drivers/net/ethernet/ibm/ibmvnic.c

   1 /**************************************************************************/
   2 /*                                                                        */
   3 /*  IBM System i and System p Virtual NIC Device Driver                   */
   4 /*  Copyright (C) 2014 IBM Corp.                                          */
   5 /*  Santiago Leon ([email protected])                                  */
   6 /*  Thomas Falcon ([email protected])                           */
   7 /*  John Allen ([email protected])                                */
   8 /*                                                                        */
   9 /*  This program is free software; you can redistribute it and/or modify  */
  10 /*  it under the terms of the GNU General Public License as published by  */
  11 /*  the Free Software Foundation; either version 2 of the License, or     */
  12 /*  (at your option) any later version.                                   */
  13 /*                                                                        */
  14 /*  This program is distributed in the hope that it will be useful,       */
  15 /*  but WITHOUT ANY WARRANTY; without even the implied warranty of        */
  16 /*  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         */
  17 /*  GNU General Public License for more details.                          */
  18 /*                                                                        */
  19 /*  You should have received a copy of the GNU General Public License     */
  20 /*  along with this program.                                              */
  21 /*                                                                        */
  22 /* This module contains the implementation of a virtual ethernet device   */
  23 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  24 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  25 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  26 /*                                                                         */
  27 /* Messages are passed between the VNIC driver and the VNIC server using  */
  28 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  29 /* issue and receive commands that initiate communication with the server */
  30 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  31 /* are used by the driver to notify the server that a packet is           */
  32 /* ready for transmission or that a buffer has been added to receive a    */
  33 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  34 /* driver that a packet transmission has been completed or that a packet  */
  35 /* has been received and placed in a waiting buffer.                      */
  36 /*                                                                        */
  37 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  38 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  39 /* or receive has been completed, the VNIC driver is required to use      */
  40 /* "long term mapping". This entails that large, continuous DMA mapped    */
  41 /* buffers are allocated on driver initialization and these buffers are   */
  42 /* then continuously reused to pass skbs to and from the VNIC server.     */
  43 /*                                                                        */
  44 /**************************************************************************/
  45
  46 #include <linux/module.h>
  47 #include <linux/moduleparam.h>
  48 #include <linux/types.h>
  49 #include <linux/errno.h>
  50 #include <linux/completion.h>
  51 #include <linux/ioport.h>
  52 #include <linux/dma-mapping.h>
  53 #include <linux/kernel.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/skbuff.h>
  57 #include <linux/init.h>
  58 #include <linux/delay.h>
  59 #include <linux/mm.h>
  60 #include <linux/ethtool.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/if_arp.h>
  63 #include <linux/in.h>
  64 #include <linux/ip.h>
  65 #include <linux/ipv6.h>
  66 #include <linux/irq.h>
  67 #include <linux/kthread.h>
  68 #include <linux/seq_file.h>
  69 #include <linux/interrupt.h>
  70 #include <net/net_namespace.h>
  71 #include <asm/hvcall.h>
  72 #include <linux/atomic.h>
  73 #include <asm/vio.h>
  74 #include <asm/iommu.h>
  75 #include <linux/uaccess.h>
  76 #include <asm/firmware.h>
  77 #include <linux/workqueue.h>
  78 #include <linux/if_vlan.h>
  79 #include <linux/utsname.h>
  80
  81 #include "ibmvnic.h"
  82
  83 static const char ibmvnic_driver_name[] = "ibmvnic";
  84 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  85
  86 MODULE_AUTHOR("Santiago Leon");
  87 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  88 MODULE_LICENSE("GPL");
  89 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  90
  91 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  92 static int ibmvnic_remove(struct vio_dev *);
  93 static void release_sub_crqs(struct ibmvnic_adapter *);
  94 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  95 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  96 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  97 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  98 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
  99                        union sub_crq *sub_crq);
 100 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
 101 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
 102 static int enable_scrq_irq(struct ibmvnic_adapter *,
 103                            struct ibmvnic_sub_crq_queue *);
 104 static int disable_scrq_irq(struct ibmvnic_adapter *,
 105                             struct ibmvnic_sub_crq_queue *);
 106 static int pending_scrq(struct ibmvnic_adapter *,
 107                         struct ibmvnic_sub_crq_queue *);
 108 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
 109                                         struct ibmvnic_sub_crq_queue *);
 110 static int ibmvnic_poll(struct napi_struct *napi, int data);
 111 static void send_map_query(struct ibmvnic_adapter *adapter);
 112 static void send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
 113 static void send_request_unmap(struct ibmvnic_adapter *, u8);
 114 static void send_login(struct ibmvnic_adapter *adapter);
 115 static void send_cap_queries(struct ibmvnic_adapter *adapter);
 116 static int init_sub_crqs(struct ibmvnic_adapter *);
 117 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 118 static int ibmvnic_init(struct ibmvnic_adapter *);
 119 static void release_crq_queue(struct ibmvnic_adapter *);
 120 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p);
 121
 122 struct ibmvnic_stat {
 123         char name[ETH_GSTRING_LEN];
 124         int offset;
 125 };
 126
 127 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 128                              offsetof(struct ibmvnic_statistics, stat))
 129 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + off)))
 130
 131 static const struct ibmvnic_stat ibmvnic_stats[] = {
 132         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 133         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 134         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 135         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 136         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 137         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 138         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 139         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 140         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 141         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 142         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 143         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 144         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 145         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 146         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 147         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 148         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 149         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 150         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 151         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 152         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 153         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 154 };
 155
 156 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 157                           unsigned long length, unsigned long *number,
 158                           unsigned long *irq)
 159 {
 160         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 161         long rc;
 162
 163         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 164         *number = retbuf[0];
 165         *irq = retbuf[1];
 166
 167         return rc;
 168 }
 169
 170 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 171                                 struct ibmvnic_long_term_buff *ltb, int size)
 172 {
 173         struct device *dev = &adapter->vdev->dev;
 174
 175         ltb->size = size;
 176         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 177                                        GFP_KERNEL);
 178
 179         if (!ltb->buff) {
 180                 dev_err(dev, "Couldn't alloc long term buffer\n");
 181                 return -ENOMEM;
 182         }
 183         ltb->map_id = adapter->map_id;
 184         adapter->map_id++;
 185
 186         init_completion(&adapter->fw_done);
 187         send_request_map(adapter, ltb->addr,
 188                          ltb->size, ltb->map_id);
 189         wait_for_completion(&adapter->fw_done);
 190
 191         if (adapter->fw_done_rc) {
 192                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 193                         adapter->fw_done_rc);
 194                 return -1;
 195         }
 196         return 0;
 197 }
 198
 199 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 200                                 struct ibmvnic_long_term_buff *ltb)
 201 {
 202         struct device *dev = &adapter->vdev->dev;
 203
 204         if (!ltb->buff)
 205                 return;
 206
 207         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 208             adapter->reset_reason != VNIC_RESET_MOBILITY)
 209                 send_request_unmap(adapter, ltb->map_id);
 210         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 211 }
 212
 213 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 214                                 struct ibmvnic_long_term_buff *ltb)
 215 {
 216         memset(ltb->buff, 0, ltb->size);
 217
 218         init_completion(&adapter->fw_done);
 219         send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 220         wait_for_completion(&adapter->fw_done);
 221
 222         if (adapter->fw_done_rc) {
 223                 dev_info(&adapter->vdev->dev,
 224                          "Reset failed, attempting to free and reallocate buffer\n");
 225                 free_long_term_buff(adapter, ltb);
 226                 return alloc_long_term_buff(adapter, ltb, ltb->size);
 227         }
 228         return 0;
 229 }
 230
 231 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 232 {
 233         int i;
 234
 235         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 236              i++)
 237                 adapter->rx_pool[i].active = 0;
 238 }
 239
 240 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 241                               struct ibmvnic_rx_pool *pool)
 242 {
 243         int count = pool->size - atomic_read(&pool->available);
 244         struct device *dev = &adapter->vdev->dev;
 245         int buffers_added = 0;
 246         unsigned long lpar_rc;
 247         union sub_crq sub_crq;
 248         struct sk_buff *skb;
 249         unsigned int offset;
 250         dma_addr_t dma_addr;
 251         unsigned char *dst;
 252         u64 *handle_array;
 253         int shift = 0;
 254         int index;
 255         int i;
 256
 257         if (!pool->active)
 258                 return;
 259
 260         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 261                                       be32_to_cpu(adapter->login_rsp_buf->
 262                                       off_rxadd_subcrqs));
 263
 264         for (i = 0; i < count; ++i) {
 265                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 266                 if (!skb) {
 267                         dev_err(dev, "Couldn't replenish rx buff\n");
 268                         adapter->replenish_no_mem++;
 269                         break;
 270                 }
 271
 272                 index = pool->free_map[pool->next_free];
 273
 274                 if (pool->rx_buff[index].skb)
 275                         dev_err(dev, "Inconsistent free_map!\n");
 276
 277                 /* Copy the skb to the long term mapped DMA buffer */
 278                 offset = index * pool->buff_size;
 279                 dst = pool->long_term_buff.buff + offset;
 280                 memset(dst, 0, pool->buff_size);
 281                 dma_addr = pool->long_term_buff.addr + offset;
 282                 pool->rx_buff[index].data = dst;
 283
 284                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 285                 pool->rx_buff[index].dma = dma_addr;
 286                 pool->rx_buff[index].skb = skb;
 287                 pool->rx_buff[index].pool_index = pool->index;
 288                 pool->rx_buff[index].size = pool->buff_size;
 289
 290                 memset(&sub_crq, 0, sizeof(sub_crq));
 291                 sub_crq.rx_add.first = IBMVNIC_CRQ_CMD;
 292                 sub_crq.rx_add.correlator =
 293                     cpu_to_be64((u64)&pool->rx_buff[index]);
 294                 sub_crq.rx_add.ioba = cpu_to_be32(dma_addr);
 295                 sub_crq.rx_add.map_id = pool->long_term_buff.map_id;
 296
 297                 /* The length field of the sCRQ is defined to be 24 bits so the
 298                  * buffer size needs to be left shifted by a byte before it is
 299                  * converted to big endian to prevent the last byte from being
 300                  * truncated.
 301                  */
 302 #ifdef __LITTLE_ENDIAN__
 303                 shift = 8;
 304 #endif
 305                 sub_crq.rx_add.len = cpu_to_be32(pool->buff_size << shift);
 306
 307                 lpar_rc = send_subcrq(adapter, handle_array[pool->index],
 308                                       &sub_crq);
 309                 if (lpar_rc != H_SUCCESS)
 310                         goto failure;
 311
 312                 buffers_added++;
 313                 adapter->replenish_add_buff_success++;
 314                 pool->next_free = (pool->next_free + 1) % pool->size;
 315         }
 316         atomic_add(buffers_added, &pool->available);
 317         return;
 318
 319 failure:
 320         dev_info(dev, "replenish pools failure\n");
 321         pool->free_map[pool->next_free] = index;
 322         pool->rx_buff[index].skb = NULL;
 323         if (!dma_mapping_error(dev, dma_addr))
 324                 dma_unmap_single(dev, dma_addr, pool->buff_size,
 325                                  DMA_FROM_DEVICE);
 326
 327         dev_kfree_skb_any(skb);
 328         adapter->replenish_add_buff_failure++;
 329         atomic_add(buffers_added, &pool->available);
 330
 331         if (lpar_rc == H_CLOSED) {
 332                 /* Disable buffer pool replenishment and report carrier off if
 333                  * queue is closed. Firmware guarantees that a signal will
 334                  * be sent to the driver, triggering a reset.
 335                  */
 336                 deactivate_rx_pools(adapter);
 337                 netif_carrier_off(adapter->netdev);
 338         }
 339 }
 340
 341 static void replenish_pools(struct ibmvnic_adapter *adapter)
 342 {
 343         int i;
 344
 345         adapter->replenish_task_cycles++;
 346         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 347              i++) {
 348                 if (adapter->rx_pool[i].active)
 349                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 350         }
 351 }
 352
 353 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 354 {
 355         kfree(adapter->tx_stats_buffers);
 356         kfree(adapter->rx_stats_buffers);
 357 }
 358
 359 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 360 {
 361         adapter->tx_stats_buffers =
 362                                 kcalloc(adapter->req_tx_queues,
 363                                         sizeof(struct ibmvnic_tx_queue_stats),
 364                                         GFP_KERNEL);
 365         if (!adapter->tx_stats_buffers)
 366                 return -ENOMEM;
 367
 368         adapter->rx_stats_buffers =
 369                                 kcalloc(adapter->req_rx_queues,
 370                                         sizeof(struct ibmvnic_rx_queue_stats),
 371                                         GFP_KERNEL);
 372         if (!adapter->rx_stats_buffers)
 373                 return -ENOMEM;
 374
 375         return 0;
 376 }
 377
 378 static void release_stats_token(struct ibmvnic_adapter *adapter)
 379 {
 380         struct device *dev = &adapter->vdev->dev;
 381
 382         if (!adapter->stats_token)
 383                 return;
 384
 385         dma_unmap_single(dev, adapter->stats_token,
 386                          sizeof(struct ibmvnic_statistics),
 387                          DMA_FROM_DEVICE);
 388         adapter->stats_token = 0;
 389 }
 390
 391 static int init_stats_token(struct ibmvnic_adapter *adapter)
 392 {
 393         struct device *dev = &adapter->vdev->dev;
 394         dma_addr_t stok;
 395
 396         stok = dma_map_single(dev, &adapter->stats,
 397                               sizeof(struct ibmvnic_statistics),
 398                               DMA_FROM_DEVICE);
 399         if (dma_mapping_error(dev, stok)) {
 400                 dev_err(dev, "Couldn't map stats buffer\n");
 401                 return -1;
 402         }
 403
 404         adapter->stats_token = stok;
 405         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 406         return 0;
 407 }
 408
 409 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 410 {
 411         struct ibmvnic_rx_pool *rx_pool;
 412         int rx_scrqs;
 413         int i, j, rc;
 414         u64 *size_array;
 415
 416         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 417                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
 418
 419         rx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 420         for (i = 0; i < rx_scrqs; i++) {
 421                 rx_pool = &adapter->rx_pool[i];
 422
 423                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 424
 425                 if (rx_pool->buff_size != be64_to_cpu(size_array[i])) {
 426                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 427                         rx_pool->buff_size = be64_to_cpu(size_array[i]);
 428                         alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 429                                              rx_pool->size *
 430                                              rx_pool->buff_size);
 431                 } else {
 432                         rc = reset_long_term_buff(adapter,
 433                                                   &rx_pool->long_term_buff);
 434                 }
 435
 436                 if (rc)
 437                         return rc;
 438
 439                 for (j = 0; j < rx_pool->size; j++)
 440                         rx_pool->free_map[j] = j;
 441
 442                 memset(rx_pool->rx_buff, 0,
 443                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 444
 445                 atomic_set(&rx_pool->available, 0);
 446                 rx_pool->next_alloc = 0;
 447                 rx_pool->next_free = 0;
 448                 rx_pool->active = 1;
 449         }
 450
 451         return 0;
 452 }
 453
 454 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 455 {
 456         struct ibmvnic_rx_pool *rx_pool;
 457         int i, j;
 458
 459         if (!adapter->rx_pool)
 460                 return;
 461
 462         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 463                 rx_pool = &adapter->rx_pool[i];
 464
 465                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 466
 467                 kfree(rx_pool->free_map);
 468                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 469
 470                 if (!rx_pool->rx_buff)
 471                         continue;
 472
 473                 for (j = 0; j < rx_pool->size; j++) {
 474                         if (rx_pool->rx_buff[j].skb) {
 475                                 dev_kfree_skb_any(rx_pool->rx_buff[i].skb);
 476                                 rx_pool->rx_buff[i].skb = NULL;
 477                         }
 478                 }
 479
 480                 kfree(rx_pool->rx_buff);
 481         }
 482
 483         kfree(adapter->rx_pool);
 484         adapter->rx_pool = NULL;
 485         adapter->num_active_rx_pools = 0;
 486 }
 487
 488 static int init_rx_pools(struct net_device *netdev)
 489 {
 490         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 491         struct device *dev = &adapter->vdev->dev;
 492         struct ibmvnic_rx_pool *rx_pool;
 493         int rxadd_subcrqs;
 494         u64 *size_array;
 495         int i, j;
 496
 497         rxadd_subcrqs =
 498                 be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 499         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 500                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
 501
 502         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 503                                    sizeof(struct ibmvnic_rx_pool),
 504                                    GFP_KERNEL);
 505         if (!adapter->rx_pool) {
 506                 dev_err(dev, "Failed to allocate rx pools\n");
 507                 return -1;
 508         }
 509
 510         adapter->num_active_rx_pools = 0;
 511
 512         for (i = 0; i < rxadd_subcrqs; i++) {
 513                 rx_pool = &adapter->rx_pool[i];
 514
 515                 netdev_dbg(adapter->netdev,
 516                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 517                            i, adapter->req_rx_add_entries_per_subcrq,
 518                            be64_to_cpu(size_array[i]));
 519
 520                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 521                 rx_pool->index = i;
 522                 rx_pool->buff_size = be64_to_cpu(size_array[i]);
 523                 rx_pool->active = 1;
 524
 525                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 526                                             GFP_KERNEL);
 527                 if (!rx_pool->free_map) {
 528                         release_rx_pools(adapter);
 529                         return -1;
 530                 }
 531
 532                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 533                                            sizeof(struct ibmvnic_rx_buff),
 534                                            GFP_KERNEL);
 535                 if (!rx_pool->rx_buff) {
 536                         dev_err(dev, "Couldn't alloc rx buffers\n");
 537                         release_rx_pools(adapter);
 538                         return -1;
 539                 }
 540
 541                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 542                                          rx_pool->size * rx_pool->buff_size)) {
 543                         release_rx_pools(adapter);
 544                         return -1;
 545                 }
 546
 547                 for (j = 0; j < rx_pool->size; ++j)
 548                         rx_pool->free_map[j] = j;
 549
 550                 atomic_set(&rx_pool->available, 0);
 551                 rx_pool->next_alloc = 0;
 552                 rx_pool->next_free = 0;
 553         }
 554
 555         adapter->num_active_rx_pools = rxadd_subcrqs;
 556
 557         return 0;
 558 }
 559
 560 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 561 {
 562         struct ibmvnic_tx_pool *tx_pool;
 563         int tx_scrqs;
 564         int i, j, rc;
 565
 566         tx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
 567         for (i = 0; i < tx_scrqs; i++) {
 568                 netdev_dbg(adapter->netdev, "Re-setting tx_pool[%d]\n", i);
 569
 570                 tx_pool = &adapter->tx_pool[i];
 571
 572                 rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 573                 if (rc)
 574                         return rc;
 575
 576                 rc = reset_long_term_buff(adapter, &tx_pool->tso_ltb);
 577                 if (rc)
 578                         return rc;
 579
 580                 memset(tx_pool->tx_buff, 0,
 581                        adapter->req_tx_entries_per_subcrq *
 582                        sizeof(struct ibmvnic_tx_buff));
 583
 584                 for (j = 0; j < adapter->req_tx_entries_per_subcrq; j++)
 585                         tx_pool->free_map[j] = j;
 586
 587                 tx_pool->consumer_index = 0;
 588                 tx_pool->producer_index = 0;
 589                 tx_pool->tso_index = 0;
 590         }
 591
 592         return 0;
 593 }
 594
 595 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 596 {
 597         if (!adapter->vpd)
 598                 return;
 599
 600         kfree(adapter->vpd->buff);
 601         kfree(adapter->vpd);
 602 }
 603
 604 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 605 {
 606         struct ibmvnic_tx_pool *tx_pool;
 607         int i;
 608
 609         if (!adapter->tx_pool)
 610                 return;
 611
 612         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 613                 netdev_dbg(adapter->netdev, "Releasing tx_pool[%d]\n", i);
 614                 tx_pool = &adapter->tx_pool[i];
 615                 kfree(tx_pool->tx_buff);
 616                 free_long_term_buff(adapter, &tx_pool->long_term_buff);
 617                 free_long_term_buff(adapter, &tx_pool->tso_ltb);
 618                 kfree(tx_pool->free_map);
 619         }
 620
 621         kfree(adapter->tx_pool);
 622         adapter->tx_pool = NULL;
 623         adapter->num_active_tx_pools = 0;
 624 }
 625
 626 static int init_tx_pools(struct net_device *netdev)
 627 {
 628         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 629         struct device *dev = &adapter->vdev->dev;
 630         struct ibmvnic_tx_pool *tx_pool;
 631         int tx_subcrqs;
 632         int i, j;
 633
 634         tx_subcrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
 635         adapter->tx_pool = kcalloc(tx_subcrqs,
 636                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 637         if (!adapter->tx_pool)
 638                 return -1;
 639
 640         adapter->num_active_tx_pools = 0;
 641
 642         for (i = 0; i < tx_subcrqs; i++) {
 643                 tx_pool = &adapter->tx_pool[i];
 644
 645                 netdev_dbg(adapter->netdev,
 646                            "Initializing tx_pool[%d], %lld buffs\n",
 647                            i, adapter->req_tx_entries_per_subcrq);
 648
 649                 tx_pool->tx_buff = kcalloc(adapter->req_tx_entries_per_subcrq,
 650                                            sizeof(struct ibmvnic_tx_buff),
 651                                            GFP_KERNEL);
 652                 if (!tx_pool->tx_buff) {
 653                         dev_err(dev, "tx pool buffer allocation failed\n");
 654                         release_tx_pools(adapter);
 655                         return -1;
 656                 }
 657
 658                 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 659                                          adapter->req_tx_entries_per_subcrq *
 660                                          adapter->req_mtu)) {
 661                         release_tx_pools(adapter);
 662                         return -1;
 663                 }
 664
 665                 /* alloc TSO ltb */
 666                 if (alloc_long_term_buff(adapter, &tx_pool->tso_ltb,
 667                                          IBMVNIC_TSO_BUFS *
 668                                          IBMVNIC_TSO_BUF_SZ)) {
 669                         release_tx_pools(adapter);
 670                         return -1;
 671                 }
 672
 673                 tx_pool->tso_index = 0;
 674
 675                 tx_pool->free_map = kcalloc(adapter->req_tx_entries_per_subcrq,
 676                                             sizeof(int), GFP_KERNEL);
 677                 if (!tx_pool->free_map) {
 678                         release_tx_pools(adapter);
 679                         return -1;
 680                 }
 681
 682                 for (j = 0; j < adapter->req_tx_entries_per_subcrq; j++)
 683                         tx_pool->free_map[j] = j;
 684
 685                 tx_pool->consumer_index = 0;
 686                 tx_pool->producer_index = 0;
 687         }
 688
 689         adapter->num_active_tx_pools = tx_subcrqs;
 690
 691         return 0;
 692 }
 693
 694 static void release_error_buffers(struct ibmvnic_adapter *adapter)
 695 {
 696         struct device *dev = &adapter->vdev->dev;
 697         struct ibmvnic_error_buff *error_buff, *tmp;
 698         unsigned long flags;
 699
 700         spin_lock_irqsave(&adapter->error_list_lock, flags);
 701         list_for_each_entry_safe(error_buff, tmp, &adapter->errors, list) {
 702                 list_del(&error_buff->list);
 703                 dma_unmap_single(dev, error_buff->dma, error_buff->len,
 704                                  DMA_FROM_DEVICE);
 705                 kfree(error_buff->buff);
 706                 kfree(error_buff);
 707         }
 708         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
 709 }
 710
 711 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 712 {
 713         int i;
 714
 715         if (adapter->napi_enabled)
 716                 return;
 717
 718         for (i = 0; i < adapter->req_rx_queues; i++)
 719                 napi_enable(&adapter->napi[i]);
 720
 721         adapter->napi_enabled = true;
 722 }
 723
 724 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 725 {
 726         int i;
 727
 728         if (!adapter->napi_enabled)
 729                 return;
 730
 731         for (i = 0; i < adapter->req_rx_queues; i++) {
 732                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 733                 napi_disable(&adapter->napi[i]);
 734         }
 735
 736         adapter->napi_enabled = false;
 737 }
 738
 739 static int ibmvnic_login(struct net_device *netdev)
 740 {
 741         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 742         unsigned long timeout = msecs_to_jiffies(30000);
 743         struct device *dev = &adapter->vdev->dev;
 744         int rc;
 745
 746         do {
 747                 if (adapter->renegotiate) {
 748                         adapter->renegotiate = false;
 749                         release_sub_crqs(adapter);
 750
 751                         reinit_completion(&adapter->init_done);
 752                         send_cap_queries(adapter);
 753                         if (!wait_for_completion_timeout(&adapter->init_done,
 754                                                          timeout)) {
 755                                 dev_err(dev, "Capabilities query timeout\n");
 756                                 return -1;
 757                         }
 758                         rc = init_sub_crqs(adapter);
 759                         if (rc) {
 760                                 dev_err(dev,
 761                                         "Initialization of SCRQ's failed\n");
 762                                 return -1;
 763                         }
 764                         rc = init_sub_crq_irqs(adapter);
 765                         if (rc) {
 766                                 dev_err(dev,
 767                                         "Initialization of SCRQ's irqs failed\n");
 768                                 return -1;
 769                         }
 770                 }
 771
 772                 reinit_completion(&adapter->init_done);
 773                 send_login(adapter);
 774                 if (!wait_for_completion_timeout(&adapter->init_done,
 775                                                  timeout)) {
 776                         dev_err(dev, "Login timeout\n");
 777                         return -1;
 778                 }
 779         } while (adapter->renegotiate);
 780
 781         /* handle pending MAC address changes after successful login */
 782         if (adapter->mac_change_pending) {
 783                 __ibmvnic_set_mac(netdev, &adapter->desired.mac);
 784                 adapter->mac_change_pending = false;
 785         }
 786
 787         return 0;
 788 }
 789
 790 static void release_resources(struct ibmvnic_adapter *adapter)
 791 {
 792         int i;
 793
 794         release_vpd_data(adapter);
 795
 796         release_tx_pools(adapter);
 797         release_rx_pools(adapter);
 798
 799         release_stats_token(adapter);
 800         release_stats_buffers(adapter);
 801         release_error_buffers(adapter);
 802
 803         if (adapter->napi) {
 804                 for (i = 0; i < adapter->req_rx_queues; i++) {
 805                         if (&adapter->napi[i]) {
 806                                 netdev_dbg(adapter->netdev,
 807                                            "Releasing napi[%d]\n", i);
 808                                 netif_napi_del(&adapter->napi[i]);
 809                         }
 810                 }
 811         }
 812 }
 813
 814 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 815 {
 816         struct net_device *netdev = adapter->netdev;
 817         unsigned long timeout = msecs_to_jiffies(30000);
 818         union ibmvnic_crq crq;
 819         bool resend;
 820         int rc;
 821
 822         netdev_dbg(netdev, "setting link state %d\n", link_state);
 823
 824         memset(&crq, 0, sizeof(crq));
 825         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 826         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 827         crq.logical_link_state.link_state = link_state;
 828
 829         do {
 830                 resend = false;
 831
 832                 reinit_completion(&adapter->init_done);
 833                 rc = ibmvnic_send_crq(adapter, &crq);
 834                 if (rc) {
 835                         netdev_err(netdev, "Failed to set link state\n");
 836                         return rc;
 837                 }
 838
 839                 if (!wait_for_completion_timeout(&adapter->init_done,
 840                                                  timeout)) {
 841                         netdev_err(netdev, "timeout setting link state\n");
 842                         return -1;
 843                 }
 844
 845                 if (adapter->init_done_rc == 1) {
 846                         /* Partuial success, delay and re-send */
 847                         mdelay(1000);
 848                         resend = true;
 849                 }
 850         } while (resend);
 851
 852         return 0;
 853 }
 854
 855 static int set_real_num_queues(struct net_device *netdev)
 856 {
 857         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 858         int rc;
 859
 860         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
 861                    adapter->req_tx_queues, adapter->req_rx_queues);
 862
 863         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
 864         if (rc) {
 865                 netdev_err(netdev, "failed to set the number of tx queues\n");
 866                 return rc;
 867         }
 868
 869         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
 870         if (rc)
 871                 netdev_err(netdev, "failed to set the number of rx queues\n");
 872
 873         return rc;
 874 }
 875
 876 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
 877 {
 878         struct device *dev = &adapter->vdev->dev;
 879         union ibmvnic_crq crq;
 880         int len = 0;
 881
 882         if (adapter->vpd->buff)
 883                 len = adapter->vpd->len;
 884
 885         init_completion(&adapter->fw_done);
 886         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
 887         crq.get_vpd_size.cmd = GET_VPD_SIZE;
 888         ibmvnic_send_crq(adapter, &crq);
 889         wait_for_completion(&adapter->fw_done);
 890
 891         if (!adapter->vpd->len)
 892                 return -ENODATA;
 893
 894         if (!adapter->vpd->buff)
 895                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
 896         else if (adapter->vpd->len != len)
 897                 adapter->vpd->buff =
 898                         krealloc(adapter->vpd->buff,
 899                                  adapter->vpd->len, GFP_KERNEL);
 900
 901         if (!adapter->vpd->buff) {
 902                 dev_err(dev, "Could allocate VPD buffer\n");
 903                 return -ENOMEM;
 904         }
 905
 906         adapter->vpd->dma_addr =
 907                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
 908                                DMA_FROM_DEVICE);
 909         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
 910                 dev_err(dev, "Could not map VPD buffer\n");
 911                 kfree(adapter->vpd->buff);
 912                 return -ENOMEM;
 913         }
 914
 915         reinit_completion(&adapter->fw_done);
 916         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
 917         crq.get_vpd.cmd = GET_VPD;
 918         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
 919         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
 920         ibmvnic_send_crq(adapter, &crq);
 921         wait_for_completion(&adapter->fw_done);
 922
 923         return 0;
 924 }
 925
 926 static int init_resources(struct ibmvnic_adapter *adapter)
 927 {
 928         struct net_device *netdev = adapter->netdev;
 929         int i, rc;
 930
 931         rc = set_real_num_queues(netdev);
 932         if (rc)
 933                 return rc;
 934
 935         rc = init_stats_buffers(adapter);
 936         if (rc)
 937                 return rc;
 938
 939         rc = init_stats_token(adapter);
 940         if (rc)
 941                 return rc;
 942
 943         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
 944         if (!adapter->vpd)
 945                 return -ENOMEM;
 946
 947         /* Vital Product Data (VPD) */
 948         rc = ibmvnic_get_vpd(adapter);
 949         if (rc) {
 950                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
 951                 return rc;
 952         }
 953
 954         adapter->map_id = 1;
 955         adapter->napi = kcalloc(adapter->req_rx_queues,
 956                                 sizeof(struct napi_struct), GFP_KERNEL);
 957         if (!adapter->napi)
 958                 return -ENOMEM;
 959
 960         for (i = 0; i < adapter->req_rx_queues; i++) {
 961                 netdev_dbg(netdev, "Adding napi[%d]\n", i);
 962                 netif_napi_add(netdev, &adapter->napi[i], ibmvnic_poll,
 963                                NAPI_POLL_WEIGHT);
 964         }
 965
 966         send_map_query(adapter);
 967
 968         rc = init_rx_pools(netdev);
 969         if (rc)
 970                 return rc;
 971
 972         rc = init_tx_pools(netdev);
 973         return rc;
 974 }
 975
 976 static int __ibmvnic_open(struct net_device *netdev)
 977 {
 978         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 979         enum vnic_state prev_state = adapter->state;
 980         int i, rc;
 981
 982         adapter->state = VNIC_OPENING;
 983         replenish_pools(adapter);
 984         ibmvnic_napi_enable(adapter);
 985
 986         /* We're ready to receive frames, enable the sub-crq interrupts and
 987          * set the logical link state to up
 988          */
 989         for (i = 0; i < adapter->req_rx_queues; i++) {
 990                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
 991                 if (prev_state == VNIC_CLOSED)
 992                         enable_irq(adapter->rx_scrq[i]->irq);
 993                 else
 994                         enable_scrq_irq(adapter, adapter->rx_scrq[i]);
 995         }
 996
 997         for (i = 0; i < adapter->req_tx_queues; i++) {
 998                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
 999                 if (prev_state == VNIC_CLOSED)
1000                         enable_irq(adapter->tx_scrq[i]->irq);
1001                 else
1002                         enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1003         }
1004
1005         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1006         if (rc) {
1007                 for (i = 0; i < adapter->req_rx_queues; i++)
1008                         napi_disable(&adapter->napi[i]);
1009                 release_resources(adapter);
1010                 return rc;
1011         }
1012
1013         netif_tx_start_all_queues(netdev);
1014
1015         if (prev_state == VNIC_CLOSED) {
1016                 for (i = 0; i < adapter->req_rx_queues; i++)
1017                         napi_schedule(&adapter->napi[i]);
1018         }
1019
1020         adapter->state = VNIC_OPEN;
1021         return rc;
1022 }
1023
1024 static int ibmvnic_open(struct net_device *netdev)
1025 {
1026         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1027         int rc;
1028
1029         mutex_lock(&adapter->reset_lock);
1030
1031         if (adapter->state != VNIC_CLOSED) {
1032                 rc = ibmvnic_login(netdev);
1033                 if (rc) {
1034                         mutex_unlock(&adapter->reset_lock);
1035                         return rc;
1036                 }
1037
1038                 rc = init_resources(adapter);
1039                 if (rc) {
1040                         netdev_err(netdev, "failed to initialize resources\n");
1041                         release_resources(adapter);
1042                         mutex_unlock(&adapter->reset_lock);
1043                         return rc;
1044                 }
1045         }
1046
1047         rc = __ibmvnic_open(netdev);
1048         netif_carrier_on(netdev);
1049
1050         mutex_unlock(&adapter->reset_lock);
1051
1052         return rc;
1053 }
1054
1055 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1056 {
1057         struct ibmvnic_tx_pool *tx_pool;
1058         u64 tx_entries;
1059         int tx_scrqs;
1060         int i, j;
1061
1062         if (!adapter->tx_pool)
1063                 return;
1064
1065         tx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
1066         tx_entries = adapter->req_tx_entries_per_subcrq;
1067
1068         /* Free any remaining skbs in the tx buffer pools */
1069         for (i = 0; i < tx_scrqs; i++) {
1070                 tx_pool = &adapter->tx_pool[i];
1071                 if (!tx_pool)
1072                         continue;
1073
1074                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1075                 for (j = 0; j < tx_entries; j++) {
1076                         if (tx_pool->tx_buff[j].skb) {
1077                                 dev_kfree_skb_any(tx_pool->tx_buff[j].skb);
1078                                 tx_pool->tx_buff[j].skb = NULL;
1079                         }
1080                 }
1081         }
1082 }
1083
1084 static int __ibmvnic_close(struct net_device *netdev)
1085 {
1086         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1087         int rc = 0;
1088         int i;
1089
1090         adapter->state = VNIC_CLOSING;
1091
1092         /* ensure that transmissions are stopped if called by do_reset */
1093         if (adapter->resetting)
1094                 netif_tx_disable(netdev);
1095         else
1096                 netif_tx_stop_all_queues(netdev);
1097
1098         ibmvnic_napi_disable(adapter);
1099
1100         if (adapter->tx_scrq) {
1101                 for (i = 0; i < adapter->req_tx_queues; i++)
1102                         if (adapter->tx_scrq[i]->irq) {
1103                                 netdev_dbg(adapter->netdev,
1104                                            "Disabling tx_scrq[%d] irq\n", i);
1105                                 disable_irq(adapter->tx_scrq[i]->irq);
1106                         }
1107         }
1108
1109         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1110         if (rc)
1111                 return rc;
1112
1113         if (adapter->rx_scrq) {
1114                 for (i = 0; i < adapter->req_rx_queues; i++) {
1115                         int retries = 10;
1116
1117                         while (pending_scrq(adapter, adapter->rx_scrq[i])) {
1118                                 retries--;
1119                                 mdelay(100);
1120
1121                                 if (retries == 0)
1122                                         break;
1123                         }
1124
1125                         if (adapter->rx_scrq[i]->irq) {
1126                                 netdev_dbg(adapter->netdev,
1127                                            "Disabling rx_scrq[%d] irq\n", i);
1128                                 disable_irq(adapter->rx_scrq[i]->irq);
1129                         }
1130                 }
1131         }
1132
1133         clean_tx_pools(adapter);
1134         adapter->state = VNIC_CLOSED;
1135         return rc;
1136 }
1137
1138 static int ibmvnic_close(struct net_device *netdev)
1139 {
1140         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1141         int rc;
1142
1143         mutex_lock(&adapter->reset_lock);
1144         rc = __ibmvnic_close(netdev);
1145         mutex_unlock(&adapter->reset_lock);
1146
1147         return rc;
1148 }
1149
1150 /**
1151  * build_hdr_data - creates L2/L3/L4 header data buffer
1152  * @hdr_field - bitfield determining needed headers
1153  * @skb - socket buffer
1154  * @hdr_len - array of header lengths
1155  * @tot_len - total length of data
1156  *
1157  * Reads hdr_field to determine which headers are needed by firmware.
1158  * Builds a buffer containing these headers.  Saves individual header
1159  * lengths and total buffer length to be used to build descriptors.
1160  */
1161 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1162                           int *hdr_len, u8 *hdr_data)
1163 {
1164         int len = 0;
1165         u8 *hdr;
1166
1167         hdr_len[0] = sizeof(struct ethhdr);
1168
1169         if (skb->protocol == htons(ETH_P_IP)) {
1170                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1171                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1172                         hdr_len[2] = tcp_hdrlen(skb);
1173                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1174                         hdr_len[2] = sizeof(struct udphdr);
1175         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1176                 hdr_len[1] = sizeof(struct ipv6hdr);
1177                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1178                         hdr_len[2] = tcp_hdrlen(skb);
1179                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1180                         hdr_len[2] = sizeof(struct udphdr);
1181         } else if (skb->protocol == htons(ETH_P_ARP)) {
1182                 hdr_len[1] = arp_hdr_len(skb->dev);
1183                 hdr_len[2] = 0;
1184         }
1185
1186         memset(hdr_data, 0, 120);
1187         if ((hdr_field >> 6) & 1) {
1188                 hdr = skb_mac_header(skb);
1189                 memcpy(hdr_data, hdr, hdr_len[0]);
1190                 len += hdr_len[0];
1191         }
1192
1193         if ((hdr_field >> 5) & 1) {
1194                 hdr = skb_network_header(skb);
1195                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1196                 len += hdr_len[1];
1197         }
1198
1199         if ((hdr_field >> 4) & 1) {
1200                 hdr = skb_transport_header(skb);
1201                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1202                 len += hdr_len[2];
1203         }
1204         return len;
1205 }
1206
1207 /**
1208  * create_hdr_descs - create header and header extension descriptors
1209  * @hdr_field - bitfield determining needed headers
1210  * @data - buffer containing header data
1211  * @len - length of data buffer
1212  * @hdr_len - array of individual header lengths
1213  * @scrq_arr - descriptor array
1214  *
1215  * Creates header and, if needed, header extension descriptors and
1216  * places them in a descriptor array, scrq_arr
1217  */
1218
1219 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1220                             union sub_crq *scrq_arr)
1221 {
1222         union sub_crq hdr_desc;
1223         int tmp_len = len;
1224         int num_descs = 0;
1225         u8 *data, *cur;
1226         int tmp;
1227
1228         while (tmp_len > 0) {
1229                 cur = hdr_data + len - tmp_len;
1230
1231                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1232                 if (cur != hdr_data) {
1233                         data = hdr_desc.hdr_ext.data;
1234                         tmp = tmp_len > 29 ? 29 : tmp_len;
1235                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1236                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1237                         hdr_desc.hdr_ext.len = tmp;
1238                 } else {
1239                         data = hdr_desc.hdr.data;
1240                         tmp = tmp_len > 24 ? 24 : tmp_len;
1241                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1242                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1243                         hdr_desc.hdr.len = tmp;
1244                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1245                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1246                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1247                         hdr_desc.hdr.flag = hdr_field << 1;
1248                 }
1249                 memcpy(data, cur, tmp);
1250                 tmp_len -= tmp;
1251                 *scrq_arr = hdr_desc;
1252                 scrq_arr++;
1253                 num_descs++;
1254         }
1255
1256         return num_descs;
1257 }
1258
1259 /**
1260  * build_hdr_descs_arr - build a header descriptor array
1261  * @skb - socket buffer
1262  * @num_entries - number of descriptors to be sent
1263  * @subcrq - first TX descriptor
1264  * @hdr_field - bit field determining which headers will be sent
1265  *
1266  * This function will build a TX descriptor array with applicable
1267  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1268  */
1269
1270 static void build_hdr_descs_arr(struct ibmvnic_tx_buff *txbuff,
1271                                 int *num_entries, u8 hdr_field)
1272 {
1273         int hdr_len[3] = {0, 0, 0};
1274         int tot_len;
1275         u8 *hdr_data = txbuff->hdr_data;
1276
1277         tot_len = build_hdr_data(hdr_field, txbuff->skb, hdr_len,
1278                                  txbuff->hdr_data);
1279         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1280                          txbuff->indir_arr + 1);
1281 }
1282
1283 static int ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1284 {
1285         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1286         int queue_num = skb_get_queue_mapping(skb);
1287         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1288         struct device *dev = &adapter->vdev->dev;
1289         struct ibmvnic_tx_buff *tx_buff = NULL;
1290         struct ibmvnic_sub_crq_queue *tx_scrq;
1291         struct ibmvnic_tx_pool *tx_pool;
1292         unsigned int tx_send_failed = 0;
1293         unsigned int tx_map_failed = 0;
1294         unsigned int tx_dropped = 0;
1295         unsigned int tx_packets = 0;
1296         unsigned int tx_bytes = 0;
1297         dma_addr_t data_dma_addr;
1298         struct netdev_queue *txq;
1299         unsigned long lpar_rc;
1300         union sub_crq tx_crq;
1301         unsigned int offset;
1302         int num_entries = 1;
1303         unsigned char *dst;
1304         u64 *handle_array;
1305         int index = 0;
1306         u8 proto = 0;
1307         int ret = 0;
1308
1309         if (adapter->resetting) {
1310                 if (!netif_subqueue_stopped(netdev, skb))
1311                         netif_stop_subqueue(netdev, queue_num);
1312                 dev_kfree_skb_any(skb);
1313
1314                 tx_send_failed++;
1315                 tx_dropped++;
1316                 ret = NETDEV_TX_OK;
1317                 goto out;
1318         }
1319
1320         tx_pool = &adapter->tx_pool[queue_num];
1321         tx_scrq = adapter->tx_scrq[queue_num];
1322         txq = netdev_get_tx_queue(netdev, skb_get_queue_mapping(skb));
1323         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
1324                 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
1325
1326         index = tx_pool->free_map[tx_pool->consumer_index];
1327
1328         if (skb_is_gso(skb)) {
1329                 offset = tx_pool->tso_index * IBMVNIC_TSO_BUF_SZ;
1330                 dst = tx_pool->tso_ltb.buff + offset;
1331                 memset(dst, 0, IBMVNIC_TSO_BUF_SZ);
1332                 data_dma_addr = tx_pool->tso_ltb.addr + offset;
1333                 tx_pool->tso_index++;
1334                 if (tx_pool->tso_index == IBMVNIC_TSO_BUFS)
1335                         tx_pool->tso_index = 0;
1336         } else {
1337                 offset = index * adapter->req_mtu;
1338                 dst = tx_pool->long_term_buff.buff + offset;
1339                 memset(dst, 0, adapter->req_mtu);
1340                 data_dma_addr = tx_pool->long_term_buff.addr + offset;
1341         }
1342
1343         if (skb_shinfo(skb)->nr_frags) {
1344                 int cur, i;
1345
1346                 /* Copy the head */
1347                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1348                 cur = skb_headlen(skb);
1349
1350                 /* Copy the frags */
1351                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1352                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1353
1354                         memcpy(dst + cur,
1355                                page_address(skb_frag_page(frag)) +
1356                                frag->page_offset, skb_frag_size(frag));
1357                         cur += skb_frag_size(frag);
1358                 }
1359         } else {
1360                 skb_copy_from_linear_data(skb, dst, skb->len);
1361         }
1362
1363         tx_pool->consumer_index =
1364             (tx_pool->consumer_index + 1) %
1365                 adapter->req_tx_entries_per_subcrq;
1366
1367         tx_buff = &tx_pool->tx_buff[index];
1368         tx_buff->skb = skb;
1369         tx_buff->data_dma[0] = data_dma_addr;
1370         tx_buff->data_len[0] = skb->len;
1371         tx_buff->index = index;
1372         tx_buff->pool_index = queue_num;
1373         tx_buff->last_frag = true;
1374
1375         memset(&tx_crq, 0, sizeof(tx_crq));
1376         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1377         tx_crq.v1.type = IBMVNIC_TX_DESC;
1378         tx_crq.v1.n_crq_elem = 1;
1379         tx_crq.v1.n_sge = 1;
1380         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1381         tx_crq.v1.correlator = cpu_to_be32(index);
1382         if (skb_is_gso(skb))
1383                 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->tso_ltb.map_id);
1384         else
1385                 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1386         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1387         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1388
1389         if (adapter->vlan_header_insertion) {
1390                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1391                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1392         }
1393
1394         if (skb->protocol == htons(ETH_P_IP)) {
1395                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1396                 proto = ip_hdr(skb)->protocol;
1397         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1398                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1399                 proto = ipv6_hdr(skb)->nexthdr;
1400         }
1401
1402         if (proto == IPPROTO_TCP)
1403                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1404         else if (proto == IPPROTO_UDP)
1405                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1406
1407         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1408                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1409                 hdrs += 2;
1410         }
1411         if (skb_is_gso(skb)) {
1412                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1413                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1414                 hdrs += 2;
1415         }
1416         /* determine if l2/3/4 headers are sent to firmware */
1417         if ((*hdrs >> 7) & 1 &&
1418             (skb->protocol == htons(ETH_P_IP) ||
1419              skb->protocol == htons(ETH_P_IPV6) ||
1420              skb->protocol == htons(ETH_P_ARP))) {
1421                 build_hdr_descs_arr(tx_buff, &num_entries, *hdrs);
1422                 tx_crq.v1.n_crq_elem = num_entries;
1423                 tx_buff->indir_arr[0] = tx_crq;
1424                 tx_buff->indir_dma = dma_map_single(dev, tx_buff->indir_arr,
1425                                                     sizeof(tx_buff->indir_arr),
1426                                                     DMA_TO_DEVICE);
1427                 if (dma_mapping_error(dev, tx_buff->indir_dma)) {
1428                         dev_kfree_skb_any(skb);
1429                         tx_buff->skb = NULL;
1430                         if (!firmware_has_feature(FW_FEATURE_CMO))
1431                                 dev_err(dev, "tx: unable to map descriptor array\n");
1432                         tx_map_failed++;
1433                         tx_dropped++;
1434                         ret = NETDEV_TX_OK;
1435                         goto out;
1436                 }
1437                 lpar_rc = send_subcrq_indirect(adapter, handle_array[queue_num],
1438                                                (u64)tx_buff->indir_dma,
1439                                                (u64)num_entries);
1440         } else {
1441                 lpar_rc = send_subcrq(adapter, handle_array[queue_num],
1442                                       &tx_crq);
1443         }
1444         if (lpar_rc != H_SUCCESS) {
1445                 dev_err(dev, "tx failed with code %ld\n", lpar_rc);
1446
1447                 if (tx_pool->consumer_index == 0)
1448                         tx_pool->consumer_index =
1449                                 adapter->req_tx_entries_per_subcrq - 1;
1450                 else
1451                         tx_pool->consumer_index--;
1452
1453                 dev_kfree_skb_any(skb);
1454                 tx_buff->skb = NULL;
1455
1456                 if (lpar_rc == H_CLOSED) {
1457                         /* Disable TX and report carrier off if queue is closed.
1458                          * Firmware guarantees that a signal will be sent to the
1459                          * driver, triggering a reset or some other action.
1460                          */
1461                         netif_tx_stop_all_queues(netdev);
1462                         netif_carrier_off(netdev);
1463                 }
1464
1465                 tx_send_failed++;
1466                 tx_dropped++;
1467                 ret = NETDEV_TX_OK;
1468                 goto out;
1469         }
1470
1471         if (atomic_inc_return(&tx_scrq->used)
1472                                         >= adapter->req_tx_entries_per_subcrq) {
1473                 netdev_info(netdev, "Stopping queue %d\n", queue_num);
1474                 netif_stop_subqueue(netdev, queue_num);
1475         }
1476
1477         tx_packets++;
1478         tx_bytes += skb->len;
1479         txq->trans_start = jiffies;
1480         ret = NETDEV_TX_OK;
1481
1482 out:
1483         netdev->stats.tx_dropped += tx_dropped;
1484         netdev->stats.tx_bytes += tx_bytes;
1485         netdev->stats.tx_packets += tx_packets;
1486         adapter->tx_send_failed += tx_send_failed;
1487         adapter->tx_map_failed += tx_map_failed;
1488         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1489         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1490         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1491
1492         return ret;
1493 }
1494
1495 static void ibmvnic_set_multi(struct net_device *netdev)
1496 {
1497         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1498         struct netdev_hw_addr *ha;
1499         union ibmvnic_crq crq;
1500
1501         memset(&crq, 0, sizeof(crq));
1502         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1503         crq.request_capability.cmd = REQUEST_CAPABILITY;
1504
1505         if (netdev->flags & IFF_PROMISC) {
1506                 if (!adapter->promisc_supported)
1507                         return;
1508         } else {
1509                 if (netdev->flags & IFF_ALLMULTI) {
1510                         /* Accept all multicast */
1511                         memset(&crq, 0, sizeof(crq));
1512                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1513                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1514                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1515                         ibmvnic_send_crq(adapter, &crq);
1516                 } else if (netdev_mc_empty(netdev)) {
1517                         /* Reject all multicast */
1518                         memset(&crq, 0, sizeof(crq));
1519                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1520                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1521                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1522                         ibmvnic_send_crq(adapter, &crq);
1523                 } else {
1524                         /* Accept one or more multicast(s) */
1525                         netdev_for_each_mc_addr(ha, netdev) {
1526                                 memset(&crq, 0, sizeof(crq));
1527                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1528                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1529                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1530                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1531                                                 ha->addr);
1532                                 ibmvnic_send_crq(adapter, &crq);
1533                         }
1534                 }
1535         }
1536 }
1537
1538 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p)
1539 {
1540         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1541         struct sockaddr *addr = p;
1542         union ibmvnic_crq crq;
1543
1544         if (!is_valid_ether_addr(addr->sa_data))
1545                 return -EADDRNOTAVAIL;
1546
1547         memset(&crq, 0, sizeof(crq));
1548         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1549         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1550         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], addr->sa_data);
1551
1552         init_completion(&adapter->fw_done);
1553         ibmvnic_send_crq(adapter, &crq);
1554         wait_for_completion(&adapter->fw_done);
1555         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1556         return adapter->fw_done_rc ? -EIO : 0;
1557 }
1558
1559 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1560 {
1561         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1562         struct sockaddr *addr = p;
1563         int rc;
1564
1565         if (adapter->state == VNIC_PROBED) {
1566                 memcpy(&adapter->desired.mac, addr, sizeof(struct sockaddr));
1567                 adapter->mac_change_pending = true;
1568                 return 0;
1569         }
1570
1571         rc = __ibmvnic_set_mac(netdev, addr);
1572
1573         return rc;
1574 }
1575
1576 /**
1577  * do_reset returns zero if we are able to keep processing reset events, or
1578  * non-zero if we hit a fatal error and must halt.
1579  */
1580 static int do_reset(struct ibmvnic_adapter *adapter,
1581                     struct ibmvnic_rwi *rwi, u32 reset_state)
1582 {
1583         u64 old_num_rx_queues, old_num_tx_queues;
1584         struct net_device *netdev = adapter->netdev;
1585         int i, rc;
1586
1587         netdev_dbg(adapter->netdev, "Re-setting driver (%d)\n",
1588                    rwi->reset_reason);
1589
1590         netif_carrier_off(netdev);
1591         adapter->reset_reason = rwi->reset_reason;
1592
1593         old_num_rx_queues = adapter->req_rx_queues;
1594         old_num_tx_queues = adapter->req_tx_queues;
1595
1596         if (rwi->reset_reason == VNIC_RESET_MOBILITY) {
1597                 rc = ibmvnic_reenable_crq_queue(adapter);
1598                 if (rc)
1599                         return 0;
1600         }
1601
1602         rc = __ibmvnic_close(netdev);
1603         if (rc)
1604                 return rc;
1605
1606         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM ||
1607             adapter->wait_for_reset) {
1608                 release_resources(adapter);
1609                 release_sub_crqs(adapter);
1610                 release_crq_queue(adapter);
1611         }
1612
1613         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
1614                 /* remove the closed state so when we call open it appears
1615                  * we are coming from the probed state.
1616                  */
1617                 adapter->state = VNIC_PROBED;
1618
1619                 rc = ibmvnic_init(adapter);
1620                 if (rc)
1621                         return IBMVNIC_INIT_FAILED;
1622
1623                 /* If the adapter was in PROBE state prior to the reset,
1624                  * exit here.
1625                  */
1626                 if (reset_state == VNIC_PROBED)
1627                         return 0;
1628
1629                 rc = ibmvnic_login(netdev);
1630                 if (rc) {
1631                         adapter->state = VNIC_PROBED;
1632                         return 0;
1633                 }
1634
1635                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM ||
1636                     adapter->wait_for_reset) {
1637                         rc = init_resources(adapter);
1638                         if (rc)
1639                                 return rc;
1640                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
1641                            adapter->req_tx_queues != old_num_tx_queues) {
1642                         release_rx_pools(adapter);
1643                         release_tx_pools(adapter);
1644                         init_rx_pools(netdev);
1645                         init_tx_pools(netdev);
1646                 } else {
1647                         rc = reset_tx_pools(adapter);
1648                         if (rc)
1649                                 return rc;
1650
1651                         rc = reset_rx_pools(adapter);
1652                         if (rc)
1653                                 return rc;
1654
1655                         if (reset_state == VNIC_CLOSED)
1656                                 return 0;
1657                 }
1658         }
1659
1660         rc = __ibmvnic_open(netdev);
1661         if (rc) {
1662                 if (list_empty(&adapter->rwi_list))
1663                         adapter->state = VNIC_CLOSED;
1664                 else
1665                         adapter->state = reset_state;
1666
1667                 return 0;
1668         }
1669
1670         netif_carrier_on(netdev);
1671
1672         /* kick napi */
1673         for (i = 0; i < adapter->req_rx_queues; i++)
1674                 napi_schedule(&adapter->napi[i]);
1675
1676         if (adapter->reset_reason != VNIC_RESET_FAILOVER)
1677                 netdev_notify_peers(netdev);
1678
1679         return 0;
1680 }
1681
1682 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
1683 {
1684         struct ibmvnic_rwi *rwi;
1685
1686         mutex_lock(&adapter->rwi_lock);
1687
1688         if (!list_empty(&adapter->rwi_list)) {
1689                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
1690                                        list);
1691                 list_del(&rwi->list);
1692         } else {
1693                 rwi = NULL;
1694         }
1695
1696         mutex_unlock(&adapter->rwi_lock);
1697         return rwi;
1698 }
1699
1700 static void free_all_rwi(struct ibmvnic_adapter *adapter)
1701 {
1702         struct ibmvnic_rwi *rwi;
1703
1704         rwi = get_next_rwi(adapter);
1705         while (rwi) {
1706                 kfree(rwi);
1707                 rwi = get_next_rwi(adapter);
1708         }
1709 }
1710
1711 static void __ibmvnic_reset(struct work_struct *work)
1712 {
1713         struct ibmvnic_rwi *rwi;
1714         struct ibmvnic_adapter *adapter;
1715         struct net_device *netdev;
1716         u32 reset_state;
1717         int rc = 0;
1718
1719         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
1720         netdev = adapter->netdev;
1721
1722         mutex_lock(&adapter->reset_lock);
1723         adapter->resetting = true;
1724         reset_state = adapter->state;
1725
1726         rwi = get_next_rwi(adapter);
1727         while (rwi) {
1728                 rc = do_reset(adapter, rwi, reset_state);
1729                 kfree(rwi);
1730                 if (rc && rc != IBMVNIC_INIT_FAILED)
1731                         break;
1732
1733                 rwi = get_next_rwi(adapter);
1734         }
1735
1736         if (adapter->wait_for_reset) {
1737                 adapter->wait_for_reset = false;
1738                 adapter->reset_done_rc = rc;
1739                 complete(&adapter->reset_done);
1740         }
1741
1742         if (rc) {
1743                 netdev_dbg(adapter->netdev, "Reset failed\n");
1744                 free_all_rwi(adapter);
1745                 mutex_unlock(&adapter->reset_lock);
1746                 return;
1747         }
1748
1749         adapter->resetting = false;
1750         mutex_unlock(&adapter->reset_lock);
1751 }
1752
1753 static void ibmvnic_reset(struct ibmvnic_adapter *adapter,
1754                           enum ibmvnic_reset_reason reason)
1755 {
1756         struct ibmvnic_rwi *rwi, *tmp;
1757         struct net_device *netdev = adapter->netdev;
1758         struct list_head *entry;
1759
1760         if (adapter->state == VNIC_REMOVING ||
1761             adapter->state == VNIC_REMOVED) {
1762                 netdev_dbg(netdev, "Adapter removing, skipping reset\n");
1763                 return;
1764         }
1765
1766         if (adapter->state == VNIC_PROBING) {
1767                 netdev_warn(netdev, "Adapter reset during probe\n");
1768                 adapter->init_done_rc = EAGAIN;
1769                 return;
1770         }
1771
1772         mutex_lock(&adapter->rwi_lock);
1773
1774         list_for_each(entry, &adapter->rwi_list) {
1775                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
1776                 if (tmp->reset_reason == reason) {
1777                         netdev_dbg(netdev, "Skipping matching reset\n");
1778                         mutex_unlock(&adapter->rwi_lock);
1779                         return;
1780                 }
1781         }
1782
1783         rwi = kzalloc(sizeof(*rwi), GFP_KERNEL);
1784         if (!rwi) {
1785                 mutex_unlock(&adapter->rwi_lock);
1786                 ibmvnic_close(netdev);
1787                 return;
1788         }
1789
1790         rwi->reset_reason = reason;
1791         list_add_tail(&rwi->list, &adapter->rwi_list);
1792         mutex_unlock(&adapter->rwi_lock);
1793
1794         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
1795         schedule_work(&adapter->ibmvnic_reset);
1796 }
1797
1798 static void ibmvnic_tx_timeout(struct net_device *dev)
1799 {
1800         struct ibmvnic_adapter *adapter = netdev_priv(dev);
1801
1802         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
1803 }
1804
1805 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
1806                                   struct ibmvnic_rx_buff *rx_buff)
1807 {
1808         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
1809
1810         rx_buff->skb = NULL;
1811
1812         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
1813         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
1814
1815         atomic_dec(&pool->available);
1816 }
1817
1818 static int ibmvnic_poll(struct napi_struct *napi, int budget)
1819 {
1820         struct net_device *netdev = napi->dev;
1821         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1822         int scrq_num = (int)(napi - adapter->napi);
1823         int frames_processed = 0;
1824
1825 restart_poll:
1826         while (frames_processed < budget) {
1827                 struct sk_buff *skb;
1828                 struct ibmvnic_rx_buff *rx_buff;
1829                 union sub_crq *next;
1830                 u32 length;
1831                 u16 offset;
1832                 u8 flags = 0;
1833
1834                 if (unlikely(adapter->resetting)) {
1835                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
1836                         napi_complete_done(napi, frames_processed);
1837                         return frames_processed;
1838                 }
1839
1840                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
1841                         break;
1842                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
1843                 rx_buff =
1844                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
1845                                                           rx_comp.correlator);
1846                 /* do error checking */
1847                 if (next->rx_comp.rc) {
1848                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
1849                                    be16_to_cpu(next->rx_comp.rc));
1850                         /* free the entry */
1851                         next->rx_comp.first = 0;
1852                         remove_buff_from_pool(adapter, rx_buff);
1853                         continue;
1854                 }
1855
1856                 length = be32_to_cpu(next->rx_comp.len);
1857                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
1858                 flags = next->rx_comp.flags;
1859                 skb = rx_buff->skb;
1860                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
1861                                         length);
1862
1863                 /* VLAN Header has been stripped by the system firmware and
1864                  * needs to be inserted by the driver
1865                  */
1866                 if (adapter->rx_vlan_header_insertion &&
1867                     (flags & IBMVNIC_VLAN_STRIPPED))
1868                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1869                                                ntohs(next->rx_comp.vlan_tci));
1870
1871                 /* free the entry */
1872                 next->rx_comp.first = 0;
1873                 remove_buff_from_pool(adapter, rx_buff);
1874
1875                 skb_put(skb, length);
1876                 skb->protocol = eth_type_trans(skb, netdev);
1877                 skb_record_rx_queue(skb, scrq_num);
1878
1879                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
1880                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
1881                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1882                 }
1883
1884                 length = skb->len;
1885                 napi_gro_receive(napi, skb); /* send it up */
1886                 netdev->stats.rx_packets++;
1887                 netdev->stats.rx_bytes += length;
1888                 adapter->rx_stats_buffers[scrq_num].packets++;
1889                 adapter->rx_stats_buffers[scrq_num].bytes += length;
1890                 frames_processed++;
1891         }
1892
1893         if (adapter->state != VNIC_CLOSING)
1894                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
1895
1896         if (frames_processed < budget) {
1897                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
1898                 napi_complete_done(napi, frames_processed);
1899                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
1900                     napi_reschedule(napi)) {
1901                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
1902                         goto restart_poll;
1903                 }
1904         }
1905         return frames_processed;
1906 }
1907
1908 #ifdef CONFIG_NET_POLL_CONTROLLER
1909 static void ibmvnic_netpoll_controller(struct net_device *dev)
1910 {
1911         struct ibmvnic_adapter *adapter = netdev_priv(dev);
1912         int i;
1913
1914         replenish_pools(netdev_priv(dev));
1915         for (i = 0; i < adapter->req_rx_queues; i++)
1916                 ibmvnic_interrupt_rx(adapter->rx_scrq[i]->irq,
1917                                      adapter->rx_scrq[i]);
1918 }
1919 #endif
1920
1921 static int wait_for_reset(struct ibmvnic_adapter *adapter)
1922 {
1923         adapter->fallback.mtu = adapter->req_mtu;
1924         adapter->fallback.rx_queues = adapter->req_rx_queues;
1925         adapter->fallback.tx_queues = adapter->req_tx_queues;
1926         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
1927         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
1928
1929         init_completion(&adapter->reset_done);
1930         ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
1931         adapter->wait_for_reset = true;
1932         wait_for_completion(&adapter->reset_done);
1933
1934         if (adapter->reset_done_rc) {
1935                 adapter->desired.mtu = adapter->fallback.mtu;
1936                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
1937                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
1938                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
1939                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
1940
1941                 init_completion(&adapter->reset_done);
1942                 ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
1943                 wait_for_completion(&adapter->reset_done);
1944         }
1945         adapter->wait_for_reset = false;
1946
1947         return adapter->reset_done_rc;
1948 }
1949
1950 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
1951 {
1952         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1953
1954         adapter->desired.mtu = new_mtu + ETH_HLEN;
1955
1956         return wait_for_reset(adapter);
1957 }
1958
1959 static const struct net_device_ops ibmvnic_netdev_ops = {
1960         .ndo_open               = ibmvnic_open,
1961         .ndo_stop               = ibmvnic_close,
1962         .ndo_start_xmit         = ibmvnic_xmit,
1963         .ndo_set_rx_mode        = ibmvnic_set_multi,
1964         .ndo_set_mac_address    = ibmvnic_set_mac,
1965         .ndo_validate_addr      = eth_validate_addr,
1966         .ndo_tx_timeout         = ibmvnic_tx_timeout,
1967 #ifdef CONFIG_NET_POLL_CONTROLLER
1968         .ndo_poll_controller    = ibmvnic_netpoll_controller,
1969 #endif
1970         .ndo_change_mtu         = ibmvnic_change_mtu,
1971 };
1972
1973 /* ethtool functions */
1974
1975 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
1976                                       struct ethtool_link_ksettings *cmd)
1977 {
1978         u32 supported, advertising;
1979
1980         supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
1981                           SUPPORTED_FIBRE);
1982         advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
1983                             ADVERTISED_FIBRE);
1984         cmd->base.speed = SPEED_1000;
1985         cmd->base.duplex = DUPLEX_FULL;
1986         cmd->base.port = PORT_FIBRE;
1987         cmd->base.phy_address = 0;
1988         cmd->base.autoneg = AUTONEG_ENABLE;
1989
1990         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1991                                                 supported);
1992         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
1993                                                 advertising);
1994
1995         return 0;
1996 }
1997
1998 static void ibmvnic_get_drvinfo(struct net_device *netdev,
1999                                 struct ethtool_drvinfo *info)
2000 {
2001         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2002
2003         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2004         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2005         strlcpy(info->fw_version, adapter->fw_version,
2006                 sizeof(info->fw_version));
2007 }
2008
2009 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2010 {
2011         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2012
2013         return adapter->msg_enable;
2014 }
2015
2016 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2017 {
2018         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2019
2020         adapter->msg_enable = data;
2021 }
2022
2023 static u32 ibmvnic_get_link(struct net_device *netdev)
2024 {
2025         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2026
2027         /* Don't need to send a query because we request a logical link up at
2028          * init and then we wait for link state indications
2029          */
2030         return adapter->logical_link_state;
2031 }
2032
2033 static void ibmvnic_get_ringparam(struct net_device *netdev,
2034                                   struct ethtool_ringparam *ring)
2035 {
2036         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2037
2038         ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2039         ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2040         ring->rx_mini_max_pending = 0;
2041         ring->rx_jumbo_max_pending = 0;
2042         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2043         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2044         ring->rx_mini_pending = 0;
2045         ring->rx_jumbo_pending = 0;
2046 }
2047
2048 static int ibmvnic_set_ringparam(struct net_device *netdev,
2049                                  struct ethtool_ringparam *ring)
2050 {
2051         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2052
2053         if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
2054             ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
2055                 netdev_err(netdev, "Invalid request.\n");
2056                 netdev_err(netdev, "Max tx buffers = %llu\n",
2057                            adapter->max_rx_add_entries_per_subcrq);
2058                 netdev_err(netdev, "Max rx buffers = %llu\n",
2059                            adapter->max_tx_entries_per_subcrq);
2060                 return -EINVAL;
2061         }
2062
2063         adapter->desired.rx_entries = ring->rx_pending;
2064         adapter->desired.tx_entries = ring->tx_pending;
2065
2066         return wait_for_reset(adapter);
2067 }
2068
2069 static void ibmvnic_get_channels(struct net_device *netdev,
2070                                  struct ethtool_channels *channels)
2071 {
2072         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2073
2074         channels->max_rx = adapter->max_rx_queues;
2075         channels->max_tx = adapter->max_tx_queues;
2076         channels->max_other = 0;
2077         channels->max_combined = 0;
2078         channels->rx_count = adapter->req_rx_queues;
2079         channels->tx_count = adapter->req_tx_queues;
2080         channels->other_count = 0;
2081         channels->combined_count = 0;
2082 }
2083
2084 static int ibmvnic_set_channels(struct net_device *netdev,
2085                                 struct ethtool_channels *channels)
2086 {
2087         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2088
2089         adapter->desired.rx_queues = channels->rx_count;
2090         adapter->desired.tx_queues = channels->tx_count;
2091
2092         return wait_for_reset(adapter);
2093 }
2094
2095 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2096 {
2097         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2098         int i;
2099
2100         if (stringset != ETH_SS_STATS)
2101                 return;
2102
2103         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
2104                 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2105
2106         for (i = 0; i < adapter->req_tx_queues; i++) {
2107                 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2108                 data += ETH_GSTRING_LEN;
2109
2110                 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2111                 data += ETH_GSTRING_LEN;
2112
2113                 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
2114                 data += ETH_GSTRING_LEN;
2115         }
2116
2117         for (i = 0; i < adapter->req_rx_queues; i++) {
2118                 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2119                 data += ETH_GSTRING_LEN;
2120
2121                 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2122                 data += ETH_GSTRING_LEN;
2123
2124                 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2125                 data += ETH_GSTRING_LEN;
2126         }
2127 }
2128
2129 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2130 {
2131         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2132
2133         switch (sset) {
2134         case ETH_SS_STATS:
2135                 return ARRAY_SIZE(ibmvnic_stats) +
2136                        adapter->req_tx_queues * NUM_TX_STATS +
2137                        adapter->req_rx_queues * NUM_RX_STATS;
2138         default:
2139                 return -EOPNOTSUPP;
2140         }
2141 }
2142
2143 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2144                                       struct ethtool_stats *stats, u64 *data)
2145 {
2146         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2147         union ibmvnic_crq crq;
2148         int i, j;
2149
2150         memset(&crq, 0, sizeof(crq));
2151         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2152         crq.request_statistics.cmd = REQUEST_STATISTICS;
2153         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2154         crq.request_statistics.len =
2155             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2156
2157         /* Wait for data to be written */
2158         init_completion(&adapter->stats_done);
2159         ibmvnic_send_crq(adapter, &crq);
2160         wait_for_completion(&adapter->stats_done);
2161
2162         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2163                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2164                                                 ibmvnic_stats[i].offset));
2165
2166         for (j = 0; j < adapter->req_tx_queues; j++) {
2167                 data[i] = adapter->tx_stats_buffers[j].packets;
2168                 i++;
2169                 data[i] = adapter->tx_stats_buffers[j].bytes;
2170                 i++;
2171                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2172                 i++;
2173         }
2174
2175         for (j = 0; j < adapter->req_rx_queues; j++) {
2176                 data[i] = adapter->rx_stats_buffers[j].packets;
2177                 i++;
2178                 data[i] = adapter->rx_stats_buffers[j].bytes;
2179                 i++;
2180                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2181                 i++;
2182         }
2183 }
2184
2185 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2186         .get_drvinfo            = ibmvnic_get_drvinfo,
2187         .get_msglevel           = ibmvnic_get_msglevel,
2188         .set_msglevel           = ibmvnic_set_msglevel,
2189         .get_link               = ibmvnic_get_link,
2190         .get_ringparam          = ibmvnic_get_ringparam,
2191         .set_ringparam          = ibmvnic_set_ringparam,
2192         .get_channels           = ibmvnic_get_channels,
2193         .set_channels           = ibmvnic_set_channels,
2194         .get_strings            = ibmvnic_get_strings,
2195         .get_sset_count         = ibmvnic_get_sset_count,
2196         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2197         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2198 };
2199
2200 /* Routines for managing CRQs/sCRQs  */
2201
2202 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2203                                    struct ibmvnic_sub_crq_queue *scrq)
2204 {
2205         int rc;
2206
2207         if (scrq->irq) {
2208                 free_irq(scrq->irq, scrq);
2209                 irq_dispose_mapping(scrq->irq);
2210                 scrq->irq = 0;
2211         }
2212
2213         memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2214         scrq->cur = 0;
2215
2216         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2217                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2218         return rc;
2219 }
2220
2221 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2222 {
2223         int i, rc;
2224
2225         for (i = 0; i < adapter->req_tx_queues; i++) {
2226                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2227                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2228                 if (rc)
2229                         return rc;
2230         }
2231
2232         for (i = 0; i < adapter->req_rx_queues; i++) {
2233                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2234                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2235                 if (rc)
2236                         return rc;
2237         }
2238
2239         return rc;
2240 }
2241
2242 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2243                                   struct ibmvnic_sub_crq_queue *scrq)
2244 {
2245         struct device *dev = &adapter->vdev->dev;
2246         long rc;
2247
2248         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2249
2250         /* Close the sub-crqs */
2251         do {
2252                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2253                                         adapter->vdev->unit_address,
2254                                         scrq->crq_num);
2255         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2256
2257         if (rc) {
2258                 netdev_err(adapter->netdev,
2259                            "Failed to release sub-CRQ %16lx, rc = %ld\n",
2260                            scrq->crq_num, rc);
2261         }
2262
2263         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2264                          DMA_BIDIRECTIONAL);
2265         free_pages((unsigned long)scrq->msgs, 2);
2266         kfree(scrq);
2267 }
2268
2269 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2270                                                         *adapter)
2271 {
2272         struct device *dev = &adapter->vdev->dev;
2273         struct ibmvnic_sub_crq_queue *scrq;
2274         int rc;
2275
2276         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2277         if (!scrq)
2278                 return NULL;
2279
2280         scrq->msgs =
2281                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2282         if (!scrq->msgs) {
2283                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2284                 goto zero_page_failed;
2285         }
2286
2287         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2288                                          DMA_BIDIRECTIONAL);
2289         if (dma_mapping_error(dev, scrq->msg_token)) {
2290                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2291                 goto map_failed;
2292         }
2293
2294         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2295                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2296
2297         if (rc == H_RESOURCE)
2298                 rc = ibmvnic_reset_crq(adapter);
2299
2300         if (rc == H_CLOSED) {
2301                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2302         } else if (rc) {
2303                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2304                 goto reg_failed;
2305         }
2306
2307         scrq->adapter = adapter;
2308         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2309         spin_lock_init(&scrq->lock);
2310
2311         netdev_dbg(adapter->netdev,
2312                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2313                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2314
2315         return scrq;
2316
2317 reg_failed:
2318         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2319                          DMA_BIDIRECTIONAL);
2320 map_failed:
2321         free_pages((unsigned long)scrq->msgs, 2);
2322 zero_page_failed:
2323         kfree(scrq);
2324
2325         return NULL;
2326 }
2327
2328 static void release_sub_crqs(struct ibmvnic_adapter *adapter)
2329 {
2330         int i;
2331
2332         if (adapter->tx_scrq) {
2333                 for (i = 0; i < adapter->req_tx_queues; i++) {
2334                         if (!adapter->tx_scrq[i])
2335                                 continue;
2336
2337                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
2338                                    i);
2339                         if (adapter->tx_scrq[i]->irq) {
2340                                 free_irq(adapter->tx_scrq[i]->irq,
2341                                          adapter->tx_scrq[i]);
2342                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
2343                                 adapter->tx_scrq[i]->irq = 0;
2344                         }
2345
2346                         release_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2347                 }
2348
2349                 kfree(adapter->tx_scrq);
2350                 adapter->tx_scrq = NULL;
2351         }
2352
2353         if (adapter->rx_scrq) {
2354                 for (i = 0; i < adapter->req_rx_queues; i++) {
2355                         if (!adapter->rx_scrq[i])
2356                                 continue;
2357
2358                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
2359                                    i);
2360                         if (adapter->rx_scrq[i]->irq) {
2361                                 free_irq(adapter->rx_scrq[i]->irq,
2362                                          adapter->rx_scrq[i]);
2363                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
2364                                 adapter->rx_scrq[i]->irq = 0;
2365                         }
2366
2367                         release_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2368                 }
2369
2370                 kfree(adapter->rx_scrq);
2371                 adapter->rx_scrq = NULL;
2372         }
2373 }
2374
2375 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
2376                             struct ibmvnic_sub_crq_queue *scrq)
2377 {
2378         struct device *dev = &adapter->vdev->dev;
2379         unsigned long rc;
2380
2381         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2382                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2383         if (rc)
2384                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
2385                         scrq->hw_irq, rc);
2386         return rc;
2387 }
2388
2389 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
2390                            struct ibmvnic_sub_crq_queue *scrq)
2391 {
2392         struct device *dev = &adapter->vdev->dev;
2393         unsigned long rc;
2394
2395         if (scrq->hw_irq > 0x100000000ULL) {
2396                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
2397                 return 1;
2398         }
2399
2400         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2401                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2402         if (rc)
2403                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
2404                         scrq->hw_irq, rc);
2405         return rc;
2406 }
2407
2408 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
2409                                struct ibmvnic_sub_crq_queue *scrq)
2410 {
2411         struct device *dev = &adapter->vdev->dev;
2412         struct ibmvnic_tx_buff *txbuff;
2413         union sub_crq *next;
2414         int index;
2415         int i, j;
2416         u8 first;
2417
2418 restart_loop:
2419         while (pending_scrq(adapter, scrq)) {
2420                 unsigned int pool = scrq->pool_index;
2421
2422                 next = ibmvnic_next_scrq(adapter, scrq);
2423                 for (i = 0; i < next->tx_comp.num_comps; i++) {
2424                         if (next->tx_comp.rcs[i]) {
2425                                 dev_err(dev, "tx error %x\n",
2426                                         next->tx_comp.rcs[i]);
2427                                 continue;
2428                         }
2429                         index = be32_to_cpu(next->tx_comp.correlators[i]);
2430                         txbuff = &adapter->tx_pool[pool].tx_buff[index];
2431
2432                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
2433                                 if (!txbuff->data_dma[j])
2434                                         continue;
2435
2436                                 txbuff->data_dma[j] = 0;
2437                         }
2438                         /* if sub_crq was sent indirectly */
2439                         first = txbuff->indir_arr[0].generic.first;
2440                         if (first == IBMVNIC_CRQ_CMD) {
2441                                 dma_unmap_single(dev, txbuff->indir_dma,
2442                                                  sizeof(txbuff->indir_arr),
2443                                                  DMA_TO_DEVICE);
2444                         }
2445
2446                         if (txbuff->last_frag) {
2447                                 dev_kfree_skb_any(txbuff->skb);
2448                                 txbuff->skb = NULL;
2449                         }
2450
2451                         adapter->tx_pool[pool].free_map[adapter->tx_pool[pool].
2452                                                      producer_index] = index;
2453                         adapter->tx_pool[pool].producer_index =
2454                             (adapter->tx_pool[pool].producer_index + 1) %
2455                             adapter->req_tx_entries_per_subcrq;
2456                 }
2457                 /* remove tx_comp scrq*/
2458                 next->tx_comp.first = 0;
2459
2460                 if (atomic_sub_return(next->tx_comp.num_comps, &scrq->used) <=
2461                     (adapter->req_tx_entries_per_subcrq / 2) &&
2462                     __netif_subqueue_stopped(adapter->netdev,
2463                                              scrq->pool_index)) {
2464                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
2465                         netdev_info(adapter->netdev, "Started queue %d\n",
2466                                     scrq->pool_index);
2467                 }
2468         }
2469
2470         enable_scrq_irq(adapter, scrq);
2471
2472         if (pending_scrq(adapter, scrq)) {
2473                 disable_scrq_irq(adapter, scrq);
2474                 goto restart_loop;
2475         }
2476
2477         return 0;
2478 }
2479
2480 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
2481 {
2482         struct ibmvnic_sub_crq_queue *scrq = instance;
2483         struct ibmvnic_adapter *adapter = scrq->adapter;
2484
2485         disable_scrq_irq(adapter, scrq);
2486         ibmvnic_complete_tx(adapter, scrq);
2487
2488         return IRQ_HANDLED;
2489 }
2490
2491 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
2492 {
2493         struct ibmvnic_sub_crq_queue *scrq = instance;
2494         struct ibmvnic_adapter *adapter = scrq->adapter;
2495
2496         /* When booting a kdump kernel we can hit pending interrupts
2497          * prior to completing driver initialization.
2498          */
2499         if (unlikely(adapter->state != VNIC_OPEN))
2500                 return IRQ_NONE;
2501
2502         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
2503
2504         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
2505                 disable_scrq_irq(adapter, scrq);
2506                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
2507         }
2508
2509         return IRQ_HANDLED;
2510 }
2511
2512 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
2513 {
2514         struct device *dev = &adapter->vdev->dev;
2515         struct ibmvnic_sub_crq_queue *scrq;
2516         int i = 0, j = 0;
2517         int rc = 0;
2518
2519         for (i = 0; i < adapter->req_tx_queues; i++) {
2520                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
2521                            i);
2522                 scrq = adapter->tx_scrq[i];
2523                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2524
2525                 if (!scrq->irq) {
2526                         rc = -EINVAL;
2527                         dev_err(dev, "Error mapping irq\n");
2528                         goto req_tx_irq_failed;
2529                 }
2530
2531                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
2532                                  0, "ibmvnic_tx", scrq);
2533
2534                 if (rc) {
2535                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
2536                                 scrq->irq, rc);
2537                         irq_dispose_mapping(scrq->irq);
2538                         goto req_rx_irq_failed;
2539                 }
2540         }
2541
2542         for (i = 0; i < adapter->req_rx_queues; i++) {
2543                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
2544                            i);
2545                 scrq = adapter->rx_scrq[i];
2546                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2547                 if (!scrq->irq) {
2548                         rc = -EINVAL;
2549                         dev_err(dev, "Error mapping irq\n");
2550                         goto req_rx_irq_failed;
2551                 }
2552                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
2553                                  0, "ibmvnic_rx", scrq);
2554                 if (rc) {
2555                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
2556                                 scrq->irq, rc);
2557                         irq_dispose_mapping(scrq->irq);
2558                         goto req_rx_irq_failed;
2559                 }
2560         }
2561         return rc;
2562
2563 req_rx_irq_failed:
2564         for (j = 0; j < i; j++) {
2565                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
2566                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2567         }
2568         i = adapter->req_tx_queues;
2569 req_tx_irq_failed:
2570         for (j = 0; j < i; j++) {
2571                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
2572                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2573         }
2574         release_sub_crqs(adapter);
2575         return rc;
2576 }
2577
2578 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
2579 {
2580         struct device *dev = &adapter->vdev->dev;
2581         struct ibmvnic_sub_crq_queue **allqueues;
2582         int registered_queues = 0;
2583         int total_queues;
2584         int more = 0;
2585         int i;
2586
2587         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
2588
2589         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
2590         if (!allqueues)
2591                 return -1;
2592
2593         for (i = 0; i < total_queues; i++) {
2594                 allqueues[i] = init_sub_crq_queue(adapter);
2595                 if (!allqueues[i]) {
2596                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
2597                         break;
2598                 }
2599                 registered_queues++;
2600         }
2601
2602         /* Make sure we were able to register the minimum number of queues */
2603         if (registered_queues <
2604             adapter->min_tx_queues + adapter->min_rx_queues) {
2605                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
2606                 goto tx_failed;
2607         }
2608
2609         /* Distribute the failed allocated queues*/
2610         for (i = 0; i < total_queues - registered_queues + more ; i++) {
2611                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
2612                 switch (i % 3) {
2613                 case 0:
2614                         if (adapter->req_rx_queues > adapter->min_rx_queues)
2615                                 adapter->req_rx_queues--;
2616                         else
2617                                 more++;
2618                         break;
2619                 case 1:
2620                         if (adapter->req_tx_queues > adapter->min_tx_queues)
2621                                 adapter->req_tx_queues--;
2622                         else
2623                                 more++;
2624                         break;
2625                 }
2626         }
2627
2628         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
2629                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
2630         if (!adapter->tx_scrq)
2631                 goto tx_failed;
2632
2633         for (i = 0; i < adapter->req_tx_queues; i++) {
2634                 adapter->tx_scrq[i] = allqueues[i];
2635                 adapter->tx_scrq[i]->pool_index = i;
2636         }
2637
2638         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
2639                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
2640         if (!adapter->rx_scrq)
2641                 goto rx_failed;
2642
2643         for (i = 0; i < adapter->req_rx_queues; i++) {
2644                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
2645                 adapter->rx_scrq[i]->scrq_num = i;
2646         }
2647
2648         kfree(allqueues);
2649         return 0;
2650
2651 rx_failed:
2652         kfree(adapter->tx_scrq);
2653         adapter->tx_scrq = NULL;
2654 tx_failed:
2655         for (i = 0; i < registered_queues; i++)
2656                 release_sub_crq_queue(adapter, allqueues[i]);
2657         kfree(allqueues);
2658         return -1;
2659 }
2660
2661 static void ibmvnic_send_req_caps(struct ibmvnic_adapter *adapter, int retry)
2662 {
2663         struct device *dev = &adapter->vdev->dev;
2664         union ibmvnic_crq crq;
2665         int max_entries;
2666
2667         if (!retry) {
2668                 /* Sub-CRQ entries are 32 byte long */
2669                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
2670
2671                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
2672                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
2673                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
2674                         return;
2675                 }
2676
2677                 if (adapter->desired.mtu)
2678                         adapter->req_mtu = adapter->desired.mtu;
2679                 else
2680                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
2681
2682                 if (!adapter->desired.tx_entries)
2683                         adapter->desired.tx_entries =
2684                                         adapter->max_tx_entries_per_subcrq;
2685                 if (!adapter->desired.rx_entries)
2686                         adapter->desired.rx_entries =
2687                                         adapter->max_rx_add_entries_per_subcrq;
2688
2689                 max_entries = IBMVNIC_MAX_LTB_SIZE /
2690                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
2691
2692                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
2693                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
2694                         adapter->desired.tx_entries = max_entries;
2695                 }
2696
2697                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
2698                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
2699                         adapter->desired.rx_entries = max_entries;
2700                 }
2701
2702                 if (adapter->desired.tx_entries)
2703                         adapter->req_tx_entries_per_subcrq =
2704                                         adapter->desired.tx_entries;
2705                 else
2706                         adapter->req_tx_entries_per_subcrq =
2707                                         adapter->max_tx_entries_per_subcrq;
2708
2709                 if (adapter->desired.rx_entries)
2710                         adapter->req_rx_add_entries_per_subcrq =
2711                                         adapter->desired.rx_entries;
2712                 else
2713                         adapter->req_rx_add_entries_per_subcrq =
2714                                         adapter->max_rx_add_entries_per_subcrq;
2715
2716                 if (adapter->desired.tx_queues)
2717                         adapter->req_tx_queues =
2718                                         adapter->desired.tx_queues;
2719                 else
2720                         adapter->req_tx_queues =
2721                                         adapter->opt_tx_comp_sub_queues;
2722
2723                 if (adapter->desired.rx_queues)
2724                         adapter->req_rx_queues =
2725                                         adapter->desired.rx_queues;
2726                 else
2727                         adapter->req_rx_queues =
2728                                         adapter->opt_rx_comp_queues;
2729
2730                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
2731         }
2732
2733         memset(&crq, 0, sizeof(crq));
2734         crq.request_capability.first = IBMVNIC_CRQ_CMD;
2735         crq.request_capability.cmd = REQUEST_CAPABILITY;
2736
2737         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
2738         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
2739         atomic_inc(&adapter->running_cap_crqs);
2740         ibmvnic_send_crq(adapter, &crq);
2741
2742         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
2743         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
2744         atomic_inc(&adapter->running_cap_crqs);
2745         ibmvnic_send_crq(adapter, &crq);
2746
2747         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
2748         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
2749         atomic_inc(&adapter->running_cap_crqs);
2750         ibmvnic_send_crq(adapter, &crq);
2751
2752         crq.request_capability.capability =
2753             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
2754         crq.request_capability.number =
2755             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
2756         atomic_inc(&adapter->running_cap_crqs);
2757         ibmvnic_send_crq(adapter, &crq);
2758
2759         crq.request_capability.capability =
2760             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
2761         crq.request_capability.number =
2762             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
2763         atomic_inc(&adapter->running_cap_crqs);
2764         ibmvnic_send_crq(adapter, &crq);
2765
2766         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
2767         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
2768         atomic_inc(&adapter->running_cap_crqs);
2769         ibmvnic_send_crq(adapter, &crq);
2770
2771         if (adapter->netdev->flags & IFF_PROMISC) {
2772                 if (adapter->promisc_supported) {
2773                         crq.request_capability.capability =
2774                             cpu_to_be16(PROMISC_REQUESTED);
2775                         crq.request_capability.number = cpu_to_be64(1);
2776                         atomic_inc(&adapter->running_cap_crqs);
2777                         ibmvnic_send_crq(adapter, &crq);
2778                 }
2779         } else {
2780                 crq.request_capability.capability =
2781                     cpu_to_be16(PROMISC_REQUESTED);
2782                 crq.request_capability.number = cpu_to_be64(0);
2783                 atomic_inc(&adapter->running_cap_crqs);
2784                 ibmvnic_send_crq(adapter, &crq);
2785         }
2786 }
2787
2788 static int pending_scrq(struct ibmvnic_adapter *adapter,
2789                         struct ibmvnic_sub_crq_queue *scrq)
2790 {
2791         union sub_crq *entry = &scrq->msgs[scrq->cur];
2792
2793         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
2794                 return 1;
2795         else
2796                 return 0;
2797 }
2798
2799 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
2800                                         struct ibmvnic_sub_crq_queue *scrq)
2801 {
2802         union sub_crq *entry;
2803         unsigned long flags;
2804
2805         spin_lock_irqsave(&scrq->lock, flags);
2806         entry = &scrq->msgs[scrq->cur];
2807         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
2808                 if (++scrq->cur == scrq->size)
2809                         scrq->cur = 0;
2810         } else {
2811                 entry = NULL;
2812         }
2813         spin_unlock_irqrestore(&scrq->lock, flags);
2814
2815         return entry;
2816 }
2817
2818 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
2819 {
2820         struct ibmvnic_crq_queue *queue = &adapter->crq;
2821         union ibmvnic_crq *crq;
2822
2823         crq = &queue->msgs[queue->cur];
2824         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
2825                 if (++queue->cur == queue->size)
2826                         queue->cur = 0;
2827         } else {
2828                 crq = NULL;
2829         }
2830
2831         return crq;
2832 }
2833
2834 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
2835                        union sub_crq *sub_crq)
2836 {
2837         unsigned int ua = adapter->vdev->unit_address;
2838         struct device *dev = &adapter->vdev->dev;
2839         u64 *u64_crq = (u64 *)sub_crq;
2840         int rc;
2841
2842         netdev_dbg(adapter->netdev,
2843                    "Sending sCRQ %016lx: %016lx %016lx %016lx %016lx\n",
2844                    (unsigned long int)cpu_to_be64(remote_handle),
2845                    (unsigned long int)cpu_to_be64(u64_crq[0]),
2846                    (unsigned long int)cpu_to_be64(u64_crq[1]),
2847                    (unsigned long int)cpu_to_be64(u64_crq[2]),
2848                    (unsigned long int)cpu_to_be64(u64_crq[3]));
2849
2850         /* Make sure the hypervisor sees the complete request */
2851         mb();
2852
2853         rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
2854                                 cpu_to_be64(remote_handle),
2855                                 cpu_to_be64(u64_crq[0]),
2856                                 cpu_to_be64(u64_crq[1]),
2857                                 cpu_to_be64(u64_crq[2]),
2858                                 cpu_to_be64(u64_crq[3]));
2859
2860         if (rc) {
2861                 if (rc == H_CLOSED)
2862                         dev_warn(dev, "CRQ Queue closed\n");
2863                 dev_err(dev, "Send error (rc=%d)\n", rc);
2864         }
2865
2866         return rc;
2867 }
2868
2869 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
2870                                 u64 remote_handle, u64 ioba, u64 num_entries)
2871 {
2872         unsigned int ua = adapter->vdev->unit_address;
2873         struct device *dev = &adapter->vdev->dev;
2874         int rc;
2875
2876         /* Make sure the hypervisor sees the complete request */
2877         mb();
2878         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
2879                                 cpu_to_be64(remote_handle),
2880                                 ioba, num_entries);
2881
2882         if (rc) {
2883                 if (rc == H_CLOSED)
2884                         dev_warn(dev, "CRQ Queue closed\n");
2885                 dev_err(dev, "Send (indirect) error (rc=%d)\n", rc);
2886         }
2887
2888         return rc;
2889 }
2890
2891 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
2892                             union ibmvnic_crq *crq)
2893 {
2894         unsigned int ua = adapter->vdev->unit_address;
2895         struct device *dev = &adapter->vdev->dev;
2896         u64 *u64_crq = (u64 *)crq;
2897         int rc;
2898
2899         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
2900                    (unsigned long int)cpu_to_be64(u64_crq[0]),
2901                    (unsigned long int)cpu_to_be64(u64_crq[1]));
2902
2903         /* Make sure the hypervisor sees the complete request */
2904         mb();
2905
2906         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
2907                                 cpu_to_be64(u64_crq[0]),
2908                                 cpu_to_be64(u64_crq[1]));
2909
2910         if (rc) {
2911                 if (rc == H_CLOSED)
2912                         dev_warn(dev, "CRQ Queue closed\n");
2913                 dev_warn(dev, "Send error (rc=%d)\n", rc);
2914         }
2915
2916         return rc;
2917 }
2918
2919 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
2920 {
2921         union ibmvnic_crq crq;
2922
2923         memset(&crq, 0, sizeof(crq));
2924         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
2925         crq.generic.cmd = IBMVNIC_CRQ_INIT;
2926         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
2927
2928         return ibmvnic_send_crq(adapter, &crq);
2929 }
2930
2931 static int send_version_xchg(struct ibmvnic_adapter *adapter)
2932 {
2933         union ibmvnic_crq crq;
2934
2935         memset(&crq, 0, sizeof(crq));
2936         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
2937         crq.version_exchange.cmd = VERSION_EXCHANGE;
2938         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
2939
2940         return ibmvnic_send_crq(adapter, &crq);
2941 }
2942
2943 struct vnic_login_client_data {
2944         u8      type;
2945         __be16  len;
2946         char    name;
2947 } __packed;
2948
2949 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
2950 {
2951         int len;
2952
2953         /* Calculate the amount of buffer space needed for the
2954          * vnic client data in the login buffer. There are four entries,
2955          * OS name, LPAR name, device name, and a null last entry.
2956          */
2957         len = 4 * sizeof(struct vnic_login_client_data);
2958         len += 6; /* "Linux" plus NULL */
2959         len += strlen(utsname()->nodename) + 1;
2960         len += strlen(adapter->netdev->name) + 1;
2961
2962         return len;
2963 }
2964
2965 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
2966                                  struct vnic_login_client_data *vlcd)
2967 {
2968         const char *os_name = "Linux";
2969         int len;
2970
2971         /* Type 1 - LPAR OS */
2972         vlcd->type = 1;
2973         len = strlen(os_name) + 1;
2974         vlcd->len = cpu_to_be16(len);
2975         strncpy(&vlcd->name, os_name, len);
2976         vlcd = (struct vnic_login_client_data *)((char *)&vlcd->name + len);
2977
2978         /* Type 2 - LPAR name */
2979         vlcd->type = 2;
2980         len = strlen(utsname()->nodename) + 1;
2981         vlcd->len = cpu_to_be16(len);
2982         strncpy(&vlcd->name, utsname()->nodename, len);
2983         vlcd = (struct vnic_login_client_data *)((char *)&vlcd->name + len);
2984
2985         /* Type 3 - device name */
2986         vlcd->type = 3;
2987         len = strlen(adapter->netdev->name) + 1;
2988         vlcd->len = cpu_to_be16(len);
2989         strncpy(&vlcd->name, adapter->netdev->name, len);
2990 }
2991
2992 static void send_login(struct ibmvnic_adapter *adapter)
2993 {
2994         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
2995         struct ibmvnic_login_buffer *login_buffer;
2996         struct device *dev = &adapter->vdev->dev;
2997         dma_addr_t rsp_buffer_token;
2998         dma_addr_t buffer_token;
2999         size_t rsp_buffer_size;
3000         union ibmvnic_crq crq;
3001         size_t buffer_size;
3002         __be64 *tx_list_p;
3003         __be64 *rx_list_p;
3004         int client_data_len;
3005         struct vnic_login_client_data *vlcd;
3006         int i;
3007
3008         client_data_len = vnic_client_data_len(adapter);
3009
3010         buffer_size =
3011             sizeof(struct ibmvnic_login_buffer) +
3012             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3013             client_data_len;
3014
3015         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3016         if (!login_buffer)
3017                 goto buf_alloc_failed;
3018
3019         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3020                                       DMA_TO_DEVICE);
3021         if (dma_mapping_error(dev, buffer_token)) {
3022                 dev_err(dev, "Couldn't map login buffer\n");
3023                 goto buf_map_failed;
3024         }
3025
3026         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3027                           sizeof(u64) * adapter->req_tx_queues +
3028                           sizeof(u64) * adapter->req_rx_queues +
3029                           sizeof(u64) * adapter->req_rx_queues +
3030                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3031
3032         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3033         if (!login_rsp_buffer)
3034                 goto buf_rsp_alloc_failed;
3035
3036         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3037                                           rsp_buffer_size, DMA_FROM_DEVICE);
3038         if (dma_mapping_error(dev, rsp_buffer_token)) {
3039                 dev_err(dev, "Couldn't map login rsp buffer\n");
3040                 goto buf_rsp_map_failed;
3041         }
3042
3043         adapter->login_buf = login_buffer;
3044         adapter->login_buf_token = buffer_token;
3045         adapter->login_buf_sz = buffer_size;
3046         adapter->login_rsp_buf = login_rsp_buffer;
3047         adapter->login_rsp_buf_token = rsp_buffer_token;
3048         adapter->login_rsp_buf_sz = rsp_buffer_size;
3049
3050         login_buffer->len = cpu_to_be32(buffer_size);
3051         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3052         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3053         login_buffer->off_txcomp_subcrqs =
3054             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3055         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3056         login_buffer->off_rxcomp_subcrqs =
3057             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3058                         sizeof(u64) * adapter->req_tx_queues);
3059         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3060         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3061
3062         tx_list_p = (__be64 *)((char *)login_buffer +
3063                                       sizeof(struct ibmvnic_login_buffer));
3064         rx_list_p = (__be64 *)((char *)login_buffer +
3065                                       sizeof(struct ibmvnic_login_buffer) +
3066                                       sizeof(u64) * adapter->req_tx_queues);
3067
3068         for (i = 0; i < adapter->req_tx_queues; i++) {
3069                 if (adapter->tx_scrq[i]) {
3070                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3071                                                    crq_num);
3072                 }
3073         }
3074
3075         for (i = 0; i < adapter->req_rx_queues; i++) {
3076                 if (adapter->rx_scrq[i]) {
3077                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3078                                                    crq_num);
3079                 }
3080         }
3081
3082         /* Insert vNIC login client data */
3083         vlcd = (struct vnic_login_client_data *)
3084                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3085         login_buffer->client_data_offset =
3086                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3087         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3088
3089         vnic_add_client_data(adapter, vlcd);
3090
3091         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3092         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3093                 netdev_dbg(adapter->netdev, "%016lx\n",
3094                            ((unsigned long int *)(adapter->login_buf))[i]);
3095         }
3096
3097         memset(&crq, 0, sizeof(crq));
3098         crq.login.first = IBMVNIC_CRQ_CMD;
3099         crq.login.cmd = LOGIN;
3100         crq.login.ioba = cpu_to_be32(buffer_token);
3101         crq.login.len = cpu_to_be32(buffer_size);
3102         ibmvnic_send_crq(adapter, &crq);
3103
3104         return;
3105
3106 buf_rsp_map_failed:
3107         kfree(login_rsp_buffer);
3108 buf_rsp_alloc_failed:
3109         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3110 buf_map_failed:
3111         kfree(login_buffer);
3112 buf_alloc_failed:
3113         return;
3114 }
3115
3116 static void send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3117                              u32 len, u8 map_id)
3118 {
3119         union ibmvnic_crq crq;
3120
3121         memset(&crq, 0, sizeof(crq));
3122         crq.request_map.first = IBMVNIC_CRQ_CMD;
3123         crq.request_map.cmd = REQUEST_MAP;
3124         crq.request_map.map_id = map_id;
3125         crq.request_map.ioba = cpu_to_be32(addr);
3126         crq.request_map.len = cpu_to_be32(len);
3127         ibmvnic_send_crq(adapter, &crq);
3128 }
3129
3130 static void send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3131 {
3132         union ibmvnic_crq crq;
3133
3134         memset(&crq, 0, sizeof(crq));
3135         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3136         crq.request_unmap.cmd = REQUEST_UNMAP;
3137         crq.request_unmap.map_id = map_id;
3138         ibmvnic_send_crq(adapter, &crq);
3139 }
3140
3141 static void send_map_query(struct ibmvnic_adapter *adapter)
3142 {
3143         union ibmvnic_crq crq;
3144
3145         memset(&crq, 0, sizeof(crq));
3146         crq.query_map.first = IBMVNIC_CRQ_CMD;
3147         crq.query_map.cmd = QUERY_MAP;
3148         ibmvnic_send_crq(adapter, &crq);
3149 }
3150
3151 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3152 static void send_cap_queries(struct ibmvnic_adapter *adapter)
3153 {
3154         union ibmvnic_crq crq;
3155
3156         atomic_set(&adapter->running_cap_crqs, 0);
3157         memset(&crq, 0, sizeof(crq));
3158         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3159         crq.query_capability.cmd = QUERY_CAPABILITY;
3160
3161         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3162         atomic_inc(&adapter->running_cap_crqs);
3163         ibmvnic_send_crq(adapter, &crq);
3164
3165         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3166         atomic_inc(&adapter->running_cap_crqs);
3167         ibmvnic_send_crq(adapter, &crq);
3168
3169         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3170         atomic_inc(&adapter->running_cap_crqs);
3171         ibmvnic_send_crq(adapter, &crq);
3172
3173         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3174         atomic_inc(&adapter->running_cap_crqs);
3175         ibmvnic_send_crq(adapter, &crq);
3176
3177         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3178         atomic_inc(&adapter->running_cap_crqs);
3179         ibmvnic_send_crq(adapter, &crq);
3180
3181         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3182         atomic_inc(&adapter->running_cap_crqs);
3183         ibmvnic_send_crq(adapter, &crq);
3184
3185         crq.query_capability.capability =
3186             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3187         atomic_inc(&adapter->running_cap_crqs);
3188         ibmvnic_send_crq(adapter, &crq);
3189
3190         crq.query_capability.capability =
3191             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3192         atomic_inc(&adapter->running_cap_crqs);
3193         ibmvnic_send_crq(adapter, &crq);
3194
3195         crq.query_capability.capability =
3196             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3197         atomic_inc(&adapter->running_cap_crqs);
3198         ibmvnic_send_crq(adapter, &crq);
3199
3200         crq.query_capability.capability =
3201             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
3202         atomic_inc(&adapter->running_cap_crqs);
3203         ibmvnic_send_crq(adapter, &crq);
3204
3205         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
3206         atomic_inc(&adapter->running_cap_crqs);
3207         ibmvnic_send_crq(adapter, &crq);
3208
3209         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
3210         atomic_inc(&adapter->running_cap_crqs);
3211         ibmvnic_send_crq(adapter, &crq);
3212
3213         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
3214         atomic_inc(&adapter->running_cap_crqs);
3215         ibmvnic_send_crq(adapter, &crq);
3216
3217         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
3218         atomic_inc(&adapter->running_cap_crqs);
3219         ibmvnic_send_crq(adapter, &crq);
3220
3221         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
3222         atomic_inc(&adapter->running_cap_crqs);
3223         ibmvnic_send_crq(adapter, &crq);
3224
3225         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
3226         atomic_inc(&adapter->running_cap_crqs);
3227         ibmvnic_send_crq(adapter, &crq);
3228
3229         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
3230         atomic_inc(&adapter->running_cap_crqs);
3231         ibmvnic_send_crq(adapter, &crq);
3232
3233         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
3234         atomic_inc(&adapter->running_cap_crqs);
3235         ibmvnic_send_crq(adapter, &crq);
3236
3237         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
3238         atomic_inc(&adapter->running_cap_crqs);
3239         ibmvnic_send_crq(adapter, &crq);
3240
3241         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
3242         atomic_inc(&adapter->running_cap_crqs);
3243         ibmvnic_send_crq(adapter, &crq);
3244
3245         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
3246         atomic_inc(&adapter->running_cap_crqs);
3247         ibmvnic_send_crq(adapter, &crq);
3248
3249         crq.query_capability.capability =
3250                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
3251         atomic_inc(&adapter->running_cap_crqs);
3252         ibmvnic_send_crq(adapter, &crq);
3253
3254         crq.query_capability.capability =
3255                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
3256         atomic_inc(&adapter->running_cap_crqs);
3257         ibmvnic_send_crq(adapter, &crq);
3258
3259         crq.query_capability.capability =
3260                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
3261         atomic_inc(&adapter->running_cap_crqs);
3262         ibmvnic_send_crq(adapter, &crq);
3263
3264         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
3265         atomic_inc(&adapter->running_cap_crqs);
3266         ibmvnic_send_crq(adapter, &crq);
3267 }
3268
3269 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
3270                                 struct ibmvnic_adapter *adapter)
3271 {
3272         struct device *dev = &adapter->vdev->dev;
3273
3274         if (crq->get_vpd_size_rsp.rc.code) {
3275                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
3276                         crq->get_vpd_size_rsp.rc.code);
3277                 complete(&adapter->fw_done);
3278                 return;
3279         }
3280
3281         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
3282         complete(&adapter->fw_done);
3283 }
3284
3285 static void handle_vpd_rsp(union ibmvnic_crq *crq,
3286                            struct ibmvnic_adapter *adapter)
3287 {
3288         struct device *dev = &adapter->vdev->dev;
3289         unsigned char *substr = NULL, *ptr = NULL;
3290         u8 fw_level_len = 0;
3291
3292         memset(adapter->fw_version, 0, 32);
3293
3294         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
3295                          DMA_FROM_DEVICE);
3296
3297         if (crq->get_vpd_rsp.rc.code) {
3298                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
3299                         crq->get_vpd_rsp.rc.code);
3300                 goto complete;
3301         }
3302
3303         /* get the position of the firmware version info
3304          * located after the ASCII 'RM' substring in the buffer
3305          */
3306         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
3307         if (!substr) {
3308                 dev_info(dev, "No FW level provided by VPD\n");
3309                 goto complete;
3310         }
3311
3312         /* get length of firmware level ASCII substring */
3313         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
3314                 fw_level_len = *(substr + 2);
3315         } else {
3316                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
3317                 goto complete;
3318         }
3319
3320         /* copy firmware version string from vpd into adapter */
3321         if ((substr + 3 + fw_level_len) <
3322             (adapter->vpd->buff + adapter->vpd->len)) {
3323                 ptr = strncpy((char *)adapter->fw_version,
3324                               substr + 3, fw_level_len);
3325
3326                 if (!ptr)
3327                         dev_err(dev, "Failed to isolate FW level string\n");
3328         } else {
3329                 dev_info(dev, "FW substr extrapolated VPD buff\n");
3330         }
3331
3332 complete:
3333         complete(&adapter->fw_done);
3334 }
3335
3336 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
3337 {
3338         struct device *dev = &adapter->vdev->dev;
3339         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
3340         union ibmvnic_crq crq;
3341         int i;
3342
3343         dma_unmap_single(dev, adapter->ip_offload_tok,
3344                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
3345
3346         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
3347         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
3348                 netdev_dbg(adapter->netdev, "%016lx\n",
3349                            ((unsigned long int *)(buf))[i]);
3350
3351         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
3352         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
3353         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
3354                    buf->tcp_ipv4_chksum);
3355         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
3356                    buf->tcp_ipv6_chksum);
3357         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
3358                    buf->udp_ipv4_chksum);
3359         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
3360                    buf->udp_ipv6_chksum);
3361         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
3362                    buf->large_tx_ipv4);
3363         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
3364                    buf->large_tx_ipv6);
3365         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
3366                    buf->large_rx_ipv4);
3367         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
3368                    buf->large_rx_ipv6);
3369         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
3370                    buf->max_ipv4_header_size);
3371         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
3372                    buf->max_ipv6_header_size);
3373         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
3374                    buf->max_tcp_header_size);
3375         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
3376                    buf->max_udp_header_size);
3377         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
3378                    buf->max_large_tx_size);
3379         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
3380                    buf->max_large_rx_size);
3381         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
3382                    buf->ipv6_extension_header);
3383         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
3384                    buf->tcp_pseudosum_req);
3385         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
3386                    buf->num_ipv6_ext_headers);
3387         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
3388                    buf->off_ipv6_ext_headers);
3389
3390         adapter->ip_offload_ctrl_tok =
3391             dma_map_single(dev, &adapter->ip_offload_ctrl,
3392                            sizeof(adapter->ip_offload_ctrl), DMA_TO_DEVICE);
3393
3394         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
3395                 dev_err(dev, "Couldn't map ip offload control buffer\n");
3396                 return;
3397         }
3398
3399         adapter->ip_offload_ctrl.len =
3400             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3401         adapter->ip_offload_ctrl.version = cpu_to_be32(INITIAL_VERSION_IOB);
3402         adapter->ip_offload_ctrl.ipv4_chksum = buf->ipv4_chksum;
3403         adapter->ip_offload_ctrl.ipv6_chksum = buf->ipv6_chksum;
3404         adapter->ip_offload_ctrl.tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
3405         adapter->ip_offload_ctrl.udp_ipv4_chksum = buf->udp_ipv4_chksum;
3406         adapter->ip_offload_ctrl.tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
3407         adapter->ip_offload_ctrl.udp_ipv6_chksum = buf->udp_ipv6_chksum;
3408         adapter->ip_offload_ctrl.large_tx_ipv4 = buf->large_tx_ipv4;
3409         adapter->ip_offload_ctrl.large_tx_ipv6 = buf->large_tx_ipv6;
3410
3411         /* large_rx disabled for now, additional features needed */
3412         adapter->ip_offload_ctrl.large_rx_ipv4 = 0;
3413         adapter->ip_offload_ctrl.large_rx_ipv6 = 0;
3414
3415         adapter->netdev->features = NETIF_F_SG | NETIF_F_GSO;
3416
3417         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
3418                 adapter->netdev->features |= NETIF_F_IP_CSUM;
3419
3420         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
3421                 adapter->netdev->features |= NETIF_F_IPV6_CSUM;
3422
3423         if ((adapter->netdev->features &
3424             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
3425                 adapter->netdev->features |= NETIF_F_RXCSUM;
3426
3427         if (buf->large_tx_ipv4)
3428                 adapter->netdev->features |= NETIF_F_TSO;
3429         if (buf->large_tx_ipv6)
3430                 adapter->netdev->features |= NETIF_F_TSO6;
3431
3432         adapter->netdev->hw_features |= adapter->netdev->features;
3433
3434         memset(&crq, 0, sizeof(crq));
3435         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
3436         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
3437         crq.control_ip_offload.len =
3438             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3439         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
3440         ibmvnic_send_crq(adapter, &crq);
3441 }
3442
3443 static void handle_error_info_rsp(union ibmvnic_crq *crq,
3444                                   struct ibmvnic_adapter *adapter)
3445 {
3446         struct device *dev = &adapter->vdev->dev;
3447         struct ibmvnic_error_buff *error_buff, *tmp;
3448         unsigned long flags;
3449         bool found = false;
3450         int i;
3451
3452         if (!crq->request_error_rsp.rc.code) {
3453                 dev_info(dev, "Request Error Rsp returned with rc=%x\n",
3454                          crq->request_error_rsp.rc.code);
3455                 return;
3456         }
3457
3458         spin_lock_irqsave(&adapter->error_list_lock, flags);
3459         list_for_each_entry_safe(error_buff, tmp, &adapter->errors, list)
3460                 if (error_buff->error_id == crq->request_error_rsp.error_id) {
3461                         found = true;
3462                         list_del(&error_buff->list);
3463                         break;
3464                 }
3465         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
3466
3467         if (!found) {
3468                 dev_err(dev, "Couldn't find error id %x\n",
3469                         be32_to_cpu(crq->request_error_rsp.error_id));
3470                 return;
3471         }
3472
3473         dev_err(dev, "Detailed info for error id %x:",
3474                 be32_to_cpu(crq->request_error_rsp.error_id));
3475
3476         for (i = 0; i < error_buff->len; i++) {
3477                 pr_cont("%02x", (int)error_buff->buff[i]);
3478                 if (i % 8 == 7)
3479                         pr_cont(" ");
3480         }
3481         pr_cont("\n");
3482
3483         dma_unmap_single(dev, error_buff->dma, error_buff->len,
3484                          DMA_FROM_DEVICE);
3485         kfree(error_buff->buff);
3486         kfree(error_buff);
3487 }
3488
3489 static void request_error_information(struct ibmvnic_adapter *adapter,
3490                                       union ibmvnic_crq *err_crq)
3491 {
3492         struct device *dev = &adapter->vdev->dev;
3493         struct net_device *netdev = adapter->netdev;
3494         struct ibmvnic_error_buff *error_buff;
3495         unsigned long timeout = msecs_to_jiffies(30000);
3496         union ibmvnic_crq crq;
3497         unsigned long flags;
3498         int rc, detail_len;
3499
3500         error_buff = kmalloc(sizeof(*error_buff), GFP_ATOMIC);
3501         if (!error_buff)
3502                 return;
3503
3504         detail_len = be32_to_cpu(err_crq->error_indication.detail_error_sz);
3505         error_buff->buff = kmalloc(detail_len, GFP_ATOMIC);
3506         if (!error_buff->buff) {
3507                 kfree(error_buff);
3508                 return;
3509         }
3510
3511         error_buff->dma = dma_map_single(dev, error_buff->buff, detail_len,
3512                                          DMA_FROM_DEVICE);
3513         if (dma_mapping_error(dev, error_buff->dma)) {
3514                 netdev_err(netdev, "Couldn't map error buffer\n");
3515                 kfree(error_buff->buff);
3516                 kfree(error_buff);
3517                 return;
3518         }
3519
3520         error_buff->len = detail_len;
3521         error_buff->error_id = err_crq->error_indication.error_id;
3522
3523         spin_lock_irqsave(&adapter->error_list_lock, flags);
3524         list_add_tail(&error_buff->list, &adapter->errors);
3525         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
3526
3527         memset(&crq, 0, sizeof(crq));
3528         crq.request_error_info.first = IBMVNIC_CRQ_CMD;
3529         crq.request_error_info.cmd = REQUEST_ERROR_INFO;
3530         crq.request_error_info.ioba = cpu_to_be32(error_buff->dma);
3531         crq.request_error_info.len = cpu_to_be32(detail_len);
3532         crq.request_error_info.error_id = err_crq->error_indication.error_id;
3533
3534         rc = ibmvnic_send_crq(adapter, &crq);
3535         if (rc) {
3536                 netdev_err(netdev, "failed to request error information\n");
3537                 goto err_info_fail;
3538         }
3539
3540         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3541                 netdev_err(netdev, "timeout waiting for error information\n");
3542                 goto err_info_fail;
3543         }
3544
3545         return;
3546
3547 err_info_fail:
3548         spin_lock_irqsave(&adapter->error_list_lock, flags);
3549         list_del(&error_buff->list);
3550         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
3551
3552         kfree(error_buff->buff);
3553         kfree(error_buff);
3554 }
3555
3556 static void handle_error_indication(union ibmvnic_crq *crq,
3557                                     struct ibmvnic_adapter *adapter)
3558 {
3559         struct device *dev = &adapter->vdev->dev;
3560
3561         dev_err(dev, "Firmware reports %serror id %x, cause %d\n",
3562                 crq->error_indication.flags
3563                         & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
3564                 be32_to_cpu(crq->error_indication.error_id),
3565                 be16_to_cpu(crq->error_indication.error_cause));
3566
3567         if (be32_to_cpu(crq->error_indication.error_id))
3568                 request_error_information(adapter, crq);
3569
3570         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
3571                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3572         else
3573                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
3574 }
3575
3576 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
3577                                  struct ibmvnic_adapter *adapter)
3578 {
3579         struct net_device *netdev = adapter->netdev;
3580         struct device *dev = &adapter->vdev->dev;
3581         long rc;
3582
3583         rc = crq->change_mac_addr_rsp.rc.code;
3584         if (rc) {
3585                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
3586                 goto out;
3587         }
3588         memcpy(netdev->dev_addr, &crq->change_mac_addr_rsp.mac_addr[0],
3589                ETH_ALEN);
3590 out:
3591         complete(&adapter->fw_done);
3592         return rc;
3593 }
3594
3595 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
3596                                    struct ibmvnic_adapter *adapter)
3597 {
3598         struct device *dev = &adapter->vdev->dev;
3599         u64 *req_value;
3600         char *name;
3601
3602         atomic_dec(&adapter->running_cap_crqs);
3603         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
3604         case REQ_TX_QUEUES:
3605                 req_value = &adapter->req_tx_queues;
3606                 name = "tx";
3607                 break;
3608         case REQ_RX_QUEUES:
3609                 req_value = &adapter->req_rx_queues;
3610                 name = "rx";
3611                 break;
3612         case REQ_RX_ADD_QUEUES:
3613                 req_value = &adapter->req_rx_add_queues;
3614                 name = "rx_add";
3615                 break;
3616         case REQ_TX_ENTRIES_PER_SUBCRQ:
3617                 req_value = &adapter->req_tx_entries_per_subcrq;
3618                 name = "tx_entries_per_subcrq";
3619                 break;
3620         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
3621                 req_value = &adapter->req_rx_add_entries_per_subcrq;
3622                 name = "rx_add_entries_per_subcrq";
3623                 break;
3624         case REQ_MTU:
3625                 req_value = &adapter->req_mtu;
3626                 name = "mtu";
3627                 break;
3628         case PROMISC_REQUESTED:
3629                 req_value = &adapter->promisc;
3630                 name = "promisc";
3631                 break;
3632         default:
3633                 dev_err(dev, "Got invalid cap request rsp %d\n",
3634                         crq->request_capability.capability);
3635                 return;
3636         }
3637
3638         switch (crq->request_capability_rsp.rc.code) {
3639         case SUCCESS:
3640                 break;
3641         case PARTIALSUCCESS:
3642                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
3643                          *req_value,
3644                          (long int)be64_to_cpu(crq->request_capability_rsp.
3645                                                number), name);
3646
3647                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
3648                     REQ_MTU) {
3649                         pr_err("mtu of %llu is not supported. Reverting.\n",
3650                                *req_value);
3651                         *req_value = adapter->fallback.mtu;
3652                 } else {
3653                         *req_value =
3654                                 be64_to_cpu(crq->request_capability_rsp.number);
3655                 }
3656
3657                 ibmvnic_send_req_caps(adapter, 1);
3658                 return;
3659         default:
3660                 dev_err(dev, "Error %d in request cap rsp\n",
3661                         crq->request_capability_rsp.rc.code);
3662                 return;
3663         }
3664
3665         /* Done receiving requested capabilities, query IP offload support */
3666         if (atomic_read(&adapter->running_cap_crqs) == 0) {
3667                 union ibmvnic_crq newcrq;
3668                 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
3669                 struct ibmvnic_query_ip_offload_buffer *ip_offload_buf =
3670                     &adapter->ip_offload_buf;
3671
3672                 adapter->wait_capability = false;
3673                 adapter->ip_offload_tok = dma_map_single(dev, ip_offload_buf,
3674                                                          buf_sz,
3675                                                          DMA_FROM_DEVICE);
3676
3677                 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
3678                         if (!firmware_has_feature(FW_FEATURE_CMO))
3679                                 dev_err(dev, "Couldn't map offload buffer\n");
3680                         return;
3681                 }
3682
3683                 memset(&newcrq, 0, sizeof(newcrq));
3684                 newcrq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
3685                 newcrq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
3686                 newcrq.query_ip_offload.len = cpu_to_be32(buf_sz);
3687                 newcrq.query_ip_offload.ioba =
3688                     cpu_to_be32(adapter->ip_offload_tok);
3689
3690                 ibmvnic_send_crq(adapter, &newcrq);
3691         }
3692 }
3693
3694 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
3695                             struct ibmvnic_adapter *adapter)
3696 {
3697         struct device *dev = &adapter->vdev->dev;
3698         struct net_device *netdev = adapter->netdev;
3699         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
3700         struct ibmvnic_login_buffer *login = adapter->login_buf;
3701         int i;
3702
3703         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
3704                          DMA_BIDIRECTIONAL);
3705         dma_unmap_single(dev, adapter->login_rsp_buf_token,
3706                          adapter->login_rsp_buf_sz, DMA_BIDIRECTIONAL);
3707
3708         /* If the number of queues requested can't be allocated by the
3709          * server, the login response will return with code 1. We will need
3710          * to resend the login buffer with fewer queues requested.
3711          */
3712         if (login_rsp_crq->generic.rc.code) {
3713                 adapter->renegotiate = true;
3714                 complete(&adapter->init_done);
3715                 return 0;
3716         }
3717
3718         netdev->mtu = adapter->req_mtu - ETH_HLEN;
3719
3720         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
3721         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
3722                 netdev_dbg(adapter->netdev, "%016lx\n",
3723                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
3724         }
3725
3726         /* Sanity checks */
3727         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
3728             (be32_to_cpu(login->num_rxcomp_subcrqs) *
3729              adapter->req_rx_add_queues !=
3730              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
3731                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
3732                 ibmvnic_remove(adapter->vdev);
3733                 return -EIO;
3734         }
3735         complete(&adapter->init_done);
3736
3737         return 0;
3738 }
3739
3740 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
3741                                      struct ibmvnic_adapter *adapter)
3742 {
3743         struct device *dev = &adapter->vdev->dev;
3744         long rc;
3745
3746         rc = crq->request_unmap_rsp.rc.code;
3747         if (rc)
3748                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
3749 }
3750
3751 static void handle_query_map_rsp(union ibmvnic_crq *crq,
3752                                  struct ibmvnic_adapter *adapter)
3753 {
3754         struct net_device *netdev = adapter->netdev;
3755         struct device *dev = &adapter->vdev->dev;
3756         long rc;
3757
3758         rc = crq->query_map_rsp.rc.code;
3759         if (rc) {
3760                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
3761                 return;
3762         }
3763         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
3764                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
3765                    crq->query_map_rsp.free_pages);
3766 }
3767
3768 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
3769                                  struct ibmvnic_adapter *adapter)
3770 {
3771         struct net_device *netdev = adapter->netdev;
3772         struct device *dev = &adapter->vdev->dev;
3773         long rc;
3774
3775         atomic_dec(&adapter->running_cap_crqs);
3776         netdev_dbg(netdev, "Outstanding queries: %d\n",
3777                    atomic_read(&adapter->running_cap_crqs));
3778         rc = crq->query_capability.rc.code;
3779         if (rc) {
3780                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
3781                 goto out;
3782         }
3783
3784         switch (be16_to_cpu(crq->query_capability.capability)) {
3785         case MIN_TX_QUEUES:
3786                 adapter->min_tx_queues =
3787                     be64_to_cpu(crq->query_capability.number);
3788                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
3789                            adapter->min_tx_queues);
3790                 break;
3791         case MIN_RX_QUEUES:
3792                 adapter->min_rx_queues =
3793                     be64_to_cpu(crq->query_capability.number);
3794                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
3795                            adapter->min_rx_queues);
3796                 break;
3797         case MIN_RX_ADD_QUEUES:
3798                 adapter->min_rx_add_queues =
3799                     be64_to_cpu(crq->query_capability.number);
3800                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
3801                            adapter->min_rx_add_queues);
3802                 break;
3803         case MAX_TX_QUEUES:
3804                 adapter->max_tx_queues =
3805                     be64_to_cpu(crq->query_capability.number);
3806                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
3807                            adapter->max_tx_queues);
3808                 break;
3809         case MAX_RX_QUEUES:
3810                 adapter->max_rx_queues =
3811                     be64_to_cpu(crq->query_capability.number);
3812                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
3813                            adapter->max_rx_queues);
3814                 break;
3815         case MAX_RX_ADD_QUEUES:
3816                 adapter->max_rx_add_queues =
3817                     be64_to_cpu(crq->query_capability.number);
3818                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
3819                            adapter->max_rx_add_queues);
3820                 break;
3821         case MIN_TX_ENTRIES_PER_SUBCRQ:
3822                 adapter->min_tx_entries_per_subcrq =
3823                     be64_to_cpu(crq->query_capability.number);
3824                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
3825                            adapter->min_tx_entries_per_subcrq);
3826                 break;
3827         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
3828                 adapter->min_rx_add_entries_per_subcrq =
3829                     be64_to_cpu(crq->query_capability.number);
3830                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
3831                            adapter->min_rx_add_entries_per_subcrq);
3832                 break;
3833         case MAX_TX_ENTRIES_PER_SUBCRQ:
3834                 adapter->max_tx_entries_per_subcrq =
3835                     be64_to_cpu(crq->query_capability.number);
3836                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
3837                            adapter->max_tx_entries_per_subcrq);
3838                 break;
3839         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
3840                 adapter->max_rx_add_entries_per_subcrq =
3841                     be64_to_cpu(crq->query_capability.number);
3842                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
3843                            adapter->max_rx_add_entries_per_subcrq);
3844                 break;
3845         case TCP_IP_OFFLOAD:
3846                 adapter->tcp_ip_offload =
3847                     be64_to_cpu(crq->query_capability.number);
3848                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
3849                            adapter->tcp_ip_offload);
3850                 break;
3851         case PROMISC_SUPPORTED:
3852                 adapter->promisc_supported =
3853                     be64_to_cpu(crq->query_capability.number);
3854                 netdev_dbg(netdev, "promisc_supported = %lld\n",
3855                            adapter->promisc_supported);
3856                 break;
3857         case MIN_MTU:
3858                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
3859                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3860                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
3861                 break;
3862         case MAX_MTU:
3863                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
3864                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3865                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
3866                 break;
3867         case MAX_MULTICAST_FILTERS:
3868                 adapter->max_multicast_filters =
3869                     be64_to_cpu(crq->query_capability.number);
3870                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
3871                            adapter->max_multicast_filters);
3872                 break;
3873         case VLAN_HEADER_INSERTION:
3874                 adapter->vlan_header_insertion =
3875                     be64_to_cpu(crq->query_capability.number);
3876                 if (adapter->vlan_header_insertion)
3877                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
3878                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
3879                            adapter->vlan_header_insertion);
3880                 break;
3881         case RX_VLAN_HEADER_INSERTION:
3882                 adapter->rx_vlan_header_insertion =
3883                     be64_to_cpu(crq->query_capability.number);
3884                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
3885                            adapter->rx_vlan_header_insertion);
3886                 break;
3887         case MAX_TX_SG_ENTRIES:
3888                 adapter->max_tx_sg_entries =
3889                     be64_to_cpu(crq->query_capability.number);
3890                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
3891                            adapter->max_tx_sg_entries);
3892                 break;
3893         case RX_SG_SUPPORTED:
3894                 adapter->rx_sg_supported =
3895                     be64_to_cpu(crq->query_capability.number);
3896                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
3897                            adapter->rx_sg_supported);
3898                 break;
3899         case OPT_TX_COMP_SUB_QUEUES:
3900                 adapter->opt_tx_comp_sub_queues =
3901                     be64_to_cpu(crq->query_capability.number);
3902                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
3903                            adapter->opt_tx_comp_sub_queues);
3904                 break;
3905         case OPT_RX_COMP_QUEUES:
3906                 adapter->opt_rx_comp_queues =
3907                     be64_to_cpu(crq->query_capability.number);
3908                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
3909                            adapter->opt_rx_comp_queues);
3910                 break;
3911         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
3912                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
3913                     be64_to_cpu(crq->query_capability.number);
3914                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
3915                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
3916                 break;
3917         case OPT_TX_ENTRIES_PER_SUBCRQ:
3918                 adapter->opt_tx_entries_per_subcrq =
3919                     be64_to_cpu(crq->query_capability.number);
3920                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
3921                            adapter->opt_tx_entries_per_subcrq);
3922                 break;
3923         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
3924                 adapter->opt_rxba_entries_per_subcrq =
3925                     be64_to_cpu(crq->query_capability.number);
3926                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
3927                            adapter->opt_rxba_entries_per_subcrq);
3928                 break;
3929         case TX_RX_DESC_REQ:
3930                 adapter->tx_rx_desc_req = crq->query_capability.number;
3931                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
3932                            adapter->tx_rx_desc_req);
3933                 break;
3934
3935         default:
3936                 netdev_err(netdev, "Got invalid cap rsp %d\n",
3937                            crq->query_capability.capability);
3938         }
3939
3940 out:
3941         if (atomic_read(&adapter->running_cap_crqs) == 0) {
3942                 adapter->wait_capability = false;
3943                 ibmvnic_send_req_caps(adapter, 0);
3944         }
3945 }
3946
3947 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
3948                                struct ibmvnic_adapter *adapter)
3949 {
3950         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
3951         struct net_device *netdev = adapter->netdev;
3952         struct device *dev = &adapter->vdev->dev;
3953         u64 *u64_crq = (u64 *)crq;
3954         long rc;
3955
3956         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
3957                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3958                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3959         switch (gen_crq->first) {
3960         case IBMVNIC_CRQ_INIT_RSP:
3961                 switch (gen_crq->cmd) {
3962                 case IBMVNIC_CRQ_INIT:
3963                         dev_info(dev, "Partner initialized\n");
3964                         adapter->from_passive_init = true;
3965                         complete(&adapter->init_done);
3966                         break;
3967                 case IBMVNIC_CRQ_INIT_COMPLETE:
3968                         dev_info(dev, "Partner initialization complete\n");
3969                         send_version_xchg(adapter);
3970                         break;
3971                 default:
3972                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
3973                 }
3974                 return;
3975         case IBMVNIC_CRQ_XPORT_EVENT:
3976                 netif_carrier_off(netdev);
3977                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
3978                         dev_info(dev, "Migrated, re-enabling adapter\n");
3979                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
3980                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
3981                         dev_info(dev, "Backing device failover detected\n");
3982                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
3983                 } else {
3984                         /* The adapter lost the connection */
3985                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
3986                                 gen_crq->cmd);
3987                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3988                 }
3989                 return;
3990         case IBMVNIC_CRQ_CMD_RSP:
3991                 break;
3992         default:
3993                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
3994                         gen_crq->first);
3995                 return;
3996         }
3997
3998         switch (gen_crq->cmd) {
3999         case VERSION_EXCHANGE_RSP:
4000                 rc = crq->version_exchange_rsp.rc.code;
4001                 if (rc) {
4002                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4003                         break;
4004                 }
4005                 dev_info(dev, "Partner protocol version is %d\n",
4006                          crq->version_exchange_rsp.version);
4007                 if (be16_to_cpu(crq->version_exchange_rsp.version) <
4008                     ibmvnic_version)
4009                         ibmvnic_version =
4010                             be16_to_cpu(crq->version_exchange_rsp.version);
4011                 send_cap_queries(adapter);
4012                 break;
4013         case QUERY_CAPABILITY_RSP:
4014                 handle_query_cap_rsp(crq, adapter);
4015                 break;
4016         case QUERY_MAP_RSP:
4017                 handle_query_map_rsp(crq, adapter);
4018                 break;
4019         case REQUEST_MAP_RSP:
4020                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4021                 complete(&adapter->fw_done);
4022                 break;
4023         case REQUEST_UNMAP_RSP:
4024                 handle_request_unmap_rsp(crq, adapter);
4025                 break;
4026         case REQUEST_CAPABILITY_RSP:
4027                 handle_request_cap_rsp(crq, adapter);
4028                 break;
4029         case LOGIN_RSP:
4030                 netdev_dbg(netdev, "Got Login Response\n");
4031                 handle_login_rsp(crq, adapter);
4032                 break;
4033         case LOGICAL_LINK_STATE_RSP:
4034                 netdev_dbg(netdev,
4035                            "Got Logical Link State Response, state: %d rc: %d\n",
4036                            crq->logical_link_state_rsp.link_state,
4037                            crq->logical_link_state_rsp.rc.code);
4038                 adapter->logical_link_state =
4039                     crq->logical_link_state_rsp.link_state;
4040                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4041                 complete(&adapter->init_done);
4042                 break;
4043         case LINK_STATE_INDICATION:
4044                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4045                 adapter->phys_link_state =
4046                     crq->link_state_indication.phys_link_state;
4047                 adapter->logical_link_state =
4048                     crq->link_state_indication.logical_link_state;
4049                 break;
4050         case CHANGE_MAC_ADDR_RSP:
4051                 netdev_dbg(netdev, "Got MAC address change Response\n");
4052                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4053                 break;
4054         case ERROR_INDICATION:
4055                 netdev_dbg(netdev, "Got Error Indication\n");
4056                 handle_error_indication(crq, adapter);
4057                 break;
4058         case REQUEST_ERROR_RSP:
4059                 netdev_dbg(netdev, "Got Error Detail Response\n");
4060                 handle_error_info_rsp(crq, adapter);
4061                 break;
4062         case REQUEST_STATISTICS_RSP:
4063                 netdev_dbg(netdev, "Got Statistics Response\n");
4064                 complete(&adapter->stats_done);
4065                 break;
4066         case QUERY_IP_OFFLOAD_RSP:
4067                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4068                 handle_query_ip_offload_rsp(adapter);
4069                 break;
4070         case MULTICAST_CTRL_RSP:
4071                 netdev_dbg(netdev, "Got multicast control Response\n");
4072                 break;
4073         case CONTROL_IP_OFFLOAD_RSP:
4074                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4075                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4076                                  sizeof(adapter->ip_offload_ctrl),
4077                                  DMA_TO_DEVICE);
4078                 complete(&adapter->init_done);
4079                 break;
4080         case COLLECT_FW_TRACE_RSP:
4081                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4082                 complete(&adapter->fw_done);
4083                 break;
4084         case GET_VPD_SIZE_RSP:
4085                 handle_vpd_size_rsp(crq, adapter);
4086                 break;
4087         case GET_VPD_RSP:
4088                 handle_vpd_rsp(crq, adapter);
4089                 break;
4090         default:
4091                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4092                            gen_crq->cmd);
4093         }
4094 }
4095
4096 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
4097 {
4098         struct ibmvnic_adapter *adapter = instance;
4099
4100         tasklet_schedule(&adapter->tasklet);
4101         return IRQ_HANDLED;
4102 }
4103
4104 static void ibmvnic_tasklet(void *data)
4105 {
4106         struct ibmvnic_adapter *adapter = data;
4107         struct ibmvnic_crq_queue *queue = &adapter->crq;
4108         union ibmvnic_crq *crq;
4109         unsigned long flags;
4110         bool done = false;
4111
4112         spin_lock_irqsave(&queue->lock, flags);
4113         while (!done) {
4114                 /* Pull all the valid messages off the CRQ */
4115                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
4116                         ibmvnic_handle_crq(crq, adapter);
4117                         crq->generic.first = 0;
4118                 }
4119
4120                 /* remain in tasklet until all
4121                  * capabilities responses are received
4122                  */
4123                 if (!adapter->wait_capability)
4124                         done = true;
4125         }
4126         /* if capabilities CRQ's were sent in this tasklet, the following
4127          * tasklet must wait until all responses are received
4128          */
4129         if (atomic_read(&adapter->running_cap_crqs) != 0)
4130                 adapter->wait_capability = true;
4131         spin_unlock_irqrestore(&queue->lock, flags);
4132 }
4133
4134 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
4135 {
4136         struct vio_dev *vdev = adapter->vdev;
4137         int rc;
4138
4139         do {
4140                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
4141         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
4142
4143         if (rc)
4144                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
4145
4146         return rc;
4147 }
4148
4149 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
4150 {
4151         struct ibmvnic_crq_queue *crq = &adapter->crq;
4152         struct device *dev = &adapter->vdev->dev;
4153         struct vio_dev *vdev = adapter->vdev;
4154         int rc;
4155
4156         /* Close the CRQ */
4157         do {
4158                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4159         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4160
4161         /* Clean out the queue */
4162         memset(crq->msgs, 0, PAGE_SIZE);
4163         crq->cur = 0;
4164
4165         /* And re-open it again */
4166         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4167                                 crq->msg_token, PAGE_SIZE);
4168
4169         if (rc == H_CLOSED)
4170                 /* Adapter is good, but other end is not ready */
4171                 dev_warn(dev, "Partner adapter not ready\n");
4172         else if (rc != 0)
4173                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
4174
4175         return rc;
4176 }
4177
4178 static void release_crq_queue(struct ibmvnic_adapter *adapter)
4179 {
4180         struct ibmvnic_crq_queue *crq = &adapter->crq;
4181         struct vio_dev *vdev = adapter->vdev;
4182         long rc;
4183
4184         if (!crq->msgs)
4185                 return;
4186
4187         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
4188         free_irq(vdev->irq, adapter);
4189         tasklet_kill(&adapter->tasklet);
4190         do {
4191                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4192         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4193
4194         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
4195                          DMA_BIDIRECTIONAL);
4196         free_page((unsigned long)crq->msgs);
4197         crq->msgs = NULL;
4198 }
4199
4200 static int init_crq_queue(struct ibmvnic_adapter *adapter)
4201 {
4202         struct ibmvnic_crq_queue *crq = &adapter->crq;
4203         struct device *dev = &adapter->vdev->dev;
4204         struct vio_dev *vdev = adapter->vdev;
4205         int rc, retrc = -ENOMEM;
4206
4207         if (crq->msgs)
4208                 return 0;
4209
4210         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
4211         /* Should we allocate more than one page? */
4212
4213         if (!crq->msgs)
4214                 return -ENOMEM;
4215
4216         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
4217         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
4218                                         DMA_BIDIRECTIONAL);
4219         if (dma_mapping_error(dev, crq->msg_token))
4220                 goto map_failed;
4221
4222         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4223                                 crq->msg_token, PAGE_SIZE);
4224
4225         if (rc == H_RESOURCE)
4226                 /* maybe kexecing and resource is busy. try a reset */
4227                 rc = ibmvnic_reset_crq(adapter);
4228         retrc = rc;
4229
4230         if (rc == H_CLOSED) {
4231                 dev_warn(dev, "Partner adapter not ready\n");
4232         } else if (rc) {
4233                 dev_warn(dev, "Error %d opening adapter\n", rc);
4234                 goto reg_crq_failed;
4235         }
4236
4237         retrc = 0;
4238
4239         tasklet_init(&adapter->tasklet, (void *)ibmvnic_tasklet,
4240                      (unsigned long)adapter);
4241
4242         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
4243         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, IBMVNIC_NAME,
4244                          adapter);
4245         if (rc) {
4246                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
4247                         vdev->irq, rc);
4248                 goto req_irq_failed;
4249         }
4250
4251         rc = vio_enable_interrupts(vdev);
4252         if (rc) {
4253                 dev_err(dev, "Error %d enabling interrupts\n", rc);
4254                 goto req_irq_failed;
4255         }
4256
4257         crq->cur = 0;
4258         spin_lock_init(&crq->lock);
4259
4260         return retrc;
4261
4262 req_irq_failed:
4263         tasklet_kill(&adapter->tasklet);
4264         do {
4265                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4266         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4267 reg_crq_failed:
4268         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
4269 map_failed:
4270         free_page((unsigned long)crq->msgs);
4271         crq->msgs = NULL;
4272         return retrc;
4273 }
4274
4275 static int ibmvnic_init(struct ibmvnic_adapter *adapter)
4276 {
4277         struct device *dev = &adapter->vdev->dev;
4278         unsigned long timeout = msecs_to_jiffies(30000);
4279         int rc;
4280
4281         if (adapter->resetting && !adapter->wait_for_reset) {
4282                 rc = ibmvnic_reset_crq(adapter);
4283                 if (!rc)
4284                         rc = vio_enable_interrupts(adapter->vdev);
4285         } else {
4286                 rc = init_crq_queue(adapter);
4287         }
4288
4289         if (rc) {
4290                 dev_err(dev, "Couldn't initialize crq. rc=%d\n", rc);
4291                 return rc;
4292         }
4293
4294         adapter->from_passive_init = false;
4295
4296         init_completion(&adapter->init_done);
4297         adapter->init_done_rc = 0;
4298         ibmvnic_send_crq_init(adapter);
4299         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
4300                 dev_err(dev, "Initialization sequence timed out\n");
4301                 return -1;
4302         }
4303
4304         if (adapter->init_done_rc) {
4305                 release_crq_queue(adapter);
4306                 return adapter->init_done_rc;
4307         }
4308
4309         if (adapter->from_passive_init) {
4310                 adapter->state = VNIC_OPEN;
4311                 adapter->from_passive_init = false;
4312                 return -1;
4313         }
4314
4315         if (adapter->resetting && !adapter->wait_for_reset)
4316                 rc = reset_sub_crq_queues(adapter);
4317         else
4318                 rc = init_sub_crqs(adapter);
4319         if (rc) {
4320                 dev_err(dev, "Initialization of sub crqs failed\n");
4321                 release_crq_queue(adapter);
4322                 return rc;
4323         }
4324
4325         rc = init_sub_crq_irqs(adapter);
4326         if (rc) {
4327                 dev_err(dev, "Failed to initialize sub crq irqs\n");
4328                 release_crq_queue(adapter);
4329         }
4330
4331         return rc;
4332 }
4333
4334 static struct device_attribute dev_attr_failover;
4335
4336 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
4337 {
4338         struct ibmvnic_adapter *adapter;
4339         struct net_device *netdev;
4340         unsigned char *mac_addr_p;
4341         int rc;
4342
4343         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
4344                 dev->unit_address);
4345
4346         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
4347                                                         VETH_MAC_ADDR, NULL);
4348         if (!mac_addr_p) {
4349                 dev_err(&dev->dev,
4350                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
4351                         __FILE__, __LINE__);
4352                 return 0;
4353         }
4354
4355         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
4356                                    IBMVNIC_MAX_QUEUES);
4357         if (!netdev)
4358                 return -ENOMEM;
4359
4360         adapter = netdev_priv(netdev);
4361         adapter->state = VNIC_PROBING;
4362         dev_set_drvdata(&dev->dev, netdev);
4363         adapter->vdev = dev;
4364         adapter->netdev = netdev;
4365
4366         ether_addr_copy(adapter->mac_addr, mac_addr_p);
4367         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4368         netdev->irq = dev->irq;
4369         netdev->netdev_ops = &ibmvnic_netdev_ops;
4370         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
4371         SET_NETDEV_DEV(netdev, &dev->dev);
4372
4373         spin_lock_init(&adapter->stats_lock);
4374
4375         INIT_LIST_HEAD(&adapter->errors);
4376         spin_lock_init(&adapter->error_list_lock);
4377
4378         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
4379         INIT_LIST_HEAD(&adapter->rwi_list);
4380         mutex_init(&adapter->reset_lock);
4381         mutex_init(&adapter->rwi_lock);
4382         adapter->resetting = false;
4383
4384         adapter->mac_change_pending = false;
4385
4386         do {
4387                 rc = ibmvnic_init(adapter);
4388                 if (rc && rc != EAGAIN)
4389                         goto ibmvnic_init_fail;
4390         } while (rc == EAGAIN);
4391
4392         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4393         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4394         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4395
4396         rc = device_create_file(&dev->dev, &dev_attr_failover);
4397         if (rc)
4398                 goto ibmvnic_init_fail;
4399
4400         netif_carrier_off(netdev);
4401         rc = register_netdev(netdev);
4402         if (rc) {
4403                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
4404                 goto ibmvnic_register_fail;
4405         }
4406         dev_info(&dev->dev, "ibmvnic registered\n");
4407
4408         adapter->state = VNIC_PROBED;
4409
4410         adapter->wait_for_reset = false;
4411
4412         return 0;
4413
4414 ibmvnic_register_fail:
4415         device_remove_file(&dev->dev, &dev_attr_failover);
4416
4417 ibmvnic_init_fail:
4418         release_sub_crqs(adapter);
4419         release_crq_queue(adapter);
4420         free_netdev(netdev);
4421
4422         return rc;
4423 }
4424
4425 static int ibmvnic_remove(struct vio_dev *dev)
4426 {
4427         struct net_device *netdev = dev_get_drvdata(&dev->dev);
4428         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4429
4430         adapter->state = VNIC_REMOVING;
4431         unregister_netdev(netdev);
4432         mutex_lock(&adapter->reset_lock);
4433
4434         release_resources(adapter);
4435         release_sub_crqs(adapter);
4436         release_crq_queue(adapter);
4437
4438         adapter->state = VNIC_REMOVED;
4439
4440         mutex_unlock(&adapter->reset_lock);
4441         device_remove_file(&dev->dev, &dev_attr_failover);
4442         free_netdev(netdev);
4443         dev_set_drvdata(&dev->dev, NULL);
4444
4445         return 0;
4446 }
4447
4448 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
4449                               const char *buf, size_t count)
4450 {
4451         struct net_device *netdev = dev_get_drvdata(dev);
4452         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4453         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
4454         __be64 session_token;
4455         long rc;
4456
4457         if (!sysfs_streq(buf, "1"))
4458                 return -EINVAL;
4459
4460         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
4461                          H_GET_SESSION_TOKEN, 0, 0, 0);
4462         if (rc) {
4463                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
4464                            rc);
4465                 return -EINVAL;
4466         }
4467
4468         session_token = (__be64)retbuf[0];
4469         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
4470                    be64_to_cpu(session_token));
4471         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4472                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
4473         if (rc) {
4474                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
4475                            rc);
4476                 return -EINVAL;
4477         }
4478
4479         return count;
4480 }
4481
4482 static DEVICE_ATTR_WO(failover);
4483
4484 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
4485 {
4486         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
4487         struct ibmvnic_adapter *adapter;
4488         struct iommu_table *tbl;
4489         unsigned long ret = 0;
4490         int i;
4491
4492         tbl = get_iommu_table_base(&vdev->dev);
4493
4494         /* netdev inits at probe time along with the structures we need below*/
4495         if (!netdev)
4496                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
4497
4498         adapter = netdev_priv(netdev);
4499
4500         ret += PAGE_SIZE; /* the crq message queue */
4501         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
4502
4503         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
4504                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
4505
4506         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4507              i++)
4508                 ret += adapter->rx_pool[i].size *
4509                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
4510
4511         return ret;
4512 }
4513
4514 static int ibmvnic_resume(struct device *dev)
4515 {
4516         struct net_device *netdev = dev_get_drvdata(dev);
4517         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4518
4519         if (adapter->state != VNIC_OPEN)
4520                 return 0;
4521
4522         tasklet_schedule(&adapter->tasklet);
4523
4524         return 0;
4525 }
4526
4527 static const struct vio_device_id ibmvnic_device_table[] = {
4528         {"network", "IBM,vnic"},
4529         {"", "" }
4530 };
4531 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
4532
4533 static const struct dev_pm_ops ibmvnic_pm_ops = {
4534         .resume = ibmvnic_resume
4535 };
4536
4537 static struct vio_driver ibmvnic_driver = {
4538         .id_table       = ibmvnic_device_table,
4539         .probe          = ibmvnic_probe,
4540         .remove         = ibmvnic_remove,
4541         .get_desired_dma = ibmvnic_get_desired_dma,
4542         .name           = ibmvnic_driver_name,
4543         .pm             = &ibmvnic_pm_ops,
4544 };
4545
4546 /* module functions */
4547 static int __init ibmvnic_module_init(void)
4548 {
4549         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
4550                 IBMVNIC_DRIVER_VERSION);
4551
4552         return vio_register_driver(&ibmvnic_driver);
4553 }
4554
4555 static void __exit ibmvnic_module_exit(void)
4556 {
4557         vio_unregister_driver(&ibmvnic_driver);
4558 }
4559
4560 module_init(ibmvnic_module_init);
4561 module_exit(ibmvnic_module_exit);