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 *, bool);
  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 int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
 113 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 114 static int 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 int ibmvnic_reset_init(struct ibmvnic_adapter *);
 120 static void release_crq_queue(struct ibmvnic_adapter *);
 121 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p);
 122 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 123
 124 struct ibmvnic_stat {
 125         char name[ETH_GSTRING_LEN];
 126         int offset;
 127 };
 128
 129 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 130                              offsetof(struct ibmvnic_statistics, stat))
 131 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + off)))
 132
 133 static const struct ibmvnic_stat ibmvnic_stats[] = {
 134         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 135         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 136         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 137         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 138         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 139         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 140         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 141         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 142         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 143         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 144         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 145         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 146         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 147         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 148         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 149         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 150         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 151         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 152         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 153         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 154         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 155         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 156 };
 157
 158 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 159                           unsigned long length, unsigned long *number,
 160                           unsigned long *irq)
 161 {
 162         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 163         long rc;
 164
 165         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 166         *number = retbuf[0];
 167         *irq = retbuf[1];
 168
 169         return rc;
 170 }
 171
 172 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 173                                 struct ibmvnic_long_term_buff *ltb, int size)
 174 {
 175         struct device *dev = &adapter->vdev->dev;
 176         int rc;
 177
 178         ltb->size = size;
 179         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 180                                        GFP_KERNEL);
 181
 182         if (!ltb->buff) {
 183                 dev_err(dev, "Couldn't alloc long term buffer\n");
 184                 return -ENOMEM;
 185         }
 186         ltb->map_id = adapter->map_id;
 187         adapter->map_id++;
 188
 189         init_completion(&adapter->fw_done);
 190         rc = send_request_map(adapter, ltb->addr,
 191                               ltb->size, ltb->map_id);
 192         if (rc) {
 193                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 194                 return rc;
 195         }
 196         wait_for_completion(&adapter->fw_done);
 197
 198         if (adapter->fw_done_rc) {
 199                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 200                         adapter->fw_done_rc);
 201                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 202                 return -1;
 203         }
 204         return 0;
 205 }
 206
 207 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 208                                 struct ibmvnic_long_term_buff *ltb)
 209 {
 210         struct device *dev = &adapter->vdev->dev;
 211
 212         if (!ltb->buff)
 213                 return;
 214
 215         /* VIOS automatically unmaps the long term buffer at remote
 216          * end for the following resets:
 217          * FAILOVER, MOBILITY, TIMEOUT.
 218          */
 219         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 220             adapter->reset_reason != VNIC_RESET_MOBILITY &&
 221             adapter->reset_reason != VNIC_RESET_TIMEOUT)
 222                 send_request_unmap(adapter, ltb->map_id);
 223         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 224 }
 225
 226 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 227                                 struct ibmvnic_long_term_buff *ltb)
 228 {
 229         int rc;
 230
 231         memset(ltb->buff, 0, ltb->size);
 232
 233         init_completion(&adapter->fw_done);
 234         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 235         if (rc)
 236                 return rc;
 237         wait_for_completion(&adapter->fw_done);
 238
 239         if (adapter->fw_done_rc) {
 240                 dev_info(&adapter->vdev->dev,
 241                          "Reset failed, attempting to free and reallocate buffer\n");
 242                 free_long_term_buff(adapter, ltb);
 243                 return alloc_long_term_buff(adapter, ltb, ltb->size);
 244         }
 245         return 0;
 246 }
 247
 248 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 249 {
 250         int i;
 251
 252         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 253              i++)
 254                 adapter->rx_pool[i].active = 0;
 255 }
 256
 257 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 258                               struct ibmvnic_rx_pool *pool)
 259 {
 260         int count = pool->size - atomic_read(&pool->available);
 261         struct device *dev = &adapter->vdev->dev;
 262         int buffers_added = 0;
 263         unsigned long lpar_rc;
 264         union sub_crq sub_crq;
 265         struct sk_buff *skb;
 266         unsigned int offset;
 267         dma_addr_t dma_addr;
 268         unsigned char *dst;
 269         u64 *handle_array;
 270         int shift = 0;
 271         int index;
 272         int i;
 273
 274         if (!pool->active)
 275                 return;
 276
 277         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 278                                       be32_to_cpu(adapter->login_rsp_buf->
 279                                       off_rxadd_subcrqs));
 280
 281         for (i = 0; i < count; ++i) {
 282                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 283                 if (!skb) {
 284                         dev_err(dev, "Couldn't replenish rx buff\n");
 285                         adapter->replenish_no_mem++;
 286                         break;
 287                 }
 288
 289                 index = pool->free_map[pool->next_free];
 290
 291                 if (pool->rx_buff[index].skb)
 292                         dev_err(dev, "Inconsistent free_map!\n");
 293
 294                 /* Copy the skb to the long term mapped DMA buffer */
 295                 offset = index * pool->buff_size;
 296                 dst = pool->long_term_buff.buff + offset;
 297                 memset(dst, 0, pool->buff_size);
 298                 dma_addr = pool->long_term_buff.addr + offset;
 299                 pool->rx_buff[index].data = dst;
 300
 301                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 302                 pool->rx_buff[index].dma = dma_addr;
 303                 pool->rx_buff[index].skb = skb;
 304                 pool->rx_buff[index].pool_index = pool->index;
 305                 pool->rx_buff[index].size = pool->buff_size;
 306
 307                 memset(&sub_crq, 0, sizeof(sub_crq));
 308                 sub_crq.rx_add.first = IBMVNIC_CRQ_CMD;
 309                 sub_crq.rx_add.correlator =
 310                     cpu_to_be64((u64)&pool->rx_buff[index]);
 311                 sub_crq.rx_add.ioba = cpu_to_be32(dma_addr);
 312                 sub_crq.rx_add.map_id = pool->long_term_buff.map_id;
 313
 314                 /* The length field of the sCRQ is defined to be 24 bits so the
 315                  * buffer size needs to be left shifted by a byte before it is
 316                  * converted to big endian to prevent the last byte from being
 317                  * truncated.
 318                  */
 319 #ifdef __LITTLE_ENDIAN__
 320                 shift = 8;
 321 #endif
 322                 sub_crq.rx_add.len = cpu_to_be32(pool->buff_size << shift);
 323
 324                 lpar_rc = send_subcrq(adapter, handle_array[pool->index],
 325                                       &sub_crq);
 326                 if (lpar_rc != H_SUCCESS)
 327                         goto failure;
 328
 329                 buffers_added++;
 330                 adapter->replenish_add_buff_success++;
 331                 pool->next_free = (pool->next_free + 1) % pool->size;
 332         }
 333         atomic_add(buffers_added, &pool->available);
 334         return;
 335
 336 failure:
 337         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 338                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 339         pool->free_map[pool->next_free] = index;
 340         pool->rx_buff[index].skb = NULL;
 341
 342         dev_kfree_skb_any(skb);
 343         adapter->replenish_add_buff_failure++;
 344         atomic_add(buffers_added, &pool->available);
 345
 346         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 347                 /* Disable buffer pool replenishment and report carrier off if
 348                  * queue is closed or pending failover.
 349                  * Firmware guarantees that a signal will be sent to the
 350                  * driver, triggering a reset.
 351                  */
 352                 deactivate_rx_pools(adapter);
 353                 netif_carrier_off(adapter->netdev);
 354         }
 355 }
 356
 357 static void replenish_pools(struct ibmvnic_adapter *adapter)
 358 {
 359         int i;
 360
 361         adapter->replenish_task_cycles++;
 362         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 363              i++) {
 364                 if (adapter->rx_pool[i].active)
 365                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 366         }
 367 }
 368
 369 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 370 {
 371         kfree(adapter->tx_stats_buffers);
 372         kfree(adapter->rx_stats_buffers);
 373         adapter->tx_stats_buffers = NULL;
 374         adapter->rx_stats_buffers = NULL;
 375 }
 376
 377 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 378 {
 379         adapter->tx_stats_buffers =
 380                                 kcalloc(IBMVNIC_MAX_QUEUES,
 381                                         sizeof(struct ibmvnic_tx_queue_stats),
 382                                         GFP_KERNEL);
 383         if (!adapter->tx_stats_buffers)
 384                 return -ENOMEM;
 385
 386         adapter->rx_stats_buffers =
 387                                 kcalloc(IBMVNIC_MAX_QUEUES,
 388                                         sizeof(struct ibmvnic_rx_queue_stats),
 389                                         GFP_KERNEL);
 390         if (!adapter->rx_stats_buffers)
 391                 return -ENOMEM;
 392
 393         return 0;
 394 }
 395
 396 static void release_stats_token(struct ibmvnic_adapter *adapter)
 397 {
 398         struct device *dev = &adapter->vdev->dev;
 399
 400         if (!adapter->stats_token)
 401                 return;
 402
 403         dma_unmap_single(dev, adapter->stats_token,
 404                          sizeof(struct ibmvnic_statistics),
 405                          DMA_FROM_DEVICE);
 406         adapter->stats_token = 0;
 407 }
 408
 409 static int init_stats_token(struct ibmvnic_adapter *adapter)
 410 {
 411         struct device *dev = &adapter->vdev->dev;
 412         dma_addr_t stok;
 413
 414         stok = dma_map_single(dev, &adapter->stats,
 415                               sizeof(struct ibmvnic_statistics),
 416                               DMA_FROM_DEVICE);
 417         if (dma_mapping_error(dev, stok)) {
 418                 dev_err(dev, "Couldn't map stats buffer\n");
 419                 return -1;
 420         }
 421
 422         adapter->stats_token = stok;
 423         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 424         return 0;
 425 }
 426
 427 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 428 {
 429         struct ibmvnic_rx_pool *rx_pool;
 430         int rx_scrqs;
 431         int i, j, rc;
 432         u64 *size_array;
 433
 434         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 435                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
 436
 437         rx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 438         for (i = 0; i < rx_scrqs; i++) {
 439                 rx_pool = &adapter->rx_pool[i];
 440
 441                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 442
 443                 if (rx_pool->buff_size != be64_to_cpu(size_array[i])) {
 444                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 445                         rx_pool->buff_size = be64_to_cpu(size_array[i]);
 446                         rc = alloc_long_term_buff(adapter,
 447                                                   &rx_pool->long_term_buff,
 448                                                   rx_pool->size *
 449                                                   rx_pool->buff_size);
 450                 } else {
 451                         rc = reset_long_term_buff(adapter,
 452                                                   &rx_pool->long_term_buff);
 453                 }
 454
 455                 if (rc)
 456                         return rc;
 457
 458                 for (j = 0; j < rx_pool->size; j++)
 459                         rx_pool->free_map[j] = j;
 460
 461                 memset(rx_pool->rx_buff, 0,
 462                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 463
 464                 atomic_set(&rx_pool->available, 0);
 465                 rx_pool->next_alloc = 0;
 466                 rx_pool->next_free = 0;
 467                 rx_pool->active = 1;
 468         }
 469
 470         return 0;
 471 }
 472
 473 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 474 {
 475         struct ibmvnic_rx_pool *rx_pool;
 476         int i, j;
 477
 478         if (!adapter->rx_pool)
 479                 return;
 480
 481         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 482                 rx_pool = &adapter->rx_pool[i];
 483
 484                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 485
 486                 kfree(rx_pool->free_map);
 487                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 488
 489                 if (!rx_pool->rx_buff)
 490                         continue;
 491
 492                 for (j = 0; j < rx_pool->size; j++) {
 493                         if (rx_pool->rx_buff[j].skb) {
 494                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 495                                 rx_pool->rx_buff[j].skb = NULL;
 496                         }
 497                 }
 498
 499                 kfree(rx_pool->rx_buff);
 500         }
 501
 502         kfree(adapter->rx_pool);
 503         adapter->rx_pool = NULL;
 504         adapter->num_active_rx_pools = 0;
 505 }
 506
 507 static int init_rx_pools(struct net_device *netdev)
 508 {
 509         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 510         struct device *dev = &adapter->vdev->dev;
 511         struct ibmvnic_rx_pool *rx_pool;
 512         int rxadd_subcrqs;
 513         u64 *size_array;
 514         int i, j;
 515
 516         rxadd_subcrqs =
 517                 be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 518         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 519                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
 520
 521         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 522                                    sizeof(struct ibmvnic_rx_pool),
 523                                    GFP_KERNEL);
 524         if (!adapter->rx_pool) {
 525                 dev_err(dev, "Failed to allocate rx pools\n");
 526                 return -1;
 527         }
 528
 529         adapter->num_active_rx_pools = rxadd_subcrqs;
 530
 531         for (i = 0; i < rxadd_subcrqs; i++) {
 532                 rx_pool = &adapter->rx_pool[i];
 533
 534                 netdev_dbg(adapter->netdev,
 535                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 536                            i, adapter->req_rx_add_entries_per_subcrq,
 537                            be64_to_cpu(size_array[i]));
 538
 539                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 540                 rx_pool->index = i;
 541                 rx_pool->buff_size = be64_to_cpu(size_array[i]);
 542                 rx_pool->active = 1;
 543
 544                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 545                                             GFP_KERNEL);
 546                 if (!rx_pool->free_map) {
 547                         release_rx_pools(adapter);
 548                         return -1;
 549                 }
 550
 551                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 552                                            sizeof(struct ibmvnic_rx_buff),
 553                                            GFP_KERNEL);
 554                 if (!rx_pool->rx_buff) {
 555                         dev_err(dev, "Couldn't alloc rx buffers\n");
 556                         release_rx_pools(adapter);
 557                         return -1;
 558                 }
 559
 560                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 561                                          rx_pool->size * rx_pool->buff_size)) {
 562                         release_rx_pools(adapter);
 563                         return -1;
 564                 }
 565
 566                 for (j = 0; j < rx_pool->size; ++j)
 567                         rx_pool->free_map[j] = j;
 568
 569                 atomic_set(&rx_pool->available, 0);
 570                 rx_pool->next_alloc = 0;
 571                 rx_pool->next_free = 0;
 572         }
 573
 574         return 0;
 575 }
 576
 577 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 578                              struct ibmvnic_tx_pool *tx_pool)
 579 {
 580         int rc, i;
 581
 582         rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 583         if (rc)
 584                 return rc;
 585
 586         memset(tx_pool->tx_buff, 0,
 587                tx_pool->num_buffers *
 588                sizeof(struct ibmvnic_tx_buff));
 589
 590         for (i = 0; i < tx_pool->num_buffers; i++)
 591                 tx_pool->free_map[i] = i;
 592
 593         tx_pool->consumer_index = 0;
 594         tx_pool->producer_index = 0;
 595
 596         return 0;
 597 }
 598
 599 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 600 {
 601         int tx_scrqs;
 602         int i, rc;
 603
 604         tx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
 605         for (i = 0; i < tx_scrqs; i++) {
 606                 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 607                 if (rc)
 608                         return rc;
 609                 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 610                 if (rc)
 611                         return rc;
 612         }
 613
 614         return 0;
 615 }
 616
 617 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 618 {
 619         if (!adapter->vpd)
 620                 return;
 621
 622         kfree(adapter->vpd->buff);
 623         kfree(adapter->vpd);
 624
 625         adapter->vpd = NULL;
 626 }
 627
 628 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 629                                 struct ibmvnic_tx_pool *tx_pool)
 630 {
 631         kfree(tx_pool->tx_buff);
 632         kfree(tx_pool->free_map);
 633         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 634 }
 635
 636 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 637 {
 638         int i;
 639
 640         if (!adapter->tx_pool)
 641                 return;
 642
 643         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 644                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 645                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 646         }
 647
 648         kfree(adapter->tx_pool);
 649         adapter->tx_pool = NULL;
 650         kfree(adapter->tso_pool);
 651         adapter->tso_pool = NULL;
 652         adapter->num_active_tx_pools = 0;
 653 }
 654
 655 static int init_one_tx_pool(struct net_device *netdev,
 656                             struct ibmvnic_tx_pool *tx_pool,
 657                             int num_entries, int buf_size)
 658 {
 659         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 660         int i;
 661
 662         tx_pool->tx_buff = kcalloc(num_entries,
 663                                    sizeof(struct ibmvnic_tx_buff),
 664                                    GFP_KERNEL);
 665         if (!tx_pool->tx_buff)
 666                 return -1;
 667
 668         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 669                                  num_entries * buf_size))
 670                 return -1;
 671
 672         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 673         if (!tx_pool->free_map)
 674                 return -1;
 675
 676         for (i = 0; i < num_entries; i++)
 677                 tx_pool->free_map[i] = i;
 678
 679         tx_pool->consumer_index = 0;
 680         tx_pool->producer_index = 0;
 681         tx_pool->num_buffers = num_entries;
 682         tx_pool->buf_size = buf_size;
 683
 684         return 0;
 685 }
 686
 687 static int init_tx_pools(struct net_device *netdev)
 688 {
 689         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 690         int tx_subcrqs;
 691         int i, rc;
 692
 693         tx_subcrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
 694         adapter->tx_pool = kcalloc(tx_subcrqs,
 695                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 696         if (!adapter->tx_pool)
 697                 return -1;
 698
 699         adapter->tso_pool = kcalloc(tx_subcrqs,
 700                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 701         if (!adapter->tso_pool) {
 702                 kfree(adapter->tx_pool);
 703                 adapter->tx_pool = NULL;
 704                 return -1;
 705         }
 706
 707         adapter->num_active_tx_pools = tx_subcrqs;
 708
 709         for (i = 0; i < tx_subcrqs; i++) {
 710                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 711                                       adapter->req_tx_entries_per_subcrq,
 712                                       adapter->req_mtu + VLAN_HLEN);
 713                 if (rc) {
 714                         release_tx_pools(adapter);
 715                         return rc;
 716                 }
 717
 718                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 719                                       IBMVNIC_TSO_BUFS,
 720                                       IBMVNIC_TSO_BUF_SZ);
 721                 if (rc) {
 722                         release_tx_pools(adapter);
 723                         return rc;
 724                 }
 725         }
 726
 727         return 0;
 728 }
 729
 730 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 731 {
 732         int i;
 733
 734         if (adapter->napi_enabled)
 735                 return;
 736
 737         for (i = 0; i < adapter->req_rx_queues; i++)
 738                 napi_enable(&adapter->napi[i]);
 739
 740         adapter->napi_enabled = true;
 741 }
 742
 743 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 744 {
 745         int i;
 746
 747         if (!adapter->napi_enabled)
 748                 return;
 749
 750         for (i = 0; i < adapter->req_rx_queues; i++) {
 751                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 752                 napi_disable(&adapter->napi[i]);
 753         }
 754
 755         adapter->napi_enabled = false;
 756 }
 757
 758 static int init_napi(struct ibmvnic_adapter *adapter)
 759 {
 760         int i;
 761
 762         adapter->napi = kcalloc(adapter->req_rx_queues,
 763                                 sizeof(struct napi_struct), GFP_KERNEL);
 764         if (!adapter->napi)
 765                 return -ENOMEM;
 766
 767         for (i = 0; i < adapter->req_rx_queues; i++) {
 768                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 769                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 770                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 771         }
 772
 773         adapter->num_active_rx_napi = adapter->req_rx_queues;
 774         return 0;
 775 }
 776
 777 static void release_napi(struct ibmvnic_adapter *adapter)
 778 {
 779         int i;
 780
 781         if (!adapter->napi)
 782                 return;
 783
 784         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 785                 if (&adapter->napi[i]) {
 786                         netdev_dbg(adapter->netdev,
 787                                    "Releasing napi[%d]\n", i);
 788                         netif_napi_del(&adapter->napi[i]);
 789                 }
 790         }
 791
 792         kfree(adapter->napi);
 793         adapter->napi = NULL;
 794         adapter->num_active_rx_napi = 0;
 795         adapter->napi_enabled = false;
 796 }
 797
 798 static int ibmvnic_login(struct net_device *netdev)
 799 {
 800         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 801         unsigned long timeout = msecs_to_jiffies(30000);
 802         int retry_count = 0;
 803         int retries = 10;
 804         bool retry;
 805         int rc;
 806
 807         do {
 808                 retry = false;
 809                 if (retry_count > retries) {
 810                         netdev_warn(netdev, "Login attempts exceeded\n");
 811                         return -1;
 812                 }
 813
 814                 adapter->init_done_rc = 0;
 815                 reinit_completion(&adapter->init_done);
 816                 rc = send_login(adapter);
 817                 if (rc) {
 818                         netdev_warn(netdev, "Unable to login\n");
 819                         return rc;
 820                 }
 821
 822                 if (!wait_for_completion_timeout(&adapter->init_done,
 823                                                  timeout)) {
 824                         netdev_warn(netdev, "Login timed out, retrying...\n");
 825                         retry = true;
 826                         adapter->init_done_rc = 0;
 827                         retry_count++;
 828                         continue;
 829                 }
 830
 831                 if (adapter->init_done_rc == ABORTED) {
 832                         netdev_warn(netdev, "Login aborted, retrying...\n");
 833                         retry = true;
 834                         adapter->init_done_rc = 0;
 835                         retry_count++;
 836                         /* FW or device may be busy, so
 837                          * wait a bit before retrying login
 838                          */
 839                         msleep(500);
 840                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 841                         retry_count++;
 842                         release_sub_crqs(adapter, 1);
 843
 844                         retry = true;
 845                         netdev_dbg(netdev,
 846                                    "Received partial success, retrying...\n");
 847                         adapter->init_done_rc = 0;
 848                         reinit_completion(&adapter->init_done);
 849                         send_cap_queries(adapter);
 850                         if (!wait_for_completion_timeout(&adapter->init_done,
 851                                                          timeout)) {
 852                                 netdev_warn(netdev,
 853                                             "Capabilities query timed out\n");
 854                                 return -1;
 855                         }
 856
 857                         rc = init_sub_crqs(adapter);
 858                         if (rc) {
 859                                 netdev_warn(netdev,
 860                                             "SCRQ initialization failed\n");
 861                                 return -1;
 862                         }
 863
 864                         rc = init_sub_crq_irqs(adapter);
 865                         if (rc) {
 866                                 netdev_warn(netdev,
 867                                             "SCRQ irq initialization failed\n");
 868                                 return -1;
 869                         }
 870                 } else if (adapter->init_done_rc) {
 871                         netdev_warn(netdev, "Adapter login failed\n");
 872                         return -1;
 873                 }
 874         } while (retry);
 875
 876         /* handle pending MAC address changes after successful login */
 877         if (adapter->mac_change_pending) {
 878                 __ibmvnic_set_mac(netdev, &adapter->desired.mac);
 879                 adapter->mac_change_pending = false;
 880         }
 881
 882         return 0;
 883 }
 884
 885 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 886 {
 887         kfree(adapter->login_buf);
 888         adapter->login_buf = NULL;
 889 }
 890
 891 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 892 {
 893         kfree(adapter->login_rsp_buf);
 894         adapter->login_rsp_buf = NULL;
 895 }
 896
 897 static void release_resources(struct ibmvnic_adapter *adapter)
 898 {
 899         release_vpd_data(adapter);
 900
 901         release_tx_pools(adapter);
 902         release_rx_pools(adapter);
 903
 904         release_napi(adapter);
 905         release_login_rsp_buffer(adapter);
 906 }
 907
 908 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 909 {
 910         struct net_device *netdev = adapter->netdev;
 911         unsigned long timeout = msecs_to_jiffies(30000);
 912         union ibmvnic_crq crq;
 913         bool resend;
 914         int rc;
 915
 916         netdev_dbg(netdev, "setting link state %d\n", link_state);
 917
 918         memset(&crq, 0, sizeof(crq));
 919         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 920         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 921         crq.logical_link_state.link_state = link_state;
 922
 923         do {
 924                 resend = false;
 925
 926                 reinit_completion(&adapter->init_done);
 927                 rc = ibmvnic_send_crq(adapter, &crq);
 928                 if (rc) {
 929                         netdev_err(netdev, "Failed to set link state\n");
 930                         return rc;
 931                 }
 932
 933                 if (!wait_for_completion_timeout(&adapter->init_done,
 934                                                  timeout)) {
 935                         netdev_err(netdev, "timeout setting link state\n");
 936                         return -1;
 937                 }
 938
 939                 if (adapter->init_done_rc == 1) {
 940                         /* Partuial success, delay and re-send */
 941                         mdelay(1000);
 942                         resend = true;
 943                 } else if (adapter->init_done_rc) {
 944                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 945                                     adapter->init_done_rc);
 946                         return adapter->init_done_rc;
 947                 }
 948         } while (resend);
 949
 950         return 0;
 951 }
 952
 953 static int set_real_num_queues(struct net_device *netdev)
 954 {
 955         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 956         int rc;
 957
 958         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
 959                    adapter->req_tx_queues, adapter->req_rx_queues);
 960
 961         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
 962         if (rc) {
 963                 netdev_err(netdev, "failed to set the number of tx queues\n");
 964                 return rc;
 965         }
 966
 967         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
 968         if (rc)
 969                 netdev_err(netdev, "failed to set the number of rx queues\n");
 970
 971         return rc;
 972 }
 973
 974 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
 975 {
 976         struct device *dev = &adapter->vdev->dev;
 977         union ibmvnic_crq crq;
 978         int len = 0;
 979         int rc;
 980
 981         if (adapter->vpd->buff)
 982                 len = adapter->vpd->len;
 983
 984         init_completion(&adapter->fw_done);
 985         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
 986         crq.get_vpd_size.cmd = GET_VPD_SIZE;
 987         rc = ibmvnic_send_crq(adapter, &crq);
 988         if (rc)
 989                 return rc;
 990         wait_for_completion(&adapter->fw_done);
 991
 992         if (!adapter->vpd->len)
 993                 return -ENODATA;
 994
 995         if (!adapter->vpd->buff)
 996                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
 997         else if (adapter->vpd->len != len)
 998                 adapter->vpd->buff =
 999                         krealloc(adapter->vpd->buff,
1000                                  adapter->vpd->len, GFP_KERNEL);
1001
1002         if (!adapter->vpd->buff) {
1003                 dev_err(dev, "Could allocate VPD buffer\n");
1004                 return -ENOMEM;
1005         }
1006
1007         adapter->vpd->dma_addr =
1008                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1009                                DMA_FROM_DEVICE);
1010         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1011                 dev_err(dev, "Could not map VPD buffer\n");
1012                 kfree(adapter->vpd->buff);
1013                 adapter->vpd->buff = NULL;
1014                 return -ENOMEM;
1015         }
1016
1017         reinit_completion(&adapter->fw_done);
1018         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1019         crq.get_vpd.cmd = GET_VPD;
1020         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1021         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1022         rc = ibmvnic_send_crq(adapter, &crq);
1023         if (rc) {
1024                 kfree(adapter->vpd->buff);
1025                 adapter->vpd->buff = NULL;
1026                 return rc;
1027         }
1028         wait_for_completion(&adapter->fw_done);
1029
1030         return 0;
1031 }
1032
1033 static int init_resources(struct ibmvnic_adapter *adapter)
1034 {
1035         struct net_device *netdev = adapter->netdev;
1036         int rc;
1037
1038         rc = set_real_num_queues(netdev);
1039         if (rc)
1040                 return rc;
1041
1042         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1043         if (!adapter->vpd)
1044                 return -ENOMEM;
1045
1046         /* Vital Product Data (VPD) */
1047         rc = ibmvnic_get_vpd(adapter);
1048         if (rc) {
1049                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1050                 return rc;
1051         }
1052
1053         adapter->map_id = 1;
1054
1055         rc = init_napi(adapter);
1056         if (rc)
1057                 return rc;
1058
1059         send_map_query(adapter);
1060
1061         rc = init_rx_pools(netdev);
1062         if (rc)
1063                 return rc;
1064
1065         rc = init_tx_pools(netdev);
1066         return rc;
1067 }
1068
1069 static int __ibmvnic_open(struct net_device *netdev)
1070 {
1071         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1072         enum vnic_state prev_state = adapter->state;
1073         int i, rc;
1074
1075         adapter->state = VNIC_OPENING;
1076         replenish_pools(adapter);
1077         ibmvnic_napi_enable(adapter);
1078
1079         /* We're ready to receive frames, enable the sub-crq interrupts and
1080          * set the logical link state to up
1081          */
1082         for (i = 0; i < adapter->req_rx_queues; i++) {
1083                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1084                 if (prev_state == VNIC_CLOSED)
1085                         enable_irq(adapter->rx_scrq[i]->irq);
1086                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1087         }
1088
1089         for (i = 0; i < adapter->req_tx_queues; i++) {
1090                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1091                 if (prev_state == VNIC_CLOSED)
1092                         enable_irq(adapter->tx_scrq[i]->irq);
1093                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1094         }
1095
1096         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1097         if (rc) {
1098                 ibmvnic_napi_disable(adapter);
1099                 release_resources(adapter);
1100                 return rc;
1101         }
1102
1103         netif_tx_start_all_queues(netdev);
1104
1105         if (prev_state == VNIC_CLOSED) {
1106                 for (i = 0; i < adapter->req_rx_queues; i++)
1107                         napi_schedule(&adapter->napi[i]);
1108         }
1109
1110         adapter->state = VNIC_OPEN;
1111         return rc;
1112 }
1113
1114 static int ibmvnic_open(struct net_device *netdev)
1115 {
1116         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1117         int rc;
1118
1119         /* If device failover is pending, just set device state and return.
1120          * Device operation will be handled by reset routine.
1121          */
1122         if (adapter->failover_pending) {
1123                 adapter->state = VNIC_OPEN;
1124                 return 0;
1125         }
1126
1127         if (adapter->state != VNIC_CLOSED) {
1128                 rc = ibmvnic_login(netdev);
1129                 if (rc)
1130                         return rc;
1131
1132                 rc = init_resources(adapter);
1133                 if (rc) {
1134                         netdev_err(netdev, "failed to initialize resources\n");
1135                         release_resources(adapter);
1136                         return rc;
1137                 }
1138         }
1139
1140         rc = __ibmvnic_open(netdev);
1141         netif_carrier_on(netdev);
1142
1143         return rc;
1144 }
1145
1146 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1147 {
1148         struct ibmvnic_rx_pool *rx_pool;
1149         struct ibmvnic_rx_buff *rx_buff;
1150         u64 rx_entries;
1151         int rx_scrqs;
1152         int i, j;
1153
1154         if (!adapter->rx_pool)
1155                 return;
1156
1157         rx_scrqs = adapter->num_active_rx_pools;
1158         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1159
1160         /* Free any remaining skbs in the rx buffer pools */
1161         for (i = 0; i < rx_scrqs; i++) {
1162                 rx_pool = &adapter->rx_pool[i];
1163                 if (!rx_pool || !rx_pool->rx_buff)
1164                         continue;
1165
1166                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1167                 for (j = 0; j < rx_entries; j++) {
1168                         rx_buff = &rx_pool->rx_buff[j];
1169                         if (rx_buff && rx_buff->skb) {
1170                                 dev_kfree_skb_any(rx_buff->skb);
1171                                 rx_buff->skb = NULL;
1172                         }
1173                 }
1174         }
1175 }
1176
1177 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1178                               struct ibmvnic_tx_pool *tx_pool)
1179 {
1180         struct ibmvnic_tx_buff *tx_buff;
1181         u64 tx_entries;
1182         int i;
1183
1184         if (!tx_pool || !tx_pool->tx_buff)
1185                 return;
1186
1187         tx_entries = tx_pool->num_buffers;
1188
1189         for (i = 0; i < tx_entries; i++) {
1190                 tx_buff = &tx_pool->tx_buff[i];
1191                 if (tx_buff && tx_buff->skb) {
1192                         dev_kfree_skb_any(tx_buff->skb);
1193                         tx_buff->skb = NULL;
1194                 }
1195         }
1196 }
1197
1198 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1199 {
1200         int tx_scrqs;
1201         int i;
1202
1203         if (!adapter->tx_pool || !adapter->tso_pool)
1204                 return;
1205
1206         tx_scrqs = adapter->num_active_tx_pools;
1207
1208         /* Free any remaining skbs in the tx buffer pools */
1209         for (i = 0; i < tx_scrqs; i++) {
1210                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1211                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1212                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1213         }
1214 }
1215
1216 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1217 {
1218         struct net_device *netdev = adapter->netdev;
1219         int i;
1220
1221         if (adapter->tx_scrq) {
1222                 for (i = 0; i < adapter->req_tx_queues; i++)
1223                         if (adapter->tx_scrq[i]->irq) {
1224                                 netdev_dbg(netdev,
1225                                            "Disabling tx_scrq[%d] irq\n", i);
1226                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1227                                 disable_irq(adapter->tx_scrq[i]->irq);
1228                         }
1229         }
1230
1231         if (adapter->rx_scrq) {
1232                 for (i = 0; i < adapter->req_rx_queues; i++) {
1233                         if (adapter->rx_scrq[i]->irq) {
1234                                 netdev_dbg(netdev,
1235                                            "Disabling rx_scrq[%d] irq\n", i);
1236                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1237                                 disable_irq(adapter->rx_scrq[i]->irq);
1238                         }
1239                 }
1240         }
1241 }
1242
1243 static void ibmvnic_cleanup(struct net_device *netdev)
1244 {
1245         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1246
1247         /* ensure that transmissions are stopped if called by do_reset */
1248         if (adapter->resetting)
1249                 netif_tx_disable(netdev);
1250         else
1251                 netif_tx_stop_all_queues(netdev);
1252
1253         ibmvnic_napi_disable(adapter);
1254         ibmvnic_disable_irqs(adapter);
1255
1256         clean_rx_pools(adapter);
1257         clean_tx_pools(adapter);
1258 }
1259
1260 static int __ibmvnic_close(struct net_device *netdev)
1261 {
1262         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1263         int rc = 0;
1264
1265         adapter->state = VNIC_CLOSING;
1266         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1267         adapter->state = VNIC_CLOSED;
1268         return rc;
1269 }
1270
1271 static int ibmvnic_close(struct net_device *netdev)
1272 {
1273         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1274         int rc;
1275
1276         /* If device failover is pending, just set device state and return.
1277          * Device operation will be handled by reset routine.
1278          */
1279         if (adapter->failover_pending) {
1280                 adapter->state = VNIC_CLOSED;
1281                 return 0;
1282         }
1283
1284         rc = __ibmvnic_close(netdev);
1285         ibmvnic_cleanup(netdev);
1286
1287         return rc;
1288 }
1289
1290 /**
1291  * build_hdr_data - creates L2/L3/L4 header data buffer
1292  * @hdr_field - bitfield determining needed headers
1293  * @skb - socket buffer
1294  * @hdr_len - array of header lengths
1295  * @tot_len - total length of data
1296  *
1297  * Reads hdr_field to determine which headers are needed by firmware.
1298  * Builds a buffer containing these headers.  Saves individual header
1299  * lengths and total buffer length to be used to build descriptors.
1300  */
1301 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1302                           int *hdr_len, u8 *hdr_data)
1303 {
1304         int len = 0;
1305         u8 *hdr;
1306
1307         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1308                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1309         else
1310                 hdr_len[0] = sizeof(struct ethhdr);
1311
1312         if (skb->protocol == htons(ETH_P_IP)) {
1313                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1314                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1315                         hdr_len[2] = tcp_hdrlen(skb);
1316                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1317                         hdr_len[2] = sizeof(struct udphdr);
1318         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1319                 hdr_len[1] = sizeof(struct ipv6hdr);
1320                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1321                         hdr_len[2] = tcp_hdrlen(skb);
1322                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1323                         hdr_len[2] = sizeof(struct udphdr);
1324         } else if (skb->protocol == htons(ETH_P_ARP)) {
1325                 hdr_len[1] = arp_hdr_len(skb->dev);
1326                 hdr_len[2] = 0;
1327         }
1328
1329         memset(hdr_data, 0, 120);
1330         if ((hdr_field >> 6) & 1) {
1331                 hdr = skb_mac_header(skb);
1332                 memcpy(hdr_data, hdr, hdr_len[0]);
1333                 len += hdr_len[0];
1334         }
1335
1336         if ((hdr_field >> 5) & 1) {
1337                 hdr = skb_network_header(skb);
1338                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1339                 len += hdr_len[1];
1340         }
1341
1342         if ((hdr_field >> 4) & 1) {
1343                 hdr = skb_transport_header(skb);
1344                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1345                 len += hdr_len[2];
1346         }
1347         return len;
1348 }
1349
1350 /**
1351  * create_hdr_descs - create header and header extension descriptors
1352  * @hdr_field - bitfield determining needed headers
1353  * @data - buffer containing header data
1354  * @len - length of data buffer
1355  * @hdr_len - array of individual header lengths
1356  * @scrq_arr - descriptor array
1357  *
1358  * Creates header and, if needed, header extension descriptors and
1359  * places them in a descriptor array, scrq_arr
1360  */
1361
1362 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1363                             union sub_crq *scrq_arr)
1364 {
1365         union sub_crq hdr_desc;
1366         int tmp_len = len;
1367         int num_descs = 0;
1368         u8 *data, *cur;
1369         int tmp;
1370
1371         while (tmp_len > 0) {
1372                 cur = hdr_data + len - tmp_len;
1373
1374                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1375                 if (cur != hdr_data) {
1376                         data = hdr_desc.hdr_ext.data;
1377                         tmp = tmp_len > 29 ? 29 : tmp_len;
1378                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1379                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1380                         hdr_desc.hdr_ext.len = tmp;
1381                 } else {
1382                         data = hdr_desc.hdr.data;
1383                         tmp = tmp_len > 24 ? 24 : tmp_len;
1384                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1385                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1386                         hdr_desc.hdr.len = tmp;
1387                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1388                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1389                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1390                         hdr_desc.hdr.flag = hdr_field << 1;
1391                 }
1392                 memcpy(data, cur, tmp);
1393                 tmp_len -= tmp;
1394                 *scrq_arr = hdr_desc;
1395                 scrq_arr++;
1396                 num_descs++;
1397         }
1398
1399         return num_descs;
1400 }
1401
1402 /**
1403  * build_hdr_descs_arr - build a header descriptor array
1404  * @skb - socket buffer
1405  * @num_entries - number of descriptors to be sent
1406  * @subcrq - first TX descriptor
1407  * @hdr_field - bit field determining which headers will be sent
1408  *
1409  * This function will build a TX descriptor array with applicable
1410  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1411  */
1412
1413 static void build_hdr_descs_arr(struct ibmvnic_tx_buff *txbuff,
1414                                 int *num_entries, u8 hdr_field)
1415 {
1416         int hdr_len[3] = {0, 0, 0};
1417         int tot_len;
1418         u8 *hdr_data = txbuff->hdr_data;
1419
1420         tot_len = build_hdr_data(hdr_field, txbuff->skb, hdr_len,
1421                                  txbuff->hdr_data);
1422         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1423                          txbuff->indir_arr + 1);
1424 }
1425
1426 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1427                                     struct net_device *netdev)
1428 {
1429         /* For some backing devices, mishandling of small packets
1430          * can result in a loss of connection or TX stall. Device
1431          * architects recommend that no packet should be smaller
1432          * than the minimum MTU value provided to the driver, so
1433          * pad any packets to that length
1434          */
1435         if (skb->len < netdev->min_mtu)
1436                 return skb_put_padto(skb, netdev->min_mtu);
1437
1438         return 0;
1439 }
1440
1441 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1442 {
1443         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1444         int queue_num = skb_get_queue_mapping(skb);
1445         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1446         struct device *dev = &adapter->vdev->dev;
1447         struct ibmvnic_tx_buff *tx_buff = NULL;
1448         struct ibmvnic_sub_crq_queue *tx_scrq;
1449         struct ibmvnic_tx_pool *tx_pool;
1450         unsigned int tx_send_failed = 0;
1451         unsigned int tx_map_failed = 0;
1452         unsigned int tx_dropped = 0;
1453         unsigned int tx_packets = 0;
1454         unsigned int tx_bytes = 0;
1455         dma_addr_t data_dma_addr;
1456         struct netdev_queue *txq;
1457         unsigned long lpar_rc;
1458         union sub_crq tx_crq;
1459         unsigned int offset;
1460         int num_entries = 1;
1461         unsigned char *dst;
1462         u64 *handle_array;
1463         int index = 0;
1464         u8 proto = 0;
1465         netdev_tx_t ret = NETDEV_TX_OK;
1466
1467         if (adapter->resetting) {
1468                 if (!netif_subqueue_stopped(netdev, skb))
1469                         netif_stop_subqueue(netdev, queue_num);
1470                 dev_kfree_skb_any(skb);
1471
1472                 tx_send_failed++;
1473                 tx_dropped++;
1474                 ret = NETDEV_TX_OK;
1475                 goto out;
1476         }
1477
1478         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1479                 tx_dropped++;
1480                 tx_send_failed++;
1481                 ret = NETDEV_TX_OK;
1482                 goto out;
1483         }
1484         if (skb_is_gso(skb))
1485                 tx_pool = &adapter->tso_pool[queue_num];
1486         else
1487                 tx_pool = &adapter->tx_pool[queue_num];
1488
1489         tx_scrq = adapter->tx_scrq[queue_num];
1490         txq = netdev_get_tx_queue(netdev, skb_get_queue_mapping(skb));
1491         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
1492                 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
1493
1494         index = tx_pool->free_map[tx_pool->consumer_index];
1495
1496         if (index == IBMVNIC_INVALID_MAP) {
1497                 dev_kfree_skb_any(skb);
1498                 tx_send_failed++;
1499                 tx_dropped++;
1500                 ret = NETDEV_TX_OK;
1501                 goto out;
1502         }
1503
1504         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1505
1506         offset = index * tx_pool->buf_size;
1507         dst = tx_pool->long_term_buff.buff + offset;
1508         memset(dst, 0, tx_pool->buf_size);
1509         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1510
1511         if (skb_shinfo(skb)->nr_frags) {
1512                 int cur, i;
1513
1514                 /* Copy the head */
1515                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1516                 cur = skb_headlen(skb);
1517
1518                 /* Copy the frags */
1519                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1520                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1521
1522                         memcpy(dst + cur,
1523                                page_address(skb_frag_page(frag)) +
1524                                frag->page_offset, skb_frag_size(frag));
1525                         cur += skb_frag_size(frag);
1526                 }
1527         } else {
1528                 skb_copy_from_linear_data(skb, dst, skb->len);
1529         }
1530
1531         /* post changes to long_term_buff *dst before VIOS accessing it */
1532         dma_wmb();
1533
1534         tx_pool->consumer_index =
1535             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1536
1537         tx_buff = &tx_pool->tx_buff[index];
1538         tx_buff->skb = skb;
1539         tx_buff->data_dma[0] = data_dma_addr;
1540         tx_buff->data_len[0] = skb->len;
1541         tx_buff->index = index;
1542         tx_buff->pool_index = queue_num;
1543         tx_buff->last_frag = true;
1544
1545         memset(&tx_crq, 0, sizeof(tx_crq));
1546         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1547         tx_crq.v1.type = IBMVNIC_TX_DESC;
1548         tx_crq.v1.n_crq_elem = 1;
1549         tx_crq.v1.n_sge = 1;
1550         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1551
1552         if (skb_is_gso(skb))
1553                 tx_crq.v1.correlator =
1554                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1555         else
1556                 tx_crq.v1.correlator = cpu_to_be32(index);
1557         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1558         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1559         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1560
1561         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1562                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1563                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1564         }
1565
1566         if (skb->protocol == htons(ETH_P_IP)) {
1567                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1568                 proto = ip_hdr(skb)->protocol;
1569         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1570                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1571                 proto = ipv6_hdr(skb)->nexthdr;
1572         }
1573
1574         if (proto == IPPROTO_TCP)
1575                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1576         else if (proto == IPPROTO_UDP)
1577                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1578
1579         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1580                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1581                 hdrs += 2;
1582         }
1583         if (skb_is_gso(skb)) {
1584                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1585                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1586                 hdrs += 2;
1587         }
1588         /* determine if l2/3/4 headers are sent to firmware */
1589         if ((*hdrs >> 7) & 1) {
1590                 build_hdr_descs_arr(tx_buff, &num_entries, *hdrs);
1591                 tx_crq.v1.n_crq_elem = num_entries;
1592                 tx_buff->num_entries = num_entries;
1593                 tx_buff->indir_arr[0] = tx_crq;
1594                 tx_buff->indir_dma = dma_map_single(dev, tx_buff->indir_arr,
1595                                                     sizeof(tx_buff->indir_arr),
1596                                                     DMA_TO_DEVICE);
1597                 if (dma_mapping_error(dev, tx_buff->indir_dma)) {
1598                         dev_kfree_skb_any(skb);
1599                         tx_buff->skb = NULL;
1600                         if (!firmware_has_feature(FW_FEATURE_CMO))
1601                                 dev_err(dev, "tx: unable to map descriptor array\n");
1602                         tx_map_failed++;
1603                         tx_dropped++;
1604                         ret = NETDEV_TX_OK;
1605                         goto tx_err_out;
1606                 }
1607                 lpar_rc = send_subcrq_indirect(adapter, handle_array[queue_num],
1608                                                (u64)tx_buff->indir_dma,
1609                                                (u64)num_entries);
1610                 dma_unmap_single(dev, tx_buff->indir_dma,
1611                                  sizeof(tx_buff->indir_arr), DMA_TO_DEVICE);
1612         } else {
1613                 tx_buff->num_entries = num_entries;
1614                 lpar_rc = send_subcrq(adapter, handle_array[queue_num],
1615                                       &tx_crq);
1616         }
1617         if (lpar_rc != H_SUCCESS) {
1618                 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1619                         dev_err_ratelimited(dev, "tx: send failed\n");
1620                 dev_kfree_skb_any(skb);
1621                 tx_buff->skb = NULL;
1622
1623                 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1624                         /* Disable TX and report carrier off if queue is closed
1625                          * or pending failover.
1626                          * Firmware guarantees that a signal will be sent to the
1627                          * driver, triggering a reset or some other action.
1628                          */
1629                         netif_tx_stop_all_queues(netdev);
1630                         netif_carrier_off(netdev);
1631                 }
1632
1633                 tx_send_failed++;
1634                 tx_dropped++;
1635                 ret = NETDEV_TX_OK;
1636                 goto tx_err_out;
1637         }
1638
1639         if (atomic_add_return(num_entries, &tx_scrq->used)
1640                                         >= adapter->req_tx_entries_per_subcrq) {
1641                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1642                 netif_stop_subqueue(netdev, queue_num);
1643         }
1644
1645         tx_packets++;
1646         tx_bytes += skb->len;
1647         txq->trans_start = jiffies;
1648         ret = NETDEV_TX_OK;
1649         goto out;
1650
1651 tx_err_out:
1652         /* roll back consumer index and map array*/
1653         if (tx_pool->consumer_index == 0)
1654                 tx_pool->consumer_index =
1655                         tx_pool->num_buffers - 1;
1656         else
1657                 tx_pool->consumer_index--;
1658         tx_pool->free_map[tx_pool->consumer_index] = index;
1659 out:
1660         netdev->stats.tx_dropped += tx_dropped;
1661         netdev->stats.tx_bytes += tx_bytes;
1662         netdev->stats.tx_packets += tx_packets;
1663         adapter->tx_send_failed += tx_send_failed;
1664         adapter->tx_map_failed += tx_map_failed;
1665         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1666         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1667         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1668
1669         return ret;
1670 }
1671
1672 static void ibmvnic_set_multi(struct net_device *netdev)
1673 {
1674         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1675         struct netdev_hw_addr *ha;
1676         union ibmvnic_crq crq;
1677
1678         memset(&crq, 0, sizeof(crq));
1679         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1680         crq.request_capability.cmd = REQUEST_CAPABILITY;
1681
1682         if (netdev->flags & IFF_PROMISC) {
1683                 if (!adapter->promisc_supported)
1684                         return;
1685         } else {
1686                 if (netdev->flags & IFF_ALLMULTI) {
1687                         /* Accept all multicast */
1688                         memset(&crq, 0, sizeof(crq));
1689                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1690                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1691                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1692                         ibmvnic_send_crq(adapter, &crq);
1693                 } else if (netdev_mc_empty(netdev)) {
1694                         /* Reject all multicast */
1695                         memset(&crq, 0, sizeof(crq));
1696                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1697                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1698                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1699                         ibmvnic_send_crq(adapter, &crq);
1700                 } else {
1701                         /* Accept one or more multicast(s) */
1702                         netdev_for_each_mc_addr(ha, netdev) {
1703                                 memset(&crq, 0, sizeof(crq));
1704                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1705                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1706                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1707                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1708                                                 ha->addr);
1709                                 ibmvnic_send_crq(adapter, &crq);
1710                         }
1711                 }
1712         }
1713 }
1714
1715 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p)
1716 {
1717         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1718         struct sockaddr *addr = p;
1719         union ibmvnic_crq crq;
1720         int rc;
1721
1722         if (!is_valid_ether_addr(addr->sa_data))
1723                 return -EADDRNOTAVAIL;
1724
1725         memset(&crq, 0, sizeof(crq));
1726         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1727         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1728         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], addr->sa_data);
1729
1730         init_completion(&adapter->fw_done);
1731         rc = ibmvnic_send_crq(adapter, &crq);
1732         if (rc)
1733                 return rc;
1734         wait_for_completion(&adapter->fw_done);
1735         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1736         return adapter->fw_done_rc ? -EIO : 0;
1737 }
1738
1739 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1740 {
1741         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1742         struct sockaddr *addr = p;
1743         int rc;
1744
1745         if (adapter->state == VNIC_PROBED) {
1746                 memcpy(&adapter->desired.mac, addr, sizeof(struct sockaddr));
1747                 adapter->mac_change_pending = true;
1748                 return 0;
1749         }
1750
1751         rc = __ibmvnic_set_mac(netdev, addr);
1752
1753         return rc;
1754 }
1755
1756 /**
1757  * do_reset returns zero if we are able to keep processing reset events, or
1758  * non-zero if we hit a fatal error and must halt.
1759  */
1760 static int do_reset(struct ibmvnic_adapter *adapter,
1761                     struct ibmvnic_rwi *rwi, u32 reset_state)
1762 {
1763         u64 old_num_rx_queues, old_num_tx_queues;
1764         u64 old_num_rx_slots, old_num_tx_slots;
1765         struct net_device *netdev = adapter->netdev;
1766         int rc;
1767
1768         netdev_dbg(adapter->netdev, "Re-setting driver (%d)\n",
1769                    rwi->reset_reason);
1770
1771         netif_carrier_off(netdev);
1772         adapter->reset_reason = rwi->reset_reason;
1773
1774         old_num_rx_queues = adapter->req_rx_queues;
1775         old_num_tx_queues = adapter->req_tx_queues;
1776         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1777         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1778
1779         ibmvnic_cleanup(netdev);
1780
1781         if (reset_state == VNIC_OPEN &&
1782             adapter->reset_reason != VNIC_RESET_MOBILITY &&
1783             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1784                 rc = __ibmvnic_close(netdev);
1785                 if (rc)
1786                         return rc;
1787         }
1788
1789         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM ||
1790             adapter->wait_for_reset) {
1791                 release_resources(adapter);
1792                 release_sub_crqs(adapter, 1);
1793                 release_crq_queue(adapter);
1794         }
1795
1796         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
1797                 /* remove the closed state so when we call open it appears
1798                  * we are coming from the probed state.
1799                  */
1800                 adapter->state = VNIC_PROBED;
1801
1802                 if (adapter->wait_for_reset) {
1803                         rc = init_crq_queue(adapter);
1804                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
1805                         rc = ibmvnic_reenable_crq_queue(adapter);
1806                         release_sub_crqs(adapter, 1);
1807                 } else {
1808                         rc = ibmvnic_reset_crq(adapter);
1809                         if (!rc)
1810                                 rc = vio_enable_interrupts(adapter->vdev);
1811                 }
1812
1813                 if (rc) {
1814                         netdev_err(adapter->netdev,
1815                                    "Couldn't initialize crq. rc=%d\n", rc);
1816                         return rc;
1817                 }
1818
1819                 rc = ibmvnic_reset_init(adapter);
1820                 if (rc)
1821                         return IBMVNIC_INIT_FAILED;
1822
1823                 /* If the adapter was in PROBE state prior to the reset,
1824                  * exit here.
1825                  */
1826                 if (reset_state == VNIC_PROBED)
1827                         return 0;
1828
1829                 rc = ibmvnic_login(netdev);
1830                 if (rc) {
1831                         adapter->state = reset_state;
1832                         return rc;
1833                 }
1834
1835                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM ||
1836                     adapter->wait_for_reset) {
1837                         rc = init_resources(adapter);
1838                         if (rc)
1839                                 return rc;
1840                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
1841                            adapter->req_tx_queues != old_num_tx_queues ||
1842                            adapter->req_rx_add_entries_per_subcrq !=
1843                                                         old_num_rx_slots ||
1844                            adapter->req_tx_entries_per_subcrq !=
1845                                                         old_num_tx_slots) {
1846                         release_rx_pools(adapter);
1847                         release_tx_pools(adapter);
1848                         release_napi(adapter);
1849                         release_vpd_data(adapter);
1850
1851                         rc = init_resources(adapter);
1852                         if (rc)
1853                                 return rc;
1854
1855                 } else {
1856                         rc = reset_tx_pools(adapter);
1857                         if (rc)
1858                                 return rc;
1859
1860                         rc = reset_rx_pools(adapter);
1861                         if (rc)
1862                                 return rc;
1863                 }
1864                 ibmvnic_disable_irqs(adapter);
1865         }
1866         adapter->state = VNIC_CLOSED;
1867
1868         if (reset_state == VNIC_CLOSED)
1869                 return 0;
1870
1871         rc = __ibmvnic_open(netdev);
1872         if (rc) {
1873                 if (list_empty(&adapter->rwi_list))
1874                         adapter->state = VNIC_CLOSED;
1875                 else
1876                         adapter->state = reset_state;
1877
1878                 return 0;
1879         }
1880
1881         /* refresh device's multicast list */
1882         ibmvnic_set_multi(netdev);
1883
1884         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
1885             adapter->reset_reason != VNIC_RESET_CHANGE_PARAM) {
1886                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
1887                 call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
1888         }
1889
1890         netif_carrier_on(netdev);
1891
1892         return 0;
1893 }
1894
1895 static int do_hard_reset(struct ibmvnic_adapter *adapter,
1896                          struct ibmvnic_rwi *rwi, u32 reset_state)
1897 {
1898         struct net_device *netdev = adapter->netdev;
1899         int rc;
1900
1901         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
1902                    rwi->reset_reason);
1903
1904         netif_carrier_off(netdev);
1905         adapter->reset_reason = rwi->reset_reason;
1906
1907         ibmvnic_cleanup(netdev);
1908         release_resources(adapter);
1909         release_sub_crqs(adapter, 0);
1910         release_crq_queue(adapter);
1911
1912         /* remove the closed state so when we call open it appears
1913          * we are coming from the probed state.
1914          */
1915         adapter->state = VNIC_PROBED;
1916
1917         reinit_completion(&adapter->init_done);
1918         rc = init_crq_queue(adapter);
1919         if (rc) {
1920                 netdev_err(adapter->netdev,
1921                            "Couldn't initialize crq. rc=%d\n", rc);
1922                 return rc;
1923         }
1924
1925         rc = ibmvnic_init(adapter);
1926         if (rc)
1927                 return rc;
1928
1929         /* If the adapter was in PROBE state prior to the reset,
1930          * exit here.
1931          */
1932         if (reset_state == VNIC_PROBED)
1933                 return 0;
1934
1935         rc = ibmvnic_login(netdev);
1936         if (rc) {
1937                 adapter->state = VNIC_PROBED;
1938                 return 0;
1939         }
1940
1941         rc = init_resources(adapter);
1942         if (rc)
1943                 return rc;
1944
1945         ibmvnic_disable_irqs(adapter);
1946         adapter->state = VNIC_CLOSED;
1947
1948         if (reset_state == VNIC_CLOSED)
1949                 return 0;
1950
1951         rc = __ibmvnic_open(netdev);
1952         if (rc) {
1953                 if (list_empty(&adapter->rwi_list))
1954                         adapter->state = VNIC_CLOSED;
1955                 else
1956                         adapter->state = reset_state;
1957
1958                 return 0;
1959         }
1960
1961         netif_carrier_on(netdev);
1962
1963         return 0;
1964 }
1965
1966 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
1967 {
1968         struct ibmvnic_rwi *rwi;
1969         unsigned long flags;
1970
1971         spin_lock_irqsave(&adapter->rwi_lock, flags);
1972
1973         if (!list_empty(&adapter->rwi_list)) {
1974                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
1975                                        list);
1976                 list_del(&rwi->list);
1977         } else {
1978                 rwi = NULL;
1979         }
1980
1981         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
1982         return rwi;
1983 }
1984
1985 static void free_all_rwi(struct ibmvnic_adapter *adapter)
1986 {
1987         struct ibmvnic_rwi *rwi;
1988
1989         rwi = get_next_rwi(adapter);
1990         while (rwi) {
1991                 kfree(rwi);
1992                 rwi = get_next_rwi(adapter);
1993         }
1994 }
1995
1996 static void __ibmvnic_reset(struct work_struct *work)
1997 {
1998         struct ibmvnic_rwi *rwi;
1999         struct ibmvnic_adapter *adapter;
2000         struct net_device *netdev;
2001         bool we_lock_rtnl = false;
2002         u32 reset_state;
2003         int rc = 0;
2004
2005         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2006         netdev = adapter->netdev;
2007
2008         /* netif_set_real_num_xx_queues needs to take rtnl lock here
2009          * unless wait_for_reset is set, in which case the rtnl lock
2010          * has already been taken before initializing the reset
2011          */
2012         if (!adapter->wait_for_reset) {
2013                 rtnl_lock();
2014                 we_lock_rtnl = true;
2015         }
2016         reset_state = adapter->state;
2017
2018         rwi = get_next_rwi(adapter);
2019         while (rwi) {
2020                 if (adapter->state == VNIC_REMOVING ||
2021                     adapter->state == VNIC_REMOVED) {
2022                         kfree(rwi);
2023                         rc = EBUSY;
2024                         break;
2025                 }
2026
2027                 if (adapter->force_reset_recovery) {
2028                         adapter->force_reset_recovery = false;
2029                         rc = do_hard_reset(adapter, rwi, reset_state);
2030                 } else {
2031                         rc = do_reset(adapter, rwi, reset_state);
2032                 }
2033                 kfree(rwi);
2034                 if (rc && rc != IBMVNIC_INIT_FAILED &&
2035                     !adapter->force_reset_recovery)
2036                         break;
2037
2038                 rwi = get_next_rwi(adapter);
2039         }
2040
2041         if (adapter->wait_for_reset) {
2042                 adapter->wait_for_reset = false;
2043                 adapter->reset_done_rc = rc;
2044                 complete(&adapter->reset_done);
2045         }
2046
2047         if (rc) {
2048                 netdev_dbg(adapter->netdev, "Reset failed\n");
2049                 free_all_rwi(adapter);
2050         }
2051
2052         adapter->resetting = false;
2053         if (we_lock_rtnl)
2054                 rtnl_unlock();
2055 }
2056
2057 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2058                          enum ibmvnic_reset_reason reason)
2059 {
2060         struct list_head *entry, *tmp_entry;
2061         struct ibmvnic_rwi *rwi, *tmp;
2062         struct net_device *netdev = adapter->netdev;
2063         unsigned long flags;
2064         int ret;
2065
2066         if (adapter->state == VNIC_REMOVING ||
2067             adapter->state == VNIC_REMOVED ||
2068             adapter->failover_pending) {
2069                 ret = EBUSY;
2070                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2071                 goto err;
2072         }
2073
2074         if (adapter->state == VNIC_PROBING) {
2075                 netdev_warn(netdev, "Adapter reset during probe\n");
2076                 ret = adapter->init_done_rc = EAGAIN;
2077                 goto err;
2078         }
2079
2080         spin_lock_irqsave(&adapter->rwi_lock, flags);
2081
2082         list_for_each(entry, &adapter->rwi_list) {
2083                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2084                 if (tmp->reset_reason == reason) {
2085                         netdev_dbg(netdev, "Skipping matching reset\n");
2086                         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2087                         ret = EBUSY;
2088                         goto err;
2089                 }
2090         }
2091
2092         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2093         if (!rwi) {
2094                 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2095                 ibmvnic_close(netdev);
2096                 ret = ENOMEM;
2097                 goto err;
2098         }
2099         /* if we just received a transport event,
2100          * flush reset queue and process this reset
2101          */
2102         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2103                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
2104                         list_del(entry);
2105                         kfree(list_entry(entry, struct ibmvnic_rwi, list));
2106                 }
2107         }
2108         rwi->reset_reason = reason;
2109         list_add_tail(&rwi->list, &adapter->rwi_list);
2110         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2111         adapter->resetting = true;
2112         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2113         schedule_work(&adapter->ibmvnic_reset);
2114
2115         return 0;
2116 err:
2117         if (adapter->wait_for_reset)
2118                 adapter->wait_for_reset = false;
2119         return -ret;
2120 }
2121
2122 static void ibmvnic_tx_timeout(struct net_device *dev)
2123 {
2124         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2125
2126         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2127 }
2128
2129 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2130                                   struct ibmvnic_rx_buff *rx_buff)
2131 {
2132         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2133
2134         rx_buff->skb = NULL;
2135
2136         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2137         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2138
2139         atomic_dec(&pool->available);
2140 }
2141
2142 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2143 {
2144         struct net_device *netdev = napi->dev;
2145         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2146         int scrq_num = (int)(napi - adapter->napi);
2147         int frames_processed = 0;
2148
2149 restart_poll:
2150         while (frames_processed < budget) {
2151                 struct sk_buff *skb;
2152                 struct ibmvnic_rx_buff *rx_buff;
2153                 union sub_crq *next;
2154                 u32 length;
2155                 u16 offset;
2156                 u8 flags = 0;
2157
2158                 if (unlikely(adapter->resetting &&
2159                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2160                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2161                         napi_complete_done(napi, frames_processed);
2162                         return frames_processed;
2163                 }
2164
2165                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
2166                         break;
2167                 /* The queue entry at the current index is peeked at above
2168                  * to determine that there is a valid descriptor awaiting
2169                  * processing. We want to be sure that the current slot
2170                  * holds a valid descriptor before reading its contents.
2171                  */
2172                 dma_rmb();
2173                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
2174                 rx_buff =
2175                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2176                                                           rx_comp.correlator);
2177                 /* do error checking */
2178                 if (next->rx_comp.rc) {
2179                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2180                                    be16_to_cpu(next->rx_comp.rc));
2181                         /* free the entry */
2182                         next->rx_comp.first = 0;
2183                         dev_kfree_skb_any(rx_buff->skb);
2184                         remove_buff_from_pool(adapter, rx_buff);
2185                         continue;
2186                 } else if (!rx_buff->skb) {
2187                         /* free the entry */
2188                         next->rx_comp.first = 0;
2189                         remove_buff_from_pool(adapter, rx_buff);
2190                         continue;
2191                 }
2192
2193                 length = be32_to_cpu(next->rx_comp.len);
2194                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2195                 flags = next->rx_comp.flags;
2196                 skb = rx_buff->skb;
2197                 /* load long_term_buff before copying to skb */
2198                 dma_rmb();
2199                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2200                                         length);
2201
2202                 /* VLAN Header has been stripped by the system firmware and
2203                  * needs to be inserted by the driver
2204                  */
2205                 if (adapter->rx_vlan_header_insertion &&
2206                     (flags & IBMVNIC_VLAN_STRIPPED))
2207                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2208                                                ntohs(next->rx_comp.vlan_tci));
2209
2210                 /* free the entry */
2211                 next->rx_comp.first = 0;
2212                 remove_buff_from_pool(adapter, rx_buff);
2213
2214                 skb_put(skb, length);
2215                 skb->protocol = eth_type_trans(skb, netdev);
2216                 skb_record_rx_queue(skb, scrq_num);
2217
2218                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2219                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2220                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2221                 }
2222
2223                 length = skb->len;
2224                 napi_gro_receive(napi, skb); /* send it up */
2225                 netdev->stats.rx_packets++;
2226                 netdev->stats.rx_bytes += length;
2227                 adapter->rx_stats_buffers[scrq_num].packets++;
2228                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2229                 frames_processed++;
2230         }
2231
2232         if (adapter->state != VNIC_CLOSING)
2233                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2234
2235         if (frames_processed < budget) {
2236                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2237                 napi_complete_done(napi, frames_processed);
2238                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
2239                     napi_reschedule(napi)) {
2240                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2241                         goto restart_poll;
2242                 }
2243         }
2244         return frames_processed;
2245 }
2246
2247 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2248 {
2249         int rc, ret;
2250
2251         adapter->fallback.mtu = adapter->req_mtu;
2252         adapter->fallback.rx_queues = adapter->req_rx_queues;
2253         adapter->fallback.tx_queues = adapter->req_tx_queues;
2254         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2255         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2256
2257         init_completion(&adapter->reset_done);
2258         adapter->wait_for_reset = true;
2259         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2260         if (rc)
2261                 return rc;
2262         wait_for_completion(&adapter->reset_done);
2263
2264         ret = 0;
2265         if (adapter->reset_done_rc) {
2266                 ret = -EIO;
2267                 adapter->desired.mtu = adapter->fallback.mtu;
2268                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2269                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2270                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2271                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2272
2273                 init_completion(&adapter->reset_done);
2274                 adapter->wait_for_reset = true;
2275                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2276                 if (rc)
2277                         return ret;
2278                 wait_for_completion(&adapter->reset_done);
2279         }
2280         adapter->wait_for_reset = false;
2281
2282         return ret;
2283 }
2284
2285 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2286 {
2287         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2288
2289         adapter->desired.mtu = new_mtu + ETH_HLEN;
2290
2291         return wait_for_reset(adapter);
2292 }
2293
2294 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2295                                                 struct net_device *dev,
2296                                                 netdev_features_t features)
2297 {
2298         /* Some backing hardware adapters can not
2299          * handle packets with a MSS less than 224
2300          * or with only one segment.
2301          */
2302         if (skb_is_gso(skb)) {
2303                 if (skb_shinfo(skb)->gso_size < 224 ||
2304                     skb_shinfo(skb)->gso_segs == 1)
2305                         features &= ~NETIF_F_GSO_MASK;
2306         }
2307
2308         return features;
2309 }
2310
2311 static const struct net_device_ops ibmvnic_netdev_ops = {
2312         .ndo_open               = ibmvnic_open,
2313         .ndo_stop               = ibmvnic_close,
2314         .ndo_start_xmit         = ibmvnic_xmit,
2315         .ndo_set_rx_mode        = ibmvnic_set_multi,
2316         .ndo_set_mac_address    = ibmvnic_set_mac,
2317         .ndo_validate_addr      = eth_validate_addr,
2318         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2319         .ndo_change_mtu         = ibmvnic_change_mtu,
2320         .ndo_features_check     = ibmvnic_features_check,
2321 };
2322
2323 /* ethtool functions */
2324
2325 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2326                                       struct ethtool_link_ksettings *cmd)
2327 {
2328         u32 supported, advertising;
2329
2330         supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
2331                           SUPPORTED_FIBRE);
2332         advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
2333                             ADVERTISED_FIBRE);
2334         cmd->base.speed = SPEED_1000;
2335         cmd->base.duplex = DUPLEX_FULL;
2336         cmd->base.port = PORT_FIBRE;
2337         cmd->base.phy_address = 0;
2338         cmd->base.autoneg = AUTONEG_ENABLE;
2339
2340         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
2341                                                 supported);
2342         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
2343                                                 advertising);
2344
2345         return 0;
2346 }
2347
2348 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2349                                 struct ethtool_drvinfo *info)
2350 {
2351         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2352
2353         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2354         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2355         strlcpy(info->fw_version, adapter->fw_version,
2356                 sizeof(info->fw_version));
2357 }
2358
2359 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2360 {
2361         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2362
2363         return adapter->msg_enable;
2364 }
2365
2366 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2367 {
2368         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2369
2370         adapter->msg_enable = data;
2371 }
2372
2373 static u32 ibmvnic_get_link(struct net_device *netdev)
2374 {
2375         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2376
2377         /* Don't need to send a query because we request a logical link up at
2378          * init and then we wait for link state indications
2379          */
2380         return adapter->logical_link_state;
2381 }
2382
2383 static void ibmvnic_get_ringparam(struct net_device *netdev,
2384                                   struct ethtool_ringparam *ring)
2385 {
2386         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2387
2388         ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2389         ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2390         ring->rx_mini_max_pending = 0;
2391         ring->rx_jumbo_max_pending = 0;
2392         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2393         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2394         ring->rx_mini_pending = 0;
2395         ring->rx_jumbo_pending = 0;
2396 }
2397
2398 static int ibmvnic_set_ringparam(struct net_device *netdev,
2399                                  struct ethtool_ringparam *ring)
2400 {
2401         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2402
2403         if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
2404             ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
2405                 netdev_err(netdev, "Invalid request.\n");
2406                 netdev_err(netdev, "Max tx buffers = %llu\n",
2407                            adapter->max_rx_add_entries_per_subcrq);
2408                 netdev_err(netdev, "Max rx buffers = %llu\n",
2409                            adapter->max_tx_entries_per_subcrq);
2410                 return -EINVAL;
2411         }
2412
2413         adapter->desired.rx_entries = ring->rx_pending;
2414         adapter->desired.tx_entries = ring->tx_pending;
2415
2416         return wait_for_reset(adapter);
2417 }
2418
2419 static void ibmvnic_get_channels(struct net_device *netdev,
2420                                  struct ethtool_channels *channels)
2421 {
2422         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2423
2424         channels->max_rx = adapter->max_rx_queues;
2425         channels->max_tx = adapter->max_tx_queues;
2426         channels->max_other = 0;
2427         channels->max_combined = 0;
2428         channels->rx_count = adapter->req_rx_queues;
2429         channels->tx_count = adapter->req_tx_queues;
2430         channels->other_count = 0;
2431         channels->combined_count = 0;
2432 }
2433
2434 static int ibmvnic_set_channels(struct net_device *netdev,
2435                                 struct ethtool_channels *channels)
2436 {
2437         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2438
2439         adapter->desired.rx_queues = channels->rx_count;
2440         adapter->desired.tx_queues = channels->tx_count;
2441
2442         return wait_for_reset(adapter);
2443 }
2444
2445 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2446 {
2447         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2448         int i;
2449
2450         if (stringset != ETH_SS_STATS)
2451                 return;
2452
2453         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
2454                 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2455
2456         for (i = 0; i < adapter->req_tx_queues; i++) {
2457                 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2458                 data += ETH_GSTRING_LEN;
2459
2460                 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2461                 data += ETH_GSTRING_LEN;
2462
2463                 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
2464                 data += ETH_GSTRING_LEN;
2465         }
2466
2467         for (i = 0; i < adapter->req_rx_queues; i++) {
2468                 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2469                 data += ETH_GSTRING_LEN;
2470
2471                 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2472                 data += ETH_GSTRING_LEN;
2473
2474                 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2475                 data += ETH_GSTRING_LEN;
2476         }
2477 }
2478
2479 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2480 {
2481         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2482
2483         switch (sset) {
2484         case ETH_SS_STATS:
2485                 return ARRAY_SIZE(ibmvnic_stats) +
2486                        adapter->req_tx_queues * NUM_TX_STATS +
2487                        adapter->req_rx_queues * NUM_RX_STATS;
2488         default:
2489                 return -EOPNOTSUPP;
2490         }
2491 }
2492
2493 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2494                                       struct ethtool_stats *stats, u64 *data)
2495 {
2496         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2497         union ibmvnic_crq crq;
2498         int i, j;
2499         int rc;
2500
2501         memset(&crq, 0, sizeof(crq));
2502         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2503         crq.request_statistics.cmd = REQUEST_STATISTICS;
2504         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2505         crq.request_statistics.len =
2506             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2507
2508         /* Wait for data to be written */
2509         init_completion(&adapter->stats_done);
2510         rc = ibmvnic_send_crq(adapter, &crq);
2511         if (rc)
2512                 return;
2513         wait_for_completion(&adapter->stats_done);
2514
2515         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2516                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2517                                                 ibmvnic_stats[i].offset));
2518
2519         for (j = 0; j < adapter->req_tx_queues; j++) {
2520                 data[i] = adapter->tx_stats_buffers[j].packets;
2521                 i++;
2522                 data[i] = adapter->tx_stats_buffers[j].bytes;
2523                 i++;
2524                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2525                 i++;
2526         }
2527
2528         for (j = 0; j < adapter->req_rx_queues; j++) {
2529                 data[i] = adapter->rx_stats_buffers[j].packets;
2530                 i++;
2531                 data[i] = adapter->rx_stats_buffers[j].bytes;
2532                 i++;
2533                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2534                 i++;
2535         }
2536 }
2537
2538 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2539         .get_drvinfo            = ibmvnic_get_drvinfo,
2540         .get_msglevel           = ibmvnic_get_msglevel,
2541         .set_msglevel           = ibmvnic_set_msglevel,
2542         .get_link               = ibmvnic_get_link,
2543         .get_ringparam          = ibmvnic_get_ringparam,
2544         .set_ringparam          = ibmvnic_set_ringparam,
2545         .get_channels           = ibmvnic_get_channels,
2546         .set_channels           = ibmvnic_set_channels,
2547         .get_strings            = ibmvnic_get_strings,
2548         .get_sset_count         = ibmvnic_get_sset_count,
2549         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2550         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2551 };
2552
2553 /* Routines for managing CRQs/sCRQs  */
2554
2555 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2556                                    struct ibmvnic_sub_crq_queue *scrq)
2557 {
2558         int rc;
2559
2560         if (scrq->irq) {
2561                 free_irq(scrq->irq, scrq);
2562                 irq_dispose_mapping(scrq->irq);
2563                 scrq->irq = 0;
2564         }
2565
2566         memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2567         atomic_set(&scrq->used, 0);
2568         scrq->cur = 0;
2569
2570         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2571                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2572         return rc;
2573 }
2574
2575 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2576 {
2577         int i, rc;
2578
2579         if (!adapter->tx_scrq || !adapter->rx_scrq)
2580                 return -EINVAL;
2581
2582         for (i = 0; i < adapter->req_tx_queues; i++) {
2583                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2584                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2585                 if (rc)
2586                         return rc;
2587         }
2588
2589         for (i = 0; i < adapter->req_rx_queues; i++) {
2590                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2591                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2592                 if (rc)
2593                         return rc;
2594         }
2595
2596         return rc;
2597 }
2598
2599 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2600                                   struct ibmvnic_sub_crq_queue *scrq,
2601                                   bool do_h_free)
2602 {
2603         struct device *dev = &adapter->vdev->dev;
2604         long rc;
2605
2606         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2607
2608         if (do_h_free) {
2609                 /* Close the sub-crqs */
2610                 do {
2611                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2612                                                 adapter->vdev->unit_address,
2613                                                 scrq->crq_num);
2614                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2615
2616                 if (rc) {
2617                         netdev_err(adapter->netdev,
2618                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2619                                    scrq->crq_num, rc);
2620                 }
2621         }
2622
2623         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2624                          DMA_BIDIRECTIONAL);
2625         free_pages((unsigned long)scrq->msgs, 2);
2626         kfree(scrq);
2627 }
2628
2629 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2630                                                         *adapter)
2631 {
2632         struct device *dev = &adapter->vdev->dev;
2633         struct ibmvnic_sub_crq_queue *scrq;
2634         int rc;
2635
2636         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2637         if (!scrq)
2638                 return NULL;
2639
2640         scrq->msgs =
2641                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2642         if (!scrq->msgs) {
2643                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2644                 goto zero_page_failed;
2645         }
2646
2647         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2648                                          DMA_BIDIRECTIONAL);
2649         if (dma_mapping_error(dev, scrq->msg_token)) {
2650                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2651                 goto map_failed;
2652         }
2653
2654         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2655                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2656
2657         if (rc == H_RESOURCE)
2658                 rc = ibmvnic_reset_crq(adapter);
2659
2660         if (rc == H_CLOSED) {
2661                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2662         } else if (rc) {
2663                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2664                 goto reg_failed;
2665         }
2666
2667         scrq->adapter = adapter;
2668         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2669         spin_lock_init(&scrq->lock);
2670
2671         netdev_dbg(adapter->netdev,
2672                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2673                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2674
2675         return scrq;
2676
2677 reg_failed:
2678         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2679                          DMA_BIDIRECTIONAL);
2680 map_failed:
2681         free_pages((unsigned long)scrq->msgs, 2);
2682 zero_page_failed:
2683         kfree(scrq);
2684
2685         return NULL;
2686 }
2687
2688 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
2689 {
2690         int i;
2691
2692         if (adapter->tx_scrq) {
2693                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
2694                         if (!adapter->tx_scrq[i])
2695                                 continue;
2696
2697                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
2698                                    i);
2699                         if (adapter->tx_scrq[i]->irq) {
2700                                 free_irq(adapter->tx_scrq[i]->irq,
2701                                          adapter->tx_scrq[i]);
2702                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
2703                                 adapter->tx_scrq[i]->irq = 0;
2704                         }
2705
2706                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
2707                                               do_h_free);
2708                 }
2709
2710                 kfree(adapter->tx_scrq);
2711                 adapter->tx_scrq = NULL;
2712                 adapter->num_active_tx_scrqs = 0;
2713         }
2714
2715         if (adapter->rx_scrq) {
2716                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
2717                         if (!adapter->rx_scrq[i])
2718                                 continue;
2719
2720                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
2721                                    i);
2722                         if (adapter->rx_scrq[i]->irq) {
2723                                 free_irq(adapter->rx_scrq[i]->irq,
2724                                          adapter->rx_scrq[i]);
2725                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
2726                                 adapter->rx_scrq[i]->irq = 0;
2727                         }
2728
2729                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
2730                                               do_h_free);
2731                 }
2732
2733                 kfree(adapter->rx_scrq);
2734                 adapter->rx_scrq = NULL;
2735                 adapter->num_active_rx_scrqs = 0;
2736         }
2737 }
2738
2739 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
2740                             struct ibmvnic_sub_crq_queue *scrq)
2741 {
2742         struct device *dev = &adapter->vdev->dev;
2743         unsigned long rc;
2744
2745         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2746                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2747         if (rc)
2748                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
2749                         scrq->hw_irq, rc);
2750         return rc;
2751 }
2752
2753 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
2754                            struct ibmvnic_sub_crq_queue *scrq)
2755 {
2756         struct device *dev = &adapter->vdev->dev;
2757         unsigned long rc;
2758
2759         if (scrq->hw_irq > 0x100000000ULL) {
2760                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
2761                 return 1;
2762         }
2763
2764         if (adapter->resetting &&
2765             adapter->reset_reason == VNIC_RESET_MOBILITY) {
2766                 u64 val = (0xff000000) | scrq->hw_irq;
2767
2768                 rc = plpar_hcall_norets(H_EOI, val);
2769                 if (rc)
2770                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
2771                                 val, rc);
2772         }
2773
2774         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2775                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2776         if (rc)
2777                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
2778                         scrq->hw_irq, rc);
2779         return rc;
2780 }
2781
2782 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
2783                                struct ibmvnic_sub_crq_queue *scrq)
2784 {
2785         struct device *dev = &adapter->vdev->dev;
2786         struct ibmvnic_tx_pool *tx_pool;
2787         struct ibmvnic_tx_buff *txbuff;
2788         union sub_crq *next;
2789         int index;
2790         int i, j;
2791
2792 restart_loop:
2793         while (pending_scrq(adapter, scrq)) {
2794                 unsigned int pool = scrq->pool_index;
2795                 int num_entries = 0;
2796
2797                 /* The queue entry at the current index is peeked at above
2798                  * to determine that there is a valid descriptor awaiting
2799                  * processing. We want to be sure that the current slot
2800                  * holds a valid descriptor before reading its contents.
2801                  */
2802                 dma_rmb();
2803
2804                 next = ibmvnic_next_scrq(adapter, scrq);
2805                 for (i = 0; i < next->tx_comp.num_comps; i++) {
2806                         if (next->tx_comp.rcs[i])
2807                                 dev_err(dev, "tx error %x\n",
2808                                         next->tx_comp.rcs[i]);
2809                         index = be32_to_cpu(next->tx_comp.correlators[i]);
2810                         if (index & IBMVNIC_TSO_POOL_MASK) {
2811                                 tx_pool = &adapter->tso_pool[pool];
2812                                 index &= ~IBMVNIC_TSO_POOL_MASK;
2813                         } else {
2814                                 tx_pool = &adapter->tx_pool[pool];
2815                         }
2816
2817                         txbuff = &tx_pool->tx_buff[index];
2818
2819                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
2820                                 if (!txbuff->data_dma[j])
2821                                         continue;
2822
2823                                 txbuff->data_dma[j] = 0;
2824                         }
2825
2826                         if (txbuff->last_frag) {
2827                                 dev_kfree_skb_any(txbuff->skb);
2828                                 txbuff->skb = NULL;
2829                         }
2830
2831                         num_entries += txbuff->num_entries;
2832
2833                         tx_pool->free_map[tx_pool->producer_index] = index;
2834                         tx_pool->producer_index =
2835                                 (tx_pool->producer_index + 1) %
2836                                         tx_pool->num_buffers;
2837                 }
2838                 /* remove tx_comp scrq*/
2839                 next->tx_comp.first = 0;
2840
2841                 if (atomic_sub_return(num_entries, &scrq->used) <=
2842                     (adapter->req_tx_entries_per_subcrq / 2) &&
2843                     __netif_subqueue_stopped(adapter->netdev,
2844                                              scrq->pool_index)) {
2845                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
2846                         netdev_dbg(adapter->netdev, "Started queue %d\n",
2847                                    scrq->pool_index);
2848                 }
2849         }
2850
2851         enable_scrq_irq(adapter, scrq);
2852
2853         if (pending_scrq(adapter, scrq)) {
2854                 disable_scrq_irq(adapter, scrq);
2855                 goto restart_loop;
2856         }
2857
2858         return 0;
2859 }
2860
2861 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
2862 {
2863         struct ibmvnic_sub_crq_queue *scrq = instance;
2864         struct ibmvnic_adapter *adapter = scrq->adapter;
2865
2866         disable_scrq_irq(adapter, scrq);
2867         ibmvnic_complete_tx(adapter, scrq);
2868
2869         return IRQ_HANDLED;
2870 }
2871
2872 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
2873 {
2874         struct ibmvnic_sub_crq_queue *scrq = instance;
2875         struct ibmvnic_adapter *adapter = scrq->adapter;
2876
2877         /* When booting a kdump kernel we can hit pending interrupts
2878          * prior to completing driver initialization.
2879          */
2880         if (unlikely(adapter->state != VNIC_OPEN))
2881                 return IRQ_NONE;
2882
2883         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
2884
2885         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
2886                 disable_scrq_irq(adapter, scrq);
2887                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
2888         }
2889
2890         return IRQ_HANDLED;
2891 }
2892
2893 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
2894 {
2895         struct device *dev = &adapter->vdev->dev;
2896         struct ibmvnic_sub_crq_queue *scrq;
2897         int i = 0, j = 0;
2898         int rc = 0;
2899
2900         for (i = 0; i < adapter->req_tx_queues; i++) {
2901                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
2902                            i);
2903                 scrq = adapter->tx_scrq[i];
2904                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2905
2906                 if (!scrq->irq) {
2907                         rc = -EINVAL;
2908                         dev_err(dev, "Error mapping irq\n");
2909                         goto req_tx_irq_failed;
2910                 }
2911
2912                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
2913                                  0, "ibmvnic_tx", scrq);
2914
2915                 if (rc) {
2916                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
2917                                 scrq->irq, rc);
2918                         irq_dispose_mapping(scrq->irq);
2919                         goto req_tx_irq_failed;
2920                 }
2921         }
2922
2923         for (i = 0; i < adapter->req_rx_queues; i++) {
2924                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
2925                            i);
2926                 scrq = adapter->rx_scrq[i];
2927                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2928                 if (!scrq->irq) {
2929                         rc = -EINVAL;
2930                         dev_err(dev, "Error mapping irq\n");
2931                         goto req_rx_irq_failed;
2932                 }
2933                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
2934                                  0, "ibmvnic_rx", scrq);
2935                 if (rc) {
2936                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
2937                                 scrq->irq, rc);
2938                         irq_dispose_mapping(scrq->irq);
2939                         goto req_rx_irq_failed;
2940                 }
2941         }
2942         return rc;
2943
2944 req_rx_irq_failed:
2945         for (j = 0; j < i; j++) {
2946                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
2947                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2948         }
2949         i = adapter->req_tx_queues;
2950 req_tx_irq_failed:
2951         for (j = 0; j < i; j++) {
2952                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
2953                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
2954         }
2955         release_sub_crqs(adapter, 1);
2956         return rc;
2957 }
2958
2959 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
2960 {
2961         struct device *dev = &adapter->vdev->dev;
2962         struct ibmvnic_sub_crq_queue **allqueues;
2963         int registered_queues = 0;
2964         int total_queues;
2965         int more = 0;
2966         int i;
2967
2968         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
2969
2970         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
2971         if (!allqueues)
2972                 return -1;
2973
2974         for (i = 0; i < total_queues; i++) {
2975                 allqueues[i] = init_sub_crq_queue(adapter);
2976                 if (!allqueues[i]) {
2977                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
2978                         break;
2979                 }
2980                 registered_queues++;
2981         }
2982
2983         /* Make sure we were able to register the minimum number of queues */
2984         if (registered_queues <
2985             adapter->min_tx_queues + adapter->min_rx_queues) {
2986                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
2987                 goto tx_failed;
2988         }
2989
2990         /* Distribute the failed allocated queues*/
2991         for (i = 0; i < total_queues - registered_queues + more ; i++) {
2992                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
2993                 switch (i % 3) {
2994                 case 0:
2995                         if (adapter->req_rx_queues > adapter->min_rx_queues)
2996                                 adapter->req_rx_queues--;
2997                         else
2998                                 more++;
2999                         break;
3000                 case 1:
3001                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3002                                 adapter->req_tx_queues--;
3003                         else
3004                                 more++;
3005                         break;
3006                 }
3007         }
3008
3009         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3010                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3011         if (!adapter->tx_scrq)
3012                 goto tx_failed;
3013
3014         for (i = 0; i < adapter->req_tx_queues; i++) {
3015                 adapter->tx_scrq[i] = allqueues[i];
3016                 adapter->tx_scrq[i]->pool_index = i;
3017                 adapter->num_active_tx_scrqs++;
3018         }
3019
3020         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3021                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3022         if (!adapter->rx_scrq)
3023                 goto rx_failed;
3024
3025         for (i = 0; i < adapter->req_rx_queues; i++) {
3026                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3027                 adapter->rx_scrq[i]->scrq_num = i;
3028                 adapter->num_active_rx_scrqs++;
3029         }
3030
3031         kfree(allqueues);
3032         return 0;
3033
3034 rx_failed:
3035         kfree(adapter->tx_scrq);
3036         adapter->tx_scrq = NULL;
3037 tx_failed:
3038         for (i = 0; i < registered_queues; i++)
3039                 release_sub_crq_queue(adapter, allqueues[i], 1);
3040         kfree(allqueues);
3041         return -1;
3042 }
3043
3044 static void ibmvnic_send_req_caps(struct ibmvnic_adapter *adapter, int retry)
3045 {
3046         struct device *dev = &adapter->vdev->dev;
3047         union ibmvnic_crq crq;
3048         int max_entries;
3049         int cap_reqs;
3050
3051         /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
3052          * the PROMISC flag). Initialize this count upfront. When the tasklet
3053          * receives a response to all of these, it will send the next protocol
3054          * message (QUERY_IP_OFFLOAD).
3055          */
3056         if (!(adapter->netdev->flags & IFF_PROMISC) ||
3057             adapter->promisc_supported)
3058                 cap_reqs = 7;
3059         else
3060                 cap_reqs = 6;
3061
3062         if (!retry) {
3063                 /* Sub-CRQ entries are 32 byte long */
3064                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3065
3066                 atomic_set(&adapter->running_cap_crqs, cap_reqs);
3067
3068                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3069                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3070                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3071                         return;
3072                 }
3073
3074                 if (adapter->desired.mtu)
3075                         adapter->req_mtu = adapter->desired.mtu;
3076                 else
3077                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3078
3079                 if (!adapter->desired.tx_entries)
3080                         adapter->desired.tx_entries =
3081                                         adapter->max_tx_entries_per_subcrq;
3082                 if (!adapter->desired.rx_entries)
3083                         adapter->desired.rx_entries =
3084                                         adapter->max_rx_add_entries_per_subcrq;
3085
3086                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3087                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3088
3089                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3090                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3091                         adapter->desired.tx_entries = max_entries;
3092                 }
3093
3094                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3095                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3096                         adapter->desired.rx_entries = max_entries;
3097                 }
3098
3099                 if (adapter->desired.tx_entries)
3100                         adapter->req_tx_entries_per_subcrq =
3101                                         adapter->desired.tx_entries;
3102                 else
3103                         adapter->req_tx_entries_per_subcrq =
3104                                         adapter->max_tx_entries_per_subcrq;
3105
3106                 if (adapter->desired.rx_entries)
3107                         adapter->req_rx_add_entries_per_subcrq =
3108                                         adapter->desired.rx_entries;
3109                 else
3110                         adapter->req_rx_add_entries_per_subcrq =
3111                                         adapter->max_rx_add_entries_per_subcrq;
3112
3113                 if (adapter->desired.tx_queues)
3114                         adapter->req_tx_queues =
3115                                         adapter->desired.tx_queues;
3116                 else
3117                         adapter->req_tx_queues =
3118                                         adapter->opt_tx_comp_sub_queues;
3119
3120                 if (adapter->desired.rx_queues)
3121                         adapter->req_rx_queues =
3122                                         adapter->desired.rx_queues;
3123                 else
3124                         adapter->req_rx_queues =
3125                                         adapter->opt_rx_comp_queues;
3126
3127                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3128         } else {
3129                 atomic_add(cap_reqs, &adapter->running_cap_crqs);
3130         }
3131         memset(&crq, 0, sizeof(crq));
3132         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3133         crq.request_capability.cmd = REQUEST_CAPABILITY;
3134
3135         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3136         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3137         cap_reqs--;
3138         ibmvnic_send_crq(adapter, &crq);
3139
3140         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3141         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3142         cap_reqs--;
3143         ibmvnic_send_crq(adapter, &crq);
3144
3145         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3146         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3147         cap_reqs--;
3148         ibmvnic_send_crq(adapter, &crq);
3149
3150         crq.request_capability.capability =
3151             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3152         crq.request_capability.number =
3153             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3154         cap_reqs--;
3155         ibmvnic_send_crq(adapter, &crq);
3156
3157         crq.request_capability.capability =
3158             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3159         crq.request_capability.number =
3160             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3161         cap_reqs--;
3162         ibmvnic_send_crq(adapter, &crq);
3163
3164         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3165         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3166         cap_reqs--;
3167         ibmvnic_send_crq(adapter, &crq);
3168
3169         if (adapter->netdev->flags & IFF_PROMISC) {
3170                 if (adapter->promisc_supported) {
3171                         crq.request_capability.capability =
3172                             cpu_to_be16(PROMISC_REQUESTED);
3173                         crq.request_capability.number = cpu_to_be64(1);
3174                         cap_reqs--;
3175                         ibmvnic_send_crq(adapter, &crq);
3176                 }
3177         } else {
3178                 crq.request_capability.capability =
3179                     cpu_to_be16(PROMISC_REQUESTED);
3180                 crq.request_capability.number = cpu_to_be64(0);
3181                 cap_reqs--;
3182                 ibmvnic_send_crq(adapter, &crq);
3183         }
3184
3185         /* Keep at end to catch any discrepancy between expected and actual
3186          * CRQs sent.
3187          */
3188         WARN_ON(cap_reqs != 0);
3189 }
3190
3191 static int pending_scrq(struct ibmvnic_adapter *adapter,
3192                         struct ibmvnic_sub_crq_queue *scrq)
3193 {
3194         union sub_crq *entry = &scrq->msgs[scrq->cur];
3195
3196         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3197                 return 1;
3198         else
3199                 return 0;
3200 }
3201
3202 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3203                                         struct ibmvnic_sub_crq_queue *scrq)
3204 {
3205         union sub_crq *entry;
3206         unsigned long flags;
3207
3208         spin_lock_irqsave(&scrq->lock, flags);
3209         entry = &scrq->msgs[scrq->cur];
3210         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3211                 if (++scrq->cur == scrq->size)
3212                         scrq->cur = 0;
3213         } else {
3214                 entry = NULL;
3215         }
3216         spin_unlock_irqrestore(&scrq->lock, flags);
3217
3218         /* Ensure that the entire buffer descriptor has been
3219          * loaded before reading its contents
3220          */
3221         dma_rmb();
3222
3223         return entry;
3224 }
3225
3226 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3227 {
3228         struct ibmvnic_crq_queue *queue = &adapter->crq;
3229         union ibmvnic_crq *crq;
3230
3231         crq = &queue->msgs[queue->cur];
3232         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3233                 if (++queue->cur == queue->size)
3234                         queue->cur = 0;
3235         } else {
3236                 crq = NULL;
3237         }
3238
3239         return crq;
3240 }
3241
3242 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3243 {
3244         switch (rc) {
3245         case H_PARAMETER:
3246                 dev_warn_ratelimited(dev,
3247                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3248                                      func, rc);
3249                 break;
3250         case H_CLOSED:
3251                 dev_warn_ratelimited(dev,
3252                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3253                                      func, rc);
3254                 break;
3255         default:
3256                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3257                 break;
3258         }
3259 }
3260
3261 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
3262                        union sub_crq *sub_crq)
3263 {
3264         unsigned int ua = adapter->vdev->unit_address;
3265         struct device *dev = &adapter->vdev->dev;
3266         u64 *u64_crq = (u64 *)sub_crq;
3267         int rc;
3268
3269         netdev_dbg(adapter->netdev,
3270                    "Sending sCRQ %016lx: %016lx %016lx %016lx %016lx\n",
3271                    (unsigned long int)cpu_to_be64(remote_handle),
3272                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3273                    (unsigned long int)cpu_to_be64(u64_crq[1]),
3274                    (unsigned long int)cpu_to_be64(u64_crq[2]),
3275                    (unsigned long int)cpu_to_be64(u64_crq[3]));
3276
3277         /* Make sure the hypervisor sees the complete request */
3278         mb();
3279
3280         rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
3281                                 cpu_to_be64(remote_handle),
3282                                 cpu_to_be64(u64_crq[0]),
3283                                 cpu_to_be64(u64_crq[1]),
3284                                 cpu_to_be64(u64_crq[2]),
3285                                 cpu_to_be64(u64_crq[3]));
3286
3287         if (rc)
3288                 print_subcrq_error(dev, rc, __func__);
3289
3290         return rc;
3291 }
3292
3293 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3294                                 u64 remote_handle, u64 ioba, u64 num_entries)
3295 {
3296         unsigned int ua = adapter->vdev->unit_address;
3297         struct device *dev = &adapter->vdev->dev;
3298         int rc;
3299
3300         /* Make sure the hypervisor sees the complete request */
3301         mb();
3302         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3303                                 cpu_to_be64(remote_handle),
3304                                 ioba, num_entries);
3305
3306         if (rc)
3307                 print_subcrq_error(dev, rc, __func__);
3308
3309         return rc;
3310 }
3311
3312 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3313                             union ibmvnic_crq *crq)
3314 {
3315         unsigned int ua = adapter->vdev->unit_address;
3316         struct device *dev = &adapter->vdev->dev;
3317         u64 *u64_crq = (u64 *)crq;
3318         int rc;
3319
3320         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3321                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3322                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3323
3324         if (!adapter->crq.active &&
3325             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3326                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3327                 return -EINVAL;
3328         }
3329
3330         /* Make sure the hypervisor sees the complete request */
3331         mb();
3332
3333         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3334                                 cpu_to_be64(u64_crq[0]),
3335                                 cpu_to_be64(u64_crq[1]));
3336
3337         if (rc) {
3338                 if (rc == H_CLOSED) {
3339                         dev_warn(dev, "CRQ Queue closed\n");
3340                         if (adapter->resetting)
3341                                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3342                 }
3343
3344                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3345         }
3346
3347         return rc;
3348 }
3349
3350 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3351 {
3352         union ibmvnic_crq crq;
3353
3354         memset(&crq, 0, sizeof(crq));
3355         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3356         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3357         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3358
3359         return ibmvnic_send_crq(adapter, &crq);
3360 }
3361
3362 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3363 {
3364         union ibmvnic_crq crq;
3365
3366         memset(&crq, 0, sizeof(crq));
3367         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3368         crq.version_exchange.cmd = VERSION_EXCHANGE;
3369         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3370
3371         return ibmvnic_send_crq(adapter, &crq);
3372 }
3373
3374 struct vnic_login_client_data {
3375         u8      type;
3376         __be16  len;
3377         char    name[];
3378 } __packed;
3379
3380 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3381 {
3382         int len;
3383
3384         /* Calculate the amount of buffer space needed for the
3385          * vnic client data in the login buffer. There are four entries,
3386          * OS name, LPAR name, device name, and a null last entry.
3387          */
3388         len = 4 * sizeof(struct vnic_login_client_data);
3389         len += 6; /* "Linux" plus NULL */
3390         len += strlen(utsname()->nodename) + 1;
3391         len += strlen(adapter->netdev->name) + 1;
3392
3393         return len;
3394 }
3395
3396 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3397                                  struct vnic_login_client_data *vlcd)
3398 {
3399         const char *os_name = "Linux";
3400         int len;
3401
3402         /* Type 1 - LPAR OS */
3403         vlcd->type = 1;
3404         len = strlen(os_name) + 1;
3405         vlcd->len = cpu_to_be16(len);
3406         strncpy(vlcd->name, os_name, len);
3407         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3408
3409         /* Type 2 - LPAR name */
3410         vlcd->type = 2;
3411         len = strlen(utsname()->nodename) + 1;
3412         vlcd->len = cpu_to_be16(len);
3413         strncpy(vlcd->name, utsname()->nodename, len);
3414         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3415
3416         /* Type 3 - device name */
3417         vlcd->type = 3;
3418         len = strlen(adapter->netdev->name) + 1;
3419         vlcd->len = cpu_to_be16(len);
3420         strncpy(vlcd->name, adapter->netdev->name, len);
3421 }
3422
3423 static int send_login(struct ibmvnic_adapter *adapter)
3424 {
3425         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3426         struct ibmvnic_login_buffer *login_buffer;
3427         struct device *dev = &adapter->vdev->dev;
3428         dma_addr_t rsp_buffer_token;
3429         dma_addr_t buffer_token;
3430         size_t rsp_buffer_size;
3431         union ibmvnic_crq crq;
3432         size_t buffer_size;
3433         __be64 *tx_list_p;
3434         __be64 *rx_list_p;
3435         int client_data_len;
3436         struct vnic_login_client_data *vlcd;
3437         int i;
3438
3439         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3440                 netdev_err(adapter->netdev,
3441                            "RX or TX queues are not allocated, device login failed\n");
3442                 return -1;
3443         }
3444
3445         release_login_rsp_buffer(adapter);
3446         client_data_len = vnic_client_data_len(adapter);
3447
3448         buffer_size =
3449             sizeof(struct ibmvnic_login_buffer) +
3450             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3451             client_data_len;
3452
3453         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3454         if (!login_buffer)
3455                 goto buf_alloc_failed;
3456
3457         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3458                                       DMA_TO_DEVICE);
3459         if (dma_mapping_error(dev, buffer_token)) {
3460                 dev_err(dev, "Couldn't map login buffer\n");
3461                 goto buf_map_failed;
3462         }
3463
3464         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3465                           sizeof(u64) * adapter->req_tx_queues +
3466                           sizeof(u64) * adapter->req_rx_queues +
3467                           sizeof(u64) * adapter->req_rx_queues +
3468                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3469
3470         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3471         if (!login_rsp_buffer)
3472                 goto buf_rsp_alloc_failed;
3473
3474         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3475                                           rsp_buffer_size, DMA_FROM_DEVICE);
3476         if (dma_mapping_error(dev, rsp_buffer_token)) {
3477                 dev_err(dev, "Couldn't map login rsp buffer\n");
3478                 goto buf_rsp_map_failed;
3479         }
3480
3481         adapter->login_buf = login_buffer;
3482         adapter->login_buf_token = buffer_token;
3483         adapter->login_buf_sz = buffer_size;
3484         adapter->login_rsp_buf = login_rsp_buffer;
3485         adapter->login_rsp_buf_token = rsp_buffer_token;
3486         adapter->login_rsp_buf_sz = rsp_buffer_size;
3487
3488         login_buffer->len = cpu_to_be32(buffer_size);
3489         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3490         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3491         login_buffer->off_txcomp_subcrqs =
3492             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3493         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3494         login_buffer->off_rxcomp_subcrqs =
3495             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3496                         sizeof(u64) * adapter->req_tx_queues);
3497         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3498         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3499
3500         tx_list_p = (__be64 *)((char *)login_buffer +
3501                                       sizeof(struct ibmvnic_login_buffer));
3502         rx_list_p = (__be64 *)((char *)login_buffer +
3503                                       sizeof(struct ibmvnic_login_buffer) +
3504                                       sizeof(u64) * adapter->req_tx_queues);
3505
3506         for (i = 0; i < adapter->req_tx_queues; i++) {
3507                 if (adapter->tx_scrq[i]) {
3508                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3509                                                    crq_num);
3510                 }
3511         }
3512
3513         for (i = 0; i < adapter->req_rx_queues; i++) {
3514                 if (adapter->rx_scrq[i]) {
3515                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3516                                                    crq_num);
3517                 }
3518         }
3519
3520         /* Insert vNIC login client data */
3521         vlcd = (struct vnic_login_client_data *)
3522                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3523         login_buffer->client_data_offset =
3524                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3525         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3526
3527         vnic_add_client_data(adapter, vlcd);
3528
3529         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3530         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3531                 netdev_dbg(adapter->netdev, "%016lx\n",
3532                            ((unsigned long int *)(adapter->login_buf))[i]);
3533         }
3534
3535         memset(&crq, 0, sizeof(crq));
3536         crq.login.first = IBMVNIC_CRQ_CMD;
3537         crq.login.cmd = LOGIN;
3538         crq.login.ioba = cpu_to_be32(buffer_token);
3539         crq.login.len = cpu_to_be32(buffer_size);
3540         ibmvnic_send_crq(adapter, &crq);
3541
3542         return 0;
3543
3544 buf_rsp_map_failed:
3545         kfree(login_rsp_buffer);
3546 buf_rsp_alloc_failed:
3547         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3548 buf_map_failed:
3549         kfree(login_buffer);
3550 buf_alloc_failed:
3551         return -1;
3552 }
3553
3554 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3555                             u32 len, u8 map_id)
3556 {
3557         union ibmvnic_crq crq;
3558
3559         memset(&crq, 0, sizeof(crq));
3560         crq.request_map.first = IBMVNIC_CRQ_CMD;
3561         crq.request_map.cmd = REQUEST_MAP;
3562         crq.request_map.map_id = map_id;
3563         crq.request_map.ioba = cpu_to_be32(addr);
3564         crq.request_map.len = cpu_to_be32(len);
3565         return ibmvnic_send_crq(adapter, &crq);
3566 }
3567
3568 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3569 {
3570         union ibmvnic_crq crq;
3571
3572         memset(&crq, 0, sizeof(crq));
3573         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3574         crq.request_unmap.cmd = REQUEST_UNMAP;
3575         crq.request_unmap.map_id = map_id;
3576         return ibmvnic_send_crq(adapter, &crq);
3577 }
3578
3579 static void send_map_query(struct ibmvnic_adapter *adapter)
3580 {
3581         union ibmvnic_crq crq;
3582
3583         memset(&crq, 0, sizeof(crq));
3584         crq.query_map.first = IBMVNIC_CRQ_CMD;
3585         crq.query_map.cmd = QUERY_MAP;
3586         ibmvnic_send_crq(adapter, &crq);
3587 }
3588
3589 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3590 static void send_cap_queries(struct ibmvnic_adapter *adapter)
3591 {
3592         union ibmvnic_crq crq;
3593         int cap_reqs;
3594
3595         /* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
3596          * upfront. When the tasklet receives a response to all of these, it
3597          * can send out the next protocol messaage (REQUEST_CAPABILITY).
3598          */
3599         cap_reqs = 25;
3600
3601         atomic_set(&adapter->running_cap_crqs, cap_reqs);
3602
3603         memset(&crq, 0, sizeof(crq));
3604         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3605         crq.query_capability.cmd = QUERY_CAPABILITY;
3606
3607         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3608         ibmvnic_send_crq(adapter, &crq);
3609         cap_reqs--;
3610
3611         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3612         ibmvnic_send_crq(adapter, &crq);
3613         cap_reqs--;
3614
3615         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3616         ibmvnic_send_crq(adapter, &crq);
3617         cap_reqs--;
3618
3619         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3620         ibmvnic_send_crq(adapter, &crq);
3621         cap_reqs--;
3622
3623         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3624         ibmvnic_send_crq(adapter, &crq);
3625         cap_reqs--;
3626
3627         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3628         ibmvnic_send_crq(adapter, &crq);
3629         cap_reqs--;
3630
3631         crq.query_capability.capability =
3632             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3633         ibmvnic_send_crq(adapter, &crq);
3634         cap_reqs--;
3635
3636         crq.query_capability.capability =
3637             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3638         ibmvnic_send_crq(adapter, &crq);
3639         cap_reqs--;
3640
3641         crq.query_capability.capability =
3642             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3643         ibmvnic_send_crq(adapter, &crq);
3644         cap_reqs--;
3645
3646         crq.query_capability.capability =
3647             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
3648         ibmvnic_send_crq(adapter, &crq);
3649         cap_reqs--;
3650
3651         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
3652         ibmvnic_send_crq(adapter, &crq);
3653         cap_reqs--;
3654
3655         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
3656         ibmvnic_send_crq(adapter, &crq);
3657         cap_reqs--;
3658
3659         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
3660         ibmvnic_send_crq(adapter, &crq);
3661         cap_reqs--;
3662
3663         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
3664         ibmvnic_send_crq(adapter, &crq);
3665         cap_reqs--;
3666
3667         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
3668         ibmvnic_send_crq(adapter, &crq);
3669         cap_reqs--;
3670
3671         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
3672         ibmvnic_send_crq(adapter, &crq);
3673         cap_reqs--;
3674
3675         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
3676         ibmvnic_send_crq(adapter, &crq);
3677         cap_reqs--;
3678
3679         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
3680         ibmvnic_send_crq(adapter, &crq);
3681         cap_reqs--;
3682
3683         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
3684         ibmvnic_send_crq(adapter, &crq);
3685         cap_reqs--;
3686
3687         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
3688         ibmvnic_send_crq(adapter, &crq);
3689         cap_reqs--;
3690
3691         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
3692         ibmvnic_send_crq(adapter, &crq);
3693         cap_reqs--;
3694
3695         crq.query_capability.capability =
3696                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
3697         ibmvnic_send_crq(adapter, &crq);
3698         cap_reqs--;
3699
3700         crq.query_capability.capability =
3701                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
3702         ibmvnic_send_crq(adapter, &crq);
3703         cap_reqs--;
3704
3705         crq.query_capability.capability =
3706                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
3707         ibmvnic_send_crq(adapter, &crq);
3708         cap_reqs--;
3709
3710         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
3711
3712         ibmvnic_send_crq(adapter, &crq);
3713         cap_reqs--;
3714
3715         /* Keep at end to catch any discrepancy between expected and actual
3716          * CRQs sent.
3717          */
3718         WARN_ON(cap_reqs != 0);
3719 }
3720
3721 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
3722                                 struct ibmvnic_adapter *adapter)
3723 {
3724         struct device *dev = &adapter->vdev->dev;
3725
3726         if (crq->get_vpd_size_rsp.rc.code) {
3727                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
3728                         crq->get_vpd_size_rsp.rc.code);
3729                 complete(&adapter->fw_done);
3730                 return;
3731         }
3732
3733         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
3734         complete(&adapter->fw_done);
3735 }
3736
3737 static void handle_vpd_rsp(union ibmvnic_crq *crq,
3738                            struct ibmvnic_adapter *adapter)
3739 {
3740         struct device *dev = &adapter->vdev->dev;
3741         unsigned char *substr = NULL;
3742         u8 fw_level_len = 0;
3743
3744         memset(adapter->fw_version, 0, 32);
3745
3746         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
3747                          DMA_FROM_DEVICE);
3748
3749         if (crq->get_vpd_rsp.rc.code) {
3750                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
3751                         crq->get_vpd_rsp.rc.code);
3752                 goto complete;
3753         }
3754
3755         /* get the position of the firmware version info
3756          * located after the ASCII 'RM' substring in the buffer
3757          */
3758         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
3759         if (!substr) {
3760                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
3761                 goto complete;
3762         }
3763
3764         /* get length of firmware level ASCII substring */
3765         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
3766                 fw_level_len = *(substr + 2);
3767         } else {
3768                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
3769                 goto complete;
3770         }
3771
3772         /* copy firmware version string from vpd into adapter */
3773         if ((substr + 3 + fw_level_len) <
3774             (adapter->vpd->buff + adapter->vpd->len)) {
3775                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
3776         } else {
3777                 dev_info(dev, "FW substr extrapolated VPD buff\n");
3778         }
3779
3780 complete:
3781         if (adapter->fw_version[0] == '\0')
3782                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
3783         complete(&adapter->fw_done);
3784 }
3785
3786 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
3787 {
3788         struct device *dev = &adapter->vdev->dev;
3789         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
3790         union ibmvnic_crq crq;
3791         int i;
3792
3793         dma_unmap_single(dev, adapter->ip_offload_tok,
3794                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
3795
3796         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
3797         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
3798                 netdev_dbg(adapter->netdev, "%016lx\n",
3799                            ((unsigned long int *)(buf))[i]);
3800
3801         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
3802         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
3803         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
3804                    buf->tcp_ipv4_chksum);
3805         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
3806                    buf->tcp_ipv6_chksum);
3807         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
3808                    buf->udp_ipv4_chksum);
3809         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
3810                    buf->udp_ipv6_chksum);
3811         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
3812                    buf->large_tx_ipv4);
3813         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
3814                    buf->large_tx_ipv6);
3815         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
3816                    buf->large_rx_ipv4);
3817         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
3818                    buf->large_rx_ipv6);
3819         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
3820                    buf->max_ipv4_header_size);
3821         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
3822                    buf->max_ipv6_header_size);
3823         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
3824                    buf->max_tcp_header_size);
3825         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
3826                    buf->max_udp_header_size);
3827         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
3828                    buf->max_large_tx_size);
3829         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
3830                    buf->max_large_rx_size);
3831         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
3832                    buf->ipv6_extension_header);
3833         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
3834                    buf->tcp_pseudosum_req);
3835         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
3836                    buf->num_ipv6_ext_headers);
3837         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
3838                    buf->off_ipv6_ext_headers);
3839
3840         adapter->ip_offload_ctrl_tok =
3841             dma_map_single(dev, &adapter->ip_offload_ctrl,
3842                            sizeof(adapter->ip_offload_ctrl), DMA_TO_DEVICE);
3843
3844         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
3845                 dev_err(dev, "Couldn't map ip offload control buffer\n");
3846                 return;
3847         }
3848
3849         adapter->ip_offload_ctrl.len =
3850             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3851         adapter->ip_offload_ctrl.version = cpu_to_be32(INITIAL_VERSION_IOB);
3852         adapter->ip_offload_ctrl.ipv4_chksum = buf->ipv4_chksum;
3853         adapter->ip_offload_ctrl.ipv6_chksum = buf->ipv6_chksum;
3854         adapter->ip_offload_ctrl.tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
3855         adapter->ip_offload_ctrl.udp_ipv4_chksum = buf->udp_ipv4_chksum;
3856         adapter->ip_offload_ctrl.tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
3857         adapter->ip_offload_ctrl.udp_ipv6_chksum = buf->udp_ipv6_chksum;
3858         adapter->ip_offload_ctrl.large_tx_ipv4 = buf->large_tx_ipv4;
3859         adapter->ip_offload_ctrl.large_tx_ipv6 = buf->large_tx_ipv6;
3860
3861         /* large_rx disabled for now, additional features needed */
3862         adapter->ip_offload_ctrl.large_rx_ipv4 = 0;
3863         adapter->ip_offload_ctrl.large_rx_ipv6 = 0;
3864
3865         adapter->netdev->features = NETIF_F_SG | NETIF_F_GSO;
3866
3867         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
3868                 adapter->netdev->features |= NETIF_F_IP_CSUM;
3869
3870         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
3871                 adapter->netdev->features |= NETIF_F_IPV6_CSUM;
3872
3873         if ((adapter->netdev->features &
3874             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
3875                 adapter->netdev->features |= NETIF_F_RXCSUM;
3876
3877         if (buf->large_tx_ipv4)
3878                 adapter->netdev->features |= NETIF_F_TSO;
3879         if (buf->large_tx_ipv6)
3880                 adapter->netdev->features |= NETIF_F_TSO6;
3881
3882         adapter->netdev->hw_features |= adapter->netdev->features;
3883
3884         memset(&crq, 0, sizeof(crq));
3885         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
3886         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
3887         crq.control_ip_offload.len =
3888             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3889         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
3890         ibmvnic_send_crq(adapter, &crq);
3891 }
3892
3893 static const char *ibmvnic_fw_err_cause(u16 cause)
3894 {
3895         switch (cause) {
3896         case ADAPTER_PROBLEM:
3897                 return "adapter problem";
3898         case BUS_PROBLEM:
3899                 return "bus problem";
3900         case FW_PROBLEM:
3901                 return "firmware problem";
3902         case DD_PROBLEM:
3903                 return "device driver problem";
3904         case EEH_RECOVERY:
3905                 return "EEH recovery";
3906         case FW_UPDATED:
3907                 return "firmware updated";
3908         case LOW_MEMORY:
3909                 return "low Memory";
3910         default:
3911                 return "unknown";
3912         }
3913 }
3914
3915 static void handle_error_indication(union ibmvnic_crq *crq,
3916                                     struct ibmvnic_adapter *adapter)
3917 {
3918         struct device *dev = &adapter->vdev->dev;
3919         u16 cause;
3920
3921         cause = be16_to_cpu(crq->error_indication.error_cause);
3922
3923         dev_warn_ratelimited(dev,
3924                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
3925                              crq->error_indication.flags
3926                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
3927                              ibmvnic_fw_err_cause(cause));
3928
3929         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
3930                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3931         else
3932                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
3933 }
3934
3935 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
3936                                  struct ibmvnic_adapter *adapter)
3937 {
3938         struct net_device *netdev = adapter->netdev;
3939         struct device *dev = &adapter->vdev->dev;
3940         long rc;
3941
3942         rc = crq->change_mac_addr_rsp.rc.code;
3943         if (rc) {
3944                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
3945                 goto out;
3946         }
3947         memcpy(netdev->dev_addr, &crq->change_mac_addr_rsp.mac_addr[0],
3948                ETH_ALEN);
3949 out:
3950         complete(&adapter->fw_done);
3951         return rc;
3952 }
3953
3954 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
3955                                    struct ibmvnic_adapter *adapter)
3956 {
3957         struct device *dev = &adapter->vdev->dev;
3958         u64 *req_value;
3959         char *name;
3960
3961         atomic_dec(&adapter->running_cap_crqs);
3962         netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
3963                    atomic_read(&adapter->running_cap_crqs));
3964         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
3965         case REQ_TX_QUEUES:
3966                 req_value = &adapter->req_tx_queues;
3967                 name = "tx";
3968                 break;
3969         case REQ_RX_QUEUES:
3970                 req_value = &adapter->req_rx_queues;
3971                 name = "rx";
3972                 break;
3973         case REQ_RX_ADD_QUEUES:
3974                 req_value = &adapter->req_rx_add_queues;
3975                 name = "rx_add";
3976                 break;
3977         case REQ_TX_ENTRIES_PER_SUBCRQ:
3978                 req_value = &adapter->req_tx_entries_per_subcrq;
3979                 name = "tx_entries_per_subcrq";
3980                 break;
3981         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
3982                 req_value = &adapter->req_rx_add_entries_per_subcrq;
3983                 name = "rx_add_entries_per_subcrq";
3984                 break;
3985         case REQ_MTU:
3986                 req_value = &adapter->req_mtu;
3987                 name = "mtu";
3988                 break;
3989         case PROMISC_REQUESTED:
3990                 req_value = &adapter->promisc;
3991                 name = "promisc";
3992                 break;
3993         default:
3994                 dev_err(dev, "Got invalid cap request rsp %d\n",
3995                         crq->request_capability.capability);
3996                 return;
3997         }
3998
3999         switch (crq->request_capability_rsp.rc.code) {
4000         case SUCCESS:
4001                 break;
4002         case PARTIALSUCCESS:
4003                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4004                          *req_value,
4005                          (long int)be64_to_cpu(crq->request_capability_rsp.
4006                                                number), name);
4007
4008                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4009                     REQ_MTU) {
4010                         pr_err("mtu of %llu is not supported. Reverting.\n",
4011                                *req_value);
4012                         *req_value = adapter->fallback.mtu;
4013                 } else {
4014                         *req_value =
4015                                 be64_to_cpu(crq->request_capability_rsp.number);
4016                 }
4017
4018                 ibmvnic_send_req_caps(adapter, 1);
4019                 return;
4020         default:
4021                 dev_err(dev, "Error %d in request cap rsp\n",
4022                         crq->request_capability_rsp.rc.code);
4023                 return;
4024         }
4025
4026         /* Done receiving requested capabilities, query IP offload support */
4027         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4028                 union ibmvnic_crq newcrq;
4029                 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4030                 struct ibmvnic_query_ip_offload_buffer *ip_offload_buf =
4031                     &adapter->ip_offload_buf;
4032
4033                 adapter->wait_capability = false;
4034                 adapter->ip_offload_tok = dma_map_single(dev, ip_offload_buf,
4035                                                          buf_sz,
4036                                                          DMA_FROM_DEVICE);
4037
4038                 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4039                         if (!firmware_has_feature(FW_FEATURE_CMO))
4040                                 dev_err(dev, "Couldn't map offload buffer\n");
4041                         return;
4042                 }
4043
4044                 memset(&newcrq, 0, sizeof(newcrq));
4045                 newcrq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4046                 newcrq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4047                 newcrq.query_ip_offload.len = cpu_to_be32(buf_sz);
4048                 newcrq.query_ip_offload.ioba =
4049                     cpu_to_be32(adapter->ip_offload_tok);
4050
4051                 ibmvnic_send_crq(adapter, &newcrq);
4052         }
4053 }
4054
4055 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4056                             struct ibmvnic_adapter *adapter)
4057 {
4058         struct device *dev = &adapter->vdev->dev;
4059         struct net_device *netdev = adapter->netdev;
4060         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4061         struct ibmvnic_login_buffer *login = adapter->login_buf;
4062         int i;
4063
4064         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4065                          DMA_TO_DEVICE);
4066         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4067                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4068
4069         /* If the number of queues requested can't be allocated by the
4070          * server, the login response will return with code 1. We will need
4071          * to resend the login buffer with fewer queues requested.
4072          */
4073         if (login_rsp_crq->generic.rc.code) {
4074                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4075                 complete(&adapter->init_done);
4076                 return 0;
4077         }
4078
4079         if (adapter->failover_pending) {
4080                 adapter->init_done_rc = -EAGAIN;
4081                 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4082                 complete(&adapter->init_done);
4083                 /* login response buffer will be released on reset */
4084                 return 0;
4085         }
4086
4087         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4088
4089         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4090         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4091                 netdev_dbg(adapter->netdev, "%016lx\n",
4092                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4093         }
4094
4095         /* Sanity checks */
4096         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4097             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4098              adapter->req_rx_add_queues !=
4099              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4100                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4101                 ibmvnic_remove(adapter->vdev);
4102                 return -EIO;
4103         }
4104         release_login_buffer(adapter);
4105         complete(&adapter->init_done);
4106
4107         return 0;
4108 }
4109
4110 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4111                                      struct ibmvnic_adapter *adapter)
4112 {
4113         struct device *dev = &adapter->vdev->dev;
4114         long rc;
4115
4116         rc = crq->request_unmap_rsp.rc.code;
4117         if (rc)
4118                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4119 }
4120
4121 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4122                                  struct ibmvnic_adapter *adapter)
4123 {
4124         struct net_device *netdev = adapter->netdev;
4125         struct device *dev = &adapter->vdev->dev;
4126         long rc;
4127
4128         rc = crq->query_map_rsp.rc.code;
4129         if (rc) {
4130                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4131                 return;
4132         }
4133         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4134                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4135                    crq->query_map_rsp.free_pages);
4136 }
4137
4138 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4139                                  struct ibmvnic_adapter *adapter)
4140 {
4141         struct net_device *netdev = adapter->netdev;
4142         struct device *dev = &adapter->vdev->dev;
4143         long rc;
4144
4145         atomic_dec(&adapter->running_cap_crqs);
4146         netdev_dbg(netdev, "Outstanding queries: %d\n",
4147                    atomic_read(&adapter->running_cap_crqs));
4148         rc = crq->query_capability.rc.code;
4149         if (rc) {
4150                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4151                 goto out;
4152         }
4153
4154         switch (be16_to_cpu(crq->query_capability.capability)) {
4155         case MIN_TX_QUEUES:
4156                 adapter->min_tx_queues =
4157                     be64_to_cpu(crq->query_capability.number);
4158                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4159                            adapter->min_tx_queues);
4160                 break;
4161         case MIN_RX_QUEUES:
4162                 adapter->min_rx_queues =
4163                     be64_to_cpu(crq->query_capability.number);
4164                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4165                            adapter->min_rx_queues);
4166                 break;
4167         case MIN_RX_ADD_QUEUES:
4168                 adapter->min_rx_add_queues =
4169                     be64_to_cpu(crq->query_capability.number);
4170                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4171                            adapter->min_rx_add_queues);
4172                 break;
4173         case MAX_TX_QUEUES:
4174                 adapter->max_tx_queues =
4175                     be64_to_cpu(crq->query_capability.number);
4176                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4177                            adapter->max_tx_queues);
4178                 break;
4179         case MAX_RX_QUEUES:
4180                 adapter->max_rx_queues =
4181                     be64_to_cpu(crq->query_capability.number);
4182                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4183                            adapter->max_rx_queues);
4184                 break;
4185         case MAX_RX_ADD_QUEUES:
4186                 adapter->max_rx_add_queues =
4187                     be64_to_cpu(crq->query_capability.number);
4188                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4189                            adapter->max_rx_add_queues);
4190                 break;
4191         case MIN_TX_ENTRIES_PER_SUBCRQ:
4192                 adapter->min_tx_entries_per_subcrq =
4193                     be64_to_cpu(crq->query_capability.number);
4194                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4195                            adapter->min_tx_entries_per_subcrq);
4196                 break;
4197         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4198                 adapter->min_rx_add_entries_per_subcrq =
4199                     be64_to_cpu(crq->query_capability.number);
4200                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4201                            adapter->min_rx_add_entries_per_subcrq);
4202                 break;
4203         case MAX_TX_ENTRIES_PER_SUBCRQ:
4204                 adapter->max_tx_entries_per_subcrq =
4205                     be64_to_cpu(crq->query_capability.number);
4206                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4207                            adapter->max_tx_entries_per_subcrq);
4208                 break;
4209         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4210                 adapter->max_rx_add_entries_per_subcrq =
4211                     be64_to_cpu(crq->query_capability.number);
4212                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4213                            adapter->max_rx_add_entries_per_subcrq);
4214                 break;
4215         case TCP_IP_OFFLOAD:
4216                 adapter->tcp_ip_offload =
4217                     be64_to_cpu(crq->query_capability.number);
4218                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4219                            adapter->tcp_ip_offload);
4220                 break;
4221         case PROMISC_SUPPORTED:
4222                 adapter->promisc_supported =
4223                     be64_to_cpu(crq->query_capability.number);
4224                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4225                            adapter->promisc_supported);
4226                 break;
4227         case MIN_MTU:
4228                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4229                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4230                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4231                 break;
4232         case MAX_MTU:
4233                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4234                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4235                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4236                 break;
4237         case MAX_MULTICAST_FILTERS:
4238                 adapter->max_multicast_filters =
4239                     be64_to_cpu(crq->query_capability.number);
4240                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4241                            adapter->max_multicast_filters);
4242                 break;
4243         case VLAN_HEADER_INSERTION:
4244                 adapter->vlan_header_insertion =
4245                     be64_to_cpu(crq->query_capability.number);
4246                 if (adapter->vlan_header_insertion)
4247                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4248                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4249                            adapter->vlan_header_insertion);
4250                 break;
4251         case RX_VLAN_HEADER_INSERTION:
4252                 adapter->rx_vlan_header_insertion =
4253                     be64_to_cpu(crq->query_capability.number);
4254                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4255                            adapter->rx_vlan_header_insertion);
4256                 break;
4257         case MAX_TX_SG_ENTRIES:
4258                 adapter->max_tx_sg_entries =
4259                     be64_to_cpu(crq->query_capability.number);
4260                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4261                            adapter->max_tx_sg_entries);
4262                 break;
4263         case RX_SG_SUPPORTED:
4264                 adapter->rx_sg_supported =
4265                     be64_to_cpu(crq->query_capability.number);
4266                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4267                            adapter->rx_sg_supported);
4268                 break;
4269         case OPT_TX_COMP_SUB_QUEUES:
4270                 adapter->opt_tx_comp_sub_queues =
4271                     be64_to_cpu(crq->query_capability.number);
4272                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4273                            adapter->opt_tx_comp_sub_queues);
4274                 break;
4275         case OPT_RX_COMP_QUEUES:
4276                 adapter->opt_rx_comp_queues =
4277                     be64_to_cpu(crq->query_capability.number);
4278                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4279                            adapter->opt_rx_comp_queues);
4280                 break;
4281         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4282                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4283                     be64_to_cpu(crq->query_capability.number);
4284                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4285                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4286                 break;
4287         case OPT_TX_ENTRIES_PER_SUBCRQ:
4288                 adapter->opt_tx_entries_per_subcrq =
4289                     be64_to_cpu(crq->query_capability.number);
4290                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4291                            adapter->opt_tx_entries_per_subcrq);
4292                 break;
4293         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4294                 adapter->opt_rxba_entries_per_subcrq =
4295                     be64_to_cpu(crq->query_capability.number);
4296                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4297                            adapter->opt_rxba_entries_per_subcrq);
4298                 break;
4299         case TX_RX_DESC_REQ:
4300                 adapter->tx_rx_desc_req = crq->query_capability.number;
4301                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4302                            adapter->tx_rx_desc_req);
4303                 break;
4304
4305         default:
4306                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4307                            crq->query_capability.capability);
4308         }
4309
4310 out:
4311         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4312                 adapter->wait_capability = false;
4313                 ibmvnic_send_req_caps(adapter, 0);
4314         }
4315 }
4316
4317 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4318                                struct ibmvnic_adapter *adapter)
4319 {
4320         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4321         struct net_device *netdev = adapter->netdev;
4322         struct device *dev = &adapter->vdev->dev;
4323         u64 *u64_crq = (u64 *)crq;
4324         long rc;
4325
4326         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4327                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4328                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4329         switch (gen_crq->first) {
4330         case IBMVNIC_CRQ_INIT_RSP:
4331                 switch (gen_crq->cmd) {
4332                 case IBMVNIC_CRQ_INIT:
4333                         dev_info(dev, "Partner initialized\n");
4334                         adapter->from_passive_init = true;
4335                         adapter->failover_pending = false;
4336                         if (!completion_done(&adapter->init_done)) {
4337                                 complete(&adapter->init_done);
4338                                 adapter->init_done_rc = -EIO;
4339                         }
4340                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4341                         break;
4342                 case IBMVNIC_CRQ_INIT_COMPLETE:
4343                         dev_info(dev, "Partner initialization complete\n");
4344                         adapter->crq.active = true;
4345                         send_version_xchg(adapter);
4346                         break;
4347                 default:
4348                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4349                 }
4350                 return;
4351         case IBMVNIC_CRQ_XPORT_EVENT:
4352                 netif_carrier_off(netdev);
4353                 adapter->crq.active = false;
4354                 if (adapter->resetting)
4355                         adapter->force_reset_recovery = true;
4356                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4357                         dev_info(dev, "Migrated, re-enabling adapter\n");
4358                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4359                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4360                         dev_info(dev, "Backing device failover detected\n");
4361                         adapter->failover_pending = true;
4362                 } else {
4363                         /* The adapter lost the connection */
4364                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4365                                 gen_crq->cmd);
4366                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4367                 }
4368                 return;
4369         case IBMVNIC_CRQ_CMD_RSP:
4370                 break;
4371         default:
4372                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4373                         gen_crq->first);
4374                 return;
4375         }
4376
4377         switch (gen_crq->cmd) {
4378         case VERSION_EXCHANGE_RSP:
4379                 rc = crq->version_exchange_rsp.rc.code;
4380                 if (rc) {
4381                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4382                         break;
4383                 }
4384                 ibmvnic_version =
4385                             be16_to_cpu(crq->version_exchange_rsp.version);
4386                 dev_info(dev, "Partner protocol version is %d\n",
4387                          ibmvnic_version);
4388                 send_cap_queries(adapter);
4389                 break;
4390         case QUERY_CAPABILITY_RSP:
4391                 handle_query_cap_rsp(crq, adapter);
4392                 break;
4393         case QUERY_MAP_RSP:
4394                 handle_query_map_rsp(crq, adapter);
4395                 break;
4396         case REQUEST_MAP_RSP:
4397                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4398                 complete(&adapter->fw_done);
4399                 break;
4400         case REQUEST_UNMAP_RSP:
4401                 handle_request_unmap_rsp(crq, adapter);
4402                 break;
4403         case REQUEST_CAPABILITY_RSP:
4404                 handle_request_cap_rsp(crq, adapter);
4405                 break;
4406         case LOGIN_RSP:
4407                 netdev_dbg(netdev, "Got Login Response\n");
4408                 handle_login_rsp(crq, adapter);
4409                 break;
4410         case LOGICAL_LINK_STATE_RSP:
4411                 netdev_dbg(netdev,
4412                            "Got Logical Link State Response, state: %d rc: %d\n",
4413                            crq->logical_link_state_rsp.link_state,
4414                            crq->logical_link_state_rsp.rc.code);
4415                 adapter->logical_link_state =
4416                     crq->logical_link_state_rsp.link_state;
4417                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4418                 complete(&adapter->init_done);
4419                 break;
4420         case LINK_STATE_INDICATION:
4421                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4422                 adapter->phys_link_state =
4423                     crq->link_state_indication.phys_link_state;
4424                 adapter->logical_link_state =
4425                     crq->link_state_indication.logical_link_state;
4426                 break;
4427         case CHANGE_MAC_ADDR_RSP:
4428                 netdev_dbg(netdev, "Got MAC address change Response\n");
4429                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4430                 break;
4431         case ERROR_INDICATION:
4432                 netdev_dbg(netdev, "Got Error Indication\n");
4433                 handle_error_indication(crq, adapter);
4434                 break;
4435         case REQUEST_STATISTICS_RSP:
4436                 netdev_dbg(netdev, "Got Statistics Response\n");
4437                 complete(&adapter->stats_done);
4438                 break;
4439         case QUERY_IP_OFFLOAD_RSP:
4440                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4441                 handle_query_ip_offload_rsp(adapter);
4442                 break;
4443         case MULTICAST_CTRL_RSP:
4444                 netdev_dbg(netdev, "Got multicast control Response\n");
4445                 break;
4446         case CONTROL_IP_OFFLOAD_RSP:
4447                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4448                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4449                                  sizeof(adapter->ip_offload_ctrl),
4450                                  DMA_TO_DEVICE);
4451                 complete(&adapter->init_done);
4452                 break;
4453         case COLLECT_FW_TRACE_RSP:
4454                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4455                 complete(&adapter->fw_done);
4456                 break;
4457         case GET_VPD_SIZE_RSP:
4458                 handle_vpd_size_rsp(crq, adapter);
4459                 break;
4460         case GET_VPD_RSP:
4461                 handle_vpd_rsp(crq, adapter);
4462                 break;
4463         default:
4464                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4465                            gen_crq->cmd);
4466         }
4467 }
4468
4469 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
4470 {
4471         struct ibmvnic_adapter *adapter = instance;
4472
4473         tasklet_schedule(&adapter->tasklet);
4474         return IRQ_HANDLED;
4475 }
4476
4477 static void ibmvnic_tasklet(void *data)
4478 {
4479         struct ibmvnic_adapter *adapter = data;
4480         struct ibmvnic_crq_queue *queue = &adapter->crq;
4481         union ibmvnic_crq *crq;
4482         unsigned long flags;
4483         bool done = false;
4484
4485         spin_lock_irqsave(&queue->lock, flags);
4486         while (!done) {
4487                 /* Pull all the valid messages off the CRQ */
4488                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
4489                         /* This barrier makes sure ibmvnic_next_crq()'s
4490                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
4491                          * before ibmvnic_handle_crq()'s
4492                          * switch(gen_crq->first) and switch(gen_crq->cmd).
4493                          */
4494                         dma_rmb();
4495                         ibmvnic_handle_crq(crq, adapter);
4496                         crq->generic.first = 0;
4497                 }
4498         }
4499         /* if capabilities CRQ's were sent in this tasklet, the following
4500          * tasklet must wait until all responses are received
4501          */
4502         if (atomic_read(&adapter->running_cap_crqs) != 0)
4503                 adapter->wait_capability = true;
4504         spin_unlock_irqrestore(&queue->lock, flags);
4505 }
4506
4507 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
4508 {
4509         struct vio_dev *vdev = adapter->vdev;
4510         int rc;
4511
4512         do {
4513                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
4514         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
4515
4516         if (rc)
4517                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
4518
4519         return rc;
4520 }
4521
4522 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
4523 {
4524         struct ibmvnic_crq_queue *crq = &adapter->crq;
4525         struct device *dev = &adapter->vdev->dev;
4526         struct vio_dev *vdev = adapter->vdev;
4527         int rc;
4528
4529         /* Close the CRQ */
4530         do {
4531                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4532         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4533
4534         /* Clean out the queue */
4535         if (!crq->msgs)
4536                 return -EINVAL;
4537
4538         memset(crq->msgs, 0, PAGE_SIZE);
4539         crq->cur = 0;
4540         crq->active = false;
4541
4542         /* And re-open it again */
4543         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4544                                 crq->msg_token, PAGE_SIZE);
4545
4546         if (rc == H_CLOSED)
4547                 /* Adapter is good, but other end is not ready */
4548                 dev_warn(dev, "Partner adapter not ready\n");
4549         else if (rc != 0)
4550                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
4551
4552         return rc;
4553 }
4554
4555 static void release_crq_queue(struct ibmvnic_adapter *adapter)
4556 {
4557         struct ibmvnic_crq_queue *crq = &adapter->crq;
4558         struct vio_dev *vdev = adapter->vdev;
4559         long rc;
4560
4561         if (!crq->msgs)
4562                 return;
4563
4564         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
4565         free_irq(vdev->irq, adapter);
4566         tasklet_kill(&adapter->tasklet);
4567         do {
4568                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4569         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4570
4571         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
4572                          DMA_BIDIRECTIONAL);
4573         free_page((unsigned long)crq->msgs);
4574         crq->msgs = NULL;
4575         crq->active = false;
4576 }
4577
4578 static int init_crq_queue(struct ibmvnic_adapter *adapter)
4579 {
4580         struct ibmvnic_crq_queue *crq = &adapter->crq;
4581         struct device *dev = &adapter->vdev->dev;
4582         struct vio_dev *vdev = adapter->vdev;
4583         int rc, retrc = -ENOMEM;
4584
4585         if (crq->msgs)
4586                 return 0;
4587
4588         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
4589         /* Should we allocate more than one page? */
4590
4591         if (!crq->msgs)
4592                 return -ENOMEM;
4593
4594         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
4595         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
4596                                         DMA_BIDIRECTIONAL);
4597         if (dma_mapping_error(dev, crq->msg_token))
4598                 goto map_failed;
4599
4600         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4601                                 crq->msg_token, PAGE_SIZE);
4602
4603         if (rc == H_RESOURCE)
4604                 /* maybe kexecing and resource is busy. try a reset */
4605                 rc = ibmvnic_reset_crq(adapter);
4606         retrc = rc;
4607
4608         if (rc == H_CLOSED) {
4609                 dev_warn(dev, "Partner adapter not ready\n");
4610         } else if (rc) {
4611                 dev_warn(dev, "Error %d opening adapter\n", rc);
4612                 goto reg_crq_failed;
4613         }
4614
4615         retrc = 0;
4616
4617         tasklet_init(&adapter->tasklet, (void *)ibmvnic_tasklet,
4618                      (unsigned long)adapter);
4619
4620         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
4621         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, IBMVNIC_NAME,
4622                          adapter);
4623         if (rc) {
4624                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
4625                         vdev->irq, rc);
4626                 goto req_irq_failed;
4627         }
4628
4629         rc = vio_enable_interrupts(vdev);
4630         if (rc) {
4631                 dev_err(dev, "Error %d enabling interrupts\n", rc);
4632                 goto req_irq_failed;
4633         }
4634
4635         crq->cur = 0;
4636         spin_lock_init(&crq->lock);
4637
4638         /* process any CRQs that were queued before we enabled interrupts */
4639         tasklet_schedule(&adapter->tasklet);
4640
4641         return retrc;
4642
4643 req_irq_failed:
4644         tasklet_kill(&adapter->tasklet);
4645         do {
4646                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4647         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4648 reg_crq_failed:
4649         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
4650 map_failed:
4651         free_page((unsigned long)crq->msgs);
4652         crq->msgs = NULL;
4653         return retrc;
4654 }
4655
4656 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter)
4657 {
4658         struct device *dev = &adapter->vdev->dev;
4659         unsigned long timeout = msecs_to_jiffies(30000);
4660         u64 old_num_rx_queues, old_num_tx_queues;
4661         int rc;
4662
4663         adapter->from_passive_init = false;
4664
4665         old_num_rx_queues = adapter->req_rx_queues;
4666         old_num_tx_queues = adapter->req_tx_queues;
4667
4668         reinit_completion(&adapter->init_done);
4669         adapter->init_done_rc = 0;
4670         ibmvnic_send_crq_init(adapter);
4671         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
4672                 dev_err(dev, "Initialization sequence timed out\n");
4673                 return -1;
4674         }
4675
4676         if (adapter->init_done_rc) {
4677                 release_crq_queue(adapter);
4678                 return adapter->init_done_rc;
4679         }
4680
4681         if (adapter->from_passive_init) {
4682                 adapter->state = VNIC_OPEN;
4683                 adapter->from_passive_init = false;
4684                 return -1;
4685         }
4686
4687         if (adapter->resetting && !adapter->wait_for_reset &&
4688             adapter->reset_reason != VNIC_RESET_MOBILITY) {
4689                 if (adapter->req_rx_queues != old_num_rx_queues ||
4690                     adapter->req_tx_queues != old_num_tx_queues) {
4691                         release_sub_crqs(adapter, 0);
4692                         rc = init_sub_crqs(adapter);
4693                 } else {
4694                         /* no need to reinitialize completely, but we do
4695                          * need to clean up transmits that were in flight
4696                          * when we processed the reset.  Failure to do so
4697                          * will confound the upper layer, usually TCP, by
4698                          * creating the illusion of transmits that are
4699                          * awaiting completion.
4700                          */
4701                         clean_tx_pools(adapter);
4702
4703                         rc = reset_sub_crq_queues(adapter);
4704                 }
4705         } else {
4706                 rc = init_sub_crqs(adapter);
4707         }
4708
4709         if (rc) {
4710                 dev_err(dev, "Initialization of sub crqs failed\n");
4711                 release_crq_queue(adapter);
4712                 return rc;
4713         }
4714
4715         rc = init_sub_crq_irqs(adapter);
4716         if (rc) {
4717                 dev_err(dev, "Failed to initialize sub crq irqs\n");
4718                 release_crq_queue(adapter);
4719         }
4720
4721         return rc;
4722 }
4723
4724 static int ibmvnic_init(struct ibmvnic_adapter *adapter)
4725 {
4726         struct device *dev = &adapter->vdev->dev;
4727         unsigned long timeout = msecs_to_jiffies(30000);
4728         int rc;
4729
4730         adapter->from_passive_init = false;
4731
4732         adapter->init_done_rc = 0;
4733         ibmvnic_send_crq_init(adapter);
4734         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
4735                 dev_err(dev, "Initialization sequence timed out\n");
4736                 return -1;
4737         }
4738
4739         if (adapter->init_done_rc) {
4740                 release_crq_queue(adapter);
4741                 return adapter->init_done_rc;
4742         }
4743
4744         if (adapter->from_passive_init) {
4745                 adapter->state = VNIC_OPEN;
4746                 adapter->from_passive_init = false;
4747                 return -1;
4748         }
4749
4750         rc = init_sub_crqs(adapter);
4751         if (rc) {
4752                 dev_err(dev, "Initialization of sub crqs failed\n");
4753                 release_crq_queue(adapter);
4754                 return rc;
4755         }
4756
4757         rc = init_sub_crq_irqs(adapter);
4758         if (rc) {
4759                 dev_err(dev, "Failed to initialize sub crq irqs\n");
4760                 release_crq_queue(adapter);
4761         }
4762
4763         return rc;
4764 }
4765
4766 static struct device_attribute dev_attr_failover;
4767
4768 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
4769 {
4770         struct ibmvnic_adapter *adapter;
4771         struct net_device *netdev;
4772         unsigned char *mac_addr_p;
4773         int rc;
4774
4775         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
4776                 dev->unit_address);
4777
4778         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
4779                                                         VETH_MAC_ADDR, NULL);
4780         if (!mac_addr_p) {
4781                 dev_err(&dev->dev,
4782                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
4783                         __FILE__, __LINE__);
4784                 return 0;
4785         }
4786
4787         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
4788                                    IBMVNIC_MAX_QUEUES);
4789         if (!netdev)
4790                 return -ENOMEM;
4791
4792         adapter = netdev_priv(netdev);
4793         adapter->state = VNIC_PROBING;
4794         dev_set_drvdata(&dev->dev, netdev);
4795         adapter->vdev = dev;
4796         adapter->netdev = netdev;
4797
4798         ether_addr_copy(adapter->mac_addr, mac_addr_p);
4799         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4800         netdev->irq = dev->irq;
4801         netdev->netdev_ops = &ibmvnic_netdev_ops;
4802         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
4803         SET_NETDEV_DEV(netdev, &dev->dev);
4804
4805         spin_lock_init(&adapter->stats_lock);
4806
4807         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
4808         INIT_LIST_HEAD(&adapter->rwi_list);
4809         spin_lock_init(&adapter->rwi_lock);
4810         init_completion(&adapter->init_done);
4811         adapter->resetting = false;
4812
4813         adapter->mac_change_pending = false;
4814
4815         do {
4816                 rc = init_crq_queue(adapter);
4817                 if (rc) {
4818                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
4819                                 rc);
4820                         goto ibmvnic_init_fail;
4821                 }
4822
4823                 rc = ibmvnic_init(adapter);
4824                 if (rc && rc != EAGAIN)
4825                         goto ibmvnic_init_fail;
4826         } while (rc == EAGAIN);
4827
4828         rc = init_stats_buffers(adapter);
4829         if (rc)
4830                 goto ibmvnic_init_fail;
4831
4832         rc = init_stats_token(adapter);
4833         if (rc)
4834                 goto ibmvnic_stats_fail;
4835
4836         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4837         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4838         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4839
4840         rc = device_create_file(&dev->dev, &dev_attr_failover);
4841         if (rc)
4842                 goto ibmvnic_dev_file_err;
4843
4844         netif_carrier_off(netdev);
4845         rc = register_netdev(netdev);
4846         if (rc) {
4847                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
4848                 goto ibmvnic_register_fail;
4849         }
4850         dev_info(&dev->dev, "ibmvnic registered\n");
4851
4852         adapter->state = VNIC_PROBED;
4853
4854         adapter->wait_for_reset = false;
4855
4856         return 0;
4857
4858 ibmvnic_register_fail:
4859         device_remove_file(&dev->dev, &dev_attr_failover);
4860
4861 ibmvnic_dev_file_err:
4862         release_stats_token(adapter);
4863
4864 ibmvnic_stats_fail:
4865         release_stats_buffers(adapter);
4866
4867 ibmvnic_init_fail:
4868         release_sub_crqs(adapter, 1);
4869         release_crq_queue(adapter);
4870         free_netdev(netdev);
4871
4872         return rc;
4873 }
4874
4875 static int ibmvnic_remove(struct vio_dev *dev)
4876 {
4877         struct net_device *netdev = dev_get_drvdata(&dev->dev);
4878         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4879
4880         adapter->state = VNIC_REMOVING;
4881         rtnl_lock();
4882         unregister_netdevice(netdev);
4883
4884         release_resources(adapter);
4885         release_sub_crqs(adapter, 1);
4886         release_crq_queue(adapter);
4887
4888         release_stats_token(adapter);
4889         release_stats_buffers(adapter);
4890
4891         adapter->state = VNIC_REMOVED;
4892
4893         rtnl_unlock();
4894         device_remove_file(&dev->dev, &dev_attr_failover);
4895         free_netdev(netdev);
4896         dev_set_drvdata(&dev->dev, NULL);
4897
4898         return 0;
4899 }
4900
4901 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
4902                               const char *buf, size_t count)
4903 {
4904         struct net_device *netdev = dev_get_drvdata(dev);
4905         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4906         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
4907         __be64 session_token;
4908         long rc;
4909
4910         if (!sysfs_streq(buf, "1"))
4911                 return -EINVAL;
4912
4913         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
4914                          H_GET_SESSION_TOKEN, 0, 0, 0);
4915         if (rc) {
4916                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
4917                            rc);
4918                 return -EINVAL;
4919         }
4920
4921         session_token = (__be64)retbuf[0];
4922         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
4923                    be64_to_cpu(session_token));
4924         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4925                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
4926         if (rc) {
4927                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
4928                            rc);
4929                 return -EINVAL;
4930         }
4931
4932         return count;
4933 }
4934
4935 static DEVICE_ATTR_WO(failover);
4936
4937 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
4938 {
4939         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
4940         struct ibmvnic_adapter *adapter;
4941         struct iommu_table *tbl;
4942         unsigned long ret = 0;
4943         int i;
4944
4945         tbl = get_iommu_table_base(&vdev->dev);
4946
4947         /* netdev inits at probe time along with the structures we need below*/
4948         if (!netdev)
4949                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
4950
4951         adapter = netdev_priv(netdev);
4952
4953         ret += PAGE_SIZE; /* the crq message queue */
4954         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
4955
4956         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
4957                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
4958
4959         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4960              i++)
4961                 ret += adapter->rx_pool[i].size *
4962                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
4963
4964         return ret;
4965 }
4966
4967 static int ibmvnic_resume(struct device *dev)
4968 {
4969         struct net_device *netdev = dev_get_drvdata(dev);
4970         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4971
4972         if (adapter->state != VNIC_OPEN)
4973                 return 0;
4974
4975         tasklet_schedule(&adapter->tasklet);
4976
4977         return 0;
4978 }
4979
4980 static const struct vio_device_id ibmvnic_device_table[] = {
4981         {"network", "IBM,vnic"},
4982         {"", "" }
4983 };
4984 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
4985
4986 static const struct dev_pm_ops ibmvnic_pm_ops = {
4987         .resume = ibmvnic_resume
4988 };
4989
4990 static struct vio_driver ibmvnic_driver = {
4991         .id_table       = ibmvnic_device_table,
4992         .probe          = ibmvnic_probe,
4993         .remove         = ibmvnic_remove,
4994         .get_desired_dma = ibmvnic_get_desired_dma,
4995         .name           = ibmvnic_driver_name,
4996         .pm             = &ibmvnic_pm_ops,
4997 };
4998
4999 /* module functions */
5000 static int __init ibmvnic_module_init(void)
5001 {
5002         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5003                 IBMVNIC_DRIVER_VERSION);
5004
5005         return vio_register_driver(&ibmvnic_driver);
5006 }
5007
5008 static void __exit ibmvnic_module_exit(void)
5009 {
5010         vio_unregister_driver(&ibmvnic_driver);
5011 }
5012
5013 module_init(ibmvnic_module_init);
5014 module_exit(ibmvnic_module_exit);