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         ind_bufp->index = 0;
 402         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 403                 /* Disable buffer pool replenishment and report carrier off if
 404                  * queue is closed or pending failover.
 405                  * Firmware guarantees that a signal will be sent to the
 406                  * driver, triggering a reset.
 407                  */
 408                 deactivate_rx_pools(adapter);
 409                 netif_carrier_off(adapter->netdev);
 410         }
 411 }
 412
 413 static void replenish_pools(struct ibmvnic_adapter *adapter)
 414 {
 415         int i;
 416
 417         adapter->replenish_task_cycles++;
 418         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 419                 if (adapter->rx_pool[i].active)
 420                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 421         }
 422 }
 423
 424 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 425 {
 426         kfree(adapter->tx_stats_buffers);
 427         kfree(adapter->rx_stats_buffers);
 428         adapter->tx_stats_buffers = NULL;
 429         adapter->rx_stats_buffers = NULL;
 430 }
 431
 432 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 433 {
 434         adapter->tx_stats_buffers =
 435                                 kcalloc(IBMVNIC_MAX_QUEUES,
 436                                         sizeof(struct ibmvnic_tx_queue_stats),
 437                                         GFP_KERNEL);
 438         if (!adapter->tx_stats_buffers)
 439                 return -ENOMEM;
 440
 441         adapter->rx_stats_buffers =
 442                                 kcalloc(IBMVNIC_MAX_QUEUES,
 443                                         sizeof(struct ibmvnic_rx_queue_stats),
 444                                         GFP_KERNEL);
 445         if (!adapter->rx_stats_buffers)
 446                 return -ENOMEM;
 447
 448         return 0;
 449 }
 450
 451 static void release_stats_token(struct ibmvnic_adapter *adapter)
 452 {
 453         struct device *dev = &adapter->vdev->dev;
 454
 455         if (!adapter->stats_token)
 456                 return;
 457
 458         dma_unmap_single(dev, adapter->stats_token,
 459                          sizeof(struct ibmvnic_statistics),
 460                          DMA_FROM_DEVICE);
 461         adapter->stats_token = 0;
 462 }
 463
 464 static int init_stats_token(struct ibmvnic_adapter *adapter)
 465 {
 466         struct device *dev = &adapter->vdev->dev;
 467         dma_addr_t stok;
 468
 469         stok = dma_map_single(dev, &adapter->stats,
 470                               sizeof(struct ibmvnic_statistics),
 471                               DMA_FROM_DEVICE);
 472         if (dma_mapping_error(dev, stok)) {
 473                 dev_err(dev, "Couldn't map stats buffer\n");
 474                 return -1;
 475         }
 476
 477         adapter->stats_token = stok;
 478         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 479         return 0;
 480 }
 481
 482 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 483 {
 484         struct ibmvnic_rx_pool *rx_pool;
 485         u64 buff_size;
 486         int rx_scrqs;
 487         int i, j, rc;
 488
 489         if (!adapter->rx_pool)
 490                 return -1;
 491
 492         buff_size = adapter->cur_rx_buf_sz;
 493         rx_scrqs = adapter->num_active_rx_pools;
 494         for (i = 0; i < rx_scrqs; i++) {
 495                 rx_pool = &adapter->rx_pool[i];
 496
 497                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 498
 499                 if (rx_pool->buff_size != buff_size) {
 500                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 501                         rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 502                         rc = alloc_long_term_buff(adapter,
 503                                                   &rx_pool->long_term_buff,
 504                                                   rx_pool->size *
 505                                                   rx_pool->buff_size);
 506                 } else {
 507                         rc = reset_long_term_buff(adapter,
 508                                                   &rx_pool->long_term_buff);
 509                 }
 510
 511                 if (rc)
 512                         return rc;
 513
 514                 for (j = 0; j < rx_pool->size; j++)
 515                         rx_pool->free_map[j] = j;
 516
 517                 memset(rx_pool->rx_buff, 0,
 518                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 519
 520                 atomic_set(&rx_pool->available, 0);
 521                 rx_pool->next_alloc = 0;
 522                 rx_pool->next_free = 0;
 523                 rx_pool->active = 1;
 524         }
 525
 526         return 0;
 527 }
 528
 529 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 530 {
 531         struct ibmvnic_rx_pool *rx_pool;
 532         int i, j;
 533
 534         if (!adapter->rx_pool)
 535                 return;
 536
 537         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 538                 rx_pool = &adapter->rx_pool[i];
 539
 540                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 541
 542                 kfree(rx_pool->free_map);
 543                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 544
 545                 if (!rx_pool->rx_buff)
 546                         continue;
 547
 548                 for (j = 0; j < rx_pool->size; j++) {
 549                         if (rx_pool->rx_buff[j].skb) {
 550                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 551                                 rx_pool->rx_buff[j].skb = NULL;
 552                         }
 553                 }
 554
 555                 kfree(rx_pool->rx_buff);
 556         }
 557
 558         kfree(adapter->rx_pool);
 559         adapter->rx_pool = NULL;
 560         adapter->num_active_rx_pools = 0;
 561 }
 562
 563 static int init_rx_pools(struct net_device *netdev)
 564 {
 565         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 566         struct device *dev = &adapter->vdev->dev;
 567         struct ibmvnic_rx_pool *rx_pool;
 568         int rxadd_subcrqs;
 569         u64 buff_size;
 570         int i, j;
 571
 572         rxadd_subcrqs = adapter->num_active_rx_scrqs;
 573         buff_size = adapter->cur_rx_buf_sz;
 574
 575         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 576                                    sizeof(struct ibmvnic_rx_pool),
 577                                    GFP_KERNEL);
 578         if (!adapter->rx_pool) {
 579                 dev_err(dev, "Failed to allocate rx pools\n");
 580                 return -1;
 581         }
 582
 583         adapter->num_active_rx_pools = rxadd_subcrqs;
 584
 585         for (i = 0; i < rxadd_subcrqs; i++) {
 586                 rx_pool = &adapter->rx_pool[i];
 587
 588                 netdev_dbg(adapter->netdev,
 589                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 590                            i, adapter->req_rx_add_entries_per_subcrq,
 591                            buff_size);
 592
 593                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 594                 rx_pool->index = i;
 595                 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 596                 rx_pool->active = 1;
 597
 598                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 599                                             GFP_KERNEL);
 600                 if (!rx_pool->free_map) {
 601                         release_rx_pools(adapter);
 602                         return -1;
 603                 }
 604
 605                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 606                                            sizeof(struct ibmvnic_rx_buff),
 607                                            GFP_KERNEL);
 608                 if (!rx_pool->rx_buff) {
 609                         dev_err(dev, "Couldn't alloc rx buffers\n");
 610                         release_rx_pools(adapter);
 611                         return -1;
 612                 }
 613
 614                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 615                                          rx_pool->size * rx_pool->buff_size)) {
 616                         release_rx_pools(adapter);
 617                         return -1;
 618                 }
 619
 620                 for (j = 0; j < rx_pool->size; ++j)
 621                         rx_pool->free_map[j] = j;
 622
 623                 atomic_set(&rx_pool->available, 0);
 624                 rx_pool->next_alloc = 0;
 625                 rx_pool->next_free = 0;
 626         }
 627
 628         return 0;
 629 }
 630
 631 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 632                              struct ibmvnic_tx_pool *tx_pool)
 633 {
 634         int rc, i;
 635
 636         rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 637         if (rc)
 638                 return rc;
 639
 640         memset(tx_pool->tx_buff, 0,
 641                tx_pool->num_buffers *
 642                sizeof(struct ibmvnic_tx_buff));
 643
 644         for (i = 0; i < tx_pool->num_buffers; i++)
 645                 tx_pool->free_map[i] = i;
 646
 647         tx_pool->consumer_index = 0;
 648         tx_pool->producer_index = 0;
 649
 650         return 0;
 651 }
 652
 653 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 654 {
 655         int tx_scrqs;
 656         int i, rc;
 657
 658         if (!adapter->tx_pool)
 659                 return -1;
 660
 661         tx_scrqs = adapter->num_active_tx_pools;
 662         for (i = 0; i < tx_scrqs; i++) {
 663                 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 664                 if (rc)
 665                         return rc;
 666                 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 667                 if (rc)
 668                         return rc;
 669         }
 670
 671         return 0;
 672 }
 673
 674 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 675 {
 676         if (!adapter->vpd)
 677                 return;
 678
 679         kfree(adapter->vpd->buff);
 680         kfree(adapter->vpd);
 681
 682         adapter->vpd = NULL;
 683 }
 684
 685 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 686                                 struct ibmvnic_tx_pool *tx_pool)
 687 {
 688         kfree(tx_pool->tx_buff);
 689         kfree(tx_pool->free_map);
 690         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 691 }
 692
 693 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 694 {
 695         int i;
 696
 697         if (!adapter->tx_pool)
 698                 return;
 699
 700         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 701                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 702                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 703         }
 704
 705         kfree(adapter->tx_pool);
 706         adapter->tx_pool = NULL;
 707         kfree(adapter->tso_pool);
 708         adapter->tso_pool = NULL;
 709         adapter->num_active_tx_pools = 0;
 710 }
 711
 712 static int init_one_tx_pool(struct net_device *netdev,
 713                             struct ibmvnic_tx_pool *tx_pool,
 714                             int num_entries, int buf_size)
 715 {
 716         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 717         int i;
 718
 719         tx_pool->tx_buff = kcalloc(num_entries,
 720                                    sizeof(struct ibmvnic_tx_buff),
 721                                    GFP_KERNEL);
 722         if (!tx_pool->tx_buff)
 723                 return -1;
 724
 725         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 726                                  num_entries * buf_size))
 727                 return -1;
 728
 729         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 730         if (!tx_pool->free_map)
 731                 return -1;
 732
 733         for (i = 0; i < num_entries; i++)
 734                 tx_pool->free_map[i] = i;
 735
 736         tx_pool->consumer_index = 0;
 737         tx_pool->producer_index = 0;
 738         tx_pool->num_buffers = num_entries;
 739         tx_pool->buf_size = buf_size;
 740
 741         return 0;
 742 }
 743
 744 static int init_tx_pools(struct net_device *netdev)
 745 {
 746         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 747         int tx_subcrqs;
 748         u64 buff_size;
 749         int i, rc;
 750
 751         tx_subcrqs = adapter->num_active_tx_scrqs;
 752         adapter->tx_pool = kcalloc(tx_subcrqs,
 753                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 754         if (!adapter->tx_pool)
 755                 return -1;
 756
 757         adapter->tso_pool = kcalloc(tx_subcrqs,
 758                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 759         if (!adapter->tso_pool)
 760                 return -1;
 761
 762         adapter->num_active_tx_pools = tx_subcrqs;
 763
 764         for (i = 0; i < tx_subcrqs; i++) {
 765                 buff_size = adapter->req_mtu + VLAN_HLEN;
 766                 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 767                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 768                                       adapter->req_tx_entries_per_subcrq,
 769                                       buff_size);
 770                 if (rc) {
 771                         release_tx_pools(adapter);
 772                         return rc;
 773                 }
 774
 775                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 776                                       IBMVNIC_TSO_BUFS,
 777                                       IBMVNIC_TSO_BUF_SZ);
 778                 if (rc) {
 779                         release_tx_pools(adapter);
 780                         return rc;
 781                 }
 782         }
 783
 784         return 0;
 785 }
 786
 787 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 788 {
 789         int i;
 790
 791         if (adapter->napi_enabled)
 792                 return;
 793
 794         for (i = 0; i < adapter->req_rx_queues; i++)
 795                 napi_enable(&adapter->napi[i]);
 796
 797         adapter->napi_enabled = true;
 798 }
 799
 800 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 801 {
 802         int i;
 803
 804         if (!adapter->napi_enabled)
 805                 return;
 806
 807         for (i = 0; i < adapter->req_rx_queues; i++) {
 808                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 809                 napi_disable(&adapter->napi[i]);
 810         }
 811
 812         adapter->napi_enabled = false;
 813 }
 814
 815 static int init_napi(struct ibmvnic_adapter *adapter)
 816 {
 817         int i;
 818
 819         adapter->napi = kcalloc(adapter->req_rx_queues,
 820                                 sizeof(struct napi_struct), GFP_KERNEL);
 821         if (!adapter->napi)
 822                 return -ENOMEM;
 823
 824         for (i = 0; i < adapter->req_rx_queues; i++) {
 825                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 826                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 827                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 828         }
 829
 830         adapter->num_active_rx_napi = adapter->req_rx_queues;
 831         return 0;
 832 }
 833
 834 static void release_napi(struct ibmvnic_adapter *adapter)
 835 {
 836         int i;
 837
 838         if (!adapter->napi)
 839                 return;
 840
 841         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 842                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 843                 netif_napi_del(&adapter->napi[i]);
 844         }
 845
 846         kfree(adapter->napi);
 847         adapter->napi = NULL;
 848         adapter->num_active_rx_napi = 0;
 849         adapter->napi_enabled = false;
 850 }
 851
 852 static int ibmvnic_login(struct net_device *netdev)
 853 {
 854         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 855         unsigned long timeout = msecs_to_jiffies(30000);
 856         int retry_count = 0;
 857         int retries = 10;
 858         bool retry;
 859         int rc;
 860
 861         do {
 862                 retry = false;
 863                 if (retry_count > retries) {
 864                         netdev_warn(netdev, "Login attempts exceeded\n");
 865                         return -1;
 866                 }
 867
 868                 adapter->init_done_rc = 0;
 869                 reinit_completion(&adapter->init_done);
 870                 rc = send_login(adapter);
 871                 if (rc) {
 872                         netdev_warn(netdev, "Unable to login\n");
 873                         return rc;
 874                 }
 875
 876                 if (!wait_for_completion_timeout(&adapter->init_done,
 877                                                  timeout)) {
 878                         netdev_warn(netdev, "Login timed out, retrying...\n");
 879                         retry = true;
 880                         adapter->init_done_rc = 0;
 881                         retry_count++;
 882                         continue;
 883                 }
 884
 885                 if (adapter->init_done_rc == ABORTED) {
 886                         netdev_warn(netdev, "Login aborted, retrying...\n");
 887                         retry = true;
 888                         adapter->init_done_rc = 0;
 889                         retry_count++;
 890                         /* FW or device may be busy, so
 891                          * wait a bit before retrying login
 892                          */
 893                         msleep(500);
 894                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 895                         retry_count++;
 896                         release_sub_crqs(adapter, 1);
 897
 898                         retry = true;
 899                         netdev_dbg(netdev,
 900                                    "Received partial success, retrying...\n");
 901                         adapter->init_done_rc = 0;
 902                         reinit_completion(&adapter->init_done);
 903                         send_query_cap(adapter);
 904                         if (!wait_for_completion_timeout(&adapter->init_done,
 905                                                          timeout)) {
 906                                 netdev_warn(netdev,
 907                                             "Capabilities query timed out\n");
 908                                 return -1;
 909                         }
 910
 911                         rc = init_sub_crqs(adapter);
 912                         if (rc) {
 913                                 netdev_warn(netdev,
 914                                             "SCRQ initialization failed\n");
 915                                 return -1;
 916                         }
 917
 918                         rc = init_sub_crq_irqs(adapter);
 919                         if (rc) {
 920                                 netdev_warn(netdev,
 921                                             "SCRQ irq initialization failed\n");
 922                                 return -1;
 923                         }
 924                 } else if (adapter->init_done_rc) {
 925                         netdev_warn(netdev, "Adapter login failed\n");
 926                         return -1;
 927                 }
 928         } while (retry);
 929
 930         __ibmvnic_set_mac(netdev, adapter->mac_addr);
 931
 932         return 0;
 933 }
 934
 935 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 936 {
 937         kfree(adapter->login_buf);
 938         adapter->login_buf = NULL;
 939 }
 940
 941 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 942 {
 943         kfree(adapter->login_rsp_buf);
 944         adapter->login_rsp_buf = NULL;
 945 }
 946
 947 static void release_resources(struct ibmvnic_adapter *adapter)
 948 {
 949         release_vpd_data(adapter);
 950
 951         release_tx_pools(adapter);
 952         release_rx_pools(adapter);
 953
 954         release_napi(adapter);
 955         release_login_rsp_buffer(adapter);
 956 }
 957
 958 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 959 {
 960         struct net_device *netdev = adapter->netdev;
 961         unsigned long timeout = msecs_to_jiffies(30000);
 962         union ibmvnic_crq crq;
 963         bool resend;
 964         int rc;
 965
 966         netdev_dbg(netdev, "setting link state %d\n", link_state);
 967
 968         memset(&crq, 0, sizeof(crq));
 969         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 970         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 971         crq.logical_link_state.link_state = link_state;
 972
 973         do {
 974                 resend = false;
 975
 976                 reinit_completion(&adapter->init_done);
 977                 rc = ibmvnic_send_crq(adapter, &crq);
 978                 if (rc) {
 979                         netdev_err(netdev, "Failed to set link state\n");
 980                         return rc;
 981                 }
 982
 983                 if (!wait_for_completion_timeout(&adapter->init_done,
 984                                                  timeout)) {
 985                         netdev_err(netdev, "timeout setting link state\n");
 986                         return -1;
 987                 }
 988
 989                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 990                         /* Partuial success, delay and re-send */
 991                         mdelay(1000);
 992                         resend = true;
 993                 } else if (adapter->init_done_rc) {
 994                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 995                                     adapter->init_done_rc);
 996                         return adapter->init_done_rc;
 997                 }
 998         } while (resend);
 999
1000         return 0;
1001 }
1002
1003 static int set_real_num_queues(struct net_device *netdev)
1004 {
1005         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1006         int rc;
1007
1008         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1009                    adapter->req_tx_queues, adapter->req_rx_queues);
1010
1011         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1012         if (rc) {
1013                 netdev_err(netdev, "failed to set the number of tx queues\n");
1014                 return rc;
1015         }
1016
1017         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1018         if (rc)
1019                 netdev_err(netdev, "failed to set the number of rx queues\n");
1020
1021         return rc;
1022 }
1023
1024 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1025 {
1026         struct device *dev = &adapter->vdev->dev;
1027         union ibmvnic_crq crq;
1028         int len = 0;
1029         int rc;
1030
1031         if (adapter->vpd->buff)
1032                 len = adapter->vpd->len;
1033
1034         mutex_lock(&adapter->fw_lock);
1035         adapter->fw_done_rc = 0;
1036         reinit_completion(&adapter->fw_done);
1037
1038         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1039         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1040         rc = ibmvnic_send_crq(adapter, &crq);
1041         if (rc) {
1042                 mutex_unlock(&adapter->fw_lock);
1043                 return rc;
1044         }
1045
1046         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1047         if (rc) {
1048                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1049                 mutex_unlock(&adapter->fw_lock);
1050                 return rc;
1051         }
1052         mutex_unlock(&adapter->fw_lock);
1053
1054         if (!adapter->vpd->len)
1055                 return -ENODATA;
1056
1057         if (!adapter->vpd->buff)
1058                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1059         else if (adapter->vpd->len != len)
1060                 adapter->vpd->buff =
1061                         krealloc(adapter->vpd->buff,
1062                                  adapter->vpd->len, GFP_KERNEL);
1063
1064         if (!adapter->vpd->buff) {
1065                 dev_err(dev, "Could allocate VPD buffer\n");
1066                 return -ENOMEM;
1067         }
1068
1069         adapter->vpd->dma_addr =
1070                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1071                                DMA_FROM_DEVICE);
1072         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1073                 dev_err(dev, "Could not map VPD buffer\n");
1074                 kfree(adapter->vpd->buff);
1075                 adapter->vpd->buff = NULL;
1076                 return -ENOMEM;
1077         }
1078
1079         mutex_lock(&adapter->fw_lock);
1080         adapter->fw_done_rc = 0;
1081         reinit_completion(&adapter->fw_done);
1082
1083         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1084         crq.get_vpd.cmd = GET_VPD;
1085         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1086         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1087         rc = ibmvnic_send_crq(adapter, &crq);
1088         if (rc) {
1089                 kfree(adapter->vpd->buff);
1090                 adapter->vpd->buff = NULL;
1091                 mutex_unlock(&adapter->fw_lock);
1092                 return rc;
1093         }
1094
1095         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1096         if (rc) {
1097                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1098                 kfree(adapter->vpd->buff);
1099                 adapter->vpd->buff = NULL;
1100                 mutex_unlock(&adapter->fw_lock);
1101                 return rc;
1102         }
1103
1104         mutex_unlock(&adapter->fw_lock);
1105         return 0;
1106 }
1107
1108 static int init_resources(struct ibmvnic_adapter *adapter)
1109 {
1110         struct net_device *netdev = adapter->netdev;
1111         int rc;
1112
1113         rc = set_real_num_queues(netdev);
1114         if (rc)
1115                 return rc;
1116
1117         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1118         if (!adapter->vpd)
1119                 return -ENOMEM;
1120
1121         /* Vital Product Data (VPD) */
1122         rc = ibmvnic_get_vpd(adapter);
1123         if (rc) {
1124                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1125                 return rc;
1126         }
1127
1128         adapter->map_id = 1;
1129
1130         rc = init_napi(adapter);
1131         if (rc)
1132                 return rc;
1133
1134         send_query_map(adapter);
1135
1136         rc = init_rx_pools(netdev);
1137         if (rc)
1138                 return rc;
1139
1140         rc = init_tx_pools(netdev);
1141         return rc;
1142 }
1143
1144 static int __ibmvnic_open(struct net_device *netdev)
1145 {
1146         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1147         enum vnic_state prev_state = adapter->state;
1148         int i, rc;
1149
1150         adapter->state = VNIC_OPENING;
1151         replenish_pools(adapter);
1152         ibmvnic_napi_enable(adapter);
1153
1154         /* We're ready to receive frames, enable the sub-crq interrupts and
1155          * set the logical link state to up
1156          */
1157         for (i = 0; i < adapter->req_rx_queues; i++) {
1158                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1159                 if (prev_state == VNIC_CLOSED)
1160                         enable_irq(adapter->rx_scrq[i]->irq);
1161                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1162         }
1163
1164         for (i = 0; i < adapter->req_tx_queues; i++) {
1165                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1166                 if (prev_state == VNIC_CLOSED)
1167                         enable_irq(adapter->tx_scrq[i]->irq);
1168                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1169                 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1170         }
1171
1172         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1173         if (rc) {
1174                 for (i = 0; i < adapter->req_rx_queues; i++)
1175                         napi_disable(&adapter->napi[i]);
1176                 release_resources(adapter);
1177                 return rc;
1178         }
1179
1180         netif_tx_start_all_queues(netdev);
1181
1182         if (prev_state == VNIC_CLOSED) {
1183                 for (i = 0; i < adapter->req_rx_queues; i++)
1184                         napi_schedule(&adapter->napi[i]);
1185         }
1186
1187         adapter->state = VNIC_OPEN;
1188         return rc;
1189 }
1190
1191 static int ibmvnic_open(struct net_device *netdev)
1192 {
1193         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1194         int rc;
1195
1196         /* If device failover is pending, just set device state and return.
1197          * Device operation will be handled by reset routine.
1198          */
1199         if (adapter->failover_pending) {
1200                 adapter->state = VNIC_OPEN;
1201                 return 0;
1202         }
1203
1204         if (adapter->state != VNIC_CLOSED) {
1205                 rc = ibmvnic_login(netdev);
1206                 if (rc)
1207                         goto out;
1208
1209                 rc = init_resources(adapter);
1210                 if (rc) {
1211                         netdev_err(netdev, "failed to initialize resources\n");
1212                         release_resources(adapter);
1213                         goto out;
1214                 }
1215         }
1216
1217         rc = __ibmvnic_open(netdev);
1218
1219 out:
1220         /*
1221          * If open fails due to a pending failover, set device state and
1222          * return. Device operation will be handled by reset routine.
1223          */
1224         if (rc && adapter->failover_pending) {
1225                 adapter->state = VNIC_OPEN;
1226                 rc = 0;
1227         }
1228         return rc;
1229 }
1230
1231 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1232 {
1233         struct ibmvnic_rx_pool *rx_pool;
1234         struct ibmvnic_rx_buff *rx_buff;
1235         u64 rx_entries;
1236         int rx_scrqs;
1237         int i, j;
1238
1239         if (!adapter->rx_pool)
1240                 return;
1241
1242         rx_scrqs = adapter->num_active_rx_pools;
1243         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1244
1245         /* Free any remaining skbs in the rx buffer pools */
1246         for (i = 0; i < rx_scrqs; i++) {
1247                 rx_pool = &adapter->rx_pool[i];
1248                 if (!rx_pool || !rx_pool->rx_buff)
1249                         continue;
1250
1251                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1252                 for (j = 0; j < rx_entries; j++) {
1253                         rx_buff = &rx_pool->rx_buff[j];
1254                         if (rx_buff && rx_buff->skb) {
1255                                 dev_kfree_skb_any(rx_buff->skb);
1256                                 rx_buff->skb = NULL;
1257                         }
1258                 }
1259         }
1260 }
1261
1262 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1263                               struct ibmvnic_tx_pool *tx_pool)
1264 {
1265         struct ibmvnic_tx_buff *tx_buff;
1266         u64 tx_entries;
1267         int i;
1268
1269         if (!tx_pool || !tx_pool->tx_buff)
1270                 return;
1271
1272         tx_entries = tx_pool->num_buffers;
1273
1274         for (i = 0; i < tx_entries; i++) {
1275                 tx_buff = &tx_pool->tx_buff[i];
1276                 if (tx_buff && tx_buff->skb) {
1277                         dev_kfree_skb_any(tx_buff->skb);
1278                         tx_buff->skb = NULL;
1279                 }
1280         }
1281 }
1282
1283 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1284 {
1285         int tx_scrqs;
1286         int i;
1287
1288         if (!adapter->tx_pool || !adapter->tso_pool)
1289                 return;
1290
1291         tx_scrqs = adapter->num_active_tx_pools;
1292
1293         /* Free any remaining skbs in the tx buffer pools */
1294         for (i = 0; i < tx_scrqs; i++) {
1295                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1296                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1297                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1298         }
1299 }
1300
1301 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1302 {
1303         struct net_device *netdev = adapter->netdev;
1304         int i;
1305
1306         if (adapter->tx_scrq) {
1307                 for (i = 0; i < adapter->req_tx_queues; i++)
1308                         if (adapter->tx_scrq[i]->irq) {
1309                                 netdev_dbg(netdev,
1310                                            "Disabling tx_scrq[%d] irq\n", i);
1311                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1312                                 disable_irq(adapter->tx_scrq[i]->irq);
1313                         }
1314         }
1315
1316         if (adapter->rx_scrq) {
1317                 for (i = 0; i < adapter->req_rx_queues; i++) {
1318                         if (adapter->rx_scrq[i]->irq) {
1319                                 netdev_dbg(netdev,
1320                                            "Disabling rx_scrq[%d] irq\n", i);
1321                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1322                                 disable_irq(adapter->rx_scrq[i]->irq);
1323                         }
1324                 }
1325         }
1326 }
1327
1328 static void ibmvnic_cleanup(struct net_device *netdev)
1329 {
1330         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1331
1332         /* ensure that transmissions are stopped if called by do_reset */
1333         if (test_bit(0, &adapter->resetting))
1334                 netif_tx_disable(netdev);
1335         else
1336                 netif_tx_stop_all_queues(netdev);
1337
1338         ibmvnic_napi_disable(adapter);
1339         ibmvnic_disable_irqs(adapter);
1340
1341         clean_rx_pools(adapter);
1342         clean_tx_pools(adapter);
1343 }
1344
1345 static int __ibmvnic_close(struct net_device *netdev)
1346 {
1347         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1348         int rc = 0;
1349
1350         adapter->state = VNIC_CLOSING;
1351         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1352         if (rc)
1353                 return rc;
1354         adapter->state = VNIC_CLOSED;
1355         return 0;
1356 }
1357
1358 static int ibmvnic_close(struct net_device *netdev)
1359 {
1360         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1361         int rc;
1362
1363         /* If device failover is pending, just set device state and return.
1364          * Device operation will be handled by reset routine.
1365          */
1366         if (adapter->failover_pending) {
1367                 adapter->state = VNIC_CLOSED;
1368                 return 0;
1369         }
1370
1371         rc = __ibmvnic_close(netdev);
1372         ibmvnic_cleanup(netdev);
1373
1374         return rc;
1375 }
1376
1377 /**
1378  * build_hdr_data - creates L2/L3/L4 header data buffer
1379  * @hdr_field - bitfield determining needed headers
1380  * @skb - socket buffer
1381  * @hdr_len - array of header lengths
1382  * @tot_len - total length of data
1383  *
1384  * Reads hdr_field to determine which headers are needed by firmware.
1385  * Builds a buffer containing these headers.  Saves individual header
1386  * lengths and total buffer length to be used to build descriptors.
1387  */
1388 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1389                           int *hdr_len, u8 *hdr_data)
1390 {
1391         int len = 0;
1392         u8 *hdr;
1393
1394         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1395                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1396         else
1397                 hdr_len[0] = sizeof(struct ethhdr);
1398
1399         if (skb->protocol == htons(ETH_P_IP)) {
1400                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1401                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1402                         hdr_len[2] = tcp_hdrlen(skb);
1403                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1404                         hdr_len[2] = sizeof(struct udphdr);
1405         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1406                 hdr_len[1] = sizeof(struct ipv6hdr);
1407                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1408                         hdr_len[2] = tcp_hdrlen(skb);
1409                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1410                         hdr_len[2] = sizeof(struct udphdr);
1411         } else if (skb->protocol == htons(ETH_P_ARP)) {
1412                 hdr_len[1] = arp_hdr_len(skb->dev);
1413                 hdr_len[2] = 0;
1414         }
1415
1416         memset(hdr_data, 0, 120);
1417         if ((hdr_field >> 6) & 1) {
1418                 hdr = skb_mac_header(skb);
1419                 memcpy(hdr_data, hdr, hdr_len[0]);
1420                 len += hdr_len[0];
1421         }
1422
1423         if ((hdr_field >> 5) & 1) {
1424                 hdr = skb_network_header(skb);
1425                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1426                 len += hdr_len[1];
1427         }
1428
1429         if ((hdr_field >> 4) & 1) {
1430                 hdr = skb_transport_header(skb);
1431                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1432                 len += hdr_len[2];
1433         }
1434         return len;
1435 }
1436
1437 /**
1438  * create_hdr_descs - create header and header extension descriptors
1439  * @hdr_field - bitfield determining needed headers
1440  * @data - buffer containing header data
1441  * @len - length of data buffer
1442  * @hdr_len - array of individual header lengths
1443  * @scrq_arr - descriptor array
1444  *
1445  * Creates header and, if needed, header extension descriptors and
1446  * places them in a descriptor array, scrq_arr
1447  */
1448
1449 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1450                             union sub_crq *scrq_arr)
1451 {
1452         union sub_crq hdr_desc;
1453         int tmp_len = len;
1454         int num_descs = 0;
1455         u8 *data, *cur;
1456         int tmp;
1457
1458         while (tmp_len > 0) {
1459                 cur = hdr_data + len - tmp_len;
1460
1461                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1462                 if (cur != hdr_data) {
1463                         data = hdr_desc.hdr_ext.data;
1464                         tmp = tmp_len > 29 ? 29 : tmp_len;
1465                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1466                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1467                         hdr_desc.hdr_ext.len = tmp;
1468                 } else {
1469                         data = hdr_desc.hdr.data;
1470                         tmp = tmp_len > 24 ? 24 : tmp_len;
1471                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1472                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1473                         hdr_desc.hdr.len = tmp;
1474                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1475                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1476                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1477                         hdr_desc.hdr.flag = hdr_field << 1;
1478                 }
1479                 memcpy(data, cur, tmp);
1480                 tmp_len -= tmp;
1481                 *scrq_arr = hdr_desc;
1482                 scrq_arr++;
1483                 num_descs++;
1484         }
1485
1486         return num_descs;
1487 }
1488
1489 /**
1490  * build_hdr_descs_arr - build a header descriptor array
1491  * @skb - socket buffer
1492  * @num_entries - number of descriptors to be sent
1493  * @subcrq - first TX descriptor
1494  * @hdr_field - bit field determining which headers will be sent
1495  *
1496  * This function will build a TX descriptor array with applicable
1497  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1498  */
1499
1500 static void build_hdr_descs_arr(struct sk_buff *skb,
1501                                 union sub_crq *indir_arr,
1502                                 int *num_entries, u8 hdr_field)
1503 {
1504         int hdr_len[3] = {0, 0, 0};
1505         u8 hdr_data[140] = {0};
1506         int tot_len;
1507
1508         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1509                                  hdr_data);
1510         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1511                                          indir_arr + 1);
1512 }
1513
1514 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1515                                     struct net_device *netdev)
1516 {
1517         /* For some backing devices, mishandling of small packets
1518          * can result in a loss of connection or TX stall. Device
1519          * architects recommend that no packet should be smaller
1520          * than the minimum MTU value provided to the driver, so
1521          * pad any packets to that length
1522          */
1523         if (skb->len < netdev->min_mtu)
1524                 return skb_put_padto(skb, netdev->min_mtu);
1525
1526         return 0;
1527 }
1528
1529 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1530                                          struct ibmvnic_sub_crq_queue *tx_scrq)
1531 {
1532         struct ibmvnic_ind_xmit_queue *ind_bufp;
1533         struct ibmvnic_tx_buff *tx_buff;
1534         struct ibmvnic_tx_pool *tx_pool;
1535         union sub_crq tx_scrq_entry;
1536         int queue_num;
1537         int entries;
1538         int index;
1539         int i;
1540
1541         ind_bufp = &tx_scrq->ind_buf;
1542         entries = (u64)ind_bufp->index;
1543         queue_num = tx_scrq->pool_index;
1544
1545         for (i = entries - 1; i >= 0; --i) {
1546                 tx_scrq_entry = ind_bufp->indir_arr[i];
1547                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1548                         continue;
1549                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1550                 if (index & IBMVNIC_TSO_POOL_MASK) {
1551                         tx_pool = &adapter->tso_pool[queue_num];
1552                         index &= ~IBMVNIC_TSO_POOL_MASK;
1553                 } else {
1554                         tx_pool = &adapter->tx_pool[queue_num];
1555                 }
1556                 tx_pool->free_map[tx_pool->consumer_index] = index;
1557                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1558                                           tx_pool->num_buffers - 1 :
1559                                           tx_pool->consumer_index - 1;
1560                 tx_buff = &tx_pool->tx_buff[index];
1561                 adapter->netdev->stats.tx_packets--;
1562                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1563                 adapter->tx_stats_buffers[queue_num].packets--;
1564                 adapter->tx_stats_buffers[queue_num].bytes -=
1565                                                 tx_buff->skb->len;
1566                 dev_kfree_skb_any(tx_buff->skb);
1567                 tx_buff->skb = NULL;
1568                 adapter->netdev->stats.tx_dropped++;
1569         }
1570         ind_bufp->index = 0;
1571         if (atomic_sub_return(entries, &tx_scrq->used) <=
1572             (adapter->req_tx_entries_per_subcrq / 2) &&
1573             __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1574                 netif_wake_subqueue(adapter->netdev, queue_num);
1575                 netdev_dbg(adapter->netdev, "Started queue %d\n",
1576                            queue_num);
1577         }
1578 }
1579
1580 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1581                                  struct ibmvnic_sub_crq_queue *tx_scrq)
1582 {
1583         struct ibmvnic_ind_xmit_queue *ind_bufp;
1584         u64 dma_addr;
1585         u64 entries;
1586         u64 handle;
1587         int rc;
1588
1589         ind_bufp = &tx_scrq->ind_buf;
1590         dma_addr = (u64)ind_bufp->indir_dma;
1591         entries = (u64)ind_bufp->index;
1592         handle = tx_scrq->handle;
1593
1594         if (!entries)
1595                 return 0;
1596         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1597         if (rc)
1598                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1599         else
1600                 ind_bufp->index = 0;
1601         return 0;
1602 }
1603
1604 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1605 {
1606         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1607         int queue_num = skb_get_queue_mapping(skb);
1608         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1609         struct device *dev = &adapter->vdev->dev;
1610         struct ibmvnic_ind_xmit_queue *ind_bufp;
1611         struct ibmvnic_tx_buff *tx_buff = NULL;
1612         struct ibmvnic_sub_crq_queue *tx_scrq;
1613         struct ibmvnic_tx_pool *tx_pool;
1614         unsigned int tx_send_failed = 0;
1615         netdev_tx_t ret = NETDEV_TX_OK;
1616         unsigned int tx_map_failed = 0;
1617         union sub_crq indir_arr[16];
1618         unsigned int tx_dropped = 0;
1619         unsigned int tx_packets = 0;
1620         unsigned int tx_bytes = 0;
1621         dma_addr_t data_dma_addr;
1622         struct netdev_queue *txq;
1623         unsigned long lpar_rc;
1624         union sub_crq tx_crq;
1625         unsigned int offset;
1626         int num_entries = 1;
1627         unsigned char *dst;
1628         int index = 0;
1629         u8 proto = 0;
1630
1631         tx_scrq = adapter->tx_scrq[queue_num];
1632         txq = netdev_get_tx_queue(netdev, queue_num);
1633         ind_bufp = &tx_scrq->ind_buf;
1634
1635         if (test_bit(0, &adapter->resetting)) {
1636                 if (!netif_subqueue_stopped(netdev, skb))
1637                         netif_stop_subqueue(netdev, queue_num);
1638                 dev_kfree_skb_any(skb);
1639
1640                 tx_send_failed++;
1641                 tx_dropped++;
1642                 ret = NETDEV_TX_OK;
1643                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1644                 goto out;
1645         }
1646
1647         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1648                 tx_dropped++;
1649                 tx_send_failed++;
1650                 ret = NETDEV_TX_OK;
1651                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1652                 goto out;
1653         }
1654         if (skb_is_gso(skb))
1655                 tx_pool = &adapter->tso_pool[queue_num];
1656         else
1657                 tx_pool = &adapter->tx_pool[queue_num];
1658
1659         index = tx_pool->free_map[tx_pool->consumer_index];
1660
1661         if (index == IBMVNIC_INVALID_MAP) {
1662                 dev_kfree_skb_any(skb);
1663                 tx_send_failed++;
1664                 tx_dropped++;
1665                 ret = NETDEV_TX_OK;
1666                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1667                 goto out;
1668         }
1669
1670         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1671
1672         offset = index * tx_pool->buf_size;
1673         dst = tx_pool->long_term_buff.buff + offset;
1674         memset(dst, 0, tx_pool->buf_size);
1675         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1676
1677         if (skb_shinfo(skb)->nr_frags) {
1678                 int cur, i;
1679
1680                 /* Copy the head */
1681                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1682                 cur = skb_headlen(skb);
1683
1684                 /* Copy the frags */
1685                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1686                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1687
1688                         memcpy(dst + cur,
1689                                page_address(skb_frag_page(frag)) +
1690                                skb_frag_off(frag), skb_frag_size(frag));
1691                         cur += skb_frag_size(frag);
1692                 }
1693         } else {
1694                 skb_copy_from_linear_data(skb, dst, skb->len);
1695         }
1696
1697         tx_pool->consumer_index =
1698             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1699
1700         tx_buff = &tx_pool->tx_buff[index];
1701         tx_buff->skb = skb;
1702         tx_buff->index = index;
1703         tx_buff->pool_index = queue_num;
1704
1705         memset(&tx_crq, 0, sizeof(tx_crq));
1706         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1707         tx_crq.v1.type = IBMVNIC_TX_DESC;
1708         tx_crq.v1.n_crq_elem = 1;
1709         tx_crq.v1.n_sge = 1;
1710         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1711
1712         if (skb_is_gso(skb))
1713                 tx_crq.v1.correlator =
1714                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1715         else
1716                 tx_crq.v1.correlator = cpu_to_be32(index);
1717         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1718         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1719         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1720
1721         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1722                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1723                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1724         }
1725
1726         if (skb->protocol == htons(ETH_P_IP)) {
1727                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1728                 proto = ip_hdr(skb)->protocol;
1729         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1730                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1731                 proto = ipv6_hdr(skb)->nexthdr;
1732         }
1733
1734         if (proto == IPPROTO_TCP)
1735                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1736         else if (proto == IPPROTO_UDP)
1737                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1738
1739         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1740                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1741                 hdrs += 2;
1742         }
1743         if (skb_is_gso(skb)) {
1744                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1745                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1746                 hdrs += 2;
1747         }
1748
1749         if ((*hdrs >> 7) & 1)
1750                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1751
1752         tx_crq.v1.n_crq_elem = num_entries;
1753         tx_buff->num_entries = num_entries;
1754         /* flush buffer if current entry can not fit */
1755         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1756                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1757                 if (lpar_rc != H_SUCCESS)
1758                         goto tx_flush_err;
1759         }
1760
1761         indir_arr[0] = tx_crq;
1762         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1763                num_entries * sizeof(struct ibmvnic_generic_scrq));
1764         ind_bufp->index += num_entries;
1765         if (__netdev_tx_sent_queue(txq, skb->len,
1766                                    netdev_xmit_more() &&
1767                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1768                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1769                 if (lpar_rc != H_SUCCESS)
1770                         goto tx_err;
1771         }
1772
1773         if (atomic_add_return(num_entries, &tx_scrq->used)
1774                                         >= adapter->req_tx_entries_per_subcrq) {
1775                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1776                 netif_stop_subqueue(netdev, queue_num);
1777         }
1778
1779         tx_packets++;
1780         tx_bytes += skb->len;
1781         txq->trans_start = jiffies;
1782         ret = NETDEV_TX_OK;
1783         goto out;
1784
1785 tx_flush_err:
1786         dev_kfree_skb_any(skb);
1787         tx_buff->skb = NULL;
1788         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1789                                   tx_pool->num_buffers - 1 :
1790                                   tx_pool->consumer_index - 1;
1791         tx_dropped++;
1792 tx_err:
1793         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1794                 dev_err_ratelimited(dev, "tx: send failed\n");
1795
1796         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1797                 /* Disable TX and report carrier off if queue is closed
1798                  * or pending failover.
1799                  * Firmware guarantees that a signal will be sent to the
1800                  * driver, triggering a reset or some other action.
1801                  */
1802                 netif_tx_stop_all_queues(netdev);
1803                 netif_carrier_off(netdev);
1804         }
1805 out:
1806         netdev->stats.tx_dropped += tx_dropped;
1807         netdev->stats.tx_bytes += tx_bytes;
1808         netdev->stats.tx_packets += tx_packets;
1809         adapter->tx_send_failed += tx_send_failed;
1810         adapter->tx_map_failed += tx_map_failed;
1811         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1812         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1813         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1814
1815         return ret;
1816 }
1817
1818 static void ibmvnic_set_multi(struct net_device *netdev)
1819 {
1820         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1821         struct netdev_hw_addr *ha;
1822         union ibmvnic_crq crq;
1823
1824         memset(&crq, 0, sizeof(crq));
1825         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1826         crq.request_capability.cmd = REQUEST_CAPABILITY;
1827
1828         if (netdev->flags & IFF_PROMISC) {
1829                 if (!adapter->promisc_supported)
1830                         return;
1831         } else {
1832                 if (netdev->flags & IFF_ALLMULTI) {
1833                         /* Accept all multicast */
1834                         memset(&crq, 0, sizeof(crq));
1835                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1836                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1837                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1838                         ibmvnic_send_crq(adapter, &crq);
1839                 } else if (netdev_mc_empty(netdev)) {
1840                         /* Reject 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_DISABLE_ALL;
1845                         ibmvnic_send_crq(adapter, &crq);
1846                 } else {
1847                         /* Accept one or more multicast(s) */
1848                         netdev_for_each_mc_addr(ha, netdev) {
1849                                 memset(&crq, 0, sizeof(crq));
1850                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1851                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1852                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1853                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1854                                                 ha->addr);
1855                                 ibmvnic_send_crq(adapter, &crq);
1856                         }
1857                 }
1858         }
1859 }
1860
1861 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1862 {
1863         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1864         union ibmvnic_crq crq;
1865         int rc;
1866
1867         if (!is_valid_ether_addr(dev_addr)) {
1868                 rc = -EADDRNOTAVAIL;
1869                 goto err;
1870         }
1871
1872         memset(&crq, 0, sizeof(crq));
1873         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1874         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1875         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1876
1877         mutex_lock(&adapter->fw_lock);
1878         adapter->fw_done_rc = 0;
1879         reinit_completion(&adapter->fw_done);
1880
1881         rc = ibmvnic_send_crq(adapter, &crq);
1882         if (rc) {
1883                 rc = -EIO;
1884                 mutex_unlock(&adapter->fw_lock);
1885                 goto err;
1886         }
1887
1888         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1889         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1890         if (rc || adapter->fw_done_rc) {
1891                 rc = -EIO;
1892                 mutex_unlock(&adapter->fw_lock);
1893                 goto err;
1894         }
1895         mutex_unlock(&adapter->fw_lock);
1896         return 0;
1897 err:
1898         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1899         return rc;
1900 }
1901
1902 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1903 {
1904         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1905         struct sockaddr *addr = p;
1906         int rc;
1907
1908         rc = 0;
1909         if (!is_valid_ether_addr(addr->sa_data))
1910                 return -EADDRNOTAVAIL;
1911
1912         if (adapter->state != VNIC_PROBED) {
1913                 ether_addr_copy(adapter->mac_addr, addr->sa_data);
1914                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1915         }
1916
1917         return rc;
1918 }
1919
1920 /**
1921  * do_change_param_reset returns zero if we are able to keep processing reset
1922  * events, or non-zero if we hit a fatal error and must halt.
1923  */
1924 static int do_change_param_reset(struct ibmvnic_adapter *adapter,
1925                                  struct ibmvnic_rwi *rwi,
1926                                  u32 reset_state)
1927 {
1928         struct net_device *netdev = adapter->netdev;
1929         int i, rc;
1930
1931         netdev_dbg(adapter->netdev, "Change param resetting driver (%d)\n",
1932                    rwi->reset_reason);
1933
1934         netif_carrier_off(netdev);
1935         adapter->reset_reason = rwi->reset_reason;
1936
1937         ibmvnic_cleanup(netdev);
1938
1939         if (reset_state == VNIC_OPEN) {
1940                 rc = __ibmvnic_close(netdev);
1941                 if (rc)
1942                         return rc;
1943         }
1944
1945         release_resources(adapter);
1946         release_sub_crqs(adapter, 1);
1947         release_crq_queue(adapter);
1948
1949         adapter->state = VNIC_PROBED;
1950
1951         rc = init_crq_queue(adapter);
1952
1953         if (rc) {
1954                 netdev_err(adapter->netdev,
1955                            "Couldn't initialize crq. rc=%d\n", rc);
1956                 return rc;
1957         }
1958
1959         rc = ibmvnic_reset_init(adapter, true);
1960         if (rc)
1961                 return IBMVNIC_INIT_FAILED;
1962
1963         /* If the adapter was in PROBE state prior to the reset,
1964          * exit here.
1965          */
1966         if (reset_state == VNIC_PROBED)
1967                 return 0;
1968
1969         rc = ibmvnic_login(netdev);
1970         if (rc) {
1971                 adapter->state = reset_state;
1972                 return rc;
1973         }
1974
1975         rc = init_resources(adapter);
1976         if (rc)
1977                 return rc;
1978
1979         ibmvnic_disable_irqs(adapter);
1980
1981         adapter->state = VNIC_CLOSED;
1982
1983         if (reset_state == VNIC_CLOSED)
1984                 return 0;
1985
1986         rc = __ibmvnic_open(netdev);
1987         if (rc)
1988                 return IBMVNIC_OPEN_FAILED;
1989
1990         /* refresh device's multicast list */
1991         ibmvnic_set_multi(netdev);
1992
1993         /* kick napi */
1994         for (i = 0; i < adapter->req_rx_queues; i++)
1995                 napi_schedule(&adapter->napi[i]);
1996
1997         return 0;
1998 }
1999
2000 /**
2001  * do_reset returns zero if we are able to keep processing reset events, or
2002  * non-zero if we hit a fatal error and must halt.
2003  */
2004 static int do_reset(struct ibmvnic_adapter *adapter,
2005                     struct ibmvnic_rwi *rwi, u32 reset_state)
2006 {
2007         u64 old_num_rx_queues, old_num_tx_queues;
2008         u64 old_num_rx_slots, old_num_tx_slots;
2009         struct net_device *netdev = adapter->netdev;
2010         int i, rc;
2011
2012         netdev_dbg(adapter->netdev, "Re-setting driver (%d)\n",
2013                    rwi->reset_reason);
2014
2015         rtnl_lock();
2016         /*
2017          * Now that we have the rtnl lock, clear any pending failover.
2018          * This will ensure ibmvnic_open() has either completed or will
2019          * block until failover is complete.
2020          */
2021         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2022                 adapter->failover_pending = false;
2023
2024         netif_carrier_off(netdev);
2025         adapter->reset_reason = rwi->reset_reason;
2026
2027         old_num_rx_queues = adapter->req_rx_queues;
2028         old_num_tx_queues = adapter->req_tx_queues;
2029         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2030         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2031
2032         ibmvnic_cleanup(netdev);
2033
2034         if (reset_state == VNIC_OPEN &&
2035             adapter->reset_reason != VNIC_RESET_MOBILITY &&
2036             adapter->reset_reason != VNIC_RESET_FAILOVER) {
2037                 adapter->state = VNIC_CLOSING;
2038
2039                 /* Release the RTNL lock before link state change and
2040                  * re-acquire after the link state change to allow
2041                  * linkwatch_event to grab the RTNL lock and run during
2042                  * a reset.
2043                  */
2044                 rtnl_unlock();
2045                 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2046                 rtnl_lock();
2047                 if (rc)
2048                         goto out;
2049
2050                 if (adapter->state != VNIC_CLOSING) {
2051                         rc = -1;
2052                         goto out;
2053                 }
2054
2055                 adapter->state = VNIC_CLOSED;
2056         }
2057
2058         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2059                 /* remove the closed state so when we call open it appears
2060                  * we are coming from the probed state.
2061                  */
2062                 adapter->state = VNIC_PROBED;
2063
2064                 if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2065                         rc = ibmvnic_reenable_crq_queue(adapter);
2066                         release_sub_crqs(adapter, 1);
2067                 } else {
2068                         rc = ibmvnic_reset_crq(adapter);
2069                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2070                                 rc = vio_enable_interrupts(adapter->vdev);
2071                                 if (rc)
2072                                         netdev_err(adapter->netdev,
2073                                                    "Reset failed to enable interrupts. rc=%d\n",
2074                                                    rc);
2075                         }
2076                 }
2077
2078                 if (rc) {
2079                         netdev_err(adapter->netdev,
2080                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2081                         goto out;
2082                 }
2083
2084                 rc = ibmvnic_reset_init(adapter, true);
2085                 if (rc) {
2086                         rc = IBMVNIC_INIT_FAILED;
2087                         goto out;
2088                 }
2089
2090                 /* If the adapter was in PROBE state prior to the reset,
2091                  * exit here.
2092                  */
2093                 if (reset_state == VNIC_PROBED) {
2094                         rc = 0;
2095                         goto out;
2096                 }
2097
2098                 rc = ibmvnic_login(netdev);
2099                 if (rc) {
2100                         adapter->state = reset_state;
2101                         goto out;
2102                 }
2103
2104                 if (adapter->req_rx_queues != old_num_rx_queues ||
2105                     adapter->req_tx_queues != old_num_tx_queues ||
2106                     adapter->req_rx_add_entries_per_subcrq !=
2107                     old_num_rx_slots ||
2108                     adapter->req_tx_entries_per_subcrq !=
2109                     old_num_tx_slots ||
2110                     !adapter->rx_pool ||
2111                     !adapter->tso_pool ||
2112                     !adapter->tx_pool) {
2113                         release_rx_pools(adapter);
2114                         release_tx_pools(adapter);
2115                         release_napi(adapter);
2116                         release_vpd_data(adapter);
2117
2118                         rc = init_resources(adapter);
2119                         if (rc)
2120                                 goto out;
2121
2122                 } else {
2123                         rc = reset_tx_pools(adapter);
2124                         if (rc) {
2125                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2126                                                 rc);
2127                                 goto out;
2128                         }
2129
2130                         rc = reset_rx_pools(adapter);
2131                         if (rc) {
2132                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2133                                                 rc);
2134                                 goto out;
2135                         }
2136                 }
2137                 ibmvnic_disable_irqs(adapter);
2138         }
2139         adapter->state = VNIC_CLOSED;
2140
2141         if (reset_state == VNIC_CLOSED) {
2142                 rc = 0;
2143                 goto out;
2144         }
2145
2146         rc = __ibmvnic_open(netdev);
2147         if (rc) {
2148                 rc = IBMVNIC_OPEN_FAILED;
2149                 goto out;
2150         }
2151
2152         /* refresh device's multicast list */
2153         ibmvnic_set_multi(netdev);
2154
2155         /* kick napi */
2156         for (i = 0; i < adapter->req_rx_queues; i++)
2157                 napi_schedule(&adapter->napi[i]);
2158
2159         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2160             adapter->reset_reason == VNIC_RESET_MOBILITY) {
2161                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
2162                 call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
2163         }
2164
2165         rc = 0;
2166
2167 out:
2168         rtnl_unlock();
2169
2170         return rc;
2171 }
2172
2173 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2174                          struct ibmvnic_rwi *rwi, u32 reset_state)
2175 {
2176         struct net_device *netdev = adapter->netdev;
2177         int rc;
2178
2179         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2180                    rwi->reset_reason);
2181
2182         netif_carrier_off(netdev);
2183         adapter->reset_reason = rwi->reset_reason;
2184
2185         ibmvnic_cleanup(netdev);
2186         release_resources(adapter);
2187         release_sub_crqs(adapter, 0);
2188         release_crq_queue(adapter);
2189
2190         /* remove the closed state so when we call open it appears
2191          * we are coming from the probed state.
2192          */
2193         adapter->state = VNIC_PROBED;
2194
2195         reinit_completion(&adapter->init_done);
2196         rc = init_crq_queue(adapter);
2197         if (rc) {
2198                 netdev_err(adapter->netdev,
2199                            "Couldn't initialize crq. rc=%d\n", rc);
2200                 return rc;
2201         }
2202
2203         rc = ibmvnic_reset_init(adapter, false);
2204         if (rc)
2205                 return rc;
2206
2207         /* If the adapter was in PROBE state prior to the reset,
2208          * exit here.
2209          */
2210         if (reset_state == VNIC_PROBED)
2211                 return 0;
2212
2213         rc = ibmvnic_login(netdev);
2214         if (rc) {
2215                 adapter->state = VNIC_PROBED;
2216                 return 0;
2217         }
2218
2219         rc = init_resources(adapter);
2220         if (rc)
2221                 return rc;
2222
2223         ibmvnic_disable_irqs(adapter);
2224         adapter->state = VNIC_CLOSED;
2225
2226         if (reset_state == VNIC_CLOSED)
2227                 return 0;
2228
2229         rc = __ibmvnic_open(netdev);
2230         if (rc)
2231                 return IBMVNIC_OPEN_FAILED;
2232
2233         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
2234         call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
2235
2236         return 0;
2237 }
2238
2239 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2240 {
2241         struct ibmvnic_rwi *rwi;
2242         unsigned long flags;
2243
2244         spin_lock_irqsave(&adapter->rwi_lock, flags);
2245
2246         if (!list_empty(&adapter->rwi_list)) {
2247                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2248                                        list);
2249                 list_del(&rwi->list);
2250         } else {
2251                 rwi = NULL;
2252         }
2253
2254         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2255         return rwi;
2256 }
2257
2258 static void free_all_rwi(struct ibmvnic_adapter *adapter)
2259 {
2260         struct ibmvnic_rwi *rwi;
2261
2262         rwi = get_next_rwi(adapter);
2263         while (rwi) {
2264                 kfree(rwi);
2265                 rwi = get_next_rwi(adapter);
2266         }
2267 }
2268
2269 static void __ibmvnic_reset(struct work_struct *work)
2270 {
2271         struct ibmvnic_rwi *rwi;
2272         struct ibmvnic_adapter *adapter;
2273         bool saved_state = false;
2274         unsigned long flags;
2275         u32 reset_state;
2276         int rc = 0;
2277
2278         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2279
2280         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2281                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2282                                       IBMVNIC_RESET_DELAY);
2283                 return;
2284         }
2285
2286         rwi = get_next_rwi(adapter);
2287         while (rwi) {
2288                 spin_lock_irqsave(&adapter->state_lock, flags);
2289
2290                 if (adapter->state == VNIC_REMOVING ||
2291                     adapter->state == VNIC_REMOVED) {
2292                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2293                         kfree(rwi);
2294                         rc = EBUSY;
2295                         break;
2296                 }
2297
2298                 if (!saved_state) {
2299                         reset_state = adapter->state;
2300                         saved_state = true;
2301                 }
2302                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2303
2304                 if (rwi->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2305                         /* CHANGE_PARAM requestor holds rtnl_lock */
2306                         rc = do_change_param_reset(adapter, rwi, reset_state);
2307                 } else if (adapter->force_reset_recovery) {
2308                         /*
2309                          * Since we are doing a hard reset now, clear the
2310                          * failover_pending flag so we don't ignore any
2311                          * future MOBILITY or other resets.
2312                          */
2313                         adapter->failover_pending = false;
2314
2315                         /* Transport event occurred during previous reset */
2316                         if (adapter->wait_for_reset) {
2317                                 /* Previous was CHANGE_PARAM; caller locked */
2318                                 adapter->force_reset_recovery = false;
2319                                 rc = do_hard_reset(adapter, rwi, reset_state);
2320                         } else {
2321                                 rtnl_lock();
2322                                 adapter->force_reset_recovery = false;
2323                                 rc = do_hard_reset(adapter, rwi, reset_state);
2324                                 rtnl_unlock();
2325                         }
2326                 } else if (!(rwi->reset_reason == VNIC_RESET_FATAL &&
2327                                 adapter->from_passive_init)) {
2328                         rc = do_reset(adapter, rwi, reset_state);
2329                 }
2330                 kfree(rwi);
2331                 if (rc == IBMVNIC_OPEN_FAILED) {
2332                         if (list_empty(&adapter->rwi_list))
2333                                 adapter->state = VNIC_CLOSED;
2334                         else
2335                                 adapter->state = reset_state;
2336                         rc = 0;
2337                 } else if (rc && rc != IBMVNIC_INIT_FAILED &&
2338                     !adapter->force_reset_recovery)
2339                         break;
2340
2341                 rwi = get_next_rwi(adapter);
2342
2343                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2344                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2345                         adapter->force_reset_recovery = true;
2346         }
2347
2348         if (adapter->wait_for_reset) {
2349                 adapter->reset_done_rc = rc;
2350                 complete(&adapter->reset_done);
2351         }
2352
2353         if (rc) {
2354                 netdev_dbg(adapter->netdev, "Reset failed\n");
2355                 free_all_rwi(adapter);
2356         }
2357
2358         clear_bit_unlock(0, &adapter->resetting);
2359 }
2360
2361 static void __ibmvnic_delayed_reset(struct work_struct *work)
2362 {
2363         struct ibmvnic_adapter *adapter;
2364
2365         adapter = container_of(work, struct ibmvnic_adapter,
2366                                ibmvnic_delayed_reset.work);
2367         __ibmvnic_reset(&adapter->ibmvnic_reset);
2368 }
2369
2370 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2371                          enum ibmvnic_reset_reason reason)
2372 {
2373         struct list_head *entry, *tmp_entry;
2374         struct ibmvnic_rwi *rwi, *tmp;
2375         struct net_device *netdev = adapter->netdev;
2376         unsigned long flags;
2377         int ret;
2378
2379         /*
2380          * If failover is pending don't schedule any other reset.
2381          * Instead let the failover complete. If there is already a
2382          * a failover reset scheduled, we will detect and drop the
2383          * duplicate reset when walking the ->rwi_list below.
2384          */
2385         if (adapter->state == VNIC_REMOVING ||
2386             adapter->state == VNIC_REMOVED ||
2387             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2388                 ret = EBUSY;
2389                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2390                 goto err;
2391         }
2392
2393         if (adapter->state == VNIC_PROBING) {
2394                 netdev_warn(netdev, "Adapter reset during probe\n");
2395                 ret = adapter->init_done_rc = EAGAIN;
2396                 goto err;
2397         }
2398
2399         spin_lock_irqsave(&adapter->rwi_lock, flags);
2400
2401         list_for_each(entry, &adapter->rwi_list) {
2402                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2403                 if (tmp->reset_reason == reason) {
2404                         netdev_dbg(netdev, "Skipping matching reset\n");
2405                         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2406                         ret = EBUSY;
2407                         goto err;
2408                 }
2409         }
2410
2411         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2412         if (!rwi) {
2413                 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2414                 ibmvnic_close(netdev);
2415                 ret = ENOMEM;
2416                 goto err;
2417         }
2418         /* if we just received a transport event,
2419          * flush reset queue and process this reset
2420          */
2421         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2422                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2423                         list_del(entry);
2424         }
2425         rwi->reset_reason = reason;
2426         list_add_tail(&rwi->list, &adapter->rwi_list);
2427         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2428         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2429         schedule_work(&adapter->ibmvnic_reset);
2430
2431         return 0;
2432 err:
2433         return -ret;
2434 }
2435
2436 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2437 {
2438         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2439
2440         if (test_bit(0, &adapter->resetting)) {
2441                 netdev_err(adapter->netdev,
2442                            "Adapter is resetting, skip timeout reset\n");
2443                 return;
2444         }
2445
2446         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2447 }
2448
2449 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2450                                   struct ibmvnic_rx_buff *rx_buff)
2451 {
2452         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2453
2454         rx_buff->skb = NULL;
2455
2456         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2457         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2458
2459         atomic_dec(&pool->available);
2460 }
2461
2462 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2463 {
2464         struct ibmvnic_sub_crq_queue *rx_scrq;
2465         struct ibmvnic_adapter *adapter;
2466         struct net_device *netdev;
2467         int frames_processed;
2468         int scrq_num;
2469
2470         netdev = napi->dev;
2471         adapter = netdev_priv(netdev);
2472         scrq_num = (int)(napi - adapter->napi);
2473         frames_processed = 0;
2474         rx_scrq = adapter->rx_scrq[scrq_num];
2475
2476 restart_poll:
2477         while (frames_processed < budget) {
2478                 struct sk_buff *skb;
2479                 struct ibmvnic_rx_buff *rx_buff;
2480                 union sub_crq *next;
2481                 u32 length;
2482                 u16 offset;
2483                 u8 flags = 0;
2484
2485                 if (unlikely(test_bit(0, &adapter->resetting) &&
2486                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2487                         enable_scrq_irq(adapter, rx_scrq);
2488                         napi_complete_done(napi, frames_processed);
2489                         return frames_processed;
2490                 }
2491
2492                 if (!pending_scrq(adapter, rx_scrq))
2493                         break;
2494                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2495                 rx_buff =
2496                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2497                                                           rx_comp.correlator);
2498                 /* do error checking */
2499                 if (next->rx_comp.rc) {
2500                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2501                                    be16_to_cpu(next->rx_comp.rc));
2502                         /* free the entry */
2503                         next->rx_comp.first = 0;
2504                         dev_kfree_skb_any(rx_buff->skb);
2505                         remove_buff_from_pool(adapter, rx_buff);
2506                         continue;
2507                 } else if (!rx_buff->skb) {
2508                         /* free the entry */
2509                         next->rx_comp.first = 0;
2510                         remove_buff_from_pool(adapter, rx_buff);
2511                         continue;
2512                 }
2513
2514                 length = be32_to_cpu(next->rx_comp.len);
2515                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2516                 flags = next->rx_comp.flags;
2517                 skb = rx_buff->skb;
2518                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2519                                         length);
2520
2521                 /* VLAN Header has been stripped by the system firmware and
2522                  * needs to be inserted by the driver
2523                  */
2524                 if (adapter->rx_vlan_header_insertion &&
2525                     (flags & IBMVNIC_VLAN_STRIPPED))
2526                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2527                                                ntohs(next->rx_comp.vlan_tci));
2528
2529                 /* free the entry */
2530                 next->rx_comp.first = 0;
2531                 remove_buff_from_pool(adapter, rx_buff);
2532
2533                 skb_put(skb, length);
2534                 skb->protocol = eth_type_trans(skb, netdev);
2535                 skb_record_rx_queue(skb, scrq_num);
2536
2537                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2538                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2539                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2540                 }
2541
2542                 length = skb->len;
2543                 napi_gro_receive(napi, skb); /* send it up */
2544                 netdev->stats.rx_packets++;
2545                 netdev->stats.rx_bytes += length;
2546                 adapter->rx_stats_buffers[scrq_num].packets++;
2547                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2548                 frames_processed++;
2549         }
2550
2551         if (adapter->state != VNIC_CLOSING &&
2552             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2553               adapter->req_rx_add_entries_per_subcrq / 2) ||
2554               frames_processed < budget))
2555                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2556         if (frames_processed < budget) {
2557                 if (napi_complete_done(napi, frames_processed)) {
2558                         enable_scrq_irq(adapter, rx_scrq);
2559                         if (pending_scrq(adapter, rx_scrq)) {
2560                                 rmb();
2561                                 if (napi_reschedule(napi)) {
2562                                         disable_scrq_irq(adapter, rx_scrq);
2563                                         goto restart_poll;
2564                                 }
2565                         }
2566                 }
2567         }
2568         return frames_processed;
2569 }
2570
2571 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2572 {
2573         int rc, ret;
2574
2575         adapter->fallback.mtu = adapter->req_mtu;
2576         adapter->fallback.rx_queues = adapter->req_rx_queues;
2577         adapter->fallback.tx_queues = adapter->req_tx_queues;
2578         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2579         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2580
2581         reinit_completion(&adapter->reset_done);
2582         adapter->wait_for_reset = true;
2583         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2584
2585         if (rc) {
2586                 ret = rc;
2587                 goto out;
2588         }
2589         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2590         if (rc) {
2591                 ret = -ENODEV;
2592                 goto out;
2593         }
2594
2595         ret = 0;
2596         if (adapter->reset_done_rc) {
2597                 ret = -EIO;
2598                 adapter->desired.mtu = adapter->fallback.mtu;
2599                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2600                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2601                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2602                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2603
2604                 reinit_completion(&adapter->reset_done);
2605                 adapter->wait_for_reset = true;
2606                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2607                 if (rc) {
2608                         ret = rc;
2609                         goto out;
2610                 }
2611                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2612                                                  60000);
2613                 if (rc) {
2614                         ret = -ENODEV;
2615                         goto out;
2616                 }
2617         }
2618 out:
2619         adapter->wait_for_reset = false;
2620
2621         return ret;
2622 }
2623
2624 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2625 {
2626         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2627
2628         adapter->desired.mtu = new_mtu + ETH_HLEN;
2629
2630         return wait_for_reset(adapter);
2631 }
2632
2633 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2634                                                 struct net_device *dev,
2635                                                 netdev_features_t features)
2636 {
2637         /* Some backing hardware adapters can not
2638          * handle packets with a MSS less than 224
2639          * or with only one segment.
2640          */
2641         if (skb_is_gso(skb)) {
2642                 if (skb_shinfo(skb)->gso_size < 224 ||
2643                     skb_shinfo(skb)->gso_segs == 1)
2644                         features &= ~NETIF_F_GSO_MASK;
2645         }
2646
2647         return features;
2648 }
2649
2650 static const struct net_device_ops ibmvnic_netdev_ops = {
2651         .ndo_open               = ibmvnic_open,
2652         .ndo_stop               = ibmvnic_close,
2653         .ndo_start_xmit         = ibmvnic_xmit,
2654         .ndo_set_rx_mode        = ibmvnic_set_multi,
2655         .ndo_set_mac_address    = ibmvnic_set_mac,
2656         .ndo_validate_addr      = eth_validate_addr,
2657         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2658         .ndo_change_mtu         = ibmvnic_change_mtu,
2659         .ndo_features_check     = ibmvnic_features_check,
2660 };
2661
2662 /* ethtool functions */
2663
2664 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2665                                       struct ethtool_link_ksettings *cmd)
2666 {
2667         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2668         int rc;
2669
2670         rc = send_query_phys_parms(adapter);
2671         if (rc) {
2672                 adapter->speed = SPEED_UNKNOWN;
2673                 adapter->duplex = DUPLEX_UNKNOWN;
2674         }
2675         cmd->base.speed = adapter->speed;
2676         cmd->base.duplex = adapter->duplex;
2677         cmd->base.port = PORT_FIBRE;
2678         cmd->base.phy_address = 0;
2679         cmd->base.autoneg = AUTONEG_ENABLE;
2680
2681         return 0;
2682 }
2683
2684 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2685                                 struct ethtool_drvinfo *info)
2686 {
2687         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2688
2689         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2690         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2691         strlcpy(info->fw_version, adapter->fw_version,
2692                 sizeof(info->fw_version));
2693 }
2694
2695 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2696 {
2697         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2698
2699         return adapter->msg_enable;
2700 }
2701
2702 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2703 {
2704         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2705
2706         adapter->msg_enable = data;
2707 }
2708
2709 static u32 ibmvnic_get_link(struct net_device *netdev)
2710 {
2711         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2712
2713         /* Don't need to send a query because we request a logical link up at
2714          * init and then we wait for link state indications
2715          */
2716         return adapter->logical_link_state;
2717 }
2718
2719 static void ibmvnic_get_ringparam(struct net_device *netdev,
2720                                   struct ethtool_ringparam *ring)
2721 {
2722         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2723
2724         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2725                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2726                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2727         } else {
2728                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2729                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2730         }
2731         ring->rx_mini_max_pending = 0;
2732         ring->rx_jumbo_max_pending = 0;
2733         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2734         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2735         ring->rx_mini_pending = 0;
2736         ring->rx_jumbo_pending = 0;
2737 }
2738
2739 static int ibmvnic_set_ringparam(struct net_device *netdev,
2740                                  struct ethtool_ringparam *ring)
2741 {
2742         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2743         int ret;
2744
2745         ret = 0;
2746         adapter->desired.rx_entries = ring->rx_pending;
2747         adapter->desired.tx_entries = ring->tx_pending;
2748
2749         ret = wait_for_reset(adapter);
2750
2751         if (!ret &&
2752             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2753              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2754                 netdev_info(netdev,
2755                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2756                             ring->rx_pending, ring->tx_pending,
2757                             adapter->req_rx_add_entries_per_subcrq,
2758                             adapter->req_tx_entries_per_subcrq);
2759         return ret;
2760 }
2761
2762 static void ibmvnic_get_channels(struct net_device *netdev,
2763                                  struct ethtool_channels *channels)
2764 {
2765         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2766
2767         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2768                 channels->max_rx = adapter->max_rx_queues;
2769                 channels->max_tx = adapter->max_tx_queues;
2770         } else {
2771                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2772                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2773         }
2774
2775         channels->max_other = 0;
2776         channels->max_combined = 0;
2777         channels->rx_count = adapter->req_rx_queues;
2778         channels->tx_count = adapter->req_tx_queues;
2779         channels->other_count = 0;
2780         channels->combined_count = 0;
2781 }
2782
2783 static int ibmvnic_set_channels(struct net_device *netdev,
2784                                 struct ethtool_channels *channels)
2785 {
2786         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2787         int ret;
2788
2789         ret = 0;
2790         adapter->desired.rx_queues = channels->rx_count;
2791         adapter->desired.tx_queues = channels->tx_count;
2792
2793         ret = wait_for_reset(adapter);
2794
2795         if (!ret &&
2796             (adapter->req_rx_queues != channels->rx_count ||
2797              adapter->req_tx_queues != channels->tx_count))
2798                 netdev_info(netdev,
2799                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2800                             channels->rx_count, channels->tx_count,
2801                             adapter->req_rx_queues, adapter->req_tx_queues);
2802         return ret;
2803
2804 }
2805
2806 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2807 {
2808         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2809         int i;
2810
2811         switch (stringset) {
2812         case ETH_SS_STATS:
2813                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2814                                 i++, data += ETH_GSTRING_LEN)
2815                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2816
2817                 for (i = 0; i < adapter->req_tx_queues; i++) {
2818                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2819                         data += ETH_GSTRING_LEN;
2820
2821                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2822                         data += ETH_GSTRING_LEN;
2823
2824                         snprintf(data, ETH_GSTRING_LEN,
2825                                  "tx%d_dropped_packets", i);
2826                         data += ETH_GSTRING_LEN;
2827                 }
2828
2829                 for (i = 0; i < adapter->req_rx_queues; i++) {
2830                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2831                         data += ETH_GSTRING_LEN;
2832
2833                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2834                         data += ETH_GSTRING_LEN;
2835
2836                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2837                         data += ETH_GSTRING_LEN;
2838                 }
2839                 break;
2840
2841         case ETH_SS_PRIV_FLAGS:
2842                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2843                         strcpy(data + i * ETH_GSTRING_LEN,
2844                                ibmvnic_priv_flags[i]);
2845                 break;
2846         default:
2847                 return;
2848         }
2849 }
2850
2851 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2852 {
2853         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2854
2855         switch (sset) {
2856         case ETH_SS_STATS:
2857                 return ARRAY_SIZE(ibmvnic_stats) +
2858                        adapter->req_tx_queues * NUM_TX_STATS +
2859                        adapter->req_rx_queues * NUM_RX_STATS;
2860         case ETH_SS_PRIV_FLAGS:
2861                 return ARRAY_SIZE(ibmvnic_priv_flags);
2862         default:
2863                 return -EOPNOTSUPP;
2864         }
2865 }
2866
2867 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2868                                       struct ethtool_stats *stats, u64 *data)
2869 {
2870         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2871         union ibmvnic_crq crq;
2872         int i, j;
2873         int rc;
2874
2875         memset(&crq, 0, sizeof(crq));
2876         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2877         crq.request_statistics.cmd = REQUEST_STATISTICS;
2878         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2879         crq.request_statistics.len =
2880             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2881
2882         /* Wait for data to be written */
2883         reinit_completion(&adapter->stats_done);
2884         rc = ibmvnic_send_crq(adapter, &crq);
2885         if (rc)
2886                 return;
2887         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2888         if (rc)
2889                 return;
2890
2891         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2892                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2893                                                 ibmvnic_stats[i].offset));
2894
2895         for (j = 0; j < adapter->req_tx_queues; j++) {
2896                 data[i] = adapter->tx_stats_buffers[j].packets;
2897                 i++;
2898                 data[i] = adapter->tx_stats_buffers[j].bytes;
2899                 i++;
2900                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2901                 i++;
2902         }
2903
2904         for (j = 0; j < adapter->req_rx_queues; j++) {
2905                 data[i] = adapter->rx_stats_buffers[j].packets;
2906                 i++;
2907                 data[i] = adapter->rx_stats_buffers[j].bytes;
2908                 i++;
2909                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2910                 i++;
2911         }
2912 }
2913
2914 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2915 {
2916         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2917
2918         return adapter->priv_flags;
2919 }
2920
2921 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2922 {
2923         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2924         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2925
2926         if (which_maxes)
2927                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2928         else
2929                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2930
2931         return 0;
2932 }
2933 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2934         .get_drvinfo            = ibmvnic_get_drvinfo,
2935         .get_msglevel           = ibmvnic_get_msglevel,
2936         .set_msglevel           = ibmvnic_set_msglevel,
2937         .get_link               = ibmvnic_get_link,
2938         .get_ringparam          = ibmvnic_get_ringparam,
2939         .set_ringparam          = ibmvnic_set_ringparam,
2940         .get_channels           = ibmvnic_get_channels,
2941         .set_channels           = ibmvnic_set_channels,
2942         .get_strings            = ibmvnic_get_strings,
2943         .get_sset_count         = ibmvnic_get_sset_count,
2944         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2945         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2946         .get_priv_flags         = ibmvnic_get_priv_flags,
2947         .set_priv_flags         = ibmvnic_set_priv_flags,
2948 };
2949
2950 /* Routines for managing CRQs/sCRQs  */
2951
2952 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2953                                    struct ibmvnic_sub_crq_queue *scrq)
2954 {
2955         int rc;
2956
2957         if (scrq->irq) {
2958                 free_irq(scrq->irq, scrq);
2959                 irq_dispose_mapping(scrq->irq);
2960                 scrq->irq = 0;
2961         }
2962
2963         memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2964         atomic_set(&scrq->used, 0);
2965         scrq->cur = 0;
2966         scrq->ind_buf.index = 0;
2967
2968         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2969                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2970         return rc;
2971 }
2972
2973 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2974 {
2975         int i, rc;
2976
2977         if (!adapter->tx_scrq || !adapter->rx_scrq)
2978                 return -EINVAL;
2979
2980         for (i = 0; i < adapter->req_tx_queues; i++) {
2981                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2982                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2983                 if (rc)
2984                         return rc;
2985         }
2986
2987         for (i = 0; i < adapter->req_rx_queues; i++) {
2988                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2989                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2990                 if (rc)
2991                         return rc;
2992         }
2993
2994         return rc;
2995 }
2996
2997 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2998                                   struct ibmvnic_sub_crq_queue *scrq,
2999                                   bool do_h_free)
3000 {
3001         struct device *dev = &adapter->vdev->dev;
3002         long rc;
3003
3004         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3005
3006         if (do_h_free) {
3007                 /* Close the sub-crqs */
3008                 do {
3009                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3010                                                 adapter->vdev->unit_address,
3011                                                 scrq->crq_num);
3012                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3013
3014                 if (rc) {
3015                         netdev_err(adapter->netdev,
3016                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
3017                                    scrq->crq_num, rc);
3018                 }
3019         }
3020
3021         dma_free_coherent(dev,
3022                           IBMVNIC_IND_ARR_SZ,
3023                           scrq->ind_buf.indir_arr,
3024                           scrq->ind_buf.indir_dma);
3025
3026         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3027                          DMA_BIDIRECTIONAL);
3028         free_pages((unsigned long)scrq->msgs, 2);
3029         kfree(scrq);
3030 }
3031
3032 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3033                                                         *adapter)
3034 {
3035         struct device *dev = &adapter->vdev->dev;
3036         struct ibmvnic_sub_crq_queue *scrq;
3037         int rc;
3038
3039         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3040         if (!scrq)
3041                 return NULL;
3042
3043         scrq->msgs =
3044                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3045         if (!scrq->msgs) {
3046                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3047                 goto zero_page_failed;
3048         }
3049
3050         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3051                                          DMA_BIDIRECTIONAL);
3052         if (dma_mapping_error(dev, scrq->msg_token)) {
3053                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3054                 goto map_failed;
3055         }
3056
3057         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3058                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3059
3060         if (rc == H_RESOURCE)
3061                 rc = ibmvnic_reset_crq(adapter);
3062
3063         if (rc == H_CLOSED) {
3064                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3065         } else if (rc) {
3066                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3067                 goto reg_failed;
3068         }
3069
3070         scrq->adapter = adapter;
3071         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3072         scrq->ind_buf.index = 0;
3073
3074         scrq->ind_buf.indir_arr =
3075                 dma_alloc_coherent(dev,
3076                                    IBMVNIC_IND_ARR_SZ,
3077                                    &scrq->ind_buf.indir_dma,
3078                                    GFP_KERNEL);
3079
3080         if (!scrq->ind_buf.indir_arr)
3081                 goto indir_failed;
3082
3083         spin_lock_init(&scrq->lock);
3084
3085         netdev_dbg(adapter->netdev,
3086                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3087                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3088
3089         return scrq;
3090
3091 indir_failed:
3092         do {
3093                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3094                                         adapter->vdev->unit_address,
3095                                         scrq->crq_num);
3096         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3097 reg_failed:
3098         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3099                          DMA_BIDIRECTIONAL);
3100 map_failed:
3101         free_pages((unsigned long)scrq->msgs, 2);
3102 zero_page_failed:
3103         kfree(scrq);
3104
3105         return NULL;
3106 }
3107
3108 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3109 {
3110         int i;
3111
3112         if (adapter->tx_scrq) {
3113                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3114                         if (!adapter->tx_scrq[i])
3115                                 continue;
3116
3117                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3118                                    i);
3119                         if (adapter->tx_scrq[i]->irq) {
3120                                 free_irq(adapter->tx_scrq[i]->irq,
3121                                          adapter->tx_scrq[i]);
3122                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3123                                 adapter->tx_scrq[i]->irq = 0;
3124                         }
3125
3126                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3127                                               do_h_free);
3128                 }
3129
3130                 kfree(adapter->tx_scrq);
3131                 adapter->tx_scrq = NULL;
3132                 adapter->num_active_tx_scrqs = 0;
3133         }
3134
3135         if (adapter->rx_scrq) {
3136                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3137                         if (!adapter->rx_scrq[i])
3138                                 continue;
3139
3140                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3141                                    i);
3142                         if (adapter->rx_scrq[i]->irq) {
3143                                 free_irq(adapter->rx_scrq[i]->irq,
3144                                          adapter->rx_scrq[i]);
3145                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3146                                 adapter->rx_scrq[i]->irq = 0;
3147                         }
3148
3149                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3150                                               do_h_free);
3151                 }
3152
3153                 kfree(adapter->rx_scrq);
3154                 adapter->rx_scrq = NULL;
3155                 adapter->num_active_rx_scrqs = 0;
3156         }
3157 }
3158
3159 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3160                             struct ibmvnic_sub_crq_queue *scrq)
3161 {
3162         struct device *dev = &adapter->vdev->dev;
3163         unsigned long rc;
3164
3165         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3166                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3167         if (rc)
3168                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3169                         scrq->hw_irq, rc);
3170         return rc;
3171 }
3172
3173 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3174                            struct ibmvnic_sub_crq_queue *scrq)
3175 {
3176         struct device *dev = &adapter->vdev->dev;
3177         unsigned long rc;
3178
3179         if (scrq->hw_irq > 0x100000000ULL) {
3180                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3181                 return 1;
3182         }
3183
3184         if (test_bit(0, &adapter->resetting) &&
3185             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3186                 u64 val = (0xff000000) | scrq->hw_irq;
3187
3188                 rc = plpar_hcall_norets(H_EOI, val);
3189                 /* H_EOI would fail with rc = H_FUNCTION when running
3190                  * in XIVE mode which is expected, but not an error.
3191                  */
3192                 if (rc && (rc != H_FUNCTION))
3193                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3194                                 val, rc);
3195         }
3196
3197         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3198                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3199         if (rc)
3200                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3201                         scrq->hw_irq, rc);
3202         return rc;
3203 }
3204
3205 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3206                                struct ibmvnic_sub_crq_queue *scrq)
3207 {
3208         struct device *dev = &adapter->vdev->dev;
3209         struct ibmvnic_tx_pool *tx_pool;
3210         struct ibmvnic_tx_buff *txbuff;
3211         struct netdev_queue *txq;
3212         union sub_crq *next;
3213         int index;
3214         int i;
3215
3216 restart_loop:
3217         while (pending_scrq(adapter, scrq)) {
3218                 unsigned int pool = scrq->pool_index;
3219                 int num_entries = 0;
3220                 int total_bytes = 0;
3221                 int num_packets = 0;
3222
3223                 next = ibmvnic_next_scrq(adapter, scrq);
3224                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3225                         if (next->tx_comp.rcs[i]) {
3226                                 dev_err(dev, "tx error %x\n",
3227                                         next->tx_comp.rcs[i]);
3228                                 continue;
3229                         }
3230                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3231                         if (index & IBMVNIC_TSO_POOL_MASK) {
3232                                 tx_pool = &adapter->tso_pool[pool];
3233                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3234                         } else {
3235                                 tx_pool = &adapter->tx_pool[pool];
3236                         }
3237
3238                         txbuff = &tx_pool->tx_buff[index];
3239                         num_packets++;
3240                         num_entries += txbuff->num_entries;
3241                         if (txbuff->skb) {
3242                                 total_bytes += txbuff->skb->len;
3243                                 dev_consume_skb_irq(txbuff->skb);
3244                                 txbuff->skb = NULL;
3245                         } else {
3246                                 netdev_warn(adapter->netdev,
3247                                             "TX completion received with NULL socket buffer\n");
3248                         }
3249                         tx_pool->free_map[tx_pool->producer_index] = index;
3250                         tx_pool->producer_index =
3251                                 (tx_pool->producer_index + 1) %
3252                                         tx_pool->num_buffers;
3253                 }
3254                 /* remove tx_comp scrq*/
3255                 next->tx_comp.first = 0;
3256
3257                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3258                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3259
3260                 if (atomic_sub_return(num_entries, &scrq->used) <=
3261                     (adapter->req_tx_entries_per_subcrq / 2) &&
3262                     __netif_subqueue_stopped(adapter->netdev,
3263                                              scrq->pool_index)) {
3264                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3265                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3266                                    scrq->pool_index);
3267                 }
3268         }
3269
3270         enable_scrq_irq(adapter, scrq);
3271
3272         if (pending_scrq(adapter, scrq)) {
3273                 disable_scrq_irq(adapter, scrq);
3274                 goto restart_loop;
3275         }
3276
3277         return 0;
3278 }
3279
3280 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3281 {
3282         struct ibmvnic_sub_crq_queue *scrq = instance;
3283         struct ibmvnic_adapter *adapter = scrq->adapter;
3284
3285         disable_scrq_irq(adapter, scrq);
3286         ibmvnic_complete_tx(adapter, scrq);
3287
3288         return IRQ_HANDLED;
3289 }
3290
3291 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3292 {
3293         struct ibmvnic_sub_crq_queue *scrq = instance;
3294         struct ibmvnic_adapter *adapter = scrq->adapter;
3295
3296         /* When booting a kdump kernel we can hit pending interrupts
3297          * prior to completing driver initialization.
3298          */
3299         if (unlikely(adapter->state != VNIC_OPEN))
3300                 return IRQ_NONE;
3301
3302         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3303
3304         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3305                 disable_scrq_irq(adapter, scrq);
3306                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3307         }
3308
3309         return IRQ_HANDLED;
3310 }
3311
3312 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3313 {
3314         struct device *dev = &adapter->vdev->dev;
3315         struct ibmvnic_sub_crq_queue *scrq;
3316         int i = 0, j = 0;
3317         int rc = 0;
3318
3319         for (i = 0; i < adapter->req_tx_queues; i++) {
3320                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3321                            i);
3322                 scrq = adapter->tx_scrq[i];
3323                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3324
3325                 if (!scrq->irq) {
3326                         rc = -EINVAL;
3327                         dev_err(dev, "Error mapping irq\n");
3328                         goto req_tx_irq_failed;
3329                 }
3330
3331                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3332                          adapter->vdev->unit_address, i);
3333                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3334                                  0, scrq->name, scrq);
3335
3336                 if (rc) {
3337                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3338                                 scrq->irq, rc);
3339                         irq_dispose_mapping(scrq->irq);
3340                         goto req_tx_irq_failed;
3341                 }
3342         }
3343
3344         for (i = 0; i < adapter->req_rx_queues; i++) {
3345                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3346                            i);
3347                 scrq = adapter->rx_scrq[i];
3348                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3349                 if (!scrq->irq) {
3350                         rc = -EINVAL;
3351                         dev_err(dev, "Error mapping irq\n");
3352                         goto req_rx_irq_failed;
3353                 }
3354                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3355                          adapter->vdev->unit_address, i);
3356                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3357                                  0, scrq->name, scrq);
3358                 if (rc) {
3359                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3360                                 scrq->irq, rc);
3361                         irq_dispose_mapping(scrq->irq);
3362                         goto req_rx_irq_failed;
3363                 }
3364         }
3365         return rc;
3366
3367 req_rx_irq_failed:
3368         for (j = 0; j < i; j++) {
3369                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3370                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3371         }
3372         i = adapter->req_tx_queues;
3373 req_tx_irq_failed:
3374         for (j = 0; j < i; j++) {
3375                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3376                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3377         }
3378         release_sub_crqs(adapter, 1);
3379         return rc;
3380 }
3381
3382 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3383 {
3384         struct device *dev = &adapter->vdev->dev;
3385         struct ibmvnic_sub_crq_queue **allqueues;
3386         int registered_queues = 0;
3387         int total_queues;
3388         int more = 0;
3389         int i;
3390
3391         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3392
3393         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3394         if (!allqueues)
3395                 return -1;
3396
3397         for (i = 0; i < total_queues; i++) {
3398                 allqueues[i] = init_sub_crq_queue(adapter);
3399                 if (!allqueues[i]) {
3400                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3401                         break;
3402                 }
3403                 registered_queues++;
3404         }
3405
3406         /* Make sure we were able to register the minimum number of queues */
3407         if (registered_queues <
3408             adapter->min_tx_queues + adapter->min_rx_queues) {
3409                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3410                 goto tx_failed;
3411         }
3412
3413         /* Distribute the failed allocated queues*/
3414         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3415                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3416                 switch (i % 3) {
3417                 case 0:
3418                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3419                                 adapter->req_rx_queues--;
3420                         else
3421                                 more++;
3422                         break;
3423                 case 1:
3424                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3425                                 adapter->req_tx_queues--;
3426                         else
3427                                 more++;
3428                         break;
3429                 }
3430         }
3431
3432         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3433                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3434         if (!adapter->tx_scrq)
3435                 goto tx_failed;
3436
3437         for (i = 0; i < adapter->req_tx_queues; i++) {
3438                 adapter->tx_scrq[i] = allqueues[i];
3439                 adapter->tx_scrq[i]->pool_index = i;
3440                 adapter->num_active_tx_scrqs++;
3441         }
3442
3443         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3444                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3445         if (!adapter->rx_scrq)
3446                 goto rx_failed;
3447
3448         for (i = 0; i < adapter->req_rx_queues; i++) {
3449                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3450                 adapter->rx_scrq[i]->scrq_num = i;
3451                 adapter->num_active_rx_scrqs++;
3452         }
3453
3454         kfree(allqueues);
3455         return 0;
3456
3457 rx_failed:
3458         kfree(adapter->tx_scrq);
3459         adapter->tx_scrq = NULL;
3460 tx_failed:
3461         for (i = 0; i < registered_queues; i++)
3462                 release_sub_crq_queue(adapter, allqueues[i], 1);
3463         kfree(allqueues);
3464         return -1;
3465 }
3466
3467 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3468 {
3469         struct device *dev = &adapter->vdev->dev;
3470         union ibmvnic_crq crq;
3471         int max_entries;
3472
3473         if (!retry) {
3474                 /* Sub-CRQ entries are 32 byte long */
3475                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3476
3477                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3478                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3479                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3480                         return;
3481                 }
3482
3483                 if (adapter->desired.mtu)
3484                         adapter->req_mtu = adapter->desired.mtu;
3485                 else
3486                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3487
3488                 if (!adapter->desired.tx_entries)
3489                         adapter->desired.tx_entries =
3490                                         adapter->max_tx_entries_per_subcrq;
3491                 if (!adapter->desired.rx_entries)
3492                         adapter->desired.rx_entries =
3493                                         adapter->max_rx_add_entries_per_subcrq;
3494
3495                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3496                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3497
3498                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3499                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3500                         adapter->desired.tx_entries = max_entries;
3501                 }
3502
3503                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3504                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3505                         adapter->desired.rx_entries = max_entries;
3506                 }
3507
3508                 if (adapter->desired.tx_entries)
3509                         adapter->req_tx_entries_per_subcrq =
3510                                         adapter->desired.tx_entries;
3511                 else
3512                         adapter->req_tx_entries_per_subcrq =
3513                                         adapter->max_tx_entries_per_subcrq;
3514
3515                 if (adapter->desired.rx_entries)
3516                         adapter->req_rx_add_entries_per_subcrq =
3517                                         adapter->desired.rx_entries;
3518                 else
3519                         adapter->req_rx_add_entries_per_subcrq =
3520                                         adapter->max_rx_add_entries_per_subcrq;
3521
3522                 if (adapter->desired.tx_queues)
3523                         adapter->req_tx_queues =
3524                                         adapter->desired.tx_queues;
3525                 else
3526                         adapter->req_tx_queues =
3527                                         adapter->opt_tx_comp_sub_queues;
3528
3529                 if (adapter->desired.rx_queues)
3530                         adapter->req_rx_queues =
3531                                         adapter->desired.rx_queues;
3532                 else
3533                         adapter->req_rx_queues =
3534                                         adapter->opt_rx_comp_queues;
3535
3536                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3537         }
3538
3539         memset(&crq, 0, sizeof(crq));
3540         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3541         crq.request_capability.cmd = REQUEST_CAPABILITY;
3542
3543         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3544         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3545         atomic_inc(&adapter->running_cap_crqs);
3546         ibmvnic_send_crq(adapter, &crq);
3547
3548         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3549         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3550         atomic_inc(&adapter->running_cap_crqs);
3551         ibmvnic_send_crq(adapter, &crq);
3552
3553         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3554         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3555         atomic_inc(&adapter->running_cap_crqs);
3556         ibmvnic_send_crq(adapter, &crq);
3557
3558         crq.request_capability.capability =
3559             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3560         crq.request_capability.number =
3561             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3562         atomic_inc(&adapter->running_cap_crqs);
3563         ibmvnic_send_crq(adapter, &crq);
3564
3565         crq.request_capability.capability =
3566             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3567         crq.request_capability.number =
3568             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3569         atomic_inc(&adapter->running_cap_crqs);
3570         ibmvnic_send_crq(adapter, &crq);
3571
3572         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3573         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3574         atomic_inc(&adapter->running_cap_crqs);
3575         ibmvnic_send_crq(adapter, &crq);
3576
3577         if (adapter->netdev->flags & IFF_PROMISC) {
3578                 if (adapter->promisc_supported) {
3579                         crq.request_capability.capability =
3580                             cpu_to_be16(PROMISC_REQUESTED);
3581                         crq.request_capability.number = cpu_to_be64(1);
3582                         atomic_inc(&adapter->running_cap_crqs);
3583                         ibmvnic_send_crq(adapter, &crq);
3584                 }
3585         } else {
3586                 crq.request_capability.capability =
3587                     cpu_to_be16(PROMISC_REQUESTED);
3588                 crq.request_capability.number = cpu_to_be64(0);
3589                 atomic_inc(&adapter->running_cap_crqs);
3590                 ibmvnic_send_crq(adapter, &crq);
3591         }
3592 }
3593
3594 static int pending_scrq(struct ibmvnic_adapter *adapter,
3595                         struct ibmvnic_sub_crq_queue *scrq)
3596 {
3597         union sub_crq *entry = &scrq->msgs[scrq->cur];
3598
3599         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3600                 return 1;
3601         else
3602                 return 0;
3603 }
3604
3605 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3606                                         struct ibmvnic_sub_crq_queue *scrq)
3607 {
3608         union sub_crq *entry;
3609         unsigned long flags;
3610
3611         spin_lock_irqsave(&scrq->lock, flags);
3612         entry = &scrq->msgs[scrq->cur];
3613         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3614                 if (++scrq->cur == scrq->size)
3615                         scrq->cur = 0;
3616         } else {
3617                 entry = NULL;
3618         }
3619         spin_unlock_irqrestore(&scrq->lock, flags);
3620
3621         return entry;
3622 }
3623
3624 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3625 {
3626         struct ibmvnic_crq_queue *queue = &adapter->crq;
3627         union ibmvnic_crq *crq;
3628
3629         crq = &queue->msgs[queue->cur];
3630         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3631                 if (++queue->cur == queue->size)
3632                         queue->cur = 0;
3633         } else {
3634                 crq = NULL;
3635         }
3636
3637         return crq;
3638 }
3639
3640 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3641 {
3642         switch (rc) {
3643         case H_PARAMETER:
3644                 dev_warn_ratelimited(dev,
3645                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3646                                      func, rc);
3647                 break;
3648         case H_CLOSED:
3649                 dev_warn_ratelimited(dev,
3650                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3651                                      func, rc);
3652                 break;
3653         default:
3654                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3655                 break;
3656         }
3657 }
3658
3659 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3660                                 u64 remote_handle, u64 ioba, u64 num_entries)
3661 {
3662         unsigned int ua = adapter->vdev->unit_address;
3663         struct device *dev = &adapter->vdev->dev;
3664         int rc;
3665
3666         /* Make sure the hypervisor sees the complete request */
3667         mb();
3668         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3669                                 cpu_to_be64(remote_handle),
3670                                 ioba, num_entries);
3671
3672         if (rc)
3673                 print_subcrq_error(dev, rc, __func__);
3674
3675         return rc;
3676 }
3677
3678 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3679                             union ibmvnic_crq *crq)
3680 {
3681         unsigned int ua = adapter->vdev->unit_address;
3682         struct device *dev = &adapter->vdev->dev;
3683         u64 *u64_crq = (u64 *)crq;
3684         int rc;
3685
3686         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3687                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3688                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3689
3690         if (!adapter->crq.active &&
3691             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3692                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3693                 return -EINVAL;
3694         }
3695
3696         /* Make sure the hypervisor sees the complete request */
3697         mb();
3698
3699         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3700                                 cpu_to_be64(u64_crq[0]),
3701                                 cpu_to_be64(u64_crq[1]));
3702
3703         if (rc) {
3704                 if (rc == H_CLOSED) {
3705                         dev_warn(dev, "CRQ Queue closed\n");
3706                         /* do not reset, report the fail, wait for passive init from server */
3707                 }
3708
3709                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3710         }
3711
3712         return rc;
3713 }
3714
3715 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3716 {
3717         struct device *dev = &adapter->vdev->dev;
3718         union ibmvnic_crq crq;
3719         int retries = 100;
3720         int rc;
3721
3722         memset(&crq, 0, sizeof(crq));
3723         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3724         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3725         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3726
3727         do {
3728                 rc = ibmvnic_send_crq(adapter, &crq);
3729                 if (rc != H_CLOSED)
3730                         break;
3731                 retries--;
3732                 msleep(50);
3733
3734         } while (retries > 0);
3735
3736         if (rc) {
3737                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3738                 return rc;
3739         }
3740
3741         return 0;
3742 }
3743
3744 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3745 {
3746         union ibmvnic_crq crq;
3747
3748         memset(&crq, 0, sizeof(crq));
3749         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3750         crq.version_exchange.cmd = VERSION_EXCHANGE;
3751         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3752
3753         return ibmvnic_send_crq(adapter, &crq);
3754 }
3755
3756 struct vnic_login_client_data {
3757         u8      type;
3758         __be16  len;
3759         char    name[];
3760 } __packed;
3761
3762 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3763 {
3764         int len;
3765
3766         /* Calculate the amount of buffer space needed for the
3767          * vnic client data in the login buffer. There are four entries,
3768          * OS name, LPAR name, device name, and a null last entry.
3769          */
3770         len = 4 * sizeof(struct vnic_login_client_data);
3771         len += 6; /* "Linux" plus NULL */
3772         len += strlen(utsname()->nodename) + 1;
3773         len += strlen(adapter->netdev->name) + 1;
3774
3775         return len;
3776 }
3777
3778 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3779                                  struct vnic_login_client_data *vlcd)
3780 {
3781         const char *os_name = "Linux";
3782         int len;
3783
3784         /* Type 1 - LPAR OS */
3785         vlcd->type = 1;
3786         len = strlen(os_name) + 1;
3787         vlcd->len = cpu_to_be16(len);
3788         strncpy(vlcd->name, os_name, len);
3789         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3790
3791         /* Type 2 - LPAR name */
3792         vlcd->type = 2;
3793         len = strlen(utsname()->nodename) + 1;
3794         vlcd->len = cpu_to_be16(len);
3795         strncpy(vlcd->name, utsname()->nodename, len);
3796         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3797
3798         /* Type 3 - device name */
3799         vlcd->type = 3;
3800         len = strlen(adapter->netdev->name) + 1;
3801         vlcd->len = cpu_to_be16(len);
3802         strncpy(vlcd->name, adapter->netdev->name, len);
3803 }
3804
3805 static int send_login(struct ibmvnic_adapter *adapter)
3806 {
3807         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3808         struct ibmvnic_login_buffer *login_buffer;
3809         struct device *dev = &adapter->vdev->dev;
3810         dma_addr_t rsp_buffer_token;
3811         dma_addr_t buffer_token;
3812         size_t rsp_buffer_size;
3813         union ibmvnic_crq crq;
3814         size_t buffer_size;
3815         __be64 *tx_list_p;
3816         __be64 *rx_list_p;
3817         int client_data_len;
3818         struct vnic_login_client_data *vlcd;
3819         int i;
3820
3821         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3822                 netdev_err(adapter->netdev,
3823                            "RX or TX queues are not allocated, device login failed\n");
3824                 return -1;
3825         }
3826
3827         release_login_rsp_buffer(adapter);
3828         client_data_len = vnic_client_data_len(adapter);
3829
3830         buffer_size =
3831             sizeof(struct ibmvnic_login_buffer) +
3832             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3833             client_data_len;
3834
3835         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3836         if (!login_buffer)
3837                 goto buf_alloc_failed;
3838
3839         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3840                                       DMA_TO_DEVICE);
3841         if (dma_mapping_error(dev, buffer_token)) {
3842                 dev_err(dev, "Couldn't map login buffer\n");
3843                 goto buf_map_failed;
3844         }
3845
3846         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3847                           sizeof(u64) * adapter->req_tx_queues +
3848                           sizeof(u64) * adapter->req_rx_queues +
3849                           sizeof(u64) * adapter->req_rx_queues +
3850                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3851
3852         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3853         if (!login_rsp_buffer)
3854                 goto buf_rsp_alloc_failed;
3855
3856         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3857                                           rsp_buffer_size, DMA_FROM_DEVICE);
3858         if (dma_mapping_error(dev, rsp_buffer_token)) {
3859                 dev_err(dev, "Couldn't map login rsp buffer\n");
3860                 goto buf_rsp_map_failed;
3861         }
3862
3863         adapter->login_buf = login_buffer;
3864         adapter->login_buf_token = buffer_token;
3865         adapter->login_buf_sz = buffer_size;
3866         adapter->login_rsp_buf = login_rsp_buffer;
3867         adapter->login_rsp_buf_token = rsp_buffer_token;
3868         adapter->login_rsp_buf_sz = rsp_buffer_size;
3869
3870         login_buffer->len = cpu_to_be32(buffer_size);
3871         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3872         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3873         login_buffer->off_txcomp_subcrqs =
3874             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3875         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3876         login_buffer->off_rxcomp_subcrqs =
3877             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3878                         sizeof(u64) * adapter->req_tx_queues);
3879         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3880         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3881
3882         tx_list_p = (__be64 *)((char *)login_buffer +
3883                                       sizeof(struct ibmvnic_login_buffer));
3884         rx_list_p = (__be64 *)((char *)login_buffer +
3885                                       sizeof(struct ibmvnic_login_buffer) +
3886                                       sizeof(u64) * adapter->req_tx_queues);
3887
3888         for (i = 0; i < adapter->req_tx_queues; i++) {
3889                 if (adapter->tx_scrq[i]) {
3890                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3891                                                    crq_num);
3892                 }
3893         }
3894
3895         for (i = 0; i < adapter->req_rx_queues; i++) {
3896                 if (adapter->rx_scrq[i]) {
3897                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3898                                                    crq_num);
3899                 }
3900         }
3901
3902         /* Insert vNIC login client data */
3903         vlcd = (struct vnic_login_client_data *)
3904                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3905         login_buffer->client_data_offset =
3906                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3907         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3908
3909         vnic_add_client_data(adapter, vlcd);
3910
3911         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3912         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3913                 netdev_dbg(adapter->netdev, "%016lx\n",
3914                            ((unsigned long int *)(adapter->login_buf))[i]);
3915         }
3916
3917         memset(&crq, 0, sizeof(crq));
3918         crq.login.first = IBMVNIC_CRQ_CMD;
3919         crq.login.cmd = LOGIN;
3920         crq.login.ioba = cpu_to_be32(buffer_token);
3921         crq.login.len = cpu_to_be32(buffer_size);
3922         ibmvnic_send_crq(adapter, &crq);
3923
3924         return 0;
3925
3926 buf_rsp_map_failed:
3927         kfree(login_rsp_buffer);
3928 buf_rsp_alloc_failed:
3929         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3930 buf_map_failed:
3931         kfree(login_buffer);
3932 buf_alloc_failed:
3933         return -1;
3934 }
3935
3936 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3937                             u32 len, u8 map_id)
3938 {
3939         union ibmvnic_crq crq;
3940
3941         memset(&crq, 0, sizeof(crq));
3942         crq.request_map.first = IBMVNIC_CRQ_CMD;
3943         crq.request_map.cmd = REQUEST_MAP;
3944         crq.request_map.map_id = map_id;
3945         crq.request_map.ioba = cpu_to_be32(addr);
3946         crq.request_map.len = cpu_to_be32(len);
3947         return ibmvnic_send_crq(adapter, &crq);
3948 }
3949
3950 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3951 {
3952         union ibmvnic_crq crq;
3953
3954         memset(&crq, 0, sizeof(crq));
3955         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3956         crq.request_unmap.cmd = REQUEST_UNMAP;
3957         crq.request_unmap.map_id = map_id;
3958         return ibmvnic_send_crq(adapter, &crq);
3959 }
3960
3961 static void send_query_map(struct ibmvnic_adapter *adapter)
3962 {
3963         union ibmvnic_crq crq;
3964
3965         memset(&crq, 0, sizeof(crq));
3966         crq.query_map.first = IBMVNIC_CRQ_CMD;
3967         crq.query_map.cmd = QUERY_MAP;
3968         ibmvnic_send_crq(adapter, &crq);
3969 }
3970
3971 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3972 static void send_query_cap(struct ibmvnic_adapter *adapter)
3973 {
3974         union ibmvnic_crq crq;
3975
3976         atomic_set(&adapter->running_cap_crqs, 0);
3977         memset(&crq, 0, sizeof(crq));
3978         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3979         crq.query_capability.cmd = QUERY_CAPABILITY;
3980
3981         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3982         atomic_inc(&adapter->running_cap_crqs);
3983         ibmvnic_send_crq(adapter, &crq);
3984
3985         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3986         atomic_inc(&adapter->running_cap_crqs);
3987         ibmvnic_send_crq(adapter, &crq);
3988
3989         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3990         atomic_inc(&adapter->running_cap_crqs);
3991         ibmvnic_send_crq(adapter, &crq);
3992
3993         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3994         atomic_inc(&adapter->running_cap_crqs);
3995         ibmvnic_send_crq(adapter, &crq);
3996
3997         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3998         atomic_inc(&adapter->running_cap_crqs);
3999         ibmvnic_send_crq(adapter, &crq);
4000
4001         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4002         atomic_inc(&adapter->running_cap_crqs);
4003         ibmvnic_send_crq(adapter, &crq);
4004
4005         crq.query_capability.capability =
4006             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4007         atomic_inc(&adapter->running_cap_crqs);
4008         ibmvnic_send_crq(adapter, &crq);
4009
4010         crq.query_capability.capability =
4011             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4012         atomic_inc(&adapter->running_cap_crqs);
4013         ibmvnic_send_crq(adapter, &crq);
4014
4015         crq.query_capability.capability =
4016             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4017         atomic_inc(&adapter->running_cap_crqs);
4018         ibmvnic_send_crq(adapter, &crq);
4019
4020         crq.query_capability.capability =
4021             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4022         atomic_inc(&adapter->running_cap_crqs);
4023         ibmvnic_send_crq(adapter, &crq);
4024
4025         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4026         atomic_inc(&adapter->running_cap_crqs);
4027         ibmvnic_send_crq(adapter, &crq);
4028
4029         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4030         atomic_inc(&adapter->running_cap_crqs);
4031         ibmvnic_send_crq(adapter, &crq);
4032
4033         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4034         atomic_inc(&adapter->running_cap_crqs);
4035         ibmvnic_send_crq(adapter, &crq);
4036
4037         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4038         atomic_inc(&adapter->running_cap_crqs);
4039         ibmvnic_send_crq(adapter, &crq);
4040
4041         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4042         atomic_inc(&adapter->running_cap_crqs);
4043         ibmvnic_send_crq(adapter, &crq);
4044
4045         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4046         atomic_inc(&adapter->running_cap_crqs);
4047         ibmvnic_send_crq(adapter, &crq);
4048
4049         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4050         atomic_inc(&adapter->running_cap_crqs);
4051         ibmvnic_send_crq(adapter, &crq);
4052
4053         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4054         atomic_inc(&adapter->running_cap_crqs);
4055         ibmvnic_send_crq(adapter, &crq);
4056
4057         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4058         atomic_inc(&adapter->running_cap_crqs);
4059         ibmvnic_send_crq(adapter, &crq);
4060
4061         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4062         atomic_inc(&adapter->running_cap_crqs);
4063         ibmvnic_send_crq(adapter, &crq);
4064
4065         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4066         atomic_inc(&adapter->running_cap_crqs);
4067         ibmvnic_send_crq(adapter, &crq);
4068
4069         crq.query_capability.capability =
4070                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4071         atomic_inc(&adapter->running_cap_crqs);
4072         ibmvnic_send_crq(adapter, &crq);
4073
4074         crq.query_capability.capability =
4075                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4076         atomic_inc(&adapter->running_cap_crqs);
4077         ibmvnic_send_crq(adapter, &crq);
4078
4079         crq.query_capability.capability =
4080                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4081         atomic_inc(&adapter->running_cap_crqs);
4082         ibmvnic_send_crq(adapter, &crq);
4083
4084         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4085         atomic_inc(&adapter->running_cap_crqs);
4086         ibmvnic_send_crq(adapter, &crq);
4087 }
4088
4089 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4090 {
4091         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4092         struct device *dev = &adapter->vdev->dev;
4093         union ibmvnic_crq crq;
4094
4095         adapter->ip_offload_tok =
4096                 dma_map_single(dev,
4097                                &adapter->ip_offload_buf,
4098                                buf_sz,
4099                                DMA_FROM_DEVICE);
4100
4101         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4102                 if (!firmware_has_feature(FW_FEATURE_CMO))
4103                         dev_err(dev, "Couldn't map offload buffer\n");
4104                 return;
4105         }
4106
4107         memset(&crq, 0, sizeof(crq));
4108         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4109         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4110         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4111         crq.query_ip_offload.ioba =
4112             cpu_to_be32(adapter->ip_offload_tok);
4113
4114         ibmvnic_send_crq(adapter, &crq);
4115 }
4116
4117 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4118 {
4119         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4120         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4121         struct device *dev = &adapter->vdev->dev;
4122         netdev_features_t old_hw_features = 0;
4123         union ibmvnic_crq crq;
4124
4125         adapter->ip_offload_ctrl_tok =
4126                 dma_map_single(dev,
4127                                ctrl_buf,
4128                                sizeof(adapter->ip_offload_ctrl),
4129                                DMA_TO_DEVICE);
4130
4131         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4132                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4133                 return;
4134         }
4135
4136         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4137         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4138         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4139         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4140         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4141         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4142         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4143         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4144         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4145         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4146
4147         /* large_rx disabled for now, additional features needed */
4148         ctrl_buf->large_rx_ipv4 = 0;
4149         ctrl_buf->large_rx_ipv6 = 0;
4150
4151         if (adapter->state != VNIC_PROBING) {
4152                 old_hw_features = adapter->netdev->hw_features;
4153                 adapter->netdev->hw_features = 0;
4154         }
4155
4156         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4157
4158         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4159                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4160
4161         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4162                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4163
4164         if ((adapter->netdev->features &
4165             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4166                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4167
4168         if (buf->large_tx_ipv4)
4169                 adapter->netdev->hw_features |= NETIF_F_TSO;
4170         if (buf->large_tx_ipv6)
4171                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4172
4173         if (adapter->state == VNIC_PROBING) {
4174                 adapter->netdev->features |= adapter->netdev->hw_features;
4175         } else if (old_hw_features != adapter->netdev->hw_features) {
4176                 netdev_features_t tmp = 0;
4177
4178                 /* disable features no longer supported */
4179                 adapter->netdev->features &= adapter->netdev->hw_features;
4180                 /* turn on features now supported if previously enabled */
4181                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4182                         adapter->netdev->hw_features;
4183                 adapter->netdev->features |=
4184                                 tmp & adapter->netdev->wanted_features;
4185         }
4186
4187         memset(&crq, 0, sizeof(crq));
4188         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4189         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4190         crq.control_ip_offload.len =
4191             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4192         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4193         ibmvnic_send_crq(adapter, &crq);
4194 }
4195
4196 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4197                                 struct ibmvnic_adapter *adapter)
4198 {
4199         struct device *dev = &adapter->vdev->dev;
4200
4201         if (crq->get_vpd_size_rsp.rc.code) {
4202                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4203                         crq->get_vpd_size_rsp.rc.code);
4204                 complete(&adapter->fw_done);
4205                 return;
4206         }
4207
4208         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4209         complete(&adapter->fw_done);
4210 }
4211
4212 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4213                            struct ibmvnic_adapter *adapter)
4214 {
4215         struct device *dev = &adapter->vdev->dev;
4216         unsigned char *substr = NULL;
4217         u8 fw_level_len = 0;
4218
4219         memset(adapter->fw_version, 0, 32);
4220
4221         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4222                          DMA_FROM_DEVICE);
4223
4224         if (crq->get_vpd_rsp.rc.code) {
4225                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4226                         crq->get_vpd_rsp.rc.code);
4227                 goto complete;
4228         }
4229
4230         /* get the position of the firmware version info
4231          * located after the ASCII 'RM' substring in the buffer
4232          */
4233         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4234         if (!substr) {
4235                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4236                 goto complete;
4237         }
4238
4239         /* get length of firmware level ASCII substring */
4240         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4241                 fw_level_len = *(substr + 2);
4242         } else {
4243                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4244                 goto complete;
4245         }
4246
4247         /* copy firmware version string from vpd into adapter */
4248         if ((substr + 3 + fw_level_len) <
4249             (adapter->vpd->buff + adapter->vpd->len)) {
4250                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4251         } else {
4252                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4253         }
4254
4255 complete:
4256         if (adapter->fw_version[0] == '\0')
4257                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4258         complete(&adapter->fw_done);
4259 }
4260
4261 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4262 {
4263         struct device *dev = &adapter->vdev->dev;
4264         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4265         int i;
4266
4267         dma_unmap_single(dev, adapter->ip_offload_tok,
4268                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4269
4270         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4271         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4272                 netdev_dbg(adapter->netdev, "%016lx\n",
4273                            ((unsigned long int *)(buf))[i]);
4274
4275         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4276         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4277         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4278                    buf->tcp_ipv4_chksum);
4279         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4280                    buf->tcp_ipv6_chksum);
4281         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4282                    buf->udp_ipv4_chksum);
4283         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4284                    buf->udp_ipv6_chksum);
4285         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4286                    buf->large_tx_ipv4);
4287         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4288                    buf->large_tx_ipv6);
4289         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4290                    buf->large_rx_ipv4);
4291         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4292                    buf->large_rx_ipv6);
4293         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4294                    buf->max_ipv4_header_size);
4295         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4296                    buf->max_ipv6_header_size);
4297         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4298                    buf->max_tcp_header_size);
4299         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4300                    buf->max_udp_header_size);
4301         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4302                    buf->max_large_tx_size);
4303         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4304                    buf->max_large_rx_size);
4305         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4306                    buf->ipv6_extension_header);
4307         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4308                    buf->tcp_pseudosum_req);
4309         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4310                    buf->num_ipv6_ext_headers);
4311         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4312                    buf->off_ipv6_ext_headers);
4313
4314         send_control_ip_offload(adapter);
4315 }
4316
4317 static const char *ibmvnic_fw_err_cause(u16 cause)
4318 {
4319         switch (cause) {
4320         case ADAPTER_PROBLEM:
4321                 return "adapter problem";
4322         case BUS_PROBLEM:
4323                 return "bus problem";
4324         case FW_PROBLEM:
4325                 return "firmware problem";
4326         case DD_PROBLEM:
4327                 return "device driver problem";
4328         case EEH_RECOVERY:
4329                 return "EEH recovery";
4330         case FW_UPDATED:
4331                 return "firmware updated";
4332         case LOW_MEMORY:
4333                 return "low Memory";
4334         default:
4335                 return "unknown";
4336         }
4337 }
4338
4339 static void handle_error_indication(union ibmvnic_crq *crq,
4340                                     struct ibmvnic_adapter *adapter)
4341 {
4342         struct device *dev = &adapter->vdev->dev;
4343         u16 cause;
4344
4345         cause = be16_to_cpu(crq->error_indication.error_cause);
4346
4347         dev_warn_ratelimited(dev,
4348                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4349                              crq->error_indication.flags
4350                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4351                              ibmvnic_fw_err_cause(cause));
4352
4353         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4354                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4355         else
4356                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4357 }
4358
4359 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4360                                  struct ibmvnic_adapter *adapter)
4361 {
4362         struct net_device *netdev = adapter->netdev;
4363         struct device *dev = &adapter->vdev->dev;
4364         long rc;
4365
4366         rc = crq->change_mac_addr_rsp.rc.code;
4367         if (rc) {
4368                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4369                 goto out;
4370         }
4371         /* crq->change_mac_addr.mac_addr is the requested one
4372          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4373          */
4374         ether_addr_copy(netdev->dev_addr,
4375                         &crq->change_mac_addr_rsp.mac_addr[0]);
4376         ether_addr_copy(adapter->mac_addr,
4377                         &crq->change_mac_addr_rsp.mac_addr[0]);
4378 out:
4379         complete(&adapter->fw_done);
4380         return rc;
4381 }
4382
4383 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4384                                    struct ibmvnic_adapter *adapter)
4385 {
4386         struct device *dev = &adapter->vdev->dev;
4387         u64 *req_value;
4388         char *name;
4389
4390         atomic_dec(&adapter->running_cap_crqs);
4391         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4392         case REQ_TX_QUEUES:
4393                 req_value = &adapter->req_tx_queues;
4394                 name = "tx";
4395                 break;
4396         case REQ_RX_QUEUES:
4397                 req_value = &adapter->req_rx_queues;
4398                 name = "rx";
4399                 break;
4400         case REQ_RX_ADD_QUEUES:
4401                 req_value = &adapter->req_rx_add_queues;
4402                 name = "rx_add";
4403                 break;
4404         case REQ_TX_ENTRIES_PER_SUBCRQ:
4405                 req_value = &adapter->req_tx_entries_per_subcrq;
4406                 name = "tx_entries_per_subcrq";
4407                 break;
4408         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4409                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4410                 name = "rx_add_entries_per_subcrq";
4411                 break;
4412         case REQ_MTU:
4413                 req_value = &adapter->req_mtu;
4414                 name = "mtu";
4415                 break;
4416         case PROMISC_REQUESTED:
4417                 req_value = &adapter->promisc;
4418                 name = "promisc";
4419                 break;
4420         default:
4421                 dev_err(dev, "Got invalid cap request rsp %d\n",
4422                         crq->request_capability.capability);
4423                 return;
4424         }
4425
4426         switch (crq->request_capability_rsp.rc.code) {
4427         case SUCCESS:
4428                 break;
4429         case PARTIALSUCCESS:
4430                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4431                          *req_value,
4432                          (long int)be64_to_cpu(crq->request_capability_rsp.
4433                                                number), name);
4434
4435                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4436                     REQ_MTU) {
4437                         pr_err("mtu of %llu is not supported. Reverting.\n",
4438                                *req_value);
4439                         *req_value = adapter->fallback.mtu;
4440                 } else {
4441                         *req_value =
4442                                 be64_to_cpu(crq->request_capability_rsp.number);
4443                 }
4444
4445                 send_request_cap(adapter, 1);
4446                 return;
4447         default:
4448                 dev_err(dev, "Error %d in request cap rsp\n",
4449                         crq->request_capability_rsp.rc.code);
4450                 return;
4451         }
4452
4453         /* Done receiving requested capabilities, query IP offload support */
4454         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4455                 adapter->wait_capability = false;
4456                 send_query_ip_offload(adapter);
4457         }
4458 }
4459
4460 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4461                             struct ibmvnic_adapter *adapter)
4462 {
4463         struct device *dev = &adapter->vdev->dev;
4464         struct net_device *netdev = adapter->netdev;
4465         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4466         struct ibmvnic_login_buffer *login = adapter->login_buf;
4467         u64 *tx_handle_array;
4468         u64 *rx_handle_array;
4469         int num_tx_pools;
4470         int num_rx_pools;
4471         u64 *size_array;
4472         int i;
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_remove(adapter->vdev);
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                         if (!completion_done(&adapter->init_done)) {
4846                                 complete(&adapter->init_done);
4847                                 adapter->init_done_rc = -EIO;
4848                         }
4849                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4850                         break;
4851                 case IBMVNIC_CRQ_INIT_COMPLETE:
4852                         dev_info(dev, "Partner initialization complete\n");
4853                         adapter->crq.active = true;
4854                         send_version_xchg(adapter);
4855                         break;
4856                 default:
4857                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4858                 }
4859                 return;
4860         case IBMVNIC_CRQ_XPORT_EVENT:
4861                 netif_carrier_off(netdev);
4862                 adapter->crq.active = false;
4863                 /* terminate any thread waiting for a response
4864                  * from the device
4865                  */
4866                 if (!completion_done(&adapter->fw_done)) {
4867                         adapter->fw_done_rc = -EIO;
4868                         complete(&adapter->fw_done);
4869                 }
4870                 if (!completion_done(&adapter->stats_done))
4871                         complete(&adapter->stats_done);
4872                 if (test_bit(0, &adapter->resetting))
4873                         adapter->force_reset_recovery = true;
4874                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4875                         dev_info(dev, "Migrated, re-enabling adapter\n");
4876                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4877                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4878                         dev_info(dev, "Backing device failover detected\n");
4879                         adapter->failover_pending = true;
4880                 } else {
4881                         /* The adapter lost the connection */
4882                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4883                                 gen_crq->cmd);
4884                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4885                 }
4886                 return;
4887         case IBMVNIC_CRQ_CMD_RSP:
4888                 break;
4889         default:
4890                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4891                         gen_crq->first);
4892                 return;
4893         }
4894
4895         switch (gen_crq->cmd) {
4896         case VERSION_EXCHANGE_RSP:
4897                 rc = crq->version_exchange_rsp.rc.code;
4898                 if (rc) {
4899                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4900                         break;
4901                 }
4902                 ibmvnic_version =
4903                             be16_to_cpu(crq->version_exchange_rsp.version);
4904                 dev_info(dev, "Partner protocol version is %d\n",
4905                          ibmvnic_version);
4906                 send_query_cap(adapter);
4907                 break;
4908         case QUERY_CAPABILITY_RSP:
4909                 handle_query_cap_rsp(crq, adapter);
4910                 break;
4911         case QUERY_MAP_RSP:
4912                 handle_query_map_rsp(crq, adapter);
4913                 break;
4914         case REQUEST_MAP_RSP:
4915                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4916                 complete(&adapter->fw_done);
4917                 break;
4918         case REQUEST_UNMAP_RSP:
4919                 handle_request_unmap_rsp(crq, adapter);
4920                 break;
4921         case REQUEST_CAPABILITY_RSP:
4922                 handle_request_cap_rsp(crq, adapter);
4923                 break;
4924         case LOGIN_RSP:
4925                 netdev_dbg(netdev, "Got Login Response\n");
4926                 handle_login_rsp(crq, adapter);
4927                 break;
4928         case LOGICAL_LINK_STATE_RSP:
4929                 netdev_dbg(netdev,
4930                            "Got Logical Link State Response, state: %d rc: %d\n",
4931                            crq->logical_link_state_rsp.link_state,
4932                            crq->logical_link_state_rsp.rc.code);
4933                 adapter->logical_link_state =
4934                     crq->logical_link_state_rsp.link_state;
4935                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4936                 complete(&adapter->init_done);
4937                 break;
4938         case LINK_STATE_INDICATION:
4939                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4940                 adapter->phys_link_state =
4941                     crq->link_state_indication.phys_link_state;
4942                 adapter->logical_link_state =
4943                     crq->link_state_indication.logical_link_state;
4944                 if (adapter->phys_link_state && adapter->logical_link_state)
4945                         netif_carrier_on(netdev);
4946                 else
4947                         netif_carrier_off(netdev);
4948                 break;
4949         case CHANGE_MAC_ADDR_RSP:
4950                 netdev_dbg(netdev, "Got MAC address change Response\n");
4951                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4952                 break;
4953         case ERROR_INDICATION:
4954                 netdev_dbg(netdev, "Got Error Indication\n");
4955                 handle_error_indication(crq, adapter);
4956                 break;
4957         case REQUEST_STATISTICS_RSP:
4958                 netdev_dbg(netdev, "Got Statistics Response\n");
4959                 complete(&adapter->stats_done);
4960                 break;
4961         case QUERY_IP_OFFLOAD_RSP:
4962                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4963                 handle_query_ip_offload_rsp(adapter);
4964                 break;
4965         case MULTICAST_CTRL_RSP:
4966                 netdev_dbg(netdev, "Got multicast control Response\n");
4967                 break;
4968         case CONTROL_IP_OFFLOAD_RSP:
4969                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4970                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4971                                  sizeof(adapter->ip_offload_ctrl),
4972                                  DMA_TO_DEVICE);
4973                 complete(&adapter->init_done);
4974                 break;
4975         case COLLECT_FW_TRACE_RSP:
4976                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4977                 complete(&adapter->fw_done);
4978                 break;
4979         case GET_VPD_SIZE_RSP:
4980                 handle_vpd_size_rsp(crq, adapter);
4981                 break;
4982         case GET_VPD_RSP:
4983                 handle_vpd_rsp(crq, adapter);
4984                 break;
4985         case QUERY_PHYS_PARMS_RSP:
4986                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
4987                 complete(&adapter->fw_done);
4988                 break;
4989         default:
4990                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4991                            gen_crq->cmd);
4992         }
4993 }
4994
4995 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
4996 {
4997         struct ibmvnic_adapter *adapter = instance;
4998
4999         tasklet_schedule(&adapter->tasklet);
5000         return IRQ_HANDLED;
5001 }
5002
5003 static void ibmvnic_tasklet(struct tasklet_struct *t)
5004 {
5005         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5006         struct ibmvnic_crq_queue *queue = &adapter->crq;
5007         union ibmvnic_crq *crq;
5008         unsigned long flags;
5009         bool done = false;
5010
5011         spin_lock_irqsave(&queue->lock, flags);
5012         while (!done) {
5013                 /* Pull all the valid messages off the CRQ */
5014                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5015                         ibmvnic_handle_crq(crq, adapter);
5016                         crq->generic.first = 0;
5017                 }
5018
5019                 /* remain in tasklet until all
5020                  * capabilities responses are received
5021                  */
5022                 if (!adapter->wait_capability)
5023                         done = true;
5024         }
5025         /* if capabilities CRQ's were sent in this tasklet, the following
5026          * tasklet must wait until all responses are received
5027          */
5028         if (atomic_read(&adapter->running_cap_crqs) != 0)
5029                 adapter->wait_capability = true;
5030         spin_unlock_irqrestore(&queue->lock, flags);
5031 }
5032
5033 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5034 {
5035         struct vio_dev *vdev = adapter->vdev;
5036         int rc;
5037
5038         do {
5039                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5040         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5041
5042         if (rc)
5043                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5044
5045         return rc;
5046 }
5047
5048 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5049 {
5050         struct ibmvnic_crq_queue *crq = &adapter->crq;
5051         struct device *dev = &adapter->vdev->dev;
5052         struct vio_dev *vdev = adapter->vdev;
5053         int rc;
5054
5055         /* Close the CRQ */
5056         do {
5057                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5058         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5059
5060         /* Clean out the queue */
5061         if (!crq->msgs)
5062                 return -EINVAL;
5063
5064         memset(crq->msgs, 0, PAGE_SIZE);
5065         crq->cur = 0;
5066         crq->active = false;
5067
5068         /* And re-open it again */
5069         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5070                                 crq->msg_token, PAGE_SIZE);
5071
5072         if (rc == H_CLOSED)
5073                 /* Adapter is good, but other end is not ready */
5074                 dev_warn(dev, "Partner adapter not ready\n");
5075         else if (rc != 0)
5076                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5077
5078         return rc;
5079 }
5080
5081 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5082 {
5083         struct ibmvnic_crq_queue *crq = &adapter->crq;
5084         struct vio_dev *vdev = adapter->vdev;
5085         long rc;
5086
5087         if (!crq->msgs)
5088                 return;
5089
5090         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5091         free_irq(vdev->irq, adapter);
5092         tasklet_kill(&adapter->tasklet);
5093         do {
5094                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5095         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5096
5097         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5098                          DMA_BIDIRECTIONAL);
5099         free_page((unsigned long)crq->msgs);
5100         crq->msgs = NULL;
5101         crq->active = false;
5102 }
5103
5104 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5105 {
5106         struct ibmvnic_crq_queue *crq = &adapter->crq;
5107         struct device *dev = &adapter->vdev->dev;
5108         struct vio_dev *vdev = adapter->vdev;
5109         int rc, retrc = -ENOMEM;
5110
5111         if (crq->msgs)
5112                 return 0;
5113
5114         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5115         /* Should we allocate more than one page? */
5116
5117         if (!crq->msgs)
5118                 return -ENOMEM;
5119
5120         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5121         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5122                                         DMA_BIDIRECTIONAL);
5123         if (dma_mapping_error(dev, crq->msg_token))
5124                 goto map_failed;
5125
5126         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5127                                 crq->msg_token, PAGE_SIZE);
5128
5129         if (rc == H_RESOURCE)
5130                 /* maybe kexecing and resource is busy. try a reset */
5131                 rc = ibmvnic_reset_crq(adapter);
5132         retrc = rc;
5133
5134         if (rc == H_CLOSED) {
5135                 dev_warn(dev, "Partner adapter not ready\n");
5136         } else if (rc) {
5137                 dev_warn(dev, "Error %d opening adapter\n", rc);
5138                 goto reg_crq_failed;
5139         }
5140
5141         retrc = 0;
5142
5143         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5144
5145         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5146         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5147                  adapter->vdev->unit_address);
5148         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5149         if (rc) {
5150                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5151                         vdev->irq, rc);
5152                 goto req_irq_failed;
5153         }
5154
5155         rc = vio_enable_interrupts(vdev);
5156         if (rc) {
5157                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5158                 goto req_irq_failed;
5159         }
5160
5161         crq->cur = 0;
5162         spin_lock_init(&crq->lock);
5163
5164         return retrc;
5165
5166 req_irq_failed:
5167         tasklet_kill(&adapter->tasklet);
5168         do {
5169                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5170         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5171 reg_crq_failed:
5172         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5173 map_failed:
5174         free_page((unsigned long)crq->msgs);
5175         crq->msgs = NULL;
5176         return retrc;
5177 }
5178
5179 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5180 {
5181         struct device *dev = &adapter->vdev->dev;
5182         unsigned long timeout = msecs_to_jiffies(30000);
5183         u64 old_num_rx_queues, old_num_tx_queues;
5184         int rc;
5185
5186         adapter->from_passive_init = false;
5187
5188         if (reset) {
5189                 old_num_rx_queues = adapter->req_rx_queues;
5190                 old_num_tx_queues = adapter->req_tx_queues;
5191                 reinit_completion(&adapter->init_done);
5192         }
5193
5194         adapter->init_done_rc = 0;
5195         rc = ibmvnic_send_crq_init(adapter);
5196         if (rc) {
5197                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5198                 return rc;
5199         }
5200
5201         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5202                 dev_err(dev, "Initialization sequence timed out\n");
5203                 return -1;
5204         }
5205
5206         if (adapter->init_done_rc) {
5207                 release_crq_queue(adapter);
5208                 return adapter->init_done_rc;
5209         }
5210
5211         if (adapter->from_passive_init) {
5212                 adapter->state = VNIC_OPEN;
5213                 adapter->from_passive_init = false;
5214                 return -1;
5215         }
5216
5217         if (reset &&
5218             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5219             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5220                 if (adapter->req_rx_queues != old_num_rx_queues ||
5221                     adapter->req_tx_queues != old_num_tx_queues) {
5222                         release_sub_crqs(adapter, 0);
5223                         rc = init_sub_crqs(adapter);
5224                 } else {
5225                         rc = reset_sub_crq_queues(adapter);
5226                 }
5227         } else {
5228                 rc = init_sub_crqs(adapter);
5229         }
5230
5231         if (rc) {
5232                 dev_err(dev, "Initialization of sub crqs failed\n");
5233                 release_crq_queue(adapter);
5234                 return rc;
5235         }
5236
5237         rc = init_sub_crq_irqs(adapter);
5238         if (rc) {
5239                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5240                 release_crq_queue(adapter);
5241         }
5242
5243         return rc;
5244 }
5245
5246 static struct device_attribute dev_attr_failover;
5247
5248 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5249 {
5250         struct ibmvnic_adapter *adapter;
5251         struct net_device *netdev;
5252         unsigned char *mac_addr_p;
5253         int rc;
5254
5255         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5256                 dev->unit_address);
5257
5258         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5259                                                         VETH_MAC_ADDR, NULL);
5260         if (!mac_addr_p) {
5261                 dev_err(&dev->dev,
5262                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5263                         __FILE__, __LINE__);
5264                 return 0;
5265         }
5266
5267         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5268                                    IBMVNIC_MAX_QUEUES);
5269         if (!netdev)
5270                 return -ENOMEM;
5271
5272         adapter = netdev_priv(netdev);
5273         adapter->state = VNIC_PROBING;
5274         dev_set_drvdata(&dev->dev, netdev);
5275         adapter->vdev = dev;
5276         adapter->netdev = netdev;
5277
5278         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5279         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5280         netdev->irq = dev->irq;
5281         netdev->netdev_ops = &ibmvnic_netdev_ops;
5282         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5283         SET_NETDEV_DEV(netdev, &dev->dev);
5284
5285         spin_lock_init(&adapter->stats_lock);
5286
5287         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5288         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5289                           __ibmvnic_delayed_reset);
5290         INIT_LIST_HEAD(&adapter->rwi_list);
5291         spin_lock_init(&adapter->rwi_lock);
5292         spin_lock_init(&adapter->state_lock);
5293         mutex_init(&adapter->fw_lock);
5294         init_completion(&adapter->init_done);
5295         init_completion(&adapter->fw_done);
5296         init_completion(&adapter->reset_done);
5297         init_completion(&adapter->stats_done);
5298         clear_bit(0, &adapter->resetting);
5299
5300         do {
5301                 rc = init_crq_queue(adapter);
5302                 if (rc) {
5303                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5304                                 rc);
5305                         goto ibmvnic_init_fail;
5306                 }
5307
5308                 rc = ibmvnic_reset_init(adapter, false);
5309                 if (rc && rc != EAGAIN)
5310                         goto ibmvnic_init_fail;
5311         } while (rc == EAGAIN);
5312
5313         rc = init_stats_buffers(adapter);
5314         if (rc)
5315                 goto ibmvnic_init_fail;
5316
5317         rc = init_stats_token(adapter);
5318         if (rc)
5319                 goto ibmvnic_stats_fail;
5320
5321         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5322         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5323         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5324
5325         rc = device_create_file(&dev->dev, &dev_attr_failover);
5326         if (rc)
5327                 goto ibmvnic_dev_file_err;
5328
5329         netif_carrier_off(netdev);
5330         rc = register_netdev(netdev);
5331         if (rc) {
5332                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5333                 goto ibmvnic_register_fail;
5334         }
5335         dev_info(&dev->dev, "ibmvnic registered\n");
5336
5337         adapter->state = VNIC_PROBED;
5338
5339         adapter->wait_for_reset = false;
5340
5341         return 0;
5342
5343 ibmvnic_register_fail:
5344         device_remove_file(&dev->dev, &dev_attr_failover);
5345
5346 ibmvnic_dev_file_err:
5347         release_stats_token(adapter);
5348
5349 ibmvnic_stats_fail:
5350         release_stats_buffers(adapter);
5351
5352 ibmvnic_init_fail:
5353         release_sub_crqs(adapter, 1);
5354         release_crq_queue(adapter);
5355         mutex_destroy(&adapter->fw_lock);
5356         free_netdev(netdev);
5357
5358         return rc;
5359 }
5360
5361 static int ibmvnic_remove(struct vio_dev *dev)
5362 {
5363         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5364         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5365         unsigned long flags;
5366
5367         spin_lock_irqsave(&adapter->state_lock, flags);
5368         if (test_bit(0, &adapter->resetting)) {
5369                 spin_unlock_irqrestore(&adapter->state_lock, flags);
5370                 return -EBUSY;
5371         }
5372
5373         adapter->state = VNIC_REMOVING;
5374         spin_unlock_irqrestore(&adapter->state_lock, flags);
5375
5376         flush_work(&adapter->ibmvnic_reset);
5377         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5378
5379         rtnl_lock();
5380         unregister_netdevice(netdev);
5381
5382         release_resources(adapter);
5383         release_sub_crqs(adapter, 1);
5384         release_crq_queue(adapter);
5385
5386         release_stats_token(adapter);
5387         release_stats_buffers(adapter);
5388
5389         adapter->state = VNIC_REMOVED;
5390
5391         rtnl_unlock();
5392         mutex_destroy(&adapter->fw_lock);
5393         device_remove_file(&dev->dev, &dev_attr_failover);
5394         free_netdev(netdev);
5395         dev_set_drvdata(&dev->dev, NULL);
5396
5397         return 0;
5398 }
5399
5400 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5401                               const char *buf, size_t count)
5402 {
5403         struct net_device *netdev = dev_get_drvdata(dev);
5404         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5405         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5406         __be64 session_token;
5407         long rc;
5408
5409         if (!sysfs_streq(buf, "1"))
5410                 return -EINVAL;
5411
5412         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5413                          H_GET_SESSION_TOKEN, 0, 0, 0);
5414         if (rc) {
5415                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5416                            rc);
5417                 return -EINVAL;
5418         }
5419
5420         session_token = (__be64)retbuf[0];
5421         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5422                    be64_to_cpu(session_token));
5423         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5424                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5425         if (rc) {
5426                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5427                            rc);
5428                 return -EINVAL;
5429         }
5430
5431         return count;
5432 }
5433
5434 static DEVICE_ATTR_WO(failover);
5435
5436 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5437 {
5438         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5439         struct ibmvnic_adapter *adapter;
5440         struct iommu_table *tbl;
5441         unsigned long ret = 0;
5442         int i;
5443
5444         tbl = get_iommu_table_base(&vdev->dev);
5445
5446         /* netdev inits at probe time along with the structures we need below*/
5447         if (!netdev)
5448                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5449
5450         adapter = netdev_priv(netdev);
5451
5452         ret += PAGE_SIZE; /* the crq message queue */
5453         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5454
5455         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5456                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5457
5458         for (i = 0; i < adapter->num_active_rx_pools; i++)
5459                 ret += adapter->rx_pool[i].size *
5460                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5461
5462         return ret;
5463 }
5464
5465 static int ibmvnic_resume(struct device *dev)
5466 {
5467         struct net_device *netdev = dev_get_drvdata(dev);
5468         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5469
5470         if (adapter->state != VNIC_OPEN)
5471                 return 0;
5472
5473         tasklet_schedule(&adapter->tasklet);
5474
5475         return 0;
5476 }
5477
5478 static const struct vio_device_id ibmvnic_device_table[] = {
5479         {"network", "IBM,vnic"},
5480         {"", "" }
5481 };
5482 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5483
5484 static const struct dev_pm_ops ibmvnic_pm_ops = {
5485         .resume = ibmvnic_resume
5486 };
5487
5488 static struct vio_driver ibmvnic_driver = {
5489         .id_table       = ibmvnic_device_table,
5490         .probe          = ibmvnic_probe,
5491         .remove         = ibmvnic_remove,
5492         .get_desired_dma = ibmvnic_get_desired_dma,
5493         .name           = ibmvnic_driver_name,
5494         .pm             = &ibmvnic_pm_ops,
5495 };
5496
5497 /* module functions */
5498 static int __init ibmvnic_module_init(void)
5499 {
5500         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5501                 IBMVNIC_DRIVER_VERSION);
5502
5503         return vio_register_driver(&ibmvnic_driver);
5504 }
5505
5506 static void __exit ibmvnic_module_exit(void)
5507 {
5508         vio_unregister_driver(&ibmvnic_driver);
5509 }
5510
5511 module_init(ibmvnic_module_init);
5512 module_exit(ibmvnic_module_exit);