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 (santi_leon@yahoo.com)                                  */
   6 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
   7 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
   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                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
1836                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
1837                         napi_complete_done(napi, frames_processed);
1838                         return frames_processed;
1839                 }
1840
1841                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
1842                         break;
1843                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
1844                 rx_buff =
1845                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
1846                                                           rx_comp.correlator);
1847                 /* do error checking */
1848                 if (next->rx_comp.rc) {
1849                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
1850                                    be16_to_cpu(next->rx_comp.rc));
1851                         /* free the entry */
1852                         next->rx_comp.first = 0;
1853                         remove_buff_from_pool(adapter, rx_buff);
1854                         continue;
1855                 }
1856
1857                 length = be32_to_cpu(next->rx_comp.len);
1858                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
1859                 flags = next->rx_comp.flags;
1860                 skb = rx_buff->skb;
1861                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
1862                                         length);
1863
1864                 /* VLAN Header has been stripped by the system firmware and
1865                  * needs to be inserted by the driver
1866                  */
1867                 if (adapter->rx_vlan_header_insertion &&
1868                     (flags & IBMVNIC_VLAN_STRIPPED))
1869                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1870                                                ntohs(next->rx_comp.vlan_tci));
1871
1872                 /* free the entry */
1873                 next->rx_comp.first = 0;
1874                 remove_buff_from_pool(adapter, rx_buff);
1875
1876                 skb_put(skb, length);
1877                 skb->protocol = eth_type_trans(skb, netdev);
1878                 skb_record_rx_queue(skb, scrq_num);
1879
1880                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
1881                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
1882                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1883                 }
1884
1885                 length = skb->len;
1886                 napi_gro_receive(napi, skb); /* send it up */
1887                 netdev->stats.rx_packets++;
1888                 netdev->stats.rx_bytes += length;
1889                 adapter->rx_stats_buffers[scrq_num].packets++;
1890                 adapter->rx_stats_buffers[scrq_num].bytes += length;
1891                 frames_processed++;
1892         }
1893
1894         if (adapter->state != VNIC_CLOSING)
1895                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
1896
1897         if (frames_processed < budget) {
1898                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
1899                 napi_complete_done(napi, frames_processed);
1900                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
1901                     napi_reschedule(napi)) {
1902                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
1903                         goto restart_poll;
1904                 }
1905         }
1906         return frames_processed;
1907 }
1908
1909 #ifdef CONFIG_NET_POLL_CONTROLLER
1910 static void ibmvnic_netpoll_controller(struct net_device *dev)
1911 {
1912         struct ibmvnic_adapter *adapter = netdev_priv(dev);
1913         int i;
1914
1915         replenish_pools(netdev_priv(dev));
1916         for (i = 0; i < adapter->req_rx_queues; i++)
1917                 ibmvnic_interrupt_rx(adapter->rx_scrq[i]->irq,
1918                                      adapter->rx_scrq[i]);
1919 }
1920 #endif
1921
1922 static int wait_for_reset(struct ibmvnic_adapter *adapter)
1923 {
1924         adapter->fallback.mtu = adapter->req_mtu;
1925         adapter->fallback.rx_queues = adapter->req_rx_queues;
1926         adapter->fallback.tx_queues = adapter->req_tx_queues;
1927         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
1928         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
1929
1930         init_completion(&adapter->reset_done);
1931         ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
1932         adapter->wait_for_reset = true;
1933         wait_for_completion(&adapter->reset_done);
1934
1935         if (adapter->reset_done_rc) {
1936                 adapter->desired.mtu = adapter->fallback.mtu;
1937                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
1938                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
1939                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
1940                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
1941
1942                 init_completion(&adapter->reset_done);
1943                 ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
1944                 wait_for_completion(&adapter->reset_done);
1945         }
1946         adapter->wait_for_reset = false;
1947
1948         return adapter->reset_done_rc;
1949 }
1950
1951 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
1952 {
1953         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1954
1955         adapter->desired.mtu = new_mtu + ETH_HLEN;
1956
1957         return wait_for_reset(adapter);
1958 }
1959
1960 static const struct net_device_ops ibmvnic_netdev_ops = {
1961         .ndo_open               = ibmvnic_open,
1962         .ndo_stop               = ibmvnic_close,
1963         .ndo_start_xmit         = ibmvnic_xmit,
1964         .ndo_set_rx_mode        = ibmvnic_set_multi,
1965         .ndo_set_mac_address    = ibmvnic_set_mac,
1966         .ndo_validate_addr      = eth_validate_addr,
1967         .ndo_tx_timeout         = ibmvnic_tx_timeout,
1968 #ifdef CONFIG_NET_POLL_CONTROLLER
1969         .ndo_poll_controller    = ibmvnic_netpoll_controller,
1970 #endif
1971         .ndo_change_mtu         = ibmvnic_change_mtu,
1972 };
1973
1974 /* ethtool functions */
1975
1976 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
1977                                       struct ethtool_link_ksettings *cmd)
1978 {
1979         u32 supported, advertising;
1980
1981         supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
1982                           SUPPORTED_FIBRE);
1983         advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
1984                             ADVERTISED_FIBRE);
1985         cmd->base.speed = SPEED_1000;
1986         cmd->base.duplex = DUPLEX_FULL;
1987         cmd->base.port = PORT_FIBRE;
1988         cmd->base.phy_address = 0;
1989         cmd->base.autoneg = AUTONEG_ENABLE;
1990
1991         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
1992                                                 supported);
1993         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
1994                                                 advertising);
1995
1996         return 0;
1997 }
1998
1999 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2000                                 struct ethtool_drvinfo *info)
2001 {
2002         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2003
2004         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2005         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2006         strlcpy(info->fw_version, adapter->fw_version,
2007                 sizeof(info->fw_version));
2008 }
2009
2010 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2011 {
2012         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2013
2014         return adapter->msg_enable;
2015 }
2016
2017 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2018 {
2019         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2020
2021         adapter->msg_enable = data;
2022 }
2023
2024 static u32 ibmvnic_get_link(struct net_device *netdev)
2025 {
2026         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2027
2028         /* Don't need to send a query because we request a logical link up at
2029          * init and then we wait for link state indications
2030          */
2031         return adapter->logical_link_state;
2032 }
2033
2034 static void ibmvnic_get_ringparam(struct net_device *netdev,
2035                                   struct ethtool_ringparam *ring)
2036 {
2037         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2038
2039         ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2040         ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2041         ring->rx_mini_max_pending = 0;
2042         ring->rx_jumbo_max_pending = 0;
2043         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2044         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2045         ring->rx_mini_pending = 0;
2046         ring->rx_jumbo_pending = 0;
2047 }
2048
2049 static int ibmvnic_set_ringparam(struct net_device *netdev,
2050                                  struct ethtool_ringparam *ring)
2051 {
2052         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2053
2054         if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
2055             ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
2056                 netdev_err(netdev, "Invalid request.\n");
2057                 netdev_err(netdev, "Max tx buffers = %llu\n",
2058                            adapter->max_rx_add_entries_per_subcrq);
2059                 netdev_err(netdev, "Max rx buffers = %llu\n",
2060                            adapter->max_tx_entries_per_subcrq);
2061                 return -EINVAL;
2062         }
2063
2064         adapter->desired.rx_entries = ring->rx_pending;
2065         adapter->desired.tx_entries = ring->tx_pending;
2066
2067         return wait_for_reset(adapter);
2068 }
2069
2070 static void ibmvnic_get_channels(struct net_device *netdev,
2071                                  struct ethtool_channels *channels)
2072 {
2073         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2074
2075         channels->max_rx = adapter->max_rx_queues;
2076         channels->max_tx = adapter->max_tx_queues;
2077         channels->max_other = 0;
2078         channels->max_combined = 0;
2079         channels->rx_count = adapter->req_rx_queues;
2080         channels->tx_count = adapter->req_tx_queues;
2081         channels->other_count = 0;
2082         channels->combined_count = 0;
2083 }
2084
2085 static int ibmvnic_set_channels(struct net_device *netdev,
2086                                 struct ethtool_channels *channels)
2087 {
2088         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2089
2090         adapter->desired.rx_queues = channels->rx_count;
2091         adapter->desired.tx_queues = channels->tx_count;
2092
2093         return wait_for_reset(adapter);
2094 }
2095
2096 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2097 {
2098         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2099         int i;
2100
2101         if (stringset != ETH_SS_STATS)
2102                 return;
2103
2104         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
2105                 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2106
2107         for (i = 0; i < adapter->req_tx_queues; i++) {
2108                 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2109                 data += ETH_GSTRING_LEN;
2110
2111                 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2112                 data += ETH_GSTRING_LEN;
2113
2114                 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
2115                 data += ETH_GSTRING_LEN;
2116         }
2117
2118         for (i = 0; i < adapter->req_rx_queues; i++) {
2119                 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2120                 data += ETH_GSTRING_LEN;
2121
2122                 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2123                 data += ETH_GSTRING_LEN;
2124
2125                 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2126                 data += ETH_GSTRING_LEN;
2127         }
2128 }
2129
2130 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2131 {
2132         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2133
2134         switch (sset) {
2135         case ETH_SS_STATS:
2136                 return ARRAY_SIZE(ibmvnic_stats) +
2137                        adapter->req_tx_queues * NUM_TX_STATS +
2138                        adapter->req_rx_queues * NUM_RX_STATS;
2139         default:
2140                 return -EOPNOTSUPP;
2141         }
2142 }
2143
2144 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2145                                       struct ethtool_stats *stats, u64 *data)
2146 {
2147         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2148         union ibmvnic_crq crq;
2149         int i, j;
2150
2151         memset(&crq, 0, sizeof(crq));
2152         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2153         crq.request_statistics.cmd = REQUEST_STATISTICS;
2154         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2155         crq.request_statistics.len =
2156             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2157
2158         /* Wait for data to be written */
2159         init_completion(&adapter->stats_done);
2160         ibmvnic_send_crq(adapter, &crq);
2161         wait_for_completion(&adapter->stats_done);
2162
2163         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2164                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2165                                                 ibmvnic_stats[i].offset));
2166
2167         for (j = 0; j < adapter->req_tx_queues; j++) {
2168                 data[i] = adapter->tx_stats_buffers[j].packets;
2169                 i++;
2170                 data[i] = adapter->tx_stats_buffers[j].bytes;
2171                 i++;
2172                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2173                 i++;
2174         }
2175
2176         for (j = 0; j < adapter->req_rx_queues; j++) {
2177                 data[i] = adapter->rx_stats_buffers[j].packets;
2178                 i++;
2179                 data[i] = adapter->rx_stats_buffers[j].bytes;
2180                 i++;
2181                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2182                 i++;
2183         }
2184 }
2185
2186 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2187         .get_drvinfo            = ibmvnic_get_drvinfo,
2188         .get_msglevel           = ibmvnic_get_msglevel,
2189         .set_msglevel           = ibmvnic_set_msglevel,
2190         .get_link               = ibmvnic_get_link,
2191         .get_ringparam          = ibmvnic_get_ringparam,
2192         .set_ringparam          = ibmvnic_set_ringparam,
2193         .get_channels           = ibmvnic_get_channels,
2194         .set_channels           = ibmvnic_set_channels,
2195         .get_strings            = ibmvnic_get_strings,
2196         .get_sset_count         = ibmvnic_get_sset_count,
2197         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2198         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2199 };
2200
2201 /* Routines for managing CRQs/sCRQs  */
2202
2203 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2204                                    struct ibmvnic_sub_crq_queue *scrq)
2205 {
2206         int rc;
2207
2208         if (scrq->irq) {
2209                 free_irq(scrq->irq, scrq);
2210                 irq_dispose_mapping(scrq->irq);
2211                 scrq->irq = 0;
2212         }
2213
2214         memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2215         scrq->cur = 0;
2216
2217         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2218                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2219         return rc;
2220 }
2221
2222 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2223 {
2224         int i, rc;
2225
2226         for (i = 0; i < adapter->req_tx_queues; i++) {
2227                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2228                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2229                 if (rc)
2230                         return rc;
2231         }
2232
2233         for (i = 0; i < adapter->req_rx_queues; i++) {
2234                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2235                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2236                 if (rc)
2237                         return rc;
2238         }
2239
2240         return rc;
2241 }
2242
2243 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2244                                   struct ibmvnic_sub_crq_queue *scrq)
2245 {
2246         struct device *dev = &adapter->vdev->dev;
2247         long rc;
2248
2249         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2250
2251         /* Close the sub-crqs */
2252         do {
2253                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2254                                         adapter->vdev->unit_address,
2255                                         scrq->crq_num);
2256         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2257
2258         if (rc) {
2259                 netdev_err(adapter->netdev,
2260                            "Failed to release sub-CRQ %16lx, rc = %ld\n",
2261                            scrq->crq_num, rc);
2262         }
2263
2264         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2265                          DMA_BIDIRECTIONAL);
2266         free_pages((unsigned long)scrq->msgs, 2);
2267         kfree(scrq);
2268 }
2269
2270 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2271                                                         *adapter)
2272 {
2273         struct device *dev = &adapter->vdev->dev;
2274         struct ibmvnic_sub_crq_queue *scrq;
2275         int rc;
2276
2277         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2278         if (!scrq)
2279                 return NULL;
2280
2281         scrq->msgs =
2282                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2283         if (!scrq->msgs) {
2284                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2285                 goto zero_page_failed;
2286         }
2287
2288         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2289                                          DMA_BIDIRECTIONAL);
2290         if (dma_mapping_error(dev, scrq->msg_token)) {
2291                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2292                 goto map_failed;
2293         }
2294
2295         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2296                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2297
2298         if (rc == H_RESOURCE)
2299                 rc = ibmvnic_reset_crq(adapter);
2300
2301         if (rc == H_CLOSED) {
2302                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2303         } else if (rc) {
2304                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2305                 goto reg_failed;
2306         }
2307
2308         scrq->adapter = adapter;
2309         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2310         spin_lock_init(&scrq->lock);
2311
2312         netdev_dbg(adapter->netdev,
2313                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2314                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2315
2316         return scrq;
2317
2318 reg_failed:
2319         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2320                          DMA_BIDIRECTIONAL);
2321 map_failed:
2322         free_pages((unsigned long)scrq->msgs, 2);
2323 zero_page_failed:
2324         kfree(scrq);
2325
2326         return NULL;
2327 }
2328
2329 static void release_sub_crqs(struct ibmvnic_adapter *adapter)
2330 {
2331         int i;
2332
2333         if (adapter->tx_scrq) {
2334                 for (i = 0; i < adapter->req_tx_queues; i++) {
2335                         if (!adapter->tx_scrq[i])
2336                                 continue;
2337
2338                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
2339                                    i);
2340                         if (adapter->tx_scrq[i]->irq) {
2341                                 free_irq(adapter->tx_scrq[i]->irq,
2342                                          adapter->tx_scrq[i]);
2343                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
2344                                 adapter->tx_scrq[i]->irq = 0;
2345                         }
2346
2347                         release_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2348                 }
2349
2350                 kfree(adapter->tx_scrq);
2351                 adapter->tx_scrq = NULL;
2352         }
2353
2354         if (adapter->rx_scrq) {
2355                 for (i = 0; i < adapter->req_rx_queues; i++) {
2356                         if (!adapter->rx_scrq[i])
2357                                 continue;
2358
2359                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
2360                                    i);
2361                         if (adapter->rx_scrq[i]->irq) {
2362                                 free_irq(adapter->rx_scrq[i]->irq,
2363                                          adapter->rx_scrq[i]);
2364                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
2365                                 adapter->rx_scrq[i]->irq = 0;
2366                         }
2367
2368                         release_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2369                 }
2370
2371                 kfree(adapter->rx_scrq);
2372                 adapter->rx_scrq = NULL;
2373         }
2374 }
2375
2376 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
2377                             struct ibmvnic_sub_crq_queue *scrq)
2378 {
2379         struct device *dev = &adapter->vdev->dev;
2380         unsigned long rc;
2381
2382         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2383                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2384         if (rc)
2385                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
2386                         scrq->hw_irq, rc);
2387         return rc;
2388 }
2389
2390 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
2391                            struct ibmvnic_sub_crq_queue *scrq)
2392 {
2393         struct device *dev = &adapter->vdev->dev;
2394         unsigned long rc;
2395
2396         if (scrq->hw_irq > 0x100000000ULL) {
2397                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
2398                 return 1;
2399         }
2400
2401         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2402                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2403         if (rc)
2404                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
2405                         scrq->hw_irq, rc);
2406         return rc;
2407 }
2408
2409 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
2410                                struct ibmvnic_sub_crq_queue *scrq)
2411 {
2412         struct device *dev = &adapter->vdev->dev;
2413         struct ibmvnic_tx_buff *txbuff;
2414         union sub_crq *next;
2415         int index;
2416         int i, j;
2417         u8 first;
2418
2419 restart_loop:
2420         while (pending_scrq(adapter, scrq)) {
2421                 unsigned int pool = scrq->pool_index;
2422
2423                 next = ibmvnic_next_scrq(adapter, scrq);
2424                 for (i = 0; i < next->tx_comp.num_comps; i++) {
2425                         if (next->tx_comp.rcs[i]) {
2426                                 dev_err(dev, "tx error %x\n",
2427                                         next->tx_comp.rcs[i]);
2428                                 continue;
2429                         }
2430                         index = be32_to_cpu(next->tx_comp.correlators[i]);
2431                         txbuff = &adapter->tx_pool[pool].tx_buff[index];
2432
2433                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
2434                                 if (!txbuff->data_dma[j])
2435                                         continue;
2436
2437                                 txbuff->data_dma[j] = 0;
2438                         }
2439                         /* if sub_crq was sent indirectly */
2440                         first = txbuff->indir_arr[0].generic.first;
2441                         if (first == IBMVNIC_CRQ_CMD) {
2442                                 dma_unmap_single(dev, txbuff->indir_dma,
2443                                                  sizeof(txbuff->indir_arr),
2444                                                  DMA_TO_DEVICE);
2445                         }
2446
2447                         if (txbuff->last_frag) {
2448                                 dev_kfree_skb_any(txbuff->skb);
2449                                 txbuff->skb = NULL;
2450                         }
2451
2452                         adapter->tx_pool[pool].free_map[adapter->tx_pool[pool].
2453                                                      producer_index] = index;
2454                         adapter->tx_pool[pool].producer_index =
2455                             (adapter->tx_pool[pool].producer_index + 1) %
2456                             adapter->req_tx_entries_per_subcrq;
2457                 }
2458                 /* remove tx_comp scrq*/
2459                 next->tx_comp.first = 0;
2460
2461                 if (atomic_sub_return(next->tx_comp.num_comps, &scrq->used) <=
2462                     (adapter->req_tx_entries_per_subcrq / 2) &&
2463                     __netif_subqueue_stopped(adapter->netdev,
2464                                              scrq->pool_index)) {
2465                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
2466                         netdev_info(adapter->netdev, "Started queue %d\n",
2467                                     scrq->pool_index);
2468                 }
2469         }
2470
2471         enable_scrq_irq(adapter, scrq);
2472
2473         if (pending_scrq(adapter, scrq)) {
2474                 disable_scrq_irq(adapter, scrq);
2475                 goto restart_loop;
2476         }
2477
2478         return 0;
2479 }
2480
2481 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
2482 {
2483         struct ibmvnic_sub_crq_queue *scrq = instance;
2484         struct ibmvnic_adapter *adapter = scrq->adapter;
2485
2486         disable_scrq_irq(adapter, scrq);
2487         ibmvnic_complete_tx(adapter, scrq);
2488
2489         return IRQ_HANDLED;
2490 }
2491
2492 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
2493 {
2494         struct ibmvnic_sub_crq_queue *scrq = instance;
2495         struct ibmvnic_adapter *adapter = scrq->adapter;
2496
2497         /* When booting a kdump kernel we can hit pending interrupts
2498          * prior to completing driver initialization.
2499          */
2500         if (unlikely(adapter->state != VNIC_OPEN))
2501                 return IRQ_NONE;
2502
2503         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
2504
2505         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
2506                 disable_scrq_irq(adapter, scrq);
2507                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
2508         }
2509
2510         return IRQ_HANDLED;
2511 }
2512
2513 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
2514 {
2515         struct device *dev = &adapter->vdev->dev;
2516         struct ibmvnic_sub_crq_queue *scrq;
2517         int i = 0, j = 0;
2518         int rc = 0;
2519
2520         for (i = 0; i < adapter->req_tx_queues; i++) {
2521                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
2522                            i);
2523                 scrq = adapter->tx_scrq[i];
2524                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2525
2526                 if (!scrq->irq) {
2527                         rc = -EINVAL;
2528                         dev_err(dev, "Error mapping irq\n");
2529                         goto req_tx_irq_failed;
2530                 }
2531
2532                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
2533                                  0, "ibmvnic_tx", scrq);
2534
2535                 if (rc) {
2536                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
2537                                 scrq->irq, rc);
2538                         irq_dispose_mapping(scrq->irq);
2539                         goto req_rx_irq_failed;
2540                 }
2541         }
2542
2543         for (i = 0; i < adapter->req_rx_queues; i++) {
2544                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
2545                            i);
2546                 scrq = adapter->rx_scrq[i];
2547                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2548                 if (!scrq->irq) {
2549                         rc = -EINVAL;
2550                         dev_err(dev, "Error mapping irq\n");
2551                         goto req_rx_irq_failed;
2552                 }
2553                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
2554                                  0, "ibmvnic_rx", scrq);
2555                 if (rc) {
2556                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
2557                                 scrq->irq, rc);
2558                         irq_dispose_mapping(scrq->irq);
2559                         goto req_rx_irq_failed;
2560                 }
2561         }
2562         return rc;
2563
2564 req_rx_irq_failed:
2565         for (j = 0; j < i; j++) {
2566                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
2567                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2568         }
2569         i = adapter->req_tx_queues;
2570 req_tx_irq_failed:
2571         for (j = 0; j < i; j++) {
2572                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
2573                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2574         }
2575         release_sub_crqs(adapter);
2576         return rc;
2577 }
2578
2579 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
2580 {
2581         struct device *dev = &adapter->vdev->dev;
2582         struct ibmvnic_sub_crq_queue **allqueues;
2583         int registered_queues = 0;
2584         int total_queues;
2585         int more = 0;
2586         int i;
2587
2588         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
2589
2590         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
2591         if (!allqueues)
2592                 return -1;
2593
2594         for (i = 0; i < total_queues; i++) {
2595                 allqueues[i] = init_sub_crq_queue(adapter);
2596                 if (!allqueues[i]) {
2597                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
2598                         break;
2599                 }
2600                 registered_queues++;
2601         }
2602
2603         /* Make sure we were able to register the minimum number of queues */
2604         if (registered_queues <
2605             adapter->min_tx_queues + adapter->min_rx_queues) {
2606                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
2607                 goto tx_failed;
2608         }
2609
2610         /* Distribute the failed allocated queues*/
2611         for (i = 0; i < total_queues - registered_queues + more ; i++) {
2612                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
2613                 switch (i % 3) {
2614                 case 0:
2615                         if (adapter->req_rx_queues > adapter->min_rx_queues)
2616                                 adapter->req_rx_queues--;
2617                         else
2618                                 more++;
2619                         break;
2620                 case 1:
2621                         if (adapter->req_tx_queues > adapter->min_tx_queues)
2622                                 adapter->req_tx_queues--;
2623                         else
2624                                 more++;
2625                         break;
2626                 }
2627         }
2628
2629         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
2630                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
2631         if (!adapter->tx_scrq)
2632                 goto tx_failed;
2633
2634         for (i = 0; i < adapter->req_tx_queues; i++) {
2635                 adapter->tx_scrq[i] = allqueues[i];
2636                 adapter->tx_scrq[i]->pool_index = i;
2637         }
2638
2639         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
2640                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
2641         if (!adapter->rx_scrq)
2642                 goto rx_failed;
2643
2644         for (i = 0; i < adapter->req_rx_queues; i++) {
2645                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
2646                 adapter->rx_scrq[i]->scrq_num = i;
2647         }
2648
2649         kfree(allqueues);
2650         return 0;
2651
2652 rx_failed:
2653         kfree(adapter->tx_scrq);
2654         adapter->tx_scrq = NULL;
2655 tx_failed:
2656         for (i = 0; i < registered_queues; i++)
2657                 release_sub_crq_queue(adapter, allqueues[i]);
2658         kfree(allqueues);
2659         return -1;
2660 }
2661
2662 static void ibmvnic_send_req_caps(struct ibmvnic_adapter *adapter, int retry)
2663 {
2664         struct device *dev = &adapter->vdev->dev;
2665         union ibmvnic_crq crq;
2666         int max_entries;
2667
2668         if (!retry) {
2669                 /* Sub-CRQ entries are 32 byte long */
2670                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
2671
2672                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
2673                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
2674                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
2675                         return;
2676                 }
2677
2678                 if (adapter->desired.mtu)
2679                         adapter->req_mtu = adapter->desired.mtu;
2680                 else
2681                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
2682
2683                 if (!adapter->desired.tx_entries)
2684                         adapter->desired.tx_entries =
2685                                         adapter->max_tx_entries_per_subcrq;
2686                 if (!adapter->desired.rx_entries)
2687                         adapter->desired.rx_entries =
2688                                         adapter->max_rx_add_entries_per_subcrq;
2689
2690                 max_entries = IBMVNIC_MAX_LTB_SIZE /
2691                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
2692
2693                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
2694                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
2695                         adapter->desired.tx_entries = max_entries;
2696                 }
2697
2698                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
2699                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
2700                         adapter->desired.rx_entries = max_entries;
2701                 }
2702
2703                 if (adapter->desired.tx_entries)
2704                         adapter->req_tx_entries_per_subcrq =
2705                                         adapter->desired.tx_entries;
2706                 else
2707                         adapter->req_tx_entries_per_subcrq =
2708                                         adapter->max_tx_entries_per_subcrq;
2709
2710                 if (adapter->desired.rx_entries)
2711                         adapter->req_rx_add_entries_per_subcrq =
2712                                         adapter->desired.rx_entries;
2713                 else
2714                         adapter->req_rx_add_entries_per_subcrq =
2715                                         adapter->max_rx_add_entries_per_subcrq;
2716
2717                 if (adapter->desired.tx_queues)
2718                         adapter->req_tx_queues =
2719                                         adapter->desired.tx_queues;
2720                 else
2721                         adapter->req_tx_queues =
2722                                         adapter->opt_tx_comp_sub_queues;
2723
2724                 if (adapter->desired.rx_queues)
2725                         adapter->req_rx_queues =
2726                                         adapter->desired.rx_queues;
2727                 else
2728                         adapter->req_rx_queues =
2729                                         adapter->opt_rx_comp_queues;
2730
2731                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
2732         }
2733
2734         memset(&crq, 0, sizeof(crq));
2735         crq.request_capability.first = IBMVNIC_CRQ_CMD;
2736         crq.request_capability.cmd = REQUEST_CAPABILITY;
2737
2738         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
2739         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
2740         atomic_inc(&adapter->running_cap_crqs);
2741         ibmvnic_send_crq(adapter, &crq);
2742
2743         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
2744         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
2745         atomic_inc(&adapter->running_cap_crqs);
2746         ibmvnic_send_crq(adapter, &crq);
2747
2748         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
2749         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
2750         atomic_inc(&adapter->running_cap_crqs);
2751         ibmvnic_send_crq(adapter, &crq);
2752
2753         crq.request_capability.capability =
2754             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
2755         crq.request_capability.number =
2756             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
2757         atomic_inc(&adapter->running_cap_crqs);
2758         ibmvnic_send_crq(adapter, &crq);
2759
2760         crq.request_capability.capability =
2761             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
2762         crq.request_capability.number =
2763             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
2764         atomic_inc(&adapter->running_cap_crqs);
2765         ibmvnic_send_crq(adapter, &crq);
2766
2767         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
2768         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
2769         atomic_inc(&adapter->running_cap_crqs);
2770         ibmvnic_send_crq(adapter, &crq);
2771
2772         if (adapter->netdev->flags & IFF_PROMISC) {
2773                 if (adapter->promisc_supported) {
2774                         crq.request_capability.capability =
2775                             cpu_to_be16(PROMISC_REQUESTED);
2776                         crq.request_capability.number = cpu_to_be64(1);
2777                         atomic_inc(&adapter->running_cap_crqs);
2778                         ibmvnic_send_crq(adapter, &crq);
2779                 }
2780         } else {
2781                 crq.request_capability.capability =
2782                     cpu_to_be16(PROMISC_REQUESTED);
2783                 crq.request_capability.number = cpu_to_be64(0);
2784                 atomic_inc(&adapter->running_cap_crqs);
2785                 ibmvnic_send_crq(adapter, &crq);
2786         }
2787 }
2788
2789 static int pending_scrq(struct ibmvnic_adapter *adapter,
2790                         struct ibmvnic_sub_crq_queue *scrq)
2791 {
2792         union sub_crq *entry = &scrq->msgs[scrq->cur];
2793
2794         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
2795                 return 1;
2796         else
2797                 return 0;
2798 }
2799
2800 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
2801                                         struct ibmvnic_sub_crq_queue *scrq)
2802 {
2803         union sub_crq *entry;
2804         unsigned long flags;
2805
2806         spin_lock_irqsave(&scrq->lock, flags);
2807         entry = &scrq->msgs[scrq->cur];
2808         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
2809                 if (++scrq->cur == scrq->size)
2810                         scrq->cur = 0;
2811         } else {
2812                 entry = NULL;
2813         }
2814         spin_unlock_irqrestore(&scrq->lock, flags);
2815
2816         return entry;
2817 }
2818
2819 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
2820 {
2821         struct ibmvnic_crq_queue *queue = &adapter->crq;
2822         union ibmvnic_crq *crq;
2823
2824         crq = &queue->msgs[queue->cur];
2825         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
2826                 if (++queue->cur == queue->size)
2827                         queue->cur = 0;
2828         } else {
2829                 crq = NULL;
2830         }
2831
2832         return crq;
2833 }
2834
2835 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
2836                        union sub_crq *sub_crq)
2837 {
2838         unsigned int ua = adapter->vdev->unit_address;
2839         struct device *dev = &adapter->vdev->dev;
2840         u64 *u64_crq = (u64 *)sub_crq;
2841         int rc;
2842
2843         netdev_dbg(adapter->netdev,
2844                    "Sending sCRQ %016lx: %016lx %016lx %016lx %016lx\n",
2845                    (unsigned long int)cpu_to_be64(remote_handle),
2846                    (unsigned long int)cpu_to_be64(u64_crq[0]),
2847                    (unsigned long int)cpu_to_be64(u64_crq[1]),
2848                    (unsigned long int)cpu_to_be64(u64_crq[2]),
2849                    (unsigned long int)cpu_to_be64(u64_crq[3]));
2850
2851         /* Make sure the hypervisor sees the complete request */
2852         mb();
2853
2854         rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
2855                                 cpu_to_be64(remote_handle),
2856                                 cpu_to_be64(u64_crq[0]),
2857                                 cpu_to_be64(u64_crq[1]),
2858                                 cpu_to_be64(u64_crq[2]),
2859                                 cpu_to_be64(u64_crq[3]));
2860
2861         if (rc) {
2862                 if (rc == H_CLOSED)
2863                         dev_warn(dev, "CRQ Queue closed\n");
2864                 dev_err(dev, "Send error (rc=%d)\n", rc);
2865         }
2866
2867         return rc;
2868 }
2869
2870 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
2871                                 u64 remote_handle, u64 ioba, u64 num_entries)
2872 {
2873         unsigned int ua = adapter->vdev->unit_address;
2874         struct device *dev = &adapter->vdev->dev;
2875         int rc;
2876
2877         /* Make sure the hypervisor sees the complete request */
2878         mb();
2879         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
2880                                 cpu_to_be64(remote_handle),
2881                                 ioba, num_entries);
2882
2883         if (rc) {
2884                 if (rc == H_CLOSED)
2885                         dev_warn(dev, "CRQ Queue closed\n");
2886                 dev_err(dev, "Send (indirect) error (rc=%d)\n", rc);
2887         }
2888
2889         return rc;
2890 }
2891
2892 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
2893                             union ibmvnic_crq *crq)
2894 {
2895         unsigned int ua = adapter->vdev->unit_address;
2896         struct device *dev = &adapter->vdev->dev;
2897         u64 *u64_crq = (u64 *)crq;
2898         int rc;
2899
2900         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
2901                    (unsigned long int)cpu_to_be64(u64_crq[0]),
2902                    (unsigned long int)cpu_to_be64(u64_crq[1]));
2903
2904         /* Make sure the hypervisor sees the complete request */
2905         mb();
2906
2907         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
2908                                 cpu_to_be64(u64_crq[0]),
2909                                 cpu_to_be64(u64_crq[1]));
2910
2911         if (rc) {
2912                 if (rc == H_CLOSED)
2913                         dev_warn(dev, "CRQ Queue closed\n");
2914                 dev_warn(dev, "Send error (rc=%d)\n", rc);
2915         }
2916
2917         return rc;
2918 }
2919
2920 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
2921 {
2922         union ibmvnic_crq crq;
2923
2924         memset(&crq, 0, sizeof(crq));
2925         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
2926         crq.generic.cmd = IBMVNIC_CRQ_INIT;
2927         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
2928
2929         return ibmvnic_send_crq(adapter, &crq);
2930 }
2931
2932 static int send_version_xchg(struct ibmvnic_adapter *adapter)
2933 {
2934         union ibmvnic_crq crq;
2935
2936         memset(&crq, 0, sizeof(crq));
2937         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
2938         crq.version_exchange.cmd = VERSION_EXCHANGE;
2939         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
2940
2941         return ibmvnic_send_crq(adapter, &crq);
2942 }
2943
2944 struct vnic_login_client_data {
2945         u8      type;
2946         __be16  len;
2947         char    name;
2948 } __packed;
2949
2950 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
2951 {
2952         int len;
2953
2954         /* Calculate the amount of buffer space needed for the
2955          * vnic client data in the login buffer. There are four entries,
2956          * OS name, LPAR name, device name, and a null last entry.
2957          */
2958         len = 4 * sizeof(struct vnic_login_client_data);
2959         len += 6; /* "Linux" plus NULL */
2960         len += strlen(utsname()->nodename) + 1;
2961         len += strlen(adapter->netdev->name) + 1;
2962
2963         return len;
2964 }
2965
2966 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
2967                                  struct vnic_login_client_data *vlcd)
2968 {
2969         const char *os_name = "Linux";
2970         int len;
2971
2972         /* Type 1 - LPAR OS */
2973         vlcd->type = 1;
2974         len = strlen(os_name) + 1;
2975         vlcd->len = cpu_to_be16(len);
2976         strncpy(&vlcd->name, os_name, len);
2977         vlcd = (struct vnic_login_client_data *)((char *)&vlcd->name + len);
2978
2979         /* Type 2 - LPAR name */
2980         vlcd->type = 2;
2981         len = strlen(utsname()->nodename) + 1;
2982         vlcd->len = cpu_to_be16(len);
2983         strncpy(&vlcd->name, utsname()->nodename, len);
2984         vlcd = (struct vnic_login_client_data *)((char *)&vlcd->name + len);
2985
2986         /* Type 3 - device name */
2987         vlcd->type = 3;
2988         len = strlen(adapter->netdev->name) + 1;
2989         vlcd->len = cpu_to_be16(len);
2990         strncpy(&vlcd->name, adapter->netdev->name, len);
2991 }
2992
2993 static void send_login(struct ibmvnic_adapter *adapter)
2994 {
2995         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
2996         struct ibmvnic_login_buffer *login_buffer;
2997         struct device *dev = &adapter->vdev->dev;
2998         dma_addr_t rsp_buffer_token;
2999         dma_addr_t buffer_token;
3000         size_t rsp_buffer_size;
3001         union ibmvnic_crq crq;
3002         size_t buffer_size;
3003         __be64 *tx_list_p;
3004         __be64 *rx_list_p;
3005         int client_data_len;
3006         struct vnic_login_client_data *vlcd;
3007         int i;
3008
3009         client_data_len = vnic_client_data_len(adapter);
3010
3011         buffer_size =
3012             sizeof(struct ibmvnic_login_buffer) +
3013             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3014             client_data_len;
3015
3016         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3017         if (!login_buffer)
3018                 goto buf_alloc_failed;
3019
3020         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3021                                       DMA_TO_DEVICE);
3022         if (dma_mapping_error(dev, buffer_token)) {
3023                 dev_err(dev, "Couldn't map login buffer\n");
3024                 goto buf_map_failed;
3025         }
3026
3027         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3028                           sizeof(u64) * adapter->req_tx_queues +
3029                           sizeof(u64) * adapter->req_rx_queues +
3030                           sizeof(u64) * adapter->req_rx_queues +
3031                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3032
3033         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3034         if (!login_rsp_buffer)
3035                 goto buf_rsp_alloc_failed;
3036
3037         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3038                                           rsp_buffer_size, DMA_FROM_DEVICE);
3039         if (dma_mapping_error(dev, rsp_buffer_token)) {
3040                 dev_err(dev, "Couldn't map login rsp buffer\n");
3041                 goto buf_rsp_map_failed;
3042         }
3043
3044         adapter->login_buf = login_buffer;
3045         adapter->login_buf_token = buffer_token;
3046         adapter->login_buf_sz = buffer_size;
3047         adapter->login_rsp_buf = login_rsp_buffer;
3048         adapter->login_rsp_buf_token = rsp_buffer_token;
3049         adapter->login_rsp_buf_sz = rsp_buffer_size;
3050
3051         login_buffer->len = cpu_to_be32(buffer_size);
3052         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3053         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3054         login_buffer->off_txcomp_subcrqs =
3055             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3056         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3057         login_buffer->off_rxcomp_subcrqs =
3058             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3059                         sizeof(u64) * adapter->req_tx_queues);
3060         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3061         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3062
3063         tx_list_p = (__be64 *)((char *)login_buffer +
3064                                       sizeof(struct ibmvnic_login_buffer));
3065         rx_list_p = (__be64 *)((char *)login_buffer +
3066                                       sizeof(struct ibmvnic_login_buffer) +
3067                                       sizeof(u64) * adapter->req_tx_queues);
3068
3069         for (i = 0; i < adapter->req_tx_queues; i++) {
3070                 if (adapter->tx_scrq[i]) {
3071                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3072                                                    crq_num);
3073                 }
3074         }
3075
3076         for (i = 0; i < adapter->req_rx_queues; i++) {
3077                 if (adapter->rx_scrq[i]) {
3078                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3079                                                    crq_num);
3080                 }
3081         }
3082
3083         /* Insert vNIC login client data */
3084         vlcd = (struct vnic_login_client_data *)
3085                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3086         login_buffer->client_data_offset =
3087                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3088         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3089
3090         vnic_add_client_data(adapter, vlcd);
3091
3092         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3093         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3094                 netdev_dbg(adapter->netdev, "%016lx\n",
3095                            ((unsigned long int *)(adapter->login_buf))[i]);
3096         }
3097
3098         memset(&crq, 0, sizeof(crq));
3099         crq.login.first = IBMVNIC_CRQ_CMD;
3100         crq.login.cmd = LOGIN;
3101         crq.login.ioba = cpu_to_be32(buffer_token);
3102         crq.login.len = cpu_to_be32(buffer_size);
3103         ibmvnic_send_crq(adapter, &crq);
3104
3105         return;
3106
3107 buf_rsp_map_failed:
3108         kfree(login_rsp_buffer);
3109 buf_rsp_alloc_failed:
3110         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3111 buf_map_failed:
3112         kfree(login_buffer);
3113 buf_alloc_failed:
3114         return;
3115 }
3116
3117 static void send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3118                              u32 len, u8 map_id)
3119 {
3120         union ibmvnic_crq crq;
3121
3122         memset(&crq, 0, sizeof(crq));
3123         crq.request_map.first = IBMVNIC_CRQ_CMD;
3124         crq.request_map.cmd = REQUEST_MAP;
3125         crq.request_map.map_id = map_id;
3126         crq.request_map.ioba = cpu_to_be32(addr);
3127         crq.request_map.len = cpu_to_be32(len);
3128         ibmvnic_send_crq(adapter, &crq);
3129 }
3130
3131 static void send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3132 {
3133         union ibmvnic_crq crq;
3134
3135         memset(&crq, 0, sizeof(crq));
3136         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3137         crq.request_unmap.cmd = REQUEST_UNMAP;
3138         crq.request_unmap.map_id = map_id;
3139         ibmvnic_send_crq(adapter, &crq);
3140 }
3141
3142 static void send_map_query(struct ibmvnic_adapter *adapter)
3143 {
3144         union ibmvnic_crq crq;
3145
3146         memset(&crq, 0, sizeof(crq));
3147         crq.query_map.first = IBMVNIC_CRQ_CMD;
3148         crq.query_map.cmd = QUERY_MAP;
3149         ibmvnic_send_crq(adapter, &crq);
3150 }
3151
3152 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3153 static void send_cap_queries(struct ibmvnic_adapter *adapter)
3154 {
3155         union ibmvnic_crq crq;
3156
3157         atomic_set(&adapter->running_cap_crqs, 0);
3158         memset(&crq, 0, sizeof(crq));
3159         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3160         crq.query_capability.cmd = QUERY_CAPABILITY;
3161
3162         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3163         atomic_inc(&adapter->running_cap_crqs);
3164         ibmvnic_send_crq(adapter, &crq);
3165
3166         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3167         atomic_inc(&adapter->running_cap_crqs);
3168         ibmvnic_send_crq(adapter, &crq);
3169
3170         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3171         atomic_inc(&adapter->running_cap_crqs);
3172         ibmvnic_send_crq(adapter, &crq);
3173
3174         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3175         atomic_inc(&adapter->running_cap_crqs);
3176         ibmvnic_send_crq(adapter, &crq);
3177
3178         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3179         atomic_inc(&adapter->running_cap_crqs);
3180         ibmvnic_send_crq(adapter, &crq);
3181
3182         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3183         atomic_inc(&adapter->running_cap_crqs);
3184         ibmvnic_send_crq(adapter, &crq);
3185
3186         crq.query_capability.capability =
3187             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3188         atomic_inc(&adapter->running_cap_crqs);
3189         ibmvnic_send_crq(adapter, &crq);
3190
3191         crq.query_capability.capability =
3192             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3193         atomic_inc(&adapter->running_cap_crqs);
3194         ibmvnic_send_crq(adapter, &crq);
3195
3196         crq.query_capability.capability =
3197             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3198         atomic_inc(&adapter->running_cap_crqs);
3199         ibmvnic_send_crq(adapter, &crq);
3200
3201         crq.query_capability.capability =
3202             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
3203         atomic_inc(&adapter->running_cap_crqs);
3204         ibmvnic_send_crq(adapter, &crq);
3205
3206         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
3207         atomic_inc(&adapter->running_cap_crqs);
3208         ibmvnic_send_crq(adapter, &crq);
3209
3210         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
3211         atomic_inc(&adapter->running_cap_crqs);
3212         ibmvnic_send_crq(adapter, &crq);
3213
3214         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
3215         atomic_inc(&adapter->running_cap_crqs);
3216         ibmvnic_send_crq(adapter, &crq);
3217
3218         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
3219         atomic_inc(&adapter->running_cap_crqs);
3220         ibmvnic_send_crq(adapter, &crq);
3221
3222         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
3223         atomic_inc(&adapter->running_cap_crqs);
3224         ibmvnic_send_crq(adapter, &crq);
3225
3226         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
3227         atomic_inc(&adapter->running_cap_crqs);
3228         ibmvnic_send_crq(adapter, &crq);
3229
3230         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
3231         atomic_inc(&adapter->running_cap_crqs);
3232         ibmvnic_send_crq(adapter, &crq);
3233
3234         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
3235         atomic_inc(&adapter->running_cap_crqs);
3236         ibmvnic_send_crq(adapter, &crq);
3237
3238         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
3239         atomic_inc(&adapter->running_cap_crqs);
3240         ibmvnic_send_crq(adapter, &crq);
3241
3242         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
3243         atomic_inc(&adapter->running_cap_crqs);
3244         ibmvnic_send_crq(adapter, &crq);
3245
3246         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
3247         atomic_inc(&adapter->running_cap_crqs);
3248         ibmvnic_send_crq(adapter, &crq);
3249
3250         crq.query_capability.capability =
3251                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
3252         atomic_inc(&adapter->running_cap_crqs);
3253         ibmvnic_send_crq(adapter, &crq);
3254
3255         crq.query_capability.capability =
3256                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
3257         atomic_inc(&adapter->running_cap_crqs);
3258         ibmvnic_send_crq(adapter, &crq);
3259
3260         crq.query_capability.capability =
3261                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
3262         atomic_inc(&adapter->running_cap_crqs);
3263         ibmvnic_send_crq(adapter, &crq);
3264
3265         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
3266         atomic_inc(&adapter->running_cap_crqs);
3267         ibmvnic_send_crq(adapter, &crq);
3268 }
3269
3270 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
3271                                 struct ibmvnic_adapter *adapter)
3272 {
3273         struct device *dev = &adapter->vdev->dev;
3274
3275         if (crq->get_vpd_size_rsp.rc.code) {
3276                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
3277                         crq->get_vpd_size_rsp.rc.code);
3278                 complete(&adapter->fw_done);
3279                 return;
3280         }
3281
3282         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
3283         complete(&adapter->fw_done);
3284 }
3285
3286 static void handle_vpd_rsp(union ibmvnic_crq *crq,
3287                            struct ibmvnic_adapter *adapter)
3288 {
3289         struct device *dev = &adapter->vdev->dev;
3290         unsigned char *substr = NULL;
3291         u8 fw_level_len = 0;
3292
3293         memset(adapter->fw_version, 0, 32);
3294
3295         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
3296                          DMA_FROM_DEVICE);
3297
3298         if (crq->get_vpd_rsp.rc.code) {
3299                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
3300                         crq->get_vpd_rsp.rc.code);
3301                 goto complete;
3302         }
3303
3304         /* get the position of the firmware version info
3305          * located after the ASCII 'RM' substring in the buffer
3306          */
3307         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
3308         if (!substr) {
3309                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
3310                 goto complete;
3311         }
3312
3313         /* get length of firmware level ASCII substring */
3314         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
3315                 fw_level_len = *(substr + 2);
3316         } else {
3317                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
3318                 goto complete;
3319         }
3320
3321         /* copy firmware version string from vpd into adapter */
3322         if ((substr + 3 + fw_level_len) <
3323             (adapter->vpd->buff + adapter->vpd->len)) {
3324                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
3325         } else {
3326                 dev_info(dev, "FW substr extrapolated VPD buff\n");
3327         }
3328
3329 complete:
3330         if (adapter->fw_version[0] == '\0')
3331                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
3332         complete(&adapter->fw_done);
3333 }
3334
3335 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
3336 {
3337         struct device *dev = &adapter->vdev->dev;
3338         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
3339         union ibmvnic_crq crq;
3340         int i;
3341
3342         dma_unmap_single(dev, adapter->ip_offload_tok,
3343                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
3344
3345         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
3346         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
3347                 netdev_dbg(adapter->netdev, "%016lx\n",
3348                            ((unsigned long int *)(buf))[i]);
3349
3350         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
3351         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
3352         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
3353                    buf->tcp_ipv4_chksum);
3354         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
3355                    buf->tcp_ipv6_chksum);
3356         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
3357                    buf->udp_ipv4_chksum);
3358         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
3359                    buf->udp_ipv6_chksum);
3360         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
3361                    buf->large_tx_ipv4);
3362         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
3363                    buf->large_tx_ipv6);
3364         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
3365                    buf->large_rx_ipv4);
3366         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
3367                    buf->large_rx_ipv6);
3368         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
3369                    buf->max_ipv4_header_size);
3370         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
3371                    buf->max_ipv6_header_size);
3372         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
3373                    buf->max_tcp_header_size);
3374         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
3375                    buf->max_udp_header_size);
3376         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
3377                    buf->max_large_tx_size);
3378         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
3379                    buf->max_large_rx_size);
3380         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
3381                    buf->ipv6_extension_header);
3382         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
3383                    buf->tcp_pseudosum_req);
3384         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
3385                    buf->num_ipv6_ext_headers);
3386         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
3387                    buf->off_ipv6_ext_headers);
3388
3389         adapter->ip_offload_ctrl_tok =
3390             dma_map_single(dev, &adapter->ip_offload_ctrl,
3391                            sizeof(adapter->ip_offload_ctrl), DMA_TO_DEVICE);
3392
3393         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
3394                 dev_err(dev, "Couldn't map ip offload control buffer\n");
3395                 return;
3396         }
3397
3398         adapter->ip_offload_ctrl.len =
3399             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3400         adapter->ip_offload_ctrl.version = cpu_to_be32(INITIAL_VERSION_IOB);
3401         adapter->ip_offload_ctrl.ipv4_chksum = buf->ipv4_chksum;
3402         adapter->ip_offload_ctrl.ipv6_chksum = buf->ipv6_chksum;
3403         adapter->ip_offload_ctrl.tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
3404         adapter->ip_offload_ctrl.udp_ipv4_chksum = buf->udp_ipv4_chksum;
3405         adapter->ip_offload_ctrl.tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
3406         adapter->ip_offload_ctrl.udp_ipv6_chksum = buf->udp_ipv6_chksum;
3407         adapter->ip_offload_ctrl.large_tx_ipv4 = buf->large_tx_ipv4;
3408         adapter->ip_offload_ctrl.large_tx_ipv6 = buf->large_tx_ipv6;
3409
3410         /* large_rx disabled for now, additional features needed */
3411         adapter->ip_offload_ctrl.large_rx_ipv4 = 0;
3412         adapter->ip_offload_ctrl.large_rx_ipv6 = 0;
3413
3414         adapter->netdev->features = NETIF_F_SG | NETIF_F_GSO;
3415
3416         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
3417                 adapter->netdev->features |= NETIF_F_IP_CSUM;
3418
3419         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
3420                 adapter->netdev->features |= NETIF_F_IPV6_CSUM;
3421
3422         if ((adapter->netdev->features &
3423             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
3424                 adapter->netdev->features |= NETIF_F_RXCSUM;
3425
3426         if (buf->large_tx_ipv4)
3427                 adapter->netdev->features |= NETIF_F_TSO;
3428         if (buf->large_tx_ipv6)
3429                 adapter->netdev->features |= NETIF_F_TSO6;
3430
3431         adapter->netdev->hw_features |= adapter->netdev->features;
3432
3433         memset(&crq, 0, sizeof(crq));
3434         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
3435         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
3436         crq.control_ip_offload.len =
3437             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3438         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
3439         ibmvnic_send_crq(adapter, &crq);
3440 }
3441
3442 static void handle_error_info_rsp(union ibmvnic_crq *crq,
3443                                   struct ibmvnic_adapter *adapter)
3444 {
3445         struct device *dev = &adapter->vdev->dev;
3446         struct ibmvnic_error_buff *error_buff, *tmp;
3447         unsigned long flags;
3448         bool found = false;
3449         int i;
3450
3451         if (!crq->request_error_rsp.rc.code) {
3452                 dev_info(dev, "Request Error Rsp returned with rc=%x\n",
3453                          crq->request_error_rsp.rc.code);
3454                 return;
3455         }
3456
3457         spin_lock_irqsave(&adapter->error_list_lock, flags);
3458         list_for_each_entry_safe(error_buff, tmp, &adapter->errors, list)
3459                 if (error_buff->error_id == crq->request_error_rsp.error_id) {
3460                         found = true;
3461                         list_del(&error_buff->list);
3462                         break;
3463                 }
3464         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
3465
3466         if (!found) {
3467                 dev_err(dev, "Couldn't find error id %x\n",
3468                         be32_to_cpu(crq->request_error_rsp.error_id));
3469                 return;
3470         }
3471
3472         dev_err(dev, "Detailed info for error id %x:",
3473                 be32_to_cpu(crq->request_error_rsp.error_id));
3474
3475         for (i = 0; i < error_buff->len; i++) {
3476                 pr_cont("%02x", (int)error_buff->buff[i]);
3477                 if (i % 8 == 7)
3478                         pr_cont(" ");
3479         }
3480         pr_cont("\n");
3481
3482         dma_unmap_single(dev, error_buff->dma, error_buff->len,
3483                          DMA_FROM_DEVICE);
3484         kfree(error_buff->buff);
3485         kfree(error_buff);
3486 }
3487
3488 static void request_error_information(struct ibmvnic_adapter *adapter,
3489                                       union ibmvnic_crq *err_crq)
3490 {
3491         struct device *dev = &adapter->vdev->dev;
3492         struct net_device *netdev = adapter->netdev;
3493         struct ibmvnic_error_buff *error_buff;
3494         unsigned long timeout = msecs_to_jiffies(30000);
3495         union ibmvnic_crq crq;
3496         unsigned long flags;
3497         int rc, detail_len;
3498
3499         error_buff = kmalloc(sizeof(*error_buff), GFP_ATOMIC);
3500         if (!error_buff)
3501                 return;
3502
3503         detail_len = be32_to_cpu(err_crq->error_indication.detail_error_sz);
3504         error_buff->buff = kmalloc(detail_len, GFP_ATOMIC);
3505         if (!error_buff->buff) {
3506                 kfree(error_buff);
3507                 return;
3508         }
3509
3510         error_buff->dma = dma_map_single(dev, error_buff->buff, detail_len,
3511                                          DMA_FROM_DEVICE);
3512         if (dma_mapping_error(dev, error_buff->dma)) {
3513                 netdev_err(netdev, "Couldn't map error buffer\n");
3514                 kfree(error_buff->buff);
3515                 kfree(error_buff);
3516                 return;
3517         }
3518
3519         error_buff->len = detail_len;
3520         error_buff->error_id = err_crq->error_indication.error_id;
3521
3522         spin_lock_irqsave(&adapter->error_list_lock, flags);
3523         list_add_tail(&error_buff->list, &adapter->errors);
3524         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
3525
3526         memset(&crq, 0, sizeof(crq));
3527         crq.request_error_info.first = IBMVNIC_CRQ_CMD;
3528         crq.request_error_info.cmd = REQUEST_ERROR_INFO;
3529         crq.request_error_info.ioba = cpu_to_be32(error_buff->dma);
3530         crq.request_error_info.len = cpu_to_be32(detail_len);
3531         crq.request_error_info.error_id = err_crq->error_indication.error_id;
3532
3533         rc = ibmvnic_send_crq(adapter, &crq);
3534         if (rc) {
3535                 netdev_err(netdev, "failed to request error information\n");
3536                 goto err_info_fail;
3537         }
3538
3539         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3540                 netdev_err(netdev, "timeout waiting for error information\n");
3541                 goto err_info_fail;
3542         }
3543
3544         return;
3545
3546 err_info_fail:
3547         spin_lock_irqsave(&adapter->error_list_lock, flags);
3548         list_del(&error_buff->list);
3549         spin_unlock_irqrestore(&adapter->error_list_lock, flags);
3550
3551         kfree(error_buff->buff);
3552         kfree(error_buff);
3553 }
3554
3555 static void handle_error_indication(union ibmvnic_crq *crq,
3556                                     struct ibmvnic_adapter *adapter)
3557 {
3558         struct device *dev = &adapter->vdev->dev;
3559
3560         dev_err(dev, "Firmware reports %serror id %x, cause %d\n",
3561                 crq->error_indication.flags
3562                         & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
3563                 be32_to_cpu(crq->error_indication.error_id),
3564                 be16_to_cpu(crq->error_indication.error_cause));
3565
3566         if (be32_to_cpu(crq->error_indication.error_id))
3567                 request_error_information(adapter, crq);
3568
3569         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
3570                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3571         else
3572                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
3573 }
3574
3575 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
3576                                  struct ibmvnic_adapter *adapter)
3577 {
3578         struct net_device *netdev = adapter->netdev;
3579         struct device *dev = &adapter->vdev->dev;
3580         long rc;
3581
3582         rc = crq->change_mac_addr_rsp.rc.code;
3583         if (rc) {
3584                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
3585                 goto out;
3586         }
3587         memcpy(netdev->dev_addr, &crq->change_mac_addr_rsp.mac_addr[0],
3588                ETH_ALEN);
3589 out:
3590         complete(&adapter->fw_done);
3591         return rc;
3592 }
3593
3594 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
3595                                    struct ibmvnic_adapter *adapter)
3596 {
3597         struct device *dev = &adapter->vdev->dev;
3598         u64 *req_value;
3599         char *name;
3600
3601         atomic_dec(&adapter->running_cap_crqs);
3602         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
3603         case REQ_TX_QUEUES:
3604                 req_value = &adapter->req_tx_queues;
3605                 name = "tx";
3606                 break;
3607         case REQ_RX_QUEUES:
3608                 req_value = &adapter->req_rx_queues;
3609                 name = "rx";
3610                 break;
3611         case REQ_RX_ADD_QUEUES:
3612                 req_value = &adapter->req_rx_add_queues;
3613                 name = "rx_add";
3614                 break;
3615         case REQ_TX_ENTRIES_PER_SUBCRQ:
3616                 req_value = &adapter->req_tx_entries_per_subcrq;
3617                 name = "tx_entries_per_subcrq";
3618                 break;
3619         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
3620                 req_value = &adapter->req_rx_add_entries_per_subcrq;
3621                 name = "rx_add_entries_per_subcrq";
3622                 break;
3623         case REQ_MTU:
3624                 req_value = &adapter->req_mtu;
3625                 name = "mtu";
3626                 break;
3627         case PROMISC_REQUESTED:
3628                 req_value = &adapter->promisc;
3629                 name = "promisc";
3630                 break;
3631         default:
3632                 dev_err(dev, "Got invalid cap request rsp %d\n",
3633                         crq->request_capability.capability);
3634                 return;
3635         }
3636
3637         switch (crq->request_capability_rsp.rc.code) {
3638         case SUCCESS:
3639                 break;
3640         case PARTIALSUCCESS:
3641                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
3642                          *req_value,
3643                          (long int)be64_to_cpu(crq->request_capability_rsp.
3644                                                number), name);
3645
3646                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
3647                     REQ_MTU) {
3648                         pr_err("mtu of %llu is not supported. Reverting.\n",
3649                                *req_value);
3650                         *req_value = adapter->fallback.mtu;
3651                 } else {
3652                         *req_value =
3653                                 be64_to_cpu(crq->request_capability_rsp.number);
3654                 }
3655
3656                 ibmvnic_send_req_caps(adapter, 1);
3657                 return;
3658         default:
3659                 dev_err(dev, "Error %d in request cap rsp\n",
3660                         crq->request_capability_rsp.rc.code);
3661                 return;
3662         }
3663
3664         /* Done receiving requested capabilities, query IP offload support */
3665         if (atomic_read(&adapter->running_cap_crqs) == 0) {
3666                 union ibmvnic_crq newcrq;
3667                 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
3668                 struct ibmvnic_query_ip_offload_buffer *ip_offload_buf =
3669                     &adapter->ip_offload_buf;
3670
3671                 adapter->wait_capability = false;
3672                 adapter->ip_offload_tok = dma_map_single(dev, ip_offload_buf,
3673                                                          buf_sz,
3674                                                          DMA_FROM_DEVICE);
3675
3676                 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
3677                         if (!firmware_has_feature(FW_FEATURE_CMO))
3678                                 dev_err(dev, "Couldn't map offload buffer\n");
3679                         return;
3680                 }
3681
3682                 memset(&newcrq, 0, sizeof(newcrq));
3683                 newcrq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
3684                 newcrq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
3685                 newcrq.query_ip_offload.len = cpu_to_be32(buf_sz);
3686                 newcrq.query_ip_offload.ioba =
3687                     cpu_to_be32(adapter->ip_offload_tok);
3688
3689                 ibmvnic_send_crq(adapter, &newcrq);
3690         }
3691 }
3692
3693 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
3694                             struct ibmvnic_adapter *adapter)
3695 {
3696         struct device *dev = &adapter->vdev->dev;
3697         struct net_device *netdev = adapter->netdev;
3698         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
3699         struct ibmvnic_login_buffer *login = adapter->login_buf;
3700         int i;
3701
3702         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
3703                          DMA_BIDIRECTIONAL);
3704         dma_unmap_single(dev, adapter->login_rsp_buf_token,
3705                          adapter->login_rsp_buf_sz, DMA_BIDIRECTIONAL);
3706
3707         /* If the number of queues requested can't be allocated by the
3708          * server, the login response will return with code 1. We will need
3709          * to resend the login buffer with fewer queues requested.
3710          */
3711         if (login_rsp_crq->generic.rc.code) {
3712                 adapter->renegotiate = true;
3713                 complete(&adapter->init_done);
3714                 return 0;
3715         }
3716
3717         netdev->mtu = adapter->req_mtu - ETH_HLEN;
3718
3719         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
3720         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
3721                 netdev_dbg(adapter->netdev, "%016lx\n",
3722                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
3723         }
3724
3725         /* Sanity checks */
3726         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
3727             (be32_to_cpu(login->num_rxcomp_subcrqs) *
3728              adapter->req_rx_add_queues !=
3729              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
3730                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
3731                 ibmvnic_remove(adapter->vdev);
3732                 return -EIO;
3733         }
3734         complete(&adapter->init_done);
3735
3736         return 0;
3737 }
3738
3739 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
3740                                      struct ibmvnic_adapter *adapter)
3741 {
3742         struct device *dev = &adapter->vdev->dev;
3743         long rc;
3744
3745         rc = crq->request_unmap_rsp.rc.code;
3746         if (rc)
3747                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
3748 }
3749
3750 static void handle_query_map_rsp(union ibmvnic_crq *crq,
3751                                  struct ibmvnic_adapter *adapter)
3752 {
3753         struct net_device *netdev = adapter->netdev;
3754         struct device *dev = &adapter->vdev->dev;
3755         long rc;
3756
3757         rc = crq->query_map_rsp.rc.code;
3758         if (rc) {
3759                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
3760                 return;
3761         }
3762         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
3763                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
3764                    crq->query_map_rsp.free_pages);
3765 }
3766
3767 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
3768                                  struct ibmvnic_adapter *adapter)
3769 {
3770         struct net_device *netdev = adapter->netdev;
3771         struct device *dev = &adapter->vdev->dev;
3772         long rc;
3773
3774         atomic_dec(&adapter->running_cap_crqs);
3775         netdev_dbg(netdev, "Outstanding queries: %d\n",
3776                    atomic_read(&adapter->running_cap_crqs));
3777         rc = crq->query_capability.rc.code;
3778         if (rc) {
3779                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
3780                 goto out;
3781         }
3782
3783         switch (be16_to_cpu(crq->query_capability.capability)) {
3784         case MIN_TX_QUEUES:
3785                 adapter->min_tx_queues =
3786                     be64_to_cpu(crq->query_capability.number);
3787                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
3788                            adapter->min_tx_queues);
3789                 break;
3790         case MIN_RX_QUEUES:
3791                 adapter->min_rx_queues =
3792                     be64_to_cpu(crq->query_capability.number);
3793                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
3794                            adapter->min_rx_queues);
3795                 break;
3796         case MIN_RX_ADD_QUEUES:
3797                 adapter->min_rx_add_queues =
3798                     be64_to_cpu(crq->query_capability.number);
3799                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
3800                            adapter->min_rx_add_queues);
3801                 break;
3802         case MAX_TX_QUEUES:
3803                 adapter->max_tx_queues =
3804                     be64_to_cpu(crq->query_capability.number);
3805                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
3806                            adapter->max_tx_queues);
3807                 break;
3808         case MAX_RX_QUEUES:
3809                 adapter->max_rx_queues =
3810                     be64_to_cpu(crq->query_capability.number);
3811                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
3812                            adapter->max_rx_queues);
3813                 break;
3814         case MAX_RX_ADD_QUEUES:
3815                 adapter->max_rx_add_queues =
3816                     be64_to_cpu(crq->query_capability.number);
3817                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
3818                            adapter->max_rx_add_queues);
3819                 break;
3820         case MIN_TX_ENTRIES_PER_SUBCRQ:
3821                 adapter->min_tx_entries_per_subcrq =
3822                     be64_to_cpu(crq->query_capability.number);
3823                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
3824                            adapter->min_tx_entries_per_subcrq);
3825                 break;
3826         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
3827                 adapter->min_rx_add_entries_per_subcrq =
3828                     be64_to_cpu(crq->query_capability.number);
3829                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
3830                            adapter->min_rx_add_entries_per_subcrq);
3831                 break;
3832         case MAX_TX_ENTRIES_PER_SUBCRQ:
3833                 adapter->max_tx_entries_per_subcrq =
3834                     be64_to_cpu(crq->query_capability.number);
3835                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
3836                            adapter->max_tx_entries_per_subcrq);
3837                 break;
3838         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
3839                 adapter->max_rx_add_entries_per_subcrq =
3840                     be64_to_cpu(crq->query_capability.number);
3841                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
3842                            adapter->max_rx_add_entries_per_subcrq);
3843                 break;
3844         case TCP_IP_OFFLOAD:
3845                 adapter->tcp_ip_offload =
3846                     be64_to_cpu(crq->query_capability.number);
3847                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
3848                            adapter->tcp_ip_offload);
3849                 break;
3850         case PROMISC_SUPPORTED:
3851                 adapter->promisc_supported =
3852                     be64_to_cpu(crq->query_capability.number);
3853                 netdev_dbg(netdev, "promisc_supported = %lld\n",
3854                            adapter->promisc_supported);
3855                 break;
3856         case MIN_MTU:
3857                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
3858                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3859                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
3860                 break;
3861         case MAX_MTU:
3862                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
3863                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3864                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
3865                 break;
3866         case MAX_MULTICAST_FILTERS:
3867                 adapter->max_multicast_filters =
3868                     be64_to_cpu(crq->query_capability.number);
3869                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
3870                            adapter->max_multicast_filters);
3871                 break;
3872         case VLAN_HEADER_INSERTION:
3873                 adapter->vlan_header_insertion =
3874                     be64_to_cpu(crq->query_capability.number);
3875                 if (adapter->vlan_header_insertion)
3876                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
3877                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
3878                            adapter->vlan_header_insertion);
3879                 break;
3880         case RX_VLAN_HEADER_INSERTION:
3881                 adapter->rx_vlan_header_insertion =
3882                     be64_to_cpu(crq->query_capability.number);
3883                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
3884                            adapter->rx_vlan_header_insertion);
3885                 break;
3886         case MAX_TX_SG_ENTRIES:
3887                 adapter->max_tx_sg_entries =
3888                     be64_to_cpu(crq->query_capability.number);
3889                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
3890                            adapter->max_tx_sg_entries);
3891                 break;
3892         case RX_SG_SUPPORTED:
3893                 adapter->rx_sg_supported =
3894                     be64_to_cpu(crq->query_capability.number);
3895                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
3896                            adapter->rx_sg_supported);
3897                 break;
3898         case OPT_TX_COMP_SUB_QUEUES:
3899                 adapter->opt_tx_comp_sub_queues =
3900                     be64_to_cpu(crq->query_capability.number);
3901                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
3902                            adapter->opt_tx_comp_sub_queues);
3903                 break;
3904         case OPT_RX_COMP_QUEUES:
3905                 adapter->opt_rx_comp_queues =
3906                     be64_to_cpu(crq->query_capability.number);
3907                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
3908                            adapter->opt_rx_comp_queues);
3909                 break;
3910         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
3911                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
3912                     be64_to_cpu(crq->query_capability.number);
3913                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
3914                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
3915                 break;
3916         case OPT_TX_ENTRIES_PER_SUBCRQ:
3917                 adapter->opt_tx_entries_per_subcrq =
3918                     be64_to_cpu(crq->query_capability.number);
3919                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
3920                            adapter->opt_tx_entries_per_subcrq);
3921                 break;
3922         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
3923                 adapter->opt_rxba_entries_per_subcrq =
3924                     be64_to_cpu(crq->query_capability.number);
3925                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
3926                            adapter->opt_rxba_entries_per_subcrq);
3927                 break;
3928         case TX_RX_DESC_REQ:
3929                 adapter->tx_rx_desc_req = crq->query_capability.number;
3930                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
3931                            adapter->tx_rx_desc_req);
3932                 break;
3933
3934         default:
3935                 netdev_err(netdev, "Got invalid cap rsp %d\n",
3936                            crq->query_capability.capability);
3937         }
3938
3939 out:
3940         if (atomic_read(&adapter->running_cap_crqs) == 0) {
3941                 adapter->wait_capability = false;
3942                 ibmvnic_send_req_caps(adapter, 0);
3943         }
3944 }
3945
3946 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
3947                                struct ibmvnic_adapter *adapter)
3948 {
3949         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
3950         struct net_device *netdev = adapter->netdev;
3951         struct device *dev = &adapter->vdev->dev;
3952         u64 *u64_crq = (u64 *)crq;
3953         long rc;
3954
3955         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
3956                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3957                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3958         switch (gen_crq->first) {
3959         case IBMVNIC_CRQ_INIT_RSP:
3960                 switch (gen_crq->cmd) {
3961                 case IBMVNIC_CRQ_INIT:
3962                         dev_info(dev, "Partner initialized\n");
3963                         adapter->from_passive_init = true;
3964                         complete(&adapter->init_done);
3965                         break;
3966                 case IBMVNIC_CRQ_INIT_COMPLETE:
3967                         dev_info(dev, "Partner initialization complete\n");
3968                         send_version_xchg(adapter);
3969                         break;
3970                 default:
3971                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
3972                 }
3973                 return;
3974         case IBMVNIC_CRQ_XPORT_EVENT:
3975                 netif_carrier_off(netdev);
3976                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
3977                         dev_info(dev, "Migrated, re-enabling adapter\n");
3978                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
3979                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
3980                         dev_info(dev, "Backing device failover detected\n");
3981                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
3982                 } else {
3983                         /* The adapter lost the connection */
3984                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
3985                                 gen_crq->cmd);
3986                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3987                 }
3988                 return;
3989         case IBMVNIC_CRQ_CMD_RSP:
3990                 break;
3991         default:
3992                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
3993                         gen_crq->first);
3994                 return;
3995         }
3996
3997         switch (gen_crq->cmd) {
3998         case VERSION_EXCHANGE_RSP:
3999                 rc = crq->version_exchange_rsp.rc.code;
4000                 if (rc) {
4001                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4002                         break;
4003                 }
4004                 dev_info(dev, "Partner protocol version is %d\n",
4005                          crq->version_exchange_rsp.version);
4006                 if (be16_to_cpu(crq->version_exchange_rsp.version) <
4007                     ibmvnic_version)
4008                         ibmvnic_version =
4009                             be16_to_cpu(crq->version_exchange_rsp.version);
4010                 send_cap_queries(adapter);
4011                 break;
4012         case QUERY_CAPABILITY_RSP:
4013                 handle_query_cap_rsp(crq, adapter);
4014                 break;
4015         case QUERY_MAP_RSP:
4016                 handle_query_map_rsp(crq, adapter);
4017                 break;
4018         case REQUEST_MAP_RSP:
4019                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4020                 complete(&adapter->fw_done);
4021                 break;
4022         case REQUEST_UNMAP_RSP:
4023                 handle_request_unmap_rsp(crq, adapter);
4024                 break;
4025         case REQUEST_CAPABILITY_RSP:
4026                 handle_request_cap_rsp(crq, adapter);
4027                 break;
4028         case LOGIN_RSP:
4029                 netdev_dbg(netdev, "Got Login Response\n");
4030                 handle_login_rsp(crq, adapter);
4031                 break;
4032         case LOGICAL_LINK_STATE_RSP:
4033                 netdev_dbg(netdev,
4034                            "Got Logical Link State Response, state: %d rc: %d\n",
4035                            crq->logical_link_state_rsp.link_state,
4036                            crq->logical_link_state_rsp.rc.code);
4037                 adapter->logical_link_state =
4038                     crq->logical_link_state_rsp.link_state;
4039                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4040                 complete(&adapter->init_done);
4041                 break;
4042         case LINK_STATE_INDICATION:
4043                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4044                 adapter->phys_link_state =
4045                     crq->link_state_indication.phys_link_state;
4046                 adapter->logical_link_state =
4047                     crq->link_state_indication.logical_link_state;
4048                 break;
4049         case CHANGE_MAC_ADDR_RSP:
4050                 netdev_dbg(netdev, "Got MAC address change Response\n");
4051                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4052                 break;
4053         case ERROR_INDICATION:
4054                 netdev_dbg(netdev, "Got Error Indication\n");
4055                 handle_error_indication(crq, adapter);
4056                 break;
4057         case REQUEST_ERROR_RSP:
4058                 netdev_dbg(netdev, "Got Error Detail Response\n");
4059                 handle_error_info_rsp(crq, adapter);
4060                 break;
4061         case REQUEST_STATISTICS_RSP:
4062                 netdev_dbg(netdev, "Got Statistics Response\n");
4063                 complete(&adapter->stats_done);
4064                 break;
4065         case QUERY_IP_OFFLOAD_RSP:
4066                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4067                 handle_query_ip_offload_rsp(adapter);
4068                 break;
4069         case MULTICAST_CTRL_RSP:
4070                 netdev_dbg(netdev, "Got multicast control Response\n");
4071                 break;
4072         case CONTROL_IP_OFFLOAD_RSP:
4073                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4074                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4075                                  sizeof(adapter->ip_offload_ctrl),
4076                                  DMA_TO_DEVICE);
4077                 complete(&adapter->init_done);
4078                 break;
4079         case COLLECT_FW_TRACE_RSP:
4080                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4081                 complete(&adapter->fw_done);
4082                 break;
4083         case GET_VPD_SIZE_RSP:
4084                 handle_vpd_size_rsp(crq, adapter);
4085                 break;
4086         case GET_VPD_RSP:
4087                 handle_vpd_rsp(crq, adapter);
4088                 break;
4089         default:
4090                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4091                            gen_crq->cmd);
4092         }
4093 }
4094
4095 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
4096 {
4097         struct ibmvnic_adapter *adapter = instance;
4098
4099         tasklet_schedule(&adapter->tasklet);
4100         return IRQ_HANDLED;
4101 }
4102
4103 static void ibmvnic_tasklet(void *data)
4104 {
4105         struct ibmvnic_adapter *adapter = data;
4106         struct ibmvnic_crq_queue *queue = &adapter->crq;
4107         union ibmvnic_crq *crq;
4108         unsigned long flags;
4109         bool done = false;
4110
4111         spin_lock_irqsave(&queue->lock, flags);
4112         while (!done) {
4113                 /* Pull all the valid messages off the CRQ */
4114                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
4115                         ibmvnic_handle_crq(crq, adapter);
4116                         crq->generic.first = 0;
4117                 }
4118
4119                 /* remain in tasklet until all
4120                  * capabilities responses are received
4121                  */
4122                 if (!adapter->wait_capability)
4123                         done = true;
4124         }
4125         /* if capabilities CRQ's were sent in this tasklet, the following
4126          * tasklet must wait until all responses are received
4127          */
4128         if (atomic_read(&adapter->running_cap_crqs) != 0)
4129                 adapter->wait_capability = true;
4130         spin_unlock_irqrestore(&queue->lock, flags);
4131 }
4132
4133 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
4134 {
4135         struct vio_dev *vdev = adapter->vdev;
4136         int rc;
4137
4138         do {
4139                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
4140         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
4141
4142         if (rc)
4143                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
4144
4145         return rc;
4146 }
4147
4148 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
4149 {
4150         struct ibmvnic_crq_queue *crq = &adapter->crq;
4151         struct device *dev = &adapter->vdev->dev;
4152         struct vio_dev *vdev = adapter->vdev;
4153         int rc;
4154
4155         /* Close the CRQ */
4156         do {
4157                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4158         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4159
4160         /* Clean out the queue */
4161         memset(crq->msgs, 0, PAGE_SIZE);
4162         crq->cur = 0;
4163
4164         /* And re-open it again */
4165         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4166                                 crq->msg_token, PAGE_SIZE);
4167
4168         if (rc == H_CLOSED)
4169                 /* Adapter is good, but other end is not ready */
4170                 dev_warn(dev, "Partner adapter not ready\n");
4171         else if (rc != 0)
4172                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
4173
4174         return rc;
4175 }
4176
4177 static void release_crq_queue(struct ibmvnic_adapter *adapter)
4178 {
4179         struct ibmvnic_crq_queue *crq = &adapter->crq;
4180         struct vio_dev *vdev = adapter->vdev;
4181         long rc;
4182
4183         if (!crq->msgs)
4184                 return;
4185
4186         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
4187         free_irq(vdev->irq, adapter);
4188         tasklet_kill(&adapter->tasklet);
4189         do {
4190                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4191         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4192
4193         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
4194                          DMA_BIDIRECTIONAL);
4195         free_page((unsigned long)crq->msgs);
4196         crq->msgs = NULL;
4197 }
4198
4199 static int init_crq_queue(struct ibmvnic_adapter *adapter)
4200 {
4201         struct ibmvnic_crq_queue *crq = &adapter->crq;
4202         struct device *dev = &adapter->vdev->dev;
4203         struct vio_dev *vdev = adapter->vdev;
4204         int rc, retrc = -ENOMEM;
4205
4206         if (crq->msgs)
4207                 return 0;
4208
4209         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
4210         /* Should we allocate more than one page? */
4211
4212         if (!crq->msgs)
4213                 return -ENOMEM;
4214
4215         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
4216         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
4217                                         DMA_BIDIRECTIONAL);
4218         if (dma_mapping_error(dev, crq->msg_token))
4219                 goto map_failed;
4220
4221         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4222                                 crq->msg_token, PAGE_SIZE);
4223
4224         if (rc == H_RESOURCE)
4225                 /* maybe kexecing and resource is busy. try a reset */
4226                 rc = ibmvnic_reset_crq(adapter);
4227         retrc = rc;
4228
4229         if (rc == H_CLOSED) {
4230                 dev_warn(dev, "Partner adapter not ready\n");
4231         } else if (rc) {
4232                 dev_warn(dev, "Error %d opening adapter\n", rc);
4233                 goto reg_crq_failed;
4234         }
4235
4236         retrc = 0;
4237
4238         tasklet_init(&adapter->tasklet, (void *)ibmvnic_tasklet,
4239                      (unsigned long)adapter);
4240
4241         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
4242         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, IBMVNIC_NAME,
4243                          adapter);
4244         if (rc) {
4245                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
4246                         vdev->irq, rc);
4247                 goto req_irq_failed;
4248         }
4249
4250         rc = vio_enable_interrupts(vdev);
4251         if (rc) {
4252                 dev_err(dev, "Error %d enabling interrupts\n", rc);
4253                 goto req_irq_failed;
4254         }
4255
4256         crq->cur = 0;
4257         spin_lock_init(&crq->lock);
4258
4259         return retrc;
4260
4261 req_irq_failed:
4262         tasklet_kill(&adapter->tasklet);
4263         do {
4264                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4265         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4266 reg_crq_failed:
4267         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
4268 map_failed:
4269         free_page((unsigned long)crq->msgs);
4270         crq->msgs = NULL;
4271         return retrc;
4272 }
4273
4274 static int ibmvnic_init(struct ibmvnic_adapter *adapter)
4275 {
4276         struct device *dev = &adapter->vdev->dev;
4277         unsigned long timeout = msecs_to_jiffies(30000);
4278         int rc;
4279
4280         if (adapter->resetting && !adapter->wait_for_reset) {
4281                 rc = ibmvnic_reset_crq(adapter);
4282                 if (!rc)
4283                         rc = vio_enable_interrupts(adapter->vdev);
4284         } else {
4285                 rc = init_crq_queue(adapter);
4286         }
4287
4288         if (rc) {
4289                 dev_err(dev, "Couldn't initialize crq. rc=%d\n", rc);
4290                 return rc;
4291         }
4292
4293         adapter->from_passive_init = false;
4294
4295         init_completion(&adapter->init_done);
4296         adapter->init_done_rc = 0;
4297         ibmvnic_send_crq_init(adapter);
4298         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
4299                 dev_err(dev, "Initialization sequence timed out\n");
4300                 return -1;
4301         }
4302
4303         if (adapter->init_done_rc) {
4304                 release_crq_queue(adapter);
4305                 return adapter->init_done_rc;
4306         }
4307
4308         if (adapter->from_passive_init) {
4309                 adapter->state = VNIC_OPEN;
4310                 adapter->from_passive_init = false;
4311                 return -1;
4312         }
4313
4314         if (adapter->resetting && !adapter->wait_for_reset)
4315                 rc = reset_sub_crq_queues(adapter);
4316         else
4317                 rc = init_sub_crqs(adapter);
4318         if (rc) {
4319                 dev_err(dev, "Initialization of sub crqs failed\n");
4320                 release_crq_queue(adapter);
4321                 return rc;
4322         }
4323
4324         rc = init_sub_crq_irqs(adapter);
4325         if (rc) {
4326                 dev_err(dev, "Failed to initialize sub crq irqs\n");
4327                 release_crq_queue(adapter);
4328         }
4329
4330         return rc;
4331 }
4332
4333 static struct device_attribute dev_attr_failover;
4334
4335 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
4336 {
4337         struct ibmvnic_adapter *adapter;
4338         struct net_device *netdev;
4339         unsigned char *mac_addr_p;
4340         int rc;
4341
4342         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
4343                 dev->unit_address);
4344
4345         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
4346                                                         VETH_MAC_ADDR, NULL);
4347         if (!mac_addr_p) {
4348                 dev_err(&dev->dev,
4349                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
4350                         __FILE__, __LINE__);
4351                 return 0;
4352         }
4353
4354         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
4355                                    IBMVNIC_MAX_QUEUES);
4356         if (!netdev)
4357                 return -ENOMEM;
4358
4359         adapter = netdev_priv(netdev);
4360         adapter->state = VNIC_PROBING;
4361         dev_set_drvdata(&dev->dev, netdev);
4362         adapter->vdev = dev;
4363         adapter->netdev = netdev;
4364
4365         ether_addr_copy(adapter->mac_addr, mac_addr_p);
4366         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4367         netdev->irq = dev->irq;
4368         netdev->netdev_ops = &ibmvnic_netdev_ops;
4369         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
4370         SET_NETDEV_DEV(netdev, &dev->dev);
4371
4372         spin_lock_init(&adapter->stats_lock);
4373
4374         INIT_LIST_HEAD(&adapter->errors);
4375         spin_lock_init(&adapter->error_list_lock);
4376
4377         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
4378         INIT_LIST_HEAD(&adapter->rwi_list);
4379         mutex_init(&adapter->reset_lock);
4380         mutex_init(&adapter->rwi_lock);
4381         adapter->resetting = false;
4382
4383         adapter->mac_change_pending = false;
4384
4385         do {
4386                 rc = ibmvnic_init(adapter);
4387                 if (rc && rc != EAGAIN)
4388                         goto ibmvnic_init_fail;
4389         } while (rc == EAGAIN);
4390
4391         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4392         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4393         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4394
4395         rc = device_create_file(&dev->dev, &dev_attr_failover);
4396         if (rc)
4397                 goto ibmvnic_init_fail;
4398
4399         netif_carrier_off(netdev);
4400         rc = register_netdev(netdev);
4401         if (rc) {
4402                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
4403                 goto ibmvnic_register_fail;
4404         }
4405         dev_info(&dev->dev, "ibmvnic registered\n");
4406
4407         adapter->state = VNIC_PROBED;
4408
4409         adapter->wait_for_reset = false;
4410
4411         return 0;
4412
4413 ibmvnic_register_fail:
4414         device_remove_file(&dev->dev, &dev_attr_failover);
4415
4416 ibmvnic_init_fail:
4417         release_sub_crqs(adapter);
4418         release_crq_queue(adapter);
4419         free_netdev(netdev);
4420
4421         return rc;
4422 }
4423
4424 static int ibmvnic_remove(struct vio_dev *dev)
4425 {
4426         struct net_device *netdev = dev_get_drvdata(&dev->dev);
4427         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4428
4429         adapter->state = VNIC_REMOVING;
4430         unregister_netdev(netdev);
4431         mutex_lock(&adapter->reset_lock);
4432
4433         release_resources(adapter);
4434         release_sub_crqs(adapter);
4435         release_crq_queue(adapter);
4436
4437         adapter->state = VNIC_REMOVED;
4438
4439         mutex_unlock(&adapter->reset_lock);
4440         device_remove_file(&dev->dev, &dev_attr_failover);
4441         free_netdev(netdev);
4442         dev_set_drvdata(&dev->dev, NULL);
4443
4444         return 0;
4445 }
4446
4447 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
4448                               const char *buf, size_t count)
4449 {
4450         struct net_device *netdev = dev_get_drvdata(dev);
4451         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4452         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
4453         __be64 session_token;
4454         long rc;
4455
4456         if (!sysfs_streq(buf, "1"))
4457                 return -EINVAL;
4458
4459         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
4460                          H_GET_SESSION_TOKEN, 0, 0, 0);
4461         if (rc) {
4462                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
4463                            rc);
4464                 return -EINVAL;
4465         }
4466
4467         session_token = (__be64)retbuf[0];
4468         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
4469                    be64_to_cpu(session_token));
4470         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4471                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
4472         if (rc) {
4473                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
4474                            rc);
4475                 return -EINVAL;
4476         }
4477
4478         return count;
4479 }
4480
4481 static DEVICE_ATTR_WO(failover);
4482
4483 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
4484 {
4485         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
4486         struct ibmvnic_adapter *adapter;
4487         struct iommu_table *tbl;
4488         unsigned long ret = 0;
4489         int i;
4490
4491         tbl = get_iommu_table_base(&vdev->dev);
4492
4493         /* netdev inits at probe time along with the structures we need below*/
4494         if (!netdev)
4495                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
4496
4497         adapter = netdev_priv(netdev);
4498
4499         ret += PAGE_SIZE; /* the crq message queue */
4500         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
4501
4502         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
4503                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
4504
4505         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4506              i++)
4507                 ret += adapter->rx_pool[i].size *
4508                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
4509
4510         return ret;
4511 }
4512
4513 static int ibmvnic_resume(struct device *dev)
4514 {
4515         struct net_device *netdev = dev_get_drvdata(dev);
4516         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4517
4518         if (adapter->state != VNIC_OPEN)
4519                 return 0;
4520
4521         tasklet_schedule(&adapter->tasklet);
4522
4523         return 0;
4524 }
4525
4526 static const struct vio_device_id ibmvnic_device_table[] = {
4527         {"network", "IBM,vnic"},
4528         {"", "" }
4529 };
4530 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
4531
4532 static const struct dev_pm_ops ibmvnic_pm_ops = {
4533         .resume = ibmvnic_resume
4534 };
4535
4536 static struct vio_driver ibmvnic_driver = {
4537         .id_table       = ibmvnic_device_table,
4538         .probe          = ibmvnic_probe,
4539         .remove         = ibmvnic_remove,
4540         .get_desired_dma = ibmvnic_get_desired_dma,
4541         .name           = ibmvnic_driver_name,
4542         .pm             = &ibmvnic_pm_ops,
4543 };
4544
4545 /* module functions */
4546 static int __init ibmvnic_module_init(void)
4547 {
4548         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
4549                 IBMVNIC_DRIVER_VERSION);
4550
4551         return vio_register_driver(&ibmvnic_driver);
4552 }
4553
4554 static void __exit ibmvnic_module_exit(void)
4555 {
4556         vio_unregister_driver(&ibmvnic_driver);
4557 }
4558
4559 module_init(ibmvnic_module_init);
4560 module_exit(ibmvnic_module_exit);