!on);
2562         if(ret){
2563                 PMD_INIT_LOG(ERR, "Unable to set VF vlan");
2564                 return ret;
2565         }
2566         vid_idx = (uint32_t) ((vlan_id >> 5) & 0x7F);
2567         vid_bit = (uint32_t) (1 << (vlan_id & 0x1F));
2568
2569         /*Save what we set and retore it after device reset*/
2570         if (on)
2571                 shadow_vfta->vfta[vid_idx] |= vid_bit;
2572         else
2573                 shadow_vfta->vfta[vid_idx] &= ~vid_bit;
2574
2575         return 0;
2576 }
2577
2578 static void
2579 igbvf_default_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *addr)
2580 {
2581         struct e1000_hw *hw =
2582                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2583
2584         /* index is not used by rar_set() */
2585         hw->mac.ops.rar_set(hw, (void *)addr, 0);
2586 }
2587
2588
2589 static int
2590 eth_igb_rss_reta_update(struct rte_eth_dev *dev,
2591                         struct rte_eth_rss_reta_entry64 *reta_conf,
2592                         uint16_t reta_size)
2593 {
2594         uint8_t i, j, mask;
2595         uint32_t reta, r;
2596         uint16_t idx, shift;
2597         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2598
2599         if (reta_size != ETH_RSS_RETA_SIZE_128) {
2600                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2601                         "(%d) doesn't match the number hardware can supported "
2602                         "(%d)\n", reta_size, ETH_RSS_RETA_SIZE_128);
2603                 return -EINVAL;
2604         }
2605
2606         for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
2607                 idx = i / RTE_RETA_GROUP_SIZE;
2608                 shift = i % RTE_RETA_GROUP_SIZE;
2609                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2610                                                 IGB_4_BIT_MASK);
2611                 if (!mask)
2612                         continue;
2613                 if (mask == IGB_4_BIT_MASK)
2614                         r = 0;
2615                 else
2616                         r = E1000_READ_REG(hw, E1000_RETA(i >> 2));
2617                 for (j = 0, reta = 0; j < IGB_4_BIT_WIDTH; j++) {
2618                         if (mask & (0x1 << j))
2619                                 reta |= reta_conf[idx].reta[shift + j] <<
2620                                                         (CHAR_BIT * j);
2621                         else
2622                                 reta |= r & (IGB_8_BIT_MASK << (CHAR_BIT * j));
2623                 }
2624                 E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta);
2625         }
2626
2627         return 0;
2628 }
2629
2630 static int
2631 eth_igb_rss_reta_query(struct rte_eth_dev *dev,
2632                        struct rte_eth_rss_reta_entry64 *reta_conf,
2633                        uint16_t reta_size)
2634 {
2635         uint8_t i, j, mask;
2636         uint32_t reta;
2637         uint16_t idx, shift;
2638         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2639
2640         if (reta_size != ETH_RSS_RETA_SIZE_128) {
2641                 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
2642                         "(%d) doesn't match the number hardware can supported "
2643                         "(%d)\n", reta_size, ETH_RSS_RETA_SIZE_128);
2644                 return -EINVAL;
2645         }
2646
2647         for (i = 0; i < reta_size; i += IGB_4_BIT_WIDTH) {
2648                 idx = i / RTE_RETA_GROUP_SIZE;
2649                 shift = i % RTE_RETA_GROUP_SIZE;
2650                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2651                                                 IGB_4_BIT_MASK);
2652                 if (!mask)
2653                         continue;
2654                 reta = E1000_READ_REG(hw, E1000_RETA(i >> 2));
2655                 for (j = 0; j < IGB_4_BIT_WIDTH; j++) {
2656                         if (mask & (0x1 << j))
2657                                 reta_conf[idx].reta[shift + j] =
2658                                         ((reta >> (CHAR_BIT * j)) &
2659                                                 IGB_8_BIT_MASK);
2660                 }
2661         }
2662
2663         return 0;
2664 }
2665
2666 #define MAC_TYPE_FILTER_SUP(type)    do {\
2667         if ((type) != e1000_82580 && (type) != e1000_i350 &&\
2668                 (type) != e1000_82576)\
2669                 return -ENOTSUP;\
2670 } while (0)
2671
2672 static int
2673 eth_igb_syn_filter_set(struct rte_eth_dev *dev,
2674                         struct rte_eth_syn_filter *filter,
2675                         bool add)
2676 {
2677         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2678         uint32_t synqf, rfctl;
2679
2680         if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
2681                 return -EINVAL;
2682
2683         synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
2684
2685         if (add) {
2686                 if (synqf & E1000_SYN_FILTER_ENABLE)
2687                         return -EINVAL;
2688
2689                 synqf = (uint32_t)(((filter->queue << E1000_SYN_FILTER_QUEUE_SHIFT) &
2690                         E1000_SYN_FILTER_QUEUE) | E1000_SYN_FILTER_ENABLE);
2691
2692                 rfctl = E1000_READ_REG(hw, E1000_RFCTL);
2693                 if (filter->hig_pri)
2694                         rfctl |= E1000_RFCTL_SYNQFP;
2695                 else
2696                         rfctl &= ~E1000_RFCTL_SYNQFP;
2697
2698                 E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
2699         } else {
2700                 if (!(synqf & E1000_SYN_FILTER_ENABLE))
2701                         return -ENOENT;
2702                 synqf = 0;
2703         }
2704
2705         E1000_WRITE_REG(hw, E1000_SYNQF(0), synqf);
2706         E1000_WRITE_FLUSH(hw);
2707         return 0;
2708 }
2709
2710 static int
2711 eth_igb_syn_filter_get(struct rte_eth_dev *dev,
2712                         struct rte_eth_syn_filter *filter)
2713 {
2714         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2715         uint32_t synqf, rfctl;
2716
2717         synqf = E1000_READ_REG(hw, E1000_SYNQF(0));
2718         if (synqf & E1000_SYN_FILTER_ENABLE) {
2719                 rfctl = E1000_READ_REG(hw, E1000_RFCTL);
2720                 filter->hig_pri = (rfctl & E1000_RFCTL_SYNQFP) ? 1 : 0;
2721                 filter->queue = (uint8_t)((synqf & E1000_SYN_FILTER_QUEUE) >>
2722                                 E1000_SYN_FILTER_QUEUE_SHIFT);
2723                 return 0;
2724         }
2725
2726         return -ENOENT;
2727 }
2728
2729 static int
2730 eth_igb_syn_filter_handle(struct rte_eth_dev *dev,
2731                         enum rte_filter_op filter_op,
2732                         void *arg)
2733 {
2734         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2735         int ret;
2736
2737         MAC_TYPE_FILTER_SUP(hw->mac.type);
2738
2739         if (filter_op == RTE_ETH_FILTER_NOP)
2740                 return 0;
2741
2742         if (arg == NULL) {
2743                 PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u",
2744                             filter_op);
2745                 return -EINVAL;
2746         }
2747
2748         switch (filter_op) {
2749         case RTE_ETH_FILTER_ADD:
2750                 ret = eth_igb_syn_filter_set(dev,
2751                                 (struct rte_eth_syn_filter *)arg,
2752                                 TRUE);
2753                 break;
2754         case RTE_ETH_FILTER_DELETE:
2755                 ret = eth_igb_syn_filter_set(dev,
2756                                 (struct rte_eth_syn_filter *)arg,
2757                                 FALSE);
2758                 break;
2759         case RTE_ETH_FILTER_GET:
2760                 ret = eth_igb_syn_filter_get(dev,
2761                                 (struct rte_eth_syn_filter *)arg);
2762                 break;
2763         default:
2764                 PMD_DRV_LOG(ERR, "unsupported operation %u\n", filter_op);
2765                 ret = -EINVAL;
2766                 break;
2767         }
2768
2769         return ret;
2770 }
2771
2772 #define MAC_TYPE_FILTER_SUP_EXT(type)    do {\
2773         if ((type) != e1000_82580 && (type) != e1000_i350)\
2774                 return -ENOSYS; \
2775 } while (0)
2776
2777 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_2tuple_filter_info*/
2778 static inline int
2779 ntuple_filter_to_2tuple(struct rte_eth_ntuple_filter *filter,
2780                         struct e1000_2tuple_filter_info *filter_info)
2781 {
2782         if (filter->queue >= IGB_MAX_RX_QUEUE_NUM)
2783                 return -EINVAL;
2784         if (filter->priority > E1000_2TUPLE_MAX_PRI)
2785                 return -EINVAL;  /* filter index is out of range. */
2786         if (filter->tcp_flags > TCP_FLAG_ALL)
2787                 return -EINVAL;  /* flags is invalid. */
2788
2789         switch (filter->dst_port_mask) {
2790         case UINT16_MAX:
2791                 filter_info->dst_port_mask = 0;
2792                 filter_info->dst_port = filter->dst_port;
2793                 break;
2794         case 0:
2795                 filter_info->dst_port_mask = 1;
2796                 break;
2797         default:
2798                 PMD_DRV_LOG(ERR, "invalid dst_port mask.");
2799                 return -EINVAL;
2800         }
2801
2802         switch (filter->proto_mask) {
2803         case UINT8_MAX:
2804                 filter_info->proto_mask = 0;
2805                 filter_info->proto = filter->proto;
2806                 break;
2807         case 0:
2808                 filter_info->proto_mask = 1;
2809                 break;
2810         default:
2811                 PMD_DRV_LOG(ERR, "invalid protocol mask.");
2812                 return -EINVAL;
2813         }
2814
2815         filter_info->priority = (uint8_t)filter->priority;
2816         if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
2817                 filter_info->tcp_flags = filter->tcp_flags;
2818         else
2819                 filter_info->tcp_flags = 0;
2820
2821         return 0;
2822 }
2823
2824 static inline struct e1000_2tuple_filter *
2825 igb_2tuple_filter_lookup(struct e1000_2tuple_filter_list *filter_list,
2826                         struct e1000_2tuple_filter_info *key)
2827 {
2828         struct e1000_2tuple_filter *it;
2829
2830         TAILQ_FOREACH(it, filter_list, entries) {
2831                 if (memcmp(key, &it->filter_info,
2832                         sizeof(struct e1000_2tuple_filter_info)) == 0) {
2833                         return it;
2834                 }
2835         }
2836         return NULL;
2837 }
2838
2839 /*
2840  * igb_add_2tuple_filter - add a 2tuple filter
2841  *
2842  * @param
2843  * dev: Pointer to struct rte_eth_dev.
2844  * ntuple_filter: ponter to the filter that will be added.
2845  *
2846  * @return
2847  *    - On success, zero.
2848  *    - On failure, a negative value.
2849  */
2850 static int
2851 igb_add_2tuple_filter(struct rte_eth_dev *dev,
2852                         struct rte_eth_ntuple_filter *ntuple_filter)
2853 {
2854         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2855         struct e1000_filter_info *filter_info =
2856                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
2857         struct e1000_2tuple_filter *filter;
2858         uint32_t ttqf = E1000_TTQF_DISABLE_MASK;
2859         uint32_t imir, imir_ext = E1000_IMIREXT_SIZE_BP;
2860         int i, ret;
2861
2862         filter = rte_zmalloc("e1000_2tuple_filter",
2863                         sizeof(struct e1000_2tuple_filter), 0);
2864         if (filter == NULL)
2865                 return -ENOMEM;
2866
2867         ret = ntuple_filter_to_2tuple(ntuple_filter,
2868                                       &filter->filter_info);
2869         if (ret < 0) {
2870                 rte_free(filter);
2871                 return ret;
2872         }
2873         if (igb_2tuple_filter_lookup(&filter_info->twotuple_list,
2874                                          &filter->filter_info) != NULL) {
2875                 PMD_DRV_LOG(ERR, "filter exists.");
2876                 rte_free(filter);
2877                 return -EEXIST;
2878         }
2879         filter->queue = ntuple_filter->queue;
2880
2881         /*
2882          * look for an unused 2tuple filter index,
2883          * and insert the filter to list.
2884          */
2885         for (i = 0; i < E1000_MAX_TTQF_FILTERS; i++) {
2886                 if (!(filter_info->twotuple_mask & (1 << i))) {
2887                         filter_info->twotuple_mask |= 1 << i;
2888                         filter->index = i;
2889                         TAILQ_INSERT_TAIL(&filter_info->twotuple_list,
2890                                           filter,
2891                                           entries);
2892                         break;
2893                 }
2894         }
2895         if (i >= E1000_MAX_TTQF_FILTERS) {
2896                 PMD_DRV_LOG(ERR, "2tuple filters are full.");
2897                 rte_free(filter);
2898                 return -ENOSYS;
2899         }
2900
2901         imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
2902         if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
2903                 imir |= E1000_IMIR_PORT_BP;
2904         else
2905                 imir &= ~E1000_IMIR_PORT_BP;
2906
2907         imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
2908
2909         ttqf |= E1000_TTQF_QUEUE_ENABLE;
2910         ttqf |= (uint32_t)(filter->queue << E1000_TTQF_QUEUE_SHIFT);
2911         ttqf |= (uint32_t)(filter->filter_info.proto & E1000_TTQF_PROTOCOL_MASK);
2912         if (filter->filter_info.proto_mask == 0)
2913                 ttqf &= ~E1000_TTQF_MASK_ENABLE;
2914
2915         /* tcp flags bits setting. */
2916         if (filter->filter_info.tcp_flags & TCP_FLAG_ALL) {
2917                 if (filter->filter_info.tcp_flags & TCP_URG_FLAG)
2918                         imir_ext |= E1000_IMIREXT_CTRL_URG;
2919                 if (filter->filter_info.tcp_flags & TCP_ACK_FLAG)
2920                         imir_ext |= E1000_IMIREXT_CTRL_ACK;
2921                 if (filter->filter_info.tcp_flags & TCP_PSH_FLAG)
2922                         imir_ext |= E1000_IMIREXT_CTRL_PSH;
2923                 if (filter->filter_info.tcp_flags & TCP_RST_FLAG)
2924                         imir_ext |= E1000_IMIREXT_CTRL_RST;
2925                 if (filter->filter_info.tcp_flags & TCP_SYN_FLAG)
2926                         imir_ext |= E1000_IMIREXT_CTRL_SYN;
2927                 if (filter->filter_info.tcp_flags & TCP_FIN_FLAG)
2928                         imir_ext |= E1000_IMIREXT_CTRL_FIN;
2929         } else
2930                 imir_ext |= E1000_IMIREXT_CTRL_BP;
2931         E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
2932         E1000_WRITE_REG(hw, E1000_TTQF(i), ttqf);
2933         E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
2934         return 0;
2935 }
2936
2937 /*
2938  * igb_remove_2tuple_filter - remove a 2tuple filter
2939  *
2940  * @param
2941  * dev: Pointer to struct rte_eth_dev.
2942  * ntuple_filter: ponter to the filter that will be removed.
2943  *
2944  * @return
2945  *    - On success, zero.
2946  *    - On failure, a negative value.
2947  */
2948 static int
2949 igb_remove_2tuple_filter(struct rte_eth_dev *dev,
2950                         struct rte_eth_ntuple_filter *ntuple_filter)
2951 {
2952         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2953         struct e1000_filter_info *filter_info =
2954                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
2955         struct e1000_2tuple_filter_info filter_2tuple;
2956         struct e1000_2tuple_filter *filter;
2957         int ret;
2958
2959         memset(&filter_2tuple, 0, sizeof(struct e1000_2tuple_filter_info));
2960         ret = ntuple_filter_to_2tuple(ntuple_filter,
2961                                       &filter_2tuple);
2962         if (ret < 0)
2963                 return ret;
2964
2965         filter = igb_2tuple_filter_lookup(&filter_info->twotuple_list,
2966                                          &filter_2tuple);
2967         if (filter == NULL) {
2968                 PMD_DRV_LOG(ERR, "filter doesn't exist.");
2969                 return -ENOENT;
2970         }
2971
2972         filter_info->twotuple_mask &= ~(1 << filter->index);
2973         TAILQ_REMOVE(&filter_info->twotuple_list, filter, entries);
2974         rte_free(filter);
2975
2976         E1000_WRITE_REG(hw, E1000_TTQF(filter->index), E1000_TTQF_DISABLE_MASK);
2977         E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
2978         E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
2979         return 0;
2980 }
2981
2982 static inline struct e1000_flex_filter *
2983 eth_igb_flex_filter_lookup(struct e1000_flex_filter_list *filter_list,
2984                         struct e1000_flex_filter_info *key)
2985 {
2986         struct e1000_flex_filter *it;
2987
2988         TAILQ_FOREACH(it, filter_list, entries) {
2989                 if (memcmp(key, &it->filter_info,
2990                         sizeof(struct e1000_flex_filter_info)) == 0)
2991                         return it;
2992         }
2993
2994         return NULL;
2995 }
2996
2997 static int
2998 eth_igb_add_del_flex_filter(struct rte_eth_dev *dev,
2999                         struct rte_eth_flex_filter *filter,
3000                         bool add)
3001 {
3002         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3003         struct e1000_filter_info *filter_info =
3004                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3005         struct e1000_flex_filter *flex_filter, *it;
3006         uint32_t wufc, queueing, mask;
3007         uint32_t reg_off;
3008         uint8_t shift, i, j = 0;
3009
3010         flex_filter = rte_zmalloc("e1000_flex_filter",
3011                         sizeof(struct e1000_flex_filter), 0);
3012         if (flex_filter == NULL)
3013                 return -ENOMEM;
3014
3015         flex_filter->filter_info.len = filter->len;
3016         flex_filter->filter_info.priority = filter->priority;
3017         memcpy(flex_filter->filter_info.dwords, filter->bytes, filter->len);
3018         for (i = 0; i < RTE_ALIGN(filter->len, CHAR_BIT) / CHAR_BIT; i++) {
3019                 mask = 0;
3020                 /* reverse bits in flex filter's mask*/
3021                 for (shift = 0; shift < CHAR_BIT; shift++) {
3022                         if (filter->mask[i] & (0x01 << shift))
3023                                 mask |= (0x80 >> shift);
3024                 }
3025                 flex_filter->filter_info.mask[i] = mask;
3026         }
3027
3028         wufc = E1000_READ_REG(hw, E1000_WUFC);
3029         if (flex_filter->index < E1000_MAX_FHFT)
3030                 reg_off = E1000_FHFT(flex_filter->index);
3031         else
3032                 reg_off = E1000_FHFT_EXT(flex_filter->index - E1000_MAX_FHFT);
3033
3034         if (add) {
3035                 if (eth_igb_flex_filter_lookup(&filter_info->flex_list,
3036                                 &flex_filter->filter_info) != NULL) {
3037                         PMD_DRV_LOG(ERR, "filter exists.");
3038                         rte_free(flex_filter);
3039                         return -EEXIST;
3040                 }
3041                 flex_filter->queue = filter->queue;
3042                 /*
3043                  * look for an unused flex filter index
3044                  * and insert the filter into the list.
3045                  */
3046                 for (i = 0; i < E1000_MAX_FLEX_FILTERS; i++) {
3047                         if (!(filter_info->flex_mask & (1 << i))) {
3048                                 filter_info->flex_mask |= 1 << i;
3049                                 flex_filter->index = i;
3050                                 TAILQ_INSERT_TAIL(&filter_info->flex_list,
3051                                         flex_filter,
3052                                         entries);
3053                                 break;
3054                         }
3055                 }
3056                 if (i >= E1000_MAX_FLEX_FILTERS) {
3057                         PMD_DRV_LOG(ERR, "flex filters are full.");
3058                         rte_free(flex_filter);
3059                         return -ENOSYS;
3060                 }
3061
3062                 E1000_WRITE_REG(hw, E1000_WUFC, wufc | E1000_WUFC_FLEX_HQ |
3063                                 (E1000_WUFC_FLX0 << flex_filter->index));
3064                 queueing = filter->len |
3065                         (filter->queue << E1000_FHFT_QUEUEING_QUEUE_SHIFT) |
3066                         (filter->priority << E1000_FHFT_QUEUEING_PRIO_SHIFT);
3067                 E1000_WRITE_REG(hw, reg_off + E1000_FHFT_QUEUEING_OFFSET,
3068                                 queueing);
3069                 for (i = 0; i < E1000_FLEX_FILTERS_MASK_SIZE; i++) {
3070                         E1000_WRITE_REG(hw, reg_off,
3071                                         flex_filter->filter_info.dwords[j]);
3072                         reg_off += sizeof(uint32_t);
3073                         E1000_WRITE_REG(hw, reg_off,
3074                                         flex_filter->filter_info.dwords[++j]);
3075                         reg_off += sizeof(uint32_t);
3076                         E1000_WRITE_REG(hw, reg_off,
3077                                 (uint32_t)flex_filter->filter_info.mask[i]);
3078                         reg_off += sizeof(uint32_t) * 2;
3079                         ++j;
3080                 }
3081         } else {
3082                 it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
3083                                 &flex_filter->filter_info);
3084                 if (it == NULL) {
3085                         PMD_DRV_LOG(ERR, "filter doesn't exist.");
3086                         rte_free(flex_filter);
3087                         return -ENOENT;
3088                 }
3089
3090                 for (i = 0; i < E1000_FHFT_SIZE_IN_DWD; i++)
3091                         E1000_WRITE_REG(hw, reg_off + i * sizeof(uint32_t), 0);
3092                 E1000_WRITE_REG(hw, E1000_WUFC, wufc &
3093                         (~(E1000_WUFC_FLX0 << it->index)));
3094
3095                 filter_info->flex_mask &= ~(1 << it->index);
3096                 TAILQ_REMOVE(&filter_info->flex_list, it, entries);
3097                 rte_free(it);
3098                 rte_free(flex_filter);
3099         }
3100
3101         return 0;
3102 }
3103
3104 static int
3105 eth_igb_get_flex_filter(struct rte_eth_dev *dev,
3106                         struct rte_eth_flex_filter *filter)
3107 {
3108         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3109         struct e1000_filter_info *filter_info =
3110                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3111         struct e1000_flex_filter flex_filter, *it;
3112         uint32_t wufc, queueing, wufc_en = 0;
3113
3114         memset(&flex_filter, 0, sizeof(struct e1000_flex_filter));
3115         flex_filter.filter_info.len = filter->len;
3116         flex_filter.filter_info.priority = filter->priority;
3117         memcpy(flex_filter.filter_info.dwords, filter->bytes, filter->len);
3118         memcpy(flex_filter.filter_info.mask, filter->mask,
3119                         RTE_ALIGN(filter->len, sizeof(char)) / sizeof(char));
3120
3121         it = eth_igb_flex_filter_lookup(&filter_info->flex_list,
3122                                 &flex_filter.filter_info);
3123         if (it == NULL) {
3124                 PMD_DRV_LOG(ERR, "filter doesn't exist.");
3125                 return -ENOENT;
3126         }
3127
3128         wufc = E1000_READ_REG(hw, E1000_WUFC);
3129         wufc_en = E1000_WUFC_FLEX_HQ | (E1000_WUFC_FLX0 << it->index);
3130
3131         if ((wufc & wufc_en) == wufc_en) {
3132                 uint32_t reg_off = 0;
3133                 if (it->index < E1000_MAX_FHFT)
3134                         reg_off = E1000_FHFT(it->index);
3135                 else
3136                         reg_off = E1000_FHFT_EXT(it->index - E1000_MAX_FHFT);
3137
3138                 queueing = E1000_READ_REG(hw,
3139                                 reg_off + E1000_FHFT_QUEUEING_OFFSET);
3140                 filter->len = queueing & E1000_FHFT_QUEUEING_LEN;
3141                 filter->priority = (queueing & E1000_FHFT_QUEUEING_PRIO) >>
3142                         E1000_FHFT_QUEUEING_PRIO_SHIFT;
3143                 filter->queue = (queueing & E1000_FHFT_QUEUEING_QUEUE) >>
3144                         E1000_FHFT_QUEUEING_QUEUE_SHIFT;
3145                 return 0;
3146         }
3147         return -ENOENT;
3148 }
3149
3150 static int
3151 eth_igb_flex_filter_handle(struct rte_eth_dev *dev,
3152                         enum rte_filter_op filter_op,
3153                         void *arg)
3154 {
3155         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3156         struct rte_eth_flex_filter *filter;
3157         int ret = 0;
3158
3159         MAC_TYPE_FILTER_SUP_EXT(hw->mac.type);
3160
3161         if (filter_op == RTE_ETH_FILTER_NOP)
3162                 return ret;
3163
3164         if (arg == NULL) {
3165                 PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u",
3166                             filter_op);
3167                 return -EINVAL;
3168         }
3169
3170         filter = (struct rte_eth_flex_filter *)arg;
3171         if (filter->len == 0 || filter->len > E1000_MAX_FLEX_FILTER_LEN
3172             || filter->len % sizeof(uint64_t) != 0) {
3173                 PMD_DRV_LOG(ERR, "filter's length is out of range");
3174                 return -EINVAL;
3175         }
3176         if (filter->priority > E1000_MAX_FLEX_FILTER_PRI) {
3177                 PMD_DRV_LOG(ERR, "filter's priority is out of range");
3178                 return -EINVAL;
3179         }
3180
3181         switch (filter_op) {
3182         case RTE_ETH_FILTER_ADD:
3183                 ret = eth_igb_add_del_flex_filter(dev, filter, TRUE);
3184                 break;
3185         case RTE_ETH_FILTER_DELETE:
3186                 ret = eth_igb_add_del_flex_filter(dev, filter, FALSE);
3187                 break;
3188         case RTE_ETH_FILTER_GET:
3189                 ret = eth_igb_get_flex_filter(dev, filter);
3190                 break;
3191         default:
3192                 PMD_DRV_LOG(ERR, "unsupported operation %u", filter_op);
3193                 ret = -EINVAL;
3194                 break;
3195         }
3196
3197         return ret;
3198 }
3199
3200 /* translate elements in struct rte_eth_ntuple_filter to struct e1000_5tuple_filter_info*/
3201 static inline int
3202 ntuple_filter_to_5tuple_82576(struct rte_eth_ntuple_filter *filter,
3203                         struct e1000_5tuple_filter_info *filter_info)
3204 {
3205         if (filter->queue >= IGB_MAX_RX_QUEUE_NUM_82576)
3206                 return -EINVAL;
3207         if (filter->priority > E1000_2TUPLE_MAX_PRI)
3208                 return -EINVAL;  /* filter index is out of range. */
3209         if (filter->tcp_flags > TCP_FLAG_ALL)
3210                 return -EINVAL;  /* flags is invalid. */
3211
3212         switch (filter->dst_ip_mask) {
3213         case UINT32_MAX:
3214                 filter_info->dst_ip_mask = 0;
3215                 filter_info->dst_ip = filter->dst_ip;
3216                 break;
3217         case 0:
3218                 filter_info->dst_ip_mask = 1;
3219                 break;
3220         default:
3221                 PMD_DRV_LOG(ERR, "invalid dst_ip mask.");
3222                 return -EINVAL;
3223         }
3224
3225         switch (filter->src_ip_mask) {
3226         case UINT32_MAX:
3227                 filter_info->src_ip_mask = 0;
3228                 filter_info->src_ip = filter->src_ip;
3229                 break;
3230         case 0:
3231                 filter_info->src_ip_mask = 1;
3232                 break;
3233         default:
3234                 PMD_DRV_LOG(ERR, "invalid src_ip mask.");
3235                 return -EINVAL;
3236         }
3237
3238         switch (filter->dst_port_mask) {
3239         case UINT16_MAX:
3240                 filter_info->dst_port_mask = 0;
3241                 filter_info->dst_port = filter->dst_port;
3242                 break;
3243         case 0:
3244                 filter_info->dst_port_mask = 1;
3245                 break;
3246         default:
3247                 PMD_DRV_LOG(ERR, "invalid dst_port mask.");
3248                 return -EINVAL;
3249         }
3250
3251         switch (filter->src_port_mask) {
3252         case UINT16_MAX:
3253                 filter_info->src_port_mask = 0;
3254                 filter_info->src_port = filter->src_port;
3255                 break;
3256         case 0:
3257                 filter_info->src_port_mask = 1;
3258                 break;
3259         default:
3260                 PMD_DRV_LOG(ERR, "invalid src_port mask.");
3261                 return -EINVAL;
3262         }
3263
3264         switch (filter->proto_mask) {
3265         case UINT8_MAX:
3266                 filter_info->proto_mask = 0;
3267                 filter_info->proto = filter->proto;
3268                 break;
3269         case 0:
3270                 filter_info->proto_mask = 1;
3271                 break;
3272         default:
3273                 PMD_DRV_LOG(ERR, "invalid protocol mask.");
3274                 return -EINVAL;
3275         }
3276
3277         filter_info->priority = (uint8_t)filter->priority;
3278         if (filter->flags & RTE_NTUPLE_FLAGS_TCP_FLAG)
3279                 filter_info->tcp_flags = filter->tcp_flags;
3280         else
3281                 filter_info->tcp_flags = 0;
3282
3283         return 0;
3284 }
3285
3286 static inline struct e1000_5tuple_filter *
3287 igb_5tuple_filter_lookup_82576(struct e1000_5tuple_filter_list *filter_list,
3288                         struct e1000_5tuple_filter_info *key)
3289 {
3290         struct e1000_5tuple_filter *it;
3291
3292         TAILQ_FOREACH(it, filter_list, entries) {
3293                 if (memcmp(key, &it->filter_info,
3294                         sizeof(struct e1000_5tuple_filter_info)) == 0) {
3295                         return it;
3296                 }
3297         }
3298         return NULL;
3299 }
3300
3301 /*
3302  * igb_add_5tuple_filter_82576 - add a 5tuple filter
3303  *
3304  * @param
3305  * dev: Pointer to struct rte_eth_dev.
3306  * ntuple_filter: ponter to the filter that will be added.
3307  *
3308  * @return
3309  *    - On success, zero.
3310  *    - On failure, a negative value.
3311  */
3312 static int
3313 igb_add_5tuple_filter_82576(struct rte_eth_dev *dev,
3314                         struct rte_eth_ntuple_filter *ntuple_filter)
3315 {
3316         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3317         struct e1000_filter_info *filter_info =
3318                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3319         struct e1000_5tuple_filter *filter;
3320         uint32_t ftqf = E1000_FTQF_VF_BP | E1000_FTQF_MASK;
3321         uint32_t spqf, imir, imir_ext = E1000_IMIREXT_SIZE_BP;
3322         uint8_t i;
3323         int ret;
3324
3325         filter = rte_zmalloc("e1000_5tuple_filter",
3326                         sizeof(struct e1000_5tuple_filter), 0);
3327         if (filter == NULL)
3328                 return -ENOMEM;
3329
3330         ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
3331                                             &filter->filter_info);
3332         if (ret < 0) {
3333                 rte_free(filter);
3334                 return ret;
3335         }
3336
3337         if (igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
3338                                          &filter->filter_info) != NULL) {
3339                 PMD_DRV_LOG(ERR, "filter exists.");
3340                 rte_free(filter);
3341                 return -EEXIST;
3342         }
3343         filter->queue = ntuple_filter->queue;
3344
3345         /*
3346          * look for an unused 5tuple filter index,
3347          * and insert the filter to list.
3348          */
3349         for (i = 0; i < E1000_MAX_FTQF_FILTERS; i++) {
3350                 if (!(filter_info->fivetuple_mask & (1 << i))) {
3351                         filter_info->fivetuple_mask |= 1 << i;
3352                         filter->index = i;
3353                         TAILQ_INSERT_TAIL(&filter_info->fivetuple_list,
3354                                           filter,
3355                                           entries);
3356                         break;
3357                 }
3358         }
3359         if (i >= E1000_MAX_FTQF_FILTERS) {
3360                 PMD_DRV_LOG(ERR, "5tuple filters are full.");
3361                 rte_free(filter);
3362                 return -ENOSYS;
3363         }
3364
3365         ftqf |= filter->filter_info.proto & E1000_FTQF_PROTOCOL_MASK;
3366         if (filter->filter_info.src_ip_mask == 0) /* 0b means compare. */
3367                 ftqf &= ~E1000_FTQF_MASK_SOURCE_ADDR_BP;
3368         if (filter->filter_info.dst_ip_mask == 0)
3369                 ftqf &= ~E1000_FTQF_MASK_DEST_ADDR_BP;
3370         if (filter->filter_info.src_port_mask == 0)
3371                 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
3372         if (filter->filter_info.proto_mask == 0)
3373                 ftqf &= ~E1000_FTQF_MASK_PROTO_BP;
3374         ftqf |= (filter->queue << E1000_FTQF_QUEUE_SHIFT) &
3375                 E1000_FTQF_QUEUE_MASK;
3376         ftqf |= E1000_FTQF_QUEUE_ENABLE;
3377         E1000_WRITE_REG(hw, E1000_FTQF(i), ftqf);
3378         E1000_WRITE_REG(hw, E1000_DAQF(i), filter->filter_info.dst_ip);
3379         E1000_WRITE_REG(hw, E1000_SAQF(i), filter->filter_info.src_ip);
3380
3381         spqf = filter->filter_info.src_port & E1000_SPQF_SRCPORT;
3382         E1000_WRITE_REG(hw, E1000_SPQF(i), spqf);
3383
3384         imir = (uint32_t)(filter->filter_info.dst_port & E1000_IMIR_DSTPORT);
3385         if (filter->filter_info.dst_port_mask == 1) /* 1b means not compare. */
3386                 imir |= E1000_IMIR_PORT_BP;
3387         else
3388                 imir &= ~E1000_IMIR_PORT_BP;
3389         imir |= filter->filter_info.priority << E1000_IMIR_PRIORITY_SHIFT;
3390
3391         /* tcp flags bits setting. */
3392         if (filter->filter_info.tcp_flags & TCP_FLAG_ALL) {
3393                 if (filter->filter_info.tcp_flags & TCP_URG_FLAG)
3394                         imir_ext |= E1000_IMIREXT_CTRL_URG;
3395                 if (filter->filter_info.tcp_flags & TCP_ACK_FLAG)
3396                         imir_ext |= E1000_IMIREXT_CTRL_ACK;
3397                 if (filter->filter_info.tcp_flags & TCP_PSH_FLAG)
3398                         imir_ext |= E1000_IMIREXT_CTRL_PSH;
3399                 if (filter->filter_info.tcp_flags & TCP_RST_FLAG)
3400                         imir_ext |= E1000_IMIREXT_CTRL_RST;
3401                 if (filter->filter_info.tcp_flags & TCP_SYN_FLAG)
3402                         imir_ext |= E1000_IMIREXT_CTRL_SYN;
3403                 if (filter->filter_info.tcp_flags & TCP_FIN_FLAG)
3404                         imir_ext |= E1000_IMIREXT_CTRL_FIN;
3405         } else
3406                 imir_ext |= E1000_IMIREXT_CTRL_BP;
3407         E1000_WRITE_REG(hw, E1000_IMIR(i), imir);
3408         E1000_WRITE_REG(hw, E1000_IMIREXT(i), imir_ext);
3409         return 0;
3410 }
3411
3412 /*
3413  * igb_remove_5tuple_filter_82576 - remove a 5tuple filter
3414  *
3415  * @param
3416  * dev: Pointer to struct rte_eth_dev.
3417  * ntuple_filter: ponter to the filter that will be removed.
3418  *
3419  * @return
3420  *    - On success, zero.
3421  *    - On failure, a negative value.
3422  */
3423 static int
3424 igb_remove_5tuple_filter_82576(struct rte_eth_dev *dev,
3425                                 struct rte_eth_ntuple_filter *ntuple_filter)
3426 {
3427         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3428         struct e1000_filter_info *filter_info =
3429                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3430         struct e1000_5tuple_filter_info filter_5tuple;
3431         struct e1000_5tuple_filter *filter;
3432         int ret;
3433
3434         memset(&filter_5tuple, 0, sizeof(struct e1000_5tuple_filter_info));
3435         ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
3436                                             &filter_5tuple);
3437         if (ret < 0)
3438                 return ret;
3439
3440         filter = igb_5tuple_filter_lookup_82576(&filter_info->fivetuple_list,
3441                                          &filter_5tuple);
3442         if (filter == NULL) {
3443                 PMD_DRV_LOG(ERR, "filter doesn't exist.");
3444                 return -ENOENT;
3445         }
3446
3447         filter_info->fivetuple_mask &= ~(1 << filter->index);
3448         TAILQ_REMOVE(&filter_info->fivetuple_list, filter, entries);
3449         rte_free(filter);
3450
3451         E1000_WRITE_REG(hw, E1000_FTQF(filter->index),
3452                         E1000_FTQF_VF_BP | E1000_FTQF_MASK);
3453         E1000_WRITE_REG(hw, E1000_DAQF(filter->index), 0);
3454         E1000_WRITE_REG(hw, E1000_SAQF(filter->index), 0);
3455         E1000_WRITE_REG(hw, E1000_SPQF(filter->index), 0);
3456         E1000_WRITE_REG(hw, E1000_IMIR(filter->index), 0);
3457         E1000_WRITE_REG(hw, E1000_IMIREXT(filter->index), 0);
3458         return 0;
3459 }
3460
3461 static int
3462 eth_igb_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
3463 {
3464         uint32_t rctl;
3465         struct e1000_hw *hw;
3466         struct rte_eth_dev_info dev_info;
3467         uint32_t frame_size = mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN +
3468                                      VLAN_TAG_SIZE);
3469
3470         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3471
3472 #ifdef RTE_LIBRTE_82571_SUPPORT
3473         /* XXX: not bigger than max_rx_pktlen */
3474         if (hw->mac.type == e1000_82571)
3475                 return -ENOTSUP;
3476 #endif
3477         eth_igb_infos_get(dev, &dev_info);
3478
3479         /* check that mtu is within the allowed range */
3480         if ((mtu < ETHER_MIN_MTU) ||
3481             (frame_size > dev_info.max_rx_pktlen))
3482                 return -EINVAL;
3483
3484         /* refuse mtu that requires the support of scattered packets when this
3485          * feature has not been enabled before. */
3486         if (!dev->data->scattered_rx &&
3487             frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)
3488                 return -EINVAL;
3489
3490         rctl = E1000_READ_REG(hw, E1000_RCTL);
3491
3492         /* switch to jumbo mode if needed */
3493         if (frame_size > ETHER_MAX_LEN) {
3494                 dev->data->dev_conf.rxmode.jumbo_frame = 1;
3495                 rctl |= E1000_RCTL_LPE;
3496         } else {
3497                 dev->data->dev_conf.rxmode.jumbo_frame = 0;
3498                 rctl &= ~E1000_RCTL_LPE;
3499         }
3500         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3501
3502         /* update max frame size */
3503         dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
3504
3505         E1000_WRITE_REG(hw, E1000_RLPML,
3506                         dev->data->dev_conf.rxmode.max_rx_pkt_len);
3507
3508         return 0;
3509 }
3510
3511 /*
3512  * igb_add_del_ntuple_filter - add or delete a ntuple filter
3513  *
3514  * @param
3515  * dev: Pointer to struct rte_eth_dev.
3516  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
3517  * add: if true, add filter, if false, remove filter
3518  *
3519  * @return
3520  *    - On success, zero.
3521  *    - On failure, a negative value.
3522  */
3523 static int
3524 igb_add_del_ntuple_filter(struct rte_eth_dev *dev,
3525                         struct rte_eth_ntuple_filter *ntuple_filter,
3526                         bool add)
3527 {
3528         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3529         int ret;
3530
3531         switch (ntuple_filter->flags) {
3532         case RTE_5TUPLE_FLAGS:
3533         case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
3534                 if (hw->mac.type != e1000_82576)
3535                         return -ENOTSUP;
3536                 if (add)
3537                         ret = igb_add_5tuple_filter_82576(dev,
3538                                                           ntuple_filter);
3539                 else
3540                         ret = igb_remove_5tuple_filter_82576(dev,
3541                                                              ntuple_filter);
3542                 break;
3543         case RTE_2TUPLE_FLAGS:
3544         case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
3545                 if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350)
3546                         return -ENOTSUP;
3547                 if (add)
3548                         ret = igb_add_2tuple_filter(dev, ntuple_filter);
3549                 else
3550                         ret = igb_remove_2tuple_filter(dev, ntuple_filter);
3551                 break;
3552         default:
3553                 ret = -EINVAL;
3554                 break;
3555         }
3556
3557         return ret;
3558 }
3559
3560 /*
3561  * igb_get_ntuple_filter - get a ntuple filter
3562  *
3563  * @param
3564  * dev: Pointer to struct rte_eth_dev.
3565  * ntuple_filter: Pointer to struct rte_eth_ntuple_filter
3566  *
3567  * @return
3568  *    - On success, zero.
3569  *    - On failure, a negative value.
3570  */
3571 static int
3572 igb_get_ntuple_filter(struct rte_eth_dev *dev,
3573                         struct rte_eth_ntuple_filter *ntuple_filter)
3574 {
3575         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3576         struct e1000_filter_info *filter_info =
3577                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3578         struct e1000_5tuple_filter_info filter_5tuple;
3579         struct e1000_2tuple_filter_info filter_2tuple;
3580         struct e1000_5tuple_filter *p_5tuple_filter;
3581         struct e1000_2tuple_filter *p_2tuple_filter;
3582         int ret;
3583
3584         switch (ntuple_filter->flags) {
3585         case RTE_5TUPLE_FLAGS:
3586         case (RTE_5TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
3587                 if (hw->mac.type != e1000_82576)
3588                         return -ENOTSUP;
3589                 memset(&filter_5tuple,
3590                         0,
3591                         sizeof(struct e1000_5tuple_filter_info));
3592                 ret = ntuple_filter_to_5tuple_82576(ntuple_filter,
3593                                                     &filter_5tuple);
3594                 if (ret < 0)
3595                         return ret;
3596                 p_5tuple_filter = igb_5tuple_filter_lookup_82576(
3597                                         &filter_info->fivetuple_list,
3598                                         &filter_5tuple);
3599                 if (p_5tuple_filter == NULL) {
3600                         PMD_DRV_LOG(ERR, "filter doesn't exist.");
3601                         return -ENOENT;
3602                 }
3603                 ntuple_filter->queue = p_5tuple_filter->queue;
3604                 break;
3605         case RTE_2TUPLE_FLAGS:
3606         case (RTE_2TUPLE_FLAGS | RTE_NTUPLE_FLAGS_TCP_FLAG):
3607                 if (hw->mac.type != e1000_82580 && hw->mac.type != e1000_i350)
3608                         return -ENOTSUP;
3609                 memset(&filter_2tuple,
3610                         0,
3611                         sizeof(struct e1000_2tuple_filter_info));
3612                 ret = ntuple_filter_to_2tuple(ntuple_filter, &filter_2tuple);
3613                 if (ret < 0)
3614                         return ret;
3615                 p_2tuple_filter = igb_2tuple_filter_lookup(
3616                                         &filter_info->twotuple_list,
3617                                         &filter_2tuple);
3618                 if (p_2tuple_filter == NULL) {
3619                         PMD_DRV_LOG(ERR, "filter doesn't exist.");
3620                         return -ENOENT;
3621                 }
3622                 ntuple_filter->queue = p_2tuple_filter->queue;
3623                 break;
3624         default:
3625                 ret = -EINVAL;
3626                 break;
3627         }
3628
3629         return 0;
3630 }
3631
3632 /*
3633  * igb_ntuple_filter_handle - Handle operations for ntuple filter.
3634  * @dev: pointer to rte_eth_dev structure
3635  * @filter_op:operation will be taken.
3636  * @arg: a pointer to specific structure corresponding to the filter_op
3637  */
3638 static int
3639 igb_ntuple_filter_handle(struct rte_eth_dev *dev,
3640                                 enum rte_filter_op filter_op,
3641                                 void *arg)
3642 {
3643         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3644         int ret;
3645
3646         MAC_TYPE_FILTER_SUP(hw->mac.type);
3647
3648         if (filter_op == RTE_ETH_FILTER_NOP)
3649                 return 0;
3650
3651         if (arg == NULL) {
3652                 PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u.",
3653                             filter_op);
3654                 return -EINVAL;
3655         }
3656
3657         switch (filter_op) {
3658         case RTE_ETH_FILTER_ADD:
3659                 ret = igb_add_del_ntuple_filter(dev,
3660                         (struct rte_eth_ntuple_filter *)arg,
3661                         TRUE);
3662                 break;
3663         case RTE_ETH_FILTER_DELETE:
3664                 ret = igb_add_del_ntuple_filter(dev,
3665                         (struct rte_eth_ntuple_filter *)arg,
3666                         FALSE);
3667                 break;
3668         case RTE_ETH_FILTER_GET:
3669                 ret = igb_get_ntuple_filter(dev,
3670                         (struct rte_eth_ntuple_filter *)arg);
3671                 break;
3672         default:
3673                 PMD_DRV_LOG(ERR, "unsupported operation %u.", filter_op);
3674                 ret = -EINVAL;
3675                 break;
3676         }
3677         return ret;
3678 }
3679
3680 static inline int
3681 igb_ethertype_filter_lookup(struct e1000_filter_info *filter_info,
3682                         uint16_t ethertype)
3683 {
3684         int i;
3685
3686         for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
3687                 if (filter_info->ethertype_filters[i] == ethertype &&
3688                     (filter_info->ethertype_mask & (1 << i)))
3689                         return i;
3690         }
3691         return -1;
3692 }
3693
3694 static inline int
3695 igb_ethertype_filter_insert(struct e1000_filter_info *filter_info,
3696                         uint16_t ethertype)
3697 {
3698         int i;
3699
3700         for (i = 0; i < E1000_MAX_ETQF_FILTERS; i++) {
3701                 if (!(filter_info->ethertype_mask & (1 << i))) {
3702                         filter_info->ethertype_mask |= 1 << i;
3703                         filter_info->ethertype_filters[i] = ethertype;
3704                         return i;
3705                 }
3706         }
3707         return -1;
3708 }
3709
3710 static inline int
3711 igb_ethertype_filter_remove(struct e1000_filter_info *filter_info,
3712                         uint8_t idx)
3713 {
3714         if (idx >= E1000_MAX_ETQF_FILTERS)
3715                 return -1;
3716         filter_info->ethertype_mask &= ~(1 << idx);
3717         filter_info->ethertype_filters[idx] = 0;
3718         return idx;
3719 }
3720
3721
3722 static int
3723 igb_add_del_ethertype_filter(struct rte_eth_dev *dev,
3724                         struct rte_eth_ethertype_filter *filter,
3725                         bool add)
3726 {
3727         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3728         struct e1000_filter_info *filter_info =
3729                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3730         uint32_t etqf = 0;
3731         int ret;
3732
3733         if (filter->ether_type == ETHER_TYPE_IPv4 ||
3734                 filter->ether_type == ETHER_TYPE_IPv6) {
3735                 PMD_DRV_LOG(ERR, "unsupported ether_type(0x%04x) in"
3736                         " ethertype filter.", filter->ether_type);
3737                 return -EINVAL;
3738         }
3739
3740         if (filter->flags & RTE_ETHTYPE_FLAGS_MAC) {
3741                 PMD_DRV_LOG(ERR, "mac compare is unsupported.");
3742                 return -EINVAL;
3743         }
3744         if (filter->flags & RTE_ETHTYPE_FLAGS_DROP) {
3745                 PMD_DRV_LOG(ERR, "drop option is unsupported.");
3746                 return -EINVAL;
3747         }
3748
3749         ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
3750         if (ret >= 0 && add) {
3751                 PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter exists.",
3752                             filter->ether_type);
3753                 return -EEXIST;
3754         }
3755         if (ret < 0 && !add) {
3756                 PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
3757                             filter->ether_type);
3758                 return -ENOENT;
3759         }
3760
3761         if (add) {
3762                 ret = igb_ethertype_filter_insert(filter_info,
3763                         filter->ether_type);
3764                 if (ret < 0) {
3765                         PMD_DRV_LOG(ERR, "ethertype filters are full.");
3766                         return -ENOSYS;
3767                 }
3768
3769                 etqf |= E1000_ETQF_FILTER_ENABLE | E1000_ETQF_QUEUE_ENABLE;
3770                 etqf |= (uint32_t)(filter->ether_type & E1000_ETQF_ETHERTYPE);
3771                 etqf |= filter->queue << E1000_ETQF_QUEUE_SHIFT;
3772         } else {
3773                 ret = igb_ethertype_filter_remove(filter_info, (uint8_t)ret);
3774                 if (ret < 0)
3775                         return -ENOSYS;
3776         }
3777         E1000_WRITE_REG(hw, E1000_ETQF(ret), etqf);
3778         E1000_WRITE_FLUSH(hw);
3779
3780         return 0;
3781 }
3782
3783 static int
3784 igb_get_ethertype_filter(struct rte_eth_dev *dev,
3785                         struct rte_eth_ethertype_filter *filter)
3786 {
3787         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3788         struct e1000_filter_info *filter_info =
3789                 E1000_DEV_PRIVATE_TO_FILTER_INFO(dev->data->dev_private);
3790         uint32_t etqf;
3791         int ret;
3792
3793         ret = igb_ethertype_filter_lookup(filter_info, filter->ether_type);
3794         if (ret < 0) {
3795                 PMD_DRV_LOG(ERR, "ethertype (0x%04x) filter doesn't exist.",
3796                             filter->ether_type);
3797                 return -ENOENT;
3798         }
3799
3800         etqf = E1000_READ_REG(hw, E1000_ETQF(ret));
3801         if (etqf & E1000_ETQF_FILTER_ENABLE) {
3802                 filter->ether_type = etqf & E1000_ETQF_ETHERTYPE;
3803                 filter->flags = 0;
3804                 filter->queue = (etqf & E1000_ETQF_QUEUE) >>
3805                                 E1000_ETQF_QUEUE_SHIFT;
3806                 return 0;
3807         }
3808
3809         return -ENOENT;
3810 }
3811
3812 /*
3813  * igb_ethertype_filter_handle - Handle operations for ethertype filter.
3814  * @dev: pointer to rte_eth_dev structure
3815  * @filter_op:operation will be taken.
3816  * @arg: a pointer to specific structure corresponding to the filter_op
3817  */
3818 static int
3819 igb_ethertype_filter_handle(struct rte_eth_dev *dev,
3820                                 enum rte_filter_op filter_op,
3821                                 void *arg)
3822 {
3823         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3824         int ret;
3825
3826         MAC_TYPE_FILTER_SUP(hw->mac.type);
3827
3828         if (filter_op == RTE_ETH_FILTER_NOP)
3829                 return 0;
3830
3831         if (arg == NULL) {
3832                 PMD_DRV_LOG(ERR, "arg shouldn't be NULL for operation %u.",
3833                             filter_op);
3834                 return -EINVAL;
3835         }
3836
3837         switch (filter_op) {
3838         case RTE_ETH_FILTER_ADD:
3839                 ret = igb_add_del_ethertype_filter(dev,
3840                         (struct rte_eth_ethertype_filter *)arg,
3841                         TRUE);
3842                 break;
3843         case RTE_ETH_FILTER_DELETE:
3844                 ret = igb_add_del_ethertype_filter(dev,
3845                         (struct rte_eth_ethertype_filter *)arg,
3846                         FALSE);
3847                 break;
3848         case RTE_ETH_FILTER_GET:
3849                 ret = igb_get_ethertype_filter(dev,
3850                         (struct rte_eth_ethertype_filter *)arg);
3851                 break;
3852         default:
3853                 PMD_DRV_LOG(ERR, "unsupported operation %u.", filter_op);
3854                 ret = -EINVAL;
3855                 break;
3856         }
3857         return ret;
3858 }
3859
3860 static int
3861 eth_igb_filter_ctrl(struct rte_eth_dev *dev,
3862                      enum rte_filter_type filter_type,
3863                      enum rte_filter_op filter_op,
3864                      void *arg)
3865 {
3866         int ret = -EINVAL;
3867
3868         switch (filter_type) {
3869         case RTE_ETH_FILTER_NTUPLE:
3870                 ret = igb_ntuple_filter_handle(dev, filter_op, arg);
3871                 break;
3872         case RTE_ETH_FILTER_ETHERTYPE:
3873                 ret = igb_ethertype_filter_handle(dev, filter_op, arg);
3874                 break;
3875         case RTE_ETH_FILTER_SYN:
3876                 ret = eth_igb_syn_filter_handle(dev, filter_op, arg);
3877                 break;
3878         case RTE_ETH_FILTER_FLEXIBLE:
3879                 ret = eth_igb_flex_filter_handle(dev, filter_op, arg);
3880                 break;
3881         default:
3882                 PMD_DRV_LOG(WARNING, "Filter type (%d) not supported",
3883                                                         filter_type);
3884                 break;
3885         }
3886
3887         return ret;
3888 }
3889
3890 static int
3891 eth_igb_set_mc_addr_list(struct rte_eth_dev *dev,
3892                          struct ether_addr *mc_addr_set,
3893                          uint32_t nb_mc_addr)
3894 {
3895         struct e1000_hw *hw;
3896
3897         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3898         e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
3899         return 0;
3900 }
3901
3902 static int
3903 igb_timesync_enable(struct rte_eth_dev *dev)
3904 {
3905         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3906         uint32_t tsync_ctl;
3907
3908         /* Start incrementing the register used to timestamp PTP packets. */
3909         E1000_WRITE_REG(hw, E1000_TIMINCA, E1000_TIMINCA_INIT);
3910
3911         /* Enable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
3912         E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588),
3913                         (ETHER_TYPE_1588 |
3914                          E1000_ETQF_FILTER_ENABLE |
3915                          E1000_ETQF_1588));
3916
3917         /* Enable timestamping of received PTP packets. */
3918         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
3919         tsync_ctl |= E1000_TSYNCRXCTL_ENABLED;
3920         E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
3921
3922         /* Enable Timestamping of transmitted PTP packets. */
3923         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
3924         tsync_ctl |= E1000_TSYNCTXCTL_ENABLED;
3925         E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
3926
3927         return 0;
3928 }
3929
3930 static int
3931 igb_timesync_disable(struct rte_eth_dev *dev)
3932 {
3933         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3934         uint32_t tsync_ctl;
3935
3936         /* Disable timestamping of transmitted PTP packets. */
3937         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
3938         tsync_ctl &= ~E1000_TSYNCTXCTL_ENABLED;
3939         E1000_WRITE_REG(hw, E1000_TSYNCTXCTL, tsync_ctl);
3940
3941         /* Disable timestamping of received PTP packets. */
3942         tsync_ctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
3943         tsync_ctl &= ~E1000_TSYNCRXCTL_ENABLED;
3944         E1000_WRITE_REG(hw, E1000_TSYNCRXCTL, tsync_ctl);
3945
3946         /* Disable L2 filtering of IEEE1588/802.1AS Ethernet frame types. */
3947         E1000_WRITE_REG(hw, E1000_ETQF(E1000_ETQF_FILTER_1588), 0);
3948
3949         /* Stop incrementating the System Time registers. */
3950         E1000_WRITE_REG(hw, E1000_TIMINCA, 0);
3951
3952         return 0;
3953 }
3954
3955 static int
3956 igb_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
3957                                struct timespec *timestamp,
3958                                uint32_t flags __rte_unused)
3959 {
3960         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3961         uint32_t tsync_rxctl;
3962         uint32_t rx_stmpl;
3963         uint32_t rx_stmph;
3964
3965         tsync_rxctl = E1000_READ_REG(hw, E1000_TSYNCRXCTL);
3966         if ((tsync_rxctl & E1000_TSYNCRXCTL_VALID) == 0)
3967                 return -EINVAL;
3968
3969         rx_stmpl = E1000_READ_REG(hw, E1000_RXSTMPL);
3970         rx_stmph = E1000_READ_REG(hw, E1000_RXSTMPH);
3971
3972         timestamp->tv_sec = (uint64_t)(((uint64_t)rx_stmph << 32) | rx_stmpl);
3973         timestamp->tv_nsec = 0;
3974
3975         return  0;
3976 }
3977
3978 static int
3979 igb_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
3980                                struct timespec *timestamp)
3981 {
3982         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3983         uint32_t tsync_txctl;
3984         uint32_t tx_stmpl;
3985         uint32_t tx_stmph;
3986
3987         tsync_txctl = E1000_READ_REG(hw, E1000_TSYNCTXCTL);
3988         if ((tsync_txctl & E1000_TSYNCTXCTL_VALID) == 0)
3989                 return -EINVAL;
3990
3991         tx_stmpl = E1000_READ_REG(hw, E1000_TXSTMPL);
3992         tx_stmph = E1000_READ_REG(hw, E1000_TXSTMPH);
3993
3994         timestamp->tv_sec = (uint64_t)(((uint64_t)tx_stmph << 32) | tx_stmpl);
3995         timestamp->tv_nsec = 0;
3996
3997         return  0;
3998 }
3999
4000 static int
4001 eth_igb_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4002 {
4003         int count = 0;
4004         int g_ind = 0;
4005         const struct reg_info *reg_group;
4006
4007         while ((reg_group = igb_regs[g_ind++]))
4008                 count += igb_reg_group_count(reg_group);
4009
4010         return count;
4011 }
4012
4013 static int
4014 igbvf_get_reg_length(struct rte_eth_dev *dev __rte_unused)
4015 {
4016         int count = 0;
4017         int g_ind = 0;
4018         const struct reg_info *reg_group;
4019
4020         while ((reg_group = igbvf_regs[g_ind++]))
4021                 count += igb_reg_group_count(reg_group);
4022
4023         return count;
4024 }
4025
4026 static int
4027 eth_igb_get_regs(struct rte_eth_dev *dev,
4028         struct rte_dev_reg_info *regs)
4029 {
4030         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4031         uint32_t *data = regs->data;
4032         int g_ind = 0;
4033         int count = 0;
4034         const struct reg_info *reg_group;
4035
4036         /* Support only full register dump */
4037         if ((regs->length == 0) ||
4038             (regs->length == (uint32_t)eth_igb_get_reg_length(dev))) {
4039                 regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4040                         hw->device_id;
4041                 while ((reg_group = igb_regs[g_ind++]))
4042                         count += igb_read_regs_group(dev, &data[count],
4043                                                         reg_group);
4044                 return 0;
4045         }
4046
4047         return -ENOTSUP;
4048 }
4049
4050 static int
4051 igbvf_get_regs(struct rte_eth_dev *dev,
4052         struct rte_dev_reg_info *regs)
4053 {
4054         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4055         uint32_t *data = regs->data;
4056         int g_ind = 0;
4057         int count = 0;
4058         const struct reg_info *reg_group;
4059
4060         /* Support only full register dump */
4061         if ((regs->length == 0) ||
4062             (regs->length == (uint32_t)igbvf_get_reg_length(dev))) {
4063                 regs->version = hw->mac.type << 24 | hw->revision_id << 16 |
4064                         hw->device_id;
4065                 while ((reg_group = igbvf_regs[g_ind++]))
4066                         count += igb_read_regs_group(dev, &data[count],
4067                                                         reg_group);
4068                 return 0;
4069         }
4070
4071         return -ENOTSUP;
4072 }
4073
4074 static int
4075 eth_igb_get_eeprom_length(struct rte_eth_dev *dev)
4076 {
4077         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4078
4079         /* Return unit is byte count */
4080         return hw->nvm.word_size * 2;
4081 }
4082
4083 static int
4084 eth_igb_get_eeprom(struct rte_eth_dev *dev,
4085         struct rte_dev_eeprom_info *in_eeprom)
4086 {
4087         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4088         struct e1000_nvm_info *nvm = &hw->nvm;
4089         uint16_t *data = in_eeprom->data;
4090         int first, length;
4091
4092         first = in_eeprom->offset >> 1;
4093         length = in_eeprom->length >> 1;
4094         if ((first >= hw->nvm.word_size) ||
4095             ((first + length) >= hw->nvm.word_size))
4096                 return -EINVAL;
4097
4098         in_eeprom->magic = hw->vendor_id |
4099                 ((uint32_t)hw->device_id << 16);
4100
4101         if ((nvm->ops.read) == NULL)
4102                 return -ENOTSUP;
4103
4104         return nvm->ops.read(hw, first, length, data);
4105 }
4106
4107 static int
4108 eth_igb_set_eeprom(struct rte_eth_dev *dev,
4109         struct rte_dev_eeprom_info *in_eeprom)
4110 {
4111         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4112         struct e1000_nvm_info *nvm = &hw->nvm;
4113         uint16_t *data = in_eeprom->data;
4114         int first, length;
4115
4116         first = in_eeprom->offset >> 1;
4117         length = in_eeprom->length >> 1;
4118         if ((first >= hw->nvm.word_size) ||
4119             ((first + length) >= hw->nvm.word_size))
4120                 return -EINVAL;
4121
4122         in_eeprom->magic = (uint32_t)hw->vendor_id |
4123                 ((uint32_t)hw->device_id << 16);
4124
4125         if ((nvm->ops.write) == NULL)
4126                 return -ENOTSUP;
4127         return nvm->ops.write(hw,  first, length, data);
4128 }
4129
4130 static struct rte_driver pmd_igb_drv = {
4131         .type = PMD_PDEV,
4132         .init = rte_igb_pmd_init,
4133 };
4134
4135 static struct rte_driver pmd_igbvf_drv = {
4136         .type = PMD_PDEV,
4137         .init = rte_igbvf_pmd_init,
4138 };
4139
4140 #ifdef RTE_NEXT_ABI
4141 static int
4142 eth_igb_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
4143 {
4144         struct e1000_hw *hw =
4145                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4146         uint32_t mask = 1 << queue_id;
4147
4148         E1000_WRITE_REG(hw, E1000_EIMC, mask);
4149         E1000_WRITE_FLUSH(hw);
4150
4151         return 0;
4152 }
4153
4154 static int
4155 eth_igb_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
4156 {
4157         struct e1000_hw *hw =
4158                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4159         uint32_t mask = 1 << queue_id;
4160         uint32_t regval;
4161
4162         regval = E1000_READ_REG(hw, E1000_EIMS);
4163         E1000_WRITE_REG(hw, E1000_EIMS, regval | mask);
4164         E1000_WRITE_FLUSH(hw);
4165
4166         rte_intr_enable(&dev->pci_dev->intr_handle);
4167
4168         return 0;
4169 }
4170
4171 static void
4172 eth_igb_write_ivar(struct e1000_hw *hw, uint8_t  msix_vector,
4173                    uint8_t index, uint8_t offset)
4174 {
4175         uint32_t val = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4176
4177         /* clear bits */
4178         val &= ~((uint32_t)0xFF << offset);
4179
4180         /* write vector and valid bit */
4181         val |= (msix_vector | E1000_IVAR_VALID) << offset;
4182
4183         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, val);
4184 }
4185
4186 static void
4187 eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
4188                            uint8_t queue, uint8_t msix_vector)
4189 {
4190         uint32_t tmp = 0;
4191
4192         if (hw->mac.type == e1000_82575) {
4193                 if (direction == 0)
4194                         tmp = E1000_EICR_RX_QUEUE0 << queue;
4195                 else if (direction == 1)
4196                         tmp = E1000_EICR_TX_QUEUE0 << queue;
4197                 E1000_WRITE_REG(hw, E1000_MSIXBM(msix_vector), tmp);
4198         } else if (hw->mac.type == e1000_82576) {
4199                 if ((direction == 0) || (direction == 1))
4200                         eth_igb_write_ivar(hw, msix_vector, queue & 0x7,
4201                                            ((queue & 0x8) << 1) +
4202                                            8 * direction);
4203         } else if ((hw->mac.type == e1000_82580) ||
4204                         (hw->mac.type == e1000_i350) ||
4205                         (hw->mac.type == e1000_i354) ||
4206                         (hw->mac.type == e1000_i210) ||
4207                         (hw->mac.type == e1000_i211)) {
4208                 if ((direction == 0) || (direction == 1))
4209                         eth_igb_write_ivar(hw, msix_vector,
4210                                            queue >> 1,
4211                                            ((queue & 0x1) << 4) +
4212                                            8 * direction);
4213         }
4214 }
4215 #endif
4216
4217 /* Sets up the hardware to generate MSI-X interrupts properly
4218  * @hw
4219  *  board private structure
4220  */
4221 static void
4222 eth_igb_configure_msix_intr(struct rte_eth_dev *dev)
4223 {
4224 #ifdef RTE_NEXT_ABI
4225         int queue_id;
4226         uint32_t tmpval, regval, intr_mask;
4227         struct e1000_hw *hw =
4228                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
4229         uint32_t vec = 0;
4230 #endif
4231         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
4232
4233         /* won't configure msix register if no mapping is done
4234          * between intr vector and event fd
4235          */
4236         if (!rte_intr_dp_is_en(intr_handle))
4237                 return;
4238
4239 #ifdef RTE_NEXT_ABI
4240         /* set interrupt vector for other causes */
4241         if (hw->mac.type == e1000_82575) {
4242                 tmpval = E1000_READ_REG(hw, E1000_CTRL_EXT);
4243                 /* enable MSI-X PBA support */
4244                 tmpval |= E1000_CTRL_EXT_PBA_CLR;
4245
4246                 /* Auto-Mask interrupts upon ICR read */
4247                 tmpval |= E1000_CTRL_EXT_EIAME;
4248                 tmpval |= E1000_CTRL_EXT_IRCA;
4249
4250                 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmpval);
4251
4252                 /* enable msix_other interrupt */
4253                 E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0), 0, E1000_EIMS_OTHER);
4254                 regval = E1000_READ_REG(hw, E1000_EIAC);
4255                 E1000_WRITE_REG(hw, E1000_EIAC, regval | E1000_EIMS_OTHER);
4256                 regval = E1000_READ_REG(hw, E1000_EIAM);
4257                 E1000_WRITE_REG(hw, E1000_EIMS, regval | E1000_EIMS_OTHER);
4258         } else if ((hw->mac.type == e1000_82576) ||
4259                         (hw->mac.type == e1000_82580) ||
4260                         (hw->mac.type == e1000_i350) ||
4261                         (hw->mac.type == e1000_i354) ||
4262                         (hw->mac.type == e1000_i210) ||
4263                         (hw->mac.type == e1000_i211)) {
4264                 /* turn on MSI-X capability first */
4265                 E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_MSIX_MODE |
4266                                         E1000_GPIE_PBA | E1000_GPIE_EIAME |
4267                                         E1000_GPIE_NSICR);
4268
4269                 intr_mask = (1 << intr_handle->max_intr) - 1;
4270                 regval = E1000_READ_REG(hw, E1000_EIAC);
4271                 E1000_WRITE_REG(hw, E1000_EIAC, regval | intr_mask);
4272
4273                 /* enable msix_other interrupt */
4274                 regval = E1000_READ_REG(hw, E1000_EIMS);
4275                 E1000_WRITE_REG(hw, E1000_EIMS, regval | intr_mask);
4276                 tmpval = (dev->data->nb_rx_queues | E1000_IVAR_VALID) << 8;
4277                 E1000_WRITE_REG(hw, E1000_IVAR_MISC, tmpval);
4278         }
4279
4280         /* use EIAM to auto-mask when MSI-X interrupt
4281          * is asserted, this saves a register write for every interrupt
4282          */
4283         intr_mask = (1 << intr_handle->nb_efd) - 1;
4284         regval = E1000_READ_REG(hw, E1000_EIAM);
4285         E1000_WRITE_REG(hw, E1000_EIAM, regval | intr_mask);
4286
4287         for (queue_id = 0; queue_id < dev->data->nb_rx_queues; queue_id++) {
4288                 eth_igb_assign_msix_vector(hw, 0, queue_id, vec);
4289                 intr_handle->intr_vec[queue_id] = vec;
4290                 if (vec < intr_handle->nb_efd - 1)
4291                         vec++;
4292         }
4293
4294         E1000_WRITE_FLUSH(hw);
4295 #endif
4296 }
4297
4298 PMD_REGISTER_DRIVER(pmd_igb_drv);
4299 PMD_REGISTER_DRIVER(pmd_igbvf_drv);