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 void release_sub_crqs(struct ibmvnic_adapter *, bool);
  82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  88 static int enable_scrq_irq(struct ibmvnic_adapter *,
  89                            struct ibmvnic_sub_crq_queue *);
  90 static int disable_scrq_irq(struct ibmvnic_adapter *,
  91                             struct ibmvnic_sub_crq_queue *);
  92 static int pending_scrq(struct ibmvnic_adapter *,
  93                         struct ibmvnic_sub_crq_queue *);
  94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  95                                         struct ibmvnic_sub_crq_queue *);
  96 static int ibmvnic_poll(struct napi_struct *napi, int data);
  97 static void send_query_map(struct ibmvnic_adapter *adapter);
  98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
  99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 100 static int send_login(struct ibmvnic_adapter *adapter);
 101 static void send_query_cap(struct ibmvnic_adapter *adapter);
 102 static int init_sub_crqs(struct ibmvnic_adapter *);
 103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 105 static void release_crq_queue(struct ibmvnic_adapter *);
 106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
 107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 109
 110 struct ibmvnic_stat {
 111         char name[ETH_GSTRING_LEN];
 112         int offset;
 113 };
 114
 115 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 116                              offsetof(struct ibmvnic_statistics, stat))
 117 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
 118
 119 static const struct ibmvnic_stat ibmvnic_stats[] = {
 120         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 121         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 122         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 123         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 124         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 125         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 126         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 127         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 128         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 129         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 130         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 131         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 132         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 133         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 134         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 135         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 136         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 137         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 138         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 139         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 140         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 141         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 142 };
 143
 144 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 145                           unsigned long length, unsigned long *number,
 146                           unsigned long *irq)
 147 {
 148         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 149         long rc;
 150
 151         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 152         *number = retbuf[0];
 153         *irq = retbuf[1];
 154
 155         return rc;
 156 }
 157
 158 /**
 159  * ibmvnic_wait_for_completion - Check device state and wait for completion
 160  * @adapter: private device data
 161  * @comp_done: completion structure to wait for
 162  * @timeout: time to wait in milliseconds
 163  *
 164  * Wait for a completion signal or until the timeout limit is reached
 165  * while checking that the device is still active.
 166  */
 167 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 168                                        struct completion *comp_done,
 169                                        unsigned long timeout)
 170 {
 171         struct net_device *netdev;
 172         unsigned long div_timeout;
 173         u8 retry;
 174
 175         netdev = adapter->netdev;
 176         retry = 5;
 177         div_timeout = msecs_to_jiffies(timeout / retry);
 178         while (true) {
 179                 if (!adapter->crq.active) {
 180                         netdev_err(netdev, "Device down!\n");
 181                         return -ENODEV;
 182                 }
 183                 if (!retry--)
 184                         break;
 185                 if (wait_for_completion_timeout(comp_done, div_timeout))
 186                         return 0;
 187         }
 188         netdev_err(netdev, "Operation timed out.\n");
 189         return -ETIMEDOUT;
 190 }
 191
 192 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 193                                 struct ibmvnic_long_term_buff *ltb, int size)
 194 {
 195         struct device *dev = &adapter->vdev->dev;
 196         int rc;
 197
 198         ltb->size = size;
 199         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 200                                        GFP_KERNEL);
 201
 202         if (!ltb->buff) {
 203                 dev_err(dev, "Couldn't alloc long term buffer\n");
 204                 return -ENOMEM;
 205         }
 206         ltb->map_id = adapter->map_id;
 207         adapter->map_id++;
 208
 209         mutex_lock(&adapter->fw_lock);
 210         adapter->fw_done_rc = 0;
 211         reinit_completion(&adapter->fw_done);
 212         rc = send_request_map(adapter, ltb->addr,
 213                               ltb->size, ltb->map_id);
 214         if (rc) {
 215                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 216                 mutex_unlock(&adapter->fw_lock);
 217                 return rc;
 218         }
 219
 220         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 221         if (rc) {
 222                 dev_err(dev,
 223                         "Long term map request aborted or timed out,rc = %d\n",
 224                         rc);
 225                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 226                 mutex_unlock(&adapter->fw_lock);
 227                 return rc;
 228         }
 229
 230         if (adapter->fw_done_rc) {
 231                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 232                         adapter->fw_done_rc);
 233                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 234                 mutex_unlock(&adapter->fw_lock);
 235                 return -1;
 236         }
 237         mutex_unlock(&adapter->fw_lock);
 238         return 0;
 239 }
 240
 241 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 242                                 struct ibmvnic_long_term_buff *ltb)
 243 {
 244         struct device *dev = &adapter->vdev->dev;
 245
 246         if (!ltb->buff)
 247                 return;
 248
 249         /* VIOS automatically unmaps the long term buffer at remote
 250          * end for the following resets:
 251          * FAILOVER, MOBILITY, TIMEOUT.
 252          */
 253         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 254             adapter->reset_reason != VNIC_RESET_MOBILITY &&
 255             adapter->reset_reason != VNIC_RESET_TIMEOUT)
 256                 send_request_unmap(adapter, ltb->map_id);
 257         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 258 }
 259
 260 static int reset_long_term_buff(struct ibmvnic_long_term_buff *ltb)
 261 {
 262         if (!ltb->buff)
 263                 return -EINVAL;
 264
 265         memset(ltb->buff, 0, ltb->size);
 266         return 0;
 267 }
 268
 269 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 270 {
 271         int i;
 272
 273         for (i = 0; i < adapter->num_active_rx_pools; i++)
 274                 adapter->rx_pool[i].active = 0;
 275 }
 276
 277 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 278                               struct ibmvnic_rx_pool *pool)
 279 {
 280         int count = pool->size - atomic_read(&pool->available);
 281         u64 handle = adapter->rx_scrq[pool->index]->handle;
 282         struct device *dev = &adapter->vdev->dev;
 283         struct ibmvnic_ind_xmit_queue *ind_bufp;
 284         struct ibmvnic_sub_crq_queue *rx_scrq;
 285         union sub_crq *sub_crq;
 286         int buffers_added = 0;
 287         unsigned long lpar_rc;
 288         struct sk_buff *skb;
 289         unsigned int offset;
 290         dma_addr_t dma_addr;
 291         unsigned char *dst;
 292         int shift = 0;
 293         int index;
 294         int i;
 295
 296         if (!pool->active)
 297                 return;
 298
 299         rx_scrq = adapter->rx_scrq[pool->index];
 300         ind_bufp = &rx_scrq->ind_buf;
 301         for (i = 0; i < count; ++i) {
 302                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 303                 if (!skb) {
 304                         dev_err(dev, "Couldn't replenish rx buff\n");
 305                         adapter->replenish_no_mem++;
 306                         break;
 307                 }
 308
 309                 index = pool->free_map[pool->next_free];
 310
 311                 if (pool->rx_buff[index].skb)
 312                         dev_err(dev, "Inconsistent free_map!\n");
 313
 314                 /* Copy the skb to the long term mapped DMA buffer */
 315                 offset = index * pool->buff_size;
 316                 dst = pool->long_term_buff.buff + offset;
 317                 memset(dst, 0, pool->buff_size);
 318                 dma_addr = pool->long_term_buff.addr + offset;
 319                 pool->rx_buff[index].data = dst;
 320
 321                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 322                 pool->rx_buff[index].dma = dma_addr;
 323                 pool->rx_buff[index].skb = skb;
 324                 pool->rx_buff[index].pool_index = pool->index;
 325                 pool->rx_buff[index].size = pool->buff_size;
 326
 327                 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
 328                 memset(sub_crq, 0, sizeof(*sub_crq));
 329                 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
 330                 sub_crq->rx_add.correlator =
 331                     cpu_to_be64((u64)&pool->rx_buff[index]);
 332                 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
 333                 sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
 334
 335                 /* The length field of the sCRQ is defined to be 24 bits so the
 336                  * buffer size needs to be left shifted by a byte before it is
 337                  * converted to big endian to prevent the last byte from being
 338                  * truncated.
 339                  */
 340 #ifdef __LITTLE_ENDIAN__
 341                 shift = 8;
 342 #endif
 343                 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
 344                 pool->next_free = (pool->next_free + 1) % pool->size;
 345                 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
 346                     i == count - 1) {
 347                         lpar_rc =
 348                                 send_subcrq_indirect(adapter, handle,
 349                                                      (u64)ind_bufp->indir_dma,
 350                                                      (u64)ind_bufp->index);
 351                         if (lpar_rc != H_SUCCESS)
 352                                 goto failure;
 353                         buffers_added += ind_bufp->index;
 354                         adapter->replenish_add_buff_success += ind_bufp->index;
 355                         ind_bufp->index = 0;
 356                 }
 357         }
 358         atomic_add(buffers_added, &pool->available);
 359         return;
 360
 361 failure:
 362         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 363                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 364         for (i = ind_bufp->index - 1; i >= 0; --i) {
 365                 struct ibmvnic_rx_buff *rx_buff;
 366
 367                 pool->next_free = pool->next_free == 0 ?
 368                                   pool->size - 1 : pool->next_free - 1;
 369                 sub_crq = &ind_bufp->indir_arr[i];
 370                 rx_buff = (struct ibmvnic_rx_buff *)
 371                                 be64_to_cpu(sub_crq->rx_add.correlator);
 372                 index = (int)(rx_buff - pool->rx_buff);
 373                 pool->free_map[pool->next_free] = index;
 374                 dev_kfree_skb_any(pool->rx_buff[index].skb);
 375                 pool->rx_buff[index].skb = NULL;
 376         }
 377         adapter->replenish_add_buff_failure += ind_bufp->index;
 378         atomic_add(buffers_added, &pool->available);
 379         ind_bufp->index = 0;
 380         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 381                 /* Disable buffer pool replenishment and report carrier off if
 382                  * queue is closed or pending failover.
 383                  * Firmware guarantees that a signal will be sent to the
 384                  * driver, triggering a reset.
 385                  */
 386                 deactivate_rx_pools(adapter);
 387                 netif_carrier_off(adapter->netdev);
 388         }
 389 }
 390
 391 static void replenish_pools(struct ibmvnic_adapter *adapter)
 392 {
 393         int i;
 394
 395         adapter->replenish_task_cycles++;
 396         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 397                 if (adapter->rx_pool[i].active)
 398                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 399         }
 400
 401         netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 402 }
 403
 404 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 405 {
 406         kfree(adapter->tx_stats_buffers);
 407         kfree(adapter->rx_stats_buffers);
 408         adapter->tx_stats_buffers = NULL;
 409         adapter->rx_stats_buffers = NULL;
 410 }
 411
 412 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 413 {
 414         adapter->tx_stats_buffers =
 415                                 kcalloc(IBMVNIC_MAX_QUEUES,
 416                                         sizeof(struct ibmvnic_tx_queue_stats),
 417                                         GFP_KERNEL);
 418         if (!adapter->tx_stats_buffers)
 419                 return -ENOMEM;
 420
 421         adapter->rx_stats_buffers =
 422                                 kcalloc(IBMVNIC_MAX_QUEUES,
 423                                         sizeof(struct ibmvnic_rx_queue_stats),
 424                                         GFP_KERNEL);
 425         if (!adapter->rx_stats_buffers)
 426                 return -ENOMEM;
 427
 428         return 0;
 429 }
 430
 431 static void release_stats_token(struct ibmvnic_adapter *adapter)
 432 {
 433         struct device *dev = &adapter->vdev->dev;
 434
 435         if (!adapter->stats_token)
 436                 return;
 437
 438         dma_unmap_single(dev, adapter->stats_token,
 439                          sizeof(struct ibmvnic_statistics),
 440                          DMA_FROM_DEVICE);
 441         adapter->stats_token = 0;
 442 }
 443
 444 static int init_stats_token(struct ibmvnic_adapter *adapter)
 445 {
 446         struct device *dev = &adapter->vdev->dev;
 447         dma_addr_t stok;
 448
 449         stok = dma_map_single(dev, &adapter->stats,
 450                               sizeof(struct ibmvnic_statistics),
 451                               DMA_FROM_DEVICE);
 452         if (dma_mapping_error(dev, stok)) {
 453                 dev_err(dev, "Couldn't map stats buffer\n");
 454                 return -1;
 455         }
 456
 457         adapter->stats_token = stok;
 458         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 459         return 0;
 460 }
 461
 462 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 463 {
 464         struct ibmvnic_rx_pool *rx_pool;
 465         u64 buff_size;
 466         int rx_scrqs;
 467         int i, j, rc;
 468
 469         if (!adapter->rx_pool)
 470                 return -1;
 471
 472         buff_size = adapter->cur_rx_buf_sz;
 473         rx_scrqs = adapter->num_active_rx_pools;
 474         for (i = 0; i < rx_scrqs; i++) {
 475                 rx_pool = &adapter->rx_pool[i];
 476
 477                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 478
 479                 if (rx_pool->buff_size != buff_size) {
 480                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 481                         rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 482                         rc = alloc_long_term_buff(adapter,
 483                                                   &rx_pool->long_term_buff,
 484                                                   rx_pool->size *
 485                                                   rx_pool->buff_size);
 486                 } else {
 487                         rc = reset_long_term_buff(&rx_pool->long_term_buff);
 488                 }
 489
 490                 if (rc)
 491                         return rc;
 492
 493                 for (j = 0; j < rx_pool->size; j++)
 494                         rx_pool->free_map[j] = j;
 495
 496                 memset(rx_pool->rx_buff, 0,
 497                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 498
 499                 atomic_set(&rx_pool->available, 0);
 500                 rx_pool->next_alloc = 0;
 501                 rx_pool->next_free = 0;
 502                 rx_pool->active = 1;
 503         }
 504
 505         return 0;
 506 }
 507
 508 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 509 {
 510         struct ibmvnic_rx_pool *rx_pool;
 511         int i, j;
 512
 513         if (!adapter->rx_pool)
 514                 return;
 515
 516         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 517                 rx_pool = &adapter->rx_pool[i];
 518
 519                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 520
 521                 kfree(rx_pool->free_map);
 522                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 523
 524                 if (!rx_pool->rx_buff)
 525                         continue;
 526
 527                 for (j = 0; j < rx_pool->size; j++) {
 528                         if (rx_pool->rx_buff[j].skb) {
 529                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 530                                 rx_pool->rx_buff[j].skb = NULL;
 531                         }
 532                 }
 533
 534                 kfree(rx_pool->rx_buff);
 535         }
 536
 537         kfree(adapter->rx_pool);
 538         adapter->rx_pool = NULL;
 539         adapter->num_active_rx_pools = 0;
 540 }
 541
 542 static int init_rx_pools(struct net_device *netdev)
 543 {
 544         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 545         struct device *dev = &adapter->vdev->dev;
 546         struct ibmvnic_rx_pool *rx_pool;
 547         int rxadd_subcrqs;
 548         u64 buff_size;
 549         int i, j;
 550
 551         rxadd_subcrqs = adapter->num_active_rx_scrqs;
 552         buff_size = adapter->cur_rx_buf_sz;
 553
 554         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 555                                    sizeof(struct ibmvnic_rx_pool),
 556                                    GFP_KERNEL);
 557         if (!adapter->rx_pool) {
 558                 dev_err(dev, "Failed to allocate rx pools\n");
 559                 return -1;
 560         }
 561
 562         adapter->num_active_rx_pools = rxadd_subcrqs;
 563
 564         for (i = 0; i < rxadd_subcrqs; i++) {
 565                 rx_pool = &adapter->rx_pool[i];
 566
 567                 netdev_dbg(adapter->netdev,
 568                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 569                            i, adapter->req_rx_add_entries_per_subcrq,
 570                            buff_size);
 571
 572                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 573                 rx_pool->index = i;
 574                 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 575                 rx_pool->active = 1;
 576
 577                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 578                                             GFP_KERNEL);
 579                 if (!rx_pool->free_map) {
 580                         release_rx_pools(adapter);
 581                         return -1;
 582                 }
 583
 584                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 585                                            sizeof(struct ibmvnic_rx_buff),
 586                                            GFP_KERNEL);
 587                 if (!rx_pool->rx_buff) {
 588                         dev_err(dev, "Couldn't alloc rx buffers\n");
 589                         release_rx_pools(adapter);
 590                         return -1;
 591                 }
 592
 593                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 594                                          rx_pool->size * rx_pool->buff_size)) {
 595                         release_rx_pools(adapter);
 596                         return -1;
 597                 }
 598
 599                 for (j = 0; j < rx_pool->size; ++j)
 600                         rx_pool->free_map[j] = j;
 601
 602                 atomic_set(&rx_pool->available, 0);
 603                 rx_pool->next_alloc = 0;
 604                 rx_pool->next_free = 0;
 605         }
 606
 607         return 0;
 608 }
 609
 610 static int reset_one_tx_pool(struct ibmvnic_tx_pool *tx_pool)
 611 {
 612         int rc, i;
 613
 614         rc = reset_long_term_buff(&tx_pool->long_term_buff);
 615         if (rc)
 616                 return rc;
 617
 618         memset(tx_pool->tx_buff, 0,
 619                tx_pool->num_buffers *
 620                sizeof(struct ibmvnic_tx_buff));
 621
 622         for (i = 0; i < tx_pool->num_buffers; i++)
 623                 tx_pool->free_map[i] = i;
 624
 625         tx_pool->consumer_index = 0;
 626         tx_pool->producer_index = 0;
 627
 628         return 0;
 629 }
 630
 631 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 632 {
 633         int tx_scrqs;
 634         int i, rc;
 635
 636         if (!adapter->tx_pool)
 637                 return -1;
 638
 639         tx_scrqs = adapter->num_active_tx_pools;
 640         for (i = 0; i < tx_scrqs; i++) {
 641                 rc = reset_one_tx_pool(&adapter->tso_pool[i]);
 642                 if (rc)
 643                         return rc;
 644                 rc = reset_one_tx_pool(&adapter->tx_pool[i]);
 645                 if (rc)
 646                         return rc;
 647         }
 648
 649         return 0;
 650 }
 651
 652 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 653 {
 654         if (!adapter->vpd)
 655                 return;
 656
 657         kfree(adapter->vpd->buff);
 658         kfree(adapter->vpd);
 659
 660         adapter->vpd = NULL;
 661 }
 662
 663 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 664                                 struct ibmvnic_tx_pool *tx_pool)
 665 {
 666         kfree(tx_pool->tx_buff);
 667         kfree(tx_pool->free_map);
 668         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 669 }
 670
 671 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 672 {
 673         int i;
 674
 675         if (!adapter->tx_pool)
 676                 return;
 677
 678         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 679                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 680                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 681         }
 682
 683         kfree(adapter->tx_pool);
 684         adapter->tx_pool = NULL;
 685         kfree(adapter->tso_pool);
 686         adapter->tso_pool = NULL;
 687         adapter->num_active_tx_pools = 0;
 688 }
 689
 690 static int init_one_tx_pool(struct net_device *netdev,
 691                             struct ibmvnic_tx_pool *tx_pool,
 692                             int num_entries, int buf_size)
 693 {
 694         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 695         int i;
 696
 697         tx_pool->tx_buff = kcalloc(num_entries,
 698                                    sizeof(struct ibmvnic_tx_buff),
 699                                    GFP_KERNEL);
 700         if (!tx_pool->tx_buff)
 701                 return -1;
 702
 703         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 704                                  num_entries * buf_size))
 705                 return -1;
 706
 707         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 708         if (!tx_pool->free_map)
 709                 return -1;
 710
 711         for (i = 0; i < num_entries; i++)
 712                 tx_pool->free_map[i] = i;
 713
 714         tx_pool->consumer_index = 0;
 715         tx_pool->producer_index = 0;
 716         tx_pool->num_buffers = num_entries;
 717         tx_pool->buf_size = buf_size;
 718
 719         return 0;
 720 }
 721
 722 static int init_tx_pools(struct net_device *netdev)
 723 {
 724         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 725         int tx_subcrqs;
 726         u64 buff_size;
 727         int i, rc;
 728
 729         tx_subcrqs = adapter->num_active_tx_scrqs;
 730         adapter->tx_pool = kcalloc(tx_subcrqs,
 731                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 732         if (!adapter->tx_pool)
 733                 return -1;
 734
 735         adapter->tso_pool = kcalloc(tx_subcrqs,
 736                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 737         if (!adapter->tso_pool)
 738                 return -1;
 739
 740         adapter->num_active_tx_pools = tx_subcrqs;
 741
 742         for (i = 0; i < tx_subcrqs; i++) {
 743                 buff_size = adapter->req_mtu + VLAN_HLEN;
 744                 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 745                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 746                                       adapter->req_tx_entries_per_subcrq,
 747                                       buff_size);
 748                 if (rc) {
 749                         release_tx_pools(adapter);
 750                         return rc;
 751                 }
 752
 753                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 754                                       IBMVNIC_TSO_BUFS,
 755                                       IBMVNIC_TSO_BUF_SZ);
 756                 if (rc) {
 757                         release_tx_pools(adapter);
 758                         return rc;
 759                 }
 760         }
 761
 762         return 0;
 763 }
 764
 765 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 766 {
 767         int i;
 768
 769         if (adapter->napi_enabled)
 770                 return;
 771
 772         for (i = 0; i < adapter->req_rx_queues; i++)
 773                 napi_enable(&adapter->napi[i]);
 774
 775         adapter->napi_enabled = true;
 776 }
 777
 778 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 779 {
 780         int i;
 781
 782         if (!adapter->napi_enabled)
 783                 return;
 784
 785         for (i = 0; i < adapter->req_rx_queues; i++) {
 786                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 787                 napi_disable(&adapter->napi[i]);
 788         }
 789
 790         adapter->napi_enabled = false;
 791 }
 792
 793 static int init_napi(struct ibmvnic_adapter *adapter)
 794 {
 795         int i;
 796
 797         adapter->napi = kcalloc(adapter->req_rx_queues,
 798                                 sizeof(struct napi_struct), GFP_KERNEL);
 799         if (!adapter->napi)
 800                 return -ENOMEM;
 801
 802         for (i = 0; i < adapter->req_rx_queues; i++) {
 803                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 804                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 805                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 806         }
 807
 808         adapter->num_active_rx_napi = adapter->req_rx_queues;
 809         return 0;
 810 }
 811
 812 static void release_napi(struct ibmvnic_adapter *adapter)
 813 {
 814         int i;
 815
 816         if (!adapter->napi)
 817                 return;
 818
 819         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 820                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 821                 netif_napi_del(&adapter->napi[i]);
 822         }
 823
 824         kfree(adapter->napi);
 825         adapter->napi = NULL;
 826         adapter->num_active_rx_napi = 0;
 827         adapter->napi_enabled = false;
 828 }
 829
 830 static int ibmvnic_login(struct net_device *netdev)
 831 {
 832         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 833         unsigned long timeout = msecs_to_jiffies(20000);
 834         int retry_count = 0;
 835         int retries = 10;
 836         bool retry;
 837         int rc;
 838
 839         do {
 840                 retry = false;
 841                 if (retry_count > retries) {
 842                         netdev_warn(netdev, "Login attempts exceeded\n");
 843                         return -1;
 844                 }
 845
 846                 adapter->init_done_rc = 0;
 847                 reinit_completion(&adapter->init_done);
 848                 rc = send_login(adapter);
 849                 if (rc)
 850                         return rc;
 851
 852                 if (!wait_for_completion_timeout(&adapter->init_done,
 853                                                  timeout)) {
 854                         netdev_warn(netdev, "Login timed out, retrying...\n");
 855                         retry = true;
 856                         adapter->init_done_rc = 0;
 857                         retry_count++;
 858                         continue;
 859                 }
 860
 861                 if (adapter->init_done_rc == ABORTED) {
 862                         netdev_warn(netdev, "Login aborted, retrying...\n");
 863                         retry = true;
 864                         adapter->init_done_rc = 0;
 865                         retry_count++;
 866                         /* FW or device may be busy, so
 867                          * wait a bit before retrying login
 868                          */
 869                         msleep(500);
 870                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 871                         retry_count++;
 872                         release_sub_crqs(adapter, 1);
 873
 874                         retry = true;
 875                         netdev_dbg(netdev,
 876                                    "Received partial success, retrying...\n");
 877                         adapter->init_done_rc = 0;
 878                         reinit_completion(&adapter->init_done);
 879                         send_query_cap(adapter);
 880                         if (!wait_for_completion_timeout(&adapter->init_done,
 881                                                          timeout)) {
 882                                 netdev_warn(netdev,
 883                                             "Capabilities query timed out\n");
 884                                 return -1;
 885                         }
 886
 887                         rc = init_sub_crqs(adapter);
 888                         if (rc) {
 889                                 netdev_warn(netdev,
 890                                             "SCRQ initialization failed\n");
 891                                 return -1;
 892                         }
 893
 894                         rc = init_sub_crq_irqs(adapter);
 895                         if (rc) {
 896                                 netdev_warn(netdev,
 897                                             "SCRQ irq initialization failed\n");
 898                                 return -1;
 899                         }
 900                 } else if (adapter->init_done_rc) {
 901                         netdev_warn(netdev, "Adapter login failed\n");
 902                         return -1;
 903                 }
 904         } while (retry);
 905
 906         __ibmvnic_set_mac(netdev, adapter->mac_addr);
 907
 908         netdev_dbg(netdev, "[S:%d] Login succeeded\n", adapter->state);
 909         return 0;
 910 }
 911
 912 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 913 {
 914         kfree(adapter->login_buf);
 915         adapter->login_buf = NULL;
 916 }
 917
 918 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 919 {
 920         kfree(adapter->login_rsp_buf);
 921         adapter->login_rsp_buf = NULL;
 922 }
 923
 924 static void release_resources(struct ibmvnic_adapter *adapter)
 925 {
 926         release_vpd_data(adapter);
 927
 928         release_tx_pools(adapter);
 929         release_rx_pools(adapter);
 930
 931         release_napi(adapter);
 932         release_login_buffer(adapter);
 933         release_login_rsp_buffer(adapter);
 934 }
 935
 936 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 937 {
 938         struct net_device *netdev = adapter->netdev;
 939         unsigned long timeout = msecs_to_jiffies(20000);
 940         union ibmvnic_crq crq;
 941         bool resend;
 942         int rc;
 943
 944         netdev_dbg(netdev, "setting link state %d\n", link_state);
 945
 946         memset(&crq, 0, sizeof(crq));
 947         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 948         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 949         crq.logical_link_state.link_state = link_state;
 950
 951         do {
 952                 resend = false;
 953
 954                 reinit_completion(&adapter->init_done);
 955                 rc = ibmvnic_send_crq(adapter, &crq);
 956                 if (rc) {
 957                         netdev_err(netdev, "Failed to set link state\n");
 958                         return rc;
 959                 }
 960
 961                 if (!wait_for_completion_timeout(&adapter->init_done,
 962                                                  timeout)) {
 963                         netdev_err(netdev, "timeout setting link state\n");
 964                         return -1;
 965                 }
 966
 967                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 968                         /* Partuial success, delay and re-send */
 969                         mdelay(1000);
 970                         resend = true;
 971                 } else if (adapter->init_done_rc) {
 972                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 973                                     adapter->init_done_rc);
 974                         return adapter->init_done_rc;
 975                 }
 976         } while (resend);
 977
 978         return 0;
 979 }
 980
 981 static int set_real_num_queues(struct net_device *netdev)
 982 {
 983         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 984         int rc;
 985
 986         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
 987                    adapter->req_tx_queues, adapter->req_rx_queues);
 988
 989         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
 990         if (rc) {
 991                 netdev_err(netdev, "failed to set the number of tx queues\n");
 992                 return rc;
 993         }
 994
 995         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
 996         if (rc)
 997                 netdev_err(netdev, "failed to set the number of rx queues\n");
 998
 999         return rc;
1000 }
1001
1002 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1003 {
1004         struct device *dev = &adapter->vdev->dev;
1005         union ibmvnic_crq crq;
1006         int len = 0;
1007         int rc;
1008
1009         if (adapter->vpd->buff)
1010                 len = adapter->vpd->len;
1011
1012         mutex_lock(&adapter->fw_lock);
1013         adapter->fw_done_rc = 0;
1014         reinit_completion(&adapter->fw_done);
1015
1016         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1017         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1018         rc = ibmvnic_send_crq(adapter, &crq);
1019         if (rc) {
1020                 mutex_unlock(&adapter->fw_lock);
1021                 return rc;
1022         }
1023
1024         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1025         if (rc) {
1026                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1027                 mutex_unlock(&adapter->fw_lock);
1028                 return rc;
1029         }
1030         mutex_unlock(&adapter->fw_lock);
1031
1032         if (!adapter->vpd->len)
1033                 return -ENODATA;
1034
1035         if (!adapter->vpd->buff)
1036                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1037         else if (adapter->vpd->len != len)
1038                 adapter->vpd->buff =
1039                         krealloc(adapter->vpd->buff,
1040                                  adapter->vpd->len, GFP_KERNEL);
1041
1042         if (!adapter->vpd->buff) {
1043                 dev_err(dev, "Could allocate VPD buffer\n");
1044                 return -ENOMEM;
1045         }
1046
1047         adapter->vpd->dma_addr =
1048                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1049                                DMA_FROM_DEVICE);
1050         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1051                 dev_err(dev, "Could not map VPD buffer\n");
1052                 kfree(adapter->vpd->buff);
1053                 adapter->vpd->buff = NULL;
1054                 return -ENOMEM;
1055         }
1056
1057         mutex_lock(&adapter->fw_lock);
1058         adapter->fw_done_rc = 0;
1059         reinit_completion(&adapter->fw_done);
1060
1061         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1062         crq.get_vpd.cmd = GET_VPD;
1063         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1064         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1065         rc = ibmvnic_send_crq(adapter, &crq);
1066         if (rc) {
1067                 kfree(adapter->vpd->buff);
1068                 adapter->vpd->buff = NULL;
1069                 mutex_unlock(&adapter->fw_lock);
1070                 return rc;
1071         }
1072
1073         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1074         if (rc) {
1075                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1076                 kfree(adapter->vpd->buff);
1077                 adapter->vpd->buff = NULL;
1078                 mutex_unlock(&adapter->fw_lock);
1079                 return rc;
1080         }
1081
1082         mutex_unlock(&adapter->fw_lock);
1083         return 0;
1084 }
1085
1086 static int init_resources(struct ibmvnic_adapter *adapter)
1087 {
1088         struct net_device *netdev = adapter->netdev;
1089         int rc;
1090
1091         rc = set_real_num_queues(netdev);
1092         if (rc)
1093                 return rc;
1094
1095         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1096         if (!adapter->vpd)
1097                 return -ENOMEM;
1098
1099         /* Vital Product Data (VPD) */
1100         rc = ibmvnic_get_vpd(adapter);
1101         if (rc) {
1102                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1103                 return rc;
1104         }
1105
1106         adapter->map_id = 1;
1107
1108         rc = init_napi(adapter);
1109         if (rc)
1110                 return rc;
1111
1112         send_query_map(adapter);
1113
1114         rc = init_rx_pools(netdev);
1115         if (rc)
1116                 return rc;
1117
1118         rc = init_tx_pools(netdev);
1119         return rc;
1120 }
1121
1122 static int __ibmvnic_open(struct net_device *netdev)
1123 {
1124         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1125         enum vnic_state prev_state = adapter->state;
1126         int i, rc;
1127
1128         adapter->state = VNIC_OPENING;
1129         replenish_pools(adapter);
1130         ibmvnic_napi_enable(adapter);
1131
1132         /* We're ready to receive frames, enable the sub-crq interrupts and
1133          * set the logical link state to up
1134          */
1135         for (i = 0; i < adapter->req_rx_queues; i++) {
1136                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1137                 if (prev_state == VNIC_CLOSED)
1138                         enable_irq(adapter->rx_scrq[i]->irq);
1139                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1140         }
1141
1142         for (i = 0; i < adapter->req_tx_queues; i++) {
1143                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1144                 if (prev_state == VNIC_CLOSED)
1145                         enable_irq(adapter->tx_scrq[i]->irq);
1146                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1147                 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1148         }
1149
1150         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1151         if (rc) {
1152                 for (i = 0; i < adapter->req_rx_queues; i++)
1153                         napi_disable(&adapter->napi[i]);
1154                 release_resources(adapter);
1155                 return rc;
1156         }
1157
1158         netif_tx_start_all_queues(netdev);
1159
1160         if (prev_state == VNIC_CLOSED) {
1161                 for (i = 0; i < adapter->req_rx_queues; i++)
1162                         napi_schedule(&adapter->napi[i]);
1163         }
1164
1165         adapter->state = VNIC_OPEN;
1166         return rc;
1167 }
1168
1169 static int ibmvnic_open(struct net_device *netdev)
1170 {
1171         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1172         int rc;
1173
1174         ASSERT_RTNL();
1175
1176         /* If device failover is pending or we are about to reset, just set
1177          * device state and return. Device operation will be handled by reset
1178          * routine.
1179          *
1180          * It should be safe to overwrite the adapter->state here. Since
1181          * we hold the rtnl, either the reset has not actually started or
1182          * the rtnl got dropped during the set_link_state() in do_reset().
1183          * In the former case, no one else is changing the state (again we
1184          * have the rtnl) and in the latter case, do_reset() will detect and
1185          * honor our setting below.
1186          */
1187         if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1188                 netdev_dbg(netdev, "[S:%d FOP:%d] Resetting, deferring open\n",
1189                            adapter->state, adapter->failover_pending);
1190                 adapter->state = VNIC_OPEN;
1191                 rc = 0;
1192                 goto out;
1193         }
1194
1195         if (adapter->state != VNIC_CLOSED) {
1196                 rc = ibmvnic_login(netdev);
1197                 if (rc)
1198                         goto out;
1199
1200                 rc = init_resources(adapter);
1201                 if (rc) {
1202                         netdev_err(netdev, "failed to initialize resources\n");
1203                         release_resources(adapter);
1204                         goto out;
1205                 }
1206         }
1207
1208         rc = __ibmvnic_open(netdev);
1209
1210 out:
1211         /* If open failed and there is a pending failover or in-progress reset,
1212          * set device state and return. Device operation will be handled by
1213          * reset routine. See also comments above regarding rtnl.
1214          */
1215         if (rc &&
1216             (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1217                 adapter->state = VNIC_OPEN;
1218                 rc = 0;
1219         }
1220         return rc;
1221 }
1222
1223 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1224 {
1225         struct ibmvnic_rx_pool *rx_pool;
1226         struct ibmvnic_rx_buff *rx_buff;
1227         u64 rx_entries;
1228         int rx_scrqs;
1229         int i, j;
1230
1231         if (!adapter->rx_pool)
1232                 return;
1233
1234         rx_scrqs = adapter->num_active_rx_pools;
1235         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1236
1237         /* Free any remaining skbs in the rx buffer pools */
1238         for (i = 0; i < rx_scrqs; i++) {
1239                 rx_pool = &adapter->rx_pool[i];
1240                 if (!rx_pool || !rx_pool->rx_buff)
1241                         continue;
1242
1243                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1244                 for (j = 0; j < rx_entries; j++) {
1245                         rx_buff = &rx_pool->rx_buff[j];
1246                         if (rx_buff && rx_buff->skb) {
1247                                 dev_kfree_skb_any(rx_buff->skb);
1248                                 rx_buff->skb = NULL;
1249                         }
1250                 }
1251         }
1252 }
1253
1254 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1255                               struct ibmvnic_tx_pool *tx_pool)
1256 {
1257         struct ibmvnic_tx_buff *tx_buff;
1258         u64 tx_entries;
1259         int i;
1260
1261         if (!tx_pool || !tx_pool->tx_buff)
1262                 return;
1263
1264         tx_entries = tx_pool->num_buffers;
1265
1266         for (i = 0; i < tx_entries; i++) {
1267                 tx_buff = &tx_pool->tx_buff[i];
1268                 if (tx_buff && tx_buff->skb) {
1269                         dev_kfree_skb_any(tx_buff->skb);
1270                         tx_buff->skb = NULL;
1271                 }
1272         }
1273 }
1274
1275 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1276 {
1277         int tx_scrqs;
1278         int i;
1279
1280         if (!adapter->tx_pool || !adapter->tso_pool)
1281                 return;
1282
1283         tx_scrqs = adapter->num_active_tx_pools;
1284
1285         /* Free any remaining skbs in the tx buffer pools */
1286         for (i = 0; i < tx_scrqs; i++) {
1287                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1288                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1289                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1290         }
1291 }
1292
1293 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1294 {
1295         struct net_device *netdev = adapter->netdev;
1296         int i;
1297
1298         if (adapter->tx_scrq) {
1299                 for (i = 0; i < adapter->req_tx_queues; i++)
1300                         if (adapter->tx_scrq[i]->irq) {
1301                                 netdev_dbg(netdev,
1302                                            "Disabling tx_scrq[%d] irq\n", i);
1303                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1304                                 disable_irq(adapter->tx_scrq[i]->irq);
1305                         }
1306         }
1307
1308         if (adapter->rx_scrq) {
1309                 for (i = 0; i < adapter->req_rx_queues; i++) {
1310                         if (adapter->rx_scrq[i]->irq) {
1311                                 netdev_dbg(netdev,
1312                                            "Disabling rx_scrq[%d] irq\n", i);
1313                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1314                                 disable_irq(adapter->rx_scrq[i]->irq);
1315                         }
1316                 }
1317         }
1318 }
1319
1320 static void ibmvnic_cleanup(struct net_device *netdev)
1321 {
1322         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1323
1324         /* ensure that transmissions are stopped if called by do_reset */
1325         if (test_bit(0, &adapter->resetting))
1326                 netif_tx_disable(netdev);
1327         else
1328                 netif_tx_stop_all_queues(netdev);
1329
1330         ibmvnic_napi_disable(adapter);
1331         ibmvnic_disable_irqs(adapter);
1332
1333         clean_rx_pools(adapter);
1334         clean_tx_pools(adapter);
1335 }
1336
1337 static int __ibmvnic_close(struct net_device *netdev)
1338 {
1339         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1340         int rc = 0;
1341
1342         adapter->state = VNIC_CLOSING;
1343         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1344         adapter->state = VNIC_CLOSED;
1345         return rc;
1346 }
1347
1348 static int ibmvnic_close(struct net_device *netdev)
1349 {
1350         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1351         int rc;
1352
1353         netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1354                    adapter->state, adapter->failover_pending,
1355                    adapter->force_reset_recovery);
1356
1357         /* If device failover is pending, just set device state and return.
1358          * Device operation will be handled by reset routine.
1359          */
1360         if (adapter->failover_pending) {
1361                 adapter->state = VNIC_CLOSED;
1362                 return 0;
1363         }
1364
1365         rc = __ibmvnic_close(netdev);
1366         ibmvnic_cleanup(netdev);
1367
1368         return rc;
1369 }
1370
1371 /**
1372  * build_hdr_data - creates L2/L3/L4 header data buffer
1373  * @hdr_field: bitfield determining needed headers
1374  * @skb: socket buffer
1375  * @hdr_len: array of header lengths
1376  * @hdr_data: buffer to write the header to
1377  *
1378  * Reads hdr_field to determine which headers are needed by firmware.
1379  * Builds a buffer containing these headers.  Saves individual header
1380  * lengths and total buffer length to be used to build descriptors.
1381  */
1382 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1383                           int *hdr_len, u8 *hdr_data)
1384 {
1385         int len = 0;
1386         u8 *hdr;
1387
1388         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1389                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1390         else
1391                 hdr_len[0] = sizeof(struct ethhdr);
1392
1393         if (skb->protocol == htons(ETH_P_IP)) {
1394                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1395                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1396                         hdr_len[2] = tcp_hdrlen(skb);
1397                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1398                         hdr_len[2] = sizeof(struct udphdr);
1399         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1400                 hdr_len[1] = sizeof(struct ipv6hdr);
1401                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1402                         hdr_len[2] = tcp_hdrlen(skb);
1403                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1404                         hdr_len[2] = sizeof(struct udphdr);
1405         } else if (skb->protocol == htons(ETH_P_ARP)) {
1406                 hdr_len[1] = arp_hdr_len(skb->dev);
1407                 hdr_len[2] = 0;
1408         }
1409
1410         memset(hdr_data, 0, 120);
1411         if ((hdr_field >> 6) & 1) {
1412                 hdr = skb_mac_header(skb);
1413                 memcpy(hdr_data, hdr, hdr_len[0]);
1414                 len += hdr_len[0];
1415         }
1416
1417         if ((hdr_field >> 5) & 1) {
1418                 hdr = skb_network_header(skb);
1419                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1420                 len += hdr_len[1];
1421         }
1422
1423         if ((hdr_field >> 4) & 1) {
1424                 hdr = skb_transport_header(skb);
1425                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1426                 len += hdr_len[2];
1427         }
1428         return len;
1429 }
1430
1431 /**
1432  * create_hdr_descs - create header and header extension descriptors
1433  * @hdr_field: bitfield determining needed headers
1434  * @hdr_data: buffer containing header data
1435  * @len: length of data buffer
1436  * @hdr_len: array of individual header lengths
1437  * @scrq_arr: descriptor array
1438  *
1439  * Creates header and, if needed, header extension descriptors and
1440  * places them in a descriptor array, scrq_arr
1441  */
1442
1443 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1444                             union sub_crq *scrq_arr)
1445 {
1446         union sub_crq hdr_desc;
1447         int tmp_len = len;
1448         int num_descs = 0;
1449         u8 *data, *cur;
1450         int tmp;
1451
1452         while (tmp_len > 0) {
1453                 cur = hdr_data + len - tmp_len;
1454
1455                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1456                 if (cur != hdr_data) {
1457                         data = hdr_desc.hdr_ext.data;
1458                         tmp = tmp_len > 29 ? 29 : tmp_len;
1459                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1460                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1461                         hdr_desc.hdr_ext.len = tmp;
1462                 } else {
1463                         data = hdr_desc.hdr.data;
1464                         tmp = tmp_len > 24 ? 24 : tmp_len;
1465                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1466                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1467                         hdr_desc.hdr.len = tmp;
1468                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1469                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1470                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1471                         hdr_desc.hdr.flag = hdr_field << 1;
1472                 }
1473                 memcpy(data, cur, tmp);
1474                 tmp_len -= tmp;
1475                 *scrq_arr = hdr_desc;
1476                 scrq_arr++;
1477                 num_descs++;
1478         }
1479
1480         return num_descs;
1481 }
1482
1483 /**
1484  * build_hdr_descs_arr - build a header descriptor array
1485  * @txbuff: tx buffer
1486  * @num_entries: number of descriptors to be sent
1487  * @hdr_field: bit field determining which headers will be sent
1488  *
1489  * This function will build a TX descriptor array with applicable
1490  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1491  */
1492
1493 static void build_hdr_descs_arr(struct sk_buff *skb,
1494                                 union sub_crq *indir_arr,
1495                                 int *num_entries, u8 hdr_field)
1496 {
1497         int hdr_len[3] = {0, 0, 0};
1498         u8 hdr_data[140] = {0};
1499         int tot_len;
1500
1501         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1502                                  hdr_data);
1503         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1504                                          indir_arr + 1);
1505 }
1506
1507 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1508                                     struct net_device *netdev)
1509 {
1510         /* For some backing devices, mishandling of small packets
1511          * can result in a loss of connection or TX stall. Device
1512          * architects recommend that no packet should be smaller
1513          * than the minimum MTU value provided to the driver, so
1514          * pad any packets to that length
1515          */
1516         if (skb->len < netdev->min_mtu)
1517                 return skb_put_padto(skb, netdev->min_mtu);
1518
1519         return 0;
1520 }
1521
1522 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1523                                          struct ibmvnic_sub_crq_queue *tx_scrq)
1524 {
1525         struct ibmvnic_ind_xmit_queue *ind_bufp;
1526         struct ibmvnic_tx_buff *tx_buff;
1527         struct ibmvnic_tx_pool *tx_pool;
1528         union sub_crq tx_scrq_entry;
1529         int queue_num;
1530         int entries;
1531         int index;
1532         int i;
1533
1534         ind_bufp = &tx_scrq->ind_buf;
1535         entries = (u64)ind_bufp->index;
1536         queue_num = tx_scrq->pool_index;
1537
1538         for (i = entries - 1; i >= 0; --i) {
1539                 tx_scrq_entry = ind_bufp->indir_arr[i];
1540                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1541                         continue;
1542                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1543                 if (index & IBMVNIC_TSO_POOL_MASK) {
1544                         tx_pool = &adapter->tso_pool[queue_num];
1545                         index &= ~IBMVNIC_TSO_POOL_MASK;
1546                 } else {
1547                         tx_pool = &adapter->tx_pool[queue_num];
1548                 }
1549                 tx_pool->free_map[tx_pool->consumer_index] = index;
1550                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1551                                           tx_pool->num_buffers - 1 :
1552                                           tx_pool->consumer_index - 1;
1553                 tx_buff = &tx_pool->tx_buff[index];
1554                 adapter->netdev->stats.tx_packets--;
1555                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1556                 adapter->tx_stats_buffers[queue_num].packets--;
1557                 adapter->tx_stats_buffers[queue_num].bytes -=
1558                                                 tx_buff->skb->len;
1559                 dev_kfree_skb_any(tx_buff->skb);
1560                 tx_buff->skb = NULL;
1561                 adapter->netdev->stats.tx_dropped++;
1562         }
1563         ind_bufp->index = 0;
1564         if (atomic_sub_return(entries, &tx_scrq->used) <=
1565             (adapter->req_tx_entries_per_subcrq / 2) &&
1566             __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1567                 netif_wake_subqueue(adapter->netdev, queue_num);
1568                 netdev_dbg(adapter->netdev, "Started queue %d\n",
1569                            queue_num);
1570         }
1571 }
1572
1573 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1574                                  struct ibmvnic_sub_crq_queue *tx_scrq)
1575 {
1576         struct ibmvnic_ind_xmit_queue *ind_bufp;
1577         u64 dma_addr;
1578         u64 entries;
1579         u64 handle;
1580         int rc;
1581
1582         ind_bufp = &tx_scrq->ind_buf;
1583         dma_addr = (u64)ind_bufp->indir_dma;
1584         entries = (u64)ind_bufp->index;
1585         handle = tx_scrq->handle;
1586
1587         if (!entries)
1588                 return 0;
1589         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1590         if (rc)
1591                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1592         else
1593                 ind_bufp->index = 0;
1594         return 0;
1595 }
1596
1597 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1598 {
1599         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1600         int queue_num = skb_get_queue_mapping(skb);
1601         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1602         struct device *dev = &adapter->vdev->dev;
1603         struct ibmvnic_ind_xmit_queue *ind_bufp;
1604         struct ibmvnic_tx_buff *tx_buff = NULL;
1605         struct ibmvnic_sub_crq_queue *tx_scrq;
1606         struct ibmvnic_tx_pool *tx_pool;
1607         unsigned int tx_send_failed = 0;
1608         netdev_tx_t ret = NETDEV_TX_OK;
1609         unsigned int tx_map_failed = 0;
1610         union sub_crq indir_arr[16];
1611         unsigned int tx_dropped = 0;
1612         unsigned int tx_packets = 0;
1613         unsigned int tx_bytes = 0;
1614         dma_addr_t data_dma_addr;
1615         struct netdev_queue *txq;
1616         unsigned long lpar_rc;
1617         union sub_crq tx_crq;
1618         unsigned int offset;
1619         int num_entries = 1;
1620         unsigned char *dst;
1621         int index = 0;
1622         u8 proto = 0;
1623
1624         tx_scrq = adapter->tx_scrq[queue_num];
1625         txq = netdev_get_tx_queue(netdev, queue_num);
1626         ind_bufp = &tx_scrq->ind_buf;
1627
1628         if (test_bit(0, &adapter->resetting)) {
1629                 if (!netif_subqueue_stopped(netdev, skb))
1630                         netif_stop_subqueue(netdev, queue_num);
1631                 dev_kfree_skb_any(skb);
1632
1633                 tx_send_failed++;
1634                 tx_dropped++;
1635                 ret = NETDEV_TX_OK;
1636                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1637                 goto out;
1638         }
1639
1640         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1641                 tx_dropped++;
1642                 tx_send_failed++;
1643                 ret = NETDEV_TX_OK;
1644                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1645                 goto out;
1646         }
1647         if (skb_is_gso(skb))
1648                 tx_pool = &adapter->tso_pool[queue_num];
1649         else
1650                 tx_pool = &adapter->tx_pool[queue_num];
1651
1652         index = tx_pool->free_map[tx_pool->consumer_index];
1653
1654         if (index == IBMVNIC_INVALID_MAP) {
1655                 dev_kfree_skb_any(skb);
1656                 tx_send_failed++;
1657                 tx_dropped++;
1658                 ret = NETDEV_TX_OK;
1659                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1660                 goto out;
1661         }
1662
1663         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1664
1665         offset = index * tx_pool->buf_size;
1666         dst = tx_pool->long_term_buff.buff + offset;
1667         memset(dst, 0, tx_pool->buf_size);
1668         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1669
1670         if (skb_shinfo(skb)->nr_frags) {
1671                 int cur, i;
1672
1673                 /* Copy the head */
1674                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1675                 cur = skb_headlen(skb);
1676
1677                 /* Copy the frags */
1678                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1679                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1680
1681                         memcpy(dst + cur,
1682                                page_address(skb_frag_page(frag)) +
1683                                skb_frag_off(frag), skb_frag_size(frag));
1684                         cur += skb_frag_size(frag);
1685                 }
1686         } else {
1687                 skb_copy_from_linear_data(skb, dst, skb->len);
1688         }
1689
1690         /* post changes to long_term_buff *dst before VIOS accessing it */
1691         dma_wmb();
1692
1693         tx_pool->consumer_index =
1694             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1695
1696         tx_buff = &tx_pool->tx_buff[index];
1697         tx_buff->skb = skb;
1698         tx_buff->index = index;
1699         tx_buff->pool_index = queue_num;
1700
1701         memset(&tx_crq, 0, sizeof(tx_crq));
1702         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1703         tx_crq.v1.type = IBMVNIC_TX_DESC;
1704         tx_crq.v1.n_crq_elem = 1;
1705         tx_crq.v1.n_sge = 1;
1706         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1707
1708         if (skb_is_gso(skb))
1709                 tx_crq.v1.correlator =
1710                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1711         else
1712                 tx_crq.v1.correlator = cpu_to_be32(index);
1713         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1714         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1715         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1716
1717         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1718                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1719                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1720         }
1721
1722         if (skb->protocol == htons(ETH_P_IP)) {
1723                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1724                 proto = ip_hdr(skb)->protocol;
1725         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1726                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1727                 proto = ipv6_hdr(skb)->nexthdr;
1728         }
1729
1730         if (proto == IPPROTO_TCP)
1731                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1732         else if (proto == IPPROTO_UDP)
1733                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1734
1735         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1736                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1737                 hdrs += 2;
1738         }
1739         if (skb_is_gso(skb)) {
1740                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1741                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1742                 hdrs += 2;
1743         }
1744
1745         if ((*hdrs >> 7) & 1)
1746                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1747
1748         tx_crq.v1.n_crq_elem = num_entries;
1749         tx_buff->num_entries = num_entries;
1750         /* flush buffer if current entry can not fit */
1751         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1752                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1753                 if (lpar_rc != H_SUCCESS)
1754                         goto tx_flush_err;
1755         }
1756
1757         indir_arr[0] = tx_crq;
1758         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1759                num_entries * sizeof(struct ibmvnic_generic_scrq));
1760         ind_bufp->index += num_entries;
1761         if (__netdev_tx_sent_queue(txq, skb->len,
1762                                    netdev_xmit_more() &&
1763                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1764                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1765                 if (lpar_rc != H_SUCCESS)
1766                         goto tx_err;
1767         }
1768
1769         if (atomic_add_return(num_entries, &tx_scrq->used)
1770                                         >= adapter->req_tx_entries_per_subcrq) {
1771                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1772                 netif_stop_subqueue(netdev, queue_num);
1773         }
1774
1775         tx_packets++;
1776         tx_bytes += skb->len;
1777         txq->trans_start = jiffies;
1778         ret = NETDEV_TX_OK;
1779         goto out;
1780
1781 tx_flush_err:
1782         dev_kfree_skb_any(skb);
1783         tx_buff->skb = NULL;
1784         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1785                                   tx_pool->num_buffers - 1 :
1786                                   tx_pool->consumer_index - 1;
1787         tx_dropped++;
1788 tx_err:
1789         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1790                 dev_err_ratelimited(dev, "tx: send failed\n");
1791
1792         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1793                 /* Disable TX and report carrier off if queue is closed
1794                  * or pending failover.
1795                  * Firmware guarantees that a signal will be sent to the
1796                  * driver, triggering a reset or some other action.
1797                  */
1798                 netif_tx_stop_all_queues(netdev);
1799                 netif_carrier_off(netdev);
1800         }
1801 out:
1802         netdev->stats.tx_dropped += tx_dropped;
1803         netdev->stats.tx_bytes += tx_bytes;
1804         netdev->stats.tx_packets += tx_packets;
1805         adapter->tx_send_failed += tx_send_failed;
1806         adapter->tx_map_failed += tx_map_failed;
1807         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1808         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1809         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1810
1811         return ret;
1812 }
1813
1814 static void ibmvnic_set_multi(struct net_device *netdev)
1815 {
1816         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1817         struct netdev_hw_addr *ha;
1818         union ibmvnic_crq crq;
1819
1820         memset(&crq, 0, sizeof(crq));
1821         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1822         crq.request_capability.cmd = REQUEST_CAPABILITY;
1823
1824         if (netdev->flags & IFF_PROMISC) {
1825                 if (!adapter->promisc_supported)
1826                         return;
1827         } else {
1828                 if (netdev->flags & IFF_ALLMULTI) {
1829                         /* Accept all multicast */
1830                         memset(&crq, 0, sizeof(crq));
1831                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1832                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1833                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1834                         ibmvnic_send_crq(adapter, &crq);
1835                 } else if (netdev_mc_empty(netdev)) {
1836                         /* Reject all multicast */
1837                         memset(&crq, 0, sizeof(crq));
1838                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1839                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1840                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1841                         ibmvnic_send_crq(adapter, &crq);
1842                 } else {
1843                         /* Accept one or more multicast(s) */
1844                         netdev_for_each_mc_addr(ha, netdev) {
1845                                 memset(&crq, 0, sizeof(crq));
1846                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1847                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1848                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1849                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1850                                                 ha->addr);
1851                                 ibmvnic_send_crq(adapter, &crq);
1852                         }
1853                 }
1854         }
1855 }
1856
1857 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1858 {
1859         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1860         union ibmvnic_crq crq;
1861         int rc;
1862
1863         if (!is_valid_ether_addr(dev_addr)) {
1864                 rc = -EADDRNOTAVAIL;
1865                 goto err;
1866         }
1867
1868         memset(&crq, 0, sizeof(crq));
1869         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1870         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1871         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1872
1873         mutex_lock(&adapter->fw_lock);
1874         adapter->fw_done_rc = 0;
1875         reinit_completion(&adapter->fw_done);
1876
1877         rc = ibmvnic_send_crq(adapter, &crq);
1878         if (rc) {
1879                 rc = -EIO;
1880                 mutex_unlock(&adapter->fw_lock);
1881                 goto err;
1882         }
1883
1884         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1885         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1886         if (rc || adapter->fw_done_rc) {
1887                 rc = -EIO;
1888                 mutex_unlock(&adapter->fw_lock);
1889                 goto err;
1890         }
1891         mutex_unlock(&adapter->fw_lock);
1892         return 0;
1893 err:
1894         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1895         return rc;
1896 }
1897
1898 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1899 {
1900         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1901         struct sockaddr *addr = p;
1902         int rc;
1903
1904         rc = 0;
1905         if (!is_valid_ether_addr(addr->sa_data))
1906                 return -EADDRNOTAVAIL;
1907
1908         ether_addr_copy(adapter->mac_addr, addr->sa_data);
1909         if (adapter->state != VNIC_PROBED)
1910                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1911
1912         return rc;
1913 }
1914
1915 /*
1916  * do_reset returns zero if we are able to keep processing reset events, or
1917  * non-zero if we hit a fatal error and must halt.
1918  */
1919 static int do_reset(struct ibmvnic_adapter *adapter,
1920                     struct ibmvnic_rwi *rwi, u32 reset_state)
1921 {
1922         u64 old_num_rx_queues, old_num_tx_queues;
1923         u64 old_num_rx_slots, old_num_tx_slots;
1924         struct net_device *netdev = adapter->netdev;
1925         int i, rc;
1926
1927         netdev_dbg(adapter->netdev,
1928                    "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1929                    adapter->state, adapter->failover_pending,
1930                    rwi->reset_reason, reset_state);
1931
1932         adapter->reset_reason = rwi->reset_reason;
1933         /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
1934         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
1935                 rtnl_lock();
1936
1937         /* Now that we have the rtnl lock, clear any pending failover.
1938          * This will ensure ibmvnic_open() has either completed or will
1939          * block until failover is complete.
1940          */
1941         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1942                 adapter->failover_pending = false;
1943
1944         /* read the state and check (again) after getting rtnl */
1945         reset_state = adapter->state;
1946
1947         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
1948                 rc = -EBUSY;
1949                 goto out;
1950         }
1951
1952         netif_carrier_off(netdev);
1953
1954         old_num_rx_queues = adapter->req_rx_queues;
1955         old_num_tx_queues = adapter->req_tx_queues;
1956         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1957         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1958
1959         ibmvnic_cleanup(netdev);
1960
1961         if (reset_state == VNIC_OPEN &&
1962             adapter->reset_reason != VNIC_RESET_MOBILITY &&
1963             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1964                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
1965                         rc = __ibmvnic_close(netdev);
1966                         if (rc)
1967                                 goto out;
1968                 } else {
1969                         adapter->state = VNIC_CLOSING;
1970
1971                         /* Release the RTNL lock before link state change and
1972                          * re-acquire after the link state change to allow
1973                          * linkwatch_event to grab the RTNL lock and run during
1974                          * a reset.
1975                          */
1976                         rtnl_unlock();
1977                         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1978                         rtnl_lock();
1979                         if (rc)
1980                                 goto out;
1981
1982                         if (adapter->state == VNIC_OPEN) {
1983                                 /* When we dropped rtnl, ibmvnic_open() got
1984                                  * it and noticed that we are resetting and
1985                                  * set the adapter state to OPEN. Update our
1986                                  * new "target" state, and resume the reset
1987                                  * from VNIC_CLOSING state.
1988                                  */
1989                                 netdev_dbg(netdev,
1990                                            "Open changed state from %d, updating.\n",
1991                                            reset_state);
1992                                 reset_state = VNIC_OPEN;
1993                                 adapter->state = VNIC_CLOSING;
1994                         }
1995
1996                         if (adapter->state != VNIC_CLOSING) {
1997                                 /* If someone else changed the adapter state
1998                                  * when we dropped the rtnl, fail the reset
1999                                  */
2000                                 rc = -1;
2001                                 goto out;
2002                         }
2003                         adapter->state = VNIC_CLOSED;
2004                 }
2005         }
2006
2007         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2008                 release_resources(adapter);
2009                 release_sub_crqs(adapter, 1);
2010                 release_crq_queue(adapter);
2011         }
2012
2013         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2014                 /* remove the closed state so when we call open it appears
2015                  * we are coming from the probed state.
2016                  */
2017                 adapter->state = VNIC_PROBED;
2018
2019                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2020                         rc = init_crq_queue(adapter);
2021                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2022                         rc = ibmvnic_reenable_crq_queue(adapter);
2023                         release_sub_crqs(adapter, 1);
2024                 } else {
2025                         rc = ibmvnic_reset_crq(adapter);
2026                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2027                                 rc = vio_enable_interrupts(adapter->vdev);
2028                                 if (rc)
2029                                         netdev_err(adapter->netdev,
2030                                                    "Reset failed to enable interrupts. rc=%d\n",
2031                                                    rc);
2032                         }
2033                 }
2034
2035                 if (rc) {
2036                         netdev_err(adapter->netdev,
2037                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2038                         goto out;
2039                 }
2040
2041                 rc = ibmvnic_reset_init(adapter, true);
2042                 if (rc) {
2043                         rc = IBMVNIC_INIT_FAILED;
2044                         goto out;
2045                 }
2046
2047                 /* If the adapter was in PROBE state prior to the reset,
2048                  * exit here.
2049                  */
2050                 if (reset_state == VNIC_PROBED) {
2051                         rc = 0;
2052                         goto out;
2053                 }
2054
2055                 rc = ibmvnic_login(netdev);
2056                 if (rc)
2057                         goto out;
2058
2059                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2060                         rc = init_resources(adapter);
2061                         if (rc)
2062                                 goto out;
2063                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2064                     adapter->req_tx_queues != old_num_tx_queues ||
2065                     adapter->req_rx_add_entries_per_subcrq !=
2066                     old_num_rx_slots ||
2067                     adapter->req_tx_entries_per_subcrq !=
2068                     old_num_tx_slots ||
2069                     !adapter->rx_pool ||
2070                     !adapter->tso_pool ||
2071                     !adapter->tx_pool) {
2072                         release_rx_pools(adapter);
2073                         release_tx_pools(adapter);
2074                         release_napi(adapter);
2075                         release_vpd_data(adapter);
2076
2077                         rc = init_resources(adapter);
2078                         if (rc)
2079                                 goto out;
2080
2081                 } else {
2082                         rc = reset_tx_pools(adapter);
2083                         if (rc) {
2084                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2085                                            rc);
2086                                 goto out;
2087                         }
2088
2089                         rc = reset_rx_pools(adapter);
2090                         if (rc) {
2091                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2092                                            rc);
2093                                 goto out;
2094                         }
2095                 }
2096                 ibmvnic_disable_irqs(adapter);
2097         }
2098         adapter->state = VNIC_CLOSED;
2099
2100         if (reset_state == VNIC_CLOSED) {
2101                 rc = 0;
2102                 goto out;
2103         }
2104
2105         rc = __ibmvnic_open(netdev);
2106         if (rc) {
2107                 rc = IBMVNIC_OPEN_FAILED;
2108                 goto out;
2109         }
2110
2111         /* refresh device's multicast list */
2112         ibmvnic_set_multi(netdev);
2113
2114         /* kick napi */
2115         for (i = 0; i < adapter->req_rx_queues; i++)
2116                 napi_schedule(&adapter->napi[i]);
2117
2118         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2119             adapter->reset_reason == VNIC_RESET_MOBILITY)
2120                 __netdev_notify_peers(netdev);
2121
2122         rc = 0;
2123
2124 out:
2125         /* restore the adapter state if reset failed */
2126         if (rc)
2127                 adapter->state = reset_state;
2128         /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2129         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2130                 rtnl_unlock();
2131
2132         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2133                    adapter->state, adapter->failover_pending, rc);
2134         return rc;
2135 }
2136
2137 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2138                          struct ibmvnic_rwi *rwi, u32 reset_state)
2139 {
2140         struct net_device *netdev = adapter->netdev;
2141         int rc;
2142
2143         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2144                    rwi->reset_reason);
2145
2146         /* read the state and check (again) after getting rtnl */
2147         reset_state = adapter->state;
2148
2149         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2150                 rc = -EBUSY;
2151                 goto out;
2152         }
2153
2154         netif_carrier_off(netdev);
2155         adapter->reset_reason = rwi->reset_reason;
2156
2157         ibmvnic_cleanup(netdev);
2158         release_resources(adapter);
2159         release_sub_crqs(adapter, 0);
2160         release_crq_queue(adapter);
2161
2162         /* remove the closed state so when we call open it appears
2163          * we are coming from the probed state.
2164          */
2165         adapter->state = VNIC_PROBED;
2166
2167         reinit_completion(&adapter->init_done);
2168         rc = init_crq_queue(adapter);
2169         if (rc) {
2170                 netdev_err(adapter->netdev,
2171                            "Couldn't initialize crq. rc=%d\n", rc);
2172                 goto out;
2173         }
2174
2175         rc = ibmvnic_reset_init(adapter, false);
2176         if (rc)
2177                 goto out;
2178
2179         /* If the adapter was in PROBE state prior to the reset,
2180          * exit here.
2181          */
2182         if (reset_state == VNIC_PROBED)
2183                 goto out;
2184
2185         rc = ibmvnic_login(netdev);
2186         if (rc)
2187                 goto out;
2188
2189         rc = init_resources(adapter);
2190         if (rc)
2191                 goto out;
2192
2193         ibmvnic_disable_irqs(adapter);
2194         adapter->state = VNIC_CLOSED;
2195
2196         if (reset_state == VNIC_CLOSED)
2197                 goto out;
2198
2199         rc = __ibmvnic_open(netdev);
2200         if (rc) {
2201                 rc = IBMVNIC_OPEN_FAILED;
2202                 goto out;
2203         }
2204
2205         __netdev_notify_peers(netdev);
2206 out:
2207         /* restore adapter state if reset failed */
2208         if (rc)
2209                 adapter->state = reset_state;
2210         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2211                    adapter->state, adapter->failover_pending, rc);
2212         return rc;
2213 }
2214
2215 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2216 {
2217         struct ibmvnic_rwi *rwi;
2218         unsigned long flags;
2219
2220         spin_lock_irqsave(&adapter->rwi_lock, flags);
2221
2222         if (!list_empty(&adapter->rwi_list)) {
2223                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2224                                        list);
2225                 list_del(&rwi->list);
2226         } else {
2227                 rwi = NULL;
2228         }
2229
2230         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2231         return rwi;
2232 }
2233
2234 static void __ibmvnic_reset(struct work_struct *work)
2235 {
2236         struct ibmvnic_rwi *rwi;
2237         struct ibmvnic_adapter *adapter;
2238         bool saved_state = false;
2239         unsigned long flags;
2240         u32 reset_state;
2241         int rc = 0;
2242
2243         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2244
2245         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2246                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2247                                       IBMVNIC_RESET_DELAY);
2248                 return;
2249         }
2250
2251         rwi = get_next_rwi(adapter);
2252         while (rwi) {
2253                 spin_lock_irqsave(&adapter->state_lock, flags);
2254
2255                 if (adapter->state == VNIC_REMOVING ||
2256                     adapter->state == VNIC_REMOVED) {
2257                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2258                         kfree(rwi);
2259                         rc = EBUSY;
2260                         break;
2261                 }
2262
2263                 if (!saved_state) {
2264                         reset_state = adapter->state;
2265                         saved_state = true;
2266                 }
2267                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2268
2269                 if (adapter->force_reset_recovery) {
2270                         /* Since we are doing a hard reset now, clear the
2271                          * failover_pending flag so we don't ignore any
2272                          * future MOBILITY or other resets.
2273                          */
2274                         adapter->failover_pending = false;
2275
2276                         /* Transport event occurred during previous reset */
2277                         if (adapter->wait_for_reset) {
2278                                 /* Previous was CHANGE_PARAM; caller locked */
2279                                 adapter->force_reset_recovery = false;
2280                                 rc = do_hard_reset(adapter, rwi, reset_state);
2281                         } else {
2282                                 rtnl_lock();
2283                                 adapter->force_reset_recovery = false;
2284                                 rc = do_hard_reset(adapter, rwi, reset_state);
2285                                 rtnl_unlock();
2286                         }
2287                         if (rc) {
2288                                 /* give backing device time to settle down */
2289                                 netdev_dbg(adapter->netdev,
2290                                            "[S:%d] Hard reset failed, waiting 60 secs\n",
2291                                            adapter->state);
2292                                 set_current_state(TASK_UNINTERRUPTIBLE);
2293                                 schedule_timeout(60 * HZ);
2294                         }
2295                 } else {
2296                         rc = do_reset(adapter, rwi, reset_state);
2297                 }
2298                 kfree(rwi);
2299                 adapter->last_reset_time = jiffies;
2300
2301                 if (rc)
2302                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2303
2304                 rwi = get_next_rwi(adapter);
2305
2306                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2307                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2308                         adapter->force_reset_recovery = true;
2309         }
2310
2311         if (adapter->wait_for_reset) {
2312                 adapter->reset_done_rc = rc;
2313                 complete(&adapter->reset_done);
2314         }
2315
2316         clear_bit_unlock(0, &adapter->resetting);
2317
2318         netdev_dbg(adapter->netdev,
2319                    "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2320                    adapter->state, adapter->force_reset_recovery,
2321                    adapter->wait_for_reset);
2322 }
2323
2324 static void __ibmvnic_delayed_reset(struct work_struct *work)
2325 {
2326         struct ibmvnic_adapter *adapter;
2327
2328         adapter = container_of(work, struct ibmvnic_adapter,
2329                                ibmvnic_delayed_reset.work);
2330         __ibmvnic_reset(&adapter->ibmvnic_reset);
2331 }
2332
2333 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2334                          enum ibmvnic_reset_reason reason)
2335 {
2336         struct list_head *entry, *tmp_entry;
2337         struct ibmvnic_rwi *rwi, *tmp;
2338         struct net_device *netdev = adapter->netdev;
2339         unsigned long flags;
2340         int ret;
2341
2342         spin_lock_irqsave(&adapter->rwi_lock, flags);
2343
2344         /* If failover is pending don't schedule any other reset.
2345          * Instead let the failover complete. If there is already a
2346          * a failover reset scheduled, we will detect and drop the
2347          * duplicate reset when walking the ->rwi_list below.
2348          */
2349         if (adapter->state == VNIC_REMOVING ||
2350             adapter->state == VNIC_REMOVED ||
2351             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2352                 ret = EBUSY;
2353                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2354                 goto err;
2355         }
2356
2357         if (adapter->state == VNIC_PROBING) {
2358                 netdev_warn(netdev, "Adapter reset during probe\n");
2359                 adapter->init_done_rc = EAGAIN;
2360                 ret = EAGAIN;
2361                 goto err;
2362         }
2363
2364         list_for_each(entry, &adapter->rwi_list) {
2365                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2366                 if (tmp->reset_reason == reason) {
2367                         netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2368                                    reason);
2369                         ret = EBUSY;
2370                         goto err;
2371                 }
2372         }
2373
2374         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2375         if (!rwi) {
2376                 ret = ENOMEM;
2377                 goto err;
2378         }
2379         /* if we just received a transport event,
2380          * flush reset queue and process this reset
2381          */
2382         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2383                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2384                         list_del(entry);
2385         }
2386         rwi->reset_reason = reason;
2387         list_add_tail(&rwi->list, &adapter->rwi_list);
2388         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2389         schedule_work(&adapter->ibmvnic_reset);
2390
2391         ret = 0;
2392 err:
2393         /* ibmvnic_close() below can block, so drop the lock first */
2394         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2395
2396         if (ret == ENOMEM)
2397                 ibmvnic_close(netdev);
2398
2399         return -ret;
2400 }
2401
2402 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2403 {
2404         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2405
2406         if (test_bit(0, &adapter->resetting)) {
2407                 netdev_err(adapter->netdev,
2408                            "Adapter is resetting, skip timeout reset\n");
2409                 return;
2410         }
2411         /* No queuing up reset until at least 5 seconds (default watchdog val)
2412          * after last reset
2413          */
2414         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2415                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2416                 return;
2417         }
2418         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2419 }
2420
2421 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2422                                   struct ibmvnic_rx_buff *rx_buff)
2423 {
2424         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2425
2426         rx_buff->skb = NULL;
2427
2428         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2429         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2430
2431         atomic_dec(&pool->available);
2432 }
2433
2434 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2435 {
2436         struct ibmvnic_sub_crq_queue *rx_scrq;
2437         struct ibmvnic_adapter *adapter;
2438         struct net_device *netdev;
2439         int frames_processed;
2440         int scrq_num;
2441
2442         netdev = napi->dev;
2443         adapter = netdev_priv(netdev);
2444         scrq_num = (int)(napi - adapter->napi);
2445         frames_processed = 0;
2446         rx_scrq = adapter->rx_scrq[scrq_num];
2447
2448 restart_poll:
2449         while (frames_processed < budget) {
2450                 struct sk_buff *skb;
2451                 struct ibmvnic_rx_buff *rx_buff;
2452                 union sub_crq *next;
2453                 u32 length;
2454                 u16 offset;
2455                 u8 flags = 0;
2456
2457                 if (unlikely(test_bit(0, &adapter->resetting) &&
2458                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2459                         enable_scrq_irq(adapter, rx_scrq);
2460                         napi_complete_done(napi, frames_processed);
2461                         return frames_processed;
2462                 }
2463
2464                 if (!pending_scrq(adapter, rx_scrq))
2465                         break;
2466                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2467                 rx_buff = (struct ibmvnic_rx_buff *)
2468                           be64_to_cpu(next->rx_comp.correlator);
2469                 /* do error checking */
2470                 if (next->rx_comp.rc) {
2471                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2472                                    be16_to_cpu(next->rx_comp.rc));
2473                         /* free the entry */
2474                         next->rx_comp.first = 0;
2475                         dev_kfree_skb_any(rx_buff->skb);
2476                         remove_buff_from_pool(adapter, rx_buff);
2477                         continue;
2478                 } else if (!rx_buff->skb) {
2479                         /* free the entry */
2480                         next->rx_comp.first = 0;
2481                         remove_buff_from_pool(adapter, rx_buff);
2482                         continue;
2483                 }
2484
2485                 length = be32_to_cpu(next->rx_comp.len);
2486                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2487                 flags = next->rx_comp.flags;
2488                 skb = rx_buff->skb;
2489                 /* load long_term_buff before copying to skb */
2490                 dma_rmb();
2491                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2492                                         length);
2493
2494                 /* VLAN Header has been stripped by the system firmware and
2495                  * needs to be inserted by the driver
2496                  */
2497                 if (adapter->rx_vlan_header_insertion &&
2498                     (flags & IBMVNIC_VLAN_STRIPPED))
2499                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2500                                                ntohs(next->rx_comp.vlan_tci));
2501
2502                 /* free the entry */
2503                 next->rx_comp.first = 0;
2504                 remove_buff_from_pool(adapter, rx_buff);
2505
2506                 skb_put(skb, length);
2507                 skb->protocol = eth_type_trans(skb, netdev);
2508                 skb_record_rx_queue(skb, scrq_num);
2509
2510                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2511                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2512                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2513                 }
2514
2515                 length = skb->len;
2516                 napi_gro_receive(napi, skb); /* send it up */
2517                 netdev->stats.rx_packets++;
2518                 netdev->stats.rx_bytes += length;
2519                 adapter->rx_stats_buffers[scrq_num].packets++;
2520                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2521                 frames_processed++;
2522         }
2523
2524         if (adapter->state != VNIC_CLOSING &&
2525             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2526               adapter->req_rx_add_entries_per_subcrq / 2) ||
2527               frames_processed < budget))
2528                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2529         if (frames_processed < budget) {
2530                 if (napi_complete_done(napi, frames_processed)) {
2531                         enable_scrq_irq(adapter, rx_scrq);
2532                         if (pending_scrq(adapter, rx_scrq)) {
2533                                 if (napi_reschedule(napi)) {
2534                                         disable_scrq_irq(adapter, rx_scrq);
2535                                         goto restart_poll;
2536                                 }
2537                         }
2538                 }
2539         }
2540         return frames_processed;
2541 }
2542
2543 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2544 {
2545         int rc, ret;
2546
2547         adapter->fallback.mtu = adapter->req_mtu;
2548         adapter->fallback.rx_queues = adapter->req_rx_queues;
2549         adapter->fallback.tx_queues = adapter->req_tx_queues;
2550         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2551         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2552
2553         reinit_completion(&adapter->reset_done);
2554         adapter->wait_for_reset = true;
2555         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2556
2557         if (rc) {
2558                 ret = rc;
2559                 goto out;
2560         }
2561         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2562         if (rc) {
2563                 ret = -ENODEV;
2564                 goto out;
2565         }
2566
2567         ret = 0;
2568         if (adapter->reset_done_rc) {
2569                 ret = -EIO;
2570                 adapter->desired.mtu = adapter->fallback.mtu;
2571                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2572                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2573                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2574                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2575
2576                 reinit_completion(&adapter->reset_done);
2577                 adapter->wait_for_reset = true;
2578                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2579                 if (rc) {
2580                         ret = rc;
2581                         goto out;
2582                 }
2583                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2584                                                  60000);
2585                 if (rc) {
2586                         ret = -ENODEV;
2587                         goto out;
2588                 }
2589         }
2590 out:
2591         adapter->wait_for_reset = false;
2592
2593         return ret;
2594 }
2595
2596 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2597 {
2598         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2599
2600         adapter->desired.mtu = new_mtu + ETH_HLEN;
2601
2602         return wait_for_reset(adapter);
2603 }
2604
2605 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2606                                                 struct net_device *dev,
2607                                                 netdev_features_t features)
2608 {
2609         /* Some backing hardware adapters can not
2610          * handle packets with a MSS less than 224
2611          * or with only one segment.
2612          */
2613         if (skb_is_gso(skb)) {
2614                 if (skb_shinfo(skb)->gso_size < 224 ||
2615                     skb_shinfo(skb)->gso_segs == 1)
2616                         features &= ~NETIF_F_GSO_MASK;
2617         }
2618
2619         return features;
2620 }
2621
2622 static const struct net_device_ops ibmvnic_netdev_ops = {
2623         .ndo_open               = ibmvnic_open,
2624         .ndo_stop               = ibmvnic_close,
2625         .ndo_start_xmit         = ibmvnic_xmit,
2626         .ndo_set_rx_mode        = ibmvnic_set_multi,
2627         .ndo_set_mac_address    = ibmvnic_set_mac,
2628         .ndo_validate_addr      = eth_validate_addr,
2629         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2630         .ndo_change_mtu         = ibmvnic_change_mtu,
2631         .ndo_features_check     = ibmvnic_features_check,
2632 };
2633
2634 /* ethtool functions */
2635
2636 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2637                                       struct ethtool_link_ksettings *cmd)
2638 {
2639         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2640         int rc;
2641
2642         rc = send_query_phys_parms(adapter);
2643         if (rc) {
2644                 adapter->speed = SPEED_UNKNOWN;
2645                 adapter->duplex = DUPLEX_UNKNOWN;
2646         }
2647         cmd->base.speed = adapter->speed;
2648         cmd->base.duplex = adapter->duplex;
2649         cmd->base.port = PORT_FIBRE;
2650         cmd->base.phy_address = 0;
2651         cmd->base.autoneg = AUTONEG_ENABLE;
2652
2653         return 0;
2654 }
2655
2656 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2657                                 struct ethtool_drvinfo *info)
2658 {
2659         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2660
2661         strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2662         strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2663         strscpy(info->fw_version, adapter->fw_version,
2664                 sizeof(info->fw_version));
2665 }
2666
2667 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2668 {
2669         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2670
2671         return adapter->msg_enable;
2672 }
2673
2674 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2675 {
2676         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2677
2678         adapter->msg_enable = data;
2679 }
2680
2681 static u32 ibmvnic_get_link(struct net_device *netdev)
2682 {
2683         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2684
2685         /* Don't need to send a query because we request a logical link up at
2686          * init and then we wait for link state indications
2687          */
2688         return adapter->logical_link_state;
2689 }
2690
2691 static void ibmvnic_get_ringparam(struct net_device *netdev,
2692                                   struct ethtool_ringparam *ring)
2693 {
2694         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2695
2696         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2697                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2698                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2699         } else {
2700                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2701                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2702         }
2703         ring->rx_mini_max_pending = 0;
2704         ring->rx_jumbo_max_pending = 0;
2705         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2706         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2707         ring->rx_mini_pending = 0;
2708         ring->rx_jumbo_pending = 0;
2709 }
2710
2711 static int ibmvnic_set_ringparam(struct net_device *netdev,
2712                                  struct ethtool_ringparam *ring)
2713 {
2714         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2715         int ret;
2716
2717         ret = 0;
2718         adapter->desired.rx_entries = ring->rx_pending;
2719         adapter->desired.tx_entries = ring->tx_pending;
2720
2721         ret = wait_for_reset(adapter);
2722
2723         if (!ret &&
2724             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2725              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2726                 netdev_info(netdev,
2727                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2728                             ring->rx_pending, ring->tx_pending,
2729                             adapter->req_rx_add_entries_per_subcrq,
2730                             adapter->req_tx_entries_per_subcrq);
2731         return ret;
2732 }
2733
2734 static void ibmvnic_get_channels(struct net_device *netdev,
2735                                  struct ethtool_channels *channels)
2736 {
2737         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2738
2739         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2740                 channels->max_rx = adapter->max_rx_queues;
2741                 channels->max_tx = adapter->max_tx_queues;
2742         } else {
2743                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2744                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2745         }
2746
2747         channels->max_other = 0;
2748         channels->max_combined = 0;
2749         channels->rx_count = adapter->req_rx_queues;
2750         channels->tx_count = adapter->req_tx_queues;
2751         channels->other_count = 0;
2752         channels->combined_count = 0;
2753 }
2754
2755 static int ibmvnic_set_channels(struct net_device *netdev,
2756                                 struct ethtool_channels *channels)
2757 {
2758         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2759         int ret;
2760
2761         ret = 0;
2762         adapter->desired.rx_queues = channels->rx_count;
2763         adapter->desired.tx_queues = channels->tx_count;
2764
2765         ret = wait_for_reset(adapter);
2766
2767         if (!ret &&
2768             (adapter->req_rx_queues != channels->rx_count ||
2769              adapter->req_tx_queues != channels->tx_count))
2770                 netdev_info(netdev,
2771                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2772                             channels->rx_count, channels->tx_count,
2773                             adapter->req_rx_queues, adapter->req_tx_queues);
2774         return ret;
2775 }
2776
2777 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2778 {
2779         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2780         int i;
2781
2782         switch (stringset) {
2783         case ETH_SS_STATS:
2784                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2785                                 i++, data += ETH_GSTRING_LEN)
2786                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2787
2788                 for (i = 0; i < adapter->req_tx_queues; i++) {
2789                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2790                         data += ETH_GSTRING_LEN;
2791
2792                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2793                         data += ETH_GSTRING_LEN;
2794
2795                         snprintf(data, ETH_GSTRING_LEN,
2796                                  "tx%d_dropped_packets", i);
2797                         data += ETH_GSTRING_LEN;
2798                 }
2799
2800                 for (i = 0; i < adapter->req_rx_queues; i++) {
2801                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2802                         data += ETH_GSTRING_LEN;
2803
2804                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2805                         data += ETH_GSTRING_LEN;
2806
2807                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2808                         data += ETH_GSTRING_LEN;
2809                 }
2810                 break;
2811
2812         case ETH_SS_PRIV_FLAGS:
2813                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2814                         strcpy(data + i * ETH_GSTRING_LEN,
2815                                ibmvnic_priv_flags[i]);
2816                 break;
2817         default:
2818                 return;
2819         }
2820 }
2821
2822 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2823 {
2824         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2825
2826         switch (sset) {
2827         case ETH_SS_STATS:
2828                 return ARRAY_SIZE(ibmvnic_stats) +
2829                        adapter->req_tx_queues * NUM_TX_STATS +
2830                        adapter->req_rx_queues * NUM_RX_STATS;
2831         case ETH_SS_PRIV_FLAGS:
2832                 return ARRAY_SIZE(ibmvnic_priv_flags);
2833         default:
2834                 return -EOPNOTSUPP;
2835         }
2836 }
2837
2838 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2839                                       struct ethtool_stats *stats, u64 *data)
2840 {
2841         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2842         union ibmvnic_crq crq;
2843         int i, j;
2844         int rc;
2845
2846         memset(&crq, 0, sizeof(crq));
2847         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2848         crq.request_statistics.cmd = REQUEST_STATISTICS;
2849         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2850         crq.request_statistics.len =
2851             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2852
2853         /* Wait for data to be written */
2854         reinit_completion(&adapter->stats_done);
2855         rc = ibmvnic_send_crq(adapter, &crq);
2856         if (rc)
2857                 return;
2858         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2859         if (rc)
2860                 return;
2861
2862         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2863                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
2864                                       (adapter, ibmvnic_stats[i].offset));
2865
2866         for (j = 0; j < adapter->req_tx_queues; j++) {
2867                 data[i] = adapter->tx_stats_buffers[j].packets;
2868                 i++;
2869                 data[i] = adapter->tx_stats_buffers[j].bytes;
2870                 i++;
2871                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2872                 i++;
2873         }
2874
2875         for (j = 0; j < adapter->req_rx_queues; j++) {
2876                 data[i] = adapter->rx_stats_buffers[j].packets;
2877                 i++;
2878                 data[i] = adapter->rx_stats_buffers[j].bytes;
2879                 i++;
2880                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2881                 i++;
2882         }
2883 }
2884
2885 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2886 {
2887         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2888
2889         return adapter->priv_flags;
2890 }
2891
2892 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2893 {
2894         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2895         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2896
2897         if (which_maxes)
2898                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2899         else
2900                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2901
2902         return 0;
2903 }
2904
2905 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2906         .get_drvinfo            = ibmvnic_get_drvinfo,
2907         .get_msglevel           = ibmvnic_get_msglevel,
2908         .set_msglevel           = ibmvnic_set_msglevel,
2909         .get_link               = ibmvnic_get_link,
2910         .get_ringparam          = ibmvnic_get_ringparam,
2911         .set_ringparam          = ibmvnic_set_ringparam,
2912         .get_channels           = ibmvnic_get_channels,
2913         .set_channels           = ibmvnic_set_channels,
2914         .get_strings            = ibmvnic_get_strings,
2915         .get_sset_count         = ibmvnic_get_sset_count,
2916         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2917         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2918         .get_priv_flags         = ibmvnic_get_priv_flags,
2919         .set_priv_flags         = ibmvnic_set_priv_flags,
2920 };
2921
2922 /* Routines for managing CRQs/sCRQs  */
2923
2924 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2925                                    struct ibmvnic_sub_crq_queue *scrq)
2926 {
2927         int rc;
2928
2929         if (!scrq) {
2930                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2931                 return -EINVAL;
2932         }
2933
2934         if (scrq->irq) {
2935                 free_irq(scrq->irq, scrq);
2936                 irq_dispose_mapping(scrq->irq);
2937                 scrq->irq = 0;
2938         }
2939
2940         if (scrq->msgs) {
2941                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2942                 atomic_set(&scrq->used, 0);
2943                 scrq->cur = 0;
2944                 scrq->ind_buf.index = 0;
2945         } else {
2946                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2947                 return -EINVAL;
2948         }
2949
2950         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2951                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2952         return rc;
2953 }
2954
2955 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2956 {
2957         int i, rc;
2958
2959         if (!adapter->tx_scrq || !adapter->rx_scrq)
2960                 return -EINVAL;
2961
2962         for (i = 0; i < adapter->req_tx_queues; i++) {
2963                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2964                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2965                 if (rc)
2966                         return rc;
2967         }
2968
2969         for (i = 0; i < adapter->req_rx_queues; i++) {
2970                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2971                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2972                 if (rc)
2973                         return rc;
2974         }
2975
2976         return rc;
2977 }
2978
2979 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2980                                   struct ibmvnic_sub_crq_queue *scrq,
2981                                   bool do_h_free)
2982 {
2983         struct device *dev = &adapter->vdev->dev;
2984         long rc;
2985
2986         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2987
2988         if (do_h_free) {
2989                 /* Close the sub-crqs */
2990                 do {
2991                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2992                                                 adapter->vdev->unit_address,
2993                                                 scrq->crq_num);
2994                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2995
2996                 if (rc) {
2997                         netdev_err(adapter->netdev,
2998                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2999                                    scrq->crq_num, rc);
3000                 }
3001         }
3002
3003         dma_free_coherent(dev,
3004                           IBMVNIC_IND_ARR_SZ,
3005                           scrq->ind_buf.indir_arr,
3006                           scrq->ind_buf.indir_dma);
3007
3008         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3009                          DMA_BIDIRECTIONAL);
3010         free_pages((unsigned long)scrq->msgs, 2);
3011         kfree(scrq);
3012 }
3013
3014 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3015                                                         *adapter)
3016 {
3017         struct device *dev = &adapter->vdev->dev;
3018         struct ibmvnic_sub_crq_queue *scrq;
3019         int rc;
3020
3021         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3022         if (!scrq)
3023                 return NULL;
3024
3025         scrq->msgs =
3026                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3027         if (!scrq->msgs) {
3028                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3029                 goto zero_page_failed;
3030         }
3031
3032         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3033                                          DMA_BIDIRECTIONAL);
3034         if (dma_mapping_error(dev, scrq->msg_token)) {
3035                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3036                 goto map_failed;
3037         }
3038
3039         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3040                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3041
3042         if (rc == H_RESOURCE)
3043                 rc = ibmvnic_reset_crq(adapter);
3044
3045         if (rc == H_CLOSED) {
3046                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3047         } else if (rc) {
3048                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3049                 goto reg_failed;
3050         }
3051
3052         scrq->adapter = adapter;
3053         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3054         scrq->ind_buf.index = 0;
3055
3056         scrq->ind_buf.indir_arr =
3057                 dma_alloc_coherent(dev,
3058                                    IBMVNIC_IND_ARR_SZ,
3059                                    &scrq->ind_buf.indir_dma,
3060                                    GFP_KERNEL);
3061
3062         if (!scrq->ind_buf.indir_arr)
3063                 goto indir_failed;
3064
3065         spin_lock_init(&scrq->lock);
3066
3067         netdev_dbg(adapter->netdev,
3068                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3069                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3070
3071         return scrq;
3072
3073 indir_failed:
3074         do {
3075                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3076                                         adapter->vdev->unit_address,
3077                                         scrq->crq_num);
3078         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3079 reg_failed:
3080         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3081                          DMA_BIDIRECTIONAL);
3082 map_failed:
3083         free_pages((unsigned long)scrq->msgs, 2);
3084 zero_page_failed:
3085         kfree(scrq);
3086
3087         return NULL;
3088 }
3089
3090 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3091 {
3092         int i;
3093
3094         if (adapter->tx_scrq) {
3095                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3096                         if (!adapter->tx_scrq[i])
3097                                 continue;
3098
3099                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3100                                    i);
3101                         if (adapter->tx_scrq[i]->irq) {
3102                                 free_irq(adapter->tx_scrq[i]->irq,
3103                                          adapter->tx_scrq[i]);
3104                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3105                                 adapter->tx_scrq[i]->irq = 0;
3106                         }
3107
3108                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3109                                               do_h_free);
3110                 }
3111
3112                 kfree(adapter->tx_scrq);
3113                 adapter->tx_scrq = NULL;
3114                 adapter->num_active_tx_scrqs = 0;
3115         }
3116
3117         if (adapter->rx_scrq) {
3118                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3119                         if (!adapter->rx_scrq[i])
3120                                 continue;
3121
3122                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3123                                    i);
3124                         if (adapter->rx_scrq[i]->irq) {
3125                                 free_irq(adapter->rx_scrq[i]->irq,
3126                                          adapter->rx_scrq[i]);
3127                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3128                                 adapter->rx_scrq[i]->irq = 0;
3129                         }
3130
3131                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3132                                               do_h_free);
3133                 }
3134
3135                 kfree(adapter->rx_scrq);
3136                 adapter->rx_scrq = NULL;
3137                 adapter->num_active_rx_scrqs = 0;
3138         }
3139 }
3140
3141 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3142                             struct ibmvnic_sub_crq_queue *scrq)
3143 {
3144         struct device *dev = &adapter->vdev->dev;
3145         unsigned long rc;
3146
3147         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3148                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3149         if (rc)
3150                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3151                         scrq->hw_irq, rc);
3152         return rc;
3153 }
3154
3155 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3156                            struct ibmvnic_sub_crq_queue *scrq)
3157 {
3158         struct device *dev = &adapter->vdev->dev;
3159         unsigned long rc;
3160
3161         if (scrq->hw_irq > 0x100000000ULL) {
3162                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3163                 return 1;
3164         }
3165
3166         if (test_bit(0, &adapter->resetting) &&
3167             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3168                 u64 val = (0xff000000) | scrq->hw_irq;
3169
3170                 rc = plpar_hcall_norets(H_EOI, val);
3171                 /* H_EOI would fail with rc = H_FUNCTION when running
3172                  * in XIVE mode which is expected, but not an error.
3173                  */
3174                 if (rc && rc != H_FUNCTION)
3175                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3176                                 val, rc);
3177         }
3178
3179         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3180                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3181         if (rc)
3182                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3183                         scrq->hw_irq, rc);
3184         return rc;
3185 }
3186
3187 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3188                                struct ibmvnic_sub_crq_queue *scrq)
3189 {
3190         struct device *dev = &adapter->vdev->dev;
3191         struct ibmvnic_tx_pool *tx_pool;
3192         struct ibmvnic_tx_buff *txbuff;
3193         struct netdev_queue *txq;
3194         union sub_crq *next;
3195         int index;
3196         int i;
3197
3198 restart_loop:
3199         while (pending_scrq(adapter, scrq)) {
3200                 unsigned int pool = scrq->pool_index;
3201                 int num_entries = 0;
3202                 int total_bytes = 0;
3203                 int num_packets = 0;
3204
3205                 next = ibmvnic_next_scrq(adapter, scrq);
3206                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3207                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3208                         if (index & IBMVNIC_TSO_POOL_MASK) {
3209                                 tx_pool = &adapter->tso_pool[pool];
3210                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3211                         } else {
3212                                 tx_pool = &adapter->tx_pool[pool];
3213                         }
3214
3215                         txbuff = &tx_pool->tx_buff[index];
3216                         num_packets++;
3217                         num_entries += txbuff->num_entries;
3218                         if (txbuff->skb) {
3219                                 total_bytes += txbuff->skb->len;
3220                                 if (next->tx_comp.rcs[i]) {
3221                                         dev_err(dev, "tx error %x\n",
3222                                                 next->tx_comp.rcs[i]);
3223                                         dev_kfree_skb_irq(txbuff->skb);
3224                                 } else {
3225                                         dev_consume_skb_irq(txbuff->skb);
3226                                 }
3227                                 txbuff->skb = NULL;
3228                         } else {
3229                                 netdev_warn(adapter->netdev,
3230                                             "TX completion received with NULL socket buffer\n");
3231                         }
3232                         tx_pool->free_map[tx_pool->producer_index] = index;
3233                         tx_pool->producer_index =
3234                                 (tx_pool->producer_index + 1) %
3235                                         tx_pool->num_buffers;
3236                 }
3237                 /* remove tx_comp scrq*/
3238                 next->tx_comp.first = 0;
3239
3240                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3241                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3242
3243                 if (atomic_sub_return(num_entries, &scrq->used) <=
3244                     (adapter->req_tx_entries_per_subcrq / 2) &&
3245                     __netif_subqueue_stopped(adapter->netdev,
3246                                              scrq->pool_index)) {
3247                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3248                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3249                                    scrq->pool_index);
3250                 }
3251         }
3252
3253         enable_scrq_irq(adapter, scrq);
3254
3255         if (pending_scrq(adapter, scrq)) {
3256                 disable_scrq_irq(adapter, scrq);
3257                 goto restart_loop;
3258         }
3259
3260         return 0;
3261 }
3262
3263 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3264 {
3265         struct ibmvnic_sub_crq_queue *scrq = instance;
3266         struct ibmvnic_adapter *adapter = scrq->adapter;
3267
3268         disable_scrq_irq(adapter, scrq);
3269         ibmvnic_complete_tx(adapter, scrq);
3270
3271         return IRQ_HANDLED;
3272 }
3273
3274 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3275 {
3276         struct ibmvnic_sub_crq_queue *scrq = instance;
3277         struct ibmvnic_adapter *adapter = scrq->adapter;
3278
3279         /* When booting a kdump kernel we can hit pending interrupts
3280          * prior to completing driver initialization.
3281          */
3282         if (unlikely(adapter->state != VNIC_OPEN))
3283                 return IRQ_NONE;
3284
3285         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3286
3287         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3288                 disable_scrq_irq(adapter, scrq);
3289                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3290         }
3291
3292         return IRQ_HANDLED;
3293 }
3294
3295 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3296 {
3297         struct device *dev = &adapter->vdev->dev;
3298         struct ibmvnic_sub_crq_queue *scrq;
3299         int i = 0, j = 0;
3300         int rc = 0;
3301
3302         for (i = 0; i < adapter->req_tx_queues; i++) {
3303                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3304                            i);
3305                 scrq = adapter->tx_scrq[i];
3306                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3307
3308                 if (!scrq->irq) {
3309                         rc = -EINVAL;
3310                         dev_err(dev, "Error mapping irq\n");
3311                         goto req_tx_irq_failed;
3312                 }
3313
3314                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3315                          adapter->vdev->unit_address, i);
3316                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3317                                  0, scrq->name, scrq);
3318
3319                 if (rc) {
3320                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3321                                 scrq->irq, rc);
3322                         irq_dispose_mapping(scrq->irq);
3323                         goto req_tx_irq_failed;
3324                 }
3325         }
3326
3327         for (i = 0; i < adapter->req_rx_queues; i++) {
3328                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3329                            i);
3330                 scrq = adapter->rx_scrq[i];
3331                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3332                 if (!scrq->irq) {
3333                         rc = -EINVAL;
3334                         dev_err(dev, "Error mapping irq\n");
3335                         goto req_rx_irq_failed;
3336                 }
3337                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3338                          adapter->vdev->unit_address, i);
3339                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3340                                  0, scrq->name, scrq);
3341                 if (rc) {
3342                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3343                                 scrq->irq, rc);
3344                         irq_dispose_mapping(scrq->irq);
3345                         goto req_rx_irq_failed;
3346                 }
3347         }
3348         return rc;
3349
3350 req_rx_irq_failed:
3351         for (j = 0; j < i; j++) {
3352                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3353                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3354         }
3355         i = adapter->req_tx_queues;
3356 req_tx_irq_failed:
3357         for (j = 0; j < i; j++) {
3358                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3359                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3360         }
3361         release_sub_crqs(adapter, 1);
3362         return rc;
3363 }
3364
3365 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3366 {
3367         struct device *dev = &adapter->vdev->dev;
3368         struct ibmvnic_sub_crq_queue **allqueues;
3369         int registered_queues = 0;
3370         int total_queues;
3371         int more = 0;
3372         int i;
3373
3374         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3375
3376         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3377         if (!allqueues)
3378                 return -1;
3379
3380         for (i = 0; i < total_queues; i++) {
3381                 allqueues[i] = init_sub_crq_queue(adapter);
3382                 if (!allqueues[i]) {
3383                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3384                         break;
3385                 }
3386                 registered_queues++;
3387         }
3388
3389         /* Make sure we were able to register the minimum number of queues */
3390         if (registered_queues <
3391             adapter->min_tx_queues + adapter->min_rx_queues) {
3392                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3393                 goto tx_failed;
3394         }
3395
3396         /* Distribute the failed allocated queues*/
3397         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3398                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3399                 switch (i % 3) {
3400                 case 0:
3401                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3402                                 adapter->req_rx_queues--;
3403                         else
3404                                 more++;
3405                         break;
3406                 case 1:
3407                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3408                                 adapter->req_tx_queues--;
3409                         else
3410                                 more++;
3411                         break;
3412                 }
3413         }
3414
3415         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3416                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3417         if (!adapter->tx_scrq)
3418                 goto tx_failed;
3419
3420         for (i = 0; i < adapter->req_tx_queues; i++) {
3421                 adapter->tx_scrq[i] = allqueues[i];
3422                 adapter->tx_scrq[i]->pool_index = i;
3423                 adapter->num_active_tx_scrqs++;
3424         }
3425
3426         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3427                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3428         if (!adapter->rx_scrq)
3429                 goto rx_failed;
3430
3431         for (i = 0; i < adapter->req_rx_queues; i++) {
3432                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3433                 adapter->rx_scrq[i]->scrq_num = i;
3434                 adapter->num_active_rx_scrqs++;
3435         }
3436
3437         kfree(allqueues);
3438         return 0;
3439
3440 rx_failed:
3441         kfree(adapter->tx_scrq);
3442         adapter->tx_scrq = NULL;
3443 tx_failed:
3444         for (i = 0; i < registered_queues; i++)
3445                 release_sub_crq_queue(adapter, allqueues[i], 1);
3446         kfree(allqueues);
3447         return -1;
3448 }
3449
3450 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3451 {
3452         struct device *dev = &adapter->vdev->dev;
3453         union ibmvnic_crq crq;
3454         int max_entries;
3455
3456         if (!retry) {
3457                 /* Sub-CRQ entries are 32 byte long */
3458                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3459
3460                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3461                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3462                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3463                         return;
3464                 }
3465
3466                 if (adapter->desired.mtu)
3467                         adapter->req_mtu = adapter->desired.mtu;
3468                 else
3469                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3470
3471                 if (!adapter->desired.tx_entries)
3472                         adapter->desired.tx_entries =
3473                                         adapter->max_tx_entries_per_subcrq;
3474                 if (!adapter->desired.rx_entries)
3475                         adapter->desired.rx_entries =
3476                                         adapter->max_rx_add_entries_per_subcrq;
3477
3478                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3479                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3480
3481                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3482                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3483                         adapter->desired.tx_entries = max_entries;
3484                 }
3485
3486                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3487                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3488                         adapter->desired.rx_entries = max_entries;
3489                 }
3490
3491                 if (adapter->desired.tx_entries)
3492                         adapter->req_tx_entries_per_subcrq =
3493                                         adapter->desired.tx_entries;
3494                 else
3495                         adapter->req_tx_entries_per_subcrq =
3496                                         adapter->max_tx_entries_per_subcrq;
3497
3498                 if (adapter->desired.rx_entries)
3499                         adapter->req_rx_add_entries_per_subcrq =
3500                                         adapter->desired.rx_entries;
3501                 else
3502                         adapter->req_rx_add_entries_per_subcrq =
3503                                         adapter->max_rx_add_entries_per_subcrq;
3504
3505                 if (adapter->desired.tx_queues)
3506                         adapter->req_tx_queues =
3507                                         adapter->desired.tx_queues;
3508                 else
3509                         adapter->req_tx_queues =
3510                                         adapter->opt_tx_comp_sub_queues;
3511
3512                 if (adapter->desired.rx_queues)
3513                         adapter->req_rx_queues =
3514                                         adapter->desired.rx_queues;
3515                 else
3516                         adapter->req_rx_queues =
3517                                         adapter->opt_rx_comp_queues;
3518
3519                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3520         }
3521
3522         memset(&crq, 0, sizeof(crq));
3523         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3524         crq.request_capability.cmd = REQUEST_CAPABILITY;
3525
3526         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3527         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3528         atomic_inc(&adapter->running_cap_crqs);
3529         ibmvnic_send_crq(adapter, &crq);
3530
3531         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3532         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3533         atomic_inc(&adapter->running_cap_crqs);
3534         ibmvnic_send_crq(adapter, &crq);
3535
3536         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3537         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3538         atomic_inc(&adapter->running_cap_crqs);
3539         ibmvnic_send_crq(adapter, &crq);
3540
3541         crq.request_capability.capability =
3542             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3543         crq.request_capability.number =
3544             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3545         atomic_inc(&adapter->running_cap_crqs);
3546         ibmvnic_send_crq(adapter, &crq);
3547
3548         crq.request_capability.capability =
3549             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3550         crq.request_capability.number =
3551             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3552         atomic_inc(&adapter->running_cap_crqs);
3553         ibmvnic_send_crq(adapter, &crq);
3554
3555         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3556         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3557         atomic_inc(&adapter->running_cap_crqs);
3558         ibmvnic_send_crq(adapter, &crq);
3559
3560         if (adapter->netdev->flags & IFF_PROMISC) {
3561                 if (adapter->promisc_supported) {
3562                         crq.request_capability.capability =
3563                             cpu_to_be16(PROMISC_REQUESTED);
3564                         crq.request_capability.number = cpu_to_be64(1);
3565                         atomic_inc(&adapter->running_cap_crqs);
3566                         ibmvnic_send_crq(adapter, &crq);
3567                 }
3568         } else {
3569                 crq.request_capability.capability =
3570                     cpu_to_be16(PROMISC_REQUESTED);
3571                 crq.request_capability.number = cpu_to_be64(0);
3572                 atomic_inc(&adapter->running_cap_crqs);
3573                 ibmvnic_send_crq(adapter, &crq);
3574         }
3575 }
3576
3577 static int pending_scrq(struct ibmvnic_adapter *adapter,
3578                         struct ibmvnic_sub_crq_queue *scrq)
3579 {
3580         union sub_crq *entry = &scrq->msgs[scrq->cur];
3581         int rc;
3582
3583         rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3584
3585         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3586          * contents of the SCRQ descriptor
3587          */
3588         dma_rmb();
3589
3590         return rc;
3591 }
3592
3593 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3594                                         struct ibmvnic_sub_crq_queue *scrq)
3595 {
3596         union sub_crq *entry;
3597         unsigned long flags;
3598
3599         spin_lock_irqsave(&scrq->lock, flags);
3600         entry = &scrq->msgs[scrq->cur];
3601         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3602                 if (++scrq->cur == scrq->size)
3603                         scrq->cur = 0;
3604         } else {
3605                 entry = NULL;
3606         }
3607         spin_unlock_irqrestore(&scrq->lock, flags);
3608
3609         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3610          * contents of the SCRQ descriptor
3611          */
3612         dma_rmb();
3613
3614         return entry;
3615 }
3616
3617 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3618 {
3619         struct ibmvnic_crq_queue *queue = &adapter->crq;
3620         union ibmvnic_crq *crq;
3621
3622         crq = &queue->msgs[queue->cur];
3623         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3624                 if (++queue->cur == queue->size)
3625                         queue->cur = 0;
3626         } else {
3627                 crq = NULL;
3628         }
3629
3630         return crq;
3631 }
3632
3633 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3634 {
3635         switch (rc) {
3636         case H_PARAMETER:
3637                 dev_warn_ratelimited(dev,
3638                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3639                                      func, rc);
3640                 break;
3641         case H_CLOSED:
3642                 dev_warn_ratelimited(dev,
3643                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3644                                      func, rc);
3645                 break;
3646         default:
3647                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3648                 break;
3649         }
3650 }
3651
3652 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3653                                 u64 remote_handle, u64 ioba, u64 num_entries)
3654 {
3655         unsigned int ua = adapter->vdev->unit_address;
3656         struct device *dev = &adapter->vdev->dev;
3657         int rc;
3658
3659         /* Make sure the hypervisor sees the complete request */
3660         dma_wmb();
3661         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3662                                 cpu_to_be64(remote_handle),
3663                                 ioba, num_entries);
3664
3665         if (rc)
3666                 print_subcrq_error(dev, rc, __func__);
3667
3668         return rc;
3669 }
3670
3671 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3672                             union ibmvnic_crq *crq)
3673 {
3674         unsigned int ua = adapter->vdev->unit_address;
3675         struct device *dev = &adapter->vdev->dev;
3676         u64 *u64_crq = (u64 *)crq;
3677         int rc;
3678
3679         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3680                    (unsigned long)cpu_to_be64(u64_crq[0]),
3681                    (unsigned long)cpu_to_be64(u64_crq[1]));
3682
3683         if (!adapter->crq.active &&
3684             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3685                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3686                 return -EINVAL;
3687         }
3688
3689         /* Make sure the hypervisor sees the complete request */
3690         dma_wmb();
3691
3692         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3693                                 cpu_to_be64(u64_crq[0]),
3694                                 cpu_to_be64(u64_crq[1]));
3695
3696         if (rc) {
3697                 if (rc == H_CLOSED) {
3698                         dev_warn(dev, "CRQ Queue closed\n");
3699                         /* do not reset, report the fail, wait for passive init from server */
3700                 }
3701
3702                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3703         }
3704
3705         return rc;
3706 }
3707
3708 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3709 {
3710         struct device *dev = &adapter->vdev->dev;
3711         union ibmvnic_crq crq;
3712         int retries = 100;
3713         int rc;
3714
3715         memset(&crq, 0, sizeof(crq));
3716         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3717         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3718         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3719
3720         do {
3721                 rc = ibmvnic_send_crq(adapter, &crq);
3722                 if (rc != H_CLOSED)
3723                         break;
3724                 retries--;
3725                 msleep(50);
3726
3727         } while (retries > 0);
3728
3729         if (rc) {
3730                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3731                 return rc;
3732         }
3733
3734         return 0;
3735 }
3736
3737 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3738 {
3739         union ibmvnic_crq crq;
3740
3741         memset(&crq, 0, sizeof(crq));
3742         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3743         crq.version_exchange.cmd = VERSION_EXCHANGE;
3744         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3745
3746         return ibmvnic_send_crq(adapter, &crq);
3747 }
3748
3749 struct vnic_login_client_data {
3750         u8      type;
3751         __be16  len;
3752         char    name[];
3753 } __packed;
3754
3755 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3756 {
3757         int len;
3758
3759         /* Calculate the amount of buffer space needed for the
3760          * vnic client data in the login buffer. There are four entries,
3761          * OS name, LPAR name, device name, and a null last entry.
3762          */
3763         len = 4 * sizeof(struct vnic_login_client_data);
3764         len += 6; /* "Linux" plus NULL */
3765         len += strlen(utsname()->nodename) + 1;
3766         len += strlen(adapter->netdev->name) + 1;
3767
3768         return len;
3769 }
3770
3771 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3772                                  struct vnic_login_client_data *vlcd)
3773 {
3774         const char *os_name = "Linux";
3775         int len;
3776
3777         /* Type 1 - LPAR OS */
3778         vlcd->type = 1;
3779         len = strlen(os_name) + 1;
3780         vlcd->len = cpu_to_be16(len);
3781         strncpy(vlcd->name, os_name, len);
3782         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3783
3784         /* Type 2 - LPAR name */
3785         vlcd->type = 2;
3786         len = strlen(utsname()->nodename) + 1;
3787         vlcd->len = cpu_to_be16(len);
3788         strncpy(vlcd->name, utsname()->nodename, len);
3789         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3790
3791         /* Type 3 - device name */
3792         vlcd->type = 3;
3793         len = strlen(adapter->netdev->name) + 1;
3794         vlcd->len = cpu_to_be16(len);
3795         strncpy(vlcd->name, adapter->netdev->name, len);
3796 }
3797
3798 static int send_login(struct ibmvnic_adapter *adapter)
3799 {
3800         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3801         struct ibmvnic_login_buffer *login_buffer;
3802         struct device *dev = &adapter->vdev->dev;
3803         struct vnic_login_client_data *vlcd;
3804         dma_addr_t rsp_buffer_token;
3805         dma_addr_t buffer_token;
3806         size_t rsp_buffer_size;
3807         union ibmvnic_crq crq;
3808         int client_data_len;
3809         size_t buffer_size;
3810         __be64 *tx_list_p;
3811         __be64 *rx_list_p;
3812         int rc;
3813         int i;
3814
3815         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3816                 netdev_err(adapter->netdev,
3817                            "RX or TX queues are not allocated, device login failed\n");
3818                 return -1;
3819         }
3820
3821         release_login_buffer(adapter);
3822         release_login_rsp_buffer(adapter);
3823
3824         client_data_len = vnic_client_data_len(adapter);
3825
3826         buffer_size =
3827             sizeof(struct ibmvnic_login_buffer) +
3828             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3829             client_data_len;
3830
3831         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3832         if (!login_buffer)
3833                 goto buf_alloc_failed;
3834
3835         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3836                                       DMA_TO_DEVICE);
3837         if (dma_mapping_error(dev, buffer_token)) {
3838                 dev_err(dev, "Couldn't map login buffer\n");
3839                 goto buf_map_failed;
3840         }
3841
3842         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3843                           sizeof(u64) * adapter->req_tx_queues +
3844                           sizeof(u64) * adapter->req_rx_queues +
3845                           sizeof(u64) * adapter->req_rx_queues +
3846                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3847
3848         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3849         if (!login_rsp_buffer)
3850                 goto buf_rsp_alloc_failed;
3851
3852         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3853                                           rsp_buffer_size, DMA_FROM_DEVICE);
3854         if (dma_mapping_error(dev, rsp_buffer_token)) {
3855                 dev_err(dev, "Couldn't map login rsp buffer\n");
3856                 goto buf_rsp_map_failed;
3857         }
3858
3859         adapter->login_buf = login_buffer;
3860         adapter->login_buf_token = buffer_token;
3861         adapter->login_buf_sz = buffer_size;
3862         adapter->login_rsp_buf = login_rsp_buffer;
3863         adapter->login_rsp_buf_token = rsp_buffer_token;
3864         adapter->login_rsp_buf_sz = rsp_buffer_size;
3865
3866         login_buffer->len = cpu_to_be32(buffer_size);
3867         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3868         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3869         login_buffer->off_txcomp_subcrqs =
3870             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3871         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3872         login_buffer->off_rxcomp_subcrqs =
3873             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3874                         sizeof(u64) * adapter->req_tx_queues);
3875         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3876         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3877
3878         tx_list_p = (__be64 *)((char *)login_buffer +
3879                                       sizeof(struct ibmvnic_login_buffer));
3880         rx_list_p = (__be64 *)((char *)login_buffer +
3881                                       sizeof(struct ibmvnic_login_buffer) +
3882                                       sizeof(u64) * adapter->req_tx_queues);
3883
3884         for (i = 0; i < adapter->req_tx_queues; i++) {
3885                 if (adapter->tx_scrq[i]) {
3886                         tx_list_p[i] =
3887                                 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
3888                 }
3889         }
3890
3891         for (i = 0; i < adapter->req_rx_queues; i++) {
3892                 if (adapter->rx_scrq[i]) {
3893                         rx_list_p[i] =
3894                                 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
3895                 }
3896         }
3897
3898         /* Insert vNIC login client data */
3899         vlcd = (struct vnic_login_client_data *)
3900                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3901         login_buffer->client_data_offset =
3902                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3903         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3904
3905         vnic_add_client_data(adapter, vlcd);
3906
3907         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3908         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3909                 netdev_dbg(adapter->netdev, "%016lx\n",
3910                            ((unsigned long *)(adapter->login_buf))[i]);
3911         }
3912
3913         memset(&crq, 0, sizeof(crq));
3914         crq.login.first = IBMVNIC_CRQ_CMD;
3915         crq.login.cmd = LOGIN;
3916         crq.login.ioba = cpu_to_be32(buffer_token);
3917         crq.login.len = cpu_to_be32(buffer_size);
3918
3919         adapter->login_pending = true;
3920         rc = ibmvnic_send_crq(adapter, &crq);
3921         if (rc) {
3922                 adapter->login_pending = false;
3923                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3924                 goto buf_rsp_map_failed;
3925         }
3926
3927         return 0;
3928
3929 buf_rsp_map_failed:
3930         kfree(login_rsp_buffer);
3931         adapter->login_rsp_buf = NULL;
3932 buf_rsp_alloc_failed:
3933         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3934 buf_map_failed:
3935         kfree(login_buffer);
3936         adapter->login_buf = NULL;
3937 buf_alloc_failed:
3938         return -1;
3939 }
3940
3941 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3942                             u32 len, u8 map_id)
3943 {
3944         union ibmvnic_crq crq;
3945
3946         memset(&crq, 0, sizeof(crq));
3947         crq.request_map.first = IBMVNIC_CRQ_CMD;
3948         crq.request_map.cmd = REQUEST_MAP;
3949         crq.request_map.map_id = map_id;
3950         crq.request_map.ioba = cpu_to_be32(addr);
3951         crq.request_map.len = cpu_to_be32(len);
3952         return ibmvnic_send_crq(adapter, &crq);
3953 }
3954
3955 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3956 {
3957         union ibmvnic_crq crq;
3958
3959         memset(&crq, 0, sizeof(crq));
3960         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3961         crq.request_unmap.cmd = REQUEST_UNMAP;
3962         crq.request_unmap.map_id = map_id;
3963         return ibmvnic_send_crq(adapter, &crq);
3964 }
3965
3966 static void send_query_map(struct ibmvnic_adapter *adapter)
3967 {
3968         union ibmvnic_crq crq;
3969
3970         memset(&crq, 0, sizeof(crq));
3971         crq.query_map.first = IBMVNIC_CRQ_CMD;
3972         crq.query_map.cmd = QUERY_MAP;
3973         ibmvnic_send_crq(adapter, &crq);
3974 }
3975
3976 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3977 static void send_query_cap(struct ibmvnic_adapter *adapter)
3978 {
3979         union ibmvnic_crq crq;
3980
3981         atomic_set(&adapter->running_cap_crqs, 0);
3982         memset(&crq, 0, sizeof(crq));
3983         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3984         crq.query_capability.cmd = QUERY_CAPABILITY;
3985
3986         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3987         atomic_inc(&adapter->running_cap_crqs);
3988         ibmvnic_send_crq(adapter, &crq);
3989
3990         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3991         atomic_inc(&adapter->running_cap_crqs);
3992         ibmvnic_send_crq(adapter, &crq);
3993
3994         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3995         atomic_inc(&adapter->running_cap_crqs);
3996         ibmvnic_send_crq(adapter, &crq);
3997
3998         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3999         atomic_inc(&adapter->running_cap_crqs);
4000         ibmvnic_send_crq(adapter, &crq);
4001
4002         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4003         atomic_inc(&adapter->running_cap_crqs);
4004         ibmvnic_send_crq(adapter, &crq);
4005
4006         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4007         atomic_inc(&adapter->running_cap_crqs);
4008         ibmvnic_send_crq(adapter, &crq);
4009
4010         crq.query_capability.capability =
4011             cpu_to_be16(MIN_TX_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(MIN_RX_ADD_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_TX_ENTRIES_PER_SUBCRQ);
4022         atomic_inc(&adapter->running_cap_crqs);
4023         ibmvnic_send_crq(adapter, &crq);
4024
4025         crq.query_capability.capability =
4026             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4027         atomic_inc(&adapter->running_cap_crqs);
4028         ibmvnic_send_crq(adapter, &crq);
4029
4030         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4031         atomic_inc(&adapter->running_cap_crqs);
4032         ibmvnic_send_crq(adapter, &crq);
4033
4034         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4035         atomic_inc(&adapter->running_cap_crqs);
4036         ibmvnic_send_crq(adapter, &crq);
4037
4038         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4039         atomic_inc(&adapter->running_cap_crqs);
4040         ibmvnic_send_crq(adapter, &crq);
4041
4042         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4043         atomic_inc(&adapter->running_cap_crqs);
4044         ibmvnic_send_crq(adapter, &crq);
4045
4046         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4047         atomic_inc(&adapter->running_cap_crqs);
4048         ibmvnic_send_crq(adapter, &crq);
4049
4050         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4051         atomic_inc(&adapter->running_cap_crqs);
4052         ibmvnic_send_crq(adapter, &crq);
4053
4054         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4055         atomic_inc(&adapter->running_cap_crqs);
4056         ibmvnic_send_crq(adapter, &crq);
4057
4058         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4059         atomic_inc(&adapter->running_cap_crqs);
4060         ibmvnic_send_crq(adapter, &crq);
4061
4062         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4063         atomic_inc(&adapter->running_cap_crqs);
4064         ibmvnic_send_crq(adapter, &crq);
4065
4066         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4067         atomic_inc(&adapter->running_cap_crqs);
4068         ibmvnic_send_crq(adapter, &crq);
4069
4070         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4071         atomic_inc(&adapter->running_cap_crqs);
4072         ibmvnic_send_crq(adapter, &crq);
4073
4074         crq.query_capability.capability =
4075                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4076         atomic_inc(&adapter->running_cap_crqs);
4077         ibmvnic_send_crq(adapter, &crq);
4078
4079         crq.query_capability.capability =
4080                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4081         atomic_inc(&adapter->running_cap_crqs);
4082         ibmvnic_send_crq(adapter, &crq);
4083
4084         crq.query_capability.capability =
4085                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4086         atomic_inc(&adapter->running_cap_crqs);
4087         ibmvnic_send_crq(adapter, &crq);
4088
4089         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4090         atomic_inc(&adapter->running_cap_crqs);
4091         ibmvnic_send_crq(adapter, &crq);
4092 }
4093
4094 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4095 {
4096         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4097         struct device *dev = &adapter->vdev->dev;
4098         union ibmvnic_crq crq;
4099
4100         adapter->ip_offload_tok =
4101                 dma_map_single(dev,
4102                                &adapter->ip_offload_buf,
4103                                buf_sz,
4104                                DMA_FROM_DEVICE);
4105
4106         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4107                 if (!firmware_has_feature(FW_FEATURE_CMO))
4108                         dev_err(dev, "Couldn't map offload buffer\n");
4109                 return;
4110         }
4111
4112         memset(&crq, 0, sizeof(crq));
4113         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4114         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4115         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4116         crq.query_ip_offload.ioba =
4117             cpu_to_be32(adapter->ip_offload_tok);
4118
4119         ibmvnic_send_crq(adapter, &crq);
4120 }
4121
4122 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4123 {
4124         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4125         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4126         struct device *dev = &adapter->vdev->dev;
4127         netdev_features_t old_hw_features = 0;
4128         union ibmvnic_crq crq;
4129
4130         adapter->ip_offload_ctrl_tok =
4131                 dma_map_single(dev,
4132                                ctrl_buf,
4133                                sizeof(adapter->ip_offload_ctrl),
4134                                DMA_TO_DEVICE);
4135
4136         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4137                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4138                 return;
4139         }
4140
4141         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4142         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4143         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4144         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4145         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4146         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4147         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4148         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4149         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4150         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4151
4152         /* large_rx disabled for now, additional features needed */
4153         ctrl_buf->large_rx_ipv4 = 0;
4154         ctrl_buf->large_rx_ipv6 = 0;
4155
4156         if (adapter->state != VNIC_PROBING) {
4157                 old_hw_features = adapter->netdev->hw_features;
4158                 adapter->netdev->hw_features = 0;
4159         }
4160
4161         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4162
4163         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4164                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4165
4166         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4167                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4168
4169         if ((adapter->netdev->features &
4170             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4171                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4172
4173         if (buf->large_tx_ipv4)
4174                 adapter->netdev->hw_features |= NETIF_F_TSO;
4175         if (buf->large_tx_ipv6)
4176                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4177
4178         if (adapter->state == VNIC_PROBING) {
4179                 adapter->netdev->features |= adapter->netdev->hw_features;
4180         } else if (old_hw_features != adapter->netdev->hw_features) {
4181                 netdev_features_t tmp = 0;
4182
4183                 /* disable features no longer supported */
4184                 adapter->netdev->features &= adapter->netdev->hw_features;
4185                 /* turn on features now supported if previously enabled */
4186                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4187                         adapter->netdev->hw_features;
4188                 adapter->netdev->features |=
4189                                 tmp & adapter->netdev->wanted_features;
4190         }
4191
4192         memset(&crq, 0, sizeof(crq));
4193         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4194         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4195         crq.control_ip_offload.len =
4196             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4197         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4198         ibmvnic_send_crq(adapter, &crq);
4199 }
4200
4201 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4202                                 struct ibmvnic_adapter *adapter)
4203 {
4204         struct device *dev = &adapter->vdev->dev;
4205
4206         if (crq->get_vpd_size_rsp.rc.code) {
4207                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4208                         crq->get_vpd_size_rsp.rc.code);
4209                 complete(&adapter->fw_done);
4210                 return;
4211         }
4212
4213         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4214         complete(&adapter->fw_done);
4215 }
4216
4217 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4218                            struct ibmvnic_adapter *adapter)
4219 {
4220         struct device *dev = &adapter->vdev->dev;
4221         unsigned char *substr = NULL;
4222         u8 fw_level_len = 0;
4223
4224         memset(adapter->fw_version, 0, 32);
4225
4226         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4227                          DMA_FROM_DEVICE);
4228
4229         if (crq->get_vpd_rsp.rc.code) {
4230                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4231                         crq->get_vpd_rsp.rc.code);
4232                 goto complete;
4233         }
4234
4235         /* get the position of the firmware version info
4236          * located after the ASCII 'RM' substring in the buffer
4237          */
4238         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4239         if (!substr) {
4240                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4241                 goto complete;
4242         }
4243
4244         /* get length of firmware level ASCII substring */
4245         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4246                 fw_level_len = *(substr + 2);
4247         } else {
4248                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4249                 goto complete;
4250         }
4251
4252         /* copy firmware version string from vpd into adapter */
4253         if ((substr + 3 + fw_level_len) <
4254             (adapter->vpd->buff + adapter->vpd->len)) {
4255                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4256         } else {
4257                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4258         }
4259
4260 complete:
4261         if (adapter->fw_version[0] == '\0')
4262                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4263         complete(&adapter->fw_done);
4264 }
4265
4266 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4267 {
4268         struct device *dev = &adapter->vdev->dev;
4269         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4270         int i;
4271
4272         dma_unmap_single(dev, adapter->ip_offload_tok,
4273                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4274
4275         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4276         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4277                 netdev_dbg(adapter->netdev, "%016lx\n",
4278                            ((unsigned long *)(buf))[i]);
4279
4280         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4281         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4282         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4283                    buf->tcp_ipv4_chksum);
4284         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4285                    buf->tcp_ipv6_chksum);
4286         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4287                    buf->udp_ipv4_chksum);
4288         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4289                    buf->udp_ipv6_chksum);
4290         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4291                    buf->large_tx_ipv4);
4292         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4293                    buf->large_tx_ipv6);
4294         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4295                    buf->large_rx_ipv4);
4296         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4297                    buf->large_rx_ipv6);
4298         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4299                    buf->max_ipv4_header_size);
4300         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4301                    buf->max_ipv6_header_size);
4302         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4303                    buf->max_tcp_header_size);
4304         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4305                    buf->max_udp_header_size);
4306         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4307                    buf->max_large_tx_size);
4308         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4309                    buf->max_large_rx_size);
4310         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4311                    buf->ipv6_extension_header);
4312         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4313                    buf->tcp_pseudosum_req);
4314         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4315                    buf->num_ipv6_ext_headers);
4316         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4317                    buf->off_ipv6_ext_headers);
4318
4319         send_control_ip_offload(adapter);
4320 }
4321
4322 static const char *ibmvnic_fw_err_cause(u16 cause)
4323 {
4324         switch (cause) {
4325         case ADAPTER_PROBLEM:
4326                 return "adapter problem";
4327         case BUS_PROBLEM:
4328                 return "bus problem";
4329         case FW_PROBLEM:
4330                 return "firmware problem";
4331         case DD_PROBLEM:
4332                 return "device driver problem";
4333         case EEH_RECOVERY:
4334                 return "EEH recovery";
4335         case FW_UPDATED:
4336                 return "firmware updated";
4337         case LOW_MEMORY:
4338                 return "low Memory";
4339         default:
4340                 return "unknown";
4341         }
4342 }
4343
4344 static void handle_error_indication(union ibmvnic_crq *crq,
4345                                     struct ibmvnic_adapter *adapter)
4346 {
4347         struct device *dev = &adapter->vdev->dev;
4348         u16 cause;
4349
4350         cause = be16_to_cpu(crq->error_indication.error_cause);
4351
4352         dev_warn_ratelimited(dev,
4353                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4354                              crq->error_indication.flags
4355                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4356                              ibmvnic_fw_err_cause(cause));
4357
4358         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4359                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4360         else
4361                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4362 }
4363
4364 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4365                                  struct ibmvnic_adapter *adapter)
4366 {
4367         struct net_device *netdev = adapter->netdev;
4368         struct device *dev = &adapter->vdev->dev;
4369         long rc;
4370
4371         rc = crq->change_mac_addr_rsp.rc.code;
4372         if (rc) {
4373                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4374                 goto out;
4375         }
4376         /* crq->change_mac_addr.mac_addr is the requested one
4377          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4378          */
4379         ether_addr_copy(netdev->dev_addr,
4380                         &crq->change_mac_addr_rsp.mac_addr[0]);
4381         ether_addr_copy(adapter->mac_addr,
4382                         &crq->change_mac_addr_rsp.mac_addr[0]);
4383 out:
4384         complete(&adapter->fw_done);
4385         return rc;
4386 }
4387
4388 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4389                                    struct ibmvnic_adapter *adapter)
4390 {
4391         struct device *dev = &adapter->vdev->dev;
4392         u64 *req_value;
4393         char *name;
4394
4395         atomic_dec(&adapter->running_cap_crqs);
4396         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4397         case REQ_TX_QUEUES:
4398                 req_value = &adapter->req_tx_queues;
4399                 name = "tx";
4400                 break;
4401         case REQ_RX_QUEUES:
4402                 req_value = &adapter->req_rx_queues;
4403                 name = "rx";
4404                 break;
4405         case REQ_RX_ADD_QUEUES:
4406                 req_value = &adapter->req_rx_add_queues;
4407                 name = "rx_add";
4408                 break;
4409         case REQ_TX_ENTRIES_PER_SUBCRQ:
4410                 req_value = &adapter->req_tx_entries_per_subcrq;
4411                 name = "tx_entries_per_subcrq";
4412                 break;
4413         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4414                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4415                 name = "rx_add_entries_per_subcrq";
4416                 break;
4417         case REQ_MTU:
4418                 req_value = &adapter->req_mtu;
4419                 name = "mtu";
4420                 break;
4421         case PROMISC_REQUESTED:
4422                 req_value = &adapter->promisc;
4423                 name = "promisc";
4424                 break;
4425         default:
4426                 dev_err(dev, "Got invalid cap request rsp %d\n",
4427                         crq->request_capability.capability);
4428                 return;
4429         }
4430
4431         switch (crq->request_capability_rsp.rc.code) {
4432         case SUCCESS:
4433                 break;
4434         case PARTIALSUCCESS:
4435                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4436                          *req_value,
4437                          (long)be64_to_cpu(crq->request_capability_rsp.number),
4438                          name);
4439
4440                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4441                     REQ_MTU) {
4442                         pr_err("mtu of %llu is not supported. Reverting.\n",
4443                                *req_value);
4444                         *req_value = adapter->fallback.mtu;
4445                 } else {
4446                         *req_value =
4447                                 be64_to_cpu(crq->request_capability_rsp.number);
4448                 }
4449
4450                 send_request_cap(adapter, 1);
4451                 return;
4452         default:
4453                 dev_err(dev, "Error %d in request cap rsp\n",
4454                         crq->request_capability_rsp.rc.code);
4455                 return;
4456         }
4457
4458         /* Done receiving requested capabilities, query IP offload support */
4459         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4460                 adapter->wait_capability = false;
4461                 send_query_ip_offload(adapter);
4462         }
4463 }
4464
4465 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4466                             struct ibmvnic_adapter *adapter)
4467 {
4468         struct device *dev = &adapter->vdev->dev;
4469         struct net_device *netdev = adapter->netdev;
4470         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4471         struct ibmvnic_login_buffer *login = adapter->login_buf;
4472         u64 *tx_handle_array;
4473         u64 *rx_handle_array;
4474         int num_tx_pools;
4475         int num_rx_pools;
4476         u64 *size_array;
4477         int i;
4478
4479         /* CHECK: Test/set of login_pending does not need to be atomic
4480          * because only ibmvnic_tasklet tests/clears this.
4481          */
4482         if (!adapter->login_pending) {
4483                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4484                 return 0;
4485         }
4486         adapter->login_pending = false;
4487
4488         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4489                          DMA_TO_DEVICE);
4490         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4491                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4492
4493         /* If the number of queues requested can't be allocated by the
4494          * server, the login response will return with code 1. We will need
4495          * to resend the login buffer with fewer queues requested.
4496          */
4497         if (login_rsp_crq->generic.rc.code) {
4498                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4499                 complete(&adapter->init_done);
4500                 return 0;
4501         }
4502
4503         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4504
4505         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4506         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4507                 netdev_dbg(adapter->netdev, "%016lx\n",
4508                            ((unsigned long *)(adapter->login_rsp_buf))[i]);
4509         }
4510
4511         /* Sanity checks */
4512         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4513             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4514              adapter->req_rx_add_queues !=
4515              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4516                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4517                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4518                 return -EIO;
4519         }
4520         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4521                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4522         /* variable buffer sizes are not supported, so just read the
4523          * first entry.
4524          */
4525         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4526
4527         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4528         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4529
4530         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4531                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4532         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4533                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4534
4535         for (i = 0; i < num_tx_pools; i++)
4536                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4537
4538         for (i = 0; i < num_rx_pools; i++)
4539                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4540
4541         adapter->num_active_tx_scrqs = num_tx_pools;
4542         adapter->num_active_rx_scrqs = num_rx_pools;
4543         release_login_rsp_buffer(adapter);
4544         release_login_buffer(adapter);
4545         complete(&adapter->init_done);
4546
4547         return 0;
4548 }
4549
4550 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4551                                      struct ibmvnic_adapter *adapter)
4552 {
4553         struct device *dev = &adapter->vdev->dev;
4554         long rc;
4555
4556         rc = crq->request_unmap_rsp.rc.code;
4557         if (rc)
4558                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4559 }
4560
4561 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4562                                  struct ibmvnic_adapter *adapter)
4563 {
4564         struct net_device *netdev = adapter->netdev;
4565         struct device *dev = &adapter->vdev->dev;
4566         long rc;
4567
4568         rc = crq->query_map_rsp.rc.code;
4569         if (rc) {
4570                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4571                 return;
4572         }
4573         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4574                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4575                    crq->query_map_rsp.free_pages);
4576 }
4577
4578 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4579                                  struct ibmvnic_adapter *adapter)
4580 {
4581         struct net_device *netdev = adapter->netdev;
4582         struct device *dev = &adapter->vdev->dev;
4583         long rc;
4584
4585         atomic_dec(&adapter->running_cap_crqs);
4586         netdev_dbg(netdev, "Outstanding queries: %d\n",
4587                    atomic_read(&adapter->running_cap_crqs));
4588         rc = crq->query_capability.rc.code;
4589         if (rc) {
4590                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4591                 goto out;
4592         }
4593
4594         switch (be16_to_cpu(crq->query_capability.capability)) {
4595         case MIN_TX_QUEUES:
4596                 adapter->min_tx_queues =
4597                     be64_to_cpu(crq->query_capability.number);
4598                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4599                            adapter->min_tx_queues);
4600                 break;
4601         case MIN_RX_QUEUES:
4602                 adapter->min_rx_queues =
4603                     be64_to_cpu(crq->query_capability.number);
4604                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4605                            adapter->min_rx_queues);
4606                 break;
4607         case MIN_RX_ADD_QUEUES:
4608                 adapter->min_rx_add_queues =
4609                     be64_to_cpu(crq->query_capability.number);
4610                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4611                            adapter->min_rx_add_queues);
4612                 break;
4613         case MAX_TX_QUEUES:
4614                 adapter->max_tx_queues =
4615                     be64_to_cpu(crq->query_capability.number);
4616                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4617                            adapter->max_tx_queues);
4618                 break;
4619         case MAX_RX_QUEUES:
4620                 adapter->max_rx_queues =
4621                     be64_to_cpu(crq->query_capability.number);
4622                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4623                            adapter->max_rx_queues);
4624                 break;
4625         case MAX_RX_ADD_QUEUES:
4626                 adapter->max_rx_add_queues =
4627                     be64_to_cpu(crq->query_capability.number);
4628                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4629                            adapter->max_rx_add_queues);
4630                 break;
4631         case MIN_TX_ENTRIES_PER_SUBCRQ:
4632                 adapter->min_tx_entries_per_subcrq =
4633                     be64_to_cpu(crq->query_capability.number);
4634                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4635                            adapter->min_tx_entries_per_subcrq);
4636                 break;
4637         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4638                 adapter->min_rx_add_entries_per_subcrq =
4639                     be64_to_cpu(crq->query_capability.number);
4640                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4641                            adapter->min_rx_add_entries_per_subcrq);
4642                 break;
4643         case MAX_TX_ENTRIES_PER_SUBCRQ:
4644                 adapter->max_tx_entries_per_subcrq =
4645                     be64_to_cpu(crq->query_capability.number);
4646                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4647                            adapter->max_tx_entries_per_subcrq);
4648                 break;
4649         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4650                 adapter->max_rx_add_entries_per_subcrq =
4651                     be64_to_cpu(crq->query_capability.number);
4652                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4653                            adapter->max_rx_add_entries_per_subcrq);
4654                 break;
4655         case TCP_IP_OFFLOAD:
4656                 adapter->tcp_ip_offload =
4657                     be64_to_cpu(crq->query_capability.number);
4658                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4659                            adapter->tcp_ip_offload);
4660                 break;
4661         case PROMISC_SUPPORTED:
4662                 adapter->promisc_supported =
4663                     be64_to_cpu(crq->query_capability.number);
4664                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4665                            adapter->promisc_supported);
4666                 break;
4667         case MIN_MTU:
4668                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4669                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4670                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4671                 break;
4672         case MAX_MTU:
4673                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4674                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4675                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4676                 break;
4677         case MAX_MULTICAST_FILTERS:
4678                 adapter->max_multicast_filters =
4679                     be64_to_cpu(crq->query_capability.number);
4680                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4681                            adapter->max_multicast_filters);
4682                 break;
4683         case VLAN_HEADER_INSERTION:
4684                 adapter->vlan_header_insertion =
4685                     be64_to_cpu(crq->query_capability.number);
4686                 if (adapter->vlan_header_insertion)
4687                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4688                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4689                            adapter->vlan_header_insertion);
4690                 break;
4691         case RX_VLAN_HEADER_INSERTION:
4692                 adapter->rx_vlan_header_insertion =
4693                     be64_to_cpu(crq->query_capability.number);
4694                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4695                            adapter->rx_vlan_header_insertion);
4696                 break;
4697         case MAX_TX_SG_ENTRIES:
4698                 adapter->max_tx_sg_entries =
4699                     be64_to_cpu(crq->query_capability.number);
4700                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4701                            adapter->max_tx_sg_entries);
4702                 break;
4703         case RX_SG_SUPPORTED:
4704                 adapter->rx_sg_supported =
4705                     be64_to_cpu(crq->query_capability.number);
4706                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4707                            adapter->rx_sg_supported);
4708                 break;
4709         case OPT_TX_COMP_SUB_QUEUES:
4710                 adapter->opt_tx_comp_sub_queues =
4711                     be64_to_cpu(crq->query_capability.number);
4712                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4713                            adapter->opt_tx_comp_sub_queues);
4714                 break;
4715         case OPT_RX_COMP_QUEUES:
4716                 adapter->opt_rx_comp_queues =
4717                     be64_to_cpu(crq->query_capability.number);
4718                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4719                            adapter->opt_rx_comp_queues);
4720                 break;
4721         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4722                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4723                     be64_to_cpu(crq->query_capability.number);
4724                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4725                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4726                 break;
4727         case OPT_TX_ENTRIES_PER_SUBCRQ:
4728                 adapter->opt_tx_entries_per_subcrq =
4729                     be64_to_cpu(crq->query_capability.number);
4730                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4731                            adapter->opt_tx_entries_per_subcrq);
4732                 break;
4733         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4734                 adapter->opt_rxba_entries_per_subcrq =
4735                     be64_to_cpu(crq->query_capability.number);
4736                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4737                            adapter->opt_rxba_entries_per_subcrq);
4738                 break;
4739         case TX_RX_DESC_REQ:
4740                 adapter->tx_rx_desc_req = crq->query_capability.number;
4741                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4742                            adapter->tx_rx_desc_req);
4743                 break;
4744
4745         default:
4746                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4747                            crq->query_capability.capability);
4748         }
4749
4750 out:
4751         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4752                 adapter->wait_capability = false;
4753                 send_request_cap(adapter, 0);
4754         }
4755 }
4756
4757 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4758 {
4759         union ibmvnic_crq crq;
4760         int rc;
4761
4762         memset(&crq, 0, sizeof(crq));
4763         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4764         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4765
4766         mutex_lock(&adapter->fw_lock);
4767         adapter->fw_done_rc = 0;
4768         reinit_completion(&adapter->fw_done);
4769
4770         rc = ibmvnic_send_crq(adapter, &crq);
4771         if (rc) {
4772                 mutex_unlock(&adapter->fw_lock);
4773                 return rc;
4774         }
4775
4776         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4777         if (rc) {
4778                 mutex_unlock(&adapter->fw_lock);
4779                 return rc;
4780         }
4781
4782         mutex_unlock(&adapter->fw_lock);
4783         return adapter->fw_done_rc ? -EIO : 0;
4784 }
4785
4786 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4787                                        struct ibmvnic_adapter *adapter)
4788 {
4789         struct net_device *netdev = adapter->netdev;
4790         int rc;
4791         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4792
4793         rc = crq->query_phys_parms_rsp.rc.code;
4794         if (rc) {
4795                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4796                 return rc;
4797         }
4798         switch (rspeed) {
4799         case IBMVNIC_10MBPS:
4800                 adapter->speed = SPEED_10;
4801                 break;
4802         case IBMVNIC_100MBPS:
4803                 adapter->speed = SPEED_100;
4804                 break;
4805         case IBMVNIC_1GBPS:
4806                 adapter->speed = SPEED_1000;
4807                 break;
4808         case IBMVNIC_10GBPS:
4809                 adapter->speed = SPEED_10000;
4810                 break;
4811         case IBMVNIC_25GBPS:
4812                 adapter->speed = SPEED_25000;
4813                 break;
4814         case IBMVNIC_40GBPS:
4815                 adapter->speed = SPEED_40000;
4816                 break;
4817         case IBMVNIC_50GBPS:
4818                 adapter->speed = SPEED_50000;
4819                 break;
4820         case IBMVNIC_100GBPS:
4821                 adapter->speed = SPEED_100000;
4822                 break;
4823         case IBMVNIC_200GBPS:
4824                 adapter->speed = SPEED_200000;
4825                 break;
4826         default:
4827                 if (netif_carrier_ok(netdev))
4828                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4829                 adapter->speed = SPEED_UNKNOWN;
4830         }
4831         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4832                 adapter->duplex = DUPLEX_FULL;
4833         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4834                 adapter->duplex = DUPLEX_HALF;
4835         else
4836                 adapter->duplex = DUPLEX_UNKNOWN;
4837
4838         return rc;
4839 }
4840
4841 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4842                                struct ibmvnic_adapter *adapter)
4843 {
4844         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4845         struct net_device *netdev = adapter->netdev;
4846         struct device *dev = &adapter->vdev->dev;
4847         u64 *u64_crq = (u64 *)crq;
4848         long rc;
4849
4850         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4851                    (unsigned long)cpu_to_be64(u64_crq[0]),
4852                    (unsigned long)cpu_to_be64(u64_crq[1]));
4853         switch (gen_crq->first) {
4854         case IBMVNIC_CRQ_INIT_RSP:
4855                 switch (gen_crq->cmd) {
4856                 case IBMVNIC_CRQ_INIT:
4857                         dev_info(dev, "Partner initialized\n");
4858                         adapter->from_passive_init = true;
4859                         /* Discard any stale login responses from prev reset.
4860                          * CHECK: should we clear even on INIT_COMPLETE?
4861                          */
4862                         adapter->login_pending = false;
4863
4864                         if (!completion_done(&adapter->init_done)) {
4865                                 complete(&adapter->init_done);
4866                                 adapter->init_done_rc = -EIO;
4867                         }
4868                         rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4869                         if (rc && rc != -EBUSY) {
4870                                 /* We were unable to schedule the failover
4871                                  * reset either because the adapter was still
4872                                  * probing (eg: during kexec) or we could not
4873                                  * allocate memory. Clear the failover_pending
4874                                  * flag since no one else will. We ignore
4875                                  * EBUSY because it means either FAILOVER reset
4876                                  * is already scheduled or the adapter is
4877                                  * being removed.
4878                                  */
4879                                 netdev_err(netdev,
4880                                            "Error %ld scheduling failover reset\n",
4881                                            rc);
4882                                 adapter->failover_pending = false;
4883                         }
4884                         break;
4885                 case IBMVNIC_CRQ_INIT_COMPLETE:
4886                         dev_info(dev, "Partner initialization complete\n");
4887                         adapter->crq.active = true;
4888                         send_version_xchg(adapter);
4889                         break;
4890                 default:
4891                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4892                 }
4893                 return;
4894         case IBMVNIC_CRQ_XPORT_EVENT:
4895                 netif_carrier_off(netdev);
4896                 adapter->crq.active = false;
4897                 /* terminate any thread waiting for a response
4898                  * from the device
4899                  */
4900                 if (!completion_done(&adapter->fw_done)) {
4901                         adapter->fw_done_rc = -EIO;
4902                         complete(&adapter->fw_done);
4903                 }
4904                 if (!completion_done(&adapter->stats_done))
4905                         complete(&adapter->stats_done);
4906                 if (test_bit(0, &adapter->resetting))
4907                         adapter->force_reset_recovery = true;
4908                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4909                         dev_info(dev, "Migrated, re-enabling adapter\n");
4910                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4911                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4912                         dev_info(dev, "Backing device failover detected\n");
4913                         adapter->failover_pending = true;
4914                 } else {
4915                         /* The adapter lost the connection */
4916                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4917                                 gen_crq->cmd);
4918                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4919                 }
4920                 return;
4921         case IBMVNIC_CRQ_CMD_RSP:
4922                 break;
4923         default:
4924                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4925                         gen_crq->first);
4926                 return;
4927         }
4928
4929         switch (gen_crq->cmd) {
4930         case VERSION_EXCHANGE_RSP:
4931                 rc = crq->version_exchange_rsp.rc.code;
4932                 if (rc) {
4933                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4934                         break;
4935                 }
4936                 ibmvnic_version =
4937                             be16_to_cpu(crq->version_exchange_rsp.version);
4938                 dev_info(dev, "Partner protocol version is %d\n",
4939                          ibmvnic_version);
4940                 send_query_cap(adapter);
4941                 break;
4942         case QUERY_CAPABILITY_RSP:
4943                 handle_query_cap_rsp(crq, adapter);
4944                 break;
4945         case QUERY_MAP_RSP:
4946                 handle_query_map_rsp(crq, adapter);
4947                 break;
4948         case REQUEST_MAP_RSP:
4949                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4950                 complete(&adapter->fw_done);
4951                 break;
4952         case REQUEST_UNMAP_RSP:
4953                 handle_request_unmap_rsp(crq, adapter);
4954                 break;
4955         case REQUEST_CAPABILITY_RSP:
4956                 handle_request_cap_rsp(crq, adapter);
4957                 break;
4958         case LOGIN_RSP:
4959                 netdev_dbg(netdev, "Got Login Response\n");
4960                 handle_login_rsp(crq, adapter);
4961                 break;
4962         case LOGICAL_LINK_STATE_RSP:
4963                 netdev_dbg(netdev,
4964                            "Got Logical Link State Response, state: %d rc: %d\n",
4965                            crq->logical_link_state_rsp.link_state,
4966                            crq->logical_link_state_rsp.rc.code);
4967                 adapter->logical_link_state =
4968                     crq->logical_link_state_rsp.link_state;
4969                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4970                 complete(&adapter->init_done);
4971                 break;
4972         case LINK_STATE_INDICATION:
4973                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4974                 adapter->phys_link_state =
4975                     crq->link_state_indication.phys_link_state;
4976                 adapter->logical_link_state =
4977                     crq->link_state_indication.logical_link_state;
4978                 if (adapter->phys_link_state && adapter->logical_link_state)
4979                         netif_carrier_on(netdev);
4980                 else
4981                         netif_carrier_off(netdev);
4982                 break;
4983         case CHANGE_MAC_ADDR_RSP:
4984                 netdev_dbg(netdev, "Got MAC address change Response\n");
4985                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4986                 break;
4987         case ERROR_INDICATION:
4988                 netdev_dbg(netdev, "Got Error Indication\n");
4989                 handle_error_indication(crq, adapter);
4990                 break;
4991         case REQUEST_STATISTICS_RSP:
4992                 netdev_dbg(netdev, "Got Statistics Response\n");
4993                 complete(&adapter->stats_done);
4994                 break;
4995         case QUERY_IP_OFFLOAD_RSP:
4996                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4997                 handle_query_ip_offload_rsp(adapter);
4998                 break;
4999         case MULTICAST_CTRL_RSP:
5000                 netdev_dbg(netdev, "Got multicast control Response\n");
5001                 break;
5002         case CONTROL_IP_OFFLOAD_RSP:
5003                 netdev_dbg(netdev, "Got Control IP offload Response\n");
5004                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5005                                  sizeof(adapter->ip_offload_ctrl),
5006                                  DMA_TO_DEVICE);
5007                 complete(&adapter->init_done);
5008                 break;
5009         case COLLECT_FW_TRACE_RSP:
5010                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5011                 complete(&adapter->fw_done);
5012                 break;
5013         case GET_VPD_SIZE_RSP:
5014                 handle_vpd_size_rsp(crq, adapter);
5015                 break;
5016         case GET_VPD_RSP:
5017                 handle_vpd_rsp(crq, adapter);
5018                 break;
5019         case QUERY_PHYS_PARMS_RSP:
5020                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5021                 complete(&adapter->fw_done);
5022                 break;
5023         default:
5024                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5025                            gen_crq->cmd);
5026         }
5027 }
5028
5029 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5030 {
5031         struct ibmvnic_adapter *adapter = instance;
5032
5033         tasklet_schedule(&adapter->tasklet);
5034         return IRQ_HANDLED;
5035 }
5036
5037 static void ibmvnic_tasklet(struct tasklet_struct *t)
5038 {
5039         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5040         struct ibmvnic_crq_queue *queue = &adapter->crq;
5041         union ibmvnic_crq *crq;
5042         unsigned long flags;
5043         bool done = false;
5044
5045         spin_lock_irqsave(&queue->lock, flags);
5046         while (!done) {
5047                 /* Pull all the valid messages off the CRQ */
5048                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5049                         /* This barrier makes sure ibmvnic_next_crq()'s
5050                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5051                          * before ibmvnic_handle_crq()'s
5052                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5053                          */
5054                         dma_rmb();
5055                         ibmvnic_handle_crq(crq, adapter);
5056                         crq->generic.first = 0;
5057                 }
5058
5059                 /* remain in tasklet until all
5060                  * capabilities responses are received
5061                  */
5062                 if (!adapter->wait_capability)
5063                         done = true;
5064         }
5065         /* if capabilities CRQ's were sent in this tasklet, the following
5066          * tasklet must wait until all responses are received
5067          */
5068         if (atomic_read(&adapter->running_cap_crqs) != 0)
5069                 adapter->wait_capability = true;
5070         spin_unlock_irqrestore(&queue->lock, flags);
5071 }
5072
5073 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5074 {
5075         struct vio_dev *vdev = adapter->vdev;
5076         int rc;
5077
5078         do {
5079                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5080         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5081
5082         if (rc)
5083                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5084
5085         return rc;
5086 }
5087
5088 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5089 {
5090         struct ibmvnic_crq_queue *crq = &adapter->crq;
5091         struct device *dev = &adapter->vdev->dev;
5092         struct vio_dev *vdev = adapter->vdev;
5093         int rc;
5094
5095         /* Close the CRQ */
5096         do {
5097                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5098         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5099
5100         /* Clean out the queue */
5101         if (!crq->msgs)
5102                 return -EINVAL;
5103
5104         memset(crq->msgs, 0, PAGE_SIZE);
5105         crq->cur = 0;
5106         crq->active = false;
5107
5108         /* And re-open it again */
5109         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5110                                 crq->msg_token, PAGE_SIZE);
5111
5112         if (rc == H_CLOSED)
5113                 /* Adapter is good, but other end is not ready */
5114                 dev_warn(dev, "Partner adapter not ready\n");
5115         else if (rc != 0)
5116                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5117
5118         return rc;
5119 }
5120
5121 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5122 {
5123         struct ibmvnic_crq_queue *crq = &adapter->crq;
5124         struct vio_dev *vdev = adapter->vdev;
5125         long rc;
5126
5127         if (!crq->msgs)
5128                 return;
5129
5130         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5131         free_irq(vdev->irq, adapter);
5132         tasklet_kill(&adapter->tasklet);
5133         do {
5134                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5135         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5136
5137         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5138                          DMA_BIDIRECTIONAL);
5139         free_page((unsigned long)crq->msgs);
5140         crq->msgs = NULL;
5141         crq->active = false;
5142 }
5143
5144 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5145 {
5146         struct ibmvnic_crq_queue *crq = &adapter->crq;
5147         struct device *dev = &adapter->vdev->dev;
5148         struct vio_dev *vdev = adapter->vdev;
5149         int rc, retrc = -ENOMEM;
5150
5151         if (crq->msgs)
5152                 return 0;
5153
5154         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5155         /* Should we allocate more than one page? */
5156
5157         if (!crq->msgs)
5158                 return -ENOMEM;
5159
5160         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5161         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5162                                         DMA_BIDIRECTIONAL);
5163         if (dma_mapping_error(dev, crq->msg_token))
5164                 goto map_failed;
5165
5166         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5167                                 crq->msg_token, PAGE_SIZE);
5168
5169         if (rc == H_RESOURCE)
5170                 /* maybe kexecing and resource is busy. try a reset */
5171                 rc = ibmvnic_reset_crq(adapter);
5172         retrc = rc;
5173
5174         if (rc == H_CLOSED) {
5175                 dev_warn(dev, "Partner adapter not ready\n");
5176         } else if (rc) {
5177                 dev_warn(dev, "Error %d opening adapter\n", rc);
5178                 goto reg_crq_failed;
5179         }
5180
5181         retrc = 0;
5182
5183         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5184
5185         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5186         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5187                  adapter->vdev->unit_address);
5188         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5189         if (rc) {
5190                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5191                         vdev->irq, rc);
5192                 goto req_irq_failed;
5193         }
5194
5195         rc = vio_enable_interrupts(vdev);
5196         if (rc) {
5197                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5198                 goto req_irq_failed;
5199         }
5200
5201         crq->cur = 0;
5202         spin_lock_init(&crq->lock);
5203
5204         return retrc;
5205
5206 req_irq_failed:
5207         tasklet_kill(&adapter->tasklet);
5208         do {
5209                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5210         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5211 reg_crq_failed:
5212         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5213 map_failed:
5214         free_page((unsigned long)crq->msgs);
5215         crq->msgs = NULL;
5216         return retrc;
5217 }
5218
5219 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5220 {
5221         struct device *dev = &adapter->vdev->dev;
5222         unsigned long timeout = msecs_to_jiffies(20000);
5223         u64 old_num_rx_queues = adapter->req_rx_queues;
5224         u64 old_num_tx_queues = adapter->req_tx_queues;
5225         int rc;
5226
5227         adapter->from_passive_init = false;
5228
5229         if (reset)
5230                 reinit_completion(&adapter->init_done);
5231
5232         adapter->init_done_rc = 0;
5233         rc = ibmvnic_send_crq_init(adapter);
5234         if (rc) {
5235                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5236                 return rc;
5237         }
5238
5239         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5240                 dev_err(dev, "Initialization sequence timed out\n");
5241                 return -1;
5242         }
5243
5244         if (adapter->init_done_rc) {
5245                 release_crq_queue(adapter);
5246                 return adapter->init_done_rc;
5247         }
5248
5249         if (adapter->from_passive_init) {
5250                 adapter->state = VNIC_OPEN;
5251                 adapter->from_passive_init = false;
5252                 return -1;
5253         }
5254
5255         if (reset &&
5256             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5257             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5258                 if (adapter->req_rx_queues != old_num_rx_queues ||
5259                     adapter->req_tx_queues != old_num_tx_queues) {
5260                         release_sub_crqs(adapter, 0);
5261                         rc = init_sub_crqs(adapter);
5262                 } else {
5263                         rc = reset_sub_crq_queues(adapter);
5264                 }
5265         } else {
5266                 rc = init_sub_crqs(adapter);
5267         }
5268
5269         if (rc) {
5270                 dev_err(dev, "Initialization of sub crqs failed\n");
5271                 release_crq_queue(adapter);
5272                 return rc;
5273         }
5274
5275         rc = init_sub_crq_irqs(adapter);
5276         if (rc) {
5277                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5278                 release_crq_queue(adapter);
5279         }
5280
5281         return rc;
5282 }
5283
5284 static struct device_attribute dev_attr_failover;
5285
5286 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5287 {
5288         struct ibmvnic_adapter *adapter;
5289         struct net_device *netdev;
5290         unsigned char *mac_addr_p;
5291         int rc;
5292
5293         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5294                 dev->unit_address);
5295
5296         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5297                                                         VETH_MAC_ADDR, NULL);
5298         if (!mac_addr_p) {
5299                 dev_err(&dev->dev,
5300                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5301                         __FILE__, __LINE__);
5302                 return 0;
5303         }
5304
5305         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5306                                    IBMVNIC_MAX_QUEUES);
5307         if (!netdev)
5308                 return -ENOMEM;
5309
5310         adapter = netdev_priv(netdev);
5311         adapter->state = VNIC_PROBING;
5312         dev_set_drvdata(&dev->dev, netdev);
5313         adapter->vdev = dev;
5314         adapter->netdev = netdev;
5315         adapter->login_pending = false;
5316
5317         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5318         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5319         netdev->irq = dev->irq;
5320         netdev->netdev_ops = &ibmvnic_netdev_ops;
5321         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5322         SET_NETDEV_DEV(netdev, &dev->dev);
5323
5324         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5325         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5326                           __ibmvnic_delayed_reset);
5327         INIT_LIST_HEAD(&adapter->rwi_list);
5328         spin_lock_init(&adapter->rwi_lock);
5329         spin_lock_init(&adapter->state_lock);
5330         mutex_init(&adapter->fw_lock);
5331         init_completion(&adapter->init_done);
5332         init_completion(&adapter->fw_done);
5333         init_completion(&adapter->reset_done);
5334         init_completion(&adapter->stats_done);
5335         clear_bit(0, &adapter->resetting);
5336
5337         do {
5338                 rc = init_crq_queue(adapter);
5339                 if (rc) {
5340                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5341                                 rc);
5342                         goto ibmvnic_init_fail;
5343                 }
5344
5345                 rc = ibmvnic_reset_init(adapter, false);
5346                 if (rc && rc != EAGAIN)
5347                         goto ibmvnic_init_fail;
5348         } while (rc == EAGAIN);
5349
5350         rc = init_stats_buffers(adapter);
5351         if (rc)
5352                 goto ibmvnic_init_fail;
5353
5354         rc = init_stats_token(adapter);
5355         if (rc)
5356                 goto ibmvnic_stats_fail;
5357
5358         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5359         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5360         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5361
5362         rc = device_create_file(&dev->dev, &dev_attr_failover);
5363         if (rc)
5364                 goto ibmvnic_dev_file_err;
5365
5366         netif_carrier_off(netdev);
5367         rc = register_netdev(netdev);
5368         if (rc) {
5369                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5370                 goto ibmvnic_register_fail;
5371         }
5372         dev_info(&dev->dev, "ibmvnic registered\n");
5373
5374         adapter->state = VNIC_PROBED;
5375
5376         adapter->wait_for_reset = false;
5377         adapter->last_reset_time = jiffies;
5378         return 0;
5379
5380 ibmvnic_register_fail:
5381         device_remove_file(&dev->dev, &dev_attr_failover);
5382
5383 ibmvnic_dev_file_err:
5384         release_stats_token(adapter);
5385
5386 ibmvnic_stats_fail:
5387         release_stats_buffers(adapter);
5388
5389 ibmvnic_init_fail:
5390         release_sub_crqs(adapter, 1);
5391         release_crq_queue(adapter);
5392         mutex_destroy(&adapter->fw_lock);
5393         free_netdev(netdev);
5394
5395         return rc;
5396 }
5397
5398 static void ibmvnic_remove(struct vio_dev *dev)
5399 {
5400         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5401         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5402         unsigned long flags;
5403
5404         spin_lock_irqsave(&adapter->state_lock, flags);
5405
5406         /* If ibmvnic_reset() is scheduling a reset, wait for it to
5407          * finish. Then, set the state to REMOVING to prevent it from
5408          * scheduling any more work and to have reset functions ignore
5409          * any resets that have already been scheduled. Drop the lock
5410          * after setting state, so __ibmvnic_reset() which is called
5411          * from the flush_work() below, can make progress.
5412          */
5413         spin_lock(&adapter->rwi_lock);
5414         adapter->state = VNIC_REMOVING;
5415         spin_unlock(&adapter->rwi_lock);
5416
5417         spin_unlock_irqrestore(&adapter->state_lock, flags);
5418
5419         flush_work(&adapter->ibmvnic_reset);
5420         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5421
5422         rtnl_lock();
5423         unregister_netdevice(netdev);
5424
5425         release_resources(adapter);
5426         release_sub_crqs(adapter, 1);
5427         release_crq_queue(adapter);
5428
5429         release_stats_token(adapter);
5430         release_stats_buffers(adapter);
5431
5432         adapter->state = VNIC_REMOVED;
5433
5434         rtnl_unlock();
5435         mutex_destroy(&adapter->fw_lock);
5436         device_remove_file(&dev->dev, &dev_attr_failover);
5437         free_netdev(netdev);
5438         dev_set_drvdata(&dev->dev, NULL);
5439 }
5440
5441 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5442                               const char *buf, size_t count)
5443 {
5444         struct net_device *netdev = dev_get_drvdata(dev);
5445         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5446         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5447         __be64 session_token;
5448         long rc;
5449
5450         if (!sysfs_streq(buf, "1"))
5451                 return -EINVAL;
5452
5453         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5454                          H_GET_SESSION_TOKEN, 0, 0, 0);
5455         if (rc) {
5456                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5457                            rc);
5458                 return -EINVAL;
5459         }
5460
5461         session_token = (__be64)retbuf[0];
5462         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5463                    be64_to_cpu(session_token));
5464         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5465                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5466         if (rc) {
5467                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5468                            rc);
5469                 return -EINVAL;
5470         }
5471
5472         return count;
5473 }
5474
5475 static DEVICE_ATTR_WO(failover);
5476
5477 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5478 {
5479         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5480         struct ibmvnic_adapter *adapter;
5481         struct iommu_table *tbl;
5482         unsigned long ret = 0;
5483         int i;
5484
5485         tbl = get_iommu_table_base(&vdev->dev);
5486
5487         /* netdev inits at probe time along with the structures we need below*/
5488         if (!netdev)
5489                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5490
5491         adapter = netdev_priv(netdev);
5492
5493         ret += PAGE_SIZE; /* the crq message queue */
5494         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5495
5496         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5497                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5498
5499         for (i = 0; i < adapter->num_active_rx_pools; i++)
5500                 ret += adapter->rx_pool[i].size *
5501                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5502
5503         return ret;
5504 }
5505
5506 static int ibmvnic_resume(struct device *dev)
5507 {
5508         struct net_device *netdev = dev_get_drvdata(dev);
5509         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5510
5511         if (adapter->state != VNIC_OPEN)
5512                 return 0;
5513
5514         tasklet_schedule(&adapter->tasklet);
5515
5516         return 0;
5517 }
5518
5519 static const struct vio_device_id ibmvnic_device_table[] = {
5520         {"network", "IBM,vnic"},
5521         {"", "" }
5522 };
5523 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5524
5525 static const struct dev_pm_ops ibmvnic_pm_ops = {
5526         .resume = ibmvnic_resume
5527 };
5528
5529 static struct vio_driver ibmvnic_driver = {
5530         .id_table       = ibmvnic_device_table,
5531         .probe          = ibmvnic_probe,
5532         .remove         = ibmvnic_remove,
5533         .get_desired_dma = ibmvnic_get_desired_dma,
5534         .name           = ibmvnic_driver_name,
5535         .pm             = &ibmvnic_pm_ops,
5536 };
5537
5538 /* module functions */
5539 static int __init ibmvnic_module_init(void)
5540 {
5541         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5542                 IBMVNIC_DRIVER_VERSION);
5543
5544         return vio_register_driver(&ibmvnic_driver);
5545 }
5546
5547 static void __exit ibmvnic_module_exit(void)
5548 {
5549         vio_unregister_driver(&ibmvnic_driver);
5550 }
5551
5552 module_init(ibmvnic_module_init);
5553 module_exit(ibmvnic_module_exit);