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 (santi_leon@yahoo.com)                                  */
   7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
   8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
   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                 /* The queue entry at the current index is peeked at above
2448                  * to determine that there is a valid descriptor awaiting
2449                  * processing. We want to be sure that the current slot
2450                  * holds a valid descriptor before reading its contents.
2451                  */
2452                 dma_rmb();
2453                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2454                 rx_buff =
2455                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2456                                                           rx_comp.correlator);
2457                 /* do error checking */
2458                 if (next->rx_comp.rc) {
2459                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2460                                    be16_to_cpu(next->rx_comp.rc));
2461                         /* free the entry */
2462                         next->rx_comp.first = 0;
2463                         dev_kfree_skb_any(rx_buff->skb);
2464                         remove_buff_from_pool(adapter, rx_buff);
2465                         continue;
2466                 } else if (!rx_buff->skb) {
2467                         /* free the entry */
2468                         next->rx_comp.first = 0;
2469                         remove_buff_from_pool(adapter, rx_buff);
2470                         continue;
2471                 }
2472
2473                 length = be32_to_cpu(next->rx_comp.len);
2474                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2475                 flags = next->rx_comp.flags;
2476                 skb = rx_buff->skb;
2477                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2478                                         length);
2479
2480                 /* VLAN Header has been stripped by the system firmware and
2481                  * needs to be inserted by the driver
2482                  */
2483                 if (adapter->rx_vlan_header_insertion &&
2484                     (flags & IBMVNIC_VLAN_STRIPPED))
2485                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2486                                                ntohs(next->rx_comp.vlan_tci));
2487
2488                 /* free the entry */
2489                 next->rx_comp.first = 0;
2490                 remove_buff_from_pool(adapter, rx_buff);
2491
2492                 skb_put(skb, length);
2493                 skb->protocol = eth_type_trans(skb, netdev);
2494                 skb_record_rx_queue(skb, scrq_num);
2495
2496                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2497                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2498                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2499                 }
2500
2501                 length = skb->len;
2502                 napi_gro_receive(napi, skb); /* send it up */
2503                 netdev->stats.rx_packets++;
2504                 netdev->stats.rx_bytes += length;
2505                 adapter->rx_stats_buffers[scrq_num].packets++;
2506                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2507                 frames_processed++;
2508         }
2509
2510         if (adapter->state != VNIC_CLOSING &&
2511             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2512               adapter->req_rx_add_entries_per_subcrq / 2) ||
2513               frames_processed < budget))
2514                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2515         if (frames_processed < budget) {
2516                 if (napi_complete_done(napi, frames_processed)) {
2517                         enable_scrq_irq(adapter, rx_scrq);
2518                         if (pending_scrq(adapter, rx_scrq)) {
2519                                 rmb();
2520                                 if (napi_reschedule(napi)) {
2521                                         disable_scrq_irq(adapter, rx_scrq);
2522                                         goto restart_poll;
2523                                 }
2524                         }
2525                 }
2526         }
2527         return frames_processed;
2528 }
2529
2530 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2531 {
2532         int rc, ret;
2533
2534         adapter->fallback.mtu = adapter->req_mtu;
2535         adapter->fallback.rx_queues = adapter->req_rx_queues;
2536         adapter->fallback.tx_queues = adapter->req_tx_queues;
2537         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2538         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2539
2540         reinit_completion(&adapter->reset_done);
2541         adapter->wait_for_reset = true;
2542         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2543
2544         if (rc) {
2545                 ret = rc;
2546                 goto out;
2547         }
2548         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2549         if (rc) {
2550                 ret = -ENODEV;
2551                 goto out;
2552         }
2553
2554         ret = 0;
2555         if (adapter->reset_done_rc) {
2556                 ret = -EIO;
2557                 adapter->desired.mtu = adapter->fallback.mtu;
2558                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2559                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2560                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2561                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2562
2563                 reinit_completion(&adapter->reset_done);
2564                 adapter->wait_for_reset = true;
2565                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2566                 if (rc) {
2567                         ret = rc;
2568                         goto out;
2569                 }
2570                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2571                                                  60000);
2572                 if (rc) {
2573                         ret = -ENODEV;
2574                         goto out;
2575                 }
2576         }
2577 out:
2578         adapter->wait_for_reset = false;
2579
2580         return ret;
2581 }
2582
2583 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2584 {
2585         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2586
2587         adapter->desired.mtu = new_mtu + ETH_HLEN;
2588
2589         return wait_for_reset(adapter);
2590 }
2591
2592 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2593                                                 struct net_device *dev,
2594                                                 netdev_features_t features)
2595 {
2596         /* Some backing hardware adapters can not
2597          * handle packets with a MSS less than 224
2598          * or with only one segment.
2599          */
2600         if (skb_is_gso(skb)) {
2601                 if (skb_shinfo(skb)->gso_size < 224 ||
2602                     skb_shinfo(skb)->gso_segs == 1)
2603                         features &= ~NETIF_F_GSO_MASK;
2604         }
2605
2606         return features;
2607 }
2608
2609 static const struct net_device_ops ibmvnic_netdev_ops = {
2610         .ndo_open               = ibmvnic_open,
2611         .ndo_stop               = ibmvnic_close,
2612         .ndo_start_xmit         = ibmvnic_xmit,
2613         .ndo_set_rx_mode        = ibmvnic_set_multi,
2614         .ndo_set_mac_address    = ibmvnic_set_mac,
2615         .ndo_validate_addr      = eth_validate_addr,
2616         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2617         .ndo_change_mtu         = ibmvnic_change_mtu,
2618         .ndo_features_check     = ibmvnic_features_check,
2619 };
2620
2621 /* ethtool functions */
2622
2623 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2624                                       struct ethtool_link_ksettings *cmd)
2625 {
2626         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2627         int rc;
2628
2629         rc = send_query_phys_parms(adapter);
2630         if (rc) {
2631                 adapter->speed = SPEED_UNKNOWN;
2632                 adapter->duplex = DUPLEX_UNKNOWN;
2633         }
2634         cmd->base.speed = adapter->speed;
2635         cmd->base.duplex = adapter->duplex;
2636         cmd->base.port = PORT_FIBRE;
2637         cmd->base.phy_address = 0;
2638         cmd->base.autoneg = AUTONEG_ENABLE;
2639
2640         return 0;
2641 }
2642
2643 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2644                                 struct ethtool_drvinfo *info)
2645 {
2646         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2647
2648         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2649         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2650         strlcpy(info->fw_version, adapter->fw_version,
2651                 sizeof(info->fw_version));
2652 }
2653
2654 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2655 {
2656         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2657
2658         return adapter->msg_enable;
2659 }
2660
2661 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2662 {
2663         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2664
2665         adapter->msg_enable = data;
2666 }
2667
2668 static u32 ibmvnic_get_link(struct net_device *netdev)
2669 {
2670         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2671
2672         /* Don't need to send a query because we request a logical link up at
2673          * init and then we wait for link state indications
2674          */
2675         return adapter->logical_link_state;
2676 }
2677
2678 static void ibmvnic_get_ringparam(struct net_device *netdev,
2679                                   struct ethtool_ringparam *ring)
2680 {
2681         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2682
2683         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2684                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2685                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2686         } else {
2687                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2688                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2689         }
2690         ring->rx_mini_max_pending = 0;
2691         ring->rx_jumbo_max_pending = 0;
2692         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2693         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2694         ring->rx_mini_pending = 0;
2695         ring->rx_jumbo_pending = 0;
2696 }
2697
2698 static int ibmvnic_set_ringparam(struct net_device *netdev,
2699                                  struct ethtool_ringparam *ring)
2700 {
2701         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2702         int ret;
2703
2704         ret = 0;
2705         adapter->desired.rx_entries = ring->rx_pending;
2706         adapter->desired.tx_entries = ring->tx_pending;
2707
2708         ret = wait_for_reset(adapter);
2709
2710         if (!ret &&
2711             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2712              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2713                 netdev_info(netdev,
2714                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2715                             ring->rx_pending, ring->tx_pending,
2716                             adapter->req_rx_add_entries_per_subcrq,
2717                             adapter->req_tx_entries_per_subcrq);
2718         return ret;
2719 }
2720
2721 static void ibmvnic_get_channels(struct net_device *netdev,
2722                                  struct ethtool_channels *channels)
2723 {
2724         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2725
2726         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2727                 channels->max_rx = adapter->max_rx_queues;
2728                 channels->max_tx = adapter->max_tx_queues;
2729         } else {
2730                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2731                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2732         }
2733
2734         channels->max_other = 0;
2735         channels->max_combined = 0;
2736         channels->rx_count = adapter->req_rx_queues;
2737         channels->tx_count = adapter->req_tx_queues;
2738         channels->other_count = 0;
2739         channels->combined_count = 0;
2740 }
2741
2742 static int ibmvnic_set_channels(struct net_device *netdev,
2743                                 struct ethtool_channels *channels)
2744 {
2745         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2746         int ret;
2747
2748         ret = 0;
2749         adapter->desired.rx_queues = channels->rx_count;
2750         adapter->desired.tx_queues = channels->tx_count;
2751
2752         ret = wait_for_reset(adapter);
2753
2754         if (!ret &&
2755             (adapter->req_rx_queues != channels->rx_count ||
2756              adapter->req_tx_queues != channels->tx_count))
2757                 netdev_info(netdev,
2758                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2759                             channels->rx_count, channels->tx_count,
2760                             adapter->req_rx_queues, adapter->req_tx_queues);
2761         return ret;
2762
2763 }
2764
2765 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2766 {
2767         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2768         int i;
2769
2770         switch (stringset) {
2771         case ETH_SS_STATS:
2772                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2773                                 i++, data += ETH_GSTRING_LEN)
2774                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2775
2776                 for (i = 0; i < adapter->req_tx_queues; i++) {
2777                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2778                         data += ETH_GSTRING_LEN;
2779
2780                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2781                         data += ETH_GSTRING_LEN;
2782
2783                         snprintf(data, ETH_GSTRING_LEN,
2784                                  "tx%d_dropped_packets", i);
2785                         data += ETH_GSTRING_LEN;
2786                 }
2787
2788                 for (i = 0; i < adapter->req_rx_queues; i++) {
2789                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2790                         data += ETH_GSTRING_LEN;
2791
2792                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2793                         data += ETH_GSTRING_LEN;
2794
2795                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2796                         data += ETH_GSTRING_LEN;
2797                 }
2798                 break;
2799
2800         case ETH_SS_PRIV_FLAGS:
2801                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2802                         strcpy(data + i * ETH_GSTRING_LEN,
2803                                ibmvnic_priv_flags[i]);
2804                 break;
2805         default:
2806                 return;
2807         }
2808 }
2809
2810 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2811 {
2812         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2813
2814         switch (sset) {
2815         case ETH_SS_STATS:
2816                 return ARRAY_SIZE(ibmvnic_stats) +
2817                        adapter->req_tx_queues * NUM_TX_STATS +
2818                        adapter->req_rx_queues * NUM_RX_STATS;
2819         case ETH_SS_PRIV_FLAGS:
2820                 return ARRAY_SIZE(ibmvnic_priv_flags);
2821         default:
2822                 return -EOPNOTSUPP;
2823         }
2824 }
2825
2826 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2827                                       struct ethtool_stats *stats, u64 *data)
2828 {
2829         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2830         union ibmvnic_crq crq;
2831         int i, j;
2832         int rc;
2833
2834         memset(&crq, 0, sizeof(crq));
2835         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2836         crq.request_statistics.cmd = REQUEST_STATISTICS;
2837         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2838         crq.request_statistics.len =
2839             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2840
2841         /* Wait for data to be written */
2842         reinit_completion(&adapter->stats_done);
2843         rc = ibmvnic_send_crq(adapter, &crq);
2844         if (rc)
2845                 return;
2846         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2847         if (rc)
2848                 return;
2849
2850         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2851                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2852                                                 ibmvnic_stats[i].offset));
2853
2854         for (j = 0; j < adapter->req_tx_queues; j++) {
2855                 data[i] = adapter->tx_stats_buffers[j].packets;
2856                 i++;
2857                 data[i] = adapter->tx_stats_buffers[j].bytes;
2858                 i++;
2859                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2860                 i++;
2861         }
2862
2863         for (j = 0; j < adapter->req_rx_queues; j++) {
2864                 data[i] = adapter->rx_stats_buffers[j].packets;
2865                 i++;
2866                 data[i] = adapter->rx_stats_buffers[j].bytes;
2867                 i++;
2868                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2869                 i++;
2870         }
2871 }
2872
2873 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2874 {
2875         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2876
2877         return adapter->priv_flags;
2878 }
2879
2880 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2881 {
2882         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2883         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2884
2885         if (which_maxes)
2886                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2887         else
2888                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2889
2890         return 0;
2891 }
2892 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2893         .get_drvinfo            = ibmvnic_get_drvinfo,
2894         .get_msglevel           = ibmvnic_get_msglevel,
2895         .set_msglevel           = ibmvnic_set_msglevel,
2896         .get_link               = ibmvnic_get_link,
2897         .get_ringparam          = ibmvnic_get_ringparam,
2898         .set_ringparam          = ibmvnic_set_ringparam,
2899         .get_channels           = ibmvnic_get_channels,
2900         .set_channels           = ibmvnic_set_channels,
2901         .get_strings            = ibmvnic_get_strings,
2902         .get_sset_count         = ibmvnic_get_sset_count,
2903         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2904         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2905         .get_priv_flags         = ibmvnic_get_priv_flags,
2906         .set_priv_flags         = ibmvnic_set_priv_flags,
2907 };
2908
2909 /* Routines for managing CRQs/sCRQs  */
2910
2911 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2912                                    struct ibmvnic_sub_crq_queue *scrq)
2913 {
2914         int rc;
2915
2916         if (!scrq) {
2917                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2918                 return -EINVAL;
2919         }
2920
2921         if (scrq->irq) {
2922                 free_irq(scrq->irq, scrq);
2923                 irq_dispose_mapping(scrq->irq);
2924                 scrq->irq = 0;
2925         }
2926
2927         if (scrq->msgs) {
2928                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2929                 atomic_set(&scrq->used, 0);
2930                 scrq->cur = 0;
2931                 scrq->ind_buf.index = 0;
2932         } else {
2933                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2934                 return -EINVAL;
2935         }
2936
2937         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2938                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2939         return rc;
2940 }
2941
2942 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2943 {
2944         int i, rc;
2945
2946         if (!adapter->tx_scrq || !adapter->rx_scrq)
2947                 return -EINVAL;
2948
2949         for (i = 0; i < adapter->req_tx_queues; i++) {
2950                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2951                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2952                 if (rc)
2953                         return rc;
2954         }
2955
2956         for (i = 0; i < adapter->req_rx_queues; i++) {
2957                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2958                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2959                 if (rc)
2960                         return rc;
2961         }
2962
2963         return rc;
2964 }
2965
2966 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2967                                   struct ibmvnic_sub_crq_queue *scrq,
2968                                   bool do_h_free)
2969 {
2970         struct device *dev = &adapter->vdev->dev;
2971         long rc;
2972
2973         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2974
2975         if (do_h_free) {
2976                 /* Close the sub-crqs */
2977                 do {
2978                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2979                                                 adapter->vdev->unit_address,
2980                                                 scrq->crq_num);
2981                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2982
2983                 if (rc) {
2984                         netdev_err(adapter->netdev,
2985                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2986                                    scrq->crq_num, rc);
2987                 }
2988         }
2989
2990         dma_free_coherent(dev,
2991                           IBMVNIC_IND_ARR_SZ,
2992                           scrq->ind_buf.indir_arr,
2993                           scrq->ind_buf.indir_dma);
2994
2995         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2996                          DMA_BIDIRECTIONAL);
2997         free_pages((unsigned long)scrq->msgs, 2);
2998         kfree(scrq);
2999 }
3000
3001 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3002                                                         *adapter)
3003 {
3004         struct device *dev = &adapter->vdev->dev;
3005         struct ibmvnic_sub_crq_queue *scrq;
3006         int rc;
3007
3008         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3009         if (!scrq)
3010                 return NULL;
3011
3012         scrq->msgs =
3013                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3014         if (!scrq->msgs) {
3015                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3016                 goto zero_page_failed;
3017         }
3018
3019         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3020                                          DMA_BIDIRECTIONAL);
3021         if (dma_mapping_error(dev, scrq->msg_token)) {
3022                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3023                 goto map_failed;
3024         }
3025
3026         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3027                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3028
3029         if (rc == H_RESOURCE)
3030                 rc = ibmvnic_reset_crq(adapter);
3031
3032         if (rc == H_CLOSED) {
3033                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3034         } else if (rc) {
3035                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3036                 goto reg_failed;
3037         }
3038
3039         scrq->adapter = adapter;
3040         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3041         scrq->ind_buf.index = 0;
3042
3043         scrq->ind_buf.indir_arr =
3044                 dma_alloc_coherent(dev,
3045                                    IBMVNIC_IND_ARR_SZ,
3046                                    &scrq->ind_buf.indir_dma,
3047                                    GFP_KERNEL);
3048
3049         if (!scrq->ind_buf.indir_arr)
3050                 goto indir_failed;
3051
3052         spin_lock_init(&scrq->lock);
3053
3054         netdev_dbg(adapter->netdev,
3055                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3056                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3057
3058         return scrq;
3059
3060 indir_failed:
3061         do {
3062                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3063                                         adapter->vdev->unit_address,
3064                                         scrq->crq_num);
3065         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3066 reg_failed:
3067         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3068                          DMA_BIDIRECTIONAL);
3069 map_failed:
3070         free_pages((unsigned long)scrq->msgs, 2);
3071 zero_page_failed:
3072         kfree(scrq);
3073
3074         return NULL;
3075 }
3076
3077 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3078 {
3079         int i;
3080
3081         if (adapter->tx_scrq) {
3082                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3083                         if (!adapter->tx_scrq[i])
3084                                 continue;
3085
3086                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3087                                    i);
3088                         if (adapter->tx_scrq[i]->irq) {
3089                                 free_irq(adapter->tx_scrq[i]->irq,
3090                                          adapter->tx_scrq[i]);
3091                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3092                                 adapter->tx_scrq[i]->irq = 0;
3093                         }
3094
3095                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3096                                               do_h_free);
3097                 }
3098
3099                 kfree(adapter->tx_scrq);
3100                 adapter->tx_scrq = NULL;
3101                 adapter->num_active_tx_scrqs = 0;
3102         }
3103
3104         if (adapter->rx_scrq) {
3105                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3106                         if (!adapter->rx_scrq[i])
3107                                 continue;
3108
3109                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3110                                    i);
3111                         if (adapter->rx_scrq[i]->irq) {
3112                                 free_irq(adapter->rx_scrq[i]->irq,
3113                                          adapter->rx_scrq[i]);
3114                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3115                                 adapter->rx_scrq[i]->irq = 0;
3116                         }
3117
3118                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3119                                               do_h_free);
3120                 }
3121
3122                 kfree(adapter->rx_scrq);
3123                 adapter->rx_scrq = NULL;
3124                 adapter->num_active_rx_scrqs = 0;
3125         }
3126 }
3127
3128 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3129                             struct ibmvnic_sub_crq_queue *scrq)
3130 {
3131         struct device *dev = &adapter->vdev->dev;
3132         unsigned long rc;
3133
3134         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3135                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3136         if (rc)
3137                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3138                         scrq->hw_irq, rc);
3139         return rc;
3140 }
3141
3142 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3143                            struct ibmvnic_sub_crq_queue *scrq)
3144 {
3145         struct device *dev = &adapter->vdev->dev;
3146         unsigned long rc;
3147
3148         if (scrq->hw_irq > 0x100000000ULL) {
3149                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3150                 return 1;
3151         }
3152
3153         if (test_bit(0, &adapter->resetting) &&
3154             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3155                 u64 val = (0xff000000) | scrq->hw_irq;
3156
3157                 rc = plpar_hcall_norets(H_EOI, val);
3158                 /* H_EOI would fail with rc = H_FUNCTION when running
3159                  * in XIVE mode which is expected, but not an error.
3160                  */
3161                 if (rc && (rc != H_FUNCTION))
3162                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3163                                 val, rc);
3164         }
3165
3166         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3167                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3168         if (rc)
3169                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3170                         scrq->hw_irq, rc);
3171         return rc;
3172 }
3173
3174 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3175                                struct ibmvnic_sub_crq_queue *scrq)
3176 {
3177         struct device *dev = &adapter->vdev->dev;
3178         struct ibmvnic_tx_pool *tx_pool;
3179         struct ibmvnic_tx_buff *txbuff;
3180         struct netdev_queue *txq;
3181         union sub_crq *next;
3182         int index;
3183         int i;
3184
3185 restart_loop:
3186         while (pending_scrq(adapter, scrq)) {
3187                 unsigned int pool = scrq->pool_index;
3188                 int num_entries = 0;
3189                 int total_bytes = 0;
3190                 int num_packets = 0;
3191
3192                 /* The queue entry at the current index is peeked at above
3193                  * to determine that there is a valid descriptor awaiting
3194                  * processing. We want to be sure that the current slot
3195                  * holds a valid descriptor before reading its contents.
3196                  */
3197                 dma_rmb();
3198
3199                 next = ibmvnic_next_scrq(adapter, scrq);
3200                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3201                         if (next->tx_comp.rcs[i])
3202                                 dev_err(dev, "tx error %x\n",
3203                                         next->tx_comp.rcs[i]);
3204                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3205                         if (index & IBMVNIC_TSO_POOL_MASK) {
3206                                 tx_pool = &adapter->tso_pool[pool];
3207                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3208                         } else {
3209                                 tx_pool = &adapter->tx_pool[pool];
3210                         }
3211
3212                         txbuff = &tx_pool->tx_buff[index];
3213                         num_packets++;
3214                         num_entries += txbuff->num_entries;
3215                         if (txbuff->skb) {
3216                                 total_bytes += txbuff->skb->len;
3217                                 dev_consume_skb_irq(txbuff->skb);
3218                                 txbuff->skb = NULL;
3219                         } else {
3220                                 netdev_warn(adapter->netdev,
3221                                             "TX completion received with NULL socket buffer\n");
3222                         }
3223                         tx_pool->free_map[tx_pool->producer_index] = index;
3224                         tx_pool->producer_index =
3225                                 (tx_pool->producer_index + 1) %
3226                                         tx_pool->num_buffers;
3227                 }
3228                 /* remove tx_comp scrq*/
3229                 next->tx_comp.first = 0;
3230
3231                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3232                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3233
3234                 if (atomic_sub_return(num_entries, &scrq->used) <=
3235                     (adapter->req_tx_entries_per_subcrq / 2) &&
3236                     __netif_subqueue_stopped(adapter->netdev,
3237                                              scrq->pool_index)) {
3238                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3239                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3240                                    scrq->pool_index);
3241                 }
3242         }
3243
3244         enable_scrq_irq(adapter, scrq);
3245
3246         if (pending_scrq(adapter, scrq)) {
3247                 disable_scrq_irq(adapter, scrq);
3248                 goto restart_loop;
3249         }
3250
3251         return 0;
3252 }
3253
3254 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3255 {
3256         struct ibmvnic_sub_crq_queue *scrq = instance;
3257         struct ibmvnic_adapter *adapter = scrq->adapter;
3258
3259         disable_scrq_irq(adapter, scrq);
3260         ibmvnic_complete_tx(adapter, scrq);
3261
3262         return IRQ_HANDLED;
3263 }
3264
3265 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3266 {
3267         struct ibmvnic_sub_crq_queue *scrq = instance;
3268         struct ibmvnic_adapter *adapter = scrq->adapter;
3269
3270         /* When booting a kdump kernel we can hit pending interrupts
3271          * prior to completing driver initialization.
3272          */
3273         if (unlikely(adapter->state != VNIC_OPEN))
3274                 return IRQ_NONE;
3275
3276         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3277
3278         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3279                 disable_scrq_irq(adapter, scrq);
3280                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3281         }
3282
3283         return IRQ_HANDLED;
3284 }
3285
3286 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3287 {
3288         struct device *dev = &adapter->vdev->dev;
3289         struct ibmvnic_sub_crq_queue *scrq;
3290         int i = 0, j = 0;
3291         int rc = 0;
3292
3293         for (i = 0; i < adapter->req_tx_queues; i++) {
3294                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3295                            i);
3296                 scrq = adapter->tx_scrq[i];
3297                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3298
3299                 if (!scrq->irq) {
3300                         rc = -EINVAL;
3301                         dev_err(dev, "Error mapping irq\n");
3302                         goto req_tx_irq_failed;
3303                 }
3304
3305                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3306                          adapter->vdev->unit_address, i);
3307                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3308                                  0, scrq->name, scrq);
3309
3310                 if (rc) {
3311                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3312                                 scrq->irq, rc);
3313                         irq_dispose_mapping(scrq->irq);
3314                         goto req_tx_irq_failed;
3315                 }
3316         }
3317
3318         for (i = 0; i < adapter->req_rx_queues; i++) {
3319                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3320                            i);
3321                 scrq = adapter->rx_scrq[i];
3322                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3323                 if (!scrq->irq) {
3324                         rc = -EINVAL;
3325                         dev_err(dev, "Error mapping irq\n");
3326                         goto req_rx_irq_failed;
3327                 }
3328                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3329                          adapter->vdev->unit_address, i);
3330                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3331                                  0, scrq->name, scrq);
3332                 if (rc) {
3333                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3334                                 scrq->irq, rc);
3335                         irq_dispose_mapping(scrq->irq);
3336                         goto req_rx_irq_failed;
3337                 }
3338         }
3339         return rc;
3340
3341 req_rx_irq_failed:
3342         for (j = 0; j < i; j++) {
3343                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3344                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3345         }
3346         i = adapter->req_tx_queues;
3347 req_tx_irq_failed:
3348         for (j = 0; j < i; j++) {
3349                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3350                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3351         }
3352         release_sub_crqs(adapter, 1);
3353         return rc;
3354 }
3355
3356 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3357 {
3358         struct device *dev = &adapter->vdev->dev;
3359         struct ibmvnic_sub_crq_queue **allqueues;
3360         int registered_queues = 0;
3361         int total_queues;
3362         int more = 0;
3363         int i;
3364
3365         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3366
3367         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3368         if (!allqueues)
3369                 return -1;
3370
3371         for (i = 0; i < total_queues; i++) {
3372                 allqueues[i] = init_sub_crq_queue(adapter);
3373                 if (!allqueues[i]) {
3374                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3375                         break;
3376                 }
3377                 registered_queues++;
3378         }
3379
3380         /* Make sure we were able to register the minimum number of queues */
3381         if (registered_queues <
3382             adapter->min_tx_queues + adapter->min_rx_queues) {
3383                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3384                 goto tx_failed;
3385         }
3386
3387         /* Distribute the failed allocated queues*/
3388         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3389                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3390                 switch (i % 3) {
3391                 case 0:
3392                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3393                                 adapter->req_rx_queues--;
3394                         else
3395                                 more++;
3396                         break;
3397                 case 1:
3398                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3399                                 adapter->req_tx_queues--;
3400                         else
3401                                 more++;
3402                         break;
3403                 }
3404         }
3405
3406         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3407                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3408         if (!adapter->tx_scrq)
3409                 goto tx_failed;
3410
3411         for (i = 0; i < adapter->req_tx_queues; i++) {
3412                 adapter->tx_scrq[i] = allqueues[i];
3413                 adapter->tx_scrq[i]->pool_index = i;
3414                 adapter->num_active_tx_scrqs++;
3415         }
3416
3417         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3418                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3419         if (!adapter->rx_scrq)
3420                 goto rx_failed;
3421
3422         for (i = 0; i < adapter->req_rx_queues; i++) {
3423                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3424                 adapter->rx_scrq[i]->scrq_num = i;
3425                 adapter->num_active_rx_scrqs++;
3426         }
3427
3428         kfree(allqueues);
3429         return 0;
3430
3431 rx_failed:
3432         kfree(adapter->tx_scrq);
3433         adapter->tx_scrq = NULL;
3434 tx_failed:
3435         for (i = 0; i < registered_queues; i++)
3436                 release_sub_crq_queue(adapter, allqueues[i], 1);
3437         kfree(allqueues);
3438         return -1;
3439 }
3440
3441 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3442 {
3443         struct device *dev = &adapter->vdev->dev;
3444         union ibmvnic_crq crq;
3445         int max_entries;
3446
3447         if (!retry) {
3448                 /* Sub-CRQ entries are 32 byte long */
3449                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3450
3451                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3452                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3453                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3454                         return;
3455                 }
3456
3457                 if (adapter->desired.mtu)
3458                         adapter->req_mtu = adapter->desired.mtu;
3459                 else
3460                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3461
3462                 if (!adapter->desired.tx_entries)
3463                         adapter->desired.tx_entries =
3464                                         adapter->max_tx_entries_per_subcrq;
3465                 if (!adapter->desired.rx_entries)
3466                         adapter->desired.rx_entries =
3467                                         adapter->max_rx_add_entries_per_subcrq;
3468
3469                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3470                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3471
3472                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3473                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3474                         adapter->desired.tx_entries = max_entries;
3475                 }
3476
3477                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3478                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3479                         adapter->desired.rx_entries = max_entries;
3480                 }
3481
3482                 if (adapter->desired.tx_entries)
3483                         adapter->req_tx_entries_per_subcrq =
3484                                         adapter->desired.tx_entries;
3485                 else
3486                         adapter->req_tx_entries_per_subcrq =
3487                                         adapter->max_tx_entries_per_subcrq;
3488
3489                 if (adapter->desired.rx_entries)
3490                         adapter->req_rx_add_entries_per_subcrq =
3491                                         adapter->desired.rx_entries;
3492                 else
3493                         adapter->req_rx_add_entries_per_subcrq =
3494                                         adapter->max_rx_add_entries_per_subcrq;
3495
3496                 if (adapter->desired.tx_queues)
3497                         adapter->req_tx_queues =
3498                                         adapter->desired.tx_queues;
3499                 else
3500                         adapter->req_tx_queues =
3501                                         adapter->opt_tx_comp_sub_queues;
3502
3503                 if (adapter->desired.rx_queues)
3504                         adapter->req_rx_queues =
3505                                         adapter->desired.rx_queues;
3506                 else
3507                         adapter->req_rx_queues =
3508                                         adapter->opt_rx_comp_queues;
3509
3510                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3511         }
3512
3513         memset(&crq, 0, sizeof(crq));
3514         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3515         crq.request_capability.cmd = REQUEST_CAPABILITY;
3516
3517         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3518         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3519         atomic_inc(&adapter->running_cap_crqs);
3520         ibmvnic_send_crq(adapter, &crq);
3521
3522         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3523         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3524         atomic_inc(&adapter->running_cap_crqs);
3525         ibmvnic_send_crq(adapter, &crq);
3526
3527         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3528         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3529         atomic_inc(&adapter->running_cap_crqs);
3530         ibmvnic_send_crq(adapter, &crq);
3531
3532         crq.request_capability.capability =
3533             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3534         crq.request_capability.number =
3535             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3536         atomic_inc(&adapter->running_cap_crqs);
3537         ibmvnic_send_crq(adapter, &crq);
3538
3539         crq.request_capability.capability =
3540             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3541         crq.request_capability.number =
3542             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3543         atomic_inc(&adapter->running_cap_crqs);
3544         ibmvnic_send_crq(adapter, &crq);
3545
3546         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3547         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3548         atomic_inc(&adapter->running_cap_crqs);
3549         ibmvnic_send_crq(adapter, &crq);
3550
3551         if (adapter->netdev->flags & IFF_PROMISC) {
3552                 if (adapter->promisc_supported) {
3553                         crq.request_capability.capability =
3554                             cpu_to_be16(PROMISC_REQUESTED);
3555                         crq.request_capability.number = cpu_to_be64(1);
3556                         atomic_inc(&adapter->running_cap_crqs);
3557                         ibmvnic_send_crq(adapter, &crq);
3558                 }
3559         } else {
3560                 crq.request_capability.capability =
3561                     cpu_to_be16(PROMISC_REQUESTED);
3562                 crq.request_capability.number = cpu_to_be64(0);
3563                 atomic_inc(&adapter->running_cap_crqs);
3564                 ibmvnic_send_crq(adapter, &crq);
3565         }
3566 }
3567
3568 static int pending_scrq(struct ibmvnic_adapter *adapter,
3569                         struct ibmvnic_sub_crq_queue *scrq)
3570 {
3571         union sub_crq *entry = &scrq->msgs[scrq->cur];
3572
3573         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3574                 return 1;
3575         else
3576                 return 0;
3577 }
3578
3579 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3580                                         struct ibmvnic_sub_crq_queue *scrq)
3581 {
3582         union sub_crq *entry;
3583         unsigned long flags;
3584
3585         spin_lock_irqsave(&scrq->lock, flags);
3586         entry = &scrq->msgs[scrq->cur];
3587         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3588                 if (++scrq->cur == scrq->size)
3589                         scrq->cur = 0;
3590         } else {
3591                 entry = NULL;
3592         }
3593         spin_unlock_irqrestore(&scrq->lock, flags);
3594
3595         /* Ensure that the entire buffer descriptor has been
3596          * loaded before reading its contents
3597          */
3598         dma_rmb();
3599
3600         return entry;
3601 }
3602
3603 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3604 {
3605         struct ibmvnic_crq_queue *queue = &adapter->crq;
3606         union ibmvnic_crq *crq;
3607
3608         crq = &queue->msgs[queue->cur];
3609         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3610                 if (++queue->cur == queue->size)
3611                         queue->cur = 0;
3612         } else {
3613                 crq = NULL;
3614         }
3615
3616         return crq;
3617 }
3618
3619 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3620 {
3621         switch (rc) {
3622         case H_PARAMETER:
3623                 dev_warn_ratelimited(dev,
3624                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3625                                      func, rc);
3626                 break;
3627         case H_CLOSED:
3628                 dev_warn_ratelimited(dev,
3629                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3630                                      func, rc);
3631                 break;
3632         default:
3633                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3634                 break;
3635         }
3636 }
3637
3638 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3639                                 u64 remote_handle, u64 ioba, u64 num_entries)
3640 {
3641         unsigned int ua = adapter->vdev->unit_address;
3642         struct device *dev = &adapter->vdev->dev;
3643         int rc;
3644
3645         /* Make sure the hypervisor sees the complete request */
3646         mb();
3647         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3648                                 cpu_to_be64(remote_handle),
3649                                 ioba, num_entries);
3650
3651         if (rc)
3652                 print_subcrq_error(dev, rc, __func__);
3653
3654         return rc;
3655 }
3656
3657 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3658                             union ibmvnic_crq *crq)
3659 {
3660         unsigned int ua = adapter->vdev->unit_address;
3661         struct device *dev = &adapter->vdev->dev;
3662         u64 *u64_crq = (u64 *)crq;
3663         int rc;
3664
3665         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3666                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3667                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3668
3669         if (!adapter->crq.active &&
3670             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3671                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3672                 return -EINVAL;
3673         }
3674
3675         /* Make sure the hypervisor sees the complete request */
3676         mb();
3677
3678         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3679                                 cpu_to_be64(u64_crq[0]),
3680                                 cpu_to_be64(u64_crq[1]));
3681
3682         if (rc) {
3683                 if (rc == H_CLOSED) {
3684                         dev_warn(dev, "CRQ Queue closed\n");
3685                         /* do not reset, report the fail, wait for passive init from server */
3686                 }
3687
3688                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3689         }
3690
3691         return rc;
3692 }
3693
3694 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3695 {
3696         struct device *dev = &adapter->vdev->dev;
3697         union ibmvnic_crq crq;
3698         int retries = 100;
3699         int rc;
3700
3701         memset(&crq, 0, sizeof(crq));
3702         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3703         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3704         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3705
3706         do {
3707                 rc = ibmvnic_send_crq(adapter, &crq);
3708                 if (rc != H_CLOSED)
3709                         break;
3710                 retries--;
3711                 msleep(50);
3712
3713         } while (retries > 0);
3714
3715         if (rc) {
3716                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3717                 return rc;
3718         }
3719
3720         return 0;
3721 }
3722
3723 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3724 {
3725         union ibmvnic_crq crq;
3726
3727         memset(&crq, 0, sizeof(crq));
3728         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3729         crq.version_exchange.cmd = VERSION_EXCHANGE;
3730         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3731
3732         return ibmvnic_send_crq(adapter, &crq);
3733 }
3734
3735 struct vnic_login_client_data {
3736         u8      type;
3737         __be16  len;
3738         char    name[];
3739 } __packed;
3740
3741 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3742 {
3743         int len;
3744
3745         /* Calculate the amount of buffer space needed for the
3746          * vnic client data in the login buffer. There are four entries,
3747          * OS name, LPAR name, device name, and a null last entry.
3748          */
3749         len = 4 * sizeof(struct vnic_login_client_data);
3750         len += 6; /* "Linux" plus NULL */
3751         len += strlen(utsname()->nodename) + 1;
3752         len += strlen(adapter->netdev->name) + 1;
3753
3754         return len;
3755 }
3756
3757 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3758                                  struct vnic_login_client_data *vlcd)
3759 {
3760         const char *os_name = "Linux";
3761         int len;
3762
3763         /* Type 1 - LPAR OS */
3764         vlcd->type = 1;
3765         len = strlen(os_name) + 1;
3766         vlcd->len = cpu_to_be16(len);
3767         strncpy(vlcd->name, os_name, len);
3768         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3769
3770         /* Type 2 - LPAR name */
3771         vlcd->type = 2;
3772         len = strlen(utsname()->nodename) + 1;
3773         vlcd->len = cpu_to_be16(len);
3774         strncpy(vlcd->name, utsname()->nodename, len);
3775         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3776
3777         /* Type 3 - device name */
3778         vlcd->type = 3;
3779         len = strlen(adapter->netdev->name) + 1;
3780         vlcd->len = cpu_to_be16(len);
3781         strncpy(vlcd->name, adapter->netdev->name, len);
3782 }
3783
3784 static int send_login(struct ibmvnic_adapter *adapter)
3785 {
3786         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3787         struct ibmvnic_login_buffer *login_buffer;
3788         struct device *dev = &adapter->vdev->dev;
3789         struct vnic_login_client_data *vlcd;
3790         dma_addr_t rsp_buffer_token;
3791         dma_addr_t buffer_token;
3792         size_t rsp_buffer_size;
3793         union ibmvnic_crq crq;
3794         int client_data_len;
3795         size_t buffer_size;
3796         __be64 *tx_list_p;
3797         __be64 *rx_list_p;
3798         int rc;
3799         int i;
3800
3801         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3802                 netdev_err(adapter->netdev,
3803                            "RX or TX queues are not allocated, device login failed\n");
3804                 return -1;
3805         }
3806
3807         release_login_buffer(adapter);
3808         release_login_rsp_buffer(adapter);
3809
3810         client_data_len = vnic_client_data_len(adapter);
3811
3812         buffer_size =
3813             sizeof(struct ibmvnic_login_buffer) +
3814             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3815             client_data_len;
3816
3817         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3818         if (!login_buffer)
3819                 goto buf_alloc_failed;
3820
3821         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3822                                       DMA_TO_DEVICE);
3823         if (dma_mapping_error(dev, buffer_token)) {
3824                 dev_err(dev, "Couldn't map login buffer\n");
3825                 goto buf_map_failed;
3826         }
3827
3828         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3829                           sizeof(u64) * adapter->req_tx_queues +
3830                           sizeof(u64) * adapter->req_rx_queues +
3831                           sizeof(u64) * adapter->req_rx_queues +
3832                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3833
3834         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3835         if (!login_rsp_buffer)
3836                 goto buf_rsp_alloc_failed;
3837
3838         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3839                                           rsp_buffer_size, DMA_FROM_DEVICE);
3840         if (dma_mapping_error(dev, rsp_buffer_token)) {
3841                 dev_err(dev, "Couldn't map login rsp buffer\n");
3842                 goto buf_rsp_map_failed;
3843         }
3844
3845         adapter->login_buf = login_buffer;
3846         adapter->login_buf_token = buffer_token;
3847         adapter->login_buf_sz = buffer_size;
3848         adapter->login_rsp_buf = login_rsp_buffer;
3849         adapter->login_rsp_buf_token = rsp_buffer_token;
3850         adapter->login_rsp_buf_sz = rsp_buffer_size;
3851
3852         login_buffer->len = cpu_to_be32(buffer_size);
3853         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3854         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3855         login_buffer->off_txcomp_subcrqs =
3856             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3857         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3858         login_buffer->off_rxcomp_subcrqs =
3859             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3860                         sizeof(u64) * adapter->req_tx_queues);
3861         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3862         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3863
3864         tx_list_p = (__be64 *)((char *)login_buffer +
3865                                       sizeof(struct ibmvnic_login_buffer));
3866         rx_list_p = (__be64 *)((char *)login_buffer +
3867                                       sizeof(struct ibmvnic_login_buffer) +
3868                                       sizeof(u64) * adapter->req_tx_queues);
3869
3870         for (i = 0; i < adapter->req_tx_queues; i++) {
3871                 if (adapter->tx_scrq[i]) {
3872                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3873                                                    crq_num);
3874                 }
3875         }
3876
3877         for (i = 0; i < adapter->req_rx_queues; i++) {
3878                 if (adapter->rx_scrq[i]) {
3879                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3880                                                    crq_num);
3881                 }
3882         }
3883
3884         /* Insert vNIC login client data */
3885         vlcd = (struct vnic_login_client_data *)
3886                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3887         login_buffer->client_data_offset =
3888                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3889         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3890
3891         vnic_add_client_data(adapter, vlcd);
3892
3893         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3894         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3895                 netdev_dbg(adapter->netdev, "%016lx\n",
3896                            ((unsigned long int *)(adapter->login_buf))[i]);
3897         }
3898
3899         memset(&crq, 0, sizeof(crq));
3900         crq.login.first = IBMVNIC_CRQ_CMD;
3901         crq.login.cmd = LOGIN;
3902         crq.login.ioba = cpu_to_be32(buffer_token);
3903         crq.login.len = cpu_to_be32(buffer_size);
3904
3905         adapter->login_pending = true;
3906         rc = ibmvnic_send_crq(adapter, &crq);
3907         if (rc) {
3908                 adapter->login_pending = false;
3909                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3910                 goto buf_rsp_map_failed;
3911         }
3912
3913         return 0;
3914
3915 buf_rsp_map_failed:
3916         kfree(login_rsp_buffer);
3917         adapter->login_rsp_buf = NULL;
3918 buf_rsp_alloc_failed:
3919         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3920 buf_map_failed:
3921         kfree(login_buffer);
3922         adapter->login_buf = NULL;
3923 buf_alloc_failed:
3924         return -1;
3925 }
3926
3927 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3928                             u32 len, u8 map_id)
3929 {
3930         union ibmvnic_crq crq;
3931
3932         memset(&crq, 0, sizeof(crq));
3933         crq.request_map.first = IBMVNIC_CRQ_CMD;
3934         crq.request_map.cmd = REQUEST_MAP;
3935         crq.request_map.map_id = map_id;
3936         crq.request_map.ioba = cpu_to_be32(addr);
3937         crq.request_map.len = cpu_to_be32(len);
3938         return ibmvnic_send_crq(adapter, &crq);
3939 }
3940
3941 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3942 {
3943         union ibmvnic_crq crq;
3944
3945         memset(&crq, 0, sizeof(crq));
3946         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3947         crq.request_unmap.cmd = REQUEST_UNMAP;
3948         crq.request_unmap.map_id = map_id;
3949         return ibmvnic_send_crq(adapter, &crq);
3950 }
3951
3952 static void send_query_map(struct ibmvnic_adapter *adapter)
3953 {
3954         union ibmvnic_crq crq;
3955
3956         memset(&crq, 0, sizeof(crq));
3957         crq.query_map.first = IBMVNIC_CRQ_CMD;
3958         crq.query_map.cmd = QUERY_MAP;
3959         ibmvnic_send_crq(adapter, &crq);
3960 }
3961
3962 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3963 static void send_query_cap(struct ibmvnic_adapter *adapter)
3964 {
3965         union ibmvnic_crq crq;
3966
3967         atomic_set(&adapter->running_cap_crqs, 0);
3968         memset(&crq, 0, sizeof(crq));
3969         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3970         crq.query_capability.cmd = QUERY_CAPABILITY;
3971
3972         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3973         atomic_inc(&adapter->running_cap_crqs);
3974         ibmvnic_send_crq(adapter, &crq);
3975
3976         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3977         atomic_inc(&adapter->running_cap_crqs);
3978         ibmvnic_send_crq(adapter, &crq);
3979
3980         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3981         atomic_inc(&adapter->running_cap_crqs);
3982         ibmvnic_send_crq(adapter, &crq);
3983
3984         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3985         atomic_inc(&adapter->running_cap_crqs);
3986         ibmvnic_send_crq(adapter, &crq);
3987
3988         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3989         atomic_inc(&adapter->running_cap_crqs);
3990         ibmvnic_send_crq(adapter, &crq);
3991
3992         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3993         atomic_inc(&adapter->running_cap_crqs);
3994         ibmvnic_send_crq(adapter, &crq);
3995
3996         crq.query_capability.capability =
3997             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3998         atomic_inc(&adapter->running_cap_crqs);
3999         ibmvnic_send_crq(adapter, &crq);
4000
4001         crq.query_capability.capability =
4002             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4003         atomic_inc(&adapter->running_cap_crqs);
4004         ibmvnic_send_crq(adapter, &crq);
4005
4006         crq.query_capability.capability =
4007             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4008         atomic_inc(&adapter->running_cap_crqs);
4009         ibmvnic_send_crq(adapter, &crq);
4010
4011         crq.query_capability.capability =
4012             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4013         atomic_inc(&adapter->running_cap_crqs);
4014         ibmvnic_send_crq(adapter, &crq);
4015
4016         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4017         atomic_inc(&adapter->running_cap_crqs);
4018         ibmvnic_send_crq(adapter, &crq);
4019
4020         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4021         atomic_inc(&adapter->running_cap_crqs);
4022         ibmvnic_send_crq(adapter, &crq);
4023
4024         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4025         atomic_inc(&adapter->running_cap_crqs);
4026         ibmvnic_send_crq(adapter, &crq);
4027
4028         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4029         atomic_inc(&adapter->running_cap_crqs);
4030         ibmvnic_send_crq(adapter, &crq);
4031
4032         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4033         atomic_inc(&adapter->running_cap_crqs);
4034         ibmvnic_send_crq(adapter, &crq);
4035
4036         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4037         atomic_inc(&adapter->running_cap_crqs);
4038         ibmvnic_send_crq(adapter, &crq);
4039
4040         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4041         atomic_inc(&adapter->running_cap_crqs);
4042         ibmvnic_send_crq(adapter, &crq);
4043
4044         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4045         atomic_inc(&adapter->running_cap_crqs);
4046         ibmvnic_send_crq(adapter, &crq);
4047
4048         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4049         atomic_inc(&adapter->running_cap_crqs);
4050         ibmvnic_send_crq(adapter, &crq);
4051
4052         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4053         atomic_inc(&adapter->running_cap_crqs);
4054         ibmvnic_send_crq(adapter, &crq);
4055
4056         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4057         atomic_inc(&adapter->running_cap_crqs);
4058         ibmvnic_send_crq(adapter, &crq);
4059
4060         crq.query_capability.capability =
4061                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4062         atomic_inc(&adapter->running_cap_crqs);
4063         ibmvnic_send_crq(adapter, &crq);
4064
4065         crq.query_capability.capability =
4066                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4067         atomic_inc(&adapter->running_cap_crqs);
4068         ibmvnic_send_crq(adapter, &crq);
4069
4070         crq.query_capability.capability =
4071                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4072         atomic_inc(&adapter->running_cap_crqs);
4073         ibmvnic_send_crq(adapter, &crq);
4074
4075         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4076         atomic_inc(&adapter->running_cap_crqs);
4077         ibmvnic_send_crq(adapter, &crq);
4078 }
4079
4080 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4081 {
4082         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4083         struct device *dev = &adapter->vdev->dev;
4084         union ibmvnic_crq crq;
4085
4086         adapter->ip_offload_tok =
4087                 dma_map_single(dev,
4088                                &adapter->ip_offload_buf,
4089                                buf_sz,
4090                                DMA_FROM_DEVICE);
4091
4092         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4093                 if (!firmware_has_feature(FW_FEATURE_CMO))
4094                         dev_err(dev, "Couldn't map offload buffer\n");
4095                 return;
4096         }
4097
4098         memset(&crq, 0, sizeof(crq));
4099         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4100         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4101         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4102         crq.query_ip_offload.ioba =
4103             cpu_to_be32(adapter->ip_offload_tok);
4104
4105         ibmvnic_send_crq(adapter, &crq);
4106 }
4107
4108 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4109 {
4110         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4111         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4112         struct device *dev = &adapter->vdev->dev;
4113         netdev_features_t old_hw_features = 0;
4114         union ibmvnic_crq crq;
4115
4116         adapter->ip_offload_ctrl_tok =
4117                 dma_map_single(dev,
4118                                ctrl_buf,
4119                                sizeof(adapter->ip_offload_ctrl),
4120                                DMA_TO_DEVICE);
4121
4122         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4123                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4124                 return;
4125         }
4126
4127         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4128         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4129         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4130         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4131         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4132         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4133         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4134         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4135         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4136         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4137
4138         /* large_rx disabled for now, additional features needed */
4139         ctrl_buf->large_rx_ipv4 = 0;
4140         ctrl_buf->large_rx_ipv6 = 0;
4141
4142         if (adapter->state != VNIC_PROBING) {
4143                 old_hw_features = adapter->netdev->hw_features;
4144                 adapter->netdev->hw_features = 0;
4145         }
4146
4147         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4148
4149         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4150                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4151
4152         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4153                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4154
4155         if ((adapter->netdev->features &
4156             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4157                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4158
4159         if (buf->large_tx_ipv4)
4160                 adapter->netdev->hw_features |= NETIF_F_TSO;
4161         if (buf->large_tx_ipv6)
4162                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4163
4164         if (adapter->state == VNIC_PROBING) {
4165                 adapter->netdev->features |= adapter->netdev->hw_features;
4166         } else if (old_hw_features != adapter->netdev->hw_features) {
4167                 netdev_features_t tmp = 0;
4168
4169                 /* disable features no longer supported */
4170                 adapter->netdev->features &= adapter->netdev->hw_features;
4171                 /* turn on features now supported if previously enabled */
4172                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4173                         adapter->netdev->hw_features;
4174                 adapter->netdev->features |=
4175                                 tmp & adapter->netdev->wanted_features;
4176         }
4177
4178         memset(&crq, 0, sizeof(crq));
4179         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4180         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4181         crq.control_ip_offload.len =
4182             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4183         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4184         ibmvnic_send_crq(adapter, &crq);
4185 }
4186
4187 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4188                                 struct ibmvnic_adapter *adapter)
4189 {
4190         struct device *dev = &adapter->vdev->dev;
4191
4192         if (crq->get_vpd_size_rsp.rc.code) {
4193                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4194                         crq->get_vpd_size_rsp.rc.code);
4195                 complete(&adapter->fw_done);
4196                 return;
4197         }
4198
4199         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4200         complete(&adapter->fw_done);
4201 }
4202
4203 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4204                            struct ibmvnic_adapter *adapter)
4205 {
4206         struct device *dev = &adapter->vdev->dev;
4207         unsigned char *substr = NULL;
4208         u8 fw_level_len = 0;
4209
4210         memset(adapter->fw_version, 0, 32);
4211
4212         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4213                          DMA_FROM_DEVICE);
4214
4215         if (crq->get_vpd_rsp.rc.code) {
4216                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4217                         crq->get_vpd_rsp.rc.code);
4218                 goto complete;
4219         }
4220
4221         /* get the position of the firmware version info
4222          * located after the ASCII 'RM' substring in the buffer
4223          */
4224         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4225         if (!substr) {
4226                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4227                 goto complete;
4228         }
4229
4230         /* get length of firmware level ASCII substring */
4231         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4232                 fw_level_len = *(substr + 2);
4233         } else {
4234                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4235                 goto complete;
4236         }
4237
4238         /* copy firmware version string from vpd into adapter */
4239         if ((substr + 3 + fw_level_len) <
4240             (adapter->vpd->buff + adapter->vpd->len)) {
4241                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4242         } else {
4243                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4244         }
4245
4246 complete:
4247         if (adapter->fw_version[0] == '\0')
4248                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4249         complete(&adapter->fw_done);
4250 }
4251
4252 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4253 {
4254         struct device *dev = &adapter->vdev->dev;
4255         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4256         int i;
4257
4258         dma_unmap_single(dev, adapter->ip_offload_tok,
4259                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4260
4261         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4262         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4263                 netdev_dbg(adapter->netdev, "%016lx\n",
4264                            ((unsigned long int *)(buf))[i]);
4265
4266         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4267         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4268         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4269                    buf->tcp_ipv4_chksum);
4270         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4271                    buf->tcp_ipv6_chksum);
4272         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4273                    buf->udp_ipv4_chksum);
4274         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4275                    buf->udp_ipv6_chksum);
4276         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4277                    buf->large_tx_ipv4);
4278         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4279                    buf->large_tx_ipv6);
4280         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4281                    buf->large_rx_ipv4);
4282         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4283                    buf->large_rx_ipv6);
4284         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4285                    buf->max_ipv4_header_size);
4286         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4287                    buf->max_ipv6_header_size);
4288         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4289                    buf->max_tcp_header_size);
4290         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4291                    buf->max_udp_header_size);
4292         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4293                    buf->max_large_tx_size);
4294         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4295                    buf->max_large_rx_size);
4296         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4297                    buf->ipv6_extension_header);
4298         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4299                    buf->tcp_pseudosum_req);
4300         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4301                    buf->num_ipv6_ext_headers);
4302         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4303                    buf->off_ipv6_ext_headers);
4304
4305         send_control_ip_offload(adapter);
4306 }
4307
4308 static const char *ibmvnic_fw_err_cause(u16 cause)
4309 {
4310         switch (cause) {
4311         case ADAPTER_PROBLEM:
4312                 return "adapter problem";
4313         case BUS_PROBLEM:
4314                 return "bus problem";
4315         case FW_PROBLEM:
4316                 return "firmware problem";
4317         case DD_PROBLEM:
4318                 return "device driver problem";
4319         case EEH_RECOVERY:
4320                 return "EEH recovery";
4321         case FW_UPDATED:
4322                 return "firmware updated";
4323         case LOW_MEMORY:
4324                 return "low Memory";
4325         default:
4326                 return "unknown";
4327         }
4328 }
4329
4330 static void handle_error_indication(union ibmvnic_crq *crq,
4331                                     struct ibmvnic_adapter *adapter)
4332 {
4333         struct device *dev = &adapter->vdev->dev;
4334         u16 cause;
4335
4336         cause = be16_to_cpu(crq->error_indication.error_cause);
4337
4338         dev_warn_ratelimited(dev,
4339                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4340                              crq->error_indication.flags
4341                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4342                              ibmvnic_fw_err_cause(cause));
4343
4344         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4345                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4346         else
4347                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4348 }
4349
4350 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4351                                  struct ibmvnic_adapter *adapter)
4352 {
4353         struct net_device *netdev = adapter->netdev;
4354         struct device *dev = &adapter->vdev->dev;
4355         long rc;
4356
4357         rc = crq->change_mac_addr_rsp.rc.code;
4358         if (rc) {
4359                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4360                 goto out;
4361         }
4362         /* crq->change_mac_addr.mac_addr is the requested one
4363          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4364          */
4365         ether_addr_copy(netdev->dev_addr,
4366                         &crq->change_mac_addr_rsp.mac_addr[0]);
4367         ether_addr_copy(adapter->mac_addr,
4368                         &crq->change_mac_addr_rsp.mac_addr[0]);
4369 out:
4370         complete(&adapter->fw_done);
4371         return rc;
4372 }
4373
4374 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4375                                    struct ibmvnic_adapter *adapter)
4376 {
4377         struct device *dev = &adapter->vdev->dev;
4378         u64 *req_value;
4379         char *name;
4380
4381         atomic_dec(&adapter->running_cap_crqs);
4382         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4383         case REQ_TX_QUEUES:
4384                 req_value = &adapter->req_tx_queues;
4385                 name = "tx";
4386                 break;
4387         case REQ_RX_QUEUES:
4388                 req_value = &adapter->req_rx_queues;
4389                 name = "rx";
4390                 break;
4391         case REQ_RX_ADD_QUEUES:
4392                 req_value = &adapter->req_rx_add_queues;
4393                 name = "rx_add";
4394                 break;
4395         case REQ_TX_ENTRIES_PER_SUBCRQ:
4396                 req_value = &adapter->req_tx_entries_per_subcrq;
4397                 name = "tx_entries_per_subcrq";
4398                 break;
4399         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4400                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4401                 name = "rx_add_entries_per_subcrq";
4402                 break;
4403         case REQ_MTU:
4404                 req_value = &adapter->req_mtu;
4405                 name = "mtu";
4406                 break;
4407         case PROMISC_REQUESTED:
4408                 req_value = &adapter->promisc;
4409                 name = "promisc";
4410                 break;
4411         default:
4412                 dev_err(dev, "Got invalid cap request rsp %d\n",
4413                         crq->request_capability.capability);
4414                 return;
4415         }
4416
4417         switch (crq->request_capability_rsp.rc.code) {
4418         case SUCCESS:
4419                 break;
4420         case PARTIALSUCCESS:
4421                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4422                          *req_value,
4423                          (long int)be64_to_cpu(crq->request_capability_rsp.
4424                                                number), name);
4425
4426                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4427                     REQ_MTU) {
4428                         pr_err("mtu of %llu is not supported. Reverting.\n",
4429                                *req_value);
4430                         *req_value = adapter->fallback.mtu;
4431                 } else {
4432                         *req_value =
4433                                 be64_to_cpu(crq->request_capability_rsp.number);
4434                 }
4435
4436                 send_request_cap(adapter, 1);
4437                 return;
4438         default:
4439                 dev_err(dev, "Error %d in request cap rsp\n",
4440                         crq->request_capability_rsp.rc.code);
4441                 return;
4442         }
4443
4444         /* Done receiving requested capabilities, query IP offload support */
4445         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4446                 adapter->wait_capability = false;
4447                 send_query_ip_offload(adapter);
4448         }
4449 }
4450
4451 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4452                             struct ibmvnic_adapter *adapter)
4453 {
4454         struct device *dev = &adapter->vdev->dev;
4455         struct net_device *netdev = adapter->netdev;
4456         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4457         struct ibmvnic_login_buffer *login = adapter->login_buf;
4458         u64 *tx_handle_array;
4459         u64 *rx_handle_array;
4460         int num_tx_pools;
4461         int num_rx_pools;
4462         u64 *size_array;
4463         int i;
4464
4465         /* CHECK: Test/set of login_pending does not need to be atomic
4466          * because only ibmvnic_tasklet tests/clears this.
4467          */
4468         if (!adapter->login_pending) {
4469                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4470                 return 0;
4471         }
4472         adapter->login_pending = false;
4473
4474         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4475                          DMA_TO_DEVICE);
4476         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4477                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4478
4479         /* If the number of queues requested can't be allocated by the
4480          * server, the login response will return with code 1. We will need
4481          * to resend the login buffer with fewer queues requested.
4482          */
4483         if (login_rsp_crq->generic.rc.code) {
4484                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4485                 complete(&adapter->init_done);
4486                 return 0;
4487         }
4488
4489         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4490
4491         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4492         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4493                 netdev_dbg(adapter->netdev, "%016lx\n",
4494                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4495         }
4496
4497         /* Sanity checks */
4498         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4499             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4500              adapter->req_rx_add_queues !=
4501              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4502                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4503                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4504                 return -EIO;
4505         }
4506         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4507                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4508         /* variable buffer sizes are not supported, so just read the
4509          * first entry.
4510          */
4511         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4512
4513         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4514         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4515
4516         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4517                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4518         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4519                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4520
4521         for (i = 0; i < num_tx_pools; i++)
4522                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4523
4524         for (i = 0; i < num_rx_pools; i++)
4525                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4526
4527         adapter->num_active_tx_scrqs = num_tx_pools;
4528         adapter->num_active_rx_scrqs = num_rx_pools;
4529         release_login_rsp_buffer(adapter);
4530         release_login_buffer(adapter);
4531         complete(&adapter->init_done);
4532
4533         return 0;
4534 }
4535
4536 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4537                                      struct ibmvnic_adapter *adapter)
4538 {
4539         struct device *dev = &adapter->vdev->dev;
4540         long rc;
4541
4542         rc = crq->request_unmap_rsp.rc.code;
4543         if (rc)
4544                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4545 }
4546
4547 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4548                                  struct ibmvnic_adapter *adapter)
4549 {
4550         struct net_device *netdev = adapter->netdev;
4551         struct device *dev = &adapter->vdev->dev;
4552         long rc;
4553
4554         rc = crq->query_map_rsp.rc.code;
4555         if (rc) {
4556                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4557                 return;
4558         }
4559         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4560                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4561                    crq->query_map_rsp.free_pages);
4562 }
4563
4564 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4565                                  struct ibmvnic_adapter *adapter)
4566 {
4567         struct net_device *netdev = adapter->netdev;
4568         struct device *dev = &adapter->vdev->dev;
4569         long rc;
4570
4571         atomic_dec(&adapter->running_cap_crqs);
4572         netdev_dbg(netdev, "Outstanding queries: %d\n",
4573                    atomic_read(&adapter->running_cap_crqs));
4574         rc = crq->query_capability.rc.code;
4575         if (rc) {
4576                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4577                 goto out;
4578         }
4579
4580         switch (be16_to_cpu(crq->query_capability.capability)) {
4581         case MIN_TX_QUEUES:
4582                 adapter->min_tx_queues =
4583                     be64_to_cpu(crq->query_capability.number);
4584                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4585                            adapter->min_tx_queues);
4586                 break;
4587         case MIN_RX_QUEUES:
4588                 adapter->min_rx_queues =
4589                     be64_to_cpu(crq->query_capability.number);
4590                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4591                            adapter->min_rx_queues);
4592                 break;
4593         case MIN_RX_ADD_QUEUES:
4594                 adapter->min_rx_add_queues =
4595                     be64_to_cpu(crq->query_capability.number);
4596                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4597                            adapter->min_rx_add_queues);
4598                 break;
4599         case MAX_TX_QUEUES:
4600                 adapter->max_tx_queues =
4601                     be64_to_cpu(crq->query_capability.number);
4602                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4603                            adapter->max_tx_queues);
4604                 break;
4605         case MAX_RX_QUEUES:
4606                 adapter->max_rx_queues =
4607                     be64_to_cpu(crq->query_capability.number);
4608                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4609                            adapter->max_rx_queues);
4610                 break;
4611         case MAX_RX_ADD_QUEUES:
4612                 adapter->max_rx_add_queues =
4613                     be64_to_cpu(crq->query_capability.number);
4614                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4615                            adapter->max_rx_add_queues);
4616                 break;
4617         case MIN_TX_ENTRIES_PER_SUBCRQ:
4618                 adapter->min_tx_entries_per_subcrq =
4619                     be64_to_cpu(crq->query_capability.number);
4620                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4621                            adapter->min_tx_entries_per_subcrq);
4622                 break;
4623         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4624                 adapter->min_rx_add_entries_per_subcrq =
4625                     be64_to_cpu(crq->query_capability.number);
4626                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4627                            adapter->min_rx_add_entries_per_subcrq);
4628                 break;
4629         case MAX_TX_ENTRIES_PER_SUBCRQ:
4630                 adapter->max_tx_entries_per_subcrq =
4631                     be64_to_cpu(crq->query_capability.number);
4632                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4633                            adapter->max_tx_entries_per_subcrq);
4634                 break;
4635         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4636                 adapter->max_rx_add_entries_per_subcrq =
4637                     be64_to_cpu(crq->query_capability.number);
4638                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4639                            adapter->max_rx_add_entries_per_subcrq);
4640                 break;
4641         case TCP_IP_OFFLOAD:
4642                 adapter->tcp_ip_offload =
4643                     be64_to_cpu(crq->query_capability.number);
4644                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4645                            adapter->tcp_ip_offload);
4646                 break;
4647         case PROMISC_SUPPORTED:
4648                 adapter->promisc_supported =
4649                     be64_to_cpu(crq->query_capability.number);
4650                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4651                            adapter->promisc_supported);
4652                 break;
4653         case MIN_MTU:
4654                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4655                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4656                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4657                 break;
4658         case MAX_MTU:
4659                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4660                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4661                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4662                 break;
4663         case MAX_MULTICAST_FILTERS:
4664                 adapter->max_multicast_filters =
4665                     be64_to_cpu(crq->query_capability.number);
4666                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4667                            adapter->max_multicast_filters);
4668                 break;
4669         case VLAN_HEADER_INSERTION:
4670                 adapter->vlan_header_insertion =
4671                     be64_to_cpu(crq->query_capability.number);
4672                 if (adapter->vlan_header_insertion)
4673                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4674                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4675                            adapter->vlan_header_insertion);
4676                 break;
4677         case RX_VLAN_HEADER_INSERTION:
4678                 adapter->rx_vlan_header_insertion =
4679                     be64_to_cpu(crq->query_capability.number);
4680                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4681                            adapter->rx_vlan_header_insertion);
4682                 break;
4683         case MAX_TX_SG_ENTRIES:
4684                 adapter->max_tx_sg_entries =
4685                     be64_to_cpu(crq->query_capability.number);
4686                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4687                            adapter->max_tx_sg_entries);
4688                 break;
4689         case RX_SG_SUPPORTED:
4690                 adapter->rx_sg_supported =
4691                     be64_to_cpu(crq->query_capability.number);
4692                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4693                            adapter->rx_sg_supported);
4694                 break;
4695         case OPT_TX_COMP_SUB_QUEUES:
4696                 adapter->opt_tx_comp_sub_queues =
4697                     be64_to_cpu(crq->query_capability.number);
4698                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4699                            adapter->opt_tx_comp_sub_queues);
4700                 break;
4701         case OPT_RX_COMP_QUEUES:
4702                 adapter->opt_rx_comp_queues =
4703                     be64_to_cpu(crq->query_capability.number);
4704                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4705                            adapter->opt_rx_comp_queues);
4706                 break;
4707         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4708                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4709                     be64_to_cpu(crq->query_capability.number);
4710                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4711                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4712                 break;
4713         case OPT_TX_ENTRIES_PER_SUBCRQ:
4714                 adapter->opt_tx_entries_per_subcrq =
4715                     be64_to_cpu(crq->query_capability.number);
4716                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4717                            adapter->opt_tx_entries_per_subcrq);
4718                 break;
4719         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4720                 adapter->opt_rxba_entries_per_subcrq =
4721                     be64_to_cpu(crq->query_capability.number);
4722                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4723                            adapter->opt_rxba_entries_per_subcrq);
4724                 break;
4725         case TX_RX_DESC_REQ:
4726                 adapter->tx_rx_desc_req = crq->query_capability.number;
4727                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4728                            adapter->tx_rx_desc_req);
4729                 break;
4730
4731         default:
4732                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4733                            crq->query_capability.capability);
4734         }
4735
4736 out:
4737         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4738                 adapter->wait_capability = false;
4739                 send_request_cap(adapter, 0);
4740         }
4741 }
4742
4743 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4744 {
4745         union ibmvnic_crq crq;
4746         int rc;
4747
4748         memset(&crq, 0, sizeof(crq));
4749         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4750         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4751
4752         mutex_lock(&adapter->fw_lock);
4753         adapter->fw_done_rc = 0;
4754         reinit_completion(&adapter->fw_done);
4755
4756         rc = ibmvnic_send_crq(adapter, &crq);
4757         if (rc) {
4758                 mutex_unlock(&adapter->fw_lock);
4759                 return rc;
4760         }
4761
4762         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4763         if (rc) {
4764                 mutex_unlock(&adapter->fw_lock);
4765                 return rc;
4766         }
4767
4768         mutex_unlock(&adapter->fw_lock);
4769         return adapter->fw_done_rc ? -EIO : 0;
4770 }
4771
4772 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4773                                        struct ibmvnic_adapter *adapter)
4774 {
4775         struct net_device *netdev = adapter->netdev;
4776         int rc;
4777         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4778
4779         rc = crq->query_phys_parms_rsp.rc.code;
4780         if (rc) {
4781                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4782                 return rc;
4783         }
4784         switch (rspeed) {
4785         case IBMVNIC_10MBPS:
4786                 adapter->speed = SPEED_10;
4787                 break;
4788         case IBMVNIC_100MBPS:
4789                 adapter->speed = SPEED_100;
4790                 break;
4791         case IBMVNIC_1GBPS:
4792                 adapter->speed = SPEED_1000;
4793                 break;
4794         case IBMVNIC_10GBPS:
4795                 adapter->speed = SPEED_10000;
4796                 break;
4797         case IBMVNIC_25GBPS:
4798                 adapter->speed = SPEED_25000;
4799                 break;
4800         case IBMVNIC_40GBPS:
4801                 adapter->speed = SPEED_40000;
4802                 break;
4803         case IBMVNIC_50GBPS:
4804                 adapter->speed = SPEED_50000;
4805                 break;
4806         case IBMVNIC_100GBPS:
4807                 adapter->speed = SPEED_100000;
4808                 break;
4809         case IBMVNIC_200GBPS:
4810                 adapter->speed = SPEED_200000;
4811                 break;
4812         default:
4813                 if (netif_carrier_ok(netdev))
4814                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4815                 adapter->speed = SPEED_UNKNOWN;
4816         }
4817         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4818                 adapter->duplex = DUPLEX_FULL;
4819         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4820                 adapter->duplex = DUPLEX_HALF;
4821         else
4822                 adapter->duplex = DUPLEX_UNKNOWN;
4823
4824         return rc;
4825 }
4826
4827 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4828                                struct ibmvnic_adapter *adapter)
4829 {
4830         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4831         struct net_device *netdev = adapter->netdev;
4832         struct device *dev = &adapter->vdev->dev;
4833         u64 *u64_crq = (u64 *)crq;
4834         long rc;
4835
4836         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4837                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4838                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4839         switch (gen_crq->first) {
4840         case IBMVNIC_CRQ_INIT_RSP:
4841                 switch (gen_crq->cmd) {
4842                 case IBMVNIC_CRQ_INIT:
4843                         dev_info(dev, "Partner initialized\n");
4844                         adapter->from_passive_init = true;
4845                         /* Discard any stale login responses from prev reset.
4846                          * CHECK: should we clear even on INIT_COMPLETE?
4847                          */
4848                         adapter->login_pending = false;
4849
4850                         if (!completion_done(&adapter->init_done)) {
4851                                 complete(&adapter->init_done);
4852                                 adapter->init_done_rc = -EIO;
4853                         }
4854                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4855                         break;
4856                 case IBMVNIC_CRQ_INIT_COMPLETE:
4857                         dev_info(dev, "Partner initialization complete\n");
4858                         adapter->crq.active = true;
4859                         send_version_xchg(adapter);
4860                         break;
4861                 default:
4862                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4863                 }
4864                 return;
4865         case IBMVNIC_CRQ_XPORT_EVENT:
4866                 netif_carrier_off(netdev);
4867                 adapter->crq.active = false;
4868                 /* terminate any thread waiting for a response
4869                  * from the device
4870                  */
4871                 if (!completion_done(&adapter->fw_done)) {
4872                         adapter->fw_done_rc = -EIO;
4873                         complete(&adapter->fw_done);
4874                 }
4875                 if (!completion_done(&adapter->stats_done))
4876                         complete(&adapter->stats_done);
4877                 if (test_bit(0, &adapter->resetting))
4878                         adapter->force_reset_recovery = true;
4879                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4880                         dev_info(dev, "Migrated, re-enabling adapter\n");
4881                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4882                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4883                         dev_info(dev, "Backing device failover detected\n");
4884                         adapter->failover_pending = true;
4885                 } else {
4886                         /* The adapter lost the connection */
4887                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4888                                 gen_crq->cmd);
4889                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4890                 }
4891                 return;
4892         case IBMVNIC_CRQ_CMD_RSP:
4893                 break;
4894         default:
4895                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4896                         gen_crq->first);
4897                 return;
4898         }
4899
4900         switch (gen_crq->cmd) {
4901         case VERSION_EXCHANGE_RSP:
4902                 rc = crq->version_exchange_rsp.rc.code;
4903                 if (rc) {
4904                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4905                         break;
4906                 }
4907                 ibmvnic_version =
4908                             be16_to_cpu(crq->version_exchange_rsp.version);
4909                 dev_info(dev, "Partner protocol version is %d\n",
4910                          ibmvnic_version);
4911                 send_query_cap(adapter);
4912                 break;
4913         case QUERY_CAPABILITY_RSP:
4914                 handle_query_cap_rsp(crq, adapter);
4915                 break;
4916         case QUERY_MAP_RSP:
4917                 handle_query_map_rsp(crq, adapter);
4918                 break;
4919         case REQUEST_MAP_RSP:
4920                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4921                 complete(&adapter->fw_done);
4922                 break;
4923         case REQUEST_UNMAP_RSP:
4924                 handle_request_unmap_rsp(crq, adapter);
4925                 break;
4926         case REQUEST_CAPABILITY_RSP:
4927                 handle_request_cap_rsp(crq, adapter);
4928                 break;
4929         case LOGIN_RSP:
4930                 netdev_dbg(netdev, "Got Login Response\n");
4931                 handle_login_rsp(crq, adapter);
4932                 break;
4933         case LOGICAL_LINK_STATE_RSP:
4934                 netdev_dbg(netdev,
4935                            "Got Logical Link State Response, state: %d rc: %d\n",
4936                            crq->logical_link_state_rsp.link_state,
4937                            crq->logical_link_state_rsp.rc.code);
4938                 adapter->logical_link_state =
4939                     crq->logical_link_state_rsp.link_state;
4940                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4941                 complete(&adapter->init_done);
4942                 break;
4943         case LINK_STATE_INDICATION:
4944                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4945                 adapter->phys_link_state =
4946                     crq->link_state_indication.phys_link_state;
4947                 adapter->logical_link_state =
4948                     crq->link_state_indication.logical_link_state;
4949                 if (adapter->phys_link_state && adapter->logical_link_state)
4950                         netif_carrier_on(netdev);
4951                 else
4952                         netif_carrier_off(netdev);
4953                 break;
4954         case CHANGE_MAC_ADDR_RSP:
4955                 netdev_dbg(netdev, "Got MAC address change Response\n");
4956                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4957                 break;
4958         case ERROR_INDICATION:
4959                 netdev_dbg(netdev, "Got Error Indication\n");
4960                 handle_error_indication(crq, adapter);
4961                 break;
4962         case REQUEST_STATISTICS_RSP:
4963                 netdev_dbg(netdev, "Got Statistics Response\n");
4964                 complete(&adapter->stats_done);
4965                 break;
4966         case QUERY_IP_OFFLOAD_RSP:
4967                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4968                 handle_query_ip_offload_rsp(adapter);
4969                 break;
4970         case MULTICAST_CTRL_RSP:
4971                 netdev_dbg(netdev, "Got multicast control Response\n");
4972                 break;
4973         case CONTROL_IP_OFFLOAD_RSP:
4974                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4975                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4976                                  sizeof(adapter->ip_offload_ctrl),
4977                                  DMA_TO_DEVICE);
4978                 complete(&adapter->init_done);
4979                 break;
4980         case COLLECT_FW_TRACE_RSP:
4981                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4982                 complete(&adapter->fw_done);
4983                 break;
4984         case GET_VPD_SIZE_RSP:
4985                 handle_vpd_size_rsp(crq, adapter);
4986                 break;
4987         case GET_VPD_RSP:
4988                 handle_vpd_rsp(crq, adapter);
4989                 break;
4990         case QUERY_PHYS_PARMS_RSP:
4991                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
4992                 complete(&adapter->fw_done);
4993                 break;
4994         default:
4995                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4996                            gen_crq->cmd);
4997         }
4998 }
4999
5000 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5001 {
5002         struct ibmvnic_adapter *adapter = instance;
5003
5004         tasklet_schedule(&adapter->tasklet);
5005         return IRQ_HANDLED;
5006 }
5007
5008 static void ibmvnic_tasklet(struct tasklet_struct *t)
5009 {
5010         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5011         struct ibmvnic_crq_queue *queue = &adapter->crq;
5012         union ibmvnic_crq *crq;
5013         unsigned long flags;
5014         bool done = false;
5015
5016         spin_lock_irqsave(&queue->lock, flags);
5017         while (!done) {
5018                 /* Pull all the valid messages off the CRQ */
5019                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5020                         /* This barrier makes sure ibmvnic_next_crq()'s
5021                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5022                          * before ibmvnic_handle_crq()'s
5023                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5024                          */
5025                         dma_rmb();
5026                         ibmvnic_handle_crq(crq, adapter);
5027                         crq->generic.first = 0;
5028                 }
5029
5030                 /* remain in tasklet until all
5031                  * capabilities responses are received
5032                  */
5033                 if (!adapter->wait_capability)
5034                         done = true;
5035         }
5036         /* if capabilities CRQ's were sent in this tasklet, the following
5037          * tasklet must wait until all responses are received
5038          */
5039         if (atomic_read(&adapter->running_cap_crqs) != 0)
5040                 adapter->wait_capability = true;
5041         spin_unlock_irqrestore(&queue->lock, flags);
5042 }
5043
5044 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5045 {
5046         struct vio_dev *vdev = adapter->vdev;
5047         int rc;
5048
5049         do {
5050                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5051         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5052
5053         if (rc)
5054                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5055
5056         return rc;
5057 }
5058
5059 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5060 {
5061         struct ibmvnic_crq_queue *crq = &adapter->crq;
5062         struct device *dev = &adapter->vdev->dev;
5063         struct vio_dev *vdev = adapter->vdev;
5064         int rc;
5065
5066         /* Close the CRQ */
5067         do {
5068                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5069         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5070
5071         /* Clean out the queue */
5072         if (!crq->msgs)
5073                 return -EINVAL;
5074
5075         memset(crq->msgs, 0, PAGE_SIZE);
5076         crq->cur = 0;
5077         crq->active = false;
5078
5079         /* And re-open it again */
5080         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5081                                 crq->msg_token, PAGE_SIZE);
5082
5083         if (rc == H_CLOSED)
5084                 /* Adapter is good, but other end is not ready */
5085                 dev_warn(dev, "Partner adapter not ready\n");
5086         else if (rc != 0)
5087                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5088
5089         return rc;
5090 }
5091
5092 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5093 {
5094         struct ibmvnic_crq_queue *crq = &adapter->crq;
5095         struct vio_dev *vdev = adapter->vdev;
5096         long rc;
5097
5098         if (!crq->msgs)
5099                 return;
5100
5101         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5102         free_irq(vdev->irq, adapter);
5103         tasklet_kill(&adapter->tasklet);
5104         do {
5105                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5106         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5107
5108         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5109                          DMA_BIDIRECTIONAL);
5110         free_page((unsigned long)crq->msgs);
5111         crq->msgs = NULL;
5112         crq->active = false;
5113 }
5114
5115 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5116 {
5117         struct ibmvnic_crq_queue *crq = &adapter->crq;
5118         struct device *dev = &adapter->vdev->dev;
5119         struct vio_dev *vdev = adapter->vdev;
5120         int rc, retrc = -ENOMEM;
5121
5122         if (crq->msgs)
5123                 return 0;
5124
5125         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5126         /* Should we allocate more than one page? */
5127
5128         if (!crq->msgs)
5129                 return -ENOMEM;
5130
5131         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5132         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5133                                         DMA_BIDIRECTIONAL);
5134         if (dma_mapping_error(dev, crq->msg_token))
5135                 goto map_failed;
5136
5137         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5138                                 crq->msg_token, PAGE_SIZE);
5139
5140         if (rc == H_RESOURCE)
5141                 /* maybe kexecing and resource is busy. try a reset */
5142                 rc = ibmvnic_reset_crq(adapter);
5143         retrc = rc;
5144
5145         if (rc == H_CLOSED) {
5146                 dev_warn(dev, "Partner adapter not ready\n");
5147         } else if (rc) {
5148                 dev_warn(dev, "Error %d opening adapter\n", rc);
5149                 goto reg_crq_failed;
5150         }
5151
5152         retrc = 0;
5153
5154         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5155
5156         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5157         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5158                  adapter->vdev->unit_address);
5159         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5160         if (rc) {
5161                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5162                         vdev->irq, rc);
5163                 goto req_irq_failed;
5164         }
5165
5166         rc = vio_enable_interrupts(vdev);
5167         if (rc) {
5168                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5169                 goto req_irq_failed;
5170         }
5171
5172         crq->cur = 0;
5173         spin_lock_init(&crq->lock);
5174
5175         return retrc;
5176
5177 req_irq_failed:
5178         tasklet_kill(&adapter->tasklet);
5179         do {
5180                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5181         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5182 reg_crq_failed:
5183         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5184 map_failed:
5185         free_page((unsigned long)crq->msgs);
5186         crq->msgs = NULL;
5187         return retrc;
5188 }
5189
5190 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5191 {
5192         struct device *dev = &adapter->vdev->dev;
5193         unsigned long timeout = msecs_to_jiffies(20000);
5194         u64 old_num_rx_queues, old_num_tx_queues;
5195         int rc;
5196
5197         adapter->from_passive_init = false;
5198
5199         if (reset) {
5200                 old_num_rx_queues = adapter->req_rx_queues;
5201                 old_num_tx_queues = adapter->req_tx_queues;
5202                 reinit_completion(&adapter->init_done);
5203         }
5204
5205         adapter->init_done_rc = 0;
5206         rc = ibmvnic_send_crq_init(adapter);
5207         if (rc) {
5208                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5209                 return rc;
5210         }
5211
5212         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5213                 dev_err(dev, "Initialization sequence timed out\n");
5214                 return -1;
5215         }
5216
5217         if (adapter->init_done_rc) {
5218                 release_crq_queue(adapter);
5219                 return adapter->init_done_rc;
5220         }
5221
5222         if (adapter->from_passive_init) {
5223                 adapter->state = VNIC_OPEN;
5224                 adapter->from_passive_init = false;
5225                 return -1;
5226         }
5227
5228         if (reset &&
5229             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5230             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5231                 if (adapter->req_rx_queues != old_num_rx_queues ||
5232                     adapter->req_tx_queues != old_num_tx_queues) {
5233                         release_sub_crqs(adapter, 0);
5234                         rc = init_sub_crqs(adapter);
5235                 } else {
5236                         rc = reset_sub_crq_queues(adapter);
5237                 }
5238         } else {
5239                 rc = init_sub_crqs(adapter);
5240         }
5241
5242         if (rc) {
5243                 dev_err(dev, "Initialization of sub crqs failed\n");
5244                 release_crq_queue(adapter);
5245                 return rc;
5246         }
5247
5248         rc = init_sub_crq_irqs(adapter);
5249         if (rc) {
5250                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5251                 release_crq_queue(adapter);
5252         }
5253
5254         return rc;
5255 }
5256
5257 static struct device_attribute dev_attr_failover;
5258
5259 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5260 {
5261         struct ibmvnic_adapter *adapter;
5262         struct net_device *netdev;
5263         unsigned char *mac_addr_p;
5264         int rc;
5265
5266         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5267                 dev->unit_address);
5268
5269         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5270                                                         VETH_MAC_ADDR, NULL);
5271         if (!mac_addr_p) {
5272                 dev_err(&dev->dev,
5273                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5274                         __FILE__, __LINE__);
5275                 return 0;
5276         }
5277
5278         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5279                                    IBMVNIC_MAX_QUEUES);
5280         if (!netdev)
5281                 return -ENOMEM;
5282
5283         adapter = netdev_priv(netdev);
5284         adapter->state = VNIC_PROBING;
5285         dev_set_drvdata(&dev->dev, netdev);
5286         adapter->vdev = dev;
5287         adapter->netdev = netdev;
5288         adapter->login_pending = false;
5289
5290         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5291         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5292         netdev->irq = dev->irq;
5293         netdev->netdev_ops = &ibmvnic_netdev_ops;
5294         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5295         SET_NETDEV_DEV(netdev, &dev->dev);
5296
5297         spin_lock_init(&adapter->stats_lock);
5298
5299         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5300         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5301                           __ibmvnic_delayed_reset);
5302         INIT_LIST_HEAD(&adapter->rwi_list);
5303         spin_lock_init(&adapter->rwi_lock);
5304         spin_lock_init(&adapter->state_lock);
5305         mutex_init(&adapter->fw_lock);
5306         init_completion(&adapter->init_done);
5307         init_completion(&adapter->fw_done);
5308         init_completion(&adapter->reset_done);
5309         init_completion(&adapter->stats_done);
5310         clear_bit(0, &adapter->resetting);
5311
5312         do {
5313                 rc = init_crq_queue(adapter);
5314                 if (rc) {
5315                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5316                                 rc);
5317                         goto ibmvnic_init_fail;
5318                 }
5319
5320                 rc = ibmvnic_reset_init(adapter, false);
5321                 if (rc && rc != EAGAIN)
5322                         goto ibmvnic_init_fail;
5323         } while (rc == EAGAIN);
5324
5325         rc = init_stats_buffers(adapter);
5326         if (rc)
5327                 goto ibmvnic_init_fail;
5328
5329         rc = init_stats_token(adapter);
5330         if (rc)
5331                 goto ibmvnic_stats_fail;
5332
5333         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5334         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5335         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5336
5337         rc = device_create_file(&dev->dev, &dev_attr_failover);
5338         if (rc)
5339                 goto ibmvnic_dev_file_err;
5340
5341         netif_carrier_off(netdev);
5342         rc = register_netdev(netdev);
5343         if (rc) {
5344                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5345                 goto ibmvnic_register_fail;
5346         }
5347         dev_info(&dev->dev, "ibmvnic registered\n");
5348
5349         adapter->state = VNIC_PROBED;
5350
5351         adapter->wait_for_reset = false;
5352         adapter->last_reset_time = jiffies;
5353         return 0;
5354
5355 ibmvnic_register_fail:
5356         device_remove_file(&dev->dev, &dev_attr_failover);
5357
5358 ibmvnic_dev_file_err:
5359         release_stats_token(adapter);
5360
5361 ibmvnic_stats_fail:
5362         release_stats_buffers(adapter);
5363
5364 ibmvnic_init_fail:
5365         release_sub_crqs(adapter, 1);
5366         release_crq_queue(adapter);
5367         mutex_destroy(&adapter->fw_lock);
5368         free_netdev(netdev);
5369
5370         return rc;
5371 }
5372
5373 static int ibmvnic_remove(struct vio_dev *dev)
5374 {
5375         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5376         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5377         unsigned long flags;
5378
5379         spin_lock_irqsave(&adapter->state_lock, flags);
5380         if (test_bit(0, &adapter->resetting)) {
5381                 spin_unlock_irqrestore(&adapter->state_lock, flags);
5382                 return -EBUSY;
5383         }
5384
5385         adapter->state = VNIC_REMOVING;
5386         spin_unlock_irqrestore(&adapter->state_lock, flags);
5387
5388         flush_work(&adapter->ibmvnic_reset);
5389         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5390
5391         rtnl_lock();
5392         unregister_netdevice(netdev);
5393
5394         release_resources(adapter);
5395         release_sub_crqs(adapter, 1);
5396         release_crq_queue(adapter);
5397
5398         release_stats_token(adapter);
5399         release_stats_buffers(adapter);
5400
5401         adapter->state = VNIC_REMOVED;
5402
5403         rtnl_unlock();
5404         mutex_destroy(&adapter->fw_lock);
5405         device_remove_file(&dev->dev, &dev_attr_failover);
5406         free_netdev(netdev);
5407         dev_set_drvdata(&dev->dev, NULL);
5408
5409         return 0;
5410 }
5411
5412 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5413                               const char *buf, size_t count)
5414 {
5415         struct net_device *netdev = dev_get_drvdata(dev);
5416         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5417         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5418         __be64 session_token;
5419         long rc;
5420
5421         if (!sysfs_streq(buf, "1"))
5422                 return -EINVAL;
5423
5424         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5425                          H_GET_SESSION_TOKEN, 0, 0, 0);
5426         if (rc) {
5427                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5428                            rc);
5429                 return -EINVAL;
5430         }
5431
5432         session_token = (__be64)retbuf[0];
5433         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5434                    be64_to_cpu(session_token));
5435         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5436                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5437         if (rc) {
5438                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5439                            rc);
5440                 return -EINVAL;
5441         }
5442
5443         return count;
5444 }
5445
5446 static DEVICE_ATTR_WO(failover);
5447
5448 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5449 {
5450         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5451         struct ibmvnic_adapter *adapter;
5452         struct iommu_table *tbl;
5453         unsigned long ret = 0;
5454         int i;
5455
5456         tbl = get_iommu_table_base(&vdev->dev);
5457
5458         /* netdev inits at probe time along with the structures we need below*/
5459         if (!netdev)
5460                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5461
5462         adapter = netdev_priv(netdev);
5463
5464         ret += PAGE_SIZE; /* the crq message queue */
5465         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5466
5467         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5468                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5469
5470         for (i = 0; i < adapter->num_active_rx_pools; i++)
5471                 ret += adapter->rx_pool[i].size *
5472                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5473
5474         return ret;
5475 }
5476
5477 static int ibmvnic_resume(struct device *dev)
5478 {
5479         struct net_device *netdev = dev_get_drvdata(dev);
5480         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5481
5482         if (adapter->state != VNIC_OPEN)
5483                 return 0;
5484
5485         tasklet_schedule(&adapter->tasklet);
5486
5487         return 0;
5488 }
5489
5490 static const struct vio_device_id ibmvnic_device_table[] = {
5491         {"network", "IBM,vnic"},
5492         {"", "" }
5493 };
5494 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5495
5496 static const struct dev_pm_ops ibmvnic_pm_ops = {
5497         .resume = ibmvnic_resume
5498 };
5499
5500 static struct vio_driver ibmvnic_driver = {
5501         .id_table       = ibmvnic_device_table,
5502         .probe          = ibmvnic_probe,
5503         .remove         = ibmvnic_remove,
5504         .get_desired_dma = ibmvnic_get_desired_dma,
5505         .name           = ibmvnic_driver_name,
5506         .pm             = &ibmvnic_pm_ops,
5507 };
5508
5509 /* module functions */
5510 static int __init ibmvnic_module_init(void)
5511 {
5512         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5513                 IBMVNIC_DRIVER_VERSION);
5514
5515         return vio_register_driver(&ibmvnic_driver);
5516 }
5517
5518 static void __exit ibmvnic_module_exit(void)
5519 {
5520         vio_unregister_driver(&ibmvnic_driver);
5521 }
5522
5523 module_init(ibmvnic_module_init);
5524 module_exit(ibmvnic_module_exit);