drivers/net/ixgbe/ixgbe_pf.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <errno.h>
   7 #include <stdint.h>
   8 #include <stdlib.h>
   9 #include <unistd.h>
  10 #include <stdarg.h>
  11 #include <inttypes.h>
  12
  13 #include <rte_interrupts.h>
  14 #include <rte_log.h>
  15 #include <rte_debug.h>
  16 #include <rte_eal.h>
  17 #include <rte_ether.h>
  18 #include <ethdev_driver.h>
  19 #include <rte_memcpy.h>
  20 #include <rte_malloc.h>
  21 #include <rte_random.h>
  22
  23 #include "base/ixgbe_common.h"
  24 #include "ixgbe_ethdev.h"
  25 #include "rte_pmd_ixgbe.h"
  26
  27 #define IXGBE_MAX_VFTA     (128)
  28 #define IXGBE_VF_MSG_SIZE_DEFAULT 1
  29 #define IXGBE_VF_GET_QUEUE_MSG_SIZE 5
  30 #define IXGBE_ETHERTYPE_FLOW_CTRL 0x8808
  31
  32 static inline uint16_t
  33 dev_num_vf(struct rte_eth_dev *eth_dev)
  34 {
  35         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
  36
  37         return pci_dev->max_vfs;
  38 }
  39
  40 static inline
  41 int ixgbe_vf_perm_addr_gen(struct rte_eth_dev *dev, uint16_t vf_num)
  42 {
  43         unsigned char vf_mac_addr[RTE_ETHER_ADDR_LEN];
  44         struct ixgbe_vf_info *vfinfo =
  45                 *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
  46         uint16_t vfn;
  47
  48         for (vfn = 0; vfn < vf_num; vfn++) {
  49                 rte_eth_random_addr(vf_mac_addr);
  50                 /* keep the random address as default */
  51                 memcpy(vfinfo[vfn].vf_mac_addresses, vf_mac_addr,
  52                            RTE_ETHER_ADDR_LEN);
  53         }
  54
  55         return 0;
  56 }
  57
  58 static inline int
  59 ixgbe_mb_intr_setup(struct rte_eth_dev *dev)
  60 {
  61         struct ixgbe_interrupt *intr =
  62                 IXGBE_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
  63
  64         intr->mask |= IXGBE_EICR_MAILBOX;
  65
  66         return 0;
  67 }
  68
  69 int ixgbe_pf_host_init(struct rte_eth_dev *eth_dev)
  70 {
  71         struct ixgbe_vf_info **vfinfo =
  72                 IXGBE_DEV_PRIVATE_TO_P_VFDATA(eth_dev->data->dev_private);
  73         struct ixgbe_uta_info *uta_info =
  74         IXGBE_DEV_PRIVATE_TO_UTA(eth_dev->data->dev_private);
  75         struct ixgbe_hw *hw =
  76                 IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
  77         uint16_t vf_num;
  78         uint8_t nb_queue;
  79         int ret = 0;
  80
  81         PMD_INIT_FUNC_TRACE();
  82
  83         RTE_ETH_DEV_SRIOV(eth_dev).active = 0;
  84         vf_num = dev_num_vf(eth_dev);
  85         if (vf_num == 0)
  86                 return ret;
  87
  88         *vfinfo = rte_zmalloc("vf_info", sizeof(struct ixgbe_vf_info) * vf_num, 0);
  89         if (*vfinfo == NULL) {
  90                 PMD_INIT_LOG(ERR,
  91                         "Cannot allocate memory for private VF data");
  92                 return -ENOMEM;
  93         }
  94
  95         ret = rte_eth_switch_domain_alloc(&(*vfinfo)->switch_domain_id);
  96         if (ret) {
  97                 PMD_INIT_LOG(ERR,
  98                         "failed to allocate switch domain for device %d", ret);
  99                 rte_free(*vfinfo);
 100                 *vfinfo = NULL;
 101                 return ret;
 102         }
 103
 104         memset(uta_info, 0, sizeof(struct ixgbe_uta_info));
 105         hw->mac.mc_filter_type = 0;
 106
 107         if (vf_num >= ETH_32_POOLS) {
 108                 nb_queue = 2;
 109                 RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_64_POOLS;
 110         } else if (vf_num >= ETH_16_POOLS) {
 111                 nb_queue = 4;
 112                 RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_32_POOLS;
 113         } else {
 114                 nb_queue = 8;
 115                 RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_16_POOLS;
 116         }
 117
 118         RTE_ETH_DEV_SRIOV(eth_dev).nb_q_per_pool = nb_queue;
 119         RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx = vf_num;
 120         RTE_ETH_DEV_SRIOV(eth_dev).def_pool_q_idx = (uint16_t)(vf_num * nb_queue);
 121
 122         ixgbe_vf_perm_addr_gen(eth_dev, vf_num);
 123
 124         /* init_mailbox_params */
 125         hw->mbx.ops.init_params(hw);
 126
 127         /* set mb interrupt mask */
 128         ixgbe_mb_intr_setup(eth_dev);
 129
 130         return ret;
 131 }
 132
 133 void ixgbe_pf_host_uninit(struct rte_eth_dev *eth_dev)
 134 {
 135         struct ixgbe_vf_info **vfinfo;
 136         uint16_t vf_num;
 137         int ret;
 138
 139         PMD_INIT_FUNC_TRACE();
 140
 141         RTE_ETH_DEV_SRIOV(eth_dev).active = 0;
 142         RTE_ETH_DEV_SRIOV(eth_dev).nb_q_per_pool = 0;
 143         RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx = 0;
 144         RTE_ETH_DEV_SRIOV(eth_dev).def_pool_q_idx = 0;
 145
 146         vf_num = dev_num_vf(eth_dev);
 147         if (vf_num == 0)
 148                 return;
 149
 150         vfinfo = IXGBE_DEV_PRIVATE_TO_P_VFDATA(eth_dev->data->dev_private);
 151         if (*vfinfo == NULL)
 152                 return;
 153
 154         ret = rte_eth_switch_domain_free((*vfinfo)->switch_domain_id);
 155         if (ret)
 156                 PMD_INIT_LOG(WARNING, "failed to free switch domain: %d", ret);
 157
 158         rte_free(*vfinfo);
 159         *vfinfo = NULL;
 160 }
 161
 162 static void
 163 ixgbe_add_tx_flow_control_drop_filter(struct rte_eth_dev *eth_dev)
 164 {
 165         struct ixgbe_hw *hw =
 166                 IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 167         struct ixgbe_filter_info *filter_info =
 168                 IXGBE_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
 169         uint16_t vf_num;
 170         int i;
 171         struct ixgbe_ethertype_filter ethertype_filter;
 172
 173         if (!hw->mac.ops.set_ethertype_anti_spoofing) {
 174                 PMD_DRV_LOG(INFO, "ether type anti-spoofing is not supported.\n");
 175                 return;
 176         }
 177
 178         i = ixgbe_ethertype_filter_lookup(filter_info,
 179                                           IXGBE_ETHERTYPE_FLOW_CTRL);
 180         if (i >= 0) {
 181                 PMD_DRV_LOG(ERR, "A ether type filter entity for flow control already exists!\n");
 182                 return;
 183         }
 184
 185         ethertype_filter.ethertype = IXGBE_ETHERTYPE_FLOW_CTRL;
 186         ethertype_filter.etqf = IXGBE_ETQF_FILTER_EN |
 187                                 IXGBE_ETQF_TX_ANTISPOOF |
 188                                 IXGBE_ETHERTYPE_FLOW_CTRL;
 189         ethertype_filter.etqs = 0;
 190         ethertype_filter.conf = TRUE;
 191         i = ixgbe_ethertype_filter_insert(filter_info,
 192                                           &ethertype_filter);
 193         if (i < 0) {
 194                 PMD_DRV_LOG(ERR, "Cannot find an unused ether type filter entity for flow control.\n");
 195                 return;
 196         }
 197
 198         IXGBE_WRITE_REG(hw, IXGBE_ETQF(i),
 199                         (IXGBE_ETQF_FILTER_EN |
 200                         IXGBE_ETQF_TX_ANTISPOOF |
 201                         IXGBE_ETHERTYPE_FLOW_CTRL));
 202
 203         vf_num = dev_num_vf(eth_dev);
 204         for (i = 0; i < vf_num; i++)
 205                 hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
 206 }
 207
 208 int ixgbe_pf_host_configure(struct rte_eth_dev *eth_dev)
 209 {
 210         uint32_t vtctl, fcrth;
 211         uint32_t vfre_slot, vfre_offset;
 212         uint16_t vf_num;
 213         const uint8_t VFRE_SHIFT = 5;  /* VFRE 32 bits per slot */
 214         const uint8_t VFRE_MASK = (uint8_t)((1U << VFRE_SHIFT) - 1);
 215         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 216         uint32_t gpie, gcr_ext;
 217         uint32_t vlanctrl;
 218         int i;
 219
 220         vf_num = dev_num_vf(eth_dev);
 221         if (vf_num == 0)
 222                 return -1;
 223
 224         /* enable VMDq and set the default pool for PF */
 225         vtctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
 226         vtctl |= IXGBE_VMD_CTL_VMDQ_EN;
 227         vtctl &= ~IXGBE_VT_CTL_POOL_MASK;
 228         vtctl |= RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx
 229                 << IXGBE_VT_CTL_POOL_SHIFT;
 230         vtctl |= IXGBE_VT_CTL_REPLEN;
 231         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
 232
 233         vfre_offset = vf_num & VFRE_MASK;
 234         vfre_slot = (vf_num >> VFRE_SHIFT) > 0 ? 1 : 0;
 235
 236         /* Enable pools reserved to PF only */
 237         IXGBE_WRITE_REG(hw, IXGBE_VFRE(vfre_slot), (~0U) << vfre_offset);
 238         IXGBE_WRITE_REG(hw, IXGBE_VFRE(vfre_slot ^ 1), vfre_slot - 1);
 239         IXGBE_WRITE_REG(hw, IXGBE_VFTE(vfre_slot), (~0U) << vfre_offset);
 240         IXGBE_WRITE_REG(hw, IXGBE_VFTE(vfre_slot ^ 1), vfre_slot - 1);
 241
 242         /* PFDMA Tx General Switch Control Enables VMDQ loopback */
 243         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
 244
 245         /* clear VMDq map to perment rar 0 */
 246         hw->mac.ops.clear_vmdq(hw, 0, IXGBE_CLEAR_VMDQ_ALL);
 247
 248         /* clear VMDq map to scan rar 127 */
 249         IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(hw->mac.num_rar_entries), 0);
 250         IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(hw->mac.num_rar_entries), 0);
 251
 252         /* set VMDq map to default PF pool */
 253         hw->mac.ops.set_vmdq(hw, 0, RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx);
 254
 255         /*
 256          * SW msut set GCR_EXT.VT_Mode the same as GPIE.VT_Mode
 257          */
 258         gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
 259         gcr_ext &= ~IXGBE_GCR_EXT_VT_MODE_MASK;
 260
 261         gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
 262         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
 263         gpie |= IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT;
 264
 265         switch (RTE_ETH_DEV_SRIOV(eth_dev).active) {
 266         case ETH_64_POOLS:
 267                 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
 268                 gpie |= IXGBE_GPIE_VTMODE_64;
 269                 break;
 270         case ETH_32_POOLS:
 271                 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_32;
 272                 gpie |= IXGBE_GPIE_VTMODE_32;
 273                 break;
 274         case ETH_16_POOLS:
 275                 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_16;
 276                 gpie |= IXGBE_GPIE_VTMODE_16;
 277                 break;
 278         }
 279
 280         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
 281         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
 282
 283         /*
 284          * enable vlan filtering and allow all vlan tags through
 285          */
 286         vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
 287         vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
 288         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
 289
 290         /* VFTA - enable all vlan filters */
 291         for (i = 0; i < IXGBE_MAX_VFTA; i++)
 292                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
 293
 294         /* Enable MAC Anti-Spoofing */
 295         hw->mac.ops.set_mac_anti_spoofing(hw, FALSE, vf_num);
 296
 297         /* set flow control threshold to max to avoid tx switch hang */
 298         for (i = 0; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
 299                 IXGBE_WRITE_REG(hw, IXGBE_FCRTL_82599(i), 0);
 300                 fcrth = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i)) - 32;
 301                 IXGBE_WRITE_REG(hw, IXGBE_FCRTH_82599(i), fcrth);
 302         }
 303
 304         ixgbe_add_tx_flow_control_drop_filter(eth_dev);
 305
 306         return 0;
 307 }
 308
 309 static void
 310 set_rx_mode(struct rte_eth_dev *dev)
 311 {
 312         struct rte_eth_dev_data *dev_data = dev->data;
 313         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 314         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
 315         uint16_t vfn = dev_num_vf(dev);
 316
 317         /* Check for Promiscuous and All Multicast modes */
 318         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
 319
 320         /* set all bits that we expect to always be set */
 321         fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
 322         fctrl |= IXGBE_FCTRL_BAM;
 323
 324         /* clear the bits we are changing the status of */
 325         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
 326
 327         if (dev_data->promiscuous) {
 328                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
 329                 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
 330         } else {
 331                 if (dev_data->all_multicast) {
 332                         fctrl |= IXGBE_FCTRL_MPE;
 333                         vmolr |= IXGBE_VMOLR_MPE;
 334                 } else {
 335                         vmolr |= IXGBE_VMOLR_ROMPE;
 336                 }
 337         }
 338
 339         if (hw->mac.type != ixgbe_mac_82598EB) {
 340                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(vfn)) &
 341                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
 342                            IXGBE_VMOLR_ROPE);
 343                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vfn), vmolr);
 344         }
 345
 346         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
 347
 348         ixgbe_vlan_hw_strip_config(dev);
 349 }
 350
 351 static inline void
 352 ixgbe_vf_reset_event(struct rte_eth_dev *dev, uint16_t vf)
 353 {
 354         struct ixgbe_hw *hw =
 355                 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 356         struct ixgbe_vf_info *vfinfo =
 357                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 358         int rar_entry = hw->mac.num_rar_entries - (vf + 1);
 359         uint32_t vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
 360
 361         vmolr |= (IXGBE_VMOLR_ROPE |
 362                         IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
 363         IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
 364
 365         IXGBE_WRITE_REG(hw, IXGBE_VMVIR(vf), 0);
 366
 367         /* reset multicast table array for vf */
 368         vfinfo[vf].num_vf_mc_hashes = 0;
 369
 370         /* reset rx mode */
 371         set_rx_mode(dev);
 372
 373         hw->mac.ops.clear_rar(hw, rar_entry);
 374 }
 375
 376 static inline void
 377 ixgbe_vf_reset_msg(struct rte_eth_dev *dev, uint16_t vf)
 378 {
 379         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 380         uint32_t reg;
 381         uint32_t reg_offset, vf_shift;
 382         const uint8_t VFRE_SHIFT = 5;  /* VFRE 32 bits per slot */
 383         const uint8_t VFRE_MASK = (uint8_t)((1U << VFRE_SHIFT) - 1);
 384         uint8_t  nb_q_per_pool;
 385         int i;
 386
 387         vf_shift = vf & VFRE_MASK;
 388         reg_offset = (vf >> VFRE_SHIFT) > 0 ? 1 : 0;
 389
 390         /* enable transmit for vf */
 391         reg = IXGBE_READ_REG(hw, IXGBE_VFTE(reg_offset));
 392         reg |= (reg | (1 << vf_shift));
 393         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), reg);
 394
 395         /* enable all queue drop for IOV */
 396         nb_q_per_pool = RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
 397         for (i = vf * nb_q_per_pool; i < (vf + 1) * nb_q_per_pool; i++) {
 398                 IXGBE_WRITE_FLUSH(hw);
 399                 reg = IXGBE_QDE_ENABLE | IXGBE_QDE_WRITE;
 400                 reg |= i << IXGBE_QDE_IDX_SHIFT;
 401                 IXGBE_WRITE_REG(hw, IXGBE_QDE, reg);
 402         }
 403
 404         /* enable receive for vf */
 405         reg = IXGBE_READ_REG(hw, IXGBE_VFRE(reg_offset));
 406         reg |= (reg | (1 << vf_shift));
 407         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), reg);
 408
 409         /* Enable counting of spoofed packets in the SSVPC register */
 410         reg = IXGBE_READ_REG(hw, IXGBE_VMECM(reg_offset));
 411         reg |= (1 << vf_shift);
 412         IXGBE_WRITE_REG(hw, IXGBE_VMECM(reg_offset), reg);
 413
 414         ixgbe_vf_reset_event(dev, vf);
 415 }
 416
 417 static int
 418 ixgbe_disable_vf_mc_promisc(struct rte_eth_dev *dev, uint32_t vf)
 419 {
 420         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 421         uint32_t vmolr;
 422
 423         vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
 424
 425         PMD_DRV_LOG(INFO, "VF %u: disabling multicast promiscuous\n", vf);
 426
 427         vmolr &= ~IXGBE_VMOLR_MPE;
 428
 429         IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
 430
 431         return 0;
 432 }
 433
 434 static int
 435 ixgbe_vf_reset(struct rte_eth_dev *dev, uint16_t vf, uint32_t *msgbuf)
 436 {
 437         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 438         struct ixgbe_vf_info *vfinfo =
 439                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 440         unsigned char *vf_mac = vfinfo[vf].vf_mac_addresses;
 441         int rar_entry = hw->mac.num_rar_entries - (vf + 1);
 442         uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
 443
 444         ixgbe_vf_reset_msg(dev, vf);
 445
 446         hw->mac.ops.set_rar(hw, rar_entry, vf_mac, vf, IXGBE_RAH_AV);
 447
 448         /* Disable multicast promiscuous at reset */
 449         ixgbe_disable_vf_mc_promisc(dev, vf);
 450
 451         /* reply to reset with ack and vf mac address */
 452         msgbuf[0] = IXGBE_VF_RESET | IXGBE_VT_MSGTYPE_ACK;
 453         rte_memcpy(new_mac, vf_mac, RTE_ETHER_ADDR_LEN);
 454         /*
 455          * Piggyback the multicast filter type so VF can compute the
 456          * correct vectors
 457          */
 458         msgbuf[3] = hw->mac.mc_filter_type;
 459         ixgbe_write_mbx(hw, msgbuf, IXGBE_VF_PERMADDR_MSG_LEN, vf);
 460
 461         return 0;
 462 }
 463
 464 static int
 465 ixgbe_vf_set_mac_addr(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 466 {
 467         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 468         struct ixgbe_vf_info *vfinfo =
 469                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 470         int rar_entry = hw->mac.num_rar_entries - (vf + 1);
 471         uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
 472
 473         if (rte_is_valid_assigned_ether_addr(
 474                         (struct rte_ether_addr *)new_mac)) {
 475                 rte_memcpy(vfinfo[vf].vf_mac_addresses, new_mac, 6);
 476                 return hw->mac.ops.set_rar(hw, rar_entry, new_mac, vf, IXGBE_RAH_AV);
 477         }
 478         return -1;
 479 }
 480
 481 static int
 482 ixgbe_vf_set_multicast(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 483 {
 484         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 485         struct ixgbe_vf_info *vfinfo =
 486                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 487         int nb_entries = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK) >>
 488                 IXGBE_VT_MSGINFO_SHIFT;
 489         uint16_t *hash_list = (uint16_t *)&msgbuf[1];
 490         uint32_t mta_idx;
 491         uint32_t mta_shift;
 492         const uint32_t IXGBE_MTA_INDEX_MASK = 0x7F;
 493         const uint32_t IXGBE_MTA_BIT_SHIFT = 5;
 494         const uint32_t IXGBE_MTA_BIT_MASK = (0x1 << IXGBE_MTA_BIT_SHIFT) - 1;
 495         uint32_t reg_val;
 496         int i;
 497         u32 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
 498
 499         /* Disable multicast promiscuous first */
 500         ixgbe_disable_vf_mc_promisc(dev, vf);
 501
 502         /* only so many hash values supported */
 503         nb_entries = RTE_MIN(nb_entries, IXGBE_MAX_VF_MC_ENTRIES);
 504
 505         /* store the mc entries  */
 506         vfinfo->num_vf_mc_hashes = (uint16_t)nb_entries;
 507         for (i = 0; i < nb_entries; i++) {
 508                 vfinfo->vf_mc_hashes[i] = hash_list[i];
 509         }
 510
 511         if (nb_entries == 0) {
 512                 vmolr &= ~IXGBE_VMOLR_ROMPE;
 513                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
 514                 return 0;
 515         }
 516
 517         for (i = 0; i < vfinfo->num_vf_mc_hashes; i++) {
 518                 mta_idx = (vfinfo->vf_mc_hashes[i] >> IXGBE_MTA_BIT_SHIFT)
 519                                 & IXGBE_MTA_INDEX_MASK;
 520                 mta_shift = vfinfo->vf_mc_hashes[i] & IXGBE_MTA_BIT_MASK;
 521                 reg_val = IXGBE_READ_REG(hw, IXGBE_MTA(mta_idx));
 522                 reg_val |= (1 << mta_shift);
 523                 IXGBE_WRITE_REG(hw, IXGBE_MTA(mta_idx), reg_val);
 524         }
 525
 526         vmolr |= IXGBE_VMOLR_ROMPE;
 527         IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
 528
 529         return 0;
 530 }
 531
 532 static int
 533 ixgbe_vf_set_vlan(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 534 {
 535         int add, vid;
 536         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 537         struct ixgbe_vf_info *vfinfo =
 538                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 539
 540         add = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK)
 541                 >> IXGBE_VT_MSGINFO_SHIFT;
 542         vid = (msgbuf[1] & IXGBE_VLVF_VLANID_MASK);
 543
 544         if (add)
 545                 vfinfo[vf].vlan_count++;
 546         else if (vfinfo[vf].vlan_count)
 547                 vfinfo[vf].vlan_count--;
 548         return hw->mac.ops.set_vfta(hw, vid, vf, (bool)add, false);
 549 }
 550
 551 static int
 552 ixgbe_set_vf_lpe(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 553 {
 554         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 555         uint32_t max_frame = msgbuf[1];
 556         uint32_t max_frs;
 557         uint32_t hlreg0;
 558
 559         /* X540 and X550 support jumbo frames in IOV mode */
 560         if (hw->mac.type != ixgbe_mac_X540 &&
 561                 hw->mac.type != ixgbe_mac_X550 &&
 562                 hw->mac.type != ixgbe_mac_X550EM_x &&
 563                 hw->mac.type != ixgbe_mac_X550EM_a) {
 564                 struct ixgbe_vf_info *vfinfo =
 565                         *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
 566
 567                 switch (vfinfo[vf].api_version) {
 568                 case ixgbe_mbox_api_11:
 569                 case ixgbe_mbox_api_12:
 570                 case ixgbe_mbox_api_13:
 571                          /**
 572                           * Version 1.1&1.2&1.3 supports jumbo frames on VFs
 573                           * if PF has jumbo frames enabled which means legacy
 574                           * VFs are disabled.
 575                           */
 576                         if (dev->data->mtu > RTE_ETHER_MTU)
 577                                 break;
 578                         /* fall through */
 579                 default:
 580                         /**
 581                          * If the PF or VF are running w/ jumbo frames enabled,
 582                          * we return -1 as we cannot support jumbo frames on
 583                          * legacy VFs.
 584                          */
 585                         if (max_frame > IXGBE_ETH_MAX_LEN ||
 586                                         dev->data->mtu > RTE_ETHER_MTU)
 587                                 return -1;
 588                         break;
 589                 }
 590         }
 591
 592         if (max_frame < RTE_ETHER_MIN_LEN ||
 593                         max_frame > RTE_ETHER_MAX_JUMBO_FRAME_LEN)
 594                 return -1;
 595
 596         max_frs = (IXGBE_READ_REG(hw, IXGBE_MAXFRS) &
 597                    IXGBE_MHADD_MFS_MASK) >> IXGBE_MHADD_MFS_SHIFT;
 598         if (max_frs < max_frame) {
 599                 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
 600                 if (max_frame > IXGBE_ETH_MAX_LEN)
 601                         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
 602                 else
 603                         hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
 604                 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
 605
 606                 max_frs = max_frame << IXGBE_MHADD_MFS_SHIFT;
 607                 IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, max_frs);
 608         }
 609
 610         return 0;
 611 }
 612
 613 static int
 614 ixgbe_negotiate_vf_api(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 615 {
 616         uint32_t api_version = msgbuf[1];
 617         struct ixgbe_vf_info *vfinfo =
 618                 *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
 619
 620         switch (api_version) {
 621         case ixgbe_mbox_api_10:
 622         case ixgbe_mbox_api_11:
 623         case ixgbe_mbox_api_12:
 624         case ixgbe_mbox_api_13:
 625                 vfinfo[vf].api_version = (uint8_t)api_version;
 626                 return 0;
 627         default:
 628                 break;
 629         }
 630
 631         PMD_DRV_LOG(ERR, "Negotiate invalid api version %u from VF %d\n",
 632                 api_version, vf);
 633
 634         return -1;
 635 }
 636
 637 static int
 638 ixgbe_get_vf_queues(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 639 {
 640         struct ixgbe_vf_info *vfinfo =
 641                 *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
 642         uint32_t default_q = vf * RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
 643         struct rte_eth_conf *eth_conf;
 644         struct rte_eth_vmdq_dcb_tx_conf *vmdq_dcb_tx_conf;
 645         u8 num_tcs;
 646         struct ixgbe_hw *hw;
 647         u32 vmvir;
 648 #define IXGBE_VMVIR_VLANA_MASK          0xC0000000
 649 #define IXGBE_VMVIR_VLAN_VID_MASK       0x00000FFF
 650 #define IXGBE_VMVIR_VLAN_UP_MASK        0x0000E000
 651 #define VLAN_PRIO_SHIFT                 13
 652         u32 vlana;
 653         u32 vid;
 654         u32 user_priority;
 655
 656         /* Verify if the PF supports the mbox APIs version or not */
 657         switch (vfinfo[vf].api_version) {
 658         case ixgbe_mbox_api_20:
 659         case ixgbe_mbox_api_11:
 660         case ixgbe_mbox_api_12:
 661         case ixgbe_mbox_api_13:
 662                 break;
 663         default:
 664                 return -1;
 665         }
 666
 667         /* Notify VF of Rx and Tx queue number */
 668         msgbuf[IXGBE_VF_RX_QUEUES] = RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
 669         msgbuf[IXGBE_VF_TX_QUEUES] = RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
 670
 671         /* Notify VF of default queue */
 672         msgbuf[IXGBE_VF_DEF_QUEUE] = default_q;
 673
 674         /* Notify VF of number of DCB traffic classes */
 675         eth_conf = &dev->data->dev_conf;
 676         switch (eth_conf->txmode.mq_mode) {
 677         case ETH_MQ_TX_NONE:
 678         case ETH_MQ_TX_DCB:
 679                 PMD_DRV_LOG(ERR, "PF must work with virtualization for VF %u"
 680                         ", but its tx mode = %d\n", vf,
 681                         eth_conf->txmode.mq_mode);
 682                 return -1;
 683
 684         case ETH_MQ_TX_VMDQ_DCB:
 685                 vmdq_dcb_tx_conf = &eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
 686                 switch (vmdq_dcb_tx_conf->nb_queue_pools) {
 687                 case ETH_16_POOLS:
 688                         num_tcs = ETH_8_TCS;
 689                         break;
 690                 case ETH_32_POOLS:
 691                         num_tcs = ETH_4_TCS;
 692                         break;
 693                 default:
 694                         return -1;
 695                 }
 696                 break;
 697
 698         /* ETH_MQ_TX_VMDQ_ONLY,  DCB not enabled */
 699         case ETH_MQ_TX_VMDQ_ONLY:
 700                 hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 701                 vmvir = IXGBE_READ_REG(hw, IXGBE_VMVIR(vf));
 702                 vlana = vmvir & IXGBE_VMVIR_VLANA_MASK;
 703                 vid = vmvir & IXGBE_VMVIR_VLAN_VID_MASK;
 704                 user_priority =
 705                         (vmvir & IXGBE_VMVIR_VLAN_UP_MASK) >> VLAN_PRIO_SHIFT;
 706                 if ((vlana == IXGBE_VMVIR_VLANA_DEFAULT) &&
 707                         ((vid !=  0) || (user_priority != 0)))
 708                         num_tcs = 1;
 709                 else
 710                         num_tcs = 0;
 711                 break;
 712
 713         default:
 714                 PMD_DRV_LOG(ERR, "PF work with invalid mode = %d\n",
 715                         eth_conf->txmode.mq_mode);
 716                 return -1;
 717         }
 718         msgbuf[IXGBE_VF_TRANS_VLAN] = num_tcs;
 719
 720         return 0;
 721 }
 722
 723 static int
 724 ixgbe_set_vf_mc_promisc(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 725 {
 726         struct ixgbe_vf_info *vfinfo =
 727                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 728         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 729         int xcast_mode = msgbuf[1];     /* msgbuf contains the flag to enable */
 730         u32 vmolr, fctrl, disable, enable;
 731
 732         switch (vfinfo[vf].api_version) {
 733         case ixgbe_mbox_api_12:
 734                 /* promisc introduced in 1.3 version */
 735                 if (xcast_mode == IXGBEVF_XCAST_MODE_PROMISC)
 736                         return -EOPNOTSUPP;
 737                 break;
 738                 /* Fall threw */
 739         case ixgbe_mbox_api_13:
 740                 break;
 741         default:
 742                 return -1;
 743         }
 744
 745         if (vfinfo[vf].xcast_mode == xcast_mode)
 746                 goto out;
 747
 748         switch (xcast_mode) {
 749         case IXGBEVF_XCAST_MODE_NONE:
 750                 disable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE |
 751                           IXGBE_VMOLR_MPE | IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
 752                 enable = 0;
 753                 break;
 754         case IXGBEVF_XCAST_MODE_MULTI:
 755                 disable = IXGBE_VMOLR_MPE | IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
 756                 enable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE;
 757                 break;
 758         case IXGBEVF_XCAST_MODE_ALLMULTI:
 759                 disable = IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
 760                 enable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_MPE;
 761                 break;
 762         case IXGBEVF_XCAST_MODE_PROMISC:
 763                 if (hw->mac.type <= ixgbe_mac_82599EB)
 764                         return -1;
 765
 766                 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
 767                 if (!(fctrl & IXGBE_FCTRL_UPE)) {
 768                         /* VF promisc requires PF in promisc */
 769                         PMD_DRV_LOG(ERR,
 770                                "Enabling VF promisc requires PF in promisc\n");
 771                         return -1;
 772                 }
 773
 774                 disable = 0;
 775                 enable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE |
 776                          IXGBE_VMOLR_MPE | IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
 777                 break;
 778         default:
 779                 return -1;
 780         }
 781
 782         vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
 783         vmolr &= ~disable;
 784         vmolr |= enable;
 785         IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
 786         vfinfo[vf].xcast_mode = xcast_mode;
 787
 788 out:
 789         msgbuf[1] = xcast_mode;
 790
 791         return 0;
 792 }
 793
 794 static int
 795 ixgbe_set_vf_macvlan_msg(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 796 {
 797         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 798         struct ixgbe_vf_info *vf_info =
 799                 *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
 800         uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
 801         int index = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK) >>
 802                     IXGBE_VT_MSGINFO_SHIFT;
 803
 804         if (index) {
 805                 if (!rte_is_valid_assigned_ether_addr(
 806                         (struct rte_ether_addr *)new_mac)) {
 807                         PMD_DRV_LOG(ERR, "set invalid mac vf:%d\n", vf);
 808                         return -1;
 809                 }
 810
 811                 vf_info[vf].mac_count++;
 812
 813                 hw->mac.ops.set_rar(hw, vf_info[vf].mac_count,
 814                                 new_mac, vf, IXGBE_RAH_AV);
 815         } else {
 816                 if (vf_info[vf].mac_count) {
 817                         hw->mac.ops.clear_rar(hw, vf_info[vf].mac_count);
 818                         vf_info[vf].mac_count = 0;
 819                 }
 820         }
 821         return 0;
 822 }
 823
 824 static int
 825 ixgbe_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
 826 {
 827         uint16_t mbx_size = IXGBE_VFMAILBOX_SIZE;
 828         uint16_t msg_size = IXGBE_VF_MSG_SIZE_DEFAULT;
 829         uint32_t msgbuf[IXGBE_VFMAILBOX_SIZE];
 830         int32_t retval;
 831         struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 832         struct ixgbe_vf_info *vfinfo =
 833                 *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
 834         struct rte_pmd_ixgbe_mb_event_param ret_param;
 835
 836         retval = ixgbe_read_mbx(hw, msgbuf, mbx_size, vf);
 837         if (retval) {
 838                 PMD_DRV_LOG(ERR, "Error mbx recv msg from VF %d", vf);
 839                 return retval;
 840         }
 841
 842         /* do nothing with the message already been processed */
 843         if (msgbuf[0] & (IXGBE_VT_MSGTYPE_ACK | IXGBE_VT_MSGTYPE_NACK))
 844                 return retval;
 845
 846         /* flush the ack before we write any messages back */
 847         IXGBE_WRITE_FLUSH(hw);
 848
 849         /**
 850          * initialise structure to send to user application
 851          * will return response from user in retval field
 852          */
 853         ret_param.retval = RTE_PMD_IXGBE_MB_EVENT_PROCEED;
 854         ret_param.vfid = vf;
 855         ret_param.msg_type = msgbuf[0] & 0xFFFF;
 856         ret_param.msg = (void *)msgbuf;
 857
 858         /* perform VF reset */
 859         if (msgbuf[0] == IXGBE_VF_RESET) {
 860                 int ret = ixgbe_vf_reset(dev, vf, msgbuf);
 861
 862                 vfinfo[vf].clear_to_send = true;
 863
 864                 /* notify application about VF reset */
 865                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX,
 866                                               &ret_param);
 867                 return ret;
 868         }
 869
 870         /**
 871          * ask user application if we allowed to perform those functions
 872          * if we get ret_param.retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED
 873          * then business as usual,
 874          * if 0, do nothing and send ACK to VF
 875          * if ret_param.retval > 1, do nothing and send NAK to VF
 876          */
 877         rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX, &ret_param);
 878
 879         retval = ret_param.retval;
 880
 881         /* check & process VF to PF mailbox message */
 882         switch ((msgbuf[0] & 0xFFFF)) {
 883         case IXGBE_VF_SET_MAC_ADDR:
 884                 if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
 885                         retval = ixgbe_vf_set_mac_addr(dev, vf, msgbuf);
 886                 break;
 887         case IXGBE_VF_SET_MULTICAST:
 888                 if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
 889                         retval = ixgbe_vf_set_multicast(dev, vf, msgbuf);
 890                 break;
 891         case IXGBE_VF_SET_LPE:
 892                 if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
 893                         retval = ixgbe_set_vf_lpe(dev, vf, msgbuf);
 894                 break;
 895         case IXGBE_VF_SET_VLAN:
 896                 if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
 897                         retval = ixgbe_vf_set_vlan(dev, vf, msgbuf);
 898                 break;
 899         case IXGBE_VF_API_NEGOTIATE:
 900                 retval = ixgbe_negotiate_vf_api(dev, vf, msgbuf);
 901                 break;
 902         case IXGBE_VF_GET_QUEUES:
 903                 retval = ixgbe_get_vf_queues(dev, vf, msgbuf);
 904                 msg_size = IXGBE_VF_GET_QUEUE_MSG_SIZE;
 905                 break;
 906         case IXGBE_VF_UPDATE_XCAST_MODE:
 907                 if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
 908                         retval = ixgbe_set_vf_mc_promisc(dev, vf, msgbuf);
 909                 break;
 910         case IXGBE_VF_SET_MACVLAN:
 911                 if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
 912                         retval = ixgbe_set_vf_macvlan_msg(dev, vf, msgbuf);
 913                 break;
 914         default:
 915                 PMD_DRV_LOG(DEBUG, "Unhandled Msg %8.8x", (unsigned)msgbuf[0]);
 916                 retval = IXGBE_ERR_MBX;
 917                 break;
 918         }
 919
 920         /* response the VF according to the message process result */
 921         if (retval)
 922                 msgbuf[0] |= IXGBE_VT_MSGTYPE_NACK;
 923         else
 924                 msgbuf[0] |= IXGBE_VT_MSGTYPE_ACK;
 925
 926         msgbuf[0] |= IXGBE_VT_MSGTYPE_CTS;
 927
 928         ixgbe_write_mbx(hw, msgbuf, msg_size, vf);
 929
 930         return retval;
 931 }
 932
 933 static inline void
 934 ixgbe_rcv_ack_from_vf(struct rte_eth_dev *dev, uint16_t vf)
 935 {
 936         uint32_t msg = IXGBE_VT_MSGTYPE_NACK;
 937         struct ixgbe_hw *hw =
 938                 IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 939         struct ixgbe_vf_info *vfinfo =
 940                 *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
 941
 942         if (!vfinfo[vf].clear_to_send)
 943                 ixgbe_write_mbx(hw, &msg, 1, vf);
 944 }
 945
 946 void ixgbe_pf_mbx_process(struct rte_eth_dev *eth_dev)
 947 {
 948         uint16_t vf;
 949         struct ixgbe_hw *hw =
 950                 IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 951
 952         for (vf = 0; vf < dev_num_vf(eth_dev); vf++) {
 953                 /* check & process vf function level reset */
 954                 if (!ixgbe_check_for_rst(hw, vf))
 955                         ixgbe_vf_reset_event(eth_dev, vf);
 956
 957                 /* check & process vf mailbox messages */
 958                 if (!ixgbe_check_for_msg(hw, vf))
 959                         ixgbe_rcv_msg_from_vf(eth_dev, vf);
 960
 961                 /* check & process acks from vf */
 962                 if (!ixgbe_check_for_ack(hw, vf))
 963                         ixgbe_rcv_ack_from_vf(eth_dev, vf);
 964         }
 965 }