drivers/net/ethernet/intel/e1000e/netdev.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
   3
   4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   5
   6 #include <linux/module.h>
   7 #include <linux/types.h>
   8 #include <linux/init.h>
   9 #include <linux/pci.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/delay.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/interrupt.h>
  15 #include <linux/tcp.h>
  16 #include <linux/ipv6.h>
  17 #include <linux/slab.h>
  18 #include <net/checksum.h>
  19 #include <net/ip6_checksum.h>
  20 #include <linux/ethtool.h>
  21 #include <linux/if_vlan.h>
  22 #include <linux/cpu.h>
  23 #include <linux/smp.h>
  24 #include <linux/pm_qos.h>
  25 #include <linux/pm_runtime.h>
  26 #include <linux/aer.h>
  27 #include <linux/prefetch.h>
  28 #include <linux/suspend.h>
  29
  30 #include "e1000.h"
  31
  32 char e1000e_driver_name[] = "e1000e";
  33
  34 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  35 static int debug = -1;
  36 module_param(debug, int, 0);
  37 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  38
  39 static const struct e1000_info *e1000_info_tbl[] = {
  40         [board_82571]           = &e1000_82571_info,
  41         [board_82572]           = &e1000_82572_info,
  42         [board_82573]           = &e1000_82573_info,
  43         [board_82574]           = &e1000_82574_info,
  44         [board_82583]           = &e1000_82583_info,
  45         [board_80003es2lan]     = &e1000_es2_info,
  46         [board_ich8lan]         = &e1000_ich8_info,
  47         [board_ich9lan]         = &e1000_ich9_info,
  48         [board_ich10lan]        = &e1000_ich10_info,
  49         [board_pchlan]          = &e1000_pch_info,
  50         [board_pch2lan]         = &e1000_pch2_info,
  51         [board_pch_lpt]         = &e1000_pch_lpt_info,
  52         [board_pch_spt]         = &e1000_pch_spt_info,
  53         [board_pch_cnp]         = &e1000_pch_cnp_info,
  54 };
  55
  56 struct e1000_reg_info {
  57         u32 ofs;
  58         char *name;
  59 };
  60
  61 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  62         /* General Registers */
  63         {E1000_CTRL, "CTRL"},
  64         {E1000_STATUS, "STATUS"},
  65         {E1000_CTRL_EXT, "CTRL_EXT"},
  66
  67         /* Interrupt Registers */
  68         {E1000_ICR, "ICR"},
  69
  70         /* Rx Registers */
  71         {E1000_RCTL, "RCTL"},
  72         {E1000_RDLEN(0), "RDLEN"},
  73         {E1000_RDH(0), "RDH"},
  74         {E1000_RDT(0), "RDT"},
  75         {E1000_RDTR, "RDTR"},
  76         {E1000_RXDCTL(0), "RXDCTL"},
  77         {E1000_ERT, "ERT"},
  78         {E1000_RDBAL(0), "RDBAL"},
  79         {E1000_RDBAH(0), "RDBAH"},
  80         {E1000_RDFH, "RDFH"},
  81         {E1000_RDFT, "RDFT"},
  82         {E1000_RDFHS, "RDFHS"},
  83         {E1000_RDFTS, "RDFTS"},
  84         {E1000_RDFPC, "RDFPC"},
  85
  86         /* Tx Registers */
  87         {E1000_TCTL, "TCTL"},
  88         {E1000_TDBAL(0), "TDBAL"},
  89         {E1000_TDBAH(0), "TDBAH"},
  90         {E1000_TDLEN(0), "TDLEN"},
  91         {E1000_TDH(0), "TDH"},
  92         {E1000_TDT(0), "TDT"},
  93         {E1000_TIDV, "TIDV"},
  94         {E1000_TXDCTL(0), "TXDCTL"},
  95         {E1000_TADV, "TADV"},
  96         {E1000_TARC(0), "TARC"},
  97         {E1000_TDFH, "TDFH"},
  98         {E1000_TDFT, "TDFT"},
  99         {E1000_TDFHS, "TDFHS"},
 100         {E1000_TDFTS, "TDFTS"},
 101         {E1000_TDFPC, "TDFPC"},
 102
 103         /* List Terminator */
 104         {0, NULL}
 105 };
 106
 107 /**
 108  * __ew32_prepare - prepare to write to MAC CSR register on certain parts
 109  * @hw: pointer to the HW structure
 110  *
 111  * When updating the MAC CSR registers, the Manageability Engine (ME) could
 112  * be accessing the registers at the same time.  Normally, this is handled in
 113  * h/w by an arbiter but on some parts there is a bug that acknowledges Host
 114  * accesses later than it should which could result in the register to have
 115  * an incorrect value.  Workaround this by checking the FWSM register which
 116  * has bit 24 set while ME is accessing MAC CSR registers, wait if it is set
 117  * and try again a number of times.
 118  **/
 119 static void __ew32_prepare(struct e1000_hw *hw)
 120 {
 121         s32 i = E1000_ICH_FWSM_PCIM2PCI_COUNT;
 122
 123         while ((er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI) && --i)
 124                 udelay(50);
 125 }
 126
 127 void __ew32(struct e1000_hw *hw, unsigned long reg, u32 val)
 128 {
 129         if (hw->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 130                 __ew32_prepare(hw);
 131
 132         writel(val, hw->hw_addr + reg);
 133 }
 134
 135 /**
 136  * e1000_regdump - register printout routine
 137  * @hw: pointer to the HW structure
 138  * @reginfo: pointer to the register info table
 139  **/
 140 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 141 {
 142         int n = 0;
 143         char rname[16];
 144         u32 regs[8];
 145
 146         switch (reginfo->ofs) {
 147         case E1000_RXDCTL(0):
 148                 for (n = 0; n < 2; n++)
 149                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 150                 break;
 151         case E1000_TXDCTL(0):
 152                 for (n = 0; n < 2; n++)
 153                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 154                 break;
 155         case E1000_TARC(0):
 156                 for (n = 0; n < 2; n++)
 157                         regs[n] = __er32(hw, E1000_TARC(n));
 158                 break;
 159         default:
 160                 pr_info("%-15s %08x\n",
 161                         reginfo->name, __er32(hw, reginfo->ofs));
 162                 return;
 163         }
 164
 165         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 166         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 167 }
 168
 169 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
 170                                  struct e1000_buffer *bi)
 171 {
 172         int i;
 173         struct e1000_ps_page *ps_page;
 174
 175         for (i = 0; i < adapter->rx_ps_pages; i++) {
 176                 ps_page = &bi->ps_pages[i];
 177
 178                 if (ps_page->page) {
 179                         pr_info("packet dump for ps_page %d:\n", i);
 180                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 181                                        16, 1, page_address(ps_page->page),
 182                                        PAGE_SIZE, true);
 183                 }
 184         }
 185 }
 186
 187 /**
 188  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 189  * @adapter: board private structure
 190  **/
 191 static void e1000e_dump(struct e1000_adapter *adapter)
 192 {
 193         struct net_device *netdev = adapter->netdev;
 194         struct e1000_hw *hw = &adapter->hw;
 195         struct e1000_reg_info *reginfo;
 196         struct e1000_ring *tx_ring = adapter->tx_ring;
 197         struct e1000_tx_desc *tx_desc;
 198         struct my_u0 {
 199                 __le64 a;
 200                 __le64 b;
 201         } *u0;
 202         struct e1000_buffer *buffer_info;
 203         struct e1000_ring *rx_ring = adapter->rx_ring;
 204         union e1000_rx_desc_packet_split *rx_desc_ps;
 205         union e1000_rx_desc_extended *rx_desc;
 206         struct my_u1 {
 207                 __le64 a;
 208                 __le64 b;
 209                 __le64 c;
 210                 __le64 d;
 211         } *u1;
 212         u32 staterr;
 213         int i = 0;
 214
 215         if (!netif_msg_hw(adapter))
 216                 return;
 217
 218         /* Print netdevice Info */
 219         if (netdev) {
 220                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 221                 pr_info("Device Name     state            trans_start\n");
 222                 pr_info("%-15s %016lX %016lX\n", netdev->name,
 223                         netdev->state, dev_trans_start(netdev));
 224         }
 225
 226         /* Print Registers */
 227         dev_info(&adapter->pdev->dev, "Register Dump\n");
 228         pr_info(" Register Name   Value\n");
 229         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 230              reginfo->name; reginfo++) {
 231                 e1000_regdump(hw, reginfo);
 232         }
 233
 234         /* Print Tx Ring Summary */
 235         if (!netdev || !netif_running(netdev))
 236                 return;
 237
 238         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 239         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 240         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 241         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 242                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 243                 (unsigned long long)buffer_info->dma,
 244                 buffer_info->length,
 245                 buffer_info->next_to_watch,
 246                 (unsigned long long)buffer_info->time_stamp);
 247
 248         /* Print Tx Ring */
 249         if (!netif_msg_tx_done(adapter))
 250                 goto rx_ring_summary;
 251
 252         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 253
 254         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 255          *
 256          * Legacy Transmit Descriptor
 257          *   +--------------------------------------------------------------+
 258          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 259          *   +--------------------------------------------------------------+
 260          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 261          *   +--------------------------------------------------------------+
 262          *   63       48 47        36 35    32 31     24 23    16 15        0
 263          *
 264          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 265          *   63      48 47    40 39       32 31             16 15    8 7      0
 266          *   +----------------------------------------------------------------+
 267          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 268          *   +----------------------------------------------------------------+
 269          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 270          *   +----------------------------------------------------------------+
 271          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 272          *
 273          * Extended Data Descriptor (DTYP=0x1)
 274          *   +----------------------------------------------------------------+
 275          * 0 |                     Buffer Address [63:0]                      |
 276          *   +----------------------------------------------------------------+
 277          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 278          *   +----------------------------------------------------------------+
 279          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 280          */
 281         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 282         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 283         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 284         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 285                 const char *next_desc;
 286                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 287                 buffer_info = &tx_ring->buffer_info[i];
 288                 u0 = (struct my_u0 *)tx_desc;
 289                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 290                         next_desc = " NTC/U";
 291                 else if (i == tx_ring->next_to_use)
 292                         next_desc = " NTU";
 293                 else if (i == tx_ring->next_to_clean)
 294                         next_desc = " NTC";
 295                 else
 296                         next_desc = "";
 297                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 298                         (!(le64_to_cpu(u0->b) & BIT(29)) ? 'l' :
 299                          ((le64_to_cpu(u0->b) & BIT(20)) ? 'd' : 'c')),
 300                         i,
 301                         (unsigned long long)le64_to_cpu(u0->a),
 302                         (unsigned long long)le64_to_cpu(u0->b),
 303                         (unsigned long long)buffer_info->dma,
 304                         buffer_info->length, buffer_info->next_to_watch,
 305                         (unsigned long long)buffer_info->time_stamp,
 306                         buffer_info->skb, next_desc);
 307
 308                 if (netif_msg_pktdata(adapter) && buffer_info->skb)
 309                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 310                                        16, 1, buffer_info->skb->data,
 311                                        buffer_info->skb->len, true);
 312         }
 313
 314         /* Print Rx Ring Summary */
 315 rx_ring_summary:
 316         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 317         pr_info("Queue [NTU] [NTC]\n");
 318         pr_info(" %5d %5X %5X\n",
 319                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 320
 321         /* Print Rx Ring */
 322         if (!netif_msg_rx_status(adapter))
 323                 return;
 324
 325         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 326         switch (adapter->rx_ps_pages) {
 327         case 1:
 328         case 2:
 329         case 3:
 330                 /* [Extended] Packet Split Receive Descriptor Format
 331                  *
 332                  *    +-----------------------------------------------------+
 333                  *  0 |                Buffer Address 0 [63:0]              |
 334                  *    +-----------------------------------------------------+
 335                  *  8 |                Buffer Address 1 [63:0]              |
 336                  *    +-----------------------------------------------------+
 337                  * 16 |                Buffer Address 2 [63:0]              |
 338                  *    +-----------------------------------------------------+
 339                  * 24 |                Buffer Address 3 [63:0]              |
 340                  *    +-----------------------------------------------------+
 341                  */
 342                 pr_info("R  [desc]      [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] [bi->skb] <-- Ext Pkt Split format\n");
 343                 /* [Extended] Receive Descriptor (Write-Back) Format
 344                  *
 345                  *   63       48 47    32 31     13 12    8 7    4 3        0
 346                  *   +------------------------------------------------------+
 347                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 348                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 349                  *   +------------------------------------------------------+
 350                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 351                  *   +------------------------------------------------------+
 352                  *   63       48 47    32 31            20 19               0
 353                  */
 354                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 355                 for (i = 0; i < rx_ring->count; i++) {
 356                         const char *next_desc;
 357                         buffer_info = &rx_ring->buffer_info[i];
 358                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 359                         u1 = (struct my_u1 *)rx_desc_ps;
 360                         staterr =
 361                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 362
 363                         if (i == rx_ring->next_to_use)
 364                                 next_desc = " NTU";
 365                         else if (i == rx_ring->next_to_clean)
 366                                 next_desc = " NTC";
 367                         else
 368                                 next_desc = "";
 369
 370                         if (staterr & E1000_RXD_STAT_DD) {
 371                                 /* Descriptor Done */
 372                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 373                                         "RWB", i,
 374                                         (unsigned long long)le64_to_cpu(u1->a),
 375                                         (unsigned long long)le64_to_cpu(u1->b),
 376                                         (unsigned long long)le64_to_cpu(u1->c),
 377                                         (unsigned long long)le64_to_cpu(u1->d),
 378                                         buffer_info->skb, next_desc);
 379                         } else {
 380                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 381                                         "R  ", i,
 382                                         (unsigned long long)le64_to_cpu(u1->a),
 383                                         (unsigned long long)le64_to_cpu(u1->b),
 384                                         (unsigned long long)le64_to_cpu(u1->c),
 385                                         (unsigned long long)le64_to_cpu(u1->d),
 386                                         (unsigned long long)buffer_info->dma,
 387                                         buffer_info->skb, next_desc);
 388
 389                                 if (netif_msg_pktdata(adapter))
 390                                         e1000e_dump_ps_pages(adapter,
 391                                                              buffer_info);
 392                         }
 393                 }
 394                 break;
 395         default:
 396         case 0:
 397                 /* Extended Receive Descriptor (Read) Format
 398                  *
 399                  *   +-----------------------------------------------------+
 400                  * 0 |                Buffer Address [63:0]                |
 401                  *   +-----------------------------------------------------+
 402                  * 8 |                      Reserved                       |
 403                  *   +-----------------------------------------------------+
 404                  */
 405                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 406                 /* Extended Receive Descriptor (Write-Back) Format
 407                  *
 408                  *   63       48 47    32 31    24 23            4 3        0
 409                  *   +------------------------------------------------------+
 410                  *   |     RSS Hash      |        |               |         |
 411                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 412                  *   | Packet   | IP     |        |               |  Type   |
 413                  *   | Checksum | Ident  |        |               |         |
 414                  *   +------------------------------------------------------+
 415                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 416                  *   +------------------------------------------------------+
 417                  *   63       48 47    32 31            20 19               0
 418                  */
 419                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 420
 421                 for (i = 0; i < rx_ring->count; i++) {
 422                         const char *next_desc;
 423
 424                         buffer_info = &rx_ring->buffer_info[i];
 425                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 426                         u1 = (struct my_u1 *)rx_desc;
 427                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 428
 429                         if (i == rx_ring->next_to_use)
 430                                 next_desc = " NTU";
 431                         else if (i == rx_ring->next_to_clean)
 432                                 next_desc = " NTC";
 433                         else
 434                                 next_desc = "";
 435
 436                         if (staterr & E1000_RXD_STAT_DD) {
 437                                 /* Descriptor Done */
 438                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 439                                         "RWB", i,
 440                                         (unsigned long long)le64_to_cpu(u1->a),
 441                                         (unsigned long long)le64_to_cpu(u1->b),
 442                                         buffer_info->skb, next_desc);
 443                         } else {
 444                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 445                                         "R  ", i,
 446                                         (unsigned long long)le64_to_cpu(u1->a),
 447                                         (unsigned long long)le64_to_cpu(u1->b),
 448                                         (unsigned long long)buffer_info->dma,
 449                                         buffer_info->skb, next_desc);
 450
 451                                 if (netif_msg_pktdata(adapter) &&
 452                                     buffer_info->skb)
 453                                         print_hex_dump(KERN_INFO, "",
 454                                                        DUMP_PREFIX_ADDRESS, 16,
 455                                                        1,
 456                                                        buffer_info->skb->data,
 457                                                        adapter->rx_buffer_len,
 458                                                        true);
 459                         }
 460                 }
 461         }
 462 }
 463
 464 /**
 465  * e1000_desc_unused - calculate if we have unused descriptors
 466  * @ring: pointer to ring struct to perform calculation on
 467  **/
 468 static int e1000_desc_unused(struct e1000_ring *ring)
 469 {
 470         if (ring->next_to_clean > ring->next_to_use)
 471                 return ring->next_to_clean - ring->next_to_use - 1;
 472
 473         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 474 }
 475
 476 /**
 477  * e1000e_systim_to_hwtstamp - convert system time value to hw time stamp
 478  * @adapter: board private structure
 479  * @hwtstamps: time stamp structure to update
 480  * @systim: unsigned 64bit system time value.
 481  *
 482  * Convert the system time value stored in the RX/TXSTMP registers into a
 483  * hwtstamp which can be used by the upper level time stamping functions.
 484  *
 485  * The 'systim_lock' spinlock is used to protect the consistency of the
 486  * system time value. This is needed because reading the 64 bit time
 487  * value involves reading two 32 bit registers. The first read latches the
 488  * value.
 489  **/
 490 static void e1000e_systim_to_hwtstamp(struct e1000_adapter *adapter,
 491                                       struct skb_shared_hwtstamps *hwtstamps,
 492                                       u64 systim)
 493 {
 494         u64 ns;
 495         unsigned long flags;
 496
 497         spin_lock_irqsave(&adapter->systim_lock, flags);
 498         ns = timecounter_cyc2time(&adapter->tc, systim);
 499         spin_unlock_irqrestore(&adapter->systim_lock, flags);
 500
 501         memset(hwtstamps, 0, sizeof(*hwtstamps));
 502         hwtstamps->hwtstamp = ns_to_ktime(ns);
 503 }
 504
 505 /**
 506  * e1000e_rx_hwtstamp - utility function which checks for Rx time stamp
 507  * @adapter: board private structure
 508  * @status: descriptor extended error and status field
 509  * @skb: particular skb to include time stamp
 510  *
 511  * If the time stamp is valid, convert it into the timecounter ns value
 512  * and store that result into the shhwtstamps structure which is passed
 513  * up the network stack.
 514  **/
 515 static void e1000e_rx_hwtstamp(struct e1000_adapter *adapter, u32 status,
 516                                struct sk_buff *skb)
 517 {
 518         struct e1000_hw *hw = &adapter->hw;
 519         u64 rxstmp;
 520
 521         if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP) ||
 522             !(status & E1000_RXDEXT_STATERR_TST) ||
 523             !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
 524                 return;
 525
 526         /* The Rx time stamp registers contain the time stamp.  No other
 527          * received packet will be time stamped until the Rx time stamp
 528          * registers are read.  Because only one packet can be time stamped
 529          * at a time, the register values must belong to this packet and
 530          * therefore none of the other additional attributes need to be
 531          * compared.
 532          */
 533         rxstmp = (u64)er32(RXSTMPL);
 534         rxstmp |= (u64)er32(RXSTMPH) << 32;
 535         e1000e_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), rxstmp);
 536
 537         adapter->flags2 &= ~FLAG2_CHECK_RX_HWTSTAMP;
 538 }
 539
 540 /**
 541  * e1000_receive_skb - helper function to handle Rx indications
 542  * @adapter: board private structure
 543  * @netdev: pointer to netdev struct
 544  * @staterr: descriptor extended error and status field as written by hardware
 545  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 546  * @skb: pointer to sk_buff to be indicated to stack
 547  **/
 548 static void e1000_receive_skb(struct e1000_adapter *adapter,
 549                               struct net_device *netdev, struct sk_buff *skb,
 550                               u32 staterr, __le16 vlan)
 551 {
 552         u16 tag = le16_to_cpu(vlan);
 553
 554         e1000e_rx_hwtstamp(adapter, staterr, skb);
 555
 556         skb->protocol = eth_type_trans(skb, netdev);
 557
 558         if (staterr & E1000_RXD_STAT_VP)
 559                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
 560
 561         napi_gro_receive(&adapter->napi, skb);
 562 }
 563
 564 /**
 565  * e1000_rx_checksum - Receive Checksum Offload
 566  * @adapter: board private structure
 567  * @status_err: receive descriptor status and error fields
 568  * @skb: socket buffer with received data
 569  **/
 570 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 571                               struct sk_buff *skb)
 572 {
 573         u16 status = (u16)status_err;
 574         u8 errors = (u8)(status_err >> 24);
 575
 576         skb_checksum_none_assert(skb);
 577
 578         /* Rx checksum disabled */
 579         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 580                 return;
 581
 582         /* Ignore Checksum bit is set */
 583         if (status & E1000_RXD_STAT_IXSM)
 584                 return;
 585
 586         /* TCP/UDP checksum error bit or IP checksum error bit is set */
 587         if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
 588                 /* let the stack verify checksum errors */
 589                 adapter->hw_csum_err++;
 590                 return;
 591         }
 592
 593         /* TCP/UDP Checksum has not been calculated */
 594         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 595                 return;
 596
 597         /* It must be a TCP or UDP packet with a valid checksum */
 598         skb->ip_summed = CHECKSUM_UNNECESSARY;
 599         adapter->hw_csum_good++;
 600 }
 601
 602 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
 603 {
 604         struct e1000_adapter *adapter = rx_ring->adapter;
 605         struct e1000_hw *hw = &adapter->hw;
 606
 607         __ew32_prepare(hw);
 608         writel(i, rx_ring->tail);
 609
 610         if (unlikely(i != readl(rx_ring->tail))) {
 611                 u32 rctl = er32(RCTL);
 612
 613                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 614                 e_err("ME firmware caused invalid RDT - resetting\n");
 615                 schedule_work(&adapter->reset_task);
 616         }
 617 }
 618
 619 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
 620 {
 621         struct e1000_adapter *adapter = tx_ring->adapter;
 622         struct e1000_hw *hw = &adapter->hw;
 623
 624         __ew32_prepare(hw);
 625         writel(i, tx_ring->tail);
 626
 627         if (unlikely(i != readl(tx_ring->tail))) {
 628                 u32 tctl = er32(TCTL);
 629
 630                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 631                 e_err("ME firmware caused invalid TDT - resetting\n");
 632                 schedule_work(&adapter->reset_task);
 633         }
 634 }
 635
 636 /**
 637  * e1000_alloc_rx_buffers - Replace used receive buffers
 638  * @rx_ring: Rx descriptor ring
 639  * @cleaned_count: number to reallocate
 640  * @gfp: flags for allocation
 641  **/
 642 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
 643                                    int cleaned_count, gfp_t gfp)
 644 {
 645         struct e1000_adapter *adapter = rx_ring->adapter;
 646         struct net_device *netdev = adapter->netdev;
 647         struct pci_dev *pdev = adapter->pdev;
 648         union e1000_rx_desc_extended *rx_desc;
 649         struct e1000_buffer *buffer_info;
 650         struct sk_buff *skb;
 651         unsigned int i;
 652         unsigned int bufsz = adapter->rx_buffer_len;
 653
 654         i = rx_ring->next_to_use;
 655         buffer_info = &rx_ring->buffer_info[i];
 656
 657         while (cleaned_count--) {
 658                 skb = buffer_info->skb;
 659                 if (skb) {
 660                         skb_trim(skb, 0);
 661                         goto map_skb;
 662                 }
 663
 664                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 665                 if (!skb) {
 666                         /* Better luck next round */
 667                         adapter->alloc_rx_buff_failed++;
 668                         break;
 669                 }
 670
 671                 buffer_info->skb = skb;
 672 map_skb:
 673                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 674                                                   adapter->rx_buffer_len,
 675                                                   DMA_FROM_DEVICE);
 676                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 677                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 678                         adapter->rx_dma_failed++;
 679                         break;
 680                 }
 681
 682                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 683                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 684
 685                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 686                         /* Force memory writes to complete before letting h/w
 687                          * know there are new descriptors to fetch.  (Only
 688                          * applicable for weak-ordered memory model archs,
 689                          * such as IA-64).
 690                          */
 691                         wmb();
 692                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 693                                 e1000e_update_rdt_wa(rx_ring, i);
 694                         else
 695                                 writel(i, rx_ring->tail);
 696                 }
 697                 i++;
 698                 if (i == rx_ring->count)
 699                         i = 0;
 700                 buffer_info = &rx_ring->buffer_info[i];
 701         }
 702
 703         rx_ring->next_to_use = i;
 704 }
 705
 706 /**
 707  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 708  * @rx_ring: Rx descriptor ring
 709  * @cleaned_count: number to reallocate
 710  * @gfp: flags for allocation
 711  **/
 712 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
 713                                       int cleaned_count, gfp_t gfp)
 714 {
 715         struct e1000_adapter *adapter = rx_ring->adapter;
 716         struct net_device *netdev = adapter->netdev;
 717         struct pci_dev *pdev = adapter->pdev;
 718         union e1000_rx_desc_packet_split *rx_desc;
 719         struct e1000_buffer *buffer_info;
 720         struct e1000_ps_page *ps_page;
 721         struct sk_buff *skb;
 722         unsigned int i, j;
 723
 724         i = rx_ring->next_to_use;
 725         buffer_info = &rx_ring->buffer_info[i];
 726
 727         while (cleaned_count--) {
 728                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 729
 730                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 731                         ps_page = &buffer_info->ps_pages[j];
 732                         if (j >= adapter->rx_ps_pages) {
 733                                 /* all unused desc entries get hw null ptr */
 734                                 rx_desc->read.buffer_addr[j + 1] =
 735                                     ~cpu_to_le64(0);
 736                                 continue;
 737                         }
 738                         if (!ps_page->page) {
 739                                 ps_page->page = alloc_page(gfp);
 740                                 if (!ps_page->page) {
 741                                         adapter->alloc_rx_buff_failed++;
 742                                         goto no_buffers;
 743                                 }
 744                                 ps_page->dma = dma_map_page(&pdev->dev,
 745                                                             ps_page->page,
 746                                                             0, PAGE_SIZE,
 747                                                             DMA_FROM_DEVICE);
 748                                 if (dma_mapping_error(&pdev->dev,
 749                                                       ps_page->dma)) {
 750                                         dev_err(&adapter->pdev->dev,
 751                                                 "Rx DMA page map failed\n");
 752                                         adapter->rx_dma_failed++;
 753                                         goto no_buffers;
 754                                 }
 755                         }
 756                         /* Refresh the desc even if buffer_addrs
 757                          * didn't change because each write-back
 758                          * erases this info.
 759                          */
 760                         rx_desc->read.buffer_addr[j + 1] =
 761                             cpu_to_le64(ps_page->dma);
 762                 }
 763
 764                 skb = __netdev_alloc_skb_ip_align(netdev, adapter->rx_ps_bsize0,
 765                                                   gfp);
 766
 767                 if (!skb) {
 768                         adapter->alloc_rx_buff_failed++;
 769                         break;
 770                 }
 771
 772                 buffer_info->skb = skb;
 773                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 774                                                   adapter->rx_ps_bsize0,
 775                                                   DMA_FROM_DEVICE);
 776                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 777                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 778                         adapter->rx_dma_failed++;
 779                         /* cleanup skb */
 780                         dev_kfree_skb_any(skb);
 781                         buffer_info->skb = NULL;
 782                         break;
 783                 }
 784
 785                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 786
 787                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 788                         /* Force memory writes to complete before letting h/w
 789                          * know there are new descriptors to fetch.  (Only
 790                          * applicable for weak-ordered memory model archs,
 791                          * such as IA-64).
 792                          */
 793                         wmb();
 794                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 795                                 e1000e_update_rdt_wa(rx_ring, i << 1);
 796                         else
 797                                 writel(i << 1, rx_ring->tail);
 798                 }
 799
 800                 i++;
 801                 if (i == rx_ring->count)
 802                         i = 0;
 803                 buffer_info = &rx_ring->buffer_info[i];
 804         }
 805
 806 no_buffers:
 807         rx_ring->next_to_use = i;
 808 }
 809
 810 /**
 811  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 812  * @rx_ring: Rx descriptor ring
 813  * @cleaned_count: number of buffers to allocate this pass
 814  * @gfp: flags for allocation
 815  **/
 816
 817 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
 818                                          int cleaned_count, gfp_t gfp)
 819 {
 820         struct e1000_adapter *adapter = rx_ring->adapter;
 821         struct net_device *netdev = adapter->netdev;
 822         struct pci_dev *pdev = adapter->pdev;
 823         union e1000_rx_desc_extended *rx_desc;
 824         struct e1000_buffer *buffer_info;
 825         struct sk_buff *skb;
 826         unsigned int i;
 827         unsigned int bufsz = 256 - 16;  /* for skb_reserve */
 828
 829         i = rx_ring->next_to_use;
 830         buffer_info = &rx_ring->buffer_info[i];
 831
 832         while (cleaned_count--) {
 833                 skb = buffer_info->skb;
 834                 if (skb) {
 835                         skb_trim(skb, 0);
 836                         goto check_page;
 837                 }
 838
 839                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 840                 if (unlikely(!skb)) {
 841                         /* Better luck next round */
 842                         adapter->alloc_rx_buff_failed++;
 843                         break;
 844                 }
 845
 846                 buffer_info->skb = skb;
 847 check_page:
 848                 /* allocate a new page if necessary */
 849                 if (!buffer_info->page) {
 850                         buffer_info->page = alloc_page(gfp);
 851                         if (unlikely(!buffer_info->page)) {
 852                                 adapter->alloc_rx_buff_failed++;
 853                                 break;
 854                         }
 855                 }
 856
 857                 if (!buffer_info->dma) {
 858                         buffer_info->dma = dma_map_page(&pdev->dev,
 859                                                         buffer_info->page, 0,
 860                                                         PAGE_SIZE,
 861                                                         DMA_FROM_DEVICE);
 862                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 863                                 adapter->alloc_rx_buff_failed++;
 864                                 break;
 865                         }
 866                 }
 867
 868                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 869                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 870
 871                 if (unlikely(++i == rx_ring->count))
 872                         i = 0;
 873                 buffer_info = &rx_ring->buffer_info[i];
 874         }
 875
 876         if (likely(rx_ring->next_to_use != i)) {
 877                 rx_ring->next_to_use = i;
 878                 if (unlikely(i-- == 0))
 879                         i = (rx_ring->count - 1);
 880
 881                 /* Force memory writes to complete before letting h/w
 882                  * know there are new descriptors to fetch.  (Only
 883                  * applicable for weak-ordered memory model archs,
 884                  * such as IA-64).
 885                  */
 886                 wmb();
 887                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 888                         e1000e_update_rdt_wa(rx_ring, i);
 889                 else
 890                         writel(i, rx_ring->tail);
 891         }
 892 }
 893
 894 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
 895                                  struct sk_buff *skb)
 896 {
 897         if (netdev->features & NETIF_F_RXHASH)
 898                 skb_set_hash(skb, le32_to_cpu(rss), PKT_HASH_TYPE_L3);
 899 }
 900
 901 /**
 902  * e1000_clean_rx_irq - Send received data up the network stack
 903  * @rx_ring: Rx descriptor ring
 904  * @work_done: output parameter for indicating completed work
 905  * @work_to_do: how many packets we can clean
 906  *
 907  * the return value indicates whether actual cleaning was done, there
 908  * is no guarantee that everything was cleaned
 909  **/
 910 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 911                                int work_to_do)
 912 {
 913         struct e1000_adapter *adapter = rx_ring->adapter;
 914         struct net_device *netdev = adapter->netdev;
 915         struct pci_dev *pdev = adapter->pdev;
 916         struct e1000_hw *hw = &adapter->hw;
 917         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 918         struct e1000_buffer *buffer_info, *next_buffer;
 919         u32 length, staterr;
 920         unsigned int i;
 921         int cleaned_count = 0;
 922         bool cleaned = false;
 923         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 924
 925         i = rx_ring->next_to_clean;
 926         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 927         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 928         buffer_info = &rx_ring->buffer_info[i];
 929
 930         while (staterr & E1000_RXD_STAT_DD) {
 931                 struct sk_buff *skb;
 932
 933                 if (*work_done >= work_to_do)
 934                         break;
 935                 (*work_done)++;
 936                 dma_rmb();      /* read descriptor and rx_buffer_info after status DD */
 937
 938                 skb = buffer_info->skb;
 939                 buffer_info->skb = NULL;
 940
 941                 prefetch(skb->data - NET_IP_ALIGN);
 942
 943                 i++;
 944                 if (i == rx_ring->count)
 945                         i = 0;
 946                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 947                 prefetch(next_rxd);
 948
 949                 next_buffer = &rx_ring->buffer_info[i];
 950
 951                 cleaned = true;
 952                 cleaned_count++;
 953                 dma_unmap_single(&pdev->dev, buffer_info->dma,
 954                                  adapter->rx_buffer_len, DMA_FROM_DEVICE);
 955                 buffer_info->dma = 0;
 956
 957                 length = le16_to_cpu(rx_desc->wb.upper.length);
 958
 959                 /* !EOP means multiple descriptors were used to store a single
 960                  * packet, if that's the case we need to toss it.  In fact, we
 961                  * need to toss every packet with the EOP bit clear and the
 962                  * next frame that _does_ have the EOP bit set, as it is by
 963                  * definition only a frame fragment
 964                  */
 965                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 966                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 967
 968                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 969                         /* All receives must fit into a single buffer */
 970                         e_dbg("Receive packet consumed multiple buffers\n");
 971                         /* recycle */
 972                         buffer_info->skb = skb;
 973                         if (staterr & E1000_RXD_STAT_EOP)
 974                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 975                         goto next_desc;
 976                 }
 977
 978                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
 979                              !(netdev->features & NETIF_F_RXALL))) {
 980                         /* recycle */
 981                         buffer_info->skb = skb;
 982                         goto next_desc;
 983                 }
 984
 985                 /* adjust length to remove Ethernet CRC */
 986                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
 987                         /* If configured to store CRC, don't subtract FCS,
 988                          * but keep the FCS bytes out of the total_rx_bytes
 989                          * counter
 990                          */
 991                         if (netdev->features & NETIF_F_RXFCS)
 992                                 total_rx_bytes -= 4;
 993                         else
 994                                 length -= 4;
 995                 }
 996
 997                 total_rx_bytes += length;
 998                 total_rx_packets++;
 999
1000                 /* code added for copybreak, this should improve
1001                  * performance for small packets with large amounts
1002                  * of reassembly being done in the stack
1003                  */
1004                 if (length < copybreak) {
1005                         struct sk_buff *new_skb =
1006                                 napi_alloc_skb(&adapter->napi, length);
1007                         if (new_skb) {
1008                                 skb_copy_to_linear_data_offset(new_skb,
1009                                                                -NET_IP_ALIGN,
1010                                                                (skb->data -
1011                                                                 NET_IP_ALIGN),
1012                                                                (length +
1013                                                                 NET_IP_ALIGN));
1014                                 /* save the skb in buffer_info as good */
1015                                 buffer_info->skb = skb;
1016                                 skb = new_skb;
1017                         }
1018                         /* else just continue with the old one */
1019                 }
1020                 /* end copybreak code */
1021                 skb_put(skb, length);
1022
1023                 /* Receive Checksum Offload */
1024                 e1000_rx_checksum(adapter, staterr, skb);
1025
1026                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1027
1028                 e1000_receive_skb(adapter, netdev, skb, staterr,
1029                                   rx_desc->wb.upper.vlan);
1030
1031 next_desc:
1032                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1033
1034                 /* return some buffers to hardware, one at a time is too slow */
1035                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1036                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1037                                               GFP_ATOMIC);
1038                         cleaned_count = 0;
1039                 }
1040
1041                 /* use prefetched values */
1042                 rx_desc = next_rxd;
1043                 buffer_info = next_buffer;
1044
1045                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1046         }
1047         rx_ring->next_to_clean = i;
1048
1049         cleaned_count = e1000_desc_unused(rx_ring);
1050         if (cleaned_count)
1051                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1052
1053         adapter->total_rx_bytes += total_rx_bytes;
1054         adapter->total_rx_packets += total_rx_packets;
1055         return cleaned;
1056 }
1057
1058 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
1059                             struct e1000_buffer *buffer_info,
1060                             bool drop)
1061 {
1062         struct e1000_adapter *adapter = tx_ring->adapter;
1063
1064         if (buffer_info->dma) {
1065                 if (buffer_info->mapped_as_page)
1066                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1067                                        buffer_info->length, DMA_TO_DEVICE);
1068                 else
1069                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1070                                          buffer_info->length, DMA_TO_DEVICE);
1071                 buffer_info->dma = 0;
1072         }
1073         if (buffer_info->skb) {
1074                 if (drop)
1075                         dev_kfree_skb_any(buffer_info->skb);
1076                 else
1077                         dev_consume_skb_any(buffer_info->skb);
1078                 buffer_info->skb = NULL;
1079         }
1080         buffer_info->time_stamp = 0;
1081 }
1082
1083 static void e1000_print_hw_hang(struct work_struct *work)
1084 {
1085         struct e1000_adapter *adapter = container_of(work,
1086                                                      struct e1000_adapter,
1087                                                      print_hang_task);
1088         struct net_device *netdev = adapter->netdev;
1089         struct e1000_ring *tx_ring = adapter->tx_ring;
1090         unsigned int i = tx_ring->next_to_clean;
1091         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1092         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1093         struct e1000_hw *hw = &adapter->hw;
1094         u16 phy_status, phy_1000t_status, phy_ext_status;
1095         u16 pci_status;
1096
1097         if (test_bit(__E1000_DOWN, &adapter->state))
1098                 return;
1099
1100         if (!adapter->tx_hang_recheck && (adapter->flags2 & FLAG2_DMA_BURST)) {
1101                 /* May be block on write-back, flush and detect again
1102                  * flush pending descriptor writebacks to memory
1103                  */
1104                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1105                 /* execute the writes immediately */
1106                 e1e_flush();
1107                 /* Due to rare timing issues, write to TIDV again to ensure
1108                  * the write is successful
1109                  */
1110                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1111                 /* execute the writes immediately */
1112                 e1e_flush();
1113                 adapter->tx_hang_recheck = true;
1114                 return;
1115         }
1116         adapter->tx_hang_recheck = false;
1117
1118         if (er32(TDH(0)) == er32(TDT(0))) {
1119                 e_dbg("false hang detected, ignoring\n");
1120                 return;
1121         }
1122
1123         /* Real hang detected */
1124         netif_stop_queue(netdev);
1125
1126         e1e_rphy(hw, MII_BMSR, &phy_status);
1127         e1e_rphy(hw, MII_STAT1000, &phy_1000t_status);
1128         e1e_rphy(hw, MII_ESTATUS, &phy_ext_status);
1129
1130         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1131
1132         /* detected Hardware unit hang */
1133         e_err("Detected Hardware Unit Hang:\n"
1134               "  TDH                  <%x>\n"
1135               "  TDT                  <%x>\n"
1136               "  next_to_use          <%x>\n"
1137               "  next_to_clean        <%x>\n"
1138               "buffer_info[next_to_clean]:\n"
1139               "  time_stamp           <%lx>\n"
1140               "  next_to_watch        <%x>\n"
1141               "  jiffies              <%lx>\n"
1142               "  next_to_watch.status <%x>\n"
1143               "MAC Status             <%x>\n"
1144               "PHY Status             <%x>\n"
1145               "PHY 1000BASE-T Status  <%x>\n"
1146               "PHY Extended Status    <%x>\n"
1147               "PCI Status             <%x>\n",
1148               readl(tx_ring->head), readl(tx_ring->tail), tx_ring->next_to_use,
1149               tx_ring->next_to_clean, tx_ring->buffer_info[eop].time_stamp,
1150               eop, jiffies, eop_desc->upper.fields.status, er32(STATUS),
1151               phy_status, phy_1000t_status, phy_ext_status, pci_status);
1152
1153         e1000e_dump(adapter);
1154
1155         /* Suggest workaround for known h/w issue */
1156         if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1157                 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1158 }
1159
1160 /**
1161  * e1000e_tx_hwtstamp_work - check for Tx time stamp
1162  * @work: pointer to work struct
1163  *
1164  * This work function polls the TSYNCTXCTL valid bit to determine when a
1165  * timestamp has been taken for the current stored skb.  The timestamp must
1166  * be for this skb because only one such packet is allowed in the queue.
1167  */
1168 static void e1000e_tx_hwtstamp_work(struct work_struct *work)
1169 {
1170         struct e1000_adapter *adapter = container_of(work, struct e1000_adapter,
1171                                                      tx_hwtstamp_work);
1172         struct e1000_hw *hw = &adapter->hw;
1173
1174         if (er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID) {
1175                 struct sk_buff *skb = adapter->tx_hwtstamp_skb;
1176                 struct skb_shared_hwtstamps shhwtstamps;
1177                 u64 txstmp;
1178
1179                 txstmp = er32(TXSTMPL);
1180                 txstmp |= (u64)er32(TXSTMPH) << 32;
1181
1182                 e1000e_systim_to_hwtstamp(adapter, &shhwtstamps, txstmp);
1183
1184                 /* Clear the global tx_hwtstamp_skb pointer and force writes
1185                  * prior to notifying the stack of a Tx timestamp.
1186                  */
1187                 adapter->tx_hwtstamp_skb = NULL;
1188                 wmb(); /* force write prior to skb_tstamp_tx */
1189
1190                 skb_tstamp_tx(skb, &shhwtstamps);
1191                 dev_consume_skb_any(skb);
1192         } else if (time_after(jiffies, adapter->tx_hwtstamp_start
1193                               + adapter->tx_timeout_factor * HZ)) {
1194                 dev_kfree_skb_any(adapter->tx_hwtstamp_skb);
1195                 adapter->tx_hwtstamp_skb = NULL;
1196                 adapter->tx_hwtstamp_timeouts++;
1197                 e_warn("clearing Tx timestamp hang\n");
1198         } else {
1199                 /* reschedule to check later */
1200                 schedule_work(&adapter->tx_hwtstamp_work);
1201         }
1202 }
1203
1204 /**
1205  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1206  * @tx_ring: Tx descriptor ring
1207  *
1208  * the return value indicates whether actual cleaning was done, there
1209  * is no guarantee that everything was cleaned
1210  **/
1211 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1212 {
1213         struct e1000_adapter *adapter = tx_ring->adapter;
1214         struct net_device *netdev = adapter->netdev;
1215         struct e1000_hw *hw = &adapter->hw;
1216         struct e1000_tx_desc *tx_desc, *eop_desc;
1217         struct e1000_buffer *buffer_info;
1218         unsigned int i, eop;
1219         unsigned int count = 0;
1220         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1221         unsigned int bytes_compl = 0, pkts_compl = 0;
1222
1223         i = tx_ring->next_to_clean;
1224         eop = tx_ring->buffer_info[i].next_to_watch;
1225         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1226
1227         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1228                (count < tx_ring->count)) {
1229                 bool cleaned = false;
1230
1231                 dma_rmb();              /* read buffer_info after eop_desc */
1232                 for (; !cleaned; count++) {
1233                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1234                         buffer_info = &tx_ring->buffer_info[i];
1235                         cleaned = (i == eop);
1236
1237                         if (cleaned) {
1238                                 total_tx_packets += buffer_info->segs;
1239                                 total_tx_bytes += buffer_info->bytecount;
1240                                 if (buffer_info->skb) {
1241                                         bytes_compl += buffer_info->skb->len;
1242                                         pkts_compl++;
1243                                 }
1244                         }
1245
1246                         e1000_put_txbuf(tx_ring, buffer_info, false);
1247                         tx_desc->upper.data = 0;
1248
1249                         i++;
1250                         if (i == tx_ring->count)
1251                                 i = 0;
1252                 }
1253
1254                 if (i == tx_ring->next_to_use)
1255                         break;
1256                 eop = tx_ring->buffer_info[i].next_to_watch;
1257                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1258         }
1259
1260         tx_ring->next_to_clean = i;
1261
1262         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1263
1264 #define TX_WAKE_THRESHOLD 32
1265         if (count && netif_carrier_ok(netdev) &&
1266             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1267                 /* Make sure that anybody stopping the queue after this
1268                  * sees the new next_to_clean.
1269                  */
1270                 smp_mb();
1271
1272                 if (netif_queue_stopped(netdev) &&
1273                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1274                         netif_wake_queue(netdev);
1275                         ++adapter->restart_queue;
1276                 }
1277         }
1278
1279         if (adapter->detect_tx_hung) {
1280                 /* Detect a transmit hang in hardware, this serializes the
1281                  * check with the clearing of time_stamp and movement of i
1282                  */
1283                 adapter->detect_tx_hung = false;
1284                 if (tx_ring->buffer_info[i].time_stamp &&
1285                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1286                                + (adapter->tx_timeout_factor * HZ)) &&
1287                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1288                         schedule_work(&adapter->print_hang_task);
1289                 else
1290                         adapter->tx_hang_recheck = false;
1291         }
1292         adapter->total_tx_bytes += total_tx_bytes;
1293         adapter->total_tx_packets += total_tx_packets;
1294         return count < tx_ring->count;
1295 }
1296
1297 /**
1298  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1299  * @rx_ring: Rx descriptor ring
1300  * @work_done: output parameter for indicating completed work
1301  * @work_to_do: how many packets we can clean
1302  *
1303  * the return value indicates whether actual cleaning was done, there
1304  * is no guarantee that everything was cleaned
1305  **/
1306 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1307                                   int work_to_do)
1308 {
1309         struct e1000_adapter *adapter = rx_ring->adapter;
1310         struct e1000_hw *hw = &adapter->hw;
1311         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1312         struct net_device *netdev = adapter->netdev;
1313         struct pci_dev *pdev = adapter->pdev;
1314         struct e1000_buffer *buffer_info, *next_buffer;
1315         struct e1000_ps_page *ps_page;
1316         struct sk_buff *skb;
1317         unsigned int i, j;
1318         u32 length, staterr;
1319         int cleaned_count = 0;
1320         bool cleaned = false;
1321         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1322
1323         i = rx_ring->next_to_clean;
1324         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1325         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1326         buffer_info = &rx_ring->buffer_info[i];
1327
1328         while (staterr & E1000_RXD_STAT_DD) {
1329                 if (*work_done >= work_to_do)
1330                         break;
1331                 (*work_done)++;
1332                 skb = buffer_info->skb;
1333                 dma_rmb();      /* read descriptor and rx_buffer_info after status DD */
1334
1335                 /* in the packet split case this is header only */
1336                 prefetch(skb->data - NET_IP_ALIGN);
1337
1338                 i++;
1339                 if (i == rx_ring->count)
1340                         i = 0;
1341                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1342                 prefetch(next_rxd);
1343
1344                 next_buffer = &rx_ring->buffer_info[i];
1345
1346                 cleaned = true;
1347                 cleaned_count++;
1348                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1349                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1350                 buffer_info->dma = 0;
1351
1352                 /* see !EOP comment in other Rx routine */
1353                 if (!(staterr & E1000_RXD_STAT_EOP))
1354                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1355
1356                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1357                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1358                         dev_kfree_skb_irq(skb);
1359                         if (staterr & E1000_RXD_STAT_EOP)
1360                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1361                         goto next_desc;
1362                 }
1363
1364                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1365                              !(netdev->features & NETIF_F_RXALL))) {
1366                         dev_kfree_skb_irq(skb);
1367                         goto next_desc;
1368                 }
1369
1370                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1371
1372                 if (!length) {
1373                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1374                         dev_kfree_skb_irq(skb);
1375                         goto next_desc;
1376                 }
1377
1378                 /* Good Receive */
1379                 skb_put(skb, length);
1380
1381                 {
1382                         /* this looks ugly, but it seems compiler issues make
1383                          * it more efficient than reusing j
1384                          */
1385                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1386
1387                         /* page alloc/put takes too long and effects small
1388                          * packet throughput, so unsplit small packets and
1389                          * save the alloc/put only valid in softirq (napi)
1390                          * context to call kmap_*
1391                          */
1392                         if (l1 && (l1 <= copybreak) &&
1393                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1394                                 u8 *vaddr;
1395
1396                                 ps_page = &buffer_info->ps_pages[0];
1397
1398                                 /* there is no documentation about how to call
1399                                  * kmap_atomic, so we can't hold the mapping
1400                                  * very long
1401                                  */
1402                                 dma_sync_single_for_cpu(&pdev->dev,
1403                                                         ps_page->dma,
1404                                                         PAGE_SIZE,
1405                                                         DMA_FROM_DEVICE);
1406                                 vaddr = kmap_atomic(ps_page->page);
1407                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1408                                 kunmap_atomic(vaddr);
1409                                 dma_sync_single_for_device(&pdev->dev,
1410                                                            ps_page->dma,
1411                                                            PAGE_SIZE,
1412                                                            DMA_FROM_DEVICE);
1413
1414                                 /* remove the CRC */
1415                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1416                                         if (!(netdev->features & NETIF_F_RXFCS))
1417                                                 l1 -= 4;
1418                                 }
1419
1420                                 skb_put(skb, l1);
1421                                 goto copydone;
1422                         }       /* if */
1423                 }
1424
1425                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1426                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1427                         if (!length)
1428                                 break;
1429
1430                         ps_page = &buffer_info->ps_pages[j];
1431                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1432                                        DMA_FROM_DEVICE);
1433                         ps_page->dma = 0;
1434                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1435                         ps_page->page = NULL;
1436                         skb->len += length;
1437                         skb->data_len += length;
1438                         skb->truesize += PAGE_SIZE;
1439                 }
1440
1441                 /* strip the ethernet crc, problem is we're using pages now so
1442                  * this whole operation can get a little cpu intensive
1443                  */
1444                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1445                         if (!(netdev->features & NETIF_F_RXFCS))
1446                                 pskb_trim(skb, skb->len - 4);
1447                 }
1448
1449 copydone:
1450                 total_rx_bytes += skb->len;
1451                 total_rx_packets++;
1452
1453                 e1000_rx_checksum(adapter, staterr, skb);
1454
1455                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1456
1457                 if (rx_desc->wb.upper.header_status &
1458                     cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1459                         adapter->rx_hdr_split++;
1460
1461                 e1000_receive_skb(adapter, netdev, skb, staterr,
1462                                   rx_desc->wb.middle.vlan);
1463
1464 next_desc:
1465                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1466                 buffer_info->skb = NULL;
1467
1468                 /* return some buffers to hardware, one at a time is too slow */
1469                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1470                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1471                                               GFP_ATOMIC);
1472                         cleaned_count = 0;
1473                 }
1474
1475                 /* use prefetched values */
1476                 rx_desc = next_rxd;
1477                 buffer_info = next_buffer;
1478
1479                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1480         }
1481         rx_ring->next_to_clean = i;
1482
1483         cleaned_count = e1000_desc_unused(rx_ring);
1484         if (cleaned_count)
1485                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1486
1487         adapter->total_rx_bytes += total_rx_bytes;
1488         adapter->total_rx_packets += total_rx_packets;
1489         return cleaned;
1490 }
1491
1492 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1493                                u16 length)
1494 {
1495         bi->page = NULL;
1496         skb->len += length;
1497         skb->data_len += length;
1498         skb->truesize += PAGE_SIZE;
1499 }
1500
1501 /**
1502  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1503  * @rx_ring: Rx descriptor ring
1504  * @work_done: output parameter for indicating completed work
1505  * @work_to_do: how many packets we can clean
1506  *
1507  * the return value indicates whether actual cleaning was done, there
1508  * is no guarantee that everything was cleaned
1509  **/
1510 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1511                                      int work_to_do)
1512 {
1513         struct e1000_adapter *adapter = rx_ring->adapter;
1514         struct net_device *netdev = adapter->netdev;
1515         struct pci_dev *pdev = adapter->pdev;
1516         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1517         struct e1000_buffer *buffer_info, *next_buffer;
1518         u32 length, staterr;
1519         unsigned int i;
1520         int cleaned_count = 0;
1521         bool cleaned = false;
1522         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1523         struct skb_shared_info *shinfo;
1524
1525         i = rx_ring->next_to_clean;
1526         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1527         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1528         buffer_info = &rx_ring->buffer_info[i];
1529
1530         while (staterr & E1000_RXD_STAT_DD) {
1531                 struct sk_buff *skb;
1532
1533                 if (*work_done >= work_to_do)
1534                         break;
1535                 (*work_done)++;
1536                 dma_rmb();      /* read descriptor and rx_buffer_info after status DD */
1537
1538                 skb = buffer_info->skb;
1539                 buffer_info->skb = NULL;
1540
1541                 ++i;
1542                 if (i == rx_ring->count)
1543                         i = 0;
1544                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1545                 prefetch(next_rxd);
1546
1547                 next_buffer = &rx_ring->buffer_info[i];
1548
1549                 cleaned = true;
1550                 cleaned_count++;
1551                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1552                                DMA_FROM_DEVICE);
1553                 buffer_info->dma = 0;
1554
1555                 length = le16_to_cpu(rx_desc->wb.upper.length);
1556
1557                 /* errors is only valid for DD + EOP descriptors */
1558                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1559                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1560                               !(netdev->features & NETIF_F_RXALL)))) {
1561                         /* recycle both page and skb */
1562                         buffer_info->skb = skb;
1563                         /* an error means any chain goes out the window too */
1564                         if (rx_ring->rx_skb_top)
1565                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1566                         rx_ring->rx_skb_top = NULL;
1567                         goto next_desc;
1568                 }
1569 #define rxtop (rx_ring->rx_skb_top)
1570                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1571                         /* this descriptor is only the beginning (or middle) */
1572                         if (!rxtop) {
1573                                 /* this is the beginning of a chain */
1574                                 rxtop = skb;
1575                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1576                                                    0, length);
1577                         } else {
1578                                 /* this is the middle of a chain */
1579                                 shinfo = skb_shinfo(rxtop);
1580                                 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1581                                                    buffer_info->page, 0,
1582                                                    length);
1583                                 /* re-use the skb, only consumed the page */
1584                                 buffer_info->skb = skb;
1585                         }
1586                         e1000_consume_page(buffer_info, rxtop, length);
1587                         goto next_desc;
1588                 } else {
1589                         if (rxtop) {
1590                                 /* end of the chain */
1591                                 shinfo = skb_shinfo(rxtop);
1592                                 skb_fill_page_desc(rxtop, shinfo->nr_frags,
1593                                                    buffer_info->page, 0,
1594                                                    length);
1595                                 /* re-use the current skb, we only consumed the
1596                                  * page
1597                                  */
1598                                 buffer_info->skb = skb;
1599                                 skb = rxtop;
1600                                 rxtop = NULL;
1601                                 e1000_consume_page(buffer_info, skb, length);
1602                         } else {
1603                                 /* no chain, got EOP, this buf is the packet
1604                                  * copybreak to save the put_page/alloc_page
1605                                  */
1606                                 if (length <= copybreak &&
1607                                     skb_tailroom(skb) >= length) {
1608                                         u8 *vaddr;
1609                                         vaddr = kmap_atomic(buffer_info->page);
1610                                         memcpy(skb_tail_pointer(skb), vaddr,
1611                                                length);
1612                                         kunmap_atomic(vaddr);
1613                                         /* re-use the page, so don't erase
1614                                          * buffer_info->page
1615                                          */
1616                                         skb_put(skb, length);
1617                                 } else {
1618                                         skb_fill_page_desc(skb, 0,
1619                                                            buffer_info->page, 0,
1620                                                            length);
1621                                         e1000_consume_page(buffer_info, skb,
1622                                                            length);
1623                                 }
1624                         }
1625                 }
1626
1627                 /* Receive Checksum Offload */
1628                 e1000_rx_checksum(adapter, staterr, skb);
1629
1630                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1631
1632                 /* probably a little skewed due to removing CRC */
1633                 total_rx_bytes += skb->len;
1634                 total_rx_packets++;
1635
1636                 /* eth type trans needs skb->data to point to something */
1637                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1638                         e_err("pskb_may_pull failed.\n");
1639                         dev_kfree_skb_irq(skb);
1640                         goto next_desc;
1641                 }
1642
1643                 e1000_receive_skb(adapter, netdev, skb, staterr,
1644                                   rx_desc->wb.upper.vlan);
1645
1646 next_desc:
1647                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1648
1649                 /* return some buffers to hardware, one at a time is too slow */
1650                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1651                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1652                                               GFP_ATOMIC);
1653                         cleaned_count = 0;
1654                 }
1655
1656                 /* use prefetched values */
1657                 rx_desc = next_rxd;
1658                 buffer_info = next_buffer;
1659
1660                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1661         }
1662         rx_ring->next_to_clean = i;
1663
1664         cleaned_count = e1000_desc_unused(rx_ring);
1665         if (cleaned_count)
1666                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1667
1668         adapter->total_rx_bytes += total_rx_bytes;
1669         adapter->total_rx_packets += total_rx_packets;
1670         return cleaned;
1671 }
1672
1673 /**
1674  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1675  * @rx_ring: Rx descriptor ring
1676  **/
1677 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1678 {
1679         struct e1000_adapter *adapter = rx_ring->adapter;
1680         struct e1000_buffer *buffer_info;
1681         struct e1000_ps_page *ps_page;
1682         struct pci_dev *pdev = adapter->pdev;
1683         unsigned int i, j;
1684
1685         /* Free all the Rx ring sk_buffs */
1686         for (i = 0; i < rx_ring->count; i++) {
1687                 buffer_info = &rx_ring->buffer_info[i];
1688                 if (buffer_info->dma) {
1689                         if (adapter->clean_rx == e1000_clean_rx_irq)
1690                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1691                                                  adapter->rx_buffer_len,
1692                                                  DMA_FROM_DEVICE);
1693                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1694                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1695                                                PAGE_SIZE, DMA_FROM_DEVICE);
1696                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1697                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1698                                                  adapter->rx_ps_bsize0,
1699                                                  DMA_FROM_DEVICE);
1700                         buffer_info->dma = 0;
1701                 }
1702
1703                 if (buffer_info->page) {
1704                         put_page(buffer_info->page);
1705                         buffer_info->page = NULL;
1706                 }
1707
1708                 if (buffer_info->skb) {
1709                         dev_kfree_skb(buffer_info->skb);
1710                         buffer_info->skb = NULL;
1711                 }
1712
1713                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1714                         ps_page = &buffer_info->ps_pages[j];
1715                         if (!ps_page->page)
1716                                 break;
1717                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1718                                        DMA_FROM_DEVICE);
1719                         ps_page->dma = 0;
1720                         put_page(ps_page->page);
1721                         ps_page->page = NULL;
1722                 }
1723         }
1724
1725         /* there also may be some cached data from a chained receive */
1726         if (rx_ring->rx_skb_top) {
1727                 dev_kfree_skb(rx_ring->rx_skb_top);
1728                 rx_ring->rx_skb_top = NULL;
1729         }
1730
1731         /* Zero out the descriptor ring */
1732         memset(rx_ring->desc, 0, rx_ring->size);
1733
1734         rx_ring->next_to_clean = 0;
1735         rx_ring->next_to_use = 0;
1736         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1737 }
1738
1739 static void e1000e_downshift_workaround(struct work_struct *work)
1740 {
1741         struct e1000_adapter *adapter = container_of(work,
1742                                                      struct e1000_adapter,
1743                                                      downshift_task);
1744
1745         if (test_bit(__E1000_DOWN, &adapter->state))
1746                 return;
1747
1748         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1749 }
1750
1751 /**
1752  * e1000_intr_msi - Interrupt Handler
1753  * @irq: interrupt number
1754  * @data: pointer to a network interface device structure
1755  **/
1756 static irqreturn_t e1000_intr_msi(int __always_unused irq, void *data)
1757 {
1758         struct net_device *netdev = data;
1759         struct e1000_adapter *adapter = netdev_priv(netdev);
1760         struct e1000_hw *hw = &adapter->hw;
1761         u32 icr = er32(ICR);
1762
1763         /* read ICR disables interrupts using IAM */
1764         if (icr & E1000_ICR_LSC) {
1765                 hw->mac.get_link_status = true;
1766                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1767                  * disconnect (LSC) before accessing any PHY registers
1768                  */
1769                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1770                     (!(er32(STATUS) & E1000_STATUS_LU)))
1771                         schedule_work(&adapter->downshift_task);
1772
1773                 /* 80003ES2LAN workaround-- For packet buffer work-around on
1774                  * link down event; disable receives here in the ISR and reset
1775                  * adapter in watchdog
1776                  */
1777                 if (netif_carrier_ok(netdev) &&
1778                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1779                         /* disable receives */
1780                         u32 rctl = er32(RCTL);
1781
1782                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1783                         adapter->flags |= FLAG_RESTART_NOW;
1784                 }
1785                 /* guard against interrupt when we're going down */
1786                 if (!test_bit(__E1000_DOWN, &adapter->state))
1787                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1788         }
1789
1790         /* Reset on uncorrectable ECC error */
1791         if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1792                 u32 pbeccsts = er32(PBECCSTS);
1793
1794                 adapter->corr_errors +=
1795                     pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1796                 adapter->uncorr_errors +=
1797                     (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1798                     E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1799
1800                 /* Do the reset outside of interrupt context */
1801                 schedule_work(&adapter->reset_task);
1802
1803                 /* return immediately since reset is imminent */
1804                 return IRQ_HANDLED;
1805         }
1806
1807         if (napi_schedule_prep(&adapter->napi)) {
1808                 adapter->total_tx_bytes = 0;
1809                 adapter->total_tx_packets = 0;
1810                 adapter->total_rx_bytes = 0;
1811                 adapter->total_rx_packets = 0;
1812                 __napi_schedule(&adapter->napi);
1813         }
1814
1815         return IRQ_HANDLED;
1816 }
1817
1818 /**
1819  * e1000_intr - Interrupt Handler
1820  * @irq: interrupt number
1821  * @data: pointer to a network interface device structure
1822  **/
1823 static irqreturn_t e1000_intr(int __always_unused irq, void *data)
1824 {
1825         struct net_device *netdev = data;
1826         struct e1000_adapter *adapter = netdev_priv(netdev);
1827         struct e1000_hw *hw = &adapter->hw;
1828         u32 rctl, icr = er32(ICR);
1829
1830         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1831                 return IRQ_NONE;        /* Not our interrupt */
1832
1833         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1834          * not set, then the adapter didn't send an interrupt
1835          */
1836         if (!(icr & E1000_ICR_INT_ASSERTED))
1837                 return IRQ_NONE;
1838
1839         /* Interrupt Auto-Mask...upon reading ICR,
1840          * interrupts are masked.  No need for the
1841          * IMC write
1842          */
1843
1844         if (icr & E1000_ICR_LSC) {
1845                 hw->mac.get_link_status = true;
1846                 /* ICH8 workaround-- Call gig speed drop workaround on cable
1847                  * disconnect (LSC) before accessing any PHY registers
1848                  */
1849                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1850                     (!(er32(STATUS) & E1000_STATUS_LU)))
1851                         schedule_work(&adapter->downshift_task);
1852
1853                 /* 80003ES2LAN workaround--
1854                  * For packet buffer work-around on link down event;
1855                  * disable receives here in the ISR and
1856                  * reset adapter in watchdog
1857                  */
1858                 if (netif_carrier_ok(netdev) &&
1859                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1860                         /* disable receives */
1861                         rctl = er32(RCTL);
1862                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1863                         adapter->flags |= FLAG_RESTART_NOW;
1864                 }
1865                 /* guard against interrupt when we're going down */
1866                 if (!test_bit(__E1000_DOWN, &adapter->state))
1867                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1868         }
1869
1870         /* Reset on uncorrectable ECC error */
1871         if ((icr & E1000_ICR_ECCER) && (hw->mac.type >= e1000_pch_lpt)) {
1872                 u32 pbeccsts = er32(PBECCSTS);
1873
1874                 adapter->corr_errors +=
1875                     pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
1876                 adapter->uncorr_errors +=
1877                     (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
1878                     E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
1879
1880                 /* Do the reset outside of interrupt context */
1881                 schedule_work(&adapter->reset_task);
1882
1883                 /* return immediately since reset is imminent */
1884                 return IRQ_HANDLED;
1885         }
1886
1887         if (napi_schedule_prep(&adapter->napi)) {
1888                 adapter->total_tx_bytes = 0;
1889                 adapter->total_tx_packets = 0;
1890                 adapter->total_rx_bytes = 0;
1891                 adapter->total_rx_packets = 0;
1892                 __napi_schedule(&adapter->napi);
1893         }
1894
1895         return IRQ_HANDLED;
1896 }
1897
1898 static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
1899 {
1900         struct net_device *netdev = data;
1901         struct e1000_adapter *adapter = netdev_priv(netdev);
1902         struct e1000_hw *hw = &adapter->hw;
1903         u32 icr = er32(ICR);
1904
1905         if (icr & adapter->eiac_mask)
1906                 ew32(ICS, (icr & adapter->eiac_mask));
1907
1908         if (icr & E1000_ICR_LSC) {
1909                 hw->mac.get_link_status = true;
1910                 /* guard against interrupt when we're going down */
1911                 if (!test_bit(__E1000_DOWN, &adapter->state))
1912                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1913         }
1914
1915         if (!test_bit(__E1000_DOWN, &adapter->state))
1916                 ew32(IMS, E1000_IMS_OTHER | IMS_OTHER_MASK);
1917
1918         return IRQ_HANDLED;
1919 }
1920
1921 static irqreturn_t e1000_intr_msix_tx(int __always_unused irq, void *data)
1922 {
1923         struct net_device *netdev = data;
1924         struct e1000_adapter *adapter = netdev_priv(netdev);
1925         struct e1000_hw *hw = &adapter->hw;
1926         struct e1000_ring *tx_ring = adapter->tx_ring;
1927
1928         adapter->total_tx_bytes = 0;
1929         adapter->total_tx_packets = 0;
1930
1931         if (!e1000_clean_tx_irq(tx_ring))
1932                 /* Ring was not completely cleaned, so fire another interrupt */
1933                 ew32(ICS, tx_ring->ims_val);
1934
1935         if (!test_bit(__E1000_DOWN, &adapter->state))
1936                 ew32(IMS, adapter->tx_ring->ims_val);
1937
1938         return IRQ_HANDLED;
1939 }
1940
1941 static irqreturn_t e1000_intr_msix_rx(int __always_unused irq, void *data)
1942 {
1943         struct net_device *netdev = data;
1944         struct e1000_adapter *adapter = netdev_priv(netdev);
1945         struct e1000_ring *rx_ring = adapter->rx_ring;
1946
1947         /* Write the ITR value calculated at the end of the
1948          * previous interrupt.
1949          */
1950         if (rx_ring->set_itr) {
1951                 u32 itr = rx_ring->itr_val ?
1952                           1000000000 / (rx_ring->itr_val * 256) : 0;
1953
1954                 writel(itr, rx_ring->itr_register);
1955                 rx_ring->set_itr = 0;
1956         }
1957
1958         if (napi_schedule_prep(&adapter->napi)) {
1959                 adapter->total_rx_bytes = 0;
1960                 adapter->total_rx_packets = 0;
1961                 __napi_schedule(&adapter->napi);
1962         }
1963         return IRQ_HANDLED;
1964 }
1965
1966 /**
1967  * e1000_configure_msix - Configure MSI-X hardware
1968  * @adapter: board private structure
1969  *
1970  * e1000_configure_msix sets up the hardware to properly
1971  * generate MSI-X interrupts.
1972  **/
1973 static void e1000_configure_msix(struct e1000_adapter *adapter)
1974 {
1975         struct e1000_hw *hw = &adapter->hw;
1976         struct e1000_ring *rx_ring = adapter->rx_ring;
1977         struct e1000_ring *tx_ring = adapter->tx_ring;
1978         int vector = 0;
1979         u32 ctrl_ext, ivar = 0;
1980
1981         adapter->eiac_mask = 0;
1982
1983         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1984         if (hw->mac.type == e1000_82574) {
1985                 u32 rfctl = er32(RFCTL);
1986
1987                 rfctl |= E1000_RFCTL_ACK_DIS;
1988                 ew32(RFCTL, rfctl);
1989         }
1990
1991         /* Configure Rx vector */
1992         rx_ring->ims_val = E1000_IMS_RXQ0;
1993         adapter->eiac_mask |= rx_ring->ims_val;
1994         if (rx_ring->itr_val)
1995                 writel(1000000000 / (rx_ring->itr_val * 256),
1996                        rx_ring->itr_register);
1997         else
1998                 writel(1, rx_ring->itr_register);
1999         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
2000
2001         /* Configure Tx vector */
2002         tx_ring->ims_val = E1000_IMS_TXQ0;
2003         vector++;
2004         if (tx_ring->itr_val)
2005                 writel(1000000000 / (tx_ring->itr_val * 256),
2006                        tx_ring->itr_register);
2007         else
2008                 writel(1, tx_ring->itr_register);
2009         adapter->eiac_mask |= tx_ring->ims_val;
2010         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
2011
2012         /* set vector for Other Causes, e.g. link changes */
2013         vector++;
2014         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
2015         if (rx_ring->itr_val)
2016                 writel(1000000000 / (rx_ring->itr_val * 256),
2017                        hw->hw_addr + E1000_EITR_82574(vector));
2018         else
2019                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
2020
2021         /* Cause Tx interrupts on every write back */
2022         ivar |= BIT(31);
2023
2024         ew32(IVAR, ivar);
2025
2026         /* enable MSI-X PBA support */
2027         ctrl_ext = er32(CTRL_EXT) & ~E1000_CTRL_EXT_IAME;
2028         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR | E1000_CTRL_EXT_EIAME;
2029         ew32(CTRL_EXT, ctrl_ext);
2030         e1e_flush();
2031 }
2032
2033 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
2034 {
2035         if (adapter->msix_entries) {
2036                 pci_disable_msix(adapter->pdev);
2037                 kfree(adapter->msix_entries);
2038                 adapter->msix_entries = NULL;
2039         } else if (adapter->flags & FLAG_MSI_ENABLED) {
2040                 pci_disable_msi(adapter->pdev);
2041                 adapter->flags &= ~FLAG_MSI_ENABLED;
2042         }
2043 }
2044
2045 /**
2046  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
2047  * @adapter: board private structure
2048  *
2049  * Attempt to configure interrupts using the best available
2050  * capabilities of the hardware and kernel.
2051  **/
2052 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
2053 {
2054         int err;
2055         int i;
2056
2057         switch (adapter->int_mode) {
2058         case E1000E_INT_MODE_MSIX:
2059                 if (adapter->flags & FLAG_HAS_MSIX) {
2060                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
2061                         adapter->msix_entries = kcalloc(adapter->num_vectors,
2062                                                         sizeof(struct
2063                                                                msix_entry),
2064                                                         GFP_KERNEL);
2065                         if (adapter->msix_entries) {
2066                                 struct e1000_adapter *a = adapter;
2067
2068                                 for (i = 0; i < adapter->num_vectors; i++)
2069                                         adapter->msix_entries[i].entry = i;
2070
2071                                 err = pci_enable_msix_range(a->pdev,
2072                                                             a->msix_entries,
2073                                                             a->num_vectors,
2074                                                             a->num_vectors);
2075                                 if (err > 0)
2076                                         return;
2077                         }
2078                         /* MSI-X failed, so fall through and try MSI */
2079                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
2080                         e1000e_reset_interrupt_capability(adapter);
2081                 }
2082                 adapter->int_mode = E1000E_INT_MODE_MSI;
2083                 fallthrough;
2084         case E1000E_INT_MODE_MSI:
2085                 if (!pci_enable_msi(adapter->pdev)) {
2086                         adapter->flags |= FLAG_MSI_ENABLED;
2087                 } else {
2088                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
2089                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
2090                 }
2091                 fallthrough;
2092         case E1000E_INT_MODE_LEGACY:
2093                 /* Don't do anything; this is the system default */
2094                 break;
2095         }
2096
2097         /* store the number of vectors being used */
2098         adapter->num_vectors = 1;
2099 }
2100
2101 /**
2102  * e1000_request_msix - Initialize MSI-X interrupts
2103  * @adapter: board private structure
2104  *
2105  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
2106  * kernel.
2107  **/
2108 static int e1000_request_msix(struct e1000_adapter *adapter)
2109 {
2110         struct net_device *netdev = adapter->netdev;
2111         int err = 0, vector = 0;
2112
2113         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2114                 snprintf(adapter->rx_ring->name,
2115                          sizeof(adapter->rx_ring->name) - 1,
2116                          "%.14s-rx-0", netdev->name);
2117         else
2118                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
2119         err = request_irq(adapter->msix_entries[vector].vector,
2120                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
2121                           netdev);
2122         if (err)
2123                 return err;
2124         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
2125             E1000_EITR_82574(vector);
2126         adapter->rx_ring->itr_val = adapter->itr;
2127         vector++;
2128
2129         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2130                 snprintf(adapter->tx_ring->name,
2131                          sizeof(adapter->tx_ring->name) - 1,
2132                          "%.14s-tx-0", netdev->name);
2133         else
2134                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2135         err = request_irq(adapter->msix_entries[vector].vector,
2136                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2137                           netdev);
2138         if (err)
2139                 return err;
2140         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2141             E1000_EITR_82574(vector);
2142         adapter->tx_ring->itr_val = adapter->itr;
2143         vector++;
2144
2145         err = request_irq(adapter->msix_entries[vector].vector,
2146                           e1000_msix_other, 0, netdev->name, netdev);
2147         if (err)
2148                 return err;
2149
2150         e1000_configure_msix(adapter);
2151
2152         return 0;
2153 }
2154
2155 /**
2156  * e1000_request_irq - initialize interrupts
2157  * @adapter: board private structure
2158  *
2159  * Attempts to configure interrupts using the best available
2160  * capabilities of the hardware and kernel.
2161  **/
2162 static int e1000_request_irq(struct e1000_adapter *adapter)
2163 {
2164         struct net_device *netdev = adapter->netdev;
2165         int err;
2166
2167         if (adapter->msix_entries) {
2168                 err = e1000_request_msix(adapter);
2169                 if (!err)
2170                         return err;
2171                 /* fall back to MSI */
2172                 e1000e_reset_interrupt_capability(adapter);
2173                 adapter->int_mode = E1000E_INT_MODE_MSI;
2174                 e1000e_set_interrupt_capability(adapter);
2175         }
2176         if (adapter->flags & FLAG_MSI_ENABLED) {
2177                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2178                                   netdev->name, netdev);
2179                 if (!err)
2180                         return err;
2181
2182                 /* fall back to legacy interrupt */
2183                 e1000e_reset_interrupt_capability(adapter);
2184                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2185         }
2186
2187         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2188                           netdev->name, netdev);
2189         if (err)
2190                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2191
2192         return err;
2193 }
2194
2195 static void e1000_free_irq(struct e1000_adapter *adapter)
2196 {
2197         struct net_device *netdev = adapter->netdev;
2198
2199         if (adapter->msix_entries) {
2200                 int vector = 0;
2201
2202                 free_irq(adapter->msix_entries[vector].vector, netdev);
2203                 vector++;
2204
2205                 free_irq(adapter->msix_entries[vector].vector, netdev);
2206                 vector++;
2207
2208                 /* Other Causes interrupt vector */
2209                 free_irq(adapter->msix_entries[vector].vector, netdev);
2210                 return;
2211         }
2212
2213         free_irq(adapter->pdev->irq, netdev);
2214 }
2215
2216 /**
2217  * e1000_irq_disable - Mask off interrupt generation on the NIC
2218  * @adapter: board private structure
2219  **/
2220 static void e1000_irq_disable(struct e1000_adapter *adapter)
2221 {
2222         struct e1000_hw *hw = &adapter->hw;
2223
2224         ew32(IMC, ~0);
2225         if (adapter->msix_entries)
2226                 ew32(EIAC_82574, 0);
2227         e1e_flush();
2228
2229         if (adapter->msix_entries) {
2230                 int i;
2231
2232                 for (i = 0; i < adapter->num_vectors; i++)
2233                         synchronize_irq(adapter->msix_entries[i].vector);
2234         } else {
2235                 synchronize_irq(adapter->pdev->irq);
2236         }
2237 }
2238
2239 /**
2240  * e1000_irq_enable - Enable default interrupt generation settings
2241  * @adapter: board private structure
2242  **/
2243 static void e1000_irq_enable(struct e1000_adapter *adapter)
2244 {
2245         struct e1000_hw *hw = &adapter->hw;
2246
2247         if (adapter->msix_entries) {
2248                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2249                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER |
2250                      IMS_OTHER_MASK);
2251         } else if (hw->mac.type >= e1000_pch_lpt) {
2252                 ew32(IMS, IMS_ENABLE_MASK | E1000_IMS_ECCER);
2253         } else {
2254                 ew32(IMS, IMS_ENABLE_MASK);
2255         }
2256         e1e_flush();
2257 }
2258
2259 /**
2260  * e1000e_get_hw_control - get control of the h/w from f/w
2261  * @adapter: address of board private structure
2262  *
2263  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2264  * For ASF and Pass Through versions of f/w this means that
2265  * the driver is loaded. For AMT version (only with 82573)
2266  * of the f/w this means that the network i/f is open.
2267  **/
2268 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2269 {
2270         struct e1000_hw *hw = &adapter->hw;
2271         u32 ctrl_ext;
2272         u32 swsm;
2273
2274         /* Let firmware know the driver has taken over */
2275         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2276                 swsm = er32(SWSM);
2277                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2278         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2279                 ctrl_ext = er32(CTRL_EXT);
2280                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2281         }
2282 }
2283
2284 /**
2285  * e1000e_release_hw_control - release control of the h/w to f/w
2286  * @adapter: address of board private structure
2287  *
2288  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2289  * For ASF and Pass Through versions of f/w this means that the
2290  * driver is no longer loaded. For AMT version (only with 82573) i
2291  * of the f/w this means that the network i/f is closed.
2292  *
2293  **/
2294 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2295 {
2296         struct e1000_hw *hw = &adapter->hw;
2297         u32 ctrl_ext;
2298         u32 swsm;
2299
2300         /* Let firmware taken over control of h/w */
2301         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2302                 swsm = er32(SWSM);
2303                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2304         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2305                 ctrl_ext = er32(CTRL_EXT);
2306                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2307         }
2308 }
2309
2310 /**
2311  * e1000_alloc_ring_dma - allocate memory for a ring structure
2312  * @adapter: board private structure
2313  * @ring: ring struct for which to allocate dma
2314  **/
2315 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2316                                 struct e1000_ring *ring)
2317 {
2318         struct pci_dev *pdev = adapter->pdev;
2319
2320         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2321                                         GFP_KERNEL);
2322         if (!ring->desc)
2323                 return -ENOMEM;
2324
2325         return 0;
2326 }
2327
2328 /**
2329  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2330  * @tx_ring: Tx descriptor ring
2331  *
2332  * Return 0 on success, negative on failure
2333  **/
2334 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2335 {
2336         struct e1000_adapter *adapter = tx_ring->adapter;
2337         int err = -ENOMEM, size;
2338
2339         size = sizeof(struct e1000_buffer) * tx_ring->count;
2340         tx_ring->buffer_info = vzalloc(size);
2341         if (!tx_ring->buffer_info)
2342                 goto err;
2343
2344         /* round up to nearest 4K */
2345         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2346         tx_ring->size = ALIGN(tx_ring->size, 4096);
2347
2348         err = e1000_alloc_ring_dma(adapter, tx_ring);
2349         if (err)
2350                 goto err;
2351
2352         tx_ring->next_to_use = 0;
2353         tx_ring->next_to_clean = 0;
2354
2355         return 0;
2356 err:
2357         vfree(tx_ring->buffer_info);
2358         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2359         return err;
2360 }
2361
2362 /**
2363  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2364  * @rx_ring: Rx descriptor ring
2365  *
2366  * Returns 0 on success, negative on failure
2367  **/
2368 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2369 {
2370         struct e1000_adapter *adapter = rx_ring->adapter;
2371         struct e1000_buffer *buffer_info;
2372         int i, size, desc_len, err = -ENOMEM;
2373
2374         size = sizeof(struct e1000_buffer) * rx_ring->count;
2375         rx_ring->buffer_info = vzalloc(size);
2376         if (!rx_ring->buffer_info)
2377                 goto err;
2378
2379         for (i = 0; i < rx_ring->count; i++) {
2380                 buffer_info = &rx_ring->buffer_info[i];
2381                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2382                                                 sizeof(struct e1000_ps_page),
2383                                                 GFP_KERNEL);
2384                 if (!buffer_info->ps_pages)
2385                         goto err_pages;
2386         }
2387
2388         desc_len = sizeof(union e1000_rx_desc_packet_split);
2389
2390         /* Round up to nearest 4K */
2391         rx_ring->size = rx_ring->count * desc_len;
2392         rx_ring->size = ALIGN(rx_ring->size, 4096);
2393
2394         err = e1000_alloc_ring_dma(adapter, rx_ring);
2395         if (err)
2396                 goto err_pages;
2397
2398         rx_ring->next_to_clean = 0;
2399         rx_ring->next_to_use = 0;
2400         rx_ring->rx_skb_top = NULL;
2401
2402         return 0;
2403
2404 err_pages:
2405         for (i = 0; i < rx_ring->count; i++) {
2406                 buffer_info = &rx_ring->buffer_info[i];
2407                 kfree(buffer_info->ps_pages);
2408         }
2409 err:
2410         vfree(rx_ring->buffer_info);
2411         e_err("Unable to allocate memory for the receive descriptor ring\n");
2412         return err;
2413 }
2414
2415 /**
2416  * e1000_clean_tx_ring - Free Tx Buffers
2417  * @tx_ring: Tx descriptor ring
2418  **/
2419 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2420 {
2421         struct e1000_adapter *adapter = tx_ring->adapter;
2422         struct e1000_buffer *buffer_info;
2423         unsigned long size;
2424         unsigned int i;
2425
2426         for (i = 0; i < tx_ring->count; i++) {
2427                 buffer_info = &tx_ring->buffer_info[i];
2428                 e1000_put_txbuf(tx_ring, buffer_info, false);
2429         }
2430
2431         netdev_reset_queue(adapter->netdev);
2432         size = sizeof(struct e1000_buffer) * tx_ring->count;
2433         memset(tx_ring->buffer_info, 0, size);
2434
2435         memset(tx_ring->desc, 0, tx_ring->size);
2436
2437         tx_ring->next_to_use = 0;
2438         tx_ring->next_to_clean = 0;
2439 }
2440
2441 /**
2442  * e1000e_free_tx_resources - Free Tx Resources per Queue
2443  * @tx_ring: Tx descriptor ring
2444  *
2445  * Free all transmit software resources
2446  **/
2447 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2448 {
2449         struct e1000_adapter *adapter = tx_ring->adapter;
2450         struct pci_dev *pdev = adapter->pdev;
2451
2452         e1000_clean_tx_ring(tx_ring);
2453
2454         vfree(tx_ring->buffer_info);
2455         tx_ring->buffer_info = NULL;
2456
2457         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2458                           tx_ring->dma);
2459         tx_ring->desc = NULL;
2460 }
2461
2462 /**
2463  * e1000e_free_rx_resources - Free Rx Resources
2464  * @rx_ring: Rx descriptor ring
2465  *
2466  * Free all receive software resources
2467  **/
2468 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2469 {
2470         struct e1000_adapter *adapter = rx_ring->adapter;
2471         struct pci_dev *pdev = adapter->pdev;
2472         int i;
2473
2474         e1000_clean_rx_ring(rx_ring);
2475
2476         for (i = 0; i < rx_ring->count; i++)
2477                 kfree(rx_ring->buffer_info[i].ps_pages);
2478
2479         vfree(rx_ring->buffer_info);
2480         rx_ring->buffer_info = NULL;
2481
2482         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2483                           rx_ring->dma);
2484         rx_ring->desc = NULL;
2485 }
2486
2487 /**
2488  * e1000_update_itr - update the dynamic ITR value based on statistics
2489  * @itr_setting: current adapter->itr
2490  * @packets: the number of packets during this measurement interval
2491  * @bytes: the number of bytes during this measurement interval
2492  *
2493  *      Stores a new ITR value based on packets and byte
2494  *      counts during the last interrupt.  The advantage of per interrupt
2495  *      computation is faster updates and more accurate ITR for the current
2496  *      traffic pattern.  Constants in this function were computed
2497  *      based on theoretical maximum wire speed and thresholds were set based
2498  *      on testing data as well as attempting to minimize response time
2499  *      while increasing bulk throughput.  This functionality is controlled
2500  *      by the InterruptThrottleRate module parameter.
2501  **/
2502 static unsigned int e1000_update_itr(u16 itr_setting, int packets, int bytes)
2503 {
2504         unsigned int retval = itr_setting;
2505
2506         if (packets == 0)
2507                 return itr_setting;
2508
2509         switch (itr_setting) {
2510         case lowest_latency:
2511                 /* handle TSO and jumbo frames */
2512                 if (bytes / packets > 8000)
2513                         retval = bulk_latency;
2514                 else if ((packets < 5) && (bytes > 512))
2515                         retval = low_latency;
2516                 break;
2517         case low_latency:       /* 50 usec aka 20000 ints/s */
2518                 if (bytes > 10000) {
2519                         /* this if handles the TSO accounting */
2520                         if (bytes / packets > 8000)
2521                                 retval = bulk_latency;
2522                         else if ((packets < 10) || ((bytes / packets) > 1200))
2523                                 retval = bulk_latency;
2524                         else if ((packets > 35))
2525                                 retval = lowest_latency;
2526                 } else if (bytes / packets > 2000) {
2527                         retval = bulk_latency;
2528                 } else if (packets <= 2 && bytes < 512) {
2529                         retval = lowest_latency;
2530                 }
2531                 break;
2532         case bulk_latency:      /* 250 usec aka 4000 ints/s */
2533                 if (bytes > 25000) {
2534                         if (packets > 35)
2535                                 retval = low_latency;
2536                 } else if (bytes < 6000) {
2537                         retval = low_latency;
2538                 }
2539                 break;
2540         }
2541
2542         return retval;
2543 }
2544
2545 static void e1000_set_itr(struct e1000_adapter *adapter)
2546 {
2547         u16 current_itr;
2548         u32 new_itr = adapter->itr;
2549
2550         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2551         if (adapter->link_speed != SPEED_1000) {
2552                 current_itr = 0;
2553                 new_itr = 4000;
2554                 goto set_itr_now;
2555         }
2556
2557         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2558                 new_itr = 0;
2559                 goto set_itr_now;
2560         }
2561
2562         adapter->tx_itr = e1000_update_itr(adapter->tx_itr,
2563                                            adapter->total_tx_packets,
2564                                            adapter->total_tx_bytes);
2565         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2566         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2567                 adapter->tx_itr = low_latency;
2568
2569         adapter->rx_itr = e1000_update_itr(adapter->rx_itr,
2570                                            adapter->total_rx_packets,
2571                                            adapter->total_rx_bytes);
2572         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2573         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2574                 adapter->rx_itr = low_latency;
2575
2576         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2577
2578         /* counts and packets in update_itr are dependent on these numbers */
2579         switch (current_itr) {
2580         case lowest_latency:
2581                 new_itr = 70000;
2582                 break;
2583         case low_latency:
2584                 new_itr = 20000;        /* aka hwitr = ~200 */
2585                 break;
2586         case bulk_latency:
2587                 new_itr = 4000;
2588                 break;
2589         default:
2590                 break;
2591         }
2592
2593 set_itr_now:
2594         if (new_itr != adapter->itr) {
2595                 /* this attempts to bias the interrupt rate towards Bulk
2596                  * by adding intermediate steps when interrupt rate is
2597                  * increasing
2598                  */
2599                 new_itr = new_itr > adapter->itr ?
2600                     min(adapter->itr + (new_itr >> 2), new_itr) : new_itr;
2601                 adapter->itr = new_itr;
2602                 adapter->rx_ring->itr_val = new_itr;
2603                 if (adapter->msix_entries)
2604                         adapter->rx_ring->set_itr = 1;
2605                 else
2606                         e1000e_write_itr(adapter, new_itr);
2607         }
2608 }
2609
2610 /**
2611  * e1000e_write_itr - write the ITR value to the appropriate registers
2612  * @adapter: address of board private structure
2613  * @itr: new ITR value to program
2614  *
2615  * e1000e_write_itr determines if the adapter is in MSI-X mode
2616  * and, if so, writes the EITR registers with the ITR value.
2617  * Otherwise, it writes the ITR value into the ITR register.
2618  **/
2619 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2620 {
2621         struct e1000_hw *hw = &adapter->hw;
2622         u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2623
2624         if (adapter->msix_entries) {
2625                 int vector;
2626
2627                 for (vector = 0; vector < adapter->num_vectors; vector++)
2628                         writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2629         } else {
2630                 ew32(ITR, new_itr);
2631         }
2632 }
2633
2634 /**
2635  * e1000_alloc_queues - Allocate memory for all rings
2636  * @adapter: board private structure to initialize
2637  **/
2638 static int e1000_alloc_queues(struct e1000_adapter *adapter)
2639 {
2640         int size = sizeof(struct e1000_ring);
2641
2642         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2643         if (!adapter->tx_ring)
2644                 goto err;
2645         adapter->tx_ring->count = adapter->tx_ring_count;
2646         adapter->tx_ring->adapter = adapter;
2647
2648         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2649         if (!adapter->rx_ring)
2650                 goto err;
2651         adapter->rx_ring->count = adapter->rx_ring_count;
2652         adapter->rx_ring->adapter = adapter;
2653
2654         return 0;
2655 err:
2656         e_err("Unable to allocate memory for queues\n");
2657         kfree(adapter->rx_ring);
2658         kfree(adapter->tx_ring);
2659         return -ENOMEM;
2660 }
2661
2662 /**
2663  * e1000e_poll - NAPI Rx polling callback
2664  * @napi: struct associated with this polling callback
2665  * @budget: number of packets driver is allowed to process this poll
2666  **/
2667 static int e1000e_poll(struct napi_struct *napi, int budget)
2668 {
2669         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2670                                                      napi);
2671         struct e1000_hw *hw = &adapter->hw;
2672         struct net_device *poll_dev = adapter->netdev;
2673         int tx_cleaned = 1, work_done = 0;
2674
2675         adapter = netdev_priv(poll_dev);
2676
2677         if (!adapter->msix_entries ||
2678             (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2679                 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2680
2681         adapter->clean_rx(adapter->rx_ring, &work_done, budget);
2682
2683         if (!tx_cleaned || work_done == budget)
2684                 return budget;
2685
2686         /* Exit the polling mode, but don't re-enable interrupts if stack might
2687          * poll us due to busy-polling
2688          */
2689         if (likely(napi_complete_done(napi, work_done))) {
2690                 if (adapter->itr_setting & 3)
2691                         e1000_set_itr(adapter);
2692                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2693                         if (adapter->msix_entries)
2694                                 ew32(IMS, adapter->rx_ring->ims_val);
2695                         else
2696                                 e1000_irq_enable(adapter);
2697                 }
2698         }
2699
2700         return work_done;
2701 }
2702
2703 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
2704                                  __always_unused __be16 proto, u16 vid)
2705 {
2706         struct e1000_adapter *adapter = netdev_priv(netdev);
2707         struct e1000_hw *hw = &adapter->hw;
2708         u32 vfta, index;
2709
2710         /* don't update vlan cookie if already programmed */
2711         if ((adapter->hw.mng_cookie.status &
2712              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2713             (vid == adapter->mng_vlan_id))
2714                 return 0;
2715
2716         /* add VID to filter table */
2717         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2718                 index = (vid >> 5) & 0x7F;
2719                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2720                 vfta |= BIT((vid & 0x1F));
2721                 hw->mac.ops.write_vfta(hw, index, vfta);
2722         }
2723
2724         set_bit(vid, adapter->active_vlans);
2725
2726         return 0;
2727 }
2728
2729 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
2730                                   __always_unused __be16 proto, u16 vid)
2731 {
2732         struct e1000_adapter *adapter = netdev_priv(netdev);
2733         struct e1000_hw *hw = &adapter->hw;
2734         u32 vfta, index;
2735
2736         if ((adapter->hw.mng_cookie.status &
2737              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2738             (vid == adapter->mng_vlan_id)) {
2739                 /* release control to f/w */
2740                 e1000e_release_hw_control(adapter);
2741                 return 0;
2742         }
2743
2744         /* remove VID from filter table */
2745         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2746                 index = (vid >> 5) & 0x7F;
2747                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2748                 vfta &= ~BIT((vid & 0x1F));
2749                 hw->mac.ops.write_vfta(hw, index, vfta);
2750         }
2751
2752         clear_bit(vid, adapter->active_vlans);
2753
2754         return 0;
2755 }
2756
2757 /**
2758  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2759  * @adapter: board private structure to initialize
2760  **/
2761 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2762 {
2763         struct net_device *netdev = adapter->netdev;
2764         struct e1000_hw *hw = &adapter->hw;
2765         u32 rctl;
2766
2767         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2768                 /* disable VLAN receive filtering */
2769                 rctl = er32(RCTL);
2770                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2771                 ew32(RCTL, rctl);
2772
2773                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2774                         e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
2775                                                adapter->mng_vlan_id);
2776                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2777                 }
2778         }
2779 }
2780
2781 /**
2782  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2783  * @adapter: board private structure to initialize
2784  **/
2785 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2786 {
2787         struct e1000_hw *hw = &adapter->hw;
2788         u32 rctl;
2789
2790         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2791                 /* enable VLAN receive filtering */
2792                 rctl = er32(RCTL);
2793                 rctl |= E1000_RCTL_VFE;
2794                 rctl &= ~E1000_RCTL_CFIEN;
2795                 ew32(RCTL, rctl);
2796         }
2797 }
2798
2799 /**
2800  * e1000e_vlan_strip_disable - helper to disable HW VLAN stripping
2801  * @adapter: board private structure to initialize
2802  **/
2803 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2804 {
2805         struct e1000_hw *hw = &adapter->hw;
2806         u32 ctrl;
2807
2808         /* disable VLAN tag insert/strip */
2809         ctrl = er32(CTRL);
2810         ctrl &= ~E1000_CTRL_VME;
2811         ew32(CTRL, ctrl);
2812 }
2813
2814 /**
2815  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2816  * @adapter: board private structure to initialize
2817  **/
2818 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2819 {
2820         struct e1000_hw *hw = &adapter->hw;
2821         u32 ctrl;
2822
2823         /* enable VLAN tag insert/strip */
2824         ctrl = er32(CTRL);
2825         ctrl |= E1000_CTRL_VME;
2826         ew32(CTRL, ctrl);
2827 }
2828
2829 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2830 {
2831         struct net_device *netdev = adapter->netdev;
2832         u16 vid = adapter->hw.mng_cookie.vlan_id;
2833         u16 old_vid = adapter->mng_vlan_id;
2834
2835         if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2836                 e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
2837                 adapter->mng_vlan_id = vid;
2838         }
2839
2840         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2841                 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q), old_vid);
2842 }
2843
2844 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2845 {
2846         u16 vid;
2847
2848         e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
2849
2850         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2851             e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2852 }
2853
2854 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2855 {
2856         struct e1000_hw *hw = &adapter->hw;
2857         u32 manc, manc2h, mdef, i, j;
2858
2859         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2860                 return;
2861
2862         manc = er32(MANC);
2863
2864         /* enable receiving management packets to the host. this will probably
2865          * generate destination unreachable messages from the host OS, but
2866          * the packets will be handled on SMBUS
2867          */
2868         manc |= E1000_MANC_EN_MNG2HOST;
2869         manc2h = er32(MANC2H);
2870
2871         switch (hw->mac.type) {
2872         default:
2873                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2874                 break;
2875         case e1000_82574:
2876         case e1000_82583:
2877                 /* Check if IPMI pass-through decision filter already exists;
2878                  * if so, enable it.
2879                  */
2880                 for (i = 0, j = 0; i < 8; i++) {
2881                         mdef = er32(MDEF(i));
2882
2883                         /* Ignore filters with anything other than IPMI ports */
2884                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2885                                 continue;
2886
2887                         /* Enable this decision filter in MANC2H */
2888                         if (mdef)
2889                                 manc2h |= BIT(i);
2890
2891                         j |= mdef;
2892                 }
2893
2894                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2895                         break;
2896
2897                 /* Create new decision filter in an empty filter */
2898                 for (i = 0, j = 0; i < 8; i++)
2899                         if (er32(MDEF(i)) == 0) {
2900                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2901                                                E1000_MDEF_PORT_664));
2902                                 manc2h |= BIT(1);
2903                                 j++;
2904                                 break;
2905                         }
2906
2907                 if (!j)
2908                         e_warn("Unable to create IPMI pass-through filter\n");
2909                 break;
2910         }
2911
2912         ew32(MANC2H, manc2h);
2913         ew32(MANC, manc);
2914 }
2915
2916 /**
2917  * e1000_configure_tx - Configure Transmit Unit after Reset
2918  * @adapter: board private structure
2919  *
2920  * Configure the Tx unit of the MAC after a reset.
2921  **/
2922 static void e1000_configure_tx(struct e1000_adapter *adapter)
2923 {
2924         struct e1000_hw *hw = &adapter->hw;
2925         struct e1000_ring *tx_ring = adapter->tx_ring;
2926         u64 tdba;
2927         u32 tdlen, tctl, tarc;
2928
2929         /* Setup the HW Tx Head and Tail descriptor pointers */
2930         tdba = tx_ring->dma;
2931         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2932         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2933         ew32(TDBAH(0), (tdba >> 32));
2934         ew32(TDLEN(0), tdlen);
2935         ew32(TDH(0), 0);
2936         ew32(TDT(0), 0);
2937         tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2938         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2939
2940         writel(0, tx_ring->head);
2941         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2942                 e1000e_update_tdt_wa(tx_ring, 0);
2943         else
2944                 writel(0, tx_ring->tail);
2945
2946         /* Set the Tx Interrupt Delay register */
2947         ew32(TIDV, adapter->tx_int_delay);
2948         /* Tx irq moderation */
2949         ew32(TADV, adapter->tx_abs_int_delay);
2950
2951         if (adapter->flags2 & FLAG2_DMA_BURST) {
2952                 u32 txdctl = er32(TXDCTL(0));
2953
2954                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2955                             E1000_TXDCTL_WTHRESH);
2956                 /* set up some performance related parameters to encourage the
2957                  * hardware to use the bus more efficiently in bursts, depends
2958                  * on the tx_int_delay to be enabled,
2959                  * wthresh = 1 ==> burst write is disabled to avoid Tx stalls
2960                  * hthresh = 1 ==> prefetch when one or more available
2961                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2962                  * BEWARE: this seems to work but should be considered first if
2963                  * there are Tx hangs or other Tx related bugs
2964                  */
2965                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2966                 ew32(TXDCTL(0), txdctl);
2967         }
2968         /* erratum work around: set txdctl the same for both queues */
2969         ew32(TXDCTL(1), er32(TXDCTL(0)));
2970
2971         /* Program the Transmit Control Register */
2972         tctl = er32(TCTL);
2973         tctl &= ~E1000_TCTL_CT;
2974         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2975                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2976
2977         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2978                 tarc = er32(TARC(0));
2979                 /* set the speed mode bit, we'll clear it if we're not at
2980                  * gigabit link later
2981                  */
2982 #define SPEED_MODE_BIT BIT(21)
2983                 tarc |= SPEED_MODE_BIT;
2984                 ew32(TARC(0), tarc);
2985         }
2986
2987         /* errata: program both queues to unweighted RR */
2988         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2989                 tarc = er32(TARC(0));
2990                 tarc |= 1;
2991                 ew32(TARC(0), tarc);
2992                 tarc = er32(TARC(1));
2993                 tarc |= 1;
2994                 ew32(TARC(1), tarc);
2995         }
2996
2997         /* Setup Transmit Descriptor Settings for eop descriptor */
2998         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2999
3000         /* only set IDE if we are delaying interrupts using the timers */
3001         if (adapter->tx_int_delay)
3002                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
3003
3004         /* enable Report Status bit */
3005         adapter->txd_cmd |= E1000_TXD_CMD_RS;
3006
3007         ew32(TCTL, tctl);
3008
3009         hw->mac.ops.config_collision_dist(hw);
3010
3011         /* SPT and KBL Si errata workaround to avoid data corruption */
3012         if (hw->mac.type == e1000_pch_spt) {
3013                 u32 reg_val;
3014
3015                 reg_val = er32(IOSFPC);
3016                 reg_val |= E1000_RCTL_RDMTS_HEX;
3017                 ew32(IOSFPC, reg_val);
3018
3019                 reg_val = er32(TARC(0));
3020                 /* SPT and KBL Si errata workaround to avoid Tx hang.
3021                  * Dropping the number of outstanding requests from
3022                  * 3 to 2 in order to avoid a buffer overrun.
3023                  */
3024                 reg_val &= ~E1000_TARC0_CB_MULTIQ_3_REQ;
3025                 reg_val |= E1000_TARC0_CB_MULTIQ_2_REQ;
3026                 ew32(TARC(0), reg_val);
3027         }
3028 }
3029
3030 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
3031                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
3032
3033 /**
3034  * e1000_setup_rctl - configure the receive control registers
3035  * @adapter: Board private structure
3036  **/
3037 static void e1000_setup_rctl(struct e1000_adapter *adapter)
3038 {
3039         struct e1000_hw *hw = &adapter->hw;
3040         u32 rctl, rfctl;
3041         u32 pages = 0;
3042
3043         /* Workaround Si errata on PCHx - configure jumbo frame flow.
3044          * If jumbo frames not set, program related MAC/PHY registers
3045          * to h/w defaults
3046          */
3047         if (hw->mac.type >= e1000_pch2lan) {
3048                 s32 ret_val;
3049
3050                 if (adapter->netdev->mtu > ETH_DATA_LEN)
3051                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
3052                 else
3053                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
3054
3055                 if (ret_val)
3056                         e_dbg("failed to enable|disable jumbo frame workaround mode\n");
3057         }
3058
3059         /* Program MC offset vector base */
3060         rctl = er32(RCTL);
3061         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3062         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
3063             E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
3064             (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3065
3066         /* Do not Store bad packets */
3067         rctl &= ~E1000_RCTL_SBP;
3068
3069         /* Enable Long Packet receive */
3070         if (adapter->netdev->mtu <= ETH_DATA_LEN)
3071                 rctl &= ~E1000_RCTL_LPE;
3072         else
3073                 rctl |= E1000_RCTL_LPE;
3074
3075         /* Some systems expect that the CRC is included in SMBUS traffic. The
3076          * hardware strips the CRC before sending to both SMBUS (BMC) and to
3077          * host memory when this is enabled
3078          */
3079         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
3080                 rctl |= E1000_RCTL_SECRC;
3081
3082         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
3083         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
3084                 u16 phy_data;
3085
3086                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
3087                 phy_data &= 0xfff8;
3088                 phy_data |= BIT(2);
3089                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
3090
3091                 e1e_rphy(hw, 22, &phy_data);
3092                 phy_data &= 0x0fff;
3093                 phy_data |= BIT(14);
3094                 e1e_wphy(hw, 0x10, 0x2823);
3095                 e1e_wphy(hw, 0x11, 0x0003);
3096                 e1e_wphy(hw, 22, phy_data);
3097         }
3098
3099         /* Setup buffer sizes */
3100         rctl &= ~E1000_RCTL_SZ_4096;
3101         rctl |= E1000_RCTL_BSEX;
3102         switch (adapter->rx_buffer_len) {
3103         case 2048:
3104         default:
3105                 rctl |= E1000_RCTL_SZ_2048;
3106                 rctl &= ~E1000_RCTL_BSEX;
3107                 break;
3108         case 4096:
3109                 rctl |= E1000_RCTL_SZ_4096;
3110                 break;
3111         case 8192:
3112                 rctl |= E1000_RCTL_SZ_8192;
3113                 break;
3114         case 16384:
3115                 rctl |= E1000_RCTL_SZ_16384;
3116                 break;
3117         }
3118
3119         /* Enable Extended Status in all Receive Descriptors */
3120         rfctl = er32(RFCTL);
3121         rfctl |= E1000_RFCTL_EXTEN;
3122         ew32(RFCTL, rfctl);
3123
3124         /* 82571 and greater support packet-split where the protocol
3125          * header is placed in skb->data and the packet data is
3126          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
3127          * In the case of a non-split, skb->data is linearly filled,
3128          * followed by the page buffers.  Therefore, skb->data is
3129          * sized to hold the largest protocol header.
3130          *
3131          * allocations using alloc_page take too long for regular MTU
3132          * so only enable packet split for jumbo frames
3133          *
3134          * Using pages when the page size is greater than 16k wastes
3135          * a lot of memory, since we allocate 3 pages at all times
3136          * per packet.
3137          */
3138         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
3139         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
3140                 adapter->rx_ps_pages = pages;
3141         else
3142                 adapter->rx_ps_pages = 0;
3143
3144         if (adapter->rx_ps_pages) {
3145                 u32 psrctl = 0;
3146
3147                 /* Enable Packet split descriptors */
3148                 rctl |= E1000_RCTL_DTYP_PS;
3149
3150                 psrctl |= adapter->rx_ps_bsize0 >> E1000_PSRCTL_BSIZE0_SHIFT;
3151
3152                 switch (adapter->rx_ps_pages) {
3153                 case 3:
3154                         psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE3_SHIFT;
3155                         fallthrough;
3156                 case 2:
3157                         psrctl |= PAGE_SIZE << E1000_PSRCTL_BSIZE2_SHIFT;
3158                         fallthrough;
3159                 case 1:
3160                         psrctl |= PAGE_SIZE >> E1000_PSRCTL_BSIZE1_SHIFT;
3161                         break;
3162                 }
3163
3164                 ew32(PSRCTL, psrctl);
3165         }
3166
3167         /* This is useful for sniffing bad packets. */
3168         if (adapter->netdev->features & NETIF_F_RXALL) {
3169                 /* UPE and MPE will be handled by normal PROMISC logic
3170                  * in e1000e_set_rx_mode
3171                  */
3172                 rctl |= (E1000_RCTL_SBP |       /* Receive bad packets */
3173                          E1000_RCTL_BAM |       /* RX All Bcast Pkts */
3174                          E1000_RCTL_PMCF);      /* RX All MAC Ctrl Pkts */
3175
3176                 rctl &= ~(E1000_RCTL_VFE |      /* Disable VLAN filter */
3177                           E1000_RCTL_DPF |      /* Allow filtered pause */
3178                           E1000_RCTL_CFIEN);    /* Dis VLAN CFIEN Filter */
3179                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3180                  * and that breaks VLANs.
3181                  */
3182         }
3183
3184         ew32(RCTL, rctl);
3185         /* just started the receive unit, no need to restart */
3186         adapter->flags &= ~FLAG_RESTART_NOW;
3187 }
3188
3189 /**
3190  * e1000_configure_rx - Configure Receive Unit after Reset
3191  * @adapter: board private structure
3192  *
3193  * Configure the Rx unit of the MAC after a reset.
3194  **/
3195 static void e1000_configure_rx(struct e1000_adapter *adapter)
3196 {
3197         struct e1000_hw *hw = &adapter->hw;
3198         struct e1000_ring *rx_ring = adapter->rx_ring;
3199         u64 rdba;
3200         u32 rdlen, rctl, rxcsum, ctrl_ext;
3201
3202         if (adapter->rx_ps_pages) {
3203                 /* this is a 32 byte descriptor */
3204                 rdlen = rx_ring->count *
3205                     sizeof(union e1000_rx_desc_packet_split);
3206                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3207                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3208         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3209                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3210                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3211                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3212         } else {
3213                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3214                 adapter->clean_rx = e1000_clean_rx_irq;
3215                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3216         }
3217
3218         /* disable receives while setting up the descriptors */
3219         rctl = er32(RCTL);
3220         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3221                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3222         e1e_flush();
3223         usleep_range(10000, 11000);
3224
3225         if (adapter->flags2 & FLAG2_DMA_BURST) {
3226                 /* set the writeback threshold (only takes effect if the RDTR
3227                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3228                  * enable prefetching of 0x20 Rx descriptors
3229                  * granularity = 01
3230                  * wthresh = 04,
3231                  * hthresh = 04,
3232                  * pthresh = 0x20
3233                  */
3234                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3235                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3236         }
3237
3238         /* set the Receive Delay Timer Register */
3239         ew32(RDTR, adapter->rx_int_delay);
3240
3241         /* irq moderation */
3242         ew32(RADV, adapter->rx_abs_int_delay);
3243         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3244                 e1000e_write_itr(adapter, adapter->itr);
3245
3246         ctrl_ext = er32(CTRL_EXT);
3247         /* Auto-Mask interrupts upon ICR access */
3248         ctrl_ext |= E1000_CTRL_EXT_IAME;
3249         ew32(IAM, 0xffffffff);
3250         ew32(CTRL_EXT, ctrl_ext);
3251         e1e_flush();
3252
3253         /* Setup the HW Rx Head and Tail Descriptor Pointers and
3254          * the Base and Length of the Rx Descriptor Ring
3255          */
3256         rdba = rx_ring->dma;
3257         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3258         ew32(RDBAH(0), (rdba >> 32));
3259         ew32(RDLEN(0), rdlen);
3260         ew32(RDH(0), 0);
3261         ew32(RDT(0), 0);
3262         rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3263         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3264
3265         writel(0, rx_ring->head);
3266         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
3267                 e1000e_update_rdt_wa(rx_ring, 0);
3268         else
3269                 writel(0, rx_ring->tail);
3270
3271         /* Enable Receive Checksum Offload for TCP and UDP */
3272         rxcsum = er32(RXCSUM);
3273         if (adapter->netdev->features & NETIF_F_RXCSUM)
3274                 rxcsum |= E1000_RXCSUM_TUOFL;
3275         else
3276                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3277         ew32(RXCSUM, rxcsum);
3278
3279         /* With jumbo frames, excessive C-state transition latencies result
3280          * in dropped transactions.
3281          */
3282         if (adapter->netdev->mtu > ETH_DATA_LEN) {
3283                 u32 lat =
3284                     ((er32(PBA) & E1000_PBA_RXA_MASK) * 1024 -
3285                      adapter->max_frame_size) * 8 / 1000;
3286
3287                 if (adapter->flags & FLAG_IS_ICH) {
3288                         u32 rxdctl = er32(RXDCTL(0));
3289
3290                         ew32(RXDCTL(0), rxdctl | 0x3 | BIT(8));
3291                 }
3292
3293                 dev_info(&adapter->pdev->dev,
3294                          "Some CPU C-states have been disabled in order to enable jumbo frames\n");
3295                 cpu_latency_qos_update_request(&adapter->pm_qos_req, lat);
3296         } else {
3297                 cpu_latency_qos_update_request(&adapter->pm_qos_req,
3298                                                PM_QOS_DEFAULT_VALUE);
3299         }
3300
3301         /* Enable Receives */
3302         ew32(RCTL, rctl);
3303 }
3304
3305 /**
3306  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3307  * @netdev: network interface device structure
3308  *
3309  * Writes multicast address list to the MTA hash table.
3310  * Returns: -ENOMEM on failure
3311  *                0 on no addresses written
3312  *                X on writing X addresses to MTA
3313  */
3314 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3315 {
3316         struct e1000_adapter *adapter = netdev_priv(netdev);
3317         struct e1000_hw *hw = &adapter->hw;
3318         struct netdev_hw_addr *ha;
3319         u8 *mta_list;
3320         int i;
3321
3322         if (netdev_mc_empty(netdev)) {
3323                 /* nothing to program, so clear mc list */
3324                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3325                 return 0;
3326         }
3327
3328         mta_list = kcalloc(netdev_mc_count(netdev), ETH_ALEN, GFP_ATOMIC);
3329         if (!mta_list)
3330                 return -ENOMEM;
3331
3332         /* update_mc_addr_list expects a packed array of only addresses. */
3333         i = 0;
3334         netdev_for_each_mc_addr(ha, netdev)
3335             memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3336
3337         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3338         kfree(mta_list);
3339
3340         return netdev_mc_count(netdev);
3341 }
3342
3343 /**
3344  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3345  * @netdev: network interface device structure
3346  *
3347  * Writes unicast address list to the RAR table.
3348  * Returns: -ENOMEM on failure/insufficient address space
3349  *                0 on no addresses written
3350  *                X on writing X addresses to the RAR table
3351  **/
3352 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3353 {
3354         struct e1000_adapter *adapter = netdev_priv(netdev);
3355         struct e1000_hw *hw = &adapter->hw;
3356         unsigned int rar_entries;
3357         int count = 0;
3358
3359         rar_entries = hw->mac.ops.rar_get_count(hw);
3360
3361         /* save a rar entry for our hardware address */
3362         rar_entries--;
3363
3364         /* save a rar entry for the LAA workaround */
3365         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3366                 rar_entries--;
3367
3368         /* return ENOMEM indicating insufficient memory for addresses */
3369         if (netdev_uc_count(netdev) > rar_entries)
3370                 return -ENOMEM;
3371
3372         if (!netdev_uc_empty(netdev) && rar_entries) {
3373                 struct netdev_hw_addr *ha;
3374
3375                 /* write the addresses in reverse order to avoid write
3376                  * combining
3377                  */
3378                 netdev_for_each_uc_addr(ha, netdev) {
3379                         int ret_val;
3380
3381                         if (!rar_entries)
3382                                 break;
3383                         ret_val = hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3384                         if (ret_val < 0)
3385                                 return -ENOMEM;
3386                         count++;
3387                 }
3388         }
3389
3390         /* zero out the remaining RAR entries not used above */
3391         for (; rar_entries > 0; rar_entries--) {
3392                 ew32(RAH(rar_entries), 0);
3393                 ew32(RAL(rar_entries), 0);
3394         }
3395         e1e_flush();
3396
3397         return count;
3398 }
3399
3400 /**
3401  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3402  * @netdev: network interface device structure
3403  *
3404  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3405  * address list or the network interface flags are updated.  This routine is
3406  * responsible for configuring the hardware for proper unicast, multicast,
3407  * promiscuous mode, and all-multi behavior.
3408  **/
3409 static void e1000e_set_rx_mode(struct net_device *netdev)
3410 {
3411         struct e1000_adapter *adapter = netdev_priv(netdev);
3412         struct e1000_hw *hw = &adapter->hw;
3413         u32 rctl;
3414
3415         if (pm_runtime_suspended(netdev->dev.parent))
3416                 return;
3417
3418         /* Check for Promiscuous and All Multicast modes */
3419         rctl = er32(RCTL);
3420
3421         /* clear the affected bits */
3422         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3423
3424         if (netdev->flags & IFF_PROMISC) {
3425                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3426                 /* Do not hardware filter VLANs in promisc mode */
3427                 e1000e_vlan_filter_disable(adapter);
3428         } else {
3429                 int count;
3430
3431                 if (netdev->flags & IFF_ALLMULTI) {
3432                         rctl |= E1000_RCTL_MPE;
3433                 } else {
3434                         /* Write addresses to the MTA, if the attempt fails
3435                          * then we should just turn on promiscuous mode so
3436                          * that we can at least receive multicast traffic
3437                          */
3438                         count = e1000e_write_mc_addr_list(netdev);
3439                         if (count < 0)
3440                                 rctl |= E1000_RCTL_MPE;
3441                 }
3442                 e1000e_vlan_filter_enable(adapter);
3443                 /* Write addresses to available RAR registers, if there is not
3444                  * sufficient space to store all the addresses then enable
3445                  * unicast promiscuous mode
3446                  */
3447                 count = e1000e_write_uc_addr_list(netdev);
3448                 if (count < 0)
3449                         rctl |= E1000_RCTL_UPE;
3450         }
3451
3452         ew32(RCTL, rctl);
3453
3454         if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3455                 e1000e_vlan_strip_enable(adapter);
3456         else
3457                 e1000e_vlan_strip_disable(adapter);
3458 }
3459
3460 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3461 {
3462         struct e1000_hw *hw = &adapter->hw;
3463         u32 mrqc, rxcsum;
3464         u32 rss_key[10];
3465         int i;
3466
3467         netdev_rss_key_fill(rss_key, sizeof(rss_key));
3468         for (i = 0; i < 10; i++)
3469                 ew32(RSSRK(i), rss_key[i]);
3470
3471         /* Direct all traffic to queue 0 */
3472         for (i = 0; i < 32; i++)
3473                 ew32(RETA(i), 0);
3474
3475         /* Disable raw packet checksumming so that RSS hash is placed in
3476          * descriptor on writeback.
3477          */
3478         rxcsum = er32(RXCSUM);
3479         rxcsum |= E1000_RXCSUM_PCSD;
3480
3481         ew32(RXCSUM, rxcsum);
3482
3483         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3484                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3485                 E1000_MRQC_RSS_FIELD_IPV6 |
3486                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3487                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3488
3489         ew32(MRQC, mrqc);
3490 }
3491
3492 /**
3493  * e1000e_get_base_timinca - get default SYSTIM time increment attributes
3494  * @adapter: board private structure
3495  * @timinca: pointer to returned time increment attributes
3496  *
3497  * Get attributes for incrementing the System Time Register SYSTIML/H at
3498  * the default base frequency, and set the cyclecounter shift value.
3499  **/
3500 s32 e1000e_get_base_timinca(struct e1000_adapter *adapter, u32 *timinca)
3501 {
3502         struct e1000_hw *hw = &adapter->hw;
3503         u32 incvalue, incperiod, shift;
3504
3505         /* Make sure clock is enabled on I217/I218/I219  before checking
3506          * the frequency
3507          */
3508         if ((hw->mac.type >= e1000_pch_lpt) &&
3509             !(er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) &&
3510             !(er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_ENABLED)) {
3511                 u32 fextnvm7 = er32(FEXTNVM7);
3512
3513                 if (!(fextnvm7 & BIT(0))) {
3514                         ew32(FEXTNVM7, fextnvm7 | BIT(0));
3515                         e1e_flush();
3516                 }
3517         }
3518
3519         switch (hw->mac.type) {
3520         case e1000_pch2lan:
3521                 /* Stable 96MHz frequency */
3522                 incperiod = INCPERIOD_96MHZ;
3523                 incvalue = INCVALUE_96MHZ;
3524                 shift = INCVALUE_SHIFT_96MHZ;
3525                 adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3526                 break;
3527         case e1000_pch_lpt:
3528                 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3529                         /* Stable 96MHz frequency */
3530                         incperiod = INCPERIOD_96MHZ;
3531                         incvalue = INCVALUE_96MHZ;
3532                         shift = INCVALUE_SHIFT_96MHZ;
3533                         adapter->cc.shift = shift + INCPERIOD_SHIFT_96MHZ;
3534                 } else {
3535                         /* Stable 25MHz frequency */
3536                         incperiod = INCPERIOD_25MHZ;
3537                         incvalue = INCVALUE_25MHZ;
3538                         shift = INCVALUE_SHIFT_25MHZ;
3539                         adapter->cc.shift = shift;
3540                 }
3541                 break;
3542         case e1000_pch_spt:
3543                 /* Stable 24MHz frequency */
3544                 incperiod = INCPERIOD_24MHZ;
3545                 incvalue = INCVALUE_24MHZ;
3546                 shift = INCVALUE_SHIFT_24MHZ;
3547                 adapter->cc.shift = shift;
3548                 break;
3549         case e1000_pch_cnp:
3550         case e1000_pch_tgp:
3551         case e1000_pch_adp:
3552         case e1000_pch_mtp:
3553                 if (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_SYSCFI) {
3554                         /* Stable 24MHz frequency */
3555                         incperiod = INCPERIOD_24MHZ;
3556                         incvalue = INCVALUE_24MHZ;
3557                         shift = INCVALUE_SHIFT_24MHZ;
3558                         adapter->cc.shift = shift;
3559                 } else {
3560                         /* Stable 38400KHz frequency */
3561                         incperiod = INCPERIOD_38400KHZ;
3562                         incvalue = INCVALUE_38400KHZ;
3563                         shift = INCVALUE_SHIFT_38400KHZ;
3564                         adapter->cc.shift = shift;
3565                 }
3566                 break;
3567         case e1000_82574:
3568         case e1000_82583:
3569                 /* Stable 25MHz frequency */
3570                 incperiod = INCPERIOD_25MHZ;
3571                 incvalue = INCVALUE_25MHZ;
3572                 shift = INCVALUE_SHIFT_25MHZ;
3573                 adapter->cc.shift = shift;
3574                 break;
3575         default:
3576                 return -EINVAL;
3577         }
3578
3579         *timinca = ((incperiod << E1000_TIMINCA_INCPERIOD_SHIFT) |
3580                     ((incvalue << shift) & E1000_TIMINCA_INCVALUE_MASK));
3581
3582         return 0;
3583 }
3584
3585 /**
3586  * e1000e_config_hwtstamp - configure the hwtstamp registers and enable/disable
3587  * @adapter: board private structure
3588  * @config: timestamp configuration
3589  *
3590  * Outgoing time stamping can be enabled and disabled. Play nice and
3591  * disable it when requested, although it shouldn't cause any overhead
3592  * when no packet needs it. At most one packet in the queue may be
3593  * marked for time stamping, otherwise it would be impossible to tell
3594  * for sure to which packet the hardware time stamp belongs.
3595  *
3596  * Incoming time stamping has to be configured via the hardware filters.
3597  * Not all combinations are supported, in particular event type has to be
3598  * specified. Matching the kind of event packet is not supported, with the
3599  * exception of "all V2 events regardless of level 2 or 4".
3600  **/
3601 static int e1000e_config_hwtstamp(struct e1000_adapter *adapter,
3602                                   struct hwtstamp_config *config)
3603 {
3604         struct e1000_hw *hw = &adapter->hw;
3605         u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
3606         u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
3607         u32 rxmtrl = 0;
3608         u16 rxudp = 0;
3609         bool is_l4 = false;
3610         bool is_l2 = false;
3611         u32 regval;
3612
3613         if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3614                 return -EINVAL;
3615
3616         /* flags reserved for future extensions - must be zero */
3617         if (config->flags)
3618                 return -EINVAL;
3619
3620         switch (config->tx_type) {
3621         case HWTSTAMP_TX_OFF:
3622                 tsync_tx_ctl = 0;
3623                 break;
3624         case HWTSTAMP_TX_ON:
3625                 break;
3626         default:
3627                 return -ERANGE;
3628         }
3629
3630         switch (config->rx_filter) {
3631         case HWTSTAMP_FILTER_NONE:
3632                 tsync_rx_ctl = 0;
3633                 break;
3634         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3635                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3636                 rxmtrl = E1000_RXMTRL_PTP_V1_SYNC_MESSAGE;
3637                 is_l4 = true;
3638                 break;
3639         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3640                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
3641                 rxmtrl = E1000_RXMTRL_PTP_V1_DELAY_REQ_MESSAGE;
3642                 is_l4 = true;
3643                 break;
3644         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3645                 /* Also time stamps V2 L2 Path Delay Request/Response */
3646                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3647                 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3648                 is_l2 = true;
3649                 break;
3650         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3651                 /* Also time stamps V2 L2 Path Delay Request/Response. */
3652                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_V2;
3653                 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3654                 is_l2 = true;
3655                 break;
3656         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3657                 /* Hardware cannot filter just V2 L4 Sync messages */
3658                 fallthrough;
3659         case HWTSTAMP_FILTER_PTP_V2_SYNC:
3660                 /* Also time stamps V2 Path Delay Request/Response. */
3661                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3662                 rxmtrl = E1000_RXMTRL_PTP_V2_SYNC_MESSAGE;
3663                 is_l2 = true;
3664                 is_l4 = true;
3665                 break;
3666         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3667                 /* Hardware cannot filter just V2 L4 Delay Request messages */
3668                 fallthrough;
3669         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3670                 /* Also time stamps V2 Path Delay Request/Response. */
3671                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
3672                 rxmtrl = E1000_RXMTRL_PTP_V2_DELAY_REQ_MESSAGE;
3673                 is_l2 = true;
3674                 is_l4 = true;
3675                 break;
3676         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3677         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3678                 /* Hardware cannot filter just V2 L4 or L2 Event messages */
3679                 fallthrough;
3680         case HWTSTAMP_FILTER_PTP_V2_EVENT:
3681                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
3682                 config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
3683                 is_l2 = true;
3684                 is_l4 = true;
3685                 break;
3686         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3687                 /* For V1, the hardware can only filter Sync messages or
3688                  * Delay Request messages but not both so fall-through to
3689                  * time stamp all packets.
3690                  */
3691                 fallthrough;
3692         case HWTSTAMP_FILTER_NTP_ALL:
3693         case HWTSTAMP_FILTER_ALL:
3694                 is_l2 = true;
3695                 is_l4 = true;
3696                 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
3697                 config->rx_filter = HWTSTAMP_FILTER_ALL;
3698                 break;
3699         default:
3700                 return -ERANGE;
3701         }
3702
3703         adapter->hwtstamp_config = *config;
3704
3705         /* enable/disable Tx h/w time stamping */
3706         regval = er32(TSYNCTXCTL);
3707         regval &= ~E1000_TSYNCTXCTL_ENABLED;
3708         regval |= tsync_tx_ctl;
3709         ew32(TSYNCTXCTL, regval);
3710         if ((er32(TSYNCTXCTL) & E1000_TSYNCTXCTL_ENABLED) !=
3711             (regval & E1000_TSYNCTXCTL_ENABLED)) {
3712                 e_err("Timesync Tx Control register not set as expected\n");
3713                 return -EAGAIN;
3714         }
3715
3716         /* enable/disable Rx h/w time stamping */
3717         regval = er32(TSYNCRXCTL);
3718         regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
3719         regval |= tsync_rx_ctl;
3720         ew32(TSYNCRXCTL, regval);
3721         if ((er32(TSYNCRXCTL) & (E1000_TSYNCRXCTL_ENABLED |
3722                                  E1000_TSYNCRXCTL_TYPE_MASK)) !=
3723             (regval & (E1000_TSYNCRXCTL_ENABLED |
3724                        E1000_TSYNCRXCTL_TYPE_MASK))) {
3725                 e_err("Timesync Rx Control register not set as expected\n");
3726                 return -EAGAIN;
3727         }
3728
3729         /* L2: define ethertype filter for time stamped packets */
3730         if (is_l2)
3731                 rxmtrl |= ETH_P_1588;
3732
3733         /* define which PTP packets get time stamped */
3734         ew32(RXMTRL, rxmtrl);
3735
3736         /* Filter by destination port */
3737         if (is_l4) {
3738                 rxudp = PTP_EV_PORT;
3739                 cpu_to_be16s(&rxudp);
3740         }
3741         ew32(RXUDP, rxudp);
3742
3743         e1e_flush();
3744
3745         /* Clear TSYNCRXCTL_VALID & TSYNCTXCTL_VALID bit */
3746         er32(RXSTMPH);
3747         er32(TXSTMPH);
3748
3749         return 0;
3750 }
3751
3752 /**
3753  * e1000_configure - configure the hardware for Rx and Tx
3754  * @adapter: private board structure
3755  **/
3756 static void e1000_configure(struct e1000_adapter *adapter)
3757 {
3758         struct e1000_ring *rx_ring = adapter->rx_ring;
3759
3760         e1000e_set_rx_mode(adapter->netdev);
3761
3762         e1000_restore_vlan(adapter);
3763         e1000_init_manageability_pt(adapter);
3764
3765         e1000_configure_tx(adapter);
3766
3767         if (adapter->netdev->features & NETIF_F_RXHASH)
3768                 e1000e_setup_rss_hash(adapter);
3769         e1000_setup_rctl(adapter);
3770         e1000_configure_rx(adapter);
3771         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3772 }
3773
3774 /**
3775  * e1000e_power_up_phy - restore link in case the phy was powered down
3776  * @adapter: address of board private structure
3777  *
3778  * The phy may be powered down to save power and turn off link when the
3779  * driver is unloaded and wake on lan is not enabled (among others)
3780  * *** this routine MUST be followed by a call to e1000e_reset ***
3781  **/
3782 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3783 {
3784         if (adapter->hw.phy.ops.power_up)
3785                 adapter->hw.phy.ops.power_up(&adapter->hw);
3786
3787         adapter->hw.mac.ops.setup_link(&adapter->hw);
3788 }
3789
3790 /**
3791  * e1000_power_down_phy - Power down the PHY
3792  * @adapter: board private structure
3793  *
3794  * Power down the PHY so no link is implied when interface is down.
3795  * The PHY cannot be powered down if management or WoL is active.
3796  */
3797 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3798 {
3799         if (adapter->hw.phy.ops.power_down)
3800                 adapter->hw.phy.ops.power_down(&adapter->hw);
3801 }
3802
3803 /**
3804  * e1000_flush_tx_ring - remove all descriptors from the tx_ring
3805  * @adapter: board private structure
3806  *
3807  * We want to clear all pending descriptors from the TX ring.
3808  * zeroing happens when the HW reads the regs. We  assign the ring itself as
3809  * the data of the next descriptor. We don't care about the data we are about
3810  * to reset the HW.
3811  */
3812 static void e1000_flush_tx_ring(struct e1000_adapter *adapter)
3813 {
3814         struct e1000_hw *hw = &adapter->hw;
3815         struct e1000_ring *tx_ring = adapter->tx_ring;
3816         struct e1000_tx_desc *tx_desc = NULL;
3817         u32 tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
3818         u16 size = 512;
3819
3820         tctl = er32(TCTL);
3821         ew32(TCTL, tctl | E1000_TCTL_EN);
3822         tdt = er32(TDT(0));
3823         BUG_ON(tdt != tx_ring->next_to_use);
3824         tx_desc =  E1000_TX_DESC(*tx_ring, tx_ring->next_to_use);
3825         tx_desc->buffer_addr = cpu_to_le64(tx_ring->dma);
3826
3827         tx_desc->lower.data = cpu_to_le32(txd_lower | size);
3828         tx_desc->upper.data = 0;
3829         /* flush descriptors to memory before notifying the HW */
3830         wmb();
3831         tx_ring->next_to_use++;
3832         if (tx_ring->next_to_use == tx_ring->count)
3833                 tx_ring->next_to_use = 0;
3834         ew32(TDT(0), tx_ring->next_to_use);
3835         usleep_range(200, 250);
3836 }
3837
3838 /**
3839  * e1000_flush_rx_ring - remove all descriptors from the rx_ring
3840  * @adapter: board private structure
3841  *
3842  * Mark all descriptors in the RX ring as consumed and disable the rx ring
3843  */
3844 static void e1000_flush_rx_ring(struct e1000_adapter *adapter)
3845 {
3846         u32 rctl, rxdctl;
3847         struct e1000_hw *hw = &adapter->hw;
3848
3849         rctl = er32(RCTL);
3850         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3851         e1e_flush();
3852         usleep_range(100, 150);
3853
3854         rxdctl = er32(RXDCTL(0));
3855         /* zero the lower 14 bits (prefetch and host thresholds) */
3856         rxdctl &= 0xffffc000;
3857
3858         /* update thresholds: prefetch threshold to 31, host threshold to 1
3859          * and make sure the granularity is "descriptors" and not "cache lines"
3860          */
3861         rxdctl |= (0x1F | BIT(8) | E1000_RXDCTL_THRESH_UNIT_DESC);
3862
3863         ew32(RXDCTL(0), rxdctl);
3864         /* momentarily enable the RX ring for the changes to take effect */
3865         ew32(RCTL, rctl | E1000_RCTL_EN);
3866         e1e_flush();
3867         usleep_range(100, 150);
3868         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3869 }
3870
3871 /**
3872  * e1000_flush_desc_rings - remove all descriptors from the descriptor rings
3873  * @adapter: board private structure
3874  *
3875  * In i219, the descriptor rings must be emptied before resetting the HW
3876  * or before changing the device state to D3 during runtime (runtime PM).
3877  *
3878  * Failure to do this will cause the HW to enter a unit hang state which can
3879  * only be released by PCI reset on the device
3880  *
3881  */
3882
3883 static void e1000_flush_desc_rings(struct e1000_adapter *adapter)
3884 {
3885         u16 hang_state;
3886         u32 fext_nvm11, tdlen;
3887         struct e1000_hw *hw = &adapter->hw;
3888
3889         /* First, disable MULR fix in FEXTNVM11 */
3890         fext_nvm11 = er32(FEXTNVM11);
3891         fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
3892         ew32(FEXTNVM11, fext_nvm11);
3893         /* do nothing if we're not in faulty state, or if the queue is empty */
3894         tdlen = er32(TDLEN(0));
3895         pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3896                              &hang_state);
3897         if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
3898                 return;
3899         e1000_flush_tx_ring(adapter);
3900         /* recheck, maybe the fault is caused by the rx ring */
3901         pci_read_config_word(adapter->pdev, PCICFG_DESC_RING_STATUS,
3902                              &hang_state);
3903         if (hang_state & FLUSH_DESC_REQUIRED)
3904                 e1000_flush_rx_ring(adapter);
3905 }
3906
3907 /**
3908  * e1000e_systim_reset - reset the timesync registers after a hardware reset
3909  * @adapter: board private structure
3910  *
3911  * When the MAC is reset, all hardware bits for timesync will be reset to the
3912  * default values. This function will restore the settings last in place.
3913  * Since the clock SYSTIME registers are reset, we will simply restore the
3914  * cyclecounter to the kernel real clock time.
3915  **/
3916 static void e1000e_systim_reset(struct e1000_adapter *adapter)
3917 {
3918         struct ptp_clock_info *info = &adapter->ptp_clock_info;
3919         struct e1000_hw *hw = &adapter->hw;
3920         unsigned long flags;
3921         u32 timinca;
3922         s32 ret_val;
3923
3924         if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
3925                 return;
3926
3927         if (info->adjfreq) {
3928                 /* restore the previous ptp frequency delta */
3929                 ret_val = info->adjfreq(info, adapter->ptp_delta);
3930         } else {
3931                 /* set the default base frequency if no adjustment possible */
3932                 ret_val = e1000e_get_base_timinca(adapter, &timinca);
3933                 if (!ret_val)
3934                         ew32(TIMINCA, timinca);
3935         }
3936
3937         if (ret_val) {
3938                 dev_warn(&adapter->pdev->dev,
3939                          "Failed to restore TIMINCA clock rate delta: %d\n",
3940                          ret_val);
3941                 return;
3942         }
3943
3944         /* reset the systim ns time counter */
3945         spin_lock_irqsave(&adapter->systim_lock, flags);
3946         timecounter_init(&adapter->tc, &adapter->cc,
3947                          ktime_to_ns(ktime_get_real()));
3948         spin_unlock_irqrestore(&adapter->systim_lock, flags);
3949
3950         /* restore the previous hwtstamp configuration settings */
3951         e1000e_config_hwtstamp(adapter, &adapter->hwtstamp_config);
3952 }
3953
3954 /**
3955  * e1000e_reset - bring the hardware into a known good state
3956  * @adapter: board private structure
3957  *
3958  * This function boots the hardware and enables some settings that
3959  * require a configuration cycle of the hardware - those cannot be
3960  * set/changed during runtime. After reset the device needs to be
3961  * properly configured for Rx, Tx etc.
3962  */
3963 void e1000e_reset(struct e1000_adapter *adapter)
3964 {
3965         struct e1000_mac_info *mac = &adapter->hw.mac;
3966         struct e1000_fc_info *fc = &adapter->hw.fc;
3967         struct e1000_hw *hw = &adapter->hw;
3968         u32 tx_space, min_tx_space, min_rx_space;
3969         u32 pba = adapter->pba;
3970         u16 hwm;
3971
3972         /* reset Packet Buffer Allocation to default */
3973         ew32(PBA, pba);
3974
3975         if (adapter->max_frame_size > (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN)) {
3976                 /* To maintain wire speed transmits, the Tx FIFO should be
3977                  * large enough to accommodate two full transmit packets,
3978                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3979                  * the Rx FIFO should be large enough to accommodate at least
3980                  * one full receive packet and is similarly rounded up and
3981                  * expressed in KB.
3982                  */
3983                 pba = er32(PBA);
3984                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3985                 tx_space = pba >> 16;
3986                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3987                 pba &= 0xffff;
3988                 /* the Tx fifo also stores 16 bytes of information about the Tx
3989                  * but don't include ethernet FCS because hardware appends it
3990                  */
3991                 min_tx_space = (adapter->max_frame_size +
3992                                 sizeof(struct e1000_tx_desc) - ETH_FCS_LEN) * 2;
3993                 min_tx_space = ALIGN(min_tx_space, 1024);
3994                 min_tx_space >>= 10;
3995                 /* software strips receive CRC, so leave room for it */
3996                 min_rx_space = adapter->max_frame_size;
3997                 min_rx_space = ALIGN(min_rx_space, 1024);
3998                 min_rx_space >>= 10;
3999
4000                 /* If current Tx allocation is less than the min Tx FIFO size,
4001                  * and the min Tx FIFO size is less than the current Rx FIFO
4002                  * allocation, take space away from current Rx allocation
4003                  */
4004                 if ((tx_space < min_tx_space) &&
4005                     ((min_tx_space - tx_space) < pba)) {
4006                         pba -= min_tx_space - tx_space;
4007
4008                         /* if short on Rx space, Rx wins and must trump Tx
4009                          * adjustment
4010                          */
4011                         if (pba < min_rx_space)
4012                                 pba = min_rx_space;
4013                 }
4014
4015                 ew32(PBA, pba);
4016         }
4017
4018         /* flow control settings
4019          *
4020          * The high water mark must be low enough to fit one full frame
4021          * (or the size used for early receive) above it in the Rx FIFO.
4022          * Set it to the lower of:
4023          * - 90% of the Rx FIFO size, and
4024          * - the full Rx FIFO size minus one full frame
4025          */
4026         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
4027                 fc->pause_time = 0xFFFF;
4028         else
4029                 fc->pause_time = E1000_FC_PAUSE_TIME;
4030         fc->send_xon = true;
4031         fc->current_mode = fc->requested_mode;
4032
4033         switch (hw->mac.type) {
4034         case e1000_ich9lan:
4035         case e1000_ich10lan:
4036                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4037                         pba = 14;
4038                         ew32(PBA, pba);
4039                         fc->high_water = 0x2800;
4040                         fc->low_water = fc->high_water - 8;
4041                         break;
4042                 }
4043                 fallthrough;
4044         default:
4045                 hwm = min(((pba << 10) * 9 / 10),
4046                           ((pba << 10) - adapter->max_frame_size));
4047
4048                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
4049                 fc->low_water = fc->high_water - 8;
4050                 break;
4051         case e1000_pchlan:
4052                 /* Workaround PCH LOM adapter hangs with certain network
4053                  * loads.  If hangs persist, try disabling Tx flow control.
4054                  */
4055                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
4056                         fc->high_water = 0x3500;
4057                         fc->low_water = 0x1500;
4058                 } else {
4059                         fc->high_water = 0x5000;
4060                         fc->low_water = 0x3000;
4061                 }
4062                 fc->refresh_time = 0x1000;
4063                 break;
4064         case e1000_pch2lan:
4065         case e1000_pch_lpt:
4066         case e1000_pch_spt:
4067         case e1000_pch_cnp:
4068         case e1000_pch_tgp:
4069         case e1000_pch_adp:
4070         case e1000_pch_mtp:
4071                 fc->refresh_time = 0xFFFF;
4072                 fc->pause_time = 0xFFFF;
4073
4074                 if (adapter->netdev->mtu <= ETH_DATA_LEN) {
4075                         fc->high_water = 0x05C20;
4076                         fc->low_water = 0x05048;
4077                         break;
4078                 }
4079
4080                 pba = 14;
4081                 ew32(PBA, pba);
4082                 fc->high_water = ((pba << 10) * 9 / 10) & E1000_FCRTH_RTH;
4083                 fc->low_water = ((pba << 10) * 8 / 10) & E1000_FCRTL_RTL;
4084                 break;
4085         }
4086
4087         /* Alignment of Tx data is on an arbitrary byte boundary with the
4088          * maximum size per Tx descriptor limited only to the transmit
4089          * allocation of the packet buffer minus 96 bytes with an upper
4090          * limit of 24KB due to receive synchronization limitations.
4091          */
4092         adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
4093                                        24 << 10);
4094
4095         /* Disable Adaptive Interrupt Moderation if 2 full packets cannot
4096          * fit in receive buffer.
4097          */
4098         if (adapter->itr_setting & 0x3) {
4099                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
4100                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
4101                                 dev_info(&adapter->pdev->dev,
4102                                          "Interrupt Throttle Rate off\n");
4103                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
4104                                 e1000e_write_itr(adapter, 0);
4105                         }
4106                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
4107                         dev_info(&adapter->pdev->dev,
4108                                  "Interrupt Throttle Rate on\n");
4109                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
4110                         adapter->itr = 20000;
4111                         e1000e_write_itr(adapter, adapter->itr);
4112                 }
4113         }
4114
4115         if (hw->mac.type >= e1000_pch_spt)
4116                 e1000_flush_desc_rings(adapter);
4117         /* Allow time for pending master requests to run */
4118         mac->ops.reset_hw(hw);
4119
4120         /* For parts with AMT enabled, let the firmware know
4121          * that the network interface is in control
4122          */
4123         if (adapter->flags & FLAG_HAS_AMT)
4124                 e1000e_get_hw_control(adapter);
4125
4126         ew32(WUC, 0);
4127
4128         if (mac->ops.init_hw(hw))
4129                 e_err("Hardware Error\n");
4130
4131         e1000_update_mng_vlan(adapter);
4132
4133         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
4134         ew32(VET, ETH_P_8021Q);
4135
4136         e1000e_reset_adaptive(hw);
4137
4138         /* restore systim and hwtstamp settings */
4139         e1000e_systim_reset(adapter);
4140
4141         /* Set EEE advertisement as appropriate */
4142         if (adapter->flags2 & FLAG2_HAS_EEE) {
4143                 s32 ret_val;
4144                 u16 adv_addr;
4145
4146                 switch (hw->phy.type) {
4147                 case e1000_phy_82579:
4148                         adv_addr = I82579_EEE_ADVERTISEMENT;
4149                         break;
4150                 case e1000_phy_i217:
4151                         adv_addr = I217_EEE_ADVERTISEMENT;
4152                         break;
4153                 default:
4154                         dev_err(&adapter->pdev->dev,
4155                                 "Invalid PHY type setting EEE advertisement\n");
4156                         return;
4157                 }
4158
4159                 ret_val = hw->phy.ops.acquire(hw);
4160                 if (ret_val) {
4161                         dev_err(&adapter->pdev->dev,
4162                                 "EEE advertisement - unable to acquire PHY\n");
4163                         return;
4164                 }
4165
4166                 e1000_write_emi_reg_locked(hw, adv_addr,
4167                                            hw->dev_spec.ich8lan.eee_disable ?
4168                                            0 : adapter->eee_advert);
4169
4170                 hw->phy.ops.release(hw);
4171         }
4172
4173         if (!netif_running(adapter->netdev) &&
4174             !test_bit(__E1000_TESTING, &adapter->state))
4175                 e1000_power_down_phy(adapter);
4176
4177         e1000_get_phy_info(hw);
4178
4179         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
4180             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
4181                 u16 phy_data = 0;
4182                 /* speed up time to link by disabling smart power down, ignore
4183                  * the return value of this function because there is nothing
4184                  * different we would do if it failed
4185                  */
4186                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
4187                 phy_data &= ~IGP02E1000_PM_SPD;
4188                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
4189         }
4190         if (hw->mac.type >= e1000_pch_spt && adapter->int_mode == 0) {
4191                 u32 reg;
4192
4193                 /* Fextnvm7 @ 0xe4[2] = 1 */
4194                 reg = er32(FEXTNVM7);
4195                 reg |= E1000_FEXTNVM7_SIDE_CLK_UNGATE;
4196                 ew32(FEXTNVM7, reg);
4197                 /* Fextnvm9 @ 0x5bb4[13:12] = 11 */
4198                 reg = er32(FEXTNVM9);
4199                 reg |= E1000_FEXTNVM9_IOSFSB_CLKGATE_DIS |
4200                        E1000_FEXTNVM9_IOSFSB_CLKREQ_DIS;
4201                 ew32(FEXTNVM9, reg);
4202         }
4203
4204 }
4205
4206 /**
4207  * e1000e_trigger_lsc - trigger an LSC interrupt
4208  * @adapter:
4209  *
4210  * Fire a link status change interrupt to start the watchdog.
4211  **/
4212 static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
4213 {
4214         struct e1000_hw *hw = &adapter->hw;
4215
4216         if (adapter->msix_entries)
4217                 ew32(ICS, E1000_ICS_LSC | E1000_ICS_OTHER);
4218         else
4219                 ew32(ICS, E1000_ICS_LSC);
4220 }
4221
4222 void e1000e_up(struct e1000_adapter *adapter)
4223 {
4224         /* hardware has been reset, we need to reload some things */
4225         e1000_configure(adapter);
4226
4227         clear_bit(__E1000_DOWN, &adapter->state);
4228
4229         if (adapter->msix_entries)
4230                 e1000_configure_msix(adapter);
4231         e1000_irq_enable(adapter);
4232
4233         /* Tx queue started by watchdog timer when link is up */
4234
4235         e1000e_trigger_lsc(adapter);
4236 }
4237
4238 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
4239 {
4240         struct e1000_hw *hw = &adapter->hw;
4241
4242         if (!(adapter->flags2 & FLAG2_DMA_BURST))
4243                 return;
4244
4245         /* flush pending descriptor writebacks to memory */
4246         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4247         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4248
4249         /* execute the writes immediately */
4250         e1e_flush();
4251
4252         /* due to rare timing issues, write to TIDV/RDTR again to ensure the
4253          * write is successful
4254          */
4255         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4256         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4257
4258         /* execute the writes immediately */
4259         e1e_flush();
4260 }
4261
4262 static void e1000e_update_stats(struct e1000_adapter *adapter);
4263
4264 /**
4265  * e1000e_down - quiesce the device and optionally reset the hardware
4266  * @adapter: board private structure
4267  * @reset: boolean flag to reset the hardware or not
4268  */
4269 void e1000e_down(struct e1000_adapter *adapter, bool reset)
4270 {
4271         struct net_device *netdev = adapter->netdev;
4272         struct e1000_hw *hw = &adapter->hw;
4273         u32 tctl, rctl;
4274
4275         /* signal that we're down so the interrupt handler does not
4276          * reschedule our watchdog timer
4277          */
4278         set_bit(__E1000_DOWN, &adapter->state);
4279
4280         netif_carrier_off(netdev);
4281
4282         /* disable receives in the hardware */
4283         rctl = er32(RCTL);
4284         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
4285                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
4286         /* flush and sleep below */
4287
4288         netif_stop_queue(netdev);
4289
4290         /* disable transmits in the hardware */
4291         tctl = er32(TCTL);
4292         tctl &= ~E1000_TCTL_EN;
4293         ew32(TCTL, tctl);
4294
4295         /* flush both disables and wait for them to finish */
4296         e1e_flush();
4297         usleep_range(10000, 11000);
4298
4299         e1000_irq_disable(adapter);
4300
4301         napi_synchronize(&adapter->napi);
4302
4303         del_timer_sync(&adapter->watchdog_timer);
4304         del_timer_sync(&adapter->phy_info_timer);
4305
4306         spin_lock(&adapter->stats64_lock);
4307         e1000e_update_stats(adapter);
4308         spin_unlock(&adapter->stats64_lock);
4309
4310         e1000e_flush_descriptors(adapter);
4311
4312         adapter->link_speed = 0;
4313         adapter->link_duplex = 0;
4314
4315         /* Disable Si errata workaround on PCHx for jumbo frame flow */
4316         if ((hw->mac.type >= e1000_pch2lan) &&
4317             (adapter->netdev->mtu > ETH_DATA_LEN) &&
4318             e1000_lv_jumbo_workaround_ich8lan(hw, false))
4319                 e_dbg("failed to disable jumbo frame workaround mode\n");
4320
4321         if (!pci_channel_offline(adapter->pdev)) {
4322                 if (reset)
4323                         e1000e_reset(adapter);
4324                 else if (hw->mac.type >= e1000_pch_spt)
4325                         e1000_flush_desc_rings(adapter);
4326         }
4327         e1000_clean_tx_ring(adapter->tx_ring);
4328         e1000_clean_rx_ring(adapter->rx_ring);
4329 }
4330
4331 void e1000e_reinit_locked(struct e1000_adapter *adapter)
4332 {
4333         might_sleep();
4334         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4335                 usleep_range(1000, 1100);
4336         e1000e_down(adapter, true);
4337         e1000e_up(adapter);
4338         clear_bit(__E1000_RESETTING, &adapter->state);
4339 }
4340
4341 /**
4342  * e1000e_sanitize_systim - sanitize raw cycle counter reads
4343  * @hw: pointer to the HW structure
4344  * @systim: PHC time value read, sanitized and returned
4345  * @sts: structure to hold system time before and after reading SYSTIML,
4346  * may be NULL
4347  *
4348  * Errata for 82574/82583 possible bad bits read from SYSTIMH/L:
4349  * check to see that the time is incrementing at a reasonable
4350  * rate and is a multiple of incvalue.
4351  **/
4352 static u64 e1000e_sanitize_systim(struct e1000_hw *hw, u64 systim,
4353                                   struct ptp_system_timestamp *sts)
4354 {
4355         u64 time_delta, rem, temp;
4356         u64 systim_next;
4357         u32 incvalue;
4358         int i;
4359
4360         incvalue = er32(TIMINCA) & E1000_TIMINCA_INCVALUE_MASK;
4361         for (i = 0; i < E1000_MAX_82574_SYSTIM_REREADS; i++) {
4362                 /* latch SYSTIMH on read of SYSTIML */
4363                 ptp_read_system_prets(sts);
4364                 systim_next = (u64)er32(SYSTIML);
4365                 ptp_read_system_postts(sts);
4366                 systim_next |= (u64)er32(SYSTIMH) << 32;
4367
4368                 time_delta = systim_next - systim;
4369                 temp = time_delta;
4370                 /* VMWare users have seen incvalue of zero, don't div / 0 */
4371                 rem = incvalue ? do_div(temp, incvalue) : (time_delta != 0);
4372
4373                 systim = systim_next;
4374
4375                 if ((time_delta < E1000_82574_SYSTIM_EPSILON) && (rem == 0))
4376                         break;
4377         }
4378
4379         return systim;
4380 }
4381
4382 /**
4383  * e1000e_read_systim - read SYSTIM register
4384  * @adapter: board private structure
4385  * @sts: structure which will contain system time before and after reading
4386  * SYSTIML, may be NULL
4387  **/
4388 u64 e1000e_read_systim(struct e1000_adapter *adapter,
4389                        struct ptp_system_timestamp *sts)
4390 {
4391         struct e1000_hw *hw = &adapter->hw;
4392         u32 systimel, systimel_2, systimeh;
4393         u64 systim;
4394         /* SYSTIMH latching upon SYSTIML read does not work well.
4395          * This means that if SYSTIML overflows after we read it but before
4396          * we read SYSTIMH, the value of SYSTIMH has been incremented and we
4397          * will experience a huge non linear increment in the systime value
4398          * to fix that we test for overflow and if true, we re-read systime.
4399          */
4400         ptp_read_system_prets(sts);
4401         systimel = er32(SYSTIML);
4402         ptp_read_system_postts(sts);
4403         systimeh = er32(SYSTIMH);
4404         /* Is systimel is so large that overflow is possible? */
4405         if (systimel >= (u32)0xffffffff - E1000_TIMINCA_INCVALUE_MASK) {
4406                 ptp_read_system_prets(sts);
4407                 systimel_2 = er32(SYSTIML);
4408                 ptp_read_system_postts(sts);
4409                 if (systimel > systimel_2) {
4410                         /* There was an overflow, read again SYSTIMH, and use
4411                          * systimel_2
4412                          */
4413                         systimeh = er32(SYSTIMH);
4414                         systimel = systimel_2;
4415                 }
4416         }
4417         systim = (u64)systimel;
4418         systim |= (u64)systimeh << 32;
4419
4420         if (adapter->flags2 & FLAG2_CHECK_SYSTIM_OVERFLOW)
4421                 systim = e1000e_sanitize_systim(hw, systim, sts);
4422
4423         return systim;
4424 }
4425
4426 /**
4427  * e1000e_cyclecounter_read - read raw cycle counter (used by time counter)
4428  * @cc: cyclecounter structure
4429  **/
4430 static u64 e1000e_cyclecounter_read(const struct cyclecounter *cc)
4431 {
4432         struct e1000_adapter *adapter = container_of(cc, struct e1000_adapter,
4433                                                      cc);
4434
4435         return e1000e_read_systim(adapter, NULL);
4436 }
4437
4438 /**
4439  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
4440  * @adapter: board private structure to initialize
4441  *
4442  * e1000_sw_init initializes the Adapter private data structure.
4443  * Fields are initialized based on PCI device information and
4444  * OS network device settings (MTU size).
4445  **/
4446 static int e1000_sw_init(struct e1000_adapter *adapter)
4447 {
4448         struct net_device *netdev = adapter->netdev;
4449
4450         adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
4451         adapter->rx_ps_bsize0 = 128;
4452         adapter->max_frame_size = netdev->mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
4453         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4454         adapter->tx_ring_count = E1000_DEFAULT_TXD;
4455         adapter->rx_ring_count = E1000_DEFAULT_RXD;
4456
4457         spin_lock_init(&adapter->stats64_lock);
4458
4459         e1000e_set_interrupt_capability(adapter);
4460
4461         if (e1000_alloc_queues(adapter))
4462                 return -ENOMEM;
4463
4464         /* Setup hardware time stamping cyclecounter */
4465         if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
4466                 adapter->cc.read = e1000e_cyclecounter_read;
4467                 adapter->cc.mask = CYCLECOUNTER_MASK(64);
4468                 adapter->cc.mult = 1;
4469                 /* cc.shift set in e1000e_get_base_tininca() */
4470
4471                 spin_lock_init(&adapter->systim_lock);
4472                 INIT_WORK(&adapter->tx_hwtstamp_work, e1000e_tx_hwtstamp_work);
4473         }
4474
4475         /* Explicitly disable IRQ since the NIC can be in any state. */
4476         e1000_irq_disable(adapter);
4477
4478         set_bit(__E1000_DOWN, &adapter->state);
4479         return 0;
4480 }
4481
4482 /**
4483  * e1000_intr_msi_test - Interrupt Handler
4484  * @irq: interrupt number
4485  * @data: pointer to a network interface device structure
4486  **/
4487 static irqreturn_t e1000_intr_msi_test(int __always_unused irq, void *data)
4488 {
4489         struct net_device *netdev = data;
4490         struct e1000_adapter *adapter = netdev_priv(netdev);
4491         struct e1000_hw *hw = &adapter->hw;
4492         u32 icr = er32(ICR);
4493
4494         e_dbg("icr is %08X\n", icr);
4495         if (icr & E1000_ICR_RXSEQ) {
4496                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
4497                 /* Force memory writes to complete before acknowledging the
4498                  * interrupt is handled.
4499                  */
4500                 wmb();
4501         }
4502
4503         return IRQ_HANDLED;
4504 }
4505
4506 /**
4507  * e1000_test_msi_interrupt - Returns 0 for successful test
4508  * @adapter: board private struct
4509  *
4510  * code flow taken from tg3.c
4511  **/
4512 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
4513 {
4514         struct net_device *netdev = adapter->netdev;
4515         struct e1000_hw *hw = &adapter->hw;
4516         int err;
4517
4518         /* poll_enable hasn't been called yet, so don't need disable */
4519         /* clear any pending events */
4520         er32(ICR);
4521
4522         /* free the real vector and request a test handler */
4523         e1000_free_irq(adapter);
4524         e1000e_reset_interrupt_capability(adapter);
4525
4526         /* Assume that the test fails, if it succeeds then the test
4527          * MSI irq handler will unset this flag
4528          */
4529         adapter->flags |= FLAG_MSI_TEST_FAILED;
4530
4531         err = pci_enable_msi(adapter->pdev);
4532         if (err)
4533                 goto msi_test_failed;
4534
4535         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
4536                           netdev->name, netdev);
4537         if (err) {
4538                 pci_disable_msi(adapter->pdev);
4539                 goto msi_test_failed;
4540         }
4541
4542         /* Force memory writes to complete before enabling and firing an
4543          * interrupt.
4544          */
4545         wmb();
4546
4547         e1000_irq_enable(adapter);
4548
4549         /* fire an unusual interrupt on the test handler */
4550         ew32(ICS, E1000_ICS_RXSEQ);
4551         e1e_flush();
4552         msleep(100);
4553
4554         e1000_irq_disable(adapter);
4555
4556         rmb();                  /* read flags after interrupt has been fired */
4557
4558         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
4559                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
4560                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
4561         } else {
4562                 e_dbg("MSI interrupt test succeeded!\n");
4563         }
4564
4565         free_irq(adapter->pdev->irq, netdev);
4566         pci_disable_msi(adapter->pdev);
4567
4568 msi_test_failed:
4569         e1000e_set_interrupt_capability(adapter);
4570         return e1000_request_irq(adapter);
4571 }
4572
4573 /**
4574  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
4575  * @adapter: board private struct
4576  *
4577  * code flow taken from tg3.c, called with e1000 interrupts disabled.
4578  **/
4579 static int e1000_test_msi(struct e1000_adapter *adapter)
4580 {
4581         int err;
4582         u16 pci_cmd;
4583
4584         if (!(adapter->flags & FLAG_MSI_ENABLED))
4585                 return 0;
4586
4587         /* disable SERR in case the MSI write causes a master abort */
4588         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4589         if (pci_cmd & PCI_COMMAND_SERR)
4590                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
4591                                       pci_cmd & ~PCI_COMMAND_SERR);
4592
4593         err = e1000_test_msi_interrupt(adapter);
4594
4595         /* re-enable SERR */
4596         if (pci_cmd & PCI_COMMAND_SERR) {
4597                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
4598                 pci_cmd |= PCI_COMMAND_SERR;
4599                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
4600         }
4601
4602         return err;
4603 }
4604
4605 /**
4606  * e1000e_open - Called when a network interface is made active
4607  * @netdev: network interface device structure
4608  *
4609  * Returns 0 on success, negative value on failure
4610  *
4611  * The open entry point is called when a network interface is made
4612  * active by the system (IFF_UP).  At this point all resources needed
4613  * for transmit and receive operations are allocated, the interrupt
4614  * handler is registered with the OS, the watchdog timer is started,
4615  * and the stack is notified that the interface is ready.
4616  **/
4617 int e1000e_open(struct net_device *netdev)
4618 {
4619         struct e1000_adapter *adapter = netdev_priv(netdev);
4620         struct e1000_hw *hw = &adapter->hw;
4621         struct pci_dev *pdev = adapter->pdev;
4622         int err;
4623
4624         /* disallow open during test */
4625         if (test_bit(__E1000_TESTING, &adapter->state))
4626                 return -EBUSY;
4627
4628         pm_runtime_get_sync(&pdev->dev);
4629
4630         netif_carrier_off(netdev);
4631         netif_stop_queue(netdev);
4632
4633         /* allocate transmit descriptors */
4634         err = e1000e_setup_tx_resources(adapter->tx_ring);
4635         if (err)
4636                 goto err_setup_tx;
4637
4638         /* allocate receive descriptors */
4639         err = e1000e_setup_rx_resources(adapter->rx_ring);
4640         if (err)
4641                 goto err_setup_rx;
4642
4643         /* If AMT is enabled, let the firmware know that the network
4644          * interface is now open and reset the part to a known state.
4645          */
4646         if (adapter->flags & FLAG_HAS_AMT) {
4647                 e1000e_get_hw_control(adapter);
4648                 e1000e_reset(adapter);
4649         }
4650
4651         e1000e_power_up_phy(adapter);
4652
4653         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4654         if ((adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
4655                 e1000_update_mng_vlan(adapter);
4656
4657         /* DMA latency requirement to workaround jumbo issue */
4658         cpu_latency_qos_add_request(&adapter->pm_qos_req, PM_QOS_DEFAULT_VALUE);
4659
4660         /* before we allocate an interrupt, we must be ready to handle it.
4661          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4662          * as soon as we call pci_request_irq, so we have to setup our
4663          * clean_rx handler before we do so.
4664          */
4665         e1000_configure(adapter);
4666
4667         err = e1000_request_irq(adapter);
4668         if (err)
4669                 goto err_req_irq;
4670
4671         /* Work around PCIe errata with MSI interrupts causing some chipsets to
4672          * ignore e1000e MSI messages, which means we need to test our MSI
4673          * interrupt now
4674          */
4675         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
4676                 err = e1000_test_msi(adapter);
4677                 if (err) {
4678                         e_err("Interrupt allocation failed\n");
4679                         goto err_req_irq;
4680                 }
4681         }
4682
4683         /* From here on the code is the same as e1000e_up() */
4684         clear_bit(__E1000_DOWN, &adapter->state);
4685
4686         napi_enable(&adapter->napi);
4687
4688         e1000_irq_enable(adapter);
4689
4690         adapter->tx_hang_recheck = false;
4691
4692         hw->mac.get_link_status = true;
4693         pm_runtime_put(&pdev->dev);
4694
4695         e1000e_trigger_lsc(adapter);
4696
4697         return 0;
4698
4699 err_req_irq:
4700         cpu_latency_qos_remove_request(&adapter->pm_qos_req);
4701         e1000e_release_hw_control(adapter);
4702         e1000_power_down_phy(adapter);
4703         e1000e_free_rx_resources(adapter->rx_ring);
4704 err_setup_rx:
4705         e1000e_free_tx_resources(adapter->tx_ring);
4706 err_setup_tx:
4707         e1000e_reset(adapter);
4708         pm_runtime_put_sync(&pdev->dev);
4709
4710         return err;
4711 }
4712
4713 /**
4714  * e1000e_close - Disables a network interface
4715  * @netdev: network interface device structure
4716  *
4717  * Returns 0, this is not allowed to fail
4718  *
4719  * The close entry point is called when an interface is de-activated
4720  * by the OS.  The hardware is still under the drivers control, but
4721  * needs to be disabled.  A global MAC reset is issued to stop the
4722  * hardware, and all transmit and receive resources are freed.
4723  **/
4724 int e1000e_close(struct net_device *netdev)
4725 {
4726         struct e1000_adapter *adapter = netdev_priv(netdev);
4727         struct pci_dev *pdev = adapter->pdev;
4728         int count = E1000_CHECK_RESET_COUNT;
4729
4730         while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
4731                 usleep_range(10000, 11000);
4732
4733         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4734
4735         pm_runtime_get_sync(&pdev->dev);
4736
4737         if (netif_device_present(netdev)) {
4738                 e1000e_down(adapter, true);
4739                 e1000_free_irq(adapter);
4740
4741                 /* Link status message must follow this format */
4742                 netdev_info(netdev, "NIC Link is Down\n");
4743         }
4744
4745         napi_disable(&adapter->napi);
4746
4747         e1000e_free_tx_resources(adapter->tx_ring);
4748         e1000e_free_rx_resources(adapter->rx_ring);
4749
4750         /* kill manageability vlan ID if supported, but not if a vlan with
4751          * the same ID is registered on the host OS (let 8021q kill it)
4752          */
4753         if (adapter->hw.mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4754                 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
4755                                        adapter->mng_vlan_id);
4756
4757         /* If AMT is enabled, let the firmware know that the network
4758          * interface is now closed
4759          */
4760         if ((adapter->flags & FLAG_HAS_AMT) &&
4761             !test_bit(__E1000_TESTING, &adapter->state))
4762                 e1000e_release_hw_control(adapter);
4763
4764         cpu_latency_qos_remove_request(&adapter->pm_qos_req);
4765
4766         pm_runtime_put_sync(&pdev->dev);
4767
4768         return 0;
4769 }
4770
4771 /**
4772  * e1000_set_mac - Change the Ethernet Address of the NIC
4773  * @netdev: network interface device structure
4774  * @p: pointer to an address structure
4775  *
4776  * Returns 0 on success, negative on failure
4777  **/
4778 static int e1000_set_mac(struct net_device *netdev, void *p)
4779 {
4780         struct e1000_adapter *adapter = netdev_priv(netdev);
4781         struct e1000_hw *hw = &adapter->hw;
4782         struct sockaddr *addr = p;
4783
4784         if (!is_valid_ether_addr(addr->sa_data))
4785                 return -EADDRNOTAVAIL;
4786
4787         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4788         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4789
4790         hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4791
4792         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4793                 /* activate the work around */
4794                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4795
4796                 /* Hold a copy of the LAA in RAR[14] This is done so that
4797                  * between the time RAR[0] gets clobbered  and the time it
4798                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4799                  * of the RARs and no incoming packets directed to this port
4800                  * are dropped. Eventually the LAA will be in RAR[0] and
4801                  * RAR[14]
4802                  */
4803                 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4804                                     adapter->hw.mac.rar_entry_count - 1);
4805         }
4806
4807         return 0;
4808 }
4809
4810 /**
4811  * e1000e_update_phy_task - work thread to update phy
4812  * @work: pointer to our work struct
4813  *
4814  * this worker thread exists because we must acquire a
4815  * semaphore to read the phy, which we could msleep while
4816  * waiting for it, and we can't msleep in a timer.
4817  **/
4818 static void e1000e_update_phy_task(struct work_struct *work)
4819 {
4820         struct e1000_adapter *adapter = container_of(work,
4821                                                      struct e1000_adapter,
4822                                                      update_phy_task);
4823         struct e1000_hw *hw = &adapter->hw;
4824
4825         if (test_bit(__E1000_DOWN, &adapter->state))
4826                 return;
4827
4828         e1000_get_phy_info(hw);
4829
4830         /* Enable EEE on 82579 after link up */
4831         if (hw->phy.type >= e1000_phy_82579)
4832                 e1000_set_eee_pchlan(hw);
4833 }
4834
4835 /**
4836  * e1000_update_phy_info - timre call-back to update PHY info
4837  * @t: pointer to timer_list containing private info adapter
4838  *
4839  * Need to wait a few seconds after link up to get diagnostic information from
4840  * the phy
4841  **/
4842 static void e1000_update_phy_info(struct timer_list *t)
4843 {
4844         struct e1000_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4845
4846         if (test_bit(__E1000_DOWN, &adapter->state))
4847                 return;
4848
4849         schedule_work(&adapter->update_phy_task);
4850 }
4851
4852 /**
4853  * e1000e_update_phy_stats - Update the PHY statistics counters
4854  * @adapter: board private structure
4855  *
4856  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4857  **/
4858 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4859 {
4860         struct e1000_hw *hw = &adapter->hw;
4861         s32 ret_val;
4862         u16 phy_data;
4863
4864         ret_val = hw->phy.ops.acquire(hw);
4865         if (ret_val)
4866                 return;
4867
4868         /* A page set is expensive so check if already on desired page.
4869          * If not, set to the page with the PHY status registers.
4870          */
4871         hw->phy.addr = 1;
4872         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4873                                            &phy_data);
4874         if (ret_val)
4875                 goto release;
4876         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4877                 ret_val = hw->phy.ops.set_page(hw,
4878                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4879                 if (ret_val)
4880                         goto release;
4881         }
4882
4883         /* Single Collision Count */
4884         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4885         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4886         if (!ret_val)
4887                 adapter->stats.scc += phy_data;
4888
4889         /* Excessive Collision Count */
4890         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4891         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4892         if (!ret_val)
4893                 adapter->stats.ecol += phy_data;
4894
4895         /* Multiple Collision Count */
4896         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4897         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4898         if (!ret_val)
4899                 adapter->stats.mcc += phy_data;
4900
4901         /* Late Collision Count */
4902         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4903         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4904         if (!ret_val)
4905                 adapter->stats.latecol += phy_data;
4906
4907         /* Collision Count - also used for adaptive IFS */
4908         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4909         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4910         if (!ret_val)
4911                 hw->mac.collision_delta = phy_data;
4912
4913         /* Defer Count */
4914         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4915         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4916         if (!ret_val)
4917                 adapter->stats.dc += phy_data;
4918
4919         /* Transmit with no CRS */
4920         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4921         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4922         if (!ret_val)
4923                 adapter->stats.tncrs += phy_data;
4924
4925 release:
4926         hw->phy.ops.release(hw);
4927 }
4928
4929 /**
4930  * e1000e_update_stats - Update the board statistics counters
4931  * @adapter: board private structure
4932  **/
4933 static void e1000e_update_stats(struct e1000_adapter *adapter)
4934 {
4935         struct net_device *netdev = adapter->netdev;
4936         struct e1000_hw *hw = &adapter->hw;
4937         struct pci_dev *pdev = adapter->pdev;
4938
4939         /* Prevent stats update while adapter is being reset, or if the pci
4940          * connection is down.
4941          */
4942         if (adapter->link_speed == 0)
4943                 return;
4944         if (pci_channel_offline(pdev))
4945                 return;
4946
4947         adapter->stats.crcerrs += er32(CRCERRS);
4948         adapter->stats.gprc += er32(GPRC);
4949         adapter->stats.gorc += er32(GORCL);
4950         er32(GORCH);            /* Clear gorc */
4951         adapter->stats.bprc += er32(BPRC);
4952         adapter->stats.mprc += er32(MPRC);
4953         adapter->stats.roc += er32(ROC);
4954
4955         adapter->stats.mpc += er32(MPC);
4956
4957         /* Half-duplex statistics */
4958         if (adapter->link_duplex == HALF_DUPLEX) {
4959                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4960                         e1000e_update_phy_stats(adapter);
4961                 } else {
4962                         adapter->stats.scc += er32(SCC);
4963                         adapter->stats.ecol += er32(ECOL);
4964                         adapter->stats.mcc += er32(MCC);
4965                         adapter->stats.latecol += er32(LATECOL);
4966                         adapter->stats.dc += er32(DC);
4967
4968                         hw->mac.collision_delta = er32(COLC);
4969
4970                         if ((hw->mac.type != e1000_82574) &&
4971                             (hw->mac.type != e1000_82583))
4972                                 adapter->stats.tncrs += er32(TNCRS);
4973                 }
4974                 adapter->stats.colc += hw->mac.collision_delta;
4975         }
4976
4977         adapter->stats.xonrxc += er32(XONRXC);
4978         adapter->stats.xontxc += er32(XONTXC);
4979         adapter->stats.xoffrxc += er32(XOFFRXC);
4980         adapter->stats.xofftxc += er32(XOFFTXC);
4981         adapter->stats.gptc += er32(GPTC);
4982         adapter->stats.gotc += er32(GOTCL);
4983         er32(GOTCH);            /* Clear gotc */
4984         adapter->stats.rnbc += er32(RNBC);
4985         adapter->stats.ruc += er32(RUC);
4986
4987         adapter->stats.mptc += er32(MPTC);
4988         adapter->stats.bptc += er32(BPTC);
4989
4990         /* used for adaptive IFS */
4991
4992         hw->mac.tx_packet_delta = er32(TPT);
4993         adapter->stats.tpt += hw->mac.tx_packet_delta;
4994
4995         adapter->stats.algnerrc += er32(ALGNERRC);
4996         adapter->stats.rxerrc += er32(RXERRC);
4997         adapter->stats.cexterr += er32(CEXTERR);
4998         adapter->stats.tsctc += er32(TSCTC);
4999         adapter->stats.tsctfc += er32(TSCTFC);
5000
5001         /* Fill out the OS statistics structure */
5002         netdev->stats.multicast = adapter->stats.mprc;
5003         netdev->stats.collisions = adapter->stats.colc;
5004
5005         /* Rx Errors */
5006
5007         /* RLEC on some newer hardware can be incorrect so build
5008          * our own version based on RUC and ROC
5009          */
5010         netdev->stats.rx_errors = adapter->stats.rxerrc +
5011             adapter->stats.crcerrs + adapter->stats.algnerrc +
5012             adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
5013         netdev->stats.rx_length_errors = adapter->stats.ruc +
5014             adapter->stats.roc;
5015         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5016         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
5017         netdev->stats.rx_missed_errors = adapter->stats.mpc;
5018
5019         /* Tx Errors */
5020         netdev->stats.tx_errors = adapter->stats.ecol + adapter->stats.latecol;
5021         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
5022         netdev->stats.tx_window_errors = adapter->stats.latecol;
5023         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
5024
5025         /* Tx Dropped needs to be maintained elsewhere */
5026
5027         /* Management Stats */
5028         adapter->stats.mgptc += er32(MGTPTC);
5029         adapter->stats.mgprc += er32(MGTPRC);
5030         adapter->stats.mgpdc += er32(MGTPDC);
5031
5032         /* Correctable ECC Errors */
5033         if (hw->mac.type >= e1000_pch_lpt) {
5034                 u32 pbeccsts = er32(PBECCSTS);
5035
5036                 adapter->corr_errors +=
5037                     pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
5038                 adapter->uncorr_errors +=
5039                     (pbeccsts & E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
5040                     E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
5041         }
5042 }
5043
5044 /**
5045  * e1000_phy_read_status - Update the PHY register status snapshot
5046  * @adapter: board private structure
5047  **/
5048 static void e1000_phy_read_status(struct e1000_adapter *adapter)
5049 {
5050         struct e1000_hw *hw = &adapter->hw;
5051         struct e1000_phy_regs *phy = &adapter->phy_regs;
5052
5053         if (!pm_runtime_suspended((&adapter->pdev->dev)->parent) &&
5054             (er32(STATUS) & E1000_STATUS_LU) &&
5055             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
5056                 int ret_val;
5057
5058                 ret_val = e1e_rphy(hw, MII_BMCR, &phy->bmcr);
5059                 ret_val |= e1e_rphy(hw, MII_BMSR, &phy->bmsr);
5060                 ret_val |= e1e_rphy(hw, MII_ADVERTISE, &phy->advertise);
5061                 ret_val |= e1e_rphy(hw, MII_LPA, &phy->lpa);
5062                 ret_val |= e1e_rphy(hw, MII_EXPANSION, &phy->expansion);
5063                 ret_val |= e1e_rphy(hw, MII_CTRL1000, &phy->ctrl1000);
5064                 ret_val |= e1e_rphy(hw, MII_STAT1000, &phy->stat1000);
5065                 ret_val |= e1e_rphy(hw, MII_ESTATUS, &phy->estatus);
5066                 if (ret_val)
5067                         e_warn("Error reading PHY register\n");
5068         } else {
5069                 /* Do not read PHY registers if link is not up
5070                  * Set values to typical power-on defaults
5071                  */
5072                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
5073                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
5074                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
5075                              BMSR_ERCAP);
5076                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
5077                                   ADVERTISE_ALL | ADVERTISE_CSMA);
5078                 phy->lpa = 0;
5079                 phy->expansion = EXPANSION_ENABLENPAGE;
5080                 phy->ctrl1000 = ADVERTISE_1000FULL;
5081                 phy->stat1000 = 0;
5082                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
5083         }
5084 }
5085
5086 static void e1000_print_link_info(struct e1000_adapter *adapter)
5087 {
5088         struct e1000_hw *hw = &adapter->hw;
5089         u32 ctrl = er32(CTRL);
5090
5091         /* Link status message must follow this format for user tools */
5092         netdev_info(adapter->netdev,
5093                     "NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5094                     adapter->link_speed,
5095                     adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
5096                     (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
5097                     (ctrl & E1000_CTRL_RFCE) ? "Rx" :
5098                     (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
5099 }
5100
5101 static bool e1000e_has_link(struct e1000_adapter *adapter)
5102 {
5103         struct e1000_hw *hw = &adapter->hw;
5104         bool link_active = false;
5105         s32 ret_val = 0;
5106
5107         /* get_link_status is set on LSC (link status) interrupt or
5108          * Rx sequence error interrupt.  get_link_status will stay
5109          * true until the check_for_link establishes link
5110          * for copper adapters ONLY
5111          */
5112         switch (hw->phy.media_type) {
5113         case e1000_media_type_copper:
5114                 if (hw->mac.get_link_status) {
5115                         ret_val = hw->mac.ops.check_for_link(hw);
5116                         link_active = !hw->mac.get_link_status;
5117                 } else {
5118                         link_active = true;
5119                 }
5120                 break;
5121         case e1000_media_type_fiber:
5122                 ret_val = hw->mac.ops.check_for_link(hw);
5123                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
5124                 break;
5125         case e1000_media_type_internal_serdes:
5126                 ret_val = hw->mac.ops.check_for_link(hw);
5127                 link_active = hw->mac.serdes_has_link;
5128                 break;
5129         default:
5130         case e1000_media_type_unknown:
5131                 break;
5132         }
5133
5134         if ((ret_val == -E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
5135             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
5136                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
5137                 e_info("Gigabit has been disabled, downgrading speed\n");
5138         }
5139
5140         return link_active;
5141 }
5142
5143 static void e1000e_enable_receives(struct e1000_adapter *adapter)
5144 {
5145         /* make sure the receive unit is started */
5146         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5147             (adapter->flags & FLAG_RESTART_NOW)) {
5148                 struct e1000_hw *hw = &adapter->hw;
5149                 u32 rctl = er32(RCTL);
5150
5151                 ew32(RCTL, rctl | E1000_RCTL_EN);
5152                 adapter->flags &= ~FLAG_RESTART_NOW;
5153         }
5154 }
5155
5156 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
5157 {
5158         struct e1000_hw *hw = &adapter->hw;
5159
5160         /* With 82574 controllers, PHY needs to be checked periodically
5161          * for hung state and reset, if two calls return true
5162          */
5163         if (e1000_check_phy_82574(hw))
5164                 adapter->phy_hang_count++;
5165         else
5166                 adapter->phy_hang_count = 0;
5167
5168         if (adapter->phy_hang_count > 1) {
5169                 adapter->phy_hang_count = 0;
5170                 e_dbg("PHY appears hung - resetting\n");
5171                 schedule_work(&adapter->reset_task);
5172         }
5173 }
5174
5175 /**
5176  * e1000_watchdog - Timer Call-back
5177  * @t: pointer to timer_list containing private info adapter
5178  **/
5179 static void e1000_watchdog(struct timer_list *t)
5180 {
5181         struct e1000_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5182
5183         /* Do the rest outside of interrupt context */
5184         schedule_work(&adapter->watchdog_task);
5185
5186         /* TODO: make this use queue_delayed_work() */
5187 }
5188
5189 static void e1000_watchdog_task(struct work_struct *work)
5190 {
5191         struct e1000_adapter *adapter = container_of(work,
5192                                                      struct e1000_adapter,
5193                                                      watchdog_task);
5194         struct net_device *netdev = adapter->netdev;
5195         struct e1000_mac_info *mac = &adapter->hw.mac;
5196         struct e1000_phy_info *phy = &adapter->hw.phy;
5197         struct e1000_ring *tx_ring = adapter->tx_ring;
5198         u32 dmoff_exit_timeout = 100, tries = 0;
5199         struct e1000_hw *hw = &adapter->hw;
5200         u32 link, tctl, pcim_state;
5201
5202         if (test_bit(__E1000_DOWN, &adapter->state))
5203                 return;
5204
5205         link = e1000e_has_link(adapter);
5206         if ((netif_carrier_ok(netdev)) && link) {
5207                 /* Cancel scheduled suspend requests. */
5208                 pm_runtime_resume(netdev->dev.parent);
5209
5210                 e1000e_enable_receives(adapter);
5211                 goto link_up;
5212         }
5213
5214         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
5215             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
5216                 e1000_update_mng_vlan(adapter);
5217
5218         if (link) {
5219                 if (!netif_carrier_ok(netdev)) {
5220                         bool txb2b = true;
5221
5222                         /* Cancel scheduled suspend requests. */
5223                         pm_runtime_resume(netdev->dev.parent);
5224
5225                         /* Checking if MAC is in DMoff state*/
5226                         if (er32(FWSM) & E1000_ICH_FWSM_FW_VALID) {
5227                                 pcim_state = er32(STATUS);
5228                                 while (pcim_state & E1000_STATUS_PCIM_STATE) {
5229                                         if (tries++ == dmoff_exit_timeout) {
5230                                                 e_dbg("Error in exiting dmoff\n");
5231                                                 break;
5232                                         }
5233                                         usleep_range(10000, 20000);
5234                                         pcim_state = er32(STATUS);
5235
5236                                         /* Checking if MAC exited DMoff state */
5237                                         if (!(pcim_state & E1000_STATUS_PCIM_STATE))
5238                                                 e1000_phy_hw_reset(&adapter->hw);
5239                                 }
5240                         }
5241
5242                         /* update snapshot of PHY registers on LSC */
5243                         e1000_phy_read_status(adapter);
5244                         mac->ops.get_link_up_info(&adapter->hw,
5245                                                   &adapter->link_speed,
5246                                                   &adapter->link_duplex);
5247                         e1000_print_link_info(adapter);
5248
5249                         /* check if SmartSpeed worked */
5250                         e1000e_check_downshift(hw);
5251                         if (phy->speed_downgraded)
5252                                 netdev_warn(netdev,
5253                                             "Link Speed was downgraded by SmartSpeed\n");
5254
5255                         /* On supported PHYs, check for duplex mismatch only
5256                          * if link has autonegotiated at 10/100 half
5257                          */
5258                         if ((hw->phy.type == e1000_phy_igp_3 ||
5259                              hw->phy.type == e1000_phy_bm) &&
5260                             hw->mac.autoneg &&
5261                             (adapter->link_speed == SPEED_10 ||
5262                              adapter->link_speed == SPEED_100) &&
5263                             (adapter->link_duplex == HALF_DUPLEX)) {
5264                                 u16 autoneg_exp;
5265
5266                                 e1e_rphy(hw, MII_EXPANSION, &autoneg_exp);
5267
5268                                 if (!(autoneg_exp & EXPANSION_NWAY))
5269                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
5270                         }
5271
5272                         /* adjust timeout factor according to speed/duplex */
5273                         adapter->tx_timeout_factor = 1;
5274                         switch (adapter->link_speed) {
5275                         case SPEED_10:
5276                                 txb2b = false;
5277                                 adapter->tx_timeout_factor = 16;
5278                                 break;
5279                         case SPEED_100:
5280                                 txb2b = false;
5281                                 adapter->tx_timeout_factor = 10;
5282                                 break;
5283                         }
5284
5285                         /* workaround: re-program speed mode bit after
5286                          * link-up event
5287                          */
5288                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
5289                             !txb2b) {
5290                                 u32 tarc0;
5291
5292                                 tarc0 = er32(TARC(0));
5293                                 tarc0 &= ~SPEED_MODE_BIT;
5294                                 ew32(TARC(0), tarc0);
5295                         }
5296
5297                         /* disable TSO for pcie and 10/100 speeds, to avoid
5298                          * some hardware issues
5299                          */
5300                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
5301                                 switch (adapter->link_speed) {
5302                                 case SPEED_10:
5303                                 case SPEED_100:
5304                                         e_info("10/100 speed: disabling TSO\n");
5305                                         netdev->features &= ~NETIF_F_TSO;
5306                                         netdev->features &= ~NETIF_F_TSO6;
5307                                         break;
5308                                 case SPEED_1000:
5309                                         netdev->features |= NETIF_F_TSO;
5310                                         netdev->features |= NETIF_F_TSO6;
5311                                         break;
5312                                 default:
5313                                         /* oops */
5314                                         break;
5315                                 }
5316                                 if (hw->mac.type == e1000_pch_spt) {
5317                                         netdev->features &= ~NETIF_F_TSO;
5318                                         netdev->features &= ~NETIF_F_TSO6;
5319                                 }
5320                         }
5321
5322                         /* enable transmits in the hardware, need to do this
5323                          * after setting TARC(0)
5324                          */
5325                         tctl = er32(TCTL);
5326                         tctl |= E1000_TCTL_EN;
5327                         ew32(TCTL, tctl);
5328
5329                         /* Perform any post-link-up configuration before
5330                          * reporting link up.
5331                          */
5332                         if (phy->ops.cfg_on_link_up)
5333                                 phy->ops.cfg_on_link_up(hw);
5334
5335                         netif_wake_queue(netdev);
5336                         netif_carrier_on(netdev);
5337
5338                         if (!test_bit(__E1000_DOWN, &adapter->state))
5339                                 mod_timer(&adapter->phy_info_timer,
5340                                           round_jiffies(jiffies + 2 * HZ));
5341                 }
5342         } else {
5343                 if (netif_carrier_ok(netdev)) {
5344                         adapter->link_speed = 0;
5345                         adapter->link_duplex = 0;
5346                         /* Link status message must follow this format */
5347                         netdev_info(netdev, "NIC Link is Down\n");
5348                         netif_carrier_off(netdev);
5349                         netif_stop_queue(netdev);
5350                         if (!test_bit(__E1000_DOWN, &adapter->state))
5351                                 mod_timer(&adapter->phy_info_timer,
5352                                           round_jiffies(jiffies + 2 * HZ));
5353
5354                         /* 8000ES2LAN requires a Rx packet buffer work-around
5355                          * on link down event; reset the controller to flush
5356                          * the Rx packet buffer.
5357                          */
5358                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
5359                                 adapter->flags |= FLAG_RESTART_NOW;
5360                         else
5361                                 pm_schedule_suspend(netdev->dev.parent,
5362                                                     LINK_TIMEOUT);
5363                 }
5364         }
5365
5366 link_up:
5367         spin_lock(&adapter->stats64_lock);
5368         e1000e_update_stats(adapter);
5369
5370         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
5371         adapter->tpt_old = adapter->stats.tpt;
5372         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
5373         adapter->colc_old = adapter->stats.colc;
5374
5375         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
5376         adapter->gorc_old = adapter->stats.gorc;
5377         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
5378         adapter->gotc_old = adapter->stats.gotc;
5379         spin_unlock(&adapter->stats64_lock);
5380
5381         /* If the link is lost the controller stops DMA, but
5382          * if there is queued Tx work it cannot be done.  So
5383          * reset the controller to flush the Tx packet buffers.
5384          */
5385         if (!netif_carrier_ok(netdev) &&
5386             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
5387                 adapter->flags |= FLAG_RESTART_NOW;
5388
5389         /* If reset is necessary, do it outside of interrupt context. */
5390         if (adapter->flags & FLAG_RESTART_NOW) {
5391                 schedule_work(&adapter->reset_task);
5392                 /* return immediately since reset is imminent */
5393                 return;
5394         }
5395
5396         e1000e_update_adaptive(&adapter->hw);
5397
5398         /* Simple mode for Interrupt Throttle Rate (ITR) */
5399         if (adapter->itr_setting == 4) {
5400                 /* Symmetric Tx/Rx gets a reduced ITR=2000;
5401                  * Total asymmetrical Tx or Rx gets ITR=8000;
5402                  * everyone else is between 2000-8000.
5403                  */
5404                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
5405                 u32 dif = (adapter->gotc > adapter->gorc ?
5406                            adapter->gotc - adapter->gorc :
5407                            adapter->gorc - adapter->gotc) / 10000;
5408                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
5409
5410                 e1000e_write_itr(adapter, itr);
5411         }
5412
5413         /* Cause software interrupt to ensure Rx ring is cleaned */
5414         if (adapter->msix_entries)
5415                 ew32(ICS, adapter->rx_ring->ims_val);
5416         else
5417                 ew32(ICS, E1000_ICS_RXDMT0);
5418
5419         /* flush pending descriptors to memory before detecting Tx hang */
5420         e1000e_flush_descriptors(adapter);
5421
5422         /* Force detection of hung controller every watchdog period */
5423         adapter->detect_tx_hung = true;
5424
5425         /* With 82571 controllers, LAA may be overwritten due to controller
5426          * reset from the other port. Set the appropriate LAA in RAR[0]
5427          */
5428         if (e1000e_get_laa_state_82571(hw))
5429                 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
5430
5431         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
5432                 e1000e_check_82574_phy_workaround(adapter);
5433
5434         /* Clear valid timestamp stuck in RXSTMPL/H due to a Rx error */
5435         if (adapter->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
5436                 if ((adapter->flags2 & FLAG2_CHECK_RX_HWTSTAMP) &&
5437                     (er32(TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID)) {
5438                         er32(RXSTMPH);
5439                         adapter->rx_hwtstamp_cleared++;
5440                 } else {
5441                         adapter->flags2 |= FLAG2_CHECK_RX_HWTSTAMP;
5442                 }
5443         }
5444
5445         /* Reset the timer */
5446         if (!test_bit(__E1000_DOWN, &adapter->state))
5447                 mod_timer(&adapter->watchdog_timer,
5448                           round_jiffies(jiffies + 2 * HZ));
5449 }
5450
5451 #define E1000_TX_FLAGS_CSUM             0x00000001
5452 #define E1000_TX_FLAGS_VLAN             0x00000002
5453 #define E1000_TX_FLAGS_TSO              0x00000004
5454 #define E1000_TX_FLAGS_IPV4             0x00000008
5455 #define E1000_TX_FLAGS_NO_FCS           0x00000010
5456 #define E1000_TX_FLAGS_HWTSTAMP         0x00000020
5457 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
5458 #define E1000_TX_FLAGS_VLAN_SHIFT       16
5459
5460 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb,
5461                      __be16 protocol)
5462 {
5463         struct e1000_context_desc *context_desc;
5464         struct e1000_buffer *buffer_info;
5465         unsigned int i;
5466         u32 cmd_length = 0;
5467         u16 ipcse = 0, mss;
5468         u8 ipcss, ipcso, tucss, tucso, hdr_len;
5469         int err;
5470
5471         if (!skb_is_gso(skb))
5472                 return 0;
5473
5474         err = skb_cow_head(skb, 0);
5475         if (err < 0)
5476                 return err;
5477
5478         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5479         mss = skb_shinfo(skb)->gso_size;
5480         if (protocol == htons(ETH_P_IP)) {
5481                 struct iphdr *iph = ip_hdr(skb);
5482                 iph->tot_len = 0;
5483                 iph->check = 0;
5484                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
5485                                                          0, IPPROTO_TCP, 0);
5486                 cmd_length = E1000_TXD_CMD_IP;
5487                 ipcse = skb_transport_offset(skb) - 1;
5488         } else if (skb_is_gso_v6(skb)) {
5489                 tcp_v6_gso_csum_prep(skb);
5490                 ipcse = 0;
5491         }
5492         ipcss = skb_network_offset(skb);
5493         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
5494         tucss = skb_transport_offset(skb);
5495         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
5496
5497         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
5498                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
5499
5500         i = tx_ring->next_to_use;
5501         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5502         buffer_info = &tx_ring->buffer_info[i];
5503
5504         context_desc->lower_setup.ip_fields.ipcss = ipcss;
5505         context_desc->lower_setup.ip_fields.ipcso = ipcso;
5506         context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
5507         context_desc->upper_setup.tcp_fields.tucss = tucss;
5508         context_desc->upper_setup.tcp_fields.tucso = tucso;
5509         context_desc->upper_setup.tcp_fields.tucse = 0;
5510         context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
5511         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
5512         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
5513
5514         buffer_info->time_stamp = jiffies;
5515         buffer_info->next_to_watch = i;
5516
5517         i++;
5518         if (i == tx_ring->count)
5519                 i = 0;
5520         tx_ring->next_to_use = i;
5521
5522         return 1;
5523 }
5524
5525 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb,
5526                           __be16 protocol)
5527 {
5528         struct e1000_adapter *adapter = tx_ring->adapter;
5529         struct e1000_context_desc *context_desc;
5530         struct e1000_buffer *buffer_info;
5531         unsigned int i;
5532         u8 css;
5533         u32 cmd_len = E1000_TXD_CMD_DEXT;
5534
5535         if (skb->ip_summed != CHECKSUM_PARTIAL)
5536                 return false;
5537
5538         switch (protocol) {
5539         case cpu_to_be16(ETH_P_IP):
5540                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5541                         cmd_len |= E1000_TXD_CMD_TCP;
5542                 break;
5543         case cpu_to_be16(ETH_P_IPV6):
5544                 /* XXX not handling all IPV6 headers */
5545                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5546                         cmd_len |= E1000_TXD_CMD_TCP;
5547                 break;
5548         default:
5549                 if (unlikely(net_ratelimit()))
5550                         e_warn("checksum_partial proto=%x!\n",
5551                                be16_to_cpu(protocol));
5552                 break;
5553         }
5554
5555         css = skb_checksum_start_offset(skb);
5556
5557         i = tx_ring->next_to_use;
5558         buffer_info = &tx_ring->buffer_info[i];
5559         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
5560
5561         context_desc->lower_setup.ip_config = 0;
5562         context_desc->upper_setup.tcp_fields.tucss = css;
5563         context_desc->upper_setup.tcp_fields.tucso = css + skb->csum_offset;
5564         context_desc->upper_setup.tcp_fields.tucse = 0;
5565         context_desc->tcp_seg_setup.data = 0;
5566         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
5567
5568         buffer_info->time_stamp = jiffies;
5569         buffer_info->next_to_watch = i;
5570
5571         i++;
5572         if (i == tx_ring->count)
5573                 i = 0;
5574         tx_ring->next_to_use = i;
5575
5576         return true;
5577 }
5578
5579 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
5580                         unsigned int first, unsigned int max_per_txd,
5581                         unsigned int nr_frags)
5582 {
5583         struct e1000_adapter *adapter = tx_ring->adapter;
5584         struct pci_dev *pdev = adapter->pdev;
5585         struct e1000_buffer *buffer_info;
5586         unsigned int len = skb_headlen(skb);
5587         unsigned int offset = 0, size, count = 0, i;
5588         unsigned int f, bytecount, segs;
5589
5590         i = tx_ring->next_to_use;
5591
5592         while (len) {
5593                 buffer_info = &tx_ring->buffer_info[i];
5594                 size = min(len, max_per_txd);
5595
5596                 buffer_info->length = size;
5597                 buffer_info->time_stamp = jiffies;
5598                 buffer_info->next_to_watch = i;
5599                 buffer_info->dma = dma_map_single(&pdev->dev,
5600                                                   skb->data + offset,
5601                                                   size, DMA_TO_DEVICE);
5602                 buffer_info->mapped_as_page = false;
5603                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5604                         goto dma_error;
5605
5606                 len -= size;
5607                 offset += size;
5608                 count++;
5609
5610                 if (len) {
5611                         i++;
5612                         if (i == tx_ring->count)
5613                                 i = 0;
5614                 }
5615         }
5616
5617         for (f = 0; f < nr_frags; f++) {
5618                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
5619
5620                 len = skb_frag_size(frag);
5621                 offset = 0;
5622
5623                 while (len) {
5624                         i++;
5625                         if (i == tx_ring->count)
5626                                 i = 0;
5627
5628                         buffer_info = &tx_ring->buffer_info[i];
5629                         size = min(len, max_per_txd);
5630
5631                         buffer_info->length = size;
5632                         buffer_info->time_stamp = jiffies;
5633                         buffer_info->next_to_watch = i;
5634                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
5635                                                             offset, size,
5636                                                             DMA_TO_DEVICE);
5637                         buffer_info->mapped_as_page = true;
5638                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
5639                                 goto dma_error;
5640
5641                         len -= size;
5642                         offset += size;
5643                         count++;
5644                 }
5645         }
5646
5647         segs = skb_shinfo(skb)->gso_segs ? : 1;
5648         /* multiply data chunks by size of headers */
5649         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
5650
5651         tx_ring->buffer_info[i].skb = skb;
5652         tx_ring->buffer_info[i].segs = segs;
5653         tx_ring->buffer_info[i].bytecount = bytecount;
5654         tx_ring->buffer_info[first].next_to_watch = i;
5655
5656         return count;
5657
5658 dma_error:
5659         dev_err(&pdev->dev, "Tx DMA map failed\n");
5660         buffer_info->dma = 0;
5661         if (count)
5662                 count--;
5663
5664         while (count--) {
5665                 if (i == 0)
5666                         i += tx_ring->count;
5667                 i--;
5668                 buffer_info = &tx_ring->buffer_info[i];
5669                 e1000_put_txbuf(tx_ring, buffer_info, true);
5670         }
5671
5672         return 0;
5673 }
5674
5675 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
5676 {
5677         struct e1000_adapter *adapter = tx_ring->adapter;
5678         struct e1000_tx_desc *tx_desc = NULL;
5679         struct e1000_buffer *buffer_info;
5680         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
5681         unsigned int i;
5682
5683         if (tx_flags & E1000_TX_FLAGS_TSO) {
5684                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
5685                     E1000_TXD_CMD_TSE;
5686                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5687
5688                 if (tx_flags & E1000_TX_FLAGS_IPV4)
5689                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
5690         }
5691
5692         if (tx_flags & E1000_TX_FLAGS_CSUM) {
5693                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5694                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
5695         }
5696
5697         if (tx_flags & E1000_TX_FLAGS_VLAN) {
5698                 txd_lower |= E1000_TXD_CMD_VLE;
5699                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
5700         }
5701
5702         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5703                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
5704
5705         if (unlikely(tx_flags & E1000_TX_FLAGS_HWTSTAMP)) {
5706                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
5707                 txd_upper |= E1000_TXD_EXTCMD_TSTAMP;
5708         }
5709
5710         i = tx_ring->next_to_use;
5711
5712         do {
5713                 buffer_info = &tx_ring->buffer_info[i];
5714                 tx_desc = E1000_TX_DESC(*tx_ring, i);
5715                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
5716                 tx_desc->lower.data = cpu_to_le32(txd_lower |
5717                                                   buffer_info->length);
5718                 tx_desc->upper.data = cpu_to_le32(txd_upper);
5719
5720                 i++;
5721                 if (i == tx_ring->count)
5722                         i = 0;
5723         } while (--count > 0);
5724
5725         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
5726
5727         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
5728         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
5729                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
5730
5731         /* Force memory writes to complete before letting h/w
5732          * know there are new descriptors to fetch.  (Only
5733          * applicable for weak-ordered memory model archs,
5734          * such as IA-64).
5735          */
5736         wmb();
5737
5738         tx_ring->next_to_use = i;
5739 }
5740
5741 #define MINIMUM_DHCP_PACKET_SIZE 282
5742 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
5743                                     struct sk_buff *skb)
5744 {
5745         struct e1000_hw *hw = &adapter->hw;
5746         u16 length, offset;
5747
5748         if (skb_vlan_tag_present(skb) &&
5749             !((skb_vlan_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
5750               (adapter->hw.mng_cookie.status &
5751                E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
5752                 return 0;
5753
5754         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
5755                 return 0;
5756
5757         if (((struct ethhdr *)skb->data)->h_proto != htons(ETH_P_IP))
5758                 return 0;
5759
5760         {
5761                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data + 14);
5762                 struct udphdr *udp;
5763
5764                 if (ip->protocol != IPPROTO_UDP)
5765                         return 0;
5766
5767                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
5768                 if (ntohs(udp->dest) != 67)
5769                         return 0;
5770
5771                 offset = (u8 *)udp + 8 - skb->data;
5772                 length = skb->len - offset;
5773                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5774         }
5775
5776         return 0;
5777 }
5778
5779 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5780 {
5781         struct e1000_adapter *adapter = tx_ring->adapter;
5782
5783         netif_stop_queue(adapter->netdev);
5784         /* Herbert's original patch had:
5785          *  smp_mb__after_netif_stop_queue();
5786          * but since that doesn't exist yet, just open code it.
5787          */
5788         smp_mb();
5789
5790         /* We need to check again in a case another CPU has just
5791          * made room available.
5792          */
5793         if (e1000_desc_unused(tx_ring) < size)
5794                 return -EBUSY;
5795
5796         /* A reprieve! */
5797         netif_start_queue(adapter->netdev);
5798         ++adapter->restart_queue;
5799         return 0;
5800 }
5801
5802 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5803 {
5804         BUG_ON(size > tx_ring->count);
5805
5806         if (e1000_desc_unused(tx_ring) >= size)
5807                 return 0;
5808         return __e1000_maybe_stop_tx(tx_ring, size);
5809 }
5810
5811 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5812                                     struct net_device *netdev)
5813 {
5814         struct e1000_adapter *adapter = netdev_priv(netdev);
5815         struct e1000_ring *tx_ring = adapter->tx_ring;
5816         unsigned int first;
5817         unsigned int tx_flags = 0;
5818         unsigned int len = skb_headlen(skb);
5819         unsigned int nr_frags;
5820         unsigned int mss;
5821         int count = 0;
5822         int tso;
5823         unsigned int f;
5824         __be16 protocol = vlan_get_protocol(skb);
5825
5826         if (test_bit(__E1000_DOWN, &adapter->state)) {
5827                 dev_kfree_skb_any(skb);
5828                 return NETDEV_TX_OK;
5829         }
5830
5831         if (skb->len <= 0) {
5832                 dev_kfree_skb_any(skb);
5833                 return NETDEV_TX_OK;
5834         }
5835
5836         /* The minimum packet size with TCTL.PSP set is 17 bytes so
5837          * pad skb in order to meet this minimum size requirement
5838          */
5839         if (skb_put_padto(skb, 17))
5840                 return NETDEV_TX_OK;
5841
5842         mss = skb_shinfo(skb)->gso_size;
5843         if (mss) {
5844                 u8 hdr_len;
5845
5846                 /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
5847                  * points to just header, pull a few bytes of payload from
5848                  * frags into skb->data
5849                  */
5850                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5851                 /* we do this workaround for ES2LAN, but it is un-necessary,
5852                  * avoiding it could save a lot of cycles
5853                  */
5854                 if (skb->data_len && (hdr_len == len)) {
5855                         unsigned int pull_size;
5856
5857                         pull_size = min_t(unsigned int, 4, skb->data_len);
5858                         if (!__pskb_pull_tail(skb, pull_size)) {
5859                                 e_err("__pskb_pull_tail failed.\n");
5860                                 dev_kfree_skb_any(skb);
5861                                 return NETDEV_TX_OK;
5862                         }
5863                         len = skb_headlen(skb);
5864                 }
5865         }
5866
5867         /* reserve a descriptor for the offload context */
5868         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5869                 count++;
5870         count++;
5871
5872         count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5873
5874         nr_frags = skb_shinfo(skb)->nr_frags;
5875         for (f = 0; f < nr_frags; f++)
5876                 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5877                                       adapter->tx_fifo_limit);
5878
5879         if (adapter->hw.mac.tx_pkt_filtering)
5880                 e1000_transfer_dhcp_info(adapter, skb);
5881
5882         /* need: count + 2 desc gap to keep tail from touching
5883          * head, otherwise try next time
5884          */
5885         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5886                 return NETDEV_TX_BUSY;
5887
5888         if (skb_vlan_tag_present(skb)) {
5889                 tx_flags |= E1000_TX_FLAGS_VLAN;
5890                 tx_flags |= (skb_vlan_tag_get(skb) <<
5891                              E1000_TX_FLAGS_VLAN_SHIFT);
5892         }
5893
5894         first = tx_ring->next_to_use;
5895
5896         tso = e1000_tso(tx_ring, skb, protocol);
5897         if (tso < 0) {
5898                 dev_kfree_skb_any(skb);
5899                 return NETDEV_TX_OK;
5900         }
5901
5902         if (tso)
5903                 tx_flags |= E1000_TX_FLAGS_TSO;
5904         else if (e1000_tx_csum(tx_ring, skb, protocol))
5905                 tx_flags |= E1000_TX_FLAGS_CSUM;
5906
5907         /* Old method was to assume IPv4 packet by default if TSO was enabled.
5908          * 82571 hardware supports TSO capabilities for IPv6 as well...
5909          * no longer assume, we must.
5910          */
5911         if (protocol == htons(ETH_P_IP))
5912                 tx_flags |= E1000_TX_FLAGS_IPV4;
5913
5914         if (unlikely(skb->no_fcs))
5915                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5916
5917         /* if count is 0 then mapping error has occurred */
5918         count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5919                              nr_frags);
5920         if (count) {
5921                 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
5922                     (adapter->flags & FLAG_HAS_HW_TIMESTAMP)) {
5923                         if (!adapter->tx_hwtstamp_skb) {
5924                                 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5925                                 tx_flags |= E1000_TX_FLAGS_HWTSTAMP;
5926                                 adapter->tx_hwtstamp_skb = skb_get(skb);
5927                                 adapter->tx_hwtstamp_start = jiffies;
5928                                 schedule_work(&adapter->tx_hwtstamp_work);
5929                         } else {
5930                                 adapter->tx_hwtstamp_skipped++;
5931                         }
5932                 }
5933
5934                 skb_tx_timestamp(skb);
5935
5936                 netdev_sent_queue(netdev, skb->len);
5937                 e1000_tx_queue(tx_ring, tx_flags, count);
5938                 /* Make sure there is space in the ring for the next send. */
5939                 e1000_maybe_stop_tx(tx_ring,
5940                                     (MAX_SKB_FRAGS *
5941                                      DIV_ROUND_UP(PAGE_SIZE,
5942                                                   adapter->tx_fifo_limit) + 2));
5943
5944                 if (!netdev_xmit_more() ||
5945                     netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
5946                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
5947                                 e1000e_update_tdt_wa(tx_ring,
5948                                                      tx_ring->next_to_use);
5949                         else
5950                                 writel(tx_ring->next_to_use, tx_ring->tail);
5951                 }
5952         } else {
5953                 dev_kfree_skb_any(skb);
5954                 tx_ring->buffer_info[first].time_stamp = 0;
5955                 tx_ring->next_to_use = first;
5956         }
5957
5958         return NETDEV_TX_OK;
5959 }
5960
5961 /**
5962  * e1000_tx_timeout - Respond to a Tx Hang
5963  * @netdev: network interface device structure
5964  * @txqueue: index of the hung queue (unused)
5965  **/
5966 static void e1000_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
5967 {
5968         struct e1000_adapter *adapter = netdev_priv(netdev);
5969
5970         /* Do the reset outside of interrupt context */
5971         adapter->tx_timeout_count++;
5972         schedule_work(&adapter->reset_task);
5973 }
5974
5975 static void e1000_reset_task(struct work_struct *work)
5976 {
5977         struct e1000_adapter *adapter;
5978         adapter = container_of(work, struct e1000_adapter, reset_task);
5979
5980         rtnl_lock();
5981         /* don't run the task if already down */
5982         if (test_bit(__E1000_DOWN, &adapter->state)) {
5983                 rtnl_unlock();
5984                 return;
5985         }
5986
5987         if (!(adapter->flags & FLAG_RESTART_NOW)) {
5988                 e1000e_dump(adapter);
5989                 e_err("Reset adapter unexpectedly\n");
5990         }
5991         e1000e_reinit_locked(adapter);
5992         rtnl_unlock();
5993 }
5994
5995 /**
5996  * e1000e_get_stats64 - Get System Network Statistics
5997  * @netdev: network interface device structure
5998  * @stats: rtnl_link_stats64 pointer
5999  *
6000  * Returns the address of the device statistics structure.
6001  **/
6002 void e1000e_get_stats64(struct net_device *netdev,
6003                         struct rtnl_link_stats64 *stats)
6004 {
6005         struct e1000_adapter *adapter = netdev_priv(netdev);
6006
6007         spin_lock(&adapter->stats64_lock);
6008         e1000e_update_stats(adapter);
6009         /* Fill out the OS statistics structure */
6010         stats->rx_bytes = adapter->stats.gorc;
6011         stats->rx_packets = adapter->stats.gprc;
6012         stats->tx_bytes = adapter->stats.gotc;
6013         stats->tx_packets = adapter->stats.gptc;
6014         stats->multicast = adapter->stats.mprc;
6015         stats->collisions = adapter->stats.colc;
6016
6017         /* Rx Errors */
6018
6019         /* RLEC on some newer hardware can be incorrect so build
6020          * our own version based on RUC and ROC
6021          */
6022         stats->rx_errors = adapter->stats.rxerrc +
6023             adapter->stats.crcerrs + adapter->stats.algnerrc +
6024             adapter->stats.ruc + adapter->stats.roc + adapter->stats.cexterr;
6025         stats->rx_length_errors = adapter->stats.ruc + adapter->stats.roc;
6026         stats->rx_crc_errors = adapter->stats.crcerrs;
6027         stats->rx_frame_errors = adapter->stats.algnerrc;
6028         stats->rx_missed_errors = adapter->stats.mpc;
6029
6030         /* Tx Errors */
6031         stats->tx_errors = adapter->stats.ecol + adapter->stats.latecol;
6032         stats->tx_aborted_errors = adapter->stats.ecol;
6033         stats->tx_window_errors = adapter->stats.latecol;
6034         stats->tx_carrier_errors = adapter->stats.tncrs;
6035
6036         /* Tx Dropped needs to be maintained elsewhere */
6037
6038         spin_unlock(&adapter->stats64_lock);
6039 }
6040
6041 /**
6042  * e1000_change_mtu - Change the Maximum Transfer Unit
6043  * @netdev: network interface device structure
6044  * @new_mtu: new value for maximum frame size
6045  *
6046  * Returns 0 on success, negative on failure
6047  **/
6048 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
6049 {
6050         struct e1000_adapter *adapter = netdev_priv(netdev);
6051         int max_frame = new_mtu + VLAN_ETH_HLEN + ETH_FCS_LEN;
6052
6053         /* Jumbo frame support */
6054         if ((new_mtu > ETH_DATA_LEN) &&
6055             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
6056                 e_err("Jumbo Frames not supported.\n");
6057                 return -EINVAL;
6058         }
6059
6060         /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
6061         if ((adapter->hw.mac.type >= e1000_pch2lan) &&
6062             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
6063             (new_mtu > ETH_DATA_LEN)) {
6064                 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
6065                 return -EINVAL;
6066         }
6067
6068         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
6069                 usleep_range(1000, 1100);
6070         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
6071         adapter->max_frame_size = max_frame;
6072         netdev_dbg(netdev, "changing MTU from %d to %d\n",
6073                    netdev->mtu, new_mtu);
6074         netdev->mtu = new_mtu;
6075
6076         pm_runtime_get_sync(netdev->dev.parent);
6077
6078         if (netif_running(netdev))
6079                 e1000e_down(adapter, true);
6080
6081         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
6082          * means we reserve 2 more, this pushes us to allocate from the next
6083          * larger slab size.
6084          * i.e. RXBUFFER_2048 --> size-4096 slab
6085          * However with the new *_jumbo_rx* routines, jumbo receives will use
6086          * fragmented skbs
6087          */
6088
6089         if (max_frame <= 2048)
6090                 adapter->rx_buffer_len = 2048;
6091         else
6092                 adapter->rx_buffer_len = 4096;
6093
6094         /* adjust allocation if LPE protects us, and we aren't using SBP */
6095         if (max_frame <= (VLAN_ETH_FRAME_LEN + ETH_FCS_LEN))
6096                 adapter->rx_buffer_len = VLAN_ETH_FRAME_LEN + ETH_FCS_LEN;
6097
6098         if (netif_running(netdev))
6099                 e1000e_up(adapter);
6100         else
6101                 e1000e_reset(adapter);
6102
6103         pm_runtime_put_sync(netdev->dev.parent);
6104
6105         clear_bit(__E1000_RESETTING, &adapter->state);
6106
6107         return 0;
6108 }
6109
6110 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
6111                            int cmd)
6112 {
6113         struct e1000_adapter *adapter = netdev_priv(netdev);
6114         struct mii_ioctl_data *data = if_mii(ifr);
6115
6116         if (adapter->hw.phy.media_type != e1000_media_type_copper)
6117                 return -EOPNOTSUPP;
6118
6119         switch (cmd) {
6120         case SIOCGMIIPHY:
6121                 data->phy_id = adapter->hw.phy.addr;
6122                 break;
6123         case SIOCGMIIREG:
6124                 e1000_phy_read_status(adapter);
6125
6126                 switch (data->reg_num & 0x1F) {
6127                 case MII_BMCR:
6128                         data->val_out = adapter->phy_regs.bmcr;
6129                         break;
6130                 case MII_BMSR:
6131                         data->val_out = adapter->phy_regs.bmsr;
6132                         break;
6133                 case MII_PHYSID1:
6134                         data->val_out = (adapter->hw.phy.id >> 16);
6135                         break;
6136                 case MII_PHYSID2:
6137                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
6138                         break;
6139                 case MII_ADVERTISE:
6140                         data->val_out = adapter->phy_regs.advertise;
6141                         break;
6142                 case MII_LPA:
6143                         data->val_out = adapter->phy_regs.lpa;
6144                         break;
6145                 case MII_EXPANSION:
6146                         data->val_out = adapter->phy_regs.expansion;
6147                         break;
6148                 case MII_CTRL1000:
6149                         data->val_out = adapter->phy_regs.ctrl1000;
6150                         break;
6151                 case MII_STAT1000:
6152                         data->val_out = adapter->phy_regs.stat1000;
6153                         break;
6154                 case MII_ESTATUS:
6155                         data->val_out = adapter->phy_regs.estatus;
6156                         break;
6157                 default:
6158                         return -EIO;
6159                 }
6160                 break;
6161         case SIOCSMIIREG:
6162         default:
6163                 return -EOPNOTSUPP;
6164         }
6165         return 0;
6166 }
6167
6168 /**
6169  * e1000e_hwtstamp_set - control hardware time stamping
6170  * @netdev: network interface device structure
6171  * @ifr: interface request
6172  *
6173  * Outgoing time stamping can be enabled and disabled. Play nice and
6174  * disable it when requested, although it shouldn't cause any overhead
6175  * when no packet needs it. At most one packet in the queue may be
6176  * marked for time stamping, otherwise it would be impossible to tell
6177  * for sure to which packet the hardware time stamp belongs.
6178  *
6179  * Incoming time stamping has to be configured via the hardware filters.
6180  * Not all combinations are supported, in particular event type has to be
6181  * specified. Matching the kind of event packet is not supported, with the
6182  * exception of "all V2 events regardless of level 2 or 4".
6183  **/
6184 static int e1000e_hwtstamp_set(struct net_device *netdev, struct ifreq *ifr)
6185 {
6186         struct e1000_adapter *adapter = netdev_priv(netdev);
6187         struct hwtstamp_config config;
6188         int ret_val;
6189
6190         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6191                 return -EFAULT;
6192
6193         ret_val = e1000e_config_hwtstamp(adapter, &config);
6194         if (ret_val)
6195                 return ret_val;
6196
6197         switch (config.rx_filter) {
6198         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6199         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6200         case HWTSTAMP_FILTER_PTP_V2_SYNC:
6201         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6202         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6203         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6204                 /* With V2 type filters which specify a Sync or Delay Request,
6205                  * Path Delay Request/Response messages are also time stamped
6206                  * by hardware so notify the caller the requested packets plus
6207                  * some others are time stamped.
6208                  */
6209                 config.rx_filter = HWTSTAMP_FILTER_SOME;
6210                 break;
6211         default:
6212                 break;
6213         }
6214
6215         return copy_to_user(ifr->ifr_data, &config,
6216                             sizeof(config)) ? -EFAULT : 0;
6217 }
6218
6219 static int e1000e_hwtstamp_get(struct net_device *netdev, struct ifreq *ifr)
6220 {
6221         struct e1000_adapter *adapter = netdev_priv(netdev);
6222
6223         return copy_to_user(ifr->ifr_data, &adapter->hwtstamp_config,
6224                             sizeof(adapter->hwtstamp_config)) ? -EFAULT : 0;
6225 }
6226
6227 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6228 {
6229         switch (cmd) {
6230         case SIOCGMIIPHY:
6231         case SIOCGMIIREG:
6232         case SIOCSMIIREG:
6233                 return e1000_mii_ioctl(netdev, ifr, cmd);
6234         case SIOCSHWTSTAMP:
6235                 return e1000e_hwtstamp_set(netdev, ifr);
6236         case SIOCGHWTSTAMP:
6237                 return e1000e_hwtstamp_get(netdev, ifr);
6238         default:
6239                 return -EOPNOTSUPP;
6240         }
6241 }
6242
6243 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
6244 {
6245         struct e1000_hw *hw = &adapter->hw;
6246         u32 i, mac_reg, wuc;
6247         u16 phy_reg, wuc_enable;
6248         int retval;
6249
6250         /* copy MAC RARs to PHY RARs */
6251         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
6252
6253         retval = hw->phy.ops.acquire(hw);
6254         if (retval) {
6255                 e_err("Could not acquire PHY\n");
6256                 return retval;
6257         }
6258
6259         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
6260         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6261         if (retval)
6262                 goto release;
6263
6264         /* copy MAC MTA to PHY MTA - only needed for pchlan */
6265         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
6266                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
6267                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
6268                                            (u16)(mac_reg & 0xFFFF));
6269                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
6270                                            (u16)((mac_reg >> 16) & 0xFFFF));
6271         }
6272
6273         /* configure PHY Rx Control register */
6274         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
6275         mac_reg = er32(RCTL);
6276         if (mac_reg & E1000_RCTL_UPE)
6277                 phy_reg |= BM_RCTL_UPE;
6278         if (mac_reg & E1000_RCTL_MPE)
6279                 phy_reg |= BM_RCTL_MPE;
6280         phy_reg &= ~(BM_RCTL_MO_MASK);
6281         if (mac_reg & E1000_RCTL_MO_3)
6282                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
6283                             << BM_RCTL_MO_SHIFT);
6284         if (mac_reg & E1000_RCTL_BAM)
6285                 phy_reg |= BM_RCTL_BAM;
6286         if (mac_reg & E1000_RCTL_PMCF)
6287                 phy_reg |= BM_RCTL_PMCF;
6288         mac_reg = er32(CTRL);
6289         if (mac_reg & E1000_CTRL_RFCE)
6290                 phy_reg |= BM_RCTL_RFCE;
6291         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
6292
6293         wuc = E1000_WUC_PME_EN;
6294         if (wufc & (E1000_WUFC_MAG | E1000_WUFC_LNKC))
6295                 wuc |= E1000_WUC_APME;
6296
6297         /* enable PHY wakeup in MAC register */
6298         ew32(WUFC, wufc);
6299         ew32(WUC, (E1000_WUC_PHY_WAKE | E1000_WUC_APMPME |
6300                    E1000_WUC_PME_STATUS | wuc));
6301
6302         /* configure and enable PHY wakeup in PHY registers */
6303         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
6304         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, wuc);
6305
6306         /* activate PHY wakeup */
6307         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
6308         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
6309         if (retval)
6310                 e_err("Could not set PHY Host Wakeup bit\n");
6311 release:
6312         hw->phy.ops.release(hw);
6313
6314         return retval;
6315 }
6316
6317 static void e1000e_flush_lpic(struct pci_dev *pdev)
6318 {
6319         struct net_device *netdev = pci_get_drvdata(pdev);
6320         struct e1000_adapter *adapter = netdev_priv(netdev);
6321         struct e1000_hw *hw = &adapter->hw;
6322         u32 ret_val;
6323
6324         pm_runtime_get_sync(netdev->dev.parent);
6325
6326         ret_val = hw->phy.ops.acquire(hw);
6327         if (ret_val)
6328                 goto fl_out;
6329
6330         pr_info("EEE TX LPI TIMER: %08X\n",
6331                 er32(LPIC) >> E1000_LPIC_LPIET_SHIFT);
6332
6333         hw->phy.ops.release(hw);
6334
6335 fl_out:
6336         pm_runtime_put_sync(netdev->dev.parent);
6337 }
6338
6339 /* S0ix implementation */
6340 static void e1000e_s0ix_entry_flow(struct e1000_adapter *adapter)
6341 {
6342         struct e1000_hw *hw = &adapter->hw;
6343         u32 mac_data;
6344         u16 phy_data;
6345
6346         /* Disable the periodic inband message,
6347          * don't request PCIe clock in K1 page770_17[10:9] = 10b
6348          */
6349         e1e_rphy(hw, HV_PM_CTRL, &phy_data);
6350         phy_data &= ~HV_PM_CTRL_K1_CLK_REQ;
6351         phy_data |= BIT(10);
6352         e1e_wphy(hw, HV_PM_CTRL, phy_data);
6353
6354         /* Make sure we don't exit K1 every time a new packet arrives
6355          * 772_29[5] = 1 CS_Mode_Stay_In_K1
6356          */
6357         e1e_rphy(hw, I217_CGFREG, &phy_data);
6358         phy_data |= BIT(5);
6359         e1e_wphy(hw, I217_CGFREG, phy_data);
6360
6361         /* Change the MAC/PHY interface to SMBus
6362          * Force the SMBus in PHY page769_23[0] = 1
6363          * Force the SMBus in MAC CTRL_EXT[11] = 1
6364          */
6365         e1e_rphy(hw, CV_SMB_CTRL, &phy_data);
6366         phy_data |= CV_SMB_CTRL_FORCE_SMBUS;
6367         e1e_wphy(hw, CV_SMB_CTRL, phy_data);
6368         mac_data = er32(CTRL_EXT);
6369         mac_data |= E1000_CTRL_EXT_FORCE_SMBUS;
6370         ew32(CTRL_EXT, mac_data);
6371
6372         /* DFT control: PHY bit: page769_20[0] = 1
6373          * Gate PPW via EXTCNF_CTRL - set 0x0F00[7] = 1
6374          */
6375         e1e_rphy(hw, I82579_DFT_CTRL, &phy_data);
6376         phy_data |= BIT(0);
6377         e1e_wphy(hw, I82579_DFT_CTRL, phy_data);
6378
6379         mac_data = er32(EXTCNF_CTRL);
6380         mac_data |= E1000_EXTCNF_CTRL_GATE_PHY_CFG;
6381         ew32(EXTCNF_CTRL, mac_data);
6382
6383         /* Check MAC Tx/Rx packet buffer pointers.
6384          * Reset MAC Tx/Rx packet buffer pointers to suppress any
6385          * pending traffic indication that would prevent power gating.
6386          */
6387         mac_data = er32(TDFH);
6388         if (mac_data)
6389                 ew32(TDFH, 0);
6390         mac_data = er32(TDFT);
6391         if (mac_data)
6392                 ew32(TDFT, 0);
6393         mac_data = er32(TDFHS);
6394         if (mac_data)
6395                 ew32(TDFHS, 0);
6396         mac_data = er32(TDFTS);
6397         if (mac_data)
6398                 ew32(TDFTS, 0);
6399         mac_data = er32(TDFPC);
6400         if (mac_data)
6401                 ew32(TDFPC, 0);
6402         mac_data = er32(RDFH);
6403         if (mac_data)
6404                 ew32(RDFH, 0);
6405         mac_data = er32(RDFT);
6406         if (mac_data)
6407                 ew32(RDFT, 0);
6408         mac_data = er32(RDFHS);
6409         if (mac_data)
6410                 ew32(RDFHS, 0);
6411         mac_data = er32(RDFTS);
6412         if (mac_data)
6413                 ew32(RDFTS, 0);
6414         mac_data = er32(RDFPC);
6415         if (mac_data)
6416                 ew32(RDFPC, 0);
6417
6418         /* Enable the Dynamic Power Gating in the MAC */
6419         mac_data = er32(FEXTNVM7);
6420         mac_data |= BIT(22);
6421         ew32(FEXTNVM7, mac_data);
6422
6423         /* Disable the time synchronization clock */
6424         mac_data = er32(FEXTNVM7);
6425         mac_data |= BIT(31);
6426         mac_data &= ~BIT(0);
6427         ew32(FEXTNVM7, mac_data);
6428
6429         /* Dynamic Power Gating Enable */
6430         mac_data = er32(CTRL_EXT);
6431         mac_data |= BIT(3);
6432         ew32(CTRL_EXT, mac_data);
6433
6434         /* Disable disconnected cable conditioning for Power Gating */
6435         mac_data = er32(DPGFR);
6436         mac_data |= BIT(2);
6437         ew32(DPGFR, mac_data);
6438
6439         /* Don't wake from dynamic Power Gating with clock request */
6440         mac_data = er32(FEXTNVM12);
6441         mac_data |= BIT(12);
6442         ew32(FEXTNVM12, mac_data);
6443
6444         /* Ungate PGCB clock */
6445         mac_data = er32(FEXTNVM9);
6446         mac_data &= ~BIT(28);
6447         ew32(FEXTNVM9, mac_data);
6448
6449         /* Enable K1 off to enable mPHY Power Gating */
6450         mac_data = er32(FEXTNVM6);
6451         mac_data |= BIT(31);
6452         ew32(FEXTNVM6, mac_data);
6453
6454         /* Enable mPHY power gating for any link and speed */
6455         mac_data = er32(FEXTNVM8);
6456         mac_data |= BIT(9);
6457         ew32(FEXTNVM8, mac_data);
6458
6459         /* Enable the Dynamic Clock Gating in the DMA and MAC */
6460         mac_data = er32(CTRL_EXT);
6461         mac_data |= E1000_CTRL_EXT_DMA_DYN_CLK_EN;
6462         ew32(CTRL_EXT, mac_data);
6463
6464         /* No MAC DPG gating SLP_S0 in modern standby
6465          * Switch the logic of the lanphypc to use PMC counter
6466          */
6467         mac_data = er32(FEXTNVM5);
6468         mac_data |= BIT(7);
6469         ew32(FEXTNVM5, mac_data);
6470 }
6471
6472 static void e1000e_s0ix_exit_flow(struct e1000_adapter *adapter)
6473 {
6474         struct e1000_hw *hw = &adapter->hw;
6475         u32 mac_data;
6476         u16 phy_data;
6477
6478         /* Disable the Dynamic Power Gating in the MAC */
6479         mac_data = er32(FEXTNVM7);
6480         mac_data &= 0xFFBFFFFF;
6481         ew32(FEXTNVM7, mac_data);
6482
6483         /* Enable the time synchronization clock */
6484         mac_data = er32(FEXTNVM7);
6485         mac_data |= BIT(0);
6486         ew32(FEXTNVM7, mac_data);
6487
6488         /* Disable mPHY power gating for any link and speed */
6489         mac_data = er32(FEXTNVM8);
6490         mac_data &= ~BIT(9);
6491         ew32(FEXTNVM8, mac_data);
6492
6493         /* Disable K1 off */
6494         mac_data = er32(FEXTNVM6);
6495         mac_data &= ~BIT(31);
6496         ew32(FEXTNVM6, mac_data);
6497
6498         /* Disable Ungate PGCB clock */
6499         mac_data = er32(FEXTNVM9);
6500         mac_data |= BIT(28);
6501         ew32(FEXTNVM9, mac_data);
6502
6503         /* Cancel not waking from dynamic
6504          * Power Gating with clock request
6505          */
6506         mac_data = er32(FEXTNVM12);
6507         mac_data &= ~BIT(12);
6508         ew32(FEXTNVM12, mac_data);
6509
6510         /* Cancel disable disconnected cable conditioning
6511          * for Power Gating
6512          */
6513         mac_data = er32(DPGFR);
6514         mac_data &= ~BIT(2);
6515         ew32(DPGFR, mac_data);
6516
6517         /* Disable Dynamic Power Gating */
6518         mac_data = er32(CTRL_EXT);
6519         mac_data &= 0xFFFFFFF7;
6520         ew32(CTRL_EXT, mac_data);
6521
6522         /* Disable the Dynamic Clock Gating in the DMA and MAC */
6523         mac_data = er32(CTRL_EXT);
6524         mac_data &= 0xFFF7FFFF;
6525         ew32(CTRL_EXT, mac_data);
6526
6527         /* Revert the lanphypc logic to use the internal Gbe counter
6528          * and not the PMC counter
6529          */
6530         mac_data = er32(FEXTNVM5);
6531         mac_data &= 0xFFFFFF7F;
6532         ew32(FEXTNVM5, mac_data);
6533
6534         /* Enable the periodic inband message,
6535          * Request PCIe clock in K1 page770_17[10:9] =01b
6536          */
6537         e1e_rphy(hw, HV_PM_CTRL, &phy_data);
6538         phy_data &= 0xFBFF;
6539         phy_data |= HV_PM_CTRL_K1_CLK_REQ;
6540         e1e_wphy(hw, HV_PM_CTRL, phy_data);
6541
6542         /* Return back configuration
6543          * 772_29[5] = 0 CS_Mode_Stay_In_K1
6544          */
6545         e1e_rphy(hw, I217_CGFREG, &phy_data);
6546         phy_data &= 0xFFDF;
6547         e1e_wphy(hw, I217_CGFREG, phy_data);
6548
6549         /* Change the MAC/PHY interface to Kumeran
6550          * Unforce the SMBus in PHY page769_23[0] = 0
6551          * Unforce the SMBus in MAC CTRL_EXT[11] = 0
6552          */
6553         e1e_rphy(hw, CV_SMB_CTRL, &phy_data);
6554         phy_data &= ~CV_SMB_CTRL_FORCE_SMBUS;
6555         e1e_wphy(hw, CV_SMB_CTRL, phy_data);
6556         mac_data = er32(CTRL_EXT);
6557         mac_data &= ~E1000_CTRL_EXT_FORCE_SMBUS;
6558         ew32(CTRL_EXT, mac_data);
6559 }
6560
6561 static int e1000e_pm_freeze(struct device *dev)
6562 {
6563         struct net_device *netdev = dev_get_drvdata(dev);
6564         struct e1000_adapter *adapter = netdev_priv(netdev);
6565         bool present;
6566
6567         rtnl_lock();
6568
6569         present = netif_device_present(netdev);
6570         netif_device_detach(netdev);
6571
6572         if (present && netif_running(netdev)) {
6573                 int count = E1000_CHECK_RESET_COUNT;
6574
6575                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
6576                         usleep_range(10000, 11000);
6577
6578                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
6579
6580                 /* Quiesce the device without resetting the hardware */
6581                 e1000e_down(adapter, false);
6582                 e1000_free_irq(adapter);
6583         }
6584         rtnl_unlock();
6585
6586         e1000e_reset_interrupt_capability(adapter);
6587
6588         /* Allow time for pending master requests to run */
6589         e1000e_disable_pcie_master(&adapter->hw);
6590
6591         return 0;
6592 }
6593
6594 static int __e1000_shutdown(struct pci_dev *pdev, bool runtime)
6595 {
6596         struct net_device *netdev = pci_get_drvdata(pdev);
6597         struct e1000_adapter *adapter = netdev_priv(netdev);
6598         struct e1000_hw *hw = &adapter->hw;
6599         u32 ctrl, ctrl_ext, rctl, status, wufc;
6600         int retval = 0;
6601
6602         /* Runtime suspend should only enable wakeup for link changes */
6603         if (runtime)
6604                 wufc = E1000_WUFC_LNKC;
6605         else if (device_may_wakeup(&pdev->dev))
6606                 wufc = adapter->wol;
6607         else
6608                 wufc = 0;
6609
6610         status = er32(STATUS);
6611         if (status & E1000_STATUS_LU)
6612                 wufc &= ~E1000_WUFC_LNKC;
6613
6614         if (wufc) {
6615                 e1000_setup_rctl(adapter);
6616                 e1000e_set_rx_mode(netdev);
6617
6618                 /* turn on all-multi mode if wake on multicast is enabled */
6619                 if (wufc & E1000_WUFC_MC) {
6620                         rctl = er32(RCTL);
6621                         rctl |= E1000_RCTL_MPE;
6622                         ew32(RCTL, rctl);
6623                 }
6624
6625                 ctrl = er32(CTRL);
6626                 ctrl |= E1000_CTRL_ADVD3WUC;
6627                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
6628                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
6629                 ew32(CTRL, ctrl);
6630
6631                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
6632                     adapter->hw.phy.media_type ==
6633                     e1000_media_type_internal_serdes) {
6634                         /* keep the laser running in D3 */
6635                         ctrl_ext = er32(CTRL_EXT);
6636                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
6637                         ew32(CTRL_EXT, ctrl_ext);
6638                 }
6639
6640                 if (!runtime)
6641                         e1000e_power_up_phy(adapter);
6642
6643                 if (adapter->flags & FLAG_IS_ICH)
6644                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
6645
6646                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6647                         /* enable wakeup by the PHY */
6648                         retval = e1000_init_phy_wakeup(adapter, wufc);
6649                         if (retval)
6650                                 return retval;
6651                 } else {
6652                         /* enable wakeup by the MAC */
6653                         ew32(WUFC, wufc);
6654                         ew32(WUC, E1000_WUC_PME_EN);
6655                 }
6656         } else {
6657                 ew32(WUC, 0);
6658                 ew32(WUFC, 0);
6659
6660                 e1000_power_down_phy(adapter);
6661         }
6662
6663         if (adapter->hw.phy.type == e1000_phy_igp_3) {
6664                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
6665         } else if (hw->mac.type >= e1000_pch_lpt) {
6666                 if (wufc && !(wufc & (E1000_WUFC_EX | E1000_WUFC_MC | E1000_WUFC_BC)))
6667                         /* ULP does not support wake from unicast, multicast
6668                          * or broadcast.
6669                          */
6670                         retval = e1000_enable_ulp_lpt_lp(hw, !runtime);
6671
6672                 if (retval)
6673                         return retval;
6674         }
6675
6676         /* Ensure that the appropriate bits are set in LPI_CTRL
6677          * for EEE in Sx
6678          */
6679         if ((hw->phy.type >= e1000_phy_i217) &&
6680             adapter->eee_advert && hw->dev_spec.ich8lan.eee_lp_ability) {
6681                 u16 lpi_ctrl = 0;
6682
6683                 retval = hw->phy.ops.acquire(hw);
6684                 if (!retval) {
6685                         retval = e1e_rphy_locked(hw, I82579_LPI_CTRL,
6686                                                  &lpi_ctrl);
6687                         if (!retval) {
6688                                 if (adapter->eee_advert &
6689                                     hw->dev_spec.ich8lan.eee_lp_ability &
6690                                     I82579_EEE_100_SUPPORTED)
6691                                         lpi_ctrl |= I82579_LPI_CTRL_100_ENABLE;
6692                                 if (adapter->eee_advert &
6693                                     hw->dev_spec.ich8lan.eee_lp_ability &
6694                                     I82579_EEE_1000_SUPPORTED)
6695                                         lpi_ctrl |= I82579_LPI_CTRL_1000_ENABLE;
6696
6697                                 retval = e1e_wphy_locked(hw, I82579_LPI_CTRL,
6698                                                          lpi_ctrl);
6699                         }
6700                 }
6701                 hw->phy.ops.release(hw);
6702         }
6703
6704         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
6705          * would have already happened in close and is redundant.
6706          */
6707         e1000e_release_hw_control(adapter);
6708
6709         pci_clear_master(pdev);
6710
6711         /* The pci-e switch on some quad port adapters will report a
6712          * correctable error when the MAC transitions from D0 to D3.  To
6713          * prevent this we need to mask off the correctable errors on the
6714          * downstream port of the pci-e switch.
6715          *
6716          * We don't have the associated upstream bridge while assigning
6717          * the PCI device into guest. For example, the KVM on power is
6718          * one of the cases.
6719          */
6720         if (adapter->flags & FLAG_IS_QUAD_PORT) {
6721                 struct pci_dev *us_dev = pdev->bus->self;
6722                 u16 devctl;
6723
6724                 if (!us_dev)
6725                         return 0;
6726
6727                 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
6728                 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
6729                                            (devctl & ~PCI_EXP_DEVCTL_CERE));
6730
6731                 pci_save_state(pdev);
6732                 pci_prepare_to_sleep(pdev);
6733
6734                 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
6735         }
6736
6737         return 0;
6738 }
6739
6740 /**
6741  * __e1000e_disable_aspm - Disable ASPM states
6742  * @pdev: pointer to PCI device struct
6743  * @state: bit-mask of ASPM states to disable
6744  * @locked: indication if this context holds pci_bus_sem locked.
6745  *
6746  * Some devices *must* have certain ASPM states disabled per hardware errata.
6747  **/
6748 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state, int locked)
6749 {
6750         struct pci_dev *parent = pdev->bus->self;
6751         u16 aspm_dis_mask = 0;
6752         u16 pdev_aspmc, parent_aspmc;
6753
6754         switch (state) {
6755         case PCIE_LINK_STATE_L0S:
6756         case PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1:
6757                 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L0S;
6758                 fallthrough; /* can't have L1 without L0s */
6759         case PCIE_LINK_STATE_L1:
6760                 aspm_dis_mask |= PCI_EXP_LNKCTL_ASPM_L1;
6761                 break;
6762         default:
6763                 return;
6764         }
6765
6766         pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6767         pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6768
6769         if (parent) {
6770                 pcie_capability_read_word(parent, PCI_EXP_LNKCTL,
6771                                           &parent_aspmc);
6772                 parent_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6773         }
6774
6775         /* Nothing to do if the ASPM states to be disabled already are */
6776         if (!(pdev_aspmc & aspm_dis_mask) &&
6777             (!parent || !(parent_aspmc & aspm_dis_mask)))
6778                 return;
6779
6780         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
6781                  (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L0S) ?
6782                  "L0s" : "",
6783                  (aspm_dis_mask & pdev_aspmc & PCI_EXP_LNKCTL_ASPM_L1) ?
6784                  "L1" : "");
6785
6786 #ifdef CONFIG_PCIEASPM
6787         if (locked)
6788                 pci_disable_link_state_locked(pdev, state);
6789         else
6790                 pci_disable_link_state(pdev, state);
6791
6792         /* Double-check ASPM control.  If not disabled by the above, the
6793          * BIOS is preventing that from happening (or CONFIG_PCIEASPM is
6794          * not enabled); override by writing PCI config space directly.
6795          */
6796         pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &pdev_aspmc);
6797         pdev_aspmc &= PCI_EXP_LNKCTL_ASPMC;
6798
6799         if (!(aspm_dis_mask & pdev_aspmc))
6800                 return;
6801 #endif
6802
6803         /* Both device and parent should have the same ASPM setting.
6804          * Disable ASPM in downstream component first and then upstream.
6805          */
6806         pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, aspm_dis_mask);
6807
6808         if (parent)
6809                 pcie_capability_clear_word(parent, PCI_EXP_LNKCTL,
6810                                            aspm_dis_mask);
6811 }
6812
6813 /**
6814  * e1000e_disable_aspm - Disable ASPM states.
6815  * @pdev: pointer to PCI device struct
6816  * @state: bit-mask of ASPM states to disable
6817  *
6818  * This function acquires the pci_bus_sem!
6819  * Some devices *must* have certain ASPM states disabled per hardware errata.
6820  **/
6821 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
6822 {
6823         __e1000e_disable_aspm(pdev, state, 0);
6824 }
6825
6826 /**
6827  * e1000e_disable_aspm_locked - Disable ASPM states.
6828  * @pdev: pointer to PCI device struct
6829  * @state: bit-mask of ASPM states to disable
6830  *
6831  * This function must be called with pci_bus_sem acquired!
6832  * Some devices *must* have certain ASPM states disabled per hardware errata.
6833  **/
6834 static void e1000e_disable_aspm_locked(struct pci_dev *pdev, u16 state)
6835 {
6836         __e1000e_disable_aspm(pdev, state, 1);
6837 }
6838
6839 static int e1000e_pm_thaw(struct device *dev)
6840 {
6841         struct net_device *netdev = dev_get_drvdata(dev);
6842         struct e1000_adapter *adapter = netdev_priv(netdev);
6843         int rc = 0;
6844
6845         e1000e_set_interrupt_capability(adapter);
6846
6847         rtnl_lock();
6848         if (netif_running(netdev)) {
6849                 rc = e1000_request_irq(adapter);
6850                 if (rc)
6851                         goto err_irq;
6852
6853                 e1000e_up(adapter);
6854         }
6855
6856         netif_device_attach(netdev);
6857 err_irq:
6858         rtnl_unlock();
6859
6860         return rc;
6861 }
6862
6863 static int __e1000_resume(struct pci_dev *pdev)
6864 {
6865         struct net_device *netdev = pci_get_drvdata(pdev);
6866         struct e1000_adapter *adapter = netdev_priv(netdev);
6867         struct e1000_hw *hw = &adapter->hw;
6868         u16 aspm_disable_flag = 0;
6869
6870         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
6871                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6872         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
6873                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6874         if (aspm_disable_flag)
6875                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6876
6877         pci_set_master(pdev);
6878
6879         if (hw->mac.type >= e1000_pch2lan)
6880                 e1000_resume_workarounds_pchlan(&adapter->hw);
6881
6882         e1000e_power_up_phy(adapter);
6883
6884         /* report the system wakeup cause from S3/S4 */
6885         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
6886                 u16 phy_data;
6887
6888                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
6889                 if (phy_data) {
6890                         e_info("PHY Wakeup cause - %s\n",
6891                                phy_data & E1000_WUS_EX ? "Unicast Packet" :
6892                                phy_data & E1000_WUS_MC ? "Multicast Packet" :
6893                                phy_data & E1000_WUS_BC ? "Broadcast Packet" :
6894                                phy_data & E1000_WUS_MAG ? "Magic Packet" :
6895                                phy_data & E1000_WUS_LNKC ?
6896                                "Link Status Change" : "other");
6897                 }
6898                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
6899         } else {
6900                 u32 wus = er32(WUS);
6901
6902                 if (wus) {
6903                         e_info("MAC Wakeup cause - %s\n",
6904                                wus & E1000_WUS_EX ? "Unicast Packet" :
6905                                wus & E1000_WUS_MC ? "Multicast Packet" :
6906                                wus & E1000_WUS_BC ? "Broadcast Packet" :
6907                                wus & E1000_WUS_MAG ? "Magic Packet" :
6908                                wus & E1000_WUS_LNKC ? "Link Status Change" :
6909                                "other");
6910                 }
6911                 ew32(WUS, ~0);
6912         }
6913
6914         e1000e_reset(adapter);
6915
6916         e1000_init_manageability_pt(adapter);
6917
6918         /* If the controller has AMT, do not set DRV_LOAD until the interface
6919          * is up.  For all other cases, let the f/w know that the h/w is now
6920          * under the control of the driver.
6921          */
6922         if (!(adapter->flags & FLAG_HAS_AMT))
6923                 e1000e_get_hw_control(adapter);
6924
6925         return 0;
6926 }
6927
6928 static __maybe_unused int e1000e_pm_prepare(struct device *dev)
6929 {
6930         return pm_runtime_suspended(dev) &&
6931                 pm_suspend_via_firmware();
6932 }
6933
6934 static __maybe_unused int e1000e_pm_suspend(struct device *dev)
6935 {
6936         struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6937         struct e1000_adapter *adapter = netdev_priv(netdev);
6938         struct pci_dev *pdev = to_pci_dev(dev);
6939         int rc;
6940
6941         e1000e_flush_lpic(pdev);
6942
6943         e1000e_pm_freeze(dev);
6944
6945         rc = __e1000_shutdown(pdev, false);
6946         if (rc) {
6947                 e1000e_pm_thaw(dev);
6948         } else {
6949                 /* Introduce S0ix implementation */
6950                 if (adapter->flags2 & FLAG2_ENABLE_S0IX_FLOWS)
6951                         e1000e_s0ix_entry_flow(adapter);
6952         }
6953
6954         return rc;
6955 }
6956
6957 static __maybe_unused int e1000e_pm_resume(struct device *dev)
6958 {
6959         struct net_device *netdev = pci_get_drvdata(to_pci_dev(dev));
6960         struct e1000_adapter *adapter = netdev_priv(netdev);
6961         struct pci_dev *pdev = to_pci_dev(dev);
6962         int rc;
6963
6964         /* Introduce S0ix implementation */
6965         if (adapter->flags2 & FLAG2_ENABLE_S0IX_FLOWS)
6966                 e1000e_s0ix_exit_flow(adapter);
6967
6968         rc = __e1000_resume(pdev);
6969         if (rc)
6970                 return rc;
6971
6972         return e1000e_pm_thaw(dev);
6973 }
6974
6975 static __maybe_unused int e1000e_pm_runtime_idle(struct device *dev)
6976 {
6977         struct net_device *netdev = dev_get_drvdata(dev);
6978         struct e1000_adapter *adapter = netdev_priv(netdev);
6979         u16 eee_lp;
6980
6981         eee_lp = adapter->hw.dev_spec.ich8lan.eee_lp_ability;
6982
6983         if (!e1000e_has_link(adapter)) {
6984                 adapter->hw.dev_spec.ich8lan.eee_lp_ability = eee_lp;
6985                 pm_schedule_suspend(dev, 5 * MSEC_PER_SEC);
6986         }
6987
6988         return -EBUSY;
6989 }
6990
6991 static __maybe_unused int e1000e_pm_runtime_resume(struct device *dev)
6992 {
6993         struct pci_dev *pdev = to_pci_dev(dev);
6994         struct net_device *netdev = pci_get_drvdata(pdev);
6995         struct e1000_adapter *adapter = netdev_priv(netdev);
6996         int rc;
6997
6998         rc = __e1000_resume(pdev);
6999         if (rc)
7000                 return rc;
7001
7002         if (netdev->flags & IFF_UP)
7003                 e1000e_up(adapter);
7004
7005         return rc;
7006 }
7007
7008 static __maybe_unused int e1000e_pm_runtime_suspend(struct device *dev)
7009 {
7010         struct pci_dev *pdev = to_pci_dev(dev);
7011         struct net_device *netdev = pci_get_drvdata(pdev);
7012         struct e1000_adapter *adapter = netdev_priv(netdev);
7013
7014         if (netdev->flags & IFF_UP) {
7015                 int count = E1000_CHECK_RESET_COUNT;
7016
7017                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
7018                         usleep_range(10000, 11000);
7019
7020                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
7021
7022                 /* Down the device without resetting the hardware */
7023                 e1000e_down(adapter, false);
7024         }
7025
7026         if (__e1000_shutdown(pdev, true)) {
7027                 e1000e_pm_runtime_resume(dev);
7028                 return -EBUSY;
7029         }
7030
7031         return 0;
7032 }
7033
7034 static void e1000_shutdown(struct pci_dev *pdev)
7035 {
7036         e1000e_flush_lpic(pdev);
7037
7038         e1000e_pm_freeze(&pdev->dev);
7039
7040         __e1000_shutdown(pdev, false);
7041 }
7042
7043 #ifdef CONFIG_NET_POLL_CONTROLLER
7044
7045 static irqreturn_t e1000_intr_msix(int __always_unused irq, void *data)
7046 {
7047         struct net_device *netdev = data;
7048         struct e1000_adapter *adapter = netdev_priv(netdev);
7049
7050         if (adapter->msix_entries) {
7051                 int vector, msix_irq;
7052
7053                 vector = 0;
7054                 msix_irq = adapter->msix_entries[vector].vector;
7055                 if (disable_hardirq(msix_irq))
7056                         e1000_intr_msix_rx(msix_irq, netdev);
7057                 enable_irq(msix_irq);
7058
7059                 vector++;
7060                 msix_irq = adapter->msix_entries[vector].vector;
7061                 if (disable_hardirq(msix_irq))
7062                         e1000_intr_msix_tx(msix_irq, netdev);
7063                 enable_irq(msix_irq);
7064
7065                 vector++;
7066                 msix_irq = adapter->msix_entries[vector].vector;
7067                 if (disable_hardirq(msix_irq))
7068                         e1000_msix_other(msix_irq, netdev);
7069                 enable_irq(msix_irq);
7070         }
7071
7072         return IRQ_HANDLED;
7073 }
7074
7075 /**
7076  * e1000_netpoll
7077  * @netdev: network interface device structure
7078  *
7079  * Polling 'interrupt' - used by things like netconsole to send skbs
7080  * without having to re-enable interrupts. It's not called while
7081  * the interrupt routine is executing.
7082  */
7083 static void e1000_netpoll(struct net_device *netdev)
7084 {
7085         struct e1000_adapter *adapter = netdev_priv(netdev);
7086
7087         switch (adapter->int_mode) {
7088         case E1000E_INT_MODE_MSIX:
7089                 e1000_intr_msix(adapter->pdev->irq, netdev);
7090                 break;
7091         case E1000E_INT_MODE_MSI:
7092                 if (disable_hardirq(adapter->pdev->irq))
7093                         e1000_intr_msi(adapter->pdev->irq, netdev);
7094                 enable_irq(adapter->pdev->irq);
7095                 break;
7096         default:                /* E1000E_INT_MODE_LEGACY */
7097                 if (disable_hardirq(adapter->pdev->irq))
7098                         e1000_intr(adapter->pdev->irq, netdev);
7099                 enable_irq(adapter->pdev->irq);
7100                 break;
7101         }
7102 }
7103 #endif
7104
7105 /**
7106  * e1000_io_error_detected - called when PCI error is detected
7107  * @pdev: Pointer to PCI device
7108  * @state: The current pci connection state
7109  *
7110  * This function is called after a PCI bus error affecting
7111  * this device has been detected.
7112  */
7113 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
7114                                                 pci_channel_state_t state)
7115 {
7116         e1000e_pm_freeze(&pdev->dev);
7117
7118         if (state == pci_channel_io_perm_failure)
7119                 return PCI_ERS_RESULT_DISCONNECT;
7120
7121         pci_disable_device(pdev);
7122
7123         /* Request a slot reset. */
7124         return PCI_ERS_RESULT_NEED_RESET;
7125 }
7126
7127 /**
7128  * e1000_io_slot_reset - called after the pci bus has been reset.
7129  * @pdev: Pointer to PCI device
7130  *
7131  * Restart the card from scratch, as if from a cold-boot. Implementation
7132  * resembles the first-half of the e1000e_pm_resume routine.
7133  */
7134 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
7135 {
7136         struct net_device *netdev = pci_get_drvdata(pdev);
7137         struct e1000_adapter *adapter = netdev_priv(netdev);
7138         struct e1000_hw *hw = &adapter->hw;
7139         u16 aspm_disable_flag = 0;
7140         int err;
7141         pci_ers_result_t result;
7142
7143         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
7144                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7145         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
7146                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7147         if (aspm_disable_flag)
7148                 e1000e_disable_aspm_locked(pdev, aspm_disable_flag);
7149
7150         err = pci_enable_device_mem(pdev);
7151         if (err) {
7152                 dev_err(&pdev->dev,
7153                         "Cannot re-enable PCI device after reset.\n");
7154                 result = PCI_ERS_RESULT_DISCONNECT;
7155         } else {
7156                 pdev->state_saved = true;
7157                 pci_restore_state(pdev);
7158                 pci_set_master(pdev);
7159
7160                 pci_enable_wake(pdev, PCI_D3hot, 0);
7161                 pci_enable_wake(pdev, PCI_D3cold, 0);
7162
7163                 e1000e_reset(adapter);
7164                 ew32(WUS, ~0);
7165                 result = PCI_ERS_RESULT_RECOVERED;
7166         }
7167
7168         return result;
7169 }
7170
7171 /**
7172  * e1000_io_resume - called when traffic can start flowing again.
7173  * @pdev: Pointer to PCI device
7174  *
7175  * This callback is called when the error recovery driver tells us that
7176  * its OK to resume normal operation. Implementation resembles the
7177  * second-half of the e1000e_pm_resume routine.
7178  */
7179 static void e1000_io_resume(struct pci_dev *pdev)
7180 {
7181         struct net_device *netdev = pci_get_drvdata(pdev);
7182         struct e1000_adapter *adapter = netdev_priv(netdev);
7183
7184         e1000_init_manageability_pt(adapter);
7185
7186         e1000e_pm_thaw(&pdev->dev);
7187
7188         /* If the controller has AMT, do not set DRV_LOAD until the interface
7189          * is up.  For all other cases, let the f/w know that the h/w is now
7190          * under the control of the driver.
7191          */
7192         if (!(adapter->flags & FLAG_HAS_AMT))
7193                 e1000e_get_hw_control(adapter);
7194 }
7195
7196 static void e1000_print_device_info(struct e1000_adapter *adapter)
7197 {
7198         struct e1000_hw *hw = &adapter->hw;
7199         struct net_device *netdev = adapter->netdev;
7200         u32 ret_val;
7201         u8 pba_str[E1000_PBANUM_LENGTH];
7202
7203         /* print bus type/speed/width info */
7204         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
7205                /* bus width */
7206                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
7207                 "Width x1"),
7208                /* MAC address */
7209                netdev->dev_addr);
7210         e_info("Intel(R) PRO/%s Network Connection\n",
7211                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
7212         ret_val = e1000_read_pba_string_generic(hw, pba_str,
7213                                                 E1000_PBANUM_LENGTH);
7214         if (ret_val)
7215                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
7216         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
7217                hw->mac.type, hw->phy.type, pba_str);
7218 }
7219
7220 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
7221 {
7222         struct e1000_hw *hw = &adapter->hw;
7223         int ret_val;
7224         u16 buf = 0;
7225
7226         if (hw->mac.type != e1000_82573)
7227                 return;
7228
7229         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
7230         le16_to_cpus(&buf);
7231         if (!ret_val && (!(buf & BIT(0)))) {
7232                 /* Deep Smart Power Down (DSPD) */
7233                 dev_warn(&adapter->pdev->dev,
7234                          "Warning: detected DSPD enabled in EEPROM\n");
7235         }
7236 }
7237
7238 static netdev_features_t e1000_fix_features(struct net_device *netdev,
7239                                             netdev_features_t features)
7240 {
7241         struct e1000_adapter *adapter = netdev_priv(netdev);
7242         struct e1000_hw *hw = &adapter->hw;
7243
7244         /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
7245         if ((hw->mac.type >= e1000_pch2lan) && (netdev->mtu > ETH_DATA_LEN))
7246                 features &= ~NETIF_F_RXFCS;
7247
7248         /* Since there is no support for separate Rx/Tx vlan accel
7249          * enable/disable make sure Tx flag is always in same state as Rx.
7250          */
7251         if (features & NETIF_F_HW_VLAN_CTAG_RX)
7252                 features |= NETIF_F_HW_VLAN_CTAG_TX;
7253         else
7254                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
7255
7256         return features;
7257 }
7258
7259 static int e1000_set_features(struct net_device *netdev,
7260                               netdev_features_t features)
7261 {
7262         struct e1000_adapter *adapter = netdev_priv(netdev);
7263         netdev_features_t changed = features ^ netdev->features;
7264
7265         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
7266                 adapter->flags |= FLAG_TSO_FORCE;
7267
7268         if (!(changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_TX |
7269                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
7270                          NETIF_F_RXALL)))
7271                 return 0;
7272
7273         if (changed & NETIF_F_RXFCS) {
7274                 if (features & NETIF_F_RXFCS) {
7275                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
7276                 } else {
7277                         /* We need to take it back to defaults, which might mean
7278                          * stripping is still disabled at the adapter level.
7279                          */
7280                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
7281                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
7282                         else
7283                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
7284                 }
7285         }
7286
7287         netdev->features = features;
7288
7289         if (netif_running(netdev))
7290                 e1000e_reinit_locked(adapter);
7291         else
7292                 e1000e_reset(adapter);
7293
7294         return 1;
7295 }
7296
7297 static const struct net_device_ops e1000e_netdev_ops = {
7298         .ndo_open               = e1000e_open,
7299         .ndo_stop               = e1000e_close,
7300         .ndo_start_xmit         = e1000_xmit_frame,
7301         .ndo_get_stats64        = e1000e_get_stats64,
7302         .ndo_set_rx_mode        = e1000e_set_rx_mode,
7303         .ndo_set_mac_address    = e1000_set_mac,
7304         .ndo_change_mtu         = e1000_change_mtu,
7305         .ndo_do_ioctl           = e1000_ioctl,
7306         .ndo_tx_timeout         = e1000_tx_timeout,
7307         .ndo_validate_addr      = eth_validate_addr,
7308
7309         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
7310         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
7311 #ifdef CONFIG_NET_POLL_CONTROLLER
7312         .ndo_poll_controller    = e1000_netpoll,
7313 #endif
7314         .ndo_set_features = e1000_set_features,
7315         .ndo_fix_features = e1000_fix_features,
7316         .ndo_features_check     = passthru_features_check,
7317 };
7318
7319 /**
7320  * e1000_probe - Device Initialization Routine
7321  * @pdev: PCI device information struct
7322  * @ent: entry in e1000_pci_tbl
7323  *
7324  * Returns 0 on success, negative on failure
7325  *
7326  * e1000_probe initializes an adapter identified by a pci_dev structure.
7327  * The OS initialization, configuring of the adapter private structure,
7328  * and a hardware reset occur.
7329  **/
7330 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7331 {
7332         struct net_device *netdev;
7333         struct e1000_adapter *adapter;
7334         struct e1000_hw *hw;
7335         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
7336         resource_size_t mmio_start, mmio_len;
7337         resource_size_t flash_start, flash_len;
7338         static int cards_found;
7339         u16 aspm_disable_flag = 0;
7340         int bars, i, err, pci_using_dac;
7341         u16 eeprom_data = 0;
7342         u16 eeprom_apme_mask = E1000_EEPROM_APME;
7343         s32 ret_val = 0;
7344
7345         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
7346                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
7347         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
7348                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
7349         if (aspm_disable_flag)
7350                 e1000e_disable_aspm(pdev, aspm_disable_flag);
7351
7352         err = pci_enable_device_mem(pdev);
7353         if (err)
7354                 return err;
7355
7356         pci_using_dac = 0;
7357         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
7358         if (!err) {
7359                 pci_using_dac = 1;
7360         } else {
7361                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
7362                 if (err) {
7363                         dev_err(&pdev->dev,
7364                                 "No usable DMA configuration, aborting\n");
7365                         goto err_dma;
7366                 }
7367         }
7368
7369         bars = pci_select_bars(pdev, IORESOURCE_MEM);
7370         err = pci_request_selected_regions_exclusive(pdev, bars,
7371                                                      e1000e_driver_name);
7372         if (err)
7373                 goto err_pci_reg;
7374
7375         /* AER (Advanced Error Reporting) hooks */
7376         pci_enable_pcie_error_reporting(pdev);
7377
7378         pci_set_master(pdev);
7379         /* PCI config space info */
7380         err = pci_save_state(pdev);
7381         if (err)
7382                 goto err_alloc_etherdev;
7383
7384         err = -ENOMEM;
7385         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
7386         if (!netdev)
7387                 goto err_alloc_etherdev;
7388
7389         SET_NETDEV_DEV(netdev, &pdev->dev);
7390
7391         netdev->irq = pdev->irq;
7392
7393         pci_set_drvdata(pdev, netdev);
7394         adapter = netdev_priv(netdev);
7395         hw = &adapter->hw;
7396         adapter->netdev = netdev;
7397         adapter->pdev = pdev;
7398         adapter->ei = ei;
7399         adapter->pba = ei->pba;
7400         adapter->flags = ei->flags;
7401         adapter->flags2 = ei->flags2;
7402         adapter->hw.adapter = adapter;
7403         adapter->hw.mac.type = ei->mac;
7404         adapter->max_hw_frame_size = ei->max_hw_frame_size;
7405         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7406
7407         mmio_start = pci_resource_start(pdev, 0);
7408         mmio_len = pci_resource_len(pdev, 0);
7409
7410         err = -EIO;
7411         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
7412         if (!adapter->hw.hw_addr)
7413                 goto err_ioremap;
7414
7415         if ((adapter->flags & FLAG_HAS_FLASH) &&
7416             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM) &&
7417             (hw->mac.type < e1000_pch_spt)) {
7418                 flash_start = pci_resource_start(pdev, 1);
7419                 flash_len = pci_resource_len(pdev, 1);
7420                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
7421                 if (!adapter->hw.flash_address)
7422                         goto err_flashmap;
7423         }
7424
7425         /* Set default EEE advertisement */
7426         if (adapter->flags2 & FLAG2_HAS_EEE)
7427                 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
7428
7429         /* construct the net_device struct */
7430         netdev->netdev_ops = &e1000e_netdev_ops;
7431         e1000e_set_ethtool_ops(netdev);
7432         netdev->watchdog_timeo = 5 * HZ;
7433         netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
7434         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
7435
7436         netdev->mem_start = mmio_start;
7437         netdev->mem_end = mmio_start + mmio_len;
7438
7439         adapter->bd_number = cards_found++;
7440
7441         e1000e_check_options(adapter);
7442
7443         /* setup adapter struct */
7444         err = e1000_sw_init(adapter);
7445         if (err)
7446                 goto err_sw_init;
7447
7448         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
7449         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
7450         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
7451
7452         err = ei->get_variants(adapter);
7453         if (err)
7454                 goto err_hw_init;
7455
7456         if ((adapter->flags & FLAG_IS_ICH) &&
7457             (adapter->flags & FLAG_READ_ONLY_NVM) &&
7458             (hw->mac.type < e1000_pch_spt))
7459                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
7460
7461         hw->mac.ops.get_bus_info(&adapter->hw);
7462
7463         adapter->hw.phy.autoneg_wait_to_complete = 0;
7464
7465         /* Copper options */
7466         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
7467                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
7468                 adapter->hw.phy.disable_polarity_correction = 0;
7469                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
7470         }
7471
7472         if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
7473                 dev_info(&pdev->dev,
7474                          "PHY reset is blocked due to SOL/IDER session.\n");
7475
7476         /* Set initial default active device features */
7477         netdev->features = (NETIF_F_SG |
7478                             NETIF_F_HW_VLAN_CTAG_RX |
7479                             NETIF_F_HW_VLAN_CTAG_TX |
7480                             NETIF_F_TSO |
7481                             NETIF_F_TSO6 |
7482                             NETIF_F_RXHASH |
7483                             NETIF_F_RXCSUM |
7484                             NETIF_F_HW_CSUM);
7485
7486         /* Set user-changeable features (subset of all device features) */
7487         netdev->hw_features = netdev->features;
7488         netdev->hw_features |= NETIF_F_RXFCS;
7489         netdev->priv_flags |= IFF_SUPP_NOFCS;
7490         netdev->hw_features |= NETIF_F_RXALL;
7491
7492         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
7493                 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
7494
7495         netdev->vlan_features |= (NETIF_F_SG |
7496                                   NETIF_F_TSO |
7497                                   NETIF_F_TSO6 |
7498                                   NETIF_F_HW_CSUM);
7499
7500         netdev->priv_flags |= IFF_UNICAST_FLT;
7501
7502         if (pci_using_dac) {
7503                 netdev->features |= NETIF_F_HIGHDMA;
7504                 netdev->vlan_features |= NETIF_F_HIGHDMA;
7505         }
7506
7507         /* MTU range: 68 - max_hw_frame_size */
7508         netdev->min_mtu = ETH_MIN_MTU;
7509         netdev->max_mtu = adapter->max_hw_frame_size -
7510                           (VLAN_ETH_HLEN + ETH_FCS_LEN);
7511
7512         if (e1000e_enable_mng_pass_thru(&adapter->hw))
7513                 adapter->flags |= FLAG_MNG_PT_ENABLED;
7514
7515         /* before reading the NVM, reset the controller to
7516          * put the device in a known good starting state
7517          */
7518         adapter->hw.mac.ops.reset_hw(&adapter->hw);
7519
7520         /* systems with ASPM and others may see the checksum fail on the first
7521          * attempt. Let's give it a few tries
7522          */
7523         for (i = 0;; i++) {
7524                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
7525                         break;
7526                 if (i == 2) {
7527                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
7528                         err = -EIO;
7529                         goto err_eeprom;
7530                 }
7531         }
7532
7533         e1000_eeprom_checks(adapter);
7534
7535         /* copy the MAC address */
7536         if (e1000e_read_mac_addr(&adapter->hw))
7537                 dev_err(&pdev->dev,
7538                         "NVM Read Error while reading MAC address\n");
7539
7540         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
7541
7542         if (!is_valid_ether_addr(netdev->dev_addr)) {
7543                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
7544                         netdev->dev_addr);
7545                 err = -EIO;
7546                 goto err_eeprom;
7547         }
7548
7549         timer_setup(&adapter->watchdog_timer, e1000_watchdog, 0);
7550         timer_setup(&adapter->phy_info_timer, e1000_update_phy_info, 0);
7551
7552         INIT_WORK(&adapter->reset_task, e1000_reset_task);
7553         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
7554         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
7555         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
7556         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
7557
7558         /* Initialize link parameters. User can change them with ethtool */
7559         adapter->hw.mac.autoneg = 1;
7560         adapter->fc_autoneg = true;
7561         adapter->hw.fc.requested_mode = e1000_fc_default;
7562         adapter->hw.fc.current_mode = e1000_fc_default;
7563         adapter->hw.phy.autoneg_advertised = 0x2f;
7564
7565         /* Initial Wake on LAN setting - If APM wake is enabled in
7566          * the EEPROM, enable the ACPI Magic Packet filter
7567          */
7568         if (adapter->flags & FLAG_APME_IN_WUC) {
7569                 /* APME bit in EEPROM is mapped to WUC.APME */
7570                 eeprom_data = er32(WUC);
7571                 eeprom_apme_mask = E1000_WUC_APME;
7572                 if ((hw->mac.type > e1000_ich10lan) &&
7573                     (eeprom_data & E1000_WUC_PHY_WAKE))
7574                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
7575         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
7576                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
7577                     (adapter->hw.bus.func == 1))
7578                         ret_val = e1000_read_nvm(&adapter->hw,
7579                                               NVM_INIT_CONTROL3_PORT_B,
7580                                               1, &eeprom_data);
7581                 else
7582                         ret_val = e1000_read_nvm(&adapter->hw,
7583                                               NVM_INIT_CONTROL3_PORT_A,
7584                                               1, &eeprom_data);
7585         }
7586
7587         /* fetch WoL from EEPROM */
7588         if (ret_val)
7589                 e_dbg("NVM read error getting WoL initial values: %d\n", ret_val);
7590         else if (eeprom_data & eeprom_apme_mask)
7591                 adapter->eeprom_wol |= E1000_WUFC_MAG;
7592
7593         /* now that we have the eeprom settings, apply the special cases
7594          * where the eeprom may be wrong or the board simply won't support
7595          * wake on lan on a particular port
7596          */
7597         if (!(adapter->flags & FLAG_HAS_WOL))
7598                 adapter->eeprom_wol = 0;
7599
7600         /* initialize the wol settings based on the eeprom settings */
7601         adapter->wol = adapter->eeprom_wol;
7602
7603         /* make sure adapter isn't asleep if manageability is enabled */
7604         if (adapter->wol || (adapter->flags & FLAG_MNG_PT_ENABLED) ||
7605             (hw->mac.ops.check_mng_mode(hw)))
7606                 device_wakeup_enable(&pdev->dev);
7607
7608         /* save off EEPROM version number */
7609         ret_val = e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
7610
7611         if (ret_val) {
7612                 e_dbg("NVM read error getting EEPROM version: %d\n", ret_val);
7613                 adapter->eeprom_vers = 0;
7614         }
7615
7616         /* init PTP hardware clock */
7617         e1000e_ptp_init(adapter);
7618
7619         /* reset the hardware with the new settings */
7620         e1000e_reset(adapter);
7621
7622         /* If the controller has AMT, do not set DRV_LOAD until the interface
7623          * is up.  For all other cases, let the f/w know that the h/w is now
7624          * under the control of the driver.
7625          */
7626         if (!(adapter->flags & FLAG_HAS_AMT))
7627                 e1000e_get_hw_control(adapter);
7628
7629         if (hw->mac.type >= e1000_pch_cnp)
7630                 adapter->flags2 |= FLAG2_ENABLE_S0IX_FLOWS;
7631
7632         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
7633         err = register_netdev(netdev);
7634         if (err)
7635                 goto err_register;
7636
7637         /* carrier off reporting is important to ethtool even BEFORE open */
7638         netif_carrier_off(netdev);
7639
7640         e1000_print_device_info(adapter);
7641
7642         dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_SMART_PREPARE);
7643
7644         if (pci_dev_run_wake(pdev) && hw->mac.type != e1000_pch_cnp)
7645                 pm_runtime_put_noidle(&pdev->dev);
7646
7647         return 0;
7648
7649 err_register:
7650         if (!(adapter->flags & FLAG_HAS_AMT))
7651                 e1000e_release_hw_control(adapter);
7652 err_eeprom:
7653         if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
7654                 e1000_phy_hw_reset(&adapter->hw);
7655 err_hw_init:
7656         kfree(adapter->tx_ring);
7657         kfree(adapter->rx_ring);
7658 err_sw_init:
7659         if ((adapter->hw.flash_address) && (hw->mac.type < e1000_pch_spt))
7660                 iounmap(adapter->hw.flash_address);
7661         e1000e_reset_interrupt_capability(adapter);
7662 err_flashmap:
7663         iounmap(adapter->hw.hw_addr);
7664 err_ioremap:
7665         free_netdev(netdev);
7666 err_alloc_etherdev:
7667         pci_release_mem_regions(pdev);
7668 err_pci_reg:
7669 err_dma:
7670         pci_disable_device(pdev);
7671         return err;
7672 }
7673
7674 /**
7675  * e1000_remove - Device Removal Routine
7676  * @pdev: PCI device information struct
7677  *
7678  * e1000_remove is called by the PCI subsystem to alert the driver
7679  * that it should release a PCI device.  The could be caused by a
7680  * Hot-Plug event, or because the driver is going to be removed from
7681  * memory.
7682  **/
7683 static void e1000_remove(struct pci_dev *pdev)
7684 {
7685         struct net_device *netdev = pci_get_drvdata(pdev);
7686         struct e1000_adapter *adapter = netdev_priv(netdev);
7687
7688         e1000e_ptp_remove(adapter);
7689
7690         /* The timers may be rescheduled, so explicitly disable them
7691          * from being rescheduled.
7692          */
7693         set_bit(__E1000_DOWN, &adapter->state);
7694         del_timer_sync(&adapter->watchdog_timer);
7695         del_timer_sync(&adapter->phy_info_timer);
7696
7697         cancel_work_sync(&adapter->reset_task);
7698         cancel_work_sync(&adapter->watchdog_task);
7699         cancel_work_sync(&adapter->downshift_task);
7700         cancel_work_sync(&adapter->update_phy_task);
7701         cancel_work_sync(&adapter->print_hang_task);
7702
7703         if (adapter->flags & FLAG_HAS_HW_TIMESTAMP) {
7704                 cancel_work_sync(&adapter->tx_hwtstamp_work);
7705                 if (adapter->tx_hwtstamp_skb) {
7706                         dev_consume_skb_any(adapter->tx_hwtstamp_skb);
7707                         adapter->tx_hwtstamp_skb = NULL;
7708                 }
7709         }
7710
7711         unregister_netdev(netdev);
7712
7713         if (pci_dev_run_wake(pdev))
7714                 pm_runtime_get_noresume(&pdev->dev);
7715
7716         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
7717          * would have already happened in close and is redundant.
7718          */
7719         e1000e_release_hw_control(adapter);
7720
7721         e1000e_reset_interrupt_capability(adapter);
7722         kfree(adapter->tx_ring);
7723         kfree(adapter->rx_ring);
7724
7725         iounmap(adapter->hw.hw_addr);
7726         if ((adapter->hw.flash_address) &&
7727             (adapter->hw.mac.type < e1000_pch_spt))
7728                 iounmap(adapter->hw.flash_address);
7729         pci_release_mem_regions(pdev);
7730
7731         free_netdev(netdev);
7732
7733         /* AER disable */
7734         pci_disable_pcie_error_reporting(pdev);
7735
7736         pci_disable_device(pdev);
7737 }
7738
7739 /* PCI Error Recovery (ERS) */
7740 static const struct pci_error_handlers e1000_err_handler = {
7741         .error_detected = e1000_io_error_detected,
7742         .slot_reset = e1000_io_slot_reset,
7743         .resume = e1000_io_resume,
7744 };
7745
7746 static const struct pci_device_id e1000_pci_tbl[] = {
7747         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
7748         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
7749         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
7750         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP),
7751           board_82571 },
7752         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
7753         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
7754         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
7755         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
7756         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
7757
7758         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
7759         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
7760         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
7761         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
7762
7763         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
7764         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
7765         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
7766
7767         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
7768         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
7769         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
7770
7771         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
7772           board_80003es2lan },
7773         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
7774           board_80003es2lan },
7775         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
7776           board_80003es2lan },
7777         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
7778           board_80003es2lan },
7779
7780         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
7781         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
7782         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
7783         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
7784         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
7785         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
7786         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
7787         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
7788
7789         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
7790         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
7791         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
7792         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
7793         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
7794         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
7795         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
7796         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
7797         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
7798
7799         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
7800         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
7801         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
7802
7803         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
7804         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
7805         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
7806
7807         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
7808         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
7809         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
7810         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
7811
7812         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
7813         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
7814
7815         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
7816         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
7817         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_LM), board_pch_lpt },
7818         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPTLP_I218_V), board_pch_lpt },
7819         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM2), board_pch_lpt },
7820         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V2), board_pch_lpt },
7821         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_LM3), board_pch_lpt },
7822         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_I218_V3), board_pch_lpt },
7823         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM), board_pch_spt },
7824         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V), board_pch_spt },
7825         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM2), board_pch_spt },
7826         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V2), board_pch_spt },
7827         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LBG_I219_LM3), board_pch_spt },
7828         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM4), board_pch_spt },
7829         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V4), board_pch_spt },
7830         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_LM5), board_pch_spt },
7831         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_SPT_I219_V5), board_pch_spt },
7832         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM6), board_pch_cnp },
7833         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V6), board_pch_cnp },
7834         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_LM7), board_pch_cnp },
7835         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CNP_I219_V7), board_pch_cnp },
7836         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM8), board_pch_cnp },
7837         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V8), board_pch_cnp },
7838         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_LM9), board_pch_cnp },
7839         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ICP_I219_V9), board_pch_cnp },
7840         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_LM10), board_pch_cnp },
7841         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_V10), board_pch_cnp },
7842         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_LM11), board_pch_cnp },
7843         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_V11), board_pch_cnp },
7844         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_LM12), board_pch_spt },
7845         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_CMP_I219_V12), board_pch_spt },
7846         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_LM13), board_pch_cnp },
7847         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_V13), board_pch_cnp },
7848         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_LM14), board_pch_cnp },
7849         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_V14), board_pch_cnp },
7850         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_LM15), board_pch_cnp },
7851         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_TGP_I219_V15), board_pch_cnp },
7852         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_LM16), board_pch_cnp },
7853         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_V16), board_pch_cnp },
7854         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_LM17), board_pch_cnp },
7855         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_ADP_I219_V17), board_pch_cnp },
7856         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_LM18), board_pch_cnp },
7857         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_V18), board_pch_cnp },
7858         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_LM19), board_pch_cnp },
7859         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_MTP_I219_V19), board_pch_cnp },
7860
7861         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
7862 };
7863 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
7864
7865 static const struct dev_pm_ops e1000_pm_ops = {
7866 #ifdef CONFIG_PM_SLEEP
7867         .prepare        = e1000e_pm_prepare,
7868         .suspend        = e1000e_pm_suspend,
7869         .resume         = e1000e_pm_resume,
7870         .freeze         = e1000e_pm_freeze,
7871         .thaw           = e1000e_pm_thaw,
7872         .poweroff       = e1000e_pm_suspend,
7873         .restore        = e1000e_pm_resume,
7874 #endif
7875         SET_RUNTIME_PM_OPS(e1000e_pm_runtime_suspend, e1000e_pm_runtime_resume,
7876                            e1000e_pm_runtime_idle)
7877 };
7878
7879 /* PCI Device API Driver */
7880 static struct pci_driver e1000_driver = {
7881         .name     = e1000e_driver_name,
7882         .id_table = e1000_pci_tbl,
7883         .probe    = e1000_probe,
7884         .remove   = e1000_remove,
7885         .driver   = {
7886                 .pm = &e1000_pm_ops,
7887         },
7888         .shutdown = e1000_shutdown,
7889         .err_handler = &e1000_err_handler
7890 };
7891
7892 /**
7893  * e1000_init_module - Driver Registration Routine
7894  *
7895  * e1000_init_module is the first routine called when the driver is
7896  * loaded. All it does is register with the PCI subsystem.
7897  **/
7898 static int __init e1000_init_module(void)
7899 {
7900         pr_info("Intel(R) PRO/1000 Network Driver\n");
7901         pr_info("Copyright(c) 1999 - 2015 Intel Corporation.\n");
7902
7903         return pci_register_driver(&e1000_driver);
7904 }
7905 module_init(e1000_init_module);
7906
7907 /**
7908  * e1000_exit_module - Driver Exit Cleanup Routine
7909  *
7910  * e1000_exit_module is called just before the driver is removed
7911  * from memory.
7912  **/
7913 static void __exit e1000_exit_module(void)
7914 {
7915         pci_unregister_driver(&e1000_driver);
7916 }
7917 module_exit(e1000_exit_module);
7918
7919 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
7920 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
7921 MODULE_LICENSE("GPL v2");
7922
7923 /* netdev.c */