drivers/net/ethernet/ibm/ibmvnic.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /**************************************************************************/
   3 /*                                                                        */
   4 /*  IBM System i and System p Virtual NIC Device Driver                   */
   5 /*  Copyright (C) 2014 IBM Corp.                                          */
   6 /*  Santiago Leon ([email protected])                                  */
   7 /*  Thomas Falcon ([email protected])                           */
   8 /*  John Allen ([email protected])                                */
   9 /*                                                                        */
  10 /*                                                                        */
  11 /* This module contains the implementation of a virtual ethernet device   */
  12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  15 /*                                                                         */
  16 /* Messages are passed between the VNIC driver and the VNIC server using  */
  17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  18 /* issue and receive commands that initiate communication with the server */
  19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  20 /* are used by the driver to notify the server that a packet is           */
  21 /* ready for transmission or that a buffer has been added to receive a    */
  22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  23 /* driver that a packet transmission has been completed or that a packet  */
  24 /* has been received and placed in a waiting buffer.                      */
  25 /*                                                                        */
  26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  28 /* or receive has been completed, the VNIC driver is required to use      */
  29 /* "long term mapping". This entails that large, continuous DMA mapped    */
  30 /* buffers are allocated on driver initialization and these buffers are   */
  31 /* then continuously reused to pass skbs to and from the VNIC server.     */
  32 /*                                                                        */
  33 /**************************************************************************/
  34
  35 #include <linux/module.h>
  36 #include <linux/moduleparam.h>
  37 #include <linux/types.h>
  38 #include <linux/errno.h>
  39 #include <linux/completion.h>
  40 #include <linux/ioport.h>
  41 #include <linux/dma-mapping.h>
  42 #include <linux/kernel.h>
  43 #include <linux/netdevice.h>
  44 #include <linux/etherdevice.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/init.h>
  47 #include <linux/delay.h>
  48 #include <linux/mm.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/if_arp.h>
  52 #include <linux/in.h>
  53 #include <linux/ip.h>
  54 #include <linux/ipv6.h>
  55 #include <linux/irq.h>
  56 #include <linux/kthread.h>
  57 #include <linux/seq_file.h>
  58 #include <linux/interrupt.h>
  59 #include <net/net_namespace.h>
  60 #include <asm/hvcall.h>
  61 #include <linux/atomic.h>
  62 #include <asm/vio.h>
  63 #include <asm/iommu.h>
  64 #include <linux/uaccess.h>
  65 #include <asm/firmware.h>
  66 #include <linux/workqueue.h>
  67 #include <linux/if_vlan.h>
  68 #include <linux/utsname.h>
  69
  70 #include "ibmvnic.h"
  71
  72 static const char ibmvnic_driver_name[] = "ibmvnic";
  73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  74
  75 MODULE_AUTHOR("Santiago Leon");
  76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  77 MODULE_LICENSE("GPL");
  78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  79
  80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  81 static int ibmvnic_remove(struct vio_dev *);
  82 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  83 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  84 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  85 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  86 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  87 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  88 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  89 static int enable_scrq_irq(struct ibmvnic_adapter *,
  90                            struct ibmvnic_sub_crq_queue *);
  91 static int disable_scrq_irq(struct ibmvnic_adapter *,
  92                             struct ibmvnic_sub_crq_queue *);
  93 static int pending_scrq(struct ibmvnic_adapter *,
  94                         struct ibmvnic_sub_crq_queue *);
  95 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  96                                         struct ibmvnic_sub_crq_queue *);
  97 static int ibmvnic_poll(struct napi_struct *napi, int data);
  98 static void send_query_map(struct ibmvnic_adapter *adapter);
  99 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
 100 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 101 static int send_login(struct ibmvnic_adapter *adapter);
 102 static void send_query_cap(struct ibmvnic_adapter *adapter);
 103 static int init_sub_crqs(struct ibmvnic_adapter *);
 104 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 105 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 106 static void release_crq_queue(struct ibmvnic_adapter *);
 107 static int __ibmvnic_set_mac(struct net_device *, u8 *);
 108 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 109 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 110
 111 struct ibmvnic_stat {
 112         char name[ETH_GSTRING_LEN];
 113         int offset;
 114 };
 115
 116 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 117                              offsetof(struct ibmvnic_statistics, stat))
 118 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + off)))
 119
 120 static const struct ibmvnic_stat ibmvnic_stats[] = {
 121         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 122         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 123         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 124         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 125         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 126         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 127         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 128         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 129         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 130         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 131         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 132         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 133         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 134         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 135         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 136         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 137         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 138         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 139         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 140         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 141         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 142         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 143 };
 144
 145 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 146                           unsigned long length, unsigned long *number,
 147                           unsigned long *irq)
 148 {
 149         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 150         long rc;
 151
 152         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 153         *number = retbuf[0];
 154         *irq = retbuf[1];
 155
 156         return rc;
 157 }
 158
 159 /**
 160  * ibmvnic_wait_for_completion - Check device state and wait for completion
 161  * @adapter: private device data
 162  * @comp_done: completion structure to wait for
 163  * @timeout: time to wait in milliseconds
 164  *
 165  * Wait for a completion signal or until the timeout limit is reached
 166  * while checking that the device is still active.
 167  */
 168 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 169                                        struct completion *comp_done,
 170                                        unsigned long timeout)
 171 {
 172         struct net_device *netdev;
 173         unsigned long div_timeout;
 174         u8 retry;
 175
 176         netdev = adapter->netdev;
 177         retry = 5;
 178         div_timeout = msecs_to_jiffies(timeout / retry);
 179         while (true) {
 180                 if (!adapter->crq.active) {
 181                         netdev_err(netdev, "Device down!\n");
 182                         return -ENODEV;
 183                 }
 184                 if (!retry--)
 185                         break;
 186                 if (wait_for_completion_timeout(comp_done, div_timeout))
 187                         return 0;
 188         }
 189         netdev_err(netdev, "Operation timed out.\n");
 190         return -ETIMEDOUT;
 191 }
 192
 193 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 194                                 struct ibmvnic_long_term_buff *ltb, int size)
 195 {
 196         struct device *dev = &adapter->vdev->dev;
 197         int rc;
 198
 199         ltb->size = size;
 200         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 201                                        GFP_KERNEL);
 202
 203         if (!ltb->buff) {
 204                 dev_err(dev, "Couldn't alloc long term buffer\n");
 205                 return -ENOMEM;
 206         }
 207         ltb->map_id = adapter->map_id;
 208         adapter->map_id++;
 209
 210         mutex_lock(&adapter->fw_lock);
 211         adapter->fw_done_rc = 0;
 212         reinit_completion(&adapter->fw_done);
 213         rc = send_request_map(adapter, ltb->addr,
 214                               ltb->size, ltb->map_id);
 215         if (rc) {
 216                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 217                 mutex_unlock(&adapter->fw_lock);
 218                 return rc;
 219         }
 220
 221         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 222         if (rc) {
 223                 dev_err(dev,
 224                         "Long term map request aborted or timed out,rc = %d\n",
 225                         rc);
 226                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 227                 mutex_unlock(&adapter->fw_lock);
 228                 return rc;
 229         }
 230
 231         if (adapter->fw_done_rc) {
 232                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 233                         adapter->fw_done_rc);
 234                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 235                 mutex_unlock(&adapter->fw_lock);
 236                 return -1;
 237         }
 238         mutex_unlock(&adapter->fw_lock);
 239         return 0;
 240 }
 241
 242 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 243                                 struct ibmvnic_long_term_buff *ltb)
 244 {
 245         struct device *dev = &adapter->vdev->dev;
 246
 247         if (!ltb->buff)
 248                 return;
 249
 250         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 251             adapter->reset_reason != VNIC_RESET_MOBILITY)
 252                 send_request_unmap(adapter, ltb->map_id);
 253         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 254 }
 255
 256 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 257                                 struct ibmvnic_long_term_buff *ltb)
 258 {
 259         struct device *dev = &adapter->vdev->dev;
 260         int rc;
 261
 262         memset(ltb->buff, 0, ltb->size);
 263
 264         mutex_lock(&adapter->fw_lock);
 265         adapter->fw_done_rc = 0;
 266
 267         reinit_completion(&adapter->fw_done);
 268         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 269         if (rc) {
 270                 mutex_unlock(&adapter->fw_lock);
 271                 return rc;
 272         }
 273
 274         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 275         if (rc) {
 276                 dev_info(dev,
 277                          "Reset failed, long term map request timed out or aborted\n");
 278                 mutex_unlock(&adapter->fw_lock);
 279                 return rc;
 280         }
 281
 282         if (adapter->fw_done_rc) {
 283                 dev_info(dev,
 284                          "Reset failed, attempting to free and reallocate buffer\n");
 285                 free_long_term_buff(adapter, ltb);
 286                 mutex_unlock(&adapter->fw_lock);
 287                 return alloc_long_term_buff(adapter, ltb, ltb->size);
 288         }
 289         mutex_unlock(&adapter->fw_lock);
 290         return 0;
 291 }
 292
 293 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 294 {
 295         int i;
 296
 297         for (i = 0; i < adapter->num_active_rx_pools; i++)
 298                 adapter->rx_pool[i].active = 0;
 299 }
 300
 301 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 302                               struct ibmvnic_rx_pool *pool)
 303 {
 304         int count = pool->size - atomic_read(&pool->available);
 305         u64 handle = adapter->rx_scrq[pool->index]->handle;
 306         struct device *dev = &adapter->vdev->dev;
 307         struct ibmvnic_ind_xmit_queue *ind_bufp;
 308         struct ibmvnic_sub_crq_queue *rx_scrq;
 309         union sub_crq *sub_crq;
 310         int buffers_added = 0;
 311         unsigned long lpar_rc;
 312         struct sk_buff *skb;
 313         unsigned int offset;
 314         dma_addr_t dma_addr;
 315         unsigned char *dst;
 316         int shift = 0;
 317         int index;
 318         int i;
 319
 320         if (!pool->active)
 321                 return;
 322
 323         rx_scrq = adapter->rx_scrq[pool->index];
 324         ind_bufp = &rx_scrq->ind_buf;
 325         for (i = 0; i < count; ++i) {
 326                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 327                 if (!skb) {
 328                         dev_err(dev, "Couldn't replenish rx buff\n");
 329                         adapter->replenish_no_mem++;
 330                         break;
 331                 }
 332
 333                 index = pool->free_map[pool->next_free];
 334
 335                 if (pool->rx_buff[index].skb)
 336                         dev_err(dev, "Inconsistent free_map!\n");
 337
 338                 /* Copy the skb to the long term mapped DMA buffer */
 339                 offset = index * pool->buff_size;
 340                 dst = pool->long_term_buff.buff + offset;
 341                 memset(dst, 0, pool->buff_size);
 342                 dma_addr = pool->long_term_buff.addr + offset;
 343                 pool->rx_buff[index].data = dst;
 344
 345                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 346                 pool->rx_buff[index].dma = dma_addr;
 347                 pool->rx_buff[index].skb = skb;
 348                 pool->rx_buff[index].pool_index = pool->index;
 349                 pool->rx_buff[index].size = pool->buff_size;
 350
 351                 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
 352                 memset(sub_crq, 0, sizeof(*sub_crq));
 353                 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
 354                 sub_crq->rx_add.correlator =
 355                     cpu_to_be64((u64)&pool->rx_buff[index]);
 356                 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
 357                 sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
 358
 359                 /* The length field of the sCRQ is defined to be 24 bits so the
 360                  * buffer size needs to be left shifted by a byte before it is
 361                  * converted to big endian to prevent the last byte from being
 362                  * truncated.
 363                  */
 364 #ifdef __LITTLE_ENDIAN__
 365                 shift = 8;
 366 #endif
 367                 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
 368                 pool->next_free = (pool->next_free + 1) % pool->size;
 369                 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
 370                     i == count - 1) {
 371                         lpar_rc =
 372                                 send_subcrq_indirect(adapter, handle,
 373                                                      (u64)ind_bufp->indir_dma,
 374                                                      (u64)ind_bufp->index);
 375                         if (lpar_rc != H_SUCCESS)
 376                                 goto failure;
 377                         buffers_added += ind_bufp->index;
 378                         adapter->replenish_add_buff_success += ind_bufp->index;
 379                         ind_bufp->index = 0;
 380                 }
 381         }
 382         atomic_add(buffers_added, &pool->available);
 383         return;
 384
 385 failure:
 386         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 387                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 388         for (i = ind_bufp->index - 1; i >= 0; --i) {
 389                 struct ibmvnic_rx_buff *rx_buff;
 390
 391                 pool->next_free = pool->next_free == 0 ?
 392                                   pool->size - 1 : pool->next_free - 1;
 393                 sub_crq = &ind_bufp->indir_arr[i];
 394                 rx_buff = (struct ibmvnic_rx_buff *)
 395                                 be64_to_cpu(sub_crq->rx_add.correlator);
 396                 index = (int)(rx_buff - pool->rx_buff);
 397                 pool->free_map[pool->next_free] = index;
 398                 dev_kfree_skb_any(pool->rx_buff[index].skb);
 399                 pool->rx_buff[index].skb = NULL;
 400         }
 401         adapter->replenish_add_buff_failure += ind_bufp->index;
 402         atomic_add(buffers_added, &pool->available);
 403         ind_bufp->index = 0;
 404         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 405                 /* Disable buffer pool replenishment and report carrier off if
 406                  * queue is closed or pending failover.
 407                  * Firmware guarantees that a signal will be sent to the
 408                  * driver, triggering a reset.
 409                  */
 410                 deactivate_rx_pools(adapter);
 411                 netif_carrier_off(adapter->netdev);
 412         }
 413 }
 414
 415 static void replenish_pools(struct ibmvnic_adapter *adapter)
 416 {
 417         int i;
 418
 419         adapter->replenish_task_cycles++;
 420         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 421                 if (adapter->rx_pool[i].active)
 422                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 423         }
 424
 425         netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 426 }
 427
 428 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 429 {
 430         kfree(adapter->tx_stats_buffers);
 431         kfree(adapter->rx_stats_buffers);
 432         adapter->tx_stats_buffers = NULL;
 433         adapter->rx_stats_buffers = NULL;
 434 }
 435
 436 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 437 {
 438         adapter->tx_stats_buffers =
 439                                 kcalloc(IBMVNIC_MAX_QUEUES,
 440                                         sizeof(struct ibmvnic_tx_queue_stats),
 441                                         GFP_KERNEL);
 442         if (!adapter->tx_stats_buffers)
 443                 return -ENOMEM;
 444
 445         adapter->rx_stats_buffers =
 446                                 kcalloc(IBMVNIC_MAX_QUEUES,
 447                                         sizeof(struct ibmvnic_rx_queue_stats),
 448                                         GFP_KERNEL);
 449         if (!adapter->rx_stats_buffers)
 450                 return -ENOMEM;
 451
 452         return 0;
 453 }
 454
 455 static void release_stats_token(struct ibmvnic_adapter *adapter)
 456 {
 457         struct device *dev = &adapter->vdev->dev;
 458
 459         if (!adapter->stats_token)
 460                 return;
 461
 462         dma_unmap_single(dev, adapter->stats_token,
 463                          sizeof(struct ibmvnic_statistics),
 464                          DMA_FROM_DEVICE);
 465         adapter->stats_token = 0;
 466 }
 467
 468 static int init_stats_token(struct ibmvnic_adapter *adapter)
 469 {
 470         struct device *dev = &adapter->vdev->dev;
 471         dma_addr_t stok;
 472
 473         stok = dma_map_single(dev, &adapter->stats,
 474                               sizeof(struct ibmvnic_statistics),
 475                               DMA_FROM_DEVICE);
 476         if (dma_mapping_error(dev, stok)) {
 477                 dev_err(dev, "Couldn't map stats buffer\n");
 478                 return -1;
 479         }
 480
 481         adapter->stats_token = stok;
 482         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 483         return 0;
 484 }
 485
 486 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 487 {
 488         struct ibmvnic_rx_pool *rx_pool;
 489         u64 buff_size;
 490         int rx_scrqs;
 491         int i, j, rc;
 492
 493         if (!adapter->rx_pool)
 494                 return -1;
 495
 496         buff_size = adapter->cur_rx_buf_sz;
 497         rx_scrqs = adapter->num_active_rx_pools;
 498         for (i = 0; i < rx_scrqs; i++) {
 499                 rx_pool = &adapter->rx_pool[i];
 500
 501                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 502
 503                 if (rx_pool->buff_size != buff_size) {
 504                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 505                         rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 506                         rc = alloc_long_term_buff(adapter,
 507                                                   &rx_pool->long_term_buff,
 508                                                   rx_pool->size *
 509                                                   rx_pool->buff_size);
 510                 } else {
 511                         rc = reset_long_term_buff(adapter,
 512                                                   &rx_pool->long_term_buff);
 513                 }
 514
 515                 if (rc)
 516                         return rc;
 517
 518                 for (j = 0; j < rx_pool->size; j++)
 519                         rx_pool->free_map[j] = j;
 520
 521                 memset(rx_pool->rx_buff, 0,
 522                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 523
 524                 atomic_set(&rx_pool->available, 0);
 525                 rx_pool->next_alloc = 0;
 526                 rx_pool->next_free = 0;
 527                 rx_pool->active = 1;
 528         }
 529
 530         return 0;
 531 }
 532
 533 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 534 {
 535         struct ibmvnic_rx_pool *rx_pool;
 536         int i, j;
 537
 538         if (!adapter->rx_pool)
 539                 return;
 540
 541         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 542                 rx_pool = &adapter->rx_pool[i];
 543
 544                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 545
 546                 kfree(rx_pool->free_map);
 547                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 548
 549                 if (!rx_pool->rx_buff)
 550                         continue;
 551
 552                 for (j = 0; j < rx_pool->size; j++) {
 553                         if (rx_pool->rx_buff[j].skb) {
 554                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 555                                 rx_pool->rx_buff[j].skb = NULL;
 556                         }
 557                 }
 558
 559                 kfree(rx_pool->rx_buff);
 560         }
 561
 562         kfree(adapter->rx_pool);
 563         adapter->rx_pool = NULL;
 564         adapter->num_active_rx_pools = 0;
 565 }
 566
 567 static int init_rx_pools(struct net_device *netdev)
 568 {
 569         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 570         struct device *dev = &adapter->vdev->dev;
 571         struct ibmvnic_rx_pool *rx_pool;
 572         int rxadd_subcrqs;
 573         u64 buff_size;
 574         int i, j;
 575
 576         rxadd_subcrqs = adapter->num_active_rx_scrqs;
 577         buff_size = adapter->cur_rx_buf_sz;
 578
 579         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 580                                    sizeof(struct ibmvnic_rx_pool),
 581                                    GFP_KERNEL);
 582         if (!adapter->rx_pool) {
 583                 dev_err(dev, "Failed to allocate rx pools\n");
 584                 return -1;
 585         }
 586
 587         adapter->num_active_rx_pools = rxadd_subcrqs;
 588
 589         for (i = 0; i < rxadd_subcrqs; i++) {
 590                 rx_pool = &adapter->rx_pool[i];
 591
 592                 netdev_dbg(adapter->netdev,
 593                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 594                            i, adapter->req_rx_add_entries_per_subcrq,
 595                            buff_size);
 596
 597                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 598                 rx_pool->index = i;
 599                 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 600                 rx_pool->active = 1;
 601
 602                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 603                                             GFP_KERNEL);
 604                 if (!rx_pool->free_map) {
 605                         release_rx_pools(adapter);
 606                         return -1;
 607                 }
 608
 609                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 610                                            sizeof(struct ibmvnic_rx_buff),
 611                                            GFP_KERNEL);
 612                 if (!rx_pool->rx_buff) {
 613                         dev_err(dev, "Couldn't alloc rx buffers\n");
 614                         release_rx_pools(adapter);
 615                         return -1;
 616                 }
 617
 618                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 619                                          rx_pool->size * rx_pool->buff_size)) {
 620                         release_rx_pools(adapter);
 621                         return -1;
 622                 }
 623
 624                 for (j = 0; j < rx_pool->size; ++j)
 625                         rx_pool->free_map[j] = j;
 626
 627                 atomic_set(&rx_pool->available, 0);
 628                 rx_pool->next_alloc = 0;
 629                 rx_pool->next_free = 0;
 630         }
 631
 632         return 0;
 633 }
 634
 635 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 636                              struct ibmvnic_tx_pool *tx_pool)
 637 {
 638         int rc, i;
 639
 640         rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 641         if (rc)
 642                 return rc;
 643
 644         memset(tx_pool->tx_buff, 0,
 645                tx_pool->num_buffers *
 646                sizeof(struct ibmvnic_tx_buff));
 647
 648         for (i = 0; i < tx_pool->num_buffers; i++)
 649                 tx_pool->free_map[i] = i;
 650
 651         tx_pool->consumer_index = 0;
 652         tx_pool->producer_index = 0;
 653
 654         return 0;
 655 }
 656
 657 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 658 {
 659         int tx_scrqs;
 660         int i, rc;
 661
 662         if (!adapter->tx_pool)
 663                 return -1;
 664
 665         tx_scrqs = adapter->num_active_tx_pools;
 666         for (i = 0; i < tx_scrqs; i++) {
 667                 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 668                 if (rc)
 669                         return rc;
 670                 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 671                 if (rc)
 672                         return rc;
 673         }
 674
 675         return 0;
 676 }
 677
 678 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 679 {
 680         if (!adapter->vpd)
 681                 return;
 682
 683         kfree(adapter->vpd->buff);
 684         kfree(adapter->vpd);
 685
 686         adapter->vpd = NULL;
 687 }
 688
 689 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 690                                 struct ibmvnic_tx_pool *tx_pool)
 691 {
 692         kfree(tx_pool->tx_buff);
 693         kfree(tx_pool->free_map);
 694         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 695 }
 696
 697 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 698 {
 699         int i;
 700
 701         if (!adapter->tx_pool)
 702                 return;
 703
 704         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 705                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 706                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 707         }
 708
 709         kfree(adapter->tx_pool);
 710         adapter->tx_pool = NULL;
 711         kfree(adapter->tso_pool);
 712         adapter->tso_pool = NULL;
 713         adapter->num_active_tx_pools = 0;
 714 }
 715
 716 static int init_one_tx_pool(struct net_device *netdev,
 717                             struct ibmvnic_tx_pool *tx_pool,
 718                             int num_entries, int buf_size)
 719 {
 720         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 721         int i;
 722
 723         tx_pool->tx_buff = kcalloc(num_entries,
 724                                    sizeof(struct ibmvnic_tx_buff),
 725                                    GFP_KERNEL);
 726         if (!tx_pool->tx_buff)
 727                 return -1;
 728
 729         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 730                                  num_entries * buf_size))
 731                 return -1;
 732
 733         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 734         if (!tx_pool->free_map)
 735                 return -1;
 736
 737         for (i = 0; i < num_entries; i++)
 738                 tx_pool->free_map[i] = i;
 739
 740         tx_pool->consumer_index = 0;
 741         tx_pool->producer_index = 0;
 742         tx_pool->num_buffers = num_entries;
 743         tx_pool->buf_size = buf_size;
 744
 745         return 0;
 746 }
 747
 748 static int init_tx_pools(struct net_device *netdev)
 749 {
 750         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 751         int tx_subcrqs;
 752         u64 buff_size;
 753         int i, rc;
 754
 755         tx_subcrqs = adapter->num_active_tx_scrqs;
 756         adapter->tx_pool = kcalloc(tx_subcrqs,
 757                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 758         if (!adapter->tx_pool)
 759                 return -1;
 760
 761         adapter->tso_pool = kcalloc(tx_subcrqs,
 762                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 763         if (!adapter->tso_pool)
 764                 return -1;
 765
 766         adapter->num_active_tx_pools = tx_subcrqs;
 767
 768         for (i = 0; i < tx_subcrqs; i++) {
 769                 buff_size = adapter->req_mtu + VLAN_HLEN;
 770                 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 771                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 772                                       adapter->req_tx_entries_per_subcrq,
 773                                       buff_size);
 774                 if (rc) {
 775                         release_tx_pools(adapter);
 776                         return rc;
 777                 }
 778
 779                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 780                                       IBMVNIC_TSO_BUFS,
 781                                       IBMVNIC_TSO_BUF_SZ);
 782                 if (rc) {
 783                         release_tx_pools(adapter);
 784                         return rc;
 785                 }
 786         }
 787
 788         return 0;
 789 }
 790
 791 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 792 {
 793         int i;
 794
 795         if (adapter->napi_enabled)
 796                 return;
 797
 798         for (i = 0; i < adapter->req_rx_queues; i++)
 799                 napi_enable(&adapter->napi[i]);
 800
 801         adapter->napi_enabled = true;
 802 }
 803
 804 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 805 {
 806         int i;
 807
 808         if (!adapter->napi_enabled)
 809                 return;
 810
 811         for (i = 0; i < adapter->req_rx_queues; i++) {
 812                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 813                 napi_disable(&adapter->napi[i]);
 814         }
 815
 816         adapter->napi_enabled = false;
 817 }
 818
 819 static int init_napi(struct ibmvnic_adapter *adapter)
 820 {
 821         int i;
 822
 823         adapter->napi = kcalloc(adapter->req_rx_queues,
 824                                 sizeof(struct napi_struct), GFP_KERNEL);
 825         if (!adapter->napi)
 826                 return -ENOMEM;
 827
 828         for (i = 0; i < adapter->req_rx_queues; i++) {
 829                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 830                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 831                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 832         }
 833
 834         adapter->num_active_rx_napi = adapter->req_rx_queues;
 835         return 0;
 836 }
 837
 838 static void release_napi(struct ibmvnic_adapter *adapter)
 839 {
 840         int i;
 841
 842         if (!adapter->napi)
 843                 return;
 844
 845         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 846                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 847                 netif_napi_del(&adapter->napi[i]);
 848         }
 849
 850         kfree(adapter->napi);
 851         adapter->napi = NULL;
 852         adapter->num_active_rx_napi = 0;
 853         adapter->napi_enabled = false;
 854 }
 855
 856 static int ibmvnic_login(struct net_device *netdev)
 857 {
 858         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 859         unsigned long timeout = msecs_to_jiffies(20000);
 860         int retry_count = 0;
 861         int retries = 10;
 862         bool retry;
 863         int rc;
 864
 865         do {
 866                 retry = false;
 867                 if (retry_count > retries) {
 868                         netdev_warn(netdev, "Login attempts exceeded\n");
 869                         return -1;
 870                 }
 871
 872                 adapter->init_done_rc = 0;
 873                 reinit_completion(&adapter->init_done);
 874                 rc = send_login(adapter);
 875                 if (rc)
 876                         return rc;
 877
 878                 if (!wait_for_completion_timeout(&adapter->init_done,
 879                                                  timeout)) {
 880                         netdev_warn(netdev, "Login timed out, retrying...\n");
 881                         retry = true;
 882                         adapter->init_done_rc = 0;
 883                         retry_count++;
 884                         continue;
 885                 }
 886
 887                 if (adapter->init_done_rc == ABORTED) {
 888                         netdev_warn(netdev, "Login aborted, retrying...\n");
 889                         retry = true;
 890                         adapter->init_done_rc = 0;
 891                         retry_count++;
 892                         /* FW or device may be busy, so
 893                          * wait a bit before retrying login
 894                          */
 895                         msleep(500);
 896                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 897                         retry_count++;
 898                         release_sub_crqs(adapter, 1);
 899
 900                         retry = true;
 901                         netdev_dbg(netdev,
 902                                    "Received partial success, retrying...\n");
 903                         adapter->init_done_rc = 0;
 904                         reinit_completion(&adapter->init_done);
 905                         send_query_cap(adapter);
 906                         if (!wait_for_completion_timeout(&adapter->init_done,
 907                                                          timeout)) {
 908                                 netdev_warn(netdev,
 909                                             "Capabilities query timed out\n");
 910                                 return -1;
 911                         }
 912
 913                         rc = init_sub_crqs(adapter);
 914                         if (rc) {
 915                                 netdev_warn(netdev,
 916                                             "SCRQ initialization failed\n");
 917                                 return -1;
 918                         }
 919
 920                         rc = init_sub_crq_irqs(adapter);
 921                         if (rc) {
 922                                 netdev_warn(netdev,
 923                                             "SCRQ irq initialization failed\n");
 924                                 return -1;
 925                         }
 926                 } else if (adapter->init_done_rc) {
 927                         netdev_warn(netdev, "Adapter login failed\n");
 928                         return -1;
 929                 }
 930         } while (retry);
 931
 932         __ibmvnic_set_mac(netdev, adapter->mac_addr);
 933
 934         netdev_dbg(netdev, "[S:%d] Login succeeded\n", adapter->state);
 935         return 0;
 936 }
 937
 938 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 939 {
 940         kfree(adapter->login_buf);
 941         adapter->login_buf = NULL;
 942 }
 943
 944 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 945 {
 946         kfree(adapter->login_rsp_buf);
 947         adapter->login_rsp_buf = NULL;
 948 }
 949
 950 static void release_resources(struct ibmvnic_adapter *adapter)
 951 {
 952         release_vpd_data(adapter);
 953
 954         release_tx_pools(adapter);
 955         release_rx_pools(adapter);
 956
 957         release_napi(adapter);
 958         release_login_buffer(adapter);
 959         release_login_rsp_buffer(adapter);
 960 }
 961
 962 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 963 {
 964         struct net_device *netdev = adapter->netdev;
 965         unsigned long timeout = msecs_to_jiffies(20000);
 966         union ibmvnic_crq crq;
 967         bool resend;
 968         int rc;
 969
 970         netdev_dbg(netdev, "setting link state %d\n", link_state);
 971
 972         memset(&crq, 0, sizeof(crq));
 973         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 974         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 975         crq.logical_link_state.link_state = link_state;
 976
 977         do {
 978                 resend = false;
 979
 980                 reinit_completion(&adapter->init_done);
 981                 rc = ibmvnic_send_crq(adapter, &crq);
 982                 if (rc) {
 983                         netdev_err(netdev, "Failed to set link state\n");
 984                         return rc;
 985                 }
 986
 987                 if (!wait_for_completion_timeout(&adapter->init_done,
 988                                                  timeout)) {
 989                         netdev_err(netdev, "timeout setting link state\n");
 990                         return -1;
 991                 }
 992
 993                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 994                         /* Partuial success, delay and re-send */
 995                         mdelay(1000);
 996                         resend = true;
 997                 } else if (adapter->init_done_rc) {
 998                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 999                                     adapter->init_done_rc);
1000                         return adapter->init_done_rc;
1001                 }
1002         } while (resend);
1003
1004         return 0;
1005 }
1006
1007 static int set_real_num_queues(struct net_device *netdev)
1008 {
1009         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1010         int rc;
1011
1012         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1013                    adapter->req_tx_queues, adapter->req_rx_queues);
1014
1015         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1016         if (rc) {
1017                 netdev_err(netdev, "failed to set the number of tx queues\n");
1018                 return rc;
1019         }
1020
1021         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1022         if (rc)
1023                 netdev_err(netdev, "failed to set the number of rx queues\n");
1024
1025         return rc;
1026 }
1027
1028 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1029 {
1030         struct device *dev = &adapter->vdev->dev;
1031         union ibmvnic_crq crq;
1032         int len = 0;
1033         int rc;
1034
1035         if (adapter->vpd->buff)
1036                 len = adapter->vpd->len;
1037
1038         mutex_lock(&adapter->fw_lock);
1039         adapter->fw_done_rc = 0;
1040         reinit_completion(&adapter->fw_done);
1041
1042         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1043         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1044         rc = ibmvnic_send_crq(adapter, &crq);
1045         if (rc) {
1046                 mutex_unlock(&adapter->fw_lock);
1047                 return rc;
1048         }
1049
1050         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1051         if (rc) {
1052                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1053                 mutex_unlock(&adapter->fw_lock);
1054                 return rc;
1055         }
1056         mutex_unlock(&adapter->fw_lock);
1057
1058         if (!adapter->vpd->len)
1059                 return -ENODATA;
1060
1061         if (!adapter->vpd->buff)
1062                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1063         else if (adapter->vpd->len != len)
1064                 adapter->vpd->buff =
1065                         krealloc(adapter->vpd->buff,
1066                                  adapter->vpd->len, GFP_KERNEL);
1067
1068         if (!adapter->vpd->buff) {
1069                 dev_err(dev, "Could allocate VPD buffer\n");
1070                 return -ENOMEM;
1071         }
1072
1073         adapter->vpd->dma_addr =
1074                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1075                                DMA_FROM_DEVICE);
1076         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1077                 dev_err(dev, "Could not map VPD buffer\n");
1078                 kfree(adapter->vpd->buff);
1079                 adapter->vpd->buff = NULL;
1080                 return -ENOMEM;
1081         }
1082
1083         mutex_lock(&adapter->fw_lock);
1084         adapter->fw_done_rc = 0;
1085         reinit_completion(&adapter->fw_done);
1086
1087         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1088         crq.get_vpd.cmd = GET_VPD;
1089         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1090         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1091         rc = ibmvnic_send_crq(adapter, &crq);
1092         if (rc) {
1093                 kfree(adapter->vpd->buff);
1094                 adapter->vpd->buff = NULL;
1095                 mutex_unlock(&adapter->fw_lock);
1096                 return rc;
1097         }
1098
1099         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1100         if (rc) {
1101                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1102                 kfree(adapter->vpd->buff);
1103                 adapter->vpd->buff = NULL;
1104                 mutex_unlock(&adapter->fw_lock);
1105                 return rc;
1106         }
1107
1108         mutex_unlock(&adapter->fw_lock);
1109         return 0;
1110 }
1111
1112 static int init_resources(struct ibmvnic_adapter *adapter)
1113 {
1114         struct net_device *netdev = adapter->netdev;
1115         int rc;
1116
1117         rc = set_real_num_queues(netdev);
1118         if (rc)
1119                 return rc;
1120
1121         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1122         if (!adapter->vpd)
1123                 return -ENOMEM;
1124
1125         /* Vital Product Data (VPD) */
1126         rc = ibmvnic_get_vpd(adapter);
1127         if (rc) {
1128                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1129                 return rc;
1130         }
1131
1132         adapter->map_id = 1;
1133
1134         rc = init_napi(adapter);
1135         if (rc)
1136                 return rc;
1137
1138         send_query_map(adapter);
1139
1140         rc = init_rx_pools(netdev);
1141         if (rc)
1142                 return rc;
1143
1144         rc = init_tx_pools(netdev);
1145         return rc;
1146 }
1147
1148 static int __ibmvnic_open(struct net_device *netdev)
1149 {
1150         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1151         enum vnic_state prev_state = adapter->state;
1152         int i, rc;
1153
1154         adapter->state = VNIC_OPENING;
1155         replenish_pools(adapter);
1156         ibmvnic_napi_enable(adapter);
1157
1158         /* We're ready to receive frames, enable the sub-crq interrupts and
1159          * set the logical link state to up
1160          */
1161         for (i = 0; i < adapter->req_rx_queues; i++) {
1162                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1163                 if (prev_state == VNIC_CLOSED)
1164                         enable_irq(adapter->rx_scrq[i]->irq);
1165                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1166         }
1167
1168         for (i = 0; i < adapter->req_tx_queues; i++) {
1169                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1170                 if (prev_state == VNIC_CLOSED)
1171                         enable_irq(adapter->tx_scrq[i]->irq);
1172                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1173                 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1174         }
1175
1176         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1177         if (rc) {
1178                 for (i = 0; i < adapter->req_rx_queues; i++)
1179                         napi_disable(&adapter->napi[i]);
1180                 release_resources(adapter);
1181                 return rc;
1182         }
1183
1184         netif_tx_start_all_queues(netdev);
1185
1186         if (prev_state == VNIC_CLOSED) {
1187                 for (i = 0; i < adapter->req_rx_queues; i++)
1188                         napi_schedule(&adapter->napi[i]);
1189         }
1190
1191         adapter->state = VNIC_OPEN;
1192         return rc;
1193 }
1194
1195 static int ibmvnic_open(struct net_device *netdev)
1196 {
1197         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1198         int rc;
1199
1200         /* If device failover is pending, just set device state and return.
1201          * Device operation will be handled by reset routine.
1202          */
1203         if (adapter->failover_pending) {
1204                 adapter->state = VNIC_OPEN;
1205                 return 0;
1206         }
1207
1208         if (adapter->state != VNIC_CLOSED) {
1209                 rc = ibmvnic_login(netdev);
1210                 if (rc)
1211                         goto out;
1212
1213                 rc = init_resources(adapter);
1214                 if (rc) {
1215                         netdev_err(netdev, "failed to initialize resources\n");
1216                         release_resources(adapter);
1217                         goto out;
1218                 }
1219         }
1220
1221         rc = __ibmvnic_open(netdev);
1222
1223 out:
1224         /*
1225          * If open fails due to a pending failover, set device state and
1226          * return. Device operation will be handled by reset routine.
1227          */
1228         if (rc && adapter->failover_pending) {
1229                 adapter->state = VNIC_OPEN;
1230                 rc = 0;
1231         }
1232         return rc;
1233 }
1234
1235 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1236 {
1237         struct ibmvnic_rx_pool *rx_pool;
1238         struct ibmvnic_rx_buff *rx_buff;
1239         u64 rx_entries;
1240         int rx_scrqs;
1241         int i, j;
1242
1243         if (!adapter->rx_pool)
1244                 return;
1245
1246         rx_scrqs = adapter->num_active_rx_pools;
1247         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1248
1249         /* Free any remaining skbs in the rx buffer pools */
1250         for (i = 0; i < rx_scrqs; i++) {
1251                 rx_pool = &adapter->rx_pool[i];
1252                 if (!rx_pool || !rx_pool->rx_buff)
1253                         continue;
1254
1255                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1256                 for (j = 0; j < rx_entries; j++) {
1257                         rx_buff = &rx_pool->rx_buff[j];
1258                         if (rx_buff && rx_buff->skb) {
1259                                 dev_kfree_skb_any(rx_buff->skb);
1260                                 rx_buff->skb = NULL;
1261                         }
1262                 }
1263         }
1264 }
1265
1266 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1267                               struct ibmvnic_tx_pool *tx_pool)
1268 {
1269         struct ibmvnic_tx_buff *tx_buff;
1270         u64 tx_entries;
1271         int i;
1272
1273         if (!tx_pool || !tx_pool->tx_buff)
1274                 return;
1275
1276         tx_entries = tx_pool->num_buffers;
1277
1278         for (i = 0; i < tx_entries; i++) {
1279                 tx_buff = &tx_pool->tx_buff[i];
1280                 if (tx_buff && tx_buff->skb) {
1281                         dev_kfree_skb_any(tx_buff->skb);
1282                         tx_buff->skb = NULL;
1283                 }
1284         }
1285 }
1286
1287 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1288 {
1289         int tx_scrqs;
1290         int i;
1291
1292         if (!adapter->tx_pool || !adapter->tso_pool)
1293                 return;
1294
1295         tx_scrqs = adapter->num_active_tx_pools;
1296
1297         /* Free any remaining skbs in the tx buffer pools */
1298         for (i = 0; i < tx_scrqs; i++) {
1299                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1300                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1301                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1302         }
1303 }
1304
1305 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1306 {
1307         struct net_device *netdev = adapter->netdev;
1308         int i;
1309
1310         if (adapter->tx_scrq) {
1311                 for (i = 0; i < adapter->req_tx_queues; i++)
1312                         if (adapter->tx_scrq[i]->irq) {
1313                                 netdev_dbg(netdev,
1314                                            "Disabling tx_scrq[%d] irq\n", i);
1315                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1316                                 disable_irq(adapter->tx_scrq[i]->irq);
1317                         }
1318         }
1319
1320         if (adapter->rx_scrq) {
1321                 for (i = 0; i < adapter->req_rx_queues; i++) {
1322                         if (adapter->rx_scrq[i]->irq) {
1323                                 netdev_dbg(netdev,
1324                                            "Disabling rx_scrq[%d] irq\n", i);
1325                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1326                                 disable_irq(adapter->rx_scrq[i]->irq);
1327                         }
1328                 }
1329         }
1330 }
1331
1332 static void ibmvnic_cleanup(struct net_device *netdev)
1333 {
1334         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1335
1336         /* ensure that transmissions are stopped if called by do_reset */
1337         if (test_bit(0, &adapter->resetting))
1338                 netif_tx_disable(netdev);
1339         else
1340                 netif_tx_stop_all_queues(netdev);
1341
1342         ibmvnic_napi_disable(adapter);
1343         ibmvnic_disable_irqs(adapter);
1344
1345         clean_rx_pools(adapter);
1346         clean_tx_pools(adapter);
1347 }
1348
1349 static int __ibmvnic_close(struct net_device *netdev)
1350 {
1351         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1352         int rc = 0;
1353
1354         adapter->state = VNIC_CLOSING;
1355         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1356         if (rc)
1357                 return rc;
1358         adapter->state = VNIC_CLOSED;
1359         return 0;
1360 }
1361
1362 static int ibmvnic_close(struct net_device *netdev)
1363 {
1364         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1365         int rc;
1366
1367         netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1368                    adapter->state, adapter->failover_pending,
1369                    adapter->force_reset_recovery);
1370
1371         /* If device failover is pending, just set device state and return.
1372          * Device operation will be handled by reset routine.
1373          */
1374         if (adapter->failover_pending) {
1375                 adapter->state = VNIC_CLOSED;
1376                 return 0;
1377         }
1378
1379         rc = __ibmvnic_close(netdev);
1380         ibmvnic_cleanup(netdev);
1381
1382         return rc;
1383 }
1384
1385 /**
1386  * build_hdr_data - creates L2/L3/L4 header data buffer
1387  * @hdr_field: bitfield determining needed headers
1388  * @skb: socket buffer
1389  * @hdr_len: array of header lengths
1390  * @hdr_data: buffer to write the header to
1391  *
1392  * Reads hdr_field to determine which headers are needed by firmware.
1393  * Builds a buffer containing these headers.  Saves individual header
1394  * lengths and total buffer length to be used to build descriptors.
1395  */
1396 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1397                           int *hdr_len, u8 *hdr_data)
1398 {
1399         int len = 0;
1400         u8 *hdr;
1401
1402         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1403                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1404         else
1405                 hdr_len[0] = sizeof(struct ethhdr);
1406
1407         if (skb->protocol == htons(ETH_P_IP)) {
1408                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1409                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1410                         hdr_len[2] = tcp_hdrlen(skb);
1411                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1412                         hdr_len[2] = sizeof(struct udphdr);
1413         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1414                 hdr_len[1] = sizeof(struct ipv6hdr);
1415                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1416                         hdr_len[2] = tcp_hdrlen(skb);
1417                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1418                         hdr_len[2] = sizeof(struct udphdr);
1419         } else if (skb->protocol == htons(ETH_P_ARP)) {
1420                 hdr_len[1] = arp_hdr_len(skb->dev);
1421                 hdr_len[2] = 0;
1422         }
1423
1424         memset(hdr_data, 0, 120);
1425         if ((hdr_field >> 6) & 1) {
1426                 hdr = skb_mac_header(skb);
1427                 memcpy(hdr_data, hdr, hdr_len[0]);
1428                 len += hdr_len[0];
1429         }
1430
1431         if ((hdr_field >> 5) & 1) {
1432                 hdr = skb_network_header(skb);
1433                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1434                 len += hdr_len[1];
1435         }
1436
1437         if ((hdr_field >> 4) & 1) {
1438                 hdr = skb_transport_header(skb);
1439                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1440                 len += hdr_len[2];
1441         }
1442         return len;
1443 }
1444
1445 /**
1446  * create_hdr_descs - create header and header extension descriptors
1447  * @hdr_field: bitfield determining needed headers
1448  * @hdr_data: buffer containing header data
1449  * @len: length of data buffer
1450  * @hdr_len: array of individual header lengths
1451  * @scrq_arr: descriptor array
1452  *
1453  * Creates header and, if needed, header extension descriptors and
1454  * places them in a descriptor array, scrq_arr
1455  */
1456
1457 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1458                             union sub_crq *scrq_arr)
1459 {
1460         union sub_crq hdr_desc;
1461         int tmp_len = len;
1462         int num_descs = 0;
1463         u8 *data, *cur;
1464         int tmp;
1465
1466         while (tmp_len > 0) {
1467                 cur = hdr_data + len - tmp_len;
1468
1469                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1470                 if (cur != hdr_data) {
1471                         data = hdr_desc.hdr_ext.data;
1472                         tmp = tmp_len > 29 ? 29 : tmp_len;
1473                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1474                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1475                         hdr_desc.hdr_ext.len = tmp;
1476                 } else {
1477                         data = hdr_desc.hdr.data;
1478                         tmp = tmp_len > 24 ? 24 : tmp_len;
1479                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1480                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1481                         hdr_desc.hdr.len = tmp;
1482                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1483                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1484                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1485                         hdr_desc.hdr.flag = hdr_field << 1;
1486                 }
1487                 memcpy(data, cur, tmp);
1488                 tmp_len -= tmp;
1489                 *scrq_arr = hdr_desc;
1490                 scrq_arr++;
1491                 num_descs++;
1492         }
1493
1494         return num_descs;
1495 }
1496
1497 /**
1498  * build_hdr_descs_arr - build a header descriptor array
1499  * @txbuff: tx buffer
1500  * @num_entries: number of descriptors to be sent
1501  * @hdr_field: bit field determining which headers will be sent
1502  *
1503  * This function will build a TX descriptor array with applicable
1504  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1505  */
1506
1507 static void build_hdr_descs_arr(struct sk_buff *skb,
1508                                 union sub_crq *indir_arr,
1509                                 int *num_entries, u8 hdr_field)
1510 {
1511         int hdr_len[3] = {0, 0, 0};
1512         u8 hdr_data[140] = {0};
1513         int tot_len;
1514
1515         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1516                                  hdr_data);
1517         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1518                                          indir_arr + 1);
1519 }
1520
1521 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1522                                     struct net_device *netdev)
1523 {
1524         /* For some backing devices, mishandling of small packets
1525          * can result in a loss of connection or TX stall. Device
1526          * architects recommend that no packet should be smaller
1527          * than the minimum MTU value provided to the driver, so
1528          * pad any packets to that length
1529          */
1530         if (skb->len < netdev->min_mtu)
1531                 return skb_put_padto(skb, netdev->min_mtu);
1532
1533         return 0;
1534 }
1535
1536 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1537                                          struct ibmvnic_sub_crq_queue *tx_scrq)
1538 {
1539         struct ibmvnic_ind_xmit_queue *ind_bufp;
1540         struct ibmvnic_tx_buff *tx_buff;
1541         struct ibmvnic_tx_pool *tx_pool;
1542         union sub_crq tx_scrq_entry;
1543         int queue_num;
1544         int entries;
1545         int index;
1546         int i;
1547
1548         ind_bufp = &tx_scrq->ind_buf;
1549         entries = (u64)ind_bufp->index;
1550         queue_num = tx_scrq->pool_index;
1551
1552         for (i = entries - 1; i >= 0; --i) {
1553                 tx_scrq_entry = ind_bufp->indir_arr[i];
1554                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1555                         continue;
1556                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1557                 if (index & IBMVNIC_TSO_POOL_MASK) {
1558                         tx_pool = &adapter->tso_pool[queue_num];
1559                         index &= ~IBMVNIC_TSO_POOL_MASK;
1560                 } else {
1561                         tx_pool = &adapter->tx_pool[queue_num];
1562                 }
1563                 tx_pool->free_map[tx_pool->consumer_index] = index;
1564                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1565                                           tx_pool->num_buffers - 1 :
1566                                           tx_pool->consumer_index - 1;
1567                 tx_buff = &tx_pool->tx_buff[index];
1568                 adapter->netdev->stats.tx_packets--;
1569                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1570                 adapter->tx_stats_buffers[queue_num].packets--;
1571                 adapter->tx_stats_buffers[queue_num].bytes -=
1572                                                 tx_buff->skb->len;
1573                 dev_kfree_skb_any(tx_buff->skb);
1574                 tx_buff->skb = NULL;
1575                 adapter->netdev->stats.tx_dropped++;
1576         }
1577         ind_bufp->index = 0;
1578         if (atomic_sub_return(entries, &tx_scrq->used) <=
1579             (adapter->req_tx_entries_per_subcrq / 2) &&
1580             __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1581                 netif_wake_subqueue(adapter->netdev, queue_num);
1582                 netdev_dbg(adapter->netdev, "Started queue %d\n",
1583                            queue_num);
1584         }
1585 }
1586
1587 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1588                                  struct ibmvnic_sub_crq_queue *tx_scrq)
1589 {
1590         struct ibmvnic_ind_xmit_queue *ind_bufp;
1591         u64 dma_addr;
1592         u64 entries;
1593         u64 handle;
1594         int rc;
1595
1596         ind_bufp = &tx_scrq->ind_buf;
1597         dma_addr = (u64)ind_bufp->indir_dma;
1598         entries = (u64)ind_bufp->index;
1599         handle = tx_scrq->handle;
1600
1601         if (!entries)
1602                 return 0;
1603         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1604         if (rc)
1605                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1606         else
1607                 ind_bufp->index = 0;
1608         return 0;
1609 }
1610
1611 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1612 {
1613         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1614         int queue_num = skb_get_queue_mapping(skb);
1615         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1616         struct device *dev = &adapter->vdev->dev;
1617         struct ibmvnic_ind_xmit_queue *ind_bufp;
1618         struct ibmvnic_tx_buff *tx_buff = NULL;
1619         struct ibmvnic_sub_crq_queue *tx_scrq;
1620         struct ibmvnic_tx_pool *tx_pool;
1621         unsigned int tx_send_failed = 0;
1622         netdev_tx_t ret = NETDEV_TX_OK;
1623         unsigned int tx_map_failed = 0;
1624         union sub_crq indir_arr[16];
1625         unsigned int tx_dropped = 0;
1626         unsigned int tx_packets = 0;
1627         unsigned int tx_bytes = 0;
1628         dma_addr_t data_dma_addr;
1629         struct netdev_queue *txq;
1630         unsigned long lpar_rc;
1631         union sub_crq tx_crq;
1632         unsigned int offset;
1633         int num_entries = 1;
1634         unsigned char *dst;
1635         int index = 0;
1636         u8 proto = 0;
1637
1638         tx_scrq = adapter->tx_scrq[queue_num];
1639         txq = netdev_get_tx_queue(netdev, queue_num);
1640         ind_bufp = &tx_scrq->ind_buf;
1641
1642         if (test_bit(0, &adapter->resetting)) {
1643                 if (!netif_subqueue_stopped(netdev, skb))
1644                         netif_stop_subqueue(netdev, queue_num);
1645                 dev_kfree_skb_any(skb);
1646
1647                 tx_send_failed++;
1648                 tx_dropped++;
1649                 ret = NETDEV_TX_OK;
1650                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1651                 goto out;
1652         }
1653
1654         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1655                 tx_dropped++;
1656                 tx_send_failed++;
1657                 ret = NETDEV_TX_OK;
1658                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1659                 goto out;
1660         }
1661         if (skb_is_gso(skb))
1662                 tx_pool = &adapter->tso_pool[queue_num];
1663         else
1664                 tx_pool = &adapter->tx_pool[queue_num];
1665
1666         index = tx_pool->free_map[tx_pool->consumer_index];
1667
1668         if (index == IBMVNIC_INVALID_MAP) {
1669                 dev_kfree_skb_any(skb);
1670                 tx_send_failed++;
1671                 tx_dropped++;
1672                 ret = NETDEV_TX_OK;
1673                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1674                 goto out;
1675         }
1676
1677         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1678
1679         offset = index * tx_pool->buf_size;
1680         dst = tx_pool->long_term_buff.buff + offset;
1681         memset(dst, 0, tx_pool->buf_size);
1682         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1683
1684         if (skb_shinfo(skb)->nr_frags) {
1685                 int cur, i;
1686
1687                 /* Copy the head */
1688                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1689                 cur = skb_headlen(skb);
1690
1691                 /* Copy the frags */
1692                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1693                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1694
1695                         memcpy(dst + cur,
1696                                page_address(skb_frag_page(frag)) +
1697                                skb_frag_off(frag), skb_frag_size(frag));
1698                         cur += skb_frag_size(frag);
1699                 }
1700         } else {
1701                 skb_copy_from_linear_data(skb, dst, skb->len);
1702         }
1703
1704         tx_pool->consumer_index =
1705             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1706
1707         tx_buff = &tx_pool->tx_buff[index];
1708         tx_buff->skb = skb;
1709         tx_buff->index = index;
1710         tx_buff->pool_index = queue_num;
1711
1712         memset(&tx_crq, 0, sizeof(tx_crq));
1713         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1714         tx_crq.v1.type = IBMVNIC_TX_DESC;
1715         tx_crq.v1.n_crq_elem = 1;
1716         tx_crq.v1.n_sge = 1;
1717         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1718
1719         if (skb_is_gso(skb))
1720                 tx_crq.v1.correlator =
1721                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1722         else
1723                 tx_crq.v1.correlator = cpu_to_be32(index);
1724         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1725         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1726         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1727
1728         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1729                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1730                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1731         }
1732
1733         if (skb->protocol == htons(ETH_P_IP)) {
1734                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1735                 proto = ip_hdr(skb)->protocol;
1736         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1737                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1738                 proto = ipv6_hdr(skb)->nexthdr;
1739         }
1740
1741         if (proto == IPPROTO_TCP)
1742                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1743         else if (proto == IPPROTO_UDP)
1744                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1745
1746         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1747                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1748                 hdrs += 2;
1749         }
1750         if (skb_is_gso(skb)) {
1751                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1752                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1753                 hdrs += 2;
1754         }
1755
1756         if ((*hdrs >> 7) & 1)
1757                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1758
1759         tx_crq.v1.n_crq_elem = num_entries;
1760         tx_buff->num_entries = num_entries;
1761         /* flush buffer if current entry can not fit */
1762         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1763                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1764                 if (lpar_rc != H_SUCCESS)
1765                         goto tx_flush_err;
1766         }
1767
1768         indir_arr[0] = tx_crq;
1769         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1770                num_entries * sizeof(struct ibmvnic_generic_scrq));
1771         ind_bufp->index += num_entries;
1772         if (__netdev_tx_sent_queue(txq, skb->len,
1773                                    netdev_xmit_more() &&
1774                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1775                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1776                 if (lpar_rc != H_SUCCESS)
1777                         goto tx_err;
1778         }
1779
1780         if (atomic_add_return(num_entries, &tx_scrq->used)
1781                                         >= adapter->req_tx_entries_per_subcrq) {
1782                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1783                 netif_stop_subqueue(netdev, queue_num);
1784         }
1785
1786         tx_packets++;
1787         tx_bytes += skb->len;
1788         txq->trans_start = jiffies;
1789         ret = NETDEV_TX_OK;
1790         goto out;
1791
1792 tx_flush_err:
1793         dev_kfree_skb_any(skb);
1794         tx_buff->skb = NULL;
1795         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1796                                   tx_pool->num_buffers - 1 :
1797                                   tx_pool->consumer_index - 1;
1798         tx_dropped++;
1799 tx_err:
1800         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1801                 dev_err_ratelimited(dev, "tx: send failed\n");
1802
1803         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1804                 /* Disable TX and report carrier off if queue is closed
1805                  * or pending failover.
1806                  * Firmware guarantees that a signal will be sent to the
1807                  * driver, triggering a reset or some other action.
1808                  */
1809                 netif_tx_stop_all_queues(netdev);
1810                 netif_carrier_off(netdev);
1811         }
1812 out:
1813         netdev->stats.tx_dropped += tx_dropped;
1814         netdev->stats.tx_bytes += tx_bytes;
1815         netdev->stats.tx_packets += tx_packets;
1816         adapter->tx_send_failed += tx_send_failed;
1817         adapter->tx_map_failed += tx_map_failed;
1818         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1819         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1820         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1821
1822         return ret;
1823 }
1824
1825 static void ibmvnic_set_multi(struct net_device *netdev)
1826 {
1827         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1828         struct netdev_hw_addr *ha;
1829         union ibmvnic_crq crq;
1830
1831         memset(&crq, 0, sizeof(crq));
1832         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1833         crq.request_capability.cmd = REQUEST_CAPABILITY;
1834
1835         if (netdev->flags & IFF_PROMISC) {
1836                 if (!adapter->promisc_supported)
1837                         return;
1838         } else {
1839                 if (netdev->flags & IFF_ALLMULTI) {
1840                         /* Accept all multicast */
1841                         memset(&crq, 0, sizeof(crq));
1842                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1843                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1844                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1845                         ibmvnic_send_crq(adapter, &crq);
1846                 } else if (netdev_mc_empty(netdev)) {
1847                         /* Reject all multicast */
1848                         memset(&crq, 0, sizeof(crq));
1849                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1850                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1851                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1852                         ibmvnic_send_crq(adapter, &crq);
1853                 } else {
1854                         /* Accept one or more multicast(s) */
1855                         netdev_for_each_mc_addr(ha, netdev) {
1856                                 memset(&crq, 0, sizeof(crq));
1857                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1858                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1859                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1860                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1861                                                 ha->addr);
1862                                 ibmvnic_send_crq(adapter, &crq);
1863                         }
1864                 }
1865         }
1866 }
1867
1868 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1869 {
1870         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1871         union ibmvnic_crq crq;
1872         int rc;
1873
1874         if (!is_valid_ether_addr(dev_addr)) {
1875                 rc = -EADDRNOTAVAIL;
1876                 goto err;
1877         }
1878
1879         memset(&crq, 0, sizeof(crq));
1880         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1881         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1882         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1883
1884         mutex_lock(&adapter->fw_lock);
1885         adapter->fw_done_rc = 0;
1886         reinit_completion(&adapter->fw_done);
1887
1888         rc = ibmvnic_send_crq(adapter, &crq);
1889         if (rc) {
1890                 rc = -EIO;
1891                 mutex_unlock(&adapter->fw_lock);
1892                 goto err;
1893         }
1894
1895         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1896         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1897         if (rc || adapter->fw_done_rc) {
1898                 rc = -EIO;
1899                 mutex_unlock(&adapter->fw_lock);
1900                 goto err;
1901         }
1902         mutex_unlock(&adapter->fw_lock);
1903         return 0;
1904 err:
1905         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1906         return rc;
1907 }
1908
1909 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1910 {
1911         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1912         struct sockaddr *addr = p;
1913         int rc;
1914
1915         rc = 0;
1916         if (!is_valid_ether_addr(addr->sa_data))
1917                 return -EADDRNOTAVAIL;
1918
1919         if (adapter->state != VNIC_PROBED) {
1920                 ether_addr_copy(adapter->mac_addr, addr->sa_data);
1921                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1922         }
1923
1924         return rc;
1925 }
1926
1927 /*
1928  * do_reset returns zero if we are able to keep processing reset events, or
1929  * non-zero if we hit a fatal error and must halt.
1930  */
1931 static int do_reset(struct ibmvnic_adapter *adapter,
1932                     struct ibmvnic_rwi *rwi, u32 reset_state)
1933 {
1934         u64 old_num_rx_queues, old_num_tx_queues;
1935         u64 old_num_rx_slots, old_num_tx_slots;
1936         struct net_device *netdev = adapter->netdev;
1937         int i, rc;
1938
1939         netdev_dbg(adapter->netdev,
1940                    "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1941                    adapter->state, adapter->failover_pending,
1942                    rwi->reset_reason, reset_state);
1943
1944         adapter->reset_reason = rwi->reset_reason;
1945         /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
1946         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
1947                 rtnl_lock();
1948
1949         /*
1950          * Now that we have the rtnl lock, clear any pending failover.
1951          * This will ensure ibmvnic_open() has either completed or will
1952          * block until failover is complete.
1953          */
1954         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1955                 adapter->failover_pending = false;
1956
1957         netif_carrier_off(netdev);
1958
1959         old_num_rx_queues = adapter->req_rx_queues;
1960         old_num_tx_queues = adapter->req_tx_queues;
1961         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1962         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1963
1964         ibmvnic_cleanup(netdev);
1965
1966         if (reset_state == VNIC_OPEN &&
1967             adapter->reset_reason != VNIC_RESET_MOBILITY &&
1968             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1969                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
1970                         rc = __ibmvnic_close(netdev);
1971                         if (rc)
1972                                 goto out;
1973                 } else {
1974                         adapter->state = VNIC_CLOSING;
1975
1976                         /* Release the RTNL lock before link state change and
1977                          * re-acquire after the link state change to allow
1978                          * linkwatch_event to grab the RTNL lock and run during
1979                          * a reset.
1980                          */
1981                         rtnl_unlock();
1982                         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1983                         rtnl_lock();
1984                         if (rc)
1985                                 goto out;
1986
1987                         if (adapter->state != VNIC_CLOSING) {
1988                                 rc = -1;
1989                                 goto out;
1990                         }
1991
1992                         adapter->state = VNIC_CLOSED;
1993                 }
1994         }
1995
1996         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
1997                 release_resources(adapter);
1998                 release_sub_crqs(adapter, 1);
1999                 release_crq_queue(adapter);
2000         }
2001
2002         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2003                 /* remove the closed state so when we call open it appears
2004                  * we are coming from the probed state.
2005                  */
2006                 adapter->state = VNIC_PROBED;
2007
2008                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2009                         rc = init_crq_queue(adapter);
2010                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2011                         rc = ibmvnic_reenable_crq_queue(adapter);
2012                         release_sub_crqs(adapter, 1);
2013                 } else {
2014                         rc = ibmvnic_reset_crq(adapter);
2015                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2016                                 rc = vio_enable_interrupts(adapter->vdev);
2017                                 if (rc)
2018                                         netdev_err(adapter->netdev,
2019                                                    "Reset failed to enable interrupts. rc=%d\n",
2020                                                    rc);
2021                         }
2022                 }
2023
2024                 if (rc) {
2025                         netdev_err(adapter->netdev,
2026                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2027                         goto out;
2028                 }
2029
2030                 rc = ibmvnic_reset_init(adapter, true);
2031                 if (rc) {
2032                         rc = IBMVNIC_INIT_FAILED;
2033                         goto out;
2034                 }
2035
2036                 /* If the adapter was in PROBE state prior to the reset,
2037                  * exit here.
2038                  */
2039                 if (reset_state == VNIC_PROBED) {
2040                         rc = 0;
2041                         goto out;
2042                 }
2043
2044                 rc = ibmvnic_login(netdev);
2045                 if (rc) {
2046                         goto out;
2047                 }
2048
2049                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2050                         rc = init_resources(adapter);
2051                         if (rc)
2052                                 goto out;
2053                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2054                     adapter->req_tx_queues != old_num_tx_queues ||
2055                     adapter->req_rx_add_entries_per_subcrq !=
2056                     old_num_rx_slots ||
2057                     adapter->req_tx_entries_per_subcrq !=
2058                     old_num_tx_slots ||
2059                     !adapter->rx_pool ||
2060                     !adapter->tso_pool ||
2061                     !adapter->tx_pool) {
2062                         release_rx_pools(adapter);
2063                         release_tx_pools(adapter);
2064                         release_napi(adapter);
2065                         release_vpd_data(adapter);
2066
2067                         rc = init_resources(adapter);
2068                         if (rc)
2069                                 goto out;
2070
2071                 } else {
2072                         rc = reset_tx_pools(adapter);
2073                         if (rc) {
2074                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2075                                                 rc);
2076                                 goto out;
2077                         }
2078
2079                         rc = reset_rx_pools(adapter);
2080                         if (rc) {
2081                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2082                                                 rc);
2083                                 goto out;
2084                         }
2085                 }
2086                 ibmvnic_disable_irqs(adapter);
2087         }
2088         adapter->state = VNIC_CLOSED;
2089
2090         if (reset_state == VNIC_CLOSED) {
2091                 rc = 0;
2092                 goto out;
2093         }
2094
2095         rc = __ibmvnic_open(netdev);
2096         if (rc) {
2097                 rc = IBMVNIC_OPEN_FAILED;
2098                 goto out;
2099         }
2100
2101         /* refresh device's multicast list */
2102         ibmvnic_set_multi(netdev);
2103
2104         /* kick napi */
2105         for (i = 0; i < adapter->req_rx_queues; i++)
2106                 napi_schedule(&adapter->napi[i]);
2107
2108         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2109             adapter->reset_reason == VNIC_RESET_MOBILITY)
2110                 __netdev_notify_peers(netdev);
2111
2112         rc = 0;
2113
2114 out:
2115         /* restore the adapter state if reset failed */
2116         if (rc)
2117                 adapter->state = reset_state;
2118         /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2119         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2120                 rtnl_unlock();
2121
2122         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2123                    adapter->state, adapter->failover_pending, rc);
2124         return rc;
2125 }
2126
2127 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2128                          struct ibmvnic_rwi *rwi, u32 reset_state)
2129 {
2130         struct net_device *netdev = adapter->netdev;
2131         int rc;
2132
2133         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2134                    rwi->reset_reason);
2135
2136         netif_carrier_off(netdev);
2137         adapter->reset_reason = rwi->reset_reason;
2138
2139         ibmvnic_cleanup(netdev);
2140         release_resources(adapter);
2141         release_sub_crqs(adapter, 0);
2142         release_crq_queue(adapter);
2143
2144         /* remove the closed state so when we call open it appears
2145          * we are coming from the probed state.
2146          */
2147         adapter->state = VNIC_PROBED;
2148
2149         reinit_completion(&adapter->init_done);
2150         rc = init_crq_queue(adapter);
2151         if (rc) {
2152                 netdev_err(adapter->netdev,
2153                            "Couldn't initialize crq. rc=%d\n", rc);
2154                 goto out;
2155         }
2156
2157         rc = ibmvnic_reset_init(adapter, false);
2158         if (rc)
2159                 goto out;
2160
2161         /* If the adapter was in PROBE state prior to the reset,
2162          * exit here.
2163          */
2164         if (reset_state == VNIC_PROBED)
2165                 goto out;
2166
2167         rc = ibmvnic_login(netdev);
2168         if (rc)
2169                 goto out;
2170
2171         rc = init_resources(adapter);
2172         if (rc)
2173                 goto out;
2174
2175         ibmvnic_disable_irqs(adapter);
2176         adapter->state = VNIC_CLOSED;
2177
2178         if (reset_state == VNIC_CLOSED)
2179                 goto out;
2180
2181         rc = __ibmvnic_open(netdev);
2182         if (rc) {
2183                 rc = IBMVNIC_OPEN_FAILED;
2184                 goto out;
2185         }
2186
2187         __netdev_notify_peers(netdev);
2188 out:
2189         /* restore adapter state if reset failed */
2190         if (rc)
2191                 adapter->state = reset_state;
2192         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2193                    adapter->state, adapter->failover_pending, rc);
2194         return rc;
2195 }
2196
2197 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2198 {
2199         struct ibmvnic_rwi *rwi;
2200         unsigned long flags;
2201
2202         spin_lock_irqsave(&adapter->rwi_lock, flags);
2203
2204         if (!list_empty(&adapter->rwi_list)) {
2205                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2206                                        list);
2207                 list_del(&rwi->list);
2208         } else {
2209                 rwi = NULL;
2210         }
2211
2212         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2213         return rwi;
2214 }
2215
2216 static void __ibmvnic_reset(struct work_struct *work)
2217 {
2218         struct ibmvnic_rwi *rwi;
2219         struct ibmvnic_adapter *adapter;
2220         bool saved_state = false;
2221         unsigned long flags;
2222         u32 reset_state;
2223         int rc = 0;
2224
2225         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2226
2227         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2228                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2229                                       IBMVNIC_RESET_DELAY);
2230                 return;
2231         }
2232
2233         rwi = get_next_rwi(adapter);
2234         while (rwi) {
2235                 spin_lock_irqsave(&adapter->state_lock, flags);
2236
2237                 if (adapter->state == VNIC_REMOVING ||
2238                     adapter->state == VNIC_REMOVED) {
2239                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2240                         kfree(rwi);
2241                         rc = EBUSY;
2242                         break;
2243                 }
2244
2245                 if (!saved_state) {
2246                         reset_state = adapter->state;
2247                         saved_state = true;
2248                 }
2249                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2250
2251                 if (adapter->force_reset_recovery) {
2252                         /*
2253                          * Since we are doing a hard reset now, clear the
2254                          * failover_pending flag so we don't ignore any
2255                          * future MOBILITY or other resets.
2256                          */
2257                         adapter->failover_pending = false;
2258
2259                         /* Transport event occurred during previous reset */
2260                         if (adapter->wait_for_reset) {
2261                                 /* Previous was CHANGE_PARAM; caller locked */
2262                                 adapter->force_reset_recovery = false;
2263                                 rc = do_hard_reset(adapter, rwi, reset_state);
2264                         } else {
2265                                 rtnl_lock();
2266                                 adapter->force_reset_recovery = false;
2267                                 rc = do_hard_reset(adapter, rwi, reset_state);
2268                                 rtnl_unlock();
2269                         }
2270                         if (rc) {
2271                                 /* give backing device time to settle down */
2272                                 netdev_dbg(adapter->netdev,
2273                                            "[S:%d] Hard reset failed, waiting 60 secs\n",
2274                                            adapter->state);
2275                                 set_current_state(TASK_UNINTERRUPTIBLE);
2276                                 schedule_timeout(60 * HZ);
2277                         }
2278                 } else {
2279                         rc = do_reset(adapter, rwi, reset_state);
2280                 }
2281                 kfree(rwi);
2282                 adapter->last_reset_time = jiffies;
2283
2284                 if (rc)
2285                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2286
2287                 rwi = get_next_rwi(adapter);
2288
2289                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2290                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2291                         adapter->force_reset_recovery = true;
2292         }
2293
2294         if (adapter->wait_for_reset) {
2295                 adapter->reset_done_rc = rc;
2296                 complete(&adapter->reset_done);
2297         }
2298
2299         clear_bit_unlock(0, &adapter->resetting);
2300
2301         netdev_dbg(adapter->netdev,
2302                    "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2303                    adapter->state, adapter->force_reset_recovery,
2304                    adapter->wait_for_reset);
2305 }
2306
2307 static void __ibmvnic_delayed_reset(struct work_struct *work)
2308 {
2309         struct ibmvnic_adapter *adapter;
2310
2311         adapter = container_of(work, struct ibmvnic_adapter,
2312                                ibmvnic_delayed_reset.work);
2313         __ibmvnic_reset(&adapter->ibmvnic_reset);
2314 }
2315
2316 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2317                          enum ibmvnic_reset_reason reason)
2318 {
2319         struct list_head *entry, *tmp_entry;
2320         struct ibmvnic_rwi *rwi, *tmp;
2321         struct net_device *netdev = adapter->netdev;
2322         unsigned long flags;
2323         int ret;
2324
2325         /*
2326          * If failover is pending don't schedule any other reset.
2327          * Instead let the failover complete. If there is already a
2328          * a failover reset scheduled, we will detect and drop the
2329          * duplicate reset when walking the ->rwi_list below.
2330          */
2331         if (adapter->state == VNIC_REMOVING ||
2332             adapter->state == VNIC_REMOVED ||
2333             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2334                 ret = EBUSY;
2335                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2336                 goto err;
2337         }
2338
2339         if (adapter->state == VNIC_PROBING) {
2340                 netdev_warn(netdev, "Adapter reset during probe\n");
2341                 ret = adapter->init_done_rc = EAGAIN;
2342                 goto err;
2343         }
2344
2345         spin_lock_irqsave(&adapter->rwi_lock, flags);
2346
2347         list_for_each(entry, &adapter->rwi_list) {
2348                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2349                 if (tmp->reset_reason == reason) {
2350                         netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2351                                    reason);
2352                         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2353                         ret = EBUSY;
2354                         goto err;
2355                 }
2356         }
2357
2358         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2359         if (!rwi) {
2360                 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2361                 ibmvnic_close(netdev);
2362                 ret = ENOMEM;
2363                 goto err;
2364         }
2365         /* if we just received a transport event,
2366          * flush reset queue and process this reset
2367          */
2368         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2369                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2370                         list_del(entry);
2371         }
2372         rwi->reset_reason = reason;
2373         list_add_tail(&rwi->list, &adapter->rwi_list);
2374         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2375         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2376         schedule_work(&adapter->ibmvnic_reset);
2377
2378         return 0;
2379 err:
2380         return -ret;
2381 }
2382
2383 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2384 {
2385         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2386
2387         if (test_bit(0, &adapter->resetting)) {
2388                 netdev_err(adapter->netdev,
2389                            "Adapter is resetting, skip timeout reset\n");
2390                 return;
2391         }
2392         /* No queuing up reset until at least 5 seconds (default watchdog val)
2393          * after last reset
2394          */
2395         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2396                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2397                 return;
2398         }
2399         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2400 }
2401
2402 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2403                                   struct ibmvnic_rx_buff *rx_buff)
2404 {
2405         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2406
2407         rx_buff->skb = NULL;
2408
2409         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2410         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2411
2412         atomic_dec(&pool->available);
2413 }
2414
2415 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2416 {
2417         struct ibmvnic_sub_crq_queue *rx_scrq;
2418         struct ibmvnic_adapter *adapter;
2419         struct net_device *netdev;
2420         int frames_processed;
2421         int scrq_num;
2422
2423         netdev = napi->dev;
2424         adapter = netdev_priv(netdev);
2425         scrq_num = (int)(napi - adapter->napi);
2426         frames_processed = 0;
2427         rx_scrq = adapter->rx_scrq[scrq_num];
2428
2429 restart_poll:
2430         while (frames_processed < budget) {
2431                 struct sk_buff *skb;
2432                 struct ibmvnic_rx_buff *rx_buff;
2433                 union sub_crq *next;
2434                 u32 length;
2435                 u16 offset;
2436                 u8 flags = 0;
2437
2438                 if (unlikely(test_bit(0, &adapter->resetting) &&
2439                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2440                         enable_scrq_irq(adapter, rx_scrq);
2441                         napi_complete_done(napi, frames_processed);
2442                         return frames_processed;
2443                 }
2444
2445                 if (!pending_scrq(adapter, rx_scrq))
2446                         break;
2447                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2448                 rx_buff =
2449                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2450                                                           rx_comp.correlator);
2451                 /* do error checking */
2452                 if (next->rx_comp.rc) {
2453                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2454                                    be16_to_cpu(next->rx_comp.rc));
2455                         /* free the entry */
2456                         next->rx_comp.first = 0;
2457                         dev_kfree_skb_any(rx_buff->skb);
2458                         remove_buff_from_pool(adapter, rx_buff);
2459                         continue;
2460                 } else if (!rx_buff->skb) {
2461                         /* free the entry */
2462                         next->rx_comp.first = 0;
2463                         remove_buff_from_pool(adapter, rx_buff);
2464                         continue;
2465                 }
2466
2467                 length = be32_to_cpu(next->rx_comp.len);
2468                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2469                 flags = next->rx_comp.flags;
2470                 skb = rx_buff->skb;
2471                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2472                                         length);
2473
2474                 /* VLAN Header has been stripped by the system firmware and
2475                  * needs to be inserted by the driver
2476                  */
2477                 if (adapter->rx_vlan_header_insertion &&
2478                     (flags & IBMVNIC_VLAN_STRIPPED))
2479                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2480                                                ntohs(next->rx_comp.vlan_tci));
2481
2482                 /* free the entry */
2483                 next->rx_comp.first = 0;
2484                 remove_buff_from_pool(adapter, rx_buff);
2485
2486                 skb_put(skb, length);
2487                 skb->protocol = eth_type_trans(skb, netdev);
2488                 skb_record_rx_queue(skb, scrq_num);
2489
2490                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2491                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2492                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2493                 }
2494
2495                 length = skb->len;
2496                 napi_gro_receive(napi, skb); /* send it up */
2497                 netdev->stats.rx_packets++;
2498                 netdev->stats.rx_bytes += length;
2499                 adapter->rx_stats_buffers[scrq_num].packets++;
2500                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2501                 frames_processed++;
2502         }
2503
2504         if (adapter->state != VNIC_CLOSING &&
2505             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2506               adapter->req_rx_add_entries_per_subcrq / 2) ||
2507               frames_processed < budget))
2508                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2509         if (frames_processed < budget) {
2510                 if (napi_complete_done(napi, frames_processed)) {
2511                         enable_scrq_irq(adapter, rx_scrq);
2512                         if (pending_scrq(adapter, rx_scrq)) {
2513                                 if (napi_reschedule(napi)) {
2514                                         disable_scrq_irq(adapter, rx_scrq);
2515                                         goto restart_poll;
2516                                 }
2517                         }
2518                 }
2519         }
2520         return frames_processed;
2521 }
2522
2523 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2524 {
2525         int rc, ret;
2526
2527         adapter->fallback.mtu = adapter->req_mtu;
2528         adapter->fallback.rx_queues = adapter->req_rx_queues;
2529         adapter->fallback.tx_queues = adapter->req_tx_queues;
2530         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2531         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2532
2533         reinit_completion(&adapter->reset_done);
2534         adapter->wait_for_reset = true;
2535         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2536
2537         if (rc) {
2538                 ret = rc;
2539                 goto out;
2540         }
2541         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2542         if (rc) {
2543                 ret = -ENODEV;
2544                 goto out;
2545         }
2546
2547         ret = 0;
2548         if (adapter->reset_done_rc) {
2549                 ret = -EIO;
2550                 adapter->desired.mtu = adapter->fallback.mtu;
2551                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2552                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2553                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2554                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2555
2556                 reinit_completion(&adapter->reset_done);
2557                 adapter->wait_for_reset = true;
2558                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2559                 if (rc) {
2560                         ret = rc;
2561                         goto out;
2562                 }
2563                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2564                                                  60000);
2565                 if (rc) {
2566                         ret = -ENODEV;
2567                         goto out;
2568                 }
2569         }
2570 out:
2571         adapter->wait_for_reset = false;
2572
2573         return ret;
2574 }
2575
2576 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2577 {
2578         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2579
2580         adapter->desired.mtu = new_mtu + ETH_HLEN;
2581
2582         return wait_for_reset(adapter);
2583 }
2584
2585 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2586                                                 struct net_device *dev,
2587                                                 netdev_features_t features)
2588 {
2589         /* Some backing hardware adapters can not
2590          * handle packets with a MSS less than 224
2591          * or with only one segment.
2592          */
2593         if (skb_is_gso(skb)) {
2594                 if (skb_shinfo(skb)->gso_size < 224 ||
2595                     skb_shinfo(skb)->gso_segs == 1)
2596                         features &= ~NETIF_F_GSO_MASK;
2597         }
2598
2599         return features;
2600 }
2601
2602 static const struct net_device_ops ibmvnic_netdev_ops = {
2603         .ndo_open               = ibmvnic_open,
2604         .ndo_stop               = ibmvnic_close,
2605         .ndo_start_xmit         = ibmvnic_xmit,
2606         .ndo_set_rx_mode        = ibmvnic_set_multi,
2607         .ndo_set_mac_address    = ibmvnic_set_mac,
2608         .ndo_validate_addr      = eth_validate_addr,
2609         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2610         .ndo_change_mtu         = ibmvnic_change_mtu,
2611         .ndo_features_check     = ibmvnic_features_check,
2612 };
2613
2614 /* ethtool functions */
2615
2616 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2617                                       struct ethtool_link_ksettings *cmd)
2618 {
2619         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2620         int rc;
2621
2622         rc = send_query_phys_parms(adapter);
2623         if (rc) {
2624                 adapter->speed = SPEED_UNKNOWN;
2625                 adapter->duplex = DUPLEX_UNKNOWN;
2626         }
2627         cmd->base.speed = adapter->speed;
2628         cmd->base.duplex = adapter->duplex;
2629         cmd->base.port = PORT_FIBRE;
2630         cmd->base.phy_address = 0;
2631         cmd->base.autoneg = AUTONEG_ENABLE;
2632
2633         return 0;
2634 }
2635
2636 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2637                                 struct ethtool_drvinfo *info)
2638 {
2639         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2640
2641         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2642         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2643         strlcpy(info->fw_version, adapter->fw_version,
2644                 sizeof(info->fw_version));
2645 }
2646
2647 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2648 {
2649         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2650
2651         return adapter->msg_enable;
2652 }
2653
2654 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2655 {
2656         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2657
2658         adapter->msg_enable = data;
2659 }
2660
2661 static u32 ibmvnic_get_link(struct net_device *netdev)
2662 {
2663         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2664
2665         /* Don't need to send a query because we request a logical link up at
2666          * init and then we wait for link state indications
2667          */
2668         return adapter->logical_link_state;
2669 }
2670
2671 static void ibmvnic_get_ringparam(struct net_device *netdev,
2672                                   struct ethtool_ringparam *ring)
2673 {
2674         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2675
2676         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2677                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2678                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2679         } else {
2680                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2681                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2682         }
2683         ring->rx_mini_max_pending = 0;
2684         ring->rx_jumbo_max_pending = 0;
2685         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2686         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2687         ring->rx_mini_pending = 0;
2688         ring->rx_jumbo_pending = 0;
2689 }
2690
2691 static int ibmvnic_set_ringparam(struct net_device *netdev,
2692                                  struct ethtool_ringparam *ring)
2693 {
2694         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2695         int ret;
2696
2697         ret = 0;
2698         adapter->desired.rx_entries = ring->rx_pending;
2699         adapter->desired.tx_entries = ring->tx_pending;
2700
2701         ret = wait_for_reset(adapter);
2702
2703         if (!ret &&
2704             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2705              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2706                 netdev_info(netdev,
2707                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2708                             ring->rx_pending, ring->tx_pending,
2709                             adapter->req_rx_add_entries_per_subcrq,
2710                             adapter->req_tx_entries_per_subcrq);
2711         return ret;
2712 }
2713
2714 static void ibmvnic_get_channels(struct net_device *netdev,
2715                                  struct ethtool_channels *channels)
2716 {
2717         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2718
2719         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2720                 channels->max_rx = adapter->max_rx_queues;
2721                 channels->max_tx = adapter->max_tx_queues;
2722         } else {
2723                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2724                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2725         }
2726
2727         channels->max_other = 0;
2728         channels->max_combined = 0;
2729         channels->rx_count = adapter->req_rx_queues;
2730         channels->tx_count = adapter->req_tx_queues;
2731         channels->other_count = 0;
2732         channels->combined_count = 0;
2733 }
2734
2735 static int ibmvnic_set_channels(struct net_device *netdev,
2736                                 struct ethtool_channels *channels)
2737 {
2738         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2739         int ret;
2740
2741         ret = 0;
2742         adapter->desired.rx_queues = channels->rx_count;
2743         adapter->desired.tx_queues = channels->tx_count;
2744
2745         ret = wait_for_reset(adapter);
2746
2747         if (!ret &&
2748             (adapter->req_rx_queues != channels->rx_count ||
2749              adapter->req_tx_queues != channels->tx_count))
2750                 netdev_info(netdev,
2751                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2752                             channels->rx_count, channels->tx_count,
2753                             adapter->req_rx_queues, adapter->req_tx_queues);
2754         return ret;
2755
2756 }
2757
2758 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2759 {
2760         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2761         int i;
2762
2763         switch (stringset) {
2764         case ETH_SS_STATS:
2765                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2766                                 i++, data += ETH_GSTRING_LEN)
2767                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2768
2769                 for (i = 0; i < adapter->req_tx_queues; i++) {
2770                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2771                         data += ETH_GSTRING_LEN;
2772
2773                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2774                         data += ETH_GSTRING_LEN;
2775
2776                         snprintf(data, ETH_GSTRING_LEN,
2777                                  "tx%d_dropped_packets", i);
2778                         data += ETH_GSTRING_LEN;
2779                 }
2780
2781                 for (i = 0; i < adapter->req_rx_queues; i++) {
2782                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2783                         data += ETH_GSTRING_LEN;
2784
2785                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2786                         data += ETH_GSTRING_LEN;
2787
2788                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2789                         data += ETH_GSTRING_LEN;
2790                 }
2791                 break;
2792
2793         case ETH_SS_PRIV_FLAGS:
2794                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2795                         strcpy(data + i * ETH_GSTRING_LEN,
2796                                ibmvnic_priv_flags[i]);
2797                 break;
2798         default:
2799                 return;
2800         }
2801 }
2802
2803 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2804 {
2805         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2806
2807         switch (sset) {
2808         case ETH_SS_STATS:
2809                 return ARRAY_SIZE(ibmvnic_stats) +
2810                        adapter->req_tx_queues * NUM_TX_STATS +
2811                        adapter->req_rx_queues * NUM_RX_STATS;
2812         case ETH_SS_PRIV_FLAGS:
2813                 return ARRAY_SIZE(ibmvnic_priv_flags);
2814         default:
2815                 return -EOPNOTSUPP;
2816         }
2817 }
2818
2819 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2820                                       struct ethtool_stats *stats, u64 *data)
2821 {
2822         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2823         union ibmvnic_crq crq;
2824         int i, j;
2825         int rc;
2826
2827         memset(&crq, 0, sizeof(crq));
2828         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2829         crq.request_statistics.cmd = REQUEST_STATISTICS;
2830         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2831         crq.request_statistics.len =
2832             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2833
2834         /* Wait for data to be written */
2835         reinit_completion(&adapter->stats_done);
2836         rc = ibmvnic_send_crq(adapter, &crq);
2837         if (rc)
2838                 return;
2839         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2840         if (rc)
2841                 return;
2842
2843         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2844                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2845                                                 ibmvnic_stats[i].offset));
2846
2847         for (j = 0; j < adapter->req_tx_queues; j++) {
2848                 data[i] = adapter->tx_stats_buffers[j].packets;
2849                 i++;
2850                 data[i] = adapter->tx_stats_buffers[j].bytes;
2851                 i++;
2852                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2853                 i++;
2854         }
2855
2856         for (j = 0; j < adapter->req_rx_queues; j++) {
2857                 data[i] = adapter->rx_stats_buffers[j].packets;
2858                 i++;
2859                 data[i] = adapter->rx_stats_buffers[j].bytes;
2860                 i++;
2861                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2862                 i++;
2863         }
2864 }
2865
2866 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2867 {
2868         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2869
2870         return adapter->priv_flags;
2871 }
2872
2873 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2874 {
2875         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2876         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2877
2878         if (which_maxes)
2879                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2880         else
2881                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2882
2883         return 0;
2884 }
2885 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2886         .get_drvinfo            = ibmvnic_get_drvinfo,
2887         .get_msglevel           = ibmvnic_get_msglevel,
2888         .set_msglevel           = ibmvnic_set_msglevel,
2889         .get_link               = ibmvnic_get_link,
2890         .get_ringparam          = ibmvnic_get_ringparam,
2891         .set_ringparam          = ibmvnic_set_ringparam,
2892         .get_channels           = ibmvnic_get_channels,
2893         .set_channels           = ibmvnic_set_channels,
2894         .get_strings            = ibmvnic_get_strings,
2895         .get_sset_count         = ibmvnic_get_sset_count,
2896         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2897         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2898         .get_priv_flags         = ibmvnic_get_priv_flags,
2899         .set_priv_flags         = ibmvnic_set_priv_flags,
2900 };
2901
2902 /* Routines for managing CRQs/sCRQs  */
2903
2904 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2905                                    struct ibmvnic_sub_crq_queue *scrq)
2906 {
2907         int rc;
2908
2909         if (!scrq) {
2910                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2911                 return -EINVAL;
2912         }
2913
2914         if (scrq->irq) {
2915                 free_irq(scrq->irq, scrq);
2916                 irq_dispose_mapping(scrq->irq);
2917                 scrq->irq = 0;
2918         }
2919
2920         if (scrq->msgs) {
2921                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2922                 atomic_set(&scrq->used, 0);
2923                 scrq->cur = 0;
2924                 scrq->ind_buf.index = 0;
2925         } else {
2926                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2927                 return -EINVAL;
2928         }
2929
2930         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2931                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2932         return rc;
2933 }
2934
2935 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2936 {
2937         int i, rc;
2938
2939         if (!adapter->tx_scrq || !adapter->rx_scrq)
2940                 return -EINVAL;
2941
2942         for (i = 0; i < adapter->req_tx_queues; i++) {
2943                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2944                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2945                 if (rc)
2946                         return rc;
2947         }
2948
2949         for (i = 0; i < adapter->req_rx_queues; i++) {
2950                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2951                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2952                 if (rc)
2953                         return rc;
2954         }
2955
2956         return rc;
2957 }
2958
2959 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2960                                   struct ibmvnic_sub_crq_queue *scrq,
2961                                   bool do_h_free)
2962 {
2963         struct device *dev = &adapter->vdev->dev;
2964         long rc;
2965
2966         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2967
2968         if (do_h_free) {
2969                 /* Close the sub-crqs */
2970                 do {
2971                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2972                                                 adapter->vdev->unit_address,
2973                                                 scrq->crq_num);
2974                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2975
2976                 if (rc) {
2977                         netdev_err(adapter->netdev,
2978                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2979                                    scrq->crq_num, rc);
2980                 }
2981         }
2982
2983         dma_free_coherent(dev,
2984                           IBMVNIC_IND_ARR_SZ,
2985                           scrq->ind_buf.indir_arr,
2986                           scrq->ind_buf.indir_dma);
2987
2988         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2989                          DMA_BIDIRECTIONAL);
2990         free_pages((unsigned long)scrq->msgs, 2);
2991         kfree(scrq);
2992 }
2993
2994 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2995                                                         *adapter)
2996 {
2997         struct device *dev = &adapter->vdev->dev;
2998         struct ibmvnic_sub_crq_queue *scrq;
2999         int rc;
3000
3001         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3002         if (!scrq)
3003                 return NULL;
3004
3005         scrq->msgs =
3006                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3007         if (!scrq->msgs) {
3008                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3009                 goto zero_page_failed;
3010         }
3011
3012         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3013                                          DMA_BIDIRECTIONAL);
3014         if (dma_mapping_error(dev, scrq->msg_token)) {
3015                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3016                 goto map_failed;
3017         }
3018
3019         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3020                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3021
3022         if (rc == H_RESOURCE)
3023                 rc = ibmvnic_reset_crq(adapter);
3024
3025         if (rc == H_CLOSED) {
3026                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3027         } else if (rc) {
3028                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3029                 goto reg_failed;
3030         }
3031
3032         scrq->adapter = adapter;
3033         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3034         scrq->ind_buf.index = 0;
3035
3036         scrq->ind_buf.indir_arr =
3037                 dma_alloc_coherent(dev,
3038                                    IBMVNIC_IND_ARR_SZ,
3039                                    &scrq->ind_buf.indir_dma,
3040                                    GFP_KERNEL);
3041
3042         if (!scrq->ind_buf.indir_arr)
3043                 goto indir_failed;
3044
3045         spin_lock_init(&scrq->lock);
3046
3047         netdev_dbg(adapter->netdev,
3048                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3049                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3050
3051         return scrq;
3052
3053 indir_failed:
3054         do {
3055                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3056                                         adapter->vdev->unit_address,
3057                                         scrq->crq_num);
3058         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3059 reg_failed:
3060         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3061                          DMA_BIDIRECTIONAL);
3062 map_failed:
3063         free_pages((unsigned long)scrq->msgs, 2);
3064 zero_page_failed:
3065         kfree(scrq);
3066
3067         return NULL;
3068 }
3069
3070 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3071 {
3072         int i;
3073
3074         if (adapter->tx_scrq) {
3075                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3076                         if (!adapter->tx_scrq[i])
3077                                 continue;
3078
3079                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3080                                    i);
3081                         if (adapter->tx_scrq[i]->irq) {
3082                                 free_irq(adapter->tx_scrq[i]->irq,
3083                                          adapter->tx_scrq[i]);
3084                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3085                                 adapter->tx_scrq[i]->irq = 0;
3086                         }
3087
3088                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3089                                               do_h_free);
3090                 }
3091
3092                 kfree(adapter->tx_scrq);
3093                 adapter->tx_scrq = NULL;
3094                 adapter->num_active_tx_scrqs = 0;
3095         }
3096
3097         if (adapter->rx_scrq) {
3098                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3099                         if (!adapter->rx_scrq[i])
3100                                 continue;
3101
3102                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3103                                    i);
3104                         if (adapter->rx_scrq[i]->irq) {
3105                                 free_irq(adapter->rx_scrq[i]->irq,
3106                                          adapter->rx_scrq[i]);
3107                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3108                                 adapter->rx_scrq[i]->irq = 0;
3109                         }
3110
3111                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3112                                               do_h_free);
3113                 }
3114
3115                 kfree(adapter->rx_scrq);
3116                 adapter->rx_scrq = NULL;
3117                 adapter->num_active_rx_scrqs = 0;
3118         }
3119 }
3120
3121 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3122                             struct ibmvnic_sub_crq_queue *scrq)
3123 {
3124         struct device *dev = &adapter->vdev->dev;
3125         unsigned long rc;
3126
3127         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3128                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3129         if (rc)
3130                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3131                         scrq->hw_irq, rc);
3132         return rc;
3133 }
3134
3135 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3136                            struct ibmvnic_sub_crq_queue *scrq)
3137 {
3138         struct device *dev = &adapter->vdev->dev;
3139         unsigned long rc;
3140
3141         if (scrq->hw_irq > 0x100000000ULL) {
3142                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3143                 return 1;
3144         }
3145
3146         if (test_bit(0, &adapter->resetting) &&
3147             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3148                 u64 val = (0xff000000) | scrq->hw_irq;
3149
3150                 rc = plpar_hcall_norets(H_EOI, val);
3151                 /* H_EOI would fail with rc = H_FUNCTION when running
3152                  * in XIVE mode which is expected, but not an error.
3153                  */
3154                 if (rc && (rc != H_FUNCTION))
3155                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3156                                 val, rc);
3157         }
3158
3159         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3160                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3161         if (rc)
3162                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3163                         scrq->hw_irq, rc);
3164         return rc;
3165 }
3166
3167 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3168                                struct ibmvnic_sub_crq_queue *scrq)
3169 {
3170         struct device *dev = &adapter->vdev->dev;
3171         struct ibmvnic_tx_pool *tx_pool;
3172         struct ibmvnic_tx_buff *txbuff;
3173         struct netdev_queue *txq;
3174         union sub_crq *next;
3175         int index;
3176         int i;
3177
3178 restart_loop:
3179         while (pending_scrq(adapter, scrq)) {
3180                 unsigned int pool = scrq->pool_index;
3181                 int num_entries = 0;
3182                 int total_bytes = 0;
3183                 int num_packets = 0;
3184
3185                 next = ibmvnic_next_scrq(adapter, scrq);
3186                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3187                         if (next->tx_comp.rcs[i])
3188                                 dev_err(dev, "tx error %x\n",
3189                                         next->tx_comp.rcs[i]);
3190                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3191                         if (index & IBMVNIC_TSO_POOL_MASK) {
3192                                 tx_pool = &adapter->tso_pool[pool];
3193                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3194                         } else {
3195                                 tx_pool = &adapter->tx_pool[pool];
3196                         }
3197
3198                         txbuff = &tx_pool->tx_buff[index];
3199                         num_packets++;
3200                         num_entries += txbuff->num_entries;
3201                         if (txbuff->skb) {
3202                                 total_bytes += txbuff->skb->len;
3203                                 dev_consume_skb_irq(txbuff->skb);
3204                                 txbuff->skb = NULL;
3205                         } else {
3206                                 netdev_warn(adapter->netdev,
3207                                             "TX completion received with NULL socket buffer\n");
3208                         }
3209                         tx_pool->free_map[tx_pool->producer_index] = index;
3210                         tx_pool->producer_index =
3211                                 (tx_pool->producer_index + 1) %
3212                                         tx_pool->num_buffers;
3213                 }
3214                 /* remove tx_comp scrq*/
3215                 next->tx_comp.first = 0;
3216
3217                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3218                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3219
3220                 if (atomic_sub_return(num_entries, &scrq->used) <=
3221                     (adapter->req_tx_entries_per_subcrq / 2) &&
3222                     __netif_subqueue_stopped(adapter->netdev,
3223                                              scrq->pool_index)) {
3224                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3225                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3226                                    scrq->pool_index);
3227                 }
3228         }
3229
3230         enable_scrq_irq(adapter, scrq);
3231
3232         if (pending_scrq(adapter, scrq)) {
3233                 disable_scrq_irq(adapter, scrq);
3234                 goto restart_loop;
3235         }
3236
3237         return 0;
3238 }
3239
3240 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3241 {
3242         struct ibmvnic_sub_crq_queue *scrq = instance;
3243         struct ibmvnic_adapter *adapter = scrq->adapter;
3244
3245         disable_scrq_irq(adapter, scrq);
3246         ibmvnic_complete_tx(adapter, scrq);
3247
3248         return IRQ_HANDLED;
3249 }
3250
3251 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3252 {
3253         struct ibmvnic_sub_crq_queue *scrq = instance;
3254         struct ibmvnic_adapter *adapter = scrq->adapter;
3255
3256         /* When booting a kdump kernel we can hit pending interrupts
3257          * prior to completing driver initialization.
3258          */
3259         if (unlikely(adapter->state != VNIC_OPEN))
3260                 return IRQ_NONE;
3261
3262         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3263
3264         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3265                 disable_scrq_irq(adapter, scrq);
3266                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3267         }
3268
3269         return IRQ_HANDLED;
3270 }
3271
3272 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3273 {
3274         struct device *dev = &adapter->vdev->dev;
3275         struct ibmvnic_sub_crq_queue *scrq;
3276         int i = 0, j = 0;
3277         int rc = 0;
3278
3279         for (i = 0; i < adapter->req_tx_queues; i++) {
3280                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3281                            i);
3282                 scrq = adapter->tx_scrq[i];
3283                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3284
3285                 if (!scrq->irq) {
3286                         rc = -EINVAL;
3287                         dev_err(dev, "Error mapping irq\n");
3288                         goto req_tx_irq_failed;
3289                 }
3290
3291                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3292                          adapter->vdev->unit_address, i);
3293                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3294                                  0, scrq->name, scrq);
3295
3296                 if (rc) {
3297                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3298                                 scrq->irq, rc);
3299                         irq_dispose_mapping(scrq->irq);
3300                         goto req_tx_irq_failed;
3301                 }
3302         }
3303
3304         for (i = 0; i < adapter->req_rx_queues; i++) {
3305                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3306                            i);
3307                 scrq = adapter->rx_scrq[i];
3308                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3309                 if (!scrq->irq) {
3310                         rc = -EINVAL;
3311                         dev_err(dev, "Error mapping irq\n");
3312                         goto req_rx_irq_failed;
3313                 }
3314                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3315                          adapter->vdev->unit_address, i);
3316                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3317                                  0, scrq->name, scrq);
3318                 if (rc) {
3319                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3320                                 scrq->irq, rc);
3321                         irq_dispose_mapping(scrq->irq);
3322                         goto req_rx_irq_failed;
3323                 }
3324         }
3325         return rc;
3326
3327 req_rx_irq_failed:
3328         for (j = 0; j < i; j++) {
3329                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3330                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3331         }
3332         i = adapter->req_tx_queues;
3333 req_tx_irq_failed:
3334         for (j = 0; j < i; j++) {
3335                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3336                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3337         }
3338         release_sub_crqs(adapter, 1);
3339         return rc;
3340 }
3341
3342 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3343 {
3344         struct device *dev = &adapter->vdev->dev;
3345         struct ibmvnic_sub_crq_queue **allqueues;
3346         int registered_queues = 0;
3347         int total_queues;
3348         int more = 0;
3349         int i;
3350
3351         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3352
3353         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3354         if (!allqueues)
3355                 return -1;
3356
3357         for (i = 0; i < total_queues; i++) {
3358                 allqueues[i] = init_sub_crq_queue(adapter);
3359                 if (!allqueues[i]) {
3360                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3361                         break;
3362                 }
3363                 registered_queues++;
3364         }
3365
3366         /* Make sure we were able to register the minimum number of queues */
3367         if (registered_queues <
3368             adapter->min_tx_queues + adapter->min_rx_queues) {
3369                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3370                 goto tx_failed;
3371         }
3372
3373         /* Distribute the failed allocated queues*/
3374         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3375                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3376                 switch (i % 3) {
3377                 case 0:
3378                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3379                                 adapter->req_rx_queues--;
3380                         else
3381                                 more++;
3382                         break;
3383                 case 1:
3384                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3385                                 adapter->req_tx_queues--;
3386                         else
3387                                 more++;
3388                         break;
3389                 }
3390         }
3391
3392         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3393                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3394         if (!adapter->tx_scrq)
3395                 goto tx_failed;
3396
3397         for (i = 0; i < adapter->req_tx_queues; i++) {
3398                 adapter->tx_scrq[i] = allqueues[i];
3399                 adapter->tx_scrq[i]->pool_index = i;
3400                 adapter->num_active_tx_scrqs++;
3401         }
3402
3403         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3404                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3405         if (!adapter->rx_scrq)
3406                 goto rx_failed;
3407
3408         for (i = 0; i < adapter->req_rx_queues; i++) {
3409                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3410                 adapter->rx_scrq[i]->scrq_num = i;
3411                 adapter->num_active_rx_scrqs++;
3412         }
3413
3414         kfree(allqueues);
3415         return 0;
3416
3417 rx_failed:
3418         kfree(adapter->tx_scrq);
3419         adapter->tx_scrq = NULL;
3420 tx_failed:
3421         for (i = 0; i < registered_queues; i++)
3422                 release_sub_crq_queue(adapter, allqueues[i], 1);
3423         kfree(allqueues);
3424         return -1;
3425 }
3426
3427 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3428 {
3429         struct device *dev = &adapter->vdev->dev;
3430         union ibmvnic_crq crq;
3431         int max_entries;
3432
3433         if (!retry) {
3434                 /* Sub-CRQ entries are 32 byte long */
3435                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3436
3437                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3438                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3439                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3440                         return;
3441                 }
3442
3443                 if (adapter->desired.mtu)
3444                         adapter->req_mtu = adapter->desired.mtu;
3445                 else
3446                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3447
3448                 if (!adapter->desired.tx_entries)
3449                         adapter->desired.tx_entries =
3450                                         adapter->max_tx_entries_per_subcrq;
3451                 if (!adapter->desired.rx_entries)
3452                         adapter->desired.rx_entries =
3453                                         adapter->max_rx_add_entries_per_subcrq;
3454
3455                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3456                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3457
3458                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3459                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3460                         adapter->desired.tx_entries = max_entries;
3461                 }
3462
3463                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3464                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3465                         adapter->desired.rx_entries = max_entries;
3466                 }
3467
3468                 if (adapter->desired.tx_entries)
3469                         adapter->req_tx_entries_per_subcrq =
3470                                         adapter->desired.tx_entries;
3471                 else
3472                         adapter->req_tx_entries_per_subcrq =
3473                                         adapter->max_tx_entries_per_subcrq;
3474
3475                 if (adapter->desired.rx_entries)
3476                         adapter->req_rx_add_entries_per_subcrq =
3477                                         adapter->desired.rx_entries;
3478                 else
3479                         adapter->req_rx_add_entries_per_subcrq =
3480                                         adapter->max_rx_add_entries_per_subcrq;
3481
3482                 if (adapter->desired.tx_queues)
3483                         adapter->req_tx_queues =
3484                                         adapter->desired.tx_queues;
3485                 else
3486                         adapter->req_tx_queues =
3487                                         adapter->opt_tx_comp_sub_queues;
3488
3489                 if (adapter->desired.rx_queues)
3490                         adapter->req_rx_queues =
3491                                         adapter->desired.rx_queues;
3492                 else
3493                         adapter->req_rx_queues =
3494                                         adapter->opt_rx_comp_queues;
3495
3496                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3497         }
3498
3499         memset(&crq, 0, sizeof(crq));
3500         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3501         crq.request_capability.cmd = REQUEST_CAPABILITY;
3502
3503         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3504         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3505         atomic_inc(&adapter->running_cap_crqs);
3506         ibmvnic_send_crq(adapter, &crq);
3507
3508         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3509         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3510         atomic_inc(&adapter->running_cap_crqs);
3511         ibmvnic_send_crq(adapter, &crq);
3512
3513         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3514         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3515         atomic_inc(&adapter->running_cap_crqs);
3516         ibmvnic_send_crq(adapter, &crq);
3517
3518         crq.request_capability.capability =
3519             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3520         crq.request_capability.number =
3521             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3522         atomic_inc(&adapter->running_cap_crqs);
3523         ibmvnic_send_crq(adapter, &crq);
3524
3525         crq.request_capability.capability =
3526             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3527         crq.request_capability.number =
3528             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3529         atomic_inc(&adapter->running_cap_crqs);
3530         ibmvnic_send_crq(adapter, &crq);
3531
3532         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3533         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3534         atomic_inc(&adapter->running_cap_crqs);
3535         ibmvnic_send_crq(adapter, &crq);
3536
3537         if (adapter->netdev->flags & IFF_PROMISC) {
3538                 if (adapter->promisc_supported) {
3539                         crq.request_capability.capability =
3540                             cpu_to_be16(PROMISC_REQUESTED);
3541                         crq.request_capability.number = cpu_to_be64(1);
3542                         atomic_inc(&adapter->running_cap_crqs);
3543                         ibmvnic_send_crq(adapter, &crq);
3544                 }
3545         } else {
3546                 crq.request_capability.capability =
3547                     cpu_to_be16(PROMISC_REQUESTED);
3548                 crq.request_capability.number = cpu_to_be64(0);
3549                 atomic_inc(&adapter->running_cap_crqs);
3550                 ibmvnic_send_crq(adapter, &crq);
3551         }
3552 }
3553
3554 static int pending_scrq(struct ibmvnic_adapter *adapter,
3555                         struct ibmvnic_sub_crq_queue *scrq)
3556 {
3557         union sub_crq *entry = &scrq->msgs[scrq->cur];
3558         int rc;
3559
3560         rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3561
3562         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3563          * contents of the SCRQ descriptor
3564          */
3565         dma_rmb();
3566
3567         return rc;
3568 }
3569
3570 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3571                                         struct ibmvnic_sub_crq_queue *scrq)
3572 {
3573         union sub_crq *entry;
3574         unsigned long flags;
3575
3576         spin_lock_irqsave(&scrq->lock, flags);
3577         entry = &scrq->msgs[scrq->cur];
3578         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3579                 if (++scrq->cur == scrq->size)
3580                         scrq->cur = 0;
3581         } else {
3582                 entry = NULL;
3583         }
3584         spin_unlock_irqrestore(&scrq->lock, flags);
3585
3586         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3587          * contents of the SCRQ descriptor
3588          */
3589         dma_rmb();
3590
3591         return entry;
3592 }
3593
3594 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3595 {
3596         struct ibmvnic_crq_queue *queue = &adapter->crq;
3597         union ibmvnic_crq *crq;
3598
3599         crq = &queue->msgs[queue->cur];
3600         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3601                 if (++queue->cur == queue->size)
3602                         queue->cur = 0;
3603         } else {
3604                 crq = NULL;
3605         }
3606
3607         return crq;
3608 }
3609
3610 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3611 {
3612         switch (rc) {
3613         case H_PARAMETER:
3614                 dev_warn_ratelimited(dev,
3615                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3616                                      func, rc);
3617                 break;
3618         case H_CLOSED:
3619                 dev_warn_ratelimited(dev,
3620                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3621                                      func, rc);
3622                 break;
3623         default:
3624                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3625                 break;
3626         }
3627 }
3628
3629 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3630                                 u64 remote_handle, u64 ioba, u64 num_entries)
3631 {
3632         unsigned int ua = adapter->vdev->unit_address;
3633         struct device *dev = &adapter->vdev->dev;
3634         int rc;
3635
3636         /* Make sure the hypervisor sees the complete request */
3637         mb();
3638         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3639                                 cpu_to_be64(remote_handle),
3640                                 ioba, num_entries);
3641
3642         if (rc)
3643                 print_subcrq_error(dev, rc, __func__);
3644
3645         return rc;
3646 }
3647
3648 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3649                             union ibmvnic_crq *crq)
3650 {
3651         unsigned int ua = adapter->vdev->unit_address;
3652         struct device *dev = &adapter->vdev->dev;
3653         u64 *u64_crq = (u64 *)crq;
3654         int rc;
3655
3656         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3657                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3658                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3659
3660         if (!adapter->crq.active &&
3661             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3662                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3663                 return -EINVAL;
3664         }
3665
3666         /* Make sure the hypervisor sees the complete request */
3667         mb();
3668
3669         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3670                                 cpu_to_be64(u64_crq[0]),
3671                                 cpu_to_be64(u64_crq[1]));
3672
3673         if (rc) {
3674                 if (rc == H_CLOSED) {
3675                         dev_warn(dev, "CRQ Queue closed\n");
3676                         /* do not reset, report the fail, wait for passive init from server */
3677                 }
3678
3679                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3680         }
3681
3682         return rc;
3683 }
3684
3685 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3686 {
3687         struct device *dev = &adapter->vdev->dev;
3688         union ibmvnic_crq crq;
3689         int retries = 100;
3690         int rc;
3691
3692         memset(&crq, 0, sizeof(crq));
3693         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3694         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3695         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3696
3697         do {
3698                 rc = ibmvnic_send_crq(adapter, &crq);
3699                 if (rc != H_CLOSED)
3700                         break;
3701                 retries--;
3702                 msleep(50);
3703
3704         } while (retries > 0);
3705
3706         if (rc) {
3707                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3708                 return rc;
3709         }
3710
3711         return 0;
3712 }
3713
3714 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3715 {
3716         union ibmvnic_crq crq;
3717
3718         memset(&crq, 0, sizeof(crq));
3719         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3720         crq.version_exchange.cmd = VERSION_EXCHANGE;
3721         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3722
3723         return ibmvnic_send_crq(adapter, &crq);
3724 }
3725
3726 struct vnic_login_client_data {
3727         u8      type;
3728         __be16  len;
3729         char    name[];
3730 } __packed;
3731
3732 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3733 {
3734         int len;
3735
3736         /* Calculate the amount of buffer space needed for the
3737          * vnic client data in the login buffer. There are four entries,
3738          * OS name, LPAR name, device name, and a null last entry.
3739          */
3740         len = 4 * sizeof(struct vnic_login_client_data);
3741         len += 6; /* "Linux" plus NULL */
3742         len += strlen(utsname()->nodename) + 1;
3743         len += strlen(adapter->netdev->name) + 1;
3744
3745         return len;
3746 }
3747
3748 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3749                                  struct vnic_login_client_data *vlcd)
3750 {
3751         const char *os_name = "Linux";
3752         int len;
3753
3754         /* Type 1 - LPAR OS */
3755         vlcd->type = 1;
3756         len = strlen(os_name) + 1;
3757         vlcd->len = cpu_to_be16(len);
3758         strncpy(vlcd->name, os_name, len);
3759         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3760
3761         /* Type 2 - LPAR name */
3762         vlcd->type = 2;
3763         len = strlen(utsname()->nodename) + 1;
3764         vlcd->len = cpu_to_be16(len);
3765         strncpy(vlcd->name, utsname()->nodename, len);
3766         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3767
3768         /* Type 3 - device name */
3769         vlcd->type = 3;
3770         len = strlen(adapter->netdev->name) + 1;
3771         vlcd->len = cpu_to_be16(len);
3772         strncpy(vlcd->name, adapter->netdev->name, len);
3773 }
3774
3775 static int send_login(struct ibmvnic_adapter *adapter)
3776 {
3777         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3778         struct ibmvnic_login_buffer *login_buffer;
3779         struct device *dev = &adapter->vdev->dev;
3780         struct vnic_login_client_data *vlcd;
3781         dma_addr_t rsp_buffer_token;
3782         dma_addr_t buffer_token;
3783         size_t rsp_buffer_size;
3784         union ibmvnic_crq crq;
3785         int client_data_len;
3786         size_t buffer_size;
3787         __be64 *tx_list_p;
3788         __be64 *rx_list_p;
3789         int rc;
3790         int i;
3791
3792         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3793                 netdev_err(adapter->netdev,
3794                            "RX or TX queues are not allocated, device login failed\n");
3795                 return -1;
3796         }
3797
3798         release_login_buffer(adapter);
3799         release_login_rsp_buffer(adapter);
3800
3801         client_data_len = vnic_client_data_len(adapter);
3802
3803         buffer_size =
3804             sizeof(struct ibmvnic_login_buffer) +
3805             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3806             client_data_len;
3807
3808         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3809         if (!login_buffer)
3810                 goto buf_alloc_failed;
3811
3812         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3813                                       DMA_TO_DEVICE);
3814         if (dma_mapping_error(dev, buffer_token)) {
3815                 dev_err(dev, "Couldn't map login buffer\n");
3816                 goto buf_map_failed;
3817         }
3818
3819         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3820                           sizeof(u64) * adapter->req_tx_queues +
3821                           sizeof(u64) * adapter->req_rx_queues +
3822                           sizeof(u64) * adapter->req_rx_queues +
3823                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3824
3825         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3826         if (!login_rsp_buffer)
3827                 goto buf_rsp_alloc_failed;
3828
3829         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3830                                           rsp_buffer_size, DMA_FROM_DEVICE);
3831         if (dma_mapping_error(dev, rsp_buffer_token)) {
3832                 dev_err(dev, "Couldn't map login rsp buffer\n");
3833                 goto buf_rsp_map_failed;
3834         }
3835
3836         adapter->login_buf = login_buffer;
3837         adapter->login_buf_token = buffer_token;
3838         adapter->login_buf_sz = buffer_size;
3839         adapter->login_rsp_buf = login_rsp_buffer;
3840         adapter->login_rsp_buf_token = rsp_buffer_token;
3841         adapter->login_rsp_buf_sz = rsp_buffer_size;
3842
3843         login_buffer->len = cpu_to_be32(buffer_size);
3844         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3845         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3846         login_buffer->off_txcomp_subcrqs =
3847             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3848         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3849         login_buffer->off_rxcomp_subcrqs =
3850             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3851                         sizeof(u64) * adapter->req_tx_queues);
3852         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3853         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3854
3855         tx_list_p = (__be64 *)((char *)login_buffer +
3856                                       sizeof(struct ibmvnic_login_buffer));
3857         rx_list_p = (__be64 *)((char *)login_buffer +
3858                                       sizeof(struct ibmvnic_login_buffer) +
3859                                       sizeof(u64) * adapter->req_tx_queues);
3860
3861         for (i = 0; i < adapter->req_tx_queues; i++) {
3862                 if (adapter->tx_scrq[i]) {
3863                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3864                                                    crq_num);
3865                 }
3866         }
3867
3868         for (i = 0; i < adapter->req_rx_queues; i++) {
3869                 if (adapter->rx_scrq[i]) {
3870                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3871                                                    crq_num);
3872                 }
3873         }
3874
3875         /* Insert vNIC login client data */
3876         vlcd = (struct vnic_login_client_data *)
3877                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3878         login_buffer->client_data_offset =
3879                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3880         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3881
3882         vnic_add_client_data(adapter, vlcd);
3883
3884         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3885         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3886                 netdev_dbg(adapter->netdev, "%016lx\n",
3887                            ((unsigned long int *)(adapter->login_buf))[i]);
3888         }
3889
3890         memset(&crq, 0, sizeof(crq));
3891         crq.login.first = IBMVNIC_CRQ_CMD;
3892         crq.login.cmd = LOGIN;
3893         crq.login.ioba = cpu_to_be32(buffer_token);
3894         crq.login.len = cpu_to_be32(buffer_size);
3895
3896         adapter->login_pending = true;
3897         rc = ibmvnic_send_crq(adapter, &crq);
3898         if (rc) {
3899                 adapter->login_pending = false;
3900                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3901                 goto buf_rsp_map_failed;
3902         }
3903
3904         return 0;
3905
3906 buf_rsp_map_failed:
3907         kfree(login_rsp_buffer);
3908         adapter->login_rsp_buf = NULL;
3909 buf_rsp_alloc_failed:
3910         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3911 buf_map_failed:
3912         kfree(login_buffer);
3913         adapter->login_buf = NULL;
3914 buf_alloc_failed:
3915         return -1;
3916 }
3917
3918 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3919                             u32 len, u8 map_id)
3920 {
3921         union ibmvnic_crq crq;
3922
3923         memset(&crq, 0, sizeof(crq));
3924         crq.request_map.first = IBMVNIC_CRQ_CMD;
3925         crq.request_map.cmd = REQUEST_MAP;
3926         crq.request_map.map_id = map_id;
3927         crq.request_map.ioba = cpu_to_be32(addr);
3928         crq.request_map.len = cpu_to_be32(len);
3929         return ibmvnic_send_crq(adapter, &crq);
3930 }
3931
3932 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3933 {
3934         union ibmvnic_crq crq;
3935
3936         memset(&crq, 0, sizeof(crq));
3937         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3938         crq.request_unmap.cmd = REQUEST_UNMAP;
3939         crq.request_unmap.map_id = map_id;
3940         return ibmvnic_send_crq(adapter, &crq);
3941 }
3942
3943 static void send_query_map(struct ibmvnic_adapter *adapter)
3944 {
3945         union ibmvnic_crq crq;
3946
3947         memset(&crq, 0, sizeof(crq));
3948         crq.query_map.first = IBMVNIC_CRQ_CMD;
3949         crq.query_map.cmd = QUERY_MAP;
3950         ibmvnic_send_crq(adapter, &crq);
3951 }
3952
3953 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3954 static void send_query_cap(struct ibmvnic_adapter *adapter)
3955 {
3956         union ibmvnic_crq crq;
3957
3958         atomic_set(&adapter->running_cap_crqs, 0);
3959         memset(&crq, 0, sizeof(crq));
3960         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3961         crq.query_capability.cmd = QUERY_CAPABILITY;
3962
3963         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3964         atomic_inc(&adapter->running_cap_crqs);
3965         ibmvnic_send_crq(adapter, &crq);
3966
3967         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3968         atomic_inc(&adapter->running_cap_crqs);
3969         ibmvnic_send_crq(adapter, &crq);
3970
3971         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3972         atomic_inc(&adapter->running_cap_crqs);
3973         ibmvnic_send_crq(adapter, &crq);
3974
3975         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3976         atomic_inc(&adapter->running_cap_crqs);
3977         ibmvnic_send_crq(adapter, &crq);
3978
3979         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3980         atomic_inc(&adapter->running_cap_crqs);
3981         ibmvnic_send_crq(adapter, &crq);
3982
3983         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3984         atomic_inc(&adapter->running_cap_crqs);
3985         ibmvnic_send_crq(adapter, &crq);
3986
3987         crq.query_capability.capability =
3988             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3989         atomic_inc(&adapter->running_cap_crqs);
3990         ibmvnic_send_crq(adapter, &crq);
3991
3992         crq.query_capability.capability =
3993             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3994         atomic_inc(&adapter->running_cap_crqs);
3995         ibmvnic_send_crq(adapter, &crq);
3996
3997         crq.query_capability.capability =
3998             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3999         atomic_inc(&adapter->running_cap_crqs);
4000         ibmvnic_send_crq(adapter, &crq);
4001
4002         crq.query_capability.capability =
4003             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4004         atomic_inc(&adapter->running_cap_crqs);
4005         ibmvnic_send_crq(adapter, &crq);
4006
4007         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4008         atomic_inc(&adapter->running_cap_crqs);
4009         ibmvnic_send_crq(adapter, &crq);
4010
4011         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4012         atomic_inc(&adapter->running_cap_crqs);
4013         ibmvnic_send_crq(adapter, &crq);
4014
4015         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4016         atomic_inc(&adapter->running_cap_crqs);
4017         ibmvnic_send_crq(adapter, &crq);
4018
4019         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4020         atomic_inc(&adapter->running_cap_crqs);
4021         ibmvnic_send_crq(adapter, &crq);
4022
4023         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4024         atomic_inc(&adapter->running_cap_crqs);
4025         ibmvnic_send_crq(adapter, &crq);
4026
4027         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4028         atomic_inc(&adapter->running_cap_crqs);
4029         ibmvnic_send_crq(adapter, &crq);
4030
4031         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4032         atomic_inc(&adapter->running_cap_crqs);
4033         ibmvnic_send_crq(adapter, &crq);
4034
4035         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4036         atomic_inc(&adapter->running_cap_crqs);
4037         ibmvnic_send_crq(adapter, &crq);
4038
4039         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4040         atomic_inc(&adapter->running_cap_crqs);
4041         ibmvnic_send_crq(adapter, &crq);
4042
4043         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4044         atomic_inc(&adapter->running_cap_crqs);
4045         ibmvnic_send_crq(adapter, &crq);
4046
4047         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4048         atomic_inc(&adapter->running_cap_crqs);
4049         ibmvnic_send_crq(adapter, &crq);
4050
4051         crq.query_capability.capability =
4052                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4053         atomic_inc(&adapter->running_cap_crqs);
4054         ibmvnic_send_crq(adapter, &crq);
4055
4056         crq.query_capability.capability =
4057                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4058         atomic_inc(&adapter->running_cap_crqs);
4059         ibmvnic_send_crq(adapter, &crq);
4060
4061         crq.query_capability.capability =
4062                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4063         atomic_inc(&adapter->running_cap_crqs);
4064         ibmvnic_send_crq(adapter, &crq);
4065
4066         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4067         atomic_inc(&adapter->running_cap_crqs);
4068         ibmvnic_send_crq(adapter, &crq);
4069 }
4070
4071 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4072 {
4073         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4074         struct device *dev = &adapter->vdev->dev;
4075         union ibmvnic_crq crq;
4076
4077         adapter->ip_offload_tok =
4078                 dma_map_single(dev,
4079                                &adapter->ip_offload_buf,
4080                                buf_sz,
4081                                DMA_FROM_DEVICE);
4082
4083         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4084                 if (!firmware_has_feature(FW_FEATURE_CMO))
4085                         dev_err(dev, "Couldn't map offload buffer\n");
4086                 return;
4087         }
4088
4089         memset(&crq, 0, sizeof(crq));
4090         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4091         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4092         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4093         crq.query_ip_offload.ioba =
4094             cpu_to_be32(adapter->ip_offload_tok);
4095
4096         ibmvnic_send_crq(adapter, &crq);
4097 }
4098
4099 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4100 {
4101         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4102         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4103         struct device *dev = &adapter->vdev->dev;
4104         netdev_features_t old_hw_features = 0;
4105         union ibmvnic_crq crq;
4106
4107         adapter->ip_offload_ctrl_tok =
4108                 dma_map_single(dev,
4109                                ctrl_buf,
4110                                sizeof(adapter->ip_offload_ctrl),
4111                                DMA_TO_DEVICE);
4112
4113         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4114                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4115                 return;
4116         }
4117
4118         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4119         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4120         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4121         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4122         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4123         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4124         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4125         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4126         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4127         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4128
4129         /* large_rx disabled for now, additional features needed */
4130         ctrl_buf->large_rx_ipv4 = 0;
4131         ctrl_buf->large_rx_ipv6 = 0;
4132
4133         if (adapter->state != VNIC_PROBING) {
4134                 old_hw_features = adapter->netdev->hw_features;
4135                 adapter->netdev->hw_features = 0;
4136         }
4137
4138         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4139
4140         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4141                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4142
4143         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4144                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4145
4146         if ((adapter->netdev->features &
4147             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4148                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4149
4150         if (buf->large_tx_ipv4)
4151                 adapter->netdev->hw_features |= NETIF_F_TSO;
4152         if (buf->large_tx_ipv6)
4153                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4154
4155         if (adapter->state == VNIC_PROBING) {
4156                 adapter->netdev->features |= adapter->netdev->hw_features;
4157         } else if (old_hw_features != adapter->netdev->hw_features) {
4158                 netdev_features_t tmp = 0;
4159
4160                 /* disable features no longer supported */
4161                 adapter->netdev->features &= adapter->netdev->hw_features;
4162                 /* turn on features now supported if previously enabled */
4163                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4164                         adapter->netdev->hw_features;
4165                 adapter->netdev->features |=
4166                                 tmp & adapter->netdev->wanted_features;
4167         }
4168
4169         memset(&crq, 0, sizeof(crq));
4170         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4171         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4172         crq.control_ip_offload.len =
4173             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4174         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4175         ibmvnic_send_crq(adapter, &crq);
4176 }
4177
4178 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4179                                 struct ibmvnic_adapter *adapter)
4180 {
4181         struct device *dev = &adapter->vdev->dev;
4182
4183         if (crq->get_vpd_size_rsp.rc.code) {
4184                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4185                         crq->get_vpd_size_rsp.rc.code);
4186                 complete(&adapter->fw_done);
4187                 return;
4188         }
4189
4190         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4191         complete(&adapter->fw_done);
4192 }
4193
4194 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4195                            struct ibmvnic_adapter *adapter)
4196 {
4197         struct device *dev = &adapter->vdev->dev;
4198         unsigned char *substr = NULL;
4199         u8 fw_level_len = 0;
4200
4201         memset(adapter->fw_version, 0, 32);
4202
4203         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4204                          DMA_FROM_DEVICE);
4205
4206         if (crq->get_vpd_rsp.rc.code) {
4207                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4208                         crq->get_vpd_rsp.rc.code);
4209                 goto complete;
4210         }
4211
4212         /* get the position of the firmware version info
4213          * located after the ASCII 'RM' substring in the buffer
4214          */
4215         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4216         if (!substr) {
4217                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4218                 goto complete;
4219         }
4220
4221         /* get length of firmware level ASCII substring */
4222         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4223                 fw_level_len = *(substr + 2);
4224         } else {
4225                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4226                 goto complete;
4227         }
4228
4229         /* copy firmware version string from vpd into adapter */
4230         if ((substr + 3 + fw_level_len) <
4231             (adapter->vpd->buff + adapter->vpd->len)) {
4232                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4233         } else {
4234                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4235         }
4236
4237 complete:
4238         if (adapter->fw_version[0] == '\0')
4239                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4240         complete(&adapter->fw_done);
4241 }
4242
4243 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4244 {
4245         struct device *dev = &adapter->vdev->dev;
4246         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4247         int i;
4248
4249         dma_unmap_single(dev, adapter->ip_offload_tok,
4250                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4251
4252         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4253         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4254                 netdev_dbg(adapter->netdev, "%016lx\n",
4255                            ((unsigned long int *)(buf))[i]);
4256
4257         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4258         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4259         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4260                    buf->tcp_ipv4_chksum);
4261         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4262                    buf->tcp_ipv6_chksum);
4263         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4264                    buf->udp_ipv4_chksum);
4265         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4266                    buf->udp_ipv6_chksum);
4267         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4268                    buf->large_tx_ipv4);
4269         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4270                    buf->large_tx_ipv6);
4271         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4272                    buf->large_rx_ipv4);
4273         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4274                    buf->large_rx_ipv6);
4275         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4276                    buf->max_ipv4_header_size);
4277         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4278                    buf->max_ipv6_header_size);
4279         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4280                    buf->max_tcp_header_size);
4281         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4282                    buf->max_udp_header_size);
4283         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4284                    buf->max_large_tx_size);
4285         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4286                    buf->max_large_rx_size);
4287         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4288                    buf->ipv6_extension_header);
4289         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4290                    buf->tcp_pseudosum_req);
4291         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4292                    buf->num_ipv6_ext_headers);
4293         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4294                    buf->off_ipv6_ext_headers);
4295
4296         send_control_ip_offload(adapter);
4297 }
4298
4299 static const char *ibmvnic_fw_err_cause(u16 cause)
4300 {
4301         switch (cause) {
4302         case ADAPTER_PROBLEM:
4303                 return "adapter problem";
4304         case BUS_PROBLEM:
4305                 return "bus problem";
4306         case FW_PROBLEM:
4307                 return "firmware problem";
4308         case DD_PROBLEM:
4309                 return "device driver problem";
4310         case EEH_RECOVERY:
4311                 return "EEH recovery";
4312         case FW_UPDATED:
4313                 return "firmware updated";
4314         case LOW_MEMORY:
4315                 return "low Memory";
4316         default:
4317                 return "unknown";
4318         }
4319 }
4320
4321 static void handle_error_indication(union ibmvnic_crq *crq,
4322                                     struct ibmvnic_adapter *adapter)
4323 {
4324         struct device *dev = &adapter->vdev->dev;
4325         u16 cause;
4326
4327         cause = be16_to_cpu(crq->error_indication.error_cause);
4328
4329         dev_warn_ratelimited(dev,
4330                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4331                              crq->error_indication.flags
4332                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4333                              ibmvnic_fw_err_cause(cause));
4334
4335         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4336                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4337         else
4338                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4339 }
4340
4341 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4342                                  struct ibmvnic_adapter *adapter)
4343 {
4344         struct net_device *netdev = adapter->netdev;
4345         struct device *dev = &adapter->vdev->dev;
4346         long rc;
4347
4348         rc = crq->change_mac_addr_rsp.rc.code;
4349         if (rc) {
4350                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4351                 goto out;
4352         }
4353         /* crq->change_mac_addr.mac_addr is the requested one
4354          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4355          */
4356         ether_addr_copy(netdev->dev_addr,
4357                         &crq->change_mac_addr_rsp.mac_addr[0]);
4358         ether_addr_copy(adapter->mac_addr,
4359                         &crq->change_mac_addr_rsp.mac_addr[0]);
4360 out:
4361         complete(&adapter->fw_done);
4362         return rc;
4363 }
4364
4365 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4366                                    struct ibmvnic_adapter *adapter)
4367 {
4368         struct device *dev = &adapter->vdev->dev;
4369         u64 *req_value;
4370         char *name;
4371
4372         atomic_dec(&adapter->running_cap_crqs);
4373         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4374         case REQ_TX_QUEUES:
4375                 req_value = &adapter->req_tx_queues;
4376                 name = "tx";
4377                 break;
4378         case REQ_RX_QUEUES:
4379                 req_value = &adapter->req_rx_queues;
4380                 name = "rx";
4381                 break;
4382         case REQ_RX_ADD_QUEUES:
4383                 req_value = &adapter->req_rx_add_queues;
4384                 name = "rx_add";
4385                 break;
4386         case REQ_TX_ENTRIES_PER_SUBCRQ:
4387                 req_value = &adapter->req_tx_entries_per_subcrq;
4388                 name = "tx_entries_per_subcrq";
4389                 break;
4390         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4391                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4392                 name = "rx_add_entries_per_subcrq";
4393                 break;
4394         case REQ_MTU:
4395                 req_value = &adapter->req_mtu;
4396                 name = "mtu";
4397                 break;
4398         case PROMISC_REQUESTED:
4399                 req_value = &adapter->promisc;
4400                 name = "promisc";
4401                 break;
4402         default:
4403                 dev_err(dev, "Got invalid cap request rsp %d\n",
4404                         crq->request_capability.capability);
4405                 return;
4406         }
4407
4408         switch (crq->request_capability_rsp.rc.code) {
4409         case SUCCESS:
4410                 break;
4411         case PARTIALSUCCESS:
4412                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4413                          *req_value,
4414                          (long int)be64_to_cpu(crq->request_capability_rsp.
4415                                                number), name);
4416
4417                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4418                     REQ_MTU) {
4419                         pr_err("mtu of %llu is not supported. Reverting.\n",
4420                                *req_value);
4421                         *req_value = adapter->fallback.mtu;
4422                 } else {
4423                         *req_value =
4424                                 be64_to_cpu(crq->request_capability_rsp.number);
4425                 }
4426
4427                 send_request_cap(adapter, 1);
4428                 return;
4429         default:
4430                 dev_err(dev, "Error %d in request cap rsp\n",
4431                         crq->request_capability_rsp.rc.code);
4432                 return;
4433         }
4434
4435         /* Done receiving requested capabilities, query IP offload support */
4436         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4437                 adapter->wait_capability = false;
4438                 send_query_ip_offload(adapter);
4439         }
4440 }
4441
4442 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4443                             struct ibmvnic_adapter *adapter)
4444 {
4445         struct device *dev = &adapter->vdev->dev;
4446         struct net_device *netdev = adapter->netdev;
4447         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4448         struct ibmvnic_login_buffer *login = adapter->login_buf;
4449         u64 *tx_handle_array;
4450         u64 *rx_handle_array;
4451         int num_tx_pools;
4452         int num_rx_pools;
4453         u64 *size_array;
4454         int i;
4455
4456         /* CHECK: Test/set of login_pending does not need to be atomic
4457          * because only ibmvnic_tasklet tests/clears this.
4458          */
4459         if (!adapter->login_pending) {
4460                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4461                 return 0;
4462         }
4463         adapter->login_pending = false;
4464
4465         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4466                          DMA_TO_DEVICE);
4467         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4468                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4469
4470         /* If the number of queues requested can't be allocated by the
4471          * server, the login response will return with code 1. We will need
4472          * to resend the login buffer with fewer queues requested.
4473          */
4474         if (login_rsp_crq->generic.rc.code) {
4475                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4476                 complete(&adapter->init_done);
4477                 return 0;
4478         }
4479
4480         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4481
4482         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4483         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4484                 netdev_dbg(adapter->netdev, "%016lx\n",
4485                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4486         }
4487
4488         /* Sanity checks */
4489         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4490             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4491              adapter->req_rx_add_queues !=
4492              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4493                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4494                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4495                 return -EIO;
4496         }
4497         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4498                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4499         /* variable buffer sizes are not supported, so just read the
4500          * first entry.
4501          */
4502         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4503
4504         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4505         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4506
4507         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4508                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4509         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4510                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4511
4512         for (i = 0; i < num_tx_pools; i++)
4513                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4514
4515         for (i = 0; i < num_rx_pools; i++)
4516                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4517
4518         adapter->num_active_tx_scrqs = num_tx_pools;
4519         adapter->num_active_rx_scrqs = num_rx_pools;
4520         release_login_rsp_buffer(adapter);
4521         release_login_buffer(adapter);
4522         complete(&adapter->init_done);
4523
4524         return 0;
4525 }
4526
4527 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4528                                      struct ibmvnic_adapter *adapter)
4529 {
4530         struct device *dev = &adapter->vdev->dev;
4531         long rc;
4532
4533         rc = crq->request_unmap_rsp.rc.code;
4534         if (rc)
4535                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4536 }
4537
4538 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4539                                  struct ibmvnic_adapter *adapter)
4540 {
4541         struct net_device *netdev = adapter->netdev;
4542         struct device *dev = &adapter->vdev->dev;
4543         long rc;
4544
4545         rc = crq->query_map_rsp.rc.code;
4546         if (rc) {
4547                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4548                 return;
4549         }
4550         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4551                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4552                    crq->query_map_rsp.free_pages);
4553 }
4554
4555 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4556                                  struct ibmvnic_adapter *adapter)
4557 {
4558         struct net_device *netdev = adapter->netdev;
4559         struct device *dev = &adapter->vdev->dev;
4560         long rc;
4561
4562         atomic_dec(&adapter->running_cap_crqs);
4563         netdev_dbg(netdev, "Outstanding queries: %d\n",
4564                    atomic_read(&adapter->running_cap_crqs));
4565         rc = crq->query_capability.rc.code;
4566         if (rc) {
4567                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4568                 goto out;
4569         }
4570
4571         switch (be16_to_cpu(crq->query_capability.capability)) {
4572         case MIN_TX_QUEUES:
4573                 adapter->min_tx_queues =
4574                     be64_to_cpu(crq->query_capability.number);
4575                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4576                            adapter->min_tx_queues);
4577                 break;
4578         case MIN_RX_QUEUES:
4579                 adapter->min_rx_queues =
4580                     be64_to_cpu(crq->query_capability.number);
4581                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4582                            adapter->min_rx_queues);
4583                 break;
4584         case MIN_RX_ADD_QUEUES:
4585                 adapter->min_rx_add_queues =
4586                     be64_to_cpu(crq->query_capability.number);
4587                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4588                            adapter->min_rx_add_queues);
4589                 break;
4590         case MAX_TX_QUEUES:
4591                 adapter->max_tx_queues =
4592                     be64_to_cpu(crq->query_capability.number);
4593                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4594                            adapter->max_tx_queues);
4595                 break;
4596         case MAX_RX_QUEUES:
4597                 adapter->max_rx_queues =
4598                     be64_to_cpu(crq->query_capability.number);
4599                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4600                            adapter->max_rx_queues);
4601                 break;
4602         case MAX_RX_ADD_QUEUES:
4603                 adapter->max_rx_add_queues =
4604                     be64_to_cpu(crq->query_capability.number);
4605                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4606                            adapter->max_rx_add_queues);
4607                 break;
4608         case MIN_TX_ENTRIES_PER_SUBCRQ:
4609                 adapter->min_tx_entries_per_subcrq =
4610                     be64_to_cpu(crq->query_capability.number);
4611                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4612                            adapter->min_tx_entries_per_subcrq);
4613                 break;
4614         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4615                 adapter->min_rx_add_entries_per_subcrq =
4616                     be64_to_cpu(crq->query_capability.number);
4617                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4618                            adapter->min_rx_add_entries_per_subcrq);
4619                 break;
4620         case MAX_TX_ENTRIES_PER_SUBCRQ:
4621                 adapter->max_tx_entries_per_subcrq =
4622                     be64_to_cpu(crq->query_capability.number);
4623                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4624                            adapter->max_tx_entries_per_subcrq);
4625                 break;
4626         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4627                 adapter->max_rx_add_entries_per_subcrq =
4628                     be64_to_cpu(crq->query_capability.number);
4629                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4630                            adapter->max_rx_add_entries_per_subcrq);
4631                 break;
4632         case TCP_IP_OFFLOAD:
4633                 adapter->tcp_ip_offload =
4634                     be64_to_cpu(crq->query_capability.number);
4635                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4636                            adapter->tcp_ip_offload);
4637                 break;
4638         case PROMISC_SUPPORTED:
4639                 adapter->promisc_supported =
4640                     be64_to_cpu(crq->query_capability.number);
4641                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4642                            adapter->promisc_supported);
4643                 break;
4644         case MIN_MTU:
4645                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4646                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4647                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4648                 break;
4649         case MAX_MTU:
4650                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4651                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4652                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4653                 break;
4654         case MAX_MULTICAST_FILTERS:
4655                 adapter->max_multicast_filters =
4656                     be64_to_cpu(crq->query_capability.number);
4657                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4658                            adapter->max_multicast_filters);
4659                 break;
4660         case VLAN_HEADER_INSERTION:
4661                 adapter->vlan_header_insertion =
4662                     be64_to_cpu(crq->query_capability.number);
4663                 if (adapter->vlan_header_insertion)
4664                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4665                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4666                            adapter->vlan_header_insertion);
4667                 break;
4668         case RX_VLAN_HEADER_INSERTION:
4669                 adapter->rx_vlan_header_insertion =
4670                     be64_to_cpu(crq->query_capability.number);
4671                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4672                            adapter->rx_vlan_header_insertion);
4673                 break;
4674         case MAX_TX_SG_ENTRIES:
4675                 adapter->max_tx_sg_entries =
4676                     be64_to_cpu(crq->query_capability.number);
4677                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4678                            adapter->max_tx_sg_entries);
4679                 break;
4680         case RX_SG_SUPPORTED:
4681                 adapter->rx_sg_supported =
4682                     be64_to_cpu(crq->query_capability.number);
4683                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4684                            adapter->rx_sg_supported);
4685                 break;
4686         case OPT_TX_COMP_SUB_QUEUES:
4687                 adapter->opt_tx_comp_sub_queues =
4688                     be64_to_cpu(crq->query_capability.number);
4689                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4690                            adapter->opt_tx_comp_sub_queues);
4691                 break;
4692         case OPT_RX_COMP_QUEUES:
4693                 adapter->opt_rx_comp_queues =
4694                     be64_to_cpu(crq->query_capability.number);
4695                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4696                            adapter->opt_rx_comp_queues);
4697                 break;
4698         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4699                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4700                     be64_to_cpu(crq->query_capability.number);
4701                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4702                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4703                 break;
4704         case OPT_TX_ENTRIES_PER_SUBCRQ:
4705                 adapter->opt_tx_entries_per_subcrq =
4706                     be64_to_cpu(crq->query_capability.number);
4707                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4708                            adapter->opt_tx_entries_per_subcrq);
4709                 break;
4710         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4711                 adapter->opt_rxba_entries_per_subcrq =
4712                     be64_to_cpu(crq->query_capability.number);
4713                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4714                            adapter->opt_rxba_entries_per_subcrq);
4715                 break;
4716         case TX_RX_DESC_REQ:
4717                 adapter->tx_rx_desc_req = crq->query_capability.number;
4718                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4719                            adapter->tx_rx_desc_req);
4720                 break;
4721
4722         default:
4723                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4724                            crq->query_capability.capability);
4725         }
4726
4727 out:
4728         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4729                 adapter->wait_capability = false;
4730                 send_request_cap(adapter, 0);
4731         }
4732 }
4733
4734 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4735 {
4736         union ibmvnic_crq crq;
4737         int rc;
4738
4739         memset(&crq, 0, sizeof(crq));
4740         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4741         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4742
4743         mutex_lock(&adapter->fw_lock);
4744         adapter->fw_done_rc = 0;
4745         reinit_completion(&adapter->fw_done);
4746
4747         rc = ibmvnic_send_crq(adapter, &crq);
4748         if (rc) {
4749                 mutex_unlock(&adapter->fw_lock);
4750                 return rc;
4751         }
4752
4753         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4754         if (rc) {
4755                 mutex_unlock(&adapter->fw_lock);
4756                 return rc;
4757         }
4758
4759         mutex_unlock(&adapter->fw_lock);
4760         return adapter->fw_done_rc ? -EIO : 0;
4761 }
4762
4763 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4764                                        struct ibmvnic_adapter *adapter)
4765 {
4766         struct net_device *netdev = adapter->netdev;
4767         int rc;
4768         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4769
4770         rc = crq->query_phys_parms_rsp.rc.code;
4771         if (rc) {
4772                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4773                 return rc;
4774         }
4775         switch (rspeed) {
4776         case IBMVNIC_10MBPS:
4777                 adapter->speed = SPEED_10;
4778                 break;
4779         case IBMVNIC_100MBPS:
4780                 adapter->speed = SPEED_100;
4781                 break;
4782         case IBMVNIC_1GBPS:
4783                 adapter->speed = SPEED_1000;
4784                 break;
4785         case IBMVNIC_10GBPS:
4786                 adapter->speed = SPEED_10000;
4787                 break;
4788         case IBMVNIC_25GBPS:
4789                 adapter->speed = SPEED_25000;
4790                 break;
4791         case IBMVNIC_40GBPS:
4792                 adapter->speed = SPEED_40000;
4793                 break;
4794         case IBMVNIC_50GBPS:
4795                 adapter->speed = SPEED_50000;
4796                 break;
4797         case IBMVNIC_100GBPS:
4798                 adapter->speed = SPEED_100000;
4799                 break;
4800         case IBMVNIC_200GBPS:
4801                 adapter->speed = SPEED_200000;
4802                 break;
4803         default:
4804                 if (netif_carrier_ok(netdev))
4805                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4806                 adapter->speed = SPEED_UNKNOWN;
4807         }
4808         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4809                 adapter->duplex = DUPLEX_FULL;
4810         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4811                 adapter->duplex = DUPLEX_HALF;
4812         else
4813                 adapter->duplex = DUPLEX_UNKNOWN;
4814
4815         return rc;
4816 }
4817
4818 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4819                                struct ibmvnic_adapter *adapter)
4820 {
4821         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4822         struct net_device *netdev = adapter->netdev;
4823         struct device *dev = &adapter->vdev->dev;
4824         u64 *u64_crq = (u64 *)crq;
4825         long rc;
4826
4827         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4828                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4829                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4830         switch (gen_crq->first) {
4831         case IBMVNIC_CRQ_INIT_RSP:
4832                 switch (gen_crq->cmd) {
4833                 case IBMVNIC_CRQ_INIT:
4834                         dev_info(dev, "Partner initialized\n");
4835                         adapter->from_passive_init = true;
4836                         /* Discard any stale login responses from prev reset.
4837                          * CHECK: should we clear even on INIT_COMPLETE?
4838                          */
4839                         adapter->login_pending = false;
4840
4841                         if (!completion_done(&adapter->init_done)) {
4842                                 complete(&adapter->init_done);
4843                                 adapter->init_done_rc = -EIO;
4844                         }
4845                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4846                         break;
4847                 case IBMVNIC_CRQ_INIT_COMPLETE:
4848                         dev_info(dev, "Partner initialization complete\n");
4849                         adapter->crq.active = true;
4850                         send_version_xchg(adapter);
4851                         break;
4852                 default:
4853                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4854                 }
4855                 return;
4856         case IBMVNIC_CRQ_XPORT_EVENT:
4857                 netif_carrier_off(netdev);
4858                 adapter->crq.active = false;
4859                 /* terminate any thread waiting for a response
4860                  * from the device
4861                  */
4862                 if (!completion_done(&adapter->fw_done)) {
4863                         adapter->fw_done_rc = -EIO;
4864                         complete(&adapter->fw_done);
4865                 }
4866                 if (!completion_done(&adapter->stats_done))
4867                         complete(&adapter->stats_done);
4868                 if (test_bit(0, &adapter->resetting))
4869                         adapter->force_reset_recovery = true;
4870                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4871                         dev_info(dev, "Migrated, re-enabling adapter\n");
4872                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4873                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4874                         dev_info(dev, "Backing device failover detected\n");
4875                         adapter->failover_pending = true;
4876                 } else {
4877                         /* The adapter lost the connection */
4878                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4879                                 gen_crq->cmd);
4880                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4881                 }
4882                 return;
4883         case IBMVNIC_CRQ_CMD_RSP:
4884                 break;
4885         default:
4886                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4887                         gen_crq->first);
4888                 return;
4889         }
4890
4891         switch (gen_crq->cmd) {
4892         case VERSION_EXCHANGE_RSP:
4893                 rc = crq->version_exchange_rsp.rc.code;
4894                 if (rc) {
4895                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4896                         break;
4897                 }
4898                 ibmvnic_version =
4899                             be16_to_cpu(crq->version_exchange_rsp.version);
4900                 dev_info(dev, "Partner protocol version is %d\n",
4901                          ibmvnic_version);
4902                 send_query_cap(adapter);
4903                 break;
4904         case QUERY_CAPABILITY_RSP:
4905                 handle_query_cap_rsp(crq, adapter);
4906                 break;
4907         case QUERY_MAP_RSP:
4908                 handle_query_map_rsp(crq, adapter);
4909                 break;
4910         case REQUEST_MAP_RSP:
4911                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4912                 complete(&adapter->fw_done);
4913                 break;
4914         case REQUEST_UNMAP_RSP:
4915                 handle_request_unmap_rsp(crq, adapter);
4916                 break;
4917         case REQUEST_CAPABILITY_RSP:
4918                 handle_request_cap_rsp(crq, adapter);
4919                 break;
4920         case LOGIN_RSP:
4921                 netdev_dbg(netdev, "Got Login Response\n");
4922                 handle_login_rsp(crq, adapter);
4923                 break;
4924         case LOGICAL_LINK_STATE_RSP:
4925                 netdev_dbg(netdev,
4926                            "Got Logical Link State Response, state: %d rc: %d\n",
4927                            crq->logical_link_state_rsp.link_state,
4928                            crq->logical_link_state_rsp.rc.code);
4929                 adapter->logical_link_state =
4930                     crq->logical_link_state_rsp.link_state;
4931                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4932                 complete(&adapter->init_done);
4933                 break;
4934         case LINK_STATE_INDICATION:
4935                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4936                 adapter->phys_link_state =
4937                     crq->link_state_indication.phys_link_state;
4938                 adapter->logical_link_state =
4939                     crq->link_state_indication.logical_link_state;
4940                 if (adapter->phys_link_state && adapter->logical_link_state)
4941                         netif_carrier_on(netdev);
4942                 else
4943                         netif_carrier_off(netdev);
4944                 break;
4945         case CHANGE_MAC_ADDR_RSP:
4946                 netdev_dbg(netdev, "Got MAC address change Response\n");
4947                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4948                 break;
4949         case ERROR_INDICATION:
4950                 netdev_dbg(netdev, "Got Error Indication\n");
4951                 handle_error_indication(crq, adapter);
4952                 break;
4953         case REQUEST_STATISTICS_RSP:
4954                 netdev_dbg(netdev, "Got Statistics Response\n");
4955                 complete(&adapter->stats_done);
4956                 break;
4957         case QUERY_IP_OFFLOAD_RSP:
4958                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4959                 handle_query_ip_offload_rsp(adapter);
4960                 break;
4961         case MULTICAST_CTRL_RSP:
4962                 netdev_dbg(netdev, "Got multicast control Response\n");
4963                 break;
4964         case CONTROL_IP_OFFLOAD_RSP:
4965                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4966                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4967                                  sizeof(adapter->ip_offload_ctrl),
4968                                  DMA_TO_DEVICE);
4969                 complete(&adapter->init_done);
4970                 break;
4971         case COLLECT_FW_TRACE_RSP:
4972                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4973                 complete(&adapter->fw_done);
4974                 break;
4975         case GET_VPD_SIZE_RSP:
4976                 handle_vpd_size_rsp(crq, adapter);
4977                 break;
4978         case GET_VPD_RSP:
4979                 handle_vpd_rsp(crq, adapter);
4980                 break;
4981         case QUERY_PHYS_PARMS_RSP:
4982                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
4983                 complete(&adapter->fw_done);
4984                 break;
4985         default:
4986                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4987                            gen_crq->cmd);
4988         }
4989 }
4990
4991 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
4992 {
4993         struct ibmvnic_adapter *adapter = instance;
4994
4995         tasklet_schedule(&adapter->tasklet);
4996         return IRQ_HANDLED;
4997 }
4998
4999 static void ibmvnic_tasklet(struct tasklet_struct *t)
5000 {
5001         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5002         struct ibmvnic_crq_queue *queue = &adapter->crq;
5003         union ibmvnic_crq *crq;
5004         unsigned long flags;
5005         bool done = false;
5006
5007         spin_lock_irqsave(&queue->lock, flags);
5008         while (!done) {
5009                 /* Pull all the valid messages off the CRQ */
5010                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5011                         /* This barrier makes sure ibmvnic_next_crq()'s
5012                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5013                          * before ibmvnic_handle_crq()'s
5014                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5015                          */
5016                         dma_rmb();
5017                         ibmvnic_handle_crq(crq, adapter);
5018                         crq->generic.first = 0;
5019                 }
5020
5021                 /* remain in tasklet until all
5022                  * capabilities responses are received
5023                  */
5024                 if (!adapter->wait_capability)
5025                         done = true;
5026         }
5027         /* if capabilities CRQ's were sent in this tasklet, the following
5028          * tasklet must wait until all responses are received
5029          */
5030         if (atomic_read(&adapter->running_cap_crqs) != 0)
5031                 adapter->wait_capability = true;
5032         spin_unlock_irqrestore(&queue->lock, flags);
5033 }
5034
5035 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5036 {
5037         struct vio_dev *vdev = adapter->vdev;
5038         int rc;
5039
5040         do {
5041                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5042         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5043
5044         if (rc)
5045                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5046
5047         return rc;
5048 }
5049
5050 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5051 {
5052         struct ibmvnic_crq_queue *crq = &adapter->crq;
5053         struct device *dev = &adapter->vdev->dev;
5054         struct vio_dev *vdev = adapter->vdev;
5055         int rc;
5056
5057         /* Close the CRQ */
5058         do {
5059                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5060         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5061
5062         /* Clean out the queue */
5063         if (!crq->msgs)
5064                 return -EINVAL;
5065
5066         memset(crq->msgs, 0, PAGE_SIZE);
5067         crq->cur = 0;
5068         crq->active = false;
5069
5070         /* And re-open it again */
5071         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5072                                 crq->msg_token, PAGE_SIZE);
5073
5074         if (rc == H_CLOSED)
5075                 /* Adapter is good, but other end is not ready */
5076                 dev_warn(dev, "Partner adapter not ready\n");
5077         else if (rc != 0)
5078                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5079
5080         return rc;
5081 }
5082
5083 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5084 {
5085         struct ibmvnic_crq_queue *crq = &adapter->crq;
5086         struct vio_dev *vdev = adapter->vdev;
5087         long rc;
5088
5089         if (!crq->msgs)
5090                 return;
5091
5092         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5093         free_irq(vdev->irq, adapter);
5094         tasklet_kill(&adapter->tasklet);
5095         do {
5096                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5097         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5098
5099         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5100                          DMA_BIDIRECTIONAL);
5101         free_page((unsigned long)crq->msgs);
5102         crq->msgs = NULL;
5103         crq->active = false;
5104 }
5105
5106 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5107 {
5108         struct ibmvnic_crq_queue *crq = &adapter->crq;
5109         struct device *dev = &adapter->vdev->dev;
5110         struct vio_dev *vdev = adapter->vdev;
5111         int rc, retrc = -ENOMEM;
5112
5113         if (crq->msgs)
5114                 return 0;
5115
5116         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5117         /* Should we allocate more than one page? */
5118
5119         if (!crq->msgs)
5120                 return -ENOMEM;
5121
5122         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5123         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5124                                         DMA_BIDIRECTIONAL);
5125         if (dma_mapping_error(dev, crq->msg_token))
5126                 goto map_failed;
5127
5128         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5129                                 crq->msg_token, PAGE_SIZE);
5130
5131         if (rc == H_RESOURCE)
5132                 /* maybe kexecing and resource is busy. try a reset */
5133                 rc = ibmvnic_reset_crq(adapter);
5134         retrc = rc;
5135
5136         if (rc == H_CLOSED) {
5137                 dev_warn(dev, "Partner adapter not ready\n");
5138         } else if (rc) {
5139                 dev_warn(dev, "Error %d opening adapter\n", rc);
5140                 goto reg_crq_failed;
5141         }
5142
5143         retrc = 0;
5144
5145         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5146
5147         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5148         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5149                  adapter->vdev->unit_address);
5150         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5151         if (rc) {
5152                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5153                         vdev->irq, rc);
5154                 goto req_irq_failed;
5155         }
5156
5157         rc = vio_enable_interrupts(vdev);
5158         if (rc) {
5159                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5160                 goto req_irq_failed;
5161         }
5162
5163         crq->cur = 0;
5164         spin_lock_init(&crq->lock);
5165
5166         return retrc;
5167
5168 req_irq_failed:
5169         tasklet_kill(&adapter->tasklet);
5170         do {
5171                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5172         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5173 reg_crq_failed:
5174         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5175 map_failed:
5176         free_page((unsigned long)crq->msgs);
5177         crq->msgs = NULL;
5178         return retrc;
5179 }
5180
5181 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5182 {
5183         struct device *dev = &adapter->vdev->dev;
5184         unsigned long timeout = msecs_to_jiffies(20000);
5185         u64 old_num_rx_queues, old_num_tx_queues;
5186         int rc;
5187
5188         adapter->from_passive_init = false;
5189
5190         if (reset) {
5191                 old_num_rx_queues = adapter->req_rx_queues;
5192                 old_num_tx_queues = adapter->req_tx_queues;
5193                 reinit_completion(&adapter->init_done);
5194         }
5195
5196         adapter->init_done_rc = 0;
5197         rc = ibmvnic_send_crq_init(adapter);
5198         if (rc) {
5199                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5200                 return rc;
5201         }
5202
5203         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5204                 dev_err(dev, "Initialization sequence timed out\n");
5205                 return -1;
5206         }
5207
5208         if (adapter->init_done_rc) {
5209                 release_crq_queue(adapter);
5210                 return adapter->init_done_rc;
5211         }
5212
5213         if (adapter->from_passive_init) {
5214                 adapter->state = VNIC_OPEN;
5215                 adapter->from_passive_init = false;
5216                 return -1;
5217         }
5218
5219         if (reset &&
5220             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5221             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5222                 if (adapter->req_rx_queues != old_num_rx_queues ||
5223                     adapter->req_tx_queues != old_num_tx_queues) {
5224                         release_sub_crqs(adapter, 0);
5225                         rc = init_sub_crqs(adapter);
5226                 } else {
5227                         rc = reset_sub_crq_queues(adapter);
5228                 }
5229         } else {
5230                 rc = init_sub_crqs(adapter);
5231         }
5232
5233         if (rc) {
5234                 dev_err(dev, "Initialization of sub crqs failed\n");
5235                 release_crq_queue(adapter);
5236                 return rc;
5237         }
5238
5239         rc = init_sub_crq_irqs(adapter);
5240         if (rc) {
5241                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5242                 release_crq_queue(adapter);
5243         }
5244
5245         return rc;
5246 }
5247
5248 static struct device_attribute dev_attr_failover;
5249
5250 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5251 {
5252         struct ibmvnic_adapter *adapter;
5253         struct net_device *netdev;
5254         unsigned char *mac_addr_p;
5255         int rc;
5256
5257         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5258                 dev->unit_address);
5259
5260         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5261                                                         VETH_MAC_ADDR, NULL);
5262         if (!mac_addr_p) {
5263                 dev_err(&dev->dev,
5264                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5265                         __FILE__, __LINE__);
5266                 return 0;
5267         }
5268
5269         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5270                                    IBMVNIC_MAX_QUEUES);
5271         if (!netdev)
5272                 return -ENOMEM;
5273
5274         adapter = netdev_priv(netdev);
5275         adapter->state = VNIC_PROBING;
5276         dev_set_drvdata(&dev->dev, netdev);
5277         adapter->vdev = dev;
5278         adapter->netdev = netdev;
5279         adapter->login_pending = false;
5280
5281         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5282         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5283         netdev->irq = dev->irq;
5284         netdev->netdev_ops = &ibmvnic_netdev_ops;
5285         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5286         SET_NETDEV_DEV(netdev, &dev->dev);
5287
5288         spin_lock_init(&adapter->stats_lock);
5289
5290         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5291         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5292                           __ibmvnic_delayed_reset);
5293         INIT_LIST_HEAD(&adapter->rwi_list);
5294         spin_lock_init(&adapter->rwi_lock);
5295         spin_lock_init(&adapter->state_lock);
5296         mutex_init(&adapter->fw_lock);
5297         init_completion(&adapter->init_done);
5298         init_completion(&adapter->fw_done);
5299         init_completion(&adapter->reset_done);
5300         init_completion(&adapter->stats_done);
5301         clear_bit(0, &adapter->resetting);
5302
5303         do {
5304                 rc = init_crq_queue(adapter);
5305                 if (rc) {
5306                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5307                                 rc);
5308                         goto ibmvnic_init_fail;
5309                 }
5310
5311                 rc = ibmvnic_reset_init(adapter, false);
5312                 if (rc && rc != EAGAIN)
5313                         goto ibmvnic_init_fail;
5314         } while (rc == EAGAIN);
5315
5316         rc = init_stats_buffers(adapter);
5317         if (rc)
5318                 goto ibmvnic_init_fail;
5319
5320         rc = init_stats_token(adapter);
5321         if (rc)
5322                 goto ibmvnic_stats_fail;
5323
5324         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5325         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5326         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5327
5328         rc = device_create_file(&dev->dev, &dev_attr_failover);
5329         if (rc)
5330                 goto ibmvnic_dev_file_err;
5331
5332         netif_carrier_off(netdev);
5333         rc = register_netdev(netdev);
5334         if (rc) {
5335                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5336                 goto ibmvnic_register_fail;
5337         }
5338         dev_info(&dev->dev, "ibmvnic registered\n");
5339
5340         adapter->state = VNIC_PROBED;
5341
5342         adapter->wait_for_reset = false;
5343         adapter->last_reset_time = jiffies;
5344         return 0;
5345
5346 ibmvnic_register_fail:
5347         device_remove_file(&dev->dev, &dev_attr_failover);
5348
5349 ibmvnic_dev_file_err:
5350         release_stats_token(adapter);
5351
5352 ibmvnic_stats_fail:
5353         release_stats_buffers(adapter);
5354
5355 ibmvnic_init_fail:
5356         release_sub_crqs(adapter, 1);
5357         release_crq_queue(adapter);
5358         mutex_destroy(&adapter->fw_lock);
5359         free_netdev(netdev);
5360
5361         return rc;
5362 }
5363
5364 static int ibmvnic_remove(struct vio_dev *dev)
5365 {
5366         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5367         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5368         unsigned long flags;
5369
5370         spin_lock_irqsave(&adapter->state_lock, flags);
5371         adapter->state = VNIC_REMOVING;
5372         spin_unlock_irqrestore(&adapter->state_lock, flags);
5373
5374         flush_work(&adapter->ibmvnic_reset);
5375         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5376
5377         rtnl_lock();
5378         unregister_netdevice(netdev);
5379
5380         release_resources(adapter);
5381         release_sub_crqs(adapter, 1);
5382         release_crq_queue(adapter);
5383
5384         release_stats_token(adapter);
5385         release_stats_buffers(adapter);
5386
5387         adapter->state = VNIC_REMOVED;
5388
5389         rtnl_unlock();
5390         mutex_destroy(&adapter->fw_lock);
5391         device_remove_file(&dev->dev, &dev_attr_failover);
5392         free_netdev(netdev);
5393         dev_set_drvdata(&dev->dev, NULL);
5394
5395         return 0;
5396 }
5397
5398 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5399                               const char *buf, size_t count)
5400 {
5401         struct net_device *netdev = dev_get_drvdata(dev);
5402         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5403         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5404         __be64 session_token;
5405         long rc;
5406
5407         if (!sysfs_streq(buf, "1"))
5408                 return -EINVAL;
5409
5410         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5411                          H_GET_SESSION_TOKEN, 0, 0, 0);
5412         if (rc) {
5413                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5414                            rc);
5415                 return -EINVAL;
5416         }
5417
5418         session_token = (__be64)retbuf[0];
5419         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5420                    be64_to_cpu(session_token));
5421         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5422                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5423         if (rc) {
5424                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5425                            rc);
5426                 return -EINVAL;
5427         }
5428
5429         return count;
5430 }
5431
5432 static DEVICE_ATTR_WO(failover);
5433
5434 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5435 {
5436         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5437         struct ibmvnic_adapter *adapter;
5438         struct iommu_table *tbl;
5439         unsigned long ret = 0;
5440         int i;
5441
5442         tbl = get_iommu_table_base(&vdev->dev);
5443
5444         /* netdev inits at probe time along with the structures we need below*/
5445         if (!netdev)
5446                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5447
5448         adapter = netdev_priv(netdev);
5449
5450         ret += PAGE_SIZE; /* the crq message queue */
5451         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5452
5453         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5454                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5455
5456         for (i = 0; i < adapter->num_active_rx_pools; i++)
5457                 ret += adapter->rx_pool[i].size *
5458                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5459
5460         return ret;
5461 }
5462
5463 static int ibmvnic_resume(struct device *dev)
5464 {
5465         struct net_device *netdev = dev_get_drvdata(dev);
5466         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5467
5468         if (adapter->state != VNIC_OPEN)
5469                 return 0;
5470
5471         tasklet_schedule(&adapter->tasklet);
5472
5473         return 0;
5474 }
5475
5476 static const struct vio_device_id ibmvnic_device_table[] = {
5477         {"network", "IBM,vnic"},
5478         {"", "" }
5479 };
5480 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5481
5482 static const struct dev_pm_ops ibmvnic_pm_ops = {
5483         .resume = ibmvnic_resume
5484 };
5485
5486 static struct vio_driver ibmvnic_driver = {
5487         .id_table       = ibmvnic_device_table,
5488         .probe          = ibmvnic_probe,
5489         .remove         = ibmvnic_remove,
5490         .get_desired_dma = ibmvnic_get_desired_dma,
5491         .name           = ibmvnic_driver_name,
5492         .pm             = &ibmvnic_pm_ops,
5493 };
5494
5495 /* module functions */
5496 static int __init ibmvnic_module_init(void)
5497 {
5498         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5499                 IBMVNIC_DRIVER_VERSION);
5500
5501         return vio_register_driver(&ibmvnic_driver);
5502 }
5503
5504 static void __exit ibmvnic_module_exit(void)
5505 {
5506         vio_unregister_driver(&ibmvnic_driver);
5507 }
5508
5509 module_init(ibmvnic_module_init);
5510 module_exit(ibmvnic_module_exit);